Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0
2 : /*
3 : * linux/fs/namei.c
4 : *
5 : * Copyright (C) 1991, 1992 Linus Torvalds
6 : */
7 :
8 : /*
9 : * Some corrections by tytso.
10 : */
11 :
12 : /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
13 : * lookup logic.
14 : */
15 : /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
16 : */
17 :
18 : #include <linux/init.h>
19 : #include <linux/export.h>
20 : #include <linux/kernel.h>
21 : #include <linux/slab.h>
22 : #include <linux/fs.h>
23 : #include <linux/namei.h>
24 : #include <linux/pagemap.h>
25 : #include <linux/fsnotify.h>
26 : #include <linux/personality.h>
27 : #include <linux/security.h>
28 : #include <linux/ima.h>
29 : #include <linux/syscalls.h>
30 : #include <linux/mount.h>
31 : #include <linux/audit.h>
32 : #include <linux/capability.h>
33 : #include <linux/file.h>
34 : #include <linux/fcntl.h>
35 : #include <linux/device_cgroup.h>
36 : #include <linux/fs_struct.h>
37 : #include <linux/posix_acl.h>
38 : #include <linux/hash.h>
39 : #include <linux/bitops.h>
40 : #include <linux/init_task.h>
41 : #include <linux/uaccess.h>
42 :
43 : #include "internal.h"
44 : #include "mount.h"
45 :
46 : /* [Feb-1997 T. Schoebel-Theuer]
47 : * Fundamental changes in the pathname lookup mechanisms (namei)
48 : * were necessary because of omirr. The reason is that omirr needs
49 : * to know the _real_ pathname, not the user-supplied one, in case
50 : * of symlinks (and also when transname replacements occur).
51 : *
52 : * The new code replaces the old recursive symlink resolution with
53 : * an iterative one (in case of non-nested symlink chains). It does
54 : * this with calls to <fs>_follow_link().
55 : * As a side effect, dir_namei(), _namei() and follow_link() are now
56 : * replaced with a single function lookup_dentry() that can handle all
57 : * the special cases of the former code.
58 : *
59 : * With the new dcache, the pathname is stored at each inode, at least as
60 : * long as the refcount of the inode is positive. As a side effect, the
61 : * size of the dcache depends on the inode cache and thus is dynamic.
62 : *
63 : * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
64 : * resolution to correspond with current state of the code.
65 : *
66 : * Note that the symlink resolution is not *completely* iterative.
67 : * There is still a significant amount of tail- and mid- recursion in
68 : * the algorithm. Also, note that <fs>_readlink() is not used in
69 : * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
70 : * may return different results than <fs>_follow_link(). Many virtual
71 : * filesystems (including /proc) exhibit this behavior.
72 : */
73 :
74 : /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
75 : * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
76 : * and the name already exists in form of a symlink, try to create the new
77 : * name indicated by the symlink. The old code always complained that the
78 : * name already exists, due to not following the symlink even if its target
79 : * is nonexistent. The new semantics affects also mknod() and link() when
80 : * the name is a symlink pointing to a non-existent name.
81 : *
82 : * I don't know which semantics is the right one, since I have no access
83 : * to standards. But I found by trial that HP-UX 9.0 has the full "new"
84 : * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
85 : * "old" one. Personally, I think the new semantics is much more logical.
86 : * Note that "ln old new" where "new" is a symlink pointing to a non-existing
87 : * file does succeed in both HP-UX and SunOs, but not in Solaris
88 : * and in the old Linux semantics.
89 : */
90 :
91 : /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
92 : * semantics. See the comments in "open_namei" and "do_link" below.
93 : *
94 : * [10-Sep-98 Alan Modra] Another symlink change.
95 : */
96 :
97 : /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
98 : * inside the path - always follow.
99 : * in the last component in creation/removal/renaming - never follow.
100 : * if LOOKUP_FOLLOW passed - follow.
101 : * if the pathname has trailing slashes - follow.
102 : * otherwise - don't follow.
103 : * (applied in that order).
104 : *
105 : * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
106 : * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
107 : * During the 2.4 we need to fix the userland stuff depending on it -
108 : * hopefully we will be able to get rid of that wart in 2.5. So far only
109 : * XEmacs seems to be relying on it...
110 : */
111 : /*
112 : * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
113 : * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
114 : * any extra contention...
115 : */
116 :
117 : /* In order to reduce some races, while at the same time doing additional
118 : * checking and hopefully speeding things up, we copy filenames to the
119 : * kernel data space before using them..
120 : *
121 : * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
122 : * PATH_MAX includes the nul terminator --RR.
123 : */
124 :
125 : #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
126 :
127 : struct filename *
128 0 : getname_flags(const char __user *filename, int flags, int *empty)
129 : {
130 : struct filename *result;
131 : char *kname;
132 : int len;
133 :
134 0 : result = audit_reusename(filename);
135 : if (result)
136 : return result;
137 :
138 0 : result = __getname();
139 0 : if (unlikely(!result))
140 : return ERR_PTR(-ENOMEM);
141 :
142 : /*
143 : * First, try to embed the struct filename inside the names_cache
144 : * allocation
145 : */
146 0 : kname = (char *)result->iname;
147 0 : result->name = kname;
148 :
149 0 : len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
150 0 : if (unlikely(len < 0)) {
151 0 : __putname(result);
152 0 : return ERR_PTR(len);
153 : }
154 :
155 : /*
156 : * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
157 : * separate struct filename so we can dedicate the entire
158 : * names_cache allocation for the pathname, and re-do the copy from
159 : * userland.
160 : */
161 0 : if (unlikely(len == EMBEDDED_NAME_MAX)) {
162 0 : const size_t size = offsetof(struct filename, iname[1]);
163 0 : kname = (char *)result;
164 :
165 : /*
166 : * size is chosen that way we to guarantee that
167 : * result->iname[0] is within the same object and that
168 : * kname can't be equal to result->iname, no matter what.
169 : */
170 0 : result = kzalloc(size, GFP_KERNEL);
171 0 : if (unlikely(!result)) {
172 0 : __putname(kname);
173 0 : return ERR_PTR(-ENOMEM);
174 : }
175 0 : result->name = kname;
176 0 : len = strncpy_from_user(kname, filename, PATH_MAX);
177 0 : if (unlikely(len < 0)) {
178 0 : __putname(kname);
179 0 : kfree(result);
180 0 : return ERR_PTR(len);
181 : }
182 0 : if (unlikely(len == PATH_MAX)) {
183 0 : __putname(kname);
184 0 : kfree(result);
185 0 : return ERR_PTR(-ENAMETOOLONG);
186 : }
187 : }
188 :
189 0 : result->refcnt = 1;
190 : /* The empty path is special. */
191 0 : if (unlikely(!len)) {
192 0 : if (empty)
193 0 : *empty = 1;
194 0 : if (!(flags & LOOKUP_EMPTY)) {
195 0 : putname(result);
196 0 : return ERR_PTR(-ENOENT);
197 : }
198 : }
199 :
200 0 : result->uptr = filename;
201 0 : result->aname = NULL;
202 0 : audit_getname(result);
203 0 : return result;
204 : }
205 :
206 : struct filename *
207 0 : getname_uflags(const char __user *filename, int uflags)
208 : {
209 0 : int flags = (uflags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
210 :
211 0 : return getname_flags(filename, flags, NULL);
212 : }
213 :
214 : struct filename *
215 0 : getname(const char __user * filename)
216 : {
217 0 : return getname_flags(filename, 0, NULL);
218 : }
219 :
220 : struct filename *
221 3 : getname_kernel(const char * filename)
222 : {
223 : struct filename *result;
224 3 : int len = strlen(filename) + 1;
225 :
226 3 : result = __getname();
227 3 : if (unlikely(!result))
228 : return ERR_PTR(-ENOMEM);
229 :
230 3 : if (len <= EMBEDDED_NAME_MAX) {
231 3 : result->name = (char *)result->iname;
232 0 : } else if (len <= PATH_MAX) {
233 0 : const size_t size = offsetof(struct filename, iname[1]);
234 : struct filename *tmp;
235 :
236 0 : tmp = kmalloc(size, GFP_KERNEL);
237 0 : if (unlikely(!tmp)) {
238 0 : __putname(result);
239 0 : return ERR_PTR(-ENOMEM);
240 : }
241 0 : tmp->name = (char *)result;
242 0 : result = tmp;
243 : } else {
244 0 : __putname(result);
245 0 : return ERR_PTR(-ENAMETOOLONG);
246 : }
247 3 : memcpy((char *)result->name, filename, len);
248 3 : result->uptr = NULL;
249 3 : result->aname = NULL;
250 3 : result->refcnt = 1;
251 3 : audit_getname(result);
252 :
253 3 : return result;
254 : }
255 :
256 3 : void putname(struct filename *name)
257 : {
258 3 : if (IS_ERR(name))
259 : return;
260 :
261 3 : BUG_ON(name->refcnt <= 0);
262 :
263 3 : if (--name->refcnt > 0)
264 : return;
265 :
266 3 : if (name->name != name->iname) {
267 0 : __putname(name->name);
268 0 : kfree(name);
269 : } else
270 3 : __putname(name);
271 : }
272 :
273 : /**
274 : * check_acl - perform ACL permission checking
275 : * @mnt_userns: user namespace of the mount the inode was found from
276 : * @inode: inode to check permissions on
277 : * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
278 : *
279 : * This function performs the ACL permission checking. Since this function
280 : * retrieve POSIX acls it needs to know whether it is called from a blocking or
281 : * non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
282 : *
283 : * If the inode has been found through an idmapped mount the user namespace of
284 : * the vfsmount must be passed through @mnt_userns. This function will then take
285 : * care to map the inode according to @mnt_userns before checking permissions.
286 : * On non-idmapped mounts or if permission checking is to be performed on the
287 : * raw inode simply passs init_user_ns.
288 : */
289 : static int check_acl(struct user_namespace *mnt_userns,
290 : struct inode *inode, int mask)
291 : {
292 : #ifdef CONFIG_FS_POSIX_ACL
293 : struct posix_acl *acl;
294 :
295 : if (mask & MAY_NOT_BLOCK) {
296 : acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
297 : if (!acl)
298 : return -EAGAIN;
299 : /* no ->get_acl() calls in RCU mode... */
300 : if (is_uncached_acl(acl))
301 : return -ECHILD;
302 : return posix_acl_permission(mnt_userns, inode, acl, mask);
303 : }
304 :
305 : acl = get_acl(inode, ACL_TYPE_ACCESS);
306 : if (IS_ERR(acl))
307 : return PTR_ERR(acl);
308 : if (acl) {
309 : int error = posix_acl_permission(mnt_userns, inode, acl, mask);
310 : posix_acl_release(acl);
311 : return error;
312 : }
313 : #endif
314 :
315 : return -EAGAIN;
316 : }
317 :
318 : /**
319 : * acl_permission_check - perform basic UNIX permission checking
320 : * @mnt_userns: user namespace of the mount the inode was found from
321 : * @inode: inode to check permissions on
322 : * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
323 : *
324 : * This function performs the basic UNIX permission checking. Since this
325 : * function may retrieve POSIX acls it needs to know whether it is called from a
326 : * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
327 : *
328 : * If the inode has been found through an idmapped mount the user namespace of
329 : * the vfsmount must be passed through @mnt_userns. This function will then take
330 : * care to map the inode according to @mnt_userns before checking permissions.
331 : * On non-idmapped mounts or if permission checking is to be performed on the
332 : * raw inode simply passs init_user_ns.
333 : */
334 7 : static int acl_permission_check(struct user_namespace *mnt_userns,
335 : struct inode *inode, int mask)
336 : {
337 7 : unsigned int mode = inode->i_mode;
338 : kuid_t i_uid;
339 :
340 : /* Are we the owner? If so, ACL's don't matter */
341 7 : i_uid = i_uid_into_mnt(mnt_userns, inode);
342 14 : if (likely(uid_eq(current_fsuid(), i_uid))) {
343 7 : mask &= 7;
344 7 : mode >>= 6;
345 7 : return (mask & ~mode) ? -EACCES : 0;
346 : }
347 :
348 : /* Do we have ACL's? */
349 : if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
350 : int error = check_acl(mnt_userns, inode, mask);
351 : if (error != -EAGAIN)
352 : return error;
353 : }
354 :
355 : /* Only RWX matters for group/other mode bits */
356 0 : mask &= 7;
357 :
358 : /*
359 : * Are the group permissions different from
360 : * the other permissions in the bits we care
361 : * about? Need to check group ownership if so.
362 : */
363 0 : if (mask & (mode ^ (mode >> 3))) {
364 0 : kgid_t kgid = i_gid_into_mnt(mnt_userns, inode);
365 0 : if (in_group_p(kgid))
366 0 : mode >>= 3;
367 : }
368 :
369 : /* Bits in 'mode' clear that we require? */
370 0 : return (mask & ~mode) ? -EACCES : 0;
371 : }
372 :
373 : /**
374 : * generic_permission - check for access rights on a Posix-like filesystem
375 : * @mnt_userns: user namespace of the mount the inode was found from
376 : * @inode: inode to check access rights for
377 : * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC,
378 : * %MAY_NOT_BLOCK ...)
379 : *
380 : * Used to check for read/write/execute permissions on a file.
381 : * We use "fsuid" for this, letting us set arbitrary permissions
382 : * for filesystem access without changing the "normal" uids which
383 : * are used for other things.
384 : *
385 : * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
386 : * request cannot be satisfied (eg. requires blocking or too much complexity).
387 : * It would then be called again in ref-walk mode.
388 : *
389 : * If the inode has been found through an idmapped mount the user namespace of
390 : * the vfsmount must be passed through @mnt_userns. This function will then take
391 : * care to map the inode according to @mnt_userns before checking permissions.
392 : * On non-idmapped mounts or if permission checking is to be performed on the
393 : * raw inode simply passs init_user_ns.
394 : */
395 7 : int generic_permission(struct user_namespace *mnt_userns, struct inode *inode,
396 : int mask)
397 : {
398 : int ret;
399 :
400 : /*
401 : * Do the basic permission checks.
402 : */
403 7 : ret = acl_permission_check(mnt_userns, inode, mask);
404 7 : if (ret != -EACCES)
405 : return ret;
406 :
407 0 : if (S_ISDIR(inode->i_mode)) {
408 : /* DACs are overridable for directories */
409 0 : if (!(mask & MAY_WRITE))
410 0 : if (capable_wrt_inode_uidgid(mnt_userns, inode,
411 : CAP_DAC_READ_SEARCH))
412 : return 0;
413 0 : if (capable_wrt_inode_uidgid(mnt_userns, inode,
414 : CAP_DAC_OVERRIDE))
415 : return 0;
416 0 : return -EACCES;
417 : }
418 :
419 : /*
420 : * Searching includes executable on directories, else just read.
421 : */
422 0 : mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
423 0 : if (mask == MAY_READ)
424 0 : if (capable_wrt_inode_uidgid(mnt_userns, inode,
425 : CAP_DAC_READ_SEARCH))
426 : return 0;
427 : /*
428 : * Read/write DACs are always overridable.
429 : * Executable DACs are overridable when there is
430 : * at least one exec bit set.
431 : */
432 0 : if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
433 0 : if (capable_wrt_inode_uidgid(mnt_userns, inode,
434 : CAP_DAC_OVERRIDE))
435 : return 0;
436 :
437 : return -EACCES;
438 : }
439 : EXPORT_SYMBOL(generic_permission);
440 :
441 : /**
442 : * do_inode_permission - UNIX permission checking
443 : * @mnt_userns: user namespace of the mount the inode was found from
444 : * @inode: inode to check permissions on
445 : * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
446 : *
447 : * We _really_ want to just do "generic_permission()" without
448 : * even looking at the inode->i_op values. So we keep a cache
449 : * flag in inode->i_opflags, that says "this has not special
450 : * permission function, use the fast case".
451 : */
452 7 : static inline int do_inode_permission(struct user_namespace *mnt_userns,
453 : struct inode *inode, int mask)
454 : {
455 7 : if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
456 2 : if (likely(inode->i_op->permission))
457 0 : return inode->i_op->permission(mnt_userns, inode, mask);
458 :
459 : /* This gets set once for the inode lifetime */
460 4 : spin_lock(&inode->i_lock);
461 2 : inode->i_opflags |= IOP_FASTPERM;
462 2 : spin_unlock(&inode->i_lock);
463 : }
464 7 : return generic_permission(mnt_userns, inode, mask);
465 : }
466 :
467 : /**
468 : * sb_permission - Check superblock-level permissions
469 : * @sb: Superblock of inode to check permission on
470 : * @inode: Inode to check permission on
471 : * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
472 : *
473 : * Separate out file-system wide checks from inode-specific permission checks.
474 : */
475 : static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
476 : {
477 7 : if (unlikely(mask & MAY_WRITE)) {
478 3 : umode_t mode = inode->i_mode;
479 :
480 : /* Nobody gets write access to a read-only fs. */
481 6 : if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
482 : return -EROFS;
483 : }
484 : return 0;
485 : }
486 :
487 : /**
488 : * inode_permission - Check for access rights to a given inode
489 : * @mnt_userns: User namespace of the mount the inode was found from
490 : * @inode: Inode to check permission on
491 : * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
492 : *
493 : * Check for read/write/execute permissions on an inode. We use fs[ug]id for
494 : * this, letting us set arbitrary permissions for filesystem access without
495 : * changing the "normal" UIDs which are used for other things.
496 : *
497 : * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
498 : */
499 7 : int inode_permission(struct user_namespace *mnt_userns,
500 : struct inode *inode, int mask)
501 : {
502 : int retval;
503 :
504 14 : retval = sb_permission(inode->i_sb, inode, mask);
505 7 : if (retval)
506 : return retval;
507 :
508 7 : if (unlikely(mask & MAY_WRITE)) {
509 : /*
510 : * Nobody gets write access to an immutable file.
511 : */
512 3 : if (IS_IMMUTABLE(inode))
513 : return -EPERM;
514 :
515 : /*
516 : * Updating mtime will likely cause i_uid and i_gid to be
517 : * written back improperly if their true value is unknown
518 : * to the vfs.
519 : */
520 3 : if (HAS_UNMAPPED_ID(mnt_userns, inode))
521 : return -EACCES;
522 : }
523 :
524 7 : retval = do_inode_permission(mnt_userns, inode, mask);
525 7 : if (retval)
526 : return retval;
527 :
528 7 : retval = devcgroup_inode_permission(inode, mask);
529 : if (retval)
530 : return retval;
531 :
532 7 : return security_inode_permission(inode, mask);
533 : }
534 : EXPORT_SYMBOL(inode_permission);
535 :
536 : /**
537 : * path_get - get a reference to a path
538 : * @path: path to get the reference to
539 : *
540 : * Given a path increment the reference count to the dentry and the vfsmount.
541 : */
542 2 : void path_get(const struct path *path)
543 : {
544 2 : mntget(path->mnt);
545 4 : dget(path->dentry);
546 2 : }
547 : EXPORT_SYMBOL(path_get);
548 :
549 : /**
550 : * path_put - put a reference to a path
551 : * @path: path to put the reference to
552 : *
553 : * Given a path decrement the reference count to the dentry and the vfsmount.
554 : */
555 0 : void path_put(const struct path *path)
556 : {
557 6 : dput(path->dentry);
558 6 : mntput(path->mnt);
559 0 : }
560 : EXPORT_SYMBOL(path_put);
561 :
562 : #define EMBEDDED_LEVELS 2
563 : struct nameidata {
564 : struct path path;
565 : struct qstr last;
566 : struct path root;
567 : struct inode *inode; /* path.dentry.d_inode */
568 : unsigned int flags, state;
569 : unsigned seq, m_seq, r_seq;
570 : int last_type;
571 : unsigned depth;
572 : int total_link_count;
573 : struct saved {
574 : struct path link;
575 : struct delayed_call done;
576 : const char *name;
577 : unsigned seq;
578 : } *stack, internal[EMBEDDED_LEVELS];
579 : struct filename *name;
580 : struct nameidata *saved;
581 : unsigned root_seq;
582 : int dfd;
583 : kuid_t dir_uid;
584 : umode_t dir_mode;
585 : } __randomize_layout;
586 :
587 : #define ND_ROOT_PRESET 1
588 : #define ND_ROOT_GRABBED 2
589 : #define ND_JUMPED 4
590 :
591 : static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name)
592 : {
593 3 : struct nameidata *old = current->nameidata;
594 3 : p->stack = p->internal;
595 3 : p->depth = 0;
596 3 : p->dfd = dfd;
597 3 : p->name = name;
598 3 : p->path.mnt = NULL;
599 3 : p->path.dentry = NULL;
600 3 : p->total_link_count = old ? old->total_link_count : 0;
601 3 : p->saved = old;
602 3 : current->nameidata = p;
603 : }
604 :
605 : static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name,
606 : const struct path *root)
607 : {
608 3 : __set_nameidata(p, dfd, name);
609 3 : p->state = 0;
610 0 : if (unlikely(root)) {
611 0 : p->state = ND_ROOT_PRESET;
612 0 : p->root = *root;
613 : }
614 : }
615 :
616 3 : static void restore_nameidata(void)
617 : {
618 3 : struct nameidata *now = current->nameidata, *old = now->saved;
619 :
620 3 : current->nameidata = old;
621 3 : if (old)
622 0 : old->total_link_count = now->total_link_count;
623 3 : if (now->stack != now->internal)
624 0 : kfree(now->stack);
625 3 : }
626 :
627 0 : static bool nd_alloc_stack(struct nameidata *nd)
628 : {
629 : struct saved *p;
630 :
631 0 : p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
632 0 : nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL);
633 0 : if (unlikely(!p))
634 : return false;
635 0 : memcpy(p, nd->internal, sizeof(nd->internal));
636 0 : nd->stack = p;
637 0 : return true;
638 : }
639 :
640 : /**
641 : * path_connected - Verify that a dentry is below mnt.mnt_root
642 : *
643 : * Rename can sometimes move a file or directory outside of a bind
644 : * mount, path_connected allows those cases to be detected.
645 : */
646 : static bool path_connected(struct vfsmount *mnt, struct dentry *dentry)
647 : {
648 0 : struct super_block *sb = mnt->mnt_sb;
649 :
650 : /* Bind mounts can have disconnected paths */
651 0 : if (mnt->mnt_root == sb->s_root)
652 : return true;
653 :
654 0 : return is_subdir(dentry, mnt->mnt_root);
655 : }
656 :
657 : static void drop_links(struct nameidata *nd)
658 : {
659 3 : int i = nd->depth;
660 3 : while (i--) {
661 0 : struct saved *last = nd->stack + i;
662 0 : do_delayed_call(&last->done);
663 0 : clear_delayed_call(&last->done);
664 : }
665 : }
666 :
667 3 : static void terminate_walk(struct nameidata *nd)
668 : {
669 6 : drop_links(nd);
670 3 : if (!(nd->flags & LOOKUP_RCU)) {
671 : int i;
672 6 : path_put(&nd->path);
673 3 : for (i = 0; i < nd->depth; i++)
674 0 : path_put(&nd->stack[i].link);
675 3 : if (nd->state & ND_ROOT_GRABBED) {
676 0 : path_put(&nd->root);
677 0 : nd->state &= ~ND_ROOT_GRABBED;
678 : }
679 : } else {
680 0 : nd->flags &= ~LOOKUP_RCU;
681 : rcu_read_unlock();
682 : }
683 3 : nd->depth = 0;
684 3 : nd->path.mnt = NULL;
685 3 : nd->path.dentry = NULL;
686 3 : }
687 :
688 : /* path_put is needed afterwards regardless of success or failure */
689 3 : static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq)
690 : {
691 3 : int res = __legitimize_mnt(path->mnt, mseq);
692 3 : if (unlikely(res)) {
693 0 : if (res > 0)
694 0 : path->mnt = NULL;
695 0 : path->dentry = NULL;
696 0 : return false;
697 : }
698 3 : if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
699 0 : path->dentry = NULL;
700 0 : return false;
701 : }
702 9 : return !read_seqcount_retry(&path->dentry->d_seq, seq);
703 : }
704 :
705 : static inline bool legitimize_path(struct nameidata *nd,
706 : struct path *path, unsigned seq)
707 : {
708 3 : return __legitimize_path(path, seq, nd->m_seq);
709 : }
710 :
711 3 : static bool legitimize_links(struct nameidata *nd)
712 : {
713 : int i;
714 3 : if (unlikely(nd->flags & LOOKUP_CACHED)) {
715 0 : drop_links(nd);
716 0 : nd->depth = 0;
717 0 : return false;
718 : }
719 0 : for (i = 0; i < nd->depth; i++) {
720 0 : struct saved *last = nd->stack + i;
721 0 : if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
722 0 : drop_links(nd);
723 0 : nd->depth = i + 1;
724 0 : return false;
725 : }
726 : }
727 : return true;
728 : }
729 :
730 3 : static bool legitimize_root(struct nameidata *nd)
731 : {
732 : /*
733 : * For scoped-lookups (where nd->root has been zeroed), we need to
734 : * restart the whole lookup from scratch -- because set_root() is wrong
735 : * for these lookups (nd->dfd is the root, not the filesystem root).
736 : */
737 3 : if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED))
738 : return false;
739 : /* Nothing to do if nd->root is zero or is managed by the VFS user. */
740 3 : if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET))
741 : return true;
742 0 : nd->state |= ND_ROOT_GRABBED;
743 0 : return legitimize_path(nd, &nd->root, nd->root_seq);
744 : }
745 :
746 : /*
747 : * Path walking has 2 modes, rcu-walk and ref-walk (see
748 : * Documentation/filesystems/path-lookup.txt). In situations when we can't
749 : * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
750 : * normal reference counts on dentries and vfsmounts to transition to ref-walk
751 : * mode. Refcounts are grabbed at the last known good point before rcu-walk
752 : * got stuck, so ref-walk may continue from there. If this is not successful
753 : * (eg. a seqcount has changed), then failure is returned and it's up to caller
754 : * to restart the path walk from the beginning in ref-walk mode.
755 : */
756 :
757 : /**
758 : * try_to_unlazy - try to switch to ref-walk mode.
759 : * @nd: nameidata pathwalk data
760 : * Returns: true on success, false on failure
761 : *
762 : * try_to_unlazy attempts to legitimize the current nd->path and nd->root
763 : * for ref-walk mode.
764 : * Must be called from rcu-walk context.
765 : * Nothing should touch nameidata between try_to_unlazy() failure and
766 : * terminate_walk().
767 : */
768 3 : static bool try_to_unlazy(struct nameidata *nd)
769 : {
770 3 : struct dentry *parent = nd->path.dentry;
771 :
772 3 : BUG_ON(!(nd->flags & LOOKUP_RCU));
773 :
774 3 : nd->flags &= ~LOOKUP_RCU;
775 3 : if (unlikely(!legitimize_links(nd)))
776 : goto out1;
777 6 : if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
778 : goto out;
779 3 : if (unlikely(!legitimize_root(nd)))
780 : goto out;
781 : rcu_read_unlock();
782 3 : BUG_ON(nd->inode != parent->d_inode);
783 : return true;
784 :
785 : out1:
786 0 : nd->path.mnt = NULL;
787 0 : nd->path.dentry = NULL;
788 : out:
789 : rcu_read_unlock();
790 0 : return false;
791 : }
792 :
793 : /**
794 : * try_to_unlazy_next - try to switch to ref-walk mode.
795 : * @nd: nameidata pathwalk data
796 : * @dentry: next dentry to step into
797 : * @seq: seq number to check @dentry against
798 : * Returns: true on success, false on failure
799 : *
800 : * Similar to to try_to_unlazy(), but here we have the next dentry already
801 : * picked by rcu-walk and want to legitimize that in addition to the current
802 : * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context.
803 : * Nothing should touch nameidata between try_to_unlazy_next() failure and
804 : * terminate_walk().
805 : */
806 0 : static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry, unsigned seq)
807 : {
808 0 : BUG_ON(!(nd->flags & LOOKUP_RCU));
809 :
810 0 : nd->flags &= ~LOOKUP_RCU;
811 0 : if (unlikely(!legitimize_links(nd)))
812 : goto out2;
813 0 : if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
814 : goto out2;
815 0 : if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
816 : goto out1;
817 :
818 : /*
819 : * We need to move both the parent and the dentry from the RCU domain
820 : * to be properly refcounted. And the sequence number in the dentry
821 : * validates *both* dentry counters, since we checked the sequence
822 : * number of the parent after we got the child sequence number. So we
823 : * know the parent must still be valid if the child sequence number is
824 : */
825 0 : if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
826 : goto out;
827 0 : if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
828 : goto out_dput;
829 : /*
830 : * Sequence counts matched. Now make sure that the root is
831 : * still valid and get it if required.
832 : */
833 0 : if (unlikely(!legitimize_root(nd)))
834 : goto out_dput;
835 : rcu_read_unlock();
836 0 : return true;
837 :
838 : out2:
839 0 : nd->path.mnt = NULL;
840 : out1:
841 0 : nd->path.dentry = NULL;
842 : out:
843 : rcu_read_unlock();
844 0 : return false;
845 : out_dput:
846 : rcu_read_unlock();
847 0 : dput(dentry);
848 0 : return false;
849 : }
850 :
851 : static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
852 : {
853 1 : if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
854 0 : return dentry->d_op->d_revalidate(dentry, flags);
855 : else
856 : return 1;
857 : }
858 :
859 : /**
860 : * complete_walk - successful completion of path walk
861 : * @nd: pointer nameidata
862 : *
863 : * If we had been in RCU mode, drop out of it and legitimize nd->path.
864 : * Revalidate the final result, unless we'd already done that during
865 : * the path walk or the filesystem doesn't ask for it. Return 0 on
866 : * success, -error on failure. In case of failure caller does not
867 : * need to drop nd->path.
868 : */
869 3 : static int complete_walk(struct nameidata *nd)
870 : {
871 3 : struct dentry *dentry = nd->path.dentry;
872 : int status;
873 :
874 3 : if (nd->flags & LOOKUP_RCU) {
875 : /*
876 : * We don't want to zero nd->root for scoped-lookups or
877 : * externally-managed nd->root.
878 : */
879 3 : if (!(nd->state & ND_ROOT_PRESET))
880 3 : if (!(nd->flags & LOOKUP_IS_SCOPED))
881 3 : nd->root.mnt = NULL;
882 3 : nd->flags &= ~LOOKUP_CACHED;
883 3 : if (!try_to_unlazy(nd))
884 : return -ECHILD;
885 : }
886 :
887 3 : if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
888 : /*
889 : * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
890 : * ever step outside the root during lookup" and should already
891 : * be guaranteed by the rest of namei, we want to avoid a namei
892 : * BUG resulting in userspace being given a path that was not
893 : * scoped within the root at some point during the lookup.
894 : *
895 : * So, do a final sanity-check to make sure that in the
896 : * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
897 : * we won't silently return an fd completely outside of the
898 : * requested root to userspace.
899 : *
900 : * Userspace could move the path outside the root after this
901 : * check, but as discussed elsewhere this is not a concern (the
902 : * resolved file was inside the root at some point).
903 : */
904 0 : if (!path_is_under(&nd->path, &nd->root))
905 : return -EXDEV;
906 : }
907 :
908 3 : if (likely(!(nd->state & ND_JUMPED)))
909 : return 0;
910 :
911 0 : if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
912 : return 0;
913 :
914 0 : status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
915 0 : if (status > 0)
916 : return 0;
917 :
918 0 : if (!status)
919 0 : status = -ESTALE;
920 :
921 : return status;
922 : }
923 :
924 3 : static int set_root(struct nameidata *nd)
925 : {
926 3 : struct fs_struct *fs = current->fs;
927 :
928 : /*
929 : * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
930 : * still have to ensure it doesn't happen because it will cause a breakout
931 : * from the dirfd.
932 : */
933 3 : if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED))
934 : return -ENOTRECOVERABLE;
935 :
936 3 : if (nd->flags & LOOKUP_RCU) {
937 : unsigned seq;
938 :
939 : do {
940 6 : seq = read_seqcount_begin(&fs->seq);
941 3 : nd->root = fs->root;
942 9 : nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
943 9 : } while (read_seqcount_retry(&fs->seq, seq));
944 : } else {
945 0 : get_fs_root(fs, &nd->root);
946 0 : nd->state |= ND_ROOT_GRABBED;
947 : }
948 : return 0;
949 : }
950 :
951 3 : static int nd_jump_root(struct nameidata *nd)
952 : {
953 3 : if (unlikely(nd->flags & LOOKUP_BENEATH))
954 : return -EXDEV;
955 3 : if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
956 : /* Absolute path arguments to path_init() are allowed. */
957 0 : if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt)
958 : return -EXDEV;
959 : }
960 3 : if (!nd->root.mnt) {
961 3 : int error = set_root(nd);
962 3 : if (error)
963 : return error;
964 : }
965 3 : if (nd->flags & LOOKUP_RCU) {
966 : struct dentry *d;
967 3 : nd->path = nd->root;
968 3 : d = nd->path.dentry;
969 3 : nd->inode = d->d_inode;
970 3 : nd->seq = nd->root_seq;
971 9 : if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
972 : return -ECHILD;
973 : } else {
974 0 : path_put(&nd->path);
975 0 : nd->path = nd->root;
976 0 : path_get(&nd->path);
977 0 : nd->inode = nd->path.dentry->d_inode;
978 : }
979 3 : nd->state |= ND_JUMPED;
980 3 : return 0;
981 : }
982 :
983 : /*
984 : * Helper to directly jump to a known parsed path from ->get_link,
985 : * caller must have taken a reference to path beforehand.
986 : */
987 0 : int nd_jump_link(struct path *path)
988 : {
989 0 : int error = -ELOOP;
990 0 : struct nameidata *nd = current->nameidata;
991 :
992 0 : if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS))
993 : goto err;
994 :
995 0 : error = -EXDEV;
996 0 : if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
997 0 : if (nd->path.mnt != path->mnt)
998 : goto err;
999 : }
1000 : /* Not currently safe for scoped-lookups. */
1001 0 : if (unlikely(nd->flags & LOOKUP_IS_SCOPED))
1002 : goto err;
1003 :
1004 0 : path_put(&nd->path);
1005 0 : nd->path = *path;
1006 0 : nd->inode = nd->path.dentry->d_inode;
1007 0 : nd->state |= ND_JUMPED;
1008 0 : return 0;
1009 :
1010 : err:
1011 0 : path_put(path);
1012 0 : return error;
1013 : }
1014 :
1015 0 : static inline void put_link(struct nameidata *nd)
1016 : {
1017 0 : struct saved *last = nd->stack + --nd->depth;
1018 0 : do_delayed_call(&last->done);
1019 0 : if (!(nd->flags & LOOKUP_RCU))
1020 0 : path_put(&last->link);
1021 0 : }
1022 :
1023 : static int sysctl_protected_symlinks __read_mostly;
1024 : static int sysctl_protected_hardlinks __read_mostly;
1025 : static int sysctl_protected_fifos __read_mostly;
1026 : static int sysctl_protected_regular __read_mostly;
1027 :
1028 : #ifdef CONFIG_SYSCTL
1029 : static struct ctl_table namei_sysctls[] = {
1030 : {
1031 : .procname = "protected_symlinks",
1032 : .data = &sysctl_protected_symlinks,
1033 : .maxlen = sizeof(int),
1034 : .mode = 0600,
1035 : .proc_handler = proc_dointvec_minmax,
1036 : .extra1 = SYSCTL_ZERO,
1037 : .extra2 = SYSCTL_ONE,
1038 : },
1039 : {
1040 : .procname = "protected_hardlinks",
1041 : .data = &sysctl_protected_hardlinks,
1042 : .maxlen = sizeof(int),
1043 : .mode = 0600,
1044 : .proc_handler = proc_dointvec_minmax,
1045 : .extra1 = SYSCTL_ZERO,
1046 : .extra2 = SYSCTL_ONE,
1047 : },
1048 : {
1049 : .procname = "protected_fifos",
1050 : .data = &sysctl_protected_fifos,
1051 : .maxlen = sizeof(int),
1052 : .mode = 0600,
1053 : .proc_handler = proc_dointvec_minmax,
1054 : .extra1 = SYSCTL_ZERO,
1055 : .extra2 = SYSCTL_TWO,
1056 : },
1057 : {
1058 : .procname = "protected_regular",
1059 : .data = &sysctl_protected_regular,
1060 : .maxlen = sizeof(int),
1061 : .mode = 0600,
1062 : .proc_handler = proc_dointvec_minmax,
1063 : .extra1 = SYSCTL_ZERO,
1064 : .extra2 = SYSCTL_TWO,
1065 : },
1066 : { }
1067 : };
1068 :
1069 1 : static int __init init_fs_namei_sysctls(void)
1070 : {
1071 1 : register_sysctl_init("fs", namei_sysctls);
1072 1 : return 0;
1073 : }
1074 : fs_initcall(init_fs_namei_sysctls);
1075 :
1076 : #endif /* CONFIG_SYSCTL */
1077 :
1078 : /**
1079 : * may_follow_link - Check symlink following for unsafe situations
1080 : * @nd: nameidata pathwalk data
1081 : *
1082 : * In the case of the sysctl_protected_symlinks sysctl being enabled,
1083 : * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
1084 : * in a sticky world-writable directory. This is to protect privileged
1085 : * processes from failing races against path names that may change out
1086 : * from under them by way of other users creating malicious symlinks.
1087 : * It will permit symlinks to be followed only when outside a sticky
1088 : * world-writable directory, or when the uid of the symlink and follower
1089 : * match, or when the directory owner matches the symlink's owner.
1090 : *
1091 : * Returns 0 if following the symlink is allowed, -ve on error.
1092 : */
1093 0 : static inline int may_follow_link(struct nameidata *nd, const struct inode *inode)
1094 : {
1095 : struct user_namespace *mnt_userns;
1096 : kuid_t i_uid;
1097 :
1098 0 : if (!sysctl_protected_symlinks)
1099 : return 0;
1100 :
1101 0 : mnt_userns = mnt_user_ns(nd->path.mnt);
1102 0 : i_uid = i_uid_into_mnt(mnt_userns, inode);
1103 : /* Allowed if owner and follower match. */
1104 0 : if (uid_eq(current_cred()->fsuid, i_uid))
1105 : return 0;
1106 :
1107 : /* Allowed if parent directory not sticky and world-writable. */
1108 0 : if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
1109 : return 0;
1110 :
1111 : /* Allowed if parent directory and link owner match. */
1112 0 : if (uid_valid(nd->dir_uid) && uid_eq(nd->dir_uid, i_uid))
1113 : return 0;
1114 :
1115 0 : if (nd->flags & LOOKUP_RCU)
1116 : return -ECHILD;
1117 :
1118 0 : audit_inode(nd->name, nd->stack[0].link.dentry, 0);
1119 0 : audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link");
1120 0 : return -EACCES;
1121 : }
1122 :
1123 : /**
1124 : * safe_hardlink_source - Check for safe hardlink conditions
1125 : * @mnt_userns: user namespace of the mount the inode was found from
1126 : * @inode: the source inode to hardlink from
1127 : *
1128 : * Return false if at least one of the following conditions:
1129 : * - inode is not a regular file
1130 : * - inode is setuid
1131 : * - inode is setgid and group-exec
1132 : * - access failure for read and write
1133 : *
1134 : * Otherwise returns true.
1135 : */
1136 0 : static bool safe_hardlink_source(struct user_namespace *mnt_userns,
1137 : struct inode *inode)
1138 : {
1139 0 : umode_t mode = inode->i_mode;
1140 :
1141 : /* Special files should not get pinned to the filesystem. */
1142 0 : if (!S_ISREG(mode))
1143 : return false;
1144 :
1145 : /* Setuid files should not get pinned to the filesystem. */
1146 0 : if (mode & S_ISUID)
1147 : return false;
1148 :
1149 : /* Executable setgid files should not get pinned to the filesystem. */
1150 0 : if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
1151 : return false;
1152 :
1153 : /* Hardlinking to unreadable or unwritable sources is dangerous. */
1154 0 : if (inode_permission(mnt_userns, inode, MAY_READ | MAY_WRITE))
1155 : return false;
1156 :
1157 0 : return true;
1158 : }
1159 :
1160 : /**
1161 : * may_linkat - Check permissions for creating a hardlink
1162 : * @mnt_userns: user namespace of the mount the inode was found from
1163 : * @link: the source to hardlink from
1164 : *
1165 : * Block hardlink when all of:
1166 : * - sysctl_protected_hardlinks enabled
1167 : * - fsuid does not match inode
1168 : * - hardlink source is unsafe (see safe_hardlink_source() above)
1169 : * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1170 : *
1171 : * If the inode has been found through an idmapped mount the user namespace of
1172 : * the vfsmount must be passed through @mnt_userns. This function will then take
1173 : * care to map the inode according to @mnt_userns before checking permissions.
1174 : * On non-idmapped mounts or if permission checking is to be performed on the
1175 : * raw inode simply passs init_user_ns.
1176 : *
1177 : * Returns 0 if successful, -ve on error.
1178 : */
1179 0 : int may_linkat(struct user_namespace *mnt_userns, struct path *link)
1180 : {
1181 0 : struct inode *inode = link->dentry->d_inode;
1182 :
1183 : /* Inode writeback is not safe when the uid or gid are invalid. */
1184 0 : if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) ||
1185 0 : !gid_valid(i_gid_into_mnt(mnt_userns, inode)))
1186 : return -EOVERFLOW;
1187 :
1188 0 : if (!sysctl_protected_hardlinks)
1189 : return 0;
1190 :
1191 : /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1192 : * otherwise, it must be a safe source.
1193 : */
1194 0 : if (safe_hardlink_source(mnt_userns, inode) ||
1195 0 : inode_owner_or_capable(mnt_userns, inode))
1196 : return 0;
1197 :
1198 : audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
1199 : return -EPERM;
1200 : }
1201 :
1202 : /**
1203 : * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1204 : * should be allowed, or not, on files that already
1205 : * exist.
1206 : * @mnt_userns: user namespace of the mount the inode was found from
1207 : * @nd: nameidata pathwalk data
1208 : * @inode: the inode of the file to open
1209 : *
1210 : * Block an O_CREAT open of a FIFO (or a regular file) when:
1211 : * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1212 : * - the file already exists
1213 : * - we are in a sticky directory
1214 : * - we don't own the file
1215 : * - the owner of the directory doesn't own the file
1216 : * - the directory is world writable
1217 : * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1218 : * the directory doesn't have to be world writable: being group writable will
1219 : * be enough.
1220 : *
1221 : * If the inode has been found through an idmapped mount the user namespace of
1222 : * the vfsmount must be passed through @mnt_userns. This function will then take
1223 : * care to map the inode according to @mnt_userns before checking permissions.
1224 : * On non-idmapped mounts or if permission checking is to be performed on the
1225 : * raw inode simply passs init_user_ns.
1226 : *
1227 : * Returns 0 if the open is allowed, -ve on error.
1228 : */
1229 0 : static int may_create_in_sticky(struct user_namespace *mnt_userns,
1230 : struct nameidata *nd, struct inode *const inode)
1231 : {
1232 0 : umode_t dir_mode = nd->dir_mode;
1233 0 : kuid_t dir_uid = nd->dir_uid;
1234 :
1235 0 : if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1236 0 : (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1237 0 : likely(!(dir_mode & S_ISVTX)) ||
1238 0 : uid_eq(i_uid_into_mnt(mnt_userns, inode), dir_uid) ||
1239 0 : uid_eq(current_fsuid(), i_uid_into_mnt(mnt_userns, inode)))
1240 : return 0;
1241 :
1242 0 : if (likely(dir_mode & 0002) ||
1243 0 : (dir_mode & 0020 &&
1244 0 : ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1245 0 : (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1246 0 : const char *operation = S_ISFIFO(inode->i_mode) ?
1247 : "sticky_create_fifo" :
1248 : "sticky_create_regular";
1249 : audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
1250 : return -EACCES;
1251 : }
1252 : return 0;
1253 : }
1254 :
1255 : /*
1256 : * follow_up - Find the mountpoint of path's vfsmount
1257 : *
1258 : * Given a path, find the mountpoint of its source file system.
1259 : * Replace @path with the path of the mountpoint in the parent mount.
1260 : * Up is towards /.
1261 : *
1262 : * Return 1 if we went up a level and 0 if we were already at the
1263 : * root.
1264 : */
1265 0 : int follow_up(struct path *path)
1266 : {
1267 0 : struct mount *mnt = real_mount(path->mnt);
1268 : struct mount *parent;
1269 : struct dentry *mountpoint;
1270 :
1271 0 : read_seqlock_excl(&mount_lock);
1272 0 : parent = mnt->mnt_parent;
1273 0 : if (parent == mnt) {
1274 0 : read_sequnlock_excl(&mount_lock);
1275 0 : return 0;
1276 : }
1277 0 : mntget(&parent->mnt);
1278 0 : mountpoint = dget(mnt->mnt_mountpoint);
1279 0 : read_sequnlock_excl(&mount_lock);
1280 0 : dput(path->dentry);
1281 0 : path->dentry = mountpoint;
1282 0 : mntput(path->mnt);
1283 0 : path->mnt = &parent->mnt;
1284 0 : return 1;
1285 : }
1286 : EXPORT_SYMBOL(follow_up);
1287 :
1288 : static bool choose_mountpoint_rcu(struct mount *m, const struct path *root,
1289 : struct path *path, unsigned *seqp)
1290 : {
1291 0 : while (mnt_has_parent(m)) {
1292 0 : struct dentry *mountpoint = m->mnt_mountpoint;
1293 :
1294 0 : m = m->mnt_parent;
1295 0 : if (unlikely(root->dentry == mountpoint &&
1296 : root->mnt == &m->mnt))
1297 : break;
1298 0 : if (mountpoint != m->mnt.mnt_root) {
1299 0 : path->mnt = &m->mnt;
1300 0 : path->dentry = mountpoint;
1301 0 : *seqp = read_seqcount_begin(&mountpoint->d_seq);
1302 : return true;
1303 : }
1304 : }
1305 : return false;
1306 : }
1307 :
1308 0 : static bool choose_mountpoint(struct mount *m, const struct path *root,
1309 : struct path *path)
1310 : {
1311 : bool found;
1312 :
1313 : rcu_read_lock();
1314 : while (1) {
1315 0 : unsigned seq, mseq = read_seqbegin(&mount_lock);
1316 :
1317 0 : found = choose_mountpoint_rcu(m, root, path, &seq);
1318 0 : if (unlikely(!found)) {
1319 0 : if (!read_seqretry(&mount_lock, mseq))
1320 : break;
1321 : } else {
1322 0 : if (likely(__legitimize_path(path, seq, mseq)))
1323 : break;
1324 0 : rcu_read_unlock();
1325 0 : path_put(path);
1326 : rcu_read_lock();
1327 : }
1328 : }
1329 : rcu_read_unlock();
1330 0 : return found;
1331 : }
1332 :
1333 : /*
1334 : * Perform an automount
1335 : * - return -EISDIR to tell follow_managed() to stop and return the path we
1336 : * were called with.
1337 : */
1338 0 : static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1339 : {
1340 0 : struct dentry *dentry = path->dentry;
1341 :
1342 : /* We don't want to mount if someone's just doing a stat -
1343 : * unless they're stat'ing a directory and appended a '/' to
1344 : * the name.
1345 : *
1346 : * We do, however, want to mount if someone wants to open or
1347 : * create a file of any type under the mountpoint, wants to
1348 : * traverse through the mountpoint or wants to open the
1349 : * mounted directory. Also, autofs may mark negative dentries
1350 : * as being automount points. These will need the attentions
1351 : * of the daemon to instantiate them before they can be used.
1352 : */
1353 0 : if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1354 0 : LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1355 0 : dentry->d_inode)
1356 : return -EISDIR;
1357 :
1358 0 : if (count && (*count)++ >= MAXSYMLINKS)
1359 : return -ELOOP;
1360 :
1361 0 : return finish_automount(dentry->d_op->d_automount(path), path);
1362 : }
1363 :
1364 : /*
1365 : * mount traversal - out-of-line part. One note on ->d_flags accesses -
1366 : * dentries are pinned but not locked here, so negative dentry can go
1367 : * positive right under us. Use of smp_load_acquire() provides a barrier
1368 : * sufficient for ->d_inode and ->d_flags consistency.
1369 : */
1370 0 : static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped,
1371 : int *count, unsigned lookup_flags)
1372 : {
1373 0 : struct vfsmount *mnt = path->mnt;
1374 0 : bool need_mntput = false;
1375 0 : int ret = 0;
1376 :
1377 0 : while (flags & DCACHE_MANAGED_DENTRY) {
1378 : /* Allow the filesystem to manage the transit without i_mutex
1379 : * being held. */
1380 0 : if (flags & DCACHE_MANAGE_TRANSIT) {
1381 0 : ret = path->dentry->d_op->d_manage(path, false);
1382 0 : flags = smp_load_acquire(&path->dentry->d_flags);
1383 0 : if (ret < 0)
1384 : break;
1385 : }
1386 :
1387 0 : if (flags & DCACHE_MOUNTED) { // something's mounted on it..
1388 0 : struct vfsmount *mounted = lookup_mnt(path);
1389 0 : if (mounted) { // ... in our namespace
1390 0 : dput(path->dentry);
1391 0 : if (need_mntput)
1392 0 : mntput(path->mnt);
1393 0 : path->mnt = mounted;
1394 0 : path->dentry = dget(mounted->mnt_root);
1395 : // here we know it's positive
1396 0 : flags = path->dentry->d_flags;
1397 0 : need_mntput = true;
1398 0 : continue;
1399 : }
1400 : }
1401 :
1402 0 : if (!(flags & DCACHE_NEED_AUTOMOUNT))
1403 : break;
1404 :
1405 : // uncovered automount point
1406 0 : ret = follow_automount(path, count, lookup_flags);
1407 0 : flags = smp_load_acquire(&path->dentry->d_flags);
1408 0 : if (ret < 0)
1409 : break;
1410 : }
1411 :
1412 0 : if (ret == -EISDIR)
1413 0 : ret = 0;
1414 : // possible if you race with several mount --move
1415 0 : if (need_mntput && path->mnt == mnt)
1416 0 : mntput(path->mnt);
1417 0 : if (!ret && unlikely(d_flags_negative(flags)))
1418 0 : ret = -ENOENT;
1419 0 : *jumped = need_mntput;
1420 0 : return ret;
1421 : }
1422 :
1423 0 : static inline int traverse_mounts(struct path *path, bool *jumped,
1424 : int *count, unsigned lookup_flags)
1425 : {
1426 0 : unsigned flags = smp_load_acquire(&path->dentry->d_flags);
1427 :
1428 : /* fastpath */
1429 0 : if (likely(!(flags & DCACHE_MANAGED_DENTRY))) {
1430 0 : *jumped = false;
1431 0 : if (unlikely(d_flags_negative(flags)))
1432 : return -ENOENT;
1433 0 : return 0;
1434 : }
1435 0 : return __traverse_mounts(path, flags, jumped, count, lookup_flags);
1436 : }
1437 :
1438 0 : int follow_down_one(struct path *path)
1439 : {
1440 : struct vfsmount *mounted;
1441 :
1442 0 : mounted = lookup_mnt(path);
1443 0 : if (mounted) {
1444 0 : dput(path->dentry);
1445 0 : mntput(path->mnt);
1446 0 : path->mnt = mounted;
1447 0 : path->dentry = dget(mounted->mnt_root);
1448 0 : return 1;
1449 : }
1450 : return 0;
1451 : }
1452 : EXPORT_SYMBOL(follow_down_one);
1453 :
1454 : /*
1455 : * Follow down to the covering mount currently visible to userspace. At each
1456 : * point, the filesystem owning that dentry may be queried as to whether the
1457 : * caller is permitted to proceed or not.
1458 : */
1459 0 : int follow_down(struct path *path)
1460 : {
1461 0 : struct vfsmount *mnt = path->mnt;
1462 : bool jumped;
1463 0 : int ret = traverse_mounts(path, &jumped, NULL, 0);
1464 :
1465 0 : if (path->mnt != mnt)
1466 0 : mntput(mnt);
1467 0 : return ret;
1468 : }
1469 : EXPORT_SYMBOL(follow_down);
1470 :
1471 : /*
1472 : * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1473 : * we meet a managed dentry that would need blocking.
1474 : */
1475 1 : static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1476 : struct inode **inode, unsigned *seqp)
1477 : {
1478 1 : struct dentry *dentry = path->dentry;
1479 1 : unsigned int flags = dentry->d_flags;
1480 :
1481 1 : if (likely(!(flags & DCACHE_MANAGED_DENTRY)))
1482 : return true;
1483 :
1484 0 : if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1485 : return false;
1486 :
1487 : for (;;) {
1488 : /*
1489 : * Don't forget we might have a non-mountpoint managed dentry
1490 : * that wants to block transit.
1491 : */
1492 0 : if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) {
1493 0 : int res = dentry->d_op->d_manage(path, true);
1494 0 : if (res)
1495 0 : return res == -EISDIR;
1496 0 : flags = dentry->d_flags;
1497 : }
1498 :
1499 0 : if (flags & DCACHE_MOUNTED) {
1500 0 : struct mount *mounted = __lookup_mnt(path->mnt, dentry);
1501 0 : if (mounted) {
1502 0 : path->mnt = &mounted->mnt;
1503 0 : dentry = path->dentry = mounted->mnt.mnt_root;
1504 0 : nd->state |= ND_JUMPED;
1505 0 : *seqp = read_seqcount_begin(&dentry->d_seq);
1506 0 : *inode = dentry->d_inode;
1507 : /*
1508 : * We don't need to re-check ->d_seq after this
1509 : * ->d_inode read - there will be an RCU delay
1510 : * between mount hash removal and ->mnt_root
1511 : * becoming unpinned.
1512 : */
1513 0 : flags = dentry->d_flags;
1514 0 : continue;
1515 : }
1516 0 : if (read_seqretry(&mount_lock, nd->m_seq))
1517 : return false;
1518 : }
1519 0 : return !(flags & DCACHE_NEED_AUTOMOUNT);
1520 : }
1521 : }
1522 :
1523 1 : static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1524 : struct path *path, struct inode **inode,
1525 : unsigned int *seqp)
1526 : {
1527 : bool jumped;
1528 : int ret;
1529 :
1530 1 : path->mnt = nd->path.mnt;
1531 1 : path->dentry = dentry;
1532 1 : if (nd->flags & LOOKUP_RCU) {
1533 1 : unsigned int seq = *seqp;
1534 1 : if (unlikely(!*inode))
1535 : return -ENOENT;
1536 1 : if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1537 : return 0;
1538 0 : if (!try_to_unlazy_next(nd, dentry, seq))
1539 : return -ECHILD;
1540 : // *path might've been clobbered by __follow_mount_rcu()
1541 0 : path->mnt = nd->path.mnt;
1542 0 : path->dentry = dentry;
1543 : }
1544 0 : ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags);
1545 0 : if (jumped) {
1546 0 : if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1547 : ret = -EXDEV;
1548 : else
1549 0 : nd->state |= ND_JUMPED;
1550 : }
1551 0 : if (unlikely(ret)) {
1552 0 : dput(path->dentry);
1553 0 : if (path->mnt != nd->path.mnt)
1554 0 : mntput(path->mnt);
1555 : } else {
1556 0 : *inode = d_backing_inode(path->dentry);
1557 0 : *seqp = 0; /* out of RCU mode, so the value doesn't matter */
1558 : }
1559 : return ret;
1560 : }
1561 :
1562 : /*
1563 : * This looks up the name in dcache and possibly revalidates the found dentry.
1564 : * NULL is returned if the dentry does not exist in the cache.
1565 : */
1566 3 : static struct dentry *lookup_dcache(const struct qstr *name,
1567 : struct dentry *dir,
1568 : unsigned int flags)
1569 : {
1570 3 : struct dentry *dentry = d_lookup(dir, name);
1571 3 : if (dentry) {
1572 0 : int error = d_revalidate(dentry, flags);
1573 0 : if (unlikely(error <= 0)) {
1574 0 : if (!error)
1575 0 : d_invalidate(dentry);
1576 0 : dput(dentry);
1577 0 : return ERR_PTR(error);
1578 : }
1579 : }
1580 : return dentry;
1581 : }
1582 :
1583 : /*
1584 : * Parent directory has inode locked exclusive. This is one
1585 : * and only case when ->lookup() gets called on non in-lookup
1586 : * dentries - as the matter of fact, this only gets called
1587 : * when directory is guaranteed to have no in-lookup children
1588 : * at all.
1589 : */
1590 3 : static struct dentry *__lookup_hash(const struct qstr *name,
1591 : struct dentry *base, unsigned int flags)
1592 : {
1593 3 : struct dentry *dentry = lookup_dcache(name, base, flags);
1594 : struct dentry *old;
1595 3 : struct inode *dir = base->d_inode;
1596 :
1597 3 : if (dentry)
1598 : return dentry;
1599 :
1600 : /* Don't create child dentry for a dead directory. */
1601 3 : if (unlikely(IS_DEADDIR(dir)))
1602 : return ERR_PTR(-ENOENT);
1603 :
1604 3 : dentry = d_alloc(base, name);
1605 3 : if (unlikely(!dentry))
1606 : return ERR_PTR(-ENOMEM);
1607 :
1608 3 : old = dir->i_op->lookup(dir, dentry, flags);
1609 3 : if (unlikely(old)) {
1610 0 : dput(dentry);
1611 0 : dentry = old;
1612 : }
1613 : return dentry;
1614 : }
1615 :
1616 1 : static struct dentry *lookup_fast(struct nameidata *nd,
1617 : struct inode **inode,
1618 : unsigned *seqp)
1619 : {
1620 1 : struct dentry *dentry, *parent = nd->path.dentry;
1621 1 : int status = 1;
1622 :
1623 : /*
1624 : * Rename seqlock is not required here because in the off chance
1625 : * of a false negative due to a concurrent rename, the caller is
1626 : * going to fall back to non-racy lookup.
1627 : */
1628 1 : if (nd->flags & LOOKUP_RCU) {
1629 : unsigned seq;
1630 1 : dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1631 1 : if (unlikely(!dentry)) {
1632 0 : if (!try_to_unlazy(nd))
1633 1 : return ERR_PTR(-ECHILD);
1634 0 : return NULL;
1635 : }
1636 :
1637 : /*
1638 : * This sequence count validates that the inode matches
1639 : * the dentry name information from lookup.
1640 : */
1641 2 : *inode = d_backing_inode(dentry);
1642 3 : if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1643 : return ERR_PTR(-ECHILD);
1644 :
1645 : /*
1646 : * This sequence count validates that the parent had no
1647 : * changes while we did the lookup of the dentry above.
1648 : *
1649 : * The memory barrier in read_seqcount_begin of child is
1650 : * enough, we can use __read_seqcount_retry here.
1651 : */
1652 3 : if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1653 : return ERR_PTR(-ECHILD);
1654 :
1655 1 : *seqp = seq;
1656 2 : status = d_revalidate(dentry, nd->flags);
1657 1 : if (likely(status > 0))
1658 : return dentry;
1659 0 : if (!try_to_unlazy_next(nd, dentry, seq))
1660 : return ERR_PTR(-ECHILD);
1661 0 : if (status == -ECHILD)
1662 : /* we'd been told to redo it in non-rcu mode */
1663 0 : status = d_revalidate(dentry, nd->flags);
1664 : } else {
1665 0 : dentry = __d_lookup(parent, &nd->last);
1666 0 : if (unlikely(!dentry))
1667 : return NULL;
1668 0 : status = d_revalidate(dentry, nd->flags);
1669 : }
1670 0 : if (unlikely(status <= 0)) {
1671 0 : if (!status)
1672 0 : d_invalidate(dentry);
1673 0 : dput(dentry);
1674 0 : return ERR_PTR(status);
1675 : }
1676 : return dentry;
1677 : }
1678 :
1679 : /* Fast lookup failed, do it the slow way */
1680 0 : static struct dentry *__lookup_slow(const struct qstr *name,
1681 : struct dentry *dir,
1682 : unsigned int flags)
1683 : {
1684 : struct dentry *dentry, *old;
1685 0 : struct inode *inode = dir->d_inode;
1686 0 : DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1687 :
1688 : /* Don't go there if it's already dead */
1689 0 : if (unlikely(IS_DEADDIR(inode)))
1690 : return ERR_PTR(-ENOENT);
1691 : again:
1692 0 : dentry = d_alloc_parallel(dir, name, &wq);
1693 0 : if (IS_ERR(dentry))
1694 : return dentry;
1695 0 : if (unlikely(!d_in_lookup(dentry))) {
1696 0 : int error = d_revalidate(dentry, flags);
1697 0 : if (unlikely(error <= 0)) {
1698 0 : if (!error) {
1699 0 : d_invalidate(dentry);
1700 0 : dput(dentry);
1701 0 : goto again;
1702 : }
1703 0 : dput(dentry);
1704 0 : dentry = ERR_PTR(error);
1705 : }
1706 : } else {
1707 0 : old = inode->i_op->lookup(inode, dentry, flags);
1708 0 : d_lookup_done(dentry);
1709 0 : if (unlikely(old)) {
1710 0 : dput(dentry);
1711 0 : dentry = old;
1712 : }
1713 : }
1714 : return dentry;
1715 : }
1716 :
1717 0 : static struct dentry *lookup_slow(const struct qstr *name,
1718 : struct dentry *dir,
1719 : unsigned int flags)
1720 : {
1721 0 : struct inode *inode = dir->d_inode;
1722 : struct dentry *res;
1723 0 : inode_lock_shared(inode);
1724 0 : res = __lookup_slow(name, dir, flags);
1725 0 : inode_unlock_shared(inode);
1726 0 : return res;
1727 : }
1728 :
1729 4 : static inline int may_lookup(struct user_namespace *mnt_userns,
1730 : struct nameidata *nd)
1731 : {
1732 4 : if (nd->flags & LOOKUP_RCU) {
1733 4 : int err = inode_permission(mnt_userns, nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1734 4 : if (err != -ECHILD || !try_to_unlazy(nd))
1735 : return err;
1736 : }
1737 0 : return inode_permission(mnt_userns, nd->inode, MAY_EXEC);
1738 : }
1739 :
1740 0 : static int reserve_stack(struct nameidata *nd, struct path *link, unsigned seq)
1741 : {
1742 0 : if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1743 : return -ELOOP;
1744 :
1745 0 : if (likely(nd->depth != EMBEDDED_LEVELS))
1746 : return 0;
1747 0 : if (likely(nd->stack != nd->internal))
1748 : return 0;
1749 0 : if (likely(nd_alloc_stack(nd)))
1750 : return 0;
1751 :
1752 0 : if (nd->flags & LOOKUP_RCU) {
1753 : // we need to grab link before we do unlazy. And we can't skip
1754 : // unlazy even if we fail to grab the link - cleanup needs it
1755 0 : bool grabbed_link = legitimize_path(nd, link, seq);
1756 :
1757 0 : if (!try_to_unlazy(nd) != 0 || !grabbed_link)
1758 : return -ECHILD;
1759 :
1760 0 : if (nd_alloc_stack(nd))
1761 : return 0;
1762 : }
1763 : return -ENOMEM;
1764 : }
1765 :
1766 : enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1767 :
1768 0 : static const char *pick_link(struct nameidata *nd, struct path *link,
1769 : struct inode *inode, unsigned seq, int flags)
1770 : {
1771 : struct saved *last;
1772 : const char *res;
1773 0 : int error = reserve_stack(nd, link, seq);
1774 :
1775 0 : if (unlikely(error)) {
1776 0 : if (!(nd->flags & LOOKUP_RCU))
1777 : path_put(link);
1778 0 : return ERR_PTR(error);
1779 : }
1780 0 : last = nd->stack + nd->depth++;
1781 0 : last->link = *link;
1782 0 : clear_delayed_call(&last->done);
1783 0 : last->seq = seq;
1784 :
1785 0 : if (flags & WALK_TRAILING) {
1786 0 : error = may_follow_link(nd, inode);
1787 0 : if (unlikely(error))
1788 0 : return ERR_PTR(error);
1789 : }
1790 :
1791 0 : if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) ||
1792 0 : unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW))
1793 : return ERR_PTR(-ELOOP);
1794 :
1795 0 : if (!(nd->flags & LOOKUP_RCU)) {
1796 0 : touch_atime(&last->link);
1797 0 : cond_resched();
1798 0 : } else if (atime_needs_update(&last->link, inode)) {
1799 0 : if (!try_to_unlazy(nd))
1800 : return ERR_PTR(-ECHILD);
1801 0 : touch_atime(&last->link);
1802 : }
1803 :
1804 0 : error = security_inode_follow_link(link->dentry, inode,
1805 0 : nd->flags & LOOKUP_RCU);
1806 : if (unlikely(error))
1807 : return ERR_PTR(error);
1808 :
1809 0 : res = READ_ONCE(inode->i_link);
1810 0 : if (!res) {
1811 : const char * (*get)(struct dentry *, struct inode *,
1812 : struct delayed_call *);
1813 0 : get = inode->i_op->get_link;
1814 0 : if (nd->flags & LOOKUP_RCU) {
1815 0 : res = get(NULL, inode, &last->done);
1816 0 : if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd))
1817 0 : res = get(link->dentry, inode, &last->done);
1818 : } else {
1819 0 : res = get(link->dentry, inode, &last->done);
1820 : }
1821 0 : if (!res)
1822 : goto all_done;
1823 0 : if (IS_ERR(res))
1824 : return res;
1825 : }
1826 0 : if (*res == '/') {
1827 0 : error = nd_jump_root(nd);
1828 0 : if (unlikely(error))
1829 0 : return ERR_PTR(error);
1830 0 : while (unlikely(*++res == '/'))
1831 : ;
1832 : }
1833 0 : if (*res)
1834 : return res;
1835 : all_done: // pure jump
1836 0 : put_link(nd);
1837 0 : return NULL;
1838 : }
1839 :
1840 : /*
1841 : * Do we need to follow links? We _really_ want to be able
1842 : * to do this check without having to look at inode->i_op,
1843 : * so we keep a cache of "no, this doesn't need follow_link"
1844 : * for the common case.
1845 : */
1846 1 : static const char *step_into(struct nameidata *nd, int flags,
1847 : struct dentry *dentry, struct inode *inode, unsigned seq)
1848 : {
1849 : struct path path;
1850 1 : int err = handle_mounts(nd, dentry, &path, &inode, &seq);
1851 :
1852 1 : if (err < 0)
1853 0 : return ERR_PTR(err);
1854 2 : if (likely(!d_is_symlink(path.dentry)) ||
1855 0 : ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
1856 0 : (flags & WALK_NOFOLLOW)) {
1857 : /* not a symlink or should not follow */
1858 1 : if (!(nd->flags & LOOKUP_RCU)) {
1859 0 : dput(nd->path.dentry);
1860 0 : if (nd->path.mnt != path.mnt)
1861 0 : mntput(nd->path.mnt);
1862 : }
1863 1 : nd->path = path;
1864 1 : nd->inode = inode;
1865 1 : nd->seq = seq;
1866 1 : return NULL;
1867 : }
1868 0 : if (nd->flags & LOOKUP_RCU) {
1869 : /* make sure that d_is_symlink above matches inode */
1870 0 : if (read_seqcount_retry(&path.dentry->d_seq, seq))
1871 : return ERR_PTR(-ECHILD);
1872 : } else {
1873 0 : if (path.mnt == nd->path.mnt)
1874 0 : mntget(path.mnt);
1875 : }
1876 0 : return pick_link(nd, &path, inode, seq, flags);
1877 : }
1878 :
1879 0 : static struct dentry *follow_dotdot_rcu(struct nameidata *nd,
1880 : struct inode **inodep,
1881 : unsigned *seqp)
1882 : {
1883 : struct dentry *parent, *old;
1884 :
1885 0 : if (path_equal(&nd->path, &nd->root))
1886 : goto in_root;
1887 0 : if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1888 : struct path path;
1889 : unsigned seq;
1890 0 : if (!choose_mountpoint_rcu(real_mount(nd->path.mnt),
1891 0 : &nd->root, &path, &seq))
1892 : goto in_root;
1893 0 : if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1894 : return ERR_PTR(-ECHILD);
1895 0 : nd->path = path;
1896 0 : nd->inode = path.dentry->d_inode;
1897 0 : nd->seq = seq;
1898 0 : if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1899 : return ERR_PTR(-ECHILD);
1900 : /* we know that mountpoint was pinned */
1901 : }
1902 0 : old = nd->path.dentry;
1903 0 : parent = old->d_parent;
1904 0 : *inodep = parent->d_inode;
1905 0 : *seqp = read_seqcount_begin(&parent->d_seq);
1906 0 : if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1907 : return ERR_PTR(-ECHILD);
1908 0 : if (unlikely(!path_connected(nd->path.mnt, parent)))
1909 : return ERR_PTR(-ECHILD);
1910 0 : return parent;
1911 : in_root:
1912 0 : if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1913 : return ERR_PTR(-ECHILD);
1914 0 : if (unlikely(nd->flags & LOOKUP_BENEATH))
1915 : return ERR_PTR(-ECHILD);
1916 0 : return NULL;
1917 : }
1918 :
1919 0 : static struct dentry *follow_dotdot(struct nameidata *nd,
1920 : struct inode **inodep,
1921 : unsigned *seqp)
1922 : {
1923 : struct dentry *parent;
1924 :
1925 0 : if (path_equal(&nd->path, &nd->root))
1926 : goto in_root;
1927 0 : if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1928 : struct path path;
1929 :
1930 0 : if (!choose_mountpoint(real_mount(nd->path.mnt),
1931 0 : &nd->root, &path))
1932 : goto in_root;
1933 0 : path_put(&nd->path);
1934 0 : nd->path = path;
1935 0 : nd->inode = path.dentry->d_inode;
1936 0 : if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1937 0 : return ERR_PTR(-EXDEV);
1938 : }
1939 : /* rare case of legitimate dget_parent()... */
1940 0 : parent = dget_parent(nd->path.dentry);
1941 0 : if (unlikely(!path_connected(nd->path.mnt, parent))) {
1942 0 : dput(parent);
1943 0 : return ERR_PTR(-ENOENT);
1944 : }
1945 0 : *seqp = 0;
1946 0 : *inodep = parent->d_inode;
1947 0 : return parent;
1948 :
1949 : in_root:
1950 0 : if (unlikely(nd->flags & LOOKUP_BENEATH))
1951 : return ERR_PTR(-EXDEV);
1952 0 : dget(nd->path.dentry);
1953 : return NULL;
1954 : }
1955 :
1956 0 : static const char *handle_dots(struct nameidata *nd, int type)
1957 : {
1958 0 : if (type == LAST_DOTDOT) {
1959 0 : const char *error = NULL;
1960 : struct dentry *parent;
1961 : struct inode *inode;
1962 : unsigned seq;
1963 :
1964 0 : if (!nd->root.mnt) {
1965 0 : error = ERR_PTR(set_root(nd));
1966 0 : if (error)
1967 0 : return error;
1968 : }
1969 0 : if (nd->flags & LOOKUP_RCU)
1970 0 : parent = follow_dotdot_rcu(nd, &inode, &seq);
1971 : else
1972 0 : parent = follow_dotdot(nd, &inode, &seq);
1973 0 : if (IS_ERR(parent))
1974 : return ERR_CAST(parent);
1975 0 : if (unlikely(!parent))
1976 0 : error = step_into(nd, WALK_NOFOLLOW,
1977 : nd->path.dentry, nd->inode, nd->seq);
1978 : else
1979 0 : error = step_into(nd, WALK_NOFOLLOW,
1980 : parent, inode, seq);
1981 0 : if (unlikely(error))
1982 : return error;
1983 :
1984 0 : if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1985 : /*
1986 : * If there was a racing rename or mount along our
1987 : * path, then we can't be sure that ".." hasn't jumped
1988 : * above nd->root (and so userspace should retry or use
1989 : * some fallback).
1990 : */
1991 0 : smp_rmb();
1992 0 : if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq)))
1993 : return ERR_PTR(-EAGAIN);
1994 0 : if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq)))
1995 : return ERR_PTR(-EAGAIN);
1996 : }
1997 : }
1998 : return NULL;
1999 : }
2000 :
2001 1 : static const char *walk_component(struct nameidata *nd, int flags)
2002 : {
2003 : struct dentry *dentry;
2004 : struct inode *inode;
2005 : unsigned seq;
2006 : /*
2007 : * "." and ".." are special - ".." especially so because it has
2008 : * to be able to know about the current root directory and
2009 : * parent relationships.
2010 : */
2011 1 : if (unlikely(nd->last_type != LAST_NORM)) {
2012 0 : if (!(flags & WALK_MORE) && nd->depth)
2013 0 : put_link(nd);
2014 0 : return handle_dots(nd, nd->last_type);
2015 : }
2016 1 : dentry = lookup_fast(nd, &inode, &seq);
2017 1 : if (IS_ERR(dentry))
2018 : return ERR_CAST(dentry);
2019 1 : if (unlikely(!dentry)) {
2020 0 : dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
2021 0 : if (IS_ERR(dentry))
2022 : return ERR_CAST(dentry);
2023 : }
2024 1 : if (!(flags & WALK_MORE) && nd->depth)
2025 0 : put_link(nd);
2026 1 : return step_into(nd, flags, dentry, inode, seq);
2027 : }
2028 :
2029 : /*
2030 : * We can do the critical dentry name comparison and hashing
2031 : * operations one word at a time, but we are limited to:
2032 : *
2033 : * - Architectures with fast unaligned word accesses. We could
2034 : * do a "get_unaligned()" if this helps and is sufficiently
2035 : * fast.
2036 : *
2037 : * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
2038 : * do not trap on the (extremely unlikely) case of a page
2039 : * crossing operation.
2040 : *
2041 : * - Furthermore, we need an efficient 64-bit compile for the
2042 : * 64-bit case in order to generate the "number of bytes in
2043 : * the final mask". Again, that could be replaced with a
2044 : * efficient population count instruction or similar.
2045 : */
2046 : #ifdef CONFIG_DCACHE_WORD_ACCESS
2047 :
2048 : #include <asm/word-at-a-time.h>
2049 :
2050 : #ifdef HASH_MIX
2051 :
2052 : /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
2053 :
2054 : #elif defined(CONFIG_64BIT)
2055 : /*
2056 : * Register pressure in the mixing function is an issue, particularly
2057 : * on 32-bit x86, but almost any function requires one state value and
2058 : * one temporary. Instead, use a function designed for two state values
2059 : * and no temporaries.
2060 : *
2061 : * This function cannot create a collision in only two iterations, so
2062 : * we have two iterations to achieve avalanche. In those two iterations,
2063 : * we have six layers of mixing, which is enough to spread one bit's
2064 : * influence out to 2^6 = 64 state bits.
2065 : *
2066 : * Rotate constants are scored by considering either 64 one-bit input
2067 : * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
2068 : * probability of that delta causing a change to each of the 128 output
2069 : * bits, using a sample of random initial states.
2070 : *
2071 : * The Shannon entropy of the computed probabilities is then summed
2072 : * to produce a score. Ideally, any input change has a 50% chance of
2073 : * toggling any given output bit.
2074 : *
2075 : * Mixing scores (in bits) for (12,45):
2076 : * Input delta: 1-bit 2-bit
2077 : * 1 round: 713.3 42542.6
2078 : * 2 rounds: 2753.7 140389.8
2079 : * 3 rounds: 5954.1 233458.2
2080 : * 4 rounds: 7862.6 256672.2
2081 : * Perfect: 8192 258048
2082 : * (64*128) (64*63/2 * 128)
2083 : */
2084 : #define HASH_MIX(x, y, a) \
2085 : ( x ^= (a), \
2086 : y ^= x, x = rol64(x,12),\
2087 : x += y, y = rol64(y,45),\
2088 : y *= 9 )
2089 :
2090 : /*
2091 : * Fold two longs into one 32-bit hash value. This must be fast, but
2092 : * latency isn't quite as critical, as there is a fair bit of additional
2093 : * work done before the hash value is used.
2094 : */
2095 : static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2096 : {
2097 6 : y ^= x * GOLDEN_RATIO_64;
2098 6 : y *= GOLDEN_RATIO_64;
2099 6 : return y >> 32;
2100 : }
2101 :
2102 : #else /* 32-bit case */
2103 :
2104 : /*
2105 : * Mixing scores (in bits) for (7,20):
2106 : * Input delta: 1-bit 2-bit
2107 : * 1 round: 330.3 9201.6
2108 : * 2 rounds: 1246.4 25475.4
2109 : * 3 rounds: 1907.1 31295.1
2110 : * 4 rounds: 2042.3 31718.6
2111 : * Perfect: 2048 31744
2112 : * (32*64) (32*31/2 * 64)
2113 : */
2114 : #define HASH_MIX(x, y, a) \
2115 : ( x ^= (a), \
2116 : y ^= x, x = rol32(x, 7),\
2117 : x += y, y = rol32(y,20),\
2118 : y *= 9 )
2119 :
2120 : static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2121 : {
2122 : /* Use arch-optimized multiply if one exists */
2123 : return __hash_32(y ^ __hash_32(x));
2124 : }
2125 :
2126 : #endif
2127 :
2128 : /*
2129 : * Return the hash of a string of known length. This is carfully
2130 : * designed to match hash_name(), which is the more critical function.
2131 : * In particular, we must end by hashing a final word containing 0..7
2132 : * payload bytes, to match the way that hash_name() iterates until it
2133 : * finds the delimiter after the name.
2134 : */
2135 0 : unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2136 : {
2137 0 : unsigned long a, x = 0, y = (unsigned long)salt;
2138 :
2139 : for (;;) {
2140 0 : if (!len)
2141 : goto done;
2142 0 : a = load_unaligned_zeropad(name);
2143 0 : if (len < sizeof(unsigned long))
2144 : break;
2145 0 : HASH_MIX(x, y, a);
2146 0 : name += sizeof(unsigned long);
2147 0 : len -= sizeof(unsigned long);
2148 : }
2149 0 : x ^= a & bytemask_from_count(len);
2150 : done:
2151 0 : return fold_hash(x, y);
2152 : }
2153 : EXPORT_SYMBOL(full_name_hash);
2154 :
2155 : /* Return the "hash_len" (hash and length) of a null-terminated string */
2156 2 : u64 hashlen_string(const void *salt, const char *name)
2157 : {
2158 2 : unsigned long a = 0, x = 0, y = (unsigned long)salt;
2159 : unsigned long adata, mask, len;
2160 2 : const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2161 :
2162 2 : len = 0;
2163 2 : goto inside;
2164 :
2165 : do {
2166 3 : HASH_MIX(x, y, a);
2167 1 : len += sizeof(unsigned long);
2168 : inside:
2169 6 : a = load_unaligned_zeropad(name+len);
2170 3 : } while (!has_zero(a, &adata, &constants));
2171 :
2172 2 : adata = prep_zero_mask(a, adata, &constants);
2173 4 : mask = create_zero_mask(adata);
2174 2 : x ^= a & zero_bytemask(mask);
2175 :
2176 4 : return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2177 : }
2178 : EXPORT_SYMBOL(hashlen_string);
2179 :
2180 : /*
2181 : * Calculate the length and hash of the path component, and
2182 : * return the "hash_len" as the result.
2183 : */
2184 4 : static inline u64 hash_name(const void *salt, const char *name)
2185 : {
2186 4 : unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2187 : unsigned long adata, bdata, mask, len;
2188 4 : const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2189 :
2190 4 : len = 0;
2191 4 : goto inside;
2192 :
2193 : do {
2194 0 : HASH_MIX(x, y, a);
2195 0 : len += sizeof(unsigned long);
2196 : inside:
2197 8 : a = load_unaligned_zeropad(name+len);
2198 4 : b = a ^ REPEAT_BYTE('/');
2199 8 : } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2200 :
2201 4 : adata = prep_zero_mask(a, adata, &constants);
2202 4 : bdata = prep_zero_mask(b, bdata, &constants);
2203 8 : mask = create_zero_mask(adata | bdata);
2204 4 : x ^= a & zero_bytemask(mask);
2205 :
2206 8 : return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2207 : }
2208 :
2209 : #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2210 :
2211 : /* Return the hash of a string of known length */
2212 : unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2213 : {
2214 : unsigned long hash = init_name_hash(salt);
2215 : while (len--)
2216 : hash = partial_name_hash((unsigned char)*name++, hash);
2217 : return end_name_hash(hash);
2218 : }
2219 : EXPORT_SYMBOL(full_name_hash);
2220 :
2221 : /* Return the "hash_len" (hash and length) of a null-terminated string */
2222 : u64 hashlen_string(const void *salt, const char *name)
2223 : {
2224 : unsigned long hash = init_name_hash(salt);
2225 : unsigned long len = 0, c;
2226 :
2227 : c = (unsigned char)*name;
2228 : while (c) {
2229 : len++;
2230 : hash = partial_name_hash(c, hash);
2231 : c = (unsigned char)name[len];
2232 : }
2233 : return hashlen_create(end_name_hash(hash), len);
2234 : }
2235 : EXPORT_SYMBOL(hashlen_string);
2236 :
2237 : /*
2238 : * We know there's a real path component here of at least
2239 : * one character.
2240 : */
2241 : static inline u64 hash_name(const void *salt, const char *name)
2242 : {
2243 : unsigned long hash = init_name_hash(salt);
2244 : unsigned long len = 0, c;
2245 :
2246 : c = (unsigned char)*name;
2247 : do {
2248 : len++;
2249 : hash = partial_name_hash(c, hash);
2250 : c = (unsigned char)name[len];
2251 : } while (c && c != '/');
2252 : return hashlen_create(end_name_hash(hash), len);
2253 : }
2254 :
2255 : #endif
2256 :
2257 : /*
2258 : * Name resolution.
2259 : * This is the basic name resolution function, turning a pathname into
2260 : * the final dentry. We expect 'base' to be positive and a directory.
2261 : *
2262 : * Returns 0 and nd will have valid dentry and mnt on success.
2263 : * Returns error and drops reference to input namei data on failure.
2264 : */
2265 3 : static int link_path_walk(const char *name, struct nameidata *nd)
2266 : {
2267 3 : int depth = 0; // depth <= nd->depth
2268 : int err;
2269 :
2270 3 : nd->last_type = LAST_ROOT;
2271 3 : nd->flags |= LOOKUP_PARENT;
2272 3 : if (IS_ERR(name))
2273 0 : return PTR_ERR(name);
2274 6 : while (*name=='/')
2275 3 : name++;
2276 3 : if (!*name) {
2277 0 : nd->dir_mode = 0; // short-circuit the 'hardening' idiocy
2278 0 : return 0;
2279 : }
2280 :
2281 : /* At this point we know we have a real path component. */
2282 : for(;;) {
2283 : struct user_namespace *mnt_userns;
2284 : const char *link;
2285 : u64 hash_len;
2286 : int type;
2287 :
2288 8 : mnt_userns = mnt_user_ns(nd->path.mnt);
2289 4 : err = may_lookup(mnt_userns, nd);
2290 4 : if (err)
2291 : return err;
2292 :
2293 4 : hash_len = hash_name(nd->path.dentry, name);
2294 :
2295 4 : type = LAST_NORM;
2296 4 : if (name[0] == '.') switch (hashlen_len(hash_len)) {
2297 : case 2:
2298 0 : if (name[1] == '.') {
2299 0 : type = LAST_DOTDOT;
2300 0 : nd->state |= ND_JUMPED;
2301 : }
2302 : break;
2303 : case 1:
2304 0 : type = LAST_DOT;
2305 : }
2306 4 : if (likely(type == LAST_NORM)) {
2307 4 : struct dentry *parent = nd->path.dentry;
2308 4 : nd->state &= ~ND_JUMPED;
2309 4 : if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2310 0 : struct qstr this = { { .hash_len = hash_len }, .name = name };
2311 0 : err = parent->d_op->d_hash(parent, &this);
2312 0 : if (err < 0)
2313 0 : return err;
2314 0 : hash_len = this.hash_len;
2315 0 : name = this.name;
2316 : }
2317 : }
2318 :
2319 4 : nd->last.hash_len = hash_len;
2320 4 : nd->last.name = name;
2321 4 : nd->last_type = type;
2322 :
2323 4 : name += hashlen_len(hash_len);
2324 4 : if (!*name)
2325 : goto OK;
2326 : /*
2327 : * If it wasn't NUL, we know it was '/'. Skip that
2328 : * slash, and continue until no more slashes.
2329 : */
2330 : do {
2331 1 : name++;
2332 1 : } while (unlikely(*name == '/'));
2333 1 : if (unlikely(!*name)) {
2334 : OK:
2335 : /* pathname or trailing symlink, done */
2336 3 : if (!depth) {
2337 6 : nd->dir_uid = i_uid_into_mnt(mnt_userns, nd->inode);
2338 3 : nd->dir_mode = nd->inode->i_mode;
2339 3 : nd->flags &= ~LOOKUP_PARENT;
2340 3 : return 0;
2341 : }
2342 : /* last component of nested symlink */
2343 0 : name = nd->stack[--depth].name;
2344 0 : link = walk_component(nd, 0);
2345 : } else {
2346 : /* not the last component */
2347 1 : link = walk_component(nd, WALK_MORE);
2348 : }
2349 1 : if (unlikely(link)) {
2350 0 : if (IS_ERR(link))
2351 0 : return PTR_ERR(link);
2352 : /* a symlink to follow */
2353 0 : nd->stack[depth++].name = name;
2354 0 : name = link;
2355 0 : continue;
2356 : }
2357 2 : if (unlikely(!d_can_lookup(nd->path.dentry))) {
2358 0 : if (nd->flags & LOOKUP_RCU) {
2359 0 : if (!try_to_unlazy(nd))
2360 : return -ECHILD;
2361 : }
2362 : return -ENOTDIR;
2363 : }
2364 : }
2365 : }
2366 :
2367 : /* must be paired with terminate_walk() */
2368 3 : static const char *path_init(struct nameidata *nd, unsigned flags)
2369 : {
2370 : int error;
2371 3 : const char *s = nd->name->name;
2372 :
2373 : /* LOOKUP_CACHED requires RCU, ask caller to retry */
2374 3 : if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED)
2375 : return ERR_PTR(-EAGAIN);
2376 :
2377 3 : if (!*s)
2378 0 : flags &= ~LOOKUP_RCU;
2379 3 : if (flags & LOOKUP_RCU)
2380 : rcu_read_lock();
2381 :
2382 3 : nd->flags = flags;
2383 3 : nd->state |= ND_JUMPED;
2384 :
2385 6 : nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2386 6 : nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2387 3 : smp_rmb();
2388 :
2389 3 : if (nd->state & ND_ROOT_PRESET) {
2390 0 : struct dentry *root = nd->root.dentry;
2391 0 : struct inode *inode = root->d_inode;
2392 0 : if (*s && unlikely(!d_can_lookup(root)))
2393 : return ERR_PTR(-ENOTDIR);
2394 0 : nd->path = nd->root;
2395 0 : nd->inode = inode;
2396 0 : if (flags & LOOKUP_RCU) {
2397 0 : nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2398 0 : nd->root_seq = nd->seq;
2399 : } else {
2400 0 : path_get(&nd->path);
2401 : }
2402 : return s;
2403 : }
2404 :
2405 3 : nd->root.mnt = NULL;
2406 :
2407 : /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2408 3 : if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2409 3 : error = nd_jump_root(nd);
2410 3 : if (unlikely(error))
2411 0 : return ERR_PTR(error);
2412 : return s;
2413 : }
2414 :
2415 : /* Relative pathname -- get the starting-point it is relative to. */
2416 0 : if (nd->dfd == AT_FDCWD) {
2417 0 : if (flags & LOOKUP_RCU) {
2418 0 : struct fs_struct *fs = current->fs;
2419 : unsigned seq;
2420 :
2421 : do {
2422 0 : seq = read_seqcount_begin(&fs->seq);
2423 0 : nd->path = fs->pwd;
2424 0 : nd->inode = nd->path.dentry->d_inode;
2425 0 : nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2426 0 : } while (read_seqcount_retry(&fs->seq, seq));
2427 : } else {
2428 0 : get_fs_pwd(current->fs, &nd->path);
2429 0 : nd->inode = nd->path.dentry->d_inode;
2430 : }
2431 : } else {
2432 : /* Caller must check execute permissions on the starting path component */
2433 0 : struct fd f = fdget_raw(nd->dfd);
2434 : struct dentry *dentry;
2435 :
2436 0 : if (!f.file)
2437 0 : return ERR_PTR(-EBADF);
2438 :
2439 0 : dentry = f.file->f_path.dentry;
2440 :
2441 0 : if (*s && unlikely(!d_can_lookup(dentry))) {
2442 0 : fdput(f);
2443 : return ERR_PTR(-ENOTDIR);
2444 : }
2445 :
2446 0 : nd->path = f.file->f_path;
2447 0 : if (flags & LOOKUP_RCU) {
2448 0 : nd->inode = nd->path.dentry->d_inode;
2449 0 : nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2450 : } else {
2451 0 : path_get(&nd->path);
2452 0 : nd->inode = nd->path.dentry->d_inode;
2453 : }
2454 0 : fdput(f);
2455 : }
2456 :
2457 : /* For scoped-lookups we need to set the root to the dirfd as well. */
2458 0 : if (flags & LOOKUP_IS_SCOPED) {
2459 0 : nd->root = nd->path;
2460 0 : if (flags & LOOKUP_RCU) {
2461 0 : nd->root_seq = nd->seq;
2462 : } else {
2463 0 : path_get(&nd->root);
2464 0 : nd->state |= ND_ROOT_GRABBED;
2465 : }
2466 : }
2467 : return s;
2468 : }
2469 :
2470 0 : static inline const char *lookup_last(struct nameidata *nd)
2471 : {
2472 0 : if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2473 0 : nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2474 :
2475 0 : return walk_component(nd, WALK_TRAILING);
2476 : }
2477 :
2478 0 : static int handle_lookup_down(struct nameidata *nd)
2479 : {
2480 0 : if (!(nd->flags & LOOKUP_RCU))
2481 0 : dget(nd->path.dentry);
2482 0 : return PTR_ERR(step_into(nd, WALK_NOFOLLOW,
2483 : nd->path.dentry, nd->inode, nd->seq));
2484 : }
2485 :
2486 : /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2487 0 : static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2488 : {
2489 0 : const char *s = path_init(nd, flags);
2490 : int err;
2491 :
2492 0 : if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2493 0 : err = handle_lookup_down(nd);
2494 0 : if (unlikely(err < 0))
2495 0 : s = ERR_PTR(err);
2496 : }
2497 :
2498 0 : while (!(err = link_path_walk(s, nd)) &&
2499 : (s = lookup_last(nd)) != NULL)
2500 : ;
2501 0 : if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2502 0 : err = handle_lookup_down(nd);
2503 0 : nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please...
2504 : }
2505 0 : if (!err)
2506 0 : err = complete_walk(nd);
2507 :
2508 0 : if (!err && nd->flags & LOOKUP_DIRECTORY)
2509 0 : if (!d_can_lookup(nd->path.dentry))
2510 0 : err = -ENOTDIR;
2511 0 : if (!err) {
2512 0 : *path = nd->path;
2513 0 : nd->path.mnt = NULL;
2514 0 : nd->path.dentry = NULL;
2515 : }
2516 0 : terminate_walk(nd);
2517 0 : return err;
2518 : }
2519 :
2520 0 : int filename_lookup(int dfd, struct filename *name, unsigned flags,
2521 : struct path *path, struct path *root)
2522 : {
2523 : int retval;
2524 : struct nameidata nd;
2525 0 : if (IS_ERR(name))
2526 0 : return PTR_ERR(name);
2527 0 : set_nameidata(&nd, dfd, name, root);
2528 0 : retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2529 0 : if (unlikely(retval == -ECHILD))
2530 0 : retval = path_lookupat(&nd, flags, path);
2531 0 : if (unlikely(retval == -ESTALE))
2532 0 : retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2533 :
2534 : if (likely(!retval))
2535 : audit_inode(name, path->dentry,
2536 : flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2537 0 : restore_nameidata();
2538 0 : return retval;
2539 : }
2540 :
2541 : /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2542 3 : static int path_parentat(struct nameidata *nd, unsigned flags,
2543 : struct path *parent)
2544 : {
2545 3 : const char *s = path_init(nd, flags);
2546 3 : int err = link_path_walk(s, nd);
2547 3 : if (!err)
2548 3 : err = complete_walk(nd);
2549 3 : if (!err) {
2550 3 : *parent = nd->path;
2551 3 : nd->path.mnt = NULL;
2552 3 : nd->path.dentry = NULL;
2553 : }
2554 3 : terminate_walk(nd);
2555 3 : return err;
2556 : }
2557 :
2558 : /* Note: this does not consume "name" */
2559 3 : static int filename_parentat(int dfd, struct filename *name,
2560 : unsigned int flags, struct path *parent,
2561 : struct qstr *last, int *type)
2562 : {
2563 : int retval;
2564 : struct nameidata nd;
2565 :
2566 3 : if (IS_ERR(name))
2567 0 : return PTR_ERR(name);
2568 3 : set_nameidata(&nd, dfd, name, NULL);
2569 3 : retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2570 3 : if (unlikely(retval == -ECHILD))
2571 0 : retval = path_parentat(&nd, flags, parent);
2572 3 : if (unlikely(retval == -ESTALE))
2573 0 : retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2574 3 : if (likely(!retval)) {
2575 3 : *last = nd.last;
2576 3 : *type = nd.last_type;
2577 3 : audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2578 : }
2579 3 : restore_nameidata();
2580 3 : return retval;
2581 : }
2582 :
2583 : /* does lookup, returns the object with parent locked */
2584 0 : static struct dentry *__kern_path_locked(struct filename *name, struct path *path)
2585 : {
2586 : struct dentry *d;
2587 : struct qstr last;
2588 : int type, error;
2589 :
2590 0 : error = filename_parentat(AT_FDCWD, name, 0, path, &last, &type);
2591 0 : if (error)
2592 0 : return ERR_PTR(error);
2593 0 : if (unlikely(type != LAST_NORM)) {
2594 0 : path_put(path);
2595 0 : return ERR_PTR(-EINVAL);
2596 : }
2597 0 : inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2598 0 : d = __lookup_hash(&last, path->dentry, 0);
2599 0 : if (IS_ERR(d)) {
2600 0 : inode_unlock(path->dentry->d_inode);
2601 : path_put(path);
2602 : }
2603 : return d;
2604 : }
2605 :
2606 0 : struct dentry *kern_path_locked(const char *name, struct path *path)
2607 : {
2608 0 : struct filename *filename = getname_kernel(name);
2609 0 : struct dentry *res = __kern_path_locked(filename, path);
2610 :
2611 0 : putname(filename);
2612 0 : return res;
2613 : }
2614 :
2615 0 : int kern_path(const char *name, unsigned int flags, struct path *path)
2616 : {
2617 0 : struct filename *filename = getname_kernel(name);
2618 0 : int ret = filename_lookup(AT_FDCWD, filename, flags, path, NULL);
2619 :
2620 0 : putname(filename);
2621 0 : return ret;
2622 :
2623 : }
2624 : EXPORT_SYMBOL(kern_path);
2625 :
2626 : /**
2627 : * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2628 : * @dentry: pointer to dentry of the base directory
2629 : * @mnt: pointer to vfs mount of the base directory
2630 : * @name: pointer to file name
2631 : * @flags: lookup flags
2632 : * @path: pointer to struct path to fill
2633 : */
2634 0 : int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2635 : const char *name, unsigned int flags,
2636 : struct path *path)
2637 : {
2638 : struct filename *filename;
2639 0 : struct path root = {.mnt = mnt, .dentry = dentry};
2640 : int ret;
2641 :
2642 0 : filename = getname_kernel(name);
2643 : /* the first argument of filename_lookup() is ignored with root */
2644 0 : ret = filename_lookup(AT_FDCWD, filename, flags, path, &root);
2645 0 : putname(filename);
2646 0 : return ret;
2647 : }
2648 : EXPORT_SYMBOL(vfs_path_lookup);
2649 :
2650 0 : static int lookup_one_common(struct user_namespace *mnt_userns,
2651 : const char *name, struct dentry *base, int len,
2652 : struct qstr *this)
2653 : {
2654 0 : this->name = name;
2655 0 : this->len = len;
2656 0 : this->hash = full_name_hash(base, name, len);
2657 0 : if (!len)
2658 : return -EACCES;
2659 :
2660 0 : if (unlikely(name[0] == '.')) {
2661 0 : if (len < 2 || (len == 2 && name[1] == '.'))
2662 : return -EACCES;
2663 : }
2664 :
2665 0 : while (len--) {
2666 0 : unsigned int c = *(const unsigned char *)name++;
2667 0 : if (c == '/' || c == '\0')
2668 : return -EACCES;
2669 : }
2670 : /*
2671 : * See if the low-level filesystem might want
2672 : * to use its own hash..
2673 : */
2674 0 : if (base->d_flags & DCACHE_OP_HASH) {
2675 0 : int err = base->d_op->d_hash(base, this);
2676 0 : if (err < 0)
2677 : return err;
2678 : }
2679 :
2680 0 : return inode_permission(mnt_userns, base->d_inode, MAY_EXEC);
2681 : }
2682 :
2683 : /**
2684 : * try_lookup_one_len - filesystem helper to lookup single pathname component
2685 : * @name: pathname component to lookup
2686 : * @base: base directory to lookup from
2687 : * @len: maximum length @len should be interpreted to
2688 : *
2689 : * Look up a dentry by name in the dcache, returning NULL if it does not
2690 : * currently exist. The function does not try to create a dentry.
2691 : *
2692 : * Note that this routine is purely a helper for filesystem usage and should
2693 : * not be called by generic code.
2694 : *
2695 : * The caller must hold base->i_mutex.
2696 : */
2697 0 : struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2698 : {
2699 : struct qstr this;
2700 : int err;
2701 :
2702 0 : WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2703 :
2704 0 : err = lookup_one_common(&init_user_ns, name, base, len, &this);
2705 0 : if (err)
2706 0 : return ERR_PTR(err);
2707 :
2708 0 : return lookup_dcache(&this, base, 0);
2709 : }
2710 : EXPORT_SYMBOL(try_lookup_one_len);
2711 :
2712 : /**
2713 : * lookup_one_len - filesystem helper to lookup single pathname component
2714 : * @name: pathname component to lookup
2715 : * @base: base directory to lookup from
2716 : * @len: maximum length @len should be interpreted to
2717 : *
2718 : * Note that this routine is purely a helper for filesystem usage and should
2719 : * not be called by generic code.
2720 : *
2721 : * The caller must hold base->i_mutex.
2722 : */
2723 0 : struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2724 : {
2725 : struct dentry *dentry;
2726 : struct qstr this;
2727 : int err;
2728 :
2729 0 : WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2730 :
2731 0 : err = lookup_one_common(&init_user_ns, name, base, len, &this);
2732 0 : if (err)
2733 0 : return ERR_PTR(err);
2734 :
2735 0 : dentry = lookup_dcache(&this, base, 0);
2736 0 : return dentry ? dentry : __lookup_slow(&this, base, 0);
2737 : }
2738 : EXPORT_SYMBOL(lookup_one_len);
2739 :
2740 : /**
2741 : * lookup_one - filesystem helper to lookup single pathname component
2742 : * @mnt_userns: user namespace of the mount the lookup is performed from
2743 : * @name: pathname component to lookup
2744 : * @base: base directory to lookup from
2745 : * @len: maximum length @len should be interpreted to
2746 : *
2747 : * Note that this routine is purely a helper for filesystem usage and should
2748 : * not be called by generic code.
2749 : *
2750 : * The caller must hold base->i_mutex.
2751 : */
2752 0 : struct dentry *lookup_one(struct user_namespace *mnt_userns, const char *name,
2753 : struct dentry *base, int len)
2754 : {
2755 : struct dentry *dentry;
2756 : struct qstr this;
2757 : int err;
2758 :
2759 0 : WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2760 :
2761 0 : err = lookup_one_common(mnt_userns, name, base, len, &this);
2762 0 : if (err)
2763 0 : return ERR_PTR(err);
2764 :
2765 0 : dentry = lookup_dcache(&this, base, 0);
2766 0 : return dentry ? dentry : __lookup_slow(&this, base, 0);
2767 : }
2768 : EXPORT_SYMBOL(lookup_one);
2769 :
2770 : /**
2771 : * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2772 : * @name: pathname component to lookup
2773 : * @base: base directory to lookup from
2774 : * @len: maximum length @len should be interpreted to
2775 : *
2776 : * Note that this routine is purely a helper for filesystem usage and should
2777 : * not be called by generic code.
2778 : *
2779 : * Unlike lookup_one_len, it should be called without the parent
2780 : * i_mutex held, and will take the i_mutex itself if necessary.
2781 : */
2782 0 : struct dentry *lookup_one_len_unlocked(const char *name,
2783 : struct dentry *base, int len)
2784 : {
2785 : struct qstr this;
2786 : int err;
2787 : struct dentry *ret;
2788 :
2789 0 : err = lookup_one_common(&init_user_ns, name, base, len, &this);
2790 0 : if (err)
2791 0 : return ERR_PTR(err);
2792 :
2793 0 : ret = lookup_dcache(&this, base, 0);
2794 0 : if (!ret)
2795 0 : ret = lookup_slow(&this, base, 0);
2796 : return ret;
2797 : }
2798 : EXPORT_SYMBOL(lookup_one_len_unlocked);
2799 :
2800 : /*
2801 : * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2802 : * on negatives. Returns known positive or ERR_PTR(); that's what
2803 : * most of the users want. Note that pinned negative with unlocked parent
2804 : * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2805 : * need to be very careful; pinned positives have ->d_inode stable, so
2806 : * this one avoids such problems.
2807 : */
2808 0 : struct dentry *lookup_positive_unlocked(const char *name,
2809 : struct dentry *base, int len)
2810 : {
2811 0 : struct dentry *ret = lookup_one_len_unlocked(name, base, len);
2812 0 : if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2813 0 : dput(ret);
2814 0 : ret = ERR_PTR(-ENOENT);
2815 : }
2816 0 : return ret;
2817 : }
2818 : EXPORT_SYMBOL(lookup_positive_unlocked);
2819 :
2820 : #ifdef CONFIG_UNIX98_PTYS
2821 0 : int path_pts(struct path *path)
2822 : {
2823 : /* Find something mounted on "pts" in the same directory as
2824 : * the input path.
2825 : */
2826 0 : struct dentry *parent = dget_parent(path->dentry);
2827 : struct dentry *child;
2828 0 : struct qstr this = QSTR_INIT("pts", 3);
2829 :
2830 0 : if (unlikely(!path_connected(path->mnt, parent))) {
2831 0 : dput(parent);
2832 0 : return -ENOENT;
2833 : }
2834 0 : dput(path->dentry);
2835 0 : path->dentry = parent;
2836 0 : child = d_hash_and_lookup(parent, &this);
2837 0 : if (!child)
2838 : return -ENOENT;
2839 :
2840 0 : path->dentry = child;
2841 0 : dput(parent);
2842 0 : follow_down(path);
2843 0 : return 0;
2844 : }
2845 : #endif
2846 :
2847 0 : int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2848 : struct path *path, int *empty)
2849 : {
2850 0 : struct filename *filename = getname_flags(name, flags, empty);
2851 0 : int ret = filename_lookup(dfd, filename, flags, path, NULL);
2852 :
2853 0 : putname(filename);
2854 0 : return ret;
2855 : }
2856 : EXPORT_SYMBOL(user_path_at_empty);
2857 :
2858 0 : int __check_sticky(struct user_namespace *mnt_userns, struct inode *dir,
2859 : struct inode *inode)
2860 : {
2861 0 : kuid_t fsuid = current_fsuid();
2862 :
2863 0 : if (uid_eq(i_uid_into_mnt(mnt_userns, inode), fsuid))
2864 : return 0;
2865 0 : if (uid_eq(i_uid_into_mnt(mnt_userns, dir), fsuid))
2866 : return 0;
2867 0 : return !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FOWNER);
2868 : }
2869 : EXPORT_SYMBOL(__check_sticky);
2870 :
2871 : /*
2872 : * Check whether we can remove a link victim from directory dir, check
2873 : * whether the type of victim is right.
2874 : * 1. We can't do it if dir is read-only (done in permission())
2875 : * 2. We should have write and exec permissions on dir
2876 : * 3. We can't remove anything from append-only dir
2877 : * 4. We can't do anything with immutable dir (done in permission())
2878 : * 5. If the sticky bit on dir is set we should either
2879 : * a. be owner of dir, or
2880 : * b. be owner of victim, or
2881 : * c. have CAP_FOWNER capability
2882 : * 6. If the victim is append-only or immutable we can't do antyhing with
2883 : * links pointing to it.
2884 : * 7. If the victim has an unknown uid or gid we can't change the inode.
2885 : * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2886 : * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2887 : * 10. We can't remove a root or mountpoint.
2888 : * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2889 : * nfs_async_unlink().
2890 : */
2891 0 : static int may_delete(struct user_namespace *mnt_userns, struct inode *dir,
2892 : struct dentry *victim, bool isdir)
2893 : {
2894 0 : struct inode *inode = d_backing_inode(victim);
2895 : int error;
2896 :
2897 0 : if (d_is_negative(victim))
2898 : return -ENOENT;
2899 0 : BUG_ON(!inode);
2900 :
2901 0 : BUG_ON(victim->d_parent->d_inode != dir);
2902 :
2903 : /* Inode writeback is not safe when the uid or gid are invalid. */
2904 0 : if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) ||
2905 0 : !gid_valid(i_gid_into_mnt(mnt_userns, inode)))
2906 : return -EOVERFLOW;
2907 :
2908 0 : audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2909 :
2910 0 : error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
2911 0 : if (error)
2912 : return error;
2913 0 : if (IS_APPEND(dir))
2914 : return -EPERM;
2915 :
2916 0 : if (check_sticky(mnt_userns, dir, inode) || IS_APPEND(inode) ||
2917 0 : IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) ||
2918 0 : HAS_UNMAPPED_ID(mnt_userns, inode))
2919 : return -EPERM;
2920 0 : if (isdir) {
2921 0 : if (!d_is_dir(victim))
2922 : return -ENOTDIR;
2923 0 : if (IS_ROOT(victim))
2924 : return -EBUSY;
2925 0 : } else if (d_is_dir(victim))
2926 : return -EISDIR;
2927 0 : if (IS_DEADDIR(dir))
2928 : return -ENOENT;
2929 0 : if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2930 : return -EBUSY;
2931 0 : return 0;
2932 : }
2933 :
2934 : /* Check whether we can create an object with dentry child in directory
2935 : * dir.
2936 : * 1. We can't do it if child already exists (open has special treatment for
2937 : * this case, but since we are inlined it's OK)
2938 : * 2. We can't do it if dir is read-only (done in permission())
2939 : * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2940 : * 4. We should have write and exec permissions on dir
2941 : * 5. We can't do it if dir is immutable (done in permission())
2942 : */
2943 3 : static inline int may_create(struct user_namespace *mnt_userns,
2944 : struct inode *dir, struct dentry *child)
2945 : {
2946 3 : audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2947 3 : if (child->d_inode)
2948 : return -EEXIST;
2949 3 : if (IS_DEADDIR(dir))
2950 : return -ENOENT;
2951 3 : if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns))
2952 : return -EOVERFLOW;
2953 :
2954 3 : return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
2955 : }
2956 :
2957 : /*
2958 : * p1 and p2 should be directories on the same fs.
2959 : */
2960 0 : struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2961 : {
2962 : struct dentry *p;
2963 :
2964 0 : if (p1 == p2) {
2965 0 : inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2966 0 : return NULL;
2967 : }
2968 :
2969 0 : mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2970 :
2971 0 : p = d_ancestor(p2, p1);
2972 0 : if (p) {
2973 0 : inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2974 0 : inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2975 0 : return p;
2976 : }
2977 :
2978 0 : p = d_ancestor(p1, p2);
2979 0 : if (p) {
2980 0 : inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2981 0 : inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2982 0 : return p;
2983 : }
2984 :
2985 0 : inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2986 0 : inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2987 0 : return NULL;
2988 : }
2989 : EXPORT_SYMBOL(lock_rename);
2990 :
2991 0 : void unlock_rename(struct dentry *p1, struct dentry *p2)
2992 : {
2993 0 : inode_unlock(p1->d_inode);
2994 0 : if (p1 != p2) {
2995 0 : inode_unlock(p2->d_inode);
2996 0 : mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2997 : }
2998 0 : }
2999 : EXPORT_SYMBOL(unlock_rename);
3000 :
3001 : /**
3002 : * vfs_create - create new file
3003 : * @mnt_userns: user namespace of the mount the inode was found from
3004 : * @dir: inode of @dentry
3005 : * @dentry: pointer to dentry of the base directory
3006 : * @mode: mode of the new file
3007 : * @want_excl: whether the file must not yet exist
3008 : *
3009 : * Create a new file.
3010 : *
3011 : * If the inode has been found through an idmapped mount the user namespace of
3012 : * the vfsmount must be passed through @mnt_userns. This function will then take
3013 : * care to map the inode according to @mnt_userns before checking permissions.
3014 : * On non-idmapped mounts or if permission checking is to be performed on the
3015 : * raw inode simply passs init_user_ns.
3016 : */
3017 0 : int vfs_create(struct user_namespace *mnt_userns, struct inode *dir,
3018 : struct dentry *dentry, umode_t mode, bool want_excl)
3019 : {
3020 0 : int error = may_create(mnt_userns, dir, dentry);
3021 0 : if (error)
3022 : return error;
3023 :
3024 0 : if (!dir->i_op->create)
3025 : return -EACCES; /* shouldn't it be ENOSYS? */
3026 0 : mode &= S_IALLUGO;
3027 0 : mode |= S_IFREG;
3028 0 : error = security_inode_create(dir, dentry, mode);
3029 : if (error)
3030 : return error;
3031 0 : error = dir->i_op->create(mnt_userns, dir, dentry, mode, want_excl);
3032 0 : if (!error)
3033 : fsnotify_create(dir, dentry);
3034 : return error;
3035 : }
3036 : EXPORT_SYMBOL(vfs_create);
3037 :
3038 0 : int vfs_mkobj(struct dentry *dentry, umode_t mode,
3039 : int (*f)(struct dentry *, umode_t, void *),
3040 : void *arg)
3041 : {
3042 0 : struct inode *dir = dentry->d_parent->d_inode;
3043 0 : int error = may_create(&init_user_ns, dir, dentry);
3044 0 : if (error)
3045 : return error;
3046 :
3047 0 : mode &= S_IALLUGO;
3048 0 : mode |= S_IFREG;
3049 0 : error = security_inode_create(dir, dentry, mode);
3050 : if (error)
3051 : return error;
3052 0 : error = f(dentry, mode, arg);
3053 0 : if (!error)
3054 : fsnotify_create(dir, dentry);
3055 : return error;
3056 : }
3057 : EXPORT_SYMBOL(vfs_mkobj);
3058 :
3059 0 : bool may_open_dev(const struct path *path)
3060 : {
3061 0 : return !(path->mnt->mnt_flags & MNT_NODEV) &&
3062 0 : !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
3063 : }
3064 :
3065 0 : static int may_open(struct user_namespace *mnt_userns, const struct path *path,
3066 : int acc_mode, int flag)
3067 : {
3068 0 : struct dentry *dentry = path->dentry;
3069 0 : struct inode *inode = dentry->d_inode;
3070 : int error;
3071 :
3072 0 : if (!inode)
3073 : return -ENOENT;
3074 :
3075 0 : switch (inode->i_mode & S_IFMT) {
3076 : case S_IFLNK:
3077 : return -ELOOP;
3078 : case S_IFDIR:
3079 0 : if (acc_mode & MAY_WRITE)
3080 : return -EISDIR;
3081 0 : if (acc_mode & MAY_EXEC)
3082 : return -EACCES;
3083 : break;
3084 : case S_IFBLK:
3085 : case S_IFCHR:
3086 0 : if (!may_open_dev(path))
3087 : return -EACCES;
3088 : fallthrough;
3089 : case S_IFIFO:
3090 : case S_IFSOCK:
3091 0 : if (acc_mode & MAY_EXEC)
3092 : return -EACCES;
3093 0 : flag &= ~O_TRUNC;
3094 0 : break;
3095 : case S_IFREG:
3096 0 : if ((acc_mode & MAY_EXEC) && path_noexec(path))
3097 : return -EACCES;
3098 : break;
3099 : }
3100 :
3101 0 : error = inode_permission(mnt_userns, inode, MAY_OPEN | acc_mode);
3102 0 : if (error)
3103 : return error;
3104 :
3105 : /*
3106 : * An append-only file must be opened in append mode for writing.
3107 : */
3108 0 : if (IS_APPEND(inode)) {
3109 0 : if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
3110 : return -EPERM;
3111 0 : if (flag & O_TRUNC)
3112 : return -EPERM;
3113 : }
3114 :
3115 : /* O_NOATIME can only be set by the owner or superuser */
3116 0 : if (flag & O_NOATIME && !inode_owner_or_capable(mnt_userns, inode))
3117 : return -EPERM;
3118 :
3119 : return 0;
3120 : }
3121 :
3122 0 : static int handle_truncate(struct user_namespace *mnt_userns, struct file *filp)
3123 : {
3124 0 : const struct path *path = &filp->f_path;
3125 0 : struct inode *inode = path->dentry->d_inode;
3126 0 : int error = get_write_access(inode);
3127 0 : if (error)
3128 : return error;
3129 :
3130 0 : error = security_path_truncate(path);
3131 : if (!error) {
3132 0 : error = do_truncate(mnt_userns, path->dentry, 0,
3133 : ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3134 : filp);
3135 : }
3136 0 : put_write_access(inode);
3137 0 : return error;
3138 : }
3139 :
3140 : static inline int open_to_namei_flags(int flag)
3141 : {
3142 0 : if ((flag & O_ACCMODE) == 3)
3143 0 : flag--;
3144 : return flag;
3145 : }
3146 :
3147 0 : static int may_o_create(struct user_namespace *mnt_userns,
3148 : const struct path *dir, struct dentry *dentry,
3149 : umode_t mode)
3150 : {
3151 0 : int error = security_path_mknod(dir, dentry, mode, 0);
3152 : if (error)
3153 : return error;
3154 :
3155 0 : if (!fsuidgid_has_mapping(dir->dentry->d_sb, mnt_userns))
3156 : return -EOVERFLOW;
3157 :
3158 0 : error = inode_permission(mnt_userns, dir->dentry->d_inode,
3159 : MAY_WRITE | MAY_EXEC);
3160 0 : if (error)
3161 : return error;
3162 :
3163 0 : return security_inode_create(dir->dentry->d_inode, dentry, mode);
3164 : }
3165 :
3166 : /*
3167 : * Attempt to atomically look up, create and open a file from a negative
3168 : * dentry.
3169 : *
3170 : * Returns 0 if successful. The file will have been created and attached to
3171 : * @file by the filesystem calling finish_open().
3172 : *
3173 : * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3174 : * be set. The caller will need to perform the open themselves. @path will
3175 : * have been updated to point to the new dentry. This may be negative.
3176 : *
3177 : * Returns an error code otherwise.
3178 : */
3179 0 : static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
3180 : struct file *file,
3181 : int open_flag, umode_t mode)
3182 : {
3183 0 : struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3184 0 : struct inode *dir = nd->path.dentry->d_inode;
3185 : int error;
3186 :
3187 0 : if (nd->flags & LOOKUP_DIRECTORY)
3188 0 : open_flag |= O_DIRECTORY;
3189 :
3190 0 : file->f_path.dentry = DENTRY_NOT_SET;
3191 0 : file->f_path.mnt = nd->path.mnt;
3192 0 : error = dir->i_op->atomic_open(dir, dentry, file,
3193 0 : open_to_namei_flags(open_flag), mode);
3194 0 : d_lookup_done(dentry);
3195 0 : if (!error) {
3196 0 : if (file->f_mode & FMODE_OPENED) {
3197 0 : if (unlikely(dentry != file->f_path.dentry)) {
3198 0 : dput(dentry);
3199 0 : dentry = dget(file->f_path.dentry);
3200 : }
3201 0 : } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3202 : error = -EIO;
3203 : } else {
3204 0 : if (file->f_path.dentry) {
3205 0 : dput(dentry);
3206 0 : dentry = file->f_path.dentry;
3207 : }
3208 0 : if (unlikely(d_is_negative(dentry)))
3209 0 : error = -ENOENT;
3210 : }
3211 : }
3212 0 : if (error) {
3213 0 : dput(dentry);
3214 0 : dentry = ERR_PTR(error);
3215 : }
3216 0 : return dentry;
3217 : }
3218 :
3219 : /*
3220 : * Look up and maybe create and open the last component.
3221 : *
3222 : * Must be called with parent locked (exclusive in O_CREAT case).
3223 : *
3224 : * Returns 0 on success, that is, if
3225 : * the file was successfully atomically created (if necessary) and opened, or
3226 : * the file was not completely opened at this time, though lookups and
3227 : * creations were performed.
3228 : * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3229 : * In the latter case dentry returned in @path might be negative if O_CREAT
3230 : * hadn't been specified.
3231 : *
3232 : * An error code is returned on failure.
3233 : */
3234 0 : static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3235 : const struct open_flags *op,
3236 : bool got_write)
3237 : {
3238 : struct user_namespace *mnt_userns;
3239 0 : struct dentry *dir = nd->path.dentry;
3240 0 : struct inode *dir_inode = dir->d_inode;
3241 0 : int open_flag = op->open_flag;
3242 : struct dentry *dentry;
3243 0 : int error, create_error = 0;
3244 0 : umode_t mode = op->mode;
3245 0 : DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3246 :
3247 0 : if (unlikely(IS_DEADDIR(dir_inode)))
3248 : return ERR_PTR(-ENOENT);
3249 :
3250 0 : file->f_mode &= ~FMODE_CREATED;
3251 0 : dentry = d_lookup(dir, &nd->last);
3252 : for (;;) {
3253 0 : if (!dentry) {
3254 0 : dentry = d_alloc_parallel(dir, &nd->last, &wq);
3255 0 : if (IS_ERR(dentry))
3256 : return dentry;
3257 : }
3258 0 : if (d_in_lookup(dentry))
3259 : break;
3260 :
3261 0 : error = d_revalidate(dentry, nd->flags);
3262 0 : if (likely(error > 0))
3263 : break;
3264 0 : if (error)
3265 : goto out_dput;
3266 0 : d_invalidate(dentry);
3267 0 : dput(dentry);
3268 0 : dentry = NULL;
3269 : }
3270 0 : if (dentry->d_inode) {
3271 : /* Cached positive dentry: will open in f_op->open */
3272 : return dentry;
3273 : }
3274 :
3275 : /*
3276 : * Checking write permission is tricky, bacuse we don't know if we are
3277 : * going to actually need it: O_CREAT opens should work as long as the
3278 : * file exists. But checking existence breaks atomicity. The trick is
3279 : * to check access and if not granted clear O_CREAT from the flags.
3280 : *
3281 : * Another problem is returing the "right" error value (e.g. for an
3282 : * O_EXCL open we want to return EEXIST not EROFS).
3283 : */
3284 0 : if (unlikely(!got_write))
3285 0 : open_flag &= ~O_TRUNC;
3286 0 : mnt_userns = mnt_user_ns(nd->path.mnt);
3287 0 : if (open_flag & O_CREAT) {
3288 0 : if (open_flag & O_EXCL)
3289 0 : open_flag &= ~O_TRUNC;
3290 0 : if (!IS_POSIXACL(dir->d_inode))
3291 0 : mode &= ~current_umask();
3292 0 : if (likely(got_write))
3293 0 : create_error = may_o_create(mnt_userns, &nd->path,
3294 : dentry, mode);
3295 : else
3296 : create_error = -EROFS;
3297 : }
3298 0 : if (create_error)
3299 0 : open_flag &= ~O_CREAT;
3300 0 : if (dir_inode->i_op->atomic_open) {
3301 0 : dentry = atomic_open(nd, dentry, file, open_flag, mode);
3302 0 : if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3303 0 : dentry = ERR_PTR(create_error);
3304 : return dentry;
3305 : }
3306 :
3307 0 : if (d_in_lookup(dentry)) {
3308 0 : struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3309 : nd->flags);
3310 0 : d_lookup_done(dentry);
3311 0 : if (unlikely(res)) {
3312 0 : if (IS_ERR(res)) {
3313 0 : error = PTR_ERR(res);
3314 0 : goto out_dput;
3315 : }
3316 0 : dput(dentry);
3317 0 : dentry = res;
3318 : }
3319 : }
3320 :
3321 : /* Negative dentry, just create the file */
3322 0 : if (!dentry->d_inode && (open_flag & O_CREAT)) {
3323 0 : file->f_mode |= FMODE_CREATED;
3324 0 : audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3325 0 : if (!dir_inode->i_op->create) {
3326 : error = -EACCES;
3327 : goto out_dput;
3328 : }
3329 :
3330 0 : error = dir_inode->i_op->create(mnt_userns, dir_inode, dentry,
3331 0 : mode, open_flag & O_EXCL);
3332 0 : if (error)
3333 : goto out_dput;
3334 : }
3335 0 : if (unlikely(create_error) && !dentry->d_inode) {
3336 : error = create_error;
3337 : goto out_dput;
3338 : }
3339 : return dentry;
3340 :
3341 : out_dput:
3342 0 : dput(dentry);
3343 0 : return ERR_PTR(error);
3344 : }
3345 :
3346 0 : static const char *open_last_lookups(struct nameidata *nd,
3347 : struct file *file, const struct open_flags *op)
3348 : {
3349 0 : struct dentry *dir = nd->path.dentry;
3350 0 : int open_flag = op->open_flag;
3351 0 : bool got_write = false;
3352 : unsigned seq;
3353 : struct inode *inode;
3354 : struct dentry *dentry;
3355 : const char *res;
3356 :
3357 0 : nd->flags |= op->intent;
3358 :
3359 0 : if (nd->last_type != LAST_NORM) {
3360 0 : if (nd->depth)
3361 0 : put_link(nd);
3362 0 : return handle_dots(nd, nd->last_type);
3363 : }
3364 :
3365 0 : if (!(open_flag & O_CREAT)) {
3366 0 : if (nd->last.name[nd->last.len])
3367 0 : nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3368 : /* we _can_ be in RCU mode here */
3369 0 : dentry = lookup_fast(nd, &inode, &seq);
3370 0 : if (IS_ERR(dentry))
3371 : return ERR_CAST(dentry);
3372 0 : if (likely(dentry))
3373 : goto finish_lookup;
3374 :
3375 0 : BUG_ON(nd->flags & LOOKUP_RCU);
3376 : } else {
3377 : /* create side of things */
3378 0 : if (nd->flags & LOOKUP_RCU) {
3379 0 : if (!try_to_unlazy(nd))
3380 : return ERR_PTR(-ECHILD);
3381 : }
3382 0 : audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3383 : /* trailing slashes? */
3384 0 : if (unlikely(nd->last.name[nd->last.len]))
3385 : return ERR_PTR(-EISDIR);
3386 : }
3387 :
3388 0 : if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3389 0 : got_write = !mnt_want_write(nd->path.mnt);
3390 : /*
3391 : * do _not_ fail yet - we might not need that or fail with
3392 : * a different error; let lookup_open() decide; we'll be
3393 : * dropping this one anyway.
3394 : */
3395 : }
3396 0 : if (open_flag & O_CREAT)
3397 0 : inode_lock(dir->d_inode);
3398 : else
3399 0 : inode_lock_shared(dir->d_inode);
3400 0 : dentry = lookup_open(nd, file, op, got_write);
3401 0 : if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED))
3402 0 : fsnotify_create(dir->d_inode, dentry);
3403 0 : if (open_flag & O_CREAT)
3404 0 : inode_unlock(dir->d_inode);
3405 : else
3406 0 : inode_unlock_shared(dir->d_inode);
3407 :
3408 0 : if (got_write)
3409 0 : mnt_drop_write(nd->path.mnt);
3410 :
3411 0 : if (IS_ERR(dentry))
3412 : return ERR_CAST(dentry);
3413 :
3414 0 : if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
3415 0 : dput(nd->path.dentry);
3416 0 : nd->path.dentry = dentry;
3417 0 : return NULL;
3418 : }
3419 :
3420 : finish_lookup:
3421 0 : if (nd->depth)
3422 0 : put_link(nd);
3423 0 : res = step_into(nd, WALK_TRAILING, dentry, inode, seq);
3424 0 : if (unlikely(res))
3425 0 : nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3426 : return res;
3427 : }
3428 :
3429 : /*
3430 : * Handle the last step of open()
3431 : */
3432 0 : static int do_open(struct nameidata *nd,
3433 : struct file *file, const struct open_flags *op)
3434 : {
3435 : struct user_namespace *mnt_userns;
3436 0 : int open_flag = op->open_flag;
3437 : bool do_truncate;
3438 : int acc_mode;
3439 : int error;
3440 :
3441 0 : if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3442 0 : error = complete_walk(nd);
3443 0 : if (error)
3444 : return error;
3445 : }
3446 : if (!(file->f_mode & FMODE_CREATED))
3447 : audit_inode(nd->name, nd->path.dentry, 0);
3448 0 : mnt_userns = mnt_user_ns(nd->path.mnt);
3449 0 : if (open_flag & O_CREAT) {
3450 0 : if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3451 : return -EEXIST;
3452 0 : if (d_is_dir(nd->path.dentry))
3453 : return -EISDIR;
3454 0 : error = may_create_in_sticky(mnt_userns, nd,
3455 : d_backing_inode(nd->path.dentry));
3456 0 : if (unlikely(error))
3457 : return error;
3458 : }
3459 0 : if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3460 : return -ENOTDIR;
3461 :
3462 0 : do_truncate = false;
3463 0 : acc_mode = op->acc_mode;
3464 0 : if (file->f_mode & FMODE_CREATED) {
3465 : /* Don't check for write permission, don't truncate */
3466 0 : open_flag &= ~O_TRUNC;
3467 0 : acc_mode = 0;
3468 0 : } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) {
3469 0 : error = mnt_want_write(nd->path.mnt);
3470 0 : if (error)
3471 : return error;
3472 : do_truncate = true;
3473 : }
3474 0 : error = may_open(mnt_userns, &nd->path, acc_mode, open_flag);
3475 0 : if (!error && !(file->f_mode & FMODE_OPENED))
3476 0 : error = vfs_open(&nd->path, file);
3477 0 : if (!error)
3478 0 : error = ima_file_check(file, op->acc_mode);
3479 0 : if (!error && do_truncate)
3480 0 : error = handle_truncate(mnt_userns, file);
3481 0 : if (unlikely(error > 0)) {
3482 0 : WARN_ON(1);
3483 0 : error = -EINVAL;
3484 : }
3485 0 : if (do_truncate)
3486 0 : mnt_drop_write(nd->path.mnt);
3487 : return error;
3488 : }
3489 :
3490 : /**
3491 : * vfs_tmpfile - create tmpfile
3492 : * @mnt_userns: user namespace of the mount the inode was found from
3493 : * @dentry: pointer to dentry of the base directory
3494 : * @mode: mode of the new tmpfile
3495 : * @open_flag: flags
3496 : *
3497 : * Create a temporary file.
3498 : *
3499 : * If the inode has been found through an idmapped mount the user namespace of
3500 : * the vfsmount must be passed through @mnt_userns. This function will then take
3501 : * care to map the inode according to @mnt_userns before checking permissions.
3502 : * On non-idmapped mounts or if permission checking is to be performed on the
3503 : * raw inode simply passs init_user_ns.
3504 : */
3505 0 : struct dentry *vfs_tmpfile(struct user_namespace *mnt_userns,
3506 : struct dentry *dentry, umode_t mode, int open_flag)
3507 : {
3508 0 : struct dentry *child = NULL;
3509 0 : struct inode *dir = dentry->d_inode;
3510 : struct inode *inode;
3511 : int error;
3512 :
3513 : /* we want directory to be writable */
3514 0 : error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
3515 0 : if (error)
3516 : goto out_err;
3517 0 : error = -EOPNOTSUPP;
3518 0 : if (!dir->i_op->tmpfile)
3519 : goto out_err;
3520 0 : error = -ENOMEM;
3521 0 : child = d_alloc(dentry, &slash_name);
3522 0 : if (unlikely(!child))
3523 : goto out_err;
3524 0 : error = dir->i_op->tmpfile(mnt_userns, dir, child, mode);
3525 0 : if (error)
3526 : goto out_err;
3527 0 : error = -ENOENT;
3528 0 : inode = child->d_inode;
3529 0 : if (unlikely(!inode))
3530 : goto out_err;
3531 0 : if (!(open_flag & O_EXCL)) {
3532 0 : spin_lock(&inode->i_lock);
3533 0 : inode->i_state |= I_LINKABLE;
3534 0 : spin_unlock(&inode->i_lock);
3535 : }
3536 : ima_post_create_tmpfile(mnt_userns, inode);
3537 : return child;
3538 :
3539 : out_err:
3540 0 : dput(child);
3541 0 : return ERR_PTR(error);
3542 : }
3543 : EXPORT_SYMBOL(vfs_tmpfile);
3544 :
3545 0 : static int do_tmpfile(struct nameidata *nd, unsigned flags,
3546 : const struct open_flags *op,
3547 : struct file *file)
3548 : {
3549 : struct user_namespace *mnt_userns;
3550 : struct dentry *child;
3551 : struct path path;
3552 0 : int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3553 0 : if (unlikely(error))
3554 : return error;
3555 0 : error = mnt_want_write(path.mnt);
3556 0 : if (unlikely(error))
3557 : goto out;
3558 0 : mnt_userns = mnt_user_ns(path.mnt);
3559 0 : child = vfs_tmpfile(mnt_userns, path.dentry, op->mode, op->open_flag);
3560 0 : error = PTR_ERR(child);
3561 0 : if (IS_ERR(child))
3562 : goto out2;
3563 0 : dput(path.dentry);
3564 0 : path.dentry = child;
3565 0 : audit_inode(nd->name, child, 0);
3566 : /* Don't check for other permissions, the inode was just created */
3567 0 : error = may_open(mnt_userns, &path, 0, op->open_flag);
3568 0 : if (!error)
3569 0 : error = vfs_open(&path, file);
3570 : out2:
3571 0 : mnt_drop_write(path.mnt);
3572 : out:
3573 0 : path_put(&path);
3574 : return error;
3575 : }
3576 :
3577 0 : static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3578 : {
3579 : struct path path;
3580 0 : int error = path_lookupat(nd, flags, &path);
3581 0 : if (!error) {
3582 0 : audit_inode(nd->name, path.dentry, 0);
3583 0 : error = vfs_open(&path, file);
3584 : path_put(&path);
3585 : }
3586 0 : return error;
3587 : }
3588 :
3589 0 : static struct file *path_openat(struct nameidata *nd,
3590 : const struct open_flags *op, unsigned flags)
3591 : {
3592 : struct file *file;
3593 : int error;
3594 :
3595 0 : file = alloc_empty_file(op->open_flag, current_cred());
3596 0 : if (IS_ERR(file))
3597 : return file;
3598 :
3599 0 : if (unlikely(file->f_flags & __O_TMPFILE)) {
3600 0 : error = do_tmpfile(nd, flags, op, file);
3601 0 : } else if (unlikely(file->f_flags & O_PATH)) {
3602 0 : error = do_o_path(nd, flags, file);
3603 : } else {
3604 0 : const char *s = path_init(nd, flags);
3605 0 : while (!(error = link_path_walk(s, nd)) &&
3606 : (s = open_last_lookups(nd, file, op)) != NULL)
3607 : ;
3608 0 : if (!error)
3609 0 : error = do_open(nd, file, op);
3610 0 : terminate_walk(nd);
3611 : }
3612 0 : if (likely(!error)) {
3613 0 : if (likely(file->f_mode & FMODE_OPENED))
3614 : return file;
3615 0 : WARN_ON(1);
3616 0 : error = -EINVAL;
3617 : }
3618 0 : fput(file);
3619 0 : if (error == -EOPENSTALE) {
3620 0 : if (flags & LOOKUP_RCU)
3621 : error = -ECHILD;
3622 : else
3623 0 : error = -ESTALE;
3624 : }
3625 0 : return ERR_PTR(error);
3626 : }
3627 :
3628 0 : struct file *do_filp_open(int dfd, struct filename *pathname,
3629 : const struct open_flags *op)
3630 : {
3631 : struct nameidata nd;
3632 0 : int flags = op->lookup_flags;
3633 : struct file *filp;
3634 :
3635 0 : set_nameidata(&nd, dfd, pathname, NULL);
3636 0 : filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3637 0 : if (unlikely(filp == ERR_PTR(-ECHILD)))
3638 0 : filp = path_openat(&nd, op, flags);
3639 0 : if (unlikely(filp == ERR_PTR(-ESTALE)))
3640 0 : filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3641 0 : restore_nameidata();
3642 0 : return filp;
3643 : }
3644 :
3645 0 : struct file *do_file_open_root(const struct path *root,
3646 : const char *name, const struct open_flags *op)
3647 : {
3648 : struct nameidata nd;
3649 : struct file *file;
3650 : struct filename *filename;
3651 0 : int flags = op->lookup_flags;
3652 :
3653 0 : if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN)
3654 : return ERR_PTR(-ELOOP);
3655 :
3656 0 : filename = getname_kernel(name);
3657 0 : if (IS_ERR(filename))
3658 : return ERR_CAST(filename);
3659 :
3660 0 : set_nameidata(&nd, -1, filename, root);
3661 0 : file = path_openat(&nd, op, flags | LOOKUP_RCU);
3662 0 : if (unlikely(file == ERR_PTR(-ECHILD)))
3663 0 : file = path_openat(&nd, op, flags);
3664 0 : if (unlikely(file == ERR_PTR(-ESTALE)))
3665 0 : file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3666 0 : restore_nameidata();
3667 0 : putname(filename);
3668 0 : return file;
3669 : }
3670 :
3671 3 : static struct dentry *filename_create(int dfd, struct filename *name,
3672 : struct path *path, unsigned int lookup_flags)
3673 : {
3674 3 : struct dentry *dentry = ERR_PTR(-EEXIST);
3675 : struct qstr last;
3676 3 : bool want_dir = lookup_flags & LOOKUP_DIRECTORY;
3677 3 : unsigned int reval_flag = lookup_flags & LOOKUP_REVAL;
3678 3 : unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL;
3679 : int type;
3680 : int err2;
3681 : int error;
3682 :
3683 3 : error = filename_parentat(dfd, name, reval_flag, path, &last, &type);
3684 3 : if (error)
3685 0 : return ERR_PTR(error);
3686 :
3687 : /*
3688 : * Yucky last component or no last component at all?
3689 : * (foo/., foo/.., /////)
3690 : */
3691 3 : if (unlikely(type != LAST_NORM))
3692 : goto out;
3693 :
3694 : /* don't fail immediately if it's r/o, at least try to report other errors */
3695 3 : err2 = mnt_want_write(path->mnt);
3696 : /*
3697 : * Do the final lookup. Suppress 'create' if there is a trailing
3698 : * '/', and a directory wasn't requested.
3699 : */
3700 3 : if (last.name[last.len] && !want_dir)
3701 0 : create_flags = 0;
3702 6 : inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3703 3 : dentry = __lookup_hash(&last, path->dentry, reval_flag | create_flags);
3704 3 : if (IS_ERR(dentry))
3705 : goto unlock;
3706 :
3707 3 : error = -EEXIST;
3708 3 : if (d_is_positive(dentry))
3709 : goto fail;
3710 :
3711 : /*
3712 : * Special case - lookup gave negative, but... we had foo/bar/
3713 : * From the vfs_mknod() POV we just have a negative dentry -
3714 : * all is fine. Let's be bastards - you had / on the end, you've
3715 : * been asking for (non-existent) directory. -ENOENT for you.
3716 : */
3717 3 : if (unlikely(!create_flags)) {
3718 : error = -ENOENT;
3719 : goto fail;
3720 : }
3721 3 : if (unlikely(err2)) {
3722 : error = err2;
3723 : goto fail;
3724 : }
3725 : return dentry;
3726 : fail:
3727 0 : dput(dentry);
3728 0 : dentry = ERR_PTR(error);
3729 : unlock:
3730 0 : inode_unlock(path->dentry->d_inode);
3731 0 : if (!err2)
3732 0 : mnt_drop_write(path->mnt);
3733 : out:
3734 0 : path_put(path);
3735 0 : return dentry;
3736 : }
3737 :
3738 3 : struct dentry *kern_path_create(int dfd, const char *pathname,
3739 : struct path *path, unsigned int lookup_flags)
3740 : {
3741 3 : struct filename *filename = getname_kernel(pathname);
3742 3 : struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3743 :
3744 3 : putname(filename);
3745 3 : return res;
3746 : }
3747 : EXPORT_SYMBOL(kern_path_create);
3748 :
3749 3 : void done_path_create(struct path *path, struct dentry *dentry)
3750 : {
3751 3 : dput(dentry);
3752 6 : inode_unlock(path->dentry->d_inode);
3753 3 : mnt_drop_write(path->mnt);
3754 3 : path_put(path);
3755 3 : }
3756 : EXPORT_SYMBOL(done_path_create);
3757 :
3758 0 : inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3759 : struct path *path, unsigned int lookup_flags)
3760 : {
3761 0 : struct filename *filename = getname(pathname);
3762 0 : struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3763 :
3764 0 : putname(filename);
3765 0 : return res;
3766 : }
3767 : EXPORT_SYMBOL(user_path_create);
3768 :
3769 : /**
3770 : * vfs_mknod - create device node or file
3771 : * @mnt_userns: user namespace of the mount the inode was found from
3772 : * @dir: inode of @dentry
3773 : * @dentry: pointer to dentry of the base directory
3774 : * @mode: mode of the new device node or file
3775 : * @dev: device number of device to create
3776 : *
3777 : * Create a device node or file.
3778 : *
3779 : * If the inode has been found through an idmapped mount the user namespace of
3780 : * the vfsmount must be passed through @mnt_userns. This function will then take
3781 : * care to map the inode according to @mnt_userns before checking permissions.
3782 : * On non-idmapped mounts or if permission checking is to be performed on the
3783 : * raw inode simply passs init_user_ns.
3784 : */
3785 1 : int vfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
3786 : struct dentry *dentry, umode_t mode, dev_t dev)
3787 : {
3788 1 : bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV;
3789 1 : int error = may_create(mnt_userns, dir, dentry);
3790 :
3791 1 : if (error)
3792 : return error;
3793 :
3794 2 : if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout &&
3795 1 : !capable(CAP_MKNOD))
3796 : return -EPERM;
3797 :
3798 1 : if (!dir->i_op->mknod)
3799 : return -EPERM;
3800 :
3801 1 : error = devcgroup_inode_mknod(mode, dev);
3802 : if (error)
3803 : return error;
3804 :
3805 1 : error = security_inode_mknod(dir, dentry, mode, dev);
3806 : if (error)
3807 : return error;
3808 :
3809 1 : error = dir->i_op->mknod(mnt_userns, dir, dentry, mode, dev);
3810 1 : if (!error)
3811 : fsnotify_create(dir, dentry);
3812 : return error;
3813 : }
3814 : EXPORT_SYMBOL(vfs_mknod);
3815 :
3816 : static int may_mknod(umode_t mode)
3817 : {
3818 0 : switch (mode & S_IFMT) {
3819 : case S_IFREG:
3820 : case S_IFCHR:
3821 : case S_IFBLK:
3822 : case S_IFIFO:
3823 : case S_IFSOCK:
3824 : case 0: /* zero mode translates to S_IFREG */
3825 : return 0;
3826 : case S_IFDIR:
3827 : return -EPERM;
3828 : default:
3829 : return -EINVAL;
3830 : }
3831 : }
3832 :
3833 0 : static int do_mknodat(int dfd, struct filename *name, umode_t mode,
3834 : unsigned int dev)
3835 : {
3836 : struct user_namespace *mnt_userns;
3837 : struct dentry *dentry;
3838 : struct path path;
3839 : int error;
3840 0 : unsigned int lookup_flags = 0;
3841 :
3842 0 : error = may_mknod(mode);
3843 0 : if (error)
3844 : goto out1;
3845 : retry:
3846 0 : dentry = filename_create(dfd, name, &path, lookup_flags);
3847 0 : error = PTR_ERR(dentry);
3848 0 : if (IS_ERR(dentry))
3849 : goto out1;
3850 :
3851 0 : if (!IS_POSIXACL(path.dentry->d_inode))
3852 0 : mode &= ~current_umask();
3853 0 : error = security_path_mknod(&path, dentry, mode, dev);
3854 : if (error)
3855 : goto out2;
3856 :
3857 0 : mnt_userns = mnt_user_ns(path.mnt);
3858 0 : switch (mode & S_IFMT) {
3859 : case 0: case S_IFREG:
3860 0 : error = vfs_create(mnt_userns, path.dentry->d_inode,
3861 : dentry, mode, true);
3862 : if (!error)
3863 : ima_post_path_mknod(mnt_userns, dentry);
3864 : break;
3865 : case S_IFCHR: case S_IFBLK:
3866 0 : error = vfs_mknod(mnt_userns, path.dentry->d_inode,
3867 : dentry, mode, new_decode_dev(dev));
3868 0 : break;
3869 : case S_IFIFO: case S_IFSOCK:
3870 0 : error = vfs_mknod(mnt_userns, path.dentry->d_inode,
3871 : dentry, mode, 0);
3872 0 : break;
3873 : }
3874 : out2:
3875 0 : done_path_create(&path, dentry);
3876 0 : if (retry_estale(error, lookup_flags)) {
3877 : lookup_flags |= LOOKUP_REVAL;
3878 : goto retry;
3879 : }
3880 : out1:
3881 0 : putname(name);
3882 0 : return error;
3883 : }
3884 :
3885 0 : SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3886 : unsigned int, dev)
3887 : {
3888 0 : return do_mknodat(dfd, getname(filename), mode, dev);
3889 : }
3890 :
3891 0 : SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3892 : {
3893 0 : return do_mknodat(AT_FDCWD, getname(filename), mode, dev);
3894 : }
3895 :
3896 : /**
3897 : * vfs_mkdir - create directory
3898 : * @mnt_userns: user namespace of the mount the inode was found from
3899 : * @dir: inode of @dentry
3900 : * @dentry: pointer to dentry of the base directory
3901 : * @mode: mode of the new directory
3902 : *
3903 : * Create a directory.
3904 : *
3905 : * If the inode has been found through an idmapped mount the user namespace of
3906 : * the vfsmount must be passed through @mnt_userns. This function will then take
3907 : * care to map the inode according to @mnt_userns before checking permissions.
3908 : * On non-idmapped mounts or if permission checking is to be performed on the
3909 : * raw inode simply passs init_user_ns.
3910 : */
3911 2 : int vfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
3912 : struct dentry *dentry, umode_t mode)
3913 : {
3914 2 : int error = may_create(mnt_userns, dir, dentry);
3915 2 : unsigned max_links = dir->i_sb->s_max_links;
3916 :
3917 2 : if (error)
3918 : return error;
3919 :
3920 2 : if (!dir->i_op->mkdir)
3921 : return -EPERM;
3922 :
3923 2 : mode &= (S_IRWXUGO|S_ISVTX);
3924 2 : error = security_inode_mkdir(dir, dentry, mode);
3925 : if (error)
3926 : return error;
3927 :
3928 2 : if (max_links && dir->i_nlink >= max_links)
3929 : return -EMLINK;
3930 :
3931 2 : error = dir->i_op->mkdir(mnt_userns, dir, dentry, mode);
3932 2 : if (!error)
3933 : fsnotify_mkdir(dir, dentry);
3934 : return error;
3935 : }
3936 : EXPORT_SYMBOL(vfs_mkdir);
3937 :
3938 0 : int do_mkdirat(int dfd, struct filename *name, umode_t mode)
3939 : {
3940 : struct dentry *dentry;
3941 : struct path path;
3942 : int error;
3943 0 : unsigned int lookup_flags = LOOKUP_DIRECTORY;
3944 :
3945 : retry:
3946 0 : dentry = filename_create(dfd, name, &path, lookup_flags);
3947 0 : error = PTR_ERR(dentry);
3948 0 : if (IS_ERR(dentry))
3949 : goto out_putname;
3950 :
3951 0 : if (!IS_POSIXACL(path.dentry->d_inode))
3952 0 : mode &= ~current_umask();
3953 0 : error = security_path_mkdir(&path, dentry, mode);
3954 : if (!error) {
3955 : struct user_namespace *mnt_userns;
3956 0 : mnt_userns = mnt_user_ns(path.mnt);
3957 0 : error = vfs_mkdir(mnt_userns, path.dentry->d_inode, dentry,
3958 : mode);
3959 : }
3960 0 : done_path_create(&path, dentry);
3961 0 : if (retry_estale(error, lookup_flags)) {
3962 : lookup_flags |= LOOKUP_REVAL;
3963 : goto retry;
3964 : }
3965 : out_putname:
3966 0 : putname(name);
3967 0 : return error;
3968 : }
3969 :
3970 0 : SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3971 : {
3972 0 : return do_mkdirat(dfd, getname(pathname), mode);
3973 : }
3974 :
3975 0 : SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3976 : {
3977 0 : return do_mkdirat(AT_FDCWD, getname(pathname), mode);
3978 : }
3979 :
3980 : /**
3981 : * vfs_rmdir - remove directory
3982 : * @mnt_userns: user namespace of the mount the inode was found from
3983 : * @dir: inode of @dentry
3984 : * @dentry: pointer to dentry of the base directory
3985 : *
3986 : * Remove a directory.
3987 : *
3988 : * If the inode has been found through an idmapped mount the user namespace of
3989 : * the vfsmount must be passed through @mnt_userns. This function will then take
3990 : * care to map the inode according to @mnt_userns before checking permissions.
3991 : * On non-idmapped mounts or if permission checking is to be performed on the
3992 : * raw inode simply passs init_user_ns.
3993 : */
3994 0 : int vfs_rmdir(struct user_namespace *mnt_userns, struct inode *dir,
3995 : struct dentry *dentry)
3996 : {
3997 0 : int error = may_delete(mnt_userns, dir, dentry, 1);
3998 :
3999 0 : if (error)
4000 : return error;
4001 :
4002 0 : if (!dir->i_op->rmdir)
4003 : return -EPERM;
4004 :
4005 0 : dget(dentry);
4006 0 : inode_lock(dentry->d_inode);
4007 :
4008 0 : error = -EBUSY;
4009 0 : if (is_local_mountpoint(dentry) ||
4010 0 : (dentry->d_inode->i_flags & S_KERNEL_FILE))
4011 : goto out;
4012 :
4013 0 : error = security_inode_rmdir(dir, dentry);
4014 : if (error)
4015 : goto out;
4016 :
4017 0 : error = dir->i_op->rmdir(dir, dentry);
4018 0 : if (error)
4019 : goto out;
4020 :
4021 0 : shrink_dcache_parent(dentry);
4022 0 : dentry->d_inode->i_flags |= S_DEAD;
4023 0 : dont_mount(dentry);
4024 : detach_mounts(dentry);
4025 :
4026 : out:
4027 0 : inode_unlock(dentry->d_inode);
4028 0 : dput(dentry);
4029 0 : if (!error)
4030 0 : d_delete_notify(dir, dentry);
4031 : return error;
4032 : }
4033 : EXPORT_SYMBOL(vfs_rmdir);
4034 :
4035 0 : int do_rmdir(int dfd, struct filename *name)
4036 : {
4037 : struct user_namespace *mnt_userns;
4038 : int error;
4039 : struct dentry *dentry;
4040 : struct path path;
4041 : struct qstr last;
4042 : int type;
4043 0 : unsigned int lookup_flags = 0;
4044 : retry:
4045 0 : error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4046 0 : if (error)
4047 : goto exit1;
4048 :
4049 0 : switch (type) {
4050 : case LAST_DOTDOT:
4051 : error = -ENOTEMPTY;
4052 : goto exit2;
4053 : case LAST_DOT:
4054 0 : error = -EINVAL;
4055 0 : goto exit2;
4056 : case LAST_ROOT:
4057 0 : error = -EBUSY;
4058 0 : goto exit2;
4059 : }
4060 :
4061 0 : error = mnt_want_write(path.mnt);
4062 0 : if (error)
4063 : goto exit2;
4064 :
4065 0 : inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4066 0 : dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4067 0 : error = PTR_ERR(dentry);
4068 0 : if (IS_ERR(dentry))
4069 : goto exit3;
4070 0 : if (!dentry->d_inode) {
4071 : error = -ENOENT;
4072 : goto exit4;
4073 : }
4074 0 : error = security_path_rmdir(&path, dentry);
4075 : if (error)
4076 : goto exit4;
4077 0 : mnt_userns = mnt_user_ns(path.mnt);
4078 0 : error = vfs_rmdir(mnt_userns, path.dentry->d_inode, dentry);
4079 : exit4:
4080 0 : dput(dentry);
4081 : exit3:
4082 0 : inode_unlock(path.dentry->d_inode);
4083 0 : mnt_drop_write(path.mnt);
4084 : exit2:
4085 0 : path_put(&path);
4086 0 : if (retry_estale(error, lookup_flags)) {
4087 : lookup_flags |= LOOKUP_REVAL;
4088 : goto retry;
4089 : }
4090 : exit1:
4091 0 : putname(name);
4092 0 : return error;
4093 : }
4094 :
4095 0 : SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
4096 : {
4097 0 : return do_rmdir(AT_FDCWD, getname(pathname));
4098 : }
4099 :
4100 : /**
4101 : * vfs_unlink - unlink a filesystem object
4102 : * @mnt_userns: user namespace of the mount the inode was found from
4103 : * @dir: parent directory
4104 : * @dentry: victim
4105 : * @delegated_inode: returns victim inode, if the inode is delegated.
4106 : *
4107 : * The caller must hold dir->i_mutex.
4108 : *
4109 : * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
4110 : * return a reference to the inode in delegated_inode. The caller
4111 : * should then break the delegation on that inode and retry. Because
4112 : * breaking a delegation may take a long time, the caller should drop
4113 : * dir->i_mutex before doing so.
4114 : *
4115 : * Alternatively, a caller may pass NULL for delegated_inode. This may
4116 : * be appropriate for callers that expect the underlying filesystem not
4117 : * to be NFS exported.
4118 : *
4119 : * If the inode has been found through an idmapped mount the user namespace of
4120 : * the vfsmount must be passed through @mnt_userns. This function will then take
4121 : * care to map the inode according to @mnt_userns before checking permissions.
4122 : * On non-idmapped mounts or if permission checking is to be performed on the
4123 : * raw inode simply passs init_user_ns.
4124 : */
4125 0 : int vfs_unlink(struct user_namespace *mnt_userns, struct inode *dir,
4126 : struct dentry *dentry, struct inode **delegated_inode)
4127 : {
4128 0 : struct inode *target = dentry->d_inode;
4129 0 : int error = may_delete(mnt_userns, dir, dentry, 0);
4130 :
4131 0 : if (error)
4132 : return error;
4133 :
4134 0 : if (!dir->i_op->unlink)
4135 : return -EPERM;
4136 :
4137 0 : inode_lock(target);
4138 0 : if (IS_SWAPFILE(target))
4139 : error = -EPERM;
4140 0 : else if (is_local_mountpoint(dentry))
4141 : error = -EBUSY;
4142 : else {
4143 0 : error = security_inode_unlink(dir, dentry);
4144 : if (!error) {
4145 0 : error = try_break_deleg(target, delegated_inode);
4146 0 : if (error)
4147 : goto out;
4148 0 : error = dir->i_op->unlink(dir, dentry);
4149 0 : if (!error) {
4150 0 : dont_mount(dentry);
4151 : detach_mounts(dentry);
4152 : }
4153 : }
4154 : }
4155 : out:
4156 0 : inode_unlock(target);
4157 :
4158 : /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4159 0 : if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) {
4160 0 : fsnotify_unlink(dir, dentry);
4161 0 : } else if (!error) {
4162 0 : fsnotify_link_count(target);
4163 0 : d_delete_notify(dir, dentry);
4164 : }
4165 :
4166 : return error;
4167 : }
4168 : EXPORT_SYMBOL(vfs_unlink);
4169 :
4170 : /*
4171 : * Make sure that the actual truncation of the file will occur outside its
4172 : * directory's i_mutex. Truncate can take a long time if there is a lot of
4173 : * writeout happening, and we don't want to prevent access to the directory
4174 : * while waiting on the I/O.
4175 : */
4176 0 : int do_unlinkat(int dfd, struct filename *name)
4177 : {
4178 : int error;
4179 : struct dentry *dentry;
4180 : struct path path;
4181 : struct qstr last;
4182 : int type;
4183 0 : struct inode *inode = NULL;
4184 0 : struct inode *delegated_inode = NULL;
4185 0 : unsigned int lookup_flags = 0;
4186 : retry:
4187 0 : error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4188 0 : if (error)
4189 : goto exit1;
4190 :
4191 0 : error = -EISDIR;
4192 0 : if (type != LAST_NORM)
4193 : goto exit2;
4194 :
4195 0 : error = mnt_want_write(path.mnt);
4196 0 : if (error)
4197 : goto exit2;
4198 : retry_deleg:
4199 0 : inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4200 0 : dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4201 0 : error = PTR_ERR(dentry);
4202 0 : if (!IS_ERR(dentry)) {
4203 : struct user_namespace *mnt_userns;
4204 :
4205 : /* Why not before? Because we want correct error value */
4206 0 : if (last.name[last.len])
4207 : goto slashes;
4208 0 : inode = dentry->d_inode;
4209 0 : if (d_is_negative(dentry))
4210 : goto slashes;
4211 0 : ihold(inode);
4212 0 : error = security_path_unlink(&path, dentry);
4213 : if (error)
4214 : goto exit3;
4215 0 : mnt_userns = mnt_user_ns(path.mnt);
4216 0 : error = vfs_unlink(mnt_userns, path.dentry->d_inode, dentry,
4217 : &delegated_inode);
4218 : exit3:
4219 0 : dput(dentry);
4220 : }
4221 0 : inode_unlock(path.dentry->d_inode);
4222 0 : if (inode)
4223 0 : iput(inode); /* truncate the inode here */
4224 0 : inode = NULL;
4225 0 : if (delegated_inode) {
4226 0 : error = break_deleg_wait(&delegated_inode);
4227 0 : if (!error)
4228 : goto retry_deleg;
4229 : }
4230 0 : mnt_drop_write(path.mnt);
4231 : exit2:
4232 0 : path_put(&path);
4233 0 : if (retry_estale(error, lookup_flags)) {
4234 : lookup_flags |= LOOKUP_REVAL;
4235 : inode = NULL;
4236 : goto retry;
4237 : }
4238 : exit1:
4239 0 : putname(name);
4240 0 : return error;
4241 :
4242 : slashes:
4243 0 : if (d_is_negative(dentry))
4244 : error = -ENOENT;
4245 0 : else if (d_is_dir(dentry))
4246 : error = -EISDIR;
4247 : else
4248 0 : error = -ENOTDIR;
4249 : goto exit3;
4250 : }
4251 :
4252 0 : SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4253 : {
4254 0 : if ((flag & ~AT_REMOVEDIR) != 0)
4255 : return -EINVAL;
4256 :
4257 0 : if (flag & AT_REMOVEDIR)
4258 0 : return do_rmdir(dfd, getname(pathname));
4259 0 : return do_unlinkat(dfd, getname(pathname));
4260 : }
4261 :
4262 0 : SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4263 : {
4264 0 : return do_unlinkat(AT_FDCWD, getname(pathname));
4265 : }
4266 :
4267 : /**
4268 : * vfs_symlink - create symlink
4269 : * @mnt_userns: user namespace of the mount the inode was found from
4270 : * @dir: inode of @dentry
4271 : * @dentry: pointer to dentry of the base directory
4272 : * @oldname: name of the file to link to
4273 : *
4274 : * Create a symlink.
4275 : *
4276 : * If the inode has been found through an idmapped mount the user namespace of
4277 : * the vfsmount must be passed through @mnt_userns. This function will then take
4278 : * care to map the inode according to @mnt_userns before checking permissions.
4279 : * On non-idmapped mounts or if permission checking is to be performed on the
4280 : * raw inode simply passs init_user_ns.
4281 : */
4282 0 : int vfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
4283 : struct dentry *dentry, const char *oldname)
4284 : {
4285 0 : int error = may_create(mnt_userns, dir, dentry);
4286 :
4287 0 : if (error)
4288 : return error;
4289 :
4290 0 : if (!dir->i_op->symlink)
4291 : return -EPERM;
4292 :
4293 0 : error = security_inode_symlink(dir, dentry, oldname);
4294 : if (error)
4295 : return error;
4296 :
4297 0 : error = dir->i_op->symlink(mnt_userns, dir, dentry, oldname);
4298 0 : if (!error)
4299 : fsnotify_create(dir, dentry);
4300 : return error;
4301 : }
4302 : EXPORT_SYMBOL(vfs_symlink);
4303 :
4304 0 : int do_symlinkat(struct filename *from, int newdfd, struct filename *to)
4305 : {
4306 : int error;
4307 : struct dentry *dentry;
4308 : struct path path;
4309 0 : unsigned int lookup_flags = 0;
4310 :
4311 0 : if (IS_ERR(from)) {
4312 0 : error = PTR_ERR(from);
4313 0 : goto out_putnames;
4314 : }
4315 : retry:
4316 0 : dentry = filename_create(newdfd, to, &path, lookup_flags);
4317 0 : error = PTR_ERR(dentry);
4318 0 : if (IS_ERR(dentry))
4319 : goto out_putnames;
4320 :
4321 0 : error = security_path_symlink(&path, dentry, from->name);
4322 : if (!error) {
4323 : struct user_namespace *mnt_userns;
4324 :
4325 0 : mnt_userns = mnt_user_ns(path.mnt);
4326 0 : error = vfs_symlink(mnt_userns, path.dentry->d_inode, dentry,
4327 : from->name);
4328 : }
4329 0 : done_path_create(&path, dentry);
4330 0 : if (retry_estale(error, lookup_flags)) {
4331 : lookup_flags |= LOOKUP_REVAL;
4332 : goto retry;
4333 : }
4334 : out_putnames:
4335 0 : putname(to);
4336 0 : putname(from);
4337 0 : return error;
4338 : }
4339 :
4340 0 : SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4341 : int, newdfd, const char __user *, newname)
4342 : {
4343 0 : return do_symlinkat(getname(oldname), newdfd, getname(newname));
4344 : }
4345 :
4346 0 : SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4347 : {
4348 0 : return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname));
4349 : }
4350 :
4351 : /**
4352 : * vfs_link - create a new link
4353 : * @old_dentry: object to be linked
4354 : * @mnt_userns: the user namespace of the mount
4355 : * @dir: new parent
4356 : * @new_dentry: where to create the new link
4357 : * @delegated_inode: returns inode needing a delegation break
4358 : *
4359 : * The caller must hold dir->i_mutex
4360 : *
4361 : * If vfs_link discovers a delegation on the to-be-linked file in need
4362 : * of breaking, it will return -EWOULDBLOCK and return a reference to the
4363 : * inode in delegated_inode. The caller should then break the delegation
4364 : * and retry. Because breaking a delegation may take a long time, the
4365 : * caller should drop the i_mutex before doing so.
4366 : *
4367 : * Alternatively, a caller may pass NULL for delegated_inode. This may
4368 : * be appropriate for callers that expect the underlying filesystem not
4369 : * to be NFS exported.
4370 : *
4371 : * If the inode has been found through an idmapped mount the user namespace of
4372 : * the vfsmount must be passed through @mnt_userns. This function will then take
4373 : * care to map the inode according to @mnt_userns before checking permissions.
4374 : * On non-idmapped mounts or if permission checking is to be performed on the
4375 : * raw inode simply passs init_user_ns.
4376 : */
4377 0 : int vfs_link(struct dentry *old_dentry, struct user_namespace *mnt_userns,
4378 : struct inode *dir, struct dentry *new_dentry,
4379 : struct inode **delegated_inode)
4380 : {
4381 0 : struct inode *inode = old_dentry->d_inode;
4382 0 : unsigned max_links = dir->i_sb->s_max_links;
4383 : int error;
4384 :
4385 0 : if (!inode)
4386 : return -ENOENT;
4387 :
4388 0 : error = may_create(mnt_userns, dir, new_dentry);
4389 0 : if (error)
4390 : return error;
4391 :
4392 0 : if (dir->i_sb != inode->i_sb)
4393 : return -EXDEV;
4394 :
4395 : /*
4396 : * A link to an append-only or immutable file cannot be created.
4397 : */
4398 0 : if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4399 : return -EPERM;
4400 : /*
4401 : * Updating the link count will likely cause i_uid and i_gid to
4402 : * be writen back improperly if their true value is unknown to
4403 : * the vfs.
4404 : */
4405 0 : if (HAS_UNMAPPED_ID(mnt_userns, inode))
4406 : return -EPERM;
4407 0 : if (!dir->i_op->link)
4408 : return -EPERM;
4409 0 : if (S_ISDIR(inode->i_mode))
4410 : return -EPERM;
4411 :
4412 0 : error = security_inode_link(old_dentry, dir, new_dentry);
4413 : if (error)
4414 : return error;
4415 :
4416 0 : inode_lock(inode);
4417 : /* Make sure we don't allow creating hardlink to an unlinked file */
4418 0 : if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4419 : error = -ENOENT;
4420 0 : else if (max_links && inode->i_nlink >= max_links)
4421 : error = -EMLINK;
4422 : else {
4423 0 : error = try_break_deleg(inode, delegated_inode);
4424 0 : if (!error)
4425 0 : error = dir->i_op->link(old_dentry, dir, new_dentry);
4426 : }
4427 :
4428 0 : if (!error && (inode->i_state & I_LINKABLE)) {
4429 0 : spin_lock(&inode->i_lock);
4430 0 : inode->i_state &= ~I_LINKABLE;
4431 0 : spin_unlock(&inode->i_lock);
4432 : }
4433 0 : inode_unlock(inode);
4434 0 : if (!error)
4435 0 : fsnotify_link(dir, inode, new_dentry);
4436 : return error;
4437 : }
4438 : EXPORT_SYMBOL(vfs_link);
4439 :
4440 : /*
4441 : * Hardlinks are often used in delicate situations. We avoid
4442 : * security-related surprises by not following symlinks on the
4443 : * newname. --KAB
4444 : *
4445 : * We don't follow them on the oldname either to be compatible
4446 : * with linux 2.0, and to avoid hard-linking to directories
4447 : * and other special files. --ADM
4448 : */
4449 0 : int do_linkat(int olddfd, struct filename *old, int newdfd,
4450 : struct filename *new, int flags)
4451 : {
4452 : struct user_namespace *mnt_userns;
4453 : struct dentry *new_dentry;
4454 : struct path old_path, new_path;
4455 0 : struct inode *delegated_inode = NULL;
4456 0 : int how = 0;
4457 : int error;
4458 :
4459 0 : if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) {
4460 : error = -EINVAL;
4461 : goto out_putnames;
4462 : }
4463 : /*
4464 : * To use null names we require CAP_DAC_READ_SEARCH
4465 : * This ensures that not everyone will be able to create
4466 : * handlink using the passed filedescriptor.
4467 : */
4468 0 : if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) {
4469 : error = -ENOENT;
4470 : goto out_putnames;
4471 : }
4472 :
4473 0 : if (flags & AT_SYMLINK_FOLLOW)
4474 0 : how |= LOOKUP_FOLLOW;
4475 : retry:
4476 0 : error = filename_lookup(olddfd, old, how, &old_path, NULL);
4477 0 : if (error)
4478 : goto out_putnames;
4479 :
4480 0 : new_dentry = filename_create(newdfd, new, &new_path,
4481 : (how & LOOKUP_REVAL));
4482 0 : error = PTR_ERR(new_dentry);
4483 0 : if (IS_ERR(new_dentry))
4484 : goto out_putpath;
4485 :
4486 0 : error = -EXDEV;
4487 0 : if (old_path.mnt != new_path.mnt)
4488 : goto out_dput;
4489 0 : mnt_userns = mnt_user_ns(new_path.mnt);
4490 0 : error = may_linkat(mnt_userns, &old_path);
4491 0 : if (unlikely(error))
4492 : goto out_dput;
4493 0 : error = security_path_link(old_path.dentry, &new_path, new_dentry);
4494 : if (error)
4495 : goto out_dput;
4496 0 : error = vfs_link(old_path.dentry, mnt_userns, new_path.dentry->d_inode,
4497 : new_dentry, &delegated_inode);
4498 : out_dput:
4499 0 : done_path_create(&new_path, new_dentry);
4500 0 : if (delegated_inode) {
4501 0 : error = break_deleg_wait(&delegated_inode);
4502 0 : if (!error) {
4503 : path_put(&old_path);
4504 : goto retry;
4505 : }
4506 : }
4507 0 : if (retry_estale(error, how)) {
4508 0 : path_put(&old_path);
4509 0 : how |= LOOKUP_REVAL;
4510 0 : goto retry;
4511 : }
4512 : out_putpath:
4513 : path_put(&old_path);
4514 : out_putnames:
4515 0 : putname(old);
4516 0 : putname(new);
4517 :
4518 0 : return error;
4519 : }
4520 :
4521 0 : SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4522 : int, newdfd, const char __user *, newname, int, flags)
4523 : {
4524 0 : return do_linkat(olddfd, getname_uflags(oldname, flags),
4525 : newdfd, getname(newname), flags);
4526 : }
4527 :
4528 0 : SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4529 : {
4530 0 : return do_linkat(AT_FDCWD, getname(oldname), AT_FDCWD, getname(newname), 0);
4531 : }
4532 :
4533 : /**
4534 : * vfs_rename - rename a filesystem object
4535 : * @rd: pointer to &struct renamedata info
4536 : *
4537 : * The caller must hold multiple mutexes--see lock_rename()).
4538 : *
4539 : * If vfs_rename discovers a delegation in need of breaking at either
4540 : * the source or destination, it will return -EWOULDBLOCK and return a
4541 : * reference to the inode in delegated_inode. The caller should then
4542 : * break the delegation and retry. Because breaking a delegation may
4543 : * take a long time, the caller should drop all locks before doing
4544 : * so.
4545 : *
4546 : * Alternatively, a caller may pass NULL for delegated_inode. This may
4547 : * be appropriate for callers that expect the underlying filesystem not
4548 : * to be NFS exported.
4549 : *
4550 : * The worst of all namespace operations - renaming directory. "Perverted"
4551 : * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4552 : * Problems:
4553 : *
4554 : * a) we can get into loop creation.
4555 : * b) race potential - two innocent renames can create a loop together.
4556 : * That's where 4.4 screws up. Current fix: serialization on
4557 : * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4558 : * story.
4559 : * c) we have to lock _four_ objects - parents and victim (if it exists),
4560 : * and source (if it is not a directory).
4561 : * And that - after we got ->i_mutex on parents (until then we don't know
4562 : * whether the target exists). Solution: try to be smart with locking
4563 : * order for inodes. We rely on the fact that tree topology may change
4564 : * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4565 : * move will be locked. Thus we can rank directories by the tree
4566 : * (ancestors first) and rank all non-directories after them.
4567 : * That works since everybody except rename does "lock parent, lookup,
4568 : * lock child" and rename is under ->s_vfs_rename_mutex.
4569 : * HOWEVER, it relies on the assumption that any object with ->lookup()
4570 : * has no more than 1 dentry. If "hybrid" objects will ever appear,
4571 : * we'd better make sure that there's no link(2) for them.
4572 : * d) conversion from fhandle to dentry may come in the wrong moment - when
4573 : * we are removing the target. Solution: we will have to grab ->i_mutex
4574 : * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4575 : * ->i_mutex on parents, which works but leads to some truly excessive
4576 : * locking].
4577 : */
4578 0 : int vfs_rename(struct renamedata *rd)
4579 : {
4580 : int error;
4581 0 : struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir;
4582 0 : struct dentry *old_dentry = rd->old_dentry;
4583 0 : struct dentry *new_dentry = rd->new_dentry;
4584 0 : struct inode **delegated_inode = rd->delegated_inode;
4585 0 : unsigned int flags = rd->flags;
4586 0 : bool is_dir = d_is_dir(old_dentry);
4587 0 : struct inode *source = old_dentry->d_inode;
4588 0 : struct inode *target = new_dentry->d_inode;
4589 0 : bool new_is_dir = false;
4590 0 : unsigned max_links = new_dir->i_sb->s_max_links;
4591 : struct name_snapshot old_name;
4592 :
4593 0 : if (source == target)
4594 : return 0;
4595 :
4596 0 : error = may_delete(rd->old_mnt_userns, old_dir, old_dentry, is_dir);
4597 0 : if (error)
4598 : return error;
4599 :
4600 0 : if (!target) {
4601 0 : error = may_create(rd->new_mnt_userns, new_dir, new_dentry);
4602 : } else {
4603 0 : new_is_dir = d_is_dir(new_dentry);
4604 :
4605 0 : if (!(flags & RENAME_EXCHANGE))
4606 0 : error = may_delete(rd->new_mnt_userns, new_dir,
4607 : new_dentry, is_dir);
4608 : else
4609 0 : error = may_delete(rd->new_mnt_userns, new_dir,
4610 : new_dentry, new_is_dir);
4611 : }
4612 0 : if (error)
4613 : return error;
4614 :
4615 0 : if (!old_dir->i_op->rename)
4616 : return -EPERM;
4617 :
4618 : /*
4619 : * If we are going to change the parent - check write permissions,
4620 : * we'll need to flip '..'.
4621 : */
4622 0 : if (new_dir != old_dir) {
4623 0 : if (is_dir) {
4624 0 : error = inode_permission(rd->old_mnt_userns, source,
4625 : MAY_WRITE);
4626 0 : if (error)
4627 : return error;
4628 : }
4629 0 : if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4630 0 : error = inode_permission(rd->new_mnt_userns, target,
4631 : MAY_WRITE);
4632 0 : if (error)
4633 : return error;
4634 : }
4635 : }
4636 :
4637 0 : error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4638 : flags);
4639 : if (error)
4640 : return error;
4641 :
4642 0 : take_dentry_name_snapshot(&old_name, old_dentry);
4643 0 : dget(new_dentry);
4644 0 : if (!is_dir || (flags & RENAME_EXCHANGE))
4645 0 : lock_two_nondirectories(source, target);
4646 0 : else if (target)
4647 : inode_lock(target);
4648 :
4649 0 : error = -EPERM;
4650 0 : if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target)))
4651 : goto out;
4652 :
4653 0 : error = -EBUSY;
4654 0 : if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4655 : goto out;
4656 :
4657 0 : if (max_links && new_dir != old_dir) {
4658 0 : error = -EMLINK;
4659 0 : if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4660 : goto out;
4661 0 : if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4662 0 : old_dir->i_nlink >= max_links)
4663 : goto out;
4664 : }
4665 0 : if (!is_dir) {
4666 0 : error = try_break_deleg(source, delegated_inode);
4667 0 : if (error)
4668 : goto out;
4669 : }
4670 0 : if (target && !new_is_dir) {
4671 0 : error = try_break_deleg(target, delegated_inode);
4672 0 : if (error)
4673 : goto out;
4674 : }
4675 0 : error = old_dir->i_op->rename(rd->new_mnt_userns, old_dir, old_dentry,
4676 : new_dir, new_dentry, flags);
4677 0 : if (error)
4678 : goto out;
4679 :
4680 0 : if (!(flags & RENAME_EXCHANGE) && target) {
4681 0 : if (is_dir) {
4682 0 : shrink_dcache_parent(new_dentry);
4683 0 : target->i_flags |= S_DEAD;
4684 : }
4685 0 : dont_mount(new_dentry);
4686 : detach_mounts(new_dentry);
4687 : }
4688 0 : if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4689 0 : if (!(flags & RENAME_EXCHANGE))
4690 0 : d_move(old_dentry, new_dentry);
4691 : else
4692 0 : d_exchange(old_dentry, new_dentry);
4693 : }
4694 : out:
4695 0 : if (!is_dir || (flags & RENAME_EXCHANGE))
4696 0 : unlock_two_nondirectories(source, target);
4697 0 : else if (target)
4698 : inode_unlock(target);
4699 0 : dput(new_dentry);
4700 0 : if (!error) {
4701 0 : fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4702 0 : !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4703 0 : if (flags & RENAME_EXCHANGE) {
4704 0 : fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4705 : new_is_dir, NULL, new_dentry);
4706 : }
4707 : }
4708 0 : release_dentry_name_snapshot(&old_name);
4709 :
4710 0 : return error;
4711 : }
4712 : EXPORT_SYMBOL(vfs_rename);
4713 :
4714 0 : int do_renameat2(int olddfd, struct filename *from, int newdfd,
4715 : struct filename *to, unsigned int flags)
4716 : {
4717 : struct renamedata rd;
4718 : struct dentry *old_dentry, *new_dentry;
4719 : struct dentry *trap;
4720 : struct path old_path, new_path;
4721 : struct qstr old_last, new_last;
4722 : int old_type, new_type;
4723 0 : struct inode *delegated_inode = NULL;
4724 0 : unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4725 0 : bool should_retry = false;
4726 0 : int error = -EINVAL;
4727 :
4728 0 : if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4729 : goto put_names;
4730 :
4731 0 : if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4732 0 : (flags & RENAME_EXCHANGE))
4733 : goto put_names;
4734 :
4735 0 : if (flags & RENAME_EXCHANGE)
4736 0 : target_flags = 0;
4737 :
4738 : retry:
4739 0 : error = filename_parentat(olddfd, from, lookup_flags, &old_path,
4740 : &old_last, &old_type);
4741 0 : if (error)
4742 : goto put_names;
4743 :
4744 0 : error = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last,
4745 : &new_type);
4746 0 : if (error)
4747 : goto exit1;
4748 :
4749 0 : error = -EXDEV;
4750 0 : if (old_path.mnt != new_path.mnt)
4751 : goto exit2;
4752 :
4753 0 : error = -EBUSY;
4754 0 : if (old_type != LAST_NORM)
4755 : goto exit2;
4756 :
4757 0 : if (flags & RENAME_NOREPLACE)
4758 0 : error = -EEXIST;
4759 0 : if (new_type != LAST_NORM)
4760 : goto exit2;
4761 :
4762 0 : error = mnt_want_write(old_path.mnt);
4763 0 : if (error)
4764 : goto exit2;
4765 :
4766 : retry_deleg:
4767 0 : trap = lock_rename(new_path.dentry, old_path.dentry);
4768 :
4769 0 : old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4770 0 : error = PTR_ERR(old_dentry);
4771 0 : if (IS_ERR(old_dentry))
4772 : goto exit3;
4773 : /* source must exist */
4774 0 : error = -ENOENT;
4775 0 : if (d_is_negative(old_dentry))
4776 : goto exit4;
4777 0 : new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4778 0 : error = PTR_ERR(new_dentry);
4779 0 : if (IS_ERR(new_dentry))
4780 : goto exit4;
4781 0 : error = -EEXIST;
4782 0 : if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4783 : goto exit5;
4784 0 : if (flags & RENAME_EXCHANGE) {
4785 0 : error = -ENOENT;
4786 0 : if (d_is_negative(new_dentry))
4787 : goto exit5;
4788 :
4789 0 : if (!d_is_dir(new_dentry)) {
4790 0 : error = -ENOTDIR;
4791 0 : if (new_last.name[new_last.len])
4792 : goto exit5;
4793 : }
4794 : }
4795 : /* unless the source is a directory trailing slashes give -ENOTDIR */
4796 0 : if (!d_is_dir(old_dentry)) {
4797 0 : error = -ENOTDIR;
4798 0 : if (old_last.name[old_last.len])
4799 : goto exit5;
4800 0 : if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4801 : goto exit5;
4802 : }
4803 : /* source should not be ancestor of target */
4804 0 : error = -EINVAL;
4805 0 : if (old_dentry == trap)
4806 : goto exit5;
4807 : /* target should not be an ancestor of source */
4808 0 : if (!(flags & RENAME_EXCHANGE))
4809 0 : error = -ENOTEMPTY;
4810 0 : if (new_dentry == trap)
4811 : goto exit5;
4812 :
4813 0 : error = security_path_rename(&old_path, old_dentry,
4814 : &new_path, new_dentry, flags);
4815 : if (error)
4816 : goto exit5;
4817 :
4818 0 : rd.old_dir = old_path.dentry->d_inode;
4819 0 : rd.old_dentry = old_dentry;
4820 0 : rd.old_mnt_userns = mnt_user_ns(old_path.mnt);
4821 0 : rd.new_dir = new_path.dentry->d_inode;
4822 0 : rd.new_dentry = new_dentry;
4823 0 : rd.new_mnt_userns = mnt_user_ns(new_path.mnt);
4824 0 : rd.delegated_inode = &delegated_inode;
4825 0 : rd.flags = flags;
4826 0 : error = vfs_rename(&rd);
4827 : exit5:
4828 0 : dput(new_dentry);
4829 : exit4:
4830 0 : dput(old_dentry);
4831 : exit3:
4832 0 : unlock_rename(new_path.dentry, old_path.dentry);
4833 0 : if (delegated_inode) {
4834 0 : error = break_deleg_wait(&delegated_inode);
4835 0 : if (!error)
4836 : goto retry_deleg;
4837 : }
4838 0 : mnt_drop_write(old_path.mnt);
4839 : exit2:
4840 0 : if (retry_estale(error, lookup_flags))
4841 0 : should_retry = true;
4842 : path_put(&new_path);
4843 : exit1:
4844 0 : path_put(&old_path);
4845 0 : if (should_retry) {
4846 : should_retry = false;
4847 : lookup_flags |= LOOKUP_REVAL;
4848 : goto retry;
4849 : }
4850 : put_names:
4851 0 : putname(from);
4852 0 : putname(to);
4853 0 : return error;
4854 : }
4855 :
4856 0 : SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4857 : int, newdfd, const char __user *, newname, unsigned int, flags)
4858 : {
4859 0 : return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4860 : flags);
4861 : }
4862 :
4863 0 : SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4864 : int, newdfd, const char __user *, newname)
4865 : {
4866 0 : return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4867 : 0);
4868 : }
4869 :
4870 0 : SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4871 : {
4872 0 : return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD,
4873 : getname(newname), 0);
4874 : }
4875 :
4876 0 : int readlink_copy(char __user *buffer, int buflen, const char *link)
4877 : {
4878 0 : int len = PTR_ERR(link);
4879 0 : if (IS_ERR(link))
4880 : goto out;
4881 :
4882 0 : len = strlen(link);
4883 0 : if (len > (unsigned) buflen)
4884 0 : len = buflen;
4885 0 : if (copy_to_user(buffer, link, len))
4886 0 : len = -EFAULT;
4887 : out:
4888 0 : return len;
4889 : }
4890 :
4891 : /**
4892 : * vfs_readlink - copy symlink body into userspace buffer
4893 : * @dentry: dentry on which to get symbolic link
4894 : * @buffer: user memory pointer
4895 : * @buflen: size of buffer
4896 : *
4897 : * Does not touch atime. That's up to the caller if necessary
4898 : *
4899 : * Does not call security hook.
4900 : */
4901 0 : int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4902 : {
4903 0 : struct inode *inode = d_inode(dentry);
4904 0 : DEFINE_DELAYED_CALL(done);
4905 : const char *link;
4906 : int res;
4907 :
4908 0 : if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4909 0 : if (unlikely(inode->i_op->readlink))
4910 0 : return inode->i_op->readlink(dentry, buffer, buflen);
4911 :
4912 0 : if (!d_is_symlink(dentry))
4913 : return -EINVAL;
4914 :
4915 0 : spin_lock(&inode->i_lock);
4916 0 : inode->i_opflags |= IOP_DEFAULT_READLINK;
4917 0 : spin_unlock(&inode->i_lock);
4918 : }
4919 :
4920 0 : link = READ_ONCE(inode->i_link);
4921 0 : if (!link) {
4922 0 : link = inode->i_op->get_link(dentry, inode, &done);
4923 0 : if (IS_ERR(link))
4924 0 : return PTR_ERR(link);
4925 : }
4926 0 : res = readlink_copy(buffer, buflen, link);
4927 : do_delayed_call(&done);
4928 : return res;
4929 : }
4930 : EXPORT_SYMBOL(vfs_readlink);
4931 :
4932 : /**
4933 : * vfs_get_link - get symlink body
4934 : * @dentry: dentry on which to get symbolic link
4935 : * @done: caller needs to free returned data with this
4936 : *
4937 : * Calls security hook and i_op->get_link() on the supplied inode.
4938 : *
4939 : * It does not touch atime. That's up to the caller if necessary.
4940 : *
4941 : * Does not work on "special" symlinks like /proc/$$/fd/N
4942 : */
4943 0 : const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4944 : {
4945 0 : const char *res = ERR_PTR(-EINVAL);
4946 0 : struct inode *inode = d_inode(dentry);
4947 :
4948 0 : if (d_is_symlink(dentry)) {
4949 0 : res = ERR_PTR(security_inode_readlink(dentry));
4950 : if (!res)
4951 0 : res = inode->i_op->get_link(dentry, inode, done);
4952 : }
4953 0 : return res;
4954 : }
4955 : EXPORT_SYMBOL(vfs_get_link);
4956 :
4957 : /* get the link contents into pagecache */
4958 0 : const char *page_get_link(struct dentry *dentry, struct inode *inode,
4959 : struct delayed_call *callback)
4960 : {
4961 : char *kaddr;
4962 : struct page *page;
4963 0 : struct address_space *mapping = inode->i_mapping;
4964 :
4965 0 : if (!dentry) {
4966 0 : page = find_get_page(mapping, 0);
4967 0 : if (!page)
4968 : return ERR_PTR(-ECHILD);
4969 0 : if (!PageUptodate(page)) {
4970 0 : put_page(page);
4971 0 : return ERR_PTR(-ECHILD);
4972 : }
4973 : } else {
4974 0 : page = read_mapping_page(mapping, 0, NULL);
4975 0 : if (IS_ERR(page))
4976 : return (char*)page;
4977 : }
4978 0 : set_delayed_call(callback, page_put_link, page);
4979 0 : BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4980 0 : kaddr = page_address(page);
4981 0 : nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4982 0 : return kaddr;
4983 : }
4984 :
4985 : EXPORT_SYMBOL(page_get_link);
4986 :
4987 0 : void page_put_link(void *arg)
4988 : {
4989 0 : put_page(arg);
4990 0 : }
4991 : EXPORT_SYMBOL(page_put_link);
4992 :
4993 0 : int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4994 : {
4995 0 : DEFINE_DELAYED_CALL(done);
4996 0 : int res = readlink_copy(buffer, buflen,
4997 : page_get_link(dentry, d_inode(dentry),
4998 : &done));
4999 0 : do_delayed_call(&done);
5000 0 : return res;
5001 : }
5002 : EXPORT_SYMBOL(page_readlink);
5003 :
5004 : /*
5005 : * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
5006 : */
5007 0 : int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
5008 : {
5009 0 : struct address_space *mapping = inode->i_mapping;
5010 : struct page *page;
5011 : void *fsdata;
5012 : int err;
5013 0 : unsigned int flags = 0;
5014 0 : if (nofs)
5015 0 : flags |= AOP_FLAG_NOFS;
5016 :
5017 : retry:
5018 0 : err = pagecache_write_begin(NULL, mapping, 0, len-1,
5019 : flags, &page, &fsdata);
5020 0 : if (err)
5021 : goto fail;
5022 :
5023 0 : memcpy(page_address(page), symname, len-1);
5024 :
5025 0 : err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
5026 : page, fsdata);
5027 0 : if (err < 0)
5028 : goto fail;
5029 0 : if (err < len-1)
5030 : goto retry;
5031 :
5032 0 : mark_inode_dirty(inode);
5033 0 : return 0;
5034 : fail:
5035 : return err;
5036 : }
5037 : EXPORT_SYMBOL(__page_symlink);
5038 :
5039 0 : int page_symlink(struct inode *inode, const char *symname, int len)
5040 : {
5041 0 : return __page_symlink(inode, symname, len,
5042 0 : !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
5043 : }
5044 : EXPORT_SYMBOL(page_symlink);
5045 :
5046 : const struct inode_operations page_symlink_inode_operations = {
5047 : .get_link = page_get_link,
5048 : };
5049 : EXPORT_SYMBOL(page_symlink_inode_operations);
|
