xref: /linux/fs/read_write.c (revision 7c7e33b799ac169e5fab8abfc6819fce8b26d53d)
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  * @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 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 	if (unlikely((ssize_t) count < 0))
358 		return -EINVAL;
359 
360 	if (ppos) {
361 		loff_t pos = *ppos;
362 
363 		if (unlikely(pos < 0)) {
364 			if (!unsigned_offsets(file))
365 				return -EINVAL;
366 			if (count >= -pos) /* both values are in 0..LLONG_MAX */
367 				return -EOVERFLOW;
368 		} else if (unlikely((loff_t) (pos + count) < 0)) {
369 			if (!unsigned_offsets(file))
370 				return -EINVAL;
371 		}
372 	}
373 
374 	return security_file_permission(file,
375 				read_write == READ ? MAY_READ : MAY_WRITE);
376 }
377 EXPORT_SYMBOL(rw_verify_area);
378 
379 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
380 {
381 	struct kiocb kiocb;
382 	struct iov_iter iter;
383 	ssize_t ret;
384 
385 	init_sync_kiocb(&kiocb, filp);
386 	kiocb.ki_pos = (ppos ? *ppos : 0);
387 	iov_iter_ubuf(&iter, ITER_DEST, buf, len);
388 
389 	ret = call_read_iter(filp, &kiocb, &iter);
390 	BUG_ON(ret == -EIOCBQUEUED);
391 	if (ppos)
392 		*ppos = kiocb.ki_pos;
393 	return ret;
394 }
395 
396 static int warn_unsupported(struct file *file, const char *op)
397 {
398 	pr_warn_ratelimited(
399 		"kernel %s not supported for file %pD4 (pid: %d comm: %.20s)\n",
400 		op, file, current->pid, current->comm);
401 	return -EINVAL;
402 }
403 
404 ssize_t __kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
405 {
406 	struct kvec iov = {
407 		.iov_base	= buf,
408 		.iov_len	= min_t(size_t, count, MAX_RW_COUNT),
409 	};
410 	struct kiocb kiocb;
411 	struct iov_iter iter;
412 	ssize_t ret;
413 
414 	if (WARN_ON_ONCE(!(file->f_mode & FMODE_READ)))
415 		return -EINVAL;
416 	if (!(file->f_mode & FMODE_CAN_READ))
417 		return -EINVAL;
418 	/*
419 	 * Also fail if ->read_iter and ->read are both wired up as that
420 	 * implies very convoluted semantics.
421 	 */
422 	if (unlikely(!file->f_op->read_iter || file->f_op->read))
423 		return warn_unsupported(file, "read");
424 
425 	init_sync_kiocb(&kiocb, file);
426 	kiocb.ki_pos = pos ? *pos : 0;
427 	iov_iter_kvec(&iter, ITER_DEST, &iov, 1, iov.iov_len);
428 	ret = file->f_op->read_iter(&kiocb, &iter);
429 	if (ret > 0) {
430 		if (pos)
431 			*pos = kiocb.ki_pos;
432 		fsnotify_access(file);
433 		add_rchar(current, ret);
434 	}
435 	inc_syscr(current);
436 	return ret;
437 }
438 
439 ssize_t kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
440 {
441 	ssize_t ret;
442 
443 	ret = rw_verify_area(READ, file, pos, count);
444 	if (ret)
445 		return ret;
446 	return __kernel_read(file, buf, count, pos);
447 }
448 EXPORT_SYMBOL(kernel_read);
449 
450 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
451 {
452 	ssize_t ret;
453 
454 	if (!(file->f_mode & FMODE_READ))
455 		return -EBADF;
456 	if (!(file->f_mode & FMODE_CAN_READ))
457 		return -EINVAL;
458 	if (unlikely(!access_ok(buf, count)))
459 		return -EFAULT;
460 
461 	ret = rw_verify_area(READ, file, pos, count);
462 	if (ret)
463 		return ret;
464 	if (count > MAX_RW_COUNT)
465 		count =  MAX_RW_COUNT;
466 
467 	if (file->f_op->read)
468 		ret = file->f_op->read(file, buf, count, pos);
469 	else if (file->f_op->read_iter)
470 		ret = new_sync_read(file, buf, count, pos);
471 	else
472 		ret = -EINVAL;
473 	if (ret > 0) {
474 		fsnotify_access(file);
475 		add_rchar(current, ret);
476 	}
477 	inc_syscr(current);
478 	return ret;
479 }
480 
481 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
482 {
483 	struct kiocb kiocb;
484 	struct iov_iter iter;
485 	ssize_t ret;
486 
487 	init_sync_kiocb(&kiocb, filp);
488 	kiocb.ki_pos = (ppos ? *ppos : 0);
489 	iov_iter_ubuf(&iter, ITER_SOURCE, (void __user *)buf, len);
490 
491 	ret = call_write_iter(filp, &kiocb, &iter);
492 	BUG_ON(ret == -EIOCBQUEUED);
493 	if (ret > 0 && ppos)
494 		*ppos = kiocb.ki_pos;
495 	return ret;
496 }
497 
498 /* caller is responsible for file_start_write/file_end_write */
499 ssize_t __kernel_write_iter(struct file *file, struct iov_iter *from, loff_t *pos)
500 {
501 	struct kiocb kiocb;
502 	ssize_t ret;
503 
504 	if (WARN_ON_ONCE(!(file->f_mode & FMODE_WRITE)))
505 		return -EBADF;
506 	if (!(file->f_mode & FMODE_CAN_WRITE))
507 		return -EINVAL;
508 	/*
509 	 * Also fail if ->write_iter and ->write are both wired up as that
510 	 * implies very convoluted semantics.
511 	 */
512 	if (unlikely(!file->f_op->write_iter || file->f_op->write))
513 		return warn_unsupported(file, "write");
514 
515 	init_sync_kiocb(&kiocb, file);
516 	kiocb.ki_pos = pos ? *pos : 0;
517 	ret = file->f_op->write_iter(&kiocb, from);
518 	if (ret > 0) {
519 		if (pos)
520 			*pos = kiocb.ki_pos;
521 		fsnotify_modify(file);
522 		add_wchar(current, ret);
523 	}
524 	inc_syscw(current);
525 	return ret;
526 }
527 
528 /* caller is responsible for file_start_write/file_end_write */
529 ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t *pos)
530 {
531 	struct kvec iov = {
532 		.iov_base	= (void *)buf,
533 		.iov_len	= min_t(size_t, count, MAX_RW_COUNT),
534 	};
535 	struct iov_iter iter;
536 	iov_iter_kvec(&iter, ITER_SOURCE, &iov, 1, iov.iov_len);
537 	return __kernel_write_iter(file, &iter, pos);
538 }
539 /*
540  * This "EXPORT_SYMBOL_GPL()" is more of a "EXPORT_SYMBOL_DONTUSE()",
541  * but autofs is one of the few internal kernel users that actually
542  * wants this _and_ can be built as a module. So we need to export
543  * this symbol for autofs, even though it really isn't appropriate
544  * for any other kernel modules.
545  */
546 EXPORT_SYMBOL_GPL(__kernel_write);
547 
548 ssize_t kernel_write(struct file *file, const void *buf, size_t count,
549 			    loff_t *pos)
550 {
551 	ssize_t ret;
552 
553 	ret = rw_verify_area(WRITE, file, pos, count);
554 	if (ret)
555 		return ret;
556 
557 	file_start_write(file);
558 	ret =  __kernel_write(file, buf, count, pos);
559 	file_end_write(file);
560 	return ret;
561 }
562 EXPORT_SYMBOL(kernel_write);
563 
564 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
565 {
566 	ssize_t ret;
567 
568 	if (!(file->f_mode & FMODE_WRITE))
569 		return -EBADF;
570 	if (!(file->f_mode & FMODE_CAN_WRITE))
571 		return -EINVAL;
572 	if (unlikely(!access_ok(buf, count)))
573 		return -EFAULT;
574 
575 	ret = rw_verify_area(WRITE, file, pos, count);
576 	if (ret)
577 		return ret;
578 	if (count > MAX_RW_COUNT)
579 		count =  MAX_RW_COUNT;
580 	file_start_write(file);
581 	if (file->f_op->write)
582 		ret = file->f_op->write(file, buf, count, pos);
583 	else if (file->f_op->write_iter)
584 		ret = new_sync_write(file, buf, count, pos);
585 	else
586 		ret = -EINVAL;
587 	if (ret > 0) {
588 		fsnotify_modify(file);
589 		add_wchar(current, ret);
590 	}
591 	inc_syscw(current);
592 	file_end_write(file);
593 	return ret;
594 }
595 
596 /* file_ppos returns &file->f_pos or NULL if file is stream */
597 static inline loff_t *file_ppos(struct file *file)
598 {
599 	return file->f_mode & FMODE_STREAM ? NULL : &file->f_pos;
600 }
601 
602 ssize_t ksys_read(unsigned int fd, char __user *buf, size_t count)
603 {
604 	struct fd f = fdget_pos(fd);
605 	ssize_t ret = -EBADF;
606 
607 	if (f.file) {
608 		loff_t pos, *ppos = file_ppos(f.file);
609 		if (ppos) {
610 			pos = *ppos;
611 			ppos = &pos;
612 		}
613 		ret = vfs_read(f.file, buf, count, ppos);
614 		if (ret >= 0 && ppos)
615 			f.file->f_pos = pos;
616 		fdput_pos(f);
617 	}
618 	return ret;
619 }
620 
621 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
622 {
623 	return ksys_read(fd, buf, count);
624 }
625 
626 ssize_t ksys_write(unsigned int fd, const char __user *buf, size_t count)
627 {
628 	struct fd f = fdget_pos(fd);
629 	ssize_t ret = -EBADF;
630 
631 	if (f.file) {
632 		loff_t pos, *ppos = file_ppos(f.file);
633 		if (ppos) {
634 			pos = *ppos;
635 			ppos = &pos;
636 		}
637 		ret = vfs_write(f.file, buf, count, ppos);
638 		if (ret >= 0 && ppos)
639 			f.file->f_pos = pos;
640 		fdput_pos(f);
641 	}
642 
643 	return ret;
644 }
645 
646 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
647 		size_t, count)
648 {
649 	return ksys_write(fd, buf, count);
650 }
651 
652 ssize_t ksys_pread64(unsigned int fd, char __user *buf, size_t count,
653 		     loff_t pos)
654 {
655 	struct fd f;
656 	ssize_t ret = -EBADF;
657 
658 	if (pos < 0)
659 		return -EINVAL;
660 
661 	f = fdget(fd);
662 	if (f.file) {
663 		ret = -ESPIPE;
664 		if (f.file->f_mode & FMODE_PREAD)
665 			ret = vfs_read(f.file, buf, count, &pos);
666 		fdput(f);
667 	}
668 
669 	return ret;
670 }
671 
672 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
673 			size_t, count, loff_t, pos)
674 {
675 	return ksys_pread64(fd, buf, count, pos);
676 }
677 
678 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_PREAD64)
679 COMPAT_SYSCALL_DEFINE5(pread64, unsigned int, fd, char __user *, buf,
680 		       size_t, count, compat_arg_u64_dual(pos))
681 {
682 	return ksys_pread64(fd, buf, count, compat_arg_u64_glue(pos));
683 }
684 #endif
685 
686 ssize_t ksys_pwrite64(unsigned int fd, const char __user *buf,
687 		      size_t count, loff_t pos)
688 {
689 	struct fd f;
690 	ssize_t ret = -EBADF;
691 
692 	if (pos < 0)
693 		return -EINVAL;
694 
695 	f = fdget(fd);
696 	if (f.file) {
697 		ret = -ESPIPE;
698 		if (f.file->f_mode & FMODE_PWRITE)
699 			ret = vfs_write(f.file, buf, count, &pos);
700 		fdput(f);
701 	}
702 
703 	return ret;
704 }
705 
706 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
707 			 size_t, count, loff_t, pos)
708 {
709 	return ksys_pwrite64(fd, buf, count, pos);
710 }
711 
712 #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_PWRITE64)
713 COMPAT_SYSCALL_DEFINE5(pwrite64, unsigned int, fd, const char __user *, buf,
714 		       size_t, count, compat_arg_u64_dual(pos))
715 {
716 	return ksys_pwrite64(fd, buf, count, compat_arg_u64_glue(pos));
717 }
718 #endif
719 
720 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
721 		loff_t *ppos, int type, rwf_t flags)
722 {
723 	struct kiocb kiocb;
724 	ssize_t ret;
725 
726 	init_sync_kiocb(&kiocb, filp);
727 	ret = kiocb_set_rw_flags(&kiocb, flags);
728 	if (ret)
729 		return ret;
730 	kiocb.ki_pos = (ppos ? *ppos : 0);
731 
732 	if (type == READ)
733 		ret = call_read_iter(filp, &kiocb, iter);
734 	else
735 		ret = call_write_iter(filp, &kiocb, iter);
736 	BUG_ON(ret == -EIOCBQUEUED);
737 	if (ppos)
738 		*ppos = kiocb.ki_pos;
739 	return ret;
740 }
741 
742 /* Do it by hand, with file-ops */
743 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
744 		loff_t *ppos, int type, rwf_t flags)
745 {
746 	ssize_t ret = 0;
747 
748 	if (flags & ~RWF_HIPRI)
749 		return -EOPNOTSUPP;
750 
751 	while (iov_iter_count(iter)) {
752 		ssize_t nr;
753 
754 		if (type == READ) {
755 			nr = filp->f_op->read(filp, iter_iov_addr(iter),
756 						iter_iov_len(iter), ppos);
757 		} else {
758 			nr = filp->f_op->write(filp, iter_iov_addr(iter),
759 						iter_iov_len(iter), ppos);
760 		}
761 
762 		if (nr < 0) {
763 			if (!ret)
764 				ret = nr;
765 			break;
766 		}
767 		ret += nr;
768 		if (nr != iter_iov_len(iter))
769 			break;
770 		iov_iter_advance(iter, nr);
771 	}
772 
773 	return ret;
774 }
775 
776 static ssize_t do_iter_read(struct file *file, struct iov_iter *iter,
777 		loff_t *pos, rwf_t flags)
778 {
779 	size_t tot_len;
780 	ssize_t ret = 0;
781 
782 	if (!(file->f_mode & FMODE_READ))
783 		return -EBADF;
784 	if (!(file->f_mode & FMODE_CAN_READ))
785 		return -EINVAL;
786 
787 	tot_len = iov_iter_count(iter);
788 	if (!tot_len)
789 		goto out;
790 	ret = rw_verify_area(READ, file, pos, tot_len);
791 	if (ret < 0)
792 		return ret;
793 
794 	if (file->f_op->read_iter)
795 		ret = do_iter_readv_writev(file, iter, pos, READ, flags);
796 	else
797 		ret = do_loop_readv_writev(file, iter, pos, READ, flags);
798 out:
799 	if (ret >= 0)
800 		fsnotify_access(file);
801 	return ret;
802 }
803 
804 ssize_t vfs_iocb_iter_read(struct file *file, struct kiocb *iocb,
805 			   struct iov_iter *iter)
806 {
807 	size_t tot_len;
808 	ssize_t ret = 0;
809 
810 	if (!file->f_op->read_iter)
811 		return -EINVAL;
812 	if (!(file->f_mode & FMODE_READ))
813 		return -EBADF;
814 	if (!(file->f_mode & FMODE_CAN_READ))
815 		return -EINVAL;
816 
817 	tot_len = iov_iter_count(iter);
818 	if (!tot_len)
819 		goto out;
820 	ret = rw_verify_area(READ, file, &iocb->ki_pos, tot_len);
821 	if (ret < 0)
822 		return ret;
823 
824 	ret = call_read_iter(file, iocb, iter);
825 out:
826 	if (ret >= 0)
827 		fsnotify_access(file);
828 	return ret;
829 }
830 EXPORT_SYMBOL(vfs_iocb_iter_read);
831 
832 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos,
833 		rwf_t flags)
834 {
835 	if (!file->f_op->read_iter)
836 		return -EINVAL;
837 	return do_iter_read(file, iter, ppos, flags);
838 }
839 EXPORT_SYMBOL(vfs_iter_read);
840 
841 static ssize_t do_iter_write(struct file *file, struct iov_iter *iter,
842 		loff_t *pos, rwf_t flags)
843 {
844 	size_t tot_len;
845 	ssize_t ret = 0;
846 
847 	if (!(file->f_mode & FMODE_WRITE))
848 		return -EBADF;
849 	if (!(file->f_mode & FMODE_CAN_WRITE))
850 		return -EINVAL;
851 
852 	tot_len = iov_iter_count(iter);
853 	if (!tot_len)
854 		return 0;
855 	ret = rw_verify_area(WRITE, file, pos, tot_len);
856 	if (ret < 0)
857 		return ret;
858 
859 	if (file->f_op->write_iter)
860 		ret = do_iter_readv_writev(file, iter, pos, WRITE, flags);
861 	else
862 		ret = do_loop_readv_writev(file, iter, pos, WRITE, flags);
863 	if (ret > 0)
864 		fsnotify_modify(file);
865 	return ret;
866 }
867 
868 ssize_t vfs_iocb_iter_write(struct file *file, struct kiocb *iocb,
869 			    struct iov_iter *iter)
870 {
871 	size_t tot_len;
872 	ssize_t ret = 0;
873 
874 	if (!file->f_op->write_iter)
875 		return -EINVAL;
876 	if (!(file->f_mode & FMODE_WRITE))
877 		return -EBADF;
878 	if (!(file->f_mode & FMODE_CAN_WRITE))
879 		return -EINVAL;
880 
881 	tot_len = iov_iter_count(iter);
882 	if (!tot_len)
883 		return 0;
884 	ret = rw_verify_area(WRITE, file, &iocb->ki_pos, tot_len);
885 	if (ret < 0)
886 		return ret;
887 
888 	ret = call_write_iter(file, iocb, iter);
889 	if (ret > 0)
890 		fsnotify_modify(file);
891 
892 	return ret;
893 }
894 EXPORT_SYMBOL(vfs_iocb_iter_write);
895 
896 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos,
897 		rwf_t flags)
898 {
899 	if (!file->f_op->write_iter)
900 		return -EINVAL;
901 	return do_iter_write(file, iter, ppos, flags);
902 }
903 EXPORT_SYMBOL(vfs_iter_write);
904 
905 static ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
906 		  unsigned long vlen, loff_t *pos, rwf_t flags)
907 {
908 	struct iovec iovstack[UIO_FASTIOV];
909 	struct iovec *iov = iovstack;
910 	struct iov_iter iter;
911 	ssize_t ret;
912 
913 	ret = import_iovec(ITER_DEST, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
914 	if (ret >= 0) {
915 		ret = do_iter_read(file, &iter, pos, flags);
916 		kfree(iov);
917 	}
918 
919 	return ret;
920 }
921 
922 static ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
923 		   unsigned long vlen, loff_t *pos, rwf_t flags)
924 {
925 	struct iovec iovstack[UIO_FASTIOV];
926 	struct iovec *iov = iovstack;
927 	struct iov_iter iter;
928 	ssize_t ret;
929 
930 	ret = import_iovec(ITER_SOURCE, vec, vlen, ARRAY_SIZE(iovstack), &iov, &iter);
931 	if (ret >= 0) {
932 		file_start_write(file);
933 		ret = do_iter_write(file, &iter, pos, flags);
934 		file_end_write(file);
935 		kfree(iov);
936 	}
937 	return ret;
938 }
939 
940 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
941 			unsigned long vlen, rwf_t flags)
942 {
943 	struct fd f = fdget_pos(fd);
944 	ssize_t ret = -EBADF;
945 
946 	if (f.file) {
947 		loff_t pos, *ppos = file_ppos(f.file);
948 		if (ppos) {
949 			pos = *ppos;
950 			ppos = &pos;
951 		}
952 		ret = vfs_readv(f.file, vec, vlen, ppos, flags);
953 		if (ret >= 0 && ppos)
954 			f.file->f_pos = pos;
955 		fdput_pos(f);
956 	}
957 
958 	if (ret > 0)
959 		add_rchar(current, ret);
960 	inc_syscr(current);
961 	return ret;
962 }
963 
964 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
965 			 unsigned long vlen, rwf_t flags)
966 {
967 	struct fd f = fdget_pos(fd);
968 	ssize_t ret = -EBADF;
969 
970 	if (f.file) {
971 		loff_t pos, *ppos = file_ppos(f.file);
972 		if (ppos) {
973 			pos = *ppos;
974 			ppos = &pos;
975 		}
976 		ret = vfs_writev(f.file, vec, vlen, ppos, flags);
977 		if (ret >= 0 && ppos)
978 			f.file->f_pos = pos;
979 		fdput_pos(f);
980 	}
981 
982 	if (ret > 0)
983 		add_wchar(current, ret);
984 	inc_syscw(current);
985 	return ret;
986 }
987 
988 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
989 {
990 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
991 	return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
992 }
993 
994 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
995 			 unsigned long vlen, loff_t pos, rwf_t flags)
996 {
997 	struct fd f;
998 	ssize_t ret = -EBADF;
999 
1000 	if (pos < 0)
1001 		return -EINVAL;
1002 
1003 	f = fdget(fd);
1004 	if (f.file) {
1005 		ret = -ESPIPE;
1006 		if (f.file->f_mode & FMODE_PREAD)
1007 			ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1008 		fdput(f);
1009 	}
1010 
1011 	if (ret > 0)
1012 		add_rchar(current, ret);
1013 	inc_syscr(current);
1014 	return ret;
1015 }
1016 
1017 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
1018 			  unsigned long vlen, loff_t pos, rwf_t flags)
1019 {
1020 	struct fd f;
1021 	ssize_t ret = -EBADF;
1022 
1023 	if (pos < 0)
1024 		return -EINVAL;
1025 
1026 	f = fdget(fd);
1027 	if (f.file) {
1028 		ret = -ESPIPE;
1029 		if (f.file->f_mode & FMODE_PWRITE)
1030 			ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1031 		fdput(f);
1032 	}
1033 
1034 	if (ret > 0)
1035 		add_wchar(current, ret);
1036 	inc_syscw(current);
1037 	return ret;
1038 }
1039 
1040 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1041 		unsigned long, vlen)
1042 {
1043 	return do_readv(fd, vec, vlen, 0);
1044 }
1045 
1046 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1047 		unsigned long, vlen)
1048 {
1049 	return do_writev(fd, vec, vlen, 0);
1050 }
1051 
1052 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1053 		unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1054 {
1055 	loff_t pos = pos_from_hilo(pos_h, pos_l);
1056 
1057 	return do_preadv(fd, vec, vlen, pos, 0);
1058 }
1059 
1060 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1061 		unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1062 		rwf_t, flags)
1063 {
1064 	loff_t pos = pos_from_hilo(pos_h, pos_l);
1065 
1066 	if (pos == -1)
1067 		return do_readv(fd, vec, vlen, flags);
1068 
1069 	return do_preadv(fd, vec, vlen, pos, flags);
1070 }
1071 
1072 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1073 		unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1074 {
1075 	loff_t pos = pos_from_hilo(pos_h, pos_l);
1076 
1077 	return do_pwritev(fd, vec, vlen, pos, 0);
1078 }
1079 
1080 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1081 		unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1082 		rwf_t, flags)
1083 {
1084 	loff_t pos = pos_from_hilo(pos_h, pos_l);
1085 
1086 	if (pos == -1)
1087 		return do_writev(fd, vec, vlen, flags);
1088 
1089 	return do_pwritev(fd, vec, vlen, pos, flags);
1090 }
1091 
1092 /*
1093  * Various compat syscalls.  Note that they all pretend to take a native
1094  * iovec - import_iovec will properly treat those as compat_iovecs based on
1095  * in_compat_syscall().
1096  */
1097 #ifdef CONFIG_COMPAT
1098 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1099 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1100 		const struct iovec __user *, vec,
1101 		unsigned long, vlen, loff_t, pos)
1102 {
1103 	return do_preadv(fd, vec, vlen, pos, 0);
1104 }
1105 #endif
1106 
1107 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1108 		const struct iovec __user *, vec,
1109 		compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1110 {
1111 	loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1112 
1113 	return do_preadv(fd, vec, vlen, pos, 0);
1114 }
1115 
1116 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1117 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1118 		const struct iovec __user *, vec,
1119 		unsigned long, vlen, loff_t, pos, rwf_t, flags)
1120 {
1121 	if (pos == -1)
1122 		return do_readv(fd, vec, vlen, flags);
1123 	return do_preadv(fd, vec, vlen, pos, flags);
1124 }
1125 #endif
1126 
1127 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1128 		const struct iovec __user *, vec,
1129 		compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1130 		rwf_t, flags)
1131 {
1132 	loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1133 
1134 	if (pos == -1)
1135 		return do_readv(fd, vec, vlen, flags);
1136 	return do_preadv(fd, vec, vlen, pos, flags);
1137 }
1138 
1139 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1140 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1141 		const struct iovec __user *, vec,
1142 		unsigned long, vlen, loff_t, pos)
1143 {
1144 	return do_pwritev(fd, vec, vlen, pos, 0);
1145 }
1146 #endif
1147 
1148 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1149 		const struct iovec __user *,vec,
1150 		compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1151 {
1152 	loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1153 
1154 	return do_pwritev(fd, vec, vlen, pos, 0);
1155 }
1156 
1157 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1158 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1159 		const struct iovec __user *, vec,
1160 		unsigned long, vlen, loff_t, pos, rwf_t, flags)
1161 {
1162 	if (pos == -1)
1163 		return do_writev(fd, vec, vlen, flags);
1164 	return do_pwritev(fd, vec, vlen, pos, flags);
1165 }
1166 #endif
1167 
1168 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1169 		const struct iovec __user *,vec,
1170 		compat_ulong_t, vlen, u32, pos_low, u32, pos_high, rwf_t, flags)
1171 {
1172 	loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1173 
1174 	if (pos == -1)
1175 		return do_writev(fd, vec, vlen, flags);
1176 	return do_pwritev(fd, vec, vlen, pos, flags);
1177 }
1178 #endif /* CONFIG_COMPAT */
1179 
1180 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1181 		  	   size_t count, loff_t max)
1182 {
1183 	struct fd in, out;
1184 	struct inode *in_inode, *out_inode;
1185 	struct pipe_inode_info *opipe;
1186 	loff_t pos;
1187 	loff_t out_pos;
1188 	ssize_t retval;
1189 	int fl;
1190 
1191 	/*
1192 	 * Get input file, and verify that it is ok..
1193 	 */
1194 	retval = -EBADF;
1195 	in = fdget(in_fd);
1196 	if (!in.file)
1197 		goto out;
1198 	if (!(in.file->f_mode & FMODE_READ))
1199 		goto fput_in;
1200 	retval = -ESPIPE;
1201 	if (!ppos) {
1202 		pos = in.file->f_pos;
1203 	} else {
1204 		pos = *ppos;
1205 		if (!(in.file->f_mode & FMODE_PREAD))
1206 			goto fput_in;
1207 	}
1208 	retval = rw_verify_area(READ, in.file, &pos, count);
1209 	if (retval < 0)
1210 		goto fput_in;
1211 	if (count > MAX_RW_COUNT)
1212 		count =  MAX_RW_COUNT;
1213 
1214 	/*
1215 	 * Get output file, and verify that it is ok..
1216 	 */
1217 	retval = -EBADF;
1218 	out = fdget(out_fd);
1219 	if (!out.file)
1220 		goto fput_in;
1221 	if (!(out.file->f_mode & FMODE_WRITE))
1222 		goto fput_out;
1223 	in_inode = file_inode(in.file);
1224 	out_inode = file_inode(out.file);
1225 	out_pos = out.file->f_pos;
1226 
1227 	if (!max)
1228 		max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1229 
1230 	if (unlikely(pos + count > max)) {
1231 		retval = -EOVERFLOW;
1232 		if (pos >= max)
1233 			goto fput_out;
1234 		count = max - pos;
1235 	}
1236 
1237 	fl = 0;
1238 #if 0
1239 	/*
1240 	 * We need to debate whether we can enable this or not. The
1241 	 * man page documents EAGAIN return for the output at least,
1242 	 * and the application is arguably buggy if it doesn't expect
1243 	 * EAGAIN on a non-blocking file descriptor.
1244 	 */
1245 	if (in.file->f_flags & O_NONBLOCK)
1246 		fl = SPLICE_F_NONBLOCK;
1247 #endif
1248 	opipe = get_pipe_info(out.file, true);
1249 	if (!opipe) {
1250 		retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1251 		if (retval < 0)
1252 			goto fput_out;
1253 		file_start_write(out.file);
1254 		retval = do_splice_direct(in.file, &pos, out.file, &out_pos,
1255 					  count, fl);
1256 		file_end_write(out.file);
1257 	} else {
1258 		if (out.file->f_flags & O_NONBLOCK)
1259 			fl |= SPLICE_F_NONBLOCK;
1260 
1261 		retval = splice_file_to_pipe(in.file, opipe, &pos, count, fl);
1262 	}
1263 
1264 	if (retval > 0) {
1265 		add_rchar(current, retval);
1266 		add_wchar(current, retval);
1267 		fsnotify_access(in.file);
1268 		fsnotify_modify(out.file);
1269 		out.file->f_pos = out_pos;
1270 		if (ppos)
1271 			*ppos = pos;
1272 		else
1273 			in.file->f_pos = pos;
1274 	}
1275 
1276 	inc_syscr(current);
1277 	inc_syscw(current);
1278 	if (pos > max)
1279 		retval = -EOVERFLOW;
1280 
1281 fput_out:
1282 	fdput(out);
1283 fput_in:
1284 	fdput(in);
1285 out:
1286 	return retval;
1287 }
1288 
1289 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1290 {
1291 	loff_t pos;
1292 	off_t off;
1293 	ssize_t ret;
1294 
1295 	if (offset) {
1296 		if (unlikely(get_user(off, offset)))
1297 			return -EFAULT;
1298 		pos = off;
1299 		ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1300 		if (unlikely(put_user(pos, offset)))
1301 			return -EFAULT;
1302 		return ret;
1303 	}
1304 
1305 	return do_sendfile(out_fd, in_fd, NULL, count, 0);
1306 }
1307 
1308 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1309 {
1310 	loff_t pos;
1311 	ssize_t ret;
1312 
1313 	if (offset) {
1314 		if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1315 			return -EFAULT;
1316 		ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1317 		if (unlikely(put_user(pos, offset)))
1318 			return -EFAULT;
1319 		return ret;
1320 	}
1321 
1322 	return do_sendfile(out_fd, in_fd, NULL, count, 0);
1323 }
1324 
1325 #ifdef CONFIG_COMPAT
1326 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1327 		compat_off_t __user *, offset, compat_size_t, count)
1328 {
1329 	loff_t pos;
1330 	off_t off;
1331 	ssize_t ret;
1332 
1333 	if (offset) {
1334 		if (unlikely(get_user(off, offset)))
1335 			return -EFAULT;
1336 		pos = off;
1337 		ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1338 		if (unlikely(put_user(pos, offset)))
1339 			return -EFAULT;
1340 		return ret;
1341 	}
1342 
1343 	return do_sendfile(out_fd, in_fd, NULL, count, 0);
1344 }
1345 
1346 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1347 		compat_loff_t __user *, offset, compat_size_t, count)
1348 {
1349 	loff_t pos;
1350 	ssize_t ret;
1351 
1352 	if (offset) {
1353 		if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1354 			return -EFAULT;
1355 		ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1356 		if (unlikely(put_user(pos, offset)))
1357 			return -EFAULT;
1358 		return ret;
1359 	}
1360 
1361 	return do_sendfile(out_fd, in_fd, NULL, count, 0);
1362 }
1363 #endif
1364 
1365 /**
1366  * generic_copy_file_range - copy data between two files
1367  * @file_in:	file structure to read from
1368  * @pos_in:	file offset to read from
1369  * @file_out:	file structure to write data to
1370  * @pos_out:	file offset to write data to
1371  * @len:	amount of data to copy
1372  * @flags:	copy flags
1373  *
1374  * This is a generic filesystem helper to copy data from one file to another.
1375  * It has no constraints on the source or destination file owners - the files
1376  * can belong to different superblocks and different filesystem types. Short
1377  * copies are allowed.
1378  *
1379  * This should be called from the @file_out filesystem, as per the
1380  * ->copy_file_range() method.
1381  *
1382  * Returns the number of bytes copied or a negative error indicating the
1383  * failure.
1384  */
1385 
1386 ssize_t generic_copy_file_range(struct file *file_in, loff_t pos_in,
1387 				struct file *file_out, loff_t pos_out,
1388 				size_t len, unsigned int flags)
1389 {
1390 	lockdep_assert(sb_write_started(file_inode(file_out)->i_sb));
1391 
1392 	return do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1393 				len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1394 }
1395 EXPORT_SYMBOL(generic_copy_file_range);
1396 
1397 /*
1398  * Performs necessary checks before doing a file copy
1399  *
1400  * Can adjust amount of bytes to copy via @req_count argument.
1401  * Returns appropriate error code that caller should return or
1402  * zero in case the copy should be allowed.
1403  */
1404 static int generic_copy_file_checks(struct file *file_in, loff_t pos_in,
1405 				    struct file *file_out, loff_t pos_out,
1406 				    size_t *req_count, unsigned int flags)
1407 {
1408 	struct inode *inode_in = file_inode(file_in);
1409 	struct inode *inode_out = file_inode(file_out);
1410 	uint64_t count = *req_count;
1411 	loff_t size_in;
1412 	int ret;
1413 
1414 	ret = generic_file_rw_checks(file_in, file_out);
1415 	if (ret)
1416 		return ret;
1417 
1418 	/*
1419 	 * We allow some filesystems to handle cross sb copy, but passing
1420 	 * a file of the wrong filesystem type to filesystem driver can result
1421 	 * in an attempt to dereference the wrong type of ->private_data, so
1422 	 * avoid doing that until we really have a good reason.
1423 	 *
1424 	 * nfs and cifs define several different file_system_type structures
1425 	 * and several different sets of file_operations, but they all end up
1426 	 * using the same ->copy_file_range() function pointer.
1427 	 */
1428 	if (flags & COPY_FILE_SPLICE) {
1429 		/* cross sb splice is allowed */
1430 	} else if (file_out->f_op->copy_file_range) {
1431 		if (file_in->f_op->copy_file_range !=
1432 		    file_out->f_op->copy_file_range)
1433 			return -EXDEV;
1434 	} else if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb) {
1435 		return -EXDEV;
1436 	}
1437 
1438 	/* Don't touch certain kinds of inodes */
1439 	if (IS_IMMUTABLE(inode_out))
1440 		return -EPERM;
1441 
1442 	if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1443 		return -ETXTBSY;
1444 
1445 	/* Ensure offsets don't wrap. */
1446 	if (pos_in + count < pos_in || pos_out + count < pos_out)
1447 		return -EOVERFLOW;
1448 
1449 	/* Shorten the copy to EOF */
1450 	size_in = i_size_read(inode_in);
1451 	if (pos_in >= size_in)
1452 		count = 0;
1453 	else
1454 		count = min(count, size_in - (uint64_t)pos_in);
1455 
1456 	ret = generic_write_check_limits(file_out, pos_out, &count);
1457 	if (ret)
1458 		return ret;
1459 
1460 	/* Don't allow overlapped copying within the same file. */
1461 	if (inode_in == inode_out &&
1462 	    pos_out + count > pos_in &&
1463 	    pos_out < pos_in + count)
1464 		return -EINVAL;
1465 
1466 	*req_count = count;
1467 	return 0;
1468 }
1469 
1470 /*
1471  * copy_file_range() differs from regular file read and write in that it
1472  * specifically allows return partial success.  When it does so is up to
1473  * the copy_file_range method.
1474  */
1475 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1476 			    struct file *file_out, loff_t pos_out,
1477 			    size_t len, unsigned int flags)
1478 {
1479 	ssize_t ret;
1480 	bool splice = flags & COPY_FILE_SPLICE;
1481 
1482 	if (flags & ~COPY_FILE_SPLICE)
1483 		return -EINVAL;
1484 
1485 	ret = generic_copy_file_checks(file_in, pos_in, file_out, pos_out, &len,
1486 				       flags);
1487 	if (unlikely(ret))
1488 		return ret;
1489 
1490 	ret = rw_verify_area(READ, file_in, &pos_in, len);
1491 	if (unlikely(ret))
1492 		return ret;
1493 
1494 	ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1495 	if (unlikely(ret))
1496 		return ret;
1497 
1498 	if (len == 0)
1499 		return 0;
1500 
1501 	file_start_write(file_out);
1502 
1503 	/*
1504 	 * Cloning is supported by more file systems, so we implement copy on
1505 	 * same sb using clone, but for filesystems where both clone and copy
1506 	 * are supported (e.g. nfs,cifs), we only call the copy method.
1507 	 */
1508 	if (!splice && file_out->f_op->copy_file_range) {
1509 		ret = file_out->f_op->copy_file_range(file_in, pos_in,
1510 						      file_out, pos_out,
1511 						      len, flags);
1512 		goto done;
1513 	}
1514 
1515 	if (!splice && file_in->f_op->remap_file_range &&
1516 	    file_inode(file_in)->i_sb == file_inode(file_out)->i_sb) {
1517 		ret = file_in->f_op->remap_file_range(file_in, pos_in,
1518 				file_out, pos_out,
1519 				min_t(loff_t, MAX_RW_COUNT, len),
1520 				REMAP_FILE_CAN_SHORTEN);
1521 		if (ret > 0)
1522 			goto done;
1523 	}
1524 
1525 	/*
1526 	 * We can get here for same sb copy of filesystems that do not implement
1527 	 * ->copy_file_range() in case filesystem does not support clone or in
1528 	 * case filesystem supports clone but rejected the clone request (e.g.
1529 	 * because it was not block aligned).
1530 	 *
1531 	 * In both cases, fall back to kernel copy so we are able to maintain a
1532 	 * consistent story about which filesystems support copy_file_range()
1533 	 * and which filesystems do not, that will allow userspace tools to
1534 	 * make consistent desicions w.r.t using copy_file_range().
1535 	 *
1536 	 * We also get here if caller (e.g. nfsd) requested COPY_FILE_SPLICE.
1537 	 */
1538 	ret = generic_copy_file_range(file_in, pos_in, file_out, pos_out, len,
1539 				      flags);
1540 
1541 done:
1542 	if (ret > 0) {
1543 		fsnotify_access(file_in);
1544 		add_rchar(current, ret);
1545 		fsnotify_modify(file_out);
1546 		add_wchar(current, ret);
1547 	}
1548 
1549 	inc_syscr(current);
1550 	inc_syscw(current);
1551 
1552 	file_end_write(file_out);
1553 
1554 	return ret;
1555 }
1556 EXPORT_SYMBOL(vfs_copy_file_range);
1557 
1558 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1559 		int, fd_out, loff_t __user *, off_out,
1560 		size_t, len, unsigned int, flags)
1561 {
1562 	loff_t pos_in;
1563 	loff_t pos_out;
1564 	struct fd f_in;
1565 	struct fd f_out;
1566 	ssize_t ret = -EBADF;
1567 
1568 	f_in = fdget(fd_in);
1569 	if (!f_in.file)
1570 		goto out2;
1571 
1572 	f_out = fdget(fd_out);
1573 	if (!f_out.file)
1574 		goto out1;
1575 
1576 	ret = -EFAULT;
1577 	if (off_in) {
1578 		if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1579 			goto out;
1580 	} else {
1581 		pos_in = f_in.file->f_pos;
1582 	}
1583 
1584 	if (off_out) {
1585 		if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1586 			goto out;
1587 	} else {
1588 		pos_out = f_out.file->f_pos;
1589 	}
1590 
1591 	ret = -EINVAL;
1592 	if (flags != 0)
1593 		goto out;
1594 
1595 	ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1596 				  flags);
1597 	if (ret > 0) {
1598 		pos_in += ret;
1599 		pos_out += ret;
1600 
1601 		if (off_in) {
1602 			if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1603 				ret = -EFAULT;
1604 		} else {
1605 			f_in.file->f_pos = pos_in;
1606 		}
1607 
1608 		if (off_out) {
1609 			if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1610 				ret = -EFAULT;
1611 		} else {
1612 			f_out.file->f_pos = pos_out;
1613 		}
1614 	}
1615 
1616 out:
1617 	fdput(f_out);
1618 out1:
1619 	fdput(f_in);
1620 out2:
1621 	return ret;
1622 }
1623 
1624 /*
1625  * Don't operate on ranges the page cache doesn't support, and don't exceed the
1626  * LFS limits.  If pos is under the limit it becomes a short access.  If it
1627  * exceeds the limit we return -EFBIG.
1628  */
1629 int generic_write_check_limits(struct file *file, loff_t pos, loff_t *count)
1630 {
1631 	struct inode *inode = file->f_mapping->host;
1632 	loff_t max_size = inode->i_sb->s_maxbytes;
1633 	loff_t limit = rlimit(RLIMIT_FSIZE);
1634 
1635 	if (limit != RLIM_INFINITY) {
1636 		if (pos >= limit) {
1637 			send_sig(SIGXFSZ, current, 0);
1638 			return -EFBIG;
1639 		}
1640 		*count = min(*count, limit - pos);
1641 	}
1642 
1643 	if (!(file->f_flags & O_LARGEFILE))
1644 		max_size = MAX_NON_LFS;
1645 
1646 	if (unlikely(pos >= max_size))
1647 		return -EFBIG;
1648 
1649 	*count = min(*count, max_size - pos);
1650 
1651 	return 0;
1652 }
1653 
1654 /* Like generic_write_checks(), but takes size of write instead of iter. */
1655 int generic_write_checks_count(struct kiocb *iocb, loff_t *count)
1656 {
1657 	struct file *file = iocb->ki_filp;
1658 	struct inode *inode = file->f_mapping->host;
1659 
1660 	if (IS_SWAPFILE(inode))
1661 		return -ETXTBSY;
1662 
1663 	if (!*count)
1664 		return 0;
1665 
1666 	if (iocb->ki_flags & IOCB_APPEND)
1667 		iocb->ki_pos = i_size_read(inode);
1668 
1669 	if ((iocb->ki_flags & IOCB_NOWAIT) &&
1670 	    !((iocb->ki_flags & IOCB_DIRECT) ||
1671 	      (file->f_mode & FMODE_BUF_WASYNC)))
1672 		return -EINVAL;
1673 
1674 	return generic_write_check_limits(iocb->ki_filp, iocb->ki_pos, count);
1675 }
1676 EXPORT_SYMBOL(generic_write_checks_count);
1677 
1678 /*
1679  * Performs necessary checks before doing a write
1680  *
1681  * Can adjust writing position or amount of bytes to write.
1682  * Returns appropriate error code that caller should return or
1683  * zero in case that write should be allowed.
1684  */
1685 ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
1686 {
1687 	loff_t count = iov_iter_count(from);
1688 	int ret;
1689 
1690 	ret = generic_write_checks_count(iocb, &count);
1691 	if (ret)
1692 		return ret;
1693 
1694 	iov_iter_truncate(from, count);
1695 	return iov_iter_count(from);
1696 }
1697 EXPORT_SYMBOL(generic_write_checks);
1698 
1699 /*
1700  * Performs common checks before doing a file copy/clone
1701  * from @file_in to @file_out.
1702  */
1703 int generic_file_rw_checks(struct file *file_in, struct file *file_out)
1704 {
1705 	struct inode *inode_in = file_inode(file_in);
1706 	struct inode *inode_out = file_inode(file_out);
1707 
1708 	/* Don't copy dirs, pipes, sockets... */
1709 	if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1710 		return -EISDIR;
1711 	if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1712 		return -EINVAL;
1713 
1714 	if (!(file_in->f_mode & FMODE_READ) ||
1715 	    !(file_out->f_mode & FMODE_WRITE) ||
1716 	    (file_out->f_flags & O_APPEND))
1717 		return -EBADF;
1718 
1719 	return 0;
1720 }
1721