xref: /linux/fs/read_write.c (revision 9e56ff53b4115875667760445b028357848b4748)
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	= filemap_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  * @maxsize:	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 default_llseek(struct file *file, loff_t offset, int whence)
231 {
232 	struct inode *inode = file_inode(file);
233 	loff_t retval;
234 
235 	inode_lock(inode);
236 	switch (whence) {
237 		case SEEK_END:
238 			offset += i_size_read(inode);
239 			break;
240 		case SEEK_CUR:
241 			if (offset == 0) {
242 				retval = file->f_pos;
243 				goto out;
244 			}
245 			offset += file->f_pos;
246 			break;
247 		case SEEK_DATA:
248 			/*
249 			 * In the generic case the entire file is data, so as
250 			 * long as offset isn't at the end of the file then the
251 			 * offset is data.
252 			 */
253 			if (offset >= inode->i_size) {
254 				retval = -ENXIO;
255 				goto out;
256 			}
257 			break;
258 		case SEEK_HOLE:
259 			/*
260 			 * There is a virtual hole at the end of the file, so
261 			 * as long as offset isn't i_size or larger, return
262 			 * i_size.
263 			 */
264 			if (offset >= inode->i_size) {
265 				retval = -ENXIO;
266 				goto out;
267 			}
268 			offset = inode->i_size;
269 			break;
270 	}
271 	retval = -EINVAL;
272 	if (offset >= 0 || unsigned_offsets(file)) {
273 		if (offset != file->f_pos) {
274 			file->f_pos = offset;
275 			file->f_version = 0;
276 		}
277 		retval = offset;
278 	}
279 out:
280 	inode_unlock(inode);
281 	return retval;
282 }
283 EXPORT_SYMBOL(default_llseek);
284 
285 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
286 {
287 	if (!(file->f_mode & FMODE_LSEEK))
288 		return -ESPIPE;
289 	return file->f_op->llseek(file, offset, whence);
290 }
291 EXPORT_SYMBOL(vfs_llseek);
292 
293 static off_t ksys_lseek(unsigned int fd, off_t offset, unsigned int whence)
294 {
295 	off_t retval;
296 	struct fd f = fdget_pos(fd);
297 	if (!f.file)
298 		return -EBADF;
299 
300 	retval = -EINVAL;
301 	if (whence <= SEEK_MAX) {
302 		loff_t res = vfs_llseek(f.file, offset, whence);
303 		retval = res;
304 		if (res != (loff_t)retval)
305 			retval = -EOVERFLOW;	/* LFS: should only happen on 32 bit platforms */
306 	}
307 	fdput_pos(f);
308 	return retval;
309 }
310 
311 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
312 {
313 	return ksys_lseek(fd, offset, whence);
314 }
315 
316 #ifdef CONFIG_COMPAT
317 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
318 {
319 	return ksys_lseek(fd, offset, whence);
320 }
321 #endif
322 
323 #if !defined(CONFIG_64BIT) || defined(CONFIG_COMPAT) || \
324 	defined(__ARCH_WANT_SYS_LLSEEK)
325 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
326 		unsigned long, offset_low, loff_t __user *, result,
327 		unsigned int, whence)
328 {
329 	int retval;
330 	struct fd f = fdget_pos(fd);
331 	loff_t offset;
332 
333 	if (!f.file)
334 		return -EBADF;
335 
336 	retval = -EINVAL;
337 	if (whence > SEEK_MAX)
338 		goto out_putf;
339 
340 	offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
341 			whence);
342 
343 	retval = (int)offset;
344 	if (offset >= 0) {
345 		retval = -EFAULT;
346 		if (!copy_to_user(result, &offset, sizeof(offset)))
347 			retval = 0;
348 	}
349 out_putf:
350 	fdput_pos(f);
351 	return retval;
352 }
353 #endif
354 
355 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
356 {
357 	int mask = read_write == READ ? MAY_READ : MAY_WRITE;
358 	int ret;
359 
360 	if (unlikely((ssize_t) count < 0))
361 		return -EINVAL;
362 
363 	if (ppos) {
364 		loff_t pos = *ppos;
365 
366 		if (unlikely(pos < 0)) {
367 			if (!unsigned_offsets(file))
368 				return -EINVAL;
369 			if (count >= -pos) /* both values are in 0..LLONG_MAX */
370 				return -EOVERFLOW;
371 		} else if (unlikely((loff_t) (pos + count) < 0)) {
372 			if (!unsigned_offsets(file))
373 				return -EINVAL;
374 		}
375 	}
376 
377 	ret = security_file_permission(file, mask);
378 	if (ret)
379 		return ret;
380 
381 	return fsnotify_file_area_perm(file, mask, ppos, count);
382 }
383 EXPORT_SYMBOL(rw_verify_area);
384 
385 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
386 {
387 	struct kiocb kiocb;
388 	struct iov_iter iter;
389 	ssize_t ret;
390 
391 	init_sync_kiocb(&kiocb, filp);
392 	kiocb.ki_pos = (ppos ? *ppos : 0);
393 	iov_iter_ubuf(&iter, ITER_DEST, buf, len);
394 
395 	ret = call_read_iter(filp, &kiocb, &iter);
396 	BUG_ON(ret == -EIOCBQUEUED);
397 	if (ppos)
398 		*ppos = kiocb.ki_pos;
399 	return ret;
400 }
401 
402 static int warn_unsupported(struct file *file, const char *op)
403 {
404 	pr_warn_ratelimited(
405 		"kernel %s not supported for file %pD4 (pid: %d comm: %.20s)\n",
406 		op, file, current->pid, current->comm);
407 	return -EINVAL;
408 }
409 
410 ssize_t __kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
411 {
412 	struct kvec iov = {
413 		.iov_base	= buf,
414 		.iov_len	= min_t(size_t, count, MAX_RW_COUNT),
415 	};
416 	struct kiocb kiocb;
417 	struct iov_iter iter;
418 	ssize_t ret;
419 
420 	if (WARN_ON_ONCE(!(file->f_mode & FMODE_READ)))
421 		return -EINVAL;
422 	if (!(file->f_mode & FMODE_CAN_READ))
423 		return -EINVAL;
424 	/*
425 	 * Also fail if ->read_iter and ->read are both wired up as that
426 	 * implies very convoluted semantics.
427 	 */
428 	if (unlikely(!file->f_op->read_iter || file->f_op->read))
429 		return warn_unsupported(file, "read");
430 
431 	init_sync_kiocb(&kiocb, file);
432 	kiocb.ki_pos = pos ? *pos : 0;
433 	iov_iter_kvec(&iter, ITER_DEST, &iov, 1, iov.iov_len);
434 	ret = file->f_op->read_iter(&kiocb, &iter);
435 	if (ret > 0) {
436 		if (pos)
437 			*pos = kiocb.ki_pos;
438 		fsnotify_access(file);
439 		add_rchar(current, ret);
440 	}
441 	inc_syscr(current);
442 	return ret;
443 }
444 
445 ssize_t kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
446 {
447 	ssize_t ret;
448 
449 	ret = rw_verify_area(READ, file, pos, count);
450 	if (ret)
451 		return ret;
452 	return __kernel_read(file, buf, count, pos);
453 }
454 EXPORT_SYMBOL(kernel_read);
455 
456 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
457 {
458 	ssize_t ret;
459 
460 	if (!(file->f_mode & FMODE_READ))
461 		return -EBADF;
462 	if (!(file->f_mode & FMODE_CAN_READ))
463 		return -EINVAL;
464 	if (unlikely(!access_ok(buf, count)))
465 		return -EFAULT;
466 
467 	ret = rw_verify_area(READ, file, pos, count);
468 	if (ret)
469 		return ret;
470 	if (count > MAX_RW_COUNT)
471 		count =  MAX_RW_COUNT;
472 
473 	if (file->f_op->read)
474 		ret = file->f_op->read(file, buf, count, pos);
475 	else if (file->f_op->read_iter)
476 		ret = new_sync_read(file, buf, count, pos);
477 	else
478 		ret = -EINVAL;
479 	if (ret > 0) {
480 		fsnotify_access(file);
481 		add_rchar(current, ret);
482 	}
483 	inc_syscr(current);
484 	return ret;
485 }
486 
487 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
488 {
489 	struct kiocb kiocb;
490 	struct iov_iter iter;
491 	ssize_t ret;
492 
493 	init_sync_kiocb(&kiocb, filp);
494 	kiocb.ki_pos = (ppos ? *ppos : 0);
495 	iov_iter_ubuf(&iter, ITER_SOURCE, (void __user *)buf, len);
496 
497 	ret = call_write_iter(filp, &kiocb, &iter);
498 	BUG_ON(ret == -EIOCBQUEUED);
499 	if (ret > 0 && ppos)
500 		*ppos = kiocb.ki_pos;
501 	return ret;
502 }
503 
504 /* caller is responsible for file_start_write/file_end_write */
505 ssize_t __kernel_write_iter(struct file *file, struct iov_iter *from, loff_t *pos)
506 {
507 	struct kiocb kiocb;
508 	ssize_t ret;
509 
510 	if (WARN_ON_ONCE(!(file->f_mode & FMODE_WRITE)))
511 		return -EBADF;
512 	if (!(file->f_mode & FMODE_CAN_WRITE))
513 		return -EINVAL;
514 	/*
515 	 * Also fail if ->write_iter and ->write are both wired up as that
516 	 * implies very convoluted semantics.
517 	 */
518 	if (unlikely(!file->f_op->write_iter || file->f_op->write))
519 		return warn_unsupported(file, "write");
520 
521 	init_sync_kiocb(&kiocb, file);
522 	kiocb.ki_pos = pos ? *pos : 0;
523 	ret = file->f_op->write_iter(&kiocb, from);
524 	if (ret > 0) {
525 		if (pos)
526 			*pos = kiocb.ki_pos;
527 		fsnotify_modify(file);
528 		add_wchar(current, ret);
529 	}
530 	inc_syscw(current);
531 	return ret;
532 }
533 
534 /* caller is responsible for file_start_write/file_end_write */
535 ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t *pos)
536 {
537 	struct kvec iov = {
538 		.iov_base	= (void *)buf,
539 		.iov_len	= min_t(size_t, count, MAX_RW_COUNT),
540 	};
541 	struct iov_iter iter;
542 	iov_iter_kvec(&iter, ITER_SOURCE, &iov, 1, iov.iov_len);
543 	return __kernel_write_iter(file, &iter, pos);
544 }
545 /*
546  * This "EXPORT_SYMBOL_GPL()" is more of a "EXPORT_SYMBOL_DONTUSE()",
547  * but autofs is one of the few internal kernel users that actually
548  * wants this _and_ can be built as a module. So we need to export
549  * this symbol for autofs, even though it really isn't appropriate
550  * for any other kernel modules.
551  */
552 EXPORT_SYMBOL_GPL(__kernel_write);
553 
554 ssize_t kernel_write(struct file *file, const void *buf, size_t count,
555 			    loff_t *pos)
556 {
557 	ssize_t ret;
558 
559 	ret = rw_verify_area(WRITE, file, pos, count);
560 	if (ret)
561 		return ret;
562 
563 	file_start_write(file);
564 	ret =  __kernel_write(file, buf, count, pos);
565 	file_end_write(file);
566 	return ret;
567 }
568 EXPORT_SYMBOL(kernel_write);
569 
570 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
571 {
572 	ssize_t ret;
573 
574 	if (!(file->f_mode & FMODE_WRITE))
575 		return -EBADF;
576 	if (!(file->f_mode & FMODE_CAN_WRITE))
577 		return -EINVAL;
578 	if (unlikely(!access_ok(buf, count)))
579 		return -EFAULT;
580 
581 	ret = rw_verify_area(WRITE, file, pos, count);
582 	if (ret)
583 		return ret;
584 	if (count > MAX_RW_COUNT)
585 		count =  MAX_RW_COUNT;
586 	file_start_write(file);
587 	if (file->f_op->write)
588 		ret = file->f_op->write(file, buf, count, pos);
589 	else if (file->f_op->write_iter)
590 		ret = new_sync_write(file, buf, count, pos);
591 	else
592 		ret = -EINVAL;
593 	if (ret > 0) {
594 		fsnotify_modify(file);
595 		add_wchar(current, ret);
596 	}
597 	inc_syscw(current);
598 	file_end_write(file);
599 	return ret;
600 }
601 
602 /* file_ppos returns &file->f_pos or NULL if file is stream */
603 static inline loff_t *file_ppos(struct file *file)
604 {
605 	return file->f_mode & FMODE_STREAM ? NULL : &file->f_pos;
606 }
607 
608 ssize_t ksys_read(unsigned int fd, char __user *buf, size_t count)
609 {
610 	struct fd f = fdget_pos(fd);
611 	ssize_t ret = -EBADF;
612 
613 	if (f.file) {
614 		loff_t pos, *ppos = file_ppos(f.file);
615 		if (ppos) {
616 			pos = *ppos;
617 			ppos = &pos;
618 		}
619 		ret = vfs_read(f.file, buf, count, ppos);
620 		if (ret >= 0 && ppos)
621 			f.file->f_pos = pos;
622 		fdput_pos(f);
623 	}
624 	return ret;
625 }
626 
627 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
628 {
629 	return ksys_read(fd, buf, count);
630 }
631 
632 ssize_t ksys_write(unsigned int fd, const char __user *buf, size_t count)
633 {
634 	struct fd f = fdget_pos(fd);
635 	ssize_t ret = -EBADF;
636 
637 	if (f.file) {
638 		loff_t pos, *ppos = file_ppos(f.file);
639 		if (ppos) {
640 			pos = *ppos;
641 			ppos = &pos;
642 		}
643 		ret = vfs_write(f.file, buf, count, ppos);
644 		if (ret >= 0 && ppos)
645 			f.file->f_pos = pos;
646 		fdput_pos(f);
647 	}
648 
649 	return ret;
650 }
651 
652 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
653 		size_t, count)
654 {
655 	return ksys_write(fd, buf, count);
656 }
657 
658 ssize_t ksys_pread64(unsigned int fd, char __user *buf, size_t count,
659 		     loff_t pos)
660 {
661 	struct fd f;
662 	ssize_t ret = -EBADF;
663 
664 	if (pos < 0)
665 		return -EINVAL;
666 
667 	f = fdget(fd);
668 	if (f.file) {
669 		ret = -ESPIPE;
670 		if (f.file->f_mode & FMODE_PREAD)
671 			ret = vfs_read(f.file, buf, count, &pos);
672 		fdput(f);
673 	}
674 
675 	return ret;
676 }
677 
678 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
679 			size_t, count, loff_t, pos)
680 {
681 	return ksys_pread64(fd, buf, count, pos);
682 }
683 
684 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_PREAD64)
685 COMPAT_SYSCALL_DEFINE5(pread64, unsigned int, fd, char __user *, buf,
686 		       size_t, count, compat_arg_u64_dual(pos))
687 {
688 	return ksys_pread64(fd, buf, count, compat_arg_u64_glue(pos));
689 }
690 #endif
691 
692 ssize_t ksys_pwrite64(unsigned int fd, const char __user *buf,
693 		      size_t count, loff_t pos)
694 {
695 	struct fd f;
696 	ssize_t ret = -EBADF;
697 
698 	if (pos < 0)
699 		return -EINVAL;
700 
701 	f = fdget(fd);
702 	if (f.file) {
703 		ret = -ESPIPE;
704 		if (f.file->f_mode & FMODE_PWRITE)
705 			ret = vfs_write(f.file, buf, count, &pos);
706 		fdput(f);
707 	}
708 
709 	return ret;
710 }
711 
712 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
713 			 size_t, count, loff_t, pos)
714 {
715 	return ksys_pwrite64(fd, buf, count, pos);
716 }
717 
718 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_PWRITE64)
719 COMPAT_SYSCALL_DEFINE5(pwrite64, unsigned int, fd, const char __user *, buf,
720 		       size_t, count, compat_arg_u64_dual(pos))
721 {
722 	return ksys_pwrite64(fd, buf, count, compat_arg_u64_glue(pos));
723 }
724 #endif
725 
726 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
727 		loff_t *ppos, int type, rwf_t flags)
728 {
729 	struct kiocb kiocb;
730 	ssize_t ret;
731 
732 	init_sync_kiocb(&kiocb, filp);
733 	ret = kiocb_set_rw_flags(&kiocb, flags);
734 	if (ret)
735 		return ret;
736 	kiocb.ki_pos = (ppos ? *ppos : 0);
737 
738 	if (type == READ)
739 		ret = call_read_iter(filp, &kiocb, iter);
740 	else
741 		ret = call_write_iter(filp, &kiocb, iter);
742 	BUG_ON(ret == -EIOCBQUEUED);
743 	if (ppos)
744 		*ppos = kiocb.ki_pos;
745 	return ret;
746 }
747 
748 /* Do it by hand, with file-ops */
749 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
750 		loff_t *ppos, int type, rwf_t flags)
751 {
752 	ssize_t ret = 0;
753 
754 	if (flags & ~RWF_HIPRI)
755 		return -EOPNOTSUPP;
756 
757 	while (iov_iter_count(iter)) {
758 		ssize_t nr;
759 
760 		if (type == READ) {
761 			nr = filp->f_op->read(filp, iter_iov_addr(iter),
762 						iter_iov_len(iter), ppos);
763 		} else {
764 			nr = filp->f_op->write(filp, iter_iov_addr(iter),
765 						iter_iov_len(iter), ppos);
766 		}
767 
768 		if (nr < 0) {
769 			if (!ret)
770 				ret = nr;
771 			break;
772 		}
773 		ret += nr;
774 		if (nr != iter_iov_len(iter))
775 			break;
776 		iov_iter_advance(iter, nr);
777 	}
778 
779 	return ret;
780 }
781 
782 ssize_t vfs_iocb_iter_read(struct file *file, struct kiocb *iocb,
783 			   struct iov_iter *iter)
784 {
785 	size_t tot_len;
786 	ssize_t ret = 0;
787 
788 	if (!file->f_op->read_iter)
789 		return -EINVAL;
790 	if (!(file->f_mode & FMODE_READ))
791 		return -EBADF;
792 	if (!(file->f_mode & FMODE_CAN_READ))
793 		return -EINVAL;
794 
795 	tot_len = iov_iter_count(iter);
796 	if (!tot_len)
797 		goto out;
798 	ret = rw_verify_area(READ, file, &iocb->ki_pos, tot_len);
799 	if (ret < 0)
800 		return ret;
801 
802 	ret = call_read_iter(file, iocb, iter);
803 out:
804 	if (ret >= 0)
805 		fsnotify_access(file);
806 	return ret;
807 }
808 EXPORT_SYMBOL(vfs_iocb_iter_read);
809 
810 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos,
811 		      rwf_t flags)
812 {
813 	size_t tot_len;
814 	ssize_t ret = 0;
815 
816 	if (!file->f_op->read_iter)
817 		return -EINVAL;
818 	if (!(file->f_mode & FMODE_READ))
819 		return -EBADF;
820 	if (!(file->f_mode & FMODE_CAN_READ))
821 		return -EINVAL;
822 
823 	tot_len = iov_iter_count(iter);
824 	if (!tot_len)
825 		goto out;
826 	ret = rw_verify_area(READ, file, ppos, tot_len);
827 	if (ret < 0)
828 		return ret;
829 
830 	ret = do_iter_readv_writev(file, iter, ppos, READ, flags);
831 out:
832 	if (ret >= 0)
833 		fsnotify_access(file);
834 	return ret;
835 }
836 EXPORT_SYMBOL(vfs_iter_read);
837 
838 /*
839  * Caller is responsible for calling kiocb_end_write() on completion
840  * if async iocb was queued.
841  */
842 ssize_t vfs_iocb_iter_write(struct file *file, struct kiocb *iocb,
843 			    struct iov_iter *iter)
844 {
845 	size_t tot_len;
846 	ssize_t ret = 0;
847 
848 	if (!file->f_op->write_iter)
849 		return -EINVAL;
850 	if (!(file->f_mode & FMODE_WRITE))
851 		return -EBADF;
852 	if (!(file->f_mode & FMODE_CAN_WRITE))
853 		return -EINVAL;
854 
855 	tot_len = iov_iter_count(iter);
856 	if (!tot_len)
857 		return 0;
858 	ret = rw_verify_area(WRITE, file, &iocb->ki_pos, tot_len);
859 	if (ret < 0)
860 		return ret;
861 
862 	kiocb_start_write(iocb);
863 	ret = call_write_iter(file, iocb, iter);
864 	if (ret != -EIOCBQUEUED)
865 		kiocb_end_write(iocb);
866 	if (ret > 0)
867 		fsnotify_modify(file);
868 
869 	return ret;
870 }
871 EXPORT_SYMBOL(vfs_iocb_iter_write);
872 
873 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos,
874 		       rwf_t flags)
875 {
876 	size_t tot_len;
877 	ssize_t ret;
878 
879 	if (!(file->f_mode & FMODE_WRITE))
880 		return -EBADF;
881 	if (!(file->f_mode & FMODE_CAN_WRITE))
882 		return -EINVAL;
883 	if (!file->f_op->write_iter)
884 		return -EINVAL;
885 
886 	tot_len = iov_iter_count(iter);
887 	if (!tot_len)
888 		return 0;
889 
890 	ret = rw_verify_area(WRITE, file, ppos, tot_len);
891 	if (ret < 0)
892 		return ret;
893 
894 	file_start_write(file);
895 	ret = do_iter_readv_writev(file, iter, ppos, WRITE, flags);
896 	if (ret > 0)
897 		fsnotify_modify(file);
898 	file_end_write(file);
899 
900 	return ret;
901 }
902 EXPORT_SYMBOL(vfs_iter_write);
903 
904 static ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
905 			 unsigned long vlen, loff_t *pos, rwf_t flags)
906 {
907 	struct iovec iovstack[UIO_FASTIOV];
908 	struct iovec *iov = iovstack;
909 	struct iov_iter iter;
910 	size_t tot_len;
911 	ssize_t ret = 0;
912 
913 	if (!(file->f_mode & FMODE_READ))
914 		return -EBADF;
915 	if (!(file->f_mode & FMODE_CAN_READ))
916 		return -EINVAL;
917 
918 	ret = import_iovec(ITER_DEST, vec, vlen, ARRAY_SIZE(iovstack), &iov,
919 			   &iter);
920 	if (ret < 0)
921 		return ret;
922 
923 	tot_len = iov_iter_count(&iter);
924 	if (!tot_len)
925 		goto out;
926 
927 	ret = rw_verify_area(READ, file, pos, tot_len);
928 	if (ret < 0)
929 		goto out;
930 
931 	if (file->f_op->read_iter)
932 		ret = do_iter_readv_writev(file, &iter, pos, READ, flags);
933 	else
934 		ret = do_loop_readv_writev(file, &iter, pos, READ, flags);
935 out:
936 	if (ret >= 0)
937 		fsnotify_access(file);
938 	kfree(iov);
939 	return ret;
940 }
941 
942 static ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
943 			  unsigned long vlen, loff_t *pos, rwf_t flags)
944 {
945 	struct iovec iovstack[UIO_FASTIOV];
946 	struct iovec *iov = iovstack;
947 	struct iov_iter iter;
948 	size_t tot_len;
949 	ssize_t ret = 0;
950 
951 	if (!(file->f_mode & FMODE_WRITE))
952 		return -EBADF;
953 	if (!(file->f_mode & FMODE_CAN_WRITE))
954 		return -EINVAL;
955 
956 	ret = import_iovec(ITER_SOURCE, vec, vlen, ARRAY_SIZE(iovstack), &iov,
957 			   &iter);
958 	if (ret < 0)
959 		return ret;
960 
961 	tot_len = iov_iter_count(&iter);
962 	if (!tot_len)
963 		goto out;
964 
965 	ret = rw_verify_area(WRITE, file, pos, tot_len);
966 	if (ret < 0)
967 		goto out;
968 
969 	file_start_write(file);
970 	if (file->f_op->write_iter)
971 		ret = do_iter_readv_writev(file, &iter, pos, WRITE, flags);
972 	else
973 		ret = do_loop_readv_writev(file, &iter, pos, WRITE, flags);
974 	if (ret > 0)
975 		fsnotify_modify(file);
976 	file_end_write(file);
977 out:
978 	kfree(iov);
979 	return ret;
980 }
981 
982 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
983 			unsigned long vlen, rwf_t flags)
984 {
985 	struct fd f = fdget_pos(fd);
986 	ssize_t ret = -EBADF;
987 
988 	if (f.file) {
989 		loff_t pos, *ppos = file_ppos(f.file);
990 		if (ppos) {
991 			pos = *ppos;
992 			ppos = &pos;
993 		}
994 		ret = vfs_readv(f.file, vec, vlen, ppos, flags);
995 		if (ret >= 0 && ppos)
996 			f.file->f_pos = pos;
997 		fdput_pos(f);
998 	}
999 
1000 	if (ret > 0)
1001 		add_rchar(current, ret);
1002 	inc_syscr(current);
1003 	return ret;
1004 }
1005 
1006 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
1007 			 unsigned long vlen, rwf_t flags)
1008 {
1009 	struct fd f = fdget_pos(fd);
1010 	ssize_t ret = -EBADF;
1011 
1012 	if (f.file) {
1013 		loff_t pos, *ppos = file_ppos(f.file);
1014 		if (ppos) {
1015 			pos = *ppos;
1016 			ppos = &pos;
1017 		}
1018 		ret = vfs_writev(f.file, vec, vlen, ppos, flags);
1019 		if (ret >= 0 && ppos)
1020 			f.file->f_pos = pos;
1021 		fdput_pos(f);
1022 	}
1023 
1024 	if (ret > 0)
1025 		add_wchar(current, ret);
1026 	inc_syscw(current);
1027 	return ret;
1028 }
1029 
1030 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
1031 {
1032 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
1033 	return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
1034 }
1035 
1036 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
1037 			 unsigned long vlen, loff_t pos, rwf_t flags)
1038 {
1039 	struct fd f;
1040 	ssize_t ret = -EBADF;
1041 
1042 	if (pos < 0)
1043 		return -EINVAL;
1044 
1045 	f = fdget(fd);
1046 	if (f.file) {
1047 		ret = -ESPIPE;
1048 		if (f.file->f_mode & FMODE_PREAD)
1049 			ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1050 		fdput(f);
1051 	}
1052 
1053 	if (ret > 0)
1054 		add_rchar(current, ret);
1055 	inc_syscr(current);
1056 	return ret;
1057 }
1058 
1059 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
1060 			  unsigned long vlen, loff_t pos, rwf_t flags)
1061 {
1062 	struct fd f;
1063 	ssize_t ret = -EBADF;
1064 
1065 	if (pos < 0)
1066 		return -EINVAL;
1067 
1068 	f = fdget(fd);
1069 	if (f.file) {
1070 		ret = -ESPIPE;
1071 		if (f.file->f_mode & FMODE_PWRITE)
1072 			ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1073 		fdput(f);
1074 	}
1075 
1076 	if (ret > 0)
1077 		add_wchar(current, ret);
1078 	inc_syscw(current);
1079 	return ret;
1080 }
1081 
1082 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1083 		unsigned long, vlen)
1084 {
1085 	return do_readv(fd, vec, vlen, 0);
1086 }
1087 
1088 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1089 		unsigned long, vlen)
1090 {
1091 	return do_writev(fd, vec, vlen, 0);
1092 }
1093 
1094 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1095 		unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1096 {
1097 	loff_t pos = pos_from_hilo(pos_h, pos_l);
1098 
1099 	return do_preadv(fd, vec, vlen, pos, 0);
1100 }
1101 
1102 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1103 		unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1104 		rwf_t, flags)
1105 {
1106 	loff_t pos = pos_from_hilo(pos_h, pos_l);
1107 
1108 	if (pos == -1)
1109 		return do_readv(fd, vec, vlen, flags);
1110 
1111 	return do_preadv(fd, vec, vlen, pos, flags);
1112 }
1113 
1114 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1115 		unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1116 {
1117 	loff_t pos = pos_from_hilo(pos_h, pos_l);
1118 
1119 	return do_pwritev(fd, vec, vlen, pos, 0);
1120 }
1121 
1122 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1123 		unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1124 		rwf_t, flags)
1125 {
1126 	loff_t pos = pos_from_hilo(pos_h, pos_l);
1127 
1128 	if (pos == -1)
1129 		return do_writev(fd, vec, vlen, flags);
1130 
1131 	return do_pwritev(fd, vec, vlen, pos, flags);
1132 }
1133 
1134 /*
1135  * Various compat syscalls.  Note that they all pretend to take a native
1136  * iovec - import_iovec will properly treat those as compat_iovecs based on
1137  * in_compat_syscall().
1138  */
1139 #ifdef CONFIG_COMPAT
1140 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1141 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1142 		const struct iovec __user *, vec,
1143 		unsigned long, vlen, loff_t, pos)
1144 {
1145 	return do_preadv(fd, vec, vlen, pos, 0);
1146 }
1147 #endif
1148 
1149 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1150 		const struct iovec __user *, vec,
1151 		compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1152 {
1153 	loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1154 
1155 	return do_preadv(fd, vec, vlen, pos, 0);
1156 }
1157 
1158 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1159 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1160 		const struct iovec __user *, vec,
1161 		unsigned long, vlen, loff_t, pos, rwf_t, flags)
1162 {
1163 	if (pos == -1)
1164 		return do_readv(fd, vec, vlen, flags);
1165 	return do_preadv(fd, vec, vlen, pos, flags);
1166 }
1167 #endif
1168 
1169 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1170 		const struct iovec __user *, vec,
1171 		compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1172 		rwf_t, flags)
1173 {
1174 	loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1175 
1176 	if (pos == -1)
1177 		return do_readv(fd, vec, vlen, flags);
1178 	return do_preadv(fd, vec, vlen, pos, flags);
1179 }
1180 
1181 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1182 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1183 		const struct iovec __user *, vec,
1184 		unsigned long, vlen, loff_t, pos)
1185 {
1186 	return do_pwritev(fd, vec, vlen, pos, 0);
1187 }
1188 #endif
1189 
1190 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1191 		const struct iovec __user *,vec,
1192 		compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1193 {
1194 	loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1195 
1196 	return do_pwritev(fd, vec, vlen, pos, 0);
1197 }
1198 
1199 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1200 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1201 		const struct iovec __user *, vec,
1202 		unsigned long, vlen, loff_t, pos, rwf_t, flags)
1203 {
1204 	if (pos == -1)
1205 		return do_writev(fd, vec, vlen, flags);
1206 	return do_pwritev(fd, vec, vlen, pos, flags);
1207 }
1208 #endif
1209 
1210 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1211 		const struct iovec __user *,vec,
1212 		compat_ulong_t, vlen, u32, pos_low, u32, pos_high, rwf_t, flags)
1213 {
1214 	loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1215 
1216 	if (pos == -1)
1217 		return do_writev(fd, vec, vlen, flags);
1218 	return do_pwritev(fd, vec, vlen, pos, flags);
1219 }
1220 #endif /* CONFIG_COMPAT */
1221 
1222 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1223 			   size_t count, loff_t max)
1224 {
1225 	struct fd in, out;
1226 	struct inode *in_inode, *out_inode;
1227 	struct pipe_inode_info *opipe;
1228 	loff_t pos;
1229 	loff_t out_pos;
1230 	ssize_t retval;
1231 	int fl;
1232 
1233 	/*
1234 	 * Get input file, and verify that it is ok..
1235 	 */
1236 	retval = -EBADF;
1237 	in = fdget(in_fd);
1238 	if (!in.file)
1239 		goto out;
1240 	if (!(in.file->f_mode & FMODE_READ))
1241 		goto fput_in;
1242 	retval = -ESPIPE;
1243 	if (!ppos) {
1244 		pos = in.file->f_pos;
1245 	} else {
1246 		pos = *ppos;
1247 		if (!(in.file->f_mode & FMODE_PREAD))
1248 			goto fput_in;
1249 	}
1250 	retval = rw_verify_area(READ, in.file, &pos, count);
1251 	if (retval < 0)
1252 		goto fput_in;
1253 	if (count > MAX_RW_COUNT)
1254 		count =  MAX_RW_COUNT;
1255 
1256 	/*
1257 	 * Get output file, and verify that it is ok..
1258 	 */
1259 	retval = -EBADF;
1260 	out = fdget(out_fd);
1261 	if (!out.file)
1262 		goto fput_in;
1263 	if (!(out.file->f_mode & FMODE_WRITE))
1264 		goto fput_out;
1265 	in_inode = file_inode(in.file);
1266 	out_inode = file_inode(out.file);
1267 	out_pos = out.file->f_pos;
1268 
1269 	if (!max)
1270 		max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1271 
1272 	if (unlikely(pos + count > max)) {
1273 		retval = -EOVERFLOW;
1274 		if (pos >= max)
1275 			goto fput_out;
1276 		count = max - pos;
1277 	}
1278 
1279 	fl = 0;
1280 #if 0
1281 	/*
1282 	 * We need to debate whether we can enable this or not. The
1283 	 * man page documents EAGAIN return for the output at least,
1284 	 * and the application is arguably buggy if it doesn't expect
1285 	 * EAGAIN on a non-blocking file descriptor.
1286 	 */
1287 	if (in.file->f_flags & O_NONBLOCK)
1288 		fl = SPLICE_F_NONBLOCK;
1289 #endif
1290 	opipe = get_pipe_info(out.file, true);
1291 	if (!opipe) {
1292 		retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1293 		if (retval < 0)
1294 			goto fput_out;
1295 		retval = do_splice_direct(in.file, &pos, out.file, &out_pos,
1296 					  count, fl);
1297 	} else {
1298 		if (out.file->f_flags & O_NONBLOCK)
1299 			fl |= SPLICE_F_NONBLOCK;
1300 
1301 		retval = splice_file_to_pipe(in.file, opipe, &pos, count, fl);
1302 	}
1303 
1304 	if (retval > 0) {
1305 		add_rchar(current, retval);
1306 		add_wchar(current, retval);
1307 		fsnotify_access(in.file);
1308 		fsnotify_modify(out.file);
1309 		out.file->f_pos = out_pos;
1310 		if (ppos)
1311 			*ppos = pos;
1312 		else
1313 			in.file->f_pos = pos;
1314 	}
1315 
1316 	inc_syscr(current);
1317 	inc_syscw(current);
1318 	if (pos > max)
1319 		retval = -EOVERFLOW;
1320 
1321 fput_out:
1322 	fdput(out);
1323 fput_in:
1324 	fdput(in);
1325 out:
1326 	return retval;
1327 }
1328 
1329 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1330 {
1331 	loff_t pos;
1332 	off_t off;
1333 	ssize_t ret;
1334 
1335 	if (offset) {
1336 		if (unlikely(get_user(off, offset)))
1337 			return -EFAULT;
1338 		pos = off;
1339 		ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1340 		if (unlikely(put_user(pos, offset)))
1341 			return -EFAULT;
1342 		return ret;
1343 	}
1344 
1345 	return do_sendfile(out_fd, in_fd, NULL, count, 0);
1346 }
1347 
1348 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1349 {
1350 	loff_t pos;
1351 	ssize_t ret;
1352 
1353 	if (offset) {
1354 		if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1355 			return -EFAULT;
1356 		ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1357 		if (unlikely(put_user(pos, offset)))
1358 			return -EFAULT;
1359 		return ret;
1360 	}
1361 
1362 	return do_sendfile(out_fd, in_fd, NULL, count, 0);
1363 }
1364 
1365 #ifdef CONFIG_COMPAT
1366 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1367 		compat_off_t __user *, offset, compat_size_t, count)
1368 {
1369 	loff_t pos;
1370 	off_t off;
1371 	ssize_t ret;
1372 
1373 	if (offset) {
1374 		if (unlikely(get_user(off, offset)))
1375 			return -EFAULT;
1376 		pos = off;
1377 		ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1378 		if (unlikely(put_user(pos, offset)))
1379 			return -EFAULT;
1380 		return ret;
1381 	}
1382 
1383 	return do_sendfile(out_fd, in_fd, NULL, count, 0);
1384 }
1385 
1386 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1387 		compat_loff_t __user *, offset, compat_size_t, count)
1388 {
1389 	loff_t pos;
1390 	ssize_t ret;
1391 
1392 	if (offset) {
1393 		if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1394 			return -EFAULT;
1395 		ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1396 		if (unlikely(put_user(pos, offset)))
1397 			return -EFAULT;
1398 		return ret;
1399 	}
1400 
1401 	return do_sendfile(out_fd, in_fd, NULL, count, 0);
1402 }
1403 #endif
1404 
1405 /*
1406  * Performs necessary checks before doing a file copy
1407  *
1408  * Can adjust amount of bytes to copy via @req_count argument.
1409  * Returns appropriate error code that caller should return or
1410  * zero in case the copy should be allowed.
1411  */
1412 static int generic_copy_file_checks(struct file *file_in, loff_t pos_in,
1413 				    struct file *file_out, loff_t pos_out,
1414 				    size_t *req_count, unsigned int flags)
1415 {
1416 	struct inode *inode_in = file_inode(file_in);
1417 	struct inode *inode_out = file_inode(file_out);
1418 	uint64_t count = *req_count;
1419 	loff_t size_in;
1420 	int ret;
1421 
1422 	ret = generic_file_rw_checks(file_in, file_out);
1423 	if (ret)
1424 		return ret;
1425 
1426 	/*
1427 	 * We allow some filesystems to handle cross sb copy, but passing
1428 	 * a file of the wrong filesystem type to filesystem driver can result
1429 	 * in an attempt to dereference the wrong type of ->private_data, so
1430 	 * avoid doing that until we really have a good reason.
1431 	 *
1432 	 * nfs and cifs define several different file_system_type structures
1433 	 * and several different sets of file_operations, but they all end up
1434 	 * using the same ->copy_file_range() function pointer.
1435 	 */
1436 	if (flags & COPY_FILE_SPLICE) {
1437 		/* cross sb splice is allowed */
1438 	} else if (file_out->f_op->copy_file_range) {
1439 		if (file_in->f_op->copy_file_range !=
1440 		    file_out->f_op->copy_file_range)
1441 			return -EXDEV;
1442 	} else if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb) {
1443 		return -EXDEV;
1444 	}
1445 
1446 	/* Don't touch certain kinds of inodes */
1447 	if (IS_IMMUTABLE(inode_out))
1448 		return -EPERM;
1449 
1450 	if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1451 		return -ETXTBSY;
1452 
1453 	/* Ensure offsets don't wrap. */
1454 	if (pos_in + count < pos_in || pos_out + count < pos_out)
1455 		return -EOVERFLOW;
1456 
1457 	/* Shorten the copy to EOF */
1458 	size_in = i_size_read(inode_in);
1459 	if (pos_in >= size_in)
1460 		count = 0;
1461 	else
1462 		count = min(count, size_in - (uint64_t)pos_in);
1463 
1464 	ret = generic_write_check_limits(file_out, pos_out, &count);
1465 	if (ret)
1466 		return ret;
1467 
1468 	/* Don't allow overlapped copying within the same file. */
1469 	if (inode_in == inode_out &&
1470 	    pos_out + count > pos_in &&
1471 	    pos_out < pos_in + count)
1472 		return -EINVAL;
1473 
1474 	*req_count = count;
1475 	return 0;
1476 }
1477 
1478 /*
1479  * copy_file_range() differs from regular file read and write in that it
1480  * specifically allows return partial success.  When it does so is up to
1481  * the copy_file_range method.
1482  */
1483 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1484 			    struct file *file_out, loff_t pos_out,
1485 			    size_t len, unsigned int flags)
1486 {
1487 	ssize_t ret;
1488 	bool splice = flags & COPY_FILE_SPLICE;
1489 	bool samesb = file_inode(file_in)->i_sb == file_inode(file_out)->i_sb;
1490 
1491 	if (flags & ~COPY_FILE_SPLICE)
1492 		return -EINVAL;
1493 
1494 	ret = generic_copy_file_checks(file_in, pos_in, file_out, pos_out, &len,
1495 				       flags);
1496 	if (unlikely(ret))
1497 		return ret;
1498 
1499 	ret = rw_verify_area(READ, file_in, &pos_in, len);
1500 	if (unlikely(ret))
1501 		return ret;
1502 
1503 	ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1504 	if (unlikely(ret))
1505 		return ret;
1506 
1507 	if (len == 0)
1508 		return 0;
1509 
1510 	file_start_write(file_out);
1511 
1512 	/*
1513 	 * Cloning is supported by more file systems, so we implement copy on
1514 	 * same sb using clone, but for filesystems where both clone and copy
1515 	 * are supported (e.g. nfs,cifs), we only call the copy method.
1516 	 */
1517 	if (!splice && file_out->f_op->copy_file_range) {
1518 		ret = file_out->f_op->copy_file_range(file_in, pos_in,
1519 						      file_out, pos_out,
1520 						      len, flags);
1521 	} else if (!splice && file_in->f_op->remap_file_range && samesb) {
1522 		ret = file_in->f_op->remap_file_range(file_in, pos_in,
1523 				file_out, pos_out,
1524 				min_t(loff_t, MAX_RW_COUNT, len),
1525 				REMAP_FILE_CAN_SHORTEN);
1526 		/* fallback to splice */
1527 		if (ret <= 0)
1528 			splice = true;
1529 	} else if (samesb) {
1530 		/* Fallback to splice for same sb copy for backward compat */
1531 		splice = true;
1532 	}
1533 
1534 	file_end_write(file_out);
1535 
1536 	if (!splice)
1537 		goto done;
1538 
1539 	/*
1540 	 * We can get here for same sb copy of filesystems that do not implement
1541 	 * ->copy_file_range() in case filesystem does not support clone or in
1542 	 * case filesystem supports clone but rejected the clone request (e.g.
1543 	 * because it was not block aligned).
1544 	 *
1545 	 * In both cases, fall back to kernel copy so we are able to maintain a
1546 	 * consistent story about which filesystems support copy_file_range()
1547 	 * and which filesystems do not, that will allow userspace tools to
1548 	 * make consistent desicions w.r.t using copy_file_range().
1549 	 *
1550 	 * We also get here if caller (e.g. nfsd) requested COPY_FILE_SPLICE
1551 	 * for server-side-copy between any two sb.
1552 	 *
1553 	 * In any case, we call do_splice_direct() and not splice_file_range(),
1554 	 * without file_start_write() held, to avoid possible deadlocks related
1555 	 * to splicing from input file, while file_start_write() is held on
1556 	 * the output file on a different sb.
1557 	 */
1558 	ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1559 			       min_t(size_t, len, MAX_RW_COUNT), 0);
1560 done:
1561 	if (ret > 0) {
1562 		fsnotify_access(file_in);
1563 		add_rchar(current, ret);
1564 		fsnotify_modify(file_out);
1565 		add_wchar(current, ret);
1566 	}
1567 
1568 	inc_syscr(current);
1569 	inc_syscw(current);
1570 
1571 	return ret;
1572 }
1573 EXPORT_SYMBOL(vfs_copy_file_range);
1574 
1575 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1576 		int, fd_out, loff_t __user *, off_out,
1577 		size_t, len, unsigned int, flags)
1578 {
1579 	loff_t pos_in;
1580 	loff_t pos_out;
1581 	struct fd f_in;
1582 	struct fd f_out;
1583 	ssize_t ret = -EBADF;
1584 
1585 	f_in = fdget(fd_in);
1586 	if (!f_in.file)
1587 		goto out2;
1588 
1589 	f_out = fdget(fd_out);
1590 	if (!f_out.file)
1591 		goto out1;
1592 
1593 	ret = -EFAULT;
1594 	if (off_in) {
1595 		if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1596 			goto out;
1597 	} else {
1598 		pos_in = f_in.file->f_pos;
1599 	}
1600 
1601 	if (off_out) {
1602 		if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1603 			goto out;
1604 	} else {
1605 		pos_out = f_out.file->f_pos;
1606 	}
1607 
1608 	ret = -EINVAL;
1609 	if (flags != 0)
1610 		goto out;
1611 
1612 	ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1613 				  flags);
1614 	if (ret > 0) {
1615 		pos_in += ret;
1616 		pos_out += ret;
1617 
1618 		if (off_in) {
1619 			if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1620 				ret = -EFAULT;
1621 		} else {
1622 			f_in.file->f_pos = pos_in;
1623 		}
1624 
1625 		if (off_out) {
1626 			if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1627 				ret = -EFAULT;
1628 		} else {
1629 			f_out.file->f_pos = pos_out;
1630 		}
1631 	}
1632 
1633 out:
1634 	fdput(f_out);
1635 out1:
1636 	fdput(f_in);
1637 out2:
1638 	return ret;
1639 }
1640 
1641 /*
1642  * Don't operate on ranges the page cache doesn't support, and don't exceed the
1643  * LFS limits.  If pos is under the limit it becomes a short access.  If it
1644  * exceeds the limit we return -EFBIG.
1645  */
1646 int generic_write_check_limits(struct file *file, loff_t pos, loff_t *count)
1647 {
1648 	struct inode *inode = file->f_mapping->host;
1649 	loff_t max_size = inode->i_sb->s_maxbytes;
1650 	loff_t limit = rlimit(RLIMIT_FSIZE);
1651 
1652 	if (limit != RLIM_INFINITY) {
1653 		if (pos >= limit) {
1654 			send_sig(SIGXFSZ, current, 0);
1655 			return -EFBIG;
1656 		}
1657 		*count = min(*count, limit - pos);
1658 	}
1659 
1660 	if (!(file->f_flags & O_LARGEFILE))
1661 		max_size = MAX_NON_LFS;
1662 
1663 	if (unlikely(pos >= max_size))
1664 		return -EFBIG;
1665 
1666 	*count = min(*count, max_size - pos);
1667 
1668 	return 0;
1669 }
1670 
1671 /* Like generic_write_checks(), but takes size of write instead of iter. */
1672 int generic_write_checks_count(struct kiocb *iocb, loff_t *count)
1673 {
1674 	struct file *file = iocb->ki_filp;
1675 	struct inode *inode = file->f_mapping->host;
1676 
1677 	if (IS_SWAPFILE(inode))
1678 		return -ETXTBSY;
1679 
1680 	if (!*count)
1681 		return 0;
1682 
1683 	if (iocb->ki_flags & IOCB_APPEND)
1684 		iocb->ki_pos = i_size_read(inode);
1685 
1686 	if ((iocb->ki_flags & IOCB_NOWAIT) &&
1687 	    !((iocb->ki_flags & IOCB_DIRECT) ||
1688 	      (file->f_mode & FMODE_BUF_WASYNC)))
1689 		return -EINVAL;
1690 
1691 	return generic_write_check_limits(iocb->ki_filp, iocb->ki_pos, count);
1692 }
1693 EXPORT_SYMBOL(generic_write_checks_count);
1694 
1695 /*
1696  * Performs necessary checks before doing a write
1697  *
1698  * Can adjust writing position or amount of bytes to write.
1699  * Returns appropriate error code that caller should return or
1700  * zero in case that write should be allowed.
1701  */
1702 ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
1703 {
1704 	loff_t count = iov_iter_count(from);
1705 	int ret;
1706 
1707 	ret = generic_write_checks_count(iocb, &count);
1708 	if (ret)
1709 		return ret;
1710 
1711 	iov_iter_truncate(from, count);
1712 	return iov_iter_count(from);
1713 }
1714 EXPORT_SYMBOL(generic_write_checks);
1715 
1716 /*
1717  * Performs common checks before doing a file copy/clone
1718  * from @file_in to @file_out.
1719  */
1720 int generic_file_rw_checks(struct file *file_in, struct file *file_out)
1721 {
1722 	struct inode *inode_in = file_inode(file_in);
1723 	struct inode *inode_out = file_inode(file_out);
1724 
1725 	/* Don't copy dirs, pipes, sockets... */
1726 	if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1727 		return -EISDIR;
1728 	if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1729 		return -EINVAL;
1730 
1731 	if (!(file_in->f_mode & FMODE_READ) ||
1732 	    !(file_out->f_mode & FMODE_WRITE) ||
1733 	    (file_out->f_flags & O_APPEND))
1734 		return -EBADF;
1735 
1736 	return 0;
1737 }
1738