]> git.ipfire.org Git - people/arne_f/kernel.git/blob - fs/read_write.c
projects / people / arne_f / kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
bnxt_en: Reset device on RX buffer errors.
[people/arne_f/kernel.git] / fs / read_write.c
1 /*
2 * linux/fs/read_write.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7 #include <linux/slab.h>
8 #include <linux/stat.h>
9 #include <linux/fcntl.h>
10 #include <linux/file.h>
11 #include <linux/uio.h>
12 #include <linux/fsnotify.h>
13 #include <linux/security.h>
14 #include <linux/export.h>
15 #include <linux/syscalls.h>
16 #include <linux/pagemap.h>
17 #include <linux/splice.h>
18 #include <linux/compat.h>
19 #include <linux/mount.h>
20 #include <linux/fs.h>
21 #include "internal.h"
22
23 #include <asm/uaccess.h>
24 #include <asm/unistd.h>
25
26 typedef ssize_t (*io_fn_t)(struct file *, char __user *, size_t, loff_t *);
27 typedef ssize_t (*iter_fn_t)(struct kiocb *, struct iov_iter *);
28
29 const struct file_operations generic_ro_fops = {
30 .llseek = generic_file_llseek,
31 .read_iter = generic_file_read_iter,
32 .mmap = generic_file_readonly_mmap,
33 .splice_read = generic_file_splice_read,
34 };
35
36 EXPORT_SYMBOL(generic_ro_fops);
37
38 static inline int unsigned_offsets(struct file *file)
39 {
40 return file->f_mode & FMODE_UNSIGNED_OFFSET;
41 }
42
43 /**
44 * vfs_setpos - update the file offset for lseek
45 * @file: file structure in question
46 * @offset: file offset to seek to
47 * @maxsize: maximum file size
48 *
49 * This is a low-level filesystem helper for updating the file offset to
50 * the value specified by @offset if the given offset is valid and it is
51 * not equal to the current file offset.
52 *
53 * Return the specified offset on success and -EINVAL on invalid offset.
54 */
55 loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
56 {
57 if (offset < 0 && !unsigned_offsets(file))
58 return -EINVAL;
59 if (offset > maxsize)
60 return -EINVAL;
61
62 if (offset != file->f_pos) {
63 file->f_pos = offset;
64 file->f_version = 0;
65 }
66 return offset;
67 }
68 EXPORT_SYMBOL(vfs_setpos);
69
70 /**
71 * generic_file_llseek_size - generic llseek implementation for regular files
72 * @file: file structure to seek on
73 * @offset: file offset to seek to
74 * @whence: type of seek
75 * @size: max size of this file in file system
76 * @eof: offset used for SEEK_END position
77 *
78 * This is a variant of generic_file_llseek that allows passing in a custom
79 * maximum file size and a custom EOF position, for e.g. hashed directories
80 *
81 * Synchronization:
82 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
83 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
84 * read/writes behave like SEEK_SET against seeks.
85 */
86 loff_t
87 generic_file_llseek_size(struct file *file, loff_t offset, int whence,
88 loff_t maxsize, loff_t eof)
89 {
90 switch (whence) {
91 case SEEK_END:
92 offset += eof;
93 break;
94 case SEEK_CUR:
95 /*
96 * Here we special-case the lseek(fd, 0, SEEK_CUR)
97 * position-querying operation. Avoid rewriting the "same"
98 * f_pos value back to the file because a concurrent read(),
99 * write() or lseek() might have altered it
100 */
101 if (offset == 0)
102 return file->f_pos;
103 /*
104 * f_lock protects against read/modify/write race with other
105 * SEEK_CURs. Note that parallel writes and reads behave
106 * like SEEK_SET.
107 */
108 spin_lock(&file->f_lock);
109 offset = vfs_setpos(file, file->f_pos + offset, maxsize);
110 spin_unlock(&file->f_lock);
111 return offset;
112 case SEEK_DATA:
113 /*
114 * In the generic case the entire file is data, so as long as
115 * offset isn't at the end of the file then the offset is data.
116 */
117 if ((unsigned long long)offset >= eof)
118 return -ENXIO;
119 break;
120 case SEEK_HOLE:
121 /*
122 * There is a virtual hole at the end of the file, so as long as
123 * offset isn't i_size or larger, return i_size.
124 */
125 if ((unsigned long long)offset >= eof)
126 return -ENXIO;
127 offset = eof;
128 break;
129 }
130
131 return vfs_setpos(file, offset, maxsize);
132 }
133 EXPORT_SYMBOL(generic_file_llseek_size);
134
135 /**
136 * generic_file_llseek - generic llseek implementation for regular files
137 * @file: file structure to seek on
138 * @offset: file offset to seek to
139 * @whence: type of seek
140 *
141 * This is a generic implemenation of ->llseek useable for all normal local
142 * filesystems. It just updates the file offset to the value specified by
143 * @offset and @whence.
144 */
145 loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
146 {
147 struct inode *inode = file->f_mapping->host;
148
149 return generic_file_llseek_size(file, offset, whence,
150 inode->i_sb->s_maxbytes,
151 i_size_read(inode));
152 }
153 EXPORT_SYMBOL(generic_file_llseek);
154
155 /**
156 * fixed_size_llseek - llseek implementation for fixed-sized devices
157 * @file: file structure to seek on
158 * @offset: file offset to seek to
159 * @whence: type of seek
160 * @size: size of the file
161 *
162 */
163 loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
164 {
165 switch (whence) {
166 case SEEK_SET: case SEEK_CUR: case SEEK_END:
167 return generic_file_llseek_size(file, offset, whence,
168 size, size);
169 default:
170 return -EINVAL;
171 }
172 }
173 EXPORT_SYMBOL(fixed_size_llseek);
174
175 /**
176 * no_seek_end_llseek - llseek implementation for fixed-sized devices
177 * @file: file structure to seek on
178 * @offset: file offset to seek to
179 * @whence: type of seek
180 *
181 */
182 loff_t no_seek_end_llseek(struct file *file, loff_t offset, int whence)
183 {
184 switch (whence) {
185 case SEEK_SET: case SEEK_CUR:
186 return generic_file_llseek_size(file, offset, whence,
187 OFFSET_MAX, 0);
188 default:
189 return -EINVAL;
190 }
191 }
192 EXPORT_SYMBOL(no_seek_end_llseek);
193
194 /**
195 * no_seek_end_llseek_size - llseek implementation for fixed-sized devices
196 * @file: file structure to seek on
197 * @offset: file offset to seek to
198 * @whence: type of seek
199 * @size: maximal offset allowed
200 *
201 */
202 loff_t no_seek_end_llseek_size(struct file *file, loff_t offset, int whence, loff_t size)
203 {
204 switch (whence) {
205 case SEEK_SET: case SEEK_CUR:
206 return generic_file_llseek_size(file, offset, whence,
207 size, 0);
208 default:
209 return -EINVAL;
210 }
211 }
212 EXPORT_SYMBOL(no_seek_end_llseek_size);
213
214 /**
215 * noop_llseek - No Operation Performed llseek implementation
216 * @file: file structure to seek on
217 * @offset: file offset to seek to
218 * @whence: type of seek
219 *
220 * This is an implementation of ->llseek useable for the rare special case when
221 * userspace expects the seek to succeed but the (device) file is actually not
222 * able to perform the seek. In this case you use noop_llseek() instead of
223 * falling back to the default implementation of ->llseek.
224 */
225 loff_t noop_llseek(struct file *file, loff_t offset, int whence)
226 {
227 return file->f_pos;
228 }
229 EXPORT_SYMBOL(noop_llseek);
230
231 loff_t no_llseek(struct file *file, loff_t offset, int whence)
232 {
233 return -ESPIPE;
234 }
235 EXPORT_SYMBOL(no_llseek);
236
237 loff_t default_llseek(struct file *file, loff_t offset, int whence)
238 {
239 struct inode *inode = file_inode(file);
240 loff_t retval;
241
242 inode_lock(inode);
243 switch (whence) {
244 case SEEK_END:
245 offset += i_size_read(inode);
246 break;
247 case SEEK_CUR:
248 if (offset == 0) {
249 retval = file->f_pos;
250 goto out;
251 }
252 offset += file->f_pos;
253 break;
254 case SEEK_DATA:
255 /*
256 * In the generic case the entire file is data, so as
257 * long as offset isn't at the end of the file then the
258 * offset is data.
259 */
260 if (offset >= inode->i_size) {
261 retval = -ENXIO;
262 goto out;
263 }
264 break;
265 case SEEK_HOLE:
266 /*
267 * There is a virtual hole at the end of the file, so
268 * as long as offset isn't i_size or larger, return
269 * i_size.
270 */
271 if (offset >= inode->i_size) {
272 retval = -ENXIO;
273 goto out;
274 }
275 offset = inode->i_size;
276 break;
277 }
278 retval = -EINVAL;
279 if (offset >= 0 || unsigned_offsets(file)) {
280 if (offset != file->f_pos) {
281 file->f_pos = offset;
282 file->f_version = 0;
283 }
284 retval = offset;
285 }
286 out:
287 inode_unlock(inode);
288 return retval;
289 }
290 EXPORT_SYMBOL(default_llseek);
291
292 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
293 {
294 loff_t (*fn)(struct file *, loff_t, int);
295
296 fn = no_llseek;
297 if (file->f_mode & FMODE_LSEEK) {
298 if (file->f_op->llseek)
299 fn = file->f_op->llseek;
300 }
301 return fn(file, offset, whence);
302 }
303 EXPORT_SYMBOL(vfs_llseek);
304
305 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
306 {
307 off_t retval;
308 struct fd f = fdget_pos(fd);
309 if (!f.file)
310 return -EBADF;
311
312 retval = -EINVAL;
313 if (whence <= SEEK_MAX) {
314 loff_t res = vfs_llseek(f.file, offset, whence);
315 retval = res;
316 if (res != (loff_t)retval)
317 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
318 }
319 fdput_pos(f);
320 return retval;
321 }
322
323 #ifdef CONFIG_COMPAT
324 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
325 {
326 return sys_lseek(fd, offset, whence);
327 }
328 #endif
329
330 #ifdef __ARCH_WANT_SYS_LLSEEK
331 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
332 unsigned long, offset_low, loff_t __user *, result,
333 unsigned int, whence)
334 {
335 int retval;
336 struct fd f = fdget_pos(fd);
337 loff_t offset;
338
339 if (!f.file)
340 return -EBADF;
341
342 retval = -EINVAL;
343 if (whence > SEEK_MAX)
344 goto out_putf;
345
346 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
347 whence);
348
349 retval = (int)offset;
350 if (offset >= 0) {
351 retval = -EFAULT;
352 if (!copy_to_user(result, &offset, sizeof(offset)))
353 retval = 0;
354 }
355 out_putf:
356 fdput_pos(f);
357 return retval;
358 }
359 #endif
360
361 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos)
362 {
363 struct kiocb kiocb;
364 ssize_t ret;
365
366 if (!file->f_op->read_iter)
367 return -EINVAL;
368
369 init_sync_kiocb(&kiocb, file);
370 kiocb.ki_pos = *ppos;
371
372 iter->type |= READ;
373 ret = file->f_op->read_iter(&kiocb, iter);
374 BUG_ON(ret == -EIOCBQUEUED);
375 if (ret > 0)
376 *ppos = kiocb.ki_pos;
377 return ret;
378 }
379 EXPORT_SYMBOL(vfs_iter_read);
380
381 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos)
382 {
383 struct kiocb kiocb;
384 ssize_t ret;
385
386 if (!file->f_op->write_iter)
387 return -EINVAL;
388
389 init_sync_kiocb(&kiocb, file);
390 kiocb.ki_pos = *ppos;
391
392 iter->type |= WRITE;
393 ret = file->f_op->write_iter(&kiocb, iter);
394 BUG_ON(ret == -EIOCBQUEUED);
395 if (ret > 0) {
396 *ppos = kiocb.ki_pos;
397 fsnotify_modify(file);
398 }
399 return ret;
400 }
401 EXPORT_SYMBOL(vfs_iter_write);
402
403 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
404 {
405 struct inode *inode;
406 loff_t pos;
407 int retval = -EINVAL;
408
409 inode = file_inode(file);
410 if (unlikely((ssize_t) count < 0))
411 return retval;
412 pos = *ppos;
413 if (unlikely(pos < 0)) {
414 if (!unsigned_offsets(file))
415 return retval;
416 if (count >= -pos) /* both values are in 0..LLONG_MAX */
417 return -EOVERFLOW;
418 } else if (unlikely((loff_t) (pos + count) < 0)) {
419 if (!unsigned_offsets(file))
420 return retval;
421 }
422
423 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
424 retval = locks_mandatory_area(inode, file, pos, pos + count - 1,
425 read_write == READ ? F_RDLCK : F_WRLCK);
426 if (retval < 0)
427 return retval;
428 }
429 return security_file_permission(file,
430 read_write == READ ? MAY_READ : MAY_WRITE);
431 }
432
433 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
434 {
435 struct iovec iov = { .iov_base = buf, .iov_len = len };
436 struct kiocb kiocb;
437 struct iov_iter iter;
438 ssize_t ret;
439
440 init_sync_kiocb(&kiocb, filp);
441 kiocb.ki_pos = *ppos;
442 iov_iter_init(&iter, READ, &iov, 1, len);
443
444 ret = filp->f_op->read_iter(&kiocb, &iter);
445 BUG_ON(ret == -EIOCBQUEUED);
446 *ppos = kiocb.ki_pos;
447 return ret;
448 }
449
450 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
451 loff_t *pos)
452 {
453 if (file->f_op->read)
454 return file->f_op->read(file, buf, count, pos);
455 else if (file->f_op->read_iter)
456 return new_sync_read(file, buf, count, pos);
457 else
458 return -EINVAL;
459 }
460 EXPORT_SYMBOL(__vfs_read);
461
462 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
463 {
464 ssize_t ret;
465
466 if (!(file->f_mode & FMODE_READ))
467 return -EBADF;
468 if (!(file->f_mode & FMODE_CAN_READ))
469 return -EINVAL;
470 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
471 return -EFAULT;
472
473 ret = rw_verify_area(READ, file, pos, count);
474 if (!ret) {
475 if (count > MAX_RW_COUNT)
476 count = MAX_RW_COUNT;
477 ret = __vfs_read(file, buf, count, pos);
478 if (ret > 0) {
479 fsnotify_access(file);
480 add_rchar(current, ret);
481 }
482 inc_syscr(current);
483 }
484
485 return ret;
486 }
487
488 EXPORT_SYMBOL(vfs_read);
489
490 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
491 {
492 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
493 struct kiocb kiocb;
494 struct iov_iter iter;
495 ssize_t ret;
496
497 init_sync_kiocb(&kiocb, filp);
498 kiocb.ki_pos = *ppos;
499 iov_iter_init(&iter, WRITE, &iov, 1, len);
500
501 ret = filp->f_op->write_iter(&kiocb, &iter);
502 BUG_ON(ret == -EIOCBQUEUED);
503 if (ret > 0)
504 *ppos = kiocb.ki_pos;
505 return ret;
506 }
507
508 ssize_t __vfs_write(struct file *file, const char __user *p, size_t count,
509 loff_t *pos)
510 {
511 if (file->f_op->write)
512 return file->f_op->write(file, p, count, pos);
513 else if (file->f_op->write_iter)
514 return new_sync_write(file, p, count, pos);
515 else
516 return -EINVAL;
517 }
518 EXPORT_SYMBOL(__vfs_write);
519
520 ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
521 {
522 mm_segment_t old_fs;
523 const char __user *p;
524 ssize_t ret;
525
526 if (!(file->f_mode & FMODE_CAN_WRITE))
527 return -EINVAL;
528
529 old_fs = get_fs();
530 set_fs(get_ds());
531 p = (__force const char __user *)buf;
532 if (count > MAX_RW_COUNT)
533 count = MAX_RW_COUNT;
534 ret = __vfs_write(file, p, count, pos);
535 set_fs(old_fs);
536 if (ret > 0) {
537 fsnotify_modify(file);
538 add_wchar(current, ret);
539 }
540 inc_syscw(current);
541 return ret;
542 }
543
544 EXPORT_SYMBOL(__kernel_write);
545
546 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
547 {
548 ssize_t ret;
549
550 if (!(file->f_mode & FMODE_WRITE))
551 return -EBADF;
552 if (!(file->f_mode & FMODE_CAN_WRITE))
553 return -EINVAL;
554 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
555 return -EFAULT;
556
557 ret = rw_verify_area(WRITE, file, pos, count);
558 if (!ret) {
559 if (count > MAX_RW_COUNT)
560 count = MAX_RW_COUNT;
561 file_start_write(file);
562 ret = __vfs_write(file, buf, count, pos);
563 if (ret > 0) {
564 fsnotify_modify(file);
565 add_wchar(current, ret);
566 }
567 inc_syscw(current);
568 file_end_write(file);
569 }
570
571 return ret;
572 }
573
574 EXPORT_SYMBOL(vfs_write);
575
576 static inline loff_t file_pos_read(struct file *file)
577 {
578 return file->f_pos;
579 }
580
581 static inline void file_pos_write(struct file *file, loff_t pos)
582 {
583 file->f_pos = pos;
584 }
585
586 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
587 {
588 struct fd f = fdget_pos(fd);
589 ssize_t ret = -EBADF;
590
591 if (f.file) {
592 loff_t pos = file_pos_read(f.file);
593 ret = vfs_read(f.file, buf, count, &pos);
594 if (ret >= 0)
595 file_pos_write(f.file, pos);
596 fdput_pos(f);
597 }
598 return ret;
599 }
600
601 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
602 size_t, count)
603 {
604 struct fd f = fdget_pos(fd);
605 ssize_t ret = -EBADF;
606
607 if (f.file) {
608 loff_t pos = file_pos_read(f.file);
609 ret = vfs_write(f.file, buf, count, &pos);
610 if (ret >= 0)
611 file_pos_write(f.file, pos);
612 fdput_pos(f);
613 }
614
615 return ret;
616 }
617
618 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
619 size_t, count, loff_t, pos)
620 {
621 struct fd f;
622 ssize_t ret = -EBADF;
623
624 if (pos < 0)
625 return -EINVAL;
626
627 f = fdget(fd);
628 if (f.file) {
629 ret = -ESPIPE;
630 if (f.file->f_mode & FMODE_PREAD)
631 ret = vfs_read(f.file, buf, count, &pos);
632 fdput(f);
633 }
634
635 return ret;
636 }
637
638 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
639 size_t, count, loff_t, pos)
640 {
641 struct fd f;
642 ssize_t ret = -EBADF;
643
644 if (pos < 0)
645 return -EINVAL;
646
647 f = fdget(fd);
648 if (f.file) {
649 ret = -ESPIPE;
650 if (f.file->f_mode & FMODE_PWRITE)
651 ret = vfs_write(f.file, buf, count, &pos);
652 fdput(f);
653 }
654
655 return ret;
656 }
657
658 /*
659 * Reduce an iovec's length in-place. Return the resulting number of segments
660 */
661 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
662 {
663 unsigned long seg = 0;
664 size_t len = 0;
665
666 while (seg < nr_segs) {
667 seg++;
668 if (len + iov->iov_len >= to) {
669 iov->iov_len = to - len;
670 break;
671 }
672 len += iov->iov_len;
673 iov++;
674 }
675 return seg;
676 }
677 EXPORT_SYMBOL(iov_shorten);
678
679 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
680 loff_t *ppos, iter_fn_t fn, int flags)
681 {
682 struct kiocb kiocb;
683 ssize_t ret;
684
685 if (flags & ~(RWF_HIPRI | RWF_DSYNC | RWF_SYNC))
686 return -EOPNOTSUPP;
687
688 init_sync_kiocb(&kiocb, filp);
689 if (flags & RWF_HIPRI)
690 kiocb.ki_flags |= IOCB_HIPRI;
691 if (flags & RWF_DSYNC)
692 kiocb.ki_flags |= IOCB_DSYNC;
693 if (flags & RWF_SYNC)
694 kiocb.ki_flags |= (IOCB_DSYNC | IOCB_SYNC);
695 kiocb.ki_pos = *ppos;
696
697 ret = fn(&kiocb, iter);
698 BUG_ON(ret == -EIOCBQUEUED);
699 *ppos = kiocb.ki_pos;
700 return ret;
701 }
702
703 /* Do it by hand, with file-ops */
704 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
705 loff_t *ppos, io_fn_t fn, int flags)
706 {
707 ssize_t ret = 0;
708
709 if (flags & ~RWF_HIPRI)
710 return -EOPNOTSUPP;
711
712 while (iov_iter_count(iter)) {
713 struct iovec iovec = iov_iter_iovec(iter);
714 ssize_t nr;
715
716 nr = fn(filp, iovec.iov_base, iovec.iov_len, ppos);
717
718 if (nr < 0) {
719 if (!ret)
720 ret = nr;
721 break;
722 }
723 ret += nr;
724 if (nr != iovec.iov_len)
725 break;
726 iov_iter_advance(iter, nr);
727 }
728
729 return ret;
730 }
731
732 /* A write operation does a read from user space and vice versa */
733 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
734
735 /**
736 * rw_copy_check_uvector() - Copy an array of &struct iovec from userspace
737 * into the kernel and check that it is valid.
738 *
739 * @type: One of %CHECK_IOVEC_ONLY, %READ, or %WRITE.
740 * @uvector: Pointer to the userspace array.
741 * @nr_segs: Number of elements in userspace array.
742 * @fast_segs: Number of elements in @fast_pointer.
743 * @fast_pointer: Pointer to (usually small on-stack) kernel array.
744 * @ret_pointer: (output parameter) Pointer to a variable that will point to
745 * either @fast_pointer, a newly allocated kernel array, or NULL,
746 * depending on which array was used.
747 *
748 * This function copies an array of &struct iovec of @nr_segs from
749 * userspace into the kernel and checks that each element is valid (e.g.
750 * it does not point to a kernel address or cause overflow by being too
751 * large, etc.).
752 *
753 * As an optimization, the caller may provide a pointer to a small
754 * on-stack array in @fast_pointer, typically %UIO_FASTIOV elements long
755 * (the size of this array, or 0 if unused, should be given in @fast_segs).
756 *
757 * @ret_pointer will always point to the array that was used, so the
758 * caller must take care not to call kfree() on it e.g. in case the
759 * @fast_pointer array was used and it was allocated on the stack.
760 *
761 * Return: The total number of bytes covered by the iovec array on success
762 * or a negative error code on error.
763 */
764 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
765 unsigned long nr_segs, unsigned long fast_segs,
766 struct iovec *fast_pointer,
767 struct iovec **ret_pointer)
768 {
769 unsigned long seg;
770 ssize_t ret;
771 struct iovec *iov = fast_pointer;
772
773 /*
774 * SuS says "The readv() function *may* fail if the iovcnt argument
775 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
776 * traditionally returned zero for zero segments, so...
777 */
778 if (nr_segs == 0) {
779 ret = 0;
780 goto out;
781 }
782
783 /*
784 * First get the "struct iovec" from user memory and
785 * verify all the pointers
786 */
787 if (nr_segs > UIO_MAXIOV) {
788 ret = -EINVAL;
789 goto out;
790 }
791 if (nr_segs > fast_segs) {
792 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
793 if (iov == NULL) {
794 ret = -ENOMEM;
795 goto out;
796 }
797 }
798 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
799 ret = -EFAULT;
800 goto out;
801 }
802
803 /*
804 * According to the Single Unix Specification we should return EINVAL
805 * if an element length is < 0 when cast to ssize_t or if the
806 * total length would overflow the ssize_t return value of the
807 * system call.
808 *
809 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
810 * overflow case.
811 */
812 ret = 0;
813 for (seg = 0; seg < nr_segs; seg++) {
814 void __user *buf = iov[seg].iov_base;
815 ssize_t len = (ssize_t)iov[seg].iov_len;
816
817 /* see if we we're about to use an invalid len or if
818 * it's about to overflow ssize_t */
819 if (len < 0) {
820 ret = -EINVAL;
821 goto out;
822 }
823 if (type >= 0
824 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
825 ret = -EFAULT;
826 goto out;
827 }
828 if (len > MAX_RW_COUNT - ret) {
829 len = MAX_RW_COUNT - ret;
830 iov[seg].iov_len = len;
831 }
832 ret += len;
833 }
834 out:
835 *ret_pointer = iov;
836 return ret;
837 }
838
839 static ssize_t do_readv_writev(int type, struct file *file,
840 const struct iovec __user * uvector,
841 unsigned long nr_segs, loff_t *pos,
842 int flags)
843 {
844 size_t tot_len;
845 struct iovec iovstack[UIO_FASTIOV];
846 struct iovec *iov = iovstack;
847 struct iov_iter iter;
848 ssize_t ret;
849 io_fn_t fn;
850 iter_fn_t iter_fn;
851
852 ret = import_iovec(type, uvector, nr_segs,
853 ARRAY_SIZE(iovstack), &iov, &iter);
854 if (ret < 0)
855 return ret;
856
857 tot_len = iov_iter_count(&iter);
858 if (!tot_len)
859 goto out;
860 ret = rw_verify_area(type, file, pos, tot_len);
861 if (ret < 0)
862 goto out;
863
864 if (type == READ) {
865 fn = file->f_op->read;
866 iter_fn = file->f_op->read_iter;
867 } else {
868 fn = (io_fn_t)file->f_op->write;
869 iter_fn = file->f_op->write_iter;
870 file_start_write(file);
871 }
872
873 if (iter_fn)
874 ret = do_iter_readv_writev(file, &iter, pos, iter_fn, flags);
875 else
876 ret = do_loop_readv_writev(file, &iter, pos, fn, flags);
877
878 if (type != READ)
879 file_end_write(file);
880
881 out:
882 kfree(iov);
883 if ((ret + (type == READ)) > 0) {
884 if (type == READ)
885 fsnotify_access(file);
886 else
887 fsnotify_modify(file);
888 }
889 return ret;
890 }
891
892 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
893 unsigned long vlen, loff_t *pos, int flags)
894 {
895 if (!(file->f_mode & FMODE_READ))
896 return -EBADF;
897 if (!(file->f_mode & FMODE_CAN_READ))
898 return -EINVAL;
899
900 return do_readv_writev(READ, file, vec, vlen, pos, flags);
901 }
902
903 EXPORT_SYMBOL(vfs_readv);
904
905 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
906 unsigned long vlen, loff_t *pos, int flags)
907 {
908 if (!(file->f_mode & FMODE_WRITE))
909 return -EBADF;
910 if (!(file->f_mode & FMODE_CAN_WRITE))
911 return -EINVAL;
912
913 return do_readv_writev(WRITE, file, vec, vlen, pos, flags);
914 }
915
916 EXPORT_SYMBOL(vfs_writev);
917
918 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
919 unsigned long vlen, int flags)
920 {
921 struct fd f = fdget_pos(fd);
922 ssize_t ret = -EBADF;
923
924 if (f.file) {
925 loff_t pos = file_pos_read(f.file);
926 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
927 if (ret >= 0)
928 file_pos_write(f.file, pos);
929 fdput_pos(f);
930 }
931
932 if (ret > 0)
933 add_rchar(current, ret);
934 inc_syscr(current);
935 return ret;
936 }
937
938 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
939 unsigned long vlen, int flags)
940 {
941 struct fd f = fdget_pos(fd);
942 ssize_t ret = -EBADF;
943
944 if (f.file) {
945 loff_t pos = file_pos_read(f.file);
946 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
947 if (ret >= 0)
948 file_pos_write(f.file, pos);
949 fdput_pos(f);
950 }
951
952 if (ret > 0)
953 add_wchar(current, ret);
954 inc_syscw(current);
955 return ret;
956 }
957
958 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
959 {
960 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
961 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
962 }
963
964 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
965 unsigned long vlen, loff_t pos, int flags)
966 {
967 struct fd f;
968 ssize_t ret = -EBADF;
969
970 if (pos < 0)
971 return -EINVAL;
972
973 f = fdget(fd);
974 if (f.file) {
975 ret = -ESPIPE;
976 if (f.file->f_mode & FMODE_PREAD)
977 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
978 fdput(f);
979 }
980
981 if (ret > 0)
982 add_rchar(current, ret);
983 inc_syscr(current);
984 return ret;
985 }
986
987 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
988 unsigned long vlen, loff_t pos, int flags)
989 {
990 struct fd f;
991 ssize_t ret = -EBADF;
992
993 if (pos < 0)
994 return -EINVAL;
995
996 f = fdget(fd);
997 if (f.file) {
998 ret = -ESPIPE;
999 if (f.file->f_mode & FMODE_PWRITE)
1000 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
1001 fdput(f);
1002 }
1003
1004 if (ret > 0)
1005 add_wchar(current, ret);
1006 inc_syscw(current);
1007 return ret;
1008 }
1009
1010 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1011 unsigned long, vlen)
1012 {
1013 return do_readv(fd, vec, vlen, 0);
1014 }
1015
1016 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1017 unsigned long, vlen)
1018 {
1019 return do_writev(fd, vec, vlen, 0);
1020 }
1021
1022 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1023 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1024 {
1025 loff_t pos = pos_from_hilo(pos_h, pos_l);
1026
1027 return do_preadv(fd, vec, vlen, pos, 0);
1028 }
1029
1030 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1031 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1032 int, flags)
1033 {
1034 loff_t pos = pos_from_hilo(pos_h, pos_l);
1035
1036 if (pos == -1)
1037 return do_readv(fd, vec, vlen, flags);
1038
1039 return do_preadv(fd, vec, vlen, pos, flags);
1040 }
1041
1042 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1043 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1044 {
1045 loff_t pos = pos_from_hilo(pos_h, pos_l);
1046
1047 return do_pwritev(fd, vec, vlen, pos, 0);
1048 }
1049
1050 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1051 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1052 int, flags)
1053 {
1054 loff_t pos = pos_from_hilo(pos_h, pos_l);
1055
1056 if (pos == -1)
1057 return do_writev(fd, vec, vlen, flags);
1058
1059 return do_pwritev(fd, vec, vlen, pos, flags);
1060 }
1061
1062 #ifdef CONFIG_COMPAT
1063
1064 static ssize_t compat_do_readv_writev(int type, struct file *file,
1065 const struct compat_iovec __user *uvector,
1066 unsigned long nr_segs, loff_t *pos,
1067 int flags)
1068 {
1069 compat_ssize_t tot_len;
1070 struct iovec iovstack[UIO_FASTIOV];
1071 struct iovec *iov = iovstack;
1072 struct iov_iter iter;
1073 ssize_t ret;
1074 io_fn_t fn;
1075 iter_fn_t iter_fn;
1076
1077 ret = compat_import_iovec(type, uvector, nr_segs,
1078 UIO_FASTIOV, &iov, &iter);
1079 if (ret < 0)
1080 return ret;
1081
1082 tot_len = iov_iter_count(&iter);
1083 if (!tot_len)
1084 goto out;
1085 ret = rw_verify_area(type, file, pos, tot_len);
1086 if (ret < 0)
1087 goto out;
1088
1089 if (type == READ) {
1090 fn = file->f_op->read;
1091 iter_fn = file->f_op->read_iter;
1092 } else {
1093 fn = (io_fn_t)file->f_op->write;
1094 iter_fn = file->f_op->write_iter;
1095 file_start_write(file);
1096 }
1097
1098 if (iter_fn)
1099 ret = do_iter_readv_writev(file, &iter, pos, iter_fn, flags);
1100 else
1101 ret = do_loop_readv_writev(file, &iter, pos, fn, flags);
1102
1103 if (type != READ)
1104 file_end_write(file);
1105
1106 out:
1107 kfree(iov);
1108 if ((ret + (type == READ)) > 0) {
1109 if (type == READ)
1110 fsnotify_access(file);
1111 else
1112 fsnotify_modify(file);
1113 }
1114 return ret;
1115 }
1116
1117 static size_t compat_readv(struct file *file,
1118 const struct compat_iovec __user *vec,
1119 unsigned long vlen, loff_t *pos, int flags)
1120 {
1121 ssize_t ret = -EBADF;
1122
1123 if (!(file->f_mode & FMODE_READ))
1124 goto out;
1125
1126 ret = -EINVAL;
1127 if (!(file->f_mode & FMODE_CAN_READ))
1128 goto out;
1129
1130 ret = compat_do_readv_writev(READ, file, vec, vlen, pos, flags);
1131
1132 out:
1133 if (ret > 0)
1134 add_rchar(current, ret);
1135 inc_syscr(current);
1136 return ret;
1137 }
1138
1139 static size_t do_compat_readv(compat_ulong_t fd,
1140 const struct compat_iovec __user *vec,
1141 compat_ulong_t vlen, int flags)
1142 {
1143 struct fd f = fdget_pos(fd);
1144 ssize_t ret;
1145 loff_t pos;
1146
1147 if (!f.file)
1148 return -EBADF;
1149 pos = f.file->f_pos;
1150 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1151 if (ret >= 0)
1152 f.file->f_pos = pos;
1153 fdput_pos(f);
1154 return ret;
1155
1156 }
1157
1158 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1159 const struct compat_iovec __user *,vec,
1160 compat_ulong_t, vlen)
1161 {
1162 return do_compat_readv(fd, vec, vlen, 0);
1163 }
1164
1165 static long do_compat_preadv64(unsigned long fd,
1166 const struct compat_iovec __user *vec,
1167 unsigned long vlen, loff_t pos, int flags)
1168 {
1169 struct fd f;
1170 ssize_t ret;
1171
1172 if (pos < 0)
1173 return -EINVAL;
1174 f = fdget(fd);
1175 if (!f.file)
1176 return -EBADF;
1177 ret = -ESPIPE;
1178 if (f.file->f_mode & FMODE_PREAD)
1179 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1180 fdput(f);
1181 return ret;
1182 }
1183
1184 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1185 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1186 const struct compat_iovec __user *,vec,
1187 unsigned long, vlen, loff_t, pos)
1188 {
1189 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1190 }
1191 #endif
1192
1193 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1194 const struct compat_iovec __user *,vec,
1195 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1196 {
1197 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1198
1199 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1200 }
1201
1202 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1203 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1204 const struct compat_iovec __user *,vec,
1205 unsigned long, vlen, loff_t, pos, int, flags)
1206 {
1207 if (pos == -1)
1208 return do_compat_readv(fd, vec, vlen, flags);
1209
1210 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1211 }
1212 #endif
1213
1214 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1215 const struct compat_iovec __user *,vec,
1216 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1217 int, flags)
1218 {
1219 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1220
1221 if (pos == -1)
1222 return do_compat_readv(fd, vec, vlen, flags);
1223
1224 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1225 }
1226
1227 static size_t compat_writev(struct file *file,
1228 const struct compat_iovec __user *vec,
1229 unsigned long vlen, loff_t *pos, int flags)
1230 {
1231 ssize_t ret = -EBADF;
1232
1233 if (!(file->f_mode & FMODE_WRITE))
1234 goto out;
1235
1236 ret = -EINVAL;
1237 if (!(file->f_mode & FMODE_CAN_WRITE))
1238 goto out;
1239
1240 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos, flags);
1241
1242 out:
1243 if (ret > 0)
1244 add_wchar(current, ret);
1245 inc_syscw(current);
1246 return ret;
1247 }
1248
1249 static size_t do_compat_writev(compat_ulong_t fd,
1250 const struct compat_iovec __user* vec,
1251 compat_ulong_t vlen, int flags)
1252 {
1253 struct fd f = fdget_pos(fd);
1254 ssize_t ret;
1255 loff_t pos;
1256
1257 if (!f.file)
1258 return -EBADF;
1259 pos = f.file->f_pos;
1260 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1261 if (ret >= 0)
1262 f.file->f_pos = pos;
1263 fdput_pos(f);
1264 return ret;
1265 }
1266
1267 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1268 const struct compat_iovec __user *, vec,
1269 compat_ulong_t, vlen)
1270 {
1271 return do_compat_writev(fd, vec, vlen, 0);
1272 }
1273
1274 static long do_compat_pwritev64(unsigned long fd,
1275 const struct compat_iovec __user *vec,
1276 unsigned long vlen, loff_t pos, int flags)
1277 {
1278 struct fd f;
1279 ssize_t ret;
1280
1281 if (pos < 0)
1282 return -EINVAL;
1283 f = fdget(fd);
1284 if (!f.file)
1285 return -EBADF;
1286 ret = -ESPIPE;
1287 if (f.file->f_mode & FMODE_PWRITE)
1288 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1289 fdput(f);
1290 return ret;
1291 }
1292
1293 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1294 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1295 const struct compat_iovec __user *,vec,
1296 unsigned long, vlen, loff_t, pos)
1297 {
1298 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1299 }
1300 #endif
1301
1302 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1303 const struct compat_iovec __user *,vec,
1304 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1305 {
1306 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1307
1308 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1309 }
1310
1311 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1312 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1313 const struct compat_iovec __user *,vec,
1314 unsigned long, vlen, loff_t, pos, int, flags)
1315 {
1316 if (pos == -1)
1317 return do_compat_writev(fd, vec, vlen, flags);
1318
1319 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1320 }
1321 #endif
1322
1323 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1324 const struct compat_iovec __user *,vec,
1325 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, int, flags)
1326 {
1327 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1328
1329 if (pos == -1)
1330 return do_compat_writev(fd, vec, vlen, flags);
1331
1332 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1333 }
1334
1335 #endif
1336
1337 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1338 size_t count, loff_t max)
1339 {
1340 struct fd in, out;
1341 struct inode *in_inode, *out_inode;
1342 loff_t pos;
1343 loff_t out_pos;
1344 ssize_t retval;
1345 int fl;
1346
1347 /*
1348 * Get input file, and verify that it is ok..
1349 */
1350 retval = -EBADF;
1351 in = fdget(in_fd);
1352 if (!in.file)
1353 goto out;
1354 if (!(in.file->f_mode & FMODE_READ))
1355 goto fput_in;
1356 retval = -ESPIPE;
1357 if (!ppos) {
1358 pos = in.file->f_pos;
1359 } else {
1360 pos = *ppos;
1361 if (!(in.file->f_mode & FMODE_PREAD))
1362 goto fput_in;
1363 }
1364 retval = rw_verify_area(READ, in.file, &pos, count);
1365 if (retval < 0)
1366 goto fput_in;
1367 if (count > MAX_RW_COUNT)
1368 count = MAX_RW_COUNT;
1369
1370 /*
1371 * Get output file, and verify that it is ok..
1372 */
1373 retval = -EBADF;
1374 out = fdget(out_fd);
1375 if (!out.file)
1376 goto fput_in;
1377 if (!(out.file->f_mode & FMODE_WRITE))
1378 goto fput_out;
1379 retval = -EINVAL;
1380 in_inode = file_inode(in.file);
1381 out_inode = file_inode(out.file);
1382 out_pos = out.file->f_pos;
1383 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1384 if (retval < 0)
1385 goto fput_out;
1386
1387 if (!max)
1388 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1389
1390 if (unlikely(pos + count > max)) {
1391 retval = -EOVERFLOW;
1392 if (pos >= max)
1393 goto fput_out;
1394 count = max - pos;
1395 }
1396
1397 fl = 0;
1398 #if 0
1399 /*
1400 * We need to debate whether we can enable this or not. The
1401 * man page documents EAGAIN return for the output at least,
1402 * and the application is arguably buggy if it doesn't expect
1403 * EAGAIN on a non-blocking file descriptor.
1404 */
1405 if (in.file->f_flags & O_NONBLOCK)
1406 fl = SPLICE_F_NONBLOCK;
1407 #endif
1408 file_start_write(out.file);
1409 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1410 file_end_write(out.file);
1411
1412 if (retval > 0) {
1413 add_rchar(current, retval);
1414 add_wchar(current, retval);
1415 fsnotify_access(in.file);
1416 fsnotify_modify(out.file);
1417 out.file->f_pos = out_pos;
1418 if (ppos)
1419 *ppos = pos;
1420 else
1421 in.file->f_pos = pos;
1422 }
1423
1424 inc_syscr(current);
1425 inc_syscw(current);
1426 if (pos > max)
1427 retval = -EOVERFLOW;
1428
1429 fput_out:
1430 fdput(out);
1431 fput_in:
1432 fdput(in);
1433 out:
1434 return retval;
1435 }
1436
1437 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1438 {
1439 loff_t pos;
1440 off_t off;
1441 ssize_t ret;
1442
1443 if (offset) {
1444 if (unlikely(get_user(off, offset)))
1445 return -EFAULT;
1446 pos = off;
1447 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1448 if (unlikely(put_user(pos, offset)))
1449 return -EFAULT;
1450 return ret;
1451 }
1452
1453 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1454 }
1455
1456 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1457 {
1458 loff_t pos;
1459 ssize_t ret;
1460
1461 if (offset) {
1462 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1463 return -EFAULT;
1464 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1465 if (unlikely(put_user(pos, offset)))
1466 return -EFAULT;
1467 return ret;
1468 }
1469
1470 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1471 }
1472
1473 #ifdef CONFIG_COMPAT
1474 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1475 compat_off_t __user *, offset, compat_size_t, count)
1476 {
1477 loff_t pos;
1478 off_t off;
1479 ssize_t ret;
1480
1481 if (offset) {
1482 if (unlikely(get_user(off, offset)))
1483 return -EFAULT;
1484 pos = off;
1485 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1486 if (unlikely(put_user(pos, offset)))
1487 return -EFAULT;
1488 return ret;
1489 }
1490
1491 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1492 }
1493
1494 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1495 compat_loff_t __user *, offset, compat_size_t, count)
1496 {
1497 loff_t pos;
1498 ssize_t ret;
1499
1500 if (offset) {
1501 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1502 return -EFAULT;
1503 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1504 if (unlikely(put_user(pos, offset)))
1505 return -EFAULT;
1506 return ret;
1507 }
1508
1509 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1510 }
1511 #endif
1512
1513 /*
1514 * copy_file_range() differs from regular file read and write in that it
1515 * specifically allows return partial success. When it does so is up to
1516 * the copy_file_range method.
1517 */
1518 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1519 struct file *file_out, loff_t pos_out,
1520 size_t len, unsigned int flags)
1521 {
1522 struct inode *inode_in = file_inode(file_in);
1523 struct inode *inode_out = file_inode(file_out);
1524 ssize_t ret;
1525
1526 if (flags != 0)
1527 return -EINVAL;
1528
1529 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1530 return -EISDIR;
1531 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1532 return -EINVAL;
1533
1534 ret = rw_verify_area(READ, file_in, &pos_in, len);
1535 if (unlikely(ret))
1536 return ret;
1537
1538 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1539 if (unlikely(ret))
1540 return ret;
1541
1542 if (!(file_in->f_mode & FMODE_READ) ||
1543 !(file_out->f_mode & FMODE_WRITE) ||
1544 (file_out->f_flags & O_APPEND))
1545 return -EBADF;
1546
1547 /* this could be relaxed once a method supports cross-fs copies */
1548 if (inode_in->i_sb != inode_out->i_sb)
1549 return -EXDEV;
1550
1551 if (len == 0)
1552 return 0;
1553
1554 ret = mnt_want_write_file(file_out);
1555 if (ret)
1556 return ret;
1557
1558 ret = -EOPNOTSUPP;
1559 if (file_out->f_op->copy_file_range)
1560 ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out,
1561 pos_out, len, flags);
1562 if (ret == -EOPNOTSUPP)
1563 ret = do_splice_direct(file_in, &pos_in, file_out, &pos_out,
1564 len > MAX_RW_COUNT ? MAX_RW_COUNT : len, 0);
1565
1566 if (ret > 0) {
1567 fsnotify_access(file_in);
1568 add_rchar(current, ret);
1569 fsnotify_modify(file_out);
1570 add_wchar(current, ret);
1571 }
1572 inc_syscr(current);
1573 inc_syscw(current);
1574
1575 mnt_drop_write_file(file_out);
1576
1577 return ret;
1578 }
1579 EXPORT_SYMBOL(vfs_copy_file_range);
1580
1581 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1582 int, fd_out, loff_t __user *, off_out,
1583 size_t, len, unsigned int, flags)
1584 {
1585 loff_t pos_in;
1586 loff_t pos_out;
1587 struct fd f_in;
1588 struct fd f_out;
1589 ssize_t ret = -EBADF;
1590
1591 f_in = fdget(fd_in);
1592 if (!f_in.file)
1593 goto out2;
1594
1595 f_out = fdget(fd_out);
1596 if (!f_out.file)
1597 goto out1;
1598
1599 ret = -EFAULT;
1600 if (off_in) {
1601 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1602 goto out;
1603 } else {
1604 pos_in = f_in.file->f_pos;
1605 }
1606
1607 if (off_out) {
1608 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1609 goto out;
1610 } else {
1611 pos_out = f_out.file->f_pos;
1612 }
1613
1614 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1615 flags);
1616 if (ret > 0) {
1617 pos_in += ret;
1618 pos_out += ret;
1619
1620 if (off_in) {
1621 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1622 ret = -EFAULT;
1623 } else {
1624 f_in.file->f_pos = pos_in;
1625 }
1626
1627 if (off_out) {
1628 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1629 ret = -EFAULT;
1630 } else {
1631 f_out.file->f_pos = pos_out;
1632 }
1633 }
1634
1635 out:
1636 fdput(f_out);
1637 out1:
1638 fdput(f_in);
1639 out2:
1640 return ret;
1641 }
1642
1643 static int clone_verify_area(struct file *file, loff_t pos, u64 len, bool write)
1644 {
1645 struct inode *inode = file_inode(file);
1646
1647 if (unlikely(pos < 0))
1648 return -EINVAL;
1649
1650 if (unlikely((loff_t) (pos + len) < 0))
1651 return -EINVAL;
1652
1653 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1654 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1655 int retval;
1656
1657 retval = locks_mandatory_area(inode, file, pos, end,
1658 write ? F_WRLCK : F_RDLCK);
1659 if (retval < 0)
1660 return retval;
1661 }
1662
1663 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1664 }
1665
1666 int vfs_clone_file_range(struct file *file_in, loff_t pos_in,
1667 struct file *file_out, loff_t pos_out, u64 len)
1668 {
1669 struct inode *inode_in = file_inode(file_in);
1670 struct inode *inode_out = file_inode(file_out);
1671 int ret;
1672
1673 if (inode_in->i_sb != inode_out->i_sb ||
1674 file_in->f_path.mnt != file_out->f_path.mnt)
1675 return -EXDEV;
1676
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 if (!(file_in->f_mode & FMODE_READ) ||
1683 !(file_out->f_mode & FMODE_WRITE) ||
1684 (file_out->f_flags & O_APPEND))
1685 return -EBADF;
1686
1687 if (!file_in->f_op->clone_file_range)
1688 return -EOPNOTSUPP;
1689
1690 ret = clone_verify_area(file_in, pos_in, len, false);
1691 if (ret)
1692 return ret;
1693
1694 ret = clone_verify_area(file_out, pos_out, len, true);
1695 if (ret)
1696 return ret;
1697
1698 if (pos_in + len > i_size_read(inode_in))
1699 return -EINVAL;
1700
1701 ret = mnt_want_write_file(file_out);
1702 if (ret)
1703 return ret;
1704
1705 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1706 file_out, pos_out, len);
1707 if (!ret) {
1708 fsnotify_access(file_in);
1709 fsnotify_modify(file_out);
1710 }
1711
1712 mnt_drop_write_file(file_out);
1713 return ret;
1714 }
1715 EXPORT_SYMBOL(vfs_clone_file_range);
1716
1717 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
1718 {
1719 struct file_dedupe_range_info *info;
1720 struct inode *src = file_inode(file);
1721 u64 off;
1722 u64 len;
1723 int i;
1724 int ret;
1725 bool is_admin = capable(CAP_SYS_ADMIN);
1726 u16 count = same->dest_count;
1727 struct file *dst_file;
1728 loff_t dst_off;
1729 ssize_t deduped;
1730
1731 if (!(file->f_mode & FMODE_READ))
1732 return -EINVAL;
1733
1734 if (same->reserved1 || same->reserved2)
1735 return -EINVAL;
1736
1737 off = same->src_offset;
1738 len = same->src_length;
1739
1740 ret = -EISDIR;
1741 if (S_ISDIR(src->i_mode))
1742 goto out;
1743
1744 ret = -EINVAL;
1745 if (!S_ISREG(src->i_mode))
1746 goto out;
1747
1748 ret = clone_verify_area(file, off, len, false);
1749 if (ret < 0)
1750 goto out;
1751 ret = 0;
1752
1753 /* pre-format output fields to sane values */
1754 for (i = 0; i < count; i++) {
1755 same->info[i].bytes_deduped = 0ULL;
1756 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
1757 }
1758
1759 for (i = 0, info = same->info; i < count; i++, info++) {
1760 struct inode *dst;
1761 struct fd dst_fd = fdget(info->dest_fd);
1762
1763 dst_file = dst_fd.file;
1764 if (!dst_file) {
1765 info->status = -EBADF;
1766 goto next_loop;
1767 }
1768 dst = file_inode(dst_file);
1769
1770 ret = mnt_want_write_file(dst_file);
1771 if (ret) {
1772 info->status = ret;
1773 goto next_loop;
1774 }
1775
1776 dst_off = info->dest_offset;
1777 ret = clone_verify_area(dst_file, dst_off, len, true);
1778 if (ret < 0) {
1779 info->status = ret;
1780 goto next_file;
1781 }
1782 ret = 0;
1783
1784 if (info->reserved) {
1785 info->status = -EINVAL;
1786 } else if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) {
1787 info->status = -EINVAL;
1788 } else if (file->f_path.mnt != dst_file->f_path.mnt) {
1789 info->status = -EXDEV;
1790 } else if (S_ISDIR(dst->i_mode)) {
1791 info->status = -EISDIR;
1792 } else if (dst_file->f_op->dedupe_file_range == NULL) {
1793 info->status = -EINVAL;
1794 } else {
1795 deduped = dst_file->f_op->dedupe_file_range(file, off,
1796 len, dst_file,
1797 info->dest_offset);
1798 if (deduped == -EBADE)
1799 info->status = FILE_DEDUPE_RANGE_DIFFERS;
1800 else if (deduped < 0)
1801 info->status = deduped;
1802 else
1803 info->bytes_deduped += deduped;
1804 }
1805
1806 next_file:
1807 mnt_drop_write_file(dst_file);
1808 next_loop:
1809 fdput(dst_fd);
1810
1811 if (fatal_signal_pending(current))
1812 goto out;
1813 }
1814
1815 out:
1816 return ret;
1817 }
1818 EXPORT_SYMBOL(vfs_dedupe_file_range);
Arne's tree of linux kernel.