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 <linux/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, and file capabilities on truncate too */
55 ret = dentry_needs_remove_privs(dentry);
56 if (ret < 0)
57 return ret;
58 if (ret)
59 newattrs.ia_valid |= ret | ATTR_FORCE;
60
61 inode_lock(dentry->d_inode);
62 /* Note any delegations or leases have already been broken: */
63 ret = notify_change(dentry, &newattrs, NULL);
64 inode_unlock(dentry->d_inode);
65 return ret;
66}
67
68long vfs_truncate(const struct path *path, loff_t length)
69{
70 struct inode *inode;
71 struct dentry *upperdentry;
72 long error;
73
74 inode = path->dentry->d_inode;
75
76 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
77 if (S_ISDIR(inode->i_mode))
78 return -EISDIR;
79 if (!S_ISREG(inode->i_mode))
80 return -EINVAL;
81
82 error = mnt_want_write(path->mnt);
83 if (error)
84 goto out;
85
86 error = inode_permission(inode, MAY_WRITE);
87 if (error)
88 goto mnt_drop_write_and_out;
89
90 error = -EPERM;
91 if (IS_APPEND(inode))
92 goto mnt_drop_write_and_out;
93
94 /*
95 * If this is an overlayfs then do as if opening the file so we get
96 * write access on the upper inode, not on the overlay inode. For
97 * non-overlay filesystems d_real() is an identity function.
98 */
99 upperdentry = d_real(path->dentry, NULL, O_WRONLY);
100 error = PTR_ERR(upperdentry);
101 if (IS_ERR(upperdentry))
102 goto mnt_drop_write_and_out;
103
104 error = get_write_access(upperdentry->d_inode);
105 if (error)
106 goto mnt_drop_write_and_out;
107
108 /*
109 * Make sure that there are no leases. get_write_access() protects
110 * against the truncate racing with a lease-granting setlease().
111 */
112 error = break_lease(inode, O_WRONLY);
113 if (error)
114 goto put_write_and_out;
115
116 error = locks_verify_truncate(inode, NULL, length);
117 if (!error)
118 error = security_path_truncate(path);
119 if (!error)
120 error = do_truncate(path->dentry, length, 0, NULL);
121
122put_write_and_out:
123 put_write_access(upperdentry->d_inode);
124mnt_drop_write_and_out:
125 mnt_drop_write(path->mnt);
126out:
127 return error;
128}
129EXPORT_SYMBOL_GPL(vfs_truncate);
130
131static long do_sys_truncate(const char __user *pathname, loff_t length)
132{
133 unsigned int lookup_flags = LOOKUP_FOLLOW;
134 struct path path;
135 int error;
136
137 if (length < 0) /* sorry, but loff_t says... */
138 return -EINVAL;
139
140retry:
141 error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
142 if (!error) {
143 error = vfs_truncate(&path, length);
144 path_put(&path);
145 }
146 if (retry_estale(error, lookup_flags)) {
147 lookup_flags |= LOOKUP_REVAL;
148 goto retry;
149 }
150 return error;
151}
152
153SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
154{
155 return do_sys_truncate(path, length);
156}
157
158#ifdef CONFIG_COMPAT
159COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
160{
161 return do_sys_truncate(path, length);
162}
163#endif
164
165static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
166{
167 struct inode *inode;
168 struct dentry *dentry;
169 struct fd f;
170 int error;
171
172 error = -EINVAL;
173 if (length < 0)
174 goto out;
175 error = -EBADF;
176 f = fdget(fd);
177 if (!f.file)
178 goto out;
179
180 /* explicitly opened as large or we are on 64-bit box */
181 if (f.file->f_flags & O_LARGEFILE)
182 small = 0;
183
184 dentry = f.file->f_path.dentry;
185 inode = dentry->d_inode;
186 error = -EINVAL;
187 if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
188 goto out_putf;
189
190 error = -EINVAL;
191 /* Cannot ftruncate over 2^31 bytes without large file support */
192 if (small && length > MAX_NON_LFS)
193 goto out_putf;
194
195 error = -EPERM;
196 if (IS_APPEND(inode))
197 goto out_putf;
198
199 sb_start_write(inode->i_sb);
200 error = locks_verify_truncate(inode, f.file, length);
201 if (!error)
202 error = security_path_truncate(&f.file->f_path);
203 if (!error)
204 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
205 sb_end_write(inode->i_sb);
206out_putf:
207 fdput(f);
208out:
209 return error;
210}
211
212SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
213{
214 return do_sys_ftruncate(fd, length, 1);
215}
216
217#ifdef CONFIG_COMPAT
218COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
219{
220 return do_sys_ftruncate(fd, length, 1);
221}
222#endif
223
224/* LFS versions of truncate are only needed on 32 bit machines */
225#if BITS_PER_LONG == 32
226SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
227{
228 return do_sys_truncate(path, length);
229}
230
231SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
232{
233 return do_sys_ftruncate(fd, length, 0);
234}
235#endif /* BITS_PER_LONG == 32 */
236
237
238int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
239{
240 struct inode *inode = file_inode(file);
241 long ret;
242
243 if (offset < 0 || len <= 0)
244 return -EINVAL;
245
246 /* Return error if mode is not supported */
247 if (mode & ~FALLOC_FL_SUPPORTED_MASK)
248 return -EOPNOTSUPP;
249
250 /* Punch hole and zero range are mutually exclusive */
251 if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
252 (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
253 return -EOPNOTSUPP;
254
255 /* Punch hole must have keep size set */
256 if ((mode & FALLOC_FL_PUNCH_HOLE) &&
257 !(mode & FALLOC_FL_KEEP_SIZE))
258 return -EOPNOTSUPP;
259
260 /* Collapse range should only be used exclusively. */
261 if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
262 (mode & ~FALLOC_FL_COLLAPSE_RANGE))
263 return -EINVAL;
264
265 /* Insert range should only be used exclusively. */
266 if ((mode & FALLOC_FL_INSERT_RANGE) &&
267 (mode & ~FALLOC_FL_INSERT_RANGE))
268 return -EINVAL;
269
270 /* Unshare range should only be used with allocate mode. */
271 if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
272 (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
273 return -EINVAL;
274
275 if (!(file->f_mode & FMODE_WRITE))
276 return -EBADF;
277
278 /*
279 * We can only allow pure fallocate on append only files
280 */
281 if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
282 return -EPERM;
283
284 if (IS_IMMUTABLE(inode))
285 return -EPERM;
286
287 /*
288 * We cannot allow any fallocate operation on an active swapfile
289 */
290 if (IS_SWAPFILE(inode))
291 return -ETXTBSY;
292
293 /*
294 * Revalidate the write permissions, in case security policy has
295 * changed since the files were opened.
296 */
297 ret = security_file_permission(file, MAY_WRITE);
298 if (ret)
299 return ret;
300
301 if (S_ISFIFO(inode->i_mode))
302 return -ESPIPE;
303
304 /*
305 * Let individual file system decide if it supports preallocation
306 * for directories or not.
307 */
308 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode) &&
309 !S_ISBLK(inode->i_mode))
310 return -ENODEV;
311
312 /* Check for wrap through zero too */
313 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
314 return -EFBIG;
315
316 if (!file->f_op->fallocate)
317 return -EOPNOTSUPP;
318
319 sb_start_write(inode->i_sb);
320 ret = file->f_op->fallocate(file, mode, offset, len);
321
322 /*
323 * Create inotify and fanotify events.
324 *
325 * To keep the logic simple always create events if fallocate succeeds.
326 * This implies that events are even created if the file size remains
327 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
328 */
329 if (ret == 0)
330 fsnotify_modify(file);
331
332 sb_end_write(inode->i_sb);
333 return ret;
334}
335EXPORT_SYMBOL_GPL(vfs_fallocate);
336
337SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
338{
339 struct fd f = fdget(fd);
340 int error = -EBADF;
341
342 if (f.file) {
343 error = vfs_fallocate(f.file, mode, offset, len);
344 fdput(f);
345 }
346 return error;
347}
348
349/*
350 * access() needs to use the real uid/gid, not the effective uid/gid.
351 * We do this by temporarily clearing all FS-related capabilities and
352 * switching the fsuid/fsgid around to the real ones.
353 */
354SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
355{
356 const struct cred *old_cred;
357 struct cred *override_cred;
358 struct path path;
359 struct inode *inode;
360 int res;
361 unsigned int lookup_flags = LOOKUP_FOLLOW;
362
363 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
364 return -EINVAL;
365
366 override_cred = prepare_creds();
367 if (!override_cred)
368 return -ENOMEM;
369
370 override_cred->fsuid = override_cred->uid;
371 override_cred->fsgid = override_cred->gid;
372
373 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
374 /* Clear the capabilities if we switch to a non-root user */
375 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
376 if (!uid_eq(override_cred->uid, root_uid))
377 cap_clear(override_cred->cap_effective);
378 else
379 override_cred->cap_effective =
380 override_cred->cap_permitted;
381 }
382
383 old_cred = override_creds(override_cred);
384retry:
385 res = user_path_at(dfd, filename, lookup_flags, &path);
386 if (res)
387 goto out;
388
389 inode = d_backing_inode(path.dentry);
390
391 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
392 /*
393 * MAY_EXEC on regular files is denied if the fs is mounted
394 * with the "noexec" flag.
395 */
396 res = -EACCES;
397 if (path_noexec(&path))
398 goto out_path_release;
399 }
400
401 res = inode_permission(inode, mode | MAY_ACCESS);
402 /* SuS v2 requires we report a read only fs too */
403 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
404 goto out_path_release;
405 /*
406 * This is a rare case where using __mnt_is_readonly()
407 * is OK without a mnt_want/drop_write() pair. Since
408 * no actual write to the fs is performed here, we do
409 * not need to telegraph to that to anyone.
410 *
411 * By doing this, we accept that this access is
412 * inherently racy and know that the fs may change
413 * state before we even see this result.
414 */
415 if (__mnt_is_readonly(path.mnt))
416 res = -EROFS;
417
418out_path_release:
419 path_put(&path);
420 if (retry_estale(res, lookup_flags)) {
421 lookup_flags |= LOOKUP_REVAL;
422 goto retry;
423 }
424out:
425 revert_creds(old_cred);
426 put_cred(override_cred);
427 return res;
428}
429
430SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
431{
432 return sys_faccessat(AT_FDCWD, filename, mode);
433}
434
435SYSCALL_DEFINE1(chdir, const char __user *, filename)
436{
437 struct path path;
438 int error;
439 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
440retry:
441 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
442 if (error)
443 goto out;
444
445 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
446 if (error)
447 goto dput_and_out;
448
449 set_fs_pwd(current->fs, &path);
450
451dput_and_out:
452 path_put(&path);
453 if (retry_estale(error, lookup_flags)) {
454 lookup_flags |= LOOKUP_REVAL;
455 goto retry;
456 }
457out:
458 return error;
459}
460
461SYSCALL_DEFINE1(fchdir, unsigned int, fd)
462{
463 struct fd f = fdget_raw(fd);
464 struct inode *inode;
465 int error = -EBADF;
466
467 error = -EBADF;
468 if (!f.file)
469 goto out;
470
471 inode = file_inode(f.file);
472
473 error = -ENOTDIR;
474 if (!S_ISDIR(inode->i_mode))
475 goto out_putf;
476
477 error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
478 if (!error)
479 set_fs_pwd(current->fs, &f.file->f_path);
480out_putf:
481 fdput(f);
482out:
483 return error;
484}
485
486SYSCALL_DEFINE1(chroot, const char __user *, filename)
487{
488 struct path path;
489 int error;
490 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
491retry:
492 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
493 if (error)
494 goto out;
495
496 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
497 if (error)
498 goto dput_and_out;
499
500 error = -EPERM;
501 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
502 goto dput_and_out;
503 error = security_path_chroot(&path);
504 if (error)
505 goto dput_and_out;
506
507 set_fs_root(current->fs, &path);
508 error = 0;
509dput_and_out:
510 path_put(&path);
511 if (retry_estale(error, lookup_flags)) {
512 lookup_flags |= LOOKUP_REVAL;
513 goto retry;
514 }
515out:
516 return error;
517}
518
519static int chmod_common(const struct path *path, umode_t mode)
520{
521 struct inode *inode = path->dentry->d_inode;
522 struct inode *delegated_inode = NULL;
523 struct iattr newattrs;
524 int error;
525
526 error = mnt_want_write(path->mnt);
527 if (error)
528 return error;
529retry_deleg:
530 inode_lock(inode);
531 error = security_path_chmod(path, mode);
532 if (error)
533 goto out_unlock;
534 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
535 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
536 error = notify_change(path->dentry, &newattrs, &delegated_inode);
537out_unlock:
538 inode_unlock(inode);
539 if (delegated_inode) {
540 error = break_deleg_wait(&delegated_inode);
541 if (!error)
542 goto retry_deleg;
543 }
544 mnt_drop_write(path->mnt);
545 return error;
546}
547
548SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
549{
550 struct fd f = fdget(fd);
551 int err = -EBADF;
552
553 if (f.file) {
554 audit_file(f.file);
555 err = chmod_common(&f.file->f_path, mode);
556 fdput(f);
557 }
558 return err;
559}
560
561SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
562{
563 struct path path;
564 int error;
565 unsigned int lookup_flags = LOOKUP_FOLLOW;
566retry:
567 error = user_path_at(dfd, filename, lookup_flags, &path);
568 if (!error) {
569 error = chmod_common(&path, mode);
570 path_put(&path);
571 if (retry_estale(error, lookup_flags)) {
572 lookup_flags |= LOOKUP_REVAL;
573 goto retry;
574 }
575 }
576 return error;
577}
578
579SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
580{
581 return sys_fchmodat(AT_FDCWD, filename, mode);
582}
583
584static int chown_common(const struct path *path, uid_t user, gid_t group)
585{
586 struct inode *inode = path->dentry->d_inode;
587 struct inode *delegated_inode = NULL;
588 int error;
589 struct iattr newattrs;
590 kuid_t uid;
591 kgid_t gid;
592
593 uid = make_kuid(current_user_ns(), user);
594 gid = make_kgid(current_user_ns(), group);
595
596retry_deleg:
597 newattrs.ia_valid = ATTR_CTIME;
598 if (user != (uid_t) -1) {
599 if (!uid_valid(uid))
600 return -EINVAL;
601 newattrs.ia_valid |= ATTR_UID;
602 newattrs.ia_uid = uid;
603 }
604 if (group != (gid_t) -1) {
605 if (!gid_valid(gid))
606 return -EINVAL;
607 newattrs.ia_valid |= ATTR_GID;
608 newattrs.ia_gid = gid;
609 }
610 if (!S_ISDIR(inode->i_mode))
611 newattrs.ia_valid |=
612 ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
613 inode_lock(inode);
614 error = security_path_chown(path, uid, gid);
615 if (!error)
616 error = notify_change(path->dentry, &newattrs, &delegated_inode);
617 inode_unlock(inode);
618 if (delegated_inode) {
619 error = break_deleg_wait(&delegated_inode);
620 if (!error)
621 goto retry_deleg;
622 }
623 return error;
624}
625
626SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
627 gid_t, group, int, flag)
628{
629 struct path path;
630 int error = -EINVAL;
631 int lookup_flags;
632
633 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
634 goto out;
635
636 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
637 if (flag & AT_EMPTY_PATH)
638 lookup_flags |= LOOKUP_EMPTY;
639retry:
640 error = user_path_at(dfd, filename, lookup_flags, &path);
641 if (error)
642 goto out;
643 error = mnt_want_write(path.mnt);
644 if (error)
645 goto out_release;
646 error = chown_common(&path, user, group);
647 mnt_drop_write(path.mnt);
648out_release:
649 path_put(&path);
650 if (retry_estale(error, lookup_flags)) {
651 lookup_flags |= LOOKUP_REVAL;
652 goto retry;
653 }
654out:
655 return error;
656}
657
658SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
659{
660 return sys_fchownat(AT_FDCWD, filename, user, group, 0);
661}
662
663SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
664{
665 return sys_fchownat(AT_FDCWD, filename, user, group,
666 AT_SYMLINK_NOFOLLOW);
667}
668
669SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
670{
671 struct fd f = fdget(fd);
672 int error = -EBADF;
673
674 if (!f.file)
675 goto out;
676
677 error = mnt_want_write_file(f.file);
678 if (error)
679 goto out_fput;
680 audit_file(f.file);
681 error = chown_common(&f.file->f_path, user, group);
682 mnt_drop_write_file(f.file);
683out_fput:
684 fdput(f);
685out:
686 return error;
687}
688
689int open_check_o_direct(struct file *f)
690{
691 /* NB: we're sure to have correct a_ops only after f_op->open */
692 if (f->f_flags & O_DIRECT) {
693 if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
694 return -EINVAL;
695 }
696 return 0;
697}
698
699static int do_dentry_open(struct file *f,
700 struct inode *inode,
701 int (*open)(struct inode *, struct file *),
702 const struct cred *cred)
703{
704 static const struct file_operations empty_fops = {};
705 int error;
706
707 f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
708 FMODE_PREAD | FMODE_PWRITE;
709
710 path_get(&f->f_path);
711 f->f_inode = inode;
712 f->f_mapping = inode->i_mapping;
713
714 if (unlikely(f->f_flags & O_PATH)) {
715 f->f_mode = FMODE_PATH;
716 f->f_op = &empty_fops;
717 return 0;
718 }
719
720 if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
721 error = get_write_access(inode);
722 if (unlikely(error))
723 goto cleanup_file;
724 error = __mnt_want_write(f->f_path.mnt);
725 if (unlikely(error)) {
726 put_write_access(inode);
727 goto cleanup_file;
728 }
729 f->f_mode |= FMODE_WRITER;
730 }
731
732 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
733 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
734 f->f_mode |= FMODE_ATOMIC_POS;
735
736 f->f_op = fops_get(inode->i_fop);
737 if (unlikely(WARN_ON(!f->f_op))) {
738 error = -ENODEV;
739 goto cleanup_all;
740 }
741
742 error = security_file_open(f, cred);
743 if (error)
744 goto cleanup_all;
745
746 error = break_lease(locks_inode(f), f->f_flags);
747 if (error)
748 goto cleanup_all;
749
750 if (!open)
751 open = f->f_op->open;
752 if (open) {
753 error = open(inode, f);
754 if (error)
755 goto cleanup_all;
756 }
757 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
758 i_readcount_inc(inode);
759 if ((f->f_mode & FMODE_READ) &&
760 likely(f->f_op->read || f->f_op->read_iter))
761 f->f_mode |= FMODE_CAN_READ;
762 if ((f->f_mode & FMODE_WRITE) &&
763 likely(f->f_op->write || f->f_op->write_iter))
764 f->f_mode |= FMODE_CAN_WRITE;
765
766 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
767
768 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
769
770 return 0;
771
772cleanup_all:
773 fops_put(f->f_op);
774 if (f->f_mode & FMODE_WRITER) {
775 put_write_access(inode);
776 __mnt_drop_write(f->f_path.mnt);
777 }
778cleanup_file:
779 path_put(&f->f_path);
780 f->f_path.mnt = NULL;
781 f->f_path.dentry = NULL;
782 f->f_inode = NULL;
783 return error;
784}
785
786/**
787 * finish_open - finish opening a file
788 * @file: file pointer
789 * @dentry: pointer to dentry
790 * @open: open callback
791 * @opened: state of open
792 *
793 * This can be used to finish opening a file passed to i_op->atomic_open().
794 *
795 * If the open callback is set to NULL, then the standard f_op->open()
796 * filesystem callback is substituted.
797 *
798 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
799 * the return value of d_splice_alias(), then the caller needs to perform dput()
800 * on it after finish_open().
801 *
802 * On successful return @file is a fully instantiated open file. After this, if
803 * an error occurs in ->atomic_open(), it needs to clean up with fput().
804 *
805 * Returns zero on success or -errno if the open failed.
806 */
807int finish_open(struct file *file, struct dentry *dentry,
808 int (*open)(struct inode *, struct file *),
809 int *opened)
810{
811 int error;
812 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
813
814 file->f_path.dentry = dentry;
815 error = do_dentry_open(file, d_backing_inode(dentry), open,
816 current_cred());
817 if (!error)
818 *opened |= FILE_OPENED;
819
820 return error;
821}
822EXPORT_SYMBOL(finish_open);
823
824/**
825 * finish_no_open - finish ->atomic_open() without opening the file
826 *
827 * @file: file pointer
828 * @dentry: dentry or NULL (as returned from ->lookup())
829 *
830 * This can be used to set the result of a successful lookup in ->atomic_open().
831 *
832 * NB: unlike finish_open() this function does consume the dentry reference and
833 * the caller need not dput() it.
834 *
835 * Returns "1" which must be the return value of ->atomic_open() after having
836 * called this function.
837 */
838int finish_no_open(struct file *file, struct dentry *dentry)
839{
840 file->f_path.dentry = dentry;
841 return 1;
842}
843EXPORT_SYMBOL(finish_no_open);
844
845char *file_path(struct file *filp, char *buf, int buflen)
846{
847 return d_path(&filp->f_path, buf, buflen);
848}
849EXPORT_SYMBOL(file_path);
850
851/**
852 * vfs_open - open the file at the given path
853 * @path: path to open
854 * @file: newly allocated file with f_flag initialized
855 * @cred: credentials to use
856 */
857int vfs_open(const struct path *path, struct file *file,
858 const struct cred *cred)
859{
860 struct dentry *dentry = d_real(path->dentry, NULL, file->f_flags);
861
862 if (IS_ERR(dentry))
863 return PTR_ERR(dentry);
864
865 file->f_path = *path;
866 return do_dentry_open(file, d_backing_inode(dentry), NULL, cred);
867}
868
869struct file *dentry_open(const struct path *path, int flags,
870 const struct cred *cred)
871{
872 int error;
873 struct file *f;
874
875 validate_creds(cred);
876
877 /* We must always pass in a valid mount pointer. */
878 BUG_ON(!path->mnt);
879
880 f = get_empty_filp();
881 if (!IS_ERR(f)) {
882 f->f_flags = flags;
883 error = vfs_open(path, f, cred);
884 if (!error) {
885 /* from now on we need fput() to dispose of f */
886 error = open_check_o_direct(f);
887 if (error) {
888 fput(f);
889 f = ERR_PTR(error);
890 }
891 } else {
892 put_filp(f);
893 f = ERR_PTR(error);
894 }
895 }
896 return f;
897}
898EXPORT_SYMBOL(dentry_open);
899
900static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
901{
902 int lookup_flags = 0;
903 int acc_mode = ACC_MODE(flags);
904
905 if (flags & (O_CREAT | __O_TMPFILE))
906 op->mode = (mode & S_IALLUGO) | S_IFREG;
907 else
908 op->mode = 0;
909
910 /* Must never be set by userspace */
911 flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
912
913 /*
914 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
915 * check for O_DSYNC if the need any syncing at all we enforce it's
916 * always set instead of having to deal with possibly weird behaviour
917 * for malicious applications setting only __O_SYNC.
918 */
919 if (flags & __O_SYNC)
920 flags |= O_DSYNC;
921
922 if (flags & __O_TMPFILE) {
923 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
924 return -EINVAL;
925 if (!(acc_mode & MAY_WRITE))
926 return -EINVAL;
927 } else if (flags & O_PATH) {
928 /*
929 * If we have O_PATH in the open flag. Then we
930 * cannot have anything other than the below set of flags
931 */
932 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
933 acc_mode = 0;
934 }
935
936 op->open_flag = flags;
937
938 /* O_TRUNC implies we need access checks for write permissions */
939 if (flags & O_TRUNC)
940 acc_mode |= MAY_WRITE;
941
942 /* Allow the LSM permission hook to distinguish append
943 access from general write access. */
944 if (flags & O_APPEND)
945 acc_mode |= MAY_APPEND;
946
947 op->acc_mode = acc_mode;
948
949 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
950
951 if (flags & O_CREAT) {
952 op->intent |= LOOKUP_CREATE;
953 if (flags & O_EXCL)
954 op->intent |= LOOKUP_EXCL;
955 }
956
957 if (flags & O_DIRECTORY)
958 lookup_flags |= LOOKUP_DIRECTORY;
959 if (!(flags & O_NOFOLLOW))
960 lookup_flags |= LOOKUP_FOLLOW;
961 op->lookup_flags = lookup_flags;
962 return 0;
963}
964
965/**
966 * file_open_name - open file and return file pointer
967 *
968 * @name: struct filename containing path to open
969 * @flags: open flags as per the open(2) second argument
970 * @mode: mode for the new file if O_CREAT is set, else ignored
971 *
972 * This is the helper to open a file from kernelspace if you really
973 * have to. But in generally you should not do this, so please move
974 * along, nothing to see here..
975 */
976struct file *file_open_name(struct filename *name, int flags, umode_t mode)
977{
978 struct open_flags op;
979 int err = build_open_flags(flags, mode, &op);
980 return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
981}
982
983/**
984 * filp_open - open file and return file pointer
985 *
986 * @filename: path to open
987 * @flags: open flags as per the open(2) second argument
988 * @mode: mode for the new file if O_CREAT is set, else ignored
989 *
990 * This is the helper to open a file from kernelspace if you really
991 * have to. But in generally you should not do this, so please move
992 * along, nothing to see here..
993 */
994struct file *filp_open(const char *filename, int flags, umode_t mode)
995{
996 struct filename *name = getname_kernel(filename);
997 struct file *file = ERR_CAST(name);
998
999 if (!IS_ERR(name)) {
1000 file = file_open_name(name, flags, mode);
1001 putname(name);
1002 }
1003 return file;
1004}
1005EXPORT_SYMBOL(filp_open);
1006
1007struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1008 const char *filename, int flags, umode_t mode)
1009{
1010 struct open_flags op;
1011 int err = build_open_flags(flags, mode, &op);
1012 if (err)
1013 return ERR_PTR(err);
1014 return do_file_open_root(dentry, mnt, filename, &op);
1015}
1016EXPORT_SYMBOL(file_open_root);
1017
1018struct file *filp_clone_open(struct file *oldfile)
1019{
1020 struct file *file;
1021 int retval;
1022
1023 file = get_empty_filp();
1024 if (IS_ERR(file))
1025 return file;
1026
1027 file->f_flags = oldfile->f_flags;
1028 retval = vfs_open(&oldfile->f_path, file, oldfile->f_cred);
1029 if (retval) {
1030 put_filp(file);
1031 return ERR_PTR(retval);
1032 }
1033
1034 return file;
1035}
1036EXPORT_SYMBOL(filp_clone_open);
1037
1038long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1039{
1040 struct open_flags op;
1041 int fd = build_open_flags(flags, mode, &op);
1042 struct filename *tmp;
1043
1044 if (fd)
1045 return fd;
1046
1047 tmp = getname(filename);
1048 if (IS_ERR(tmp))
1049 return PTR_ERR(tmp);
1050
1051 fd = get_unused_fd_flags(flags);
1052 if (fd >= 0) {
1053 struct file *f = do_filp_open(dfd, tmp, &op);
1054 if (IS_ERR(f)) {
1055 put_unused_fd(fd);
1056 fd = PTR_ERR(f);
1057 } else {
1058 fsnotify_open(f);
1059 fd_install(fd, f);
1060 }
1061 }
1062 putname(tmp);
1063 return fd;
1064}
1065
1066SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1067{
1068 if (force_o_largefile())
1069 flags |= O_LARGEFILE;
1070
1071 return do_sys_open(AT_FDCWD, filename, flags, mode);
1072}
1073
1074SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1075 umode_t, mode)
1076{
1077 if (force_o_largefile())
1078 flags |= O_LARGEFILE;
1079
1080 return do_sys_open(dfd, filename, flags, mode);
1081}
1082
1083#ifndef __alpha__
1084
1085/*
1086 * For backward compatibility? Maybe this should be moved
1087 * into arch/i386 instead?
1088 */
1089SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1090{
1091 return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1092}
1093
1094#endif
1095
1096/*
1097 * "id" is the POSIX thread ID. We use the
1098 * files pointer for this..
1099 */
1100int filp_close(struct file *filp, fl_owner_t id)
1101{
1102 int retval = 0;
1103
1104 if (!file_count(filp)) {
1105 printk(KERN_ERR "VFS: Close: file count is 0\n");
1106 return 0;
1107 }
1108
1109 if (filp->f_op->flush)
1110 retval = filp->f_op->flush(filp, id);
1111
1112 if (likely(!(filp->f_mode & FMODE_PATH))) {
1113 dnotify_flush(filp, id);
1114 locks_remove_posix(filp, id);
1115 }
1116 fput(filp);
1117 return retval;
1118}
1119
1120EXPORT_SYMBOL(filp_close);
1121
1122/*
1123 * Careful here! We test whether the file pointer is NULL before
1124 * releasing the fd. This ensures that one clone task can't release
1125 * an fd while another clone is opening it.
1126 */
1127SYSCALL_DEFINE1(close, unsigned int, fd)
1128{
1129 int retval = __close_fd(current->files, fd);
1130
1131 /* can't restart close syscall because file table entry was cleared */
1132 if (unlikely(retval == -ERESTARTSYS ||
1133 retval == -ERESTARTNOINTR ||
1134 retval == -ERESTARTNOHAND ||
1135 retval == -ERESTART_RESTARTBLOCK))
1136 retval = -EINTR;
1137
1138 return retval;
1139}
1140EXPORT_SYMBOL(sys_close);
1141
1142/*
1143 * This routine simulates a hangup on the tty, to arrange that users
1144 * are given clean terminals at login time.
1145 */
1146SYSCALL_DEFINE0(vhangup)
1147{
1148 if (capable(CAP_SYS_TTY_CONFIG)) {
1149 tty_vhangup_self();
1150 return 0;
1151 }
1152 return -EPERM;
1153}
1154
1155/*
1156 * Called when an inode is about to be open.
1157 * We use this to disallow opening large files on 32bit systems if
1158 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1159 * on this flag in sys_open.
1160 */
1161int generic_file_open(struct inode * inode, struct file * filp)
1162{
1163 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1164 return -EOVERFLOW;
1165 return 0;
1166}
1167
1168EXPORT_SYMBOL(generic_file_open);
1169
1170/*
1171 * This is used by subsystems that don't want seekable
1172 * file descriptors. The function is not supposed to ever fail, the only
1173 * reason it returns an 'int' and not 'void' is so that it can be plugged
1174 * directly into file_operations structure.
1175 */
1176int nonseekable_open(struct inode *inode, struct file *filp)
1177{
1178 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1179 return 0;
1180}
1181
1182EXPORT_SYMBOL(nonseekable_open);
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
34#include "internal.h"
35
36int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
37 struct file *filp)
38{
39 int ret;
40 struct iattr newattrs;
41
42 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
43 if (length < 0)
44 return -EINVAL;
45
46 newattrs.ia_size = length;
47 newattrs.ia_valid = ATTR_SIZE | time_attrs;
48 if (filp) {
49 newattrs.ia_file = filp;
50 newattrs.ia_valid |= ATTR_FILE;
51 }
52
53 /* Remove suid/sgid on truncate too */
54 ret = should_remove_suid(dentry);
55 if (ret)
56 newattrs.ia_valid |= ret | ATTR_FORCE;
57
58 mutex_lock(&dentry->d_inode->i_mutex);
59 ret = notify_change(dentry, &newattrs);
60 mutex_unlock(&dentry->d_inode->i_mutex);
61 return ret;
62}
63
64static long do_sys_truncate(const char __user *pathname, loff_t length)
65{
66 struct path path;
67 struct inode *inode;
68 int error;
69
70 error = -EINVAL;
71 if (length < 0) /* sorry, but loff_t says... */
72 goto out;
73
74 error = user_path(pathname, &path);
75 if (error)
76 goto out;
77 inode = path.dentry->d_inode;
78
79 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
80 error = -EISDIR;
81 if (S_ISDIR(inode->i_mode))
82 goto dput_and_out;
83
84 error = -EINVAL;
85 if (!S_ISREG(inode->i_mode))
86 goto dput_and_out;
87
88 error = mnt_want_write(path.mnt);
89 if (error)
90 goto dput_and_out;
91
92 error = inode_permission(inode, MAY_WRITE);
93 if (error)
94 goto mnt_drop_write_and_out;
95
96 error = -EPERM;
97 if (IS_APPEND(inode))
98 goto mnt_drop_write_and_out;
99
100 error = get_write_access(inode);
101 if (error)
102 goto mnt_drop_write_and_out;
103
104 /*
105 * Make sure that there are no leases. get_write_access() protects
106 * against the truncate racing with a lease-granting setlease().
107 */
108 error = break_lease(inode, O_WRONLY);
109 if (error)
110 goto put_write_and_out;
111
112 error = locks_verify_truncate(inode, NULL, length);
113 if (!error)
114 error = security_path_truncate(&path);
115 if (!error)
116 error = do_truncate(path.dentry, length, 0, NULL);
117
118put_write_and_out:
119 put_write_access(inode);
120mnt_drop_write_and_out:
121 mnt_drop_write(path.mnt);
122dput_and_out:
123 path_put(&path);
124out:
125 return error;
126}
127
128SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
129{
130 return do_sys_truncate(path, length);
131}
132
133static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
134{
135 struct inode * inode;
136 struct dentry *dentry;
137 struct file * file;
138 int error;
139
140 error = -EINVAL;
141 if (length < 0)
142 goto out;
143 error = -EBADF;
144 file = fget(fd);
145 if (!file)
146 goto out;
147
148 /* explicitly opened as large or we are on 64-bit box */
149 if (file->f_flags & O_LARGEFILE)
150 small = 0;
151
152 dentry = file->f_path.dentry;
153 inode = dentry->d_inode;
154 error = -EINVAL;
155 if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
156 goto out_putf;
157
158 error = -EINVAL;
159 /* Cannot ftruncate over 2^31 bytes without large file support */
160 if (small && length > MAX_NON_LFS)
161 goto out_putf;
162
163 error = -EPERM;
164 if (IS_APPEND(inode))
165 goto out_putf;
166
167 error = locks_verify_truncate(inode, file, length);
168 if (!error)
169 error = security_path_truncate(&file->f_path);
170 if (!error)
171 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
172out_putf:
173 fput(file);
174out:
175 return error;
176}
177
178SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
179{
180 long ret = do_sys_ftruncate(fd, length, 1);
181 /* avoid REGPARM breakage on x86: */
182 asmlinkage_protect(2, ret, fd, length);
183 return ret;
184}
185
186/* LFS versions of truncate are only needed on 32 bit machines */
187#if BITS_PER_LONG == 32
188SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
189{
190 return do_sys_truncate(path, length);
191}
192#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
193asmlinkage long SyS_truncate64(long path, loff_t length)
194{
195 return SYSC_truncate64((const char __user *) path, length);
196}
197SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
198#endif
199
200SYSCALL_DEFINE(ftruncate64)(unsigned int fd, loff_t length)
201{
202 long ret = do_sys_ftruncate(fd, length, 0);
203 /* avoid REGPARM breakage on x86: */
204 asmlinkage_protect(2, ret, fd, length);
205 return ret;
206}
207#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
208asmlinkage long SyS_ftruncate64(long fd, loff_t length)
209{
210 return SYSC_ftruncate64((unsigned int) fd, length);
211}
212SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
213#endif
214#endif /* BITS_PER_LONG == 32 */
215
216
217int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
218{
219 struct inode *inode = file->f_path.dentry->d_inode;
220 long ret;
221
222 if (offset < 0 || len <= 0)
223 return -EINVAL;
224
225 /* Return error if mode is not supported */
226 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
227 return -EOPNOTSUPP;
228
229 /* Punch hole must have keep size set */
230 if ((mode & FALLOC_FL_PUNCH_HOLE) &&
231 !(mode & FALLOC_FL_KEEP_SIZE))
232 return -EOPNOTSUPP;
233
234 if (!(file->f_mode & FMODE_WRITE))
235 return -EBADF;
236
237 /* It's not possible punch hole on append only file */
238 if (mode & FALLOC_FL_PUNCH_HOLE && IS_APPEND(inode))
239 return -EPERM;
240
241 if (IS_IMMUTABLE(inode))
242 return -EPERM;
243
244 /*
245 * Revalidate the write permissions, in case security policy has
246 * changed since the files were opened.
247 */
248 ret = security_file_permission(file, MAY_WRITE);
249 if (ret)
250 return ret;
251
252 if (S_ISFIFO(inode->i_mode))
253 return -ESPIPE;
254
255 /*
256 * Let individual file system decide if it supports preallocation
257 * for directories or not.
258 */
259 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
260 return -ENODEV;
261
262 /* Check for wrap through zero too */
263 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
264 return -EFBIG;
265
266 if (!file->f_op->fallocate)
267 return -EOPNOTSUPP;
268
269 return file->f_op->fallocate(file, mode, offset, len);
270}
271
272SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
273{
274 struct file *file;
275 int error = -EBADF;
276
277 file = fget(fd);
278 if (file) {
279 error = do_fallocate(file, mode, offset, len);
280 fput(file);
281 }
282
283 return error;
284}
285
286#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
287asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
288{
289 return SYSC_fallocate((int)fd, (int)mode, offset, len);
290}
291SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
292#endif
293
294/*
295 * access() needs to use the real uid/gid, not the effective uid/gid.
296 * We do this by temporarily clearing all FS-related capabilities and
297 * switching the fsuid/fsgid around to the real ones.
298 */
299SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
300{
301 const struct cred *old_cred;
302 struct cred *override_cred;
303 struct path path;
304 struct inode *inode;
305 int res;
306
307 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
308 return -EINVAL;
309
310 override_cred = prepare_creds();
311 if (!override_cred)
312 return -ENOMEM;
313
314 override_cred->fsuid = override_cred->uid;
315 override_cred->fsgid = override_cred->gid;
316
317 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
318 /* Clear the capabilities if we switch to a non-root user */
319 if (override_cred->uid)
320 cap_clear(override_cred->cap_effective);
321 else
322 override_cred->cap_effective =
323 override_cred->cap_permitted;
324 }
325
326 old_cred = override_creds(override_cred);
327
328 res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
329 if (res)
330 goto out;
331
332 inode = path.dentry->d_inode;
333
334 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
335 /*
336 * MAY_EXEC on regular files is denied if the fs is mounted
337 * with the "noexec" flag.
338 */
339 res = -EACCES;
340 if (path.mnt->mnt_flags & MNT_NOEXEC)
341 goto out_path_release;
342 }
343
344 res = inode_permission(inode, mode | MAY_ACCESS);
345 /* SuS v2 requires we report a read only fs too */
346 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
347 goto out_path_release;
348 /*
349 * This is a rare case where using __mnt_is_readonly()
350 * is OK without a mnt_want/drop_write() pair. Since
351 * no actual write to the fs is performed here, we do
352 * not need to telegraph to that to anyone.
353 *
354 * By doing this, we accept that this access is
355 * inherently racy and know that the fs may change
356 * state before we even see this result.
357 */
358 if (__mnt_is_readonly(path.mnt))
359 res = -EROFS;
360
361out_path_release:
362 path_put(&path);
363out:
364 revert_creds(old_cred);
365 put_cred(override_cred);
366 return res;
367}
368
369SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
370{
371 return sys_faccessat(AT_FDCWD, filename, mode);
372}
373
374SYSCALL_DEFINE1(chdir, const char __user *, filename)
375{
376 struct path path;
377 int error;
378
379 error = user_path_dir(filename, &path);
380 if (error)
381 goto out;
382
383 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
384 if (error)
385 goto dput_and_out;
386
387 set_fs_pwd(current->fs, &path);
388
389dput_and_out:
390 path_put(&path);
391out:
392 return error;
393}
394
395SYSCALL_DEFINE1(fchdir, unsigned int, fd)
396{
397 struct file *file;
398 struct inode *inode;
399 int error;
400
401 error = -EBADF;
402 file = fget(fd);
403 if (!file)
404 goto out;
405
406 inode = file->f_path.dentry->d_inode;
407
408 error = -ENOTDIR;
409 if (!S_ISDIR(inode->i_mode))
410 goto out_putf;
411
412 error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
413 if (!error)
414 set_fs_pwd(current->fs, &file->f_path);
415out_putf:
416 fput(file);
417out:
418 return error;
419}
420
421SYSCALL_DEFINE1(chroot, const char __user *, filename)
422{
423 struct path path;
424 int error;
425
426 error = user_path_dir(filename, &path);
427 if (error)
428 goto out;
429
430 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
431 if (error)
432 goto dput_and_out;
433
434 error = -EPERM;
435 if (!capable(CAP_SYS_CHROOT))
436 goto dput_and_out;
437 error = security_path_chroot(&path);
438 if (error)
439 goto dput_and_out;
440
441 set_fs_root(current->fs, &path);
442 error = 0;
443dput_and_out:
444 path_put(&path);
445out:
446 return error;
447}
448
449static int chmod_common(struct path *path, umode_t mode)
450{
451 struct inode *inode = path->dentry->d_inode;
452 struct iattr newattrs;
453 int error;
454
455 error = mnt_want_write(path->mnt);
456 if (error)
457 return error;
458 mutex_lock(&inode->i_mutex);
459 error = security_path_chmod(path->dentry, path->mnt, mode);
460 if (error)
461 goto out_unlock;
462 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
463 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
464 error = notify_change(path->dentry, &newattrs);
465out_unlock:
466 mutex_unlock(&inode->i_mutex);
467 mnt_drop_write(path->mnt);
468 return error;
469}
470
471SYSCALL_DEFINE2(fchmod, unsigned int, fd, mode_t, mode)
472{
473 struct file * file;
474 int err = -EBADF;
475
476 file = fget(fd);
477 if (file) {
478 audit_inode(NULL, file->f_path.dentry);
479 err = chmod_common(&file->f_path, mode);
480 fput(file);
481 }
482 return err;
483}
484
485SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, mode_t, mode)
486{
487 struct path path;
488 int error;
489
490 error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
491 if (!error) {
492 error = chmod_common(&path, mode);
493 path_put(&path);
494 }
495 return error;
496}
497
498SYSCALL_DEFINE2(chmod, const char __user *, filename, mode_t, mode)
499{
500 return sys_fchmodat(AT_FDCWD, filename, mode);
501}
502
503static int chown_common(struct path *path, uid_t user, gid_t group)
504{
505 struct inode *inode = path->dentry->d_inode;
506 int error;
507 struct iattr newattrs;
508
509 newattrs.ia_valid = ATTR_CTIME;
510 if (user != (uid_t) -1) {
511 newattrs.ia_valid |= ATTR_UID;
512 newattrs.ia_uid = user;
513 }
514 if (group != (gid_t) -1) {
515 newattrs.ia_valid |= ATTR_GID;
516 newattrs.ia_gid = group;
517 }
518 if (!S_ISDIR(inode->i_mode))
519 newattrs.ia_valid |=
520 ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
521 mutex_lock(&inode->i_mutex);
522 error = security_path_chown(path, user, group);
523 if (!error)
524 error = notify_change(path->dentry, &newattrs);
525 mutex_unlock(&inode->i_mutex);
526
527 return error;
528}
529
530SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
531{
532 struct path path;
533 int error;
534
535 error = user_path(filename, &path);
536 if (error)
537 goto out;
538 error = mnt_want_write(path.mnt);
539 if (error)
540 goto out_release;
541 error = chown_common(&path, user, group);
542 mnt_drop_write(path.mnt);
543out_release:
544 path_put(&path);
545out:
546 return error;
547}
548
549SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
550 gid_t, group, int, flag)
551{
552 struct path path;
553 int error = -EINVAL;
554 int lookup_flags;
555
556 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
557 goto out;
558
559 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
560 if (flag & AT_EMPTY_PATH)
561 lookup_flags |= LOOKUP_EMPTY;
562 error = user_path_at(dfd, filename, lookup_flags, &path);
563 if (error)
564 goto out;
565 error = mnt_want_write(path.mnt);
566 if (error)
567 goto out_release;
568 error = chown_common(&path, user, group);
569 mnt_drop_write(path.mnt);
570out_release:
571 path_put(&path);
572out:
573 return error;
574}
575
576SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
577{
578 struct path path;
579 int error;
580
581 error = user_lpath(filename, &path);
582 if (error)
583 goto out;
584 error = mnt_want_write(path.mnt);
585 if (error)
586 goto out_release;
587 error = chown_common(&path, user, group);
588 mnt_drop_write(path.mnt);
589out_release:
590 path_put(&path);
591out:
592 return error;
593}
594
595SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
596{
597 struct file * file;
598 int error = -EBADF;
599 struct dentry * dentry;
600
601 file = fget(fd);
602 if (!file)
603 goto out;
604
605 error = mnt_want_write_file(file);
606 if (error)
607 goto out_fput;
608 dentry = file->f_path.dentry;
609 audit_inode(NULL, dentry);
610 error = chown_common(&file->f_path, user, group);
611 mnt_drop_write(file->f_path.mnt);
612out_fput:
613 fput(file);
614out:
615 return error;
616}
617
618/*
619 * You have to be very careful that these write
620 * counts get cleaned up in error cases and
621 * upon __fput(). This should probably never
622 * be called outside of __dentry_open().
623 */
624static inline int __get_file_write_access(struct inode *inode,
625 struct vfsmount *mnt)
626{
627 int error;
628 error = get_write_access(inode);
629 if (error)
630 return error;
631 /*
632 * Do not take mount writer counts on
633 * special files since no writes to
634 * the mount itself will occur.
635 */
636 if (!special_file(inode->i_mode)) {
637 /*
638 * Balanced in __fput()
639 */
640 error = mnt_want_write(mnt);
641 if (error)
642 put_write_access(inode);
643 }
644 return error;
645}
646
647static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
648 struct file *f,
649 int (*open)(struct inode *, struct file *),
650 const struct cred *cred)
651{
652 static const struct file_operations empty_fops = {};
653 struct inode *inode;
654 int error;
655
656 f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
657 FMODE_PREAD | FMODE_PWRITE;
658
659 if (unlikely(f->f_flags & O_PATH))
660 f->f_mode = FMODE_PATH;
661
662 inode = dentry->d_inode;
663 if (f->f_mode & FMODE_WRITE) {
664 error = __get_file_write_access(inode, mnt);
665 if (error)
666 goto cleanup_file;
667 if (!special_file(inode->i_mode))
668 file_take_write(f);
669 }
670
671 f->f_mapping = inode->i_mapping;
672 f->f_path.dentry = dentry;
673 f->f_path.mnt = mnt;
674 f->f_pos = 0;
675 file_sb_list_add(f, inode->i_sb);
676
677 if (unlikely(f->f_mode & FMODE_PATH)) {
678 f->f_op = &empty_fops;
679 return f;
680 }
681
682 f->f_op = fops_get(inode->i_fop);
683
684 error = security_dentry_open(f, cred);
685 if (error)
686 goto cleanup_all;
687
688 if (!open && f->f_op)
689 open = f->f_op->open;
690 if (open) {
691 error = open(inode, f);
692 if (error)
693 goto cleanup_all;
694 }
695 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
696 i_readcount_inc(inode);
697
698 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
699
700 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
701
702 /* NB: we're sure to have correct a_ops only after f_op->open */
703 if (f->f_flags & O_DIRECT) {
704 if (!f->f_mapping->a_ops ||
705 ((!f->f_mapping->a_ops->direct_IO) &&
706 (!f->f_mapping->a_ops->get_xip_mem))) {
707 fput(f);
708 f = ERR_PTR(-EINVAL);
709 }
710 }
711
712 return f;
713
714cleanup_all:
715 fops_put(f->f_op);
716 if (f->f_mode & FMODE_WRITE) {
717 put_write_access(inode);
718 if (!special_file(inode->i_mode)) {
719 /*
720 * We don't consider this a real
721 * mnt_want/drop_write() pair
722 * because it all happenend right
723 * here, so just reset the state.
724 */
725 file_reset_write(f);
726 mnt_drop_write(mnt);
727 }
728 }
729 file_sb_list_del(f);
730 f->f_path.dentry = NULL;
731 f->f_path.mnt = NULL;
732cleanup_file:
733 put_filp(f);
734 dput(dentry);
735 mntput(mnt);
736 return ERR_PTR(error);
737}
738
739/**
740 * lookup_instantiate_filp - instantiates the open intent filp
741 * @nd: pointer to nameidata
742 * @dentry: pointer to dentry
743 * @open: open callback
744 *
745 * Helper for filesystems that want to use lookup open intents and pass back
746 * a fully instantiated struct file to the caller.
747 * This function is meant to be called from within a filesystem's
748 * lookup method.
749 * Beware of calling it for non-regular files! Those ->open methods might block
750 * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
751 * leading to a deadlock, as nobody can open that fifo anymore, because
752 * another process to open fifo will block on locked parent when doing lookup).
753 * Note that in case of error, nd->intent.open.file is destroyed, but the
754 * path information remains valid.
755 * If the open callback is set to NULL, then the standard f_op->open()
756 * filesystem callback is substituted.
757 */
758struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
759 int (*open)(struct inode *, struct file *))
760{
761 const struct cred *cred = current_cred();
762
763 if (IS_ERR(nd->intent.open.file))
764 goto out;
765 if (IS_ERR(dentry))
766 goto out_err;
767 nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
768 nd->intent.open.file,
769 open, cred);
770out:
771 return nd->intent.open.file;
772out_err:
773 release_open_intent(nd);
774 nd->intent.open.file = ERR_CAST(dentry);
775 goto out;
776}
777EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
778
779/**
780 * nameidata_to_filp - convert a nameidata to an open filp.
781 * @nd: pointer to nameidata
782 * @flags: open flags
783 *
784 * Note that this function destroys the original nameidata
785 */
786struct file *nameidata_to_filp(struct nameidata *nd)
787{
788 const struct cred *cred = current_cred();
789 struct file *filp;
790
791 /* Pick up the filp from the open intent */
792 filp = nd->intent.open.file;
793 nd->intent.open.file = NULL;
794
795 /* Has the filesystem initialised the file for us? */
796 if (filp->f_path.dentry == NULL) {
797 path_get(&nd->path);
798 filp = __dentry_open(nd->path.dentry, nd->path.mnt, filp,
799 NULL, cred);
800 }
801 return filp;
802}
803
804/*
805 * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
806 * error.
807 */
808struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
809 const struct cred *cred)
810{
811 int error;
812 struct file *f;
813
814 validate_creds(cred);
815
816 /* We must always pass in a valid mount pointer. */
817 BUG_ON(!mnt);
818
819 error = -ENFILE;
820 f = get_empty_filp();
821 if (f == NULL) {
822 dput(dentry);
823 mntput(mnt);
824 return ERR_PTR(error);
825 }
826
827 f->f_flags = flags;
828 return __dentry_open(dentry, mnt, f, NULL, cred);
829}
830EXPORT_SYMBOL(dentry_open);
831
832static void __put_unused_fd(struct files_struct *files, unsigned int fd)
833{
834 struct fdtable *fdt = files_fdtable(files);
835 __FD_CLR(fd, fdt->open_fds);
836 if (fd < files->next_fd)
837 files->next_fd = fd;
838}
839
840void put_unused_fd(unsigned int fd)
841{
842 struct files_struct *files = current->files;
843 spin_lock(&files->file_lock);
844 __put_unused_fd(files, fd);
845 spin_unlock(&files->file_lock);
846}
847
848EXPORT_SYMBOL(put_unused_fd);
849
850/*
851 * Install a file pointer in the fd array.
852 *
853 * The VFS is full of places where we drop the files lock between
854 * setting the open_fds bitmap and installing the file in the file
855 * array. At any such point, we are vulnerable to a dup2() race
856 * installing a file in the array before us. We need to detect this and
857 * fput() the struct file we are about to overwrite in this case.
858 *
859 * It should never happen - if we allow dup2() do it, _really_ bad things
860 * will follow.
861 */
862
863void fd_install(unsigned int fd, struct file *file)
864{
865 struct files_struct *files = current->files;
866 struct fdtable *fdt;
867 spin_lock(&files->file_lock);
868 fdt = files_fdtable(files);
869 BUG_ON(fdt->fd[fd] != NULL);
870 rcu_assign_pointer(fdt->fd[fd], file);
871 spin_unlock(&files->file_lock);
872}
873
874EXPORT_SYMBOL(fd_install);
875
876static inline int build_open_flags(int flags, int mode, struct open_flags *op)
877{
878 int lookup_flags = 0;
879 int acc_mode;
880
881 if (!(flags & O_CREAT))
882 mode = 0;
883 op->mode = mode;
884
885 /* Must never be set by userspace */
886 flags &= ~FMODE_NONOTIFY;
887
888 /*
889 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
890 * check for O_DSYNC if the need any syncing at all we enforce it's
891 * always set instead of having to deal with possibly weird behaviour
892 * for malicious applications setting only __O_SYNC.
893 */
894 if (flags & __O_SYNC)
895 flags |= O_DSYNC;
896
897 /*
898 * If we have O_PATH in the open flag. Then we
899 * cannot have anything other than the below set of flags
900 */
901 if (flags & O_PATH) {
902 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
903 acc_mode = 0;
904 } else {
905 acc_mode = MAY_OPEN | ACC_MODE(flags);
906 }
907
908 op->open_flag = flags;
909
910 /* O_TRUNC implies we need access checks for write permissions */
911 if (flags & O_TRUNC)
912 acc_mode |= MAY_WRITE;
913
914 /* Allow the LSM permission hook to distinguish append
915 access from general write access. */
916 if (flags & O_APPEND)
917 acc_mode |= MAY_APPEND;
918
919 op->acc_mode = acc_mode;
920
921 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
922
923 if (flags & O_CREAT) {
924 op->intent |= LOOKUP_CREATE;
925 if (flags & O_EXCL)
926 op->intent |= LOOKUP_EXCL;
927 }
928
929 if (flags & O_DIRECTORY)
930 lookup_flags |= LOOKUP_DIRECTORY;
931 if (!(flags & O_NOFOLLOW))
932 lookup_flags |= LOOKUP_FOLLOW;
933 return lookup_flags;
934}
935
936/**
937 * filp_open - open file and return file pointer
938 *
939 * @filename: path to open
940 * @flags: open flags as per the open(2) second argument
941 * @mode: mode for the new file if O_CREAT is set, else ignored
942 *
943 * This is the helper to open a file from kernelspace if you really
944 * have to. But in generally you should not do this, so please move
945 * along, nothing to see here..
946 */
947struct file *filp_open(const char *filename, int flags, int mode)
948{
949 struct open_flags op;
950 int lookup = build_open_flags(flags, mode, &op);
951 return do_filp_open(AT_FDCWD, filename, &op, lookup);
952}
953EXPORT_SYMBOL(filp_open);
954
955struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
956 const char *filename, int flags)
957{
958 struct open_flags op;
959 int lookup = build_open_flags(flags, 0, &op);
960 if (flags & O_CREAT)
961 return ERR_PTR(-EINVAL);
962 if (!filename && (flags & O_DIRECTORY))
963 if (!dentry->d_inode->i_op->lookup)
964 return ERR_PTR(-ENOTDIR);
965 return do_file_open_root(dentry, mnt, filename, &op, lookup);
966}
967EXPORT_SYMBOL(file_open_root);
968
969long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
970{
971 struct open_flags op;
972 int lookup = build_open_flags(flags, mode, &op);
973 char *tmp = getname(filename);
974 int fd = PTR_ERR(tmp);
975
976 if (!IS_ERR(tmp)) {
977 fd = get_unused_fd_flags(flags);
978 if (fd >= 0) {
979 struct file *f = do_filp_open(dfd, tmp, &op, lookup);
980 if (IS_ERR(f)) {
981 put_unused_fd(fd);
982 fd = PTR_ERR(f);
983 } else {
984 fsnotify_open(f);
985 fd_install(fd, f);
986 }
987 }
988 putname(tmp);
989 }
990 return fd;
991}
992
993SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, int, mode)
994{
995 long ret;
996
997 if (force_o_largefile())
998 flags |= O_LARGEFILE;
999
1000 ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1001 /* avoid REGPARM breakage on x86: */
1002 asmlinkage_protect(3, ret, filename, flags, mode);
1003 return ret;
1004}
1005
1006SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1007 int, mode)
1008{
1009 long ret;
1010
1011 if (force_o_largefile())
1012 flags |= O_LARGEFILE;
1013
1014 ret = do_sys_open(dfd, filename, flags, mode);
1015 /* avoid REGPARM breakage on x86: */
1016 asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1017 return ret;
1018}
1019
1020#ifndef __alpha__
1021
1022/*
1023 * For backward compatibility? Maybe this should be moved
1024 * into arch/i386 instead?
1025 */
1026SYSCALL_DEFINE2(creat, const char __user *, pathname, int, mode)
1027{
1028 return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1029}
1030
1031#endif
1032
1033/*
1034 * "id" is the POSIX thread ID. We use the
1035 * files pointer for this..
1036 */
1037int filp_close(struct file *filp, fl_owner_t id)
1038{
1039 int retval = 0;
1040
1041 if (!file_count(filp)) {
1042 printk(KERN_ERR "VFS: Close: file count is 0\n");
1043 return 0;
1044 }
1045
1046 if (filp->f_op && filp->f_op->flush)
1047 retval = filp->f_op->flush(filp, id);
1048
1049 if (likely(!(filp->f_mode & FMODE_PATH))) {
1050 dnotify_flush(filp, id);
1051 locks_remove_posix(filp, id);
1052 }
1053 fput(filp);
1054 return retval;
1055}
1056
1057EXPORT_SYMBOL(filp_close);
1058
1059/*
1060 * Careful here! We test whether the file pointer is NULL before
1061 * releasing the fd. This ensures that one clone task can't release
1062 * an fd while another clone is opening it.
1063 */
1064SYSCALL_DEFINE1(close, unsigned int, fd)
1065{
1066 struct file * filp;
1067 struct files_struct *files = current->files;
1068 struct fdtable *fdt;
1069 int retval;
1070
1071 spin_lock(&files->file_lock);
1072 fdt = files_fdtable(files);
1073 if (fd >= fdt->max_fds)
1074 goto out_unlock;
1075 filp = fdt->fd[fd];
1076 if (!filp)
1077 goto out_unlock;
1078 rcu_assign_pointer(fdt->fd[fd], NULL);
1079 FD_CLR(fd, fdt->close_on_exec);
1080 __put_unused_fd(files, fd);
1081 spin_unlock(&files->file_lock);
1082 retval = filp_close(filp, files);
1083
1084 /* can't restart close syscall because file table entry was cleared */
1085 if (unlikely(retval == -ERESTARTSYS ||
1086 retval == -ERESTARTNOINTR ||
1087 retval == -ERESTARTNOHAND ||
1088 retval == -ERESTART_RESTARTBLOCK))
1089 retval = -EINTR;
1090
1091 return retval;
1092
1093out_unlock:
1094 spin_unlock(&files->file_lock);
1095 return -EBADF;
1096}
1097EXPORT_SYMBOL(sys_close);
1098
1099/*
1100 * This routine simulates a hangup on the tty, to arrange that users
1101 * are given clean terminals at login time.
1102 */
1103SYSCALL_DEFINE0(vhangup)
1104{
1105 if (capable(CAP_SYS_TTY_CONFIG)) {
1106 tty_vhangup_self();
1107 return 0;
1108 }
1109 return -EPERM;
1110}
1111
1112/*
1113 * Called when an inode is about to be open.
1114 * We use this to disallow opening large files on 32bit systems if
1115 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1116 * on this flag in sys_open.
1117 */
1118int generic_file_open(struct inode * inode, struct file * filp)
1119{
1120 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1121 return -EOVERFLOW;
1122 return 0;
1123}
1124
1125EXPORT_SYMBOL(generic_file_open);
1126
1127/*
1128 * This is used by subsystems that don't want seekable
1129 * file descriptors. The function is not supposed to ever fail, the only
1130 * reason it returns an 'int' and not 'void' is so that it can be plugged
1131 * directly into file_operations structure.
1132 */
1133int nonseekable_open(struct inode *inode, struct file *filp)
1134{
1135 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1136 return 0;
1137}
1138
1139EXPORT_SYMBOL(nonseekable_open);