?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ARM: tct_hammer_defconfig: Save defconfig
[linux-2.6/btrfs-unstable.git] / fs / read_write.c
blob47c1d4484df9e6b69333215efc10d7a5db64210f
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/sched/xacct.h>
10 #include <linux/fcntl.h>
11 #include <linux/file.h>
12 #include <linux/uio.h>
13 #include <linux/fsnotify.h>
14 #include <linux/security.h>
15 #include <linux/export.h>
16 #include <linux/syscalls.h>
17 #include <linux/pagemap.h>
18 #include <linux/splice.h>
19 #include <linux/compat.h>
20 #include <linux/mount.h>
21 #include <linux/fs.h>
22 #include "internal.h"
23
24 #include <linux/uaccess.h>
25 #include <asm/unistd.h>
26
27 const struct file_operations generic_ro_fops = {
28 .llseek = generic_file_llseek,
29 .read_iter = generic_file_read_iter,
30 .mmap = generic_file_readonly_mmap,
31 .splice_read = generic_file_splice_read,
32 };
33
34 EXPORT_SYMBOL(generic_ro_fops);
35
36 static inline int unsigned_offsets(struct file *file)
37 {
38 return file->f_mode & FMODE_UNSIGNED_OFFSET;
39 }
40
41 /**
42 * vfs_setpos - update the file offset for lseek
43 * @file: file structure in question
44 * @offset: file offset to seek to
45 * @maxsize: maximum file size
46 *
47 * This is a low-level filesystem helper for updating the file offset to
48 * the value specified by @offset if the given offset is valid and it is
49 * not equal to the current file offset.
50 *
51 * Return the specified offset on success and -EINVAL on invalid offset.
52 */
53 loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
54 {
55 if (offset < 0 && !unsigned_offsets(file))
56 return -EINVAL;
57 if (offset > maxsize)
58 return -EINVAL;
59
60 if (offset != file->f_pos) {
61 file->f_pos = offset;
62 file->f_version = 0;
63 }
64 return offset;
65 }
66 EXPORT_SYMBOL(vfs_setpos);
67
68 /**
69 * generic_file_llseek_size - generic llseek implementation for regular files
70 * @file: file structure to seek on
71 * @offset: file offset to seek to
72 * @whence: type of seek
73 * @size: max size of this file in file system
74 * @eof: offset used for SEEK_END position
75 *
76 * This is a variant of generic_file_llseek that allows passing in a custom
77 * maximum file size and a custom EOF position, for e.g. hashed directories
78 *
79 * Synchronization:
80 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
81 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
82 * read/writes behave like SEEK_SET against seeks.
83 */
84 loff_t
85 generic_file_llseek_size(struct file *file, loff_t offset, int whence,
86 loff_t maxsize, loff_t eof)
87 {
88 switch (whence) {
89 case SEEK_END:
90 offset += eof;
91 break;
92 case SEEK_CUR:
93 /*
94 * Here we special-case the lseek(fd, 0, SEEK_CUR)
95 * position-querying operation. Avoid rewriting the "same"
96 * f_pos value back to the file because a concurrent read(),
97 * write() or lseek() might have altered it
98 */
99 if (offset == 0)
100 return file->f_pos;
101 /*
102 * f_lock protects against read/modify/write race with other
103 * SEEK_CURs. Note that parallel writes and reads behave
104 * like SEEK_SET.
105 */
106 spin_lock(&file->f_lock);
107 offset = vfs_setpos(file, file->f_pos + offset, maxsize);
108 spin_unlock(&file->f_lock);
109 return offset;
110 case SEEK_DATA:
111 /*
112 * In the generic case the entire file is data, so as long as
113 * offset isn't at the end of the file then the offset is data.
114 */
115 if (offset >= eof)
116 return -ENXIO;
117 break;
118 case SEEK_HOLE:
119 /*
120 * There is a virtual hole at the end of the file, so as long as
121 * offset isn't i_size or larger, return i_size.
122 */
123 if (offset >= eof)
124 return -ENXIO;
125 offset = eof;
126 break;
127 }
128
129 return vfs_setpos(file, offset, maxsize);
130 }
131 EXPORT_SYMBOL(generic_file_llseek_size);
132
133 /**
134 * generic_file_llseek - generic llseek implementation for regular files
135 * @file: file structure to seek on
136 * @offset: file offset to seek to
137 * @whence: type of seek
138 *
139 * This is a generic implemenation of ->llseek useable for all normal local
140 * filesystems. It just updates the file offset to the value specified by
141 * @offset and @whence.
142 */
143 loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
144 {
145 struct inode *inode = file->f_mapping->host;
146
147 return generic_file_llseek_size(file, offset, whence,
148 inode->i_sb->s_maxbytes,
149 i_size_read(inode));
150 }
151 EXPORT_SYMBOL(generic_file_llseek);
152
153 /**
154 * fixed_size_llseek - llseek implementation for fixed-sized devices
155 * @file: file structure to seek on
156 * @offset: file offset to seek to
157 * @whence: type of seek
158 * @size: size of the file
159 *
160 */
161 loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
162 {
163 switch (whence) {
164 case SEEK_SET: case SEEK_CUR: case SEEK_END:
165 return generic_file_llseek_size(file, offset, whence,
166 size, size);
167 default:
168 return -EINVAL;
169 }
170 }
171 EXPORT_SYMBOL(fixed_size_llseek);
172
173 /**
174 * no_seek_end_llseek - llseek implementation for fixed-sized devices
175 * @file: file structure to seek on
176 * @offset: file offset to seek to
177 * @whence: type of seek
178 *
179 */
180 loff_t no_seek_end_llseek(struct file *file, loff_t offset, int whence)
181 {
182 switch (whence) {
183 case SEEK_SET: case SEEK_CUR:
184 return generic_file_llseek_size(file, offset, whence,
185 OFFSET_MAX, 0);
186 default:
187 return -EINVAL;
188 }
189 }
190 EXPORT_SYMBOL(no_seek_end_llseek);
191
192 /**
193 * no_seek_end_llseek_size - llseek implementation for fixed-sized devices
194 * @file: file structure to seek on
195 * @offset: file offset to seek to
196 * @whence: type of seek
197 * @size: maximal offset allowed
198 *
199 */
200 loff_t no_seek_end_llseek_size(struct file *file, loff_t offset, int whence, loff_t size)
201 {
202 switch (whence) {
203 case SEEK_SET: case SEEK_CUR:
204 return generic_file_llseek_size(file, offset, whence,
205 size, 0);
206 default:
207 return -EINVAL;
208 }
209 }
210 EXPORT_SYMBOL(no_seek_end_llseek_size);
211
212 /**
213 * noop_llseek - No Operation Performed llseek implementation
214 * @file: file structure to seek on
215 * @offset: file offset to seek to
216 * @whence: type of seek
217 *
218 * This is an implementation of ->llseek useable for the rare special case when
219 * userspace expects the seek to succeed but the (device) file is actually not
220 * able to perform the seek. In this case you use noop_llseek() instead of
221 * falling back to the default implementation of ->llseek.
222 */
223 loff_t noop_llseek(struct file *file, loff_t offset, int whence)
224 {
225 return file->f_pos;
226 }
227 EXPORT_SYMBOL(noop_llseek);
228
229 loff_t no_llseek(struct file *file, loff_t offset, int whence)
230 {
231 return -ESPIPE;
232 }
233 EXPORT_SYMBOL(no_llseek);
234
235 loff_t default_llseek(struct file *file, loff_t offset, int whence)
236 {
237 struct inode *inode = file_inode(file);
238 loff_t retval;
239
240 inode_lock(inode);
241 switch (whence) {
242 case SEEK_END:
243 offset += i_size_read(inode);
244 break;
245 case SEEK_CUR:
246 if (offset == 0) {
247 retval = file->f_pos;
248 goto out;
249 }
250 offset += file->f_pos;
251 break;
252 case SEEK_DATA:
253 /*
254 * In the generic case the entire file is data, so as
255 * long as offset isn't at the end of the file then the
256 * offset is data.
257 */
258 if (offset >= inode->i_size) {
259 retval = -ENXIO;
260 goto out;
261 }
262 break;
263 case SEEK_HOLE:
264 /*
265 * There is a virtual hole at the end of the file, so
266 * as long as offset isn't i_size or larger, return
267 * i_size.
268 */
269 if (offset >= inode->i_size) {
270 retval = -ENXIO;
271 goto out;
272 }
273 offset = inode->i_size;
274 break;
275 }
276 retval = -EINVAL;
277 if (offset >= 0 || unsigned_offsets(file)) {
278 if (offset != file->f_pos) {
279 file->f_pos = offset;
280 file->f_version = 0;
281 }
282 retval = offset;
283 }
284 out:
285 inode_unlock(inode);
286 return retval;
287 }
288 EXPORT_SYMBOL(default_llseek);
289
290 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
291 {
292 loff_t (*fn)(struct file *, loff_t, int);
293
294 fn = no_llseek;
295 if (file->f_mode & FMODE_LSEEK) {
296 if (file->f_op->llseek)
297 fn = file->f_op->llseek;
298 }
299 return fn(file, offset, whence);
300 }
301 EXPORT_SYMBOL(vfs_llseek);
302
303 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
304 {
305 off_t retval;
306 struct fd f = fdget_pos(fd);
307 if (!f.file)
308 return -EBADF;
309
310 retval = -EINVAL;
311 if (whence <= SEEK_MAX) {
312 loff_t res = vfs_llseek(f.file, offset, whence);
313 retval = res;
314 if (res != (loff_t)retval)
315 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
316 }
317 fdput_pos(f);
318 return retval;
319 }
320
321 #ifdef CONFIG_COMPAT
322 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
323 {
324 return sys_lseek(fd, offset, whence);
325 }
326 #endif
327
328 #ifdef __ARCH_WANT_SYS_LLSEEK
329 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
330 unsigned long, offset_low, loff_t __user *, result,
331 unsigned int, whence)
332 {
333 int retval;
334 struct fd f = fdget_pos(fd);
335 loff_t offset;
336
337 if (!f.file)
338 return -EBADF;
339
340 retval = -EINVAL;
341 if (whence > SEEK_MAX)
342 goto out_putf;
343
344 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
345 whence);
346
347 retval = (int)offset;
348 if (offset >= 0) {
349 retval = -EFAULT;
350 if (!copy_to_user(result, &offset, sizeof(offset)))
351 retval = 0;
352 }
353 out_putf:
354 fdput_pos(f);
355 return retval;
356 }
357 #endif
358
359 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos)
360 {
361 struct kiocb kiocb;
362 ssize_t ret;
363
364 if (!file->f_op->read_iter)
365 return -EINVAL;
366
367 init_sync_kiocb(&kiocb, file);
368 kiocb.ki_pos = *ppos;
369
370 iter->type |= READ;
371 ret = call_read_iter(file, &kiocb, iter);
372 BUG_ON(ret == -EIOCBQUEUED);
373 if (ret > 0)
374 *ppos = kiocb.ki_pos;
375 return ret;
376 }
377 EXPORT_SYMBOL(vfs_iter_read);
378
379 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos)
380 {
381 struct kiocb kiocb;
382 ssize_t ret;
383
384 if (!file->f_op->write_iter)
385 return -EINVAL;
386
387 init_sync_kiocb(&kiocb, file);
388 kiocb.ki_pos = *ppos;
389
390 iter->type |= WRITE;
391 ret = call_write_iter(file, &kiocb, iter);
392 BUG_ON(ret == -EIOCBQUEUED);
393 if (ret > 0)
394 *ppos = kiocb.ki_pos;
395 return ret;
396 }
397 EXPORT_SYMBOL(vfs_iter_write);
398
399 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
400 {
401 struct inode *inode;
402 loff_t pos;
403 int retval = -EINVAL;
404
405 inode = file_inode(file);
406 if (unlikely((ssize_t) count < 0))
407 return retval;
408 pos = *ppos;
409 if (unlikely(pos < 0)) {
410 if (!unsigned_offsets(file))
411 return retval;
412 if (count >= -pos) /* both values are in 0..LLONG_MAX */
413 return -EOVERFLOW;
414 } else if (unlikely((loff_t) (pos + count) < 0)) {
415 if (!unsigned_offsets(file))
416 return retval;
417 }
418
419 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
420 retval = locks_mandatory_area(inode, file, pos, pos + count - 1,
421 read_write == READ ? F_RDLCK : F_WRLCK);
422 if (retval < 0)
423 return retval;
424 }
425 return security_file_permission(file,
426 read_write == READ ? MAY_READ : MAY_WRITE);
427 }
428
429 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
430 {
431 struct iovec iov = { .iov_base = buf, .iov_len = len };
432 struct kiocb kiocb;
433 struct iov_iter iter;
434 ssize_t ret;
435
436 init_sync_kiocb(&kiocb, filp);
437 kiocb.ki_pos = *ppos;
438 iov_iter_init(&iter, READ, &iov, 1, len);
439
440 ret = call_read_iter(filp, &kiocb, &iter);
441 BUG_ON(ret == -EIOCBQUEUED);
442 *ppos = kiocb.ki_pos;
443 return ret;
444 }
445
446 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
447 loff_t *pos)
448 {
449 if (file->f_op->read)
450 return file->f_op->read(file, buf, count, pos);
451 else if (file->f_op->read_iter)
452 return new_sync_read(file, buf, count, pos);
453 else
454 return -EINVAL;
455 }
456 EXPORT_SYMBOL(__vfs_read);
457
458 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
459 {
460 ssize_t ret;
461
462 if (!(file->f_mode & FMODE_READ))
463 return -EBADF;
464 if (!(file->f_mode & FMODE_CAN_READ))
465 return -EINVAL;
466 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
467 return -EFAULT;
468
469 ret = rw_verify_area(READ, file, pos, count);
470 if (!ret) {
471 if (count > MAX_RW_COUNT)
472 count = MAX_RW_COUNT;
473 ret = __vfs_read(file, buf, count, pos);
474 if (ret > 0) {
475 fsnotify_access(file);
476 add_rchar(current, ret);
477 }
478 inc_syscr(current);
479 }
480
481 return ret;
482 }
483
484 EXPORT_SYMBOL(vfs_read);
485
486 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
487 {
488 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
489 struct kiocb kiocb;
490 struct iov_iter iter;
491 ssize_t ret;
492
493 init_sync_kiocb(&kiocb, filp);
494 kiocb.ki_pos = *ppos;
495 iov_iter_init(&iter, WRITE, &iov, 1, len);
496
497 ret = call_write_iter(filp, &kiocb, &iter);
498 BUG_ON(ret == -EIOCBQUEUED);
499 if (ret > 0)
500 *ppos = kiocb.ki_pos;
501 return ret;
502 }
503
504 ssize_t __vfs_write(struct file *file, const char __user *p, size_t count,
505 loff_t *pos)
506 {
507 if (file->f_op->write)
508 return file->f_op->write(file, p, count, pos);
509 else if (file->f_op->write_iter)
510 return new_sync_write(file, p, count, pos);
511 else
512 return -EINVAL;
513 }
514 EXPORT_SYMBOL(__vfs_write);
515
516 ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
517 {
518 mm_segment_t old_fs;
519 const char __user *p;
520 ssize_t ret;
521
522 if (!(file->f_mode & FMODE_CAN_WRITE))
523 return -EINVAL;
524
525 old_fs = get_fs();
526 set_fs(get_ds());
527 p = (__force const char __user *)buf;
528 if (count > MAX_RW_COUNT)
529 count = MAX_RW_COUNT;
530 ret = __vfs_write(file, p, count, pos);
531 set_fs(old_fs);
532 if (ret > 0) {
533 fsnotify_modify(file);
534 add_wchar(current, ret);
535 }
536 inc_syscw(current);
537 return ret;
538 }
539
540 EXPORT_SYMBOL(__kernel_write);
541
542 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
543 {
544 ssize_t ret;
545
546 if (!(file->f_mode & FMODE_WRITE))
547 return -EBADF;
548 if (!(file->f_mode & FMODE_CAN_WRITE))
549 return -EINVAL;
550 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
551 return -EFAULT;
552
553 ret = rw_verify_area(WRITE, file, pos, count);
554 if (!ret) {
555 if (count > MAX_RW_COUNT)
556 count = MAX_RW_COUNT;
557 file_start_write(file);
558 ret = __vfs_write(file, buf, count, pos);
559 if (ret > 0) {
560 fsnotify_modify(file);
561 add_wchar(current, ret);
562 }
563 inc_syscw(current);
564 file_end_write(file);
565 }
566
567 return ret;
568 }
569
570 EXPORT_SYMBOL(vfs_write);
571
572 static inline loff_t file_pos_read(struct file *file)
573 {
574 return file->f_pos;
575 }
576
577 static inline void file_pos_write(struct file *file, loff_t pos)
578 {
579 file->f_pos = pos;
580 }
581
582 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
583 {
584 struct fd f = fdget_pos(fd);
585 ssize_t ret = -EBADF;
586
587 if (f.file) {
588 loff_t pos = file_pos_read(f.file);
589 ret = vfs_read(f.file, buf, count, &pos);
590 if (ret >= 0)
591 file_pos_write(f.file, pos);
592 fdput_pos(f);
593 }
594 return ret;
595 }
596
597 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
598 size_t, count)
599 {
600 struct fd f = fdget_pos(fd);
601 ssize_t ret = -EBADF;
602
603 if (f.file) {
604 loff_t pos = file_pos_read(f.file);
605 ret = vfs_write(f.file, buf, count, &pos);
606 if (ret >= 0)
607 file_pos_write(f.file, pos);
608 fdput_pos(f);
609 }
610
611 return ret;
612 }
613
614 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
615 size_t, count, loff_t, pos)
616 {
617 struct fd f;
618 ssize_t ret = -EBADF;
619
620 if (pos < 0)
621 return -EINVAL;
622
623 f = fdget(fd);
624 if (f.file) {
625 ret = -ESPIPE;
626 if (f.file->f_mode & FMODE_PREAD)
627 ret = vfs_read(f.file, buf, count, &pos);
628 fdput(f);
629 }
630
631 return ret;
632 }
633
634 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
635 size_t, count, loff_t, pos)
636 {
637 struct fd f;
638 ssize_t ret = -EBADF;
639
640 if (pos < 0)
641 return -EINVAL;
642
643 f = fdget(fd);
644 if (f.file) {
645 ret = -ESPIPE;
646 if (f.file->f_mode & FMODE_PWRITE)
647 ret = vfs_write(f.file, buf, count, &pos);
648 fdput(f);
649 }
650
651 return ret;
652 }
653
654 /*
655 * Reduce an iovec's length in-place. Return the resulting number of segments
656 */
657 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
658 {
659 unsigned long seg = 0;
660 size_t len = 0;
661
662 while (seg < nr_segs) {
663 seg++;
664 if (len + iov->iov_len >= to) {
665 iov->iov_len = to - len;
666 break;
667 }
668 len += iov->iov_len;
669 iov++;
670 }
671 return seg;
672 }
673 EXPORT_SYMBOL(iov_shorten);
674
675 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
676 loff_t *ppos, int type, int flags)
677 {
678 struct kiocb kiocb;
679 ssize_t ret;
680
681 if (flags & ~(RWF_HIPRI | RWF_DSYNC | RWF_SYNC))
682 return -EOPNOTSUPP;
683
684 init_sync_kiocb(&kiocb, filp);
685 if (flags & RWF_HIPRI)
686 kiocb.ki_flags |= IOCB_HIPRI;
687 if (flags & RWF_DSYNC)
688 kiocb.ki_flags |= IOCB_DSYNC;
689 if (flags & RWF_SYNC)
690 kiocb.ki_flags |= (IOCB_DSYNC | IOCB_SYNC);
691 kiocb.ki_pos = *ppos;
692
693 if (type == READ)
694 ret = call_read_iter(filp, &kiocb, iter);
695 else
696 ret = call_write_iter(filp, &kiocb, iter);
697 BUG_ON(ret == -EIOCBQUEUED);
698 *ppos = kiocb.ki_pos;
699 return ret;
700 }
701
702 /* Do it by hand, with file-ops */
703 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
704 loff_t *ppos, int type, int flags)
705 {
706 ssize_t ret = 0;
707
708 if (flags & ~RWF_HIPRI)
709 return -EOPNOTSUPP;
710
711 while (iov_iter_count(iter)) {
712 struct iovec iovec = iov_iter_iovec(iter);
713 ssize_t nr;
714
715 if (type == READ) {
716 nr = filp->f_op->read(filp, iovec.iov_base,
717 iovec.iov_len, ppos);
718 } else {
719 nr = filp->f_op->write(filp, iovec.iov_base,
720 iovec.iov_len, ppos);
721 }
722
723 if (nr < 0) {
724 if (!ret)
725 ret = nr;
726 break;
727 }
728 ret += nr;
729 if (nr != iovec.iov_len)
730 break;
731 iov_iter_advance(iter, nr);
732 }
733
734 return ret;
735 }
736
737 /* A write operation does a read from user space and vice versa */
738 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
739
740 /**
741 * rw_copy_check_uvector() - Copy an array of &struct iovec from userspace
742 * into the kernel and check that it is valid.
743 *
744 * @type: One of %CHECK_IOVEC_ONLY, %READ, or %WRITE.
745 * @uvector: Pointer to the userspace array.
746 * @nr_segs: Number of elements in userspace array.
747 * @fast_segs: Number of elements in @fast_pointer.
748 * @fast_pointer: Pointer to (usually small on-stack) kernel array.
749 * @ret_pointer: (output parameter) Pointer to a variable that will point to
750 * either @fast_pointer, a newly allocated kernel array, or NULL,
751 * depending on which array was used.
752 *
753 * This function copies an array of &struct iovec of @nr_segs from
754 * userspace into the kernel and checks that each element is valid (e.g.
755 * it does not point to a kernel address or cause overflow by being too
756 * large, etc.).
757 *
758 * As an optimization, the caller may provide a pointer to a small
759 * on-stack array in @fast_pointer, typically %UIO_FASTIOV elements long
760 * (the size of this array, or 0 if unused, should be given in @fast_segs).
761 *
762 * @ret_pointer will always point to the array that was used, so the
763 * caller must take care not to call kfree() on it e.g. in case the
764 * @fast_pointer array was used and it was allocated on the stack.
765 *
766 * Return: The total number of bytes covered by the iovec array on success
767 * or a negative error code on error.
768 */
769 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
770 unsigned long nr_segs, unsigned long fast_segs,
771 struct iovec *fast_pointer,
772 struct iovec **ret_pointer)
773 {
774 unsigned long seg;
775 ssize_t ret;
776 struct iovec *iov = fast_pointer;
777
778 /*
779 * SuS says "The readv() function *may* fail if the iovcnt argument
780 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
781 * traditionally returned zero for zero segments, so...
782 */
783 if (nr_segs == 0) {
784 ret = 0;
785 goto out;
786 }
787
788 /*
789 * First get the "struct iovec" from user memory and
790 * verify all the pointers
791 */
792 if (nr_segs > UIO_MAXIOV) {
793 ret = -EINVAL;
794 goto out;
795 }
796 if (nr_segs > fast_segs) {
797 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
798 if (iov == NULL) {
799 ret = -ENOMEM;
800 goto out;
801 }
802 }
803 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
804 ret = -EFAULT;
805 goto out;
806 }
807
808 /*
809 * According to the Single Unix Specification we should return EINVAL
810 * if an element length is < 0 when cast to ssize_t or if the
811 * total length would overflow the ssize_t return value of the
812 * system call.
813 *
814 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
815 * overflow case.
816 */
817 ret = 0;
818 for (seg = 0; seg < nr_segs; seg++) {
819 void __user *buf = iov[seg].iov_base;
820 ssize_t len = (ssize_t)iov[seg].iov_len;
821
822 /* see if we we're about to use an invalid len or if
823 * it's about to overflow ssize_t */
824 if (len < 0) {
825 ret = -EINVAL;
826 goto out;
827 }
828 if (type >= 0
829 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
830 ret = -EFAULT;
831 goto out;
832 }
833 if (len > MAX_RW_COUNT - ret) {
834 len = MAX_RW_COUNT - ret;
835 iov[seg].iov_len = len;
836 }
837 ret += len;
838 }
839 out:
840 *ret_pointer = iov;
841 return ret;
842 }
843
844 #ifdef CONFIG_COMPAT
845 ssize_t compat_rw_copy_check_uvector(int type,
846 const struct compat_iovec __user *uvector, unsigned long nr_segs,
847 unsigned long fast_segs, struct iovec *fast_pointer,
848 struct iovec **ret_pointer)
849 {
850 compat_ssize_t tot_len;
851 struct iovec *iov = *ret_pointer = fast_pointer;
852 ssize_t ret = 0;
853 int seg;
854
855 /*
856 * SuS says "The readv() function *may* fail if the iovcnt argument
857 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
858 * traditionally returned zero for zero segments, so...
859 */
860 if (nr_segs == 0)
861 goto out;
862
863 ret = -EINVAL;
864 if (nr_segs > UIO_MAXIOV)
865 goto out;
866 if (nr_segs > fast_segs) {
867 ret = -ENOMEM;
868 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
869 if (iov == NULL)
870 goto out;
871 }
872 *ret_pointer = iov;
873
874 ret = -EFAULT;
875 if (!access_ok(VERIFY_READ, uvector, nr_segs*sizeof(*uvector)))
876 goto out;
877
878 /*
879 * Single unix specification:
880 * We should -EINVAL if an element length is not >= 0 and fitting an
881 * ssize_t.
882 *
883 * In Linux, the total length is limited to MAX_RW_COUNT, there is
884 * no overflow possibility.
885 */
886 tot_len = 0;
887 ret = -EINVAL;
888 for (seg = 0; seg < nr_segs; seg++) {
889 compat_uptr_t buf;
890 compat_ssize_t len;
891
892 if (__get_user(len, &uvector->iov_len) ||
893 __get_user(buf, &uvector->iov_base)) {
894 ret = -EFAULT;
895 goto out;
896 }
897 if (len < 0) /* size_t not fitting in compat_ssize_t .. */
898 goto out;
899 if (type >= 0 &&
900 !access_ok(vrfy_dir(type), compat_ptr(buf), len)) {
901 ret = -EFAULT;
902 goto out;
903 }
904 if (len > MAX_RW_COUNT - tot_len)
905 len = MAX_RW_COUNT - tot_len;
906 tot_len += len;
907 iov->iov_base = compat_ptr(buf);
908 iov->iov_len = (compat_size_t) len;
909 uvector++;
910 iov++;
911 }
912 ret = tot_len;
913
914 out:
915 return ret;
916 }
917 #endif
918
919 static ssize_t __do_readv_writev(int type, struct file *file,
920 struct iov_iter *iter, loff_t *pos, int flags)
921 {
922 size_t tot_len;
923 ssize_t ret = 0;
924
925 tot_len = iov_iter_count(iter);
926 if (!tot_len)
927 goto out;
928 ret = rw_verify_area(type, file, pos, tot_len);
929 if (ret < 0)
930 goto out;
931
932 if (type != READ)
933 file_start_write(file);
934
935 if ((type == READ && file->f_op->read_iter) ||
936 (type == WRITE && file->f_op->write_iter))
937 ret = do_iter_readv_writev(file, iter, pos, type, flags);
938 else
939 ret = do_loop_readv_writev(file, iter, pos, type, flags);
940
941 if (type != READ)
942 file_end_write(file);
943
944 out:
945 if ((ret + (type == READ)) > 0) {
946 if (type == READ)
947 fsnotify_access(file);
948 else
949 fsnotify_modify(file);
950 }
951 return ret;
952 }
953
954 static ssize_t do_readv_writev(int type, struct file *file,
955 const struct iovec __user *uvector,
956 unsigned long nr_segs, loff_t *pos,
957 int flags)
958 {
959 struct iovec iovstack[UIO_FASTIOV];
960 struct iovec *iov = iovstack;
961 struct iov_iter iter;
962 ssize_t ret;
963
964 ret = import_iovec(type, uvector, nr_segs,
965 ARRAY_SIZE(iovstack), &iov, &iter);
966 if (ret < 0)
967 return ret;
968
969 ret = __do_readv_writev(type, file, &iter, pos, flags);
970 kfree(iov);
971
972 return ret;
973 }
974
975 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
976 unsigned long vlen, loff_t *pos, int flags)
977 {
978 if (!(file->f_mode & FMODE_READ))
979 return -EBADF;
980 if (!(file->f_mode & FMODE_CAN_READ))
981 return -EINVAL;
982
983 return do_readv_writev(READ, file, vec, vlen, pos, flags);
984 }
985
986 EXPORT_SYMBOL(vfs_readv);
987
988 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
989 unsigned long vlen, loff_t *pos, int flags)
990 {
991 if (!(file->f_mode & FMODE_WRITE))
992 return -EBADF;
993 if (!(file->f_mode & FMODE_CAN_WRITE))
994 return -EINVAL;
995
996 return do_readv_writev(WRITE, file, vec, vlen, pos, flags);
997 }
998
999 EXPORT_SYMBOL(vfs_writev);
1000
1001 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
1002 unsigned long vlen, int flags)
1003 {
1004 struct fd f = fdget_pos(fd);
1005 ssize_t ret = -EBADF;
1006
1007 if (f.file) {
1008 loff_t pos = file_pos_read(f.file);
1009 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1010 if (ret >= 0)
1011 file_pos_write(f.file, pos);
1012 fdput_pos(f);
1013 }
1014
1015 if (ret > 0)
1016 add_rchar(current, ret);
1017 inc_syscr(current);
1018 return ret;
1019 }
1020
1021 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
1022 unsigned long vlen, int flags)
1023 {
1024 struct fd f = fdget_pos(fd);
1025 ssize_t ret = -EBADF;
1026
1027 if (f.file) {
1028 loff_t pos = file_pos_read(f.file);
1029 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1030 if (ret >= 0)
1031 file_pos_write(f.file, pos);
1032 fdput_pos(f);
1033 }
1034
1035 if (ret > 0)
1036 add_wchar(current, ret);
1037 inc_syscw(current);
1038 return ret;
1039 }
1040
1041 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
1042 {
1043 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
1044 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
1045 }
1046
1047 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
1048 unsigned long vlen, loff_t pos, int flags)
1049 {
1050 struct fd f;
1051 ssize_t ret = -EBADF;
1052
1053 if (pos < 0)
1054 return -EINVAL;
1055
1056 f = fdget(fd);
1057 if (f.file) {
1058 ret = -ESPIPE;
1059 if (f.file->f_mode & FMODE_PREAD)
1060 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
1061 fdput(f);
1062 }
1063
1064 if (ret > 0)
1065 add_rchar(current, ret);
1066 inc_syscr(current);
1067 return ret;
1068 }
1069
1070 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
1071 unsigned long vlen, loff_t pos, int flags)
1072 {
1073 struct fd f;
1074 ssize_t ret = -EBADF;
1075
1076 if (pos < 0)
1077 return -EINVAL;
1078
1079 f = fdget(fd);
1080 if (f.file) {
1081 ret = -ESPIPE;
1082 if (f.file->f_mode & FMODE_PWRITE)
1083 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1084 fdput(f);
1085 }
1086
1087 if (ret > 0)
1088 add_wchar(current, ret);
1089 inc_syscw(current);
1090 return ret;
1091 }
1092
1093 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1094 unsigned long, vlen)
1095 {
1096 return do_readv(fd, vec, vlen, 0);
1097 }
1098
1099 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1100 unsigned long, vlen)
1101 {
1102 return do_writev(fd, vec, vlen, 0);
1103 }
1104
1105 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1106 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1107 {
1108 loff_t pos = pos_from_hilo(pos_h, pos_l);
1109
1110 return do_preadv(fd, vec, vlen, pos, 0);
1111 }
1112
1113 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1114 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1115 int, flags)
1116 {
1117 loff_t pos = pos_from_hilo(pos_h, pos_l);
1118
1119 if (pos == -1)
1120 return do_readv(fd, vec, vlen, flags);
1121
1122 return do_preadv(fd, vec, vlen, pos, flags);
1123 }
1124
1125 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1126 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1127 {
1128 loff_t pos = pos_from_hilo(pos_h, pos_l);
1129
1130 return do_pwritev(fd, vec, vlen, pos, 0);
1131 }
1132
1133 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1134 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1135 int, flags)
1136 {
1137 loff_t pos = pos_from_hilo(pos_h, pos_l);
1138
1139 if (pos == -1)
1140 return do_writev(fd, vec, vlen, flags);
1141
1142 return do_pwritev(fd, vec, vlen, pos, flags);
1143 }
1144
1145 #ifdef CONFIG_COMPAT
1146
1147 static ssize_t compat_do_readv_writev(int type, struct file *file,
1148 const struct compat_iovec __user *uvector,
1149 unsigned long nr_segs, loff_t *pos,
1150 int flags)
1151 {
1152 struct iovec iovstack[UIO_FASTIOV];
1153 struct iovec *iov = iovstack;
1154 struct iov_iter iter;
1155 ssize_t ret;
1156
1157 ret = compat_import_iovec(type, uvector, nr_segs,
1158 UIO_FASTIOV, &iov, &iter);
1159 if (ret < 0)
1160 return ret;
1161
1162 ret = __do_readv_writev(type, file, &iter, pos, flags);
1163 kfree(iov);
1164
1165 return ret;
1166 }
1167
1168 static size_t compat_readv(struct file *file,
1169 const struct compat_iovec __user *vec,
1170 unsigned long vlen, loff_t *pos, int flags)
1171 {
1172 ssize_t ret = -EBADF;
1173
1174 if (!(file->f_mode & FMODE_READ))
1175 goto out;
1176
1177 ret = -EINVAL;
1178 if (!(file->f_mode & FMODE_CAN_READ))
1179 goto out;
1180
1181 ret = compat_do_readv_writev(READ, file, vec, vlen, pos, flags);
1182
1183 out:
1184 if (ret > 0)
1185 add_rchar(current, ret);
1186 inc_syscr(current);
1187 return ret;
1188 }
1189
1190 static size_t do_compat_readv(compat_ulong_t fd,
1191 const struct compat_iovec __user *vec,
1192 compat_ulong_t vlen, int flags)
1193 {
1194 struct fd f = fdget_pos(fd);
1195 ssize_t ret;
1196 loff_t pos;
1197
1198 if (!f.file)
1199 return -EBADF;
1200 pos = f.file->f_pos;
1201 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1202 if (ret >= 0)
1203 f.file->f_pos = pos;
1204 fdput_pos(f);
1205 return ret;
1206
1207 }
1208
1209 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1210 const struct compat_iovec __user *,vec,
1211 compat_ulong_t, vlen)
1212 {
1213 return do_compat_readv(fd, vec, vlen, 0);
1214 }
1215
1216 static long do_compat_preadv64(unsigned long fd,
1217 const struct compat_iovec __user *vec,
1218 unsigned long vlen, loff_t pos, int flags)
1219 {
1220 struct fd f;
1221 ssize_t ret;
1222
1223 if (pos < 0)
1224 return -EINVAL;
1225 f = fdget(fd);
1226 if (!f.file)
1227 return -EBADF;
1228 ret = -ESPIPE;
1229 if (f.file->f_mode & FMODE_PREAD)
1230 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1231 fdput(f);
1232 return ret;
1233 }
1234
1235 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1236 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1237 const struct compat_iovec __user *,vec,
1238 unsigned long, vlen, loff_t, pos)
1239 {
1240 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1241 }
1242 #endif
1243
1244 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1245 const struct compat_iovec __user *,vec,
1246 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1247 {
1248 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1249
1250 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1251 }
1252
1253 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1254 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1255 const struct compat_iovec __user *,vec,
1256 unsigned long, vlen, loff_t, pos, int, flags)
1257 {
1258 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1259 }
1260 #endif
1261
1262 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1263 const struct compat_iovec __user *,vec,
1264 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1265 int, flags)
1266 {
1267 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1268
1269 if (pos == -1)
1270 return do_compat_readv(fd, vec, vlen, flags);
1271
1272 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1273 }
1274
1275 static size_t compat_writev(struct file *file,
1276 const struct compat_iovec __user *vec,
1277 unsigned long vlen, loff_t *pos, int flags)
1278 {
1279 ssize_t ret = -EBADF;
1280
1281 if (!(file->f_mode & FMODE_WRITE))
1282 goto out;
1283
1284 ret = -EINVAL;
1285 if (!(file->f_mode & FMODE_CAN_WRITE))
1286 goto out;
1287
1288 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos, 0);
1289
1290 out:
1291 if (ret > 0)
1292 add_wchar(current, ret);
1293 inc_syscw(current);
1294 return ret;
1295 }
1296
1297 static size_t do_compat_writev(compat_ulong_t fd,
1298 const struct compat_iovec __user* vec,
1299 compat_ulong_t vlen, int flags)
1300 {
1301 struct fd f = fdget_pos(fd);
1302 ssize_t ret;
1303 loff_t pos;
1304
1305 if (!f.file)
1306 return -EBADF;
1307 pos = f.file->f_pos;
1308 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1309 if (ret >= 0)
1310 f.file->f_pos = pos;
1311 fdput_pos(f);
1312 return ret;
1313 }
1314
1315 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1316 const struct compat_iovec __user *, vec,
1317 compat_ulong_t, vlen)
1318 {
1319 return do_compat_writev(fd, vec, vlen, 0);
1320 }
1321
1322 static long do_compat_pwritev64(unsigned long fd,
1323 const struct compat_iovec __user *vec,
1324 unsigned long vlen, loff_t pos, int flags)
1325 {
1326 struct fd f;
1327 ssize_t ret;
1328
1329 if (pos < 0)
1330 return -EINVAL;
1331 f = fdget(fd);
1332 if (!f.file)
1333 return -EBADF;
1334 ret = -ESPIPE;
1335 if (f.file->f_mode & FMODE_PWRITE)
1336 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1337 fdput(f);
1338 return ret;
1339 }
1340
1341 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1342 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1343 const struct compat_iovec __user *,vec,
1344 unsigned long, vlen, loff_t, pos)
1345 {
1346 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1347 }
1348 #endif
1349
1350 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1351 const struct compat_iovec __user *,vec,
1352 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1353 {
1354 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1355
1356 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1357 }
1358
1359 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1360 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1361 const struct compat_iovec __user *,vec,
1362 unsigned long, vlen, loff_t, pos, int, flags)
1363 {
1364 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1365 }
1366 #endif
1367
1368 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1369 const struct compat_iovec __user *,vec,
1370 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, int, flags)
1371 {
1372 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1373
1374 if (pos == -1)
1375 return do_compat_writev(fd, vec, vlen, flags);
1376
1377 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1378 }
1379
1380 #endif
1381
1382 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1383 size_t count, loff_t max)
1384 {
1385 struct fd in, out;
1386 struct inode *in_inode, *out_inode;
1387 loff_t pos;
1388 loff_t out_pos;
1389 ssize_t retval;
1390 int fl;
1391
1392 /*
1393 * Get input file, and verify that it is ok..
1394 */
1395 retval = -EBADF;
1396 in = fdget(in_fd);
1397 if (!in.file)
1398 goto out;
1399 if (!(in.file->f_mode & FMODE_READ))
1400 goto fput_in;
1401 retval = -ESPIPE;
1402 if (!ppos) {
1403 pos = in.file->f_pos;
1404 } else {
1405 pos = *ppos;
1406 if (!(in.file->f_mode & FMODE_PREAD))
1407 goto fput_in;
1408 }
1409 retval = rw_verify_area(READ, in.file, &pos, count);
1410 if (retval < 0)
1411 goto fput_in;
1412 if (count > MAX_RW_COUNT)
1413 count = MAX_RW_COUNT;
1414
1415 /*
1416 * Get output file, and verify that it is ok..
1417 */
1418 retval = -EBADF;
1419 out = fdget(out_fd);
1420 if (!out.file)
1421 goto fput_in;
1422 if (!(out.file->f_mode & FMODE_WRITE))
1423 goto fput_out;
1424 retval = -EINVAL;
1425 in_inode = file_inode(in.file);
1426 out_inode = file_inode(out.file);
1427 out_pos = out.file->f_pos;
1428 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1429 if (retval < 0)
1430 goto fput_out;
1431
1432 if (!max)
1433 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1434
1435 if (unlikely(pos + count > max)) {
1436 retval = -EOVERFLOW;
1437 if (pos >= max)
1438 goto fput_out;
1439 count = max - pos;
1440 }
1441
1442 fl = 0;
1443 #if 0
1444 /*
1445 * We need to debate whether we can enable this or not. The
1446 * man page documents EAGAIN return for the output at least,
1447 * and the application is arguably buggy if it doesn't expect
1448 * EAGAIN on a non-blocking file descriptor.
1449 */
1450 if (in.file->f_flags & O_NONBLOCK)
1451 fl = SPLICE_F_NONBLOCK;
1452 #endif
1453 file_start_write(out.file);
1454 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1455 file_end_write(out.file);
1456
1457 if (retval > 0) {
1458 add_rchar(current, retval);
1459 add_wchar(current, retval);
1460 fsnotify_access(in.file);
1461 fsnotify_modify(out.file);
1462 out.file->f_pos = out_pos;
1463 if (ppos)
1464 *ppos = pos;
1465 else
1466 in.file->f_pos = pos;
1467 }
1468
1469 inc_syscr(current);
1470 inc_syscw(current);
1471 if (pos > max)
1472 retval = -EOVERFLOW;
1473
1474 fput_out:
1475 fdput(out);
1476 fput_in:
1477 fdput(in);
1478 out:
1479 return retval;
1480 }
1481
1482 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1483 {
1484 loff_t pos;
1485 off_t off;
1486 ssize_t ret;
1487
1488 if (offset) {
1489 if (unlikely(get_user(off, offset)))
1490 return -EFAULT;
1491 pos = off;
1492 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1493 if (unlikely(put_user(pos, offset)))
1494 return -EFAULT;
1495 return ret;
1496 }
1497
1498 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1499 }
1500
1501 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1502 {
1503 loff_t pos;
1504 ssize_t ret;
1505
1506 if (offset) {
1507 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1508 return -EFAULT;
1509 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1510 if (unlikely(put_user(pos, offset)))
1511 return -EFAULT;
1512 return ret;
1513 }
1514
1515 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1516 }
1517
1518 #ifdef CONFIG_COMPAT
1519 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1520 compat_off_t __user *, offset, compat_size_t, count)
1521 {
1522 loff_t pos;
1523 off_t off;
1524 ssize_t ret;
1525
1526 if (offset) {
1527 if (unlikely(get_user(off, offset)))
1528 return -EFAULT;
1529 pos = off;
1530 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1531 if (unlikely(put_user(pos, offset)))
1532 return -EFAULT;
1533 return ret;
1534 }
1535
1536 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1537 }
1538
1539 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1540 compat_loff_t __user *, offset, compat_size_t, count)
1541 {
1542 loff_t pos;
1543 ssize_t ret;
1544
1545 if (offset) {
1546 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1547 return -EFAULT;
1548 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1549 if (unlikely(put_user(pos, offset)))
1550 return -EFAULT;
1551 return ret;
1552 }
1553
1554 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1555 }
1556 #endif
1557
1558 /*
1559 * copy_file_range() differs from regular file read and write in that it
1560 * specifically allows return partial success. When it does so is up to
1561 * the copy_file_range method.
1562 */
1563 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1564 struct file *file_out, loff_t pos_out,
1565 size_t len, unsigned int flags)
1566 {
1567 struct inode *inode_in = file_inode(file_in);
1568 struct inode *inode_out = file_inode(file_out);
1569 ssize_t ret;
1570
1571 if (flags != 0)
1572 return -EINVAL;
1573
1574 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1575 return -EISDIR;
1576 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1577 return -EINVAL;
1578
1579 ret = rw_verify_area(READ, file_in, &pos_in, len);
1580 if (unlikely(ret))
1581 return ret;
1582
1583 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1584 if (unlikely(ret))
1585 return ret;
1586
1587 if (!(file_in->f_mode & FMODE_READ) ||
1588 !(file_out->f_mode & FMODE_WRITE) ||
1589 (file_out->f_flags & O_APPEND))
1590 return -EBADF;
1591
1592 /* this could be relaxed once a method supports cross-fs copies */
1593 if (inode_in->i_sb != inode_out->i_sb)
1594 return -EXDEV;
1595
1596 if (len == 0)
1597 return 0;
1598
1599 file_start_write(file_out);
1600
1601 /*
1602 * Try cloning first, this is supported by more file systems, and
1603 * more efficient if both clone and copy are supported (e.g. NFS).
1604 */
1605 if (file_in->f_op->clone_file_range) {
1606 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1607 file_out, pos_out, len);
1608 if (ret == 0) {
1609 ret = len;
1610 goto done;
1611 }
1612 }
1613
1614 if (file_out->f_op->copy_file_range) {
1615 ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out,
1616 pos_out, len, flags);
1617 if (ret != -EOPNOTSUPP)
1618 goto done;
1619 }
1620
1621 ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1622 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1623
1624 done:
1625 if (ret > 0) {
1626 fsnotify_access(file_in);
1627 add_rchar(current, ret);
1628 fsnotify_modify(file_out);
1629 add_wchar(current, ret);
1630 }
1631
1632 inc_syscr(current);
1633 inc_syscw(current);
1634
1635 file_end_write(file_out);
1636
1637 return ret;
1638 }
1639 EXPORT_SYMBOL(vfs_copy_file_range);
1640
1641 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1642 int, fd_out, loff_t __user *, off_out,
1643 size_t, len, unsigned int, flags)
1644 {
1645 loff_t pos_in;
1646 loff_t pos_out;
1647 struct fd f_in;
1648 struct fd f_out;
1649 ssize_t ret = -EBADF;
1650
1651 f_in = fdget(fd_in);
1652 if (!f_in.file)
1653 goto out2;
1654
1655 f_out = fdget(fd_out);
1656 if (!f_out.file)
1657 goto out1;
1658
1659 ret = -EFAULT;
1660 if (off_in) {
1661 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1662 goto out;
1663 } else {
1664 pos_in = f_in.file->f_pos;
1665 }
1666
1667 if (off_out) {
1668 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1669 goto out;
1670 } else {
1671 pos_out = f_out.file->f_pos;
1672 }
1673
1674 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1675 flags);
1676 if (ret > 0) {
1677 pos_in += ret;
1678 pos_out += ret;
1679
1680 if (off_in) {
1681 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1682 ret = -EFAULT;
1683 } else {
1684 f_in.file->f_pos = pos_in;
1685 }
1686
1687 if (off_out) {
1688 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1689 ret = -EFAULT;
1690 } else {
1691 f_out.file->f_pos = pos_out;
1692 }
1693 }
1694
1695 out:
1696 fdput(f_out);
1697 out1:
1698 fdput(f_in);
1699 out2:
1700 return ret;
1701 }
1702
1703 static int clone_verify_area(struct file *file, loff_t pos, u64 len, bool write)
1704 {
1705 struct inode *inode = file_inode(file);
1706
1707 if (unlikely(pos < 0))
1708 return -EINVAL;
1709
1710 if (unlikely((loff_t) (pos + len) < 0))
1711 return -EINVAL;
1712
1713 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1714 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1715 int retval;
1716
1717 retval = locks_mandatory_area(inode, file, pos, end,
1718 write ? F_WRLCK : F_RDLCK);
1719 if (retval < 0)
1720 return retval;
1721 }
1722
1723 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1724 }
1725
1726 /*
1727 * Check that the two inodes are eligible for cloning, the ranges make
1728 * sense, and then flush all dirty data. Caller must ensure that the
1729 * inodes have been locked against any other modifications.
1730 *
1731 * Returns: 0 for "nothing to clone", 1 for "something to clone", or
1732 * the usual negative error code.
1733 */
1734 int vfs_clone_file_prep_inodes(struct inode *inode_in, loff_t pos_in,
1735 struct inode *inode_out, loff_t pos_out,
1736 u64 *len, bool is_dedupe)
1737 {
1738 loff_t bs = inode_out->i_sb->s_blocksize;
1739 loff_t blen;
1740 loff_t isize;
1741 bool same_inode = (inode_in == inode_out);
1742 int ret;
1743
1744 /* Don't touch certain kinds of inodes */
1745 if (IS_IMMUTABLE(inode_out))
1746 return -EPERM;
1747
1748 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1749 return -ETXTBSY;
1750
1751 /* Don't reflink dirs, pipes, sockets... */
1752 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1753 return -EISDIR;
1754 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1755 return -EINVAL;
1756
1757 /* Are we going all the way to the end? */
1758 isize = i_size_read(inode_in);
1759 if (isize == 0)
1760 return 0;
1761
1762 /* Zero length dedupe exits immediately; reflink goes to EOF. */
1763 if (*len == 0) {
1764 if (is_dedupe || pos_in == isize)
1765 return 0;
1766 if (pos_in > isize)
1767 return -EINVAL;
1768 *len = isize - pos_in;
1769 }
1770
1771 /* Ensure offsets don't wrap and the input is inside i_size */
1772 if (pos_in + *len < pos_in || pos_out + *len < pos_out ||
1773 pos_in + *len > isize)
1774 return -EINVAL;
1775
1776 /* Don't allow dedupe past EOF in the dest file */
1777 if (is_dedupe) {
1778 loff_t disize;
1779
1780 disize = i_size_read(inode_out);
1781 if (pos_out >= disize || pos_out + *len > disize)
1782 return -EINVAL;
1783 }
1784
1785 /* If we're linking to EOF, continue to the block boundary. */
1786 if (pos_in + *len == isize)
1787 blen = ALIGN(isize, bs) - pos_in;
1788 else
1789 blen = *len;
1790
1791 /* Only reflink if we're aligned to block boundaries */
1792 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_in + blen, bs) ||
1793 !IS_ALIGNED(pos_out, bs) || !IS_ALIGNED(pos_out + blen, bs))
1794 return -EINVAL;
1795
1796 /* Don't allow overlapped reflink within the same file */
1797 if (same_inode) {
1798 if (pos_out + blen > pos_in && pos_out < pos_in + blen)
1799 return -EINVAL;
1800 }
1801
1802 /* Wait for the completion of any pending IOs on both files */
1803 inode_dio_wait(inode_in);
1804 if (!same_inode)
1805 inode_dio_wait(inode_out);
1806
1807 ret = filemap_write_and_wait_range(inode_in->i_mapping,
1808 pos_in, pos_in + *len - 1);
1809 if (ret)
1810 return ret;
1811
1812 ret = filemap_write_and_wait_range(inode_out->i_mapping,
1813 pos_out, pos_out + *len - 1);
1814 if (ret)
1815 return ret;
1816
1817 /*
1818 * Check that the extents are the same.
1819 */
1820 if (is_dedupe) {
1821 bool is_same = false;
1822
1823 ret = vfs_dedupe_file_range_compare(inode_in, pos_in,
1824 inode_out, pos_out, *len, &is_same);
1825 if (ret)
1826 return ret;
1827 if (!is_same)
1828 return -EBADE;
1829 }
1830
1831 return 1;
1832 }
1833 EXPORT_SYMBOL(vfs_clone_file_prep_inodes);
1834
1835 int vfs_clone_file_range(struct file *file_in, loff_t pos_in,
1836 struct file *file_out, loff_t pos_out, u64 len)
1837 {
1838 struct inode *inode_in = file_inode(file_in);
1839 struct inode *inode_out = file_inode(file_out);
1840 int ret;
1841
1842 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1843 return -EISDIR;
1844 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1845 return -EINVAL;
1846
1847 /*
1848 * FICLONE/FICLONERANGE ioctls enforce that src and dest files are on
1849 * the same mount. Practically, they only need to be on the same file
1850 * system.
1851 */
1852 if (inode_in->i_sb != inode_out->i_sb)
1853 return -EXDEV;
1854
1855 if (!(file_in->f_mode & FMODE_READ) ||
1856 !(file_out->f_mode & FMODE_WRITE) ||
1857 (file_out->f_flags & O_APPEND))
1858 return -EBADF;
1859
1860 if (!file_in->f_op->clone_file_range)
1861 return -EOPNOTSUPP;
1862
1863 ret = clone_verify_area(file_in, pos_in, len, false);
1864 if (ret)
1865 return ret;
1866
1867 ret = clone_verify_area(file_out, pos_out, len, true);
1868 if (ret)
1869 return ret;
1870
1871 if (pos_in + len > i_size_read(inode_in))
1872 return -EINVAL;
1873
1874 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1875 file_out, pos_out, len);
1876 if (!ret) {
1877 fsnotify_access(file_in);
1878 fsnotify_modify(file_out);
1879 }
1880
1881 return ret;
1882 }
1883 EXPORT_SYMBOL(vfs_clone_file_range);
1884
1885 /*
1886 * Read a page's worth of file data into the page cache. Return the page
1887 * locked.
1888 */
1889 static struct page *vfs_dedupe_get_page(struct inode *inode, loff_t offset)
1890 {
1891 struct address_space *mapping;
1892 struct page *page;
1893 pgoff_t n;
1894
1895 n = offset >> PAGE_SHIFT;
1896 mapping = inode->i_mapping;
1897 page = read_mapping_page(mapping, n, NULL);
1898 if (IS_ERR(page))
1899 return page;
1900 if (!PageUptodate(page)) {
1901 put_page(page);
1902 return ERR_PTR(-EIO);
1903 }
1904 lock_page(page);
1905 return page;
1906 }
1907
1908 /*
1909 * Compare extents of two files to see if they are the same.
1910 * Caller must have locked both inodes to prevent write races.
1911 */
1912 int vfs_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
1913 struct inode *dest, loff_t destoff,
1914 loff_t len, bool *is_same)
1915 {
1916 loff_t src_poff;
1917 loff_t dest_poff;
1918 void *src_addr;
1919 void *dest_addr;
1920 struct page *src_page;
1921 struct page *dest_page;
1922 loff_t cmp_len;
1923 bool same;
1924 int error;
1925
1926 error = -EINVAL;
1927 same = true;
1928 while (len) {
1929 src_poff = srcoff & (PAGE_SIZE - 1);
1930 dest_poff = destoff & (PAGE_SIZE - 1);
1931 cmp_len = min(PAGE_SIZE - src_poff,
1932 PAGE_SIZE - dest_poff);
1933 cmp_len = min(cmp_len, len);
1934 if (cmp_len <= 0)
1935 goto out_error;
1936
1937 src_page = vfs_dedupe_get_page(src, srcoff);
1938 if (IS_ERR(src_page)) {
1939 error = PTR_ERR(src_page);
1940 goto out_error;
1941 }
1942 dest_page = vfs_dedupe_get_page(dest, destoff);
1943 if (IS_ERR(dest_page)) {
1944 error = PTR_ERR(dest_page);
1945 unlock_page(src_page);
1946 put_page(src_page);
1947 goto out_error;
1948 }
1949 src_addr = kmap_atomic(src_page);
1950 dest_addr = kmap_atomic(dest_page);
1951
1952 flush_dcache_page(src_page);
1953 flush_dcache_page(dest_page);
1954
1955 if (memcmp(src_addr + src_poff, dest_addr + dest_poff, cmp_len))
1956 same = false;
1957
1958 kunmap_atomic(dest_addr);
1959 kunmap_atomic(src_addr);
1960 unlock_page(dest_page);
1961 unlock_page(src_page);
1962 put_page(dest_page);
1963 put_page(src_page);
1964
1965 if (!same)
1966 break;
1967
1968 srcoff += cmp_len;
1969 destoff += cmp_len;
1970 len -= cmp_len;
1971 }
1972
1973 *is_same = same;
1974 return 0;
1975
1976 out_error:
1977 return error;
1978 }
1979 EXPORT_SYMBOL(vfs_dedupe_file_range_compare);
1980
1981 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
1982 {
1983 struct file_dedupe_range_info *info;
1984 struct inode *src = file_inode(file);
1985 u64 off;
1986 u64 len;
1987 int i;
1988 int ret;
1989 bool is_admin = capable(CAP_SYS_ADMIN);
1990 u16 count = same->dest_count;
1991 struct file *dst_file;
1992 loff_t dst_off;
1993 ssize_t deduped;
1994
1995 if (!(file->f_mode & FMODE_READ))
1996 return -EINVAL;
1997
1998 if (same->reserved1 || same->reserved2)
1999 return -EINVAL;
2000
2001 off = same->src_offset;
2002 len = same->src_length;
2003
2004 ret = -EISDIR;
2005 if (S_ISDIR(src->i_mode))
2006 goto out;
2007
2008 ret = -EINVAL;
2009 if (!S_ISREG(src->i_mode))
2010 goto out;
2011
2012 ret = clone_verify_area(file, off, len, false);
2013 if (ret < 0)
2014 goto out;
2015 ret = 0;
2016
2017 if (off + len > i_size_read(src))
2018 return -EINVAL;
2019
2020 /* pre-format output fields to sane values */
2021 for (i = 0; i < count; i++) {
2022 same->info[i].bytes_deduped = 0ULL;
2023 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
2024 }
2025
2026 for (i = 0, info = same->info; i < count; i++, info++) {
2027 struct inode *dst;
2028 struct fd dst_fd = fdget(info->dest_fd);
2029
2030 dst_file = dst_fd.file;
2031 if (!dst_file) {
2032 info->status = -EBADF;
2033 goto next_loop;
2034 }
2035 dst = file_inode(dst_file);
2036
2037 ret = mnt_want_write_file(dst_file);
2038 if (ret) {
2039 info->status = ret;
2040 goto next_loop;
2041 }
2042
2043 dst_off = info->dest_offset;
2044 ret = clone_verify_area(dst_file, dst_off, len, true);
2045 if (ret < 0) {
2046 info->status = ret;
2047 goto next_file;
2048 }
2049 ret = 0;
2050
2051 if (info->reserved) {
2052 info->status = -EINVAL;
2053 } else if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) {
2054 info->status = -EINVAL;
2055 } else if (file->f_path.mnt != dst_file->f_path.mnt) {
2056 info->status = -EXDEV;
2057 } else if (S_ISDIR(dst->i_mode)) {
2058 info->status = -EISDIR;
2059 } else if (dst_file->f_op->dedupe_file_range == NULL) {
2060 info->status = -EINVAL;
2061 } else {
2062 deduped = dst_file->f_op->dedupe_file_range(file, off,
2063 len, dst_file,
2064 info->dest_offset);
2065 if (deduped == -EBADE)
2066 info->status = FILE_DEDUPE_RANGE_DIFFERS;
2067 else if (deduped < 0)
2068 info->status = deduped;
2069 else
2070 info->bytes_deduped += deduped;
2071 }
2072
2073 next_file:
2074 mnt_drop_write_file(dst_file);
2075 next_loop:
2076 fdput(dst_fd);
2077
2078 if (fatal_signal_pending(current))
2079 goto out;
2080 }
2081
2082 out:
2083 return ret;
2084 }
2085 EXPORT_SYMBOL(vfs_dedupe_file_range);
(deprecated) Btrfs devel tree