Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0
2 : /*
3 : * linux/fs/namei.c
4 : *
5 : * Copyright (C) 1991, 1992 Linus Torvalds
6 : */
7 :
8 : /*
9 : * Some corrections by tytso.
10 : */
11 :
12 : /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
13 : * lookup logic.
14 : */
15 : /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
16 : */
17 :
18 : #include <linux/init.h>
19 : #include <linux/export.h>
20 : #include <linux/kernel.h>
21 : #include <linux/slab.h>
22 : #include <linux/fs.h>
23 : #include <linux/filelock.h>
24 : #include <linux/namei.h>
25 : #include <linux/pagemap.h>
26 : #include <linux/sched/mm.h>
27 : #include <linux/fsnotify.h>
28 : #include <linux/personality.h>
29 : #include <linux/security.h>
30 : #include <linux/ima.h>
31 : #include <linux/syscalls.h>
32 : #include <linux/mount.h>
33 : #include <linux/audit.h>
34 : #include <linux/capability.h>
35 : #include <linux/file.h>
36 : #include <linux/fcntl.h>
37 : #include <linux/device_cgroup.h>
38 : #include <linux/fs_struct.h>
39 : #include <linux/posix_acl.h>
40 : #include <linux/hash.h>
41 : #include <linux/bitops.h>
42 : #include <linux/init_task.h>
43 : #include <linux/uaccess.h>
44 :
45 : #include "internal.h"
46 : #include "mount.h"
47 :
48 : /* [Feb-1997 T. Schoebel-Theuer]
49 : * Fundamental changes in the pathname lookup mechanisms (namei)
50 : * were necessary because of omirr. The reason is that omirr needs
51 : * to know the _real_ pathname, not the user-supplied one, in case
52 : * of symlinks (and also when transname replacements occur).
53 : *
54 : * The new code replaces the old recursive symlink resolution with
55 : * an iterative one (in case of non-nested symlink chains). It does
56 : * this with calls to <fs>_follow_link().
57 : * As a side effect, dir_namei(), _namei() and follow_link() are now
58 : * replaced with a single function lookup_dentry() that can handle all
59 : * the special cases of the former code.
60 : *
61 : * With the new dcache, the pathname is stored at each inode, at least as
62 : * long as the refcount of the inode is positive. As a side effect, the
63 : * size of the dcache depends on the inode cache and thus is dynamic.
64 : *
65 : * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
66 : * resolution to correspond with current state of the code.
67 : *
68 : * Note that the symlink resolution is not *completely* iterative.
69 : * There is still a significant amount of tail- and mid- recursion in
70 : * the algorithm. Also, note that <fs>_readlink() is not used in
71 : * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
72 : * may return different results than <fs>_follow_link(). Many virtual
73 : * filesystems (including /proc) exhibit this behavior.
74 : */
75 :
76 : /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
77 : * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
78 : * and the name already exists in form of a symlink, try to create the new
79 : * name indicated by the symlink. The old code always complained that the
80 : * name already exists, due to not following the symlink even if its target
81 : * is nonexistent. The new semantics affects also mknod() and link() when
82 : * the name is a symlink pointing to a non-existent name.
83 : *
84 : * I don't know which semantics is the right one, since I have no access
85 : * to standards. But I found by trial that HP-UX 9.0 has the full "new"
86 : * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
87 : * "old" one. Personally, I think the new semantics is much more logical.
88 : * Note that "ln old new" where "new" is a symlink pointing to a non-existing
89 : * file does succeed in both HP-UX and SunOs, but not in Solaris
90 : * and in the old Linux semantics.
91 : */
92 :
93 : /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
94 : * semantics. See the comments in "open_namei" and "do_link" below.
95 : *
96 : * [10-Sep-98 Alan Modra] Another symlink change.
97 : */
98 :
99 : /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
100 : * inside the path - always follow.
101 : * in the last component in creation/removal/renaming - never follow.
102 : * if LOOKUP_FOLLOW passed - follow.
103 : * if the pathname has trailing slashes - follow.
104 : * otherwise - don't follow.
105 : * (applied in that order).
106 : *
107 : * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
108 : * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
109 : * During the 2.4 we need to fix the userland stuff depending on it -
110 : * hopefully we will be able to get rid of that wart in 2.5. So far only
111 : * XEmacs seems to be relying on it...
112 : */
113 : /*
114 : * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
115 : * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
116 : * any extra contention...
117 : */
118 :
119 : /* In order to reduce some races, while at the same time doing additional
120 : * checking and hopefully speeding things up, we copy filenames to the
121 : * kernel data space before using them..
122 : *
123 : * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
124 : * PATH_MAX includes the nul terminator --RR.
125 : */
126 :
127 : #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
128 :
129 : struct filename *
130 4231639341 : getname_flags(const char __user *filename, int flags, int *empty)
131 : {
132 4231639341 : struct filename *result;
133 4231639341 : char *kname;
134 4231639341 : int len;
135 :
136 4231639341 : result = audit_reusename(filename);
137 4231639341 : if (result)
138 : return result;
139 :
140 4231639341 : result = __getname();
141 4233347678 : if (unlikely(!result))
142 : return ERR_PTR(-ENOMEM);
143 :
144 : /*
145 : * First, try to embed the struct filename inside the names_cache
146 : * allocation
147 : */
148 4233347678 : kname = (char *)result->iname;
149 4233347678 : result->name = kname;
150 :
151 4233347678 : len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
152 4234868895 : if (unlikely(len < 0)) {
153 34 : __putname(result);
154 34 : return ERR_PTR(len);
155 : }
156 :
157 : /*
158 : * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
159 : * separate struct filename so we can dedicate the entire
160 : * names_cache allocation for the pathname, and re-do the copy from
161 : * userland.
162 : */
163 4234868861 : if (unlikely(len == EMBEDDED_NAME_MAX)) {
164 2189220 : const size_t size = offsetof(struct filename, iname[1]);
165 2189220 : kname = (char *)result;
166 :
167 : /*
168 : * size is chosen that way we to guarantee that
169 : * result->iname[0] is within the same object and that
170 : * kname can't be equal to result->iname, no matter what.
171 : */
172 2189220 : result = kzalloc(size, GFP_KERNEL);
173 2189222 : if (unlikely(!result)) {
174 0 : __putname(kname);
175 0 : return ERR_PTR(-ENOMEM);
176 : }
177 2189222 : result->name = kname;
178 2189222 : len = strncpy_from_user(kname, filename, PATH_MAX);
179 2189221 : if (unlikely(len < 0)) {
180 0 : __putname(kname);
181 0 : kfree(result);
182 0 : return ERR_PTR(len);
183 : }
184 2189221 : if (unlikely(len == PATH_MAX)) {
185 0 : __putname(kname);
186 0 : kfree(result);
187 0 : return ERR_PTR(-ENAMETOOLONG);
188 : }
189 : }
190 :
191 4234868862 : result->refcnt = 1;
192 : /* The empty path is special. */
193 4234868862 : if (unlikely(!len)) {
194 491862315 : if (empty)
195 0 : *empty = 1;
196 491862315 : if (!(flags & LOOKUP_EMPTY)) {
197 81969370 : putname(result);
198 81969370 : return ERR_PTR(-ENOENT);
199 : }
200 : }
201 :
202 4152899492 : result->uptr = filename;
203 4152899492 : result->aname = NULL;
204 4152899492 : audit_getname(result);
205 4152899492 : return result;
206 : }
207 :
208 : struct filename *
209 12 : getname_uflags(const char __user *filename, int uflags)
210 : {
211 3179623 : int flags = (uflags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
212 :
213 12 : return getname_flags(filename, flags, NULL);
214 : }
215 :
216 : struct filename *
217 437076871 : getname(const char __user * filename)
218 : {
219 702510293 : return getname_flags(filename, 0, NULL);
220 : }
221 :
222 : struct filename *
223 17303934 : getname_kernel(const char * filename)
224 : {
225 17303934 : struct filename *result;
226 17303934 : int len = strlen(filename) + 1;
227 :
228 17303934 : result = __getname();
229 17304210 : if (unlikely(!result))
230 : return ERR_PTR(-ENOMEM);
231 :
232 17304210 : if (len <= EMBEDDED_NAME_MAX) {
233 17304210 : result->name = (char *)result->iname;
234 0 : } else if (len <= PATH_MAX) {
235 0 : const size_t size = offsetof(struct filename, iname[1]);
236 0 : struct filename *tmp;
237 :
238 0 : tmp = kmalloc(size, GFP_KERNEL);
239 0 : if (unlikely(!tmp)) {
240 0 : __putname(result);
241 0 : return ERR_PTR(-ENOMEM);
242 : }
243 0 : tmp->name = (char *)result;
244 0 : result = tmp;
245 : } else {
246 0 : __putname(result);
247 0 : return ERR_PTR(-ENAMETOOLONG);
248 : }
249 34608420 : memcpy((char *)result->name, filename, len);
250 17304210 : result->uptr = NULL;
251 17304210 : result->aname = NULL;
252 17304210 : result->refcnt = 1;
253 17304210 : audit_getname(result);
254 :
255 17304210 : return result;
256 : }
257 : EXPORT_SYMBOL(getname_kernel);
258 :
259 4333372592 : void putname(struct filename *name)
260 : {
261 4333372592 : if (IS_ERR(name))
262 : return;
263 :
264 4251406306 : BUG_ON(name->refcnt <= 0);
265 :
266 4251406306 : if (--name->refcnt > 0)
267 : return;
268 :
269 4251406306 : if (name->name != name->iname) {
270 2189224 : __putname(name->name);
271 2189224 : kfree(name);
272 : } else
273 4249217082 : __putname(name);
274 : }
275 : EXPORT_SYMBOL(putname);
276 :
277 : /**
278 : * check_acl - perform ACL permission checking
279 : * @idmap: idmap of the mount the inode was found from
280 : * @inode: inode to check permissions on
281 : * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
282 : *
283 : * This function performs the ACL permission checking. Since this function
284 : * retrieve POSIX acls it needs to know whether it is called from a blocking or
285 : * non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
286 : *
287 : * If the inode has been found through an idmapped mount the idmap of
288 : * the vfsmount must be passed through @idmap. This function will then take
289 : * care to map the inode according to @idmap before checking permissions.
290 : * On non-idmapped mounts or if permission checking is to be performed on the
291 : * raw inode simply passs @nop_mnt_idmap.
292 : */
293 4869599739 : static int check_acl(struct mnt_idmap *idmap,
294 : struct inode *inode, int mask)
295 : {
296 : #ifdef CONFIG_FS_POSIX_ACL
297 4869599739 : struct posix_acl *acl;
298 :
299 4869599739 : if (mask & MAY_NOT_BLOCK) {
300 4748960508 : acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
301 4749301944 : if (!acl)
302 : return -EAGAIN;
303 : /* no ->get_inode_acl() calls in RCU mode... */
304 20317 : if (is_uncached_acl(acl))
305 : return -ECHILD;
306 62 : return posix_acl_permission(idmap, inode, acl, mask);
307 : }
308 :
309 120639231 : acl = get_inode_acl(inode, ACL_TYPE_ACCESS);
310 120637154 : if (IS_ERR(acl))
311 97 : return PTR_ERR(acl);
312 120637057 : if (acl) {
313 88 : int error = posix_acl_permission(idmap, inode, acl, mask);
314 88 : posix_acl_release(acl);
315 88 : return error;
316 : }
317 : #endif
318 :
319 : return -EAGAIN;
320 : }
321 :
322 : /**
323 : * acl_permission_check - perform basic UNIX permission checking
324 : * @idmap: idmap of the mount the inode was found from
325 : * @inode: inode to check permissions on
326 : * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
327 : *
328 : * This function performs the basic UNIX permission checking. Since this
329 : * function may retrieve POSIX acls it needs to know whether it is called from a
330 : * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
331 : *
332 : * If the inode has been found through an idmapped mount the idmap of
333 : * the vfsmount must be passed through @idmap. This function will then take
334 : * care to map the inode according to @idmap before checking permissions.
335 : * On non-idmapped mounts or if permission checking is to be performed on the
336 : * raw inode simply passs @nop_mnt_idmap.
337 : */
338 24117881563 : static int acl_permission_check(struct mnt_idmap *idmap,
339 : struct inode *inode, int mask)
340 : {
341 24117881563 : unsigned int mode = inode->i_mode;
342 24117881563 : vfsuid_t vfsuid;
343 :
344 : /* Are we the owner? If so, ACL's don't matter */
345 24117881563 : vfsuid = i_uid_into_vfsuid(idmap, inode);
346 48413853586 : if (likely(vfsuid_eq_kuid(vfsuid, current_fsuid()))) {
347 19330641486 : mask &= 7;
348 19330641486 : mode >>= 6;
349 19330641486 : return (mask & ~mode) ? -EACCES : 0;
350 : }
351 :
352 : /* Do we have ACL's? */
353 4876385321 : if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
354 4869428527 : int error = check_acl(idmap, inode, mask);
355 4869847536 : if (error != -EAGAIN)
356 : return error;
357 : }
358 :
359 : /* Only RWX matters for group/other mode bits */
360 4876767271 : mask &= 7;
361 :
362 : /*
363 : * Are the group permissions different from
364 : * the other permissions in the bits we care
365 : * about? Need to check group ownership if so.
366 : */
367 4876767271 : if (mask & (mode ^ (mode >> 3))) {
368 1636498 : vfsgid_t vfsgid = i_gid_into_vfsgid(idmap, inode);
369 1636490 : if (vfsgid_in_group_p(vfsgid))
370 269 : mode >>= 3;
371 : }
372 :
373 : /* Bits in 'mode' clear that we require? */
374 4876767178 : return (mask & ~mode) ? -EACCES : 0;
375 : }
376 :
377 : /**
378 : * generic_permission - check for access rights on a Posix-like filesystem
379 : * @idmap: idmap of the mount the inode was found from
380 : * @inode: inode to check access rights for
381 : * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC,
382 : * %MAY_NOT_BLOCK ...)
383 : *
384 : * Used to check for read/write/execute permissions on a file.
385 : * We use "fsuid" for this, letting us set arbitrary permissions
386 : * for filesystem access without changing the "normal" uids which
387 : * are used for other things.
388 : *
389 : * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
390 : * request cannot be satisfied (eg. requires blocking or too much complexity).
391 : * It would then be called again in ref-walk mode.
392 : *
393 : * If the inode has been found through an idmapped mount the idmap of
394 : * the vfsmount must be passed through @idmap. This function will then take
395 : * care to map the inode according to @idmap before checking permissions.
396 : * On non-idmapped mounts or if permission checking is to be performed on the
397 : * raw inode simply passs @nop_mnt_idmap.
398 : */
399 24160632381 : int generic_permission(struct mnt_idmap *idmap, struct inode *inode,
400 : int mask)
401 : {
402 24160632381 : int ret;
403 :
404 : /*
405 : * Do the basic permission checks.
406 : */
407 24160632381 : ret = acl_permission_check(idmap, inode, mask);
408 24207038180 : if (ret != -EACCES)
409 : return ret;
410 :
411 2525217 : if (S_ISDIR(inode->i_mode)) {
412 : /* DACs are overridable for directories */
413 1820989 : if (!(mask & MAY_WRITE))
414 74428 : if (capable_wrt_inode_uidgid(idmap, inode,
415 : CAP_DAC_READ_SEARCH))
416 : return 0;
417 1746981 : if (capable_wrt_inode_uidgid(idmap, inode,
418 : CAP_DAC_OVERRIDE))
419 : return 0;
420 560 : return -EACCES;
421 : }
422 :
423 : /*
424 : * Searching includes executable on directories, else just read.
425 : */
426 704228 : mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
427 704228 : if (mask == MAY_READ)
428 1303 : if (capable_wrt_inode_uidgid(idmap, inode,
429 : CAP_DAC_READ_SEARCH))
430 : return 0;
431 : /*
432 : * Read/write DACs are always overridable.
433 : * Executable DACs are overridable when there is
434 : * at least one exec bit set.
435 : */
436 703955 : if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
437 698579 : if (capable_wrt_inode_uidgid(idmap, inode,
438 : CAP_DAC_OVERRIDE))
439 689227 : return 0;
440 :
441 : return -EACCES;
442 : }
443 : EXPORT_SYMBOL(generic_permission);
444 :
445 : /**
446 : * do_inode_permission - UNIX permission checking
447 : * @idmap: idmap of the mount the inode was found from
448 : * @inode: inode to check permissions on
449 : * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
450 : *
451 : * We _really_ want to just do "generic_permission()" without
452 : * even looking at the inode->i_op values. So we keep a cache
453 : * flag in inode->i_opflags, that says "this has not special
454 : * permission function, use the fast case".
455 : */
456 26794858445 : static inline int do_inode_permission(struct mnt_idmap *idmap,
457 : struct inode *inode, int mask)
458 : {
459 26794858445 : if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
460 3001872519 : if (likely(inode->i_op->permission))
461 2963174381 : return inode->i_op->permission(idmap, inode, mask);
462 :
463 : /* This gets set once for the inode lifetime */
464 38698138 : spin_lock(&inode->i_lock);
465 38698431 : inode->i_opflags |= IOP_FASTPERM;
466 38698431 : spin_unlock(&inode->i_lock);
467 : }
468 23831682982 : return generic_permission(idmap, inode, mask);
469 : }
470 :
471 : /**
472 : * sb_permission - Check superblock-level permissions
473 : * @sb: Superblock of inode to check permission on
474 : * @inode: Inode to check permission on
475 : * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
476 : *
477 : * Separate out file-system wide checks from inode-specific permission checks.
478 : */
479 26875933409 : static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
480 : {
481 26875933409 : if (unlikely(mask & MAY_WRITE)) {
482 741738848 : umode_t mode = inode->i_mode;
483 :
484 : /* Nobody gets write access to a read-only fs. */
485 741738848 : if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
486 2 : return -EROFS;
487 : }
488 : return 0;
489 : }
490 :
491 : /**
492 : * inode_permission - Check for access rights to a given inode
493 : * @idmap: idmap of the mount the inode was found from
494 : * @inode: Inode to check permission on
495 : * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
496 : *
497 : * Check for read/write/execute permissions on an inode. We use fs[ug]id for
498 : * this, letting us set arbitrary permissions for filesystem access without
499 : * changing the "normal" UIDs which are used for other things.
500 : *
501 : * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
502 : */
503 26856255673 : int inode_permission(struct mnt_idmap *idmap,
504 : struct inode *inode, int mask)
505 : {
506 26856255673 : int retval;
507 :
508 26856255673 : retval = sb_permission(inode->i_sb, inode, mask);
509 26856255673 : if (retval)
510 : return retval;
511 :
512 26856255671 : if (unlikely(mask & MAY_WRITE)) {
513 : /*
514 : * Nobody gets write access to an immutable file.
515 : */
516 741752870 : if (IS_IMMUTABLE(inode))
517 : return -EPERM;
518 :
519 : /*
520 : * Updating mtime will likely cause i_uid and i_gid to be
521 : * written back improperly if their true value is unknown
522 : * to the vfs.
523 : */
524 741752774 : if (HAS_UNMAPPED_ID(idmap, inode))
525 : return -EACCES;
526 : }
527 :
528 26856296373 : retval = do_inode_permission(idmap, inode, mask);
529 26872180277 : if (retval)
530 : return retval;
531 :
532 26798349573 : retval = devcgroup_inode_permission(inode, mask);
533 26806355010 : if (retval)
534 0 : return retval;
535 :
536 : return security_inode_permission(inode, mask);
537 : }
538 : EXPORT_SYMBOL(inode_permission);
539 :
540 : /**
541 : * path_get - get a reference to a path
542 : * @path: path to get the reference to
543 : *
544 : * Given a path increment the reference count to the dentry and the vfsmount.
545 : */
546 1283096507 : void path_get(const struct path *path)
547 : {
548 1283096507 : mntget(path->mnt);
549 1283268499 : dget(path->dentry);
550 1283296565 : }
551 : EXPORT_SYMBOL(path_get);
552 :
553 : /**
554 : * path_put - put a reference to a path
555 : * @path: path to put the reference to
556 : *
557 : * Given a path decrement the reference count to the dentry and the vfsmount.
558 : */
559 7510259374 : void path_put(const struct path *path)
560 : {
561 7510259374 : dput(path->dentry);
562 7516290541 : mntput(path->mnt);
563 7509672020 : }
564 : EXPORT_SYMBOL(path_put);
565 :
566 : #define EMBEDDED_LEVELS 2
567 : struct nameidata {
568 : struct path path;
569 : struct qstr last;
570 : struct path root;
571 : struct inode *inode; /* path.dentry.d_inode */
572 : unsigned int flags, state;
573 : unsigned seq, next_seq, m_seq, r_seq;
574 : int last_type;
575 : unsigned depth;
576 : int total_link_count;
577 : struct saved {
578 : struct path link;
579 : struct delayed_call done;
580 : const char *name;
581 : unsigned seq;
582 : } *stack, internal[EMBEDDED_LEVELS];
583 : struct filename *name;
584 : struct nameidata *saved;
585 : unsigned root_seq;
586 : int dfd;
587 : vfsuid_t dir_vfsuid;
588 : umode_t dir_mode;
589 : } __randomize_layout;
590 :
591 : #define ND_ROOT_PRESET 1
592 : #define ND_ROOT_GRABBED 2
593 : #define ND_JUMPED 4
594 :
595 3931964958 : static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name)
596 : {
597 3931964958 : struct nameidata *old = current->nameidata;
598 3931964958 : p->stack = p->internal;
599 3931964958 : p->depth = 0;
600 3931964958 : p->dfd = dfd;
601 3931964958 : p->name = name;
602 3931964958 : p->path.mnt = NULL;
603 3931964958 : p->path.dentry = NULL;
604 3931964958 : p->total_link_count = old ? old->total_link_count : 0;
605 3931964958 : p->saved = old;
606 3931964958 : current->nameidata = p;
607 3931964958 : }
608 :
609 : static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name,
610 : const struct path *root)
611 : {
612 3933993648 : __set_nameidata(p, dfd, name);
613 3933551777 : p->state = 0;
614 3486418685 : if (unlikely(root)) {
615 6376 : p->state = ND_ROOT_PRESET;
616 6376 : p->root = *root;
617 : }
618 : }
619 :
620 3932552352 : static void restore_nameidata(void)
621 : {
622 3932552352 : struct nameidata *now = current->nameidata, *old = now->saved;
623 :
624 3932552352 : current->nameidata = old;
625 3932552352 : if (old)
626 0 : old->total_link_count = now->total_link_count;
627 3932552352 : if (now->stack != now->internal)
628 0 : kfree(now->stack);
629 3932552352 : }
630 :
631 0 : static bool nd_alloc_stack(struct nameidata *nd)
632 : {
633 0 : struct saved *p;
634 :
635 0 : p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
636 0 : nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL);
637 0 : if (unlikely(!p))
638 : return false;
639 0 : memcpy(p, nd->internal, sizeof(nd->internal));
640 0 : nd->stack = p;
641 0 : return true;
642 : }
643 :
644 : /**
645 : * path_connected - Verify that a dentry is below mnt.mnt_root
646 : *
647 : * Rename can sometimes move a file or directory outside of a bind
648 : * mount, path_connected allows those cases to be detected.
649 : */
650 138950078 : static bool path_connected(struct vfsmount *mnt, struct dentry *dentry)
651 : {
652 138950078 : struct super_block *sb = mnt->mnt_sb;
653 :
654 : /* Bind mounts can have disconnected paths */
655 138950078 : if (mnt->mnt_root == sb->s_root)
656 : return true;
657 :
658 4464606 : return is_subdir(dentry, mnt->mnt_root);
659 : }
660 :
661 3932881605 : static void drop_links(struct nameidata *nd)
662 : {
663 3932881605 : int i = nd->depth;
664 3934104271 : while (i--) {
665 98796 : struct saved *last = nd->stack + i;
666 98796 : do_delayed_call(&last->done);
667 1222666 : clear_delayed_call(&last->done);
668 : }
669 3934005475 : }
670 :
671 : static void leave_rcu(struct nameidata *nd)
672 : {
673 3525826204 : nd->flags &= ~LOOKUP_RCU;
674 3525826204 : nd->seq = nd->next_seq = 0;
675 3525826204 : rcu_read_unlock();
676 168447716 : }
677 :
678 3933207355 : static void terminate_walk(struct nameidata *nd)
679 : {
680 3933207355 : drop_links(nd);
681 3934058299 : if (!(nd->flags & LOOKUP_RCU)) {
682 3765609694 : int i;
683 3765609694 : path_put(&nd->path);
684 7531134330 : for (i = 0; i < nd->depth; i++)
685 27251 : path_put(&nd->stack[i].link);
686 3765667204 : if (nd->state & ND_ROOT_GRABBED) {
687 340795637 : path_put(&nd->root);
688 340852624 : nd->state &= ~ND_ROOT_GRABBED;
689 : }
690 : } else {
691 168448605 : leave_rcu(nd);
692 : }
693 3934171907 : nd->depth = 0;
694 3934171907 : nd->path.mnt = NULL;
695 3934171907 : nd->path.dentry = NULL;
696 3934171907 : }
697 :
698 : /* path_put is needed afterwards regardless of success or failure */
699 3694735438 : static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq)
700 : {
701 3694735438 : int res = __legitimize_mnt(path->mnt, mseq);
702 3696538048 : if (unlikely(res)) {
703 2966 : if (res > 0)
704 2700 : path->mnt = NULL;
705 2966 : path->dentry = NULL;
706 2966 : return false;
707 : }
708 3696535082 : if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
709 13 : path->dentry = NULL;
710 13 : return false;
711 : }
712 3695209486 : return !read_seqcount_retry(&path->dentry->d_seq, seq);
713 : }
714 :
715 : static inline bool legitimize_path(struct nameidata *nd,
716 : struct path *path, unsigned seq)
717 : {
718 3694368797 : return __legitimize_path(path, seq, nd->m_seq);
719 : }
720 :
721 3355621438 : static bool legitimize_links(struct nameidata *nd)
722 : {
723 3355621438 : int i;
724 3355621438 : if (unlikely(nd->flags & LOOKUP_CACHED)) {
725 0 : drop_links(nd);
726 0 : nd->depth = 0;
727 0 : return false;
728 : }
729 3363963782 : for (i = 0; i < nd->depth; i++) {
730 7774244 : struct saved *last = nd->stack + i;
731 7774244 : if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
732 17 : drop_links(nd);
733 17 : nd->depth = i + 1;
734 17 : return false;
735 : }
736 : }
737 : return true;
738 : }
739 :
740 3357320382 : static bool legitimize_root(struct nameidata *nd)
741 : {
742 : /* Nothing to do if nd->root is zero or is managed by the VFS user. */
743 3357320382 : if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET))
744 : return true;
745 338696228 : nd->state |= ND_ROOT_GRABBED;
746 338696228 : return legitimize_path(nd, &nd->root, nd->root_seq);
747 : }
748 :
749 : /*
750 : * Path walking has 2 modes, rcu-walk and ref-walk (see
751 : * Documentation/filesystems/path-lookup.txt). In situations when we can't
752 : * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
753 : * normal reference counts on dentries and vfsmounts to transition to ref-walk
754 : * mode. Refcounts are grabbed at the last known good point before rcu-walk
755 : * got stuck, so ref-walk may continue from there. If this is not successful
756 : * (eg. a seqcount has changed), then failure is returned and it's up to caller
757 : * to restart the path walk from the beginning in ref-walk mode.
758 : */
759 :
760 : /**
761 : * try_to_unlazy - try to switch to ref-walk mode.
762 : * @nd: nameidata pathwalk data
763 : * Returns: true on success, false on failure
764 : *
765 : * try_to_unlazy attempts to legitimize the current nd->path and nd->root
766 : * for ref-walk mode.
767 : * Must be called from rcu-walk context.
768 : * Nothing should touch nameidata between try_to_unlazy() failure and
769 : * terminate_walk().
770 : */
771 3347763374 : static bool try_to_unlazy(struct nameidata *nd)
772 : {
773 3347763374 : struct dentry *parent = nd->path.dentry;
774 :
775 3347763374 : BUG_ON(!(nd->flags & LOOKUP_RCU));
776 :
777 3347763374 : if (unlikely(!legitimize_links(nd)))
778 17 : goto out1;
779 3347898325 : if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
780 3058 : goto out;
781 3349777352 : if (unlikely(!legitimize_root(nd)))
782 33 : goto out;
783 3349917495 : leave_rcu(nd);
784 3349901424 : BUG_ON(nd->inode != parent->d_inode);
785 : return true;
786 :
787 : out1:
788 17 : nd->path.mnt = NULL;
789 17 : nd->path.dentry = NULL;
790 3108 : out:
791 3108 : leave_rcu(nd);
792 3108 : return false;
793 : }
794 :
795 : /**
796 : * try_to_unlazy_next - try to switch to ref-walk mode.
797 : * @nd: nameidata pathwalk data
798 : * @dentry: next dentry to step into
799 : * Returns: true on success, false on failure
800 : *
801 : * Similar to try_to_unlazy(), but here we have the next dentry already
802 : * picked by rcu-walk and want to legitimize that in addition to the current
803 : * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context.
804 : * Nothing should touch nameidata between try_to_unlazy_next() failure and
805 : * terminate_walk().
806 : */
807 7456400 : static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry)
808 : {
809 7456400 : int res;
810 7456400 : BUG_ON(!(nd->flags & LOOKUP_RCU));
811 :
812 7456400 : if (unlikely(!legitimize_links(nd)))
813 0 : goto out2;
814 7456431 : res = __legitimize_mnt(nd->path.mnt, nd->m_seq);
815 7456436 : if (unlikely(res)) {
816 1907 : if (res > 0)
817 1900 : goto out2;
818 7 : goto out1;
819 : }
820 7454529 : if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
821 0 : goto out1;
822 :
823 : /*
824 : * We need to move both the parent and the dentry from the RCU domain
825 : * to be properly refcounted. And the sequence number in the dentry
826 : * validates *both* dentry counters, since we checked the sequence
827 : * number of the parent after we got the child sequence number. So we
828 : * know the parent must still be valid if the child sequence number is
829 : */
830 7455084 : if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
831 0 : goto out;
832 7455064 : if (read_seqcount_retry(&dentry->d_seq, nd->next_seq))
833 9 : goto out_dput;
834 : /*
835 : * Sequence counts matched. Now make sure that the root is
836 : * still valid and get it if required.
837 : */
838 7455072 : if (unlikely(!legitimize_root(nd)))
839 13 : goto out_dput;
840 7455067 : leave_rcu(nd);
841 7455067 : return true;
842 :
843 1900 : out2:
844 1900 : nd->path.mnt = NULL;
845 1907 : out1:
846 1907 : nd->path.dentry = NULL;
847 1907 : out:
848 1907 : leave_rcu(nd);
849 1907 : return false;
850 22 : out_dput:
851 22 : leave_rcu(nd);
852 22 : dput(dentry);
853 22 : return false;
854 : }
855 :
856 24053195740 : static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
857 : {
858 24053195740 : if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
859 2790905436 : return dentry->d_op->d_revalidate(dentry, flags);
860 : else
861 : return 1;
862 : }
863 :
864 : /**
865 : * complete_walk - successful completion of path walk
866 : * @nd: pointer nameidata
867 : *
868 : * If we had been in RCU mode, drop out of it and legitimize nd->path.
869 : * Revalidate the final result, unless we'd already done that during
870 : * the path walk or the filesystem doesn't ask for it. Return 0 on
871 : * success, -error on failure. In case of failure caller does not
872 : * need to drop nd->path.
873 : */
874 3474981132 : static int complete_walk(struct nameidata *nd)
875 : {
876 3474981132 : struct dentry *dentry = nd->path.dentry;
877 3474981132 : int status;
878 :
879 3474981132 : if (nd->flags & LOOKUP_RCU) {
880 : /*
881 : * We don't want to zero nd->root for scoped-lookups or
882 : * externally-managed nd->root.
883 : */
884 2924903190 : if (!(nd->state & ND_ROOT_PRESET))
885 2924988328 : if (!(nd->flags & LOOKUP_IS_SCOPED))
886 2924969453 : nd->root.mnt = NULL;
887 2924903190 : nd->flags &= ~LOOKUP_CACHED;
888 2924903190 : if (!try_to_unlazy(nd))
889 : return -ECHILD;
890 : }
891 :
892 3476353809 : if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
893 : /*
894 : * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
895 : * ever step outside the root during lookup" and should already
896 : * be guaranteed by the rest of namei, we want to avoid a namei
897 : * BUG resulting in userspace being given a path that was not
898 : * scoped within the root at some point during the lookup.
899 : *
900 : * So, do a final sanity-check to make sure that in the
901 : * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
902 : * we won't silently return an fd completely outside of the
903 : * requested root to userspace.
904 : *
905 : * Userspace could move the path outside the root after this
906 : * check, but as discussed elsewhere this is not a concern (the
907 : * resolved file was inside the root at some point).
908 : */
909 0 : if (!path_is_under(&nd->path, &nd->root))
910 : return -EXDEV;
911 : }
912 :
913 3476353809 : if (likely(!(nd->state & ND_JUMPED)))
914 : return 0;
915 :
916 1754271161 : if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
917 : return 0;
918 :
919 145047562 : status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
920 145043570 : if (status > 0)
921 : return 0;
922 :
923 0 : if (!status)
924 0 : status = -ESTALE;
925 :
926 : return status;
927 : }
928 :
929 1391264694 : static int set_root(struct nameidata *nd)
930 : {
931 1391264694 : struct fs_struct *fs = current->fs;
932 :
933 : /*
934 : * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
935 : * still have to ensure it doesn't happen because it will cause a breakout
936 : * from the dirfd.
937 : */
938 1391264694 : if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED))
939 : return -ENOTRECOVERABLE;
940 :
941 1391264694 : if (nd->flags & LOOKUP_RCU) {
942 : unsigned seq;
943 :
944 : do {
945 1389126582 : seq = read_seqcount_begin(&fs->seq);
946 1389432265 : nd->root = fs->root;
947 1389432265 : nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
948 1389459601 : } while (read_seqcount_retry(&fs->seq, seq));
949 : } else {
950 2138112 : get_fs_root(fs, &nd->root);
951 2138228 : nd->state |= ND_ROOT_GRABBED;
952 : }
953 : return 0;
954 : }
955 :
956 1273454473 : static int nd_jump_root(struct nameidata *nd)
957 : {
958 1273454473 : if (unlikely(nd->flags & LOOKUP_BENEATH))
959 : return -EXDEV;
960 1273454473 : if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
961 : /* Absolute path arguments to path_init() are allowed. */
962 0 : if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt)
963 : return -EXDEV;
964 : }
965 1273454473 : if (!nd->root.mnt) {
966 1260459883 : int error = set_root(nd);
967 1260576723 : if (error)
968 : return error;
969 : }
970 1273571313 : if (nd->flags & LOOKUP_RCU) {
971 1267983973 : struct dentry *d;
972 1267983973 : nd->path = nd->root;
973 1267983973 : d = nd->path.dentry;
974 1267983973 : nd->inode = d->d_inode;
975 1267983973 : nd->seq = nd->root_seq;
976 1267983973 : if (read_seqcount_retry(&d->d_seq, nd->seq))
977 : return -ECHILD;
978 : } else {
979 5587340 : path_put(&nd->path);
980 5587434 : nd->path = nd->root;
981 5587434 : path_get(&nd->path);
982 5587435 : nd->inode = nd->path.dentry->d_inode;
983 : }
984 1273524273 : nd->state |= ND_JUMPED;
985 1273524273 : return 0;
986 : }
987 :
988 : /*
989 : * Helper to directly jump to a known parsed path from ->get_link,
990 : * caller must have taken a reference to path beforehand.
991 : */
992 750929 : int nd_jump_link(const struct path *path)
993 : {
994 750929 : int error = -ELOOP;
995 750929 : struct nameidata *nd = current->nameidata;
996 :
997 750929 : if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS))
998 0 : goto err;
999 :
1000 750929 : error = -EXDEV;
1001 750929 : if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
1002 0 : if (nd->path.mnt != path->mnt)
1003 0 : goto err;
1004 : }
1005 : /* Not currently safe for scoped-lookups. */
1006 750929 : if (unlikely(nd->flags & LOOKUP_IS_SCOPED))
1007 0 : goto err;
1008 :
1009 750929 : path_put(&nd->path);
1010 750929 : nd->path = *path;
1011 750929 : nd->inode = nd->path.dentry->d_inode;
1012 750929 : nd->state |= ND_JUMPED;
1013 750929 : return 0;
1014 :
1015 0 : err:
1016 0 : path_put(path);
1017 0 : return error;
1018 : }
1019 :
1020 97403963 : static inline void put_link(struct nameidata *nd)
1021 : {
1022 97403963 : struct saved *last = nd->stack + --nd->depth;
1023 97403963 : do_delayed_call(&last->done);
1024 97407696 : if (!(nd->flags & LOOKUP_RCU))
1025 11154702 : path_put(&last->link);
1026 97407855 : }
1027 :
1028 : static int sysctl_protected_symlinks __read_mostly;
1029 : static int sysctl_protected_hardlinks __read_mostly;
1030 : static int sysctl_protected_fifos __read_mostly;
1031 : static int sysctl_protected_regular __read_mostly;
1032 :
1033 : #ifdef CONFIG_SYSCTL
1034 : static struct ctl_table namei_sysctls[] = {
1035 : {
1036 : .procname = "protected_symlinks",
1037 : .data = &sysctl_protected_symlinks,
1038 : .maxlen = sizeof(int),
1039 : .mode = 0644,
1040 : .proc_handler = proc_dointvec_minmax,
1041 : .extra1 = SYSCTL_ZERO,
1042 : .extra2 = SYSCTL_ONE,
1043 : },
1044 : {
1045 : .procname = "protected_hardlinks",
1046 : .data = &sysctl_protected_hardlinks,
1047 : .maxlen = sizeof(int),
1048 : .mode = 0644,
1049 : .proc_handler = proc_dointvec_minmax,
1050 : .extra1 = SYSCTL_ZERO,
1051 : .extra2 = SYSCTL_ONE,
1052 : },
1053 : {
1054 : .procname = "protected_fifos",
1055 : .data = &sysctl_protected_fifos,
1056 : .maxlen = sizeof(int),
1057 : .mode = 0644,
1058 : .proc_handler = proc_dointvec_minmax,
1059 : .extra1 = SYSCTL_ZERO,
1060 : .extra2 = SYSCTL_TWO,
1061 : },
1062 : {
1063 : .procname = "protected_regular",
1064 : .data = &sysctl_protected_regular,
1065 : .maxlen = sizeof(int),
1066 : .mode = 0644,
1067 : .proc_handler = proc_dointvec_minmax,
1068 : .extra1 = SYSCTL_ZERO,
1069 : .extra2 = SYSCTL_TWO,
1070 : },
1071 : { }
1072 : };
1073 :
1074 0 : static int __init init_fs_namei_sysctls(void)
1075 : {
1076 0 : register_sysctl_init("fs", namei_sysctls);
1077 0 : return 0;
1078 : }
1079 : fs_initcall(init_fs_namei_sysctls);
1080 :
1081 : #endif /* CONFIG_SYSCTL */
1082 :
1083 : /**
1084 : * may_follow_link - Check symlink following for unsafe situations
1085 : * @nd: nameidata pathwalk data
1086 : *
1087 : * In the case of the sysctl_protected_symlinks sysctl being enabled,
1088 : * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
1089 : * in a sticky world-writable directory. This is to protect privileged
1090 : * processes from failing races against path names that may change out
1091 : * from under them by way of other users creating malicious symlinks.
1092 : * It will permit symlinks to be followed only when outside a sticky
1093 : * world-writable directory, or when the uid of the symlink and follower
1094 : * match, or when the directory owner matches the symlink's owner.
1095 : *
1096 : * Returns 0 if following the symlink is allowed, -ve on error.
1097 : */
1098 38191768 : static inline int may_follow_link(struct nameidata *nd, const struct inode *inode)
1099 : {
1100 38191768 : struct mnt_idmap *idmap;
1101 38191768 : vfsuid_t vfsuid;
1102 :
1103 38191768 : if (!sysctl_protected_symlinks)
1104 : return 0;
1105 :
1106 38192699 : idmap = mnt_idmap(nd->path.mnt);
1107 38192334 : vfsuid = i_uid_into_vfsuid(idmap, inode);
1108 : /* Allowed if owner and follower match. */
1109 76384905 : if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
1110 : return 0;
1111 :
1112 : /* Allowed if parent directory not sticky and world-writable. */
1113 165854 : if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
1114 : return 0;
1115 :
1116 : /* Allowed if parent directory and link owner match. */
1117 8 : if (vfsuid_valid(nd->dir_vfsuid) && vfsuid_eq(nd->dir_vfsuid, vfsuid))
1118 : return 0;
1119 :
1120 4 : if (nd->flags & LOOKUP_RCU)
1121 2 : return -ECHILD;
1122 :
1123 : audit_inode(nd->name, nd->stack[0].link.dentry, 0);
1124 : audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link");
1125 : return -EACCES;
1126 : }
1127 :
1128 : /**
1129 : * safe_hardlink_source - Check for safe hardlink conditions
1130 : * @idmap: idmap of the mount the inode was found from
1131 : * @inode: the source inode to hardlink from
1132 : *
1133 : * Return false if at least one of the following conditions:
1134 : * - inode is not a regular file
1135 : * - inode is setuid
1136 : * - inode is setgid and group-exec
1137 : * - access failure for read and write
1138 : *
1139 : * Otherwise returns true.
1140 : */
1141 3103646 : static bool safe_hardlink_source(struct mnt_idmap *idmap,
1142 : struct inode *inode)
1143 : {
1144 3103646 : umode_t mode = inode->i_mode;
1145 :
1146 : /* Special files should not get pinned to the filesystem. */
1147 3103646 : if (!S_ISREG(mode))
1148 : return false;
1149 :
1150 : /* Setuid files should not get pinned to the filesystem. */
1151 1957250 : if (mode & S_ISUID)
1152 : return false;
1153 :
1154 : /* Executable setgid files should not get pinned to the filesystem. */
1155 1957250 : if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
1156 : return false;
1157 :
1158 : /* Hardlinking to unreadable or unwritable sources is dangerous. */
1159 1957250 : if (inode_permission(idmap, inode, MAY_READ | MAY_WRITE))
1160 6 : return false;
1161 :
1162 : return true;
1163 : }
1164 :
1165 : /**
1166 : * may_linkat - Check permissions for creating a hardlink
1167 : * @idmap: idmap of the mount the inode was found from
1168 : * @link: the source to hardlink from
1169 : *
1170 : * Block hardlink when all of:
1171 : * - sysctl_protected_hardlinks enabled
1172 : * - fsuid does not match inode
1173 : * - hardlink source is unsafe (see safe_hardlink_source() above)
1174 : * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1175 : *
1176 : * If the inode has been found through an idmapped mount the idmap of
1177 : * the vfsmount must be passed through @idmap. This function will then take
1178 : * care to map the inode according to @idmap before checking permissions.
1179 : * On non-idmapped mounts or if permission checking is to be performed on the
1180 : * raw inode simply pass @nop_mnt_idmap.
1181 : *
1182 : * Returns 0 if successful, -ve on error.
1183 : */
1184 3103664 : int may_linkat(struct mnt_idmap *idmap, const struct path *link)
1185 : {
1186 3103664 : struct inode *inode = link->dentry->d_inode;
1187 :
1188 : /* Inode writeback is not safe when the uid or gid are invalid. */
1189 6207311 : if (!vfsuid_valid(i_uid_into_vfsuid(idmap, inode)) ||
1190 : !vfsgid_valid(i_gid_into_vfsgid(idmap, inode)))
1191 0 : return -EOVERFLOW;
1192 :
1193 3103647 : if (!sysctl_protected_hardlinks)
1194 : return 0;
1195 :
1196 : /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1197 : * otherwise, it must be a safe source.
1198 : */
1199 4250065 : if (safe_hardlink_source(idmap, inode) ||
1200 1146419 : inode_owner_or_capable(idmap, inode))
1201 3102024 : return 0;
1202 :
1203 : audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
1204 : return -EPERM;
1205 : }
1206 :
1207 : /**
1208 : * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1209 : * should be allowed, or not, on files that already
1210 : * exist.
1211 : * @idmap: idmap of the mount the inode was found from
1212 : * @nd: nameidata pathwalk data
1213 : * @inode: the inode of the file to open
1214 : *
1215 : * Block an O_CREAT open of a FIFO (or a regular file) when:
1216 : * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1217 : * - the file already exists
1218 : * - we are in a sticky directory
1219 : * - we don't own the file
1220 : * - the owner of the directory doesn't own the file
1221 : * - the directory is world writable
1222 : * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1223 : * the directory doesn't have to be world writable: being group writable will
1224 : * be enough.
1225 : *
1226 : * If the inode has been found through an idmapped mount the idmap of
1227 : * the vfsmount must be passed through @idmap. This function will then take
1228 : * care to map the inode according to @idmap before checking permissions.
1229 : * On non-idmapped mounts or if permission checking is to be performed on the
1230 : * raw inode simply pass @nop_mnt_idmap.
1231 : *
1232 : * Returns 0 if the open is allowed, -ve on error.
1233 : */
1234 16632601 : static int may_create_in_sticky(struct mnt_idmap *idmap,
1235 : struct nameidata *nd, struct inode *const inode)
1236 : {
1237 16632601 : umode_t dir_mode = nd->dir_mode;
1238 16632601 : vfsuid_t dir_vfsuid = nd->dir_vfsuid;
1239 :
1240 16632601 : if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1241 16632597 : (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1242 16870746 : likely(!(dir_mode & S_ISVTX)) ||
1243 18 : vfsuid_eq(i_uid_into_vfsuid(idmap, inode), dir_vfsuid) ||
1244 18 : vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode), current_fsuid()))
1245 16632585 : return 0;
1246 :
1247 16 : if (likely(dir_mode & 0002) ||
1248 8 : (dir_mode & 0020 &&
1249 8 : ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1250 6 : (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1251 12 : const char *operation = S_ISFIFO(inode->i_mode) ?
1252 : "sticky_create_fifo" :
1253 : "sticky_create_regular";
1254 12 : audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
1255 12 : return -EACCES;
1256 : }
1257 : return 0;
1258 : }
1259 :
1260 : /*
1261 : * follow_up - Find the mountpoint of path's vfsmount
1262 : *
1263 : * Given a path, find the mountpoint of its source file system.
1264 : * Replace @path with the path of the mountpoint in the parent mount.
1265 : * Up is towards /.
1266 : *
1267 : * Return 1 if we went up a level and 0 if we were already at the
1268 : * root.
1269 : */
1270 0 : int follow_up(struct path *path)
1271 : {
1272 0 : struct mount *mnt = real_mount(path->mnt);
1273 0 : struct mount *parent;
1274 0 : struct dentry *mountpoint;
1275 :
1276 0 : read_seqlock_excl(&mount_lock);
1277 0 : parent = mnt->mnt_parent;
1278 0 : if (parent == mnt) {
1279 0 : read_sequnlock_excl(&mount_lock);
1280 0 : return 0;
1281 : }
1282 0 : mntget(&parent->mnt);
1283 0 : mountpoint = dget(mnt->mnt_mountpoint);
1284 0 : read_sequnlock_excl(&mount_lock);
1285 0 : dput(path->dentry);
1286 0 : path->dentry = mountpoint;
1287 0 : mntput(path->mnt);
1288 0 : path->mnt = &parent->mnt;
1289 0 : return 1;
1290 : }
1291 : EXPORT_SYMBOL(follow_up);
1292 :
1293 341 : static bool choose_mountpoint_rcu(struct mount *m, const struct path *root,
1294 : struct path *path, unsigned *seqp)
1295 : {
1296 341 : while (mnt_has_parent(m)) {
1297 341 : struct dentry *mountpoint = m->mnt_mountpoint;
1298 :
1299 341 : m = m->mnt_parent;
1300 341 : if (unlikely(root->dentry == mountpoint &&
1301 : root->mnt == &m->mnt))
1302 : break;
1303 341 : if (mountpoint != m->mnt.mnt_root) {
1304 341 : path->mnt = &m->mnt;
1305 341 : path->dentry = mountpoint;
1306 341 : *seqp = read_seqcount_begin(&mountpoint->d_seq);
1307 341 : return true;
1308 : }
1309 : }
1310 : return false;
1311 : }
1312 :
1313 328 : static bool choose_mountpoint(struct mount *m, const struct path *root,
1314 : struct path *path)
1315 : {
1316 328 : bool found;
1317 :
1318 328 : rcu_read_lock();
1319 0 : while (1) {
1320 328 : unsigned seq, mseq = read_seqbegin(&mount_lock);
1321 :
1322 328 : found = choose_mountpoint_rcu(m, root, path, &seq);
1323 328 : if (unlikely(!found)) {
1324 0 : if (!read_seqretry(&mount_lock, mseq))
1325 : break;
1326 : } else {
1327 328 : if (likely(__legitimize_path(path, seq, mseq)))
1328 : break;
1329 0 : rcu_read_unlock();
1330 0 : path_put(path);
1331 0 : rcu_read_lock();
1332 : }
1333 : }
1334 328 : rcu_read_unlock();
1335 328 : return found;
1336 : }
1337 :
1338 : /*
1339 : * Perform an automount
1340 : * - return -EISDIR to tell follow_managed() to stop and return the path we
1341 : * were called with.
1342 : */
1343 2 : static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1344 : {
1345 2 : struct dentry *dentry = path->dentry;
1346 :
1347 : /* We don't want to mount if someone's just doing a stat -
1348 : * unless they're stat'ing a directory and appended a '/' to
1349 : * the name.
1350 : *
1351 : * We do, however, want to mount if someone wants to open or
1352 : * create a file of any type under the mountpoint, wants to
1353 : * traverse through the mountpoint or wants to open the
1354 : * mounted directory. Also, autofs may mark negative dentries
1355 : * as being automount points. These will need the attentions
1356 : * of the daemon to instantiate them before they can be used.
1357 : */
1358 2 : if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1359 0 : LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1360 0 : dentry->d_inode)
1361 : return -EISDIR;
1362 :
1363 2 : if (count && (*count)++ >= MAXSYMLINKS)
1364 : return -ELOOP;
1365 :
1366 2 : return finish_automount(dentry->d_op->d_automount(path), path);
1367 : }
1368 :
1369 : /*
1370 : * mount traversal - out-of-line part. One note on ->d_flags accesses -
1371 : * dentries are pinned but not locked here, so negative dentry can go
1372 : * positive right under us. Use of smp_load_acquire() provides a barrier
1373 : * sufficient for ->d_inode and ->d_flags consistency.
1374 : */
1375 21259525 : static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped,
1376 : int *count, unsigned lookup_flags)
1377 : {
1378 21259525 : struct vfsmount *mnt = path->mnt;
1379 21259525 : bool need_mntput = false;
1380 21259525 : int ret = 0;
1381 :
1382 42514894 : while (flags & DCACHE_MANAGED_DENTRY) {
1383 : /* Allow the filesystem to manage the transit without i_mutex
1384 : * being held. */
1385 21273768 : if (flags & DCACHE_MANAGE_TRANSIT) {
1386 0 : ret = path->dentry->d_op->d_manage(path, false);
1387 0 : flags = smp_load_acquire(&path->dentry->d_flags);
1388 0 : if (ret < 0)
1389 : break;
1390 : }
1391 :
1392 21273768 : if (flags & DCACHE_MOUNTED) { // something's mounted on it..
1393 21273766 : struct vfsmount *mounted = lookup_mnt(path);
1394 21273918 : if (mounted) { // ... in our namespace
1395 21255252 : dput(path->dentry);
1396 21255295 : if (need_mntput)
1397 0 : mntput(path->mnt);
1398 21255295 : path->mnt = mounted;
1399 21255295 : path->dentry = dget(mounted->mnt_root);
1400 : // here we know it's positive
1401 21255132 : flags = path->dentry->d_flags;
1402 21255132 : need_mntput = true;
1403 21255132 : continue;
1404 : }
1405 : }
1406 :
1407 18668 : if (!(flags & DCACHE_NEED_AUTOMOUNT))
1408 : break;
1409 :
1410 : // uncovered automount point
1411 2 : ret = follow_automount(path, count, lookup_flags);
1412 2 : flags = smp_load_acquire(&path->dentry->d_flags);
1413 263 : if (ret < 0)
1414 : break;
1415 : }
1416 :
1417 21259792 : if (ret == -EISDIR)
1418 0 : ret = 0;
1419 : // possible if you race with several mount --move
1420 21259792 : if (need_mntput && path->mnt == mnt)
1421 0 : mntput(path->mnt);
1422 21259792 : if (!ret && unlikely(d_flags_negative(flags)))
1423 0 : ret = -ENOENT;
1424 21259792 : *jumped = need_mntput;
1425 21259792 : return ret;
1426 : }
1427 :
1428 539450377 : static inline int traverse_mounts(struct path *path, bool *jumped,
1429 : int *count, unsigned lookup_flags)
1430 : {
1431 539450377 : unsigned flags = smp_load_acquire(&path->dentry->d_flags);
1432 :
1433 : /* fastpath */
1434 539562676 : if (likely(!(flags & DCACHE_MANAGED_DENTRY))) {
1435 518303030 : *jumped = false;
1436 518303030 : if (unlikely(d_flags_negative(flags)))
1437 : return -ENOENT;
1438 240677088 : return 0;
1439 : }
1440 21259646 : return __traverse_mounts(path, flags, jumped, count, lookup_flags);
1441 : }
1442 :
1443 564 : int follow_down_one(struct path *path)
1444 : {
1445 564 : struct vfsmount *mounted;
1446 :
1447 564 : mounted = lookup_mnt(path);
1448 564 : if (mounted) {
1449 282 : dput(path->dentry);
1450 282 : mntput(path->mnt);
1451 282 : path->mnt = mounted;
1452 282 : path->dentry = dget(mounted->mnt_root);
1453 282 : return 1;
1454 : }
1455 : return 0;
1456 : }
1457 : EXPORT_SYMBOL(follow_down_one);
1458 :
1459 : /*
1460 : * Follow down to the covering mount currently visible to userspace. At each
1461 : * point, the filesystem owning that dentry may be queried as to whether the
1462 : * caller is permitted to proceed or not.
1463 : */
1464 0 : int follow_down(struct path *path, unsigned int flags)
1465 : {
1466 0 : struct vfsmount *mnt = path->mnt;
1467 0 : bool jumped;
1468 0 : int ret = traverse_mounts(path, &jumped, NULL, flags);
1469 :
1470 0 : if (path->mnt != mnt)
1471 0 : mntput(mnt);
1472 0 : return ret;
1473 : }
1474 : EXPORT_SYMBOL(follow_down);
1475 :
1476 : /*
1477 : * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1478 : * we meet a managed dentry that would need blocking.
1479 : */
1480 23704411231 : static bool __follow_mount_rcu(struct nameidata *nd, struct path *path)
1481 : {
1482 23704411231 : struct dentry *dentry = path->dentry;
1483 23704411231 : unsigned int flags = dentry->d_flags;
1484 :
1485 23704411231 : if (likely(!(flags & DCACHE_MANAGED_DENTRY)))
1486 : return true;
1487 :
1488 872394116 : if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1489 : return false;
1490 :
1491 2590384104 : for (;;) {
1492 : /*
1493 : * Don't forget we might have a non-mountpoint managed dentry
1494 : * that wants to block transit.
1495 : */
1496 1731389110 : if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) {
1497 0 : int res = dentry->d_op->d_manage(path, true);
1498 0 : if (res)
1499 0 : return res == -EISDIR;
1500 0 : flags = dentry->d_flags;
1501 : }
1502 :
1503 1731389110 : if (flags & DCACHE_MOUNTED) {
1504 1135055265 : struct mount *mounted = __lookup_mnt(path->mnt, dentry);
1505 1134695739 : if (mounted) {
1506 858562301 : path->mnt = &mounted->mnt;
1507 858562301 : dentry = path->dentry = mounted->mnt.mnt_root;
1508 858562301 : nd->state |= ND_JUMPED;
1509 858562301 : nd->next_seq = read_seqcount_begin(&dentry->d_seq);
1510 858995429 : flags = dentry->d_flags;
1511 : // makes sure that non-RCU pathwalk could reach
1512 : // this state.
1513 858995429 : if (read_seqretry(&mount_lock, nd->m_seq))
1514 : return false;
1515 858994994 : continue;
1516 : }
1517 276133438 : if (read_seqretry(&mount_lock, nd->m_seq))
1518 : return false;
1519 : }
1520 872476771 : return !(flags & DCACHE_NEED_AUTOMOUNT);
1521 : }
1522 : }
1523 :
1524 24243490805 : static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1525 : struct path *path)
1526 : {
1527 24243490805 : bool jumped;
1528 24243490805 : int ret;
1529 :
1530 24243490805 : path->mnt = nd->path.mnt;
1531 24243490805 : path->dentry = dentry;
1532 24243490805 : if (nd->flags & LOOKUP_RCU) {
1533 23712036673 : unsigned int seq = nd->next_seq;
1534 23712036673 : if (likely(__follow_mount_rcu(nd, path)))
1535 : return 0;
1536 : // *path and nd->next_seq might've been clobbered
1537 1836 : path->mnt = nd->path.mnt;
1538 1836 : path->dentry = dentry;
1539 1836 : nd->next_seq = seq;
1540 1836 : if (!try_to_unlazy_next(nd, dentry))
1541 : return -ECHILD;
1542 : }
1543 531454132 : ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags);
1544 539457520 : if (jumped) {
1545 21255343 : if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1546 : ret = -EXDEV;
1547 : else
1548 21255343 : nd->state |= ND_JUMPED;
1549 : }
1550 539457520 : if (unlikely(ret)) {
1551 277624259 : dput(path->dentry);
1552 277626260 : if (path->mnt != nd->path.mnt)
1553 0 : mntput(path->mnt);
1554 : }
1555 : return ret;
1556 : }
1557 :
1558 : /*
1559 : * This looks up the name in dcache and possibly revalidates the found dentry.
1560 : * NULL is returned if the dentry does not exist in the cache.
1561 : */
1562 337311713 : static struct dentry *lookup_dcache(const struct qstr *name,
1563 : struct dentry *dir,
1564 : unsigned int flags)
1565 : {
1566 337311713 : struct dentry *dentry = d_lookup(dir, name);
1567 337266277 : if (dentry) {
1568 285730566 : int error = d_revalidate(dentry, flags);
1569 285765799 : if (unlikely(error <= 0)) {
1570 139 : if (!error)
1571 139 : d_invalidate(dentry);
1572 139 : dput(dentry);
1573 139 : return ERR_PTR(error);
1574 : }
1575 : }
1576 : return dentry;
1577 : }
1578 :
1579 : /*
1580 : * Parent directory has inode locked exclusive. This is one
1581 : * and only case when ->lookup() gets called on non in-lookup
1582 : * dentries - as the matter of fact, this only gets called
1583 : * when directory is guaranteed to have no in-lookup children
1584 : * at all.
1585 : */
1586 333275390 : struct dentry *lookup_one_qstr_excl(const struct qstr *name,
1587 : struct dentry *base,
1588 : unsigned int flags)
1589 : {
1590 333275390 : struct dentry *dentry = lookup_dcache(name, base, flags);
1591 333249201 : struct dentry *old;
1592 333249201 : struct inode *dir = base->d_inode;
1593 :
1594 333249201 : if (dentry)
1595 : return dentry;
1596 :
1597 : /* Don't create child dentry for a dead directory. */
1598 47583883 : if (unlikely(IS_DEADDIR(dir)))
1599 : return ERR_PTR(-ENOENT);
1600 :
1601 47583883 : dentry = d_alloc(base, name);
1602 47581286 : if (unlikely(!dentry))
1603 : return ERR_PTR(-ENOMEM);
1604 :
1605 47581286 : old = dir->i_op->lookup(dir, dentry, flags);
1606 47585371 : if (unlikely(old)) {
1607 65610 : dput(dentry);
1608 65610 : dentry = old;
1609 : }
1610 : return dentry;
1611 : }
1612 : EXPORT_SYMBOL(lookup_one_qstr_excl);
1613 :
1614 24085921250 : static struct dentry *lookup_fast(struct nameidata *nd)
1615 : {
1616 24085921250 : struct dentry *dentry, *parent = nd->path.dentry;
1617 24085921250 : int status = 1;
1618 :
1619 : /*
1620 : * Rename seqlock is not required here because in the off chance
1621 : * of a false negative due to a concurrent rename, the caller is
1622 : * going to fall back to non-racy lookup.
1623 : */
1624 24085921250 : if (nd->flags & LOOKUP_RCU) {
1625 23895057884 : dentry = __d_lookup_rcu(parent, &nd->last, &nd->next_seq);
1626 23911237794 : if (unlikely(!dentry)) {
1627 331541294 : if (!try_to_unlazy(nd))
1628 : return ERR_PTR(-ECHILD);
1629 331557372 : return NULL;
1630 : }
1631 :
1632 : /*
1633 : * This sequence count validates that the parent had no
1634 : * changes while we did the lookup of the dentry above.
1635 : */
1636 23579696500 : if (read_seqcount_retry(&parent->d_seq, nd->seq))
1637 : return ERR_PTR(-ECHILD);
1638 :
1639 23598190778 : status = d_revalidate(dentry, nd->flags);
1640 23602229274 : if (likely(status > 0))
1641 : return dentry;
1642 7454567 : if (!try_to_unlazy_next(nd, dentry))
1643 : return ERR_PTR(-ECHILD);
1644 7455063 : if (status == -ECHILD)
1645 : /* we'd been told to redo it in non-rcu mode */
1646 7454958 : status = d_revalidate(dentry, nd->flags);
1647 : } else {
1648 190863366 : dentry = __d_lookup(parent, &nd->last);
1649 190834970 : if (unlikely(!dentry))
1650 : return NULL;
1651 171210984 : status = d_revalidate(dentry, nd->flags);
1652 : }
1653 178659751 : if (unlikely(status <= 0)) {
1654 497914 : if (!status)
1655 497913 : d_invalidate(dentry);
1656 497915 : dput(dentry);
1657 497916 : return ERR_PTR(status);
1658 : }
1659 : return dentry;
1660 : }
1661 :
1662 : /* Fast lookup failed, do it the slow way */
1663 302489731 : static struct dentry *__lookup_slow(const struct qstr *name,
1664 : struct dentry *dir,
1665 : unsigned int flags)
1666 : {
1667 302489731 : struct dentry *dentry, *old;
1668 302489731 : struct inode *inode = dir->d_inode;
1669 302489731 : DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1670 :
1671 : /* Don't go there if it's already dead */
1672 302489731 : if (unlikely(IS_DEADDIR(inode)))
1673 : return ERR_PTR(-ENOENT);
1674 302489731 : again:
1675 302489755 : dentry = d_alloc_parallel(dir, name, &wq);
1676 302494458 : if (IS_ERR(dentry))
1677 0 : return dentry;
1678 302494458 : if (unlikely(!d_in_lookup(dentry))) {
1679 594590 : int error = d_revalidate(dentry, flags);
1680 594590 : if (unlikely(error <= 0)) {
1681 24 : if (!error) {
1682 24 : d_invalidate(dentry);
1683 24 : dput(dentry);
1684 24 : goto again;
1685 : }
1686 0 : dput(dentry);
1687 0 : dentry = ERR_PTR(error);
1688 : }
1689 : } else {
1690 301899868 : old = inode->i_op->lookup(inode, dentry, flags);
1691 301874233 : d_lookup_done(dentry);
1692 301895258 : if (unlikely(old)) {
1693 3825106 : dput(dentry);
1694 3825106 : dentry = old;
1695 : }
1696 : }
1697 : return dentry;
1698 : }
1699 :
1700 302275759 : static struct dentry *lookup_slow(const struct qstr *name,
1701 : struct dentry *dir,
1702 : unsigned int flags)
1703 : {
1704 302275759 : struct inode *inode = dir->d_inode;
1705 302275759 : struct dentry *res;
1706 302275759 : inode_lock_shared(inode);
1707 302301070 : res = __lookup_slow(name, dir, flags);
1708 302295902 : inode_unlock_shared(inode);
1709 302301774 : return res;
1710 : }
1711 :
1712 25508430903 : static inline int may_lookup(struct mnt_idmap *idmap,
1713 : struct nameidata *nd)
1714 : {
1715 25508430903 : if (nd->flags & LOOKUP_RCU) {
1716 25380515434 : int err = inode_permission(idmap, nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1717 25350938902 : if (err != -ECHILD || !try_to_unlazy(nd))
1718 25281263004 : return err;
1719 : }
1720 197602524 : return inode_permission(idmap, nd->inode, MAY_EXEC);
1721 : }
1722 :
1723 97509953 : static int reserve_stack(struct nameidata *nd, struct path *link)
1724 : {
1725 97509953 : if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1726 : return -ELOOP;
1727 :
1728 97508475 : if (likely(nd->depth != EMBEDDED_LEVELS))
1729 : return 0;
1730 0 : if (likely(nd->stack != nd->internal))
1731 : return 0;
1732 0 : if (likely(nd_alloc_stack(nd)))
1733 : return 0;
1734 :
1735 0 : if (nd->flags & LOOKUP_RCU) {
1736 : // we need to grab link before we do unlazy. And we can't skip
1737 : // unlazy even if we fail to grab the link - cleanup needs it
1738 0 : bool grabbed_link = legitimize_path(nd, link, nd->next_seq);
1739 :
1740 0 : if (!try_to_unlazy(nd) || !grabbed_link)
1741 : return -ECHILD;
1742 :
1743 0 : if (nd_alloc_stack(nd))
1744 0 : return 0;
1745 : }
1746 : return -ENOMEM;
1747 : }
1748 :
1749 : enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1750 :
1751 97510842 : static const char *pick_link(struct nameidata *nd, struct path *link,
1752 : struct inode *inode, int flags)
1753 : {
1754 97510842 : struct saved *last;
1755 97510842 : const char *res;
1756 97510842 : int error = reserve_stack(nd, link);
1757 :
1758 97510067 : if (unlikely(error)) {
1759 1478 : if (!(nd->flags & LOOKUP_RCU))
1760 1478 : path_put(link);
1761 1478 : return ERR_PTR(error);
1762 : }
1763 97508589 : last = nd->stack + nd->depth++;
1764 97508589 : last->link = *link;
1765 97508589 : clear_delayed_call(&last->done);
1766 97508589 : last->seq = nd->next_seq;
1767 :
1768 97508589 : if (flags & WALK_TRAILING) {
1769 38193588 : error = may_follow_link(nd, inode);
1770 38191032 : if (unlikely(error))
1771 4 : return ERR_PTR(error);
1772 : }
1773 :
1774 97506029 : if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) ||
1775 97506029 : unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW))
1776 : return ERR_PTR(-ELOOP);
1777 :
1778 97506029 : if (!(nd->flags & LOOKUP_RCU)) {
1779 2911286 : touch_atime(&last->link);
1780 2911111 : cond_resched();
1781 94594743 : } else if (atime_needs_update(&last->link, inode)) {
1782 156068 : if (!try_to_unlazy(nd))
1783 : return ERR_PTR(-ECHILD);
1784 156096 : touch_atime(&last->link);
1785 : }
1786 :
1787 97480576 : error = security_inode_follow_link(link->dentry, inode,
1788 97480576 : nd->flags & LOOKUP_RCU);
1789 97480576 : if (unlikely(error))
1790 : return ERR_PTR(error);
1791 :
1792 97480576 : res = READ_ONCE(inode->i_link);
1793 97480576 : if (!res) {
1794 96523365 : const char * (*get)(struct dentry *, struct inode *,
1795 : struct delayed_call *);
1796 96523365 : get = inode->i_op->get_link;
1797 96523365 : if (nd->flags & LOOKUP_RCU) {
1798 93595563 : res = get(NULL, inode, &last->done);
1799 93604276 : if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd))
1800 5754403 : res = get(link->dentry, inode, &last->done);
1801 : } else {
1802 2927802 : res = get(link->dentry, inode, &last->done);
1803 : }
1804 96529854 : if (!res)
1805 750929 : goto all_done;
1806 95778925 : if (IS_ERR(res))
1807 : return res;
1808 : }
1809 96729785 : if (*res == '/') {
1810 12836225 : error = nd_jump_root(nd);
1811 12836343 : if (unlikely(error))
1812 0 : return ERR_PTR(error);
1813 12836343 : while (unlikely(*++res == '/'))
1814 : ;
1815 : }
1816 96729903 : if (*res)
1817 : return res;
1818 0 : all_done: // pure jump
1819 750929 : put_link(nd);
1820 750929 : return NULL;
1821 : }
1822 :
1823 : /*
1824 : * Do we need to follow links? We _really_ want to be able
1825 : * to do this check without having to look at inode->i_op,
1826 : * so we keep a cache of "no, this doesn't need follow_link"
1827 : * for the common case.
1828 : *
1829 : * NOTE: dentry must be what nd->next_seq had been sampled from.
1830 : */
1831 24235701710 : static const char *step_into(struct nameidata *nd, int flags,
1832 : struct dentry *dentry)
1833 : {
1834 24235701710 : struct path path;
1835 24235701710 : struct inode *inode;
1836 24235701710 : int err = handle_mounts(nd, dentry, &path);
1837 :
1838 24231388354 : if (err < 0)
1839 277610409 : return ERR_PTR(err);
1840 23953777945 : inode = path.dentry->d_inode;
1841 23953777945 : if (likely(!d_is_symlink(path.dentry)) ||
1842 367462152 : ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
1843 97503909 : (flags & WALK_NOFOLLOW)) {
1844 : /* not a symlink or should not follow */
1845 23856274036 : if (nd->flags & LOOKUP_RCU) {
1846 23597301805 : if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1847 : return ERR_PTR(-ECHILD);
1848 23617860465 : if (unlikely(!inode))
1849 : return ERR_PTR(-ENOENT);
1850 : } else {
1851 258972231 : dput(nd->path.dentry);
1852 259060241 : if (nd->path.mnt != path.mnt)
1853 21255299 : mntput(nd->path.mnt);
1854 : }
1855 23708481568 : nd->path = path;
1856 23708481568 : nd->inode = inode;
1857 23708481568 : nd->seq = nd->next_seq;
1858 23708481568 : return NULL;
1859 : }
1860 97503909 : if (nd->flags & LOOKUP_RCU) {
1861 : /* make sure that d_is_symlink above matches inode */
1862 94591243 : if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq))
1863 : return ERR_PTR(-ECHILD);
1864 : } else {
1865 2912666 : if (path.mnt == nd->path.mnt)
1866 2912844 : mntget(path.mnt);
1867 : }
1868 97512186 : return pick_link(nd, &path, inode, flags);
1869 : }
1870 :
1871 133864939 : static struct dentry *follow_dotdot_rcu(struct nameidata *nd)
1872 : {
1873 133864939 : struct dentry *parent, *old;
1874 :
1875 133864939 : if (path_equal(&nd->path, &nd->root))
1876 0 : goto in_root;
1877 133864939 : if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1878 13 : struct path path;
1879 13 : unsigned seq;
1880 13 : if (!choose_mountpoint_rcu(real_mount(nd->path.mnt),
1881 13 : &nd->root, &path, &seq))
1882 0 : goto in_root;
1883 13 : if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1884 0 : return ERR_PTR(-ECHILD);
1885 13 : nd->path = path;
1886 13 : nd->inode = path.dentry->d_inode;
1887 13 : nd->seq = seq;
1888 : // makes sure that non-RCU pathwalk could reach this state
1889 13 : if (read_seqretry(&mount_lock, nd->m_seq))
1890 : return ERR_PTR(-ECHILD);
1891 : /* we know that mountpoint was pinned */
1892 : }
1893 133864939 : old = nd->path.dentry;
1894 133864939 : parent = old->d_parent;
1895 133864939 : nd->next_seq = read_seqcount_begin(&parent->d_seq);
1896 : // makes sure that non-RCU pathwalk could reach this state
1897 133878633 : if (read_seqcount_retry(&old->d_seq, nd->seq))
1898 : return ERR_PTR(-ECHILD);
1899 133878682 : if (unlikely(!path_connected(nd->path.mnt, parent)))
1900 0 : return ERR_PTR(-ECHILD);
1901 : return parent;
1902 0 : in_root:
1903 0 : if (read_seqretry(&mount_lock, nd->m_seq))
1904 : return ERR_PTR(-ECHILD);
1905 0 : if (unlikely(nd->flags & LOOKUP_BENEATH))
1906 : return ERR_PTR(-ECHILD);
1907 0 : nd->next_seq = nd->seq;
1908 0 : return nd->path.dentry;
1909 : }
1910 :
1911 5070945 : static struct dentry *follow_dotdot(struct nameidata *nd)
1912 : {
1913 5070945 : struct dentry *parent;
1914 :
1915 5070945 : if (path_equal(&nd->path, &nd->root))
1916 0 : goto in_root;
1917 5070945 : if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1918 328 : struct path path;
1919 :
1920 328 : if (!choose_mountpoint(real_mount(nd->path.mnt),
1921 328 : &nd->root, &path))
1922 0 : goto in_root;
1923 328 : path_put(&nd->path);
1924 328 : nd->path = path;
1925 328 : nd->inode = path.dentry->d_inode;
1926 328 : if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1927 0 : return ERR_PTR(-EXDEV);
1928 : }
1929 : /* rare case of legitimate dget_parent()... */
1930 5070945 : parent = dget_parent(nd->path.dentry);
1931 5071790 : if (unlikely(!path_connected(nd->path.mnt, parent))) {
1932 0 : dput(parent);
1933 0 : return ERR_PTR(-ENOENT);
1934 : }
1935 : return parent;
1936 :
1937 0 : in_root:
1938 0 : if (unlikely(nd->flags & LOOKUP_BENEATH))
1939 : return ERR_PTR(-EXDEV);
1940 0 : return dget(nd->path.dentry);
1941 : }
1942 :
1943 1476096124 : static const char *handle_dots(struct nameidata *nd, int type)
1944 : {
1945 1476096124 : if (type == LAST_DOTDOT) {
1946 138947953 : const char *error = NULL;
1947 138947953 : struct dentry *parent;
1948 :
1949 138947953 : if (!nd->root.mnt) {
1950 130935417 : error = ERR_PTR(set_root(nd));
1951 130937072 : if (error)
1952 : return error;
1953 : }
1954 138949608 : if (nd->flags & LOOKUP_RCU)
1955 133878675 : parent = follow_dotdot_rcu(nd);
1956 : else
1957 5070933 : parent = follow_dotdot(nd);
1958 138950345 : if (IS_ERR(parent))
1959 : return ERR_CAST(parent);
1960 138950345 : error = step_into(nd, WALK_NOFOLLOW, parent);
1961 138949639 : if (unlikely(error))
1962 : return error;
1963 :
1964 138949639 : if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1965 : /*
1966 : * If there was a racing rename or mount along our
1967 : * path, then we can't be sure that ".." hasn't jumped
1968 : * above nd->root (and so userspace should retry or use
1969 : * some fallback).
1970 : */
1971 0 : smp_rmb();
1972 0 : if (__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))
1973 : return ERR_PTR(-EAGAIN);
1974 0 : if (__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))
1975 0 : return ERR_PTR(-EAGAIN);
1976 : }
1977 : }
1978 : return NULL;
1979 : }
1980 :
1981 25128157824 : static const char *walk_component(struct nameidata *nd, int flags)
1982 : {
1983 25128157824 : struct dentry *dentry;
1984 : /*
1985 : * "." and ".." are special - ".." especially so because it has
1986 : * to be able to know about the current root directory and
1987 : * parent relationships.
1988 : */
1989 25128157824 : if (unlikely(nd->last_type != LAST_NORM)) {
1990 1468367854 : if (!(flags & WALK_MORE) && nd->depth)
1991 0 : put_link(nd);
1992 1468367854 : return handle_dots(nd, nd->last_type);
1993 : }
1994 23659789970 : dentry = lookup_fast(nd);
1995 23719140573 : if (IS_ERR(dentry))
1996 : return ERR_CAST(dentry);
1997 23719140241 : if (unlikely(!dentry)) {
1998 298530410 : dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
1999 298538629 : if (IS_ERR(dentry))
2000 : return ERR_CAST(dentry);
2001 : }
2002 23719073997 : if (!(flags & WALK_MORE) && nd->depth)
2003 63406447 : put_link(nd);
2004 23719074431 : return step_into(nd, flags, dentry);
2005 : }
2006 :
2007 : /*
2008 : * We can do the critical dentry name comparison and hashing
2009 : * operations one word at a time, but we are limited to:
2010 : *
2011 : * - Architectures with fast unaligned word accesses. We could
2012 : * do a "get_unaligned()" if this helps and is sufficiently
2013 : * fast.
2014 : *
2015 : * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
2016 : * do not trap on the (extremely unlikely) case of a page
2017 : * crossing operation.
2018 : *
2019 : * - Furthermore, we need an efficient 64-bit compile for the
2020 : * 64-bit case in order to generate the "number of bytes in
2021 : * the final mask". Again, that could be replaced with a
2022 : * efficient population count instruction or similar.
2023 : */
2024 : #ifdef CONFIG_DCACHE_WORD_ACCESS
2025 :
2026 : #include <asm/word-at-a-time.h>
2027 :
2028 : #ifdef HASH_MIX
2029 :
2030 : /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
2031 :
2032 : #elif defined(CONFIG_64BIT)
2033 : /*
2034 : * Register pressure in the mixing function is an issue, particularly
2035 : * on 32-bit x86, but almost any function requires one state value and
2036 : * one temporary. Instead, use a function designed for two state values
2037 : * and no temporaries.
2038 : *
2039 : * This function cannot create a collision in only two iterations, so
2040 : * we have two iterations to achieve avalanche. In those two iterations,
2041 : * we have six layers of mixing, which is enough to spread one bit's
2042 : * influence out to 2^6 = 64 state bits.
2043 : *
2044 : * Rotate constants are scored by considering either 64 one-bit input
2045 : * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
2046 : * probability of that delta causing a change to each of the 128 output
2047 : * bits, using a sample of random initial states.
2048 : *
2049 : * The Shannon entropy of the computed probabilities is then summed
2050 : * to produce a score. Ideally, any input change has a 50% chance of
2051 : * toggling any given output bit.
2052 : *
2053 : * Mixing scores (in bits) for (12,45):
2054 : * Input delta: 1-bit 2-bit
2055 : * 1 round: 713.3 42542.6
2056 : * 2 rounds: 2753.7 140389.8
2057 : * 3 rounds: 5954.1 233458.2
2058 : * 4 rounds: 7862.6 256672.2
2059 : * Perfect: 8192 258048
2060 : * (64*128) (64*63/2 * 128)
2061 : */
2062 : #define HASH_MIX(x, y, a) \
2063 : ( x ^= (a), \
2064 : y ^= x, x = rol64(x,12),\
2065 : x += y, y = rol64(y,45),\
2066 : y *= 9 )
2067 :
2068 : /*
2069 : * Fold two longs into one 32-bit hash value. This must be fast, but
2070 : * latency isn't quite as critical, as there is a fair bit of additional
2071 : * work done before the hash value is used.
2072 : */
2073 : static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2074 : {
2075 25520394001 : y ^= x * GOLDEN_RATIO_64;
2076 25520394001 : y *= GOLDEN_RATIO_64;
2077 25520394001 : return y >> 32;
2078 : }
2079 :
2080 : #else /* 32-bit case */
2081 :
2082 : /*
2083 : * Mixing scores (in bits) for (7,20):
2084 : * Input delta: 1-bit 2-bit
2085 : * 1 round: 330.3 9201.6
2086 : * 2 rounds: 1246.4 25475.4
2087 : * 3 rounds: 1907.1 31295.1
2088 : * 4 rounds: 2042.3 31718.6
2089 : * Perfect: 2048 31744
2090 : * (32*64) (32*31/2 * 64)
2091 : */
2092 : #define HASH_MIX(x, y, a) \
2093 : ( x ^= (a), \
2094 : y ^= x, x = rol32(x, 7),\
2095 : x += y, y = rol32(y,20),\
2096 : y *= 9 )
2097 :
2098 : static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2099 : {
2100 : /* Use arch-optimized multiply if one exists */
2101 : return __hash_32(y ^ __hash_32(x));
2102 : }
2103 :
2104 : #endif
2105 :
2106 : /*
2107 : * Return the hash of a string of known length. This is carfully
2108 : * designed to match hash_name(), which is the more critical function.
2109 : * In particular, we must end by hashing a final word containing 0..7
2110 : * payload bytes, to match the way that hash_name() iterates until it
2111 : * finds the delimiter after the name.
2112 : */
2113 27650643 : unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2114 : {
2115 27650643 : unsigned long a, x = 0, y = (unsigned long)salt;
2116 :
2117 29049883 : for (;;) {
2118 28350263 : if (!len)
2119 102034 : goto done;
2120 28248229 : a = load_unaligned_zeropad(name);
2121 28248229 : if (len < sizeof(unsigned long))
2122 : break;
2123 699620 : HASH_MIX(x, y, a);
2124 699620 : name += sizeof(unsigned long);
2125 699620 : len -= sizeof(unsigned long);
2126 : }
2127 27548609 : x ^= a & bytemask_from_count(len);
2128 27650643 : done:
2129 27650643 : return fold_hash(x, y);
2130 : }
2131 : EXPORT_SYMBOL(full_name_hash);
2132 :
2133 : /* Return the "hash_len" (hash and length) of a null-terminated string */
2134 8 : u64 hashlen_string(const void *salt, const char *name)
2135 : {
2136 8 : unsigned long a = 0, x = 0, y = (unsigned long)salt;
2137 8 : unsigned long adata, mask, len;
2138 8 : const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2139 :
2140 8 : len = 0;
2141 8 : goto inside;
2142 :
2143 4 : do {
2144 4 : HASH_MIX(x, y, a);
2145 4 : len += sizeof(unsigned long);
2146 12 : inside:
2147 12 : a = load_unaligned_zeropad(name+len);
2148 12 : } while (!has_zero(a, &adata, &constants));
2149 :
2150 8 : adata = prep_zero_mask(a, adata, &constants);
2151 8 : mask = create_zero_mask(adata);
2152 8 : x ^= a & zero_bytemask(mask);
2153 :
2154 8 : return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2155 : }
2156 : EXPORT_SYMBOL(hashlen_string);
2157 :
2158 : /*
2159 : * Calculate the length and hash of the path component, and
2160 : * return the "hash_len" as the result.
2161 : */
2162 25492743350 : static inline u64 hash_name(const void *salt, const char *name)
2163 : {
2164 25492743350 : unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2165 25492743350 : unsigned long adata, bdata, mask, len;
2166 25492743350 : const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2167 :
2168 25492743350 : len = 0;
2169 25492743350 : goto inside;
2170 :
2171 646025081 : do {
2172 646025081 : HASH_MIX(x, y, a);
2173 646025081 : len += sizeof(unsigned long);
2174 26138768431 : inside:
2175 26138768431 : a = load_unaligned_zeropad(name+len);
2176 26138768431 : b = a ^ REPEAT_BYTE('/');
2177 26138768431 : } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2178 :
2179 25492743350 : adata = prep_zero_mask(a, adata, &constants);
2180 25492743350 : bdata = prep_zero_mask(b, bdata, &constants);
2181 25492743350 : mask = create_zero_mask(adata | bdata);
2182 25492743350 : x ^= a & zero_bytemask(mask);
2183 :
2184 25492743350 : return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2185 : }
2186 :
2187 : #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2188 :
2189 : /* Return the hash of a string of known length */
2190 : unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2191 : {
2192 : unsigned long hash = init_name_hash(salt);
2193 : while (len--)
2194 : hash = partial_name_hash((unsigned char)*name++, hash);
2195 : return end_name_hash(hash);
2196 : }
2197 : EXPORT_SYMBOL(full_name_hash);
2198 :
2199 : /* Return the "hash_len" (hash and length) of a null-terminated string */
2200 : u64 hashlen_string(const void *salt, const char *name)
2201 : {
2202 : unsigned long hash = init_name_hash(salt);
2203 : unsigned long len = 0, c;
2204 :
2205 : c = (unsigned char)*name;
2206 : while (c) {
2207 : len++;
2208 : hash = partial_name_hash(c, hash);
2209 : c = (unsigned char)name[len];
2210 : }
2211 : return hashlen_create(end_name_hash(hash), len);
2212 : }
2213 : EXPORT_SYMBOL(hashlen_string);
2214 :
2215 : /*
2216 : * We know there's a real path component here of at least
2217 : * one character.
2218 : */
2219 : static inline u64 hash_name(const void *salt, const char *name)
2220 : {
2221 : unsigned long hash = init_name_hash(salt);
2222 : unsigned long len = 0, c;
2223 :
2224 : c = (unsigned char)*name;
2225 : do {
2226 : len++;
2227 : hash = partial_name_hash(c, hash);
2228 : c = (unsigned char)name[len];
2229 : } while (c && c != '/');
2230 : return hashlen_create(end_name_hash(hash), len);
2231 : }
2232 :
2233 : #endif
2234 :
2235 : /*
2236 : * Name resolution.
2237 : * This is the basic name resolution function, turning a pathname into
2238 : * the final dentry. We expect 'base' to be positive and a directory.
2239 : *
2240 : * Returns 0 and nd will have valid dentry and mnt on success.
2241 : * Returns error and drops reference to input namei data on failure.
2242 : */
2243 4323302342 : static int link_path_walk(const char *name, struct nameidata *nd)
2244 : {
2245 4323302342 : int depth = 0; // depth <= nd->depth
2246 4323302342 : int err;
2247 :
2248 4323302342 : nd->last_type = LAST_ROOT;
2249 4323302342 : nd->flags |= LOOKUP_PARENT;
2250 4323302342 : if (IS_ERR(name))
2251 350954279 : return PTR_ERR(name);
2252 5232930615 : while (*name=='/')
2253 1260582552 : name++;
2254 3972348063 : if (!*name) {
2255 413530230 : nd->dir_mode = 0; // short-circuit the 'hardening' idiocy
2256 413530230 : return 0;
2257 : }
2258 :
2259 : /* At this point we know we have a real path component. */
2260 25539642815 : for(;;) {
2261 25539642815 : struct mnt_idmap *idmap;
2262 25539642815 : const char *link;
2263 25539642815 : u64 hash_len;
2264 25539642815 : int type;
2265 :
2266 25539642815 : idmap = mnt_idmap(nd->path.mnt);
2267 25505214717 : err = may_lookup(idmap, nd);
2268 25461053246 : if (err)
2269 662 : return err;
2270 :
2271 25461052584 : hash_len = hash_name(nd->path.dentry, name);
2272 :
2273 25422233209 : type = LAST_NORM;
2274 25422233209 : if (name[0] == '.') switch (hashlen_len(hash_len)) {
2275 138949637 : case 2:
2276 138949637 : if (name[1] == '.') {
2277 138949923 : type = LAST_DOTDOT;
2278 138949923 : nd->state |= ND_JUMPED;
2279 : }
2280 : break;
2281 923647950 : case 1:
2282 923647950 : type = LAST_DOT;
2283 : }
2284 25422233209 : if (likely(type == LAST_NORM)) {
2285 24359506805 : struct dentry *parent = nd->path.dentry;
2286 24359506805 : nd->state &= ~ND_JUMPED;
2287 24359506805 : if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2288 2 : struct qstr this = { { .hash_len = hash_len }, .name = name };
2289 2 : err = parent->d_op->d_hash(parent, &this);
2290 2 : if (err < 0)
2291 0 : return err;
2292 2 : hash_len = this.hash_len;
2293 2 : name = this.name;
2294 : }
2295 : }
2296 :
2297 25422233209 : nd->last.hash_len = hash_len;
2298 25422233209 : nd->last.name = name;
2299 25422233209 : nd->last_type = type;
2300 :
2301 25422233209 : name += hashlen_len(hash_len);
2302 25422233209 : if (!*name)
2303 3519975933 : goto OK;
2304 : /*
2305 : * If it wasn't NUL, we know it was '/'. Skip that
2306 : * slash, and continue until no more slashes.
2307 : */
2308 21903233429 : do {
2309 21903233429 : name++;
2310 21903233429 : } while (unlikely(*name == '/'));
2311 21902257276 : if (unlikely(!*name)) {
2312 133060 : OK:
2313 : /* pathname or trailing symlink, done */
2314 3520108993 : if (!depth) {
2315 3461209670 : nd->dir_vfsuid = i_uid_into_vfsuid(idmap, nd->inode);
2316 3462168106 : nd->dir_mode = nd->inode->i_mode;
2317 3462168106 : nd->flags &= ~LOOKUP_PARENT;
2318 3462168106 : return 0;
2319 : }
2320 : /* last component of nested symlink */
2321 58899323 : name = nd->stack[--depth].name;
2322 58899323 : link = walk_component(nd, 0);
2323 : } else {
2324 : /* not the last component */
2325 21902124216 : link = walk_component(nd, WALK_MORE);
2326 : }
2327 22076444748 : if (unlikely(link)) {
2328 154925322 : if (IS_ERR(link))
2329 95619666 : return PTR_ERR(link);
2330 : /* a symlink to follow */
2331 59305656 : nd->stack[depth++].name = name;
2332 59305656 : name = link;
2333 59305656 : continue;
2334 : }
2335 21921519426 : if (unlikely(!d_can_lookup(nd->path.dentry))) {
2336 100 : if (nd->flags & LOOKUP_RCU) {
2337 100 : if (!try_to_unlazy(nd))
2338 : return -ECHILD;
2339 : }
2340 100 : return -ENOTDIR;
2341 : }
2342 : }
2343 : }
2344 :
2345 : /* must be paired with terminate_walk() */
2346 3932341199 : static const char *path_init(struct nameidata *nd, unsigned flags)
2347 : {
2348 3932341199 : int error;
2349 3932341199 : const char *s = nd->name->name;
2350 :
2351 : /* LOOKUP_CACHED requires RCU, ask caller to retry */
2352 3932341199 : if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED)
2353 : return ERR_PTR(-EAGAIN);
2354 :
2355 3932341199 : if (!*s)
2356 409918038 : flags &= ~LOOKUP_RCU;
2357 3932341199 : if (flags & LOOKUP_RCU)
2358 3522401988 : rcu_read_lock();
2359 : else
2360 409939211 : nd->seq = nd->next_seq = 0;
2361 :
2362 3932463608 : nd->flags = flags;
2363 3932463608 : nd->state |= ND_JUMPED;
2364 :
2365 3942556312 : nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2366 4390578345 : nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2367 3933793088 : smp_rmb();
2368 :
2369 3935091458 : if (nd->state & ND_ROOT_PRESET) {
2370 6376 : struct dentry *root = nd->root.dentry;
2371 6376 : struct inode *inode = root->d_inode;
2372 6376 : if (*s && unlikely(!d_can_lookup(root)))
2373 : return ERR_PTR(-ENOTDIR);
2374 6376 : nd->path = nd->root;
2375 6376 : nd->inode = inode;
2376 6376 : if (flags & LOOKUP_RCU) {
2377 0 : nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2378 0 : nd->root_seq = nd->seq;
2379 : } else {
2380 6376 : path_get(&nd->path);
2381 : }
2382 6376 : return s;
2383 : }
2384 :
2385 3935085082 : nd->root.mnt = NULL;
2386 :
2387 : /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2388 3935085082 : if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2389 1260574668 : error = nd_jump_root(nd);
2390 1260672215 : if (unlikely(error))
2391 0 : return ERR_PTR(error);
2392 : return s;
2393 : }
2394 :
2395 : /* Relative pathname -- get the starting-point it is relative to. */
2396 2674510414 : if (nd->dfd == AT_FDCWD) {
2397 2236228522 : if (flags & LOOKUP_RCU) {
2398 2236227913 : struct fs_struct *fs = current->fs;
2399 : unsigned seq;
2400 :
2401 : do {
2402 2236227913 : seq = read_seqcount_begin(&fs->seq);
2403 2236365909 : nd->path = fs->pwd;
2404 2236365909 : nd->inode = nd->path.dentry->d_inode;
2405 2236366202 : nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2406 2236450202 : } while (read_seqcount_retry(&fs->seq, seq));
2407 : } else {
2408 609 : get_fs_pwd(current->fs, &nd->path);
2409 609 : nd->inode = nd->path.dentry->d_inode;
2410 : }
2411 : } else {
2412 : /* Caller must check execute permissions on the starting path component */
2413 438281892 : struct fd f = fdget_raw(nd->dfd);
2414 438295044 : struct dentry *dentry;
2415 :
2416 438295044 : if (!f.file)
2417 0 : return ERR_PTR(-EBADF);
2418 :
2419 438295044 : dentry = f.file->f_path.dentry;
2420 :
2421 438295044 : if (*s && unlikely(!d_can_lookup(dentry))) {
2422 0 : fdput(f);
2423 0 : return ERR_PTR(-ENOTDIR);
2424 : }
2425 :
2426 438295044 : nd->path = f.file->f_path;
2427 438295044 : if (flags & LOOKUP_RCU) {
2428 28382441 : nd->inode = nd->path.dentry->d_inode;
2429 28382441 : nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2430 : } else {
2431 409912603 : path_get(&nd->path);
2432 409912384 : nd->inode = nd->path.dentry->d_inode;
2433 : }
2434 438298570 : fdput(f);
2435 : }
2436 :
2437 : /* For scoped-lookups we need to set the root to the dirfd as well. */
2438 2674960242 : if (flags & LOOKUP_IS_SCOPED) {
2439 0 : nd->root = nd->path;
2440 0 : if (flags & LOOKUP_RCU) {
2441 0 : nd->root_seq = nd->seq;
2442 : } else {
2443 0 : path_get(&nd->root);
2444 0 : nd->state |= ND_ROOT_GRABBED;
2445 : }
2446 : }
2447 : return s;
2448 : }
2449 :
2450 3118424880 : static inline const char *lookup_last(struct nameidata *nd)
2451 : {
2452 3118424880 : if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2453 59142 : nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2454 :
2455 3118424880 : return walk_component(nd, WALK_TRAILING);
2456 : }
2457 :
2458 54114 : static int handle_lookup_down(struct nameidata *nd)
2459 : {
2460 54114 : if (!(nd->flags & LOOKUP_RCU))
2461 30 : dget(nd->path.dentry);
2462 54114 : nd->next_seq = nd->seq;
2463 54114 : return PTR_ERR(step_into(nd, WALK_NOFOLLOW, nd->path.dentry));
2464 : }
2465 :
2466 : /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2467 3171746519 : static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2468 : {
2469 3171746519 : const char *s = path_init(nd, flags);
2470 3174051327 : int err;
2471 :
2472 3174051327 : if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2473 0 : err = handle_lookup_down(nd);
2474 0 : if (unlikely(err < 0))
2475 0 : s = ERR_PTR(err);
2476 : }
2477 :
2478 6291776920 : while (!(err = link_path_walk(s, nd)) &&
2479 3117469558 : (s = lookup_last(nd)) != NULL)
2480 : ;
2481 3173365243 : if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2482 54114 : err = handle_lookup_down(nd);
2483 54114 : nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please...
2484 : }
2485 3173365243 : if (!err)
2486 2767795843 : err = complete_walk(nd);
2487 :
2488 3173994332 : if (!err && nd->flags & LOOKUP_DIRECTORY)
2489 260677183 : if (!d_can_lookup(nd->path.dentry))
2490 : err = -ENOTDIR;
2491 3173994284 : if (!err) {
2492 2768273479 : *path = nd->path;
2493 2768273479 : nd->path.mnt = NULL;
2494 2768273479 : nd->path.dentry = NULL;
2495 : }
2496 3173994332 : terminate_walk(nd);
2497 3173106033 : return err;
2498 : }
2499 :
2500 3235646495 : int filename_lookup(int dfd, struct filename *name, unsigned flags,
2501 : struct path *path, struct path *root)
2502 : {
2503 3235646495 : int retval;
2504 3235646495 : struct nameidata nd;
2505 3235646495 : if (IS_ERR(name))
2506 81938608 : return PTR_ERR(name);
2507 3153707887 : set_nameidata(&nd, dfd, name, root);
2508 3153215074 : retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2509 3153636084 : if (unlikely(retval == -ECHILD))
2510 11786 : retval = path_lookupat(&nd, flags, path);
2511 3153636084 : if (unlikely(retval == -ESTALE))
2512 0 : retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2513 :
2514 3153636084 : if (likely(!retval))
2515 : audit_inode(name, path->dentry,
2516 : flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2517 3153636084 : restore_nameidata();
2518 3153636084 : return retval;
2519 : }
2520 :
2521 : /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2522 333243203 : static int path_parentat(struct nameidata *nd, unsigned flags,
2523 : struct path *parent)
2524 : {
2525 333243203 : const char *s = path_init(nd, flags);
2526 333281614 : int err = link_path_walk(s, nd);
2527 333268395 : if (!err)
2528 333265479 : err = complete_walk(nd);
2529 333300770 : if (!err) {
2530 333297380 : *parent = nd->path;
2531 333297380 : nd->path.mnt = NULL;
2532 333297380 : nd->path.dentry = NULL;
2533 : }
2534 333300770 : terminate_walk(nd);
2535 333299556 : return err;
2536 : }
2537 :
2538 : /* Note: this does not consume "name" */
2539 333165647 : static int __filename_parentat(int dfd, struct filename *name,
2540 : unsigned int flags, struct path *parent,
2541 : struct qstr *last, int *type,
2542 : const struct path *root)
2543 : {
2544 333165647 : int retval;
2545 333165647 : struct nameidata nd;
2546 :
2547 333165647 : if (IS_ERR(name))
2548 206 : return PTR_ERR(name);
2549 333165441 : set_nameidata(&nd, dfd, name, root);
2550 333197235 : retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2551 333298852 : if (unlikely(retval == -ECHILD))
2552 220 : retval = path_parentat(&nd, flags, parent);
2553 333298852 : if (unlikely(retval == -ESTALE))
2554 0 : retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2555 333298852 : if (likely(!retval)) {
2556 333289373 : *last = nd.last;
2557 333289373 : *type = nd.last_type;
2558 333289373 : audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2559 : }
2560 333298852 : restore_nameidata();
2561 333298852 : return retval;
2562 : }
2563 :
2564 : static int filename_parentat(int dfd, struct filename *name,
2565 : unsigned int flags, struct path *parent,
2566 : struct qstr *last, int *type)
2567 : {
2568 333237292 : return __filename_parentat(dfd, name, flags, parent, last, type, NULL);
2569 : }
2570 :
2571 : /* does lookup, returns the object with parent locked */
2572 319 : static struct dentry *__kern_path_locked(struct filename *name, struct path *path)
2573 : {
2574 319 : struct dentry *d;
2575 319 : struct qstr last;
2576 319 : int type, error;
2577 :
2578 319 : error = filename_parentat(AT_FDCWD, name, 0, path, &last, &type);
2579 319 : if (error)
2580 0 : return ERR_PTR(error);
2581 319 : if (unlikely(type != LAST_NORM)) {
2582 0 : path_put(path);
2583 0 : return ERR_PTR(-EINVAL);
2584 : }
2585 319 : inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2586 319 : d = lookup_one_qstr_excl(&last, path->dentry, 0);
2587 319 : if (IS_ERR(d)) {
2588 0 : inode_unlock(path->dentry->d_inode);
2589 0 : path_put(path);
2590 : }
2591 : return d;
2592 : }
2593 :
2594 319 : struct dentry *kern_path_locked(const char *name, struct path *path)
2595 : {
2596 319 : struct filename *filename = getname_kernel(name);
2597 319 : struct dentry *res = __kern_path_locked(filename, path);
2598 :
2599 319 : putname(filename);
2600 319 : return res;
2601 : }
2602 :
2603 6971139 : int kern_path(const char *name, unsigned int flags, struct path *path)
2604 : {
2605 6971139 : struct filename *filename = getname_kernel(name);
2606 6971185 : int ret = filename_lookup(AT_FDCWD, filename, flags, path, NULL);
2607 :
2608 6971347 : putname(filename);
2609 6971310 : return ret;
2610 :
2611 : }
2612 : EXPORT_SYMBOL(kern_path);
2613 :
2614 : /**
2615 : * vfs_path_parent_lookup - lookup a parent path relative to a dentry-vfsmount pair
2616 : * @filename: filename structure
2617 : * @flags: lookup flags
2618 : * @parent: pointer to struct path to fill
2619 : * @last: last component
2620 : * @type: type of the last component
2621 : * @root: pointer to struct path of the base directory
2622 : */
2623 0 : int vfs_path_parent_lookup(struct filename *filename, unsigned int flags,
2624 : struct path *parent, struct qstr *last, int *type,
2625 : const struct path *root)
2626 : {
2627 0 : return __filename_parentat(AT_FDCWD, filename, flags, parent, last,
2628 : type, root);
2629 : }
2630 : EXPORT_SYMBOL(vfs_path_parent_lookup);
2631 :
2632 : /**
2633 : * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2634 : * @dentry: pointer to dentry of the base directory
2635 : * @mnt: pointer to vfs mount of the base directory
2636 : * @name: pointer to file name
2637 : * @flags: lookup flags
2638 : * @path: pointer to struct path to fill
2639 : */
2640 0 : int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2641 : const char *name, unsigned int flags,
2642 : struct path *path)
2643 : {
2644 0 : struct filename *filename;
2645 0 : struct path root = {.mnt = mnt, .dentry = dentry};
2646 0 : int ret;
2647 :
2648 0 : filename = getname_kernel(name);
2649 : /* the first argument of filename_lookup() is ignored with root */
2650 0 : ret = filename_lookup(AT_FDCWD, filename, flags, path, &root);
2651 0 : putname(filename);
2652 0 : return ret;
2653 : }
2654 : EXPORT_SYMBOL(vfs_path_lookup);
2655 :
2656 4044095 : static int lookup_one_common(struct mnt_idmap *idmap,
2657 : const char *name, struct dentry *base, int len,
2658 : struct qstr *this)
2659 : {
2660 4044095 : this->name = name;
2661 4044095 : this->len = len;
2662 4044095 : this->hash = full_name_hash(base, name, len);
2663 4044095 : if (!len)
2664 : return -EACCES;
2665 :
2666 4044095 : if (unlikely(name[0] == '.')) {
2667 1062 : if (len < 2 || (len == 2 && name[1] == '.'))
2668 : return -EACCES;
2669 : }
2670 :
2671 22806811 : while (len--) {
2672 18762716 : unsigned int c = *(const unsigned char *)name++;
2673 18762716 : if (c == '/' || c == '\0')
2674 : return -EACCES;
2675 : }
2676 : /*
2677 : * See if the low-level filesystem might want
2678 : * to use its own hash..
2679 : */
2680 4044095 : if (base->d_flags & DCACHE_OP_HASH) {
2681 0 : int err = base->d_op->d_hash(base, this);
2682 0 : if (err < 0)
2683 : return err;
2684 : }
2685 :
2686 4044095 : return inode_permission(idmap, base->d_inode, MAY_EXEC);
2687 : }
2688 :
2689 : /**
2690 : * try_lookup_one_len - filesystem helper to lookup single pathname component
2691 : * @name: pathname component to lookup
2692 : * @base: base directory to lookup from
2693 : * @len: maximum length @len should be interpreted to
2694 : *
2695 : * Look up a dentry by name in the dcache, returning NULL if it does not
2696 : * currently exist. The function does not try to create a dentry.
2697 : *
2698 : * Note that this routine is purely a helper for filesystem usage and should
2699 : * not be called by generic code.
2700 : *
2701 : * The caller must hold base->i_mutex.
2702 : */
2703 0 : struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2704 : {
2705 0 : struct qstr this;
2706 0 : int err;
2707 :
2708 0 : WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2709 :
2710 0 : err = lookup_one_common(&nop_mnt_idmap, name, base, len, &this);
2711 0 : if (err)
2712 0 : return ERR_PTR(err);
2713 :
2714 0 : return lookup_dcache(&this, base, 0);
2715 : }
2716 : EXPORT_SYMBOL(try_lookup_one_len);
2717 :
2718 : /**
2719 : * lookup_one_len - filesystem helper to lookup single pathname component
2720 : * @name: pathname component to lookup
2721 : * @base: base directory to lookup from
2722 : * @len: maximum length @len should be interpreted to
2723 : *
2724 : * Note that this routine is purely a helper for filesystem usage and should
2725 : * not be called by generic code.
2726 : *
2727 : * The caller must hold base->i_mutex.
2728 : */
2729 108475 : struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2730 : {
2731 108475 : struct dentry *dentry;
2732 108475 : struct qstr this;
2733 108475 : int err;
2734 :
2735 108475 : WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2736 :
2737 108475 : err = lookup_one_common(&nop_mnt_idmap, name, base, len, &this);
2738 108475 : if (err)
2739 0 : return ERR_PTR(err);
2740 :
2741 108475 : dentry = lookup_dcache(&this, base, 0);
2742 108475 : return dentry ? dentry : __lookup_slow(&this, base, 0);
2743 : }
2744 : EXPORT_SYMBOL(lookup_one_len);
2745 :
2746 : /**
2747 : * lookup_one - filesystem helper to lookup single pathname component
2748 : * @idmap: idmap of the mount the lookup is performed from
2749 : * @name: pathname component to lookup
2750 : * @base: base directory to lookup from
2751 : * @len: maximum length @len should be interpreted to
2752 : *
2753 : * Note that this routine is purely a helper for filesystem usage and should
2754 : * not be called by generic code.
2755 : *
2756 : * The caller must hold base->i_mutex.
2757 : */
2758 124379 : struct dentry *lookup_one(struct mnt_idmap *idmap, const char *name,
2759 : struct dentry *base, int len)
2760 : {
2761 124379 : struct dentry *dentry;
2762 124379 : struct qstr this;
2763 124379 : int err;
2764 :
2765 124379 : WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2766 :
2767 124379 : err = lookup_one_common(idmap, name, base, len, &this);
2768 124377 : if (err)
2769 0 : return ERR_PTR(err);
2770 :
2771 124377 : dentry = lookup_dcache(&this, base, 0);
2772 124383 : return dentry ? dentry : __lookup_slow(&this, base, 0);
2773 : }
2774 : EXPORT_SYMBOL(lookup_one);
2775 :
2776 : /**
2777 : * lookup_one_unlocked - filesystem helper to lookup single pathname component
2778 : * @idmap: idmap of the mount the lookup is performed from
2779 : * @name: pathname component to lookup
2780 : * @base: base directory to lookup from
2781 : * @len: maximum length @len should be interpreted to
2782 : *
2783 : * Note that this routine is purely a helper for filesystem usage and should
2784 : * not be called by generic code.
2785 : *
2786 : * Unlike lookup_one_len, it should be called without the parent
2787 : * i_mutex held, and will take the i_mutex itself if necessary.
2788 : */
2789 3811152 : struct dentry *lookup_one_unlocked(struct mnt_idmap *idmap,
2790 : const char *name, struct dentry *base,
2791 : int len)
2792 : {
2793 3811152 : struct qstr this;
2794 3811152 : int err;
2795 3811152 : struct dentry *ret;
2796 :
2797 3811152 : err = lookup_one_common(idmap, name, base, len, &this);
2798 3810951 : if (err)
2799 0 : return ERR_PTR(err);
2800 :
2801 3810951 : ret = lookup_dcache(&this, base, 0);
2802 3811211 : if (!ret)
2803 3760763 : ret = lookup_slow(&this, base, 0);
2804 : return ret;
2805 : }
2806 : EXPORT_SYMBOL(lookup_one_unlocked);
2807 :
2808 : /**
2809 : * lookup_one_positive_unlocked - filesystem helper to lookup single
2810 : * pathname component
2811 : * @idmap: idmap of the mount the lookup is performed from
2812 : * @name: pathname component to lookup
2813 : * @base: base directory to lookup from
2814 : * @len: maximum length @len should be interpreted to
2815 : *
2816 : * This helper will yield ERR_PTR(-ENOENT) on negatives. The helper returns
2817 : * known positive or ERR_PTR(). This is what most of the users want.
2818 : *
2819 : * Note that pinned negative with unlocked parent _can_ become positive at any
2820 : * time, so callers of lookup_one_unlocked() need to be very careful; pinned
2821 : * positives have >d_inode stable, so this one avoids such problems.
2822 : *
2823 : * Note that this routine is purely a helper for filesystem usage and should
2824 : * not be called by generic code.
2825 : *
2826 : * The helper should be called without i_mutex held.
2827 : */
2828 16233 : struct dentry *lookup_one_positive_unlocked(struct mnt_idmap *idmap,
2829 : const char *name,
2830 : struct dentry *base, int len)
2831 : {
2832 16233 : struct dentry *ret = lookup_one_unlocked(idmap, name, base, len);
2833 :
2834 16233 : if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2835 8029 : dput(ret);
2836 8029 : ret = ERR_PTR(-ENOENT);
2837 : }
2838 16233 : return ret;
2839 : }
2840 : EXPORT_SYMBOL(lookup_one_positive_unlocked);
2841 :
2842 : /**
2843 : * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2844 : * @name: pathname component to lookup
2845 : * @base: base directory to lookup from
2846 : * @len: maximum length @len should be interpreted to
2847 : *
2848 : * Note that this routine is purely a helper for filesystem usage and should
2849 : * not be called by generic code.
2850 : *
2851 : * Unlike lookup_one_len, it should be called without the parent
2852 : * i_mutex held, and will take the i_mutex itself if necessary.
2853 : */
2854 0 : struct dentry *lookup_one_len_unlocked(const char *name,
2855 : struct dentry *base, int len)
2856 : {
2857 0 : return lookup_one_unlocked(&nop_mnt_idmap, name, base, len);
2858 : }
2859 : EXPORT_SYMBOL(lookup_one_len_unlocked);
2860 :
2861 : /*
2862 : * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2863 : * on negatives. Returns known positive or ERR_PTR(); that's what
2864 : * most of the users want. Note that pinned negative with unlocked parent
2865 : * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2866 : * need to be very careful; pinned positives have ->d_inode stable, so
2867 : * this one avoids such problems.
2868 : */
2869 676 : struct dentry *lookup_positive_unlocked(const char *name,
2870 : struct dentry *base, int len)
2871 : {
2872 676 : return lookup_one_positive_unlocked(&nop_mnt_idmap, name, base, len);
2873 : }
2874 : EXPORT_SYMBOL(lookup_positive_unlocked);
2875 :
2876 : #ifdef CONFIG_UNIX98_PTYS
2877 0 : int path_pts(struct path *path)
2878 : {
2879 : /* Find something mounted on "pts" in the same directory as
2880 : * the input path.
2881 : */
2882 0 : struct dentry *parent = dget_parent(path->dentry);
2883 0 : struct dentry *child;
2884 0 : struct qstr this = QSTR_INIT("pts", 3);
2885 :
2886 0 : if (unlikely(!path_connected(path->mnt, parent))) {
2887 0 : dput(parent);
2888 0 : return -ENOENT;
2889 : }
2890 0 : dput(path->dentry);
2891 0 : path->dentry = parent;
2892 0 : child = d_hash_and_lookup(parent, &this);
2893 0 : if (!child)
2894 : return -ENOENT;
2895 :
2896 0 : path->dentry = child;
2897 0 : dput(parent);
2898 0 : follow_down(path, 0);
2899 0 : return 0;
2900 : }
2901 : #endif
2902 :
2903 1650957285 : int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2904 : struct path *path, int *empty)
2905 : {
2906 1650957285 : struct filename *filename = getname_flags(name, flags, empty);
2907 1651639187 : int ret = filename_lookup(dfd, filename, flags, path, NULL);
2908 :
2909 1651503275 : putname(filename);
2910 1651855624 : return ret;
2911 : }
2912 : EXPORT_SYMBOL(user_path_at_empty);
2913 :
2914 217073 : int __check_sticky(struct mnt_idmap *idmap, struct inode *dir,
2915 : struct inode *inode)
2916 : {
2917 217073 : kuid_t fsuid = current_fsuid();
2918 :
2919 434146 : if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode), fsuid))
2920 : return 0;
2921 12 : if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, dir), fsuid))
2922 : return 0;
2923 6 : return !capable_wrt_inode_uidgid(idmap, inode, CAP_FOWNER);
2924 : }
2925 : EXPORT_SYMBOL(__check_sticky);
2926 :
2927 : /*
2928 : * Check whether we can remove a link victim from directory dir, check
2929 : * whether the type of victim is right.
2930 : * 1. We can't do it if dir is read-only (done in permission())
2931 : * 2. We should have write and exec permissions on dir
2932 : * 3. We can't remove anything from append-only dir
2933 : * 4. We can't do anything with immutable dir (done in permission())
2934 : * 5. If the sticky bit on dir is set we should either
2935 : * a. be owner of dir, or
2936 : * b. be owner of victim, or
2937 : * c. have CAP_FOWNER capability
2938 : * 6. If the victim is append-only or immutable we can't do antyhing with
2939 : * links pointing to it.
2940 : * 7. If the victim has an unknown uid or gid we can't change the inode.
2941 : * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2942 : * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2943 : * 10. We can't remove a root or mountpoint.
2944 : * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2945 : * nfs_async_unlink().
2946 : */
2947 66811782 : static int may_delete(struct mnt_idmap *idmap, struct inode *dir,
2948 : struct dentry *victim, bool isdir)
2949 : {
2950 66811782 : struct inode *inode = d_backing_inode(victim);
2951 66811782 : int error;
2952 :
2953 66811782 : if (d_is_negative(victim))
2954 : return -ENOENT;
2955 66811782 : BUG_ON(!inode);
2956 :
2957 66811782 : BUG_ON(victim->d_parent->d_inode != dir);
2958 :
2959 : /* Inode writeback is not safe when the uid or gid are invalid. */
2960 133624484 : if (!vfsuid_valid(i_uid_into_vfsuid(idmap, inode)) ||
2961 : !vfsgid_valid(i_gid_into_vfsgid(idmap, inode)))
2962 0 : return -EOVERFLOW;
2963 :
2964 66812702 : audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2965 :
2966 66812702 : error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
2967 66810525 : if (error)
2968 : return error;
2969 66810509 : if (IS_APPEND(dir))
2970 : return -EPERM;
2971 :
2972 66810479 : if (check_sticky(idmap, dir, inode) || IS_APPEND(inode) ||
2973 133626519 : IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) ||
2974 66813169 : HAS_UNMAPPED_ID(idmap, inode))
2975 24 : return -EPERM;
2976 66813326 : if (isdir) {
2977 10298211 : if (!d_is_dir(victim))
2978 : return -ENOTDIR;
2979 10286025 : if (IS_ROOT(victim))
2980 : return -EBUSY;
2981 113038822 : } else if (d_is_dir(victim))
2982 : return -EISDIR;
2983 66803639 : if (IS_DEADDIR(dir))
2984 : return -ENOENT;
2985 66803639 : if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2986 0 : return -EBUSY;
2987 : return 0;
2988 : }
2989 :
2990 : /* Check whether we can create an object with dentry child in directory
2991 : * dir.
2992 : * 1. We can't do it if child already exists (open has special treatment for
2993 : * this case, but since we are inlined it's OK)
2994 : * 2. We can't do it if dir is read-only (done in permission())
2995 : * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2996 : * 4. We should have write and exec permissions on dir
2997 : * 5. We can't do it if dir is immutable (done in permission())
2998 : */
2999 263804085 : static inline int may_create(struct mnt_idmap *idmap,
3000 : struct inode *dir, struct dentry *child)
3001 : {
3002 263804085 : audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
3003 263804085 : if (child->d_inode)
3004 : return -EEXIST;
3005 263804085 : if (IS_DEADDIR(dir))
3006 : return -ENOENT;
3007 263804085 : if (!fsuidgid_has_mapping(dir->i_sb, idmap))
3008 : return -EOVERFLOW;
3009 :
3010 263809331 : return inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
3011 : }
3012 :
3013 21495130 : static struct dentry *lock_two_directories(struct dentry *p1, struct dentry *p2)
3014 : {
3015 21495130 : struct dentry *p;
3016 :
3017 21495130 : p = d_ancestor(p2, p1);
3018 21495130 : if (p) {
3019 911881 : inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
3020 911881 : inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
3021 911881 : return p;
3022 : }
3023 :
3024 20583249 : p = d_ancestor(p1, p2);
3025 20583249 : if (p) {
3026 572839 : inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3027 572839 : inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
3028 572839 : return p;
3029 : }
3030 :
3031 20010410 : lock_two_inodes(p1->d_inode, p2->d_inode,
3032 : I_MUTEX_PARENT, I_MUTEX_PARENT2);
3033 20010410 : return NULL;
3034 : }
3035 :
3036 : /*
3037 : * p1 and p2 should be directories on the same fs.
3038 : */
3039 23553859 : struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
3040 : {
3041 23553859 : if (p1 == p2) {
3042 2061370 : inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3043 2061370 : return NULL;
3044 : }
3045 :
3046 21492489 : mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
3047 21495130 : return lock_two_directories(p1, p2);
3048 : }
3049 : EXPORT_SYMBOL(lock_rename);
3050 :
3051 : /*
3052 : * c1 and p2 should be on the same fs.
3053 : */
3054 0 : struct dentry *lock_rename_child(struct dentry *c1, struct dentry *p2)
3055 : {
3056 0 : if (READ_ONCE(c1->d_parent) == p2) {
3057 : /*
3058 : * hopefully won't need to touch ->s_vfs_rename_mutex at all.
3059 : */
3060 0 : inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
3061 : /*
3062 : * now that p2 is locked, nobody can move in or out of it,
3063 : * so the test below is safe.
3064 : */
3065 0 : if (likely(c1->d_parent == p2))
3066 : return NULL;
3067 :
3068 : /*
3069 : * c1 got moved out of p2 while we'd been taking locks;
3070 : * unlock and fall back to slow case.
3071 : */
3072 0 : inode_unlock(p2->d_inode);
3073 : }
3074 :
3075 0 : mutex_lock(&c1->d_sb->s_vfs_rename_mutex);
3076 : /*
3077 : * nobody can move out of any directories on this fs.
3078 : */
3079 0 : if (likely(c1->d_parent != p2))
3080 0 : return lock_two_directories(c1->d_parent, p2);
3081 :
3082 : /*
3083 : * c1 got moved into p2 while we were taking locks;
3084 : * we need p2 locked and ->s_vfs_rename_mutex unlocked,
3085 : * for consistency with lock_rename().
3086 : */
3087 0 : inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
3088 0 : mutex_unlock(&c1->d_sb->s_vfs_rename_mutex);
3089 0 : return NULL;
3090 : }
3091 : EXPORT_SYMBOL(lock_rename_child);
3092 :
3093 23556478 : void unlock_rename(struct dentry *p1, struct dentry *p2)
3094 : {
3095 23556478 : inode_unlock(p1->d_inode);
3096 23556495 : if (p1 != p2) {
3097 21495130 : inode_unlock(p2->d_inode);
3098 21495130 : mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
3099 : }
3100 23556477 : }
3101 : EXPORT_SYMBOL(unlock_rename);
3102 :
3103 : /**
3104 : * mode_strip_umask - handle vfs umask stripping
3105 : * @dir: parent directory of the new inode
3106 : * @mode: mode of the new inode to be created in @dir
3107 : *
3108 : * Umask stripping depends on whether or not the filesystem supports POSIX
3109 : * ACLs. If the filesystem doesn't support it umask stripping is done directly
3110 : * in here. If the filesystem does support POSIX ACLs umask stripping is
3111 : * deferred until the filesystem calls posix_acl_create().
3112 : *
3113 : * Returns: mode
3114 : */
3115 26376888 : static inline umode_t mode_strip_umask(const struct inode *dir, umode_t mode)
3116 : {
3117 26376888 : if (!IS_POSIXACL(dir))
3118 47775 : mode &= ~current_umask();
3119 26376888 : return mode;
3120 : }
3121 :
3122 : /**
3123 : * vfs_prepare_mode - prepare the mode to be used for a new inode
3124 : * @idmap: idmap of the mount the inode was found from
3125 : * @dir: parent directory of the new inode
3126 : * @mode: mode of the new inode
3127 : * @mask_perms: allowed permission by the vfs
3128 : * @type: type of file to be created
3129 : *
3130 : * This helper consolidates and enforces vfs restrictions on the @mode of a new
3131 : * object to be created.
3132 : *
3133 : * Umask stripping depends on whether the filesystem supports POSIX ACLs (see
3134 : * the kernel documentation for mode_strip_umask()). Moving umask stripping
3135 : * after setgid stripping allows the same ordering for both non-POSIX ACL and
3136 : * POSIX ACL supporting filesystems.
3137 : *
3138 : * Note that it's currently valid for @type to be 0 if a directory is created.
3139 : * Filesystems raise that flag individually and we need to check whether each
3140 : * filesystem can deal with receiving S_IFDIR from the vfs before we enforce a
3141 : * non-zero type.
3142 : *
3143 : * Returns: mode to be passed to the filesystem
3144 : */
3145 19971831 : static inline umode_t vfs_prepare_mode(struct mnt_idmap *idmap,
3146 : const struct inode *dir, umode_t mode,
3147 : umode_t mask_perms, umode_t type)
3148 : {
3149 19971831 : mode = mode_strip_sgid(idmap, dir, mode);
3150 19973903 : mode = mode_strip_umask(dir, mode);
3151 :
3152 : /*
3153 : * Apply the vfs mandated allowed permission mask and set the type of
3154 : * file to be created before we call into the filesystem.
3155 : */
3156 19973856 : mode &= (mask_perms & ~S_IFMT);
3157 19973856 : mode |= (type & S_IFMT);
3158 :
3159 19973856 : return mode;
3160 : }
3161 :
3162 : /**
3163 : * vfs_create - create new file
3164 : * @idmap: idmap of the mount the inode was found from
3165 : * @dir: inode of @dentry
3166 : * @dentry: pointer to dentry of the base directory
3167 : * @mode: mode of the new file
3168 : * @want_excl: whether the file must not yet exist
3169 : *
3170 : * Create a new file.
3171 : *
3172 : * If the inode has been found through an idmapped mount the idmap of
3173 : * the vfsmount must be passed through @idmap. This function will then take
3174 : * care to map the inode according to @idmap before checking permissions.
3175 : * On non-idmapped mounts or if permission checking is to be performed on the
3176 : * raw inode simply passs @nop_mnt_idmap.
3177 : */
3178 44992 : int vfs_create(struct mnt_idmap *idmap, struct inode *dir,
3179 : struct dentry *dentry, umode_t mode, bool want_excl)
3180 : {
3181 44992 : int error;
3182 :
3183 44992 : error = may_create(idmap, dir, dentry);
3184 44992 : if (error)
3185 : return error;
3186 :
3187 44990 : if (!dir->i_op->create)
3188 : return -EACCES; /* shouldn't it be ENOSYS? */
3189 :
3190 44990 : mode = vfs_prepare_mode(idmap, dir, mode, S_IALLUGO, S_IFREG);
3191 44990 : error = security_inode_create(dir, dentry, mode);
3192 44990 : if (error)
3193 : return error;
3194 44990 : error = dir->i_op->create(idmap, dir, dentry, mode, want_excl);
3195 44990 : if (!error)
3196 44990 : fsnotify_create(dir, dentry);
3197 : return error;
3198 : }
3199 : EXPORT_SYMBOL(vfs_create);
3200 :
3201 0 : int vfs_mkobj(struct dentry *dentry, umode_t mode,
3202 : int (*f)(struct dentry *, umode_t, void *),
3203 : void *arg)
3204 : {
3205 0 : struct inode *dir = dentry->d_parent->d_inode;
3206 0 : int error = may_create(&nop_mnt_idmap, dir, dentry);
3207 0 : if (error)
3208 : return error;
3209 :
3210 0 : mode &= S_IALLUGO;
3211 0 : mode |= S_IFREG;
3212 0 : error = security_inode_create(dir, dentry, mode);
3213 0 : if (error)
3214 : return error;
3215 0 : error = f(dentry, mode, arg);
3216 0 : if (!error)
3217 0 : fsnotify_create(dir, dentry);
3218 : return error;
3219 : }
3220 : EXPORT_SYMBOL(vfs_mkobj);
3221 :
3222 4122483 : bool may_open_dev(const struct path *path)
3223 : {
3224 4122483 : return !(path->mnt->mnt_flags & MNT_NODEV) &&
3225 4122483 : !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
3226 : }
3227 :
3228 388293910 : static int may_open(struct mnt_idmap *idmap, const struct path *path,
3229 : int acc_mode, int flag)
3230 : {
3231 388293910 : struct dentry *dentry = path->dentry;
3232 388293910 : struct inode *inode = dentry->d_inode;
3233 388293910 : int error;
3234 :
3235 388293910 : if (!inode)
3236 : return -ENOENT;
3237 :
3238 388293910 : switch (inode->i_mode & S_IFMT) {
3239 : case S_IFLNK:
3240 : return -ELOOP;
3241 12843508 : case S_IFDIR:
3242 12843508 : if (acc_mode & MAY_WRITE)
3243 : return -EISDIR;
3244 11661080 : if (acc_mode & MAY_EXEC)
3245 : return -EACCES;
3246 : break;
3247 3819075 : case S_IFBLK:
3248 : case S_IFCHR:
3249 3819075 : if (!may_open_dev(path))
3250 : return -EACCES;
3251 4284239 : fallthrough;
3252 : case S_IFIFO:
3253 : case S_IFSOCK:
3254 4284239 : if (acc_mode & MAY_EXEC)
3255 : return -EACCES;
3256 4284239 : flag &= ~O_TRUNC;
3257 4284239 : break;
3258 371166163 : case S_IFREG:
3259 371166163 : if ((acc_mode & MAY_EXEC) && path_noexec(path))
3260 : return -EACCES;
3261 : break;
3262 : }
3263 :
3264 387111471 : error = inode_permission(idmap, inode, MAY_OPEN | acc_mode);
3265 387113935 : if (error)
3266 : return error;
3267 :
3268 : /*
3269 : * An append-only file must be opened in append mode for writing.
3270 : */
3271 387104614 : if (IS_APPEND(inode)) {
3272 170 : if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
3273 : return -EPERM;
3274 154 : if (flag & O_TRUNC)
3275 : return -EPERM;
3276 : }
3277 :
3278 : /* O_NOATIME can only be set by the owner or superuser */
3279 387104590 : if (flag & O_NOATIME && !inode_owner_or_capable(idmap, inode))
3280 0 : return -EPERM;
3281 :
3282 : return 0;
3283 : }
3284 :
3285 437137 : static int handle_truncate(struct mnt_idmap *idmap, struct file *filp)
3286 : {
3287 437137 : const struct path *path = &filp->f_path;
3288 437137 : struct inode *inode = path->dentry->d_inode;
3289 437137 : int error = get_write_access(inode);
3290 437138 : if (error)
3291 : return error;
3292 :
3293 437138 : error = security_file_truncate(filp);
3294 437138 : if (!error) {
3295 437138 : error = do_truncate(idmap, path->dentry, 0,
3296 : ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3297 : filp);
3298 : }
3299 437139 : put_write_access(inode);
3300 437139 : return error;
3301 : }
3302 :
3303 : static inline int open_to_namei_flags(int flag)
3304 : {
3305 18774 : if ((flag & O_ACCMODE) == 3)
3306 0 : flag--;
3307 18774 : return flag;
3308 : }
3309 :
3310 12551456 : static int may_o_create(struct mnt_idmap *idmap,
3311 : const struct path *dir, struct dentry *dentry,
3312 : umode_t mode)
3313 : {
3314 12551456 : int error = security_path_mknod(dir, dentry, mode, 0);
3315 12551456 : if (error)
3316 : return error;
3317 :
3318 12551456 : if (!fsuidgid_has_mapping(dir->dentry->d_sb, idmap))
3319 : return -EOVERFLOW;
3320 :
3321 12551328 : error = inode_permission(idmap, dir->dentry->d_inode,
3322 : MAY_WRITE | MAY_EXEC);
3323 12551234 : if (error)
3324 46 : return error;
3325 :
3326 : return security_inode_create(dir->dentry->d_inode, dentry, mode);
3327 : }
3328 :
3329 : /*
3330 : * Attempt to atomically look up, create and open a file from a negative
3331 : * dentry.
3332 : *
3333 : * Returns 0 if successful. The file will have been created and attached to
3334 : * @file by the filesystem calling finish_open().
3335 : *
3336 : * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3337 : * be set. The caller will need to perform the open themselves. @path will
3338 : * have been updated to point to the new dentry. This may be negative.
3339 : *
3340 : * Returns an error code otherwise.
3341 : */
3342 18774 : static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
3343 : struct file *file,
3344 : int open_flag, umode_t mode)
3345 : {
3346 18774 : struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3347 18774 : struct inode *dir = nd->path.dentry->d_inode;
3348 18774 : int error;
3349 :
3350 18774 : if (nd->flags & LOOKUP_DIRECTORY)
3351 9 : open_flag |= O_DIRECTORY;
3352 :
3353 18774 : file->f_path.dentry = DENTRY_NOT_SET;
3354 18774 : file->f_path.mnt = nd->path.mnt;
3355 18774 : error = dir->i_op->atomic_open(dir, dentry, file,
3356 : open_to_namei_flags(open_flag), mode);
3357 18774 : d_lookup_done(dentry);
3358 18774 : if (!error) {
3359 16254 : if (file->f_mode & FMODE_OPENED) {
3360 8393 : if (unlikely(dentry != file->f_path.dentry)) {
3361 0 : dput(dentry);
3362 0 : dentry = dget(file->f_path.dentry);
3363 : }
3364 7861 : } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3365 : error = -EIO;
3366 : } else {
3367 7861 : if (file->f_path.dentry) {
3368 0 : dput(dentry);
3369 0 : dentry = file->f_path.dentry;
3370 : }
3371 7861 : if (unlikely(d_is_negative(dentry)))
3372 : error = -ENOENT;
3373 : }
3374 : }
3375 18774 : if (error) {
3376 2520 : dput(dentry);
3377 2520 : dentry = ERR_PTR(error);
3378 : }
3379 18774 : return dentry;
3380 : }
3381 :
3382 : /*
3383 : * Look up and maybe create and open the last component.
3384 : *
3385 : * Must be called with parent locked (exclusive in O_CREAT case).
3386 : *
3387 : * Returns 0 on success, that is, if
3388 : * the file was successfully atomically created (if necessary) and opened, or
3389 : * the file was not completely opened at this time, though lookups and
3390 : * creations were performed.
3391 : * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3392 : * In the latter case dentry returned in @path might be negative if O_CREAT
3393 : * hadn't been specified.
3394 : *
3395 : * An error code is returned on failure.
3396 : */
3397 70225063 : static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3398 : const struct open_flags *op,
3399 : bool got_write)
3400 : {
3401 70225063 : struct mnt_idmap *idmap;
3402 70225063 : struct dentry *dir = nd->path.dentry;
3403 70225063 : struct inode *dir_inode = dir->d_inode;
3404 70225063 : int open_flag = op->open_flag;
3405 70225063 : struct dentry *dentry;
3406 70225063 : int error, create_error = 0;
3407 70225063 : umode_t mode = op->mode;
3408 70225063 : DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3409 :
3410 70225063 : if (unlikely(IS_DEADDIR(dir_inode)))
3411 : return ERR_PTR(-ENOENT);
3412 :
3413 70225063 : file->f_mode &= ~FMODE_CREATED;
3414 70225063 : dentry = d_lookup(dir, &nd->last);
3415 70226144 : for (;;) {
3416 70226142 : if (!dentry) {
3417 65093900 : dentry = d_alloc_parallel(dir, &nd->last, &wq);
3418 65092185 : if (IS_ERR(dentry))
3419 : return dentry;
3420 : }
3421 70224429 : if (d_in_lookup(dentry))
3422 : break;
3423 :
3424 5134747 : error = d_revalidate(dentry, nd->flags);
3425 5134750 : if (likely(error > 0))
3426 : break;
3427 2 : if (error)
3428 0 : goto out_dput;
3429 2 : d_invalidate(dentry);
3430 2 : dput(dentry);
3431 2 : dentry = NULL;
3432 : }
3433 70224430 : if (dentry->d_inode) {
3434 : /* Cached positive dentry: will open in f_op->open */
3435 : return dentry;
3436 : }
3437 :
3438 : /*
3439 : * Checking write permission is tricky, bacuse we don't know if we are
3440 : * going to actually need it: O_CREAT opens should work as long as the
3441 : * file exists. But checking existence breaks atomicity. The trick is
3442 : * to check access and if not granted clear O_CREAT from the flags.
3443 : *
3444 : * Another problem is returing the "right" error value (e.g. for an
3445 : * O_EXCL open we want to return EEXIST not EROFS).
3446 : */
3447 65686919 : if (unlikely(!got_write))
3448 6131035 : open_flag &= ~O_TRUNC;
3449 65686919 : idmap = mnt_idmap(nd->path.mnt);
3450 65685315 : if (open_flag & O_CREAT) {
3451 12551943 : if (open_flag & O_EXCL)
3452 2640652 : open_flag &= ~O_TRUNC;
3453 12551943 : mode = vfs_prepare_mode(idmap, dir->d_inode, mode, mode, mode);
3454 12551855 : if (likely(got_write))
3455 12551410 : create_error = may_o_create(idmap, &nd->path,
3456 : dentry, mode);
3457 : else
3458 : create_error = -EROFS;
3459 : }
3460 65685171 : if (create_error)
3461 493 : open_flag &= ~O_CREAT;
3462 65685171 : if (dir_inode->i_op->atomic_open) {
3463 18774 : dentry = atomic_open(nd, dentry, file, open_flag, mode);
3464 18774 : if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3465 24 : dentry = ERR_PTR(create_error);
3466 18774 : return dentry;
3467 : }
3468 :
3469 65666397 : if (d_in_lookup(dentry)) {
3470 65072729 : struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3471 : nd->flags);
3472 65072712 : d_lookup_done(dentry);
3473 65072884 : if (unlikely(res)) {
3474 150803 : if (IS_ERR(res)) {
3475 128245 : error = PTR_ERR(res);
3476 128245 : goto out_dput;
3477 : }
3478 22558 : dput(dentry);
3479 22558 : dentry = res;
3480 : }
3481 : }
3482 :
3483 : /* Negative dentry, just create the file */
3484 65538307 : if (!dentry->d_inode && (open_flag & O_CREAT)) {
3485 12194849 : file->f_mode |= FMODE_CREATED;
3486 12194849 : audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3487 12194849 : if (!dir_inode->i_op->create) {
3488 0 : error = -EACCES;
3489 0 : goto out_dput;
3490 : }
3491 :
3492 12194849 : error = dir_inode->i_op->create(idmap, dir_inode, dentry,
3493 12194849 : mode, open_flag & O_EXCL);
3494 12194555 : if (error)
3495 299527 : goto out_dput;
3496 : }
3497 65238486 : if (unlikely(create_error) && !dentry->d_inode) {
3498 423 : error = create_error;
3499 423 : goto out_dput;
3500 : }
3501 : return dentry;
3502 :
3503 428195 : out_dput:
3504 428195 : dput(dentry);
3505 428194 : return ERR_PTR(error);
3506 : }
3507 :
3508 424448405 : static const char *open_last_lookups(struct nameidata *nd,
3509 : struct file *file, const struct open_flags *op)
3510 : {
3511 424448405 : struct dentry *dir = nd->path.dentry;
3512 424448405 : int open_flag = op->open_flag;
3513 424448405 : bool got_write = false;
3514 424448405 : struct dentry *dentry;
3515 424448405 : const char *res;
3516 :
3517 424448405 : nd->flags |= op->intent;
3518 :
3519 424448405 : if (nd->last_type != LAST_NORM) {
3520 7688671 : if (nd->depth)
3521 0 : put_link(nd);
3522 7688671 : return handle_dots(nd, nd->last_type);
3523 : }
3524 :
3525 416759734 : if (!(open_flag & O_CREAT)) {
3526 399674833 : if (nd->last.name[nd->last.len])
3527 73844 : nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3528 : /* we _can_ be in RCU mode here */
3529 399674833 : dentry = lookup_fast(nd);
3530 399686461 : if (IS_ERR(dentry))
3531 : return ERR_CAST(dentry);
3532 399686372 : if (likely(dentry))
3533 346549766 : goto finish_lookup;
3534 :
3535 53136606 : BUG_ON(nd->flags & LOOKUP_RCU);
3536 : } else {
3537 : /* create side of things */
3538 17084901 : if (nd->flags & LOOKUP_RCU) {
3539 16506561 : if (!try_to_unlazy(nd))
3540 : return ERR_PTR(-ECHILD);
3541 : }
3542 17088146 : audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3543 : /* trailing slashes? */
3544 17088146 : if (unlikely(nd->last.name[nd->last.len]))
3545 : return ERR_PTR(-EISDIR);
3546 : }
3547 :
3548 70224752 : if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3549 64092417 : got_write = !mnt_want_write(nd->path.mnt);
3550 : /*
3551 : * do _not_ fail yet - we might not need that or fail with
3552 : * a different error; let lookup_open() decide; we'll be
3553 : * dropping this one anyway.
3554 : */
3555 : }
3556 70225551 : if (open_flag & O_CREAT)
3557 17088412 : inode_lock(dir->d_inode);
3558 : else
3559 53137139 : inode_lock_shared(dir->d_inode);
3560 70226388 : dentry = lookup_open(nd, file, op, got_write);
3561 70226074 : if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED))
3562 11895248 : fsnotify_create(dir->d_inode, dentry);
3563 70226081 : if (open_flag & O_CREAT)
3564 17088912 : inode_unlock(dir->d_inode);
3565 : else
3566 53137169 : inode_unlock_shared(dir->d_inode);
3567 :
3568 70225027 : if (got_write)
3569 64092679 : mnt_drop_write(nd->path.mnt);
3570 :
3571 70225090 : if (IS_ERR(dentry))
3572 : return ERR_CAST(dentry);
3573 :
3574 69794374 : if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
3575 11903684 : dput(nd->path.dentry);
3576 11903719 : nd->path.dentry = dentry;
3577 11903719 : return NULL;
3578 : }
3579 :
3580 57890690 : finish_lookup:
3581 404440456 : if (nd->depth)
3582 33251133 : put_link(nd);
3583 404440698 : res = step_into(nd, WALK_TRAILING, dentry);
3584 404420241 : if (unlikely(res))
3585 36632664 : nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3586 : return res;
3587 : }
3588 :
3589 : /*
3590 : * Handle the last step of open()
3591 : */
3592 387400689 : static int do_open(struct nameidata *nd,
3593 : struct file *file, const struct open_flags *op)
3594 : {
3595 387400689 : struct mnt_idmap *idmap;
3596 387400689 : int open_flag = op->open_flag;
3597 387400689 : bool do_truncate;
3598 387400689 : int acc_mode;
3599 387400689 : int error;
3600 :
3601 387400689 : if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3602 375501515 : error = complete_walk(nd);
3603 375507149 : if (error)
3604 : return error;
3605 : }
3606 387404106 : if (!(file->f_mode & FMODE_CREATED))
3607 : audit_inode(nd->name, nd->path.dentry, 0);
3608 387404106 : idmap = mnt_idmap(nd->path.mnt);
3609 387410490 : if (open_flag & O_CREAT) {
3610 16700075 : if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3611 : return -EEXIST;
3612 33267321 : if (d_is_dir(nd->path.dentry))
3613 : return -EISDIR;
3614 16632566 : error = may_create_in_sticky(idmap, nd,
3615 : d_backing_inode(nd->path.dentry));
3616 16632337 : if (unlikely(error))
3617 : return error;
3618 : }
3619 387342740 : if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3620 : return -ENOTDIR;
3621 :
3622 387342682 : do_truncate = false;
3623 387342682 : acc_mode = op->acc_mode;
3624 387342682 : if (file->f_mode & FMODE_CREATED) {
3625 : /* Don't check for write permission, don't truncate */
3626 11895279 : open_flag &= ~O_TRUNC;
3627 11895279 : acc_mode = 0;
3628 375447403 : } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) {
3629 437262 : error = mnt_want_write(nd->path.mnt);
3630 437263 : if (error)
3631 : return error;
3632 : do_truncate = true;
3633 : }
3634 387342683 : error = may_open(idmap, &nd->path, acc_mode, open_flag);
3635 387321602 : if (!error && !(file->f_mode & FMODE_OPENED))
3636 386127711 : error = vfs_open(&nd->path, file);
3637 387329883 : if (!error)
3638 385564986 : error = ima_file_check(file, op->acc_mode);
3639 387329883 : if (!error && do_truncate)
3640 437137 : error = handle_truncate(idmap, file);
3641 387329884 : if (unlikely(error > 0)) {
3642 0 : WARN_ON(1);
3643 0 : error = -EINVAL;
3644 : }
3645 387329884 : if (do_truncate)
3646 437262 : mnt_drop_write(nd->path.mnt);
3647 : return error;
3648 : }
3649 :
3650 : /**
3651 : * vfs_tmpfile - create tmpfile
3652 : * @idmap: idmap of the mount the inode was found from
3653 : * @parentpath: pointer to the path of the base directory
3654 : * @file: file descriptor of the new tmpfile
3655 : * @mode: mode of the new tmpfile
3656 : *
3657 : * Create a temporary file.
3658 : *
3659 : * If the inode has been found through an idmapped mount the idmap of
3660 : * the vfsmount must be passed through @idmap. This function will then take
3661 : * care to map the inode according to @idmap before checking permissions.
3662 : * On non-idmapped mounts or if permission checking is to be performed on the
3663 : * raw inode simply passs @nop_mnt_idmap.
3664 : */
3665 976214 : static int vfs_tmpfile(struct mnt_idmap *idmap,
3666 : const struct path *parentpath,
3667 : struct file *file, umode_t mode)
3668 : {
3669 976214 : struct dentry *child;
3670 976214 : struct inode *dir = d_inode(parentpath->dentry);
3671 976214 : struct inode *inode;
3672 976214 : int error;
3673 976214 : int open_flag = file->f_flags;
3674 :
3675 : /* we want directory to be writable */
3676 976214 : error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC);
3677 976242 : if (error)
3678 : return error;
3679 976242 : if (!dir->i_op->tmpfile)
3680 : return -EOPNOTSUPP;
3681 976242 : child = d_alloc(parentpath->dentry, &slash_name);
3682 974594 : if (unlikely(!child))
3683 : return -ENOMEM;
3684 974594 : file->f_path.mnt = parentpath->mnt;
3685 974594 : file->f_path.dentry = child;
3686 974594 : mode = vfs_prepare_mode(idmap, dir, mode, mode, mode);
3687 976474 : error = dir->i_op->tmpfile(idmap, dir, file, mode);
3688 976494 : dput(child);
3689 976497 : if (error)
3690 : return error;
3691 : /* Don't check for other permissions, the inode was just created */
3692 976497 : error = may_open(idmap, &file->f_path, 0, file->f_flags);
3693 976456 : if (error)
3694 : return error;
3695 976456 : inode = file_inode(file);
3696 976456 : if (!(open_flag & O_EXCL)) {
3697 976296 : spin_lock(&inode->i_lock);
3698 976329 : inode->i_state |= I_LINKABLE;
3699 976329 : spin_unlock(&inode->i_lock);
3700 : }
3701 : ima_post_create_tmpfile(idmap, inode);
3702 : return 0;
3703 : }
3704 :
3705 : /**
3706 : * kernel_tmpfile_open - open a tmpfile for kernel internal use
3707 : * @idmap: idmap of the mount the inode was found from
3708 : * @parentpath: path of the base directory
3709 : * @mode: mode of the new tmpfile
3710 : * @open_flag: flags
3711 : * @cred: credentials for open
3712 : *
3713 : * Create and open a temporary file. The file is not accounted in nr_files,
3714 : * hence this is only for kernel internal use, and must not be installed into
3715 : * file tables or such.
3716 : */
3717 16408 : struct file *kernel_tmpfile_open(struct mnt_idmap *idmap,
3718 : const struct path *parentpath,
3719 : umode_t mode, int open_flag,
3720 : const struct cred *cred)
3721 : {
3722 16408 : struct file *file;
3723 16408 : int error;
3724 :
3725 16408 : file = alloc_empty_file_noaccount(open_flag, cred);
3726 16408 : if (IS_ERR(file))
3727 : return file;
3728 :
3729 16408 : error = vfs_tmpfile(idmap, parentpath, file, mode);
3730 16408 : if (error) {
3731 0 : fput(file);
3732 0 : file = ERR_PTR(error);
3733 : }
3734 : return file;
3735 : }
3736 : EXPORT_SYMBOL(kernel_tmpfile_open);
3737 :
3738 959974 : static int do_tmpfile(struct nameidata *nd, unsigned flags,
3739 : const struct open_flags *op,
3740 : struct file *file)
3741 : {
3742 959974 : struct path path;
3743 959974 : int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3744 :
3745 960005 : if (unlikely(error))
3746 : return error;
3747 960005 : error = mnt_want_write(path.mnt);
3748 960073 : if (unlikely(error))
3749 0 : goto out;
3750 960073 : error = vfs_tmpfile(mnt_idmap(path.mnt), &path, file, op->mode);
3751 959843 : if (error)
3752 : goto out2;
3753 959843 : audit_inode(nd->name, file->f_path.dentry, 0);
3754 : out2:
3755 959843 : mnt_drop_write(path.mnt);
3756 960060 : out:
3757 960060 : path_put(&path);
3758 960060 : return error;
3759 : }
3760 :
3761 17961626 : static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3762 : {
3763 17961626 : struct path path;
3764 17961626 : int error = path_lookupat(nd, flags, &path);
3765 17960709 : if (!error) {
3766 17396531 : audit_inode(nd->name, path.dentry, 0);
3767 17396531 : error = vfs_open(&path, file);
3768 17393184 : path_put(&path);
3769 : }
3770 17963647 : return error;
3771 : }
3772 :
3773 447115647 : static struct file *path_openat(struct nameidata *nd,
3774 : const struct open_flags *op, unsigned flags)
3775 : {
3776 447115647 : struct file *file;
3777 447115647 : int error;
3778 :
3779 447115647 : file = alloc_empty_file(op->open_flag, current_cred());
3780 447149844 : if (IS_ERR(file))
3781 : return file;
3782 :
3783 447149844 : if (unlikely(file->f_flags & __O_TMPFILE)) {
3784 959983 : error = do_tmpfile(nd, flags, op, file);
3785 446189861 : } else if (unlikely(file->f_flags & O_PATH)) {
3786 17960932 : error = do_o_path(nd, flags, file);
3787 : } else {
3788 428228929 : const char *s = path_init(nd, flags);
3789 852674797 : while (!(error = link_path_walk(s, nd)) &&
3790 424455413 : (s = open_last_lookups(nd, file, op)) != NULL)
3791 : ;
3792 428229902 : if (!error)
3793 387404039 : error = do_open(nd, file, op);
3794 428227163 : terminate_walk(nd);
3795 : }
3796 447158158 : if (likely(!error)) {
3797 403919030 : if (likely(file->f_mode & FMODE_OPENED))
3798 : return file;
3799 0 : WARN_ON(1);
3800 0 : error = -EINVAL;
3801 : }
3802 43239128 : fput(file);
3803 43238856 : if (error == -EOPENSTALE) {
3804 0 : if (flags & LOOKUP_RCU)
3805 : error = -ECHILD;
3806 : else
3807 0 : error = -ESTALE;
3808 : }
3809 43238856 : return ERR_PTR(error);
3810 : }
3811 :
3812 447113944 : struct file *do_filp_open(int dfd, struct filename *pathname,
3813 : const struct open_flags *op)
3814 : {
3815 447113944 : struct nameidata nd;
3816 447113944 : int flags = op->lookup_flags;
3817 447113944 : struct file *filp;
3818 :
3819 447113944 : set_nameidata(&nd, dfd, pathname, NULL);
3820 447133092 : filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3821 447137186 : if (unlikely(filp == ERR_PTR(-ECHILD)))
3822 3196 : filp = path_openat(&nd, op, flags);
3823 447137186 : if (unlikely(filp == ERR_PTR(-ESTALE)))
3824 0 : filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3825 447137186 : restore_nameidata();
3826 447138563 : return filp;
3827 : }
3828 :
3829 6376 : struct file *do_file_open_root(const struct path *root,
3830 : const char *name, const struct open_flags *op)
3831 : {
3832 6376 : struct nameidata nd;
3833 6376 : struct file *file;
3834 6376 : struct filename *filename;
3835 6376 : int flags = op->lookup_flags;
3836 :
3837 6376 : if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN)
3838 : return ERR_PTR(-ELOOP);
3839 :
3840 6376 : filename = getname_kernel(name);
3841 6376 : if (IS_ERR(filename))
3842 : return ERR_CAST(filename);
3843 :
3844 6376 : set_nameidata(&nd, -1, filename, root);
3845 6376 : file = path_openat(&nd, op, flags | LOOKUP_RCU);
3846 6376 : if (unlikely(file == ERR_PTR(-ECHILD)))
3847 0 : file = path_openat(&nd, op, flags);
3848 6376 : if (unlikely(file == ERR_PTR(-ESTALE)))
3849 0 : file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3850 6376 : restore_nameidata();
3851 6376 : putname(filename);
3852 6376 : return file;
3853 : }
3854 :
3855 249006014 : static struct dentry *filename_create(int dfd, struct filename *name,
3856 : struct path *path, unsigned int lookup_flags)
3857 : {
3858 249006014 : struct dentry *dentry = ERR_PTR(-EEXIST);
3859 249006014 : struct qstr last;
3860 249006014 : bool want_dir = lookup_flags & LOOKUP_DIRECTORY;
3861 249006014 : unsigned int reval_flag = lookup_flags & LOOKUP_REVAL;
3862 249006014 : unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL;
3863 249006014 : int type;
3864 249006014 : int err2;
3865 249006014 : int error;
3866 :
3867 249006014 : error = filename_parentat(dfd, name, reval_flag, path, &last, &type);
3868 249073229 : if (error)
3869 2 : return ERR_PTR(error);
3870 :
3871 : /*
3872 : * Yucky last component or no last component at all?
3873 : * (foo/., foo/.., /////)
3874 : */
3875 249073227 : if (unlikely(type != LAST_NORM))
3876 0 : goto out;
3877 :
3878 : /* don't fail immediately if it's r/o, at least try to report other errors */
3879 249073227 : err2 = mnt_want_write(path->mnt);
3880 : /*
3881 : * Do the final lookup. Suppress 'create' if there is a trailing
3882 : * '/', and a directory wasn't requested.
3883 : */
3884 249070836 : if (last.name[last.len] && !want_dir)
3885 28 : create_flags = 0;
3886 249070836 : inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3887 249057896 : dentry = lookup_one_qstr_excl(&last, path->dentry,
3888 : reval_flag | create_flags);
3889 249053134 : if (IS_ERR(dentry))
3890 15368 : goto unlock;
3891 :
3892 249037766 : error = -EEXIST;
3893 249037766 : if (d_is_positive(dentry))
3894 1154533 : goto fail;
3895 :
3896 : /*
3897 : * Special case - lookup gave negative, but... we had foo/bar/
3898 : * From the vfs_mknod() POV we just have a negative dentry -
3899 : * all is fine. Let's be bastards - you had / on the end, you've
3900 : * been asking for (non-existent) directory. -ENOENT for you.
3901 : */
3902 247883233 : if (unlikely(!create_flags)) {
3903 0 : error = -ENOENT;
3904 0 : goto fail;
3905 : }
3906 247883233 : if (unlikely(err2)) {
3907 0 : error = err2;
3908 0 : goto fail;
3909 : }
3910 : return dentry;
3911 1154533 : fail:
3912 1154533 : dput(dentry);
3913 1154539 : dentry = ERR_PTR(error);
3914 1169907 : unlock:
3915 1169907 : inode_unlock(path->dentry->d_inode);
3916 1169906 : if (!err2)
3917 1132415 : mnt_drop_write(path->mnt);
3918 37491 : out:
3919 1169900 : path_put(path);
3920 1169900 : return dentry;
3921 : }
3922 :
3923 1257 : struct dentry *kern_path_create(int dfd, const char *pathname,
3924 : struct path *path, unsigned int lookup_flags)
3925 : {
3926 1257 : struct filename *filename = getname_kernel(pathname);
3927 1257 : struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3928 :
3929 1257 : putname(filename);
3930 1257 : return res;
3931 : }
3932 : EXPORT_SYMBOL(kern_path_create);
3933 :
3934 247856179 : void done_path_create(struct path *path, struct dentry *dentry)
3935 : {
3936 247856179 : dput(dentry);
3937 247882008 : inode_unlock(path->dentry->d_inode);
3938 247873684 : mnt_drop_write(path->mnt);
3939 247878348 : path_put(path);
3940 247893075 : }
3941 : EXPORT_SYMBOL(done_path_create);
3942 :
3943 0 : inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3944 : struct path *path, unsigned int lookup_flags)
3945 : {
3946 0 : struct filename *filename = getname(pathname);
3947 0 : struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3948 :
3949 0 : putname(filename);
3950 0 : return res;
3951 : }
3952 : EXPORT_SYMBOL(user_path_create);
3953 :
3954 : /**
3955 : * vfs_mknod - create device node or file
3956 : * @idmap: idmap of the mount the inode was found from
3957 : * @dir: inode of @dentry
3958 : * @dentry: pointer to dentry of the base directory
3959 : * @mode: mode of the new device node or file
3960 : * @dev: device number of device to create
3961 : *
3962 : * Create a device node or file.
3963 : *
3964 : * If the inode has been found through an idmapped mount the idmap of
3965 : * the vfsmount must be passed through @idmap. This function will then take
3966 : * care to map the inode according to @idmap before checking permissions.
3967 : * On non-idmapped mounts or if permission checking is to be performed on the
3968 : * raw inode simply passs @nop_mnt_idmap.
3969 : */
3970 3060055 : int vfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
3971 : struct dentry *dentry, umode_t mode, dev_t dev)
3972 : {
3973 3060055 : bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV;
3974 3060055 : int error = may_create(idmap, dir, dentry);
3975 :
3976 3060050 : if (error)
3977 : return error;
3978 :
3979 3060561 : if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout &&
3980 515 : !capable(CAP_MKNOD))
3981 : return -EPERM;
3982 :
3983 3059908 : if (!dir->i_op->mknod)
3984 : return -EPERM;
3985 :
3986 3059908 : mode = vfs_prepare_mode(idmap, dir, mode, mode, mode);
3987 3059912 : error = devcgroup_inode_mknod(mode, dev);
3988 3059907 : if (error)
3989 : return error;
3990 :
3991 3059907 : error = security_inode_mknod(dir, dentry, mode, dev);
3992 3059907 : if (error)
3993 : return error;
3994 :
3995 3059907 : error = dir->i_op->mknod(idmap, dir, dentry, mode, dev);
3996 3059893 : if (!error)
3997 2995837 : fsnotify_create(dir, dentry);
3998 : return error;
3999 : }
4000 : EXPORT_SYMBOL(vfs_mknod);
4001 :
4002 3319613 : static int may_mknod(umode_t mode)
4003 : {
4004 3319613 : switch (mode & S_IFMT) {
4005 : case S_IFREG:
4006 : case S_IFCHR:
4007 : case S_IFBLK:
4008 : case S_IFIFO:
4009 : case S_IFSOCK:
4010 : case 0: /* zero mode translates to S_IFREG */
4011 : return 0;
4012 0 : case S_IFDIR:
4013 0 : return -EPERM;
4014 0 : default:
4015 0 : return -EINVAL;
4016 : }
4017 : }
4018 :
4019 3319626 : static int do_mknodat(int dfd, struct filename *name, umode_t mode,
4020 : unsigned int dev)
4021 : {
4022 3319626 : struct mnt_idmap *idmap;
4023 3319626 : struct dentry *dentry;
4024 3319626 : struct path path;
4025 3319626 : int error;
4026 3319626 : unsigned int lookup_flags = 0;
4027 :
4028 3319626 : error = may_mknod(mode);
4029 3319626 : if (error)
4030 0 : goto out1;
4031 3319626 : retry:
4032 3319624 : dentry = filename_create(dfd, name, &path, lookup_flags);
4033 3319648 : error = PTR_ERR(dentry);
4034 3319648 : if (IS_ERR(dentry))
4035 240698 : goto out1;
4036 :
4037 3078950 : error = security_path_mknod(&path, dentry,
4038 3078950 : mode_strip_umask(path.dentry->d_inode, mode), dev);
4039 3078950 : if (error)
4040 : goto out2;
4041 :
4042 3078950 : idmap = mnt_idmap(path.mnt);
4043 3078950 : switch (mode & S_IFMT) {
4044 20144 : case 0: case S_IFREG:
4045 20144 : error = vfs_create(idmap, path.dentry->d_inode,
4046 : dentry, mode, true);
4047 20144 : if (!error)
4048 : ima_post_path_mknod(idmap, dentry);
4049 : break;
4050 3052950 : case S_IFCHR: case S_IFBLK:
4051 3052950 : error = vfs_mknod(idmap, path.dentry->d_inode,
4052 : dentry, mode, new_decode_dev(dev));
4053 3052950 : break;
4054 5856 : case S_IFIFO: case S_IFSOCK:
4055 5856 : error = vfs_mknod(idmap, path.dentry->d_inode,
4056 : dentry, mode, 0);
4057 5856 : break;
4058 : }
4059 3078926 : out2:
4060 3078926 : done_path_create(&path, dentry);
4061 6157870 : if (retry_estale(error, lookup_flags)) {
4062 0 : lookup_flags |= LOOKUP_REVAL;
4063 0 : goto retry;
4064 : }
4065 3078937 : out1:
4066 3319635 : putname(name);
4067 3319634 : return error;
4068 : }
4069 :
4070 6639228 : SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
4071 : unsigned int, dev)
4072 : {
4073 3319624 : return do_mknodat(dfd, getname(filename), mode, dev);
4074 : }
4075 :
4076 0 : SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
4077 : {
4078 0 : return do_mknodat(AT_FDCWD, getname(filename), mode, dev);
4079 : }
4080 :
4081 : /**
4082 : * vfs_mkdir - create directory
4083 : * @idmap: idmap of the mount the inode was found from
4084 : * @dir: inode of @dentry
4085 : * @dentry: pointer to dentry of the base directory
4086 : * @mode: mode of the new directory
4087 : *
4088 : * Create a directory.
4089 : *
4090 : * If the inode has been found through an idmapped mount the idmap of
4091 : * the vfsmount must be passed through @idmap. This function will then take
4092 : * care to map the inode according to @idmap before checking permissions.
4093 : * On non-idmapped mounts or if permission checking is to be performed on the
4094 : * raw inode simply passs @nop_mnt_idmap.
4095 : */
4096 3340740 : int vfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
4097 : struct dentry *dentry, umode_t mode)
4098 : {
4099 3340740 : int error;
4100 3340740 : unsigned max_links = dir->i_sb->s_max_links;
4101 :
4102 3340740 : error = may_create(idmap, dir, dentry);
4103 3340723 : if (error)
4104 : return error;
4105 :
4106 3340717 : if (!dir->i_op->mkdir)
4107 : return -EPERM;
4108 :
4109 3340717 : mode = vfs_prepare_mode(idmap, dir, mode, S_IRWXUGO | S_ISVTX, 0);
4110 3340748 : error = security_inode_mkdir(dir, dentry, mode);
4111 3340748 : if (error)
4112 : return error;
4113 :
4114 3340748 : if (max_links && dir->i_nlink >= max_links)
4115 : return -EMLINK;
4116 :
4117 3340748 : error = dir->i_op->mkdir(idmap, dir, dentry, mode);
4118 3340712 : if (!error)
4119 3275567 : fsnotify_mkdir(dir, dentry);
4120 : return error;
4121 : }
4122 : EXPORT_SYMBOL(vfs_mkdir);
4123 :
4124 3771113 : int do_mkdirat(int dfd, struct filename *name, umode_t mode)
4125 : {
4126 3771113 : struct dentry *dentry;
4127 3771113 : struct path path;
4128 3771113 : int error;
4129 3771113 : unsigned int lookup_flags = LOOKUP_DIRECTORY;
4130 :
4131 3771111 : retry:
4132 3771111 : dentry = filename_create(dfd, name, &path, lookup_flags);
4133 3771153 : error = PTR_ERR(dentry);
4134 3771153 : if (IS_ERR(dentry))
4135 446859 : goto out_putname;
4136 :
4137 3324294 : error = security_path_mkdir(&path, dentry,
4138 3324294 : mode_strip_umask(path.dentry->d_inode, mode));
4139 3324295 : if (!error) {
4140 3324295 : error = vfs_mkdir(mnt_idmap(path.mnt), path.dentry->d_inode,
4141 : dentry, mode);
4142 : }
4143 3324272 : done_path_create(&path, dentry);
4144 6648568 : if (retry_estale(error, lookup_flags)) {
4145 1 : lookup_flags |= LOOKUP_REVAL;
4146 1 : goto retry;
4147 : }
4148 3324286 : out_putname:
4149 3771145 : putname(name);
4150 3771146 : return error;
4151 : }
4152 :
4153 7542208 : SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
4154 : {
4155 3771113 : return do_mkdirat(dfd, getname(pathname), mode);
4156 : }
4157 :
4158 0 : SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
4159 : {
4160 0 : return do_mkdirat(AT_FDCWD, getname(pathname), mode);
4161 : }
4162 :
4163 : /**
4164 : * vfs_rmdir - remove directory
4165 : * @idmap: idmap of the mount the inode was found from
4166 : * @dir: inode of @dentry
4167 : * @dentry: pointer to dentry of the base directory
4168 : *
4169 : * Remove a directory.
4170 : *
4171 : * If the inode has been found through an idmapped mount the idmap of
4172 : * the vfsmount must be passed through @idmap. This function will then take
4173 : * care to map the inode according to @idmap before checking permissions.
4174 : * On non-idmapped mounts or if permission checking is to be performed on the
4175 : * raw inode simply passs @nop_mnt_idmap.
4176 : */
4177 1425150 : int vfs_rmdir(struct mnt_idmap *idmap, struct inode *dir,
4178 : struct dentry *dentry)
4179 : {
4180 1425150 : int error = may_delete(idmap, dir, dentry, 1);
4181 :
4182 1425152 : if (error)
4183 : return error;
4184 :
4185 1420176 : if (!dir->i_op->rmdir)
4186 : return -EPERM;
4187 :
4188 1420176 : dget(dentry);
4189 1420177 : inode_lock(dentry->d_inode);
4190 :
4191 1420176 : error = -EBUSY;
4192 1420176 : if (is_local_mountpoint(dentry) ||
4193 1420170 : (dentry->d_inode->i_flags & S_KERNEL_FILE))
4194 8 : goto out;
4195 :
4196 1420170 : error = security_inode_rmdir(dir, dentry);
4197 1420170 : if (error)
4198 : goto out;
4199 :
4200 1420170 : error = dir->i_op->rmdir(dir, dentry);
4201 1420166 : if (error)
4202 1107051 : goto out;
4203 :
4204 313115 : shrink_dcache_parent(dentry);
4205 313115 : dentry->d_inode->i_flags |= S_DEAD;
4206 313115 : dont_mount(dentry);
4207 313115 : detach_mounts(dentry);
4208 :
4209 1420174 : out:
4210 1420174 : inode_unlock(dentry->d_inode);
4211 1420170 : dput(dentry);
4212 1420177 : if (!error)
4213 313115 : d_delete_notify(dir, dentry);
4214 : return error;
4215 : }
4216 : EXPORT_SYMBOL(vfs_rmdir);
4217 :
4218 1473246 : int do_rmdir(int dfd, struct filename *name)
4219 : {
4220 1473246 : int error;
4221 1473246 : struct dentry *dentry;
4222 1473246 : struct path path;
4223 1473246 : struct qstr last;
4224 1473246 : int type;
4225 1473246 : unsigned int lookup_flags = 0;
4226 1473241 : retry:
4227 1473241 : error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4228 1473253 : if (error)
4229 0 : goto exit1;
4230 :
4231 1473253 : switch (type) {
4232 : case LAST_DOTDOT:
4233 : error = -ENOTEMPTY;
4234 : goto exit2;
4235 : case LAST_DOT:
4236 : error = -EINVAL;
4237 : goto exit2;
4238 : case LAST_ROOT:
4239 : error = -EBUSY;
4240 : goto exit2;
4241 : }
4242 :
4243 1473253 : error = mnt_want_write(path.mnt);
4244 1473253 : if (error)
4245 0 : goto exit2;
4246 :
4247 1473253 : inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4248 1473254 : dentry = lookup_one_qstr_excl(&last, path.dentry, lookup_flags);
4249 1473248 : error = PTR_ERR(dentry);
4250 1473248 : if (IS_ERR(dentry))
4251 0 : goto exit3;
4252 1473248 : if (!dentry->d_inode) {
4253 48157 : error = -ENOENT;
4254 48157 : goto exit4;
4255 : }
4256 1425091 : error = security_path_rmdir(&path, dentry);
4257 1425091 : if (error)
4258 : goto exit4;
4259 1425091 : error = vfs_rmdir(mnt_idmap(path.mnt), path.dentry->d_inode, dentry);
4260 1473247 : exit4:
4261 1473247 : dput(dentry);
4262 1473250 : exit3:
4263 1473250 : inode_unlock(path.dentry->d_inode);
4264 1473237 : mnt_drop_write(path.mnt);
4265 1473241 : exit2:
4266 1473241 : path_put(&path);
4267 2946492 : if (retry_estale(error, lookup_flags)) {
4268 0 : lookup_flags |= LOOKUP_REVAL;
4269 0 : goto retry;
4270 : }
4271 1473251 : exit1:
4272 1473251 : putname(name);
4273 1473249 : return error;
4274 : }
4275 :
4276 0 : SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
4277 : {
4278 0 : return do_rmdir(AT_FDCWD, getname(pathname));
4279 : }
4280 :
4281 : /**
4282 : * vfs_unlink - unlink a filesystem object
4283 : * @idmap: idmap of the mount the inode was found from
4284 : * @dir: parent directory
4285 : * @dentry: victim
4286 : * @delegated_inode: returns victim inode, if the inode is delegated.
4287 : *
4288 : * The caller must hold dir->i_mutex.
4289 : *
4290 : * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
4291 : * return a reference to the inode in delegated_inode. The caller
4292 : * should then break the delegation on that inode and retry. Because
4293 : * breaking a delegation may take a long time, the caller should drop
4294 : * dir->i_mutex before doing so.
4295 : *
4296 : * Alternatively, a caller may pass NULL for delegated_inode. This may
4297 : * be appropriate for callers that expect the underlying filesystem not
4298 : * to be NFS exported.
4299 : *
4300 : * If the inode has been found through an idmapped mount the idmap of
4301 : * the vfsmount must be passed through @idmap. This function will then take
4302 : * care to map the inode according to @idmap before checking permissions.
4303 : * On non-idmapped mounts or if permission checking is to be performed on the
4304 : * raw inode simply passs @nop_mnt_idmap.
4305 : */
4306 34922372 : int vfs_unlink(struct mnt_idmap *idmap, struct inode *dir,
4307 : struct dentry *dentry, struct inode **delegated_inode)
4308 : {
4309 34922372 : struct inode *target = dentry->d_inode;
4310 34922372 : int error = may_delete(idmap, dir, dentry, 0);
4311 :
4312 34922371 : if (error)
4313 : return error;
4314 :
4315 34920054 : if (!dir->i_op->unlink)
4316 : return -EPERM;
4317 :
4318 34920054 : inode_lock(target);
4319 34920838 : if (IS_SWAPFILE(target))
4320 : error = -EPERM;
4321 34920660 : else if (is_local_mountpoint(dentry))
4322 : error = -EBUSY;
4323 : else {
4324 34920901 : error = security_inode_unlink(dir, dentry);
4325 34920901 : if (!error) {
4326 34920901 : error = try_break_deleg(target, delegated_inode);
4327 34920773 : if (error)
4328 0 : goto out;
4329 34920773 : error = dir->i_op->unlink(dir, dentry);
4330 34920955 : if (!error) {
4331 34920435 : dont_mount(dentry);
4332 34920739 : detach_mounts(dentry);
4333 : }
4334 : }
4335 : }
4336 520 : out:
4337 34921462 : inode_unlock(target);
4338 :
4339 : /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4340 34921288 : if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) {
4341 0 : fsnotify_unlink(dir, dentry);
4342 34921288 : } else if (!error) {
4343 34920614 : fsnotify_link_count(target);
4344 34920030 : d_delete_notify(dir, dentry);
4345 : }
4346 :
4347 : return error;
4348 : }
4349 : EXPORT_SYMBOL(vfs_unlink);
4350 :
4351 : /*
4352 : * Make sure that the actual truncation of the file will occur outside its
4353 : * directory's i_mutex. Truncate can take a long time if there is a lot of
4354 : * writeout happening, and we don't want to prevent access to the directory
4355 : * while waiting on the I/O.
4356 : */
4357 35739754 : int do_unlinkat(int dfd, struct filename *name)
4358 : {
4359 35739754 : int error;
4360 35739754 : struct dentry *dentry;
4361 35739754 : struct path path;
4362 35739754 : struct qstr last;
4363 35739754 : int type;
4364 35739754 : struct inode *inode = NULL;
4365 35739754 : struct inode *delegated_inode = NULL;
4366 35739754 : unsigned int lookup_flags = 0;
4367 35738600 : retry:
4368 35738600 : error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4369 35741743 : if (error)
4370 5681 : goto exit1;
4371 :
4372 35736062 : error = -EISDIR;
4373 35736062 : if (type != LAST_NORM)
4374 0 : goto exit2;
4375 :
4376 35736062 : error = mnt_want_write(path.mnt);
4377 35736265 : if (error)
4378 421 : goto exit2;
4379 35735844 : retry_deleg:
4380 35735844 : inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4381 35735861 : dentry = lookup_one_qstr_excl(&last, path.dentry, lookup_flags);
4382 35733495 : error = PTR_ERR(dentry);
4383 35733495 : if (!IS_ERR(dentry)) {
4384 :
4385 : /* Why not before? Because we want correct error value */
4386 35712545 : if (last.name[last.len])
4387 0 : goto slashes;
4388 35712545 : inode = dentry->d_inode;
4389 35712545 : if (d_is_negative(dentry))
4390 808389 : goto slashes;
4391 34904156 : ihold(inode);
4392 34905773 : error = security_path_unlink(&path, dentry);
4393 34905773 : if (error)
4394 : goto exit3;
4395 34905773 : error = vfs_unlink(mnt_idmap(path.mnt), path.dentry->d_inode,
4396 : dentry, &delegated_inode);
4397 35714778 : exit3:
4398 35714778 : dput(dentry);
4399 : }
4400 35735322 : inode_unlock(path.dentry->d_inode);
4401 35735836 : if (inode)
4402 34906512 : iput(inode); /* truncate the inode here */
4403 35734730 : inode = NULL;
4404 35734730 : if (delegated_inode) {
4405 0 : error = break_deleg_wait(&delegated_inode);
4406 0 : if (!error)
4407 0 : goto retry_deleg;
4408 : }
4409 35734730 : mnt_drop_write(path.mnt);
4410 35735508 : exit2:
4411 35735508 : path_put(&path);
4412 71469558 : if (retry_estale(error, lookup_flags)) {
4413 0 : lookup_flags |= LOOKUP_REVAL;
4414 0 : inode = NULL;
4415 0 : goto retry;
4416 : }
4417 35735933 : exit1:
4418 35741614 : putname(name);
4419 35740676 : return error;
4420 :
4421 808389 : slashes:
4422 808389 : if (d_is_negative(dentry))
4423 : error = -ENOENT;
4424 0 : else if (d_is_dir(dentry))
4425 : error = -EISDIR;
4426 : else
4427 0 : error = -ENOTDIR;
4428 808389 : goto exit3;
4429 : }
4430 :
4431 74384353 : SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4432 : {
4433 37192563 : if ((flag & ~AT_REMOVEDIR) != 0)
4434 : return -EINVAL;
4435 :
4436 37192563 : if (flag & AT_REMOVEDIR)
4437 1473245 : return do_rmdir(dfd, getname(pathname));
4438 35719318 : return do_unlinkat(dfd, getname(pathname));
4439 : }
4440 :
4441 0 : SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4442 : {
4443 0 : return do_unlinkat(AT_FDCWD, getname(pathname));
4444 : }
4445 :
4446 : /**
4447 : * vfs_symlink - create symlink
4448 : * @idmap: idmap of the mount the inode was found from
4449 : * @dir: inode of @dentry
4450 : * @dentry: pointer to dentry of the base directory
4451 : * @oldname: name of the file to link to
4452 : *
4453 : * Create a symlink.
4454 : *
4455 : * If the inode has been found through an idmapped mount the idmap of
4456 : * the vfsmount must be passed through @idmap. This function will then take
4457 : * care to map the inode according to @idmap before checking permissions.
4458 : * On non-idmapped mounts or if permission checking is to be performed on the
4459 : * raw inode simply passs @nop_mnt_idmap.
4460 : */
4461 238382916 : int vfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
4462 : struct dentry *dentry, const char *oldname)
4463 : {
4464 238382916 : int error;
4465 :
4466 238382916 : error = may_create(idmap, dir, dentry);
4467 238382388 : if (error)
4468 : return error;
4469 :
4470 238382382 : if (!dir->i_op->symlink)
4471 : return -EPERM;
4472 :
4473 238382382 : error = security_inode_symlink(dir, dentry, oldname);
4474 238382382 : if (error)
4475 : return error;
4476 :
4477 238382382 : error = dir->i_op->symlink(idmap, dir, dentry, oldname);
4478 238348089 : if (!error)
4479 27635376 : fsnotify_create(dir, dentry);
4480 : return error;
4481 : }
4482 : EXPORT_SYMBOL(vfs_symlink);
4483 :
4484 238753100 : int do_symlinkat(struct filename *from, int newdfd, struct filename *to)
4485 : {
4486 238753100 : int error;
4487 238753100 : struct dentry *dentry;
4488 238753100 : struct path path;
4489 238753100 : unsigned int lookup_flags = 0;
4490 :
4491 238753100 : if (IS_ERR(from)) {
4492 26917 : error = PTR_ERR(from);
4493 26917 : goto out_putnames;
4494 : }
4495 238726183 : retry:
4496 238713995 : dentry = filename_create(newdfd, to, &path, lookup_flags);
4497 238786895 : error = PTR_ERR(dentry);
4498 238786895 : if (IS_ERR(dentry))
4499 410765 : goto out_putnames;
4500 :
4501 238376130 : error = security_path_symlink(&path, dentry, from->name);
4502 238376130 : if (!error)
4503 238376130 : error = vfs_symlink(mnt_idmap(path.mnt), path.dentry->d_inode,
4504 : dentry, from->name);
4505 238360272 : done_path_create(&path, dentry);
4506 476760110 : if (retry_estale(error, lookup_flags)) {
4507 0 : lookup_flags |= LOOKUP_REVAL;
4508 0 : goto retry;
4509 : }
4510 238392243 : out_putnames:
4511 238829925 : putname(to);
4512 238829475 : putname(from);
4513 238828767 : return error;
4514 : }
4515 :
4516 477592857 : SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4517 : int, newdfd, const char __user *, newname)
4518 : {
4519 477602355 : return do_symlinkat(getname(oldname), newdfd, getname(newname));
4520 : }
4521 :
4522 0 : SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4523 : {
4524 0 : return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname));
4525 : }
4526 :
4527 : /**
4528 : * vfs_link - create a new link
4529 : * @old_dentry: object to be linked
4530 : * @idmap: idmap of the mount
4531 : * @dir: new parent
4532 : * @new_dentry: where to create the new link
4533 : * @delegated_inode: returns inode needing a delegation break
4534 : *
4535 : * The caller must hold dir->i_mutex
4536 : *
4537 : * If vfs_link discovers a delegation on the to-be-linked file in need
4538 : * of breaking, it will return -EWOULDBLOCK and return a reference to the
4539 : * inode in delegated_inode. The caller should then break the delegation
4540 : * and retry. Because breaking a delegation may take a long time, the
4541 : * caller should drop the i_mutex before doing so.
4542 : *
4543 : * Alternatively, a caller may pass NULL for delegated_inode. This may
4544 : * be appropriate for callers that expect the underlying filesystem not
4545 : * to be NFS exported.
4546 : *
4547 : * If the inode has been found through an idmapped mount the idmap of
4548 : * the vfsmount must be passed through @idmap. This function will then take
4549 : * care to map the inode according to @idmap before checking permissions.
4550 : * On non-idmapped mounts or if permission checking is to be performed on the
4551 : * raw inode simply passs @nop_mnt_idmap.
4552 : */
4553 3110260 : int vfs_link(struct dentry *old_dentry, struct mnt_idmap *idmap,
4554 : struct inode *dir, struct dentry *new_dentry,
4555 : struct inode **delegated_inode)
4556 : {
4557 3110260 : struct inode *inode = old_dentry->d_inode;
4558 3110260 : unsigned max_links = dir->i_sb->s_max_links;
4559 3110260 : int error;
4560 :
4561 3110260 : if (!inode)
4562 : return -ENOENT;
4563 :
4564 3110260 : error = may_create(idmap, dir, new_dentry);
4565 3110265 : if (error)
4566 : return error;
4567 :
4568 3110259 : if (dir->i_sb != inode->i_sb)
4569 : return -EXDEV;
4570 :
4571 : /*
4572 : * A link to an append-only or immutable file cannot be created.
4573 : */
4574 3110259 : if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4575 : return -EPERM;
4576 : /*
4577 : * Updating the link count will likely cause i_uid and i_gid to
4578 : * be writen back improperly if their true value is unknown to
4579 : * the vfs.
4580 : */
4581 3110253 : if (HAS_UNMAPPED_ID(idmap, inode))
4582 : return -EPERM;
4583 3110260 : if (!dir->i_op->link)
4584 : return -EPERM;
4585 3110260 : if (S_ISDIR(inode->i_mode))
4586 : return -EPERM;
4587 :
4588 3110260 : error = security_inode_link(old_dentry, dir, new_dentry);
4589 3110260 : if (error)
4590 : return error;
4591 :
4592 3110260 : inode_lock(inode);
4593 : /* Make sure we don't allow creating hardlink to an unlinked file */
4594 3110257 : if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4595 : error = -ENOENT;
4596 3110161 : else if (max_links && inode->i_nlink >= max_links)
4597 : error = -EMLINK;
4598 : else {
4599 3110161 : error = try_break_deleg(inode, delegated_inode);
4600 3110155 : if (!error)
4601 3110165 : error = dir->i_op->link(old_dentry, dir, new_dentry);
4602 : }
4603 :
4604 3110156 : if (!error && (inode->i_state & I_LINKABLE)) {
4605 8208 : spin_lock(&inode->i_lock);
4606 8208 : inode->i_state &= ~I_LINKABLE;
4607 8208 : spin_unlock(&inode->i_lock);
4608 : }
4609 3110252 : inode_unlock(inode);
4610 3110259 : if (!error)
4611 3109997 : fsnotify_link(dir, inode, new_dentry);
4612 : return error;
4613 : }
4614 : EXPORT_SYMBOL(vfs_link);
4615 :
4616 : /*
4617 : * Hardlinks are often used in delicate situations. We avoid
4618 : * security-related surprises by not following symlinks on the
4619 : * newname. --KAB
4620 : *
4621 : * We don't follow them on the oldname either to be compatible
4622 : * with linux 2.0, and to avoid hard-linking to directories
4623 : * and other special files. --ADM
4624 : */
4625 3179591 : int do_linkat(int olddfd, struct filename *old, int newdfd,
4626 : struct filename *new, int flags)
4627 : {
4628 3179591 : struct mnt_idmap *idmap;
4629 3179591 : struct dentry *new_dentry;
4630 3179591 : struct path old_path, new_path;
4631 3179591 : struct inode *delegated_inode = NULL;
4632 3179591 : int how = 0;
4633 3179591 : int error;
4634 :
4635 3179591 : if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) {
4636 0 : error = -EINVAL;
4637 0 : goto out_putnames;
4638 : }
4639 : /*
4640 : * To use null names we require CAP_DAC_READ_SEARCH
4641 : * This ensures that not everyone will be able to create
4642 : * handlink using the passed filedescriptor.
4643 : */
4644 3179591 : if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) {
4645 0 : error = -ENOENT;
4646 0 : goto out_putnames;
4647 : }
4648 :
4649 3179591 : if (flags & AT_SYMLINK_FOLLOW)
4650 600030 : how |= LOOKUP_FOLLOW;
4651 2579561 : retry:
4652 3179591 : error = filename_lookup(olddfd, old, how, &old_path, NULL);
4653 3179619 : if (error)
4654 4364 : goto out_putnames;
4655 :
4656 3175255 : new_dentry = filename_create(newdfd, new, &new_path,
4657 : (how & LOOKUP_REVAL));
4658 3175249 : error = PTR_ERR(new_dentry);
4659 3175249 : if (IS_ERR(new_dentry))
4660 71586 : goto out_putpath;
4661 :
4662 3103663 : error = -EXDEV;
4663 3103663 : if (old_path.mnt != new_path.mnt)
4664 4 : goto out_dput;
4665 3103659 : idmap = mnt_idmap(new_path.mnt);
4666 3103657 : error = may_linkat(idmap, &old_path);
4667 3103653 : if (unlikely(error))
4668 1628 : goto out_dput;
4669 3102025 : error = security_path_link(old_path.dentry, &new_path, new_dentry);
4670 3102025 : if (error)
4671 : goto out_dput;
4672 3102025 : error = vfs_link(old_path.dentry, idmap, new_path.dentry->d_inode,
4673 : new_dentry, &delegated_inode);
4674 3103674 : out_dput:
4675 3103674 : done_path_create(&new_path, new_dentry);
4676 3103658 : if (delegated_inode) {
4677 0 : error = break_deleg_wait(&delegated_inode);
4678 0 : if (!error) {
4679 0 : path_put(&old_path);
4680 0 : goto retry;
4681 : }
4682 : }
4683 6207316 : if (retry_estale(error, how)) {
4684 0 : path_put(&old_path);
4685 0 : how |= LOOKUP_REVAL;
4686 0 : goto retry;
4687 : }
4688 3103674 : out_putpath:
4689 3175260 : path_put(&old_path);
4690 3179621 : out_putnames:
4691 3179621 : putname(old);
4692 3179621 : putname(new);
4693 :
4694 3179611 : return error;
4695 : }
4696 :
4697 6359201 : SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4698 : int, newdfd, const char __user *, newname, int, flags)
4699 : {
4700 6359210 : return do_linkat(olddfd, getname_uflags(oldname, flags),
4701 : newdfd, getname(newname), flags);
4702 : }
4703 :
4704 0 : SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4705 : {
4706 0 : return do_linkat(AT_FDCWD, getname(oldname), AT_FDCWD, getname(newname), 0);
4707 : }
4708 :
4709 : /**
4710 : * vfs_rename - rename a filesystem object
4711 : * @rd: pointer to &struct renamedata info
4712 : *
4713 : * The caller must hold multiple mutexes--see lock_rename()).
4714 : *
4715 : * If vfs_rename discovers a delegation in need of breaking at either
4716 : * the source or destination, it will return -EWOULDBLOCK and return a
4717 : * reference to the inode in delegated_inode. The caller should then
4718 : * break the delegation and retry. Because breaking a delegation may
4719 : * take a long time, the caller should drop all locks before doing
4720 : * so.
4721 : *
4722 : * Alternatively, a caller may pass NULL for delegated_inode. This may
4723 : * be appropriate for callers that expect the underlying filesystem not
4724 : * to be NFS exported.
4725 : *
4726 : * The worst of all namespace operations - renaming directory. "Perverted"
4727 : * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4728 : * Problems:
4729 : *
4730 : * a) we can get into loop creation.
4731 : * b) race potential - two innocent renames can create a loop together.
4732 : * That's where 4.4 screws up. Current fix: serialization on
4733 : * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4734 : * story.
4735 : * c) we have to lock _four_ objects - parents and victim (if it exists),
4736 : * and source.
4737 : * And that - after we got ->i_mutex on parents (until then we don't know
4738 : * whether the target exists). Solution: try to be smart with locking
4739 : * order for inodes. We rely on the fact that tree topology may change
4740 : * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4741 : * move will be locked. Thus we can rank directories by the tree
4742 : * (ancestors first) and rank all non-directories after them.
4743 : * That works since everybody except rename does "lock parent, lookup,
4744 : * lock child" and rename is under ->s_vfs_rename_mutex.
4745 : * HOWEVER, it relies on the assumption that any object with ->lookup()
4746 : * has no more than 1 dentry. If "hybrid" objects will ever appear,
4747 : * we'd better make sure that there's no link(2) for them.
4748 : * d) conversion from fhandle to dentry may come in the wrong moment - when
4749 : * we are removing the target. Solution: we will have to grab ->i_mutex
4750 : * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4751 : * ->i_mutex on parents, which works but leads to some truly excessive
4752 : * locking].
4753 : */
4754 23255970 : int vfs_rename(struct renamedata *rd)
4755 : {
4756 23255970 : int error;
4757 23255970 : struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir;
4758 23255970 : struct dentry *old_dentry = rd->old_dentry;
4759 23255970 : struct dentry *new_dentry = rd->new_dentry;
4760 23255970 : struct inode **delegated_inode = rd->delegated_inode;
4761 23255970 : unsigned int flags = rd->flags;
4762 23255970 : bool is_dir = d_is_dir(old_dentry);
4763 23255970 : struct inode *source = old_dentry->d_inode;
4764 23255970 : struct inode *target = new_dentry->d_inode;
4765 23255970 : bool new_is_dir = false;
4766 23255970 : unsigned max_links = new_dir->i_sb->s_max_links;
4767 23255970 : struct name_snapshot old_name;
4768 :
4769 23255970 : if (source == target)
4770 : return 0;
4771 :
4772 23169314 : error = may_delete(rd->old_mnt_idmap, old_dir, old_dentry, is_dir);
4773 23169337 : if (error)
4774 : return error;
4775 :
4776 23169319 : if (!target) {
4777 15873220 : error = may_create(rd->new_mnt_idmap, new_dir, new_dentry);
4778 : } else {
4779 7296099 : new_is_dir = d_is_dir(new_dentry);
4780 :
4781 7296099 : if (!(flags & RENAME_EXCHANGE))
4782 583451 : error = may_delete(rd->new_mnt_idmap, new_dir,
4783 : new_dentry, is_dir);
4784 : else
4785 6712648 : error = may_delete(rd->new_mnt_idmap, new_dir,
4786 : new_dentry, new_is_dir);
4787 : }
4788 23169315 : if (error)
4789 : return error;
4790 :
4791 23166865 : if (!old_dir->i_op->rename)
4792 : return -EPERM;
4793 :
4794 : /*
4795 : * If we are going to change the parent - check write permissions,
4796 : * we'll need to flip '..'.
4797 : */
4798 23166865 : if (new_dir != old_dir) {
4799 21332699 : if (is_dir) {
4800 6724569 : error = inode_permission(rd->old_mnt_idmap, source,
4801 : MAY_WRITE);
4802 6724569 : if (error)
4803 : return error;
4804 : }
4805 21332699 : if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4806 2119438 : error = inode_permission(rd->new_mnt_idmap, target,
4807 : MAY_WRITE);
4808 2119438 : if (error)
4809 : return error;
4810 : }
4811 : }
4812 :
4813 23166865 : error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4814 : flags);
4815 23166865 : if (error)
4816 : return error;
4817 :
4818 23166865 : take_dentry_name_snapshot(&old_name, old_dentry);
4819 23166863 : dget(new_dentry);
4820 : /*
4821 : * Lock all moved children. Moved directories may need to change parent
4822 : * pointer so they need the lock to prevent against concurrent
4823 : * directory changes moving parent pointer. For regular files we've
4824 : * historically always done this. The lockdep locking subclasses are
4825 : * somewhat arbitrary but RENAME_EXCHANGE in particular can swap
4826 : * regular files and directories so it's difficult to tell which
4827 : * subclasses to use.
4828 : */
4829 23166862 : lock_two_inodes(source, target, I_MUTEX_NORMAL, I_MUTEX_NONDIR2);
4830 :
4831 23166865 : error = -EPERM;
4832 23166865 : if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target)))
4833 0 : goto out;
4834 :
4835 23166865 : error = -EBUSY;
4836 23166865 : if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4837 0 : goto out;
4838 :
4839 23166864 : if (max_links && new_dir != old_dir) {
4840 21332697 : error = -EMLINK;
4841 21332697 : if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4842 0 : goto out;
4843 21332697 : if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4844 8 : old_dir->i_nlink >= max_links)
4845 0 : goto out;
4846 : }
4847 23166864 : if (!is_dir) {
4848 16426089 : error = try_break_deleg(source, delegated_inode);
4849 16426088 : if (error)
4850 0 : goto out;
4851 : }
4852 23166863 : if (target && !new_is_dir) {
4853 5169694 : error = try_break_deleg(target, delegated_inode);
4854 5169695 : if (error)
4855 0 : goto out;
4856 : }
4857 23166864 : error = old_dir->i_op->rename(rd->new_mnt_idmap, old_dir, old_dentry,
4858 : new_dir, new_dentry, flags);
4859 23166871 : if (error)
4860 39118 : goto out;
4861 :
4862 23127753 : if (!(flags & RENAME_EXCHANGE) && target) {
4863 580260 : if (is_dir) {
4864 284 : shrink_dcache_parent(new_dentry);
4865 284 : target->i_flags |= S_DEAD;
4866 : }
4867 580260 : dont_mount(new_dentry);
4868 580260 : detach_mounts(new_dentry);
4869 : }
4870 23127753 : if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4871 23127753 : if (!(flags & RENAME_EXCHANGE))
4872 16415535 : d_move(old_dentry, new_dentry);
4873 : else
4874 6712218 : d_exchange(old_dentry, new_dentry);
4875 : }
4876 0 : out:
4877 23166872 : inode_unlock(source);
4878 23166870 : if (target)
4879 7293656 : inode_unlock(target);
4880 23166870 : dput(new_dentry);
4881 23166868 : if (!error) {
4882 23127754 : fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4883 23127754 : !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4884 23127754 : if (flags & RENAME_EXCHANGE) {
4885 6712218 : fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4886 : new_is_dir, NULL, new_dentry);
4887 : }
4888 : }
4889 23166868 : release_dentry_name_snapshot(&old_name);
4890 :
4891 23166868 : return error;
4892 : }
4893 : EXPORT_SYMBOL(vfs_rename);
4894 :
4895 23508526 : int do_renameat2(int olddfd, struct filename *from, int newdfd,
4896 : struct filename *to, unsigned int flags)
4897 : {
4898 23508526 : struct renamedata rd;
4899 23508526 : struct dentry *old_dentry, *new_dentry;
4900 23508526 : struct dentry *trap;
4901 23508526 : struct path old_path, new_path;
4902 23508526 : struct qstr old_last, new_last;
4903 23508526 : int old_type, new_type;
4904 23508526 : struct inode *delegated_inode = NULL;
4905 23508526 : unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4906 23508526 : bool should_retry = false;
4907 23508526 : int error = -EINVAL;
4908 :
4909 23508526 : if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4910 0 : goto put_names;
4911 :
4912 23508526 : if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4913 8489754 : (flags & RENAME_EXCHANGE))
4914 0 : goto put_names;
4915 :
4916 23508526 : if (flags & RENAME_EXCHANGE)
4917 6798171 : target_flags = 0;
4918 :
4919 23508526 : retry:
4920 23508526 : error = filename_parentat(olddfd, from, lookup_flags, &old_path,
4921 : &old_last, &old_type);
4922 23510594 : if (error)
4923 2 : goto put_names;
4924 :
4925 23510592 : error = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last,
4926 : &new_type);
4927 23508582 : if (error)
4928 4000 : goto exit1;
4929 :
4930 23504582 : error = -EXDEV;
4931 23504582 : if (old_path.mnt != new_path.mnt)
4932 20 : goto exit2;
4933 :
4934 23504562 : error = -EBUSY;
4935 23504562 : if (old_type != LAST_NORM)
4936 0 : goto exit2;
4937 :
4938 23504562 : if (flags & RENAME_NOREPLACE)
4939 7033901 : error = -EEXIST;
4940 23504562 : if (new_type != LAST_NORM)
4941 0 : goto exit2;
4942 :
4943 23504562 : error = mnt_want_write(old_path.mnt);
4944 23506919 : if (error)
4945 0 : goto exit2;
4946 :
4947 23506919 : retry_deleg:
4948 23506919 : trap = lock_rename(new_path.dentry, old_path.dentry);
4949 :
4950 23507032 : old_dentry = lookup_one_qstr_excl(&old_last, old_path.dentry,
4951 : lookup_flags);
4952 23507027 : error = PTR_ERR(old_dentry);
4953 23507027 : if (IS_ERR(old_dentry))
4954 26 : goto exit3;
4955 : /* source must exist */
4956 23507001 : error = -ENOENT;
4957 23507001 : if (d_is_negative(old_dentry))
4958 3320 : goto exit4;
4959 23503681 : new_dentry = lookup_one_qstr_excl(&new_last, new_path.dentry,
4960 : lookup_flags | target_flags);
4961 23503687 : error = PTR_ERR(new_dentry);
4962 23503687 : if (IS_ERR(new_dentry))
4963 1076 : goto exit4;
4964 23502611 : error = -EEXIST;
4965 23502611 : if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4966 151253 : goto exit5;
4967 23351358 : if (flags & RENAME_EXCHANGE) {
4968 6799067 : error = -ENOENT;
4969 6799067 : if (d_is_negative(new_dentry))
4970 16 : goto exit5;
4971 :
4972 11466869 : if (!d_is_dir(new_dentry)) {
4973 4667818 : error = -ENOTDIR;
4974 4667818 : if (new_last.name[new_last.len])
4975 0 : goto exit5;
4976 : }
4977 : }
4978 : /* unless the source is a directory trailing slashes give -ENOTDIR */
4979 39816730 : if (!d_is_dir(old_dentry)) {
4980 16465389 : error = -ENOTDIR;
4981 16465389 : if (old_last.name[old_last.len])
4982 0 : goto exit5;
4983 16465389 : if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4984 0 : goto exit5;
4985 : }
4986 : /* source should not be ancestor of target */
4987 23351342 : error = -EINVAL;
4988 23351342 : if (old_dentry == trap)
4989 144822 : goto exit5;
4990 : /* target should not be an ancestor of source */
4991 23206520 : if (!(flags & RENAME_EXCHANGE))
4992 16407479 : error = -ENOTEMPTY;
4993 23206520 : if (new_dentry == trap)
4994 0 : goto exit5;
4995 :
4996 23206520 : error = security_path_rename(&old_path, old_dentry,
4997 : &new_path, new_dentry, flags);
4998 23206520 : if (error)
4999 : goto exit5;
5000 :
5001 23206520 : rd.old_dir = old_path.dentry->d_inode;
5002 23206520 : rd.old_dentry = old_dentry;
5003 23206520 : rd.old_mnt_idmap = mnt_idmap(old_path.mnt);
5004 23206524 : rd.new_dir = new_path.dentry->d_inode;
5005 23206524 : rd.new_dentry = new_dentry;
5006 23206524 : rd.new_mnt_idmap = mnt_idmap(new_path.mnt);
5007 23206525 : rd.delegated_inode = &delegated_inode;
5008 23206525 : rd.flags = flags;
5009 23206525 : error = vfs_rename(&rd);
5010 23502382 : exit5:
5011 23502382 : dput(new_dentry);
5012 23506988 : exit4:
5013 23506988 : dput(old_dentry);
5014 23507022 : exit3:
5015 23507022 : unlock_rename(new_path.dentry, old_path.dentry);
5016 23506990 : if (delegated_inode) {
5017 0 : error = break_deleg_wait(&delegated_inode);
5018 0 : if (!error)
5019 0 : goto retry_deleg;
5020 : }
5021 23506990 : mnt_drop_write(old_path.mnt);
5022 23506634 : exit2:
5023 23506654 : if (retry_estale(error, lookup_flags))
5024 0 : should_retry = true;
5025 23506654 : path_put(&new_path);
5026 23510033 : exit1:
5027 23510033 : path_put(&old_path);
5028 23511050 : if (should_retry) {
5029 0 : should_retry = false;
5030 0 : lookup_flags |= LOOKUP_REVAL;
5031 0 : goto retry;
5032 : }
5033 23511050 : put_names:
5034 23511052 : putname(from);
5035 23510969 : putname(to);
5036 23511037 : return error;
5037 : }
5038 :
5039 30575946 : SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
5040 : int, newdfd, const char __user *, newname, unsigned int, flags)
5041 : {
5042 30576941 : return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
5043 : flags);
5044 : }
5045 :
5046 16442443 : SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
5047 : int, newdfd, const char __user *, newname)
5048 : {
5049 16442922 : return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
5050 : 0);
5051 : }
5052 :
5053 0 : SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5054 : {
5055 0 : return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD,
5056 : getname(newname), 0);
5057 : }
5058 :
5059 267808962 : int readlink_copy(char __user *buffer, int buflen, const char *link)
5060 : {
5061 267808962 : int len = PTR_ERR(link);
5062 267808962 : if (IS_ERR(link))
5063 0 : goto out;
5064 :
5065 267808962 : len = strlen(link);
5066 267808962 : if (len > (unsigned) buflen)
5067 8 : len = buflen;
5068 535693094 : if (copy_to_user(buffer, link, len))
5069 0 : len = -EFAULT;
5070 267884132 : out:
5071 267884132 : return len;
5072 : }
5073 :
5074 : /**
5075 : * vfs_readlink - copy symlink body into userspace buffer
5076 : * @dentry: dentry on which to get symbolic link
5077 : * @buffer: user memory pointer
5078 : * @buflen: size of buffer
5079 : *
5080 : * Does not touch atime. That's up to the caller if necessary
5081 : *
5082 : * Does not call security hook.
5083 : */
5084 267912436 : int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5085 : {
5086 267912436 : struct inode *inode = d_inode(dentry);
5087 267912436 : DEFINE_DELAYED_CALL(done);
5088 267912436 : const char *link;
5089 267912436 : int res;
5090 :
5091 267912436 : if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
5092 24895870 : if (unlikely(inode->i_op->readlink))
5093 137990 : return inode->i_op->readlink(dentry, buffer, buflen);
5094 :
5095 24757880 : if (!d_is_symlink(dentry))
5096 : return -EINVAL;
5097 :
5098 24757880 : spin_lock(&inode->i_lock);
5099 24759135 : inode->i_opflags |= IOP_DEFAULT_READLINK;
5100 24759135 : spin_unlock(&inode->i_lock);
5101 : }
5102 :
5103 267775643 : link = READ_ONCE(inode->i_link);
5104 267775643 : if (!link) {
5105 267423301 : link = inode->i_op->get_link(dentry, inode, &done);
5106 267444835 : if (IS_ERR(link))
5107 33 : return PTR_ERR(link);
5108 : }
5109 267797144 : res = readlink_copy(buffer, buflen, link);
5110 267847843 : do_delayed_call(&done);
5111 : return res;
5112 : }
5113 : EXPORT_SYMBOL(vfs_readlink);
5114 :
5115 : /**
5116 : * vfs_get_link - get symlink body
5117 : * @dentry: dentry on which to get symbolic link
5118 : * @done: caller needs to free returned data with this
5119 : *
5120 : * Calls security hook and i_op->get_link() on the supplied inode.
5121 : *
5122 : * It does not touch atime. That's up to the caller if necessary.
5123 : *
5124 : * Does not work on "special" symlinks like /proc/$$/fd/N
5125 : */
5126 5643883 : const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
5127 : {
5128 5643883 : const char *res = ERR_PTR(-EINVAL);
5129 5643883 : struct inode *inode = d_inode(dentry);
5130 :
5131 5643883 : if (d_is_symlink(dentry)) {
5132 5643883 : res = ERR_PTR(security_inode_readlink(dentry));
5133 5643883 : if (!res)
5134 5643883 : res = inode->i_op->get_link(dentry, inode, done);
5135 : }
5136 5643877 : return res;
5137 : }
5138 : EXPORT_SYMBOL(vfs_get_link);
5139 :
5140 : /* get the link contents into pagecache */
5141 22 : const char *page_get_link(struct dentry *dentry, struct inode *inode,
5142 : struct delayed_call *callback)
5143 : {
5144 22 : char *kaddr;
5145 22 : struct page *page;
5146 22 : struct address_space *mapping = inode->i_mapping;
5147 :
5148 22 : if (!dentry) {
5149 1 : page = find_get_page(mapping, 0);
5150 1 : if (!page)
5151 : return ERR_PTR(-ECHILD);
5152 1 : if (!PageUptodate(page)) {
5153 1 : put_page(page);
5154 1 : return ERR_PTR(-ECHILD);
5155 : }
5156 : } else {
5157 21 : page = read_mapping_page(mapping, 0, NULL);
5158 21 : if (IS_ERR(page))
5159 : return (char*)page;
5160 : }
5161 21 : set_delayed_call(callback, page_put_link, page);
5162 21 : BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
5163 21 : kaddr = page_address(page);
5164 21 : nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
5165 21 : return kaddr;
5166 : }
5167 :
5168 : EXPORT_SYMBOL(page_get_link);
5169 :
5170 87905590 : void page_put_link(void *arg)
5171 : {
5172 87905590 : put_page(arg);
5173 87907082 : }
5174 : EXPORT_SYMBOL(page_put_link);
5175 :
5176 0 : int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5177 : {
5178 0 : DEFINE_DELAYED_CALL(done);
5179 0 : int res = readlink_copy(buffer, buflen,
5180 : page_get_link(dentry, d_inode(dentry),
5181 : &done));
5182 0 : do_delayed_call(&done);
5183 0 : return res;
5184 : }
5185 : EXPORT_SYMBOL(page_readlink);
5186 :
5187 11 : int page_symlink(struct inode *inode, const char *symname, int len)
5188 : {
5189 11 : struct address_space *mapping = inode->i_mapping;
5190 11 : const struct address_space_operations *aops = mapping->a_ops;
5191 11 : bool nofs = !mapping_gfp_constraint(mapping, __GFP_FS);
5192 11 : struct page *page;
5193 11 : void *fsdata = NULL;
5194 11 : int err;
5195 11 : unsigned int flags;
5196 :
5197 11 : retry:
5198 11 : if (nofs)
5199 0 : flags = memalloc_nofs_save();
5200 11 : err = aops->write_begin(NULL, mapping, 0, len-1, &page, &fsdata);
5201 11 : if (nofs)
5202 0 : memalloc_nofs_restore(flags);
5203 11 : if (err)
5204 0 : goto fail;
5205 :
5206 22 : memcpy(page_address(page), symname, len-1);
5207 :
5208 11 : err = aops->write_end(NULL, mapping, 0, len-1, len-1,
5209 : page, fsdata);
5210 11 : if (err < 0)
5211 0 : goto fail;
5212 11 : if (err < len-1)
5213 0 : goto retry;
5214 :
5215 11 : mark_inode_dirty(inode);
5216 11 : return 0;
5217 : fail:
5218 : return err;
5219 : }
5220 : EXPORT_SYMBOL(page_symlink);
5221 :
5222 : const struct inode_operations page_symlink_inode_operations = {
5223 : .get_link = page_get_link,
5224 : };
5225 : EXPORT_SYMBOL(page_symlink_inode_operations);
|
