1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/namei.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8/*
9 * Some corrections by tytso.
10 */
11
12/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
13 * lookup logic.
14 */
15/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
16 */
17
18#include <linux/init.h>
19#include <linux/export.h>
20#include <linux/slab.h>
21#include <linux/wordpart.h>
22#include <linux/fs.h>
23#include <linux/filelock.h>
24#include <linux/namei.h>
25#include <linux/pagemap.h>
26#include <linux/sched/mm.h>
27#include <linux/fsnotify.h>
28#include <linux/personality.h>
29#include <linux/security.h>
30#include <linux/syscalls.h>
31#include <linux/mount.h>
32#include <linux/audit.h>
33#include <linux/capability.h>
34#include <linux/file.h>
35#include <linux/fcntl.h>
36#include <linux/device_cgroup.h>
37#include <linux/fs_struct.h>
38#include <linux/posix_acl.h>
39#include <linux/hash.h>
40#include <linux/bitops.h>
41#include <linux/init_task.h>
42#include <linux/uaccess.h>
43
44#include "internal.h"
45#include "mount.h"
46
47/* [Feb-1997 T. Schoebel-Theuer]
48 * Fundamental changes in the pathname lookup mechanisms (namei)
49 * were necessary because of omirr. The reason is that omirr needs
50 * to know the _real_ pathname, not the user-supplied one, in case
51 * of symlinks (and also when transname replacements occur).
52 *
53 * The new code replaces the old recursive symlink resolution with
54 * an iterative one (in case of non-nested symlink chains). It does
55 * this with calls to <fs>_follow_link().
56 * As a side effect, dir_namei(), _namei() and follow_link() are now
57 * replaced with a single function lookup_dentry() that can handle all
58 * the special cases of the former code.
59 *
60 * With the new dcache, the pathname is stored at each inode, at least as
61 * long as the refcount of the inode is positive. As a side effect, the
62 * size of the dcache depends on the inode cache and thus is dynamic.
63 *
64 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
65 * resolution to correspond with current state of the code.
66 *
67 * Note that the symlink resolution is not *completely* iterative.
68 * There is still a significant amount of tail- and mid- recursion in
69 * the algorithm. Also, note that <fs>_readlink() is not used in
70 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
71 * may return different results than <fs>_follow_link(). Many virtual
72 * filesystems (including /proc) exhibit this behavior.
73 */
74
75/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
76 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
77 * and the name already exists in form of a symlink, try to create the new
78 * name indicated by the symlink. The old code always complained that the
79 * name already exists, due to not following the symlink even if its target
80 * is nonexistent. The new semantics affects also mknod() and link() when
81 * the name is a symlink pointing to a non-existent name.
82 *
83 * I don't know which semantics is the right one, since I have no access
84 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
85 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
86 * "old" one. Personally, I think the new semantics is much more logical.
87 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
88 * file does succeed in both HP-UX and SunOs, but not in Solaris
89 * and in the old Linux semantics.
90 */
91
92/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
93 * semantics. See the comments in "open_namei" and "do_link" below.
94 *
95 * [10-Sep-98 Alan Modra] Another symlink change.
96 */
97
98/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
99 * inside the path - always follow.
100 * in the last component in creation/removal/renaming - never follow.
101 * if LOOKUP_FOLLOW passed - follow.
102 * if the pathname has trailing slashes - follow.
103 * otherwise - don't follow.
104 * (applied in that order).
105 *
106 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
107 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
108 * During the 2.4 we need to fix the userland stuff depending on it -
109 * hopefully we will be able to get rid of that wart in 2.5. So far only
110 * XEmacs seems to be relying on it...
111 */
112/*
113 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
114 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
115 * any extra contention...
116 */
117
118/* In order to reduce some races, while at the same time doing additional
119 * checking and hopefully speeding things up, we copy filenames to the
120 * kernel data space before using them..
121 *
122 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
123 * PATH_MAX includes the nul terminator --RR.
124 */
125
126#define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
127
128static inline void initname(struct filename *name, const char __user *uptr)
129{
130 name->uptr = uptr;
131 name->aname = NULL;
132 atomic_set(v: &name->refcnt, i: 1);
133}
134
135struct filename *
136getname_flags(const char __user *filename, int flags)
137{
138 struct filename *result;
139 char *kname;
140 int len;
141
142 result = audit_reusename(name: filename);
143 if (result)
144 return result;
145
146 result = __getname();
147 if (unlikely(!result))
148 return ERR_PTR(error: -ENOMEM);
149
150 /*
151 * First, try to embed the struct filename inside the names_cache
152 * allocation
153 */
154 kname = (char *)result->iname;
155 result->name = kname;
156
157 len = strncpy_from_user(dst: kname, src: filename, EMBEDDED_NAME_MAX);
158 /*
159 * Handle both empty path and copy failure in one go.
160 */
161 if (unlikely(len <= 0)) {
162 if (unlikely(len < 0)) {
163 __putname(result);
164 return ERR_PTR(error: len);
165 }
166
167 /* The empty path is special. */
168 if (!(flags & LOOKUP_EMPTY)) {
169 __putname(result);
170 return ERR_PTR(error: -ENOENT);
171 }
172 }
173
174 /*
175 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
176 * separate struct filename so we can dedicate the entire
177 * names_cache allocation for the pathname, and re-do the copy from
178 * userland.
179 */
180 if (unlikely(len == EMBEDDED_NAME_MAX)) {
181 const size_t size = offsetof(struct filename, iname[1]);
182 kname = (char *)result;
183
184 /*
185 * size is chosen that way we to guarantee that
186 * result->iname[0] is within the same object and that
187 * kname can't be equal to result->iname, no matter what.
188 */
189 result = kzalloc(size, GFP_KERNEL);
190 if (unlikely(!result)) {
191 __putname(kname);
192 return ERR_PTR(error: -ENOMEM);
193 }
194 result->name = kname;
195 len = strncpy_from_user(dst: kname, src: filename, PATH_MAX);
196 if (unlikely(len < 0)) {
197 __putname(kname);
198 kfree(objp: result);
199 return ERR_PTR(error: len);
200 }
201 /* The empty path is special. */
202 if (unlikely(!len) && !(flags & LOOKUP_EMPTY)) {
203 __putname(kname);
204 kfree(objp: result);
205 return ERR_PTR(error: -ENOENT);
206 }
207 if (unlikely(len == PATH_MAX)) {
208 __putname(kname);
209 kfree(objp: result);
210 return ERR_PTR(error: -ENAMETOOLONG);
211 }
212 }
213 initname(name: result, uptr: filename);
214 audit_getname(name: result);
215 return result;
216}
217
218struct filename *getname_uflags(const char __user *filename, int uflags)
219{
220 int flags = (uflags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
221
222 return getname_flags(filename, flags);
223}
224
225struct filename *__getname_maybe_null(const char __user *pathname)
226{
227 struct filename *name;
228 char c;
229
230 /* try to save on allocations; loss on um, though */
231 if (get_user(c, pathname))
232 return ERR_PTR(error: -EFAULT);
233 if (!c)
234 return NULL;
235
236 name = getname_flags(filename: pathname, LOOKUP_EMPTY);
237 if (!IS_ERR(ptr: name) && !(name->name[0])) {
238 putname(name);
239 name = NULL;
240 }
241 return name;
242}
243
244struct filename *getname_kernel(const char * filename)
245{
246 struct filename *result;
247 int len = strlen(filename) + 1;
248
249 result = __getname();
250 if (unlikely(!result))
251 return ERR_PTR(error: -ENOMEM);
252
253 if (len <= EMBEDDED_NAME_MAX) {
254 result->name = (char *)result->iname;
255 } else if (len <= PATH_MAX) {
256 const size_t size = offsetof(struct filename, iname[1]);
257 struct filename *tmp;
258
259 tmp = kmalloc(size, GFP_KERNEL);
260 if (unlikely(!tmp)) {
261 __putname(result);
262 return ERR_PTR(error: -ENOMEM);
263 }
264 tmp->name = (char *)result;
265 result = tmp;
266 } else {
267 __putname(result);
268 return ERR_PTR(error: -ENAMETOOLONG);
269 }
270 memcpy((char *)result->name, filename, len);
271 initname(name: result, NULL);
272 audit_getname(name: result);
273 return result;
274}
275EXPORT_SYMBOL(getname_kernel);
276
277void putname(struct filename *name)
278{
279 int refcnt;
280
281 if (IS_ERR_OR_NULL(ptr: name))
282 return;
283
284 refcnt = atomic_read(v: &name->refcnt);
285 if (refcnt != 1) {
286 if (WARN_ON_ONCE(!refcnt))
287 return;
288
289 if (!atomic_dec_and_test(v: &name->refcnt))
290 return;
291 }
292
293 if (name->name != name->iname) {
294 __putname(name->name);
295 kfree(objp: name);
296 } else
297 __putname(name);
298}
299EXPORT_SYMBOL(putname);
300
301/**
302 * check_acl - perform ACL permission checking
303 * @idmap: idmap of the mount the inode was found from
304 * @inode: inode to check permissions on
305 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
306 *
307 * This function performs the ACL permission checking. Since this function
308 * retrieve POSIX acls it needs to know whether it is called from a blocking or
309 * non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
310 *
311 * If the inode has been found through an idmapped mount the idmap of
312 * the vfsmount must be passed through @idmap. This function will then take
313 * care to map the inode according to @idmap before checking permissions.
314 * On non-idmapped mounts or if permission checking is to be performed on the
315 * raw inode simply pass @nop_mnt_idmap.
316 */
317static int check_acl(struct mnt_idmap *idmap,
318 struct inode *inode, int mask)
319{
320#ifdef CONFIG_FS_POSIX_ACL
321 struct posix_acl *acl;
322
323 if (mask & MAY_NOT_BLOCK) {
324 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
325 if (!acl)
326 return -EAGAIN;
327 /* no ->get_inode_acl() calls in RCU mode... */
328 if (is_uncached_acl(acl))
329 return -ECHILD;
330 return posix_acl_permission(idmap, inode, acl, mask);
331 }
332
333 acl = get_inode_acl(inode, ACL_TYPE_ACCESS);
334 if (IS_ERR(ptr: acl))
335 return PTR_ERR(ptr: acl);
336 if (acl) {
337 int error = posix_acl_permission(idmap, inode, acl, mask);
338 posix_acl_release(acl);
339 return error;
340 }
341#endif
342
343 return -EAGAIN;
344}
345
346/*
347 * Very quick optimistic "we know we have no ACL's" check.
348 *
349 * Note that this is purely for ACL_TYPE_ACCESS, and purely
350 * for the "we have cached that there are no ACLs" case.
351 *
352 * If this returns true, we know there are no ACLs. But if
353 * it returns false, we might still not have ACLs (it could
354 * be the is_uncached_acl() case).
355 */
356static inline bool no_acl_inode(struct inode *inode)
357{
358#ifdef CONFIG_FS_POSIX_ACL
359 return likely(!READ_ONCE(inode->i_acl));
360#else
361 return true;
362#endif
363}
364
365/**
366 * acl_permission_check - perform basic UNIX permission checking
367 * @idmap: idmap of the mount the inode was found from
368 * @inode: inode to check permissions on
369 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
370 *
371 * This function performs the basic UNIX permission checking. Since this
372 * function may retrieve POSIX acls it needs to know whether it is called from a
373 * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
374 *
375 * If the inode has been found through an idmapped mount the idmap of
376 * the vfsmount must be passed through @idmap. This function will then take
377 * care to map the inode according to @idmap before checking permissions.
378 * On non-idmapped mounts or if permission checking is to be performed on the
379 * raw inode simply pass @nop_mnt_idmap.
380 */
381static int acl_permission_check(struct mnt_idmap *idmap,
382 struct inode *inode, int mask)
383{
384 unsigned int mode = inode->i_mode;
385 vfsuid_t vfsuid;
386
387 /*
388 * Common cheap case: everybody has the requested
389 * rights, and there are no ACLs to check. No need
390 * to do any owner/group checks in that case.
391 *
392 * - 'mask&7' is the requested permission bit set
393 * - multiplying by 0111 spreads them out to all of ugo
394 * - '& ~mode' looks for missing inode permission bits
395 * - the '!' is for "no missing permissions"
396 *
397 * After that, we just need to check that there are no
398 * ACL's on the inode - do the 'IS_POSIXACL()' check last
399 * because it will dereference the ->i_sb pointer and we
400 * want to avoid that if at all possible.
401 */
402 if (!((mask & 7) * 0111 & ~mode)) {
403 if (no_acl_inode(inode))
404 return 0;
405 if (!IS_POSIXACL(inode))
406 return 0;
407 }
408
409 /* Are we the owner? If so, ACL's don't matter */
410 vfsuid = i_uid_into_vfsuid(idmap, inode);
411 if (likely(vfsuid_eq_kuid(vfsuid, current_fsuid()))) {
412 mask &= 7;
413 mode >>= 6;
414 return (mask & ~mode) ? -EACCES : 0;
415 }
416
417 /* Do we have ACL's? */
418 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
419 int error = check_acl(idmap, inode, mask);
420 if (error != -EAGAIN)
421 return error;
422 }
423
424 /* Only RWX matters for group/other mode bits */
425 mask &= 7;
426
427 /*
428 * Are the group permissions different from
429 * the other permissions in the bits we care
430 * about? Need to check group ownership if so.
431 */
432 if (mask & (mode ^ (mode >> 3))) {
433 vfsgid_t vfsgid = i_gid_into_vfsgid(idmap, inode);
434 if (vfsgid_in_group_p(vfsgid))
435 mode >>= 3;
436 }
437
438 /* Bits in 'mode' clear that we require? */
439 return (mask & ~mode) ? -EACCES : 0;
440}
441
442/**
443 * generic_permission - check for access rights on a Posix-like filesystem
444 * @idmap: idmap of the mount the inode was found from
445 * @inode: inode to check access rights for
446 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC,
447 * %MAY_NOT_BLOCK ...)
448 *
449 * Used to check for read/write/execute permissions on a file.
450 * We use "fsuid" for this, letting us set arbitrary permissions
451 * for filesystem access without changing the "normal" uids which
452 * are used for other things.
453 *
454 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
455 * request cannot be satisfied (eg. requires blocking or too much complexity).
456 * It would then be called again in ref-walk mode.
457 *
458 * If the inode has been found through an idmapped mount the idmap of
459 * the vfsmount must be passed through @idmap. This function will then take
460 * care to map the inode according to @idmap before checking permissions.
461 * On non-idmapped mounts or if permission checking is to be performed on the
462 * raw inode simply pass @nop_mnt_idmap.
463 */
464int generic_permission(struct mnt_idmap *idmap, struct inode *inode,
465 int mask)
466{
467 int ret;
468
469 /*
470 * Do the basic permission checks.
471 */
472 ret = acl_permission_check(idmap, inode, mask);
473 if (ret != -EACCES)
474 return ret;
475
476 if (S_ISDIR(inode->i_mode)) {
477 /* DACs are overridable for directories */
478 if (!(mask & MAY_WRITE))
479 if (capable_wrt_inode_uidgid(idmap, inode,
480 CAP_DAC_READ_SEARCH))
481 return 0;
482 if (capable_wrt_inode_uidgid(idmap, inode,
483 CAP_DAC_OVERRIDE))
484 return 0;
485 return -EACCES;
486 }
487
488 /*
489 * Searching includes executable on directories, else just read.
490 */
491 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
492 if (mask == MAY_READ)
493 if (capable_wrt_inode_uidgid(idmap, inode,
494 CAP_DAC_READ_SEARCH))
495 return 0;
496 /*
497 * Read/write DACs are always overridable.
498 * Executable DACs are overridable when there is
499 * at least one exec bit set.
500 */
501 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
502 if (capable_wrt_inode_uidgid(idmap, inode,
503 CAP_DAC_OVERRIDE))
504 return 0;
505
506 return -EACCES;
507}
508EXPORT_SYMBOL(generic_permission);
509
510/**
511 * do_inode_permission - UNIX permission checking
512 * @idmap: idmap of the mount the inode was found from
513 * @inode: inode to check permissions on
514 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
515 *
516 * We _really_ want to just do "generic_permission()" without
517 * even looking at the inode->i_op values. So we keep a cache
518 * flag in inode->i_opflags, that says "this has not special
519 * permission function, use the fast case".
520 */
521static inline int do_inode_permission(struct mnt_idmap *idmap,
522 struct inode *inode, int mask)
523{
524 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
525 if (likely(inode->i_op->permission))
526 return inode->i_op->permission(idmap, inode, mask);
527
528 /* This gets set once for the inode lifetime */
529 spin_lock(lock: &inode->i_lock);
530 inode->i_opflags |= IOP_FASTPERM;
531 spin_unlock(lock: &inode->i_lock);
532 }
533 return generic_permission(idmap, inode, mask);
534}
535
536/**
537 * sb_permission - Check superblock-level permissions
538 * @sb: Superblock of inode to check permission on
539 * @inode: Inode to check permission on
540 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
541 *
542 * Separate out file-system wide checks from inode-specific permission checks.
543 */
544static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
545{
546 if (unlikely(mask & MAY_WRITE)) {
547 umode_t mode = inode->i_mode;
548
549 /* Nobody gets write access to a read-only fs. */
550 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
551 return -EROFS;
552 }
553 return 0;
554}
555
556/**
557 * inode_permission - Check for access rights to a given inode
558 * @idmap: idmap of the mount the inode was found from
559 * @inode: Inode to check permission on
560 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
561 *
562 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
563 * this, letting us set arbitrary permissions for filesystem access without
564 * changing the "normal" UIDs which are used for other things.
565 *
566 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
567 */
568int inode_permission(struct mnt_idmap *idmap,
569 struct inode *inode, int mask)
570{
571 int retval;
572
573 retval = sb_permission(sb: inode->i_sb, inode, mask);
574 if (unlikely(retval))
575 return retval;
576
577 if (unlikely(mask & MAY_WRITE)) {
578 /*
579 * Nobody gets write access to an immutable file.
580 */
581 if (unlikely(IS_IMMUTABLE(inode)))
582 return -EPERM;
583
584 /*
585 * Updating mtime will likely cause i_uid and i_gid to be
586 * written back improperly if their true value is unknown
587 * to the vfs.
588 */
589 if (unlikely(HAS_UNMAPPED_ID(idmap, inode)))
590 return -EACCES;
591 }
592
593 retval = do_inode_permission(idmap, inode, mask);
594 if (unlikely(retval))
595 return retval;
596
597 retval = devcgroup_inode_permission(inode, mask);
598 if (unlikely(retval))
599 return retval;
600
601 return security_inode_permission(inode, mask);
602}
603EXPORT_SYMBOL(inode_permission);
604
605/**
606 * path_get - get a reference to a path
607 * @path: path to get the reference to
608 *
609 * Given a path increment the reference count to the dentry and the vfsmount.
610 */
611void path_get(const struct path *path)
612{
613 mntget(mnt: path->mnt);
614 dget(dentry: path->dentry);
615}
616EXPORT_SYMBOL(path_get);
617
618/**
619 * path_put - put a reference to a path
620 * @path: path to put the reference to
621 *
622 * Given a path decrement the reference count to the dentry and the vfsmount.
623 */
624void path_put(const struct path *path)
625{
626 dput(path->dentry);
627 mntput(mnt: path->mnt);
628}
629EXPORT_SYMBOL(path_put);
630
631#define EMBEDDED_LEVELS 2
632struct nameidata {
633 struct path path;
634 struct qstr last;
635 struct path root;
636 struct inode *inode; /* path.dentry.d_inode */
637 unsigned int flags, state;
638 unsigned seq, next_seq, m_seq, r_seq;
639 int last_type;
640 unsigned depth;
641 int total_link_count;
642 struct saved {
643 struct path link;
644 struct delayed_call done;
645 const char *name;
646 unsigned seq;
647 } *stack, internal[EMBEDDED_LEVELS];
648 struct filename *name;
649 const char *pathname;
650 struct nameidata *saved;
651 unsigned root_seq;
652 int dfd;
653 vfsuid_t dir_vfsuid;
654 umode_t dir_mode;
655} __randomize_layout;
656
657#define ND_ROOT_PRESET 1
658#define ND_ROOT_GRABBED 2
659#define ND_JUMPED 4
660
661static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name)
662{
663 struct nameidata *old = current->nameidata;
664 p->stack = p->internal;
665 p->depth = 0;
666 p->dfd = dfd;
667 p->name = name;
668 p->pathname = likely(name) ? name->name : "";
669 p->path.mnt = NULL;
670 p->path.dentry = NULL;
671 p->total_link_count = old ? old->total_link_count : 0;
672 p->saved = old;
673 current->nameidata = p;
674}
675
676static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name,
677 const struct path *root)
678{
679 __set_nameidata(p, dfd, name);
680 p->state = 0;
681 if (unlikely(root)) {
682 p->state = ND_ROOT_PRESET;
683 p->root = *root;
684 }
685}
686
687static void restore_nameidata(void)
688{
689 struct nameidata *now = current->nameidata, *old = now->saved;
690
691 current->nameidata = old;
692 if (old)
693 old->total_link_count = now->total_link_count;
694 if (now->stack != now->internal)
695 kfree(objp: now->stack);
696}
697
698static bool nd_alloc_stack(struct nameidata *nd)
699{
700 struct saved *p;
701
702 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
703 nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL);
704 if (unlikely(!p))
705 return false;
706 memcpy(p, nd->internal, sizeof(nd->internal));
707 nd->stack = p;
708 return true;
709}
710
711/**
712 * path_connected - Verify that a dentry is below mnt.mnt_root
713 * @mnt: The mountpoint to check.
714 * @dentry: The dentry to check.
715 *
716 * Rename can sometimes move a file or directory outside of a bind
717 * mount, path_connected allows those cases to be detected.
718 */
719static bool path_connected(struct vfsmount *mnt, struct dentry *dentry)
720{
721 struct super_block *sb = mnt->mnt_sb;
722
723 /* Bind mounts can have disconnected paths */
724 if (mnt->mnt_root == sb->s_root)
725 return true;
726
727 return is_subdir(dentry, mnt->mnt_root);
728}
729
730static void drop_links(struct nameidata *nd)
731{
732 int i = nd->depth;
733 while (i--) {
734 struct saved *last = nd->stack + i;
735 do_delayed_call(call: &last->done);
736 clear_delayed_call(call: &last->done);
737 }
738}
739
740static void leave_rcu(struct nameidata *nd)
741{
742 nd->flags &= ~LOOKUP_RCU;
743 nd->seq = nd->next_seq = 0;
744 rcu_read_unlock();
745}
746
747static void terminate_walk(struct nameidata *nd)
748{
749 drop_links(nd);
750 if (!(nd->flags & LOOKUP_RCU)) {
751 int i;
752 path_put(&nd->path);
753 for (i = 0; i < nd->depth; i++)
754 path_put(&nd->stack[i].link);
755 if (nd->state & ND_ROOT_GRABBED) {
756 path_put(&nd->root);
757 nd->state &= ~ND_ROOT_GRABBED;
758 }
759 } else {
760 leave_rcu(nd);
761 }
762 nd->depth = 0;
763 nd->path.mnt = NULL;
764 nd->path.dentry = NULL;
765}
766
767/* path_put is needed afterwards regardless of success or failure */
768static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq)
769{
770 int res = __legitimize_mnt(path->mnt, mseq);
771 if (unlikely(res)) {
772 if (res > 0)
773 path->mnt = NULL;
774 path->dentry = NULL;
775 return false;
776 }
777 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
778 path->dentry = NULL;
779 return false;
780 }
781 return !read_seqcount_retry(&path->dentry->d_seq, seq);
782}
783
784static inline bool legitimize_path(struct nameidata *nd,
785 struct path *path, unsigned seq)
786{
787 return __legitimize_path(path, seq, mseq: nd->m_seq);
788}
789
790static bool legitimize_links(struct nameidata *nd)
791{
792 int i;
793 if (unlikely(nd->flags & LOOKUP_CACHED)) {
794 drop_links(nd);
795 nd->depth = 0;
796 return false;
797 }
798 for (i = 0; i < nd->depth; i++) {
799 struct saved *last = nd->stack + i;
800 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
801 drop_links(nd);
802 nd->depth = i + 1;
803 return false;
804 }
805 }
806 return true;
807}
808
809static bool legitimize_root(struct nameidata *nd)
810{
811 /* Nothing to do if nd->root is zero or is managed by the VFS user. */
812 if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET))
813 return true;
814 nd->state |= ND_ROOT_GRABBED;
815 return legitimize_path(nd, path: &nd->root, seq: nd->root_seq);
816}
817
818/*
819 * Path walking has 2 modes, rcu-walk and ref-walk (see
820 * Documentation/filesystems/path-lookup.txt). In situations when we can't
821 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
822 * normal reference counts on dentries and vfsmounts to transition to ref-walk
823 * mode. Refcounts are grabbed at the last known good point before rcu-walk
824 * got stuck, so ref-walk may continue from there. If this is not successful
825 * (eg. a seqcount has changed), then failure is returned and it's up to caller
826 * to restart the path walk from the beginning in ref-walk mode.
827 */
828
829/**
830 * try_to_unlazy - try to switch to ref-walk mode.
831 * @nd: nameidata pathwalk data
832 * Returns: true on success, false on failure
833 *
834 * try_to_unlazy attempts to legitimize the current nd->path and nd->root
835 * for ref-walk mode.
836 * Must be called from rcu-walk context.
837 * Nothing should touch nameidata between try_to_unlazy() failure and
838 * terminate_walk().
839 */
840static bool try_to_unlazy(struct nameidata *nd)
841{
842 struct dentry *parent = nd->path.dentry;
843
844 BUG_ON(!(nd->flags & LOOKUP_RCU));
845
846 if (unlikely(!legitimize_links(nd)))
847 goto out1;
848 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
849 goto out;
850 if (unlikely(!legitimize_root(nd)))
851 goto out;
852 leave_rcu(nd);
853 BUG_ON(nd->inode != parent->d_inode);
854 return true;
855
856out1:
857 nd->path.mnt = NULL;
858 nd->path.dentry = NULL;
859out:
860 leave_rcu(nd);
861 return false;
862}
863
864/**
865 * try_to_unlazy_next - try to switch to ref-walk mode.
866 * @nd: nameidata pathwalk data
867 * @dentry: next dentry to step into
868 * Returns: true on success, false on failure
869 *
870 * Similar to try_to_unlazy(), but here we have the next dentry already
871 * picked by rcu-walk and want to legitimize that in addition to the current
872 * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context.
873 * Nothing should touch nameidata between try_to_unlazy_next() failure and
874 * terminate_walk().
875 */
876static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry)
877{
878 int res;
879 BUG_ON(!(nd->flags & LOOKUP_RCU));
880
881 if (unlikely(!legitimize_links(nd)))
882 goto out2;
883 res = __legitimize_mnt(nd->path.mnt, nd->m_seq);
884 if (unlikely(res)) {
885 if (res > 0)
886 goto out2;
887 goto out1;
888 }
889 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
890 goto out1;
891
892 /*
893 * We need to move both the parent and the dentry from the RCU domain
894 * to be properly refcounted. And the sequence number in the dentry
895 * validates *both* dentry counters, since we checked the sequence
896 * number of the parent after we got the child sequence number. So we
897 * know the parent must still be valid if the child sequence number is
898 */
899 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
900 goto out;
901 if (read_seqcount_retry(&dentry->d_seq, nd->next_seq))
902 goto out_dput;
903 /*
904 * Sequence counts matched. Now make sure that the root is
905 * still valid and get it if required.
906 */
907 if (unlikely(!legitimize_root(nd)))
908 goto out_dput;
909 leave_rcu(nd);
910 return true;
911
912out2:
913 nd->path.mnt = NULL;
914out1:
915 nd->path.dentry = NULL;
916out:
917 leave_rcu(nd);
918 return false;
919out_dput:
920 leave_rcu(nd);
921 dput(dentry);
922 return false;
923}
924
925static inline int d_revalidate(struct inode *dir, const struct qstr *name,
926 struct dentry *dentry, unsigned int flags)
927{
928 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
929 return dentry->d_op->d_revalidate(dir, name, dentry, flags);
930 else
931 return 1;
932}
933
934/**
935 * complete_walk - successful completion of path walk
936 * @nd: pointer nameidata
937 *
938 * If we had been in RCU mode, drop out of it and legitimize nd->path.
939 * Revalidate the final result, unless we'd already done that during
940 * the path walk or the filesystem doesn't ask for it. Return 0 on
941 * success, -error on failure. In case of failure caller does not
942 * need to drop nd->path.
943 */
944static int complete_walk(struct nameidata *nd)
945{
946 struct dentry *dentry = nd->path.dentry;
947 int status;
948
949 if (nd->flags & LOOKUP_RCU) {
950 /*
951 * We don't want to zero nd->root for scoped-lookups or
952 * externally-managed nd->root.
953 */
954 if (!(nd->state & ND_ROOT_PRESET))
955 if (!(nd->flags & LOOKUP_IS_SCOPED))
956 nd->root.mnt = NULL;
957 nd->flags &= ~LOOKUP_CACHED;
958 if (!try_to_unlazy(nd))
959 return -ECHILD;
960 }
961
962 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
963 /*
964 * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
965 * ever step outside the root during lookup" and should already
966 * be guaranteed by the rest of namei, we want to avoid a namei
967 * BUG resulting in userspace being given a path that was not
968 * scoped within the root at some point during the lookup.
969 *
970 * So, do a final sanity-check to make sure that in the
971 * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
972 * we won't silently return an fd completely outside of the
973 * requested root to userspace.
974 *
975 * Userspace could move the path outside the root after this
976 * check, but as discussed elsewhere this is not a concern (the
977 * resolved file was inside the root at some point).
978 */
979 if (!path_is_under(&nd->path, &nd->root))
980 return -EXDEV;
981 }
982
983 if (likely(!(nd->state & ND_JUMPED)))
984 return 0;
985
986 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
987 return 0;
988
989 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
990 if (status > 0)
991 return 0;
992
993 if (!status)
994 status = -ESTALE;
995
996 return status;
997}
998
999static int set_root(struct nameidata *nd)
1000{
1001 struct fs_struct *fs = current->fs;
1002
1003 /*
1004 * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
1005 * still have to ensure it doesn't happen because it will cause a breakout
1006 * from the dirfd.
1007 */
1008 if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED))
1009 return -ENOTRECOVERABLE;
1010
1011 if (nd->flags & LOOKUP_RCU) {
1012 unsigned seq;
1013
1014 do {
1015 seq = read_seqcount_begin(&fs->seq);
1016 nd->root = fs->root;
1017 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
1018 } while (read_seqcount_retry(&fs->seq, seq));
1019 } else {
1020 get_fs_root(fs, root: &nd->root);
1021 nd->state |= ND_ROOT_GRABBED;
1022 }
1023 return 0;
1024}
1025
1026static int nd_jump_root(struct nameidata *nd)
1027{
1028 if (unlikely(nd->flags & LOOKUP_BENEATH))
1029 return -EXDEV;
1030 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
1031 /* Absolute path arguments to path_init() are allowed. */
1032 if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt)
1033 return -EXDEV;
1034 }
1035 if (!nd->root.mnt) {
1036 int error = set_root(nd);
1037 if (error)
1038 return error;
1039 }
1040 if (nd->flags & LOOKUP_RCU) {
1041 struct dentry *d;
1042 nd->path = nd->root;
1043 d = nd->path.dentry;
1044 nd->inode = d->d_inode;
1045 nd->seq = nd->root_seq;
1046 if (read_seqcount_retry(&d->d_seq, nd->seq))
1047 return -ECHILD;
1048 } else {
1049 path_put(&nd->path);
1050 nd->path = nd->root;
1051 path_get(&nd->path);
1052 nd->inode = nd->path.dentry->d_inode;
1053 }
1054 nd->state |= ND_JUMPED;
1055 return 0;
1056}
1057
1058/*
1059 * Helper to directly jump to a known parsed path from ->get_link,
1060 * caller must have taken a reference to path beforehand.
1061 */
1062int nd_jump_link(const struct path *path)
1063{
1064 int error = -ELOOP;
1065 struct nameidata *nd = current->nameidata;
1066
1067 if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS))
1068 goto err;
1069
1070 error = -EXDEV;
1071 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
1072 if (nd->path.mnt != path->mnt)
1073 goto err;
1074 }
1075 /* Not currently safe for scoped-lookups. */
1076 if (unlikely(nd->flags & LOOKUP_IS_SCOPED))
1077 goto err;
1078
1079 path_put(&nd->path);
1080 nd->path = *path;
1081 nd->inode = nd->path.dentry->d_inode;
1082 nd->state |= ND_JUMPED;
1083 return 0;
1084
1085err:
1086 path_put(path);
1087 return error;
1088}
1089
1090static inline void put_link(struct nameidata *nd)
1091{
1092 struct saved *last = nd->stack + --nd->depth;
1093 do_delayed_call(call: &last->done);
1094 if (!(nd->flags & LOOKUP_RCU))
1095 path_put(&last->link);
1096}
1097
1098static int sysctl_protected_symlinks __read_mostly;
1099static int sysctl_protected_hardlinks __read_mostly;
1100static int sysctl_protected_fifos __read_mostly;
1101static int sysctl_protected_regular __read_mostly;
1102
1103#ifdef CONFIG_SYSCTL
1104static const struct ctl_table namei_sysctls[] = {
1105 {
1106 .procname = "protected_symlinks",
1107 .data = &sysctl_protected_symlinks,
1108 .maxlen = sizeof(int),
1109 .mode = 0644,
1110 .proc_handler = proc_dointvec_minmax,
1111 .extra1 = SYSCTL_ZERO,
1112 .extra2 = SYSCTL_ONE,
1113 },
1114 {
1115 .procname = "protected_hardlinks",
1116 .data = &sysctl_protected_hardlinks,
1117 .maxlen = sizeof(int),
1118 .mode = 0644,
1119 .proc_handler = proc_dointvec_minmax,
1120 .extra1 = SYSCTL_ZERO,
1121 .extra2 = SYSCTL_ONE,
1122 },
1123 {
1124 .procname = "protected_fifos",
1125 .data = &sysctl_protected_fifos,
1126 .maxlen = sizeof(int),
1127 .mode = 0644,
1128 .proc_handler = proc_dointvec_minmax,
1129 .extra1 = SYSCTL_ZERO,
1130 .extra2 = SYSCTL_TWO,
1131 },
1132 {
1133 .procname = "protected_regular",
1134 .data = &sysctl_protected_regular,
1135 .maxlen = sizeof(int),
1136 .mode = 0644,
1137 .proc_handler = proc_dointvec_minmax,
1138 .extra1 = SYSCTL_ZERO,
1139 .extra2 = SYSCTL_TWO,
1140 },
1141};
1142
1143static int __init init_fs_namei_sysctls(void)
1144{
1145 register_sysctl_init("fs", namei_sysctls);
1146 return 0;
1147}
1148fs_initcall(init_fs_namei_sysctls);
1149
1150#endif /* CONFIG_SYSCTL */
1151
1152/**
1153 * may_follow_link - Check symlink following for unsafe situations
1154 * @nd: nameidata pathwalk data
1155 * @inode: Used for idmapping.
1156 *
1157 * In the case of the sysctl_protected_symlinks sysctl being enabled,
1158 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
1159 * in a sticky world-writable directory. This is to protect privileged
1160 * processes from failing races against path names that may change out
1161 * from under them by way of other users creating malicious symlinks.
1162 * It will permit symlinks to be followed only when outside a sticky
1163 * world-writable directory, or when the uid of the symlink and follower
1164 * match, or when the directory owner matches the symlink's owner.
1165 *
1166 * Returns 0 if following the symlink is allowed, -ve on error.
1167 */
1168static inline int may_follow_link(struct nameidata *nd, const struct inode *inode)
1169{
1170 struct mnt_idmap *idmap;
1171 vfsuid_t vfsuid;
1172
1173 if (!sysctl_protected_symlinks)
1174 return 0;
1175
1176 idmap = mnt_idmap(mnt: nd->path.mnt);
1177 vfsuid = i_uid_into_vfsuid(idmap, inode);
1178 /* Allowed if owner and follower match. */
1179 if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
1180 return 0;
1181
1182 /* Allowed if parent directory not sticky and world-writable. */
1183 if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
1184 return 0;
1185
1186 /* Allowed if parent directory and link owner match. */
1187 if (vfsuid_valid(uid: nd->dir_vfsuid) && vfsuid_eq(left: nd->dir_vfsuid, right: vfsuid))
1188 return 0;
1189
1190 if (nd->flags & LOOKUP_RCU)
1191 return -ECHILD;
1192
1193 audit_inode(name: nd->name, dentry: nd->stack[0].link.dentry, aflags: 0);
1194 audit_log_path_denied(AUDIT_ANOM_LINK, operation: "follow_link");
1195 return -EACCES;
1196}
1197
1198/**
1199 * safe_hardlink_source - Check for safe hardlink conditions
1200 * @idmap: idmap of the mount the inode was found from
1201 * @inode: the source inode to hardlink from
1202 *
1203 * Return false if at least one of the following conditions:
1204 * - inode is not a regular file
1205 * - inode is setuid
1206 * - inode is setgid and group-exec
1207 * - access failure for read and write
1208 *
1209 * Otherwise returns true.
1210 */
1211static bool safe_hardlink_source(struct mnt_idmap *idmap,
1212 struct inode *inode)
1213{
1214 umode_t mode = inode->i_mode;
1215
1216 /* Special files should not get pinned to the filesystem. */
1217 if (!S_ISREG(mode))
1218 return false;
1219
1220 /* Setuid files should not get pinned to the filesystem. */
1221 if (mode & S_ISUID)
1222 return false;
1223
1224 /* Executable setgid files should not get pinned to the filesystem. */
1225 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
1226 return false;
1227
1228 /* Hardlinking to unreadable or unwritable sources is dangerous. */
1229 if (inode_permission(idmap, inode, MAY_READ | MAY_WRITE))
1230 return false;
1231
1232 return true;
1233}
1234
1235/**
1236 * may_linkat - Check permissions for creating a hardlink
1237 * @idmap: idmap of the mount the inode was found from
1238 * @link: the source to hardlink from
1239 *
1240 * Block hardlink when all of:
1241 * - sysctl_protected_hardlinks enabled
1242 * - fsuid does not match inode
1243 * - hardlink source is unsafe (see safe_hardlink_source() above)
1244 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1245 *
1246 * If the inode has been found through an idmapped mount the idmap of
1247 * the vfsmount must be passed through @idmap. This function will then take
1248 * care to map the inode according to @idmap before checking permissions.
1249 * On non-idmapped mounts or if permission checking is to be performed on the
1250 * raw inode simply pass @nop_mnt_idmap.
1251 *
1252 * Returns 0 if successful, -ve on error.
1253 */
1254int may_linkat(struct mnt_idmap *idmap, const struct path *link)
1255{
1256 struct inode *inode = link->dentry->d_inode;
1257
1258 /* Inode writeback is not safe when the uid or gid are invalid. */
1259 if (!vfsuid_valid(uid: i_uid_into_vfsuid(idmap, inode)) ||
1260 !vfsgid_valid(gid: i_gid_into_vfsgid(idmap, inode)))
1261 return -EOVERFLOW;
1262
1263 if (!sysctl_protected_hardlinks)
1264 return 0;
1265
1266 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1267 * otherwise, it must be a safe source.
1268 */
1269 if (safe_hardlink_source(idmap, inode) ||
1270 inode_owner_or_capable(idmap, inode))
1271 return 0;
1272
1273 audit_log_path_denied(AUDIT_ANOM_LINK, operation: "linkat");
1274 return -EPERM;
1275}
1276
1277/**
1278 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1279 * should be allowed, or not, on files that already
1280 * exist.
1281 * @idmap: idmap of the mount the inode was found from
1282 * @nd: nameidata pathwalk data
1283 * @inode: the inode of the file to open
1284 *
1285 * Block an O_CREAT open of a FIFO (or a regular file) when:
1286 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1287 * - the file already exists
1288 * - we are in a sticky directory
1289 * - we don't own the file
1290 * - the owner of the directory doesn't own the file
1291 * - the directory is world writable
1292 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1293 * the directory doesn't have to be world writable: being group writable will
1294 * be enough.
1295 *
1296 * If the inode has been found through an idmapped mount the idmap of
1297 * the vfsmount must be passed through @idmap. This function will then take
1298 * care to map the inode according to @idmap before checking permissions.
1299 * On non-idmapped mounts or if permission checking is to be performed on the
1300 * raw inode simply pass @nop_mnt_idmap.
1301 *
1302 * Returns 0 if the open is allowed, -ve on error.
1303 */
1304static int may_create_in_sticky(struct mnt_idmap *idmap, struct nameidata *nd,
1305 struct inode *const inode)
1306{
1307 umode_t dir_mode = nd->dir_mode;
1308 vfsuid_t dir_vfsuid = nd->dir_vfsuid, i_vfsuid;
1309
1310 if (likely(!(dir_mode & S_ISVTX)))
1311 return 0;
1312
1313 if (S_ISREG(inode->i_mode) && !sysctl_protected_regular)
1314 return 0;
1315
1316 if (S_ISFIFO(inode->i_mode) && !sysctl_protected_fifos)
1317 return 0;
1318
1319 i_vfsuid = i_uid_into_vfsuid(idmap, inode);
1320
1321 if (vfsuid_eq(left: i_vfsuid, right: dir_vfsuid))
1322 return 0;
1323
1324 if (vfsuid_eq_kuid(vfsuid: i_vfsuid, current_fsuid()))
1325 return 0;
1326
1327 if (likely(dir_mode & 0002)) {
1328 audit_log_path_denied(AUDIT_ANOM_CREAT, operation: "sticky_create");
1329 return -EACCES;
1330 }
1331
1332 if (dir_mode & 0020) {
1333 if (sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) {
1334 audit_log_path_denied(AUDIT_ANOM_CREAT,
1335 operation: "sticky_create_fifo");
1336 return -EACCES;
1337 }
1338
1339 if (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode)) {
1340 audit_log_path_denied(AUDIT_ANOM_CREAT,
1341 operation: "sticky_create_regular");
1342 return -EACCES;
1343 }
1344 }
1345
1346 return 0;
1347}
1348
1349/*
1350 * follow_up - Find the mountpoint of path's vfsmount
1351 *
1352 * Given a path, find the mountpoint of its source file system.
1353 * Replace @path with the path of the mountpoint in the parent mount.
1354 * Up is towards /.
1355 *
1356 * Return 1 if we went up a level and 0 if we were already at the
1357 * root.
1358 */
1359int follow_up(struct path *path)
1360{
1361 struct mount *mnt = real_mount(mnt: path->mnt);
1362 struct mount *parent;
1363 struct dentry *mountpoint;
1364
1365 read_seqlock_excl(sl: &mount_lock);
1366 parent = mnt->mnt_parent;
1367 if (parent == mnt) {
1368 read_sequnlock_excl(sl: &mount_lock);
1369 return 0;
1370 }
1371 mntget(mnt: &parent->mnt);
1372 mountpoint = dget(dentry: mnt->mnt_mountpoint);
1373 read_sequnlock_excl(sl: &mount_lock);
1374 dput(path->dentry);
1375 path->dentry = mountpoint;
1376 mntput(mnt: path->mnt);
1377 path->mnt = &parent->mnt;
1378 return 1;
1379}
1380EXPORT_SYMBOL(follow_up);
1381
1382static bool choose_mountpoint_rcu(struct mount *m, const struct path *root,
1383 struct path *path, unsigned *seqp)
1384{
1385 while (mnt_has_parent(mnt: m)) {
1386 struct dentry *mountpoint = m->mnt_mountpoint;
1387
1388 m = m->mnt_parent;
1389 if (unlikely(root->dentry == mountpoint &&
1390 root->mnt == &m->mnt))
1391 break;
1392 if (mountpoint != m->mnt.mnt_root) {
1393 path->mnt = &m->mnt;
1394 path->dentry = mountpoint;
1395 *seqp = read_seqcount_begin(&mountpoint->d_seq);
1396 return true;
1397 }
1398 }
1399 return false;
1400}
1401
1402static bool choose_mountpoint(struct mount *m, const struct path *root,
1403 struct path *path)
1404{
1405 bool found;
1406
1407 rcu_read_lock();
1408 while (1) {
1409 unsigned seq, mseq = read_seqbegin(sl: &mount_lock);
1410
1411 found = choose_mountpoint_rcu(m, root, path, seqp: &seq);
1412 if (unlikely(!found)) {
1413 if (!read_seqretry(sl: &mount_lock, start: mseq))
1414 break;
1415 } else {
1416 if (likely(__legitimize_path(path, seq, mseq)))
1417 break;
1418 rcu_read_unlock();
1419 path_put(path);
1420 rcu_read_lock();
1421 }
1422 }
1423 rcu_read_unlock();
1424 return found;
1425}
1426
1427/*
1428 * Perform an automount
1429 * - return -EISDIR to tell follow_managed() to stop and return the path we
1430 * were called with.
1431 */
1432static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1433{
1434 struct dentry *dentry = path->dentry;
1435
1436 /* We don't want to mount if someone's just doing a stat -
1437 * unless they're stat'ing a directory and appended a '/' to
1438 * the name.
1439 *
1440 * We do, however, want to mount if someone wants to open or
1441 * create a file of any type under the mountpoint, wants to
1442 * traverse through the mountpoint or wants to open the
1443 * mounted directory. Also, autofs may mark negative dentries
1444 * as being automount points. These will need the attentions
1445 * of the daemon to instantiate them before they can be used.
1446 */
1447 if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1448 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1449 dentry->d_inode)
1450 return -EISDIR;
1451
1452 if (count && (*count)++ >= MAXSYMLINKS)
1453 return -ELOOP;
1454
1455 return finish_automount(dentry->d_op->d_automount(path), path);
1456}
1457
1458/*
1459 * mount traversal - out-of-line part. One note on ->d_flags accesses -
1460 * dentries are pinned but not locked here, so negative dentry can go
1461 * positive right under us. Use of smp_load_acquire() provides a barrier
1462 * sufficient for ->d_inode and ->d_flags consistency.
1463 */
1464static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped,
1465 int *count, unsigned lookup_flags)
1466{
1467 struct vfsmount *mnt = path->mnt;
1468 bool need_mntput = false;
1469 int ret = 0;
1470
1471 while (flags & DCACHE_MANAGED_DENTRY) {
1472 /* Allow the filesystem to manage the transit without i_mutex
1473 * being held. */
1474 if (flags & DCACHE_MANAGE_TRANSIT) {
1475 ret = path->dentry->d_op->d_manage(path, false);
1476 flags = smp_load_acquire(&path->dentry->d_flags);
1477 if (ret < 0)
1478 break;
1479 }
1480
1481 if (flags & DCACHE_MOUNTED) { // something's mounted on it..
1482 struct vfsmount *mounted = lookup_mnt(path);
1483 if (mounted) { // ... in our namespace
1484 dput(path->dentry);
1485 if (need_mntput)
1486 mntput(mnt: path->mnt);
1487 path->mnt = mounted;
1488 path->dentry = dget(dentry: mounted->mnt_root);
1489 // here we know it's positive
1490 flags = path->dentry->d_flags;
1491 need_mntput = true;
1492 continue;
1493 }
1494 }
1495
1496 if (!(flags & DCACHE_NEED_AUTOMOUNT))
1497 break;
1498
1499 // uncovered automount point
1500 ret = follow_automount(path, count, lookup_flags);
1501 flags = smp_load_acquire(&path->dentry->d_flags);
1502 if (ret < 0)
1503 break;
1504 }
1505
1506 if (ret == -EISDIR)
1507 ret = 0;
1508 // possible if you race with several mount --move
1509 if (need_mntput && path->mnt == mnt)
1510 mntput(mnt: path->mnt);
1511 if (!ret && unlikely(d_flags_negative(flags)))
1512 ret = -ENOENT;
1513 *jumped = need_mntput;
1514 return ret;
1515}
1516
1517static inline int traverse_mounts(struct path *path, bool *jumped,
1518 int *count, unsigned lookup_flags)
1519{
1520 unsigned flags = smp_load_acquire(&path->dentry->d_flags);
1521
1522 /* fastpath */
1523 if (likely(!(flags & DCACHE_MANAGED_DENTRY))) {
1524 *jumped = false;
1525 if (unlikely(d_flags_negative(flags)))
1526 return -ENOENT;
1527 return 0;
1528 }
1529 return __traverse_mounts(path, flags, jumped, count, lookup_flags);
1530}
1531
1532int follow_down_one(struct path *path)
1533{
1534 struct vfsmount *mounted;
1535
1536 mounted = lookup_mnt(path);
1537 if (mounted) {
1538 dput(path->dentry);
1539 mntput(mnt: path->mnt);
1540 path->mnt = mounted;
1541 path->dentry = dget(dentry: mounted->mnt_root);
1542 return 1;
1543 }
1544 return 0;
1545}
1546EXPORT_SYMBOL(follow_down_one);
1547
1548/*
1549 * Follow down to the covering mount currently visible to userspace. At each
1550 * point, the filesystem owning that dentry may be queried as to whether the
1551 * caller is permitted to proceed or not.
1552 */
1553int follow_down(struct path *path, unsigned int flags)
1554{
1555 struct vfsmount *mnt = path->mnt;
1556 bool jumped;
1557 int ret = traverse_mounts(path, jumped: &jumped, NULL, lookup_flags: flags);
1558
1559 if (path->mnt != mnt)
1560 mntput(mnt);
1561 return ret;
1562}
1563EXPORT_SYMBOL(follow_down);
1564
1565/*
1566 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1567 * we meet a managed dentry that would need blocking.
1568 */
1569static bool __follow_mount_rcu(struct nameidata *nd, struct path *path)
1570{
1571 struct dentry *dentry = path->dentry;
1572 unsigned int flags = dentry->d_flags;
1573
1574 if (likely(!(flags & DCACHE_MANAGED_DENTRY)))
1575 return true;
1576
1577 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1578 return false;
1579
1580 for (;;) {
1581 /*
1582 * Don't forget we might have a non-mountpoint managed dentry
1583 * that wants to block transit.
1584 */
1585 if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) {
1586 int res = dentry->d_op->d_manage(path, true);
1587 if (res)
1588 return res == -EISDIR;
1589 flags = dentry->d_flags;
1590 }
1591
1592 if (flags & DCACHE_MOUNTED) {
1593 struct mount *mounted = __lookup_mnt(path->mnt, dentry);
1594 if (mounted) {
1595 path->mnt = &mounted->mnt;
1596 dentry = path->dentry = mounted->mnt.mnt_root;
1597 nd->state |= ND_JUMPED;
1598 nd->next_seq = read_seqcount_begin(&dentry->d_seq);
1599 flags = dentry->d_flags;
1600 // makes sure that non-RCU pathwalk could reach
1601 // this state.
1602 if (read_seqretry(sl: &mount_lock, start: nd->m_seq))
1603 return false;
1604 continue;
1605 }
1606 if (read_seqretry(sl: &mount_lock, start: nd->m_seq))
1607 return false;
1608 }
1609 return !(flags & DCACHE_NEED_AUTOMOUNT);
1610 }
1611}
1612
1613static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1614 struct path *path)
1615{
1616 bool jumped;
1617 int ret;
1618
1619 path->mnt = nd->path.mnt;
1620 path->dentry = dentry;
1621 if (nd->flags & LOOKUP_RCU) {
1622 unsigned int seq = nd->next_seq;
1623 if (likely(__follow_mount_rcu(nd, path)))
1624 return 0;
1625 // *path and nd->next_seq might've been clobbered
1626 path->mnt = nd->path.mnt;
1627 path->dentry = dentry;
1628 nd->next_seq = seq;
1629 if (!try_to_unlazy_next(nd, dentry))
1630 return -ECHILD;
1631 }
1632 ret = traverse_mounts(path, jumped: &jumped, count: &nd->total_link_count, lookup_flags: nd->flags);
1633 if (jumped) {
1634 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1635 ret = -EXDEV;
1636 else
1637 nd->state |= ND_JUMPED;
1638 }
1639 if (unlikely(ret)) {
1640 dput(path->dentry);
1641 if (path->mnt != nd->path.mnt)
1642 mntput(mnt: path->mnt);
1643 }
1644 return ret;
1645}
1646
1647/*
1648 * This looks up the name in dcache and possibly revalidates the found dentry.
1649 * NULL is returned if the dentry does not exist in the cache.
1650 */
1651static struct dentry *lookup_dcache(const struct qstr *name,
1652 struct dentry *dir,
1653 unsigned int flags)
1654{
1655 struct dentry *dentry = d_lookup(dir, name);
1656 if (dentry) {
1657 int error = d_revalidate(dir: dir->d_inode, name, dentry, flags);
1658 if (unlikely(error <= 0)) {
1659 if (!error)
1660 d_invalidate(dentry);
1661 dput(dentry);
1662 return ERR_PTR(error);
1663 }
1664 }
1665 return dentry;
1666}
1667
1668static struct dentry *lookup_one_qstr_excl_raw(const struct qstr *name,
1669 struct dentry *base,
1670 unsigned int flags)
1671{
1672 struct dentry *dentry;
1673 struct dentry *old;
1674 struct inode *dir;
1675
1676 dentry = lookup_dcache(name, dir: base, flags);
1677 if (dentry)
1678 return dentry;
1679
1680 /* Don't create child dentry for a dead directory. */
1681 dir = base->d_inode;
1682 if (unlikely(IS_DEADDIR(dir)))
1683 return ERR_PTR(error: -ENOENT);
1684
1685 dentry = d_alloc(base, name);
1686 if (unlikely(!dentry))
1687 return ERR_PTR(error: -ENOMEM);
1688
1689 old = dir->i_op->lookup(dir, dentry, flags);
1690 if (unlikely(old)) {
1691 dput(dentry);
1692 dentry = old;
1693 }
1694 return dentry;
1695}
1696
1697/*
1698 * Parent directory has inode locked exclusive. This is one
1699 * and only case when ->lookup() gets called on non in-lookup
1700 * dentries - as the matter of fact, this only gets called
1701 * when directory is guaranteed to have no in-lookup children
1702 * at all.
1703 * Will return -ENOENT if name isn't found and LOOKUP_CREATE wasn't passed.
1704 * Will return -EEXIST if name is found and LOOKUP_EXCL was passed.
1705 */
1706struct dentry *lookup_one_qstr_excl(const struct qstr *name,
1707 struct dentry *base, unsigned int flags)
1708{
1709 struct dentry *dentry;
1710
1711 dentry = lookup_one_qstr_excl_raw(name, base, flags);
1712 if (IS_ERR(ptr: dentry))
1713 return dentry;
1714 if (d_is_negative(dentry) && !(flags & LOOKUP_CREATE)) {
1715 dput(dentry);
1716 return ERR_PTR(error: -ENOENT);
1717 }
1718 if (d_is_positive(dentry) && (flags & LOOKUP_EXCL)) {
1719 dput(dentry);
1720 return ERR_PTR(error: -EEXIST);
1721 }
1722 return dentry;
1723}
1724EXPORT_SYMBOL(lookup_one_qstr_excl);
1725
1726/**
1727 * lookup_fast - do fast lockless (but racy) lookup of a dentry
1728 * @nd: current nameidata
1729 *
1730 * Do a fast, but racy lookup in the dcache for the given dentry, and
1731 * revalidate it. Returns a valid dentry pointer or NULL if one wasn't
1732 * found. On error, an ERR_PTR will be returned.
1733 *
1734 * If this function returns a valid dentry and the walk is no longer
1735 * lazy, the dentry will carry a reference that must later be put. If
1736 * RCU mode is still in force, then this is not the case and the dentry
1737 * must be legitimized before use. If this returns NULL, then the walk
1738 * will no longer be in RCU mode.
1739 */
1740static struct dentry *lookup_fast(struct nameidata *nd)
1741{
1742 struct dentry *dentry, *parent = nd->path.dentry;
1743 int status = 1;
1744
1745 /*
1746 * Rename seqlock is not required here because in the off chance
1747 * of a false negative due to a concurrent rename, the caller is
1748 * going to fall back to non-racy lookup.
1749 */
1750 if (nd->flags & LOOKUP_RCU) {
1751 dentry = __d_lookup_rcu(parent, name: &nd->last, seq: &nd->next_seq);
1752 if (unlikely(!dentry)) {
1753 if (!try_to_unlazy(nd))
1754 return ERR_PTR(error: -ECHILD);
1755 return NULL;
1756 }
1757
1758 /*
1759 * This sequence count validates that the parent had no
1760 * changes while we did the lookup of the dentry above.
1761 */
1762 if (read_seqcount_retry(&parent->d_seq, nd->seq))
1763 return ERR_PTR(error: -ECHILD);
1764
1765 status = d_revalidate(dir: nd->inode, name: &nd->last, dentry, flags: nd->flags);
1766 if (likely(status > 0))
1767 return dentry;
1768 if (!try_to_unlazy_next(nd, dentry))
1769 return ERR_PTR(error: -ECHILD);
1770 if (status == -ECHILD)
1771 /* we'd been told to redo it in non-rcu mode */
1772 status = d_revalidate(dir: nd->inode, name: &nd->last,
1773 dentry, flags: nd->flags);
1774 } else {
1775 dentry = __d_lookup(parent, &nd->last);
1776 if (unlikely(!dentry))
1777 return NULL;
1778 status = d_revalidate(dir: nd->inode, name: &nd->last, dentry, flags: nd->flags);
1779 }
1780 if (unlikely(status <= 0)) {
1781 if (!status)
1782 d_invalidate(dentry);
1783 dput(dentry);
1784 return ERR_PTR(error: status);
1785 }
1786 return dentry;
1787}
1788
1789/* Fast lookup failed, do it the slow way */
1790static struct dentry *__lookup_slow(const struct qstr *name,
1791 struct dentry *dir,
1792 unsigned int flags)
1793{
1794 struct dentry *dentry, *old;
1795 struct inode *inode = dir->d_inode;
1796 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1797
1798 /* Don't go there if it's already dead */
1799 if (unlikely(IS_DEADDIR(inode)))
1800 return ERR_PTR(error: -ENOENT);
1801again:
1802 dentry = d_alloc_parallel(dir, name, &wq);
1803 if (IS_ERR(ptr: dentry))
1804 return dentry;
1805 if (unlikely(!d_in_lookup(dentry))) {
1806 int error = d_revalidate(dir: inode, name, dentry, flags);
1807 if (unlikely(error <= 0)) {
1808 if (!error) {
1809 d_invalidate(dentry);
1810 dput(dentry);
1811 goto again;
1812 }
1813 dput(dentry);
1814 dentry = ERR_PTR(error);
1815 }
1816 } else {
1817 old = inode->i_op->lookup(inode, dentry, flags);
1818 d_lookup_done(dentry);
1819 if (unlikely(old)) {
1820 dput(dentry);
1821 dentry = old;
1822 }
1823 }
1824 return dentry;
1825}
1826
1827static struct dentry *lookup_slow(const struct qstr *name,
1828 struct dentry *dir,
1829 unsigned int flags)
1830{
1831 struct inode *inode = dir->d_inode;
1832 struct dentry *res;
1833 inode_lock_shared(inode);
1834 res = __lookup_slow(name, dir, flags);
1835 inode_unlock_shared(inode);
1836 return res;
1837}
1838
1839static inline int may_lookup(struct mnt_idmap *idmap,
1840 struct nameidata *restrict nd)
1841{
1842 int err, mask;
1843
1844 mask = nd->flags & LOOKUP_RCU ? MAY_NOT_BLOCK : 0;
1845 err = inode_permission(idmap, nd->inode, mask | MAY_EXEC);
1846 if (likely(!err))
1847 return 0;
1848
1849 // If we failed, and we weren't in LOOKUP_RCU, it's final
1850 if (!(nd->flags & LOOKUP_RCU))
1851 return err;
1852
1853 // Drop out of RCU mode to make sure it wasn't transient
1854 if (!try_to_unlazy(nd))
1855 return -ECHILD; // redo it all non-lazy
1856
1857 if (err != -ECHILD) // hard error
1858 return err;
1859
1860 return inode_permission(idmap, nd->inode, MAY_EXEC);
1861}
1862
1863static int reserve_stack(struct nameidata *nd, struct path *link)
1864{
1865 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1866 return -ELOOP;
1867
1868 if (likely(nd->depth != EMBEDDED_LEVELS))
1869 return 0;
1870 if (likely(nd->stack != nd->internal))
1871 return 0;
1872 if (likely(nd_alloc_stack(nd)))
1873 return 0;
1874
1875 if (nd->flags & LOOKUP_RCU) {
1876 // we need to grab link before we do unlazy. And we can't skip
1877 // unlazy even if we fail to grab the link - cleanup needs it
1878 bool grabbed_link = legitimize_path(nd, path: link, seq: nd->next_seq);
1879
1880 if (!try_to_unlazy(nd) || !grabbed_link)
1881 return -ECHILD;
1882
1883 if (nd_alloc_stack(nd))
1884 return 0;
1885 }
1886 return -ENOMEM;
1887}
1888
1889enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1890
1891static const char *pick_link(struct nameidata *nd, struct path *link,
1892 struct inode *inode, int flags)
1893{
1894 struct saved *last;
1895 const char *res;
1896 int error = reserve_stack(nd, link);
1897
1898 if (unlikely(error)) {
1899 if (!(nd->flags & LOOKUP_RCU))
1900 path_put(link);
1901 return ERR_PTR(error);
1902 }
1903 last = nd->stack + nd->depth++;
1904 last->link = *link;
1905 clear_delayed_call(call: &last->done);
1906 last->seq = nd->next_seq;
1907
1908 if (flags & WALK_TRAILING) {
1909 error = may_follow_link(nd, inode);
1910 if (unlikely(error))
1911 return ERR_PTR(error);
1912 }
1913
1914 if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) ||
1915 unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW))
1916 return ERR_PTR(error: -ELOOP);
1917
1918 if (unlikely(atime_needs_update(&last->link, inode))) {
1919 if (nd->flags & LOOKUP_RCU) {
1920 if (!try_to_unlazy(nd))
1921 return ERR_PTR(error: -ECHILD);
1922 }
1923 touch_atime(&last->link);
1924 cond_resched();
1925 }
1926
1927 error = security_inode_follow_link(dentry: link->dentry, inode,
1928 rcu: nd->flags & LOOKUP_RCU);
1929 if (unlikely(error))
1930 return ERR_PTR(error);
1931
1932 res = READ_ONCE(inode->i_link);
1933 if (!res) {
1934 const char * (*get)(struct dentry *, struct inode *,
1935 struct delayed_call *);
1936 get = inode->i_op->get_link;
1937 if (nd->flags & LOOKUP_RCU) {
1938 res = get(NULL, inode, &last->done);
1939 if (res == ERR_PTR(error: -ECHILD) && try_to_unlazy(nd))
1940 res = get(link->dentry, inode, &last->done);
1941 } else {
1942 res = get(link->dentry, inode, &last->done);
1943 }
1944 if (!res)
1945 goto all_done;
1946 if (IS_ERR(ptr: res))
1947 return res;
1948 }
1949 if (*res == '/') {
1950 error = nd_jump_root(nd);
1951 if (unlikely(error))
1952 return ERR_PTR(error);
1953 while (unlikely(*++res == '/'))
1954 ;
1955 }
1956 if (*res)
1957 return res;
1958all_done: // pure jump
1959 put_link(nd);
1960 return NULL;
1961}
1962
1963/*
1964 * Do we need to follow links? We _really_ want to be able
1965 * to do this check without having to look at inode->i_op,
1966 * so we keep a cache of "no, this doesn't need follow_link"
1967 * for the common case.
1968 *
1969 * NOTE: dentry must be what nd->next_seq had been sampled from.
1970 */
1971static const char *step_into(struct nameidata *nd, int flags,
1972 struct dentry *dentry)
1973{
1974 struct path path;
1975 struct inode *inode;
1976 int err = handle_mounts(nd, dentry, path: &path);
1977
1978 if (err < 0)
1979 return ERR_PTR(error: err);
1980 inode = path.dentry->d_inode;
1981 if (likely(!d_is_symlink(path.dentry)) ||
1982 ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
1983 (flags & WALK_NOFOLLOW)) {
1984 /* not a symlink or should not follow */
1985 if (nd->flags & LOOKUP_RCU) {
1986 if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1987 return ERR_PTR(error: -ECHILD);
1988 if (unlikely(!inode))
1989 return ERR_PTR(error: -ENOENT);
1990 } else {
1991 dput(nd->path.dentry);
1992 if (nd->path.mnt != path.mnt)
1993 mntput(mnt: nd->path.mnt);
1994 }
1995 nd->path = path;
1996 nd->inode = inode;
1997 nd->seq = nd->next_seq;
1998 return NULL;
1999 }
2000 if (nd->flags & LOOKUP_RCU) {
2001 /* make sure that d_is_symlink above matches inode */
2002 if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
2003 return ERR_PTR(error: -ECHILD);
2004 } else {
2005 if (path.mnt == nd->path.mnt)
2006 mntget(mnt: path.mnt);
2007 }
2008 return pick_link(nd, link: &path, inode, flags);
2009}
2010
2011static struct dentry *follow_dotdot_rcu(struct nameidata *nd)
2012{
2013 struct dentry *parent, *old;
2014
2015 if (path_equal(path1: &nd->path, path2: &nd->root))
2016 goto in_root;
2017 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
2018 struct path path;
2019 unsigned seq;
2020 if (!choose_mountpoint_rcu(m: real_mount(mnt: nd->path.mnt),
2021 root: &nd->root, path: &path, seqp: &seq))
2022 goto in_root;
2023 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
2024 return ERR_PTR(error: -ECHILD);
2025 nd->path = path;
2026 nd->inode = path.dentry->d_inode;
2027 nd->seq = seq;
2028 // makes sure that non-RCU pathwalk could reach this state
2029 if (read_seqretry(sl: &mount_lock, start: nd->m_seq))
2030 return ERR_PTR(error: -ECHILD);
2031 /* we know that mountpoint was pinned */
2032 }
2033 old = nd->path.dentry;
2034 parent = old->d_parent;
2035 nd->next_seq = read_seqcount_begin(&parent->d_seq);
2036 // makes sure that non-RCU pathwalk could reach this state
2037 if (read_seqcount_retry(&old->d_seq, nd->seq))
2038 return ERR_PTR(error: -ECHILD);
2039 if (unlikely(!path_connected(nd->path.mnt, parent)))
2040 return ERR_PTR(error: -ECHILD);
2041 return parent;
2042in_root:
2043 if (read_seqretry(sl: &mount_lock, start: nd->m_seq))
2044 return ERR_PTR(error: -ECHILD);
2045 if (unlikely(nd->flags & LOOKUP_BENEATH))
2046 return ERR_PTR(error: -ECHILD);
2047 nd->next_seq = nd->seq;
2048 return nd->path.dentry;
2049}
2050
2051static struct dentry *follow_dotdot(struct nameidata *nd)
2052{
2053 struct dentry *parent;
2054
2055 if (path_equal(path1: &nd->path, path2: &nd->root))
2056 goto in_root;
2057 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
2058 struct path path;
2059
2060 if (!choose_mountpoint(m: real_mount(mnt: nd->path.mnt),
2061 root: &nd->root, path: &path))
2062 goto in_root;
2063 path_put(&nd->path);
2064 nd->path = path;
2065 nd->inode = path.dentry->d_inode;
2066 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
2067 return ERR_PTR(error: -EXDEV);
2068 }
2069 /* rare case of legitimate dget_parent()... */
2070 parent = dget_parent(dentry: nd->path.dentry);
2071 if (unlikely(!path_connected(nd->path.mnt, parent))) {
2072 dput(parent);
2073 return ERR_PTR(error: -ENOENT);
2074 }
2075 return parent;
2076
2077in_root:
2078 if (unlikely(nd->flags & LOOKUP_BENEATH))
2079 return ERR_PTR(error: -EXDEV);
2080 return dget(dentry: nd->path.dentry);
2081}
2082
2083static const char *handle_dots(struct nameidata *nd, int type)
2084{
2085 if (type == LAST_DOTDOT) {
2086 const char *error = NULL;
2087 struct dentry *parent;
2088
2089 if (!nd->root.mnt) {
2090 error = ERR_PTR(error: set_root(nd));
2091 if (error)
2092 return error;
2093 }
2094 if (nd->flags & LOOKUP_RCU)
2095 parent = follow_dotdot_rcu(nd);
2096 else
2097 parent = follow_dotdot(nd);
2098 if (IS_ERR(ptr: parent))
2099 return ERR_CAST(ptr: parent);
2100 error = step_into(nd, flags: WALK_NOFOLLOW, dentry: parent);
2101 if (unlikely(error))
2102 return error;
2103
2104 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
2105 /*
2106 * If there was a racing rename or mount along our
2107 * path, then we can't be sure that ".." hasn't jumped
2108 * above nd->root (and so userspace should retry or use
2109 * some fallback).
2110 */
2111 smp_rmb();
2112 if (__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))
2113 return ERR_PTR(error: -EAGAIN);
2114 if (__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))
2115 return ERR_PTR(error: -EAGAIN);
2116 }
2117 }
2118 return NULL;
2119}
2120
2121static const char *walk_component(struct nameidata *nd, int flags)
2122{
2123 struct dentry *dentry;
2124 /*
2125 * "." and ".." are special - ".." especially so because it has
2126 * to be able to know about the current root directory and
2127 * parent relationships.
2128 */
2129 if (unlikely(nd->last_type != LAST_NORM)) {
2130 if (!(flags & WALK_MORE) && nd->depth)
2131 put_link(nd);
2132 return handle_dots(nd, type: nd->last_type);
2133 }
2134 dentry = lookup_fast(nd);
2135 if (IS_ERR(ptr: dentry))
2136 return ERR_CAST(ptr: dentry);
2137 if (unlikely(!dentry)) {
2138 dentry = lookup_slow(name: &nd->last, dir: nd->path.dentry, flags: nd->flags);
2139 if (IS_ERR(ptr: dentry))
2140 return ERR_CAST(ptr: dentry);
2141 }
2142 if (!(flags & WALK_MORE) && nd->depth)
2143 put_link(nd);
2144 return step_into(nd, flags, dentry);
2145}
2146
2147/*
2148 * We can do the critical dentry name comparison and hashing
2149 * operations one word at a time, but we are limited to:
2150 *
2151 * - Architectures with fast unaligned word accesses. We could
2152 * do a "get_unaligned()" if this helps and is sufficiently
2153 * fast.
2154 *
2155 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
2156 * do not trap on the (extremely unlikely) case of a page
2157 * crossing operation.
2158 *
2159 * - Furthermore, we need an efficient 64-bit compile for the
2160 * 64-bit case in order to generate the "number of bytes in
2161 * the final mask". Again, that could be replaced with a
2162 * efficient population count instruction or similar.
2163 */
2164#ifdef CONFIG_DCACHE_WORD_ACCESS
2165
2166#include <asm/word-at-a-time.h>
2167
2168#ifdef HASH_MIX
2169
2170/* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
2171
2172#elif defined(CONFIG_64BIT)
2173/*
2174 * Register pressure in the mixing function is an issue, particularly
2175 * on 32-bit x86, but almost any function requires one state value and
2176 * one temporary. Instead, use a function designed for two state values
2177 * and no temporaries.
2178 *
2179 * This function cannot create a collision in only two iterations, so
2180 * we have two iterations to achieve avalanche. In those two iterations,
2181 * we have six layers of mixing, which is enough to spread one bit's
2182 * influence out to 2^6 = 64 state bits.
2183 *
2184 * Rotate constants are scored by considering either 64 one-bit input
2185 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
2186 * probability of that delta causing a change to each of the 128 output
2187 * bits, using a sample of random initial states.
2188 *
2189 * The Shannon entropy of the computed probabilities is then summed
2190 * to produce a score. Ideally, any input change has a 50% chance of
2191 * toggling any given output bit.
2192 *
2193 * Mixing scores (in bits) for (12,45):
2194 * Input delta: 1-bit 2-bit
2195 * 1 round: 713.3 42542.6
2196 * 2 rounds: 2753.7 140389.8
2197 * 3 rounds: 5954.1 233458.2
2198 * 4 rounds: 7862.6 256672.2
2199 * Perfect: 8192 258048
2200 * (64*128) (64*63/2 * 128)
2201 */
2202#define HASH_MIX(x, y, a) \
2203 ( x ^= (a), \
2204 y ^= x, x = rol64(x,12),\
2205 x += y, y = rol64(y,45),\
2206 y *= 9 )
2207
2208/*
2209 * Fold two longs into one 32-bit hash value. This must be fast, but
2210 * latency isn't quite as critical, as there is a fair bit of additional
2211 * work done before the hash value is used.
2212 */
2213static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2214{
2215 y ^= x * GOLDEN_RATIO_64;
2216 y *= GOLDEN_RATIO_64;
2217 return y >> 32;
2218}
2219
2220#else /* 32-bit case */
2221
2222/*
2223 * Mixing scores (in bits) for (7,20):
2224 * Input delta: 1-bit 2-bit
2225 * 1 round: 330.3 9201.6
2226 * 2 rounds: 1246.4 25475.4
2227 * 3 rounds: 1907.1 31295.1
2228 * 4 rounds: 2042.3 31718.6
2229 * Perfect: 2048 31744
2230 * (32*64) (32*31/2 * 64)
2231 */
2232#define HASH_MIX(x, y, a) \
2233 ( x ^= (a), \
2234 y ^= x, x = rol32(x, 7),\
2235 x += y, y = rol32(y,20),\
2236 y *= 9 )
2237
2238static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2239{
2240 /* Use arch-optimized multiply if one exists */
2241 return __hash_32(y ^ __hash_32(x));
2242}
2243
2244#endif
2245
2246/*
2247 * Return the hash of a string of known length. This is carfully
2248 * designed to match hash_name(), which is the more critical function.
2249 * In particular, we must end by hashing a final word containing 0..7
2250 * payload bytes, to match the way that hash_name() iterates until it
2251 * finds the delimiter after the name.
2252 */
2253unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2254{
2255 unsigned long a, x = 0, y = (unsigned long)salt;
2256
2257 for (;;) {
2258 if (!len)
2259 goto done;
2260 a = load_unaligned_zeropad(addr: name);
2261 if (len < sizeof(unsigned long))
2262 break;
2263 HASH_MIX(x, y, a);
2264 name += sizeof(unsigned long);
2265 len -= sizeof(unsigned long);
2266 }
2267 x ^= a & bytemask_from_count(len);
2268done:
2269 return fold_hash(x, y);
2270}
2271EXPORT_SYMBOL(full_name_hash);
2272
2273/* Return the "hash_len" (hash and length) of a null-terminated string */
2274u64 hashlen_string(const void *salt, const char *name)
2275{
2276 unsigned long a = 0, x = 0, y = (unsigned long)salt;
2277 unsigned long adata, mask, len;
2278 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2279
2280 len = 0;
2281 goto inside;
2282
2283 do {
2284 HASH_MIX(x, y, a);
2285 len += sizeof(unsigned long);
2286inside:
2287 a = load_unaligned_zeropad(addr: name+len);
2288 } while (!has_zero(a, bits: &adata, c: &constants));
2289
2290 adata = prep_zero_mask(a, bits: adata, c: &constants);
2291 mask = create_zero_mask(adata);
2292 x ^= a & zero_bytemask(bits: mask);
2293
2294 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2295}
2296EXPORT_SYMBOL(hashlen_string);
2297
2298/*
2299 * Calculate the length and hash of the path component, and
2300 * return the length as the result.
2301 */
2302static inline const char *hash_name(struct nameidata *nd,
2303 const char *name,
2304 unsigned long *lastword)
2305{
2306 unsigned long a, b, x, y = (unsigned long)nd->path.dentry;
2307 unsigned long adata, bdata, mask, len;
2308 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2309
2310 /*
2311 * The first iteration is special, because it can result in
2312 * '.' and '..' and has no mixing other than the final fold.
2313 */
2314 a = load_unaligned_zeropad(addr: name);
2315 b = a ^ REPEAT_BYTE('/');
2316 if (has_zero(a, bits: &adata, c: &constants) | has_zero(a: b, bits: &bdata, c: &constants)) {
2317 adata = prep_zero_mask(a, bits: adata, c: &constants);
2318 bdata = prep_zero_mask(a: b, bits: bdata, c: &constants);
2319 mask = create_zero_mask(adata | bdata);
2320 a &= zero_bytemask(bits: mask);
2321 *lastword = a;
2322 len = find_zero(mask);
2323 nd->last.hash = fold_hash(x: a, y);
2324 nd->last.len = len;
2325 return name + len;
2326 }
2327
2328 len = 0;
2329 x = 0;
2330 do {
2331 HASH_MIX(x, y, a);
2332 len += sizeof(unsigned long);
2333 a = load_unaligned_zeropad(addr: name+len);
2334 b = a ^ REPEAT_BYTE('/');
2335 } while (!(has_zero(a, bits: &adata, c: &constants) | has_zero(a: b, bits: &bdata, c: &constants)));
2336
2337 adata = prep_zero_mask(a, bits: adata, c: &constants);
2338 bdata = prep_zero_mask(a: b, bits: bdata, c: &constants);
2339 mask = create_zero_mask(adata | bdata);
2340 a &= zero_bytemask(bits: mask);
2341 x ^= a;
2342 len += find_zero(mask);
2343 *lastword = 0; // Multi-word components cannot be DOT or DOTDOT
2344
2345 nd->last.hash = fold_hash(x, y);
2346 nd->last.len = len;
2347 return name + len;
2348}
2349
2350/*
2351 * Note that the 'last' word is always zero-masked, but
2352 * was loaded as a possibly big-endian word.
2353 */
2354#ifdef __BIG_ENDIAN
2355 #define LAST_WORD_IS_DOT (0x2eul << (BITS_PER_LONG-8))
2356 #define LAST_WORD_IS_DOTDOT (0x2e2eul << (BITS_PER_LONG-16))
2357#endif
2358
2359#else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2360
2361/* Return the hash of a string of known length */
2362unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2363{
2364 unsigned long hash = init_name_hash(salt);
2365 while (len--)
2366 hash = partial_name_hash((unsigned char)*name++, hash);
2367 return end_name_hash(hash);
2368}
2369EXPORT_SYMBOL(full_name_hash);
2370
2371/* Return the "hash_len" (hash and length) of a null-terminated string */
2372u64 hashlen_string(const void *salt, const char *name)
2373{
2374 unsigned long hash = init_name_hash(salt);
2375 unsigned long len = 0, c;
2376
2377 c = (unsigned char)*name;
2378 while (c) {
2379 len++;
2380 hash = partial_name_hash(c, hash);
2381 c = (unsigned char)name[len];
2382 }
2383 return hashlen_create(end_name_hash(hash), len);
2384}
2385EXPORT_SYMBOL(hashlen_string);
2386
2387/*
2388 * We know there's a real path component here of at least
2389 * one character.
2390 */
2391static inline const char *hash_name(struct nameidata *nd, const char *name, unsigned long *lastword)
2392{
2393 unsigned long hash = init_name_hash(nd->path.dentry);
2394 unsigned long len = 0, c, last = 0;
2395
2396 c = (unsigned char)*name;
2397 do {
2398 last = (last << 8) + c;
2399 len++;
2400 hash = partial_name_hash(c, hash);
2401 c = (unsigned char)name[len];
2402 } while (c && c != '/');
2403
2404 // This is reliable for DOT or DOTDOT, since the component
2405 // cannot contain NUL characters - top bits being zero means
2406 // we cannot have had any other pathnames.
2407 *lastword = last;
2408 nd->last.hash = end_name_hash(hash);
2409 nd->last.len = len;
2410 return name + len;
2411}
2412
2413#endif
2414
2415#ifndef LAST_WORD_IS_DOT
2416 #define LAST_WORD_IS_DOT 0x2e
2417 #define LAST_WORD_IS_DOTDOT 0x2e2e
2418#endif
2419
2420/*
2421 * Name resolution.
2422 * This is the basic name resolution function, turning a pathname into
2423 * the final dentry. We expect 'base' to be positive and a directory.
2424 *
2425 * Returns 0 and nd will have valid dentry and mnt on success.
2426 * Returns error and drops reference to input namei data on failure.
2427 */
2428static int link_path_walk(const char *name, struct nameidata *nd)
2429{
2430 int depth = 0; // depth <= nd->depth
2431 int err;
2432
2433 nd->last_type = LAST_ROOT;
2434 nd->flags |= LOOKUP_PARENT;
2435 if (IS_ERR(ptr: name))
2436 return PTR_ERR(ptr: name);
2437 if (*name == '/') {
2438 do {
2439 name++;
2440 } while (unlikely(*name == '/'));
2441 }
2442 if (unlikely(!*name)) {
2443 nd->dir_mode = 0; // short-circuit the 'hardening' idiocy
2444 return 0;
2445 }
2446
2447 /* At this point we know we have a real path component. */
2448 for(;;) {
2449 struct mnt_idmap *idmap;
2450 const char *link;
2451 unsigned long lastword;
2452
2453 idmap = mnt_idmap(mnt: nd->path.mnt);
2454 err = may_lookup(idmap, nd);
2455 if (unlikely(err))
2456 return err;
2457
2458 nd->last.name = name;
2459 name = hash_name(nd, name, lastword: &lastword);
2460
2461 switch(lastword) {
2462 case LAST_WORD_IS_DOTDOT:
2463 nd->last_type = LAST_DOTDOT;
2464 nd->state |= ND_JUMPED;
2465 break;
2466
2467 case LAST_WORD_IS_DOT:
2468 nd->last_type = LAST_DOT;
2469 break;
2470
2471 default:
2472 nd->last_type = LAST_NORM;
2473 nd->state &= ~ND_JUMPED;
2474
2475 struct dentry *parent = nd->path.dentry;
2476 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2477 err = parent->d_op->d_hash(parent, &nd->last);
2478 if (err < 0)
2479 return err;
2480 }
2481 }
2482
2483 if (!*name)
2484 goto OK;
2485 /*
2486 * If it wasn't NUL, we know it was '/'. Skip that
2487 * slash, and continue until no more slashes.
2488 */
2489 do {
2490 name++;
2491 } while (unlikely(*name == '/'));
2492 if (unlikely(!*name)) {
2493OK:
2494 /* pathname or trailing symlink, done */
2495 if (!depth) {
2496 nd->dir_vfsuid = i_uid_into_vfsuid(idmap, inode: nd->inode);
2497 nd->dir_mode = nd->inode->i_mode;
2498 nd->flags &= ~LOOKUP_PARENT;
2499 return 0;
2500 }
2501 /* last component of nested symlink */
2502 name = nd->stack[--depth].name;
2503 link = walk_component(nd, flags: 0);
2504 } else {
2505 /* not the last component */
2506 link = walk_component(nd, flags: WALK_MORE);
2507 }
2508 if (unlikely(link)) {
2509 if (IS_ERR(ptr: link))
2510 return PTR_ERR(ptr: link);
2511 /* a symlink to follow */
2512 nd->stack[depth++].name = name;
2513 name = link;
2514 continue;
2515 }
2516 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2517 if (nd->flags & LOOKUP_RCU) {
2518 if (!try_to_unlazy(nd))
2519 return -ECHILD;
2520 }
2521 return -ENOTDIR;
2522 }
2523 }
2524}
2525
2526/* must be paired with terminate_walk() */
2527static const char *path_init(struct nameidata *nd, unsigned flags)
2528{
2529 int error;
2530 const char *s = nd->pathname;
2531
2532 /* LOOKUP_CACHED requires RCU, ask caller to retry */
2533 if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED)
2534 return ERR_PTR(error: -EAGAIN);
2535
2536 if (!*s)
2537 flags &= ~LOOKUP_RCU;
2538 if (flags & LOOKUP_RCU)
2539 rcu_read_lock();
2540 else
2541 nd->seq = nd->next_seq = 0;
2542
2543 nd->flags = flags;
2544 nd->state |= ND_JUMPED;
2545
2546 nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2547 nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2548 smp_rmb();
2549
2550 if (nd->state & ND_ROOT_PRESET) {
2551 struct dentry *root = nd->root.dentry;
2552 struct inode *inode = root->d_inode;
2553 if (*s && unlikely(!d_can_lookup(root)))
2554 return ERR_PTR(error: -ENOTDIR);
2555 nd->path = nd->root;
2556 nd->inode = inode;
2557 if (flags & LOOKUP_RCU) {
2558 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2559 nd->root_seq = nd->seq;
2560 } else {
2561 path_get(&nd->path);
2562 }
2563 return s;
2564 }
2565
2566 nd->root.mnt = NULL;
2567
2568 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2569 if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2570 error = nd_jump_root(nd);
2571 if (unlikely(error))
2572 return ERR_PTR(error);
2573 return s;
2574 }
2575
2576 /* Relative pathname -- get the starting-point it is relative to. */
2577 if (nd->dfd == AT_FDCWD) {
2578 if (flags & LOOKUP_RCU) {
2579 struct fs_struct *fs = current->fs;
2580 unsigned seq;
2581
2582 do {
2583 seq = read_seqcount_begin(&fs->seq);
2584 nd->path = fs->pwd;
2585 nd->inode = nd->path.dentry->d_inode;
2586 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2587 } while (read_seqcount_retry(&fs->seq, seq));
2588 } else {
2589 get_fs_pwd(current->fs, pwd: &nd->path);
2590 nd->inode = nd->path.dentry->d_inode;
2591 }
2592 } else {
2593 /* Caller must check execute permissions on the starting path component */
2594 CLASS(fd_raw, f)(fd: nd->dfd);
2595 struct dentry *dentry;
2596
2597 if (fd_empty(f))
2598 return ERR_PTR(error: -EBADF);
2599
2600 if (flags & LOOKUP_LINKAT_EMPTY) {
2601 if (fd_file(f)->f_cred != current_cred() &&
2602 !ns_capable(fd_file(f)->f_cred->user_ns, CAP_DAC_READ_SEARCH))
2603 return ERR_PTR(error: -ENOENT);
2604 }
2605
2606 dentry = fd_file(f)->f_path.dentry;
2607
2608 if (*s && unlikely(!d_can_lookup(dentry)))
2609 return ERR_PTR(error: -ENOTDIR);
2610
2611 nd->path = fd_file(f)->f_path;
2612 if (flags & LOOKUP_RCU) {
2613 nd->inode = nd->path.dentry->d_inode;
2614 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2615 } else {
2616 path_get(&nd->path);
2617 nd->inode = nd->path.dentry->d_inode;
2618 }
2619 }
2620
2621 /* For scoped-lookups we need to set the root to the dirfd as well. */
2622 if (flags & LOOKUP_IS_SCOPED) {
2623 nd->root = nd->path;
2624 if (flags & LOOKUP_RCU) {
2625 nd->root_seq = nd->seq;
2626 } else {
2627 path_get(&nd->root);
2628 nd->state |= ND_ROOT_GRABBED;
2629 }
2630 }
2631 return s;
2632}
2633
2634static inline const char *lookup_last(struct nameidata *nd)
2635{
2636 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2637 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2638
2639 return walk_component(nd, flags: WALK_TRAILING);
2640}
2641
2642static int handle_lookup_down(struct nameidata *nd)
2643{
2644 if (!(nd->flags & LOOKUP_RCU))
2645 dget(dentry: nd->path.dentry);
2646 nd->next_seq = nd->seq;
2647 return PTR_ERR(ptr: step_into(nd, flags: WALK_NOFOLLOW, dentry: nd->path.dentry));
2648}
2649
2650/* Returns 0 and nd will be valid on success; Returns error, otherwise. */
2651static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2652{
2653 const char *s = path_init(nd, flags);
2654 int err;
2655
2656 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(ptr: s)) {
2657 err = handle_lookup_down(nd);
2658 if (unlikely(err < 0))
2659 s = ERR_PTR(error: err);
2660 }
2661
2662 while (!(err = link_path_walk(name: s, nd)) &&
2663 (s = lookup_last(nd)) != NULL)
2664 ;
2665 if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2666 err = handle_lookup_down(nd);
2667 nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please...
2668 }
2669 if (!err)
2670 err = complete_walk(nd);
2671
2672 if (!err && nd->flags & LOOKUP_DIRECTORY)
2673 if (!d_can_lookup(dentry: nd->path.dentry))
2674 err = -ENOTDIR;
2675 if (!err) {
2676 *path = nd->path;
2677 nd->path.mnt = NULL;
2678 nd->path.dentry = NULL;
2679 }
2680 terminate_walk(nd);
2681 return err;
2682}
2683
2684int filename_lookup(int dfd, struct filename *name, unsigned flags,
2685 struct path *path, struct path *root)
2686{
2687 int retval;
2688 struct nameidata nd;
2689 if (IS_ERR(ptr: name))
2690 return PTR_ERR(ptr: name);
2691 set_nameidata(p: &nd, dfd, name, root);
2692 retval = path_lookupat(nd: &nd, flags: flags | LOOKUP_RCU, path);
2693 if (unlikely(retval == -ECHILD))
2694 retval = path_lookupat(nd: &nd, flags, path);
2695 if (unlikely(retval == -ESTALE))
2696 retval = path_lookupat(nd: &nd, flags: flags | LOOKUP_REVAL, path);
2697
2698 if (likely(!retval))
2699 audit_inode(name, dentry: path->dentry,
2700 aflags: flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2701 restore_nameidata();
2702 return retval;
2703}
2704
2705/* Returns 0 and nd will be valid on success; Returns error, otherwise. */
2706static int path_parentat(struct nameidata *nd, unsigned flags,
2707 struct path *parent)
2708{
2709 const char *s = path_init(nd, flags);
2710 int err = link_path_walk(name: s, nd);
2711 if (!err)
2712 err = complete_walk(nd);
2713 if (!err) {
2714 *parent = nd->path;
2715 nd->path.mnt = NULL;
2716 nd->path.dentry = NULL;
2717 }
2718 terminate_walk(nd);
2719 return err;
2720}
2721
2722/* Note: this does not consume "name" */
2723static int __filename_parentat(int dfd, struct filename *name,
2724 unsigned int flags, struct path *parent,
2725 struct qstr *last, int *type,
2726 const struct path *root)
2727{
2728 int retval;
2729 struct nameidata nd;
2730
2731 if (IS_ERR(ptr: name))
2732 return PTR_ERR(ptr: name);
2733 set_nameidata(p: &nd, dfd, name, root);
2734 retval = path_parentat(nd: &nd, flags: flags | LOOKUP_RCU, parent);
2735 if (unlikely(retval == -ECHILD))
2736 retval = path_parentat(nd: &nd, flags, parent);
2737 if (unlikely(retval == -ESTALE))
2738 retval = path_parentat(nd: &nd, flags: flags | LOOKUP_REVAL, parent);
2739 if (likely(!retval)) {
2740 *last = nd.last;
2741 *type = nd.last_type;
2742 audit_inode(name, dentry: parent->dentry, AUDIT_INODE_PARENT);
2743 }
2744 restore_nameidata();
2745 return retval;
2746}
2747
2748static int filename_parentat(int dfd, struct filename *name,
2749 unsigned int flags, struct path *parent,
2750 struct qstr *last, int *type)
2751{
2752 return __filename_parentat(dfd, name, flags, parent, last, type, NULL);
2753}
2754
2755/* does lookup, returns the object with parent locked */
2756static struct dentry *__kern_path_locked(int dfd, struct filename *name, struct path *path)
2757{
2758 struct path parent_path __free(path_put) = {};
2759 struct dentry *d;
2760 struct qstr last;
2761 int type, error;
2762
2763 error = filename_parentat(dfd, name, flags: 0, parent: &parent_path, last: &last, type: &type);
2764 if (error)
2765 return ERR_PTR(error);
2766 if (unlikely(type != LAST_NORM))
2767 return ERR_PTR(error: -EINVAL);
2768 inode_lock_nested(inode: parent_path.dentry->d_inode, subclass: I_MUTEX_PARENT);
2769 d = lookup_one_qstr_excl(&last, parent_path.dentry, 0);
2770 if (IS_ERR(ptr: d)) {
2771 inode_unlock(inode: parent_path.dentry->d_inode);
2772 return d;
2773 }
2774 path->dentry = no_free_ptr(parent_path.dentry);
2775 path->mnt = no_free_ptr(parent_path.mnt);
2776 return d;
2777}
2778
2779struct dentry *kern_path_locked_negative(const char *name, struct path *path)
2780{
2781 struct path parent_path __free(path_put) = {};
2782 struct filename *filename __free(putname) = getname_kernel(name);
2783 struct dentry *d;
2784 struct qstr last;
2785 int type, error;
2786
2787 error = filename_parentat(AT_FDCWD, name: filename, flags: 0, parent: &parent_path, last: &last, type: &type);
2788 if (error)
2789 return ERR_PTR(error);
2790 if (unlikely(type != LAST_NORM))
2791 return ERR_PTR(error: -EINVAL);
2792 inode_lock_nested(inode: parent_path.dentry->d_inode, subclass: I_MUTEX_PARENT);
2793 d = lookup_one_qstr_excl_raw(name: &last, base: parent_path.dentry, flags: 0);
2794 if (IS_ERR(ptr: d)) {
2795 inode_unlock(inode: parent_path.dentry->d_inode);
2796 return d;
2797 }
2798 path->dentry = no_free_ptr(parent_path.dentry);
2799 path->mnt = no_free_ptr(parent_path.mnt);
2800 return d;
2801}
2802
2803struct dentry *kern_path_locked(const char *name, struct path *path)
2804{
2805 struct filename *filename = getname_kernel(name);
2806 struct dentry *res = __kern_path_locked(AT_FDCWD, name: filename, path);
2807
2808 putname(filename);
2809 return res;
2810}
2811
2812struct dentry *user_path_locked_at(int dfd, const char __user *name, struct path *path)
2813{
2814 struct filename *filename = getname(name);
2815 struct dentry *res = __kern_path_locked(dfd, name: filename, path);
2816
2817 putname(filename);
2818 return res;
2819}
2820EXPORT_SYMBOL(user_path_locked_at);
2821
2822int kern_path(const char *name, unsigned int flags, struct path *path)
2823{
2824 struct filename *filename = getname_kernel(name);
2825 int ret = filename_lookup(AT_FDCWD, name: filename, flags, path, NULL);
2826
2827 putname(filename);
2828 return ret;
2829
2830}
2831EXPORT_SYMBOL(kern_path);
2832
2833/**
2834 * vfs_path_parent_lookup - lookup a parent path relative to a dentry-vfsmount pair
2835 * @filename: filename structure
2836 * @flags: lookup flags
2837 * @parent: pointer to struct path to fill
2838 * @last: last component
2839 * @type: type of the last component
2840 * @root: pointer to struct path of the base directory
2841 */
2842int vfs_path_parent_lookup(struct filename *filename, unsigned int flags,
2843 struct path *parent, struct qstr *last, int *type,
2844 const struct path *root)
2845{
2846 return __filename_parentat(AT_FDCWD, name: filename, flags, parent, last,
2847 type, root);
2848}
2849EXPORT_SYMBOL(vfs_path_parent_lookup);
2850
2851/**
2852 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2853 * @dentry: pointer to dentry of the base directory
2854 * @mnt: pointer to vfs mount of the base directory
2855 * @name: pointer to file name
2856 * @flags: lookup flags
2857 * @path: pointer to struct path to fill
2858 */
2859int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2860 const char *name, unsigned int flags,
2861 struct path *path)
2862{
2863 struct filename *filename;
2864 struct path root = {.mnt = mnt, .dentry = dentry};
2865 int ret;
2866
2867 filename = getname_kernel(name);
2868 /* the first argument of filename_lookup() is ignored with root */
2869 ret = filename_lookup(AT_FDCWD, name: filename, flags, path, root: &root);
2870 putname(filename);
2871 return ret;
2872}
2873EXPORT_SYMBOL(vfs_path_lookup);
2874
2875static int lookup_noperm_common(struct qstr *qname, struct dentry *base)
2876{
2877 const char *name = qname->name;
2878 u32 len = qname->len;
2879
2880 qname->hash = full_name_hash(base, name, len);
2881 if (!len)
2882 return -EACCES;
2883
2884 if (is_dot_dotdot(name, len))
2885 return -EACCES;
2886
2887 while (len--) {
2888 unsigned int c = *(const unsigned char *)name++;
2889 if (c == '/' || c == '\0')
2890 return -EACCES;
2891 }
2892 /*
2893 * See if the low-level filesystem might want
2894 * to use its own hash..
2895 */
2896 if (base->d_flags & DCACHE_OP_HASH) {
2897 int err = base->d_op->d_hash(base, qname);
2898 if (err < 0)
2899 return err;
2900 }
2901 return 0;
2902}
2903
2904static int lookup_one_common(struct mnt_idmap *idmap,
2905 struct qstr *qname, struct dentry *base)
2906{
2907 int err;
2908 err = lookup_noperm_common(qname, base);
2909 if (err < 0)
2910 return err;
2911 return inode_permission(idmap, base->d_inode, MAY_EXEC);
2912}
2913
2914/**
2915 * try_lookup_noperm - filesystem helper to lookup single pathname component
2916 * @name: qstr storing pathname component to lookup
2917 * @base: base directory to lookup from
2918 *
2919 * Look up a dentry by name in the dcache, returning NULL if it does not
2920 * currently exist. The function does not try to create a dentry.
2921 *
2922 * Note that this routine is purely a helper for filesystem usage and should
2923 * not be called by generic code. It does no permission checking.
2924 *
2925 * No locks need be held - only a counted reference to @base is needed.
2926 *
2927 */
2928struct dentry *try_lookup_noperm(struct qstr *name, struct dentry *base)
2929{
2930 int err;
2931
2932 err = lookup_noperm_common(qname: name, base);
2933 if (err)
2934 return ERR_PTR(error: err);
2935
2936 return lookup_dcache(name, dir: base, flags: 0);
2937}
2938EXPORT_SYMBOL(try_lookup_noperm);
2939
2940/**
2941 * lookup_noperm - filesystem helper to lookup single pathname component
2942 * @name: qstr storing pathname component to lookup
2943 * @base: base directory to lookup from
2944 *
2945 * Note that this routine is purely a helper for filesystem usage and should
2946 * not be called by generic code. It does no permission checking.
2947 *
2948 * The caller must hold base->i_mutex.
2949 */
2950struct dentry *lookup_noperm(struct qstr *name, struct dentry *base)
2951{
2952 struct dentry *dentry;
2953 int err;
2954
2955 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2956
2957 err = lookup_noperm_common(qname: name, base);
2958 if (err)
2959 return ERR_PTR(error: err);
2960
2961 dentry = lookup_dcache(name, dir: base, flags: 0);
2962 return dentry ? dentry : __lookup_slow(name, dir: base, flags: 0);
2963}
2964EXPORT_SYMBOL(lookup_noperm);
2965
2966/**
2967 * lookup_one - lookup single pathname component
2968 * @idmap: idmap of the mount the lookup is performed from
2969 * @name: qstr holding pathname component to lookup
2970 * @base: base directory to lookup from
2971 *
2972 * This can be used for in-kernel filesystem clients such as file servers.
2973 *
2974 * The caller must hold base->i_mutex.
2975 */
2976struct dentry *lookup_one(struct mnt_idmap *idmap, struct qstr *name,
2977 struct dentry *base)
2978{
2979 struct dentry *dentry;
2980 int err;
2981
2982 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2983
2984 err = lookup_one_common(idmap, qname: name, base);
2985 if (err)
2986 return ERR_PTR(error: err);
2987
2988 dentry = lookup_dcache(name, dir: base, flags: 0);
2989 return dentry ? dentry : __lookup_slow(name, dir: base, flags: 0);
2990}
2991EXPORT_SYMBOL(lookup_one);
2992
2993/**
2994 * lookup_one_unlocked - lookup single pathname component
2995 * @idmap: idmap of the mount the lookup is performed from
2996 * @name: qstr olding pathname component to lookup
2997 * @base: base directory to lookup from
2998 *
2999 * This can be used for in-kernel filesystem clients such as file servers.
3000 *
3001 * Unlike lookup_one, it should be called without the parent
3002 * i_rwsem held, and will take the i_rwsem itself if necessary.
3003 */
3004struct dentry *lookup_one_unlocked(struct mnt_idmap *idmap, struct qstr *name,
3005 struct dentry *base)
3006{
3007 int err;
3008 struct dentry *ret;
3009
3010 err = lookup_one_common(idmap, qname: name, base);
3011 if (err)
3012 return ERR_PTR(error: err);
3013
3014 ret = lookup_dcache(name, dir: base, flags: 0);
3015 if (!ret)
3016 ret = lookup_slow(name, dir: base, flags: 0);
3017 return ret;
3018}
3019EXPORT_SYMBOL(lookup_one_unlocked);
3020
3021/**
3022 * lookup_one_positive_unlocked - lookup single pathname component
3023 * @idmap: idmap of the mount the lookup is performed from
3024 * @name: qstr holding pathname component to lookup
3025 * @base: base directory to lookup from
3026 *
3027 * This helper will yield ERR_PTR(-ENOENT) on negatives. The helper returns
3028 * known positive or ERR_PTR(). This is what most of the users want.
3029 *
3030 * Note that pinned negative with unlocked parent _can_ become positive at any
3031 * time, so callers of lookup_one_unlocked() need to be very careful; pinned
3032 * positives have >d_inode stable, so this one avoids such problems.
3033 *
3034 * This can be used for in-kernel filesystem clients such as file servers.
3035 *
3036 * The helper should be called without i_rwsem held.
3037 */
3038struct dentry *lookup_one_positive_unlocked(struct mnt_idmap *idmap,
3039 struct qstr *name,
3040 struct dentry *base)
3041{
3042 struct dentry *ret = lookup_one_unlocked(idmap, name, base);
3043
3044 if (!IS_ERR(ptr: ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
3045 dput(ret);
3046 ret = ERR_PTR(error: -ENOENT);
3047 }
3048 return ret;
3049}
3050EXPORT_SYMBOL(lookup_one_positive_unlocked);
3051
3052/**
3053 * lookup_noperm_unlocked - filesystem helper to lookup single pathname component
3054 * @name: pathname component to lookup
3055 * @base: base directory to lookup from
3056 *
3057 * Note that this routine is purely a helper for filesystem usage and should
3058 * not be called by generic code. It does no permission checking.
3059 *
3060 * Unlike lookup_noperm, it should be called without the parent
3061 * i_rwsem held, and will take the i_rwsem itself if necessary.
3062 */
3063struct dentry *lookup_noperm_unlocked(struct qstr *name, struct dentry *base)
3064{
3065 struct dentry *ret;
3066
3067 ret = try_lookup_noperm(name, base);
3068 if (!ret)
3069 ret = lookup_slow(name, dir: base, flags: 0);
3070 return ret;
3071}
3072EXPORT_SYMBOL(lookup_noperm_unlocked);
3073
3074/*
3075 * Like lookup_noperm_unlocked(), except that it yields ERR_PTR(-ENOENT)
3076 * on negatives. Returns known positive or ERR_PTR(); that's what
3077 * most of the users want. Note that pinned negative with unlocked parent
3078 * _can_ become positive at any time, so callers of lookup_noperm_unlocked()
3079 * need to be very careful; pinned positives have ->d_inode stable, so
3080 * this one avoids such problems.
3081 */
3082struct dentry *lookup_noperm_positive_unlocked(struct qstr *name,
3083 struct dentry *base)
3084{
3085 struct dentry *ret;
3086
3087 ret = lookup_noperm_unlocked(name, base);
3088 if (!IS_ERR(ptr: ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
3089 dput(ret);
3090 ret = ERR_PTR(error: -ENOENT);
3091 }
3092 return ret;
3093}
3094EXPORT_SYMBOL(lookup_noperm_positive_unlocked);
3095
3096#ifdef CONFIG_UNIX98_PTYS
3097int path_pts(struct path *path)
3098{
3099 /* Find something mounted on "pts" in the same directory as
3100 * the input path.
3101 */
3102 struct dentry *parent = dget_parent(dentry: path->dentry);
3103 struct dentry *child;
3104 struct qstr this = QSTR_INIT("pts", 3);
3105
3106 if (unlikely(!path_connected(path->mnt, parent))) {
3107 dput(parent);
3108 return -ENOENT;
3109 }
3110 dput(path->dentry);
3111 path->dentry = parent;
3112 child = d_hash_and_lookup(parent, &this);
3113 if (IS_ERR_OR_NULL(ptr: child))
3114 return -ENOENT;
3115
3116 path->dentry = child;
3117 dput(parent);
3118 follow_down(path, 0);
3119 return 0;
3120}
3121#endif
3122
3123int user_path_at(int dfd, const char __user *name, unsigned flags,
3124 struct path *path)
3125{
3126 struct filename *filename = getname_flags(filename: name, flags);
3127 int ret = filename_lookup(dfd, name: filename, flags, path, NULL);
3128
3129 putname(filename);
3130 return ret;
3131}
3132EXPORT_SYMBOL(user_path_at);
3133
3134int __check_sticky(struct mnt_idmap *idmap, struct inode *dir,
3135 struct inode *inode)
3136{
3137 kuid_t fsuid = current_fsuid();
3138
3139 if (vfsuid_eq_kuid(vfsuid: i_uid_into_vfsuid(idmap, inode), kuid: fsuid))
3140 return 0;
3141 if (vfsuid_eq_kuid(vfsuid: i_uid_into_vfsuid(idmap, inode: dir), kuid: fsuid))
3142 return 0;
3143 return !capable_wrt_inode_uidgid(idmap, inode, CAP_FOWNER);
3144}
3145EXPORT_SYMBOL(__check_sticky);
3146
3147/*
3148 * Check whether we can remove a link victim from directory dir, check
3149 * whether the type of victim is right.
3150 * 1. We can't do it if dir is read-only (done in permission())
3151 * 2. We should have write and exec permissions on dir
3152 * 3. We can't remove anything from append-only dir
3153 * 4. We can't do anything with immutable dir (done in permission())
3154 * 5. If the sticky bit on dir is set we should either
3155 * a. be owner of dir, or
3156 * b. be owner of victim, or
3157 * c. have CAP_FOWNER capability
3158 * 6. If the victim is append-only or immutable we can't do antyhing with
3159 * links pointing to it.
3160 * 7. If the victim has an unknown uid or gid we can't change the inode.
3161 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
3162 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
3163 * 10. We can't remove a root or mountpoint.
3164 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
3165 * nfs_async_unlink().
3166 */
3167static int may_delete(struct mnt_idmap *idmap, struct inode *dir,
3168 struct dentry *victim, bool isdir)
3169{
3170 struct inode *inode = d_backing_inode(upper: victim);
3171 int error;
3172
3173 if (d_is_negative(dentry: victim))
3174 return -ENOENT;
3175 BUG_ON(!inode);
3176
3177 BUG_ON(victim->d_parent->d_inode != dir);
3178
3179 /* Inode writeback is not safe when the uid or gid are invalid. */
3180 if (!vfsuid_valid(uid: i_uid_into_vfsuid(idmap, inode)) ||
3181 !vfsgid_valid(gid: i_gid_into_vfsgid(idmap, inode)))
3182 return -EOVERFLOW;
3183
3184 audit_inode_child(parent: dir, dentry: victim, AUDIT_TYPE_CHILD_DELETE);
3185
3186 error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
3187 if (error)
3188 return error;
3189 if (IS_APPEND(dir))
3190 return -EPERM;
3191
3192 if (check_sticky(idmap, dir, inode) || IS_APPEND(inode) ||
3193 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) ||
3194 HAS_UNMAPPED_ID(idmap, inode))
3195 return -EPERM;
3196 if (isdir) {
3197 if (!d_is_dir(dentry: victim))
3198 return -ENOTDIR;
3199 if (IS_ROOT(victim))
3200 return -EBUSY;
3201 } else if (d_is_dir(dentry: victim))
3202 return -EISDIR;
3203 if (IS_DEADDIR(dir))
3204 return -ENOENT;
3205 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
3206 return -EBUSY;
3207 return 0;
3208}
3209
3210/* Check whether we can create an object with dentry child in directory
3211 * dir.
3212 * 1. We can't do it if child already exists (open has special treatment for
3213 * this case, but since we are inlined it's OK)
3214 * 2. We can't do it if dir is read-only (done in permission())
3215 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
3216 * 4. We should have write and exec permissions on dir
3217 * 5. We can't do it if dir is immutable (done in permission())
3218 */
3219static inline int may_create(struct mnt_idmap *idmap,
3220 struct inode *dir, struct dentry *child)
3221{
3222 audit_inode_child(parent: dir, dentry: child, AUDIT_TYPE_CHILD_CREATE);
3223 if (child->d_inode)
3224 return -EEXIST;
3225 if (IS_DEADDIR(dir))
3226 return -ENOENT;
3227 if (!fsuidgid_has_mapping(sb: dir->i_sb, idmap))
3228 return -EOVERFLOW;
3229
3230 return inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
3231}
3232
3233// p1 != p2, both are on the same filesystem, ->s_vfs_rename_mutex is held
3234static struct dentry *lock_two_directories(struct dentry *p1, struct dentry *p2)
3235{
3236 struct dentry *p = p1, *q = p2, *r;
3237
3238 while ((r = p->d_parent) != p2 && r != p)
3239 p = r;
3240 if (r == p2) {
3241 // p is a child of p2 and an ancestor of p1 or p1 itself
3242 inode_lock_nested(inode: p2->d_inode, subclass: I_MUTEX_PARENT);
3243 inode_lock_nested(inode: p1->d_inode, subclass: I_MUTEX_PARENT2);
3244 return p;
3245 }
3246 // p is the root of connected component that contains p1
3247 // p2 does not occur on the path from p to p1
3248 while ((r = q->d_parent) != p1 && r != p && r != q)
3249 q = r;
3250 if (r == p1) {
3251 // q is a child of p1 and an ancestor of p2 or p2 itself
3252 inode_lock_nested(inode: p1->d_inode, subclass: I_MUTEX_PARENT);
3253 inode_lock_nested(inode: p2->d_inode, subclass: I_MUTEX_PARENT2);
3254 return q;
3255 } else if (likely(r == p)) {
3256 // both p2 and p1 are descendents of p
3257 inode_lock_nested(inode: p1->d_inode, subclass: I_MUTEX_PARENT);
3258 inode_lock_nested(inode: p2->d_inode, subclass: I_MUTEX_PARENT2);
3259 return NULL;
3260 } else { // no common ancestor at the time we'd been called
3261 mutex_unlock(lock: &p1->d_sb->s_vfs_rename_mutex);
3262 return ERR_PTR(error: -EXDEV);
3263 }
3264}
3265
3266/*
3267 * p1 and p2 should be directories on the same fs.
3268 */
3269struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
3270{
3271 if (p1 == p2) {
3272 inode_lock_nested(inode: p1->d_inode, subclass: I_MUTEX_PARENT);
3273 return NULL;
3274 }
3275
3276 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
3277 return lock_two_directories(p1, p2);
3278}
3279EXPORT_SYMBOL(lock_rename);
3280
3281/*
3282 * c1 and p2 should be on the same fs.
3283 */
3284struct dentry *lock_rename_child(struct dentry *c1, struct dentry *p2)
3285{
3286 if (READ_ONCE(c1->d_parent) == p2) {
3287 /*
3288 * hopefully won't need to touch ->s_vfs_rename_mutex at all.
3289 */
3290 inode_lock_nested(inode: p2->d_inode, subclass: I_MUTEX_PARENT);
3291 /*
3292 * now that p2 is locked, nobody can move in or out of it,
3293 * so the test below is safe.
3294 */
3295 if (likely(c1->d_parent == p2))
3296 return NULL;
3297
3298 /*
3299 * c1 got moved out of p2 while we'd been taking locks;
3300 * unlock and fall back to slow case.
3301 */
3302 inode_unlock(inode: p2->d_inode);
3303 }
3304
3305 mutex_lock(&c1->d_sb->s_vfs_rename_mutex);
3306 /*
3307 * nobody can move out of any directories on this fs.
3308 */
3309 if (likely(c1->d_parent != p2))
3310 return lock_two_directories(p1: c1->d_parent, p2);
3311
3312 /*
3313 * c1 got moved into p2 while we were taking locks;
3314 * we need p2 locked and ->s_vfs_rename_mutex unlocked,
3315 * for consistency with lock_rename().
3316 */
3317 inode_lock_nested(inode: p2->d_inode, subclass: I_MUTEX_PARENT);
3318 mutex_unlock(lock: &c1->d_sb->s_vfs_rename_mutex);
3319 return NULL;
3320}
3321EXPORT_SYMBOL(lock_rename_child);
3322
3323void unlock_rename(struct dentry *p1, struct dentry *p2)
3324{
3325 inode_unlock(inode: p1->d_inode);
3326 if (p1 != p2) {
3327 inode_unlock(inode: p2->d_inode);
3328 mutex_unlock(lock: &p1->d_sb->s_vfs_rename_mutex);
3329 }
3330}
3331EXPORT_SYMBOL(unlock_rename);
3332
3333/**
3334 * vfs_prepare_mode - prepare the mode to be used for a new inode
3335 * @idmap: idmap of the mount the inode was found from
3336 * @dir: parent directory of the new inode
3337 * @mode: mode of the new inode
3338 * @mask_perms: allowed permission by the vfs
3339 * @type: type of file to be created
3340 *
3341 * This helper consolidates and enforces vfs restrictions on the @mode of a new
3342 * object to be created.
3343 *
3344 * Umask stripping depends on whether the filesystem supports POSIX ACLs (see
3345 * the kernel documentation for mode_strip_umask()). Moving umask stripping
3346 * after setgid stripping allows the same ordering for both non-POSIX ACL and
3347 * POSIX ACL supporting filesystems.
3348 *
3349 * Note that it's currently valid for @type to be 0 if a directory is created.
3350 * Filesystems raise that flag individually and we need to check whether each
3351 * filesystem can deal with receiving S_IFDIR from the vfs before we enforce a
3352 * non-zero type.
3353 *
3354 * Returns: mode to be passed to the filesystem
3355 */
3356static inline umode_t vfs_prepare_mode(struct mnt_idmap *idmap,
3357 const struct inode *dir, umode_t mode,
3358 umode_t mask_perms, umode_t type)
3359{
3360 mode = mode_strip_sgid(idmap, dir, mode);
3361 mode = mode_strip_umask(dir, mode);
3362
3363 /*
3364 * Apply the vfs mandated allowed permission mask and set the type of
3365 * file to be created before we call into the filesystem.
3366 */
3367 mode &= (mask_perms & ~S_IFMT);
3368 mode |= (type & S_IFMT);
3369
3370 return mode;
3371}
3372
3373/**
3374 * vfs_create - create new file
3375 * @idmap: idmap of the mount the inode was found from
3376 * @dir: inode of the parent directory
3377 * @dentry: dentry of the child file
3378 * @mode: mode of the child file
3379 * @want_excl: whether the file must not yet exist
3380 *
3381 * Create a new file.
3382 *
3383 * If the inode has been found through an idmapped mount the idmap of
3384 * the vfsmount must be passed through @idmap. This function will then take
3385 * care to map the inode according to @idmap before checking permissions.
3386 * On non-idmapped mounts or if permission checking is to be performed on the
3387 * raw inode simply pass @nop_mnt_idmap.
3388 */
3389int vfs_create(struct mnt_idmap *idmap, struct inode *dir,
3390 struct dentry *dentry, umode_t mode, bool want_excl)
3391{
3392 int error;
3393
3394 error = may_create(idmap, dir, child: dentry);
3395 if (error)
3396 return error;
3397
3398 if (!dir->i_op->create)
3399 return -EACCES; /* shouldn't it be ENOSYS? */
3400
3401 mode = vfs_prepare_mode(idmap, dir, mode, S_IALLUGO, S_IFREG);
3402 error = security_inode_create(dir, dentry, mode);
3403 if (error)
3404 return error;
3405 error = dir->i_op->create(idmap, dir, dentry, mode, want_excl);
3406 if (!error)
3407 fsnotify_create(dir, dentry);
3408 return error;
3409}
3410EXPORT_SYMBOL(vfs_create);
3411
3412int vfs_mkobj(struct dentry *dentry, umode_t mode,
3413 int (*f)(struct dentry *, umode_t, void *),
3414 void *arg)
3415{
3416 struct inode *dir = dentry->d_parent->d_inode;
3417 int error = may_create(idmap: &nop_mnt_idmap, dir, child: dentry);
3418 if (error)
3419 return error;
3420
3421 mode &= S_IALLUGO;
3422 mode |= S_IFREG;
3423 error = security_inode_create(dir, dentry, mode);
3424 if (error)
3425 return error;
3426 error = f(dentry, mode, arg);
3427 if (!error)
3428 fsnotify_create(dir, dentry);
3429 return error;
3430}
3431EXPORT_SYMBOL(vfs_mkobj);
3432
3433bool may_open_dev(const struct path *path)
3434{
3435 return !(path->mnt->mnt_flags & MNT_NODEV) &&
3436 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
3437}
3438
3439static int may_open(struct mnt_idmap *idmap, const struct path *path,
3440 int acc_mode, int flag)
3441{
3442 struct dentry *dentry = path->dentry;
3443 struct inode *inode = dentry->d_inode;
3444 int error;
3445
3446 if (!inode)
3447 return -ENOENT;
3448
3449 switch (inode->i_mode & S_IFMT) {
3450 case S_IFLNK:
3451 return -ELOOP;
3452 case S_IFDIR:
3453 if (acc_mode & MAY_WRITE)
3454 return -EISDIR;
3455 if (acc_mode & MAY_EXEC)
3456 return -EACCES;
3457 break;
3458 case S_IFBLK:
3459 case S_IFCHR:
3460 if (!may_open_dev(path))
3461 return -EACCES;
3462 fallthrough;
3463 case S_IFIFO:
3464 case S_IFSOCK:
3465 if (acc_mode & MAY_EXEC)
3466 return -EACCES;
3467 flag &= ~O_TRUNC;
3468 break;
3469 case S_IFREG:
3470 if ((acc_mode & MAY_EXEC) && path_noexec(path))
3471 return -EACCES;
3472 break;
3473 default:
3474 VFS_BUG_ON_INODE(1, inode);
3475 }
3476
3477 error = inode_permission(idmap, inode, MAY_OPEN | acc_mode);
3478 if (error)
3479 return error;
3480
3481 /*
3482 * An append-only file must be opened in append mode for writing.
3483 */
3484 if (IS_APPEND(inode)) {
3485 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
3486 return -EPERM;
3487 if (flag & O_TRUNC)
3488 return -EPERM;
3489 }
3490
3491 /* O_NOATIME can only be set by the owner or superuser */
3492 if (flag & O_NOATIME && !inode_owner_or_capable(idmap, inode))
3493 return -EPERM;
3494
3495 return 0;
3496}
3497
3498static int handle_truncate(struct mnt_idmap *idmap, struct file *filp)
3499{
3500 const struct path *path = &filp->f_path;
3501 struct inode *inode = path->dentry->d_inode;
3502 int error = get_write_access(inode);
3503 if (error)
3504 return error;
3505
3506 error = security_file_truncate(file: filp);
3507 if (!error) {
3508 error = do_truncate(idmap, path->dentry, start: 0,
3509 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3510 filp);
3511 }
3512 put_write_access(inode);
3513 return error;
3514}
3515
3516static inline int open_to_namei_flags(int flag)
3517{
3518 if ((flag & O_ACCMODE) == 3)
3519 flag--;
3520 return flag;
3521}
3522
3523static int may_o_create(struct mnt_idmap *idmap,
3524 const struct path *dir, struct dentry *dentry,
3525 umode_t mode)
3526{
3527 int error = security_path_mknod(dir, dentry, mode, dev: 0);
3528 if (error)
3529 return error;
3530
3531 if (!fsuidgid_has_mapping(sb: dir->dentry->d_sb, idmap))
3532 return -EOVERFLOW;
3533
3534 error = inode_permission(idmap, dir->dentry->d_inode,
3535 MAY_WRITE | MAY_EXEC);
3536 if (error)
3537 return error;
3538
3539 return security_inode_create(dir: dir->dentry->d_inode, dentry, mode);
3540}
3541
3542/*
3543 * Attempt to atomically look up, create and open a file from a negative
3544 * dentry.
3545 *
3546 * Returns 0 if successful. The file will have been created and attached to
3547 * @file by the filesystem calling finish_open().
3548 *
3549 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3550 * be set. The caller will need to perform the open themselves. @path will
3551 * have been updated to point to the new dentry. This may be negative.
3552 *
3553 * Returns an error code otherwise.
3554 */
3555static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
3556 struct file *file,
3557 int open_flag, umode_t mode)
3558{
3559 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3560 struct inode *dir = nd->path.dentry->d_inode;
3561 int error;
3562
3563 if (nd->flags & LOOKUP_DIRECTORY)
3564 open_flag |= O_DIRECTORY;
3565
3566 file->f_path.dentry = DENTRY_NOT_SET;
3567 file->f_path.mnt = nd->path.mnt;
3568 error = dir->i_op->atomic_open(dir, dentry, file,
3569 open_to_namei_flags(flag: open_flag), mode);
3570 d_lookup_done(dentry);
3571 if (!error) {
3572 if (file->f_mode & FMODE_OPENED) {
3573 if (unlikely(dentry != file->f_path.dentry)) {
3574 dput(dentry);
3575 dentry = dget(dentry: file->f_path.dentry);
3576 }
3577 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3578 error = -EIO;
3579 } else {
3580 if (file->f_path.dentry) {
3581 dput(dentry);
3582 dentry = file->f_path.dentry;
3583 }
3584 if (unlikely(d_is_negative(dentry)))
3585 error = -ENOENT;
3586 }
3587 }
3588 if (error) {
3589 dput(dentry);
3590 dentry = ERR_PTR(error);
3591 }
3592 return dentry;
3593}
3594
3595/*
3596 * Look up and maybe create and open the last component.
3597 *
3598 * Must be called with parent locked (exclusive in O_CREAT case).
3599 *
3600 * Returns 0 on success, that is, if
3601 * the file was successfully atomically created (if necessary) and opened, or
3602 * the file was not completely opened at this time, though lookups and
3603 * creations were performed.
3604 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3605 * In the latter case dentry returned in @path might be negative if O_CREAT
3606 * hadn't been specified.
3607 *
3608 * An error code is returned on failure.
3609 */
3610static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3611 const struct open_flags *op,
3612 bool got_write)
3613{
3614 struct mnt_idmap *idmap;
3615 struct dentry *dir = nd->path.dentry;
3616 struct inode *dir_inode = dir->d_inode;
3617 int open_flag = op->open_flag;
3618 struct dentry *dentry;
3619 int error, create_error = 0;
3620 umode_t mode = op->mode;
3621 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3622
3623 if (unlikely(IS_DEADDIR(dir_inode)))
3624 return ERR_PTR(error: -ENOENT);
3625
3626 file->f_mode &= ~FMODE_CREATED;
3627 dentry = d_lookup(dir, &nd->last);
3628 for (;;) {
3629 if (!dentry) {
3630 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3631 if (IS_ERR(ptr: dentry))
3632 return dentry;
3633 }
3634 if (d_in_lookup(dentry))
3635 break;
3636
3637 error = d_revalidate(dir: dir_inode, name: &nd->last, dentry, flags: nd->flags);
3638 if (likely(error > 0))
3639 break;
3640 if (error)
3641 goto out_dput;
3642 d_invalidate(dentry);
3643 dput(dentry);
3644 dentry = NULL;
3645 }
3646 if (dentry->d_inode) {
3647 /* Cached positive dentry: will open in f_op->open */
3648 return dentry;
3649 }
3650
3651 if (open_flag & O_CREAT)
3652 audit_inode(name: nd->name, dentry: dir, AUDIT_INODE_PARENT);
3653
3654 /*
3655 * Checking write permission is tricky, bacuse we don't know if we are
3656 * going to actually need it: O_CREAT opens should work as long as the
3657 * file exists. But checking existence breaks atomicity. The trick is
3658 * to check access and if not granted clear O_CREAT from the flags.
3659 *
3660 * Another problem is returing the "right" error value (e.g. for an
3661 * O_EXCL open we want to return EEXIST not EROFS).
3662 */
3663 if (unlikely(!got_write))
3664 open_flag &= ~O_TRUNC;
3665 idmap = mnt_idmap(mnt: nd->path.mnt);
3666 if (open_flag & O_CREAT) {
3667 if (open_flag & O_EXCL)
3668 open_flag &= ~O_TRUNC;
3669 mode = vfs_prepare_mode(idmap, dir: dir->d_inode, mode, mask_perms: mode, type: mode);
3670 if (likely(got_write))
3671 create_error = may_o_create(idmap, dir: &nd->path,
3672 dentry, mode);
3673 else
3674 create_error = -EROFS;
3675 }
3676 if (create_error)
3677 open_flag &= ~O_CREAT;
3678 if (dir_inode->i_op->atomic_open) {
3679 dentry = atomic_open(nd, dentry, file, open_flag, mode);
3680 if (unlikely(create_error) && dentry == ERR_PTR(error: -ENOENT))
3681 dentry = ERR_PTR(error: create_error);
3682 return dentry;
3683 }
3684
3685 if (d_in_lookup(dentry)) {
3686 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3687 nd->flags);
3688 d_lookup_done(dentry);
3689 if (unlikely(res)) {
3690 if (IS_ERR(ptr: res)) {
3691 error = PTR_ERR(ptr: res);
3692 goto out_dput;
3693 }
3694 dput(dentry);
3695 dentry = res;
3696 }
3697 }
3698
3699 /* Negative dentry, just create the file */
3700 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3701 file->f_mode |= FMODE_CREATED;
3702 audit_inode_child(parent: dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3703 if (!dir_inode->i_op->create) {
3704 error = -EACCES;
3705 goto out_dput;
3706 }
3707
3708 error = dir_inode->i_op->create(idmap, dir_inode, dentry,
3709 mode, open_flag & O_EXCL);
3710 if (error)
3711 goto out_dput;
3712 }
3713 if (unlikely(create_error) && !dentry->d_inode) {
3714 error = create_error;
3715 goto out_dput;
3716 }
3717 return dentry;
3718
3719out_dput:
3720 dput(dentry);
3721 return ERR_PTR(error);
3722}
3723
3724static inline bool trailing_slashes(struct nameidata *nd)
3725{
3726 return (bool)nd->last.name[nd->last.len];
3727}
3728
3729static struct dentry *lookup_fast_for_open(struct nameidata *nd, int open_flag)
3730{
3731 struct dentry *dentry;
3732
3733 if (open_flag & O_CREAT) {
3734 if (trailing_slashes(nd))
3735 return ERR_PTR(error: -EISDIR);
3736
3737 /* Don't bother on an O_EXCL create */
3738 if (open_flag & O_EXCL)
3739 return NULL;
3740 }
3741
3742 if (trailing_slashes(nd))
3743 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3744
3745 dentry = lookup_fast(nd);
3746 if (IS_ERR_OR_NULL(ptr: dentry))
3747 return dentry;
3748
3749 if (open_flag & O_CREAT) {
3750 /* Discard negative dentries. Need inode_lock to do the create */
3751 if (!dentry->d_inode) {
3752 if (!(nd->flags & LOOKUP_RCU))
3753 dput(dentry);
3754 dentry = NULL;
3755 }
3756 }
3757 return dentry;
3758}
3759
3760static const char *open_last_lookups(struct nameidata *nd,
3761 struct file *file, const struct open_flags *op)
3762{
3763 struct dentry *dir = nd->path.dentry;
3764 int open_flag = op->open_flag;
3765 bool got_write = false;
3766 struct dentry *dentry;
3767 const char *res;
3768
3769 nd->flags |= op->intent;
3770
3771 if (nd->last_type != LAST_NORM) {
3772 if (nd->depth)
3773 put_link(nd);
3774 return handle_dots(nd, type: nd->last_type);
3775 }
3776
3777 /* We _can_ be in RCU mode here */
3778 dentry = lookup_fast_for_open(nd, open_flag);
3779 if (IS_ERR(ptr: dentry))
3780 return ERR_CAST(ptr: dentry);
3781
3782 if (likely(dentry))
3783 goto finish_lookup;
3784
3785 if (!(open_flag & O_CREAT)) {
3786 if (WARN_ON_ONCE(nd->flags & LOOKUP_RCU))
3787 return ERR_PTR(error: -ECHILD);
3788 } else {
3789 if (nd->flags & LOOKUP_RCU) {
3790 if (!try_to_unlazy(nd))
3791 return ERR_PTR(error: -ECHILD);
3792 }
3793 }
3794
3795 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3796 got_write = !mnt_want_write(mnt: nd->path.mnt);
3797 /*
3798 * do _not_ fail yet - we might not need that or fail with
3799 * a different error; let lookup_open() decide; we'll be
3800 * dropping this one anyway.
3801 */
3802 }
3803 if (open_flag & O_CREAT)
3804 inode_lock(inode: dir->d_inode);
3805 else
3806 inode_lock_shared(inode: dir->d_inode);
3807 dentry = lookup_open(nd, file, op, got_write);
3808 if (!IS_ERR(ptr: dentry)) {
3809 if (file->f_mode & FMODE_CREATED)
3810 fsnotify_create(dir: dir->d_inode, dentry);
3811 if (file->f_mode & FMODE_OPENED)
3812 fsnotify_open(file);
3813 }
3814 if (open_flag & O_CREAT)
3815 inode_unlock(inode: dir->d_inode);
3816 else
3817 inode_unlock_shared(inode: dir->d_inode);
3818
3819 if (got_write)
3820 mnt_drop_write(mnt: nd->path.mnt);
3821
3822 if (IS_ERR(ptr: dentry))
3823 return ERR_CAST(ptr: dentry);
3824
3825 if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
3826 dput(nd->path.dentry);
3827 nd->path.dentry = dentry;
3828 return NULL;
3829 }
3830
3831finish_lookup:
3832 if (nd->depth)
3833 put_link(nd);
3834 res = step_into(nd, flags: WALK_TRAILING, dentry);
3835 if (unlikely(res))
3836 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3837 return res;
3838}
3839
3840/*
3841 * Handle the last step of open()
3842 */
3843static int do_open(struct nameidata *nd,
3844 struct file *file, const struct open_flags *op)
3845{
3846 struct mnt_idmap *idmap;
3847 int open_flag = op->open_flag;
3848 bool do_truncate;
3849 int acc_mode;
3850 int error;
3851
3852 if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3853 error = complete_walk(nd);
3854 if (error)
3855 return error;
3856 }
3857 if (!(file->f_mode & FMODE_CREATED))
3858 audit_inode(name: nd->name, dentry: nd->path.dentry, aflags: 0);
3859 idmap = mnt_idmap(mnt: nd->path.mnt);
3860 if (open_flag & O_CREAT) {
3861 if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3862 return -EEXIST;
3863 if (d_is_dir(dentry: nd->path.dentry))
3864 return -EISDIR;
3865 error = may_create_in_sticky(idmap, nd,
3866 inode: d_backing_inode(upper: nd->path.dentry));
3867 if (unlikely(error))
3868 return error;
3869 }
3870 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(dentry: nd->path.dentry))
3871 return -ENOTDIR;
3872
3873 do_truncate = false;
3874 acc_mode = op->acc_mode;
3875 if (file->f_mode & FMODE_CREATED) {
3876 /* Don't check for write permission, don't truncate */
3877 open_flag &= ~O_TRUNC;
3878 acc_mode = 0;
3879 } else if (d_is_reg(dentry: nd->path.dentry) && open_flag & O_TRUNC) {
3880 error = mnt_want_write(mnt: nd->path.mnt);
3881 if (error)
3882 return error;
3883 do_truncate = true;
3884 }
3885 error = may_open(idmap, path: &nd->path, acc_mode, flag: open_flag);
3886 if (!error && !(file->f_mode & FMODE_OPENED))
3887 error = vfs_open(&nd->path, file);
3888 if (!error)
3889 error = security_file_post_open(file, mask: op->acc_mode);
3890 if (!error && do_truncate)
3891 error = handle_truncate(idmap, filp: file);
3892 if (unlikely(error > 0)) {
3893 WARN_ON(1);
3894 error = -EINVAL;
3895 }
3896 if (do_truncate)
3897 mnt_drop_write(mnt: nd->path.mnt);
3898 return error;
3899}
3900
3901/**
3902 * vfs_tmpfile - create tmpfile
3903 * @idmap: idmap of the mount the inode was found from
3904 * @parentpath: pointer to the path of the base directory
3905 * @file: file descriptor of the new tmpfile
3906 * @mode: mode of the new tmpfile
3907 *
3908 * Create a temporary file.
3909 *
3910 * If the inode has been found through an idmapped mount the idmap of
3911 * the vfsmount must be passed through @idmap. This function will then take
3912 * care to map the inode according to @idmap before checking permissions.
3913 * On non-idmapped mounts or if permission checking is to be performed on the
3914 * raw inode simply pass @nop_mnt_idmap.
3915 */
3916int vfs_tmpfile(struct mnt_idmap *idmap,
3917 const struct path *parentpath,
3918 struct file *file, umode_t mode)
3919{
3920 struct dentry *child;
3921 struct inode *dir = d_inode(dentry: parentpath->dentry);
3922 struct inode *inode;
3923 int error;
3924 int open_flag = file->f_flags;
3925
3926 /* we want directory to be writable */
3927 error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
3928 if (error)
3929 return error;
3930 if (!dir->i_op->tmpfile)
3931 return -EOPNOTSUPP;
3932 child = d_alloc(parentpath->dentry, &slash_name);
3933 if (unlikely(!child))
3934 return -ENOMEM;
3935 file->f_path.mnt = parentpath->mnt;
3936 file->f_path.dentry = child;
3937 mode = vfs_prepare_mode(idmap, dir, mode, mask_perms: mode, type: mode);
3938 error = dir->i_op->tmpfile(idmap, dir, file, mode);
3939 dput(child);
3940 if (file->f_mode & FMODE_OPENED)
3941 fsnotify_open(file);
3942 if (error)
3943 return error;
3944 /* Don't check for other permissions, the inode was just created */
3945 error = may_open(idmap, path: &file->f_path, acc_mode: 0, flag: file->f_flags);
3946 if (error)
3947 return error;
3948 inode = file_inode(f: file);
3949 if (!(open_flag & O_EXCL)) {
3950 spin_lock(lock: &inode->i_lock);
3951 inode->i_state |= I_LINKABLE;
3952 spin_unlock(lock: &inode->i_lock);
3953 }
3954 security_inode_post_create_tmpfile(idmap, inode);
3955 return 0;
3956}
3957
3958/**
3959 * kernel_tmpfile_open - open a tmpfile for kernel internal use
3960 * @idmap: idmap of the mount the inode was found from
3961 * @parentpath: path of the base directory
3962 * @mode: mode of the new tmpfile
3963 * @open_flag: flags
3964 * @cred: credentials for open
3965 *
3966 * Create and open a temporary file. The file is not accounted in nr_files,
3967 * hence this is only for kernel internal use, and must not be installed into
3968 * file tables or such.
3969 */
3970struct file *kernel_tmpfile_open(struct mnt_idmap *idmap,
3971 const struct path *parentpath,
3972 umode_t mode, int open_flag,
3973 const struct cred *cred)
3974{
3975 struct file *file;
3976 int error;
3977
3978 file = alloc_empty_file_noaccount(flags: open_flag, cred);
3979 if (IS_ERR(ptr: file))
3980 return file;
3981
3982 error = vfs_tmpfile(idmap, parentpath, file, mode);
3983 if (error) {
3984 fput(file);
3985 file = ERR_PTR(error);
3986 }
3987 return file;
3988}
3989EXPORT_SYMBOL(kernel_tmpfile_open);
3990
3991static int do_tmpfile(struct nameidata *nd, unsigned flags,
3992 const struct open_flags *op,
3993 struct file *file)
3994{
3995 struct path path;
3996 int error = path_lookupat(nd, flags: flags | LOOKUP_DIRECTORY, path: &path);
3997
3998 if (unlikely(error))
3999 return error;
4000 error = mnt_want_write(mnt: path.mnt);
4001 if (unlikely(error))
4002 goto out;
4003 error = vfs_tmpfile(idmap: mnt_idmap(mnt: path.mnt), parentpath: &path, file, mode: op->mode);
4004 if (error)
4005 goto out2;
4006 audit_inode(name: nd->name, dentry: file->f_path.dentry, aflags: 0);
4007out2:
4008 mnt_drop_write(mnt: path.mnt);
4009out:
4010 path_put(&path);
4011 return error;
4012}
4013
4014static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
4015{
4016 struct path path;
4017 int error = path_lookupat(nd, flags, path: &path);
4018 if (!error) {
4019 audit_inode(name: nd->name, dentry: path.dentry, aflags: 0);
4020 error = vfs_open(&path, file);
4021 path_put(&path);
4022 }
4023 return error;
4024}
4025
4026static struct file *path_openat(struct nameidata *nd,
4027 const struct open_flags *op, unsigned flags)
4028{
4029 struct file *file;
4030 int error;
4031
4032 file = alloc_empty_file(flags: op->open_flag, current_cred());
4033 if (IS_ERR(ptr: file))
4034 return file;
4035
4036 if (unlikely(file->f_flags & __O_TMPFILE)) {
4037 error = do_tmpfile(nd, flags, op, file);
4038 } else if (unlikely(file->f_flags & O_PATH)) {
4039 error = do_o_path(nd, flags, file);
4040 } else {
4041 const char *s = path_init(nd, flags);
4042 while (!(error = link_path_walk(name: s, nd)) &&
4043 (s = open_last_lookups(nd, file, op)) != NULL)
4044 ;
4045 if (!error)
4046 error = do_open(nd, file, op);
4047 terminate_walk(nd);
4048 }
4049 if (likely(!error)) {
4050 if (likely(file->f_mode & FMODE_OPENED))
4051 return file;
4052 WARN_ON(1);
4053 error = -EINVAL;
4054 }
4055 fput_close(file);
4056 if (error == -EOPENSTALE) {
4057 if (flags & LOOKUP_RCU)
4058 error = -ECHILD;
4059 else
4060 error = -ESTALE;
4061 }
4062 return ERR_PTR(error);
4063}
4064
4065struct file *do_filp_open(int dfd, struct filename *pathname,
4066 const struct open_flags *op)
4067{
4068 struct nameidata nd;
4069 int flags = op->lookup_flags;
4070 struct file *filp;
4071
4072 set_nameidata(p: &nd, dfd, name: pathname, NULL);
4073 filp = path_openat(nd: &nd, op, flags: flags | LOOKUP_RCU);
4074 if (unlikely(filp == ERR_PTR(-ECHILD)))
4075 filp = path_openat(nd: &nd, op, flags);
4076 if (unlikely(filp == ERR_PTR(-ESTALE)))
4077 filp = path_openat(nd: &nd, op, flags: flags | LOOKUP_REVAL);
4078 restore_nameidata();
4079 return filp;
4080}
4081
4082struct file *do_file_open_root(const struct path *root,
4083 const char *name, const struct open_flags *op)
4084{
4085 struct nameidata nd;
4086 struct file *file;
4087 struct filename *filename;
4088 int flags = op->lookup_flags;
4089
4090 if (d_is_symlink(dentry: root->dentry) && op->intent & LOOKUP_OPEN)
4091 return ERR_PTR(error: -ELOOP);
4092
4093 filename = getname_kernel(name);
4094 if (IS_ERR(ptr: filename))
4095 return ERR_CAST(ptr: filename);
4096
4097 set_nameidata(p: &nd, dfd: -1, name: filename, root);
4098 file = path_openat(nd: &nd, op, flags: flags | LOOKUP_RCU);
4099 if (unlikely(file == ERR_PTR(-ECHILD)))
4100 file = path_openat(nd: &nd, op, flags);
4101 if (unlikely(file == ERR_PTR(-ESTALE)))
4102 file = path_openat(nd: &nd, op, flags: flags | LOOKUP_REVAL);
4103 restore_nameidata();
4104 putname(filename);
4105 return file;
4106}
4107
4108static struct dentry *filename_create(int dfd, struct filename *name,
4109 struct path *path, unsigned int lookup_flags)
4110{
4111 struct dentry *dentry = ERR_PTR(error: -EEXIST);
4112 struct qstr last;
4113 bool want_dir = lookup_flags & LOOKUP_DIRECTORY;
4114 unsigned int reval_flag = lookup_flags & LOOKUP_REVAL;
4115 unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL;
4116 int type;
4117 int err2;
4118 int error;
4119
4120 error = filename_parentat(dfd, name, flags: reval_flag, parent: path, last: &last, type: &type);
4121 if (error)
4122 return ERR_PTR(error);
4123
4124 /*
4125 * Yucky last component or no last component at all?
4126 * (foo/., foo/.., /////)
4127 */
4128 if (unlikely(type != LAST_NORM))
4129 goto out;
4130
4131 /* don't fail immediately if it's r/o, at least try to report other errors */
4132 err2 = mnt_want_write(mnt: path->mnt);
4133 /*
4134 * Do the final lookup. Suppress 'create' if there is a trailing
4135 * '/', and a directory wasn't requested.
4136 */
4137 if (last.name[last.len] && !want_dir)
4138 create_flags &= ~LOOKUP_CREATE;
4139 inode_lock_nested(inode: path->dentry->d_inode, subclass: I_MUTEX_PARENT);
4140 dentry = lookup_one_qstr_excl(&last, path->dentry,
4141 reval_flag | create_flags);
4142 if (IS_ERR(ptr: dentry))
4143 goto unlock;
4144
4145 if (unlikely(err2)) {
4146 error = err2;
4147 goto fail;
4148 }
4149 return dentry;
4150fail:
4151 dput(dentry);
4152 dentry = ERR_PTR(error);
4153unlock:
4154 inode_unlock(inode: path->dentry->d_inode);
4155 if (!err2)
4156 mnt_drop_write(mnt: path->mnt);
4157out:
4158 path_put(path);
4159 return dentry;
4160}
4161
4162struct dentry *kern_path_create(int dfd, const char *pathname,
4163 struct path *path, unsigned int lookup_flags)
4164{
4165 struct filename *filename = getname_kernel(pathname);
4166 struct dentry *res = filename_create(dfd, name: filename, path, lookup_flags);
4167
4168 putname(filename);
4169 return res;
4170}
4171EXPORT_SYMBOL(kern_path_create);
4172
4173void done_path_create(struct path *path, struct dentry *dentry)
4174{
4175 if (!IS_ERR(ptr: dentry))
4176 dput(dentry);
4177 inode_unlock(inode: path->dentry->d_inode);
4178 mnt_drop_write(mnt: path->mnt);
4179 path_put(path);
4180}
4181EXPORT_SYMBOL(done_path_create);
4182
4183inline struct dentry *user_path_create(int dfd, const char __user *pathname,
4184 struct path *path, unsigned int lookup_flags)
4185{
4186 struct filename *filename = getname(name: pathname);
4187 struct dentry *res = filename_create(dfd, name: filename, path, lookup_flags);
4188
4189 putname(filename);
4190 return res;
4191}
4192EXPORT_SYMBOL(user_path_create);
4193
4194/**
4195 * vfs_mknod - create device node or file
4196 * @idmap: idmap of the mount the inode was found from
4197 * @dir: inode of the parent directory
4198 * @dentry: dentry of the child device node
4199 * @mode: mode of the child device node
4200 * @dev: device number of device to create
4201 *
4202 * Create a device node or file.
4203 *
4204 * If the inode has been found through an idmapped mount the idmap of
4205 * the vfsmount must be passed through @idmap. This function will then take
4206 * care to map the inode according to @idmap before checking permissions.
4207 * On non-idmapped mounts or if permission checking is to be performed on the
4208 * raw inode simply pass @nop_mnt_idmap.
4209 */
4210int vfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
4211 struct dentry *dentry, umode_t mode, dev_t dev)
4212{
4213 bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV;
4214 int error = may_create(idmap, dir, child: dentry);
4215
4216 if (error)
4217 return error;
4218
4219 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout &&
4220 !capable(CAP_MKNOD))
4221 return -EPERM;
4222
4223 if (!dir->i_op->mknod)
4224 return -EPERM;
4225
4226 mode = vfs_prepare_mode(idmap, dir, mode, mask_perms: mode, type: mode);
4227 error = devcgroup_inode_mknod(mode, dev);
4228 if (error)
4229 return error;
4230
4231 error = security_inode_mknod(dir, dentry, mode, dev);
4232 if (error)
4233 return error;
4234
4235 error = dir->i_op->mknod(idmap, dir, dentry, mode, dev);
4236 if (!error)
4237 fsnotify_create(dir, dentry);
4238 return error;
4239}
4240EXPORT_SYMBOL(vfs_mknod);
4241
4242static int may_mknod(umode_t mode)
4243{
4244 switch (mode & S_IFMT) {
4245 case S_IFREG:
4246 case S_IFCHR:
4247 case S_IFBLK:
4248 case S_IFIFO:
4249 case S_IFSOCK:
4250 case 0: /* zero mode translates to S_IFREG */
4251 return 0;
4252 case S_IFDIR:
4253 return -EPERM;
4254 default:
4255 return -EINVAL;
4256 }
4257}
4258
4259static int do_mknodat(int dfd, struct filename *name, umode_t mode,
4260 unsigned int dev)
4261{
4262 struct mnt_idmap *idmap;
4263 struct dentry *dentry;
4264 struct path path;
4265 int error;
4266 unsigned int lookup_flags = 0;
4267
4268 error = may_mknod(mode);
4269 if (error)
4270 goto out1;
4271retry:
4272 dentry = filename_create(dfd, name, path: &path, lookup_flags);
4273 error = PTR_ERR(ptr: dentry);
4274 if (IS_ERR(ptr: dentry))
4275 goto out1;
4276
4277 error = security_path_mknod(dir: &path, dentry,
4278 mode: mode_strip_umask(dir: path.dentry->d_inode, mode), dev);
4279 if (error)
4280 goto out2;
4281
4282 idmap = mnt_idmap(mnt: path.mnt);
4283 switch (mode & S_IFMT) {
4284 case 0: case S_IFREG:
4285 error = vfs_create(idmap, path.dentry->d_inode,
4286 dentry, mode, true);
4287 if (!error)
4288 security_path_post_mknod(idmap, dentry);
4289 break;
4290 case S_IFCHR: case S_IFBLK:
4291 error = vfs_mknod(idmap, path.dentry->d_inode,
4292 dentry, mode, new_decode_dev(dev));
4293 break;
4294 case S_IFIFO: case S_IFSOCK:
4295 error = vfs_mknod(idmap, path.dentry->d_inode,
4296 dentry, mode, 0);
4297 break;
4298 }
4299out2:
4300 done_path_create(&path, dentry);
4301 if (retry_estale(error, flags: lookup_flags)) {
4302 lookup_flags |= LOOKUP_REVAL;
4303 goto retry;
4304 }
4305out1:
4306 putname(name);
4307 return error;
4308}
4309
4310SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
4311 unsigned int, dev)
4312{
4313 return do_mknodat(dfd, name: getname(name: filename), mode, dev);
4314}
4315
4316SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
4317{
4318 return do_mknodat(AT_FDCWD, name: getname(name: filename), mode, dev);
4319}
4320
4321/**
4322 * vfs_mkdir - create directory returning correct dentry if possible
4323 * @idmap: idmap of the mount the inode was found from
4324 * @dir: inode of the parent directory
4325 * @dentry: dentry of the child directory
4326 * @mode: mode of the child directory
4327 *
4328 * Create a directory.
4329 *
4330 * If the inode has been found through an idmapped mount the idmap of
4331 * the vfsmount must be passed through @idmap. This function will then take
4332 * care to map the inode according to @idmap before checking permissions.
4333 * On non-idmapped mounts or if permission checking is to be performed on the
4334 * raw inode simply pass @nop_mnt_idmap.
4335 *
4336 * In the event that the filesystem does not use the *@dentry but leaves it
4337 * negative or unhashes it and possibly splices a different one returning it,
4338 * the original dentry is dput() and the alternate is returned.
4339 *
4340 * In case of an error the dentry is dput() and an ERR_PTR() is returned.
4341 */
4342struct dentry *vfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
4343 struct dentry *dentry, umode_t mode)
4344{
4345 int error;
4346 unsigned max_links = dir->i_sb->s_max_links;
4347 struct dentry *de;
4348
4349 error = may_create(idmap, dir, child: dentry);
4350 if (error)
4351 goto err;
4352
4353 error = -EPERM;
4354 if (!dir->i_op->mkdir)
4355 goto err;
4356
4357 mode = vfs_prepare_mode(idmap, dir, mode, S_IRWXUGO | S_ISVTX, type: 0);
4358 error = security_inode_mkdir(dir, dentry, mode);
4359 if (error)
4360 goto err;
4361
4362 error = -EMLINK;
4363 if (max_links && dir->i_nlink >= max_links)
4364 goto err;
4365
4366 de = dir->i_op->mkdir(idmap, dir, dentry, mode);
4367 error = PTR_ERR(ptr: de);
4368 if (IS_ERR(ptr: de))
4369 goto err;
4370 if (de) {
4371 dput(dentry);
4372 dentry = de;
4373 }
4374 fsnotify_mkdir(dir, dentry);
4375 return dentry;
4376
4377err:
4378 dput(dentry);
4379 return ERR_PTR(error);
4380}
4381EXPORT_SYMBOL(vfs_mkdir);
4382
4383int do_mkdirat(int dfd, struct filename *name, umode_t mode)
4384{
4385 struct dentry *dentry;
4386 struct path path;
4387 int error;
4388 unsigned int lookup_flags = LOOKUP_DIRECTORY;
4389
4390retry:
4391 dentry = filename_create(dfd, name, path: &path, lookup_flags);
4392 error = PTR_ERR(ptr: dentry);
4393 if (IS_ERR(ptr: dentry))
4394 goto out_putname;
4395
4396 error = security_path_mkdir(dir: &path, dentry,
4397 mode: mode_strip_umask(dir: path.dentry->d_inode, mode));
4398 if (!error) {
4399 dentry = vfs_mkdir(mnt_idmap(mnt: path.mnt), path.dentry->d_inode,
4400 dentry, mode);
4401 if (IS_ERR(ptr: dentry))
4402 error = PTR_ERR(ptr: dentry);
4403 }
4404 done_path_create(&path, dentry);
4405 if (retry_estale(error, flags: lookup_flags)) {
4406 lookup_flags |= LOOKUP_REVAL;
4407 goto retry;
4408 }
4409out_putname:
4410 putname(name);
4411 return error;
4412}
4413
4414SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
4415{
4416 return do_mkdirat(dfd, name: getname(name: pathname), mode);
4417}
4418
4419SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
4420{
4421 return do_mkdirat(AT_FDCWD, name: getname(name: pathname), mode);
4422}
4423
4424/**
4425 * vfs_rmdir - remove directory
4426 * @idmap: idmap of the mount the inode was found from
4427 * @dir: inode of the parent directory
4428 * @dentry: dentry of the child directory
4429 *
4430 * Remove a directory.
4431 *
4432 * If the inode has been found through an idmapped mount the idmap of
4433 * the vfsmount must be passed through @idmap. This function will then take
4434 * care to map the inode according to @idmap before checking permissions.
4435 * On non-idmapped mounts or if permission checking is to be performed on the
4436 * raw inode simply pass @nop_mnt_idmap.
4437 */
4438int vfs_rmdir(struct mnt_idmap *idmap, struct inode *dir,
4439 struct dentry *dentry)
4440{
4441 int error = may_delete(idmap, dir, victim: dentry, isdir: 1);
4442
4443 if (error)
4444 return error;
4445
4446 if (!dir->i_op->rmdir)
4447 return -EPERM;
4448
4449 dget(dentry);
4450 inode_lock(inode: dentry->d_inode);
4451
4452 error = -EBUSY;
4453 if (is_local_mountpoint(dentry) ||
4454 (dentry->d_inode->i_flags & S_KERNEL_FILE))
4455 goto out;
4456
4457 error = security_inode_rmdir(dir, dentry);
4458 if (error)
4459 goto out;
4460
4461 error = dir->i_op->rmdir(dir, dentry);
4462 if (error)
4463 goto out;
4464
4465 shrink_dcache_parent(dentry);
4466 dentry->d_inode->i_flags |= S_DEAD;
4467 dont_mount(dentry);
4468 detach_mounts(dentry);
4469
4470out:
4471 inode_unlock(inode: dentry->d_inode);
4472 dput(dentry);
4473 if (!error)
4474 d_delete_notify(dir, dentry);
4475 return error;
4476}
4477EXPORT_SYMBOL(vfs_rmdir);
4478
4479int do_rmdir(int dfd, struct filename *name)
4480{
4481 int error;
4482 struct dentry *dentry;
4483 struct path path;
4484 struct qstr last;
4485 int type;
4486 unsigned int lookup_flags = 0;
4487retry:
4488 error = filename_parentat(dfd, name, flags: lookup_flags, parent: &path, last: &last, type: &type);
4489 if (error)
4490 goto exit1;
4491
4492 switch (type) {
4493 case LAST_DOTDOT:
4494 error = -ENOTEMPTY;
4495 goto exit2;
4496 case LAST_DOT:
4497 error = -EINVAL;
4498 goto exit2;
4499 case LAST_ROOT:
4500 error = -EBUSY;
4501 goto exit2;
4502 }
4503
4504 error = mnt_want_write(mnt: path.mnt);
4505 if (error)
4506 goto exit2;
4507
4508 inode_lock_nested(inode: path.dentry->d_inode, subclass: I_MUTEX_PARENT);
4509 dentry = lookup_one_qstr_excl(&last, path.dentry, lookup_flags);
4510 error = PTR_ERR(ptr: dentry);
4511 if (IS_ERR(ptr: dentry))
4512 goto exit3;
4513 error = security_path_rmdir(dir: &path, dentry);
4514 if (error)
4515 goto exit4;
4516 error = vfs_rmdir(mnt_idmap(mnt: path.mnt), path.dentry->d_inode, dentry);
4517exit4:
4518 dput(dentry);
4519exit3:
4520 inode_unlock(inode: path.dentry->d_inode);
4521 mnt_drop_write(mnt: path.mnt);
4522exit2:
4523 path_put(&path);
4524 if (retry_estale(error, flags: lookup_flags)) {
4525 lookup_flags |= LOOKUP_REVAL;
4526 goto retry;
4527 }
4528exit1:
4529 putname(name);
4530 return error;
4531}
4532
4533SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
4534{
4535 return do_rmdir(AT_FDCWD, name: getname(name: pathname));
4536}
4537
4538/**
4539 * vfs_unlink - unlink a filesystem object
4540 * @idmap: idmap of the mount the inode was found from
4541 * @dir: parent directory
4542 * @dentry: victim
4543 * @delegated_inode: returns victim inode, if the inode is delegated.
4544 *
4545 * The caller must hold dir->i_mutex.
4546 *
4547 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
4548 * return a reference to the inode in delegated_inode. The caller
4549 * should then break the delegation on that inode and retry. Because
4550 * breaking a delegation may take a long time, the caller should drop
4551 * dir->i_mutex before doing so.
4552 *
4553 * Alternatively, a caller may pass NULL for delegated_inode. This may
4554 * be appropriate for callers that expect the underlying filesystem not
4555 * to be NFS exported.
4556 *
4557 * If the inode has been found through an idmapped mount the idmap of
4558 * the vfsmount must be passed through @idmap. This function will then take
4559 * care to map the inode according to @idmap before checking permissions.
4560 * On non-idmapped mounts or if permission checking is to be performed on the
4561 * raw inode simply pass @nop_mnt_idmap.
4562 */
4563int vfs_unlink(struct mnt_idmap *idmap, struct inode *dir,
4564 struct dentry *dentry, struct inode **delegated_inode)
4565{
4566 struct inode *target = dentry->d_inode;
4567 int error = may_delete(idmap, dir, victim: dentry, isdir: 0);
4568
4569 if (error)
4570 return error;
4571
4572 if (!dir->i_op->unlink)
4573 return -EPERM;
4574
4575 inode_lock(inode: target);
4576 if (IS_SWAPFILE(target))
4577 error = -EPERM;
4578 else if (is_local_mountpoint(dentry))
4579 error = -EBUSY;
4580 else {
4581 error = security_inode_unlink(dir, dentry);
4582 if (!error) {
4583 error = try_break_deleg(inode: target, delegated_inode);
4584 if (error)
4585 goto out;
4586 error = dir->i_op->unlink(dir, dentry);
4587 if (!error) {
4588 dont_mount(dentry);
4589 detach_mounts(dentry);
4590 }
4591 }
4592 }
4593out:
4594 inode_unlock(inode: target);
4595
4596 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4597 if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) {
4598 fsnotify_unlink(dir, dentry);
4599 } else if (!error) {
4600 fsnotify_link_count(inode: target);
4601 d_delete_notify(dir, dentry);
4602 }
4603
4604 return error;
4605}
4606EXPORT_SYMBOL(vfs_unlink);
4607
4608/*
4609 * Make sure that the actual truncation of the file will occur outside its
4610 * directory's i_mutex. Truncate can take a long time if there is a lot of
4611 * writeout happening, and we don't want to prevent access to the directory
4612 * while waiting on the I/O.
4613 */
4614int do_unlinkat(int dfd, struct filename *name)
4615{
4616 int error;
4617 struct dentry *dentry;
4618 struct path path;
4619 struct qstr last;
4620 int type;
4621 struct inode *inode = NULL;
4622 struct inode *delegated_inode = NULL;
4623 unsigned int lookup_flags = 0;
4624retry:
4625 error = filename_parentat(dfd, name, flags: lookup_flags, parent: &path, last: &last, type: &type);
4626 if (error)
4627 goto exit1;
4628
4629 error = -EISDIR;
4630 if (type != LAST_NORM)
4631 goto exit2;
4632
4633 error = mnt_want_write(mnt: path.mnt);
4634 if (error)
4635 goto exit2;
4636retry_deleg:
4637 inode_lock_nested(inode: path.dentry->d_inode, subclass: I_MUTEX_PARENT);
4638 dentry = lookup_one_qstr_excl(&last, path.dentry, lookup_flags);
4639 error = PTR_ERR(ptr: dentry);
4640 if (!IS_ERR(ptr: dentry)) {
4641
4642 /* Why not before? Because we want correct error value */
4643 if (last.name[last.len])
4644 goto slashes;
4645 inode = dentry->d_inode;
4646 ihold(inode);
4647 error = security_path_unlink(dir: &path, dentry);
4648 if (error)
4649 goto exit3;
4650 error = vfs_unlink(mnt_idmap(mnt: path.mnt), path.dentry->d_inode,
4651 dentry, &delegated_inode);
4652exit3:
4653 dput(dentry);
4654 }
4655 inode_unlock(inode: path.dentry->d_inode);
4656 if (inode)
4657 iput(inode); /* truncate the inode here */
4658 inode = NULL;
4659 if (delegated_inode) {
4660 error = break_deleg_wait(delegated_inode: &delegated_inode);
4661 if (!error)
4662 goto retry_deleg;
4663 }
4664 mnt_drop_write(mnt: path.mnt);
4665exit2:
4666 path_put(&path);
4667 if (retry_estale(error, flags: lookup_flags)) {
4668 lookup_flags |= LOOKUP_REVAL;
4669 inode = NULL;
4670 goto retry;
4671 }
4672exit1:
4673 putname(name);
4674 return error;
4675
4676slashes:
4677 if (d_is_dir(dentry))
4678 error = -EISDIR;
4679 else
4680 error = -ENOTDIR;
4681 goto exit3;
4682}
4683
4684SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4685{
4686 if ((flag & ~AT_REMOVEDIR) != 0)
4687 return -EINVAL;
4688
4689 if (flag & AT_REMOVEDIR)
4690 return do_rmdir(dfd, name: getname(name: pathname));
4691 return do_unlinkat(dfd, name: getname(name: pathname));
4692}
4693
4694SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4695{
4696 return do_unlinkat(AT_FDCWD, name: getname(name: pathname));
4697}
4698
4699/**
4700 * vfs_symlink - create symlink
4701 * @idmap: idmap of the mount the inode was found from
4702 * @dir: inode of the parent directory
4703 * @dentry: dentry of the child symlink file
4704 * @oldname: name of the file to link to
4705 *
4706 * Create a symlink.
4707 *
4708 * If the inode has been found through an idmapped mount the idmap of
4709 * the vfsmount must be passed through @idmap. This function will then take
4710 * care to map the inode according to @idmap before checking permissions.
4711 * On non-idmapped mounts or if permission checking is to be performed on the
4712 * raw inode simply pass @nop_mnt_idmap.
4713 */
4714int vfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
4715 struct dentry *dentry, const char *oldname)
4716{
4717 int error;
4718
4719 error = may_create(idmap, dir, child: dentry);
4720 if (error)
4721 return error;
4722
4723 if (!dir->i_op->symlink)
4724 return -EPERM;
4725
4726 error = security_inode_symlink(dir, dentry, old_name: oldname);
4727 if (error)
4728 return error;
4729
4730 error = dir->i_op->symlink(idmap, dir, dentry, oldname);
4731 if (!error)
4732 fsnotify_create(dir, dentry);
4733 return error;
4734}
4735EXPORT_SYMBOL(vfs_symlink);
4736
4737int do_symlinkat(struct filename *from, int newdfd, struct filename *to)
4738{
4739 int error;
4740 struct dentry *dentry;
4741 struct path path;
4742 unsigned int lookup_flags = 0;
4743
4744 if (IS_ERR(ptr: from)) {
4745 error = PTR_ERR(ptr: from);
4746 goto out_putnames;
4747 }
4748retry:
4749 dentry = filename_create(dfd: newdfd, name: to, path: &path, lookup_flags);
4750 error = PTR_ERR(ptr: dentry);
4751 if (IS_ERR(ptr: dentry))
4752 goto out_putnames;
4753
4754 error = security_path_symlink(dir: &path, dentry, old_name: from->name);
4755 if (!error)
4756 error = vfs_symlink(mnt_idmap(mnt: path.mnt), path.dentry->d_inode,
4757 dentry, from->name);
4758 done_path_create(&path, dentry);
4759 if (retry_estale(error, flags: lookup_flags)) {
4760 lookup_flags |= LOOKUP_REVAL;
4761 goto retry;
4762 }
4763out_putnames:
4764 putname(to);
4765 putname(from);
4766 return error;
4767}
4768
4769SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4770 int, newdfd, const char __user *, newname)
4771{
4772 return do_symlinkat(from: getname(name: oldname), newdfd, to: getname(name: newname));
4773}
4774
4775SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4776{
4777 return do_symlinkat(from: getname(name: oldname), AT_FDCWD, to: getname(name: newname));
4778}
4779
4780/**
4781 * vfs_link - create a new link
4782 * @old_dentry: object to be linked
4783 * @idmap: idmap of the mount
4784 * @dir: new parent
4785 * @new_dentry: where to create the new link
4786 * @delegated_inode: returns inode needing a delegation break
4787 *
4788 * The caller must hold dir->i_mutex
4789 *
4790 * If vfs_link discovers a delegation on the to-be-linked file in need
4791 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4792 * inode in delegated_inode. The caller should then break the delegation
4793 * and retry. Because breaking a delegation may take a long time, the
4794 * caller should drop the i_mutex before doing so.
4795 *
4796 * Alternatively, a caller may pass NULL for delegated_inode. This may
4797 * be appropriate for callers that expect the underlying filesystem not
4798 * to be NFS exported.
4799 *
4800 * If the inode has been found through an idmapped mount the idmap of
4801 * the vfsmount must be passed through @idmap. This function will then take
4802 * care to map the inode according to @idmap before checking permissions.
4803 * On non-idmapped mounts or if permission checking is to be performed on the
4804 * raw inode simply pass @nop_mnt_idmap.
4805 */
4806int vfs_link(struct dentry *old_dentry, struct mnt_idmap *idmap,
4807 struct inode *dir, struct dentry *new_dentry,
4808 struct inode **delegated_inode)
4809{
4810 struct inode *inode = old_dentry->d_inode;
4811 unsigned max_links = dir->i_sb->s_max_links;
4812 int error;
4813
4814 if (!inode)
4815 return -ENOENT;
4816
4817 error = may_create(idmap, dir, child: new_dentry);
4818 if (error)
4819 return error;
4820
4821 if (dir->i_sb != inode->i_sb)
4822 return -EXDEV;
4823
4824 /*
4825 * A link to an append-only or immutable file cannot be created.
4826 */
4827 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4828 return -EPERM;
4829 /*
4830 * Updating the link count will likely cause i_uid and i_gid to
4831 * be writen back improperly if their true value is unknown to
4832 * the vfs.
4833 */
4834 if (HAS_UNMAPPED_ID(idmap, inode))
4835 return -EPERM;
4836 if (!dir->i_op->link)
4837 return -EPERM;
4838 if (S_ISDIR(inode->i_mode))
4839 return -EPERM;
4840
4841 error = security_inode_link(old_dentry, dir, new_dentry);
4842 if (error)
4843 return error;
4844
4845 inode_lock(inode);
4846 /* Make sure we don't allow creating hardlink to an unlinked file */
4847 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4848 error = -ENOENT;
4849 else if (max_links && inode->i_nlink >= max_links)
4850 error = -EMLINK;
4851 else {
4852 error = try_break_deleg(inode, delegated_inode);
4853 if (!error)
4854 error = dir->i_op->link(old_dentry, dir, new_dentry);
4855 }
4856
4857 if (!error && (inode->i_state & I_LINKABLE)) {
4858 spin_lock(lock: &inode->i_lock);
4859 inode->i_state &= ~I_LINKABLE;
4860 spin_unlock(lock: &inode->i_lock);
4861 }
4862 inode_unlock(inode);
4863 if (!error)
4864 fsnotify_link(dir, inode, new_dentry);
4865 return error;
4866}
4867EXPORT_SYMBOL(vfs_link);
4868
4869/*
4870 * Hardlinks are often used in delicate situations. We avoid
4871 * security-related surprises by not following symlinks on the
4872 * newname. --KAB
4873 *
4874 * We don't follow them on the oldname either to be compatible
4875 * with linux 2.0, and to avoid hard-linking to directories
4876 * and other special files. --ADM
4877 */
4878int do_linkat(int olddfd, struct filename *old, int newdfd,
4879 struct filename *new, int flags)
4880{
4881 struct mnt_idmap *idmap;
4882 struct dentry *new_dentry;
4883 struct path old_path, new_path;
4884 struct inode *delegated_inode = NULL;
4885 int how = 0;
4886 int error;
4887
4888 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) {
4889 error = -EINVAL;
4890 goto out_putnames;
4891 }
4892 /*
4893 * To use null names we require CAP_DAC_READ_SEARCH or
4894 * that the open-time creds of the dfd matches current.
4895 * This ensures that not everyone will be able to create
4896 * a hardlink using the passed file descriptor.
4897 */
4898 if (flags & AT_EMPTY_PATH)
4899 how |= LOOKUP_LINKAT_EMPTY;
4900
4901 if (flags & AT_SYMLINK_FOLLOW)
4902 how |= LOOKUP_FOLLOW;
4903retry:
4904 error = filename_lookup(dfd: olddfd, name: old, flags: how, path: &old_path, NULL);
4905 if (error)
4906 goto out_putnames;
4907
4908 new_dentry = filename_create(dfd: newdfd, name: new, path: &new_path,
4909 lookup_flags: (how & LOOKUP_REVAL));
4910 error = PTR_ERR(ptr: new_dentry);
4911 if (IS_ERR(ptr: new_dentry))
4912 goto out_putpath;
4913
4914 error = -EXDEV;
4915 if (old_path.mnt != new_path.mnt)
4916 goto out_dput;
4917 idmap = mnt_idmap(mnt: new_path.mnt);
4918 error = may_linkat(idmap, link: &old_path);
4919 if (unlikely(error))
4920 goto out_dput;
4921 error = security_path_link(old_dentry: old_path.dentry, new_dir: &new_path, new_dentry);
4922 if (error)
4923 goto out_dput;
4924 error = vfs_link(old_path.dentry, idmap, new_path.dentry->d_inode,
4925 new_dentry, &delegated_inode);
4926out_dput:
4927 done_path_create(&new_path, new_dentry);
4928 if (delegated_inode) {
4929 error = break_deleg_wait(delegated_inode: &delegated_inode);
4930 if (!error) {
4931 path_put(&old_path);
4932 goto retry;
4933 }
4934 }
4935 if (retry_estale(error, flags: how)) {
4936 path_put(&old_path);
4937 how |= LOOKUP_REVAL;
4938 goto retry;
4939 }
4940out_putpath:
4941 path_put(&old_path);
4942out_putnames:
4943 putname(old);
4944 putname(new);
4945
4946 return error;
4947}
4948
4949SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4950 int, newdfd, const char __user *, newname, int, flags)
4951{
4952 return do_linkat(olddfd, old: getname_uflags(filename: oldname, uflags: flags),
4953 newdfd, new: getname(name: newname), flags);
4954}
4955
4956SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4957{
4958 return do_linkat(AT_FDCWD, old: getname(name: oldname), AT_FDCWD, new: getname(name: newname), flags: 0);
4959}
4960
4961/**
4962 * vfs_rename - rename a filesystem object
4963 * @rd: pointer to &struct renamedata info
4964 *
4965 * The caller must hold multiple mutexes--see lock_rename()).
4966 *
4967 * If vfs_rename discovers a delegation in need of breaking at either
4968 * the source or destination, it will return -EWOULDBLOCK and return a
4969 * reference to the inode in delegated_inode. The caller should then
4970 * break the delegation and retry. Because breaking a delegation may
4971 * take a long time, the caller should drop all locks before doing
4972 * so.
4973 *
4974 * Alternatively, a caller may pass NULL for delegated_inode. This may
4975 * be appropriate for callers that expect the underlying filesystem not
4976 * to be NFS exported.
4977 *
4978 * The worst of all namespace operations - renaming directory. "Perverted"
4979 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4980 * Problems:
4981 *
4982 * a) we can get into loop creation.
4983 * b) race potential - two innocent renames can create a loop together.
4984 * That's where 4.4BSD screws up. Current fix: serialization on
4985 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4986 * story.
4987 * c) we may have to lock up to _four_ objects - parents and victim (if it exists),
4988 * and source (if it's a non-directory or a subdirectory that moves to
4989 * different parent).
4990 * And that - after we got ->i_mutex on parents (until then we don't know
4991 * whether the target exists). Solution: try to be smart with locking
4992 * order for inodes. We rely on the fact that tree topology may change
4993 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4994 * move will be locked. Thus we can rank directories by the tree
4995 * (ancestors first) and rank all non-directories after them.
4996 * That works since everybody except rename does "lock parent, lookup,
4997 * lock child" and rename is under ->s_vfs_rename_mutex.
4998 * HOWEVER, it relies on the assumption that any object with ->lookup()
4999 * has no more than 1 dentry. If "hybrid" objects will ever appear,
5000 * we'd better make sure that there's no link(2) for them.
5001 * d) conversion from fhandle to dentry may come in the wrong moment - when
5002 * we are removing the target. Solution: we will have to grab ->i_mutex
5003 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
5004 * ->i_mutex on parents, which works but leads to some truly excessive
5005 * locking].
5006 */
5007int vfs_rename(struct renamedata *rd)
5008{
5009 int error;
5010 struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir;
5011 struct dentry *old_dentry = rd->old_dentry;
5012 struct dentry *new_dentry = rd->new_dentry;
5013 struct inode **delegated_inode = rd->delegated_inode;
5014 unsigned int flags = rd->flags;
5015 bool is_dir = d_is_dir(dentry: old_dentry);
5016 struct inode *source = old_dentry->d_inode;
5017 struct inode *target = new_dentry->d_inode;
5018 bool new_is_dir = false;
5019 unsigned max_links = new_dir->i_sb->s_max_links;
5020 struct name_snapshot old_name;
5021 bool lock_old_subdir, lock_new_subdir;
5022
5023 if (source == target)
5024 return 0;
5025
5026 error = may_delete(idmap: rd->old_mnt_idmap, dir: old_dir, victim: old_dentry, isdir: is_dir);
5027 if (error)
5028 return error;
5029
5030 if (!target) {
5031 error = may_create(idmap: rd->new_mnt_idmap, dir: new_dir, child: new_dentry);
5032 } else {
5033 new_is_dir = d_is_dir(dentry: new_dentry);
5034
5035 if (!(flags & RENAME_EXCHANGE))
5036 error = may_delete(idmap: rd->new_mnt_idmap, dir: new_dir,
5037 victim: new_dentry, isdir: is_dir);
5038 else
5039 error = may_delete(idmap: rd->new_mnt_idmap, dir: new_dir,
5040 victim: new_dentry, isdir: new_is_dir);
5041 }
5042 if (error)
5043 return error;
5044
5045 if (!old_dir->i_op->rename)
5046 return -EPERM;
5047
5048 /*
5049 * If we are going to change the parent - check write permissions,
5050 * we'll need to flip '..'.
5051 */
5052 if (new_dir != old_dir) {
5053 if (is_dir) {
5054 error = inode_permission(rd->old_mnt_idmap, source,
5055 MAY_WRITE);
5056 if (error)
5057 return error;
5058 }
5059 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
5060 error = inode_permission(rd->new_mnt_idmap, target,
5061 MAY_WRITE);
5062 if (error)
5063 return error;
5064 }
5065 }
5066
5067 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
5068 flags);
5069 if (error)
5070 return error;
5071
5072 take_dentry_name_snapshot(&old_name, old_dentry);
5073 dget(dentry: new_dentry);
5074 /*
5075 * Lock children.
5076 * The source subdirectory needs to be locked on cross-directory
5077 * rename or cross-directory exchange since its parent changes.
5078 * The target subdirectory needs to be locked on cross-directory
5079 * exchange due to parent change and on any rename due to becoming
5080 * a victim.
5081 * Non-directories need locking in all cases (for NFS reasons);
5082 * they get locked after any subdirectories (in inode address order).
5083 *
5084 * NOTE: WE ONLY LOCK UNRELATED DIRECTORIES IN CROSS-DIRECTORY CASE.
5085 * NEVER, EVER DO THAT WITHOUT ->s_vfs_rename_mutex.
5086 */
5087 lock_old_subdir = new_dir != old_dir;
5088 lock_new_subdir = new_dir != old_dir || !(flags & RENAME_EXCHANGE);
5089 if (is_dir) {
5090 if (lock_old_subdir)
5091 inode_lock_nested(inode: source, subclass: I_MUTEX_CHILD);
5092 if (target && (!new_is_dir || lock_new_subdir))
5093 inode_lock(inode: target);
5094 } else if (new_is_dir) {
5095 if (lock_new_subdir)
5096 inode_lock_nested(inode: target, subclass: I_MUTEX_CHILD);
5097 inode_lock(inode: source);
5098 } else {
5099 lock_two_nondirectories(source, target);
5100 }
5101
5102 error = -EPERM;
5103 if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target)))
5104 goto out;
5105
5106 error = -EBUSY;
5107 if (is_local_mountpoint(dentry: old_dentry) || is_local_mountpoint(dentry: new_dentry))
5108 goto out;
5109
5110 if (max_links && new_dir != old_dir) {
5111 error = -EMLINK;
5112 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
5113 goto out;
5114 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
5115 old_dir->i_nlink >= max_links)
5116 goto out;
5117 }
5118 if (!is_dir) {
5119 error = try_break_deleg(inode: source, delegated_inode);
5120 if (error)
5121 goto out;
5122 }
5123 if (target && !new_is_dir) {
5124 error = try_break_deleg(inode: target, delegated_inode);
5125 if (error)
5126 goto out;
5127 }
5128 error = old_dir->i_op->rename(rd->new_mnt_idmap, old_dir, old_dentry,
5129 new_dir, new_dentry, flags);
5130 if (error)
5131 goto out;
5132
5133 if (!(flags & RENAME_EXCHANGE) && target) {
5134 if (is_dir) {
5135 shrink_dcache_parent(new_dentry);
5136 target->i_flags |= S_DEAD;
5137 }
5138 dont_mount(dentry: new_dentry);
5139 detach_mounts(dentry: new_dentry);
5140 }
5141 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
5142 if (!(flags & RENAME_EXCHANGE))
5143 d_move(old_dentry, new_dentry);
5144 else
5145 d_exchange(old_dentry, new_dentry);
5146 }
5147out:
5148 if (!is_dir || lock_old_subdir)
5149 inode_unlock(inode: source);
5150 if (target && (!new_is_dir || lock_new_subdir))
5151 inode_unlock(inode: target);
5152 dput(new_dentry);
5153 if (!error) {
5154 fsnotify_move(old_dir, new_dir, old_name: &old_name.name, isdir: is_dir,
5155 target: !(flags & RENAME_EXCHANGE) ? target : NULL, moved: old_dentry);
5156 if (flags & RENAME_EXCHANGE) {
5157 fsnotify_move(old_dir: new_dir, new_dir: old_dir, old_name: &old_dentry->d_name,
5158 isdir: new_is_dir, NULL, moved: new_dentry);
5159 }
5160 }
5161 release_dentry_name_snapshot(&old_name);
5162
5163 return error;
5164}
5165EXPORT_SYMBOL(vfs_rename);
5166
5167int do_renameat2(int olddfd, struct filename *from, int newdfd,
5168 struct filename *to, unsigned int flags)
5169{
5170 struct renamedata rd;
5171 struct dentry *old_dentry, *new_dentry;
5172 struct dentry *trap;
5173 struct path old_path, new_path;
5174 struct qstr old_last, new_last;
5175 int old_type, new_type;
5176 struct inode *delegated_inode = NULL;
5177 unsigned int lookup_flags = 0, target_flags =
5178 LOOKUP_RENAME_TARGET | LOOKUP_CREATE;
5179 bool should_retry = false;
5180 int error = -EINVAL;
5181
5182 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
5183 goto put_names;
5184
5185 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
5186 (flags & RENAME_EXCHANGE))
5187 goto put_names;
5188
5189 if (flags & RENAME_EXCHANGE)
5190 target_flags = 0;
5191 if (flags & RENAME_NOREPLACE)
5192 target_flags |= LOOKUP_EXCL;
5193
5194retry:
5195 error = filename_parentat(dfd: olddfd, name: from, flags: lookup_flags, parent: &old_path,
5196 last: &old_last, type: &old_type);
5197 if (error)
5198 goto put_names;
5199
5200 error = filename_parentat(dfd: newdfd, name: to, flags: lookup_flags, parent: &new_path, last: &new_last,
5201 type: &new_type);
5202 if (error)
5203 goto exit1;
5204
5205 error = -EXDEV;
5206 if (old_path.mnt != new_path.mnt)
5207 goto exit2;
5208
5209 error = -EBUSY;
5210 if (old_type != LAST_NORM)
5211 goto exit2;
5212
5213 if (flags & RENAME_NOREPLACE)
5214 error = -EEXIST;
5215 if (new_type != LAST_NORM)
5216 goto exit2;
5217
5218 error = mnt_want_write(mnt: old_path.mnt);
5219 if (error)
5220 goto exit2;
5221
5222retry_deleg:
5223 trap = lock_rename(new_path.dentry, old_path.dentry);
5224 if (IS_ERR(ptr: trap)) {
5225 error = PTR_ERR(ptr: trap);
5226 goto exit_lock_rename;
5227 }
5228
5229 old_dentry = lookup_one_qstr_excl(&old_last, old_path.dentry,
5230 lookup_flags);
5231 error = PTR_ERR(ptr: old_dentry);
5232 if (IS_ERR(ptr: old_dentry))
5233 goto exit3;
5234 new_dentry = lookup_one_qstr_excl(&new_last, new_path.dentry,
5235 lookup_flags | target_flags);
5236 error = PTR_ERR(ptr: new_dentry);
5237 if (IS_ERR(ptr: new_dentry))
5238 goto exit4;
5239 if (flags & RENAME_EXCHANGE) {
5240 if (!d_is_dir(dentry: new_dentry)) {
5241 error = -ENOTDIR;
5242 if (new_last.name[new_last.len])
5243 goto exit5;
5244 }
5245 }
5246 /* unless the source is a directory trailing slashes give -ENOTDIR */
5247 if (!d_is_dir(dentry: old_dentry)) {
5248 error = -ENOTDIR;
5249 if (old_last.name[old_last.len])
5250 goto exit5;
5251 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
5252 goto exit5;
5253 }
5254 /* source should not be ancestor of target */
5255 error = -EINVAL;
5256 if (old_dentry == trap)
5257 goto exit5;
5258 /* target should not be an ancestor of source */
5259 if (!(flags & RENAME_EXCHANGE))
5260 error = -ENOTEMPTY;
5261 if (new_dentry == trap)
5262 goto exit5;
5263
5264 error = security_path_rename(old_dir: &old_path, old_dentry,
5265 new_dir: &new_path, new_dentry, flags);
5266 if (error)
5267 goto exit5;
5268
5269 rd.old_dir = old_path.dentry->d_inode;
5270 rd.old_dentry = old_dentry;
5271 rd.old_mnt_idmap = mnt_idmap(mnt: old_path.mnt);
5272 rd.new_dir = new_path.dentry->d_inode;
5273 rd.new_dentry = new_dentry;
5274 rd.new_mnt_idmap = mnt_idmap(mnt: new_path.mnt);
5275 rd.delegated_inode = &delegated_inode;
5276 rd.flags = flags;
5277 error = vfs_rename(&rd);
5278exit5:
5279 dput(new_dentry);
5280exit4:
5281 dput(old_dentry);
5282exit3:
5283 unlock_rename(new_path.dentry, old_path.dentry);
5284exit_lock_rename:
5285 if (delegated_inode) {
5286 error = break_deleg_wait(delegated_inode: &delegated_inode);
5287 if (!error)
5288 goto retry_deleg;
5289 }
5290 mnt_drop_write(mnt: old_path.mnt);
5291exit2:
5292 if (retry_estale(error, flags: lookup_flags))
5293 should_retry = true;
5294 path_put(&new_path);
5295exit1:
5296 path_put(&old_path);
5297 if (should_retry) {
5298 should_retry = false;
5299 lookup_flags |= LOOKUP_REVAL;
5300 goto retry;
5301 }
5302put_names:
5303 putname(from);
5304 putname(to);
5305 return error;
5306}
5307
5308SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
5309 int, newdfd, const char __user *, newname, unsigned int, flags)
5310{
5311 return do_renameat2(olddfd, from: getname(name: oldname), newdfd, to: getname(name: newname),
5312 flags);
5313}
5314
5315SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
5316 int, newdfd, const char __user *, newname)
5317{
5318 return do_renameat2(olddfd, from: getname(name: oldname), newdfd, to: getname(name: newname),
5319 flags: 0);
5320}
5321
5322SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5323{
5324 return do_renameat2(AT_FDCWD, from: getname(name: oldname), AT_FDCWD,
5325 to: getname(name: newname), flags: 0);
5326}
5327
5328int readlink_copy(char __user *buffer, int buflen, const char *link, int linklen)
5329{
5330 int copylen;
5331
5332 copylen = linklen;
5333 if (unlikely(copylen > (unsigned) buflen))
5334 copylen = buflen;
5335 if (copy_to_user(to: buffer, from: link, n: copylen))
5336 copylen = -EFAULT;
5337 return copylen;
5338}
5339
5340/**
5341 * vfs_readlink - copy symlink body into userspace buffer
5342 * @dentry: dentry on which to get symbolic link
5343 * @buffer: user memory pointer
5344 * @buflen: size of buffer
5345 *
5346 * Does not touch atime. That's up to the caller if necessary
5347 *
5348 * Does not call security hook.
5349 */
5350int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5351{
5352 struct inode *inode = d_inode(dentry);
5353 DEFINE_DELAYED_CALL(done);
5354 const char *link;
5355 int res;
5356
5357 if (inode->i_opflags & IOP_CACHED_LINK)
5358 return readlink_copy(buffer, buflen, link: inode->i_link, linklen: inode->i_linklen);
5359
5360 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
5361 if (unlikely(inode->i_op->readlink))
5362 return inode->i_op->readlink(dentry, buffer, buflen);
5363
5364 if (!d_is_symlink(dentry))
5365 return -EINVAL;
5366
5367 spin_lock(lock: &inode->i_lock);
5368 inode->i_opflags |= IOP_DEFAULT_READLINK;
5369 spin_unlock(lock: &inode->i_lock);
5370 }
5371
5372 link = READ_ONCE(inode->i_link);
5373 if (!link) {
5374 link = inode->i_op->get_link(dentry, inode, &done);
5375 if (IS_ERR(ptr: link))
5376 return PTR_ERR(ptr: link);
5377 }
5378 res = readlink_copy(buffer, buflen, link, strlen(link));
5379 do_delayed_call(call: &done);
5380 return res;
5381}
5382EXPORT_SYMBOL(vfs_readlink);
5383
5384/**
5385 * vfs_get_link - get symlink body
5386 * @dentry: dentry on which to get symbolic link
5387 * @done: caller needs to free returned data with this
5388 *
5389 * Calls security hook and i_op->get_link() on the supplied inode.
5390 *
5391 * It does not touch atime. That's up to the caller if necessary.
5392 *
5393 * Does not work on "special" symlinks like /proc/$$/fd/N
5394 */
5395const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
5396{
5397 const char *res = ERR_PTR(error: -EINVAL);
5398 struct inode *inode = d_inode(dentry);
5399
5400 if (d_is_symlink(dentry)) {
5401 res = ERR_PTR(error: security_inode_readlink(dentry));
5402 if (!res)
5403 res = inode->i_op->get_link(dentry, inode, done);
5404 }
5405 return res;
5406}
5407EXPORT_SYMBOL(vfs_get_link);
5408
5409/* get the link contents into pagecache */
5410static char *__page_get_link(struct dentry *dentry, struct inode *inode,
5411 struct delayed_call *callback)
5412{
5413 struct folio *folio;
5414 struct address_space *mapping = inode->i_mapping;
5415
5416 if (!dentry) {
5417 folio = filemap_get_folio(mapping, index: 0);
5418 if (IS_ERR(ptr: folio))
5419 return ERR_PTR(error: -ECHILD);
5420 if (!folio_test_uptodate(folio)) {
5421 folio_put(folio);
5422 return ERR_PTR(error: -ECHILD);
5423 }
5424 } else {
5425 folio = read_mapping_folio(mapping, index: 0, NULL);
5426 if (IS_ERR(ptr: folio))
5427 return ERR_CAST(ptr: folio);
5428 }
5429 set_delayed_call(call: callback, fn: page_put_link, arg: folio);
5430 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
5431 return folio_address(folio);
5432}
5433
5434const char *page_get_link_raw(struct dentry *dentry, struct inode *inode,
5435 struct delayed_call *callback)
5436{
5437 return __page_get_link(dentry, inode, callback);
5438}
5439EXPORT_SYMBOL_GPL(page_get_link_raw);
5440
5441/**
5442 * page_get_link() - An implementation of the get_link inode_operation.
5443 * @dentry: The directory entry which is the symlink.
5444 * @inode: The inode for the symlink.
5445 * @callback: Used to drop the reference to the symlink.
5446 *
5447 * Filesystems which store their symlinks in the page cache should use
5448 * this to implement the get_link() member of their inode_operations.
5449 *
5450 * Return: A pointer to the NUL-terminated symlink.
5451 */
5452const char *page_get_link(struct dentry *dentry, struct inode *inode,
5453 struct delayed_call *callback)
5454{
5455 char *kaddr = __page_get_link(dentry, inode, callback);
5456
5457 if (!IS_ERR(ptr: kaddr))
5458 nd_terminate_link(name: kaddr, len: inode->i_size, PAGE_SIZE - 1);
5459 return kaddr;
5460}
5461EXPORT_SYMBOL(page_get_link);
5462
5463/**
5464 * page_put_link() - Drop the reference to the symlink.
5465 * @arg: The folio which contains the symlink.
5466 *
5467 * This is used internally by page_get_link(). It is exported for use
5468 * by filesystems which need to implement a variant of page_get_link()
5469 * themselves. Despite the apparent symmetry, filesystems which use
5470 * page_get_link() do not need to call page_put_link().
5471 *
5472 * The argument, while it has a void pointer type, must be a pointer to
5473 * the folio which was retrieved from the page cache. The delayed_call
5474 * infrastructure is used to drop the reference count once the caller
5475 * is done with the symlink.
5476 */
5477void page_put_link(void *arg)
5478{
5479 folio_put(folio: arg);
5480}
5481EXPORT_SYMBOL(page_put_link);
5482
5483int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5484{
5485 const char *link;
5486 int res;
5487
5488 DEFINE_DELAYED_CALL(done);
5489 link = page_get_link(dentry, d_inode(dentry), &done);
5490 res = PTR_ERR(ptr: link);
5491 if (!IS_ERR(ptr: link))
5492 res = readlink_copy(buffer, buflen, link, strlen(link));
5493 do_delayed_call(call: &done);
5494 return res;
5495}
5496EXPORT_SYMBOL(page_readlink);
5497
5498int page_symlink(struct inode *inode, const char *symname, int len)
5499{
5500 struct address_space *mapping = inode->i_mapping;
5501 const struct address_space_operations *aops = mapping->a_ops;
5502 bool nofs = !mapping_gfp_constraint(mapping, __GFP_FS);
5503 struct folio *folio;
5504 void *fsdata = NULL;
5505 int err;
5506 unsigned int flags;
5507
5508retry:
5509 if (nofs)
5510 flags = memalloc_nofs_save();
5511 err = aops->write_begin(NULL, mapping, 0, len-1, &folio, &fsdata);
5512 if (nofs)
5513 memalloc_nofs_restore(flags);
5514 if (err)
5515 goto fail;
5516
5517 memcpy(folio_address(folio), symname, len - 1);
5518
5519 err = aops->write_end(NULL, mapping, 0, len - 1, len - 1,
5520 folio, fsdata);
5521 if (err < 0)
5522 goto fail;
5523 if (err < len-1)
5524 goto retry;
5525
5526 mark_inode_dirty(inode);
5527 return 0;
5528fail:
5529 return err;
5530}
5531EXPORT_SYMBOL(page_symlink);
5532
5533const struct inode_operations page_symlink_inode_operations = {
5534 .get_link = page_get_link,
5535};
5536EXPORT_SYMBOL(page_symlink_inode_operations);
5537

Provided by KDAB

Privacy Policy
Improve your Profiling and Debugging skills
Find out more

source code of linux/fs/namei.c