xref: /openbmc/linux/fs/read_write.c (revision 29c37341)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/read_write.c
4  *
5  *  Copyright (C) 1991, 1992  Linus Torvalds
6  */
7 
8 #include <linux/slab.h>
9 #include <linux/stat.h>
10 #include <linux/sched/xacct.h>
11 #include <linux/fcntl.h>
12 #include <linux/file.h>
13 #include <linux/uio.h>
14 #include <linux/fsnotify.h>
15 #include <linux/security.h>
16 #include <linux/export.h>
17 #include <linux/syscalls.h>
18 #include <linux/pagemap.h>
19 #include <linux/splice.h>
20 #include <linux/compat.h>
21 #include <linux/mount.h>
22 #include <linux/fs.h>
23 #include "internal.h"
24 
25 #include <linux/uaccess.h>
26 #include <asm/unistd.h>
27 
28 const struct file_operations generic_ro_fops = {
29 	.llseek		= generic_file_llseek,
30 	.read_iter	= generic_file_read_iter,
31 	.mmap		= generic_file_readonly_mmap,
32 	.splice_read	= generic_file_splice_read,
33 };
34 
35 EXPORT_SYMBOL(generic_ro_fops);
36 
37 static inline bool unsigned_offsets(struct file *file)
38 {
39 	return file->f_mode & FMODE_UNSIGNED_OFFSET;
40 }
41 
42 /**
43  * vfs_setpos - update the file offset for lseek
44  * @file:	file structure in question
45  * @offset:	file offset to seek to
46  * @maxsize:	maximum file size
47  *
48  * This is a low-level filesystem helper for updating the file offset to
49  * the value specified by @offset if the given offset is valid and it is
50  * not equal to the current file offset.
51  *
52  * Return the specified offset on success and -EINVAL on invalid offset.
53  */
54 loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
55 {
56 	if (offset < 0 && !unsigned_offsets(file))
57 		return -EINVAL;
58 	if (offset > maxsize)
59 		return -EINVAL;
60 
61 	if (offset != file->f_pos) {
62 		file->f_pos = offset;
63 		file->f_version = 0;
64 	}
65 	return offset;
66 }
67 EXPORT_SYMBOL(vfs_setpos);
68 
69 /**
70  * generic_file_llseek_size - generic llseek implementation for regular files
71  * @file:	file structure to seek on
72  * @offset:	file offset to seek to
73  * @whence:	type of seek
74  * @size:	max size of this file in file system
75  * @eof:	offset used for SEEK_END position
76  *
77  * This is a variant of generic_file_llseek that allows passing in a custom
78  * maximum file size and a custom EOF position, for e.g. hashed directories
79  *
80  * Synchronization:
81  * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
82  * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
83  * read/writes behave like SEEK_SET against seeks.
84  */
85 loff_t
86 generic_file_llseek_size(struct file *file, loff_t offset, int whence,
87 		loff_t maxsize, loff_t eof)
88 {
89 	switch (whence) {
90 	case SEEK_END:
91 		offset += eof;
92 		break;
93 	case SEEK_CUR:
94 		/*
95 		 * Here we special-case the lseek(fd, 0, SEEK_CUR)
96 		 * position-querying operation.  Avoid rewriting the "same"
97 		 * f_pos value back to the file because a concurrent read(),
98 		 * write() or lseek() might have altered it
99 		 */
100 		if (offset == 0)
101 			return file->f_pos;
102 		/*
103 		 * f_lock protects against read/modify/write race with other
104 		 * SEEK_CURs. Note that parallel writes and reads behave
105 		 * like SEEK_SET.
106 		 */
107 		spin_lock(&file->f_lock);
108 		offset = vfs_setpos(file, file->f_pos + offset, maxsize);
109 		spin_unlock(&file->f_lock);
110 		return offset;
111 	case SEEK_DATA:
112 		/*
113 		 * In the generic case the entire file is data, so as long as
114 		 * offset isn't at the end of the file then the offset is data.
115 		 */
116 		if ((unsigned long long)offset >= eof)
117 			return -ENXIO;
118 		break;
119 	case SEEK_HOLE:
120 		/*
121 		 * There is a virtual hole at the end of the file, so as long as
122 		 * offset isn't i_size or larger, return i_size.
123 		 */
124 		if ((unsigned long long)offset >= eof)
125 			return -ENXIO;
126 		offset = eof;
127 		break;
128 	}
129 
130 	return vfs_setpos(file, offset, maxsize);
131 }
132 EXPORT_SYMBOL(generic_file_llseek_size);
133 
134 /**
135  * generic_file_llseek - generic llseek implementation for regular files
136  * @file:	file structure to seek on
137  * @offset:	file offset to seek to
138  * @whence:	type of seek
139  *
140  * This is a generic implemenation of ->llseek useable for all normal local
141  * filesystems.  It just updates the file offset to the value specified by
142  * @offset and @whence.
143  */
144 loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
145 {
146 	struct inode *inode = file->f_mapping->host;
147 
148 	return generic_file_llseek_size(file, offset, whence,
149 					inode->i_sb->s_maxbytes,
150 					i_size_read(inode));
151 }
152 EXPORT_SYMBOL(generic_file_llseek);
153 
154 /**
155  * fixed_size_llseek - llseek implementation for fixed-sized devices
156  * @file:	file structure to seek on
157  * @offset:	file offset to seek to
158  * @whence:	type of seek
159  * @size:	size of the file
160  *
161  */
162 loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
163 {
164 	switch (whence) {
165 	case SEEK_SET: case SEEK_CUR: case SEEK_END:
166 		return generic_file_llseek_size(file, offset, whence,
167 						size, size);
168 	default:
169 		return -EINVAL;
170 	}
171 }
172 EXPORT_SYMBOL(fixed_size_llseek);
173 
174 /**
175  * no_seek_end_llseek - llseek implementation for fixed-sized devices
176  * @file:	file structure to seek on
177  * @offset:	file offset to seek to
178  * @whence:	type of seek
179  *
180  */
181 loff_t no_seek_end_llseek(struct file *file, loff_t offset, int whence)
182 {
183 	switch (whence) {
184 	case SEEK_SET: case SEEK_CUR:
185 		return generic_file_llseek_size(file, offset, whence,
186 						OFFSET_MAX, 0);
187 	default:
188 		return -EINVAL;
189 	}
190 }
191 EXPORT_SYMBOL(no_seek_end_llseek);
192 
193 /**
194  * no_seek_end_llseek_size - llseek implementation for fixed-sized devices
195  * @file:	file structure to seek on
196  * @offset:	file offset to seek to
197  * @whence:	type of seek
198  * @size:	maximal offset allowed
199  *
200  */
201 loff_t no_seek_end_llseek_size(struct file *file, loff_t offset, int whence, loff_t size)
202 {
203 	switch (whence) {
204 	case SEEK_SET: case SEEK_CUR:
205 		return generic_file_llseek_size(file, offset, whence,
206 						size, 0);
207 	default:
208 		return -EINVAL;
209 	}
210 }
211 EXPORT_SYMBOL(no_seek_end_llseek_size);
212 
213 /**
214  * noop_llseek - No Operation Performed llseek implementation
215  * @file:	file structure to seek on
216  * @offset:	file offset to seek to
217  * @whence:	type of seek
218  *
219  * This is an implementation of ->llseek useable for the rare special case when
220  * userspace expects the seek to succeed but the (device) file is actually not
221  * able to perform the seek. In this case you use noop_llseek() instead of
222  * falling back to the default implementation of ->llseek.
223  */
224 loff_t noop_llseek(struct file *file, loff_t offset, int whence)
225 {
226 	return file->f_pos;
227 }
228 EXPORT_SYMBOL(noop_llseek);
229 
230 loff_t no_llseek(struct file *file, loff_t offset, int whence)
231 {
232 	return -ESPIPE;
233 }
234 EXPORT_SYMBOL(no_llseek);
235 
236 loff_t default_llseek(struct file *file, loff_t offset, int whence)
237 {
238 	struct inode *inode = file_inode(file);
239 	loff_t retval;
240 
241 	inode_lock(inode);
242 	switch (whence) {
243 		case SEEK_END:
244 			offset += i_size_read(inode);
245 			break;
246 		case SEEK_CUR:
247 			if (offset == 0) {
248 				retval = file->f_pos;
249 				goto out;
250 			}
251 			offset += file->f_pos;
252 			break;
253 		case SEEK_DATA:
254 			/*
255 			 * In the generic case the entire file is data, so as
256 			 * long as offset isn't at the end of the file then the
257 			 * offset is data.
258 			 */
259 			if (offset >= inode->i_size) {
260 				retval = -ENXIO;
261 				goto out;
262 			}
263 			break;
264 		case SEEK_HOLE:
265 			/*
266 			 * There is a virtual hole at the end of the file, so
267 			 * as long as offset isn't i_size or larger, return
268 			 * i_size.
269 			 */
270 			if (offset >= inode->i_size) {
271 				retval = -ENXIO;
272 				goto out;
273 			}
274 			offset = inode->i_size;
275 			break;
276 	}
277 	retval = -EINVAL;
278 	if (offset >= 0 || unsigned_offsets(file)) {
279 		if (offset != file->f_pos) {
280 			file->f_pos = offset;
281 			file->f_version = 0;
282 		}
283 		retval = offset;
284 	}
285 out:
286 	inode_unlock(inode);
287 	return retval;
288 }
289 EXPORT_SYMBOL(default_llseek);
290 
291 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
292 {
293 	loff_t (*fn)(struct file *, loff_t, int);
294 
295 	fn = no_llseek;
296 	if (file->f_mode & FMODE_LSEEK) {
297 		if (file->f_op->llseek)
298 			fn = file->f_op->llseek;
299 	}
300 	return fn(file, offset, whence);
301 }
302 EXPORT_SYMBOL(vfs_llseek);
303 
304 static off_t ksys_lseek(unsigned int fd, off_t offset, unsigned int whence)
305 {
306 	off_t retval;
307 	struct fd f = fdget_pos(fd);
308 	if (!f.file)
309 		return -EBADF;
310 
311 	retval = -EINVAL;
312 	if (whence <= SEEK_MAX) {
313 		loff_t res = vfs_llseek(f.file, offset, whence);
314 		retval = res;
315 		if (res != (loff_t)retval)
316 			retval = -EOVERFLOW;	/* LFS: should only happen on 32 bit platforms */
317 	}
318 	fdput_pos(f);
319 	return retval;
320 }
321 
322 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
323 {
324 	return ksys_lseek(fd, offset, whence);
325 }
326 
327 #ifdef CONFIG_COMPAT
328 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
329 {
330 	return ksys_lseek(fd, offset, whence);
331 }
332 #endif
333 
334 #if !defined(CONFIG_64BIT) || defined(CONFIG_COMPAT) || \
335 	defined(__ARCH_WANT_SYS_LLSEEK)
336 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
337 		unsigned long, offset_low, loff_t __user *, result,
338 		unsigned int, whence)
339 {
340 	int retval;
341 	struct fd f = fdget_pos(fd);
342 	loff_t offset;
343 
344 	if (!f.file)
345 		return -EBADF;
346 
347 	retval = -EINVAL;
348 	if (whence > SEEK_MAX)
349 		goto out_putf;
350 
351 	offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
352 			whence);
353 
354 	retval = (int)offset;
355 	if (offset >= 0) {
356 		retval = -EFAULT;
357 		if (!copy_to_user(result, &offset, sizeof(offset)))
358 			retval = 0;
359 	}
360 out_putf:
361 	fdput_pos(f);
362 	return retval;
363 }
364 #endif
365 
366 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
367 {
368 	struct inode *inode;
369 	int retval = -EINVAL;
370 
371 	inode = file_inode(file);
372 	if (unlikely((ssize_t) count < 0))
373 		return retval;
374 
375 	/*
376 	 * ranged mandatory locking does not apply to streams - it makes sense
377 	 * only for files where position has a meaning.
378 	 */
379 	if (ppos) {
380 		loff_t pos = *ppos;
381 
382 		if (unlikely(pos < 0)) {
383 			if (!unsigned_offsets(file))
384 				return retval;
385 			if (count >= -pos) /* both values are in 0..LLONG_MAX */
386 				return -EOVERFLOW;
387 		} else if (unlikely((loff_t) (pos + count) < 0)) {
388 			if (!unsigned_offsets(file))
389 				return retval;
390 		}
391 
392 		if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
393 			retval = locks_mandatory_area(inode, file, pos, pos + count - 1,
394 					read_write == READ ? F_RDLCK : F_WRLCK);
395 			if (retval < 0)
396 				return retval;
397 		}
398 	}
399 
400 	return security_file_permission(file,
401 				read_write == READ ? MAY_READ : MAY_WRITE);
402 }
403 
404 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
405 {
406 	struct iovec iov = { .iov_base = buf, .iov_len = len };
407 	struct kiocb kiocb;
408 	struct iov_iter iter;
409 	ssize_t ret;
410 
411 	init_sync_kiocb(&kiocb, filp);
412 	kiocb.ki_pos = (ppos ? *ppos : 0);
413 	iov_iter_init(&iter, READ, &iov, 1, len);
414 
415 	ret = call_read_iter(filp, &kiocb, &iter);
416 	BUG_ON(ret == -EIOCBQUEUED);
417 	if (ppos)
418 		*ppos = kiocb.ki_pos;
419 	return ret;
420 }
421 
422 ssize_t __kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
423 {
424 	mm_segment_t old_fs = get_fs();
425 	ssize_t ret;
426 
427 	if (WARN_ON_ONCE(!(file->f_mode & FMODE_READ)))
428 		return -EINVAL;
429 	if (!(file->f_mode & FMODE_CAN_READ))
430 		return -EINVAL;
431 
432 	if (count > MAX_RW_COUNT)
433 		count =  MAX_RW_COUNT;
434 	set_fs(KERNEL_DS);
435 	if (file->f_op->read)
436 		ret = file->f_op->read(file, (void __user *)buf, count, pos);
437 	else if (file->f_op->read_iter)
438 		ret = new_sync_read(file, (void __user *)buf, count, pos);
439 	else
440 		ret = -EINVAL;
441 	set_fs(old_fs);
442 	if (ret > 0) {
443 		fsnotify_access(file);
444 		add_rchar(current, ret);
445 	}
446 	inc_syscr(current);
447 	return ret;
448 }
449 
450 ssize_t kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
451 {
452 	ssize_t ret;
453 
454 	ret = rw_verify_area(READ, file, pos, count);
455 	if (ret)
456 		return ret;
457 	return __kernel_read(file, buf, count, pos);
458 }
459 EXPORT_SYMBOL(kernel_read);
460 
461 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
462 {
463 	ssize_t ret;
464 
465 	if (!(file->f_mode & FMODE_READ))
466 		return -EBADF;
467 	if (!(file->f_mode & FMODE_CAN_READ))
468 		return -EINVAL;
469 	if (unlikely(!access_ok(buf, count)))
470 		return -EFAULT;
471 
472 	ret = rw_verify_area(READ, file, pos, count);
473 	if (ret)
474 		return ret;
475 	if (count > MAX_RW_COUNT)
476 		count =  MAX_RW_COUNT;
477 
478 	if (file->f_op->read)
479 		ret = file->f_op->read(file, buf, count, pos);
480 	else if (file->f_op->read_iter)
481 		ret = new_sync_read(file, buf, count, pos);
482 	else
483 		ret = -EINVAL;
484 	if (ret > 0) {
485 		fsnotify_access(file);
486 		add_rchar(current, ret);
487 	}
488 	inc_syscr(current);
489 	return ret;
490 }
491 
492 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
493 {
494 	struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
495 	struct kiocb kiocb;
496 	struct iov_iter iter;
497 	ssize_t ret;
498 
499 	init_sync_kiocb(&kiocb, filp);
500 	kiocb.ki_pos = (ppos ? *ppos : 0);
501 	iov_iter_init(&iter, WRITE, &iov, 1, len);
502 
503 	ret = call_write_iter(filp, &kiocb, &iter);
504 	BUG_ON(ret == -EIOCBQUEUED);
505 	if (ret > 0 && ppos)
506 		*ppos = kiocb.ki_pos;
507 	return ret;
508 }
509 
510 /* caller is responsible for file_start_write/file_end_write */
511 ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t *pos)
512 {
513 	mm_segment_t old_fs;
514 	const char __user *p;
515 	ssize_t ret;
516 
517 	if (WARN_ON_ONCE(!(file->f_mode & FMODE_WRITE)))
518 		return -EBADF;
519 	if (!(file->f_mode & FMODE_CAN_WRITE))
520 		return -EINVAL;
521 
522 	old_fs = get_fs();
523 	set_fs(KERNEL_DS);
524 	p = (__force const char __user *)buf;
525 	if (count > MAX_RW_COUNT)
526 		count =  MAX_RW_COUNT;
527 	if (file->f_op->write)
528 		ret = file->f_op->write(file, p, count, pos);
529 	else if (file->f_op->write_iter)
530 		ret = new_sync_write(file, p, count, pos);
531 	else
532 		ret = -EINVAL;
533 	set_fs(old_fs);
534 	if (ret > 0) {
535 		fsnotify_modify(file);
536 		add_wchar(current, ret);
537 	}
538 	inc_syscw(current);
539 	return ret;
540 }
541 
542 ssize_t kernel_write(struct file *file, const void *buf, size_t count,
543 			    loff_t *pos)
544 {
545 	ssize_t ret;
546 
547 	ret = rw_verify_area(WRITE, file, pos, count);
548 	if (ret)
549 		return ret;
550 
551 	file_start_write(file);
552 	ret =  __kernel_write(file, buf, count, pos);
553 	file_end_write(file);
554 	return ret;
555 }
556 EXPORT_SYMBOL(kernel_write);
557 
558 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
559 {
560 	ssize_t ret;
561 
562 	if (!(file->f_mode & FMODE_WRITE))
563 		return -EBADF;
564 	if (!(file->f_mode & FMODE_CAN_WRITE))
565 		return -EINVAL;
566 	if (unlikely(!access_ok(buf, count)))
567 		return -EFAULT;
568 
569 	ret = rw_verify_area(WRITE, file, pos, count);
570 	if (ret)
571 		return ret;
572 	if (count > MAX_RW_COUNT)
573 		count =  MAX_RW_COUNT;
574 	file_start_write(file);
575 	if (file->f_op->write)
576 		ret = file->f_op->write(file, buf, count, pos);
577 	else if (file->f_op->write_iter)
578 		ret = new_sync_write(file, buf, count, pos);
579 	else
580 		ret = -EINVAL;
581 	if (ret > 0) {
582 		fsnotify_modify(file);
583 		add_wchar(current, ret);
584 	}
585 	inc_syscw(current);
586 	file_end_write(file);
587 	return ret;
588 }
589 
590 /* file_ppos returns &file->f_pos or NULL if file is stream */
591 static inline loff_t *file_ppos(struct file *file)
592 {
593 	return file->f_mode & FMODE_STREAM ? NULL : &file->f_pos;
594 }
595 
596 ssize_t ksys_read(unsigned int fd, char __user *buf, size_t count)
597 {
598 	struct fd f = fdget_pos(fd);
599 	ssize_t ret = -EBADF;
600 
601 	if (f.file) {
602 		loff_t pos, *ppos = file_ppos(f.file);
603 		if (ppos) {
604 			pos = *ppos;
605 			ppos = &pos;
606 		}
607 		ret = vfs_read(f.file, buf, count, ppos);
608 		if (ret >= 0 && ppos)
609 			f.file->f_pos = pos;
610 		fdput_pos(f);
611 	}
612 	return ret;
613 }
614 
615 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
616 {
617 	return ksys_read(fd, buf, count);
618 }
619 
620 ssize_t ksys_write(unsigned int fd, const char __user *buf, size_t count)
621 {
622 	struct fd f = fdget_pos(fd);
623 	ssize_t ret = -EBADF;
624 
625 	if (f.file) {
626 		loff_t pos, *ppos = file_ppos(f.file);
627 		if (ppos) {
628 			pos = *ppos;
629 			ppos = &pos;
630 		}
631 		ret = vfs_write(f.file, buf, count, ppos);
632 		if (ret >= 0 && ppos)
633 			f.file->f_pos = pos;
634 		fdput_pos(f);
635 	}
636 
637 	return ret;
638 }
639 
640 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
641 		size_t, count)
642 {
643 	return ksys_write(fd, buf, count);
644 }
645 
646 ssize_t ksys_pread64(unsigned int fd, char __user *buf, size_t count,
647 		     loff_t pos)
648 {
649 	struct fd f;
650 	ssize_t ret = -EBADF;
651 
652 	if (pos < 0)
653 		return -EINVAL;
654 
655 	f = fdget(fd);
656 	if (f.file) {
657 		ret = -ESPIPE;
658 		if (f.file->f_mode & FMODE_PREAD)
659 			ret = vfs_read(f.file, buf, count, &pos);
660 		fdput(f);
661 	}
662 
663 	return ret;
664 }
665 
666 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
667 			size_t, count, loff_t, pos)
668 {
669 	return ksys_pread64(fd, buf, count, pos);
670 }
671 
672 ssize_t ksys_pwrite64(unsigned int fd, const char __user *buf,
673 		      size_t count, loff_t pos)
674 {
675 	struct fd f;
676 	ssize_t ret = -EBADF;
677 
678 	if (pos < 0)
679 		return -EINVAL;
680 
681 	f = fdget(fd);
682 	if (f.file) {
683 		ret = -ESPIPE;
684 		if (f.file->f_mode & FMODE_PWRITE)
685 			ret = vfs_write(f.file, buf, count, &pos);
686 		fdput(f);
687 	}
688 
689 	return ret;
690 }
691 
692 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
693 			 size_t, count, loff_t, pos)
694 {
695 	return ksys_pwrite64(fd, buf, count, pos);
696 }
697 
698 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
699 		loff_t *ppos, int type, rwf_t flags)
700 {
701 	struct kiocb kiocb;
702 	ssize_t ret;
703 
704 	init_sync_kiocb(&kiocb, filp);
705 	ret = kiocb_set_rw_flags(&kiocb, flags);
706 	if (ret)
707 		return ret;
708 	kiocb.ki_pos = (ppos ? *ppos : 0);
709 
710 	if (type == READ)
711 		ret = call_read_iter(filp, &kiocb, iter);
712 	else
713 		ret = call_write_iter(filp, &kiocb, iter);
714 	BUG_ON(ret == -EIOCBQUEUED);
715 	if (ppos)
716 		*ppos = kiocb.ki_pos;
717 	return ret;
718 }
719 
720 /* Do it by hand, with file-ops */
721 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
722 		loff_t *ppos, int type, rwf_t flags)
723 {
724 	ssize_t ret = 0;
725 
726 	if (flags & ~RWF_HIPRI)
727 		return -EOPNOTSUPP;
728 
729 	while (iov_iter_count(iter)) {
730 		struct iovec iovec = iov_iter_iovec(iter);
731 		ssize_t nr;
732 
733 		if (type == READ) {
734 			nr = filp->f_op->read(filp, iovec.iov_base,
735 					      iovec.iov_len, ppos);
736 		} else {
737 			nr = filp->f_op->write(filp, iovec.iov_base,
738 					       iovec.iov_len, ppos);
739 		}
740 
741 		if (nr < 0) {
742 			if (!ret)
743 				ret = nr;
744 			break;
745 		}
746 		ret += nr;
747 		if (nr != iovec.iov_len)
748 			break;
749 		iov_iter_advance(iter, nr);
750 	}
751 
752 	return ret;
753 }
754 
755 /**
756  * rw_copy_check_uvector() - Copy an array of &struct iovec from userspace
757  *     into the kernel and check that it is valid.
758  *
759  * @type: One of %CHECK_IOVEC_ONLY, %READ, or %WRITE.
760  * @uvector: Pointer to the userspace array.
761  * @nr_segs: Number of elements in userspace array.
762  * @fast_segs: Number of elements in @fast_pointer.
763  * @fast_pointer: Pointer to (usually small on-stack) kernel array.
764  * @ret_pointer: (output parameter) Pointer to a variable that will point to
765  *     either @fast_pointer, a newly allocated kernel array, or NULL,
766  *     depending on which array was used.
767  *
768  * This function copies an array of &struct iovec of @nr_segs from
769  * userspace into the kernel and checks that each element is valid (e.g.
770  * it does not point to a kernel address or cause overflow by being too
771  * large, etc.).
772  *
773  * As an optimization, the caller may provide a pointer to a small
774  * on-stack array in @fast_pointer, typically %UIO_FASTIOV elements long
775  * (the size of this array, or 0 if unused, should be given in @fast_segs).
776  *
777  * @ret_pointer will always point to the array that was used, so the
778  * caller must take care not to call kfree() on it e.g. in case the
779  * @fast_pointer array was used and it was allocated on the stack.
780  *
781  * Return: The total number of bytes covered by the iovec array on success
782  *   or a negative error code on error.
783  */
784 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
785 			      unsigned long nr_segs, unsigned long fast_segs,
786 			      struct iovec *fast_pointer,
787 			      struct iovec **ret_pointer)
788 {
789 	unsigned long seg;
790 	ssize_t ret;
791 	struct iovec *iov = fast_pointer;
792 
793 	/*
794 	 * SuS says "The readv() function *may* fail if the iovcnt argument
795 	 * was less than or equal to 0, or greater than {IOV_MAX}.  Linux has
796 	 * traditionally returned zero for zero segments, so...
797 	 */
798 	if (nr_segs == 0) {
799 		ret = 0;
800 		goto out;
801 	}
802 
803 	/*
804 	 * First get the "struct iovec" from user memory and
805 	 * verify all the pointers
806 	 */
807 	if (nr_segs > UIO_MAXIOV) {
808 		ret = -EINVAL;
809 		goto out;
810 	}
811 	if (nr_segs > fast_segs) {
812 		iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
813 		if (iov == NULL) {
814 			ret = -ENOMEM;
815 			goto out;
816 		}
817 	}
818 	if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
819 		ret = -EFAULT;
820 		goto out;
821 	}
822 
823 	/*
824 	 * According to the Single Unix Specification we should return EINVAL
825 	 * if an element length is < 0 when cast to ssize_t or if the
826 	 * total length would overflow the ssize_t return value of the
827 	 * system call.
828 	 *
829 	 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
830 	 * overflow case.
831 	 */
832 	ret = 0;
833 	for (seg = 0; seg < nr_segs; seg++) {
834 		void __user *buf = iov[seg].iov_base;
835 		ssize_t len = (ssize_t)iov[seg].iov_len;
836 
837 		/* see if we we're about to use an invalid len or if
838 		 * it's about to overflow ssize_t */
839 		if (len < 0) {
840 			ret = -EINVAL;
841 			goto out;
842 		}
843 		if (type >= 0
844 		    && unlikely(!access_ok(buf, len))) {
845 			ret = -EFAULT;
846 			goto out;
847 		}
848 		if (len > MAX_RW_COUNT - ret) {
849 			len = MAX_RW_COUNT - ret;
850 			iov[seg].iov_len = len;
851 		}
852 		ret += len;
853 	}
854 out:
855 	*ret_pointer = iov;
856 	return ret;
857 }
858 
859 #ifdef CONFIG_COMPAT
860 ssize_t compat_rw_copy_check_uvector(int type,
861 		const struct compat_iovec __user *uvector, unsigned long nr_segs,
862 		unsigned long fast_segs, struct iovec *fast_pointer,
863 		struct iovec **ret_pointer)
864 {
865 	compat_ssize_t tot_len;
866 	struct iovec *iov = *ret_pointer = fast_pointer;
867 	ssize_t ret = 0;
868 	int seg;
869 
870 	/*
871 	 * SuS says "The readv() function *may* fail if the iovcnt argument
872 	 * was less than or equal to 0, or greater than {IOV_MAX}.  Linux has
873 	 * traditionally returned zero for zero segments, so...
874 	 */
875 	if (nr_segs == 0)
876 		goto out;
877 
878 	ret = -EINVAL;
879 	if (nr_segs > UIO_MAXIOV)
880 		goto out;
881 	if (nr_segs > fast_segs) {
882 		ret = -ENOMEM;
883 		iov = kmalloc_array(nr_segs, sizeof(struct iovec), GFP_KERNEL);
884 		if (iov == NULL)
885 			goto out;
886 	}
887 	*ret_pointer = iov;
888 
889 	ret = -EFAULT;
890 	if (!access_ok(uvector, nr_segs*sizeof(*uvector)))
891 		goto out;
892 
893 	/*
894 	 * Single unix specification:
895 	 * We should -EINVAL if an element length is not >= 0 and fitting an
896 	 * ssize_t.
897 	 *
898 	 * In Linux, the total length is limited to MAX_RW_COUNT, there is
899 	 * no overflow possibility.
900 	 */
901 	tot_len = 0;
902 	ret = -EINVAL;
903 	for (seg = 0; seg < nr_segs; seg++) {
904 		compat_uptr_t buf;
905 		compat_ssize_t len;
906 
907 		if (__get_user(len, &uvector->iov_len) ||
908 		   __get_user(buf, &uvector->iov_base)) {
909 			ret = -EFAULT;
910 			goto out;
911 		}
912 		if (len < 0)	/* size_t not fitting in compat_ssize_t .. */
913 			goto out;
914 		if (type >= 0 &&
915 		    !access_ok(compat_ptr(buf), len)) {
916 			ret = -EFAULT;
917 			goto out;
918 		}
919 		if (len > MAX_RW_COUNT - tot_len)
920 			len = MAX_RW_COUNT - tot_len;
921 		tot_len += len;
922 		iov->iov_base = compat_ptr(buf);
923 		iov->iov_len = (compat_size_t) len;
924 		uvector++;
925 		iov++;
926 	}
927 	ret = tot_len;
928 
929 out:
930 	return ret;
931 }
932 #endif
933 
934 static ssize_t do_iter_read(struct file *file, struct iov_iter *iter,
935 		loff_t *pos, rwf_t flags)
936 {
937 	size_t tot_len;
938 	ssize_t ret = 0;
939 
940 	if (!(file->f_mode & FMODE_READ))
941 		return -EBADF;
942 	if (!(file->f_mode & FMODE_CAN_READ))
943 		return -EINVAL;
944 
945 	tot_len = iov_iter_count(iter);
946 	if (!tot_len)
947 		goto out;
948 	ret = rw_verify_area(READ, file, pos, tot_len);
949 	if (ret < 0)
950 		return ret;
951 
952 	if (file->f_op->read_iter)
953 		ret = do_iter_readv_writev(file, iter, pos, READ, flags);
954 	else
955 		ret = do_loop_readv_writev(file, iter, pos, READ, flags);
956 out:
957 	if (ret >= 0)
958 		fsnotify_access(file);
959 	return ret;
960 }
961 
962 ssize_t vfs_iocb_iter_read(struct file *file, struct kiocb *iocb,
963 			   struct iov_iter *iter)
964 {
965 	size_t tot_len;
966 	ssize_t ret = 0;
967 
968 	if (!file->f_op->read_iter)
969 		return -EINVAL;
970 	if (!(file->f_mode & FMODE_READ))
971 		return -EBADF;
972 	if (!(file->f_mode & FMODE_CAN_READ))
973 		return -EINVAL;
974 
975 	tot_len = iov_iter_count(iter);
976 	if (!tot_len)
977 		goto out;
978 	ret = rw_verify_area(READ, file, &iocb->ki_pos, tot_len);
979 	if (ret < 0)
980 		return ret;
981 
982 	ret = call_read_iter(file, iocb, iter);
983 out:
984 	if (ret >= 0)
985 		fsnotify_access(file);
986 	return ret;
987 }
988 EXPORT_SYMBOL(vfs_iocb_iter_read);
989 
990 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos,
991 		rwf_t flags)
992 {
993 	if (!file->f_op->read_iter)
994 		return -EINVAL;
995 	return do_iter_read(file, iter, ppos, flags);
996 }
997 EXPORT_SYMBOL(vfs_iter_read);
998 
999 static ssize_t do_iter_write(struct file *file, struct iov_iter *iter,
1000 		loff_t *pos, rwf_t flags)
1001 {
1002 	size_t tot_len;
1003 	ssize_t ret = 0;
1004 
1005 	if (!(file->f_mode & FMODE_WRITE))
1006 		return -EBADF;
1007 	if (!(file->f_mode & FMODE_CAN_WRITE))
1008 		return -EINVAL;
1009 
1010 	tot_len = iov_iter_count(iter);
1011 	if (!tot_len)
1012 		return 0;
1013 	ret = rw_verify_area(WRITE, file, pos, tot_len);
1014 	if (ret < 0)
1015 		return ret;
1016 
1017 	if (file->f_op->write_iter)
1018 		ret = do_iter_readv_writev(file, iter, pos, WRITE, flags);
1019 	else
1020 		ret = do_loop_readv_writev(file, iter, pos, WRITE, flags);
1021 	if (ret > 0)
1022 		fsnotify_modify(file);
1023 	return ret;
1024 }
1025 
1026 ssize_t vfs_iocb_iter_write(struct file *file, struct kiocb *iocb,
1027 			    struct iov_iter *iter)
1028 {
1029 	size_t tot_len;
1030 	ssize_t ret = 0;
1031 
1032 	if (!file->f_op->write_iter)
1033 		return -EINVAL;
1034 	if (!(file->f_mode & FMODE_WRITE))
1035 		return -EBADF;
1036 	if (!(file->f_mode & FMODE_CAN_WRITE))
1037 		return -EINVAL;
1038 
1039 	tot_len = iov_iter_count(iter);
1040 	if (!tot_len)
1041 		return 0;
1042 	ret = rw_verify_area(WRITE, file, &iocb->ki_pos, tot_len);
1043 	if (ret < 0)
1044 		return ret;
1045 
1046 	ret = call_write_iter(file, iocb, iter);
1047 	if (ret > 0)
1048 		fsnotify_modify(file);
1049 
1050 	return ret;
1051 }
1052 EXPORT_SYMBOL(vfs_iocb_iter_write);
1053 
1054 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos,
1055 		rwf_t flags)
1056 {
1057 	if (!file->f_op->write_iter)
1058 		return -EINVAL;
1059 	return do_iter_write(file, iter, ppos, flags);
1060 }
1061 EXPORT_SYMBOL(vfs_iter_write);
1062 
1063 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
1064 		  unsigned long vlen, loff_t *pos, rwf_t flags)
1065 {
1066 	struct iovec iovstack[UIO_FASTIOV];
1067 	struct iovec *iov = iovstack;
1068 	struct iov_iter iter;
1069 	ssize_t ret;
1070 
1071 	ret = import_iovec(READ, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
1072 	if (ret >= 0) {
1073 		ret = do_iter_read(file, &iter, pos, flags);
1074 		kfree(iov);
1075 	}
1076 
1077 	return ret;
1078 }
1079 
1080 static ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
1081 		   unsigned long vlen, loff_t *pos, rwf_t flags)
1082 {
1083 	struct iovec iovstack[UIO_FASTIOV];
1084 	struct iovec *iov = iovstack;
1085 	struct iov_iter iter;
1086 	ssize_t ret;
1087 
1088 	ret = import_iovec(WRITE, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
1089 	if (ret >= 0) {
1090 		file_start_write(file);
1091 		ret = do_iter_write(file, &iter, pos, flags);
1092 		file_end_write(file);
1093 		kfree(iov);
1094 	}
1095 	return ret;
1096 }
1097 
1098 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
1099 			unsigned long vlen, rwf_t flags)
1100 {
1101 	struct fd f = fdget_pos(fd);
1102 	ssize_t ret = -EBADF;
1103 
1104 	if (f.file) {
1105 		loff_t pos, *ppos = file_ppos(f.file);
1106 		if (ppos) {
1107 			pos = *ppos;
1108 			ppos = &pos;
1109 		}
1110 		ret = vfs_readv(f.file, vec, vlen, ppos, flags);
1111 		if (ret >= 0 && ppos)
1112 			f.file->f_pos = pos;
1113 		fdput_pos(f);
1114 	}
1115 
1116 	if (ret > 0)
1117 		add_rchar(current, ret);
1118 	inc_syscr(current);
1119 	return ret;
1120 }
1121 
1122 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
1123 			 unsigned long vlen, rwf_t flags)
1124 {
1125 	struct fd f = fdget_pos(fd);
1126 	ssize_t ret = -EBADF;
1127 
1128 	if (f.file) {
1129 		loff_t pos, *ppos = file_ppos(f.file);
1130 		if (ppos) {
1131 			pos = *ppos;
1132 			ppos = &pos;
1133 		}
1134 		ret = vfs_writev(f.file, vec, vlen, ppos, flags);
1135 		if (ret >= 0 && ppos)
1136 			f.file->f_pos = pos;
1137 		fdput_pos(f);
1138 	}
1139 
1140 	if (ret > 0)
1141 		add_wchar(current, ret);
1142 	inc_syscw(current);
1143 	return ret;
1144 }
1145 
1146 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
1147 {
1148 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
1149 	return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
1150 }
1151 
1152 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
1153 			 unsigned long vlen, loff_t pos, rwf_t flags)
1154 {
1155 	struct fd f;
1156 	ssize_t ret = -EBADF;
1157 
1158 	if (pos < 0)
1159 		return -EINVAL;
1160 
1161 	f = fdget(fd);
1162 	if (f.file) {
1163 		ret = -ESPIPE;
1164 		if (f.file->f_mode & FMODE_PREAD)
1165 			ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1166 		fdput(f);
1167 	}
1168 
1169 	if (ret > 0)
1170 		add_rchar(current, ret);
1171 	inc_syscr(current);
1172 	return ret;
1173 }
1174 
1175 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
1176 			  unsigned long vlen, loff_t pos, rwf_t flags)
1177 {
1178 	struct fd f;
1179 	ssize_t ret = -EBADF;
1180 
1181 	if (pos < 0)
1182 		return -EINVAL;
1183 
1184 	f = fdget(fd);
1185 	if (f.file) {
1186 		ret = -ESPIPE;
1187 		if (f.file->f_mode & FMODE_PWRITE)
1188 			ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1189 		fdput(f);
1190 	}
1191 
1192 	if (ret > 0)
1193 		add_wchar(current, ret);
1194 	inc_syscw(current);
1195 	return ret;
1196 }
1197 
1198 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1199 		unsigned long, vlen)
1200 {
1201 	return do_readv(fd, vec, vlen, 0);
1202 }
1203 
1204 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1205 		unsigned long, vlen)
1206 {
1207 	return do_writev(fd, vec, vlen, 0);
1208 }
1209 
1210 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1211 		unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1212 {
1213 	loff_t pos = pos_from_hilo(pos_h, pos_l);
1214 
1215 	return do_preadv(fd, vec, vlen, pos, 0);
1216 }
1217 
1218 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1219 		unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1220 		rwf_t, flags)
1221 {
1222 	loff_t pos = pos_from_hilo(pos_h, pos_l);
1223 
1224 	if (pos == -1)
1225 		return do_readv(fd, vec, vlen, flags);
1226 
1227 	return do_preadv(fd, vec, vlen, pos, flags);
1228 }
1229 
1230 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1231 		unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1232 {
1233 	loff_t pos = pos_from_hilo(pos_h, pos_l);
1234 
1235 	return do_pwritev(fd, vec, vlen, pos, 0);
1236 }
1237 
1238 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1239 		unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1240 		rwf_t, flags)
1241 {
1242 	loff_t pos = pos_from_hilo(pos_h, pos_l);
1243 
1244 	if (pos == -1)
1245 		return do_writev(fd, vec, vlen, flags);
1246 
1247 	return do_pwritev(fd, vec, vlen, pos, flags);
1248 }
1249 
1250 #ifdef CONFIG_COMPAT
1251 static size_t compat_readv(struct file *file,
1252 			   const struct compat_iovec __user *vec,
1253 			   unsigned long vlen, loff_t *pos, rwf_t flags)
1254 {
1255 	struct iovec iovstack[UIO_FASTIOV];
1256 	struct iovec *iov = iovstack;
1257 	struct iov_iter iter;
1258 	ssize_t ret;
1259 
1260 	ret = compat_import_iovec(READ, vec, vlen, UIO_FASTIOV, &iov, &iter);
1261 	if (ret >= 0) {
1262 		ret = do_iter_read(file, &iter, pos, flags);
1263 		kfree(iov);
1264 	}
1265 	if (ret > 0)
1266 		add_rchar(current, ret);
1267 	inc_syscr(current);
1268 	return ret;
1269 }
1270 
1271 static size_t do_compat_readv(compat_ulong_t fd,
1272 				 const struct compat_iovec __user *vec,
1273 				 compat_ulong_t vlen, rwf_t flags)
1274 {
1275 	struct fd f = fdget_pos(fd);
1276 	ssize_t ret;
1277 	loff_t pos;
1278 
1279 	if (!f.file)
1280 		return -EBADF;
1281 	pos = f.file->f_pos;
1282 	ret = compat_readv(f.file, vec, vlen, &pos, flags);
1283 	if (ret >= 0)
1284 		f.file->f_pos = pos;
1285 	fdput_pos(f);
1286 	return ret;
1287 
1288 }
1289 
1290 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1291 		const struct compat_iovec __user *,vec,
1292 		compat_ulong_t, vlen)
1293 {
1294 	return do_compat_readv(fd, vec, vlen, 0);
1295 }
1296 
1297 static long do_compat_preadv64(unsigned long fd,
1298 				  const struct compat_iovec __user *vec,
1299 				  unsigned long vlen, loff_t pos, rwf_t flags)
1300 {
1301 	struct fd f;
1302 	ssize_t ret;
1303 
1304 	if (pos < 0)
1305 		return -EINVAL;
1306 	f = fdget(fd);
1307 	if (!f.file)
1308 		return -EBADF;
1309 	ret = -ESPIPE;
1310 	if (f.file->f_mode & FMODE_PREAD)
1311 		ret = compat_readv(f.file, vec, vlen, &pos, flags);
1312 	fdput(f);
1313 	return ret;
1314 }
1315 
1316 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1317 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1318 		const struct compat_iovec __user *,vec,
1319 		unsigned long, vlen, loff_t, pos)
1320 {
1321 	return do_compat_preadv64(fd, vec, vlen, pos, 0);
1322 }
1323 #endif
1324 
1325 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1326 		const struct compat_iovec __user *,vec,
1327 		compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1328 {
1329 	loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1330 
1331 	return do_compat_preadv64(fd, vec, vlen, pos, 0);
1332 }
1333 
1334 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1335 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1336 		const struct compat_iovec __user *,vec,
1337 		unsigned long, vlen, loff_t, pos, rwf_t, flags)
1338 {
1339 	if (pos == -1)
1340 		return do_compat_readv(fd, vec, vlen, flags);
1341 
1342 	return do_compat_preadv64(fd, vec, vlen, pos, flags);
1343 }
1344 #endif
1345 
1346 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1347 		const struct compat_iovec __user *,vec,
1348 		compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1349 		rwf_t, flags)
1350 {
1351 	loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1352 
1353 	if (pos == -1)
1354 		return do_compat_readv(fd, vec, vlen, flags);
1355 
1356 	return do_compat_preadv64(fd, vec, vlen, pos, flags);
1357 }
1358 
1359 static size_t compat_writev(struct file *file,
1360 			    const struct compat_iovec __user *vec,
1361 			    unsigned long vlen, loff_t *pos, rwf_t flags)
1362 {
1363 	struct iovec iovstack[UIO_FASTIOV];
1364 	struct iovec *iov = iovstack;
1365 	struct iov_iter iter;
1366 	ssize_t ret;
1367 
1368 	ret = compat_import_iovec(WRITE, vec, vlen, UIO_FASTIOV, &iov, &iter);
1369 	if (ret >= 0) {
1370 		file_start_write(file);
1371 		ret = do_iter_write(file, &iter, pos, flags);
1372 		file_end_write(file);
1373 		kfree(iov);
1374 	}
1375 	if (ret > 0)
1376 		add_wchar(current, ret);
1377 	inc_syscw(current);
1378 	return ret;
1379 }
1380 
1381 static size_t do_compat_writev(compat_ulong_t fd,
1382 				  const struct compat_iovec __user* vec,
1383 				  compat_ulong_t vlen, rwf_t flags)
1384 {
1385 	struct fd f = fdget_pos(fd);
1386 	ssize_t ret;
1387 	loff_t pos;
1388 
1389 	if (!f.file)
1390 		return -EBADF;
1391 	pos = f.file->f_pos;
1392 	ret = compat_writev(f.file, vec, vlen, &pos, flags);
1393 	if (ret >= 0)
1394 		f.file->f_pos = pos;
1395 	fdput_pos(f);
1396 	return ret;
1397 }
1398 
1399 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1400 		const struct compat_iovec __user *, vec,
1401 		compat_ulong_t, vlen)
1402 {
1403 	return do_compat_writev(fd, vec, vlen, 0);
1404 }
1405 
1406 static long do_compat_pwritev64(unsigned long fd,
1407 				   const struct compat_iovec __user *vec,
1408 				   unsigned long vlen, loff_t pos, rwf_t flags)
1409 {
1410 	struct fd f;
1411 	ssize_t ret;
1412 
1413 	if (pos < 0)
1414 		return -EINVAL;
1415 	f = fdget(fd);
1416 	if (!f.file)
1417 		return -EBADF;
1418 	ret = -ESPIPE;
1419 	if (f.file->f_mode & FMODE_PWRITE)
1420 		ret = compat_writev(f.file, vec, vlen, &pos, flags);
1421 	fdput(f);
1422 	return ret;
1423 }
1424 
1425 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1426 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1427 		const struct compat_iovec __user *,vec,
1428 		unsigned long, vlen, loff_t, pos)
1429 {
1430 	return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1431 }
1432 #endif
1433 
1434 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1435 		const struct compat_iovec __user *,vec,
1436 		compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1437 {
1438 	loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1439 
1440 	return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1441 }
1442 
1443 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1444 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1445 		const struct compat_iovec __user *,vec,
1446 		unsigned long, vlen, loff_t, pos, rwf_t, flags)
1447 {
1448 	if (pos == -1)
1449 		return do_compat_writev(fd, vec, vlen, flags);
1450 
1451 	return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1452 }
1453 #endif
1454 
1455 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1456 		const struct compat_iovec __user *,vec,
1457 		compat_ulong_t, vlen, u32, pos_low, u32, pos_high, rwf_t, flags)
1458 {
1459 	loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1460 
1461 	if (pos == -1)
1462 		return do_compat_writev(fd, vec, vlen, flags);
1463 
1464 	return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1465 }
1466 
1467 #endif
1468 
1469 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1470 		  	   size_t count, loff_t max)
1471 {
1472 	struct fd in, out;
1473 	struct inode *in_inode, *out_inode;
1474 	loff_t pos;
1475 	loff_t out_pos;
1476 	ssize_t retval;
1477 	int fl;
1478 
1479 	/*
1480 	 * Get input file, and verify that it is ok..
1481 	 */
1482 	retval = -EBADF;
1483 	in = fdget(in_fd);
1484 	if (!in.file)
1485 		goto out;
1486 	if (!(in.file->f_mode & FMODE_READ))
1487 		goto fput_in;
1488 	retval = -ESPIPE;
1489 	if (!ppos) {
1490 		pos = in.file->f_pos;
1491 	} else {
1492 		pos = *ppos;
1493 		if (!(in.file->f_mode & FMODE_PREAD))
1494 			goto fput_in;
1495 	}
1496 	retval = rw_verify_area(READ, in.file, &pos, count);
1497 	if (retval < 0)
1498 		goto fput_in;
1499 	if (count > MAX_RW_COUNT)
1500 		count =  MAX_RW_COUNT;
1501 
1502 	/*
1503 	 * Get output file, and verify that it is ok..
1504 	 */
1505 	retval = -EBADF;
1506 	out = fdget(out_fd);
1507 	if (!out.file)
1508 		goto fput_in;
1509 	if (!(out.file->f_mode & FMODE_WRITE))
1510 		goto fput_out;
1511 	in_inode = file_inode(in.file);
1512 	out_inode = file_inode(out.file);
1513 	out_pos = out.file->f_pos;
1514 	retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1515 	if (retval < 0)
1516 		goto fput_out;
1517 
1518 	if (!max)
1519 		max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1520 
1521 	if (unlikely(pos + count > max)) {
1522 		retval = -EOVERFLOW;
1523 		if (pos >= max)
1524 			goto fput_out;
1525 		count = max - pos;
1526 	}
1527 
1528 	fl = 0;
1529 #if 0
1530 	/*
1531 	 * We need to debate whether we can enable this or not. The
1532 	 * man page documents EAGAIN return for the output at least,
1533 	 * and the application is arguably buggy if it doesn't expect
1534 	 * EAGAIN on a non-blocking file descriptor.
1535 	 */
1536 	if (in.file->f_flags & O_NONBLOCK)
1537 		fl = SPLICE_F_NONBLOCK;
1538 #endif
1539 	file_start_write(out.file);
1540 	retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1541 	file_end_write(out.file);
1542 
1543 	if (retval > 0) {
1544 		add_rchar(current, retval);
1545 		add_wchar(current, retval);
1546 		fsnotify_access(in.file);
1547 		fsnotify_modify(out.file);
1548 		out.file->f_pos = out_pos;
1549 		if (ppos)
1550 			*ppos = pos;
1551 		else
1552 			in.file->f_pos = pos;
1553 	}
1554 
1555 	inc_syscr(current);
1556 	inc_syscw(current);
1557 	if (pos > max)
1558 		retval = -EOVERFLOW;
1559 
1560 fput_out:
1561 	fdput(out);
1562 fput_in:
1563 	fdput(in);
1564 out:
1565 	return retval;
1566 }
1567 
1568 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1569 {
1570 	loff_t pos;
1571 	off_t off;
1572 	ssize_t ret;
1573 
1574 	if (offset) {
1575 		if (unlikely(get_user(off, offset)))
1576 			return -EFAULT;
1577 		pos = off;
1578 		ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1579 		if (unlikely(put_user(pos, offset)))
1580 			return -EFAULT;
1581 		return ret;
1582 	}
1583 
1584 	return do_sendfile(out_fd, in_fd, NULL, count, 0);
1585 }
1586 
1587 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1588 {
1589 	loff_t pos;
1590 	ssize_t ret;
1591 
1592 	if (offset) {
1593 		if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1594 			return -EFAULT;
1595 		ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1596 		if (unlikely(put_user(pos, offset)))
1597 			return -EFAULT;
1598 		return ret;
1599 	}
1600 
1601 	return do_sendfile(out_fd, in_fd, NULL, count, 0);
1602 }
1603 
1604 #ifdef CONFIG_COMPAT
1605 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1606 		compat_off_t __user *, offset, compat_size_t, count)
1607 {
1608 	loff_t pos;
1609 	off_t off;
1610 	ssize_t ret;
1611 
1612 	if (offset) {
1613 		if (unlikely(get_user(off, offset)))
1614 			return -EFAULT;
1615 		pos = off;
1616 		ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1617 		if (unlikely(put_user(pos, offset)))
1618 			return -EFAULT;
1619 		return ret;
1620 	}
1621 
1622 	return do_sendfile(out_fd, in_fd, NULL, count, 0);
1623 }
1624 
1625 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1626 		compat_loff_t __user *, offset, compat_size_t, count)
1627 {
1628 	loff_t pos;
1629 	ssize_t ret;
1630 
1631 	if (offset) {
1632 		if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1633 			return -EFAULT;
1634 		ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1635 		if (unlikely(put_user(pos, offset)))
1636 			return -EFAULT;
1637 		return ret;
1638 	}
1639 
1640 	return do_sendfile(out_fd, in_fd, NULL, count, 0);
1641 }
1642 #endif
1643 
1644 /**
1645  * generic_copy_file_range - copy data between two files
1646  * @file_in:	file structure to read from
1647  * @pos_in:	file offset to read from
1648  * @file_out:	file structure to write data to
1649  * @pos_out:	file offset to write data to
1650  * @len:	amount of data to copy
1651  * @flags:	copy flags
1652  *
1653  * This is a generic filesystem helper to copy data from one file to another.
1654  * It has no constraints on the source or destination file owners - the files
1655  * can belong to different superblocks and different filesystem types. Short
1656  * copies are allowed.
1657  *
1658  * This should be called from the @file_out filesystem, as per the
1659  * ->copy_file_range() method.
1660  *
1661  * Returns the number of bytes copied or a negative error indicating the
1662  * failure.
1663  */
1664 
1665 ssize_t generic_copy_file_range(struct file *file_in, loff_t pos_in,
1666 				struct file *file_out, loff_t pos_out,
1667 				size_t len, unsigned int flags)
1668 {
1669 	return do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1670 				len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1671 }
1672 EXPORT_SYMBOL(generic_copy_file_range);
1673 
1674 static ssize_t do_copy_file_range(struct file *file_in, loff_t pos_in,
1675 				  struct file *file_out, loff_t pos_out,
1676 				  size_t len, unsigned int flags)
1677 {
1678 	/*
1679 	 * Although we now allow filesystems to handle cross sb copy, passing
1680 	 * a file of the wrong filesystem type to filesystem driver can result
1681 	 * in an attempt to dereference the wrong type of ->private_data, so
1682 	 * avoid doing that until we really have a good reason.  NFS defines
1683 	 * several different file_system_type structures, but they all end up
1684 	 * using the same ->copy_file_range() function pointer.
1685 	 */
1686 	if (file_out->f_op->copy_file_range &&
1687 	    file_out->f_op->copy_file_range == file_in->f_op->copy_file_range)
1688 		return file_out->f_op->copy_file_range(file_in, pos_in,
1689 						       file_out, pos_out,
1690 						       len, flags);
1691 
1692 	return generic_copy_file_range(file_in, pos_in, file_out, pos_out, len,
1693 				       flags);
1694 }
1695 
1696 /*
1697  * copy_file_range() differs from regular file read and write in that it
1698  * specifically allows return partial success.  When it does so is up to
1699  * the copy_file_range method.
1700  */
1701 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1702 			    struct file *file_out, loff_t pos_out,
1703 			    size_t len, unsigned int flags)
1704 {
1705 	ssize_t ret;
1706 
1707 	if (flags != 0)
1708 		return -EINVAL;
1709 
1710 	ret = generic_copy_file_checks(file_in, pos_in, file_out, pos_out, &len,
1711 				       flags);
1712 	if (unlikely(ret))
1713 		return ret;
1714 
1715 	ret = rw_verify_area(READ, file_in, &pos_in, len);
1716 	if (unlikely(ret))
1717 		return ret;
1718 
1719 	ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1720 	if (unlikely(ret))
1721 		return ret;
1722 
1723 	if (len == 0)
1724 		return 0;
1725 
1726 	file_start_write(file_out);
1727 
1728 	/*
1729 	 * Try cloning first, this is supported by more file systems, and
1730 	 * more efficient if both clone and copy are supported (e.g. NFS).
1731 	 */
1732 	if (file_in->f_op->remap_file_range &&
1733 	    file_inode(file_in)->i_sb == file_inode(file_out)->i_sb) {
1734 		loff_t cloned;
1735 
1736 		cloned = file_in->f_op->remap_file_range(file_in, pos_in,
1737 				file_out, pos_out,
1738 				min_t(loff_t, MAX_RW_COUNT, len),
1739 				REMAP_FILE_CAN_SHORTEN);
1740 		if (cloned > 0) {
1741 			ret = cloned;
1742 			goto done;
1743 		}
1744 	}
1745 
1746 	ret = do_copy_file_range(file_in, pos_in, file_out, pos_out, len,
1747 				flags);
1748 	WARN_ON_ONCE(ret == -EOPNOTSUPP);
1749 done:
1750 	if (ret > 0) {
1751 		fsnotify_access(file_in);
1752 		add_rchar(current, ret);
1753 		fsnotify_modify(file_out);
1754 		add_wchar(current, ret);
1755 	}
1756 
1757 	inc_syscr(current);
1758 	inc_syscw(current);
1759 
1760 	file_end_write(file_out);
1761 
1762 	return ret;
1763 }
1764 EXPORT_SYMBOL(vfs_copy_file_range);
1765 
1766 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1767 		int, fd_out, loff_t __user *, off_out,
1768 		size_t, len, unsigned int, flags)
1769 {
1770 	loff_t pos_in;
1771 	loff_t pos_out;
1772 	struct fd f_in;
1773 	struct fd f_out;
1774 	ssize_t ret = -EBADF;
1775 
1776 	f_in = fdget(fd_in);
1777 	if (!f_in.file)
1778 		goto out2;
1779 
1780 	f_out = fdget(fd_out);
1781 	if (!f_out.file)
1782 		goto out1;
1783 
1784 	ret = -EFAULT;
1785 	if (off_in) {
1786 		if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1787 			goto out;
1788 	} else {
1789 		pos_in = f_in.file->f_pos;
1790 	}
1791 
1792 	if (off_out) {
1793 		if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1794 			goto out;
1795 	} else {
1796 		pos_out = f_out.file->f_pos;
1797 	}
1798 
1799 	ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1800 				  flags);
1801 	if (ret > 0) {
1802 		pos_in += ret;
1803 		pos_out += ret;
1804 
1805 		if (off_in) {
1806 			if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1807 				ret = -EFAULT;
1808 		} else {
1809 			f_in.file->f_pos = pos_in;
1810 		}
1811 
1812 		if (off_out) {
1813 			if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1814 				ret = -EFAULT;
1815 		} else {
1816 			f_out.file->f_pos = pos_out;
1817 		}
1818 	}
1819 
1820 out:
1821 	fdput(f_out);
1822 out1:
1823 	fdput(f_in);
1824 out2:
1825 	return ret;
1826 }
1827 
1828 static int remap_verify_area(struct file *file, loff_t pos, loff_t len,
1829 			     bool write)
1830 {
1831 	struct inode *inode = file_inode(file);
1832 
1833 	if (unlikely(pos < 0 || len < 0))
1834 		return -EINVAL;
1835 
1836 	 if (unlikely((loff_t) (pos + len) < 0))
1837 		return -EINVAL;
1838 
1839 	if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1840 		loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1841 		int retval;
1842 
1843 		retval = locks_mandatory_area(inode, file, pos, end,
1844 				write ? F_WRLCK : F_RDLCK);
1845 		if (retval < 0)
1846 			return retval;
1847 	}
1848 
1849 	return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1850 }
1851 /*
1852  * Ensure that we don't remap a partial EOF block in the middle of something
1853  * else.  Assume that the offsets have already been checked for block
1854  * alignment.
1855  *
1856  * For clone we only link a partial EOF block above or at the destination file's
1857  * EOF.  For deduplication we accept a partial EOF block only if it ends at the
1858  * destination file's EOF (can not link it into the middle of a file).
1859  *
1860  * Shorten the request if possible.
1861  */
1862 static int generic_remap_check_len(struct inode *inode_in,
1863 				   struct inode *inode_out,
1864 				   loff_t pos_out,
1865 				   loff_t *len,
1866 				   unsigned int remap_flags)
1867 {
1868 	u64 blkmask = i_blocksize(inode_in) - 1;
1869 	loff_t new_len = *len;
1870 
1871 	if ((*len & blkmask) == 0)
1872 		return 0;
1873 
1874 	if (pos_out + *len < i_size_read(inode_out))
1875 		new_len &= ~blkmask;
1876 
1877 	if (new_len == *len)
1878 		return 0;
1879 
1880 	if (remap_flags & REMAP_FILE_CAN_SHORTEN) {
1881 		*len = new_len;
1882 		return 0;
1883 	}
1884 
1885 	return (remap_flags & REMAP_FILE_DEDUP) ? -EBADE : -EINVAL;
1886 }
1887 
1888 /* Read a page's worth of file data into the page cache. */
1889 static struct page *vfs_dedupe_get_page(struct inode *inode, loff_t offset)
1890 {
1891 	struct page *page;
1892 
1893 	page = read_mapping_page(inode->i_mapping, offset >> PAGE_SHIFT, NULL);
1894 	if (IS_ERR(page))
1895 		return page;
1896 	if (!PageUptodate(page)) {
1897 		put_page(page);
1898 		return ERR_PTR(-EIO);
1899 	}
1900 	return page;
1901 }
1902 
1903 /*
1904  * Lock two pages, ensuring that we lock in offset order if the pages are from
1905  * the same file.
1906  */
1907 static void vfs_lock_two_pages(struct page *page1, struct page *page2)
1908 {
1909 	/* Always lock in order of increasing index. */
1910 	if (page1->index > page2->index)
1911 		swap(page1, page2);
1912 
1913 	lock_page(page1);
1914 	if (page1 != page2)
1915 		lock_page(page2);
1916 }
1917 
1918 /* Unlock two pages, being careful not to unlock the same page twice. */
1919 static void vfs_unlock_two_pages(struct page *page1, struct page *page2)
1920 {
1921 	unlock_page(page1);
1922 	if (page1 != page2)
1923 		unlock_page(page2);
1924 }
1925 
1926 /*
1927  * Compare extents of two files to see if they are the same.
1928  * Caller must have locked both inodes to prevent write races.
1929  */
1930 static int vfs_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
1931 					 struct inode *dest, loff_t destoff,
1932 					 loff_t len, bool *is_same)
1933 {
1934 	loff_t src_poff;
1935 	loff_t dest_poff;
1936 	void *src_addr;
1937 	void *dest_addr;
1938 	struct page *src_page;
1939 	struct page *dest_page;
1940 	loff_t cmp_len;
1941 	bool same;
1942 	int error;
1943 
1944 	error = -EINVAL;
1945 	same = true;
1946 	while (len) {
1947 		src_poff = srcoff & (PAGE_SIZE - 1);
1948 		dest_poff = destoff & (PAGE_SIZE - 1);
1949 		cmp_len = min(PAGE_SIZE - src_poff,
1950 			      PAGE_SIZE - dest_poff);
1951 		cmp_len = min(cmp_len, len);
1952 		if (cmp_len <= 0)
1953 			goto out_error;
1954 
1955 		src_page = vfs_dedupe_get_page(src, srcoff);
1956 		if (IS_ERR(src_page)) {
1957 			error = PTR_ERR(src_page);
1958 			goto out_error;
1959 		}
1960 		dest_page = vfs_dedupe_get_page(dest, destoff);
1961 		if (IS_ERR(dest_page)) {
1962 			error = PTR_ERR(dest_page);
1963 			put_page(src_page);
1964 			goto out_error;
1965 		}
1966 
1967 		vfs_lock_two_pages(src_page, dest_page);
1968 
1969 		/*
1970 		 * Now that we've locked both pages, make sure they're still
1971 		 * mapped to the file data we're interested in.  If not,
1972 		 * someone is invalidating pages on us and we lose.
1973 		 */
1974 		if (!PageUptodate(src_page) || !PageUptodate(dest_page) ||
1975 		    src_page->mapping != src->i_mapping ||
1976 		    dest_page->mapping != dest->i_mapping) {
1977 			same = false;
1978 			goto unlock;
1979 		}
1980 
1981 		src_addr = kmap_atomic(src_page);
1982 		dest_addr = kmap_atomic(dest_page);
1983 
1984 		flush_dcache_page(src_page);
1985 		flush_dcache_page(dest_page);
1986 
1987 		if (memcmp(src_addr + src_poff, dest_addr + dest_poff, cmp_len))
1988 			same = false;
1989 
1990 		kunmap_atomic(dest_addr);
1991 		kunmap_atomic(src_addr);
1992 unlock:
1993 		vfs_unlock_two_pages(src_page, dest_page);
1994 		put_page(dest_page);
1995 		put_page(src_page);
1996 
1997 		if (!same)
1998 			break;
1999 
2000 		srcoff += cmp_len;
2001 		destoff += cmp_len;
2002 		len -= cmp_len;
2003 	}
2004 
2005 	*is_same = same;
2006 	return 0;
2007 
2008 out_error:
2009 	return error;
2010 }
2011 
2012 /*
2013  * Check that the two inodes are eligible for cloning, the ranges make
2014  * sense, and then flush all dirty data.  Caller must ensure that the
2015  * inodes have been locked against any other modifications.
2016  *
2017  * If there's an error, then the usual negative error code is returned.
2018  * Otherwise returns 0 with *len set to the request length.
2019  */
2020 int generic_remap_file_range_prep(struct file *file_in, loff_t pos_in,
2021 				  struct file *file_out, loff_t pos_out,
2022 				  loff_t *len, unsigned int remap_flags)
2023 {
2024 	struct inode *inode_in = file_inode(file_in);
2025 	struct inode *inode_out = file_inode(file_out);
2026 	bool same_inode = (inode_in == inode_out);
2027 	int ret;
2028 
2029 	/* Don't touch certain kinds of inodes */
2030 	if (IS_IMMUTABLE(inode_out))
2031 		return -EPERM;
2032 
2033 	if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
2034 		return -ETXTBSY;
2035 
2036 	/* Don't reflink dirs, pipes, sockets... */
2037 	if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
2038 		return -EISDIR;
2039 	if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
2040 		return -EINVAL;
2041 
2042 	/* Zero length dedupe exits immediately; reflink goes to EOF. */
2043 	if (*len == 0) {
2044 		loff_t isize = i_size_read(inode_in);
2045 
2046 		if ((remap_flags & REMAP_FILE_DEDUP) || pos_in == isize)
2047 			return 0;
2048 		if (pos_in > isize)
2049 			return -EINVAL;
2050 		*len = isize - pos_in;
2051 		if (*len == 0)
2052 			return 0;
2053 	}
2054 
2055 	/* Check that we don't violate system file offset limits. */
2056 	ret = generic_remap_checks(file_in, pos_in, file_out, pos_out, len,
2057 			remap_flags);
2058 	if (ret)
2059 		return ret;
2060 
2061 	/* Wait for the completion of any pending IOs on both files */
2062 	inode_dio_wait(inode_in);
2063 	if (!same_inode)
2064 		inode_dio_wait(inode_out);
2065 
2066 	ret = filemap_write_and_wait_range(inode_in->i_mapping,
2067 			pos_in, pos_in + *len - 1);
2068 	if (ret)
2069 		return ret;
2070 
2071 	ret = filemap_write_and_wait_range(inode_out->i_mapping,
2072 			pos_out, pos_out + *len - 1);
2073 	if (ret)
2074 		return ret;
2075 
2076 	/*
2077 	 * Check that the extents are the same.
2078 	 */
2079 	if (remap_flags & REMAP_FILE_DEDUP) {
2080 		bool		is_same = false;
2081 
2082 		ret = vfs_dedupe_file_range_compare(inode_in, pos_in,
2083 				inode_out, pos_out, *len, &is_same);
2084 		if (ret)
2085 			return ret;
2086 		if (!is_same)
2087 			return -EBADE;
2088 	}
2089 
2090 	ret = generic_remap_check_len(inode_in, inode_out, pos_out, len,
2091 			remap_flags);
2092 	if (ret)
2093 		return ret;
2094 
2095 	/* If can't alter the file contents, we're done. */
2096 	if (!(remap_flags & REMAP_FILE_DEDUP))
2097 		ret = file_modified(file_out);
2098 
2099 	return ret;
2100 }
2101 EXPORT_SYMBOL(generic_remap_file_range_prep);
2102 
2103 loff_t do_clone_file_range(struct file *file_in, loff_t pos_in,
2104 			   struct file *file_out, loff_t pos_out,
2105 			   loff_t len, unsigned int remap_flags)
2106 {
2107 	loff_t ret;
2108 
2109 	WARN_ON_ONCE(remap_flags & REMAP_FILE_DEDUP);
2110 
2111 	/*
2112 	 * FICLONE/FICLONERANGE ioctls enforce that src and dest files are on
2113 	 * the same mount. Practically, they only need to be on the same file
2114 	 * system.
2115 	 */
2116 	if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb)
2117 		return -EXDEV;
2118 
2119 	ret = generic_file_rw_checks(file_in, file_out);
2120 	if (ret < 0)
2121 		return ret;
2122 
2123 	if (!file_in->f_op->remap_file_range)
2124 		return -EOPNOTSUPP;
2125 
2126 	ret = remap_verify_area(file_in, pos_in, len, false);
2127 	if (ret)
2128 		return ret;
2129 
2130 	ret = remap_verify_area(file_out, pos_out, len, true);
2131 	if (ret)
2132 		return ret;
2133 
2134 	ret = file_in->f_op->remap_file_range(file_in, pos_in,
2135 			file_out, pos_out, len, remap_flags);
2136 	if (ret < 0)
2137 		return ret;
2138 
2139 	fsnotify_access(file_in);
2140 	fsnotify_modify(file_out);
2141 	return ret;
2142 }
2143 EXPORT_SYMBOL(do_clone_file_range);
2144 
2145 loff_t vfs_clone_file_range(struct file *file_in, loff_t pos_in,
2146 			    struct file *file_out, loff_t pos_out,
2147 			    loff_t len, unsigned int remap_flags)
2148 {
2149 	loff_t ret;
2150 
2151 	file_start_write(file_out);
2152 	ret = do_clone_file_range(file_in, pos_in, file_out, pos_out, len,
2153 				  remap_flags);
2154 	file_end_write(file_out);
2155 
2156 	return ret;
2157 }
2158 EXPORT_SYMBOL(vfs_clone_file_range);
2159 
2160 /* Check whether we are allowed to dedupe the destination file */
2161 static bool allow_file_dedupe(struct file *file)
2162 {
2163 	if (capable(CAP_SYS_ADMIN))
2164 		return true;
2165 	if (file->f_mode & FMODE_WRITE)
2166 		return true;
2167 	if (uid_eq(current_fsuid(), file_inode(file)->i_uid))
2168 		return true;
2169 	if (!inode_permission(file_inode(file), MAY_WRITE))
2170 		return true;
2171 	return false;
2172 }
2173 
2174 loff_t vfs_dedupe_file_range_one(struct file *src_file, loff_t src_pos,
2175 				 struct file *dst_file, loff_t dst_pos,
2176 				 loff_t len, unsigned int remap_flags)
2177 {
2178 	loff_t ret;
2179 
2180 	WARN_ON_ONCE(remap_flags & ~(REMAP_FILE_DEDUP |
2181 				     REMAP_FILE_CAN_SHORTEN));
2182 
2183 	ret = mnt_want_write_file(dst_file);
2184 	if (ret)
2185 		return ret;
2186 
2187 	ret = remap_verify_area(dst_file, dst_pos, len, true);
2188 	if (ret < 0)
2189 		goto out_drop_write;
2190 
2191 	ret = -EPERM;
2192 	if (!allow_file_dedupe(dst_file))
2193 		goto out_drop_write;
2194 
2195 	ret = -EXDEV;
2196 	if (src_file->f_path.mnt != dst_file->f_path.mnt)
2197 		goto out_drop_write;
2198 
2199 	ret = -EISDIR;
2200 	if (S_ISDIR(file_inode(dst_file)->i_mode))
2201 		goto out_drop_write;
2202 
2203 	ret = -EINVAL;
2204 	if (!dst_file->f_op->remap_file_range)
2205 		goto out_drop_write;
2206 
2207 	if (len == 0) {
2208 		ret = 0;
2209 		goto out_drop_write;
2210 	}
2211 
2212 	ret = dst_file->f_op->remap_file_range(src_file, src_pos, dst_file,
2213 			dst_pos, len, remap_flags | REMAP_FILE_DEDUP);
2214 out_drop_write:
2215 	mnt_drop_write_file(dst_file);
2216 
2217 	return ret;
2218 }
2219 EXPORT_SYMBOL(vfs_dedupe_file_range_one);
2220 
2221 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
2222 {
2223 	struct file_dedupe_range_info *info;
2224 	struct inode *src = file_inode(file);
2225 	u64 off;
2226 	u64 len;
2227 	int i;
2228 	int ret;
2229 	u16 count = same->dest_count;
2230 	loff_t deduped;
2231 
2232 	if (!(file->f_mode & FMODE_READ))
2233 		return -EINVAL;
2234 
2235 	if (same->reserved1 || same->reserved2)
2236 		return -EINVAL;
2237 
2238 	off = same->src_offset;
2239 	len = same->src_length;
2240 
2241 	if (S_ISDIR(src->i_mode))
2242 		return -EISDIR;
2243 
2244 	if (!S_ISREG(src->i_mode))
2245 		return -EINVAL;
2246 
2247 	if (!file->f_op->remap_file_range)
2248 		return -EOPNOTSUPP;
2249 
2250 	ret = remap_verify_area(file, off, len, false);
2251 	if (ret < 0)
2252 		return ret;
2253 	ret = 0;
2254 
2255 	if (off + len > i_size_read(src))
2256 		return -EINVAL;
2257 
2258 	/* Arbitrary 1G limit on a single dedupe request, can be raised. */
2259 	len = min_t(u64, len, 1 << 30);
2260 
2261 	/* pre-format output fields to sane values */
2262 	for (i = 0; i < count; i++) {
2263 		same->info[i].bytes_deduped = 0ULL;
2264 		same->info[i].status = FILE_DEDUPE_RANGE_SAME;
2265 	}
2266 
2267 	for (i = 0, info = same->info; i < count; i++, info++) {
2268 		struct fd dst_fd = fdget(info->dest_fd);
2269 		struct file *dst_file = dst_fd.file;
2270 
2271 		if (!dst_file) {
2272 			info->status = -EBADF;
2273 			goto next_loop;
2274 		}
2275 
2276 		if (info->reserved) {
2277 			info->status = -EINVAL;
2278 			goto next_fdput;
2279 		}
2280 
2281 		deduped = vfs_dedupe_file_range_one(file, off, dst_file,
2282 						    info->dest_offset, len,
2283 						    REMAP_FILE_CAN_SHORTEN);
2284 		if (deduped == -EBADE)
2285 			info->status = FILE_DEDUPE_RANGE_DIFFERS;
2286 		else if (deduped < 0)
2287 			info->status = deduped;
2288 		else
2289 			info->bytes_deduped = len;
2290 
2291 next_fdput:
2292 		fdput(dst_fd);
2293 next_loop:
2294 		if (fatal_signal_pending(current))
2295 			break;
2296 	}
2297 	return ret;
2298 }
2299 EXPORT_SYMBOL(vfs_dedupe_file_range);
2300