1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/namei.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8/*
9 * Some corrections by tytso.
10 */
11
12/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
13 * lookup logic.
14 */
15/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
16 */
17
18#include <linux/init.h>
19#include <linux/export.h>
20#include <linux/kernel.h>
21#include <linux/slab.h>
22#include <linux/fs.h>
23#include <linux/filelock.h>
24#include <linux/namei.h>
25#include <linux/pagemap.h>
26#include <linux/sched/mm.h>
27#include <linux/fsnotify.h>
28#include <linux/personality.h>
29#include <linux/security.h>
30#include <linux/ima.h>
31#include <linux/syscalls.h>
32#include <linux/mount.h>
33#include <linux/audit.h>
34#include <linux/capability.h>
35#include <linux/file.h>
36#include <linux/fcntl.h>
37#include <linux/device_cgroup.h>
38#include <linux/fs_struct.h>
39#include <linux/posix_acl.h>
40#include <linux/hash.h>
41#include <linux/bitops.h>
42#include <linux/init_task.h>
43#include <linux/uaccess.h>
44
45#include "internal.h"
46#include "mount.h"
47
48/* [Feb-1997 T. Schoebel-Theuer]
49 * Fundamental changes in the pathname lookup mechanisms (namei)
50 * were necessary because of omirr. The reason is that omirr needs
51 * to know the _real_ pathname, not the user-supplied one, in case
52 * of symlinks (and also when transname replacements occur).
53 *
54 * The new code replaces the old recursive symlink resolution with
55 * an iterative one (in case of non-nested symlink chains). It does
56 * this with calls to <fs>_follow_link().
57 * As a side effect, dir_namei(), _namei() and follow_link() are now
58 * replaced with a single function lookup_dentry() that can handle all
59 * the special cases of the former code.
60 *
61 * With the new dcache, the pathname is stored at each inode, at least as
62 * long as the refcount of the inode is positive. As a side effect, the
63 * size of the dcache depends on the inode cache and thus is dynamic.
64 *
65 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
66 * resolution to correspond with current state of the code.
67 *
68 * Note that the symlink resolution is not *completely* iterative.
69 * There is still a significant amount of tail- and mid- recursion in
70 * the algorithm. Also, note that <fs>_readlink() is not used in
71 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
72 * may return different results than <fs>_follow_link(). Many virtual
73 * filesystems (including /proc) exhibit this behavior.
74 */
75
76/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
77 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
78 * and the name already exists in form of a symlink, try to create the new
79 * name indicated by the symlink. The old code always complained that the
80 * name already exists, due to not following the symlink even if its target
81 * is nonexistent. The new semantics affects also mknod() and link() when
82 * the name is a symlink pointing to a non-existent name.
83 *
84 * I don't know which semantics is the right one, since I have no access
85 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
86 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
87 * "old" one. Personally, I think the new semantics is much more logical.
88 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
89 * file does succeed in both HP-UX and SunOs, but not in Solaris
90 * and in the old Linux semantics.
91 */
92
93/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
94 * semantics. See the comments in "open_namei" and "do_link" below.
95 *
96 * [10-Sep-98 Alan Modra] Another symlink change.
97 */
98
99/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
100 * inside the path - always follow.
101 * in the last component in creation/removal/renaming - never follow.
102 * if LOOKUP_FOLLOW passed - follow.
103 * if the pathname has trailing slashes - follow.
104 * otherwise - don't follow.
105 * (applied in that order).
106 *
107 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
108 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
109 * During the 2.4 we need to fix the userland stuff depending on it -
110 * hopefully we will be able to get rid of that wart in 2.5. So far only
111 * XEmacs seems to be relying on it...
112 */
113/*
114 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
115 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
116 * any extra contention...
117 */
118
119/* In order to reduce some races, while at the same time doing additional
120 * checking and hopefully speeding things up, we copy filenames to the
121 * kernel data space before using them..
122 *
123 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
124 * PATH_MAX includes the nul terminator --RR.
125 */
126
127#define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
128
129struct filename *
130getname_flags(const char __user *filename, int flags, int *empty)
131{
132 struct filename *result;
133 char *kname;
134 int len;
135
136 result = audit_reusename(name: filename);
137 if (result)
138 return result;
139
140 result = __getname();
141 if (unlikely(!result))
142 return ERR_PTR(error: -ENOMEM);
143
144 /*
145 * First, try to embed the struct filename inside the names_cache
146 * allocation
147 */
148 kname = (char *)result->iname;
149 result->name = kname;
150
151 len = strncpy_from_user(dst: kname, src: filename, EMBEDDED_NAME_MAX);
152 if (unlikely(len < 0)) {
153 __putname(result);
154 return ERR_PTR(error: len);
155 }
156
157 /*
158 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
159 * separate struct filename so we can dedicate the entire
160 * names_cache allocation for the pathname, and re-do the copy from
161 * userland.
162 */
163 if (unlikely(len == EMBEDDED_NAME_MAX)) {
164 const size_t size = offsetof(struct filename, iname[1]);
165 kname = (char *)result;
166
167 /*
168 * size is chosen that way we to guarantee that
169 * result->iname[0] is within the same object and that
170 * kname can't be equal to result->iname, no matter what.
171 */
172 result = kzalloc(size, GFP_KERNEL);
173 if (unlikely(!result)) {
174 __putname(kname);
175 return ERR_PTR(error: -ENOMEM);
176 }
177 result->name = kname;
178 len = strncpy_from_user(dst: kname, src: filename, PATH_MAX);
179 if (unlikely(len < 0)) {
180 __putname(kname);
181 kfree(objp: result);
182 return ERR_PTR(error: len);
183 }
184 if (unlikely(len == PATH_MAX)) {
185 __putname(kname);
186 kfree(objp: result);
187 return ERR_PTR(error: -ENAMETOOLONG);
188 }
189 }
190
191 atomic_set(v: &result->refcnt, i: 1);
192 /* The empty path is special. */
193 if (unlikely(!len)) {
194 if (empty)
195 *empty = 1;
196 if (!(flags & LOOKUP_EMPTY)) {
197 putname(name: result);
198 return ERR_PTR(error: -ENOENT);
199 }
200 }
201
202 result->uptr = filename;
203 result->aname = NULL;
204 audit_getname(name: result);
205 return result;
206}
207
208struct filename *
209getname_uflags(const char __user *filename, int uflags)
210{
211 int flags = (uflags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
212
213 return getname_flags(filename, flags, NULL);
214}
215
216struct filename *
217getname(const char __user * filename)
218{
219 return getname_flags(filename, flags: 0, NULL);
220}
221
222struct filename *
223getname_kernel(const char * filename)
224{
225 struct filename *result;
226 int len = strlen(filename) + 1;
227
228 result = __getname();
229 if (unlikely(!result))
230 return ERR_PTR(error: -ENOMEM);
231
232 if (len <= EMBEDDED_NAME_MAX) {
233 result->name = (char *)result->iname;
234 } else if (len <= PATH_MAX) {
235 const size_t size = offsetof(struct filename, iname[1]);
236 struct filename *tmp;
237
238 tmp = kmalloc(size, GFP_KERNEL);
239 if (unlikely(!tmp)) {
240 __putname(result);
241 return ERR_PTR(error: -ENOMEM);
242 }
243 tmp->name = (char *)result;
244 result = tmp;
245 } else {
246 __putname(result);
247 return ERR_PTR(error: -ENAMETOOLONG);
248 }
249 memcpy((char *)result->name, filename, len);
250 result->uptr = NULL;
251 result->aname = NULL;
252 atomic_set(v: &result->refcnt, i: 1);
253 audit_getname(name: result);
254
255 return result;
256}
257EXPORT_SYMBOL(getname_kernel);
258
259void putname(struct filename *name)
260{
261 if (IS_ERR(ptr: name))
262 return;
263
264 if (WARN_ON_ONCE(!atomic_read(&name->refcnt)))
265 return;
266
267 if (!atomic_dec_and_test(v: &name->refcnt))
268 return;
269
270 if (name->name != name->iname) {
271 __putname(name->name);
272 kfree(objp: name);
273 } else
274 __putname(name);
275}
276EXPORT_SYMBOL(putname);
277
278/**
279 * check_acl - perform ACL permission checking
280 * @idmap: idmap of the mount the inode was found from
281 * @inode: inode to check permissions on
282 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
283 *
284 * This function performs the ACL permission checking. Since this function
285 * retrieve POSIX acls it needs to know whether it is called from a blocking or
286 * non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
287 *
288 * If the inode has been found through an idmapped mount the idmap of
289 * the vfsmount must be passed through @idmap. This function will then take
290 * care to map the inode according to @idmap before checking permissions.
291 * On non-idmapped mounts or if permission checking is to be performed on the
292 * raw inode simply passs @nop_mnt_idmap.
293 */
294static int check_acl(struct mnt_idmap *idmap,
295 struct inode *inode, int mask)
296{
297#ifdef CONFIG_FS_POSIX_ACL
298 struct posix_acl *acl;
299
300 if (mask & MAY_NOT_BLOCK) {
301 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
302 if (!acl)
303 return -EAGAIN;
304 /* no ->get_inode_acl() calls in RCU mode... */
305 if (is_uncached_acl(acl))
306 return -ECHILD;
307 return posix_acl_permission(idmap, inode, acl, mask);
308 }
309
310 acl = get_inode_acl(inode, ACL_TYPE_ACCESS);
311 if (IS_ERR(ptr: acl))
312 return PTR_ERR(ptr: acl);
313 if (acl) {
314 int error = posix_acl_permission(idmap, inode, acl, mask);
315 posix_acl_release(acl);
316 return error;
317 }
318#endif
319
320 return -EAGAIN;
321}
322
323/**
324 * acl_permission_check - perform basic UNIX permission checking
325 * @idmap: idmap of the mount the inode was found from
326 * @inode: inode to check permissions on
327 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
328 *
329 * This function performs the basic UNIX permission checking. Since this
330 * function may retrieve POSIX acls it needs to know whether it is called from a
331 * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
332 *
333 * If the inode has been found through an idmapped mount the idmap of
334 * the vfsmount must be passed through @idmap. This function will then take
335 * care to map the inode according to @idmap before checking permissions.
336 * On non-idmapped mounts or if permission checking is to be performed on the
337 * raw inode simply passs @nop_mnt_idmap.
338 */
339static int acl_permission_check(struct mnt_idmap *idmap,
340 struct inode *inode, int mask)
341{
342 unsigned int mode = inode->i_mode;
343 vfsuid_t vfsuid;
344
345 /* Are we the owner? If so, ACL's don't matter */
346 vfsuid = i_uid_into_vfsuid(idmap, inode);
347 if (likely(vfsuid_eq_kuid(vfsuid, current_fsuid()))) {
348 mask &= 7;
349 mode >>= 6;
350 return (mask & ~mode) ? -EACCES : 0;
351 }
352
353 /* Do we have ACL's? */
354 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
355 int error = check_acl(idmap, inode, mask);
356 if (error != -EAGAIN)
357 return error;
358 }
359
360 /* Only RWX matters for group/other mode bits */
361 mask &= 7;
362
363 /*
364 * Are the group permissions different from
365 * the other permissions in the bits we care
366 * about? Need to check group ownership if so.
367 */
368 if (mask & (mode ^ (mode >> 3))) {
369 vfsgid_t vfsgid = i_gid_into_vfsgid(idmap, inode);
370 if (vfsgid_in_group_p(vfsgid))
371 mode >>= 3;
372 }
373
374 /* Bits in 'mode' clear that we require? */
375 return (mask & ~mode) ? -EACCES : 0;
376}
377
378/**
379 * generic_permission - check for access rights on a Posix-like filesystem
380 * @idmap: idmap of the mount the inode was found from
381 * @inode: inode to check access rights for
382 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC,
383 * %MAY_NOT_BLOCK ...)
384 *
385 * Used to check for read/write/execute permissions on a file.
386 * We use "fsuid" for this, letting us set arbitrary permissions
387 * for filesystem access without changing the "normal" uids which
388 * are used for other things.
389 *
390 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
391 * request cannot be satisfied (eg. requires blocking or too much complexity).
392 * It would then be called again in ref-walk mode.
393 *
394 * If the inode has been found through an idmapped mount the idmap of
395 * the vfsmount must be passed through @idmap. This function will then take
396 * care to map the inode according to @idmap before checking permissions.
397 * On non-idmapped mounts or if permission checking is to be performed on the
398 * raw inode simply passs @nop_mnt_idmap.
399 */
400int generic_permission(struct mnt_idmap *idmap, struct inode *inode,
401 int mask)
402{
403 int ret;
404
405 /*
406 * Do the basic permission checks.
407 */
408 ret = acl_permission_check(idmap, inode, mask);
409 if (ret != -EACCES)
410 return ret;
411
412 if (S_ISDIR(inode->i_mode)) {
413 /* DACs are overridable for directories */
414 if (!(mask & MAY_WRITE))
415 if (capable_wrt_inode_uidgid(idmap, inode,
416 CAP_DAC_READ_SEARCH))
417 return 0;
418 if (capable_wrt_inode_uidgid(idmap, inode,
419 CAP_DAC_OVERRIDE))
420 return 0;
421 return -EACCES;
422 }
423
424 /*
425 * Searching includes executable on directories, else just read.
426 */
427 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
428 if (mask == MAY_READ)
429 if (capable_wrt_inode_uidgid(idmap, inode,
430 CAP_DAC_READ_SEARCH))
431 return 0;
432 /*
433 * Read/write DACs are always overridable.
434 * Executable DACs are overridable when there is
435 * at least one exec bit set.
436 */
437 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
438 if (capable_wrt_inode_uidgid(idmap, inode,
439 CAP_DAC_OVERRIDE))
440 return 0;
441
442 return -EACCES;
443}
444EXPORT_SYMBOL(generic_permission);
445
446/**
447 * do_inode_permission - UNIX permission checking
448 * @idmap: idmap of the mount the inode was found from
449 * @inode: inode to check permissions on
450 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
451 *
452 * We _really_ want to just do "generic_permission()" without
453 * even looking at the inode->i_op values. So we keep a cache
454 * flag in inode->i_opflags, that says "this has not special
455 * permission function, use the fast case".
456 */
457static inline int do_inode_permission(struct mnt_idmap *idmap,
458 struct inode *inode, int mask)
459{
460 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
461 if (likely(inode->i_op->permission))
462 return inode->i_op->permission(idmap, inode, mask);
463
464 /* This gets set once for the inode lifetime */
465 spin_lock(lock: &inode->i_lock);
466 inode->i_opflags |= IOP_FASTPERM;
467 spin_unlock(lock: &inode->i_lock);
468 }
469 return generic_permission(idmap, inode, mask);
470}
471
472/**
473 * sb_permission - Check superblock-level permissions
474 * @sb: Superblock of inode to check permission on
475 * @inode: Inode to check permission on
476 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
477 *
478 * Separate out file-system wide checks from inode-specific permission checks.
479 */
480static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
481{
482 if (unlikely(mask & MAY_WRITE)) {
483 umode_t mode = inode->i_mode;
484
485 /* Nobody gets write access to a read-only fs. */
486 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
487 return -EROFS;
488 }
489 return 0;
490}
491
492/**
493 * inode_permission - Check for access rights to a given inode
494 * @idmap: idmap of the mount the inode was found from
495 * @inode: Inode to check permission on
496 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
497 *
498 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
499 * this, letting us set arbitrary permissions for filesystem access without
500 * changing the "normal" UIDs which are used for other things.
501 *
502 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
503 */
504int inode_permission(struct mnt_idmap *idmap,
505 struct inode *inode, int mask)
506{
507 int retval;
508
509 retval = sb_permission(sb: inode->i_sb, inode, mask);
510 if (retval)
511 return retval;
512
513 if (unlikely(mask & MAY_WRITE)) {
514 /*
515 * Nobody gets write access to an immutable file.
516 */
517 if (IS_IMMUTABLE(inode))
518 return -EPERM;
519
520 /*
521 * Updating mtime will likely cause i_uid and i_gid to be
522 * written back improperly if their true value is unknown
523 * to the vfs.
524 */
525 if (HAS_UNMAPPED_ID(idmap, inode))
526 return -EACCES;
527 }
528
529 retval = do_inode_permission(idmap, inode, mask);
530 if (retval)
531 return retval;
532
533 retval = devcgroup_inode_permission(inode, mask);
534 if (retval)
535 return retval;
536
537 return security_inode_permission(inode, mask);
538}
539EXPORT_SYMBOL(inode_permission);
540
541/**
542 * path_get - get a reference to a path
543 * @path: path to get the reference to
544 *
545 * Given a path increment the reference count to the dentry and the vfsmount.
546 */
547void path_get(const struct path *path)
548{
549 mntget(mnt: path->mnt);
550 dget(dentry: path->dentry);
551}
552EXPORT_SYMBOL(path_get);
553
554/**
555 * path_put - put a reference to a path
556 * @path: path to put the reference to
557 *
558 * Given a path decrement the reference count to the dentry and the vfsmount.
559 */
560void path_put(const struct path *path)
561{
562 dput(path->dentry);
563 mntput(mnt: path->mnt);
564}
565EXPORT_SYMBOL(path_put);
566
567#define EMBEDDED_LEVELS 2
568struct nameidata {
569 struct path path;
570 struct qstr last;
571 struct path root;
572 struct inode *inode; /* path.dentry.d_inode */
573 unsigned int flags, state;
574 unsigned seq, next_seq, m_seq, r_seq;
575 int last_type;
576 unsigned depth;
577 int total_link_count;
578 struct saved {
579 struct path link;
580 struct delayed_call done;
581 const char *name;
582 unsigned seq;
583 } *stack, internal[EMBEDDED_LEVELS];
584 struct filename *name;
585 struct nameidata *saved;
586 unsigned root_seq;
587 int dfd;
588 vfsuid_t dir_vfsuid;
589 umode_t dir_mode;
590} __randomize_layout;
591
592#define ND_ROOT_PRESET 1
593#define ND_ROOT_GRABBED 2
594#define ND_JUMPED 4
595
596static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name)
597{
598 struct nameidata *old = current->nameidata;
599 p->stack = p->internal;
600 p->depth = 0;
601 p->dfd = dfd;
602 p->name = name;
603 p->path.mnt = NULL;
604 p->path.dentry = NULL;
605 p->total_link_count = old ? old->total_link_count : 0;
606 p->saved = old;
607 current->nameidata = p;
608}
609
610static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name,
611 const struct path *root)
612{
613 __set_nameidata(p, dfd, name);
614 p->state = 0;
615 if (unlikely(root)) {
616 p->state = ND_ROOT_PRESET;
617 p->root = *root;
618 }
619}
620
621static void restore_nameidata(void)
622{
623 struct nameidata *now = current->nameidata, *old = now->saved;
624
625 current->nameidata = old;
626 if (old)
627 old->total_link_count = now->total_link_count;
628 if (now->stack != now->internal)
629 kfree(objp: now->stack);
630}
631
632static bool nd_alloc_stack(struct nameidata *nd)
633{
634 struct saved *p;
635
636 p= kmalloc_array(MAXSYMLINKS, size: sizeof(struct saved),
637 flags: nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL);
638 if (unlikely(!p))
639 return false;
640 memcpy(p, nd->internal, sizeof(nd->internal));
641 nd->stack = p;
642 return true;
643}
644
645/**
646 * path_connected - Verify that a dentry is below mnt.mnt_root
647 * @mnt: The mountpoint to check.
648 * @dentry: The dentry to check.
649 *
650 * Rename can sometimes move a file or directory outside of a bind
651 * mount, path_connected allows those cases to be detected.
652 */
653static bool path_connected(struct vfsmount *mnt, struct dentry *dentry)
654{
655 struct super_block *sb = mnt->mnt_sb;
656
657 /* Bind mounts can have disconnected paths */
658 if (mnt->mnt_root == sb->s_root)
659 return true;
660
661 return is_subdir(dentry, mnt->mnt_root);
662}
663
664static void drop_links(struct nameidata *nd)
665{
666 int i = nd->depth;
667 while (i--) {
668 struct saved *last = nd->stack + i;
669 do_delayed_call(call: &last->done);
670 clear_delayed_call(call: &last->done);
671 }
672}
673
674static void leave_rcu(struct nameidata *nd)
675{
676 nd->flags &= ~LOOKUP_RCU;
677 nd->seq = nd->next_seq = 0;
678 rcu_read_unlock();
679}
680
681static void terminate_walk(struct nameidata *nd)
682{
683 drop_links(nd);
684 if (!(nd->flags & LOOKUP_RCU)) {
685 int i;
686 path_put(&nd->path);
687 for (i = 0; i < nd->depth; i++)
688 path_put(&nd->stack[i].link);
689 if (nd->state & ND_ROOT_GRABBED) {
690 path_put(&nd->root);
691 nd->state &= ~ND_ROOT_GRABBED;
692 }
693 } else {
694 leave_rcu(nd);
695 }
696 nd->depth = 0;
697 nd->path.mnt = NULL;
698 nd->path.dentry = NULL;
699}
700
701/* path_put is needed afterwards regardless of success or failure */
702static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq)
703{
704 int res = __legitimize_mnt(path->mnt, mseq);
705 if (unlikely(res)) {
706 if (res > 0)
707 path->mnt = NULL;
708 path->dentry = NULL;
709 return false;
710 }
711 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
712 path->dentry = NULL;
713 return false;
714 }
715 return !read_seqcount_retry(&path->dentry->d_seq, seq);
716}
717
718static inline bool legitimize_path(struct nameidata *nd,
719 struct path *path, unsigned seq)
720{
721 return __legitimize_path(path, seq, mseq: nd->m_seq);
722}
723
724static bool legitimize_links(struct nameidata *nd)
725{
726 int i;
727 if (unlikely(nd->flags & LOOKUP_CACHED)) {
728 drop_links(nd);
729 nd->depth = 0;
730 return false;
731 }
732 for (i = 0; i < nd->depth; i++) {
733 struct saved *last = nd->stack + i;
734 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
735 drop_links(nd);
736 nd->depth = i + 1;
737 return false;
738 }
739 }
740 return true;
741}
742
743static bool legitimize_root(struct nameidata *nd)
744{
745 /* Nothing to do if nd->root is zero or is managed by the VFS user. */
746 if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET))
747 return true;
748 nd->state |= ND_ROOT_GRABBED;
749 return legitimize_path(nd, path: &nd->root, seq: nd->root_seq);
750}
751
752/*
753 * Path walking has 2 modes, rcu-walk and ref-walk (see
754 * Documentation/filesystems/path-lookup.txt). In situations when we can't
755 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
756 * normal reference counts on dentries and vfsmounts to transition to ref-walk
757 * mode. Refcounts are grabbed at the last known good point before rcu-walk
758 * got stuck, so ref-walk may continue from there. If this is not successful
759 * (eg. a seqcount has changed), then failure is returned and it's up to caller
760 * to restart the path walk from the beginning in ref-walk mode.
761 */
762
763/**
764 * try_to_unlazy - try to switch to ref-walk mode.
765 * @nd: nameidata pathwalk data
766 * Returns: true on success, false on failure
767 *
768 * try_to_unlazy attempts to legitimize the current nd->path and nd->root
769 * for ref-walk mode.
770 * Must be called from rcu-walk context.
771 * Nothing should touch nameidata between try_to_unlazy() failure and
772 * terminate_walk().
773 */
774static bool try_to_unlazy(struct nameidata *nd)
775{
776 struct dentry *parent = nd->path.dentry;
777
778 BUG_ON(!(nd->flags & LOOKUP_RCU));
779
780 if (unlikely(!legitimize_links(nd)))
781 goto out1;
782 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
783 goto out;
784 if (unlikely(!legitimize_root(nd)))
785 goto out;
786 leave_rcu(nd);
787 BUG_ON(nd->inode != parent->d_inode);
788 return true;
789
790out1:
791 nd->path.mnt = NULL;
792 nd->path.dentry = NULL;
793out:
794 leave_rcu(nd);
795 return false;
796}
797
798/**
799 * try_to_unlazy_next - try to switch to ref-walk mode.
800 * @nd: nameidata pathwalk data
801 * @dentry: next dentry to step into
802 * Returns: true on success, false on failure
803 *
804 * Similar to try_to_unlazy(), but here we have the next dentry already
805 * picked by rcu-walk and want to legitimize that in addition to the current
806 * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context.
807 * Nothing should touch nameidata between try_to_unlazy_next() failure and
808 * terminate_walk().
809 */
810static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry)
811{
812 int res;
813 BUG_ON(!(nd->flags & LOOKUP_RCU));
814
815 if (unlikely(!legitimize_links(nd)))
816 goto out2;
817 res = __legitimize_mnt(nd->path.mnt, nd->m_seq);
818 if (unlikely(res)) {
819 if (res > 0)
820 goto out2;
821 goto out1;
822 }
823 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
824 goto out1;
825
826 /*
827 * We need to move both the parent and the dentry from the RCU domain
828 * to be properly refcounted. And the sequence number in the dentry
829 * validates *both* dentry counters, since we checked the sequence
830 * number of the parent after we got the child sequence number. So we
831 * know the parent must still be valid if the child sequence number is
832 */
833 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
834 goto out;
835 if (read_seqcount_retry(&dentry->d_seq, nd->next_seq))
836 goto out_dput;
837 /*
838 * Sequence counts matched. Now make sure that the root is
839 * still valid and get it if required.
840 */
841 if (unlikely(!legitimize_root(nd)))
842 goto out_dput;
843 leave_rcu(nd);
844 return true;
845
846out2:
847 nd->path.mnt = NULL;
848out1:
849 nd->path.dentry = NULL;
850out:
851 leave_rcu(nd);
852 return false;
853out_dput:
854 leave_rcu(nd);
855 dput(dentry);
856 return false;
857}
858
859static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
860{
861 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
862 return dentry->d_op->d_revalidate(dentry, flags);
863 else
864 return 1;
865}
866
867/**
868 * complete_walk - successful completion of path walk
869 * @nd: pointer nameidata
870 *
871 * If we had been in RCU mode, drop out of it and legitimize nd->path.
872 * Revalidate the final result, unless we'd already done that during
873 * the path walk or the filesystem doesn't ask for it. Return 0 on
874 * success, -error on failure. In case of failure caller does not
875 * need to drop nd->path.
876 */
877static int complete_walk(struct nameidata *nd)
878{
879 struct dentry *dentry = nd->path.dentry;
880 int status;
881
882 if (nd->flags & LOOKUP_RCU) {
883 /*
884 * We don't want to zero nd->root for scoped-lookups or
885 * externally-managed nd->root.
886 */
887 if (!(nd->state & ND_ROOT_PRESET))
888 if (!(nd->flags & LOOKUP_IS_SCOPED))
889 nd->root.mnt = NULL;
890 nd->flags &= ~LOOKUP_CACHED;
891 if (!try_to_unlazy(nd))
892 return -ECHILD;
893 }
894
895 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
896 /*
897 * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
898 * ever step outside the root during lookup" and should already
899 * be guaranteed by the rest of namei, we want to avoid a namei
900 * BUG resulting in userspace being given a path that was not
901 * scoped within the root at some point during the lookup.
902 *
903 * So, do a final sanity-check to make sure that in the
904 * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
905 * we won't silently return an fd completely outside of the
906 * requested root to userspace.
907 *
908 * Userspace could move the path outside the root after this
909 * check, but as discussed elsewhere this is not a concern (the
910 * resolved file was inside the root at some point).
911 */
912 if (!path_is_under(&nd->path, &nd->root))
913 return -EXDEV;
914 }
915
916 if (likely(!(nd->state & ND_JUMPED)))
917 return 0;
918
919 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
920 return 0;
921
922 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
923 if (status > 0)
924 return 0;
925
926 if (!status)
927 status = -ESTALE;
928
929 return status;
930}
931
932static int set_root(struct nameidata *nd)
933{
934 struct fs_struct *fs = current->fs;
935
936 /*
937 * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
938 * still have to ensure it doesn't happen because it will cause a breakout
939 * from the dirfd.
940 */
941 if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED))
942 return -ENOTRECOVERABLE;
943
944 if (nd->flags & LOOKUP_RCU) {
945 unsigned seq;
946
947 do {
948 seq = read_seqcount_begin(&fs->seq);
949 nd->root = fs->root;
950 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
951 } while (read_seqcount_retry(&fs->seq, seq));
952 } else {
953 get_fs_root(fs, root: &nd->root);
954 nd->state |= ND_ROOT_GRABBED;
955 }
956 return 0;
957}
958
959static int nd_jump_root(struct nameidata *nd)
960{
961 if (unlikely(nd->flags & LOOKUP_BENEATH))
962 return -EXDEV;
963 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
964 /* Absolute path arguments to path_init() are allowed. */
965 if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt)
966 return -EXDEV;
967 }
968 if (!nd->root.mnt) {
969 int error = set_root(nd);
970 if (error)
971 return error;
972 }
973 if (nd->flags & LOOKUP_RCU) {
974 struct dentry *d;
975 nd->path = nd->root;
976 d = nd->path.dentry;
977 nd->inode = d->d_inode;
978 nd->seq = nd->root_seq;
979 if (read_seqcount_retry(&d->d_seq, nd->seq))
980 return -ECHILD;
981 } else {
982 path_put(&nd->path);
983 nd->path = nd->root;
984 path_get(&nd->path);
985 nd->inode = nd->path.dentry->d_inode;
986 }
987 nd->state |= ND_JUMPED;
988 return 0;
989}
990
991/*
992 * Helper to directly jump to a known parsed path from ->get_link,
993 * caller must have taken a reference to path beforehand.
994 */
995int nd_jump_link(const struct path *path)
996{
997 int error = -ELOOP;
998 struct nameidata *nd = current->nameidata;
999
1000 if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS))
1001 goto err;
1002
1003 error = -EXDEV;
1004 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
1005 if (nd->path.mnt != path->mnt)
1006 goto err;
1007 }
1008 /* Not currently safe for scoped-lookups. */
1009 if (unlikely(nd->flags & LOOKUP_IS_SCOPED))
1010 goto err;
1011
1012 path_put(&nd->path);
1013 nd->path = *path;
1014 nd->inode = nd->path.dentry->d_inode;
1015 nd->state |= ND_JUMPED;
1016 return 0;
1017
1018err:
1019 path_put(path);
1020 return error;
1021}
1022
1023static inline void put_link(struct nameidata *nd)
1024{
1025 struct saved *last = nd->stack + --nd->depth;
1026 do_delayed_call(call: &last->done);
1027 if (!(nd->flags & LOOKUP_RCU))
1028 path_put(&last->link);
1029}
1030
1031static int sysctl_protected_symlinks __read_mostly;
1032static int sysctl_protected_hardlinks __read_mostly;
1033static int sysctl_protected_fifos __read_mostly;
1034static int sysctl_protected_regular __read_mostly;
1035
1036#ifdef CONFIG_SYSCTL
1037static struct ctl_table namei_sysctls[] = {
1038 {
1039 .procname = "protected_symlinks",
1040 .data = &sysctl_protected_symlinks,
1041 .maxlen = sizeof(int),
1042 .mode = 0644,
1043 .proc_handler = proc_dointvec_minmax,
1044 .extra1 = SYSCTL_ZERO,
1045 .extra2 = SYSCTL_ONE,
1046 },
1047 {
1048 .procname = "protected_hardlinks",
1049 .data = &sysctl_protected_hardlinks,
1050 .maxlen = sizeof(int),
1051 .mode = 0644,
1052 .proc_handler = proc_dointvec_minmax,
1053 .extra1 = SYSCTL_ZERO,
1054 .extra2 = SYSCTL_ONE,
1055 },
1056 {
1057 .procname = "protected_fifos",
1058 .data = &sysctl_protected_fifos,
1059 .maxlen = sizeof(int),
1060 .mode = 0644,
1061 .proc_handler = proc_dointvec_minmax,
1062 .extra1 = SYSCTL_ZERO,
1063 .extra2 = SYSCTL_TWO,
1064 },
1065 {
1066 .procname = "protected_regular",
1067 .data = &sysctl_protected_regular,
1068 .maxlen = sizeof(int),
1069 .mode = 0644,
1070 .proc_handler = proc_dointvec_minmax,
1071 .extra1 = SYSCTL_ZERO,
1072 .extra2 = SYSCTL_TWO,
1073 },
1074 { }
1075};
1076
1077static int __init init_fs_namei_sysctls(void)
1078{
1079 register_sysctl_init("fs", namei_sysctls);
1080 return 0;
1081}
1082fs_initcall(init_fs_namei_sysctls);
1083
1084#endif /* CONFIG_SYSCTL */
1085
1086/**
1087 * may_follow_link - Check symlink following for unsafe situations
1088 * @nd: nameidata pathwalk data
1089 * @inode: Used for idmapping.
1090 *
1091 * In the case of the sysctl_protected_symlinks sysctl being enabled,
1092 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
1093 * in a sticky world-writable directory. This is to protect privileged
1094 * processes from failing races against path names that may change out
1095 * from under them by way of other users creating malicious symlinks.
1096 * It will permit symlinks to be followed only when outside a sticky
1097 * world-writable directory, or when the uid of the symlink and follower
1098 * match, or when the directory owner matches the symlink's owner.
1099 *
1100 * Returns 0 if following the symlink is allowed, -ve on error.
1101 */
1102static inline int may_follow_link(struct nameidata *nd, const struct inode *inode)
1103{
1104 struct mnt_idmap *idmap;
1105 vfsuid_t vfsuid;
1106
1107 if (!sysctl_protected_symlinks)
1108 return 0;
1109
1110 idmap = mnt_idmap(mnt: nd->path.mnt);
1111 vfsuid = i_uid_into_vfsuid(idmap, inode);
1112 /* Allowed if owner and follower match. */
1113 if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
1114 return 0;
1115
1116 /* Allowed if parent directory not sticky and world-writable. */
1117 if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
1118 return 0;
1119
1120 /* Allowed if parent directory and link owner match. */
1121 if (vfsuid_valid(uid: nd->dir_vfsuid) && vfsuid_eq(left: nd->dir_vfsuid, right: vfsuid))
1122 return 0;
1123
1124 if (nd->flags & LOOKUP_RCU)
1125 return -ECHILD;
1126
1127 audit_inode(name: nd->name, dentry: nd->stack[0].link.dentry, aflags: 0);
1128 audit_log_path_denied(AUDIT_ANOM_LINK, operation: "follow_link");
1129 return -EACCES;
1130}
1131
1132/**
1133 * safe_hardlink_source - Check for safe hardlink conditions
1134 * @idmap: idmap of the mount the inode was found from
1135 * @inode: the source inode to hardlink from
1136 *
1137 * Return false if at least one of the following conditions:
1138 * - inode is not a regular file
1139 * - inode is setuid
1140 * - inode is setgid and group-exec
1141 * - access failure for read and write
1142 *
1143 * Otherwise returns true.
1144 */
1145static bool safe_hardlink_source(struct mnt_idmap *idmap,
1146 struct inode *inode)
1147{
1148 umode_t mode = inode->i_mode;
1149
1150 /* Special files should not get pinned to the filesystem. */
1151 if (!S_ISREG(mode))
1152 return false;
1153
1154 /* Setuid files should not get pinned to the filesystem. */
1155 if (mode & S_ISUID)
1156 return false;
1157
1158 /* Executable setgid files should not get pinned to the filesystem. */
1159 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
1160 return false;
1161
1162 /* Hardlinking to unreadable or unwritable sources is dangerous. */
1163 if (inode_permission(idmap, inode, MAY_READ | MAY_WRITE))
1164 return false;
1165
1166 return true;
1167}
1168
1169/**
1170 * may_linkat - Check permissions for creating a hardlink
1171 * @idmap: idmap of the mount the inode was found from
1172 * @link: the source to hardlink from
1173 *
1174 * Block hardlink when all of:
1175 * - sysctl_protected_hardlinks enabled
1176 * - fsuid does not match inode
1177 * - hardlink source is unsafe (see safe_hardlink_source() above)
1178 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1179 *
1180 * If the inode has been found through an idmapped mount the idmap of
1181 * the vfsmount must be passed through @idmap. This function will then take
1182 * care to map the inode according to @idmap before checking permissions.
1183 * On non-idmapped mounts or if permission checking is to be performed on the
1184 * raw inode simply pass @nop_mnt_idmap.
1185 *
1186 * Returns 0 if successful, -ve on error.
1187 */
1188int may_linkat(struct mnt_idmap *idmap, const struct path *link)
1189{
1190 struct inode *inode = link->dentry->d_inode;
1191
1192 /* Inode writeback is not safe when the uid or gid are invalid. */
1193 if (!vfsuid_valid(uid: i_uid_into_vfsuid(idmap, inode)) ||
1194 !vfsgid_valid(gid: i_gid_into_vfsgid(idmap, inode)))
1195 return -EOVERFLOW;
1196
1197 if (!sysctl_protected_hardlinks)
1198 return 0;
1199
1200 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1201 * otherwise, it must be a safe source.
1202 */
1203 if (safe_hardlink_source(idmap, inode) ||
1204 inode_owner_or_capable(idmap, inode))
1205 return 0;
1206
1207 audit_log_path_denied(AUDIT_ANOM_LINK, operation: "linkat");
1208 return -EPERM;
1209}
1210
1211/**
1212 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1213 * should be allowed, or not, on files that already
1214 * exist.
1215 * @idmap: idmap of the mount the inode was found from
1216 * @nd: nameidata pathwalk data
1217 * @inode: the inode of the file to open
1218 *
1219 * Block an O_CREAT open of a FIFO (or a regular file) when:
1220 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1221 * - the file already exists
1222 * - we are in a sticky directory
1223 * - we don't own the file
1224 * - the owner of the directory doesn't own the file
1225 * - the directory is world writable
1226 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1227 * the directory doesn't have to be world writable: being group writable will
1228 * be enough.
1229 *
1230 * If the inode has been found through an idmapped mount the idmap of
1231 * the vfsmount must be passed through @idmap. This function will then take
1232 * care to map the inode according to @idmap before checking permissions.
1233 * On non-idmapped mounts or if permission checking is to be performed on the
1234 * raw inode simply pass @nop_mnt_idmap.
1235 *
1236 * Returns 0 if the open is allowed, -ve on error.
1237 */
1238static int may_create_in_sticky(struct mnt_idmap *idmap,
1239 struct nameidata *nd, struct inode *const inode)
1240{
1241 umode_t dir_mode = nd->dir_mode;
1242 vfsuid_t dir_vfsuid = nd->dir_vfsuid;
1243
1244 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1245 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1246 likely(!(dir_mode & S_ISVTX)) ||
1247 vfsuid_eq(left: i_uid_into_vfsuid(idmap, inode), right: dir_vfsuid) ||
1248 vfsuid_eq_kuid(vfsuid: i_uid_into_vfsuid(idmap, inode), current_fsuid()))
1249 return 0;
1250
1251 if (likely(dir_mode & 0002) ||
1252 (dir_mode & 0020 &&
1253 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1254 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1255 const char *operation = S_ISFIFO(inode->i_mode) ?
1256 "sticky_create_fifo" :
1257 "sticky_create_regular";
1258 audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
1259 return -EACCES;
1260 }
1261 return 0;
1262}
1263
1264/*
1265 * follow_up - Find the mountpoint of path's vfsmount
1266 *
1267 * Given a path, find the mountpoint of its source file system.
1268 * Replace @path with the path of the mountpoint in the parent mount.
1269 * Up is towards /.
1270 *
1271 * Return 1 if we went up a level and 0 if we were already at the
1272 * root.
1273 */
1274int follow_up(struct path *path)
1275{
1276 struct mount *mnt = real_mount(mnt: path->mnt);
1277 struct mount *parent;
1278 struct dentry *mountpoint;
1279
1280 read_seqlock_excl(sl: &mount_lock);
1281 parent = mnt->mnt_parent;
1282 if (parent == mnt) {
1283 read_sequnlock_excl(sl: &mount_lock);
1284 return 0;
1285 }
1286 mntget(mnt: &parent->mnt);
1287 mountpoint = dget(dentry: mnt->mnt_mountpoint);
1288 read_sequnlock_excl(sl: &mount_lock);
1289 dput(path->dentry);
1290 path->dentry = mountpoint;
1291 mntput(mnt: path->mnt);
1292 path->mnt = &parent->mnt;
1293 return 1;
1294}
1295EXPORT_SYMBOL(follow_up);
1296
1297static bool choose_mountpoint_rcu(struct mount *m, const struct path *root,
1298 struct path *path, unsigned *seqp)
1299{
1300 while (mnt_has_parent(mnt: m)) {
1301 struct dentry *mountpoint = m->mnt_mountpoint;
1302
1303 m = m->mnt_parent;
1304 if (unlikely(root->dentry == mountpoint &&
1305 root->mnt == &m->mnt))
1306 break;
1307 if (mountpoint != m->mnt.mnt_root) {
1308 path->mnt = &m->mnt;
1309 path->dentry = mountpoint;
1310 *seqp = read_seqcount_begin(&mountpoint->d_seq);
1311 return true;
1312 }
1313 }
1314 return false;
1315}
1316
1317static bool choose_mountpoint(struct mount *m, const struct path *root,
1318 struct path *path)
1319{
1320 bool found;
1321
1322 rcu_read_lock();
1323 while (1) {
1324 unsigned seq, mseq = read_seqbegin(sl: &mount_lock);
1325
1326 found = choose_mountpoint_rcu(m, root, path, seqp: &seq);
1327 if (unlikely(!found)) {
1328 if (!read_seqretry(sl: &mount_lock, start: mseq))
1329 break;
1330 } else {
1331 if (likely(__legitimize_path(path, seq, mseq)))
1332 break;
1333 rcu_read_unlock();
1334 path_put(path);
1335 rcu_read_lock();
1336 }
1337 }
1338 rcu_read_unlock();
1339 return found;
1340}
1341
1342/*
1343 * Perform an automount
1344 * - return -EISDIR to tell follow_managed() to stop and return the path we
1345 * were called with.
1346 */
1347static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1348{
1349 struct dentry *dentry = path->dentry;
1350
1351 /* We don't want to mount if someone's just doing a stat -
1352 * unless they're stat'ing a directory and appended a '/' to
1353 * the name.
1354 *
1355 * We do, however, want to mount if someone wants to open or
1356 * create a file of any type under the mountpoint, wants to
1357 * traverse through the mountpoint or wants to open the
1358 * mounted directory. Also, autofs may mark negative dentries
1359 * as being automount points. These will need the attentions
1360 * of the daemon to instantiate them before they can be used.
1361 */
1362 if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1363 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1364 dentry->d_inode)
1365 return -EISDIR;
1366
1367 if (count && (*count)++ >= MAXSYMLINKS)
1368 return -ELOOP;
1369
1370 return finish_automount(dentry->d_op->d_automount(path), path);
1371}
1372
1373/*
1374 * mount traversal - out-of-line part. One note on ->d_flags accesses -
1375 * dentries are pinned but not locked here, so negative dentry can go
1376 * positive right under us. Use of smp_load_acquire() provides a barrier
1377 * sufficient for ->d_inode and ->d_flags consistency.
1378 */
1379static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped,
1380 int *count, unsigned lookup_flags)
1381{
1382 struct vfsmount *mnt = path->mnt;
1383 bool need_mntput = false;
1384 int ret = 0;
1385
1386 while (flags & DCACHE_MANAGED_DENTRY) {
1387 /* Allow the filesystem to manage the transit without i_mutex
1388 * being held. */
1389 if (flags & DCACHE_MANAGE_TRANSIT) {
1390 ret = path->dentry->d_op->d_manage(path, false);
1391 flags = smp_load_acquire(&path->dentry->d_flags);
1392 if (ret < 0)
1393 break;
1394 }
1395
1396 if (flags & DCACHE_MOUNTED) { // something's mounted on it..
1397 struct vfsmount *mounted = lookup_mnt(path);
1398 if (mounted) { // ... in our namespace
1399 dput(path->dentry);
1400 if (need_mntput)
1401 mntput(mnt: path->mnt);
1402 path->mnt = mounted;
1403 path->dentry = dget(dentry: mounted->mnt_root);
1404 // here we know it's positive
1405 flags = path->dentry->d_flags;
1406 need_mntput = true;
1407 continue;
1408 }
1409 }
1410
1411 if (!(flags & DCACHE_NEED_AUTOMOUNT))
1412 break;
1413
1414 // uncovered automount point
1415 ret = follow_automount(path, count, lookup_flags);
1416 flags = smp_load_acquire(&path->dentry->d_flags);
1417 if (ret < 0)
1418 break;
1419 }
1420
1421 if (ret == -EISDIR)
1422 ret = 0;
1423 // possible if you race with several mount --move
1424 if (need_mntput && path->mnt == mnt)
1425 mntput(mnt: path->mnt);
1426 if (!ret && unlikely(d_flags_negative(flags)))
1427 ret = -ENOENT;
1428 *jumped = need_mntput;
1429 return ret;
1430}
1431
1432static inline int traverse_mounts(struct path *path, bool *jumped,
1433 int *count, unsigned lookup_flags)
1434{
1435 unsigned flags = smp_load_acquire(&path->dentry->d_flags);
1436
1437 /* fastpath */
1438 if (likely(!(flags & DCACHE_MANAGED_DENTRY))) {
1439 *jumped = false;
1440 if (unlikely(d_flags_negative(flags)))
1441 return -ENOENT;
1442 return 0;
1443 }
1444 return __traverse_mounts(path, flags, jumped, count, lookup_flags);
1445}
1446
1447int follow_down_one(struct path *path)
1448{
1449 struct vfsmount *mounted;
1450
1451 mounted = lookup_mnt(path);
1452 if (mounted) {
1453 dput(path->dentry);
1454 mntput(mnt: path->mnt);
1455 path->mnt = mounted;
1456 path->dentry = dget(dentry: mounted->mnt_root);
1457 return 1;
1458 }
1459 return 0;
1460}
1461EXPORT_SYMBOL(follow_down_one);
1462
1463/*
1464 * Follow down to the covering mount currently visible to userspace. At each
1465 * point, the filesystem owning that dentry may be queried as to whether the
1466 * caller is permitted to proceed or not.
1467 */
1468int follow_down(struct path *path, unsigned int flags)
1469{
1470 struct vfsmount *mnt = path->mnt;
1471 bool jumped;
1472 int ret = traverse_mounts(path, jumped: &jumped, NULL, lookup_flags: flags);
1473
1474 if (path->mnt != mnt)
1475 mntput(mnt);
1476 return ret;
1477}
1478EXPORT_SYMBOL(follow_down);
1479
1480/*
1481 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1482 * we meet a managed dentry that would need blocking.
1483 */
1484static bool __follow_mount_rcu(struct nameidata *nd, struct path *path)
1485{
1486 struct dentry *dentry = path->dentry;
1487 unsigned int flags = dentry->d_flags;
1488
1489 if (likely(!(flags & DCACHE_MANAGED_DENTRY)))
1490 return true;
1491
1492 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1493 return false;
1494
1495 for (;;) {
1496 /*
1497 * Don't forget we might have a non-mountpoint managed dentry
1498 * that wants to block transit.
1499 */
1500 if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) {
1501 int res = dentry->d_op->d_manage(path, true);
1502 if (res)
1503 return res == -EISDIR;
1504 flags = dentry->d_flags;
1505 }
1506
1507 if (flags & DCACHE_MOUNTED) {
1508 struct mount *mounted = __lookup_mnt(path->mnt, dentry);
1509 if (mounted) {
1510 path->mnt = &mounted->mnt;
1511 dentry = path->dentry = mounted->mnt.mnt_root;
1512 nd->state |= ND_JUMPED;
1513 nd->next_seq = read_seqcount_begin(&dentry->d_seq);
1514 flags = dentry->d_flags;
1515 // makes sure that non-RCU pathwalk could reach
1516 // this state.
1517 if (read_seqretry(sl: &mount_lock, start: nd->m_seq))
1518 return false;
1519 continue;
1520 }
1521 if (read_seqretry(sl: &mount_lock, start: nd->m_seq))
1522 return false;
1523 }
1524 return !(flags & DCACHE_NEED_AUTOMOUNT);
1525 }
1526}
1527
1528static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1529 struct path *path)
1530{
1531 bool jumped;
1532 int ret;
1533
1534 path->mnt = nd->path.mnt;
1535 path->dentry = dentry;
1536 if (nd->flags & LOOKUP_RCU) {
1537 unsigned int seq = nd->next_seq;
1538 if (likely(__follow_mount_rcu(nd, path)))
1539 return 0;
1540 // *path and nd->next_seq might've been clobbered
1541 path->mnt = nd->path.mnt;
1542 path->dentry = dentry;
1543 nd->next_seq = seq;
1544 if (!try_to_unlazy_next(nd, dentry))
1545 return -ECHILD;
1546 }
1547 ret = traverse_mounts(path, jumped: &jumped, count: &nd->total_link_count, lookup_flags: nd->flags);
1548 if (jumped) {
1549 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1550 ret = -EXDEV;
1551 else
1552 nd->state |= ND_JUMPED;
1553 }
1554 if (unlikely(ret)) {
1555 dput(path->dentry);
1556 if (path->mnt != nd->path.mnt)
1557 mntput(mnt: path->mnt);
1558 }
1559 return ret;
1560}
1561
1562/*
1563 * This looks up the name in dcache and possibly revalidates the found dentry.
1564 * NULL is returned if the dentry does not exist in the cache.
1565 */
1566static struct dentry *lookup_dcache(const struct qstr *name,
1567 struct dentry *dir,
1568 unsigned int flags)
1569{
1570 struct dentry *dentry = d_lookup(dir, name);
1571 if (dentry) {
1572 int error = d_revalidate(dentry, flags);
1573 if (unlikely(error <= 0)) {
1574 if (!error)
1575 d_invalidate(dentry);
1576 dput(dentry);
1577 return ERR_PTR(error);
1578 }
1579 }
1580 return dentry;
1581}
1582
1583/*
1584 * Parent directory has inode locked exclusive. This is one
1585 * and only case when ->lookup() gets called on non in-lookup
1586 * dentries - as the matter of fact, this only gets called
1587 * when directory is guaranteed to have no in-lookup children
1588 * at all.
1589 */
1590struct dentry *lookup_one_qstr_excl(const struct qstr *name,
1591 struct dentry *base,
1592 unsigned int flags)
1593{
1594 struct dentry *dentry = lookup_dcache(name, dir: base, flags);
1595 struct dentry *old;
1596 struct inode *dir = base->d_inode;
1597
1598 if (dentry)
1599 return dentry;
1600
1601 /* Don't create child dentry for a dead directory. */
1602 if (unlikely(IS_DEADDIR(dir)))
1603 return ERR_PTR(error: -ENOENT);
1604
1605 dentry = d_alloc(base, name);
1606 if (unlikely(!dentry))
1607 return ERR_PTR(error: -ENOMEM);
1608
1609 old = dir->i_op->lookup(dir, dentry, flags);
1610 if (unlikely(old)) {
1611 dput(dentry);
1612 dentry = old;
1613 }
1614 return dentry;
1615}
1616EXPORT_SYMBOL(lookup_one_qstr_excl);
1617
1618static struct dentry *lookup_fast(struct nameidata *nd)
1619{
1620 struct dentry *dentry, *parent = nd->path.dentry;
1621 int status = 1;
1622
1623 /*
1624 * Rename seqlock is not required here because in the off chance
1625 * of a false negative due to a concurrent rename, the caller is
1626 * going to fall back to non-racy lookup.
1627 */
1628 if (nd->flags & LOOKUP_RCU) {
1629 dentry = __d_lookup_rcu(parent, name: &nd->last, seq: &nd->next_seq);
1630 if (unlikely(!dentry)) {
1631 if (!try_to_unlazy(nd))
1632 return ERR_PTR(error: -ECHILD);
1633 return NULL;
1634 }
1635
1636 /*
1637 * This sequence count validates that the parent had no
1638 * changes while we did the lookup of the dentry above.
1639 */
1640 if (read_seqcount_retry(&parent->d_seq, nd->seq))
1641 return ERR_PTR(error: -ECHILD);
1642
1643 status = d_revalidate(dentry, flags: nd->flags);
1644 if (likely(status > 0))
1645 return dentry;
1646 if (!try_to_unlazy_next(nd, dentry))
1647 return ERR_PTR(error: -ECHILD);
1648 if (status == -ECHILD)
1649 /* we'd been told to redo it in non-rcu mode */
1650 status = d_revalidate(dentry, flags: nd->flags);
1651 } else {
1652 dentry = __d_lookup(parent, &nd->last);
1653 if (unlikely(!dentry))
1654 return NULL;
1655 status = d_revalidate(dentry, flags: nd->flags);
1656 }
1657 if (unlikely(status <= 0)) {
1658 if (!status)
1659 d_invalidate(dentry);
1660 dput(dentry);
1661 return ERR_PTR(error: status);
1662 }
1663 return dentry;
1664}
1665
1666/* Fast lookup failed, do it the slow way */
1667static struct dentry *__lookup_slow(const struct qstr *name,
1668 struct dentry *dir,
1669 unsigned int flags)
1670{
1671 struct dentry *dentry, *old;
1672 struct inode *inode = dir->d_inode;
1673 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1674
1675 /* Don't go there if it's already dead */
1676 if (unlikely(IS_DEADDIR(inode)))
1677 return ERR_PTR(error: -ENOENT);
1678again:
1679 dentry = d_alloc_parallel(dir, name, &wq);
1680 if (IS_ERR(ptr: dentry))
1681 return dentry;
1682 if (unlikely(!d_in_lookup(dentry))) {
1683 int error = d_revalidate(dentry, flags);
1684 if (unlikely(error <= 0)) {
1685 if (!error) {
1686 d_invalidate(dentry);
1687 dput(dentry);
1688 goto again;
1689 }
1690 dput(dentry);
1691 dentry = ERR_PTR(error);
1692 }
1693 } else {
1694 old = inode->i_op->lookup(inode, dentry, flags);
1695 d_lookup_done(dentry);
1696 if (unlikely(old)) {
1697 dput(dentry);
1698 dentry = old;
1699 }
1700 }
1701 return dentry;
1702}
1703
1704static struct dentry *lookup_slow(const struct qstr *name,
1705 struct dentry *dir,
1706 unsigned int flags)
1707{
1708 struct inode *inode = dir->d_inode;
1709 struct dentry *res;
1710 inode_lock_shared(inode);
1711 res = __lookup_slow(name, dir, flags);
1712 inode_unlock_shared(inode);
1713 return res;
1714}
1715
1716static inline int may_lookup(struct mnt_idmap *idmap,
1717 struct nameidata *nd)
1718{
1719 if (nd->flags & LOOKUP_RCU) {
1720 int err = inode_permission(idmap, nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1721 if (err != -ECHILD || !try_to_unlazy(nd))
1722 return err;
1723 }
1724 return inode_permission(idmap, nd->inode, MAY_EXEC);
1725}
1726
1727static int reserve_stack(struct nameidata *nd, struct path *link)
1728{
1729 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1730 return -ELOOP;
1731
1732 if (likely(nd->depth != EMBEDDED_LEVELS))
1733 return 0;
1734 if (likely(nd->stack != nd->internal))
1735 return 0;
1736 if (likely(nd_alloc_stack(nd)))
1737 return 0;
1738
1739 if (nd->flags & LOOKUP_RCU) {
1740 // we need to grab link before we do unlazy. And we can't skip
1741 // unlazy even if we fail to grab the link - cleanup needs it
1742 bool grabbed_link = legitimize_path(nd, path: link, seq: nd->next_seq);
1743
1744 if (!try_to_unlazy(nd) || !grabbed_link)
1745 return -ECHILD;
1746
1747 if (nd_alloc_stack(nd))
1748 return 0;
1749 }
1750 return -ENOMEM;
1751}
1752
1753enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1754
1755static const char *pick_link(struct nameidata *nd, struct path *link,
1756 struct inode *inode, int flags)
1757{
1758 struct saved *last;
1759 const char *res;
1760 int error = reserve_stack(nd, link);
1761
1762 if (unlikely(error)) {
1763 if (!(nd->flags & LOOKUP_RCU))
1764 path_put(link);
1765 return ERR_PTR(error);
1766 }
1767 last = nd->stack + nd->depth++;
1768 last->link = *link;
1769 clear_delayed_call(call: &last->done);
1770 last->seq = nd->next_seq;
1771
1772 if (flags & WALK_TRAILING) {
1773 error = may_follow_link(nd, inode);
1774 if (unlikely(error))
1775 return ERR_PTR(error);
1776 }
1777
1778 if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) ||
1779 unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW))
1780 return ERR_PTR(error: -ELOOP);
1781
1782 if (!(nd->flags & LOOKUP_RCU)) {
1783 touch_atime(&last->link);
1784 cond_resched();
1785 } else if (atime_needs_update(&last->link, inode)) {
1786 if (!try_to_unlazy(nd))
1787 return ERR_PTR(error: -ECHILD);
1788 touch_atime(&last->link);
1789 }
1790
1791 error = security_inode_follow_link(dentry: link->dentry, inode,
1792 rcu: nd->flags & LOOKUP_RCU);
1793 if (unlikely(error))
1794 return ERR_PTR(error);
1795
1796 res = READ_ONCE(inode->i_link);
1797 if (!res) {
1798 const char * (*get)(struct dentry *, struct inode *,
1799 struct delayed_call *);
1800 get = inode->i_op->get_link;
1801 if (nd->flags & LOOKUP_RCU) {
1802 res = get(NULL, inode, &last->done);
1803 if (res == ERR_PTR(error: -ECHILD) && try_to_unlazy(nd))
1804 res = get(link->dentry, inode, &last->done);
1805 } else {
1806 res = get(link->dentry, inode, &last->done);
1807 }
1808 if (!res)
1809 goto all_done;
1810 if (IS_ERR(ptr: res))
1811 return res;
1812 }
1813 if (*res == '/') {
1814 error = nd_jump_root(nd);
1815 if (unlikely(error))
1816 return ERR_PTR(error);
1817 while (unlikely(*++res == '/'))
1818 ;
1819 }
1820 if (*res)
1821 return res;
1822all_done: // pure jump
1823 put_link(nd);
1824 return NULL;
1825}
1826
1827/*
1828 * Do we need to follow links? We _really_ want to be able
1829 * to do this check without having to look at inode->i_op,
1830 * so we keep a cache of "no, this doesn't need follow_link"
1831 * for the common case.
1832 *
1833 * NOTE: dentry must be what nd->next_seq had been sampled from.
1834 */
1835static const char *step_into(struct nameidata *nd, int flags,
1836 struct dentry *dentry)
1837{
1838 struct path path;
1839 struct inode *inode;
1840 int err = handle_mounts(nd, dentry, path: &path);
1841
1842 if (err < 0)
1843 return ERR_PTR(error: err);
1844 inode = path.dentry->d_inode;
1845 if (likely(!d_is_symlink(path.dentry)) ||
1846 ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
1847 (flags & WALK_NOFOLLOW)) {
1848 /* not a symlink or should not follow */
1849 if (nd->flags & LOOKUP_RCU) {
1850 if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1851 return ERR_PTR(error: -ECHILD);
1852 if (unlikely(!inode))
1853 return ERR_PTR(error: -ENOENT);
1854 } else {
1855 dput(nd->path.dentry);
1856 if (nd->path.mnt != path.mnt)
1857 mntput(mnt: nd->path.mnt);
1858 }
1859 nd->path = path;
1860 nd->inode = inode;
1861 nd->seq = nd->next_seq;
1862 return NULL;
1863 }
1864 if (nd->flags & LOOKUP_RCU) {
1865 /* make sure that d_is_symlink above matches inode */
1866 if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1867 return ERR_PTR(error: -ECHILD);
1868 } else {
1869 if (path.mnt == nd->path.mnt)
1870 mntget(mnt: path.mnt);
1871 }
1872 return pick_link(nd, link: &path, inode, flags);
1873}
1874
1875static struct dentry *follow_dotdot_rcu(struct nameidata *nd)
1876{
1877 struct dentry *parent, *old;
1878
1879 if (path_equal(path1: &nd->path, path2: &nd->root))
1880 goto in_root;
1881 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1882 struct path path;
1883 unsigned seq;
1884 if (!choose_mountpoint_rcu(m: real_mount(mnt: nd->path.mnt),
1885 root: &nd->root, path: &path, seqp: &seq))
1886 goto in_root;
1887 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1888 return ERR_PTR(error: -ECHILD);
1889 nd->path = path;
1890 nd->inode = path.dentry->d_inode;
1891 nd->seq = seq;
1892 // makes sure that non-RCU pathwalk could reach this state
1893 if (read_seqretry(sl: &mount_lock, start: nd->m_seq))
1894 return ERR_PTR(error: -ECHILD);
1895 /* we know that mountpoint was pinned */
1896 }
1897 old = nd->path.dentry;
1898 parent = old->d_parent;
1899 nd->next_seq = read_seqcount_begin(&parent->d_seq);
1900 // makes sure that non-RCU pathwalk could reach this state
1901 if (read_seqcount_retry(&old->d_seq, nd->seq))
1902 return ERR_PTR(error: -ECHILD);
1903 if (unlikely(!path_connected(nd->path.mnt, parent)))
1904 return ERR_PTR(error: -ECHILD);
1905 return parent;
1906in_root:
1907 if (read_seqretry(sl: &mount_lock, start: nd->m_seq))
1908 return ERR_PTR(error: -ECHILD);
1909 if (unlikely(nd->flags & LOOKUP_BENEATH))
1910 return ERR_PTR(error: -ECHILD);
1911 nd->next_seq = nd->seq;
1912 return nd->path.dentry;
1913}
1914
1915static struct dentry *follow_dotdot(struct nameidata *nd)
1916{
1917 struct dentry *parent;
1918
1919 if (path_equal(path1: &nd->path, path2: &nd->root))
1920 goto in_root;
1921 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1922 struct path path;
1923
1924 if (!choose_mountpoint(m: real_mount(mnt: nd->path.mnt),
1925 root: &nd->root, path: &path))
1926 goto in_root;
1927 path_put(&nd->path);
1928 nd->path = path;
1929 nd->inode = path.dentry->d_inode;
1930 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1931 return ERR_PTR(error: -EXDEV);
1932 }
1933 /* rare case of legitimate dget_parent()... */
1934 parent = dget_parent(dentry: nd->path.dentry);
1935 if (unlikely(!path_connected(nd->path.mnt, parent))) {
1936 dput(parent);
1937 return ERR_PTR(error: -ENOENT);
1938 }
1939 return parent;
1940
1941in_root:
1942 if (unlikely(nd->flags & LOOKUP_BENEATH))
1943 return ERR_PTR(error: -EXDEV);
1944 return dget(dentry: nd->path.dentry);
1945}
1946
1947static const char *handle_dots(struct nameidata *nd, int type)
1948{
1949 if (type == LAST_DOTDOT) {
1950 const char *error = NULL;
1951 struct dentry *parent;
1952
1953 if (!nd->root.mnt) {
1954 error = ERR_PTR(error: set_root(nd));
1955 if (error)
1956 return error;
1957 }
1958 if (nd->flags & LOOKUP_RCU)
1959 parent = follow_dotdot_rcu(nd);
1960 else
1961 parent = follow_dotdot(nd);
1962 if (IS_ERR(ptr: parent))
1963 return ERR_CAST(ptr: parent);
1964 error = step_into(nd, flags: WALK_NOFOLLOW, dentry: parent);
1965 if (unlikely(error))
1966 return error;
1967
1968 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1969 /*
1970 * If there was a racing rename or mount along our
1971 * path, then we can't be sure that ".." hasn't jumped
1972 * above nd->root (and so userspace should retry or use
1973 * some fallback).
1974 */
1975 smp_rmb();
1976 if (__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))
1977 return ERR_PTR(error: -EAGAIN);
1978 if (__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))
1979 return ERR_PTR(error: -EAGAIN);
1980 }
1981 }
1982 return NULL;
1983}
1984
1985static const char *walk_component(struct nameidata *nd, int flags)
1986{
1987 struct dentry *dentry;
1988 /*
1989 * "." and ".." are special - ".." especially so because it has
1990 * to be able to know about the current root directory and
1991 * parent relationships.
1992 */
1993 if (unlikely(nd->last_type != LAST_NORM)) {
1994 if (!(flags & WALK_MORE) && nd->depth)
1995 put_link(nd);
1996 return handle_dots(nd, type: nd->last_type);
1997 }
1998 dentry = lookup_fast(nd);
1999 if (IS_ERR(ptr: dentry))
2000 return ERR_CAST(ptr: dentry);
2001 if (unlikely(!dentry)) {
2002 dentry = lookup_slow(name: &nd->last, dir: nd->path.dentry, flags: nd->flags);
2003 if (IS_ERR(ptr: dentry))
2004 return ERR_CAST(ptr: dentry);
2005 }
2006 if (!(flags & WALK_MORE) && nd->depth)
2007 put_link(nd);
2008 return step_into(nd, flags, dentry);
2009}
2010
2011/*
2012 * We can do the critical dentry name comparison and hashing
2013 * operations one word at a time, but we are limited to:
2014 *
2015 * - Architectures with fast unaligned word accesses. We could
2016 * do a "get_unaligned()" if this helps and is sufficiently
2017 * fast.
2018 *
2019 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
2020 * do not trap on the (extremely unlikely) case of a page
2021 * crossing operation.
2022 *
2023 * - Furthermore, we need an efficient 64-bit compile for the
2024 * 64-bit case in order to generate the "number of bytes in
2025 * the final mask". Again, that could be replaced with a
2026 * efficient population count instruction or similar.
2027 */
2028#ifdef CONFIG_DCACHE_WORD_ACCESS
2029
2030#include <asm/word-at-a-time.h>
2031
2032#ifdef HASH_MIX
2033
2034/* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
2035
2036#elif defined(CONFIG_64BIT)
2037/*
2038 * Register pressure in the mixing function is an issue, particularly
2039 * on 32-bit x86, but almost any function requires one state value and
2040 * one temporary. Instead, use a function designed for two state values
2041 * and no temporaries.
2042 *
2043 * This function cannot create a collision in only two iterations, so
2044 * we have two iterations to achieve avalanche. In those two iterations,
2045 * we have six layers of mixing, which is enough to spread one bit's
2046 * influence out to 2^6 = 64 state bits.
2047 *
2048 * Rotate constants are scored by considering either 64 one-bit input
2049 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
2050 * probability of that delta causing a change to each of the 128 output
2051 * bits, using a sample of random initial states.
2052 *
2053 * The Shannon entropy of the computed probabilities is then summed
2054 * to produce a score. Ideally, any input change has a 50% chance of
2055 * toggling any given output bit.
2056 *
2057 * Mixing scores (in bits) for (12,45):
2058 * Input delta: 1-bit 2-bit
2059 * 1 round: 713.3 42542.6
2060 * 2 rounds: 2753.7 140389.8
2061 * 3 rounds: 5954.1 233458.2
2062 * 4 rounds: 7862.6 256672.2
2063 * Perfect: 8192 258048
2064 * (64*128) (64*63/2 * 128)
2065 */
2066#define HASH_MIX(x, y, a) \
2067 ( x ^= (a), \
2068 y ^= x, x = rol64(x,12),\
2069 x += y, y = rol64(y,45),\
2070 y *= 9 )
2071
2072/*
2073 * Fold two longs into one 32-bit hash value. This must be fast, but
2074 * latency isn't quite as critical, as there is a fair bit of additional
2075 * work done before the hash value is used.
2076 */
2077static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2078{
2079 y ^= x * GOLDEN_RATIO_64;
2080 y *= GOLDEN_RATIO_64;
2081 return y >> 32;
2082}
2083
2084#else /* 32-bit case */
2085
2086/*
2087 * Mixing scores (in bits) for (7,20):
2088 * Input delta: 1-bit 2-bit
2089 * 1 round: 330.3 9201.6
2090 * 2 rounds: 1246.4 25475.4
2091 * 3 rounds: 1907.1 31295.1
2092 * 4 rounds: 2042.3 31718.6
2093 * Perfect: 2048 31744
2094 * (32*64) (32*31/2 * 64)
2095 */
2096#define HASH_MIX(x, y, a) \
2097 ( x ^= (a), \
2098 y ^= x, x = rol32(x, 7),\
2099 x += y, y = rol32(y,20),\
2100 y *= 9 )
2101
2102static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2103{
2104 /* Use arch-optimized multiply if one exists */
2105 return __hash_32(y ^ __hash_32(x));
2106}
2107
2108#endif
2109
2110/*
2111 * Return the hash of a string of known length. This is carfully
2112 * designed to match hash_name(), which is the more critical function.
2113 * In particular, we must end by hashing a final word containing 0..7
2114 * payload bytes, to match the way that hash_name() iterates until it
2115 * finds the delimiter after the name.
2116 */
2117unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2118{
2119 unsigned long a, x = 0, y = (unsigned long)salt;
2120
2121 for (;;) {
2122 if (!len)
2123 goto done;
2124 a = load_unaligned_zeropad(addr: name);
2125 if (len < sizeof(unsigned long))
2126 break;
2127 HASH_MIX(x, y, a);
2128 name += sizeof(unsigned long);
2129 len -= sizeof(unsigned long);
2130 }
2131 x ^= a & bytemask_from_count(len);
2132done:
2133 return fold_hash(x, y);
2134}
2135EXPORT_SYMBOL(full_name_hash);
2136
2137/* Return the "hash_len" (hash and length) of a null-terminated string */
2138u64 hashlen_string(const void *salt, const char *name)
2139{
2140 unsigned long a = 0, x = 0, y = (unsigned long)salt;
2141 unsigned long adata, mask, len;
2142 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2143
2144 len = 0;
2145 goto inside;
2146
2147 do {
2148 HASH_MIX(x, y, a);
2149 len += sizeof(unsigned long);
2150inside:
2151 a = load_unaligned_zeropad(addr: name+len);
2152 } while (!has_zero(a, bits: &adata, c: &constants));
2153
2154 adata = prep_zero_mask(a, bits: adata, c: &constants);
2155 mask = create_zero_mask(bits: adata);
2156 x ^= a & zero_bytemask(mask);
2157
2158 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2159}
2160EXPORT_SYMBOL(hashlen_string);
2161
2162/*
2163 * Calculate the length and hash of the path component, and
2164 * return the "hash_len" as the result.
2165 */
2166static inline u64 hash_name(const void *salt, const char *name)
2167{
2168 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2169 unsigned long adata, bdata, mask, len;
2170 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2171
2172 len = 0;
2173 goto inside;
2174
2175 do {
2176 HASH_MIX(x, y, a);
2177 len += sizeof(unsigned long);
2178inside:
2179 a = load_unaligned_zeropad(addr: name+len);
2180 b = a ^ REPEAT_BYTE('/');
2181 } while (!(has_zero(a, bits: &adata, c: &constants) | has_zero(a: b, bits: &bdata, c: &constants)));
2182
2183 adata = prep_zero_mask(a, bits: adata, c: &constants);
2184 bdata = prep_zero_mask(a: b, bits: bdata, c: &constants);
2185 mask = create_zero_mask(bits: adata | bdata);
2186 x ^= a & zero_bytemask(mask);
2187
2188 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2189}
2190
2191#else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2192
2193/* Return the hash of a string of known length */
2194unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2195{
2196 unsigned long hash = init_name_hash(salt);
2197 while (len--)
2198 hash = partial_name_hash((unsigned char)*name++, hash);
2199 return end_name_hash(hash);
2200}
2201EXPORT_SYMBOL(full_name_hash);
2202
2203/* Return the "hash_len" (hash and length) of a null-terminated string */
2204u64 hashlen_string(const void *salt, const char *name)
2205{
2206 unsigned long hash = init_name_hash(salt);
2207 unsigned long len = 0, c;
2208
2209 c = (unsigned char)*name;
2210 while (c) {
2211 len++;
2212 hash = partial_name_hash(c, hash);
2213 c = (unsigned char)name[len];
2214 }
2215 return hashlen_create(end_name_hash(hash), len);
2216}
2217EXPORT_SYMBOL(hashlen_string);
2218
2219/*
2220 * We know there's a real path component here of at least
2221 * one character.
2222 */
2223static inline u64 hash_name(const void *salt, const char *name)
2224{
2225 unsigned long hash = init_name_hash(salt);
2226 unsigned long len = 0, c;
2227
2228 c = (unsigned char)*name;
2229 do {
2230 len++;
2231 hash = partial_name_hash(c, hash);
2232 c = (unsigned char)name[len];
2233 } while (c && c != '/');
2234 return hashlen_create(end_name_hash(hash), len);
2235}
2236
2237#endif
2238
2239/*
2240 * Name resolution.
2241 * This is the basic name resolution function, turning a pathname into
2242 * the final dentry. We expect 'base' to be positive and a directory.
2243 *
2244 * Returns 0 and nd will have valid dentry and mnt on success.
2245 * Returns error and drops reference to input namei data on failure.
2246 */
2247static int link_path_walk(const char *name, struct nameidata *nd)
2248{
2249 int depth = 0; // depth <= nd->depth
2250 int err;
2251
2252 nd->last_type = LAST_ROOT;
2253 nd->flags |= LOOKUP_PARENT;
2254 if (IS_ERR(ptr: name))
2255 return PTR_ERR(ptr: name);
2256 while (*name=='/')
2257 name++;
2258 if (!*name) {
2259 nd->dir_mode = 0; // short-circuit the 'hardening' idiocy
2260 return 0;
2261 }
2262
2263 /* At this point we know we have a real path component. */
2264 for(;;) {
2265 struct mnt_idmap *idmap;
2266 const char *link;
2267 u64 hash_len;
2268 int type;
2269
2270 idmap = mnt_idmap(mnt: nd->path.mnt);
2271 err = may_lookup(idmap, nd);
2272 if (err)
2273 return err;
2274
2275 hash_len = hash_name(salt: nd->path.dentry, name);
2276
2277 type = LAST_NORM;
2278 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2279 case 2:
2280 if (name[1] == '.') {
2281 type = LAST_DOTDOT;
2282 nd->state |= ND_JUMPED;
2283 }
2284 break;
2285 case 1:
2286 type = LAST_DOT;
2287 }
2288 if (likely(type == LAST_NORM)) {
2289 struct dentry *parent = nd->path.dentry;
2290 nd->state &= ~ND_JUMPED;
2291 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2292 struct qstr this = { { .hash_len = hash_len }, .name = name };
2293 err = parent->d_op->d_hash(parent, &this);
2294 if (err < 0)
2295 return err;
2296 hash_len = this.hash_len;
2297 name = this.name;
2298 }
2299 }
2300
2301 nd->last.hash_len = hash_len;
2302 nd->last.name = name;
2303 nd->last_type = type;
2304
2305 name += hashlen_len(hash_len);
2306 if (!*name)
2307 goto OK;
2308 /*
2309 * If it wasn't NUL, we know it was '/'. Skip that
2310 * slash, and continue until no more slashes.
2311 */
2312 do {
2313 name++;
2314 } while (unlikely(*name == '/'));
2315 if (unlikely(!*name)) {
2316OK:
2317 /* pathname or trailing symlink, done */
2318 if (!depth) {
2319 nd->dir_vfsuid = i_uid_into_vfsuid(idmap, inode: nd->inode);
2320 nd->dir_mode = nd->inode->i_mode;
2321 nd->flags &= ~LOOKUP_PARENT;
2322 return 0;
2323 }
2324 /* last component of nested symlink */
2325 name = nd->stack[--depth].name;
2326 link = walk_component(nd, flags: 0);
2327 } else {
2328 /* not the last component */
2329 link = walk_component(nd, flags: WALK_MORE);
2330 }
2331 if (unlikely(link)) {
2332 if (IS_ERR(ptr: link))
2333 return PTR_ERR(ptr: link);
2334 /* a symlink to follow */
2335 nd->stack[depth++].name = name;
2336 name = link;
2337 continue;
2338 }
2339 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2340 if (nd->flags & LOOKUP_RCU) {
2341 if (!try_to_unlazy(nd))
2342 return -ECHILD;
2343 }
2344 return -ENOTDIR;
2345 }
2346 }
2347}
2348
2349/* must be paired with terminate_walk() */
2350static const char *path_init(struct nameidata *nd, unsigned flags)
2351{
2352 int error;
2353 const char *s = nd->name->name;
2354
2355 /* LOOKUP_CACHED requires RCU, ask caller to retry */
2356 if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED)
2357 return ERR_PTR(error: -EAGAIN);
2358
2359 if (!*s)
2360 flags &= ~LOOKUP_RCU;
2361 if (flags & LOOKUP_RCU)
2362 rcu_read_lock();
2363 else
2364 nd->seq = nd->next_seq = 0;
2365
2366 nd->flags = flags;
2367 nd->state |= ND_JUMPED;
2368
2369 nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2370 nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2371 smp_rmb();
2372
2373 if (nd->state & ND_ROOT_PRESET) {
2374 struct dentry *root = nd->root.dentry;
2375 struct inode *inode = root->d_inode;
2376 if (*s && unlikely(!d_can_lookup(root)))
2377 return ERR_PTR(error: -ENOTDIR);
2378 nd->path = nd->root;
2379 nd->inode = inode;
2380 if (flags & LOOKUP_RCU) {
2381 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2382 nd->root_seq = nd->seq;
2383 } else {
2384 path_get(&nd->path);
2385 }
2386 return s;
2387 }
2388
2389 nd->root.mnt = NULL;
2390
2391 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2392 if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2393 error = nd_jump_root(nd);
2394 if (unlikely(error))
2395 return ERR_PTR(error);
2396 return s;
2397 }
2398
2399 /* Relative pathname -- get the starting-point it is relative to. */
2400 if (nd->dfd == AT_FDCWD) {
2401 if (flags & LOOKUP_RCU) {
2402 struct fs_struct *fs = current->fs;
2403 unsigned seq;
2404
2405 do {
2406 seq = read_seqcount_begin(&fs->seq);
2407 nd->path = fs->pwd;
2408 nd->inode = nd->path.dentry->d_inode;
2409 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2410 } while (read_seqcount_retry(&fs->seq, seq));
2411 } else {
2412 get_fs_pwd(current->fs, pwd: &nd->path);
2413 nd->inode = nd->path.dentry->d_inode;
2414 }
2415 } else {
2416 /* Caller must check execute permissions on the starting path component */
2417 struct fd f = fdget_raw(fd: nd->dfd);
2418 struct dentry *dentry;
2419
2420 if (!f.file)
2421 return ERR_PTR(error: -EBADF);
2422
2423 dentry = f.file->f_path.dentry;
2424
2425 if (*s && unlikely(!d_can_lookup(dentry))) {
2426 fdput(fd: f);
2427 return ERR_PTR(error: -ENOTDIR);
2428 }
2429
2430 nd->path = f.file->f_path;
2431 if (flags & LOOKUP_RCU) {
2432 nd->inode = nd->path.dentry->d_inode;
2433 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2434 } else {
2435 path_get(&nd->path);
2436 nd->inode = nd->path.dentry->d_inode;
2437 }
2438 fdput(fd: f);
2439 }
2440
2441 /* For scoped-lookups we need to set the root to the dirfd as well. */
2442 if (flags & LOOKUP_IS_SCOPED) {
2443 nd->root = nd->path;
2444 if (flags & LOOKUP_RCU) {
2445 nd->root_seq = nd->seq;
2446 } else {
2447 path_get(&nd->root);
2448 nd->state |= ND_ROOT_GRABBED;
2449 }
2450 }
2451 return s;
2452}
2453
2454static inline const char *lookup_last(struct nameidata *nd)
2455{
2456 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2457 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2458
2459 return walk_component(nd, flags: WALK_TRAILING);
2460}
2461
2462static int handle_lookup_down(struct nameidata *nd)
2463{
2464 if (!(nd->flags & LOOKUP_RCU))
2465 dget(dentry: nd->path.dentry);
2466 nd->next_seq = nd->seq;
2467 return PTR_ERR(ptr: step_into(nd, flags: WALK_NOFOLLOW, dentry: nd->path.dentry));
2468}
2469
2470/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2471static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2472{
2473 const char *s = path_init(nd, flags);
2474 int err;
2475
2476 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(ptr: s)) {
2477 err = handle_lookup_down(nd);
2478 if (unlikely(err < 0))
2479 s = ERR_PTR(error: err);
2480 }
2481
2482 while (!(err = link_path_walk(name: s, nd)) &&
2483 (s = lookup_last(nd)) != NULL)
2484 ;
2485 if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2486 err = handle_lookup_down(nd);
2487 nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please...
2488 }
2489 if (!err)
2490 err = complete_walk(nd);
2491
2492 if (!err && nd->flags & LOOKUP_DIRECTORY)
2493 if (!d_can_lookup(dentry: nd->path.dentry))
2494 err = -ENOTDIR;
2495 if (!err) {
2496 *path = nd->path;
2497 nd->path.mnt = NULL;
2498 nd->path.dentry = NULL;
2499 }
2500 terminate_walk(nd);
2501 return err;
2502}
2503
2504int filename_lookup(int dfd, struct filename *name, unsigned flags,
2505 struct path *path, struct path *root)
2506{
2507 int retval;
2508 struct nameidata nd;
2509 if (IS_ERR(ptr: name))
2510 return PTR_ERR(ptr: name);
2511 set_nameidata(p: &nd, dfd, name, root);
2512 retval = path_lookupat(nd: &nd, flags: flags | LOOKUP_RCU, path);
2513 if (unlikely(retval == -ECHILD))
2514 retval = path_lookupat(nd: &nd, flags, path);
2515 if (unlikely(retval == -ESTALE))
2516 retval = path_lookupat(nd: &nd, flags: flags | LOOKUP_REVAL, path);
2517
2518 if (likely(!retval))
2519 audit_inode(name, dentry: path->dentry,
2520 aflags: flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2521 restore_nameidata();
2522 return retval;
2523}
2524
2525/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2526static int path_parentat(struct nameidata *nd, unsigned flags,
2527 struct path *parent)
2528{
2529 const char *s = path_init(nd, flags);
2530 int err = link_path_walk(name: s, nd);
2531 if (!err)
2532 err = complete_walk(nd);
2533 if (!err) {
2534 *parent = nd->path;
2535 nd->path.mnt = NULL;
2536 nd->path.dentry = NULL;
2537 }
2538 terminate_walk(nd);
2539 return err;
2540}
2541
2542/* Note: this does not consume "name" */
2543static int __filename_parentat(int dfd, struct filename *name,
2544 unsigned int flags, struct path *parent,
2545 struct qstr *last, int *type,
2546 const struct path *root)
2547{
2548 int retval;
2549 struct nameidata nd;
2550
2551 if (IS_ERR(ptr: name))
2552 return PTR_ERR(ptr: name);
2553 set_nameidata(p: &nd, dfd, name, root);
2554 retval = path_parentat(nd: &nd, flags: flags | LOOKUP_RCU, parent);
2555 if (unlikely(retval == -ECHILD))
2556 retval = path_parentat(nd: &nd, flags, parent);
2557 if (unlikely(retval == -ESTALE))
2558 retval = path_parentat(nd: &nd, flags: flags | LOOKUP_REVAL, parent);
2559 if (likely(!retval)) {
2560 *last = nd.last;
2561 *type = nd.last_type;
2562 audit_inode(name, dentry: parent->dentry, AUDIT_INODE_PARENT);
2563 }
2564 restore_nameidata();
2565 return retval;
2566}
2567
2568static int filename_parentat(int dfd, struct filename *name,
2569 unsigned int flags, struct path *parent,
2570 struct qstr *last, int *type)
2571{
2572 return __filename_parentat(dfd, name, flags, parent, last, type, NULL);
2573}
2574
2575/* does lookup, returns the object with parent locked */
2576static struct dentry *__kern_path_locked(struct filename *name, struct path *path)
2577{
2578 struct dentry *d;
2579 struct qstr last;
2580 int type, error;
2581
2582 error = filename_parentat(AT_FDCWD, name, flags: 0, parent: path, last: &last, type: &type);
2583 if (error)
2584 return ERR_PTR(error);
2585 if (unlikely(type != LAST_NORM)) {
2586 path_put(path);
2587 return ERR_PTR(error: -EINVAL);
2588 }
2589 inode_lock_nested(inode: path->dentry->d_inode, subclass: I_MUTEX_PARENT);
2590 d = lookup_one_qstr_excl(&last, path->dentry, 0);
2591 if (IS_ERR(ptr: d)) {
2592 inode_unlock(inode: path->dentry->d_inode);
2593 path_put(path);
2594 }
2595 return d;
2596}
2597
2598struct dentry *kern_path_locked(const char *name, struct path *path)
2599{
2600 struct filename *filename = getname_kernel(name);
2601 struct dentry *res = __kern_path_locked(name: filename, path);
2602
2603 putname(filename);
2604 return res;
2605}
2606
2607int kern_path(const char *name, unsigned int flags, struct path *path)
2608{
2609 struct filename *filename = getname_kernel(name);
2610 int ret = filename_lookup(AT_FDCWD, name: filename, flags, path, NULL);
2611
2612 putname(filename);
2613 return ret;
2614
2615}
2616EXPORT_SYMBOL(kern_path);
2617
2618/**
2619 * vfs_path_parent_lookup - lookup a parent path relative to a dentry-vfsmount pair
2620 * @filename: filename structure
2621 * @flags: lookup flags
2622 * @parent: pointer to struct path to fill
2623 * @last: last component
2624 * @type: type of the last component
2625 * @root: pointer to struct path of the base directory
2626 */
2627int vfs_path_parent_lookup(struct filename *filename, unsigned int flags,
2628 struct path *parent, struct qstr *last, int *type,
2629 const struct path *root)
2630{
2631 return __filename_parentat(AT_FDCWD, name: filename, flags, parent, last,
2632 type, root);
2633}
2634EXPORT_SYMBOL(vfs_path_parent_lookup);
2635
2636/**
2637 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2638 * @dentry: pointer to dentry of the base directory
2639 * @mnt: pointer to vfs mount of the base directory
2640 * @name: pointer to file name
2641 * @flags: lookup flags
2642 * @path: pointer to struct path to fill
2643 */
2644int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2645 const char *name, unsigned int flags,
2646 struct path *path)
2647{
2648 struct filename *filename;
2649 struct path root = {.mnt = mnt, .dentry = dentry};
2650 int ret;
2651
2652 filename = getname_kernel(name);
2653 /* the first argument of filename_lookup() is ignored with root */
2654 ret = filename_lookup(AT_FDCWD, name: filename, flags, path, root: &root);
2655 putname(filename);
2656 return ret;
2657}
2658EXPORT_SYMBOL(vfs_path_lookup);
2659
2660static int lookup_one_common(struct mnt_idmap *idmap,
2661 const char *name, struct dentry *base, int len,
2662 struct qstr *this)
2663{
2664 this->name = name;
2665 this->len = len;
2666 this->hash = full_name_hash(base, name, len);
2667 if (!len)
2668 return -EACCES;
2669
2670 if (unlikely(name[0] == '.')) {
2671 if (len < 2 || (len == 2 && name[1] == '.'))
2672 return -EACCES;
2673 }
2674
2675 while (len--) {
2676 unsigned int c = *(const unsigned char *)name++;
2677 if (c == '/' || c == '\0')
2678 return -EACCES;
2679 }
2680 /*
2681 * See if the low-level filesystem might want
2682 * to use its own hash..
2683 */
2684 if (base->d_flags & DCACHE_OP_HASH) {
2685 int err = base->d_op->d_hash(base, this);
2686 if (err < 0)
2687 return err;
2688 }
2689
2690 return inode_permission(idmap, base->d_inode, MAY_EXEC);
2691}
2692
2693/**
2694 * try_lookup_one_len - filesystem helper to lookup single pathname component
2695 * @name: pathname component to lookup
2696 * @base: base directory to lookup from
2697 * @len: maximum length @len should be interpreted to
2698 *
2699 * Look up a dentry by name in the dcache, returning NULL if it does not
2700 * currently exist. The function does not try to create a dentry.
2701 *
2702 * Note that this routine is purely a helper for filesystem usage and should
2703 * not be called by generic code.
2704 *
2705 * The caller must hold base->i_mutex.
2706 */
2707struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2708{
2709 struct qstr this;
2710 int err;
2711
2712 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2713
2714 err = lookup_one_common(idmap: &nop_mnt_idmap, name, base, len, this: &this);
2715 if (err)
2716 return ERR_PTR(error: err);
2717
2718 return lookup_dcache(name: &this, dir: base, flags: 0);
2719}
2720EXPORT_SYMBOL(try_lookup_one_len);
2721
2722/**
2723 * lookup_one_len - filesystem helper to lookup single pathname component
2724 * @name: pathname component to lookup
2725 * @base: base directory to lookup from
2726 * @len: maximum length @len should be interpreted to
2727 *
2728 * Note that this routine is purely a helper for filesystem usage and should
2729 * not be called by generic code.
2730 *
2731 * The caller must hold base->i_mutex.
2732 */
2733struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2734{
2735 struct dentry *dentry;
2736 struct qstr this;
2737 int err;
2738
2739 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2740
2741 err = lookup_one_common(idmap: &nop_mnt_idmap, name, base, len, this: &this);
2742 if (err)
2743 return ERR_PTR(error: err);
2744
2745 dentry = lookup_dcache(name: &this, dir: base, flags: 0);
2746 return dentry ? dentry : __lookup_slow(name: &this, dir: base, flags: 0);
2747}
2748EXPORT_SYMBOL(lookup_one_len);
2749
2750/**
2751 * lookup_one - filesystem helper to lookup single pathname component
2752 * @idmap: idmap of the mount the lookup is performed from
2753 * @name: pathname component to lookup
2754 * @base: base directory to lookup from
2755 * @len: maximum length @len should be interpreted to
2756 *
2757 * Note that this routine is purely a helper for filesystem usage and should
2758 * not be called by generic code.
2759 *
2760 * The caller must hold base->i_mutex.
2761 */
2762struct dentry *lookup_one(struct mnt_idmap *idmap, const char *name,
2763 struct dentry *base, int len)
2764{
2765 struct dentry *dentry;
2766 struct qstr this;
2767 int err;
2768
2769 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2770
2771 err = lookup_one_common(idmap, name, base, len, this: &this);
2772 if (err)
2773 return ERR_PTR(error: err);
2774
2775 dentry = lookup_dcache(name: &this, dir: base, flags: 0);
2776 return dentry ? dentry : __lookup_slow(name: &this, dir: base, flags: 0);
2777}
2778EXPORT_SYMBOL(lookup_one);
2779
2780/**
2781 * lookup_one_unlocked - filesystem helper to lookup single pathname component
2782 * @idmap: idmap of the mount the lookup is performed from
2783 * @name: pathname component to lookup
2784 * @base: base directory to lookup from
2785 * @len: maximum length @len should be interpreted to
2786 *
2787 * Note that this routine is purely a helper for filesystem usage and should
2788 * not be called by generic code.
2789 *
2790 * Unlike lookup_one_len, it should be called without the parent
2791 * i_mutex held, and will take the i_mutex itself if necessary.
2792 */
2793struct dentry *lookup_one_unlocked(struct mnt_idmap *idmap,
2794 const char *name, struct dentry *base,
2795 int len)
2796{
2797 struct qstr this;
2798 int err;
2799 struct dentry *ret;
2800
2801 err = lookup_one_common(idmap, name, base, len, this: &this);
2802 if (err)
2803 return ERR_PTR(error: err);
2804
2805 ret = lookup_dcache(name: &this, dir: base, flags: 0);
2806 if (!ret)
2807 ret = lookup_slow(name: &this, dir: base, flags: 0);
2808 return ret;
2809}
2810EXPORT_SYMBOL(lookup_one_unlocked);
2811
2812/**
2813 * lookup_one_positive_unlocked - filesystem helper to lookup single
2814 * pathname component
2815 * @idmap: idmap of the mount the lookup is performed from
2816 * @name: pathname component to lookup
2817 * @base: base directory to lookup from
2818 * @len: maximum length @len should be interpreted to
2819 *
2820 * This helper will yield ERR_PTR(-ENOENT) on negatives. The helper returns
2821 * known positive or ERR_PTR(). This is what most of the users want.
2822 *
2823 * Note that pinned negative with unlocked parent _can_ become positive at any
2824 * time, so callers of lookup_one_unlocked() need to be very careful; pinned
2825 * positives have >d_inode stable, so this one avoids such problems.
2826 *
2827 * Note that this routine is purely a helper for filesystem usage and should
2828 * not be called by generic code.
2829 *
2830 * The helper should be called without i_mutex held.
2831 */
2832struct dentry *lookup_one_positive_unlocked(struct mnt_idmap *idmap,
2833 const char *name,
2834 struct dentry *base, int len)
2835{
2836 struct dentry *ret = lookup_one_unlocked(idmap, name, base, len);
2837
2838 if (!IS_ERR(ptr: ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2839 dput(ret);
2840 ret = ERR_PTR(error: -ENOENT);
2841 }
2842 return ret;
2843}
2844EXPORT_SYMBOL(lookup_one_positive_unlocked);
2845
2846/**
2847 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2848 * @name: pathname component to lookup
2849 * @base: base directory to lookup from
2850 * @len: maximum length @len should be interpreted to
2851 *
2852 * Note that this routine is purely a helper for filesystem usage and should
2853 * not be called by generic code.
2854 *
2855 * Unlike lookup_one_len, it should be called without the parent
2856 * i_mutex held, and will take the i_mutex itself if necessary.
2857 */
2858struct dentry *lookup_one_len_unlocked(const char *name,
2859 struct dentry *base, int len)
2860{
2861 return lookup_one_unlocked(&nop_mnt_idmap, name, base, len);
2862}
2863EXPORT_SYMBOL(lookup_one_len_unlocked);
2864
2865/*
2866 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2867 * on negatives. Returns known positive or ERR_PTR(); that's what
2868 * most of the users want. Note that pinned negative with unlocked parent
2869 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2870 * need to be very careful; pinned positives have ->d_inode stable, so
2871 * this one avoids such problems.
2872 */
2873struct dentry *lookup_positive_unlocked(const char *name,
2874 struct dentry *base, int len)
2875{
2876 return lookup_one_positive_unlocked(&nop_mnt_idmap, name, base, len);
2877}
2878EXPORT_SYMBOL(lookup_positive_unlocked);
2879
2880#ifdef CONFIG_UNIX98_PTYS
2881int path_pts(struct path *path)
2882{
2883 /* Find something mounted on "pts" in the same directory as
2884 * the input path.
2885 */
2886 struct dentry *parent = dget_parent(dentry: path->dentry);
2887 struct dentry *child;
2888 struct qstr this = QSTR_INIT("pts", 3);
2889
2890 if (unlikely(!path_connected(path->mnt, parent))) {
2891 dput(parent);
2892 return -ENOENT;
2893 }
2894 dput(path->dentry);
2895 path->dentry = parent;
2896 child = d_hash_and_lookup(parent, &this);
2897 if (IS_ERR_OR_NULL(ptr: child))
2898 return -ENOENT;
2899
2900 path->dentry = child;
2901 dput(parent);
2902 follow_down(path, 0);
2903 return 0;
2904}
2905#endif
2906
2907int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2908 struct path *path, int *empty)
2909{
2910 struct filename *filename = getname_flags(filename: name, flags, empty);
2911 int ret = filename_lookup(dfd, name: filename, flags, path, NULL);
2912
2913 putname(filename);
2914 return ret;
2915}
2916EXPORT_SYMBOL(user_path_at_empty);
2917
2918int __check_sticky(struct mnt_idmap *idmap, struct inode *dir,
2919 struct inode *inode)
2920{
2921 kuid_t fsuid = current_fsuid();
2922
2923 if (vfsuid_eq_kuid(vfsuid: i_uid_into_vfsuid(idmap, inode), kuid: fsuid))
2924 return 0;
2925 if (vfsuid_eq_kuid(vfsuid: i_uid_into_vfsuid(idmap, inode: dir), kuid: fsuid))
2926 return 0;
2927 return !capable_wrt_inode_uidgid(idmap, inode, CAP_FOWNER);
2928}
2929EXPORT_SYMBOL(__check_sticky);
2930
2931/*
2932 * Check whether we can remove a link victim from directory dir, check
2933 * whether the type of victim is right.
2934 * 1. We can't do it if dir is read-only (done in permission())
2935 * 2. We should have write and exec permissions on dir
2936 * 3. We can't remove anything from append-only dir
2937 * 4. We can't do anything with immutable dir (done in permission())
2938 * 5. If the sticky bit on dir is set we should either
2939 * a. be owner of dir, or
2940 * b. be owner of victim, or
2941 * c. have CAP_FOWNER capability
2942 * 6. If the victim is append-only or immutable we can't do antyhing with
2943 * links pointing to it.
2944 * 7. If the victim has an unknown uid or gid we can't change the inode.
2945 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2946 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2947 * 10. We can't remove a root or mountpoint.
2948 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2949 * nfs_async_unlink().
2950 */
2951static int may_delete(struct mnt_idmap *idmap, struct inode *dir,
2952 struct dentry *victim, bool isdir)
2953{
2954 struct inode *inode = d_backing_inode(upper: victim);
2955 int error;
2956
2957 if (d_is_negative(dentry: victim))
2958 return -ENOENT;
2959 BUG_ON(!inode);
2960
2961 BUG_ON(victim->d_parent->d_inode != dir);
2962
2963 /* Inode writeback is not safe when the uid or gid are invalid. */
2964 if (!vfsuid_valid(uid: i_uid_into_vfsuid(idmap, inode)) ||
2965 !vfsgid_valid(gid: i_gid_into_vfsgid(idmap, inode)))
2966 return -EOVERFLOW;
2967
2968 audit_inode_child(parent: dir, dentry: victim, AUDIT_TYPE_CHILD_DELETE);
2969
2970 error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
2971 if (error)
2972 return error;
2973 if (IS_APPEND(dir))
2974 return -EPERM;
2975
2976 if (check_sticky(idmap, dir, inode) || IS_APPEND(inode) ||
2977 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) ||
2978 HAS_UNMAPPED_ID(idmap, inode))
2979 return -EPERM;
2980 if (isdir) {
2981 if (!d_is_dir(dentry: victim))
2982 return -ENOTDIR;
2983 if (IS_ROOT(victim))
2984 return -EBUSY;
2985 } else if (d_is_dir(dentry: victim))
2986 return -EISDIR;
2987 if (IS_DEADDIR(dir))
2988 return -ENOENT;
2989 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2990 return -EBUSY;
2991 return 0;
2992}
2993
2994/* Check whether we can create an object with dentry child in directory
2995 * dir.
2996 * 1. We can't do it if child already exists (open has special treatment for
2997 * this case, but since we are inlined it's OK)
2998 * 2. We can't do it if dir is read-only (done in permission())
2999 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
3000 * 4. We should have write and exec permissions on dir
3001 * 5. We can't do it if dir is immutable (done in permission())
3002 */
3003static inline int may_create(struct mnt_idmap *idmap,
3004 struct inode *dir, struct dentry *child)
3005{
3006 audit_inode_child(parent: dir, dentry: child, AUDIT_TYPE_CHILD_CREATE);
3007 if (child->d_inode)
3008 return -EEXIST;
3009 if (IS_DEADDIR(dir))
3010 return -ENOENT;
3011 if (!fsuidgid_has_mapping(sb: dir->i_sb, idmap))
3012 return -EOVERFLOW;
3013
3014 return inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
3015}
3016
3017static struct dentry *lock_two_directories(struct dentry *p1, struct dentry *p2)
3018{
3019 struct dentry *p;
3020
3021 p = d_ancestor(p2, p1);
3022 if (p) {
3023 inode_lock_nested(inode: p2->d_inode, subclass: I_MUTEX_PARENT);
3024 inode_lock_nested(inode: p1->d_inode, subclass: I_MUTEX_CHILD);
3025 return p;
3026 }
3027
3028 p = d_ancestor(p1, p2);
3029 if (p) {
3030 inode_lock_nested(inode: p1->d_inode, subclass: I_MUTEX_PARENT);
3031 inode_lock_nested(inode: p2->d_inode, subclass: I_MUTEX_CHILD);
3032 return p;
3033 }
3034
3035 lock_two_inodes(inode1: p1->d_inode, inode2: p2->d_inode,
3036 subclass1: I_MUTEX_PARENT, subclass2: I_MUTEX_PARENT2);
3037 return NULL;
3038}
3039
3040/*
3041 * p1 and p2 should be directories on the same fs.
3042 */
3043struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
3044{
3045 if (p1 == p2) {
3046 inode_lock_nested(inode: p1->d_inode, subclass: I_MUTEX_PARENT);
3047 return NULL;
3048 }
3049
3050 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
3051 return lock_two_directories(p1, p2);
3052}
3053EXPORT_SYMBOL(lock_rename);
3054
3055/*
3056 * c1 and p2 should be on the same fs.
3057 */
3058struct dentry *lock_rename_child(struct dentry *c1, struct dentry *p2)
3059{
3060 if (READ_ONCE(c1->d_parent) == p2) {
3061 /*
3062 * hopefully won't need to touch ->s_vfs_rename_mutex at all.
3063 */
3064 inode_lock_nested(inode: p2->d_inode, subclass: I_MUTEX_PARENT);
3065 /*
3066 * now that p2 is locked, nobody can move in or out of it,
3067 * so the test below is safe.
3068 */
3069 if (likely(c1->d_parent == p2))
3070 return NULL;
3071
3072 /*
3073 * c1 got moved out of p2 while we'd been taking locks;
3074 * unlock and fall back to slow case.
3075 */
3076 inode_unlock(inode: p2->d_inode);
3077 }
3078
3079 mutex_lock(&c1->d_sb->s_vfs_rename_mutex);
3080 /*
3081 * nobody can move out of any directories on this fs.
3082 */
3083 if (likely(c1->d_parent != p2))
3084 return lock_two_directories(p1: c1->d_parent, p2);
3085
3086 /*
3087 * c1 got moved into p2 while we were taking locks;
3088 * we need p2 locked and ->s_vfs_rename_mutex unlocked,
3089 * for consistency with lock_rename().
3090 */
3091 inode_lock_nested(inode: p2->d_inode, subclass: I_MUTEX_PARENT);
3092 mutex_unlock(lock: &c1->d_sb->s_vfs_rename_mutex);
3093 return NULL;
3094}
3095EXPORT_SYMBOL(lock_rename_child);
3096
3097void unlock_rename(struct dentry *p1, struct dentry *p2)
3098{
3099 inode_unlock(inode: p1->d_inode);
3100 if (p1 != p2) {
3101 inode_unlock(inode: p2->d_inode);
3102 mutex_unlock(lock: &p1->d_sb->s_vfs_rename_mutex);
3103 }
3104}
3105EXPORT_SYMBOL(unlock_rename);
3106
3107/**
3108 * vfs_prepare_mode - prepare the mode to be used for a new inode
3109 * @idmap: idmap of the mount the inode was found from
3110 * @dir: parent directory of the new inode
3111 * @mode: mode of the new inode
3112 * @mask_perms: allowed permission by the vfs
3113 * @type: type of file to be created
3114 *
3115 * This helper consolidates and enforces vfs restrictions on the @mode of a new
3116 * object to be created.
3117 *
3118 * Umask stripping depends on whether the filesystem supports POSIX ACLs (see
3119 * the kernel documentation for mode_strip_umask()). Moving umask stripping
3120 * after setgid stripping allows the same ordering for both non-POSIX ACL and
3121 * POSIX ACL supporting filesystems.
3122 *
3123 * Note that it's currently valid for @type to be 0 if a directory is created.
3124 * Filesystems raise that flag individually and we need to check whether each
3125 * filesystem can deal with receiving S_IFDIR from the vfs before we enforce a
3126 * non-zero type.
3127 *
3128 * Returns: mode to be passed to the filesystem
3129 */
3130static inline umode_t vfs_prepare_mode(struct mnt_idmap *idmap,
3131 const struct inode *dir, umode_t mode,
3132 umode_t mask_perms, umode_t type)
3133{
3134 mode = mode_strip_sgid(idmap, dir, mode);
3135 mode = mode_strip_umask(dir, mode);
3136
3137 /*
3138 * Apply the vfs mandated allowed permission mask and set the type of
3139 * file to be created before we call into the filesystem.
3140 */
3141 mode &= (mask_perms & ~S_IFMT);
3142 mode |= (type & S_IFMT);
3143
3144 return mode;
3145}
3146
3147/**
3148 * vfs_create - create new file
3149 * @idmap: idmap of the mount the inode was found from
3150 * @dir: inode of @dentry
3151 * @dentry: pointer to dentry of the base directory
3152 * @mode: mode of the new file
3153 * @want_excl: whether the file must not yet exist
3154 *
3155 * Create a new file.
3156 *
3157 * If the inode has been found through an idmapped mount the idmap of
3158 * the vfsmount must be passed through @idmap. This function will then take
3159 * care to map the inode according to @idmap before checking permissions.
3160 * On non-idmapped mounts or if permission checking is to be performed on the
3161 * raw inode simply passs @nop_mnt_idmap.
3162 */
3163int vfs_create(struct mnt_idmap *idmap, struct inode *dir,
3164 struct dentry *dentry, umode_t mode, bool want_excl)
3165{
3166 int error;
3167
3168 error = may_create(idmap, dir, child: dentry);
3169 if (error)
3170 return error;
3171
3172 if (!dir->i_op->create)
3173 return -EACCES; /* shouldn't it be ENOSYS? */
3174
3175 mode = vfs_prepare_mode(idmap, dir, mode, S_IALLUGO, S_IFREG);
3176 error = security_inode_create(dir, dentry, mode);
3177 if (error)
3178 return error;
3179 error = dir->i_op->create(idmap, dir, dentry, mode, want_excl);
3180 if (!error)
3181 fsnotify_create(dir, dentry);
3182 return error;
3183}
3184EXPORT_SYMBOL(vfs_create);
3185
3186int vfs_mkobj(struct dentry *dentry, umode_t mode,
3187 int (*f)(struct dentry *, umode_t, void *),
3188 void *arg)
3189{
3190 struct inode *dir = dentry->d_parent->d_inode;
3191 int error = may_create(idmap: &nop_mnt_idmap, dir, child: dentry);
3192 if (error)
3193 return error;
3194
3195 mode &= S_IALLUGO;
3196 mode |= S_IFREG;
3197 error = security_inode_create(dir, dentry, mode);
3198 if (error)
3199 return error;
3200 error = f(dentry, mode, arg);
3201 if (!error)
3202 fsnotify_create(dir, dentry);
3203 return error;
3204}
3205EXPORT_SYMBOL(vfs_mkobj);
3206
3207bool may_open_dev(const struct path *path)
3208{
3209 return !(path->mnt->mnt_flags & MNT_NODEV) &&
3210 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
3211}
3212
3213static int may_open(struct mnt_idmap *idmap, const struct path *path,
3214 int acc_mode, int flag)
3215{
3216 struct dentry *dentry = path->dentry;
3217 struct inode *inode = dentry->d_inode;
3218 int error;
3219
3220 if (!inode)
3221 return -ENOENT;
3222
3223 switch (inode->i_mode & S_IFMT) {
3224 case S_IFLNK:
3225 return -ELOOP;
3226 case S_IFDIR:
3227 if (acc_mode & MAY_WRITE)
3228 return -EISDIR;
3229 if (acc_mode & MAY_EXEC)
3230 return -EACCES;
3231 break;
3232 case S_IFBLK:
3233 case S_IFCHR:
3234 if (!may_open_dev(path))
3235 return -EACCES;
3236 fallthrough;
3237 case S_IFIFO:
3238 case S_IFSOCK:
3239 if (acc_mode & MAY_EXEC)
3240 return -EACCES;
3241 flag &= ~O_TRUNC;
3242 break;
3243 case S_IFREG:
3244 if ((acc_mode & MAY_EXEC) && path_noexec(path))
3245 return -EACCES;
3246 break;
3247 }
3248
3249 error = inode_permission(idmap, inode, MAY_OPEN | acc_mode);
3250 if (error)
3251 return error;
3252
3253 /*
3254 * An append-only file must be opened in append mode for writing.
3255 */
3256 if (IS_APPEND(inode)) {
3257 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
3258 return -EPERM;
3259 if (flag & O_TRUNC)
3260 return -EPERM;
3261 }
3262
3263 /* O_NOATIME can only be set by the owner or superuser */
3264 if (flag & O_NOATIME && !inode_owner_or_capable(idmap, inode))
3265 return -EPERM;
3266
3267 return 0;
3268}
3269
3270static int handle_truncate(struct mnt_idmap *idmap, struct file *filp)
3271{
3272 const struct path *path = &filp->f_path;
3273 struct inode *inode = path->dentry->d_inode;
3274 int error = get_write_access(inode);
3275 if (error)
3276 return error;
3277
3278 error = security_file_truncate(file: filp);
3279 if (!error) {
3280 error = do_truncate(idmap, path->dentry, start: 0,
3281 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3282 filp);
3283 }
3284 put_write_access(inode);
3285 return error;
3286}
3287
3288static inline int open_to_namei_flags(int flag)
3289{
3290 if ((flag & O_ACCMODE) == 3)
3291 flag--;
3292 return flag;
3293}
3294
3295static int may_o_create(struct mnt_idmap *idmap,
3296 const struct path *dir, struct dentry *dentry,
3297 umode_t mode)
3298{
3299 int error = security_path_mknod(dir, dentry, mode, dev: 0);
3300 if (error)
3301 return error;
3302
3303 if (!fsuidgid_has_mapping(sb: dir->dentry->d_sb, idmap))
3304 return -EOVERFLOW;
3305
3306 error = inode_permission(idmap, dir->dentry->d_inode,
3307 MAY_WRITE | MAY_EXEC);
3308 if (error)
3309 return error;
3310
3311 return security_inode_create(dir: dir->dentry->d_inode, dentry, mode);
3312}
3313
3314/*
3315 * Attempt to atomically look up, create and open a file from a negative
3316 * dentry.
3317 *
3318 * Returns 0 if successful. The file will have been created and attached to
3319 * @file by the filesystem calling finish_open().
3320 *
3321 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3322 * be set. The caller will need to perform the open themselves. @path will
3323 * have been updated to point to the new dentry. This may be negative.
3324 *
3325 * Returns an error code otherwise.
3326 */
3327static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
3328 struct file *file,
3329 int open_flag, umode_t mode)
3330{
3331 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3332 struct inode *dir = nd->path.dentry->d_inode;
3333 int error;
3334
3335 if (nd->flags & LOOKUP_DIRECTORY)
3336 open_flag |= O_DIRECTORY;
3337
3338 file->f_path.dentry = DENTRY_NOT_SET;
3339 file->f_path.mnt = nd->path.mnt;
3340 error = dir->i_op->atomic_open(dir, dentry, file,
3341 open_to_namei_flags(flag: open_flag), mode);
3342 d_lookup_done(dentry);
3343 if (!error) {
3344 if (file->f_mode & FMODE_OPENED) {
3345 if (unlikely(dentry != file->f_path.dentry)) {
3346 dput(dentry);
3347 dentry = dget(dentry: file->f_path.dentry);
3348 }
3349 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3350 error = -EIO;
3351 } else {
3352 if (file->f_path.dentry) {
3353 dput(dentry);
3354 dentry = file->f_path.dentry;
3355 }
3356 if (unlikely(d_is_negative(dentry)))
3357 error = -ENOENT;
3358 }
3359 }
3360 if (error) {
3361 dput(dentry);
3362 dentry = ERR_PTR(error);
3363 }
3364 return dentry;
3365}
3366
3367/*
3368 * Look up and maybe create and open the last component.
3369 *
3370 * Must be called with parent locked (exclusive in O_CREAT case).
3371 *
3372 * Returns 0 on success, that is, if
3373 * the file was successfully atomically created (if necessary) and opened, or
3374 * the file was not completely opened at this time, though lookups and
3375 * creations were performed.
3376 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3377 * In the latter case dentry returned in @path might be negative if O_CREAT
3378 * hadn't been specified.
3379 *
3380 * An error code is returned on failure.
3381 */
3382static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3383 const struct open_flags *op,
3384 bool got_write)
3385{
3386 struct mnt_idmap *idmap;
3387 struct dentry *dir = nd->path.dentry;
3388 struct inode *dir_inode = dir->d_inode;
3389 int open_flag = op->open_flag;
3390 struct dentry *dentry;
3391 int error, create_error = 0;
3392 umode_t mode = op->mode;
3393 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3394
3395 if (unlikely(IS_DEADDIR(dir_inode)))
3396 return ERR_PTR(error: -ENOENT);
3397
3398 file->f_mode &= ~FMODE_CREATED;
3399 dentry = d_lookup(dir, &nd->last);
3400 for (;;) {
3401 if (!dentry) {
3402 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3403 if (IS_ERR(ptr: dentry))
3404 return dentry;
3405 }
3406 if (d_in_lookup(dentry))
3407 break;
3408
3409 error = d_revalidate(dentry, flags: nd->flags);
3410 if (likely(error > 0))
3411 break;
3412 if (error)
3413 goto out_dput;
3414 d_invalidate(dentry);
3415 dput(dentry);
3416 dentry = NULL;
3417 }
3418 if (dentry->d_inode) {
3419 /* Cached positive dentry: will open in f_op->open */
3420 return dentry;
3421 }
3422
3423 /*
3424 * Checking write permission is tricky, bacuse we don't know if we are
3425 * going to actually need it: O_CREAT opens should work as long as the
3426 * file exists. But checking existence breaks atomicity. The trick is
3427 * to check access and if not granted clear O_CREAT from the flags.
3428 *
3429 * Another problem is returing the "right" error value (e.g. for an
3430 * O_EXCL open we want to return EEXIST not EROFS).
3431 */
3432 if (unlikely(!got_write))
3433 open_flag &= ~O_TRUNC;
3434 idmap = mnt_idmap(mnt: nd->path.mnt);
3435 if (open_flag & O_CREAT) {
3436 if (open_flag & O_EXCL)
3437 open_flag &= ~O_TRUNC;
3438 mode = vfs_prepare_mode(idmap, dir: dir->d_inode, mode, mask_perms: mode, type: mode);
3439 if (likely(got_write))
3440 create_error = may_o_create(idmap, dir: &nd->path,
3441 dentry, mode);
3442 else
3443 create_error = -EROFS;
3444 }
3445 if (create_error)
3446 open_flag &= ~O_CREAT;
3447 if (dir_inode->i_op->atomic_open) {
3448 dentry = atomic_open(nd, dentry, file, open_flag, mode);
3449 if (unlikely(create_error) && dentry == ERR_PTR(error: -ENOENT))
3450 dentry = ERR_PTR(error: create_error);
3451 return dentry;
3452 }
3453
3454 if (d_in_lookup(dentry)) {
3455 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3456 nd->flags);
3457 d_lookup_done(dentry);
3458 if (unlikely(res)) {
3459 if (IS_ERR(ptr: res)) {
3460 error = PTR_ERR(ptr: res);
3461 goto out_dput;
3462 }
3463 dput(dentry);
3464 dentry = res;
3465 }
3466 }
3467
3468 /* Negative dentry, just create the file */
3469 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3470 file->f_mode |= FMODE_CREATED;
3471 audit_inode_child(parent: dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3472 if (!dir_inode->i_op->create) {
3473 error = -EACCES;
3474 goto out_dput;
3475 }
3476
3477 error = dir_inode->i_op->create(idmap, dir_inode, dentry,
3478 mode, open_flag & O_EXCL);
3479 if (error)
3480 goto out_dput;
3481 }
3482 if (unlikely(create_error) && !dentry->d_inode) {
3483 error = create_error;
3484 goto out_dput;
3485 }
3486 return dentry;
3487
3488out_dput:
3489 dput(dentry);
3490 return ERR_PTR(error);
3491}
3492
3493static const char *open_last_lookups(struct nameidata *nd,
3494 struct file *file, const struct open_flags *op)
3495{
3496 struct dentry *dir = nd->path.dentry;
3497 int open_flag = op->open_flag;
3498 bool got_write = false;
3499 struct dentry *dentry;
3500 const char *res;
3501
3502 nd->flags |= op->intent;
3503
3504 if (nd->last_type != LAST_NORM) {
3505 if (nd->depth)
3506 put_link(nd);
3507 return handle_dots(nd, type: nd->last_type);
3508 }
3509
3510 if (!(open_flag & O_CREAT)) {
3511 if (nd->last.name[nd->last.len])
3512 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3513 /* we _can_ be in RCU mode here */
3514 dentry = lookup_fast(nd);
3515 if (IS_ERR(ptr: dentry))
3516 return ERR_CAST(ptr: dentry);
3517 if (likely(dentry))
3518 goto finish_lookup;
3519
3520 if (WARN_ON_ONCE(nd->flags & LOOKUP_RCU))
3521 return ERR_PTR(error: -ECHILD);
3522 } else {
3523 /* create side of things */
3524 if (nd->flags & LOOKUP_RCU) {
3525 if (!try_to_unlazy(nd))
3526 return ERR_PTR(error: -ECHILD);
3527 }
3528 audit_inode(name: nd->name, dentry: dir, AUDIT_INODE_PARENT);
3529 /* trailing slashes? */
3530 if (unlikely(nd->last.name[nd->last.len]))
3531 return ERR_PTR(error: -EISDIR);
3532 }
3533
3534 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3535 got_write = !mnt_want_write(mnt: nd->path.mnt);
3536 /*
3537 * do _not_ fail yet - we might not need that or fail with
3538 * a different error; let lookup_open() decide; we'll be
3539 * dropping this one anyway.
3540 */
3541 }
3542 if (open_flag & O_CREAT)
3543 inode_lock(inode: dir->d_inode);
3544 else
3545 inode_lock_shared(inode: dir->d_inode);
3546 dentry = lookup_open(nd, file, op, got_write);
3547 if (!IS_ERR(ptr: dentry) && (file->f_mode & FMODE_CREATED))
3548 fsnotify_create(dir: dir->d_inode, dentry);
3549 if (open_flag & O_CREAT)
3550 inode_unlock(inode: dir->d_inode);
3551 else
3552 inode_unlock_shared(inode: dir->d_inode);
3553
3554 if (got_write)
3555 mnt_drop_write(mnt: nd->path.mnt);
3556
3557 if (IS_ERR(ptr: dentry))
3558 return ERR_CAST(ptr: dentry);
3559
3560 if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
3561 dput(nd->path.dentry);
3562 nd->path.dentry = dentry;
3563 return NULL;
3564 }
3565
3566finish_lookup:
3567 if (nd->depth)
3568 put_link(nd);
3569 res = step_into(nd, flags: WALK_TRAILING, dentry);
3570 if (unlikely(res))
3571 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3572 return res;
3573}
3574
3575/*
3576 * Handle the last step of open()
3577 */
3578static int do_open(struct nameidata *nd,
3579 struct file *file, const struct open_flags *op)
3580{
3581 struct mnt_idmap *idmap;
3582 int open_flag = op->open_flag;
3583 bool do_truncate;
3584 int acc_mode;
3585 int error;
3586
3587 if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3588 error = complete_walk(nd);
3589 if (error)
3590 return error;
3591 }
3592 if (!(file->f_mode & FMODE_CREATED))
3593 audit_inode(name: nd->name, dentry: nd->path.dentry, aflags: 0);
3594 idmap = mnt_idmap(mnt: nd->path.mnt);
3595 if (open_flag & O_CREAT) {
3596 if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3597 return -EEXIST;
3598 if (d_is_dir(dentry: nd->path.dentry))
3599 return -EISDIR;
3600 error = may_create_in_sticky(idmap, nd,
3601 inode: d_backing_inode(upper: nd->path.dentry));
3602 if (unlikely(error))
3603 return error;
3604 }
3605 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(dentry: nd->path.dentry))
3606 return -ENOTDIR;
3607
3608 do_truncate = false;
3609 acc_mode = op->acc_mode;
3610 if (file->f_mode & FMODE_CREATED) {
3611 /* Don't check for write permission, don't truncate */
3612 open_flag &= ~O_TRUNC;
3613 acc_mode = 0;
3614 } else if (d_is_reg(dentry: nd->path.dentry) && open_flag & O_TRUNC) {
3615 error = mnt_want_write(mnt: nd->path.mnt);
3616 if (error)
3617 return error;
3618 do_truncate = true;
3619 }
3620 error = may_open(idmap, path: &nd->path, acc_mode, flag: open_flag);
3621 if (!error && !(file->f_mode & FMODE_OPENED))
3622 error = vfs_open(&nd->path, file);
3623 if (!error)
3624 error = ima_file_check(file, mask: op->acc_mode);
3625 if (!error && do_truncate)
3626 error = handle_truncate(idmap, filp: file);
3627 if (unlikely(error > 0)) {
3628 WARN_ON(1);
3629 error = -EINVAL;
3630 }
3631 if (do_truncate)
3632 mnt_drop_write(mnt: nd->path.mnt);
3633 return error;
3634}
3635
3636/**
3637 * vfs_tmpfile - create tmpfile
3638 * @idmap: idmap of the mount the inode was found from
3639 * @parentpath: pointer to the path of the base directory
3640 * @file: file descriptor of the new tmpfile
3641 * @mode: mode of the new tmpfile
3642 *
3643 * Create a temporary file.
3644 *
3645 * If the inode has been found through an idmapped mount the idmap of
3646 * the vfsmount must be passed through @idmap. This function will then take
3647 * care to map the inode according to @idmap before checking permissions.
3648 * On non-idmapped mounts or if permission checking is to be performed on the
3649 * raw inode simply passs @nop_mnt_idmap.
3650 */
3651static int vfs_tmpfile(struct mnt_idmap *idmap,
3652 const struct path *parentpath,
3653 struct file *file, umode_t mode)
3654{
3655 struct dentry *child;
3656 struct inode *dir = d_inode(dentry: parentpath->dentry);
3657 struct inode *inode;
3658 int error;
3659 int open_flag = file->f_flags;
3660
3661 /* we want directory to be writable */
3662 error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
3663 if (error)
3664 return error;
3665 if (!dir->i_op->tmpfile)
3666 return -EOPNOTSUPP;
3667 child = d_alloc(parentpath->dentry, &slash_name);
3668 if (unlikely(!child))
3669 return -ENOMEM;
3670 file->f_path.mnt = parentpath->mnt;
3671 file->f_path.dentry = child;
3672 mode = vfs_prepare_mode(idmap, dir, mode, mask_perms: mode, type: mode);
3673 error = dir->i_op->tmpfile(idmap, dir, file, mode);
3674 dput(child);
3675 if (error)
3676 return error;
3677 /* Don't check for other permissions, the inode was just created */
3678 error = may_open(idmap, path: &file->f_path, acc_mode: 0, flag: file->f_flags);
3679 if (error)
3680 return error;
3681 inode = file_inode(f: file);
3682 if (!(open_flag & O_EXCL)) {
3683 spin_lock(lock: &inode->i_lock);
3684 inode->i_state |= I_LINKABLE;
3685 spin_unlock(lock: &inode->i_lock);
3686 }
3687 ima_post_create_tmpfile(idmap, inode);
3688 return 0;
3689}
3690
3691/**
3692 * kernel_tmpfile_open - open a tmpfile for kernel internal use
3693 * @idmap: idmap of the mount the inode was found from
3694 * @parentpath: path of the base directory
3695 * @mode: mode of the new tmpfile
3696 * @open_flag: flags
3697 * @cred: credentials for open
3698 *
3699 * Create and open a temporary file. The file is not accounted in nr_files,
3700 * hence this is only for kernel internal use, and must not be installed into
3701 * file tables or such.
3702 */
3703struct file *kernel_tmpfile_open(struct mnt_idmap *idmap,
3704 const struct path *parentpath,
3705 umode_t mode, int open_flag,
3706 const struct cred *cred)
3707{
3708 struct file *file;
3709 int error;
3710
3711 file = alloc_empty_file_noaccount(flags: open_flag, cred);
3712 if (IS_ERR(ptr: file))
3713 return file;
3714
3715 error = vfs_tmpfile(idmap, parentpath, file, mode);
3716 if (error) {
3717 fput(file);
3718 file = ERR_PTR(error);
3719 }
3720 return file;
3721}
3722EXPORT_SYMBOL(kernel_tmpfile_open);
3723
3724static int do_tmpfile(struct nameidata *nd, unsigned flags,
3725 const struct open_flags *op,
3726 struct file *file)
3727{
3728 struct path path;
3729 int error = path_lookupat(nd, flags: flags | LOOKUP_DIRECTORY, path: &path);
3730
3731 if (unlikely(error))
3732 return error;
3733 error = mnt_want_write(mnt: path.mnt);
3734 if (unlikely(error))
3735 goto out;
3736 error = vfs_tmpfile(idmap: mnt_idmap(mnt: path.mnt), parentpath: &path, file, mode: op->mode);
3737 if (error)
3738 goto out2;
3739 audit_inode(name: nd->name, dentry: file->f_path.dentry, aflags: 0);
3740out2:
3741 mnt_drop_write(mnt: path.mnt);
3742out:
3743 path_put(&path);
3744 return error;
3745}
3746
3747static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3748{
3749 struct path path;
3750 int error = path_lookupat(nd, flags, path: &path);
3751 if (!error) {
3752 audit_inode(name: nd->name, dentry: path.dentry, aflags: 0);
3753 error = vfs_open(&path, file);
3754 path_put(&path);
3755 }
3756 return error;
3757}
3758
3759static struct file *path_openat(struct nameidata *nd,
3760 const struct open_flags *op, unsigned flags)
3761{
3762 struct file *file;
3763 int error;
3764
3765 file = alloc_empty_file(flags: op->open_flag, current_cred());
3766 if (IS_ERR(ptr: file))
3767 return file;
3768
3769 if (unlikely(file->f_flags & __O_TMPFILE)) {
3770 error = do_tmpfile(nd, flags, op, file);
3771 } else if (unlikely(file->f_flags & O_PATH)) {
3772 error = do_o_path(nd, flags, file);
3773 } else {
3774 const char *s = path_init(nd, flags);
3775 while (!(error = link_path_walk(name: s, nd)) &&
3776 (s = open_last_lookups(nd, file, op)) != NULL)
3777 ;
3778 if (!error)
3779 error = do_open(nd, file, op);
3780 terminate_walk(nd);
3781 }
3782 if (likely(!error)) {
3783 if (likely(file->f_mode & FMODE_OPENED))
3784 return file;
3785 WARN_ON(1);
3786 error = -EINVAL;
3787 }
3788 if (unlikely(file->f_mode & FMODE_OPENED))
3789 fput(file);
3790 else
3791 release_empty_file(f: file);
3792 if (error == -EOPENSTALE) {
3793 if (flags & LOOKUP_RCU)
3794 error = -ECHILD;
3795 else
3796 error = -ESTALE;
3797 }
3798 return ERR_PTR(error);
3799}
3800
3801struct file *do_filp_open(int dfd, struct filename *pathname,
3802 const struct open_flags *op)
3803{
3804 struct nameidata nd;
3805 int flags = op->lookup_flags;
3806 struct file *filp;
3807
3808 set_nameidata(p: &nd, dfd, name: pathname, NULL);
3809 filp = path_openat(nd: &nd, op, flags: flags | LOOKUP_RCU);
3810 if (unlikely(filp == ERR_PTR(-ECHILD)))
3811 filp = path_openat(nd: &nd, op, flags);
3812 if (unlikely(filp == ERR_PTR(-ESTALE)))
3813 filp = path_openat(nd: &nd, op, flags: flags | LOOKUP_REVAL);
3814 restore_nameidata();
3815 return filp;
3816}
3817
3818struct file *do_file_open_root(const struct path *root,
3819 const char *name, const struct open_flags *op)
3820{
3821 struct nameidata nd;
3822 struct file *file;
3823 struct filename *filename;
3824 int flags = op->lookup_flags;
3825
3826 if (d_is_symlink(dentry: root->dentry) && op->intent & LOOKUP_OPEN)
3827 return ERR_PTR(error: -ELOOP);
3828
3829 filename = getname_kernel(name);
3830 if (IS_ERR(ptr: filename))
3831 return ERR_CAST(ptr: filename);
3832
3833 set_nameidata(p: &nd, dfd: -1, name: filename, root);
3834 file = path_openat(nd: &nd, op, flags: flags | LOOKUP_RCU);
3835 if (unlikely(file == ERR_PTR(-ECHILD)))
3836 file = path_openat(nd: &nd, op, flags);
3837 if (unlikely(file == ERR_PTR(-ESTALE)))
3838 file = path_openat(nd: &nd, op, flags: flags | LOOKUP_REVAL);
3839 restore_nameidata();
3840 putname(filename);
3841 return file;
3842}
3843
3844static struct dentry *filename_create(int dfd, struct filename *name,
3845 struct path *path, unsigned int lookup_flags)
3846{
3847 struct dentry *dentry = ERR_PTR(error: -EEXIST);
3848 struct qstr last;
3849 bool want_dir = lookup_flags & LOOKUP_DIRECTORY;
3850 unsigned int reval_flag = lookup_flags & LOOKUP_REVAL;
3851 unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL;
3852 int type;
3853 int err2;
3854 int error;
3855
3856 error = filename_parentat(dfd, name, flags: reval_flag, parent: path, last: &last, type: &type);
3857 if (error)
3858 return ERR_PTR(error);
3859
3860 /*
3861 * Yucky last component or no last component at all?
3862 * (foo/., foo/.., /////)
3863 */
3864 if (unlikely(type != LAST_NORM))
3865 goto out;
3866
3867 /* don't fail immediately if it's r/o, at least try to report other errors */
3868 err2 = mnt_want_write(mnt: path->mnt);
3869 /*
3870 * Do the final lookup. Suppress 'create' if there is a trailing
3871 * '/', and a directory wasn't requested.
3872 */
3873 if (last.name[last.len] && !want_dir)
3874 create_flags = 0;
3875 inode_lock_nested(inode: path->dentry->d_inode, subclass: I_MUTEX_PARENT);
3876 dentry = lookup_one_qstr_excl(&last, path->dentry,
3877 reval_flag | create_flags);
3878 if (IS_ERR(ptr: dentry))
3879 goto unlock;
3880
3881 error = -EEXIST;
3882 if (d_is_positive(dentry))
3883 goto fail;
3884
3885 /*
3886 * Special case - lookup gave negative, but... we had foo/bar/
3887 * From the vfs_mknod() POV we just have a negative dentry -
3888 * all is fine. Let's be bastards - you had / on the end, you've
3889 * been asking for (non-existent) directory. -ENOENT for you.
3890 */
3891 if (unlikely(!create_flags)) {
3892 error = -ENOENT;
3893 goto fail;
3894 }
3895 if (unlikely(err2)) {
3896 error = err2;
3897 goto fail;
3898 }
3899 return dentry;
3900fail:
3901 dput(dentry);
3902 dentry = ERR_PTR(error);
3903unlock:
3904 inode_unlock(inode: path->dentry->d_inode);
3905 if (!err2)
3906 mnt_drop_write(mnt: path->mnt);
3907out:
3908 path_put(path);
3909 return dentry;
3910}
3911
3912struct dentry *kern_path_create(int dfd, const char *pathname,
3913 struct path *path, unsigned int lookup_flags)
3914{
3915 struct filename *filename = getname_kernel(pathname);
3916 struct dentry *res = filename_create(dfd, name: filename, path, lookup_flags);
3917
3918 putname(filename);
3919 return res;
3920}
3921EXPORT_SYMBOL(kern_path_create);
3922
3923void done_path_create(struct path *path, struct dentry *dentry)
3924{
3925 dput(dentry);
3926 inode_unlock(inode: path->dentry->d_inode);
3927 mnt_drop_write(mnt: path->mnt);
3928 path_put(path);
3929}
3930EXPORT_SYMBOL(done_path_create);
3931
3932inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3933 struct path *path, unsigned int lookup_flags)
3934{
3935 struct filename *filename = getname(filename: pathname);
3936 struct dentry *res = filename_create(dfd, name: filename, path, lookup_flags);
3937
3938 putname(filename);
3939 return res;
3940}
3941EXPORT_SYMBOL(user_path_create);
3942
3943/**
3944 * vfs_mknod - create device node or file
3945 * @idmap: idmap of the mount the inode was found from
3946 * @dir: inode of @dentry
3947 * @dentry: pointer to dentry of the base directory
3948 * @mode: mode of the new device node or file
3949 * @dev: device number of device to create
3950 *
3951 * Create a device node or file.
3952 *
3953 * If the inode has been found through an idmapped mount the idmap of
3954 * the vfsmount must be passed through @idmap. This function will then take
3955 * care to map the inode according to @idmap before checking permissions.
3956 * On non-idmapped mounts or if permission checking is to be performed on the
3957 * raw inode simply passs @nop_mnt_idmap.
3958 */
3959int vfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
3960 struct dentry *dentry, umode_t mode, dev_t dev)
3961{
3962 bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV;
3963 int error = may_create(idmap, dir, child: dentry);
3964
3965 if (error)
3966 return error;
3967
3968 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout &&
3969 !capable(CAP_MKNOD))
3970 return -EPERM;
3971
3972 if (!dir->i_op->mknod)
3973 return -EPERM;
3974
3975 mode = vfs_prepare_mode(idmap, dir, mode, mask_perms: mode, type: mode);
3976 error = devcgroup_inode_mknod(mode, dev);
3977 if (error)
3978 return error;
3979
3980 error = security_inode_mknod(dir, dentry, mode, dev);
3981 if (error)
3982 return error;
3983
3984 error = dir->i_op->mknod(idmap, dir, dentry, mode, dev);
3985 if (!error)
3986 fsnotify_create(dir, dentry);
3987 return error;
3988}
3989EXPORT_SYMBOL(vfs_mknod);
3990
3991static int may_mknod(umode_t mode)
3992{
3993 switch (mode & S_IFMT) {
3994 case S_IFREG:
3995 case S_IFCHR:
3996 case S_IFBLK:
3997 case S_IFIFO:
3998 case S_IFSOCK:
3999 case 0: /* zero mode translates to S_IFREG */
4000 return 0;
4001 case S_IFDIR:
4002 return -EPERM;
4003 default:
4004 return -EINVAL;
4005 }
4006}
4007
4008static int do_mknodat(int dfd, struct filename *name, umode_t mode,
4009 unsigned int dev)
4010{
4011 struct mnt_idmap *idmap;
4012 struct dentry *dentry;
4013 struct path path;
4014 int error;
4015 unsigned int lookup_flags = 0;
4016
4017 error = may_mknod(mode);
4018 if (error)
4019 goto out1;
4020retry:
4021 dentry = filename_create(dfd, name, path: &path, lookup_flags);
4022 error = PTR_ERR(ptr: dentry);
4023 if (IS_ERR(ptr: dentry))
4024 goto out1;
4025
4026 error = security_path_mknod(dir: &path, dentry,
4027 mode: mode_strip_umask(dir: path.dentry->d_inode, mode), dev);
4028 if (error)
4029 goto out2;
4030
4031 idmap = mnt_idmap(mnt: path.mnt);
4032 switch (mode & S_IFMT) {
4033 case 0: case S_IFREG:
4034 error = vfs_create(idmap, path.dentry->d_inode,
4035 dentry, mode, true);
4036 if (!error)
4037 ima_post_path_mknod(idmap, dentry);
4038 break;
4039 case S_IFCHR: case S_IFBLK:
4040 error = vfs_mknod(idmap, path.dentry->d_inode,
4041 dentry, mode, new_decode_dev(dev));
4042 break;
4043 case S_IFIFO: case S_IFSOCK:
4044 error = vfs_mknod(idmap, path.dentry->d_inode,
4045 dentry, mode, 0);
4046 break;
4047 }
4048out2:
4049 done_path_create(&path, dentry);
4050 if (retry_estale(error, flags: lookup_flags)) {
4051 lookup_flags |= LOOKUP_REVAL;
4052 goto retry;
4053 }
4054out1:
4055 putname(name);
4056 return error;
4057}
4058
4059SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
4060 unsigned int, dev)
4061{
4062 return do_mknodat(dfd, name: getname(filename), mode, dev);
4063}
4064
4065SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
4066{
4067 return do_mknodat(AT_FDCWD, name: getname(filename), mode, dev);
4068}
4069
4070/**
4071 * vfs_mkdir - create directory
4072 * @idmap: idmap of the mount the inode was found from
4073 * @dir: inode of @dentry
4074 * @dentry: pointer to dentry of the base directory
4075 * @mode: mode of the new directory
4076 *
4077 * Create a directory.
4078 *
4079 * If the inode has been found through an idmapped mount the idmap of
4080 * the vfsmount must be passed through @idmap. This function will then take
4081 * care to map the inode according to @idmap before checking permissions.
4082 * On non-idmapped mounts or if permission checking is to be performed on the
4083 * raw inode simply passs @nop_mnt_idmap.
4084 */
4085int vfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
4086 struct dentry *dentry, umode_t mode)
4087{
4088 int error;
4089 unsigned max_links = dir->i_sb->s_max_links;
4090
4091 error = may_create(idmap, dir, child: dentry);
4092 if (error)
4093 return error;
4094
4095 if (!dir->i_op->mkdir)
4096 return -EPERM;
4097
4098 mode = vfs_prepare_mode(idmap, dir, mode, S_IRWXUGO | S_ISVTX, type: 0);
4099 error = security_inode_mkdir(dir, dentry, mode);
4100 if (error)
4101 return error;
4102
4103 if (max_links && dir->i_nlink >= max_links)
4104 return -EMLINK;
4105
4106 error = dir->i_op->mkdir(idmap, dir, dentry, mode);
4107 if (!error)
4108 fsnotify_mkdir(dir, dentry);
4109 return error;
4110}
4111EXPORT_SYMBOL(vfs_mkdir);
4112
4113int do_mkdirat(int dfd, struct filename *name, umode_t mode)
4114{
4115 struct dentry *dentry;
4116 struct path path;
4117 int error;
4118 unsigned int lookup_flags = LOOKUP_DIRECTORY;
4119
4120retry:
4121 dentry = filename_create(dfd, name, path: &path, lookup_flags);
4122 error = PTR_ERR(ptr: dentry);
4123 if (IS_ERR(ptr: dentry))
4124 goto out_putname;
4125
4126 error = security_path_mkdir(dir: &path, dentry,
4127 mode: mode_strip_umask(dir: path.dentry->d_inode, mode));
4128 if (!error) {
4129 error = vfs_mkdir(mnt_idmap(mnt: path.mnt), path.dentry->d_inode,
4130 dentry, mode);
4131 }
4132 done_path_create(&path, dentry);
4133 if (retry_estale(error, flags: lookup_flags)) {
4134 lookup_flags |= LOOKUP_REVAL;
4135 goto retry;
4136 }
4137out_putname:
4138 putname(name);
4139 return error;
4140}
4141
4142SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
4143{
4144 return do_mkdirat(dfd, name: getname(filename: pathname), mode);
4145}
4146
4147SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
4148{
4149 return do_mkdirat(AT_FDCWD, name: getname(filename: pathname), mode);
4150}
4151
4152/**
4153 * vfs_rmdir - remove directory
4154 * @idmap: idmap of the mount the inode was found from
4155 * @dir: inode of @dentry
4156 * @dentry: pointer to dentry of the base directory
4157 *
4158 * Remove a directory.
4159 *
4160 * If the inode has been found through an idmapped mount the idmap of
4161 * the vfsmount must be passed through @idmap. This function will then take
4162 * care to map the inode according to @idmap before checking permissions.
4163 * On non-idmapped mounts or if permission checking is to be performed on the
4164 * raw inode simply passs @nop_mnt_idmap.
4165 */
4166int vfs_rmdir(struct mnt_idmap *idmap, struct inode *dir,
4167 struct dentry *dentry)
4168{
4169 int error = may_delete(idmap, dir, victim: dentry, isdir: 1);
4170
4171 if (error)
4172 return error;
4173
4174 if (!dir->i_op->rmdir)
4175 return -EPERM;
4176
4177 dget(dentry);
4178 inode_lock(inode: dentry->d_inode);
4179
4180 error = -EBUSY;
4181 if (is_local_mountpoint(dentry) ||
4182 (dentry->d_inode->i_flags & S_KERNEL_FILE))
4183 goto out;
4184
4185 error = security_inode_rmdir(dir, dentry);
4186 if (error)
4187 goto out;
4188
4189 error = dir->i_op->rmdir(dir, dentry);
4190 if (error)
4191 goto out;
4192
4193 shrink_dcache_parent(dentry);
4194 dentry->d_inode->i_flags |= S_DEAD;
4195 dont_mount(dentry);
4196 detach_mounts(dentry);
4197
4198out:
4199 inode_unlock(inode: dentry->d_inode);
4200 dput(dentry);
4201 if (!error)
4202 d_delete_notify(dir, dentry);
4203 return error;
4204}
4205EXPORT_SYMBOL(vfs_rmdir);
4206
4207int do_rmdir(int dfd, struct filename *name)
4208{
4209 int error;
4210 struct dentry *dentry;
4211 struct path path;
4212 struct qstr last;
4213 int type;
4214 unsigned int lookup_flags = 0;
4215retry:
4216 error = filename_parentat(dfd, name, flags: lookup_flags, parent: &path, last: &last, type: &type);
4217 if (error)
4218 goto exit1;
4219
4220 switch (type) {
4221 case LAST_DOTDOT:
4222 error = -ENOTEMPTY;
4223 goto exit2;
4224 case LAST_DOT:
4225 error = -EINVAL;
4226 goto exit2;
4227 case LAST_ROOT:
4228 error = -EBUSY;
4229 goto exit2;
4230 }
4231
4232 error = mnt_want_write(mnt: path.mnt);
4233 if (error)
4234 goto exit2;
4235
4236 inode_lock_nested(inode: path.dentry->d_inode, subclass: I_MUTEX_PARENT);
4237 dentry = lookup_one_qstr_excl(&last, path.dentry, lookup_flags);
4238 error = PTR_ERR(ptr: dentry);
4239 if (IS_ERR(ptr: dentry))
4240 goto exit3;
4241 if (!dentry->d_inode) {
4242 error = -ENOENT;
4243 goto exit4;
4244 }
4245 error = security_path_rmdir(dir: &path, dentry);
4246 if (error)
4247 goto exit4;
4248 error = vfs_rmdir(mnt_idmap(mnt: path.mnt), path.dentry->d_inode, dentry);
4249exit4:
4250 dput(dentry);
4251exit3:
4252 inode_unlock(inode: path.dentry->d_inode);
4253 mnt_drop_write(mnt: path.mnt);
4254exit2:
4255 path_put(&path);
4256 if (retry_estale(error, flags: lookup_flags)) {
4257 lookup_flags |= LOOKUP_REVAL;
4258 goto retry;
4259 }
4260exit1:
4261 putname(name);
4262 return error;
4263}
4264
4265SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
4266{
4267 return do_rmdir(AT_FDCWD, name: getname(filename: pathname));
4268}
4269
4270/**
4271 * vfs_unlink - unlink a filesystem object
4272 * @idmap: idmap of the mount the inode was found from
4273 * @dir: parent directory
4274 * @dentry: victim
4275 * @delegated_inode: returns victim inode, if the inode is delegated.
4276 *
4277 * The caller must hold dir->i_mutex.
4278 *
4279 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
4280 * return a reference to the inode in delegated_inode. The caller
4281 * should then break the delegation on that inode and retry. Because
4282 * breaking a delegation may take a long time, the caller should drop
4283 * dir->i_mutex before doing so.
4284 *
4285 * Alternatively, a caller may pass NULL for delegated_inode. This may
4286 * be appropriate for callers that expect the underlying filesystem not
4287 * to be NFS exported.
4288 *
4289 * If the inode has been found through an idmapped mount the idmap of
4290 * the vfsmount must be passed through @idmap. This function will then take
4291 * care to map the inode according to @idmap before checking permissions.
4292 * On non-idmapped mounts or if permission checking is to be performed on the
4293 * raw inode simply passs @nop_mnt_idmap.
4294 */
4295int vfs_unlink(struct mnt_idmap *idmap, struct inode *dir,
4296 struct dentry *dentry, struct inode **delegated_inode)
4297{
4298 struct inode *target = dentry->d_inode;
4299 int error = may_delete(idmap, dir, victim: dentry, isdir: 0);
4300
4301 if (error)
4302 return error;
4303
4304 if (!dir->i_op->unlink)
4305 return -EPERM;
4306
4307 inode_lock(inode: target);
4308 if (IS_SWAPFILE(target))
4309 error = -EPERM;
4310 else if (is_local_mountpoint(dentry))
4311 error = -EBUSY;
4312 else {
4313 error = security_inode_unlink(dir, dentry);
4314 if (!error) {
4315 error = try_break_deleg(inode: target, delegated_inode);
4316 if (error)
4317 goto out;
4318 error = dir->i_op->unlink(dir, dentry);
4319 if (!error) {
4320 dont_mount(dentry);
4321 detach_mounts(dentry);
4322 }
4323 }
4324 }
4325out:
4326 inode_unlock(inode: target);
4327
4328 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4329 if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) {
4330 fsnotify_unlink(dir, dentry);
4331 } else if (!error) {
4332 fsnotify_link_count(inode: target);
4333 d_delete_notify(dir, dentry);
4334 }
4335
4336 return error;
4337}
4338EXPORT_SYMBOL(vfs_unlink);
4339
4340/*
4341 * Make sure that the actual truncation of the file will occur outside its
4342 * directory's i_mutex. Truncate can take a long time if there is a lot of
4343 * writeout happening, and we don't want to prevent access to the directory
4344 * while waiting on the I/O.
4345 */
4346int do_unlinkat(int dfd, struct filename *name)
4347{
4348 int error;
4349 struct dentry *dentry;
4350 struct path path;
4351 struct qstr last;
4352 int type;
4353 struct inode *inode = NULL;
4354 struct inode *delegated_inode = NULL;
4355 unsigned int lookup_flags = 0;
4356retry:
4357 error = filename_parentat(dfd, name, flags: lookup_flags, parent: &path, last: &last, type: &type);
4358 if (error)
4359 goto exit1;
4360
4361 error = -EISDIR;
4362 if (type != LAST_NORM)
4363 goto exit2;
4364
4365 error = mnt_want_write(mnt: path.mnt);
4366 if (error)
4367 goto exit2;
4368retry_deleg:
4369 inode_lock_nested(inode: path.dentry->d_inode, subclass: I_MUTEX_PARENT);
4370 dentry = lookup_one_qstr_excl(&last, path.dentry, lookup_flags);
4371 error = PTR_ERR(ptr: dentry);
4372 if (!IS_ERR(ptr: dentry)) {
4373
4374 /* Why not before? Because we want correct error value */
4375 if (last.name[last.len] || d_is_negative(dentry))
4376 goto slashes;
4377 inode = dentry->d_inode;
4378 ihold(inode);
4379 error = security_path_unlink(dir: &path, dentry);
4380 if (error)
4381 goto exit3;
4382 error = vfs_unlink(mnt_idmap(mnt: path.mnt), path.dentry->d_inode,
4383 dentry, &delegated_inode);
4384exit3:
4385 dput(dentry);
4386 }
4387 inode_unlock(inode: path.dentry->d_inode);
4388 if (inode)
4389 iput(inode); /* truncate the inode here */
4390 inode = NULL;
4391 if (delegated_inode) {
4392 error = break_deleg_wait(delegated_inode: &delegated_inode);
4393 if (!error)
4394 goto retry_deleg;
4395 }
4396 mnt_drop_write(mnt: path.mnt);
4397exit2:
4398 path_put(&path);
4399 if (retry_estale(error, flags: lookup_flags)) {
4400 lookup_flags |= LOOKUP_REVAL;
4401 inode = NULL;
4402 goto retry;
4403 }
4404exit1:
4405 putname(name);
4406 return error;
4407
4408slashes:
4409 if (d_is_negative(dentry))
4410 error = -ENOENT;
4411 else if (d_is_dir(dentry))
4412 error = -EISDIR;
4413 else
4414 error = -ENOTDIR;
4415 goto exit3;
4416}
4417
4418SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4419{
4420 if ((flag & ~AT_REMOVEDIR) != 0)
4421 return -EINVAL;
4422
4423 if (flag & AT_REMOVEDIR)
4424 return do_rmdir(dfd, name: getname(filename: pathname));
4425 return do_unlinkat(dfd, name: getname(filename: pathname));
4426}
4427
4428SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4429{
4430 return do_unlinkat(AT_FDCWD, name: getname(filename: pathname));
4431}
4432
4433/**
4434 * vfs_symlink - create symlink
4435 * @idmap: idmap of the mount the inode was found from
4436 * @dir: inode of @dentry
4437 * @dentry: pointer to dentry of the base directory
4438 * @oldname: name of the file to link to
4439 *
4440 * Create a symlink.
4441 *
4442 * If the inode has been found through an idmapped mount the idmap of
4443 * the vfsmount must be passed through @idmap. This function will then take
4444 * care to map the inode according to @idmap before checking permissions.
4445 * On non-idmapped mounts or if permission checking is to be performed on the
4446 * raw inode simply passs @nop_mnt_idmap.
4447 */
4448int vfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
4449 struct dentry *dentry, const char *oldname)
4450{
4451 int error;
4452
4453 error = may_create(idmap, dir, child: dentry);
4454 if (error)
4455 return error;
4456
4457 if (!dir->i_op->symlink)
4458 return -EPERM;
4459
4460 error = security_inode_symlink(dir, dentry, old_name: oldname);
4461 if (error)
4462 return error;
4463
4464 error = dir->i_op->symlink(idmap, dir, dentry, oldname);
4465 if (!error)
4466 fsnotify_create(dir, dentry);
4467 return error;
4468}
4469EXPORT_SYMBOL(vfs_symlink);
4470
4471int do_symlinkat(struct filename *from, int newdfd, struct filename *to)
4472{
4473 int error;
4474 struct dentry *dentry;
4475 struct path path;
4476 unsigned int lookup_flags = 0;
4477
4478 if (IS_ERR(ptr: from)) {
4479 error = PTR_ERR(ptr: from);
4480 goto out_putnames;
4481 }
4482retry:
4483 dentry = filename_create(dfd: newdfd, name: to, path: &path, lookup_flags);
4484 error = PTR_ERR(ptr: dentry);
4485 if (IS_ERR(ptr: dentry))
4486 goto out_putnames;
4487
4488 error = security_path_symlink(dir: &path, dentry, old_name: from->name);
4489 if (!error)
4490 error = vfs_symlink(mnt_idmap(mnt: path.mnt), path.dentry->d_inode,
4491 dentry, from->name);
4492 done_path_create(&path, dentry);
4493 if (retry_estale(error, flags: lookup_flags)) {
4494 lookup_flags |= LOOKUP_REVAL;
4495 goto retry;
4496 }
4497out_putnames:
4498 putname(to);
4499 putname(from);
4500 return error;
4501}
4502
4503SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4504 int, newdfd, const char __user *, newname)
4505{
4506 return do_symlinkat(from: getname(filename: oldname), newdfd, to: getname(filename: newname));
4507}
4508
4509SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4510{
4511 return do_symlinkat(from: getname(filename: oldname), AT_FDCWD, to: getname(filename: newname));
4512}
4513
4514/**
4515 * vfs_link - create a new link
4516 * @old_dentry: object to be linked
4517 * @idmap: idmap of the mount
4518 * @dir: new parent
4519 * @new_dentry: where to create the new link
4520 * @delegated_inode: returns inode needing a delegation break
4521 *
4522 * The caller must hold dir->i_mutex
4523 *
4524 * If vfs_link discovers a delegation on the to-be-linked file in need
4525 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4526 * inode in delegated_inode. The caller should then break the delegation
4527 * and retry. Because breaking a delegation may take a long time, the
4528 * caller should drop the i_mutex before doing so.
4529 *
4530 * Alternatively, a caller may pass NULL for delegated_inode. This may
4531 * be appropriate for callers that expect the underlying filesystem not
4532 * to be NFS exported.
4533 *
4534 * If the inode has been found through an idmapped mount the idmap of
4535 * the vfsmount must be passed through @idmap. This function will then take
4536 * care to map the inode according to @idmap before checking permissions.
4537 * On non-idmapped mounts or if permission checking is to be performed on the
4538 * raw inode simply passs @nop_mnt_idmap.
4539 */
4540int vfs_link(struct dentry *old_dentry, struct mnt_idmap *idmap,
4541 struct inode *dir, struct dentry *new_dentry,
4542 struct inode **delegated_inode)
4543{
4544 struct inode *inode = old_dentry->d_inode;
4545 unsigned max_links = dir->i_sb->s_max_links;
4546 int error;
4547
4548 if (!inode)
4549 return -ENOENT;
4550
4551 error = may_create(idmap, dir, child: new_dentry);
4552 if (error)
4553 return error;
4554
4555 if (dir->i_sb != inode->i_sb)
4556 return -EXDEV;
4557
4558 /*
4559 * A link to an append-only or immutable file cannot be created.
4560 */
4561 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4562 return -EPERM;
4563 /*
4564 * Updating the link count will likely cause i_uid and i_gid to
4565 * be writen back improperly if their true value is unknown to
4566 * the vfs.
4567 */
4568 if (HAS_UNMAPPED_ID(idmap, inode))
4569 return -EPERM;
4570 if (!dir->i_op->link)
4571 return -EPERM;
4572 if (S_ISDIR(inode->i_mode))
4573 return -EPERM;
4574
4575 error = security_inode_link(old_dentry, dir, new_dentry);
4576 if (error)
4577 return error;
4578
4579 inode_lock(inode);
4580 /* Make sure we don't allow creating hardlink to an unlinked file */
4581 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4582 error = -ENOENT;
4583 else if (max_links && inode->i_nlink >= max_links)
4584 error = -EMLINK;
4585 else {
4586 error = try_break_deleg(inode, delegated_inode);
4587 if (!error)
4588 error = dir->i_op->link(old_dentry, dir, new_dentry);
4589 }
4590
4591 if (!error && (inode->i_state & I_LINKABLE)) {
4592 spin_lock(lock: &inode->i_lock);
4593 inode->i_state &= ~I_LINKABLE;
4594 spin_unlock(lock: &inode->i_lock);
4595 }
4596 inode_unlock(inode);
4597 if (!error)
4598 fsnotify_link(dir, inode, new_dentry);
4599 return error;
4600}
4601EXPORT_SYMBOL(vfs_link);
4602
4603/*
4604 * Hardlinks are often used in delicate situations. We avoid
4605 * security-related surprises by not following symlinks on the
4606 * newname. --KAB
4607 *
4608 * We don't follow them on the oldname either to be compatible
4609 * with linux 2.0, and to avoid hard-linking to directories
4610 * and other special files. --ADM
4611 */
4612int do_linkat(int olddfd, struct filename *old, int newdfd,
4613 struct filename *new, int flags)
4614{
4615 struct mnt_idmap *idmap;
4616 struct dentry *new_dentry;
4617 struct path old_path, new_path;
4618 struct inode *delegated_inode = NULL;
4619 int how = 0;
4620 int error;
4621
4622 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) {
4623 error = -EINVAL;
4624 goto out_putnames;
4625 }
4626 /*
4627 * To use null names we require CAP_DAC_READ_SEARCH
4628 * This ensures that not everyone will be able to create
4629 * handlink using the passed filedescriptor.
4630 */
4631 if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) {
4632 error = -ENOENT;
4633 goto out_putnames;
4634 }
4635
4636 if (flags & AT_SYMLINK_FOLLOW)
4637 how |= LOOKUP_FOLLOW;
4638retry:
4639 error = filename_lookup(dfd: olddfd, name: old, flags: how, path: &old_path, NULL);
4640 if (error)
4641 goto out_putnames;
4642
4643 new_dentry = filename_create(dfd: newdfd, name: new, path: &new_path,
4644 lookup_flags: (how & LOOKUP_REVAL));
4645 error = PTR_ERR(ptr: new_dentry);
4646 if (IS_ERR(ptr: new_dentry))
4647 goto out_putpath;
4648
4649 error = -EXDEV;
4650 if (old_path.mnt != new_path.mnt)
4651 goto out_dput;
4652 idmap = mnt_idmap(mnt: new_path.mnt);
4653 error = may_linkat(idmap, link: &old_path);
4654 if (unlikely(error))
4655 goto out_dput;
4656 error = security_path_link(old_dentry: old_path.dentry, new_dir: &new_path, new_dentry);
4657 if (error)
4658 goto out_dput;
4659 error = vfs_link(old_path.dentry, idmap, new_path.dentry->d_inode,
4660 new_dentry, &delegated_inode);
4661out_dput:
4662 done_path_create(&new_path, new_dentry);
4663 if (delegated_inode) {
4664 error = break_deleg_wait(delegated_inode: &delegated_inode);
4665 if (!error) {
4666 path_put(&old_path);
4667 goto retry;
4668 }
4669 }
4670 if (retry_estale(error, flags: how)) {
4671 path_put(&old_path);
4672 how |= LOOKUP_REVAL;
4673 goto retry;
4674 }
4675out_putpath:
4676 path_put(&old_path);
4677out_putnames:
4678 putname(old);
4679 putname(new);
4680
4681 return error;
4682}
4683
4684SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4685 int, newdfd, const char __user *, newname, int, flags)
4686{
4687 return do_linkat(olddfd, old: getname_uflags(filename: oldname, uflags: flags),
4688 newdfd, new: getname(filename: newname), flags);
4689}
4690
4691SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4692{
4693 return do_linkat(AT_FDCWD, old: getname(filename: oldname), AT_FDCWD, new: getname(filename: newname), flags: 0);
4694}
4695
4696/**
4697 * vfs_rename - rename a filesystem object
4698 * @rd: pointer to &struct renamedata info
4699 *
4700 * The caller must hold multiple mutexes--see lock_rename()).
4701 *
4702 * If vfs_rename discovers a delegation in need of breaking at either
4703 * the source or destination, it will return -EWOULDBLOCK and return a
4704 * reference to the inode in delegated_inode. The caller should then
4705 * break the delegation and retry. Because breaking a delegation may
4706 * take a long time, the caller should drop all locks before doing
4707 * so.
4708 *
4709 * Alternatively, a caller may pass NULL for delegated_inode. This may
4710 * be appropriate for callers that expect the underlying filesystem not
4711 * to be NFS exported.
4712 *
4713 * The worst of all namespace operations - renaming directory. "Perverted"
4714 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4715 * Problems:
4716 *
4717 * a) we can get into loop creation.
4718 * b) race potential - two innocent renames can create a loop together.
4719 * That's where 4.4 screws up. Current fix: serialization on
4720 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4721 * story.
4722 * c) we have to lock _four_ objects - parents and victim (if it exists),
4723 * and source.
4724 * And that - after we got ->i_mutex on parents (until then we don't know
4725 * whether the target exists). Solution: try to be smart with locking
4726 * order for inodes. We rely on the fact that tree topology may change
4727 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4728 * move will be locked. Thus we can rank directories by the tree
4729 * (ancestors first) and rank all non-directories after them.
4730 * That works since everybody except rename does "lock parent, lookup,
4731 * lock child" and rename is under ->s_vfs_rename_mutex.
4732 * HOWEVER, it relies on the assumption that any object with ->lookup()
4733 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4734 * we'd better make sure that there's no link(2) for them.
4735 * d) conversion from fhandle to dentry may come in the wrong moment - when
4736 * we are removing the target. Solution: we will have to grab ->i_mutex
4737 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4738 * ->i_mutex on parents, which works but leads to some truly excessive
4739 * locking].
4740 */
4741int vfs_rename(struct renamedata *rd)
4742{
4743 int error;
4744 struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir;
4745 struct dentry *old_dentry = rd->old_dentry;
4746 struct dentry *new_dentry = rd->new_dentry;
4747 struct inode **delegated_inode = rd->delegated_inode;
4748 unsigned int flags = rd->flags;
4749 bool is_dir = d_is_dir(dentry: old_dentry);
4750 struct inode *source = old_dentry->d_inode;
4751 struct inode *target = new_dentry->d_inode;
4752 bool new_is_dir = false;
4753 unsigned max_links = new_dir->i_sb->s_max_links;
4754 struct name_snapshot old_name;
4755
4756 if (source == target)
4757 return 0;
4758
4759 error = may_delete(idmap: rd->old_mnt_idmap, dir: old_dir, victim: old_dentry, isdir: is_dir);
4760 if (error)
4761 return error;
4762
4763 if (!target) {
4764 error = may_create(idmap: rd->new_mnt_idmap, dir: new_dir, child: new_dentry);
4765 } else {
4766 new_is_dir = d_is_dir(dentry: new_dentry);
4767
4768 if (!(flags & RENAME_EXCHANGE))
4769 error = may_delete(idmap: rd->new_mnt_idmap, dir: new_dir,
4770 victim: new_dentry, isdir: is_dir);
4771 else
4772 error = may_delete(idmap: rd->new_mnt_idmap, dir: new_dir,
4773 victim: new_dentry, isdir: new_is_dir);
4774 }
4775 if (error)
4776 return error;
4777
4778 if (!old_dir->i_op->rename)
4779 return -EPERM;
4780
4781 /*
4782 * If we are going to change the parent - check write permissions,
4783 * we'll need to flip '..'.
4784 */
4785 if (new_dir != old_dir) {
4786 if (is_dir) {
4787 error = inode_permission(rd->old_mnt_idmap, source,
4788 MAY_WRITE);
4789 if (error)
4790 return error;
4791 }
4792 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4793 error = inode_permission(rd->new_mnt_idmap, target,
4794 MAY_WRITE);
4795 if (error)
4796 return error;
4797 }
4798 }
4799
4800 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4801 flags);
4802 if (error)
4803 return error;
4804
4805 take_dentry_name_snapshot(&old_name, old_dentry);
4806 dget(dentry: new_dentry);
4807 /*
4808 * Lock all moved children. Moved directories may need to change parent
4809 * pointer so they need the lock to prevent against concurrent
4810 * directory changes moving parent pointer. For regular files we've
4811 * historically always done this. The lockdep locking subclasses are
4812 * somewhat arbitrary but RENAME_EXCHANGE in particular can swap
4813 * regular files and directories so it's difficult to tell which
4814 * subclasses to use.
4815 */
4816 lock_two_inodes(inode1: source, inode2: target, subclass1: I_MUTEX_NORMAL, subclass2: I_MUTEX_NONDIR2);
4817
4818 error = -EPERM;
4819 if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target)))
4820 goto out;
4821
4822 error = -EBUSY;
4823 if (is_local_mountpoint(dentry: old_dentry) || is_local_mountpoint(dentry: new_dentry))
4824 goto out;
4825
4826 if (max_links && new_dir != old_dir) {
4827 error = -EMLINK;
4828 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4829 goto out;
4830 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4831 old_dir->i_nlink >= max_links)
4832 goto out;
4833 }
4834 if (!is_dir) {
4835 error = try_break_deleg(inode: source, delegated_inode);
4836 if (error)
4837 goto out;
4838 }
4839 if (target && !new_is_dir) {
4840 error = try_break_deleg(inode: target, delegated_inode);
4841 if (error)
4842 goto out;
4843 }
4844 error = old_dir->i_op->rename(rd->new_mnt_idmap, old_dir, old_dentry,
4845 new_dir, new_dentry, flags);
4846 if (error)
4847 goto out;
4848
4849 if (!(flags & RENAME_EXCHANGE) && target) {
4850 if (is_dir) {
4851 shrink_dcache_parent(new_dentry);
4852 target->i_flags |= S_DEAD;
4853 }
4854 dont_mount(dentry: new_dentry);
4855 detach_mounts(dentry: new_dentry);
4856 }
4857 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4858 if (!(flags & RENAME_EXCHANGE))
4859 d_move(old_dentry, new_dentry);
4860 else
4861 d_exchange(old_dentry, new_dentry);
4862 }
4863out:
4864 inode_unlock(inode: source);
4865 if (target)
4866 inode_unlock(inode: target);
4867 dput(new_dentry);
4868 if (!error) {
4869 fsnotify_move(old_dir, new_dir, old_name: &old_name.name, isdir: is_dir,
4870 target: !(flags & RENAME_EXCHANGE) ? target : NULL, moved: old_dentry);
4871 if (flags & RENAME_EXCHANGE) {
4872 fsnotify_move(old_dir: new_dir, new_dir: old_dir, old_name: &old_dentry->d_name,
4873 isdir: new_is_dir, NULL, moved: new_dentry);
4874 }
4875 }
4876 release_dentry_name_snapshot(&old_name);
4877
4878 return error;
4879}
4880EXPORT_SYMBOL(vfs_rename);
4881
4882int do_renameat2(int olddfd, struct filename *from, int newdfd,
4883 struct filename *to, unsigned int flags)
4884{
4885 struct renamedata rd;
4886 struct dentry *old_dentry, *new_dentry;
4887 struct dentry *trap;
4888 struct path old_path, new_path;
4889 struct qstr old_last, new_last;
4890 int old_type, new_type;
4891 struct inode *delegated_inode = NULL;
4892 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4893 bool should_retry = false;
4894 int error = -EINVAL;
4895
4896 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4897 goto put_names;
4898
4899 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4900 (flags & RENAME_EXCHANGE))
4901 goto put_names;
4902
4903 if (flags & RENAME_EXCHANGE)
4904 target_flags = 0;
4905
4906retry:
4907 error = filename_parentat(dfd: olddfd, name: from, flags: lookup_flags, parent: &old_path,
4908 last: &old_last, type: &old_type);
4909 if (error)
4910 goto put_names;
4911
4912 error = filename_parentat(dfd: newdfd, name: to, flags: lookup_flags, parent: &new_path, last: &new_last,
4913 type: &new_type);
4914 if (error)
4915 goto exit1;
4916
4917 error = -EXDEV;
4918 if (old_path.mnt != new_path.mnt)
4919 goto exit2;
4920
4921 error = -EBUSY;
4922 if (old_type != LAST_NORM)
4923 goto exit2;
4924
4925 if (flags & RENAME_NOREPLACE)
4926 error = -EEXIST;
4927 if (new_type != LAST_NORM)
4928 goto exit2;
4929
4930 error = mnt_want_write(mnt: old_path.mnt);
4931 if (error)
4932 goto exit2;
4933
4934retry_deleg:
4935 trap = lock_rename(new_path.dentry, old_path.dentry);
4936
4937 old_dentry = lookup_one_qstr_excl(&old_last, old_path.dentry,
4938 lookup_flags);
4939 error = PTR_ERR(ptr: old_dentry);
4940 if (IS_ERR(ptr: old_dentry))
4941 goto exit3;
4942 /* source must exist */
4943 error = -ENOENT;
4944 if (d_is_negative(dentry: old_dentry))
4945 goto exit4;
4946 new_dentry = lookup_one_qstr_excl(&new_last, new_path.dentry,
4947 lookup_flags | target_flags);
4948 error = PTR_ERR(ptr: new_dentry);
4949 if (IS_ERR(ptr: new_dentry))
4950 goto exit4;
4951 error = -EEXIST;
4952 if ((flags & RENAME_NOREPLACE) && d_is_positive(dentry: new_dentry))
4953 goto exit5;
4954 if (flags & RENAME_EXCHANGE) {
4955 error = -ENOENT;
4956 if (d_is_negative(dentry: new_dentry))
4957 goto exit5;
4958
4959 if (!d_is_dir(dentry: new_dentry)) {
4960 error = -ENOTDIR;
4961 if (new_last.name[new_last.len])
4962 goto exit5;
4963 }
4964 }
4965 /* unless the source is a directory trailing slashes give -ENOTDIR */
4966 if (!d_is_dir(dentry: old_dentry)) {
4967 error = -ENOTDIR;
4968 if (old_last.name[old_last.len])
4969 goto exit5;
4970 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4971 goto exit5;
4972 }
4973 /* source should not be ancestor of target */
4974 error = -EINVAL;
4975 if (old_dentry == trap)
4976 goto exit5;
4977 /* target should not be an ancestor of source */
4978 if (!(flags & RENAME_EXCHANGE))
4979 error = -ENOTEMPTY;
4980 if (new_dentry == trap)
4981 goto exit5;
4982
4983 error = security_path_rename(old_dir: &old_path, old_dentry,
4984 new_dir: &new_path, new_dentry, flags);
4985 if (error)
4986 goto exit5;
4987
4988 rd.old_dir = old_path.dentry->d_inode;
4989 rd.old_dentry = old_dentry;
4990 rd.old_mnt_idmap = mnt_idmap(mnt: old_path.mnt);
4991 rd.new_dir = new_path.dentry->d_inode;
4992 rd.new_dentry = new_dentry;
4993 rd.new_mnt_idmap = mnt_idmap(mnt: new_path.mnt);
4994 rd.delegated_inode = &delegated_inode;
4995 rd.flags = flags;
4996 error = vfs_rename(&rd);
4997exit5:
4998 dput(new_dentry);
4999exit4:
5000 dput(old_dentry);
5001exit3:
5002 unlock_rename(new_path.dentry, old_path.dentry);
5003 if (delegated_inode) {
5004 error = break_deleg_wait(delegated_inode: &delegated_inode);
5005 if (!error)
5006 goto retry_deleg;
5007 }
5008 mnt_drop_write(mnt: old_path.mnt);
5009exit2:
5010 if (retry_estale(error, flags: lookup_flags))
5011 should_retry = true;
5012 path_put(&new_path);
5013exit1:
5014 path_put(&old_path);
5015 if (should_retry) {
5016 should_retry = false;
5017 lookup_flags |= LOOKUP_REVAL;
5018 goto retry;
5019 }
5020put_names:
5021 putname(from);
5022 putname(to);
5023 return error;
5024}
5025
5026SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
5027 int, newdfd, const char __user *, newname, unsigned int, flags)
5028{
5029 return do_renameat2(olddfd, from: getname(filename: oldname), newdfd, to: getname(filename: newname),
5030 flags);
5031}
5032
5033SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
5034 int, newdfd, const char __user *, newname)
5035{
5036 return do_renameat2(olddfd, from: getname(filename: oldname), newdfd, to: getname(filename: newname),
5037 flags: 0);
5038}
5039
5040SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5041{
5042 return do_renameat2(AT_FDCWD, from: getname(filename: oldname), AT_FDCWD,
5043 to: getname(filename: newname), flags: 0);
5044}
5045
5046int readlink_copy(char __user *buffer, int buflen, const char *link)
5047{
5048 int len = PTR_ERR(ptr: link);
5049 if (IS_ERR(ptr: link))
5050 goto out;
5051
5052 len = strlen(link);
5053 if (len > (unsigned) buflen)
5054 len = buflen;
5055 if (copy_to_user(to: buffer, from: link, n: len))
5056 len = -EFAULT;
5057out:
5058 return len;
5059}
5060
5061/**
5062 * vfs_readlink - copy symlink body into userspace buffer
5063 * @dentry: dentry on which to get symbolic link
5064 * @buffer: user memory pointer
5065 * @buflen: size of buffer
5066 *
5067 * Does not touch atime. That's up to the caller if necessary
5068 *
5069 * Does not call security hook.
5070 */
5071int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5072{
5073 struct inode *inode = d_inode(dentry);
5074 DEFINE_DELAYED_CALL(done);
5075 const char *link;
5076 int res;
5077
5078 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
5079 if (unlikely(inode->i_op->readlink))
5080 return inode->i_op->readlink(dentry, buffer, buflen);
5081
5082 if (!d_is_symlink(dentry))
5083 return -EINVAL;
5084
5085 spin_lock(lock: &inode->i_lock);
5086 inode->i_opflags |= IOP_DEFAULT_READLINK;
5087 spin_unlock(lock: &inode->i_lock);
5088 }
5089
5090 link = READ_ONCE(inode->i_link);
5091 if (!link) {
5092 link = inode->i_op->get_link(dentry, inode, &done);
5093 if (IS_ERR(ptr: link))
5094 return PTR_ERR(ptr: link);
5095 }
5096 res = readlink_copy(buffer, buflen, link);
5097 do_delayed_call(call: &done);
5098 return res;
5099}
5100EXPORT_SYMBOL(vfs_readlink);
5101
5102/**
5103 * vfs_get_link - get symlink body
5104 * @dentry: dentry on which to get symbolic link
5105 * @done: caller needs to free returned data with this
5106 *
5107 * Calls security hook and i_op->get_link() on the supplied inode.
5108 *
5109 * It does not touch atime. That's up to the caller if necessary.
5110 *
5111 * Does not work on "special" symlinks like /proc/$$/fd/N
5112 */
5113const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
5114{
5115 const char *res = ERR_PTR(error: -EINVAL);
5116 struct inode *inode = d_inode(dentry);
5117
5118 if (d_is_symlink(dentry)) {
5119 res = ERR_PTR(error: security_inode_readlink(dentry));
5120 if (!res)
5121 res = inode->i_op->get_link(dentry, inode, done);
5122 }
5123 return res;
5124}
5125EXPORT_SYMBOL(vfs_get_link);
5126
5127/* get the link contents into pagecache */
5128const char *page_get_link(struct dentry *dentry, struct inode *inode,
5129 struct delayed_call *callback)
5130{
5131 char *kaddr;
5132 struct page *page;
5133 struct address_space *mapping = inode->i_mapping;
5134
5135 if (!dentry) {
5136 page = find_get_page(mapping, offset: 0);
5137 if (!page)
5138 return ERR_PTR(error: -ECHILD);
5139 if (!PageUptodate(page)) {
5140 put_page(page);
5141 return ERR_PTR(error: -ECHILD);
5142 }
5143 } else {
5144 page = read_mapping_page(mapping, index: 0, NULL);
5145 if (IS_ERR(ptr: page))
5146 return (char*)page;
5147 }
5148 set_delayed_call(call: callback, fn: page_put_link, arg: page);
5149 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
5150 kaddr = page_address(page);
5151 nd_terminate_link(name: kaddr, len: inode->i_size, PAGE_SIZE - 1);
5152 return kaddr;
5153}
5154
5155EXPORT_SYMBOL(page_get_link);
5156
5157void page_put_link(void *arg)
5158{
5159 put_page(page: arg);
5160}
5161EXPORT_SYMBOL(page_put_link);
5162
5163int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5164{
5165 DEFINE_DELAYED_CALL(done);
5166 int res = readlink_copy(buffer, buflen,
5167 link: page_get_link(dentry, d_inode(dentry),
5168 &done));
5169 do_delayed_call(call: &done);
5170 return res;
5171}
5172EXPORT_SYMBOL(page_readlink);
5173
5174int page_symlink(struct inode *inode, const char *symname, int len)
5175{
5176 struct address_space *mapping = inode->i_mapping;
5177 const struct address_space_operations *aops = mapping->a_ops;
5178 bool nofs = !mapping_gfp_constraint(mapping, __GFP_FS);
5179 struct page *page;
5180 void *fsdata = NULL;
5181 int err;
5182 unsigned int flags;
5183
5184retry:
5185 if (nofs)
5186 flags = memalloc_nofs_save();
5187 err = aops->write_begin(NULL, mapping, 0, len-1, &page, &fsdata);
5188 if (nofs)
5189 memalloc_nofs_restore(flags);
5190 if (err)
5191 goto fail;
5192
5193 memcpy(page_address(page), symname, len-1);
5194
5195 err = aops->write_end(NULL, mapping, 0, len-1, len-1,
5196 page, fsdata);
5197 if (err < 0)
5198 goto fail;
5199 if (err < len-1)
5200 goto retry;
5201
5202 mark_inode_dirty(inode);
5203 return 0;
5204fail:
5205 return err;
5206}
5207EXPORT_SYMBOL(page_symlink);
5208
5209const struct inode_operations page_symlink_inode_operations = {
5210 .get_link = page_get_link,
5211};
5212EXPORT_SYMBOL(page_symlink_inode_operations);
5213

source code of linux/fs/namei.c