Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
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 ret = inode_lock_killable(dentry->d_inode);
64 if (ret)
65 return ret;
66
67 /* Note any delegations or leases have already been broken: */
68 ret = notify_change(idmap, dentry, &newattrs, NULL);
69 inode_unlock(dentry->d_inode);
70 return ret;
71}
72
73int vfs_truncate(const struct path *path, loff_t length)
74{
75 struct mnt_idmap *idmap;
76 struct inode *inode;
77 int error;
78
79 inode = path->dentry->d_inode;
80
81 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
82 if (S_ISDIR(inode->i_mode))
83 return -EISDIR;
84 if (!S_ISREG(inode->i_mode))
85 return -EINVAL;
86
87 idmap = mnt_idmap(path->mnt);
88 error = inode_permission(idmap, inode, MAY_WRITE);
89 if (error)
90 return error;
91
92 error = fsnotify_truncate_perm(path, length);
93 if (error)
94 return error;
95
96 error = mnt_want_write(path->mnt);
97 if (error)
98 return error;
99
100 error = -EPERM;
101 if (IS_APPEND(inode))
102 goto mnt_drop_write_and_out;
103
104 error = get_write_access(inode);
105 if (error)
106 goto mnt_drop_write_and_out;
107
108 /*
109 * Make sure that there are no leases. get_write_access() protects
110 * against the truncate racing with a lease-granting setlease().
111 */
112 error = break_lease(inode, O_WRONLY);
113 if (error)
114 goto put_write_and_out;
115
116 error = security_path_truncate(path);
117 if (!error)
118 error = do_truncate(idmap, path->dentry, length, 0, NULL);
119
120put_write_and_out:
121 put_write_access(inode);
122mnt_drop_write_and_out:
123 mnt_drop_write(path->mnt);
124
125 return error;
126}
127EXPORT_SYMBOL_GPL(vfs_truncate);
128
129int do_sys_truncate(const char __user *pathname, loff_t length)
130{
131 unsigned int lookup_flags = LOOKUP_FOLLOW;
132 struct path path;
133 int error;
134
135 if (length < 0) /* sorry, but loff_t says... */
136 return -EINVAL;
137
138 CLASS(filename, name)(pathname);
139retry:
140 error = filename_lookup(AT_FDCWD, name, lookup_flags, &path, NULL);
141 if (!error) {
142 error = vfs_truncate(&path, length);
143 path_put(&path);
144 if (retry_estale(error, lookup_flags)) {
145 lookup_flags |= LOOKUP_REVAL;
146 goto retry;
147 }
148 }
149 return error;
150}
151
152SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
153{
154 return do_sys_truncate(path, length);
155}
156
157#ifdef CONFIG_COMPAT
158COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
159{
160 return do_sys_truncate(path, length);
161}
162#endif
163
164int do_ftruncate(struct file *file, loff_t length, int small)
165{
166 struct inode *inode;
167 struct dentry *dentry;
168 int error;
169
170 /* explicitly opened as large or we are on 64-bit box */
171 if (file->f_flags & O_LARGEFILE)
172 small = 0;
173
174 dentry = file->f_path.dentry;
175 inode = dentry->d_inode;
176 if (!S_ISREG(inode->i_mode) || !(file->f_mode & FMODE_WRITE))
177 return -EINVAL;
178
179 /* Cannot ftruncate over 2^31 bytes without large file support */
180 if (small && length > MAX_NON_LFS)
181 return -EINVAL;
182
183 /* Check IS_APPEND on real upper inode */
184 if (IS_APPEND(file_inode(file)))
185 return -EPERM;
186
187 error = security_file_truncate(file);
188 if (error)
189 return error;
190
191 error = fsnotify_truncate_perm(&file->f_path, length);
192 if (error)
193 return error;
194
195 scoped_guard(super_write, inode->i_sb)
196 return do_truncate(file_mnt_idmap(file), dentry, length,
197 ATTR_MTIME | ATTR_CTIME, file);
198}
199
200int do_sys_ftruncate(unsigned int fd, loff_t length, int small)
201{
202 if (length < 0)
203 return -EINVAL;
204 CLASS(fd, f)(fd);
205 if (fd_empty(f))
206 return -EBADF;
207
208 return do_ftruncate(fd_file(f), length, small);
209}
210
211SYSCALL_DEFINE2(ftruncate, unsigned int, fd, off_t, length)
212{
213 return do_sys_ftruncate(fd, length, 1);
214}
215
216#ifdef CONFIG_COMPAT
217COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_off_t, length)
218{
219 return do_sys_ftruncate(fd, length, 1);
220}
221#endif
222
223/* LFS versions of truncate are only needed on 32 bit machines */
224#if BITS_PER_LONG == 32
225SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
226{
227 return do_sys_truncate(path, length);
228}
229
230SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
231{
232 return do_sys_ftruncate(fd, length, 0);
233}
234#endif /* BITS_PER_LONG == 32 */
235
236#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_TRUNCATE64)
237COMPAT_SYSCALL_DEFINE3(truncate64, const char __user *, pathname,
238 compat_arg_u64_dual(length))
239{
240 return ksys_truncate(pathname, compat_arg_u64_glue(length));
241}
242#endif
243
244#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FTRUNCATE64)
245COMPAT_SYSCALL_DEFINE3(ftruncate64, unsigned int, fd,
246 compat_arg_u64_dual(length))
247{
248 return ksys_ftruncate(fd, compat_arg_u64_glue(length));
249}
250#endif
251
252int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
253{
254 struct inode *inode = file_inode(file);
255 int ret;
256 loff_t sum;
257
258 if (offset < 0 || len <= 0)
259 return -EINVAL;
260
261 if (mode & ~(FALLOC_FL_MODE_MASK | FALLOC_FL_KEEP_SIZE))
262 return -EOPNOTSUPP;
263
264 /*
265 * Modes are exclusive, even if that is not obvious from the encoding
266 * as bit masks and the mix with the flag in the same namespace.
267 *
268 * To make things even more complicated, FALLOC_FL_ALLOCATE_RANGE is
269 * encoded as no bit set.
270 */
271 switch (mode & FALLOC_FL_MODE_MASK) {
272 case FALLOC_FL_ALLOCATE_RANGE:
273 case FALLOC_FL_UNSHARE_RANGE:
274 case FALLOC_FL_ZERO_RANGE:
275 break;
276 case FALLOC_FL_PUNCH_HOLE:
277 if (!(mode & FALLOC_FL_KEEP_SIZE))
278 return -EOPNOTSUPP;
279 break;
280 case FALLOC_FL_COLLAPSE_RANGE:
281 case FALLOC_FL_INSERT_RANGE:
282 case FALLOC_FL_WRITE_ZEROES:
283 if (mode & FALLOC_FL_KEEP_SIZE)
284 return -EOPNOTSUPP;
285 break;
286 default:
287 return -EOPNOTSUPP;
288 }
289
290 if (!(file->f_mode & FMODE_WRITE))
291 return -EBADF;
292
293 /*
294 * On append-only files only space preallocation is supported.
295 */
296 if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
297 return -EPERM;
298
299 if (IS_IMMUTABLE(inode))
300 return -EPERM;
301
302 /*
303 * We cannot allow any fallocate operation on an active swapfile
304 */
305 if (IS_SWAPFILE(inode))
306 return -ETXTBSY;
307
308 /*
309 * Revalidate the write permissions, in case security policy has
310 * changed since the files were opened.
311 */
312 ret = security_file_permission(file, MAY_WRITE);
313 if (ret)
314 return ret;
315
316 ret = fsnotify_file_area_perm(file, MAY_WRITE, &offset, len);
317 if (ret)
318 return ret;
319
320 if (S_ISFIFO(inode->i_mode))
321 return -ESPIPE;
322
323 if (S_ISDIR(inode->i_mode))
324 return -EISDIR;
325
326 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
327 return -ENODEV;
328
329 /* Check for wraparound */
330 if (check_add_overflow(offset, len, &sum))
331 return -EFBIG;
332
333 if (sum > inode->i_sb->s_maxbytes)
334 return -EFBIG;
335
336 if (!file->f_op->fallocate)
337 return -EOPNOTSUPP;
338
339 file_start_write(file);
340 ret = file->f_op->fallocate(file, mode, offset, len);
341
342 /*
343 * Create inotify and fanotify events.
344 *
345 * To keep the logic simple always create events if fallocate succeeds.
346 * This implies that events are even created if the file size remains
347 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
348 */
349 if (ret == 0)
350 fsnotify_modify(file);
351
352 file_end_write(file);
353 return ret;
354}
355EXPORT_SYMBOL_GPL(vfs_fallocate);
356
357int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
358{
359 CLASS(fd, f)(fd);
360
361 if (fd_empty(f))
362 return -EBADF;
363
364 return vfs_fallocate(fd_file(f), mode, offset, len);
365}
366
367SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
368{
369 return ksys_fallocate(fd, mode, offset, len);
370}
371
372#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FALLOCATE)
373COMPAT_SYSCALL_DEFINE6(fallocate, int, fd, int, mode, compat_arg_u64_dual(offset),
374 compat_arg_u64_dual(len))
375{
376 return ksys_fallocate(fd, mode, compat_arg_u64_glue(offset),
377 compat_arg_u64_glue(len));
378}
379#endif
380
381/*
382 * access() needs to use the real uid/gid, not the effective uid/gid.
383 * We do this by temporarily clearing all FS-related capabilities and
384 * switching the fsuid/fsgid around to the real ones.
385 *
386 * Creating new credentials is expensive, so we try to skip doing it,
387 * which we can if the result would match what we already got.
388 */
389static bool access_need_override_creds(int flags)
390{
391 const struct cred *cred;
392
393 if (flags & AT_EACCESS)
394 return false;
395
396 cred = current_cred();
397 if (!uid_eq(cred->fsuid, cred->uid) ||
398 !gid_eq(cred->fsgid, cred->gid))
399 return true;
400
401 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
402 kuid_t root_uid = make_kuid(cred->user_ns, 0);
403 if (!uid_eq(cred->uid, root_uid)) {
404 if (!cap_isclear(cred->cap_effective))
405 return true;
406 } else {
407 if (!cap_isidentical(cred->cap_effective,
408 cred->cap_permitted))
409 return true;
410 }
411 }
412
413 return false;
414}
415
416static const struct cred *access_override_creds(void)
417{
418 struct cred *override_cred;
419
420 override_cred = prepare_creds();
421 if (!override_cred)
422 return NULL;
423
424 /*
425 * XXX access_need_override_creds performs checks in hopes of skipping
426 * this work. Make sure it stays in sync if making any changes in this
427 * routine.
428 */
429
430 override_cred->fsuid = override_cred->uid;
431 override_cred->fsgid = override_cred->gid;
432
433 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
434 /* Clear the capabilities if we switch to a non-root user */
435 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
436 if (!uid_eq(override_cred->uid, root_uid))
437 cap_clear(override_cred->cap_effective);
438 else
439 override_cred->cap_effective =
440 override_cred->cap_permitted;
441 }
442
443 /*
444 * The new set of credentials can *only* be used in
445 * task-synchronous circumstances, and does not need
446 * RCU freeing, unless somebody then takes a separate
447 * reference to it.
448 *
449 * NOTE! This is _only_ true because this credential
450 * is used purely for override_creds() that installs
451 * it as the subjective cred. Other threads will be
452 * accessing ->real_cred, not the subjective cred.
453 *
454 * If somebody _does_ make a copy of this (using the
455 * 'get_current_cred()' function), that will clear the
456 * non_rcu field, because now that other user may be
457 * expecting RCU freeing. But normal thread-synchronous
458 * cred accesses will keep things non-racy to avoid RCU
459 * freeing.
460 */
461 override_cred->non_rcu = 1;
462 return override_creds(override_cred);
463}
464
465static int do_faccessat(int dfd, const char __user *filename, int mode, int flags)
466{
467 struct path path;
468 struct inode *inode;
469 int res;
470 unsigned int lookup_flags = LOOKUP_FOLLOW;
471 const struct cred *old_cred = NULL;
472
473 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
474 return -EINVAL;
475
476 if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
477 return -EINVAL;
478
479 if (flags & AT_SYMLINK_NOFOLLOW)
480 lookup_flags &= ~LOOKUP_FOLLOW;
481
482 if (access_need_override_creds(flags)) {
483 old_cred = access_override_creds();
484 if (!old_cred)
485 return -ENOMEM;
486 }
487
488 CLASS(filename_uflags, name)(filename, flags);
489retry:
490 res = filename_lookup(dfd, name, lookup_flags, &path, NULL);
491 if (res)
492 goto out;
493
494 inode = d_backing_inode(path.dentry);
495
496 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
497 /*
498 * MAY_EXEC on regular files is denied if the fs is mounted
499 * with the "noexec" flag.
500 */
501 res = -EACCES;
502 if (path_noexec(&path))
503 goto out_path_release;
504 }
505
506 res = inode_permission(mnt_idmap(path.mnt), inode, mode | MAY_ACCESS);
507 /* SuS v2 requires we report a read only fs too */
508 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
509 goto out_path_release;
510 /*
511 * This is a rare case where using __mnt_is_readonly()
512 * is OK without a mnt_want/drop_write() pair. Since
513 * no actual write to the fs is performed here, we do
514 * not need to telegraph to that to anyone.
515 *
516 * By doing this, we accept that this access is
517 * inherently racy and know that the fs may change
518 * state before we even see this result.
519 */
520 if (__mnt_is_readonly(path.mnt))
521 res = -EROFS;
522
523out_path_release:
524 path_put(&path);
525 if (retry_estale(res, lookup_flags)) {
526 lookup_flags |= LOOKUP_REVAL;
527 goto retry;
528 }
529out:
530 if (old_cred)
531 put_cred(revert_creds(old_cred));
532
533 return res;
534}
535
536SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
537{
538 return do_faccessat(dfd, filename, mode, 0);
539}
540
541SYSCALL_DEFINE4(faccessat2, int, dfd, const char __user *, filename, int, mode,
542 int, flags)
543{
544 return do_faccessat(dfd, filename, mode, flags);
545}
546
547SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
548{
549 return do_faccessat(AT_FDCWD, filename, mode, 0);
550}
551
552SYSCALL_DEFINE1(chdir, const char __user *, filename)
553{
554 struct path path;
555 int error;
556 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
557 CLASS(filename, name)(filename);
558retry:
559 error = filename_lookup(AT_FDCWD, name, lookup_flags, &path, NULL);
560 if (!error) {
561 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
562 if (!error)
563 set_fs_pwd(current->fs, &path);
564 path_put(&path);
565 if (retry_estale(error, lookup_flags)) {
566 lookup_flags |= LOOKUP_REVAL;
567 goto retry;
568 }
569 }
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;
595 CLASS(filename, name)(filename);
596retry:
597 error = filename_lookup(AT_FDCWD, name, lookup_flags, &path, NULL);
598 if (error)
599 return error;
600
601 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
602 if (error)
603 goto dput_and_out;
604
605 error = -EPERM;
606 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
607 goto dput_and_out;
608 error = security_path_chroot(&path);
609 if (!error)
610 set_fs_root(current->fs, &path);
611dput_and_out:
612 path_put(&path);
613 if (retry_estale(error, lookup_flags)) {
614 lookup_flags |= LOOKUP_REVAL;
615 goto retry;
616 }
617 return error;
618}
619
620int chmod_common(const struct path *path, umode_t mode)
621{
622 struct inode *inode = path->dentry->d_inode;
623 struct delegated_inode delegated_inode = { };
624 struct iattr newattrs;
625 int error;
626
627 error = mnt_want_write(path->mnt);
628 if (error)
629 return error;
630retry_deleg:
631 error = inode_lock_killable(inode);
632 if (error)
633 goto out_mnt_unlock;
634 error = security_path_chmod(path, mode);
635 if (error)
636 goto out_unlock;
637 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
638 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
639 error = notify_change(mnt_idmap(path->mnt), path->dentry,
640 &newattrs, &delegated_inode);
641out_unlock:
642 inode_unlock(inode);
643 if (is_delegated(&delegated_inode)) {
644 error = break_deleg_wait(&delegated_inode);
645 if (!error)
646 goto retry_deleg;
647 }
648out_mnt_unlock:
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 CLASS(filename_uflags, name)(filename, flags);
681retry:
682 error = filename_lookup(dfd, name, lookup_flags, &path, NULL);
683 if (!error) {
684 error = chmod_common(&path, mode);
685 path_put(&path);
686 if (retry_estale(error, lookup_flags)) {
687 lookup_flags |= LOOKUP_REVAL;
688 goto retry;
689 }
690 }
691 return error;
692}
693
694SYSCALL_DEFINE4(fchmodat2, int, dfd, const char __user *, filename,
695 umode_t, mode, unsigned int, flags)
696{
697 return do_fchmodat(dfd, filename, mode, flags);
698}
699
700SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
701 umode_t, mode)
702{
703 return do_fchmodat(dfd, filename, mode, 0);
704}
705
706SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
707{
708 return do_fchmodat(AT_FDCWD, filename, mode, 0);
709}
710
711/*
712 * Check whether @kuid is valid and if so generate and set vfsuid_t in
713 * ia_vfsuid.
714 *
715 * Return: true if @kuid is valid, false if not.
716 */
717static inline bool setattr_vfsuid(struct iattr *attr, kuid_t kuid)
718{
719 if (!uid_valid(kuid))
720 return false;
721 attr->ia_valid |= ATTR_UID;
722 attr->ia_vfsuid = VFSUIDT_INIT(kuid);
723 return true;
724}
725
726/*
727 * Check whether @kgid is valid and if so generate and set vfsgid_t in
728 * ia_vfsgid.
729 *
730 * Return: true if @kgid is valid, false if not.
731 */
732static inline bool setattr_vfsgid(struct iattr *attr, kgid_t kgid)
733{
734 if (!gid_valid(kgid))
735 return false;
736 attr->ia_valid |= ATTR_GID;
737 attr->ia_vfsgid = VFSGIDT_INIT(kgid);
738 return true;
739}
740
741int chown_common(const struct path *path, uid_t user, gid_t group)
742{
743 struct mnt_idmap *idmap;
744 struct user_namespace *fs_userns;
745 struct inode *inode = path->dentry->d_inode;
746 struct delegated_inode delegated_inode = { };
747 int error;
748 struct iattr newattrs;
749 kuid_t uid;
750 kgid_t gid;
751
752 uid = make_kuid(current_user_ns(), user);
753 gid = make_kgid(current_user_ns(), group);
754
755 idmap = mnt_idmap(path->mnt);
756 fs_userns = i_user_ns(inode);
757
758retry_deleg:
759 newattrs.ia_vfsuid = INVALID_VFSUID;
760 newattrs.ia_vfsgid = INVALID_VFSGID;
761 newattrs.ia_valid = ATTR_CTIME;
762 if ((user != (uid_t)-1) && !setattr_vfsuid(&newattrs, uid))
763 return -EINVAL;
764 if ((group != (gid_t)-1) && !setattr_vfsgid(&newattrs, gid))
765 return -EINVAL;
766 error = inode_lock_killable(inode);
767 if (error)
768 return error;
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 (is_delegated(&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;
794 int lookup_flags;
795
796 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
797 return -EINVAL;
798
799 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
800 CLASS(filename_uflags, name)(filename, flag);
801retry:
802 error = filename_lookup(dfd, name, lookup_flags, &path, NULL);
803 if (!error) {
804 error = mnt_want_write(path.mnt);
805 if (!error) {
806 error = chown_common(&path, user, group);
807 mnt_drop_write(path.mnt);
808 }
809 path_put(&path);
810 if (retry_estale(error, lookup_flags)) {
811 lookup_flags |= LOOKUP_REVAL;
812 goto retry;
813 }
814 }
815 return error;
816}
817
818SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
819 gid_t, group, int, flag)
820{
821 return do_fchownat(dfd, filename, user, group, flag);
822}
823
824SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
825{
826 return do_fchownat(AT_FDCWD, filename, user, group, 0);
827}
828
829SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
830{
831 return do_fchownat(AT_FDCWD, filename, user, group,
832 AT_SYMLINK_NOFOLLOW);
833}
834
835int vfs_fchown(struct file *file, uid_t user, gid_t group)
836{
837 int error;
838
839 error = mnt_want_write_file(file);
840 if (error)
841 return error;
842 audit_file(file);
843 error = chown_common(&file->f_path, user, group);
844 mnt_drop_write_file(file);
845 return error;
846}
847
848int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
849{
850 CLASS(fd, f)(fd);
851
852 if (fd_empty(f))
853 return -EBADF;
854
855 return vfs_fchown(fd_file(f), user, group);
856}
857
858SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
859{
860 return ksys_fchown(fd, user, group);
861}
862
863static inline int file_get_write_access(struct file *f)
864{
865 int error;
866
867 error = get_write_access(f->f_inode);
868 if (unlikely(error))
869 return error;
870 error = mnt_get_write_access(f->f_path.mnt);
871 if (unlikely(error))
872 goto cleanup_inode;
873 if (unlikely(f->f_mode & FMODE_BACKING)) {
874 error = mnt_get_write_access(backing_file_user_path(f)->mnt);
875 if (unlikely(error))
876 goto cleanup_mnt;
877 }
878 return 0;
879
880cleanup_mnt:
881 mnt_put_write_access(f->f_path.mnt);
882cleanup_inode:
883 put_write_access(f->f_inode);
884 return error;
885}
886
887static int do_dentry_open(struct file *f,
888 int (*open)(struct inode *, struct file *))
889{
890 static const struct file_operations empty_fops = {};
891 struct inode *inode = f->f_path.dentry->d_inode;
892 int error;
893
894 path_get(&f->f_path);
895 f->f_inode = inode;
896 f->f_mapping = inode->i_mapping;
897 f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
898 f->f_sb_err = file_sample_sb_err(f);
899
900 if (unlikely(f->f_flags & O_PATH)) {
901 f->f_mode = FMODE_PATH | FMODE_OPENED;
902 file_set_fsnotify_mode(f, FMODE_NONOTIFY);
903 f->f_op = &empty_fops;
904 return 0;
905 }
906
907 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) {
908 i_readcount_inc(inode);
909 } else if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
910 error = file_get_write_access(f);
911 if (unlikely(error))
912 goto cleanup_file;
913 f->f_mode |= FMODE_WRITER;
914 }
915
916 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
917 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
918 f->f_mode |= FMODE_ATOMIC_POS;
919
920 f->f_op = fops_get(inode->i_fop);
921 if (WARN_ON(!f->f_op)) {
922 error = -ENODEV;
923 goto cleanup_all;
924 }
925
926 error = security_file_open(f);
927 if (unlikely(error))
928 goto cleanup_all;
929
930 /*
931 * Call fsnotify open permission hook and set FMODE_NONOTIFY_* bits
932 * according to existing permission watches.
933 * If FMODE_NONOTIFY mode was already set for an fanotify fd or for a
934 * pseudo file, this call will not change the mode.
935 */
936 error = fsnotify_open_perm_and_set_mode(f);
937 if (unlikely(error))
938 goto cleanup_all;
939
940 error = break_lease(file_inode(f), f->f_flags);
941 if (unlikely(error))
942 goto cleanup_all;
943
944 /* normally all 3 are set; ->open() can clear them if needed */
945 f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
946 if (!open)
947 open = f->f_op->open;
948 if (open) {
949 error = open(inode, f);
950 if (error)
951 goto cleanup_all;
952 }
953 f->f_mode |= FMODE_OPENED;
954 if ((f->f_mode & FMODE_READ) &&
955 likely(f->f_op->read || f->f_op->read_iter))
956 f->f_mode |= FMODE_CAN_READ;
957 if ((f->f_mode & FMODE_WRITE) &&
958 likely(f->f_op->write || f->f_op->write_iter))
959 f->f_mode |= FMODE_CAN_WRITE;
960 if ((f->f_mode & FMODE_LSEEK) && !f->f_op->llseek)
961 f->f_mode &= ~FMODE_LSEEK;
962 if (f->f_mapping->a_ops && f->f_mapping->a_ops->direct_IO)
963 f->f_mode |= FMODE_CAN_ODIRECT;
964
965 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
966 f->f_iocb_flags = iocb_flags(f);
967
968 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
969
970 if ((f->f_flags & O_DIRECT) && !(f->f_mode & FMODE_CAN_ODIRECT))
971 return -EINVAL;
972
973 /*
974 * XXX: Huge page cache doesn't support writing yet. Drop all page
975 * cache for this file before processing writes.
976 */
977 if (f->f_mode & FMODE_WRITE) {
978 /*
979 * Depends on full fence from get_write_access() to synchronize
980 * against collapse_file() regarding i_writecount and nr_thps
981 * updates. Ensures subsequent insertion of THPs into the page
982 * cache will fail.
983 */
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 return 0;
1001
1002cleanup_all:
1003 if (WARN_ON_ONCE(error > 0))
1004 error = -EINVAL;
1005 fops_put(f->f_op);
1006 put_file_access(f);
1007cleanup_file:
1008 path_put(&f->f_path);
1009 f->__f_path.mnt = NULL;
1010 f->__f_path.dentry = NULL;
1011 f->f_inode = NULL;
1012 return error;
1013}
1014
1015/**
1016 * finish_open - finish opening a file
1017 * @file: file pointer
1018 * @dentry: pointer to dentry
1019 * @open: open callback
1020 *
1021 * This can be used to finish opening a file passed to i_op->atomic_open().
1022 *
1023 * If the open callback is set to NULL, then the standard f_op->open()
1024 * filesystem callback is substituted.
1025 *
1026 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
1027 * the return value of d_splice_alias(), then the caller needs to perform dput()
1028 * on it after finish_open().
1029 *
1030 * Returns zero on success or -errno if the open failed.
1031 */
1032int finish_open(struct file *file, struct dentry *dentry,
1033 int (*open)(struct inode *, struct file *))
1034{
1035 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
1036
1037 file->__f_path.dentry = dentry;
1038 return do_dentry_open(file, open);
1039}
1040EXPORT_SYMBOL(finish_open);
1041
1042/**
1043 * finish_no_open - finish ->atomic_open() without opening the file
1044 *
1045 * @file: file pointer
1046 * @dentry: dentry, ERR_PTR(-E...) or NULL (as returned from ->lookup())
1047 *
1048 * This can be used to set the result of a lookup in ->atomic_open().
1049 *
1050 * NB: unlike finish_open() this function does consume the dentry reference and
1051 * the caller need not dput() it.
1052 *
1053 * Returns 0 or -E..., which must be the return value of ->atomic_open() after
1054 * having called this function.
1055 */
1056int finish_no_open(struct file *file, struct dentry *dentry)
1057{
1058 if (IS_ERR(dentry))
1059 return PTR_ERR(dentry);
1060 file->__f_path.dentry = dentry;
1061 return 0;
1062}
1063EXPORT_SYMBOL(finish_no_open);
1064
1065char *file_path(struct file *filp, char *buf, int buflen)
1066{
1067 return d_path(&filp->f_path, buf, buflen);
1068}
1069EXPORT_SYMBOL(file_path);
1070
1071/**
1072 * vfs_open - open the file at the given path
1073 * @path: path to open
1074 * @file: newly allocated file with f_flag initialized
1075 */
1076int vfs_open(const struct path *path, struct file *file)
1077{
1078 int ret;
1079
1080 file->__f_path = *path;
1081 ret = do_dentry_open(file, NULL);
1082 if (!ret) {
1083 /*
1084 * Once we return a file with FMODE_OPENED, __fput() will call
1085 * fsnotify_close(), so we need fsnotify_open() here for
1086 * symmetry.
1087 */
1088 fsnotify_open(file);
1089 }
1090 return ret;
1091}
1092
1093struct file *dentry_open(const struct path *path, int flags,
1094 const struct cred *cred)
1095{
1096 int error;
1097 struct file *f;
1098
1099 /* We must always pass in a valid mount pointer. */
1100 BUG_ON(!path->mnt);
1101
1102 f = alloc_empty_file(flags, cred);
1103 if (!IS_ERR(f)) {
1104 error = vfs_open(path, f);
1105 if (error) {
1106 fput(f);
1107 f = ERR_PTR(error);
1108 }
1109 }
1110 return f;
1111}
1112EXPORT_SYMBOL(dentry_open);
1113
1114struct file *dentry_open_nonotify(const struct path *path, int flags,
1115 const struct cred *cred)
1116{
1117 struct file *f = alloc_empty_file(flags, cred);
1118 if (!IS_ERR(f)) {
1119 int error;
1120
1121 file_set_fsnotify_mode(f, FMODE_NONOTIFY);
1122 error = vfs_open(path, f);
1123 if (error) {
1124 fput(f);
1125 f = ERR_PTR(error);
1126 }
1127 }
1128 return f;
1129}
1130
1131/**
1132 * kernel_file_open - open a file for kernel internal use
1133 * @path: path of the file to open
1134 * @flags: open flags
1135 * @cred: credentials for open
1136 *
1137 * Open a file for use by in-kernel consumers. The file is not accounted
1138 * against nr_files and must not be installed into the file descriptor
1139 * table.
1140 *
1141 * Return: Opened file on success, an error pointer on failure.
1142 */
1143struct file *kernel_file_open(const struct path *path, int flags,
1144 const struct cred *cred)
1145{
1146 struct file *f;
1147 int error;
1148
1149 f = alloc_empty_file_noaccount(flags, cred);
1150 if (IS_ERR(f))
1151 return f;
1152
1153 error = vfs_open(path, f);
1154 if (error) {
1155 fput(f);
1156 return ERR_PTR(error);
1157 }
1158 return f;
1159}
1160EXPORT_SYMBOL_GPL(kernel_file_open);
1161
1162#define WILL_CREATE(flags) (flags & (O_CREAT | __O_TMPFILE))
1163#define O_PATH_FLAGS (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1164
1165inline struct open_how build_open_how(int flags, umode_t mode)
1166{
1167 struct open_how how = {
1168 .flags = flags & VALID_OPEN_FLAGS,
1169 .mode = mode & S_IALLUGO,
1170 };
1171
1172 /* O_PATH beats everything else. */
1173 if (how.flags & O_PATH)
1174 how.flags &= O_PATH_FLAGS;
1175 /* Modes should only be set for create-like flags. */
1176 if (!WILL_CREATE(how.flags))
1177 how.mode = 0;
1178 return how;
1179}
1180
1181inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1182{
1183 u64 flags = how->flags;
1184 u64 strip = O_CLOEXEC;
1185 int lookup_flags = 0;
1186 int acc_mode = ACC_MODE(flags);
1187
1188 BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1189 "struct open_flags doesn't yet handle flags > 32 bits");
1190
1191 /*
1192 * Strip flags that aren't relevant in determining struct open_flags.
1193 */
1194 flags &= ~strip;
1195
1196 /*
1197 * Older syscalls implicitly clear all of the invalid flags or argument
1198 * values before calling build_open_flags(), but openat2(2) checks all
1199 * of its arguments.
1200 */
1201 if (flags & ~VALID_OPEN_FLAGS)
1202 return -EINVAL;
1203 if (how->resolve & ~VALID_RESOLVE_FLAGS)
1204 return -EINVAL;
1205
1206 /* Scoping flags are mutually exclusive. */
1207 if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1208 return -EINVAL;
1209
1210 /* Deal with the mode. */
1211 if (WILL_CREATE(flags)) {
1212 if (how->mode & ~S_IALLUGO)
1213 return -EINVAL;
1214 op->mode = how->mode | S_IFREG;
1215 } else {
1216 if (how->mode != 0)
1217 return -EINVAL;
1218 op->mode = 0;
1219 }
1220
1221 /*
1222 * Block bugs where O_DIRECTORY | O_CREAT created regular files.
1223 * Note, that blocking O_DIRECTORY | O_CREAT here also protects
1224 * O_TMPFILE below which requires O_DIRECTORY being raised.
1225 */
1226 if ((flags & (O_DIRECTORY | O_CREAT)) == (O_DIRECTORY | O_CREAT))
1227 return -EINVAL;
1228
1229 /* Now handle the creative implementation of O_TMPFILE. */
1230 if (flags & __O_TMPFILE) {
1231 /*
1232 * In order to ensure programs get explicit errors when trying
1233 * to use O_TMPFILE on old kernels we enforce that O_DIRECTORY
1234 * is raised alongside __O_TMPFILE.
1235 */
1236 if (!(flags & O_DIRECTORY))
1237 return -EINVAL;
1238 if (!(acc_mode & MAY_WRITE))
1239 return -EINVAL;
1240 }
1241 if (flags & O_PATH) {
1242 /* O_PATH only permits certain other flags to be set. */
1243 if (flags & ~O_PATH_FLAGS)
1244 return -EINVAL;
1245 acc_mode = 0;
1246 }
1247
1248 /*
1249 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1250 * check for O_DSYNC if the need any syncing at all we enforce it's
1251 * always set instead of having to deal with possibly weird behaviour
1252 * for malicious applications setting only __O_SYNC.
1253 */
1254 if (flags & __O_SYNC)
1255 flags |= O_DSYNC;
1256
1257 op->open_flag = flags;
1258
1259 /* O_TRUNC implies we need access checks for write permissions */
1260 if (flags & O_TRUNC)
1261 acc_mode |= MAY_WRITE;
1262
1263 /* Allow the LSM permission hook to distinguish append
1264 access from general write access. */
1265 if (flags & O_APPEND)
1266 acc_mode |= MAY_APPEND;
1267
1268 op->acc_mode = acc_mode;
1269
1270 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1271
1272 if (flags & O_CREAT) {
1273 op->intent |= LOOKUP_CREATE;
1274 if (flags & O_EXCL) {
1275 op->intent |= LOOKUP_EXCL;
1276 flags |= O_NOFOLLOW;
1277 }
1278 }
1279
1280 if (flags & O_DIRECTORY)
1281 lookup_flags |= LOOKUP_DIRECTORY;
1282 if (!(flags & O_NOFOLLOW))
1283 lookup_flags |= LOOKUP_FOLLOW;
1284
1285 if (how->resolve & RESOLVE_NO_XDEV)
1286 lookup_flags |= LOOKUP_NO_XDEV;
1287 if (how->resolve & RESOLVE_NO_MAGICLINKS)
1288 lookup_flags |= LOOKUP_NO_MAGICLINKS;
1289 if (how->resolve & RESOLVE_NO_SYMLINKS)
1290 lookup_flags |= LOOKUP_NO_SYMLINKS;
1291 if (how->resolve & RESOLVE_BENEATH)
1292 lookup_flags |= LOOKUP_BENEATH;
1293 if (how->resolve & RESOLVE_IN_ROOT)
1294 lookup_flags |= LOOKUP_IN_ROOT;
1295 if (how->resolve & RESOLVE_CACHED) {
1296 /* Don't bother even trying for create/truncate/tmpfile open */
1297 if (flags & (O_TRUNC | O_CREAT | __O_TMPFILE))
1298 return -EAGAIN;
1299 lookup_flags |= LOOKUP_CACHED;
1300 }
1301
1302 op->lookup_flags = lookup_flags;
1303 return 0;
1304}
1305
1306/**
1307 * file_open_name - open file and return file pointer
1308 *
1309 * @name: struct filename containing path to open
1310 * @flags: open flags as per the open(2) second argument
1311 * @mode: mode for the new file if O_CREAT is set, else ignored
1312 *
1313 * This is the helper to open a file from kernelspace if you really
1314 * have to. But in generally you should not do this, so please move
1315 * along, nothing to see here..
1316 */
1317struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1318{
1319 struct open_flags op;
1320 struct open_how how = build_open_how(flags, mode);
1321 int err = build_open_flags(&how, &op);
1322 if (err)
1323 return ERR_PTR(err);
1324 return do_file_open(AT_FDCWD, name, &op);
1325}
1326
1327/**
1328 * filp_open - open file and return file pointer
1329 *
1330 * @filename: path to open
1331 * @flags: open flags as per the open(2) second argument
1332 * @mode: mode for the new file if O_CREAT is set, else ignored
1333 *
1334 * This is the helper to open a file from kernelspace if you really
1335 * have to. But in generally you should not do this, so please move
1336 * along, nothing to see here..
1337 */
1338struct file *filp_open(const char *filename, int flags, umode_t mode)
1339{
1340 CLASS(filename_kernel, name)(filename);
1341 return file_open_name(name, flags, mode);
1342}
1343EXPORT_SYMBOL(filp_open);
1344
1345struct file *file_open_root(const struct path *root,
1346 const char *filename, int flags, umode_t mode)
1347{
1348 struct open_flags op;
1349 struct open_how how = build_open_how(flags, mode);
1350 int err = build_open_flags(&how, &op);
1351 if (err)
1352 return ERR_PTR(err);
1353 return do_file_open_root(root, filename, &op);
1354}
1355EXPORT_SYMBOL(file_open_root);
1356
1357static int do_sys_openat2(int dfd, const char __user *filename,
1358 struct open_how *how)
1359{
1360 struct open_flags op;
1361 int err = build_open_flags(how, &op);
1362 if (unlikely(err))
1363 return err;
1364
1365 CLASS(filename, name)(filename);
1366 return FD_ADD(how->flags, do_file_open(dfd, name, &op));
1367}
1368
1369int do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1370{
1371 struct open_how how = build_open_how(flags, mode);
1372 return do_sys_openat2(dfd, filename, &how);
1373}
1374
1375
1376SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1377{
1378 if (force_o_largefile())
1379 flags |= O_LARGEFILE;
1380 return do_sys_open(AT_FDCWD, filename, flags, mode);
1381}
1382
1383SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1384 umode_t, mode)
1385{
1386 if (force_o_largefile())
1387 flags |= O_LARGEFILE;
1388 return do_sys_open(dfd, filename, flags, mode);
1389}
1390
1391SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1392 struct open_how __user *, how, size_t, usize)
1393{
1394 int err;
1395 struct open_how tmp;
1396
1397 BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1398 BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1399
1400 if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1401 return -EINVAL;
1402 if (unlikely(usize > PAGE_SIZE))
1403 return -E2BIG;
1404
1405 err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1406 if (err)
1407 return err;
1408
1409 audit_openat2_how(&tmp);
1410
1411 /* O_LARGEFILE is only allowed for non-O_PATH. */
1412 if (!(tmp.flags & O_PATH) && force_o_largefile())
1413 tmp.flags |= O_LARGEFILE;
1414
1415 return do_sys_openat2(dfd, filename, &tmp);
1416}
1417
1418#ifdef CONFIG_COMPAT
1419/*
1420 * Exactly like sys_open(), except that it doesn't set the
1421 * O_LARGEFILE flag.
1422 */
1423COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1424{
1425 return do_sys_open(AT_FDCWD, filename, flags, mode);
1426}
1427
1428/*
1429 * Exactly like sys_openat(), except that it doesn't set the
1430 * O_LARGEFILE flag.
1431 */
1432COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1433{
1434 return do_sys_open(dfd, filename, flags, mode);
1435}
1436#endif
1437
1438#ifndef __alpha__
1439
1440/*
1441 * For backward compatibility? Maybe this should be moved
1442 * into arch/i386 instead?
1443 */
1444SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1445{
1446 int flags = O_CREAT | O_WRONLY | O_TRUNC;
1447
1448 if (force_o_largefile())
1449 flags |= O_LARGEFILE;
1450 return do_sys_open(AT_FDCWD, pathname, flags, mode);
1451}
1452#endif
1453
1454/*
1455 * "id" is the POSIX thread ID. We use the
1456 * files pointer for this..
1457 */
1458static int filp_flush(struct file *filp, fl_owner_t id)
1459{
1460 int retval = 0;
1461
1462 if (CHECK_DATA_CORRUPTION(file_count(filp) == 0, filp,
1463 "VFS: Close: file count is 0 (f_op=%ps)",
1464 filp->f_op)) {
1465 return 0;
1466 }
1467
1468 if (filp->f_op->flush)
1469 retval = filp->f_op->flush(filp, id);
1470
1471 if (likely(!(filp->f_mode & FMODE_PATH))) {
1472 dnotify_flush(filp, id);
1473 locks_remove_posix(filp, id);
1474 }
1475 return retval;
1476}
1477
1478int filp_close(struct file *filp, fl_owner_t id)
1479{
1480 int retval;
1481
1482 retval = filp_flush(filp, id);
1483 fput_close(filp);
1484
1485 return retval;
1486}
1487EXPORT_SYMBOL(filp_close);
1488
1489/*
1490 * Careful here! We test whether the file pointer is NULL before
1491 * releasing the fd. This ensures that one clone task can't release
1492 * an fd while another clone is opening it.
1493 */
1494SYSCALL_DEFINE1(close, unsigned int, fd)
1495{
1496 int retval;
1497 struct file *file;
1498
1499 file = file_close_fd(fd);
1500 if (!file)
1501 return -EBADF;
1502
1503 retval = filp_flush(file, current->files);
1504
1505 /*
1506 * We're returning to user space. Don't bother
1507 * with any delayed fput() cases.
1508 */
1509 fput_close_sync(file);
1510
1511 if (likely(retval == 0))
1512 return 0;
1513
1514 /* can't restart close syscall because file table entry was cleared */
1515 if (retval == -ERESTARTSYS ||
1516 retval == -ERESTARTNOINTR ||
1517 retval == -ERESTARTNOHAND ||
1518 retval == -ERESTART_RESTARTBLOCK)
1519 retval = -EINTR;
1520
1521 return retval;
1522}
1523
1524/*
1525 * This routine simulates a hangup on the tty, to arrange that users
1526 * are given clean terminals at login time.
1527 */
1528SYSCALL_DEFINE0(vhangup)
1529{
1530 if (capable(CAP_SYS_TTY_CONFIG)) {
1531 tty_vhangup_self();
1532 return 0;
1533 }
1534 return -EPERM;
1535}
1536
1537/*
1538 * Called when an inode is about to be open.
1539 * We use this to disallow opening large files on 32bit systems if
1540 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1541 * on this flag in sys_open.
1542 */
1543int generic_file_open(struct inode * inode, struct file * filp)
1544{
1545 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1546 return -EOVERFLOW;
1547 return 0;
1548}
1549
1550EXPORT_SYMBOL(generic_file_open);
1551
1552/*
1553 * This is used by subsystems that don't want seekable
1554 * file descriptors. The function is not supposed to ever fail, the only
1555 * reason it returns an 'int' and not 'void' is so that it can be plugged
1556 * directly into file_operations structure.
1557 */
1558int nonseekable_open(struct inode *inode, struct file *filp)
1559{
1560 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1561 return 0;
1562}
1563
1564EXPORT_SYMBOL(nonseekable_open);
1565
1566/*
1567 * stream_open is used by subsystems that want stream-like file descriptors.
1568 * Such file descriptors are not seekable and don't have notion of position
1569 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1570 * Contrary to file descriptors of other regular files, .read() and .write()
1571 * can run simultaneously.
1572 *
1573 * stream_open never fails and is marked to return int so that it could be
1574 * directly used as file_operations.open .
1575 */
1576int stream_open(struct inode *inode, struct file *filp)
1577{
1578 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1579 filp->f_mode |= FMODE_STREAM;
1580 return 0;
1581}
1582
1583EXPORT_SYMBOL(stream_open);