?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ibmvnic: Move login to its own routine
[linux-2.6/btrfs-unstable.git] / fs / read_write.c
blobc4f88afbc67f49ae9e451e06a81c45c37f8171f5
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 static ssize_t __do_readv_writev(int type, struct file *file,
845 struct iov_iter *iter, loff_t *pos, int flags)
846 {
847 size_t tot_len;
848 ssize_t ret = 0;
849
850 tot_len = iov_iter_count(iter);
851 if (!tot_len)
852 goto out;
853 ret = rw_verify_area(type, file, pos, tot_len);
854 if (ret < 0)
855 goto out;
856
857 if (type != READ)
858 file_start_write(file);
859
860 if ((type == READ && file->f_op->read_iter) ||
861 (type == WRITE && file->f_op->write_iter))
862 ret = do_iter_readv_writev(file, iter, pos, type, flags);
863 else
864 ret = do_loop_readv_writev(file, iter, pos, type, flags);
865
866 if (type != READ)
867 file_end_write(file);
868
869 out:
870 if ((ret + (type == READ)) > 0) {
871 if (type == READ)
872 fsnotify_access(file);
873 else
874 fsnotify_modify(file);
875 }
876 return ret;
877 }
878
879 static ssize_t do_readv_writev(int type, struct file *file,
880 const struct iovec __user *uvector,
881 unsigned long nr_segs, loff_t *pos,
882 int flags)
883 {
884 struct iovec iovstack[UIO_FASTIOV];
885 struct iovec *iov = iovstack;
886 struct iov_iter iter;
887 ssize_t ret;
888
889 ret = import_iovec(type, uvector, nr_segs,
890 ARRAY_SIZE(iovstack), &iov, &iter);
891 if (ret < 0)
892 return ret;
893
894 ret = __do_readv_writev(type, file, &iter, pos, flags);
895 kfree(iov);
896
897 return ret;
898 }
899
900 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
901 unsigned long vlen, loff_t *pos, int flags)
902 {
903 if (!(file->f_mode & FMODE_READ))
904 return -EBADF;
905 if (!(file->f_mode & FMODE_CAN_READ))
906 return -EINVAL;
907
908 return do_readv_writev(READ, file, vec, vlen, pos, flags);
909 }
910
911 EXPORT_SYMBOL(vfs_readv);
912
913 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
914 unsigned long vlen, loff_t *pos, int flags)
915 {
916 if (!(file->f_mode & FMODE_WRITE))
917 return -EBADF;
918 if (!(file->f_mode & FMODE_CAN_WRITE))
919 return -EINVAL;
920
921 return do_readv_writev(WRITE, file, vec, vlen, pos, flags);
922 }
923
924 EXPORT_SYMBOL(vfs_writev);
925
926 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
927 unsigned long vlen, int flags)
928 {
929 struct fd f = fdget_pos(fd);
930 ssize_t ret = -EBADF;
931
932 if (f.file) {
933 loff_t pos = file_pos_read(f.file);
934 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
935 if (ret >= 0)
936 file_pos_write(f.file, pos);
937 fdput_pos(f);
938 }
939
940 if (ret > 0)
941 add_rchar(current, ret);
942 inc_syscr(current);
943 return ret;
944 }
945
946 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
947 unsigned long vlen, int flags)
948 {
949 struct fd f = fdget_pos(fd);
950 ssize_t ret = -EBADF;
951
952 if (f.file) {
953 loff_t pos = file_pos_read(f.file);
954 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
955 if (ret >= 0)
956 file_pos_write(f.file, pos);
957 fdput_pos(f);
958 }
959
960 if (ret > 0)
961 add_wchar(current, ret);
962 inc_syscw(current);
963 return ret;
964 }
965
966 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
967 {
968 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
969 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
970 }
971
972 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
973 unsigned long vlen, loff_t pos, int flags)
974 {
975 struct fd f;
976 ssize_t ret = -EBADF;
977
978 if (pos < 0)
979 return -EINVAL;
980
981 f = fdget(fd);
982 if (f.file) {
983 ret = -ESPIPE;
984 if (f.file->f_mode & FMODE_PREAD)
985 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
986 fdput(f);
987 }
988
989 if (ret > 0)
990 add_rchar(current, ret);
991 inc_syscr(current);
992 return ret;
993 }
994
995 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
996 unsigned long vlen, loff_t pos, int flags)
997 {
998 struct fd f;
999 ssize_t ret = -EBADF;
1000
1001 if (pos < 0)
1002 return -EINVAL;
1003
1004 f = fdget(fd);
1005 if (f.file) {
1006 ret = -ESPIPE;
1007 if (f.file->f_mode & FMODE_PWRITE)
1008 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1009 fdput(f);
1010 }
1011
1012 if (ret > 0)
1013 add_wchar(current, ret);
1014 inc_syscw(current);
1015 return ret;
1016 }
1017
1018 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1019 unsigned long, vlen)
1020 {
1021 return do_readv(fd, vec, vlen, 0);
1022 }
1023
1024 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1025 unsigned long, vlen)
1026 {
1027 return do_writev(fd, vec, vlen, 0);
1028 }
1029
1030 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1031 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1032 {
1033 loff_t pos = pos_from_hilo(pos_h, pos_l);
1034
1035 return do_preadv(fd, vec, vlen, pos, 0);
1036 }
1037
1038 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1039 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1040 int, flags)
1041 {
1042 loff_t pos = pos_from_hilo(pos_h, pos_l);
1043
1044 if (pos == -1)
1045 return do_readv(fd, vec, vlen, flags);
1046
1047 return do_preadv(fd, vec, vlen, pos, flags);
1048 }
1049
1050 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1051 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1052 {
1053 loff_t pos = pos_from_hilo(pos_h, pos_l);
1054
1055 return do_pwritev(fd, vec, vlen, pos, 0);
1056 }
1057
1058 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1059 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1060 int, flags)
1061 {
1062 loff_t pos = pos_from_hilo(pos_h, pos_l);
1063
1064 if (pos == -1)
1065 return do_writev(fd, vec, vlen, flags);
1066
1067 return do_pwritev(fd, vec, vlen, pos, flags);
1068 }
1069
1070 #ifdef CONFIG_COMPAT
1071
1072 static ssize_t compat_do_readv_writev(int type, struct file *file,
1073 const struct compat_iovec __user *uvector,
1074 unsigned long nr_segs, loff_t *pos,
1075 int flags)
1076 {
1077 struct iovec iovstack[UIO_FASTIOV];
1078 struct iovec *iov = iovstack;
1079 struct iov_iter iter;
1080 ssize_t ret;
1081
1082 ret = compat_import_iovec(type, uvector, nr_segs,
1083 UIO_FASTIOV, &iov, &iter);
1084 if (ret < 0)
1085 return ret;
1086
1087 ret = __do_readv_writev(type, file, &iter, pos, flags);
1088 kfree(iov);
1089
1090 return ret;
1091 }
1092
1093 static size_t compat_readv(struct file *file,
1094 const struct compat_iovec __user *vec,
1095 unsigned long vlen, loff_t *pos, int flags)
1096 {
1097 ssize_t ret = -EBADF;
1098
1099 if (!(file->f_mode & FMODE_READ))
1100 goto out;
1101
1102 ret = -EINVAL;
1103 if (!(file->f_mode & FMODE_CAN_READ))
1104 goto out;
1105
1106 ret = compat_do_readv_writev(READ, file, vec, vlen, pos, flags);
1107
1108 out:
1109 if (ret > 0)
1110 add_rchar(current, ret);
1111 inc_syscr(current);
1112 return ret;
1113 }
1114
1115 static size_t do_compat_readv(compat_ulong_t fd,
1116 const struct compat_iovec __user *vec,
1117 compat_ulong_t vlen, int flags)
1118 {
1119 struct fd f = fdget_pos(fd);
1120 ssize_t ret;
1121 loff_t pos;
1122
1123 if (!f.file)
1124 return -EBADF;
1125 pos = f.file->f_pos;
1126 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1127 if (ret >= 0)
1128 f.file->f_pos = pos;
1129 fdput_pos(f);
1130 return ret;
1131
1132 }
1133
1134 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1135 const struct compat_iovec __user *,vec,
1136 compat_ulong_t, vlen)
1137 {
1138 return do_compat_readv(fd, vec, vlen, 0);
1139 }
1140
1141 static long do_compat_preadv64(unsigned long fd,
1142 const struct compat_iovec __user *vec,
1143 unsigned long vlen, loff_t pos, int flags)
1144 {
1145 struct fd f;
1146 ssize_t ret;
1147
1148 if (pos < 0)
1149 return -EINVAL;
1150 f = fdget(fd);
1151 if (!f.file)
1152 return -EBADF;
1153 ret = -ESPIPE;
1154 if (f.file->f_mode & FMODE_PREAD)
1155 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1156 fdput(f);
1157 return ret;
1158 }
1159
1160 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1161 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1162 const struct compat_iovec __user *,vec,
1163 unsigned long, vlen, loff_t, pos)
1164 {
1165 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1166 }
1167 #endif
1168
1169 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1170 const struct compat_iovec __user *,vec,
1171 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1172 {
1173 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1174
1175 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1176 }
1177
1178 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1179 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1180 const struct compat_iovec __user *,vec,
1181 unsigned long, vlen, loff_t, pos, int, flags)
1182 {
1183 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1184 }
1185 #endif
1186
1187 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1188 const struct compat_iovec __user *,vec,
1189 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1190 int, flags)
1191 {
1192 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1193
1194 if (pos == -1)
1195 return do_compat_readv(fd, vec, vlen, flags);
1196
1197 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1198 }
1199
1200 static size_t compat_writev(struct file *file,
1201 const struct compat_iovec __user *vec,
1202 unsigned long vlen, loff_t *pos, int flags)
1203 {
1204 ssize_t ret = -EBADF;
1205
1206 if (!(file->f_mode & FMODE_WRITE))
1207 goto out;
1208
1209 ret = -EINVAL;
1210 if (!(file->f_mode & FMODE_CAN_WRITE))
1211 goto out;
1212
1213 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos, 0);
1214
1215 out:
1216 if (ret > 0)
1217 add_wchar(current, ret);
1218 inc_syscw(current);
1219 return ret;
1220 }
1221
1222 static size_t do_compat_writev(compat_ulong_t fd,
1223 const struct compat_iovec __user* vec,
1224 compat_ulong_t vlen, int flags)
1225 {
1226 struct fd f = fdget_pos(fd);
1227 ssize_t ret;
1228 loff_t pos;
1229
1230 if (!f.file)
1231 return -EBADF;
1232 pos = f.file->f_pos;
1233 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1234 if (ret >= 0)
1235 f.file->f_pos = pos;
1236 fdput_pos(f);
1237 return ret;
1238 }
1239
1240 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1241 const struct compat_iovec __user *, vec,
1242 compat_ulong_t, vlen)
1243 {
1244 return do_compat_writev(fd, vec, vlen, 0);
1245 }
1246
1247 static long do_compat_pwritev64(unsigned long fd,
1248 const struct compat_iovec __user *vec,
1249 unsigned long vlen, loff_t pos, int flags)
1250 {
1251 struct fd f;
1252 ssize_t ret;
1253
1254 if (pos < 0)
1255 return -EINVAL;
1256 f = fdget(fd);
1257 if (!f.file)
1258 return -EBADF;
1259 ret = -ESPIPE;
1260 if (f.file->f_mode & FMODE_PWRITE)
1261 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1262 fdput(f);
1263 return ret;
1264 }
1265
1266 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1267 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1268 const struct compat_iovec __user *,vec,
1269 unsigned long, vlen, loff_t, pos)
1270 {
1271 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1272 }
1273 #endif
1274
1275 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1276 const struct compat_iovec __user *,vec,
1277 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1278 {
1279 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1280
1281 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1282 }
1283
1284 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1285 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1286 const struct compat_iovec __user *,vec,
1287 unsigned long, vlen, loff_t, pos, int, flags)
1288 {
1289 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1290 }
1291 #endif
1292
1293 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1294 const struct compat_iovec __user *,vec,
1295 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, int, flags)
1296 {
1297 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1298
1299 if (pos == -1)
1300 return do_compat_writev(fd, vec, vlen, flags);
1301
1302 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1303 }
1304
1305 #endif
1306
1307 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1308 size_t count, loff_t max)
1309 {
1310 struct fd in, out;
1311 struct inode *in_inode, *out_inode;
1312 loff_t pos;
1313 loff_t out_pos;
1314 ssize_t retval;
1315 int fl;
1316
1317 /*
1318 * Get input file, and verify that it is ok..
1319 */
1320 retval = -EBADF;
1321 in = fdget(in_fd);
1322 if (!in.file)
1323 goto out;
1324 if (!(in.file->f_mode & FMODE_READ))
1325 goto fput_in;
1326 retval = -ESPIPE;
1327 if (!ppos) {
1328 pos = in.file->f_pos;
1329 } else {
1330 pos = *ppos;
1331 if (!(in.file->f_mode & FMODE_PREAD))
1332 goto fput_in;
1333 }
1334 retval = rw_verify_area(READ, in.file, &pos, count);
1335 if (retval < 0)
1336 goto fput_in;
1337 if (count > MAX_RW_COUNT)
1338 count = MAX_RW_COUNT;
1339
1340 /*
1341 * Get output file, and verify that it is ok..
1342 */
1343 retval = -EBADF;
1344 out = fdget(out_fd);
1345 if (!out.file)
1346 goto fput_in;
1347 if (!(out.file->f_mode & FMODE_WRITE))
1348 goto fput_out;
1349 retval = -EINVAL;
1350 in_inode = file_inode(in.file);
1351 out_inode = file_inode(out.file);
1352 out_pos = out.file->f_pos;
1353 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1354 if (retval < 0)
1355 goto fput_out;
1356
1357 if (!max)
1358 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1359
1360 if (unlikely(pos + count > max)) {
1361 retval = -EOVERFLOW;
1362 if (pos >= max)
1363 goto fput_out;
1364 count = max - pos;
1365 }
1366
1367 fl = 0;
1368 #if 0
1369 /*
1370 * We need to debate whether we can enable this or not. The
1371 * man page documents EAGAIN return for the output at least,
1372 * and the application is arguably buggy if it doesn't expect
1373 * EAGAIN on a non-blocking file descriptor.
1374 */
1375 if (in.file->f_flags & O_NONBLOCK)
1376 fl = SPLICE_F_NONBLOCK;
1377 #endif
1378 file_start_write(out.file);
1379 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1380 file_end_write(out.file);
1381
1382 if (retval > 0) {
1383 add_rchar(current, retval);
1384 add_wchar(current, retval);
1385 fsnotify_access(in.file);
1386 fsnotify_modify(out.file);
1387 out.file->f_pos = out_pos;
1388 if (ppos)
1389 *ppos = pos;
1390 else
1391 in.file->f_pos = pos;
1392 }
1393
1394 inc_syscr(current);
1395 inc_syscw(current);
1396 if (pos > max)
1397 retval = -EOVERFLOW;
1398
1399 fput_out:
1400 fdput(out);
1401 fput_in:
1402 fdput(in);
1403 out:
1404 return retval;
1405 }
1406
1407 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1408 {
1409 loff_t pos;
1410 off_t off;
1411 ssize_t ret;
1412
1413 if (offset) {
1414 if (unlikely(get_user(off, offset)))
1415 return -EFAULT;
1416 pos = off;
1417 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1418 if (unlikely(put_user(pos, offset)))
1419 return -EFAULT;
1420 return ret;
1421 }
1422
1423 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1424 }
1425
1426 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1427 {
1428 loff_t pos;
1429 ssize_t ret;
1430
1431 if (offset) {
1432 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1433 return -EFAULT;
1434 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1435 if (unlikely(put_user(pos, offset)))
1436 return -EFAULT;
1437 return ret;
1438 }
1439
1440 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1441 }
1442
1443 #ifdef CONFIG_COMPAT
1444 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1445 compat_off_t __user *, offset, compat_size_t, count)
1446 {
1447 loff_t pos;
1448 off_t off;
1449 ssize_t ret;
1450
1451 if (offset) {
1452 if (unlikely(get_user(off, offset)))
1453 return -EFAULT;
1454 pos = off;
1455 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1456 if (unlikely(put_user(pos, offset)))
1457 return -EFAULT;
1458 return ret;
1459 }
1460
1461 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1462 }
1463
1464 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1465 compat_loff_t __user *, offset, compat_size_t, count)
1466 {
1467 loff_t pos;
1468 ssize_t ret;
1469
1470 if (offset) {
1471 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1472 return -EFAULT;
1473 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1474 if (unlikely(put_user(pos, offset)))
1475 return -EFAULT;
1476 return ret;
1477 }
1478
1479 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1480 }
1481 #endif
1482
1483 /*
1484 * copy_file_range() differs from regular file read and write in that it
1485 * specifically allows return partial success. When it does so is up to
1486 * the copy_file_range method.
1487 */
1488 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1489 struct file *file_out, loff_t pos_out,
1490 size_t len, unsigned int flags)
1491 {
1492 struct inode *inode_in = file_inode(file_in);
1493 struct inode *inode_out = file_inode(file_out);
1494 ssize_t ret;
1495
1496 if (flags != 0)
1497 return -EINVAL;
1498
1499 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1500 return -EISDIR;
1501 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1502 return -EINVAL;
1503
1504 ret = rw_verify_area(READ, file_in, &pos_in, len);
1505 if (unlikely(ret))
1506 return ret;
1507
1508 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1509 if (unlikely(ret))
1510 return ret;
1511
1512 if (!(file_in->f_mode & FMODE_READ) ||
1513 !(file_out->f_mode & FMODE_WRITE) ||
1514 (file_out->f_flags & O_APPEND))
1515 return -EBADF;
1516
1517 /* this could be relaxed once a method supports cross-fs copies */
1518 if (inode_in->i_sb != inode_out->i_sb)
1519 return -EXDEV;
1520
1521 if (len == 0)
1522 return 0;
1523
1524 file_start_write(file_out);
1525
1526 /*
1527 * Try cloning first, this is supported by more file systems, and
1528 * more efficient if both clone and copy are supported (e.g. NFS).
1529 */
1530 if (file_in->f_op->clone_file_range) {
1531 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1532 file_out, pos_out, len);
1533 if (ret == 0) {
1534 ret = len;
1535 goto done;
1536 }
1537 }
1538
1539 if (file_out->f_op->copy_file_range) {
1540 ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out,
1541 pos_out, len, flags);
1542 if (ret != -EOPNOTSUPP)
1543 goto done;
1544 }
1545
1546 ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1547 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1548
1549 done:
1550 if (ret > 0) {
1551 fsnotify_access(file_in);
1552 add_rchar(current, ret);
1553 fsnotify_modify(file_out);
1554 add_wchar(current, ret);
1555 }
1556
1557 inc_syscr(current);
1558 inc_syscw(current);
1559
1560 file_end_write(file_out);
1561
1562 return ret;
1563 }
1564 EXPORT_SYMBOL(vfs_copy_file_range);
1565
1566 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1567 int, fd_out, loff_t __user *, off_out,
1568 size_t, len, unsigned int, flags)
1569 {
1570 loff_t pos_in;
1571 loff_t pos_out;
1572 struct fd f_in;
1573 struct fd f_out;
1574 ssize_t ret = -EBADF;
1575
1576 f_in = fdget(fd_in);
1577 if (!f_in.file)
1578 goto out2;
1579
1580 f_out = fdget(fd_out);
1581 if (!f_out.file)
1582 goto out1;
1583
1584 ret = -EFAULT;
1585 if (off_in) {
1586 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1587 goto out;
1588 } else {
1589 pos_in = f_in.file->f_pos;
1590 }
1591
1592 if (off_out) {
1593 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1594 goto out;
1595 } else {
1596 pos_out = f_out.file->f_pos;
1597 }
1598
1599 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1600 flags);
1601 if (ret > 0) {
1602 pos_in += ret;
1603 pos_out += ret;
1604
1605 if (off_in) {
1606 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1607 ret = -EFAULT;
1608 } else {
1609 f_in.file->f_pos = pos_in;
1610 }
1611
1612 if (off_out) {
1613 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1614 ret = -EFAULT;
1615 } else {
1616 f_out.file->f_pos = pos_out;
1617 }
1618 }
1619
1620 out:
1621 fdput(f_out);
1622 out1:
1623 fdput(f_in);
1624 out2:
1625 return ret;
1626 }
1627
1628 static int clone_verify_area(struct file *file, loff_t pos, u64 len, bool write)
1629 {
1630 struct inode *inode = file_inode(file);
1631
1632 if (unlikely(pos < 0))
1633 return -EINVAL;
1634
1635 if (unlikely((loff_t) (pos + len) < 0))
1636 return -EINVAL;
1637
1638 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1639 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1640 int retval;
1641
1642 retval = locks_mandatory_area(inode, file, pos, end,
1643 write ? F_WRLCK : F_RDLCK);
1644 if (retval < 0)
1645 return retval;
1646 }
1647
1648 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1649 }
1650
1651 /*
1652 * Check that the two inodes are eligible for cloning, the ranges make
1653 * sense, and then flush all dirty data. Caller must ensure that the
1654 * inodes have been locked against any other modifications.
1655 *
1656 * Returns: 0 for "nothing to clone", 1 for "something to clone", or
1657 * the usual negative error code.
1658 */
1659 int vfs_clone_file_prep_inodes(struct inode *inode_in, loff_t pos_in,
1660 struct inode *inode_out, loff_t pos_out,
1661 u64 *len, bool is_dedupe)
1662 {
1663 loff_t bs = inode_out->i_sb->s_blocksize;
1664 loff_t blen;
1665 loff_t isize;
1666 bool same_inode = (inode_in == inode_out);
1667 int ret;
1668
1669 /* Don't touch certain kinds of inodes */
1670 if (IS_IMMUTABLE(inode_out))
1671 return -EPERM;
1672
1673 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1674 return -ETXTBSY;
1675
1676 /* Don't reflink dirs, pipes, sockets... */
1677 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1678 return -EISDIR;
1679 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1680 return -EINVAL;
1681
1682 /* Are we going all the way to the end? */
1683 isize = i_size_read(inode_in);
1684 if (isize == 0)
1685 return 0;
1686
1687 /* Zero length dedupe exits immediately; reflink goes to EOF. */
1688 if (*len == 0) {
1689 if (is_dedupe || pos_in == isize)
1690 return 0;
1691 if (pos_in > isize)
1692 return -EINVAL;
1693 *len = isize - pos_in;
1694 }
1695
1696 /* Ensure offsets don't wrap and the input is inside i_size */
1697 if (pos_in + *len < pos_in || pos_out + *len < pos_out ||
1698 pos_in + *len > isize)
1699 return -EINVAL;
1700
1701 /* Don't allow dedupe past EOF in the dest file */
1702 if (is_dedupe) {
1703 loff_t disize;
1704
1705 disize = i_size_read(inode_out);
1706 if (pos_out >= disize || pos_out + *len > disize)
1707 return -EINVAL;
1708 }
1709
1710 /* If we're linking to EOF, continue to the block boundary. */
1711 if (pos_in + *len == isize)
1712 blen = ALIGN(isize, bs) - pos_in;
1713 else
1714 blen = *len;
1715
1716 /* Only reflink if we're aligned to block boundaries */
1717 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_in + blen, bs) ||
1718 !IS_ALIGNED(pos_out, bs) || !IS_ALIGNED(pos_out + blen, bs))
1719 return -EINVAL;
1720
1721 /* Don't allow overlapped reflink within the same file */
1722 if (same_inode) {
1723 if (pos_out + blen > pos_in && pos_out < pos_in + blen)
1724 return -EINVAL;
1725 }
1726
1727 /* Wait for the completion of any pending IOs on both files */
1728 inode_dio_wait(inode_in);
1729 if (!same_inode)
1730 inode_dio_wait(inode_out);
1731
1732 ret = filemap_write_and_wait_range(inode_in->i_mapping,
1733 pos_in, pos_in + *len - 1);
1734 if (ret)
1735 return ret;
1736
1737 ret = filemap_write_and_wait_range(inode_out->i_mapping,
1738 pos_out, pos_out + *len - 1);
1739 if (ret)
1740 return ret;
1741
1742 /*
1743 * Check that the extents are the same.
1744 */
1745 if (is_dedupe) {
1746 bool is_same = false;
1747
1748 ret = vfs_dedupe_file_range_compare(inode_in, pos_in,
1749 inode_out, pos_out, *len, &is_same);
1750 if (ret)
1751 return ret;
1752 if (!is_same)
1753 return -EBADE;
1754 }
1755
1756 return 1;
1757 }
1758 EXPORT_SYMBOL(vfs_clone_file_prep_inodes);
1759
1760 int vfs_clone_file_range(struct file *file_in, loff_t pos_in,
1761 struct file *file_out, loff_t pos_out, u64 len)
1762 {
1763 struct inode *inode_in = file_inode(file_in);
1764 struct inode *inode_out = file_inode(file_out);
1765 int ret;
1766
1767 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1768 return -EISDIR;
1769 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1770 return -EINVAL;
1771
1772 /*
1773 * FICLONE/FICLONERANGE ioctls enforce that src and dest files are on
1774 * the same mount. Practically, they only need to be on the same file
1775 * system.
1776 */
1777 if (inode_in->i_sb != inode_out->i_sb)
1778 return -EXDEV;
1779
1780 if (!(file_in->f_mode & FMODE_READ) ||
1781 !(file_out->f_mode & FMODE_WRITE) ||
1782 (file_out->f_flags & O_APPEND))
1783 return -EBADF;
1784
1785 if (!file_in->f_op->clone_file_range)
1786 return -EOPNOTSUPP;
1787
1788 ret = clone_verify_area(file_in, pos_in, len, false);
1789 if (ret)
1790 return ret;
1791
1792 ret = clone_verify_area(file_out, pos_out, len, true);
1793 if (ret)
1794 return ret;
1795
1796 if (pos_in + len > i_size_read(inode_in))
1797 return -EINVAL;
1798
1799 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1800 file_out, pos_out, len);
1801 if (!ret) {
1802 fsnotify_access(file_in);
1803 fsnotify_modify(file_out);
1804 }
1805
1806 return ret;
1807 }
1808 EXPORT_SYMBOL(vfs_clone_file_range);
1809
1810 /*
1811 * Read a page's worth of file data into the page cache. Return the page
1812 * locked.
1813 */
1814 static struct page *vfs_dedupe_get_page(struct inode *inode, loff_t offset)
1815 {
1816 struct address_space *mapping;
1817 struct page *page;
1818 pgoff_t n;
1819
1820 n = offset >> PAGE_SHIFT;
1821 mapping = inode->i_mapping;
1822 page = read_mapping_page(mapping, n, NULL);
1823 if (IS_ERR(page))
1824 return page;
1825 if (!PageUptodate(page)) {
1826 put_page(page);
1827 return ERR_PTR(-EIO);
1828 }
1829 lock_page(page);
1830 return page;
1831 }
1832
1833 /*
1834 * Compare extents of two files to see if they are the same.
1835 * Caller must have locked both inodes to prevent write races.
1836 */
1837 int vfs_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
1838 struct inode *dest, loff_t destoff,
1839 loff_t len, bool *is_same)
1840 {
1841 loff_t src_poff;
1842 loff_t dest_poff;
1843 void *src_addr;
1844 void *dest_addr;
1845 struct page *src_page;
1846 struct page *dest_page;
1847 loff_t cmp_len;
1848 bool same;
1849 int error;
1850
1851 error = -EINVAL;
1852 same = true;
1853 while (len) {
1854 src_poff = srcoff & (PAGE_SIZE - 1);
1855 dest_poff = destoff & (PAGE_SIZE - 1);
1856 cmp_len = min(PAGE_SIZE - src_poff,
1857 PAGE_SIZE - dest_poff);
1858 cmp_len = min(cmp_len, len);
1859 if (cmp_len <= 0)
1860 goto out_error;
1861
1862 src_page = vfs_dedupe_get_page(src, srcoff);
1863 if (IS_ERR(src_page)) {
1864 error = PTR_ERR(src_page);
1865 goto out_error;
1866 }
1867 dest_page = vfs_dedupe_get_page(dest, destoff);
1868 if (IS_ERR(dest_page)) {
1869 error = PTR_ERR(dest_page);
1870 unlock_page(src_page);
1871 put_page(src_page);
1872 goto out_error;
1873 }
1874 src_addr = kmap_atomic(src_page);
1875 dest_addr = kmap_atomic(dest_page);
1876
1877 flush_dcache_page(src_page);
1878 flush_dcache_page(dest_page);
1879
1880 if (memcmp(src_addr + src_poff, dest_addr + dest_poff, cmp_len))
1881 same = false;
1882
1883 kunmap_atomic(dest_addr);
1884 kunmap_atomic(src_addr);
1885 unlock_page(dest_page);
1886 unlock_page(src_page);
1887 put_page(dest_page);
1888 put_page(src_page);
1889
1890 if (!same)
1891 break;
1892
1893 srcoff += cmp_len;
1894 destoff += cmp_len;
1895 len -= cmp_len;
1896 }
1897
1898 *is_same = same;
1899 return 0;
1900
1901 out_error:
1902 return error;
1903 }
1904 EXPORT_SYMBOL(vfs_dedupe_file_range_compare);
1905
1906 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
1907 {
1908 struct file_dedupe_range_info *info;
1909 struct inode *src = file_inode(file);
1910 u64 off;
1911 u64 len;
1912 int i;
1913 int ret;
1914 bool is_admin = capable(CAP_SYS_ADMIN);
1915 u16 count = same->dest_count;
1916 struct file *dst_file;
1917 loff_t dst_off;
1918 ssize_t deduped;
1919
1920 if (!(file->f_mode & FMODE_READ))
1921 return -EINVAL;
1922
1923 if (same->reserved1 || same->reserved2)
1924 return -EINVAL;
1925
1926 off = same->src_offset;
1927 len = same->src_length;
1928
1929 ret = -EISDIR;
1930 if (S_ISDIR(src->i_mode))
1931 goto out;
1932
1933 ret = -EINVAL;
1934 if (!S_ISREG(src->i_mode))
1935 goto out;
1936
1937 ret = clone_verify_area(file, off, len, false);
1938 if (ret < 0)
1939 goto out;
1940 ret = 0;
1941
1942 if (off + len > i_size_read(src))
1943 return -EINVAL;
1944
1945 /* pre-format output fields to sane values */
1946 for (i = 0; i < count; i++) {
1947 same->info[i].bytes_deduped = 0ULL;
1948 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
1949 }
1950
1951 for (i = 0, info = same->info; i < count; i++, info++) {
1952 struct inode *dst;
1953 struct fd dst_fd = fdget(info->dest_fd);
1954
1955 dst_file = dst_fd.file;
1956 if (!dst_file) {
1957 info->status = -EBADF;
1958 goto next_loop;
1959 }
1960 dst = file_inode(dst_file);
1961
1962 ret = mnt_want_write_file(dst_file);
1963 if (ret) {
1964 info->status = ret;
1965 goto next_loop;
1966 }
1967
1968 dst_off = info->dest_offset;
1969 ret = clone_verify_area(dst_file, dst_off, len, true);
1970 if (ret < 0) {
1971 info->status = ret;
1972 goto next_file;
1973 }
1974 ret = 0;
1975
1976 if (info->reserved) {
1977 info->status = -EINVAL;
1978 } else if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) {
1979 info->status = -EINVAL;
1980 } else if (file->f_path.mnt != dst_file->f_path.mnt) {
1981 info->status = -EXDEV;
1982 } else if (S_ISDIR(dst->i_mode)) {
1983 info->status = -EISDIR;
1984 } else if (dst_file->f_op->dedupe_file_range == NULL) {
1985 info->status = -EINVAL;
1986 } else {
1987 deduped = dst_file->f_op->dedupe_file_range(file, off,
1988 len, dst_file,
1989 info->dest_offset);
1990 if (deduped == -EBADE)
1991 info->status = FILE_DEDUPE_RANGE_DIFFERS;
1992 else if (deduped < 0)
1993 info->status = deduped;
1994 else
1995 info->bytes_deduped += deduped;
1996 }
1997
1998 next_file:
1999 mnt_drop_write_file(dst_file);
2000 next_loop:
2001 fdput(dst_fd);
2002
2003 if (fatal_signal_pending(current))
2004 goto out;
2005 }
2006
2007 out:
2008 return ret;
2009 }
2010 EXPORT_SYMBOL(vfs_dedupe_file_range);
(deprecated) Btrfs devel tree