?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
arm64: kgdb: Fix single-step exception handling oops
[linux/fpc-iii.git] / fs / read_write.c
blob9e1fd4c20e8962bc1c06c92ba309d3035174eb92
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 ((unsigned long long)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 ((unsigned long long)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 fsnotify_modify(file);
398 }
399 return ret;
400 }
401 EXPORT_SYMBOL(vfs_iter_write);
402
403 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
404 {
405 struct inode *inode;
406 loff_t pos;
407 int retval = -EINVAL;
408
409 inode = file_inode(file);
410 if (unlikely((ssize_t) count < 0))
411 return retval;
412 pos = *ppos;
413 if (unlikely(pos < 0)) {
414 if (!unsigned_offsets(file))
415 return retval;
416 if (count >= -pos) /* both values are in 0..LLONG_MAX */
417 return -EOVERFLOW;
418 } else if (unlikely((loff_t) (pos + count) < 0)) {
419 if (!unsigned_offsets(file))
420 return retval;
421 }
422
423 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
424 retval = locks_mandatory_area(inode, file, pos, pos + count - 1,
425 read_write == READ ? F_RDLCK : F_WRLCK);
426 if (retval < 0)
427 return retval;
428 }
429 return security_file_permission(file,
430 read_write == READ ? MAY_READ : MAY_WRITE);
431 }
432
433 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
434 {
435 struct iovec iov = { .iov_base = buf, .iov_len = len };
436 struct kiocb kiocb;
437 struct iov_iter iter;
438 ssize_t ret;
439
440 init_sync_kiocb(&kiocb, filp);
441 kiocb.ki_pos = *ppos;
442 iov_iter_init(&iter, READ, &iov, 1, len);
443
444 ret = filp->f_op->read_iter(&kiocb, &iter);
445 BUG_ON(ret == -EIOCBQUEUED);
446 *ppos = kiocb.ki_pos;
447 return ret;
448 }
449
450 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
451 loff_t *pos)
452 {
453 if (file->f_op->read)
454 return file->f_op->read(file, buf, count, pos);
455 else if (file->f_op->read_iter)
456 return new_sync_read(file, buf, count, pos);
457 else
458 return -EINVAL;
459 }
460 EXPORT_SYMBOL(__vfs_read);
461
462 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
463 {
464 ssize_t ret;
465
466 if (!(file->f_mode & FMODE_READ))
467 return -EBADF;
468 if (!(file->f_mode & FMODE_CAN_READ))
469 return -EINVAL;
470 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
471 return -EFAULT;
472
473 ret = rw_verify_area(READ, file, pos, count);
474 if (!ret) {
475 if (count > MAX_RW_COUNT)
476 count = MAX_RW_COUNT;
477 ret = __vfs_read(file, buf, count, pos);
478 if (ret > 0) {
479 fsnotify_access(file);
480 add_rchar(current, ret);
481 }
482 inc_syscr(current);
483 }
484
485 return ret;
486 }
487
488 EXPORT_SYMBOL(vfs_read);
489
490 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
491 {
492 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
493 struct kiocb kiocb;
494 struct iov_iter iter;
495 ssize_t ret;
496
497 init_sync_kiocb(&kiocb, filp);
498 kiocb.ki_pos = *ppos;
499 iov_iter_init(&iter, WRITE, &iov, 1, len);
500
501 ret = filp->f_op->write_iter(&kiocb, &iter);
502 BUG_ON(ret == -EIOCBQUEUED);
503 if (ret > 0)
504 *ppos = kiocb.ki_pos;
505 return ret;
506 }
507
508 ssize_t __vfs_write(struct file *file, const char __user *p, size_t count,
509 loff_t *pos)
510 {
511 if (file->f_op->write)
512 return file->f_op->write(file, p, count, pos);
513 else if (file->f_op->write_iter)
514 return new_sync_write(file, p, count, pos);
515 else
516 return -EINVAL;
517 }
518 EXPORT_SYMBOL(__vfs_write);
519
520 ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
521 {
522 mm_segment_t old_fs;
523 const char __user *p;
524 ssize_t ret;
525
526 if (!(file->f_mode & FMODE_CAN_WRITE))
527 return -EINVAL;
528
529 old_fs = get_fs();
530 set_fs(get_ds());
531 p = (__force const char __user *)buf;
532 if (count > MAX_RW_COUNT)
533 count = MAX_RW_COUNT;
534 ret = __vfs_write(file, p, count, pos);
535 set_fs(old_fs);
536 if (ret > 0) {
537 fsnotify_modify(file);
538 add_wchar(current, ret);
539 }
540 inc_syscw(current);
541 return ret;
542 }
543
544 EXPORT_SYMBOL(__kernel_write);
545
546 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
547 {
548 ssize_t ret;
549
550 if (!(file->f_mode & FMODE_WRITE))
551 return -EBADF;
552 if (!(file->f_mode & FMODE_CAN_WRITE))
553 return -EINVAL;
554 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
555 return -EFAULT;
556
557 ret = rw_verify_area(WRITE, file, pos, count);
558 if (!ret) {
559 if (count > MAX_RW_COUNT)
560 count = MAX_RW_COUNT;
561 file_start_write(file);
562 ret = __vfs_write(file, buf, count, pos);
563 if (ret > 0) {
564 fsnotify_modify(file);
565 add_wchar(current, ret);
566 }
567 inc_syscw(current);
568 file_end_write(file);
569 }
570
571 return ret;
572 }
573
574 EXPORT_SYMBOL(vfs_write);
575
576 static inline loff_t file_pos_read(struct file *file)
577 {
578 return file->f_mode & FMODE_STREAM ? 0 : file->f_pos;
579 }
580
581 static inline void file_pos_write(struct file *file, loff_t pos)
582 {
583 if ((file->f_mode & FMODE_STREAM) == 0)
584 file->f_pos = pos;
585 }
586
587 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
588 {
589 struct fd f = fdget_pos(fd);
590 ssize_t ret = -EBADF;
591
592 if (f.file) {
593 loff_t pos = file_pos_read(f.file);
594 ret = vfs_read(f.file, buf, count, &pos);
595 if (ret >= 0)
596 file_pos_write(f.file, pos);
597 fdput_pos(f);
598 }
599 return ret;
600 }
601
602 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
603 size_t, count)
604 {
605 struct fd f = fdget_pos(fd);
606 ssize_t ret = -EBADF;
607
608 if (f.file) {
609 loff_t pos = file_pos_read(f.file);
610 ret = vfs_write(f.file, buf, count, &pos);
611 if (ret >= 0)
612 file_pos_write(f.file, pos);
613 fdput_pos(f);
614 }
615
616 return ret;
617 }
618
619 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
620 size_t, count, loff_t, pos)
621 {
622 struct fd f;
623 ssize_t ret = -EBADF;
624
625 if (pos < 0)
626 return -EINVAL;
627
628 f = fdget(fd);
629 if (f.file) {
630 ret = -ESPIPE;
631 if (f.file->f_mode & FMODE_PREAD)
632 ret = vfs_read(f.file, buf, count, &pos);
633 fdput(f);
634 }
635
636 return ret;
637 }
638
639 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
640 size_t, count, loff_t, pos)
641 {
642 struct fd f;
643 ssize_t ret = -EBADF;
644
645 if (pos < 0)
646 return -EINVAL;
647
648 f = fdget(fd);
649 if (f.file) {
650 ret = -ESPIPE;
651 if (f.file->f_mode & FMODE_PWRITE)
652 ret = vfs_write(f.file, buf, count, &pos);
653 fdput(f);
654 }
655
656 return ret;
657 }
658
659 /*
660 * Reduce an iovec's length in-place. Return the resulting number of segments
661 */
662 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
663 {
664 unsigned long seg = 0;
665 size_t len = 0;
666
667 while (seg < nr_segs) {
668 seg++;
669 if (len + iov->iov_len >= to) {
670 iov->iov_len = to - len;
671 break;
672 }
673 len += iov->iov_len;
674 iov++;
675 }
676 return seg;
677 }
678 EXPORT_SYMBOL(iov_shorten);
679
680 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
681 loff_t *ppos, iter_fn_t fn, int flags)
682 {
683 struct kiocb kiocb;
684 ssize_t ret;
685
686 if (flags & ~(RWF_HIPRI | RWF_DSYNC | RWF_SYNC))
687 return -EOPNOTSUPP;
688
689 init_sync_kiocb(&kiocb, filp);
690 if (flags & RWF_HIPRI)
691 kiocb.ki_flags |= IOCB_HIPRI;
692 if (flags & RWF_DSYNC)
693 kiocb.ki_flags |= IOCB_DSYNC;
694 if (flags & RWF_SYNC)
695 kiocb.ki_flags |= (IOCB_DSYNC | IOCB_SYNC);
696 kiocb.ki_pos = *ppos;
697
698 ret = fn(&kiocb, iter);
699 BUG_ON(ret == -EIOCBQUEUED);
700 *ppos = kiocb.ki_pos;
701 return ret;
702 }
703
704 /* Do it by hand, with file-ops */
705 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
706 loff_t *ppos, io_fn_t fn, int flags)
707 {
708 ssize_t ret = 0;
709
710 if (flags & ~RWF_HIPRI)
711 return -EOPNOTSUPP;
712
713 while (iov_iter_count(iter)) {
714 struct iovec iovec = iov_iter_iovec(iter);
715 ssize_t nr;
716
717 nr = fn(filp, iovec.iov_base, iovec.iov_len, ppos);
718
719 if (nr < 0) {
720 if (!ret)
721 ret = nr;
722 break;
723 }
724 ret += nr;
725 if (nr != iovec.iov_len)
726 break;
727 iov_iter_advance(iter, nr);
728 }
729
730 return ret;
731 }
732
733 /* A write operation does a read from user space and vice versa */
734 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
735
736 /**
737 * rw_copy_check_uvector() - Copy an array of &struct iovec from userspace
738 * into the kernel and check that it is valid.
739 *
740 * @type: One of %CHECK_IOVEC_ONLY, %READ, or %WRITE.
741 * @uvector: Pointer to the userspace array.
742 * @nr_segs: Number of elements in userspace array.
743 * @fast_segs: Number of elements in @fast_pointer.
744 * @fast_pointer: Pointer to (usually small on-stack) kernel array.
745 * @ret_pointer: (output parameter) Pointer to a variable that will point to
746 * either @fast_pointer, a newly allocated kernel array, or NULL,
747 * depending on which array was used.
748 *
749 * This function copies an array of &struct iovec of @nr_segs from
750 * userspace into the kernel and checks that each element is valid (e.g.
751 * it does not point to a kernel address or cause overflow by being too
752 * large, etc.).
753 *
754 * As an optimization, the caller may provide a pointer to a small
755 * on-stack array in @fast_pointer, typically %UIO_FASTIOV elements long
756 * (the size of this array, or 0 if unused, should be given in @fast_segs).
757 *
758 * @ret_pointer will always point to the array that was used, so the
759 * caller must take care not to call kfree() on it e.g. in case the
760 * @fast_pointer array was used and it was allocated on the stack.
761 *
762 * Return: The total number of bytes covered by the iovec array on success
763 * or a negative error code on error.
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 int flags)
844 {
845 size_t tot_len;
846 struct iovec iovstack[UIO_FASTIOV];
847 struct iovec *iov = iovstack;
848 struct iov_iter iter;
849 ssize_t ret;
850 io_fn_t fn;
851 iter_fn_t iter_fn;
852
853 ret = import_iovec(type, uvector, nr_segs,
854 ARRAY_SIZE(iovstack), &iov, &iter);
855 if (ret < 0)
856 return ret;
857
858 tot_len = iov_iter_count(&iter);
859 if (!tot_len)
860 goto out;
861 ret = rw_verify_area(type, file, pos, tot_len);
862 if (ret < 0)
863 goto out;
864
865 if (type == READ) {
866 fn = file->f_op->read;
867 iter_fn = file->f_op->read_iter;
868 } else {
869 fn = (io_fn_t)file->f_op->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, &iter, pos, iter_fn, flags);
876 else
877 ret = do_loop_readv_writev(file, &iter, pos, fn, flags);
878
879 if (type != READ)
880 file_end_write(file);
881
882 out:
883 kfree(iov);
884 if ((ret + (type == READ)) > 0) {
885 if (type == READ)
886 fsnotify_access(file);
887 else
888 fsnotify_modify(file);
889 }
890 return ret;
891 }
892
893 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
894 unsigned long vlen, loff_t *pos, int flags)
895 {
896 if (!(file->f_mode & FMODE_READ))
897 return -EBADF;
898 if (!(file->f_mode & FMODE_CAN_READ))
899 return -EINVAL;
900
901 return do_readv_writev(READ, file, vec, vlen, pos, flags);
902 }
903
904 EXPORT_SYMBOL(vfs_readv);
905
906 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
907 unsigned long vlen, loff_t *pos, int flags)
908 {
909 if (!(file->f_mode & FMODE_WRITE))
910 return -EBADF;
911 if (!(file->f_mode & FMODE_CAN_WRITE))
912 return -EINVAL;
913
914 return do_readv_writev(WRITE, file, vec, vlen, pos, flags);
915 }
916
917 EXPORT_SYMBOL(vfs_writev);
918
919 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
920 unsigned long vlen, int flags)
921 {
922 struct fd f = fdget_pos(fd);
923 ssize_t ret = -EBADF;
924
925 if (f.file) {
926 loff_t pos = file_pos_read(f.file);
927 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
928 if (ret >= 0)
929 file_pos_write(f.file, pos);
930 fdput_pos(f);
931 }
932
933 if (ret > 0)
934 add_rchar(current, ret);
935 inc_syscr(current);
936 return ret;
937 }
938
939 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
940 unsigned long vlen, int flags)
941 {
942 struct fd f = fdget_pos(fd);
943 ssize_t ret = -EBADF;
944
945 if (f.file) {
946 loff_t pos = file_pos_read(f.file);
947 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
948 if (ret >= 0)
949 file_pos_write(f.file, pos);
950 fdput_pos(f);
951 }
952
953 if (ret > 0)
954 add_wchar(current, ret);
955 inc_syscw(current);
956 return ret;
957 }
958
959 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
960 {
961 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
962 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
963 }
964
965 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
966 unsigned long vlen, loff_t pos, int flags)
967 {
968 struct fd f;
969 ssize_t ret = -EBADF;
970
971 if (pos < 0)
972 return -EINVAL;
973
974 f = fdget(fd);
975 if (f.file) {
976 ret = -ESPIPE;
977 if (f.file->f_mode & FMODE_PREAD)
978 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
979 fdput(f);
980 }
981
982 if (ret > 0)
983 add_rchar(current, ret);
984 inc_syscr(current);
985 return ret;
986 }
987
988 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
989 unsigned long vlen, loff_t pos, int flags)
990 {
991 struct fd f;
992 ssize_t ret = -EBADF;
993
994 if (pos < 0)
995 return -EINVAL;
996
997 f = fdget(fd);
998 if (f.file) {
999 ret = -ESPIPE;
1000 if (f.file->f_mode & FMODE_PWRITE)
1001 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1002 fdput(f);
1003 }
1004
1005 if (ret > 0)
1006 add_wchar(current, ret);
1007 inc_syscw(current);
1008 return ret;
1009 }
1010
1011 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1012 unsigned long, vlen)
1013 {
1014 return do_readv(fd, vec, vlen, 0);
1015 }
1016
1017 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1018 unsigned long, vlen)
1019 {
1020 return do_writev(fd, vec, vlen, 0);
1021 }
1022
1023 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1024 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1025 {
1026 loff_t pos = pos_from_hilo(pos_h, pos_l);
1027
1028 return do_preadv(fd, vec, vlen, pos, 0);
1029 }
1030
1031 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1032 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1033 int, flags)
1034 {
1035 loff_t pos = pos_from_hilo(pos_h, pos_l);
1036
1037 if (pos == -1)
1038 return do_readv(fd, vec, vlen, flags);
1039
1040 return do_preadv(fd, vec, vlen, pos, flags);
1041 }
1042
1043 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1044 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1045 {
1046 loff_t pos = pos_from_hilo(pos_h, pos_l);
1047
1048 return do_pwritev(fd, vec, vlen, pos, 0);
1049 }
1050
1051 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1052 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1053 int, flags)
1054 {
1055 loff_t pos = pos_from_hilo(pos_h, pos_l);
1056
1057 if (pos == -1)
1058 return do_writev(fd, vec, vlen, flags);
1059
1060 return do_pwritev(fd, vec, vlen, pos, flags);
1061 }
1062
1063 #ifdef CONFIG_COMPAT
1064
1065 static ssize_t compat_do_readv_writev(int type, struct file *file,
1066 const struct compat_iovec __user *uvector,
1067 unsigned long nr_segs, loff_t *pos,
1068 int flags)
1069 {
1070 compat_ssize_t tot_len;
1071 struct iovec iovstack[UIO_FASTIOV];
1072 struct iovec *iov = iovstack;
1073 struct iov_iter iter;
1074 ssize_t ret;
1075 io_fn_t fn;
1076 iter_fn_t iter_fn;
1077
1078 ret = compat_import_iovec(type, uvector, nr_segs,
1079 UIO_FASTIOV, &iov, &iter);
1080 if (ret < 0)
1081 return ret;
1082
1083 tot_len = iov_iter_count(&iter);
1084 if (!tot_len)
1085 goto out;
1086 ret = rw_verify_area(type, file, pos, tot_len);
1087 if (ret < 0)
1088 goto out;
1089
1090 if (type == READ) {
1091 fn = file->f_op->read;
1092 iter_fn = file->f_op->read_iter;
1093 } else {
1094 fn = (io_fn_t)file->f_op->write;
1095 iter_fn = file->f_op->write_iter;
1096 file_start_write(file);
1097 }
1098
1099 if (iter_fn)
1100 ret = do_iter_readv_writev(file, &iter, pos, iter_fn, flags);
1101 else
1102 ret = do_loop_readv_writev(file, &iter, pos, fn, flags);
1103
1104 if (type != READ)
1105 file_end_write(file);
1106
1107 out:
1108 kfree(iov);
1109 if ((ret + (type == READ)) > 0) {
1110 if (type == READ)
1111 fsnotify_access(file);
1112 else
1113 fsnotify_modify(file);
1114 }
1115 return ret;
1116 }
1117
1118 static size_t compat_readv(struct file *file,
1119 const struct compat_iovec __user *vec,
1120 unsigned long vlen, loff_t *pos, int flags)
1121 {
1122 ssize_t ret = -EBADF;
1123
1124 if (!(file->f_mode & FMODE_READ))
1125 goto out;
1126
1127 ret = -EINVAL;
1128 if (!(file->f_mode & FMODE_CAN_READ))
1129 goto out;
1130
1131 ret = compat_do_readv_writev(READ, file, vec, vlen, pos, flags);
1132
1133 out:
1134 if (ret > 0)
1135 add_rchar(current, ret);
1136 inc_syscr(current);
1137 return ret;
1138 }
1139
1140 static size_t do_compat_readv(compat_ulong_t fd,
1141 const struct compat_iovec __user *vec,
1142 compat_ulong_t vlen, int flags)
1143 {
1144 struct fd f = fdget_pos(fd);
1145 ssize_t ret;
1146 loff_t pos;
1147
1148 if (!f.file)
1149 return -EBADF;
1150 pos = f.file->f_pos;
1151 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1152 if (ret >= 0)
1153 f.file->f_pos = pos;
1154 fdput_pos(f);
1155 return ret;
1156
1157 }
1158
1159 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1160 const struct compat_iovec __user *,vec,
1161 compat_ulong_t, vlen)
1162 {
1163 return do_compat_readv(fd, vec, vlen, 0);
1164 }
1165
1166 static long do_compat_preadv64(unsigned long fd,
1167 const struct compat_iovec __user *vec,
1168 unsigned long vlen, loff_t pos, int flags)
1169 {
1170 struct fd f;
1171 ssize_t ret;
1172
1173 if (pos < 0)
1174 return -EINVAL;
1175 f = fdget(fd);
1176 if (!f.file)
1177 return -EBADF;
1178 ret = -ESPIPE;
1179 if (f.file->f_mode & FMODE_PREAD)
1180 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1181 fdput(f);
1182 return ret;
1183 }
1184
1185 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1186 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1187 const struct compat_iovec __user *,vec,
1188 unsigned long, vlen, loff_t, pos)
1189 {
1190 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1191 }
1192 #endif
1193
1194 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1195 const struct compat_iovec __user *,vec,
1196 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1197 {
1198 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1199
1200 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1201 }
1202
1203 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1204 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1205 const struct compat_iovec __user *,vec,
1206 unsigned long, vlen, loff_t, pos, int, flags)
1207 {
1208 if (pos == -1)
1209 return do_compat_readv(fd, vec, vlen, flags);
1210
1211 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1212 }
1213 #endif
1214
1215 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1216 const struct compat_iovec __user *,vec,
1217 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1218 int, flags)
1219 {
1220 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1221
1222 if (pos == -1)
1223 return do_compat_readv(fd, vec, vlen, flags);
1224
1225 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1226 }
1227
1228 static size_t compat_writev(struct file *file,
1229 const struct compat_iovec __user *vec,
1230 unsigned long vlen, loff_t *pos, int flags)
1231 {
1232 ssize_t ret = -EBADF;
1233
1234 if (!(file->f_mode & FMODE_WRITE))
1235 goto out;
1236
1237 ret = -EINVAL;
1238 if (!(file->f_mode & FMODE_CAN_WRITE))
1239 goto out;
1240
1241 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos, flags);
1242
1243 out:
1244 if (ret > 0)
1245 add_wchar(current, ret);
1246 inc_syscw(current);
1247 return ret;
1248 }
1249
1250 static size_t do_compat_writev(compat_ulong_t fd,
1251 const struct compat_iovec __user* vec,
1252 compat_ulong_t vlen, int flags)
1253 {
1254 struct fd f = fdget_pos(fd);
1255 ssize_t ret;
1256 loff_t pos;
1257
1258 if (!f.file)
1259 return -EBADF;
1260 pos = f.file->f_pos;
1261 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1262 if (ret >= 0)
1263 f.file->f_pos = pos;
1264 fdput_pos(f);
1265 return ret;
1266 }
1267
1268 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1269 const struct compat_iovec __user *, vec,
1270 compat_ulong_t, vlen)
1271 {
1272 return do_compat_writev(fd, vec, vlen, 0);
1273 }
1274
1275 static long do_compat_pwritev64(unsigned long fd,
1276 const struct compat_iovec __user *vec,
1277 unsigned long vlen, loff_t pos, int flags)
1278 {
1279 struct fd f;
1280 ssize_t ret;
1281
1282 if (pos < 0)
1283 return -EINVAL;
1284 f = fdget(fd);
1285 if (!f.file)
1286 return -EBADF;
1287 ret = -ESPIPE;
1288 if (f.file->f_mode & FMODE_PWRITE)
1289 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1290 fdput(f);
1291 return ret;
1292 }
1293
1294 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1295 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1296 const struct compat_iovec __user *,vec,
1297 unsigned long, vlen, loff_t, pos)
1298 {
1299 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1300 }
1301 #endif
1302
1303 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1304 const struct compat_iovec __user *,vec,
1305 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1306 {
1307 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1308
1309 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1310 }
1311
1312 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1313 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1314 const struct compat_iovec __user *,vec,
1315 unsigned long, vlen, loff_t, pos, int, flags)
1316 {
1317 if (pos == -1)
1318 return do_compat_writev(fd, vec, vlen, flags);
1319
1320 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1321 }
1322 #endif
1323
1324 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1325 const struct compat_iovec __user *,vec,
1326 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, int, flags)
1327 {
1328 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1329
1330 if (pos == -1)
1331 return do_compat_writev(fd, vec, vlen, flags);
1332
1333 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1334 }
1335
1336 #endif
1337
1338 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1339 size_t count, loff_t max)
1340 {
1341 struct fd in, out;
1342 struct inode *in_inode, *out_inode;
1343 loff_t pos;
1344 loff_t out_pos;
1345 ssize_t retval;
1346 int fl;
1347
1348 /*
1349 * Get input file, and verify that it is ok..
1350 */
1351 retval = -EBADF;
1352 in = fdget(in_fd);
1353 if (!in.file)
1354 goto out;
1355 if (!(in.file->f_mode & FMODE_READ))
1356 goto fput_in;
1357 retval = -ESPIPE;
1358 if (!ppos) {
1359 pos = in.file->f_pos;
1360 } else {
1361 pos = *ppos;
1362 if (!(in.file->f_mode & FMODE_PREAD))
1363 goto fput_in;
1364 }
1365 retval = rw_verify_area(READ, in.file, &pos, count);
1366 if (retval < 0)
1367 goto fput_in;
1368 if (count > MAX_RW_COUNT)
1369 count = MAX_RW_COUNT;
1370
1371 /*
1372 * Get output file, and verify that it is ok..
1373 */
1374 retval = -EBADF;
1375 out = fdget(out_fd);
1376 if (!out.file)
1377 goto fput_in;
1378 if (!(out.file->f_mode & FMODE_WRITE))
1379 goto fput_out;
1380 retval = -EINVAL;
1381 in_inode = file_inode(in.file);
1382 out_inode = file_inode(out.file);
1383 out_pos = out.file->f_pos;
1384 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1385 if (retval < 0)
1386 goto fput_out;
1387
1388 if (!max)
1389 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1390
1391 if (unlikely(pos + count > max)) {
1392 retval = -EOVERFLOW;
1393 if (pos >= max)
1394 goto fput_out;
1395 count = max - pos;
1396 }
1397
1398 fl = 0;
1399 #if 0
1400 /*
1401 * We need to debate whether we can enable this or not. The
1402 * man page documents EAGAIN return for the output at least,
1403 * and the application is arguably buggy if it doesn't expect
1404 * EAGAIN on a non-blocking file descriptor.
1405 */
1406 if (in.file->f_flags & O_NONBLOCK)
1407 fl = SPLICE_F_NONBLOCK;
1408 #endif
1409 file_start_write(out.file);
1410 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1411 file_end_write(out.file);
1412
1413 if (retval > 0) {
1414 add_rchar(current, retval);
1415 add_wchar(current, retval);
1416 fsnotify_access(in.file);
1417 fsnotify_modify(out.file);
1418 out.file->f_pos = out_pos;
1419 if (ppos)
1420 *ppos = pos;
1421 else
1422 in.file->f_pos = pos;
1423 }
1424
1425 inc_syscr(current);
1426 inc_syscw(current);
1427 if (pos > max)
1428 retval = -EOVERFLOW;
1429
1430 fput_out:
1431 fdput(out);
1432 fput_in:
1433 fdput(in);
1434 out:
1435 return retval;
1436 }
1437
1438 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1439 {
1440 loff_t pos;
1441 off_t off;
1442 ssize_t ret;
1443
1444 if (offset) {
1445 if (unlikely(get_user(off, offset)))
1446 return -EFAULT;
1447 pos = off;
1448 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1449 if (unlikely(put_user(pos, offset)))
1450 return -EFAULT;
1451 return ret;
1452 }
1453
1454 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1455 }
1456
1457 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1458 {
1459 loff_t pos;
1460 ssize_t ret;
1461
1462 if (offset) {
1463 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1464 return -EFAULT;
1465 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1466 if (unlikely(put_user(pos, offset)))
1467 return -EFAULT;
1468 return ret;
1469 }
1470
1471 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1472 }
1473
1474 #ifdef CONFIG_COMPAT
1475 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1476 compat_off_t __user *, offset, compat_size_t, count)
1477 {
1478 loff_t pos;
1479 off_t off;
1480 ssize_t ret;
1481
1482 if (offset) {
1483 if (unlikely(get_user(off, offset)))
1484 return -EFAULT;
1485 pos = off;
1486 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1487 if (unlikely(put_user(pos, offset)))
1488 return -EFAULT;
1489 return ret;
1490 }
1491
1492 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1493 }
1494
1495 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1496 compat_loff_t __user *, offset, compat_size_t, count)
1497 {
1498 loff_t pos;
1499 ssize_t ret;
1500
1501 if (offset) {
1502 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1503 return -EFAULT;
1504 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1505 if (unlikely(put_user(pos, offset)))
1506 return -EFAULT;
1507 return ret;
1508 }
1509
1510 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1511 }
1512 #endif
1513
1514 /*
1515 * copy_file_range() differs from regular file read and write in that it
1516 * specifically allows return partial success. When it does so is up to
1517 * the copy_file_range method.
1518 */
1519 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1520 struct file *file_out, loff_t pos_out,
1521 size_t len, unsigned int flags)
1522 {
1523 struct inode *inode_in = file_inode(file_in);
1524 struct inode *inode_out = file_inode(file_out);
1525 ssize_t ret;
1526
1527 if (flags != 0)
1528 return -EINVAL;
1529
1530 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1531 return -EISDIR;
1532 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1533 return -EINVAL;
1534
1535 ret = rw_verify_area(READ, file_in, &pos_in, len);
1536 if (unlikely(ret))
1537 return ret;
1538
1539 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1540 if (unlikely(ret))
1541 return ret;
1542
1543 if (!(file_in->f_mode & FMODE_READ) ||
1544 !(file_out->f_mode & FMODE_WRITE) ||
1545 (file_out->f_flags & O_APPEND))
1546 return -EBADF;
1547
1548 /* this could be relaxed once a method supports cross-fs copies */
1549 if (inode_in->i_sb != inode_out->i_sb)
1550 return -EXDEV;
1551
1552 if (len == 0)
1553 return 0;
1554
1555 ret = mnt_want_write_file(file_out);
1556 if (ret)
1557 return ret;
1558
1559 ret = -EOPNOTSUPP;
1560 if (file_out->f_op->copy_file_range)
1561 ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out,
1562 pos_out, len, flags);
1563 if (ret == -EOPNOTSUPP)
1564 ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1565 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1566
1567 if (ret > 0) {
1568 fsnotify_access(file_in);
1569 add_rchar(current, ret);
1570 fsnotify_modify(file_out);
1571 add_wchar(current, ret);
1572 }
1573 inc_syscr(current);
1574 inc_syscw(current);
1575
1576 mnt_drop_write_file(file_out);
1577
1578 return ret;
1579 }
1580 EXPORT_SYMBOL(vfs_copy_file_range);
1581
1582 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1583 int, fd_out, loff_t __user *, off_out,
1584 size_t, len, unsigned int, flags)
1585 {
1586 loff_t pos_in;
1587 loff_t pos_out;
1588 struct fd f_in;
1589 struct fd f_out;
1590 ssize_t ret = -EBADF;
1591
1592 f_in = fdget(fd_in);
1593 if (!f_in.file)
1594 goto out2;
1595
1596 f_out = fdget(fd_out);
1597 if (!f_out.file)
1598 goto out1;
1599
1600 ret = -EFAULT;
1601 if (off_in) {
1602 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1603 goto out;
1604 } else {
1605 pos_in = f_in.file->f_pos;
1606 }
1607
1608 if (off_out) {
1609 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1610 goto out;
1611 } else {
1612 pos_out = f_out.file->f_pos;
1613 }
1614
1615 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1616 flags);
1617 if (ret > 0) {
1618 pos_in += ret;
1619 pos_out += ret;
1620
1621 if (off_in) {
1622 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1623 ret = -EFAULT;
1624 } else {
1625 f_in.file->f_pos = pos_in;
1626 }
1627
1628 if (off_out) {
1629 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1630 ret = -EFAULT;
1631 } else {
1632 f_out.file->f_pos = pos_out;
1633 }
1634 }
1635
1636 out:
1637 fdput(f_out);
1638 out1:
1639 fdput(f_in);
1640 out2:
1641 return ret;
1642 }
1643
1644 static int clone_verify_area(struct file *file, loff_t pos, u64 len, bool write)
1645 {
1646 struct inode *inode = file_inode(file);
1647
1648 if (unlikely(pos < 0))
1649 return -EINVAL;
1650
1651 if (unlikely((loff_t) (pos + len) < 0))
1652 return -EINVAL;
1653
1654 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1655 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1656 int retval;
1657
1658 retval = locks_mandatory_area(inode, file, pos, end,
1659 write ? F_WRLCK : F_RDLCK);
1660 if (retval < 0)
1661 return retval;
1662 }
1663
1664 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1665 }
1666
1667 int vfs_clone_file_range(struct file *file_in, loff_t pos_in,
1668 struct file *file_out, loff_t pos_out, u64 len)
1669 {
1670 struct inode *inode_in = file_inode(file_in);
1671 struct inode *inode_out = file_inode(file_out);
1672 int ret;
1673
1674 if (inode_in->i_sb != inode_out->i_sb ||
1675 file_in->f_path.mnt != file_out->f_path.mnt)
1676 return -EXDEV;
1677
1678 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1679 return -EISDIR;
1680 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1681 return -EINVAL;
1682
1683 if (!(file_in->f_mode & FMODE_READ) ||
1684 !(file_out->f_mode & FMODE_WRITE) ||
1685 (file_out->f_flags & O_APPEND))
1686 return -EBADF;
1687
1688 if (!file_in->f_op->clone_file_range)
1689 return -EOPNOTSUPP;
1690
1691 ret = clone_verify_area(file_in, pos_in, len, false);
1692 if (ret)
1693 return ret;
1694
1695 ret = clone_verify_area(file_out, pos_out, len, true);
1696 if (ret)
1697 return ret;
1698
1699 if (pos_in + len > i_size_read(inode_in))
1700 return -EINVAL;
1701
1702 ret = mnt_want_write_file(file_out);
1703 if (ret)
1704 return ret;
1705
1706 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1707 file_out, pos_out, len);
1708 if (!ret) {
1709 fsnotify_access(file_in);
1710 fsnotify_modify(file_out);
1711 }
1712
1713 mnt_drop_write_file(file_out);
1714 return ret;
1715 }
1716 EXPORT_SYMBOL(vfs_clone_file_range);
1717
1718 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
1719 {
1720 struct file_dedupe_range_info *info;
1721 struct inode *src = file_inode(file);
1722 u64 off;
1723 u64 len;
1724 int i;
1725 int ret;
1726 bool is_admin = capable(CAP_SYS_ADMIN);
1727 u16 count = same->dest_count;
1728 struct file *dst_file;
1729 loff_t dst_off;
1730 ssize_t deduped;
1731
1732 if (!(file->f_mode & FMODE_READ))
1733 return -EINVAL;
1734
1735 if (same->reserved1 || same->reserved2)
1736 return -EINVAL;
1737
1738 off = same->src_offset;
1739 len = same->src_length;
1740
1741 ret = -EISDIR;
1742 if (S_ISDIR(src->i_mode))
1743 goto out;
1744
1745 ret = -EINVAL;
1746 if (!S_ISREG(src->i_mode))
1747 goto out;
1748
1749 ret = clone_verify_area(file, off, len, false);
1750 if (ret < 0)
1751 goto out;
1752 ret = 0;
1753
1754 /* pre-format output fields to sane values */
1755 for (i = 0; i < count; i++) {
1756 same->info[i].bytes_deduped = 0ULL;
1757 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
1758 }
1759
1760 for (i = 0, info = same->info; i < count; i++, info++) {
1761 struct inode *dst;
1762 struct fd dst_fd = fdget(info->dest_fd);
1763
1764 dst_file = dst_fd.file;
1765 if (!dst_file) {
1766 info->status = -EBADF;
1767 goto next_loop;
1768 }
1769 dst = file_inode(dst_file);
1770
1771 ret = mnt_want_write_file(dst_file);
1772 if (ret) {
1773 info->status = ret;
1774 goto next_loop;
1775 }
1776
1777 dst_off = info->dest_offset;
1778 ret = clone_verify_area(dst_file, dst_off, len, true);
1779 if (ret < 0) {
1780 info->status = ret;
1781 goto next_file;
1782 }
1783 ret = 0;
1784
1785 if (info->reserved) {
1786 info->status = -EINVAL;
1787 } else if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) {
1788 info->status = -EINVAL;
1789 } else if (file->f_path.mnt != dst_file->f_path.mnt) {
1790 info->status = -EXDEV;
1791 } else if (S_ISDIR(dst->i_mode)) {
1792 info->status = -EISDIR;
1793 } else if (dst_file->f_op->dedupe_file_range == NULL) {
1794 info->status = -EINVAL;
1795 } else {
1796 deduped = dst_file->f_op->dedupe_file_range(file, off,
1797 len, dst_file,
1798 info->dest_offset);
1799 if (deduped == -EBADE)
1800 info->status = FILE_DEDUPE_RANGE_DIFFERS;
1801 else if (deduped < 0)
1802 info->status = deduped;
1803 else
1804 info->bytes_deduped += deduped;
1805 }
1806
1807 next_file:
1808 mnt_drop_write_file(dst_file);
1809 next_loop:
1810 fdput(dst_fd);
1811
1812 if (fatal_signal_pending(current))
1813 goto out;
1814 }
1815
1816 out:
1817 return ret;
1818 }
1819 EXPORT_SYMBOL(vfs_dedupe_file_range);
Linux with added FPC-III support