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