]> git.ipfire.org Git - thirdparty/linux.git/blob - fs/open.c
netfs: Allocate multipage folios in the writepath
[thirdparty/linux.git] / fs / open.c
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 #include <linux/filelock.h>
37
38 #include "internal.h"
39
40 int do_truncate(struct mnt_idmap *idmap, struct dentry *dentry,
41 loff_t length, unsigned int time_attrs, struct file *filp)
42 {
43 int ret;
44 struct iattr newattrs;
45
46 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
47 if (length < 0)
48 return -EINVAL;
49
50 newattrs.ia_size = length;
51 newattrs.ia_valid = ATTR_SIZE | time_attrs;
52 if (filp) {
53 newattrs.ia_file = filp;
54 newattrs.ia_valid |= ATTR_FILE;
55 }
56
57 /* Remove suid, sgid, and file capabilities on truncate too */
58 ret = dentry_needs_remove_privs(idmap, dentry);
59 if (ret < 0)
60 return ret;
61 if (ret)
62 newattrs.ia_valid |= ret | ATTR_FORCE;
63
64 inode_lock(dentry->d_inode);
65 /* Note any delegations or leases have already been broken: */
66 ret = notify_change(idmap, dentry, &newattrs, NULL);
67 inode_unlock(dentry->d_inode);
68 return ret;
69 }
70
71 long vfs_truncate(const struct path *path, loff_t length)
72 {
73 struct mnt_idmap *idmap;
74 struct inode *inode;
75 long error;
76
77 inode = path->dentry->d_inode;
78
79 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
80 if (S_ISDIR(inode->i_mode))
81 return -EISDIR;
82 if (!S_ISREG(inode->i_mode))
83 return -EINVAL;
84
85 error = mnt_want_write(path->mnt);
86 if (error)
87 goto out;
88
89 idmap = mnt_idmap(path->mnt);
90 error = inode_permission(idmap, inode, MAY_WRITE);
91 if (error)
92 goto mnt_drop_write_and_out;
93
94 error = -EPERM;
95 if (IS_APPEND(inode))
96 goto mnt_drop_write_and_out;
97
98 error = get_write_access(inode);
99 if (error)
100 goto mnt_drop_write_and_out;
101
102 /*
103 * Make sure that there are no leases. get_write_access() protects
104 * against the truncate racing with a lease-granting setlease().
105 */
106 error = break_lease(inode, O_WRONLY);
107 if (error)
108 goto put_write_and_out;
109
110 error = security_path_truncate(path);
111 if (!error)
112 error = do_truncate(idmap, path->dentry, length, 0, NULL);
113
114 put_write_and_out:
115 put_write_access(inode);
116 mnt_drop_write_and_out:
117 mnt_drop_write(path->mnt);
118 out:
119 return error;
120 }
121 EXPORT_SYMBOL_GPL(vfs_truncate);
122
123 long do_sys_truncate(const char __user *pathname, loff_t length)
124 {
125 unsigned int lookup_flags = LOOKUP_FOLLOW;
126 struct path path;
127 int error;
128
129 if (length < 0) /* sorry, but loff_t says... */
130 return -EINVAL;
131
132 retry:
133 error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
134 if (!error) {
135 error = vfs_truncate(&path, length);
136 path_put(&path);
137 }
138 if (retry_estale(error, lookup_flags)) {
139 lookup_flags |= LOOKUP_REVAL;
140 goto retry;
141 }
142 return error;
143 }
144
145 SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
146 {
147 return do_sys_truncate(path, length);
148 }
149
150 #ifdef CONFIG_COMPAT
151 COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
152 {
153 return do_sys_truncate(path, length);
154 }
155 #endif
156
157 long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
158 {
159 struct inode *inode;
160 struct dentry *dentry;
161 struct fd f;
162 int error;
163
164 error = -EINVAL;
165 if (length < 0)
166 goto out;
167 error = -EBADF;
168 f = fdget(fd);
169 if (!f.file)
170 goto out;
171
172 /* explicitly opened as large or we are on 64-bit box */
173 if (f.file->f_flags & O_LARGEFILE)
174 small = 0;
175
176 dentry = f.file->f_path.dentry;
177 inode = dentry->d_inode;
178 error = -EINVAL;
179 if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
180 goto out_putf;
181
182 error = -EINVAL;
183 /* Cannot ftruncate over 2^31 bytes without large file support */
184 if (small && length > MAX_NON_LFS)
185 goto out_putf;
186
187 error = -EPERM;
188 /* Check IS_APPEND on real upper inode */
189 if (IS_APPEND(file_inode(f.file)))
190 goto out_putf;
191 sb_start_write(inode->i_sb);
192 error = security_file_truncate(f.file);
193 if (!error)
194 error = do_truncate(file_mnt_idmap(f.file), dentry, length,
195 ATTR_MTIME | ATTR_CTIME, f.file);
196 sb_end_write(inode->i_sb);
197 out_putf:
198 fdput(f);
199 out:
200 return error;
201 }
202
203 SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
204 {
205 return do_sys_ftruncate(fd, length, 1);
206 }
207
208 #ifdef CONFIG_COMPAT
209 COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
210 {
211 return do_sys_ftruncate(fd, length, 1);
212 }
213 #endif
214
215 /* LFS versions of truncate are only needed on 32 bit machines */
216 #if BITS_PER_LONG == 32
217 SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
218 {
219 return do_sys_truncate(path, length);
220 }
221
222 SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
223 {
224 return do_sys_ftruncate(fd, length, 0);
225 }
226 #endif /* BITS_PER_LONG == 32 */
227
228 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_TRUNCATE64)
229 COMPAT_SYSCALL_DEFINE3(truncate64, const char __user *, pathname,
230 compat_arg_u64_dual(length))
231 {
232 return ksys_truncate(pathname, compat_arg_u64_glue(length));
233 }
234 #endif
235
236 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FTRUNCATE64)
237 COMPAT_SYSCALL_DEFINE3(ftruncate64, unsigned int, fd,
238 compat_arg_u64_dual(length))
239 {
240 return ksys_ftruncate(fd, compat_arg_u64_glue(length));
241 }
242 #endif
243
244 int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
245 {
246 struct inode *inode = file_inode(file);
247 long ret;
248
249 if (offset < 0 || len <= 0)
250 return -EINVAL;
251
252 /* Return error if mode is not supported */
253 if (mode & ~FALLOC_FL_SUPPORTED_MASK)
254 return -EOPNOTSUPP;
255
256 /* Punch hole and zero range are mutually exclusive */
257 if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
258 (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
259 return -EOPNOTSUPP;
260
261 /* Punch hole must have keep size set */
262 if ((mode & FALLOC_FL_PUNCH_HOLE) &&
263 !(mode & FALLOC_FL_KEEP_SIZE))
264 return -EOPNOTSUPP;
265
266 /* Collapse range should only be used exclusively. */
267 if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
268 (mode & ~FALLOC_FL_COLLAPSE_RANGE))
269 return -EINVAL;
270
271 /* Insert range should only be used exclusively. */
272 if ((mode & FALLOC_FL_INSERT_RANGE) &&
273 (mode & ~FALLOC_FL_INSERT_RANGE))
274 return -EINVAL;
275
276 /* Unshare range should only be used with allocate mode. */
277 if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
278 (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
279 return -EINVAL;
280
281 if (!(file->f_mode & FMODE_WRITE))
282 return -EBADF;
283
284 /*
285 * We can only allow pure fallocate on append only files
286 */
287 if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
288 return -EPERM;
289
290 if (IS_IMMUTABLE(inode))
291 return -EPERM;
292
293 /*
294 * We cannot allow any fallocate operation on an active swapfile
295 */
296 if (IS_SWAPFILE(inode))
297 return -ETXTBSY;
298
299 /*
300 * Revalidate the write permissions, in case security policy has
301 * changed since the files were opened.
302 */
303 ret = security_file_permission(file, MAY_WRITE);
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 wrap through zero too */
317 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
318 return -EFBIG;
319
320 if (!file->f_op->fallocate)
321 return -EOPNOTSUPP;
322
323 file_start_write(file);
324 ret = file->f_op->fallocate(file, mode, offset, len);
325
326 /*
327 * Create inotify and fanotify events.
328 *
329 * To keep the logic simple always create events if fallocate succeeds.
330 * This implies that events are even created if the file size remains
331 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
332 */
333 if (ret == 0)
334 fsnotify_modify(file);
335
336 file_end_write(file);
337 return ret;
338 }
339 EXPORT_SYMBOL_GPL(vfs_fallocate);
340
341 int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
342 {
343 struct fd f = fdget(fd);
344 int error = -EBADF;
345
346 if (f.file) {
347 error = vfs_fallocate(f.file, mode, offset, len);
348 fdput(f);
349 }
350 return error;
351 }
352
353 SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
354 {
355 return ksys_fallocate(fd, mode, offset, len);
356 }
357
358 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FALLOCATE)
359 COMPAT_SYSCALL_DEFINE6(fallocate, int, fd, int, mode, compat_arg_u64_dual(offset),
360 compat_arg_u64_dual(len))
361 {
362 return ksys_fallocate(fd, mode, compat_arg_u64_glue(offset),
363 compat_arg_u64_glue(len));
364 }
365 #endif
366
367 /*
368 * access() needs to use the real uid/gid, not the effective uid/gid.
369 * We do this by temporarily clearing all FS-related capabilities and
370 * switching the fsuid/fsgid around to the real ones.
371 *
372 * Creating new credentials is expensive, so we try to skip doing it,
373 * which we can if the result would match what we already got.
374 */
375 static bool access_need_override_creds(int flags)
376 {
377 const struct cred *cred;
378
379 if (flags & AT_EACCESS)
380 return false;
381
382 cred = current_cred();
383 if (!uid_eq(cred->fsuid, cred->uid) ||
384 !gid_eq(cred->fsgid, cred->gid))
385 return true;
386
387 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
388 kuid_t root_uid = make_kuid(cred->user_ns, 0);
389 if (!uid_eq(cred->uid, root_uid)) {
390 if (!cap_isclear(cred->cap_effective))
391 return true;
392 } else {
393 if (!cap_isidentical(cred->cap_effective,
394 cred->cap_permitted))
395 return true;
396 }
397 }
398
399 return false;
400 }
401
402 static const struct cred *access_override_creds(void)
403 {
404 const struct cred *old_cred;
405 struct cred *override_cred;
406
407 override_cred = prepare_creds();
408 if (!override_cred)
409 return NULL;
410
411 /*
412 * XXX access_need_override_creds performs checks in hopes of skipping
413 * this work. Make sure it stays in sync if making any changes in this
414 * routine.
415 */
416
417 override_cred->fsuid = override_cred->uid;
418 override_cred->fsgid = override_cred->gid;
419
420 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
421 /* Clear the capabilities if we switch to a non-root user */
422 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
423 if (!uid_eq(override_cred->uid, root_uid))
424 cap_clear(override_cred->cap_effective);
425 else
426 override_cred->cap_effective =
427 override_cred->cap_permitted;
428 }
429
430 /*
431 * The new set of credentials can *only* be used in
432 * task-synchronous circumstances, and does not need
433 * RCU freeing, unless somebody then takes a separate
434 * reference to it.
435 *
436 * NOTE! This is _only_ true because this credential
437 * is used purely for override_creds() that installs
438 * it as the subjective cred. Other threads will be
439 * accessing ->real_cred, not the subjective cred.
440 *
441 * If somebody _does_ make a copy of this (using the
442 * 'get_current_cred()' function), that will clear the
443 * non_rcu field, because now that other user may be
444 * expecting RCU freeing. But normal thread-synchronous
445 * cred accesses will keep things non-RCY.
446 */
447 override_cred->non_rcu = 1;
448
449 old_cred = override_creds(override_cred);
450
451 /* override_cred() gets its own ref */
452 put_cred(override_cred);
453
454 return old_cred;
455 }
456
457 static long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
458 {
459 struct path path;
460 struct inode *inode;
461 int res;
462 unsigned int lookup_flags = LOOKUP_FOLLOW;
463 const struct cred *old_cred = NULL;
464
465 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
466 return -EINVAL;
467
468 if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
469 return -EINVAL;
470
471 if (flags & AT_SYMLINK_NOFOLLOW)
472 lookup_flags &= ~LOOKUP_FOLLOW;
473 if (flags & AT_EMPTY_PATH)
474 lookup_flags |= LOOKUP_EMPTY;
475
476 if (access_need_override_creds(flags)) {
477 old_cred = access_override_creds();
478 if (!old_cred)
479 return -ENOMEM;
480 }
481
482 retry:
483 res = user_path_at(dfd, filename, lookup_flags, &path);
484 if (res)
485 goto out;
486
487 inode = d_backing_inode(path.dentry);
488
489 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
490 /*
491 * MAY_EXEC on regular files is denied if the fs is mounted
492 * with the "noexec" flag.
493 */
494 res = -EACCES;
495 if (path_noexec(&path))
496 goto out_path_release;
497 }
498
499 res = inode_permission(mnt_idmap(path.mnt), inode, mode | MAY_ACCESS);
500 /* SuS v2 requires we report a read only fs too */
501 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
502 goto out_path_release;
503 /*
504 * This is a rare case where using __mnt_is_readonly()
505 * is OK without a mnt_want/drop_write() pair. Since
506 * no actual write to the fs is performed here, we do
507 * not need to telegraph to that to anyone.
508 *
509 * By doing this, we accept that this access is
510 * inherently racy and know that the fs may change
511 * state before we even see this result.
512 */
513 if (__mnt_is_readonly(path.mnt))
514 res = -EROFS;
515
516 out_path_release:
517 path_put(&path);
518 if (retry_estale(res, lookup_flags)) {
519 lookup_flags |= LOOKUP_REVAL;
520 goto retry;
521 }
522 out:
523 if (old_cred)
524 revert_creds(old_cred);
525
526 return res;
527 }
528
529 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
530 {
531 return do_faccessat(dfd, filename, mode, 0);
532 }
533
534 SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
535 int, flags)
536 {
537 return do_faccessat(dfd, filename, mode, flags);
538 }
539
540 SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
541 {
542 return do_faccessat(AT_FDCWD, filename, mode, 0);
543 }
544
545 SYSCALL_DEFINE1(chdir, const char __user *, filename)
546 {
547 struct path path;
548 int error;
549 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
550 retry:
551 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
552 if (error)
553 goto out;
554
555 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
556 if (error)
557 goto dput_and_out;
558
559 set_fs_pwd(current->fs, &path);
560
561 dput_and_out:
562 path_put(&path);
563 if (retry_estale(error, lookup_flags)) {
564 lookup_flags |= LOOKUP_REVAL;
565 goto retry;
566 }
567 out:
568 return error;
569 }
570
571 SYSCALL_DEFINE1(fchdir, unsigned int, fd)
572 {
573 struct fd f = fdget_raw(fd);
574 int error;
575
576 error = -EBADF;
577 if (!f.file)
578 goto out;
579
580 error = -ENOTDIR;
581 if (!d_can_lookup(f.file->f_path.dentry))
582 goto out_putf;
583
584 error = file_permission(f.file, MAY_EXEC | MAY_CHDIR);
585 if (!error)
586 set_fs_pwd(current->fs, &f.file->f_path);
587 out_putf:
588 fdput(f);
589 out:
590 return error;
591 }
592
593 SYSCALL_DEFINE1(chroot, const char __user *, filename)
594 {
595 struct path path;
596 int error;
597 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
598 retry:
599 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
600 if (error)
601 goto out;
602
603 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
604 if (error)
605 goto dput_and_out;
606
607 error = -EPERM;
608 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
609 goto dput_and_out;
610 error = security_path_chroot(&path);
611 if (error)
612 goto dput_and_out;
613
614 set_fs_root(current->fs, &path);
615 error = 0;
616 dput_and_out:
617 path_put(&path);
618 if (retry_estale(error, lookup_flags)) {
619 lookup_flags |= LOOKUP_REVAL;
620 goto retry;
621 }
622 out:
623 return error;
624 }
625
626 int chmod_common(const struct path *path, umode_t mode)
627 {
628 struct inode *inode = path->dentry->d_inode;
629 struct inode *delegated_inode = NULL;
630 struct iattr newattrs;
631 int error;
632
633 error = mnt_want_write(path->mnt);
634 if (error)
635 return error;
636 retry_deleg:
637 inode_lock(inode);
638 error = security_path_chmod(path, mode);
639 if (error)
640 goto out_unlock;
641 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
642 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
643 error = notify_change(mnt_idmap(path->mnt), path->dentry,
644 &newattrs, &delegated_inode);
645 out_unlock:
646 inode_unlock(inode);
647 if (delegated_inode) {
648 error = break_deleg_wait(&delegated_inode);
649 if (!error)
650 goto retry_deleg;
651 }
652 mnt_drop_write(path->mnt);
653 return error;
654 }
655
656 int vfs_fchmod(struct file *file, umode_t mode)
657 {
658 audit_file(file);
659 return chmod_common(&file->f_path, mode);
660 }
661
662 SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
663 {
664 struct fd f = fdget(fd);
665 int err = -EBADF;
666
667 if (f.file) {
668 err = vfs_fchmod(f.file, mode);
669 fdput(f);
670 }
671 return err;
672 }
673
674 static int do_fchmodat(int dfd, const char __user *filename, umode_t mode,
675 unsigned int flags)
676 {
677 struct path path;
678 int error;
679 unsigned int lookup_flags;
680
681 if (unlikely(flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)))
682 return -EINVAL;
683
684 lookup_flags = (flags & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
685 if (flags & AT_EMPTY_PATH)
686 lookup_flags |= LOOKUP_EMPTY;
687
688 retry:
689 error = user_path_at(dfd, filename, lookup_flags, &path);
690 if (!error) {
691 error = chmod_common(&path, mode);
692 path_put(&path);
693 if (retry_estale(error, lookup_flags)) {
694 lookup_flags |= LOOKUP_REVAL;
695 goto retry;
696 }
697 }
698 return error;
699 }
700
701 SYSCALL_DEFINE4(fchmodat2, int, dfd, const char __user *, filename,
702 umode_t, mode, unsigned int, flags)
703 {
704 return do_fchmodat(dfd, filename, mode, flags);
705 }
706
707 SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
708 umode_t, mode)
709 {
710 return do_fchmodat(dfd, filename, mode, 0);
711 }
712
713 SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
714 {
715 return do_fchmodat(AT_FDCWD, filename, mode, 0);
716 }
717
718 /*
719 * Check whether @kuid is valid and if so generate and set vfsuid_t in
720 * ia_vfsuid.
721 *
722 * Return: true if @kuid is valid, false if not.
723 */
724 static inline bool setattr_vfsuid(struct iattr *attr, kuid_t kuid)
725 {
726 if (!uid_valid(kuid))
727 return false;
728 attr->ia_valid |= ATTR_UID;
729 attr->ia_vfsuid = VFSUIDT_INIT(kuid);
730 return true;
731 }
732
733 /*
734 * Check whether @kgid is valid and if so generate and set vfsgid_t in
735 * ia_vfsgid.
736 *
737 * Return: true if @kgid is valid, false if not.
738 */
739 static inline bool setattr_vfsgid(struct iattr *attr, kgid_t kgid)
740 {
741 if (!gid_valid(kgid))
742 return false;
743 attr->ia_valid |= ATTR_GID;
744 attr->ia_vfsgid = VFSGIDT_INIT(kgid);
745 return true;
746 }
747
748 int chown_common(const struct path *path, uid_t user, gid_t group)
749 {
750 struct mnt_idmap *idmap;
751 struct user_namespace *fs_userns;
752 struct inode *inode = path->dentry->d_inode;
753 struct inode *delegated_inode = NULL;
754 int error;
755 struct iattr newattrs;
756 kuid_t uid;
757 kgid_t gid;
758
759 uid = make_kuid(current_user_ns(), user);
760 gid = make_kgid(current_user_ns(), group);
761
762 idmap = mnt_idmap(path->mnt);
763 fs_userns = i_user_ns(inode);
764
765 retry_deleg:
766 newattrs.ia_vfsuid = INVALID_VFSUID;
767 newattrs.ia_vfsgid = INVALID_VFSGID;
768 newattrs.ia_valid = ATTR_CTIME;
769 if ((user != (uid_t)-1) && !setattr_vfsuid(&newattrs, uid))
770 return -EINVAL;
771 if ((group != (gid_t)-1) && !setattr_vfsgid(&newattrs, gid))
772 return -EINVAL;
773 inode_lock(inode);
774 if (!S_ISDIR(inode->i_mode))
775 newattrs.ia_valid |= ATTR_KILL_SUID | ATTR_KILL_PRIV |
776 setattr_should_drop_sgid(idmap, inode);
777 /* Continue to send actual fs values, not the mount values. */
778 error = security_path_chown(
779 path,
780 from_vfsuid(idmap, fs_userns, newattrs.ia_vfsuid),
781 from_vfsgid(idmap, fs_userns, newattrs.ia_vfsgid));
782 if (!error)
783 error = notify_change(idmap, path->dentry, &newattrs,
784 &delegated_inode);
785 inode_unlock(inode);
786 if (delegated_inode) {
787 error = break_deleg_wait(&delegated_inode);
788 if (!error)
789 goto retry_deleg;
790 }
791 return error;
792 }
793
794 int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
795 int flag)
796 {
797 struct path path;
798 int error = -EINVAL;
799 int lookup_flags;
800
801 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
802 goto out;
803
804 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
805 if (flag & AT_EMPTY_PATH)
806 lookup_flags |= LOOKUP_EMPTY;
807 retry:
808 error = user_path_at(dfd, filename, lookup_flags, &path);
809 if (error)
810 goto out;
811 error = mnt_want_write(path.mnt);
812 if (error)
813 goto out_release;
814 error = chown_common(&path, user, group);
815 mnt_drop_write(path.mnt);
816 out_release:
817 path_put(&path);
818 if (retry_estale(error, lookup_flags)) {
819 lookup_flags |= LOOKUP_REVAL;
820 goto retry;
821 }
822 out:
823 return error;
824 }
825
826 SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
827 gid_t, group, int, flag)
828 {
829 return do_fchownat(dfd, filename, user, group, flag);
830 }
831
832 SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
833 {
834 return do_fchownat(AT_FDCWD, filename, user, group, 0);
835 }
836
837 SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
838 {
839 return do_fchownat(AT_FDCWD, filename, user, group,
840 AT_SYMLINK_NOFOLLOW);
841 }
842
843 int vfs_fchown(struct file *file, uid_t user, gid_t group)
844 {
845 int error;
846
847 error = mnt_want_write_file(file);
848 if (error)
849 return error;
850 audit_file(file);
851 error = chown_common(&file->f_path, user, group);
852 mnt_drop_write_file(file);
853 return error;
854 }
855
856 int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
857 {
858 struct fd f = fdget(fd);
859 int error = -EBADF;
860
861 if (f.file) {
862 error = vfs_fchown(f.file, user, group);
863 fdput(f);
864 }
865 return error;
866 }
867
868 SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
869 {
870 return ksys_fchown(fd, user, group);
871 }
872
873 static inline int file_get_write_access(struct file *f)
874 {
875 int error;
876
877 error = get_write_access(f->f_inode);
878 if (unlikely(error))
879 return error;
880 error = mnt_get_write_access(f->f_path.mnt);
881 if (unlikely(error))
882 goto cleanup_inode;
883 if (unlikely(f->f_mode & FMODE_BACKING)) {
884 error = mnt_get_write_access(backing_file_user_path(f)->mnt);
885 if (unlikely(error))
886 goto cleanup_mnt;
887 }
888 return 0;
889
890 cleanup_mnt:
891 mnt_put_write_access(f->f_path.mnt);
892 cleanup_inode:
893 put_write_access(f->f_inode);
894 return error;
895 }
896
897 static int do_dentry_open(struct file *f,
898 struct inode *inode,
899 int (*open)(struct inode *, struct file *))
900 {
901 static const struct file_operations empty_fops = {};
902 int error;
903
904 path_get(&f->f_path);
905 f->f_inode = inode;
906 f->f_mapping = inode->i_mapping;
907 f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
908 f->f_sb_err = file_sample_sb_err(f);
909
910 if (unlikely(f->f_flags & O_PATH)) {
911 f->f_mode = FMODE_PATH | FMODE_OPENED;
912 f->f_op = &empty_fops;
913 return 0;
914 }
915
916 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) {
917 i_readcount_inc(inode);
918 } else if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
919 error = file_get_write_access(f);
920 if (unlikely(error))
921 goto cleanup_file;
922 f->f_mode |= FMODE_WRITER;
923 }
924
925 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
926 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
927 f->f_mode |= FMODE_ATOMIC_POS;
928
929 f->f_op = fops_get(inode->i_fop);
930 if (WARN_ON(!f->f_op)) {
931 error = -ENODEV;
932 goto cleanup_all;
933 }
934
935 error = security_file_open(f);
936 if (error)
937 goto cleanup_all;
938
939 error = break_lease(file_inode(f), f->f_flags);
940 if (error)
941 goto cleanup_all;
942
943 /* normally all 3 are set; ->open() can clear them if needed */
944 f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
945 if (!open)
946 open = f->f_op->open;
947 if (open) {
948 error = open(inode, f);
949 if (error)
950 goto cleanup_all;
951 }
952 f->f_mode |= FMODE_OPENED;
953 if ((f->f_mode & FMODE_READ) &&
954 likely(f->f_op->read || f->f_op->read_iter))
955 f->f_mode |= FMODE_CAN_READ;
956 if ((f->f_mode & FMODE_WRITE) &&
957 likely(f->f_op->write || f->f_op->write_iter))
958 f->f_mode |= FMODE_CAN_WRITE;
959 if ((f->f_mode & FMODE_LSEEK) && !f->f_op->llseek)
960 f->f_mode &= ~FMODE_LSEEK;
961 if (f->f_mapping->a_ops && f->f_mapping->a_ops->direct_IO)
962 f->f_mode |= FMODE_CAN_ODIRECT;
963
964 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
965 f->f_iocb_flags = iocb_flags(f);
966
967 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
968
969 if ((f->f_flags & O_DIRECT) && !(f->f_mode & FMODE_CAN_ODIRECT))
970 return -EINVAL;
971
972 /*
973 * XXX: Huge page cache doesn't support writing yet. Drop all page
974 * cache for this file before processing writes.
975 */
976 if (f->f_mode & FMODE_WRITE) {
977 /*
978 * Paired with smp_mb() in collapse_file() to ensure nr_thps
979 * is up to date and the update to i_writecount by
980 * get_write_access() is visible. Ensures subsequent insertion
981 * of THPs into the page cache will fail.
982 */
983 smp_mb();
984 if (filemap_nr_thps(inode->i_mapping)) {
985 struct address_space *mapping = inode->i_mapping;
986
987 filemap_invalidate_lock(inode->i_mapping);
988 /*
989 * unmap_mapping_range just need to be called once
990 * here, because the private pages is not need to be
991 * unmapped mapping (e.g. data segment of dynamic
992 * shared libraries here).
993 */
994 unmap_mapping_range(mapping, 0, 0, 0);
995 truncate_inode_pages(mapping, 0);
996 filemap_invalidate_unlock(inode->i_mapping);
997 }
998 }
999
1000 /*
1001 * Once we return a file with FMODE_OPENED, __fput() will call
1002 * fsnotify_close(), so we need fsnotify_open() here for symmetry.
1003 */
1004 fsnotify_open(f);
1005 return 0;
1006
1007 cleanup_all:
1008 if (WARN_ON_ONCE(error > 0))
1009 error = -EINVAL;
1010 fops_put(f->f_op);
1011 put_file_access(f);
1012 cleanup_file:
1013 path_put(&f->f_path);
1014 f->f_path.mnt = NULL;
1015 f->f_path.dentry = NULL;
1016 f->f_inode = NULL;
1017 return error;
1018 }
1019
1020 /**
1021 * finish_open - finish opening a file
1022 * @file: file pointer
1023 * @dentry: pointer to dentry
1024 * @open: open callback
1025 *
1026 * This can be used to finish opening a file passed to i_op->atomic_open().
1027 *
1028 * If the open callback is set to NULL, then the standard f_op->open()
1029 * filesystem callback is substituted.
1030 *
1031 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
1032 * the return value of d_splice_alias(), then the caller needs to perform dput()
1033 * on it after finish_open().
1034 *
1035 * Returns zero on success or -errno if the open failed.
1036 */
1037 int finish_open(struct file *file, struct dentry *dentry,
1038 int (*open)(struct inode *, struct file *))
1039 {
1040 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
1041
1042 file->f_path.dentry = dentry;
1043 return do_dentry_open(file, d_backing_inode(dentry), open);
1044 }
1045 EXPORT_SYMBOL(finish_open);
1046
1047 /**
1048 * finish_no_open - finish ->atomic_open() without opening the file
1049 *
1050 * @file: file pointer
1051 * @dentry: dentry or NULL (as returned from ->lookup())
1052 *
1053 * This can be used to set the result of a successful lookup in ->atomic_open().
1054 *
1055 * NB: unlike finish_open() this function does consume the dentry reference and
1056 * the caller need not dput() it.
1057 *
1058 * Returns "0" which must be the return value of ->atomic_open() after having
1059 * called this function.
1060 */
1061 int finish_no_open(struct file *file, struct dentry *dentry)
1062 {
1063 file->f_path.dentry = dentry;
1064 return 0;
1065 }
1066 EXPORT_SYMBOL(finish_no_open);
1067
1068 char *file_path(struct file *filp, char *buf, int buflen)
1069 {
1070 return d_path(&filp->f_path, buf, buflen);
1071 }
1072 EXPORT_SYMBOL(file_path);
1073
1074 /**
1075 * vfs_open - open the file at the given path
1076 * @path: path to open
1077 * @file: newly allocated file with f_flag initialized
1078 */
1079 int vfs_open(const struct path *path, struct file *file)
1080 {
1081 file->f_path = *path;
1082 return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
1083 }
1084
1085 struct file *dentry_open(const struct path *path, int flags,
1086 const struct cred *cred)
1087 {
1088 int error;
1089 struct file *f;
1090
1091 /* We must always pass in a valid mount pointer. */
1092 BUG_ON(!path->mnt);
1093
1094 f = alloc_empty_file(flags, cred);
1095 if (!IS_ERR(f)) {
1096 error = vfs_open(path, f);
1097 if (error) {
1098 fput(f);
1099 f = ERR_PTR(error);
1100 }
1101 }
1102 return f;
1103 }
1104 EXPORT_SYMBOL(dentry_open);
1105
1106 /**
1107 * dentry_create - Create and open a file
1108 * @path: path to create
1109 * @flags: O_ flags
1110 * @mode: mode bits for new file
1111 * @cred: credentials to use
1112 *
1113 * Caller must hold the parent directory's lock, and have prepared
1114 * a negative dentry, placed in @path->dentry, for the new file.
1115 *
1116 * Caller sets @path->mnt to the vfsmount of the filesystem where
1117 * the new file is to be created. The parent directory and the
1118 * negative dentry must reside on the same filesystem instance.
1119 *
1120 * On success, returns a "struct file *". Otherwise a ERR_PTR
1121 * is returned.
1122 */
1123 struct file *dentry_create(const struct path *path, int flags, umode_t mode,
1124 const struct cred *cred)
1125 {
1126 struct file *f;
1127 int error;
1128
1129 f = alloc_empty_file(flags, cred);
1130 if (IS_ERR(f))
1131 return f;
1132
1133 error = vfs_create(mnt_idmap(path->mnt),
1134 d_inode(path->dentry->d_parent),
1135 path->dentry, mode, true);
1136 if (!error)
1137 error = vfs_open(path, f);
1138
1139 if (unlikely(error)) {
1140 fput(f);
1141 return ERR_PTR(error);
1142 }
1143 return f;
1144 }
1145 EXPORT_SYMBOL(dentry_create);
1146
1147 /**
1148 * kernel_file_open - open a file for kernel internal use
1149 * @path: path of the file to open
1150 * @flags: open flags
1151 * @inode: the inode
1152 * @cred: credentials for open
1153 *
1154 * Open a file for use by in-kernel consumers. The file is not accounted
1155 * against nr_files and must not be installed into the file descriptor
1156 * table.
1157 *
1158 * Return: Opened file on success, an error pointer on failure.
1159 */
1160 struct file *kernel_file_open(const struct path *path, int flags,
1161 struct inode *inode, const struct cred *cred)
1162 {
1163 struct file *f;
1164 int error;
1165
1166 f = alloc_empty_file_noaccount(flags, cred);
1167 if (IS_ERR(f))
1168 return f;
1169
1170 f->f_path = *path;
1171 error = do_dentry_open(f, inode, NULL);
1172 if (error) {
1173 fput(f);
1174 f = ERR_PTR(error);
1175 }
1176 return f;
1177 }
1178 EXPORT_SYMBOL_GPL(kernel_file_open);
1179
1180 /**
1181 * backing_file_open - open a backing file for kernel internal use
1182 * @user_path: path that the user reuqested to open
1183 * @flags: open flags
1184 * @real_path: path of the backing file
1185 * @cred: credentials for open
1186 *
1187 * Open a backing file for a stackable filesystem (e.g., overlayfs).
1188 * @user_path may be on the stackable filesystem and @real_path on the
1189 * underlying filesystem. In this case, we want to be able to return the
1190 * @user_path of the stackable filesystem. This is done by embedding the
1191 * returned file into a container structure that also stores the stacked
1192 * file's path, which can be retrieved using backing_file_user_path().
1193 */
1194 struct file *backing_file_open(const struct path *user_path, int flags,
1195 const struct path *real_path,
1196 const struct cred *cred)
1197 {
1198 struct file *f;
1199 int error;
1200
1201 f = alloc_empty_backing_file(flags, cred);
1202 if (IS_ERR(f))
1203 return f;
1204
1205 path_get(user_path);
1206 *backing_file_user_path(f) = *user_path;
1207 f->f_path = *real_path;
1208 error = do_dentry_open(f, d_inode(real_path->dentry), NULL);
1209 if (error) {
1210 fput(f);
1211 f = ERR_PTR(error);
1212 }
1213
1214 return f;
1215 }
1216 EXPORT_SYMBOL_GPL(backing_file_open);
1217
1218 #define WILL_CREATE(flags) (flags & (O_CREAT | __O_TMPFILE))
1219 #define O_PATH_FLAGS (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1220
1221 inline struct open_how build_open_how(int flags, umode_t mode)
1222 {
1223 struct open_how how = {
1224 .flags = flags & VALID_OPEN_FLAGS,
1225 .mode = mode & S_IALLUGO,
1226 };
1227
1228 /* O_PATH beats everything else. */
1229 if (how.flags & O_PATH)
1230 how.flags &= O_PATH_FLAGS;
1231 /* Modes should only be set for create-like flags. */
1232 if (!WILL_CREATE(how.flags))
1233 how.mode = 0;
1234 return how;
1235 }
1236
1237 inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1238 {
1239 u64 flags = how->flags;
1240 u64 strip = __FMODE_NONOTIFY | O_CLOEXEC;
1241 int lookup_flags = 0;
1242 int acc_mode = ACC_MODE(flags);
1243
1244 BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1245 "struct open_flags doesn't yet handle flags > 32 bits");
1246
1247 /*
1248 * Strip flags that either shouldn't be set by userspace like
1249 * FMODE_NONOTIFY or that aren't relevant in determining struct
1250 * open_flags like O_CLOEXEC.
1251 */
1252 flags &= ~strip;
1253
1254 /*
1255 * Older syscalls implicitly clear all of the invalid flags or argument
1256 * values before calling build_open_flags(), but openat2(2) checks all
1257 * of its arguments.
1258 */
1259 if (flags & ~VALID_OPEN_FLAGS)
1260 return -EINVAL;
1261 if (how->resolve & ~VALID_RESOLVE_FLAGS)
1262 return -EINVAL;
1263
1264 /* Scoping flags are mutually exclusive. */
1265 if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1266 return -EINVAL;
1267
1268 /* Deal with the mode. */
1269 if (WILL_CREATE(flags)) {
1270 if (how->mode & ~S_IALLUGO)
1271 return -EINVAL;
1272 op->mode = how->mode | S_IFREG;
1273 } else {
1274 if (how->mode != 0)
1275 return -EINVAL;
1276 op->mode = 0;
1277 }
1278
1279 /*
1280 * Block bugs where O_DIRECTORY | O_CREAT created regular files.
1281 * Note, that blocking O_DIRECTORY | O_CREAT here also protects
1282 * O_TMPFILE below which requires O_DIRECTORY being raised.
1283 */
1284 if ((flags & (O_DIRECTORY | O_CREAT)) == (O_DIRECTORY | O_CREAT))
1285 return -EINVAL;
1286
1287 /* Now handle the creative implementation of O_TMPFILE. */
1288 if (flags & __O_TMPFILE) {
1289 /*
1290 * In order to ensure programs get explicit errors when trying
1291 * to use O_TMPFILE on old kernels we enforce that O_DIRECTORY
1292 * is raised alongside __O_TMPFILE.
1293 */
1294 if (!(flags & O_DIRECTORY))
1295 return -EINVAL;
1296 if (!(acc_mode & MAY_WRITE))
1297 return -EINVAL;
1298 }
1299 if (flags & O_PATH) {
1300 /* O_PATH only permits certain other flags to be set. */
1301 if (flags & ~O_PATH_FLAGS)
1302 return -EINVAL;
1303 acc_mode = 0;
1304 }
1305
1306 /*
1307 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1308 * check for O_DSYNC if the need any syncing at all we enforce it's
1309 * always set instead of having to deal with possibly weird behaviour
1310 * for malicious applications setting only __O_SYNC.
1311 */
1312 if (flags & __O_SYNC)
1313 flags |= O_DSYNC;
1314
1315 op->open_flag = flags;
1316
1317 /* O_TRUNC implies we need access checks for write permissions */
1318 if (flags & O_TRUNC)
1319 acc_mode |= MAY_WRITE;
1320
1321 /* Allow the LSM permission hook to distinguish append
1322 access from general write access. */
1323 if (flags & O_APPEND)
1324 acc_mode |= MAY_APPEND;
1325
1326 op->acc_mode = acc_mode;
1327
1328 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1329
1330 if (flags & O_CREAT) {
1331 op->intent |= LOOKUP_CREATE;
1332 if (flags & O_EXCL) {
1333 op->intent |= LOOKUP_EXCL;
1334 flags |= O_NOFOLLOW;
1335 }
1336 }
1337
1338 if (flags & O_DIRECTORY)
1339 lookup_flags |= LOOKUP_DIRECTORY;
1340 if (!(flags & O_NOFOLLOW))
1341 lookup_flags |= LOOKUP_FOLLOW;
1342
1343 if (how->resolve & RESOLVE_NO_XDEV)
1344 lookup_flags |= LOOKUP_NO_XDEV;
1345 if (how->resolve & RESOLVE_NO_MAGICLINKS)
1346 lookup_flags |= LOOKUP_NO_MAGICLINKS;
1347 if (how->resolve & RESOLVE_NO_SYMLINKS)
1348 lookup_flags |= LOOKUP_NO_SYMLINKS;
1349 if (how->resolve & RESOLVE_BENEATH)
1350 lookup_flags |= LOOKUP_BENEATH;
1351 if (how->resolve & RESOLVE_IN_ROOT)
1352 lookup_flags |= LOOKUP_IN_ROOT;
1353 if (how->resolve & RESOLVE_CACHED) {
1354 /* Don't bother even trying for create/truncate/tmpfile open */
1355 if (flags & (O_TRUNC | O_CREAT | __O_TMPFILE))
1356 return -EAGAIN;
1357 lookup_flags |= LOOKUP_CACHED;
1358 }
1359
1360 op->lookup_flags = lookup_flags;
1361 return 0;
1362 }
1363
1364 /**
1365 * file_open_name - open file and return file pointer
1366 *
1367 * @name: struct filename containing path to open
1368 * @flags: open flags as per the open(2) second argument
1369 * @mode: mode for the new file if O_CREAT is set, else ignored
1370 *
1371 * This is the helper to open a file from kernelspace if you really
1372 * have to. But in generally you should not do this, so please move
1373 * along, nothing to see here..
1374 */
1375 struct file *file_open_name(struct filename *name, 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_filp_open(AT_FDCWD, name, &op);
1383 }
1384
1385 /**
1386 * filp_open - open file and return file pointer
1387 *
1388 * @filename: path to open
1389 * @flags: open flags as per the open(2) second argument
1390 * @mode: mode for the new file if O_CREAT is set, else ignored
1391 *
1392 * This is the helper to open a file from kernelspace if you really
1393 * have to. But in generally you should not do this, so please move
1394 * along, nothing to see here..
1395 */
1396 struct file *filp_open(const char *filename, int flags, umode_t mode)
1397 {
1398 struct filename *name = getname_kernel(filename);
1399 struct file *file = ERR_CAST(name);
1400
1401 if (!IS_ERR(name)) {
1402 file = file_open_name(name, flags, mode);
1403 putname(name);
1404 }
1405 return file;
1406 }
1407 EXPORT_SYMBOL(filp_open);
1408
1409 struct file *file_open_root(const struct path *root,
1410 const char *filename, int flags, umode_t mode)
1411 {
1412 struct open_flags op;
1413 struct open_how how = build_open_how(flags, mode);
1414 int err = build_open_flags(&how, &op);
1415 if (err)
1416 return ERR_PTR(err);
1417 return do_file_open_root(root, filename, &op);
1418 }
1419 EXPORT_SYMBOL(file_open_root);
1420
1421 static long do_sys_openat2(int dfd, const char __user *filename,
1422 struct open_how *how)
1423 {
1424 struct open_flags op;
1425 int fd = build_open_flags(how, &op);
1426 struct filename *tmp;
1427
1428 if (fd)
1429 return fd;
1430
1431 tmp = getname(filename);
1432 if (IS_ERR(tmp))
1433 return PTR_ERR(tmp);
1434
1435 fd = get_unused_fd_flags(how->flags);
1436 if (fd >= 0) {
1437 struct file *f = do_filp_open(dfd, tmp, &op);
1438 if (IS_ERR(f)) {
1439 put_unused_fd(fd);
1440 fd = PTR_ERR(f);
1441 } else {
1442 fd_install(fd, f);
1443 }
1444 }
1445 putname(tmp);
1446 return fd;
1447 }
1448
1449 long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1450 {
1451 struct open_how how = build_open_how(flags, mode);
1452 return do_sys_openat2(dfd, filename, &how);
1453 }
1454
1455
1456 SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1457 {
1458 if (force_o_largefile())
1459 flags |= O_LARGEFILE;
1460 return do_sys_open(AT_FDCWD, filename, flags, mode);
1461 }
1462
1463 SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1464 umode_t, mode)
1465 {
1466 if (force_o_largefile())
1467 flags |= O_LARGEFILE;
1468 return do_sys_open(dfd, filename, flags, mode);
1469 }
1470
1471 SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1472 struct open_how __user *, how, size_t, usize)
1473 {
1474 int err;
1475 struct open_how tmp;
1476
1477 BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1478 BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1479
1480 if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1481 return -EINVAL;
1482
1483 err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1484 if (err)
1485 return err;
1486
1487 audit_openat2_how(&tmp);
1488
1489 /* O_LARGEFILE is only allowed for non-O_PATH. */
1490 if (!(tmp.flags & O_PATH) && force_o_largefile())
1491 tmp.flags |= O_LARGEFILE;
1492
1493 return do_sys_openat2(dfd, filename, &tmp);
1494 }
1495
1496 #ifdef CONFIG_COMPAT
1497 /*
1498 * Exactly like sys_open(), except that it doesn't set the
1499 * O_LARGEFILE flag.
1500 */
1501 COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1502 {
1503 return do_sys_open(AT_FDCWD, filename, flags, mode);
1504 }
1505
1506 /*
1507 * Exactly like sys_openat(), except that it doesn't set the
1508 * O_LARGEFILE flag.
1509 */
1510 COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1511 {
1512 return do_sys_open(dfd, filename, flags, mode);
1513 }
1514 #endif
1515
1516 #ifndef __alpha__
1517
1518 /*
1519 * For backward compatibility? Maybe this should be moved
1520 * into arch/i386 instead?
1521 */
1522 SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1523 {
1524 int flags = O_CREAT | O_WRONLY | O_TRUNC;
1525
1526 if (force_o_largefile())
1527 flags |= O_LARGEFILE;
1528 return do_sys_open(AT_FDCWD, pathname, flags, mode);
1529 }
1530 #endif
1531
1532 /*
1533 * "id" is the POSIX thread ID. We use the
1534 * files pointer for this..
1535 */
1536 static int filp_flush(struct file *filp, fl_owner_t id)
1537 {
1538 int retval = 0;
1539
1540 if (CHECK_DATA_CORRUPTION(file_count(filp) == 0,
1541 "VFS: Close: file count is 0 (f_op=%ps)",
1542 filp->f_op)) {
1543 return 0;
1544 }
1545
1546 if (filp->f_op->flush)
1547 retval = filp->f_op->flush(filp, id);
1548
1549 if (likely(!(filp->f_mode & FMODE_PATH))) {
1550 dnotify_flush(filp, id);
1551 locks_remove_posix(filp, id);
1552 }
1553 return retval;
1554 }
1555
1556 int filp_close(struct file *filp, fl_owner_t id)
1557 {
1558 int retval;
1559
1560 retval = filp_flush(filp, id);
1561 fput(filp);
1562
1563 return retval;
1564 }
1565 EXPORT_SYMBOL(filp_close);
1566
1567 /*
1568 * Careful here! We test whether the file pointer is NULL before
1569 * releasing the fd. This ensures that one clone task can't release
1570 * an fd while another clone is opening it.
1571 */
1572 SYSCALL_DEFINE1(close, unsigned int, fd)
1573 {
1574 int retval;
1575 struct file *file;
1576
1577 file = close_fd_get_file(fd);
1578 if (!file)
1579 return -EBADF;
1580
1581 retval = filp_flush(file, current->files);
1582
1583 /*
1584 * We're returning to user space. Don't bother
1585 * with any delayed fput() cases.
1586 */
1587 __fput_sync(file);
1588
1589 /* can't restart close syscall because file table entry was cleared */
1590 if (unlikely(retval == -ERESTARTSYS ||
1591 retval == -ERESTARTNOINTR ||
1592 retval == -ERESTARTNOHAND ||
1593 retval == -ERESTART_RESTARTBLOCK))
1594 retval = -EINTR;
1595
1596 return retval;
1597 }
1598
1599 /**
1600 * sys_close_range() - Close all file descriptors in a given range.
1601 *
1602 * @fd: starting file descriptor to close
1603 * @max_fd: last file descriptor to close
1604 * @flags: reserved for future extensions
1605 *
1606 * This closes a range of file descriptors. All file descriptors
1607 * from @fd up to and including @max_fd are closed.
1608 * Currently, errors to close a given file descriptor are ignored.
1609 */
1610 SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd,
1611 unsigned int, flags)
1612 {
1613 return __close_range(fd, max_fd, flags);
1614 }
1615
1616 /*
1617 * This routine simulates a hangup on the tty, to arrange that users
1618 * are given clean terminals at login time.
1619 */
1620 SYSCALL_DEFINE0(vhangup)
1621 {
1622 if (capable(CAP_SYS_TTY_CONFIG)) {
1623 tty_vhangup_self();
1624 return 0;
1625 }
1626 return -EPERM;
1627 }
1628
1629 /*
1630 * Called when an inode is about to be open.
1631 * We use this to disallow opening large files on 32bit systems if
1632 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1633 * on this flag in sys_open.
1634 */
1635 int generic_file_open(struct inode * inode, struct file * filp)
1636 {
1637 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1638 return -EOVERFLOW;
1639 return 0;
1640 }
1641
1642 EXPORT_SYMBOL(generic_file_open);
1643
1644 /*
1645 * This is used by subsystems that don't want seekable
1646 * file descriptors. The function is not supposed to ever fail, the only
1647 * reason it returns an 'int' and not 'void' is so that it can be plugged
1648 * directly into file_operations structure.
1649 */
1650 int nonseekable_open(struct inode *inode, struct file *filp)
1651 {
1652 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1653 return 0;
1654 }
1655
1656 EXPORT_SYMBOL(nonseekable_open);
1657
1658 /*
1659 * stream_open is used by subsystems that want stream-like file descriptors.
1660 * Such file descriptors are not seekable and don't have notion of position
1661 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1662 * Contrary to file descriptors of other regular files, .read() and .write()
1663 * can run simultaneously.
1664 *
1665 * stream_open never fails and is marked to return int so that it could be
1666 * directly used as file_operations.open .
1667 */
1668 int stream_open(struct inode *inode, struct file *filp)
1669 {
1670 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1671 filp->f_mode |= FMODE_STREAM;
1672 return 0;
1673 }
1674
1675 EXPORT_SYMBOL(stream_open);