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