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/filelock.h>
24#include <linux/namei.h>
25#include <linux/pagemap.h>
26#include <linux/sched/mm.h>
27#include <linux/fsnotify.h>
28#include <linux/personality.h>
29#include <linux/security.h>
30#include <linux/ima.h>
31#include <linux/syscalls.h>
32#include <linux/mount.h>
33#include <linux/audit.h>
34#include <linux/capability.h>
35#include <linux/file.h>
36#include <linux/fcntl.h>
37#include <linux/device_cgroup.h>
38#include <linux/fs_struct.h>
39#include <linux/posix_acl.h>
40#include <linux/hash.h>
41#include <linux/bitops.h>
42#include <linux/init_task.h>
43#include <linux/uaccess.h>
44
45#include "internal.h"
46#include "mount.h"
47
48/* [Feb-1997 T. Schoebel-Theuer]
49 * Fundamental changes in the pathname lookup mechanisms (namei)
50 * were necessary because of omirr. The reason is that omirr needs
51 * to know the _real_ pathname, not the user-supplied one, in case
52 * of symlinks (and also when transname replacements occur).
53 *
54 * The new code replaces the old recursive symlink resolution with
55 * an iterative one (in case of non-nested symlink chains). It does
56 * this with calls to <fs>_follow_link().
57 * As a side effect, dir_namei(), _namei() and follow_link() are now
58 * replaced with a single function lookup_dentry() that can handle all
59 * the special cases of the former code.
60 *
61 * With the new dcache, the pathname is stored at each inode, at least as
62 * long as the refcount of the inode is positive. As a side effect, the
63 * size of the dcache depends on the inode cache and thus is dynamic.
64 *
65 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
66 * resolution to correspond with current state of the code.
67 *
68 * Note that the symlink resolution is not *completely* iterative.
69 * There is still a significant amount of tail- and mid- recursion in
70 * the algorithm. Also, note that <fs>_readlink() is not used in
71 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
72 * may return different results than <fs>_follow_link(). Many virtual
73 * filesystems (including /proc) exhibit this behavior.
74 */
75
76/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
77 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
78 * and the name already exists in form of a symlink, try to create the new
79 * name indicated by the symlink. The old code always complained that the
80 * name already exists, due to not following the symlink even if its target
81 * is nonexistent. The new semantics affects also mknod() and link() when
82 * the name is a symlink pointing to a non-existent name.
83 *
84 * I don't know which semantics is the right one, since I have no access
85 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
86 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
87 * "old" one. Personally, I think the new semantics is much more logical.
88 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
89 * file does succeed in both HP-UX and SunOs, but not in Solaris
90 * and in the old Linux semantics.
91 */
92
93/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
94 * semantics. See the comments in "open_namei" and "do_link" below.
95 *
96 * [10-Sep-98 Alan Modra] Another symlink change.
97 */
98
99/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
100 * inside the path - always follow.
101 * in the last component in creation/removal/renaming - never follow.
102 * if LOOKUP_FOLLOW passed - follow.
103 * if the pathname has trailing slashes - follow.
104 * otherwise - don't follow.
105 * (applied in that order).
106 *
107 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
108 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
109 * During the 2.4 we need to fix the userland stuff depending on it -
110 * hopefully we will be able to get rid of that wart in 2.5. So far only
111 * XEmacs seems to be relying on it...
112 */
113/*
114 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
115 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
116 * any extra contention...
117 */
118
119/* In order to reduce some races, while at the same time doing additional
120 * checking and hopefully speeding things up, we copy filenames to the
121 * kernel data space before using them..
122 *
123 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
124 * PATH_MAX includes the nul terminator --RR.
125 */
126
127#define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
128
129struct filename *
130getname_flags(const char __user *filename, int flags, int *empty)
131{
132 struct filename *result;
133 char *kname;
134 int len;
135
136 result = audit_reusename(filename);
137 if (result)
138 return result;
139
140 result = __getname();
141 if (unlikely(!result))
142 return ERR_PTR(-ENOMEM);
143
144 /*
145 * First, try to embed the struct filename inside the names_cache
146 * allocation
147 */
148 kname = (char *)result->iname;
149 result->name = kname;
150
151 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
152 if (unlikely(len < 0)) {
153 __putname(result);
154 return ERR_PTR(len);
155 }
156
157 /*
158 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
159 * separate struct filename so we can dedicate the entire
160 * names_cache allocation for the pathname, and re-do the copy from
161 * userland.
162 */
163 if (unlikely(len == EMBEDDED_NAME_MAX)) {
164 const size_t size = offsetof(struct filename, iname[1]);
165 kname = (char *)result;
166
167 /*
168 * size is chosen that way we to guarantee that
169 * result->iname[0] is within the same object and that
170 * kname can't be equal to result->iname, no matter what.
171 */
172 result = kzalloc(size, GFP_KERNEL);
173 if (unlikely(!result)) {
174 __putname(kname);
175 return ERR_PTR(-ENOMEM);
176 }
177 result->name = kname;
178 len = strncpy_from_user(kname, filename, PATH_MAX);
179 if (unlikely(len < 0)) {
180 __putname(kname);
181 kfree(result);
182 return ERR_PTR(len);
183 }
184 if (unlikely(len == PATH_MAX)) {
185 __putname(kname);
186 kfree(result);
187 return ERR_PTR(-ENAMETOOLONG);
188 }
189 }
190
191 atomic_set(&result->refcnt, 1);
192 /* The empty path is special. */
193 if (unlikely(!len)) {
194 if (empty)
195 *empty = 1;
196 if (!(flags & LOOKUP_EMPTY)) {
197 putname(result);
198 return ERR_PTR(-ENOENT);
199 }
200 }
201
202 result->uptr = filename;
203 result->aname = NULL;
204 audit_getname(result);
205 return result;
206}
207
208struct filename *
209getname_uflags(const char __user *filename, int uflags)
210{
211 int flags = (uflags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
212
213 return getname_flags(filename, flags, NULL);
214}
215
216struct filename *
217getname(const char __user * filename)
218{
219 return getname_flags(filename, 0, NULL);
220}
221
222struct filename *
223getname_kernel(const char * filename)
224{
225 struct filename *result;
226 int len = strlen(filename) + 1;
227
228 result = __getname();
229 if (unlikely(!result))
230 return ERR_PTR(-ENOMEM);
231
232 if (len <= EMBEDDED_NAME_MAX) {
233 result->name = (char *)result->iname;
234 } else if (len <= PATH_MAX) {
235 const size_t size = offsetof(struct filename, iname[1]);
236 struct filename *tmp;
237
238 tmp = kmalloc(size, GFP_KERNEL);
239 if (unlikely(!tmp)) {
240 __putname(result);
241 return ERR_PTR(-ENOMEM);
242 }
243 tmp->name = (char *)result;
244 result = tmp;
245 } else {
246 __putname(result);
247 return ERR_PTR(-ENAMETOOLONG);
248 }
249 memcpy((char *)result->name, filename, len);
250 result->uptr = NULL;
251 result->aname = NULL;
252 atomic_set(&result->refcnt, 1);
253 audit_getname(result);
254
255 return result;
256}
257EXPORT_SYMBOL(getname_kernel);
258
259void putname(struct filename *name)
260{
261 if (IS_ERR(name))
262 return;
263
264 if (WARN_ON_ONCE(!atomic_read(&name->refcnt)))
265 return;
266
267 if (!atomic_dec_and_test(&name->refcnt))
268 return;
269
270 if (name->name != name->iname) {
271 __putname(name->name);
272 kfree(name);
273 } else
274 __putname(name);
275}
276EXPORT_SYMBOL(putname);
277
278/**
279 * check_acl - perform ACL permission checking
280 * @idmap: idmap of the mount the inode was found from
281 * @inode: inode to check permissions on
282 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
283 *
284 * This function performs the ACL permission checking. Since this function
285 * retrieve POSIX acls it needs to know whether it is called from a blocking or
286 * non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
287 *
288 * If the inode has been found through an idmapped mount the idmap of
289 * the vfsmount must be passed through @idmap. This function will then take
290 * care to map the inode according to @idmap before checking permissions.
291 * On non-idmapped mounts or if permission checking is to be performed on the
292 * raw inode simply pass @nop_mnt_idmap.
293 */
294static int check_acl(struct mnt_idmap *idmap,
295 struct inode *inode, int mask)
296{
297#ifdef CONFIG_FS_POSIX_ACL
298 struct posix_acl *acl;
299
300 if (mask & MAY_NOT_BLOCK) {
301 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
302 if (!acl)
303 return -EAGAIN;
304 /* no ->get_inode_acl() calls in RCU mode... */
305 if (is_uncached_acl(acl))
306 return -ECHILD;
307 return posix_acl_permission(idmap, inode, acl, mask);
308 }
309
310 acl = get_inode_acl(inode, ACL_TYPE_ACCESS);
311 if (IS_ERR(acl))
312 return PTR_ERR(acl);
313 if (acl) {
314 int error = posix_acl_permission(idmap, inode, acl, mask);
315 posix_acl_release(acl);
316 return error;
317 }
318#endif
319
320 return -EAGAIN;
321}
322
323/**
324 * acl_permission_check - perform basic UNIX permission checking
325 * @idmap: idmap of the mount the inode was found from
326 * @inode: inode to check permissions on
327 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
328 *
329 * This function performs the basic UNIX permission checking. Since this
330 * function may retrieve POSIX acls it needs to know whether it is called from a
331 * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
332 *
333 * If the inode has been found through an idmapped mount the idmap of
334 * the vfsmount must be passed through @idmap. This function will then take
335 * care to map the inode according to @idmap before checking permissions.
336 * On non-idmapped mounts or if permission checking is to be performed on the
337 * raw inode simply pass @nop_mnt_idmap.
338 */
339static int acl_permission_check(struct mnt_idmap *idmap,
340 struct inode *inode, int mask)
341{
342 unsigned int mode = inode->i_mode;
343 vfsuid_t vfsuid;
344
345 /* Are we the owner? If so, ACL's don't matter */
346 vfsuid = i_uid_into_vfsuid(idmap, inode);
347 if (likely(vfsuid_eq_kuid(vfsuid, current_fsuid()))) {
348 mask &= 7;
349 mode >>= 6;
350 return (mask & ~mode) ? -EACCES : 0;
351 }
352
353 /* Do we have ACL's? */
354 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
355 int error = check_acl(idmap, inode, mask);
356 if (error != -EAGAIN)
357 return error;
358 }
359
360 /* Only RWX matters for group/other mode bits */
361 mask &= 7;
362
363 /*
364 * Are the group permissions different from
365 * the other permissions in the bits we care
366 * about? Need to check group ownership if so.
367 */
368 if (mask & (mode ^ (mode >> 3))) {
369 vfsgid_t vfsgid = i_gid_into_vfsgid(idmap, inode);
370 if (vfsgid_in_group_p(vfsgid))
371 mode >>= 3;
372 }
373
374 /* Bits in 'mode' clear that we require? */
375 return (mask & ~mode) ? -EACCES : 0;
376}
377
378/**
379 * generic_permission - check for access rights on a Posix-like filesystem
380 * @idmap: idmap of the mount the inode was found from
381 * @inode: inode to check access rights for
382 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC,
383 * %MAY_NOT_BLOCK ...)
384 *
385 * Used to check for read/write/execute permissions on a file.
386 * We use "fsuid" for this, letting us set arbitrary permissions
387 * for filesystem access without changing the "normal" uids which
388 * are used for other things.
389 *
390 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
391 * request cannot be satisfied (eg. requires blocking or too much complexity).
392 * It would then be called again in ref-walk mode.
393 *
394 * If the inode has been found through an idmapped mount the idmap of
395 * the vfsmount must be passed through @idmap. This function will then take
396 * care to map the inode according to @idmap before checking permissions.
397 * On non-idmapped mounts or if permission checking is to be performed on the
398 * raw inode simply pass @nop_mnt_idmap.
399 */
400int generic_permission(struct mnt_idmap *idmap, struct inode *inode,
401 int mask)
402{
403 int ret;
404
405 /*
406 * Do the basic permission checks.
407 */
408 ret = acl_permission_check(idmap, inode, mask);
409 if (ret != -EACCES)
410 return ret;
411
412 if (S_ISDIR(inode->i_mode)) {
413 /* DACs are overridable for directories */
414 if (!(mask & MAY_WRITE))
415 if (capable_wrt_inode_uidgid(idmap, inode,
416 CAP_DAC_READ_SEARCH))
417 return 0;
418 if (capable_wrt_inode_uidgid(idmap, inode,
419 CAP_DAC_OVERRIDE))
420 return 0;
421 return -EACCES;
422 }
423
424 /*
425 * Searching includes executable on directories, else just read.
426 */
427 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
428 if (mask == MAY_READ)
429 if (capable_wrt_inode_uidgid(idmap, inode,
430 CAP_DAC_READ_SEARCH))
431 return 0;
432 /*
433 * Read/write DACs are always overridable.
434 * Executable DACs are overridable when there is
435 * at least one exec bit set.
436 */
437 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
438 if (capable_wrt_inode_uidgid(idmap, inode,
439 CAP_DAC_OVERRIDE))
440 return 0;
441
442 return -EACCES;
443}
444EXPORT_SYMBOL(generic_permission);
445
446/**
447 * do_inode_permission - UNIX permission checking
448 * @idmap: idmap of the mount the inode was found from
449 * @inode: inode to check permissions on
450 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
451 *
452 * We _really_ want to just do "generic_permission()" without
453 * even looking at the inode->i_op values. So we keep a cache
454 * flag in inode->i_opflags, that says "this has not special
455 * permission function, use the fast case".
456 */
457static inline int do_inode_permission(struct mnt_idmap *idmap,
458 struct inode *inode, int mask)
459{
460 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
461 if (likely(inode->i_op->permission))
462 return inode->i_op->permission(idmap, inode, mask);
463
464 /* This gets set once for the inode lifetime */
465 spin_lock(&inode->i_lock);
466 inode->i_opflags |= IOP_FASTPERM;
467 spin_unlock(&inode->i_lock);
468 }
469 return generic_permission(idmap, inode, mask);
470}
471
472/**
473 * sb_permission - Check superblock-level permissions
474 * @sb: Superblock of inode to check permission on
475 * @inode: Inode to check permission on
476 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
477 *
478 * Separate out file-system wide checks from inode-specific permission checks.
479 */
480static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
481{
482 if (unlikely(mask & MAY_WRITE)) {
483 umode_t mode = inode->i_mode;
484
485 /* Nobody gets write access to a read-only fs. */
486 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
487 return -EROFS;
488 }
489 return 0;
490}
491
492/**
493 * inode_permission - Check for access rights to a given inode
494 * @idmap: idmap of the mount the inode was found from
495 * @inode: Inode to check permission on
496 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
497 *
498 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
499 * this, letting us set arbitrary permissions for filesystem access without
500 * changing the "normal" UIDs which are used for other things.
501 *
502 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
503 */
504int inode_permission(struct mnt_idmap *idmap,
505 struct inode *inode, int mask)
506{
507 int retval;
508
509 retval = sb_permission(inode->i_sb, inode, mask);
510 if (retval)
511 return retval;
512
513 if (unlikely(mask & MAY_WRITE)) {
514 /*
515 * Nobody gets write access to an immutable file.
516 */
517 if (IS_IMMUTABLE(inode))
518 return -EPERM;
519
520 /*
521 * Updating mtime will likely cause i_uid and i_gid to be
522 * written back improperly if their true value is unknown
523 * to the vfs.
524 */
525 if (HAS_UNMAPPED_ID(idmap, inode))
526 return -EACCES;
527 }
528
529 retval = do_inode_permission(idmap, inode, mask);
530 if (retval)
531 return retval;
532
533 retval = devcgroup_inode_permission(inode, mask);
534 if (retval)
535 return retval;
536
537 return security_inode_permission(inode, mask);
538}
539EXPORT_SYMBOL(inode_permission);
540
541/**
542 * path_get - get a reference to a path
543 * @path: path to get the reference to
544 *
545 * Given a path increment the reference count to the dentry and the vfsmount.
546 */
547void path_get(const struct path *path)
548{
549 mntget(path->mnt);
550 dget(path->dentry);
551}
552EXPORT_SYMBOL(path_get);
553
554/**
555 * path_put - put a reference to a path
556 * @path: path to put the reference to
557 *
558 * Given a path decrement the reference count to the dentry and the vfsmount.
559 */
560void path_put(const struct path *path)
561{
562 dput(path->dentry);
563 mntput(path->mnt);
564}
565EXPORT_SYMBOL(path_put);
566
567#define EMBEDDED_LEVELS 2
568struct nameidata {
569 struct path path;
570 struct qstr last;
571 struct path root;
572 struct inode *inode; /* path.dentry.d_inode */
573 unsigned int flags, state;
574 unsigned seq, next_seq, m_seq, r_seq;
575 int last_type;
576 unsigned depth;
577 int total_link_count;
578 struct saved {
579 struct path link;
580 struct delayed_call done;
581 const char *name;
582 unsigned seq;
583 } *stack, internal[EMBEDDED_LEVELS];
584 struct filename *name;
585 struct nameidata *saved;
586 unsigned root_seq;
587 int dfd;
588 vfsuid_t dir_vfsuid;
589 umode_t dir_mode;
590} __randomize_layout;
591
592#define ND_ROOT_PRESET 1
593#define ND_ROOT_GRABBED 2
594#define ND_JUMPED 4
595
596static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name)
597{
598 struct nameidata *old = current->nameidata;
599 p->stack = p->internal;
600 p->depth = 0;
601 p->dfd = dfd;
602 p->name = name;
603 p->path.mnt = NULL;
604 p->path.dentry = NULL;
605 p->total_link_count = old ? old->total_link_count : 0;
606 p->saved = old;
607 current->nameidata = p;
608}
609
610static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name,
611 const struct path *root)
612{
613 __set_nameidata(p, dfd, name);
614 p->state = 0;
615 if (unlikely(root)) {
616 p->state = ND_ROOT_PRESET;
617 p->root = *root;
618 }
619}
620
621static void restore_nameidata(void)
622{
623 struct nameidata *now = current->nameidata, *old = now->saved;
624
625 current->nameidata = old;
626 if (old)
627 old->total_link_count = now->total_link_count;
628 if (now->stack != now->internal)
629 kfree(now->stack);
630}
631
632static bool nd_alloc_stack(struct nameidata *nd)
633{
634 struct saved *p;
635
636 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
637 nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL);
638 if (unlikely(!p))
639 return false;
640 memcpy(p, nd->internal, sizeof(nd->internal));
641 nd->stack = p;
642 return true;
643}
644
645/**
646 * path_connected - Verify that a dentry is below mnt.mnt_root
647 * @mnt: The mountpoint to check.
648 * @dentry: The dentry to check.
649 *
650 * Rename can sometimes move a file or directory outside of a bind
651 * mount, path_connected allows those cases to be detected.
652 */
653static bool path_connected(struct vfsmount *mnt, struct dentry *dentry)
654{
655 struct super_block *sb = mnt->mnt_sb;
656
657 /* Bind mounts can have disconnected paths */
658 if (mnt->mnt_root == sb->s_root)
659 return true;
660
661 return is_subdir(dentry, mnt->mnt_root);
662}
663
664static void drop_links(struct nameidata *nd)
665{
666 int i = nd->depth;
667 while (i--) {
668 struct saved *last = nd->stack + i;
669 do_delayed_call(&last->done);
670 clear_delayed_call(&last->done);
671 }
672}
673
674static void leave_rcu(struct nameidata *nd)
675{
676 nd->flags &= ~LOOKUP_RCU;
677 nd->seq = nd->next_seq = 0;
678 rcu_read_unlock();
679}
680
681static void terminate_walk(struct nameidata *nd)
682{
683 drop_links(nd);
684 if (!(nd->flags & LOOKUP_RCU)) {
685 int i;
686 path_put(&nd->path);
687 for (i = 0; i < nd->depth; i++)
688 path_put(&nd->stack[i].link);
689 if (nd->state & ND_ROOT_GRABBED) {
690 path_put(&nd->root);
691 nd->state &= ~ND_ROOT_GRABBED;
692 }
693 } else {
694 leave_rcu(nd);
695 }
696 nd->depth = 0;
697 nd->path.mnt = NULL;
698 nd->path.dentry = NULL;
699}
700
701/* path_put is needed afterwards regardless of success or failure */
702static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq)
703{
704 int res = __legitimize_mnt(path->mnt, mseq);
705 if (unlikely(res)) {
706 if (res > 0)
707 path->mnt = NULL;
708 path->dentry = NULL;
709 return false;
710 }
711 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
712 path->dentry = NULL;
713 return false;
714 }
715 return !read_seqcount_retry(&path->dentry->d_seq, seq);
716}
717
718static inline bool legitimize_path(struct nameidata *nd,
719 struct path *path, unsigned seq)
720{
721 return __legitimize_path(path, seq, nd->m_seq);
722}
723
724static bool legitimize_links(struct nameidata *nd)
725{
726 int i;
727 if (unlikely(nd->flags & LOOKUP_CACHED)) {
728 drop_links(nd);
729 nd->depth = 0;
730 return false;
731 }
732 for (i = 0; i < nd->depth; i++) {
733 struct saved *last = nd->stack + i;
734 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
735 drop_links(nd);
736 nd->depth = i + 1;
737 return false;
738 }
739 }
740 return true;
741}
742
743static bool legitimize_root(struct nameidata *nd)
744{
745 /* Nothing to do if nd->root is zero or is managed by the VFS user. */
746 if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET))
747 return true;
748 nd->state |= ND_ROOT_GRABBED;
749 return legitimize_path(nd, &nd->root, nd->root_seq);
750}
751
752/*
753 * Path walking has 2 modes, rcu-walk and ref-walk (see
754 * Documentation/filesystems/path-lookup.txt). In situations when we can't
755 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
756 * normal reference counts on dentries and vfsmounts to transition to ref-walk
757 * mode. Refcounts are grabbed at the last known good point before rcu-walk
758 * got stuck, so ref-walk may continue from there. If this is not successful
759 * (eg. a seqcount has changed), then failure is returned and it's up to caller
760 * to restart the path walk from the beginning in ref-walk mode.
761 */
762
763/**
764 * try_to_unlazy - try to switch to ref-walk mode.
765 * @nd: nameidata pathwalk data
766 * Returns: true on success, false on failure
767 *
768 * try_to_unlazy attempts to legitimize the current nd->path and nd->root
769 * for ref-walk mode.
770 * Must be called from rcu-walk context.
771 * Nothing should touch nameidata between try_to_unlazy() failure and
772 * terminate_walk().
773 */
774static bool try_to_unlazy(struct nameidata *nd)
775{
776 struct dentry *parent = nd->path.dentry;
777
778 BUG_ON(!(nd->flags & LOOKUP_RCU));
779
780 if (unlikely(!legitimize_links(nd)))
781 goto out1;
782 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
783 goto out;
784 if (unlikely(!legitimize_root(nd)))
785 goto out;
786 leave_rcu(nd);
787 BUG_ON(nd->inode != parent->d_inode);
788 return true;
789
790out1:
791 nd->path.mnt = NULL;
792 nd->path.dentry = NULL;
793out:
794 leave_rcu(nd);
795 return false;
796}
797
798/**
799 * try_to_unlazy_next - try to switch to ref-walk mode.
800 * @nd: nameidata pathwalk data
801 * @dentry: next dentry to step into
802 * Returns: true on success, false on failure
803 *
804 * Similar to try_to_unlazy(), but here we have the next dentry already
805 * picked by rcu-walk and want to legitimize that in addition to the current
806 * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context.
807 * Nothing should touch nameidata between try_to_unlazy_next() failure and
808 * terminate_walk().
809 */
810static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry)
811{
812 int res;
813 BUG_ON(!(nd->flags & LOOKUP_RCU));
814
815 if (unlikely(!legitimize_links(nd)))
816 goto out2;
817 res = __legitimize_mnt(nd->path.mnt, nd->m_seq);
818 if (unlikely(res)) {
819 if (res > 0)
820 goto out2;
821 goto out1;
822 }
823 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
824 goto out1;
825
826 /*
827 * We need to move both the parent and the dentry from the RCU domain
828 * to be properly refcounted. And the sequence number in the dentry
829 * validates *both* dentry counters, since we checked the sequence
830 * number of the parent after we got the child sequence number. So we
831 * know the parent must still be valid if the child sequence number is
832 */
833 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
834 goto out;
835 if (read_seqcount_retry(&dentry->d_seq, nd->next_seq))
836 goto out_dput;
837 /*
838 * Sequence counts matched. Now make sure that the root is
839 * still valid and get it if required.
840 */
841 if (unlikely(!legitimize_root(nd)))
842 goto out_dput;
843 leave_rcu(nd);
844 return true;
845
846out2:
847 nd->path.mnt = NULL;
848out1:
849 nd->path.dentry = NULL;
850out:
851 leave_rcu(nd);
852 return false;
853out_dput:
854 leave_rcu(nd);
855 dput(dentry);
856 return false;
857}
858
859static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
860{
861 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
862 return dentry->d_op->d_revalidate(dentry, flags);
863 else
864 return 1;
865}
866
867/**
868 * complete_walk - successful completion of path walk
869 * @nd: pointer nameidata
870 *
871 * If we had been in RCU mode, drop out of it and legitimize nd->path.
872 * Revalidate the final result, unless we'd already done that during
873 * the path walk or the filesystem doesn't ask for it. Return 0 on
874 * success, -error on failure. In case of failure caller does not
875 * need to drop nd->path.
876 */
877static int complete_walk(struct nameidata *nd)
878{
879 struct dentry *dentry = nd->path.dentry;
880 int status;
881
882 if (nd->flags & LOOKUP_RCU) {
883 /*
884 * We don't want to zero nd->root for scoped-lookups or
885 * externally-managed nd->root.
886 */
887 if (!(nd->state & ND_ROOT_PRESET))
888 if (!(nd->flags & LOOKUP_IS_SCOPED))
889 nd->root.mnt = NULL;
890 nd->flags &= ~LOOKUP_CACHED;
891 if (!try_to_unlazy(nd))
892 return -ECHILD;
893 }
894
895 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
896 /*
897 * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
898 * ever step outside the root during lookup" and should already
899 * be guaranteed by the rest of namei, we want to avoid a namei
900 * BUG resulting in userspace being given a path that was not
901 * scoped within the root at some point during the lookup.
902 *
903 * So, do a final sanity-check to make sure that in the
904 * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
905 * we won't silently return an fd completely outside of the
906 * requested root to userspace.
907 *
908 * Userspace could move the path outside the root after this
909 * check, but as discussed elsewhere this is not a concern (the
910 * resolved file was inside the root at some point).
911 */
912 if (!path_is_under(&nd->path, &nd->root))
913 return -EXDEV;
914 }
915
916 if (likely(!(nd->state & ND_JUMPED)))
917 return 0;
918
919 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
920 return 0;
921
922 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
923 if (status > 0)
924 return 0;
925
926 if (!status)
927 status = -ESTALE;
928
929 return status;
930}
931
932static int set_root(struct nameidata *nd)
933{
934 struct fs_struct *fs = current->fs;
935
936 /*
937 * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
938 * still have to ensure it doesn't happen because it will cause a breakout
939 * from the dirfd.
940 */
941 if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED))
942 return -ENOTRECOVERABLE;
943
944 if (nd->flags & LOOKUP_RCU) {
945 unsigned seq;
946
947 do {
948 seq = read_seqcount_begin(&fs->seq);
949 nd->root = fs->root;
950 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
951 } while (read_seqcount_retry(&fs->seq, seq));
952 } else {
953 get_fs_root(fs, &nd->root);
954 nd->state |= ND_ROOT_GRABBED;
955 }
956 return 0;
957}
958
959static int nd_jump_root(struct nameidata *nd)
960{
961 if (unlikely(nd->flags & LOOKUP_BENEATH))
962 return -EXDEV;
963 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
964 /* Absolute path arguments to path_init() are allowed. */
965 if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt)
966 return -EXDEV;
967 }
968 if (!nd->root.mnt) {
969 int error = set_root(nd);
970 if (error)
971 return error;
972 }
973 if (nd->flags & LOOKUP_RCU) {
974 struct dentry *d;
975 nd->path = nd->root;
976 d = nd->path.dentry;
977 nd->inode = d->d_inode;
978 nd->seq = nd->root_seq;
979 if (read_seqcount_retry(&d->d_seq, nd->seq))
980 return -ECHILD;
981 } else {
982 path_put(&nd->path);
983 nd->path = nd->root;
984 path_get(&nd->path);
985 nd->inode = nd->path.dentry->d_inode;
986 }
987 nd->state |= ND_JUMPED;
988 return 0;
989}
990
991/*
992 * Helper to directly jump to a known parsed path from ->get_link,
993 * caller must have taken a reference to path beforehand.
994 */
995int nd_jump_link(const struct path *path)
996{
997 int error = -ELOOP;
998 struct nameidata *nd = current->nameidata;
999
1000 if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS))
1001 goto err;
1002
1003 error = -EXDEV;
1004 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
1005 if (nd->path.mnt != path->mnt)
1006 goto err;
1007 }
1008 /* Not currently safe for scoped-lookups. */
1009 if (unlikely(nd->flags & LOOKUP_IS_SCOPED))
1010 goto err;
1011
1012 path_put(&nd->path);
1013 nd->path = *path;
1014 nd->inode = nd->path.dentry->d_inode;
1015 nd->state |= ND_JUMPED;
1016 return 0;
1017
1018err:
1019 path_put(path);
1020 return error;
1021}
1022
1023static inline void put_link(struct nameidata *nd)
1024{
1025 struct saved *last = nd->stack + --nd->depth;
1026 do_delayed_call(&last->done);
1027 if (!(nd->flags & LOOKUP_RCU))
1028 path_put(&last->link);
1029}
1030
1031static int sysctl_protected_symlinks __read_mostly;
1032static int sysctl_protected_hardlinks __read_mostly;
1033static int sysctl_protected_fifos __read_mostly;
1034static int sysctl_protected_regular __read_mostly;
1035
1036#ifdef CONFIG_SYSCTL
1037static struct ctl_table namei_sysctls[] = {
1038 {
1039 .procname = "protected_symlinks",
1040 .data = &sysctl_protected_symlinks,
1041 .maxlen = sizeof(int),
1042 .mode = 0644,
1043 .proc_handler = proc_dointvec_minmax,
1044 .extra1 = SYSCTL_ZERO,
1045 .extra2 = SYSCTL_ONE,
1046 },
1047 {
1048 .procname = "protected_hardlinks",
1049 .data = &sysctl_protected_hardlinks,
1050 .maxlen = sizeof(int),
1051 .mode = 0644,
1052 .proc_handler = proc_dointvec_minmax,
1053 .extra1 = SYSCTL_ZERO,
1054 .extra2 = SYSCTL_ONE,
1055 },
1056 {
1057 .procname = "protected_fifos",
1058 .data = &sysctl_protected_fifos,
1059 .maxlen = sizeof(int),
1060 .mode = 0644,
1061 .proc_handler = proc_dointvec_minmax,
1062 .extra1 = SYSCTL_ZERO,
1063 .extra2 = SYSCTL_TWO,
1064 },
1065 {
1066 .procname = "protected_regular",
1067 .data = &sysctl_protected_regular,
1068 .maxlen = sizeof(int),
1069 .mode = 0644,
1070 .proc_handler = proc_dointvec_minmax,
1071 .extra1 = SYSCTL_ZERO,
1072 .extra2 = SYSCTL_TWO,
1073 },
1074};
1075
1076static int __init init_fs_namei_sysctls(void)
1077{
1078 register_sysctl_init("fs", namei_sysctls);
1079 return 0;
1080}
1081fs_initcall(init_fs_namei_sysctls);
1082
1083#endif /* CONFIG_SYSCTL */
1084
1085/**
1086 * may_follow_link - Check symlink following for unsafe situations
1087 * @nd: nameidata pathwalk data
1088 * @inode: Used for idmapping.
1089 *
1090 * In the case of the sysctl_protected_symlinks sysctl being enabled,
1091 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
1092 * in a sticky world-writable directory. This is to protect privileged
1093 * processes from failing races against path names that may change out
1094 * from under them by way of other users creating malicious symlinks.
1095 * It will permit symlinks to be followed only when outside a sticky
1096 * world-writable directory, or when the uid of the symlink and follower
1097 * match, or when the directory owner matches the symlink's owner.
1098 *
1099 * Returns 0 if following the symlink is allowed, -ve on error.
1100 */
1101static inline int may_follow_link(struct nameidata *nd, const struct inode *inode)
1102{
1103 struct mnt_idmap *idmap;
1104 vfsuid_t vfsuid;
1105
1106 if (!sysctl_protected_symlinks)
1107 return 0;
1108
1109 idmap = mnt_idmap(nd->path.mnt);
1110 vfsuid = i_uid_into_vfsuid(idmap, inode);
1111 /* Allowed if owner and follower match. */
1112 if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
1113 return 0;
1114
1115 /* Allowed if parent directory not sticky and world-writable. */
1116 if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
1117 return 0;
1118
1119 /* Allowed if parent directory and link owner match. */
1120 if (vfsuid_valid(nd->dir_vfsuid) && vfsuid_eq(nd->dir_vfsuid, vfsuid))
1121 return 0;
1122
1123 if (nd->flags & LOOKUP_RCU)
1124 return -ECHILD;
1125
1126 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
1127 audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link");
1128 return -EACCES;
1129}
1130
1131/**
1132 * safe_hardlink_source - Check for safe hardlink conditions
1133 * @idmap: idmap of the mount the inode was found from
1134 * @inode: the source inode to hardlink from
1135 *
1136 * Return false if at least one of the following conditions:
1137 * - inode is not a regular file
1138 * - inode is setuid
1139 * - inode is setgid and group-exec
1140 * - access failure for read and write
1141 *
1142 * Otherwise returns true.
1143 */
1144static bool safe_hardlink_source(struct mnt_idmap *idmap,
1145 struct inode *inode)
1146{
1147 umode_t mode = inode->i_mode;
1148
1149 /* Special files should not get pinned to the filesystem. */
1150 if (!S_ISREG(mode))
1151 return false;
1152
1153 /* Setuid files should not get pinned to the filesystem. */
1154 if (mode & S_ISUID)
1155 return false;
1156
1157 /* Executable setgid files should not get pinned to the filesystem. */
1158 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
1159 return false;
1160
1161 /* Hardlinking to unreadable or unwritable sources is dangerous. */
1162 if (inode_permission(idmap, inode, MAY_READ | MAY_WRITE))
1163 return false;
1164
1165 return true;
1166}
1167
1168/**
1169 * may_linkat - Check permissions for creating a hardlink
1170 * @idmap: idmap of the mount the inode was found from
1171 * @link: the source to hardlink from
1172 *
1173 * Block hardlink when all of:
1174 * - sysctl_protected_hardlinks enabled
1175 * - fsuid does not match inode
1176 * - hardlink source is unsafe (see safe_hardlink_source() above)
1177 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1178 *
1179 * If the inode has been found through an idmapped mount the idmap of
1180 * the vfsmount must be passed through @idmap. This function will then take
1181 * care to map the inode according to @idmap before checking permissions.
1182 * On non-idmapped mounts or if permission checking is to be performed on the
1183 * raw inode simply pass @nop_mnt_idmap.
1184 *
1185 * Returns 0 if successful, -ve on error.
1186 */
1187int may_linkat(struct mnt_idmap *idmap, const struct path *link)
1188{
1189 struct inode *inode = link->dentry->d_inode;
1190
1191 /* Inode writeback is not safe when the uid or gid are invalid. */
1192 if (!vfsuid_valid(i_uid_into_vfsuid(idmap, inode)) ||
1193 !vfsgid_valid(i_gid_into_vfsgid(idmap, inode)))
1194 return -EOVERFLOW;
1195
1196 if (!sysctl_protected_hardlinks)
1197 return 0;
1198
1199 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1200 * otherwise, it must be a safe source.
1201 */
1202 if (safe_hardlink_source(idmap, inode) ||
1203 inode_owner_or_capable(idmap, inode))
1204 return 0;
1205
1206 audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
1207 return -EPERM;
1208}
1209
1210/**
1211 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1212 * should be allowed, or not, on files that already
1213 * exist.
1214 * @idmap: idmap of the mount the inode was found from
1215 * @nd: nameidata pathwalk data
1216 * @inode: the inode of the file to open
1217 *
1218 * Block an O_CREAT open of a FIFO (or a regular file) when:
1219 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1220 * - the file already exists
1221 * - we are in a sticky directory
1222 * - we don't own the file
1223 * - the owner of the directory doesn't own the file
1224 * - the directory is world writable
1225 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1226 * the directory doesn't have to be world writable: being group writable will
1227 * be enough.
1228 *
1229 * If the inode has been found through an idmapped mount the idmap of
1230 * the vfsmount must be passed through @idmap. This function will then take
1231 * care to map the inode according to @idmap before checking permissions.
1232 * On non-idmapped mounts or if permission checking is to be performed on the
1233 * raw inode simply pass @nop_mnt_idmap.
1234 *
1235 * Returns 0 if the open is allowed, -ve on error.
1236 */
1237static int may_create_in_sticky(struct mnt_idmap *idmap,
1238 struct nameidata *nd, struct inode *const inode)
1239{
1240 umode_t dir_mode = nd->dir_mode;
1241 vfsuid_t dir_vfsuid = nd->dir_vfsuid;
1242
1243 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1244 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1245 likely(!(dir_mode & S_ISVTX)) ||
1246 vfsuid_eq(i_uid_into_vfsuid(idmap, inode), dir_vfsuid) ||
1247 vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode), current_fsuid()))
1248 return 0;
1249
1250 if (likely(dir_mode & 0002) ||
1251 (dir_mode & 0020 &&
1252 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1253 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1254 const char *operation = S_ISFIFO(inode->i_mode) ?
1255 "sticky_create_fifo" :
1256 "sticky_create_regular";
1257 audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
1258 return -EACCES;
1259 }
1260 return 0;
1261}
1262
1263/*
1264 * follow_up - Find the mountpoint of path's vfsmount
1265 *
1266 * Given a path, find the mountpoint of its source file system.
1267 * Replace @path with the path of the mountpoint in the parent mount.
1268 * Up is towards /.
1269 *
1270 * Return 1 if we went up a level and 0 if we were already at the
1271 * root.
1272 */
1273int follow_up(struct path *path)
1274{
1275 struct mount *mnt = real_mount(path->mnt);
1276 struct mount *parent;
1277 struct dentry *mountpoint;
1278
1279 read_seqlock_excl(&mount_lock);
1280 parent = mnt->mnt_parent;
1281 if (parent == mnt) {
1282 read_sequnlock_excl(&mount_lock);
1283 return 0;
1284 }
1285 mntget(&parent->mnt);
1286 mountpoint = dget(mnt->mnt_mountpoint);
1287 read_sequnlock_excl(&mount_lock);
1288 dput(path->dentry);
1289 path->dentry = mountpoint;
1290 mntput(path->mnt);
1291 path->mnt = &parent->mnt;
1292 return 1;
1293}
1294EXPORT_SYMBOL(follow_up);
1295
1296static bool choose_mountpoint_rcu(struct mount *m, const struct path *root,
1297 struct path *path, unsigned *seqp)
1298{
1299 while (mnt_has_parent(m)) {
1300 struct dentry *mountpoint = m->mnt_mountpoint;
1301
1302 m = m->mnt_parent;
1303 if (unlikely(root->dentry == mountpoint &&
1304 root->mnt == &m->mnt))
1305 break;
1306 if (mountpoint != m->mnt.mnt_root) {
1307 path->mnt = &m->mnt;
1308 path->dentry = mountpoint;
1309 *seqp = read_seqcount_begin(&mountpoint->d_seq);
1310 return true;
1311 }
1312 }
1313 return false;
1314}
1315
1316static bool choose_mountpoint(struct mount *m, const struct path *root,
1317 struct path *path)
1318{
1319 bool found;
1320
1321 rcu_read_lock();
1322 while (1) {
1323 unsigned seq, mseq = read_seqbegin(&mount_lock);
1324
1325 found = choose_mountpoint_rcu(m, root, path, &seq);
1326 if (unlikely(!found)) {
1327 if (!read_seqretry(&mount_lock, mseq))
1328 break;
1329 } else {
1330 if (likely(__legitimize_path(path, seq, mseq)))
1331 break;
1332 rcu_read_unlock();
1333 path_put(path);
1334 rcu_read_lock();
1335 }
1336 }
1337 rcu_read_unlock();
1338 return found;
1339}
1340
1341/*
1342 * Perform an automount
1343 * - return -EISDIR to tell follow_managed() to stop and return the path we
1344 * were called with.
1345 */
1346static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1347{
1348 struct dentry *dentry = path->dentry;
1349
1350 /* We don't want to mount if someone's just doing a stat -
1351 * unless they're stat'ing a directory and appended a '/' to
1352 * the name.
1353 *
1354 * We do, however, want to mount if someone wants to open or
1355 * create a file of any type under the mountpoint, wants to
1356 * traverse through the mountpoint or wants to open the
1357 * mounted directory. Also, autofs may mark negative dentries
1358 * as being automount points. These will need the attentions
1359 * of the daemon to instantiate them before they can be used.
1360 */
1361 if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1362 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1363 dentry->d_inode)
1364 return -EISDIR;
1365
1366 if (count && (*count)++ >= MAXSYMLINKS)
1367 return -ELOOP;
1368
1369 return finish_automount(dentry->d_op->d_automount(path), path);
1370}
1371
1372/*
1373 * mount traversal - out-of-line part. One note on ->d_flags accesses -
1374 * dentries are pinned but not locked here, so negative dentry can go
1375 * positive right under us. Use of smp_load_acquire() provides a barrier
1376 * sufficient for ->d_inode and ->d_flags consistency.
1377 */
1378static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped,
1379 int *count, unsigned lookup_flags)
1380{
1381 struct vfsmount *mnt = path->mnt;
1382 bool need_mntput = false;
1383 int ret = 0;
1384
1385 while (flags & DCACHE_MANAGED_DENTRY) {
1386 /* Allow the filesystem to manage the transit without i_mutex
1387 * being held. */
1388 if (flags & DCACHE_MANAGE_TRANSIT) {
1389 ret = path->dentry->d_op->d_manage(path, false);
1390 flags = smp_load_acquire(&path->dentry->d_flags);
1391 if (ret < 0)
1392 break;
1393 }
1394
1395 if (flags & DCACHE_MOUNTED) { // something's mounted on it..
1396 struct vfsmount *mounted = lookup_mnt(path);
1397 if (mounted) { // ... in our namespace
1398 dput(path->dentry);
1399 if (need_mntput)
1400 mntput(path->mnt);
1401 path->mnt = mounted;
1402 path->dentry = dget(mounted->mnt_root);
1403 // here we know it's positive
1404 flags = path->dentry->d_flags;
1405 need_mntput = true;
1406 continue;
1407 }
1408 }
1409
1410 if (!(flags & DCACHE_NEED_AUTOMOUNT))
1411 break;
1412
1413 // uncovered automount point
1414 ret = follow_automount(path, count, lookup_flags);
1415 flags = smp_load_acquire(&path->dentry->d_flags);
1416 if (ret < 0)
1417 break;
1418 }
1419
1420 if (ret == -EISDIR)
1421 ret = 0;
1422 // possible if you race with several mount --move
1423 if (need_mntput && path->mnt == mnt)
1424 mntput(path->mnt);
1425 if (!ret && unlikely(d_flags_negative(flags)))
1426 ret = -ENOENT;
1427 *jumped = need_mntput;
1428 return ret;
1429}
1430
1431static inline int traverse_mounts(struct path *path, bool *jumped,
1432 int *count, unsigned lookup_flags)
1433{
1434 unsigned flags = smp_load_acquire(&path->dentry->d_flags);
1435
1436 /* fastpath */
1437 if (likely(!(flags & DCACHE_MANAGED_DENTRY))) {
1438 *jumped = false;
1439 if (unlikely(d_flags_negative(flags)))
1440 return -ENOENT;
1441 return 0;
1442 }
1443 return __traverse_mounts(path, flags, jumped, count, lookup_flags);
1444}
1445
1446int follow_down_one(struct path *path)
1447{
1448 struct vfsmount *mounted;
1449
1450 mounted = lookup_mnt(path);
1451 if (mounted) {
1452 dput(path->dentry);
1453 mntput(path->mnt);
1454 path->mnt = mounted;
1455 path->dentry = dget(mounted->mnt_root);
1456 return 1;
1457 }
1458 return 0;
1459}
1460EXPORT_SYMBOL(follow_down_one);
1461
1462/*
1463 * Follow down to the covering mount currently visible to userspace. At each
1464 * point, the filesystem owning that dentry may be queried as to whether the
1465 * caller is permitted to proceed or not.
1466 */
1467int follow_down(struct path *path, unsigned int flags)
1468{
1469 struct vfsmount *mnt = path->mnt;
1470 bool jumped;
1471 int ret = traverse_mounts(path, &jumped, NULL, flags);
1472
1473 if (path->mnt != mnt)
1474 mntput(mnt);
1475 return ret;
1476}
1477EXPORT_SYMBOL(follow_down);
1478
1479/*
1480 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1481 * we meet a managed dentry that would need blocking.
1482 */
1483static bool __follow_mount_rcu(struct nameidata *nd, struct path *path)
1484{
1485 struct dentry *dentry = path->dentry;
1486 unsigned int flags = dentry->d_flags;
1487
1488 if (likely(!(flags & DCACHE_MANAGED_DENTRY)))
1489 return true;
1490
1491 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1492 return false;
1493
1494 for (;;) {
1495 /*
1496 * Don't forget we might have a non-mountpoint managed dentry
1497 * that wants to block transit.
1498 */
1499 if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) {
1500 int res = dentry->d_op->d_manage(path, true);
1501 if (res)
1502 return res == -EISDIR;
1503 flags = dentry->d_flags;
1504 }
1505
1506 if (flags & DCACHE_MOUNTED) {
1507 struct mount *mounted = __lookup_mnt(path->mnt, dentry);
1508 if (mounted) {
1509 path->mnt = &mounted->mnt;
1510 dentry = path->dentry = mounted->mnt.mnt_root;
1511 nd->state |= ND_JUMPED;
1512 nd->next_seq = read_seqcount_begin(&dentry->d_seq);
1513 flags = dentry->d_flags;
1514 // makes sure that non-RCU pathwalk could reach
1515 // this state.
1516 if (read_seqretry(&mount_lock, nd->m_seq))
1517 return false;
1518 continue;
1519 }
1520 if (read_seqretry(&mount_lock, nd->m_seq))
1521 return false;
1522 }
1523 return !(flags & DCACHE_NEED_AUTOMOUNT);
1524 }
1525}
1526
1527static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1528 struct path *path)
1529{
1530 bool jumped;
1531 int ret;
1532
1533 path->mnt = nd->path.mnt;
1534 path->dentry = dentry;
1535 if (nd->flags & LOOKUP_RCU) {
1536 unsigned int seq = nd->next_seq;
1537 if (likely(__follow_mount_rcu(nd, path)))
1538 return 0;
1539 // *path and nd->next_seq might've been clobbered
1540 path->mnt = nd->path.mnt;
1541 path->dentry = dentry;
1542 nd->next_seq = seq;
1543 if (!try_to_unlazy_next(nd, dentry))
1544 return -ECHILD;
1545 }
1546 ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags);
1547 if (jumped) {
1548 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1549 ret = -EXDEV;
1550 else
1551 nd->state |= ND_JUMPED;
1552 }
1553 if (unlikely(ret)) {
1554 dput(path->dentry);
1555 if (path->mnt != nd->path.mnt)
1556 mntput(path->mnt);
1557 }
1558 return ret;
1559}
1560
1561/*
1562 * This looks up the name in dcache and possibly revalidates the found dentry.
1563 * NULL is returned if the dentry does not exist in the cache.
1564 */
1565static struct dentry *lookup_dcache(const struct qstr *name,
1566 struct dentry *dir,
1567 unsigned int flags)
1568{
1569 struct dentry *dentry = d_lookup(dir, name);
1570 if (dentry) {
1571 int error = d_revalidate(dentry, flags);
1572 if (unlikely(error <= 0)) {
1573 if (!error)
1574 d_invalidate(dentry);
1575 dput(dentry);
1576 return ERR_PTR(error);
1577 }
1578 }
1579 return dentry;
1580}
1581
1582/*
1583 * Parent directory has inode locked exclusive. This is one
1584 * and only case when ->lookup() gets called on non in-lookup
1585 * dentries - as the matter of fact, this only gets called
1586 * when directory is guaranteed to have no in-lookup children
1587 * at all.
1588 */
1589struct dentry *lookup_one_qstr_excl(const struct qstr *name,
1590 struct dentry *base,
1591 unsigned int flags)
1592{
1593 struct dentry *dentry = lookup_dcache(name, base, flags);
1594 struct dentry *old;
1595 struct inode *dir = base->d_inode;
1596
1597 if (dentry)
1598 return dentry;
1599
1600 /* Don't create child dentry for a dead directory. */
1601 if (unlikely(IS_DEADDIR(dir)))
1602 return ERR_PTR(-ENOENT);
1603
1604 dentry = d_alloc(base, name);
1605 if (unlikely(!dentry))
1606 return ERR_PTR(-ENOMEM);
1607
1608 old = dir->i_op->lookup(dir, dentry, flags);
1609 if (unlikely(old)) {
1610 dput(dentry);
1611 dentry = old;
1612 }
1613 return dentry;
1614}
1615EXPORT_SYMBOL(lookup_one_qstr_excl);
1616
1617static struct dentry *lookup_fast(struct nameidata *nd)
1618{
1619 struct dentry *dentry, *parent = nd->path.dentry;
1620 int status = 1;
1621
1622 /*
1623 * Rename seqlock is not required here because in the off chance
1624 * of a false negative due to a concurrent rename, the caller is
1625 * going to fall back to non-racy lookup.
1626 */
1627 if (nd->flags & LOOKUP_RCU) {
1628 dentry = __d_lookup_rcu(parent, &nd->last, &nd->next_seq);
1629 if (unlikely(!dentry)) {
1630 if (!try_to_unlazy(nd))
1631 return ERR_PTR(-ECHILD);
1632 return NULL;
1633 }
1634
1635 /*
1636 * This sequence count validates that the parent had no
1637 * changes while we did the lookup of the dentry above.
1638 */
1639 if (read_seqcount_retry(&parent->d_seq, nd->seq))
1640 return ERR_PTR(-ECHILD);
1641
1642 status = d_revalidate(dentry, nd->flags);
1643 if (likely(status > 0))
1644 return dentry;
1645 if (!try_to_unlazy_next(nd, dentry))
1646 return ERR_PTR(-ECHILD);
1647 if (status == -ECHILD)
1648 /* we'd been told to redo it in non-rcu mode */
1649 status = d_revalidate(dentry, nd->flags);
1650 } else {
1651 dentry = __d_lookup(parent, &nd->last);
1652 if (unlikely(!dentry))
1653 return NULL;
1654 status = d_revalidate(dentry, nd->flags);
1655 }
1656 if (unlikely(status <= 0)) {
1657 if (!status)
1658 d_invalidate(dentry);
1659 dput(dentry);
1660 return ERR_PTR(status);
1661 }
1662 return dentry;
1663}
1664
1665/* Fast lookup failed, do it the slow way */
1666static struct dentry *__lookup_slow(const struct qstr *name,
1667 struct dentry *dir,
1668 unsigned int flags)
1669{
1670 struct dentry *dentry, *old;
1671 struct inode *inode = dir->d_inode;
1672 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1673
1674 /* Don't go there if it's already dead */
1675 if (unlikely(IS_DEADDIR(inode)))
1676 return ERR_PTR(-ENOENT);
1677again:
1678 dentry = d_alloc_parallel(dir, name, &wq);
1679 if (IS_ERR(dentry))
1680 return dentry;
1681 if (unlikely(!d_in_lookup(dentry))) {
1682 int error = d_revalidate(dentry, flags);
1683 if (unlikely(error <= 0)) {
1684 if (!error) {
1685 d_invalidate(dentry);
1686 dput(dentry);
1687 goto again;
1688 }
1689 dput(dentry);
1690 dentry = ERR_PTR(error);
1691 }
1692 } else {
1693 old = inode->i_op->lookup(inode, dentry, flags);
1694 d_lookup_done(dentry);
1695 if (unlikely(old)) {
1696 dput(dentry);
1697 dentry = old;
1698 }
1699 }
1700 return dentry;
1701}
1702
1703static struct dentry *lookup_slow(const struct qstr *name,
1704 struct dentry *dir,
1705 unsigned int flags)
1706{
1707 struct inode *inode = dir->d_inode;
1708 struct dentry *res;
1709 inode_lock_shared(inode);
1710 res = __lookup_slow(name, dir, flags);
1711 inode_unlock_shared(inode);
1712 return res;
1713}
1714
1715static inline int may_lookup(struct mnt_idmap *idmap,
1716 struct nameidata *nd)
1717{
1718 if (nd->flags & LOOKUP_RCU) {
1719 int err = inode_permission(idmap, nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1720 if (!err) // success, keep going
1721 return 0;
1722 if (!try_to_unlazy(nd))
1723 return -ECHILD; // redo it all non-lazy
1724 if (err != -ECHILD) // hard error
1725 return err;
1726 }
1727 return inode_permission(idmap, nd->inode, MAY_EXEC);
1728}
1729
1730static int reserve_stack(struct nameidata *nd, struct path *link)
1731{
1732 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1733 return -ELOOP;
1734
1735 if (likely(nd->depth != EMBEDDED_LEVELS))
1736 return 0;
1737 if (likely(nd->stack != nd->internal))
1738 return 0;
1739 if (likely(nd_alloc_stack(nd)))
1740 return 0;
1741
1742 if (nd->flags & LOOKUP_RCU) {
1743 // we need to grab link before we do unlazy. And we can't skip
1744 // unlazy even if we fail to grab the link - cleanup needs it
1745 bool grabbed_link = legitimize_path(nd, link, nd->next_seq);
1746
1747 if (!try_to_unlazy(nd) || !grabbed_link)
1748 return -ECHILD;
1749
1750 if (nd_alloc_stack(nd))
1751 return 0;
1752 }
1753 return -ENOMEM;
1754}
1755
1756enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1757
1758static const char *pick_link(struct nameidata *nd, struct path *link,
1759 struct inode *inode, int flags)
1760{
1761 struct saved *last;
1762 const char *res;
1763 int error = reserve_stack(nd, link);
1764
1765 if (unlikely(error)) {
1766 if (!(nd->flags & LOOKUP_RCU))
1767 path_put(link);
1768 return ERR_PTR(error);
1769 }
1770 last = nd->stack + nd->depth++;
1771 last->link = *link;
1772 clear_delayed_call(&last->done);
1773 last->seq = nd->next_seq;
1774
1775 if (flags & WALK_TRAILING) {
1776 error = may_follow_link(nd, inode);
1777 if (unlikely(error))
1778 return ERR_PTR(error);
1779 }
1780
1781 if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) ||
1782 unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW))
1783 return ERR_PTR(-ELOOP);
1784
1785 if (!(nd->flags & LOOKUP_RCU)) {
1786 touch_atime(&last->link);
1787 cond_resched();
1788 } else if (atime_needs_update(&last->link, inode)) {
1789 if (!try_to_unlazy(nd))
1790 return ERR_PTR(-ECHILD);
1791 touch_atime(&last->link);
1792 }
1793
1794 error = security_inode_follow_link(link->dentry, inode,
1795 nd->flags & LOOKUP_RCU);
1796 if (unlikely(error))
1797 return ERR_PTR(error);
1798
1799 res = READ_ONCE(inode->i_link);
1800 if (!res) {
1801 const char * (*get)(struct dentry *, struct inode *,
1802 struct delayed_call *);
1803 get = inode->i_op->get_link;
1804 if (nd->flags & LOOKUP_RCU) {
1805 res = get(NULL, inode, &last->done);
1806 if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd))
1807 res = get(link->dentry, inode, &last->done);
1808 } else {
1809 res = get(link->dentry, inode, &last->done);
1810 }
1811 if (!res)
1812 goto all_done;
1813 if (IS_ERR(res))
1814 return res;
1815 }
1816 if (*res == '/') {
1817 error = nd_jump_root(nd);
1818 if (unlikely(error))
1819 return ERR_PTR(error);
1820 while (unlikely(*++res == '/'))
1821 ;
1822 }
1823 if (*res)
1824 return res;
1825all_done: // pure jump
1826 put_link(nd);
1827 return NULL;
1828}
1829
1830/*
1831 * Do we need to follow links? We _really_ want to be able
1832 * to do this check without having to look at inode->i_op,
1833 * so we keep a cache of "no, this doesn't need follow_link"
1834 * for the common case.
1835 *
1836 * NOTE: dentry must be what nd->next_seq had been sampled from.
1837 */
1838static const char *step_into(struct nameidata *nd, int flags,
1839 struct dentry *dentry)
1840{
1841 struct path path;
1842 struct inode *inode;
1843 int err = handle_mounts(nd, dentry, &path);
1844
1845 if (err < 0)
1846 return ERR_PTR(err);
1847 inode = path.dentry->d_inode;
1848 if (likely(!d_is_symlink(path.dentry)) ||
1849 ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
1850 (flags & WALK_NOFOLLOW)) {
1851 /* not a symlink or should not follow */
1852 if (nd->flags & LOOKUP_RCU) {
1853 if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1854 return ERR_PTR(-ECHILD);
1855 if (unlikely(!inode))
1856 return ERR_PTR(-ENOENT);
1857 } else {
1858 dput(nd->path.dentry);
1859 if (nd->path.mnt != path.mnt)
1860 mntput(nd->path.mnt);
1861 }
1862 nd->path = path;
1863 nd->inode = inode;
1864 nd->seq = nd->next_seq;
1865 return NULL;
1866 }
1867 if (nd->flags & LOOKUP_RCU) {
1868 /* make sure that d_is_symlink above matches inode */
1869 if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1870 return ERR_PTR(-ECHILD);
1871 } else {
1872 if (path.mnt == nd->path.mnt)
1873 mntget(path.mnt);
1874 }
1875 return pick_link(nd, &path, inode, flags);
1876}
1877
1878static struct dentry *follow_dotdot_rcu(struct nameidata *nd)
1879{
1880 struct dentry *parent, *old;
1881
1882 if (path_equal(&nd->path, &nd->root))
1883 goto in_root;
1884 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1885 struct path path;
1886 unsigned seq;
1887 if (!choose_mountpoint_rcu(real_mount(nd->path.mnt),
1888 &nd->root, &path, &seq))
1889 goto in_root;
1890 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1891 return ERR_PTR(-ECHILD);
1892 nd->path = path;
1893 nd->inode = path.dentry->d_inode;
1894 nd->seq = seq;
1895 // makes sure that non-RCU pathwalk could reach this state
1896 if (read_seqretry(&mount_lock, nd->m_seq))
1897 return ERR_PTR(-ECHILD);
1898 /* we know that mountpoint was pinned */
1899 }
1900 old = nd->path.dentry;
1901 parent = old->d_parent;
1902 nd->next_seq = read_seqcount_begin(&parent->d_seq);
1903 // makes sure that non-RCU pathwalk could reach this state
1904 if (read_seqcount_retry(&old->d_seq, nd->seq))
1905 return ERR_PTR(-ECHILD);
1906 if (unlikely(!path_connected(nd->path.mnt, parent)))
1907 return ERR_PTR(-ECHILD);
1908 return parent;
1909in_root:
1910 if (read_seqretry(&mount_lock, nd->m_seq))
1911 return ERR_PTR(-ECHILD);
1912 if (unlikely(nd->flags & LOOKUP_BENEATH))
1913 return ERR_PTR(-ECHILD);
1914 nd->next_seq = nd->seq;
1915 return nd->path.dentry;
1916}
1917
1918static struct dentry *follow_dotdot(struct nameidata *nd)
1919{
1920 struct dentry *parent;
1921
1922 if (path_equal(&nd->path, &nd->root))
1923 goto in_root;
1924 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1925 struct path path;
1926
1927 if (!choose_mountpoint(real_mount(nd->path.mnt),
1928 &nd->root, &path))
1929 goto in_root;
1930 path_put(&nd->path);
1931 nd->path = path;
1932 nd->inode = path.dentry->d_inode;
1933 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1934 return ERR_PTR(-EXDEV);
1935 }
1936 /* rare case of legitimate dget_parent()... */
1937 parent = dget_parent(nd->path.dentry);
1938 if (unlikely(!path_connected(nd->path.mnt, parent))) {
1939 dput(parent);
1940 return ERR_PTR(-ENOENT);
1941 }
1942 return parent;
1943
1944in_root:
1945 if (unlikely(nd->flags & LOOKUP_BENEATH))
1946 return ERR_PTR(-EXDEV);
1947 return dget(nd->path.dentry);
1948}
1949
1950static const char *handle_dots(struct nameidata *nd, int type)
1951{
1952 if (type == LAST_DOTDOT) {
1953 const char *error = NULL;
1954 struct dentry *parent;
1955
1956 if (!nd->root.mnt) {
1957 error = ERR_PTR(set_root(nd));
1958 if (error)
1959 return error;
1960 }
1961 if (nd->flags & LOOKUP_RCU)
1962 parent = follow_dotdot_rcu(nd);
1963 else
1964 parent = follow_dotdot(nd);
1965 if (IS_ERR(parent))
1966 return ERR_CAST(parent);
1967 error = step_into(nd, WALK_NOFOLLOW, parent);
1968 if (unlikely(error))
1969 return error;
1970
1971 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1972 /*
1973 * If there was a racing rename or mount along our
1974 * path, then we can't be sure that ".." hasn't jumped
1975 * above nd->root (and so userspace should retry or use
1976 * some fallback).
1977 */
1978 smp_rmb();
1979 if (__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))
1980 return ERR_PTR(-EAGAIN);
1981 if (__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))
1982 return ERR_PTR(-EAGAIN);
1983 }
1984 }
1985 return NULL;
1986}
1987
1988static const char *walk_component(struct nameidata *nd, int flags)
1989{
1990 struct dentry *dentry;
1991 /*
1992 * "." and ".." are special - ".." especially so because it has
1993 * to be able to know about the current root directory and
1994 * parent relationships.
1995 */
1996 if (unlikely(nd->last_type != LAST_NORM)) {
1997 if (!(flags & WALK_MORE) && nd->depth)
1998 put_link(nd);
1999 return handle_dots(nd, nd->last_type);
2000 }
2001 dentry = lookup_fast(nd);
2002 if (IS_ERR(dentry))
2003 return ERR_CAST(dentry);
2004 if (unlikely(!dentry)) {
2005 dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
2006 if (IS_ERR(dentry))
2007 return ERR_CAST(dentry);
2008 }
2009 if (!(flags & WALK_MORE) && nd->depth)
2010 put_link(nd);
2011 return step_into(nd, flags, dentry);
2012}
2013
2014/*
2015 * We can do the critical dentry name comparison and hashing
2016 * operations one word at a time, but we are limited to:
2017 *
2018 * - Architectures with fast unaligned word accesses. We could
2019 * do a "get_unaligned()" if this helps and is sufficiently
2020 * fast.
2021 *
2022 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
2023 * do not trap on the (extremely unlikely) case of a page
2024 * crossing operation.
2025 *
2026 * - Furthermore, we need an efficient 64-bit compile for the
2027 * 64-bit case in order to generate the "number of bytes in
2028 * the final mask". Again, that could be replaced with a
2029 * efficient population count instruction or similar.
2030 */
2031#ifdef CONFIG_DCACHE_WORD_ACCESS
2032
2033#include <asm/word-at-a-time.h>
2034
2035#ifdef HASH_MIX
2036
2037/* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
2038
2039#elif defined(CONFIG_64BIT)
2040/*
2041 * Register pressure in the mixing function is an issue, particularly
2042 * on 32-bit x86, but almost any function requires one state value and
2043 * one temporary. Instead, use a function designed for two state values
2044 * and no temporaries.
2045 *
2046 * This function cannot create a collision in only two iterations, so
2047 * we have two iterations to achieve avalanche. In those two iterations,
2048 * we have six layers of mixing, which is enough to spread one bit's
2049 * influence out to 2^6 = 64 state bits.
2050 *
2051 * Rotate constants are scored by considering either 64 one-bit input
2052 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
2053 * probability of that delta causing a change to each of the 128 output
2054 * bits, using a sample of random initial states.
2055 *
2056 * The Shannon entropy of the computed probabilities is then summed
2057 * to produce a score. Ideally, any input change has a 50% chance of
2058 * toggling any given output bit.
2059 *
2060 * Mixing scores (in bits) for (12,45):
2061 * Input delta: 1-bit 2-bit
2062 * 1 round: 713.3 42542.6
2063 * 2 rounds: 2753.7 140389.8
2064 * 3 rounds: 5954.1 233458.2
2065 * 4 rounds: 7862.6 256672.2
2066 * Perfect: 8192 258048
2067 * (64*128) (64*63/2 * 128)
2068 */
2069#define HASH_MIX(x, y, a) \
2070 ( x ^= (a), \
2071 y ^= x, x = rol64(x,12),\
2072 x += y, y = rol64(y,45),\
2073 y *= 9 )
2074
2075/*
2076 * Fold two longs into one 32-bit hash value. This must be fast, but
2077 * latency isn't quite as critical, as there is a fair bit of additional
2078 * work done before the hash value is used.
2079 */
2080static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2081{
2082 y ^= x * GOLDEN_RATIO_64;
2083 y *= GOLDEN_RATIO_64;
2084 return y >> 32;
2085}
2086
2087#else /* 32-bit case */
2088
2089/*
2090 * Mixing scores (in bits) for (7,20):
2091 * Input delta: 1-bit 2-bit
2092 * 1 round: 330.3 9201.6
2093 * 2 rounds: 1246.4 25475.4
2094 * 3 rounds: 1907.1 31295.1
2095 * 4 rounds: 2042.3 31718.6
2096 * Perfect: 2048 31744
2097 * (32*64) (32*31/2 * 64)
2098 */
2099#define HASH_MIX(x, y, a) \
2100 ( x ^= (a), \
2101 y ^= x, x = rol32(x, 7),\
2102 x += y, y = rol32(y,20),\
2103 y *= 9 )
2104
2105static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2106{
2107 /* Use arch-optimized multiply if one exists */
2108 return __hash_32(y ^ __hash_32(x));
2109}
2110
2111#endif
2112
2113/*
2114 * Return the hash of a string of known length. This is carfully
2115 * designed to match hash_name(), which is the more critical function.
2116 * In particular, we must end by hashing a final word containing 0..7
2117 * payload bytes, to match the way that hash_name() iterates until it
2118 * finds the delimiter after the name.
2119 */
2120unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2121{
2122 unsigned long a, x = 0, y = (unsigned long)salt;
2123
2124 for (;;) {
2125 if (!len)
2126 goto done;
2127 a = load_unaligned_zeropad(name);
2128 if (len < sizeof(unsigned long))
2129 break;
2130 HASH_MIX(x, y, a);
2131 name += sizeof(unsigned long);
2132 len -= sizeof(unsigned long);
2133 }
2134 x ^= a & bytemask_from_count(len);
2135done:
2136 return fold_hash(x, y);
2137}
2138EXPORT_SYMBOL(full_name_hash);
2139
2140/* Return the "hash_len" (hash and length) of a null-terminated string */
2141u64 hashlen_string(const void *salt, const char *name)
2142{
2143 unsigned long a = 0, x = 0, y = (unsigned long)salt;
2144 unsigned long adata, mask, len;
2145 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2146
2147 len = 0;
2148 goto inside;
2149
2150 do {
2151 HASH_MIX(x, y, a);
2152 len += sizeof(unsigned long);
2153inside:
2154 a = load_unaligned_zeropad(name+len);
2155 } while (!has_zero(a, &adata, &constants));
2156
2157 adata = prep_zero_mask(a, adata, &constants);
2158 mask = create_zero_mask(adata);
2159 x ^= a & zero_bytemask(mask);
2160
2161 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2162}
2163EXPORT_SYMBOL(hashlen_string);
2164
2165/*
2166 * Calculate the length and hash of the path component, and
2167 * return the "hash_len" as the result.
2168 */
2169static inline u64 hash_name(const void *salt, const char *name)
2170{
2171 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2172 unsigned long adata, bdata, mask, len;
2173 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2174
2175 len = 0;
2176 goto inside;
2177
2178 do {
2179 HASH_MIX(x, y, a);
2180 len += sizeof(unsigned long);
2181inside:
2182 a = load_unaligned_zeropad(name+len);
2183 b = a ^ REPEAT_BYTE('/');
2184 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2185
2186 adata = prep_zero_mask(a, adata, &constants);
2187 bdata = prep_zero_mask(b, bdata, &constants);
2188 mask = create_zero_mask(adata | bdata);
2189 x ^= a & zero_bytemask(mask);
2190
2191 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2192}
2193
2194#else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2195
2196/* Return the hash of a string of known length */
2197unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2198{
2199 unsigned long hash = init_name_hash(salt);
2200 while (len--)
2201 hash = partial_name_hash((unsigned char)*name++, hash);
2202 return end_name_hash(hash);
2203}
2204EXPORT_SYMBOL(full_name_hash);
2205
2206/* Return the "hash_len" (hash and length) of a null-terminated string */
2207u64 hashlen_string(const void *salt, const char *name)
2208{
2209 unsigned long hash = init_name_hash(salt);
2210 unsigned long len = 0, c;
2211
2212 c = (unsigned char)*name;
2213 while (c) {
2214 len++;
2215 hash = partial_name_hash(c, hash);
2216 c = (unsigned char)name[len];
2217 }
2218 return hashlen_create(end_name_hash(hash), len);
2219}
2220EXPORT_SYMBOL(hashlen_string);
2221
2222/*
2223 * We know there's a real path component here of at least
2224 * one character.
2225 */
2226static inline u64 hash_name(const void *salt, const char *name)
2227{
2228 unsigned long hash = init_name_hash(salt);
2229 unsigned long len = 0, c;
2230
2231 c = (unsigned char)*name;
2232 do {
2233 len++;
2234 hash = partial_name_hash(c, hash);
2235 c = (unsigned char)name[len];
2236 } while (c && c != '/');
2237 return hashlen_create(end_name_hash(hash), len);
2238}
2239
2240#endif
2241
2242/*
2243 * Name resolution.
2244 * This is the basic name resolution function, turning a pathname into
2245 * the final dentry. We expect 'base' to be positive and a directory.
2246 *
2247 * Returns 0 and nd will have valid dentry and mnt on success.
2248 * Returns error and drops reference to input namei data on failure.
2249 */
2250static int link_path_walk(const char *name, struct nameidata *nd)
2251{
2252 int depth = 0; // depth <= nd->depth
2253 int err;
2254
2255 nd->last_type = LAST_ROOT;
2256 nd->flags |= LOOKUP_PARENT;
2257 if (IS_ERR(name))
2258 return PTR_ERR(name);
2259 while (*name=='/')
2260 name++;
2261 if (!*name) {
2262 nd->dir_mode = 0; // short-circuit the 'hardening' idiocy
2263 return 0;
2264 }
2265
2266 /* At this point we know we have a real path component. */
2267 for(;;) {
2268 struct mnt_idmap *idmap;
2269 const char *link;
2270 u64 hash_len;
2271 int type;
2272
2273 idmap = mnt_idmap(nd->path.mnt);
2274 err = may_lookup(idmap, nd);
2275 if (err)
2276 return err;
2277
2278 hash_len = hash_name(nd->path.dentry, name);
2279
2280 type = LAST_NORM;
2281 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2282 case 2:
2283 if (name[1] == '.') {
2284 type = LAST_DOTDOT;
2285 nd->state |= ND_JUMPED;
2286 }
2287 break;
2288 case 1:
2289 type = LAST_DOT;
2290 }
2291 if (likely(type == LAST_NORM)) {
2292 struct dentry *parent = nd->path.dentry;
2293 nd->state &= ~ND_JUMPED;
2294 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2295 struct qstr this = { { .hash_len = hash_len }, .name = name };
2296 err = parent->d_op->d_hash(parent, &this);
2297 if (err < 0)
2298 return err;
2299 hash_len = this.hash_len;
2300 name = this.name;
2301 }
2302 }
2303
2304 nd->last.hash_len = hash_len;
2305 nd->last.name = name;
2306 nd->last_type = type;
2307
2308 name += hashlen_len(hash_len);
2309 if (!*name)
2310 goto OK;
2311 /*
2312 * If it wasn't NUL, we know it was '/'. Skip that
2313 * slash, and continue until no more slashes.
2314 */
2315 do {
2316 name++;
2317 } while (unlikely(*name == '/'));
2318 if (unlikely(!*name)) {
2319OK:
2320 /* pathname or trailing symlink, done */
2321 if (!depth) {
2322 nd->dir_vfsuid = i_uid_into_vfsuid(idmap, nd->inode);
2323 nd->dir_mode = nd->inode->i_mode;
2324 nd->flags &= ~LOOKUP_PARENT;
2325 return 0;
2326 }
2327 /* last component of nested symlink */
2328 name = nd->stack[--depth].name;
2329 link = walk_component(nd, 0);
2330 } else {
2331 /* not the last component */
2332 link = walk_component(nd, WALK_MORE);
2333 }
2334 if (unlikely(link)) {
2335 if (IS_ERR(link))
2336 return PTR_ERR(link);
2337 /* a symlink to follow */
2338 nd->stack[depth++].name = name;
2339 name = link;
2340 continue;
2341 }
2342 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2343 if (nd->flags & LOOKUP_RCU) {
2344 if (!try_to_unlazy(nd))
2345 return -ECHILD;
2346 }
2347 return -ENOTDIR;
2348 }
2349 }
2350}
2351
2352/* must be paired with terminate_walk() */
2353static const char *path_init(struct nameidata *nd, unsigned flags)
2354{
2355 int error;
2356 const char *s = nd->name->name;
2357
2358 /* LOOKUP_CACHED requires RCU, ask caller to retry */
2359 if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED)
2360 return ERR_PTR(-EAGAIN);
2361
2362 if (!*s)
2363 flags &= ~LOOKUP_RCU;
2364 if (flags & LOOKUP_RCU)
2365 rcu_read_lock();
2366 else
2367 nd->seq = nd->next_seq = 0;
2368
2369 nd->flags = flags;
2370 nd->state |= ND_JUMPED;
2371
2372 nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2373 nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2374 smp_rmb();
2375
2376 if (nd->state & ND_ROOT_PRESET) {
2377 struct dentry *root = nd->root.dentry;
2378 struct inode *inode = root->d_inode;
2379 if (*s && unlikely(!d_can_lookup(root)))
2380 return ERR_PTR(-ENOTDIR);
2381 nd->path = nd->root;
2382 nd->inode = inode;
2383 if (flags & LOOKUP_RCU) {
2384 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2385 nd->root_seq = nd->seq;
2386 } else {
2387 path_get(&nd->path);
2388 }
2389 return s;
2390 }
2391
2392 nd->root.mnt = NULL;
2393
2394 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2395 if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2396 error = nd_jump_root(nd);
2397 if (unlikely(error))
2398 return ERR_PTR(error);
2399 return s;
2400 }
2401
2402 /* Relative pathname -- get the starting-point it is relative to. */
2403 if (nd->dfd == AT_FDCWD) {
2404 if (flags & LOOKUP_RCU) {
2405 struct fs_struct *fs = current->fs;
2406 unsigned seq;
2407
2408 do {
2409 seq = read_seqcount_begin(&fs->seq);
2410 nd->path = fs->pwd;
2411 nd->inode = nd->path.dentry->d_inode;
2412 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2413 } while (read_seqcount_retry(&fs->seq, seq));
2414 } else {
2415 get_fs_pwd(current->fs, &nd->path);
2416 nd->inode = nd->path.dentry->d_inode;
2417 }
2418 } else {
2419 /* Caller must check execute permissions on the starting path component */
2420 struct fd f = fdget_raw(nd->dfd);
2421 struct dentry *dentry;
2422
2423 if (!f.file)
2424 return ERR_PTR(-EBADF);
2425
2426 dentry = f.file->f_path.dentry;
2427
2428 if (*s && unlikely(!d_can_lookup(dentry))) {
2429 fdput(f);
2430 return ERR_PTR(-ENOTDIR);
2431 }
2432
2433 nd->path = f.file->f_path;
2434 if (flags & LOOKUP_RCU) {
2435 nd->inode = nd->path.dentry->d_inode;
2436 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2437 } else {
2438 path_get(&nd->path);
2439 nd->inode = nd->path.dentry->d_inode;
2440 }
2441 fdput(f);
2442 }
2443
2444 /* For scoped-lookups we need to set the root to the dirfd as well. */
2445 if (flags & LOOKUP_IS_SCOPED) {
2446 nd->root = nd->path;
2447 if (flags & LOOKUP_RCU) {
2448 nd->root_seq = nd->seq;
2449 } else {
2450 path_get(&nd->root);
2451 nd->state |= ND_ROOT_GRABBED;
2452 }
2453 }
2454 return s;
2455}
2456
2457static inline const char *lookup_last(struct nameidata *nd)
2458{
2459 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2460 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2461
2462 return walk_component(nd, WALK_TRAILING);
2463}
2464
2465static int handle_lookup_down(struct nameidata *nd)
2466{
2467 if (!(nd->flags & LOOKUP_RCU))
2468 dget(nd->path.dentry);
2469 nd->next_seq = nd->seq;
2470 return PTR_ERR(step_into(nd, WALK_NOFOLLOW, nd->path.dentry));
2471}
2472
2473/* Returns 0 and nd will be valid on success; Returns error, otherwise. */
2474static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2475{
2476 const char *s = path_init(nd, flags);
2477 int err;
2478
2479 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2480 err = handle_lookup_down(nd);
2481 if (unlikely(err < 0))
2482 s = ERR_PTR(err);
2483 }
2484
2485 while (!(err = link_path_walk(s, nd)) &&
2486 (s = lookup_last(nd)) != NULL)
2487 ;
2488 if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2489 err = handle_lookup_down(nd);
2490 nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please...
2491 }
2492 if (!err)
2493 err = complete_walk(nd);
2494
2495 if (!err && nd->flags & LOOKUP_DIRECTORY)
2496 if (!d_can_lookup(nd->path.dentry))
2497 err = -ENOTDIR;
2498 if (!err) {
2499 *path = nd->path;
2500 nd->path.mnt = NULL;
2501 nd->path.dentry = NULL;
2502 }
2503 terminate_walk(nd);
2504 return err;
2505}
2506
2507int filename_lookup(int dfd, struct filename *name, unsigned flags,
2508 struct path *path, struct path *root)
2509{
2510 int retval;
2511 struct nameidata nd;
2512 if (IS_ERR(name))
2513 return PTR_ERR(name);
2514 set_nameidata(&nd, dfd, name, root);
2515 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2516 if (unlikely(retval == -ECHILD))
2517 retval = path_lookupat(&nd, flags, path);
2518 if (unlikely(retval == -ESTALE))
2519 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2520
2521 if (likely(!retval))
2522 audit_inode(name, path->dentry,
2523 flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2524 restore_nameidata();
2525 return retval;
2526}
2527
2528/* Returns 0 and nd will be valid on success; Returns error, otherwise. */
2529static int path_parentat(struct nameidata *nd, unsigned flags,
2530 struct path *parent)
2531{
2532 const char *s = path_init(nd, flags);
2533 int err = link_path_walk(s, nd);
2534 if (!err)
2535 err = complete_walk(nd);
2536 if (!err) {
2537 *parent = nd->path;
2538 nd->path.mnt = NULL;
2539 nd->path.dentry = NULL;
2540 }
2541 terminate_walk(nd);
2542 return err;
2543}
2544
2545/* Note: this does not consume "name" */
2546static int __filename_parentat(int dfd, struct filename *name,
2547 unsigned int flags, struct path *parent,
2548 struct qstr *last, int *type,
2549 const struct path *root)
2550{
2551 int retval;
2552 struct nameidata nd;
2553
2554 if (IS_ERR(name))
2555 return PTR_ERR(name);
2556 set_nameidata(&nd, dfd, name, root);
2557 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2558 if (unlikely(retval == -ECHILD))
2559 retval = path_parentat(&nd, flags, parent);
2560 if (unlikely(retval == -ESTALE))
2561 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2562 if (likely(!retval)) {
2563 *last = nd.last;
2564 *type = nd.last_type;
2565 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2566 }
2567 restore_nameidata();
2568 return retval;
2569}
2570
2571static int filename_parentat(int dfd, struct filename *name,
2572 unsigned int flags, struct path *parent,
2573 struct qstr *last, int *type)
2574{
2575 return __filename_parentat(dfd, name, flags, parent, last, type, NULL);
2576}
2577
2578/* does lookup, returns the object with parent locked */
2579static struct dentry *__kern_path_locked(int dfd, struct filename *name, struct path *path)
2580{
2581 struct dentry *d;
2582 struct qstr last;
2583 int type, error;
2584
2585 error = filename_parentat(dfd, name, 0, path, &last, &type);
2586 if (error)
2587 return ERR_PTR(error);
2588 if (unlikely(type != LAST_NORM)) {
2589 path_put(path);
2590 return ERR_PTR(-EINVAL);
2591 }
2592 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2593 d = lookup_one_qstr_excl(&last, path->dentry, 0);
2594 if (IS_ERR(d)) {
2595 inode_unlock(path->dentry->d_inode);
2596 path_put(path);
2597 }
2598 return d;
2599}
2600
2601struct dentry *kern_path_locked(const char *name, struct path *path)
2602{
2603 struct filename *filename = getname_kernel(name);
2604 struct dentry *res = __kern_path_locked(AT_FDCWD, filename, path);
2605
2606 putname(filename);
2607 return res;
2608}
2609
2610struct dentry *user_path_locked_at(int dfd, const char __user *name, struct path *path)
2611{
2612 struct filename *filename = getname(name);
2613 struct dentry *res = __kern_path_locked(dfd, filename, path);
2614
2615 putname(filename);
2616 return res;
2617}
2618EXPORT_SYMBOL(user_path_locked_at);
2619
2620int kern_path(const char *name, unsigned int flags, struct path *path)
2621{
2622 struct filename *filename = getname_kernel(name);
2623 int ret = filename_lookup(AT_FDCWD, filename, flags, path, NULL);
2624
2625 putname(filename);
2626 return ret;
2627
2628}
2629EXPORT_SYMBOL(kern_path);
2630
2631/**
2632 * vfs_path_parent_lookup - lookup a parent path relative to a dentry-vfsmount pair
2633 * @filename: filename structure
2634 * @flags: lookup flags
2635 * @parent: pointer to struct path to fill
2636 * @last: last component
2637 * @type: type of the last component
2638 * @root: pointer to struct path of the base directory
2639 */
2640int vfs_path_parent_lookup(struct filename *filename, unsigned int flags,
2641 struct path *parent, struct qstr *last, int *type,
2642 const struct path *root)
2643{
2644 return __filename_parentat(AT_FDCWD, filename, flags, parent, last,
2645 type, root);
2646}
2647EXPORT_SYMBOL(vfs_path_parent_lookup);
2648
2649/**
2650 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2651 * @dentry: pointer to dentry of the base directory
2652 * @mnt: pointer to vfs mount of the base directory
2653 * @name: pointer to file name
2654 * @flags: lookup flags
2655 * @path: pointer to struct path to fill
2656 */
2657int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2658 const char *name, unsigned int flags,
2659 struct path *path)
2660{
2661 struct filename *filename;
2662 struct path root = {.mnt = mnt, .dentry = dentry};
2663 int ret;
2664
2665 filename = getname_kernel(name);
2666 /* the first argument of filename_lookup() is ignored with root */
2667 ret = filename_lookup(AT_FDCWD, filename, flags, path, &root);
2668 putname(filename);
2669 return ret;
2670}
2671EXPORT_SYMBOL(vfs_path_lookup);
2672
2673static int lookup_one_common(struct mnt_idmap *idmap,
2674 const char *name, struct dentry *base, int len,
2675 struct qstr *this)
2676{
2677 this->name = name;
2678 this->len = len;
2679 this->hash = full_name_hash(base, name, len);
2680 if (!len)
2681 return -EACCES;
2682
2683 if (unlikely(name[0] == '.')) {
2684 if (len < 2 || (len == 2 && name[1] == '.'))
2685 return -EACCES;
2686 }
2687
2688 while (len--) {
2689 unsigned int c = *(const unsigned char *)name++;
2690 if (c == '/' || c == '\0')
2691 return -EACCES;
2692 }
2693 /*
2694 * See if the low-level filesystem might want
2695 * to use its own hash..
2696 */
2697 if (base->d_flags & DCACHE_OP_HASH) {
2698 int err = base->d_op->d_hash(base, this);
2699 if (err < 0)
2700 return err;
2701 }
2702
2703 return inode_permission(idmap, base->d_inode, MAY_EXEC);
2704}
2705
2706/**
2707 * try_lookup_one_len - filesystem helper to lookup single pathname component
2708 * @name: pathname component to lookup
2709 * @base: base directory to lookup from
2710 * @len: maximum length @len should be interpreted to
2711 *
2712 * Look up a dentry by name in the dcache, returning NULL if it does not
2713 * currently exist. The function does not try to create a dentry.
2714 *
2715 * Note that this routine is purely a helper for filesystem usage and should
2716 * not be called by generic code.
2717 *
2718 * The caller must hold base->i_mutex.
2719 */
2720struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2721{
2722 struct qstr this;
2723 int err;
2724
2725 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2726
2727 err = lookup_one_common(&nop_mnt_idmap, name, base, len, &this);
2728 if (err)
2729 return ERR_PTR(err);
2730
2731 return lookup_dcache(&this, base, 0);
2732}
2733EXPORT_SYMBOL(try_lookup_one_len);
2734
2735/**
2736 * lookup_one_len - filesystem helper to lookup single pathname component
2737 * @name: pathname component to lookup
2738 * @base: base directory to lookup from
2739 * @len: maximum length @len should be interpreted to
2740 *
2741 * Note that this routine is purely a helper for filesystem usage and should
2742 * not be called by generic code.
2743 *
2744 * The caller must hold base->i_mutex.
2745 */
2746struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2747{
2748 struct dentry *dentry;
2749 struct qstr this;
2750 int err;
2751
2752 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2753
2754 err = lookup_one_common(&nop_mnt_idmap, name, base, len, &this);
2755 if (err)
2756 return ERR_PTR(err);
2757
2758 dentry = lookup_dcache(&this, base, 0);
2759 return dentry ? dentry : __lookup_slow(&this, base, 0);
2760}
2761EXPORT_SYMBOL(lookup_one_len);
2762
2763/**
2764 * lookup_one - filesystem helper to lookup single pathname component
2765 * @idmap: idmap of the mount the lookup is performed from
2766 * @name: pathname component to lookup
2767 * @base: base directory to lookup from
2768 * @len: maximum length @len should be interpreted to
2769 *
2770 * Note that this routine is purely a helper for filesystem usage and should
2771 * not be called by generic code.
2772 *
2773 * The caller must hold base->i_mutex.
2774 */
2775struct dentry *lookup_one(struct mnt_idmap *idmap, const char *name,
2776 struct dentry *base, int len)
2777{
2778 struct dentry *dentry;
2779 struct qstr this;
2780 int err;
2781
2782 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2783
2784 err = lookup_one_common(idmap, name, base, len, &this);
2785 if (err)
2786 return ERR_PTR(err);
2787
2788 dentry = lookup_dcache(&this, base, 0);
2789 return dentry ? dentry : __lookup_slow(&this, base, 0);
2790}
2791EXPORT_SYMBOL(lookup_one);
2792
2793/**
2794 * lookup_one_unlocked - filesystem helper to lookup single pathname component
2795 * @idmap: idmap of the mount the lookup is performed from
2796 * @name: pathname component to lookup
2797 * @base: base directory to lookup from
2798 * @len: maximum length @len should be interpreted to
2799 *
2800 * Note that this routine is purely a helper for filesystem usage and should
2801 * not be called by generic code.
2802 *
2803 * Unlike lookup_one_len, it should be called without the parent
2804 * i_mutex held, and will take the i_mutex itself if necessary.
2805 */
2806struct dentry *lookup_one_unlocked(struct mnt_idmap *idmap,
2807 const char *name, struct dentry *base,
2808 int len)
2809{
2810 struct qstr this;
2811 int err;
2812 struct dentry *ret;
2813
2814 err = lookup_one_common(idmap, name, base, len, &this);
2815 if (err)
2816 return ERR_PTR(err);
2817
2818 ret = lookup_dcache(&this, base, 0);
2819 if (!ret)
2820 ret = lookup_slow(&this, base, 0);
2821 return ret;
2822}
2823EXPORT_SYMBOL(lookup_one_unlocked);
2824
2825/**
2826 * lookup_one_positive_unlocked - filesystem helper to lookup single
2827 * pathname component
2828 * @idmap: idmap of the mount the lookup is performed from
2829 * @name: pathname component to lookup
2830 * @base: base directory to lookup from
2831 * @len: maximum length @len should be interpreted to
2832 *
2833 * This helper will yield ERR_PTR(-ENOENT) on negatives. The helper returns
2834 * known positive or ERR_PTR(). This is what most of the users want.
2835 *
2836 * Note that pinned negative with unlocked parent _can_ become positive at any
2837 * time, so callers of lookup_one_unlocked() need to be very careful; pinned
2838 * positives have >d_inode stable, so this one avoids such problems.
2839 *
2840 * Note that this routine is purely a helper for filesystem usage and should
2841 * not be called by generic code.
2842 *
2843 * The helper should be called without i_mutex held.
2844 */
2845struct dentry *lookup_one_positive_unlocked(struct mnt_idmap *idmap,
2846 const char *name,
2847 struct dentry *base, int len)
2848{
2849 struct dentry *ret = lookup_one_unlocked(idmap, name, base, len);
2850
2851 if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2852 dput(ret);
2853 ret = ERR_PTR(-ENOENT);
2854 }
2855 return ret;
2856}
2857EXPORT_SYMBOL(lookup_one_positive_unlocked);
2858
2859/**
2860 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2861 * @name: pathname component to lookup
2862 * @base: base directory to lookup from
2863 * @len: maximum length @len should be interpreted to
2864 *
2865 * Note that this routine is purely a helper for filesystem usage and should
2866 * not be called by generic code.
2867 *
2868 * Unlike lookup_one_len, it should be called without the parent
2869 * i_mutex held, and will take the i_mutex itself if necessary.
2870 */
2871struct dentry *lookup_one_len_unlocked(const char *name,
2872 struct dentry *base, int len)
2873{
2874 return lookup_one_unlocked(&nop_mnt_idmap, name, base, len);
2875}
2876EXPORT_SYMBOL(lookup_one_len_unlocked);
2877
2878/*
2879 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2880 * on negatives. Returns known positive or ERR_PTR(); that's what
2881 * most of the users want. Note that pinned negative with unlocked parent
2882 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2883 * need to be very careful; pinned positives have ->d_inode stable, so
2884 * this one avoids such problems.
2885 */
2886struct dentry *lookup_positive_unlocked(const char *name,
2887 struct dentry *base, int len)
2888{
2889 return lookup_one_positive_unlocked(&nop_mnt_idmap, name, base, len);
2890}
2891EXPORT_SYMBOL(lookup_positive_unlocked);
2892
2893#ifdef CONFIG_UNIX98_PTYS
2894int path_pts(struct path *path)
2895{
2896 /* Find something mounted on "pts" in the same directory as
2897 * the input path.
2898 */
2899 struct dentry *parent = dget_parent(path->dentry);
2900 struct dentry *child;
2901 struct qstr this = QSTR_INIT("pts", 3);
2902
2903 if (unlikely(!path_connected(path->mnt, parent))) {
2904 dput(parent);
2905 return -ENOENT;
2906 }
2907 dput(path->dentry);
2908 path->dentry = parent;
2909 child = d_hash_and_lookup(parent, &this);
2910 if (IS_ERR_OR_NULL(child))
2911 return -ENOENT;
2912
2913 path->dentry = child;
2914 dput(parent);
2915 follow_down(path, 0);
2916 return 0;
2917}
2918#endif
2919
2920int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2921 struct path *path, int *empty)
2922{
2923 struct filename *filename = getname_flags(name, flags, empty);
2924 int ret = filename_lookup(dfd, filename, flags, path, NULL);
2925
2926 putname(filename);
2927 return ret;
2928}
2929EXPORT_SYMBOL(user_path_at_empty);
2930
2931int __check_sticky(struct mnt_idmap *idmap, struct inode *dir,
2932 struct inode *inode)
2933{
2934 kuid_t fsuid = current_fsuid();
2935
2936 if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode), fsuid))
2937 return 0;
2938 if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, dir), fsuid))
2939 return 0;
2940 return !capable_wrt_inode_uidgid(idmap, inode, CAP_FOWNER);
2941}
2942EXPORT_SYMBOL(__check_sticky);
2943
2944/*
2945 * Check whether we can remove a link victim from directory dir, check
2946 * whether the type of victim is right.
2947 * 1. We can't do it if dir is read-only (done in permission())
2948 * 2. We should have write and exec permissions on dir
2949 * 3. We can't remove anything from append-only dir
2950 * 4. We can't do anything with immutable dir (done in permission())
2951 * 5. If the sticky bit on dir is set we should either
2952 * a. be owner of dir, or
2953 * b. be owner of victim, or
2954 * c. have CAP_FOWNER capability
2955 * 6. If the victim is append-only or immutable we can't do antyhing with
2956 * links pointing to it.
2957 * 7. If the victim has an unknown uid or gid we can't change the inode.
2958 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2959 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2960 * 10. We can't remove a root or mountpoint.
2961 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2962 * nfs_async_unlink().
2963 */
2964static int may_delete(struct mnt_idmap *idmap, struct inode *dir,
2965 struct dentry *victim, bool isdir)
2966{
2967 struct inode *inode = d_backing_inode(victim);
2968 int error;
2969
2970 if (d_is_negative(victim))
2971 return -ENOENT;
2972 BUG_ON(!inode);
2973
2974 BUG_ON(victim->d_parent->d_inode != dir);
2975
2976 /* Inode writeback is not safe when the uid or gid are invalid. */
2977 if (!vfsuid_valid(i_uid_into_vfsuid(idmap, inode)) ||
2978 !vfsgid_valid(i_gid_into_vfsgid(idmap, inode)))
2979 return -EOVERFLOW;
2980
2981 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2982
2983 error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
2984 if (error)
2985 return error;
2986 if (IS_APPEND(dir))
2987 return -EPERM;
2988
2989 if (check_sticky(idmap, dir, inode) || IS_APPEND(inode) ||
2990 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) ||
2991 HAS_UNMAPPED_ID(idmap, inode))
2992 return -EPERM;
2993 if (isdir) {
2994 if (!d_is_dir(victim))
2995 return -ENOTDIR;
2996 if (IS_ROOT(victim))
2997 return -EBUSY;
2998 } else if (d_is_dir(victim))
2999 return -EISDIR;
3000 if (IS_DEADDIR(dir))
3001 return -ENOENT;
3002 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
3003 return -EBUSY;
3004 return 0;
3005}
3006
3007/* Check whether we can create an object with dentry child in directory
3008 * dir.
3009 * 1. We can't do it if child already exists (open has special treatment for
3010 * this case, but since we are inlined it's OK)
3011 * 2. We can't do it if dir is read-only (done in permission())
3012 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
3013 * 4. We should have write and exec permissions on dir
3014 * 5. We can't do it if dir is immutable (done in permission())
3015 */
3016static inline int may_create(struct mnt_idmap *idmap,
3017 struct inode *dir, struct dentry *child)
3018{
3019 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
3020 if (child->d_inode)
3021 return -EEXIST;
3022 if (IS_DEADDIR(dir))
3023 return -ENOENT;
3024 if (!fsuidgid_has_mapping(dir->i_sb, idmap))
3025 return -EOVERFLOW;
3026
3027 return inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
3028}
3029
3030// p1 != p2, both are on the same filesystem, ->s_vfs_rename_mutex is held
3031static struct dentry *lock_two_directories(struct dentry *p1, struct dentry *p2)
3032{
3033 struct dentry *p = p1, *q = p2, *r;
3034
3035 while ((r = p->d_parent) != p2 && r != p)
3036 p = r;
3037 if (r == p2) {
3038 // p is a child of p2 and an ancestor of p1 or p1 itself
3039 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
3040 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT2);
3041 return p;
3042 }
3043 // p is the root of connected component that contains p1
3044 // p2 does not occur on the path from p to p1
3045 while ((r = q->d_parent) != p1 && r != p && r != q)
3046 q = r;
3047 if (r == p1) {
3048 // q is a child of p1 and an ancestor of p2 or p2 itself
3049 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3050 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
3051 return q;
3052 } else if (likely(r == p)) {
3053 // both p2 and p1 are descendents of p
3054 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3055 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
3056 return NULL;
3057 } else { // no common ancestor at the time we'd been called
3058 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
3059 return ERR_PTR(-EXDEV);
3060 }
3061}
3062
3063/*
3064 * p1 and p2 should be directories on the same fs.
3065 */
3066struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
3067{
3068 if (p1 == p2) {
3069 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3070 return NULL;
3071 }
3072
3073 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
3074 return lock_two_directories(p1, p2);
3075}
3076EXPORT_SYMBOL(lock_rename);
3077
3078/*
3079 * c1 and p2 should be on the same fs.
3080 */
3081struct dentry *lock_rename_child(struct dentry *c1, struct dentry *p2)
3082{
3083 if (READ_ONCE(c1->d_parent) == p2) {
3084 /*
3085 * hopefully won't need to touch ->s_vfs_rename_mutex at all.
3086 */
3087 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
3088 /*
3089 * now that p2 is locked, nobody can move in or out of it,
3090 * so the test below is safe.
3091 */
3092 if (likely(c1->d_parent == p2))
3093 return NULL;
3094
3095 /*
3096 * c1 got moved out of p2 while we'd been taking locks;
3097 * unlock and fall back to slow case.
3098 */
3099 inode_unlock(p2->d_inode);
3100 }
3101
3102 mutex_lock(&c1->d_sb->s_vfs_rename_mutex);
3103 /*
3104 * nobody can move out of any directories on this fs.
3105 */
3106 if (likely(c1->d_parent != p2))
3107 return lock_two_directories(c1->d_parent, p2);
3108
3109 /*
3110 * c1 got moved into p2 while we were taking locks;
3111 * we need p2 locked and ->s_vfs_rename_mutex unlocked,
3112 * for consistency with lock_rename().
3113 */
3114 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
3115 mutex_unlock(&c1->d_sb->s_vfs_rename_mutex);
3116 return NULL;
3117}
3118EXPORT_SYMBOL(lock_rename_child);
3119
3120void unlock_rename(struct dentry *p1, struct dentry *p2)
3121{
3122 inode_unlock(p1->d_inode);
3123 if (p1 != p2) {
3124 inode_unlock(p2->d_inode);
3125 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
3126 }
3127}
3128EXPORT_SYMBOL(unlock_rename);
3129
3130/**
3131 * vfs_prepare_mode - prepare the mode to be used for a new inode
3132 * @idmap: idmap of the mount the inode was found from
3133 * @dir: parent directory of the new inode
3134 * @mode: mode of the new inode
3135 * @mask_perms: allowed permission by the vfs
3136 * @type: type of file to be created
3137 *
3138 * This helper consolidates and enforces vfs restrictions on the @mode of a new
3139 * object to be created.
3140 *
3141 * Umask stripping depends on whether the filesystem supports POSIX ACLs (see
3142 * the kernel documentation for mode_strip_umask()). Moving umask stripping
3143 * after setgid stripping allows the same ordering for both non-POSIX ACL and
3144 * POSIX ACL supporting filesystems.
3145 *
3146 * Note that it's currently valid for @type to be 0 if a directory is created.
3147 * Filesystems raise that flag individually and we need to check whether each
3148 * filesystem can deal with receiving S_IFDIR from the vfs before we enforce a
3149 * non-zero type.
3150 *
3151 * Returns: mode to be passed to the filesystem
3152 */
3153static inline umode_t vfs_prepare_mode(struct mnt_idmap *idmap,
3154 const struct inode *dir, umode_t mode,
3155 umode_t mask_perms, umode_t type)
3156{
3157 mode = mode_strip_sgid(idmap, dir, mode);
3158 mode = mode_strip_umask(dir, mode);
3159
3160 /*
3161 * Apply the vfs mandated allowed permission mask and set the type of
3162 * file to be created before we call into the filesystem.
3163 */
3164 mode &= (mask_perms & ~S_IFMT);
3165 mode |= (type & S_IFMT);
3166
3167 return mode;
3168}
3169
3170/**
3171 * vfs_create - create new file
3172 * @idmap: idmap of the mount the inode was found from
3173 * @dir: inode of @dentry
3174 * @dentry: pointer to dentry of the base directory
3175 * @mode: mode of the new file
3176 * @want_excl: whether the file must not yet exist
3177 *
3178 * Create a new file.
3179 *
3180 * If the inode has been found through an idmapped mount the idmap of
3181 * the vfsmount must be passed through @idmap. This function will then take
3182 * care to map the inode according to @idmap before checking permissions.
3183 * On non-idmapped mounts or if permission checking is to be performed on the
3184 * raw inode simply pass @nop_mnt_idmap.
3185 */
3186int vfs_create(struct mnt_idmap *idmap, struct inode *dir,
3187 struct dentry *dentry, umode_t mode, bool want_excl)
3188{
3189 int error;
3190
3191 error = may_create(idmap, dir, dentry);
3192 if (error)
3193 return error;
3194
3195 if (!dir->i_op->create)
3196 return -EACCES; /* shouldn't it be ENOSYS? */
3197
3198 mode = vfs_prepare_mode(idmap, dir, mode, S_IALLUGO, S_IFREG);
3199 error = security_inode_create(dir, dentry, mode);
3200 if (error)
3201 return error;
3202 error = dir->i_op->create(idmap, dir, dentry, mode, want_excl);
3203 if (!error)
3204 fsnotify_create(dir, dentry);
3205 return error;
3206}
3207EXPORT_SYMBOL(vfs_create);
3208
3209int vfs_mkobj(struct dentry *dentry, umode_t mode,
3210 int (*f)(struct dentry *, umode_t, void *),
3211 void *arg)
3212{
3213 struct inode *dir = dentry->d_parent->d_inode;
3214 int error = may_create(&nop_mnt_idmap, dir, dentry);
3215 if (error)
3216 return error;
3217
3218 mode &= S_IALLUGO;
3219 mode |= S_IFREG;
3220 error = security_inode_create(dir, dentry, mode);
3221 if (error)
3222 return error;
3223 error = f(dentry, mode, arg);
3224 if (!error)
3225 fsnotify_create(dir, dentry);
3226 return error;
3227}
3228EXPORT_SYMBOL(vfs_mkobj);
3229
3230bool may_open_dev(const struct path *path)
3231{
3232 return !(path->mnt->mnt_flags & MNT_NODEV) &&
3233 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
3234}
3235
3236static int may_open(struct mnt_idmap *idmap, const struct path *path,
3237 int acc_mode, int flag)
3238{
3239 struct dentry *dentry = path->dentry;
3240 struct inode *inode = dentry->d_inode;
3241 int error;
3242
3243 if (!inode)
3244 return -ENOENT;
3245
3246 switch (inode->i_mode & S_IFMT) {
3247 case S_IFLNK:
3248 return -ELOOP;
3249 case S_IFDIR:
3250 if (acc_mode & MAY_WRITE)
3251 return -EISDIR;
3252 if (acc_mode & MAY_EXEC)
3253 return -EACCES;
3254 break;
3255 case S_IFBLK:
3256 case S_IFCHR:
3257 if (!may_open_dev(path))
3258 return -EACCES;
3259 fallthrough;
3260 case S_IFIFO:
3261 case S_IFSOCK:
3262 if (acc_mode & MAY_EXEC)
3263 return -EACCES;
3264 flag &= ~O_TRUNC;
3265 break;
3266 case S_IFREG:
3267 if ((acc_mode & MAY_EXEC) && path_noexec(path))
3268 return -EACCES;
3269 break;
3270 }
3271
3272 error = inode_permission(idmap, inode, MAY_OPEN | acc_mode);
3273 if (error)
3274 return error;
3275
3276 /*
3277 * An append-only file must be opened in append mode for writing.
3278 */
3279 if (IS_APPEND(inode)) {
3280 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
3281 return -EPERM;
3282 if (flag & O_TRUNC)
3283 return -EPERM;
3284 }
3285
3286 /* O_NOATIME can only be set by the owner or superuser */
3287 if (flag & O_NOATIME && !inode_owner_or_capable(idmap, inode))
3288 return -EPERM;
3289
3290 return 0;
3291}
3292
3293static int handle_truncate(struct mnt_idmap *idmap, struct file *filp)
3294{
3295 const struct path *path = &filp->f_path;
3296 struct inode *inode = path->dentry->d_inode;
3297 int error = get_write_access(inode);
3298 if (error)
3299 return error;
3300
3301 error = security_file_truncate(filp);
3302 if (!error) {
3303 error = do_truncate(idmap, path->dentry, 0,
3304 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3305 filp);
3306 }
3307 put_write_access(inode);
3308 return error;
3309}
3310
3311static inline int open_to_namei_flags(int flag)
3312{
3313 if ((flag & O_ACCMODE) == 3)
3314 flag--;
3315 return flag;
3316}
3317
3318static int may_o_create(struct mnt_idmap *idmap,
3319 const struct path *dir, struct dentry *dentry,
3320 umode_t mode)
3321{
3322 int error = security_path_mknod(dir, dentry, mode, 0);
3323 if (error)
3324 return error;
3325
3326 if (!fsuidgid_has_mapping(dir->dentry->d_sb, idmap))
3327 return -EOVERFLOW;
3328
3329 error = inode_permission(idmap, dir->dentry->d_inode,
3330 MAY_WRITE | MAY_EXEC);
3331 if (error)
3332 return error;
3333
3334 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3335}
3336
3337/*
3338 * Attempt to atomically look up, create and open a file from a negative
3339 * dentry.
3340 *
3341 * Returns 0 if successful. The file will have been created and attached to
3342 * @file by the filesystem calling finish_open().
3343 *
3344 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3345 * be set. The caller will need to perform the open themselves. @path will
3346 * have been updated to point to the new dentry. This may be negative.
3347 *
3348 * Returns an error code otherwise.
3349 */
3350static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
3351 struct file *file,
3352 int open_flag, umode_t mode)
3353{
3354 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3355 struct inode *dir = nd->path.dentry->d_inode;
3356 int error;
3357
3358 if (nd->flags & LOOKUP_DIRECTORY)
3359 open_flag |= O_DIRECTORY;
3360
3361 file->f_path.dentry = DENTRY_NOT_SET;
3362 file->f_path.mnt = nd->path.mnt;
3363 error = dir->i_op->atomic_open(dir, dentry, file,
3364 open_to_namei_flags(open_flag), mode);
3365 d_lookup_done(dentry);
3366 if (!error) {
3367 if (file->f_mode & FMODE_OPENED) {
3368 if (unlikely(dentry != file->f_path.dentry)) {
3369 dput(dentry);
3370 dentry = dget(file->f_path.dentry);
3371 }
3372 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3373 error = -EIO;
3374 } else {
3375 if (file->f_path.dentry) {
3376 dput(dentry);
3377 dentry = file->f_path.dentry;
3378 }
3379 if (unlikely(d_is_negative(dentry)))
3380 error = -ENOENT;
3381 }
3382 }
3383 if (error) {
3384 dput(dentry);
3385 dentry = ERR_PTR(error);
3386 }
3387 return dentry;
3388}
3389
3390/*
3391 * Look up and maybe create and open the last component.
3392 *
3393 * Must be called with parent locked (exclusive in O_CREAT case).
3394 *
3395 * Returns 0 on success, that is, if
3396 * the file was successfully atomically created (if necessary) and opened, or
3397 * the file was not completely opened at this time, though lookups and
3398 * creations were performed.
3399 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3400 * In the latter case dentry returned in @path might be negative if O_CREAT
3401 * hadn't been specified.
3402 *
3403 * An error code is returned on failure.
3404 */
3405static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3406 const struct open_flags *op,
3407 bool got_write)
3408{
3409 struct mnt_idmap *idmap;
3410 struct dentry *dir = nd->path.dentry;
3411 struct inode *dir_inode = dir->d_inode;
3412 int open_flag = op->open_flag;
3413 struct dentry *dentry;
3414 int error, create_error = 0;
3415 umode_t mode = op->mode;
3416 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3417
3418 if (unlikely(IS_DEADDIR(dir_inode)))
3419 return ERR_PTR(-ENOENT);
3420
3421 file->f_mode &= ~FMODE_CREATED;
3422 dentry = d_lookup(dir, &nd->last);
3423 for (;;) {
3424 if (!dentry) {
3425 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3426 if (IS_ERR(dentry))
3427 return dentry;
3428 }
3429 if (d_in_lookup(dentry))
3430 break;
3431
3432 error = d_revalidate(dentry, nd->flags);
3433 if (likely(error > 0))
3434 break;
3435 if (error)
3436 goto out_dput;
3437 d_invalidate(dentry);
3438 dput(dentry);
3439 dentry = NULL;
3440 }
3441 if (dentry->d_inode) {
3442 /* Cached positive dentry: will open in f_op->open */
3443 return dentry;
3444 }
3445
3446 /*
3447 * Checking write permission is tricky, bacuse we don't know if we are
3448 * going to actually need it: O_CREAT opens should work as long as the
3449 * file exists. But checking existence breaks atomicity. The trick is
3450 * to check access and if not granted clear O_CREAT from the flags.
3451 *
3452 * Another problem is returing the "right" error value (e.g. for an
3453 * O_EXCL open we want to return EEXIST not EROFS).
3454 */
3455 if (unlikely(!got_write))
3456 open_flag &= ~O_TRUNC;
3457 idmap = mnt_idmap(nd->path.mnt);
3458 if (open_flag & O_CREAT) {
3459 if (open_flag & O_EXCL)
3460 open_flag &= ~O_TRUNC;
3461 mode = vfs_prepare_mode(idmap, dir->d_inode, mode, mode, mode);
3462 if (likely(got_write))
3463 create_error = may_o_create(idmap, &nd->path,
3464 dentry, mode);
3465 else
3466 create_error = -EROFS;
3467 }
3468 if (create_error)
3469 open_flag &= ~O_CREAT;
3470 if (dir_inode->i_op->atomic_open) {
3471 dentry = atomic_open(nd, dentry, file, open_flag, mode);
3472 if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3473 dentry = ERR_PTR(create_error);
3474 return dentry;
3475 }
3476
3477 if (d_in_lookup(dentry)) {
3478 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3479 nd->flags);
3480 d_lookup_done(dentry);
3481 if (unlikely(res)) {
3482 if (IS_ERR(res)) {
3483 error = PTR_ERR(res);
3484 goto out_dput;
3485 }
3486 dput(dentry);
3487 dentry = res;
3488 }
3489 }
3490
3491 /* Negative dentry, just create the file */
3492 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3493 file->f_mode |= FMODE_CREATED;
3494 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3495 if (!dir_inode->i_op->create) {
3496 error = -EACCES;
3497 goto out_dput;
3498 }
3499
3500 error = dir_inode->i_op->create(idmap, dir_inode, dentry,
3501 mode, open_flag & O_EXCL);
3502 if (error)
3503 goto out_dput;
3504 }
3505 if (unlikely(create_error) && !dentry->d_inode) {
3506 error = create_error;
3507 goto out_dput;
3508 }
3509 return dentry;
3510
3511out_dput:
3512 dput(dentry);
3513 return ERR_PTR(error);
3514}
3515
3516static const char *open_last_lookups(struct nameidata *nd,
3517 struct file *file, const struct open_flags *op)
3518{
3519 struct dentry *dir = nd->path.dentry;
3520 int open_flag = op->open_flag;
3521 bool got_write = false;
3522 struct dentry *dentry;
3523 const char *res;
3524
3525 nd->flags |= op->intent;
3526
3527 if (nd->last_type != LAST_NORM) {
3528 if (nd->depth)
3529 put_link(nd);
3530 return handle_dots(nd, nd->last_type);
3531 }
3532
3533 if (!(open_flag & O_CREAT)) {
3534 if (nd->last.name[nd->last.len])
3535 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3536 /* we _can_ be in RCU mode here */
3537 dentry = lookup_fast(nd);
3538 if (IS_ERR(dentry))
3539 return ERR_CAST(dentry);
3540 if (likely(dentry))
3541 goto finish_lookup;
3542
3543 if (WARN_ON_ONCE(nd->flags & LOOKUP_RCU))
3544 return ERR_PTR(-ECHILD);
3545 } else {
3546 /* create side of things */
3547 if (nd->flags & LOOKUP_RCU) {
3548 if (!try_to_unlazy(nd))
3549 return ERR_PTR(-ECHILD);
3550 }
3551 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3552 /* trailing slashes? */
3553 if (unlikely(nd->last.name[nd->last.len]))
3554 return ERR_PTR(-EISDIR);
3555 }
3556
3557 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3558 got_write = !mnt_want_write(nd->path.mnt);
3559 /*
3560 * do _not_ fail yet - we might not need that or fail with
3561 * a different error; let lookup_open() decide; we'll be
3562 * dropping this one anyway.
3563 */
3564 }
3565 if (open_flag & O_CREAT)
3566 inode_lock(dir->d_inode);
3567 else
3568 inode_lock_shared(dir->d_inode);
3569 dentry = lookup_open(nd, file, op, got_write);
3570 if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED))
3571 fsnotify_create(dir->d_inode, dentry);
3572 if (open_flag & O_CREAT)
3573 inode_unlock(dir->d_inode);
3574 else
3575 inode_unlock_shared(dir->d_inode);
3576
3577 if (got_write)
3578 mnt_drop_write(nd->path.mnt);
3579
3580 if (IS_ERR(dentry))
3581 return ERR_CAST(dentry);
3582
3583 if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
3584 dput(nd->path.dentry);
3585 nd->path.dentry = dentry;
3586 return NULL;
3587 }
3588
3589finish_lookup:
3590 if (nd->depth)
3591 put_link(nd);
3592 res = step_into(nd, WALK_TRAILING, dentry);
3593 if (unlikely(res))
3594 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3595 return res;
3596}
3597
3598/*
3599 * Handle the last step of open()
3600 */
3601static int do_open(struct nameidata *nd,
3602 struct file *file, const struct open_flags *op)
3603{
3604 struct mnt_idmap *idmap;
3605 int open_flag = op->open_flag;
3606 bool do_truncate;
3607 int acc_mode;
3608 int error;
3609
3610 if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3611 error = complete_walk(nd);
3612 if (error)
3613 return error;
3614 }
3615 if (!(file->f_mode & FMODE_CREATED))
3616 audit_inode(nd->name, nd->path.dentry, 0);
3617 idmap = mnt_idmap(nd->path.mnt);
3618 if (open_flag & O_CREAT) {
3619 if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3620 return -EEXIST;
3621 if (d_is_dir(nd->path.dentry))
3622 return -EISDIR;
3623 error = may_create_in_sticky(idmap, nd,
3624 d_backing_inode(nd->path.dentry));
3625 if (unlikely(error))
3626 return error;
3627 }
3628 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3629 return -ENOTDIR;
3630
3631 do_truncate = false;
3632 acc_mode = op->acc_mode;
3633 if (file->f_mode & FMODE_CREATED) {
3634 /* Don't check for write permission, don't truncate */
3635 open_flag &= ~O_TRUNC;
3636 acc_mode = 0;
3637 } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) {
3638 error = mnt_want_write(nd->path.mnt);
3639 if (error)
3640 return error;
3641 do_truncate = true;
3642 }
3643 error = may_open(idmap, &nd->path, acc_mode, open_flag);
3644 if (!error && !(file->f_mode & FMODE_OPENED))
3645 error = vfs_open(&nd->path, file);
3646 if (!error)
3647 error = ima_file_check(file, op->acc_mode);
3648 if (!error && do_truncate)
3649 error = handle_truncate(idmap, file);
3650 if (unlikely(error > 0)) {
3651 WARN_ON(1);
3652 error = -EINVAL;
3653 }
3654 if (do_truncate)
3655 mnt_drop_write(nd->path.mnt);
3656 return error;
3657}
3658
3659/**
3660 * vfs_tmpfile - create tmpfile
3661 * @idmap: idmap of the mount the inode was found from
3662 * @parentpath: pointer to the path of the base directory
3663 * @file: file descriptor of the new tmpfile
3664 * @mode: mode of the new tmpfile
3665 *
3666 * Create a temporary file.
3667 *
3668 * If the inode has been found through an idmapped mount the idmap of
3669 * the vfsmount must be passed through @idmap. This function will then take
3670 * care to map the inode according to @idmap before checking permissions.
3671 * On non-idmapped mounts or if permission checking is to be performed on the
3672 * raw inode simply pass @nop_mnt_idmap.
3673 */
3674static int vfs_tmpfile(struct mnt_idmap *idmap,
3675 const struct path *parentpath,
3676 struct file *file, umode_t mode)
3677{
3678 struct dentry *child;
3679 struct inode *dir = d_inode(parentpath->dentry);
3680 struct inode *inode;
3681 int error;
3682 int open_flag = file->f_flags;
3683
3684 /* we want directory to be writable */
3685 error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
3686 if (error)
3687 return error;
3688 if (!dir->i_op->tmpfile)
3689 return -EOPNOTSUPP;
3690 child = d_alloc(parentpath->dentry, &slash_name);
3691 if (unlikely(!child))
3692 return -ENOMEM;
3693 file->f_path.mnt = parentpath->mnt;
3694 file->f_path.dentry = child;
3695 mode = vfs_prepare_mode(idmap, dir, mode, mode, mode);
3696 error = dir->i_op->tmpfile(idmap, dir, file, mode);
3697 dput(child);
3698 if (error)
3699 return error;
3700 /* Don't check for other permissions, the inode was just created */
3701 error = may_open(idmap, &file->f_path, 0, file->f_flags);
3702 if (error)
3703 return error;
3704 inode = file_inode(file);
3705 if (!(open_flag & O_EXCL)) {
3706 spin_lock(&inode->i_lock);
3707 inode->i_state |= I_LINKABLE;
3708 spin_unlock(&inode->i_lock);
3709 }
3710 ima_post_create_tmpfile(idmap, inode);
3711 return 0;
3712}
3713
3714/**
3715 * kernel_tmpfile_open - open a tmpfile for kernel internal use
3716 * @idmap: idmap of the mount the inode was found from
3717 * @parentpath: path of the base directory
3718 * @mode: mode of the new tmpfile
3719 * @open_flag: flags
3720 * @cred: credentials for open
3721 *
3722 * Create and open a temporary file. The file is not accounted in nr_files,
3723 * hence this is only for kernel internal use, and must not be installed into
3724 * file tables or such.
3725 */
3726struct file *kernel_tmpfile_open(struct mnt_idmap *idmap,
3727 const struct path *parentpath,
3728 umode_t mode, int open_flag,
3729 const struct cred *cred)
3730{
3731 struct file *file;
3732 int error;
3733
3734 file = alloc_empty_file_noaccount(open_flag, cred);
3735 if (IS_ERR(file))
3736 return file;
3737
3738 error = vfs_tmpfile(idmap, parentpath, file, mode);
3739 if (error) {
3740 fput(file);
3741 file = ERR_PTR(error);
3742 }
3743 return file;
3744}
3745EXPORT_SYMBOL(kernel_tmpfile_open);
3746
3747static int do_tmpfile(struct nameidata *nd, unsigned flags,
3748 const struct open_flags *op,
3749 struct file *file)
3750{
3751 struct path path;
3752 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3753
3754 if (unlikely(error))
3755 return error;
3756 error = mnt_want_write(path.mnt);
3757 if (unlikely(error))
3758 goto out;
3759 error = vfs_tmpfile(mnt_idmap(path.mnt), &path, file, op->mode);
3760 if (error)
3761 goto out2;
3762 audit_inode(nd->name, file->f_path.dentry, 0);
3763out2:
3764 mnt_drop_write(path.mnt);
3765out:
3766 path_put(&path);
3767 return error;
3768}
3769
3770static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3771{
3772 struct path path;
3773 int error = path_lookupat(nd, flags, &path);
3774 if (!error) {
3775 audit_inode(nd->name, path.dentry, 0);
3776 error = vfs_open(&path, file);
3777 path_put(&path);
3778 }
3779 return error;
3780}
3781
3782static struct file *path_openat(struct nameidata *nd,
3783 const struct open_flags *op, unsigned flags)
3784{
3785 struct file *file;
3786 int error;
3787
3788 file = alloc_empty_file(op->open_flag, current_cred());
3789 if (IS_ERR(file))
3790 return file;
3791
3792 if (unlikely(file->f_flags & __O_TMPFILE)) {
3793 error = do_tmpfile(nd, flags, op, file);
3794 } else if (unlikely(file->f_flags & O_PATH)) {
3795 error = do_o_path(nd, flags, file);
3796 } else {
3797 const char *s = path_init(nd, flags);
3798 while (!(error = link_path_walk(s, nd)) &&
3799 (s = open_last_lookups(nd, file, op)) != NULL)
3800 ;
3801 if (!error)
3802 error = do_open(nd, file, op);
3803 terminate_walk(nd);
3804 }
3805 if (likely(!error)) {
3806 if (likely(file->f_mode & FMODE_OPENED))
3807 return file;
3808 WARN_ON(1);
3809 error = -EINVAL;
3810 }
3811 fput(file);
3812 if (error == -EOPENSTALE) {
3813 if (flags & LOOKUP_RCU)
3814 error = -ECHILD;
3815 else
3816 error = -ESTALE;
3817 }
3818 return ERR_PTR(error);
3819}
3820
3821struct file *do_filp_open(int dfd, struct filename *pathname,
3822 const struct open_flags *op)
3823{
3824 struct nameidata nd;
3825 int flags = op->lookup_flags;
3826 struct file *filp;
3827
3828 set_nameidata(&nd, dfd, pathname, NULL);
3829 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3830 if (unlikely(filp == ERR_PTR(-ECHILD)))
3831 filp = path_openat(&nd, op, flags);
3832 if (unlikely(filp == ERR_PTR(-ESTALE)))
3833 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3834 restore_nameidata();
3835 return filp;
3836}
3837
3838struct file *do_file_open_root(const struct path *root,
3839 const char *name, const struct open_flags *op)
3840{
3841 struct nameidata nd;
3842 struct file *file;
3843 struct filename *filename;
3844 int flags = op->lookup_flags;
3845
3846 if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN)
3847 return ERR_PTR(-ELOOP);
3848
3849 filename = getname_kernel(name);
3850 if (IS_ERR(filename))
3851 return ERR_CAST(filename);
3852
3853 set_nameidata(&nd, -1, filename, root);
3854 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3855 if (unlikely(file == ERR_PTR(-ECHILD)))
3856 file = path_openat(&nd, op, flags);
3857 if (unlikely(file == ERR_PTR(-ESTALE)))
3858 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3859 restore_nameidata();
3860 putname(filename);
3861 return file;
3862}
3863
3864static struct dentry *filename_create(int dfd, struct filename *name,
3865 struct path *path, unsigned int lookup_flags)
3866{
3867 struct dentry *dentry = ERR_PTR(-EEXIST);
3868 struct qstr last;
3869 bool want_dir = lookup_flags & LOOKUP_DIRECTORY;
3870 unsigned int reval_flag = lookup_flags & LOOKUP_REVAL;
3871 unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL;
3872 int type;
3873 int err2;
3874 int error;
3875
3876 error = filename_parentat(dfd, name, reval_flag, path, &last, &type);
3877 if (error)
3878 return ERR_PTR(error);
3879
3880 /*
3881 * Yucky last component or no last component at all?
3882 * (foo/., foo/.., /////)
3883 */
3884 if (unlikely(type != LAST_NORM))
3885 goto out;
3886
3887 /* don't fail immediately if it's r/o, at least try to report other errors */
3888 err2 = mnt_want_write(path->mnt);
3889 /*
3890 * Do the final lookup. Suppress 'create' if there is a trailing
3891 * '/', and a directory wasn't requested.
3892 */
3893 if (last.name[last.len] && !want_dir)
3894 create_flags = 0;
3895 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3896 dentry = lookup_one_qstr_excl(&last, path->dentry,
3897 reval_flag | create_flags);
3898 if (IS_ERR(dentry))
3899 goto unlock;
3900
3901 error = -EEXIST;
3902 if (d_is_positive(dentry))
3903 goto fail;
3904
3905 /*
3906 * Special case - lookup gave negative, but... we had foo/bar/
3907 * From the vfs_mknod() POV we just have a negative dentry -
3908 * all is fine. Let's be bastards - you had / on the end, you've
3909 * been asking for (non-existent) directory. -ENOENT for you.
3910 */
3911 if (unlikely(!create_flags)) {
3912 error = -ENOENT;
3913 goto fail;
3914 }
3915 if (unlikely(err2)) {
3916 error = err2;
3917 goto fail;
3918 }
3919 return dentry;
3920fail:
3921 dput(dentry);
3922 dentry = ERR_PTR(error);
3923unlock:
3924 inode_unlock(path->dentry->d_inode);
3925 if (!err2)
3926 mnt_drop_write(path->mnt);
3927out:
3928 path_put(path);
3929 return dentry;
3930}
3931
3932struct dentry *kern_path_create(int dfd, const char *pathname,
3933 struct path *path, unsigned int lookup_flags)
3934{
3935 struct filename *filename = getname_kernel(pathname);
3936 struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3937
3938 putname(filename);
3939 return res;
3940}
3941EXPORT_SYMBOL(kern_path_create);
3942
3943void done_path_create(struct path *path, struct dentry *dentry)
3944{
3945 dput(dentry);
3946 inode_unlock(path->dentry->d_inode);
3947 mnt_drop_write(path->mnt);
3948 path_put(path);
3949}
3950EXPORT_SYMBOL(done_path_create);
3951
3952inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3953 struct path *path, unsigned int lookup_flags)
3954{
3955 struct filename *filename = getname(pathname);
3956 struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3957
3958 putname(filename);
3959 return res;
3960}
3961EXPORT_SYMBOL(user_path_create);
3962
3963/**
3964 * vfs_mknod - create device node or file
3965 * @idmap: idmap of the mount the inode was found from
3966 * @dir: inode of @dentry
3967 * @dentry: pointer to dentry of the base directory
3968 * @mode: mode of the new device node or file
3969 * @dev: device number of device to create
3970 *
3971 * Create a device node or file.
3972 *
3973 * If the inode has been found through an idmapped mount the idmap of
3974 * the vfsmount must be passed through @idmap. This function will then take
3975 * care to map the inode according to @idmap before checking permissions.
3976 * On non-idmapped mounts or if permission checking is to be performed on the
3977 * raw inode simply pass @nop_mnt_idmap.
3978 */
3979int vfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
3980 struct dentry *dentry, umode_t mode, dev_t dev)
3981{
3982 bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV;
3983 int error = may_create(idmap, dir, dentry);
3984
3985 if (error)
3986 return error;
3987
3988 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout &&
3989 !capable(CAP_MKNOD))
3990 return -EPERM;
3991
3992 if (!dir->i_op->mknod)
3993 return -EPERM;
3994
3995 mode = vfs_prepare_mode(idmap, dir, mode, mode, mode);
3996 error = devcgroup_inode_mknod(mode, dev);
3997 if (error)
3998 return error;
3999
4000 error = security_inode_mknod(dir, dentry, mode, dev);
4001 if (error)
4002 return error;
4003
4004 error = dir->i_op->mknod(idmap, dir, dentry, mode, dev);
4005 if (!error)
4006 fsnotify_create(dir, dentry);
4007 return error;
4008}
4009EXPORT_SYMBOL(vfs_mknod);
4010
4011static int may_mknod(umode_t mode)
4012{
4013 switch (mode & S_IFMT) {
4014 case S_IFREG:
4015 case S_IFCHR:
4016 case S_IFBLK:
4017 case S_IFIFO:
4018 case S_IFSOCK:
4019 case 0: /* zero mode translates to S_IFREG */
4020 return 0;
4021 case S_IFDIR:
4022 return -EPERM;
4023 default:
4024 return -EINVAL;
4025 }
4026}
4027
4028static int do_mknodat(int dfd, struct filename *name, umode_t mode,
4029 unsigned int dev)
4030{
4031 struct mnt_idmap *idmap;
4032 struct dentry *dentry;
4033 struct path path;
4034 int error;
4035 unsigned int lookup_flags = 0;
4036
4037 error = may_mknod(mode);
4038 if (error)
4039 goto out1;
4040retry:
4041 dentry = filename_create(dfd, name, &path, lookup_flags);
4042 error = PTR_ERR(dentry);
4043 if (IS_ERR(dentry))
4044 goto out1;
4045
4046 error = security_path_mknod(&path, dentry,
4047 mode_strip_umask(path.dentry->d_inode, mode), dev);
4048 if (error)
4049 goto out2;
4050
4051 idmap = mnt_idmap(path.mnt);
4052 switch (mode & S_IFMT) {
4053 case 0: case S_IFREG:
4054 error = vfs_create(idmap, path.dentry->d_inode,
4055 dentry, mode, true);
4056 if (!error)
4057 ima_post_path_mknod(idmap, dentry);
4058 break;
4059 case S_IFCHR: case S_IFBLK:
4060 error = vfs_mknod(idmap, path.dentry->d_inode,
4061 dentry, mode, new_decode_dev(dev));
4062 break;
4063 case S_IFIFO: case S_IFSOCK:
4064 error = vfs_mknod(idmap, path.dentry->d_inode,
4065 dentry, mode, 0);
4066 break;
4067 }
4068out2:
4069 done_path_create(&path, dentry);
4070 if (retry_estale(error, lookup_flags)) {
4071 lookup_flags |= LOOKUP_REVAL;
4072 goto retry;
4073 }
4074out1:
4075 putname(name);
4076 return error;
4077}
4078
4079SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
4080 unsigned int, dev)
4081{
4082 return do_mknodat(dfd, getname(filename), mode, dev);
4083}
4084
4085SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
4086{
4087 return do_mknodat(AT_FDCWD, getname(filename), mode, dev);
4088}
4089
4090/**
4091 * vfs_mkdir - create directory
4092 * @idmap: idmap of the mount the inode was found from
4093 * @dir: inode of @dentry
4094 * @dentry: pointer to dentry of the base directory
4095 * @mode: mode of the new directory
4096 *
4097 * Create a directory.
4098 *
4099 * If the inode has been found through an idmapped mount the idmap of
4100 * the vfsmount must be passed through @idmap. This function will then take
4101 * care to map the inode according to @idmap before checking permissions.
4102 * On non-idmapped mounts or if permission checking is to be performed on the
4103 * raw inode simply pass @nop_mnt_idmap.
4104 */
4105int vfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
4106 struct dentry *dentry, umode_t mode)
4107{
4108 int error;
4109 unsigned max_links = dir->i_sb->s_max_links;
4110
4111 error = may_create(idmap, dir, dentry);
4112 if (error)
4113 return error;
4114
4115 if (!dir->i_op->mkdir)
4116 return -EPERM;
4117
4118 mode = vfs_prepare_mode(idmap, dir, mode, S_IRWXUGO | S_ISVTX, 0);
4119 error = security_inode_mkdir(dir, dentry, mode);
4120 if (error)
4121 return error;
4122
4123 if (max_links && dir->i_nlink >= max_links)
4124 return -EMLINK;
4125
4126 error = dir->i_op->mkdir(idmap, dir, dentry, mode);
4127 if (!error)
4128 fsnotify_mkdir(dir, dentry);
4129 return error;
4130}
4131EXPORT_SYMBOL(vfs_mkdir);
4132
4133int do_mkdirat(int dfd, struct filename *name, umode_t mode)
4134{
4135 struct dentry *dentry;
4136 struct path path;
4137 int error;
4138 unsigned int lookup_flags = LOOKUP_DIRECTORY;
4139
4140retry:
4141 dentry = filename_create(dfd, name, &path, lookup_flags);
4142 error = PTR_ERR(dentry);
4143 if (IS_ERR(dentry))
4144 goto out_putname;
4145
4146 error = security_path_mkdir(&path, dentry,
4147 mode_strip_umask(path.dentry->d_inode, mode));
4148 if (!error) {
4149 error = vfs_mkdir(mnt_idmap(path.mnt), path.dentry->d_inode,
4150 dentry, mode);
4151 }
4152 done_path_create(&path, dentry);
4153 if (retry_estale(error, lookup_flags)) {
4154 lookup_flags |= LOOKUP_REVAL;
4155 goto retry;
4156 }
4157out_putname:
4158 putname(name);
4159 return error;
4160}
4161
4162SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
4163{
4164 return do_mkdirat(dfd, getname(pathname), mode);
4165}
4166
4167SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
4168{
4169 return do_mkdirat(AT_FDCWD, getname(pathname), mode);
4170}
4171
4172/**
4173 * vfs_rmdir - remove directory
4174 * @idmap: idmap of the mount the inode was found from
4175 * @dir: inode of @dentry
4176 * @dentry: pointer to dentry of the base directory
4177 *
4178 * Remove a directory.
4179 *
4180 * If the inode has been found through an idmapped mount the idmap of
4181 * the vfsmount must be passed through @idmap. This function will then take
4182 * care to map the inode according to @idmap before checking permissions.
4183 * On non-idmapped mounts or if permission checking is to be performed on the
4184 * raw inode simply pass @nop_mnt_idmap.
4185 */
4186int vfs_rmdir(struct mnt_idmap *idmap, struct inode *dir,
4187 struct dentry *dentry)
4188{
4189 int error = may_delete(idmap, dir, dentry, 1);
4190
4191 if (error)
4192 return error;
4193
4194 if (!dir->i_op->rmdir)
4195 return -EPERM;
4196
4197 dget(dentry);
4198 inode_lock(dentry->d_inode);
4199
4200 error = -EBUSY;
4201 if (is_local_mountpoint(dentry) ||
4202 (dentry->d_inode->i_flags & S_KERNEL_FILE))
4203 goto out;
4204
4205 error = security_inode_rmdir(dir, dentry);
4206 if (error)
4207 goto out;
4208
4209 error = dir->i_op->rmdir(dir, dentry);
4210 if (error)
4211 goto out;
4212
4213 shrink_dcache_parent(dentry);
4214 dentry->d_inode->i_flags |= S_DEAD;
4215 dont_mount(dentry);
4216 detach_mounts(dentry);
4217
4218out:
4219 inode_unlock(dentry->d_inode);
4220 dput(dentry);
4221 if (!error)
4222 d_delete_notify(dir, dentry);
4223 return error;
4224}
4225EXPORT_SYMBOL(vfs_rmdir);
4226
4227int do_rmdir(int dfd, struct filename *name)
4228{
4229 int error;
4230 struct dentry *dentry;
4231 struct path path;
4232 struct qstr last;
4233 int type;
4234 unsigned int lookup_flags = 0;
4235retry:
4236 error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4237 if (error)
4238 goto exit1;
4239
4240 switch (type) {
4241 case LAST_DOTDOT:
4242 error = -ENOTEMPTY;
4243 goto exit2;
4244 case LAST_DOT:
4245 error = -EINVAL;
4246 goto exit2;
4247 case LAST_ROOT:
4248 error = -EBUSY;
4249 goto exit2;
4250 }
4251
4252 error = mnt_want_write(path.mnt);
4253 if (error)
4254 goto exit2;
4255
4256 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4257 dentry = lookup_one_qstr_excl(&last, path.dentry, lookup_flags);
4258 error = PTR_ERR(dentry);
4259 if (IS_ERR(dentry))
4260 goto exit3;
4261 if (!dentry->d_inode) {
4262 error = -ENOENT;
4263 goto exit4;
4264 }
4265 error = security_path_rmdir(&path, dentry);
4266 if (error)
4267 goto exit4;
4268 error = vfs_rmdir(mnt_idmap(path.mnt), path.dentry->d_inode, dentry);
4269exit4:
4270 dput(dentry);
4271exit3:
4272 inode_unlock(path.dentry->d_inode);
4273 mnt_drop_write(path.mnt);
4274exit2:
4275 path_put(&path);
4276 if (retry_estale(error, lookup_flags)) {
4277 lookup_flags |= LOOKUP_REVAL;
4278 goto retry;
4279 }
4280exit1:
4281 putname(name);
4282 return error;
4283}
4284
4285SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
4286{
4287 return do_rmdir(AT_FDCWD, getname(pathname));
4288}
4289
4290/**
4291 * vfs_unlink - unlink a filesystem object
4292 * @idmap: idmap of the mount the inode was found from
4293 * @dir: parent directory
4294 * @dentry: victim
4295 * @delegated_inode: returns victim inode, if the inode is delegated.
4296 *
4297 * The caller must hold dir->i_mutex.
4298 *
4299 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
4300 * return a reference to the inode in delegated_inode. The caller
4301 * should then break the delegation on that inode and retry. Because
4302 * breaking a delegation may take a long time, the caller should drop
4303 * dir->i_mutex before doing so.
4304 *
4305 * Alternatively, a caller may pass NULL for delegated_inode. This may
4306 * be appropriate for callers that expect the underlying filesystem not
4307 * to be NFS exported.
4308 *
4309 * If the inode has been found through an idmapped mount the idmap of
4310 * the vfsmount must be passed through @idmap. This function will then take
4311 * care to map the inode according to @idmap before checking permissions.
4312 * On non-idmapped mounts or if permission checking is to be performed on the
4313 * raw inode simply pass @nop_mnt_idmap.
4314 */
4315int vfs_unlink(struct mnt_idmap *idmap, struct inode *dir,
4316 struct dentry *dentry, struct inode **delegated_inode)
4317{
4318 struct inode *target = dentry->d_inode;
4319 int error = may_delete(idmap, dir, dentry, 0);
4320
4321 if (error)
4322 return error;
4323
4324 if (!dir->i_op->unlink)
4325 return -EPERM;
4326
4327 inode_lock(target);
4328 if (IS_SWAPFILE(target))
4329 error = -EPERM;
4330 else if (is_local_mountpoint(dentry))
4331 error = -EBUSY;
4332 else {
4333 error = security_inode_unlink(dir, dentry);
4334 if (!error) {
4335 error = try_break_deleg(target, delegated_inode);
4336 if (error)
4337 goto out;
4338 error = dir->i_op->unlink(dir, dentry);
4339 if (!error) {
4340 dont_mount(dentry);
4341 detach_mounts(dentry);
4342 }
4343 }
4344 }
4345out:
4346 inode_unlock(target);
4347
4348 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4349 if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) {
4350 fsnotify_unlink(dir, dentry);
4351 } else if (!error) {
4352 fsnotify_link_count(target);
4353 d_delete_notify(dir, dentry);
4354 }
4355
4356 return error;
4357}
4358EXPORT_SYMBOL(vfs_unlink);
4359
4360/*
4361 * Make sure that the actual truncation of the file will occur outside its
4362 * directory's i_mutex. Truncate can take a long time if there is a lot of
4363 * writeout happening, and we don't want to prevent access to the directory
4364 * while waiting on the I/O.
4365 */
4366int do_unlinkat(int dfd, struct filename *name)
4367{
4368 int error;
4369 struct dentry *dentry;
4370 struct path path;
4371 struct qstr last;
4372 int type;
4373 struct inode *inode = NULL;
4374 struct inode *delegated_inode = NULL;
4375 unsigned int lookup_flags = 0;
4376retry:
4377 error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4378 if (error)
4379 goto exit1;
4380
4381 error = -EISDIR;
4382 if (type != LAST_NORM)
4383 goto exit2;
4384
4385 error = mnt_want_write(path.mnt);
4386 if (error)
4387 goto exit2;
4388retry_deleg:
4389 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4390 dentry = lookup_one_qstr_excl(&last, path.dentry, lookup_flags);
4391 error = PTR_ERR(dentry);
4392 if (!IS_ERR(dentry)) {
4393
4394 /* Why not before? Because we want correct error value */
4395 if (last.name[last.len] || d_is_negative(dentry))
4396 goto slashes;
4397 inode = dentry->d_inode;
4398 ihold(inode);
4399 error = security_path_unlink(&path, dentry);
4400 if (error)
4401 goto exit3;
4402 error = vfs_unlink(mnt_idmap(path.mnt), path.dentry->d_inode,
4403 dentry, &delegated_inode);
4404exit3:
4405 dput(dentry);
4406 }
4407 inode_unlock(path.dentry->d_inode);
4408 if (inode)
4409 iput(inode); /* truncate the inode here */
4410 inode = NULL;
4411 if (delegated_inode) {
4412 error = break_deleg_wait(&delegated_inode);
4413 if (!error)
4414 goto retry_deleg;
4415 }
4416 mnt_drop_write(path.mnt);
4417exit2:
4418 path_put(&path);
4419 if (retry_estale(error, lookup_flags)) {
4420 lookup_flags |= LOOKUP_REVAL;
4421 inode = NULL;
4422 goto retry;
4423 }
4424exit1:
4425 putname(name);
4426 return error;
4427
4428slashes:
4429 if (d_is_negative(dentry))
4430 error = -ENOENT;
4431 else if (d_is_dir(dentry))
4432 error = -EISDIR;
4433 else
4434 error = -ENOTDIR;
4435 goto exit3;
4436}
4437
4438SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4439{
4440 if ((flag & ~AT_REMOVEDIR) != 0)
4441 return -EINVAL;
4442
4443 if (flag & AT_REMOVEDIR)
4444 return do_rmdir(dfd, getname(pathname));
4445 return do_unlinkat(dfd, getname(pathname));
4446}
4447
4448SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4449{
4450 return do_unlinkat(AT_FDCWD, getname(pathname));
4451}
4452
4453/**
4454 * vfs_symlink - create symlink
4455 * @idmap: idmap of the mount the inode was found from
4456 * @dir: inode of @dentry
4457 * @dentry: pointer to dentry of the base directory
4458 * @oldname: name of the file to link to
4459 *
4460 * Create a symlink.
4461 *
4462 * If the inode has been found through an idmapped mount the idmap of
4463 * the vfsmount must be passed through @idmap. This function will then take
4464 * care to map the inode according to @idmap before checking permissions.
4465 * On non-idmapped mounts or if permission checking is to be performed on the
4466 * raw inode simply pass @nop_mnt_idmap.
4467 */
4468int vfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
4469 struct dentry *dentry, const char *oldname)
4470{
4471 int error;
4472
4473 error = may_create(idmap, dir, dentry);
4474 if (error)
4475 return error;
4476
4477 if (!dir->i_op->symlink)
4478 return -EPERM;
4479
4480 error = security_inode_symlink(dir, dentry, oldname);
4481 if (error)
4482 return error;
4483
4484 error = dir->i_op->symlink(idmap, dir, dentry, oldname);
4485 if (!error)
4486 fsnotify_create(dir, dentry);
4487 return error;
4488}
4489EXPORT_SYMBOL(vfs_symlink);
4490
4491int do_symlinkat(struct filename *from, int newdfd, struct filename *to)
4492{
4493 int error;
4494 struct dentry *dentry;
4495 struct path path;
4496 unsigned int lookup_flags = 0;
4497
4498 if (IS_ERR(from)) {
4499 error = PTR_ERR(from);
4500 goto out_putnames;
4501 }
4502retry:
4503 dentry = filename_create(newdfd, to, &path, lookup_flags);
4504 error = PTR_ERR(dentry);
4505 if (IS_ERR(dentry))
4506 goto out_putnames;
4507
4508 error = security_path_symlink(&path, dentry, from->name);
4509 if (!error)
4510 error = vfs_symlink(mnt_idmap(path.mnt), path.dentry->d_inode,
4511 dentry, from->name);
4512 done_path_create(&path, dentry);
4513 if (retry_estale(error, lookup_flags)) {
4514 lookup_flags |= LOOKUP_REVAL;
4515 goto retry;
4516 }
4517out_putnames:
4518 putname(to);
4519 putname(from);
4520 return error;
4521}
4522
4523SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4524 int, newdfd, const char __user *, newname)
4525{
4526 return do_symlinkat(getname(oldname), newdfd, getname(newname));
4527}
4528
4529SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4530{
4531 return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname));
4532}
4533
4534/**
4535 * vfs_link - create a new link
4536 * @old_dentry: object to be linked
4537 * @idmap: idmap of the mount
4538 * @dir: new parent
4539 * @new_dentry: where to create the new link
4540 * @delegated_inode: returns inode needing a delegation break
4541 *
4542 * The caller must hold dir->i_mutex
4543 *
4544 * If vfs_link discovers a delegation on the to-be-linked file in need
4545 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4546 * inode in delegated_inode. The caller should then break the delegation
4547 * and retry. Because breaking a delegation may take a long time, the
4548 * caller should drop the i_mutex before doing so.
4549 *
4550 * Alternatively, a caller may pass NULL for delegated_inode. This may
4551 * be appropriate for callers that expect the underlying filesystem not
4552 * to be NFS exported.
4553 *
4554 * If the inode has been found through an idmapped mount the idmap of
4555 * the vfsmount must be passed through @idmap. This function will then take
4556 * care to map the inode according to @idmap before checking permissions.
4557 * On non-idmapped mounts or if permission checking is to be performed on the
4558 * raw inode simply pass @nop_mnt_idmap.
4559 */
4560int vfs_link(struct dentry *old_dentry, struct mnt_idmap *idmap,
4561 struct inode *dir, struct dentry *new_dentry,
4562 struct inode **delegated_inode)
4563{
4564 struct inode *inode = old_dentry->d_inode;
4565 unsigned max_links = dir->i_sb->s_max_links;
4566 int error;
4567
4568 if (!inode)
4569 return -ENOENT;
4570
4571 error = may_create(idmap, dir, new_dentry);
4572 if (error)
4573 return error;
4574
4575 if (dir->i_sb != inode->i_sb)
4576 return -EXDEV;
4577
4578 /*
4579 * A link to an append-only or immutable file cannot be created.
4580 */
4581 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4582 return -EPERM;
4583 /*
4584 * Updating the link count will likely cause i_uid and i_gid to
4585 * be writen back improperly if their true value is unknown to
4586 * the vfs.
4587 */
4588 if (HAS_UNMAPPED_ID(idmap, inode))
4589 return -EPERM;
4590 if (!dir->i_op->link)
4591 return -EPERM;
4592 if (S_ISDIR(inode->i_mode))
4593 return -EPERM;
4594
4595 error = security_inode_link(old_dentry, dir, new_dentry);
4596 if (error)
4597 return error;
4598
4599 inode_lock(inode);
4600 /* Make sure we don't allow creating hardlink to an unlinked file */
4601 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4602 error = -ENOENT;
4603 else if (max_links && inode->i_nlink >= max_links)
4604 error = -EMLINK;
4605 else {
4606 error = try_break_deleg(inode, delegated_inode);
4607 if (!error)
4608 error = dir->i_op->link(old_dentry, dir, new_dentry);
4609 }
4610
4611 if (!error && (inode->i_state & I_LINKABLE)) {
4612 spin_lock(&inode->i_lock);
4613 inode->i_state &= ~I_LINKABLE;
4614 spin_unlock(&inode->i_lock);
4615 }
4616 inode_unlock(inode);
4617 if (!error)
4618 fsnotify_link(dir, inode, new_dentry);
4619 return error;
4620}
4621EXPORT_SYMBOL(vfs_link);
4622
4623/*
4624 * Hardlinks are often used in delicate situations. We avoid
4625 * security-related surprises by not following symlinks on the
4626 * newname. --KAB
4627 *
4628 * We don't follow them on the oldname either to be compatible
4629 * with linux 2.0, and to avoid hard-linking to directories
4630 * and other special files. --ADM
4631 */
4632int do_linkat(int olddfd, struct filename *old, int newdfd,
4633 struct filename *new, int flags)
4634{
4635 struct mnt_idmap *idmap;
4636 struct dentry *new_dentry;
4637 struct path old_path, new_path;
4638 struct inode *delegated_inode = NULL;
4639 int how = 0;
4640 int error;
4641
4642 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) {
4643 error = -EINVAL;
4644 goto out_putnames;
4645 }
4646 /*
4647 * To use null names we require CAP_DAC_READ_SEARCH
4648 * This ensures that not everyone will be able to create
4649 * handlink using the passed filedescriptor.
4650 */
4651 if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) {
4652 error = -ENOENT;
4653 goto out_putnames;
4654 }
4655
4656 if (flags & AT_SYMLINK_FOLLOW)
4657 how |= LOOKUP_FOLLOW;
4658retry:
4659 error = filename_lookup(olddfd, old, how, &old_path, NULL);
4660 if (error)
4661 goto out_putnames;
4662
4663 new_dentry = filename_create(newdfd, new, &new_path,
4664 (how & LOOKUP_REVAL));
4665 error = PTR_ERR(new_dentry);
4666 if (IS_ERR(new_dentry))
4667 goto out_putpath;
4668
4669 error = -EXDEV;
4670 if (old_path.mnt != new_path.mnt)
4671 goto out_dput;
4672 idmap = mnt_idmap(new_path.mnt);
4673 error = may_linkat(idmap, &old_path);
4674 if (unlikely(error))
4675 goto out_dput;
4676 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4677 if (error)
4678 goto out_dput;
4679 error = vfs_link(old_path.dentry, idmap, new_path.dentry->d_inode,
4680 new_dentry, &delegated_inode);
4681out_dput:
4682 done_path_create(&new_path, new_dentry);
4683 if (delegated_inode) {
4684 error = break_deleg_wait(&delegated_inode);
4685 if (!error) {
4686 path_put(&old_path);
4687 goto retry;
4688 }
4689 }
4690 if (retry_estale(error, how)) {
4691 path_put(&old_path);
4692 how |= LOOKUP_REVAL;
4693 goto retry;
4694 }
4695out_putpath:
4696 path_put(&old_path);
4697out_putnames:
4698 putname(old);
4699 putname(new);
4700
4701 return error;
4702}
4703
4704SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4705 int, newdfd, const char __user *, newname, int, flags)
4706{
4707 return do_linkat(olddfd, getname_uflags(oldname, flags),
4708 newdfd, getname(newname), flags);
4709}
4710
4711SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4712{
4713 return do_linkat(AT_FDCWD, getname(oldname), AT_FDCWD, getname(newname), 0);
4714}
4715
4716/**
4717 * vfs_rename - rename a filesystem object
4718 * @rd: pointer to &struct renamedata info
4719 *
4720 * The caller must hold multiple mutexes--see lock_rename()).
4721 *
4722 * If vfs_rename discovers a delegation in need of breaking at either
4723 * the source or destination, it will return -EWOULDBLOCK and return a
4724 * reference to the inode in delegated_inode. The caller should then
4725 * break the delegation and retry. Because breaking a delegation may
4726 * take a long time, the caller should drop all locks before doing
4727 * so.
4728 *
4729 * Alternatively, a caller may pass NULL for delegated_inode. This may
4730 * be appropriate for callers that expect the underlying filesystem not
4731 * to be NFS exported.
4732 *
4733 * The worst of all namespace operations - renaming directory. "Perverted"
4734 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4735 * Problems:
4736 *
4737 * a) we can get into loop creation.
4738 * b) race potential - two innocent renames can create a loop together.
4739 * That's where 4.4BSD screws up. Current fix: serialization on
4740 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4741 * story.
4742 * c) we may have to lock up to _four_ objects - parents and victim (if it exists),
4743 * and source (if it's a non-directory or a subdirectory that moves to
4744 * different parent).
4745 * And that - after we got ->i_mutex on parents (until then we don't know
4746 * whether the target exists). Solution: try to be smart with locking
4747 * order for inodes. We rely on the fact that tree topology may change
4748 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4749 * move will be locked. Thus we can rank directories by the tree
4750 * (ancestors first) and rank all non-directories after them.
4751 * That works since everybody except rename does "lock parent, lookup,
4752 * lock child" and rename is under ->s_vfs_rename_mutex.
4753 * HOWEVER, it relies on the assumption that any object with ->lookup()
4754 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4755 * we'd better make sure that there's no link(2) for them.
4756 * d) conversion from fhandle to dentry may come in the wrong moment - when
4757 * we are removing the target. Solution: we will have to grab ->i_mutex
4758 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4759 * ->i_mutex on parents, which works but leads to some truly excessive
4760 * locking].
4761 */
4762int vfs_rename(struct renamedata *rd)
4763{
4764 int error;
4765 struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir;
4766 struct dentry *old_dentry = rd->old_dentry;
4767 struct dentry *new_dentry = rd->new_dentry;
4768 struct inode **delegated_inode = rd->delegated_inode;
4769 unsigned int flags = rd->flags;
4770 bool is_dir = d_is_dir(old_dentry);
4771 struct inode *source = old_dentry->d_inode;
4772 struct inode *target = new_dentry->d_inode;
4773 bool new_is_dir = false;
4774 unsigned max_links = new_dir->i_sb->s_max_links;
4775 struct name_snapshot old_name;
4776 bool lock_old_subdir, lock_new_subdir;
4777
4778 if (source == target)
4779 return 0;
4780
4781 error = may_delete(rd->old_mnt_idmap, old_dir, old_dentry, is_dir);
4782 if (error)
4783 return error;
4784
4785 if (!target) {
4786 error = may_create(rd->new_mnt_idmap, new_dir, new_dentry);
4787 } else {
4788 new_is_dir = d_is_dir(new_dentry);
4789
4790 if (!(flags & RENAME_EXCHANGE))
4791 error = may_delete(rd->new_mnt_idmap, new_dir,
4792 new_dentry, is_dir);
4793 else
4794 error = may_delete(rd->new_mnt_idmap, new_dir,
4795 new_dentry, new_is_dir);
4796 }
4797 if (error)
4798 return error;
4799
4800 if (!old_dir->i_op->rename)
4801 return -EPERM;
4802
4803 /*
4804 * If we are going to change the parent - check write permissions,
4805 * we'll need to flip '..'.
4806 */
4807 if (new_dir != old_dir) {
4808 if (is_dir) {
4809 error = inode_permission(rd->old_mnt_idmap, source,
4810 MAY_WRITE);
4811 if (error)
4812 return error;
4813 }
4814 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4815 error = inode_permission(rd->new_mnt_idmap, target,
4816 MAY_WRITE);
4817 if (error)
4818 return error;
4819 }
4820 }
4821
4822 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4823 flags);
4824 if (error)
4825 return error;
4826
4827 take_dentry_name_snapshot(&old_name, old_dentry);
4828 dget(new_dentry);
4829 /*
4830 * Lock children.
4831 * The source subdirectory needs to be locked on cross-directory
4832 * rename or cross-directory exchange since its parent changes.
4833 * The target subdirectory needs to be locked on cross-directory
4834 * exchange due to parent change and on any rename due to becoming
4835 * a victim.
4836 * Non-directories need locking in all cases (for NFS reasons);
4837 * they get locked after any subdirectories (in inode address order).
4838 *
4839 * NOTE: WE ONLY LOCK UNRELATED DIRECTORIES IN CROSS-DIRECTORY CASE.
4840 * NEVER, EVER DO THAT WITHOUT ->s_vfs_rename_mutex.
4841 */
4842 lock_old_subdir = new_dir != old_dir;
4843 lock_new_subdir = new_dir != old_dir || !(flags & RENAME_EXCHANGE);
4844 if (is_dir) {
4845 if (lock_old_subdir)
4846 inode_lock_nested(source, I_MUTEX_CHILD);
4847 if (target && (!new_is_dir || lock_new_subdir))
4848 inode_lock(target);
4849 } else if (new_is_dir) {
4850 if (lock_new_subdir)
4851 inode_lock_nested(target, I_MUTEX_CHILD);
4852 inode_lock(source);
4853 } else {
4854 lock_two_nondirectories(source, target);
4855 }
4856
4857 error = -EPERM;
4858 if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target)))
4859 goto out;
4860
4861 error = -EBUSY;
4862 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4863 goto out;
4864
4865 if (max_links && new_dir != old_dir) {
4866 error = -EMLINK;
4867 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4868 goto out;
4869 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4870 old_dir->i_nlink >= max_links)
4871 goto out;
4872 }
4873 if (!is_dir) {
4874 error = try_break_deleg(source, delegated_inode);
4875 if (error)
4876 goto out;
4877 }
4878 if (target && !new_is_dir) {
4879 error = try_break_deleg(target, delegated_inode);
4880 if (error)
4881 goto out;
4882 }
4883 error = old_dir->i_op->rename(rd->new_mnt_idmap, old_dir, old_dentry,
4884 new_dir, new_dentry, flags);
4885 if (error)
4886 goto out;
4887
4888 if (!(flags & RENAME_EXCHANGE) && target) {
4889 if (is_dir) {
4890 shrink_dcache_parent(new_dentry);
4891 target->i_flags |= S_DEAD;
4892 }
4893 dont_mount(new_dentry);
4894 detach_mounts(new_dentry);
4895 }
4896 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4897 if (!(flags & RENAME_EXCHANGE))
4898 d_move(old_dentry, new_dentry);
4899 else
4900 d_exchange(old_dentry, new_dentry);
4901 }
4902out:
4903 if (!is_dir || lock_old_subdir)
4904 inode_unlock(source);
4905 if (target && (!new_is_dir || lock_new_subdir))
4906 inode_unlock(target);
4907 dput(new_dentry);
4908 if (!error) {
4909 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4910 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4911 if (flags & RENAME_EXCHANGE) {
4912 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4913 new_is_dir, NULL, new_dentry);
4914 }
4915 }
4916 release_dentry_name_snapshot(&old_name);
4917
4918 return error;
4919}
4920EXPORT_SYMBOL(vfs_rename);
4921
4922int do_renameat2(int olddfd, struct filename *from, int newdfd,
4923 struct filename *to, unsigned int flags)
4924{
4925 struct renamedata rd;
4926 struct dentry *old_dentry, *new_dentry;
4927 struct dentry *trap;
4928 struct path old_path, new_path;
4929 struct qstr old_last, new_last;
4930 int old_type, new_type;
4931 struct inode *delegated_inode = NULL;
4932 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4933 bool should_retry = false;
4934 int error = -EINVAL;
4935
4936 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4937 goto put_names;
4938
4939 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4940 (flags & RENAME_EXCHANGE))
4941 goto put_names;
4942
4943 if (flags & RENAME_EXCHANGE)
4944 target_flags = 0;
4945
4946retry:
4947 error = filename_parentat(olddfd, from, lookup_flags, &old_path,
4948 &old_last, &old_type);
4949 if (error)
4950 goto put_names;
4951
4952 error = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last,
4953 &new_type);
4954 if (error)
4955 goto exit1;
4956
4957 error = -EXDEV;
4958 if (old_path.mnt != new_path.mnt)
4959 goto exit2;
4960
4961 error = -EBUSY;
4962 if (old_type != LAST_NORM)
4963 goto exit2;
4964
4965 if (flags & RENAME_NOREPLACE)
4966 error = -EEXIST;
4967 if (new_type != LAST_NORM)
4968 goto exit2;
4969
4970 error = mnt_want_write(old_path.mnt);
4971 if (error)
4972 goto exit2;
4973
4974retry_deleg:
4975 trap = lock_rename(new_path.dentry, old_path.dentry);
4976 if (IS_ERR(trap)) {
4977 error = PTR_ERR(trap);
4978 goto exit_lock_rename;
4979 }
4980
4981 old_dentry = lookup_one_qstr_excl(&old_last, old_path.dentry,
4982 lookup_flags);
4983 error = PTR_ERR(old_dentry);
4984 if (IS_ERR(old_dentry))
4985 goto exit3;
4986 /* source must exist */
4987 error = -ENOENT;
4988 if (d_is_negative(old_dentry))
4989 goto exit4;
4990 new_dentry = lookup_one_qstr_excl(&new_last, new_path.dentry,
4991 lookup_flags | target_flags);
4992 error = PTR_ERR(new_dentry);
4993 if (IS_ERR(new_dentry))
4994 goto exit4;
4995 error = -EEXIST;
4996 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4997 goto exit5;
4998 if (flags & RENAME_EXCHANGE) {
4999 error = -ENOENT;
5000 if (d_is_negative(new_dentry))
5001 goto exit5;
5002
5003 if (!d_is_dir(new_dentry)) {
5004 error = -ENOTDIR;
5005 if (new_last.name[new_last.len])
5006 goto exit5;
5007 }
5008 }
5009 /* unless the source is a directory trailing slashes give -ENOTDIR */
5010 if (!d_is_dir(old_dentry)) {
5011 error = -ENOTDIR;
5012 if (old_last.name[old_last.len])
5013 goto exit5;
5014 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
5015 goto exit5;
5016 }
5017 /* source should not be ancestor of target */
5018 error = -EINVAL;
5019 if (old_dentry == trap)
5020 goto exit5;
5021 /* target should not be an ancestor of source */
5022 if (!(flags & RENAME_EXCHANGE))
5023 error = -ENOTEMPTY;
5024 if (new_dentry == trap)
5025 goto exit5;
5026
5027 error = security_path_rename(&old_path, old_dentry,
5028 &new_path, new_dentry, flags);
5029 if (error)
5030 goto exit5;
5031
5032 rd.old_dir = old_path.dentry->d_inode;
5033 rd.old_dentry = old_dentry;
5034 rd.old_mnt_idmap = mnt_idmap(old_path.mnt);
5035 rd.new_dir = new_path.dentry->d_inode;
5036 rd.new_dentry = new_dentry;
5037 rd.new_mnt_idmap = mnt_idmap(new_path.mnt);
5038 rd.delegated_inode = &delegated_inode;
5039 rd.flags = flags;
5040 error = vfs_rename(&rd);
5041exit5:
5042 dput(new_dentry);
5043exit4:
5044 dput(old_dentry);
5045exit3:
5046 unlock_rename(new_path.dentry, old_path.dentry);
5047exit_lock_rename:
5048 if (delegated_inode) {
5049 error = break_deleg_wait(&delegated_inode);
5050 if (!error)
5051 goto retry_deleg;
5052 }
5053 mnt_drop_write(old_path.mnt);
5054exit2:
5055 if (retry_estale(error, lookup_flags))
5056 should_retry = true;
5057 path_put(&new_path);
5058exit1:
5059 path_put(&old_path);
5060 if (should_retry) {
5061 should_retry = false;
5062 lookup_flags |= LOOKUP_REVAL;
5063 goto retry;
5064 }
5065put_names:
5066 putname(from);
5067 putname(to);
5068 return error;
5069}
5070
5071SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
5072 int, newdfd, const char __user *, newname, unsigned int, flags)
5073{
5074 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
5075 flags);
5076}
5077
5078SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
5079 int, newdfd, const char __user *, newname)
5080{
5081 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
5082 0);
5083}
5084
5085SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5086{
5087 return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD,
5088 getname(newname), 0);
5089}
5090
5091int readlink_copy(char __user *buffer, int buflen, const char *link)
5092{
5093 int len = PTR_ERR(link);
5094 if (IS_ERR(link))
5095 goto out;
5096
5097 len = strlen(link);
5098 if (len > (unsigned) buflen)
5099 len = buflen;
5100 if (copy_to_user(buffer, link, len))
5101 len = -EFAULT;
5102out:
5103 return len;
5104}
5105
5106/**
5107 * vfs_readlink - copy symlink body into userspace buffer
5108 * @dentry: dentry on which to get symbolic link
5109 * @buffer: user memory pointer
5110 * @buflen: size of buffer
5111 *
5112 * Does not touch atime. That's up to the caller if necessary
5113 *
5114 * Does not call security hook.
5115 */
5116int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5117{
5118 struct inode *inode = d_inode(dentry);
5119 DEFINE_DELAYED_CALL(done);
5120 const char *link;
5121 int res;
5122
5123 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
5124 if (unlikely(inode->i_op->readlink))
5125 return inode->i_op->readlink(dentry, buffer, buflen);
5126
5127 if (!d_is_symlink(dentry))
5128 return -EINVAL;
5129
5130 spin_lock(&inode->i_lock);
5131 inode->i_opflags |= IOP_DEFAULT_READLINK;
5132 spin_unlock(&inode->i_lock);
5133 }
5134
5135 link = READ_ONCE(inode->i_link);
5136 if (!link) {
5137 link = inode->i_op->get_link(dentry, inode, &done);
5138 if (IS_ERR(link))
5139 return PTR_ERR(link);
5140 }
5141 res = readlink_copy(buffer, buflen, link);
5142 do_delayed_call(&done);
5143 return res;
5144}
5145EXPORT_SYMBOL(vfs_readlink);
5146
5147/**
5148 * vfs_get_link - get symlink body
5149 * @dentry: dentry on which to get symbolic link
5150 * @done: caller needs to free returned data with this
5151 *
5152 * Calls security hook and i_op->get_link() on the supplied inode.
5153 *
5154 * It does not touch atime. That's up to the caller if necessary.
5155 *
5156 * Does not work on "special" symlinks like /proc/$$/fd/N
5157 */
5158const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
5159{
5160 const char *res = ERR_PTR(-EINVAL);
5161 struct inode *inode = d_inode(dentry);
5162
5163 if (d_is_symlink(dentry)) {
5164 res = ERR_PTR(security_inode_readlink(dentry));
5165 if (!res)
5166 res = inode->i_op->get_link(dentry, inode, done);
5167 }
5168 return res;
5169}
5170EXPORT_SYMBOL(vfs_get_link);
5171
5172/* get the link contents into pagecache */
5173const char *page_get_link(struct dentry *dentry, struct inode *inode,
5174 struct delayed_call *callback)
5175{
5176 char *kaddr;
5177 struct page *page;
5178 struct address_space *mapping = inode->i_mapping;
5179
5180 if (!dentry) {
5181 page = find_get_page(mapping, 0);
5182 if (!page)
5183 return ERR_PTR(-ECHILD);
5184 if (!PageUptodate(page)) {
5185 put_page(page);
5186 return ERR_PTR(-ECHILD);
5187 }
5188 } else {
5189 page = read_mapping_page(mapping, 0, NULL);
5190 if (IS_ERR(page))
5191 return (char*)page;
5192 }
5193 set_delayed_call(callback, page_put_link, page);
5194 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
5195 kaddr = page_address(page);
5196 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
5197 return kaddr;
5198}
5199
5200EXPORT_SYMBOL(page_get_link);
5201
5202void page_put_link(void *arg)
5203{
5204 put_page(arg);
5205}
5206EXPORT_SYMBOL(page_put_link);
5207
5208int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5209{
5210 DEFINE_DELAYED_CALL(done);
5211 int res = readlink_copy(buffer, buflen,
5212 page_get_link(dentry, d_inode(dentry),
5213 &done));
5214 do_delayed_call(&done);
5215 return res;
5216}
5217EXPORT_SYMBOL(page_readlink);
5218
5219int page_symlink(struct inode *inode, const char *symname, int len)
5220{
5221 struct address_space *mapping = inode->i_mapping;
5222 const struct address_space_operations *aops = mapping->a_ops;
5223 bool nofs = !mapping_gfp_constraint(mapping, __GFP_FS);
5224 struct page *page;
5225 void *fsdata = NULL;
5226 int err;
5227 unsigned int flags;
5228
5229retry:
5230 if (nofs)
5231 flags = memalloc_nofs_save();
5232 err = aops->write_begin(NULL, mapping, 0, len-1, &page, &fsdata);
5233 if (nofs)
5234 memalloc_nofs_restore(flags);
5235 if (err)
5236 goto fail;
5237
5238 memcpy(page_address(page), symname, len-1);
5239
5240 err = aops->write_end(NULL, mapping, 0, len-1, len-1,
5241 page, fsdata);
5242 if (err < 0)
5243 goto fail;
5244 if (err < len-1)
5245 goto retry;
5246
5247 mark_inode_dirty(inode);
5248 return 0;
5249fail:
5250 return err;
5251}
5252EXPORT_SYMBOL(page_symlink);
5253
5254const struct inode_operations page_symlink_inode_operations = {
5255 .get_link = page_get_link,
5256};
5257EXPORT_SYMBOL(page_symlink_inode_operations);
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/fsnotify.h>
26#include <linux/personality.h>
27#include <linux/security.h>
28#include <linux/ima.h>
29#include <linux/syscalls.h>
30#include <linux/mount.h>
31#include <linux/audit.h>
32#include <linux/capability.h>
33#include <linux/file.h>
34#include <linux/fcntl.h>
35#include <linux/device_cgroup.h>
36#include <linux/fs_struct.h>
37#include <linux/posix_acl.h>
38#include <linux/hash.h>
39#include <linux/bitops.h>
40#include <linux/init_task.h>
41#include <linux/uaccess.h>
42
43#include "internal.h"
44#include "mount.h"
45
46/* [Feb-1997 T. Schoebel-Theuer]
47 * Fundamental changes in the pathname lookup mechanisms (namei)
48 * were necessary because of omirr. The reason is that omirr needs
49 * to know the _real_ pathname, not the user-supplied one, in case
50 * of symlinks (and also when transname replacements occur).
51 *
52 * The new code replaces the old recursive symlink resolution with
53 * an iterative one (in case of non-nested symlink chains). It does
54 * this with calls to <fs>_follow_link().
55 * As a side effect, dir_namei(), _namei() and follow_link() are now
56 * replaced with a single function lookup_dentry() that can handle all
57 * the special cases of the former code.
58 *
59 * With the new dcache, the pathname is stored at each inode, at least as
60 * long as the refcount of the inode is positive. As a side effect, the
61 * size of the dcache depends on the inode cache and thus is dynamic.
62 *
63 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
64 * resolution to correspond with current state of the code.
65 *
66 * Note that the symlink resolution is not *completely* iterative.
67 * There is still a significant amount of tail- and mid- recursion in
68 * the algorithm. Also, note that <fs>_readlink() is not used in
69 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
70 * may return different results than <fs>_follow_link(). Many virtual
71 * filesystems (including /proc) exhibit this behavior.
72 */
73
74/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
75 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
76 * and the name already exists in form of a symlink, try to create the new
77 * name indicated by the symlink. The old code always complained that the
78 * name already exists, due to not following the symlink even if its target
79 * is nonexistent. The new semantics affects also mknod() and link() when
80 * the name is a symlink pointing to a non-existent name.
81 *
82 * I don't know which semantics is the right one, since I have no access
83 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
84 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
85 * "old" one. Personally, I think the new semantics is much more logical.
86 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
87 * file does succeed in both HP-UX and SunOs, but not in Solaris
88 * and in the old Linux semantics.
89 */
90
91/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
92 * semantics. See the comments in "open_namei" and "do_link" below.
93 *
94 * [10-Sep-98 Alan Modra] Another symlink change.
95 */
96
97/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
98 * inside the path - always follow.
99 * in the last component in creation/removal/renaming - never follow.
100 * if LOOKUP_FOLLOW passed - follow.
101 * if the pathname has trailing slashes - follow.
102 * otherwise - don't follow.
103 * (applied in that order).
104 *
105 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
106 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
107 * During the 2.4 we need to fix the userland stuff depending on it -
108 * hopefully we will be able to get rid of that wart in 2.5. So far only
109 * XEmacs seems to be relying on it...
110 */
111/*
112 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
113 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
114 * any extra contention...
115 */
116
117/* In order to reduce some races, while at the same time doing additional
118 * checking and hopefully speeding things up, we copy filenames to the
119 * kernel data space before using them..
120 *
121 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
122 * PATH_MAX includes the nul terminator --RR.
123 */
124
125#define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
126
127struct filename *
128getname_flags(const char __user *filename, int flags, int *empty)
129{
130 struct filename *result;
131 char *kname;
132 int len;
133
134 result = audit_reusename(filename);
135 if (result)
136 return result;
137
138 result = __getname();
139 if (unlikely(!result))
140 return ERR_PTR(-ENOMEM);
141
142 /*
143 * First, try to embed the struct filename inside the names_cache
144 * allocation
145 */
146 kname = (char *)result->iname;
147 result->name = kname;
148
149 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
150 if (unlikely(len < 0)) {
151 __putname(result);
152 return ERR_PTR(len);
153 }
154
155 /*
156 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
157 * separate struct filename so we can dedicate the entire
158 * names_cache allocation for the pathname, and re-do the copy from
159 * userland.
160 */
161 if (unlikely(len == EMBEDDED_NAME_MAX)) {
162 const size_t size = offsetof(struct filename, iname[1]);
163 kname = (char *)result;
164
165 /*
166 * size is chosen that way we to guarantee that
167 * result->iname[0] is within the same object and that
168 * kname can't be equal to result->iname, no matter what.
169 */
170 result = kzalloc(size, GFP_KERNEL);
171 if (unlikely(!result)) {
172 __putname(kname);
173 return ERR_PTR(-ENOMEM);
174 }
175 result->name = kname;
176 len = strncpy_from_user(kname, filename, PATH_MAX);
177 if (unlikely(len < 0)) {
178 __putname(kname);
179 kfree(result);
180 return ERR_PTR(len);
181 }
182 if (unlikely(len == PATH_MAX)) {
183 __putname(kname);
184 kfree(result);
185 return ERR_PTR(-ENAMETOOLONG);
186 }
187 }
188
189 result->refcnt = 1;
190 /* The empty path is special. */
191 if (unlikely(!len)) {
192 if (empty)
193 *empty = 1;
194 if (!(flags & LOOKUP_EMPTY)) {
195 putname(result);
196 return ERR_PTR(-ENOENT);
197 }
198 }
199
200 result->uptr = filename;
201 result->aname = NULL;
202 audit_getname(result);
203 return result;
204}
205
206struct filename *
207getname(const char __user * filename)
208{
209 return getname_flags(filename, 0, NULL);
210}
211
212struct filename *
213getname_kernel(const char * filename)
214{
215 struct filename *result;
216 int len = strlen(filename) + 1;
217
218 result = __getname();
219 if (unlikely(!result))
220 return ERR_PTR(-ENOMEM);
221
222 if (len <= EMBEDDED_NAME_MAX) {
223 result->name = (char *)result->iname;
224 } else if (len <= PATH_MAX) {
225 const size_t size = offsetof(struct filename, iname[1]);
226 struct filename *tmp;
227
228 tmp = kmalloc(size, GFP_KERNEL);
229 if (unlikely(!tmp)) {
230 __putname(result);
231 return ERR_PTR(-ENOMEM);
232 }
233 tmp->name = (char *)result;
234 result = tmp;
235 } else {
236 __putname(result);
237 return ERR_PTR(-ENAMETOOLONG);
238 }
239 memcpy((char *)result->name, filename, len);
240 result->uptr = NULL;
241 result->aname = NULL;
242 result->refcnt = 1;
243 audit_getname(result);
244
245 return result;
246}
247
248void putname(struct filename *name)
249{
250 BUG_ON(name->refcnt <= 0);
251
252 if (--name->refcnt > 0)
253 return;
254
255 if (name->name != name->iname) {
256 __putname(name->name);
257 kfree(name);
258 } else
259 __putname(name);
260}
261
262static int check_acl(struct inode *inode, int mask)
263{
264#ifdef CONFIG_FS_POSIX_ACL
265 struct posix_acl *acl;
266
267 if (mask & MAY_NOT_BLOCK) {
268 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
269 if (!acl)
270 return -EAGAIN;
271 /* no ->get_acl() calls in RCU mode... */
272 if (is_uncached_acl(acl))
273 return -ECHILD;
274 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
275 }
276
277 acl = get_acl(inode, ACL_TYPE_ACCESS);
278 if (IS_ERR(acl))
279 return PTR_ERR(acl);
280 if (acl) {
281 int error = posix_acl_permission(inode, acl, mask);
282 posix_acl_release(acl);
283 return error;
284 }
285#endif
286
287 return -EAGAIN;
288}
289
290/*
291 * This does the basic permission checking
292 */
293static int acl_permission_check(struct inode *inode, int mask)
294{
295 unsigned int mode = inode->i_mode;
296
297 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
298 mode >>= 6;
299 else {
300 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
301 int error = check_acl(inode, mask);
302 if (error != -EAGAIN)
303 return error;
304 }
305
306 if (in_group_p(inode->i_gid))
307 mode >>= 3;
308 }
309
310 /*
311 * If the DACs are ok we don't need any capability check.
312 */
313 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
314 return 0;
315 return -EACCES;
316}
317
318/**
319 * generic_permission - check for access rights on a Posix-like filesystem
320 * @inode: inode to check access rights for
321 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
322 *
323 * Used to check for read/write/execute permissions on a file.
324 * We use "fsuid" for this, letting us set arbitrary permissions
325 * for filesystem access without changing the "normal" uids which
326 * are used for other things.
327 *
328 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
329 * request cannot be satisfied (eg. requires blocking or too much complexity).
330 * It would then be called again in ref-walk mode.
331 */
332int generic_permission(struct inode *inode, int mask)
333{
334 int ret;
335
336 /*
337 * Do the basic permission checks.
338 */
339 ret = acl_permission_check(inode, mask);
340 if (ret != -EACCES)
341 return ret;
342
343 if (S_ISDIR(inode->i_mode)) {
344 /* DACs are overridable for directories */
345 if (!(mask & MAY_WRITE))
346 if (capable_wrt_inode_uidgid(inode,
347 CAP_DAC_READ_SEARCH))
348 return 0;
349 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
350 return 0;
351 return -EACCES;
352 }
353
354 /*
355 * Searching includes executable on directories, else just read.
356 */
357 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
358 if (mask == MAY_READ)
359 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
360 return 0;
361 /*
362 * Read/write DACs are always overridable.
363 * Executable DACs are overridable when there is
364 * at least one exec bit set.
365 */
366 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
367 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
368 return 0;
369
370 return -EACCES;
371}
372EXPORT_SYMBOL(generic_permission);
373
374/*
375 * We _really_ want to just do "generic_permission()" without
376 * even looking at the inode->i_op values. So we keep a cache
377 * flag in inode->i_opflags, that says "this has not special
378 * permission function, use the fast case".
379 */
380static inline int do_inode_permission(struct inode *inode, int mask)
381{
382 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
383 if (likely(inode->i_op->permission))
384 return inode->i_op->permission(inode, mask);
385
386 /* This gets set once for the inode lifetime */
387 spin_lock(&inode->i_lock);
388 inode->i_opflags |= IOP_FASTPERM;
389 spin_unlock(&inode->i_lock);
390 }
391 return generic_permission(inode, mask);
392}
393
394/**
395 * sb_permission - Check superblock-level permissions
396 * @sb: Superblock of inode to check permission on
397 * @inode: Inode to check permission on
398 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
399 *
400 * Separate out file-system wide checks from inode-specific permission checks.
401 */
402static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
403{
404 if (unlikely(mask & MAY_WRITE)) {
405 umode_t mode = inode->i_mode;
406
407 /* Nobody gets write access to a read-only fs. */
408 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
409 return -EROFS;
410 }
411 return 0;
412}
413
414/**
415 * inode_permission - Check for access rights to a given inode
416 * @inode: Inode to check permission on
417 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
418 *
419 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
420 * this, letting us set arbitrary permissions for filesystem access without
421 * changing the "normal" UIDs which are used for other things.
422 *
423 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
424 */
425int inode_permission(struct inode *inode, int mask)
426{
427 int retval;
428
429 retval = sb_permission(inode->i_sb, inode, mask);
430 if (retval)
431 return retval;
432
433 if (unlikely(mask & MAY_WRITE)) {
434 /*
435 * Nobody gets write access to an immutable file.
436 */
437 if (IS_IMMUTABLE(inode))
438 return -EPERM;
439
440 /*
441 * Updating mtime will likely cause i_uid and i_gid to be
442 * written back improperly if their true value is unknown
443 * to the vfs.
444 */
445 if (HAS_UNMAPPED_ID(inode))
446 return -EACCES;
447 }
448
449 retval = do_inode_permission(inode, mask);
450 if (retval)
451 return retval;
452
453 retval = devcgroup_inode_permission(inode, mask);
454 if (retval)
455 return retval;
456
457 return security_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#define EMBEDDED_LEVELS 2
488struct nameidata {
489 struct path path;
490 struct qstr last;
491 struct path root;
492 struct inode *inode; /* path.dentry.d_inode */
493 unsigned int flags;
494 unsigned seq, m_seq;
495 int last_type;
496 unsigned depth;
497 int total_link_count;
498 struct saved {
499 struct path link;
500 struct delayed_call done;
501 const char *name;
502 unsigned seq;
503 } *stack, internal[EMBEDDED_LEVELS];
504 struct filename *name;
505 struct nameidata *saved;
506 struct inode *link_inode;
507 unsigned root_seq;
508 int dfd;
509} __randomize_layout;
510
511static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
512{
513 struct nameidata *old = current->nameidata;
514 p->stack = p->internal;
515 p->dfd = dfd;
516 p->name = name;
517 p->total_link_count = old ? old->total_link_count : 0;
518 p->saved = old;
519 current->nameidata = p;
520}
521
522static void restore_nameidata(void)
523{
524 struct nameidata *now = current->nameidata, *old = now->saved;
525
526 current->nameidata = old;
527 if (old)
528 old->total_link_count = now->total_link_count;
529 if (now->stack != now->internal)
530 kfree(now->stack);
531}
532
533static int __nd_alloc_stack(struct nameidata *nd)
534{
535 struct saved *p;
536
537 if (nd->flags & LOOKUP_RCU) {
538 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
539 GFP_ATOMIC);
540 if (unlikely(!p))
541 return -ECHILD;
542 } else {
543 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
544 GFP_KERNEL);
545 if (unlikely(!p))
546 return -ENOMEM;
547 }
548 memcpy(p, nd->internal, sizeof(nd->internal));
549 nd->stack = p;
550 return 0;
551}
552
553/**
554 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
555 * @path: nameidate to verify
556 *
557 * Rename can sometimes move a file or directory outside of a bind
558 * mount, path_connected allows those cases to be detected.
559 */
560static bool path_connected(const struct path *path)
561{
562 struct vfsmount *mnt = path->mnt;
563 struct super_block *sb = mnt->mnt_sb;
564
565 /* Bind mounts and multi-root filesystems can have disconnected paths */
566 if (!(sb->s_iflags & SB_I_MULTIROOT) && (mnt->mnt_root == sb->s_root))
567 return true;
568
569 return is_subdir(path->dentry, mnt->mnt_root);
570}
571
572static inline int nd_alloc_stack(struct nameidata *nd)
573{
574 if (likely(nd->depth != EMBEDDED_LEVELS))
575 return 0;
576 if (likely(nd->stack != nd->internal))
577 return 0;
578 return __nd_alloc_stack(nd);
579}
580
581static void drop_links(struct nameidata *nd)
582{
583 int i = nd->depth;
584 while (i--) {
585 struct saved *last = nd->stack + i;
586 do_delayed_call(&last->done);
587 clear_delayed_call(&last->done);
588 }
589}
590
591static void terminate_walk(struct nameidata *nd)
592{
593 drop_links(nd);
594 if (!(nd->flags & LOOKUP_RCU)) {
595 int i;
596 path_put(&nd->path);
597 for (i = 0; i < nd->depth; i++)
598 path_put(&nd->stack[i].link);
599 if (nd->flags & LOOKUP_ROOT_GRABBED) {
600 path_put(&nd->root);
601 nd->flags &= ~LOOKUP_ROOT_GRABBED;
602 }
603 } else {
604 nd->flags &= ~LOOKUP_RCU;
605 rcu_read_unlock();
606 }
607 nd->depth = 0;
608}
609
610/* path_put is needed afterwards regardless of success or failure */
611static bool legitimize_path(struct nameidata *nd,
612 struct path *path, unsigned seq)
613{
614 int res = __legitimize_mnt(path->mnt, nd->m_seq);
615 if (unlikely(res)) {
616 if (res > 0)
617 path->mnt = NULL;
618 path->dentry = NULL;
619 return false;
620 }
621 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
622 path->dentry = NULL;
623 return false;
624 }
625 return !read_seqcount_retry(&path->dentry->d_seq, seq);
626}
627
628static bool legitimize_links(struct nameidata *nd)
629{
630 int i;
631 for (i = 0; i < nd->depth; i++) {
632 struct saved *last = nd->stack + i;
633 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
634 drop_links(nd);
635 nd->depth = i + 1;
636 return false;
637 }
638 }
639 return true;
640}
641
642static bool legitimize_root(struct nameidata *nd)
643{
644 if (!nd->root.mnt || (nd->flags & LOOKUP_ROOT))
645 return true;
646 nd->flags |= LOOKUP_ROOT_GRABBED;
647 return legitimize_path(nd, &nd->root, nd->root_seq);
648}
649
650/*
651 * Path walking has 2 modes, rcu-walk and ref-walk (see
652 * Documentation/filesystems/path-lookup.txt). In situations when we can't
653 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
654 * normal reference counts on dentries and vfsmounts to transition to ref-walk
655 * mode. Refcounts are grabbed at the last known good point before rcu-walk
656 * got stuck, so ref-walk may continue from there. If this is not successful
657 * (eg. a seqcount has changed), then failure is returned and it's up to caller
658 * to restart the path walk from the beginning in ref-walk mode.
659 */
660
661/**
662 * unlazy_walk - try to switch to ref-walk mode.
663 * @nd: nameidata pathwalk data
664 * Returns: 0 on success, -ECHILD on failure
665 *
666 * unlazy_walk attempts to legitimize the current nd->path and nd->root
667 * for ref-walk mode.
668 * Must be called from rcu-walk context.
669 * Nothing should touch nameidata between unlazy_walk() failure and
670 * terminate_walk().
671 */
672static int unlazy_walk(struct nameidata *nd)
673{
674 struct dentry *parent = nd->path.dentry;
675
676 BUG_ON(!(nd->flags & LOOKUP_RCU));
677
678 nd->flags &= ~LOOKUP_RCU;
679 if (unlikely(!legitimize_links(nd)))
680 goto out1;
681 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
682 goto out;
683 if (unlikely(!legitimize_root(nd)))
684 goto out;
685 rcu_read_unlock();
686 BUG_ON(nd->inode != parent->d_inode);
687 return 0;
688
689out1:
690 nd->path.mnt = NULL;
691 nd->path.dentry = NULL;
692out:
693 rcu_read_unlock();
694 return -ECHILD;
695}
696
697/**
698 * unlazy_child - try to switch to ref-walk mode.
699 * @nd: nameidata pathwalk data
700 * @dentry: child of nd->path.dentry
701 * @seq: seq number to check dentry against
702 * Returns: 0 on success, -ECHILD on failure
703 *
704 * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry
705 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
706 * @nd. Must be called from rcu-walk context.
707 * Nothing should touch nameidata between unlazy_child() failure and
708 * terminate_walk().
709 */
710static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq)
711{
712 BUG_ON(!(nd->flags & LOOKUP_RCU));
713
714 nd->flags &= ~LOOKUP_RCU;
715 if (unlikely(!legitimize_links(nd)))
716 goto out2;
717 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
718 goto out2;
719 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
720 goto out1;
721
722 /*
723 * We need to move both the parent and the dentry from the RCU domain
724 * to be properly refcounted. And the sequence number in the dentry
725 * validates *both* dentry counters, since we checked the sequence
726 * number of the parent after we got the child sequence number. So we
727 * know the parent must still be valid if the child sequence number is
728 */
729 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
730 goto out;
731 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
732 goto out_dput;
733 /*
734 * Sequence counts matched. Now make sure that the root is
735 * still valid and get it if required.
736 */
737 if (unlikely(!legitimize_root(nd)))
738 goto out_dput;
739 rcu_read_unlock();
740 return 0;
741
742out2:
743 nd->path.mnt = NULL;
744out1:
745 nd->path.dentry = NULL;
746out:
747 rcu_read_unlock();
748 return -ECHILD;
749out_dput:
750 rcu_read_unlock();
751 dput(dentry);
752 return -ECHILD;
753}
754
755static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
756{
757 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
758 return dentry->d_op->d_revalidate(dentry, flags);
759 else
760 return 1;
761}
762
763/**
764 * complete_walk - successful completion of path walk
765 * @nd: pointer nameidata
766 *
767 * If we had been in RCU mode, drop out of it and legitimize nd->path.
768 * Revalidate the final result, unless we'd already done that during
769 * the path walk or the filesystem doesn't ask for it. Return 0 on
770 * success, -error on failure. In case of failure caller does not
771 * need to drop nd->path.
772 */
773static int complete_walk(struct nameidata *nd)
774{
775 struct dentry *dentry = nd->path.dentry;
776 int status;
777
778 if (nd->flags & LOOKUP_RCU) {
779 if (!(nd->flags & LOOKUP_ROOT))
780 nd->root.mnt = NULL;
781 if (unlikely(unlazy_walk(nd)))
782 return -ECHILD;
783 }
784
785 if (likely(!(nd->flags & LOOKUP_JUMPED)))
786 return 0;
787
788 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
789 return 0;
790
791 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
792 if (status > 0)
793 return 0;
794
795 if (!status)
796 status = -ESTALE;
797
798 return status;
799}
800
801static void set_root(struct nameidata *nd)
802{
803 struct fs_struct *fs = current->fs;
804
805 if (nd->flags & LOOKUP_RCU) {
806 unsigned seq;
807
808 do {
809 seq = read_seqcount_begin(&fs->seq);
810 nd->root = fs->root;
811 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
812 } while (read_seqcount_retry(&fs->seq, seq));
813 } else {
814 get_fs_root(fs, &nd->root);
815 nd->flags |= LOOKUP_ROOT_GRABBED;
816 }
817}
818
819static void path_put_conditional(struct path *path, struct nameidata *nd)
820{
821 dput(path->dentry);
822 if (path->mnt != nd->path.mnt)
823 mntput(path->mnt);
824}
825
826static inline void path_to_nameidata(const struct path *path,
827 struct nameidata *nd)
828{
829 if (!(nd->flags & LOOKUP_RCU)) {
830 dput(nd->path.dentry);
831 if (nd->path.mnt != path->mnt)
832 mntput(nd->path.mnt);
833 }
834 nd->path.mnt = path->mnt;
835 nd->path.dentry = path->dentry;
836}
837
838static int nd_jump_root(struct nameidata *nd)
839{
840 if (nd->flags & LOOKUP_RCU) {
841 struct dentry *d;
842 nd->path = nd->root;
843 d = nd->path.dentry;
844 nd->inode = d->d_inode;
845 nd->seq = nd->root_seq;
846 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
847 return -ECHILD;
848 } else {
849 path_put(&nd->path);
850 nd->path = nd->root;
851 path_get(&nd->path);
852 nd->inode = nd->path.dentry->d_inode;
853 }
854 nd->flags |= LOOKUP_JUMPED;
855 return 0;
856}
857
858/*
859 * Helper to directly jump to a known parsed path from ->get_link,
860 * caller must have taken a reference to path beforehand.
861 */
862void nd_jump_link(struct path *path)
863{
864 struct nameidata *nd = current->nameidata;
865 path_put(&nd->path);
866
867 nd->path = *path;
868 nd->inode = nd->path.dentry->d_inode;
869 nd->flags |= LOOKUP_JUMPED;
870}
871
872static inline void put_link(struct nameidata *nd)
873{
874 struct saved *last = nd->stack + --nd->depth;
875 do_delayed_call(&last->done);
876 if (!(nd->flags & LOOKUP_RCU))
877 path_put(&last->link);
878}
879
880int sysctl_protected_symlinks __read_mostly = 0;
881int sysctl_protected_hardlinks __read_mostly = 0;
882int sysctl_protected_fifos __read_mostly;
883int sysctl_protected_regular __read_mostly;
884
885/**
886 * may_follow_link - Check symlink following for unsafe situations
887 * @nd: nameidata pathwalk data
888 *
889 * In the case of the sysctl_protected_symlinks sysctl being enabled,
890 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
891 * in a sticky world-writable directory. This is to protect privileged
892 * processes from failing races against path names that may change out
893 * from under them by way of other users creating malicious symlinks.
894 * It will permit symlinks to be followed only when outside a sticky
895 * world-writable directory, or when the uid of the symlink and follower
896 * match, or when the directory owner matches the symlink's owner.
897 *
898 * Returns 0 if following the symlink is allowed, -ve on error.
899 */
900static inline int may_follow_link(struct nameidata *nd)
901{
902 const struct inode *inode;
903 const struct inode *parent;
904 kuid_t puid;
905
906 if (!sysctl_protected_symlinks)
907 return 0;
908
909 /* Allowed if owner and follower match. */
910 inode = nd->link_inode;
911 if (uid_eq(current_cred()->fsuid, inode->i_uid))
912 return 0;
913
914 /* Allowed if parent directory not sticky and world-writable. */
915 parent = nd->inode;
916 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
917 return 0;
918
919 /* Allowed if parent directory and link owner match. */
920 puid = parent->i_uid;
921 if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
922 return 0;
923
924 if (nd->flags & LOOKUP_RCU)
925 return -ECHILD;
926
927 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
928 audit_log_link_denied("follow_link");
929 return -EACCES;
930}
931
932/**
933 * safe_hardlink_source - Check for safe hardlink conditions
934 * @inode: the source inode to hardlink from
935 *
936 * Return false if at least one of the following conditions:
937 * - inode is not a regular file
938 * - inode is setuid
939 * - inode is setgid and group-exec
940 * - access failure for read and write
941 *
942 * Otherwise returns true.
943 */
944static bool safe_hardlink_source(struct inode *inode)
945{
946 umode_t mode = inode->i_mode;
947
948 /* Special files should not get pinned to the filesystem. */
949 if (!S_ISREG(mode))
950 return false;
951
952 /* Setuid files should not get pinned to the filesystem. */
953 if (mode & S_ISUID)
954 return false;
955
956 /* Executable setgid files should not get pinned to the filesystem. */
957 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
958 return false;
959
960 /* Hardlinking to unreadable or unwritable sources is dangerous. */
961 if (inode_permission(inode, MAY_READ | MAY_WRITE))
962 return false;
963
964 return true;
965}
966
967/**
968 * may_linkat - Check permissions for creating a hardlink
969 * @link: the source to hardlink from
970 *
971 * Block hardlink when all of:
972 * - sysctl_protected_hardlinks enabled
973 * - fsuid does not match inode
974 * - hardlink source is unsafe (see safe_hardlink_source() above)
975 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
976 *
977 * Returns 0 if successful, -ve on error.
978 */
979static int may_linkat(struct path *link)
980{
981 struct inode *inode = link->dentry->d_inode;
982
983 /* Inode writeback is not safe when the uid or gid are invalid. */
984 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
985 return -EOVERFLOW;
986
987 if (!sysctl_protected_hardlinks)
988 return 0;
989
990 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
991 * otherwise, it must be a safe source.
992 */
993 if (safe_hardlink_source(inode) || inode_owner_or_capable(inode))
994 return 0;
995
996 audit_log_link_denied("linkat");
997 return -EPERM;
998}
999
1000/**
1001 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1002 * should be allowed, or not, on files that already
1003 * exist.
1004 * @dir: the sticky parent directory
1005 * @inode: the inode of the file to open
1006 *
1007 * Block an O_CREAT open of a FIFO (or a regular file) when:
1008 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1009 * - the file already exists
1010 * - we are in a sticky directory
1011 * - we don't own the file
1012 * - the owner of the directory doesn't own the file
1013 * - the directory is world writable
1014 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1015 * the directory doesn't have to be world writable: being group writable will
1016 * be enough.
1017 *
1018 * Returns 0 if the open is allowed, -ve on error.
1019 */
1020static int may_create_in_sticky(struct dentry * const dir,
1021 struct inode * const inode)
1022{
1023 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1024 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1025 likely(!(dir->d_inode->i_mode & S_ISVTX)) ||
1026 uid_eq(inode->i_uid, dir->d_inode->i_uid) ||
1027 uid_eq(current_fsuid(), inode->i_uid))
1028 return 0;
1029
1030 if (likely(dir->d_inode->i_mode & 0002) ||
1031 (dir->d_inode->i_mode & 0020 &&
1032 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1033 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1034 return -EACCES;
1035 }
1036 return 0;
1037}
1038
1039static __always_inline
1040const char *get_link(struct nameidata *nd)
1041{
1042 struct saved *last = nd->stack + nd->depth - 1;
1043 struct dentry *dentry = last->link.dentry;
1044 struct inode *inode = nd->link_inode;
1045 int error;
1046 const char *res;
1047
1048 if (!(nd->flags & LOOKUP_RCU)) {
1049 touch_atime(&last->link);
1050 cond_resched();
1051 } else if (atime_needs_update(&last->link, inode)) {
1052 if (unlikely(unlazy_walk(nd)))
1053 return ERR_PTR(-ECHILD);
1054 touch_atime(&last->link);
1055 }
1056
1057 error = security_inode_follow_link(dentry, inode,
1058 nd->flags & LOOKUP_RCU);
1059 if (unlikely(error))
1060 return ERR_PTR(error);
1061
1062 nd->last_type = LAST_BIND;
1063 res = READ_ONCE(inode->i_link);
1064 if (!res) {
1065 const char * (*get)(struct dentry *, struct inode *,
1066 struct delayed_call *);
1067 get = inode->i_op->get_link;
1068 if (nd->flags & LOOKUP_RCU) {
1069 res = get(NULL, inode, &last->done);
1070 if (res == ERR_PTR(-ECHILD)) {
1071 if (unlikely(unlazy_walk(nd)))
1072 return ERR_PTR(-ECHILD);
1073 res = get(dentry, inode, &last->done);
1074 }
1075 } else {
1076 res = get(dentry, inode, &last->done);
1077 }
1078 if (IS_ERR_OR_NULL(res))
1079 return res;
1080 }
1081 if (*res == '/') {
1082 if (!nd->root.mnt)
1083 set_root(nd);
1084 if (unlikely(nd_jump_root(nd)))
1085 return ERR_PTR(-ECHILD);
1086 while (unlikely(*++res == '/'))
1087 ;
1088 }
1089 if (!*res)
1090 res = NULL;
1091 return res;
1092}
1093
1094/*
1095 * follow_up - Find the mountpoint of path's vfsmount
1096 *
1097 * Given a path, find the mountpoint of its source file system.
1098 * Replace @path with the path of the mountpoint in the parent mount.
1099 * Up is towards /.
1100 *
1101 * Return 1 if we went up a level and 0 if we were already at the
1102 * root.
1103 */
1104int follow_up(struct path *path)
1105{
1106 struct mount *mnt = real_mount(path->mnt);
1107 struct mount *parent;
1108 struct dentry *mountpoint;
1109
1110 read_seqlock_excl(&mount_lock);
1111 parent = mnt->mnt_parent;
1112 if (parent == mnt) {
1113 read_sequnlock_excl(&mount_lock);
1114 return 0;
1115 }
1116 mntget(&parent->mnt);
1117 mountpoint = dget(mnt->mnt_mountpoint);
1118 read_sequnlock_excl(&mount_lock);
1119 dput(path->dentry);
1120 path->dentry = mountpoint;
1121 mntput(path->mnt);
1122 path->mnt = &parent->mnt;
1123 return 1;
1124}
1125EXPORT_SYMBOL(follow_up);
1126
1127/*
1128 * Perform an automount
1129 * - return -EISDIR to tell follow_managed() to stop and return the path we
1130 * were called with.
1131 */
1132static int follow_automount(struct path *path, struct nameidata *nd,
1133 bool *need_mntput)
1134{
1135 struct vfsmount *mnt;
1136 int err;
1137
1138 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1139 return -EREMOTE;
1140
1141 /* We don't want to mount if someone's just doing a stat -
1142 * unless they're stat'ing a directory and appended a '/' to
1143 * the name.
1144 *
1145 * We do, however, want to mount if someone wants to open or
1146 * create a file of any type under the mountpoint, wants to
1147 * traverse through the mountpoint or wants to open the
1148 * mounted directory. Also, autofs may mark negative dentries
1149 * as being automount points. These will need the attentions
1150 * of the daemon to instantiate them before they can be used.
1151 */
1152 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1153 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1154 path->dentry->d_inode)
1155 return -EISDIR;
1156
1157 nd->total_link_count++;
1158 if (nd->total_link_count >= 40)
1159 return -ELOOP;
1160
1161 mnt = path->dentry->d_op->d_automount(path);
1162 if (IS_ERR(mnt)) {
1163 /*
1164 * The filesystem is allowed to return -EISDIR here to indicate
1165 * it doesn't want to automount. For instance, autofs would do
1166 * this so that its userspace daemon can mount on this dentry.
1167 *
1168 * However, we can only permit this if it's a terminal point in
1169 * the path being looked up; if it wasn't then the remainder of
1170 * the path is inaccessible and we should say so.
1171 */
1172 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1173 return -EREMOTE;
1174 return PTR_ERR(mnt);
1175 }
1176
1177 if (!mnt) /* mount collision */
1178 return 0;
1179
1180 if (!*need_mntput) {
1181 /* lock_mount() may release path->mnt on error */
1182 mntget(path->mnt);
1183 *need_mntput = true;
1184 }
1185 err = finish_automount(mnt, path);
1186
1187 switch (err) {
1188 case -EBUSY:
1189 /* Someone else made a mount here whilst we were busy */
1190 return 0;
1191 case 0:
1192 path_put(path);
1193 path->mnt = mnt;
1194 path->dentry = dget(mnt->mnt_root);
1195 return 0;
1196 default:
1197 return err;
1198 }
1199
1200}
1201
1202/*
1203 * Handle a dentry that is managed in some way.
1204 * - Flagged for transit management (autofs)
1205 * - Flagged as mountpoint
1206 * - Flagged as automount point
1207 *
1208 * This may only be called in refwalk mode.
1209 *
1210 * Serialization is taken care of in namespace.c
1211 */
1212static int follow_managed(struct path *path, struct nameidata *nd)
1213{
1214 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1215 unsigned managed;
1216 bool need_mntput = false;
1217 int ret = 0;
1218
1219 /* Given that we're not holding a lock here, we retain the value in a
1220 * local variable for each dentry as we look at it so that we don't see
1221 * the components of that value change under us */
1222 while (managed = READ_ONCE(path->dentry->d_flags),
1223 managed &= DCACHE_MANAGED_DENTRY,
1224 unlikely(managed != 0)) {
1225 /* Allow the filesystem to manage the transit without i_mutex
1226 * being held. */
1227 if (managed & DCACHE_MANAGE_TRANSIT) {
1228 BUG_ON(!path->dentry->d_op);
1229 BUG_ON(!path->dentry->d_op->d_manage);
1230 ret = path->dentry->d_op->d_manage(path, false);
1231 if (ret < 0)
1232 break;
1233 }
1234
1235 /* Transit to a mounted filesystem. */
1236 if (managed & DCACHE_MOUNTED) {
1237 struct vfsmount *mounted = lookup_mnt(path);
1238 if (mounted) {
1239 dput(path->dentry);
1240 if (need_mntput)
1241 mntput(path->mnt);
1242 path->mnt = mounted;
1243 path->dentry = dget(mounted->mnt_root);
1244 need_mntput = true;
1245 continue;
1246 }
1247
1248 /* Something is mounted on this dentry in another
1249 * namespace and/or whatever was mounted there in this
1250 * namespace got unmounted before lookup_mnt() could
1251 * get it */
1252 }
1253
1254 /* Handle an automount point */
1255 if (managed & DCACHE_NEED_AUTOMOUNT) {
1256 ret = follow_automount(path, nd, &need_mntput);
1257 if (ret < 0)
1258 break;
1259 continue;
1260 }
1261
1262 /* We didn't change the current path point */
1263 break;
1264 }
1265
1266 if (need_mntput && path->mnt == mnt)
1267 mntput(path->mnt);
1268 if (ret == -EISDIR || !ret)
1269 ret = 1;
1270 if (need_mntput)
1271 nd->flags |= LOOKUP_JUMPED;
1272 if (unlikely(ret < 0))
1273 path_put_conditional(path, nd);
1274 return ret;
1275}
1276
1277int follow_down_one(struct path *path)
1278{
1279 struct vfsmount *mounted;
1280
1281 mounted = lookup_mnt(path);
1282 if (mounted) {
1283 dput(path->dentry);
1284 mntput(path->mnt);
1285 path->mnt = mounted;
1286 path->dentry = dget(mounted->mnt_root);
1287 return 1;
1288 }
1289 return 0;
1290}
1291EXPORT_SYMBOL(follow_down_one);
1292
1293static inline int managed_dentry_rcu(const struct path *path)
1294{
1295 return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1296 path->dentry->d_op->d_manage(path, true) : 0;
1297}
1298
1299/*
1300 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1301 * we meet a managed dentry that would need blocking.
1302 */
1303static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1304 struct inode **inode, unsigned *seqp)
1305{
1306 for (;;) {
1307 struct mount *mounted;
1308 /*
1309 * Don't forget we might have a non-mountpoint managed dentry
1310 * that wants to block transit.
1311 */
1312 switch (managed_dentry_rcu(path)) {
1313 case -ECHILD:
1314 default:
1315 return false;
1316 case -EISDIR:
1317 return true;
1318 case 0:
1319 break;
1320 }
1321
1322 if (!d_mountpoint(path->dentry))
1323 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1324
1325 mounted = __lookup_mnt(path->mnt, path->dentry);
1326 if (!mounted)
1327 break;
1328 path->mnt = &mounted->mnt;
1329 path->dentry = mounted->mnt.mnt_root;
1330 nd->flags |= LOOKUP_JUMPED;
1331 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1332 /*
1333 * Update the inode too. We don't need to re-check the
1334 * dentry sequence number here after this d_inode read,
1335 * because a mount-point is always pinned.
1336 */
1337 *inode = path->dentry->d_inode;
1338 }
1339 return !read_seqretry(&mount_lock, nd->m_seq) &&
1340 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1341}
1342
1343static int follow_dotdot_rcu(struct nameidata *nd)
1344{
1345 struct inode *inode = nd->inode;
1346
1347 while (1) {
1348 if (path_equal(&nd->path, &nd->root))
1349 break;
1350 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1351 struct dentry *old = nd->path.dentry;
1352 struct dentry *parent = old->d_parent;
1353 unsigned seq;
1354
1355 inode = parent->d_inode;
1356 seq = read_seqcount_begin(&parent->d_seq);
1357 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1358 return -ECHILD;
1359 nd->path.dentry = parent;
1360 nd->seq = seq;
1361 if (unlikely(!path_connected(&nd->path)))
1362 return -ENOENT;
1363 break;
1364 } else {
1365 struct mount *mnt = real_mount(nd->path.mnt);
1366 struct mount *mparent = mnt->mnt_parent;
1367 struct dentry *mountpoint = mnt->mnt_mountpoint;
1368 struct inode *inode2 = mountpoint->d_inode;
1369 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1370 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1371 return -ECHILD;
1372 if (&mparent->mnt == nd->path.mnt)
1373 break;
1374 /* we know that mountpoint was pinned */
1375 nd->path.dentry = mountpoint;
1376 nd->path.mnt = &mparent->mnt;
1377 inode = inode2;
1378 nd->seq = seq;
1379 }
1380 }
1381 while (unlikely(d_mountpoint(nd->path.dentry))) {
1382 struct mount *mounted;
1383 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1384 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1385 return -ECHILD;
1386 if (!mounted)
1387 break;
1388 nd->path.mnt = &mounted->mnt;
1389 nd->path.dentry = mounted->mnt.mnt_root;
1390 inode = nd->path.dentry->d_inode;
1391 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1392 }
1393 nd->inode = inode;
1394 return 0;
1395}
1396
1397/*
1398 * Follow down to the covering mount currently visible to userspace. At each
1399 * point, the filesystem owning that dentry may be queried as to whether the
1400 * caller is permitted to proceed or not.
1401 */
1402int follow_down(struct path *path)
1403{
1404 unsigned managed;
1405 int ret;
1406
1407 while (managed = READ_ONCE(path->dentry->d_flags),
1408 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1409 /* Allow the filesystem to manage the transit without i_mutex
1410 * being held.
1411 *
1412 * We indicate to the filesystem if someone is trying to mount
1413 * something here. This gives autofs the chance to deny anyone
1414 * other than its daemon the right to mount on its
1415 * superstructure.
1416 *
1417 * The filesystem may sleep at this point.
1418 */
1419 if (managed & DCACHE_MANAGE_TRANSIT) {
1420 BUG_ON(!path->dentry->d_op);
1421 BUG_ON(!path->dentry->d_op->d_manage);
1422 ret = path->dentry->d_op->d_manage(path, false);
1423 if (ret < 0)
1424 return ret == -EISDIR ? 0 : ret;
1425 }
1426
1427 /* Transit to a mounted filesystem. */
1428 if (managed & DCACHE_MOUNTED) {
1429 struct vfsmount *mounted = lookup_mnt(path);
1430 if (!mounted)
1431 break;
1432 dput(path->dentry);
1433 mntput(path->mnt);
1434 path->mnt = mounted;
1435 path->dentry = dget(mounted->mnt_root);
1436 continue;
1437 }
1438
1439 /* Don't handle automount points here */
1440 break;
1441 }
1442 return 0;
1443}
1444EXPORT_SYMBOL(follow_down);
1445
1446/*
1447 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1448 */
1449static void follow_mount(struct path *path)
1450{
1451 while (d_mountpoint(path->dentry)) {
1452 struct vfsmount *mounted = lookup_mnt(path);
1453 if (!mounted)
1454 break;
1455 dput(path->dentry);
1456 mntput(path->mnt);
1457 path->mnt = mounted;
1458 path->dentry = dget(mounted->mnt_root);
1459 }
1460}
1461
1462static int path_parent_directory(struct path *path)
1463{
1464 struct dentry *old = path->dentry;
1465 /* rare case of legitimate dget_parent()... */
1466 path->dentry = dget_parent(path->dentry);
1467 dput(old);
1468 if (unlikely(!path_connected(path)))
1469 return -ENOENT;
1470 return 0;
1471}
1472
1473static int follow_dotdot(struct nameidata *nd)
1474{
1475 while(1) {
1476 if (path_equal(&nd->path, &nd->root))
1477 break;
1478 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1479 int ret = path_parent_directory(&nd->path);
1480 if (ret)
1481 return ret;
1482 break;
1483 }
1484 if (!follow_up(&nd->path))
1485 break;
1486 }
1487 follow_mount(&nd->path);
1488 nd->inode = nd->path.dentry->d_inode;
1489 return 0;
1490}
1491
1492/*
1493 * This looks up the name in dcache and possibly revalidates the found dentry.
1494 * NULL is returned if the dentry does not exist in the cache.
1495 */
1496static struct dentry *lookup_dcache(const struct qstr *name,
1497 struct dentry *dir,
1498 unsigned int flags)
1499{
1500 struct dentry *dentry = d_lookup(dir, name);
1501 if (dentry) {
1502 int error = d_revalidate(dentry, flags);
1503 if (unlikely(error <= 0)) {
1504 if (!error)
1505 d_invalidate(dentry);
1506 dput(dentry);
1507 return ERR_PTR(error);
1508 }
1509 }
1510 return dentry;
1511}
1512
1513/*
1514 * Parent directory has inode locked exclusive. This is one
1515 * and only case when ->lookup() gets called on non in-lookup
1516 * dentries - as the matter of fact, this only gets called
1517 * when directory is guaranteed to have no in-lookup children
1518 * at all.
1519 */
1520static struct dentry *__lookup_hash(const struct qstr *name,
1521 struct dentry *base, unsigned int flags)
1522{
1523 struct dentry *dentry = lookup_dcache(name, base, flags);
1524 struct dentry *old;
1525 struct inode *dir = base->d_inode;
1526
1527 if (dentry)
1528 return dentry;
1529
1530 /* Don't create child dentry for a dead directory. */
1531 if (unlikely(IS_DEADDIR(dir)))
1532 return ERR_PTR(-ENOENT);
1533
1534 dentry = d_alloc(base, name);
1535 if (unlikely(!dentry))
1536 return ERR_PTR(-ENOMEM);
1537
1538 old = dir->i_op->lookup(dir, dentry, flags);
1539 if (unlikely(old)) {
1540 dput(dentry);
1541 dentry = old;
1542 }
1543 return dentry;
1544}
1545
1546static int lookup_fast(struct nameidata *nd,
1547 struct path *path, struct inode **inode,
1548 unsigned *seqp)
1549{
1550 struct vfsmount *mnt = nd->path.mnt;
1551 struct dentry *dentry, *parent = nd->path.dentry;
1552 int status = 1;
1553 int err;
1554
1555 /*
1556 * Rename seqlock is not required here because in the off chance
1557 * of a false negative due to a concurrent rename, the caller is
1558 * going to fall back to non-racy lookup.
1559 */
1560 if (nd->flags & LOOKUP_RCU) {
1561 unsigned seq;
1562 bool negative;
1563 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1564 if (unlikely(!dentry)) {
1565 if (unlazy_walk(nd))
1566 return -ECHILD;
1567 return 0;
1568 }
1569
1570 /*
1571 * This sequence count validates that the inode matches
1572 * the dentry name information from lookup.
1573 */
1574 *inode = d_backing_inode(dentry);
1575 negative = d_is_negative(dentry);
1576 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1577 return -ECHILD;
1578
1579 /*
1580 * This sequence count validates that the parent had no
1581 * changes while we did the lookup of the dentry above.
1582 *
1583 * The memory barrier in read_seqcount_begin of child is
1584 * enough, we can use __read_seqcount_retry here.
1585 */
1586 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1587 return -ECHILD;
1588
1589 *seqp = seq;
1590 status = d_revalidate(dentry, nd->flags);
1591 if (likely(status > 0)) {
1592 /*
1593 * Note: do negative dentry check after revalidation in
1594 * case that drops it.
1595 */
1596 if (unlikely(negative))
1597 return -ENOENT;
1598 path->mnt = mnt;
1599 path->dentry = dentry;
1600 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1601 return 1;
1602 }
1603 if (unlazy_child(nd, dentry, seq))
1604 return -ECHILD;
1605 if (unlikely(status == -ECHILD))
1606 /* we'd been told to redo it in non-rcu mode */
1607 status = d_revalidate(dentry, nd->flags);
1608 } else {
1609 dentry = __d_lookup(parent, &nd->last);
1610 if (unlikely(!dentry))
1611 return 0;
1612 status = d_revalidate(dentry, nd->flags);
1613 }
1614 if (unlikely(status <= 0)) {
1615 if (!status)
1616 d_invalidate(dentry);
1617 dput(dentry);
1618 return status;
1619 }
1620 if (unlikely(d_is_negative(dentry))) {
1621 dput(dentry);
1622 return -ENOENT;
1623 }
1624
1625 path->mnt = mnt;
1626 path->dentry = dentry;
1627 err = follow_managed(path, nd);
1628 if (likely(err > 0))
1629 *inode = d_backing_inode(path->dentry);
1630 return err;
1631}
1632
1633/* Fast lookup failed, do it the slow way */
1634static struct dentry *__lookup_slow(const struct qstr *name,
1635 struct dentry *dir,
1636 unsigned int flags)
1637{
1638 struct dentry *dentry, *old;
1639 struct inode *inode = dir->d_inode;
1640 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1641
1642 /* Don't go there if it's already dead */
1643 if (unlikely(IS_DEADDIR(inode)))
1644 return ERR_PTR(-ENOENT);
1645again:
1646 dentry = d_alloc_parallel(dir, name, &wq);
1647 if (IS_ERR(dentry))
1648 return dentry;
1649 if (unlikely(!d_in_lookup(dentry))) {
1650 if (!(flags & LOOKUP_NO_REVAL)) {
1651 int error = d_revalidate(dentry, flags);
1652 if (unlikely(error <= 0)) {
1653 if (!error) {
1654 d_invalidate(dentry);
1655 dput(dentry);
1656 goto again;
1657 }
1658 dput(dentry);
1659 dentry = ERR_PTR(error);
1660 }
1661 }
1662 } else {
1663 old = inode->i_op->lookup(inode, dentry, flags);
1664 d_lookup_done(dentry);
1665 if (unlikely(old)) {
1666 dput(dentry);
1667 dentry = old;
1668 }
1669 }
1670 return dentry;
1671}
1672
1673static struct dentry *lookup_slow(const struct qstr *name,
1674 struct dentry *dir,
1675 unsigned int flags)
1676{
1677 struct inode *inode = dir->d_inode;
1678 struct dentry *res;
1679 inode_lock_shared(inode);
1680 res = __lookup_slow(name, dir, flags);
1681 inode_unlock_shared(inode);
1682 return res;
1683}
1684
1685static inline int may_lookup(struct nameidata *nd)
1686{
1687 if (nd->flags & LOOKUP_RCU) {
1688 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1689 if (err != -ECHILD)
1690 return err;
1691 if (unlazy_walk(nd))
1692 return -ECHILD;
1693 }
1694 return inode_permission(nd->inode, MAY_EXEC);
1695}
1696
1697static inline int handle_dots(struct nameidata *nd, int type)
1698{
1699 if (type == LAST_DOTDOT) {
1700 if (!nd->root.mnt)
1701 set_root(nd);
1702 if (nd->flags & LOOKUP_RCU) {
1703 return follow_dotdot_rcu(nd);
1704 } else
1705 return follow_dotdot(nd);
1706 }
1707 return 0;
1708}
1709
1710static int pick_link(struct nameidata *nd, struct path *link,
1711 struct inode *inode, unsigned seq)
1712{
1713 int error;
1714 struct saved *last;
1715 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1716 path_to_nameidata(link, nd);
1717 return -ELOOP;
1718 }
1719 if (!(nd->flags & LOOKUP_RCU)) {
1720 if (link->mnt == nd->path.mnt)
1721 mntget(link->mnt);
1722 }
1723 error = nd_alloc_stack(nd);
1724 if (unlikely(error)) {
1725 if (error == -ECHILD) {
1726 if (unlikely(!legitimize_path(nd, link, seq))) {
1727 drop_links(nd);
1728 nd->depth = 0;
1729 nd->flags &= ~LOOKUP_RCU;
1730 nd->path.mnt = NULL;
1731 nd->path.dentry = NULL;
1732 rcu_read_unlock();
1733 } else if (likely(unlazy_walk(nd)) == 0)
1734 error = nd_alloc_stack(nd);
1735 }
1736 if (error) {
1737 path_put(link);
1738 return error;
1739 }
1740 }
1741
1742 last = nd->stack + nd->depth++;
1743 last->link = *link;
1744 clear_delayed_call(&last->done);
1745 nd->link_inode = inode;
1746 last->seq = seq;
1747 return 1;
1748}
1749
1750enum {WALK_FOLLOW = 1, WALK_MORE = 2};
1751
1752/*
1753 * Do we need to follow links? We _really_ want to be able
1754 * to do this check without having to look at inode->i_op,
1755 * so we keep a cache of "no, this doesn't need follow_link"
1756 * for the common case.
1757 */
1758static inline int step_into(struct nameidata *nd, struct path *path,
1759 int flags, struct inode *inode, unsigned seq)
1760{
1761 if (!(flags & WALK_MORE) && nd->depth)
1762 put_link(nd);
1763 if (likely(!d_is_symlink(path->dentry)) ||
1764 !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) {
1765 /* not a symlink or should not follow */
1766 path_to_nameidata(path, nd);
1767 nd->inode = inode;
1768 nd->seq = seq;
1769 return 0;
1770 }
1771 /* make sure that d_is_symlink above matches inode */
1772 if (nd->flags & LOOKUP_RCU) {
1773 if (read_seqcount_retry(&path->dentry->d_seq, seq))
1774 return -ECHILD;
1775 }
1776 return pick_link(nd, path, inode, seq);
1777}
1778
1779static int walk_component(struct nameidata *nd, int flags)
1780{
1781 struct path path;
1782 struct inode *inode;
1783 unsigned seq;
1784 int err;
1785 /*
1786 * "." and ".." are special - ".." especially so because it has
1787 * to be able to know about the current root directory and
1788 * parent relationships.
1789 */
1790 if (unlikely(nd->last_type != LAST_NORM)) {
1791 err = handle_dots(nd, nd->last_type);
1792 if (!(flags & WALK_MORE) && nd->depth)
1793 put_link(nd);
1794 return err;
1795 }
1796 err = lookup_fast(nd, &path, &inode, &seq);
1797 if (unlikely(err <= 0)) {
1798 if (err < 0)
1799 return err;
1800 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1801 nd->flags);
1802 if (IS_ERR(path.dentry))
1803 return PTR_ERR(path.dentry);
1804
1805 path.mnt = nd->path.mnt;
1806 err = follow_managed(&path, nd);
1807 if (unlikely(err < 0))
1808 return err;
1809
1810 if (unlikely(d_is_negative(path.dentry))) {
1811 path_to_nameidata(&path, nd);
1812 return -ENOENT;
1813 }
1814
1815 seq = 0; /* we are already out of RCU mode */
1816 inode = d_backing_inode(path.dentry);
1817 }
1818
1819 return step_into(nd, &path, flags, inode, seq);
1820}
1821
1822/*
1823 * We can do the critical dentry name comparison and hashing
1824 * operations one word at a time, but we are limited to:
1825 *
1826 * - Architectures with fast unaligned word accesses. We could
1827 * do a "get_unaligned()" if this helps and is sufficiently
1828 * fast.
1829 *
1830 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1831 * do not trap on the (extremely unlikely) case of a page
1832 * crossing operation.
1833 *
1834 * - Furthermore, we need an efficient 64-bit compile for the
1835 * 64-bit case in order to generate the "number of bytes in
1836 * the final mask". Again, that could be replaced with a
1837 * efficient population count instruction or similar.
1838 */
1839#ifdef CONFIG_DCACHE_WORD_ACCESS
1840
1841#include <asm/word-at-a-time.h>
1842
1843#ifdef HASH_MIX
1844
1845/* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1846
1847#elif defined(CONFIG_64BIT)
1848/*
1849 * Register pressure in the mixing function is an issue, particularly
1850 * on 32-bit x86, but almost any function requires one state value and
1851 * one temporary. Instead, use a function designed for two state values
1852 * and no temporaries.
1853 *
1854 * This function cannot create a collision in only two iterations, so
1855 * we have two iterations to achieve avalanche. In those two iterations,
1856 * we have six layers of mixing, which is enough to spread one bit's
1857 * influence out to 2^6 = 64 state bits.
1858 *
1859 * Rotate constants are scored by considering either 64 one-bit input
1860 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1861 * probability of that delta causing a change to each of the 128 output
1862 * bits, using a sample of random initial states.
1863 *
1864 * The Shannon entropy of the computed probabilities is then summed
1865 * to produce a score. Ideally, any input change has a 50% chance of
1866 * toggling any given output bit.
1867 *
1868 * Mixing scores (in bits) for (12,45):
1869 * Input delta: 1-bit 2-bit
1870 * 1 round: 713.3 42542.6
1871 * 2 rounds: 2753.7 140389.8
1872 * 3 rounds: 5954.1 233458.2
1873 * 4 rounds: 7862.6 256672.2
1874 * Perfect: 8192 258048
1875 * (64*128) (64*63/2 * 128)
1876 */
1877#define HASH_MIX(x, y, a) \
1878 ( x ^= (a), \
1879 y ^= x, x = rol64(x,12),\
1880 x += y, y = rol64(y,45),\
1881 y *= 9 )
1882
1883/*
1884 * Fold two longs into one 32-bit hash value. This must be fast, but
1885 * latency isn't quite as critical, as there is a fair bit of additional
1886 * work done before the hash value is used.
1887 */
1888static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1889{
1890 y ^= x * GOLDEN_RATIO_64;
1891 y *= GOLDEN_RATIO_64;
1892 return y >> 32;
1893}
1894
1895#else /* 32-bit case */
1896
1897/*
1898 * Mixing scores (in bits) for (7,20):
1899 * Input delta: 1-bit 2-bit
1900 * 1 round: 330.3 9201.6
1901 * 2 rounds: 1246.4 25475.4
1902 * 3 rounds: 1907.1 31295.1
1903 * 4 rounds: 2042.3 31718.6
1904 * Perfect: 2048 31744
1905 * (32*64) (32*31/2 * 64)
1906 */
1907#define HASH_MIX(x, y, a) \
1908 ( x ^= (a), \
1909 y ^= x, x = rol32(x, 7),\
1910 x += y, y = rol32(y,20),\
1911 y *= 9 )
1912
1913static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1914{
1915 /* Use arch-optimized multiply if one exists */
1916 return __hash_32(y ^ __hash_32(x));
1917}
1918
1919#endif
1920
1921/*
1922 * Return the hash of a string of known length. This is carfully
1923 * designed to match hash_name(), which is the more critical function.
1924 * In particular, we must end by hashing a final word containing 0..7
1925 * payload bytes, to match the way that hash_name() iterates until it
1926 * finds the delimiter after the name.
1927 */
1928unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1929{
1930 unsigned long a, x = 0, y = (unsigned long)salt;
1931
1932 for (;;) {
1933 if (!len)
1934 goto done;
1935 a = load_unaligned_zeropad(name);
1936 if (len < sizeof(unsigned long))
1937 break;
1938 HASH_MIX(x, y, a);
1939 name += sizeof(unsigned long);
1940 len -= sizeof(unsigned long);
1941 }
1942 x ^= a & bytemask_from_count(len);
1943done:
1944 return fold_hash(x, y);
1945}
1946EXPORT_SYMBOL(full_name_hash);
1947
1948/* Return the "hash_len" (hash and length) of a null-terminated string */
1949u64 hashlen_string(const void *salt, const char *name)
1950{
1951 unsigned long a = 0, x = 0, y = (unsigned long)salt;
1952 unsigned long adata, mask, len;
1953 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1954
1955 len = 0;
1956 goto inside;
1957
1958 do {
1959 HASH_MIX(x, y, a);
1960 len += sizeof(unsigned long);
1961inside:
1962 a = load_unaligned_zeropad(name+len);
1963 } while (!has_zero(a, &adata, &constants));
1964
1965 adata = prep_zero_mask(a, adata, &constants);
1966 mask = create_zero_mask(adata);
1967 x ^= a & zero_bytemask(mask);
1968
1969 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1970}
1971EXPORT_SYMBOL(hashlen_string);
1972
1973/*
1974 * Calculate the length and hash of the path component, and
1975 * return the "hash_len" as the result.
1976 */
1977static inline u64 hash_name(const void *salt, const char *name)
1978{
1979 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
1980 unsigned long adata, bdata, mask, len;
1981 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1982
1983 len = 0;
1984 goto inside;
1985
1986 do {
1987 HASH_MIX(x, y, a);
1988 len += sizeof(unsigned long);
1989inside:
1990 a = load_unaligned_zeropad(name+len);
1991 b = a ^ REPEAT_BYTE('/');
1992 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1993
1994 adata = prep_zero_mask(a, adata, &constants);
1995 bdata = prep_zero_mask(b, bdata, &constants);
1996 mask = create_zero_mask(adata | bdata);
1997 x ^= a & zero_bytemask(mask);
1998
1999 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2000}
2001
2002#else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2003
2004/* Return the hash of a string of known length */
2005unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2006{
2007 unsigned long hash = init_name_hash(salt);
2008 while (len--)
2009 hash = partial_name_hash((unsigned char)*name++, hash);
2010 return end_name_hash(hash);
2011}
2012EXPORT_SYMBOL(full_name_hash);
2013
2014/* Return the "hash_len" (hash and length) of a null-terminated string */
2015u64 hashlen_string(const void *salt, const char *name)
2016{
2017 unsigned long hash = init_name_hash(salt);
2018 unsigned long len = 0, c;
2019
2020 c = (unsigned char)*name;
2021 while (c) {
2022 len++;
2023 hash = partial_name_hash(c, hash);
2024 c = (unsigned char)name[len];
2025 }
2026 return hashlen_create(end_name_hash(hash), len);
2027}
2028EXPORT_SYMBOL(hashlen_string);
2029
2030/*
2031 * We know there's a real path component here of at least
2032 * one character.
2033 */
2034static inline u64 hash_name(const void *salt, const char *name)
2035{
2036 unsigned long hash = init_name_hash(salt);
2037 unsigned long len = 0, c;
2038
2039 c = (unsigned char)*name;
2040 do {
2041 len++;
2042 hash = partial_name_hash(c, hash);
2043 c = (unsigned char)name[len];
2044 } while (c && c != '/');
2045 return hashlen_create(end_name_hash(hash), len);
2046}
2047
2048#endif
2049
2050/*
2051 * Name resolution.
2052 * This is the basic name resolution function, turning a pathname into
2053 * the final dentry. We expect 'base' to be positive and a directory.
2054 *
2055 * Returns 0 and nd will have valid dentry and mnt on success.
2056 * Returns error and drops reference to input namei data on failure.
2057 */
2058static int link_path_walk(const char *name, struct nameidata *nd)
2059{
2060 int err;
2061
2062 if (IS_ERR(name))
2063 return PTR_ERR(name);
2064 while (*name=='/')
2065 name++;
2066 if (!*name)
2067 return 0;
2068
2069 /* At this point we know we have a real path component. */
2070 for(;;) {
2071 u64 hash_len;
2072 int type;
2073
2074 err = may_lookup(nd);
2075 if (err)
2076 return err;
2077
2078 hash_len = hash_name(nd->path.dentry, name);
2079
2080 type = LAST_NORM;
2081 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2082 case 2:
2083 if (name[1] == '.') {
2084 type = LAST_DOTDOT;
2085 nd->flags |= LOOKUP_JUMPED;
2086 }
2087 break;
2088 case 1:
2089 type = LAST_DOT;
2090 }
2091 if (likely(type == LAST_NORM)) {
2092 struct dentry *parent = nd->path.dentry;
2093 nd->flags &= ~LOOKUP_JUMPED;
2094 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2095 struct qstr this = { { .hash_len = hash_len }, .name = name };
2096 err = parent->d_op->d_hash(parent, &this);
2097 if (err < 0)
2098 return err;
2099 hash_len = this.hash_len;
2100 name = this.name;
2101 }
2102 }
2103
2104 nd->last.hash_len = hash_len;
2105 nd->last.name = name;
2106 nd->last_type = type;
2107
2108 name += hashlen_len(hash_len);
2109 if (!*name)
2110 goto OK;
2111 /*
2112 * If it wasn't NUL, we know it was '/'. Skip that
2113 * slash, and continue until no more slashes.
2114 */
2115 do {
2116 name++;
2117 } while (unlikely(*name == '/'));
2118 if (unlikely(!*name)) {
2119OK:
2120 /* pathname body, done */
2121 if (!nd->depth)
2122 return 0;
2123 name = nd->stack[nd->depth - 1].name;
2124 /* trailing symlink, done */
2125 if (!name)
2126 return 0;
2127 /* last component of nested symlink */
2128 err = walk_component(nd, WALK_FOLLOW);
2129 } else {
2130 /* not the last component */
2131 err = walk_component(nd, WALK_FOLLOW | WALK_MORE);
2132 }
2133 if (err < 0)
2134 return err;
2135
2136 if (err) {
2137 const char *s = get_link(nd);
2138
2139 if (IS_ERR(s))
2140 return PTR_ERR(s);
2141 err = 0;
2142 if (unlikely(!s)) {
2143 /* jumped */
2144 put_link(nd);
2145 } else {
2146 nd->stack[nd->depth - 1].name = name;
2147 name = s;
2148 continue;
2149 }
2150 }
2151 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2152 if (nd->flags & LOOKUP_RCU) {
2153 if (unlazy_walk(nd))
2154 return -ECHILD;
2155 }
2156 return -ENOTDIR;
2157 }
2158 }
2159}
2160
2161/* must be paired with terminate_walk() */
2162static const char *path_init(struct nameidata *nd, unsigned flags)
2163{
2164 const char *s = nd->name->name;
2165
2166 if (!*s)
2167 flags &= ~LOOKUP_RCU;
2168 if (flags & LOOKUP_RCU)
2169 rcu_read_lock();
2170
2171 nd->last_type = LAST_ROOT; /* if there are only slashes... */
2172 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2173 nd->depth = 0;
2174 if (flags & LOOKUP_ROOT) {
2175 struct dentry *root = nd->root.dentry;
2176 struct inode *inode = root->d_inode;
2177 if (*s && unlikely(!d_can_lookup(root)))
2178 return ERR_PTR(-ENOTDIR);
2179 nd->path = nd->root;
2180 nd->inode = inode;
2181 if (flags & LOOKUP_RCU) {
2182 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2183 nd->root_seq = nd->seq;
2184 nd->m_seq = read_seqbegin(&mount_lock);
2185 } else {
2186 path_get(&nd->path);
2187 }
2188 return s;
2189 }
2190
2191 nd->root.mnt = NULL;
2192 nd->path.mnt = NULL;
2193 nd->path.dentry = NULL;
2194
2195 nd->m_seq = read_seqbegin(&mount_lock);
2196 if (*s == '/') {
2197 set_root(nd);
2198 if (likely(!nd_jump_root(nd)))
2199 return s;
2200 return ERR_PTR(-ECHILD);
2201 } else if (nd->dfd == AT_FDCWD) {
2202 if (flags & LOOKUP_RCU) {
2203 struct fs_struct *fs = current->fs;
2204 unsigned seq;
2205
2206 do {
2207 seq = read_seqcount_begin(&fs->seq);
2208 nd->path = fs->pwd;
2209 nd->inode = nd->path.dentry->d_inode;
2210 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2211 } while (read_seqcount_retry(&fs->seq, seq));
2212 } else {
2213 get_fs_pwd(current->fs, &nd->path);
2214 nd->inode = nd->path.dentry->d_inode;
2215 }
2216 return s;
2217 } else {
2218 /* Caller must check execute permissions on the starting path component */
2219 struct fd f = fdget_raw(nd->dfd);
2220 struct dentry *dentry;
2221
2222 if (!f.file)
2223 return ERR_PTR(-EBADF);
2224
2225 dentry = f.file->f_path.dentry;
2226
2227 if (*s && unlikely(!d_can_lookup(dentry))) {
2228 fdput(f);
2229 return ERR_PTR(-ENOTDIR);
2230 }
2231
2232 nd->path = f.file->f_path;
2233 if (flags & LOOKUP_RCU) {
2234 nd->inode = nd->path.dentry->d_inode;
2235 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2236 } else {
2237 path_get(&nd->path);
2238 nd->inode = nd->path.dentry->d_inode;
2239 }
2240 fdput(f);
2241 return s;
2242 }
2243}
2244
2245static const char *trailing_symlink(struct nameidata *nd)
2246{
2247 const char *s;
2248 int error = may_follow_link(nd);
2249 if (unlikely(error))
2250 return ERR_PTR(error);
2251 nd->flags |= LOOKUP_PARENT;
2252 nd->stack[0].name = NULL;
2253 s = get_link(nd);
2254 return s ? s : "";
2255}
2256
2257static inline int lookup_last(struct nameidata *nd)
2258{
2259 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2260 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2261
2262 nd->flags &= ~LOOKUP_PARENT;
2263 return walk_component(nd, 0);
2264}
2265
2266static int handle_lookup_down(struct nameidata *nd)
2267{
2268 struct path path = nd->path;
2269 struct inode *inode = nd->inode;
2270 unsigned seq = nd->seq;
2271 int err;
2272
2273 if (nd->flags & LOOKUP_RCU) {
2274 /*
2275 * don't bother with unlazy_walk on failure - we are
2276 * at the very beginning of walk, so we lose nothing
2277 * if we simply redo everything in non-RCU mode
2278 */
2279 if (unlikely(!__follow_mount_rcu(nd, &path, &inode, &seq)))
2280 return -ECHILD;
2281 } else {
2282 dget(path.dentry);
2283 err = follow_managed(&path, nd);
2284 if (unlikely(err < 0))
2285 return err;
2286 inode = d_backing_inode(path.dentry);
2287 seq = 0;
2288 }
2289 path_to_nameidata(&path, nd);
2290 nd->inode = inode;
2291 nd->seq = seq;
2292 return 0;
2293}
2294
2295/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2296static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2297{
2298 const char *s = path_init(nd, flags);
2299 int err;
2300
2301 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2302 err = handle_lookup_down(nd);
2303 if (unlikely(err < 0))
2304 s = ERR_PTR(err);
2305 }
2306
2307 while (!(err = link_path_walk(s, nd))
2308 && ((err = lookup_last(nd)) > 0)) {
2309 s = trailing_symlink(nd);
2310 }
2311 if (!err)
2312 err = complete_walk(nd);
2313
2314 if (!err && nd->flags & LOOKUP_DIRECTORY)
2315 if (!d_can_lookup(nd->path.dentry))
2316 err = -ENOTDIR;
2317 if (!err) {
2318 *path = nd->path;
2319 nd->path.mnt = NULL;
2320 nd->path.dentry = NULL;
2321 }
2322 terminate_walk(nd);
2323 return err;
2324}
2325
2326int filename_lookup(int dfd, struct filename *name, unsigned flags,
2327 struct path *path, struct path *root)
2328{
2329 int retval;
2330 struct nameidata nd;
2331 if (IS_ERR(name))
2332 return PTR_ERR(name);
2333 if (unlikely(root)) {
2334 nd.root = *root;
2335 flags |= LOOKUP_ROOT;
2336 }
2337 set_nameidata(&nd, dfd, name);
2338 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2339 if (unlikely(retval == -ECHILD))
2340 retval = path_lookupat(&nd, flags, path);
2341 if (unlikely(retval == -ESTALE))
2342 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2343
2344 if (likely(!retval))
2345 audit_inode(name, path->dentry, 0);
2346 restore_nameidata();
2347 putname(name);
2348 return retval;
2349}
2350
2351/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2352static int path_parentat(struct nameidata *nd, unsigned flags,
2353 struct path *parent)
2354{
2355 const char *s = path_init(nd, flags);
2356 int err = link_path_walk(s, nd);
2357 if (!err)
2358 err = complete_walk(nd);
2359 if (!err) {
2360 *parent = nd->path;
2361 nd->path.mnt = NULL;
2362 nd->path.dentry = NULL;
2363 }
2364 terminate_walk(nd);
2365 return err;
2366}
2367
2368static struct filename *filename_parentat(int dfd, struct filename *name,
2369 unsigned int flags, struct path *parent,
2370 struct qstr *last, int *type)
2371{
2372 int retval;
2373 struct nameidata nd;
2374
2375 if (IS_ERR(name))
2376 return name;
2377 set_nameidata(&nd, dfd, name);
2378 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2379 if (unlikely(retval == -ECHILD))
2380 retval = path_parentat(&nd, flags, parent);
2381 if (unlikely(retval == -ESTALE))
2382 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2383 if (likely(!retval)) {
2384 *last = nd.last;
2385 *type = nd.last_type;
2386 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2387 } else {
2388 putname(name);
2389 name = ERR_PTR(retval);
2390 }
2391 restore_nameidata();
2392 return name;
2393}
2394
2395/* does lookup, returns the object with parent locked */
2396struct dentry *kern_path_locked(const char *name, struct path *path)
2397{
2398 struct filename *filename;
2399 struct dentry *d;
2400 struct qstr last;
2401 int type;
2402
2403 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2404 &last, &type);
2405 if (IS_ERR(filename))
2406 return ERR_CAST(filename);
2407 if (unlikely(type != LAST_NORM)) {
2408 path_put(path);
2409 putname(filename);
2410 return ERR_PTR(-EINVAL);
2411 }
2412 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2413 d = __lookup_hash(&last, path->dentry, 0);
2414 if (IS_ERR(d)) {
2415 inode_unlock(path->dentry->d_inode);
2416 path_put(path);
2417 }
2418 putname(filename);
2419 return d;
2420}
2421
2422int kern_path(const char *name, unsigned int flags, struct path *path)
2423{
2424 return filename_lookup(AT_FDCWD, getname_kernel(name),
2425 flags, path, NULL);
2426}
2427EXPORT_SYMBOL(kern_path);
2428
2429/**
2430 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2431 * @dentry: pointer to dentry of the base directory
2432 * @mnt: pointer to vfs mount of the base directory
2433 * @name: pointer to file name
2434 * @flags: lookup flags
2435 * @path: pointer to struct path to fill
2436 */
2437int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2438 const char *name, unsigned int flags,
2439 struct path *path)
2440{
2441 struct path root = {.mnt = mnt, .dentry = dentry};
2442 /* the first argument of filename_lookup() is ignored with root */
2443 return filename_lookup(AT_FDCWD, getname_kernel(name),
2444 flags , path, &root);
2445}
2446EXPORT_SYMBOL(vfs_path_lookup);
2447
2448static int lookup_one_len_common(const char *name, struct dentry *base,
2449 int len, struct qstr *this)
2450{
2451 this->name = name;
2452 this->len = len;
2453 this->hash = full_name_hash(base, name, len);
2454 if (!len)
2455 return -EACCES;
2456
2457 if (unlikely(name[0] == '.')) {
2458 if (len < 2 || (len == 2 && name[1] == '.'))
2459 return -EACCES;
2460 }
2461
2462 while (len--) {
2463 unsigned int c = *(const unsigned char *)name++;
2464 if (c == '/' || c == '\0')
2465 return -EACCES;
2466 }
2467 /*
2468 * See if the low-level filesystem might want
2469 * to use its own hash..
2470 */
2471 if (base->d_flags & DCACHE_OP_HASH) {
2472 int err = base->d_op->d_hash(base, this);
2473 if (err < 0)
2474 return err;
2475 }
2476
2477 return inode_permission(base->d_inode, MAY_EXEC);
2478}
2479
2480/**
2481 * try_lookup_one_len - filesystem helper to lookup single pathname component
2482 * @name: pathname component to lookup
2483 * @base: base directory to lookup from
2484 * @len: maximum length @len should be interpreted to
2485 *
2486 * Look up a dentry by name in the dcache, returning NULL if it does not
2487 * currently exist. The function does not try to create a dentry.
2488 *
2489 * Note that this routine is purely a helper for filesystem usage and should
2490 * not be called by generic code.
2491 *
2492 * The caller must hold base->i_mutex.
2493 */
2494struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2495{
2496 struct qstr this;
2497 int err;
2498
2499 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2500
2501 err = lookup_one_len_common(name, base, len, &this);
2502 if (err)
2503 return ERR_PTR(err);
2504
2505 return lookup_dcache(&this, base, 0);
2506}
2507EXPORT_SYMBOL(try_lookup_one_len);
2508
2509/**
2510 * lookup_one_len - filesystem helper to lookup single pathname component
2511 * @name: pathname component to lookup
2512 * @base: base directory to lookup from
2513 * @len: maximum length @len should be interpreted to
2514 *
2515 * Note that this routine is purely a helper for filesystem usage and should
2516 * not be called by generic code.
2517 *
2518 * The caller must hold base->i_mutex.
2519 */
2520struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2521{
2522 struct dentry *dentry;
2523 struct qstr this;
2524 int err;
2525
2526 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2527
2528 err = lookup_one_len_common(name, base, len, &this);
2529 if (err)
2530 return ERR_PTR(err);
2531
2532 dentry = lookup_dcache(&this, base, 0);
2533 return dentry ? dentry : __lookup_slow(&this, base, 0);
2534}
2535EXPORT_SYMBOL(lookup_one_len);
2536
2537/**
2538 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2539 * @name: pathname component to lookup
2540 * @base: base directory to lookup from
2541 * @len: maximum length @len should be interpreted to
2542 *
2543 * Note that this routine is purely a helper for filesystem usage and should
2544 * not be called by generic code.
2545 *
2546 * Unlike lookup_one_len, it should be called without the parent
2547 * i_mutex held, and will take the i_mutex itself if necessary.
2548 */
2549struct dentry *lookup_one_len_unlocked(const char *name,
2550 struct dentry *base, int len)
2551{
2552 struct qstr this;
2553 int err;
2554 struct dentry *ret;
2555
2556 err = lookup_one_len_common(name, base, len, &this);
2557 if (err)
2558 return ERR_PTR(err);
2559
2560 ret = lookup_dcache(&this, base, 0);
2561 if (!ret)
2562 ret = lookup_slow(&this, base, 0);
2563 return ret;
2564}
2565EXPORT_SYMBOL(lookup_one_len_unlocked);
2566
2567#ifdef CONFIG_UNIX98_PTYS
2568int path_pts(struct path *path)
2569{
2570 /* Find something mounted on "pts" in the same directory as
2571 * the input path.
2572 */
2573 struct dentry *child, *parent;
2574 struct qstr this;
2575 int ret;
2576
2577 ret = path_parent_directory(path);
2578 if (ret)
2579 return ret;
2580
2581 parent = path->dentry;
2582 this.name = "pts";
2583 this.len = 3;
2584 child = d_hash_and_lookup(parent, &this);
2585 if (!child)
2586 return -ENOENT;
2587
2588 path->dentry = child;
2589 dput(parent);
2590 follow_mount(path);
2591 return 0;
2592}
2593#endif
2594
2595int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2596 struct path *path, int *empty)
2597{
2598 return filename_lookup(dfd, getname_flags(name, flags, empty),
2599 flags, path, NULL);
2600}
2601EXPORT_SYMBOL(user_path_at_empty);
2602
2603/**
2604 * mountpoint_last - look up last component for umount
2605 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2606 *
2607 * This is a special lookup_last function just for umount. In this case, we
2608 * need to resolve the path without doing any revalidation.
2609 *
2610 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2611 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2612 * in almost all cases, this lookup will be served out of the dcache. The only
2613 * cases where it won't are if nd->last refers to a symlink or the path is
2614 * bogus and it doesn't exist.
2615 *
2616 * Returns:
2617 * -error: if there was an error during lookup. This includes -ENOENT if the
2618 * lookup found a negative dentry.
2619 *
2620 * 0: if we successfully resolved nd->last and found it to not to be a
2621 * symlink that needs to be followed.
2622 *
2623 * 1: if we successfully resolved nd->last and found it to be a symlink
2624 * that needs to be followed.
2625 */
2626static int
2627mountpoint_last(struct nameidata *nd)
2628{
2629 int error = 0;
2630 struct dentry *dir = nd->path.dentry;
2631 struct path path;
2632
2633 /* If we're in rcuwalk, drop out of it to handle last component */
2634 if (nd->flags & LOOKUP_RCU) {
2635 if (unlazy_walk(nd))
2636 return -ECHILD;
2637 }
2638
2639 nd->flags &= ~LOOKUP_PARENT;
2640
2641 if (unlikely(nd->last_type != LAST_NORM)) {
2642 error = handle_dots(nd, nd->last_type);
2643 if (error)
2644 return error;
2645 path.dentry = dget(nd->path.dentry);
2646 } else {
2647 path.dentry = d_lookup(dir, &nd->last);
2648 if (!path.dentry) {
2649 /*
2650 * No cached dentry. Mounted dentries are pinned in the
2651 * cache, so that means that this dentry is probably
2652 * a symlink or the path doesn't actually point
2653 * to a mounted dentry.
2654 */
2655 path.dentry = lookup_slow(&nd->last, dir,
2656 nd->flags | LOOKUP_NO_REVAL);
2657 if (IS_ERR(path.dentry))
2658 return PTR_ERR(path.dentry);
2659 }
2660 }
2661 if (d_is_negative(path.dentry)) {
2662 dput(path.dentry);
2663 return -ENOENT;
2664 }
2665 path.mnt = nd->path.mnt;
2666 return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0);
2667}
2668
2669/**
2670 * path_mountpoint - look up a path to be umounted
2671 * @nd: lookup context
2672 * @flags: lookup flags
2673 * @path: pointer to container for result
2674 *
2675 * Look up the given name, but don't attempt to revalidate the last component.
2676 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2677 */
2678static int
2679path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2680{
2681 const char *s = path_init(nd, flags);
2682 int err;
2683
2684 while (!(err = link_path_walk(s, nd)) &&
2685 (err = mountpoint_last(nd)) > 0) {
2686 s = trailing_symlink(nd);
2687 }
2688 if (!err) {
2689 *path = nd->path;
2690 nd->path.mnt = NULL;
2691 nd->path.dentry = NULL;
2692 follow_mount(path);
2693 }
2694 terminate_walk(nd);
2695 return err;
2696}
2697
2698static int
2699filename_mountpoint(int dfd, struct filename *name, struct path *path,
2700 unsigned int flags)
2701{
2702 struct nameidata nd;
2703 int error;
2704 if (IS_ERR(name))
2705 return PTR_ERR(name);
2706 set_nameidata(&nd, dfd, name);
2707 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2708 if (unlikely(error == -ECHILD))
2709 error = path_mountpoint(&nd, flags, path);
2710 if (unlikely(error == -ESTALE))
2711 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2712 if (likely(!error))
2713 audit_inode(name, path->dentry, AUDIT_INODE_NOEVAL);
2714 restore_nameidata();
2715 putname(name);
2716 return error;
2717}
2718
2719/**
2720 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2721 * @dfd: directory file descriptor
2722 * @name: pathname from userland
2723 * @flags: lookup flags
2724 * @path: pointer to container to hold result
2725 *
2726 * A umount is a special case for path walking. We're not actually interested
2727 * in the inode in this situation, and ESTALE errors can be a problem. We
2728 * simply want track down the dentry and vfsmount attached at the mountpoint
2729 * and avoid revalidating the last component.
2730 *
2731 * Returns 0 and populates "path" on success.
2732 */
2733int
2734user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2735 struct path *path)
2736{
2737 return filename_mountpoint(dfd, getname(name), path, flags);
2738}
2739
2740int
2741kern_path_mountpoint(int dfd, const char *name, struct path *path,
2742 unsigned int flags)
2743{
2744 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2745}
2746EXPORT_SYMBOL(kern_path_mountpoint);
2747
2748int __check_sticky(struct inode *dir, struct inode *inode)
2749{
2750 kuid_t fsuid = current_fsuid();
2751
2752 if (uid_eq(inode->i_uid, fsuid))
2753 return 0;
2754 if (uid_eq(dir->i_uid, fsuid))
2755 return 0;
2756 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2757}
2758EXPORT_SYMBOL(__check_sticky);
2759
2760/*
2761 * Check whether we can remove a link victim from directory dir, check
2762 * whether the type of victim is right.
2763 * 1. We can't do it if dir is read-only (done in permission())
2764 * 2. We should have write and exec permissions on dir
2765 * 3. We can't remove anything from append-only dir
2766 * 4. We can't do anything with immutable dir (done in permission())
2767 * 5. If the sticky bit on dir is set we should either
2768 * a. be owner of dir, or
2769 * b. be owner of victim, or
2770 * c. have CAP_FOWNER capability
2771 * 6. If the victim is append-only or immutable we can't do antyhing with
2772 * links pointing to it.
2773 * 7. If the victim has an unknown uid or gid we can't change the inode.
2774 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2775 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2776 * 10. We can't remove a root or mountpoint.
2777 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2778 * nfs_async_unlink().
2779 */
2780static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2781{
2782 struct inode *inode = d_backing_inode(victim);
2783 int error;
2784
2785 if (d_is_negative(victim))
2786 return -ENOENT;
2787 BUG_ON(!inode);
2788
2789 BUG_ON(victim->d_parent->d_inode != dir);
2790
2791 /* Inode writeback is not safe when the uid or gid are invalid. */
2792 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
2793 return -EOVERFLOW;
2794
2795 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2796
2797 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2798 if (error)
2799 return error;
2800 if (IS_APPEND(dir))
2801 return -EPERM;
2802
2803 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2804 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2805 return -EPERM;
2806 if (isdir) {
2807 if (!d_is_dir(victim))
2808 return -ENOTDIR;
2809 if (IS_ROOT(victim))
2810 return -EBUSY;
2811 } else if (d_is_dir(victim))
2812 return -EISDIR;
2813 if (IS_DEADDIR(dir))
2814 return -ENOENT;
2815 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2816 return -EBUSY;
2817 return 0;
2818}
2819
2820/* Check whether we can create an object with dentry child in directory
2821 * dir.
2822 * 1. We can't do it if child already exists (open has special treatment for
2823 * this case, but since we are inlined it's OK)
2824 * 2. We can't do it if dir is read-only (done in permission())
2825 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2826 * 4. We should have write and exec permissions on dir
2827 * 5. We can't do it if dir is immutable (done in permission())
2828 */
2829static inline int may_create(struct inode *dir, struct dentry *child)
2830{
2831 struct user_namespace *s_user_ns;
2832 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2833 if (child->d_inode)
2834 return -EEXIST;
2835 if (IS_DEADDIR(dir))
2836 return -ENOENT;
2837 s_user_ns = dir->i_sb->s_user_ns;
2838 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2839 !kgid_has_mapping(s_user_ns, current_fsgid()))
2840 return -EOVERFLOW;
2841 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2842}
2843
2844/*
2845 * p1 and p2 should be directories on the same fs.
2846 */
2847struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2848{
2849 struct dentry *p;
2850
2851 if (p1 == p2) {
2852 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2853 return NULL;
2854 }
2855
2856 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2857
2858 p = d_ancestor(p2, p1);
2859 if (p) {
2860 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2861 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2862 return p;
2863 }
2864
2865 p = d_ancestor(p1, p2);
2866 if (p) {
2867 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2868 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2869 return p;
2870 }
2871
2872 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2873 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2874 return NULL;
2875}
2876EXPORT_SYMBOL(lock_rename);
2877
2878void unlock_rename(struct dentry *p1, struct dentry *p2)
2879{
2880 inode_unlock(p1->d_inode);
2881 if (p1 != p2) {
2882 inode_unlock(p2->d_inode);
2883 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2884 }
2885}
2886EXPORT_SYMBOL(unlock_rename);
2887
2888int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2889 bool want_excl)
2890{
2891 int error = may_create(dir, dentry);
2892 if (error)
2893 return error;
2894
2895 if (!dir->i_op->create)
2896 return -EACCES; /* shouldn't it be ENOSYS? */
2897 mode &= S_IALLUGO;
2898 mode |= S_IFREG;
2899 error = security_inode_create(dir, dentry, mode);
2900 if (error)
2901 return error;
2902 error = dir->i_op->create(dir, dentry, mode, want_excl);
2903 if (!error)
2904 fsnotify_create(dir, dentry);
2905 return error;
2906}
2907EXPORT_SYMBOL(vfs_create);
2908
2909int vfs_mkobj(struct dentry *dentry, umode_t mode,
2910 int (*f)(struct dentry *, umode_t, void *),
2911 void *arg)
2912{
2913 struct inode *dir = dentry->d_parent->d_inode;
2914 int error = may_create(dir, dentry);
2915 if (error)
2916 return error;
2917
2918 mode &= S_IALLUGO;
2919 mode |= S_IFREG;
2920 error = security_inode_create(dir, dentry, mode);
2921 if (error)
2922 return error;
2923 error = f(dentry, mode, arg);
2924 if (!error)
2925 fsnotify_create(dir, dentry);
2926 return error;
2927}
2928EXPORT_SYMBOL(vfs_mkobj);
2929
2930bool may_open_dev(const struct path *path)
2931{
2932 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2933 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2934}
2935
2936static int may_open(const struct path *path, int acc_mode, int flag)
2937{
2938 struct dentry *dentry = path->dentry;
2939 struct inode *inode = dentry->d_inode;
2940 int error;
2941
2942 if (!inode)
2943 return -ENOENT;
2944
2945 switch (inode->i_mode & S_IFMT) {
2946 case S_IFLNK:
2947 return -ELOOP;
2948 case S_IFDIR:
2949 if (acc_mode & MAY_WRITE)
2950 return -EISDIR;
2951 break;
2952 case S_IFBLK:
2953 case S_IFCHR:
2954 if (!may_open_dev(path))
2955 return -EACCES;
2956 /*FALLTHRU*/
2957 case S_IFIFO:
2958 case S_IFSOCK:
2959 flag &= ~O_TRUNC;
2960 break;
2961 }
2962
2963 error = inode_permission(inode, MAY_OPEN | acc_mode);
2964 if (error)
2965 return error;
2966
2967 /*
2968 * An append-only file must be opened in append mode for writing.
2969 */
2970 if (IS_APPEND(inode)) {
2971 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2972 return -EPERM;
2973 if (flag & O_TRUNC)
2974 return -EPERM;
2975 }
2976
2977 /* O_NOATIME can only be set by the owner or superuser */
2978 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2979 return -EPERM;
2980
2981 return 0;
2982}
2983
2984static int handle_truncate(struct file *filp)
2985{
2986 const struct path *path = &filp->f_path;
2987 struct inode *inode = path->dentry->d_inode;
2988 int error = get_write_access(inode);
2989 if (error)
2990 return error;
2991 /*
2992 * Refuse to truncate files with mandatory locks held on them.
2993 */
2994 error = locks_verify_locked(filp);
2995 if (!error)
2996 error = security_path_truncate(path);
2997 if (!error) {
2998 error = do_truncate(path->dentry, 0,
2999 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3000 filp);
3001 }
3002 put_write_access(inode);
3003 return error;
3004}
3005
3006static inline int open_to_namei_flags(int flag)
3007{
3008 if ((flag & O_ACCMODE) == 3)
3009 flag--;
3010 return flag;
3011}
3012
3013static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
3014{
3015 struct user_namespace *s_user_ns;
3016 int error = security_path_mknod(dir, dentry, mode, 0);
3017 if (error)
3018 return error;
3019
3020 s_user_ns = dir->dentry->d_sb->s_user_ns;
3021 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
3022 !kgid_has_mapping(s_user_ns, current_fsgid()))
3023 return -EOVERFLOW;
3024
3025 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
3026 if (error)
3027 return error;
3028
3029 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3030}
3031
3032/*
3033 * Attempt to atomically look up, create and open a file from a negative
3034 * dentry.
3035 *
3036 * Returns 0 if successful. The file will have been created and attached to
3037 * @file by the filesystem calling finish_open().
3038 *
3039 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3040 * be set. The caller will need to perform the open themselves. @path will
3041 * have been updated to point to the new dentry. This may be negative.
3042 *
3043 * Returns an error code otherwise.
3044 */
3045static int atomic_open(struct nameidata *nd, struct dentry *dentry,
3046 struct path *path, struct file *file,
3047 const struct open_flags *op,
3048 int open_flag, umode_t mode)
3049{
3050 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3051 struct inode *dir = nd->path.dentry->d_inode;
3052 int error;
3053
3054 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
3055 open_flag &= ~O_TRUNC;
3056
3057 if (nd->flags & LOOKUP_DIRECTORY)
3058 open_flag |= O_DIRECTORY;
3059
3060 file->f_path.dentry = DENTRY_NOT_SET;
3061 file->f_path.mnt = nd->path.mnt;
3062 error = dir->i_op->atomic_open(dir, dentry, file,
3063 open_to_namei_flags(open_flag), mode);
3064 d_lookup_done(dentry);
3065 if (!error) {
3066 if (file->f_mode & FMODE_OPENED) {
3067 /*
3068 * We didn't have the inode before the open, so check open
3069 * permission here.
3070 */
3071 int acc_mode = op->acc_mode;
3072 if (file->f_mode & FMODE_CREATED) {
3073 WARN_ON(!(open_flag & O_CREAT));
3074 fsnotify_create(dir, dentry);
3075 acc_mode = 0;
3076 }
3077 error = may_open(&file->f_path, acc_mode, open_flag);
3078 if (WARN_ON(error > 0))
3079 error = -EINVAL;
3080 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3081 error = -EIO;
3082 } else {
3083 if (file->f_path.dentry) {
3084 dput(dentry);
3085 dentry = file->f_path.dentry;
3086 }
3087 if (file->f_mode & FMODE_CREATED)
3088 fsnotify_create(dir, dentry);
3089 if (unlikely(d_is_negative(dentry))) {
3090 error = -ENOENT;
3091 } else {
3092 path->dentry = dentry;
3093 path->mnt = nd->path.mnt;
3094 return 0;
3095 }
3096 }
3097 }
3098 dput(dentry);
3099 return error;
3100}
3101
3102/*
3103 * Look up and maybe create and open the last component.
3104 *
3105 * Must be called with parent locked (exclusive in O_CREAT case).
3106 *
3107 * Returns 0 on success, that is, if
3108 * the file was successfully atomically created (if necessary) and opened, or
3109 * the file was not completely opened at this time, though lookups and
3110 * creations were performed.
3111 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3112 * In the latter case dentry returned in @path might be negative if O_CREAT
3113 * hadn't been specified.
3114 *
3115 * An error code is returned on failure.
3116 */
3117static int lookup_open(struct nameidata *nd, struct path *path,
3118 struct file *file,
3119 const struct open_flags *op,
3120 bool got_write)
3121{
3122 struct dentry *dir = nd->path.dentry;
3123 struct inode *dir_inode = dir->d_inode;
3124 int open_flag = op->open_flag;
3125 struct dentry *dentry;
3126 int error, create_error = 0;
3127 umode_t mode = op->mode;
3128 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3129
3130 if (unlikely(IS_DEADDIR(dir_inode)))
3131 return -ENOENT;
3132
3133 file->f_mode &= ~FMODE_CREATED;
3134 dentry = d_lookup(dir, &nd->last);
3135 for (;;) {
3136 if (!dentry) {
3137 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3138 if (IS_ERR(dentry))
3139 return PTR_ERR(dentry);
3140 }
3141 if (d_in_lookup(dentry))
3142 break;
3143
3144 error = d_revalidate(dentry, nd->flags);
3145 if (likely(error > 0))
3146 break;
3147 if (error)
3148 goto out_dput;
3149 d_invalidate(dentry);
3150 dput(dentry);
3151 dentry = NULL;
3152 }
3153 if (dentry->d_inode) {
3154 /* Cached positive dentry: will open in f_op->open */
3155 goto out_no_open;
3156 }
3157
3158 /*
3159 * Checking write permission is tricky, bacuse we don't know if we are
3160 * going to actually need it: O_CREAT opens should work as long as the
3161 * file exists. But checking existence breaks atomicity. The trick is
3162 * to check access and if not granted clear O_CREAT from the flags.
3163 *
3164 * Another problem is returing the "right" error value (e.g. for an
3165 * O_EXCL open we want to return EEXIST not EROFS).
3166 */
3167 if (open_flag & O_CREAT) {
3168 if (!IS_POSIXACL(dir->d_inode))
3169 mode &= ~current_umask();
3170 if (unlikely(!got_write)) {
3171 create_error = -EROFS;
3172 open_flag &= ~O_CREAT;
3173 if (open_flag & (O_EXCL | O_TRUNC))
3174 goto no_open;
3175 /* No side effects, safe to clear O_CREAT */
3176 } else {
3177 create_error = may_o_create(&nd->path, dentry, mode);
3178 if (create_error) {
3179 open_flag &= ~O_CREAT;
3180 if (open_flag & O_EXCL)
3181 goto no_open;
3182 }
3183 }
3184 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3185 unlikely(!got_write)) {
3186 /*
3187 * No O_CREATE -> atomicity not a requirement -> fall
3188 * back to lookup + open
3189 */
3190 goto no_open;
3191 }
3192
3193 if (dir_inode->i_op->atomic_open) {
3194 error = atomic_open(nd, dentry, path, file, op, open_flag,
3195 mode);
3196 if (unlikely(error == -ENOENT) && create_error)
3197 error = create_error;
3198 return error;
3199 }
3200
3201no_open:
3202 if (d_in_lookup(dentry)) {
3203 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3204 nd->flags);
3205 d_lookup_done(dentry);
3206 if (unlikely(res)) {
3207 if (IS_ERR(res)) {
3208 error = PTR_ERR(res);
3209 goto out_dput;
3210 }
3211 dput(dentry);
3212 dentry = res;
3213 }
3214 }
3215
3216 /* Negative dentry, just create the file */
3217 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3218 file->f_mode |= FMODE_CREATED;
3219 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3220 if (!dir_inode->i_op->create) {
3221 error = -EACCES;
3222 goto out_dput;
3223 }
3224 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3225 open_flag & O_EXCL);
3226 if (error)
3227 goto out_dput;
3228 fsnotify_create(dir_inode, dentry);
3229 }
3230 if (unlikely(create_error) && !dentry->d_inode) {
3231 error = create_error;
3232 goto out_dput;
3233 }
3234out_no_open:
3235 path->dentry = dentry;
3236 path->mnt = nd->path.mnt;
3237 return 0;
3238
3239out_dput:
3240 dput(dentry);
3241 return error;
3242}
3243
3244/*
3245 * Handle the last step of open()
3246 */
3247static int do_last(struct nameidata *nd,
3248 struct file *file, const struct open_flags *op)
3249{
3250 struct dentry *dir = nd->path.dentry;
3251 int open_flag = op->open_flag;
3252 bool will_truncate = (open_flag & O_TRUNC) != 0;
3253 bool got_write = false;
3254 int acc_mode = op->acc_mode;
3255 unsigned seq;
3256 struct inode *inode;
3257 struct path path;
3258 int error;
3259
3260 nd->flags &= ~LOOKUP_PARENT;
3261 nd->flags |= op->intent;
3262
3263 if (nd->last_type != LAST_NORM) {
3264 error = handle_dots(nd, nd->last_type);
3265 if (unlikely(error))
3266 return error;
3267 goto finish_open;
3268 }
3269
3270 if (!(open_flag & O_CREAT)) {
3271 if (nd->last.name[nd->last.len])
3272 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3273 /* we _can_ be in RCU mode here */
3274 error = lookup_fast(nd, &path, &inode, &seq);
3275 if (likely(error > 0))
3276 goto finish_lookup;
3277
3278 if (error < 0)
3279 return error;
3280
3281 BUG_ON(nd->inode != dir->d_inode);
3282 BUG_ON(nd->flags & LOOKUP_RCU);
3283 } else {
3284 /* create side of things */
3285 /*
3286 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3287 * has been cleared when we got to the last component we are
3288 * about to look up
3289 */
3290 error = complete_walk(nd);
3291 if (error)
3292 return error;
3293
3294 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3295 /* trailing slashes? */
3296 if (unlikely(nd->last.name[nd->last.len]))
3297 return -EISDIR;
3298 }
3299
3300 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3301 error = mnt_want_write(nd->path.mnt);
3302 if (!error)
3303 got_write = true;
3304 /*
3305 * do _not_ fail yet - we might not need that or fail with
3306 * a different error; let lookup_open() decide; we'll be
3307 * dropping this one anyway.
3308 */
3309 }
3310 if (open_flag & O_CREAT)
3311 inode_lock(dir->d_inode);
3312 else
3313 inode_lock_shared(dir->d_inode);
3314 error = lookup_open(nd, &path, file, op, got_write);
3315 if (open_flag & O_CREAT)
3316 inode_unlock(dir->d_inode);
3317 else
3318 inode_unlock_shared(dir->d_inode);
3319
3320 if (error)
3321 goto out;
3322
3323 if (file->f_mode & FMODE_OPENED) {
3324 if ((file->f_mode & FMODE_CREATED) ||
3325 !S_ISREG(file_inode(file)->i_mode))
3326 will_truncate = false;
3327
3328 audit_inode(nd->name, file->f_path.dentry, 0);
3329 goto opened;
3330 }
3331
3332 if (file->f_mode & FMODE_CREATED) {
3333 /* Don't check for write permission, don't truncate */
3334 open_flag &= ~O_TRUNC;
3335 will_truncate = false;
3336 acc_mode = 0;
3337 path_to_nameidata(&path, nd);
3338 goto finish_open_created;
3339 }
3340
3341 /*
3342 * If atomic_open() acquired write access it is dropped now due to
3343 * possible mount and symlink following (this might be optimized away if
3344 * necessary...)
3345 */
3346 if (got_write) {
3347 mnt_drop_write(nd->path.mnt);
3348 got_write = false;
3349 }
3350
3351 error = follow_managed(&path, nd);
3352 if (unlikely(error < 0))
3353 return error;
3354
3355 if (unlikely(d_is_negative(path.dentry))) {
3356 path_to_nameidata(&path, nd);
3357 return -ENOENT;
3358 }
3359
3360 /*
3361 * create/update audit record if it already exists.
3362 */
3363 audit_inode(nd->name, path.dentry, 0);
3364
3365 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3366 path_to_nameidata(&path, nd);
3367 return -EEXIST;
3368 }
3369
3370 seq = 0; /* out of RCU mode, so the value doesn't matter */
3371 inode = d_backing_inode(path.dentry);
3372finish_lookup:
3373 error = step_into(nd, &path, 0, inode, seq);
3374 if (unlikely(error))
3375 return error;
3376finish_open:
3377 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3378 error = complete_walk(nd);
3379 if (error)
3380 return error;
3381 audit_inode(nd->name, nd->path.dentry, 0);
3382 if (open_flag & O_CREAT) {
3383 error = -EISDIR;
3384 if (d_is_dir(nd->path.dentry))
3385 goto out;
3386 error = may_create_in_sticky(dir,
3387 d_backing_inode(nd->path.dentry));
3388 if (unlikely(error))
3389 goto out;
3390 }
3391 error = -ENOTDIR;
3392 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3393 goto out;
3394 if (!d_is_reg(nd->path.dentry))
3395 will_truncate = false;
3396
3397 if (will_truncate) {
3398 error = mnt_want_write(nd->path.mnt);
3399 if (error)
3400 goto out;
3401 got_write = true;
3402 }
3403finish_open_created:
3404 error = may_open(&nd->path, acc_mode, open_flag);
3405 if (error)
3406 goto out;
3407 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
3408 error = vfs_open(&nd->path, file);
3409 if (error)
3410 goto out;
3411opened:
3412 error = ima_file_check(file, op->acc_mode);
3413 if (!error && will_truncate)
3414 error = handle_truncate(file);
3415out:
3416 if (unlikely(error > 0)) {
3417 WARN_ON(1);
3418 error = -EINVAL;
3419 }
3420 if (got_write)
3421 mnt_drop_write(nd->path.mnt);
3422 return error;
3423}
3424
3425struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag)
3426{
3427 struct dentry *child = NULL;
3428 struct inode *dir = dentry->d_inode;
3429 struct inode *inode;
3430 int error;
3431
3432 /* we want directory to be writable */
3433 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3434 if (error)
3435 goto out_err;
3436 error = -EOPNOTSUPP;
3437 if (!dir->i_op->tmpfile)
3438 goto out_err;
3439 error = -ENOMEM;
3440 child = d_alloc(dentry, &slash_name);
3441 if (unlikely(!child))
3442 goto out_err;
3443 error = dir->i_op->tmpfile(dir, child, mode);
3444 if (error)
3445 goto out_err;
3446 error = -ENOENT;
3447 inode = child->d_inode;
3448 if (unlikely(!inode))
3449 goto out_err;
3450 if (!(open_flag & O_EXCL)) {
3451 spin_lock(&inode->i_lock);
3452 inode->i_state |= I_LINKABLE;
3453 spin_unlock(&inode->i_lock);
3454 }
3455 ima_post_create_tmpfile(inode);
3456 return child;
3457
3458out_err:
3459 dput(child);
3460 return ERR_PTR(error);
3461}
3462EXPORT_SYMBOL(vfs_tmpfile);
3463
3464static int do_tmpfile(struct nameidata *nd, unsigned flags,
3465 const struct open_flags *op,
3466 struct file *file)
3467{
3468 struct dentry *child;
3469 struct path path;
3470 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3471 if (unlikely(error))
3472 return error;
3473 error = mnt_want_write(path.mnt);
3474 if (unlikely(error))
3475 goto out;
3476 child = vfs_tmpfile(path.dentry, op->mode, op->open_flag);
3477 error = PTR_ERR(child);
3478 if (IS_ERR(child))
3479 goto out2;
3480 dput(path.dentry);
3481 path.dentry = child;
3482 audit_inode(nd->name, child, 0);
3483 /* Don't check for other permissions, the inode was just created */
3484 error = may_open(&path, 0, op->open_flag);
3485 if (error)
3486 goto out2;
3487 file->f_path.mnt = path.mnt;
3488 error = finish_open(file, child, NULL);
3489out2:
3490 mnt_drop_write(path.mnt);
3491out:
3492 path_put(&path);
3493 return error;
3494}
3495
3496static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3497{
3498 struct path path;
3499 int error = path_lookupat(nd, flags, &path);
3500 if (!error) {
3501 audit_inode(nd->name, path.dentry, 0);
3502 error = vfs_open(&path, file);
3503 path_put(&path);
3504 }
3505 return error;
3506}
3507
3508static struct file *path_openat(struct nameidata *nd,
3509 const struct open_flags *op, unsigned flags)
3510{
3511 struct file *file;
3512 int error;
3513
3514 file = alloc_empty_file(op->open_flag, current_cred());
3515 if (IS_ERR(file))
3516 return file;
3517
3518 if (unlikely(file->f_flags & __O_TMPFILE)) {
3519 error = do_tmpfile(nd, flags, op, file);
3520 } else if (unlikely(file->f_flags & O_PATH)) {
3521 error = do_o_path(nd, flags, file);
3522 } else {
3523 const char *s = path_init(nd, flags);
3524 while (!(error = link_path_walk(s, nd)) &&
3525 (error = do_last(nd, file, op)) > 0) {
3526 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3527 s = trailing_symlink(nd);
3528 }
3529 terminate_walk(nd);
3530 }
3531 if (likely(!error)) {
3532 if (likely(file->f_mode & FMODE_OPENED))
3533 return file;
3534 WARN_ON(1);
3535 error = -EINVAL;
3536 }
3537 fput(file);
3538 if (error == -EOPENSTALE) {
3539 if (flags & LOOKUP_RCU)
3540 error = -ECHILD;
3541 else
3542 error = -ESTALE;
3543 }
3544 return ERR_PTR(error);
3545}
3546
3547struct file *do_filp_open(int dfd, struct filename *pathname,
3548 const struct open_flags *op)
3549{
3550 struct nameidata nd;
3551 int flags = op->lookup_flags;
3552 struct file *filp;
3553
3554 set_nameidata(&nd, dfd, pathname);
3555 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3556 if (unlikely(filp == ERR_PTR(-ECHILD)))
3557 filp = path_openat(&nd, op, flags);
3558 if (unlikely(filp == ERR_PTR(-ESTALE)))
3559 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3560 restore_nameidata();
3561 return filp;
3562}
3563
3564struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3565 const char *name, const struct open_flags *op)
3566{
3567 struct nameidata nd;
3568 struct file *file;
3569 struct filename *filename;
3570 int flags = op->lookup_flags | LOOKUP_ROOT;
3571
3572 nd.root.mnt = mnt;
3573 nd.root.dentry = dentry;
3574
3575 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3576 return ERR_PTR(-ELOOP);
3577
3578 filename = getname_kernel(name);
3579 if (IS_ERR(filename))
3580 return ERR_CAST(filename);
3581
3582 set_nameidata(&nd, -1, filename);
3583 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3584 if (unlikely(file == ERR_PTR(-ECHILD)))
3585 file = path_openat(&nd, op, flags);
3586 if (unlikely(file == ERR_PTR(-ESTALE)))
3587 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3588 restore_nameidata();
3589 putname(filename);
3590 return file;
3591}
3592
3593static struct dentry *filename_create(int dfd, struct filename *name,
3594 struct path *path, unsigned int lookup_flags)
3595{
3596 struct dentry *dentry = ERR_PTR(-EEXIST);
3597 struct qstr last;
3598 int type;
3599 int err2;
3600 int error;
3601 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3602
3603 /*
3604 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3605 * other flags passed in are ignored!
3606 */
3607 lookup_flags &= LOOKUP_REVAL;
3608
3609 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3610 if (IS_ERR(name))
3611 return ERR_CAST(name);
3612
3613 /*
3614 * Yucky last component or no last component at all?
3615 * (foo/., foo/.., /////)
3616 */
3617 if (unlikely(type != LAST_NORM))
3618 goto out;
3619
3620 /* don't fail immediately if it's r/o, at least try to report other errors */
3621 err2 = mnt_want_write(path->mnt);
3622 /*
3623 * Do the final lookup.
3624 */
3625 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3626 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3627 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3628 if (IS_ERR(dentry))
3629 goto unlock;
3630
3631 error = -EEXIST;
3632 if (d_is_positive(dentry))
3633 goto fail;
3634
3635 /*
3636 * Special case - lookup gave negative, but... we had foo/bar/
3637 * From the vfs_mknod() POV we just have a negative dentry -
3638 * all is fine. Let's be bastards - you had / on the end, you've
3639 * been asking for (non-existent) directory. -ENOENT for you.
3640 */
3641 if (unlikely(!is_dir && last.name[last.len])) {
3642 error = -ENOENT;
3643 goto fail;
3644 }
3645 if (unlikely(err2)) {
3646 error = err2;
3647 goto fail;
3648 }
3649 putname(name);
3650 return dentry;
3651fail:
3652 dput(dentry);
3653 dentry = ERR_PTR(error);
3654unlock:
3655 inode_unlock(path->dentry->d_inode);
3656 if (!err2)
3657 mnt_drop_write(path->mnt);
3658out:
3659 path_put(path);
3660 putname(name);
3661 return dentry;
3662}
3663
3664struct dentry *kern_path_create(int dfd, const char *pathname,
3665 struct path *path, unsigned int lookup_flags)
3666{
3667 return filename_create(dfd, getname_kernel(pathname),
3668 path, lookup_flags);
3669}
3670EXPORT_SYMBOL(kern_path_create);
3671
3672void done_path_create(struct path *path, struct dentry *dentry)
3673{
3674 dput(dentry);
3675 inode_unlock(path->dentry->d_inode);
3676 mnt_drop_write(path->mnt);
3677 path_put(path);
3678}
3679EXPORT_SYMBOL(done_path_create);
3680
3681inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3682 struct path *path, unsigned int lookup_flags)
3683{
3684 return filename_create(dfd, getname(pathname), path, lookup_flags);
3685}
3686EXPORT_SYMBOL(user_path_create);
3687
3688int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3689{
3690 int error = may_create(dir, dentry);
3691
3692 if (error)
3693 return error;
3694
3695 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3696 return -EPERM;
3697
3698 if (!dir->i_op->mknod)
3699 return -EPERM;
3700
3701 error = devcgroup_inode_mknod(mode, dev);
3702 if (error)
3703 return error;
3704
3705 error = security_inode_mknod(dir, dentry, mode, dev);
3706 if (error)
3707 return error;
3708
3709 error = dir->i_op->mknod(dir, dentry, mode, dev);
3710 if (!error)
3711 fsnotify_create(dir, dentry);
3712 return error;
3713}
3714EXPORT_SYMBOL(vfs_mknod);
3715
3716static int may_mknod(umode_t mode)
3717{
3718 switch (mode & S_IFMT) {
3719 case S_IFREG:
3720 case S_IFCHR:
3721 case S_IFBLK:
3722 case S_IFIFO:
3723 case S_IFSOCK:
3724 case 0: /* zero mode translates to S_IFREG */
3725 return 0;
3726 case S_IFDIR:
3727 return -EPERM;
3728 default:
3729 return -EINVAL;
3730 }
3731}
3732
3733long do_mknodat(int dfd, const char __user *filename, umode_t mode,
3734 unsigned int dev)
3735{
3736 struct dentry *dentry;
3737 struct path path;
3738 int error;
3739 unsigned int lookup_flags = 0;
3740
3741 error = may_mknod(mode);
3742 if (error)
3743 return error;
3744retry:
3745 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3746 if (IS_ERR(dentry))
3747 return PTR_ERR(dentry);
3748
3749 if (!IS_POSIXACL(path.dentry->d_inode))
3750 mode &= ~current_umask();
3751 error = security_path_mknod(&path, dentry, mode, dev);
3752 if (error)
3753 goto out;
3754 switch (mode & S_IFMT) {
3755 case 0: case S_IFREG:
3756 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3757 if (!error)
3758 ima_post_path_mknod(dentry);
3759 break;
3760 case S_IFCHR: case S_IFBLK:
3761 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3762 new_decode_dev(dev));
3763 break;
3764 case S_IFIFO: case S_IFSOCK:
3765 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3766 break;
3767 }
3768out:
3769 done_path_create(&path, dentry);
3770 if (retry_estale(error, lookup_flags)) {
3771 lookup_flags |= LOOKUP_REVAL;
3772 goto retry;
3773 }
3774 return error;
3775}
3776
3777SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3778 unsigned int, dev)
3779{
3780 return do_mknodat(dfd, filename, mode, dev);
3781}
3782
3783SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3784{
3785 return do_mknodat(AT_FDCWD, filename, mode, dev);
3786}
3787
3788int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3789{
3790 int error = may_create(dir, dentry);
3791 unsigned max_links = dir->i_sb->s_max_links;
3792
3793 if (error)
3794 return error;
3795
3796 if (!dir->i_op->mkdir)
3797 return -EPERM;
3798
3799 mode &= (S_IRWXUGO|S_ISVTX);
3800 error = security_inode_mkdir(dir, dentry, mode);
3801 if (error)
3802 return error;
3803
3804 if (max_links && dir->i_nlink >= max_links)
3805 return -EMLINK;
3806
3807 error = dir->i_op->mkdir(dir, dentry, mode);
3808 if (!error)
3809 fsnotify_mkdir(dir, dentry);
3810 return error;
3811}
3812EXPORT_SYMBOL(vfs_mkdir);
3813
3814long do_mkdirat(int dfd, const char __user *pathname, umode_t mode)
3815{
3816 struct dentry *dentry;
3817 struct path path;
3818 int error;
3819 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3820
3821retry:
3822 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3823 if (IS_ERR(dentry))
3824 return PTR_ERR(dentry);
3825
3826 if (!IS_POSIXACL(path.dentry->d_inode))
3827 mode &= ~current_umask();
3828 error = security_path_mkdir(&path, dentry, mode);
3829 if (!error)
3830 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3831 done_path_create(&path, dentry);
3832 if (retry_estale(error, lookup_flags)) {
3833 lookup_flags |= LOOKUP_REVAL;
3834 goto retry;
3835 }
3836 return error;
3837}
3838
3839SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3840{
3841 return do_mkdirat(dfd, pathname, mode);
3842}
3843
3844SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3845{
3846 return do_mkdirat(AT_FDCWD, pathname, mode);
3847}
3848
3849int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3850{
3851 int error = may_delete(dir, dentry, 1);
3852
3853 if (error)
3854 return error;
3855
3856 if (!dir->i_op->rmdir)
3857 return -EPERM;
3858
3859 dget(dentry);
3860 inode_lock(dentry->d_inode);
3861
3862 error = -EBUSY;
3863 if (is_local_mountpoint(dentry))
3864 goto out;
3865
3866 error = security_inode_rmdir(dir, dentry);
3867 if (error)
3868 goto out;
3869
3870 error = dir->i_op->rmdir(dir, dentry);
3871 if (error)
3872 goto out;
3873
3874 shrink_dcache_parent(dentry);
3875 dentry->d_inode->i_flags |= S_DEAD;
3876 dont_mount(dentry);
3877 detach_mounts(dentry);
3878 fsnotify_rmdir(dir, dentry);
3879
3880out:
3881 inode_unlock(dentry->d_inode);
3882 dput(dentry);
3883 if (!error)
3884 d_delete(dentry);
3885 return error;
3886}
3887EXPORT_SYMBOL(vfs_rmdir);
3888
3889long do_rmdir(int dfd, const char __user *pathname)
3890{
3891 int error = 0;
3892 struct filename *name;
3893 struct dentry *dentry;
3894 struct path path;
3895 struct qstr last;
3896 int type;
3897 unsigned int lookup_flags = 0;
3898retry:
3899 name = filename_parentat(dfd, getname(pathname), lookup_flags,
3900 &path, &last, &type);
3901 if (IS_ERR(name))
3902 return PTR_ERR(name);
3903
3904 switch (type) {
3905 case LAST_DOTDOT:
3906 error = -ENOTEMPTY;
3907 goto exit1;
3908 case LAST_DOT:
3909 error = -EINVAL;
3910 goto exit1;
3911 case LAST_ROOT:
3912 error = -EBUSY;
3913 goto exit1;
3914 }
3915
3916 error = mnt_want_write(path.mnt);
3917 if (error)
3918 goto exit1;
3919
3920 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3921 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3922 error = PTR_ERR(dentry);
3923 if (IS_ERR(dentry))
3924 goto exit2;
3925 if (!dentry->d_inode) {
3926 error = -ENOENT;
3927 goto exit3;
3928 }
3929 error = security_path_rmdir(&path, dentry);
3930 if (error)
3931 goto exit3;
3932 error = vfs_rmdir(path.dentry->d_inode, dentry);
3933exit3:
3934 dput(dentry);
3935exit2:
3936 inode_unlock(path.dentry->d_inode);
3937 mnt_drop_write(path.mnt);
3938exit1:
3939 path_put(&path);
3940 putname(name);
3941 if (retry_estale(error, lookup_flags)) {
3942 lookup_flags |= LOOKUP_REVAL;
3943 goto retry;
3944 }
3945 return error;
3946}
3947
3948SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3949{
3950 return do_rmdir(AT_FDCWD, pathname);
3951}
3952
3953/**
3954 * vfs_unlink - unlink a filesystem object
3955 * @dir: parent directory
3956 * @dentry: victim
3957 * @delegated_inode: returns victim inode, if the inode is delegated.
3958 *
3959 * The caller must hold dir->i_mutex.
3960 *
3961 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3962 * return a reference to the inode in delegated_inode. The caller
3963 * should then break the delegation on that inode and retry. Because
3964 * breaking a delegation may take a long time, the caller should drop
3965 * dir->i_mutex before doing so.
3966 *
3967 * Alternatively, a caller may pass NULL for delegated_inode. This may
3968 * be appropriate for callers that expect the underlying filesystem not
3969 * to be NFS exported.
3970 */
3971int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3972{
3973 struct inode *target = dentry->d_inode;
3974 int error = may_delete(dir, dentry, 0);
3975
3976 if (error)
3977 return error;
3978
3979 if (!dir->i_op->unlink)
3980 return -EPERM;
3981
3982 inode_lock(target);
3983 if (is_local_mountpoint(dentry))
3984 error = -EBUSY;
3985 else {
3986 error = security_inode_unlink(dir, dentry);
3987 if (!error) {
3988 error = try_break_deleg(target, delegated_inode);
3989 if (error)
3990 goto out;
3991 error = dir->i_op->unlink(dir, dentry);
3992 if (!error) {
3993 dont_mount(dentry);
3994 detach_mounts(dentry);
3995 fsnotify_unlink(dir, dentry);
3996 }
3997 }
3998 }
3999out:
4000 inode_unlock(target);
4001
4002 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4003 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
4004 fsnotify_link_count(target);
4005 d_delete(dentry);
4006 }
4007
4008 return error;
4009}
4010EXPORT_SYMBOL(vfs_unlink);
4011
4012/*
4013 * Make sure that the actual truncation of the file will occur outside its
4014 * directory's i_mutex. Truncate can take a long time if there is a lot of
4015 * writeout happening, and we don't want to prevent access to the directory
4016 * while waiting on the I/O.
4017 */
4018long do_unlinkat(int dfd, struct filename *name)
4019{
4020 int error;
4021 struct dentry *dentry;
4022 struct path path;
4023 struct qstr last;
4024 int type;
4025 struct inode *inode = NULL;
4026 struct inode *delegated_inode = NULL;
4027 unsigned int lookup_flags = 0;
4028retry:
4029 name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4030 if (IS_ERR(name))
4031 return PTR_ERR(name);
4032
4033 error = -EISDIR;
4034 if (type != LAST_NORM)
4035 goto exit1;
4036
4037 error = mnt_want_write(path.mnt);
4038 if (error)
4039 goto exit1;
4040retry_deleg:
4041 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4042 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4043 error = PTR_ERR(dentry);
4044 if (!IS_ERR(dentry)) {
4045 /* Why not before? Because we want correct error value */
4046 if (last.name[last.len])
4047 goto slashes;
4048 inode = dentry->d_inode;
4049 if (d_is_negative(dentry))
4050 goto slashes;
4051 ihold(inode);
4052 error = security_path_unlink(&path, dentry);
4053 if (error)
4054 goto exit2;
4055 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
4056exit2:
4057 dput(dentry);
4058 }
4059 inode_unlock(path.dentry->d_inode);
4060 if (inode)
4061 iput(inode); /* truncate the inode here */
4062 inode = NULL;
4063 if (delegated_inode) {
4064 error = break_deleg_wait(&delegated_inode);
4065 if (!error)
4066 goto retry_deleg;
4067 }
4068 mnt_drop_write(path.mnt);
4069exit1:
4070 path_put(&path);
4071 if (retry_estale(error, lookup_flags)) {
4072 lookup_flags |= LOOKUP_REVAL;
4073 inode = NULL;
4074 goto retry;
4075 }
4076 putname(name);
4077 return error;
4078
4079slashes:
4080 if (d_is_negative(dentry))
4081 error = -ENOENT;
4082 else if (d_is_dir(dentry))
4083 error = -EISDIR;
4084 else
4085 error = -ENOTDIR;
4086 goto exit2;
4087}
4088
4089SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4090{
4091 if ((flag & ~AT_REMOVEDIR) != 0)
4092 return -EINVAL;
4093
4094 if (flag & AT_REMOVEDIR)
4095 return do_rmdir(dfd, pathname);
4096
4097 return do_unlinkat(dfd, getname(pathname));
4098}
4099
4100SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4101{
4102 return do_unlinkat(AT_FDCWD, getname(pathname));
4103}
4104
4105int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
4106{
4107 int error = may_create(dir, dentry);
4108
4109 if (error)
4110 return error;
4111
4112 if (!dir->i_op->symlink)
4113 return -EPERM;
4114
4115 error = security_inode_symlink(dir, dentry, oldname);
4116 if (error)
4117 return error;
4118
4119 error = dir->i_op->symlink(dir, dentry, oldname);
4120 if (!error)
4121 fsnotify_create(dir, dentry);
4122 return error;
4123}
4124EXPORT_SYMBOL(vfs_symlink);
4125
4126long do_symlinkat(const char __user *oldname, int newdfd,
4127 const char __user *newname)
4128{
4129 int error;
4130 struct filename *from;
4131 struct dentry *dentry;
4132 struct path path;
4133 unsigned int lookup_flags = 0;
4134
4135 from = getname(oldname);
4136 if (IS_ERR(from))
4137 return PTR_ERR(from);
4138retry:
4139 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4140 error = PTR_ERR(dentry);
4141 if (IS_ERR(dentry))
4142 goto out_putname;
4143
4144 error = security_path_symlink(&path, dentry, from->name);
4145 if (!error)
4146 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4147 done_path_create(&path, dentry);
4148 if (retry_estale(error, lookup_flags)) {
4149 lookup_flags |= LOOKUP_REVAL;
4150 goto retry;
4151 }
4152out_putname:
4153 putname(from);
4154 return error;
4155}
4156
4157SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4158 int, newdfd, const char __user *, newname)
4159{
4160 return do_symlinkat(oldname, newdfd, newname);
4161}
4162
4163SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4164{
4165 return do_symlinkat(oldname, AT_FDCWD, newname);
4166}
4167
4168/**
4169 * vfs_link - create a new link
4170 * @old_dentry: object to be linked
4171 * @dir: new parent
4172 * @new_dentry: where to create the new link
4173 * @delegated_inode: returns inode needing a delegation break
4174 *
4175 * The caller must hold dir->i_mutex
4176 *
4177 * If vfs_link discovers a delegation on the to-be-linked file in need
4178 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4179 * inode in delegated_inode. The caller should then break the delegation
4180 * and retry. Because breaking a delegation may take a long time, the
4181 * caller should drop the i_mutex before doing so.
4182 *
4183 * Alternatively, a caller may pass NULL for delegated_inode. This may
4184 * be appropriate for callers that expect the underlying filesystem not
4185 * to be NFS exported.
4186 */
4187int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4188{
4189 struct inode *inode = old_dentry->d_inode;
4190 unsigned max_links = dir->i_sb->s_max_links;
4191 int error;
4192
4193 if (!inode)
4194 return -ENOENT;
4195
4196 error = may_create(dir, new_dentry);
4197 if (error)
4198 return error;
4199
4200 if (dir->i_sb != inode->i_sb)
4201 return -EXDEV;
4202
4203 /*
4204 * A link to an append-only or immutable file cannot be created.
4205 */
4206 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4207 return -EPERM;
4208 /*
4209 * Updating the link count will likely cause i_uid and i_gid to
4210 * be writen back improperly if their true value is unknown to
4211 * the vfs.
4212 */
4213 if (HAS_UNMAPPED_ID(inode))
4214 return -EPERM;
4215 if (!dir->i_op->link)
4216 return -EPERM;
4217 if (S_ISDIR(inode->i_mode))
4218 return -EPERM;
4219
4220 error = security_inode_link(old_dentry, dir, new_dentry);
4221 if (error)
4222 return error;
4223
4224 inode_lock(inode);
4225 /* Make sure we don't allow creating hardlink to an unlinked file */
4226 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4227 error = -ENOENT;
4228 else if (max_links && inode->i_nlink >= max_links)
4229 error = -EMLINK;
4230 else {
4231 error = try_break_deleg(inode, delegated_inode);
4232 if (!error)
4233 error = dir->i_op->link(old_dentry, dir, new_dentry);
4234 }
4235
4236 if (!error && (inode->i_state & I_LINKABLE)) {
4237 spin_lock(&inode->i_lock);
4238 inode->i_state &= ~I_LINKABLE;
4239 spin_unlock(&inode->i_lock);
4240 }
4241 inode_unlock(inode);
4242 if (!error)
4243 fsnotify_link(dir, inode, new_dentry);
4244 return error;
4245}
4246EXPORT_SYMBOL(vfs_link);
4247
4248/*
4249 * Hardlinks are often used in delicate situations. We avoid
4250 * security-related surprises by not following symlinks on the
4251 * newname. --KAB
4252 *
4253 * We don't follow them on the oldname either to be compatible
4254 * with linux 2.0, and to avoid hard-linking to directories
4255 * and other special files. --ADM
4256 */
4257int do_linkat(int olddfd, const char __user *oldname, int newdfd,
4258 const char __user *newname, int flags)
4259{
4260 struct dentry *new_dentry;
4261 struct path old_path, new_path;
4262 struct inode *delegated_inode = NULL;
4263 int how = 0;
4264 int error;
4265
4266 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4267 return -EINVAL;
4268 /*
4269 * To use null names we require CAP_DAC_READ_SEARCH
4270 * This ensures that not everyone will be able to create
4271 * handlink using the passed filedescriptor.
4272 */
4273 if (flags & AT_EMPTY_PATH) {
4274 if (!capable(CAP_DAC_READ_SEARCH))
4275 return -ENOENT;
4276 how = LOOKUP_EMPTY;
4277 }
4278
4279 if (flags & AT_SYMLINK_FOLLOW)
4280 how |= LOOKUP_FOLLOW;
4281retry:
4282 error = user_path_at(olddfd, oldname, how, &old_path);
4283 if (error)
4284 return error;
4285
4286 new_dentry = user_path_create(newdfd, newname, &new_path,
4287 (how & LOOKUP_REVAL));
4288 error = PTR_ERR(new_dentry);
4289 if (IS_ERR(new_dentry))
4290 goto out;
4291
4292 error = -EXDEV;
4293 if (old_path.mnt != new_path.mnt)
4294 goto out_dput;
4295 error = may_linkat(&old_path);
4296 if (unlikely(error))
4297 goto out_dput;
4298 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4299 if (error)
4300 goto out_dput;
4301 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4302out_dput:
4303 done_path_create(&new_path, new_dentry);
4304 if (delegated_inode) {
4305 error = break_deleg_wait(&delegated_inode);
4306 if (!error) {
4307 path_put(&old_path);
4308 goto retry;
4309 }
4310 }
4311 if (retry_estale(error, how)) {
4312 path_put(&old_path);
4313 how |= LOOKUP_REVAL;
4314 goto retry;
4315 }
4316out:
4317 path_put(&old_path);
4318
4319 return error;
4320}
4321
4322SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4323 int, newdfd, const char __user *, newname, int, flags)
4324{
4325 return do_linkat(olddfd, oldname, newdfd, newname, flags);
4326}
4327
4328SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4329{
4330 return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4331}
4332
4333/**
4334 * vfs_rename - rename a filesystem object
4335 * @old_dir: parent of source
4336 * @old_dentry: source
4337 * @new_dir: parent of destination
4338 * @new_dentry: destination
4339 * @delegated_inode: returns an inode needing a delegation break
4340 * @flags: rename flags
4341 *
4342 * The caller must hold multiple mutexes--see lock_rename()).
4343 *
4344 * If vfs_rename discovers a delegation in need of breaking at either
4345 * the source or destination, it will return -EWOULDBLOCK and return a
4346 * reference to the inode in delegated_inode. The caller should then
4347 * break the delegation and retry. Because breaking a delegation may
4348 * take a long time, the caller should drop all locks before doing
4349 * so.
4350 *
4351 * Alternatively, a caller may pass NULL for delegated_inode. This may
4352 * be appropriate for callers that expect the underlying filesystem not
4353 * to be NFS exported.
4354 *
4355 * The worst of all namespace operations - renaming directory. "Perverted"
4356 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4357 * Problems:
4358 *
4359 * a) we can get into loop creation.
4360 * b) race potential - two innocent renames can create a loop together.
4361 * That's where 4.4 screws up. Current fix: serialization on
4362 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4363 * story.
4364 * c) we have to lock _four_ objects - parents and victim (if it exists),
4365 * and source (if it is not a directory).
4366 * And that - after we got ->i_mutex on parents (until then we don't know
4367 * whether the target exists). Solution: try to be smart with locking
4368 * order for inodes. We rely on the fact that tree topology may change
4369 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4370 * move will be locked. Thus we can rank directories by the tree
4371 * (ancestors first) and rank all non-directories after them.
4372 * That works since everybody except rename does "lock parent, lookup,
4373 * lock child" and rename is under ->s_vfs_rename_mutex.
4374 * HOWEVER, it relies on the assumption that any object with ->lookup()
4375 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4376 * we'd better make sure that there's no link(2) for them.
4377 * d) conversion from fhandle to dentry may come in the wrong moment - when
4378 * we are removing the target. Solution: we will have to grab ->i_mutex
4379 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4380 * ->i_mutex on parents, which works but leads to some truly excessive
4381 * locking].
4382 */
4383int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4384 struct inode *new_dir, struct dentry *new_dentry,
4385 struct inode **delegated_inode, unsigned int flags)
4386{
4387 int error;
4388 bool is_dir = d_is_dir(old_dentry);
4389 struct inode *source = old_dentry->d_inode;
4390 struct inode *target = new_dentry->d_inode;
4391 bool new_is_dir = false;
4392 unsigned max_links = new_dir->i_sb->s_max_links;
4393 struct name_snapshot old_name;
4394
4395 if (source == target)
4396 return 0;
4397
4398 error = may_delete(old_dir, old_dentry, is_dir);
4399 if (error)
4400 return error;
4401
4402 if (!target) {
4403 error = may_create(new_dir, new_dentry);
4404 } else {
4405 new_is_dir = d_is_dir(new_dentry);
4406
4407 if (!(flags & RENAME_EXCHANGE))
4408 error = may_delete(new_dir, new_dentry, is_dir);
4409 else
4410 error = may_delete(new_dir, new_dentry, new_is_dir);
4411 }
4412 if (error)
4413 return error;
4414
4415 if (!old_dir->i_op->rename)
4416 return -EPERM;
4417
4418 /*
4419 * If we are going to change the parent - check write permissions,
4420 * we'll need to flip '..'.
4421 */
4422 if (new_dir != old_dir) {
4423 if (is_dir) {
4424 error = inode_permission(source, MAY_WRITE);
4425 if (error)
4426 return error;
4427 }
4428 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4429 error = inode_permission(target, MAY_WRITE);
4430 if (error)
4431 return error;
4432 }
4433 }
4434
4435 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4436 flags);
4437 if (error)
4438 return error;
4439
4440 take_dentry_name_snapshot(&old_name, old_dentry);
4441 dget(new_dentry);
4442 if (!is_dir || (flags & RENAME_EXCHANGE))
4443 lock_two_nondirectories(source, target);
4444 else if (target)
4445 inode_lock(target);
4446
4447 error = -EBUSY;
4448 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4449 goto out;
4450
4451 if (max_links && new_dir != old_dir) {
4452 error = -EMLINK;
4453 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4454 goto out;
4455 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4456 old_dir->i_nlink >= max_links)
4457 goto out;
4458 }
4459 if (!is_dir) {
4460 error = try_break_deleg(source, delegated_inode);
4461 if (error)
4462 goto out;
4463 }
4464 if (target && !new_is_dir) {
4465 error = try_break_deleg(target, delegated_inode);
4466 if (error)
4467 goto out;
4468 }
4469 error = old_dir->i_op->rename(old_dir, old_dentry,
4470 new_dir, new_dentry, flags);
4471 if (error)
4472 goto out;
4473
4474 if (!(flags & RENAME_EXCHANGE) && target) {
4475 if (is_dir) {
4476 shrink_dcache_parent(new_dentry);
4477 target->i_flags |= S_DEAD;
4478 }
4479 dont_mount(new_dentry);
4480 detach_mounts(new_dentry);
4481 }
4482 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4483 if (!(flags & RENAME_EXCHANGE))
4484 d_move(old_dentry, new_dentry);
4485 else
4486 d_exchange(old_dentry, new_dentry);
4487 }
4488out:
4489 if (!is_dir || (flags & RENAME_EXCHANGE))
4490 unlock_two_nondirectories(source, target);
4491 else if (target)
4492 inode_unlock(target);
4493 dput(new_dentry);
4494 if (!error) {
4495 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4496 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4497 if (flags & RENAME_EXCHANGE) {
4498 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4499 new_is_dir, NULL, new_dentry);
4500 }
4501 }
4502 release_dentry_name_snapshot(&old_name);
4503
4504 return error;
4505}
4506EXPORT_SYMBOL(vfs_rename);
4507
4508static int do_renameat2(int olddfd, const char __user *oldname, int newdfd,
4509 const char __user *newname, unsigned int flags)
4510{
4511 struct dentry *old_dentry, *new_dentry;
4512 struct dentry *trap;
4513 struct path old_path, new_path;
4514 struct qstr old_last, new_last;
4515 int old_type, new_type;
4516 struct inode *delegated_inode = NULL;
4517 struct filename *from;
4518 struct filename *to;
4519 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4520 bool should_retry = false;
4521 int error;
4522
4523 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4524 return -EINVAL;
4525
4526 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4527 (flags & RENAME_EXCHANGE))
4528 return -EINVAL;
4529
4530 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4531 return -EPERM;
4532
4533 if (flags & RENAME_EXCHANGE)
4534 target_flags = 0;
4535
4536retry:
4537 from = filename_parentat(olddfd, getname(oldname), lookup_flags,
4538 &old_path, &old_last, &old_type);
4539 if (IS_ERR(from)) {
4540 error = PTR_ERR(from);
4541 goto exit;
4542 }
4543
4544 to = filename_parentat(newdfd, getname(newname), lookup_flags,
4545 &new_path, &new_last, &new_type);
4546 if (IS_ERR(to)) {
4547 error = PTR_ERR(to);
4548 goto exit1;
4549 }
4550
4551 error = -EXDEV;
4552 if (old_path.mnt != new_path.mnt)
4553 goto exit2;
4554
4555 error = -EBUSY;
4556 if (old_type != LAST_NORM)
4557 goto exit2;
4558
4559 if (flags & RENAME_NOREPLACE)
4560 error = -EEXIST;
4561 if (new_type != LAST_NORM)
4562 goto exit2;
4563
4564 error = mnt_want_write(old_path.mnt);
4565 if (error)
4566 goto exit2;
4567
4568retry_deleg:
4569 trap = lock_rename(new_path.dentry, old_path.dentry);
4570
4571 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4572 error = PTR_ERR(old_dentry);
4573 if (IS_ERR(old_dentry))
4574 goto exit3;
4575 /* source must exist */
4576 error = -ENOENT;
4577 if (d_is_negative(old_dentry))
4578 goto exit4;
4579 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4580 error = PTR_ERR(new_dentry);
4581 if (IS_ERR(new_dentry))
4582 goto exit4;
4583 error = -EEXIST;
4584 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4585 goto exit5;
4586 if (flags & RENAME_EXCHANGE) {
4587 error = -ENOENT;
4588 if (d_is_negative(new_dentry))
4589 goto exit5;
4590
4591 if (!d_is_dir(new_dentry)) {
4592 error = -ENOTDIR;
4593 if (new_last.name[new_last.len])
4594 goto exit5;
4595 }
4596 }
4597 /* unless the source is a directory trailing slashes give -ENOTDIR */
4598 if (!d_is_dir(old_dentry)) {
4599 error = -ENOTDIR;
4600 if (old_last.name[old_last.len])
4601 goto exit5;
4602 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4603 goto exit5;
4604 }
4605 /* source should not be ancestor of target */
4606 error = -EINVAL;
4607 if (old_dentry == trap)
4608 goto exit5;
4609 /* target should not be an ancestor of source */
4610 if (!(flags & RENAME_EXCHANGE))
4611 error = -ENOTEMPTY;
4612 if (new_dentry == trap)
4613 goto exit5;
4614
4615 error = security_path_rename(&old_path, old_dentry,
4616 &new_path, new_dentry, flags);
4617 if (error)
4618 goto exit5;
4619 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4620 new_path.dentry->d_inode, new_dentry,
4621 &delegated_inode, flags);
4622exit5:
4623 dput(new_dentry);
4624exit4:
4625 dput(old_dentry);
4626exit3:
4627 unlock_rename(new_path.dentry, old_path.dentry);
4628 if (delegated_inode) {
4629 error = break_deleg_wait(&delegated_inode);
4630 if (!error)
4631 goto retry_deleg;
4632 }
4633 mnt_drop_write(old_path.mnt);
4634exit2:
4635 if (retry_estale(error, lookup_flags))
4636 should_retry = true;
4637 path_put(&new_path);
4638 putname(to);
4639exit1:
4640 path_put(&old_path);
4641 putname(from);
4642 if (should_retry) {
4643 should_retry = false;
4644 lookup_flags |= LOOKUP_REVAL;
4645 goto retry;
4646 }
4647exit:
4648 return error;
4649}
4650
4651SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4652 int, newdfd, const char __user *, newname, unsigned int, flags)
4653{
4654 return do_renameat2(olddfd, oldname, newdfd, newname, flags);
4655}
4656
4657SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4658 int, newdfd, const char __user *, newname)
4659{
4660 return do_renameat2(olddfd, oldname, newdfd, newname, 0);
4661}
4662
4663SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4664{
4665 return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4666}
4667
4668int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4669{
4670 int error = may_create(dir, dentry);
4671 if (error)
4672 return error;
4673
4674 if (!dir->i_op->mknod)
4675 return -EPERM;
4676
4677 return dir->i_op->mknod(dir, dentry,
4678 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4679}
4680EXPORT_SYMBOL(vfs_whiteout);
4681
4682int readlink_copy(char __user *buffer, int buflen, const char *link)
4683{
4684 int len = PTR_ERR(link);
4685 if (IS_ERR(link))
4686 goto out;
4687
4688 len = strlen(link);
4689 if (len > (unsigned) buflen)
4690 len = buflen;
4691 if (copy_to_user(buffer, link, len))
4692 len = -EFAULT;
4693out:
4694 return len;
4695}
4696
4697/**
4698 * vfs_readlink - copy symlink body into userspace buffer
4699 * @dentry: dentry on which to get symbolic link
4700 * @buffer: user memory pointer
4701 * @buflen: size of buffer
4702 *
4703 * Does not touch atime. That's up to the caller if necessary
4704 *
4705 * Does not call security hook.
4706 */
4707int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4708{
4709 struct inode *inode = d_inode(dentry);
4710 DEFINE_DELAYED_CALL(done);
4711 const char *link;
4712 int res;
4713
4714 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4715 if (unlikely(inode->i_op->readlink))
4716 return inode->i_op->readlink(dentry, buffer, buflen);
4717
4718 if (!d_is_symlink(dentry))
4719 return -EINVAL;
4720
4721 spin_lock(&inode->i_lock);
4722 inode->i_opflags |= IOP_DEFAULT_READLINK;
4723 spin_unlock(&inode->i_lock);
4724 }
4725
4726 link = READ_ONCE(inode->i_link);
4727 if (!link) {
4728 link = inode->i_op->get_link(dentry, inode, &done);
4729 if (IS_ERR(link))
4730 return PTR_ERR(link);
4731 }
4732 res = readlink_copy(buffer, buflen, link);
4733 do_delayed_call(&done);
4734 return res;
4735}
4736EXPORT_SYMBOL(vfs_readlink);
4737
4738/**
4739 * vfs_get_link - get symlink body
4740 * @dentry: dentry on which to get symbolic link
4741 * @done: caller needs to free returned data with this
4742 *
4743 * Calls security hook and i_op->get_link() on the supplied inode.
4744 *
4745 * It does not touch atime. That's up to the caller if necessary.
4746 *
4747 * Does not work on "special" symlinks like /proc/$$/fd/N
4748 */
4749const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4750{
4751 const char *res = ERR_PTR(-EINVAL);
4752 struct inode *inode = d_inode(dentry);
4753
4754 if (d_is_symlink(dentry)) {
4755 res = ERR_PTR(security_inode_readlink(dentry));
4756 if (!res)
4757 res = inode->i_op->get_link(dentry, inode, done);
4758 }
4759 return res;
4760}
4761EXPORT_SYMBOL(vfs_get_link);
4762
4763/* get the link contents into pagecache */
4764const char *page_get_link(struct dentry *dentry, struct inode *inode,
4765 struct delayed_call *callback)
4766{
4767 char *kaddr;
4768 struct page *page;
4769 struct address_space *mapping = inode->i_mapping;
4770
4771 if (!dentry) {
4772 page = find_get_page(mapping, 0);
4773 if (!page)
4774 return ERR_PTR(-ECHILD);
4775 if (!PageUptodate(page)) {
4776 put_page(page);
4777 return ERR_PTR(-ECHILD);
4778 }
4779 } else {
4780 page = read_mapping_page(mapping, 0, NULL);
4781 if (IS_ERR(page))
4782 return (char*)page;
4783 }
4784 set_delayed_call(callback, page_put_link, page);
4785 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4786 kaddr = page_address(page);
4787 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4788 return kaddr;
4789}
4790
4791EXPORT_SYMBOL(page_get_link);
4792
4793void page_put_link(void *arg)
4794{
4795 put_page(arg);
4796}
4797EXPORT_SYMBOL(page_put_link);
4798
4799int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4800{
4801 DEFINE_DELAYED_CALL(done);
4802 int res = readlink_copy(buffer, buflen,
4803 page_get_link(dentry, d_inode(dentry),
4804 &done));
4805 do_delayed_call(&done);
4806 return res;
4807}
4808EXPORT_SYMBOL(page_readlink);
4809
4810/*
4811 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4812 */
4813int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4814{
4815 struct address_space *mapping = inode->i_mapping;
4816 struct page *page;
4817 void *fsdata;
4818 int err;
4819 unsigned int flags = 0;
4820 if (nofs)
4821 flags |= AOP_FLAG_NOFS;
4822
4823retry:
4824 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4825 flags, &page, &fsdata);
4826 if (err)
4827 goto fail;
4828
4829 memcpy(page_address(page), symname, len-1);
4830
4831 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4832 page, fsdata);
4833 if (err < 0)
4834 goto fail;
4835 if (err < len-1)
4836 goto retry;
4837
4838 mark_inode_dirty(inode);
4839 return 0;
4840fail:
4841 return err;
4842}
4843EXPORT_SYMBOL(__page_symlink);
4844
4845int page_symlink(struct inode *inode, const char *symname, int len)
4846{
4847 return __page_symlink(inode, symname, len,
4848 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4849}
4850EXPORT_SYMBOL(page_symlink);
4851
4852const struct inode_operations page_symlink_inode_operations = {
4853 .get_link = page_get_link,
4854};
4855EXPORT_SYMBOL(page_symlink_inode_operations);