projects/linux·blob v6.13 · refs

1810 lines · 43.3 KiB

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
28const 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
35EXPORT_SYMBOL(generic_ro_fops);
36
37static inline bool unsigned_offsets(struct file *file)
38{
39 return file->f_op->fop_flags & FOP_UNSIGNED_OFFSET;
40}
41
42/**
43 * vfs_setpos_cookie - update the file offset for lseek and reset cookie
44 * @file: file structure in question
45 * @offset: file offset to seek to
46 * @maxsize: maximum file size
47 * @cookie: cookie to reset
48 *
49 * Update the file offset to the value specified by @offset if the given
50 * offset is valid and it is not equal to the current file offset and
51 * reset the specified cookie to indicate that a seek happened.
52 *
53 * Return the specified offset on success and -EINVAL on invalid offset.
54 */
55static loff_t vfs_setpos_cookie(struct file *file, loff_t offset,
56 loff_t maxsize, u64 *cookie)
57{
58 if (offset < 0 && !unsigned_offsets(file))
59 return -EINVAL;
60 if (offset > maxsize)
61 return -EINVAL;
62
63 if (offset != file->f_pos) {
64 file->f_pos = offset;
65 if (cookie)
66 *cookie = 0;
67 }
68 return offset;
69}
70
71/**
72 * vfs_setpos - update the file offset for lseek
73 * @file: file structure in question
74 * @offset: file offset to seek to
75 * @maxsize: maximum file size
76 *
77 * This is a low-level filesystem helper for updating the file offset to
78 * the value specified by @offset if the given offset is valid and it is
79 * not equal to the current file offset.
80 *
81 * Return the specified offset on success and -EINVAL on invalid offset.
82 */
83loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
84{
85 return vfs_setpos_cookie(file, offset, maxsize, NULL);
86}
87EXPORT_SYMBOL(vfs_setpos);
88
89/**
90 * must_set_pos - check whether f_pos has to be updated
91 * @file: file to seek on
92 * @offset: offset to use
93 * @whence: type of seek operation
94 * @eof: end of file
95 *
96 * Check whether f_pos needs to be updated and update @offset according
97 * to @whence.
98 *
99 * Return: 0 if f_pos doesn't need to be updated, 1 if f_pos has to be
100 * updated, and negative error code on failure.
101 */
102static int must_set_pos(struct file *file, loff_t *offset, int whence, loff_t eof)
103{
104 switch (whence) {
105 case SEEK_END:
106 *offset += eof;
107 break;
108 case SEEK_CUR:
109 /*
110 * Here we special-case the lseek(fd, 0, SEEK_CUR)
111 * position-querying operation. Avoid rewriting the "same"
112 * f_pos value back to the file because a concurrent read(),
113 * write() or lseek() might have altered it
114 */
115 if (*offset == 0) {
116 *offset = file->f_pos;
117 return 0;
118 }
119 break;
120 case SEEK_DATA:
121 /*
122 * In the generic case the entire file is data, so as long as
123 * offset isn't at the end of the file then the offset is data.
124 */
125 if ((unsigned long long)*offset >= eof)
126 return -ENXIO;
127 break;
128 case SEEK_HOLE:
129 /*
130 * There is a virtual hole at the end of the file, so as long as
131 * offset isn't i_size or larger, return i_size.
132 */
133 if ((unsigned long long)*offset >= eof)
134 return -ENXIO;
135 *offset = eof;
136 break;
137 }
138
139 return 1;
140}
141
142/**
143 * generic_file_llseek_size - generic llseek implementation for regular files
144 * @file: file structure to seek on
145 * @offset: file offset to seek to
146 * @whence: type of seek
147 * @maxsize: max size of this file in file system
148 * @eof: offset used for SEEK_END position
149 *
150 * This is a variant of generic_file_llseek that allows passing in a custom
151 * maximum file size and a custom EOF position, for e.g. hashed directories
152 *
153 * Synchronization:
154 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
155 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
156 * read/writes behave like SEEK_SET against seeks.
157 */
158loff_t
159generic_file_llseek_size(struct file *file, loff_t offset, int whence,
160 loff_t maxsize, loff_t eof)
161{
162 int ret;
163
164 ret = must_set_pos(file, &offset, whence, eof);
165 if (ret < 0)
166 return ret;
167 if (ret == 0)
168 return offset;
169
170 if (whence == SEEK_CUR) {
171 /*
172 * f_lock protects against read/modify/write race with
173 * other SEEK_CURs. Note that parallel writes and reads
174 * behave like SEEK_SET.
175 */
176 guard(spinlock)(&file->f_lock);
177 return vfs_setpos(file, file->f_pos + offset, maxsize);
178 }
179
180 return vfs_setpos(file, offset, maxsize);
181}
182EXPORT_SYMBOL(generic_file_llseek_size);
183
184/**
185 * generic_llseek_cookie - versioned llseek implementation
186 * @file: file structure to seek on
187 * @offset: file offset to seek to
188 * @whence: type of seek
189 * @cookie: cookie to update
190 *
191 * See generic_file_llseek for a general description and locking assumptions.
192 *
193 * In contrast to generic_file_llseek, this function also resets a
194 * specified cookie to indicate a seek took place.
195 */
196loff_t generic_llseek_cookie(struct file *file, loff_t offset, int whence,
197 u64 *cookie)
198{
199 struct inode *inode = file->f_mapping->host;
200 loff_t maxsize = inode->i_sb->s_maxbytes;
201 loff_t eof = i_size_read(inode);
202 int ret;
203
204 if (WARN_ON_ONCE(!cookie))
205 return -EINVAL;
206
207 /*
208 * Require that this is only used for directories that guarantee
209 * synchronization between readdir and seek so that an update to
210 * @cookie is correctly synchronized with concurrent readdir.
211 */
212 if (WARN_ON_ONCE(!(file->f_mode & FMODE_ATOMIC_POS)))
213 return -EINVAL;
214
215 ret = must_set_pos(file, &offset, whence, eof);
216 if (ret < 0)
217 return ret;
218 if (ret == 0)
219 return offset;
220
221 /* No need to hold f_lock because we know that f_pos_lock is held. */
222 if (whence == SEEK_CUR)
223 return vfs_setpos_cookie(file, file->f_pos + offset, maxsize, cookie);
224
225 return vfs_setpos_cookie(file, offset, maxsize, cookie);
226}
227EXPORT_SYMBOL(generic_llseek_cookie);
228
229/**
230 * generic_file_llseek - generic llseek implementation for regular files
231 * @file: file structure to seek on
232 * @offset: file offset to seek to
233 * @whence: type of seek
234 *
235 * This is a generic implemenation of ->llseek useable for all normal local
236 * filesystems. It just updates the file offset to the value specified by
237 * @offset and @whence.
238 */
239loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
240{
241 struct inode *inode = file->f_mapping->host;
242
243 return generic_file_llseek_size(file, offset, whence,
244 inode->i_sb->s_maxbytes,
245 i_size_read(inode));
246}
247EXPORT_SYMBOL(generic_file_llseek);
248
249/**
250 * fixed_size_llseek - llseek implementation for fixed-sized devices
251 * @file: file structure to seek on
252 * @offset: file offset to seek to
253 * @whence: type of seek
254 * @size: size of the file
255 *
256 */
257loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
258{
259 switch (whence) {
260 case SEEK_SET: case SEEK_CUR: case SEEK_END:
261 return generic_file_llseek_size(file, offset, whence,
262 size, size);
263 default:
264 return -EINVAL;
265 }
266}
267EXPORT_SYMBOL(fixed_size_llseek);
268
269/**
270 * no_seek_end_llseek - llseek implementation for fixed-sized devices
271 * @file: file structure to seek on
272 * @offset: file offset to seek to
273 * @whence: type of seek
274 *
275 */
276loff_t no_seek_end_llseek(struct file *file, loff_t offset, int whence)
277{
278 switch (whence) {
279 case SEEK_SET: case SEEK_CUR:
280 return generic_file_llseek_size(file, offset, whence,
281 OFFSET_MAX, 0);
282 default:
283 return -EINVAL;
284 }
285}
286EXPORT_SYMBOL(no_seek_end_llseek);
287
288/**
289 * no_seek_end_llseek_size - llseek implementation for fixed-sized devices
290 * @file: file structure to seek on
291 * @offset: file offset to seek to
292 * @whence: type of seek
293 * @size: maximal offset allowed
294 *
295 */
296loff_t no_seek_end_llseek_size(struct file *file, loff_t offset, int whence, loff_t size)
297{
298 switch (whence) {
299 case SEEK_SET: case SEEK_CUR:
300 return generic_file_llseek_size(file, offset, whence,
301 size, 0);
302 default:
303 return -EINVAL;
304 }
305}
306EXPORT_SYMBOL(no_seek_end_llseek_size);
307
308/**
309 * noop_llseek - No Operation Performed llseek implementation
310 * @file: file structure to seek on
311 * @offset: file offset to seek to
312 * @whence: type of seek
313 *
314 * This is an implementation of ->llseek useable for the rare special case when
315 * userspace expects the seek to succeed but the (device) file is actually not
316 * able to perform the seek. In this case you use noop_llseek() instead of
317 * falling back to the default implementation of ->llseek.
318 */
319loff_t noop_llseek(struct file *file, loff_t offset, int whence)
320{
321 return file->f_pos;
322}
323EXPORT_SYMBOL(noop_llseek);
324
325loff_t default_llseek(struct file *file, loff_t offset, int whence)
326{
327 struct inode *inode = file_inode(file);
328 loff_t retval;
329
330 inode_lock(inode);
331 switch (whence) {
332 case SEEK_END:
333 offset += i_size_read(inode);
334 break;
335 case SEEK_CUR:
336 if (offset == 0) {
337 retval = file->f_pos;
338 goto out;
339 }
340 offset += file->f_pos;
341 break;
342 case SEEK_DATA:
343 /*
344 * In the generic case the entire file is data, so as
345 * long as offset isn't at the end of the file then the
346 * offset is data.
347 */
348 if (offset >= inode->i_size) {
349 retval = -ENXIO;
350 goto out;
351 }
352 break;
353 case SEEK_HOLE:
354 /*
355 * There is a virtual hole at the end of the file, so
356 * as long as offset isn't i_size or larger, return
357 * i_size.
358 */
359 if (offset >= inode->i_size) {
360 retval = -ENXIO;
361 goto out;
362 }
363 offset = inode->i_size;
364 break;
365 }
366 retval = -EINVAL;
367 if (offset >= 0 || unsigned_offsets(file)) {
368 if (offset != file->f_pos)
369 file->f_pos = offset;
370 retval = offset;
371 }
372out:
373 inode_unlock(inode);
374 return retval;
375}
376EXPORT_SYMBOL(default_llseek);
377
378loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
379{
380 if (!(file->f_mode & FMODE_LSEEK))
381 return -ESPIPE;
382 return file->f_op->llseek(file, offset, whence);
383}
384EXPORT_SYMBOL(vfs_llseek);
385
386static off_t ksys_lseek(unsigned int fd, off_t offset, unsigned int whence)
387{
388 off_t retval;
389 CLASS(fd_pos, f)(fd);
390 if (fd_empty(f))
391 return -EBADF;
392
393 retval = -EINVAL;
394 if (whence <= SEEK_MAX) {
395 loff_t res = vfs_llseek(fd_file(f), offset, whence);
396 retval = res;
397 if (res != (loff_t)retval)
398 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
399 }
400 return retval;
401}
402
403SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
404{
405 return ksys_lseek(fd, offset, whence);
406}
407
408#ifdef CONFIG_COMPAT
409COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
410{
411 return ksys_lseek(fd, offset, whence);
412}
413#endif
414
415#if !defined(CONFIG_64BIT) || defined(CONFIG_COMPAT) || \
416 defined(__ARCH_WANT_SYS_LLSEEK)
417SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
418 unsigned long, offset_low, loff_t __user *, result,
419 unsigned int, whence)
420{
421 int retval;
422 CLASS(fd_pos, f)(fd);
423 loff_t offset;
424
425 if (fd_empty(f))
426 return -EBADF;
427
428 if (whence > SEEK_MAX)
429 return -EINVAL;
430
431 offset = vfs_llseek(fd_file(f), ((loff_t) offset_high << 32) | offset_low,
432 whence);
433
434 retval = (int)offset;
435 if (offset >= 0) {
436 retval = -EFAULT;
437 if (!copy_to_user(result, &offset, sizeof(offset)))
438 retval = 0;
439 }
440 return retval;
441}
442#endif
443
444int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
445{
446 int mask = read_write == READ ? MAY_READ : MAY_WRITE;
447 int ret;
448
449 if (unlikely((ssize_t) count < 0))
450 return -EINVAL;
451
452 if (ppos) {
453 loff_t pos = *ppos;
454
455 if (unlikely(pos < 0)) {
456 if (!unsigned_offsets(file))
457 return -EINVAL;
458 if (count >= -pos) /* both values are in 0..LLONG_MAX */
459 return -EOVERFLOW;
460 } else if (unlikely((loff_t) (pos + count) < 0)) {
461 if (!unsigned_offsets(file))
462 return -EINVAL;
463 }
464 }
465
466 ret = security_file_permission(file, mask);
467 if (ret)
468 return ret;
469
470 return fsnotify_file_area_perm(file, mask, ppos, count);
471}
472EXPORT_SYMBOL(rw_verify_area);
473
474static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
475{
476 struct kiocb kiocb;
477 struct iov_iter iter;
478 ssize_t ret;
479
480 init_sync_kiocb(&kiocb, filp);
481 kiocb.ki_pos = (ppos ? *ppos : 0);
482 iov_iter_ubuf(&iter, ITER_DEST, buf, len);
483
484 ret = filp->f_op->read_iter(&kiocb, &iter);
485 BUG_ON(ret == -EIOCBQUEUED);
486 if (ppos)
487 *ppos = kiocb.ki_pos;
488 return ret;
489}
490
491static int warn_unsupported(struct file *file, const char *op)
492{
493 pr_warn_ratelimited(
494 "kernel %s not supported for file %pD4 (pid: %d comm: %.20s)\n",
495 op, file, current->pid, current->comm);
496 return -EINVAL;
497}
498
499ssize_t __kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
500{
501 struct kvec iov = {
502 .iov_base = buf,
503 .iov_len = min_t(size_t, count, MAX_RW_COUNT),
504 };
505 struct kiocb kiocb;
506 struct iov_iter iter;
507 ssize_t ret;
508
509 if (WARN_ON_ONCE(!(file->f_mode & FMODE_READ)))
510 return -EINVAL;
511 if (!(file->f_mode & FMODE_CAN_READ))
512 return -EINVAL;
513 /*
514 * Also fail if ->read_iter and ->read are both wired up as that
515 * implies very convoluted semantics.
516 */
517 if (unlikely(!file->f_op->read_iter || file->f_op->read))
518 return warn_unsupported(file, "read");
519
520 init_sync_kiocb(&kiocb, file);
521 kiocb.ki_pos = pos ? *pos : 0;
522 iov_iter_kvec(&iter, ITER_DEST, &iov, 1, iov.iov_len);
523 ret = file->f_op->read_iter(&kiocb, &iter);
524 if (ret > 0) {
525 if (pos)
526 *pos = kiocb.ki_pos;
527 fsnotify_access(file);
528 add_rchar(current, ret);
529 }
530 inc_syscr(current);
531 return ret;
532}
533
534ssize_t kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
535{
536 ssize_t ret;
537
538 ret = rw_verify_area(READ, file, pos, count);
539 if (ret)
540 return ret;
541 return __kernel_read(file, buf, count, pos);
542}
543EXPORT_SYMBOL(kernel_read);
544
545ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
546{
547 ssize_t ret;
548
549 if (!(file->f_mode & FMODE_READ))
550 return -EBADF;
551 if (!(file->f_mode & FMODE_CAN_READ))
552 return -EINVAL;
553 if (unlikely(!access_ok(buf, count)))
554 return -EFAULT;
555
556 ret = rw_verify_area(READ, file, pos, count);
557 if (ret)
558 return ret;
559 if (count > MAX_RW_COUNT)
560 count = MAX_RW_COUNT;
561
562 if (file->f_op->read)
563 ret = file->f_op->read(file, buf, count, pos);
564 else if (file->f_op->read_iter)
565 ret = new_sync_read(file, buf, count, pos);
566 else
567 ret = -EINVAL;
568 if (ret > 0) {
569 fsnotify_access(file);
570 add_rchar(current, ret);
571 }
572 inc_syscr(current);
573 return ret;
574}
575
576static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
577{
578 struct kiocb kiocb;
579 struct iov_iter iter;
580 ssize_t ret;
581
582 init_sync_kiocb(&kiocb, filp);
583 kiocb.ki_pos = (ppos ? *ppos : 0);
584 iov_iter_ubuf(&iter, ITER_SOURCE, (void __user *)buf, len);
585
586 ret = filp->f_op->write_iter(&kiocb, &iter);
587 BUG_ON(ret == -EIOCBQUEUED);
588 if (ret > 0 && ppos)
589 *ppos = kiocb.ki_pos;
590 return ret;
591}
592
593/* caller is responsible for file_start_write/file_end_write */
594ssize_t __kernel_write_iter(struct file *file, struct iov_iter *from, loff_t *pos)
595{
596 struct kiocb kiocb;
597 ssize_t ret;
598
599 if (WARN_ON_ONCE(!(file->f_mode & FMODE_WRITE)))
600 return -EBADF;
601 if (!(file->f_mode & FMODE_CAN_WRITE))
602 return -EINVAL;
603 /*
604 * Also fail if ->write_iter and ->write are both wired up as that
605 * implies very convoluted semantics.
606 */
607 if (unlikely(!file->f_op->write_iter || file->f_op->write))
608 return warn_unsupported(file, "write");
609
610 init_sync_kiocb(&kiocb, file);
611 kiocb.ki_pos = pos ? *pos : 0;
612 ret = file->f_op->write_iter(&kiocb, from);
613 if (ret > 0) {
614 if (pos)
615 *pos = kiocb.ki_pos;
616 fsnotify_modify(file);
617 add_wchar(current, ret);
618 }
619 inc_syscw(current);
620 return ret;
621}
622
623/* caller is responsible for file_start_write/file_end_write */
624ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t *pos)
625{
626 struct kvec iov = {
627 .iov_base = (void *)buf,
628 .iov_len = min_t(size_t, count, MAX_RW_COUNT),
629 };
630 struct iov_iter iter;
631 iov_iter_kvec(&iter, ITER_SOURCE, &iov, 1, iov.iov_len);
632 return __kernel_write_iter(file, &iter, pos);
633}
634/*
635 * This "EXPORT_SYMBOL_GPL()" is more of a "EXPORT_SYMBOL_DONTUSE()",
636 * but autofs is one of the few internal kernel users that actually
637 * wants this _and_ can be built as a module. So we need to export
638 * this symbol for autofs, even though it really isn't appropriate
639 * for any other kernel modules.
640 */
641EXPORT_SYMBOL_GPL(__kernel_write);
642
643ssize_t kernel_write(struct file *file, const void *buf, size_t count,
644 loff_t *pos)
645{
646 ssize_t ret;
647
648 ret = rw_verify_area(WRITE, file, pos, count);
649 if (ret)
650 return ret;
651
652 file_start_write(file);
653 ret = __kernel_write(file, buf, count, pos);
654 file_end_write(file);
655 return ret;
656}
657EXPORT_SYMBOL(kernel_write);
658
659ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
660{
661 ssize_t ret;
662
663 if (!(file->f_mode & FMODE_WRITE))
664 return -EBADF;
665 if (!(file->f_mode & FMODE_CAN_WRITE))
666 return -EINVAL;
667 if (unlikely(!access_ok(buf, count)))
668 return -EFAULT;
669
670 ret = rw_verify_area(WRITE, file, pos, count);
671 if (ret)
672 return ret;
673 if (count > MAX_RW_COUNT)
674 count = MAX_RW_COUNT;
675 file_start_write(file);
676 if (file->f_op->write)
677 ret = file->f_op->write(file, buf, count, pos);
678 else if (file->f_op->write_iter)
679 ret = new_sync_write(file, buf, count, pos);
680 else
681 ret = -EINVAL;
682 if (ret > 0) {
683 fsnotify_modify(file);
684 add_wchar(current, ret);
685 }
686 inc_syscw(current);
687 file_end_write(file);
688 return ret;
689}
690
691/* file_ppos returns &file->f_pos or NULL if file is stream */
692static inline loff_t *file_ppos(struct file *file)
693{
694 return file->f_mode & FMODE_STREAM ? NULL : &file->f_pos;
695}
696
697ssize_t ksys_read(unsigned int fd, char __user *buf, size_t count)
698{
699 CLASS(fd_pos, f)(fd);
700 ssize_t ret = -EBADF;
701
702 if (!fd_empty(f)) {
703 loff_t pos, *ppos = file_ppos(fd_file(f));
704 if (ppos) {
705 pos = *ppos;
706 ppos = &pos;
707 }
708 ret = vfs_read(fd_file(f), buf, count, ppos);
709 if (ret >= 0 && ppos)
710 fd_file(f)->f_pos = pos;
711 }
712 return ret;
713}
714
715SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
716{
717 return ksys_read(fd, buf, count);
718}
719
720ssize_t ksys_write(unsigned int fd, const char __user *buf, size_t count)
721{
722 CLASS(fd_pos, f)(fd);
723 ssize_t ret = -EBADF;
724
725 if (!fd_empty(f)) {
726 loff_t pos, *ppos = file_ppos(fd_file(f));
727 if (ppos) {
728 pos = *ppos;
729 ppos = &pos;
730 }
731 ret = vfs_write(fd_file(f), buf, count, ppos);
732 if (ret >= 0 && ppos)
733 fd_file(f)->f_pos = pos;
734 }
735
736 return ret;
737}
738
739SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
740 size_t, count)
741{
742 return ksys_write(fd, buf, count);
743}
744
745ssize_t ksys_pread64(unsigned int fd, char __user *buf, size_t count,
746 loff_t pos)
747{
748 if (pos < 0)
749 return -EINVAL;
750
751 CLASS(fd, f)(fd);
752 if (fd_empty(f))
753 return -EBADF;
754
755 if (fd_file(f)->f_mode & FMODE_PREAD)
756 return vfs_read(fd_file(f), buf, count, &pos);
757
758 return -ESPIPE;
759}
760
761SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
762 size_t, count, loff_t, pos)
763{
764 return ksys_pread64(fd, buf, count, pos);
765}
766
767#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_PREAD64)
768COMPAT_SYSCALL_DEFINE5(pread64, unsigned int, fd, char __user *, buf,
769 size_t, count, compat_arg_u64_dual(pos))
770{
771 return ksys_pread64(fd, buf, count, compat_arg_u64_glue(pos));
772}
773#endif
774
775ssize_t ksys_pwrite64(unsigned int fd, const char __user *buf,
776 size_t count, loff_t pos)
777{
778 if (pos < 0)
779 return -EINVAL;
780
781 CLASS(fd, f)(fd);
782 if (fd_empty(f))
783 return -EBADF;
784
785 if (fd_file(f)->f_mode & FMODE_PWRITE)
786 return vfs_write(fd_file(f), buf, count, &pos);
787
788 return -ESPIPE;
789}
790
791SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
792 size_t, count, loff_t, pos)
793{
794 return ksys_pwrite64(fd, buf, count, pos);
795}
796
797#if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_PWRITE64)
798COMPAT_SYSCALL_DEFINE5(pwrite64, unsigned int, fd, const char __user *, buf,
799 size_t, count, compat_arg_u64_dual(pos))
800{
801 return ksys_pwrite64(fd, buf, count, compat_arg_u64_glue(pos));
802}
803#endif
804
805static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
806 loff_t *ppos, int type, rwf_t flags)
807{
808 struct kiocb kiocb;
809 ssize_t ret;
810
811 init_sync_kiocb(&kiocb, filp);
812 ret = kiocb_set_rw_flags(&kiocb, flags, type);
813 if (ret)
814 return ret;
815 kiocb.ki_pos = (ppos ? *ppos : 0);
816
817 if (type == READ)
818 ret = filp->f_op->read_iter(&kiocb, iter);
819 else
820 ret = filp->f_op->write_iter(&kiocb, iter);
821 BUG_ON(ret == -EIOCBQUEUED);
822 if (ppos)
823 *ppos = kiocb.ki_pos;
824 return ret;
825}
826
827/* Do it by hand, with file-ops */
828static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
829 loff_t *ppos, int type, rwf_t flags)
830{
831 ssize_t ret = 0;
832
833 if (flags & ~RWF_HIPRI)
834 return -EOPNOTSUPP;
835
836 while (iov_iter_count(iter)) {
837 ssize_t nr;
838
839 if (type == READ) {
840 nr = filp->f_op->read(filp, iter_iov_addr(iter),
841 iter_iov_len(iter), ppos);
842 } else {
843 nr = filp->f_op->write(filp, iter_iov_addr(iter),
844 iter_iov_len(iter), ppos);
845 }
846
847 if (nr < 0) {
848 if (!ret)
849 ret = nr;
850 break;
851 }
852 ret += nr;
853 if (nr != iter_iov_len(iter))
854 break;
855 iov_iter_advance(iter, nr);
856 }
857
858 return ret;
859}
860
861ssize_t vfs_iocb_iter_read(struct file *file, struct kiocb *iocb,
862 struct iov_iter *iter)
863{
864 size_t tot_len;
865 ssize_t ret = 0;
866
867 if (!file->f_op->read_iter)
868 return -EINVAL;
869 if (!(file->f_mode & FMODE_READ))
870 return -EBADF;
871 if (!(file->f_mode & FMODE_CAN_READ))
872 return -EINVAL;
873
874 tot_len = iov_iter_count(iter);
875 if (!tot_len)
876 goto out;
877 ret = rw_verify_area(READ, file, &iocb->ki_pos, tot_len);
878 if (ret < 0)
879 return ret;
880
881 ret = file->f_op->read_iter(iocb, iter);
882out:
883 if (ret >= 0)
884 fsnotify_access(file);
885 return ret;
886}
887EXPORT_SYMBOL(vfs_iocb_iter_read);
888
889ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos,
890 rwf_t flags)
891{
892 size_t tot_len;
893 ssize_t ret = 0;
894
895 if (!file->f_op->read_iter)
896 return -EINVAL;
897 if (!(file->f_mode & FMODE_READ))
898 return -EBADF;
899 if (!(file->f_mode & FMODE_CAN_READ))
900 return -EINVAL;
901
902 tot_len = iov_iter_count(iter);
903 if (!tot_len)
904 goto out;
905 ret = rw_verify_area(READ, file, ppos, tot_len);
906 if (ret < 0)
907 return ret;
908
909 ret = do_iter_readv_writev(file, iter, ppos, READ, flags);
910out:
911 if (ret >= 0)
912 fsnotify_access(file);
913 return ret;
914}
915EXPORT_SYMBOL(vfs_iter_read);
916
917/*
918 * Caller is responsible for calling kiocb_end_write() on completion
919 * if async iocb was queued.
920 */
921ssize_t vfs_iocb_iter_write(struct file *file, struct kiocb *iocb,
922 struct iov_iter *iter)
923{
924 size_t tot_len;
925 ssize_t ret = 0;
926
927 if (!file->f_op->write_iter)
928 return -EINVAL;
929 if (!(file->f_mode & FMODE_WRITE))
930 return -EBADF;
931 if (!(file->f_mode & FMODE_CAN_WRITE))
932 return -EINVAL;
933
934 tot_len = iov_iter_count(iter);
935 if (!tot_len)
936 return 0;
937 ret = rw_verify_area(WRITE, file, &iocb->ki_pos, tot_len);
938 if (ret < 0)
939 return ret;
940
941 kiocb_start_write(iocb);
942 ret = file->f_op->write_iter(iocb, iter);
943 if (ret != -EIOCBQUEUED)
944 kiocb_end_write(iocb);
945 if (ret > 0)
946 fsnotify_modify(file);
947
948 return ret;
949}
950EXPORT_SYMBOL(vfs_iocb_iter_write);
951
952ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos,
953 rwf_t flags)
954{
955 size_t tot_len;
956 ssize_t ret;
957
958 if (!(file->f_mode & FMODE_WRITE))
959 return -EBADF;
960 if (!(file->f_mode & FMODE_CAN_WRITE))
961 return -EINVAL;
962 if (!file->f_op->write_iter)
963 return -EINVAL;
964
965 tot_len = iov_iter_count(iter);
966 if (!tot_len)
967 return 0;
968
969 ret = rw_verify_area(WRITE, file, ppos, tot_len);
970 if (ret < 0)
971 return ret;
972
973 file_start_write(file);
974 ret = do_iter_readv_writev(file, iter, ppos, WRITE, flags);
975 if (ret > 0)
976 fsnotify_modify(file);
977 file_end_write(file);
978
979 return ret;
980}
981EXPORT_SYMBOL(vfs_iter_write);
982
983static ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
984 unsigned long vlen, loff_t *pos, rwf_t flags)
985{
986 struct iovec iovstack[UIO_FASTIOV];
987 struct iovec *iov = iovstack;
988 struct iov_iter iter;
989 size_t tot_len;
990 ssize_t ret = 0;
991
992 if (!(file->f_mode & FMODE_READ))
993 return -EBADF;
994 if (!(file->f_mode & FMODE_CAN_READ))
995 return -EINVAL;
996
997 ret = import_iovec(ITER_DEST, vec, vlen, ARRAY_SIZE(iovstack), &iov,
998 &iter);
999 if (ret < 0)
1000 return ret;
1001
1002 tot_len = iov_iter_count(&iter);
1003 if (!tot_len)
1004 goto out;
1005
1006 ret = rw_verify_area(READ, file, pos, tot_len);
1007 if (ret < 0)
1008 goto out;
1009
1010 if (file->f_op->read_iter)
1011 ret = do_iter_readv_writev(file, &iter, pos, READ, flags);
1012 else
1013 ret = do_loop_readv_writev(file, &iter, pos, READ, flags);
1014out:
1015 if (ret >= 0)
1016 fsnotify_access(file);
1017 kfree(iov);
1018 return ret;
1019}
1020
1021static ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
1022 unsigned long vlen, loff_t *pos, rwf_t flags)
1023{
1024 struct iovec iovstack[UIO_FASTIOV];
1025 struct iovec *iov = iovstack;
1026 struct iov_iter iter;
1027 size_t tot_len;
1028 ssize_t ret = 0;
1029
1030 if (!(file->f_mode & FMODE_WRITE))
1031 return -EBADF;
1032 if (!(file->f_mode & FMODE_CAN_WRITE))
1033 return -EINVAL;
1034
1035 ret = import_iovec(ITER_SOURCE, vec, vlen, ARRAY_SIZE(iovstack), &iov,
1036 &iter);
1037 if (ret < 0)
1038 return ret;
1039
1040 tot_len = iov_iter_count(&iter);
1041 if (!tot_len)
1042 goto out;
1043
1044 ret = rw_verify_area(WRITE, file, pos, tot_len);
1045 if (ret < 0)
1046 goto out;
1047
1048 file_start_write(file);
1049 if (file->f_op->write_iter)
1050 ret = do_iter_readv_writev(file, &iter, pos, WRITE, flags);
1051 else
1052 ret = do_loop_readv_writev(file, &iter, pos, WRITE, flags);
1053 if (ret > 0)
1054 fsnotify_modify(file);
1055 file_end_write(file);
1056out:
1057 kfree(iov);
1058 return ret;
1059}
1060
1061static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
1062 unsigned long vlen, rwf_t flags)
1063{
1064 CLASS(fd_pos, f)(fd);
1065 ssize_t ret = -EBADF;
1066
1067 if (!fd_empty(f)) {
1068 loff_t pos, *ppos = file_ppos(fd_file(f));
1069 if (ppos) {
1070 pos = *ppos;
1071 ppos = &pos;
1072 }
1073 ret = vfs_readv(fd_file(f), vec, vlen, ppos, flags);
1074 if (ret >= 0 && ppos)
1075 fd_file(f)->f_pos = pos;
1076 }
1077
1078 if (ret > 0)
1079 add_rchar(current, ret);
1080 inc_syscr(current);
1081 return ret;
1082}
1083
1084static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
1085 unsigned long vlen, rwf_t flags)
1086{
1087 CLASS(fd_pos, f)(fd);
1088 ssize_t ret = -EBADF;
1089
1090 if (!fd_empty(f)) {
1091 loff_t pos, *ppos = file_ppos(fd_file(f));
1092 if (ppos) {
1093 pos = *ppos;
1094 ppos = &pos;
1095 }
1096 ret = vfs_writev(fd_file(f), vec, vlen, ppos, flags);
1097 if (ret >= 0 && ppos)
1098 fd_file(f)->f_pos = pos;
1099 }
1100
1101 if (ret > 0)
1102 add_wchar(current, ret);
1103 inc_syscw(current);
1104 return ret;
1105}
1106
1107static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
1108{
1109#define HALF_LONG_BITS (BITS_PER_LONG / 2)
1110 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
1111}
1112
1113static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
1114 unsigned long vlen, loff_t pos, rwf_t flags)
1115{
1116 ssize_t ret = -EBADF;
1117
1118 if (pos < 0)
1119 return -EINVAL;
1120
1121 CLASS(fd, f)(fd);
1122 if (!fd_empty(f)) {
1123 ret = -ESPIPE;
1124 if (fd_file(f)->f_mode & FMODE_PREAD)
1125 ret = vfs_readv(fd_file(f), vec, vlen, &pos, flags);
1126 }
1127
1128 if (ret > 0)
1129 add_rchar(current, ret);
1130 inc_syscr(current);
1131 return ret;
1132}
1133
1134static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
1135 unsigned long vlen, loff_t pos, rwf_t flags)
1136{
1137 ssize_t ret = -EBADF;
1138
1139 if (pos < 0)
1140 return -EINVAL;
1141
1142 CLASS(fd, f)(fd);
1143 if (!fd_empty(f)) {
1144 ret = -ESPIPE;
1145 if (fd_file(f)->f_mode & FMODE_PWRITE)
1146 ret = vfs_writev(fd_file(f), vec, vlen, &pos, flags);
1147 }
1148
1149 if (ret > 0)
1150 add_wchar(current, ret);
1151 inc_syscw(current);
1152 return ret;
1153}
1154
1155SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1156 unsigned long, vlen)
1157{
1158 return do_readv(fd, vec, vlen, 0);
1159}
1160
1161SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1162 unsigned long, vlen)
1163{
1164 return do_writev(fd, vec, vlen, 0);
1165}
1166
1167SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1168 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1169{
1170 loff_t pos = pos_from_hilo(pos_h, pos_l);
1171
1172 return do_preadv(fd, vec, vlen, pos, 0);
1173}
1174
1175SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1176 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1177 rwf_t, flags)
1178{
1179 loff_t pos = pos_from_hilo(pos_h, pos_l);
1180
1181 if (pos == -1)
1182 return do_readv(fd, vec, vlen, flags);
1183
1184 return do_preadv(fd, vec, vlen, pos, flags);
1185}
1186
1187SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1188 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1189{
1190 loff_t pos = pos_from_hilo(pos_h, pos_l);
1191
1192 return do_pwritev(fd, vec, vlen, pos, 0);
1193}
1194
1195SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1196 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1197 rwf_t, flags)
1198{
1199 loff_t pos = pos_from_hilo(pos_h, pos_l);
1200
1201 if (pos == -1)
1202 return do_writev(fd, vec, vlen, flags);
1203
1204 return do_pwritev(fd, vec, vlen, pos, flags);
1205}
1206
1207/*
1208 * Various compat syscalls. Note that they all pretend to take a native
1209 * iovec - import_iovec will properly treat those as compat_iovecs based on
1210 * in_compat_syscall().
1211 */
1212#ifdef CONFIG_COMPAT
1213#ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1214COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1215 const struct iovec __user *, vec,
1216 unsigned long, vlen, loff_t, pos)
1217{
1218 return do_preadv(fd, vec, vlen, pos, 0);
1219}
1220#endif
1221
1222COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1223 const struct iovec __user *, vec,
1224 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1225{
1226 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1227
1228 return do_preadv(fd, vec, vlen, pos, 0);
1229}
1230
1231#ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1232COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1233 const struct iovec __user *, vec,
1234 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1235{
1236 if (pos == -1)
1237 return do_readv(fd, vec, vlen, flags);
1238 return do_preadv(fd, vec, vlen, pos, flags);
1239}
1240#endif
1241
1242COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1243 const struct iovec __user *, vec,
1244 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1245 rwf_t, flags)
1246{
1247 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1248
1249 if (pos == -1)
1250 return do_readv(fd, vec, vlen, flags);
1251 return do_preadv(fd, vec, vlen, pos, flags);
1252}
1253
1254#ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1255COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1256 const struct iovec __user *, vec,
1257 unsigned long, vlen, loff_t, pos)
1258{
1259 return do_pwritev(fd, vec, vlen, pos, 0);
1260}
1261#endif
1262
1263COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1264 const struct iovec __user *,vec,
1265 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1266{
1267 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1268
1269 return do_pwritev(fd, vec, vlen, pos, 0);
1270}
1271
1272#ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1273COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1274 const struct iovec __user *, vec,
1275 unsigned long, vlen, loff_t, pos, rwf_t, flags)
1276{
1277 if (pos == -1)
1278 return do_writev(fd, vec, vlen, flags);
1279 return do_pwritev(fd, vec, vlen, pos, flags);
1280}
1281#endif
1282
1283COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1284 const struct iovec __user *,vec,
1285 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, rwf_t, flags)
1286{
1287 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1288
1289 if (pos == -1)
1290 return do_writev(fd, vec, vlen, flags);
1291 return do_pwritev(fd, vec, vlen, pos, flags);
1292}
1293#endif /* CONFIG_COMPAT */
1294
1295static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1296 size_t count, loff_t max)
1297{
1298 struct inode *in_inode, *out_inode;
1299 struct pipe_inode_info *opipe;
1300 loff_t pos;
1301 loff_t out_pos;
1302 ssize_t retval;
1303 int fl;
1304
1305 /*
1306 * Get input file, and verify that it is ok..
1307 */
1308 CLASS(fd, in)(in_fd);
1309 if (fd_empty(in))
1310 return -EBADF;
1311 if (!(fd_file(in)->f_mode & FMODE_READ))
1312 return -EBADF;
1313 if (!ppos) {
1314 pos = fd_file(in)->f_pos;
1315 } else {
1316 pos = *ppos;
1317 if (!(fd_file(in)->f_mode & FMODE_PREAD))
1318 return -ESPIPE;
1319 }
1320 retval = rw_verify_area(READ, fd_file(in), &pos, count);
1321 if (retval < 0)
1322 return retval;
1323 if (count > MAX_RW_COUNT)
1324 count = MAX_RW_COUNT;
1325
1326 /*
1327 * Get output file, and verify that it is ok..
1328 */
1329 CLASS(fd, out)(out_fd);
1330 if (fd_empty(out))
1331 return -EBADF;
1332 if (!(fd_file(out)->f_mode & FMODE_WRITE))
1333 return -EBADF;
1334 in_inode = file_inode(fd_file(in));
1335 out_inode = file_inode(fd_file(out));
1336 out_pos = fd_file(out)->f_pos;
1337
1338 if (!max)
1339 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1340
1341 if (unlikely(pos + count > max)) {
1342 if (pos >= max)
1343 return -EOVERFLOW;
1344 count = max - pos;
1345 }
1346
1347 fl = 0;
1348#if 0
1349 /*
1350 * We need to debate whether we can enable this or not. The
1351 * man page documents EAGAIN return for the output at least,
1352 * and the application is arguably buggy if it doesn't expect
1353 * EAGAIN on a non-blocking file descriptor.
1354 */
1355 if (fd_file(in)->f_flags & O_NONBLOCK)
1356 fl = SPLICE_F_NONBLOCK;
1357#endif
1358 opipe = get_pipe_info(fd_file(out), true);
1359 if (!opipe) {
1360 retval = rw_verify_area(WRITE, fd_file(out), &out_pos, count);
1361 if (retval < 0)
1362 return retval;
1363 retval = do_splice_direct(fd_file(in), &pos, fd_file(out), &out_pos,
1364 count, fl);
1365 } else {
1366 if (fd_file(out)->f_flags & O_NONBLOCK)
1367 fl |= SPLICE_F_NONBLOCK;
1368
1369 retval = splice_file_to_pipe(fd_file(in), opipe, &pos, count, fl);
1370 }
1371
1372 if (retval > 0) {
1373 add_rchar(current, retval);
1374 add_wchar(current, retval);
1375 fsnotify_access(fd_file(in));
1376 fsnotify_modify(fd_file(out));
1377 fd_file(out)->f_pos = out_pos;
1378 if (ppos)
1379 *ppos = pos;
1380 else
1381 fd_file(in)->f_pos = pos;
1382 }
1383
1384 inc_syscr(current);
1385 inc_syscw(current);
1386 if (pos > max)
1387 retval = -EOVERFLOW;
1388 return retval;
1389}
1390
1391SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1392{
1393 loff_t pos;
1394 off_t off;
1395 ssize_t ret;
1396
1397 if (offset) {
1398 if (unlikely(get_user(off, offset)))
1399 return -EFAULT;
1400 pos = off;
1401 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1402 if (unlikely(put_user(pos, offset)))
1403 return -EFAULT;
1404 return ret;
1405 }
1406
1407 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1408}
1409
1410SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1411{
1412 loff_t pos;
1413 ssize_t ret;
1414
1415 if (offset) {
1416 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1417 return -EFAULT;
1418 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1419 if (unlikely(put_user(pos, offset)))
1420 return -EFAULT;
1421 return ret;
1422 }
1423
1424 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1425}
1426
1427#ifdef CONFIG_COMPAT
1428COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1429 compat_off_t __user *, offset, compat_size_t, count)
1430{
1431 loff_t pos;
1432 off_t off;
1433 ssize_t ret;
1434
1435 if (offset) {
1436 if (unlikely(get_user(off, offset)))
1437 return -EFAULT;
1438 pos = off;
1439 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1440 if (unlikely(put_user(pos, offset)))
1441 return -EFAULT;
1442 return ret;
1443 }
1444
1445 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1446}
1447
1448COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1449 compat_loff_t __user *, offset, compat_size_t, count)
1450{
1451 loff_t pos;
1452 ssize_t ret;
1453
1454 if (offset) {
1455 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1456 return -EFAULT;
1457 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1458 if (unlikely(put_user(pos, offset)))
1459 return -EFAULT;
1460 return ret;
1461 }
1462
1463 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1464}
1465#endif
1466
1467/*
1468 * Performs necessary checks before doing a file copy
1469 *
1470 * Can adjust amount of bytes to copy via @req_count argument.
1471 * Returns appropriate error code that caller should return or
1472 * zero in case the copy should be allowed.
1473 */
1474static int generic_copy_file_checks(struct file *file_in, loff_t pos_in,
1475 struct file *file_out, loff_t pos_out,
1476 size_t *req_count, unsigned int flags)
1477{
1478 struct inode *inode_in = file_inode(file_in);
1479 struct inode *inode_out = file_inode(file_out);
1480 uint64_t count = *req_count;
1481 loff_t size_in;
1482 int ret;
1483
1484 ret = generic_file_rw_checks(file_in, file_out);
1485 if (ret)
1486 return ret;
1487
1488 /*
1489 * We allow some filesystems to handle cross sb copy, but passing
1490 * a file of the wrong filesystem type to filesystem driver can result
1491 * in an attempt to dereference the wrong type of ->private_data, so
1492 * avoid doing that until we really have a good reason.
1493 *
1494 * nfs and cifs define several different file_system_type structures
1495 * and several different sets of file_operations, but they all end up
1496 * using the same ->copy_file_range() function pointer.
1497 */
1498 if (flags & COPY_FILE_SPLICE) {
1499 /* cross sb splice is allowed */
1500 } else if (file_out->f_op->copy_file_range) {
1501 if (file_in->f_op->copy_file_range !=
1502 file_out->f_op->copy_file_range)
1503 return -EXDEV;
1504 } else if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb) {
1505 return -EXDEV;
1506 }
1507
1508 /* Don't touch certain kinds of inodes */
1509 if (IS_IMMUTABLE(inode_out))
1510 return -EPERM;
1511
1512 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1513 return -ETXTBSY;
1514
1515 /* Ensure offsets don't wrap. */
1516 if (pos_in + count < pos_in || pos_out + count < pos_out)
1517 return -EOVERFLOW;
1518
1519 /* Shorten the copy to EOF */
1520 size_in = i_size_read(inode_in);
1521 if (pos_in >= size_in)
1522 count = 0;
1523 else
1524 count = min(count, size_in - (uint64_t)pos_in);
1525
1526 ret = generic_write_check_limits(file_out, pos_out, &count);
1527 if (ret)
1528 return ret;
1529
1530 /* Don't allow overlapped copying within the same file. */
1531 if (inode_in == inode_out &&
1532 pos_out + count > pos_in &&
1533 pos_out < pos_in + count)
1534 return -EINVAL;
1535
1536 *req_count = count;
1537 return 0;
1538}
1539
1540/*
1541 * copy_file_range() differs from regular file read and write in that it
1542 * specifically allows return partial success. When it does so is up to
1543 * the copy_file_range method.
1544 */
1545ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1546 struct file *file_out, loff_t pos_out,
1547 size_t len, unsigned int flags)
1548{
1549 ssize_t ret;
1550 bool splice = flags & COPY_FILE_SPLICE;
1551 bool samesb = file_inode(file_in)->i_sb == file_inode(file_out)->i_sb;
1552
1553 if (flags & ~COPY_FILE_SPLICE)
1554 return -EINVAL;
1555
1556 ret = generic_copy_file_checks(file_in, pos_in, file_out, pos_out, &len,
1557 flags);
1558 if (unlikely(ret))
1559 return ret;
1560
1561 ret = rw_verify_area(READ, file_in, &pos_in, len);
1562 if (unlikely(ret))
1563 return ret;
1564
1565 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1566 if (unlikely(ret))
1567 return ret;
1568
1569 if (len == 0)
1570 return 0;
1571
1572 file_start_write(file_out);
1573
1574 /*
1575 * Cloning is supported by more file systems, so we implement copy on
1576 * same sb using clone, but for filesystems where both clone and copy
1577 * are supported (e.g. nfs,cifs), we only call the copy method.
1578 */
1579 if (!splice && file_out->f_op->copy_file_range) {
1580 ret = file_out->f_op->copy_file_range(file_in, pos_in,
1581 file_out, pos_out,
1582 len, flags);
1583 } else if (!splice && file_in->f_op->remap_file_range && samesb) {
1584 ret = file_in->f_op->remap_file_range(file_in, pos_in,
1585 file_out, pos_out,
1586 min_t(loff_t, MAX_RW_COUNT, len),
1587 REMAP_FILE_CAN_SHORTEN);
1588 /* fallback to splice */
1589 if (ret <= 0)
1590 splice = true;
1591 } else if (samesb) {
1592 /* Fallback to splice for same sb copy for backward compat */
1593 splice = true;
1594 }
1595
1596 file_end_write(file_out);
1597
1598 if (!splice)
1599 goto done;
1600
1601 /*
1602 * We can get here for same sb copy of filesystems that do not implement
1603 * ->copy_file_range() in case filesystem does not support clone or in
1604 * case filesystem supports clone but rejected the clone request (e.g.
1605 * because it was not block aligned).
1606 *
1607 * In both cases, fall back to kernel copy so we are able to maintain a
1608 * consistent story about which filesystems support copy_file_range()
1609 * and which filesystems do not, that will allow userspace tools to
1610 * make consistent desicions w.r.t using copy_file_range().
1611 *
1612 * We also get here if caller (e.g. nfsd) requested COPY_FILE_SPLICE
1613 * for server-side-copy between any two sb.
1614 *
1615 * In any case, we call do_splice_direct() and not splice_file_range(),
1616 * without file_start_write() held, to avoid possible deadlocks related
1617 * to splicing from input file, while file_start_write() is held on
1618 * the output file on a different sb.
1619 */
1620 ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1621 min_t(size_t, len, MAX_RW_COUNT), 0);
1622done:
1623 if (ret > 0) {
1624 fsnotify_access(file_in);
1625 add_rchar(current, ret);
1626 fsnotify_modify(file_out);
1627 add_wchar(current, ret);
1628 }
1629
1630 inc_syscr(current);
1631 inc_syscw(current);
1632
1633 return ret;
1634}
1635EXPORT_SYMBOL(vfs_copy_file_range);
1636
1637SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1638 int, fd_out, loff_t __user *, off_out,
1639 size_t, len, unsigned int, flags)
1640{
1641 loff_t pos_in;
1642 loff_t pos_out;
1643 ssize_t ret = -EBADF;
1644
1645 CLASS(fd, f_in)(fd_in);
1646 if (fd_empty(f_in))
1647 return -EBADF;
1648
1649 CLASS(fd, f_out)(fd_out);
1650 if (fd_empty(f_out))
1651 return -EBADF;
1652
1653 if (off_in) {
1654 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1655 return -EFAULT;
1656 } else {
1657 pos_in = fd_file(f_in)->f_pos;
1658 }
1659
1660 if (off_out) {
1661 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1662 return -EFAULT;
1663 } else {
1664 pos_out = fd_file(f_out)->f_pos;
1665 }
1666
1667 if (flags != 0)
1668 return -EINVAL;
1669
1670 ret = vfs_copy_file_range(fd_file(f_in), pos_in, fd_file(f_out), pos_out, len,
1671 flags);
1672 if (ret > 0) {
1673 pos_in += ret;
1674 pos_out += ret;
1675
1676 if (off_in) {
1677 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1678 ret = -EFAULT;
1679 } else {
1680 fd_file(f_in)->f_pos = pos_in;
1681 }
1682
1683 if (off_out) {
1684 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1685 ret = -EFAULT;
1686 } else {
1687 fd_file(f_out)->f_pos = pos_out;
1688 }
1689 }
1690 return ret;
1691}
1692
1693/*
1694 * Don't operate on ranges the page cache doesn't support, and don't exceed the
1695 * LFS limits. If pos is under the limit it becomes a short access. If it
1696 * exceeds the limit we return -EFBIG.
1697 */
1698int generic_write_check_limits(struct file *file, loff_t pos, loff_t *count)
1699{
1700 struct inode *inode = file->f_mapping->host;
1701 loff_t max_size = inode->i_sb->s_maxbytes;
1702 loff_t limit = rlimit(RLIMIT_FSIZE);
1703
1704 if (limit != RLIM_INFINITY) {
1705 if (pos >= limit) {
1706 send_sig(SIGXFSZ, current, 0);
1707 return -EFBIG;
1708 }
1709 *count = min(*count, limit - pos);
1710 }
1711
1712 if (!(file->f_flags & O_LARGEFILE))
1713 max_size = MAX_NON_LFS;
1714
1715 if (unlikely(pos >= max_size))
1716 return -EFBIG;
1717
1718 *count = min(*count, max_size - pos);
1719
1720 return 0;
1721}
1722EXPORT_SYMBOL_GPL(generic_write_check_limits);
1723
1724/* Like generic_write_checks(), but takes size of write instead of iter. */
1725int generic_write_checks_count(struct kiocb *iocb, loff_t *count)
1726{
1727 struct file *file = iocb->ki_filp;
1728 struct inode *inode = file->f_mapping->host;
1729
1730 if (IS_SWAPFILE(inode))
1731 return -ETXTBSY;
1732
1733 if (!*count)
1734 return 0;
1735
1736 if (iocb->ki_flags & IOCB_APPEND)
1737 iocb->ki_pos = i_size_read(inode);
1738
1739 if ((iocb->ki_flags & IOCB_NOWAIT) &&
1740 !((iocb->ki_flags & IOCB_DIRECT) ||
1741 (file->f_op->fop_flags & FOP_BUFFER_WASYNC)))
1742 return -EINVAL;
1743
1744 return generic_write_check_limits(iocb->ki_filp, iocb->ki_pos, count);
1745}
1746EXPORT_SYMBOL(generic_write_checks_count);
1747
1748/*
1749 * Performs necessary checks before doing a write
1750 *
1751 * Can adjust writing position or amount of bytes to write.
1752 * Returns appropriate error code that caller should return or
1753 * zero in case that write should be allowed.
1754 */
1755ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
1756{
1757 loff_t count = iov_iter_count(from);
1758 int ret;
1759
1760 ret = generic_write_checks_count(iocb, &count);
1761 if (ret)
1762 return ret;
1763
1764 iov_iter_truncate(from, count);
1765 return iov_iter_count(from);
1766}
1767EXPORT_SYMBOL(generic_write_checks);
1768
1769/*
1770 * Performs common checks before doing a file copy/clone
1771 * from @file_in to @file_out.
1772 */
1773int generic_file_rw_checks(struct file *file_in, struct file *file_out)
1774{
1775 struct inode *inode_in = file_inode(file_in);
1776 struct inode *inode_out = file_inode(file_out);
1777
1778 /* Don't copy dirs, pipes, sockets... */
1779 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1780 return -EISDIR;
1781 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1782 return -EINVAL;
1783
1784 if (!(file_in->f_mode & FMODE_READ) ||
1785 !(file_out->f_mode & FMODE_WRITE) ||
1786 (file_out->f_flags & O_APPEND))
1787 return -EBADF;
1788
1789 return 0;
1790}
1791
1792int generic_atomic_write_valid(struct kiocb *iocb, struct iov_iter *iter)
1793{
1794 size_t len = iov_iter_count(iter);
1795
1796 if (!iter_is_ubuf(iter))
1797 return -EINVAL;
1798
1799 if (!is_power_of_2(len))
1800 return -EINVAL;
1801
1802 if (!IS_ALIGNED(iocb->ki_pos, len))
1803 return -EINVAL;
1804
1805 if (!(iocb->ki_flags & IOCB_DIRECT))
1806 return -EOPNOTSUPP;
1807
1808 return 0;
1809}
1810EXPORT_SYMBOL_GPL(generic_atomic_write_valid);