projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'smack-for-3.19' of git://git.gitorious.org/smack-next/kernel into next
[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 /*
337 * rw_verify_area doesn't like huge counts. We limit
338 * them to something that fits in "int" so that others
339 * won't have to do range checks all the time.
340 */
341 int rw_verify_area(int read_write, struct file *file, const loff_t *ppos, size_t count)
342 {
343 struct inode *inode;
344 loff_t pos;
345 int retval = -EINVAL;
346
347 inode = file_inode(file);
348 if (unlikely((ssize_t) count < 0))
349 return retval;
350 pos = *ppos;
351 if (unlikely(pos < 0)) {
352 if (!unsigned_offsets(file))
353 return retval;
354 if (count >= -pos) /* both values are in 0..LLONG_MAX */
355 return -EOVERFLOW;
356 } else if (unlikely((loff_t) (pos + count) < 0)) {
357 if (!unsigned_offsets(file))
358 return retval;
359 }
360
361 if (unlikely(inode->i_flock && mandatory_lock(inode))) {
362 retval = locks_mandatory_area(
363 read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE,
364 inode, file, pos, count);
365 if (retval < 0)
366 return retval;
367 }
368 retval = security_file_permission(file,
369 read_write == READ ? MAY_READ : MAY_WRITE);
370 if (retval)
371 return retval;
372 return count > MAX_RW_COUNT ? MAX_RW_COUNT : count;
373 }
374
375 ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
376 {
377 struct iovec iov = { .iov_base = buf, .iov_len = len };
378 struct kiocb kiocb;
379 ssize_t ret;
380
381 init_sync_kiocb(&kiocb, filp);
382 kiocb.ki_pos = *ppos;
383 kiocb.ki_nbytes = len;
384
385 ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
386 if (-EIOCBQUEUED == ret)
387 ret = wait_on_sync_kiocb(&kiocb);
388 *ppos = kiocb.ki_pos;
389 return ret;
390 }
391
392 EXPORT_SYMBOL(do_sync_read);
393
394 ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
395 {
396 struct iovec iov = { .iov_base = buf, .iov_len = len };
397 struct kiocb kiocb;
398 struct iov_iter iter;
399 ssize_t ret;
400
401 init_sync_kiocb(&kiocb, filp);
402 kiocb.ki_pos = *ppos;
403 kiocb.ki_nbytes = len;
404 iov_iter_init(&iter, READ, &iov, 1, len);
405
406 ret = filp->f_op->read_iter(&kiocb, &iter);
407 if (-EIOCBQUEUED == ret)
408 ret = wait_on_sync_kiocb(&kiocb);
409 *ppos = kiocb.ki_pos;
410 return ret;
411 }
412
413 EXPORT_SYMBOL(new_sync_read);
414
415 ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
416 loff_t *pos)
417 {
418 ssize_t ret;
419
420 if (file->f_op->read)
421 ret = file->f_op->read(file, buf, count, pos);
422 else if (file->f_op->aio_read)
423 ret = do_sync_read(file, buf, count, pos);
424 else if (file->f_op->read_iter)
425 ret = new_sync_read(file, buf, count, pos);
426 else
427 ret = -EINVAL;
428
429 return ret;
430 }
431
432 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
433 {
434 ssize_t ret;
435
436 if (!(file->f_mode & FMODE_READ))
437 return -EBADF;
438 if (!(file->f_mode & FMODE_CAN_READ))
439 return -EINVAL;
440 if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
441 return -EFAULT;
442
443 ret = rw_verify_area(READ, file, pos, count);
444 if (ret >= 0) {
445 count = ret;
446 ret = __vfs_read(file, buf, count, pos);
447 if (ret > 0) {
448 fsnotify_access(file);
449 add_rchar(current, ret);
450 }
451 inc_syscr(current);
452 }
453
454 return ret;
455 }
456
457 EXPORT_SYMBOL(vfs_read);
458
459 ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
460 {
461 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
462 struct kiocb kiocb;
463 ssize_t ret;
464
465 init_sync_kiocb(&kiocb, filp);
466 kiocb.ki_pos = *ppos;
467 kiocb.ki_nbytes = len;
468
469 ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
470 if (-EIOCBQUEUED == ret)
471 ret = wait_on_sync_kiocb(&kiocb);
472 *ppos = kiocb.ki_pos;
473 return ret;
474 }
475
476 EXPORT_SYMBOL(do_sync_write);
477
478 ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
479 {
480 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = len };
481 struct kiocb kiocb;
482 struct iov_iter iter;
483 ssize_t ret;
484
485 init_sync_kiocb(&kiocb, filp);
486 kiocb.ki_pos = *ppos;
487 kiocb.ki_nbytes = len;
488 iov_iter_init(&iter, WRITE, &iov, 1, len);
489
490 ret = filp->f_op->write_iter(&kiocb, &iter);
491 if (-EIOCBQUEUED == ret)
492 ret = wait_on_sync_kiocb(&kiocb);
493 *ppos = kiocb.ki_pos;
494 return ret;
495 }
496
497 EXPORT_SYMBOL(new_sync_write);
498
499 ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
500 {
501 mm_segment_t old_fs;
502 const char __user *p;
503 ssize_t ret;
504
505 if (!(file->f_mode & FMODE_CAN_WRITE))
506 return -EINVAL;
507
508 old_fs = get_fs();
509 set_fs(get_ds());
510 p = (__force const char __user *)buf;
511 if (count > MAX_RW_COUNT)
512 count = MAX_RW_COUNT;
513 if (file->f_op->write)
514 ret = file->f_op->write(file, p, count, pos);
515 else if (file->f_op->aio_write)
516 ret = do_sync_write(file, p, count, pos);
517 else
518 ret = new_sync_write(file, p, count, pos);
519 set_fs(old_fs);
520 if (ret > 0) {
521 fsnotify_modify(file);
522 add_wchar(current, ret);
523 }
524 inc_syscw(current);
525 return ret;
526 }
527
528 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
529 {
530 ssize_t ret;
531
532 if (!(file->f_mode & FMODE_WRITE))
533 return -EBADF;
534 if (!(file->f_mode & FMODE_CAN_WRITE))
535 return -EINVAL;
536 if (unlikely(!access_ok(VERIFY_READ, buf, count)))
537 return -EFAULT;
538
539 ret = rw_verify_area(WRITE, file, pos, count);
540 if (ret >= 0) {
541 count = ret;
542 file_start_write(file);
543 if (file->f_op->write)
544 ret = file->f_op->write(file, buf, count, pos);
545 else if (file->f_op->aio_write)
546 ret = do_sync_write(file, buf, count, pos);
547 else
548 ret = new_sync_write(file, buf, count, pos);
549 if (ret > 0) {
550 fsnotify_modify(file);
551 add_wchar(current, ret);
552 }
553 inc_syscw(current);
554 file_end_write(file);
555 }
556
557 return ret;
558 }
559
560 EXPORT_SYMBOL(vfs_write);
561
562 static inline loff_t file_pos_read(struct file *file)
563 {
564 return file->f_pos;
565 }
566
567 static inline void file_pos_write(struct file *file, loff_t pos)
568 {
569 file->f_pos = pos;
570 }
571
572 SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
573 {
574 struct fd f = fdget_pos(fd);
575 ssize_t ret = -EBADF;
576
577 if (f.file) {
578 loff_t pos = file_pos_read(f.file);
579 ret = vfs_read(f.file, buf, count, &pos);
580 if (ret >= 0)
581 file_pos_write(f.file, pos);
582 fdput_pos(f);
583 }
584 return ret;
585 }
586
587 SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
588 size_t, count)
589 {
590 struct fd f = fdget_pos(fd);
591 ssize_t ret = -EBADF;
592
593 if (f.file) {
594 loff_t pos = file_pos_read(f.file);
595 ret = vfs_write(f.file, buf, count, &pos);
596 if (ret >= 0)
597 file_pos_write(f.file, pos);
598 fdput_pos(f);
599 }
600
601 return ret;
602 }
603
604 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
605 size_t, count, loff_t, pos)
606 {
607 struct fd f;
608 ssize_t ret = -EBADF;
609
610 if (pos < 0)
611 return -EINVAL;
612
613 f = fdget(fd);
614 if (f.file) {
615 ret = -ESPIPE;
616 if (f.file->f_mode & FMODE_PREAD)
617 ret = vfs_read(f.file, buf, count, &pos);
618 fdput(f);
619 }
620
621 return ret;
622 }
623
624 SYSCALL_DEFINE4(pwrite64, unsigned int, fd, const char __user *, buf,
625 size_t, count, loff_t, pos)
626 {
627 struct fd f;
628 ssize_t ret = -EBADF;
629
630 if (pos < 0)
631 return -EINVAL;
632
633 f = fdget(fd);
634 if (f.file) {
635 ret = -ESPIPE;
636 if (f.file->f_mode & FMODE_PWRITE)
637 ret = vfs_write(f.file, buf, count, &pos);
638 fdput(f);
639 }
640
641 return ret;
642 }
643
644 /*
645 * Reduce an iovec's length in-place. Return the resulting number of segments
646 */
647 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
648 {
649 unsigned long seg = 0;
650 size_t len = 0;
651
652 while (seg < nr_segs) {
653 seg++;
654 if (len + iov->iov_len >= to) {
655 iov->iov_len = to - len;
656 break;
657 }
658 len += iov->iov_len;
659 iov++;
660 }
661 return seg;
662 }
663 EXPORT_SYMBOL(iov_shorten);
664
665 static ssize_t do_iter_readv_writev(struct file *filp, int rw, const struct iovec *iov,
666 unsigned long nr_segs, size_t len, loff_t *ppos, iter_fn_t fn)
667 {
668 struct kiocb kiocb;
669 struct iov_iter iter;
670 ssize_t ret;
671
672 init_sync_kiocb(&kiocb, filp);
673 kiocb.ki_pos = *ppos;
674 kiocb.ki_nbytes = len;
675
676 iov_iter_init(&iter, rw, iov, nr_segs, len);
677 ret = fn(&kiocb, &iter);
678 if (ret == -EIOCBQUEUED)
679 ret = wait_on_sync_kiocb(&kiocb);
680 *ppos = kiocb.ki_pos;
681 return ret;
682 }
683
684 static ssize_t do_sync_readv_writev(struct file *filp, const struct iovec *iov,
685 unsigned long nr_segs, size_t len, loff_t *ppos, iov_fn_t fn)
686 {
687 struct kiocb kiocb;
688 ssize_t ret;
689
690 init_sync_kiocb(&kiocb, filp);
691 kiocb.ki_pos = *ppos;
692 kiocb.ki_nbytes = len;
693
694 ret = fn(&kiocb, iov, nr_segs, kiocb.ki_pos);
695 if (ret == -EIOCBQUEUED)
696 ret = wait_on_sync_kiocb(&kiocb);
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 iovec *iov,
703 unsigned long nr_segs, loff_t *ppos, io_fn_t fn)
704 {
705 struct iovec *vector = iov;
706 ssize_t ret = 0;
707
708 while (nr_segs > 0) {
709 void __user *base;
710 size_t len;
711 ssize_t nr;
712
713 base = vector->iov_base;
714 len = vector->iov_len;
715 vector++;
716 nr_segs--;
717
718 nr = fn(filp, base, len, ppos);
719
720 if (nr < 0) {
721 if (!ret)
722 ret = nr;
723 break;
724 }
725 ret += nr;
726 if (nr != len)
727 break;
728 }
729
730 return ret;
731 }
732
733 /* A write operation does a read from user space and vice versa */
734 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
735
736 ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
737 unsigned long nr_segs, unsigned long fast_segs,
738 struct iovec *fast_pointer,
739 struct iovec **ret_pointer)
740 {
741 unsigned long seg;
742 ssize_t ret;
743 struct iovec *iov = fast_pointer;
744
745 /*
746 * SuS says "The readv() function *may* fail if the iovcnt argument
747 * was less than or equal to 0, or greater than {IOV_MAX}. Linux has
748 * traditionally returned zero for zero segments, so...
749 */
750 if (nr_segs == 0) {
751 ret = 0;
752 goto out;
753 }
754
755 /*
756 * First get the "struct iovec" from user memory and
757 * verify all the pointers
758 */
759 if (nr_segs > UIO_MAXIOV) {
760 ret = -EINVAL;
761 goto out;
762 }
763 if (nr_segs > fast_segs) {
764 iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
765 if (iov == NULL) {
766 ret = -ENOMEM;
767 goto out;
768 }
769 }
770 if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector))) {
771 ret = -EFAULT;
772 goto out;
773 }
774
775 /*
776 * According to the Single Unix Specification we should return EINVAL
777 * if an element length is < 0 when cast to ssize_t or if the
778 * total length would overflow the ssize_t return value of the
779 * system call.
780 *
781 * Linux caps all read/write calls to MAX_RW_COUNT, and avoids the
782 * overflow case.
783 */
784 ret = 0;
785 for (seg = 0; seg < nr_segs; seg++) {
786 void __user *buf = iov[seg].iov_base;
787 ssize_t len = (ssize_t)iov[seg].iov_len;
788
789 /* see if we we're about to use an invalid len or if
790 * it's about to overflow ssize_t */
791 if (len < 0) {
792 ret = -EINVAL;
793 goto out;
794 }
795 if (type >= 0
796 && unlikely(!access_ok(vrfy_dir(type), buf, len))) {
797 ret = -EFAULT;
798 goto out;
799 }
800 if (len > MAX_RW_COUNT - ret) {
801 len = MAX_RW_COUNT - ret;
802 iov[seg].iov_len = len;
803 }
804 ret += len;
805 }
806 out:
807 *ret_pointer = iov;
808 return ret;
809 }
810
811 static ssize_t do_readv_writev(int type, struct file *file,
812 const struct iovec __user * uvector,
813 unsigned long nr_segs, loff_t *pos)
814 {
815 size_t tot_len;
816 struct iovec iovstack[UIO_FASTIOV];
817 struct iovec *iov = iovstack;
818 ssize_t ret;
819 io_fn_t fn;
820 iov_fn_t fnv;
821 iter_fn_t iter_fn;
822
823 ret = rw_copy_check_uvector(type, uvector, nr_segs,
824 ARRAY_SIZE(iovstack), iovstack, &iov);
825 if (ret <= 0)
826 goto out;
827
828 tot_len = ret;
829 ret = rw_verify_area(type, file, pos, tot_len);
830 if (ret < 0)
831 goto out;
832
833 fnv = NULL;
834 if (type == READ) {
835 fn = file->f_op->read;
836 fnv = file->f_op->aio_read;
837 iter_fn = file->f_op->read_iter;
838 } else {
839 fn = (io_fn_t)file->f_op->write;
840 fnv = file->f_op->aio_write;
841 iter_fn = file->f_op->write_iter;
842 file_start_write(file);
843 }
844
845 if (iter_fn)
846 ret = do_iter_readv_writev(file, type, iov, nr_segs, tot_len,
847 pos, iter_fn);
848 else if (fnv)
849 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
850 pos, fnv);
851 else
852 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
853
854 if (type != READ)
855 file_end_write(file);
856
857 out:
858 if (iov != iovstack)
859 kfree(iov);
860 if ((ret + (type == READ)) > 0) {
861 if (type == READ)
862 fsnotify_access(file);
863 else
864 fsnotify_modify(file);
865 }
866 return ret;
867 }
868
869 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
870 unsigned long vlen, loff_t *pos)
871 {
872 if (!(file->f_mode & FMODE_READ))
873 return -EBADF;
874 if (!(file->f_mode & FMODE_CAN_READ))
875 return -EINVAL;
876
877 return do_readv_writev(READ, file, vec, vlen, pos);
878 }
879
880 EXPORT_SYMBOL(vfs_readv);
881
882 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
883 unsigned long vlen, loff_t *pos)
884 {
885 if (!(file->f_mode & FMODE_WRITE))
886 return -EBADF;
887 if (!(file->f_mode & FMODE_CAN_WRITE))
888 return -EINVAL;
889
890 return do_readv_writev(WRITE, file, vec, vlen, pos);
891 }
892
893 EXPORT_SYMBOL(vfs_writev);
894
895 SYSCALL_DEFINE3(readv, unsigned long, fd, const struct iovec __user *, vec,
896 unsigned long, vlen)
897 {
898 struct fd f = fdget_pos(fd);
899 ssize_t ret = -EBADF;
900
901 if (f.file) {
902 loff_t pos = file_pos_read(f.file);
903 ret = vfs_readv(f.file, vec, vlen, &pos);
904 if (ret >= 0)
905 file_pos_write(f.file, pos);
906 fdput_pos(f);
907 }
908
909 if (ret > 0)
910 add_rchar(current, ret);
911 inc_syscr(current);
912 return ret;
913 }
914
915 SYSCALL_DEFINE3(writev, unsigned long, fd, const struct iovec __user *, vec,
916 unsigned long, vlen)
917 {
918 struct fd f = fdget_pos(fd);
919 ssize_t ret = -EBADF;
920
921 if (f.file) {
922 loff_t pos = file_pos_read(f.file);
923 ret = vfs_writev(f.file, vec, vlen, &pos);
924 if (ret >= 0)
925 file_pos_write(f.file, pos);
926 fdput_pos(f);
927 }
928
929 if (ret > 0)
930 add_wchar(current, ret);
931 inc_syscw(current);
932 return ret;
933 }
934
935 static inline loff_t pos_from_hilo(unsigned long high, unsigned long low)
936 {
937 #define HALF_LONG_BITS (BITS_PER_LONG / 2)
938 return (((loff_t)high << HALF_LONG_BITS) << HALF_LONG_BITS) | low;
939 }
940
941 SYSCALL_DEFINE5(preadv, unsigned long, fd, const struct iovec __user *, vec,
942 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
943 {
944 loff_t pos = pos_from_hilo(pos_h, pos_l);
945 struct fd f;
946 ssize_t ret = -EBADF;
947
948 if (pos < 0)
949 return -EINVAL;
950
951 f = fdget(fd);
952 if (f.file) {
953 ret = -ESPIPE;
954 if (f.file->f_mode & FMODE_PREAD)
955 ret = vfs_readv(f.file, vec, vlen, &pos);
956 fdput(f);
957 }
958
959 if (ret > 0)
960 add_rchar(current, ret);
961 inc_syscr(current);
962 return ret;
963 }
964
965 SYSCALL_DEFINE5(pwritev, unsigned long, fd, const struct iovec __user *, vec,
966 unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h)
967 {
968 loff_t pos = pos_from_hilo(pos_h, pos_l);
969 struct fd f;
970 ssize_t ret = -EBADF;
971
972 if (pos < 0)
973 return -EINVAL;
974
975 f = fdget(fd);
976 if (f.file) {
977 ret = -ESPIPE;
978 if (f.file->f_mode & FMODE_PWRITE)
979 ret = vfs_writev(f.file, vec, vlen, &pos);
980 fdput(f);
981 }
982
983 if (ret > 0)
984 add_wchar(current, ret);
985 inc_syscw(current);
986 return ret;
987 }
988
989 #ifdef CONFIG_COMPAT
990
991 static ssize_t compat_do_readv_writev(int type, struct file *file,
992 const struct compat_iovec __user *uvector,
993 unsigned long nr_segs, loff_t *pos)
994 {
995 compat_ssize_t tot_len;
996 struct iovec iovstack[UIO_FASTIOV];
997 struct iovec *iov = iovstack;
998 ssize_t ret;
999 io_fn_t fn;
1000 iov_fn_t fnv;
1001 iter_fn_t iter_fn;
1002
1003 ret = compat_rw_copy_check_uvector(type, uvector, nr_segs,
1004 UIO_FASTIOV, iovstack, &iov);
1005 if (ret <= 0)
1006 goto out;
1007
1008 tot_len = ret;
1009 ret = rw_verify_area(type, file, pos, tot_len);
1010 if (ret < 0)
1011 goto out;
1012
1013 fnv = NULL;
1014 if (type == READ) {
1015 fn = file->f_op->read;
1016 fnv = file->f_op->aio_read;
1017 iter_fn = file->f_op->read_iter;
1018 } else {
1019 fn = (io_fn_t)file->f_op->write;
1020 fnv = file->f_op->aio_write;
1021 iter_fn = file->f_op->write_iter;
1022 file_start_write(file);
1023 }
1024
1025 if (iter_fn)
1026 ret = do_iter_readv_writev(file, type, iov, nr_segs, tot_len,
1027 pos, iter_fn);
1028 else if (fnv)
1029 ret = do_sync_readv_writev(file, iov, nr_segs, tot_len,
1030 pos, fnv);
1031 else
1032 ret = do_loop_readv_writev(file, iov, nr_segs, pos, fn);
1033
1034 if (type != READ)
1035 file_end_write(file);
1036
1037 out:
1038 if (iov != iovstack)
1039 kfree(iov);
1040 if ((ret + (type == READ)) > 0) {
1041 if (type == READ)
1042 fsnotify_access(file);
1043 else
1044 fsnotify_modify(file);
1045 }
1046 return ret;
1047 }
1048
1049 static size_t compat_readv(struct file *file,
1050 const struct compat_iovec __user *vec,
1051 unsigned long vlen, loff_t *pos)
1052 {
1053 ssize_t ret = -EBADF;
1054
1055 if (!(file->f_mode & FMODE_READ))
1056 goto out;
1057
1058 ret = -EINVAL;
1059 if (!(file->f_mode & FMODE_CAN_READ))
1060 goto out;
1061
1062 ret = compat_do_readv_writev(READ, file, vec, vlen, pos);
1063
1064 out:
1065 if (ret > 0)
1066 add_rchar(current, ret);
1067 inc_syscr(current);
1068 return ret;
1069 }
1070
1071 COMPAT_SYSCALL_DEFINE3(readv, compat_ulong_t, fd,
1072 const struct compat_iovec __user *,vec,
1073 compat_ulong_t, vlen)
1074 {
1075 struct fd f = fdget_pos(fd);
1076 ssize_t ret;
1077 loff_t pos;
1078
1079 if (!f.file)
1080 return -EBADF;
1081 pos = f.file->f_pos;
1082 ret = compat_readv(f.file, vec, vlen, &pos);
1083 if (ret >= 0)
1084 f.file->f_pos = pos;
1085 fdput_pos(f);
1086 return ret;
1087 }
1088
1089 static long __compat_sys_preadv64(unsigned long fd,
1090 const struct compat_iovec __user *vec,
1091 unsigned long vlen, loff_t pos)
1092 {
1093 struct fd f;
1094 ssize_t ret;
1095
1096 if (pos < 0)
1097 return -EINVAL;
1098 f = fdget(fd);
1099 if (!f.file)
1100 return -EBADF;
1101 ret = -ESPIPE;
1102 if (f.file->f_mode & FMODE_PREAD)
1103 ret = compat_readv(f.file, vec, vlen, &pos);
1104 fdput(f);
1105 return ret;
1106 }
1107
1108 #ifdef __ARCH_WANT_COMPAT_SYS_PREADV64
1109 COMPAT_SYSCALL_DEFINE4(preadv64, unsigned long, fd,
1110 const struct compat_iovec __user *,vec,
1111 unsigned long, vlen, loff_t, pos)
1112 {
1113 return __compat_sys_preadv64(fd, vec, vlen, pos);
1114 }
1115 #endif
1116
1117 COMPAT_SYSCALL_DEFINE5(preadv, compat_ulong_t, fd,
1118 const struct compat_iovec __user *,vec,
1119 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1120 {
1121 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1122
1123 return __compat_sys_preadv64(fd, vec, vlen, pos);
1124 }
1125
1126 static size_t compat_writev(struct file *file,
1127 const struct compat_iovec __user *vec,
1128 unsigned long vlen, loff_t *pos)
1129 {
1130 ssize_t ret = -EBADF;
1131
1132 if (!(file->f_mode & FMODE_WRITE))
1133 goto out;
1134
1135 ret = -EINVAL;
1136 if (!(file->f_mode & FMODE_CAN_WRITE))
1137 goto out;
1138
1139 ret = compat_do_readv_writev(WRITE, file, vec, vlen, pos);
1140
1141 out:
1142 if (ret > 0)
1143 add_wchar(current, ret);
1144 inc_syscw(current);
1145 return ret;
1146 }
1147
1148 COMPAT_SYSCALL_DEFINE3(writev, compat_ulong_t, fd,
1149 const struct compat_iovec __user *, vec,
1150 compat_ulong_t, vlen)
1151 {
1152 struct fd f = fdget_pos(fd);
1153 ssize_t ret;
1154 loff_t pos;
1155
1156 if (!f.file)
1157 return -EBADF;
1158 pos = f.file->f_pos;
1159 ret = compat_writev(f.file, vec, vlen, &pos);
1160 if (ret >= 0)
1161 f.file->f_pos = pos;
1162 fdput_pos(f);
1163 return ret;
1164 }
1165
1166 static long __compat_sys_pwritev64(unsigned long fd,
1167 const struct compat_iovec __user *vec,
1168 unsigned long vlen, loff_t pos)
1169 {
1170 struct fd f;
1171 ssize_t ret;
1172
1173 if (pos < 0)
1174 return -EINVAL;
1175 f = fdget(fd);
1176 if (!f.file)
1177 return -EBADF;
1178 ret = -ESPIPE;
1179 if (f.file->f_mode & FMODE_PWRITE)
1180 ret = compat_writev(f.file, vec, vlen, &pos);
1181 fdput(f);
1182 return ret;
1183 }
1184
1185 #ifdef __ARCH_WANT_COMPAT_SYS_PWRITEV64
1186 COMPAT_SYSCALL_DEFINE4(pwritev64, unsigned long, fd,
1187 const struct compat_iovec __user *,vec,
1188 unsigned long, vlen, loff_t, pos)
1189 {
1190 return __compat_sys_pwritev64(fd, vec, vlen, pos);
1191 }
1192 #endif
1193
1194 COMPAT_SYSCALL_DEFINE5(pwritev, compat_ulong_t, fd,
1195 const struct compat_iovec __user *,vec,
1196 compat_ulong_t, vlen, u32, pos_low, u32, pos_high)
1197 {
1198 loff_t pos = ((loff_t)pos_high << 32) | pos_low;
1199
1200 return __compat_sys_pwritev64(fd, vec, vlen, pos);
1201 }
1202 #endif
1203
1204 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
1205 size_t count, loff_t max)
1206 {
1207 struct fd in, out;
1208 struct inode *in_inode, *out_inode;
1209 loff_t pos;
1210 loff_t out_pos;
1211 ssize_t retval;
1212 int fl;
1213
1214 /*
1215 * Get input file, and verify that it is ok..
1216 */
1217 retval = -EBADF;
1218 in = fdget(in_fd);
1219 if (!in.file)
1220 goto out;
1221 if (!(in.file->f_mode & FMODE_READ))
1222 goto fput_in;
1223 retval = -ESPIPE;
1224 if (!ppos) {
1225 pos = in.file->f_pos;
1226 } else {
1227 pos = *ppos;
1228 if (!(in.file->f_mode & FMODE_PREAD))
1229 goto fput_in;
1230 }
1231 retval = rw_verify_area(READ, in.file, &pos, count);
1232 if (retval < 0)
1233 goto fput_in;
1234 count = retval;
1235
1236 /*
1237 * Get output file, and verify that it is ok..
1238 */
1239 retval = -EBADF;
1240 out = fdget(out_fd);
1241 if (!out.file)
1242 goto fput_in;
1243 if (!(out.file->f_mode & FMODE_WRITE))
1244 goto fput_out;
1245 retval = -EINVAL;
1246 in_inode = file_inode(in.file);
1247 out_inode = file_inode(out.file);
1248 out_pos = out.file->f_pos;
1249 retval = rw_verify_area(WRITE, out.file, &out_pos, count);
1250 if (retval < 0)
1251 goto fput_out;
1252 count = retval;
1253
1254 if (!max)
1255 max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
1256
1257 if (unlikely(pos + count > max)) {
1258 retval = -EOVERFLOW;
1259 if (pos >= max)
1260 goto fput_out;
1261 count = max - pos;
1262 }
1263
1264 fl = 0;
1265 #if 0
1266 /*
1267 * We need to debate whether we can enable this or not. The
1268 * man page documents EAGAIN return for the output at least,
1269 * and the application is arguably buggy if it doesn't expect
1270 * EAGAIN on a non-blocking file descriptor.
1271 */
1272 if (in.file->f_flags & O_NONBLOCK)
1273 fl = SPLICE_F_NONBLOCK;
1274 #endif
1275 file_start_write(out.file);
1276 retval = do_splice_direct(in.file, &pos, out.file, &out_pos, count, fl);
1277 file_end_write(out.file);
1278
1279 if (retval > 0) {
1280 add_rchar(current, retval);
1281 add_wchar(current, retval);
1282 fsnotify_access(in.file);
1283 fsnotify_modify(out.file);
1284 out.file->f_pos = out_pos;
1285 if (ppos)
1286 *ppos = pos;
1287 else
1288 in.file->f_pos = pos;
1289 }
1290
1291 inc_syscr(current);
1292 inc_syscw(current);
1293 if (pos > max)
1294 retval = -EOVERFLOW;
1295
1296 fput_out:
1297 fdput(out);
1298 fput_in:
1299 fdput(in);
1300 out:
1301 return retval;
1302 }
1303
1304 SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd, off_t __user *, offset, size_t, count)
1305 {
1306 loff_t pos;
1307 off_t off;
1308 ssize_t ret;
1309
1310 if (offset) {
1311 if (unlikely(get_user(off, offset)))
1312 return -EFAULT;
1313 pos = off;
1314 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1315 if (unlikely(put_user(pos, offset)))
1316 return -EFAULT;
1317 return ret;
1318 }
1319
1320 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1321 }
1322
1323 SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd, loff_t __user *, offset, size_t, count)
1324 {
1325 loff_t pos;
1326 ssize_t ret;
1327
1328 if (offset) {
1329 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1330 return -EFAULT;
1331 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1332 if (unlikely(put_user(pos, offset)))
1333 return -EFAULT;
1334 return ret;
1335 }
1336
1337 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1338 }
1339
1340 #ifdef CONFIG_COMPAT
1341 COMPAT_SYSCALL_DEFINE4(sendfile, int, out_fd, int, in_fd,
1342 compat_off_t __user *, offset, compat_size_t, count)
1343 {
1344 loff_t pos;
1345 off_t off;
1346 ssize_t ret;
1347
1348 if (offset) {
1349 if (unlikely(get_user(off, offset)))
1350 return -EFAULT;
1351 pos = off;
1352 ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
1353 if (unlikely(put_user(pos, offset)))
1354 return -EFAULT;
1355 return ret;
1356 }
1357
1358 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1359 }
1360
1361 COMPAT_SYSCALL_DEFINE4(sendfile64, int, out_fd, int, in_fd,
1362 compat_loff_t __user *, offset, compat_size_t, count)
1363 {
1364 loff_t pos;
1365 ssize_t ret;
1366
1367 if (offset) {
1368 if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
1369 return -EFAULT;
1370 ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
1371 if (unlikely(put_user(pos, offset)))
1372 return -EFAULT;
1373 return ret;
1374 }
1375
1376 return do_sendfile(out_fd, in_fd, NULL, count, 0);
1377 }
1378 #endif
Linux kernel - Deliverables
This page took 0.06084 seconds and 6 git commands to generate.