Loading...
1/*
2 * linux/fs/open.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7#include <linux/string.h>
8#include <linux/mm.h>
9#include <linux/file.h>
10#include <linux/fdtable.h>
11#include <linux/fsnotify.h>
12#include <linux/module.h>
13#include <linux/tty.h>
14#include <linux/namei.h>
15#include <linux/backing-dev.h>
16#include <linux/capability.h>
17#include <linux/securebits.h>
18#include <linux/security.h>
19#include <linux/mount.h>
20#include <linux/fcntl.h>
21#include <linux/slab.h>
22#include <linux/uaccess.h>
23#include <linux/fs.h>
24#include <linux/personality.h>
25#include <linux/pagemap.h>
26#include <linux/syscalls.h>
27#include <linux/rcupdate.h>
28#include <linux/audit.h>
29#include <linux/falloc.h>
30#include <linux/fs_struct.h>
31#include <linux/ima.h>
32#include <linux/dnotify.h>
33#include <linux/compat.h>
34
35#include "internal.h"
36
37int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
38 struct file *filp)
39{
40 int ret;
41 struct iattr newattrs;
42
43 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
44 if (length < 0)
45 return -EINVAL;
46
47 newattrs.ia_size = length;
48 newattrs.ia_valid = ATTR_SIZE | time_attrs;
49 if (filp) {
50 newattrs.ia_file = filp;
51 newattrs.ia_valid |= ATTR_FILE;
52 }
53
54 /* Remove suid, sgid, and file capabilities on truncate too */
55 ret = dentry_needs_remove_privs(dentry);
56 if (ret < 0)
57 return ret;
58 if (ret)
59 newattrs.ia_valid |= ret | ATTR_FORCE;
60
61 inode_lock(dentry->d_inode);
62 /* Note any delegations or leases have already been broken: */
63 ret = notify_change(dentry, &newattrs, NULL);
64 inode_unlock(dentry->d_inode);
65 return ret;
66}
67
68long vfs_truncate(const struct path *path, loff_t length)
69{
70 struct inode *inode;
71 struct dentry *upperdentry;
72 long error;
73
74 inode = path->dentry->d_inode;
75
76 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
77 if (S_ISDIR(inode->i_mode))
78 return -EISDIR;
79 if (!S_ISREG(inode->i_mode))
80 return -EINVAL;
81
82 error = mnt_want_write(path->mnt);
83 if (error)
84 goto out;
85
86 error = inode_permission(inode, MAY_WRITE);
87 if (error)
88 goto mnt_drop_write_and_out;
89
90 error = -EPERM;
91 if (IS_APPEND(inode))
92 goto mnt_drop_write_and_out;
93
94 /*
95 * If this is an overlayfs then do as if opening the file so we get
96 * write access on the upper inode, not on the overlay inode. For
97 * non-overlay filesystems d_real() is an identity function.
98 */
99 upperdentry = d_real(path->dentry, NULL, O_WRONLY);
100 error = PTR_ERR(upperdentry);
101 if (IS_ERR(upperdentry))
102 goto mnt_drop_write_and_out;
103
104 error = get_write_access(upperdentry->d_inode);
105 if (error)
106 goto mnt_drop_write_and_out;
107
108 /*
109 * Make sure that there are no leases. get_write_access() protects
110 * against the truncate racing with a lease-granting setlease().
111 */
112 error = break_lease(inode, O_WRONLY);
113 if (error)
114 goto put_write_and_out;
115
116 error = locks_verify_truncate(inode, NULL, length);
117 if (!error)
118 error = security_path_truncate(path);
119 if (!error)
120 error = do_truncate(path->dentry, length, 0, NULL);
121
122put_write_and_out:
123 put_write_access(upperdentry->d_inode);
124mnt_drop_write_and_out:
125 mnt_drop_write(path->mnt);
126out:
127 return error;
128}
129EXPORT_SYMBOL_GPL(vfs_truncate);
130
131static long do_sys_truncate(const char __user *pathname, loff_t length)
132{
133 unsigned int lookup_flags = LOOKUP_FOLLOW;
134 struct path path;
135 int error;
136
137 if (length < 0) /* sorry, but loff_t says... */
138 return -EINVAL;
139
140retry:
141 error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
142 if (!error) {
143 error = vfs_truncate(&path, length);
144 path_put(&path);
145 }
146 if (retry_estale(error, lookup_flags)) {
147 lookup_flags |= LOOKUP_REVAL;
148 goto retry;
149 }
150 return error;
151}
152
153SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
154{
155 return do_sys_truncate(path, length);
156}
157
158#ifdef CONFIG_COMPAT
159COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
160{
161 return do_sys_truncate(path, length);
162}
163#endif
164
165static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
166{
167 struct inode *inode;
168 struct dentry *dentry;
169 struct fd f;
170 int error;
171
172 error = -EINVAL;
173 if (length < 0)
174 goto out;
175 error = -EBADF;
176 f = fdget(fd);
177 if (!f.file)
178 goto out;
179
180 /* explicitly opened as large or we are on 64-bit box */
181 if (f.file->f_flags & O_LARGEFILE)
182 small = 0;
183
184 dentry = f.file->f_path.dentry;
185 inode = dentry->d_inode;
186 error = -EINVAL;
187 if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
188 goto out_putf;
189
190 error = -EINVAL;
191 /* Cannot ftruncate over 2^31 bytes without large file support */
192 if (small && length > MAX_NON_LFS)
193 goto out_putf;
194
195 error = -EPERM;
196 if (IS_APPEND(inode))
197 goto out_putf;
198
199 sb_start_write(inode->i_sb);
200 error = locks_verify_truncate(inode, f.file, length);
201 if (!error)
202 error = security_path_truncate(&f.file->f_path);
203 if (!error)
204 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
205 sb_end_write(inode->i_sb);
206out_putf:
207 fdput(f);
208out:
209 return error;
210}
211
212SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
213{
214 return do_sys_ftruncate(fd, length, 1);
215}
216
217#ifdef CONFIG_COMPAT
218COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
219{
220 return do_sys_ftruncate(fd, length, 1);
221}
222#endif
223
224/* LFS versions of truncate are only needed on 32 bit machines */
225#if BITS_PER_LONG == 32
226SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
227{
228 return do_sys_truncate(path, length);
229}
230
231SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
232{
233 return do_sys_ftruncate(fd, length, 0);
234}
235#endif /* BITS_PER_LONG == 32 */
236
237
238int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
239{
240 struct inode *inode = file_inode(file);
241 long ret;
242
243 if (offset < 0 || len <= 0)
244 return -EINVAL;
245
246 /* Return error if mode is not supported */
247 if (mode & ~FALLOC_FL_SUPPORTED_MASK)
248 return -EOPNOTSUPP;
249
250 /* Punch hole and zero range are mutually exclusive */
251 if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
252 (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
253 return -EOPNOTSUPP;
254
255 /* Punch hole must have keep size set */
256 if ((mode & FALLOC_FL_PUNCH_HOLE) &&
257 !(mode & FALLOC_FL_KEEP_SIZE))
258 return -EOPNOTSUPP;
259
260 /* Collapse range should only be used exclusively. */
261 if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
262 (mode & ~FALLOC_FL_COLLAPSE_RANGE))
263 return -EINVAL;
264
265 /* Insert range should only be used exclusively. */
266 if ((mode & FALLOC_FL_INSERT_RANGE) &&
267 (mode & ~FALLOC_FL_INSERT_RANGE))
268 return -EINVAL;
269
270 /* Unshare range should only be used with allocate mode. */
271 if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
272 (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
273 return -EINVAL;
274
275 if (!(file->f_mode & FMODE_WRITE))
276 return -EBADF;
277
278 /*
279 * We can only allow pure fallocate on append only files
280 */
281 if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
282 return -EPERM;
283
284 if (IS_IMMUTABLE(inode))
285 return -EPERM;
286
287 /*
288 * We cannot allow any fallocate operation on an active swapfile
289 */
290 if (IS_SWAPFILE(inode))
291 return -ETXTBSY;
292
293 /*
294 * Revalidate the write permissions, in case security policy has
295 * changed since the files were opened.
296 */
297 ret = security_file_permission(file, MAY_WRITE);
298 if (ret)
299 return ret;
300
301 if (S_ISFIFO(inode->i_mode))
302 return -ESPIPE;
303
304 /*
305 * Let individual file system decide if it supports preallocation
306 * for directories or not.
307 */
308 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode) &&
309 !S_ISBLK(inode->i_mode))
310 return -ENODEV;
311
312 /* Check for wrap through zero too */
313 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
314 return -EFBIG;
315
316 if (!file->f_op->fallocate)
317 return -EOPNOTSUPP;
318
319 sb_start_write(inode->i_sb);
320 ret = file->f_op->fallocate(file, mode, offset, len);
321
322 /*
323 * Create inotify and fanotify events.
324 *
325 * To keep the logic simple always create events if fallocate succeeds.
326 * This implies that events are even created if the file size remains
327 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
328 */
329 if (ret == 0)
330 fsnotify_modify(file);
331
332 sb_end_write(inode->i_sb);
333 return ret;
334}
335EXPORT_SYMBOL_GPL(vfs_fallocate);
336
337SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
338{
339 struct fd f = fdget(fd);
340 int error = -EBADF;
341
342 if (f.file) {
343 error = vfs_fallocate(f.file, mode, offset, len);
344 fdput(f);
345 }
346 return error;
347}
348
349/*
350 * access() needs to use the real uid/gid, not the effective uid/gid.
351 * We do this by temporarily clearing all FS-related capabilities and
352 * switching the fsuid/fsgid around to the real ones.
353 */
354SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
355{
356 const struct cred *old_cred;
357 struct cred *override_cred;
358 struct path path;
359 struct inode *inode;
360 int res;
361 unsigned int lookup_flags = LOOKUP_FOLLOW;
362
363 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
364 return -EINVAL;
365
366 override_cred = prepare_creds();
367 if (!override_cred)
368 return -ENOMEM;
369
370 override_cred->fsuid = override_cred->uid;
371 override_cred->fsgid = override_cred->gid;
372
373 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
374 /* Clear the capabilities if we switch to a non-root user */
375 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
376 if (!uid_eq(override_cred->uid, root_uid))
377 cap_clear(override_cred->cap_effective);
378 else
379 override_cred->cap_effective =
380 override_cred->cap_permitted;
381 }
382
383 old_cred = override_creds(override_cred);
384retry:
385 res = user_path_at(dfd, filename, lookup_flags, &path);
386 if (res)
387 goto out;
388
389 inode = d_backing_inode(path.dentry);
390
391 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
392 /*
393 * MAY_EXEC on regular files is denied if the fs is mounted
394 * with the "noexec" flag.
395 */
396 res = -EACCES;
397 if (path_noexec(&path))
398 goto out_path_release;
399 }
400
401 res = inode_permission(inode, mode | MAY_ACCESS);
402 /* SuS v2 requires we report a read only fs too */
403 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
404 goto out_path_release;
405 /*
406 * This is a rare case where using __mnt_is_readonly()
407 * is OK without a mnt_want/drop_write() pair. Since
408 * no actual write to the fs is performed here, we do
409 * not need to telegraph to that to anyone.
410 *
411 * By doing this, we accept that this access is
412 * inherently racy and know that the fs may change
413 * state before we even see this result.
414 */
415 if (__mnt_is_readonly(path.mnt))
416 res = -EROFS;
417
418out_path_release:
419 path_put(&path);
420 if (retry_estale(res, lookup_flags)) {
421 lookup_flags |= LOOKUP_REVAL;
422 goto retry;
423 }
424out:
425 revert_creds(old_cred);
426 put_cred(override_cred);
427 return res;
428}
429
430SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
431{
432 return sys_faccessat(AT_FDCWD, filename, mode);
433}
434
435SYSCALL_DEFINE1(chdir, const char __user *, filename)
436{
437 struct path path;
438 int error;
439 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
440retry:
441 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
442 if (error)
443 goto out;
444
445 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
446 if (error)
447 goto dput_and_out;
448
449 set_fs_pwd(current->fs, &path);
450
451dput_and_out:
452 path_put(&path);
453 if (retry_estale(error, lookup_flags)) {
454 lookup_flags |= LOOKUP_REVAL;
455 goto retry;
456 }
457out:
458 return error;
459}
460
461SYSCALL_DEFINE1(fchdir, unsigned int, fd)
462{
463 struct fd f = fdget_raw(fd);
464 struct inode *inode;
465 int error = -EBADF;
466
467 error = -EBADF;
468 if (!f.file)
469 goto out;
470
471 inode = file_inode(f.file);
472
473 error = -ENOTDIR;
474 if (!S_ISDIR(inode->i_mode))
475 goto out_putf;
476
477 error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
478 if (!error)
479 set_fs_pwd(current->fs, &f.file->f_path);
480out_putf:
481 fdput(f);
482out:
483 return error;
484}
485
486SYSCALL_DEFINE1(chroot, const char __user *, filename)
487{
488 struct path path;
489 int error;
490 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
491retry:
492 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
493 if (error)
494 goto out;
495
496 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
497 if (error)
498 goto dput_and_out;
499
500 error = -EPERM;
501 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
502 goto dput_and_out;
503 error = security_path_chroot(&path);
504 if (error)
505 goto dput_and_out;
506
507 set_fs_root(current->fs, &path);
508 error = 0;
509dput_and_out:
510 path_put(&path);
511 if (retry_estale(error, lookup_flags)) {
512 lookup_flags |= LOOKUP_REVAL;
513 goto retry;
514 }
515out:
516 return error;
517}
518
519static int chmod_common(const struct path *path, umode_t mode)
520{
521 struct inode *inode = path->dentry->d_inode;
522 struct inode *delegated_inode = NULL;
523 struct iattr newattrs;
524 int error;
525
526 error = mnt_want_write(path->mnt);
527 if (error)
528 return error;
529retry_deleg:
530 inode_lock(inode);
531 error = security_path_chmod(path, mode);
532 if (error)
533 goto out_unlock;
534 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
535 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
536 error = notify_change(path->dentry, &newattrs, &delegated_inode);
537out_unlock:
538 inode_unlock(inode);
539 if (delegated_inode) {
540 error = break_deleg_wait(&delegated_inode);
541 if (!error)
542 goto retry_deleg;
543 }
544 mnt_drop_write(path->mnt);
545 return error;
546}
547
548SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
549{
550 struct fd f = fdget(fd);
551 int err = -EBADF;
552
553 if (f.file) {
554 audit_file(f.file);
555 err = chmod_common(&f.file->f_path, mode);
556 fdput(f);
557 }
558 return err;
559}
560
561SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
562{
563 struct path path;
564 int error;
565 unsigned int lookup_flags = LOOKUP_FOLLOW;
566retry:
567 error = user_path_at(dfd, filename, lookup_flags, &path);
568 if (!error) {
569 error = chmod_common(&path, mode);
570 path_put(&path);
571 if (retry_estale(error, lookup_flags)) {
572 lookup_flags |= LOOKUP_REVAL;
573 goto retry;
574 }
575 }
576 return error;
577}
578
579SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
580{
581 return sys_fchmodat(AT_FDCWD, filename, mode);
582}
583
584static int chown_common(const struct path *path, uid_t user, gid_t group)
585{
586 struct inode *inode = path->dentry->d_inode;
587 struct inode *delegated_inode = NULL;
588 int error;
589 struct iattr newattrs;
590 kuid_t uid;
591 kgid_t gid;
592
593 uid = make_kuid(current_user_ns(), user);
594 gid = make_kgid(current_user_ns(), group);
595
596retry_deleg:
597 newattrs.ia_valid = ATTR_CTIME;
598 if (user != (uid_t) -1) {
599 if (!uid_valid(uid))
600 return -EINVAL;
601 newattrs.ia_valid |= ATTR_UID;
602 newattrs.ia_uid = uid;
603 }
604 if (group != (gid_t) -1) {
605 if (!gid_valid(gid))
606 return -EINVAL;
607 newattrs.ia_valid |= ATTR_GID;
608 newattrs.ia_gid = gid;
609 }
610 if (!S_ISDIR(inode->i_mode))
611 newattrs.ia_valid |=
612 ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
613 inode_lock(inode);
614 error = security_path_chown(path, uid, gid);
615 if (!error)
616 error = notify_change(path->dentry, &newattrs, &delegated_inode);
617 inode_unlock(inode);
618 if (delegated_inode) {
619 error = break_deleg_wait(&delegated_inode);
620 if (!error)
621 goto retry_deleg;
622 }
623 return error;
624}
625
626SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
627 gid_t, group, int, flag)
628{
629 struct path path;
630 int error = -EINVAL;
631 int lookup_flags;
632
633 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
634 goto out;
635
636 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
637 if (flag & AT_EMPTY_PATH)
638 lookup_flags |= LOOKUP_EMPTY;
639retry:
640 error = user_path_at(dfd, filename, lookup_flags, &path);
641 if (error)
642 goto out;
643 error = mnt_want_write(path.mnt);
644 if (error)
645 goto out_release;
646 error = chown_common(&path, user, group);
647 mnt_drop_write(path.mnt);
648out_release:
649 path_put(&path);
650 if (retry_estale(error, lookup_flags)) {
651 lookup_flags |= LOOKUP_REVAL;
652 goto retry;
653 }
654out:
655 return error;
656}
657
658SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
659{
660 return sys_fchownat(AT_FDCWD, filename, user, group, 0);
661}
662
663SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
664{
665 return sys_fchownat(AT_FDCWD, filename, user, group,
666 AT_SYMLINK_NOFOLLOW);
667}
668
669SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
670{
671 struct fd f = fdget(fd);
672 int error = -EBADF;
673
674 if (!f.file)
675 goto out;
676
677 error = mnt_want_write_file(f.file);
678 if (error)
679 goto out_fput;
680 audit_file(f.file);
681 error = chown_common(&f.file->f_path, user, group);
682 mnt_drop_write_file(f.file);
683out_fput:
684 fdput(f);
685out:
686 return error;
687}
688
689int open_check_o_direct(struct file *f)
690{
691 /* NB: we're sure to have correct a_ops only after f_op->open */
692 if (f->f_flags & O_DIRECT) {
693 if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
694 return -EINVAL;
695 }
696 return 0;
697}
698
699static int do_dentry_open(struct file *f,
700 struct inode *inode,
701 int (*open)(struct inode *, struct file *),
702 const struct cred *cred)
703{
704 static const struct file_operations empty_fops = {};
705 int error;
706
707 f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
708 FMODE_PREAD | FMODE_PWRITE;
709
710 path_get(&f->f_path);
711 f->f_inode = inode;
712 f->f_mapping = inode->i_mapping;
713
714 if (unlikely(f->f_flags & O_PATH)) {
715 f->f_mode = FMODE_PATH;
716 f->f_op = &empty_fops;
717 return 0;
718 }
719
720 if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
721 error = get_write_access(inode);
722 if (unlikely(error))
723 goto cleanup_file;
724 error = __mnt_want_write(f->f_path.mnt);
725 if (unlikely(error)) {
726 put_write_access(inode);
727 goto cleanup_file;
728 }
729 f->f_mode |= FMODE_WRITER;
730 }
731
732 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
733 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
734 f->f_mode |= FMODE_ATOMIC_POS;
735
736 f->f_op = fops_get(inode->i_fop);
737 if (unlikely(WARN_ON(!f->f_op))) {
738 error = -ENODEV;
739 goto cleanup_all;
740 }
741
742 error = security_file_open(f, cred);
743 if (error)
744 goto cleanup_all;
745
746 error = break_lease(locks_inode(f), f->f_flags);
747 if (error)
748 goto cleanup_all;
749
750 if (!open)
751 open = f->f_op->open;
752 if (open) {
753 error = open(inode, f);
754 if (error)
755 goto cleanup_all;
756 }
757 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
758 i_readcount_inc(inode);
759 if ((f->f_mode & FMODE_READ) &&
760 likely(f->f_op->read || f->f_op->read_iter))
761 f->f_mode |= FMODE_CAN_READ;
762 if ((f->f_mode & FMODE_WRITE) &&
763 likely(f->f_op->write || f->f_op->write_iter))
764 f->f_mode |= FMODE_CAN_WRITE;
765
766 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
767
768 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
769
770 return 0;
771
772cleanup_all:
773 fops_put(f->f_op);
774 if (f->f_mode & FMODE_WRITER) {
775 put_write_access(inode);
776 __mnt_drop_write(f->f_path.mnt);
777 }
778cleanup_file:
779 path_put(&f->f_path);
780 f->f_path.mnt = NULL;
781 f->f_path.dentry = NULL;
782 f->f_inode = NULL;
783 return error;
784}
785
786/**
787 * finish_open - finish opening a file
788 * @file: file pointer
789 * @dentry: pointer to dentry
790 * @open: open callback
791 * @opened: state of open
792 *
793 * This can be used to finish opening a file passed to i_op->atomic_open().
794 *
795 * If the open callback is set to NULL, then the standard f_op->open()
796 * filesystem callback is substituted.
797 *
798 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
799 * the return value of d_splice_alias(), then the caller needs to perform dput()
800 * on it after finish_open().
801 *
802 * On successful return @file is a fully instantiated open file. After this, if
803 * an error occurs in ->atomic_open(), it needs to clean up with fput().
804 *
805 * Returns zero on success or -errno if the open failed.
806 */
807int finish_open(struct file *file, struct dentry *dentry,
808 int (*open)(struct inode *, struct file *),
809 int *opened)
810{
811 int error;
812 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
813
814 file->f_path.dentry = dentry;
815 error = do_dentry_open(file, d_backing_inode(dentry), open,
816 current_cred());
817 if (!error)
818 *opened |= FILE_OPENED;
819
820 return error;
821}
822EXPORT_SYMBOL(finish_open);
823
824/**
825 * finish_no_open - finish ->atomic_open() without opening the file
826 *
827 * @file: file pointer
828 * @dentry: dentry or NULL (as returned from ->lookup())
829 *
830 * This can be used to set the result of a successful lookup in ->atomic_open().
831 *
832 * NB: unlike finish_open() this function does consume the dentry reference and
833 * the caller need not dput() it.
834 *
835 * Returns "1" which must be the return value of ->atomic_open() after having
836 * called this function.
837 */
838int finish_no_open(struct file *file, struct dentry *dentry)
839{
840 file->f_path.dentry = dentry;
841 return 1;
842}
843EXPORT_SYMBOL(finish_no_open);
844
845char *file_path(struct file *filp, char *buf, int buflen)
846{
847 return d_path(&filp->f_path, buf, buflen);
848}
849EXPORT_SYMBOL(file_path);
850
851/**
852 * vfs_open - open the file at the given path
853 * @path: path to open
854 * @file: newly allocated file with f_flag initialized
855 * @cred: credentials to use
856 */
857int vfs_open(const struct path *path, struct file *file,
858 const struct cred *cred)
859{
860 struct dentry *dentry = d_real(path->dentry, NULL, file->f_flags);
861
862 if (IS_ERR(dentry))
863 return PTR_ERR(dentry);
864
865 file->f_path = *path;
866 return do_dentry_open(file, d_backing_inode(dentry), NULL, cred);
867}
868
869struct file *dentry_open(const struct path *path, int flags,
870 const struct cred *cred)
871{
872 int error;
873 struct file *f;
874
875 validate_creds(cred);
876
877 /* We must always pass in a valid mount pointer. */
878 BUG_ON(!path->mnt);
879
880 f = get_empty_filp();
881 if (!IS_ERR(f)) {
882 f->f_flags = flags;
883 error = vfs_open(path, f, cred);
884 if (!error) {
885 /* from now on we need fput() to dispose of f */
886 error = open_check_o_direct(f);
887 if (error) {
888 fput(f);
889 f = ERR_PTR(error);
890 }
891 } else {
892 put_filp(f);
893 f = ERR_PTR(error);
894 }
895 }
896 return f;
897}
898EXPORT_SYMBOL(dentry_open);
899
900static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
901{
902 int lookup_flags = 0;
903 int acc_mode = ACC_MODE(flags);
904
905 if (flags & (O_CREAT | __O_TMPFILE))
906 op->mode = (mode & S_IALLUGO) | S_IFREG;
907 else
908 op->mode = 0;
909
910 /* Must never be set by userspace */
911 flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
912
913 /*
914 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
915 * check for O_DSYNC if the need any syncing at all we enforce it's
916 * always set instead of having to deal with possibly weird behaviour
917 * for malicious applications setting only __O_SYNC.
918 */
919 if (flags & __O_SYNC)
920 flags |= O_DSYNC;
921
922 if (flags & __O_TMPFILE) {
923 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
924 return -EINVAL;
925 if (!(acc_mode & MAY_WRITE))
926 return -EINVAL;
927 } else if (flags & O_PATH) {
928 /*
929 * If we have O_PATH in the open flag. Then we
930 * cannot have anything other than the below set of flags
931 */
932 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
933 acc_mode = 0;
934 }
935
936 op->open_flag = flags;
937
938 /* O_TRUNC implies we need access checks for write permissions */
939 if (flags & O_TRUNC)
940 acc_mode |= MAY_WRITE;
941
942 /* Allow the LSM permission hook to distinguish append
943 access from general write access. */
944 if (flags & O_APPEND)
945 acc_mode |= MAY_APPEND;
946
947 op->acc_mode = acc_mode;
948
949 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
950
951 if (flags & O_CREAT) {
952 op->intent |= LOOKUP_CREATE;
953 if (flags & O_EXCL)
954 op->intent |= LOOKUP_EXCL;
955 }
956
957 if (flags & O_DIRECTORY)
958 lookup_flags |= LOOKUP_DIRECTORY;
959 if (!(flags & O_NOFOLLOW))
960 lookup_flags |= LOOKUP_FOLLOW;
961 op->lookup_flags = lookup_flags;
962 return 0;
963}
964
965/**
966 * file_open_name - open file and return file pointer
967 *
968 * @name: struct filename containing path to open
969 * @flags: open flags as per the open(2) second argument
970 * @mode: mode for the new file if O_CREAT is set, else ignored
971 *
972 * This is the helper to open a file from kernelspace if you really
973 * have to. But in generally you should not do this, so please move
974 * along, nothing to see here..
975 */
976struct file *file_open_name(struct filename *name, int flags, umode_t mode)
977{
978 struct open_flags op;
979 int err = build_open_flags(flags, mode, &op);
980 return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
981}
982
983/**
984 * filp_open - open file and return file pointer
985 *
986 * @filename: path to open
987 * @flags: open flags as per the open(2) second argument
988 * @mode: mode for the new file if O_CREAT is set, else ignored
989 *
990 * This is the helper to open a file from kernelspace if you really
991 * have to. But in generally you should not do this, so please move
992 * along, nothing to see here..
993 */
994struct file *filp_open(const char *filename, int flags, umode_t mode)
995{
996 struct filename *name = getname_kernel(filename);
997 struct file *file = ERR_CAST(name);
998
999 if (!IS_ERR(name)) {
1000 file = file_open_name(name, flags, mode);
1001 putname(name);
1002 }
1003 return file;
1004}
1005EXPORT_SYMBOL(filp_open);
1006
1007struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1008 const char *filename, int flags, umode_t mode)
1009{
1010 struct open_flags op;
1011 int err = build_open_flags(flags, mode, &op);
1012 if (err)
1013 return ERR_PTR(err);
1014 return do_file_open_root(dentry, mnt, filename, &op);
1015}
1016EXPORT_SYMBOL(file_open_root);
1017
1018struct file *filp_clone_open(struct file *oldfile)
1019{
1020 struct file *file;
1021 int retval;
1022
1023 file = get_empty_filp();
1024 if (IS_ERR(file))
1025 return file;
1026
1027 file->f_flags = oldfile->f_flags;
1028 retval = vfs_open(&oldfile->f_path, file, oldfile->f_cred);
1029 if (retval) {
1030 put_filp(file);
1031 return ERR_PTR(retval);
1032 }
1033
1034 return file;
1035}
1036EXPORT_SYMBOL(filp_clone_open);
1037
1038long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1039{
1040 struct open_flags op;
1041 int fd = build_open_flags(flags, mode, &op);
1042 struct filename *tmp;
1043
1044 if (fd)
1045 return fd;
1046
1047 tmp = getname(filename);
1048 if (IS_ERR(tmp))
1049 return PTR_ERR(tmp);
1050
1051 fd = get_unused_fd_flags(flags);
1052 if (fd >= 0) {
1053 struct file *f = do_filp_open(dfd, tmp, &op);
1054 if (IS_ERR(f)) {
1055 put_unused_fd(fd);
1056 fd = PTR_ERR(f);
1057 } else {
1058 fsnotify_open(f);
1059 fd_install(fd, f);
1060 }
1061 }
1062 putname(tmp);
1063 return fd;
1064}
1065
1066SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1067{
1068 if (force_o_largefile())
1069 flags |= O_LARGEFILE;
1070
1071 return do_sys_open(AT_FDCWD, filename, flags, mode);
1072}
1073
1074SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1075 umode_t, mode)
1076{
1077 if (force_o_largefile())
1078 flags |= O_LARGEFILE;
1079
1080 return do_sys_open(dfd, filename, flags, mode);
1081}
1082
1083#ifndef __alpha__
1084
1085/*
1086 * For backward compatibility? Maybe this should be moved
1087 * into arch/i386 instead?
1088 */
1089SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1090{
1091 return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1092}
1093
1094#endif
1095
1096/*
1097 * "id" is the POSIX thread ID. We use the
1098 * files pointer for this..
1099 */
1100int filp_close(struct file *filp, fl_owner_t id)
1101{
1102 int retval = 0;
1103
1104 if (!file_count(filp)) {
1105 printk(KERN_ERR "VFS: Close: file count is 0\n");
1106 return 0;
1107 }
1108
1109 if (filp->f_op->flush)
1110 retval = filp->f_op->flush(filp, id);
1111
1112 if (likely(!(filp->f_mode & FMODE_PATH))) {
1113 dnotify_flush(filp, id);
1114 locks_remove_posix(filp, id);
1115 }
1116 fput(filp);
1117 return retval;
1118}
1119
1120EXPORT_SYMBOL(filp_close);
1121
1122/*
1123 * Careful here! We test whether the file pointer is NULL before
1124 * releasing the fd. This ensures that one clone task can't release
1125 * an fd while another clone is opening it.
1126 */
1127SYSCALL_DEFINE1(close, unsigned int, fd)
1128{
1129 int retval = __close_fd(current->files, fd);
1130
1131 /* can't restart close syscall because file table entry was cleared */
1132 if (unlikely(retval == -ERESTARTSYS ||
1133 retval == -ERESTARTNOINTR ||
1134 retval == -ERESTARTNOHAND ||
1135 retval == -ERESTART_RESTARTBLOCK))
1136 retval = -EINTR;
1137
1138 return retval;
1139}
1140EXPORT_SYMBOL(sys_close);
1141
1142/*
1143 * This routine simulates a hangup on the tty, to arrange that users
1144 * are given clean terminals at login time.
1145 */
1146SYSCALL_DEFINE0(vhangup)
1147{
1148 if (capable(CAP_SYS_TTY_CONFIG)) {
1149 tty_vhangup_self();
1150 return 0;
1151 }
1152 return -EPERM;
1153}
1154
1155/*
1156 * Called when an inode is about to be open.
1157 * We use this to disallow opening large files on 32bit systems if
1158 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1159 * on this flag in sys_open.
1160 */
1161int generic_file_open(struct inode * inode, struct file * filp)
1162{
1163 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1164 return -EOVERFLOW;
1165 return 0;
1166}
1167
1168EXPORT_SYMBOL(generic_file_open);
1169
1170/*
1171 * This is used by subsystems that don't want seekable
1172 * file descriptors. The function is not supposed to ever fail, the only
1173 * reason it returns an 'int' and not 'void' is so that it can be plugged
1174 * directly into file_operations structure.
1175 */
1176int nonseekable_open(struct inode *inode, struct file *filp)
1177{
1178 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1179 return 0;
1180}
1181
1182EXPORT_SYMBOL(nonseekable_open);
1/*
2 * linux/fs/open.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7#include <linux/string.h>
8#include <linux/mm.h>
9#include <linux/file.h>
10#include <linux/fdtable.h>
11#include <linux/fsnotify.h>
12#include <linux/module.h>
13#include <linux/tty.h>
14#include <linux/namei.h>
15#include <linux/backing-dev.h>
16#include <linux/capability.h>
17#include <linux/securebits.h>
18#include <linux/security.h>
19#include <linux/mount.h>
20#include <linux/fcntl.h>
21#include <linux/slab.h>
22#include <asm/uaccess.h>
23#include <linux/fs.h>
24#include <linux/personality.h>
25#include <linux/pagemap.h>
26#include <linux/syscalls.h>
27#include <linux/rcupdate.h>
28#include <linux/audit.h>
29#include <linux/falloc.h>
30#include <linux/fs_struct.h>
31#include <linux/ima.h>
32#include <linux/dnotify.h>
33#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);