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