]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - fs/read_write.c
projects / mirror_ubuntu-zesty-kernel.git / blob
? search:


f9f8bb137806b20a44019073f3ecd0380449a9f3
[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 <linux/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 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
402 {
403 struct inode *inode;
404 loff_t pos;
405 int retval = -EINVAL;
406
407 inode = file_inode(file);
408 if (unlikely((ssize_t) count < 0))
409 return retval;
410 pos = *ppos;
411 if (unlikely(pos < 0)) {
412 if (!unsigned_offsets(file))
413 return retval;
414 if (count >= -pos) /* both values are in 0..LLONG_MAX */
415 return -EOVERFLOW;
416 } else if (unlikely((loff_t) (pos + count) < 0)) {
417 if (!unsigned_offsets(file))
418 return retval;
419 }
420
421 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
422 retval = locks_mandatory_area(inode, file, pos, pos + count - 1,
423 read_write == READ ? F_RDLCK : F_WRLCK);
424 if (retval < 0)
425 return retval;
426 }
427 return security_file_permission(file,
428 read_write == READ ? MAY_READ : MAY_WRITE);
429 }
430
431 static 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 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
449 loff_t *pos)
450 {
451 if (file->f_op->read)
452 return file->f_op->read(file, buf, count, pos);
453 else if (file->f_op->read_iter)
454 return new_sync_read(file, buf, count, pos);
455 else
456 return -EINVAL;
457 }
458 EXPORT_SYMBOL(__vfs_read);
459
460 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
461 {
462 ssize_t ret;
463
464 if (!(file->f_mode & FMODE_READ))
465 return -EBADF;
466 if (!(file->f_mode & FMODE_CAN_READ))
467 return -EINVAL;
468 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
469 return -EFAULT;
470
471 ret = rw_verify_area(READ, file, pos, count);
472 if (!ret) {
473 if (count > MAX_RW_COUNT)
474 count = MAX_RW_COUNT;
475 ret = __vfs_read(file, buf, count, pos);
476 if (ret > 0) {
477 fsnotify_access(file);
478 add_rchar(current, ret);
479 }
480 inc_syscr(current);
481 }
482
483 return ret;
484 }
485
486 EXPORT_SYMBOL(vfs_read);
487
488 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
489 {
490 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
491 struct kiocb kiocb;
492 struct iov_iter iter;
493 ssize_t ret;
494
495 init_sync_kiocb(&kiocb, filp);
496 kiocb.ki_pos = *ppos;
497 iov_iter_init(&iter, WRITE, &iov, 1, len);
498
499 ret = filp->f_op->write_iter(&kiocb, &iter);
500 BUG_ON(ret == -EIOCBQUEUED);
501 if (ret > 0)
502 *ppos = kiocb.ki_pos;
503 return ret;
504 }
505
506 ssize_t __vfs_write(struct file *file, const char __user *p, size_t count,
507 loff_t *pos)
508 {
509 if (file->f_op->write)
510 return file->f_op->write(file, p, count, pos);
511 else if (file->f_op->write_iter)
512 return new_sync_write(file, p, count, pos);
513 else
514 return -EINVAL;
515 }
516 EXPORT_SYMBOL(__vfs_write);
517
518 vfs_readf_t vfs_readf(struct file *file)
519 {
520 const struct file_operations *fop = file->f_op;
521
522 if (fop->read)
523 return fop->read;
524 if (fop->read_iter)
525 return new_sync_read;
526 return ERR_PTR(-ENOSYS);
527 }
528 EXPORT_SYMBOL_GPL(vfs_readf);
529
530 vfs_writef_t vfs_writef(struct file *file)
531 {
532 const struct file_operations *fop = file->f_op;
533
534 if (fop->write)
535 return fop->write;
536 if (fop->write_iter)
537 return new_sync_write;
538 return ERR_PTR(-ENOSYS);
539 }
540 EXPORT_SYMBOL_GPL(vfs_writef);
541
542 ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
543 {
544 mm_segment_t old_fs;
545 const char __user *p;
546 ssize_t ret;
547
548 if (!(file->f_mode & FMODE_CAN_WRITE))
549 return -EINVAL;
550
551 old_fs = get_fs();
552 set_fs(get_ds());
553 p = (__force const char __user *)buf;
554 if (count > MAX_RW_COUNT)
555 count = MAX_RW_COUNT;
556 ret = __vfs_write(file, p, count, pos);
557 set_fs(old_fs);
558 if (ret > 0) {
559 fsnotify_modify(file);
560 add_wchar(current, ret);
561 }
562 inc_syscw(current);
563 return ret;
564 }
565
566 EXPORT_SYMBOL(__kernel_write);
567
568 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
569 {
570 ssize_t ret;
571
572 if (!(file->f_mode & FMODE_WRITE))
573 return -EBADF;
574 if (!(file->f_mode & FMODE_CAN_WRITE))
575 return -EINVAL;
576 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
577 return -EFAULT;
578
579 ret = rw_verify_area(WRITE, file, pos, count);
580 if (!ret) {
581 if (count > MAX_RW_COUNT)
582 count = MAX_RW_COUNT;
583 file_start_write(file);
584 ret = __vfs_write(file, buf, count, pos);
585 if (ret > 0) {
586 fsnotify_modify(file);
587 add_wchar(current, ret);
588 }
589 inc_syscw(current);
590 file_end_write(file);
591 }
592
593 return ret;
594 }
595
596 EXPORT_SYMBOL(vfs_write);
597
598 static inline loff_t file_pos_read(struct file *file)
599 {
600 return file->f_pos;
601 }
602
603 static inline void file_pos_write(struct file *file, loff_t pos)
604 {
605 file->f_pos = pos;
606 }
607
608 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
609 {
610 struct fd f = fdget_pos(fd);
611 ssize_t ret = -EBADF;
612
613 if (f.file) {
614 loff_t pos = file_pos_read(f.file);
615 ret = vfs_read(f.file, buf, count, &pos);
616 if (ret >= 0)
617 file_pos_write(f.file, pos);
618 fdput_pos(f);
619 }
620 return ret;
621 }
622
623 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
624 size_t, count)
625 {
626 struct fd f = fdget_pos(fd);
627 ssize_t ret = -EBADF;
628
629 if (f.file) {
630 loff_t pos = file_pos_read(f.file);
631 ret = vfs_write(f.file, buf, count, &pos);
632 if (ret >= 0)
633 file_pos_write(f.file, pos);
634 fdput_pos(f);
635 }
636
637 return ret;
638 }
639
640 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
641 size_t, count, loff_t, pos)
642 {
643 struct fd f;
644 ssize_t ret = -EBADF;
645
646 if (pos < 0)
647 return -EINVAL;
648
649 f = fdget(fd);
650 if (f.file) {
651 ret = -ESPIPE;
652 if (f.file->f_mode & FMODE_PREAD)
653 ret = vfs_read(f.file, buf, count, &pos);
654 fdput(f);
655 }
656
657 return ret;
658 }
659
660 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
661 size_t, count, loff_t, pos)
662 {
663 struct fd f;
664 ssize_t ret = -EBADF;
665
666 if (pos < 0)
667 return -EINVAL;
668
669 f = fdget(fd);
670 if (f.file) {
671 ret = -ESPIPE;
672 if (f.file->f_mode & FMODE_PWRITE)
673 ret = vfs_write(f.file, buf, count, &pos);
674 fdput(f);
675 }
676
677 return ret;
678 }
679
680 /*
681 * Reduce an iovec's length in-place. Return the resulting number of segments
682 */
683 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
684 {
685 unsigned long seg = 0;
686 size_t len = 0;
687
688 while (seg < nr_segs) {
689 seg++;
690 if (len + iov->iov_len >= to) {
691 iov->iov_len = to - len;
692 break;
693 }
694 len += iov->iov_len;
695 iov++;
696 }
697 return seg;
698 }
699 EXPORT_SYMBOL(iov_shorten);
700
701 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
702 loff_t *ppos, iter_fn_t fn, int flags)
703 {
704 struct kiocb kiocb;
705 ssize_t ret;
706
707 if (flags & ~(RWF_HIPRI | RWF_DSYNC | RWF_SYNC))
708 return -EOPNOTSUPP;
709
710 init_sync_kiocb(&kiocb, filp);
711 if (flags & RWF_HIPRI)
712 kiocb.ki_flags |= IOCB_HIPRI;
713 if (flags & RWF_DSYNC)
714 kiocb.ki_flags |= IOCB_DSYNC;
715 if (flags & RWF_SYNC)
716 kiocb.ki_flags |= (IOCB_DSYNC | IOCB_SYNC);
717 kiocb.ki_pos = *ppos;
718
719 ret = fn(&kiocb, iter);
720 BUG_ON(ret == -EIOCBQUEUED);
721 *ppos = kiocb.ki_pos;
722 return ret;
723 }
724
725 /* Do it by hand, with file-ops */
726 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
727 loff_t *ppos, io_fn_t fn, int flags)
728 {
729 ssize_t ret = 0;
730
731 if (flags & ~RWF_HIPRI)
732 return -EOPNOTSUPP;
733
734 while (iov_iter_count(iter)) {
735 struct iovec iovec = iov_iter_iovec(iter);
736 ssize_t nr;
737
738 nr = fn(filp, iovec.iov_base, iovec.iov_len, ppos);
739
740 if (nr < 0) {
741 if (!ret)
742 ret = nr;
743 break;
744 }
745 ret += nr;
746 if (nr != iovec.iov_len)
747 break;
748 iov_iter_advance(iter, nr);
749 }
750
751 return ret;
752 }
753
754 /* A write operation does a read from user space and vice versa */
755 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
756
757 /**
758 * rw_copy_check_uvector() - Copy an array of &struct iovec from userspace
759 * into the kernel and check that it is valid.
760 *
761 * @type: One of %CHECK_IOVEC_ONLY, %READ, or %WRITE.
762 * @uvector: Pointer to the userspace array.
763 * @nr_segs: Number of elements in userspace array.
764 * @fast_segs: Number of elements in @fast_pointer.
765 * @fast_pointer: Pointer to (usually small on-stack) kernel array.
766 * @ret_pointer: (output parameter) Pointer to a variable that will point to
767 * either @fast_pointer, a newly allocated kernel array, or NULL,
768 * depending on which array was used.
769 *
770 * This function copies an array of &struct iovec of @nr_segs from
771 * userspace into the kernel and checks that each element is valid (e.g.
772 * it does not point to a kernel address or cause overflow by being too
773 * large, etc.).
774 *
775 * As an optimization, the caller may provide a pointer to a small
776 * on-stack array in @fast_pointer, typically %UIO_FASTIOV elements long
777 * (the size of this array, or 0 if unused, should be given in @fast_segs).
778 *
779 * @ret_pointer will always point to the array that was used, so the
780 * caller must take care not to call kfree() on it e.g. in case the
781 * @fast_pointer array was used and it was allocated on the stack.
782 *
783 * Return: The total number of bytes covered by the iovec array on success
784 * or a negative error code on error.
785 */
786 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
787 unsigned long nr_segs, unsigned long fast_segs,
788 struct iovec *fast_pointer,
789 struct iovec **ret_pointer)
790 {
791 unsigned long seg;
792 ssize_t ret;
793 struct iovec *iov = fast_pointer;
794
795 /*
796 * SuS says "The readv() function *may* fail if the iovcnt argument
797 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
798 * traditionally returned zero for zero segments, so...
799 */
800 if (nr_segs == 0) {
801 ret = 0;
802 goto out;
803 }
804
805 /*
806 * First get the "struct iovec" from user memory and
807 * verify all the pointers
808 */
809 if (nr_segs > UIO_MAXIOV) {
810 ret = -EINVAL;
811 goto out;
812 }
813 if (nr_segs > fast_segs) {
814 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
815 if (iov == NULL) {
816 ret = -ENOMEM;
817 goto out;
818 }
819 }
820 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
821 ret = -EFAULT;
822 goto out;
823 }
824
825 /*
826 * According to the Single Unix Specification we should return EINVAL
827 * if an element length is < 0 when cast to ssize_t or if the
828 * total length would overflow the ssize_t return value of the
829 * system call.
830 *
831 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
832 * overflow case.
833 */
834 ret = 0;
835 for (seg = 0; seg < nr_segs; seg++) {
836 void __user *buf = iov[seg].iov_base;
837 ssize_t len = (ssize_t)iov[seg].iov_len;
838
839 /* see if we we're about to use an invalid len or if
840 * it's about to overflow ssize_t */
841 if (len < 0) {
842 ret = -EINVAL;
843 goto out;
844 }
845 if (type >= 0
846 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
847 ret = -EFAULT;
848 goto out;
849 }
850 if (len > MAX_RW_COUNT - ret) {
851 len = MAX_RW_COUNT - ret;
852 iov[seg].iov_len = len;
853 }
854 ret += len;
855 }
856 out:
857 *ret_pointer = iov;
858 return ret;
859 }
860
861 static ssize_t do_readv_writev(int type, struct file *file,
862 const struct iovec __user * uvector,
863 unsigned long nr_segs, loff_t *pos,
864 int flags)
865 {
866 size_t tot_len;
867 struct iovec iovstack[UIO_FASTIOV];
868 struct iovec *iov = iovstack;
869 struct iov_iter iter;
870 ssize_t ret;
871 io_fn_t fn;
872 iter_fn_t iter_fn;
873
874 ret = import_iovec(type, uvector, nr_segs,
875 ARRAY_SIZE(iovstack), &iov, &iter);
876 if (ret < 0)
877 return ret;
878
879 tot_len = iov_iter_count(&iter);
880 if (!tot_len)
881 goto out;
882 ret = rw_verify_area(type, file, pos, tot_len);
883 if (ret < 0)
884 goto out;
885
886 if (type == READ) {
887 fn = file->f_op->read;
888 iter_fn = file->f_op->read_iter;
889 } else {
890 fn = (io_fn_t)file->f_op->write;
891 iter_fn = file->f_op->write_iter;
892 file_start_write(file);
893 }
894
895 if (iter_fn)
896 ret = do_iter_readv_writev(file, &iter, pos, iter_fn, flags);
897 else
898 ret = do_loop_readv_writev(file, &iter, pos, fn, flags);
899
900 if (type != READ)
901 file_end_write(file);
902
903 out:
904 kfree(iov);
905 if ((ret + (type == READ)) > 0) {
906 if (type == READ)
907 fsnotify_access(file);
908 else
909 fsnotify_modify(file);
910 }
911 return ret;
912 }
913
914 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
915 unsigned long vlen, loff_t *pos, int flags)
916 {
917 if (!(file->f_mode & FMODE_READ))
918 return -EBADF;
919 if (!(file->f_mode & FMODE_CAN_READ))
920 return -EINVAL;
921
922 return do_readv_writev(READ, file, vec, vlen, pos, flags);
923 }
924
925 EXPORT_SYMBOL(vfs_readv);
926
927 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
928 unsigned long vlen, loff_t *pos, int flags)
929 {
930 if (!(file->f_mode & FMODE_WRITE))
931 return -EBADF;
932 if (!(file->f_mode & FMODE_CAN_WRITE))
933 return -EINVAL;
934
935 return do_readv_writev(WRITE, file, vec, vlen, pos, flags);
936 }
937
938 EXPORT_SYMBOL(vfs_writev);
939
940 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
941 unsigned long vlen, int flags)
942 {
943 struct fd f = fdget_pos(fd);
944 ssize_t ret = -EBADF;
945
946 if (f.file) {
947 loff_t pos = file_pos_read(f.file);
948 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
949 if (ret >= 0)
950 file_pos_write(f.file, pos);
951 fdput_pos(f);
952 }
953
954 if (ret > 0)
955 add_rchar(current, ret);
956 inc_syscr(current);
957 return ret;
958 }
959
960 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
961 unsigned long vlen, int flags)
962 {
963 struct fd f = fdget_pos(fd);
964 ssize_t ret = -EBADF;
965
966 if (f.file) {
967 loff_t pos = file_pos_read(f.file);
968 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
969 if (ret >= 0)
970 file_pos_write(f.file, pos);
971 fdput_pos(f);
972 }
973
974 if (ret > 0)
975 add_wchar(current, ret);
976 inc_syscw(current);
977 return ret;
978 }
979
980 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
981 {
982 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
983 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
984 }
985
986 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
987 unsigned long vlen, loff_t pos, int flags)
988 {
989 struct fd f;
990 ssize_t ret = -EBADF;
991
992 if (pos < 0)
993 return -EINVAL;
994
995 f = fdget(fd);
996 if (f.file) {
997 ret = -ESPIPE;
998 if (f.file->f_mode & FMODE_PREAD)
999 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1000 fdput(f);
1001 }
1002
1003 if (ret > 0)
1004 add_rchar(current, ret);
1005 inc_syscr(current);
1006 return ret;
1007 }
1008
1009 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
1010 unsigned long vlen, loff_t pos, int flags)
1011 {
1012 struct fd f;
1013 ssize_t ret = -EBADF;
1014
1015 if (pos < 0)
1016 return -EINVAL;
1017
1018 f = fdget(fd);
1019 if (f.file) {
1020 ret = -ESPIPE;
1021 if (f.file->f_mode & FMODE_PWRITE)
1022 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1023 fdput(f);
1024 }
1025
1026 if (ret > 0)
1027 add_wchar(current, ret);
1028 inc_syscw(current);
1029 return ret;
1030 }
1031
1032 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1033 unsigned long, vlen)
1034 {
1035 return do_readv(fd, vec, vlen, 0);
1036 }
1037
1038 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1039 unsigned long, vlen)
1040 {
1041 return do_writev(fd, vec, vlen, 0);
1042 }
1043
1044 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1045 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1046 {
1047 loff_t pos = pos_from_hilo(pos_h, pos_l);
1048
1049 return do_preadv(fd, vec, vlen, pos, 0);
1050 }
1051
1052 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1053 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1054 int, flags)
1055 {
1056 loff_t pos = pos_from_hilo(pos_h, pos_l);
1057
1058 if (pos == -1)
1059 return do_readv(fd, vec, vlen, flags);
1060
1061 return do_preadv(fd, vec, vlen, pos, flags);
1062 }
1063
1064 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1065 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1066 {
1067 loff_t pos = pos_from_hilo(pos_h, pos_l);
1068
1069 return do_pwritev(fd, vec, vlen, pos, 0);
1070 }
1071
1072 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1073 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1074 int, flags)
1075 {
1076 loff_t pos = pos_from_hilo(pos_h, pos_l);
1077
1078 if (pos == -1)
1079 return do_writev(fd, vec, vlen, flags);
1080
1081 return do_pwritev(fd, vec, vlen, pos, flags);
1082 }
1083
1084 #ifdef CONFIG_COMPAT
1085
1086 static ssize_t compat_do_readv_writev(int type, struct file *file,
1087 const struct compat_iovec __user *uvector,
1088 unsigned long nr_segs, loff_t *pos,
1089 int flags)
1090 {
1091 compat_ssize_t tot_len;
1092 struct iovec iovstack[UIO_FASTIOV];
1093 struct iovec *iov = iovstack;
1094 struct iov_iter iter;
1095 ssize_t ret;
1096 io_fn_t fn;
1097 iter_fn_t iter_fn;
1098
1099 ret = compat_import_iovec(type, uvector, nr_segs,
1100 UIO_FASTIOV, &iov, &iter);
1101 if (ret < 0)
1102 return ret;
1103
1104 tot_len = iov_iter_count(&iter);
1105 if (!tot_len)
1106 goto out;
1107 ret = rw_verify_area(type, file, pos, tot_len);
1108 if (ret < 0)
1109 goto out;
1110
1111 if (type == READ) {
1112 fn = file->f_op->read;
1113 iter_fn = file->f_op->read_iter;
1114 } else {
1115 fn = (io_fn_t)file->f_op->write;
1116 iter_fn = file->f_op->write_iter;
1117 file_start_write(file);
1118 }
1119
1120 if (iter_fn)
1121 ret = do_iter_readv_writev(file, &iter, pos, iter_fn, flags);
1122 else
1123 ret = do_loop_readv_writev(file, &iter, pos, fn, flags);
1124
1125 if (type != READ)
1126 file_end_write(file);
1127
1128 out:
1129 kfree(iov);
1130 if ((ret + (type == READ)) > 0) {
1131 if (type == READ)
1132 fsnotify_access(file);
1133 else
1134 fsnotify_modify(file);
1135 }
1136 return ret;
1137 }
1138
1139 static size_t compat_readv(struct file *file,
1140 const struct compat_iovec __user *vec,
1141 unsigned long vlen, loff_t *pos, int flags)
1142 {
1143 ssize_t ret = -EBADF;
1144
1145 if (!(file->f_mode & FMODE_READ))
1146 goto out;
1147
1148 ret = -EINVAL;
1149 if (!(file->f_mode & FMODE_CAN_READ))
1150 goto out;
1151
1152 ret = compat_do_readv_writev(READ, file, vec, vlen, pos, flags);
1153
1154 out:
1155 if (ret > 0)
1156 add_rchar(current, ret);
1157 inc_syscr(current);
1158 return ret;
1159 }
1160
1161 static size_t do_compat_readv(compat_ulong_t fd,
1162 const struct compat_iovec __user *vec,
1163 compat_ulong_t vlen, int flags)
1164 {
1165 struct fd f = fdget_pos(fd);
1166 ssize_t ret;
1167 loff_t pos;
1168
1169 if (!f.file)
1170 return -EBADF;
1171 pos = f.file->f_pos;
1172 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1173 if (ret >= 0)
1174 f.file->f_pos = pos;
1175 fdput_pos(f);
1176 return ret;
1177
1178 }
1179
1180 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1181 const struct compat_iovec __user *,vec,
1182 compat_ulong_t, vlen)
1183 {
1184 return do_compat_readv(fd, vec, vlen, 0);
1185 }
1186
1187 static long do_compat_preadv64(unsigned long fd,
1188 const struct compat_iovec __user *vec,
1189 unsigned long vlen, loff_t pos, int flags)
1190 {
1191 struct fd f;
1192 ssize_t ret;
1193
1194 if (pos < 0)
1195 return -EINVAL;
1196 f = fdget(fd);
1197 if (!f.file)
1198 return -EBADF;
1199 ret = -ESPIPE;
1200 if (f.file->f_mode & FMODE_PREAD)
1201 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1202 fdput(f);
1203 return ret;
1204 }
1205
1206 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1207 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1208 const struct compat_iovec __user *,vec,
1209 unsigned long, vlen, loff_t, pos)
1210 {
1211 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1212 }
1213 #endif
1214
1215 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1216 const struct compat_iovec __user *,vec,
1217 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1218 {
1219 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1220
1221 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1222 }
1223
1224 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1225 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1226 const struct compat_iovec __user *,vec,
1227 unsigned long, vlen, loff_t, pos, int, flags)
1228 {
1229 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1230 }
1231 #endif
1232
1233 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1234 const struct compat_iovec __user *,vec,
1235 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1236 int, flags)
1237 {
1238 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1239
1240 if (pos == -1)
1241 return do_compat_readv(fd, vec, vlen, flags);
1242
1243 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1244 }
1245
1246 static size_t compat_writev(struct file *file,
1247 const struct compat_iovec __user *vec,
1248 unsigned long vlen, loff_t *pos, int flags)
1249 {
1250 ssize_t ret = -EBADF;
1251
1252 if (!(file->f_mode & FMODE_WRITE))
1253 goto out;
1254
1255 ret = -EINVAL;
1256 if (!(file->f_mode & FMODE_CAN_WRITE))
1257 goto out;
1258
1259 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos, 0);
1260
1261 out:
1262 if (ret > 0)
1263 add_wchar(current, ret);
1264 inc_syscw(current);
1265 return ret;
1266 }
1267
1268 static size_t do_compat_writev(compat_ulong_t fd,
1269 const struct compat_iovec __user* vec,
1270 compat_ulong_t vlen, int flags)
1271 {
1272 struct fd f = fdget_pos(fd);
1273 ssize_t ret;
1274 loff_t pos;
1275
1276 if (!f.file)
1277 return -EBADF;
1278 pos = f.file->f_pos;
1279 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1280 if (ret >= 0)
1281 f.file->f_pos = pos;
1282 fdput_pos(f);
1283 return ret;
1284 }
1285
1286 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1287 const struct compat_iovec __user *, vec,
1288 compat_ulong_t, vlen)
1289 {
1290 return do_compat_writev(fd, vec, vlen, 0);
1291 }
1292
1293 static long do_compat_pwritev64(unsigned long fd,
1294 const struct compat_iovec __user *vec,
1295 unsigned long vlen, loff_t pos, int flags)
1296 {
1297 struct fd f;
1298 ssize_t ret;
1299
1300 if (pos < 0)
1301 return -EINVAL;
1302 f = fdget(fd);
1303 if (!f.file)
1304 return -EBADF;
1305 ret = -ESPIPE;
1306 if (f.file->f_mode & FMODE_PWRITE)
1307 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1308 fdput(f);
1309 return ret;
1310 }
1311
1312 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1313 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1314 const struct compat_iovec __user *,vec,
1315 unsigned long, vlen, loff_t, pos)
1316 {
1317 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1318 }
1319 #endif
1320
1321 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1322 const struct compat_iovec __user *,vec,
1323 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1324 {
1325 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1326
1327 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1328 }
1329
1330 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1331 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1332 const struct compat_iovec __user *,vec,
1333 unsigned long, vlen, loff_t, pos, int, flags)
1334 {
1335 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1336 }
1337 #endif
1338
1339 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1340 const struct compat_iovec __user *,vec,
1341 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, int, flags)
1342 {
1343 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1344
1345 if (pos == -1)
1346 return do_compat_writev(fd, vec, vlen, flags);
1347
1348 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1349 }
1350
1351 #endif
1352
1353 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1354 size_t count, loff_t max)
1355 {
1356 struct fd in, out;
1357 struct inode *in_inode, *out_inode;
1358 loff_t pos;
1359 loff_t out_pos;
1360 ssize_t retval;
1361 int fl;
1362
1363 /*
1364 * Get input file, and verify that it is ok..
1365 */
1366 retval = -EBADF;
1367 in = fdget(in_fd);
1368 if (!in.file)
1369 goto out;
1370 if (!(in.file->f_mode & FMODE_READ))
1371 goto fput_in;
1372 retval = -ESPIPE;
1373 if (!ppos) {
1374 pos = in.file->f_pos;
1375 } else {
1376 pos = *ppos;
1377 if (!(in.file->f_mode & FMODE_PREAD))
1378 goto fput_in;
1379 }
1380 retval = rw_verify_area(READ, in.file, &pos, count);
1381 if (retval < 0)
1382 goto fput_in;
1383 if (count > MAX_RW_COUNT)
1384 count = MAX_RW_COUNT;
1385
1386 /*
1387 * Get output file, and verify that it is ok..
1388 */
1389 retval = -EBADF;
1390 out = fdget(out_fd);
1391 if (!out.file)
1392 goto fput_in;
1393 if (!(out.file->f_mode & FMODE_WRITE))
1394 goto fput_out;
1395 retval = -EINVAL;
1396 in_inode = file_inode(in.file);
1397 out_inode = file_inode(out.file);
1398 out_pos = out.file->f_pos;
1399 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1400 if (retval < 0)
1401 goto fput_out;
1402
1403 if (!max)
1404 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1405
1406 if (unlikely(pos + count > max)) {
1407 retval = -EOVERFLOW;
1408 if (pos >= max)
1409 goto fput_out;
1410 count = max - pos;
1411 }
1412
1413 fl = 0;
1414 #if 0
1415 /*
1416 * We need to debate whether we can enable this or not. The
1417 * man page documents EAGAIN return for the output at least,
1418 * and the application is arguably buggy if it doesn't expect
1419 * EAGAIN on a non-blocking file descriptor.
1420 */
1421 if (in.file->f_flags & O_NONBLOCK)
1422 fl = SPLICE_F_NONBLOCK;
1423 #endif
1424 file_start_write(out.file);
1425 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1426 file_end_write(out.file);
1427
1428 if (retval > 0) {
1429 add_rchar(current, retval);
1430 add_wchar(current, retval);
1431 fsnotify_access(in.file);
1432 fsnotify_modify(out.file);
1433 out.file->f_pos = out_pos;
1434 if (ppos)
1435 *ppos = pos;
1436 else
1437 in.file->f_pos = pos;
1438 }
1439
1440 inc_syscr(current);
1441 inc_syscw(current);
1442 if (pos > max)
1443 retval = -EOVERFLOW;
1444
1445 fput_out:
1446 fdput(out);
1447 fput_in:
1448 fdput(in);
1449 out:
1450 return retval;
1451 }
1452
1453 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1454 {
1455 loff_t pos;
1456 off_t off;
1457 ssize_t ret;
1458
1459 if (offset) {
1460 if (unlikely(get_user(off, offset)))
1461 return -EFAULT;
1462 pos = off;
1463 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1464 if (unlikely(put_user(pos, offset)))
1465 return -EFAULT;
1466 return ret;
1467 }
1468
1469 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1470 }
1471
1472 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1473 {
1474 loff_t pos;
1475 ssize_t ret;
1476
1477 if (offset) {
1478 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1479 return -EFAULT;
1480 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1481 if (unlikely(put_user(pos, offset)))
1482 return -EFAULT;
1483 return ret;
1484 }
1485
1486 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1487 }
1488
1489 #ifdef CONFIG_COMPAT
1490 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1491 compat_off_t __user *, offset, compat_size_t, count)
1492 {
1493 loff_t pos;
1494 off_t off;
1495 ssize_t ret;
1496
1497 if (offset) {
1498 if (unlikely(get_user(off, offset)))
1499 return -EFAULT;
1500 pos = off;
1501 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1502 if (unlikely(put_user(pos, offset)))
1503 return -EFAULT;
1504 return ret;
1505 }
1506
1507 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1508 }
1509
1510 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1511 compat_loff_t __user *, offset, compat_size_t, count)
1512 {
1513 loff_t pos;
1514 ssize_t ret;
1515
1516 if (offset) {
1517 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1518 return -EFAULT;
1519 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1520 if (unlikely(put_user(pos, offset)))
1521 return -EFAULT;
1522 return ret;
1523 }
1524
1525 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1526 }
1527 #endif
1528
1529 /*
1530 * copy_file_range() differs from regular file read and write in that it
1531 * specifically allows return partial success. When it does so is up to
1532 * the copy_file_range method.
1533 */
1534 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1535 struct file *file_out, loff_t pos_out,
1536 size_t len, unsigned int flags)
1537 {
1538 struct inode *inode_in = file_inode(file_in);
1539 struct inode *inode_out = file_inode(file_out);
1540 ssize_t ret;
1541
1542 if (flags != 0)
1543 return -EINVAL;
1544
1545 ret = rw_verify_area(READ, file_in, &pos_in, len);
1546 if (unlikely(ret))
1547 return ret;
1548
1549 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1550 if (unlikely(ret))
1551 return ret;
1552
1553 if (!(file_in->f_mode & FMODE_READ) ||
1554 !(file_out->f_mode & FMODE_WRITE) ||
1555 (file_out->f_flags & O_APPEND))
1556 return -EBADF;
1557
1558 /* this could be relaxed once a method supports cross-fs copies */
1559 if (inode_in->i_sb != inode_out->i_sb)
1560 return -EXDEV;
1561
1562 if (len == 0)
1563 return 0;
1564
1565 sb_start_write(inode_out->i_sb);
1566
1567 /*
1568 * Try cloning first, this is supported by more file systems, and
1569 * more efficient if both clone and copy are supported (e.g. NFS).
1570 */
1571 if (file_in->f_op->clone_file_range) {
1572 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1573 file_out, pos_out, len);
1574 if (ret == 0) {
1575 ret = len;
1576 goto done;
1577 }
1578 }
1579
1580 if (file_out->f_op->copy_file_range) {
1581 ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out,
1582 pos_out, len, flags);
1583 if (ret != -EOPNOTSUPP)
1584 goto done;
1585 }
1586
1587 ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1588 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1589
1590 done:
1591 if (ret > 0) {
1592 fsnotify_access(file_in);
1593 add_rchar(current, ret);
1594 fsnotify_modify(file_out);
1595 add_wchar(current, ret);
1596 }
1597
1598 inc_syscr(current);
1599 inc_syscw(current);
1600
1601 sb_end_write(inode_out->i_sb);
1602
1603 return ret;
1604 }
1605 EXPORT_SYMBOL(vfs_copy_file_range);
1606
1607 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1608 int, fd_out, loff_t __user *, off_out,
1609 size_t, len, unsigned int, flags)
1610 {
1611 loff_t pos_in;
1612 loff_t pos_out;
1613 struct fd f_in;
1614 struct fd f_out;
1615 ssize_t ret = -EBADF;
1616
1617 f_in = fdget(fd_in);
1618 if (!f_in.file)
1619 goto out2;
1620
1621 f_out = fdget(fd_out);
1622 if (!f_out.file)
1623 goto out1;
1624
1625 ret = -EFAULT;
1626 if (off_in) {
1627 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1628 goto out;
1629 } else {
1630 pos_in = f_in.file->f_pos;
1631 }
1632
1633 if (off_out) {
1634 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1635 goto out;
1636 } else {
1637 pos_out = f_out.file->f_pos;
1638 }
1639
1640 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1641 flags);
1642 if (ret > 0) {
1643 pos_in += ret;
1644 pos_out += ret;
1645
1646 if (off_in) {
1647 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1648 ret = -EFAULT;
1649 } else {
1650 f_in.file->f_pos = pos_in;
1651 }
1652
1653 if (off_out) {
1654 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1655 ret = -EFAULT;
1656 } else {
1657 f_out.file->f_pos = pos_out;
1658 }
1659 }
1660
1661 out:
1662 fdput(f_out);
1663 out1:
1664 fdput(f_in);
1665 out2:
1666 return ret;
1667 }
1668
1669 static int clone_verify_area(struct file *file, loff_t pos, u64 len, bool write)
1670 {
1671 struct inode *inode = file_inode(file);
1672
1673 if (unlikely(pos < 0))
1674 return -EINVAL;
1675
1676 if (unlikely((loff_t) (pos + len) < 0))
1677 return -EINVAL;
1678
1679 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1680 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1681 int retval;
1682
1683 retval = locks_mandatory_area(inode, file, pos, end,
1684 write ? F_WRLCK : F_RDLCK);
1685 if (retval < 0)
1686 return retval;
1687 }
1688
1689 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1690 }
1691
1692 /*
1693 * Check that the two inodes are eligible for cloning, the ranges make
1694 * sense, and then flush all dirty data. Caller must ensure that the
1695 * inodes have been locked against any other modifications.
1696 *
1697 * Returns: 0 for "nothing to clone", 1 for "something to clone", or
1698 * the usual negative error code.
1699 */
1700 int vfs_clone_file_prep_inodes(struct inode *inode_in, loff_t pos_in,
1701 struct inode *inode_out, loff_t pos_out,
1702 u64 *len, bool is_dedupe)
1703 {
1704 loff_t bs = inode_out->i_sb->s_blocksize;
1705 loff_t blen;
1706 loff_t isize;
1707 bool same_inode = (inode_in == inode_out);
1708 int ret;
1709
1710 /* Don't touch certain kinds of inodes */
1711 if (IS_IMMUTABLE(inode_out))
1712 return -EPERM;
1713
1714 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1715 return -ETXTBSY;
1716
1717 /* Don't reflink dirs, pipes, sockets... */
1718 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1719 return -EISDIR;
1720 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1721 return -EINVAL;
1722
1723 /* Are we going all the way to the end? */
1724 isize = i_size_read(inode_in);
1725 if (isize == 0)
1726 return 0;
1727
1728 /* Zero length dedupe exits immediately; reflink goes to EOF. */
1729 if (*len == 0) {
1730 if (is_dedupe || pos_in == isize)
1731 return 0;
1732 if (pos_in > isize)
1733 return -EINVAL;
1734 *len = isize - pos_in;
1735 }
1736
1737 /* Ensure offsets don't wrap and the input is inside i_size */
1738 if (pos_in + *len < pos_in || pos_out + *len < pos_out ||
1739 pos_in + *len > isize)
1740 return -EINVAL;
1741
1742 /* Don't allow dedupe past EOF in the dest file */
1743 if (is_dedupe) {
1744 loff_t disize;
1745
1746 disize = i_size_read(inode_out);
1747 if (pos_out >= disize || pos_out + *len > disize)
1748 return -EINVAL;
1749 }
1750
1751 /* If we're linking to EOF, continue to the block boundary. */
1752 if (pos_in + *len == isize)
1753 blen = ALIGN(isize, bs) - pos_in;
1754 else
1755 blen = *len;
1756
1757 /* Only reflink if we're aligned to block boundaries */
1758 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_in + blen, bs) ||
1759 !IS_ALIGNED(pos_out, bs) || !IS_ALIGNED(pos_out + blen, bs))
1760 return -EINVAL;
1761
1762 /* Don't allow overlapped reflink within the same file */
1763 if (same_inode) {
1764 if (pos_out + blen > pos_in && pos_out < pos_in + blen)
1765 return -EINVAL;
1766 }
1767
1768 /* Wait for the completion of any pending IOs on both files */
1769 inode_dio_wait(inode_in);
1770 if (!same_inode)
1771 inode_dio_wait(inode_out);
1772
1773 ret = filemap_write_and_wait_range(inode_in->i_mapping,
1774 pos_in, pos_in + *len - 1);
1775 if (ret)
1776 return ret;
1777
1778 ret = filemap_write_and_wait_range(inode_out->i_mapping,
1779 pos_out, pos_out + *len - 1);
1780 if (ret)
1781 return ret;
1782
1783 /*
1784 * Check that the extents are the same.
1785 */
1786 if (is_dedupe) {
1787 bool is_same = false;
1788
1789 ret = vfs_dedupe_file_range_compare(inode_in, pos_in,
1790 inode_out, pos_out, *len, &is_same);
1791 if (ret)
1792 return ret;
1793 if (!is_same)
1794 return -EBADE;
1795 }
1796
1797 return 1;
1798 }
1799 EXPORT_SYMBOL(vfs_clone_file_prep_inodes);
1800
1801 int vfs_clone_file_range(struct file *file_in, loff_t pos_in,
1802 struct file *file_out, loff_t pos_out, u64 len)
1803 {
1804 struct inode *inode_in = file_inode(file_in);
1805 struct inode *inode_out = file_inode(file_out);
1806 int ret;
1807
1808 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1809 return -EISDIR;
1810 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1811 return -EINVAL;
1812
1813 /*
1814 * FICLONE/FICLONERANGE ioctls enforce that src and dest files are on
1815 * the same mount. Practically, they only need to be on the same file
1816 * system.
1817 */
1818 if (inode_in->i_sb != inode_out->i_sb)
1819 return -EXDEV;
1820
1821 if (!(file_in->f_mode & FMODE_READ) ||
1822 !(file_out->f_mode & FMODE_WRITE) ||
1823 (file_out->f_flags & O_APPEND))
1824 return -EBADF;
1825
1826 if (!file_in->f_op->clone_file_range)
1827 return -EOPNOTSUPP;
1828
1829 ret = clone_verify_area(file_in, pos_in, len, false);
1830 if (ret)
1831 return ret;
1832
1833 ret = clone_verify_area(file_out, pos_out, len, true);
1834 if (ret)
1835 return ret;
1836
1837 if (pos_in + len > i_size_read(inode_in))
1838 return -EINVAL;
1839
1840 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1841 file_out, pos_out, len);
1842 if (!ret) {
1843 fsnotify_access(file_in);
1844 fsnotify_modify(file_out);
1845 }
1846
1847 return ret;
1848 }
1849 EXPORT_SYMBOL(vfs_clone_file_range);
1850
1851 /*
1852 * Read a page's worth of file data into the page cache. Return the page
1853 * locked.
1854 */
1855 static struct page *vfs_dedupe_get_page(struct inode *inode, loff_t offset)
1856 {
1857 struct address_space *mapping;
1858 struct page *page;
1859 pgoff_t n;
1860
1861 n = offset >> PAGE_SHIFT;
1862 mapping = inode->i_mapping;
1863 page = read_mapping_page(mapping, n, NULL);
1864 if (IS_ERR(page))
1865 return page;
1866 if (!PageUptodate(page)) {
1867 put_page(page);
1868 return ERR_PTR(-EIO);
1869 }
1870 lock_page(page);
1871 return page;
1872 }
1873
1874 /*
1875 * Compare extents of two files to see if they are the same.
1876 * Caller must have locked both inodes to prevent write races.
1877 */
1878 int vfs_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
1879 struct inode *dest, loff_t destoff,
1880 loff_t len, bool *is_same)
1881 {
1882 loff_t src_poff;
1883 loff_t dest_poff;
1884 void *src_addr;
1885 void *dest_addr;
1886 struct page *src_page;
1887 struct page *dest_page;
1888 loff_t cmp_len;
1889 bool same;
1890 int error;
1891
1892 error = -EINVAL;
1893 same = true;
1894 while (len) {
1895 src_poff = srcoff & (PAGE_SIZE - 1);
1896 dest_poff = destoff & (PAGE_SIZE - 1);
1897 cmp_len = min(PAGE_SIZE - src_poff,
1898 PAGE_SIZE - dest_poff);
1899 cmp_len = min(cmp_len, len);
1900 if (cmp_len <= 0)
1901 goto out_error;
1902
1903 src_page = vfs_dedupe_get_page(src, srcoff);
1904 if (IS_ERR(src_page)) {
1905 error = PTR_ERR(src_page);
1906 goto out_error;
1907 }
1908 dest_page = vfs_dedupe_get_page(dest, destoff);
1909 if (IS_ERR(dest_page)) {
1910 error = PTR_ERR(dest_page);
1911 unlock_page(src_page);
1912 put_page(src_page);
1913 goto out_error;
1914 }
1915 src_addr = kmap_atomic(src_page);
1916 dest_addr = kmap_atomic(dest_page);
1917
1918 flush_dcache_page(src_page);
1919 flush_dcache_page(dest_page);
1920
1921 if (memcmp(src_addr + src_poff, dest_addr + dest_poff, cmp_len))
1922 same = false;
1923
1924 kunmap_atomic(dest_addr);
1925 kunmap_atomic(src_addr);
1926 unlock_page(dest_page);
1927 unlock_page(src_page);
1928 put_page(dest_page);
1929 put_page(src_page);
1930
1931 if (!same)
1932 break;
1933
1934 srcoff += cmp_len;
1935 destoff += cmp_len;
1936 len -= cmp_len;
1937 }
1938
1939 *is_same = same;
1940 return 0;
1941
1942 out_error:
1943 return error;
1944 }
1945 EXPORT_SYMBOL(vfs_dedupe_file_range_compare);
1946
1947 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
1948 {
1949 struct file_dedupe_range_info *info;
1950 struct inode *src = file_inode(file);
1951 u64 off;
1952 u64 len;
1953 int i;
1954 int ret;
1955 bool is_admin = capable(CAP_SYS_ADMIN);
1956 u16 count = same->dest_count;
1957 struct file *dst_file;
1958 loff_t dst_off;
1959 ssize_t deduped;
1960
1961 if (!(file->f_mode & FMODE_READ))
1962 return -EINVAL;
1963
1964 if (same->reserved1 || same->reserved2)
1965 return -EINVAL;
1966
1967 off = same->src_offset;
1968 len = same->src_length;
1969
1970 ret = -EISDIR;
1971 if (S_ISDIR(src->i_mode))
1972 goto out;
1973
1974 ret = -EINVAL;
1975 if (!S_ISREG(src->i_mode))
1976 goto out;
1977
1978 ret = clone_verify_area(file, off, len, false);
1979 if (ret < 0)
1980 goto out;
1981 ret = 0;
1982
1983 if (off + len > i_size_read(src))
1984 return -EINVAL;
1985
1986 /* pre-format output fields to sane values */
1987 for (i = 0; i < count; i++) {
1988 same->info[i].bytes_deduped = 0ULL;
1989 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
1990 }
1991
1992 for (i = 0, info = same->info; i < count; i++, info++) {
1993 struct inode *dst;
1994 struct fd dst_fd = fdget(info->dest_fd);
1995
1996 dst_file = dst_fd.file;
1997 if (!dst_file) {
1998 info->status = -EBADF;
1999 goto next_loop;
2000 }
2001 dst = file_inode(dst_file);
2002
2003 ret = mnt_want_write_file(dst_file);
2004 if (ret) {
2005 info->status = ret;
2006 goto next_loop;
2007 }
2008
2009 dst_off = info->dest_offset;
2010 ret = clone_verify_area(dst_file, dst_off, len, true);
2011 if (ret < 0) {
2012 info->status = ret;
2013 goto next_file;
2014 }
2015 ret = 0;
2016
2017 if (info->reserved) {
2018 info->status = -EINVAL;
2019 } else if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) {
2020 info->status = -EINVAL;
2021 } else if (file->f_path.mnt != dst_file->f_path.mnt) {
2022 info->status = -EXDEV;
2023 } else if (S_ISDIR(dst->i_mode)) {
2024 info->status = -EISDIR;
2025 } else if (dst_file->f_op->dedupe_file_range == NULL) {
2026 info->status = -EINVAL;
2027 } else {
2028 deduped = dst_file->f_op->dedupe_file_range(file, off,
2029 len, dst_file,
2030 info->dest_offset);
2031 if (deduped == -EBADE)
2032 info->status = FILE_DEDUPE_RANGE_DIFFERS;
2033 else if (deduped < 0)
2034 info->status = deduped;
2035 else
2036 info->bytes_deduped += deduped;
2037 }
2038
2039 next_file:
2040 mnt_drop_write_file(dst_file);
2041 next_loop:
2042 fdput(dst_fd);
2043
2044 if (fatal_signal_pending(current))
2045 goto out;
2046 }
2047
2048 out:
2049 return ret;
2050 }
2051 EXPORT_SYMBOL(vfs_dedupe_file_range);
ubuntu-zesty-kernel mirror