Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0
2 : /*
3 : * linux/fs/namei.c
4 : *
5 : * Copyright (C) 1991, 1992 Linus Torvalds
6 : */
7 :
8 : /*
9 : * Some corrections by tytso.
10 : */
11 :
12 : /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
13 : * lookup logic.
14 : */
15 : /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
16 : */
17 :
18 : #include <linux/init.h>
19 : #include <linux/export.h>
20 : #include <linux/kernel.h>
21 : #include <linux/slab.h>
22 : #include <linux/fs.h>
23 : #include <linux/namei.h>
24 : #include <linux/pagemap.h>
25 : #include <linux/fsnotify.h>
26 : #include <linux/personality.h>
27 : #include <linux/security.h>
28 : #include <linux/ima.h>
29 : #include <linux/syscalls.h>
30 : #include <linux/mount.h>
31 : #include <linux/audit.h>
32 : #include <linux/capability.h>
33 : #include <linux/file.h>
34 : #include <linux/fcntl.h>
35 : #include <linux/device_cgroup.h>
36 : #include <linux/fs_struct.h>
37 : #include <linux/posix_acl.h>
38 : #include <linux/hash.h>
39 : #include <linux/bitops.h>
40 : #include <linux/init_task.h>
41 : #include <linux/uaccess.h>
42 :
43 : #include "internal.h"
44 : #include "mount.h"
45 :
46 : /* [Feb-1997 T. Schoebel-Theuer]
47 : * Fundamental changes in the pathname lookup mechanisms (namei)
48 : * were necessary because of omirr. The reason is that omirr needs
49 : * to know the _real_ pathname, not the user-supplied one, in case
50 : * of symlinks (and also when transname replacements occur).
51 : *
52 : * The new code replaces the old recursive symlink resolution with
53 : * an iterative one (in case of non-nested symlink chains). It does
54 : * this with calls to <fs>_follow_link().
55 : * As a side effect, dir_namei(), _namei() and follow_link() are now
56 : * replaced with a single function lookup_dentry() that can handle all
57 : * the special cases of the former code.
58 : *
59 : * With the new dcache, the pathname is stored at each inode, at least as
60 : * long as the refcount of the inode is positive. As a side effect, the
61 : * size of the dcache depends on the inode cache and thus is dynamic.
62 : *
63 : * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
64 : * resolution to correspond with current state of the code.
65 : *
66 : * Note that the symlink resolution is not *completely* iterative.
67 : * There is still a significant amount of tail- and mid- recursion in
68 : * the algorithm. Also, note that <fs>_readlink() is not used in
69 : * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
70 : * may return different results than <fs>_follow_link(). Many virtual
71 : * filesystems (including /proc) exhibit this behavior.
72 : */
73 :
74 : /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
75 : * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
76 : * and the name already exists in form of a symlink, try to create the new
77 : * name indicated by the symlink. The old code always complained that the
78 : * name already exists, due to not following the symlink even if its target
79 : * is nonexistent. The new semantics affects also mknod() and link() when
80 : * the name is a symlink pointing to a non-existent name.
81 : *
82 : * I don't know which semantics is the right one, since I have no access
83 : * to standards. But I found by trial that HP-UX 9.0 has the full "new"
84 : * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
85 : * "old" one. Personally, I think the new semantics is much more logical.
86 : * Note that "ln old new" where "new" is a symlink pointing to a non-existing
87 : * file does succeed in both HP-UX and SunOs, but not in Solaris
88 : * and in the old Linux semantics.
89 : */
90 :
91 : /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
92 : * semantics. See the comments in "open_namei" and "do_link" below.
93 : *
94 : * [10-Sep-98 Alan Modra] Another symlink change.
95 : */
96 :
97 : /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
98 : * inside the path - always follow.
99 : * in the last component in creation/removal/renaming - never follow.
100 : * if LOOKUP_FOLLOW passed - follow.
101 : * if the pathname has trailing slashes - follow.
102 : * otherwise - don't follow.
103 : * (applied in that order).
104 : *
105 : * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
106 : * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
107 : * During the 2.4 we need to fix the userland stuff depending on it -
108 : * hopefully we will be able to get rid of that wart in 2.5. So far only
109 : * XEmacs seems to be relying on it...
110 : */
111 : /*
112 : * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
113 : * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
114 : * any extra contention...
115 : */
116 :
117 : /* In order to reduce some races, while at the same time doing additional
118 : * checking and hopefully speeding things up, we copy filenames to the
119 : * kernel data space before using them..
120 : *
121 : * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
122 : * PATH_MAX includes the nul terminator --RR.
123 : */
124 :
125 : #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
126 :
127 : struct filename *
128 89142 : getname_flags(const char __user *filename, int flags, int *empty)
129 : {
130 89142 : struct filename *result;
131 89142 : char *kname;
132 89142 : int len;
133 :
134 89142 : result = audit_reusename(filename);
135 89142 : if (result)
136 : return result;
137 :
138 89142 : result = __getname();
139 89223 : if (unlikely(!result))
140 89211 : return ERR_PTR(-ENOMEM);
141 :
142 : /*
143 : * First, try to embed the struct filename inside the names_cache
144 : * allocation
145 : */
146 89223 : kname = (char *)result->iname;
147 89223 : result->name = kname;
148 :
149 89223 : len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
150 89211 : if (unlikely(len < 0)) {
151 0 : __putname(result);
152 0 : return ERR_PTR(len);
153 : }
154 :
155 : /*
156 : * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
157 : * separate struct filename so we can dedicate the entire
158 : * names_cache allocation for the pathname, and re-do the copy from
159 : * userland.
160 : */
161 89211 : if (unlikely(len == EMBEDDED_NAME_MAX)) {
162 0 : const size_t size = offsetof(struct filename, iname[1]);
163 0 : kname = (char *)result;
164 :
165 : /*
166 : * size is chosen that way we to guarantee that
167 : * result->iname[0] is within the same object and that
168 : * kname can't be equal to result->iname, no matter what.
169 : */
170 0 : result = kzalloc(size, GFP_KERNEL);
171 0 : if (unlikely(!result)) {
172 0 : __putname(kname);
173 0 : return ERR_PTR(-ENOMEM);
174 : }
175 0 : result->name = kname;
176 0 : len = strncpy_from_user(kname, filename, PATH_MAX);
177 0 : if (unlikely(len < 0)) {
178 0 : __putname(kname);
179 0 : kfree(result);
180 0 : return ERR_PTR(len);
181 : }
182 0 : if (unlikely(len == PATH_MAX)) {
183 0 : __putname(kname);
184 0 : kfree(result);
185 0 : return ERR_PTR(-ENAMETOOLONG);
186 : }
187 : }
188 :
189 89211 : result->refcnt = 1;
190 : /* The empty path is special. */
191 89211 : if (unlikely(!len)) {
192 34 : if (empty)
193 0 : *empty = 1;
194 34 : if (!(flags & LOOKUP_EMPTY)) {
195 0 : putname(result);
196 0 : return ERR_PTR(-ENOENT);
197 : }
198 : }
199 :
200 89211 : result->uptr = filename;
201 89211 : result->aname = NULL;
202 89211 : audit_getname(result);
203 89211 : return result;
204 : }
205 :
206 : struct filename *
207 67038 : getname(const char __user * filename)
208 : {
209 59097 : return getname_flags(filename, 0, NULL);
210 : }
211 :
212 : struct filename *
213 1874 : getname_kernel(const char * filename)
214 : {
215 1874 : struct filename *result;
216 1874 : int len = strlen(filename) + 1;
217 :
218 1874 : result = __getname();
219 1874 : if (unlikely(!result))
220 1874 : return ERR_PTR(-ENOMEM);
221 :
222 1874 : if (len <= EMBEDDED_NAME_MAX) {
223 1874 : result->name = (char *)result->iname;
224 0 : } else if (len <= PATH_MAX) {
225 0 : const size_t size = offsetof(struct filename, iname[1]);
226 0 : struct filename *tmp;
227 :
228 0 : tmp = kmalloc(size, GFP_KERNEL);
229 0 : if (unlikely(!tmp)) {
230 0 : __putname(result);
231 0 : return ERR_PTR(-ENOMEM);
232 : }
233 0 : tmp->name = (char *)result;
234 0 : result = tmp;
235 : } else {
236 0 : __putname(result);
237 0 : return ERR_PTR(-ENAMETOOLONG);
238 : }
239 1874 : memcpy((char *)result->name, filename, len);
240 1874 : result->uptr = NULL;
241 1874 : result->aname = NULL;
242 1874 : result->refcnt = 1;
243 1874 : audit_getname(result);
244 :
245 1874 : return result;
246 : }
247 :
248 91078 : void putname(struct filename *name)
249 : {
250 91078 : BUG_ON(name->refcnt <= 0);
251 :
252 91078 : if (--name->refcnt > 0)
253 : return;
254 :
255 91078 : if (name->name != name->iname) {
256 0 : __putname(name->name);
257 0 : kfree(name);
258 : } else
259 91078 : __putname(name);
260 : }
261 :
262 : /**
263 : * check_acl - perform ACL permission checking
264 : * @mnt_userns: user namespace of the mount the inode was found from
265 : * @inode: inode to check permissions on
266 : * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
267 : *
268 : * This function performs the ACL permission checking. Since this function
269 : * retrieve POSIX acls it needs to know whether it is called from a blocking or
270 : * non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
271 : *
272 : * If the inode has been found through an idmapped mount the user namespace of
273 : * the vfsmount must be passed through @mnt_userns. This function will then take
274 : * care to map the inode according to @mnt_userns before checking permissions.
275 : * On non-idmapped mounts or if permission checking is to be performed on the
276 : * raw inode simply passs init_user_ns.
277 : */
278 : static int check_acl(struct user_namespace *mnt_userns,
279 : struct inode *inode, int mask)
280 : {
281 : #ifdef CONFIG_FS_POSIX_ACL
282 : struct posix_acl *acl;
283 :
284 : if (mask & MAY_NOT_BLOCK) {
285 : acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
286 : if (!acl)
287 : return -EAGAIN;
288 : /* no ->get_acl() calls in RCU mode... */
289 : if (is_uncached_acl(acl))
290 : return -ECHILD;
291 : return posix_acl_permission(mnt_userns, inode, acl, mask);
292 : }
293 :
294 : acl = get_acl(inode, ACL_TYPE_ACCESS);
295 : if (IS_ERR(acl))
296 : return PTR_ERR(acl);
297 : if (acl) {
298 : int error = posix_acl_permission(mnt_userns, inode, acl, mask);
299 : posix_acl_release(acl);
300 : return error;
301 : }
302 : #endif
303 :
304 : return -EAGAIN;
305 : }
306 :
307 : /**
308 : * acl_permission_check - perform basic UNIX permission checking
309 : * @mnt_userns: user namespace of the mount the inode was found from
310 : * @inode: inode to check permissions on
311 : * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
312 : *
313 : * This function performs the basic UNIX permission checking. Since this
314 : * function may retrieve POSIX acls it needs to know whether it is called from a
315 : * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
316 : *
317 : * If the inode has been found through an idmapped mount the user namespace of
318 : * the vfsmount must be passed through @mnt_userns. This function will then take
319 : * care to map the inode according to @mnt_userns before checking permissions.
320 : * On non-idmapped mounts or if permission checking is to be performed on the
321 : * raw inode simply passs init_user_ns.
322 : */
323 258181 : static int acl_permission_check(struct user_namespace *mnt_userns,
324 : struct inode *inode, int mask)
325 : {
326 258181 : unsigned int mode = inode->i_mode;
327 258181 : kuid_t i_uid;
328 :
329 : /* Are we the owner? If so, ACL's don't matter */
330 258181 : i_uid = i_uid_into_mnt(mnt_userns, inode);
331 258181 : if (likely(uid_eq(current_fsuid(), i_uid))) {
332 232907 : mask &= 7;
333 232907 : mode >>= 6;
334 232907 : return (mask & ~mode) ? -EACCES : 0;
335 : }
336 :
337 : /* Do we have ACL's? */
338 25286 : if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
339 25286 : int error = check_acl(mnt_userns, inode, mask);
340 : if (error != -EAGAIN)
341 : return error;
342 : }
343 :
344 : /* Only RWX matters for group/other mode bits */
345 25286 : mask &= 7;
346 :
347 : /*
348 : * Are the group permissions different from
349 : * the other permissions in the bits we care
350 : * about? Need to check group ownership if so.
351 : */
352 25286 : if (mask & (mode ^ (mode >> 3))) {
353 78 : kgid_t kgid = i_gid_into_mnt(mnt_userns, inode);
354 78 : if (in_group_p(kgid))
355 0 : mode >>= 3;
356 : }
357 :
358 : /* Bits in 'mode' clear that we require? */
359 25286 : return (mask & ~mode) ? -EACCES : 0;
360 : }
361 :
362 : /**
363 : * generic_permission - check for access rights on a Posix-like filesystem
364 : * @mnt_userns: user namespace of the mount the inode was found from
365 : * @inode: inode to check access rights for
366 : * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC,
367 : * %MAY_NOT_BLOCK ...)
368 : *
369 : * Used to check for read/write/execute permissions on a file.
370 : * We use "fsuid" for this, letting us set arbitrary permissions
371 : * for filesystem access without changing the "normal" uids which
372 : * are used for other things.
373 : *
374 : * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
375 : * request cannot be satisfied (eg. requires blocking or too much complexity).
376 : * It would then be called again in ref-walk mode.
377 : *
378 : * If the inode has been found through an idmapped mount the user namespace of
379 : * the vfsmount must be passed through @mnt_userns. This function will then take
380 : * care to map the inode according to @mnt_userns before checking permissions.
381 : * On non-idmapped mounts or if permission checking is to be performed on the
382 : * raw inode simply passs init_user_ns.
383 : */
384 258180 : int generic_permission(struct user_namespace *mnt_userns, struct inode *inode,
385 : int mask)
386 : {
387 258180 : int ret;
388 :
389 : /*
390 : * Do the basic permission checks.
391 : */
392 258180 : ret = acl_permission_check(mnt_userns, inode, mask);
393 258193 : if (ret != -EACCES)
394 : return ret;
395 :
396 209 : if (S_ISDIR(inode->i_mode)) {
397 : /* DACs are overridable for directories */
398 139 : if (!(mask & MAY_WRITE))
399 68 : if (capable_wrt_inode_uidgid(mnt_userns, inode,
400 : CAP_DAC_READ_SEARCH))
401 : return 0;
402 78 : if (capable_wrt_inode_uidgid(mnt_userns, inode,
403 : CAP_DAC_OVERRIDE))
404 : return 0;
405 0 : return -EACCES;
406 : }
407 :
408 : /*
409 : * Searching includes executable on directories, else just read.
410 : */
411 70 : mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
412 70 : if (mask == MAY_READ)
413 51 : if (capable_wrt_inode_uidgid(mnt_userns, inode,
414 : CAP_DAC_READ_SEARCH))
415 : return 0;
416 : /*
417 : * Read/write DACs are always overridable.
418 : * Executable DACs are overridable when there is
419 : * at least one exec bit set.
420 : */
421 20 : if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
422 10 : if (capable_wrt_inode_uidgid(mnt_userns, inode,
423 : CAP_DAC_OVERRIDE))
424 3 : return 0;
425 :
426 : return -EACCES;
427 : }
428 : EXPORT_SYMBOL(generic_permission);
429 :
430 : /**
431 : * do_inode_permission - UNIX permission checking
432 : * @mnt_userns: user namespace of the mount the inode was found from
433 : * @inode: inode to check permissions on
434 : * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
435 : *
436 : * We _really_ want to just do "generic_permission()" without
437 : * even looking at the inode->i_op values. So we keep a cache
438 : * flag in inode->i_opflags, that says "this has not special
439 : * permission function, use the fast case".
440 : */
441 267836 : static inline int do_inode_permission(struct user_namespace *mnt_userns,
442 : struct inode *inode, int mask)
443 : {
444 267836 : if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
445 41528 : if (likely(inode->i_op->permission))
446 37764 : return inode->i_op->permission(mnt_userns, inode, mask);
447 :
448 : /* This gets set once for the inode lifetime */
449 3764 : spin_lock(&inode->i_lock);
450 3764 : inode->i_opflags |= IOP_FASTPERM;
451 3764 : spin_unlock(&inode->i_lock);
452 : }
453 230072 : return generic_permission(mnt_userns, inode, mask);
454 : }
455 :
456 : /**
457 : * sb_permission - Check superblock-level permissions
458 : * @sb: Superblock of inode to check permission on
459 : * @inode: Inode to check permission on
460 : * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
461 : *
462 : * Separate out file-system wide checks from inode-specific permission checks.
463 : */
464 267841 : static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
465 : {
466 267841 : if (unlikely(mask & MAY_WRITE)) {
467 6495 : umode_t mode = inode->i_mode;
468 :
469 : /* Nobody gets write access to a read-only fs. */
470 6495 : if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
471 1 : return -EROFS;
472 : }
473 : return 0;
474 : }
475 :
476 : /**
477 : * inode_permission - Check for access rights to a given inode
478 : * @mnt_userns: User namespace of the mount the inode was found from
479 : * @inode: Inode to check permission on
480 : * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
481 : *
482 : * Check for read/write/execute permissions on an inode. We use fs[ug]id for
483 : * this, letting us set arbitrary permissions for filesystem access without
484 : * changing the "normal" UIDs which are used for other things.
485 : *
486 : * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
487 : */
488 267838 : int inode_permission(struct user_namespace *mnt_userns,
489 : struct inode *inode, int mask)
490 : {
491 267838 : int retval;
492 :
493 267838 : retval = sb_permission(inode->i_sb, inode, mask);
494 267838 : if (retval)
495 : return retval;
496 :
497 267837 : if (unlikely(mask & MAY_WRITE)) {
498 : /*
499 : * Nobody gets write access to an immutable file.
500 : */
501 6494 : if (IS_IMMUTABLE(inode))
502 : return -EPERM;
503 :
504 : /*
505 : * Updating mtime will likely cause i_uid and i_gid to be
506 : * written back improperly if their true value is unknown
507 : * to the vfs.
508 : */
509 12980 : if (HAS_UNMAPPED_ID(mnt_userns, inode))
510 : return -EACCES;
511 : }
512 :
513 267833 : retval = do_inode_permission(mnt_userns, inode, mask);
514 267843 : if (retval)
515 : return retval;
516 :
517 258331 : retval = devcgroup_inode_permission(inode, mask);
518 258331 : if (retval)
519 : return retval;
520 :
521 258331 : return security_inode_permission(inode, mask);
522 : }
523 : EXPORT_SYMBOL(inode_permission);
524 :
525 : /**
526 : * path_get - get a reference to a path
527 : * @path: path to get the reference to
528 : *
529 : * Given a path increment the reference count to the dentry and the vfsmount.
530 : */
531 52519 : void path_get(const struct path *path)
532 : {
533 52519 : mntget(path->mnt);
534 52535 : dget(path->dentry);
535 52548 : }
536 : EXPORT_SYMBOL(path_get);
537 :
538 : /**
539 : * path_put - put a reference to a path
540 : * @path: path to put the reference to
541 : *
542 : * Given a path decrement the reference count to the dentry and the vfsmount.
543 : */
544 153828 : void path_put(const struct path *path)
545 : {
546 153828 : dput(path->dentry);
547 153811 : mntput(path->mnt);
548 153824 : }
549 : EXPORT_SYMBOL(path_put);
550 :
551 : #define EMBEDDED_LEVELS 2
552 : struct nameidata {
553 : struct path path;
554 : struct qstr last;
555 : struct path root;
556 : struct inode *inode; /* path.dentry.d_inode */
557 : unsigned int flags;
558 : unsigned seq, m_seq, r_seq;
559 : int last_type;
560 : unsigned depth;
561 : int total_link_count;
562 : struct saved {
563 : struct path link;
564 : struct delayed_call done;
565 : const char *name;
566 : unsigned seq;
567 : } *stack, internal[EMBEDDED_LEVELS];
568 : struct filename *name;
569 : struct nameidata *saved;
570 : unsigned root_seq;
571 : int dfd;
572 : kuid_t dir_uid;
573 : umode_t dir_mode;
574 : } __randomize_layout;
575 :
576 90853 : static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
577 : {
578 90853 : struct nameidata *old = current->nameidata;
579 90853 : p->stack = p->internal;
580 90853 : p->dfd = dfd;
581 90853 : p->name = name;
582 90853 : p->total_link_count = old ? old->total_link_count : 0;
583 90853 : p->saved = old;
584 90853 : current->nameidata = p;
585 : }
586 :
587 90859 : static void restore_nameidata(void)
588 : {
589 90859 : struct nameidata *now = current->nameidata, *old = now->saved;
590 :
591 90859 : current->nameidata = old;
592 90859 : if (old)
593 0 : old->total_link_count = now->total_link_count;
594 90859 : if (now->stack != now->internal)
595 0 : kfree(now->stack);
596 90859 : }
597 :
598 0 : static bool nd_alloc_stack(struct nameidata *nd)
599 : {
600 0 : struct saved *p;
601 :
602 0 : p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
603 0 : nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL);
604 0 : if (unlikely(!p))
605 : return false;
606 0 : memcpy(p, nd->internal, sizeof(nd->internal));
607 0 : nd->stack = p;
608 0 : return true;
609 : }
610 :
611 : /**
612 : * path_connected - Verify that a dentry is below mnt.mnt_root
613 : *
614 : * Rename can sometimes move a file or directory outside of a bind
615 : * mount, path_connected allows those cases to be detected.
616 : */
617 1163 : static bool path_connected(struct vfsmount *mnt, struct dentry *dentry)
618 : {
619 1163 : struct super_block *sb = mnt->mnt_sb;
620 :
621 : /* Bind mounts can have disconnected paths */
622 1163 : if (mnt->mnt_root == sb->s_root)
623 : return true;
624 :
625 0 : return is_subdir(dentry, mnt->mnt_root);
626 : }
627 :
628 90934 : static void drop_links(struct nameidata *nd)
629 : {
630 90934 : int i = nd->depth;
631 90934 : while (i--) {
632 12 : struct saved *last = nd->stack + i;
633 12 : do_delayed_call(&last->done);
634 90952 : clear_delayed_call(&last->done);
635 : }
636 90940 : }
637 :
638 90933 : static void terminate_walk(struct nameidata *nd)
639 : {
640 90933 : drop_links(nd);
641 90943 : if (!(nd->flags & LOOKUP_RCU)) {
642 77578 : int i;
643 77578 : path_put(&nd->path);
644 155162 : for (i = 0; i < nd->depth; i++)
645 5 : path_put(&nd->stack[i].link);
646 77579 : if (nd->flags & LOOKUP_ROOT_GRABBED) {
647 15701 : path_put(&nd->root);
648 15701 : nd->flags &= ~LOOKUP_ROOT_GRABBED;
649 : }
650 : } else {
651 13365 : nd->flags &= ~LOOKUP_RCU;
652 13365 : rcu_read_unlock();
653 : }
654 90944 : nd->depth = 0;
655 90944 : }
656 :
657 : /* path_put is needed afterwards regardless of success or failure */
658 91013 : static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq)
659 : {
660 91013 : int res = __legitimize_mnt(path->mnt, mseq);
661 91030 : if (unlikely(res)) {
662 78 : if (res > 0)
663 76 : path->mnt = NULL;
664 78 : path->dentry = NULL;
665 78 : return false;
666 : }
667 90952 : if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
668 0 : path->dentry = NULL;
669 0 : return false;
670 : }
671 90948 : return !read_seqcount_retry(&path->dentry->d_seq, seq);
672 : }
673 :
674 91018 : static inline bool legitimize_path(struct nameidata *nd,
675 : struct path *path, unsigned seq)
676 : {
677 91018 : return __legitimize_path(path, seq, nd->m_seq);
678 : }
679 :
680 77461 : static bool legitimize_links(struct nameidata *nd)
681 : {
682 77461 : int i;
683 77461 : if (unlikely(nd->flags & LOOKUP_CACHED)) {
684 0 : drop_links(nd);
685 0 : nd->depth = 0;
686 0 : return false;
687 : }
688 78118 : for (i = 0; i < nd->depth; i++) {
689 665 : struct saved *last = nd->stack + i;
690 665 : if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
691 6 : drop_links(nd);
692 6 : nd->depth = i + 1;
693 6 : return false;
694 : }
695 : }
696 : return true;
697 : }
698 :
699 77391 : static bool legitimize_root(struct nameidata *nd)
700 : {
701 : /*
702 : * For scoped-lookups (where nd->root has been zeroed), we need to
703 : * restart the whole lookup from scratch -- because set_root() is wrong
704 : * for these lookups (nd->dfd is the root, not the filesystem root).
705 : */
706 77391 : if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED))
707 : return false;
708 : /* Nothing to do if nd->root is zero or is managed by the VFS user. */
709 77391 : if (!nd->root.mnt || (nd->flags & LOOKUP_ROOT))
710 : return true;
711 15045 : nd->flags |= LOOKUP_ROOT_GRABBED;
712 15045 : return legitimize_path(nd, &nd->root, nd->root_seq);
713 : }
714 :
715 : /*
716 : * Path walking has 2 modes, rcu-walk and ref-walk (see
717 : * Documentation/filesystems/path-lookup.txt). In situations when we can't
718 : * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
719 : * normal reference counts on dentries and vfsmounts to transition to ref-walk
720 : * mode. Refcounts are grabbed at the last known good point before rcu-walk
721 : * got stuck, so ref-walk may continue from there. If this is not successful
722 : * (eg. a seqcount has changed), then failure is returned and it's up to caller
723 : * to restart the path walk from the beginning in ref-walk mode.
724 : */
725 :
726 : /**
727 : * try_to_unlazy - try to switch to ref-walk mode.
728 : * @nd: nameidata pathwalk data
729 : * Returns: true on success, false on failure
730 : *
731 : * try_to_unlazy attempts to legitimize the current nd->path and nd->root
732 : * for ref-walk mode.
733 : * Must be called from rcu-walk context.
734 : * Nothing should touch nameidata between try_to_unlazy() failure and
735 : * terminate_walk().
736 : */
737 75311 : static bool try_to_unlazy(struct nameidata *nd)
738 : {
739 75311 : struct dentry *parent = nd->path.dentry;
740 :
741 75311 : BUG_ON(!(nd->flags & LOOKUP_RCU));
742 :
743 75311 : nd->flags &= ~LOOKUP_RCU;
744 75311 : if (unlikely(!legitimize_links(nd)))
745 0 : goto out1;
746 75308 : if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
747 68 : goto out;
748 75253 : if (unlikely(!legitimize_root(nd)))
749 1 : goto out;
750 75249 : rcu_read_unlock();
751 75240 : BUG_ON(nd->inode != parent->d_inode);
752 : return true;
753 :
754 0 : out1:
755 0 : nd->path.mnt = NULL;
756 0 : nd->path.dentry = NULL;
757 69 : out:
758 69 : rcu_read_unlock();
759 69 : return false;
760 : }
761 :
762 : /**
763 : * try_to_unlazy_next - try to switch to ref-walk mode.
764 : * @nd: nameidata pathwalk data
765 : * @dentry: next dentry to step into
766 : * @seq: seq number to check @dentry against
767 : * Returns: true on success, false on failure
768 : *
769 : * Similar to to try_to_unlazy(), but here we have the next dentry already
770 : * picked by rcu-walk and want to legitimize that in addition to the current
771 : * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context.
772 : * Nothing should touch nameidata between try_to_unlazy_next() failure and
773 : * terminate_walk().
774 : */
775 2154 : static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry, unsigned seq)
776 : {
777 2154 : BUG_ON(!(nd->flags & LOOKUP_RCU));
778 :
779 2154 : nd->flags &= ~LOOKUP_RCU;
780 2154 : if (unlikely(!legitimize_links(nd)))
781 5 : goto out2;
782 2147 : if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
783 7 : goto out2;
784 2141 : if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
785 0 : goto out1;
786 :
787 : /*
788 : * We need to move both the parent and the dentry from the RCU domain
789 : * to be properly refcounted. And the sequence number in the dentry
790 : * validates *both* dentry counters, since we checked the sequence
791 : * number of the parent after we got the child sequence number. So we
792 : * know the parent must still be valid if the child sequence number is
793 : */
794 2141 : if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
795 0 : goto out;
796 2141 : if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
797 0 : goto out_dput;
798 : /*
799 : * Sequence counts matched. Now make sure that the root is
800 : * still valid and get it if required.
801 : */
802 2141 : if (unlikely(!legitimize_root(nd)))
803 3 : goto out_dput;
804 2138 : rcu_read_unlock();
805 2138 : return true;
806 :
807 12 : out2:
808 12 : nd->path.mnt = NULL;
809 12 : out1:
810 12 : nd->path.dentry = NULL;
811 12 : out:
812 12 : rcu_read_unlock();
813 12 : return false;
814 3 : out_dput:
815 3 : rcu_read_unlock();
816 3 : dput(dentry);
817 3 : return false;
818 : }
819 :
820 208230 : static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
821 : {
822 208230 : if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
823 25947 : return dentry->d_op->d_revalidate(dentry, flags);
824 : else
825 : return 1;
826 : }
827 :
828 : /**
829 : * complete_walk - successful completion of path walk
830 : * @nd: pointer nameidata
831 : *
832 : * If we had been in RCU mode, drop out of it and legitimize nd->path.
833 : * Revalidate the final result, unless we'd already done that during
834 : * the path walk or the filesystem doesn't ask for it. Return 0 on
835 : * success, -error on failure. In case of failure caller does not
836 : * need to drop nd->path.
837 : */
838 68114 : static int complete_walk(struct nameidata *nd)
839 : {
840 68114 : struct dentry *dentry = nd->path.dentry;
841 68114 : int status;
842 :
843 68114 : if (nd->flags & LOOKUP_RCU) {
844 : /*
845 : * We don't want to zero nd->root for scoped-lookups or
846 : * externally-managed nd->root.
847 : */
848 53076 : if (!(nd->flags & (LOOKUP_ROOT | LOOKUP_IS_SCOPED)))
849 53081 : nd->root.mnt = NULL;
850 53076 : nd->flags &= ~LOOKUP_CACHED;
851 53076 : if (!try_to_unlazy(nd))
852 : return -ECHILD;
853 : }
854 :
855 68059 : if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
856 : /*
857 : * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
858 : * ever step outside the root during lookup" and should already
859 : * be guaranteed by the rest of namei, we want to avoid a namei
860 : * BUG resulting in userspace being given a path that was not
861 : * scoped within the root at some point during the lookup.
862 : *
863 : * So, do a final sanity-check to make sure that in the
864 : * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
865 : * we won't silently return an fd completely outside of the
866 : * requested root to userspace.
867 : *
868 : * Userspace could move the path outside the root after this
869 : * check, but as discussed elsewhere this is not a concern (the
870 : * resolved file was inside the root at some point).
871 : */
872 0 : if (!path_is_under(&nd->path, &nd->root))
873 : return -EXDEV;
874 : }
875 :
876 68059 : if (likely(!(nd->flags & LOOKUP_JUMPED)))
877 : return 0;
878 :
879 14338 : if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
880 : return 0;
881 :
882 0 : status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
883 0 : if (status > 0)
884 : return 0;
885 :
886 0 : if (!status)
887 0 : status = -ESTALE;
888 :
889 : return status;
890 : }
891 :
892 53107 : static int set_root(struct nameidata *nd)
893 : {
894 53107 : struct fs_struct *fs = current->fs;
895 :
896 : /*
897 : * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
898 : * still have to ensure it doesn't happen because it will cause a breakout
899 : * from the dirfd.
900 : */
901 53107 : if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED))
902 : return -ENOTRECOVERABLE;
903 :
904 53107 : if (nd->flags & LOOKUP_RCU) {
905 52450 : unsigned seq;
906 :
907 52450 : do {
908 52450 : seq = read_seqcount_begin(&fs->seq);
909 52451 : nd->root = fs->root;
910 52451 : nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
911 52455 : } while (read_seqcount_retry(&fs->seq, seq));
912 : } else {
913 657 : get_fs_root(fs, &nd->root);
914 657 : nd->flags |= LOOKUP_ROOT_GRABBED;
915 : }
916 : return 0;
917 : }
918 :
919 54910 : static int nd_jump_root(struct nameidata *nd)
920 : {
921 54910 : if (unlikely(nd->flags & LOOKUP_BENEATH))
922 : return -EXDEV;
923 54910 : if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
924 : /* Absolute path arguments to path_init() are allowed. */
925 0 : if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt)
926 : return -EXDEV;
927 : }
928 54910 : if (!nd->root.mnt) {
929 52453 : int error = set_root(nd);
930 52459 : if (error)
931 : return error;
932 : }
933 54916 : if (nd->flags & LOOKUP_RCU) {
934 54665 : struct dentry *d;
935 54665 : nd->path = nd->root;
936 54665 : d = nd->path.dentry;
937 54665 : nd->inode = d->d_inode;
938 54665 : nd->seq = nd->root_seq;
939 54665 : if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
940 : return -ECHILD;
941 : } else {
942 251 : path_put(&nd->path);
943 251 : nd->path = nd->root;
944 251 : path_get(&nd->path);
945 251 : nd->inode = nd->path.dentry->d_inode;
946 : }
947 54913 : nd->flags |= LOOKUP_JUMPED;
948 54913 : return 0;
949 : }
950 :
951 : /*
952 : * Helper to directly jump to a known parsed path from ->get_link,
953 : * caller must have taken a reference to path beforehand.
954 : */
955 123 : int nd_jump_link(struct path *path)
956 : {
957 123 : int error = -ELOOP;
958 123 : struct nameidata *nd = current->nameidata;
959 :
960 123 : if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS))
961 0 : goto err;
962 :
963 123 : error = -EXDEV;
964 123 : if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
965 0 : if (nd->path.mnt != path->mnt)
966 0 : goto err;
967 : }
968 : /* Not currently safe for scoped-lookups. */
969 123 : if (unlikely(nd->flags & LOOKUP_IS_SCOPED))
970 0 : goto err;
971 :
972 123 : path_put(&nd->path);
973 123 : nd->path = *path;
974 123 : nd->inode = nd->path.dentry->d_inode;
975 123 : nd->flags |= LOOKUP_JUMPED;
976 123 : return 0;
977 :
978 0 : err:
979 0 : path_put(path);
980 0 : return error;
981 : }
982 :
983 8154 : static inline void put_link(struct nameidata *nd)
984 : {
985 8154 : struct saved *last = nd->stack + --nd->depth;
986 8154 : do_delayed_call(&last->done);
987 8152 : if (!(nd->flags & LOOKUP_RCU))
988 1859 : path_put(&last->link);
989 8154 : }
990 :
991 : int sysctl_protected_symlinks __read_mostly = 0;
992 : int sysctl_protected_hardlinks __read_mostly = 0;
993 : int sysctl_protected_fifos __read_mostly;
994 : int sysctl_protected_regular __read_mostly;
995 :
996 : /**
997 : * may_follow_link - Check symlink following for unsafe situations
998 : * @nd: nameidata pathwalk data
999 : *
1000 : * In the case of the sysctl_protected_symlinks sysctl being enabled,
1001 : * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
1002 : * in a sticky world-writable directory. This is to protect privileged
1003 : * processes from failing races against path names that may change out
1004 : * from under them by way of other users creating malicious symlinks.
1005 : * It will permit symlinks to be followed only when outside a sticky
1006 : * world-writable directory, or when the uid of the symlink and follower
1007 : * match, or when the directory owner matches the symlink's owner.
1008 : *
1009 : * Returns 0 if following the symlink is allowed, -ve on error.
1010 : */
1011 7118 : static inline int may_follow_link(struct nameidata *nd, const struct inode *inode)
1012 : {
1013 7118 : struct user_namespace *mnt_userns;
1014 7118 : kuid_t i_uid;
1015 :
1016 7118 : if (!sysctl_protected_symlinks)
1017 : return 0;
1018 :
1019 6421 : mnt_userns = mnt_user_ns(nd->path.mnt);
1020 6421 : i_uid = i_uid_into_mnt(mnt_userns, inode);
1021 : /* Allowed if owner and follower match. */
1022 6421 : if (uid_eq(current_cred()->fsuid, i_uid))
1023 : return 0;
1024 :
1025 : /* Allowed if parent directory not sticky and world-writable. */
1026 679 : if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
1027 : return 0;
1028 :
1029 : /* Allowed if parent directory and link owner match. */
1030 0 : if (uid_valid(nd->dir_uid) && uid_eq(nd->dir_uid, i_uid))
1031 : return 0;
1032 :
1033 0 : if (nd->flags & LOOKUP_RCU)
1034 0 : return -ECHILD;
1035 :
1036 7118 : audit_inode(nd->name, nd->stack[0].link.dentry, 0);
1037 7118 : audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link");
1038 : return -EACCES;
1039 : }
1040 :
1041 : /**
1042 : * safe_hardlink_source - Check for safe hardlink conditions
1043 : * @mnt_userns: user namespace of the mount the inode was found from
1044 : * @inode: the source inode to hardlink from
1045 : *
1046 : * Return false if at least one of the following conditions:
1047 : * - inode is not a regular file
1048 : * - inode is setuid
1049 : * - inode is setgid and group-exec
1050 : * - access failure for read and write
1051 : *
1052 : * Otherwise returns true.
1053 : */
1054 26 : static bool safe_hardlink_source(struct user_namespace *mnt_userns,
1055 : struct inode *inode)
1056 : {
1057 26 : umode_t mode = inode->i_mode;
1058 :
1059 : /* Special files should not get pinned to the filesystem. */
1060 26 : if (!S_ISREG(mode))
1061 : return false;
1062 :
1063 : /* Setuid files should not get pinned to the filesystem. */
1064 11 : if (mode & S_ISUID)
1065 : return false;
1066 :
1067 : /* Executable setgid files should not get pinned to the filesystem. */
1068 11 : if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
1069 : return false;
1070 :
1071 : /* Hardlinking to unreadable or unwritable sources is dangerous. */
1072 11 : if (inode_permission(mnt_userns, inode, MAY_READ | MAY_WRITE))
1073 6 : return false;
1074 :
1075 : return true;
1076 : }
1077 :
1078 : /**
1079 : * may_linkat - Check permissions for creating a hardlink
1080 : * @mnt_userns: user namespace of the mount the inode was found from
1081 : * @link: the source to hardlink from
1082 : *
1083 : * Block hardlink when all of:
1084 : * - sysctl_protected_hardlinks enabled
1085 : * - fsuid does not match inode
1086 : * - hardlink source is unsafe (see safe_hardlink_source() above)
1087 : * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1088 : *
1089 : * If the inode has been found through an idmapped mount the user namespace of
1090 : * the vfsmount must be passed through @mnt_userns. This function will then take
1091 : * care to map the inode according to @mnt_userns before checking permissions.
1092 : * On non-idmapped mounts or if permission checking is to be performed on the
1093 : * raw inode simply passs init_user_ns.
1094 : *
1095 : * Returns 0 if successful, -ve on error.
1096 : */
1097 26 : int may_linkat(struct user_namespace *mnt_userns, struct path *link)
1098 : {
1099 26 : struct inode *inode = link->dentry->d_inode;
1100 :
1101 : /* Inode writeback is not safe when the uid or gid are invalid. */
1102 26 : if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) ||
1103 26 : !gid_valid(i_gid_into_mnt(mnt_userns, inode)))
1104 : return -EOVERFLOW;
1105 :
1106 26 : if (!sysctl_protected_hardlinks)
1107 : return 0;
1108 :
1109 : /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1110 : * otherwise, it must be a safe source.
1111 : */
1112 47 : if (safe_hardlink_source(mnt_userns, inode) ||
1113 21 : inode_owner_or_capable(mnt_userns, inode))
1114 26 : return 0;
1115 :
1116 26 : audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
1117 : return -EPERM;
1118 : }
1119 :
1120 : /**
1121 : * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1122 : * should be allowed, or not, on files that already
1123 : * exist.
1124 : * @mnt_userns: user namespace of the mount the inode was found from
1125 : * @dir_mode: mode bits of directory
1126 : * @dir_uid: owner of directory
1127 : * @inode: the inode of the file to open
1128 : *
1129 : * Block an O_CREAT open of a FIFO (or a regular file) when:
1130 : * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1131 : * - the file already exists
1132 : * - we are in a sticky directory
1133 : * - we don't own the file
1134 : * - the owner of the directory doesn't own the file
1135 : * - the directory is world writable
1136 : * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1137 : * the directory doesn't have to be world writable: being group writable will
1138 : * be enough.
1139 : *
1140 : * If the inode has been found through an idmapped mount the user namespace of
1141 : * the vfsmount must be passed through @mnt_userns. This function will then take
1142 : * care to map the inode according to @mnt_userns before checking permissions.
1143 : * On non-idmapped mounts or if permission checking is to be performed on the
1144 : * raw inode simply passs init_user_ns.
1145 : *
1146 : * Returns 0 if the open is allowed, -ve on error.
1147 : */
1148 738 : static int may_create_in_sticky(struct user_namespace *mnt_userns,
1149 : struct nameidata *nd, struct inode *const inode)
1150 : {
1151 738 : umode_t dir_mode = nd->dir_mode;
1152 738 : kuid_t dir_uid = nd->dir_uid;
1153 :
1154 738 : if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1155 735 : (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1156 73 : likely(!(dir_mode & S_ISVTX)) ||
1157 0 : uid_eq(i_uid_into_mnt(mnt_userns, inode), dir_uid) ||
1158 0 : uid_eq(current_fsuid(), i_uid_into_mnt(mnt_userns, inode)))
1159 738 : return 0;
1160 :
1161 0 : if (likely(dir_mode & 0002) ||
1162 0 : (dir_mode & 0020 &&
1163 0 : ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1164 0 : (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1165 0 : const char *operation = S_ISFIFO(inode->i_mode) ?
1166 : "sticky_create_fifo" :
1167 : "sticky_create_regular";
1168 0 : audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
1169 0 : return -EACCES;
1170 : }
1171 : return 0;
1172 : }
1173 :
1174 : /*
1175 : * follow_up - Find the mountpoint of path's vfsmount
1176 : *
1177 : * Given a path, find the mountpoint of its source file system.
1178 : * Replace @path with the path of the mountpoint in the parent mount.
1179 : * Up is towards /.
1180 : *
1181 : * Return 1 if we went up a level and 0 if we were already at the
1182 : * root.
1183 : */
1184 632 : int follow_up(struct path *path)
1185 : {
1186 632 : struct mount *mnt = real_mount(path->mnt);
1187 632 : struct mount *parent;
1188 632 : struct dentry *mountpoint;
1189 :
1190 632 : read_seqlock_excl(&mount_lock);
1191 632 : parent = mnt->mnt_parent;
1192 632 : if (parent == mnt) {
1193 179 : read_sequnlock_excl(&mount_lock);
1194 179 : return 0;
1195 : }
1196 453 : mntget(&parent->mnt);
1197 453 : mountpoint = dget(mnt->mnt_mountpoint);
1198 453 : read_sequnlock_excl(&mount_lock);
1199 453 : dput(path->dentry);
1200 453 : path->dentry = mountpoint;
1201 453 : mntput(path->mnt);
1202 453 : path->mnt = &parent->mnt;
1203 453 : return 1;
1204 : }
1205 : EXPORT_SYMBOL(follow_up);
1206 :
1207 0 : static bool choose_mountpoint_rcu(struct mount *m, const struct path *root,
1208 : struct path *path, unsigned *seqp)
1209 : {
1210 0 : while (mnt_has_parent(m)) {
1211 0 : struct dentry *mountpoint = m->mnt_mountpoint;
1212 :
1213 0 : m = m->mnt_parent;
1214 0 : if (unlikely(root->dentry == mountpoint &&
1215 : root->mnt == &m->mnt))
1216 : break;
1217 0 : if (mountpoint != m->mnt.mnt_root) {
1218 0 : path->mnt = &m->mnt;
1219 0 : path->dentry = mountpoint;
1220 0 : *seqp = read_seqcount_begin(&mountpoint->d_seq);
1221 0 : return true;
1222 : }
1223 : }
1224 : return false;
1225 : }
1226 :
1227 0 : static bool choose_mountpoint(struct mount *m, const struct path *root,
1228 : struct path *path)
1229 : {
1230 0 : bool found;
1231 :
1232 0 : rcu_read_lock();
1233 0 : while (1) {
1234 0 : unsigned seq, mseq = read_seqbegin(&mount_lock);
1235 :
1236 0 : found = choose_mountpoint_rcu(m, root, path, &seq);
1237 0 : if (unlikely(!found)) {
1238 0 : if (!read_seqretry(&mount_lock, mseq))
1239 : break;
1240 : } else {
1241 0 : if (likely(__legitimize_path(path, seq, mseq)))
1242 : break;
1243 0 : rcu_read_unlock();
1244 0 : path_put(path);
1245 0 : rcu_read_lock();
1246 : }
1247 : }
1248 0 : rcu_read_unlock();
1249 0 : return found;
1250 : }
1251 :
1252 : /*
1253 : * Perform an automount
1254 : * - return -EISDIR to tell follow_managed() to stop and return the path we
1255 : * were called with.
1256 : */
1257 0 : static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1258 : {
1259 0 : struct dentry *dentry = path->dentry;
1260 :
1261 : /* We don't want to mount if someone's just doing a stat -
1262 : * unless they're stat'ing a directory and appended a '/' to
1263 : * the name.
1264 : *
1265 : * We do, however, want to mount if someone wants to open or
1266 : * create a file of any type under the mountpoint, wants to
1267 : * traverse through the mountpoint or wants to open the
1268 : * mounted directory. Also, autofs may mark negative dentries
1269 : * as being automount points. These will need the attentions
1270 : * of the daemon to instantiate them before they can be used.
1271 : */
1272 0 : if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1273 0 : LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1274 0 : dentry->d_inode)
1275 : return -EISDIR;
1276 :
1277 0 : if (count && (*count)++ >= MAXSYMLINKS)
1278 : return -ELOOP;
1279 :
1280 0 : return finish_automount(dentry->d_op->d_automount(path), path);
1281 : }
1282 :
1283 : /*
1284 : * mount traversal - out-of-line part. One note on ->d_flags accesses -
1285 : * dentries are pinned but not locked here, so negative dentry can go
1286 : * positive right under us. Use of smp_load_acquire() provides a barrier
1287 : * sufficient for ->d_inode and ->d_flags consistency.
1288 : */
1289 2897 : static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped,
1290 : int *count, unsigned lookup_flags)
1291 : {
1292 2897 : struct vfsmount *mnt = path->mnt;
1293 2897 : bool need_mntput = false;
1294 2897 : int ret = 0;
1295 :
1296 5741 : while (flags & DCACHE_MANAGED_DENTRY) {
1297 : /* Allow the filesystem to manage the transit without i_mutex
1298 : * being held. */
1299 2919 : if (flags & DCACHE_MANAGE_TRANSIT) {
1300 0 : ret = path->dentry->d_op->d_manage(path, false);
1301 0 : flags = smp_load_acquire(&path->dentry->d_flags);
1302 0 : if (ret < 0)
1303 : break;
1304 : }
1305 :
1306 2919 : if (flags & DCACHE_MOUNTED) { // something's mounted on it..
1307 2919 : struct vfsmount *mounted = lookup_mnt(path);
1308 2919 : if (mounted) { // ... in our namespace
1309 2844 : dput(path->dentry);
1310 2844 : if (need_mntput)
1311 0 : mntput(path->mnt);
1312 2844 : path->mnt = mounted;
1313 2844 : path->dentry = dget(mounted->mnt_root);
1314 : // here we know it's positive
1315 2844 : flags = path->dentry->d_flags;
1316 2844 : need_mntput = true;
1317 2844 : continue;
1318 : }
1319 : }
1320 :
1321 75 : if (!(flags & DCACHE_NEED_AUTOMOUNT))
1322 : break;
1323 :
1324 : // uncovered automount point
1325 0 : ret = follow_automount(path, count, lookup_flags);
1326 0 : flags = smp_load_acquire(&path->dentry->d_flags);
1327 0 : if (ret < 0)
1328 : break;
1329 : }
1330 :
1331 2897 : if (ret == -EISDIR)
1332 0 : ret = 0;
1333 : // possible if you race with several mount --move
1334 2897 : if (need_mntput && path->mnt == mnt)
1335 0 : mntput(path->mnt);
1336 2897 : if (!ret && unlikely(d_flags_negative(flags)))
1337 0 : ret = -ENOENT;
1338 2897 : *jumped = need_mntput;
1339 2897 : return ret;
1340 : }
1341 :
1342 45207 : static inline int traverse_mounts(struct path *path, bool *jumped,
1343 : int *count, unsigned lookup_flags)
1344 : {
1345 45207 : unsigned flags = smp_load_acquire(&path->dentry->d_flags);
1346 :
1347 : /* fastpath */
1348 45209 : if (likely(!(flags & DCACHE_MANAGED_DENTRY))) {
1349 42312 : *jumped = false;
1350 42312 : if (unlikely(d_flags_negative(flags)))
1351 : return -ENOENT;
1352 33771 : return 0;
1353 : }
1354 2897 : return __traverse_mounts(path, flags, jumped, count, lookup_flags);
1355 : }
1356 :
1357 0 : int follow_down_one(struct path *path)
1358 : {
1359 0 : struct vfsmount *mounted;
1360 :
1361 0 : mounted = lookup_mnt(path);
1362 0 : if (mounted) {
1363 0 : dput(path->dentry);
1364 0 : mntput(path->mnt);
1365 0 : path->mnt = mounted;
1366 0 : path->dentry = dget(mounted->mnt_root);
1367 0 : return 1;
1368 : }
1369 : return 0;
1370 : }
1371 : EXPORT_SYMBOL(follow_down_one);
1372 :
1373 : /*
1374 : * Follow down to the covering mount currently visible to userspace. At each
1375 : * point, the filesystem owning that dentry may be queried as to whether the
1376 : * caller is permitted to proceed or not.
1377 : */
1378 0 : int follow_down(struct path *path)
1379 : {
1380 0 : struct vfsmount *mnt = path->mnt;
1381 0 : bool jumped;
1382 0 : int ret = traverse_mounts(path, &jumped, NULL, 0);
1383 :
1384 0 : if (path->mnt != mnt)
1385 0 : mntput(mnt);
1386 0 : return ret;
1387 : }
1388 : EXPORT_SYMBOL(follow_down);
1389 :
1390 : /*
1391 : * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1392 : * we meet a managed dentry that would need blocking.
1393 : */
1394 159460 : static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1395 : struct inode **inode, unsigned *seqp)
1396 : {
1397 159460 : struct dentry *dentry = path->dentry;
1398 159460 : unsigned int flags = dentry->d_flags;
1399 :
1400 159460 : if (likely(!(flags & DCACHE_MANAGED_DENTRY)))
1401 : return true;
1402 :
1403 29752 : if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1404 : return false;
1405 :
1406 80904 : for (;;) {
1407 : /*
1408 : * Don't forget we might have a non-mountpoint managed dentry
1409 : * that wants to block transit.
1410 : */
1411 55328 : if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) {
1412 0 : int res = dentry->d_op->d_manage(path, true);
1413 0 : if (res)
1414 0 : return res == -EISDIR;
1415 0 : flags = dentry->d_flags;
1416 : }
1417 :
1418 55328 : if (flags & DCACHE_MOUNTED) {
1419 30839 : struct mount *mounted = __lookup_mnt(path->mnt, dentry);
1420 30841 : if (mounted) {
1421 25577 : path->mnt = &mounted->mnt;
1422 25577 : dentry = path->dentry = mounted->mnt.mnt_root;
1423 25577 : nd->flags |= LOOKUP_JUMPED;
1424 25577 : *seqp = read_seqcount_begin(&dentry->d_seq);
1425 25576 : *inode = dentry->d_inode;
1426 : /*
1427 : * We don't need to re-check ->d_seq after this
1428 : * ->d_inode read - there will be an RCU delay
1429 : * between mount hash removal and ->mnt_root
1430 : * becoming unpinned.
1431 : */
1432 25576 : flags = dentry->d_flags;
1433 25576 : continue;
1434 : }
1435 5264 : if (read_seqretry(&mount_lock, nd->m_seq))
1436 : return false;
1437 : }
1438 29750 : return !(flags & DCACHE_NEED_AUTOMOUNT);
1439 : }
1440 : }
1441 :
1442 218030 : static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1443 : struct path *path, struct inode **inode,
1444 : unsigned int *seqp)
1445 : {
1446 218030 : bool jumped;
1447 218030 : int ret;
1448 :
1449 218030 : path->mnt = nd->path.mnt;
1450 218030 : path->dentry = dentry;
1451 218030 : if (nd->flags & LOOKUP_RCU) {
1452 172823 : unsigned int seq = *seqp;
1453 172823 : if (unlikely(!*inode))
1454 : return -ENOENT;
1455 159458 : if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1456 : return 0;
1457 5 : if (!try_to_unlazy_next(nd, dentry, seq))
1458 : return -ECHILD;
1459 : // *path might've been clobbered by __follow_mount_rcu()
1460 0 : path->mnt = nd->path.mnt;
1461 0 : path->dentry = dentry;
1462 : }
1463 45207 : ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags);
1464 45208 : if (jumped) {
1465 2844 : if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1466 : ret = -EXDEV;
1467 : else
1468 2844 : nd->flags |= LOOKUP_JUMPED;
1469 : }
1470 45208 : if (unlikely(ret)) {
1471 8540 : dput(path->dentry);
1472 8539 : if (path->mnt != nd->path.mnt)
1473 0 : mntput(path->mnt);
1474 : } else {
1475 36668 : *inode = d_backing_inode(path->dentry);
1476 36668 : *seqp = 0; /* out of RCU mode, so the value doesn't matter */
1477 : }
1478 : return ret;
1479 : }
1480 :
1481 : /*
1482 : * This looks up the name in dcache and possibly revalidates the found dentry.
1483 : * NULL is returned if the dentry does not exist in the cache.
1484 : */
1485 12678 : static struct dentry *lookup_dcache(const struct qstr *name,
1486 : struct dentry *dir,
1487 : unsigned int flags)
1488 : {
1489 12678 : struct dentry *dentry = d_lookup(dir, name);
1490 12678 : if (dentry) {
1491 5056 : int error = d_revalidate(dentry, flags);
1492 5056 : if (unlikely(error <= 0)) {
1493 0 : if (!error)
1494 0 : d_invalidate(dentry);
1495 0 : dput(dentry);
1496 0 : return ERR_PTR(error);
1497 : }
1498 : }
1499 : return dentry;
1500 : }
1501 :
1502 : /*
1503 : * Parent directory has inode locked exclusive. This is one
1504 : * and only case when ->lookup() gets called on non in-lookup
1505 : * dentries - as the matter of fact, this only gets called
1506 : * when directory is guaranteed to have no in-lookup children
1507 : * at all.
1508 : */
1509 7115 : static struct dentry *__lookup_hash(const struct qstr *name,
1510 : struct dentry *base, unsigned int flags)
1511 : {
1512 7115 : struct dentry *dentry = lookup_dcache(name, base, flags);
1513 7115 : struct dentry *old;
1514 7115 : struct inode *dir = base->d_inode;
1515 :
1516 7115 : if (dentry)
1517 : return dentry;
1518 :
1519 : /* Don't create child dentry for a dead directory. */
1520 2085 : if (unlikely(IS_DEADDIR(dir)))
1521 7115 : return ERR_PTR(-ENOENT);
1522 :
1523 2085 : dentry = d_alloc(base, name);
1524 2085 : if (unlikely(!dentry))
1525 7115 : return ERR_PTR(-ENOMEM);
1526 :
1527 2085 : old = dir->i_op->lookup(dir, dentry, flags);
1528 2085 : if (unlikely(old)) {
1529 0 : dput(dentry);
1530 0 : dentry = old;
1531 : }
1532 : return dentry;
1533 : }
1534 :
1535 216893 : static struct dentry *lookup_fast(struct nameidata *nd,
1536 : struct inode **inode,
1537 : unsigned *seqp)
1538 : {
1539 216893 : struct dentry *dentry, *parent = nd->path.dentry;
1540 216893 : int status = 1;
1541 :
1542 : /*
1543 : * Rename seqlock is not required here because in the off chance
1544 : * of a false negative due to a concurrent rename, the caller is
1545 : * going to fall back to non-racy lookup.
1546 : */
1547 216893 : if (nd->flags & LOOKUP_RCU) {
1548 186821 : unsigned seq;
1549 186821 : dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1550 186856 : if (unlikely(!dentry)) {
1551 11984 : if (!try_to_unlazy(nd))
1552 184716 : return ERR_PTR(-ECHILD);
1553 11976 : return NULL;
1554 : }
1555 :
1556 : /*
1557 : * This sequence count validates that the inode matches
1558 : * the dentry name information from lookup.
1559 : */
1560 174872 : *inode = d_backing_inode(dentry);
1561 174872 : if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1562 184716 : return ERR_PTR(-ECHILD);
1563 :
1564 : /*
1565 : * This sequence count validates that the parent had no
1566 : * changes while we did the lookup of the dentry above.
1567 : *
1568 : * The memory barrier in read_seqcount_begin of child is
1569 : * enough, we can use __read_seqcount_retry here.
1570 : */
1571 174867 : if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1572 184716 : return ERR_PTR(-ECHILD);
1573 :
1574 174867 : *seqp = seq;
1575 174867 : status = d_revalidate(dentry, nd->flags);
1576 174872 : if (likely(status > 0))
1577 : return dentry;
1578 2149 : if (!try_to_unlazy_next(nd, dentry, seq))
1579 184716 : return ERR_PTR(-ECHILD);
1580 2138 : if (status == -ECHILD)
1581 : /* we'd been told to redo it in non-rcu mode */
1582 2138 : status = d_revalidate(dentry, nd->flags);
1583 : } else {
1584 30072 : dentry = __d_lookup(parent, &nd->last);
1585 30071 : if (unlikely(!dentry))
1586 : return NULL;
1587 26066 : status = d_revalidate(dentry, nd->flags);
1588 : }
1589 28206 : if (unlikely(status <= 0)) {
1590 0 : if (!status)
1591 0 : d_invalidate(dentry);
1592 0 : dput(dentry);
1593 0 : return ERR_PTR(status);
1594 : }
1595 : return dentry;
1596 : }
1597 :
1598 : /* Fast lookup failed, do it the slow way */
1599 18799 : static struct dentry *__lookup_slow(const struct qstr *name,
1600 : struct dentry *dir,
1601 : unsigned int flags)
1602 : {
1603 18799 : struct dentry *dentry, *old;
1604 18799 : struct inode *inode = dir->d_inode;
1605 18799 : DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1606 :
1607 : /* Don't go there if it's already dead */
1608 18799 : if (unlikely(IS_DEADDIR(inode)))
1609 18799 : return ERR_PTR(-ENOENT);
1610 18799 : again:
1611 18799 : dentry = d_alloc_parallel(dir, name, &wq);
1612 18798 : if (IS_ERR(dentry))
1613 0 : return dentry;
1614 18798 : if (unlikely(!d_in_lookup(dentry))) {
1615 4 : int error = d_revalidate(dentry, flags);
1616 4 : if (unlikely(error <= 0)) {
1617 0 : if (!error) {
1618 0 : d_invalidate(dentry);
1619 0 : dput(dentry);
1620 0 : goto again;
1621 : }
1622 0 : dput(dentry);
1623 0 : dentry = ERR_PTR(error);
1624 : }
1625 : } else {
1626 18794 : old = inode->i_op->lookup(inode, dentry, flags);
1627 18795 : d_lookup_done(dentry);
1628 18795 : if (unlikely(old)) {
1629 113 : dput(dentry);
1630 113 : dentry = old;
1631 : }
1632 : }
1633 : return dentry;
1634 : }
1635 :
1636 13269 : static struct dentry *lookup_slow(const struct qstr *name,
1637 : struct dentry *dir,
1638 : unsigned int flags)
1639 : {
1640 13269 : struct inode *inode = dir->d_inode;
1641 13269 : struct dentry *res;
1642 13269 : inode_lock_shared(inode);
1643 13270 : res = __lookup_slow(name, dir, flags);
1644 13269 : inode_unlock_shared(inode);
1645 13270 : return res;
1646 : }
1647 :
1648 226338 : static inline int may_lookup(struct user_namespace *mnt_userns,
1649 : struct nameidata *nd)
1650 : {
1651 226338 : if (nd->flags & LOOKUP_RCU) {
1652 204537 : int err = inode_permission(mnt_userns, nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1653 204539 : if (err != -ECHILD || !try_to_unlazy(nd))
1654 195049 : return err;
1655 : }
1656 31291 : return inode_permission(mnt_userns, nd->inode, MAY_EXEC);
1657 : }
1658 :
1659 8159 : static int reserve_stack(struct nameidata *nd, struct path *link, unsigned seq)
1660 : {
1661 8159 : if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1662 : return -ELOOP;
1663 :
1664 8159 : if (likely(nd->depth != EMBEDDED_LEVELS))
1665 : return 0;
1666 0 : if (likely(nd->stack != nd->internal))
1667 : return 0;
1668 0 : if (likely(nd_alloc_stack(nd)))
1669 : return 0;
1670 :
1671 0 : if (nd->flags & LOOKUP_RCU) {
1672 : // we need to grab link before we do unlazy. And we can't skip
1673 : // unlazy even if we fail to grab the link - cleanup needs it
1674 0 : bool grabbed_link = legitimize_path(nd, link, seq);
1675 :
1676 0 : if (!try_to_unlazy(nd) != 0 || !grabbed_link)
1677 : return -ECHILD;
1678 :
1679 0 : if (nd_alloc_stack(nd))
1680 0 : return 0;
1681 : }
1682 : return -ENOMEM;
1683 : }
1684 :
1685 : enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1686 :
1687 8159 : static const char *pick_link(struct nameidata *nd, struct path *link,
1688 : struct inode *inode, unsigned seq, int flags)
1689 : {
1690 8159 : struct saved *last;
1691 8159 : const char *res;
1692 8159 : int error = reserve_stack(nd, link, seq);
1693 :
1694 8159 : if (unlikely(error)) {
1695 0 : if (!(nd->flags & LOOKUP_RCU))
1696 0 : path_put(link);
1697 0 : return ERR_PTR(error);
1698 : }
1699 8159 : last = nd->stack + nd->depth++;
1700 8159 : last->link = *link;
1701 8159 : clear_delayed_call(&last->done);
1702 8159 : last->seq = seq;
1703 :
1704 8159 : if (flags & WALK_TRAILING) {
1705 7118 : error = may_follow_link(nd, inode);
1706 7118 : if (unlikely(error))
1707 0 : return ERR_PTR(error);
1708 : }
1709 :
1710 8159 : if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) ||
1711 8159 : unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW))
1712 8161 : return ERR_PTR(-ELOOP);
1713 :
1714 8159 : if (!(nd->flags & LOOKUP_RCU)) {
1715 1205 : touch_atime(&last->link);
1716 1205 : cond_resched();
1717 6954 : } else if (atime_needs_update(&last->link, inode)) {
1718 20 : if (!try_to_unlazy(nd))
1719 8161 : return ERR_PTR(-ECHILD);
1720 20 : touch_atime(&last->link);
1721 : }
1722 :
1723 16316 : error = security_inode_follow_link(link->dentry, inode,
1724 8159 : nd->flags & LOOKUP_RCU);
1725 8157 : if (unlikely(error))
1726 0 : return ERR_PTR(error);
1727 :
1728 8157 : res = READ_ONCE(inode->i_link);
1729 8157 : if (!res) {
1730 1022 : const char * (*get)(struct dentry *, struct inode *,
1731 : struct delayed_call *);
1732 1022 : get = inode->i_op->get_link;
1733 1022 : if (nd->flags & LOOKUP_RCU) {
1734 639 : res = get(NULL, inode, &last->done);
1735 642 : if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd))
1736 0 : res = get(link->dentry, inode, &last->done);
1737 : } else {
1738 383 : res = get(link->dentry, inode, &last->done);
1739 : }
1740 1026 : if (!res)
1741 123 : goto all_done;
1742 903 : if (IS_ERR(res))
1743 : return res;
1744 : }
1745 8038 : if (*res == '/') {
1746 2457 : error = nd_jump_root(nd);
1747 2457 : if (unlikely(error))
1748 0 : return ERR_PTR(error);
1749 2457 : while (unlikely(*++res == '/'))
1750 2457 : ;
1751 : }
1752 8038 : if (*res)
1753 : return res;
1754 0 : all_done: // pure jump
1755 123 : put_link(nd);
1756 123 : return NULL;
1757 : }
1758 :
1759 : /*
1760 : * Do we need to follow links? We _really_ want to be able
1761 : * to do this check without having to look at inode->i_op,
1762 : * so we keep a cache of "no, this doesn't need follow_link"
1763 : * for the common case.
1764 : */
1765 218022 : static const char *step_into(struct nameidata *nd, int flags,
1766 : struct dentry *dentry, struct inode *inode, unsigned seq)
1767 : {
1768 218022 : struct path path;
1769 218022 : int err = handle_mounts(nd, dentry, &path, &inode, &seq);
1770 :
1771 218022 : if (err < 0)
1772 21909 : return ERR_PTR(err);
1773 196113 : if (likely(!d_is_symlink(path.dentry)) ||
1774 9969 : ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
1775 8159 : (flags & WALK_NOFOLLOW)) {
1776 : /* not a symlink or should not follow */
1777 187954 : if (!(nd->flags & LOOKUP_RCU)) {
1778 35463 : dput(nd->path.dentry);
1779 35461 : if (nd->path.mnt != path.mnt)
1780 2844 : mntput(nd->path.mnt);
1781 : }
1782 187952 : nd->path = path;
1783 187952 : nd->inode = inode;
1784 187952 : nd->seq = seq;
1785 187952 : return NULL;
1786 : }
1787 8159 : if (nd->flags & LOOKUP_RCU) {
1788 : /* make sure that d_is_symlink above matches inode */
1789 6954 : if (read_seqcount_retry(&path.dentry->d_seq, seq))
1790 218021 : return ERR_PTR(-ECHILD);
1791 : } else {
1792 1205 : if (path.mnt == nd->path.mnt)
1793 1205 : mntget(path.mnt);
1794 : }
1795 8161 : return pick_link(nd, &path, inode, seq, flags);
1796 : }
1797 :
1798 107 : static struct dentry *follow_dotdot_rcu(struct nameidata *nd,
1799 : struct inode **inodep,
1800 : unsigned *seqp)
1801 : {
1802 107 : struct dentry *parent, *old;
1803 :
1804 107 : if (path_equal(&nd->path, &nd->root))
1805 4 : goto in_root;
1806 103 : if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1807 0 : struct path path;
1808 0 : unsigned seq;
1809 0 : if (!choose_mountpoint_rcu(real_mount(nd->path.mnt),
1810 0 : &nd->root, &path, &seq))
1811 0 : goto in_root;
1812 0 : if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1813 0 : return ERR_PTR(-ECHILD);
1814 0 : nd->path = path;
1815 0 : nd->inode = path.dentry->d_inode;
1816 0 : nd->seq = seq;
1817 0 : if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1818 0 : return ERR_PTR(-ECHILD);
1819 : /* we know that mountpoint was pinned */
1820 : }
1821 103 : old = nd->path.dentry;
1822 103 : parent = old->d_parent;
1823 103 : *inodep = parent->d_inode;
1824 103 : *seqp = read_seqcount_begin(&parent->d_seq);
1825 103 : if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1826 107 : return ERR_PTR(-ECHILD);
1827 103 : if (unlikely(!path_connected(nd->path.mnt, parent)))
1828 0 : return ERR_PTR(-ECHILD);
1829 : return parent;
1830 4 : in_root:
1831 4 : if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1832 107 : return ERR_PTR(-ECHILD);
1833 4 : if (unlikely(nd->flags & LOOKUP_BENEATH))
1834 0 : return ERR_PTR(-ECHILD);
1835 : return NULL;
1836 : }
1837 :
1838 1060 : static struct dentry *follow_dotdot(struct nameidata *nd,
1839 : struct inode **inodep,
1840 : unsigned *seqp)
1841 : {
1842 1060 : struct dentry *parent;
1843 :
1844 1060 : if (path_equal(&nd->path, &nd->root))
1845 0 : goto in_root;
1846 1060 : if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1847 0 : struct path path;
1848 :
1849 0 : if (!choose_mountpoint(real_mount(nd->path.mnt),
1850 0 : &nd->root, &path))
1851 0 : goto in_root;
1852 0 : path_put(&nd->path);
1853 0 : nd->path = path;
1854 0 : nd->inode = path.dentry->d_inode;
1855 0 : if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1856 0 : return ERR_PTR(-EXDEV);
1857 : }
1858 : /* rare case of legitimate dget_parent()... */
1859 1060 : parent = dget_parent(nd->path.dentry);
1860 1060 : if (unlikely(!path_connected(nd->path.mnt, parent))) {
1861 0 : dput(parent);
1862 0 : return ERR_PTR(-ENOENT);
1863 : }
1864 1060 : *seqp = 0;
1865 1060 : *inodep = parent->d_inode;
1866 1060 : return parent;
1867 :
1868 0 : in_root:
1869 0 : if (unlikely(nd->flags & LOOKUP_BENEATH))
1870 1060 : return ERR_PTR(-EXDEV);
1871 0 : dget(nd->path.dentry);
1872 : return NULL;
1873 : }
1874 :
1875 8858 : static const char *handle_dots(struct nameidata *nd, int type)
1876 : {
1877 8858 : if (type == LAST_DOTDOT) {
1878 1167 : const char *error = NULL;
1879 1167 : struct dentry *parent;
1880 1167 : struct inode *inode;
1881 1167 : unsigned seq;
1882 :
1883 1167 : if (!nd->root.mnt) {
1884 653 : error = ERR_PTR(set_root(nd));
1885 653 : if (error)
1886 0 : return error;
1887 : }
1888 1167 : if (nd->flags & LOOKUP_RCU)
1889 107 : parent = follow_dotdot_rcu(nd, &inode, &seq);
1890 : else
1891 1060 : parent = follow_dotdot(nd, &inode, &seq);
1892 1167 : if (IS_ERR(parent))
1893 0 : return ERR_CAST(parent);
1894 1167 : if (unlikely(!parent))
1895 4 : error = step_into(nd, WALK_NOFOLLOW,
1896 : nd->path.dentry, nd->inode, nd->seq);
1897 : else
1898 1163 : error = step_into(nd, WALK_NOFOLLOW,
1899 : parent, inode, seq);
1900 1167 : if (unlikely(error))
1901 : return error;
1902 :
1903 1167 : if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1904 : /*
1905 : * If there was a racing rename or mount along our
1906 : * path, then we can't be sure that ".." hasn't jumped
1907 : * above nd->root (and so userspace should retry or use
1908 : * some fallback).
1909 : */
1910 0 : smp_rmb();
1911 0 : if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq)))
1912 0 : return ERR_PTR(-EAGAIN);
1913 0 : if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq)))
1914 0 : return ERR_PTR(-EAGAIN);
1915 : }
1916 : }
1917 : return NULL;
1918 : }
1919 :
1920 201314 : static const char *walk_component(struct nameidata *nd, int flags)
1921 : {
1922 201314 : struct dentry *dentry;
1923 201314 : struct inode *inode;
1924 201314 : unsigned seq;
1925 : /*
1926 : * "." and ".." are special - ".." especially so because it has
1927 : * to be able to know about the current root directory and
1928 : * parent relationships.
1929 : */
1930 201314 : if (unlikely(nd->last_type != LAST_NORM)) {
1931 8821 : if (!(flags & WALK_MORE) && nd->depth)
1932 0 : put_link(nd);
1933 8821 : return handle_dots(nd, nd->last_type);
1934 : }
1935 192493 : dentry = lookup_fast(nd, &inode, &seq);
1936 192521 : if (IS_ERR(dentry))
1937 201318 : return ERR_CAST(dentry);
1938 192506 : if (unlikely(!dentry)) {
1939 13261 : dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
1940 13261 : if (IS_ERR(dentry))
1941 201318 : return ERR_CAST(dentry);
1942 : }
1943 192393 : if (!(flags & WALK_MORE) && nd->depth)
1944 2360 : put_link(nd);
1945 192393 : return step_into(nd, flags, dentry, inode, seq);
1946 : }
1947 :
1948 : /*
1949 : * We can do the critical dentry name comparison and hashing
1950 : * operations one word at a time, but we are limited to:
1951 : *
1952 : * - Architectures with fast unaligned word accesses. We could
1953 : * do a "get_unaligned()" if this helps and is sufficiently
1954 : * fast.
1955 : *
1956 : * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1957 : * do not trap on the (extremely unlikely) case of a page
1958 : * crossing operation.
1959 : *
1960 : * - Furthermore, we need an efficient 64-bit compile for the
1961 : * 64-bit case in order to generate the "number of bytes in
1962 : * the final mask". Again, that could be replaced with a
1963 : * efficient population count instruction or similar.
1964 : */
1965 : #ifdef CONFIG_DCACHE_WORD_ACCESS
1966 :
1967 : #include <asm/word-at-a-time.h>
1968 :
1969 : #ifdef HASH_MIX
1970 :
1971 : /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1972 :
1973 : #elif defined(CONFIG_64BIT)
1974 : /*
1975 : * Register pressure in the mixing function is an issue, particularly
1976 : * on 32-bit x86, but almost any function requires one state value and
1977 : * one temporary. Instead, use a function designed for two state values
1978 : * and no temporaries.
1979 : *
1980 : * This function cannot create a collision in only two iterations, so
1981 : * we have two iterations to achieve avalanche. In those two iterations,
1982 : * we have six layers of mixing, which is enough to spread one bit's
1983 : * influence out to 2^6 = 64 state bits.
1984 : *
1985 : * Rotate constants are scored by considering either 64 one-bit input
1986 : * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1987 : * probability of that delta causing a change to each of the 128 output
1988 : * bits, using a sample of random initial states.
1989 : *
1990 : * The Shannon entropy of the computed probabilities is then summed
1991 : * to produce a score. Ideally, any input change has a 50% chance of
1992 : * toggling any given output bit.
1993 : *
1994 : * Mixing scores (in bits) for (12,45):
1995 : * Input delta: 1-bit 2-bit
1996 : * 1 round: 713.3 42542.6
1997 : * 2 rounds: 2753.7 140389.8
1998 : * 3 rounds: 5954.1 233458.2
1999 : * 4 rounds: 7862.6 256672.2
2000 : * Perfect: 8192 258048
2001 : * (64*128) (64*63/2 * 128)
2002 : */
2003 : #define HASH_MIX(x, y, a) \
2004 : ( x ^= (a), \
2005 : y ^= x, x = rol64(x,12),\
2006 : x += y, y = rol64(y,45),\
2007 : y *= 9 )
2008 :
2009 : /*
2010 : * Fold two longs into one 32-bit hash value. This must be fast, but
2011 : * latency isn't quite as critical, as there is a fair bit of additional
2012 : * work done before the hash value is used.
2013 : */
2014 234798 : static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2015 : {
2016 234798 : y ^= x * GOLDEN_RATIO_64;
2017 234798 : y *= GOLDEN_RATIO_64;
2018 234798 : return y >> 32;
2019 : }
2020 :
2021 : #else /* 32-bit case */
2022 :
2023 : /*
2024 : * Mixing scores (in bits) for (7,20):
2025 : * Input delta: 1-bit 2-bit
2026 : * 1 round: 330.3 9201.6
2027 : * 2 rounds: 1246.4 25475.4
2028 : * 3 rounds: 1907.1 31295.1
2029 : * 4 rounds: 2042.3 31718.6
2030 : * Perfect: 2048 31744
2031 : * (32*64) (32*31/2 * 64)
2032 : */
2033 : #define HASH_MIX(x, y, a) \
2034 : ( x ^= (a), \
2035 : y ^= x, x = rol32(x, 7),\
2036 : x += y, y = rol32(y,20),\
2037 : y *= 9 )
2038 :
2039 : static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2040 : {
2041 : /* Use arch-optimized multiply if one exists */
2042 : return __hash_32(y ^ __hash_32(x));
2043 : }
2044 :
2045 : #endif
2046 :
2047 : /*
2048 : * Return the hash of a string of known length. This is carfully
2049 : * designed to match hash_name(), which is the more critical function.
2050 : * In particular, we must end by hashing a final word containing 0..7
2051 : * payload bytes, to match the way that hash_name() iterates until it
2052 : * finds the delimiter after the name.
2053 : */
2054 8448 : unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2055 : {
2056 8448 : unsigned long a, x = 0, y = (unsigned long)salt;
2057 :
2058 15226 : for (;;) {
2059 11837 : if (!len)
2060 327 : goto done;
2061 11510 : a = load_unaligned_zeropad(name);
2062 11510 : if (len < sizeof(unsigned long))
2063 : break;
2064 3389 : HASH_MIX(x, y, a);
2065 3389 : name += sizeof(unsigned long);
2066 3389 : len -= sizeof(unsigned long);
2067 : }
2068 8121 : x ^= a & bytemask_from_count(len);
2069 8448 : done:
2070 8448 : return fold_hash(x, y);
2071 : }
2072 : EXPORT_SYMBOL(full_name_hash);
2073 :
2074 : /* Return the "hash_len" (hash and length) of a null-terminated string */
2075 5 : u64 hashlen_string(const void *salt, const char *name)
2076 : {
2077 5 : unsigned long a = 0, x = 0, y = (unsigned long)salt;
2078 5 : unsigned long adata, mask, len;
2079 5 : const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2080 :
2081 5 : len = 0;
2082 5 : goto inside;
2083 :
2084 2 : do {
2085 2 : HASH_MIX(x, y, a);
2086 2 : len += sizeof(unsigned long);
2087 7 : inside:
2088 7 : a = load_unaligned_zeropad(name+len);
2089 7 : } while (!has_zero(a, &adata, &constants));
2090 :
2091 5 : adata = prep_zero_mask(a, adata, &constants);
2092 5 : mask = create_zero_mask(adata);
2093 5 : x ^= a & zero_bytemask(mask);
2094 :
2095 5 : return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2096 : }
2097 : EXPORT_SYMBOL(hashlen_string);
2098 :
2099 : /*
2100 : * Calculate the length and hash of the path component, and
2101 : * return the "hash_len" as the result.
2102 : */
2103 226345 : static inline u64 hash_name(const void *salt, const char *name)
2104 : {
2105 226345 : unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2106 226345 : unsigned long adata, bdata, mask, len;
2107 226345 : const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2108 :
2109 226345 : len = 0;
2110 226345 : goto inside;
2111 :
2112 80349 : do {
2113 80349 : HASH_MIX(x, y, a);
2114 80349 : len += sizeof(unsigned long);
2115 306694 : inside:
2116 306694 : a = load_unaligned_zeropad(name+len);
2117 306694 : b = a ^ REPEAT_BYTE('/');
2118 306694 : } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2119 :
2120 226345 : adata = prep_zero_mask(a, adata, &constants);
2121 226345 : bdata = prep_zero_mask(b, bdata, &constants);
2122 226345 : mask = create_zero_mask(adata | bdata);
2123 226345 : x ^= a & zero_bytemask(mask);
2124 :
2125 226345 : return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2126 : }
2127 :
2128 : #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2129 :
2130 : /* Return the hash of a string of known length */
2131 : unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2132 : {
2133 : unsigned long hash = init_name_hash(salt);
2134 : while (len--)
2135 : hash = partial_name_hash((unsigned char)*name++, hash);
2136 : return end_name_hash(hash);
2137 : }
2138 : EXPORT_SYMBOL(full_name_hash);
2139 :
2140 : /* Return the "hash_len" (hash and length) of a null-terminated string */
2141 : u64 hashlen_string(const void *salt, const char *name)
2142 : {
2143 : unsigned long hash = init_name_hash(salt);
2144 : unsigned long len = 0, c;
2145 :
2146 : c = (unsigned char)*name;
2147 : while (c) {
2148 : len++;
2149 : hash = partial_name_hash(c, hash);
2150 : c = (unsigned char)name[len];
2151 : }
2152 : return hashlen_create(end_name_hash(hash), len);
2153 : }
2154 : EXPORT_SYMBOL(hashlen_string);
2155 :
2156 : /*
2157 : * We know there's a real path component here of at least
2158 : * one character.
2159 : */
2160 : static inline u64 hash_name(const void *salt, const char *name)
2161 : {
2162 : unsigned long hash = init_name_hash(salt);
2163 : unsigned long len = 0, c;
2164 :
2165 : c = (unsigned char)*name;
2166 : do {
2167 : len++;
2168 : hash = partial_name_hash(c, hash);
2169 : c = (unsigned char)name[len];
2170 : } while (c && c != '/');
2171 : return hashlen_create(end_name_hash(hash), len);
2172 : }
2173 :
2174 : #endif
2175 :
2176 : /*
2177 : * Name resolution.
2178 : * This is the basic name resolution function, turning a pathname into
2179 : * the final dentry. We expect 'base' to be positive and a directory.
2180 : *
2181 : * Returns 0 and nd will have valid dentry and mnt on success.
2182 : * Returns error and drops reference to input namei data on failure.
2183 : */
2184 114110 : static int link_path_walk(const char *name, struct nameidata *nd)
2185 : {
2186 114110 : int depth = 0; // depth <= nd->depth
2187 114110 : int err;
2188 :
2189 114110 : nd->last_type = LAST_ROOT;
2190 114110 : nd->flags |= LOOKUP_PARENT;
2191 114110 : if (IS_ERR(name))
2192 16185 : return PTR_ERR(name);
2193 150375 : while (*name=='/')
2194 52450 : name++;
2195 97925 : if (!*name) {
2196 7317 : nd->dir_mode = 0; // short-circuit the 'hardening' idiocy
2197 7317 : return 0;
2198 : }
2199 :
2200 : /* At this point we know we have a real path component. */
2201 226332 : for(;;) {
2202 226332 : struct user_namespace *mnt_userns;
2203 226332 : const char *link;
2204 226332 : u64 hash_len;
2205 226332 : int type;
2206 :
2207 226332 : mnt_userns = mnt_user_ns(nd->path.mnt);
2208 226338 : err = may_lookup(mnt_userns, nd);
2209 226355 : if (err)
2210 9 : return err;
2211 :
2212 226346 : hash_len = hash_name(nd->path.dentry, name);
2213 :
2214 226350 : type = LAST_NORM;
2215 226350 : if (name[0] == '.') switch (hashlen_len(hash_len)) {
2216 1167 : case 2:
2217 1167 : if (name[1] == '.') {
2218 1167 : type = LAST_DOTDOT;
2219 1167 : nd->flags |= LOOKUP_JUMPED;
2220 : }
2221 : break;
2222 377 : case 1:
2223 377 : type = LAST_DOT;
2224 : }
2225 226350 : if (likely(type == LAST_NORM)) {
2226 224806 : struct dentry *parent = nd->path.dentry;
2227 224806 : nd->flags &= ~LOOKUP_JUMPED;
2228 224806 : if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2229 0 : struct qstr this = { { .hash_len = hash_len }, .name = name };
2230 0 : err = parent->d_op->d_hash(parent, &this);
2231 0 : if (err < 0)
2232 0 : return err;
2233 0 : hash_len = this.hash_len;
2234 0 : name = this.name;
2235 : }
2236 : }
2237 :
2238 226350 : nd->last.hash_len = hash_len;
2239 226350 : nd->last.name = name;
2240 226350 : nd->last_type = type;
2241 :
2242 226350 : name += hashlen_len(hash_len);
2243 226350 : if (!*name)
2244 85329 : goto OK;
2245 : /*
2246 : * If it wasn't NUL, we know it was '/'. Skip that
2247 : * slash, and continue until no more slashes.
2248 : */
2249 141301 : do {
2250 141301 : name++;
2251 141301 : } while (unlikely(*name == '/'));
2252 141021 : if (unlikely(!*name)) {
2253 316 : OK:
2254 : /* pathname or trailing symlink, done */
2255 85645 : if (!depth) {
2256 84604 : nd->dir_uid = i_uid_into_mnt(mnt_userns, nd->inode);
2257 84604 : nd->dir_mode = nd->inode->i_mode;
2258 84604 : nd->flags &= ~LOOKUP_PARENT;
2259 84604 : return 0;
2260 : }
2261 : /* last component of nested symlink */
2262 1041 : name = nd->stack[--depth].name;
2263 1041 : link = walk_component(nd, 0);
2264 : } else {
2265 : /* not the last component */
2266 140705 : link = walk_component(nd, WALK_MORE);
2267 : }
2268 141742 : if (unlikely(link)) {
2269 7061 : if (IS_ERR(link))
2270 6018 : return PTR_ERR(link);
2271 : /* a symlink to follow */
2272 1043 : nd->stack[depth++].name = name;
2273 1043 : name = link;
2274 1043 : continue;
2275 : }
2276 134681 : if (unlikely(!d_can_lookup(nd->path.dentry))) {
2277 0 : if (nd->flags & LOOKUP_RCU) {
2278 0 : if (!try_to_unlazy(nd))
2279 : return -ECHILD;
2280 : }
2281 0 : return -ENOTDIR;
2282 : }
2283 : }
2284 : }
2285 :
2286 : /* must be paired with terminate_walk() */
2287 90956 : static const char *path_init(struct nameidata *nd, unsigned flags)
2288 : {
2289 90956 : int error;
2290 90956 : const char *s = nd->name->name;
2291 :
2292 : /* LOOKUP_CACHED requires RCU, ask caller to retry */
2293 90956 : if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED)
2294 90956 : return ERR_PTR(-EAGAIN);
2295 :
2296 90956 : if (!*s)
2297 34 : flags &= ~LOOKUP_RCU;
2298 90956 : if (flags & LOOKUP_RCU)
2299 90838 : rcu_read_lock();
2300 :
2301 90954 : nd->flags = flags | LOOKUP_JUMPED;
2302 90954 : nd->depth = 0;
2303 :
2304 413543 : nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2305 132856 : nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2306 90948 : smp_rmb();
2307 :
2308 90949 : if (flags & LOOKUP_ROOT) {
2309 0 : struct dentry *root = nd->root.dentry;
2310 0 : struct inode *inode = root->d_inode;
2311 0 : if (*s && unlikely(!d_can_lookup(root)))
2312 90956 : return ERR_PTR(-ENOTDIR);
2313 0 : nd->path = nd->root;
2314 0 : nd->inode = inode;
2315 0 : if (flags & LOOKUP_RCU) {
2316 0 : nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2317 0 : nd->root_seq = nd->seq;
2318 : } else {
2319 0 : path_get(&nd->path);
2320 : }
2321 0 : return s;
2322 : }
2323 :
2324 90949 : nd->root.mnt = NULL;
2325 90949 : nd->path.mnt = NULL;
2326 90949 : nd->path.dentry = NULL;
2327 :
2328 : /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2329 90949 : if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2330 52454 : error = nd_jump_root(nd);
2331 52457 : if (unlikely(error))
2332 0 : return ERR_PTR(error);
2333 : return s;
2334 : }
2335 :
2336 : /* Relative pathname -- get the starting-point it is relative to. */
2337 38495 : if (nd->dfd == AT_FDCWD) {
2338 7710 : if (flags & LOOKUP_RCU) {
2339 7710 : struct fs_struct *fs = current->fs;
2340 7710 : unsigned seq;
2341 :
2342 7710 : do {
2343 7710 : seq = read_seqcount_begin(&fs->seq);
2344 7710 : nd->path = fs->pwd;
2345 7710 : nd->inode = nd->path.dentry->d_inode;
2346 7710 : nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2347 7710 : } while (read_seqcount_retry(&fs->seq, seq));
2348 : } else {
2349 0 : get_fs_pwd(current->fs, &nd->path);
2350 0 : nd->inode = nd->path.dentry->d_inode;
2351 : }
2352 : } else {
2353 : /* Caller must check execute permissions on the starting path component */
2354 30785 : struct fd f = fdget_raw(nd->dfd);
2355 30792 : struct dentry *dentry;
2356 :
2357 30792 : if (!f.file)
2358 90956 : return ERR_PTR(-EBADF);
2359 :
2360 30792 : dentry = f.file->f_path.dentry;
2361 :
2362 30792 : if (*s && unlikely(!d_can_lookup(dentry))) {
2363 0 : fdput(f);
2364 0 : return ERR_PTR(-ENOTDIR);
2365 : }
2366 :
2367 30792 : nd->path = f.file->f_path;
2368 30792 : if (flags & LOOKUP_RCU) {
2369 30713 : nd->inode = nd->path.dentry->d_inode;
2370 30713 : nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2371 : } else {
2372 79 : path_get(&nd->path);
2373 79 : nd->inode = nd->path.dentry->d_inode;
2374 : }
2375 30799 : fdput(f);
2376 : }
2377 :
2378 : /* For scoped-lookups we need to set the root to the dirfd as well. */
2379 38499 : if (flags & LOOKUP_IS_SCOPED) {
2380 0 : nd->root = nd->path;
2381 0 : if (flags & LOOKUP_RCU) {
2382 0 : nd->root_seq = nd->seq;
2383 : } else {
2384 0 : path_get(&nd->root);
2385 0 : nd->flags |= LOOKUP_ROOT_GRABBED;
2386 : }
2387 : }
2388 : return s;
2389 : }
2390 :
2391 59591 : static inline const char *lookup_last(struct nameidata *nd)
2392 : {
2393 59591 : if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2394 179 : nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2395 :
2396 59591 : return walk_component(nd, WALK_TRAILING);
2397 : }
2398 :
2399 114 : static int handle_lookup_down(struct nameidata *nd)
2400 : {
2401 114 : if (!(nd->flags & LOOKUP_RCU))
2402 114 : dget(nd->path.dentry);
2403 114 : return PTR_ERR(step_into(nd, WALK_NOFOLLOW,
2404 : nd->path.dentry, nd->inode, nd->seq));
2405 : }
2406 :
2407 : /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2408 59125 : static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2409 : {
2410 59125 : const char *s = path_init(nd, flags);
2411 59123 : int err;
2412 :
2413 59123 : if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2414 0 : err = handle_lookup_down(nd);
2415 0 : if (unlikely(err < 0))
2416 0 : s = ERR_PTR(err);
2417 : }
2418 :
2419 72706 : while (!(err = link_path_walk(s, nd)) &&
2420 59593 : (s = lookup_last(nd)) != NULL)
2421 72713 : ;
2422 59122 : if (!err)
2423 46010 : err = complete_walk(nd);
2424 :
2425 59120 : if (!err && nd->flags & LOOKUP_DIRECTORY)
2426 1610 : if (!d_can_lookup(nd->path.dentry))
2427 : err = -ENOTDIR;
2428 59120 : if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2429 114 : err = handle_lookup_down(nd);
2430 114 : nd->flags &= ~LOOKUP_JUMPED; // no d_weak_revalidate(), please...
2431 : }
2432 59120 : if (!err) {
2433 45962 : *path = nd->path;
2434 45962 : nd->path.mnt = NULL;
2435 45962 : nd->path.dentry = NULL;
2436 : }
2437 59120 : terminate_walk(nd);
2438 59119 : return err;
2439 : }
2440 :
2441 22782 : int filename_lookup(int dfd, struct filename *name, unsigned flags,
2442 : struct path *path, struct path *root)
2443 : {
2444 22782 : int retval;
2445 22782 : struct nameidata nd;
2446 22782 : if (IS_ERR(name))
2447 0 : return PTR_ERR(name);
2448 22782 : if (unlikely(root)) {
2449 0 : nd.root = *root;
2450 0 : flags |= LOOKUP_ROOT;
2451 : }
2452 22782 : set_nameidata(&nd, dfd, name);
2453 22782 : retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2454 22784 : if (unlikely(retval == -ECHILD))
2455 9 : retval = path_lookupat(&nd, flags, path);
2456 22784 : if (unlikely(retval == -ESTALE))
2457 0 : retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2458 :
2459 22784 : if (likely(!retval))
2460 22784 : audit_inode(name, path->dentry,
2461 : flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2462 22784 : restore_nameidata();
2463 22785 : putname(name);
2464 22785 : return retval;
2465 : }
2466 :
2467 : /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2468 7860 : static int path_parentat(struct nameidata *nd, unsigned flags,
2469 : struct path *parent)
2470 : {
2471 7860 : const char *s = path_init(nd, flags);
2472 7860 : int err = link_path_walk(s, nd);
2473 7860 : if (!err)
2474 7129 : err = complete_walk(nd);
2475 7860 : if (!err) {
2476 7129 : *parent = nd->path;
2477 7129 : nd->path.mnt = NULL;
2478 7129 : nd->path.dentry = NULL;
2479 : }
2480 7860 : terminate_walk(nd);
2481 7860 : return err;
2482 : }
2483 :
2484 7860 : static struct filename *filename_parentat(int dfd, struct filename *name,
2485 : unsigned int flags, struct path *parent,
2486 : struct qstr *last, int *type)
2487 : {
2488 7860 : int retval;
2489 7860 : struct nameidata nd;
2490 :
2491 7860 : if (IS_ERR(name))
2492 : return name;
2493 7860 : set_nameidata(&nd, dfd, name);
2494 7860 : retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2495 7860 : if (unlikely(retval == -ECHILD))
2496 0 : retval = path_parentat(&nd, flags, parent);
2497 7860 : if (unlikely(retval == -ESTALE))
2498 0 : retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2499 7860 : if (likely(!retval)) {
2500 7129 : *last = nd.last;
2501 7129 : *type = nd.last_type;
2502 7129 : audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2503 : } else {
2504 731 : putname(name);
2505 731 : name = ERR_PTR(retval);
2506 : }
2507 7860 : restore_nameidata();
2508 7860 : return name;
2509 : }
2510 :
2511 : /* does lookup, returns the object with parent locked */
2512 0 : struct dentry *kern_path_locked(const char *name, struct path *path)
2513 : {
2514 0 : struct filename *filename;
2515 0 : struct dentry *d;
2516 0 : struct qstr last;
2517 0 : int type;
2518 :
2519 0 : filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2520 : &last, &type);
2521 0 : if (IS_ERR(filename))
2522 0 : return ERR_CAST(filename);
2523 0 : if (unlikely(type != LAST_NORM)) {
2524 0 : path_put(path);
2525 0 : putname(filename);
2526 0 : return ERR_PTR(-EINVAL);
2527 : }
2528 0 : inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2529 0 : d = __lookup_hash(&last, path->dentry, 0);
2530 0 : if (IS_ERR(d)) {
2531 0 : inode_unlock(path->dentry->d_inode);
2532 0 : path_put(path);
2533 : }
2534 0 : putname(filename);
2535 0 : return d;
2536 : }
2537 :
2538 668 : int kern_path(const char *name, unsigned int flags, struct path *path)
2539 : {
2540 668 : return filename_lookup(AT_FDCWD, getname_kernel(name),
2541 : flags, path, NULL);
2542 : }
2543 : EXPORT_SYMBOL(kern_path);
2544 :
2545 : /**
2546 : * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2547 : * @dentry: pointer to dentry of the base directory
2548 : * @mnt: pointer to vfs mount of the base directory
2549 : * @name: pointer to file name
2550 : * @flags: lookup flags
2551 : * @path: pointer to struct path to fill
2552 : */
2553 0 : int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2554 : const char *name, unsigned int flags,
2555 : struct path *path)
2556 : {
2557 0 : struct path root = {.mnt = mnt, .dentry = dentry};
2558 : /* the first argument of filename_lookup() is ignored with root */
2559 0 : return filename_lookup(AT_FDCWD, getname_kernel(name),
2560 : flags , path, &root);
2561 : }
2562 : EXPORT_SYMBOL(vfs_path_lookup);
2563 :
2564 5563 : static int lookup_one_len_common(const char *name, struct dentry *base,
2565 : int len, struct qstr *this)
2566 : {
2567 5563 : this->name = name;
2568 5563 : this->len = len;
2569 5563 : this->hash = full_name_hash(base, name, len);
2570 5563 : if (!len)
2571 : return -EACCES;
2572 :
2573 5563 : if (unlikely(name[0] == '.')) {
2574 0 : if (len < 2 || (len == 2 && name[1] == '.'))
2575 : return -EACCES;
2576 : }
2577 :
2578 57153 : while (len--) {
2579 51590 : unsigned int c = *(const unsigned char *)name++;
2580 51590 : if (c == '/' || c == '\0')
2581 : return -EACCES;
2582 : }
2583 : /*
2584 : * See if the low-level filesystem might want
2585 : * to use its own hash..
2586 : */
2587 5563 : if (base->d_flags & DCACHE_OP_HASH) {
2588 0 : int err = base->d_op->d_hash(base, this);
2589 0 : if (err < 0)
2590 : return err;
2591 : }
2592 :
2593 5563 : return inode_permission(&init_user_ns, base->d_inode, MAY_EXEC);
2594 : }
2595 :
2596 : /**
2597 : * try_lookup_one_len - filesystem helper to lookup single pathname component
2598 : * @name: pathname component to lookup
2599 : * @base: base directory to lookup from
2600 : * @len: maximum length @len should be interpreted to
2601 : *
2602 : * Look up a dentry by name in the dcache, returning NULL if it does not
2603 : * currently exist. The function does not try to create a dentry.
2604 : *
2605 : * Note that this routine is purely a helper for filesystem usage and should
2606 : * not be called by generic code.
2607 : *
2608 : * The caller must hold base->i_mutex.
2609 : */
2610 0 : struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2611 : {
2612 0 : struct qstr this;
2613 0 : int err;
2614 :
2615 0 : WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2616 :
2617 0 : err = lookup_one_len_common(name, base, len, &this);
2618 0 : if (err)
2619 0 : return ERR_PTR(err);
2620 :
2621 0 : return lookup_dcache(&this, base, 0);
2622 : }
2623 : EXPORT_SYMBOL(try_lookup_one_len);
2624 :
2625 : /**
2626 : * lookup_one_len - filesystem helper to lookup single pathname component
2627 : * @name: pathname component to lookup
2628 : * @base: base directory to lookup from
2629 : * @len: maximum length @len should be interpreted to
2630 : *
2631 : * Note that this routine is purely a helper for filesystem usage and should
2632 : * not be called by generic code.
2633 : *
2634 : * The caller must hold base->i_mutex.
2635 : */
2636 5531 : struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2637 : {
2638 5531 : struct dentry *dentry;
2639 5531 : struct qstr this;
2640 5531 : int err;
2641 :
2642 5531 : WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2643 :
2644 5531 : err = lookup_one_len_common(name, base, len, &this);
2645 5531 : if (err)
2646 0 : return ERR_PTR(err);
2647 :
2648 5531 : dentry = lookup_dcache(&this, base, 0);
2649 5531 : return dentry ? dentry : __lookup_slow(&this, base, 0);
2650 : }
2651 : EXPORT_SYMBOL(lookup_one_len);
2652 :
2653 : /**
2654 : * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2655 : * @name: pathname component to lookup
2656 : * @base: base directory to lookup from
2657 : * @len: maximum length @len should be interpreted to
2658 : *
2659 : * Note that this routine is purely a helper for filesystem usage and should
2660 : * not be called by generic code.
2661 : *
2662 : * Unlike lookup_one_len, it should be called without the parent
2663 : * i_mutex held, and will take the i_mutex itself if necessary.
2664 : */
2665 32 : struct dentry *lookup_one_len_unlocked(const char *name,
2666 : struct dentry *base, int len)
2667 : {
2668 32 : struct qstr this;
2669 32 : int err;
2670 32 : struct dentry *ret;
2671 :
2672 32 : err = lookup_one_len_common(name, base, len, &this);
2673 32 : if (err)
2674 0 : return ERR_PTR(err);
2675 :
2676 32 : ret = lookup_dcache(&this, base, 0);
2677 32 : if (!ret)
2678 8 : ret = lookup_slow(&this, base, 0);
2679 : return ret;
2680 : }
2681 : EXPORT_SYMBOL(lookup_one_len_unlocked);
2682 :
2683 : /*
2684 : * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2685 : * on negatives. Returns known positive or ERR_PTR(); that's what
2686 : * most of the users want. Note that pinned negative with unlocked parent
2687 : * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2688 : * need to be very careful; pinned positives have ->d_inode stable, so
2689 : * this one avoids such problems.
2690 : */
2691 0 : struct dentry *lookup_positive_unlocked(const char *name,
2692 : struct dentry *base, int len)
2693 : {
2694 0 : struct dentry *ret = lookup_one_len_unlocked(name, base, len);
2695 0 : if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2696 0 : dput(ret);
2697 0 : ret = ERR_PTR(-ENOENT);
2698 : }
2699 0 : return ret;
2700 : }
2701 : EXPORT_SYMBOL(lookup_positive_unlocked);
2702 :
2703 : #ifdef CONFIG_UNIX98_PTYS
2704 0 : int path_pts(struct path *path)
2705 : {
2706 : /* Find something mounted on "pts" in the same directory as
2707 : * the input path.
2708 : */
2709 0 : struct dentry *parent = dget_parent(path->dentry);
2710 0 : struct dentry *child;
2711 0 : struct qstr this = QSTR_INIT("pts", 3);
2712 :
2713 0 : if (unlikely(!path_connected(path->mnt, parent))) {
2714 0 : dput(parent);
2715 0 : return -ENOENT;
2716 : }
2717 0 : dput(path->dentry);
2718 0 : path->dentry = parent;
2719 0 : child = d_hash_and_lookup(parent, &this);
2720 0 : if (!child)
2721 : return -ENOENT;
2722 :
2723 0 : path->dentry = child;
2724 0 : dput(parent);
2725 0 : follow_down(path);
2726 0 : return 0;
2727 : }
2728 : #endif
2729 :
2730 22116 : int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2731 : struct path *path, int *empty)
2732 : {
2733 22116 : return filename_lookup(dfd, getname_flags(name, flags, empty),
2734 : flags, path, NULL);
2735 : }
2736 : EXPORT_SYMBOL(user_path_at_empty);
2737 :
2738 21 : int __check_sticky(struct user_namespace *mnt_userns, struct inode *dir,
2739 : struct inode *inode)
2740 : {
2741 21 : kuid_t fsuid = current_fsuid();
2742 :
2743 21 : if (uid_eq(i_uid_into_mnt(mnt_userns, inode), fsuid))
2744 : return 0;
2745 0 : if (uid_eq(i_uid_into_mnt(mnt_userns, dir), fsuid))
2746 : return 0;
2747 0 : return !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FOWNER);
2748 : }
2749 : EXPORT_SYMBOL(__check_sticky);
2750 :
2751 : /*
2752 : * Check whether we can remove a link victim from directory dir, check
2753 : * whether the type of victim is right.
2754 : * 1. We can't do it if dir is read-only (done in permission())
2755 : * 2. We should have write and exec permissions on dir
2756 : * 3. We can't remove anything from append-only dir
2757 : * 4. We can't do anything with immutable dir (done in permission())
2758 : * 5. If the sticky bit on dir is set we should either
2759 : * a. be owner of dir, or
2760 : * b. be owner of victim, or
2761 : * c. have CAP_FOWNER capability
2762 : * 6. If the victim is append-only or immutable we can't do antyhing with
2763 : * links pointing to it.
2764 : * 7. If the victim has an unknown uid or gid we can't change the inode.
2765 : * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2766 : * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2767 : * 10. We can't remove a root or mountpoint.
2768 : * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2769 : * nfs_async_unlink().
2770 : */
2771 2349 : static int may_delete(struct user_namespace *mnt_userns, struct inode *dir,
2772 : struct dentry *victim, bool isdir)
2773 : {
2774 2349 : struct inode *inode = d_backing_inode(victim);
2775 2349 : int error;
2776 :
2777 2349 : if (d_is_negative(victim))
2778 : return -ENOENT;
2779 2349 : BUG_ON(!inode);
2780 :
2781 2349 : BUG_ON(victim->d_parent->d_inode != dir);
2782 :
2783 : /* Inode writeback is not safe when the uid or gid are invalid. */
2784 2349 : if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) ||
2785 2349 : !gid_valid(i_gid_into_mnt(mnt_userns, inode)))
2786 : return -EOVERFLOW;
2787 :
2788 2349 : audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2789 :
2790 2349 : error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
2791 2349 : if (error)
2792 : return error;
2793 2349 : if (IS_APPEND(dir))
2794 : return -EPERM;
2795 :
2796 2349 : if (check_sticky(mnt_userns, dir, inode) || IS_APPEND(inode) ||
2797 2349 : IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) ||
2798 2349 : HAS_UNMAPPED_ID(mnt_userns, inode))
2799 : return -EPERM;
2800 2349 : if (isdir) {
2801 1072 : if (!d_is_dir(victim))
2802 : return -ENOTDIR;
2803 1072 : if (IS_ROOT(victim))
2804 : return -EBUSY;
2805 2412 : } else if (d_is_dir(victim))
2806 : return -EISDIR;
2807 2207 : if (IS_DEADDIR(dir))
2808 : return -ENOENT;
2809 2207 : if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2810 0 : return -EBUSY;
2811 : return 0;
2812 : }
2813 :
2814 : /* Check whether we can create an object with dentry child in directory
2815 : * dir.
2816 : * 1. We can't do it if child already exists (open has special treatment for
2817 : * this case, but since we are inlined it's OK)
2818 : * 2. We can't do it if dir is read-only (done in permission())
2819 : * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2820 : * 4. We should have write and exec permissions on dir
2821 : * 5. We can't do it if dir is immutable (done in permission())
2822 : */
2823 1952 : static inline int may_create(struct user_namespace *mnt_userns,
2824 : struct inode *dir, struct dentry *child)
2825 : {
2826 1952 : struct user_namespace *s_user_ns;
2827 1952 : audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2828 1952 : if (child->d_inode)
2829 : return -EEXIST;
2830 1952 : if (IS_DEADDIR(dir))
2831 : return -ENOENT;
2832 1952 : s_user_ns = dir->i_sb->s_user_ns;
2833 3904 : if (!kuid_has_mapping(s_user_ns, fsuid_into_mnt(mnt_userns)) ||
2834 1952 : !kgid_has_mapping(s_user_ns, fsgid_into_mnt(mnt_userns)))
2835 0 : return -EOVERFLOW;
2836 1952 : return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
2837 : }
2838 :
2839 : /*
2840 : * p1 and p2 should be directories on the same fs.
2841 : */
2842 428 : struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2843 : {
2844 428 : struct dentry *p;
2845 :
2846 428 : if (p1 == p2) {
2847 414 : inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2848 414 : return NULL;
2849 : }
2850 :
2851 14 : mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2852 :
2853 14 : p = d_ancestor(p2, p1);
2854 14 : if (p) {
2855 0 : inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2856 0 : inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2857 0 : return p;
2858 : }
2859 :
2860 14 : p = d_ancestor(p1, p2);
2861 14 : if (p) {
2862 0 : inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2863 0 : inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2864 0 : return p;
2865 : }
2866 :
2867 14 : inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2868 14 : inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2869 14 : return NULL;
2870 : }
2871 : EXPORT_SYMBOL(lock_rename);
2872 :
2873 428 : void unlock_rename(struct dentry *p1, struct dentry *p2)
2874 : {
2875 428 : inode_unlock(p1->d_inode);
2876 428 : if (p1 != p2) {
2877 14 : inode_unlock(p2->d_inode);
2878 14 : mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2879 : }
2880 428 : }
2881 : EXPORT_SYMBOL(unlock_rename);
2882 :
2883 : /**
2884 : * vfs_create - create new file
2885 : * @mnt_userns: user namespace of the mount the inode was found from
2886 : * @dir: inode of @dentry
2887 : * @dentry: pointer to dentry of the base directory
2888 : * @mode: mode of the new file
2889 : * @want_excl: whether the file must not yet exist
2890 : *
2891 : * Create a new file.
2892 : *
2893 : * If the inode has been found through an idmapped mount the user namespace of
2894 : * the vfsmount must be passed through @mnt_userns. This function will then take
2895 : * care to map the inode according to @mnt_userns before checking permissions.
2896 : * On non-idmapped mounts or if permission checking is to be performed on the
2897 : * raw inode simply passs init_user_ns.
2898 : */
2899 471 : int vfs_create(struct user_namespace *mnt_userns, struct inode *dir,
2900 : struct dentry *dentry, umode_t mode, bool want_excl)
2901 : {
2902 471 : int error = may_create(mnt_userns, dir, dentry);
2903 471 : if (error)
2904 : return error;
2905 :
2906 471 : if (!dir->i_op->create)
2907 : return -EACCES; /* shouldn't it be ENOSYS? */
2908 471 : mode &= S_IALLUGO;
2909 471 : mode |= S_IFREG;
2910 471 : error = security_inode_create(dir, dentry, mode);
2911 471 : if (error)
2912 : return error;
2913 471 : error = dir->i_op->create(mnt_userns, dir, dentry, mode, want_excl);
2914 471 : if (!error)
2915 471 : fsnotify_create(dir, dentry);
2916 : return error;
2917 : }
2918 : EXPORT_SYMBOL(vfs_create);
2919 :
2920 0 : int vfs_mkobj(struct dentry *dentry, umode_t mode,
2921 : int (*f)(struct dentry *, umode_t, void *),
2922 : void *arg)
2923 : {
2924 0 : struct inode *dir = dentry->d_parent->d_inode;
2925 0 : int error = may_create(&init_user_ns, dir, dentry);
2926 0 : if (error)
2927 : return error;
2928 :
2929 0 : mode &= S_IALLUGO;
2930 0 : mode |= S_IFREG;
2931 0 : error = security_inode_create(dir, dentry, mode);
2932 0 : if (error)
2933 : return error;
2934 0 : error = f(dentry, mode, arg);
2935 0 : if (!error)
2936 0 : fsnotify_create(dir, dentry);
2937 : return error;
2938 : }
2939 : EXPORT_SYMBOL(vfs_mkobj);
2940 :
2941 442 : bool may_open_dev(const struct path *path)
2942 : {
2943 442 : return !(path->mnt->mnt_flags & MNT_NODEV) &&
2944 442 : !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2945 : }
2946 :
2947 15593 : static int may_open(struct user_namespace *mnt_userns, const struct path *path,
2948 : int acc_mode, int flag)
2949 : {
2950 15593 : struct dentry *dentry = path->dentry;
2951 15593 : struct inode *inode = dentry->d_inode;
2952 15593 : int error;
2953 :
2954 15593 : if (!inode)
2955 : return -ENOENT;
2956 :
2957 15593 : switch (inode->i_mode & S_IFMT) {
2958 : case S_IFLNK:
2959 : return -ELOOP;
2960 1388 : case S_IFDIR:
2961 1388 : if (acc_mode & MAY_WRITE)
2962 : return -EISDIR;
2963 1388 : if (acc_mode & MAY_EXEC)
2964 : return -EACCES;
2965 : break;
2966 440 : case S_IFBLK:
2967 : case S_IFCHR:
2968 440 : if (!may_open_dev(path))
2969 : return -EACCES;
2970 454 : fallthrough;
2971 : case S_IFIFO:
2972 : case S_IFSOCK:
2973 454 : if (acc_mode & MAY_EXEC)
2974 : return -EACCES;
2975 454 : flag &= ~O_TRUNC;
2976 454 : break;
2977 13740 : case S_IFREG:
2978 13740 : if ((acc_mode & MAY_EXEC) && path_noexec(path))
2979 : return -EACCES;
2980 : break;
2981 : }
2982 :
2983 15582 : error = inode_permission(mnt_userns, inode, MAY_OPEN | acc_mode);
2984 15583 : if (error)
2985 : return error;
2986 :
2987 : /*
2988 : * An append-only file must be opened in append mode for writing.
2989 : */
2990 15581 : if (IS_APPEND(inode)) {
2991 0 : if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2992 : return -EPERM;
2993 0 : if (flag & O_TRUNC)
2994 : return -EPERM;
2995 : }
2996 :
2997 : /* O_NOATIME can only be set by the owner or superuser */
2998 15581 : if (flag & O_NOATIME && !inode_owner_or_capable(mnt_userns, inode))
2999 0 : return -EPERM;
3000 :
3001 : return 0;
3002 : }
3003 :
3004 10 : static int handle_truncate(struct user_namespace *mnt_userns, struct file *filp)
3005 : {
3006 10 : const struct path *path = &filp->f_path;
3007 10 : struct inode *inode = path->dentry->d_inode;
3008 10 : int error = get_write_access(inode);
3009 10 : if (error)
3010 : return error;
3011 : /*
3012 : * Refuse to truncate files with mandatory locks held on them.
3013 : */
3014 10 : error = locks_verify_locked(filp);
3015 10 : if (!error)
3016 10 : error = security_path_truncate(path);
3017 10 : if (!error) {
3018 10 : error = do_truncate(mnt_userns, path->dentry, 0,
3019 : ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3020 : filp);
3021 : }
3022 10 : put_write_access(inode);
3023 10 : return error;
3024 : }
3025 :
3026 0 : static inline int open_to_namei_flags(int flag)
3027 : {
3028 0 : if ((flag & O_ACCMODE) == 3)
3029 0 : flag--;
3030 0 : return flag;
3031 : }
3032 :
3033 635 : static int may_o_create(struct user_namespace *mnt_userns,
3034 : const struct path *dir, struct dentry *dentry,
3035 : umode_t mode)
3036 : {
3037 635 : struct user_namespace *s_user_ns;
3038 635 : int error = security_path_mknod(dir, dentry, mode, 0);
3039 635 : if (error)
3040 : return error;
3041 :
3042 635 : s_user_ns = dir->dentry->d_sb->s_user_ns;
3043 1270 : if (!kuid_has_mapping(s_user_ns, fsuid_into_mnt(mnt_userns)) ||
3044 635 : !kgid_has_mapping(s_user_ns, fsgid_into_mnt(mnt_userns)))
3045 0 : return -EOVERFLOW;
3046 :
3047 635 : error = inode_permission(mnt_userns, dir->dentry->d_inode,
3048 : MAY_WRITE | MAY_EXEC);
3049 635 : if (error)
3050 : return error;
3051 :
3052 631 : return security_inode_create(dir->dentry->d_inode, dentry, mode);
3053 : }
3054 :
3055 : /*
3056 : * Attempt to atomically look up, create and open a file from a negative
3057 : * dentry.
3058 : *
3059 : * Returns 0 if successful. The file will have been created and attached to
3060 : * @file by the filesystem calling finish_open().
3061 : *
3062 : * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3063 : * be set. The caller will need to perform the open themselves. @path will
3064 : * have been updated to point to the new dentry. This may be negative.
3065 : *
3066 : * Returns an error code otherwise.
3067 : */
3068 0 : static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
3069 : struct file *file,
3070 : int open_flag, umode_t mode)
3071 : {
3072 0 : struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3073 0 : struct inode *dir = nd->path.dentry->d_inode;
3074 0 : int error;
3075 :
3076 0 : if (nd->flags & LOOKUP_DIRECTORY)
3077 0 : open_flag |= O_DIRECTORY;
3078 :
3079 0 : file->f_path.dentry = DENTRY_NOT_SET;
3080 0 : file->f_path.mnt = nd->path.mnt;
3081 0 : error = dir->i_op->atomic_open(dir, dentry, file,
3082 0 : open_to_namei_flags(open_flag), mode);
3083 0 : d_lookup_done(dentry);
3084 0 : if (!error) {
3085 0 : if (file->f_mode & FMODE_OPENED) {
3086 0 : if (unlikely(dentry != file->f_path.dentry)) {
3087 0 : dput(dentry);
3088 0 : dentry = dget(file->f_path.dentry);
3089 : }
3090 0 : } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3091 : error = -EIO;
3092 : } else {
3093 0 : if (file->f_path.dentry) {
3094 0 : dput(dentry);
3095 0 : dentry = file->f_path.dentry;
3096 : }
3097 0 : if (unlikely(d_is_negative(dentry)))
3098 : error = -ENOENT;
3099 : }
3100 : }
3101 0 : if (error) {
3102 0 : dput(dentry);
3103 0 : dentry = ERR_PTR(error);
3104 : }
3105 0 : return dentry;
3106 : }
3107 :
3108 : /*
3109 : * Look up and maybe create and open the last component.
3110 : *
3111 : * Must be called with parent locked (exclusive in O_CREAT case).
3112 : *
3113 : * Returns 0 on success, that is, if
3114 : * the file was successfully atomically created (if necessary) and opened, or
3115 : * the file was not completely opened at this time, though lookups and
3116 : * creations were performed.
3117 : * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3118 : * In the latter case dentry returned in @path might be negative if O_CREAT
3119 : * hadn't been specified.
3120 : *
3121 : * An error code is returned on failure.
3122 : */
3123 3465 : static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3124 : const struct open_flags *op,
3125 : bool got_write)
3126 : {
3127 3465 : struct user_namespace *mnt_userns;
3128 3465 : struct dentry *dir = nd->path.dentry;
3129 3465 : struct inode *dir_inode = dir->d_inode;
3130 3465 : int open_flag = op->open_flag;
3131 3465 : struct dentry *dentry;
3132 3465 : int error, create_error = 0;
3133 3465 : umode_t mode = op->mode;
3134 3465 : DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3135 :
3136 3465 : if (unlikely(IS_DEADDIR(dir_inode)))
3137 3465 : return ERR_PTR(-ENOENT);
3138 :
3139 3465 : file->f_mode &= ~FMODE_CREATED;
3140 3465 : dentry = d_lookup(dir, &nd->last);
3141 3465 : for (;;) {
3142 3465 : if (!dentry) {
3143 3351 : dentry = d_alloc_parallel(dir, &nd->last, &wq);
3144 3351 : if (IS_ERR(dentry))
3145 : return dentry;
3146 : }
3147 3465 : if (d_in_lookup(dentry))
3148 : break;
3149 :
3150 116 : error = d_revalidate(dentry, nd->flags);
3151 116 : if (likely(error > 0))
3152 : break;
3153 0 : if (error)
3154 0 : goto out_dput;
3155 0 : d_invalidate(dentry);
3156 0 : dput(dentry);
3157 0 : dentry = NULL;
3158 : }
3159 3465 : if (dentry->d_inode) {
3160 : /* Cached positive dentry: will open in f_op->open */
3161 : return dentry;
3162 : }
3163 :
3164 : /*
3165 : * Checking write permission is tricky, bacuse we don't know if we are
3166 : * going to actually need it: O_CREAT opens should work as long as the
3167 : * file exists. But checking existence breaks atomicity. The trick is
3168 : * to check access and if not granted clear O_CREAT from the flags.
3169 : *
3170 : * Another problem is returing the "right" error value (e.g. for an
3171 : * O_EXCL open we want to return EEXIST not EROFS).
3172 : */
3173 3356 : if (unlikely(!got_write))
3174 2562 : open_flag &= ~O_TRUNC;
3175 3356 : mnt_userns = mnt_user_ns(nd->path.mnt);
3176 3356 : if (open_flag & O_CREAT) {
3177 636 : if (open_flag & O_EXCL)
3178 313 : open_flag &= ~O_TRUNC;
3179 636 : if (!IS_POSIXACL(dir->d_inode))
3180 636 : mode &= ~current_umask();
3181 636 : if (likely(got_write))
3182 635 : create_error = may_o_create(mnt_userns, &nd->path,
3183 : dentry, mode);
3184 : else
3185 : create_error = -EROFS;
3186 : }
3187 3356 : if (create_error)
3188 5 : open_flag &= ~O_CREAT;
3189 3356 : if (dir_inode->i_op->atomic_open) {
3190 0 : dentry = atomic_open(nd, dentry, file, open_flag, mode);
3191 0 : if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3192 0 : dentry = ERR_PTR(create_error);
3193 0 : return dentry;
3194 : }
3195 :
3196 3356 : if (d_in_lookup(dentry)) {
3197 3349 : struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3198 : nd->flags);
3199 3349 : d_lookup_done(dentry);
3200 3349 : if (unlikely(res)) {
3201 163 : if (IS_ERR(res)) {
3202 163 : error = PTR_ERR(res);
3203 163 : goto out_dput;
3204 : }
3205 0 : dput(dentry);
3206 0 : dentry = res;
3207 : }
3208 : }
3209 :
3210 : /* Negative dentry, just create the file */
3211 3193 : if (!dentry->d_inode && (open_flag & O_CREAT)) {
3212 620 : file->f_mode |= FMODE_CREATED;
3213 620 : audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3214 620 : if (!dir_inode->i_op->create) {
3215 0 : error = -EACCES;
3216 0 : goto out_dput;
3217 : }
3218 :
3219 1240 : error = dir_inode->i_op->create(mnt_userns, dir_inode, dentry,
3220 620 : mode, open_flag & O_EXCL);
3221 620 : if (error)
3222 0 : goto out_dput;
3223 : }
3224 3193 : if (unlikely(create_error) && !dentry->d_inode) {
3225 1 : error = create_error;
3226 1 : goto out_dput;
3227 : }
3228 : return dentry;
3229 :
3230 164 : out_dput:
3231 164 : dput(dentry);
3232 164 : return ERR_PTR(error);
3233 : }
3234 :
3235 25191 : static const char *open_last_lookups(struct nameidata *nd,
3236 : struct file *file, const struct open_flags *op)
3237 : {
3238 25191 : struct dentry *dir = nd->path.dentry;
3239 25191 : int open_flag = op->open_flag;
3240 25191 : bool got_write = false;
3241 25191 : unsigned seq;
3242 25191 : struct inode *inode;
3243 25191 : struct dentry *dentry;
3244 25191 : const char *res;
3245 :
3246 25191 : nd->flags |= op->intent;
3247 :
3248 25191 : if (nd->last_type != LAST_NORM) {
3249 38 : if (nd->depth)
3250 0 : put_link(nd);
3251 38 : return handle_dots(nd, nd->last_type);
3252 : }
3253 :
3254 25153 : if (!(open_flag & O_CREAT)) {
3255 24408 : if (nd->last.name[nd->last.len])
3256 130 : nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3257 : /* we _can_ be in RCU mode here */
3258 24408 : dentry = lookup_fast(nd, &inode, &seq);
3259 24411 : if (IS_ERR(dentry))
3260 25192 : return ERR_CAST(dentry);
3261 24409 : if (likely(dentry))
3262 21689 : goto finish_lookup;
3263 :
3264 2720 : BUG_ON(nd->flags & LOOKUP_RCU);
3265 : } else {
3266 : /* create side of things */
3267 745 : if (nd->flags & LOOKUP_RCU) {
3268 738 : if (!try_to_unlazy(nd))
3269 25192 : return ERR_PTR(-ECHILD);
3270 : }
3271 745 : audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3272 : /* trailing slashes? */
3273 745 : if (unlikely(nd->last.name[nd->last.len]))
3274 25192 : return ERR_PTR(-EISDIR);
3275 : }
3276 :
3277 3465 : if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3278 904 : got_write = !mnt_want_write(nd->path.mnt);
3279 : /*
3280 : * do _not_ fail yet - we might not need that or fail with
3281 : * a different error; let lookup_open() decide; we'll be
3282 : * dropping this one anyway.
3283 : */
3284 : }
3285 3465 : if (open_flag & O_CREAT)
3286 745 : inode_lock(dir->d_inode);
3287 : else
3288 2720 : inode_lock_shared(dir->d_inode);
3289 3465 : dentry = lookup_open(nd, file, op, got_write);
3290 3465 : if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED))
3291 620 : fsnotify_create(dir->d_inode, dentry);
3292 3465 : if (open_flag & O_CREAT)
3293 745 : inode_unlock(dir->d_inode);
3294 : else
3295 2720 : inode_unlock_shared(dir->d_inode);
3296 :
3297 3465 : if (got_write)
3298 902 : mnt_drop_write(nd->path.mnt);
3299 :
3300 3465 : if (IS_ERR(dentry))
3301 25192 : return ERR_CAST(dentry);
3302 :
3303 3301 : if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
3304 620 : dput(nd->path.dentry);
3305 620 : nd->path.dentry = dentry;
3306 620 : return NULL;
3307 : }
3308 :
3309 2681 : finish_lookup:
3310 24370 : if (nd->depth)
3311 5671 : put_link(nd);
3312 24370 : res = step_into(nd, WALK_TRAILING, dentry, inode, seq);
3313 24368 : if (unlikely(res))
3314 9424 : nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3315 : return res;
3316 : }
3317 :
3318 : /*
3319 : * Handle the last step of open()
3320 : */
3321 15603 : static int do_open(struct nameidata *nd,
3322 : struct file *file, const struct open_flags *op)
3323 : {
3324 15603 : struct user_namespace *mnt_userns;
3325 15603 : int open_flag = op->open_flag;
3326 15603 : bool do_truncate;
3327 15603 : int acc_mode;
3328 15603 : int error;
3329 :
3330 15603 : if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3331 14983 : error = complete_walk(nd);
3332 14982 : if (error)
3333 : return error;
3334 : }
3335 15595 : if (!(file->f_mode & FMODE_CREATED))
3336 15595 : audit_inode(nd->name, nd->path.dentry, 0);
3337 15595 : mnt_userns = mnt_user_ns(nd->path.mnt);
3338 15595 : if (open_flag & O_CREAT) {
3339 741 : if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3340 : return -EEXIST;
3341 1476 : if (d_is_dir(nd->path.dentry))
3342 : return -EISDIR;
3343 1476 : error = may_create_in_sticky(mnt_userns, nd,
3344 738 : d_backing_inode(nd->path.dentry));
3345 738 : if (unlikely(error))
3346 : return error;
3347 : }
3348 15592 : if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3349 : return -ENOTDIR;
3350 :
3351 15592 : do_truncate = false;
3352 15592 : acc_mode = op->acc_mode;
3353 15592 : if (file->f_mode & FMODE_CREATED) {
3354 : /* Don't check for write permission, don't truncate */
3355 620 : open_flag &= ~O_TRUNC;
3356 620 : acc_mode = 0;
3357 14972 : } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) {
3358 10 : error = mnt_want_write(nd->path.mnt);
3359 10 : if (error)
3360 : return error;
3361 : do_truncate = true;
3362 : }
3363 15592 : error = may_open(mnt_userns, &nd->path, acc_mode, open_flag);
3364 15594 : if (!error && !(file->f_mode & FMODE_OPENED))
3365 15581 : error = vfs_open(&nd->path, file);
3366 15592 : if (!error)
3367 15361 : error = ima_file_check(file, op->acc_mode);
3368 15592 : if (!error && do_truncate)
3369 10 : error = handle_truncate(mnt_userns, file);
3370 15592 : if (unlikely(error > 0)) {
3371 0 : WARN_ON(1);
3372 0 : error = -EINVAL;
3373 : }
3374 15592 : if (do_truncate)
3375 10 : mnt_drop_write(nd->path.mnt);
3376 : return error;
3377 : }
3378 :
3379 : /**
3380 : * vfs_tmpfile - create tmpfile
3381 : * @mnt_userns: user namespace of the mount the inode was found from
3382 : * @dentry: pointer to dentry of the base directory
3383 : * @mode: mode of the new tmpfile
3384 : * @open_flags: flags
3385 : *
3386 : * Create a temporary file.
3387 : *
3388 : * If the inode has been found through an idmapped mount the user namespace of
3389 : * the vfsmount must be passed through @mnt_userns. This function will then take
3390 : * care to map the inode according to @mnt_userns before checking permissions.
3391 : * On non-idmapped mounts or if permission checking is to be performed on the
3392 : * raw inode simply passs init_user_ns.
3393 : */
3394 5 : struct dentry *vfs_tmpfile(struct user_namespace *mnt_userns,
3395 : struct dentry *dentry, umode_t mode, int open_flag)
3396 : {
3397 5 : struct dentry *child = NULL;
3398 5 : struct inode *dir = dentry->d_inode;
3399 5 : struct inode *inode;
3400 5 : int error;
3401 :
3402 : /* we want directory to be writable */
3403 5 : error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
3404 5 : if (error)
3405 0 : goto out_err;
3406 5 : error = -EOPNOTSUPP;
3407 5 : if (!dir->i_op->tmpfile)
3408 0 : goto out_err;
3409 5 : error = -ENOMEM;
3410 5 : child = d_alloc(dentry, &slash_name);
3411 5 : if (unlikely(!child))
3412 0 : goto out_err;
3413 5 : error = dir->i_op->tmpfile(mnt_userns, dir, child, mode);
3414 5 : if (error)
3415 0 : goto out_err;
3416 5 : error = -ENOENT;
3417 5 : inode = child->d_inode;
3418 5 : if (unlikely(!inode))
3419 0 : goto out_err;
3420 5 : if (!(open_flag & O_EXCL)) {
3421 5 : spin_lock(&inode->i_lock);
3422 5 : inode->i_state |= I_LINKABLE;
3423 5 : spin_unlock(&inode->i_lock);
3424 : }
3425 5 : ima_post_create_tmpfile(mnt_userns, inode);
3426 : return child;
3427 :
3428 0 : out_err:
3429 0 : dput(child);
3430 0 : return ERR_PTR(error);
3431 : }
3432 : EXPORT_SYMBOL(vfs_tmpfile);
3433 :
3434 0 : static int do_tmpfile(struct nameidata *nd, unsigned flags,
3435 : const struct open_flags *op,
3436 : struct file *file)
3437 : {
3438 0 : struct user_namespace *mnt_userns;
3439 0 : struct dentry *child;
3440 0 : struct path path;
3441 0 : int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3442 0 : if (unlikely(error))
3443 : return error;
3444 0 : error = mnt_want_write(path.mnt);
3445 0 : if (unlikely(error))
3446 0 : goto out;
3447 0 : mnt_userns = mnt_user_ns(path.mnt);
3448 0 : child = vfs_tmpfile(mnt_userns, path.dentry, op->mode, op->open_flag);
3449 0 : error = PTR_ERR(child);
3450 0 : if (IS_ERR(child))
3451 0 : goto out2;
3452 0 : dput(path.dentry);
3453 0 : path.dentry = child;
3454 0 : audit_inode(nd->name, child, 0);
3455 : /* Don't check for other permissions, the inode was just created */
3456 0 : error = may_open(mnt_userns, &path, 0, op->open_flag);
3457 0 : if (!error)
3458 0 : error = vfs_open(&path, file);
3459 0 : out2:
3460 0 : mnt_drop_write(path.mnt);
3461 0 : out:
3462 0 : path_put(&path);
3463 0 : return error;
3464 : }
3465 :
3466 36333 : static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3467 : {
3468 36333 : struct path path;
3469 36333 : int error = path_lookupat(nd, flags, &path);
3470 36324 : if (!error) {
3471 30248 : audit_inode(nd->name, path.dentry, 0);
3472 30248 : error = vfs_open(&path, file);
3473 30265 : path_put(&path);
3474 : }
3475 36338 : return error;
3476 : }
3477 :
3478 60269 : static struct file *path_openat(struct nameidata *nd,
3479 : const struct open_flags *op, unsigned flags)
3480 : {
3481 60269 : struct file *file;
3482 60269 : int error;
3483 :
3484 60269 : file = alloc_empty_file(op->open_flag, current_cred());
3485 60308 : if (IS_ERR(file))
3486 : return file;
3487 :
3488 60308 : if (unlikely(file->f_flags & __O_TMPFILE)) {
3489 0 : error = do_tmpfile(nd, flags, op, file);
3490 60308 : } else if (unlikely(file->f_flags & O_PATH)) {
3491 36333 : error = do_o_path(nd, flags, file);
3492 : } else {
3493 23975 : const char *s = path_init(nd, flags);
3494 57536 : while (!(error = link_path_walk(s, nd)) &&
3495 25191 : (s = open_last_lookups(nd, file, op)) != NULL)
3496 33562 : ;
3497 23973 : if (!error)
3498 15603 : error = do_open(nd, file, op);
3499 23973 : terminate_walk(nd);
3500 : }
3501 60297 : if (likely(!error)) {
3502 45619 : if (likely(file->f_mode & FMODE_OPENED))
3503 : return file;
3504 0 : WARN_ON(1);
3505 0 : error = -EINVAL;
3506 : }
3507 14678 : fput(file);
3508 14679 : if (error == -EOPENSTALE) {
3509 0 : if (flags & LOOKUP_RCU)
3510 : error = -ECHILD;
3511 : else
3512 0 : error = -ESTALE;
3513 : }
3514 14679 : return ERR_PTR(error);
3515 : }
3516 :
3517 60211 : struct file *do_filp_open(int dfd, struct filename *pathname,
3518 : const struct open_flags *op)
3519 : {
3520 60211 : struct nameidata nd;
3521 60211 : int flags = op->lookup_flags;
3522 60211 : struct file *filp;
3523 :
3524 60211 : set_nameidata(&nd, dfd, pathname);
3525 60211 : filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3526 60219 : if (unlikely(filp == ERR_PTR(-ECHILD)))
3527 76 : filp = path_openat(&nd, op, flags);
3528 60219 : if (unlikely(filp == ERR_PTR(-ESTALE)))
3529 0 : filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3530 60219 : restore_nameidata();
3531 60221 : return filp;
3532 : }
3533 :
3534 0 : struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3535 : const char *name, const struct open_flags *op)
3536 : {
3537 0 : struct nameidata nd;
3538 0 : struct file *file;
3539 0 : struct filename *filename;
3540 0 : int flags = op->lookup_flags | LOOKUP_ROOT;
3541 :
3542 0 : nd.root.mnt = mnt;
3543 0 : nd.root.dentry = dentry;
3544 :
3545 0 : if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3546 0 : return ERR_PTR(-ELOOP);
3547 :
3548 0 : filename = getname_kernel(name);
3549 0 : if (IS_ERR(filename))
3550 0 : return ERR_CAST(filename);
3551 :
3552 0 : set_nameidata(&nd, -1, filename);
3553 0 : file = path_openat(&nd, op, flags | LOOKUP_RCU);
3554 0 : if (unlikely(file == ERR_PTR(-ECHILD)))
3555 0 : file = path_openat(&nd, op, flags);
3556 0 : if (unlikely(file == ERR_PTR(-ESTALE)))
3557 0 : file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3558 0 : restore_nameidata();
3559 0 : putname(filename);
3560 0 : return file;
3561 : }
3562 :
3563 4979 : static struct dentry *filename_create(int dfd, struct filename *name,
3564 : struct path *path, unsigned int lookup_flags)
3565 : {
3566 4979 : struct dentry *dentry = ERR_PTR(-EEXIST);
3567 4979 : struct qstr last;
3568 4979 : int type;
3569 4979 : int err2;
3570 4979 : int error;
3571 4979 : bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3572 :
3573 : /*
3574 : * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3575 : * other flags passed in are ignored!
3576 : */
3577 4979 : lookup_flags &= LOOKUP_REVAL;
3578 :
3579 4979 : name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3580 4979 : if (IS_ERR(name))
3581 4979 : return ERR_CAST(name);
3582 :
3583 : /*
3584 : * Yucky last component or no last component at all?
3585 : * (foo/., foo/.., /////)
3586 : */
3587 4325 : if (unlikely(type != LAST_NORM))
3588 1 : goto out;
3589 :
3590 : /* don't fail immediately if it's r/o, at least try to report other errors */
3591 4324 : err2 = mnt_want_write(path->mnt);
3592 : /*
3593 : * Do the final lookup.
3594 : */
3595 4324 : lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3596 4324 : inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3597 4324 : dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3598 4324 : if (IS_ERR(dentry))
3599 0 : goto unlock;
3600 :
3601 4324 : error = -EEXIST;
3602 4324 : if (d_is_positive(dentry))
3603 2657 : goto fail;
3604 :
3605 : /*
3606 : * Special case - lookup gave negative, but... we had foo/bar/
3607 : * From the vfs_mknod() POV we just have a negative dentry -
3608 : * all is fine. Let's be bastards - you had / on the end, you've
3609 : * been asking for (non-existent) directory. -ENOENT for you.
3610 : */
3611 1667 : if (unlikely(!is_dir && last.name[last.len])) {
3612 0 : error = -ENOENT;
3613 0 : goto fail;
3614 : }
3615 1667 : if (unlikely(err2)) {
3616 0 : error = err2;
3617 0 : goto fail;
3618 : }
3619 1667 : putname(name);
3620 1667 : return dentry;
3621 2657 : fail:
3622 2657 : dput(dentry);
3623 2657 : dentry = ERR_PTR(error);
3624 2657 : unlock:
3625 2657 : inode_unlock(path->dentry->d_inode);
3626 2657 : if (!err2)
3627 2644 : mnt_drop_write(path->mnt);
3628 13 : out:
3629 2658 : path_put(path);
3630 2658 : putname(name);
3631 2658 : return dentry;
3632 : }
3633 :
3634 136 : struct dentry *kern_path_create(int dfd, const char *pathname,
3635 : struct path *path, unsigned int lookup_flags)
3636 : {
3637 136 : return filename_create(dfd, getname_kernel(pathname),
3638 : path, lookup_flags);
3639 : }
3640 : EXPORT_SYMBOL(kern_path_create);
3641 :
3642 1667 : void done_path_create(struct path *path, struct dentry *dentry)
3643 : {
3644 1667 : dput(dentry);
3645 1667 : inode_unlock(path->dentry->d_inode);
3646 1667 : mnt_drop_write(path->mnt);
3647 1667 : path_put(path);
3648 1667 : }
3649 : EXPORT_SYMBOL(done_path_create);
3650 :
3651 4843 : inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3652 : struct path *path, unsigned int lookup_flags)
3653 : {
3654 4843 : return filename_create(dfd, getname(pathname), path, lookup_flags);
3655 : }
3656 : EXPORT_SYMBOL(user_path_create);
3657 :
3658 : /**
3659 : * vfs_mknod - create device node or file
3660 : * @mnt_userns: user namespace of the mount the inode was found from
3661 : * @dir: inode of @dentry
3662 : * @dentry: pointer to dentry of the base directory
3663 : * @mode: mode of the new device node or file
3664 : * @dev: device number of device to create
3665 : *
3666 : * Create a device node or file.
3667 : *
3668 : * If the inode has been found through an idmapped mount the user namespace of
3669 : * the vfsmount must be passed through @mnt_userns. This function will then take
3670 : * care to map the inode according to @mnt_userns before checking permissions.
3671 : * On non-idmapped mounts or if permission checking is to be performed on the
3672 : * raw inode simply passs init_user_ns.
3673 : */
3674 167 : int vfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
3675 : struct dentry *dentry, umode_t mode, dev_t dev)
3676 : {
3677 167 : bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV;
3678 167 : int error = may_create(mnt_userns, dir, dentry);
3679 :
3680 167 : if (error)
3681 : return error;
3682 :
3683 302 : if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout &&
3684 135 : !capable(CAP_MKNOD))
3685 : return -EPERM;
3686 :
3687 167 : if (!dir->i_op->mknod)
3688 : return -EPERM;
3689 :
3690 167 : error = devcgroup_inode_mknod(mode, dev);
3691 167 : if (error)
3692 : return error;
3693 :
3694 167 : error = security_inode_mknod(dir, dentry, mode, dev);
3695 167 : if (error)
3696 : return error;
3697 :
3698 167 : error = dir->i_op->mknod(mnt_userns, dir, dentry, mode, dev);
3699 167 : if (!error)
3700 167 : fsnotify_create(dir, dentry);
3701 : return error;
3702 : }
3703 : EXPORT_SYMBOL(vfs_mknod);
3704 :
3705 512 : static int may_mknod(umode_t mode)
3706 : {
3707 512 : switch (mode & S_IFMT) {
3708 : case S_IFREG:
3709 : case S_IFCHR:
3710 : case S_IFBLK:
3711 : case S_IFIFO:
3712 : case S_IFSOCK:
3713 : case 0: /* zero mode translates to S_IFREG */
3714 : return 0;
3715 0 : case S_IFDIR:
3716 0 : return -EPERM;
3717 0 : default:
3718 0 : return -EINVAL;
3719 : }
3720 : }
3721 :
3722 512 : static long do_mknodat(int dfd, const char __user *filename, umode_t mode,
3723 : unsigned int dev)
3724 : {
3725 512 : struct user_namespace *mnt_userns;
3726 512 : struct dentry *dentry;
3727 512 : struct path path;
3728 512 : int error;
3729 512 : unsigned int lookup_flags = 0;
3730 :
3731 512 : error = may_mknod(mode);
3732 512 : if (error)
3733 0 : return error;
3734 512 : retry:
3735 512 : dentry = user_path_create(dfd, filename, &path, lookup_flags);
3736 512 : if (IS_ERR(dentry))
3737 0 : return PTR_ERR(dentry);
3738 :
3739 512 : if (!IS_POSIXACL(path.dentry->d_inode))
3740 512 : mode &= ~current_umask();
3741 512 : error = security_path_mknod(&path, dentry, mode, dev);
3742 512 : if (error)
3743 8 : goto out;
3744 :
3745 504 : mnt_userns = mnt_user_ns(path.mnt);
3746 504 : switch (mode & S_IFMT) {
3747 469 : case 0: case S_IFREG:
3748 469 : error = vfs_create(mnt_userns, path.dentry->d_inode,
3749 : dentry, mode, true);
3750 469 : if (!error)
3751 512 : ima_post_path_mknod(mnt_userns, dentry);
3752 : break;
3753 : case S_IFCHR: case S_IFBLK:
3754 22 : error = vfs_mknod(mnt_userns, path.dentry->d_inode,
3755 : dentry, mode, new_decode_dev(dev));
3756 22 : break;
3757 13 : case S_IFIFO: case S_IFSOCK:
3758 13 : error = vfs_mknod(mnt_userns, path.dentry->d_inode,
3759 : dentry, mode, 0);
3760 13 : break;
3761 : }
3762 512 : out:
3763 512 : done_path_create(&path, dentry);
3764 1024 : if (retry_estale(error, lookup_flags)) {
3765 0 : lookup_flags |= LOOKUP_REVAL;
3766 0 : goto retry;
3767 : }
3768 : return error;
3769 : }
3770 :
3771 0 : SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3772 : unsigned int, dev)
3773 : {
3774 0 : return do_mknodat(dfd, filename, mode, dev);
3775 : }
3776 :
3777 1024 : SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3778 : {
3779 512 : return do_mknodat(AT_FDCWD, filename, mode, dev);
3780 : }
3781 :
3782 : /**
3783 : * vfs_mkdir - create directory
3784 : * @mnt_userns: user namespace of the mount the inode was found from
3785 : * @dir: inode of @dentry
3786 : * @dentry: pointer to dentry of the base directory
3787 : * @mode: mode of the new directory
3788 : *
3789 : * Create a directory.
3790 : *
3791 : * If the inode has been found through an idmapped mount the user namespace of
3792 : * the vfsmount must be passed through @mnt_userns. This function will then take
3793 : * care to map the inode according to @mnt_userns before checking permissions.
3794 : * On non-idmapped mounts or if permission checking is to be performed on the
3795 : * raw inode simply passs init_user_ns.
3796 : */
3797 789 : int vfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
3798 : struct dentry *dentry, umode_t mode)
3799 : {
3800 789 : int error = may_create(mnt_userns, dir, dentry);
3801 789 : unsigned max_links = dir->i_sb->s_max_links;
3802 :
3803 789 : if (error)
3804 : return error;
3805 :
3806 789 : if (!dir->i_op->mkdir)
3807 : return -EPERM;
3808 :
3809 789 : mode &= (S_IRWXUGO|S_ISVTX);
3810 789 : error = security_inode_mkdir(dir, dentry, mode);
3811 789 : if (error)
3812 : return error;
3813 :
3814 789 : if (max_links && dir->i_nlink >= max_links)
3815 : return -EMLINK;
3816 :
3817 789 : error = dir->i_op->mkdir(mnt_userns, dir, dentry, mode);
3818 789 : if (!error)
3819 785 : fsnotify_mkdir(dir, dentry);
3820 : return error;
3821 : }
3822 : EXPORT_SYMBOL(vfs_mkdir);
3823 :
3824 4085 : static long do_mkdirat(int dfd, const char __user *pathname, umode_t mode)
3825 : {
3826 4085 : struct dentry *dentry;
3827 4085 : struct path path;
3828 4085 : int error;
3829 4085 : unsigned int lookup_flags = LOOKUP_DIRECTORY;
3830 :
3831 4085 : retry:
3832 4085 : dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3833 4085 : if (IS_ERR(dentry))
3834 3301 : return PTR_ERR(dentry);
3835 :
3836 784 : if (!IS_POSIXACL(path.dentry->d_inode))
3837 784 : mode &= ~current_umask();
3838 784 : error = security_path_mkdir(&path, dentry, mode);
3839 784 : if (!error) {
3840 783 : struct user_namespace *mnt_userns;
3841 783 : mnt_userns = mnt_user_ns(path.mnt);
3842 783 : error = vfs_mkdir(mnt_userns, path.dentry->d_inode, dentry,
3843 : mode);
3844 : }
3845 784 : done_path_create(&path, dentry);
3846 1568 : if (retry_estale(error, lookup_flags)) {
3847 0 : lookup_flags |= LOOKUP_REVAL;
3848 0 : goto retry;
3849 : }
3850 : return error;
3851 : }
3852 :
3853 78 : SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3854 : {
3855 39 : return do_mkdirat(dfd, pathname, mode);
3856 : }
3857 :
3858 8092 : SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3859 : {
3860 4046 : return do_mkdirat(AT_FDCWD, pathname, mode);
3861 : }
3862 :
3863 : /**
3864 : * vfs_rmdir - remove directory
3865 : * @mnt_userns: user namespace of the mount the inode was found from
3866 : * @dir: inode of @dentry
3867 : * @dentry: pointer to dentry of the base directory
3868 : *
3869 : * Remove a directory.
3870 : *
3871 : * If the inode has been found through an idmapped mount the user namespace of
3872 : * the vfsmount must be passed through @mnt_userns. This function will then take
3873 : * care to map the inode according to @mnt_userns before checking permissions.
3874 : * On non-idmapped mounts or if permission checking is to be performed on the
3875 : * raw inode simply passs init_user_ns.
3876 : */
3877 1066 : int vfs_rmdir(struct user_namespace *mnt_userns, struct inode *dir,
3878 : struct dentry *dentry)
3879 : {
3880 1066 : int error = may_delete(mnt_userns, dir, dentry, 1);
3881 :
3882 1066 : if (error)
3883 : return error;
3884 :
3885 1066 : if (!dir->i_op->rmdir)
3886 : return -EPERM;
3887 :
3888 1066 : dget(dentry);
3889 1066 : inode_lock(dentry->d_inode);
3890 :
3891 1066 : error = -EBUSY;
3892 1066 : if (is_local_mountpoint(dentry))
3893 185 : goto out;
3894 :
3895 881 : error = security_inode_rmdir(dir, dentry);
3896 881 : if (error)
3897 0 : goto out;
3898 :
3899 881 : error = dir->i_op->rmdir(dir, dentry);
3900 881 : if (error)
3901 333 : goto out;
3902 :
3903 548 : shrink_dcache_parent(dentry);
3904 548 : dentry->d_inode->i_flags |= S_DEAD;
3905 548 : dont_mount(dentry);
3906 548 : detach_mounts(dentry);
3907 548 : fsnotify_rmdir(dir, dentry);
3908 :
3909 1066 : out:
3910 1066 : inode_unlock(dentry->d_inode);
3911 1066 : dput(dentry);
3912 1066 : if (!error)
3913 548 : d_delete(dentry);
3914 : return error;
3915 : }
3916 : EXPORT_SYMBOL(vfs_rmdir);
3917 :
3918 1113 : long do_rmdir(int dfd, struct filename *name)
3919 : {
3920 1113 : struct user_namespace *mnt_userns;
3921 1113 : int error = 0;
3922 1113 : struct dentry *dentry;
3923 1113 : struct path path;
3924 1113 : struct qstr last;
3925 1113 : int type;
3926 1113 : unsigned int lookup_flags = 0;
3927 1113 : retry:
3928 1113 : name = filename_parentat(dfd, name, lookup_flags,
3929 : &path, &last, &type);
3930 1113 : if (IS_ERR(name))
3931 38 : return PTR_ERR(name);
3932 :
3933 1075 : switch (type) {
3934 : case LAST_DOTDOT:
3935 : error = -ENOTEMPTY;
3936 : goto exit1;
3937 : case LAST_DOT:
3938 : error = -EINVAL;
3939 : goto exit1;
3940 : case LAST_ROOT:
3941 : error = -EBUSY;
3942 : goto exit1;
3943 : }
3944 :
3945 1075 : error = mnt_want_write(path.mnt);
3946 1075 : if (error)
3947 0 : goto exit1;
3948 :
3949 1075 : inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3950 1075 : dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3951 1075 : error = PTR_ERR(dentry);
3952 1075 : if (IS_ERR(dentry))
3953 0 : goto exit2;
3954 1075 : if (!dentry->d_inode) {
3955 5 : error = -ENOENT;
3956 5 : goto exit3;
3957 : }
3958 1070 : error = security_path_rmdir(&path, dentry);
3959 1070 : if (error)
3960 4 : goto exit3;
3961 1066 : mnt_userns = mnt_user_ns(path.mnt);
3962 1066 : error = vfs_rmdir(mnt_userns, path.dentry->d_inode, dentry);
3963 1075 : exit3:
3964 1075 : dput(dentry);
3965 1075 : exit2:
3966 1075 : inode_unlock(path.dentry->d_inode);
3967 1075 : mnt_drop_write(path.mnt);
3968 1075 : exit1:
3969 1075 : path_put(&path);
3970 2150 : if (retry_estale(error, lookup_flags)) {
3971 0 : lookup_flags |= LOOKUP_REVAL;
3972 0 : goto retry;
3973 : }
3974 1075 : putname(name);
3975 1075 : return error;
3976 : }
3977 :
3978 2182 : SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3979 : {
3980 1091 : return do_rmdir(AT_FDCWD, getname(pathname));
3981 : }
3982 :
3983 : /**
3984 : * vfs_unlink - unlink a filesystem object
3985 : * @mnt_userns: user namespace of the mount the inode was found from
3986 : * @dir: parent directory
3987 : * @dentry: victim
3988 : * @delegated_inode: returns victim inode, if the inode is delegated.
3989 : *
3990 : * The caller must hold dir->i_mutex.
3991 : *
3992 : * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3993 : * return a reference to the inode in delegated_inode. The caller
3994 : * should then break the delegation on that inode and retry. Because
3995 : * breaking a delegation may take a long time, the caller should drop
3996 : * dir->i_mutex before doing so.
3997 : *
3998 : * Alternatively, a caller may pass NULL for delegated_inode. This may
3999 : * be appropriate for callers that expect the underlying filesystem not
4000 : * to be NFS exported.
4001 : *
4002 : * If the inode has been found through an idmapped mount the user namespace of
4003 : * the vfsmount must be passed through @mnt_userns. This function will then take
4004 : * care to map the inode according to @mnt_userns before checking permissions.
4005 : * On non-idmapped mounts or if permission checking is to be performed on the
4006 : * raw inode simply passs init_user_ns.
4007 : */
4008 796 : int vfs_unlink(struct user_namespace *mnt_userns, struct inode *dir,
4009 : struct dentry *dentry, struct inode **delegated_inode)
4010 : {
4011 796 : struct inode *target = dentry->d_inode;
4012 796 : int error = may_delete(mnt_userns, dir, dentry, 0);
4013 :
4014 796 : if (error)
4015 : return error;
4016 :
4017 654 : if (!dir->i_op->unlink)
4018 : return -EPERM;
4019 :
4020 654 : inode_lock(target);
4021 654 : if (is_local_mountpoint(dentry))
4022 : error = -EBUSY;
4023 : else {
4024 654 : error = security_inode_unlink(dir, dentry);
4025 654 : if (!error) {
4026 654 : error = try_break_deleg(target, delegated_inode);
4027 654 : if (error)
4028 0 : goto out;
4029 654 : error = dir->i_op->unlink(dir, dentry);
4030 654 : if (!error) {
4031 654 : dont_mount(dentry);
4032 654 : detach_mounts(dentry);
4033 654 : fsnotify_unlink(dir, dentry);
4034 : }
4035 : }
4036 : }
4037 0 : out:
4038 654 : inode_unlock(target);
4039 :
4040 : /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4041 654 : if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
4042 654 : fsnotify_link_count(target);
4043 654 : d_delete(dentry);
4044 : }
4045 :
4046 : return error;
4047 : }
4048 : EXPORT_SYMBOL(vfs_unlink);
4049 :
4050 : /*
4051 : * Make sure that the actual truncation of the file will occur outside its
4052 : * directory's i_mutex. Truncate can take a long time if there is a lot of
4053 : * writeout happening, and we don't want to prevent access to the directory
4054 : * while waiting on the I/O.
4055 : */
4056 918 : long do_unlinkat(int dfd, struct filename *name)
4057 : {
4058 918 : int error;
4059 918 : struct dentry *dentry;
4060 918 : struct path path;
4061 918 : struct qstr last;
4062 918 : int type;
4063 918 : struct inode *inode = NULL;
4064 918 : struct inode *delegated_inode = NULL;
4065 918 : unsigned int lookup_flags = 0;
4066 918 : retry:
4067 918 : name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4068 918 : if (IS_ERR(name))
4069 39 : return PTR_ERR(name);
4070 :
4071 879 : error = -EISDIR;
4072 879 : if (type != LAST_NORM)
4073 0 : goto exit1;
4074 :
4075 879 : error = mnt_want_write(path.mnt);
4076 879 : if (error)
4077 1 : goto exit1;
4078 878 : retry_deleg:
4079 878 : inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4080 878 : dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4081 878 : error = PTR_ERR(dentry);
4082 878 : if (!IS_ERR(dentry)) {
4083 878 : struct user_namespace *mnt_userns;
4084 :
4085 : /* Why not before? Because we want correct error value */
4086 878 : if (last.name[last.len])
4087 0 : goto slashes;
4088 878 : inode = dentry->d_inode;
4089 878 : if (d_is_negative(dentry))
4090 83 : goto slashes;
4091 795 : ihold(inode);
4092 795 : error = security_path_unlink(&path, dentry);
4093 795 : if (error)
4094 3 : goto exit2;
4095 792 : mnt_userns = mnt_user_ns(path.mnt);
4096 792 : error = vfs_unlink(mnt_userns, path.dentry->d_inode, dentry,
4097 : &delegated_inode);
4098 878 : exit2:
4099 878 : dput(dentry);
4100 : }
4101 878 : inode_unlock(path.dentry->d_inode);
4102 878 : if (inode)
4103 795 : iput(inode); /* truncate the inode here */
4104 878 : inode = NULL;
4105 878 : if (delegated_inode) {
4106 0 : error = break_deleg_wait(&delegated_inode);
4107 0 : if (!error)
4108 0 : goto retry_deleg;
4109 : }
4110 878 : mnt_drop_write(path.mnt);
4111 879 : exit1:
4112 879 : path_put(&path);
4113 1758 : if (retry_estale(error, lookup_flags)) {
4114 0 : lookup_flags |= LOOKUP_REVAL;
4115 0 : inode = NULL;
4116 0 : goto retry;
4117 : }
4118 879 : putname(name);
4119 879 : return error;
4120 :
4121 83 : slashes:
4122 83 : if (d_is_negative(dentry))
4123 : error = -ENOENT;
4124 0 : else if (d_is_dir(dentry))
4125 : error = -EISDIR;
4126 : else
4127 0 : error = -ENOTDIR;
4128 83 : goto exit2;
4129 : }
4130 :
4131 114 : SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4132 : {
4133 57 : if ((flag & ~AT_REMOVEDIR) != 0)
4134 : return -EINVAL;
4135 :
4136 57 : if (flag & AT_REMOVEDIR)
4137 22 : return do_rmdir(dfd, getname(pathname));
4138 35 : return do_unlinkat(dfd, getname(pathname));
4139 : }
4140 :
4141 1760 : SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4142 : {
4143 880 : return do_unlinkat(AT_FDCWD, getname(pathname));
4144 : }
4145 :
4146 : /**
4147 : * vfs_symlink - create symlink
4148 : * @mnt_userns: user namespace of the mount the inode was found from
4149 : * @dir: inode of @dentry
4150 : * @dentry: pointer to dentry of the base directory
4151 : * @oldname: name of the file to link to
4152 : *
4153 : * Create a symlink.
4154 : *
4155 : * If the inode has been found through an idmapped mount the user namespace of
4156 : * the vfsmount must be passed through @mnt_userns. This function will then take
4157 : * care to map the inode according to @mnt_userns before checking permissions.
4158 : * On non-idmapped mounts or if permission checking is to be performed on the
4159 : * raw inode simply passs init_user_ns.
4160 : */
4161 209 : int vfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
4162 : struct dentry *dentry, const char *oldname)
4163 : {
4164 209 : int error = may_create(mnt_userns, dir, dentry);
4165 :
4166 209 : if (error)
4167 : return error;
4168 :
4169 209 : if (!dir->i_op->symlink)
4170 : return -EPERM;
4171 :
4172 209 : error = security_inode_symlink(dir, dentry, oldname);
4173 209 : if (error)
4174 : return error;
4175 :
4176 209 : error = dir->i_op->symlink(mnt_userns, dir, dentry, oldname);
4177 209 : if (!error)
4178 209 : fsnotify_create(dir, dentry);
4179 : return error;
4180 : }
4181 : EXPORT_SYMBOL(vfs_symlink);
4182 :
4183 220 : static long do_symlinkat(const char __user *oldname, int newdfd,
4184 : const char __user *newname)
4185 : {
4186 220 : int error;
4187 220 : struct filename *from;
4188 220 : struct dentry *dentry;
4189 220 : struct path path;
4190 220 : unsigned int lookup_flags = 0;
4191 :
4192 220 : from = getname(oldname);
4193 220 : if (IS_ERR(from))
4194 0 : return PTR_ERR(from);
4195 220 : retry:
4196 220 : dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4197 220 : error = PTR_ERR(dentry);
4198 220 : if (IS_ERR(dentry))
4199 10 : goto out_putname;
4200 :
4201 210 : error = security_path_symlink(&path, dentry, from->name);
4202 210 : if (!error) {
4203 209 : struct user_namespace *mnt_userns;
4204 :
4205 209 : mnt_userns = mnt_user_ns(path.mnt);
4206 209 : error = vfs_symlink(mnt_userns, path.dentry->d_inode, dentry,
4207 : from->name);
4208 : }
4209 210 : done_path_create(&path, dentry);
4210 420 : if (retry_estale(error, lookup_flags)) {
4211 0 : lookup_flags |= LOOKUP_REVAL;
4212 0 : goto retry;
4213 : }
4214 210 : out_putname:
4215 220 : putname(from);
4216 220 : return error;
4217 : }
4218 :
4219 0 : SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4220 : int, newdfd, const char __user *, newname)
4221 : {
4222 0 : return do_symlinkat(oldname, newdfd, newname);
4223 : }
4224 :
4225 440 : SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4226 : {
4227 220 : return do_symlinkat(oldname, AT_FDCWD, newname);
4228 : }
4229 :
4230 : /**
4231 : * vfs_link - create a new link
4232 : * @old_dentry: object to be linked
4233 : * @mnt_userns: the user namespace of the mount
4234 : * @dir: new parent
4235 : * @new_dentry: where to create the new link
4236 : * @delegated_inode: returns inode needing a delegation break
4237 : *
4238 : * The caller must hold dir->i_mutex
4239 : *
4240 : * If vfs_link discovers a delegation on the to-be-linked file in need
4241 : * of breaking, it will return -EWOULDBLOCK and return a reference to the
4242 : * inode in delegated_inode. The caller should then break the delegation
4243 : * and retry. Because breaking a delegation may take a long time, the
4244 : * caller should drop the i_mutex before doing so.
4245 : *
4246 : * Alternatively, a caller may pass NULL for delegated_inode. This may
4247 : * be appropriate for callers that expect the underlying filesystem not
4248 : * to be NFS exported.
4249 : *
4250 : * If the inode has been found through an idmapped mount the user namespace of
4251 : * the vfsmount must be passed through @mnt_userns. This function will then take
4252 : * care to map the inode according to @mnt_userns before checking permissions.
4253 : * On non-idmapped mounts or if permission checking is to be performed on the
4254 : * raw inode simply passs init_user_ns.
4255 : */
4256 19 : int vfs_link(struct dentry *old_dentry, struct user_namespace *mnt_userns,
4257 : struct inode *dir, struct dentry *new_dentry,
4258 : struct inode **delegated_inode)
4259 : {
4260 19 : struct inode *inode = old_dentry->d_inode;
4261 19 : unsigned max_links = dir->i_sb->s_max_links;
4262 19 : int error;
4263 :
4264 19 : if (!inode)
4265 : return -ENOENT;
4266 :
4267 19 : error = may_create(mnt_userns, dir, new_dentry);
4268 19 : if (error)
4269 : return error;
4270 :
4271 19 : if (dir->i_sb != inode->i_sb)
4272 : return -EXDEV;
4273 :
4274 : /*
4275 : * A link to an append-only or immutable file cannot be created.
4276 : */
4277 19 : if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4278 : return -EPERM;
4279 : /*
4280 : * Updating the link count will likely cause i_uid and i_gid to
4281 : * be writen back improperly if their true value is unknown to
4282 : * the vfs.
4283 : */
4284 38 : if (HAS_UNMAPPED_ID(mnt_userns, inode))
4285 : return -EPERM;
4286 19 : if (!dir->i_op->link)
4287 : return -EPERM;
4288 19 : if (S_ISDIR(inode->i_mode))
4289 : return -EPERM;
4290 :
4291 19 : error = security_inode_link(old_dentry, dir, new_dentry);
4292 19 : if (error)
4293 : return error;
4294 :
4295 19 : inode_lock(inode);
4296 : /* Make sure we don't allow creating hardlink to an unlinked file */
4297 19 : if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4298 : error = -ENOENT;
4299 19 : else if (max_links && inode->i_nlink >= max_links)
4300 : error = -EMLINK;
4301 : else {
4302 19 : error = try_break_deleg(inode, delegated_inode);
4303 19 : if (!error)
4304 19 : error = dir->i_op->link(old_dentry, dir, new_dentry);
4305 : }
4306 :
4307 19 : if (!error && (inode->i_state & I_LINKABLE)) {
4308 3 : spin_lock(&inode->i_lock);
4309 3 : inode->i_state &= ~I_LINKABLE;
4310 3 : spin_unlock(&inode->i_lock);
4311 : }
4312 19 : inode_unlock(inode);
4313 19 : if (!error)
4314 19 : fsnotify_link(dir, inode, new_dentry);
4315 : return error;
4316 : }
4317 : EXPORT_SYMBOL(vfs_link);
4318 :
4319 : /*
4320 : * Hardlinks are often used in delicate situations. We avoid
4321 : * security-related surprises by not following symlinks on the
4322 : * newname. --KAB
4323 : *
4324 : * We don't follow them on the oldname either to be compatible
4325 : * with linux 2.0, and to avoid hard-linking to directories
4326 : * and other special files. --ADM
4327 : */
4328 26 : static int do_linkat(int olddfd, const char __user *oldname, int newdfd,
4329 : const char __user *newname, int flags)
4330 : {
4331 26 : struct user_namespace *mnt_userns;
4332 26 : struct dentry *new_dentry;
4333 26 : struct path old_path, new_path;
4334 26 : struct inode *delegated_inode = NULL;
4335 26 : int how = 0;
4336 26 : int error;
4337 :
4338 26 : if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4339 : return -EINVAL;
4340 : /*
4341 : * To use null names we require CAP_DAC_READ_SEARCH
4342 : * This ensures that not everyone will be able to create
4343 : * handlink using the passed filedescriptor.
4344 : */
4345 26 : if (flags & AT_EMPTY_PATH) {
4346 0 : if (!capable(CAP_DAC_READ_SEARCH))
4347 : return -ENOENT;
4348 : how = LOOKUP_EMPTY;
4349 : }
4350 :
4351 26 : if (flags & AT_SYMLINK_FOLLOW)
4352 0 : how |= LOOKUP_FOLLOW;
4353 26 : retry:
4354 26 : error = user_path_at(olddfd, oldname, how, &old_path);
4355 26 : if (error)
4356 0 : return error;
4357 :
4358 26 : new_dentry = user_path_create(newdfd, newname, &new_path,
4359 : (how & LOOKUP_REVAL));
4360 26 : error = PTR_ERR(new_dentry);
4361 26 : if (IS_ERR(new_dentry))
4362 0 : goto out;
4363 :
4364 26 : error = -EXDEV;
4365 26 : if (old_path.mnt != new_path.mnt)
4366 0 : goto out_dput;
4367 26 : mnt_userns = mnt_user_ns(new_path.mnt);
4368 26 : error = may_linkat(mnt_userns, &old_path);
4369 26 : if (unlikely(error))
4370 0 : goto out_dput;
4371 26 : error = security_path_link(old_path.dentry, &new_path, new_dentry);
4372 26 : if (error)
4373 10 : goto out_dput;
4374 16 : error = vfs_link(old_path.dentry, mnt_userns, new_path.dentry->d_inode,
4375 : new_dentry, &delegated_inode);
4376 26 : out_dput:
4377 26 : done_path_create(&new_path, new_dentry);
4378 26 : if (delegated_inode) {
4379 0 : error = break_deleg_wait(&delegated_inode);
4380 0 : if (!error) {
4381 0 : path_put(&old_path);
4382 0 : goto retry;
4383 : }
4384 : }
4385 52 : if (retry_estale(error, how)) {
4386 0 : path_put(&old_path);
4387 0 : how |= LOOKUP_REVAL;
4388 0 : goto retry;
4389 : }
4390 26 : out:
4391 26 : path_put(&old_path);
4392 :
4393 26 : return error;
4394 : }
4395 :
4396 0 : SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4397 : int, newdfd, const char __user *, newname, int, flags)
4398 : {
4399 0 : return do_linkat(olddfd, oldname, newdfd, newname, flags);
4400 : }
4401 :
4402 52 : SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4403 : {
4404 26 : return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4405 : }
4406 :
4407 : /**
4408 : * vfs_rename - rename a filesystem object
4409 : * @old_mnt_userns: old user namespace of the mount the inode was found from
4410 : * @old_dir: parent of source
4411 : * @old_dentry: source
4412 : * @new_mnt_userns: new user namespace of the mount the inode was found from
4413 : * @new_dir: parent of destination
4414 : * @new_dentry: destination
4415 : * @delegated_inode: returns an inode needing a delegation break
4416 : * @flags: rename flags
4417 : *
4418 : * The caller must hold multiple mutexes--see lock_rename()).
4419 : *
4420 : * If vfs_rename discovers a delegation in need of breaking at either
4421 : * the source or destination, it will return -EWOULDBLOCK and return a
4422 : * reference to the inode in delegated_inode. The caller should then
4423 : * break the delegation and retry. Because breaking a delegation may
4424 : * take a long time, the caller should drop all locks before doing
4425 : * so.
4426 : *
4427 : * Alternatively, a caller may pass NULL for delegated_inode. This may
4428 : * be appropriate for callers that expect the underlying filesystem not
4429 : * to be NFS exported.
4430 : *
4431 : * The worst of all namespace operations - renaming directory. "Perverted"
4432 : * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4433 : * Problems:
4434 : *
4435 : * a) we can get into loop creation.
4436 : * b) race potential - two innocent renames can create a loop together.
4437 : * That's where 4.4 screws up. Current fix: serialization on
4438 : * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4439 : * story.
4440 : * c) we have to lock _four_ objects - parents and victim (if it exists),
4441 : * and source (if it is not a directory).
4442 : * And that - after we got ->i_mutex on parents (until then we don't know
4443 : * whether the target exists). Solution: try to be smart with locking
4444 : * order for inodes. We rely on the fact that tree topology may change
4445 : * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4446 : * move will be locked. Thus we can rank directories by the tree
4447 : * (ancestors first) and rank all non-directories after them.
4448 : * That works since everybody except rename does "lock parent, lookup,
4449 : * lock child" and rename is under ->s_vfs_rename_mutex.
4450 : * HOWEVER, it relies on the assumption that any object with ->lookup()
4451 : * has no more than 1 dentry. If "hybrid" objects will ever appear,
4452 : * we'd better make sure that there's no link(2) for them.
4453 : * d) conversion from fhandle to dentry may come in the wrong moment - when
4454 : * we are removing the target. Solution: we will have to grab ->i_mutex
4455 : * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4456 : * ->i_mutex on parents, which works but leads to some truly excessive
4457 : * locking].
4458 : */
4459 392 : int vfs_rename(struct renamedata *rd)
4460 : {
4461 392 : int error;
4462 392 : struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir;
4463 392 : struct dentry *old_dentry = rd->old_dentry;
4464 392 : struct dentry *new_dentry = rd->new_dentry;
4465 392 : struct inode **delegated_inode = rd->delegated_inode;
4466 392 : unsigned int flags = rd->flags;
4467 392 : bool is_dir = d_is_dir(old_dentry);
4468 392 : struct inode *source = old_dentry->d_inode;
4469 392 : struct inode *target = new_dentry->d_inode;
4470 392 : bool new_is_dir = false;
4471 392 : unsigned max_links = new_dir->i_sb->s_max_links;
4472 392 : struct name_snapshot old_name;
4473 :
4474 392 : if (source == target)
4475 : return 0;
4476 :
4477 392 : error = may_delete(rd->old_mnt_userns, old_dir, old_dentry, is_dir);
4478 392 : if (error)
4479 : return error;
4480 :
4481 392 : if (!target) {
4482 297 : error = may_create(rd->new_mnt_userns, new_dir, new_dentry);
4483 : } else {
4484 95 : new_is_dir = d_is_dir(new_dentry);
4485 :
4486 95 : if (!(flags & RENAME_EXCHANGE))
4487 91 : error = may_delete(rd->new_mnt_userns, new_dir,
4488 : new_dentry, is_dir);
4489 : else
4490 4 : error = may_delete(rd->new_mnt_userns, new_dir,
4491 : new_dentry, new_is_dir);
4492 : }
4493 392 : if (error)
4494 : return error;
4495 :
4496 392 : if (!old_dir->i_op->rename)
4497 : return -EPERM;
4498 :
4499 : /*
4500 : * If we are going to change the parent - check write permissions,
4501 : * we'll need to flip '..'.
4502 : */
4503 392 : if (new_dir != old_dir) {
4504 1 : if (is_dir) {
4505 1 : error = inode_permission(rd->old_mnt_userns, source,
4506 : MAY_WRITE);
4507 1 : if (error)
4508 : return error;
4509 : }
4510 1 : if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4511 0 : error = inode_permission(rd->new_mnt_userns, target,
4512 : MAY_WRITE);
4513 0 : if (error)
4514 : return error;
4515 : }
4516 : }
4517 :
4518 392 : error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4519 : flags);
4520 392 : if (error)
4521 : return error;
4522 :
4523 392 : take_dentry_name_snapshot(&old_name, old_dentry);
4524 392 : dget(new_dentry);
4525 392 : if (!is_dir || (flags & RENAME_EXCHANGE))
4526 390 : lock_two_nondirectories(source, target);
4527 2 : else if (target)
4528 1 : inode_lock(target);
4529 :
4530 392 : error = -EBUSY;
4531 392 : if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4532 0 : goto out;
4533 :
4534 392 : if (max_links && new_dir != old_dir) {
4535 0 : error = -EMLINK;
4536 0 : if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4537 0 : goto out;
4538 0 : if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4539 0 : old_dir->i_nlink >= max_links)
4540 0 : goto out;
4541 : }
4542 392 : if (!is_dir) {
4543 388 : error = try_break_deleg(source, delegated_inode);
4544 388 : if (error)
4545 0 : goto out;
4546 : }
4547 392 : if (target && !new_is_dir) {
4548 93 : error = try_break_deleg(target, delegated_inode);
4549 93 : if (error)
4550 0 : goto out;
4551 : }
4552 392 : error = old_dir->i_op->rename(rd->new_mnt_userns, old_dir, old_dentry,
4553 : new_dir, new_dentry, flags);
4554 392 : if (error)
4555 0 : goto out;
4556 :
4557 392 : if (!(flags & RENAME_EXCHANGE) && target) {
4558 91 : if (is_dir) {
4559 1 : shrink_dcache_parent(new_dentry);
4560 1 : target->i_flags |= S_DEAD;
4561 : }
4562 91 : dont_mount(new_dentry);
4563 91 : detach_mounts(new_dentry);
4564 : }
4565 392 : if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4566 392 : if (!(flags & RENAME_EXCHANGE))
4567 388 : d_move(old_dentry, new_dentry);
4568 : else
4569 4 : d_exchange(old_dentry, new_dentry);
4570 : }
4571 0 : out:
4572 392 : if (!is_dir || (flags & RENAME_EXCHANGE))
4573 390 : unlock_two_nondirectories(source, target);
4574 2 : else if (target)
4575 1 : inode_unlock(target);
4576 392 : dput(new_dentry);
4577 392 : if (!error) {
4578 392 : fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4579 392 : !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4580 392 : if (flags & RENAME_EXCHANGE) {
4581 4 : fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4582 : new_is_dir, NULL, new_dentry);
4583 : }
4584 : }
4585 392 : release_dentry_name_snapshot(&old_name);
4586 :
4587 392 : return error;
4588 : }
4589 : EXPORT_SYMBOL(vfs_rename);
4590 :
4591 425 : int do_renameat2(int olddfd, struct filename *from, int newdfd,
4592 : struct filename *to, unsigned int flags)
4593 : {
4594 425 : struct renamedata rd;
4595 425 : struct dentry *old_dentry, *new_dentry;
4596 425 : struct dentry *trap;
4597 425 : struct path old_path, new_path;
4598 425 : struct qstr old_last, new_last;
4599 425 : int old_type, new_type;
4600 425 : struct inode *delegated_inode = NULL;
4601 425 : unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4602 425 : bool should_retry = false;
4603 425 : int error = -EINVAL;
4604 :
4605 425 : if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4606 0 : goto put_both;
4607 :
4608 425 : if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4609 9 : (flags & RENAME_EXCHANGE))
4610 0 : goto put_both;
4611 :
4612 425 : if (flags & RENAME_EXCHANGE)
4613 14 : target_flags = 0;
4614 :
4615 425 : retry:
4616 425 : from = filename_parentat(olddfd, from, lookup_flags, &old_path,
4617 : &old_last, &old_type);
4618 425 : if (IS_ERR(from)) {
4619 0 : error = PTR_ERR(from);
4620 0 : goto put_new;
4621 : }
4622 :
4623 425 : to = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last,
4624 : &new_type);
4625 425 : if (IS_ERR(to)) {
4626 0 : error = PTR_ERR(to);
4627 0 : goto exit1;
4628 : }
4629 :
4630 425 : error = -EXDEV;
4631 425 : if (old_path.mnt != new_path.mnt)
4632 0 : goto exit2;
4633 :
4634 425 : error = -EBUSY;
4635 425 : if (old_type != LAST_NORM)
4636 0 : goto exit2;
4637 :
4638 425 : if (flags & RENAME_NOREPLACE)
4639 : error = -EEXIST;
4640 425 : if (new_type != LAST_NORM)
4641 0 : goto exit2;
4642 :
4643 425 : error = mnt_want_write(old_path.mnt);
4644 425 : if (error)
4645 0 : goto exit2;
4646 :
4647 425 : retry_deleg:
4648 425 : trap = lock_rename(new_path.dentry, old_path.dentry);
4649 :
4650 425 : old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4651 425 : error = PTR_ERR(old_dentry);
4652 425 : if (IS_ERR(old_dentry))
4653 0 : goto exit3;
4654 : /* source must exist */
4655 425 : error = -ENOENT;
4656 425 : if (d_is_negative(old_dentry))
4657 12 : goto exit4;
4658 413 : new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4659 413 : error = PTR_ERR(new_dentry);
4660 413 : if (IS_ERR(new_dentry))
4661 0 : goto exit4;
4662 413 : error = -EEXIST;
4663 413 : if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4664 2 : goto exit5;
4665 411 : if (flags & RENAME_EXCHANGE) {
4666 14 : error = -ENOENT;
4667 14 : if (d_is_negative(new_dentry))
4668 0 : goto exit5;
4669 :
4670 21 : if (!d_is_dir(new_dentry)) {
4671 7 : error = -ENOTDIR;
4672 7 : if (new_last.name[new_last.len])
4673 0 : goto exit5;
4674 : }
4675 : }
4676 : /* unless the source is a directory trailing slashes give -ENOTDIR */
4677 814 : if (!d_is_dir(old_dentry)) {
4678 403 : error = -ENOTDIR;
4679 403 : if (old_last.name[old_last.len])
4680 0 : goto exit5;
4681 403 : if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4682 0 : goto exit5;
4683 : }
4684 : /* source should not be ancestor of target */
4685 411 : error = -EINVAL;
4686 411 : if (old_dentry == trap)
4687 0 : goto exit5;
4688 : /* target should not be an ancestor of source */
4689 411 : if (!(flags & RENAME_EXCHANGE))
4690 397 : error = -ENOTEMPTY;
4691 411 : if (new_dentry == trap)
4692 0 : goto exit5;
4693 :
4694 411 : error = security_path_rename(&old_path, old_dentry,
4695 : &new_path, new_dentry, flags);
4696 411 : if (error)
4697 22 : goto exit5;
4698 :
4699 389 : rd.old_dir = old_path.dentry->d_inode;
4700 389 : rd.old_dentry = old_dentry;
4701 389 : rd.old_mnt_userns = mnt_user_ns(old_path.mnt);
4702 389 : rd.new_dir = new_path.dentry->d_inode;
4703 389 : rd.new_dentry = new_dentry;
4704 389 : rd.new_mnt_userns = mnt_user_ns(new_path.mnt);
4705 389 : rd.delegated_inode = &delegated_inode;
4706 389 : rd.flags = flags;
4707 389 : error = vfs_rename(&rd);
4708 413 : exit5:
4709 413 : dput(new_dentry);
4710 425 : exit4:
4711 425 : dput(old_dentry);
4712 425 : exit3:
4713 425 : unlock_rename(new_path.dentry, old_path.dentry);
4714 425 : if (delegated_inode) {
4715 0 : error = break_deleg_wait(&delegated_inode);
4716 0 : if (!error)
4717 0 : goto retry_deleg;
4718 : }
4719 425 : mnt_drop_write(old_path.mnt);
4720 425 : exit2:
4721 425 : if (retry_estale(error, lookup_flags))
4722 0 : should_retry = true;
4723 425 : path_put(&new_path);
4724 425 : exit1:
4725 425 : path_put(&old_path);
4726 425 : if (should_retry) {
4727 0 : should_retry = false;
4728 0 : lookup_flags |= LOOKUP_REVAL;
4729 0 : goto retry;
4730 : }
4731 425 : put_both:
4732 425 : if (!IS_ERR(from))
4733 425 : putname(from);
4734 0 : put_new:
4735 425 : if (!IS_ERR(to))
4736 425 : putname(to);
4737 425 : return error;
4738 : }
4739 :
4740 46 : SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4741 : int, newdfd, const char __user *, newname, unsigned int, flags)
4742 : {
4743 23 : return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4744 : flags);
4745 : }
4746 :
4747 0 : SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4748 : int, newdfd, const char __user *, newname)
4749 : {
4750 0 : return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4751 : 0);
4752 : }
4753 :
4754 804 : SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4755 : {
4756 402 : return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD,
4757 : getname(newname), 0);
4758 : }
4759 :
4760 1106 : int readlink_copy(char __user *buffer, int buflen, const char *link)
4761 : {
4762 1106 : int len = PTR_ERR(link);
4763 1106 : if (IS_ERR(link))
4764 0 : goto out;
4765 :
4766 1106 : len = strlen(link);
4767 1106 : if (len > (unsigned) buflen)
4768 0 : len = buflen;
4769 2212 : if (copy_to_user(buffer, link, len))
4770 0 : len = -EFAULT;
4771 1106 : out:
4772 1106 : return len;
4773 : }
4774 :
4775 : /**
4776 : * vfs_readlink - copy symlink body into userspace buffer
4777 : * @dentry: dentry on which to get symbolic link
4778 : * @buffer: user memory pointer
4779 : * @buflen: size of buffer
4780 : *
4781 : * Does not touch atime. That's up to the caller if necessary
4782 : *
4783 : * Does not call security hook.
4784 : */
4785 1156 : int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4786 : {
4787 1156 : struct inode *inode = d_inode(dentry);
4788 1156 : DEFINE_DELAYED_CALL(done);
4789 1156 : const char *link;
4790 1156 : int res;
4791 :
4792 1156 : if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4793 518 : if (unlikely(inode->i_op->readlink))
4794 50 : return inode->i_op->readlink(dentry, buffer, buflen);
4795 :
4796 468 : if (!d_is_symlink(dentry))
4797 : return -EINVAL;
4798 :
4799 468 : spin_lock(&inode->i_lock);
4800 468 : inode->i_opflags |= IOP_DEFAULT_READLINK;
4801 468 : spin_unlock(&inode->i_lock);
4802 : }
4803 :
4804 1106 : link = READ_ONCE(inode->i_link);
4805 1106 : if (!link) {
4806 860 : link = inode->i_op->get_link(dentry, inode, &done);
4807 861 : if (IS_ERR(link))
4808 0 : return PTR_ERR(link);
4809 : }
4810 1107 : res = readlink_copy(buffer, buflen, link);
4811 1106 : do_delayed_call(&done);
4812 : return res;
4813 : }
4814 : EXPORT_SYMBOL(vfs_readlink);
4815 :
4816 : /**
4817 : * vfs_get_link - get symlink body
4818 : * @dentry: dentry on which to get symbolic link
4819 : * @done: caller needs to free returned data with this
4820 : *
4821 : * Calls security hook and i_op->get_link() on the supplied inode.
4822 : *
4823 : * It does not touch atime. That's up to the caller if necessary.
4824 : *
4825 : * Does not work on "special" symlinks like /proc/$$/fd/N
4826 : */
4827 0 : const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4828 : {
4829 0 : const char *res = ERR_PTR(-EINVAL);
4830 0 : struct inode *inode = d_inode(dentry);
4831 :
4832 0 : if (d_is_symlink(dentry)) {
4833 0 : res = ERR_PTR(security_inode_readlink(dentry));
4834 0 : if (!res)
4835 0 : res = inode->i_op->get_link(dentry, inode, done);
4836 : }
4837 0 : return res;
4838 : }
4839 : EXPORT_SYMBOL(vfs_get_link);
4840 :
4841 : /* get the link contents into pagecache */
4842 0 : const char *page_get_link(struct dentry *dentry, struct inode *inode,
4843 : struct delayed_call *callback)
4844 : {
4845 0 : char *kaddr;
4846 0 : struct page *page;
4847 0 : struct address_space *mapping = inode->i_mapping;
4848 :
4849 0 : if (!dentry) {
4850 0 : page = find_get_page(mapping, 0);
4851 0 : if (!page)
4852 0 : return ERR_PTR(-ECHILD);
4853 0 : if (!PageUptodate(page)) {
4854 0 : put_page(page);
4855 0 : return ERR_PTR(-ECHILD);
4856 : }
4857 : } else {
4858 0 : page = read_mapping_page(mapping, 0, NULL);
4859 0 : if (IS_ERR(page))
4860 : return (char*)page;
4861 : }
4862 0 : set_delayed_call(callback, page_put_link, page);
4863 0 : BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4864 0 : kaddr = page_address(page);
4865 0 : nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4866 0 : return kaddr;
4867 : }
4868 :
4869 : EXPORT_SYMBOL(page_get_link);
4870 :
4871 0 : void page_put_link(void *arg)
4872 : {
4873 0 : put_page(arg);
4874 0 : }
4875 : EXPORT_SYMBOL(page_put_link);
4876 :
4877 0 : int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4878 : {
4879 0 : DEFINE_DELAYED_CALL(done);
4880 0 : int res = readlink_copy(buffer, buflen,
4881 : page_get_link(dentry, d_inode(dentry),
4882 : &done));
4883 0 : do_delayed_call(&done);
4884 0 : return res;
4885 : }
4886 : EXPORT_SYMBOL(page_readlink);
4887 :
4888 : /*
4889 : * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4890 : */
4891 0 : int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4892 : {
4893 0 : struct address_space *mapping = inode->i_mapping;
4894 0 : struct page *page;
4895 0 : void *fsdata;
4896 0 : int err;
4897 0 : unsigned int flags = 0;
4898 0 : if (nofs)
4899 0 : flags |= AOP_FLAG_NOFS;
4900 :
4901 0 : retry:
4902 0 : err = pagecache_write_begin(NULL, mapping, 0, len-1,
4903 : flags, &page, &fsdata);
4904 0 : if (err)
4905 0 : goto fail;
4906 :
4907 0 : memcpy(page_address(page), symname, len-1);
4908 :
4909 0 : err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4910 : page, fsdata);
4911 0 : if (err < 0)
4912 0 : goto fail;
4913 0 : if (err < len-1)
4914 0 : goto retry;
4915 :
4916 0 : mark_inode_dirty(inode);
4917 0 : return 0;
4918 : fail:
4919 : return err;
4920 : }
4921 : EXPORT_SYMBOL(__page_symlink);
4922 :
4923 0 : int page_symlink(struct inode *inode, const char *symname, int len)
4924 : {
4925 0 : return __page_symlink(inode, symname, len,
4926 0 : !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4927 : }
4928 : EXPORT_SYMBOL(page_symlink);
4929 :
4930 : const struct inode_operations page_symlink_inode_operations = {
4931 : .get_link = page_get_link,
4932 : };
4933 : EXPORT_SYMBOL(page_symlink_inode_operations);
|
