Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/open.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8#include <linux/string.h>
9#include <linux/mm.h>
10#include <linux/file.h>
11#include <linux/fdtable.h>
12#include <linux/fsnotify.h>
13#include <linux/module.h>
14#include <linux/tty.h>
15#include <linux/namei.h>
16#include <linux/backing-dev.h>
17#include <linux/capability.h>
18#include <linux/securebits.h>
19#include <linux/security.h>
20#include <linux/mount.h>
21#include <linux/fcntl.h>
22#include <linux/slab.h>
23#include <linux/uaccess.h>
24#include <linux/fs.h>
25#include <linux/personality.h>
26#include <linux/pagemap.h>
27#include <linux/syscalls.h>
28#include <linux/rcupdate.h>
29#include <linux/audit.h>
30#include <linux/falloc.h>
31#include <linux/fs_struct.h>
32#include <linux/ima.h>
33#include <linux/dnotify.h>
34#include <linux/compat.h>
35
36#include "internal.h"
37
38int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
39 struct file *filp)
40{
41 int ret;
42 struct iattr newattrs;
43
44 /* Not pretty: "inode->i_size" shouldn't really be signed. But it is. */
45 if (length < 0)
46 return -EINVAL;
47
48 newattrs.ia_size = length;
49 newattrs.ia_valid = ATTR_SIZE | time_attrs;
50 if (filp) {
51 newattrs.ia_file = filp;
52 newattrs.ia_valid |= ATTR_FILE;
53 }
54
55 /* Remove suid, sgid, and file capabilities on truncate too */
56 ret = dentry_needs_remove_privs(dentry);
57 if (ret < 0)
58 return ret;
59 if (ret)
60 newattrs.ia_valid |= ret | ATTR_FORCE;
61
62 inode_lock(dentry->d_inode);
63 /* Note any delegations or leases have already been broken: */
64 ret = notify_change(dentry, &newattrs, NULL);
65 inode_unlock(dentry->d_inode);
66 return ret;
67}
68
69long vfs_truncate(const struct path *path, loff_t length)
70{
71 struct inode *inode;
72 long error;
73
74 inode = path->dentry->d_inode;
75
76 /* For directories it's -EISDIR, for other non-regulars - -EINVAL */
77 if (S_ISDIR(inode->i_mode))
78 return -EISDIR;
79 if (!S_ISREG(inode->i_mode))
80 return -EINVAL;
81
82 error = mnt_want_write(path->mnt);
83 if (error)
84 goto out;
85
86 error = inode_permission(inode, MAY_WRITE);
87 if (error)
88 goto mnt_drop_write_and_out;
89
90 error = -EPERM;
91 if (IS_APPEND(inode))
92 goto mnt_drop_write_and_out;
93
94 error = get_write_access(inode);
95 if (error)
96 goto mnt_drop_write_and_out;
97
98 /*
99 * Make sure that there are no leases. get_write_access() protects
100 * against the truncate racing with a lease-granting setlease().
101 */
102 error = break_lease(inode, O_WRONLY);
103 if (error)
104 goto put_write_and_out;
105
106 error = locks_verify_truncate(inode, NULL, length);
107 if (!error)
108 error = security_path_truncate(path);
109 if (!error)
110 error = do_truncate(path->dentry, length, 0, NULL);
111
112put_write_and_out:
113 put_write_access(inode);
114mnt_drop_write_and_out:
115 mnt_drop_write(path->mnt);
116out:
117 return error;
118}
119EXPORT_SYMBOL_GPL(vfs_truncate);
120
121long do_sys_truncate(const char __user *pathname, loff_t length)
122{
123 unsigned int lookup_flags = LOOKUP_FOLLOW;
124 struct path path;
125 int error;
126
127 if (length < 0) /* sorry, but loff_t says... */
128 return -EINVAL;
129
130retry:
131 error = user_path_at(AT_FDCWD, pathname, lookup_flags, &path);
132 if (!error) {
133 error = vfs_truncate(&path, length);
134 path_put(&path);
135 }
136 if (retry_estale(error, lookup_flags)) {
137 lookup_flags |= LOOKUP_REVAL;
138 goto retry;
139 }
140 return error;
141}
142
143SYSCALL_DEFINE2(truncate, const char __user *, path, long, length)
144{
145 return do_sys_truncate(path, length);
146}
147
148#ifdef CONFIG_COMPAT
149COMPAT_SYSCALL_DEFINE2(truncate, const char __user *, path, compat_off_t, length)
150{
151 return do_sys_truncate(path, length);
152}
153#endif
154
155long do_sys_ftruncate(unsigned int fd, loff_t length, int small)
156{
157 struct inode *inode;
158 struct dentry *dentry;
159 struct fd f;
160 int error;
161
162 error = -EINVAL;
163 if (length < 0)
164 goto out;
165 error = -EBADF;
166 f = fdget(fd);
167 if (!f.file)
168 goto out;
169
170 /* explicitly opened as large or we are on 64-bit box */
171 if (f.file->f_flags & O_LARGEFILE)
172 small = 0;
173
174 dentry = f.file->f_path.dentry;
175 inode = dentry->d_inode;
176 error = -EINVAL;
177 if (!S_ISREG(inode->i_mode) || !(f.file->f_mode & FMODE_WRITE))
178 goto out_putf;
179
180 error = -EINVAL;
181 /* Cannot ftruncate over 2^31 bytes without large file support */
182 if (small && length > MAX_NON_LFS)
183 goto out_putf;
184
185 error = -EPERM;
186 /* Check IS_APPEND on real upper inode */
187 if (IS_APPEND(file_inode(f.file)))
188 goto out_putf;
189
190 sb_start_write(inode->i_sb);
191 error = locks_verify_truncate(inode, f.file, length);
192 if (!error)
193 error = security_path_truncate(&f.file->f_path);
194 if (!error)
195 error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file);
196 sb_end_write(inode->i_sb);
197out_putf:
198 fdput(f);
199out:
200 return error;
201}
202
203SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
204{
205 return do_sys_ftruncate(fd, length, 1);
206}
207
208#ifdef CONFIG_COMPAT
209COMPAT_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
217SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
218{
219 return do_sys_truncate(path, length);
220}
221
222SYSCALL_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
229int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
230{
231 struct inode *inode = file_inode(file);
232 long ret;
233
234 if (offset < 0 || len <= 0)
235 return -EINVAL;
236
237 /* Return error if mode is not supported */
238 if (mode & ~FALLOC_FL_SUPPORTED_MASK)
239 return -EOPNOTSUPP;
240
241 /* Punch hole and zero range are mutually exclusive */
242 if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
243 (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
244 return -EOPNOTSUPP;
245
246 /* Punch hole must have keep size set */
247 if ((mode & FALLOC_FL_PUNCH_HOLE) &&
248 !(mode & FALLOC_FL_KEEP_SIZE))
249 return -EOPNOTSUPP;
250
251 /* Collapse range should only be used exclusively. */
252 if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
253 (mode & ~FALLOC_FL_COLLAPSE_RANGE))
254 return -EINVAL;
255
256 /* Insert range should only be used exclusively. */
257 if ((mode & FALLOC_FL_INSERT_RANGE) &&
258 (mode & ~FALLOC_FL_INSERT_RANGE))
259 return -EINVAL;
260
261 /* Unshare range should only be used with allocate mode. */
262 if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
263 (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
264 return -EINVAL;
265
266 if (!(file->f_mode & FMODE_WRITE))
267 return -EBADF;
268
269 /*
270 * We can only allow pure fallocate on append only files
271 */
272 if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
273 return -EPERM;
274
275 if (IS_IMMUTABLE(inode))
276 return -EPERM;
277
278 /*
279 * We cannot allow any fallocate operation on an active swapfile
280 */
281 if (IS_SWAPFILE(inode))
282 return -ETXTBSY;
283
284 /*
285 * Revalidate the write permissions, in case security policy has
286 * changed since the files were opened.
287 */
288 ret = security_file_permission(file, MAY_WRITE);
289 if (ret)
290 return ret;
291
292 if (S_ISFIFO(inode->i_mode))
293 return -ESPIPE;
294
295 if (S_ISDIR(inode->i_mode))
296 return -EISDIR;
297
298 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
299 return -ENODEV;
300
301 /* Check for wrap through zero too */
302 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
303 return -EFBIG;
304
305 if (!file->f_op->fallocate)
306 return -EOPNOTSUPP;
307
308 file_start_write(file);
309 ret = file->f_op->fallocate(file, mode, offset, len);
310
311 /*
312 * Create inotify and fanotify events.
313 *
314 * To keep the logic simple always create events if fallocate succeeds.
315 * This implies that events are even created if the file size remains
316 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
317 */
318 if (ret == 0)
319 fsnotify_modify(file);
320
321 file_end_write(file);
322 return ret;
323}
324EXPORT_SYMBOL_GPL(vfs_fallocate);
325
326int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
327{
328 struct fd f = fdget(fd);
329 int error = -EBADF;
330
331 if (f.file) {
332 error = vfs_fallocate(f.file, mode, offset, len);
333 fdput(f);
334 }
335 return error;
336}
337
338SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
339{
340 return ksys_fallocate(fd, mode, offset, len);
341}
342
343/*
344 * access() needs to use the real uid/gid, not the effective uid/gid.
345 * We do this by temporarily clearing all FS-related capabilities and
346 * switching the fsuid/fsgid around to the real ones.
347 */
348long do_faccessat(int dfd, const char __user *filename, int mode)
349{
350 const struct cred *old_cred;
351 struct cred *override_cred;
352 struct path path;
353 struct inode *inode;
354 int res;
355 unsigned int lookup_flags = LOOKUP_FOLLOW;
356
357 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
358 return -EINVAL;
359
360 override_cred = prepare_creds();
361 if (!override_cred)
362 return -ENOMEM;
363
364 override_cred->fsuid = override_cred->uid;
365 override_cred->fsgid = override_cred->gid;
366
367 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
368 /* Clear the capabilities if we switch to a non-root user */
369 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
370 if (!uid_eq(override_cred->uid, root_uid))
371 cap_clear(override_cred->cap_effective);
372 else
373 override_cred->cap_effective =
374 override_cred->cap_permitted;
375 }
376
377 /*
378 * The new set of credentials can *only* be used in
379 * task-synchronous circumstances, and does not need
380 * RCU freeing, unless somebody then takes a separate
381 * reference to it.
382 *
383 * NOTE! This is _only_ true because this credential
384 * is used purely for override_creds() that installs
385 * it as the subjective cred. Other threads will be
386 * accessing ->real_cred, not the subjective cred.
387 *
388 * If somebody _does_ make a copy of this (using the
389 * 'get_current_cred()' function), that will clear the
390 * non_rcu field, because now that other user may be
391 * expecting RCU freeing. But normal thread-synchronous
392 * cred accesses will keep things non-RCY.
393 */
394 override_cred->non_rcu = 1;
395
396 old_cred = override_creds(override_cred);
397retry:
398 res = user_path_at(dfd, filename, lookup_flags, &path);
399 if (res)
400 goto out;
401
402 inode = d_backing_inode(path.dentry);
403
404 if ((mode & MAY_EXEC) && S_ISREG(inode->i_mode)) {
405 /*
406 * MAY_EXEC on regular files is denied if the fs is mounted
407 * with the "noexec" flag.
408 */
409 res = -EACCES;
410 if (path_noexec(&path))
411 goto out_path_release;
412 }
413
414 res = inode_permission(inode, mode | MAY_ACCESS);
415 /* SuS v2 requires we report a read only fs too */
416 if (res || !(mode & S_IWOTH) || special_file(inode->i_mode))
417 goto out_path_release;
418 /*
419 * This is a rare case where using __mnt_is_readonly()
420 * is OK without a mnt_want/drop_write() pair. Since
421 * no actual write to the fs is performed here, we do
422 * not need to telegraph to that to anyone.
423 *
424 * By doing this, we accept that this access is
425 * inherently racy and know that the fs may change
426 * state before we even see this result.
427 */
428 if (__mnt_is_readonly(path.mnt))
429 res = -EROFS;
430
431out_path_release:
432 path_put(&path);
433 if (retry_estale(res, lookup_flags)) {
434 lookup_flags |= LOOKUP_REVAL;
435 goto retry;
436 }
437out:
438 revert_creds(old_cred);
439 put_cred(override_cred);
440 return res;
441}
442
443SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
444{
445 return do_faccessat(dfd, filename, mode);
446}
447
448SYSCALL_DEFINE2(access, const char __user *, filename, int, mode)
449{
450 return do_faccessat(AT_FDCWD, filename, mode);
451}
452
453int ksys_chdir(const char __user *filename)
454{
455 struct path path;
456 int error;
457 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
458retry:
459 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
460 if (error)
461 goto out;
462
463 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
464 if (error)
465 goto dput_and_out;
466
467 set_fs_pwd(current->fs, &path);
468
469dput_and_out:
470 path_put(&path);
471 if (retry_estale(error, lookup_flags)) {
472 lookup_flags |= LOOKUP_REVAL;
473 goto retry;
474 }
475out:
476 return error;
477}
478
479SYSCALL_DEFINE1(chdir, const char __user *, filename)
480{
481 return ksys_chdir(filename);
482}
483
484SYSCALL_DEFINE1(fchdir, unsigned int, fd)
485{
486 struct fd f = fdget_raw(fd);
487 int error;
488
489 error = -EBADF;
490 if (!f.file)
491 goto out;
492
493 error = -ENOTDIR;
494 if (!d_can_lookup(f.file->f_path.dentry))
495 goto out_putf;
496
497 error = inode_permission(file_inode(f.file), MAY_EXEC | MAY_CHDIR);
498 if (!error)
499 set_fs_pwd(current->fs, &f.file->f_path);
500out_putf:
501 fdput(f);
502out:
503 return error;
504}
505
506int ksys_chroot(const char __user *filename)
507{
508 struct path path;
509 int error;
510 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
511retry:
512 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
513 if (error)
514 goto out;
515
516 error = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_CHDIR);
517 if (error)
518 goto dput_and_out;
519
520 error = -EPERM;
521 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
522 goto dput_and_out;
523 error = security_path_chroot(&path);
524 if (error)
525 goto dput_and_out;
526
527 set_fs_root(current->fs, &path);
528 error = 0;
529dput_and_out:
530 path_put(&path);
531 if (retry_estale(error, lookup_flags)) {
532 lookup_flags |= LOOKUP_REVAL;
533 goto retry;
534 }
535out:
536 return error;
537}
538
539SYSCALL_DEFINE1(chroot, const char __user *, filename)
540{
541 return ksys_chroot(filename);
542}
543
544static int chmod_common(const struct path *path, umode_t mode)
545{
546 struct inode *inode = path->dentry->d_inode;
547 struct inode *delegated_inode = NULL;
548 struct iattr newattrs;
549 int error;
550
551 error = mnt_want_write(path->mnt);
552 if (error)
553 return error;
554retry_deleg:
555 inode_lock(inode);
556 error = security_path_chmod(path, mode);
557 if (error)
558 goto out_unlock;
559 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
560 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
561 error = notify_change(path->dentry, &newattrs, &delegated_inode);
562out_unlock:
563 inode_unlock(inode);
564 if (delegated_inode) {
565 error = break_deleg_wait(&delegated_inode);
566 if (!error)
567 goto retry_deleg;
568 }
569 mnt_drop_write(path->mnt);
570 return error;
571}
572
573int ksys_fchmod(unsigned int fd, umode_t mode)
574{
575 struct fd f = fdget(fd);
576 int err = -EBADF;
577
578 if (f.file) {
579 audit_file(f.file);
580 err = chmod_common(&f.file->f_path, mode);
581 fdput(f);
582 }
583 return err;
584}
585
586SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
587{
588 return ksys_fchmod(fd, mode);
589}
590
591int do_fchmodat(int dfd, const char __user *filename, umode_t mode)
592{
593 struct path path;
594 int error;
595 unsigned int lookup_flags = LOOKUP_FOLLOW;
596retry:
597 error = user_path_at(dfd, filename, lookup_flags, &path);
598 if (!error) {
599 error = chmod_common(&path, mode);
600 path_put(&path);
601 if (retry_estale(error, lookup_flags)) {
602 lookup_flags |= LOOKUP_REVAL;
603 goto retry;
604 }
605 }
606 return error;
607}
608
609SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
610 umode_t, mode)
611{
612 return do_fchmodat(dfd, filename, mode);
613}
614
615SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
616{
617 return do_fchmodat(AT_FDCWD, filename, mode);
618}
619
620static int chown_common(const struct path *path, uid_t user, gid_t group)
621{
622 struct inode *inode = path->dentry->d_inode;
623 struct inode *delegated_inode = NULL;
624 int error;
625 struct iattr newattrs;
626 kuid_t uid;
627 kgid_t gid;
628
629 uid = make_kuid(current_user_ns(), user);
630 gid = make_kgid(current_user_ns(), group);
631
632retry_deleg:
633 newattrs.ia_valid = ATTR_CTIME;
634 if (user != (uid_t) -1) {
635 if (!uid_valid(uid))
636 return -EINVAL;
637 newattrs.ia_valid |= ATTR_UID;
638 newattrs.ia_uid = uid;
639 }
640 if (group != (gid_t) -1) {
641 if (!gid_valid(gid))
642 return -EINVAL;
643 newattrs.ia_valid |= ATTR_GID;
644 newattrs.ia_gid = gid;
645 }
646 if (!S_ISDIR(inode->i_mode))
647 newattrs.ia_valid |=
648 ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
649 inode_lock(inode);
650 error = security_path_chown(path, uid, gid);
651 if (!error)
652 error = notify_change(path->dentry, &newattrs, &delegated_inode);
653 inode_unlock(inode);
654 if (delegated_inode) {
655 error = break_deleg_wait(&delegated_inode);
656 if (!error)
657 goto retry_deleg;
658 }
659 return error;
660}
661
662int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
663 int flag)
664{
665 struct path path;
666 int error = -EINVAL;
667 int lookup_flags;
668
669 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
670 goto out;
671
672 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
673 if (flag & AT_EMPTY_PATH)
674 lookup_flags |= LOOKUP_EMPTY;
675retry:
676 error = user_path_at(dfd, filename, lookup_flags, &path);
677 if (error)
678 goto out;
679 error = mnt_want_write(path.mnt);
680 if (error)
681 goto out_release;
682 error = chown_common(&path, user, group);
683 mnt_drop_write(path.mnt);
684out_release:
685 path_put(&path);
686 if (retry_estale(error, lookup_flags)) {
687 lookup_flags |= LOOKUP_REVAL;
688 goto retry;
689 }
690out:
691 return error;
692}
693
694SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
695 gid_t, group, int, flag)
696{
697 return do_fchownat(dfd, filename, user, group, flag);
698}
699
700SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
701{
702 return do_fchownat(AT_FDCWD, filename, user, group, 0);
703}
704
705SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
706{
707 return do_fchownat(AT_FDCWD, filename, user, group,
708 AT_SYMLINK_NOFOLLOW);
709}
710
711int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
712{
713 struct fd f = fdget(fd);
714 int error = -EBADF;
715
716 if (!f.file)
717 goto out;
718
719 error = mnt_want_write_file(f.file);
720 if (error)
721 goto out_fput;
722 audit_file(f.file);
723 error = chown_common(&f.file->f_path, user, group);
724 mnt_drop_write_file(f.file);
725out_fput:
726 fdput(f);
727out:
728 return error;
729}
730
731SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
732{
733 return ksys_fchown(fd, user, group);
734}
735
736static int do_dentry_open(struct file *f,
737 struct inode *inode,
738 int (*open)(struct inode *, struct file *))
739{
740 static const struct file_operations empty_fops = {};
741 int error;
742
743 path_get(&f->f_path);
744 f->f_inode = inode;
745 f->f_mapping = inode->i_mapping;
746
747 /* Ensure that we skip any errors that predate opening of the file */
748 f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
749
750 if (unlikely(f->f_flags & O_PATH)) {
751 f->f_mode = FMODE_PATH | FMODE_OPENED;
752 f->f_op = &empty_fops;
753 return 0;
754 }
755
756 /* Any file opened for execve()/uselib() has to be a regular file. */
757 if (unlikely(f->f_flags & FMODE_EXEC && !S_ISREG(inode->i_mode))) {
758 error = -EACCES;
759 goto cleanup_file;
760 }
761
762 if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
763 error = get_write_access(inode);
764 if (unlikely(error))
765 goto cleanup_file;
766 error = __mnt_want_write(f->f_path.mnt);
767 if (unlikely(error)) {
768 put_write_access(inode);
769 goto cleanup_file;
770 }
771 f->f_mode |= FMODE_WRITER;
772 }
773
774 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
775 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
776 f->f_mode |= FMODE_ATOMIC_POS;
777
778 f->f_op = fops_get(inode->i_fop);
779 if (WARN_ON(!f->f_op)) {
780 error = -ENODEV;
781 goto cleanup_all;
782 }
783
784 error = security_file_open(f);
785 if (error)
786 goto cleanup_all;
787
788 error = break_lease(locks_inode(f), f->f_flags);
789 if (error)
790 goto cleanup_all;
791
792 /* normally all 3 are set; ->open() can clear them if needed */
793 f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
794 if (!open)
795 open = f->f_op->open;
796 if (open) {
797 error = open(inode, f);
798 if (error)
799 goto cleanup_all;
800 }
801 f->f_mode |= FMODE_OPENED;
802 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
803 i_readcount_inc(inode);
804 if ((f->f_mode & FMODE_READ) &&
805 likely(f->f_op->read || f->f_op->read_iter))
806 f->f_mode |= FMODE_CAN_READ;
807 if ((f->f_mode & FMODE_WRITE) &&
808 likely(f->f_op->write || f->f_op->write_iter))
809 f->f_mode |= FMODE_CAN_WRITE;
810
811 f->f_write_hint = WRITE_LIFE_NOT_SET;
812 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
813
814 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
815
816 /* NB: we're sure to have correct a_ops only after f_op->open */
817 if (f->f_flags & O_DIRECT) {
818 if (!f->f_mapping->a_ops || !f->f_mapping->a_ops->direct_IO)
819 return -EINVAL;
820 }
821
822 /*
823 * XXX: Huge page cache doesn't support writing yet. Drop all page
824 * cache for this file before processing writes.
825 */
826 if ((f->f_mode & FMODE_WRITE) && filemap_nr_thps(inode->i_mapping))
827 truncate_pagecache(inode, 0);
828
829 return 0;
830
831cleanup_all:
832 if (WARN_ON_ONCE(error > 0))
833 error = -EINVAL;
834 fops_put(f->f_op);
835 if (f->f_mode & FMODE_WRITER) {
836 put_write_access(inode);
837 __mnt_drop_write(f->f_path.mnt);
838 }
839cleanup_file:
840 path_put(&f->f_path);
841 f->f_path.mnt = NULL;
842 f->f_path.dentry = NULL;
843 f->f_inode = NULL;
844 return error;
845}
846
847/**
848 * finish_open - finish opening a file
849 * @file: file pointer
850 * @dentry: pointer to dentry
851 * @open: open callback
852 * @opened: state of open
853 *
854 * This can be used to finish opening a file passed to i_op->atomic_open().
855 *
856 * If the open callback is set to NULL, then the standard f_op->open()
857 * filesystem callback is substituted.
858 *
859 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
860 * the return value of d_splice_alias(), then the caller needs to perform dput()
861 * on it after finish_open().
862 *
863 * On successful return @file is a fully instantiated open file. After this, if
864 * an error occurs in ->atomic_open(), it needs to clean up with fput().
865 *
866 * Returns zero on success or -errno if the open failed.
867 */
868int finish_open(struct file *file, struct dentry *dentry,
869 int (*open)(struct inode *, struct file *))
870{
871 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
872
873 file->f_path.dentry = dentry;
874 return do_dentry_open(file, d_backing_inode(dentry), open);
875}
876EXPORT_SYMBOL(finish_open);
877
878/**
879 * finish_no_open - finish ->atomic_open() without opening the file
880 *
881 * @file: file pointer
882 * @dentry: dentry or NULL (as returned from ->lookup())
883 *
884 * This can be used to set the result of a successful lookup in ->atomic_open().
885 *
886 * NB: unlike finish_open() this function does consume the dentry reference and
887 * the caller need not dput() it.
888 *
889 * Returns "0" which must be the return value of ->atomic_open() after having
890 * called this function.
891 */
892int finish_no_open(struct file *file, struct dentry *dentry)
893{
894 file->f_path.dentry = dentry;
895 return 0;
896}
897EXPORT_SYMBOL(finish_no_open);
898
899char *file_path(struct file *filp, char *buf, int buflen)
900{
901 return d_path(&filp->f_path, buf, buflen);
902}
903EXPORT_SYMBOL(file_path);
904
905/**
906 * vfs_open - open the file at the given path
907 * @path: path to open
908 * @file: newly allocated file with f_flag initialized
909 * @cred: credentials to use
910 */
911int vfs_open(const struct path *path, struct file *file)
912{
913 file->f_path = *path;
914 return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
915}
916
917struct file *dentry_open(const struct path *path, int flags,
918 const struct cred *cred)
919{
920 int error;
921 struct file *f;
922
923 validate_creds(cred);
924
925 /* We must always pass in a valid mount pointer. */
926 BUG_ON(!path->mnt);
927
928 f = alloc_empty_file(flags, cred);
929 if (!IS_ERR(f)) {
930 error = vfs_open(path, f);
931 if (error) {
932 fput(f);
933 f = ERR_PTR(error);
934 }
935 }
936 return f;
937}
938EXPORT_SYMBOL(dentry_open);
939
940struct file *open_with_fake_path(const struct path *path, int flags,
941 struct inode *inode, const struct cred *cred)
942{
943 struct file *f = alloc_empty_file_noaccount(flags, cred);
944 if (!IS_ERR(f)) {
945 int error;
946
947 f->f_path = *path;
948 error = do_dentry_open(f, inode, NULL);
949 if (error) {
950 fput(f);
951 f = ERR_PTR(error);
952 }
953 }
954 return f;
955}
956EXPORT_SYMBOL(open_with_fake_path);
957
958static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
959{
960 int lookup_flags = 0;
961 int acc_mode = ACC_MODE(flags);
962
963 /*
964 * Clear out all open flags we don't know about so that we don't report
965 * them in fcntl(F_GETFD) or similar interfaces.
966 */
967 flags &= VALID_OPEN_FLAGS;
968
969 if (flags & (O_CREAT | __O_TMPFILE))
970 op->mode = (mode & S_IALLUGO) | S_IFREG;
971 else
972 op->mode = 0;
973
974 /* Must never be set by userspace */
975 flags &= ~FMODE_NONOTIFY & ~O_CLOEXEC;
976
977 /*
978 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
979 * check for O_DSYNC if the need any syncing at all we enforce it's
980 * always set instead of having to deal with possibly weird behaviour
981 * for malicious applications setting only __O_SYNC.
982 */
983 if (flags & __O_SYNC)
984 flags |= O_DSYNC;
985
986 if (flags & __O_TMPFILE) {
987 if ((flags & O_TMPFILE_MASK) != O_TMPFILE)
988 return -EINVAL;
989 if (!(acc_mode & MAY_WRITE))
990 return -EINVAL;
991 } else if (flags & O_PATH) {
992 /*
993 * If we have O_PATH in the open flag. Then we
994 * cannot have anything other than the below set of flags
995 */
996 flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH;
997 acc_mode = 0;
998 }
999
1000 op->open_flag = flags;
1001
1002 /* O_TRUNC implies we need access checks for write permissions */
1003 if (flags & O_TRUNC)
1004 acc_mode |= MAY_WRITE;
1005
1006 /* Allow the LSM permission hook to distinguish append
1007 access from general write access. */
1008 if (flags & O_APPEND)
1009 acc_mode |= MAY_APPEND;
1010
1011 op->acc_mode = acc_mode;
1012
1013 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1014
1015 if (flags & O_CREAT) {
1016 op->intent |= LOOKUP_CREATE;
1017 if (flags & O_EXCL)
1018 op->intent |= LOOKUP_EXCL;
1019 }
1020
1021 if (flags & O_DIRECTORY)
1022 lookup_flags |= LOOKUP_DIRECTORY;
1023 if (!(flags & O_NOFOLLOW))
1024 lookup_flags |= LOOKUP_FOLLOW;
1025 op->lookup_flags = lookup_flags;
1026 return 0;
1027}
1028
1029/**
1030 * file_open_name - open file and return file pointer
1031 *
1032 * @name: struct filename containing path to open
1033 * @flags: open flags as per the open(2) second argument
1034 * @mode: mode for the new file if O_CREAT is set, else ignored
1035 *
1036 * This is the helper to open a file from kernelspace if you really
1037 * have to. But in generally you should not do this, so please move
1038 * along, nothing to see here..
1039 */
1040struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1041{
1042 struct open_flags op;
1043 int err = build_open_flags(flags, mode, &op);
1044 return err ? ERR_PTR(err) : do_filp_open(AT_FDCWD, name, &op);
1045}
1046
1047/**
1048 * filp_open - open file and return file pointer
1049 *
1050 * @filename: path to open
1051 * @flags: open flags as per the open(2) second argument
1052 * @mode: mode for the new file if O_CREAT is set, else ignored
1053 *
1054 * This is the helper to open a file from kernelspace if you really
1055 * have to. But in generally you should not do this, so please move
1056 * along, nothing to see here..
1057 */
1058struct file *filp_open(const char *filename, int flags, umode_t mode)
1059{
1060 struct filename *name = getname_kernel(filename);
1061 struct file *file = ERR_CAST(name);
1062
1063 if (!IS_ERR(name)) {
1064 file = file_open_name(name, flags, mode);
1065 putname(name);
1066 }
1067 return file;
1068}
1069EXPORT_SYMBOL(filp_open);
1070
1071struct file *file_open_root(struct dentry *dentry, struct vfsmount *mnt,
1072 const char *filename, int flags, umode_t mode)
1073{
1074 struct open_flags op;
1075 int err = build_open_flags(flags, mode, &op);
1076 if (err)
1077 return ERR_PTR(err);
1078 return do_file_open_root(dentry, mnt, filename, &op);
1079}
1080EXPORT_SYMBOL(file_open_root);
1081
1082long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1083{
1084 struct open_flags op;
1085 int fd = build_open_flags(flags, mode, &op);
1086 struct filename *tmp;
1087
1088 if (fd)
1089 return fd;
1090
1091 tmp = getname(filename);
1092 if (IS_ERR(tmp))
1093 return PTR_ERR(tmp);
1094
1095 fd = get_unused_fd_flags(flags);
1096 if (fd >= 0) {
1097 struct file *f = do_filp_open(dfd, tmp, &op);
1098 if (IS_ERR(f)) {
1099 put_unused_fd(fd);
1100 fd = PTR_ERR(f);
1101 } else {
1102 fsnotify_open(f);
1103 fd_install(fd, f);
1104 }
1105 }
1106 putname(tmp);
1107 return fd;
1108}
1109
1110SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1111{
1112 if (force_o_largefile())
1113 flags |= O_LARGEFILE;
1114
1115 return do_sys_open(AT_FDCWD, filename, flags, mode);
1116}
1117
1118SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1119 umode_t, mode)
1120{
1121 if (force_o_largefile())
1122 flags |= O_LARGEFILE;
1123
1124 return do_sys_open(dfd, filename, flags, mode);
1125}
1126
1127#ifdef CONFIG_COMPAT
1128/*
1129 * Exactly like sys_open(), except that it doesn't set the
1130 * O_LARGEFILE flag.
1131 */
1132COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1133{
1134 return do_sys_open(AT_FDCWD, filename, flags, mode);
1135}
1136
1137/*
1138 * Exactly like sys_openat(), except that it doesn't set the
1139 * O_LARGEFILE flag.
1140 */
1141COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1142{
1143 return do_sys_open(dfd, filename, flags, mode);
1144}
1145#endif
1146
1147#ifndef __alpha__
1148
1149/*
1150 * For backward compatibility? Maybe this should be moved
1151 * into arch/i386 instead?
1152 */
1153SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1154{
1155 return ksys_open(pathname, O_CREAT | O_WRONLY | O_TRUNC, mode);
1156}
1157
1158#endif
1159
1160/*
1161 * "id" is the POSIX thread ID. We use the
1162 * files pointer for this..
1163 */
1164int filp_close(struct file *filp, fl_owner_t id)
1165{
1166 int retval = 0;
1167
1168 if (!file_count(filp)) {
1169 printk(KERN_ERR "VFS: Close: file count is 0\n");
1170 return 0;
1171 }
1172
1173 if (filp->f_op->flush)
1174 retval = filp->f_op->flush(filp, id);
1175
1176 if (likely(!(filp->f_mode & FMODE_PATH))) {
1177 dnotify_flush(filp, id);
1178 locks_remove_posix(filp, id);
1179 }
1180 fput(filp);
1181 return retval;
1182}
1183
1184EXPORT_SYMBOL(filp_close);
1185
1186/*
1187 * Careful here! We test whether the file pointer is NULL before
1188 * releasing the fd. This ensures that one clone task can't release
1189 * an fd while another clone is opening it.
1190 */
1191SYSCALL_DEFINE1(close, unsigned int, fd)
1192{
1193 int retval = __close_fd(current->files, fd);
1194
1195 /* can't restart close syscall because file table entry was cleared */
1196 if (unlikely(retval == -ERESTARTSYS ||
1197 retval == -ERESTARTNOINTR ||
1198 retval == -ERESTARTNOHAND ||
1199 retval == -ERESTART_RESTARTBLOCK))
1200 retval = -EINTR;
1201
1202 return retval;
1203}
1204
1205/*
1206 * This routine simulates a hangup on the tty, to arrange that users
1207 * are given clean terminals at login time.
1208 */
1209SYSCALL_DEFINE0(vhangup)
1210{
1211 if (capable(CAP_SYS_TTY_CONFIG)) {
1212 tty_vhangup_self();
1213 return 0;
1214 }
1215 return -EPERM;
1216}
1217
1218/*
1219 * Called when an inode is about to be open.
1220 * We use this to disallow opening large files on 32bit systems if
1221 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1222 * on this flag in sys_open.
1223 */
1224int generic_file_open(struct inode * inode, struct file * filp)
1225{
1226 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1227 return -EOVERFLOW;
1228 return 0;
1229}
1230
1231EXPORT_SYMBOL(generic_file_open);
1232
1233/*
1234 * This is used by subsystems that don't want seekable
1235 * file descriptors. The function is not supposed to ever fail, the only
1236 * reason it returns an 'int' and not 'void' is so that it can be plugged
1237 * directly into file_operations structure.
1238 */
1239int nonseekable_open(struct inode *inode, struct file *filp)
1240{
1241 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1242 return 0;
1243}
1244
1245EXPORT_SYMBOL(nonseekable_open);
1246
1247/*
1248 * stream_open is used by subsystems that want stream-like file descriptors.
1249 * Such file descriptors are not seekable and don't have notion of position
1250 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1251 * Contrary to file descriptors of other regular files, .read() and .write()
1252 * can run simultaneously.
1253 *
1254 * stream_open never fails and is marked to return int so that it could be
1255 * directly used as file_operations.open .
1256 */
1257int stream_open(struct inode *inode, struct file *filp)
1258{
1259 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1260 filp->f_mode |= FMODE_STREAM;
1261 return 0;
1262}
1263
1264EXPORT_SYMBOL(stream_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 struct fd f;
191 int error;
192
193 if (length < 0)
194 return -EINVAL;
195 f = fdget(fd);
196 if (!f.file)
197 return -EBADF;
198
199 error = do_ftruncate(f.file, length, small);
200
201 fdput(f);
202 return error;
203}
204
205SYSCALL_DEFINE2(ftruncate, unsigned int, fd, unsigned long, length)
206{
207 return do_sys_ftruncate(fd, length, 1);
208}
209
210#ifdef CONFIG_COMPAT
211COMPAT_SYSCALL_DEFINE2(ftruncate, unsigned int, fd, compat_ulong_t, length)
212{
213 return do_sys_ftruncate(fd, length, 1);
214}
215#endif
216
217/* LFS versions of truncate are only needed on 32 bit machines */
218#if BITS_PER_LONG == 32
219SYSCALL_DEFINE2(truncate64, const char __user *, path, loff_t, length)
220{
221 return do_sys_truncate(path, length);
222}
223
224SYSCALL_DEFINE2(ftruncate64, unsigned int, fd, loff_t, length)
225{
226 return do_sys_ftruncate(fd, length, 0);
227}
228#endif /* BITS_PER_LONG == 32 */
229
230#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_TRUNCATE64)
231COMPAT_SYSCALL_DEFINE3(truncate64, const char __user *, pathname,
232 compat_arg_u64_dual(length))
233{
234 return ksys_truncate(pathname, compat_arg_u64_glue(length));
235}
236#endif
237
238#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FTRUNCATE64)
239COMPAT_SYSCALL_DEFINE3(ftruncate64, unsigned int, fd,
240 compat_arg_u64_dual(length))
241{
242 return ksys_ftruncate(fd, compat_arg_u64_glue(length));
243}
244#endif
245
246int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
247{
248 struct inode *inode = file_inode(file);
249 long ret;
250
251 if (offset < 0 || len <= 0)
252 return -EINVAL;
253
254 /* Return error if mode is not supported */
255 if (mode & ~FALLOC_FL_SUPPORTED_MASK)
256 return -EOPNOTSUPP;
257
258 /* Punch hole and zero range are mutually exclusive */
259 if ((mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) ==
260 (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))
261 return -EOPNOTSUPP;
262
263 /* Punch hole must have keep size set */
264 if ((mode & FALLOC_FL_PUNCH_HOLE) &&
265 !(mode & FALLOC_FL_KEEP_SIZE))
266 return -EOPNOTSUPP;
267
268 /* Collapse range should only be used exclusively. */
269 if ((mode & FALLOC_FL_COLLAPSE_RANGE) &&
270 (mode & ~FALLOC_FL_COLLAPSE_RANGE))
271 return -EINVAL;
272
273 /* Insert range should only be used exclusively. */
274 if ((mode & FALLOC_FL_INSERT_RANGE) &&
275 (mode & ~FALLOC_FL_INSERT_RANGE))
276 return -EINVAL;
277
278 /* Unshare range should only be used with allocate mode. */
279 if ((mode & FALLOC_FL_UNSHARE_RANGE) &&
280 (mode & ~(FALLOC_FL_UNSHARE_RANGE | FALLOC_FL_KEEP_SIZE)))
281 return -EINVAL;
282
283 if (!(file->f_mode & FMODE_WRITE))
284 return -EBADF;
285
286 /*
287 * We can only allow pure fallocate on append only files
288 */
289 if ((mode & ~FALLOC_FL_KEEP_SIZE) && IS_APPEND(inode))
290 return -EPERM;
291
292 if (IS_IMMUTABLE(inode))
293 return -EPERM;
294
295 /*
296 * We cannot allow any fallocate operation on an active swapfile
297 */
298 if (IS_SWAPFILE(inode))
299 return -ETXTBSY;
300
301 /*
302 * Revalidate the write permissions, in case security policy has
303 * changed since the files were opened.
304 */
305 ret = security_file_permission(file, MAY_WRITE);
306 if (ret)
307 return ret;
308
309 ret = fsnotify_file_area_perm(file, MAY_WRITE, &offset, len);
310 if (ret)
311 return ret;
312
313 if (S_ISFIFO(inode->i_mode))
314 return -ESPIPE;
315
316 if (S_ISDIR(inode->i_mode))
317 return -EISDIR;
318
319 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
320 return -ENODEV;
321
322 /* Check for wrap through zero too */
323 if (((offset + len) > inode->i_sb->s_maxbytes) || ((offset + len) < 0))
324 return -EFBIG;
325
326 if (!file->f_op->fallocate)
327 return -EOPNOTSUPP;
328
329 file_start_write(file);
330 ret = file->f_op->fallocate(file, mode, offset, len);
331
332 /*
333 * Create inotify and fanotify events.
334 *
335 * To keep the logic simple always create events if fallocate succeeds.
336 * This implies that events are even created if the file size remains
337 * unchanged, e.g. when using flag FALLOC_FL_KEEP_SIZE.
338 */
339 if (ret == 0)
340 fsnotify_modify(file);
341
342 file_end_write(file);
343 return ret;
344}
345EXPORT_SYMBOL_GPL(vfs_fallocate);
346
347int ksys_fallocate(int fd, int mode, loff_t offset, loff_t len)
348{
349 struct fd f = fdget(fd);
350 int error = -EBADF;
351
352 if (f.file) {
353 error = vfs_fallocate(f.file, mode, offset, len);
354 fdput(f);
355 }
356 return error;
357}
358
359SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
360{
361 return ksys_fallocate(fd, mode, offset, len);
362}
363
364#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_FALLOCATE)
365COMPAT_SYSCALL_DEFINE6(fallocate, int, fd, int, mode, compat_arg_u64_dual(offset),
366 compat_arg_u64_dual(len))
367{
368 return ksys_fallocate(fd, mode, compat_arg_u64_glue(offset),
369 compat_arg_u64_glue(len));
370}
371#endif
372
373/*
374 * access() needs to use the real uid/gid, not the effective uid/gid.
375 * We do this by temporarily clearing all FS-related capabilities and
376 * switching the fsuid/fsgid around to the real ones.
377 *
378 * Creating new credentials is expensive, so we try to skip doing it,
379 * which we can if the result would match what we already got.
380 */
381static bool access_need_override_creds(int flags)
382{
383 const struct cred *cred;
384
385 if (flags & AT_EACCESS)
386 return false;
387
388 cred = current_cred();
389 if (!uid_eq(cred->fsuid, cred->uid) ||
390 !gid_eq(cred->fsgid, cred->gid))
391 return true;
392
393 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
394 kuid_t root_uid = make_kuid(cred->user_ns, 0);
395 if (!uid_eq(cred->uid, root_uid)) {
396 if (!cap_isclear(cred->cap_effective))
397 return true;
398 } else {
399 if (!cap_isidentical(cred->cap_effective,
400 cred->cap_permitted))
401 return true;
402 }
403 }
404
405 return false;
406}
407
408static const struct cred *access_override_creds(void)
409{
410 const struct cred *old_cred;
411 struct cred *override_cred;
412
413 override_cred = prepare_creds();
414 if (!override_cred)
415 return NULL;
416
417 /*
418 * XXX access_need_override_creds performs checks in hopes of skipping
419 * this work. Make sure it stays in sync if making any changes in this
420 * routine.
421 */
422
423 override_cred->fsuid = override_cred->uid;
424 override_cred->fsgid = override_cred->gid;
425
426 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
427 /* Clear the capabilities if we switch to a non-root user */
428 kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
429 if (!uid_eq(override_cred->uid, root_uid))
430 cap_clear(override_cred->cap_effective);
431 else
432 override_cred->cap_effective =
433 override_cred->cap_permitted;
434 }
435
436 /*
437 * The new set of credentials can *only* be used in
438 * task-synchronous circumstances, and does not need
439 * RCU freeing, unless somebody then takes a separate
440 * reference to it.
441 *
442 * NOTE! This is _only_ true because this credential
443 * is used purely for override_creds() that installs
444 * it as the subjective cred. Other threads will be
445 * accessing ->real_cred, not the subjective cred.
446 *
447 * If somebody _does_ make a copy of this (using the
448 * 'get_current_cred()' function), that will clear the
449 * non_rcu field, because now that other user may be
450 * expecting RCU freeing. But normal thread-synchronous
451 * cred accesses will keep things non-racy to avoid RCU
452 * freeing.
453 */
454 override_cred->non_rcu = 1;
455
456 old_cred = override_creds(override_cred);
457
458 /* override_cred() gets its own ref */
459 put_cred(override_cred);
460
461 return old_cred;
462}
463
464static long do_faccessat(int dfd, const char __user *filename, int mode, int flags)
465{
466 struct path path;
467 struct inode *inode;
468 int res;
469 unsigned int lookup_flags = LOOKUP_FOLLOW;
470 const struct cred *old_cred = NULL;
471
472 if (mode & ~S_IRWXO) /* where's F_OK, X_OK, W_OK, R_OK? */
473 return -EINVAL;
474
475 if (flags & ~(AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH))
476 return -EINVAL;
477
478 if (flags & AT_SYMLINK_NOFOLLOW)
479 lookup_flags &= ~LOOKUP_FOLLOW;
480 if (flags & AT_EMPTY_PATH)
481 lookup_flags |= LOOKUP_EMPTY;
482
483 if (access_need_override_creds(flags)) {
484 old_cred = access_override_creds();
485 if (!old_cred)
486 return -ENOMEM;
487 }
488
489retry:
490 res = user_path_at(dfd, filename, lookup_flags, &path);
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 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;
557retry:
558 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
559 if (error)
560 goto out;
561
562 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
563 if (error)
564 goto dput_and_out;
565
566 set_fs_pwd(current->fs, &path);
567
568dput_and_out:
569 path_put(&path);
570 if (retry_estale(error, lookup_flags)) {
571 lookup_flags |= LOOKUP_REVAL;
572 goto retry;
573 }
574out:
575 return error;
576}
577
578SYSCALL_DEFINE1(fchdir, unsigned int, fd)
579{
580 struct fd f = fdget_raw(fd);
581 int error;
582
583 error = -EBADF;
584 if (!f.file)
585 goto out;
586
587 error = -ENOTDIR;
588 if (!d_can_lookup(f.file->f_path.dentry))
589 goto out_putf;
590
591 error = file_permission(f.file, MAY_EXEC | MAY_CHDIR);
592 if (!error)
593 set_fs_pwd(current->fs, &f.file->f_path);
594out_putf:
595 fdput(f);
596out:
597 return error;
598}
599
600SYSCALL_DEFINE1(chroot, const char __user *, filename)
601{
602 struct path path;
603 int error;
604 unsigned int lookup_flags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
605retry:
606 error = user_path_at(AT_FDCWD, filename, lookup_flags, &path);
607 if (error)
608 goto out;
609
610 error = path_permission(&path, MAY_EXEC | MAY_CHDIR);
611 if (error)
612 goto dput_and_out;
613
614 error = -EPERM;
615 if (!ns_capable(current_user_ns(), CAP_SYS_CHROOT))
616 goto dput_and_out;
617 error = security_path_chroot(&path);
618 if (error)
619 goto dput_and_out;
620
621 set_fs_root(current->fs, &path);
622 error = 0;
623dput_and_out:
624 path_put(&path);
625 if (retry_estale(error, lookup_flags)) {
626 lookup_flags |= LOOKUP_REVAL;
627 goto retry;
628 }
629out:
630 return error;
631}
632
633int chmod_common(const struct path *path, umode_t mode)
634{
635 struct inode *inode = path->dentry->d_inode;
636 struct inode *delegated_inode = NULL;
637 struct iattr newattrs;
638 int error;
639
640 error = mnt_want_write(path->mnt);
641 if (error)
642 return error;
643retry_deleg:
644 inode_lock(inode);
645 error = security_path_chmod(path, mode);
646 if (error)
647 goto out_unlock;
648 newattrs.ia_mode = (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
649 newattrs.ia_valid = ATTR_MODE | ATTR_CTIME;
650 error = notify_change(mnt_idmap(path->mnt), path->dentry,
651 &newattrs, &delegated_inode);
652out_unlock:
653 inode_unlock(inode);
654 if (delegated_inode) {
655 error = break_deleg_wait(&delegated_inode);
656 if (!error)
657 goto retry_deleg;
658 }
659 mnt_drop_write(path->mnt);
660 return error;
661}
662
663int vfs_fchmod(struct file *file, umode_t mode)
664{
665 audit_file(file);
666 return chmod_common(&file->f_path, mode);
667}
668
669SYSCALL_DEFINE2(fchmod, unsigned int, fd, umode_t, mode)
670{
671 struct fd f = fdget(fd);
672 int err = -EBADF;
673
674 if (f.file) {
675 err = vfs_fchmod(f.file, mode);
676 fdput(f);
677 }
678 return err;
679}
680
681static int do_fchmodat(int dfd, const char __user *filename, umode_t mode,
682 unsigned int flags)
683{
684 struct path path;
685 int error;
686 unsigned int lookup_flags;
687
688 if (unlikely(flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)))
689 return -EINVAL;
690
691 lookup_flags = (flags & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
692 if (flags & AT_EMPTY_PATH)
693 lookup_flags |= LOOKUP_EMPTY;
694
695retry:
696 error = user_path_at(dfd, filename, lookup_flags, &path);
697 if (!error) {
698 error = chmod_common(&path, mode);
699 path_put(&path);
700 if (retry_estale(error, lookup_flags)) {
701 lookup_flags |= LOOKUP_REVAL;
702 goto retry;
703 }
704 }
705 return error;
706}
707
708SYSCALL_DEFINE4(fchmodat2, int, dfd, const char __user *, filename,
709 umode_t, mode, unsigned int, flags)
710{
711 return do_fchmodat(dfd, filename, mode, flags);
712}
713
714SYSCALL_DEFINE3(fchmodat, int, dfd, const char __user *, filename,
715 umode_t, mode)
716{
717 return do_fchmodat(dfd, filename, mode, 0);
718}
719
720SYSCALL_DEFINE2(chmod, const char __user *, filename, umode_t, mode)
721{
722 return do_fchmodat(AT_FDCWD, filename, mode, 0);
723}
724
725/*
726 * Check whether @kuid is valid and if so generate and set vfsuid_t in
727 * ia_vfsuid.
728 *
729 * Return: true if @kuid is valid, false if not.
730 */
731static inline bool setattr_vfsuid(struct iattr *attr, kuid_t kuid)
732{
733 if (!uid_valid(kuid))
734 return false;
735 attr->ia_valid |= ATTR_UID;
736 attr->ia_vfsuid = VFSUIDT_INIT(kuid);
737 return true;
738}
739
740/*
741 * Check whether @kgid is valid and if so generate and set vfsgid_t in
742 * ia_vfsgid.
743 *
744 * Return: true if @kgid is valid, false if not.
745 */
746static inline bool setattr_vfsgid(struct iattr *attr, kgid_t kgid)
747{
748 if (!gid_valid(kgid))
749 return false;
750 attr->ia_valid |= ATTR_GID;
751 attr->ia_vfsgid = VFSGIDT_INIT(kgid);
752 return true;
753}
754
755int chown_common(const struct path *path, uid_t user, gid_t group)
756{
757 struct mnt_idmap *idmap;
758 struct user_namespace *fs_userns;
759 struct inode *inode = path->dentry->d_inode;
760 struct inode *delegated_inode = NULL;
761 int error;
762 struct iattr newattrs;
763 kuid_t uid;
764 kgid_t gid;
765
766 uid = make_kuid(current_user_ns(), user);
767 gid = make_kgid(current_user_ns(), group);
768
769 idmap = mnt_idmap(path->mnt);
770 fs_userns = i_user_ns(inode);
771
772retry_deleg:
773 newattrs.ia_vfsuid = INVALID_VFSUID;
774 newattrs.ia_vfsgid = INVALID_VFSGID;
775 newattrs.ia_valid = ATTR_CTIME;
776 if ((user != (uid_t)-1) && !setattr_vfsuid(&newattrs, uid))
777 return -EINVAL;
778 if ((group != (gid_t)-1) && !setattr_vfsgid(&newattrs, gid))
779 return -EINVAL;
780 inode_lock(inode);
781 if (!S_ISDIR(inode->i_mode))
782 newattrs.ia_valid |= ATTR_KILL_SUID | ATTR_KILL_PRIV |
783 setattr_should_drop_sgid(idmap, inode);
784 /* Continue to send actual fs values, not the mount values. */
785 error = security_path_chown(
786 path,
787 from_vfsuid(idmap, fs_userns, newattrs.ia_vfsuid),
788 from_vfsgid(idmap, fs_userns, newattrs.ia_vfsgid));
789 if (!error)
790 error = notify_change(idmap, path->dentry, &newattrs,
791 &delegated_inode);
792 inode_unlock(inode);
793 if (delegated_inode) {
794 error = break_deleg_wait(&delegated_inode);
795 if (!error)
796 goto retry_deleg;
797 }
798 return error;
799}
800
801int do_fchownat(int dfd, const char __user *filename, uid_t user, gid_t group,
802 int flag)
803{
804 struct path path;
805 int error = -EINVAL;
806 int lookup_flags;
807
808 if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
809 goto out;
810
811 lookup_flags = (flag & AT_SYMLINK_NOFOLLOW) ? 0 : LOOKUP_FOLLOW;
812 if (flag & AT_EMPTY_PATH)
813 lookup_flags |= LOOKUP_EMPTY;
814retry:
815 error = user_path_at(dfd, filename, lookup_flags, &path);
816 if (error)
817 goto out;
818 error = mnt_want_write(path.mnt);
819 if (error)
820 goto out_release;
821 error = chown_common(&path, user, group);
822 mnt_drop_write(path.mnt);
823out_release:
824 path_put(&path);
825 if (retry_estale(error, lookup_flags)) {
826 lookup_flags |= LOOKUP_REVAL;
827 goto retry;
828 }
829out:
830 return error;
831}
832
833SYSCALL_DEFINE5(fchownat, int, dfd, const char __user *, filename, uid_t, user,
834 gid_t, group, int, flag)
835{
836 return do_fchownat(dfd, filename, user, group, flag);
837}
838
839SYSCALL_DEFINE3(chown, const char __user *, filename, uid_t, user, gid_t, group)
840{
841 return do_fchownat(AT_FDCWD, filename, user, group, 0);
842}
843
844SYSCALL_DEFINE3(lchown, const char __user *, filename, uid_t, user, gid_t, group)
845{
846 return do_fchownat(AT_FDCWD, filename, user, group,
847 AT_SYMLINK_NOFOLLOW);
848}
849
850int vfs_fchown(struct file *file, uid_t user, gid_t group)
851{
852 int error;
853
854 error = mnt_want_write_file(file);
855 if (error)
856 return error;
857 audit_file(file);
858 error = chown_common(&file->f_path, user, group);
859 mnt_drop_write_file(file);
860 return error;
861}
862
863int ksys_fchown(unsigned int fd, uid_t user, gid_t group)
864{
865 struct fd f = fdget(fd);
866 int error = -EBADF;
867
868 if (f.file) {
869 error = vfs_fchown(f.file, user, group);
870 fdput(f);
871 }
872 return error;
873}
874
875SYSCALL_DEFINE3(fchown, unsigned int, fd, uid_t, user, gid_t, group)
876{
877 return ksys_fchown(fd, user, group);
878}
879
880static inline int file_get_write_access(struct file *f)
881{
882 int error;
883
884 error = get_write_access(f->f_inode);
885 if (unlikely(error))
886 return error;
887 error = mnt_get_write_access(f->f_path.mnt);
888 if (unlikely(error))
889 goto cleanup_inode;
890 if (unlikely(f->f_mode & FMODE_BACKING)) {
891 error = mnt_get_write_access(backing_file_user_path(f)->mnt);
892 if (unlikely(error))
893 goto cleanup_mnt;
894 }
895 return 0;
896
897cleanup_mnt:
898 mnt_put_write_access(f->f_path.mnt);
899cleanup_inode:
900 put_write_access(f->f_inode);
901 return error;
902}
903
904static int do_dentry_open(struct file *f,
905 struct inode *inode,
906 int (*open)(struct inode *, struct file *))
907{
908 static const struct file_operations empty_fops = {};
909 int error;
910
911 path_get(&f->f_path);
912 f->f_inode = inode;
913 f->f_mapping = inode->i_mapping;
914 f->f_wb_err = filemap_sample_wb_err(f->f_mapping);
915 f->f_sb_err = file_sample_sb_err(f);
916
917 if (unlikely(f->f_flags & O_PATH)) {
918 f->f_mode = FMODE_PATH | FMODE_OPENED;
919 f->f_op = &empty_fops;
920 return 0;
921 }
922
923 if ((f->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) {
924 i_readcount_inc(inode);
925 } else if (f->f_mode & FMODE_WRITE && !special_file(inode->i_mode)) {
926 error = file_get_write_access(f);
927 if (unlikely(error))
928 goto cleanup_file;
929 f->f_mode |= FMODE_WRITER;
930 }
931
932 /* POSIX.1-2008/SUSv4 Section XSI 2.9.7 */
933 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode))
934 f->f_mode |= FMODE_ATOMIC_POS;
935
936 f->f_op = fops_get(inode->i_fop);
937 if (WARN_ON(!f->f_op)) {
938 error = -ENODEV;
939 goto cleanup_all;
940 }
941
942 error = security_file_open(f);
943 if (error)
944 goto cleanup_all;
945
946 error = break_lease(file_inode(f), f->f_flags);
947 if (error)
948 goto cleanup_all;
949
950 /* normally all 3 are set; ->open() can clear them if needed */
951 f->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
952 if (!open)
953 open = f->f_op->open;
954 if (open) {
955 error = open(inode, f);
956 if (error)
957 goto cleanup_all;
958 }
959 f->f_mode |= FMODE_OPENED;
960 if ((f->f_mode & FMODE_READ) &&
961 likely(f->f_op->read || f->f_op->read_iter))
962 f->f_mode |= FMODE_CAN_READ;
963 if ((f->f_mode & FMODE_WRITE) &&
964 likely(f->f_op->write || f->f_op->write_iter))
965 f->f_mode |= FMODE_CAN_WRITE;
966 if ((f->f_mode & FMODE_LSEEK) && !f->f_op->llseek)
967 f->f_mode &= ~FMODE_LSEEK;
968 if (f->f_mapping->a_ops && f->f_mapping->a_ops->direct_IO)
969 f->f_mode |= FMODE_CAN_ODIRECT;
970
971 f->f_flags &= ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
972 f->f_iocb_flags = iocb_flags(f);
973
974 file_ra_state_init(&f->f_ra, f->f_mapping->host->i_mapping);
975
976 if ((f->f_flags & O_DIRECT) && !(f->f_mode & FMODE_CAN_ODIRECT))
977 return -EINVAL;
978
979 /*
980 * XXX: Huge page cache doesn't support writing yet. Drop all page
981 * cache for this file before processing writes.
982 */
983 if (f->f_mode & FMODE_WRITE) {
984 /*
985 * Paired with smp_mb() in collapse_file() to ensure nr_thps
986 * is up to date and the update to i_writecount by
987 * get_write_access() is visible. Ensures subsequent insertion
988 * of THPs into the page cache will fail.
989 */
990 smp_mb();
991 if (filemap_nr_thps(inode->i_mapping)) {
992 struct address_space *mapping = inode->i_mapping;
993
994 filemap_invalidate_lock(inode->i_mapping);
995 /*
996 * unmap_mapping_range just need to be called once
997 * here, because the private pages is not need to be
998 * unmapped mapping (e.g. data segment of dynamic
999 * shared libraries here).
1000 */
1001 unmap_mapping_range(mapping, 0, 0, 0);
1002 truncate_inode_pages(mapping, 0);
1003 filemap_invalidate_unlock(inode->i_mapping);
1004 }
1005 }
1006
1007 /*
1008 * Once we return a file with FMODE_OPENED, __fput() will call
1009 * fsnotify_close(), so we need fsnotify_open() here for symmetry.
1010 */
1011 fsnotify_open(f);
1012 return 0;
1013
1014cleanup_all:
1015 if (WARN_ON_ONCE(error > 0))
1016 error = -EINVAL;
1017 fops_put(f->f_op);
1018 put_file_access(f);
1019cleanup_file:
1020 path_put(&f->f_path);
1021 f->f_path.mnt = NULL;
1022 f->f_path.dentry = NULL;
1023 f->f_inode = NULL;
1024 return error;
1025}
1026
1027/**
1028 * finish_open - finish opening a file
1029 * @file: file pointer
1030 * @dentry: pointer to dentry
1031 * @open: open callback
1032 *
1033 * This can be used to finish opening a file passed to i_op->atomic_open().
1034 *
1035 * If the open callback is set to NULL, then the standard f_op->open()
1036 * filesystem callback is substituted.
1037 *
1038 * NB: the dentry reference is _not_ consumed. If, for example, the dentry is
1039 * the return value of d_splice_alias(), then the caller needs to perform dput()
1040 * on it after finish_open().
1041 *
1042 * Returns zero on success or -errno if the open failed.
1043 */
1044int finish_open(struct file *file, struct dentry *dentry,
1045 int (*open)(struct inode *, struct file *))
1046{
1047 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
1048
1049 file->f_path.dentry = dentry;
1050 return do_dentry_open(file, d_backing_inode(dentry), open);
1051}
1052EXPORT_SYMBOL(finish_open);
1053
1054/**
1055 * finish_no_open - finish ->atomic_open() without opening the file
1056 *
1057 * @file: file pointer
1058 * @dentry: dentry or NULL (as returned from ->lookup())
1059 *
1060 * This can be used to set the result of a successful lookup in ->atomic_open().
1061 *
1062 * NB: unlike finish_open() this function does consume the dentry reference and
1063 * the caller need not dput() it.
1064 *
1065 * Returns "0" which must be the return value of ->atomic_open() after having
1066 * called this function.
1067 */
1068int finish_no_open(struct file *file, struct dentry *dentry)
1069{
1070 file->f_path.dentry = dentry;
1071 return 0;
1072}
1073EXPORT_SYMBOL(finish_no_open);
1074
1075char *file_path(struct file *filp, char *buf, int buflen)
1076{
1077 return d_path(&filp->f_path, buf, buflen);
1078}
1079EXPORT_SYMBOL(file_path);
1080
1081/**
1082 * vfs_open - open the file at the given path
1083 * @path: path to open
1084 * @file: newly allocated file with f_flag initialized
1085 */
1086int vfs_open(const struct path *path, struct file *file)
1087{
1088 file->f_path = *path;
1089 return do_dentry_open(file, d_backing_inode(path->dentry), NULL);
1090}
1091
1092struct file *dentry_open(const struct path *path, int flags,
1093 const struct cred *cred)
1094{
1095 int error;
1096 struct file *f;
1097
1098 /* We must always pass in a valid mount pointer. */
1099 BUG_ON(!path->mnt);
1100
1101 f = alloc_empty_file(flags, cred);
1102 if (!IS_ERR(f)) {
1103 error = vfs_open(path, f);
1104 if (error) {
1105 fput(f);
1106 f = ERR_PTR(error);
1107 }
1108 }
1109 return f;
1110}
1111EXPORT_SYMBOL(dentry_open);
1112
1113/**
1114 * dentry_create - Create and open a file
1115 * @path: path to create
1116 * @flags: O_ flags
1117 * @mode: mode bits for new file
1118 * @cred: credentials to use
1119 *
1120 * Caller must hold the parent directory's lock, and have prepared
1121 * a negative dentry, placed in @path->dentry, for the new file.
1122 *
1123 * Caller sets @path->mnt to the vfsmount of the filesystem where
1124 * the new file is to be created. The parent directory and the
1125 * negative dentry must reside on the same filesystem instance.
1126 *
1127 * On success, returns a "struct file *". Otherwise a ERR_PTR
1128 * is returned.
1129 */
1130struct file *dentry_create(const struct path *path, int flags, umode_t mode,
1131 const struct cred *cred)
1132{
1133 struct file *f;
1134 int error;
1135
1136 f = alloc_empty_file(flags, cred);
1137 if (IS_ERR(f))
1138 return f;
1139
1140 error = vfs_create(mnt_idmap(path->mnt),
1141 d_inode(path->dentry->d_parent),
1142 path->dentry, mode, true);
1143 if (!error)
1144 error = vfs_open(path, f);
1145
1146 if (unlikely(error)) {
1147 fput(f);
1148 return ERR_PTR(error);
1149 }
1150 return f;
1151}
1152EXPORT_SYMBOL(dentry_create);
1153
1154/**
1155 * kernel_file_open - open a file for kernel internal use
1156 * @path: path of the file to open
1157 * @flags: open flags
1158 * @inode: the inode
1159 * @cred: credentials for open
1160 *
1161 * Open a file for use by in-kernel consumers. The file is not accounted
1162 * against nr_files and must not be installed into the file descriptor
1163 * table.
1164 *
1165 * Return: Opened file on success, an error pointer on failure.
1166 */
1167struct file *kernel_file_open(const struct path *path, int flags,
1168 struct inode *inode, const struct cred *cred)
1169{
1170 struct file *f;
1171 int error;
1172
1173 f = alloc_empty_file_noaccount(flags, cred);
1174 if (IS_ERR(f))
1175 return f;
1176
1177 f->f_path = *path;
1178 error = do_dentry_open(f, inode, NULL);
1179 if (error) {
1180 fput(f);
1181 f = ERR_PTR(error);
1182 }
1183 return f;
1184}
1185EXPORT_SYMBOL_GPL(kernel_file_open);
1186
1187#define WILL_CREATE(flags) (flags & (O_CREAT | __O_TMPFILE))
1188#define O_PATH_FLAGS (O_DIRECTORY | O_NOFOLLOW | O_PATH | O_CLOEXEC)
1189
1190inline struct open_how build_open_how(int flags, umode_t mode)
1191{
1192 struct open_how how = {
1193 .flags = flags & VALID_OPEN_FLAGS,
1194 .mode = mode & S_IALLUGO,
1195 };
1196
1197 /* O_PATH beats everything else. */
1198 if (how.flags & O_PATH)
1199 how.flags &= O_PATH_FLAGS;
1200 /* Modes should only be set for create-like flags. */
1201 if (!WILL_CREATE(how.flags))
1202 how.mode = 0;
1203 return how;
1204}
1205
1206inline int build_open_flags(const struct open_how *how, struct open_flags *op)
1207{
1208 u64 flags = how->flags;
1209 u64 strip = __FMODE_NONOTIFY | O_CLOEXEC;
1210 int lookup_flags = 0;
1211 int acc_mode = ACC_MODE(flags);
1212
1213 BUILD_BUG_ON_MSG(upper_32_bits(VALID_OPEN_FLAGS),
1214 "struct open_flags doesn't yet handle flags > 32 bits");
1215
1216 /*
1217 * Strip flags that either shouldn't be set by userspace like
1218 * FMODE_NONOTIFY or that aren't relevant in determining struct
1219 * open_flags like O_CLOEXEC.
1220 */
1221 flags &= ~strip;
1222
1223 /*
1224 * Older syscalls implicitly clear all of the invalid flags or argument
1225 * values before calling build_open_flags(), but openat2(2) checks all
1226 * of its arguments.
1227 */
1228 if (flags & ~VALID_OPEN_FLAGS)
1229 return -EINVAL;
1230 if (how->resolve & ~VALID_RESOLVE_FLAGS)
1231 return -EINVAL;
1232
1233 /* Scoping flags are mutually exclusive. */
1234 if ((how->resolve & RESOLVE_BENEATH) && (how->resolve & RESOLVE_IN_ROOT))
1235 return -EINVAL;
1236
1237 /* Deal with the mode. */
1238 if (WILL_CREATE(flags)) {
1239 if (how->mode & ~S_IALLUGO)
1240 return -EINVAL;
1241 op->mode = how->mode | S_IFREG;
1242 } else {
1243 if (how->mode != 0)
1244 return -EINVAL;
1245 op->mode = 0;
1246 }
1247
1248 /*
1249 * Block bugs where O_DIRECTORY | O_CREAT created regular files.
1250 * Note, that blocking O_DIRECTORY | O_CREAT here also protects
1251 * O_TMPFILE below which requires O_DIRECTORY being raised.
1252 */
1253 if ((flags & (O_DIRECTORY | O_CREAT)) == (O_DIRECTORY | O_CREAT))
1254 return -EINVAL;
1255
1256 /* Now handle the creative implementation of O_TMPFILE. */
1257 if (flags & __O_TMPFILE) {
1258 /*
1259 * In order to ensure programs get explicit errors when trying
1260 * to use O_TMPFILE on old kernels we enforce that O_DIRECTORY
1261 * is raised alongside __O_TMPFILE.
1262 */
1263 if (!(flags & O_DIRECTORY))
1264 return -EINVAL;
1265 if (!(acc_mode & MAY_WRITE))
1266 return -EINVAL;
1267 }
1268 if (flags & O_PATH) {
1269 /* O_PATH only permits certain other flags to be set. */
1270 if (flags & ~O_PATH_FLAGS)
1271 return -EINVAL;
1272 acc_mode = 0;
1273 }
1274
1275 /*
1276 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
1277 * check for O_DSYNC if the need any syncing at all we enforce it's
1278 * always set instead of having to deal with possibly weird behaviour
1279 * for malicious applications setting only __O_SYNC.
1280 */
1281 if (flags & __O_SYNC)
1282 flags |= O_DSYNC;
1283
1284 op->open_flag = flags;
1285
1286 /* O_TRUNC implies we need access checks for write permissions */
1287 if (flags & O_TRUNC)
1288 acc_mode |= MAY_WRITE;
1289
1290 /* Allow the LSM permission hook to distinguish append
1291 access from general write access. */
1292 if (flags & O_APPEND)
1293 acc_mode |= MAY_APPEND;
1294
1295 op->acc_mode = acc_mode;
1296
1297 op->intent = flags & O_PATH ? 0 : LOOKUP_OPEN;
1298
1299 if (flags & O_CREAT) {
1300 op->intent |= LOOKUP_CREATE;
1301 if (flags & O_EXCL) {
1302 op->intent |= LOOKUP_EXCL;
1303 flags |= O_NOFOLLOW;
1304 }
1305 }
1306
1307 if (flags & O_DIRECTORY)
1308 lookup_flags |= LOOKUP_DIRECTORY;
1309 if (!(flags & O_NOFOLLOW))
1310 lookup_flags |= LOOKUP_FOLLOW;
1311
1312 if (how->resolve & RESOLVE_NO_XDEV)
1313 lookup_flags |= LOOKUP_NO_XDEV;
1314 if (how->resolve & RESOLVE_NO_MAGICLINKS)
1315 lookup_flags |= LOOKUP_NO_MAGICLINKS;
1316 if (how->resolve & RESOLVE_NO_SYMLINKS)
1317 lookup_flags |= LOOKUP_NO_SYMLINKS;
1318 if (how->resolve & RESOLVE_BENEATH)
1319 lookup_flags |= LOOKUP_BENEATH;
1320 if (how->resolve & RESOLVE_IN_ROOT)
1321 lookup_flags |= LOOKUP_IN_ROOT;
1322 if (how->resolve & RESOLVE_CACHED) {
1323 /* Don't bother even trying for create/truncate/tmpfile open */
1324 if (flags & (O_TRUNC | O_CREAT | __O_TMPFILE))
1325 return -EAGAIN;
1326 lookup_flags |= LOOKUP_CACHED;
1327 }
1328
1329 op->lookup_flags = lookup_flags;
1330 return 0;
1331}
1332
1333/**
1334 * file_open_name - open file and return file pointer
1335 *
1336 * @name: struct filename containing path to open
1337 * @flags: open flags as per the open(2) second argument
1338 * @mode: mode for the new file if O_CREAT is set, else ignored
1339 *
1340 * This is the helper to open a file from kernelspace if you really
1341 * have to. But in generally you should not do this, so please move
1342 * along, nothing to see here..
1343 */
1344struct file *file_open_name(struct filename *name, int flags, umode_t mode)
1345{
1346 struct open_flags op;
1347 struct open_how how = build_open_how(flags, mode);
1348 int err = build_open_flags(&how, &op);
1349 if (err)
1350 return ERR_PTR(err);
1351 return do_filp_open(AT_FDCWD, name, &op);
1352}
1353
1354/**
1355 * filp_open - open file and return file pointer
1356 *
1357 * @filename: path to open
1358 * @flags: open flags as per the open(2) second argument
1359 * @mode: mode for the new file if O_CREAT is set, else ignored
1360 *
1361 * This is the helper to open a file from kernelspace if you really
1362 * have to. But in generally you should not do this, so please move
1363 * along, nothing to see here..
1364 */
1365struct file *filp_open(const char *filename, int flags, umode_t mode)
1366{
1367 struct filename *name = getname_kernel(filename);
1368 struct file *file = ERR_CAST(name);
1369
1370 if (!IS_ERR(name)) {
1371 file = file_open_name(name, flags, mode);
1372 putname(name);
1373 }
1374 return file;
1375}
1376EXPORT_SYMBOL(filp_open);
1377
1378struct file *file_open_root(const struct path *root,
1379 const char *filename, int flags, umode_t mode)
1380{
1381 struct open_flags op;
1382 struct open_how how = build_open_how(flags, mode);
1383 int err = build_open_flags(&how, &op);
1384 if (err)
1385 return ERR_PTR(err);
1386 return do_file_open_root(root, filename, &op);
1387}
1388EXPORT_SYMBOL(file_open_root);
1389
1390static long do_sys_openat2(int dfd, const char __user *filename,
1391 struct open_how *how)
1392{
1393 struct open_flags op;
1394 int fd = build_open_flags(how, &op);
1395 struct filename *tmp;
1396
1397 if (fd)
1398 return fd;
1399
1400 tmp = getname(filename);
1401 if (IS_ERR(tmp))
1402 return PTR_ERR(tmp);
1403
1404 fd = get_unused_fd_flags(how->flags);
1405 if (fd >= 0) {
1406 struct file *f = do_filp_open(dfd, tmp, &op);
1407 if (IS_ERR(f)) {
1408 put_unused_fd(fd);
1409 fd = PTR_ERR(f);
1410 } else {
1411 fd_install(fd, f);
1412 }
1413 }
1414 putname(tmp);
1415 return fd;
1416}
1417
1418long do_sys_open(int dfd, const char __user *filename, int flags, umode_t mode)
1419{
1420 struct open_how how = build_open_how(flags, mode);
1421 return do_sys_openat2(dfd, filename, &how);
1422}
1423
1424
1425SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1426{
1427 if (force_o_largefile())
1428 flags |= O_LARGEFILE;
1429 return do_sys_open(AT_FDCWD, filename, flags, mode);
1430}
1431
1432SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags,
1433 umode_t, mode)
1434{
1435 if (force_o_largefile())
1436 flags |= O_LARGEFILE;
1437 return do_sys_open(dfd, filename, flags, mode);
1438}
1439
1440SYSCALL_DEFINE4(openat2, int, dfd, const char __user *, filename,
1441 struct open_how __user *, how, size_t, usize)
1442{
1443 int err;
1444 struct open_how tmp;
1445
1446 BUILD_BUG_ON(sizeof(struct open_how) < OPEN_HOW_SIZE_VER0);
1447 BUILD_BUG_ON(sizeof(struct open_how) != OPEN_HOW_SIZE_LATEST);
1448
1449 if (unlikely(usize < OPEN_HOW_SIZE_VER0))
1450 return -EINVAL;
1451
1452 err = copy_struct_from_user(&tmp, sizeof(tmp), how, usize);
1453 if (err)
1454 return err;
1455
1456 audit_openat2_how(&tmp);
1457
1458 /* O_LARGEFILE is only allowed for non-O_PATH. */
1459 if (!(tmp.flags & O_PATH) && force_o_largefile())
1460 tmp.flags |= O_LARGEFILE;
1461
1462 return do_sys_openat2(dfd, filename, &tmp);
1463}
1464
1465#ifdef CONFIG_COMPAT
1466/*
1467 * Exactly like sys_open(), except that it doesn't set the
1468 * O_LARGEFILE flag.
1469 */
1470COMPAT_SYSCALL_DEFINE3(open, const char __user *, filename, int, flags, umode_t, mode)
1471{
1472 return do_sys_open(AT_FDCWD, filename, flags, mode);
1473}
1474
1475/*
1476 * Exactly like sys_openat(), except that it doesn't set the
1477 * O_LARGEFILE flag.
1478 */
1479COMPAT_SYSCALL_DEFINE4(openat, int, dfd, const char __user *, filename, int, flags, umode_t, mode)
1480{
1481 return do_sys_open(dfd, filename, flags, mode);
1482}
1483#endif
1484
1485#ifndef __alpha__
1486
1487/*
1488 * For backward compatibility? Maybe this should be moved
1489 * into arch/i386 instead?
1490 */
1491SYSCALL_DEFINE2(creat, const char __user *, pathname, umode_t, mode)
1492{
1493 int flags = O_CREAT | O_WRONLY | O_TRUNC;
1494
1495 if (force_o_largefile())
1496 flags |= O_LARGEFILE;
1497 return do_sys_open(AT_FDCWD, pathname, flags, mode);
1498}
1499#endif
1500
1501/*
1502 * "id" is the POSIX thread ID. We use the
1503 * files pointer for this..
1504 */
1505static int filp_flush(struct file *filp, fl_owner_t id)
1506{
1507 int retval = 0;
1508
1509 if (CHECK_DATA_CORRUPTION(file_count(filp) == 0,
1510 "VFS: Close: file count is 0 (f_op=%ps)",
1511 filp->f_op)) {
1512 return 0;
1513 }
1514
1515 if (filp->f_op->flush)
1516 retval = filp->f_op->flush(filp, id);
1517
1518 if (likely(!(filp->f_mode & FMODE_PATH))) {
1519 dnotify_flush(filp, id);
1520 locks_remove_posix(filp, id);
1521 }
1522 return retval;
1523}
1524
1525int filp_close(struct file *filp, fl_owner_t id)
1526{
1527 int retval;
1528
1529 retval = filp_flush(filp, id);
1530 fput(filp);
1531
1532 return retval;
1533}
1534EXPORT_SYMBOL(filp_close);
1535
1536/*
1537 * Careful here! We test whether the file pointer is NULL before
1538 * releasing the fd. This ensures that one clone task can't release
1539 * an fd while another clone is opening it.
1540 */
1541SYSCALL_DEFINE1(close, unsigned int, fd)
1542{
1543 int retval;
1544 struct file *file;
1545
1546 file = file_close_fd(fd);
1547 if (!file)
1548 return -EBADF;
1549
1550 retval = filp_flush(file, current->files);
1551
1552 /*
1553 * We're returning to user space. Don't bother
1554 * with any delayed fput() cases.
1555 */
1556 __fput_sync(file);
1557
1558 /* can't restart close syscall because file table entry was cleared */
1559 if (unlikely(retval == -ERESTARTSYS ||
1560 retval == -ERESTARTNOINTR ||
1561 retval == -ERESTARTNOHAND ||
1562 retval == -ERESTART_RESTARTBLOCK))
1563 retval = -EINTR;
1564
1565 return retval;
1566}
1567
1568/**
1569 * sys_close_range() - Close all file descriptors in a given range.
1570 *
1571 * @fd: starting file descriptor to close
1572 * @max_fd: last file descriptor to close
1573 * @flags: reserved for future extensions
1574 *
1575 * This closes a range of file descriptors. All file descriptors
1576 * from @fd up to and including @max_fd are closed.
1577 * Currently, errors to close a given file descriptor are ignored.
1578 */
1579SYSCALL_DEFINE3(close_range, unsigned int, fd, unsigned int, max_fd,
1580 unsigned int, flags)
1581{
1582 return __close_range(fd, max_fd, flags);
1583}
1584
1585/*
1586 * This routine simulates a hangup on the tty, to arrange that users
1587 * are given clean terminals at login time.
1588 */
1589SYSCALL_DEFINE0(vhangup)
1590{
1591 if (capable(CAP_SYS_TTY_CONFIG)) {
1592 tty_vhangup_self();
1593 return 0;
1594 }
1595 return -EPERM;
1596}
1597
1598/*
1599 * Called when an inode is about to be open.
1600 * We use this to disallow opening large files on 32bit systems if
1601 * the caller didn't specify O_LARGEFILE. On 64bit systems we force
1602 * on this flag in sys_open.
1603 */
1604int generic_file_open(struct inode * inode, struct file * filp)
1605{
1606 if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1607 return -EOVERFLOW;
1608 return 0;
1609}
1610
1611EXPORT_SYMBOL(generic_file_open);
1612
1613/*
1614 * This is used by subsystems that don't want seekable
1615 * file descriptors. The function is not supposed to ever fail, the only
1616 * reason it returns an 'int' and not 'void' is so that it can be plugged
1617 * directly into file_operations structure.
1618 */
1619int nonseekable_open(struct inode *inode, struct file *filp)
1620{
1621 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE);
1622 return 0;
1623}
1624
1625EXPORT_SYMBOL(nonseekable_open);
1626
1627/*
1628 * stream_open is used by subsystems that want stream-like file descriptors.
1629 * Such file descriptors are not seekable and don't have notion of position
1630 * (file.f_pos is always 0 and ppos passed to .read()/.write() is always NULL).
1631 * Contrary to file descriptors of other regular files, .read() and .write()
1632 * can run simultaneously.
1633 *
1634 * stream_open never fails and is marked to return int so that it could be
1635 * directly used as file_operations.open .
1636 */
1637int stream_open(struct inode *inode, struct file *filp)
1638{
1639 filp->f_mode &= ~(FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE | FMODE_ATOMIC_POS);
1640 filp->f_mode |= FMODE_STREAM;
1641 return 0;
1642}
1643
1644EXPORT_SYMBOL(stream_open);