projects / GitHub / exynos8895 / android_kernel_samsung_universal8895.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
fs: move struct kiocb to fs.h
[GitHub/exynos8895/android_kernel_samsung_universal8895.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 "internal.h"
20
21 #include <asm/uaccess.h>
22 #include <asm/unistd.h>
23
24 typedef ssize_t (*io_fn_t)(struct file *, char __user *, size_t, loff_t *);
25 typedef ssize_t (*iov_fn_t)(struct kiocb *, const struct iovec *,
26 unsigned long, 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 = new_sync_read,
32 .read_iter = generic_file_read_iter,
33 .mmap = generic_file_readonly_mmap,
34 .splice_read = generic_file_splice_read,
35 };
36
37 EXPORT_SYMBOL(generic_ro_fops);
38
39 static inline int unsigned_offsets(struct file *file)
40 {
41 return file->f_mode & FMODE_UNSIGNED_OFFSET;
42 }
43
44 /**
45 * vfs_setpos - update the file offset for lseek
46 * @file: file structure in question
47 * @offset: file offset to seek to
48 * @maxsize: maximum file size
49 *
50 * This is a low-level filesystem helper for updating the file offset to
51 * the value specified by @offset if the given offset is valid and it is
52 * not equal to the current file offset.
53 *
54 * Return the specified offset on success and -EINVAL on invalid offset.
55 */
56 loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
57 {
58 if (offset < 0 && !unsigned_offsets(file))
59 return -EINVAL;
60 if (offset > maxsize)
61 return -EINVAL;
62
63 if (offset != file->f_pos) {
64 file->f_pos = offset;
65 file->f_version = 0;
66 }
67 return offset;
68 }
69 EXPORT_SYMBOL(vfs_setpos);
70
71 /**
72 * generic_file_llseek_size - generic llseek implementation for regular files
73 * @file: file structure to seek on
74 * @offset: file offset to seek to
75 * @whence: type of seek
76 * @size: max size of this file in file system
77 * @eof: offset used for SEEK_END position
78 *
79 * This is a variant of generic_file_llseek that allows passing in a custom
80 * maximum file size and a custom EOF position, for e.g. hashed directories
81 *
82 * Synchronization:
83 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
84 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
85 * read/writes behave like SEEK_SET against seeks.
86 */
87 loff_t
88 generic_file_llseek_size(struct file *file, loff_t offset, int whence,
89 loff_t maxsize, loff_t eof)
90 {
91 switch (whence) {
92 case SEEK_END:
93 offset += eof;
94 break;
95 case SEEK_CUR:
96 /*
97 * Here we special-case the lseek(fd, 0, SEEK_CUR)
98 * position-querying operation. Avoid rewriting the "same"
99 * f_pos value back to the file because a concurrent read(),
100 * write() or lseek() might have altered it
101 */
102 if (offset == 0)
103 return file->f_pos;
104 /*
105 * f_lock protects against read/modify/write race with other
106 * SEEK_CURs. Note that parallel writes and reads behave
107 * like SEEK_SET.
108 */
109 spin_lock(&file->f_lock);
110 offset = vfs_setpos(file, file->f_pos + offset, maxsize);
111 spin_unlock(&file->f_lock);
112 return offset;
113 case SEEK_DATA:
114 /*
115 * In the generic case the entire file is data, so as long as
116 * offset isn't at the end of the file then the offset is data.
117 */
118 if (offset >= eof)
119 return -ENXIO;
120 break;
121 case SEEK_HOLE:
122 /*
123 * There is a virtual hole at the end of the file, so as long as
124 * offset isn't i_size or larger, return i_size.
125 */
126 if (offset >= eof)
127 return -ENXIO;
128 offset = eof;
129 break;
130 }
131
132 return vfs_setpos(file, offset, maxsize);
133 }
134 EXPORT_SYMBOL(generic_file_llseek_size);
135
136 /**
137 * generic_file_llseek - generic llseek implementation for regular files
138 * @file: file structure to seek on
139 * @offset: file offset to seek to
140 * @whence: type of seek
141 *
142 * This is a generic implemenation of ->llseek useable for all normal local
143 * filesystems. It just updates the file offset to the value specified by
144 * @offset and @whence.
145 */
146 loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
147 {
148 struct inode *inode = file->f_mapping->host;
149
150 return generic_file_llseek_size(file, offset, whence,
151 inode->i_sb->s_maxbytes,
152 i_size_read(inode));
153 }
154 EXPORT_SYMBOL(generic_file_llseek);
155
156 /**
157 * fixed_size_llseek - llseek implementation for fixed-sized devices
158 * @file: file structure to seek on
159 * @offset: file offset to seek to
160 * @whence: type of seek
161 * @size: size of the file
162 *
163 */
164 loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
165 {
166 switch (whence) {
167 case SEEK_SET: case SEEK_CUR: case SEEK_END:
168 return generic_file_llseek_size(file, offset, whence,
169 size, size);
170 default:
171 return -EINVAL;
172 }
173 }
174 EXPORT_SYMBOL(fixed_size_llseek);
175
176 /**
177 * noop_llseek - No Operation Performed llseek implementation
178 * @file: file structure to seek on
179 * @offset: file offset to seek to
180 * @whence: type of seek
181 *
182 * This is an implementation of ->llseek useable for the rare special case when
183 * userspace expects the seek to succeed but the (device) file is actually not
184 * able to perform the seek. In this case you use noop_llseek() instead of
185 * falling back to the default implementation of ->llseek.
186 */
187 loff_t noop_llseek(struct file *file, loff_t offset, int whence)
188 {
189 return file->f_pos;
190 }
191 EXPORT_SYMBOL(noop_llseek);
192
193 loff_t no_llseek(struct file *file, loff_t offset, int whence)
194 {
195 return -ESPIPE;
196 }
197 EXPORT_SYMBOL(no_llseek);
198
199 loff_t default_llseek(struct file *file, loff_t offset, int whence)
200 {
201 struct inode *inode = file_inode(file);
202 loff_t retval;
203
204 mutex_lock(&inode->i_mutex);
205 switch (whence) {
206 case SEEK_END:
207 offset += i_size_read(inode);
208 break;
209 case SEEK_CUR:
210 if (offset == 0) {
211 retval = file->f_pos;
212 goto out;
213 }
214 offset += file->f_pos;
215 break;
216 case SEEK_DATA:
217 /*
218 * In the generic case the entire file is data, so as
219 * long as offset isn't at the end of the file then the
220 * offset is data.
221 */
222 if (offset >= inode->i_size) {
223 retval = -ENXIO;
224 goto out;
225 }
226 break;
227 case SEEK_HOLE:
228 /*
229 * There is a virtual hole at the end of the file, so
230 * as long as offset isn't i_size or larger, return
231 * i_size.
232 */
233 if (offset >= inode->i_size) {
234 retval = -ENXIO;
235 goto out;
236 }
237 offset = inode->i_size;
238 break;
239 }
240 retval = -EINVAL;
241 if (offset >= 0 || unsigned_offsets(file)) {
242 if (offset != file->f_pos) {
243 file->f_pos = offset;
244 file->f_version = 0;
245 }
246 retval = offset;
247 }
248 out:
249 mutex_unlock(&inode->i_mutex);
250 return retval;
251 }
252 EXPORT_SYMBOL(default_llseek);
253
254 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
255 {
256 loff_t (*fn)(struct file *, loff_t, int);
257
258 fn = no_llseek;
259 if (file->f_mode & FMODE_LSEEK) {
260 if (file->f_op->llseek)
261 fn = file->f_op->llseek;
262 }
263 return fn(file, offset, whence);
264 }
265 EXPORT_SYMBOL(vfs_llseek);
266
267 static inline struct fd fdget_pos(int fd)
268 {
269 return __to_fd(__fdget_pos(fd));
270 }
271
272 static inline void fdput_pos(struct fd f)
273 {
274 if (f.flags & FDPUT_POS_UNLOCK)
275 mutex_unlock(&f.file->f_pos_lock);
276 fdput(f);
277 }
278
279 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
280 {
281 off_t retval;
282 struct fd f = fdget_pos(fd);
283 if (!f.file)
284 return -EBADF;
285
286 retval = -EINVAL;
287 if (whence <= SEEK_MAX) {
288 loff_t res = vfs_llseek(f.file, offset, whence);
289 retval = res;
290 if (res != (loff_t)retval)
291 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
292 }
293 fdput_pos(f);
294 return retval;
295 }
296
297 #ifdef CONFIG_COMPAT
298 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
299 {
300 return sys_lseek(fd, offset, whence);
301 }
302 #endif
303
304 #ifdef __ARCH_WANT_SYS_LLSEEK
305 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
306 unsigned long, offset_low, loff_t __user *, result,
307 unsigned int, whence)
308 {
309 int retval;
310 struct fd f = fdget_pos(fd);
311 loff_t offset;
312
313 if (!f.file)
314 return -EBADF;
315
316 retval = -EINVAL;
317 if (whence > SEEK_MAX)
318 goto out_putf;
319
320 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
321 whence);
322
323 retval = (int)offset;
324 if (offset >= 0) {
325 retval = -EFAULT;
326 if (!copy_to_user(result, &offset, sizeof(offset)))
327 retval = 0;
328 }
329 out_putf:
330 fdput_pos(f);
331 return retval;
332 }
333 #endif
334
335 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos)
336 {
337 struct kiocb kiocb;
338 ssize_t ret;
339
340 if (!file->f_op->read_iter)
341 return -EINVAL;
342
343 init_sync_kiocb(&kiocb, file);
344 kiocb.ki_pos = *ppos;
345
346 iter->type |= READ;
347 ret = file->f_op->read_iter(&kiocb, iter);
348 BUG_ON(ret == -EIOCBQUEUED);
349 if (ret > 0)
350 *ppos = kiocb.ki_pos;
351 return ret;
352 }
353 EXPORT_SYMBOL(vfs_iter_read);
354
355 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos)
356 {
357 struct kiocb kiocb;
358 ssize_t ret;
359
360 if (!file->f_op->write_iter)
361 return -EINVAL;
362
363 init_sync_kiocb(&kiocb, file);
364 kiocb.ki_pos = *ppos;
365
366 iter->type |= WRITE;
367 ret = file->f_op->write_iter(&kiocb, iter);
368 BUG_ON(ret == -EIOCBQUEUED);
369 if (ret > 0)
370 *ppos = kiocb.ki_pos;
371 return ret;
372 }
373 EXPORT_SYMBOL(vfs_iter_write);
374
375 /*
376 * rw_verify_area doesn't like huge counts. We limit
377 * them to something that fits in "int" so that others
378 * won't have to do range checks all the time.
379 */
380 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
381 {
382 struct inode *inode;
383 loff_t pos;
384 int retval = -EINVAL;
385
386 inode = file_inode(file);
387 if (unlikely((ssize_t) count < 0))
388 return retval;
389 pos = *ppos;
390 if (unlikely(pos < 0)) {
391 if (!unsigned_offsets(file))
392 return retval;
393 if (count >= -pos) /* both values are in 0..LLONG_MAX */
394 return -EOVERFLOW;
395 } else if (unlikely((loff_t) (pos + count) < 0)) {
396 if (!unsigned_offsets(file))
397 return retval;
398 }
399
400 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
401 retval = locks_mandatory_area(
402 read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE,
403 inode, file, pos, count);
404 if (retval < 0)
405 return retval;
406 }
407 retval = security_file_permission(file,
408 read_write == READ ? MAY_READ : MAY_WRITE);
409 if (retval)
410 return retval;
411 return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
412 }
413
414 ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
415 {
416 struct iovec iov = { .iov_base = buf, .iov_len = len };
417 struct kiocb kiocb;
418 ssize_t ret;
419
420 init_sync_kiocb(&kiocb, filp);
421 kiocb.ki_pos = *ppos;
422
423 ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
424 BUG_ON(ret == -EIOCBQUEUED);
425 *ppos = kiocb.ki_pos;
426 return ret;
427 }
428
429 EXPORT_SYMBOL(do_sync_read);
430
431 ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
432 {
433 struct iovec iov = { .iov_base = buf, .iov_len = len };
434 struct kiocb kiocb;
435 struct iov_iter iter;
436 ssize_t ret;
437
438 init_sync_kiocb(&kiocb, filp);
439 kiocb.ki_pos = *ppos;
440 iov_iter_init(&iter, READ, &iov, 1, len);
441
442 ret = filp->f_op->read_iter(&kiocb, &iter);
443 BUG_ON(ret == -EIOCBQUEUED);
444 *ppos = kiocb.ki_pos;
445 return ret;
446 }
447
448 EXPORT_SYMBOL(new_sync_read);
449
450 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
451 loff_t *pos)
452 {
453 ssize_t ret;
454
455 if (file->f_op->read)
456 ret = file->f_op->read(file, buf, count, pos);
457 else if (file->f_op->aio_read)
458 ret = do_sync_read(file, buf, count, pos);
459 else if (file->f_op->read_iter)
460 ret = new_sync_read(file, buf, count, pos);
461 else
462 ret = -EINVAL;
463
464 return ret;
465 }
466
467 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
468 {
469 ssize_t ret;
470
471 if (!(file->f_mode & FMODE_READ))
472 return -EBADF;
473 if (!(file->f_mode & FMODE_CAN_READ))
474 return -EINVAL;
475 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
476 return -EFAULT;
477
478 ret = rw_verify_area(READ, file, pos, count);
479 if (ret >= 0) {
480 count = ret;
481 ret = __vfs_read(file, buf, count, pos);
482 if (ret > 0) {
483 fsnotify_access(file);
484 add_rchar(current, ret);
485 }
486 inc_syscr(current);
487 }
488
489 return ret;
490 }
491
492 EXPORT_SYMBOL(vfs_read);
493
494 ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
495 {
496 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
497 struct kiocb kiocb;
498 ssize_t ret;
499
500 init_sync_kiocb(&kiocb, filp);
501 kiocb.ki_pos = *ppos;
502
503 ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
504 BUG_ON(ret == -EIOCBQUEUED);
505 *ppos = kiocb.ki_pos;
506 return ret;
507 }
508
509 EXPORT_SYMBOL(do_sync_write);
510
511 ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
512 {
513 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
514 struct kiocb kiocb;
515 struct iov_iter iter;
516 ssize_t ret;
517
518 init_sync_kiocb(&kiocb, filp);
519 kiocb.ki_pos = *ppos;
520 iov_iter_init(&iter, WRITE, &iov, 1, len);
521
522 ret = filp->f_op->write_iter(&kiocb, &iter);
523 BUG_ON(ret == -EIOCBQUEUED);
524 *ppos = kiocb.ki_pos;
525 return ret;
526 }
527
528 EXPORT_SYMBOL(new_sync_write);
529
530 ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
531 {
532 mm_segment_t old_fs;
533 const char __user *p;
534 ssize_t ret;
535
536 if (!(file->f_mode & FMODE_CAN_WRITE))
537 return -EINVAL;
538
539 old_fs = get_fs();
540 set_fs(get_ds());
541 p = (__force const char __user *)buf;
542 if (count > MAX_RW_COUNT)
543 count = MAX_RW_COUNT;
544 if (file->f_op->write)
545 ret = file->f_op->write(file, p, count, pos);
546 else if (file->f_op->aio_write)
547 ret = do_sync_write(file, p, count, pos);
548 else
549 ret = new_sync_write(file, p, count, pos);
550 set_fs(old_fs);
551 if (ret > 0) {
552 fsnotify_modify(file);
553 add_wchar(current, ret);
554 }
555 inc_syscw(current);
556 return ret;
557 }
558
559 EXPORT_SYMBOL(__kernel_write);
560
561 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
562 {
563 ssize_t ret;
564
565 if (!(file->f_mode & FMODE_WRITE))
566 return -EBADF;
567 if (!(file->f_mode & FMODE_CAN_WRITE))
568 return -EINVAL;
569 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
570 return -EFAULT;
571
572 ret = rw_verify_area(WRITE, file, pos, count);
573 if (ret >= 0) {
574 count = ret;
575 file_start_write(file);
576 if (file->f_op->write)
577 ret = file->f_op->write(file, buf, count, pos);
578 else if (file->f_op->aio_write)
579 ret = do_sync_write(file, buf, count, pos);
580 else
581 ret = new_sync_write(file, buf, count, pos);
582 if (ret > 0) {
583 fsnotify_modify(file);
584 add_wchar(current, ret);
585 }
586 inc_syscw(current);
587 file_end_write(file);
588 }
589
590 return ret;
591 }
592
593 EXPORT_SYMBOL(vfs_write);
594
595 static inline loff_t file_pos_read(struct file *file)
596 {
597 return file->f_pos;
598 }
599
600 static inline void file_pos_write(struct file *file, loff_t pos)
601 {
602 file->f_pos = pos;
603 }
604
605 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
606 {
607 struct fd f = fdget_pos(fd);
608 ssize_t ret = -EBADF;
609
610 if (f.file) {
611 loff_t pos = file_pos_read(f.file);
612 ret = vfs_read(f.file, buf, count, &pos);
613 if (ret >= 0)
614 file_pos_write(f.file, pos);
615 fdput_pos(f);
616 }
617 return ret;
618 }
619
620 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
621 size_t, count)
622 {
623 struct fd f = fdget_pos(fd);
624 ssize_t ret = -EBADF;
625
626 if (f.file) {
627 loff_t pos = file_pos_read(f.file);
628 ret = vfs_write(f.file, buf, count, &pos);
629 if (ret >= 0)
630 file_pos_write(f.file, pos);
631 fdput_pos(f);
632 }
633
634 return ret;
635 }
636
637 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
638 size_t, count, loff_t, pos)
639 {
640 struct fd f;
641 ssize_t ret = -EBADF;
642
643 if (pos < 0)
644 return -EINVAL;
645
646 f = fdget(fd);
647 if (f.file) {
648 ret = -ESPIPE;
649 if (f.file->f_mode & FMODE_PREAD)
650 ret = vfs_read(f.file, buf, count, &pos);
651 fdput(f);
652 }
653
654 return ret;
655 }
656
657 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
658 size_t, count, loff_t, pos)
659 {
660 struct fd f;
661 ssize_t ret = -EBADF;
662
663 if (pos < 0)
664 return -EINVAL;
665
666 f = fdget(fd);
667 if (f.file) {
668 ret = -ESPIPE;
669 if (f.file->f_mode & FMODE_PWRITE)
670 ret = vfs_write(f.file, buf, count, &pos);
671 fdput(f);
672 }
673
674 return ret;
675 }
676
677 /*
678 * Reduce an iovec's length in-place. Return the resulting number of segments
679 */
680 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
681 {
682 unsigned long seg = 0;
683 size_t len = 0;
684
685 while (seg < nr_segs) {
686 seg++;
687 if (len + iov->iov_len >= to) {
688 iov->iov_len = to - len;
689 break;
690 }
691 len += iov->iov_len;
692 iov++;
693 }
694 return seg;
695 }
696 EXPORT_SYMBOL(iov_shorten);
697
698 static ssize_t do_iter_readv_writev(struct file *filp, int rw, const struct iovec *iov,
699 unsigned long nr_segs, size_t len, loff_t *ppos, iter_fn_t fn)
700 {
701 struct kiocb kiocb;
702 struct iov_iter iter;
703 ssize_t ret;
704
705 init_sync_kiocb(&kiocb, filp);
706 kiocb.ki_pos = *ppos;
707
708 iov_iter_init(&iter, rw, iov, nr_segs, len);
709 ret = fn(&kiocb, &iter);
710 BUG_ON(ret == -EIOCBQUEUED);
711 *ppos = kiocb.ki_pos;
712 return ret;
713 }
714
715 static ssize_t do_sync_readv_writev(struct file *filp, const struct iovec *iov,
716 unsigned long nr_segs, size_t len, loff_t *ppos, iov_fn_t fn)
717 {
718 struct kiocb kiocb;
719 ssize_t ret;
720
721 init_sync_kiocb(&kiocb, filp);
722 kiocb.ki_pos = *ppos;
723
724 ret = fn(&kiocb, iov, nr_segs, kiocb.ki_pos);
725 BUG_ON(ret == -EIOCBQUEUED);
726 *ppos = kiocb.ki_pos;
727 return ret;
728 }
729
730 /* Do it by hand, with file-ops */
731 static ssize_t do_loop_readv_writev(struct file *filp, struct iovec *iov,
732 unsigned long nr_segs, loff_t *ppos, io_fn_t fn)
733 {
734 struct iovec *vector = iov;
735 ssize_t ret = 0;
736
737 while (nr_segs > 0) {
738 void __user *base;
739 size_t len;
740 ssize_t nr;
741
742 base = vector->iov_base;
743 len = vector->iov_len;
744 vector++;
745 nr_segs--;
746
747 nr = fn(filp, base, len, ppos);
748
749 if (nr < 0) {
750 if (!ret)
751 ret = nr;
752 break;
753 }
754 ret += nr;
755 if (nr != len)
756 break;
757 }
758
759 return ret;
760 }
761
762 /* A write operation does a read from user space and vice versa */
763 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
764
765 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
766 unsigned long nr_segs, unsigned long fast_segs,
767 struct iovec *fast_pointer,
768 struct iovec **ret_pointer)
769 {
770 unsigned long seg;
771 ssize_t ret;
772 struct iovec *iov = fast_pointer;
773
774 /*
775 * SuS says "The readv() function *may* fail if the iovcnt argument
776 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
777 * traditionally returned zero for zero segments, so...
778 */
779 if (nr_segs == 0) {
780 ret = 0;
781 goto out;
782 }
783
784 /*
785 * First get the "struct iovec" from user memory and
786 * verify all the pointers
787 */
788 if (nr_segs > UIO_MAXIOV) {
789 ret = -EINVAL;
790 goto out;
791 }
792 if (nr_segs > fast_segs) {
793 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
794 if (iov == NULL) {
795 ret = -ENOMEM;
796 goto out;
797 }
798 }
799 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
800 ret = -EFAULT;
801 goto out;
802 }
803
804 /*
805 * According to the Single Unix Specification we should return EINVAL
806 * if an element length is < 0 when cast to ssize_t or if the
807 * total length would overflow the ssize_t return value of the
808 * system call.
809 *
810 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
811 * overflow case.
812 */
813 ret = 0;
814 for (seg = 0; seg < nr_segs; seg++) {
815 void __user *buf = iov[seg].iov_base;
816 ssize_t len = (ssize_t)iov[seg].iov_len;
817
818 /* see if we we're about to use an invalid len or if
819 * it's about to overflow ssize_t */
820 if (len < 0) {
821 ret = -EINVAL;
822 goto out;
823 }
824 if (type >= 0
825 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
826 ret = -EFAULT;
827 goto out;
828 }
829 if (len > MAX_RW_COUNT - ret) {
830 len = MAX_RW_COUNT - ret;
831 iov[seg].iov_len = len;
832 }
833 ret += len;
834 }
835 out:
836 *ret_pointer = iov;
837 return ret;
838 }
839
840 static ssize_t do_readv_writev(int type, struct file *file,
841 const struct iovec __user * uvector,
842 unsigned long nr_segs, loff_t *pos)
843 {
844 size_t tot_len;
845 struct iovec iovstack[UIO_FASTIOV];
846 struct iovec *iov = iovstack;
847 ssize_t ret;
848 io_fn_t fn;
849 iov_fn_t fnv;
850 iter_fn_t iter_fn;
851
852 ret = rw_copy_check_uvector(type, uvector, nr_segs,
853 ARRAY_SIZE(iovstack), iovstack, &iov);
854 if (ret <= 0)
855 goto out;
856
857 tot_len = ret;
858 ret = rw_verify_area(type, file, pos, tot_len);
859 if (ret < 0)
860 goto out;
861
862 fnv = NULL;
863 if (type == READ) {
864 fn = file->f_op->read;
865 fnv = file->f_op->aio_read;
866 iter_fn = file->f_op->read_iter;
867 } else {
868 fn = (io_fn_t)file->f_op->write;
869 fnv = file->f_op->aio_write;
870 iter_fn = file->f_op->write_iter;
871 file_start_write(file);
872 }
873
874 if (iter_fn)
875 ret = do_iter_readv_writev(file, type, iov, nr_segs, tot_len,
876 pos, iter_fn);
877 else if (fnv)
878 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
879 pos, fnv);
880 else
881 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
882
883 if (type != READ)
884 file_end_write(file);
885
886 out:
887 if (iov != iovstack)
888 kfree(iov);
889 if ((ret + (type == READ)) > 0) {
890 if (type == READ)
891 fsnotify_access(file);
892 else
893 fsnotify_modify(file);
894 }
895 return ret;
896 }
897
898 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
899 unsigned long vlen, loff_t *pos)
900 {
901 if (!(file->f_mode & FMODE_READ))
902 return -EBADF;
903 if (!(file->f_mode & FMODE_CAN_READ))
904 return -EINVAL;
905
906 return do_readv_writev(READ, file, vec, vlen, pos);
907 }
908
909 EXPORT_SYMBOL(vfs_readv);
910
911 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
912 unsigned long vlen, loff_t *pos)
913 {
914 if (!(file->f_mode & FMODE_WRITE))
915 return -EBADF;
916 if (!(file->f_mode & FMODE_CAN_WRITE))
917 return -EINVAL;
918
919 return do_readv_writev(WRITE, file, vec, vlen, pos);
920 }
921
922 EXPORT_SYMBOL(vfs_writev);
923
924 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
925 unsigned long, vlen)
926 {
927 struct fd f = fdget_pos(fd);
928 ssize_t ret = -EBADF;
929
930 if (f.file) {
931 loff_t pos = file_pos_read(f.file);
932 ret = vfs_readv(f.file, vec, vlen, &pos);
933 if (ret >= 0)
934 file_pos_write(f.file, pos);
935 fdput_pos(f);
936 }
937
938 if (ret > 0)
939 add_rchar(current, ret);
940 inc_syscr(current);
941 return ret;
942 }
943
944 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
945 unsigned long, vlen)
946 {
947 struct fd f = fdget_pos(fd);
948 ssize_t ret = -EBADF;
949
950 if (f.file) {
951 loff_t pos = file_pos_read(f.file);
952 ret = vfs_writev(f.file, vec, vlen, &pos);
953 if (ret >= 0)
954 file_pos_write(f.file, pos);
955 fdput_pos(f);
956 }
957
958 if (ret > 0)
959 add_wchar(current, ret);
960 inc_syscw(current);
961 return ret;
962 }
963
964 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
965 {
966 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
967 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
968 }
969
970 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
971 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
972 {
973 loff_t pos = pos_from_hilo(pos_h, pos_l);
974 struct fd f;
975 ssize_t ret = -EBADF;
976
977 if (pos < 0)
978 return -EINVAL;
979
980 f = fdget(fd);
981 if (f.file) {
982 ret = -ESPIPE;
983 if (f.file->f_mode & FMODE_PREAD)
984 ret = vfs_readv(f.file, vec, vlen, &pos);
985 fdput(f);
986 }
987
988 if (ret > 0)
989 add_rchar(current, ret);
990 inc_syscr(current);
991 return ret;
992 }
993
994 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
995 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
996 {
997 loff_t pos = pos_from_hilo(pos_h, pos_l);
998 struct fd f;
999 ssize_t ret = -EBADF;
1000
1001 if (pos < 0)
1002 return -EINVAL;
1003
1004 f = fdget(fd);
1005 if (f.file) {
1006 ret = -ESPIPE;
1007 if (f.file->f_mode & FMODE_PWRITE)
1008 ret = vfs_writev(f.file, vec, vlen, &pos);
1009 fdput(f);
1010 }
1011
1012 if (ret > 0)
1013 add_wchar(current, ret);
1014 inc_syscw(current);
1015 return ret;
1016 }
1017
1018 #ifdef CONFIG_COMPAT
1019
1020 static ssize_t compat_do_readv_writev(int type, struct file *file,
1021 const struct compat_iovec __user *uvector,
1022 unsigned long nr_segs, loff_t *pos)
1023 {
1024 compat_ssize_t tot_len;
1025 struct iovec iovstack[UIO_FASTIOV];
1026 struct iovec *iov = iovstack;
1027 ssize_t ret;
1028 io_fn_t fn;
1029 iov_fn_t fnv;
1030 iter_fn_t iter_fn;
1031
1032 ret = compat_rw_copy_check_uvector(type, uvector, nr_segs,
1033 UIO_FASTIOV, iovstack, &iov);
1034 if (ret <= 0)
1035 goto out;
1036
1037 tot_len = ret;
1038 ret = rw_verify_area(type, file, pos, tot_len);
1039 if (ret < 0)
1040 goto out;
1041
1042 fnv = NULL;
1043 if (type == READ) {
1044 fn = file->f_op->read;
1045 fnv = file->f_op->aio_read;
1046 iter_fn = file->f_op->read_iter;
1047 } else {
1048 fn = (io_fn_t)file->f_op->write;
1049 fnv = file->f_op->aio_write;
1050 iter_fn = file->f_op->write_iter;
1051 file_start_write(file);
1052 }
1053
1054 if (iter_fn)
1055 ret = do_iter_readv_writev(file, type, iov, nr_segs, tot_len,
1056 pos, iter_fn);
1057 else if (fnv)
1058 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
1059 pos, fnv);
1060 else
1061 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
1062
1063 if (type != READ)
1064 file_end_write(file);
1065
1066 out:
1067 if (iov != iovstack)
1068 kfree(iov);
1069 if ((ret + (type == READ)) > 0) {
1070 if (type == READ)
1071 fsnotify_access(file);
1072 else
1073 fsnotify_modify(file);
1074 }
1075 return ret;
1076 }
1077
1078 static size_t compat_readv(struct file *file,
1079 const struct compat_iovec __user *vec,
1080 unsigned long vlen, loff_t *pos)
1081 {
1082 ssize_t ret = -EBADF;
1083
1084 if (!(file->f_mode & FMODE_READ))
1085 goto out;
1086
1087 ret = -EINVAL;
1088 if (!(file->f_mode & FMODE_CAN_READ))
1089 goto out;
1090
1091 ret = compat_do_readv_writev(READ, file, vec, vlen, pos);
1092
1093 out:
1094 if (ret > 0)
1095 add_rchar(current, ret);
1096 inc_syscr(current);
1097 return ret;
1098 }
1099
1100 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1101 const struct compat_iovec __user *,vec,
1102 compat_ulong_t, vlen)
1103 {
1104 struct fd f = fdget_pos(fd);
1105 ssize_t ret;
1106 loff_t pos;
1107
1108 if (!f.file)
1109 return -EBADF;
1110 pos = f.file->f_pos;
1111 ret = compat_readv(f.file, vec, vlen, &pos);
1112 if (ret >= 0)
1113 f.file->f_pos = pos;
1114 fdput_pos(f);
1115 return ret;
1116 }
1117
1118 static long __compat_sys_preadv64(unsigned long fd,
1119 const struct compat_iovec __user *vec,
1120 unsigned long vlen, loff_t pos)
1121 {
1122 struct fd f;
1123 ssize_t ret;
1124
1125 if (pos < 0)
1126 return -EINVAL;
1127 f = fdget(fd);
1128 if (!f.file)
1129 return -EBADF;
1130 ret = -ESPIPE;
1131 if (f.file->f_mode & FMODE_PREAD)
1132 ret = compat_readv(f.file, vec, vlen, &pos);
1133 fdput(f);
1134 return ret;
1135 }
1136
1137 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1138 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1139 const struct compat_iovec __user *,vec,
1140 unsigned long, vlen, loff_t, pos)
1141 {
1142 return __compat_sys_preadv64(fd, vec, vlen, pos);
1143 }
1144 #endif
1145
1146 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1147 const struct compat_iovec __user *,vec,
1148 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1149 {
1150 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1151
1152 return __compat_sys_preadv64(fd, vec, vlen, pos);
1153 }
1154
1155 static size_t compat_writev(struct file *file,
1156 const struct compat_iovec __user *vec,
1157 unsigned long vlen, loff_t *pos)
1158 {
1159 ssize_t ret = -EBADF;
1160
1161 if (!(file->f_mode & FMODE_WRITE))
1162 goto out;
1163
1164 ret = -EINVAL;
1165 if (!(file->f_mode & FMODE_CAN_WRITE))
1166 goto out;
1167
1168 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos);
1169
1170 out:
1171 if (ret > 0)
1172 add_wchar(current, ret);
1173 inc_syscw(current);
1174 return ret;
1175 }
1176
1177 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1178 const struct compat_iovec __user *, vec,
1179 compat_ulong_t, vlen)
1180 {
1181 struct fd f = fdget_pos(fd);
1182 ssize_t ret;
1183 loff_t pos;
1184
1185 if (!f.file)
1186 return -EBADF;
1187 pos = f.file->f_pos;
1188 ret = compat_writev(f.file, vec, vlen, &pos);
1189 if (ret >= 0)
1190 f.file->f_pos = pos;
1191 fdput_pos(f);
1192 return ret;
1193 }
1194
1195 static long __compat_sys_pwritev64(unsigned long fd,
1196 const struct compat_iovec __user *vec,
1197 unsigned long vlen, loff_t pos)
1198 {
1199 struct fd f;
1200 ssize_t ret;
1201
1202 if (pos < 0)
1203 return -EINVAL;
1204 f = fdget(fd);
1205 if (!f.file)
1206 return -EBADF;
1207 ret = -ESPIPE;
1208 if (f.file->f_mode & FMODE_PWRITE)
1209 ret = compat_writev(f.file, vec, vlen, &pos);
1210 fdput(f);
1211 return ret;
1212 }
1213
1214 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1215 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1216 const struct compat_iovec __user *,vec,
1217 unsigned long, vlen, loff_t, pos)
1218 {
1219 return __compat_sys_pwritev64(fd, vec, vlen, pos);
1220 }
1221 #endif
1222
1223 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1224 const struct compat_iovec __user *,vec,
1225 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1226 {
1227 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1228
1229 return __compat_sys_pwritev64(fd, vec, vlen, pos);
1230 }
1231 #endif
1232
1233 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1234 size_t count, loff_t max)
1235 {
1236 struct fd in, out;
1237 struct inode *in_inode, *out_inode;
1238 loff_t pos;
1239 loff_t out_pos;
1240 ssize_t retval;
1241 int fl;
1242
1243 /*
1244 * Get input file, and verify that it is ok..
1245 */
1246 retval = -EBADF;
1247 in = fdget(in_fd);
1248 if (!in.file)
1249 goto out;
1250 if (!(in.file->f_mode & FMODE_READ))
1251 goto fput_in;
1252 retval = -ESPIPE;
1253 if (!ppos) {
1254 pos = in.file->f_pos;
1255 } else {
1256 pos = *ppos;
1257 if (!(in.file->f_mode & FMODE_PREAD))
1258 goto fput_in;
1259 }
1260 retval = rw_verify_area(READ, in.file, &pos, count);
1261 if (retval < 0)
1262 goto fput_in;
1263 count = retval;
1264
1265 /*
1266 * Get output file, and verify that it is ok..
1267 */
1268 retval = -EBADF;
1269 out = fdget(out_fd);
1270 if (!out.file)
1271 goto fput_in;
1272 if (!(out.file->f_mode & FMODE_WRITE))
1273 goto fput_out;
1274 retval = -EINVAL;
1275 in_inode = file_inode(in.file);
1276 out_inode = file_inode(out.file);
1277 out_pos = out.file->f_pos;
1278 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1279 if (retval < 0)
1280 goto fput_out;
1281 count = retval;
1282
1283 if (!max)
1284 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1285
1286 if (unlikely(pos + count > max)) {
1287 retval = -EOVERFLOW;
1288 if (pos >= max)
1289 goto fput_out;
1290 count = max - pos;
1291 }
1292
1293 fl = 0;
1294 #if 0
1295 /*
1296 * We need to debate whether we can enable this or not. The
1297 * man page documents EAGAIN return for the output at least,
1298 * and the application is arguably buggy if it doesn't expect
1299 * EAGAIN on a non-blocking file descriptor.
1300 */
1301 if (in.file->f_flags & O_NONBLOCK)
1302 fl = SPLICE_F_NONBLOCK;
1303 #endif
1304 file_start_write(out.file);
1305 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1306 file_end_write(out.file);
1307
1308 if (retval > 0) {
1309 add_rchar(current, retval);
1310 add_wchar(current, retval);
1311 fsnotify_access(in.file);
1312 fsnotify_modify(out.file);
1313 out.file->f_pos = out_pos;
1314 if (ppos)
1315 *ppos = pos;
1316 else
1317 in.file->f_pos = pos;
1318 }
1319
1320 inc_syscr(current);
1321 inc_syscw(current);
1322 if (pos > max)
1323 retval = -EOVERFLOW;
1324
1325 fput_out:
1326 fdput(out);
1327 fput_in:
1328 fdput(in);
1329 out:
1330 return retval;
1331 }
1332
1333 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, 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 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1353 {
1354 loff_t pos;
1355 ssize_t ret;
1356
1357 if (offset) {
1358 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1359 return -EFAULT;
1360 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1361 if (unlikely(put_user(pos, offset)))
1362 return -EFAULT;
1363 return ret;
1364 }
1365
1366 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1367 }
1368
1369 #ifdef CONFIG_COMPAT
1370 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1371 compat_off_t __user *, offset, compat_size_t, count)
1372 {
1373 loff_t pos;
1374 off_t off;
1375 ssize_t ret;
1376
1377 if (offset) {
1378 if (unlikely(get_user(off, offset)))
1379 return -EFAULT;
1380 pos = off;
1381 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1382 if (unlikely(put_user(pos, offset)))
1383 return -EFAULT;
1384 return ret;
1385 }
1386
1387 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1388 }
1389
1390 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1391 compat_loff_t __user *, offset, compat_size_t, count)
1392 {
1393 loff_t pos;
1394 ssize_t ret;
1395
1396 if (offset) {
1397 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1398 return -EFAULT;
1399 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1400 if (unlikely(put_user(pos, offset)))
1401 return -EFAULT;
1402 return ret;
1403 }
1404
1405 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1406 }
1407 #endif