projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
macvlan: implement ndo_get_iflink
[deliverable/linux.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/aio.h>
13 #include <linux/fsnotify.h>
14 #include <linux/security.h>
15 #include <linux/export.h>
16 #include <linux/syscalls.h>
17 #include <linux/pagemap.h>
18 #include <linux/splice.h>
19 #include <linux/compat.h>
20 #include "internal.h"
21
22 #include <asm/uaccess.h>
23 #include <asm/unistd.h>
24
25 typedef ssize_t (*io_fn_t)(struct file *, char __user *, size_t, loff_t *);
26 typedef ssize_t (*iov_fn_t)(struct kiocb *, const struct iovec *,
27 unsigned long, loff_t);
28 typedef ssize_t (*iter_fn_t)(struct kiocb *, struct iov_iter *);
29
30 const struct file_operations generic_ro_fops = {
31 .llseek = generic_file_llseek,
32 .read = new_sync_read,
33 .read_iter = generic_file_read_iter,
34 .mmap = generic_file_readonly_mmap,
35 .splice_read = generic_file_splice_read,
36 };
37
38 EXPORT_SYMBOL(generic_ro_fops);
39
40 static inline int unsigned_offsets(struct file *file)
41 {
42 return file->f_mode & FMODE_UNSIGNED_OFFSET;
43 }
44
45 /**
46 * vfs_setpos - update the file offset for lseek
47 * @file: file structure in question
48 * @offset: file offset to seek to
49 * @maxsize: maximum file size
50 *
51 * This is a low-level filesystem helper for updating the file offset to
52 * the value specified by @offset if the given offset is valid and it is
53 * not equal to the current file offset.
54 *
55 * Return the specified offset on success and -EINVAL on invalid offset.
56 */
57 loff_t vfs_setpos(struct file *file, loff_t offset, loff_t maxsize)
58 {
59 if (offset < 0 && !unsigned_offsets(file))
60 return -EINVAL;
61 if (offset > maxsize)
62 return -EINVAL;
63
64 if (offset != file->f_pos) {
65 file->f_pos = offset;
66 file->f_version = 0;
67 }
68 return offset;
69 }
70 EXPORT_SYMBOL(vfs_setpos);
71
72 /**
73 * generic_file_llseek_size - generic llseek implementation for regular files
74 * @file: file structure to seek on
75 * @offset: file offset to seek to
76 * @whence: type of seek
77 * @size: max size of this file in file system
78 * @eof: offset used for SEEK_END position
79 *
80 * This is a variant of generic_file_llseek that allows passing in a custom
81 * maximum file size and a custom EOF position, for e.g. hashed directories
82 *
83 * Synchronization:
84 * SEEK_SET and SEEK_END are unsynchronized (but atomic on 64bit platforms)
85 * SEEK_CUR is synchronized against other SEEK_CURs, but not read/writes.
86 * read/writes behave like SEEK_SET against seeks.
87 */
88 loff_t
89 generic_file_llseek_size(struct file *file, loff_t offset, int whence,
90 loff_t maxsize, loff_t eof)
91 {
92 switch (whence) {
93 case SEEK_END:
94 offset += eof;
95 break;
96 case SEEK_CUR:
97 /*
98 * Here we special-case the lseek(fd, 0, SEEK_CUR)
99 * position-querying operation. Avoid rewriting the "same"
100 * f_pos value back to the file because a concurrent read(),
101 * write() or lseek() might have altered it
102 */
103 if (offset == 0)
104 return file->f_pos;
105 /*
106 * f_lock protects against read/modify/write race with other
107 * SEEK_CURs. Note that parallel writes and reads behave
108 * like SEEK_SET.
109 */
110 spin_lock(&file->f_lock);
111 offset = vfs_setpos(file, file->f_pos + offset, maxsize);
112 spin_unlock(&file->f_lock);
113 return offset;
114 case SEEK_DATA:
115 /*
116 * In the generic case the entire file is data, so as long as
117 * offset isn't at the end of the file then the offset is data.
118 */
119 if (offset >= eof)
120 return -ENXIO;
121 break;
122 case SEEK_HOLE:
123 /*
124 * There is a virtual hole at the end of the file, so as long as
125 * offset isn't i_size or larger, return i_size.
126 */
127 if (offset >= eof)
128 return -ENXIO;
129 offset = eof;
130 break;
131 }
132
133 return vfs_setpos(file, offset, maxsize);
134 }
135 EXPORT_SYMBOL(generic_file_llseek_size);
136
137 /**
138 * generic_file_llseek - generic llseek implementation for regular files
139 * @file: file structure to seek on
140 * @offset: file offset to seek to
141 * @whence: type of seek
142 *
143 * This is a generic implemenation of ->llseek useable for all normal local
144 * filesystems. It just updates the file offset to the value specified by
145 * @offset and @whence.
146 */
147 loff_t generic_file_llseek(struct file *file, loff_t offset, int whence)
148 {
149 struct inode *inode = file->f_mapping->host;
150
151 return generic_file_llseek_size(file, offset, whence,
152 inode->i_sb->s_maxbytes,
153 i_size_read(inode));
154 }
155 EXPORT_SYMBOL(generic_file_llseek);
156
157 /**
158 * fixed_size_llseek - llseek implementation for fixed-sized devices
159 * @file: file structure to seek on
160 * @offset: file offset to seek to
161 * @whence: type of seek
162 * @size: size of the file
163 *
164 */
165 loff_t fixed_size_llseek(struct file *file, loff_t offset, int whence, loff_t size)
166 {
167 switch (whence) {
168 case SEEK_SET: case SEEK_CUR: case SEEK_END:
169 return generic_file_llseek_size(file, offset, whence,
170 size, size);
171 default:
172 return -EINVAL;
173 }
174 }
175 EXPORT_SYMBOL(fixed_size_llseek);
176
177 /**
178 * noop_llseek - No Operation Performed llseek implementation
179 * @file: file structure to seek on
180 * @offset: file offset to seek to
181 * @whence: type of seek
182 *
183 * This is an implementation of ->llseek useable for the rare special case when
184 * userspace expects the seek to succeed but the (device) file is actually not
185 * able to perform the seek. In this case you use noop_llseek() instead of
186 * falling back to the default implementation of ->llseek.
187 */
188 loff_t noop_llseek(struct file *file, loff_t offset, int whence)
189 {
190 return file->f_pos;
191 }
192 EXPORT_SYMBOL(noop_llseek);
193
194 loff_t no_llseek(struct file *file, loff_t offset, int whence)
195 {
196 return -ESPIPE;
197 }
198 EXPORT_SYMBOL(no_llseek);
199
200 loff_t default_llseek(struct file *file, loff_t offset, int whence)
201 {
202 struct inode *inode = file_inode(file);
203 loff_t retval;
204
205 mutex_lock(&inode->i_mutex);
206 switch (whence) {
207 case SEEK_END:
208 offset += i_size_read(inode);
209 break;
210 case SEEK_CUR:
211 if (offset == 0) {
212 retval = file->f_pos;
213 goto out;
214 }
215 offset += file->f_pos;
216 break;
217 case SEEK_DATA:
218 /*
219 * In the generic case the entire file is data, so as
220 * long as offset isn't at the end of the file then the
221 * offset is data.
222 */
223 if (offset >= inode->i_size) {
224 retval = -ENXIO;
225 goto out;
226 }
227 break;
228 case SEEK_HOLE:
229 /*
230 * There is a virtual hole at the end of the file, so
231 * as long as offset isn't i_size or larger, return
232 * i_size.
233 */
234 if (offset >= inode->i_size) {
235 retval = -ENXIO;
236 goto out;
237 }
238 offset = inode->i_size;
239 break;
240 }
241 retval = -EINVAL;
242 if (offset >= 0 || unsigned_offsets(file)) {
243 if (offset != file->f_pos) {
244 file->f_pos = offset;
245 file->f_version = 0;
246 }
247 retval = offset;
248 }
249 out:
250 mutex_unlock(&inode->i_mutex);
251 return retval;
252 }
253 EXPORT_SYMBOL(default_llseek);
254
255 loff_t vfs_llseek(struct file *file, loff_t offset, int whence)
256 {
257 loff_t (*fn)(struct file *, loff_t, int);
258
259 fn = no_llseek;
260 if (file->f_mode & FMODE_LSEEK) {
261 if (file->f_op->llseek)
262 fn = file->f_op->llseek;
263 }
264 return fn(file, offset, whence);
265 }
266 EXPORT_SYMBOL(vfs_llseek);
267
268 static inline struct fd fdget_pos(int fd)
269 {
270 return __to_fd(__fdget_pos(fd));
271 }
272
273 static inline void fdput_pos(struct fd f)
274 {
275 if (f.flags & FDPUT_POS_UNLOCK)
276 mutex_unlock(&f.file->f_pos_lock);
277 fdput(f);
278 }
279
280 SYSCALL_DEFINE3(lseek, unsigned int, fd, off_t, offset, unsigned int, whence)
281 {
282 off_t retval;
283 struct fd f = fdget_pos(fd);
284 if (!f.file)
285 return -EBADF;
286
287 retval = -EINVAL;
288 if (whence <= SEEK_MAX) {
289 loff_t res = vfs_llseek(f.file, offset, whence);
290 retval = res;
291 if (res != (loff_t)retval)
292 retval = -EOVERFLOW; /* LFS: should only happen on 32 bit platforms */
293 }
294 fdput_pos(f);
295 return retval;
296 }
297
298 #ifdef CONFIG_COMPAT
299 COMPAT_SYSCALL_DEFINE3(lseek, unsigned int, fd, compat_off_t, offset, unsigned int, whence)
300 {
301 return sys_lseek(fd, offset, whence);
302 }
303 #endif
304
305 #ifdef __ARCH_WANT_SYS_LLSEEK
306 SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned long, offset_high,
307 unsigned long, offset_low, loff_t __user *, result,
308 unsigned int, whence)
309 {
310 int retval;
311 struct fd f = fdget_pos(fd);
312 loff_t offset;
313
314 if (!f.file)
315 return -EBADF;
316
317 retval = -EINVAL;
318 if (whence > SEEK_MAX)
319 goto out_putf;
320
321 offset = vfs_llseek(f.file, ((loff_t) offset_high << 32) | offset_low,
322 whence);
323
324 retval = (int)offset;
325 if (offset >= 0) {
326 retval = -EFAULT;
327 if (!copy_to_user(result, &offset, sizeof(offset)))
328 retval = 0;
329 }
330 out_putf:
331 fdput_pos(f);
332 return retval;
333 }
334 #endif
335
336 ssize_t vfs_iter_read(struct file *file, struct iov_iter *iter, loff_t *ppos)
337 {
338 struct kiocb kiocb;
339 ssize_t ret;
340
341 if (!file->f_op->read_iter)
342 return -EINVAL;
343
344 init_sync_kiocb(&kiocb, file);
345 kiocb.ki_pos = *ppos;
346 kiocb.ki_nbytes = iov_iter_count(iter);
347
348 iter->type |= READ;
349 ret = file->f_op->read_iter(&kiocb, iter);
350 if (ret == -EIOCBQUEUED)
351 ret = wait_on_sync_kiocb(&kiocb);
352
353 if (ret > 0)
354 *ppos = kiocb.ki_pos;
355 return ret;
356 }
357 EXPORT_SYMBOL(vfs_iter_read);
358
359 ssize_t vfs_iter_write(struct file *file, struct iov_iter *iter, loff_t *ppos)
360 {
361 struct kiocb kiocb;
362 ssize_t ret;
363
364 if (!file->f_op->write_iter)
365 return -EINVAL;
366
367 init_sync_kiocb(&kiocb, file);
368 kiocb.ki_pos = *ppos;
369 kiocb.ki_nbytes = iov_iter_count(iter);
370
371 iter->type |= WRITE;
372 ret = file->f_op->write_iter(&kiocb, iter);
373 if (ret == -EIOCBQUEUED)
374 ret = wait_on_sync_kiocb(&kiocb);
375
376 if (ret > 0)
377 *ppos = kiocb.ki_pos;
378 return ret;
379 }
380 EXPORT_SYMBOL(vfs_iter_write);
381
382 /*
383 * rw_verify_area doesn't like huge counts. We limit
384 * them to something that fits in "int" so that others
385 * won't have to do range checks all the time.
386 */
387 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
388 {
389 struct inode *inode;
390 loff_t pos;
391 int retval = -EINVAL;
392
393 inode = file_inode(file);
394 if (unlikely((ssize_t) count < 0))
395 return retval;
396 pos = *ppos;
397 if (unlikely(pos < 0)) {
398 if (!unsigned_offsets(file))
399 return retval;
400 if (count >= -pos) /* both values are in 0..LLONG_MAX */
401 return -EOVERFLOW;
402 } else if (unlikely((loff_t) (pos + count) < 0)) {
403 if (!unsigned_offsets(file))
404 return retval;
405 }
406
407 if (unlikely(inode->i_flctx && mandatory_lock(inode))) {
408 retval = locks_mandatory_area(
409 read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE,
410 inode, file, pos, count);
411 if (retval < 0)
412 return retval;
413 }
414 retval = security_file_permission(file,
415 read_write == READ ? MAY_READ : MAY_WRITE);
416 if (retval)
417 return retval;
418 return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
419 }
420
421 ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
422 {
423 struct iovec iov = { .iov_base = buf, .iov_len = len };
424 struct kiocb kiocb;
425 ssize_t ret;
426
427 init_sync_kiocb(&kiocb, filp);
428 kiocb.ki_pos = *ppos;
429 kiocb.ki_nbytes = len;
430
431 ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
432 if (-EIOCBQUEUED == ret)
433 ret = wait_on_sync_kiocb(&kiocb);
434 *ppos = kiocb.ki_pos;
435 return ret;
436 }
437
438 EXPORT_SYMBOL(do_sync_read);
439
440 ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
441 {
442 struct iovec iov = { .iov_base = buf, .iov_len = len };
443 struct kiocb kiocb;
444 struct iov_iter iter;
445 ssize_t ret;
446
447 init_sync_kiocb(&kiocb, filp);
448 kiocb.ki_pos = *ppos;
449 kiocb.ki_nbytes = len;
450 iov_iter_init(&iter, READ, &iov, 1, len);
451
452 ret = filp->f_op->read_iter(&kiocb, &iter);
453 if (-EIOCBQUEUED == ret)
454 ret = wait_on_sync_kiocb(&kiocb);
455 *ppos = kiocb.ki_pos;
456 return ret;
457 }
458
459 EXPORT_SYMBOL(new_sync_read);
460
461 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
462 loff_t *pos)
463 {
464 ssize_t ret;
465
466 if (file->f_op->read)
467 ret = file->f_op->read(file, buf, count, pos);
468 else if (file->f_op->aio_read)
469 ret = do_sync_read(file, buf, count, pos);
470 else if (file->f_op->read_iter)
471 ret = new_sync_read(file, buf, count, pos);
472 else
473 ret = -EINVAL;
474
475 return ret;
476 }
477
478 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
479 {
480 ssize_t ret;
481
482 if (!(file->f_mode & FMODE_READ))
483 return -EBADF;
484 if (!(file->f_mode & FMODE_CAN_READ))
485 return -EINVAL;
486 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
487 return -EFAULT;
488
489 ret = rw_verify_area(READ, file, pos, count);
490 if (ret >= 0) {
491 count = ret;
492 ret = __vfs_read(file, buf, count, pos);
493 if (ret > 0) {
494 fsnotify_access(file);
495 add_rchar(current, ret);
496 }
497 inc_syscr(current);
498 }
499
500 return ret;
501 }
502
503 EXPORT_SYMBOL(vfs_read);
504
505 ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
506 {
507 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
508 struct kiocb kiocb;
509 ssize_t ret;
510
511 init_sync_kiocb(&kiocb, filp);
512 kiocb.ki_pos = *ppos;
513 kiocb.ki_nbytes = len;
514
515 ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
516 if (-EIOCBQUEUED == ret)
517 ret = wait_on_sync_kiocb(&kiocb);
518 *ppos = kiocb.ki_pos;
519 return ret;
520 }
521
522 EXPORT_SYMBOL(do_sync_write);
523
524 ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
525 {
526 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
527 struct kiocb kiocb;
528 struct iov_iter iter;
529 ssize_t ret;
530
531 init_sync_kiocb(&kiocb, filp);
532 kiocb.ki_pos = *ppos;
533 kiocb.ki_nbytes = len;
534 iov_iter_init(&iter, WRITE, &iov, 1, len);
535
536 ret = filp->f_op->write_iter(&kiocb, &iter);
537 if (-EIOCBQUEUED == ret)
538 ret = wait_on_sync_kiocb(&kiocb);
539 *ppos = kiocb.ki_pos;
540 return ret;
541 }
542
543 EXPORT_SYMBOL(new_sync_write);
544
545 ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
546 {
547 mm_segment_t old_fs;
548 const char __user *p;
549 ssize_t ret;
550
551 if (!(file->f_mode & FMODE_CAN_WRITE))
552 return -EINVAL;
553
554 old_fs = get_fs();
555 set_fs(get_ds());
556 p = (__force const char __user *)buf;
557 if (count > MAX_RW_COUNT)
558 count = MAX_RW_COUNT;
559 if (file->f_op->write)
560 ret = file->f_op->write(file, p, count, pos);
561 else if (file->f_op->aio_write)
562 ret = do_sync_write(file, p, count, pos);
563 else
564 ret = new_sync_write(file, p, count, pos);
565 set_fs(old_fs);
566 if (ret > 0) {
567 fsnotify_modify(file);
568 add_wchar(current, ret);
569 }
570 inc_syscw(current);
571 return ret;
572 }
573
574 EXPORT_SYMBOL(__kernel_write);
575
576 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
577 {
578 ssize_t ret;
579
580 if (!(file->f_mode & FMODE_WRITE))
581 return -EBADF;
582 if (!(file->f_mode & FMODE_CAN_WRITE))
583 return -EINVAL;
584 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
585 return -EFAULT;
586
587 ret = rw_verify_area(WRITE, file, pos, count);
588 if (ret >= 0) {
589 count = ret;
590 file_start_write(file);
591 if (file->f_op->write)
592 ret = file->f_op->write(file, buf, count, pos);
593 else if (file->f_op->aio_write)
594 ret = do_sync_write(file, buf, count, pos);
595 else
596 ret = new_sync_write(file, buf, count, pos);
597 if (ret > 0) {
598 fsnotify_modify(file);
599 add_wchar(current, ret);
600 }
601 inc_syscw(current);
602 file_end_write(file);
603 }
604
605 return ret;
606 }
607
608 EXPORT_SYMBOL(vfs_write);
609
610 static inline loff_t file_pos_read(struct file *file)
611 {
612 return file->f_pos;
613 }
614
615 static inline void file_pos_write(struct file *file, loff_t pos)
616 {
617 file->f_pos = pos;
618 }
619
620 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
621 {
622 struct fd f = fdget_pos(fd);
623 ssize_t ret = -EBADF;
624
625 if (f.file) {
626 loff_t pos = file_pos_read(f.file);
627 ret = vfs_read(f.file, buf, count, &pos);
628 if (ret >= 0)
629 file_pos_write(f.file, pos);
630 fdput_pos(f);
631 }
632 return ret;
633 }
634
635 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
636 size_t, count)
637 {
638 struct fd f = fdget_pos(fd);
639 ssize_t ret = -EBADF;
640
641 if (f.file) {
642 loff_t pos = file_pos_read(f.file);
643 ret = vfs_write(f.file, buf, count, &pos);
644 if (ret >= 0)
645 file_pos_write(f.file, pos);
646 fdput_pos(f);
647 }
648
649 return ret;
650 }
651
652 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
653 size_t, count, loff_t, pos)
654 {
655 struct fd f;
656 ssize_t ret = -EBADF;
657
658 if (pos < 0)
659 return -EINVAL;
660
661 f = fdget(fd);
662 if (f.file) {
663 ret = -ESPIPE;
664 if (f.file->f_mode & FMODE_PREAD)
665 ret = vfs_read(f.file, buf, count, &pos);
666 fdput(f);
667 }
668
669 return ret;
670 }
671
672 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
673 size_t, count, loff_t, pos)
674 {
675 struct fd f;
676 ssize_t ret = -EBADF;
677
678 if (pos < 0)
679 return -EINVAL;
680
681 f = fdget(fd);
682 if (f.file) {
683 ret = -ESPIPE;
684 if (f.file->f_mode & FMODE_PWRITE)
685 ret = vfs_write(f.file, buf, count, &pos);
686 fdput(f);
687 }
688
689 return ret;
690 }
691
692 /*
693 * Reduce an iovec's length in-place. Return the resulting number of segments
694 */
695 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
696 {
697 unsigned long seg = 0;
698 size_t len = 0;
699
700 while (seg < nr_segs) {
701 seg++;
702 if (len + iov->iov_len >= to) {
703 iov->iov_len = to - len;
704 break;
705 }
706 len += iov->iov_len;
707 iov++;
708 }
709 return seg;
710 }
711 EXPORT_SYMBOL(iov_shorten);
712
713 static ssize_t do_iter_readv_writev(struct file *filp, int rw, const struct iovec *iov,
714 unsigned long nr_segs, size_t len, loff_t *ppos, iter_fn_t fn)
715 {
716 struct kiocb kiocb;
717 struct iov_iter iter;
718 ssize_t ret;
719
720 init_sync_kiocb(&kiocb, filp);
721 kiocb.ki_pos = *ppos;
722 kiocb.ki_nbytes = len;
723
724 iov_iter_init(&iter, rw, iov, nr_segs, len);
725 ret = fn(&kiocb, &iter);
726 if (ret == -EIOCBQUEUED)
727 ret = wait_on_sync_kiocb(&kiocb);
728 *ppos = kiocb.ki_pos;
729 return ret;
730 }
731
732 static ssize_t do_sync_readv_writev(struct file *filp, const struct iovec *iov,
733 unsigned long nr_segs, size_t len, loff_t *ppos, iov_fn_t fn)
734 {
735 struct kiocb kiocb;
736 ssize_t ret;
737
738 init_sync_kiocb(&kiocb, filp);
739 kiocb.ki_pos = *ppos;
740 kiocb.ki_nbytes = len;
741
742 ret = fn(&kiocb, iov, nr_segs, kiocb.ki_pos);
743 if (ret == -EIOCBQUEUED)
744 ret = wait_on_sync_kiocb(&kiocb);
745 *ppos = kiocb.ki_pos;
746 return ret;
747 }
748
749 /* Do it by hand, with file-ops */
750 static ssize_t do_loop_readv_writev(struct file *filp, struct iovec *iov,
751 unsigned long nr_segs, loff_t *ppos, io_fn_t fn)
752 {
753 struct iovec *vector = iov;
754 ssize_t ret = 0;
755
756 while (nr_segs > 0) {
757 void __user *base;
758 size_t len;
759 ssize_t nr;
760
761 base = vector->iov_base;
762 len = vector->iov_len;
763 vector++;
764 nr_segs--;
765
766 nr = fn(filp, base, len, ppos);
767
768 if (nr < 0) {
769 if (!ret)
770 ret = nr;
771 break;
772 }
773 ret += nr;
774 if (nr != len)
775 break;
776 }
777
778 return ret;
779 }
780
781 /* A write operation does a read from user space and vice versa */
782 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
783
784 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
785 unsigned long nr_segs, unsigned long fast_segs,
786 struct iovec *fast_pointer,
787 struct iovec **ret_pointer)
788 {
789 unsigned long seg;
790 ssize_t ret;
791 struct iovec *iov = fast_pointer;
792
793 /*
794 * SuS says "The readv() function *may* fail if the iovcnt argument
795 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
796 * traditionally returned zero for zero segments, so...
797 */
798 if (nr_segs == 0) {
799 ret = 0;
800 goto out;
801 }
802
803 /*
804 * First get the "struct iovec" from user memory and
805 * verify all the pointers
806 */
807 if (nr_segs > UIO_MAXIOV) {
808 ret = -EINVAL;
809 goto out;
810 }
811 if (nr_segs > fast_segs) {
812 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
813 if (iov == NULL) {
814 ret = -ENOMEM;
815 goto out;
816 }
817 }
818 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
819 ret = -EFAULT;
820 goto out;
821 }
822
823 /*
824 * According to the Single Unix Specification we should return EINVAL
825 * if an element length is < 0 when cast to ssize_t or if the
826 * total length would overflow the ssize_t return value of the
827 * system call.
828 *
829 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
830 * overflow case.
831 */
832 ret = 0;
833 for (seg = 0; seg < nr_segs; seg++) {
834 void __user *buf = iov[seg].iov_base;
835 ssize_t len = (ssize_t)iov[seg].iov_len;
836
837 /* see if we we're about to use an invalid len or if
838 * it's about to overflow ssize_t */
839 if (len < 0) {
840 ret = -EINVAL;
841 goto out;
842 }
843 if (type >= 0
844 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
845 ret = -EFAULT;
846 goto out;
847 }
848 if (len > MAX_RW_COUNT - ret) {
849 len = MAX_RW_COUNT - ret;
850 iov[seg].iov_len = len;
851 }
852 ret += len;
853 }
854 out:
855 *ret_pointer = iov;
856 return ret;
857 }
858
859 static ssize_t do_readv_writev(int type, struct file *file,
860 const struct iovec __user * uvector,
861 unsigned long nr_segs, loff_t *pos)
862 {
863 size_t tot_len;
864 struct iovec iovstack[UIO_FASTIOV];
865 struct iovec *iov = iovstack;
866 ssize_t ret;
867 io_fn_t fn;
868 iov_fn_t fnv;
869 iter_fn_t iter_fn;
870
871 ret = rw_copy_check_uvector(type, uvector, nr_segs,
872 ARRAY_SIZE(iovstack), iovstack, &iov);
873 if (ret <= 0)
874 goto out;
875
876 tot_len = ret;
877 ret = rw_verify_area(type, file, pos, tot_len);
878 if (ret < 0)
879 goto out;
880
881 fnv = NULL;
882 if (type == READ) {
883 fn = file->f_op->read;
884 fnv = file->f_op->aio_read;
885 iter_fn = file->f_op->read_iter;
886 } else {
887 fn = (io_fn_t)file->f_op->write;
888 fnv = file->f_op->aio_write;
889 iter_fn = file->f_op->write_iter;
890 file_start_write(file);
891 }
892
893 if (iter_fn)
894 ret = do_iter_readv_writev(file, type, iov, nr_segs, tot_len,
895 pos, iter_fn);
896 else if (fnv)
897 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
898 pos, fnv);
899 else
900 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
901
902 if (type != READ)
903 file_end_write(file);
904
905 out:
906 if (iov != iovstack)
907 kfree(iov);
908 if ((ret + (type == READ)) > 0) {
909 if (type == READ)
910 fsnotify_access(file);
911 else
912 fsnotify_modify(file);
913 }
914 return ret;
915 }
916
917 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
918 unsigned long vlen, loff_t *pos)
919 {
920 if (!(file->f_mode & FMODE_READ))
921 return -EBADF;
922 if (!(file->f_mode & FMODE_CAN_READ))
923 return -EINVAL;
924
925 return do_readv_writev(READ, file, vec, vlen, pos);
926 }
927
928 EXPORT_SYMBOL(vfs_readv);
929
930 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
931 unsigned long vlen, loff_t *pos)
932 {
933 if (!(file->f_mode & FMODE_WRITE))
934 return -EBADF;
935 if (!(file->f_mode & FMODE_CAN_WRITE))
936 return -EINVAL;
937
938 return do_readv_writev(WRITE, file, vec, vlen, pos);
939 }
940
941 EXPORT_SYMBOL(vfs_writev);
942
943 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
944 unsigned long, vlen)
945 {
946 struct fd f = fdget_pos(fd);
947 ssize_t ret = -EBADF;
948
949 if (f.file) {
950 loff_t pos = file_pos_read(f.file);
951 ret = vfs_readv(f.file, vec, vlen, &pos);
952 if (ret >= 0)
953 file_pos_write(f.file, pos);
954 fdput_pos(f);
955 }
956
957 if (ret > 0)
958 add_rchar(current, ret);
959 inc_syscr(current);
960 return ret;
961 }
962
963 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
964 unsigned long, vlen)
965 {
966 struct fd f = fdget_pos(fd);
967 ssize_t ret = -EBADF;
968
969 if (f.file) {
970 loff_t pos = file_pos_read(f.file);
971 ret = vfs_writev(f.file, vec, vlen, &pos);
972 if (ret >= 0)
973 file_pos_write(f.file, pos);
974 fdput_pos(f);
975 }
976
977 if (ret > 0)
978 add_wchar(current, ret);
979 inc_syscw(current);
980 return ret;
981 }
982
983 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
984 {
985 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
986 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
987 }
988
989 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
990 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
991 {
992 loff_t pos = pos_from_hilo(pos_h, pos_l);
993 struct fd f;
994 ssize_t ret = -EBADF;
995
996 if (pos < 0)
997 return -EINVAL;
998
999 f = fdget(fd);
1000 if (f.file) {
1001 ret = -ESPIPE;
1002 if (f.file->f_mode & FMODE_PREAD)
1003 ret = vfs_readv(f.file, vec, vlen, &pos);
1004 fdput(f);
1005 }
1006
1007 if (ret > 0)
1008 add_rchar(current, ret);
1009 inc_syscr(current);
1010 return ret;
1011 }
1012
1013 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
1014 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
1015 {
1016 loff_t pos = pos_from_hilo(pos_h, pos_l);
1017 struct fd f;
1018 ssize_t ret = -EBADF;
1019
1020 if (pos < 0)
1021 return -EINVAL;
1022
1023 f = fdget(fd);
1024 if (f.file) {
1025 ret = -ESPIPE;
1026 if (f.file->f_mode & FMODE_PWRITE)
1027 ret = vfs_writev(f.file, vec, vlen, &pos);
1028 fdput(f);
1029 }
1030
1031 if (ret > 0)
1032 add_wchar(current, ret);
1033 inc_syscw(current);
1034 return ret;
1035 }
1036
1037 #ifdef CONFIG_COMPAT
1038
1039 static ssize_t compat_do_readv_writev(int type, struct file *file,
1040 const struct compat_iovec __user *uvector,
1041 unsigned long nr_segs, loff_t *pos)
1042 {
1043 compat_ssize_t tot_len;
1044 struct iovec iovstack[UIO_FASTIOV];
1045 struct iovec *iov = iovstack;
1046 ssize_t ret;
1047 io_fn_t fn;
1048 iov_fn_t fnv;
1049 iter_fn_t iter_fn;
1050
1051 ret = compat_rw_copy_check_uvector(type, uvector, nr_segs,
1052 UIO_FASTIOV, iovstack, &iov);
1053 if (ret <= 0)
1054 goto out;
1055
1056 tot_len = ret;
1057 ret = rw_verify_area(type, file, pos, tot_len);
1058 if (ret < 0)
1059 goto out;
1060
1061 fnv = NULL;
1062 if (type == READ) {
1063 fn = file->f_op->read;
1064 fnv = file->f_op->aio_read;
1065 iter_fn = file->f_op->read_iter;
1066 } else {
1067 fn = (io_fn_t)file->f_op->write;
1068 fnv = file->f_op->aio_write;
1069 iter_fn = file->f_op->write_iter;
1070 file_start_write(file);
1071 }
1072
1073 if (iter_fn)
1074 ret = do_iter_readv_writev(file, type, iov, nr_segs, tot_len,
1075 pos, iter_fn);
1076 else if (fnv)
1077 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
1078 pos, fnv);
1079 else
1080 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
1081
1082 if (type != READ)
1083 file_end_write(file);
1084
1085 out:
1086 if (iov != iovstack)
1087 kfree(iov);
1088 if ((ret + (type == READ)) > 0) {
1089 if (type == READ)
1090 fsnotify_access(file);
1091 else
1092 fsnotify_modify(file);
1093 }
1094 return ret;
1095 }
1096
1097 static size_t compat_readv(struct file *file,
1098 const struct compat_iovec __user *vec,
1099 unsigned long vlen, loff_t *pos)
1100 {
1101 ssize_t ret = -EBADF;
1102
1103 if (!(file->f_mode & FMODE_READ))
1104 goto out;
1105
1106 ret = -EINVAL;
1107 if (!(file->f_mode & FMODE_CAN_READ))
1108 goto out;
1109
1110 ret = compat_do_readv_writev(READ, file, vec, vlen, pos);
1111
1112 out:
1113 if (ret > 0)
1114 add_rchar(current, ret);
1115 inc_syscr(current);
1116 return ret;
1117 }
1118
1119 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1120 const struct compat_iovec __user *,vec,
1121 compat_ulong_t, vlen)
1122 {
1123 struct fd f = fdget_pos(fd);
1124 ssize_t ret;
1125 loff_t pos;
1126
1127 if (!f.file)
1128 return -EBADF;
1129 pos = f.file->f_pos;
1130 ret = compat_readv(f.file, vec, vlen, &pos);
1131 if (ret >= 0)
1132 f.file->f_pos = pos;
1133 fdput_pos(f);
1134 return ret;
1135 }
1136
1137 static long __compat_sys_preadv64(unsigned long fd,
1138 const struct compat_iovec __user *vec,
1139 unsigned long vlen, loff_t pos)
1140 {
1141 struct fd f;
1142 ssize_t ret;
1143
1144 if (pos < 0)
1145 return -EINVAL;
1146 f = fdget(fd);
1147 if (!f.file)
1148 return -EBADF;
1149 ret = -ESPIPE;
1150 if (f.file->f_mode & FMODE_PREAD)
1151 ret = compat_readv(f.file, vec, vlen, &pos);
1152 fdput(f);
1153 return ret;
1154 }
1155
1156 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1157 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1158 const struct compat_iovec __user *,vec,
1159 unsigned long, vlen, loff_t, pos)
1160 {
1161 return __compat_sys_preadv64(fd, vec, vlen, pos);
1162 }
1163 #endif
1164
1165 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1166 const struct compat_iovec __user *,vec,
1167 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1168 {
1169 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1170
1171 return __compat_sys_preadv64(fd, vec, vlen, pos);
1172 }
1173
1174 static size_t compat_writev(struct file *file,
1175 const struct compat_iovec __user *vec,
1176 unsigned long vlen, loff_t *pos)
1177 {
1178 ssize_t ret = -EBADF;
1179
1180 if (!(file->f_mode & FMODE_WRITE))
1181 goto out;
1182
1183 ret = -EINVAL;
1184 if (!(file->f_mode & FMODE_CAN_WRITE))
1185 goto out;
1186
1187 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos);
1188
1189 out:
1190 if (ret > 0)
1191 add_wchar(current, ret);
1192 inc_syscw(current);
1193 return ret;
1194 }
1195
1196 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1197 const struct compat_iovec __user *, vec,
1198 compat_ulong_t, vlen)
1199 {
1200 struct fd f = fdget_pos(fd);
1201 ssize_t ret;
1202 loff_t pos;
1203
1204 if (!f.file)
1205 return -EBADF;
1206 pos = f.file->f_pos;
1207 ret = compat_writev(f.file, vec, vlen, &pos);
1208 if (ret >= 0)
1209 f.file->f_pos = pos;
1210 fdput_pos(f);
1211 return ret;
1212 }
1213
1214 static long __compat_sys_pwritev64(unsigned long fd,
1215 const struct compat_iovec __user *vec,
1216 unsigned long vlen, loff_t pos)
1217 {
1218 struct fd f;
1219 ssize_t ret;
1220
1221 if (pos < 0)
1222 return -EINVAL;
1223 f = fdget(fd);
1224 if (!f.file)
1225 return -EBADF;
1226 ret = -ESPIPE;
1227 if (f.file->f_mode & FMODE_PWRITE)
1228 ret = compat_writev(f.file, vec, vlen, &pos);
1229 fdput(f);
1230 return ret;
1231 }
1232
1233 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1234 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1235 const struct compat_iovec __user *,vec,
1236 unsigned long, vlen, loff_t, pos)
1237 {
1238 return __compat_sys_pwritev64(fd, vec, vlen, pos);
1239 }
1240 #endif
1241
1242 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1243 const struct compat_iovec __user *,vec,
1244 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1245 {
1246 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1247
1248 return __compat_sys_pwritev64(fd, vec, vlen, pos);
1249 }
1250 #endif
1251
1252 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1253 size_t count, loff_t max)
1254 {
1255 struct fd in, out;
1256 struct inode *in_inode, *out_inode;
1257 loff_t pos;
1258 loff_t out_pos;
1259 ssize_t retval;
1260 int fl;
1261
1262 /*
1263 * Get input file, and verify that it is ok..
1264 */
1265 retval = -EBADF;
1266 in = fdget(in_fd);
1267 if (!in.file)
1268 goto out;
1269 if (!(in.file->f_mode & FMODE_READ))
1270 goto fput_in;
1271 retval = -ESPIPE;
1272 if (!ppos) {
1273 pos = in.file->f_pos;
1274 } else {
1275 pos = *ppos;
1276 if (!(in.file->f_mode & FMODE_PREAD))
1277 goto fput_in;
1278 }
1279 retval = rw_verify_area(READ, in.file, &pos, count);
1280 if (retval < 0)
1281 goto fput_in;
1282 count = retval;
1283
1284 /*
1285 * Get output file, and verify that it is ok..
1286 */
1287 retval = -EBADF;
1288 out = fdget(out_fd);
1289 if (!out.file)
1290 goto fput_in;
1291 if (!(out.file->f_mode & FMODE_WRITE))
1292 goto fput_out;
1293 retval = -EINVAL;
1294 in_inode = file_inode(in.file);
1295 out_inode = file_inode(out.file);
1296 out_pos = out.file->f_pos;
1297 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1298 if (retval < 0)
1299 goto fput_out;
1300 count = retval;
1301
1302 if (!max)
1303 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1304
1305 if (unlikely(pos + count > max)) {
1306 retval = -EOVERFLOW;
1307 if (pos >= max)
1308 goto fput_out;
1309 count = max - pos;
1310 }
1311
1312 fl = 0;
1313 #if 0
1314 /*
1315 * We need to debate whether we can enable this or not. The
1316 * man page documents EAGAIN return for the output at least,
1317 * and the application is arguably buggy if it doesn't expect
1318 * EAGAIN on a non-blocking file descriptor.
1319 */
1320 if (in.file->f_flags & O_NONBLOCK)
1321 fl = SPLICE_F_NONBLOCK;
1322 #endif
1323 file_start_write(out.file);
1324 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1325 file_end_write(out.file);
1326
1327 if (retval > 0) {
1328 add_rchar(current, retval);
1329 add_wchar(current, retval);
1330 fsnotify_access(in.file);
1331 fsnotify_modify(out.file);
1332 out.file->f_pos = out_pos;
1333 if (ppos)
1334 *ppos = pos;
1335 else
1336 in.file->f_pos = pos;
1337 }
1338
1339 inc_syscr(current);
1340 inc_syscw(current);
1341 if (pos > max)
1342 retval = -EOVERFLOW;
1343
1344 fput_out:
1345 fdput(out);
1346 fput_in:
1347 fdput(in);
1348 out:
1349 return retval;
1350 }
1351
1352 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1353 {
1354 loff_t pos;
1355 off_t off;
1356 ssize_t ret;
1357
1358 if (offset) {
1359 if (unlikely(get_user(off, offset)))
1360 return -EFAULT;
1361 pos = off;
1362 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1363 if (unlikely(put_user(pos, offset)))
1364 return -EFAULT;
1365 return ret;
1366 }
1367
1368 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1369 }
1370
1371 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1372 {
1373 loff_t pos;
1374 ssize_t ret;
1375
1376 if (offset) {
1377 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1378 return -EFAULT;
1379 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1380 if (unlikely(put_user(pos, offset)))
1381 return -EFAULT;
1382 return ret;
1383 }
1384
1385 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1386 }
1387
1388 #ifdef CONFIG_COMPAT
1389 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1390 compat_off_t __user *, offset, compat_size_t, count)
1391 {
1392 loff_t pos;
1393 off_t off;
1394 ssize_t ret;
1395
1396 if (offset) {
1397 if (unlikely(get_user(off, offset)))
1398 return -EFAULT;
1399 pos = off;
1400 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1401 if (unlikely(put_user(pos, offset)))
1402 return -EFAULT;
1403 return ret;
1404 }
1405
1406 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1407 }
1408
1409 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1410 compat_loff_t __user *, offset, compat_size_t, count)
1411 {
1412 loff_t pos;
1413 ssize_t ret;
1414
1415 if (offset) {
1416 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1417 return -EFAULT;
1418 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1419 if (unlikely(put_user(pos, offset)))
1420 return -EFAULT;
1421 return ret;
1422 }
1423
1424 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1425 }
1426 #endif
Linux kernel - Deliverables
This page took 0.059919 seconds and 5 git commands to generate.