xref: /linux/fs/read_write.c (revision 20d0021394c1b070bf04b22c5bc8fdb437edd4c5)
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/smp_lock.h>
13 #include <linux/fsnotify.h>
14 #include <linux/security.h>
15 #include <linux/module.h>
16 #include <linux/syscalls.h>
17 
18 #include <asm/uaccess.h>
19 #include <asm/unistd.h>
20 
21 struct file_operations generic_ro_fops = {
22 	.llseek		= generic_file_llseek,
23 	.read		= generic_file_read,
24 	.mmap		= generic_file_readonly_mmap,
25 	.sendfile	= generic_file_sendfile,
26 };
27 
28 EXPORT_SYMBOL(generic_ro_fops);
29 
30 loff_t generic_file_llseek(struct file *file, loff_t offset, int origin)
31 {
32 	long long retval;
33 	struct inode *inode = file->f_mapping->host;
34 
35 	down(&inode->i_sem);
36 	switch (origin) {
37 		case 2:
38 			offset += inode->i_size;
39 			break;
40 		case 1:
41 			offset += file->f_pos;
42 	}
43 	retval = -EINVAL;
44 	if (offset>=0 && offset<=inode->i_sb->s_maxbytes) {
45 		if (offset != file->f_pos) {
46 			file->f_pos = offset;
47 			file->f_version = 0;
48 		}
49 		retval = offset;
50 	}
51 	up(&inode->i_sem);
52 	return retval;
53 }
54 
55 EXPORT_SYMBOL(generic_file_llseek);
56 
57 loff_t remote_llseek(struct file *file, loff_t offset, int origin)
58 {
59 	long long retval;
60 
61 	lock_kernel();
62 	switch (origin) {
63 		case 2:
64 			offset += i_size_read(file->f_dentry->d_inode);
65 			break;
66 		case 1:
67 			offset += file->f_pos;
68 	}
69 	retval = -EINVAL;
70 	if (offset>=0 && offset<=file->f_dentry->d_inode->i_sb->s_maxbytes) {
71 		if (offset != file->f_pos) {
72 			file->f_pos = offset;
73 			file->f_version = 0;
74 		}
75 		retval = offset;
76 	}
77 	unlock_kernel();
78 	return retval;
79 }
80 EXPORT_SYMBOL(remote_llseek);
81 
82 loff_t no_llseek(struct file *file, loff_t offset, int origin)
83 {
84 	return -ESPIPE;
85 }
86 EXPORT_SYMBOL(no_llseek);
87 
88 loff_t default_llseek(struct file *file, loff_t offset, int origin)
89 {
90 	long long retval;
91 
92 	lock_kernel();
93 	switch (origin) {
94 		case 2:
95 			offset += i_size_read(file->f_dentry->d_inode);
96 			break;
97 		case 1:
98 			offset += file->f_pos;
99 	}
100 	retval = -EINVAL;
101 	if (offset >= 0) {
102 		if (offset != file->f_pos) {
103 			file->f_pos = offset;
104 			file->f_version = 0;
105 		}
106 		retval = offset;
107 	}
108 	unlock_kernel();
109 	return retval;
110 }
111 EXPORT_SYMBOL(default_llseek);
112 
113 loff_t vfs_llseek(struct file *file, loff_t offset, int origin)
114 {
115 	loff_t (*fn)(struct file *, loff_t, int);
116 
117 	fn = no_llseek;
118 	if (file->f_mode & FMODE_LSEEK) {
119 		fn = default_llseek;
120 		if (file->f_op && file->f_op->llseek)
121 			fn = file->f_op->llseek;
122 	}
123 	return fn(file, offset, origin);
124 }
125 EXPORT_SYMBOL(vfs_llseek);
126 
127 asmlinkage off_t sys_lseek(unsigned int fd, off_t offset, unsigned int origin)
128 {
129 	off_t retval;
130 	struct file * file;
131 	int fput_needed;
132 
133 	retval = -EBADF;
134 	file = fget_light(fd, &fput_needed);
135 	if (!file)
136 		goto bad;
137 
138 	retval = -EINVAL;
139 	if (origin <= 2) {
140 		loff_t res = vfs_llseek(file, offset, origin);
141 		retval = res;
142 		if (res != (loff_t)retval)
143 			retval = -EOVERFLOW;	/* LFS: should only happen on 32 bit platforms */
144 	}
145 	fput_light(file, fput_needed);
146 bad:
147 	return retval;
148 }
149 
150 #ifdef __ARCH_WANT_SYS_LLSEEK
151 asmlinkage long sys_llseek(unsigned int fd, unsigned long offset_high,
152 			   unsigned long offset_low, loff_t __user * result,
153 			   unsigned int origin)
154 {
155 	int retval;
156 	struct file * file;
157 	loff_t offset;
158 	int fput_needed;
159 
160 	retval = -EBADF;
161 	file = fget_light(fd, &fput_needed);
162 	if (!file)
163 		goto bad;
164 
165 	retval = -EINVAL;
166 	if (origin > 2)
167 		goto out_putf;
168 
169 	offset = vfs_llseek(file, ((loff_t) offset_high << 32) | offset_low,
170 			origin);
171 
172 	retval = (int)offset;
173 	if (offset >= 0) {
174 		retval = -EFAULT;
175 		if (!copy_to_user(result, &offset, sizeof(offset)))
176 			retval = 0;
177 	}
178 out_putf:
179 	fput_light(file, fput_needed);
180 bad:
181 	return retval;
182 }
183 #endif
184 
185 
186 int rw_verify_area(int read_write, struct file *file, loff_t *ppos, size_t count)
187 {
188 	struct inode *inode;
189 	loff_t pos;
190 
191 	if (unlikely(count > file->f_maxcount))
192 		goto Einval;
193 	pos = *ppos;
194 	if (unlikely((pos < 0) || (loff_t) (pos + count) < 0))
195 		goto Einval;
196 
197 	inode = file->f_dentry->d_inode;
198 	if (inode->i_flock && MANDATORY_LOCK(inode))
199 		return locks_mandatory_area(read_write == READ ? FLOCK_VERIFY_READ : FLOCK_VERIFY_WRITE, inode, file, pos, count);
200 	return 0;
201 
202 Einval:
203 	return -EINVAL;
204 }
205 
206 static void wait_on_retry_sync_kiocb(struct kiocb *iocb)
207 {
208 	set_current_state(TASK_UNINTERRUPTIBLE);
209 	if (!kiocbIsKicked(iocb))
210 		schedule();
211 	else
212 		kiocbClearKicked(iocb);
213 	__set_current_state(TASK_RUNNING);
214 }
215 
216 ssize_t do_sync_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
217 {
218 	struct kiocb kiocb;
219 	ssize_t ret;
220 
221 	init_sync_kiocb(&kiocb, filp);
222 	kiocb.ki_pos = *ppos;
223 	while (-EIOCBRETRY ==
224 		(ret = filp->f_op->aio_read(&kiocb, buf, len, kiocb.ki_pos)))
225 		wait_on_retry_sync_kiocb(&kiocb);
226 
227 	if (-EIOCBQUEUED == ret)
228 		ret = wait_on_sync_kiocb(&kiocb);
229 	*ppos = kiocb.ki_pos;
230 	return ret;
231 }
232 
233 EXPORT_SYMBOL(do_sync_read);
234 
235 ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
236 {
237 	ssize_t ret;
238 
239 	if (!(file->f_mode & FMODE_READ))
240 		return -EBADF;
241 	if (!file->f_op || (!file->f_op->read && !file->f_op->aio_read))
242 		return -EINVAL;
243 	if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
244 		return -EFAULT;
245 
246 	ret = rw_verify_area(READ, file, pos, count);
247 	if (!ret) {
248 		ret = security_file_permission (file, MAY_READ);
249 		if (!ret) {
250 			if (file->f_op->read)
251 				ret = file->f_op->read(file, buf, count, pos);
252 			else
253 				ret = do_sync_read(file, buf, count, pos);
254 			if (ret > 0) {
255 				fsnotify_access(file->f_dentry);
256 				current->rchar += ret;
257 			}
258 			current->syscr++;
259 		}
260 	}
261 
262 	return ret;
263 }
264 
265 EXPORT_SYMBOL(vfs_read);
266 
267 ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
268 {
269 	struct kiocb kiocb;
270 	ssize_t ret;
271 
272 	init_sync_kiocb(&kiocb, filp);
273 	kiocb.ki_pos = *ppos;
274 	while (-EIOCBRETRY ==
275 	       (ret = filp->f_op->aio_write(&kiocb, buf, len, kiocb.ki_pos)))
276 		wait_on_retry_sync_kiocb(&kiocb);
277 
278 	if (-EIOCBQUEUED == ret)
279 		ret = wait_on_sync_kiocb(&kiocb);
280 	*ppos = kiocb.ki_pos;
281 	return ret;
282 }
283 
284 EXPORT_SYMBOL(do_sync_write);
285 
286 ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
287 {
288 	ssize_t ret;
289 
290 	if (!(file->f_mode & FMODE_WRITE))
291 		return -EBADF;
292 	if (!file->f_op || (!file->f_op->write && !file->f_op->aio_write))
293 		return -EINVAL;
294 	if (unlikely(!access_ok(VERIFY_READ, buf, count)))
295 		return -EFAULT;
296 
297 	ret = rw_verify_area(WRITE, file, pos, count);
298 	if (!ret) {
299 		ret = security_file_permission (file, MAY_WRITE);
300 		if (!ret) {
301 			if (file->f_op->write)
302 				ret = file->f_op->write(file, buf, count, pos);
303 			else
304 				ret = do_sync_write(file, buf, count, pos);
305 			if (ret > 0) {
306 				fsnotify_modify(file->f_dentry);
307 				current->wchar += ret;
308 			}
309 			current->syscw++;
310 		}
311 	}
312 
313 	return ret;
314 }
315 
316 EXPORT_SYMBOL(vfs_write);
317 
318 static inline loff_t file_pos_read(struct file *file)
319 {
320 	return file->f_pos;
321 }
322 
323 static inline void file_pos_write(struct file *file, loff_t pos)
324 {
325 	file->f_pos = pos;
326 }
327 
328 asmlinkage ssize_t sys_read(unsigned int fd, char __user * buf, size_t count)
329 {
330 	struct file *file;
331 	ssize_t ret = -EBADF;
332 	int fput_needed;
333 
334 	file = fget_light(fd, &fput_needed);
335 	if (file) {
336 		loff_t pos = file_pos_read(file);
337 		ret = vfs_read(file, buf, count, &pos);
338 		file_pos_write(file, pos);
339 		fput_light(file, fput_needed);
340 	}
341 
342 	return ret;
343 }
344 EXPORT_SYMBOL_GPL(sys_read);
345 
346 asmlinkage ssize_t sys_write(unsigned int fd, const char __user * buf, size_t count)
347 {
348 	struct file *file;
349 	ssize_t ret = -EBADF;
350 	int fput_needed;
351 
352 	file = fget_light(fd, &fput_needed);
353 	if (file) {
354 		loff_t pos = file_pos_read(file);
355 		ret = vfs_write(file, buf, count, &pos);
356 		file_pos_write(file, pos);
357 		fput_light(file, fput_needed);
358 	}
359 
360 	return ret;
361 }
362 
363 asmlinkage ssize_t sys_pread64(unsigned int fd, char __user *buf,
364 			     size_t count, loff_t pos)
365 {
366 	struct file *file;
367 	ssize_t ret = -EBADF;
368 	int fput_needed;
369 
370 	if (pos < 0)
371 		return -EINVAL;
372 
373 	file = fget_light(fd, &fput_needed);
374 	if (file) {
375 		ret = -ESPIPE;
376 		if (file->f_mode & FMODE_PREAD)
377 			ret = vfs_read(file, buf, count, &pos);
378 		fput_light(file, fput_needed);
379 	}
380 
381 	return ret;
382 }
383 
384 asmlinkage ssize_t sys_pwrite64(unsigned int fd, const char __user *buf,
385 			      size_t count, loff_t pos)
386 {
387 	struct file *file;
388 	ssize_t ret = -EBADF;
389 	int fput_needed;
390 
391 	if (pos < 0)
392 		return -EINVAL;
393 
394 	file = fget_light(fd, &fput_needed);
395 	if (file) {
396 		ret = -ESPIPE;
397 		if (file->f_mode & FMODE_PWRITE)
398 			ret = vfs_write(file, buf, count, &pos);
399 		fput_light(file, fput_needed);
400 	}
401 
402 	return ret;
403 }
404 
405 /*
406  * Reduce an iovec's length in-place.  Return the resulting number of segments
407  */
408 unsigned long iov_shorten(struct iovec *iov, unsigned long nr_segs, size_t to)
409 {
410 	unsigned long seg = 0;
411 	size_t len = 0;
412 
413 	while (seg < nr_segs) {
414 		seg++;
415 		if (len + iov->iov_len >= to) {
416 			iov->iov_len = to - len;
417 			break;
418 		}
419 		len += iov->iov_len;
420 		iov++;
421 	}
422 	return seg;
423 }
424 
425 EXPORT_SYMBOL(iov_shorten);
426 
427 /* A write operation does a read from user space and vice versa */
428 #define vrfy_dir(type) ((type) == READ ? VERIFY_WRITE : VERIFY_READ)
429 
430 static ssize_t do_readv_writev(int type, struct file *file,
431 			       const struct iovec __user * uvector,
432 			       unsigned long nr_segs, loff_t *pos)
433 {
434 	typedef ssize_t (*io_fn_t)(struct file *, char __user *, size_t, loff_t *);
435 	typedef ssize_t (*iov_fn_t)(struct file *, const struct iovec *, unsigned long, loff_t *);
436 
437 	size_t tot_len;
438 	struct iovec iovstack[UIO_FASTIOV];
439 	struct iovec *iov=iovstack, *vector;
440 	ssize_t ret;
441 	int seg;
442 	io_fn_t fn;
443 	iov_fn_t fnv;
444 
445 	/*
446 	 * SuS says "The readv() function *may* fail if the iovcnt argument
447 	 * was less than or equal to 0, or greater than {IOV_MAX}.  Linux has
448 	 * traditionally returned zero for zero segments, so...
449 	 */
450 	ret = 0;
451 	if (nr_segs == 0)
452 		goto out;
453 
454 	/*
455 	 * First get the "struct iovec" from user memory and
456 	 * verify all the pointers
457 	 */
458 	ret = -EINVAL;
459 	if ((nr_segs > UIO_MAXIOV) || (nr_segs <= 0))
460 		goto out;
461 	if (!file->f_op)
462 		goto out;
463 	if (nr_segs > UIO_FASTIOV) {
464 		ret = -ENOMEM;
465 		iov = kmalloc(nr_segs*sizeof(struct iovec), GFP_KERNEL);
466 		if (!iov)
467 			goto out;
468 	}
469 	ret = -EFAULT;
470 	if (copy_from_user(iov, uvector, nr_segs*sizeof(*uvector)))
471 		goto out;
472 
473 	/*
474 	 * Single unix specification:
475 	 * We should -EINVAL if an element length is not >= 0 and fitting an
476 	 * ssize_t.  The total length is fitting an ssize_t
477 	 *
478 	 * Be careful here because iov_len is a size_t not an ssize_t
479 	 */
480 	tot_len = 0;
481 	ret = -EINVAL;
482 	for (seg = 0; seg < nr_segs; seg++) {
483 		void __user *buf = iov[seg].iov_base;
484 		ssize_t len = (ssize_t)iov[seg].iov_len;
485 
486 		if (len < 0)	/* size_t not fitting an ssize_t .. */
487 			goto out;
488 		if (unlikely(!access_ok(vrfy_dir(type), buf, len)))
489 			goto Efault;
490 		tot_len += len;
491 		if ((ssize_t)tot_len < 0) /* maths overflow on the ssize_t */
492 			goto out;
493 	}
494 	if (tot_len == 0) {
495 		ret = 0;
496 		goto out;
497 	}
498 
499 	ret = rw_verify_area(type, file, pos, tot_len);
500 	if (ret)
501 		goto out;
502 
503 	fnv = NULL;
504 	if (type == READ) {
505 		fn = file->f_op->read;
506 		fnv = file->f_op->readv;
507 	} else {
508 		fn = (io_fn_t)file->f_op->write;
509 		fnv = file->f_op->writev;
510 	}
511 	if (fnv) {
512 		ret = fnv(file, iov, nr_segs, pos);
513 		goto out;
514 	}
515 
516 	/* Do it by hand, with file-ops */
517 	ret = 0;
518 	vector = iov;
519 	while (nr_segs > 0) {
520 		void __user * base;
521 		size_t len;
522 		ssize_t nr;
523 
524 		base = vector->iov_base;
525 		len = vector->iov_len;
526 		vector++;
527 		nr_segs--;
528 
529 		nr = fn(file, base, len, pos);
530 
531 		if (nr < 0) {
532 			if (!ret) ret = nr;
533 			break;
534 		}
535 		ret += nr;
536 		if (nr != len)
537 			break;
538 	}
539 out:
540 	if (iov != iovstack)
541 		kfree(iov);
542 	if ((ret + (type == READ)) > 0) {
543 		if (type == READ)
544 			fsnotify_access(file->f_dentry);
545 		else
546 			fsnotify_modify(file->f_dentry);
547 	}
548 	return ret;
549 Efault:
550 	ret = -EFAULT;
551 	goto out;
552 }
553 
554 ssize_t vfs_readv(struct file *file, const struct iovec __user *vec,
555 		  unsigned long vlen, loff_t *pos)
556 {
557 	if (!(file->f_mode & FMODE_READ))
558 		return -EBADF;
559 	if (!file->f_op || (!file->f_op->readv && !file->f_op->read))
560 		return -EINVAL;
561 
562 	return do_readv_writev(READ, file, vec, vlen, pos);
563 }
564 
565 EXPORT_SYMBOL(vfs_readv);
566 
567 ssize_t vfs_writev(struct file *file, const struct iovec __user *vec,
568 		   unsigned long vlen, loff_t *pos)
569 {
570 	if (!(file->f_mode & FMODE_WRITE))
571 		return -EBADF;
572 	if (!file->f_op || (!file->f_op->writev && !file->f_op->write))
573 		return -EINVAL;
574 
575 	return do_readv_writev(WRITE, file, vec, vlen, pos);
576 }
577 
578 EXPORT_SYMBOL(vfs_writev);
579 
580 asmlinkage ssize_t
581 sys_readv(unsigned long fd, const struct iovec __user *vec, unsigned long vlen)
582 {
583 	struct file *file;
584 	ssize_t ret = -EBADF;
585 	int fput_needed;
586 
587 	file = fget_light(fd, &fput_needed);
588 	if (file) {
589 		loff_t pos = file_pos_read(file);
590 		ret = vfs_readv(file, vec, vlen, &pos);
591 		file_pos_write(file, pos);
592 		fput_light(file, fput_needed);
593 	}
594 
595 	if (ret > 0)
596 		current->rchar += ret;
597 	current->syscr++;
598 	return ret;
599 }
600 
601 asmlinkage ssize_t
602 sys_writev(unsigned long fd, const struct iovec __user *vec, unsigned long vlen)
603 {
604 	struct file *file;
605 	ssize_t ret = -EBADF;
606 	int fput_needed;
607 
608 	file = fget_light(fd, &fput_needed);
609 	if (file) {
610 		loff_t pos = file_pos_read(file);
611 		ret = vfs_writev(file, vec, vlen, &pos);
612 		file_pos_write(file, pos);
613 		fput_light(file, fput_needed);
614 	}
615 
616 	if (ret > 0)
617 		current->wchar += ret;
618 	current->syscw++;
619 	return ret;
620 }
621 
622 static ssize_t do_sendfile(int out_fd, int in_fd, loff_t *ppos,
623 			   size_t count, loff_t max)
624 {
625 	struct file * in_file, * out_file;
626 	struct inode * in_inode, * out_inode;
627 	loff_t pos;
628 	ssize_t retval;
629 	int fput_needed_in, fput_needed_out;
630 
631 	/*
632 	 * Get input file, and verify that it is ok..
633 	 */
634 	retval = -EBADF;
635 	in_file = fget_light(in_fd, &fput_needed_in);
636 	if (!in_file)
637 		goto out;
638 	if (!(in_file->f_mode & FMODE_READ))
639 		goto fput_in;
640 	retval = -EINVAL;
641 	in_inode = in_file->f_dentry->d_inode;
642 	if (!in_inode)
643 		goto fput_in;
644 	if (!in_file->f_op || !in_file->f_op->sendfile)
645 		goto fput_in;
646 	retval = -ESPIPE;
647 	if (!ppos)
648 		ppos = &in_file->f_pos;
649 	else
650 		if (!(in_file->f_mode & FMODE_PREAD))
651 			goto fput_in;
652 	retval = rw_verify_area(READ, in_file, ppos, count);
653 	if (retval)
654 		goto fput_in;
655 
656 	retval = security_file_permission (in_file, MAY_READ);
657 	if (retval)
658 		goto fput_in;
659 
660 	/*
661 	 * Get output file, and verify that it is ok..
662 	 */
663 	retval = -EBADF;
664 	out_file = fget_light(out_fd, &fput_needed_out);
665 	if (!out_file)
666 		goto fput_in;
667 	if (!(out_file->f_mode & FMODE_WRITE))
668 		goto fput_out;
669 	retval = -EINVAL;
670 	if (!out_file->f_op || !out_file->f_op->sendpage)
671 		goto fput_out;
672 	out_inode = out_file->f_dentry->d_inode;
673 	retval = rw_verify_area(WRITE, out_file, &out_file->f_pos, count);
674 	if (retval)
675 		goto fput_out;
676 
677 	retval = security_file_permission (out_file, MAY_WRITE);
678 	if (retval)
679 		goto fput_out;
680 
681 	if (!max)
682 		max = min(in_inode->i_sb->s_maxbytes, out_inode->i_sb->s_maxbytes);
683 
684 	pos = *ppos;
685 	retval = -EINVAL;
686 	if (unlikely(pos < 0))
687 		goto fput_out;
688 	if (unlikely(pos + count > max)) {
689 		retval = -EOVERFLOW;
690 		if (pos >= max)
691 			goto fput_out;
692 		count = max - pos;
693 	}
694 
695 	retval = in_file->f_op->sendfile(in_file, ppos, count, file_send_actor, out_file);
696 
697 	if (retval > 0) {
698 		current->rchar += retval;
699 		current->wchar += retval;
700 	}
701 	current->syscr++;
702 	current->syscw++;
703 
704 	if (*ppos > max)
705 		retval = -EOVERFLOW;
706 
707 fput_out:
708 	fput_light(out_file, fput_needed_out);
709 fput_in:
710 	fput_light(in_file, fput_needed_in);
711 out:
712 	return retval;
713 }
714 
715 asmlinkage ssize_t sys_sendfile(int out_fd, int in_fd, off_t __user *offset, size_t count)
716 {
717 	loff_t pos;
718 	off_t off;
719 	ssize_t ret;
720 
721 	if (offset) {
722 		if (unlikely(get_user(off, offset)))
723 			return -EFAULT;
724 		pos = off;
725 		ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
726 		if (unlikely(put_user(pos, offset)))
727 			return -EFAULT;
728 		return ret;
729 	}
730 
731 	return do_sendfile(out_fd, in_fd, NULL, count, 0);
732 }
733 
734 asmlinkage ssize_t sys_sendfile64(int out_fd, int in_fd, loff_t __user *offset, size_t count)
735 {
736 	loff_t pos;
737 	ssize_t ret;
738 
739 	if (offset) {
740 		if (unlikely(copy_from_user(&pos, offset, sizeof(loff_t))))
741 			return -EFAULT;
742 		ret = do_sendfile(out_fd, in_fd, &pos, count, 0);
743 		if (unlikely(put_user(pos, offset)))
744 			return -EFAULT;
745 		return ret;
746 	}
747 
748 	return do_sendfile(out_fd, in_fd, NULL, count, 0);
749 }
750