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