Loading...
1/*
2 * linux/fs/open.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7#include <linux/string.h>
8#include <linux/mm.h>
9#include <linux/file.h>
10#include <linux/fdtable.h>
11#include <linux/fsnotify.h>
12#include <linux/module.h>
13#include <linux/tty.h>
14#include <linux/namei.h>
15#include <linux/backing-dev.h>
16#include <linux/capability.h>
17#include <linux/securebits.h>
18#include <linux/security.h>
19#include <linux/mount.h>
20#include <linux/fcntl.h>
21#include <linux/slab.h>
22#include <asm/uaccess.h>
23#include <linux/fs.h>
24#include <linux/personality.h>
25#include <linux/pagemap.h>
26#include <linux/syscalls.h>
27#include <linux/rcupdate.h>
28#include <linux/audit.h>
29#include <linux/falloc.h>
30#include <linux/fs_struct.h>
31#include <linux/ima.h>
32#include <linux/dnotify.h>
33
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);
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/dnotify.h>
33#include <linux/compat.h>
34#include <linux/mnt_idmapping.h>
35#include <linux/filelock.h>
36
37#include "internal.h"
38
39int do_truncate(struct mnt_idmap *idmap, 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(idmap, 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(idmap, 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 mnt_idmap *idmap;
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 idmap = mnt_idmap(path->mnt);
89 error = inode_permission(idmap, 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(idmap, 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_ftruncate(struct file *file, loff_t length, int small)
157{
158 struct inode *inode;
159 struct dentry *dentry;
160 int error;
161
162 /* explicitly opened as large or we are on 64-bit box */
163 if (file->f_flags & O_LARGEFILE)
164 small = 0;
165
166 dentry = file->f_path.dentry;
167 inode = dentry->d_inode;
168 if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
169 return -EINVAL;
170
171 /* Cannot ftruncate over 2^31 bytes without large file support */
172 if (small && length > MAX_NON_LFS)
173 return -EINVAL;
174
175 /* Check IS_APPEND on real upper inode */
176 if (IS_APPEND(file_inode(file)))
177 return -EPERM;
178 sb_start_write(inode->i_sb);
179 error = security_file_truncate(file);
180 if (!error)
181 error = do_truncate(file_mnt_idmap(file), dentry, length,
182 ATTR_MTIME | ATTR_CTIME, file);
183 sb_end_write(inode->i_sb);
184
185 return error;
186}
187
188long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
189{
190 if (length < 0)
191 return -EINVAL;
192 CLASS(fd, f)(fd);
193 if (fd_empty(f))
194 return -EBADF;
195
196 return do_ftruncate(fd_file(f), length, small);
197}
198
199SYSCALL_DEFINE2(ftruncate, unsigned int, fd, off_t, length)
200{
201 return do_sys_ftruncate(fd, length, 1);
202}
203
204#ifdef CONFIG_COMPAT
205COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_off_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#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_TRUNCATE64)
225COMPAT_SYSCALL_DEFINE3(truncate64, const char __user *, pathname,
226 compat_arg_u64_dual(length))
227{
228 return ksys_truncate(pathname, compat_arg_u64_glue(length));
229}
230#endif
231
232#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FTRUNCATE64)
233COMPAT_SYSCALL_DEFINE3(ftruncate64, unsigned int, fd,
234 compat_arg_u64_dual(length))
235{
236 return ksys_ftruncate(fd, compat_arg_u64_glue(length));
237}
238#endif
239
240int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
241{
242 struct inode *inode = file_inode(file);
243 long ret;
244 loff_t sum;
245
246 if (offset < 0 || len <= 0)
247 return -EINVAL;
248
249 if (mode & ~(FALLOC_FL_MODE_MASK | FALLOC_FL_KEEP_SIZE))
250 return -EOPNOTSUPP;
251
252 /*
253 * Modes are exclusive, even if that is not obvious from the encoding
254 * as bit masks and the mix with the flag in the same namespace.
255 *
256 * To make things even more complicated, FALLOC_FL_ALLOCATE_RANGE is
257 * encoded as no bit set.
258 */
259 switch (mode & FALLOC_FL_MODE_MASK) {
260 case FALLOC_FL_ALLOCATE_RANGE:
261 case FALLOC_FL_UNSHARE_RANGE:
262 case FALLOC_FL_ZERO_RANGE:
263 break;
264 case FALLOC_FL_PUNCH_HOLE:
265 if (!(mode & FALLOC_FL_KEEP_SIZE))
266 return -EOPNOTSUPP;
267 break;
268 case FALLOC_FL_COLLAPSE_RANGE:
269 case FALLOC_FL_INSERT_RANGE:
270 if (mode & FALLOC_FL_KEEP_SIZE)
271 return -EOPNOTSUPP;
272 break;
273 default:
274 return -EOPNOTSUPP;
275 }
276
277 if (!(file->f_mode & FMODE_WRITE))
278 return -EBADF;
279
280 /*
281 * On append-only files only space preallocation is supported.
282 */
283 if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
284 return -EPERM;
285
286 if (IS_IMMUTABLE(inode))
287 return -EPERM;
288
289 /*
290 * We cannot allow any fallocate operation on an active swapfile
291 */
292 if (IS_SWAPFILE(inode))
293 return -ETXTBSY;
294
295 /*
296 * Revalidate the write permissions, in case security policy has
297 * changed since the files were opened.
298 */
299 ret = security_file_permission(file, MAY_WRITE);
300 if (ret)
301 return ret;
302
303 ret = fsnotify_file_area_perm(file, MAY_WRITE, &offset, len);
304 if (ret)
305 return ret;
306
307 if (S_ISFIFO(inode->i_mode))
308 return -ESPIPE;
309
310 if (S_ISDIR(inode->i_mode))
311 return -EISDIR;
312
313 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
314 return -ENODEV;
315
316 /* Check for wraparound */
317 if (check_add_overflow(offset, len, &sum))
318 return -EFBIG;
319
320 if (sum > inode->i_sb->s_maxbytes)
321 return -EFBIG;
322
323 if (!file->f_op->fallocate)
324 return -EOPNOTSUPP;
325
326 file_start_write(file);
327 ret = file->f_op->fallocate(file, mode, offset, len);
328
329 /*
330 * Create inotify and fanotify events.
331 *
332 * To keep the logic simple always create events if fallocate succeeds.
333 * This implies that events are even created if the file size remains
334 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
335 */
336 if (ret == 0)
337 fsnotify_modify(file);
338
339 file_end_write(file);
340 return ret;
341}
342EXPORT_SYMBOL_GPL(vfs_fallocate);
343
344int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
345{
346 CLASS(fd, f)(fd);
347
348 if (fd_empty(f))
349 return -EBADF;
350
351 return vfs_fallocate(fd_file(f), mode, offset, len);
352}
353
354SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
355{
356 return ksys_fallocate(fd, mode, offset, len);
357}
358
359#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FALLOCATE)
360COMPAT_SYSCALL_DEFINE6(fallocate, int, fd, int, mode, compat_arg_u64_dual(offset),
361 compat_arg_u64_dual(len))
362{
363 return ksys_fallocate(fd, mode, compat_arg_u64_glue(offset),
364 compat_arg_u64_glue(len));
365}
366#endif
367
368/*
369 * access() needs to use the real uid/gid, not the effective uid/gid.
370 * We do this by temporarily clearing all FS-related capabilities and
371 * switching the fsuid/fsgid around to the real ones.
372 *
373 * Creating new credentials is expensive, so we try to skip doing it,
374 * which we can if the result would match what we already got.
375 */
376static bool access_need_override_creds(int flags)
377{
378 const struct cred *cred;
379
380 if (flags & AT_EACCESS)
381 return false;
382
383 cred = current_cred();
384 if (!uid_eq(cred->fsuid, cred->uid) ||
385 !gid_eq(cred->fsgid, cred->gid))
386 return true;
387
388 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
389 kuid_t root_uid = make_kuid(cred->user_ns, 0);
390 if (!uid_eq(cred->uid, root_uid)) {
391 if (!cap_isclear(cred->cap_effective))
392 return true;
393 } else {
394 if (!cap_isidentical(cred->cap_effective,
395 cred->cap_permitted))
396 return true;
397 }
398 }
399
400 return false;
401}
402
403static const struct cred *access_override_creds(void)
404{
405 const struct cred *old_cred;
406 struct cred *override_cred;
407
408 override_cred = prepare_creds();
409 if (!override_cred)
410 return NULL;
411
412 /*
413 * XXX access_need_override_creds performs checks in hopes of skipping
414 * this work. Make sure it stays in sync if making any changes in this
415 * routine.
416 */
417
418 override_cred->fsuid = override_cred->uid;
419 override_cred->fsgid = override_cred->gid;
420
421 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
422 /* Clear the capabilities if we switch to a non-root user */
423 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
424 if (!uid_eq(override_cred->uid, root_uid))
425 cap_clear(override_cred->cap_effective);
426 else
427 override_cred->cap_effective =
428 override_cred->cap_permitted;
429 }
430
431 /*
432 * The new set of credentials can *only* be used in
433 * task-synchronous circumstances, and does not need
434 * RCU freeing, unless somebody then takes a separate
435 * reference to it.
436 *
437 * NOTE! This is _only_ true because this credential
438 * is used purely for override_creds() that installs
439 * it as the subjective cred. Other threads will be
440 * accessing ->real_cred, not the subjective cred.
441 *
442 * If somebody _does_ make a copy of this (using the
443 * 'get_current_cred()' function), that will clear the
444 * non_rcu field, because now that other user may be
445 * expecting RCU freeing. But normal thread-synchronous
446 * cred accesses will keep things non-racy to avoid RCU
447 * freeing.
448 */
449 override_cred->non_rcu = 1;
450
451 old_cred = override_creds(override_cred);
452
453 /* override_cred() gets its own ref */
454 put_cred(override_cred);
455
456 return old_cred;
457}
458
459static long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
460{
461 struct path path;
462 struct inode *inode;
463 int res;
464 unsigned int lookup_flags = LOOKUP_FOLLOW;
465 const struct cred *old_cred = NULL;
466
467 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
468 return -EINVAL;
469
470 if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
471 return -EINVAL;
472
473 if (flags & AT_SYMLINK_NOFOLLOW)
474 lookup_flags &= ~LOOKUP_FOLLOW;
475 if (flags & AT_EMPTY_PATH)
476 lookup_flags |= LOOKUP_EMPTY;
477
478 if (access_need_override_creds(flags)) {
479 old_cred = access_override_creds();
480 if (!old_cred)
481 return -ENOMEM;
482 }
483
484retry:
485 res = user_path_at(dfd, filename, lookup_flags, &path);
486 if (res)
487 goto out;
488
489 inode = d_backing_inode(path.dentry);
490
491 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
492 /*
493 * MAY_EXEC on regular files is denied if the fs is mounted
494 * with the "noexec" flag.
495 */
496 res = -EACCES;
497 if (path_noexec(&path))
498 goto out_path_release;
499 }
500
501 res = inode_permission(mnt_idmap(path.mnt), inode, mode | MAY_ACCESS);
502 /* SuS v2 requires we report a read only fs too */
503 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
504 goto out_path_release;
505 /*
506 * This is a rare case where using __mnt_is_readonly()
507 * is OK without a mnt_want/drop_write() pair. Since
508 * no actual write to the fs is performed here, we do
509 * not need to telegraph to that to anyone.
510 *
511 * By doing this, we accept that this access is
512 * inherently racy and know that the fs may change
513 * state before we even see this result.
514 */
515 if (__mnt_is_readonly(path.mnt))
516 res = -EROFS;
517
518out_path_release:
519 path_put(&path);
520 if (retry_estale(res, lookup_flags)) {
521 lookup_flags |= LOOKUP_REVAL;
522 goto retry;
523 }
524out:
525 if (old_cred)
526 revert_creds(old_cred);
527
528 return res;
529}
530
531SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
532{
533 return do_faccessat(dfd, filename, mode, 0);
534}
535
536SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
537 int, flags)
538{
539 return do_faccessat(dfd, filename, mode, flags);
540}
541
542SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
543{
544 return do_faccessat(AT_FDCWD, filename, mode, 0);
545}
546
547SYSCALL_DEFINE1(chdir, const char __user *, filename)
548{
549 struct path path;
550 int error;
551 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
552retry:
553 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
554 if (error)
555 goto out;
556
557 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
558 if (error)
559 goto dput_and_out;
560
561 set_fs_pwd(current->fs, &path);
562
563dput_and_out:
564 path_put(&path);
565 if (retry_estale(error, lookup_flags)) {
566 lookup_flags |= LOOKUP_REVAL;
567 goto retry;
568 }
569out:
570 return error;
571}
572
573SYSCALL_DEFINE1(fchdir, unsigned int, fd)
574{
575 CLASS(fd_raw, f)(fd);
576 int error;
577
578 if (fd_empty(f))
579 return -EBADF;
580
581 if (!d_can_lookup(fd_file(f)->f_path.dentry))
582 return -ENOTDIR;
583
584 error = file_permission(fd_file(f), MAY_EXEC | MAY_CHDIR);
585 if (!error)
586 set_fs_pwd(current->fs, &fd_file(f)->f_path);
587 return error;
588}
589
590SYSCALL_DEFINE1(chroot, const char __user *, filename)
591{
592 struct path path;
593 int error;
594 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
595retry:
596 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
597 if (error)
598 goto out;
599
600 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
601 if (error)
602 goto dput_and_out;
603
604 error = -EPERM;
605 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
606 goto dput_and_out;
607 error = security_path_chroot(&path);
608 if (error)
609 goto dput_and_out;
610
611 set_fs_root(current->fs, &path);
612 error = 0;
613dput_and_out:
614 path_put(&path);
615 if (retry_estale(error, lookup_flags)) {
616 lookup_flags |= LOOKUP_REVAL;
617 goto retry;
618 }
619out:
620 return error;
621}
622
623int chmod_common(const struct path *path, umode_t mode)
624{
625 struct inode *inode = path->dentry->d_inode;
626 struct inode *delegated_inode = NULL;
627 struct iattr newattrs;
628 int error;
629
630 error = mnt_want_write(path->mnt);
631 if (error)
632 return error;
633retry_deleg:
634 inode_lock(inode);
635 error = security_path_chmod(path, mode);
636 if (error)
637 goto out_unlock;
638 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
639 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
640 error = notify_change(mnt_idmap(path->mnt), path->dentry,
641 &newattrs, &delegated_inode);
642out_unlock:
643 inode_unlock(inode);
644 if (delegated_inode) {
645 error = break_deleg_wait(&delegated_inode);
646 if (!error)
647 goto retry_deleg;
648 }
649 mnt_drop_write(path->mnt);
650 return error;
651}
652
653int vfs_fchmod(struct file *file, umode_t mode)
654{
655 audit_file(file);
656 return chmod_common(&file->f_path, mode);
657}
658
659SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
660{
661 CLASS(fd, f)(fd);
662
663 if (fd_empty(f))
664 return -EBADF;
665
666 return vfs_fchmod(fd_file(f), mode);
667}
668
669static int do_fchmodat(int dfd, const char __user *filename, umode_t mode,
670 unsigned int flags)
671{
672 struct path path;
673 int error;
674 unsigned int lookup_flags;
675
676 if (unlikely(flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)))
677 return -EINVAL;
678
679 lookup_flags = (flags & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
680 if (flags & AT_EMPTY_PATH)
681 lookup_flags |= LOOKUP_EMPTY;
682
683retry:
684 error = user_path_at(dfd, filename, lookup_flags, &path);
685 if (!error) {
686 error = chmod_common(&path, mode);
687 path_put(&path);
688 if (retry_estale(error, lookup_flags)) {
689 lookup_flags |= LOOKUP_REVAL;
690 goto retry;
691 }
692 }
693 return error;
694}
695
696SYSCALL_DEFINE4(fchmodat2, int, dfd, const char __user *, filename,
697 umode_t, mode, unsigned int, flags)
698{
699 return do_fchmodat(dfd, filename, mode, flags);
700}
701
702SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
703 umode_t, mode)
704{
705 return do_fchmodat(dfd, filename, mode, 0);
706}
707
708SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
709{
710 return do_fchmodat(AT_FDCWD, filename, mode, 0);
711}
712
713/*
714 * Check whether @kuid is valid and if so generate and set vfsuid_t in
715 * ia_vfsuid.
716 *
717 * Return: true if @kuid is valid, false if not.
718 */
719static inline bool setattr_vfsuid(struct iattr *attr, kuid_t kuid)
720{
721 if (!uid_valid(kuid))
722 return false;
723 attr->ia_valid |= ATTR_UID;
724 attr->ia_vfsuid = VFSUIDT_INIT(kuid);
725 return true;
726}
727
728/*
729 * Check whether @kgid is valid and if so generate and set vfsgid_t in
730 * ia_vfsgid.
731 *
732 * Return: true if @kgid is valid, false if not.
733 */
734static inline bool setattr_vfsgid(struct iattr *attr, kgid_t kgid)
735{
736 if (!gid_valid(kgid))
737 return false;
738 attr->ia_valid |= ATTR_GID;
739 attr->ia_vfsgid = VFSGIDT_INIT(kgid);
740 return true;
741}
742
743int chown_common(const struct path *path, uid_t user, gid_t group)
744{
745 struct mnt_idmap *idmap;
746 struct user_namespace *fs_userns;
747 struct inode *inode = path->dentry->d_inode;
748 struct inode *delegated_inode = NULL;
749 int error;
750 struct iattr newattrs;
751 kuid_t uid;
752 kgid_t gid;
753
754 uid = make_kuid(current_user_ns(), user);
755 gid = make_kgid(current_user_ns(), group);
756
757 idmap = mnt_idmap(path->mnt);
758 fs_userns = i_user_ns(inode);
759
760retry_deleg:
761 newattrs.ia_vfsuid = INVALID_VFSUID;
762 newattrs.ia_vfsgid = INVALID_VFSGID;
763 newattrs.ia_valid = ATTR_CTIME;
764 if ((user != (uid_t)-1) && !setattr_vfsuid(&newattrs, uid))
765 return -EINVAL;
766 if ((group != (gid_t)-1) && !setattr_vfsgid(&newattrs, gid))
767 return -EINVAL;
768 inode_lock(inode);
769 if (!S_ISDIR(inode->i_mode))
770 newattrs.ia_valid |= ATTR_KILL_SUID | ATTR_KILL_PRIV |
771 setattr_should_drop_sgid(idmap, inode);
772 /* Continue to send actual fs values, not the mount values. */
773 error = security_path_chown(
774 path,
775 from_vfsuid(idmap, fs_userns, newattrs.ia_vfsuid),
776 from_vfsgid(idmap, fs_userns, newattrs.ia_vfsgid));
777 if (!error)
778 error = notify_change(idmap, path->dentry, &newattrs,
779 &delegated_inode);
780 inode_unlock(inode);
781 if (delegated_inode) {
782 error = break_deleg_wait(&delegated_inode);
783 if (!error)
784 goto retry_deleg;
785 }
786 return error;
787}
788
789int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
790 int flag)
791{
792 struct path path;
793 int error = -EINVAL;
794 int lookup_flags;
795
796 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
797 goto out;
798
799 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
800 if (flag & AT_EMPTY_PATH)
801 lookup_flags |= LOOKUP_EMPTY;
802retry:
803 error = user_path_at(dfd, filename, lookup_flags, &path);
804 if (error)
805 goto out;
806 error = mnt_want_write(path.mnt);
807 if (error)
808 goto out_release;
809 error = chown_common(&path, user, group);
810 mnt_drop_write(path.mnt);
811out_release:
812 path_put(&path);
813 if (retry_estale(error, lookup_flags)) {
814 lookup_flags |= LOOKUP_REVAL;
815 goto retry;
816 }
817out:
818 return error;
819}
820
821SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
822 gid_t, group, int, flag)
823{
824 return do_fchownat(dfd, filename, user, group, flag);
825}
826
827SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
828{
829 return do_fchownat(AT_FDCWD, filename, user, group, 0);
830}
831
832SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
833{
834 return do_fchownat(AT_FDCWD, filename, user, group,
835 AT_SYMLINK_NOFOLLOW);
836}
837
838int vfs_fchown(struct file *file, uid_t user, gid_t group)
839{
840 int error;
841
842 error = mnt_want_write_file(file);
843 if (error)
844 return error;
845 audit_file(file);
846 error = chown_common(&file->f_path, user, group);
847 mnt_drop_write_file(file);
848 return error;
849}
850
851int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
852{
853 CLASS(fd, f)(fd);
854
855 if (fd_empty(f))
856 return -EBADF;
857
858 return vfs_fchown(fd_file(f), user, group);
859}
860
861SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
862{
863 return ksys_fchown(fd, user, group);
864}
865
866static inline int file_get_write_access(struct file *f)
867{
868 int error;
869
870 error = get_write_access(f->f_inode);
871 if (unlikely(error))
872 return error;
873 error = mnt_get_write_access(f->f_path.mnt);
874 if (unlikely(error))
875 goto cleanup_inode;
876 if (unlikely(f->f_mode & FMODE_BACKING)) {
877 error = mnt_get_write_access(backing_file_user_path(f)->mnt);
878 if (unlikely(error))
879 goto cleanup_mnt;
880 }
881 return 0;
882
883cleanup_mnt:
884 mnt_put_write_access(f->f_path.mnt);
885cleanup_inode:
886 put_write_access(f->f_inode);
887 return error;
888}
889
890static int do_dentry_open(struct file *f,
891 int (*open)(struct inode *, struct file *))
892{
893 static const struct file_operations empty_fops = {};
894 struct inode *inode = f->f_path.dentry->d_inode;
895 int error;
896
897 path_get(&f->f_path);
898 f->f_inode = inode;
899 f->f_mapping = inode->i_mapping;
900 f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
901 f->f_sb_err = file_sample_sb_err(f);
902
903 if (unlikely(f->f_flags & O_PATH)) {
904 f->f_mode = FMODE_PATH | FMODE_OPENED;
905 f->f_op = &empty_fops;
906 return 0;
907 }
908
909 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) {
910 i_readcount_inc(inode);
911 } else if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
912 error = file_get_write_access(f);
913 if (unlikely(error))
914 goto cleanup_file;
915 f->f_mode |= FMODE_WRITER;
916 }
917
918 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
919 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
920 f->f_mode |= FMODE_ATOMIC_POS;
921
922 f->f_op = fops_get(inode->i_fop);
923 if (WARN_ON(!f->f_op)) {
924 error = -ENODEV;
925 goto cleanup_all;
926 }
927
928 error = security_file_open(f);
929 if (error)
930 goto cleanup_all;
931
932 error = fsnotify_open_perm(f);
933 if (error)
934 goto cleanup_all;
935
936 error = break_lease(file_inode(f), f->f_flags);
937 if (error)
938 goto cleanup_all;
939
940 /* normally all 3 are set; ->open() can clear them if needed */
941 f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
942 if (!open)
943 open = f->f_op->open;
944 if (open) {
945 error = open(inode, f);
946 if (error)
947 goto cleanup_all;
948 }
949 f->f_mode |= FMODE_OPENED;
950 if ((f->f_mode & FMODE_READ) &&
951 likely(f->f_op->read || f->f_op->read_iter))
952 f->f_mode |= FMODE_CAN_READ;
953 if ((f->f_mode & FMODE_WRITE) &&
954 likely(f->f_op->write || f->f_op->write_iter))
955 f->f_mode |= FMODE_CAN_WRITE;
956 if ((f->f_mode & FMODE_LSEEK) && !f->f_op->llseek)
957 f->f_mode &= ~FMODE_LSEEK;
958 if (f->f_mapping->a_ops && f->f_mapping->a_ops->direct_IO)
959 f->f_mode |= FMODE_CAN_ODIRECT;
960
961 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
962 f->f_iocb_flags = iocb_flags(f);
963
964 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
965
966 if ((f->f_flags & O_DIRECT) && !(f->f_mode & FMODE_CAN_ODIRECT))
967 return -EINVAL;
968
969 /*
970 * XXX: Huge page cache doesn't support writing yet. Drop all page
971 * cache for this file before processing writes.
972 */
973 if (f->f_mode & FMODE_WRITE) {
974 /*
975 * Depends on full fence from get_write_access() to synchronize
976 * against collapse_file() regarding i_writecount and nr_thps
977 * updates. Ensures subsequent insertion of THPs into the page
978 * cache will fail.
979 */
980 if (filemap_nr_thps(inode->i_mapping)) {
981 struct address_space *mapping = inode->i_mapping;
982
983 filemap_invalidate_lock(inode->i_mapping);
984 /*
985 * unmap_mapping_range just need to be called once
986 * here, because the private pages is not need to be
987 * unmapped mapping (e.g. data segment of dynamic
988 * shared libraries here).
989 */
990 unmap_mapping_range(mapping, 0, 0, 0);
991 truncate_inode_pages(mapping, 0);
992 filemap_invalidate_unlock(inode->i_mapping);
993 }
994 }
995
996 return 0;
997
998cleanup_all:
999 if (WARN_ON_ONCE(error > 0))
1000 error = -EINVAL;
1001 fops_put(f->f_op);
1002 put_file_access(f);
1003cleanup_file:
1004 path_put(&f->f_path);
1005 f->f_path.mnt = NULL;
1006 f->f_path.dentry = NULL;
1007 f->f_inode = NULL;
1008 return error;
1009}
1010
1011/**
1012 * finish_open - finish opening a file
1013 * @file: file pointer
1014 * @dentry: pointer to dentry
1015 * @open: open callback
1016 *
1017 * This can be used to finish opening a file passed to i_op->atomic_open().
1018 *
1019 * If the open callback is set to NULL, then the standard f_op->open()
1020 * filesystem callback is substituted.
1021 *
1022 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
1023 * the return value of d_splice_alias(), then the caller needs to perform dput()
1024 * on it after finish_open().
1025 *
1026 * Returns zero on success or -errno if the open failed.
1027 */
1028int finish_open(struct file *file, struct dentry *dentry,
1029 int (*open)(struct inode *, struct file *))
1030{
1031 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
1032
1033 file->f_path.dentry = dentry;
1034 return do_dentry_open(file, open);
1035}
1036EXPORT_SYMBOL(finish_open);
1037
1038/**
1039 * finish_no_open - finish ->atomic_open() without opening the file
1040 *
1041 * @file: file pointer
1042 * @dentry: dentry or NULL (as returned from ->lookup())
1043 *
1044 * This can be used to set the result of a successful lookup in ->atomic_open().
1045 *
1046 * NB: unlike finish_open() this function does consume the dentry reference and
1047 * the caller need not dput() it.
1048 *
1049 * Returns "0" which must be the return value of ->atomic_open() after having
1050 * called this function.
1051 */
1052int finish_no_open(struct file *file, struct dentry *dentry)
1053{
1054 file->f_path.dentry = dentry;
1055 return 0;
1056}
1057EXPORT_SYMBOL(finish_no_open);
1058
1059char *file_path(struct file *filp, char *buf, int buflen)
1060{
1061 return d_path(&filp->f_path, buf, buflen);
1062}
1063EXPORT_SYMBOL(file_path);
1064
1065/**
1066 * vfs_open - open the file at the given path
1067 * @path: path to open
1068 * @file: newly allocated file with f_flag initialized
1069 */
1070int vfs_open(const struct path *path, struct file *file)
1071{
1072 int ret;
1073
1074 file->f_path = *path;
1075 ret = do_dentry_open(file, NULL);
1076 if (!ret) {
1077 /*
1078 * Once we return a file with FMODE_OPENED, __fput() will call
1079 * fsnotify_close(), so we need fsnotify_open() here for
1080 * symmetry.
1081 */
1082 fsnotify_open(file);
1083 }
1084 return ret;
1085}
1086
1087struct file *dentry_open(const struct path *path, int flags,
1088 const struct cred *cred)
1089{
1090 int error;
1091 struct file *f;
1092
1093 /* We must always pass in a valid mount pointer. */
1094 BUG_ON(!path->mnt);
1095
1096 f = alloc_empty_file(flags, cred);
1097 if (!IS_ERR(f)) {
1098 error = vfs_open(path, f);
1099 if (error) {
1100 fput(f);
1101 f = ERR_PTR(error);
1102 }
1103 }
1104 return f;
1105}
1106EXPORT_SYMBOL(dentry_open);
1107
1108/**
1109 * dentry_create - Create and open a file
1110 * @path: path to create
1111 * @flags: O_ flags
1112 * @mode: mode bits for new file
1113 * @cred: credentials to use
1114 *
1115 * Caller must hold the parent directory's lock, and have prepared
1116 * a negative dentry, placed in @path->dentry, for the new file.
1117 *
1118 * Caller sets @path->mnt to the vfsmount of the filesystem where
1119 * the new file is to be created. The parent directory and the
1120 * negative dentry must reside on the same filesystem instance.
1121 *
1122 * On success, returns a "struct file *". Otherwise a ERR_PTR
1123 * is returned.
1124 */
1125struct file *dentry_create(const struct path *path, int flags, umode_t mode,
1126 const struct cred *cred)
1127{
1128 struct file *f;
1129 int error;
1130
1131 f = alloc_empty_file(flags, cred);
1132 if (IS_ERR(f))
1133 return f;
1134
1135 error = vfs_create(mnt_idmap(path->mnt),
1136 d_inode(path->dentry->d_parent),
1137 path->dentry, mode, true);
1138 if (!error)
1139 error = vfs_open(path, f);
1140
1141 if (unlikely(error)) {
1142 fput(f);
1143 return ERR_PTR(error);
1144 }
1145 return f;
1146}
1147EXPORT_SYMBOL(dentry_create);
1148
1149/**
1150 * kernel_file_open - open a file for kernel internal use
1151 * @path: path of the file to open
1152 * @flags: open flags
1153 * @cred: credentials for open
1154 *
1155 * Open a file for use by in-kernel consumers. The file is not accounted
1156 * against nr_files and must not be installed into the file descriptor
1157 * table.
1158 *
1159 * Return: Opened file on success, an error pointer on failure.
1160 */
1161struct file *kernel_file_open(const struct path *path, int flags,
1162 const struct cred *cred)
1163{
1164 struct file *f;
1165 int error;
1166
1167 f = alloc_empty_file_noaccount(flags, cred);
1168 if (IS_ERR(f))
1169 return f;
1170
1171 f->f_path = *path;
1172 error = do_dentry_open(f, NULL);
1173 if (error) {
1174 fput(f);
1175 return ERR_PTR(error);
1176 }
1177
1178 fsnotify_open(f);
1179 return f;
1180}
1181EXPORT_SYMBOL_GPL(kernel_file_open);
1182
1183#define WILL_CREATE(flags) (flags & (O_CREAT | __O_TMPFILE))
1184#define O_PATH_FLAGS (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1185
1186inline struct open_how build_open_how(int flags, umode_t mode)
1187{
1188 struct open_how how = {
1189 .flags = flags & VALID_OPEN_FLAGS,
1190 .mode = mode & S_IALLUGO,
1191 };
1192
1193 /* O_PATH beats everything else. */
1194 if (how.flags & O_PATH)
1195 how.flags &= O_PATH_FLAGS;
1196 /* Modes should only be set for create-like flags. */
1197 if (!WILL_CREATE(how.flags))
1198 how.mode = 0;
1199 return how;
1200}
1201
1202inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1203{
1204 u64 flags = how->flags;
1205 u64 strip = __FMODE_NONOTIFY | O_CLOEXEC;
1206 int lookup_flags = 0;
1207 int acc_mode = ACC_MODE(flags);
1208
1209 BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1210 "struct open_flags doesn't yet handle flags > 32 bits");
1211
1212 /*
1213 * Strip flags that either shouldn't be set by userspace like
1214 * FMODE_NONOTIFY or that aren't relevant in determining struct
1215 * open_flags like O_CLOEXEC.
1216 */
1217 flags &= ~strip;
1218
1219 /*
1220 * Older syscalls implicitly clear all of the invalid flags or argument
1221 * values before calling build_open_flags(), but openat2(2) checks all
1222 * of its arguments.
1223 */
1224 if (flags & ~VALID_OPEN_FLAGS)
1225 return -EINVAL;
1226 if (how->resolve & ~VALID_RESOLVE_FLAGS)
1227 return -EINVAL;
1228
1229 /* Scoping flags are mutually exclusive. */
1230 if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1231 return -EINVAL;
1232
1233 /* Deal with the mode. */
1234 if (WILL_CREATE(flags)) {
1235 if (how->mode & ~S_IALLUGO)
1236 return -EINVAL;
1237 op->mode = how->mode | S_IFREG;
1238 } else {
1239 if (how->mode != 0)
1240 return -EINVAL;
1241 op->mode = 0;
1242 }
1243
1244 /*
1245 * Block bugs where O_DIRECTORY | O_CREAT created regular files.
1246 * Note, that blocking O_DIRECTORY | O_CREAT here also protects
1247 * O_TMPFILE below which requires O_DIRECTORY being raised.
1248 */
1249 if ((flags & (O_DIRECTORY | O_CREAT)) == (O_DIRECTORY | O_CREAT))
1250 return -EINVAL;
1251
1252 /* Now handle the creative implementation of O_TMPFILE. */
1253 if (flags & __O_TMPFILE) {
1254 /*
1255 * In order to ensure programs get explicit errors when trying
1256 * to use O_TMPFILE on old kernels we enforce that O_DIRECTORY
1257 * is raised alongside __O_TMPFILE.
1258 */
1259 if (!(flags & O_DIRECTORY))
1260 return -EINVAL;
1261 if (!(acc_mode & MAY_WRITE))
1262 return -EINVAL;
1263 }
1264 if (flags & O_PATH) {
1265 /* O_PATH only permits certain other flags to be set. */
1266 if (flags & ~O_PATH_FLAGS)
1267 return -EINVAL;
1268 acc_mode = 0;
1269 }
1270
1271 /*
1272 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1273 * check for O_DSYNC if the need any syncing at all we enforce it's
1274 * always set instead of having to deal with possibly weird behaviour
1275 * for malicious applications setting only __O_SYNC.
1276 */
1277 if (flags & __O_SYNC)
1278 flags |= O_DSYNC;
1279
1280 op->open_flag = flags;
1281
1282 /* O_TRUNC implies we need access checks for write permissions */
1283 if (flags & O_TRUNC)
1284 acc_mode |= MAY_WRITE;
1285
1286 /* Allow the LSM permission hook to distinguish append
1287 access from general write access. */
1288 if (flags & O_APPEND)
1289 acc_mode |= MAY_APPEND;
1290
1291 op->acc_mode = acc_mode;
1292
1293 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1294
1295 if (flags & O_CREAT) {
1296 op->intent |= LOOKUP_CREATE;
1297 if (flags & O_EXCL) {
1298 op->intent |= LOOKUP_EXCL;
1299 flags |= O_NOFOLLOW;
1300 }
1301 }
1302
1303 if (flags & O_DIRECTORY)
1304 lookup_flags |= LOOKUP_DIRECTORY;
1305 if (!(flags & O_NOFOLLOW))
1306 lookup_flags |= LOOKUP_FOLLOW;
1307
1308 if (how->resolve & RESOLVE_NO_XDEV)
1309 lookup_flags |= LOOKUP_NO_XDEV;
1310 if (how->resolve & RESOLVE_NO_MAGICLINKS)
1311 lookup_flags |= LOOKUP_NO_MAGICLINKS;
1312 if (how->resolve & RESOLVE_NO_SYMLINKS)
1313 lookup_flags |= LOOKUP_NO_SYMLINKS;
1314 if (how->resolve & RESOLVE_BENEATH)
1315 lookup_flags |= LOOKUP_BENEATH;
1316 if (how->resolve & RESOLVE_IN_ROOT)
1317 lookup_flags |= LOOKUP_IN_ROOT;
1318 if (how->resolve & RESOLVE_CACHED) {
1319 /* Don't bother even trying for create/truncate/tmpfile open */
1320 if (flags & (O_TRUNC | O_CREAT | __O_TMPFILE))
1321 return -EAGAIN;
1322 lookup_flags |= LOOKUP_CACHED;
1323 }
1324
1325 op->lookup_flags = lookup_flags;
1326 return 0;
1327}
1328
1329/**
1330 * file_open_name - open file and return file pointer
1331 *
1332 * @name: struct filename containing path to open
1333 * @flags: open flags as per the open(2) second argument
1334 * @mode: mode for the new file if O_CREAT is set, else ignored
1335 *
1336 * This is the helper to open a file from kernelspace if you really
1337 * have to. But in generally you should not do this, so please move
1338 * along, nothing to see here..
1339 */
1340struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1341{
1342 struct open_flags op;
1343 struct open_how how = build_open_how(flags, mode);
1344 int err = build_open_flags(&how, &op);
1345 if (err)
1346 return ERR_PTR(err);
1347 return do_filp_open(AT_FDCWD, name, &op);
1348}
1349
1350/**
1351 * filp_open - open file and return file pointer
1352 *
1353 * @filename: path to open
1354 * @flags: open flags as per the open(2) second argument
1355 * @mode: mode for the new file if O_CREAT is set, else ignored
1356 *
1357 * This is the helper to open a file from kernelspace if you really
1358 * have to. But in generally you should not do this, so please move
1359 * along, nothing to see here..
1360 */
1361struct file *filp_open(const char *filename, int flags, umode_t mode)
1362{
1363 struct filename *name = getname_kernel(filename);
1364 struct file *file = ERR_CAST(name);
1365
1366 if (!IS_ERR(name)) {
1367 file = file_open_name(name, flags, mode);
1368 putname(name);
1369 }
1370 return file;
1371}
1372EXPORT_SYMBOL(filp_open);
1373
1374struct file *file_open_root(const struct path *root,
1375 const char *filename, int flags, umode_t mode)
1376{
1377 struct open_flags op;
1378 struct open_how how = build_open_how(flags, mode);
1379 int err = build_open_flags(&how, &op);
1380 if (err)
1381 return ERR_PTR(err);
1382 return do_file_open_root(root, filename, &op);
1383}
1384EXPORT_SYMBOL(file_open_root);
1385
1386static long do_sys_openat2(int dfd, const char __user *filename,
1387 struct open_how *how)
1388{
1389 struct open_flags op;
1390 int fd = build_open_flags(how, &op);
1391 struct filename *tmp;
1392
1393 if (fd)
1394 return fd;
1395
1396 tmp = getname(filename);
1397 if (IS_ERR(tmp))
1398 return PTR_ERR(tmp);
1399
1400 fd = get_unused_fd_flags(how->flags);
1401 if (fd >= 0) {
1402 struct file *f = do_filp_open(dfd, tmp, &op);
1403 if (IS_ERR(f)) {
1404 put_unused_fd(fd);
1405 fd = PTR_ERR(f);
1406 } else {
1407 fd_install(fd, f);
1408 }
1409 }
1410 putname(tmp);
1411 return fd;
1412}
1413
1414long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1415{
1416 struct open_how how = build_open_how(flags, mode);
1417 return do_sys_openat2(dfd, filename, &how);
1418}
1419
1420
1421SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1422{
1423 if (force_o_largefile())
1424 flags |= O_LARGEFILE;
1425 return do_sys_open(AT_FDCWD, filename, flags, mode);
1426}
1427
1428SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1429 umode_t, mode)
1430{
1431 if (force_o_largefile())
1432 flags |= O_LARGEFILE;
1433 return do_sys_open(dfd, filename, flags, mode);
1434}
1435
1436SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1437 struct open_how __user *, how, size_t, usize)
1438{
1439 int err;
1440 struct open_how tmp;
1441
1442 BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1443 BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1444
1445 if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1446 return -EINVAL;
1447 if (unlikely(usize > PAGE_SIZE))
1448 return -E2BIG;
1449
1450 err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1451 if (err)
1452 return err;
1453
1454 audit_openat2_how(&tmp);
1455
1456 /* O_LARGEFILE is only allowed for non-O_PATH. */
1457 if (!(tmp.flags & O_PATH) && force_o_largefile())
1458 tmp.flags |= O_LARGEFILE;
1459
1460 return do_sys_openat2(dfd, filename, &tmp);
1461}
1462
1463#ifdef CONFIG_COMPAT
1464/*
1465 * Exactly like sys_open(), except that it doesn't set the
1466 * O_LARGEFILE flag.
1467 */
1468COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1469{
1470 return do_sys_open(AT_FDCWD, filename, flags, mode);
1471}
1472
1473/*
1474 * Exactly like sys_openat(), except that it doesn't set the
1475 * O_LARGEFILE flag.
1476 */
1477COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1478{
1479 return do_sys_open(dfd, filename, flags, mode);
1480}
1481#endif
1482
1483#ifndef __alpha__
1484
1485/*
1486 * For backward compatibility? Maybe this should be moved
1487 * into arch/i386 instead?
1488 */
1489SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1490{
1491 int flags = O_CREAT | O_WRONLY | O_TRUNC;
1492
1493 if (force_o_largefile())
1494 flags |= O_LARGEFILE;
1495 return do_sys_open(AT_FDCWD, pathname, flags, mode);
1496}
1497#endif
1498
1499/*
1500 * "id" is the POSIX thread ID. We use the
1501 * files pointer for this..
1502 */
1503static int filp_flush(struct file *filp, fl_owner_t id)
1504{
1505 int retval = 0;
1506
1507 if (CHECK_DATA_CORRUPTION(file_count(filp) == 0,
1508 "VFS: Close: file count is 0 (f_op=%ps)",
1509 filp->f_op)) {
1510 return 0;
1511 }
1512
1513 if (filp->f_op->flush)
1514 retval = filp->f_op->flush(filp, id);
1515
1516 if (likely(!(filp->f_mode & FMODE_PATH))) {
1517 dnotify_flush(filp, id);
1518 locks_remove_posix(filp, id);
1519 }
1520 return retval;
1521}
1522
1523int filp_close(struct file *filp, fl_owner_t id)
1524{
1525 int retval;
1526
1527 retval = filp_flush(filp, id);
1528 fput(filp);
1529
1530 return retval;
1531}
1532EXPORT_SYMBOL(filp_close);
1533
1534/*
1535 * Careful here! We test whether the file pointer is NULL before
1536 * releasing the fd. This ensures that one clone task can't release
1537 * an fd while another clone is opening it.
1538 */
1539SYSCALL_DEFINE1(close, unsigned int, fd)
1540{
1541 int retval;
1542 struct file *file;
1543
1544 file = file_close_fd(fd);
1545 if (!file)
1546 return -EBADF;
1547
1548 retval = filp_flush(file, current->files);
1549
1550 /*
1551 * We're returning to user space. Don't bother
1552 * with any delayed fput() cases.
1553 */
1554 __fput_sync(file);
1555
1556 /* can't restart close syscall because file table entry was cleared */
1557 if (unlikely(retval == -ERESTARTSYS ||
1558 retval == -ERESTARTNOINTR ||
1559 retval == -ERESTARTNOHAND ||
1560 retval == -ERESTART_RESTARTBLOCK))
1561 retval = -EINTR;
1562
1563 return retval;
1564}
1565
1566/*
1567 * This routine simulates a hangup on the tty, to arrange that users
1568 * are given clean terminals at login time.
1569 */
1570SYSCALL_DEFINE0(vhangup)
1571{
1572 if (capable(CAP_SYS_TTY_CONFIG)) {
1573 tty_vhangup_self();
1574 return 0;
1575 }
1576 return -EPERM;
1577}
1578
1579/*
1580 * Called when an inode is about to be open.
1581 * We use this to disallow opening large files on 32bit systems if
1582 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1583 * on this flag in sys_open.
1584 */
1585int generic_file_open(struct inode * inode, struct file * filp)
1586{
1587 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1588 return -EOVERFLOW;
1589 return 0;
1590}
1591
1592EXPORT_SYMBOL(generic_file_open);
1593
1594/*
1595 * This is used by subsystems that don't want seekable
1596 * file descriptors. The function is not supposed to ever fail, the only
1597 * reason it returns an 'int' and not 'void' is so that it can be plugged
1598 * directly into file_operations structure.
1599 */
1600int nonseekable_open(struct inode *inode, struct file *filp)
1601{
1602 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1603 return 0;
1604}
1605
1606EXPORT_SYMBOL(nonseekable_open);
1607
1608/*
1609 * stream_open is used by subsystems that want stream-like file descriptors.
1610 * Such file descriptors are not seekable and don't have notion of position
1611 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1612 * Contrary to file descriptors of other regular files, .read() and .write()
1613 * can run simultaneously.
1614 *
1615 * stream_open never fails and is marked to return int so that it could be
1616 * directly used as file_operations.open .
1617 */
1618int stream_open(struct inode *inode, struct file *filp)
1619{
1620 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1621 filp->f_mode |= FMODE_STREAM;
1622 return 0;
1623}
1624
1625EXPORT_SYMBOL(stream_open);