]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - fs/read_write.c
projects / mirror_ubuntu-zesty-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jikos/hid
[mirror_ubuntu-zesty-kernel.git] / fs / read_write.c
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 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
306 {
307 off_t retval;
308 struct fd f = fdget_pos(fd);
309 if (!f.file)
310 return -EBADF;
311
312 retval = -EINVAL;
313 if (whence <= SEEK_MAX) {
314 loff_t res = vfs_llseek(f.file, offset, whence);
315 retval = res;
316 if (res != (loff_t)retval)
317 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
318 }
319 fdput_pos(f);
320 return retval;
321 }
322
323 #ifdef CONFIG_COMPAT
324 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
325 {
326 return sys_lseek(fd, offset, whence);
327 }
328 #endif
329
330 #ifdef __ARCH_WANT_SYS_LLSEEK
331 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
332 unsigned long, offset_low, loff_t __user *, result,
333 unsigned int, whence)
334 {
335 int retval;
336 struct fd f = fdget_pos(fd);
337 loff_t offset;
338
339 if (!f.file)
340 return -EBADF;
341
342 retval = -EINVAL;
343 if (whence > SEEK_MAX)
344 goto out_putf;
345
346 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
347 whence);
348
349 retval = (int)offset;
350 if (offset >= 0) {
351 retval = -EFAULT;
352 if (!copy_to_user(result, &offset, sizeof(offset)))
353 retval = 0;
354 }
355 out_putf:
356 fdput_pos(f);
357 return retval;
358 }
359 #endif
360
361 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos)
362 {
363 struct kiocb kiocb;
364 ssize_t ret;
365
366 if (!file->f_op->read_iter)
367 return -EINVAL;
368
369 init_sync_kiocb(&kiocb, file);
370 kiocb.ki_pos = *ppos;
371
372 iter->type |= READ;
373 ret = file->f_op->read_iter(&kiocb, iter);
374 BUG_ON(ret == -EIOCBQUEUED);
375 if (ret > 0)
376 *ppos = kiocb.ki_pos;
377 return ret;
378 }
379 EXPORT_SYMBOL(vfs_iter_read);
380
381 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos)
382 {
383 struct kiocb kiocb;
384 ssize_t ret;
385
386 if (!file->f_op->write_iter)
387 return -EINVAL;
388
389 init_sync_kiocb(&kiocb, file);
390 kiocb.ki_pos = *ppos;
391
392 iter->type |= WRITE;
393 ret = file->f_op->write_iter(&kiocb, iter);
394 BUG_ON(ret == -EIOCBQUEUED);
395 if (ret > 0)
396 *ppos = kiocb.ki_pos;
397 return ret;
398 }
399 EXPORT_SYMBOL(vfs_iter_write);
400
401 /*
402 * rw_verify_area doesn't like huge counts. We limit
403 * them to something that fits in "int" so that others
404 * won't have to do range checks all the time.
405 */
406 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
407 {
408 struct inode *inode;
409 loff_t pos;
410 int retval = -EINVAL;
411
412 inode = file_inode(file);
413 if (unlikely((ssize_t) count < 0))
414 return retval;
415 pos = *ppos;
416 if (unlikely(pos < 0)) {
417 if (!unsigned_offsets(file))
418 return retval;
419 if (count >= -pos) /* both values are in 0..LLONG_MAX */
420 return -EOVERFLOW;
421 } else if (unlikely((loff_t) (pos + count) < 0)) {
422 if (!unsigned_offsets(file))
423 return retval;
424 }
425
426 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
427 retval = locks_mandatory_area(inode, file, pos, pos + count - 1,
428 read_write == READ ? F_RDLCK : F_WRLCK);
429 if (retval < 0)
430 return retval;
431 }
432 retval = security_file_permission(file,
433 read_write == READ ? MAY_READ : MAY_WRITE);
434 if (retval)
435 return retval;
436 return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
437 }
438
439 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
440 {
441 struct iovec iov = { .iov_base = buf, .iov_len = len };
442 struct kiocb kiocb;
443 struct iov_iter iter;
444 ssize_t ret;
445
446 init_sync_kiocb(&kiocb, filp);
447 kiocb.ki_pos = *ppos;
448 iov_iter_init(&iter, READ, &iov, 1, len);
449
450 ret = filp->f_op->read_iter(&kiocb, &iter);
451 BUG_ON(ret == -EIOCBQUEUED);
452 *ppos = kiocb.ki_pos;
453 return ret;
454 }
455
456 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
457 loff_t *pos)
458 {
459 if (file->f_op->read)
460 return file->f_op->read(file, buf, count, pos);
461 else if (file->f_op->read_iter)
462 return new_sync_read(file, buf, count, pos);
463 else
464 return -EINVAL;
465 }
466 EXPORT_SYMBOL(__vfs_read);
467
468 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
469 {
470 ssize_t ret;
471
472 if (!(file->f_mode & FMODE_READ))
473 return -EBADF;
474 if (!(file->f_mode & FMODE_CAN_READ))
475 return -EINVAL;
476 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
477 return -EFAULT;
478
479 ret = rw_verify_area(READ, file, pos, count);
480 if (ret >= 0) {
481 count = ret;
482 ret = __vfs_read(file, buf, count, pos);
483 if (ret > 0) {
484 fsnotify_access(file);
485 add_rchar(current, ret);
486 }
487 inc_syscr(current);
488 }
489
490 return ret;
491 }
492
493 EXPORT_SYMBOL(vfs_read);
494
495 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
496 {
497 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
498 struct kiocb kiocb;
499 struct iov_iter iter;
500 ssize_t ret;
501
502 init_sync_kiocb(&kiocb, filp);
503 kiocb.ki_pos = *ppos;
504 iov_iter_init(&iter, WRITE, &iov, 1, len);
505
506 ret = filp->f_op->write_iter(&kiocb, &iter);
507 BUG_ON(ret == -EIOCBQUEUED);
508 if (ret > 0)
509 *ppos = kiocb.ki_pos;
510 return ret;
511 }
512
513 ssize_t __vfs_write(struct file *file, const char __user *p, size_t count,
514 loff_t *pos)
515 {
516 if (file->f_op->write)
517 return file->f_op->write(file, p, count, pos);
518 else if (file->f_op->write_iter)
519 return new_sync_write(file, p, count, pos);
520 else
521 return -EINVAL;
522 }
523 EXPORT_SYMBOL(__vfs_write);
524
525 ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
526 {
527 mm_segment_t old_fs;
528 const char __user *p;
529 ssize_t ret;
530
531 if (!(file->f_mode & FMODE_CAN_WRITE))
532 return -EINVAL;
533
534 old_fs = get_fs();
535 set_fs(get_ds());
536 p = (__force const char __user *)buf;
537 if (count > MAX_RW_COUNT)
538 count = MAX_RW_COUNT;
539 ret = __vfs_write(file, p, count, pos);
540 set_fs(old_fs);
541 if (ret > 0) {
542 fsnotify_modify(file);
543 add_wchar(current, ret);
544 }
545 inc_syscw(current);
546 return ret;
547 }
548
549 EXPORT_SYMBOL(__kernel_write);
550
551 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
552 {
553 ssize_t ret;
554
555 if (!(file->f_mode & FMODE_WRITE))
556 return -EBADF;
557 if (!(file->f_mode & FMODE_CAN_WRITE))
558 return -EINVAL;
559 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
560 return -EFAULT;
561
562 ret = rw_verify_area(WRITE, file, pos, count);
563 if (ret >= 0) {
564 count = ret;
565 file_start_write(file);
566 ret = __vfs_write(file, buf, count, pos);
567 if (ret > 0) {
568 fsnotify_modify(file);
569 add_wchar(current, ret);
570 }
571 inc_syscw(current);
572 file_end_write(file);
573 }
574
575 return ret;
576 }
577
578 EXPORT_SYMBOL(vfs_write);
579
580 static inline loff_t file_pos_read(struct file *file)
581 {
582 return file->f_pos;
583 }
584
585 static inline void file_pos_write(struct file *file, loff_t pos)
586 {
587 file->f_pos = pos;
588 }
589
590 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
591 {
592 struct fd f = fdget_pos(fd);
593 ssize_t ret = -EBADF;
594
595 if (f.file) {
596 loff_t pos = file_pos_read(f.file);
597 ret = vfs_read(f.file, buf, count, &pos);
598 if (ret >= 0)
599 file_pos_write(f.file, pos);
600 fdput_pos(f);
601 }
602 return ret;
603 }
604
605 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
606 size_t, count)
607 {
608 struct fd f = fdget_pos(fd);
609 ssize_t ret = -EBADF;
610
611 if (f.file) {
612 loff_t pos = file_pos_read(f.file);
613 ret = vfs_write(f.file, buf, count, &pos);
614 if (ret >= 0)
615 file_pos_write(f.file, pos);
616 fdput_pos(f);
617 }
618
619 return ret;
620 }
621
622 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
623 size_t, count, loff_t, pos)
624 {
625 struct fd f;
626 ssize_t ret = -EBADF;
627
628 if (pos < 0)
629 return -EINVAL;
630
631 f = fdget(fd);
632 if (f.file) {
633 ret = -ESPIPE;
634 if (f.file->f_mode & FMODE_PREAD)
635 ret = vfs_read(f.file, buf, count, &pos);
636 fdput(f);
637 }
638
639 return ret;
640 }
641
642 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
643 size_t, count, loff_t, pos)
644 {
645 struct fd f;
646 ssize_t ret = -EBADF;
647
648 if (pos < 0)
649 return -EINVAL;
650
651 f = fdget(fd);
652 if (f.file) {
653 ret = -ESPIPE;
654 if (f.file->f_mode & FMODE_PWRITE)
655 ret = vfs_write(f.file, buf, count, &pos);
656 fdput(f);
657 }
658
659 return ret;
660 }
661
662 /*
663 * Reduce an iovec's length in-place. Return the resulting number of segments
664 */
665 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
666 {
667 unsigned long seg = 0;
668 size_t len = 0;
669
670 while (seg < nr_segs) {
671 seg++;
672 if (len + iov->iov_len >= to) {
673 iov->iov_len = to - len;
674 break;
675 }
676 len += iov->iov_len;
677 iov++;
678 }
679 return seg;
680 }
681 EXPORT_SYMBOL(iov_shorten);
682
683 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
684 loff_t *ppos, iter_fn_t fn, int flags)
685 {
686 struct kiocb kiocb;
687 ssize_t ret;
688
689 if (flags & ~(RWF_HIPRI | RWF_DSYNC | RWF_SYNC))
690 return -EOPNOTSUPP;
691
692 init_sync_kiocb(&kiocb, filp);
693 if (flags & RWF_HIPRI)
694 kiocb.ki_flags |= IOCB_HIPRI;
695 if (flags & RWF_DSYNC)
696 kiocb.ki_flags |= IOCB_DSYNC;
697 if (flags & RWF_SYNC)
698 kiocb.ki_flags |= (IOCB_DSYNC | IOCB_SYNC);
699 kiocb.ki_pos = *ppos;
700
701 ret = fn(&kiocb, iter);
702 BUG_ON(ret == -EIOCBQUEUED);
703 *ppos = kiocb.ki_pos;
704 return ret;
705 }
706
707 /* Do it by hand, with file-ops */
708 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
709 loff_t *ppos, io_fn_t fn, int flags)
710 {
711 ssize_t ret = 0;
712
713 if (flags & ~RWF_HIPRI)
714 return -EOPNOTSUPP;
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 int flags)
818 {
819 size_t tot_len;
820 struct iovec iovstack[UIO_FASTIOV];
821 struct iovec *iov = iovstack;
822 struct iov_iter iter;
823 ssize_t ret;
824 io_fn_t fn;
825 iter_fn_t iter_fn;
826
827 ret = import_iovec(type, uvector, nr_segs,
828 ARRAY_SIZE(iovstack), &iov, &iter);
829 if (ret < 0)
830 return ret;
831
832 tot_len = iov_iter_count(&iter);
833 if (!tot_len)
834 goto out;
835 ret = rw_verify_area(type, file, pos, tot_len);
836 if (ret < 0)
837 goto out;
838
839 if (type == READ) {
840 fn = file->f_op->read;
841 iter_fn = file->f_op->read_iter;
842 } else {
843 fn = (io_fn_t)file->f_op->write;
844 iter_fn = file->f_op->write_iter;
845 file_start_write(file);
846 }
847
848 if (iter_fn)
849 ret = do_iter_readv_writev(file, &iter, pos, iter_fn, flags);
850 else
851 ret = do_loop_readv_writev(file, &iter, pos, fn, flags);
852
853 if (type != READ)
854 file_end_write(file);
855
856 out:
857 kfree(iov);
858 if ((ret + (type == READ)) > 0) {
859 if (type == READ)
860 fsnotify_access(file);
861 else
862 fsnotify_modify(file);
863 }
864 return ret;
865 }
866
867 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
868 unsigned long vlen, loff_t *pos, int flags)
869 {
870 if (!(file->f_mode & FMODE_READ))
871 return -EBADF;
872 if (!(file->f_mode & FMODE_CAN_READ))
873 return -EINVAL;
874
875 return do_readv_writev(READ, file, vec, vlen, pos, flags);
876 }
877
878 EXPORT_SYMBOL(vfs_readv);
879
880 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
881 unsigned long vlen, loff_t *pos, int flags)
882 {
883 if (!(file->f_mode & FMODE_WRITE))
884 return -EBADF;
885 if (!(file->f_mode & FMODE_CAN_WRITE))
886 return -EINVAL;
887
888 return do_readv_writev(WRITE, file, vec, vlen, pos, flags);
889 }
890
891 EXPORT_SYMBOL(vfs_writev);
892
893 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
894 unsigned long vlen, int flags)
895 {
896 struct fd f = fdget_pos(fd);
897 ssize_t ret = -EBADF;
898
899 if (f.file) {
900 loff_t pos = file_pos_read(f.file);
901 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
902 if (ret >= 0)
903 file_pos_write(f.file, pos);
904 fdput_pos(f);
905 }
906
907 if (ret > 0)
908 add_rchar(current, ret);
909 inc_syscr(current);
910 return ret;
911 }
912
913 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
914 unsigned long vlen, int flags)
915 {
916 struct fd f = fdget_pos(fd);
917 ssize_t ret = -EBADF;
918
919 if (f.file) {
920 loff_t pos = file_pos_read(f.file);
921 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
922 if (ret >= 0)
923 file_pos_write(f.file, pos);
924 fdput_pos(f);
925 }
926
927 if (ret > 0)
928 add_wchar(current, ret);
929 inc_syscw(current);
930 return ret;
931 }
932
933 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
934 {
935 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
936 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
937 }
938
939 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
940 unsigned long vlen, loff_t pos, int flags)
941 {
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, flags);
953 fdput(f);
954 }
955
956 if (ret > 0)
957 add_rchar(current, ret);
958 inc_syscr(current);
959 return ret;
960 }
961
962 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
963 unsigned long vlen, loff_t pos, int flags)
964 {
965 struct fd f;
966 ssize_t ret = -EBADF;
967
968 if (pos < 0)
969 return -EINVAL;
970
971 f = fdget(fd);
972 if (f.file) {
973 ret = -ESPIPE;
974 if (f.file->f_mode & FMODE_PWRITE)
975 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
976 fdput(f);
977 }
978
979 if (ret > 0)
980 add_wchar(current, ret);
981 inc_syscw(current);
982 return ret;
983 }
984
985 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
986 unsigned long, vlen)
987 {
988 return do_readv(fd, vec, vlen, 0);
989 }
990
991 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
992 unsigned long, vlen)
993 {
994 return do_writev(fd, vec, vlen, 0);
995 }
996
997 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
998 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
999 {
1000 loff_t pos = pos_from_hilo(pos_h, pos_l);
1001
1002 return do_preadv(fd, vec, vlen, pos, 0);
1003 }
1004
1005 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1006 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1007 int, flags)
1008 {
1009 loff_t pos = pos_from_hilo(pos_h, pos_l);
1010
1011 if (pos == -1)
1012 return do_readv(fd, vec, vlen, flags);
1013
1014 return do_preadv(fd, vec, vlen, pos, flags);
1015 }
1016
1017 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1018 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1019 {
1020 loff_t pos = pos_from_hilo(pos_h, pos_l);
1021
1022 return do_pwritev(fd, vec, vlen, pos, 0);
1023 }
1024
1025 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1026 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1027 int, flags)
1028 {
1029 loff_t pos = pos_from_hilo(pos_h, pos_l);
1030
1031 if (pos == -1)
1032 return do_writev(fd, vec, vlen, flags);
1033
1034 return do_pwritev(fd, vec, vlen, pos, flags);
1035 }
1036
1037 #ifdef CONFIG_COMPAT
1038
1039 static ssize_t compat_do_readv_writev(int type, struct file *file,
1040 const struct compat_iovec __user *uvector,
1041 unsigned long nr_segs, loff_t *pos,
1042 int flags)
1043 {
1044 compat_ssize_t tot_len;
1045 struct iovec iovstack[UIO_FASTIOV];
1046 struct iovec *iov = iovstack;
1047 struct iov_iter iter;
1048 ssize_t ret;
1049 io_fn_t fn;
1050 iter_fn_t iter_fn;
1051
1052 ret = compat_import_iovec(type, uvector, nr_segs,
1053 UIO_FASTIOV, &iov, &iter);
1054 if (ret < 0)
1055 return ret;
1056
1057 tot_len = iov_iter_count(&iter);
1058 if (!tot_len)
1059 goto out;
1060 ret = rw_verify_area(type, file, pos, tot_len);
1061 if (ret < 0)
1062 goto out;
1063
1064 if (type == READ) {
1065 fn = file->f_op->read;
1066 iter_fn = file->f_op->read_iter;
1067 } else {
1068 fn = (io_fn_t)file->f_op->write;
1069 iter_fn = file->f_op->write_iter;
1070 file_start_write(file);
1071 }
1072
1073 if (iter_fn)
1074 ret = do_iter_readv_writev(file, &iter, pos, iter_fn, flags);
1075 else
1076 ret = do_loop_readv_writev(file, &iter, pos, fn, flags);
1077
1078 if (type != READ)
1079 file_end_write(file);
1080
1081 out:
1082 kfree(iov);
1083 if ((ret + (type == READ)) > 0) {
1084 if (type == READ)
1085 fsnotify_access(file);
1086 else
1087 fsnotify_modify(file);
1088 }
1089 return ret;
1090 }
1091
1092 static size_t compat_readv(struct file *file,
1093 const struct compat_iovec __user *vec,
1094 unsigned long vlen, loff_t *pos, int flags)
1095 {
1096 ssize_t ret = -EBADF;
1097
1098 if (!(file->f_mode & FMODE_READ))
1099 goto out;
1100
1101 ret = -EINVAL;
1102 if (!(file->f_mode & FMODE_CAN_READ))
1103 goto out;
1104
1105 ret = compat_do_readv_writev(READ, file, vec, vlen, pos, flags);
1106
1107 out:
1108 if (ret > 0)
1109 add_rchar(current, ret);
1110 inc_syscr(current);
1111 return ret;
1112 }
1113
1114 static size_t do_compat_readv(compat_ulong_t fd,
1115 const struct compat_iovec __user *vec,
1116 compat_ulong_t vlen, int flags)
1117 {
1118 struct fd f = fdget_pos(fd);
1119 ssize_t ret;
1120 loff_t pos;
1121
1122 if (!f.file)
1123 return -EBADF;
1124 pos = f.file->f_pos;
1125 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1126 if (ret >= 0)
1127 f.file->f_pos = pos;
1128 fdput_pos(f);
1129 return ret;
1130
1131 }
1132
1133 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1134 const struct compat_iovec __user *,vec,
1135 compat_ulong_t, vlen)
1136 {
1137 return do_compat_readv(fd, vec, vlen, 0);
1138 }
1139
1140 static long do_compat_preadv64(unsigned long fd,
1141 const struct compat_iovec __user *vec,
1142 unsigned long vlen, loff_t pos, int flags)
1143 {
1144 struct fd f;
1145 ssize_t ret;
1146
1147 if (pos < 0)
1148 return -EINVAL;
1149 f = fdget(fd);
1150 if (!f.file)
1151 return -EBADF;
1152 ret = -ESPIPE;
1153 if (f.file->f_mode & FMODE_PREAD)
1154 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1155 fdput(f);
1156 return ret;
1157 }
1158
1159 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1160 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1161 const struct compat_iovec __user *,vec,
1162 unsigned long, vlen, loff_t, pos)
1163 {
1164 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1165 }
1166 #endif
1167
1168 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1169 const struct compat_iovec __user *,vec,
1170 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1171 {
1172 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1173
1174 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1175 }
1176
1177 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1178 const struct compat_iovec __user *,vec,
1179 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1180 int, flags)
1181 {
1182 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1183
1184 if (pos == -1)
1185 return do_compat_readv(fd, vec, vlen, flags);
1186
1187 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1188 }
1189
1190 static size_t compat_writev(struct file *file,
1191 const struct compat_iovec __user *vec,
1192 unsigned long vlen, loff_t *pos, int flags)
1193 {
1194 ssize_t ret = -EBADF;
1195
1196 if (!(file->f_mode & FMODE_WRITE))
1197 goto out;
1198
1199 ret = -EINVAL;
1200 if (!(file->f_mode & FMODE_CAN_WRITE))
1201 goto out;
1202
1203 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos, 0);
1204
1205 out:
1206 if (ret > 0)
1207 add_wchar(current, ret);
1208 inc_syscw(current);
1209 return ret;
1210 }
1211
1212 static size_t do_compat_writev(compat_ulong_t fd,
1213 const struct compat_iovec __user* vec,
1214 compat_ulong_t vlen, int flags)
1215 {
1216 struct fd f = fdget_pos(fd);
1217 ssize_t ret;
1218 loff_t pos;
1219
1220 if (!f.file)
1221 return -EBADF;
1222 pos = f.file->f_pos;
1223 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1224 if (ret >= 0)
1225 f.file->f_pos = pos;
1226 fdput_pos(f);
1227 return ret;
1228 }
1229
1230 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1231 const struct compat_iovec __user *, vec,
1232 compat_ulong_t, vlen)
1233 {
1234 return do_compat_writev(fd, vec, vlen, 0);
1235 }
1236
1237 static long do_compat_pwritev64(unsigned long fd,
1238 const struct compat_iovec __user *vec,
1239 unsigned long vlen, loff_t pos, int flags)
1240 {
1241 struct fd f;
1242 ssize_t ret;
1243
1244 if (pos < 0)
1245 return -EINVAL;
1246 f = fdget(fd);
1247 if (!f.file)
1248 return -EBADF;
1249 ret = -ESPIPE;
1250 if (f.file->f_mode & FMODE_PWRITE)
1251 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1252 fdput(f);
1253 return ret;
1254 }
1255
1256 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1257 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1258 const struct compat_iovec __user *,vec,
1259 unsigned long, vlen, loff_t, pos)
1260 {
1261 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1262 }
1263 #endif
1264
1265 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1266 const struct compat_iovec __user *,vec,
1267 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1268 {
1269 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1270
1271 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1272 }
1273
1274 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1275 const struct compat_iovec __user *,vec,
1276 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, int, flags)
1277 {
1278 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1279
1280 if (pos == -1)
1281 return do_compat_writev(fd, vec, vlen, flags);
1282
1283 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1284 }
1285
1286 #endif
1287
1288 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1289 size_t count, loff_t max)
1290 {
1291 struct fd in, out;
1292 struct inode *in_inode, *out_inode;
1293 loff_t pos;
1294 loff_t out_pos;
1295 ssize_t retval;
1296 int fl;
1297
1298 /*
1299 * Get input file, and verify that it is ok..
1300 */
1301 retval = -EBADF;
1302 in = fdget(in_fd);
1303 if (!in.file)
1304 goto out;
1305 if (!(in.file->f_mode & FMODE_READ))
1306 goto fput_in;
1307 retval = -ESPIPE;
1308 if (!ppos) {
1309 pos = in.file->f_pos;
1310 } else {
1311 pos = *ppos;
1312 if (!(in.file->f_mode & FMODE_PREAD))
1313 goto fput_in;
1314 }
1315 retval = rw_verify_area(READ, in.file, &pos, count);
1316 if (retval < 0)
1317 goto fput_in;
1318 count = retval;
1319
1320 /*
1321 * Get output file, and verify that it is ok..
1322 */
1323 retval = -EBADF;
1324 out = fdget(out_fd);
1325 if (!out.file)
1326 goto fput_in;
1327 if (!(out.file->f_mode & FMODE_WRITE))
1328 goto fput_out;
1329 retval = -EINVAL;
1330 in_inode = file_inode(in.file);
1331 out_inode = file_inode(out.file);
1332 out_pos = out.file->f_pos;
1333 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1334 if (retval < 0)
1335 goto fput_out;
1336 count = retval;
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 retval = -EOVERFLOW;
1343 if (pos >= max)
1344 goto fput_out;
1345 count = max - pos;
1346 }
1347
1348 fl = 0;
1349 #if 0
1350 /*
1351 * We need to debate whether we can enable this or not. The
1352 * man page documents EAGAIN return for the output at least,
1353 * and the application is arguably buggy if it doesn't expect
1354 * EAGAIN on a non-blocking file descriptor.
1355 */
1356 if (in.file->f_flags & O_NONBLOCK)
1357 fl = SPLICE_F_NONBLOCK;
1358 #endif
1359 file_start_write(out.file);
1360 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1361 file_end_write(out.file);
1362
1363 if (retval > 0) {
1364 add_rchar(current, retval);
1365 add_wchar(current, retval);
1366 fsnotify_access(in.file);
1367 fsnotify_modify(out.file);
1368 out.file->f_pos = out_pos;
1369 if (ppos)
1370 *ppos = pos;
1371 else
1372 in.file->f_pos = pos;
1373 }
1374
1375 inc_syscr(current);
1376 inc_syscw(current);
1377 if (pos > max)
1378 retval = -EOVERFLOW;
1379
1380 fput_out:
1381 fdput(out);
1382 fput_in:
1383 fdput(in);
1384 out:
1385 return retval;
1386 }
1387
1388 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1389 {
1390 loff_t pos;
1391 off_t off;
1392 ssize_t ret;
1393
1394 if (offset) {
1395 if (unlikely(get_user(off, offset)))
1396 return -EFAULT;
1397 pos = off;
1398 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1399 if (unlikely(put_user(pos, offset)))
1400 return -EFAULT;
1401 return ret;
1402 }
1403
1404 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1405 }
1406
1407 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1408 {
1409 loff_t pos;
1410 ssize_t ret;
1411
1412 if (offset) {
1413 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1414 return -EFAULT;
1415 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1416 if (unlikely(put_user(pos, offset)))
1417 return -EFAULT;
1418 return ret;
1419 }
1420
1421 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1422 }
1423
1424 #ifdef CONFIG_COMPAT
1425 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1426 compat_off_t __user *, offset, compat_size_t, count)
1427 {
1428 loff_t pos;
1429 off_t off;
1430 ssize_t ret;
1431
1432 if (offset) {
1433 if (unlikely(get_user(off, offset)))
1434 return -EFAULT;
1435 pos = off;
1436 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1437 if (unlikely(put_user(pos, offset)))
1438 return -EFAULT;
1439 return ret;
1440 }
1441
1442 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1443 }
1444
1445 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1446 compat_loff_t __user *, offset, compat_size_t, count)
1447 {
1448 loff_t pos;
1449 ssize_t ret;
1450
1451 if (offset) {
1452 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1453 return -EFAULT;
1454 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1455 if (unlikely(put_user(pos, offset)))
1456 return -EFAULT;
1457 return ret;
1458 }
1459
1460 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1461 }
1462 #endif
1463
1464 /*
1465 * copy_file_range() differs from regular file read and write in that it
1466 * specifically allows return partial success. When it does so is up to
1467 * the copy_file_range method.
1468 */
1469 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1470 struct file *file_out, loff_t pos_out,
1471 size_t len, unsigned int flags)
1472 {
1473 struct inode *inode_in = file_inode(file_in);
1474 struct inode *inode_out = file_inode(file_out);
1475 ssize_t ret;
1476
1477 if (flags != 0)
1478 return -EINVAL;
1479
1480 /* copy_file_range allows full ssize_t len, ignoring MAX_RW_COUNT */
1481 ret = rw_verify_area(READ, file_in, &pos_in, len);
1482 if (ret >= 0)
1483 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1484 if (ret < 0)
1485 return ret;
1486
1487 if (!(file_in->f_mode & FMODE_READ) ||
1488 !(file_out->f_mode & FMODE_WRITE) ||
1489 (file_out->f_flags & O_APPEND))
1490 return -EBADF;
1491
1492 /* this could be relaxed once a method supports cross-fs copies */
1493 if (inode_in->i_sb != inode_out->i_sb)
1494 return -EXDEV;
1495
1496 if (len == 0)
1497 return 0;
1498
1499 ret = mnt_want_write_file(file_out);
1500 if (ret)
1501 return ret;
1502
1503 ret = -EOPNOTSUPP;
1504 if (file_out->f_op->copy_file_range)
1505 ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out,
1506 pos_out, len, flags);
1507 if (ret == -EOPNOTSUPP)
1508 ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1509 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1510
1511 if (ret > 0) {
1512 fsnotify_access(file_in);
1513 add_rchar(current, ret);
1514 fsnotify_modify(file_out);
1515 add_wchar(current, ret);
1516 }
1517 inc_syscr(current);
1518 inc_syscw(current);
1519
1520 mnt_drop_write_file(file_out);
1521
1522 return ret;
1523 }
1524 EXPORT_SYMBOL(vfs_copy_file_range);
1525
1526 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1527 int, fd_out, loff_t __user *, off_out,
1528 size_t, len, unsigned int, flags)
1529 {
1530 loff_t pos_in;
1531 loff_t pos_out;
1532 struct fd f_in;
1533 struct fd f_out;
1534 ssize_t ret = -EBADF;
1535
1536 f_in = fdget(fd_in);
1537 if (!f_in.file)
1538 goto out2;
1539
1540 f_out = fdget(fd_out);
1541 if (!f_out.file)
1542 goto out1;
1543
1544 ret = -EFAULT;
1545 if (off_in) {
1546 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1547 goto out;
1548 } else {
1549 pos_in = f_in.file->f_pos;
1550 }
1551
1552 if (off_out) {
1553 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1554 goto out;
1555 } else {
1556 pos_out = f_out.file->f_pos;
1557 }
1558
1559 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1560 flags);
1561 if (ret > 0) {
1562 pos_in += ret;
1563 pos_out += ret;
1564
1565 if (off_in) {
1566 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1567 ret = -EFAULT;
1568 } else {
1569 f_in.file->f_pos = pos_in;
1570 }
1571
1572 if (off_out) {
1573 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1574 ret = -EFAULT;
1575 } else {
1576 f_out.file->f_pos = pos_out;
1577 }
1578 }
1579
1580 out:
1581 fdput(f_out);
1582 out1:
1583 fdput(f_in);
1584 out2:
1585 return ret;
1586 }
1587
1588 static int clone_verify_area(struct file *file, loff_t pos, u64 len, bool write)
1589 {
1590 struct inode *inode = file_inode(file);
1591
1592 if (unlikely(pos < 0))
1593 return -EINVAL;
1594
1595 if (unlikely((loff_t) (pos + len) < 0))
1596 return -EINVAL;
1597
1598 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1599 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1600 int retval;
1601
1602 retval = locks_mandatory_area(inode, file, pos, end,
1603 write ? F_WRLCK : F_RDLCK);
1604 if (retval < 0)
1605 return retval;
1606 }
1607
1608 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1609 }
1610
1611 int vfs_clone_file_range(struct file *file_in, loff_t pos_in,
1612 struct file *file_out, loff_t pos_out, u64 len)
1613 {
1614 struct inode *inode_in = file_inode(file_in);
1615 struct inode *inode_out = file_inode(file_out);
1616 int ret;
1617
1618 if (inode_in->i_sb != inode_out->i_sb ||
1619 file_in->f_path.mnt != file_out->f_path.mnt)
1620 return -EXDEV;
1621
1622 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1623 return -EISDIR;
1624 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1625 return -EINVAL;
1626
1627 if (!(file_in->f_mode & FMODE_READ) ||
1628 !(file_out->f_mode & FMODE_WRITE) ||
1629 (file_out->f_flags & O_APPEND))
1630 return -EBADF;
1631
1632 if (!file_in->f_op->clone_file_range)
1633 return -EOPNOTSUPP;
1634
1635 ret = clone_verify_area(file_in, pos_in, len, false);
1636 if (ret)
1637 return ret;
1638
1639 ret = clone_verify_area(file_out, pos_out, len, true);
1640 if (ret)
1641 return ret;
1642
1643 if (pos_in + len > i_size_read(inode_in))
1644 return -EINVAL;
1645
1646 ret = mnt_want_write_file(file_out);
1647 if (ret)
1648 return ret;
1649
1650 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1651 file_out, pos_out, len);
1652 if (!ret) {
1653 fsnotify_access(file_in);
1654 fsnotify_modify(file_out);
1655 }
1656
1657 mnt_drop_write_file(file_out);
1658 return ret;
1659 }
1660 EXPORT_SYMBOL(vfs_clone_file_range);
1661
1662 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
1663 {
1664 struct file_dedupe_range_info *info;
1665 struct inode *src = file_inode(file);
1666 u64 off;
1667 u64 len;
1668 int i;
1669 int ret;
1670 bool is_admin = capable(CAP_SYS_ADMIN);
1671 u16 count = same->dest_count;
1672 struct file *dst_file;
1673 loff_t dst_off;
1674 ssize_t deduped;
1675
1676 if (!(file->f_mode & FMODE_READ))
1677 return -EINVAL;
1678
1679 if (same->reserved1 || same->reserved2)
1680 return -EINVAL;
1681
1682 off = same->src_offset;
1683 len = same->src_length;
1684
1685 ret = -EISDIR;
1686 if (S_ISDIR(src->i_mode))
1687 goto out;
1688
1689 ret = -EINVAL;
1690 if (!S_ISREG(src->i_mode))
1691 goto out;
1692
1693 ret = clone_verify_area(file, off, len, false);
1694 if (ret < 0)
1695 goto out;
1696 ret = 0;
1697
1698 /* pre-format output fields to sane values */
1699 for (i = 0; i < count; i++) {
1700 same->info[i].bytes_deduped = 0ULL;
1701 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
1702 }
1703
1704 for (i = 0, info = same->info; i < count; i++, info++) {
1705 struct inode *dst;
1706 struct fd dst_fd = fdget(info->dest_fd);
1707
1708 dst_file = dst_fd.file;
1709 if (!dst_file) {
1710 info->status = -EBADF;
1711 goto next_loop;
1712 }
1713 dst = file_inode(dst_file);
1714
1715 ret = mnt_want_write_file(dst_file);
1716 if (ret) {
1717 info->status = ret;
1718 goto next_loop;
1719 }
1720
1721 dst_off = info->dest_offset;
1722 ret = clone_verify_area(dst_file, dst_off, len, true);
1723 if (ret < 0) {
1724 info->status = ret;
1725 goto next_file;
1726 }
1727 ret = 0;
1728
1729 if (info->reserved) {
1730 info->status = -EINVAL;
1731 } else if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) {
1732 info->status = -EINVAL;
1733 } else if (file->f_path.mnt != dst_file->f_path.mnt) {
1734 info->status = -EXDEV;
1735 } else if (S_ISDIR(dst->i_mode)) {
1736 info->status = -EISDIR;
1737 } else if (dst_file->f_op->dedupe_file_range == NULL) {
1738 info->status = -EINVAL;
1739 } else {
1740 deduped = dst_file->f_op->dedupe_file_range(file, off,
1741 len, dst_file,
1742 info->dest_offset);
1743 if (deduped == -EBADE)
1744 info->status = FILE_DEDUPE_RANGE_DIFFERS;
1745 else if (deduped < 0)
1746 info->status = deduped;
1747 else
1748 info->bytes_deduped += deduped;
1749 }
1750
1751 next_file:
1752 mnt_drop_write_file(dst_file);
1753 next_loop:
1754 fdput(dst_fd);
1755
1756 if (fatal_signal_pending(current))
1757 goto out;
1758 }
1759
1760 out:
1761 return ret;
1762 }
1763 EXPORT_SYMBOL(vfs_dedupe_file_range);
ubuntu-zesty-kernel mirror