Loading...
1/*
2 * linux/fs/open.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7#include <linux/string.h>
8#include <linux/mm.h>
9#include <linux/file.h>
10#include <linux/fdtable.h>
11#include <linux/fsnotify.h>
12#include <linux/module.h>
13#include <linux/tty.h>
14#include <linux/namei.h>
15#include <linux/backing-dev.h>
16#include <linux/capability.h>
17#include <linux/securebits.h>
18#include <linux/security.h>
19#include <linux/mount.h>
20#include <linux/fcntl.h>
21#include <linux/slab.h>
22#include <asm/uaccess.h>
23#include <linux/fs.h>
24#include <linux/personality.h>
25#include <linux/pagemap.h>
26#include <linux/syscalls.h>
27#include <linux/rcupdate.h>
28#include <linux/audit.h>
29#include <linux/falloc.h>
30#include <linux/fs_struct.h>
31#include <linux/ima.h>
32#include <linux/dnotify.h>
33#include <linux/compat.h>
34
35#include "internal.h"
36
37int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
38 struct file *filp)
39{
40 int ret;
41 struct iattr newattrs;
42
43 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
44 if (length < 0)
45 return -EINVAL;
46
47 newattrs.ia_size = length;
48 newattrs.ia_valid = ATTR_SIZE | time_attrs;
49 if (filp) {
50 newattrs.ia_file = filp;
51 newattrs.ia_valid |= ATTR_FILE;
52 }
53
54 /* Remove suid/sgid on truncate too */
55 ret = should_remove_suid(dentry);
56 if (ret)
57 newattrs.ia_valid |= ret | ATTR_FORCE;
58
59 mutex_lock(&dentry->d_inode->i_mutex);
60 /* Note any delegations or leases have already been broken: */
61 ret = notify_change(dentry, &newattrs, NULL);
62 mutex_unlock(&dentry->d_inode->i_mutex);
63 return ret;
64}
65
66long vfs_truncate(struct path *path, loff_t length)
67{
68 struct inode *inode;
69 long error;
70
71 inode = path->dentry->d_inode;
72
73 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
74 if (S_ISDIR(inode->i_mode))
75 return -EISDIR;
76 if (!S_ISREG(inode->i_mode))
77 return -EINVAL;
78
79 error = mnt_want_write(path->mnt);
80 if (error)
81 goto out;
82
83 error = inode_permission(inode, MAY_WRITE);
84 if (error)
85 goto mnt_drop_write_and_out;
86
87 error = -EPERM;
88 if (IS_APPEND(inode))
89 goto mnt_drop_write_and_out;
90
91 error = get_write_access(inode);
92 if (error)
93 goto mnt_drop_write_and_out;
94
95 /*
96 * Make sure that there are no leases. get_write_access() protects
97 * against the truncate racing with a lease-granting setlease().
98 */
99 error = break_lease(inode, O_WRONLY);
100 if (error)
101 goto put_write_and_out;
102
103 error = locks_verify_truncate(inode, NULL, length);
104 if (!error)
105 error = security_path_truncate(path);
106 if (!error)
107 error = do_truncate(path->dentry, length, 0, NULL);
108
109put_write_and_out:
110 put_write_access(inode);
111mnt_drop_write_and_out:
112 mnt_drop_write(path->mnt);
113out:
114 return error;
115}
116EXPORT_SYMBOL_GPL(vfs_truncate);
117
118static long do_sys_truncate(const char __user *pathname, loff_t length)
119{
120 unsigned int lookup_flags = LOOKUP_FOLLOW;
121 struct path path;
122 int error;
123
124 if (length < 0) /* sorry, but loff_t says... */
125 return -EINVAL;
126
127retry:
128 error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
129 if (!error) {
130 error = vfs_truncate(&path, length);
131 path_put(&path);
132 }
133 if (retry_estale(error, lookup_flags)) {
134 lookup_flags |= LOOKUP_REVAL;
135 goto retry;
136 }
137 return error;
138}
139
140SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
141{
142 return do_sys_truncate(path, length);
143}
144
145#ifdef CONFIG_COMPAT
146COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
147{
148 return do_sys_truncate(path, length);
149}
150#endif
151
152static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
153{
154 struct inode *inode;
155 struct dentry *dentry;
156 struct fd f;
157 int error;
158
159 error = -EINVAL;
160 if (length < 0)
161 goto out;
162 error = -EBADF;
163 f = fdget(fd);
164 if (!f.file)
165 goto out;
166
167 /* explicitly opened as large or we are on 64-bit box */
168 if (f.file->f_flags & O_LARGEFILE)
169 small = 0;
170
171 dentry = f.file->f_path.dentry;
172 inode = dentry->d_inode;
173 error = -EINVAL;
174 if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
175 goto out_putf;
176
177 error = -EINVAL;
178 /* Cannot ftruncate over 2^31 bytes without large file support */
179 if (small && length > MAX_NON_LFS)
180 goto out_putf;
181
182 error = -EPERM;
183 if (IS_APPEND(inode))
184 goto out_putf;
185
186 sb_start_write(inode->i_sb);
187 error = locks_verify_truncate(inode, f.file, length);
188 if (!error)
189 error = security_path_truncate(&f.file->f_path);
190 if (!error)
191 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
192 sb_end_write(inode->i_sb);
193out_putf:
194 fdput(f);
195out:
196 return error;
197}
198
199SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
200{
201 return do_sys_ftruncate(fd, length, 1);
202}
203
204#ifdef CONFIG_COMPAT
205COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
206{
207 return do_sys_ftruncate(fd, length, 1);
208}
209#endif
210
211/* LFS versions of truncate are only needed on 32 bit machines */
212#if BITS_PER_LONG == 32
213SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
214{
215 return do_sys_truncate(path, length);
216}
217
218SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
219{
220 return do_sys_ftruncate(fd, length, 0);
221}
222#endif /* BITS_PER_LONG == 32 */
223
224
225int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
226{
227 struct inode *inode = file_inode(file);
228 long ret;
229
230 if (offset < 0 || len <= 0)
231 return -EINVAL;
232
233 /* Return error if mode is not supported */
234 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
235 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE))
236 return -EOPNOTSUPP;
237
238 /* Punch hole and zero range are mutually exclusive */
239 if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
240 (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
241 return -EOPNOTSUPP;
242
243 /* Punch hole must have keep size set */
244 if ((mode & FALLOC_FL_PUNCH_HOLE) &&
245 !(mode & FALLOC_FL_KEEP_SIZE))
246 return -EOPNOTSUPP;
247
248 /* Collapse range should only be used exclusively. */
249 if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
250 (mode & ~FALLOC_FL_COLLAPSE_RANGE))
251 return -EINVAL;
252
253 if (!(file->f_mode & FMODE_WRITE))
254 return -EBADF;
255
256 /*
257 * We can only allow pure fallocate on append only files
258 */
259 if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
260 return -EPERM;
261
262 if (IS_IMMUTABLE(inode))
263 return -EPERM;
264
265 /*
266 * We can not allow to do any fallocate operation on an active
267 * swapfile
268 */
269 if (IS_SWAPFILE(inode))
270 ret = -ETXTBSY;
271
272 /*
273 * Revalidate the write permissions, in case security policy has
274 * changed since the files were opened.
275 */
276 ret = security_file_permission(file, MAY_WRITE);
277 if (ret)
278 return ret;
279
280 if (S_ISFIFO(inode->i_mode))
281 return -ESPIPE;
282
283 /*
284 * Let individual file system decide if it supports preallocation
285 * for directories or not.
286 */
287 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
288 return -ENODEV;
289
290 /* Check for wrap through zero too */
291 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
292 return -EFBIG;
293
294 if (!file->f_op->fallocate)
295 return -EOPNOTSUPP;
296
297 sb_start_write(inode->i_sb);
298 ret = file->f_op->fallocate(file, mode, offset, len);
299 sb_end_write(inode->i_sb);
300 return ret;
301}
302
303SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
304{
305 struct fd f = fdget(fd);
306 int error = -EBADF;
307
308 if (f.file) {
309 error = do_fallocate(f.file, mode, offset, len);
310 fdput(f);
311 }
312 return error;
313}
314
315/*
316 * access() needs to use the real uid/gid, not the effective uid/gid.
317 * We do this by temporarily clearing all FS-related capabilities and
318 * switching the fsuid/fsgid around to the real ones.
319 */
320SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
321{
322 const struct cred *old_cred;
323 struct cred *override_cred;
324 struct path path;
325 struct inode *inode;
326 int res;
327 unsigned int lookup_flags = LOOKUP_FOLLOW;
328
329 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
330 return -EINVAL;
331
332 override_cred = prepare_creds();
333 if (!override_cred)
334 return -ENOMEM;
335
336 override_cred->fsuid = override_cred->uid;
337 override_cred->fsgid = override_cred->gid;
338
339 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
340 /* Clear the capabilities if we switch to a non-root user */
341 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
342 if (!uid_eq(override_cred->uid, root_uid))
343 cap_clear(override_cred->cap_effective);
344 else
345 override_cred->cap_effective =
346 override_cred->cap_permitted;
347 }
348
349 old_cred = override_creds(override_cred);
350retry:
351 res = user_path_at(dfd, filename, lookup_flags, &path);
352 if (res)
353 goto out;
354
355 inode = path.dentry->d_inode;
356
357 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
358 /*
359 * MAY_EXEC on regular files is denied if the fs is mounted
360 * with the "noexec" flag.
361 */
362 res = -EACCES;
363 if (path.mnt->mnt_flags & MNT_NOEXEC)
364 goto out_path_release;
365 }
366
367 res = inode_permission(inode, mode | MAY_ACCESS);
368 /* SuS v2 requires we report a read only fs too */
369 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
370 goto out_path_release;
371 /*
372 * This is a rare case where using __mnt_is_readonly()
373 * is OK without a mnt_want/drop_write() pair. Since
374 * no actual write to the fs is performed here, we do
375 * not need to telegraph to that to anyone.
376 *
377 * By doing this, we accept that this access is
378 * inherently racy and know that the fs may change
379 * state before we even see this result.
380 */
381 if (__mnt_is_readonly(path.mnt))
382 res = -EROFS;
383
384out_path_release:
385 path_put(&path);
386 if (retry_estale(res, lookup_flags)) {
387 lookup_flags |= LOOKUP_REVAL;
388 goto retry;
389 }
390out:
391 revert_creds(old_cred);
392 put_cred(override_cred);
393 return res;
394}
395
396SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
397{
398 return sys_faccessat(AT_FDCWD, filename, mode);
399}
400
401SYSCALL_DEFINE1(chdir, const char __user *, filename)
402{
403 struct path path;
404 int error;
405 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
406retry:
407 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
408 if (error)
409 goto out;
410
411 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
412 if (error)
413 goto dput_and_out;
414
415 set_fs_pwd(current->fs, &path);
416
417dput_and_out:
418 path_put(&path);
419 if (retry_estale(error, lookup_flags)) {
420 lookup_flags |= LOOKUP_REVAL;
421 goto retry;
422 }
423out:
424 return error;
425}
426
427SYSCALL_DEFINE1(fchdir, unsigned int, fd)
428{
429 struct fd f = fdget_raw(fd);
430 struct inode *inode;
431 int error = -EBADF;
432
433 error = -EBADF;
434 if (!f.file)
435 goto out;
436
437 inode = file_inode(f.file);
438
439 error = -ENOTDIR;
440 if (!S_ISDIR(inode->i_mode))
441 goto out_putf;
442
443 error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
444 if (!error)
445 set_fs_pwd(current->fs, &f.file->f_path);
446out_putf:
447 fdput(f);
448out:
449 return error;
450}
451
452SYSCALL_DEFINE1(chroot, const char __user *, filename)
453{
454 struct path path;
455 int error;
456 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
457retry:
458 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
459 if (error)
460 goto out;
461
462 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
463 if (error)
464 goto dput_and_out;
465
466 error = -EPERM;
467 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
468 goto dput_and_out;
469 error = security_path_chroot(&path);
470 if (error)
471 goto dput_and_out;
472
473 set_fs_root(current->fs, &path);
474 error = 0;
475dput_and_out:
476 path_put(&path);
477 if (retry_estale(error, lookup_flags)) {
478 lookup_flags |= LOOKUP_REVAL;
479 goto retry;
480 }
481out:
482 return error;
483}
484
485static int chmod_common(struct path *path, umode_t mode)
486{
487 struct inode *inode = path->dentry->d_inode;
488 struct inode *delegated_inode = NULL;
489 struct iattr newattrs;
490 int error;
491
492 error = mnt_want_write(path->mnt);
493 if (error)
494 return error;
495retry_deleg:
496 mutex_lock(&inode->i_mutex);
497 error = security_path_chmod(path, mode);
498 if (error)
499 goto out_unlock;
500 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
501 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
502 error = notify_change(path->dentry, &newattrs, &delegated_inode);
503out_unlock:
504 mutex_unlock(&inode->i_mutex);
505 if (delegated_inode) {
506 error = break_deleg_wait(&delegated_inode);
507 if (!error)
508 goto retry_deleg;
509 }
510 mnt_drop_write(path->mnt);
511 return error;
512}
513
514SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
515{
516 struct fd f = fdget(fd);
517 int err = -EBADF;
518
519 if (f.file) {
520 audit_inode(NULL, f.file->f_path.dentry, 0);
521 err = chmod_common(&f.file->f_path, mode);
522 fdput(f);
523 }
524 return err;
525}
526
527SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
528{
529 struct path path;
530 int error;
531 unsigned int lookup_flags = LOOKUP_FOLLOW;
532retry:
533 error = user_path_at(dfd, filename, lookup_flags, &path);
534 if (!error) {
535 error = chmod_common(&path, mode);
536 path_put(&path);
537 if (retry_estale(error, lookup_flags)) {
538 lookup_flags |= LOOKUP_REVAL;
539 goto retry;
540 }
541 }
542 return error;
543}
544
545SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
546{
547 return sys_fchmodat(AT_FDCWD, filename, mode);
548}
549
550static int chown_common(struct path *path, uid_t user, gid_t group)
551{
552 struct inode *inode = path->dentry->d_inode;
553 struct inode *delegated_inode = NULL;
554 int error;
555 struct iattr newattrs;
556 kuid_t uid;
557 kgid_t gid;
558
559 uid = make_kuid(current_user_ns(), user);
560 gid = make_kgid(current_user_ns(), group);
561
562 newattrs.ia_valid = ATTR_CTIME;
563 if (user != (uid_t) -1) {
564 if (!uid_valid(uid))
565 return -EINVAL;
566 newattrs.ia_valid |= ATTR_UID;
567 newattrs.ia_uid = uid;
568 }
569 if (group != (gid_t) -1) {
570 if (!gid_valid(gid))
571 return -EINVAL;
572 newattrs.ia_valid |= ATTR_GID;
573 newattrs.ia_gid = gid;
574 }
575 if (!S_ISDIR(inode->i_mode))
576 newattrs.ia_valid |=
577 ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
578retry_deleg:
579 mutex_lock(&inode->i_mutex);
580 error = security_path_chown(path, uid, gid);
581 if (!error)
582 error = notify_change(path->dentry, &newattrs, &delegated_inode);
583 mutex_unlock(&inode->i_mutex);
584 if (delegated_inode) {
585 error = break_deleg_wait(&delegated_inode);
586 if (!error)
587 goto retry_deleg;
588 }
589 return error;
590}
591
592SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
593 gid_t, group, int, flag)
594{
595 struct path path;
596 int error = -EINVAL;
597 int lookup_flags;
598
599 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
600 goto out;
601
602 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
603 if (flag & AT_EMPTY_PATH)
604 lookup_flags |= LOOKUP_EMPTY;
605retry:
606 error = user_path_at(dfd, filename, lookup_flags, &path);
607 if (error)
608 goto out;
609 error = mnt_want_write(path.mnt);
610 if (error)
611 goto out_release;
612 error = chown_common(&path, user, group);
613 mnt_drop_write(path.mnt);
614out_release:
615 path_put(&path);
616 if (retry_estale(error, lookup_flags)) {
617 lookup_flags |= LOOKUP_REVAL;
618 goto retry;
619 }
620out:
621 return error;
622}
623
624SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
625{
626 return sys_fchownat(AT_FDCWD, filename, user, group, 0);
627}
628
629SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
630{
631 return sys_fchownat(AT_FDCWD, filename, user, group,
632 AT_SYMLINK_NOFOLLOW);
633}
634
635SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
636{
637 struct fd f = fdget(fd);
638 int error = -EBADF;
639
640 if (!f.file)
641 goto out;
642
643 error = mnt_want_write_file(f.file);
644 if (error)
645 goto out_fput;
646 audit_inode(NULL, f.file->f_path.dentry, 0);
647 error = chown_common(&f.file->f_path, user, group);
648 mnt_drop_write_file(f.file);
649out_fput:
650 fdput(f);
651out:
652 return error;
653}
654
655int open_check_o_direct(struct file *f)
656{
657 /* NB: we're sure to have correct a_ops only after f_op->open */
658 if (f->f_flags & O_DIRECT) {
659 if (!f->f_mapping->a_ops ||
660 ((!f->f_mapping->a_ops->direct_IO) &&
661 (!f->f_mapping->a_ops->get_xip_mem))) {
662 return -EINVAL;
663 }
664 }
665 return 0;
666}
667
668static int do_dentry_open(struct file *f,
669 int (*open)(struct inode *, struct file *),
670 const struct cred *cred)
671{
672 static const struct file_operations empty_fops = {};
673 struct inode *inode;
674 int error;
675
676 f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
677 FMODE_PREAD | FMODE_PWRITE;
678
679 path_get(&f->f_path);
680 inode = f->f_inode = f->f_path.dentry->d_inode;
681 f->f_mapping = inode->i_mapping;
682
683 if (unlikely(f->f_flags & O_PATH)) {
684 f->f_mode = FMODE_PATH;
685 f->f_op = &empty_fops;
686 return 0;
687 }
688
689 if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
690 error = get_write_access(inode);
691 if (unlikely(error))
692 goto cleanup_file;
693 error = __mnt_want_write(f->f_path.mnt);
694 if (unlikely(error)) {
695 put_write_access(inode);
696 goto cleanup_file;
697 }
698 f->f_mode |= FMODE_WRITER;
699 }
700
701 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
702 if (S_ISREG(inode->i_mode))
703 f->f_mode |= FMODE_ATOMIC_POS;
704
705 f->f_op = fops_get(inode->i_fop);
706 if (unlikely(WARN_ON(!f->f_op))) {
707 error = -ENODEV;
708 goto cleanup_all;
709 }
710
711 error = security_file_open(f, cred);
712 if (error)
713 goto cleanup_all;
714
715 error = break_lease(inode, f->f_flags);
716 if (error)
717 goto cleanup_all;
718
719 if (!open)
720 open = f->f_op->open;
721 if (open) {
722 error = open(inode, f);
723 if (error)
724 goto cleanup_all;
725 }
726 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
727 i_readcount_inc(inode);
728
729 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
730
731 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
732
733 return 0;
734
735cleanup_all:
736 fops_put(f->f_op);
737 if (f->f_mode & FMODE_WRITER) {
738 put_write_access(inode);
739 __mnt_drop_write(f->f_path.mnt);
740 }
741cleanup_file:
742 path_put(&f->f_path);
743 f->f_path.mnt = NULL;
744 f->f_path.dentry = NULL;
745 f->f_inode = NULL;
746 return error;
747}
748
749/**
750 * finish_open - finish opening a file
751 * @file: file pointer
752 * @dentry: pointer to dentry
753 * @open: open callback
754 * @opened: state of open
755 *
756 * This can be used to finish opening a file passed to i_op->atomic_open().
757 *
758 * If the open callback is set to NULL, then the standard f_op->open()
759 * filesystem callback is substituted.
760 *
761 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
762 * the return value of d_splice_alias(), then the caller needs to perform dput()
763 * on it after finish_open().
764 *
765 * On successful return @file is a fully instantiated open file. After this, if
766 * an error occurs in ->atomic_open(), it needs to clean up with fput().
767 *
768 * Returns zero on success or -errno if the open failed.
769 */
770int finish_open(struct file *file, struct dentry *dentry,
771 int (*open)(struct inode *, struct file *),
772 int *opened)
773{
774 int error;
775 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
776
777 file->f_path.dentry = dentry;
778 error = do_dentry_open(file, open, current_cred());
779 if (!error)
780 *opened |= FILE_OPENED;
781
782 return error;
783}
784EXPORT_SYMBOL(finish_open);
785
786/**
787 * finish_no_open - finish ->atomic_open() without opening the file
788 *
789 * @file: file pointer
790 * @dentry: dentry or NULL (as returned from ->lookup())
791 *
792 * This can be used to set the result of a successful lookup in ->atomic_open().
793 *
794 * NB: unlike finish_open() this function does consume the dentry reference and
795 * the caller need not dput() it.
796 *
797 * Returns "1" which must be the return value of ->atomic_open() after having
798 * called this function.
799 */
800int finish_no_open(struct file *file, struct dentry *dentry)
801{
802 file->f_path.dentry = dentry;
803 return 1;
804}
805EXPORT_SYMBOL(finish_no_open);
806
807struct file *dentry_open(const struct path *path, int flags,
808 const struct cred *cred)
809{
810 int error;
811 struct file *f;
812
813 validate_creds(cred);
814
815 /* We must always pass in a valid mount pointer. */
816 BUG_ON(!path->mnt);
817
818 f = get_empty_filp();
819 if (!IS_ERR(f)) {
820 f->f_flags = flags;
821 f->f_path = *path;
822 error = do_dentry_open(f, NULL, cred);
823 if (!error) {
824 /* from now on we need fput() to dispose of f */
825 error = open_check_o_direct(f);
826 if (error) {
827 fput(f);
828 f = ERR_PTR(error);
829 }
830 } else {
831 put_filp(f);
832 f = ERR_PTR(error);
833 }
834 }
835 return f;
836}
837EXPORT_SYMBOL(dentry_open);
838
839static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
840{
841 int lookup_flags = 0;
842 int acc_mode;
843
844 if (flags & (O_CREAT | __O_TMPFILE))
845 op->mode = (mode & S_IALLUGO) | S_IFREG;
846 else
847 op->mode = 0;
848
849 /* Must never be set by userspace */
850 flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
851
852 /*
853 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
854 * check for O_DSYNC if the need any syncing at all we enforce it's
855 * always set instead of having to deal with possibly weird behaviour
856 * for malicious applications setting only __O_SYNC.
857 */
858 if (flags & __O_SYNC)
859 flags |= O_DSYNC;
860
861 if (flags & __O_TMPFILE) {
862 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
863 return -EINVAL;
864 acc_mode = MAY_OPEN | ACC_MODE(flags);
865 if (!(acc_mode & MAY_WRITE))
866 return -EINVAL;
867 } else if (flags & O_PATH) {
868 /*
869 * If we have O_PATH in the open flag. Then we
870 * cannot have anything other than the below set of flags
871 */
872 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
873 acc_mode = 0;
874 } else {
875 acc_mode = MAY_OPEN | ACC_MODE(flags);
876 }
877
878 op->open_flag = flags;
879
880 /* O_TRUNC implies we need access checks for write permissions */
881 if (flags & O_TRUNC)
882 acc_mode |= MAY_WRITE;
883
884 /* Allow the LSM permission hook to distinguish append
885 access from general write access. */
886 if (flags & O_APPEND)
887 acc_mode |= MAY_APPEND;
888
889 op->acc_mode = acc_mode;
890
891 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
892
893 if (flags & O_CREAT) {
894 op->intent |= LOOKUP_CREATE;
895 if (flags & O_EXCL)
896 op->intent |= LOOKUP_EXCL;
897 }
898
899 if (flags & O_DIRECTORY)
900 lookup_flags |= LOOKUP_DIRECTORY;
901 if (!(flags & O_NOFOLLOW))
902 lookup_flags |= LOOKUP_FOLLOW;
903 op->lookup_flags = lookup_flags;
904 return 0;
905}
906
907/**
908 * file_open_name - open file and return file pointer
909 *
910 * @name: struct filename containing path to open
911 * @flags: open flags as per the open(2) second argument
912 * @mode: mode for the new file if O_CREAT is set, else ignored
913 *
914 * This is the helper to open a file from kernelspace if you really
915 * have to. But in generally you should not do this, so please move
916 * along, nothing to see here..
917 */
918struct file *file_open_name(struct filename *name, int flags, umode_t mode)
919{
920 struct open_flags op;
921 int err = build_open_flags(flags, mode, &op);
922 return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
923}
924
925/**
926 * filp_open - open file and return file pointer
927 *
928 * @filename: path to open
929 * @flags: open flags as per the open(2) second argument
930 * @mode: mode for the new file if O_CREAT is set, else ignored
931 *
932 * This is the helper to open a file from kernelspace if you really
933 * have to. But in generally you should not do this, so please move
934 * along, nothing to see here..
935 */
936struct file *filp_open(const char *filename, int flags, umode_t mode)
937{
938 struct filename name = {.name = filename};
939 return file_open_name(&name, flags, mode);
940}
941EXPORT_SYMBOL(filp_open);
942
943struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
944 const char *filename, int flags)
945{
946 struct open_flags op;
947 int err = build_open_flags(flags, 0, &op);
948 if (err)
949 return ERR_PTR(err);
950 if (flags & O_CREAT)
951 return ERR_PTR(-EINVAL);
952 if (!filename && (flags & O_DIRECTORY))
953 if (!dentry->d_inode->i_op->lookup)
954 return ERR_PTR(-ENOTDIR);
955 return do_file_open_root(dentry, mnt, filename, &op);
956}
957EXPORT_SYMBOL(file_open_root);
958
959long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
960{
961 struct open_flags op;
962 int fd = build_open_flags(flags, mode, &op);
963 struct filename *tmp;
964
965 if (fd)
966 return fd;
967
968 tmp = getname(filename);
969 if (IS_ERR(tmp))
970 return PTR_ERR(tmp);
971
972 fd = get_unused_fd_flags(flags);
973 if (fd >= 0) {
974 struct file *f = do_filp_open(dfd, tmp, &op);
975 if (IS_ERR(f)) {
976 put_unused_fd(fd);
977 fd = PTR_ERR(f);
978 } else {
979 fsnotify_open(f);
980 fd_install(fd, f);
981 }
982 }
983 putname(tmp);
984 return fd;
985}
986
987SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
988{
989 if (force_o_largefile())
990 flags |= O_LARGEFILE;
991
992 return do_sys_open(AT_FDCWD, filename, flags, mode);
993}
994
995SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
996 umode_t, mode)
997{
998 if (force_o_largefile())
999 flags |= O_LARGEFILE;
1000
1001 return do_sys_open(dfd, filename, flags, mode);
1002}
1003
1004#ifndef __alpha__
1005
1006/*
1007 * For backward compatibility? Maybe this should be moved
1008 * into arch/i386 instead?
1009 */
1010SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1011{
1012 return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1013}
1014
1015#endif
1016
1017/*
1018 * "id" is the POSIX thread ID. We use the
1019 * files pointer for this..
1020 */
1021int filp_close(struct file *filp, fl_owner_t id)
1022{
1023 int retval = 0;
1024
1025 if (!file_count(filp)) {
1026 printk(KERN_ERR "VFS: Close: file count is 0\n");
1027 return 0;
1028 }
1029
1030 if (filp->f_op->flush)
1031 retval = filp->f_op->flush(filp, id);
1032
1033 if (likely(!(filp->f_mode & FMODE_PATH))) {
1034 dnotify_flush(filp, id);
1035 locks_remove_posix(filp, id);
1036 }
1037 fput(filp);
1038 return retval;
1039}
1040
1041EXPORT_SYMBOL(filp_close);
1042
1043/*
1044 * Careful here! We test whether the file pointer is NULL before
1045 * releasing the fd. This ensures that one clone task can't release
1046 * an fd while another clone is opening it.
1047 */
1048SYSCALL_DEFINE1(close, unsigned int, fd)
1049{
1050 int retval = __close_fd(current->files, fd);
1051
1052 /* can't restart close syscall because file table entry was cleared */
1053 if (unlikely(retval == -ERESTARTSYS ||
1054 retval == -ERESTARTNOINTR ||
1055 retval == -ERESTARTNOHAND ||
1056 retval == -ERESTART_RESTARTBLOCK))
1057 retval = -EINTR;
1058
1059 return retval;
1060}
1061EXPORT_SYMBOL(sys_close);
1062
1063/*
1064 * This routine simulates a hangup on the tty, to arrange that users
1065 * are given clean terminals at login time.
1066 */
1067SYSCALL_DEFINE0(vhangup)
1068{
1069 if (capable(CAP_SYS_TTY_CONFIG)) {
1070 tty_vhangup_self();
1071 return 0;
1072 }
1073 return -EPERM;
1074}
1075
1076/*
1077 * Called when an inode is about to be open.
1078 * We use this to disallow opening large files on 32bit systems if
1079 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1080 * on this flag in sys_open.
1081 */
1082int generic_file_open(struct inode * inode, struct file * filp)
1083{
1084 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1085 return -EOVERFLOW;
1086 return 0;
1087}
1088
1089EXPORT_SYMBOL(generic_file_open);
1090
1091/*
1092 * This is used by subsystems that don't want seekable
1093 * file descriptors. The function is not supposed to ever fail, the only
1094 * reason it returns an 'int' and not 'void' is so that it can be plugged
1095 * directly into file_operations structure.
1096 */
1097int nonseekable_open(struct inode *inode, struct file *filp)
1098{
1099 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1100 return 0;
1101}
1102
1103EXPORT_SYMBOL(nonseekable_open);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/open.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8#include <linux/string.h>
9#include <linux/mm.h>
10#include <linux/file.h>
11#include <linux/fdtable.h>
12#include <linux/fsnotify.h>
13#include <linux/module.h>
14#include <linux/tty.h>
15#include <linux/namei.h>
16#include <linux/backing-dev.h>
17#include <linux/capability.h>
18#include <linux/securebits.h>
19#include <linux/security.h>
20#include <linux/mount.h>
21#include <linux/fcntl.h>
22#include <linux/slab.h>
23#include <linux/uaccess.h>
24#include <linux/fs.h>
25#include <linux/personality.h>
26#include <linux/pagemap.h>
27#include <linux/syscalls.h>
28#include <linux/rcupdate.h>
29#include <linux/audit.h>
30#include <linux/falloc.h>
31#include <linux/fs_struct.h>
32#include <linux/dnotify.h>
33#include <linux/compat.h>
34#include <linux/mnt_idmapping.h>
35#include <linux/filelock.h>
36
37#include "internal.h"
38
39int do_truncate(struct mnt_idmap *idmap, struct dentry *dentry,
40 loff_t length, unsigned int time_attrs, struct file *filp)
41{
42 int ret;
43 struct iattr newattrs;
44
45 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
46 if (length < 0)
47 return -EINVAL;
48
49 newattrs.ia_size = length;
50 newattrs.ia_valid = ATTR_SIZE | time_attrs;
51 if (filp) {
52 newattrs.ia_file = filp;
53 newattrs.ia_valid |= ATTR_FILE;
54 }
55
56 /* Remove suid, sgid, and file capabilities on truncate too */
57 ret = dentry_needs_remove_privs(idmap, dentry);
58 if (ret < 0)
59 return ret;
60 if (ret)
61 newattrs.ia_valid |= ret | ATTR_FORCE;
62
63 inode_lock(dentry->d_inode);
64 /* Note any delegations or leases have already been broken: */
65 ret = notify_change(idmap, dentry, &newattrs, NULL);
66 inode_unlock(dentry->d_inode);
67 return ret;
68}
69
70long vfs_truncate(const struct path *path, loff_t length)
71{
72 struct mnt_idmap *idmap;
73 struct inode *inode;
74 long error;
75
76 inode = path->dentry->d_inode;
77
78 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
79 if (S_ISDIR(inode->i_mode))
80 return -EISDIR;
81 if (!S_ISREG(inode->i_mode))
82 return -EINVAL;
83
84 error = mnt_want_write(path->mnt);
85 if (error)
86 goto out;
87
88 idmap = mnt_idmap(path->mnt);
89 error = inode_permission(idmap, inode, MAY_WRITE);
90 if (error)
91 goto mnt_drop_write_and_out;
92
93 error = -EPERM;
94 if (IS_APPEND(inode))
95 goto mnt_drop_write_and_out;
96
97 error = get_write_access(inode);
98 if (error)
99 goto mnt_drop_write_and_out;
100
101 /*
102 * Make sure that there are no leases. get_write_access() protects
103 * against the truncate racing with a lease-granting setlease().
104 */
105 error = break_lease(inode, O_WRONLY);
106 if (error)
107 goto put_write_and_out;
108
109 error = security_path_truncate(path);
110 if (!error)
111 error = do_truncate(idmap, path->dentry, length, 0, NULL);
112
113put_write_and_out:
114 put_write_access(inode);
115mnt_drop_write_and_out:
116 mnt_drop_write(path->mnt);
117out:
118 return error;
119}
120EXPORT_SYMBOL_GPL(vfs_truncate);
121
122long do_sys_truncate(const char __user *pathname, loff_t length)
123{
124 unsigned int lookup_flags = LOOKUP_FOLLOW;
125 struct path path;
126 int error;
127
128 if (length < 0) /* sorry, but loff_t says... */
129 return -EINVAL;
130
131retry:
132 error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
133 if (!error) {
134 error = vfs_truncate(&path, length);
135 path_put(&path);
136 }
137 if (retry_estale(error, lookup_flags)) {
138 lookup_flags |= LOOKUP_REVAL;
139 goto retry;
140 }
141 return error;
142}
143
144SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
145{
146 return do_sys_truncate(path, length);
147}
148
149#ifdef CONFIG_COMPAT
150COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
151{
152 return do_sys_truncate(path, length);
153}
154#endif
155
156long do_ftruncate(struct file *file, loff_t length, int small)
157{
158 struct inode *inode;
159 struct dentry *dentry;
160 int error;
161
162 /* explicitly opened as large or we are on 64-bit box */
163 if (file->f_flags & O_LARGEFILE)
164 small = 0;
165
166 dentry = file->f_path.dentry;
167 inode = dentry->d_inode;
168 if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
169 return -EINVAL;
170
171 /* Cannot ftruncate over 2^31 bytes without large file support */
172 if (small && length > MAX_NON_LFS)
173 return -EINVAL;
174
175 /* Check IS_APPEND on real upper inode */
176 if (IS_APPEND(file_inode(file)))
177 return -EPERM;
178 sb_start_write(inode->i_sb);
179 error = security_file_truncate(file);
180 if (!error)
181 error = do_truncate(file_mnt_idmap(file), dentry, length,
182 ATTR_MTIME | ATTR_CTIME, file);
183 sb_end_write(inode->i_sb);
184
185 return error;
186}
187
188long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
189{
190 if (length < 0)
191 return -EINVAL;
192 CLASS(fd, f)(fd);
193 if (fd_empty(f))
194 return -EBADF;
195
196 return do_ftruncate(fd_file(f), length, small);
197}
198
199SYSCALL_DEFINE2(ftruncate, unsigned int, fd, off_t, length)
200{
201 return do_sys_ftruncate(fd, length, 1);
202}
203
204#ifdef CONFIG_COMPAT
205COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_off_t, length)
206{
207 return do_sys_ftruncate(fd, length, 1);
208}
209#endif
210
211/* LFS versions of truncate are only needed on 32 bit machines */
212#if BITS_PER_LONG == 32
213SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
214{
215 return do_sys_truncate(path, length);
216}
217
218SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
219{
220 return do_sys_ftruncate(fd, length, 0);
221}
222#endif /* BITS_PER_LONG == 32 */
223
224#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_TRUNCATE64)
225COMPAT_SYSCALL_DEFINE3(truncate64, const char __user *, pathname,
226 compat_arg_u64_dual(length))
227{
228 return ksys_truncate(pathname, compat_arg_u64_glue(length));
229}
230#endif
231
232#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FTRUNCATE64)
233COMPAT_SYSCALL_DEFINE3(ftruncate64, unsigned int, fd,
234 compat_arg_u64_dual(length))
235{
236 return ksys_ftruncate(fd, compat_arg_u64_glue(length));
237}
238#endif
239
240int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
241{
242 struct inode *inode = file_inode(file);
243 long ret;
244 loff_t sum;
245
246 if (offset < 0 || len <= 0)
247 return -EINVAL;
248
249 if (mode & ~(FALLOC_FL_MODE_MASK | FALLOC_FL_KEEP_SIZE))
250 return -EOPNOTSUPP;
251
252 /*
253 * Modes are exclusive, even if that is not obvious from the encoding
254 * as bit masks and the mix with the flag in the same namespace.
255 *
256 * To make things even more complicated, FALLOC_FL_ALLOCATE_RANGE is
257 * encoded as no bit set.
258 */
259 switch (mode & FALLOC_FL_MODE_MASK) {
260 case FALLOC_FL_ALLOCATE_RANGE:
261 case FALLOC_FL_UNSHARE_RANGE:
262 case FALLOC_FL_ZERO_RANGE:
263 break;
264 case FALLOC_FL_PUNCH_HOLE:
265 if (!(mode & FALLOC_FL_KEEP_SIZE))
266 return -EOPNOTSUPP;
267 break;
268 case FALLOC_FL_COLLAPSE_RANGE:
269 case FALLOC_FL_INSERT_RANGE:
270 if (mode & FALLOC_FL_KEEP_SIZE)
271 return -EOPNOTSUPP;
272 break;
273 default:
274 return -EOPNOTSUPP;
275 }
276
277 if (!(file->f_mode & FMODE_WRITE))
278 return -EBADF;
279
280 /*
281 * On append-only files only space preallocation is supported.
282 */
283 if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
284 return -EPERM;
285
286 if (IS_IMMUTABLE(inode))
287 return -EPERM;
288
289 /*
290 * We cannot allow any fallocate operation on an active swapfile
291 */
292 if (IS_SWAPFILE(inode))
293 return -ETXTBSY;
294
295 /*
296 * Revalidate the write permissions, in case security policy has
297 * changed since the files were opened.
298 */
299 ret = security_file_permission(file, MAY_WRITE);
300 if (ret)
301 return ret;
302
303 ret = fsnotify_file_area_perm(file, MAY_WRITE, &offset, len);
304 if (ret)
305 return ret;
306
307 if (S_ISFIFO(inode->i_mode))
308 return -ESPIPE;
309
310 if (S_ISDIR(inode->i_mode))
311 return -EISDIR;
312
313 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
314 return -ENODEV;
315
316 /* Check for wraparound */
317 if (check_add_overflow(offset, len, &sum))
318 return -EFBIG;
319
320 if (sum > inode->i_sb->s_maxbytes)
321 return -EFBIG;
322
323 if (!file->f_op->fallocate)
324 return -EOPNOTSUPP;
325
326 file_start_write(file);
327 ret = file->f_op->fallocate(file, mode, offset, len);
328
329 /*
330 * Create inotify and fanotify events.
331 *
332 * To keep the logic simple always create events if fallocate succeeds.
333 * This implies that events are even created if the file size remains
334 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
335 */
336 if (ret == 0)
337 fsnotify_modify(file);
338
339 file_end_write(file);
340 return ret;
341}
342EXPORT_SYMBOL_GPL(vfs_fallocate);
343
344int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
345{
346 CLASS(fd, f)(fd);
347
348 if (fd_empty(f))
349 return -EBADF;
350
351 return vfs_fallocate(fd_file(f), mode, offset, len);
352}
353
354SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
355{
356 return ksys_fallocate(fd, mode, offset, len);
357}
358
359#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FALLOCATE)
360COMPAT_SYSCALL_DEFINE6(fallocate, int, fd, int, mode, compat_arg_u64_dual(offset),
361 compat_arg_u64_dual(len))
362{
363 return ksys_fallocate(fd, mode, compat_arg_u64_glue(offset),
364 compat_arg_u64_glue(len));
365}
366#endif
367
368/*
369 * access() needs to use the real uid/gid, not the effective uid/gid.
370 * We do this by temporarily clearing all FS-related capabilities and
371 * switching the fsuid/fsgid around to the real ones.
372 *
373 * Creating new credentials is expensive, so we try to skip doing it,
374 * which we can if the result would match what we already got.
375 */
376static bool access_need_override_creds(int flags)
377{
378 const struct cred *cred;
379
380 if (flags & AT_EACCESS)
381 return false;
382
383 cred = current_cred();
384 if (!uid_eq(cred->fsuid, cred->uid) ||
385 !gid_eq(cred->fsgid, cred->gid))
386 return true;
387
388 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
389 kuid_t root_uid = make_kuid(cred->user_ns, 0);
390 if (!uid_eq(cred->uid, root_uid)) {
391 if (!cap_isclear(cred->cap_effective))
392 return true;
393 } else {
394 if (!cap_isidentical(cred->cap_effective,
395 cred->cap_permitted))
396 return true;
397 }
398 }
399
400 return false;
401}
402
403static const struct cred *access_override_creds(void)
404{
405 const struct cred *old_cred;
406 struct cred *override_cred;
407
408 override_cred = prepare_creds();
409 if (!override_cred)
410 return NULL;
411
412 /*
413 * XXX access_need_override_creds performs checks in hopes of skipping
414 * this work. Make sure it stays in sync if making any changes in this
415 * routine.
416 */
417
418 override_cred->fsuid = override_cred->uid;
419 override_cred->fsgid = override_cred->gid;
420
421 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
422 /* Clear the capabilities if we switch to a non-root user */
423 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
424 if (!uid_eq(override_cred->uid, root_uid))
425 cap_clear(override_cred->cap_effective);
426 else
427 override_cred->cap_effective =
428 override_cred->cap_permitted;
429 }
430
431 /*
432 * The new set of credentials can *only* be used in
433 * task-synchronous circumstances, and does not need
434 * RCU freeing, unless somebody then takes a separate
435 * reference to it.
436 *
437 * NOTE! This is _only_ true because this credential
438 * is used purely for override_creds() that installs
439 * it as the subjective cred. Other threads will be
440 * accessing ->real_cred, not the subjective cred.
441 *
442 * If somebody _does_ make a copy of this (using the
443 * 'get_current_cred()' function), that will clear the
444 * non_rcu field, because now that other user may be
445 * expecting RCU freeing. But normal thread-synchronous
446 * cred accesses will keep things non-racy to avoid RCU
447 * freeing.
448 */
449 override_cred->non_rcu = 1;
450
451 old_cred = override_creds(override_cred);
452
453 /* override_cred() gets its own ref */
454 put_cred(override_cred);
455
456 return old_cred;
457}
458
459static long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
460{
461 struct path path;
462 struct inode *inode;
463 int res;
464 unsigned int lookup_flags = LOOKUP_FOLLOW;
465 const struct cred *old_cred = NULL;
466
467 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
468 return -EINVAL;
469
470 if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
471 return -EINVAL;
472
473 if (flags & AT_SYMLINK_NOFOLLOW)
474 lookup_flags &= ~LOOKUP_FOLLOW;
475 if (flags & AT_EMPTY_PATH)
476 lookup_flags |= LOOKUP_EMPTY;
477
478 if (access_need_override_creds(flags)) {
479 old_cred = access_override_creds();
480 if (!old_cred)
481 return -ENOMEM;
482 }
483
484retry:
485 res = user_path_at(dfd, filename, lookup_flags, &path);
486 if (res)
487 goto out;
488
489 inode = d_backing_inode(path.dentry);
490
491 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
492 /*
493 * MAY_EXEC on regular files is denied if the fs is mounted
494 * with the "noexec" flag.
495 */
496 res = -EACCES;
497 if (path_noexec(&path))
498 goto out_path_release;
499 }
500
501 res = inode_permission(mnt_idmap(path.mnt), inode, mode | MAY_ACCESS);
502 /* SuS v2 requires we report a read only fs too */
503 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
504 goto out_path_release;
505 /*
506 * This is a rare case where using __mnt_is_readonly()
507 * is OK without a mnt_want/drop_write() pair. Since
508 * no actual write to the fs is performed here, we do
509 * not need to telegraph to that to anyone.
510 *
511 * By doing this, we accept that this access is
512 * inherently racy and know that the fs may change
513 * state before we even see this result.
514 */
515 if (__mnt_is_readonly(path.mnt))
516 res = -EROFS;
517
518out_path_release:
519 path_put(&path);
520 if (retry_estale(res, lookup_flags)) {
521 lookup_flags |= LOOKUP_REVAL;
522 goto retry;
523 }
524out:
525 if (old_cred)
526 revert_creds(old_cred);
527
528 return res;
529}
530
531SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
532{
533 return do_faccessat(dfd, filename, mode, 0);
534}
535
536SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
537 int, flags)
538{
539 return do_faccessat(dfd, filename, mode, flags);
540}
541
542SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
543{
544 return do_faccessat(AT_FDCWD, filename, mode, 0);
545}
546
547SYSCALL_DEFINE1(chdir, const char __user *, filename)
548{
549 struct path path;
550 int error;
551 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
552retry:
553 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
554 if (error)
555 goto out;
556
557 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
558 if (error)
559 goto dput_and_out;
560
561 set_fs_pwd(current->fs, &path);
562
563dput_and_out:
564 path_put(&path);
565 if (retry_estale(error, lookup_flags)) {
566 lookup_flags |= LOOKUP_REVAL;
567 goto retry;
568 }
569out:
570 return error;
571}
572
573SYSCALL_DEFINE1(fchdir, unsigned int, fd)
574{
575 CLASS(fd_raw, f)(fd);
576 int error;
577
578 if (fd_empty(f))
579 return -EBADF;
580
581 if (!d_can_lookup(fd_file(f)->f_path.dentry))
582 return -ENOTDIR;
583
584 error = file_permission(fd_file(f), MAY_EXEC | MAY_CHDIR);
585 if (!error)
586 set_fs_pwd(current->fs, &fd_file(f)->f_path);
587 return error;
588}
589
590SYSCALL_DEFINE1(chroot, const char __user *, filename)
591{
592 struct path path;
593 int error;
594 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
595retry:
596 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
597 if (error)
598 goto out;
599
600 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
601 if (error)
602 goto dput_and_out;
603
604 error = -EPERM;
605 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
606 goto dput_and_out;
607 error = security_path_chroot(&path);
608 if (error)
609 goto dput_and_out;
610
611 set_fs_root(current->fs, &path);
612 error = 0;
613dput_and_out:
614 path_put(&path);
615 if (retry_estale(error, lookup_flags)) {
616 lookup_flags |= LOOKUP_REVAL;
617 goto retry;
618 }
619out:
620 return error;
621}
622
623int chmod_common(const struct path *path, umode_t mode)
624{
625 struct inode *inode = path->dentry->d_inode;
626 struct inode *delegated_inode = NULL;
627 struct iattr newattrs;
628 int error;
629
630 error = mnt_want_write(path->mnt);
631 if (error)
632 return error;
633retry_deleg:
634 inode_lock(inode);
635 error = security_path_chmod(path, mode);
636 if (error)
637 goto out_unlock;
638 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
639 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
640 error = notify_change(mnt_idmap(path->mnt), path->dentry,
641 &newattrs, &delegated_inode);
642out_unlock:
643 inode_unlock(inode);
644 if (delegated_inode) {
645 error = break_deleg_wait(&delegated_inode);
646 if (!error)
647 goto retry_deleg;
648 }
649 mnt_drop_write(path->mnt);
650 return error;
651}
652
653int vfs_fchmod(struct file *file, umode_t mode)
654{
655 audit_file(file);
656 return chmod_common(&file->f_path, mode);
657}
658
659SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
660{
661 CLASS(fd, f)(fd);
662
663 if (fd_empty(f))
664 return -EBADF;
665
666 return vfs_fchmod(fd_file(f), mode);
667}
668
669static int do_fchmodat(int dfd, const char __user *filename, umode_t mode,
670 unsigned int flags)
671{
672 struct path path;
673 int error;
674 unsigned int lookup_flags;
675
676 if (unlikely(flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)))
677 return -EINVAL;
678
679 lookup_flags = (flags & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
680 if (flags & AT_EMPTY_PATH)
681 lookup_flags |= LOOKUP_EMPTY;
682
683retry:
684 error = user_path_at(dfd, filename, lookup_flags, &path);
685 if (!error) {
686 error = chmod_common(&path, mode);
687 path_put(&path);
688 if (retry_estale(error, lookup_flags)) {
689 lookup_flags |= LOOKUP_REVAL;
690 goto retry;
691 }
692 }
693 return error;
694}
695
696SYSCALL_DEFINE4(fchmodat2, int, dfd, const char __user *, filename,
697 umode_t, mode, unsigned int, flags)
698{
699 return do_fchmodat(dfd, filename, mode, flags);
700}
701
702SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
703 umode_t, mode)
704{
705 return do_fchmodat(dfd, filename, mode, 0);
706}
707
708SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
709{
710 return do_fchmodat(AT_FDCWD, filename, mode, 0);
711}
712
713/*
714 * Check whether @kuid is valid and if so generate and set vfsuid_t in
715 * ia_vfsuid.
716 *
717 * Return: true if @kuid is valid, false if not.
718 */
719static inline bool setattr_vfsuid(struct iattr *attr, kuid_t kuid)
720{
721 if (!uid_valid(kuid))
722 return false;
723 attr->ia_valid |= ATTR_UID;
724 attr->ia_vfsuid = VFSUIDT_INIT(kuid);
725 return true;
726}
727
728/*
729 * Check whether @kgid is valid and if so generate and set vfsgid_t in
730 * ia_vfsgid.
731 *
732 * Return: true if @kgid is valid, false if not.
733 */
734static inline bool setattr_vfsgid(struct iattr *attr, kgid_t kgid)
735{
736 if (!gid_valid(kgid))
737 return false;
738 attr->ia_valid |= ATTR_GID;
739 attr->ia_vfsgid = VFSGIDT_INIT(kgid);
740 return true;
741}
742
743int chown_common(const struct path *path, uid_t user, gid_t group)
744{
745 struct mnt_idmap *idmap;
746 struct user_namespace *fs_userns;
747 struct inode *inode = path->dentry->d_inode;
748 struct inode *delegated_inode = NULL;
749 int error;
750 struct iattr newattrs;
751 kuid_t uid;
752 kgid_t gid;
753
754 uid = make_kuid(current_user_ns(), user);
755 gid = make_kgid(current_user_ns(), group);
756
757 idmap = mnt_idmap(path->mnt);
758 fs_userns = i_user_ns(inode);
759
760retry_deleg:
761 newattrs.ia_vfsuid = INVALID_VFSUID;
762 newattrs.ia_vfsgid = INVALID_VFSGID;
763 newattrs.ia_valid = ATTR_CTIME;
764 if ((user != (uid_t)-1) && !setattr_vfsuid(&newattrs, uid))
765 return -EINVAL;
766 if ((group != (gid_t)-1) && !setattr_vfsgid(&newattrs, gid))
767 return -EINVAL;
768 inode_lock(inode);
769 if (!S_ISDIR(inode->i_mode))
770 newattrs.ia_valid |= ATTR_KILL_SUID | ATTR_KILL_PRIV |
771 setattr_should_drop_sgid(idmap, inode);
772 /* Continue to send actual fs values, not the mount values. */
773 error = security_path_chown(
774 path,
775 from_vfsuid(idmap, fs_userns, newattrs.ia_vfsuid),
776 from_vfsgid(idmap, fs_userns, newattrs.ia_vfsgid));
777 if (!error)
778 error = notify_change(idmap, path->dentry, &newattrs,
779 &delegated_inode);
780 inode_unlock(inode);
781 if (delegated_inode) {
782 error = break_deleg_wait(&delegated_inode);
783 if (!error)
784 goto retry_deleg;
785 }
786 return error;
787}
788
789int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
790 int flag)
791{
792 struct path path;
793 int error = -EINVAL;
794 int lookup_flags;
795
796 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
797 goto out;
798
799 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
800 if (flag & AT_EMPTY_PATH)
801 lookup_flags |= LOOKUP_EMPTY;
802retry:
803 error = user_path_at(dfd, filename, lookup_flags, &path);
804 if (error)
805 goto out;
806 error = mnt_want_write(path.mnt);
807 if (error)
808 goto out_release;
809 error = chown_common(&path, user, group);
810 mnt_drop_write(path.mnt);
811out_release:
812 path_put(&path);
813 if (retry_estale(error, lookup_flags)) {
814 lookup_flags |= LOOKUP_REVAL;
815 goto retry;
816 }
817out:
818 return error;
819}
820
821SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
822 gid_t, group, int, flag)
823{
824 return do_fchownat(dfd, filename, user, group, flag);
825}
826
827SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
828{
829 return do_fchownat(AT_FDCWD, filename, user, group, 0);
830}
831
832SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
833{
834 return do_fchownat(AT_FDCWD, filename, user, group,
835 AT_SYMLINK_NOFOLLOW);
836}
837
838int vfs_fchown(struct file *file, uid_t user, gid_t group)
839{
840 int error;
841
842 error = mnt_want_write_file(file);
843 if (error)
844 return error;
845 audit_file(file);
846 error = chown_common(&file->f_path, user, group);
847 mnt_drop_write_file(file);
848 return error;
849}
850
851int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
852{
853 CLASS(fd, f)(fd);
854
855 if (fd_empty(f))
856 return -EBADF;
857
858 return vfs_fchown(fd_file(f), user, group);
859}
860
861SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
862{
863 return ksys_fchown(fd, user, group);
864}
865
866static inline int file_get_write_access(struct file *f)
867{
868 int error;
869
870 error = get_write_access(f->f_inode);
871 if (unlikely(error))
872 return error;
873 error = mnt_get_write_access(f->f_path.mnt);
874 if (unlikely(error))
875 goto cleanup_inode;
876 if (unlikely(f->f_mode & FMODE_BACKING)) {
877 error = mnt_get_write_access(backing_file_user_path(f)->mnt);
878 if (unlikely(error))
879 goto cleanup_mnt;
880 }
881 return 0;
882
883cleanup_mnt:
884 mnt_put_write_access(f->f_path.mnt);
885cleanup_inode:
886 put_write_access(f->f_inode);
887 return error;
888}
889
890static int do_dentry_open(struct file *f,
891 int (*open)(struct inode *, struct file *))
892{
893 static const struct file_operations empty_fops = {};
894 struct inode *inode = f->f_path.dentry->d_inode;
895 int error;
896
897 path_get(&f->f_path);
898 f->f_inode = inode;
899 f->f_mapping = inode->i_mapping;
900 f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
901 f->f_sb_err = file_sample_sb_err(f);
902
903 if (unlikely(f->f_flags & O_PATH)) {
904 f->f_mode = FMODE_PATH | FMODE_OPENED;
905 f->f_op = &empty_fops;
906 return 0;
907 }
908
909 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) {
910 i_readcount_inc(inode);
911 } else if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
912 error = file_get_write_access(f);
913 if (unlikely(error))
914 goto cleanup_file;
915 f->f_mode |= FMODE_WRITER;
916 }
917
918 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
919 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
920 f->f_mode |= FMODE_ATOMIC_POS;
921
922 f->f_op = fops_get(inode->i_fop);
923 if (WARN_ON(!f->f_op)) {
924 error = -ENODEV;
925 goto cleanup_all;
926 }
927
928 error = security_file_open(f);
929 if (error)
930 goto cleanup_all;
931
932 error = fsnotify_open_perm(f);
933 if (error)
934 goto cleanup_all;
935
936 error = break_lease(file_inode(f), f->f_flags);
937 if (error)
938 goto cleanup_all;
939
940 /* normally all 3 are set; ->open() can clear them if needed */
941 f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
942 if (!open)
943 open = f->f_op->open;
944 if (open) {
945 error = open(inode, f);
946 if (error)
947 goto cleanup_all;
948 }
949 f->f_mode |= FMODE_OPENED;
950 if ((f->f_mode & FMODE_READ) &&
951 likely(f->f_op->read || f->f_op->read_iter))
952 f->f_mode |= FMODE_CAN_READ;
953 if ((f->f_mode & FMODE_WRITE) &&
954 likely(f->f_op->write || f->f_op->write_iter))
955 f->f_mode |= FMODE_CAN_WRITE;
956 if ((f->f_mode & FMODE_LSEEK) && !f->f_op->llseek)
957 f->f_mode &= ~FMODE_LSEEK;
958 if (f->f_mapping->a_ops && f->f_mapping->a_ops->direct_IO)
959 f->f_mode |= FMODE_CAN_ODIRECT;
960
961 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
962 f->f_iocb_flags = iocb_flags(f);
963
964 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
965
966 if ((f->f_flags & O_DIRECT) && !(f->f_mode & FMODE_CAN_ODIRECT))
967 return -EINVAL;
968
969 /*
970 * XXX: Huge page cache doesn't support writing yet. Drop all page
971 * cache for this file before processing writes.
972 */
973 if (f->f_mode & FMODE_WRITE) {
974 /*
975 * Depends on full fence from get_write_access() to synchronize
976 * against collapse_file() regarding i_writecount and nr_thps
977 * updates. Ensures subsequent insertion of THPs into the page
978 * cache will fail.
979 */
980 if (filemap_nr_thps(inode->i_mapping)) {
981 struct address_space *mapping = inode->i_mapping;
982
983 filemap_invalidate_lock(inode->i_mapping);
984 /*
985 * unmap_mapping_range just need to be called once
986 * here, because the private pages is not need to be
987 * unmapped mapping (e.g. data segment of dynamic
988 * shared libraries here).
989 */
990 unmap_mapping_range(mapping, 0, 0, 0);
991 truncate_inode_pages(mapping, 0);
992 filemap_invalidate_unlock(inode->i_mapping);
993 }
994 }
995
996 return 0;
997
998cleanup_all:
999 if (WARN_ON_ONCE(error > 0))
1000 error = -EINVAL;
1001 fops_put(f->f_op);
1002 put_file_access(f);
1003cleanup_file:
1004 path_put(&f->f_path);
1005 f->f_path.mnt = NULL;
1006 f->f_path.dentry = NULL;
1007 f->f_inode = NULL;
1008 return error;
1009}
1010
1011/**
1012 * finish_open - finish opening a file
1013 * @file: file pointer
1014 * @dentry: pointer to dentry
1015 * @open: open callback
1016 *
1017 * This can be used to finish opening a file passed to i_op->atomic_open().
1018 *
1019 * If the open callback is set to NULL, then the standard f_op->open()
1020 * filesystem callback is substituted.
1021 *
1022 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
1023 * the return value of d_splice_alias(), then the caller needs to perform dput()
1024 * on it after finish_open().
1025 *
1026 * Returns zero on success or -errno if the open failed.
1027 */
1028int finish_open(struct file *file, struct dentry *dentry,
1029 int (*open)(struct inode *, struct file *))
1030{
1031 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
1032
1033 file->f_path.dentry = dentry;
1034 return do_dentry_open(file, open);
1035}
1036EXPORT_SYMBOL(finish_open);
1037
1038/**
1039 * finish_no_open - finish ->atomic_open() without opening the file
1040 *
1041 * @file: file pointer
1042 * @dentry: dentry or NULL (as returned from ->lookup())
1043 *
1044 * This can be used to set the result of a successful lookup in ->atomic_open().
1045 *
1046 * NB: unlike finish_open() this function does consume the dentry reference and
1047 * the caller need not dput() it.
1048 *
1049 * Returns "0" which must be the return value of ->atomic_open() after having
1050 * called this function.
1051 */
1052int finish_no_open(struct file *file, struct dentry *dentry)
1053{
1054 file->f_path.dentry = dentry;
1055 return 0;
1056}
1057EXPORT_SYMBOL(finish_no_open);
1058
1059char *file_path(struct file *filp, char *buf, int buflen)
1060{
1061 return d_path(&filp->f_path, buf, buflen);
1062}
1063EXPORT_SYMBOL(file_path);
1064
1065/**
1066 * vfs_open - open the file at the given path
1067 * @path: path to open
1068 * @file: newly allocated file with f_flag initialized
1069 */
1070int vfs_open(const struct path *path, struct file *file)
1071{
1072 int ret;
1073
1074 file->f_path = *path;
1075 ret = do_dentry_open(file, NULL);
1076 if (!ret) {
1077 /*
1078 * Once we return a file with FMODE_OPENED, __fput() will call
1079 * fsnotify_close(), so we need fsnotify_open() here for
1080 * symmetry.
1081 */
1082 fsnotify_open(file);
1083 }
1084 return ret;
1085}
1086
1087struct file *dentry_open(const struct path *path, int flags,
1088 const struct cred *cred)
1089{
1090 int error;
1091 struct file *f;
1092
1093 /* We must always pass in a valid mount pointer. */
1094 BUG_ON(!path->mnt);
1095
1096 f = alloc_empty_file(flags, cred);
1097 if (!IS_ERR(f)) {
1098 error = vfs_open(path, f);
1099 if (error) {
1100 fput(f);
1101 f = ERR_PTR(error);
1102 }
1103 }
1104 return f;
1105}
1106EXPORT_SYMBOL(dentry_open);
1107
1108/**
1109 * dentry_create - Create and open a file
1110 * @path: path to create
1111 * @flags: O_ flags
1112 * @mode: mode bits for new file
1113 * @cred: credentials to use
1114 *
1115 * Caller must hold the parent directory's lock, and have prepared
1116 * a negative dentry, placed in @path->dentry, for the new file.
1117 *
1118 * Caller sets @path->mnt to the vfsmount of the filesystem where
1119 * the new file is to be created. The parent directory and the
1120 * negative dentry must reside on the same filesystem instance.
1121 *
1122 * On success, returns a "struct file *". Otherwise a ERR_PTR
1123 * is returned.
1124 */
1125struct file *dentry_create(const struct path *path, int flags, umode_t mode,
1126 const struct cred *cred)
1127{
1128 struct file *f;
1129 int error;
1130
1131 f = alloc_empty_file(flags, cred);
1132 if (IS_ERR(f))
1133 return f;
1134
1135 error = vfs_create(mnt_idmap(path->mnt),
1136 d_inode(path->dentry->d_parent),
1137 path->dentry, mode, true);
1138 if (!error)
1139 error = vfs_open(path, f);
1140
1141 if (unlikely(error)) {
1142 fput(f);
1143 return ERR_PTR(error);
1144 }
1145 return f;
1146}
1147EXPORT_SYMBOL(dentry_create);
1148
1149/**
1150 * kernel_file_open - open a file for kernel internal use
1151 * @path: path of the file to open
1152 * @flags: open flags
1153 * @cred: credentials for open
1154 *
1155 * Open a file for use by in-kernel consumers. The file is not accounted
1156 * against nr_files and must not be installed into the file descriptor
1157 * table.
1158 *
1159 * Return: Opened file on success, an error pointer on failure.
1160 */
1161struct file *kernel_file_open(const struct path *path, int flags,
1162 const struct cred *cred)
1163{
1164 struct file *f;
1165 int error;
1166
1167 f = alloc_empty_file_noaccount(flags, cred);
1168 if (IS_ERR(f))
1169 return f;
1170
1171 f->f_path = *path;
1172 error = do_dentry_open(f, NULL);
1173 if (error) {
1174 fput(f);
1175 return ERR_PTR(error);
1176 }
1177
1178 fsnotify_open(f);
1179 return f;
1180}
1181EXPORT_SYMBOL_GPL(kernel_file_open);
1182
1183#define WILL_CREATE(flags) (flags & (O_CREAT | __O_TMPFILE))
1184#define O_PATH_FLAGS (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1185
1186inline struct open_how build_open_how(int flags, umode_t mode)
1187{
1188 struct open_how how = {
1189 .flags = flags & VALID_OPEN_FLAGS,
1190 .mode = mode & S_IALLUGO,
1191 };
1192
1193 /* O_PATH beats everything else. */
1194 if (how.flags & O_PATH)
1195 how.flags &= O_PATH_FLAGS;
1196 /* Modes should only be set for create-like flags. */
1197 if (!WILL_CREATE(how.flags))
1198 how.mode = 0;
1199 return how;
1200}
1201
1202inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1203{
1204 u64 flags = how->flags;
1205 u64 strip = __FMODE_NONOTIFY | O_CLOEXEC;
1206 int lookup_flags = 0;
1207 int acc_mode = ACC_MODE(flags);
1208
1209 BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1210 "struct open_flags doesn't yet handle flags > 32 bits");
1211
1212 /*
1213 * Strip flags that either shouldn't be set by userspace like
1214 * FMODE_NONOTIFY or that aren't relevant in determining struct
1215 * open_flags like O_CLOEXEC.
1216 */
1217 flags &= ~strip;
1218
1219 /*
1220 * Older syscalls implicitly clear all of the invalid flags or argument
1221 * values before calling build_open_flags(), but openat2(2) checks all
1222 * of its arguments.
1223 */
1224 if (flags & ~VALID_OPEN_FLAGS)
1225 return -EINVAL;
1226 if (how->resolve & ~VALID_RESOLVE_FLAGS)
1227 return -EINVAL;
1228
1229 /* Scoping flags are mutually exclusive. */
1230 if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1231 return -EINVAL;
1232
1233 /* Deal with the mode. */
1234 if (WILL_CREATE(flags)) {
1235 if (how->mode & ~S_IALLUGO)
1236 return -EINVAL;
1237 op->mode = how->mode | S_IFREG;
1238 } else {
1239 if (how->mode != 0)
1240 return -EINVAL;
1241 op->mode = 0;
1242 }
1243
1244 /*
1245 * Block bugs where O_DIRECTORY | O_CREAT created regular files.
1246 * Note, that blocking O_DIRECTORY | O_CREAT here also protects
1247 * O_TMPFILE below which requires O_DIRECTORY being raised.
1248 */
1249 if ((flags & (O_DIRECTORY | O_CREAT)) == (O_DIRECTORY | O_CREAT))
1250 return -EINVAL;
1251
1252 /* Now handle the creative implementation of O_TMPFILE. */
1253 if (flags & __O_TMPFILE) {
1254 /*
1255 * In order to ensure programs get explicit errors when trying
1256 * to use O_TMPFILE on old kernels we enforce that O_DIRECTORY
1257 * is raised alongside __O_TMPFILE.
1258 */
1259 if (!(flags & O_DIRECTORY))
1260 return -EINVAL;
1261 if (!(acc_mode & MAY_WRITE))
1262 return -EINVAL;
1263 }
1264 if (flags & O_PATH) {
1265 /* O_PATH only permits certain other flags to be set. */
1266 if (flags & ~O_PATH_FLAGS)
1267 return -EINVAL;
1268 acc_mode = 0;
1269 }
1270
1271 /*
1272 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1273 * check for O_DSYNC if the need any syncing at all we enforce it's
1274 * always set instead of having to deal with possibly weird behaviour
1275 * for malicious applications setting only __O_SYNC.
1276 */
1277 if (flags & __O_SYNC)
1278 flags |= O_DSYNC;
1279
1280 op->open_flag = flags;
1281
1282 /* O_TRUNC implies we need access checks for write permissions */
1283 if (flags & O_TRUNC)
1284 acc_mode |= MAY_WRITE;
1285
1286 /* Allow the LSM permission hook to distinguish append
1287 access from general write access. */
1288 if (flags & O_APPEND)
1289 acc_mode |= MAY_APPEND;
1290
1291 op->acc_mode = acc_mode;
1292
1293 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1294
1295 if (flags & O_CREAT) {
1296 op->intent |= LOOKUP_CREATE;
1297 if (flags & O_EXCL) {
1298 op->intent |= LOOKUP_EXCL;
1299 flags |= O_NOFOLLOW;
1300 }
1301 }
1302
1303 if (flags & O_DIRECTORY)
1304 lookup_flags |= LOOKUP_DIRECTORY;
1305 if (!(flags & O_NOFOLLOW))
1306 lookup_flags |= LOOKUP_FOLLOW;
1307
1308 if (how->resolve & RESOLVE_NO_XDEV)
1309 lookup_flags |= LOOKUP_NO_XDEV;
1310 if (how->resolve & RESOLVE_NO_MAGICLINKS)
1311 lookup_flags |= LOOKUP_NO_MAGICLINKS;
1312 if (how->resolve & RESOLVE_NO_SYMLINKS)
1313 lookup_flags |= LOOKUP_NO_SYMLINKS;
1314 if (how->resolve & RESOLVE_BENEATH)
1315 lookup_flags |= LOOKUP_BENEATH;
1316 if (how->resolve & RESOLVE_IN_ROOT)
1317 lookup_flags |= LOOKUP_IN_ROOT;
1318 if (how->resolve & RESOLVE_CACHED) {
1319 /* Don't bother even trying for create/truncate/tmpfile open */
1320 if (flags & (O_TRUNC | O_CREAT | __O_TMPFILE))
1321 return -EAGAIN;
1322 lookup_flags |= LOOKUP_CACHED;
1323 }
1324
1325 op->lookup_flags = lookup_flags;
1326 return 0;
1327}
1328
1329/**
1330 * file_open_name - open file and return file pointer
1331 *
1332 * @name: struct filename containing path to open
1333 * @flags: open flags as per the open(2) second argument
1334 * @mode: mode for the new file if O_CREAT is set, else ignored
1335 *
1336 * This is the helper to open a file from kernelspace if you really
1337 * have to. But in generally you should not do this, so please move
1338 * along, nothing to see here..
1339 */
1340struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1341{
1342 struct open_flags op;
1343 struct open_how how = build_open_how(flags, mode);
1344 int err = build_open_flags(&how, &op);
1345 if (err)
1346 return ERR_PTR(err);
1347 return do_filp_open(AT_FDCWD, name, &op);
1348}
1349
1350/**
1351 * filp_open - open file and return file pointer
1352 *
1353 * @filename: path to open
1354 * @flags: open flags as per the open(2) second argument
1355 * @mode: mode for the new file if O_CREAT is set, else ignored
1356 *
1357 * This is the helper to open a file from kernelspace if you really
1358 * have to. But in generally you should not do this, so please move
1359 * along, nothing to see here..
1360 */
1361struct file *filp_open(const char *filename, int flags, umode_t mode)
1362{
1363 struct filename *name = getname_kernel(filename);
1364 struct file *file = ERR_CAST(name);
1365
1366 if (!IS_ERR(name)) {
1367 file = file_open_name(name, flags, mode);
1368 putname(name);
1369 }
1370 return file;
1371}
1372EXPORT_SYMBOL(filp_open);
1373
1374struct file *file_open_root(const struct path *root,
1375 const char *filename, int flags, umode_t mode)
1376{
1377 struct open_flags op;
1378 struct open_how how = build_open_how(flags, mode);
1379 int err = build_open_flags(&how, &op);
1380 if (err)
1381 return ERR_PTR(err);
1382 return do_file_open_root(root, filename, &op);
1383}
1384EXPORT_SYMBOL(file_open_root);
1385
1386static long do_sys_openat2(int dfd, const char __user *filename,
1387 struct open_how *how)
1388{
1389 struct open_flags op;
1390 int fd = build_open_flags(how, &op);
1391 struct filename *tmp;
1392
1393 if (fd)
1394 return fd;
1395
1396 tmp = getname(filename);
1397 if (IS_ERR(tmp))
1398 return PTR_ERR(tmp);
1399
1400 fd = get_unused_fd_flags(how->flags);
1401 if (fd >= 0) {
1402 struct file *f = do_filp_open(dfd, tmp, &op);
1403 if (IS_ERR(f)) {
1404 put_unused_fd(fd);
1405 fd = PTR_ERR(f);
1406 } else {
1407 fd_install(fd, f);
1408 }
1409 }
1410 putname(tmp);
1411 return fd;
1412}
1413
1414long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1415{
1416 struct open_how how = build_open_how(flags, mode);
1417 return do_sys_openat2(dfd, filename, &how);
1418}
1419
1420
1421SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1422{
1423 if (force_o_largefile())
1424 flags |= O_LARGEFILE;
1425 return do_sys_open(AT_FDCWD, filename, flags, mode);
1426}
1427
1428SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1429 umode_t, mode)
1430{
1431 if (force_o_largefile())
1432 flags |= O_LARGEFILE;
1433 return do_sys_open(dfd, filename, flags, mode);
1434}
1435
1436SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1437 struct open_how __user *, how, size_t, usize)
1438{
1439 int err;
1440 struct open_how tmp;
1441
1442 BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1443 BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1444
1445 if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1446 return -EINVAL;
1447 if (unlikely(usize > PAGE_SIZE))
1448 return -E2BIG;
1449
1450 err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1451 if (err)
1452 return err;
1453
1454 audit_openat2_how(&tmp);
1455
1456 /* O_LARGEFILE is only allowed for non-O_PATH. */
1457 if (!(tmp.flags & O_PATH) && force_o_largefile())
1458 tmp.flags |= O_LARGEFILE;
1459
1460 return do_sys_openat2(dfd, filename, &tmp);
1461}
1462
1463#ifdef CONFIG_COMPAT
1464/*
1465 * Exactly like sys_open(), except that it doesn't set the
1466 * O_LARGEFILE flag.
1467 */
1468COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1469{
1470 return do_sys_open(AT_FDCWD, filename, flags, mode);
1471}
1472
1473/*
1474 * Exactly like sys_openat(), except that it doesn't set the
1475 * O_LARGEFILE flag.
1476 */
1477COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1478{
1479 return do_sys_open(dfd, filename, flags, mode);
1480}
1481#endif
1482
1483#ifndef __alpha__
1484
1485/*
1486 * For backward compatibility? Maybe this should be moved
1487 * into arch/i386 instead?
1488 */
1489SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1490{
1491 int flags = O_CREAT | O_WRONLY | O_TRUNC;
1492
1493 if (force_o_largefile())
1494 flags |= O_LARGEFILE;
1495 return do_sys_open(AT_FDCWD, pathname, flags, mode);
1496}
1497#endif
1498
1499/*
1500 * "id" is the POSIX thread ID. We use the
1501 * files pointer for this..
1502 */
1503static int filp_flush(struct file *filp, fl_owner_t id)
1504{
1505 int retval = 0;
1506
1507 if (CHECK_DATA_CORRUPTION(file_count(filp) == 0,
1508 "VFS: Close: file count is 0 (f_op=%ps)",
1509 filp->f_op)) {
1510 return 0;
1511 }
1512
1513 if (filp->f_op->flush)
1514 retval = filp->f_op->flush(filp, id);
1515
1516 if (likely(!(filp->f_mode & FMODE_PATH))) {
1517 dnotify_flush(filp, id);
1518 locks_remove_posix(filp, id);
1519 }
1520 return retval;
1521}
1522
1523int filp_close(struct file *filp, fl_owner_t id)
1524{
1525 int retval;
1526
1527 retval = filp_flush(filp, id);
1528 fput(filp);
1529
1530 return retval;
1531}
1532EXPORT_SYMBOL(filp_close);
1533
1534/*
1535 * Careful here! We test whether the file pointer is NULL before
1536 * releasing the fd. This ensures that one clone task can't release
1537 * an fd while another clone is opening it.
1538 */
1539SYSCALL_DEFINE1(close, unsigned int, fd)
1540{
1541 int retval;
1542 struct file *file;
1543
1544 file = file_close_fd(fd);
1545 if (!file)
1546 return -EBADF;
1547
1548 retval = filp_flush(file, current->files);
1549
1550 /*
1551 * We're returning to user space. Don't bother
1552 * with any delayed fput() cases.
1553 */
1554 __fput_sync(file);
1555
1556 /* can't restart close syscall because file table entry was cleared */
1557 if (unlikely(retval == -ERESTARTSYS ||
1558 retval == -ERESTARTNOINTR ||
1559 retval == -ERESTARTNOHAND ||
1560 retval == -ERESTART_RESTARTBLOCK))
1561 retval = -EINTR;
1562
1563 return retval;
1564}
1565
1566/*
1567 * This routine simulates a hangup on the tty, to arrange that users
1568 * are given clean terminals at login time.
1569 */
1570SYSCALL_DEFINE0(vhangup)
1571{
1572 if (capable(CAP_SYS_TTY_CONFIG)) {
1573 tty_vhangup_self();
1574 return 0;
1575 }
1576 return -EPERM;
1577}
1578
1579/*
1580 * Called when an inode is about to be open.
1581 * We use this to disallow opening large files on 32bit systems if
1582 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1583 * on this flag in sys_open.
1584 */
1585int generic_file_open(struct inode * inode, struct file * filp)
1586{
1587 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1588 return -EOVERFLOW;
1589 return 0;
1590}
1591
1592EXPORT_SYMBOL(generic_file_open);
1593
1594/*
1595 * This is used by subsystems that don't want seekable
1596 * file descriptors. The function is not supposed to ever fail, the only
1597 * reason it returns an 'int' and not 'void' is so that it can be plugged
1598 * directly into file_operations structure.
1599 */
1600int nonseekable_open(struct inode *inode, struct file *filp)
1601{
1602 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1603 return 0;
1604}
1605
1606EXPORT_SYMBOL(nonseekable_open);
1607
1608/*
1609 * stream_open is used by subsystems that want stream-like file descriptors.
1610 * Such file descriptors are not seekable and don't have notion of position
1611 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1612 * Contrary to file descriptors of other regular files, .read() and .write()
1613 * can run simultaneously.
1614 *
1615 * stream_open never fails and is marked to return int so that it could be
1616 * directly used as file_operations.open .
1617 */
1618int stream_open(struct inode *inode, struct file *filp)
1619{
1620 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1621 filp->f_mode |= FMODE_STREAM;
1622 return 0;
1623}
1624
1625EXPORT_SYMBOL(stream_open);