]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/read_write.c
projects / mirror_ubuntu-artful-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/jack/linux-fs
[mirror_ubuntu-artful-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)
690 return -EOPNOTSUPP;
691
692 init_sync_kiocb(&kiocb, filp);
693 if (flags & RWF_HIPRI)
694 kiocb.ki_flags |= IOCB_HIPRI;
695 kiocb.ki_pos = *ppos;
696
697 ret = fn(&kiocb, iter);
698 BUG_ON(ret == -EIOCBQUEUED);
699 *ppos = kiocb.ki_pos;
700 return ret;
701 }
702
703 /* Do it by hand, with file-ops */
704 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
705 loff_t *ppos, io_fn_t fn, int flags)
706 {
707 ssize_t ret = 0;
708
709 if (flags & ~RWF_HIPRI)
710 return -EOPNOTSUPP;
711
712 while (iov_iter_count(iter)) {
713 struct iovec iovec = iov_iter_iovec(iter);
714 ssize_t nr;
715
716 nr = fn(filp, iovec.iov_base, iovec.iov_len, ppos);
717
718 if (nr < 0) {
719 if (!ret)
720 ret = nr;
721 break;
722 }
723 ret += nr;
724 if (nr != iovec.iov_len)
725 break;
726 iov_iter_advance(iter, nr);
727 }
728
729 return ret;
730 }
731
732 /* A write operation does a read from user space and vice versa */
733 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
734
735 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
736 unsigned long nr_segs, unsigned long fast_segs,
737 struct iovec *fast_pointer,
738 struct iovec **ret_pointer)
739 {
740 unsigned long seg;
741 ssize_t ret;
742 struct iovec *iov = fast_pointer;
743
744 /*
745 * SuS says "The readv() function *may* fail if the iovcnt argument
746 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
747 * traditionally returned zero for zero segments, so...
748 */
749 if (nr_segs == 0) {
750 ret = 0;
751 goto out;
752 }
753
754 /*
755 * First get the "struct iovec" from user memory and
756 * verify all the pointers
757 */
758 if (nr_segs > UIO_MAXIOV) {
759 ret = -EINVAL;
760 goto out;
761 }
762 if (nr_segs > fast_segs) {
763 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
764 if (iov == NULL) {
765 ret = -ENOMEM;
766 goto out;
767 }
768 }
769 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
770 ret = -EFAULT;
771 goto out;
772 }
773
774 /*
775 * According to the Single Unix Specification we should return EINVAL
776 * if an element length is < 0 when cast to ssize_t or if the
777 * total length would overflow the ssize_t return value of the
778 * system call.
779 *
780 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
781 * overflow case.
782 */
783 ret = 0;
784 for (seg = 0; seg < nr_segs; seg++) {
785 void __user *buf = iov[seg].iov_base;
786 ssize_t len = (ssize_t)iov[seg].iov_len;
787
788 /* see if we we're about to use an invalid len or if
789 * it's about to overflow ssize_t */
790 if (len < 0) {
791 ret = -EINVAL;
792 goto out;
793 }
794 if (type >= 0
795 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
796 ret = -EFAULT;
797 goto out;
798 }
799 if (len > MAX_RW_COUNT - ret) {
800 len = MAX_RW_COUNT - ret;
801 iov[seg].iov_len = len;
802 }
803 ret += len;
804 }
805 out:
806 *ret_pointer = iov;
807 return ret;
808 }
809
810 static ssize_t do_readv_writev(int type, struct file *file,
811 const struct iovec __user * uvector,
812 unsigned long nr_segs, loff_t *pos,
813 int flags)
814 {
815 size_t tot_len;
816 struct iovec iovstack[UIO_FASTIOV];
817 struct iovec *iov = iovstack;
818 struct iov_iter iter;
819 ssize_t ret;
820 io_fn_t fn;
821 iter_fn_t iter_fn;
822
823 ret = import_iovec(type, uvector, nr_segs,
824 ARRAY_SIZE(iovstack), &iov, &iter);
825 if (ret < 0)
826 return ret;
827
828 tot_len = iov_iter_count(&iter);
829 if (!tot_len)
830 goto out;
831 ret = rw_verify_area(type, file, pos, tot_len);
832 if (ret < 0)
833 goto out;
834
835 if (type == READ) {
836 fn = file->f_op->read;
837 iter_fn = file->f_op->read_iter;
838 } else {
839 fn = (io_fn_t)file->f_op->write;
840 iter_fn = file->f_op->write_iter;
841 file_start_write(file);
842 }
843
844 if (iter_fn)
845 ret = do_iter_readv_writev(file, &iter, pos, iter_fn, flags);
846 else
847 ret = do_loop_readv_writev(file, &iter, pos, fn, flags);
848
849 if (type != READ)
850 file_end_write(file);
851
852 out:
853 kfree(iov);
854 if ((ret + (type == READ)) > 0) {
855 if (type == READ)
856 fsnotify_access(file);
857 else
858 fsnotify_modify(file);
859 }
860 return ret;
861 }
862
863 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
864 unsigned long vlen, loff_t *pos, int flags)
865 {
866 if (!(file->f_mode & FMODE_READ))
867 return -EBADF;
868 if (!(file->f_mode & FMODE_CAN_READ))
869 return -EINVAL;
870
871 return do_readv_writev(READ, file, vec, vlen, pos, flags);
872 }
873
874 EXPORT_SYMBOL(vfs_readv);
875
876 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
877 unsigned long vlen, loff_t *pos, int flags)
878 {
879 if (!(file->f_mode & FMODE_WRITE))
880 return -EBADF;
881 if (!(file->f_mode & FMODE_CAN_WRITE))
882 return -EINVAL;
883
884 return do_readv_writev(WRITE, file, vec, vlen, pos, flags);
885 }
886
887 EXPORT_SYMBOL(vfs_writev);
888
889 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
890 unsigned long vlen, int flags)
891 {
892 struct fd f = fdget_pos(fd);
893 ssize_t ret = -EBADF;
894
895 if (f.file) {
896 loff_t pos = file_pos_read(f.file);
897 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
898 if (ret >= 0)
899 file_pos_write(f.file, pos);
900 fdput_pos(f);
901 }
902
903 if (ret > 0)
904 add_rchar(current, ret);
905 inc_syscr(current);
906 return ret;
907 }
908
909 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
910 unsigned long vlen, int flags)
911 {
912 struct fd f = fdget_pos(fd);
913 ssize_t ret = -EBADF;
914
915 if (f.file) {
916 loff_t pos = file_pos_read(f.file);
917 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
918 if (ret >= 0)
919 file_pos_write(f.file, pos);
920 fdput_pos(f);
921 }
922
923 if (ret > 0)
924 add_wchar(current, ret);
925 inc_syscw(current);
926 return ret;
927 }
928
929 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
930 {
931 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
932 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
933 }
934
935 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
936 unsigned long vlen, loff_t pos, int flags)
937 {
938 struct fd f;
939 ssize_t ret = -EBADF;
940
941 if (pos < 0)
942 return -EINVAL;
943
944 f = fdget(fd);
945 if (f.file) {
946 ret = -ESPIPE;
947 if (f.file->f_mode & FMODE_PREAD)
948 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
949 fdput(f);
950 }
951
952 if (ret > 0)
953 add_rchar(current, ret);
954 inc_syscr(current);
955 return ret;
956 }
957
958 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
959 unsigned long vlen, loff_t pos, int flags)
960 {
961 struct fd f;
962 ssize_t ret = -EBADF;
963
964 if (pos < 0)
965 return -EINVAL;
966
967 f = fdget(fd);
968 if (f.file) {
969 ret = -ESPIPE;
970 if (f.file->f_mode & FMODE_PWRITE)
971 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
972 fdput(f);
973 }
974
975 if (ret > 0)
976 add_wchar(current, ret);
977 inc_syscw(current);
978 return ret;
979 }
980
981 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
982 unsigned long, vlen)
983 {
984 return do_readv(fd, vec, vlen, 0);
985 }
986
987 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
988 unsigned long, vlen)
989 {
990 return do_writev(fd, vec, vlen, 0);
991 }
992
993 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
994 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
995 {
996 loff_t pos = pos_from_hilo(pos_h, pos_l);
997
998 return do_preadv(fd, vec, vlen, pos, 0);
999 }
1000
1001 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1002 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1003 int, flags)
1004 {
1005 loff_t pos = pos_from_hilo(pos_h, pos_l);
1006
1007 if (pos == -1)
1008 return do_readv(fd, vec, vlen, flags);
1009
1010 return do_preadv(fd, vec, vlen, pos, flags);
1011 }
1012
1013 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1014 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1015 {
1016 loff_t pos = pos_from_hilo(pos_h, pos_l);
1017
1018 return do_pwritev(fd, vec, vlen, pos, 0);
1019 }
1020
1021 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1022 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1023 int, flags)
1024 {
1025 loff_t pos = pos_from_hilo(pos_h, pos_l);
1026
1027 if (pos == -1)
1028 return do_writev(fd, vec, vlen, flags);
1029
1030 return do_pwritev(fd, vec, vlen, pos, flags);
1031 }
1032
1033 #ifdef CONFIG_COMPAT
1034
1035 static ssize_t compat_do_readv_writev(int type, struct file *file,
1036 const struct compat_iovec __user *uvector,
1037 unsigned long nr_segs, loff_t *pos,
1038 int flags)
1039 {
1040 compat_ssize_t tot_len;
1041 struct iovec iovstack[UIO_FASTIOV];
1042 struct iovec *iov = iovstack;
1043 struct iov_iter iter;
1044 ssize_t ret;
1045 io_fn_t fn;
1046 iter_fn_t iter_fn;
1047
1048 ret = compat_import_iovec(type, uvector, nr_segs,
1049 UIO_FASTIOV, &iov, &iter);
1050 if (ret < 0)
1051 return ret;
1052
1053 tot_len = iov_iter_count(&iter);
1054 if (!tot_len)
1055 goto out;
1056 ret = rw_verify_area(type, file, pos, tot_len);
1057 if (ret < 0)
1058 goto out;
1059
1060 if (type == READ) {
1061 fn = file->f_op->read;
1062 iter_fn = file->f_op->read_iter;
1063 } else {
1064 fn = (io_fn_t)file->f_op->write;
1065 iter_fn = file->f_op->write_iter;
1066 file_start_write(file);
1067 }
1068
1069 if (iter_fn)
1070 ret = do_iter_readv_writev(file, &iter, pos, iter_fn, flags);
1071 else
1072 ret = do_loop_readv_writev(file, &iter, pos, fn, flags);
1073
1074 if (type != READ)
1075 file_end_write(file);
1076
1077 out:
1078 kfree(iov);
1079 if ((ret + (type == READ)) > 0) {
1080 if (type == READ)
1081 fsnotify_access(file);
1082 else
1083 fsnotify_modify(file);
1084 }
1085 return ret;
1086 }
1087
1088 static size_t compat_readv(struct file *file,
1089 const struct compat_iovec __user *vec,
1090 unsigned long vlen, loff_t *pos, int flags)
1091 {
1092 ssize_t ret = -EBADF;
1093
1094 if (!(file->f_mode & FMODE_READ))
1095 goto out;
1096
1097 ret = -EINVAL;
1098 if (!(file->f_mode & FMODE_CAN_READ))
1099 goto out;
1100
1101 ret = compat_do_readv_writev(READ, file, vec, vlen, pos, flags);
1102
1103 out:
1104 if (ret > 0)
1105 add_rchar(current, ret);
1106 inc_syscr(current);
1107 return ret;
1108 }
1109
1110 static size_t do_compat_readv(compat_ulong_t fd,
1111 const struct compat_iovec __user *vec,
1112 compat_ulong_t vlen, int flags)
1113 {
1114 struct fd f = fdget_pos(fd);
1115 ssize_t ret;
1116 loff_t pos;
1117
1118 if (!f.file)
1119 return -EBADF;
1120 pos = f.file->f_pos;
1121 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1122 if (ret >= 0)
1123 f.file->f_pos = pos;
1124 fdput_pos(f);
1125 return ret;
1126
1127 }
1128
1129 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1130 const struct compat_iovec __user *,vec,
1131 compat_ulong_t, vlen)
1132 {
1133 return do_compat_readv(fd, vec, vlen, 0);
1134 }
1135
1136 static long do_compat_preadv64(unsigned long fd,
1137 const struct compat_iovec __user *vec,
1138 unsigned long vlen, loff_t pos, int flags)
1139 {
1140 struct fd f;
1141 ssize_t ret;
1142
1143 if (pos < 0)
1144 return -EINVAL;
1145 f = fdget(fd);
1146 if (!f.file)
1147 return -EBADF;
1148 ret = -ESPIPE;
1149 if (f.file->f_mode & FMODE_PREAD)
1150 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1151 fdput(f);
1152 return ret;
1153 }
1154
1155 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1156 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1157 const struct compat_iovec __user *,vec,
1158 unsigned long, vlen, loff_t, pos)
1159 {
1160 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1161 }
1162 #endif
1163
1164 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1165 const struct compat_iovec __user *,vec,
1166 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1167 {
1168 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1169
1170 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1171 }
1172
1173 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1174 const struct compat_iovec __user *,vec,
1175 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1176 int, flags)
1177 {
1178 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1179
1180 if (pos == -1)
1181 return do_compat_readv(fd, vec, vlen, flags);
1182
1183 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1184 }
1185
1186 static size_t compat_writev(struct file *file,
1187 const struct compat_iovec __user *vec,
1188 unsigned long vlen, loff_t *pos, int flags)
1189 {
1190 ssize_t ret = -EBADF;
1191
1192 if (!(file->f_mode & FMODE_WRITE))
1193 goto out;
1194
1195 ret = -EINVAL;
1196 if (!(file->f_mode & FMODE_CAN_WRITE))
1197 goto out;
1198
1199 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos, 0);
1200
1201 out:
1202 if (ret > 0)
1203 add_wchar(current, ret);
1204 inc_syscw(current);
1205 return ret;
1206 }
1207
1208 static size_t do_compat_writev(compat_ulong_t fd,
1209 const struct compat_iovec __user* vec,
1210 compat_ulong_t vlen, int flags)
1211 {
1212 struct fd f = fdget_pos(fd);
1213 ssize_t ret;
1214 loff_t pos;
1215
1216 if (!f.file)
1217 return -EBADF;
1218 pos = f.file->f_pos;
1219 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1220 if (ret >= 0)
1221 f.file->f_pos = pos;
1222 fdput_pos(f);
1223 return ret;
1224 }
1225
1226 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1227 const struct compat_iovec __user *, vec,
1228 compat_ulong_t, vlen)
1229 {
1230 return do_compat_writev(fd, vec, vlen, 0);
1231 }
1232
1233 static long do_compat_pwritev64(unsigned long fd,
1234 const struct compat_iovec __user *vec,
1235 unsigned long vlen, loff_t pos, int flags)
1236 {
1237 struct fd f;
1238 ssize_t ret;
1239
1240 if (pos < 0)
1241 return -EINVAL;
1242 f = fdget(fd);
1243 if (!f.file)
1244 return -EBADF;
1245 ret = -ESPIPE;
1246 if (f.file->f_mode & FMODE_PWRITE)
1247 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1248 fdput(f);
1249 return ret;
1250 }
1251
1252 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1253 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1254 const struct compat_iovec __user *,vec,
1255 unsigned long, vlen, loff_t, pos)
1256 {
1257 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1258 }
1259 #endif
1260
1261 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1262 const struct compat_iovec __user *,vec,
1263 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1264 {
1265 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1266
1267 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1268 }
1269
1270 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1271 const struct compat_iovec __user *,vec,
1272 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, int, flags)
1273 {
1274 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1275
1276 if (pos == -1)
1277 return do_compat_writev(fd, vec, vlen, flags);
1278
1279 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1280 }
1281
1282 #endif
1283
1284 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1285 size_t count, loff_t max)
1286 {
1287 struct fd in, out;
1288 struct inode *in_inode, *out_inode;
1289 loff_t pos;
1290 loff_t out_pos;
1291 ssize_t retval;
1292 int fl;
1293
1294 /*
1295 * Get input file, and verify that it is ok..
1296 */
1297 retval = -EBADF;
1298 in = fdget(in_fd);
1299 if (!in.file)
1300 goto out;
1301 if (!(in.file->f_mode & FMODE_READ))
1302 goto fput_in;
1303 retval = -ESPIPE;
1304 if (!ppos) {
1305 pos = in.file->f_pos;
1306 } else {
1307 pos = *ppos;
1308 if (!(in.file->f_mode & FMODE_PREAD))
1309 goto fput_in;
1310 }
1311 retval = rw_verify_area(READ, in.file, &pos, count);
1312 if (retval < 0)
1313 goto fput_in;
1314 count = retval;
1315
1316 /*
1317 * Get output file, and verify that it is ok..
1318 */
1319 retval = -EBADF;
1320 out = fdget(out_fd);
1321 if (!out.file)
1322 goto fput_in;
1323 if (!(out.file->f_mode & FMODE_WRITE))
1324 goto fput_out;
1325 retval = -EINVAL;
1326 in_inode = file_inode(in.file);
1327 out_inode = file_inode(out.file);
1328 out_pos = out.file->f_pos;
1329 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1330 if (retval < 0)
1331 goto fput_out;
1332 count = retval;
1333
1334 if (!max)
1335 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1336
1337 if (unlikely(pos + count > max)) {
1338 retval = -EOVERFLOW;
1339 if (pos >= max)
1340 goto fput_out;
1341 count = max - pos;
1342 }
1343
1344 fl = 0;
1345 #if 0
1346 /*
1347 * We need to debate whether we can enable this or not. The
1348 * man page documents EAGAIN return for the output at least,
1349 * and the application is arguably buggy if it doesn't expect
1350 * EAGAIN on a non-blocking file descriptor.
1351 */
1352 if (in.file->f_flags & O_NONBLOCK)
1353 fl = SPLICE_F_NONBLOCK;
1354 #endif
1355 file_start_write(out.file);
1356 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1357 file_end_write(out.file);
1358
1359 if (retval > 0) {
1360 add_rchar(current, retval);
1361 add_wchar(current, retval);
1362 fsnotify_access(in.file);
1363 fsnotify_modify(out.file);
1364 out.file->f_pos = out_pos;
1365 if (ppos)
1366 *ppos = pos;
1367 else
1368 in.file->f_pos = pos;
1369 }
1370
1371 inc_syscr(current);
1372 inc_syscw(current);
1373 if (pos > max)
1374 retval = -EOVERFLOW;
1375
1376 fput_out:
1377 fdput(out);
1378 fput_in:
1379 fdput(in);
1380 out:
1381 return retval;
1382 }
1383
1384 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1385 {
1386 loff_t pos;
1387 off_t off;
1388 ssize_t ret;
1389
1390 if (offset) {
1391 if (unlikely(get_user(off, offset)))
1392 return -EFAULT;
1393 pos = off;
1394 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1395 if (unlikely(put_user(pos, offset)))
1396 return -EFAULT;
1397 return ret;
1398 }
1399
1400 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1401 }
1402
1403 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1404 {
1405 loff_t pos;
1406 ssize_t ret;
1407
1408 if (offset) {
1409 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1410 return -EFAULT;
1411 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1412 if (unlikely(put_user(pos, offset)))
1413 return -EFAULT;
1414 return ret;
1415 }
1416
1417 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1418 }
1419
1420 #ifdef CONFIG_COMPAT
1421 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1422 compat_off_t __user *, offset, compat_size_t, count)
1423 {
1424 loff_t pos;
1425 off_t off;
1426 ssize_t ret;
1427
1428 if (offset) {
1429 if (unlikely(get_user(off, offset)))
1430 return -EFAULT;
1431 pos = off;
1432 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1433 if (unlikely(put_user(pos, offset)))
1434 return -EFAULT;
1435 return ret;
1436 }
1437
1438 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1439 }
1440
1441 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1442 compat_loff_t __user *, offset, compat_size_t, count)
1443 {
1444 loff_t pos;
1445 ssize_t ret;
1446
1447 if (offset) {
1448 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1449 return -EFAULT;
1450 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1451 if (unlikely(put_user(pos, offset)))
1452 return -EFAULT;
1453 return ret;
1454 }
1455
1456 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1457 }
1458 #endif
1459
1460 /*
1461 * copy_file_range() differs from regular file read and write in that it
1462 * specifically allows return partial success. When it does so is up to
1463 * the copy_file_range method.
1464 */
1465 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1466 struct file *file_out, loff_t pos_out,
1467 size_t len, unsigned int flags)
1468 {
1469 struct inode *inode_in = file_inode(file_in);
1470 struct inode *inode_out = file_inode(file_out);
1471 ssize_t ret;
1472
1473 if (flags != 0)
1474 return -EINVAL;
1475
1476 /* copy_file_range allows full ssize_t len, ignoring MAX_RW_COUNT */
1477 ret = rw_verify_area(READ, file_in, &pos_in, len);
1478 if (ret >= 0)
1479 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1480 if (ret < 0)
1481 return ret;
1482
1483 if (!(file_in->f_mode & FMODE_READ) ||
1484 !(file_out->f_mode & FMODE_WRITE) ||
1485 (file_out->f_flags & O_APPEND))
1486 return -EBADF;
1487
1488 /* this could be relaxed once a method supports cross-fs copies */
1489 if (inode_in->i_sb != inode_out->i_sb)
1490 return -EXDEV;
1491
1492 if (len == 0)
1493 return 0;
1494
1495 ret = mnt_want_write_file(file_out);
1496 if (ret)
1497 return ret;
1498
1499 ret = -EOPNOTSUPP;
1500 if (file_out->f_op->copy_file_range)
1501 ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out,
1502 pos_out, len, flags);
1503 if (ret == -EOPNOTSUPP)
1504 ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1505 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1506
1507 if (ret > 0) {
1508 fsnotify_access(file_in);
1509 add_rchar(current, ret);
1510 fsnotify_modify(file_out);
1511 add_wchar(current, ret);
1512 }
1513 inc_syscr(current);
1514 inc_syscw(current);
1515
1516 mnt_drop_write_file(file_out);
1517
1518 return ret;
1519 }
1520 EXPORT_SYMBOL(vfs_copy_file_range);
1521
1522 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1523 int, fd_out, loff_t __user *, off_out,
1524 size_t, len, unsigned int, flags)
1525 {
1526 loff_t pos_in;
1527 loff_t pos_out;
1528 struct fd f_in;
1529 struct fd f_out;
1530 ssize_t ret = -EBADF;
1531
1532 f_in = fdget(fd_in);
1533 if (!f_in.file)
1534 goto out2;
1535
1536 f_out = fdget(fd_out);
1537 if (!f_out.file)
1538 goto out1;
1539
1540 ret = -EFAULT;
1541 if (off_in) {
1542 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1543 goto out;
1544 } else {
1545 pos_in = f_in.file->f_pos;
1546 }
1547
1548 if (off_out) {
1549 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1550 goto out;
1551 } else {
1552 pos_out = f_out.file->f_pos;
1553 }
1554
1555 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1556 flags);
1557 if (ret > 0) {
1558 pos_in += ret;
1559 pos_out += ret;
1560
1561 if (off_in) {
1562 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1563 ret = -EFAULT;
1564 } else {
1565 f_in.file->f_pos = pos_in;
1566 }
1567
1568 if (off_out) {
1569 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1570 ret = -EFAULT;
1571 } else {
1572 f_out.file->f_pos = pos_out;
1573 }
1574 }
1575
1576 out:
1577 fdput(f_out);
1578 out1:
1579 fdput(f_in);
1580 out2:
1581 return ret;
1582 }
1583
1584 static int clone_verify_area(struct file *file, loff_t pos, u64 len, bool write)
1585 {
1586 struct inode *inode = file_inode(file);
1587
1588 if (unlikely(pos < 0))
1589 return -EINVAL;
1590
1591 if (unlikely((loff_t) (pos + len) < 0))
1592 return -EINVAL;
1593
1594 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1595 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1596 int retval;
1597
1598 retval = locks_mandatory_area(inode, file, pos, end,
1599 write ? F_WRLCK : F_RDLCK);
1600 if (retval < 0)
1601 return retval;
1602 }
1603
1604 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1605 }
1606
1607 int vfs_clone_file_range(struct file *file_in, loff_t pos_in,
1608 struct file *file_out, loff_t pos_out, u64 len)
1609 {
1610 struct inode *inode_in = file_inode(file_in);
1611 struct inode *inode_out = file_inode(file_out);
1612 int ret;
1613
1614 if (inode_in->i_sb != inode_out->i_sb ||
1615 file_in->f_path.mnt != file_out->f_path.mnt)
1616 return -EXDEV;
1617
1618 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1619 return -EISDIR;
1620 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1621 return -EINVAL;
1622
1623 if (!(file_in->f_mode & FMODE_READ) ||
1624 !(file_out->f_mode & FMODE_WRITE) ||
1625 (file_out->f_flags & O_APPEND))
1626 return -EBADF;
1627
1628 if (!file_in->f_op->clone_file_range)
1629 return -EOPNOTSUPP;
1630
1631 ret = clone_verify_area(file_in, pos_in, len, false);
1632 if (ret)
1633 return ret;
1634
1635 ret = clone_verify_area(file_out, pos_out, len, true);
1636 if (ret)
1637 return ret;
1638
1639 if (pos_in + len > i_size_read(inode_in))
1640 return -EINVAL;
1641
1642 ret = mnt_want_write_file(file_out);
1643 if (ret)
1644 return ret;
1645
1646 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1647 file_out, pos_out, len);
1648 if (!ret) {
1649 fsnotify_access(file_in);
1650 fsnotify_modify(file_out);
1651 }
1652
1653 mnt_drop_write_file(file_out);
1654 return ret;
1655 }
1656 EXPORT_SYMBOL(vfs_clone_file_range);
1657
1658 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
1659 {
1660 struct file_dedupe_range_info *info;
1661 struct inode *src = file_inode(file);
1662 u64 off;
1663 u64 len;
1664 int i;
1665 int ret;
1666 bool is_admin = capable(CAP_SYS_ADMIN);
1667 u16 count = same->dest_count;
1668 struct file *dst_file;
1669 loff_t dst_off;
1670 ssize_t deduped;
1671
1672 if (!(file->f_mode & FMODE_READ))
1673 return -EINVAL;
1674
1675 if (same->reserved1 || same->reserved2)
1676 return -EINVAL;
1677
1678 off = same->src_offset;
1679 len = same->src_length;
1680
1681 ret = -EISDIR;
1682 if (S_ISDIR(src->i_mode))
1683 goto out;
1684
1685 ret = -EINVAL;
1686 if (!S_ISREG(src->i_mode))
1687 goto out;
1688
1689 ret = clone_verify_area(file, off, len, false);
1690 if (ret < 0)
1691 goto out;
1692 ret = 0;
1693
1694 /* pre-format output fields to sane values */
1695 for (i = 0; i < count; i++) {
1696 same->info[i].bytes_deduped = 0ULL;
1697 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
1698 }
1699
1700 for (i = 0, info = same->info; i < count; i++, info++) {
1701 struct inode *dst;
1702 struct fd dst_fd = fdget(info->dest_fd);
1703
1704 dst_file = dst_fd.file;
1705 if (!dst_file) {
1706 info->status = -EBADF;
1707 goto next_loop;
1708 }
1709 dst = file_inode(dst_file);
1710
1711 ret = mnt_want_write_file(dst_file);
1712 if (ret) {
1713 info->status = ret;
1714 goto next_loop;
1715 }
1716
1717 dst_off = info->dest_offset;
1718 ret = clone_verify_area(dst_file, dst_off, len, true);
1719 if (ret < 0) {
1720 info->status = ret;
1721 goto next_file;
1722 }
1723 ret = 0;
1724
1725 if (info->reserved) {
1726 info->status = -EINVAL;
1727 } else if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) {
1728 info->status = -EINVAL;
1729 } else if (file->f_path.mnt != dst_file->f_path.mnt) {
1730 info->status = -EXDEV;
1731 } else if (S_ISDIR(dst->i_mode)) {
1732 info->status = -EISDIR;
1733 } else if (dst_file->f_op->dedupe_file_range == NULL) {
1734 info->status = -EINVAL;
1735 } else {
1736 deduped = dst_file->f_op->dedupe_file_range(file, off,
1737 len, dst_file,
1738 info->dest_offset);
1739 if (deduped == -EBADE)
1740 info->status = FILE_DEDUPE_RANGE_DIFFERS;
1741 else if (deduped < 0)
1742 info->status = deduped;
1743 else
1744 info->bytes_deduped += deduped;
1745 }
1746
1747 next_file:
1748 mnt_drop_write_file(dst_file);
1749 next_loop:
1750 fdput(dst_fd);
1751
1752 if (fatal_signal_pending(current))
1753 goto out;
1754 }
1755
1756 out:
1757 return ret;
1758 }
1759 EXPORT_SYMBOL(vfs_dedupe_file_range);
Ubuntu Artful kernel mirror