?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
mnt: Protect the mountpoint hashtable with mount_lock
[cris-mirror.git] / fs / read_write.c
blob5816d4c4cab09c22274c2cdc40f4176f16426f0a
1 /*
2 * linux/fs/read_write.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7 #include <linux/slab.h>
8 #include <linux/stat.h>
9 #include <linux/fcntl.h>
10 #include <linux/file.h>
11 #include <linux/uio.h>
12 #include <linux/fsnotify.h>
13 #include <linux/security.h>
14 #include <linux/export.h>
15 #include <linux/syscalls.h>
16 #include <linux/pagemap.h>
17 #include <linux/splice.h>
18 #include <linux/compat.h>
19 #include <linux/mount.h>
20 #include <linux/fs.h>
21 #include "internal.h"
22
23 #include <linux/uaccess.h>
24 #include <asm/unistd.h>
25
26 typedef ssize_t (*io_fn_t)(struct file *, char __user *, size_t, loff_t *);
27 typedef ssize_t (*iter_fn_t)(struct kiocb *, struct iov_iter *);
28
29 const struct file_operations generic_ro_fops = {
30 .llseek = generic_file_llseek,
31 .read_iter = generic_file_read_iter,
32 .mmap = generic_file_readonly_mmap,
33 .splice_read = generic_file_splice_read,
34 };
35
36 EXPORT_SYMBOL(generic_ro_fops);
37
38 static inline int unsigned_offsets(struct file *file)
39 {
40 return file->f_mode & FMODE_UNSIGNED_OFFSET;
41 }
42
43 /**
44 * vfs_setpos - update the file offset for lseek
45 * @file: file structure in question
46 * @offset: file offset to seek to
47 * @maxsize: maximum file size
48 *
49 * This is a low-level filesystem helper for updating the file offset to
50 * the value specified by @offset if the given offset is valid and it is
51 * not equal to the current file offset.
52 *
53 * Return the specified offset on success and -EINVAL on invalid offset.
54 */
55 loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
56 {
57 if (offset < 0 && !unsigned_offsets(file))
58 return -EINVAL;
59 if (offset > maxsize)
60 return -EINVAL;
61
62 if (offset != file->f_pos) {
63 file->f_pos = offset;
64 file->f_version = 0;
65 }
66 return offset;
67 }
68 EXPORT_SYMBOL(vfs_setpos);
69
70 /**
71 * generic_file_llseek_size - generic llseek implementation for regular files
72 * @file: file structure to seek on
73 * @offset: file offset to seek to
74 * @whence: type of seek
75 * @size: max size of this file in file system
76 * @eof: offset used for SEEK_END position
77 *
78 * This is a variant of generic_file_llseek that allows passing in a custom
79 * maximum file size and a custom EOF position, for e.g. hashed directories
80 *
81 * Synchronization:
82 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
83 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
84 * read/writes behave like SEEK_SET against seeks.
85 */
86 loff_t
87 generic_file_llseek_size(struct file *file, loff_t offset, int whence,
88 loff_t maxsize, loff_t eof)
89 {
90 switch (whence) {
91 case SEEK_END:
92 offset += eof;
93 break;
94 case SEEK_CUR:
95 /*
96 * Here we special-case the lseek(fd, 0, SEEK_CUR)
97 * position-querying operation. Avoid rewriting the "same"
98 * f_pos value back to the file because a concurrent read(),
99 * write() or lseek() might have altered it
100 */
101 if (offset == 0)
102 return file->f_pos;
103 /*
104 * f_lock protects against read/modify/write race with other
105 * SEEK_CURs. Note that parallel writes and reads behave
106 * like SEEK_SET.
107 */
108 spin_lock(&file->f_lock);
109 offset = vfs_setpos(file, file->f_pos + offset, maxsize);
110 spin_unlock(&file->f_lock);
111 return offset;
112 case SEEK_DATA:
113 /*
114 * In the generic case the entire file is data, so as long as
115 * offset isn't at the end of the file then the offset is data.
116 */
117 if (offset >= eof)
118 return -ENXIO;
119 break;
120 case SEEK_HOLE:
121 /*
122 * There is a virtual hole at the end of the file, so as long as
123 * offset isn't i_size or larger, return i_size.
124 */
125 if (offset >= eof)
126 return -ENXIO;
127 offset = eof;
128 break;
129 }
130
131 return vfs_setpos(file, offset, maxsize);
132 }
133 EXPORT_SYMBOL(generic_file_llseek_size);
134
135 /**
136 * generic_file_llseek - generic llseek implementation for regular files
137 * @file: file structure to seek on
138 * @offset: file offset to seek to
139 * @whence: type of seek
140 *
141 * This is a generic implemenation of ->llseek useable for all normal local
142 * filesystems. It just updates the file offset to the value specified by
143 * @offset and @whence.
144 */
145 loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
146 {
147 struct inode *inode = file->f_mapping->host;
148
149 return generic_file_llseek_size(file, offset, whence,
150 inode->i_sb->s_maxbytes,
151 i_size_read(inode));
152 }
153 EXPORT_SYMBOL(generic_file_llseek);
154
155 /**
156 * fixed_size_llseek - llseek implementation for fixed-sized devices
157 * @file: file structure to seek on
158 * @offset: file offset to seek to
159 * @whence: type of seek
160 * @size: size of the file
161 *
162 */
163 loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
164 {
165 switch (whence) {
166 case SEEK_SET: case SEEK_CUR: case SEEK_END:
167 return generic_file_llseek_size(file, offset, whence,
168 size, size);
169 default:
170 return -EINVAL;
171 }
172 }
173 EXPORT_SYMBOL(fixed_size_llseek);
174
175 /**
176 * no_seek_end_llseek - llseek implementation for fixed-sized devices
177 * @file: file structure to seek on
178 * @offset: file offset to seek to
179 * @whence: type of seek
180 *
181 */
182 loff_t no_seek_end_llseek(struct file *file, loff_t offset, int whence)
183 {
184 switch (whence) {
185 case SEEK_SET: case SEEK_CUR:
186 return generic_file_llseek_size(file, offset, whence,
187 OFFSET_MAX, 0);
188 default:
189 return -EINVAL;
190 }
191 }
192 EXPORT_SYMBOL(no_seek_end_llseek);
193
194 /**
195 * no_seek_end_llseek_size - llseek implementation for fixed-sized devices
196 * @file: file structure to seek on
197 * @offset: file offset to seek to
198 * @whence: type of seek
199 * @size: maximal offset allowed
200 *
201 */
202 loff_t no_seek_end_llseek_size(struct file *file, loff_t offset, int whence, loff_t size)
203 {
204 switch (whence) {
205 case SEEK_SET: case SEEK_CUR:
206 return generic_file_llseek_size(file, offset, whence,
207 size, 0);
208 default:
209 return -EINVAL;
210 }
211 }
212 EXPORT_SYMBOL(no_seek_end_llseek_size);
213
214 /**
215 * noop_llseek - No Operation Performed llseek implementation
216 * @file: file structure to seek on
217 * @offset: file offset to seek to
218 * @whence: type of seek
219 *
220 * This is an implementation of ->llseek useable for the rare special case when
221 * userspace expects the seek to succeed but the (device) file is actually not
222 * able to perform the seek. In this case you use noop_llseek() instead of
223 * falling back to the default implementation of ->llseek.
224 */
225 loff_t noop_llseek(struct file *file, loff_t offset, int whence)
226 {
227 return file->f_pos;
228 }
229 EXPORT_SYMBOL(noop_llseek);
230
231 loff_t no_llseek(struct file *file, loff_t offset, int whence)
232 {
233 return -ESPIPE;
234 }
235 EXPORT_SYMBOL(no_llseek);
236
237 loff_t default_llseek(struct file *file, loff_t offset, int whence)
238 {
239 struct inode *inode = file_inode(file);
240 loff_t retval;
241
242 inode_lock(inode);
243 switch (whence) {
244 case SEEK_END:
245 offset += i_size_read(inode);
246 break;
247 case SEEK_CUR:
248 if (offset == 0) {
249 retval = file->f_pos;
250 goto out;
251 }
252 offset += file->f_pos;
253 break;
254 case SEEK_DATA:
255 /*
256 * In the generic case the entire file is data, so as
257 * long as offset isn't at the end of the file then the
258 * offset is data.
259 */
260 if (offset >= inode->i_size) {
261 retval = -ENXIO;
262 goto out;
263 }
264 break;
265 case SEEK_HOLE:
266 /*
267 * There is a virtual hole at the end of the file, so
268 * as long as offset isn't i_size or larger, return
269 * i_size.
270 */
271 if (offset >= inode->i_size) {
272 retval = -ENXIO;
273 goto out;
274 }
275 offset = inode->i_size;
276 break;
277 }
278 retval = -EINVAL;
279 if (offset >= 0 || unsigned_offsets(file)) {
280 if (offset != file->f_pos) {
281 file->f_pos = offset;
282 file->f_version = 0;
283 }
284 retval = offset;
285 }
286 out:
287 inode_unlock(inode);
288 return retval;
289 }
290 EXPORT_SYMBOL(default_llseek);
291
292 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
293 {
294 loff_t (*fn)(struct file *, loff_t, int);
295
296 fn = no_llseek;
297 if (file->f_mode & FMODE_LSEEK) {
298 if (file->f_op->llseek)
299 fn = file->f_op->llseek;
300 }
301 return fn(file, offset, whence);
302 }
303 EXPORT_SYMBOL(vfs_llseek);
304
305 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
306 {
307 off_t retval;
308 struct fd f = fdget_pos(fd);
309 if (!f.file)
310 return -EBADF;
311
312 retval = -EINVAL;
313 if (whence <= SEEK_MAX) {
314 loff_t res = vfs_llseek(f.file, offset, whence);
315 retval = res;
316 if (res != (loff_t)retval)
317 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
318 }
319 fdput_pos(f);
320 return retval;
321 }
322
323 #ifdef CONFIG_COMPAT
324 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
325 {
326 return sys_lseek(fd, offset, whence);
327 }
328 #endif
329
330 #ifdef __ARCH_WANT_SYS_LLSEEK
331 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
332 unsigned long, offset_low, loff_t __user *, result,
333 unsigned int, whence)
334 {
335 int retval;
336 struct fd f = fdget_pos(fd);
337 loff_t offset;
338
339 if (!f.file)
340 return -EBADF;
341
342 retval = -EINVAL;
343 if (whence > SEEK_MAX)
344 goto out_putf;
345
346 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
347 whence);
348
349 retval = (int)offset;
350 if (offset >= 0) {
351 retval = -EFAULT;
352 if (!copy_to_user(result, &offset, sizeof(offset)))
353 retval = 0;
354 }
355 out_putf:
356 fdput_pos(f);
357 return retval;
358 }
359 #endif
360
361 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos)
362 {
363 struct kiocb kiocb;
364 ssize_t ret;
365
366 if (!file->f_op->read_iter)
367 return -EINVAL;
368
369 init_sync_kiocb(&kiocb, file);
370 kiocb.ki_pos = *ppos;
371
372 iter->type |= READ;
373 ret = file->f_op->read_iter(&kiocb, iter);
374 BUG_ON(ret == -EIOCBQUEUED);
375 if (ret > 0)
376 *ppos = kiocb.ki_pos;
377 return ret;
378 }
379 EXPORT_SYMBOL(vfs_iter_read);
380
381 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos)
382 {
383 struct kiocb kiocb;
384 ssize_t ret;
385
386 if (!file->f_op->write_iter)
387 return -EINVAL;
388
389 init_sync_kiocb(&kiocb, file);
390 kiocb.ki_pos = *ppos;
391
392 iter->type |= WRITE;
393 ret = file->f_op->write_iter(&kiocb, iter);
394 BUG_ON(ret == -EIOCBQUEUED);
395 if (ret > 0)
396 *ppos = kiocb.ki_pos;
397 return ret;
398 }
399 EXPORT_SYMBOL(vfs_iter_write);
400
401 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
402 {
403 struct inode *inode;
404 loff_t pos;
405 int retval = -EINVAL;
406
407 inode = file_inode(file);
408 if (unlikely((ssize_t) count < 0))
409 return retval;
410 pos = *ppos;
411 if (unlikely(pos < 0)) {
412 if (!unsigned_offsets(file))
413 return retval;
414 if (count >= -pos) /* both values are in 0..LLONG_MAX */
415 return -EOVERFLOW;
416 } else if (unlikely((loff_t) (pos + count) < 0)) {
417 if (!unsigned_offsets(file))
418 return retval;
419 }
420
421 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
422 retval = locks_mandatory_area(inode, file, pos, pos + count - 1,
423 read_write == READ ? F_RDLCK : F_WRLCK);
424 if (retval < 0)
425 return retval;
426 }
427 return security_file_permission(file,
428 read_write == READ ? MAY_READ : MAY_WRITE);
429 }
430
431 static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
432 {
433 struct iovec iov = { .iov_base = buf, .iov_len = len };
434 struct kiocb kiocb;
435 struct iov_iter iter;
436 ssize_t ret;
437
438 init_sync_kiocb(&kiocb, filp);
439 kiocb.ki_pos = *ppos;
440 iov_iter_init(&iter, READ, &iov, 1, len);
441
442 ret = filp->f_op->read_iter(&kiocb, &iter);
443 BUG_ON(ret == -EIOCBQUEUED);
444 *ppos = kiocb.ki_pos;
445 return ret;
446 }
447
448 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
449 loff_t *pos)
450 {
451 if (file->f_op->read)
452 return file->f_op->read(file, buf, count, pos);
453 else if (file->f_op->read_iter)
454 return new_sync_read(file, buf, count, pos);
455 else
456 return -EINVAL;
457 }
458 EXPORT_SYMBOL(__vfs_read);
459
460 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
461 {
462 ssize_t ret;
463
464 if (!(file->f_mode & FMODE_READ))
465 return -EBADF;
466 if (!(file->f_mode & FMODE_CAN_READ))
467 return -EINVAL;
468 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
469 return -EFAULT;
470
471 ret = rw_verify_area(READ, file, pos, count);
472 if (!ret) {
473 if (count > MAX_RW_COUNT)
474 count = MAX_RW_COUNT;
475 ret = __vfs_read(file, buf, count, pos);
476 if (ret > 0) {
477 fsnotify_access(file);
478 add_rchar(current, ret);
479 }
480 inc_syscr(current);
481 }
482
483 return ret;
484 }
485
486 EXPORT_SYMBOL(vfs_read);
487
488 static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
489 {
490 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
491 struct kiocb kiocb;
492 struct iov_iter iter;
493 ssize_t ret;
494
495 init_sync_kiocb(&kiocb, filp);
496 kiocb.ki_pos = *ppos;
497 iov_iter_init(&iter, WRITE, &iov, 1, len);
498
499 ret = filp->f_op->write_iter(&kiocb, &iter);
500 BUG_ON(ret == -EIOCBQUEUED);
501 if (ret > 0)
502 *ppos = kiocb.ki_pos;
503 return ret;
504 }
505
506 ssize_t __vfs_write(struct file *file, const char __user *p, size_t count,
507 loff_t *pos)
508 {
509 if (file->f_op->write)
510 return file->f_op->write(file, p, count, pos);
511 else if (file->f_op->write_iter)
512 return new_sync_write(file, p, count, pos);
513 else
514 return -EINVAL;
515 }
516 EXPORT_SYMBOL(__vfs_write);
517
518 ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
519 {
520 mm_segment_t old_fs;
521 const char __user *p;
522 ssize_t ret;
523
524 if (!(file->f_mode & FMODE_CAN_WRITE))
525 return -EINVAL;
526
527 old_fs = get_fs();
528 set_fs(get_ds());
529 p = (__force const char __user *)buf;
530 if (count > MAX_RW_COUNT)
531 count = MAX_RW_COUNT;
532 ret = __vfs_write(file, p, count, pos);
533 set_fs(old_fs);
534 if (ret > 0) {
535 fsnotify_modify(file);
536 add_wchar(current, ret);
537 }
538 inc_syscw(current);
539 return ret;
540 }
541
542 EXPORT_SYMBOL(__kernel_write);
543
544 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
545 {
546 ssize_t ret;
547
548 if (!(file->f_mode & FMODE_WRITE))
549 return -EBADF;
550 if (!(file->f_mode & FMODE_CAN_WRITE))
551 return -EINVAL;
552 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
553 return -EFAULT;
554
555 ret = rw_verify_area(WRITE, file, pos, count);
556 if (!ret) {
557 if (count > MAX_RW_COUNT)
558 count = MAX_RW_COUNT;
559 file_start_write(file);
560 ret = __vfs_write(file, buf, count, pos);
561 if (ret > 0) {
562 fsnotify_modify(file);
563 add_wchar(current, ret);
564 }
565 inc_syscw(current);
566 file_end_write(file);
567 }
568
569 return ret;
570 }
571
572 EXPORT_SYMBOL(vfs_write);
573
574 static inline loff_t file_pos_read(struct file *file)
575 {
576 return file->f_pos;
577 }
578
579 static inline void file_pos_write(struct file *file, loff_t pos)
580 {
581 file->f_pos = pos;
582 }
583
584 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
585 {
586 struct fd f = fdget_pos(fd);
587 ssize_t ret = -EBADF;
588
589 if (f.file) {
590 loff_t pos = file_pos_read(f.file);
591 ret = vfs_read(f.file, buf, count, &pos);
592 if (ret >= 0)
593 file_pos_write(f.file, pos);
594 fdput_pos(f);
595 }
596 return ret;
597 }
598
599 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
600 size_t, count)
601 {
602 struct fd f = fdget_pos(fd);
603 ssize_t ret = -EBADF;
604
605 if (f.file) {
606 loff_t pos = file_pos_read(f.file);
607 ret = vfs_write(f.file, buf, count, &pos);
608 if (ret >= 0)
609 file_pos_write(f.file, pos);
610 fdput_pos(f);
611 }
612
613 return ret;
614 }
615
616 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
617 size_t, count, loff_t, pos)
618 {
619 struct fd f;
620 ssize_t ret = -EBADF;
621
622 if (pos < 0)
623 return -EINVAL;
624
625 f = fdget(fd);
626 if (f.file) {
627 ret = -ESPIPE;
628 if (f.file->f_mode & FMODE_PREAD)
629 ret = vfs_read(f.file, buf, count, &pos);
630 fdput(f);
631 }
632
633 return ret;
634 }
635
636 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
637 size_t, count, loff_t, pos)
638 {
639 struct fd f;
640 ssize_t ret = -EBADF;
641
642 if (pos < 0)
643 return -EINVAL;
644
645 f = fdget(fd);
646 if (f.file) {
647 ret = -ESPIPE;
648 if (f.file->f_mode & FMODE_PWRITE)
649 ret = vfs_write(f.file, buf, count, &pos);
650 fdput(f);
651 }
652
653 return ret;
654 }
655
656 /*
657 * Reduce an iovec's length in-place. Return the resulting number of segments
658 */
659 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
660 {
661 unsigned long seg = 0;
662 size_t len = 0;
663
664 while (seg < nr_segs) {
665 seg++;
666 if (len + iov->iov_len >= to) {
667 iov->iov_len = to - len;
668 break;
669 }
670 len += iov->iov_len;
671 iov++;
672 }
673 return seg;
674 }
675 EXPORT_SYMBOL(iov_shorten);
676
677 static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
678 loff_t *ppos, iter_fn_t fn, int flags)
679 {
680 struct kiocb kiocb;
681 ssize_t ret;
682
683 if (flags & ~(RWF_HIPRI | RWF_DSYNC | RWF_SYNC))
684 return -EOPNOTSUPP;
685
686 init_sync_kiocb(&kiocb, filp);
687 if (flags & RWF_HIPRI)
688 kiocb.ki_flags |= IOCB_HIPRI;
689 if (flags & RWF_DSYNC)
690 kiocb.ki_flags |= IOCB_DSYNC;
691 if (flags & RWF_SYNC)
692 kiocb.ki_flags |= (IOCB_DSYNC | IOCB_SYNC);
693 kiocb.ki_pos = *ppos;
694
695 ret = fn(&kiocb, iter);
696 BUG_ON(ret == -EIOCBQUEUED);
697 *ppos = kiocb.ki_pos;
698 return ret;
699 }
700
701 /* Do it by hand, with file-ops */
702 static ssize_t do_loop_readv_writev(struct file *filp, struct iov_iter *iter,
703 loff_t *ppos, io_fn_t fn, int flags)
704 {
705 ssize_t ret = 0;
706
707 if (flags & ~RWF_HIPRI)
708 return -EOPNOTSUPP;
709
710 while (iov_iter_count(iter)) {
711 struct iovec iovec = iov_iter_iovec(iter);
712 ssize_t nr;
713
714 nr = fn(filp, iovec.iov_base, iovec.iov_len, ppos);
715
716 if (nr < 0) {
717 if (!ret)
718 ret = nr;
719 break;
720 }
721 ret += nr;
722 if (nr != iovec.iov_len)
723 break;
724 iov_iter_advance(iter, nr);
725 }
726
727 return ret;
728 }
729
730 /* A write operation does a read from user space and vice versa */
731 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
732
733 /**
734 * rw_copy_check_uvector() - Copy an array of &struct iovec from userspace
735 * into the kernel and check that it is valid.
736 *
737 * @type: One of %CHECK_IOVEC_ONLY, %READ, or %WRITE.
738 * @uvector: Pointer to the userspace array.
739 * @nr_segs: Number of elements in userspace array.
740 * @fast_segs: Number of elements in @fast_pointer.
741 * @fast_pointer: Pointer to (usually small on-stack) kernel array.
742 * @ret_pointer: (output parameter) Pointer to a variable that will point to
743 * either @fast_pointer, a newly allocated kernel array, or NULL,
744 * depending on which array was used.
745 *
746 * This function copies an array of &struct iovec of @nr_segs from
747 * userspace into the kernel and checks that each element is valid (e.g.
748 * it does not point to a kernel address or cause overflow by being too
749 * large, etc.).
750 *
751 * As an optimization, the caller may provide a pointer to a small
752 * on-stack array in @fast_pointer, typically %UIO_FASTIOV elements long
753 * (the size of this array, or 0 if unused, should be given in @fast_segs).
754 *
755 * @ret_pointer will always point to the array that was used, so the
756 * caller must take care not to call kfree() on it e.g. in case the
757 * @fast_pointer array was used and it was allocated on the stack.
758 *
759 * Return: The total number of bytes covered by the iovec array on success
760 * or a negative error code on error.
761 */
762 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
763 unsigned long nr_segs, unsigned long fast_segs,
764 struct iovec *fast_pointer,
765 struct iovec **ret_pointer)
766 {
767 unsigned long seg;
768 ssize_t ret;
769 struct iovec *iov = fast_pointer;
770
771 /*
772 * SuS says "The readv() function *may* fail if the iovcnt argument
773 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
774 * traditionally returned zero for zero segments, so...
775 */
776 if (nr_segs == 0) {
777 ret = 0;
778 goto out;
779 }
780
781 /*
782 * First get the "struct iovec" from user memory and
783 * verify all the pointers
784 */
785 if (nr_segs > UIO_MAXIOV) {
786 ret = -EINVAL;
787 goto out;
788 }
789 if (nr_segs > fast_segs) {
790 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
791 if (iov == NULL) {
792 ret = -ENOMEM;
793 goto out;
794 }
795 }
796 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
797 ret = -EFAULT;
798 goto out;
799 }
800
801 /*
802 * According to the Single Unix Specification we should return EINVAL
803 * if an element length is < 0 when cast to ssize_t or if the
804 * total length would overflow the ssize_t return value of the
805 * system call.
806 *
807 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
808 * overflow case.
809 */
810 ret = 0;
811 for (seg = 0; seg < nr_segs; seg++) {
812 void __user *buf = iov[seg].iov_base;
813 ssize_t len = (ssize_t)iov[seg].iov_len;
814
815 /* see if we we're about to use an invalid len or if
816 * it's about to overflow ssize_t */
817 if (len < 0) {
818 ret = -EINVAL;
819 goto out;
820 }
821 if (type >= 0
822 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
823 ret = -EFAULT;
824 goto out;
825 }
826 if (len > MAX_RW_COUNT - ret) {
827 len = MAX_RW_COUNT - ret;
828 iov[seg].iov_len = len;
829 }
830 ret += len;
831 }
832 out:
833 *ret_pointer = iov;
834 return ret;
835 }
836
837 static ssize_t do_readv_writev(int type, struct file *file,
838 const struct iovec __user * uvector,
839 unsigned long nr_segs, loff_t *pos,
840 int flags)
841 {
842 size_t tot_len;
843 struct iovec iovstack[UIO_FASTIOV];
844 struct iovec *iov = iovstack;
845 struct iov_iter iter;
846 ssize_t ret;
847 io_fn_t fn;
848 iter_fn_t iter_fn;
849
850 ret = import_iovec(type, uvector, nr_segs,
851 ARRAY_SIZE(iovstack), &iov, &iter);
852 if (ret < 0)
853 return ret;
854
855 tot_len = iov_iter_count(&iter);
856 if (!tot_len)
857 goto out;
858 ret = rw_verify_area(type, file, pos, tot_len);
859 if (ret < 0)
860 goto out;
861
862 if (type == READ) {
863 fn = file->f_op->read;
864 iter_fn = file->f_op->read_iter;
865 } else {
866 fn = (io_fn_t)file->f_op->write;
867 iter_fn = file->f_op->write_iter;
868 file_start_write(file);
869 }
870
871 if (iter_fn)
872 ret = do_iter_readv_writev(file, &iter, pos, iter_fn, flags);
873 else
874 ret = do_loop_readv_writev(file, &iter, pos, fn, flags);
875
876 if (type != READ)
877 file_end_write(file);
878
879 out:
880 kfree(iov);
881 if ((ret + (type == READ)) > 0) {
882 if (type == READ)
883 fsnotify_access(file);
884 else
885 fsnotify_modify(file);
886 }
887 return ret;
888 }
889
890 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
891 unsigned long vlen, loff_t *pos, int flags)
892 {
893 if (!(file->f_mode & FMODE_READ))
894 return -EBADF;
895 if (!(file->f_mode & FMODE_CAN_READ))
896 return -EINVAL;
897
898 return do_readv_writev(READ, file, vec, vlen, pos, flags);
899 }
900
901 EXPORT_SYMBOL(vfs_readv);
902
903 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
904 unsigned long vlen, loff_t *pos, int flags)
905 {
906 if (!(file->f_mode & FMODE_WRITE))
907 return -EBADF;
908 if (!(file->f_mode & FMODE_CAN_WRITE))
909 return -EINVAL;
910
911 return do_readv_writev(WRITE, file, vec, vlen, pos, flags);
912 }
913
914 EXPORT_SYMBOL(vfs_writev);
915
916 static ssize_t do_readv(unsigned long fd, const struct iovec __user *vec,
917 unsigned long vlen, int flags)
918 {
919 struct fd f = fdget_pos(fd);
920 ssize_t ret = -EBADF;
921
922 if (f.file) {
923 loff_t pos = file_pos_read(f.file);
924 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
925 if (ret >= 0)
926 file_pos_write(f.file, pos);
927 fdput_pos(f);
928 }
929
930 if (ret > 0)
931 add_rchar(current, ret);
932 inc_syscr(current);
933 return ret;
934 }
935
936 static ssize_t do_writev(unsigned long fd, const struct iovec __user *vec,
937 unsigned long vlen, int flags)
938 {
939 struct fd f = fdget_pos(fd);
940 ssize_t ret = -EBADF;
941
942 if (f.file) {
943 loff_t pos = file_pos_read(f.file);
944 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
945 if (ret >= 0)
946 file_pos_write(f.file, pos);
947 fdput_pos(f);
948 }
949
950 if (ret > 0)
951 add_wchar(current, ret);
952 inc_syscw(current);
953 return ret;
954 }
955
956 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
957 {
958 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
959 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
960 }
961
962 static ssize_t do_preadv(unsigned long fd, const struct iovec __user *vec,
963 unsigned long vlen, loff_t pos, int flags)
964 {
965 struct fd f;
966 ssize_t ret = -EBADF;
967
968 if (pos < 0)
969 return -EINVAL;
970
971 f = fdget(fd);
972 if (f.file) {
973 ret = -ESPIPE;
974 if (f.file->f_mode & FMODE_PREAD)
975 ret = vfs_readv(f.file, vec, vlen, &pos, flags);
976 fdput(f);
977 }
978
979 if (ret > 0)
980 add_rchar(current, ret);
981 inc_syscr(current);
982 return ret;
983 }
984
985 static ssize_t do_pwritev(unsigned long fd, const struct iovec __user *vec,
986 unsigned long vlen, loff_t pos, int flags)
987 {
988 struct fd f;
989 ssize_t ret = -EBADF;
990
991 if (pos < 0)
992 return -EINVAL;
993
994 f = fdget(fd);
995 if (f.file) {
996 ret = -ESPIPE;
997 if (f.file->f_mode & FMODE_PWRITE)
998 ret = vfs_writev(f.file, vec, vlen, &pos, flags);
999 fdput(f);
1000 }
1001
1002 if (ret > 0)
1003 add_wchar(current, ret);
1004 inc_syscw(current);
1005 return ret;
1006 }
1007
1008 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
1009 unsigned long, vlen)
1010 {
1011 return do_readv(fd, vec, vlen, 0);
1012 }
1013
1014 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
1015 unsigned long, vlen)
1016 {
1017 return do_writev(fd, vec, vlen, 0);
1018 }
1019
1020 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
1021 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1022 {
1023 loff_t pos = pos_from_hilo(pos_h, pos_l);
1024
1025 return do_preadv(fd, vec, vlen, pos, 0);
1026 }
1027
1028 SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
1029 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1030 int, flags)
1031 {
1032 loff_t pos = pos_from_hilo(pos_h, pos_l);
1033
1034 if (pos == -1)
1035 return do_readv(fd, vec, vlen, flags);
1036
1037 return do_preadv(fd, vec, vlen, pos, flags);
1038 }
1039
1040 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1041 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1042 {
1043 loff_t pos = pos_from_hilo(pos_h, pos_l);
1044
1045 return do_pwritev(fd, vec, vlen, pos, 0);
1046 }
1047
1048 SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
1049 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
1050 int, flags)
1051 {
1052 loff_t pos = pos_from_hilo(pos_h, pos_l);
1053
1054 if (pos == -1)
1055 return do_writev(fd, vec, vlen, flags);
1056
1057 return do_pwritev(fd, vec, vlen, pos, flags);
1058 }
1059
1060 #ifdef CONFIG_COMPAT
1061
1062 static ssize_t compat_do_readv_writev(int type, struct file *file,
1063 const struct compat_iovec __user *uvector,
1064 unsigned long nr_segs, loff_t *pos,
1065 int flags)
1066 {
1067 compat_ssize_t tot_len;
1068 struct iovec iovstack[UIO_FASTIOV];
1069 struct iovec *iov = iovstack;
1070 struct iov_iter iter;
1071 ssize_t ret;
1072 io_fn_t fn;
1073 iter_fn_t iter_fn;
1074
1075 ret = compat_import_iovec(type, uvector, nr_segs,
1076 UIO_FASTIOV, &iov, &iter);
1077 if (ret < 0)
1078 return ret;
1079
1080 tot_len = iov_iter_count(&iter);
1081 if (!tot_len)
1082 goto out;
1083 ret = rw_verify_area(type, file, pos, tot_len);
1084 if (ret < 0)
1085 goto out;
1086
1087 if (type == READ) {
1088 fn = file->f_op->read;
1089 iter_fn = file->f_op->read_iter;
1090 } else {
1091 fn = (io_fn_t)file->f_op->write;
1092 iter_fn = file->f_op->write_iter;
1093 file_start_write(file);
1094 }
1095
1096 if (iter_fn)
1097 ret = do_iter_readv_writev(file, &iter, pos, iter_fn, flags);
1098 else
1099 ret = do_loop_readv_writev(file, &iter, pos, fn, flags);
1100
1101 if (type != READ)
1102 file_end_write(file);
1103
1104 out:
1105 kfree(iov);
1106 if ((ret + (type == READ)) > 0) {
1107 if (type == READ)
1108 fsnotify_access(file);
1109 else
1110 fsnotify_modify(file);
1111 }
1112 return ret;
1113 }
1114
1115 static size_t compat_readv(struct file *file,
1116 const struct compat_iovec __user *vec,
1117 unsigned long vlen, loff_t *pos, int flags)
1118 {
1119 ssize_t ret = -EBADF;
1120
1121 if (!(file->f_mode & FMODE_READ))
1122 goto out;
1123
1124 ret = -EINVAL;
1125 if (!(file->f_mode & FMODE_CAN_READ))
1126 goto out;
1127
1128 ret = compat_do_readv_writev(READ, file, vec, vlen, pos, flags);
1129
1130 out:
1131 if (ret > 0)
1132 add_rchar(current, ret);
1133 inc_syscr(current);
1134 return ret;
1135 }
1136
1137 static size_t do_compat_readv(compat_ulong_t fd,
1138 const struct compat_iovec __user *vec,
1139 compat_ulong_t vlen, int flags)
1140 {
1141 struct fd f = fdget_pos(fd);
1142 ssize_t ret;
1143 loff_t pos;
1144
1145 if (!f.file)
1146 return -EBADF;
1147 pos = f.file->f_pos;
1148 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1149 if (ret >= 0)
1150 f.file->f_pos = pos;
1151 fdput_pos(f);
1152 return ret;
1153
1154 }
1155
1156 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1157 const struct compat_iovec __user *,vec,
1158 compat_ulong_t, vlen)
1159 {
1160 return do_compat_readv(fd, vec, vlen, 0);
1161 }
1162
1163 static long do_compat_preadv64(unsigned long fd,
1164 const struct compat_iovec __user *vec,
1165 unsigned long vlen, loff_t pos, int flags)
1166 {
1167 struct fd f;
1168 ssize_t ret;
1169
1170 if (pos < 0)
1171 return -EINVAL;
1172 f = fdget(fd);
1173 if (!f.file)
1174 return -EBADF;
1175 ret = -ESPIPE;
1176 if (f.file->f_mode & FMODE_PREAD)
1177 ret = compat_readv(f.file, vec, vlen, &pos, flags);
1178 fdput(f);
1179 return ret;
1180 }
1181
1182 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1183 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1184 const struct compat_iovec __user *,vec,
1185 unsigned long, vlen, loff_t, pos)
1186 {
1187 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1188 }
1189 #endif
1190
1191 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1192 const struct compat_iovec __user *,vec,
1193 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1194 {
1195 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1196
1197 return do_compat_preadv64(fd, vec, vlen, pos, 0);
1198 }
1199
1200 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64V2
1201 COMPAT_SYSCALL_DEFINE5(preadv64v2, unsigned long, fd,
1202 const struct compat_iovec __user *,vec,
1203 unsigned long, vlen, loff_t, pos, int, flags)
1204 {
1205 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1206 }
1207 #endif
1208
1209 COMPAT_SYSCALL_DEFINE6(preadv2, compat_ulong_t, fd,
1210 const struct compat_iovec __user *,vec,
1211 compat_ulong_t, vlen, u32, pos_low, u32, pos_high,
1212 int, flags)
1213 {
1214 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1215
1216 if (pos == -1)
1217 return do_compat_readv(fd, vec, vlen, flags);
1218
1219 return do_compat_preadv64(fd, vec, vlen, pos, flags);
1220 }
1221
1222 static size_t compat_writev(struct file *file,
1223 const struct compat_iovec __user *vec,
1224 unsigned long vlen, loff_t *pos, int flags)
1225 {
1226 ssize_t ret = -EBADF;
1227
1228 if (!(file->f_mode & FMODE_WRITE))
1229 goto out;
1230
1231 ret = -EINVAL;
1232 if (!(file->f_mode & FMODE_CAN_WRITE))
1233 goto out;
1234
1235 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos, 0);
1236
1237 out:
1238 if (ret > 0)
1239 add_wchar(current, ret);
1240 inc_syscw(current);
1241 return ret;
1242 }
1243
1244 static size_t do_compat_writev(compat_ulong_t fd,
1245 const struct compat_iovec __user* vec,
1246 compat_ulong_t vlen, int flags)
1247 {
1248 struct fd f = fdget_pos(fd);
1249 ssize_t ret;
1250 loff_t pos;
1251
1252 if (!f.file)
1253 return -EBADF;
1254 pos = f.file->f_pos;
1255 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1256 if (ret >= 0)
1257 f.file->f_pos = pos;
1258 fdput_pos(f);
1259 return ret;
1260 }
1261
1262 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1263 const struct compat_iovec __user *, vec,
1264 compat_ulong_t, vlen)
1265 {
1266 return do_compat_writev(fd, vec, vlen, 0);
1267 }
1268
1269 static long do_compat_pwritev64(unsigned long fd,
1270 const struct compat_iovec __user *vec,
1271 unsigned long vlen, loff_t pos, int flags)
1272 {
1273 struct fd f;
1274 ssize_t ret;
1275
1276 if (pos < 0)
1277 return -EINVAL;
1278 f = fdget(fd);
1279 if (!f.file)
1280 return -EBADF;
1281 ret = -ESPIPE;
1282 if (f.file->f_mode & FMODE_PWRITE)
1283 ret = compat_writev(f.file, vec, vlen, &pos, flags);
1284 fdput(f);
1285 return ret;
1286 }
1287
1288 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1289 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1290 const struct compat_iovec __user *,vec,
1291 unsigned long, vlen, loff_t, pos)
1292 {
1293 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1294 }
1295 #endif
1296
1297 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1298 const struct compat_iovec __user *,vec,
1299 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1300 {
1301 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1302
1303 return do_compat_pwritev64(fd, vec, vlen, pos, 0);
1304 }
1305
1306 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64V2
1307 COMPAT_SYSCALL_DEFINE5(pwritev64v2, unsigned long, fd,
1308 const struct compat_iovec __user *,vec,
1309 unsigned long, vlen, loff_t, pos, int, flags)
1310 {
1311 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1312 }
1313 #endif
1314
1315 COMPAT_SYSCALL_DEFINE6(pwritev2, compat_ulong_t, fd,
1316 const struct compat_iovec __user *,vec,
1317 compat_ulong_t, vlen, u32, pos_low, u32, pos_high, int, flags)
1318 {
1319 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1320
1321 if (pos == -1)
1322 return do_compat_writev(fd, vec, vlen, flags);
1323
1324 return do_compat_pwritev64(fd, vec, vlen, pos, flags);
1325 }
1326
1327 #endif
1328
1329 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1330 size_t count, loff_t max)
1331 {
1332 struct fd in, out;
1333 struct inode *in_inode, *out_inode;
1334 loff_t pos;
1335 loff_t out_pos;
1336 ssize_t retval;
1337 int fl;
1338
1339 /*
1340 * Get input file, and verify that it is ok..
1341 */
1342 retval = -EBADF;
1343 in = fdget(in_fd);
1344 if (!in.file)
1345 goto out;
1346 if (!(in.file->f_mode & FMODE_READ))
1347 goto fput_in;
1348 retval = -ESPIPE;
1349 if (!ppos) {
1350 pos = in.file->f_pos;
1351 } else {
1352 pos = *ppos;
1353 if (!(in.file->f_mode & FMODE_PREAD))
1354 goto fput_in;
1355 }
1356 retval = rw_verify_area(READ, in.file, &pos, count);
1357 if (retval < 0)
1358 goto fput_in;
1359 if (count > MAX_RW_COUNT)
1360 count = MAX_RW_COUNT;
1361
1362 /*
1363 * Get output file, and verify that it is ok..
1364 */
1365 retval = -EBADF;
1366 out = fdget(out_fd);
1367 if (!out.file)
1368 goto fput_in;
1369 if (!(out.file->f_mode & FMODE_WRITE))
1370 goto fput_out;
1371 retval = -EINVAL;
1372 in_inode = file_inode(in.file);
1373 out_inode = file_inode(out.file);
1374 out_pos = out.file->f_pos;
1375 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1376 if (retval < 0)
1377 goto fput_out;
1378
1379 if (!max)
1380 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1381
1382 if (unlikely(pos + count > max)) {
1383 retval = -EOVERFLOW;
1384 if (pos >= max)
1385 goto fput_out;
1386 count = max - pos;
1387 }
1388
1389 fl = 0;
1390 #if 0
1391 /*
1392 * We need to debate whether we can enable this or not. The
1393 * man page documents EAGAIN return for the output at least,
1394 * and the application is arguably buggy if it doesn't expect
1395 * EAGAIN on a non-blocking file descriptor.
1396 */
1397 if (in.file->f_flags & O_NONBLOCK)
1398 fl = SPLICE_F_NONBLOCK;
1399 #endif
1400 file_start_write(out.file);
1401 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1402 file_end_write(out.file);
1403
1404 if (retval > 0) {
1405 add_rchar(current, retval);
1406 add_wchar(current, retval);
1407 fsnotify_access(in.file);
1408 fsnotify_modify(out.file);
1409 out.file->f_pos = out_pos;
1410 if (ppos)
1411 *ppos = pos;
1412 else
1413 in.file->f_pos = pos;
1414 }
1415
1416 inc_syscr(current);
1417 inc_syscw(current);
1418 if (pos > max)
1419 retval = -EOVERFLOW;
1420
1421 fput_out:
1422 fdput(out);
1423 fput_in:
1424 fdput(in);
1425 out:
1426 return retval;
1427 }
1428
1429 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1430 {
1431 loff_t pos;
1432 off_t off;
1433 ssize_t ret;
1434
1435 if (offset) {
1436 if (unlikely(get_user(off, offset)))
1437 return -EFAULT;
1438 pos = off;
1439 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1440 if (unlikely(put_user(pos, offset)))
1441 return -EFAULT;
1442 return ret;
1443 }
1444
1445 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1446 }
1447
1448 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1449 {
1450 loff_t pos;
1451 ssize_t ret;
1452
1453 if (offset) {
1454 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1455 return -EFAULT;
1456 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1457 if (unlikely(put_user(pos, offset)))
1458 return -EFAULT;
1459 return ret;
1460 }
1461
1462 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1463 }
1464
1465 #ifdef CONFIG_COMPAT
1466 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1467 compat_off_t __user *, offset, compat_size_t, count)
1468 {
1469 loff_t pos;
1470 off_t off;
1471 ssize_t ret;
1472
1473 if (offset) {
1474 if (unlikely(get_user(off, offset)))
1475 return -EFAULT;
1476 pos = off;
1477 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1478 if (unlikely(put_user(pos, offset)))
1479 return -EFAULT;
1480 return ret;
1481 }
1482
1483 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1484 }
1485
1486 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1487 compat_loff_t __user *, offset, compat_size_t, count)
1488 {
1489 loff_t pos;
1490 ssize_t ret;
1491
1492 if (offset) {
1493 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1494 return -EFAULT;
1495 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1496 if (unlikely(put_user(pos, offset)))
1497 return -EFAULT;
1498 return ret;
1499 }
1500
1501 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1502 }
1503 #endif
1504
1505 /*
1506 * copy_file_range() differs from regular file read and write in that it
1507 * specifically allows return partial success. When it does so is up to
1508 * the copy_file_range method.
1509 */
1510 ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
1511 struct file *file_out, loff_t pos_out,
1512 size_t len, unsigned int flags)
1513 {
1514 struct inode *inode_in = file_inode(file_in);
1515 struct inode *inode_out = file_inode(file_out);
1516 ssize_t ret;
1517
1518 if (flags != 0)
1519 return -EINVAL;
1520
1521 ret = rw_verify_area(READ, file_in, &pos_in, len);
1522 if (unlikely(ret))
1523 return ret;
1524
1525 ret = rw_verify_area(WRITE, file_out, &pos_out, len);
1526 if (unlikely(ret))
1527 return ret;
1528
1529 if (!(file_in->f_mode & FMODE_READ) ||
1530 !(file_out->f_mode & FMODE_WRITE) ||
1531 (file_out->f_flags & O_APPEND))
1532 return -EBADF;
1533
1534 /* this could be relaxed once a method supports cross-fs copies */
1535 if (inode_in->i_sb != inode_out->i_sb)
1536 return -EXDEV;
1537
1538 if (len == 0)
1539 return 0;
1540
1541 sb_start_write(inode_out->i_sb);
1542
1543 /*
1544 * Try cloning first, this is supported by more file systems, and
1545 * more efficient if both clone and copy are supported (e.g. NFS).
1546 */
1547 if (file_in->f_op->clone_file_range) {
1548 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1549 file_out, pos_out, len);
1550 if (ret == 0) {
1551 ret = len;
1552 goto done;
1553 }
1554 }
1555
1556 if (file_out->f_op->copy_file_range) {
1557 ret = file_out->f_op->copy_file_range(file_in, pos_in, file_out,
1558 pos_out, len, flags);
1559 if (ret != -EOPNOTSUPP)
1560 goto done;
1561 }
1562
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 done:
1567 if (ret > 0) {
1568 fsnotify_access(file_in);
1569 add_rchar(current, ret);
1570 fsnotify_modify(file_out);
1571 add_wchar(current, ret);
1572 }
1573
1574 inc_syscr(current);
1575 inc_syscw(current);
1576
1577 sb_end_write(inode_out->i_sb);
1578
1579 return ret;
1580 }
1581 EXPORT_SYMBOL(vfs_copy_file_range);
1582
1583 SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in,
1584 int, fd_out, loff_t __user *, off_out,
1585 size_t, len, unsigned int, flags)
1586 {
1587 loff_t pos_in;
1588 loff_t pos_out;
1589 struct fd f_in;
1590 struct fd f_out;
1591 ssize_t ret = -EBADF;
1592
1593 f_in = fdget(fd_in);
1594 if (!f_in.file)
1595 goto out2;
1596
1597 f_out = fdget(fd_out);
1598 if (!f_out.file)
1599 goto out1;
1600
1601 ret = -EFAULT;
1602 if (off_in) {
1603 if (copy_from_user(&pos_in, off_in, sizeof(loff_t)))
1604 goto out;
1605 } else {
1606 pos_in = f_in.file->f_pos;
1607 }
1608
1609 if (off_out) {
1610 if (copy_from_user(&pos_out, off_out, sizeof(loff_t)))
1611 goto out;
1612 } else {
1613 pos_out = f_out.file->f_pos;
1614 }
1615
1616 ret = vfs_copy_file_range(f_in.file, pos_in, f_out.file, pos_out, len,
1617 flags);
1618 if (ret > 0) {
1619 pos_in += ret;
1620 pos_out += ret;
1621
1622 if (off_in) {
1623 if (copy_to_user(off_in, &pos_in, sizeof(loff_t)))
1624 ret = -EFAULT;
1625 } else {
1626 f_in.file->f_pos = pos_in;
1627 }
1628
1629 if (off_out) {
1630 if (copy_to_user(off_out, &pos_out, sizeof(loff_t)))
1631 ret = -EFAULT;
1632 } else {
1633 f_out.file->f_pos = pos_out;
1634 }
1635 }
1636
1637 out:
1638 fdput(f_out);
1639 out1:
1640 fdput(f_in);
1641 out2:
1642 return ret;
1643 }
1644
1645 static int clone_verify_area(struct file *file, loff_t pos, u64 len, bool write)
1646 {
1647 struct inode *inode = file_inode(file);
1648
1649 if (unlikely(pos < 0))
1650 return -EINVAL;
1651
1652 if (unlikely((loff_t) (pos + len) < 0))
1653 return -EINVAL;
1654
1655 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
1656 loff_t end = len ? pos + len - 1 : OFFSET_MAX;
1657 int retval;
1658
1659 retval = locks_mandatory_area(inode, file, pos, end,
1660 write ? F_WRLCK : F_RDLCK);
1661 if (retval < 0)
1662 return retval;
1663 }
1664
1665 return security_file_permission(file, write ? MAY_WRITE : MAY_READ);
1666 }
1667
1668 /*
1669 * Check that the two inodes are eligible for cloning, the ranges make
1670 * sense, and then flush all dirty data. Caller must ensure that the
1671 * inodes have been locked against any other modifications.
1672 *
1673 * Returns: 0 for "nothing to clone", 1 for "something to clone", or
1674 * the usual negative error code.
1675 */
1676 int vfs_clone_file_prep_inodes(struct inode *inode_in, loff_t pos_in,
1677 struct inode *inode_out, loff_t pos_out,
1678 u64 *len, bool is_dedupe)
1679 {
1680 loff_t bs = inode_out->i_sb->s_blocksize;
1681 loff_t blen;
1682 loff_t isize;
1683 bool same_inode = (inode_in == inode_out);
1684 int ret;
1685
1686 /* Don't touch certain kinds of inodes */
1687 if (IS_IMMUTABLE(inode_out))
1688 return -EPERM;
1689
1690 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
1691 return -ETXTBSY;
1692
1693 /* Don't reflink dirs, pipes, sockets... */
1694 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1695 return -EISDIR;
1696 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1697 return -EINVAL;
1698
1699 /* Are we going all the way to the end? */
1700 isize = i_size_read(inode_in);
1701 if (isize == 0)
1702 return 0;
1703
1704 /* Zero length dedupe exits immediately; reflink goes to EOF. */
1705 if (*len == 0) {
1706 if (is_dedupe || pos_in == isize)
1707 return 0;
1708 if (pos_in > isize)
1709 return -EINVAL;
1710 *len = isize - pos_in;
1711 }
1712
1713 /* Ensure offsets don't wrap and the input is inside i_size */
1714 if (pos_in + *len < pos_in || pos_out + *len < pos_out ||
1715 pos_in + *len > isize)
1716 return -EINVAL;
1717
1718 /* Don't allow dedupe past EOF in the dest file */
1719 if (is_dedupe) {
1720 loff_t disize;
1721
1722 disize = i_size_read(inode_out);
1723 if (pos_out >= disize || pos_out + *len > disize)
1724 return -EINVAL;
1725 }
1726
1727 /* If we're linking to EOF, continue to the block boundary. */
1728 if (pos_in + *len == isize)
1729 blen = ALIGN(isize, bs) - pos_in;
1730 else
1731 blen = *len;
1732
1733 /* Only reflink if we're aligned to block boundaries */
1734 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_in + blen, bs) ||
1735 !IS_ALIGNED(pos_out, bs) || !IS_ALIGNED(pos_out + blen, bs))
1736 return -EINVAL;
1737
1738 /* Don't allow overlapped reflink within the same file */
1739 if (same_inode) {
1740 if (pos_out + blen > pos_in && pos_out < pos_in + blen)
1741 return -EINVAL;
1742 }
1743
1744 /* Wait for the completion of any pending IOs on both files */
1745 inode_dio_wait(inode_in);
1746 if (!same_inode)
1747 inode_dio_wait(inode_out);
1748
1749 ret = filemap_write_and_wait_range(inode_in->i_mapping,
1750 pos_in, pos_in + *len - 1);
1751 if (ret)
1752 return ret;
1753
1754 ret = filemap_write_and_wait_range(inode_out->i_mapping,
1755 pos_out, pos_out + *len - 1);
1756 if (ret)
1757 return ret;
1758
1759 /*
1760 * Check that the extents are the same.
1761 */
1762 if (is_dedupe) {
1763 bool is_same = false;
1764
1765 ret = vfs_dedupe_file_range_compare(inode_in, pos_in,
1766 inode_out, pos_out, *len, &is_same);
1767 if (ret)
1768 return ret;
1769 if (!is_same)
1770 return -EBADE;
1771 }
1772
1773 return 1;
1774 }
1775 EXPORT_SYMBOL(vfs_clone_file_prep_inodes);
1776
1777 int vfs_clone_file_range(struct file *file_in, loff_t pos_in,
1778 struct file *file_out, loff_t pos_out, u64 len)
1779 {
1780 struct inode *inode_in = file_inode(file_in);
1781 struct inode *inode_out = file_inode(file_out);
1782 int ret;
1783
1784 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
1785 return -EISDIR;
1786 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
1787 return -EINVAL;
1788
1789 /*
1790 * FICLONE/FICLONERANGE ioctls enforce that src and dest files are on
1791 * the same mount. Practically, they only need to be on the same file
1792 * system.
1793 */
1794 if (inode_in->i_sb != inode_out->i_sb)
1795 return -EXDEV;
1796
1797 if (!(file_in->f_mode & FMODE_READ) ||
1798 !(file_out->f_mode & FMODE_WRITE) ||
1799 (file_out->f_flags & O_APPEND))
1800 return -EBADF;
1801
1802 if (!file_in->f_op->clone_file_range)
1803 return -EOPNOTSUPP;
1804
1805 ret = clone_verify_area(file_in, pos_in, len, false);
1806 if (ret)
1807 return ret;
1808
1809 ret = clone_verify_area(file_out, pos_out, len, true);
1810 if (ret)
1811 return ret;
1812
1813 if (pos_in + len > i_size_read(inode_in))
1814 return -EINVAL;
1815
1816 ret = file_in->f_op->clone_file_range(file_in, pos_in,
1817 file_out, pos_out, len);
1818 if (!ret) {
1819 fsnotify_access(file_in);
1820 fsnotify_modify(file_out);
1821 }
1822
1823 return ret;
1824 }
1825 EXPORT_SYMBOL(vfs_clone_file_range);
1826
1827 /*
1828 * Read a page's worth of file data into the page cache. Return the page
1829 * locked.
1830 */
1831 static struct page *vfs_dedupe_get_page(struct inode *inode, loff_t offset)
1832 {
1833 struct address_space *mapping;
1834 struct page *page;
1835 pgoff_t n;
1836
1837 n = offset >> PAGE_SHIFT;
1838 mapping = inode->i_mapping;
1839 page = read_mapping_page(mapping, n, NULL);
1840 if (IS_ERR(page))
1841 return page;
1842 if (!PageUptodate(page)) {
1843 put_page(page);
1844 return ERR_PTR(-EIO);
1845 }
1846 lock_page(page);
1847 return page;
1848 }
1849
1850 /*
1851 * Compare extents of two files to see if they are the same.
1852 * Caller must have locked both inodes to prevent write races.
1853 */
1854 int vfs_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
1855 struct inode *dest, loff_t destoff,
1856 loff_t len, bool *is_same)
1857 {
1858 loff_t src_poff;
1859 loff_t dest_poff;
1860 void *src_addr;
1861 void *dest_addr;
1862 struct page *src_page;
1863 struct page *dest_page;
1864 loff_t cmp_len;
1865 bool same;
1866 int error;
1867
1868 error = -EINVAL;
1869 same = true;
1870 while (len) {
1871 src_poff = srcoff & (PAGE_SIZE - 1);
1872 dest_poff = destoff & (PAGE_SIZE - 1);
1873 cmp_len = min(PAGE_SIZE - src_poff,
1874 PAGE_SIZE - dest_poff);
1875 cmp_len = min(cmp_len, len);
1876 if (cmp_len <= 0)
1877 goto out_error;
1878
1879 src_page = vfs_dedupe_get_page(src, srcoff);
1880 if (IS_ERR(src_page)) {
1881 error = PTR_ERR(src_page);
1882 goto out_error;
1883 }
1884 dest_page = vfs_dedupe_get_page(dest, destoff);
1885 if (IS_ERR(dest_page)) {
1886 error = PTR_ERR(dest_page);
1887 unlock_page(src_page);
1888 put_page(src_page);
1889 goto out_error;
1890 }
1891 src_addr = kmap_atomic(src_page);
1892 dest_addr = kmap_atomic(dest_page);
1893
1894 flush_dcache_page(src_page);
1895 flush_dcache_page(dest_page);
1896
1897 if (memcmp(src_addr + src_poff, dest_addr + dest_poff, cmp_len))
1898 same = false;
1899
1900 kunmap_atomic(dest_addr);
1901 kunmap_atomic(src_addr);
1902 unlock_page(dest_page);
1903 unlock_page(src_page);
1904 put_page(dest_page);
1905 put_page(src_page);
1906
1907 if (!same)
1908 break;
1909
1910 srcoff += cmp_len;
1911 destoff += cmp_len;
1912 len -= cmp_len;
1913 }
1914
1915 *is_same = same;
1916 return 0;
1917
1918 out_error:
1919 return error;
1920 }
1921 EXPORT_SYMBOL(vfs_dedupe_file_range_compare);
1922
1923 int vfs_dedupe_file_range(struct file *file, struct file_dedupe_range *same)
1924 {
1925 struct file_dedupe_range_info *info;
1926 struct inode *src = file_inode(file);
1927 u64 off;
1928 u64 len;
1929 int i;
1930 int ret;
1931 bool is_admin = capable(CAP_SYS_ADMIN);
1932 u16 count = same->dest_count;
1933 struct file *dst_file;
1934 loff_t dst_off;
1935 ssize_t deduped;
1936
1937 if (!(file->f_mode & FMODE_READ))
1938 return -EINVAL;
1939
1940 if (same->reserved1 || same->reserved2)
1941 return -EINVAL;
1942
1943 off = same->src_offset;
1944 len = same->src_length;
1945
1946 ret = -EISDIR;
1947 if (S_ISDIR(src->i_mode))
1948 goto out;
1949
1950 ret = -EINVAL;
1951 if (!S_ISREG(src->i_mode))
1952 goto out;
1953
1954 ret = clone_verify_area(file, off, len, false);
1955 if (ret < 0)
1956 goto out;
1957 ret = 0;
1958
1959 if (off + len > i_size_read(src))
1960 return -EINVAL;
1961
1962 /* pre-format output fields to sane values */
1963 for (i = 0; i < count; i++) {
1964 same->info[i].bytes_deduped = 0ULL;
1965 same->info[i].status = FILE_DEDUPE_RANGE_SAME;
1966 }
1967
1968 for (i = 0, info = same->info; i < count; i++, info++) {
1969 struct inode *dst;
1970 struct fd dst_fd = fdget(info->dest_fd);
1971
1972 dst_file = dst_fd.file;
1973 if (!dst_file) {
1974 info->status = -EBADF;
1975 goto next_loop;
1976 }
1977 dst = file_inode(dst_file);
1978
1979 ret = mnt_want_write_file(dst_file);
1980 if (ret) {
1981 info->status = ret;
1982 goto next_loop;
1983 }
1984
1985 dst_off = info->dest_offset;
1986 ret = clone_verify_area(dst_file, dst_off, len, true);
1987 if (ret < 0) {
1988 info->status = ret;
1989 goto next_file;
1990 }
1991 ret = 0;
1992
1993 if (info->reserved) {
1994 info->status = -EINVAL;
1995 } else if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) {
1996 info->status = -EINVAL;
1997 } else if (file->f_path.mnt != dst_file->f_path.mnt) {
1998 info->status = -EXDEV;
1999 } else if (S_ISDIR(dst->i_mode)) {
2000 info->status = -EISDIR;
2001 } else if (dst_file->f_op->dedupe_file_range == NULL) {
2002 info->status = -EINVAL;
2003 } else {
2004 deduped = dst_file->f_op->dedupe_file_range(file, off,
2005 len, dst_file,
2006 info->dest_offset);
2007 if (deduped == -EBADE)
2008 info->status = FILE_DEDUPE_RANGE_DIFFERS;
2009 else if (deduped < 0)
2010 info->status = deduped;
2011 else
2012 info->bytes_deduped += deduped;
2013 }
2014
2015 next_file:
2016 mnt_drop_write_file(dst_file);
2017 next_loop:
2018 fdput(dst_fd);
2019
2020 if (fatal_signal_pending(current))
2021 goto out;
2022 }
2023
2024 out:
2025 return ret;
2026 }
2027 EXPORT_SYMBOL(vfs_dedupe_file_range);
CRIS linux kernel tree