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