Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/open.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8#include <linux/string.h>
9#include <linux/mm.h>
10#include <linux/file.h>
11#include <linux/fdtable.h>
12#include <linux/fsnotify.h>
13#include <linux/module.h>
14#include <linux/tty.h>
15#include <linux/namei.h>
16#include <linux/backing-dev.h>
17#include <linux/capability.h>
18#include <linux/securebits.h>
19#include <linux/security.h>
20#include <linux/mount.h>
21#include <linux/fcntl.h>
22#include <linux/slab.h>
23#include <linux/uaccess.h>
24#include <linux/fs.h>
25#include <linux/personality.h>
26#include <linux/pagemap.h>
27#include <linux/syscalls.h>
28#include <linux/rcupdate.h>
29#include <linux/audit.h>
30#include <linux/falloc.h>
31#include <linux/fs_struct.h>
32#include <linux/ima.h>
33#include <linux/dnotify.h>
34#include <linux/compat.h>
35#include <linux/mnt_idmapping.h>
36
37#include "internal.h"
38
39int do_truncate(struct user_namespace *mnt_userns, struct dentry *dentry,
40 loff_t length, unsigned int time_attrs, struct file *filp)
41{
42 int ret;
43 struct iattr newattrs;
44
45 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
46 if (length < 0)
47 return -EINVAL;
48
49 newattrs.ia_size = length;
50 newattrs.ia_valid = ATTR_SIZE | time_attrs;
51 if (filp) {
52 newattrs.ia_file = filp;
53 newattrs.ia_valid |= ATTR_FILE;
54 }
55
56 /* Remove suid, sgid, and file capabilities on truncate too */
57 ret = dentry_needs_remove_privs(mnt_userns, dentry);
58 if (ret < 0)
59 return ret;
60 if (ret)
61 newattrs.ia_valid |= ret | ATTR_FORCE;
62
63 inode_lock(dentry->d_inode);
64 /* Note any delegations or leases have already been broken: */
65 ret = notify_change(mnt_userns, dentry, &newattrs, NULL);
66 inode_unlock(dentry->d_inode);
67 return ret;
68}
69
70long vfs_truncate(const struct path *path, loff_t length)
71{
72 struct user_namespace *mnt_userns;
73 struct inode *inode;
74 long error;
75
76 inode = path->dentry->d_inode;
77
78 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
79 if (S_ISDIR(inode->i_mode))
80 return -EISDIR;
81 if (!S_ISREG(inode->i_mode))
82 return -EINVAL;
83
84 error = mnt_want_write(path->mnt);
85 if (error)
86 goto out;
87
88 mnt_userns = mnt_user_ns(path->mnt);
89 error = inode_permission(mnt_userns, inode, MAY_WRITE);
90 if (error)
91 goto mnt_drop_write_and_out;
92
93 error = -EPERM;
94 if (IS_APPEND(inode))
95 goto mnt_drop_write_and_out;
96
97 error = get_write_access(inode);
98 if (error)
99 goto mnt_drop_write_and_out;
100
101 /*
102 * Make sure that there are no leases. get_write_access() protects
103 * against the truncate racing with a lease-granting setlease().
104 */
105 error = break_lease(inode, O_WRONLY);
106 if (error)
107 goto put_write_and_out;
108
109 error = security_path_truncate(path);
110 if (!error)
111 error = do_truncate(mnt_userns, path->dentry, length, 0, NULL);
112
113put_write_and_out:
114 put_write_access(inode);
115mnt_drop_write_and_out:
116 mnt_drop_write(path->mnt);
117out:
118 return error;
119}
120EXPORT_SYMBOL_GPL(vfs_truncate);
121
122long do_sys_truncate(const char __user *pathname, loff_t length)
123{
124 unsigned int lookup_flags = LOOKUP_FOLLOW;
125 struct path path;
126 int error;
127
128 if (length < 0) /* sorry, but loff_t says... */
129 return -EINVAL;
130
131retry:
132 error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
133 if (!error) {
134 error = vfs_truncate(&path, length);
135 path_put(&path);
136 }
137 if (retry_estale(error, lookup_flags)) {
138 lookup_flags |= LOOKUP_REVAL;
139 goto retry;
140 }
141 return error;
142}
143
144SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
145{
146 return do_sys_truncate(path, length);
147}
148
149#ifdef CONFIG_COMPAT
150COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
151{
152 return do_sys_truncate(path, length);
153}
154#endif
155
156long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
157{
158 struct inode *inode;
159 struct dentry *dentry;
160 struct fd f;
161 int error;
162
163 error = -EINVAL;
164 if (length < 0)
165 goto out;
166 error = -EBADF;
167 f = fdget(fd);
168 if (!f.file)
169 goto out;
170
171 /* explicitly opened as large or we are on 64-bit box */
172 if (f.file->f_flags & O_LARGEFILE)
173 small = 0;
174
175 dentry = f.file->f_path.dentry;
176 inode = dentry->d_inode;
177 error = -EINVAL;
178 if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
179 goto out_putf;
180
181 error = -EINVAL;
182 /* Cannot ftruncate over 2^31 bytes without large file support */
183 if (small && length > MAX_NON_LFS)
184 goto out_putf;
185
186 error = -EPERM;
187 /* Check IS_APPEND on real upper inode */
188 if (IS_APPEND(file_inode(f.file)))
189 goto out_putf;
190 sb_start_write(inode->i_sb);
191 error = security_file_truncate(f.file);
192 if (!error)
193 error = do_truncate(file_mnt_user_ns(f.file), dentry, length,
194 ATTR_MTIME | ATTR_CTIME, f.file);
195 sb_end_write(inode->i_sb);
196out_putf:
197 fdput(f);
198out:
199 return error;
200}
201
202SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
203{
204 return do_sys_ftruncate(fd, length, 1);
205}
206
207#ifdef CONFIG_COMPAT
208COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
209{
210 return do_sys_ftruncate(fd, length, 1);
211}
212#endif
213
214/* LFS versions of truncate are only needed on 32 bit machines */
215#if BITS_PER_LONG == 32
216SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
217{
218 return do_sys_truncate(path, length);
219}
220
221SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
222{
223 return do_sys_ftruncate(fd, length, 0);
224}
225#endif /* BITS_PER_LONG == 32 */
226
227#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_TRUNCATE64)
228COMPAT_SYSCALL_DEFINE3(truncate64, const char __user *, pathname,
229 compat_arg_u64_dual(length))
230{
231 return ksys_truncate(pathname, compat_arg_u64_glue(length));
232}
233#endif
234
235#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FTRUNCATE64)
236COMPAT_SYSCALL_DEFINE3(ftruncate64, unsigned int, fd,
237 compat_arg_u64_dual(length))
238{
239 return ksys_ftruncate(fd, compat_arg_u64_glue(length));
240}
241#endif
242
243int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
244{
245 struct inode *inode = file_inode(file);
246 long ret;
247
248 if (offset < 0 || len <= 0)
249 return -EINVAL;
250
251 /* Return error if mode is not supported */
252 if (mode & ~FALLOC_FL_SUPPORTED_MASK)
253 return -EOPNOTSUPP;
254
255 /* Punch hole and zero range are mutually exclusive */
256 if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
257 (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
258 return -EOPNOTSUPP;
259
260 /* Punch hole must have keep size set */
261 if ((mode & FALLOC_FL_PUNCH_HOLE) &&
262 !(mode & FALLOC_FL_KEEP_SIZE))
263 return -EOPNOTSUPP;
264
265 /* Collapse range should only be used exclusively. */
266 if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
267 (mode & ~FALLOC_FL_COLLAPSE_RANGE))
268 return -EINVAL;
269
270 /* Insert range should only be used exclusively. */
271 if ((mode & FALLOC_FL_INSERT_RANGE) &&
272 (mode & ~FALLOC_FL_INSERT_RANGE))
273 return -EINVAL;
274
275 /* Unshare range should only be used with allocate mode. */
276 if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
277 (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
278 return -EINVAL;
279
280 if (!(file->f_mode & FMODE_WRITE))
281 return -EBADF;
282
283 /*
284 * We can only allow pure fallocate on append only files
285 */
286 if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
287 return -EPERM;
288
289 if (IS_IMMUTABLE(inode))
290 return -EPERM;
291
292 /*
293 * We cannot allow any fallocate operation on an active swapfile
294 */
295 if (IS_SWAPFILE(inode))
296 return -ETXTBSY;
297
298 /*
299 * Revalidate the write permissions, in case security policy has
300 * changed since the files were opened.
301 */
302 ret = security_file_permission(file, MAY_WRITE);
303 if (ret)
304 return ret;
305
306 if (S_ISFIFO(inode->i_mode))
307 return -ESPIPE;
308
309 if (S_ISDIR(inode->i_mode))
310 return -EISDIR;
311
312 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
313 return -ENODEV;
314
315 /* Check for wrap through zero too */
316 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
317 return -EFBIG;
318
319 if (!file->f_op->fallocate)
320 return -EOPNOTSUPP;
321
322 file_start_write(file);
323 ret = file->f_op->fallocate(file, mode, offset, len);
324
325 /*
326 * Create inotify and fanotify events.
327 *
328 * To keep the logic simple always create events if fallocate succeeds.
329 * This implies that events are even created if the file size remains
330 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
331 */
332 if (ret == 0)
333 fsnotify_modify(file);
334
335 file_end_write(file);
336 return ret;
337}
338EXPORT_SYMBOL_GPL(vfs_fallocate);
339
340int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
341{
342 struct fd f = fdget(fd);
343 int error = -EBADF;
344
345 if (f.file) {
346 error = vfs_fallocate(f.file, mode, offset, len);
347 fdput(f);
348 }
349 return error;
350}
351
352SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
353{
354 return ksys_fallocate(fd, mode, offset, len);
355}
356
357#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FALLOCATE)
358COMPAT_SYSCALL_DEFINE6(fallocate, int, fd, int, mode, compat_arg_u64_dual(offset),
359 compat_arg_u64_dual(len))
360{
361 return ksys_fallocate(fd, mode, compat_arg_u64_glue(offset),
362 compat_arg_u64_glue(len));
363}
364#endif
365
366/*
367 * access() needs to use the real uid/gid, not the effective uid/gid.
368 * We do this by temporarily clearing all FS-related capabilities and
369 * switching the fsuid/fsgid around to the real ones.
370 */
371static const struct cred *access_override_creds(void)
372{
373 const struct cred *old_cred;
374 struct cred *override_cred;
375
376 override_cred = prepare_creds();
377 if (!override_cred)
378 return NULL;
379
380 override_cred->fsuid = override_cred->uid;
381 override_cred->fsgid = override_cred->gid;
382
383 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
384 /* Clear the capabilities if we switch to a non-root user */
385 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
386 if (!uid_eq(override_cred->uid, root_uid))
387 cap_clear(override_cred->cap_effective);
388 else
389 override_cred->cap_effective =
390 override_cred->cap_permitted;
391 }
392
393 /*
394 * The new set of credentials can *only* be used in
395 * task-synchronous circumstances, and does not need
396 * RCU freeing, unless somebody then takes a separate
397 * reference to it.
398 *
399 * NOTE! This is _only_ true because this credential
400 * is used purely for override_creds() that installs
401 * it as the subjective cred. Other threads will be
402 * accessing ->real_cred, not the subjective cred.
403 *
404 * If somebody _does_ make a copy of this (using the
405 * 'get_current_cred()' function), that will clear the
406 * non_rcu field, because now that other user may be
407 * expecting RCU freeing. But normal thread-synchronous
408 * cred accesses will keep things non-RCY.
409 */
410 override_cred->non_rcu = 1;
411
412 old_cred = override_creds(override_cred);
413
414 /* override_cred() gets its own ref */
415 put_cred(override_cred);
416
417 return old_cred;
418}
419
420static long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
421{
422 struct path path;
423 struct inode *inode;
424 int res;
425 unsigned int lookup_flags = LOOKUP_FOLLOW;
426 const struct cred *old_cred = NULL;
427
428 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
429 return -EINVAL;
430
431 if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
432 return -EINVAL;
433
434 if (flags & AT_SYMLINK_NOFOLLOW)
435 lookup_flags &= ~LOOKUP_FOLLOW;
436 if (flags & AT_EMPTY_PATH)
437 lookup_flags |= LOOKUP_EMPTY;
438
439 if (!(flags & AT_EACCESS)) {
440 old_cred = access_override_creds();
441 if (!old_cred)
442 return -ENOMEM;
443 }
444
445retry:
446 res = user_path_at(dfd, filename, lookup_flags, &path);
447 if (res)
448 goto out;
449
450 inode = d_backing_inode(path.dentry);
451
452 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
453 /*
454 * MAY_EXEC on regular files is denied if the fs is mounted
455 * with the "noexec" flag.
456 */
457 res = -EACCES;
458 if (path_noexec(&path))
459 goto out_path_release;
460 }
461
462 res = inode_permission(mnt_user_ns(path.mnt), inode, mode | MAY_ACCESS);
463 /* SuS v2 requires we report a read only fs too */
464 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
465 goto out_path_release;
466 /*
467 * This is a rare case where using __mnt_is_readonly()
468 * is OK without a mnt_want/drop_write() pair. Since
469 * no actual write to the fs is performed here, we do
470 * not need to telegraph to that to anyone.
471 *
472 * By doing this, we accept that this access is
473 * inherently racy and know that the fs may change
474 * state before we even see this result.
475 */
476 if (__mnt_is_readonly(path.mnt))
477 res = -EROFS;
478
479out_path_release:
480 path_put(&path);
481 if (retry_estale(res, lookup_flags)) {
482 lookup_flags |= LOOKUP_REVAL;
483 goto retry;
484 }
485out:
486 if (old_cred)
487 revert_creds(old_cred);
488
489 return res;
490}
491
492SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
493{
494 return do_faccessat(dfd, filename, mode, 0);
495}
496
497SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
498 int, flags)
499{
500 return do_faccessat(dfd, filename, mode, flags);
501}
502
503SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
504{
505 return do_faccessat(AT_FDCWD, filename, mode, 0);
506}
507
508SYSCALL_DEFINE1(chdir, const char __user *, filename)
509{
510 struct path path;
511 int error;
512 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
513retry:
514 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
515 if (error)
516 goto out;
517
518 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
519 if (error)
520 goto dput_and_out;
521
522 set_fs_pwd(current->fs, &path);
523
524dput_and_out:
525 path_put(&path);
526 if (retry_estale(error, lookup_flags)) {
527 lookup_flags |= LOOKUP_REVAL;
528 goto retry;
529 }
530out:
531 return error;
532}
533
534SYSCALL_DEFINE1(fchdir, unsigned int, fd)
535{
536 struct fd f = fdget_raw(fd);
537 int error;
538
539 error = -EBADF;
540 if (!f.file)
541 goto out;
542
543 error = -ENOTDIR;
544 if (!d_can_lookup(f.file->f_path.dentry))
545 goto out_putf;
546
547 error = file_permission(f.file, MAY_EXEC | MAY_CHDIR);
548 if (!error)
549 set_fs_pwd(current->fs, &f.file->f_path);
550out_putf:
551 fdput(f);
552out:
553 return error;
554}
555
556SYSCALL_DEFINE1(chroot, const char __user *, filename)
557{
558 struct path path;
559 int error;
560 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
561retry:
562 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
563 if (error)
564 goto out;
565
566 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
567 if (error)
568 goto dput_and_out;
569
570 error = -EPERM;
571 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
572 goto dput_and_out;
573 error = security_path_chroot(&path);
574 if (error)
575 goto dput_and_out;
576
577 set_fs_root(current->fs, &path);
578 error = 0;
579dput_and_out:
580 path_put(&path);
581 if (retry_estale(error, lookup_flags)) {
582 lookup_flags |= LOOKUP_REVAL;
583 goto retry;
584 }
585out:
586 return error;
587}
588
589int chmod_common(const struct path *path, umode_t mode)
590{
591 struct inode *inode = path->dentry->d_inode;
592 struct inode *delegated_inode = NULL;
593 struct iattr newattrs;
594 int error;
595
596 error = mnt_want_write(path->mnt);
597 if (error)
598 return error;
599retry_deleg:
600 inode_lock(inode);
601 error = security_path_chmod(path, mode);
602 if (error)
603 goto out_unlock;
604 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
605 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
606 error = notify_change(mnt_user_ns(path->mnt), path->dentry,
607 &newattrs, &delegated_inode);
608out_unlock:
609 inode_unlock(inode);
610 if (delegated_inode) {
611 error = break_deleg_wait(&delegated_inode);
612 if (!error)
613 goto retry_deleg;
614 }
615 mnt_drop_write(path->mnt);
616 return error;
617}
618
619int vfs_fchmod(struct file *file, umode_t mode)
620{
621 audit_file(file);
622 return chmod_common(&file->f_path, mode);
623}
624
625SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
626{
627 struct fd f = fdget(fd);
628 int err = -EBADF;
629
630 if (f.file) {
631 err = vfs_fchmod(f.file, mode);
632 fdput(f);
633 }
634 return err;
635}
636
637static int do_fchmodat(int dfd, const char __user *filename, umode_t mode)
638{
639 struct path path;
640 int error;
641 unsigned int lookup_flags = LOOKUP_FOLLOW;
642retry:
643 error = user_path_at(dfd, filename, lookup_flags, &path);
644 if (!error) {
645 error = chmod_common(&path, mode);
646 path_put(&path);
647 if (retry_estale(error, lookup_flags)) {
648 lookup_flags |= LOOKUP_REVAL;
649 goto retry;
650 }
651 }
652 return error;
653}
654
655SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
656 umode_t, mode)
657{
658 return do_fchmodat(dfd, filename, mode);
659}
660
661SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
662{
663 return do_fchmodat(AT_FDCWD, filename, mode);
664}
665
666/**
667 * setattr_vfsuid - check and set ia_fsuid attribute
668 * @kuid: new inode owner
669 *
670 * Check whether @kuid is valid and if so generate and set vfsuid_t in
671 * ia_vfsuid.
672 *
673 * Return: true if @kuid is valid, false if not.
674 */
675static inline bool setattr_vfsuid(struct iattr *attr, kuid_t kuid)
676{
677 if (!uid_valid(kuid))
678 return false;
679 attr->ia_valid |= ATTR_UID;
680 attr->ia_vfsuid = VFSUIDT_INIT(kuid);
681 return true;
682}
683
684/**
685 * setattr_vfsgid - check and set ia_fsgid attribute
686 * @kgid: new inode owner
687 *
688 * Check whether @kgid is valid and if so generate and set vfsgid_t in
689 * ia_vfsgid.
690 *
691 * Return: true if @kgid is valid, false if not.
692 */
693static inline bool setattr_vfsgid(struct iattr *attr, kgid_t kgid)
694{
695 if (!gid_valid(kgid))
696 return false;
697 attr->ia_valid |= ATTR_GID;
698 attr->ia_vfsgid = VFSGIDT_INIT(kgid);
699 return true;
700}
701
702int chown_common(const struct path *path, uid_t user, gid_t group)
703{
704 struct user_namespace *mnt_userns, *fs_userns;
705 struct inode *inode = path->dentry->d_inode;
706 struct inode *delegated_inode = NULL;
707 int error;
708 struct iattr newattrs;
709 kuid_t uid;
710 kgid_t gid;
711
712 uid = make_kuid(current_user_ns(), user);
713 gid = make_kgid(current_user_ns(), group);
714
715 mnt_userns = mnt_user_ns(path->mnt);
716 fs_userns = i_user_ns(inode);
717
718retry_deleg:
719 newattrs.ia_vfsuid = INVALID_VFSUID;
720 newattrs.ia_vfsgid = INVALID_VFSGID;
721 newattrs.ia_valid = ATTR_CTIME;
722 if ((user != (uid_t)-1) && !setattr_vfsuid(&newattrs, uid))
723 return -EINVAL;
724 if ((group != (gid_t)-1) && !setattr_vfsgid(&newattrs, gid))
725 return -EINVAL;
726 inode_lock(inode);
727 if (!S_ISDIR(inode->i_mode))
728 newattrs.ia_valid |= ATTR_KILL_SUID | ATTR_KILL_PRIV |
729 setattr_should_drop_sgid(mnt_userns, inode);
730 /* Continue to send actual fs values, not the mount values. */
731 error = security_path_chown(
732 path,
733 from_vfsuid(mnt_userns, fs_userns, newattrs.ia_vfsuid),
734 from_vfsgid(mnt_userns, fs_userns, newattrs.ia_vfsgid));
735 if (!error)
736 error = notify_change(mnt_userns, path->dentry, &newattrs,
737 &delegated_inode);
738 inode_unlock(inode);
739 if (delegated_inode) {
740 error = break_deleg_wait(&delegated_inode);
741 if (!error)
742 goto retry_deleg;
743 }
744 return error;
745}
746
747int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
748 int flag)
749{
750 struct path path;
751 int error = -EINVAL;
752 int lookup_flags;
753
754 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
755 goto out;
756
757 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
758 if (flag & AT_EMPTY_PATH)
759 lookup_flags |= LOOKUP_EMPTY;
760retry:
761 error = user_path_at(dfd, filename, lookup_flags, &path);
762 if (error)
763 goto out;
764 error = mnt_want_write(path.mnt);
765 if (error)
766 goto out_release;
767 error = chown_common(&path, user, group);
768 mnt_drop_write(path.mnt);
769out_release:
770 path_put(&path);
771 if (retry_estale(error, lookup_flags)) {
772 lookup_flags |= LOOKUP_REVAL;
773 goto retry;
774 }
775out:
776 return error;
777}
778
779SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
780 gid_t, group, int, flag)
781{
782 return do_fchownat(dfd, filename, user, group, flag);
783}
784
785SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
786{
787 return do_fchownat(AT_FDCWD, filename, user, group, 0);
788}
789
790SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
791{
792 return do_fchownat(AT_FDCWD, filename, user, group,
793 AT_SYMLINK_NOFOLLOW);
794}
795
796int vfs_fchown(struct file *file, uid_t user, gid_t group)
797{
798 int error;
799
800 error = mnt_want_write_file(file);
801 if (error)
802 return error;
803 audit_file(file);
804 error = chown_common(&file->f_path, user, group);
805 mnt_drop_write_file(file);
806 return error;
807}
808
809int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
810{
811 struct fd f = fdget(fd);
812 int error = -EBADF;
813
814 if (f.file) {
815 error = vfs_fchown(f.file, user, group);
816 fdput(f);
817 }
818 return error;
819}
820
821SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
822{
823 return ksys_fchown(fd, user, group);
824}
825
826static int do_dentry_open(struct file *f,
827 struct inode *inode,
828 int (*open)(struct inode *, struct file *))
829{
830 static const struct file_operations empty_fops = {};
831 int error;
832
833 path_get(&f->f_path);
834 f->f_inode = inode;
835 f->f_mapping = inode->i_mapping;
836 f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
837 f->f_sb_err = file_sample_sb_err(f);
838
839 if (unlikely(f->f_flags & O_PATH)) {
840 f->f_mode = FMODE_PATH | FMODE_OPENED;
841 f->f_op = &empty_fops;
842 return 0;
843 }
844
845 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) {
846 i_readcount_inc(inode);
847 } else if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
848 error = get_write_access(inode);
849 if (unlikely(error))
850 goto cleanup_file;
851 error = __mnt_want_write(f->f_path.mnt);
852 if (unlikely(error)) {
853 put_write_access(inode);
854 goto cleanup_file;
855 }
856 f->f_mode |= FMODE_WRITER;
857 }
858
859 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
860 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
861 f->f_mode |= FMODE_ATOMIC_POS;
862
863 f->f_op = fops_get(inode->i_fop);
864 if (WARN_ON(!f->f_op)) {
865 error = -ENODEV;
866 goto cleanup_all;
867 }
868
869 error = security_file_open(f);
870 if (error)
871 goto cleanup_all;
872
873 error = break_lease(locks_inode(f), f->f_flags);
874 if (error)
875 goto cleanup_all;
876
877 /* normally all 3 are set; ->open() can clear them if needed */
878 f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
879 if (!open)
880 open = f->f_op->open;
881 if (open) {
882 error = open(inode, f);
883 if (error)
884 goto cleanup_all;
885 }
886 f->f_mode |= FMODE_OPENED;
887 if ((f->f_mode & FMODE_READ) &&
888 likely(f->f_op->read || f->f_op->read_iter))
889 f->f_mode |= FMODE_CAN_READ;
890 if ((f->f_mode & FMODE_WRITE) &&
891 likely(f->f_op->write || f->f_op->write_iter))
892 f->f_mode |= FMODE_CAN_WRITE;
893 if ((f->f_mode & FMODE_LSEEK) && !f->f_op->llseek)
894 f->f_mode &= ~FMODE_LSEEK;
895 if (f->f_mapping->a_ops && f->f_mapping->a_ops->direct_IO)
896 f->f_mode |= FMODE_CAN_ODIRECT;
897
898 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
899 f->f_iocb_flags = iocb_flags(f);
900
901 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
902
903 if ((f->f_flags & O_DIRECT) && !(f->f_mode & FMODE_CAN_ODIRECT))
904 return -EINVAL;
905
906 /*
907 * XXX: Huge page cache doesn't support writing yet. Drop all page
908 * cache for this file before processing writes.
909 */
910 if (f->f_mode & FMODE_WRITE) {
911 /*
912 * Paired with smp_mb() in collapse_file() to ensure nr_thps
913 * is up to date and the update to i_writecount by
914 * get_write_access() is visible. Ensures subsequent insertion
915 * of THPs into the page cache will fail.
916 */
917 smp_mb();
918 if (filemap_nr_thps(inode->i_mapping)) {
919 struct address_space *mapping = inode->i_mapping;
920
921 filemap_invalidate_lock(inode->i_mapping);
922 /*
923 * unmap_mapping_range just need to be called once
924 * here, because the private pages is not need to be
925 * unmapped mapping (e.g. data segment of dynamic
926 * shared libraries here).
927 */
928 unmap_mapping_range(mapping, 0, 0, 0);
929 truncate_inode_pages(mapping, 0);
930 filemap_invalidate_unlock(inode->i_mapping);
931 }
932 }
933
934 return 0;
935
936cleanup_all:
937 if (WARN_ON_ONCE(error > 0))
938 error = -EINVAL;
939 fops_put(f->f_op);
940 put_file_access(f);
941cleanup_file:
942 path_put(&f->f_path);
943 f->f_path.mnt = NULL;
944 f->f_path.dentry = NULL;
945 f->f_inode = NULL;
946 return error;
947}
948
949/**
950 * finish_open - finish opening a file
951 * @file: file pointer
952 * @dentry: pointer to dentry
953 * @open: open callback
954 * @opened: state of open
955 *
956 * This can be used to finish opening a file passed to i_op->atomic_open().
957 *
958 * If the open callback is set to NULL, then the standard f_op->open()
959 * filesystem callback is substituted.
960 *
961 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
962 * the return value of d_splice_alias(), then the caller needs to perform dput()
963 * on it after finish_open().
964 *
965 * Returns zero on success or -errno if the open failed.
966 */
967int finish_open(struct file *file, struct dentry *dentry,
968 int (*open)(struct inode *, struct file *))
969{
970 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
971
972 file->f_path.dentry = dentry;
973 return do_dentry_open(file, d_backing_inode(dentry), open);
974}
975EXPORT_SYMBOL(finish_open);
976
977/**
978 * finish_no_open - finish ->atomic_open() without opening the file
979 *
980 * @file: file pointer
981 * @dentry: dentry or NULL (as returned from ->lookup())
982 *
983 * This can be used to set the result of a successful lookup in ->atomic_open().
984 *
985 * NB: unlike finish_open() this function does consume the dentry reference and
986 * the caller need not dput() it.
987 *
988 * Returns "0" which must be the return value of ->atomic_open() after having
989 * called this function.
990 */
991int finish_no_open(struct file *file, struct dentry *dentry)
992{
993 file->f_path.dentry = dentry;
994 return 0;
995}
996EXPORT_SYMBOL(finish_no_open);
997
998char *file_path(struct file *filp, char *buf, int buflen)
999{
1000 return d_path(&filp->f_path, buf, buflen);
1001}
1002EXPORT_SYMBOL(file_path);
1003
1004/**
1005 * vfs_open - open the file at the given path
1006 * @path: path to open
1007 * @file: newly allocated file with f_flag initialized
1008 * @cred: credentials to use
1009 */
1010int vfs_open(const struct path *path, struct file *file)
1011{
1012 file->f_path = *path;
1013 return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
1014}
1015
1016struct file *dentry_open(const struct path *path, int flags,
1017 const struct cred *cred)
1018{
1019 int error;
1020 struct file *f;
1021
1022 validate_creds(cred);
1023
1024 /* We must always pass in a valid mount pointer. */
1025 BUG_ON(!path->mnt);
1026
1027 f = alloc_empty_file(flags, cred);
1028 if (!IS_ERR(f)) {
1029 error = vfs_open(path, f);
1030 if (error) {
1031 fput(f);
1032 f = ERR_PTR(error);
1033 }
1034 }
1035 return f;
1036}
1037EXPORT_SYMBOL(dentry_open);
1038
1039/**
1040 * dentry_create - Create and open a file
1041 * @path: path to create
1042 * @flags: O_ flags
1043 * @mode: mode bits for new file
1044 * @cred: credentials to use
1045 *
1046 * Caller must hold the parent directory's lock, and have prepared
1047 * a negative dentry, placed in @path->dentry, for the new file.
1048 *
1049 * Caller sets @path->mnt to the vfsmount of the filesystem where
1050 * the new file is to be created. The parent directory and the
1051 * negative dentry must reside on the same filesystem instance.
1052 *
1053 * On success, returns a "struct file *". Otherwise a ERR_PTR
1054 * is returned.
1055 */
1056struct file *dentry_create(const struct path *path, int flags, umode_t mode,
1057 const struct cred *cred)
1058{
1059 struct file *f;
1060 int error;
1061
1062 validate_creds(cred);
1063 f = alloc_empty_file(flags, cred);
1064 if (IS_ERR(f))
1065 return f;
1066
1067 error = vfs_create(mnt_user_ns(path->mnt),
1068 d_inode(path->dentry->d_parent),
1069 path->dentry, mode, true);
1070 if (!error)
1071 error = vfs_open(path, f);
1072
1073 if (unlikely(error)) {
1074 fput(f);
1075 return ERR_PTR(error);
1076 }
1077 return f;
1078}
1079EXPORT_SYMBOL(dentry_create);
1080
1081struct file *open_with_fake_path(const struct path *path, int flags,
1082 struct inode *inode, const struct cred *cred)
1083{
1084 struct file *f = alloc_empty_file_noaccount(flags, cred);
1085 if (!IS_ERR(f)) {
1086 int error;
1087
1088 f->f_path = *path;
1089 error = do_dentry_open(f, inode, NULL);
1090 if (error) {
1091 fput(f);
1092 f = ERR_PTR(error);
1093 }
1094 }
1095 return f;
1096}
1097EXPORT_SYMBOL(open_with_fake_path);
1098
1099#define WILL_CREATE(flags) (flags & (O_CREAT | __O_TMPFILE))
1100#define O_PATH_FLAGS (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1101
1102inline struct open_how build_open_how(int flags, umode_t mode)
1103{
1104 struct open_how how = {
1105 .flags = flags & VALID_OPEN_FLAGS,
1106 .mode = mode & S_IALLUGO,
1107 };
1108
1109 /* O_PATH beats everything else. */
1110 if (how.flags & O_PATH)
1111 how.flags &= O_PATH_FLAGS;
1112 /* Modes should only be set for create-like flags. */
1113 if (!WILL_CREATE(how.flags))
1114 how.mode = 0;
1115 return how;
1116}
1117
1118inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1119{
1120 u64 flags = how->flags;
1121 u64 strip = FMODE_NONOTIFY | O_CLOEXEC;
1122 int lookup_flags = 0;
1123 int acc_mode = ACC_MODE(flags);
1124
1125 BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1126 "struct open_flags doesn't yet handle flags > 32 bits");
1127
1128 /*
1129 * Strip flags that either shouldn't be set by userspace like
1130 * FMODE_NONOTIFY or that aren't relevant in determining struct
1131 * open_flags like O_CLOEXEC.
1132 */
1133 flags &= ~strip;
1134
1135 /*
1136 * Older syscalls implicitly clear all of the invalid flags or argument
1137 * values before calling build_open_flags(), but openat2(2) checks all
1138 * of its arguments.
1139 */
1140 if (flags & ~VALID_OPEN_FLAGS)
1141 return -EINVAL;
1142 if (how->resolve & ~VALID_RESOLVE_FLAGS)
1143 return -EINVAL;
1144
1145 /* Scoping flags are mutually exclusive. */
1146 if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1147 return -EINVAL;
1148
1149 /* Deal with the mode. */
1150 if (WILL_CREATE(flags)) {
1151 if (how->mode & ~S_IALLUGO)
1152 return -EINVAL;
1153 op->mode = how->mode | S_IFREG;
1154 } else {
1155 if (how->mode != 0)
1156 return -EINVAL;
1157 op->mode = 0;
1158 }
1159
1160 /*
1161 * In order to ensure programs get explicit errors when trying to use
1162 * O_TMPFILE on old kernels, O_TMPFILE is implemented such that it
1163 * looks like (O_DIRECTORY|O_RDWR & ~O_CREAT) to old kernels. But we
1164 * have to require userspace to explicitly set it.
1165 */
1166 if (flags & __O_TMPFILE) {
1167 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
1168 return -EINVAL;
1169 if (!(acc_mode & MAY_WRITE))
1170 return -EINVAL;
1171 }
1172 if (flags & O_PATH) {
1173 /* O_PATH only permits certain other flags to be set. */
1174 if (flags & ~O_PATH_FLAGS)
1175 return -EINVAL;
1176 acc_mode = 0;
1177 }
1178
1179 /*
1180 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1181 * check for O_DSYNC if the need any syncing at all we enforce it's
1182 * always set instead of having to deal with possibly weird behaviour
1183 * for malicious applications setting only __O_SYNC.
1184 */
1185 if (flags & __O_SYNC)
1186 flags |= O_DSYNC;
1187
1188 op->open_flag = flags;
1189
1190 /* O_TRUNC implies we need access checks for write permissions */
1191 if (flags & O_TRUNC)
1192 acc_mode |= MAY_WRITE;
1193
1194 /* Allow the LSM permission hook to distinguish append
1195 access from general write access. */
1196 if (flags & O_APPEND)
1197 acc_mode |= MAY_APPEND;
1198
1199 op->acc_mode = acc_mode;
1200
1201 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1202
1203 if (flags & O_CREAT) {
1204 op->intent |= LOOKUP_CREATE;
1205 if (flags & O_EXCL) {
1206 op->intent |= LOOKUP_EXCL;
1207 flags |= O_NOFOLLOW;
1208 }
1209 }
1210
1211 if (flags & O_DIRECTORY)
1212 lookup_flags |= LOOKUP_DIRECTORY;
1213 if (!(flags & O_NOFOLLOW))
1214 lookup_flags |= LOOKUP_FOLLOW;
1215
1216 if (how->resolve & RESOLVE_NO_XDEV)
1217 lookup_flags |= LOOKUP_NO_XDEV;
1218 if (how->resolve & RESOLVE_NO_MAGICLINKS)
1219 lookup_flags |= LOOKUP_NO_MAGICLINKS;
1220 if (how->resolve & RESOLVE_NO_SYMLINKS)
1221 lookup_flags |= LOOKUP_NO_SYMLINKS;
1222 if (how->resolve & RESOLVE_BENEATH)
1223 lookup_flags |= LOOKUP_BENEATH;
1224 if (how->resolve & RESOLVE_IN_ROOT)
1225 lookup_flags |= LOOKUP_IN_ROOT;
1226 if (how->resolve & RESOLVE_CACHED) {
1227 /* Don't bother even trying for create/truncate/tmpfile open */
1228 if (flags & (O_TRUNC | O_CREAT | O_TMPFILE))
1229 return -EAGAIN;
1230 lookup_flags |= LOOKUP_CACHED;
1231 }
1232
1233 op->lookup_flags = lookup_flags;
1234 return 0;
1235}
1236
1237/**
1238 * file_open_name - open file and return file pointer
1239 *
1240 * @name: struct filename containing path to open
1241 * @flags: open flags as per the open(2) second argument
1242 * @mode: mode for the new file if O_CREAT is set, else ignored
1243 *
1244 * This is the helper to open a file from kernelspace if you really
1245 * have to. But in generally you should not do this, so please move
1246 * along, nothing to see here..
1247 */
1248struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1249{
1250 struct open_flags op;
1251 struct open_how how = build_open_how(flags, mode);
1252 int err = build_open_flags(&how, &op);
1253 if (err)
1254 return ERR_PTR(err);
1255 return do_filp_open(AT_FDCWD, name, &op);
1256}
1257
1258/**
1259 * filp_open - open file and return file pointer
1260 *
1261 * @filename: path to open
1262 * @flags: open flags as per the open(2) second argument
1263 * @mode: mode for the new file if O_CREAT is set, else ignored
1264 *
1265 * This is the helper to open a file from kernelspace if you really
1266 * have to. But in generally you should not do this, so please move
1267 * along, nothing to see here..
1268 */
1269struct file *filp_open(const char *filename, int flags, umode_t mode)
1270{
1271 struct filename *name = getname_kernel(filename);
1272 struct file *file = ERR_CAST(name);
1273
1274 if (!IS_ERR(name)) {
1275 file = file_open_name(name, flags, mode);
1276 putname(name);
1277 }
1278 return file;
1279}
1280EXPORT_SYMBOL(filp_open);
1281
1282struct file *file_open_root(const struct path *root,
1283 const char *filename, int flags, umode_t mode)
1284{
1285 struct open_flags op;
1286 struct open_how how = build_open_how(flags, mode);
1287 int err = build_open_flags(&how, &op);
1288 if (err)
1289 return ERR_PTR(err);
1290 return do_file_open_root(root, filename, &op);
1291}
1292EXPORT_SYMBOL(file_open_root);
1293
1294static long do_sys_openat2(int dfd, const char __user *filename,
1295 struct open_how *how)
1296{
1297 struct open_flags op;
1298 int fd = build_open_flags(how, &op);
1299 struct filename *tmp;
1300
1301 if (fd)
1302 return fd;
1303
1304 tmp = getname(filename);
1305 if (IS_ERR(tmp))
1306 return PTR_ERR(tmp);
1307
1308 fd = get_unused_fd_flags(how->flags);
1309 if (fd >= 0) {
1310 struct file *f = do_filp_open(dfd, tmp, &op);
1311 if (IS_ERR(f)) {
1312 put_unused_fd(fd);
1313 fd = PTR_ERR(f);
1314 } else {
1315 fsnotify_open(f);
1316 fd_install(fd, f);
1317 }
1318 }
1319 putname(tmp);
1320 return fd;
1321}
1322
1323long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1324{
1325 struct open_how how = build_open_how(flags, mode);
1326 return do_sys_openat2(dfd, filename, &how);
1327}
1328
1329
1330SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1331{
1332 if (force_o_largefile())
1333 flags |= O_LARGEFILE;
1334 return do_sys_open(AT_FDCWD, filename, flags, mode);
1335}
1336
1337SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1338 umode_t, mode)
1339{
1340 if (force_o_largefile())
1341 flags |= O_LARGEFILE;
1342 return do_sys_open(dfd, filename, flags, mode);
1343}
1344
1345SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1346 struct open_how __user *, how, size_t, usize)
1347{
1348 int err;
1349 struct open_how tmp;
1350
1351 BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1352 BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1353
1354 if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1355 return -EINVAL;
1356
1357 err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1358 if (err)
1359 return err;
1360
1361 audit_openat2_how(&tmp);
1362
1363 /* O_LARGEFILE is only allowed for non-O_PATH. */
1364 if (!(tmp.flags & O_PATH) && force_o_largefile())
1365 tmp.flags |= O_LARGEFILE;
1366
1367 return do_sys_openat2(dfd, filename, &tmp);
1368}
1369
1370#ifdef CONFIG_COMPAT
1371/*
1372 * Exactly like sys_open(), except that it doesn't set the
1373 * O_LARGEFILE flag.
1374 */
1375COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1376{
1377 return do_sys_open(AT_FDCWD, filename, flags, mode);
1378}
1379
1380/*
1381 * Exactly like sys_openat(), except that it doesn't set the
1382 * O_LARGEFILE flag.
1383 */
1384COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1385{
1386 return do_sys_open(dfd, filename, flags, mode);
1387}
1388#endif
1389
1390#ifndef __alpha__
1391
1392/*
1393 * For backward compatibility? Maybe this should be moved
1394 * into arch/i386 instead?
1395 */
1396SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1397{
1398 int flags = O_CREAT | O_WRONLY | O_TRUNC;
1399
1400 if (force_o_largefile())
1401 flags |= O_LARGEFILE;
1402 return do_sys_open(AT_FDCWD, pathname, flags, mode);
1403}
1404#endif
1405
1406/*
1407 * "id" is the POSIX thread ID. We use the
1408 * files pointer for this..
1409 */
1410int filp_close(struct file *filp, fl_owner_t id)
1411{
1412 int retval = 0;
1413
1414 if (!file_count(filp)) {
1415 printk(KERN_ERR "VFS: Close: file count is 0\n");
1416 return 0;
1417 }
1418
1419 if (filp->f_op->flush)
1420 retval = filp->f_op->flush(filp, id);
1421
1422 if (likely(!(filp->f_mode & FMODE_PATH))) {
1423 dnotify_flush(filp, id);
1424 locks_remove_posix(filp, id);
1425 }
1426 fput(filp);
1427 return retval;
1428}
1429
1430EXPORT_SYMBOL(filp_close);
1431
1432/*
1433 * Careful here! We test whether the file pointer is NULL before
1434 * releasing the fd. This ensures that one clone task can't release
1435 * an fd while another clone is opening it.
1436 */
1437SYSCALL_DEFINE1(close, unsigned int, fd)
1438{
1439 int retval = close_fd(fd);
1440
1441 /* can't restart close syscall because file table entry was cleared */
1442 if (unlikely(retval == -ERESTARTSYS ||
1443 retval == -ERESTARTNOINTR ||
1444 retval == -ERESTARTNOHAND ||
1445 retval == -ERESTART_RESTARTBLOCK))
1446 retval = -EINTR;
1447
1448 return retval;
1449}
1450
1451/**
1452 * close_range() - Close all file descriptors in a given range.
1453 *
1454 * @fd: starting file descriptor to close
1455 * @max_fd: last file descriptor to close
1456 * @flags: reserved for future extensions
1457 *
1458 * This closes a range of file descriptors. All file descriptors
1459 * from @fd up to and including @max_fd are closed.
1460 * Currently, errors to close a given file descriptor are ignored.
1461 */
1462SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd,
1463 unsigned int, flags)
1464{
1465 return __close_range(fd, max_fd, flags);
1466}
1467
1468/*
1469 * This routine simulates a hangup on the tty, to arrange that users
1470 * are given clean terminals at login time.
1471 */
1472SYSCALL_DEFINE0(vhangup)
1473{
1474 if (capable(CAP_SYS_TTY_CONFIG)) {
1475 tty_vhangup_self();
1476 return 0;
1477 }
1478 return -EPERM;
1479}
1480
1481/*
1482 * Called when an inode is about to be open.
1483 * We use this to disallow opening large files on 32bit systems if
1484 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1485 * on this flag in sys_open.
1486 */
1487int generic_file_open(struct inode * inode, struct file * filp)
1488{
1489 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1490 return -EOVERFLOW;
1491 return 0;
1492}
1493
1494EXPORT_SYMBOL(generic_file_open);
1495
1496/*
1497 * This is used by subsystems that don't want seekable
1498 * file descriptors. The function is not supposed to ever fail, the only
1499 * reason it returns an 'int' and not 'void' is so that it can be plugged
1500 * directly into file_operations structure.
1501 */
1502int nonseekable_open(struct inode *inode, struct file *filp)
1503{
1504 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1505 return 0;
1506}
1507
1508EXPORT_SYMBOL(nonseekable_open);
1509
1510/*
1511 * stream_open is used by subsystems that want stream-like file descriptors.
1512 * Such file descriptors are not seekable and don't have notion of position
1513 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1514 * Contrary to file descriptors of other regular files, .read() and .write()
1515 * can run simultaneously.
1516 *
1517 * stream_open never fails and is marked to return int so that it could be
1518 * directly used as file_operations.open .
1519 */
1520int stream_open(struct inode *inode, struct file *filp)
1521{
1522 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1523 filp->f_mode |= FMODE_STREAM;
1524 return 0;
1525}
1526
1527EXPORT_SYMBOL(stream_open);
1/*
2 * linux/fs/open.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7#include <linux/string.h>
8#include <linux/mm.h>
9#include <linux/file.h>
10#include <linux/fdtable.h>
11#include <linux/fsnotify.h>
12#include <linux/module.h>
13#include <linux/tty.h>
14#include <linux/namei.h>
15#include <linux/backing-dev.h>
16#include <linux/capability.h>
17#include <linux/securebits.h>
18#include <linux/security.h>
19#include <linux/mount.h>
20#include <linux/fcntl.h>
21#include <linux/slab.h>
22#include <asm/uaccess.h>
23#include <linux/fs.h>
24#include <linux/personality.h>
25#include <linux/pagemap.h>
26#include <linux/syscalls.h>
27#include <linux/rcupdate.h>
28#include <linux/audit.h>
29#include <linux/falloc.h>
30#include <linux/fs_struct.h>
31#include <linux/ima.h>
32#include <linux/dnotify.h>
33
34#include "internal.h"
35
36int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
37 struct file *filp)
38{
39 int ret;
40 struct iattr newattrs;
41
42 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
43 if (length < 0)
44 return -EINVAL;
45
46 newattrs.ia_size = length;
47 newattrs.ia_valid = ATTR_SIZE | time_attrs;
48 if (filp) {
49 newattrs.ia_file = filp;
50 newattrs.ia_valid |= ATTR_FILE;
51 }
52
53 /* Remove suid/sgid on truncate too */
54 ret = should_remove_suid(dentry);
55 if (ret)
56 newattrs.ia_valid |= ret | ATTR_FORCE;
57
58 mutex_lock(&dentry->d_inode->i_mutex);
59 ret = notify_change(dentry, &newattrs);
60 mutex_unlock(&dentry->d_inode->i_mutex);
61 return ret;
62}
63
64static long do_sys_truncate(const char __user *pathname, loff_t length)
65{
66 struct path path;
67 struct inode *inode;
68 int error;
69
70 error = -EINVAL;
71 if (length < 0) /* sorry, but loff_t says... */
72 goto out;
73
74 error = user_path(pathname, &path);
75 if (error)
76 goto out;
77 inode = path.dentry->d_inode;
78
79 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
80 error = -EISDIR;
81 if (S_ISDIR(inode->i_mode))
82 goto dput_and_out;
83
84 error = -EINVAL;
85 if (!S_ISREG(inode->i_mode))
86 goto dput_and_out;
87
88 error = mnt_want_write(path.mnt);
89 if (error)
90 goto dput_and_out;
91
92 error = inode_permission(inode, MAY_WRITE);
93 if (error)
94 goto mnt_drop_write_and_out;
95
96 error = -EPERM;
97 if (IS_APPEND(inode))
98 goto mnt_drop_write_and_out;
99
100 error = get_write_access(inode);
101 if (error)
102 goto mnt_drop_write_and_out;
103
104 /*
105 * Make sure that there are no leases. get_write_access() protects
106 * against the truncate racing with a lease-granting setlease().
107 */
108 error = break_lease(inode, O_WRONLY);
109 if (error)
110 goto put_write_and_out;
111
112 error = locks_verify_truncate(inode, NULL, length);
113 if (!error)
114 error = security_path_truncate(&path);
115 if (!error)
116 error = do_truncate(path.dentry, length, 0, NULL);
117
118put_write_and_out:
119 put_write_access(inode);
120mnt_drop_write_and_out:
121 mnt_drop_write(path.mnt);
122dput_and_out:
123 path_put(&path);
124out:
125 return error;
126}
127
128SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
129{
130 return do_sys_truncate(path, length);
131}
132
133static long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
134{
135 struct inode * inode;
136 struct dentry *dentry;
137 struct file * file;
138 int error;
139
140 error = -EINVAL;
141 if (length < 0)
142 goto out;
143 error = -EBADF;
144 file = fget(fd);
145 if (!file)
146 goto out;
147
148 /* explicitly opened as large or we are on 64-bit box */
149 if (file->f_flags & O_LARGEFILE)
150 small = 0;
151
152 dentry = file->f_path.dentry;
153 inode = dentry->d_inode;
154 error = -EINVAL;
155 if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
156 goto out_putf;
157
158 error = -EINVAL;
159 /* Cannot ftruncate over 2^31 bytes without large file support */
160 if (small && length > MAX_NON_LFS)
161 goto out_putf;
162
163 error = -EPERM;
164 if (IS_APPEND(inode))
165 goto out_putf;
166
167 error = locks_verify_truncate(inode, file, length);
168 if (!error)
169 error = security_path_truncate(&file->f_path);
170 if (!error)
171 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, file);
172out_putf:
173 fput(file);
174out:
175 return error;
176}
177
178SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
179{
180 long ret = do_sys_ftruncate(fd, length, 1);
181 /* avoid REGPARM breakage on x86: */
182 asmlinkage_protect(2, ret, fd, length);
183 return ret;
184}
185
186/* LFS versions of truncate are only needed on 32 bit machines */
187#if BITS_PER_LONG == 32
188SYSCALL_DEFINE(truncate64)(const char __user * path, loff_t length)
189{
190 return do_sys_truncate(path, length);
191}
192#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
193asmlinkage long SyS_truncate64(long path, loff_t length)
194{
195 return SYSC_truncate64((const char __user *) path, length);
196}
197SYSCALL_ALIAS(sys_truncate64, SyS_truncate64);
198#endif
199
200SYSCALL_DEFINE(ftruncate64)(unsigned int fd, loff_t length)
201{
202 long ret = do_sys_ftruncate(fd, length, 0);
203 /* avoid REGPARM breakage on x86: */
204 asmlinkage_protect(2, ret, fd, length);
205 return ret;
206}
207#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
208asmlinkage long SyS_ftruncate64(long fd, loff_t length)
209{
210 return SYSC_ftruncate64((unsigned int) fd, length);
211}
212SYSCALL_ALIAS(sys_ftruncate64, SyS_ftruncate64);
213#endif
214#endif /* BITS_PER_LONG == 32 */
215
216
217int do_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
218{
219 struct inode *inode = file->f_path.dentry->d_inode;
220 long ret;
221
222 if (offset < 0 || len <= 0)
223 return -EINVAL;
224
225 /* Return error if mode is not supported */
226 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
227 return -EOPNOTSUPP;
228
229 /* Punch hole must have keep size set */
230 if ((mode & FALLOC_FL_PUNCH_HOLE) &&
231 !(mode & FALLOC_FL_KEEP_SIZE))
232 return -EOPNOTSUPP;
233
234 if (!(file->f_mode & FMODE_WRITE))
235 return -EBADF;
236
237 /* It's not possible punch hole on append only file */
238 if (mode & FALLOC_FL_PUNCH_HOLE && IS_APPEND(inode))
239 return -EPERM;
240
241 if (IS_IMMUTABLE(inode))
242 return -EPERM;
243
244 /*
245 * Revalidate the write permissions, in case security policy has
246 * changed since the files were opened.
247 */
248 ret = security_file_permission(file, MAY_WRITE);
249 if (ret)
250 return ret;
251
252 if (S_ISFIFO(inode->i_mode))
253 return -ESPIPE;
254
255 /*
256 * Let individual file system decide if it supports preallocation
257 * for directories or not.
258 */
259 if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
260 return -ENODEV;
261
262 /* Check for wrap through zero too */
263 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
264 return -EFBIG;
265
266 if (!file->f_op->fallocate)
267 return -EOPNOTSUPP;
268
269 return file->f_op->fallocate(file, mode, offset, len);
270}
271
272SYSCALL_DEFINE(fallocate)(int fd, int mode, loff_t offset, loff_t len)
273{
274 struct file *file;
275 int error = -EBADF;
276
277 file = fget(fd);
278 if (file) {
279 error = do_fallocate(file, mode, offset, len);
280 fput(file);
281 }
282
283 return error;
284}
285
286#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
287asmlinkage long SyS_fallocate(long fd, long mode, loff_t offset, loff_t len)
288{
289 return SYSC_fallocate((int)fd, (int)mode, offset, len);
290}
291SYSCALL_ALIAS(sys_fallocate, SyS_fallocate);
292#endif
293
294/*
295 * access() needs to use the real uid/gid, not the effective uid/gid.
296 * We do this by temporarily clearing all FS-related capabilities and
297 * switching the fsuid/fsgid around to the real ones.
298 */
299SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
300{
301 const struct cred *old_cred;
302 struct cred *override_cred;
303 struct path path;
304 struct inode *inode;
305 int res;
306
307 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
308 return -EINVAL;
309
310 override_cred = prepare_creds();
311 if (!override_cred)
312 return -ENOMEM;
313
314 override_cred->fsuid = override_cred->uid;
315 override_cred->fsgid = override_cred->gid;
316
317 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
318 /* Clear the capabilities if we switch to a non-root user */
319 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
320 if (!uid_eq(override_cred->uid, root_uid))
321 cap_clear(override_cred->cap_effective);
322 else
323 override_cred->cap_effective =
324 override_cred->cap_permitted;
325 }
326
327 old_cred = override_creds(override_cred);
328
329 res = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
330 if (res)
331 goto out;
332
333 inode = path.dentry->d_inode;
334
335 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
336 /*
337 * MAY_EXEC on regular files is denied if the fs is mounted
338 * with the "noexec" flag.
339 */
340 res = -EACCES;
341 if (path.mnt->mnt_flags & MNT_NOEXEC)
342 goto out_path_release;
343 }
344
345 res = inode_permission(inode, mode | MAY_ACCESS);
346 /* SuS v2 requires we report a read only fs too */
347 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
348 goto out_path_release;
349 /*
350 * This is a rare case where using __mnt_is_readonly()
351 * is OK without a mnt_want/drop_write() pair. Since
352 * no actual write to the fs is performed here, we do
353 * not need to telegraph to that to anyone.
354 *
355 * By doing this, we accept that this access is
356 * inherently racy and know that the fs may change
357 * state before we even see this result.
358 */
359 if (__mnt_is_readonly(path.mnt))
360 res = -EROFS;
361
362out_path_release:
363 path_put(&path);
364out:
365 revert_creds(old_cred);
366 put_cred(override_cred);
367 return res;
368}
369
370SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
371{
372 return sys_faccessat(AT_FDCWD, filename, mode);
373}
374
375SYSCALL_DEFINE1(chdir, const char __user *, filename)
376{
377 struct path path;
378 int error;
379
380 error = user_path_dir(filename, &path);
381 if (error)
382 goto out;
383
384 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
385 if (error)
386 goto dput_and_out;
387
388 set_fs_pwd(current->fs, &path);
389
390dput_and_out:
391 path_put(&path);
392out:
393 return error;
394}
395
396SYSCALL_DEFINE1(fchdir, unsigned int, fd)
397{
398 struct file *file;
399 struct inode *inode;
400 int error, fput_needed;
401
402 error = -EBADF;
403 file = fget_raw_light(fd, &fput_needed);
404 if (!file)
405 goto out;
406
407 inode = file->f_path.dentry->d_inode;
408
409 error = -ENOTDIR;
410 if (!S_ISDIR(inode->i_mode))
411 goto out_putf;
412
413 error = inode_permission(inode, MAY_EXEC | MAY_CHDIR);
414 if (!error)
415 set_fs_pwd(current->fs, &file->f_path);
416out_putf:
417 fput_light(file, fput_needed);
418out:
419 return error;
420}
421
422SYSCALL_DEFINE1(chroot, const char __user *, filename)
423{
424 struct path path;
425 int error;
426
427 error = user_path_dir(filename, &path);
428 if (error)
429 goto out;
430
431 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
432 if (error)
433 goto dput_and_out;
434
435 error = -EPERM;
436 if (!capable(CAP_SYS_CHROOT))
437 goto dput_and_out;
438 error = security_path_chroot(&path);
439 if (error)
440 goto dput_and_out;
441
442 set_fs_root(current->fs, &path);
443 error = 0;
444dput_and_out:
445 path_put(&path);
446out:
447 return error;
448}
449
450static int chmod_common(struct path *path, umode_t mode)
451{
452 struct inode *inode = path->dentry->d_inode;
453 struct iattr newattrs;
454 int error;
455
456 error = mnt_want_write(path->mnt);
457 if (error)
458 return error;
459 mutex_lock(&inode->i_mutex);
460 error = security_path_chmod(path, mode);
461 if (error)
462 goto out_unlock;
463 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
464 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
465 error = notify_change(path->dentry, &newattrs);
466out_unlock:
467 mutex_unlock(&inode->i_mutex);
468 mnt_drop_write(path->mnt);
469 return error;
470}
471
472SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
473{
474 struct file * file;
475 int err = -EBADF;
476
477 file = fget(fd);
478 if (file) {
479 audit_inode(NULL, file->f_path.dentry);
480 err = chmod_common(&file->f_path, mode);
481 fput(file);
482 }
483 return err;
484}
485
486SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename, umode_t, mode)
487{
488 struct path path;
489 int error;
490
491 error = user_path_at(dfd, filename, LOOKUP_FOLLOW, &path);
492 if (!error) {
493 error = chmod_common(&path, mode);
494 path_put(&path);
495 }
496 return error;
497}
498
499SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
500{
501 return sys_fchmodat(AT_FDCWD, filename, mode);
502}
503
504static int chown_common(struct path *path, uid_t user, gid_t group)
505{
506 struct inode *inode = path->dentry->d_inode;
507 int error;
508 struct iattr newattrs;
509 kuid_t uid;
510 kgid_t gid;
511
512 uid = make_kuid(current_user_ns(), user);
513 gid = make_kgid(current_user_ns(), group);
514
515 newattrs.ia_valid = ATTR_CTIME;
516 if (user != (uid_t) -1) {
517 if (!uid_valid(uid))
518 return -EINVAL;
519 newattrs.ia_valid |= ATTR_UID;
520 newattrs.ia_uid = uid;
521 }
522 if (group != (gid_t) -1) {
523 if (!gid_valid(gid))
524 return -EINVAL;
525 newattrs.ia_valid |= ATTR_GID;
526 newattrs.ia_gid = gid;
527 }
528 if (!S_ISDIR(inode->i_mode))
529 newattrs.ia_valid |=
530 ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
531 mutex_lock(&inode->i_mutex);
532 error = security_path_chown(path, user, group);
533 if (!error)
534 error = notify_change(path->dentry, &newattrs);
535 mutex_unlock(&inode->i_mutex);
536
537 return error;
538}
539
540SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
541{
542 struct path path;
543 int error;
544
545 error = user_path(filename, &path);
546 if (error)
547 goto out;
548 error = mnt_want_write(path.mnt);
549 if (error)
550 goto out_release;
551 error = chown_common(&path, user, group);
552 mnt_drop_write(path.mnt);
553out_release:
554 path_put(&path);
555out:
556 return error;
557}
558
559SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
560 gid_t, group, int, flag)
561{
562 struct path path;
563 int error = -EINVAL;
564 int lookup_flags;
565
566 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
567 goto out;
568
569 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
570 if (flag & AT_EMPTY_PATH)
571 lookup_flags |= LOOKUP_EMPTY;
572 error = user_path_at(dfd, filename, lookup_flags, &path);
573 if (error)
574 goto out;
575 error = mnt_want_write(path.mnt);
576 if (error)
577 goto out_release;
578 error = chown_common(&path, user, group);
579 mnt_drop_write(path.mnt);
580out_release:
581 path_put(&path);
582out:
583 return error;
584}
585
586SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
587{
588 struct path path;
589 int error;
590
591 error = user_lpath(filename, &path);
592 if (error)
593 goto out;
594 error = mnt_want_write(path.mnt);
595 if (error)
596 goto out_release;
597 error = chown_common(&path, user, group);
598 mnt_drop_write(path.mnt);
599out_release:
600 path_put(&path);
601out:
602 return error;
603}
604
605SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
606{
607 struct file * file;
608 int error = -EBADF;
609 struct dentry * dentry;
610
611 file = fget(fd);
612 if (!file)
613 goto out;
614
615 error = mnt_want_write_file(file);
616 if (error)
617 goto out_fput;
618 dentry = file->f_path.dentry;
619 audit_inode(NULL, dentry);
620 error = chown_common(&file->f_path, user, group);
621 mnt_drop_write_file(file);
622out_fput:
623 fput(file);
624out:
625 return error;
626}
627
628/*
629 * You have to be very careful that these write
630 * counts get cleaned up in error cases and
631 * upon __fput(). This should probably never
632 * be called outside of __dentry_open().
633 */
634static inline int __get_file_write_access(struct inode *inode,
635 struct vfsmount *mnt)
636{
637 int error;
638 error = get_write_access(inode);
639 if (error)
640 return error;
641 /*
642 * Do not take mount writer counts on
643 * special files since no writes to
644 * the mount itself will occur.
645 */
646 if (!special_file(inode->i_mode)) {
647 /*
648 * Balanced in __fput()
649 */
650 error = mnt_want_write(mnt);
651 if (error)
652 put_write_access(inode);
653 }
654 return error;
655}
656
657int open_check_o_direct(struct file *f)
658{
659 /* NB: we're sure to have correct a_ops only after f_op->open */
660 if (f->f_flags & O_DIRECT) {
661 if (!f->f_mapping->a_ops ||
662 ((!f->f_mapping->a_ops->direct_IO) &&
663 (!f->f_mapping->a_ops->get_xip_mem))) {
664 return -EINVAL;
665 }
666 }
667 return 0;
668}
669
670static struct file *do_dentry_open(struct dentry *dentry, struct vfsmount *mnt,
671 struct file *f,
672 int (*open)(struct inode *, struct file *),
673 const struct cred *cred)
674{
675 static const struct file_operations empty_fops = {};
676 struct inode *inode;
677 int error;
678
679 f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
680 FMODE_PREAD | FMODE_PWRITE;
681
682 if (unlikely(f->f_flags & O_PATH))
683 f->f_mode = FMODE_PATH;
684
685 inode = dentry->d_inode;
686 if (f->f_mode & FMODE_WRITE) {
687 error = __get_file_write_access(inode, mnt);
688 if (error)
689 goto cleanup_file;
690 if (!special_file(inode->i_mode))
691 file_take_write(f);
692 }
693
694 f->f_mapping = inode->i_mapping;
695 f->f_path.dentry = dentry;
696 f->f_path.mnt = mnt;
697 f->f_pos = 0;
698 file_sb_list_add(f, inode->i_sb);
699
700 if (unlikely(f->f_mode & FMODE_PATH)) {
701 f->f_op = &empty_fops;
702 return f;
703 }
704
705 f->f_op = fops_get(inode->i_fop);
706
707 error = security_file_open(f, cred);
708 if (error)
709 goto cleanup_all;
710
711 error = break_lease(inode, f->f_flags);
712 if (error)
713 goto cleanup_all;
714
715 if (!open && f->f_op)
716 open = f->f_op->open;
717 if (open) {
718 error = open(inode, f);
719 if (error)
720 goto cleanup_all;
721 }
722 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
723 i_readcount_inc(inode);
724
725 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
726
727 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
728
729 return f;
730
731cleanup_all:
732 fops_put(f->f_op);
733 if (f->f_mode & FMODE_WRITE) {
734 put_write_access(inode);
735 if (!special_file(inode->i_mode)) {
736 /*
737 * We don't consider this a real
738 * mnt_want/drop_write() pair
739 * because it all happenend right
740 * here, so just reset the state.
741 */
742 file_reset_write(f);
743 mnt_drop_write(mnt);
744 }
745 }
746 file_sb_list_del(f);
747 f->f_path.dentry = NULL;
748 f->f_path.mnt = NULL;
749cleanup_file:
750 dput(dentry);
751 mntput(mnt);
752 return ERR_PTR(error);
753}
754
755static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
756 struct file *f,
757 int (*open)(struct inode *, struct file *),
758 const struct cred *cred)
759{
760 struct file *res = do_dentry_open(dentry, mnt, f, open, cred);
761 if (!IS_ERR(res)) {
762 int error = open_check_o_direct(f);
763 if (error) {
764 fput(res);
765 res = ERR_PTR(error);
766 }
767 } else {
768 put_filp(f);
769 }
770 return res;
771}
772
773/**
774 * lookup_instantiate_filp - instantiates the open intent filp
775 * @nd: pointer to nameidata
776 * @dentry: pointer to dentry
777 * @open: open callback
778 *
779 * Helper for filesystems that want to use lookup open intents and pass back
780 * a fully instantiated struct file to the caller.
781 * This function is meant to be called from within a filesystem's
782 * lookup method.
783 * Beware of calling it for non-regular files! Those ->open methods might block
784 * (e.g. in fifo_open), leaving you with parent locked (and in case of fifo,
785 * leading to a deadlock, as nobody can open that fifo anymore, because
786 * another process to open fifo will block on locked parent when doing lookup).
787 * Note that in case of error, nd->intent.open.file is destroyed, but the
788 * path information remains valid.
789 * If the open callback is set to NULL, then the standard f_op->open()
790 * filesystem callback is substituted.
791 */
792struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
793 int (*open)(struct inode *, struct file *))
794{
795 const struct cred *cred = current_cred();
796
797 if (IS_ERR(nd->intent.open.file))
798 goto out;
799 if (IS_ERR(dentry))
800 goto out_err;
801 nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
802 nd->intent.open.file,
803 open, cred);
804out:
805 return nd->intent.open.file;
806out_err:
807 release_open_intent(nd);
808 nd->intent.open.file = ERR_CAST(dentry);
809 goto out;
810}
811EXPORT_SYMBOL_GPL(lookup_instantiate_filp);
812
813/**
814 * nameidata_to_filp - convert a nameidata to an open filp.
815 * @nd: pointer to nameidata
816 * @flags: open flags
817 *
818 * Note that this function destroys the original nameidata
819 */
820struct file *nameidata_to_filp(struct nameidata *nd)
821{
822 const struct cred *cred = current_cred();
823 struct file *filp;
824
825 /* Pick up the filp from the open intent */
826 filp = nd->intent.open.file;
827
828 /* Has the filesystem initialised the file for us? */
829 if (filp->f_path.dentry != NULL) {
830 nd->intent.open.file = NULL;
831 } else {
832 struct file *res;
833
834 path_get(&nd->path);
835 res = do_dentry_open(nd->path.dentry, nd->path.mnt,
836 filp, NULL, cred);
837 if (!IS_ERR(res)) {
838 int error;
839
840 nd->intent.open.file = NULL;
841 BUG_ON(res != filp);
842
843 error = open_check_o_direct(filp);
844 if (error) {
845 fput(filp);
846 filp = ERR_PTR(error);
847 }
848 } else {
849 /* Allow nd->intent.open.file to be recycled */
850 filp = res;
851 }
852 }
853 return filp;
854}
855
856/*
857 * dentry_open() will have done dput(dentry) and mntput(mnt) if it returns an
858 * error.
859 */
860struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
861 const struct cred *cred)
862{
863 int error;
864 struct file *f;
865
866 validate_creds(cred);
867
868 /* We must always pass in a valid mount pointer. */
869 BUG_ON(!mnt);
870
871 error = -ENFILE;
872 f = get_empty_filp();
873 if (f == NULL) {
874 dput(dentry);
875 mntput(mnt);
876 return ERR_PTR(error);
877 }
878
879 f->f_flags = flags;
880 return __dentry_open(dentry, mnt, f, NULL, cred);
881}
882EXPORT_SYMBOL(dentry_open);
883
884static void __put_unused_fd(struct files_struct *files, unsigned int fd)
885{
886 struct fdtable *fdt = files_fdtable(files);
887 __clear_open_fd(fd, fdt);
888 if (fd < files->next_fd)
889 files->next_fd = fd;
890}
891
892void put_unused_fd(unsigned int fd)
893{
894 struct files_struct *files = current->files;
895 spin_lock(&files->file_lock);
896 __put_unused_fd(files, fd);
897 spin_unlock(&files->file_lock);
898}
899
900EXPORT_SYMBOL(put_unused_fd);
901
902/*
903 * Install a file pointer in the fd array.
904 *
905 * The VFS is full of places where we drop the files lock between
906 * setting the open_fds bitmap and installing the file in the file
907 * array. At any such point, we are vulnerable to a dup2() race
908 * installing a file in the array before us. We need to detect this and
909 * fput() the struct file we are about to overwrite in this case.
910 *
911 * It should never happen - if we allow dup2() do it, _really_ bad things
912 * will follow.
913 */
914
915void fd_install(unsigned int fd, struct file *file)
916{
917 struct files_struct *files = current->files;
918 struct fdtable *fdt;
919 spin_lock(&files->file_lock);
920 fdt = files_fdtable(files);
921 BUG_ON(fdt->fd[fd] != NULL);
922 rcu_assign_pointer(fdt->fd[fd], file);
923 spin_unlock(&files->file_lock);
924}
925
926EXPORT_SYMBOL(fd_install);
927
928static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
929{
930 int lookup_flags = 0;
931 int acc_mode;
932
933 if (flags & O_CREAT)
934 op->mode = (mode & S_IALLUGO) | S_IFREG;
935 else
936 op->mode = 0;
937
938 /* Must never be set by userspace */
939 flags &= ~FMODE_NONOTIFY;
940
941 /*
942 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
943 * check for O_DSYNC if the need any syncing at all we enforce it's
944 * always set instead of having to deal with possibly weird behaviour
945 * for malicious applications setting only __O_SYNC.
946 */
947 if (flags & __O_SYNC)
948 flags |= O_DSYNC;
949
950 /*
951 * If we have O_PATH in the open flag. Then we
952 * cannot have anything other than the below set of flags
953 */
954 if (flags & O_PATH) {
955 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
956 acc_mode = 0;
957 } else {
958 acc_mode = MAY_OPEN | ACC_MODE(flags);
959 }
960
961 op->open_flag = flags;
962
963 /* O_TRUNC implies we need access checks for write permissions */
964 if (flags & O_TRUNC)
965 acc_mode |= MAY_WRITE;
966
967 /* Allow the LSM permission hook to distinguish append
968 access from general write access. */
969 if (flags & O_APPEND)
970 acc_mode |= MAY_APPEND;
971
972 op->acc_mode = acc_mode;
973
974 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
975
976 if (flags & O_CREAT) {
977 op->intent |= LOOKUP_CREATE;
978 if (flags & O_EXCL)
979 op->intent |= LOOKUP_EXCL;
980 }
981
982 if (flags & O_DIRECTORY)
983 lookup_flags |= LOOKUP_DIRECTORY;
984 if (!(flags & O_NOFOLLOW))
985 lookup_flags |= LOOKUP_FOLLOW;
986 return lookup_flags;
987}
988
989/**
990 * filp_open - open file and return file pointer
991 *
992 * @filename: path to open
993 * @flags: open flags as per the open(2) second argument
994 * @mode: mode for the new file if O_CREAT is set, else ignored
995 *
996 * This is the helper to open a file from kernelspace if you really
997 * have to. But in generally you should not do this, so please move
998 * along, nothing to see here..
999 */
1000struct file *filp_open(const char *filename, int flags, umode_t mode)
1001{
1002 struct open_flags op;
1003 int lookup = build_open_flags(flags, mode, &op);
1004 return do_filp_open(AT_FDCWD, filename, &op, lookup);
1005}
1006EXPORT_SYMBOL(filp_open);
1007
1008struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1009 const char *filename, int flags)
1010{
1011 struct open_flags op;
1012 int lookup = build_open_flags(flags, 0, &op);
1013 if (flags & O_CREAT)
1014 return ERR_PTR(-EINVAL);
1015 if (!filename && (flags & O_DIRECTORY))
1016 if (!dentry->d_inode->i_op->lookup)
1017 return ERR_PTR(-ENOTDIR);
1018 return do_file_open_root(dentry, mnt, filename, &op, lookup);
1019}
1020EXPORT_SYMBOL(file_open_root);
1021
1022long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1023{
1024 struct open_flags op;
1025 int lookup = build_open_flags(flags, mode, &op);
1026 char *tmp = getname(filename);
1027 int fd = PTR_ERR(tmp);
1028
1029 if (!IS_ERR(tmp)) {
1030 fd = get_unused_fd_flags(flags);
1031 if (fd >= 0) {
1032 struct file *f = do_filp_open(dfd, tmp, &op, lookup);
1033 if (IS_ERR(f)) {
1034 put_unused_fd(fd);
1035 fd = PTR_ERR(f);
1036 } else {
1037 fsnotify_open(f);
1038 fd_install(fd, f);
1039 }
1040 }
1041 putname(tmp);
1042 }
1043 return fd;
1044}
1045
1046SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1047{
1048 long ret;
1049
1050 if (force_o_largefile())
1051 flags |= O_LARGEFILE;
1052
1053 ret = do_sys_open(AT_FDCWD, filename, flags, mode);
1054 /* avoid REGPARM breakage on x86: */
1055 asmlinkage_protect(3, ret, filename, flags, mode);
1056 return ret;
1057}
1058
1059SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1060 umode_t, mode)
1061{
1062 long ret;
1063
1064 if (force_o_largefile())
1065 flags |= O_LARGEFILE;
1066
1067 ret = do_sys_open(dfd, filename, flags, mode);
1068 /* avoid REGPARM breakage on x86: */
1069 asmlinkage_protect(4, ret, dfd, filename, flags, mode);
1070 return ret;
1071}
1072
1073#ifndef __alpha__
1074
1075/*
1076 * For backward compatibility? Maybe this should be moved
1077 * into arch/i386 instead?
1078 */
1079SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1080{
1081 return sys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1082}
1083
1084#endif
1085
1086/*
1087 * "id" is the POSIX thread ID. We use the
1088 * files pointer for this..
1089 */
1090int filp_close(struct file *filp, fl_owner_t id)
1091{
1092 int retval = 0;
1093
1094 if (!file_count(filp)) {
1095 printk(KERN_ERR "VFS: Close: file count is 0\n");
1096 return 0;
1097 }
1098
1099 if (filp->f_op && filp->f_op->flush)
1100 retval = filp->f_op->flush(filp, id);
1101
1102 if (likely(!(filp->f_mode & FMODE_PATH))) {
1103 dnotify_flush(filp, id);
1104 locks_remove_posix(filp, id);
1105 }
1106 fput(filp);
1107 return retval;
1108}
1109
1110EXPORT_SYMBOL(filp_close);
1111
1112/*
1113 * Careful here! We test whether the file pointer is NULL before
1114 * releasing the fd. This ensures that one clone task can't release
1115 * an fd while another clone is opening it.
1116 */
1117SYSCALL_DEFINE1(close, unsigned int, fd)
1118{
1119 struct file * filp;
1120 struct files_struct *files = current->files;
1121 struct fdtable *fdt;
1122 int retval;
1123
1124 spin_lock(&files->file_lock);
1125 fdt = files_fdtable(files);
1126 if (fd >= fdt->max_fds)
1127 goto out_unlock;
1128 filp = fdt->fd[fd];
1129 if (!filp)
1130 goto out_unlock;
1131 rcu_assign_pointer(fdt->fd[fd], NULL);
1132 __clear_close_on_exec(fd, fdt);
1133 __put_unused_fd(files, fd);
1134 spin_unlock(&files->file_lock);
1135 retval = filp_close(filp, files);
1136
1137 /* can't restart close syscall because file table entry was cleared */
1138 if (unlikely(retval == -ERESTARTSYS ||
1139 retval == -ERESTARTNOINTR ||
1140 retval == -ERESTARTNOHAND ||
1141 retval == -ERESTART_RESTARTBLOCK))
1142 retval = -EINTR;
1143
1144 return retval;
1145
1146out_unlock:
1147 spin_unlock(&files->file_lock);
1148 return -EBADF;
1149}
1150EXPORT_SYMBOL(sys_close);
1151
1152/*
1153 * This routine simulates a hangup on the tty, to arrange that users
1154 * are given clean terminals at login time.
1155 */
1156SYSCALL_DEFINE0(vhangup)
1157{
1158 if (capable(CAP_SYS_TTY_CONFIG)) {
1159 tty_vhangup_self();
1160 return 0;
1161 }
1162 return -EPERM;
1163}
1164
1165/*
1166 * Called when an inode is about to be open.
1167 * We use this to disallow opening large files on 32bit systems if
1168 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1169 * on this flag in sys_open.
1170 */
1171int generic_file_open(struct inode * inode, struct file * filp)
1172{
1173 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1174 return -EOVERFLOW;
1175 return 0;
1176}
1177
1178EXPORT_SYMBOL(generic_file_open);
1179
1180/*
1181 * This is used by subsystems that don't want seekable
1182 * file descriptors. The function is not supposed to ever fail, the only
1183 * reason it returns an 'int' and not 'void' is so that it can be plugged
1184 * directly into file_operations structure.
1185 */
1186int nonseekable_open(struct inode *inode, struct file *filp)
1187{
1188 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1189 return 0;
1190}
1191
1192EXPORT_SYMBOL(nonseekable_open);