?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
rmap: extend try_to_unmap() to be usable by split_huge_page()
[linux/fpc-iii.git] / fs / read_write.c
blobdadf24e5c95bb7004d105f0773430576bfc57cf5
1 /*
2 * linux/fs/read_write.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7 #include <linux/slab.h>
8 #include <linux/stat.h>
9 #include <linux/fcntl.h>
10 #include <linux/file.h>
11 #include <linux/uio.h>
12 #include <linux/fsnotify.h>
13 #include <linux/security.h>
14 #include <linux/export.h>
15 #include <linux/syscalls.h>
16 #include <linux/pagemap.h>
17 #include <linux/splice.h>
18 #include <linux/compat.h>
19 #include <linux/mount.h>
20 #include <linux/fs.h>
21 #include "internal.h"
22
23 #include <asm/uaccess.h>
24 #include <asm/unistd.h>
25
26 typedef ssize_t (*io_fn_t)(struct file *, char __user *, size_t, loff_t *);
27 typedef ssize_t (*iter_fn_t)(struct kiocb *, struct iov_iter *);
28
29 const struct file_operations generic_ro_fops = {
30 .llseek = generic_file_llseek,
31 .read_iter = generic_file_read_iter,
32 .mmap = generic_file_readonly_mmap,
33 .splice_read = generic_file_splice_read,
34 };
35
36 EXPORT_SYMBOL(generic_ro_fops);
37
38 static inline int unsigned_offsets(struct file *file)
39 {
40 return file->f_mode & FMODE_UNSIGNED_OFFSET;
41 }
42
43 /**
44 * vfs_setpos - update the file offset for lseek
45 * @file: file structure in question
46 * @offset: file offset to seek to
47 * @maxsize: maximum file size
48 *
49 * This is a low-level filesystem helper for updating the file offset to
50 * the value specified by @offset if the given offset is valid and it is
51 * not equal to the current file offset.
52 *
53 * Return the specified offset on success and -EINVAL on invalid offset.
54 */
55 loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
56 {
57 if (offset < 0 && !unsigned_offsets(file))
58 return -EINVAL;
59 if (offset > maxsize)
60 return -EINVAL;
61
62 if (offset != file->f_pos) {
63 file->f_pos = offset;
64 file->f_version = 0;
65 }
66 return offset;
67 }
68 EXPORT_SYMBOL(vfs_setpos);
69
70 /**
71 * generic_file_llseek_size - generic llseek implementation for regular files
72 * @file: file structure to seek on
73 * @offset: file offset to seek to
74 * @whence: type of seek
75 * @size: max size of this file in file system
76 * @eof: offset used for SEEK_END position
77 *
78 * This is a variant of generic_file_llseek that allows passing in a custom
79 * maximum file size and a custom EOF position, for e.g. hashed directories
80 *
81 * Synchronization:
82 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
83 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
84 * read/writes behave like SEEK_SET against seeks.
85 */
86 loff_t
87 generic_file_llseek_size(struct file *file, loff_t offset, int whence,
88 loff_t maxsize, loff_t eof)
89 {
90 switch (whence) {
91 case SEEK_END:
92 offset += eof;
93 break;
94 case SEEK_CUR:
95 /*
96 * Here we special-case the lseek(fd, 0, SEEK_CUR)
97 * position-querying operation. Avoid rewriting the "same"
98 * f_pos value back to the file because a concurrent read(),
99 * write() or lseek() might have altered it
100 */
101 if (offset == 0)
102 return file->f_pos;
103 /*
104 * f_lock protects against read/modify/write race with other
105 * SEEK_CURs. Note that parallel writes and reads behave
106 * like SEEK_SET.
107 */
108 spin_lock(&file->f_lock);
109 offset = vfs_setpos(file, file->f_pos + offset, maxsize);
110 spin_unlock(&file->f_lock);
111 return offset;
112 case SEEK_DATA:
113 /*
114 * In the generic case the entire file is data, so as long as
115 * offset isn't at the end of the file then the offset is data.
116 */
117 if (offset >= eof)
118 return -ENXIO;
119 break;
120 case SEEK_HOLE:
121 /*
122 * There is a virtual hole at the end of the file, so as long as
123 * offset isn't i_size or larger, return i_size.
124 */
125 if (offset >= eof)
126 return -ENXIO;
127 offset = eof;
128 break;
129 }
130
131 return vfs_setpos(file, offset, maxsize);
132 }
133 EXPORT_SYMBOL(generic_file_llseek_size);
134
135 /**
136 * generic_file_llseek - generic llseek implementation for regular files
137 * @file: file structure to seek on
138 * @offset: file offset to seek to
139 * @whence: type of seek
140 *
141 * This is a generic implemenation of ->llseek useable for all normal local
142 * filesystems. It just updates the file offset to the value specified by
143 * @offset and @whence.
144 */
145 loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
146 {
147 struct inode *inode = file->f_mapping->host;
148
149 return generic_file_llseek_size(file, offset, whence,
150 inode->i_sb->s_maxbytes,
151 i_size_read(inode));
152 }
153 EXPORT_SYMBOL(generic_file_llseek);
154
155 /**
156 * fixed_size_llseek - llseek implementation for fixed-sized devices
157 * @file: file structure to seek on
158 * @offset: file offset to seek to
159 * @whence: type of seek
160 * @size: size of the file
161 *
162 */
163 loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
164 {
165 switch (whence) {
166 case SEEK_SET: case SEEK_CUR: case SEEK_END:
167 return generic_file_llseek_size(file, offset, whence,
168 size, size);
169 default:
170 return -EINVAL;
171 }
172 }
173 EXPORT_SYMBOL(fixed_size_llseek);
174
175 /**
176 * no_seek_end_llseek - llseek implementation for fixed-sized devices
177 * @file: file structure to seek on
178 * @offset: file offset to seek to
179 * @whence: type of seek
180 *
181 */
182 loff_t no_seek_end_llseek(struct file *file, loff_t offset, int whence)
183 {
184 switch (whence) {
185 case SEEK_SET: case SEEK_CUR:
186 return generic_file_llseek_size(file, offset, whence,
187 OFFSET_MAX, 0);
188 default:
189 return -EINVAL;
190 }
191 }
192 EXPORT_SYMBOL(no_seek_end_llseek);
193
194 /**
195 * no_seek_end_llseek_size - llseek implementation for fixed-sized devices
196 * @file: file structure to seek on
197 * @offset: file offset to seek to
198 * @whence: type of seek
199 * @size: maximal offset allowed
200 *
201 */
202 loff_t no_seek_end_llseek_size(struct file *file, loff_t offset, int whence, loff_t size)
203 {
204 switch (whence) {
205 case SEEK_SET: case SEEK_CUR:
206 return generic_file_llseek_size(file, offset, whence,
207 size, 0);
208 default:
209 return -EINVAL;
210 }
211 }
212 EXPORT_SYMBOL(no_seek_end_llseek_size);
213
214 /**
215 * noop_llseek - No Operation Performed llseek implementation
216 * @file: file structure to seek on
217 * @offset: file offset to seek to
218 * @whence: type of seek
219 *
220 * This is an implementation of ->llseek useable for the rare special case when
221 * userspace expects the seek to succeed but the (device) file is actually not
222 * able to perform the seek. In this case you use noop_llseek() instead of
223 * falling back to the default implementation of ->llseek.
224 */
225 loff_t noop_llseek(struct file *file, loff_t offset, int whence)
226 {
227 return file->f_pos;
228 }
229 EXPORT_SYMBOL(noop_llseek);
230
231 loff_t no_llseek(struct file *file, loff_t offset, int whence)
232 {
233 return -ESPIPE;
234 }
235 EXPORT_SYMBOL(no_llseek);
236
237 loff_t default_llseek(struct file *file, loff_t offset, int whence)
238 {
239 struct inode *inode = file_inode(file);
240 loff_t retval;
241
242 inode_lock(inode);
243 switch (whence) {
244 case SEEK_END:
245 offset += i_size_read(inode);
246 break;
247 case SEEK_CUR:
248 if (offset == 0) {
249 retval = file->f_pos;
250 goto out;
251 }
252 offset += file->f_pos;
253 break;
254 case SEEK_DATA:
255 /*
256 * In the generic case the entire file is data, so as
257 * long as offset isn't at the end of the file then the
258 * offset is data.
259 */
260 if (offset >= inode->i_size) {
261 retval = -ENXIO;
262 goto out;
263 }
264 break;
265 case SEEK_HOLE:
266 /*
267 * There is a virtual hole at the end of the file, so
268 * as long as offset isn't i_size or larger, return
269 * i_size.
270 */
271 if (offset >= inode->i_size) {
272 retval = -ENXIO;
273 goto out;
274 }
275 offset = inode->i_size;
276 break;
277 }
278 retval = -EINVAL;
279 if (offset >= 0 || unsigned_offsets(file)) {
280 if (offset != file->f_pos) {
281 file->f_pos = offset;
282 file->f_version = 0;
283 }
284 retval = offset;
285 }
286 out:
287 inode_unlock(inode);
288 return retval;
289 }
290 EXPORT_SYMBOL(default_llseek);
291
292 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
293 {
294 loff_t (*fn)(struct file *, loff_t, int);
295
296 fn = no_llseek;
297 if (file->f_mode & FMODE_LSEEK) {
298 if (file->f_op->llseek)
299 fn = file->f_op->llseek;
300 }
301 return fn(file, offset, whence);
302 }
303 EXPORT_SYMBOL(vfs_llseek);
304
305 static inline struct fd fdget_pos(int fd)
306 {
307 return __to_fd(__fdget_pos(fd));
308 }
309
310 static inline void fdput_pos(struct fd f)
311 {
312 if (f.flags & FDPUT_POS_UNLOCK)
313 mutex_unlock(&f.file->f_pos_lock);
314 fdput(f);
315 }
316
317 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
318 {
319 off_t retval;
320 struct fd f = fdget_pos(fd);
321 if (!f.file)
322 return -EBADF;
323
324 retval = -EINVAL;
325 if (whence <= SEEK_MAX) {
326 loff_t res = vfs_llseek(f.file, offset, whence);
327 retval = res;
328 if (res != (loff_t)retval)
329 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
330 }
331 fdput_pos(f);
332 return retval;
333 }
334
335 #ifdef CONFIG_COMPAT
336 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
337 {
338 return sys_lseek(fd, offset, whence);
339 }
340 #endif
341
342 #ifdef __ARCH_WANT_SYS_LLSEEK
343 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
344 unsigned long, offset_low, loff_t __user *, result,
345 unsigned int, whence)
346 {
347 int retval;
348 struct fd f = fdget_pos(fd);
349 loff_t offset;
350
351 if (!f.file)
352 return -EBADF;
353
354 retval = -EINVAL;
355 if (whence > SEEK_MAX)
356 goto out_putf;
357
358 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
359 whence);
360
361 retval = (int)offset;
362 if (offset >= 0) {
363 retval = -EFAULT;
364 if (!copy_to_user(result, &offset, sizeof(offset)))
365 retval = 0;
366 }
367 out_putf:
368 fdput_pos(f);
369 return retval;
370 }
371 #endif
372
373 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos)
374 {
375 struct kiocb kiocb;
376 ssize_t ret;
377
378 if (!file->f_op->read_iter)
379 return -EINVAL;
380
381 init_sync_kiocb(&kiocb, file);
382 kiocb.ki_pos = *ppos;
383
384 iter->type |= READ;
385 ret = file->f_op->read_iter(&kiocb, iter);
386 BUG_ON(ret == -EIOCBQUEUED);
387 if (ret > 0)
388 *ppos = kiocb.ki_pos;
389 return ret;
390 }
391 EXPORT_SYMBOL(vfs_iter_read);
392
393 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos)
394 {
395 struct kiocb kiocb;
396 ssize_t ret;
397
398 if (!file->f_op->write_iter)
399 return -EINVAL;
400
401 init_sync_kiocb(&kiocb, file);
402 kiocb.ki_pos = *ppos;
403
404 iter->type |= WRITE;
405 ret = file->f_op->write_iter(&kiocb, iter);
406 BUG_ON(ret == -EIOCBQUEUED);
407 if (ret > 0)
408 *ppos = kiocb.ki_pos;
409 return ret;
410 }
411 EXPORT_SYMBOL(vfs_iter_write);
412
413 /*
414 * rw_verify_area doesn't like huge counts. We limit
415 * them to something that fits in "int" so that others
416 * won't have to do range checks all the time.
417 */
418 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
419 {
420 struct inode *inode;
421 loff_t pos;
422 int retval = -EINVAL;
423
424 inode = file_inode(file);
425 if (unlikely((ssize_t) count < 0))
426 return retval;
427 pos = *ppos;
428 if (unlikely(pos < 0)) {
429 if (!unsigned_offsets(file))
430 return retval;
431 if (count >= -pos) /* both values are in 0..LLONG_MAX */
432 return -EOVERFLOW;
433 } else if (unlikely((loff_t) (pos + count) < 0)) {
434 if (!unsigned_offsets(file))
435 return retval;
436 }
437
438 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
439 retval = locks_mandatory_area(inode, file, pos, pos + count - 1,
440 read_write == READ ? F_RDLCK : F_WRLCK);
441 if (retval < 0)
442 return retval;
443 }
444 retval = security_file_permission(file,
445 read_write == READ ? MAY_READ : MAY_WRITE);
446 if (retval)
447 return retval;
448 return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
449 }
450
451 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
452 {
453 struct iovec iov = { .iov_base = buf, .iov_len = len };
454 struct kiocb kiocb;
455 struct iov_iter iter;
456 ssize_t ret;
457
458 init_sync_kiocb(&kiocb, filp);
459 kiocb.ki_pos = *ppos;
460 iov_iter_init(&iter, READ, &iov, 1, len);
461
462 ret = filp->f_op->read_iter(&kiocb, &iter);
463 BUG_ON(ret == -EIOCBQUEUED);
464 *ppos = kiocb.ki_pos;
465 return ret;
466 }
467
468 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
469 loff_t *pos)
470 {
471 if (file->f_op->read)
472 return file->f_op->read(file, buf, count, pos);
473 else if (file->f_op->read_iter)
474 return new_sync_read(file, buf, count, pos);
475 else
476 return -EINVAL;
477 }
478 EXPORT_SYMBOL(__vfs_read);
479
480 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
481 {
482 ssize_t ret;
483
484 if (!(file->f_mode & FMODE_READ))
485 return -EBADF;
486 if (!(file->f_mode & FMODE_CAN_READ))
487 return -EINVAL;
488 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
489 return -EFAULT;
490
491 ret = rw_verify_area(READ, file, pos, count);
492 if (ret >= 0) {
493 count = ret;
494 ret = __vfs_read(file, buf, count, pos);
495 if (ret > 0) {
496 fsnotify_access(file);
497 add_rchar(current, ret);
498 }
499 inc_syscr(current);
500 }
501
502 return ret;
503 }
504
505 EXPORT_SYMBOL(vfs_read);
506
507 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
508 {
509 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
510 struct kiocb kiocb;
511 struct iov_iter iter;
512 ssize_t ret;
513
514 init_sync_kiocb(&kiocb, filp);
515 kiocb.ki_pos = *ppos;
516 iov_iter_init(&iter, WRITE, &iov, 1, len);
517
518 ret = filp->f_op->write_iter(&kiocb, &iter);
519 BUG_ON(ret == -EIOCBQUEUED);
520 if (ret > 0)
521 *ppos = kiocb.ki_pos;
522 return ret;
523 }
524
525 ssize_t __vfs_write(struct file *file, const char __user *p, size_t count,
526 loff_t *pos)
527 {
528 if (file->f_op->write)
529 return file->f_op->write(file, p, count, pos);
530 else if (file->f_op->write_iter)
531 return new_sync_write(file, p, count, pos);
532 else
533 return -EINVAL;
534 }
535 EXPORT_SYMBOL(__vfs_write);
536
537 ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
538 {
539 mm_segment_t old_fs;
540 const char __user *p;
541 ssize_t ret;
542
543 if (!(file->f_mode & FMODE_CAN_WRITE))
544 return -EINVAL;
545
546 old_fs = get_fs();
547 set_fs(get_ds());
548 p = (__force const char __user *)buf;
549 if (count > MAX_RW_COUNT)
550 count = MAX_RW_COUNT;
551 ret = __vfs_write(file, p, count, pos);
552 set_fs(old_fs);
553 if (ret > 0) {
554 fsnotify_modify(file);
555 add_wchar(current, ret);
556 }
557 inc_syscw(current);
558 return ret;
559 }
560
561 EXPORT_SYMBOL(__kernel_write);
562
563 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
564 {
565 ssize_t ret;
566
567 if (!(file->f_mode & FMODE_WRITE))
568 return -EBADF;
569 if (!(file->f_mode & FMODE_CAN_WRITE))
570 return -EINVAL;
571 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
572 return -EFAULT;
573
574 ret = rw_verify_area(WRITE, file, pos, count);
575 if (ret >= 0) {
576 count = ret;
577 file_start_write(file);
578 ret = __vfs_write(file, buf, count, pos);
579 if (ret > 0) {
580 fsnotify_modify(file);
581 add_wchar(current, ret);
582 }
583 inc_syscw(current);
584 file_end_write(file);
585 }
586
587 return ret;
588 }
589
590 EXPORT_SYMBOL(vfs_write);
591
592 static inline loff_t file_pos_read(struct file *file)
593 {
594 return file->f_pos;
595 }
596
597 static inline void file_pos_write(struct file *file, loff_t pos)
598 {
599 file->f_pos = pos;
600 }
601
602 SYSCALL_DEFINE3(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 = file_pos_read(f.file);
609 ret = vfs_read(f.file, buf, count, &pos);
610 if (ret >= 0)
611 file_pos_write(f.file, pos);
612 fdput_pos(f);
613 }
614 return ret;
615 }
616
617 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
618 size_t, count)
619 {
620 struct fd f = fdget_pos(fd);
621 ssize_t ret = -EBADF;
622
623 if (f.file) {
624 loff_t pos = file_pos_read(f.file);
625 ret = vfs_write(f.file, buf, count, &pos);
626 if (ret >= 0)
627 file_pos_write(f.file, pos);
628 fdput_pos(f);
629 }
630
631 return ret;
632 }
633
634 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
635 size_t, count, loff_t, pos)
636 {
637 struct fd f;
638 ssize_t ret = -EBADF;
639
640 if (pos < 0)
641 return -EINVAL;
642
643 f = fdget(fd);
644 if (f.file) {
645 ret = -ESPIPE;
646 if (f.file->f_mode & FMODE_PREAD)
647 ret = vfs_read(f.file, buf, count, &pos);
648 fdput(f);
649 }
650
651 return ret;
652 }
653
654 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
655 size_t, count, loff_t, pos)
656 {
657 struct fd f;
658 ssize_t ret = -EBADF;
659
660 if (pos < 0)
661 return -EINVAL;
662
663 f = fdget(fd);
664 if (f.file) {
665 ret = -ESPIPE;
666 if (f.file->f_mode & FMODE_PWRITE)
667 ret = vfs_write(f.file, buf, count, &pos);
668 fdput(f);
669 }
670
671 return ret;
672 }
673
674 /*
675 * Reduce an iovec's length in-place. Return the resulting number of segments
676 */
677 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
678 {
679 unsigned long seg = 0;
680 size_t len = 0;
681
682 while (seg < nr_segs) {
683 seg++;
684 if (len + iov->iov_len >= to) {
685 iov->iov_len = to - len;
686 break;
687 }
688 len += iov->iov_len;
689 iov++;
690 }
691 return seg;
692 }
693 EXPORT_SYMBOL(iov_shorten);
694
695 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
696 loff_t *ppos, iter_fn_t fn)
697 {
698 struct kiocb kiocb;
699 ssize_t ret;
700
701 init_sync_kiocb(&kiocb, filp);
702 kiocb.ki_pos = *ppos;
703
704 ret = fn(&kiocb, iter);
705 BUG_ON(ret == -EIOCBQUEUED);
706 *ppos = kiocb.ki_pos;
707 return ret;
708 }
709
710 /* Do it by hand, with file-ops */
711 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
712 loff_t *ppos, io_fn_t fn)
713 {
714 ssize_t ret = 0;
715
716 while (iov_iter_count(iter)) {
717 struct iovec iovec = iov_iter_iovec(iter);
718 ssize_t nr;
719
720 nr = fn(filp, iovec.iov_base, iovec.iov_len, ppos);
721
722 if (nr < 0) {
723 if (!ret)
724 ret = nr;
725 break;
726 }
727 ret += nr;
728 if (nr != iovec.iov_len)
729 break;
730 iov_iter_advance(iter, nr);
731 }
732
733 return ret;
734 }
735
736 /* A write operation does a read from user space and vice versa */
737 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
738
739 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
740 unsigned long nr_segs, unsigned long fast_segs,
741 struct iovec *fast_pointer,
742 struct iovec **ret_pointer)
743 {
744 unsigned long seg;
745 ssize_t ret;
746 struct iovec *iov = fast_pointer;
747
748 /*
749 * SuS says "The readv() function *may* fail if the iovcnt argument
750 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
751 * traditionally returned zero for zero segments, so...
752 */
753 if (nr_segs == 0) {
754 ret = 0;
755 goto out;
756 }
757
758 /*
759 * First get the "struct iovec" from user memory and
760 * verify all the pointers
761 */
762 if (nr_segs > UIO_MAXIOV) {
763 ret = -EINVAL;
764 goto out;
765 }
766 if (nr_segs > fast_segs) {
767 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
768 if (iov == NULL) {
769 ret = -ENOMEM;
770 goto out;
771 }
772 }
773 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
774 ret = -EFAULT;
775 goto out;
776 }
777
778 /*
779 * According to the Single Unix Specification we should return EINVAL
780 * if an element length is < 0 when cast to ssize_t or if the
781 * total length would overflow the ssize_t return value of the
782 * system call.
783 *
784 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
785 * overflow case.
786 */
787 ret = 0;
788 for (seg = 0; seg < nr_segs; seg++) {
789 void __user *buf = iov[seg].iov_base;
790 ssize_t len = (ssize_t)iov[seg].iov_len;
791
792 /* see if we we're about to use an invalid len or if
793 * it's about to overflow ssize_t */
794 if (len < 0) {
795 ret = -EINVAL;
796 goto out;
797 }
798 if (type >= 0
799 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
800 ret = -EFAULT;
801 goto out;
802 }
803 if (len > MAX_RW_COUNT - ret) {
804 len = MAX_RW_COUNT - ret;
805 iov[seg].iov_len = len;
806 }
807 ret += len;
808 }
809 out:
810 *ret_pointer = iov;
811 return ret;
812 }
813
814 static ssize_t do_readv_writev(int type, struct file *file,
815 const struct iovec __user * uvector,
816 unsigned long nr_segs, loff_t *pos)
817 {
818 size_t tot_len;
819 struct iovec iovstack[UIO_FASTIOV];
820 struct iovec *iov = iovstack;
821 struct iov_iter iter;
822 ssize_t ret;
823 io_fn_t fn;
824 iter_fn_t iter_fn;
825
826 ret = import_iovec(type, uvector, nr_segs,
827 ARRAY_SIZE(iovstack), &iov, &iter);
828 if (ret < 0)
829 return ret;
830
831 tot_len = iov_iter_count(&iter);
832 if (!tot_len)
833 goto out;
834 ret = rw_verify_area(type, file, pos, tot_len);
835 if (ret < 0)
836 goto out;
837
838 if (type == READ) {
839 fn = file->f_op->read;
840 iter_fn = file->f_op->read_iter;
841 } else {
842 fn = (io_fn_t)file->f_op->write;
843 iter_fn = file->f_op->write_iter;
844 file_start_write(file);
845 }
846
847 if (iter_fn)
848 ret = do_iter_readv_writev(file, &iter, pos, iter_fn);
849 else
850 ret = do_loop_readv_writev(file, &iter, pos, fn);
851
852 if (type != READ)
853 file_end_write(file);
854
855 out:
856 kfree(iov);
857 if ((ret + (type == READ)) > 0) {
858 if (type == READ)
859 fsnotify_access(file);
860 else
861 fsnotify_modify(file);
862 }
863 return ret;
864 }
865
866 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
867 unsigned long vlen, loff_t *pos)
868 {
869 if (!(file->f_mode & FMODE_READ))
870 return -EBADF;
871 if (!(file->f_mode & FMODE_CAN_READ))
872 return -EINVAL;
873
874 return do_readv_writev(READ, file, vec, vlen, pos);
875 }
876
877 EXPORT_SYMBOL(vfs_readv);
878
879 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
880 unsigned long vlen, loff_t *pos)
881 {
882 if (!(file->f_mode & FMODE_WRITE))
883 return -EBADF;
884 if (!(file->f_mode & FMODE_CAN_WRITE))
885 return -EINVAL;
886
887 return do_readv_writev(WRITE, file, vec, vlen, pos);
888 }
889
890 EXPORT_SYMBOL(vfs_writev);
891
892 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
893 unsigned long, vlen)
894 {
895 struct fd f = fdget_pos(fd);
896 ssize_t ret = -EBADF;
897
898 if (f.file) {
899 loff_t pos = file_pos_read(f.file);
900 ret = vfs_readv(f.file, vec, vlen, &pos);
901 if (ret >= 0)
902 file_pos_write(f.file, pos);
903 fdput_pos(f);
904 }
905
906 if (ret > 0)
907 add_rchar(current, ret);
908 inc_syscr(current);
909 return ret;
910 }
911
912 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
913 unsigned long, vlen)
914 {
915 struct fd f = fdget_pos(fd);
916 ssize_t ret = -EBADF;
917
918 if (f.file) {
919 loff_t pos = file_pos_read(f.file);
920 ret = vfs_writev(f.file, vec, vlen, &pos);
921 if (ret >= 0)
922 file_pos_write(f.file, pos);
923 fdput_pos(f);
924 }
925
926 if (ret > 0)
927 add_wchar(current, ret);
928 inc_syscw(current);
929 return ret;
930 }
931
932 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
933 {
934 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
935 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
936 }
937
938 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
939 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
940 {
941 loff_t pos = pos_from_hilo(pos_h, pos_l);
942 struct fd f;
943 ssize_t ret = -EBADF;
944
945 if (pos < 0)
946 return -EINVAL;
947
948 f = fdget(fd);
949 if (f.file) {
950 ret = -ESPIPE;
951 if (f.file->f_mode & FMODE_PREAD)
952 ret = vfs_readv(f.file, vec, vlen, &pos);
953 fdput(f);
954 }
955
956 if (ret > 0)
957 add_rchar(current, ret);
958 inc_syscr(current);
959 return ret;
960 }
961
962 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
963 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
964 {
965 loff_t pos = pos_from_hilo(pos_h, pos_l);
966 struct fd f;
967 ssize_t ret = -EBADF;
968
969 if (pos < 0)
970 return -EINVAL;
971
972 f = fdget(fd);
973 if (f.file) {
974 ret = -ESPIPE;
975 if (f.file->f_mode & FMODE_PWRITE)
976 ret = vfs_writev(f.file, vec, vlen, &pos);
977 fdput(f);
978 }
979
980 if (ret > 0)
981 add_wchar(current, ret);
982 inc_syscw(current);
983 return ret;
984 }
985
986 #ifdef CONFIG_COMPAT
987
988 static ssize_t compat_do_readv_writev(int type, struct file *file,
989 const struct compat_iovec __user *uvector,
990 unsigned long nr_segs, loff_t *pos)
991 {
992 compat_ssize_t tot_len;
993 struct iovec iovstack[UIO_FASTIOV];
994 struct iovec *iov = iovstack;
995 struct iov_iter iter;
996 ssize_t ret;
997 io_fn_t fn;
998 iter_fn_t iter_fn;
999
1000 ret = compat_import_iovec(type, uvector, nr_segs,
1001 UIO_FASTIOV, &iov, &iter);
1002 if (ret < 0)
1003 return ret;
1004
1005 tot_len = iov_iter_count(&iter);
1006 if (!tot_len)
1007 goto out;
1008 ret = rw_verify_area(type, file, pos, tot_len);
1009 if (ret < 0)
1010 goto out;
1011
1012 if (type == READ) {
1013 fn = file->f_op->read;
1014 iter_fn = file->f_op->read_iter;
1015 } else {
1016 fn = (io_fn_t)file->f_op->write;
1017 iter_fn = file->f_op->write_iter;
1018 file_start_write(file);
1019 }
1020
1021 if (iter_fn)
1022 ret = do_iter_readv_writev(file, &iter, pos, iter_fn);
1023 else
1024 ret = do_loop_readv_writev(file, &iter, pos, fn);
1025
1026 if (type != READ)
1027 file_end_write(file);
1028
1029 out:
1030 kfree(iov);
1031 if ((ret + (type == READ)) > 0) {
1032 if (type == READ)
1033 fsnotify_access(file);
1034 else
1035 fsnotify_modify(file);
1036 }
1037 return ret;
1038 }
1039
1040 static size_t compat_readv(struct file *file,
1041 const struct compat_iovec __user *vec,
1042 unsigned long vlen, loff_t *pos)
1043 {
1044 ssize_t ret = -EBADF;
1045
1046 if (!(file->f_mode & FMODE_READ))
1047 goto out;
1048
1049 ret = -EINVAL;
1050 if (!(file->f_mode & FMODE_CAN_READ))
1051 goto out;
1052
1053 ret = compat_do_readv_writev(READ, file, vec, vlen, pos);
1054
1055 out:
1056 if (ret > 0)
1057 add_rchar(current, ret);
1058 inc_syscr(current);
1059 return ret;
1060 }
1061
1062 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1063 const struct compat_iovec __user *,vec,
1064 compat_ulong_t, vlen)
1065 {
1066 struct fd f = fdget_pos(fd);
1067 ssize_t ret;
1068 loff_t pos;
1069
1070 if (!f.file)
1071 return -EBADF;
1072 pos = f.file->f_pos;
1073 ret = compat_readv(f.file, vec, vlen, &pos);
1074 if (ret >= 0)
1075 f.file->f_pos = pos;
1076 fdput_pos(f);
1077 return ret;
1078 }
1079
1080 static long __compat_sys_preadv64(unsigned long fd,
1081 const struct compat_iovec __user *vec,
1082 unsigned long vlen, loff_t pos)
1083 {
1084 struct fd f;
1085 ssize_t ret;
1086
1087 if (pos < 0)
1088 return -EINVAL;
1089 f = fdget(fd);
1090 if (!f.file)
1091 return -EBADF;
1092 ret = -ESPIPE;
1093 if (f.file->f_mode & FMODE_PREAD)
1094 ret = compat_readv(f.file, vec, vlen, &pos);
1095 fdput(f);
1096 return ret;
1097 }
1098
1099 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1100 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1101 const struct compat_iovec __user *,vec,
1102 unsigned long, vlen, loff_t, pos)
1103 {
1104 return __compat_sys_preadv64(fd, vec, vlen, pos);
1105 }
1106 #endif
1107
1108 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1109 const struct compat_iovec __user *,vec,
1110 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1111 {
1112 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1113
1114 return __compat_sys_preadv64(fd, vec, vlen, pos);
1115 }
1116
1117 static size_t compat_writev(struct file *file,
1118 const struct compat_iovec __user *vec,
1119 unsigned long vlen, loff_t *pos)
1120 {
1121 ssize_t ret = -EBADF;
1122
1123 if (!(file->f_mode & FMODE_WRITE))
1124 goto out;
1125
1126 ret = -EINVAL;
1127 if (!(file->f_mode & FMODE_CAN_WRITE))
1128 goto out;
1129
1130 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos);
1131
1132 out:
1133 if (ret > 0)
1134 add_wchar(current, ret);
1135 inc_syscw(current);
1136 return ret;
1137 }
1138
1139 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1140 const struct compat_iovec __user *, vec,
1141 compat_ulong_t, vlen)
1142 {
1143 struct fd f = fdget_pos(fd);
1144 ssize_t ret;
1145 loff_t pos;
1146
1147 if (!f.file)
1148 return -EBADF;
1149 pos = f.file->f_pos;
1150 ret = compat_writev(f.file, vec, vlen, &pos);
1151 if (ret >= 0)
1152 f.file->f_pos = pos;
1153 fdput_pos(f);
1154 return ret;
1155 }
1156
1157 static long __compat_sys_pwritev64(unsigned long fd,
1158 const struct compat_iovec __user *vec,
1159 unsigned long vlen, loff_t pos)
1160 {
1161 struct fd f;
1162 ssize_t ret;
1163
1164 if (pos < 0)
1165 return -EINVAL;
1166 f = fdget(fd);
1167 if (!f.file)
1168 return -EBADF;
1169 ret = -ESPIPE;
1170 if (f.file->f_mode & FMODE_PWRITE)
1171 ret = compat_writev(f.file, vec, vlen, &pos);
1172 fdput(f);
1173 return ret;
1174 }
1175
1176 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1177 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1178 const struct compat_iovec __user *,vec,
1179 unsigned long, vlen, loff_t, pos)
1180 {
1181 return __compat_sys_pwritev64(fd, vec, vlen, pos);
1182 }
1183 #endif
1184
1185 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1186 const struct compat_iovec __user *,vec,
1187 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1188 {
1189 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1190
1191 return __compat_sys_pwritev64(fd, vec, vlen, pos);
1192 }
1193 #endif
1194
1195 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1196 size_t count, loff_t max)
1197 {
1198 struct fd in, out;
1199 struct inode *in_inode, *out_inode;
1200 loff_t pos;
1201 loff_t out_pos;
1202 ssize_t retval;
1203 int fl;
1204
1205 /*
1206 * Get input file, and verify that it is ok..
1207 */
1208 retval = -EBADF;
1209 in = fdget(in_fd);
1210 if (!in.file)
1211 goto out;
1212 if (!(in.file->f_mode & FMODE_READ))
1213 goto fput_in;
1214 retval = -ESPIPE;
1215 if (!ppos) {
1216 pos = in.file->f_pos;
1217 } else {
1218 pos = *ppos;
1219 if (!(in.file->f_mode & FMODE_PREAD))
1220 goto fput_in;
1221 }
1222 retval = rw_verify_area(READ, in.file, &pos, count);
1223 if (retval < 0)
1224 goto fput_in;
1225 count = retval;
1226
1227 /*
1228 * Get output file, and verify that it is ok..
1229 */
1230 retval = -EBADF;
1231 out = fdget(out_fd);
1232 if (!out.file)
1233 goto fput_in;
1234 if (!(out.file->f_mode & FMODE_WRITE))
1235 goto fput_out;
1236 retval = -EINVAL;
1237 in_inode = file_inode(in.file);
1238 out_inode = file_inode(out.file);
1239 out_pos = out.file->f_pos;
1240 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1241 if (retval < 0)
1242 goto fput_out;
1243 count = retval;
1244
1245 if (!max)
1246 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1247
1248 if (unlikely(pos + count > max)) {
1249 retval = -EOVERFLOW;
1250 if (pos >= max)
1251 goto fput_out;
1252 count = max - pos;
1253 }
1254
1255 fl = 0;
1256 #if 0
1257 /*
1258 * We need to debate whether we can enable this or not. The
1259 * man page documents EAGAIN return for the output at least,
1260 * and the application is arguably buggy if it doesn't expect
1261 * EAGAIN on a non-blocking file descriptor.
1262 */
1263 if (in.file->f_flags & O_NONBLOCK)
1264 fl = SPLICE_F_NONBLOCK;
1265 #endif
1266 file_start_write(out.file);
1267 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1268 file_end_write(out.file);
1269
1270 if (retval > 0) {
1271 add_rchar(current, retval);
1272 add_wchar(current, retval);
1273 fsnotify_access(in.file);
1274 fsnotify_modify(out.file);
1275 out.file->f_pos = out_pos;
1276 if (ppos)
1277 *ppos = pos;
1278 else
1279 in.file->f_pos = pos;
1280 }
1281
1282 inc_syscr(current);
1283 inc_syscw(current);
1284 if (pos > max)
1285 retval = -EOVERFLOW;
1286
1287 fput_out:
1288 fdput(out);
1289 fput_in:
1290 fdput(in);
1291 out:
1292 return retval;
1293 }
1294
1295 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1296 {
1297 loff_t pos;
1298 off_t off;
1299 ssize_t ret;
1300
1301 if (offset) {
1302 if (unlikely(get_user(off, offset)))
1303 return -EFAULT;
1304 pos = off;
1305 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1306 if (unlikely(put_user(pos, offset)))
1307 return -EFAULT;
1308 return ret;
1309 }
1310
1311 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1312 }
1313
1314 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1315 {
1316 loff_t pos;
1317 ssize_t ret;
1318
1319 if (offset) {
1320 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1321 return -EFAULT;
1322 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1323 if (unlikely(put_user(pos, offset)))
1324 return -EFAULT;
1325 return ret;
1326 }
1327
1328 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1329 }
1330
1331 #ifdef CONFIG_COMPAT
1332 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1333 compat_off_t __user *, offset, compat_size_t, count)
1334 {
1335 loff_t pos;
1336 off_t off;
1337 ssize_t ret;
1338
1339 if (offset) {
1340 if (unlikely(get_user(off, offset)))
1341 return -EFAULT;
1342 pos = off;
1343 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1344 if (unlikely(put_user(pos, offset)))
1345 return -EFAULT;
1346 return ret;
1347 }
1348
1349 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1350 }
1351
1352 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1353 compat_loff_t __user *, offset, compat_size_t, count)
1354 {
1355 loff_t pos;
1356 ssize_t ret;
1357
1358 if (offset) {
1359 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1360 return -EFAULT;
1361 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1362 if (unlikely(put_user(pos, offset)))
1363 return -EFAULT;
1364 return ret;
1365 }
1366
1367 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1368 }
1369 #endif
1370
1371 /*
1372 * copy_file_range() differs from regular file read and write in that it
1373 * specifically allows return partial success. When it does so is up to
1374 * the copy_file_range method.
1375 */
1376 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1377 struct file *file_out, loff_t pos_out,
1378 size_t len, unsigned int flags)
1379 {
1380 struct inode *inode_in = file_inode(file_in);
1381 struct inode *inode_out = file_inode(file_out);
1382 ssize_t ret;
1383
1384 if (flags != 0)
1385 return -EINVAL;
1386
1387 /* copy_file_range allows full ssize_t len, ignoring MAX_RW_COUNT */
1388 ret = rw_verify_area(READ, file_in, &pos_in, len);
1389 if (ret >= 0)
1390 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1391 if (ret < 0)
1392 return ret;
1393
1394 if (!(file_in->f_mode & FMODE_READ) ||
1395 !(file_out->f_mode & FMODE_WRITE) ||
1396 (file_out->f_flags & O_APPEND))
1397 return -EBADF;
1398
1399 /* this could be relaxed once a method supports cross-fs copies */
1400 if (inode_in->i_sb != inode_out->i_sb)
1401 return -EXDEV;
1402
1403 if (len == 0)
1404 return 0;
1405
1406 ret = mnt_want_write_file(file_out);
1407 if (ret)
1408 return ret;
1409
1410 ret = -EOPNOTSUPP;
1411 if (file_out->f_op->copy_file_range)
1412 ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out,
1413 pos_out, len, flags);
1414 if (ret == -EOPNOTSUPP)
1415 ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1416 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1417
1418 if (ret > 0) {
1419 fsnotify_access(file_in);
1420 add_rchar(current, ret);
1421 fsnotify_modify(file_out);
1422 add_wchar(current, ret);
1423 }
1424 inc_syscr(current);
1425 inc_syscw(current);
1426
1427 mnt_drop_write_file(file_out);
1428
1429 return ret;
1430 }
1431 EXPORT_SYMBOL(vfs_copy_file_range);
1432
1433 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1434 int, fd_out, loff_t __user *, off_out,
1435 size_t, len, unsigned int, flags)
1436 {
1437 loff_t pos_in;
1438 loff_t pos_out;
1439 struct fd f_in;
1440 struct fd f_out;
1441 ssize_t ret = -EBADF;
1442
1443 f_in = fdget(fd_in);
1444 if (!f_in.file)
1445 goto out2;
1446
1447 f_out = fdget(fd_out);
1448 if (!f_out.file)
1449 goto out1;
1450
1451 ret = -EFAULT;
1452 if (off_in) {
1453 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1454 goto out;
1455 } else {
1456 pos_in = f_in.file->f_pos;
1457 }
1458
1459 if (off_out) {
1460 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1461 goto out;
1462 } else {
1463 pos_out = f_out.file->f_pos;
1464 }
1465
1466 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1467 flags);
1468 if (ret > 0) {
1469 pos_in += ret;
1470 pos_out += ret;
1471
1472 if (off_in) {
1473 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1474 ret = -EFAULT;
1475 } else {
1476 f_in.file->f_pos = pos_in;
1477 }
1478
1479 if (off_out) {
1480 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1481 ret = -EFAULT;
1482 } else {
1483 f_out.file->f_pos = pos_out;
1484 }
1485 }
1486
1487 out:
1488 fdput(f_out);
1489 out1:
1490 fdput(f_in);
1491 out2:
1492 return ret;
1493 }
1494
1495 static int clone_verify_area(struct file *file, loff_t pos, u64 len, bool write)
1496 {
1497 struct inode *inode = file_inode(file);
1498
1499 if (unlikely(pos < 0))
1500 return -EINVAL;
1501
1502 if (unlikely((loff_t) (pos + len) < 0))
1503 return -EINVAL;
1504
1505 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1506 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1507 int retval;
1508
1509 retval = locks_mandatory_area(inode, file, pos, end,
1510 write ? F_WRLCK : F_RDLCK);
1511 if (retval < 0)
1512 return retval;
1513 }
1514
1515 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1516 }
1517
1518 int vfs_clone_file_range(struct file *file_in, loff_t pos_in,
1519 struct file *file_out, loff_t pos_out, u64 len)
1520 {
1521 struct inode *inode_in = file_inode(file_in);
1522 struct inode *inode_out = file_inode(file_out);
1523 int ret;
1524
1525 if (inode_in->i_sb != inode_out->i_sb ||
1526 file_in->f_path.mnt != file_out->f_path.mnt)
1527 return -EXDEV;
1528
1529 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1530 return -EISDIR;
1531 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1532 return -EINVAL;
1533
1534 if (!(file_in->f_mode & FMODE_READ) ||
1535 !(file_out->f_mode & FMODE_WRITE) ||
1536 (file_out->f_flags & O_APPEND))
1537 return -EBADF;
1538
1539 if (!file_in->f_op->clone_file_range)
1540 return -EOPNOTSUPP;
1541
1542 ret = clone_verify_area(file_in, pos_in, len, false);
1543 if (ret)
1544 return ret;
1545
1546 ret = clone_verify_area(file_out, pos_out, len, true);
1547 if (ret)
1548 return ret;
1549
1550 if (pos_in + len > i_size_read(inode_in))
1551 return -EINVAL;
1552
1553 ret = mnt_want_write_file(file_out);
1554 if (ret)
1555 return ret;
1556
1557 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1558 file_out, pos_out, len);
1559 if (!ret) {
1560 fsnotify_access(file_in);
1561 fsnotify_modify(file_out);
1562 }
1563
1564 mnt_drop_write_file(file_out);
1565 return ret;
1566 }
1567 EXPORT_SYMBOL(vfs_clone_file_range);
1568
1569 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
1570 {
1571 struct file_dedupe_range_info *info;
1572 struct inode *src = file_inode(file);
1573 u64 off;
1574 u64 len;
1575 int i;
1576 int ret;
1577 bool is_admin = capable(CAP_SYS_ADMIN);
1578 u16 count = same->dest_count;
1579 struct file *dst_file;
1580 loff_t dst_off;
1581 ssize_t deduped;
1582
1583 if (!(file->f_mode & FMODE_READ))
1584 return -EINVAL;
1585
1586 if (same->reserved1 || same->reserved2)
1587 return -EINVAL;
1588
1589 off = same->src_offset;
1590 len = same->src_length;
1591
1592 ret = -EISDIR;
1593 if (S_ISDIR(src->i_mode))
1594 goto out;
1595
1596 ret = -EINVAL;
1597 if (!S_ISREG(src->i_mode))
1598 goto out;
1599
1600 ret = clone_verify_area(file, off, len, false);
1601 if (ret < 0)
1602 goto out;
1603 ret = 0;
1604
1605 /* pre-format output fields to sane values */
1606 for (i = 0; i < count; i++) {
1607 same->info[i].bytes_deduped = 0ULL;
1608 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
1609 }
1610
1611 for (i = 0, info = same->info; i < count; i++, info++) {
1612 struct inode *dst;
1613 struct fd dst_fd = fdget(info->dest_fd);
1614
1615 dst_file = dst_fd.file;
1616 if (!dst_file) {
1617 info->status = -EBADF;
1618 goto next_loop;
1619 }
1620 dst = file_inode(dst_file);
1621
1622 ret = mnt_want_write_file(dst_file);
1623 if (ret) {
1624 info->status = ret;
1625 goto next_loop;
1626 }
1627
1628 dst_off = info->dest_offset;
1629 ret = clone_verify_area(dst_file, dst_off, len, true);
1630 if (ret < 0) {
1631 info->status = ret;
1632 goto next_file;
1633 }
1634 ret = 0;
1635
1636 if (info->reserved) {
1637 info->status = -EINVAL;
1638 } else if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) {
1639 info->status = -EINVAL;
1640 } else if (file->f_path.mnt != dst_file->f_path.mnt) {
1641 info->status = -EXDEV;
1642 } else if (S_ISDIR(dst->i_mode)) {
1643 info->status = -EISDIR;
1644 } else if (dst_file->f_op->dedupe_file_range == NULL) {
1645 info->status = -EINVAL;
1646 } else {
1647 deduped = dst_file->f_op->dedupe_file_range(file, off,
1648 len, dst_file,
1649 info->dest_offset);
1650 if (deduped == -EBADE)
1651 info->status = FILE_DEDUPE_RANGE_DIFFERS;
1652 else if (deduped < 0)
1653 info->status = deduped;
1654 else
1655 info->bytes_deduped += deduped;
1656 }
1657
1658 next_file:
1659 mnt_drop_write_file(dst_file);
1660 next_loop:
1661 fdput(dst_fd);
1662
1663 if (fatal_signal_pending(current))
1664 goto out;
1665 }
1666
1667 out:
1668 return ret;
1669 }
1670 EXPORT_SYMBOL(vfs_dedupe_file_range);
Linux with added FPC-III support