xref: /openbmc/linux/block/ioctl.c (revision f9834f18)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/capability.h>
3 #include <linux/compat.h>
4 #include <linux/blkdev.h>
5 #include <linux/export.h>
6 #include <linux/gfp.h>
7 #include <linux/blkpg.h>
8 #include <linux/hdreg.h>
9 #include <linux/backing-dev.h>
10 #include <linux/fs.h>
11 #include <linux/blktrace_api.h>
12 #include <linux/pr.h>
13 #include <linux/uaccess.h>
14 
15 static int blkpg_do_ioctl(struct block_device *bdev,
16 			  struct blkpg_partition __user *upart, int op)
17 {
18 	struct block_device *bdevp;
19 	struct gendisk *disk;
20 	struct hd_struct *part, *lpart;
21 	struct blkpg_partition p;
22 	struct disk_part_iter piter;
23 	long long start, length;
24 	int partno;
25 
26 	if (!capable(CAP_SYS_ADMIN))
27 		return -EACCES;
28 	if (copy_from_user(&p, upart, sizeof(struct blkpg_partition)))
29 		return -EFAULT;
30 	disk = bdev->bd_disk;
31 	if (bdev != bdev->bd_contains)
32 		return -EINVAL;
33 	partno = p.pno;
34 	if (partno <= 0)
35 		return -EINVAL;
36 	switch (op) {
37 		case BLKPG_ADD_PARTITION:
38 			start = p.start >> 9;
39 			length = p.length >> 9;
40 			/* check for fit in a hd_struct */
41 			if (sizeof(sector_t) == sizeof(long) &&
42 			    sizeof(long long) > sizeof(long)) {
43 				long pstart = start, plength = length;
44 				if (pstart != start || plength != length
45 				    || pstart < 0 || plength < 0 || partno > 65535)
46 					return -EINVAL;
47 			}
48 			/* check if partition is aligned to blocksize */
49 			if (p.start & (bdev_logical_block_size(bdev) - 1))
50 				return -EINVAL;
51 
52 			mutex_lock(&bdev->bd_mutex);
53 
54 			/* overlap? */
55 			disk_part_iter_init(&piter, disk,
56 					    DISK_PITER_INCL_EMPTY);
57 			while ((part = disk_part_iter_next(&piter))) {
58 				if (!(start + length <= part->start_sect ||
59 				      start >= part->start_sect + part->nr_sects)) {
60 					disk_part_iter_exit(&piter);
61 					mutex_unlock(&bdev->bd_mutex);
62 					return -EBUSY;
63 				}
64 			}
65 			disk_part_iter_exit(&piter);
66 
67 			/* all seems OK */
68 			part = add_partition(disk, partno, start, length,
69 					     ADDPART_FLAG_NONE, NULL);
70 			mutex_unlock(&bdev->bd_mutex);
71 			return PTR_ERR_OR_ZERO(part);
72 		case BLKPG_DEL_PARTITION:
73 			part = disk_get_part(disk, partno);
74 			if (!part)
75 				return -ENXIO;
76 
77 			bdevp = bdget(part_devt(part));
78 			disk_put_part(part);
79 			if (!bdevp)
80 				return -ENOMEM;
81 
82 			mutex_lock(&bdevp->bd_mutex);
83 			if (bdevp->bd_openers) {
84 				mutex_unlock(&bdevp->bd_mutex);
85 				bdput(bdevp);
86 				return -EBUSY;
87 			}
88 			/* all seems OK */
89 			fsync_bdev(bdevp);
90 			invalidate_bdev(bdevp);
91 
92 			mutex_lock_nested(&bdev->bd_mutex, 1);
93 			delete_partition(disk, partno);
94 			mutex_unlock(&bdev->bd_mutex);
95 			mutex_unlock(&bdevp->bd_mutex);
96 			bdput(bdevp);
97 
98 			return 0;
99 		case BLKPG_RESIZE_PARTITION:
100 			start = p.start >> 9;
101 			/* new length of partition in bytes */
102 			length = p.length >> 9;
103 			/* check for fit in a hd_struct */
104 			if (sizeof(sector_t) == sizeof(long) &&
105 			    sizeof(long long) > sizeof(long)) {
106 				long pstart = start, plength = length;
107 				if (pstart != start || plength != length
108 				    || pstart < 0 || plength < 0)
109 					return -EINVAL;
110 			}
111 			part = disk_get_part(disk, partno);
112 			if (!part)
113 				return -ENXIO;
114 			bdevp = bdget(part_devt(part));
115 			if (!bdevp) {
116 				disk_put_part(part);
117 				return -ENOMEM;
118 			}
119 			mutex_lock(&bdevp->bd_mutex);
120 			mutex_lock_nested(&bdev->bd_mutex, 1);
121 			if (start != part->start_sect) {
122 				mutex_unlock(&bdevp->bd_mutex);
123 				mutex_unlock(&bdev->bd_mutex);
124 				bdput(bdevp);
125 				disk_put_part(part);
126 				return -EINVAL;
127 			}
128 			/* overlap? */
129 			disk_part_iter_init(&piter, disk,
130 					    DISK_PITER_INCL_EMPTY);
131 			while ((lpart = disk_part_iter_next(&piter))) {
132 				if (lpart->partno != partno &&
133 				   !(start + length <= lpart->start_sect ||
134 				   start >= lpart->start_sect + lpart->nr_sects)
135 				   ) {
136 					disk_part_iter_exit(&piter);
137 					mutex_unlock(&bdevp->bd_mutex);
138 					mutex_unlock(&bdev->bd_mutex);
139 					bdput(bdevp);
140 					disk_put_part(part);
141 					return -EBUSY;
142 				}
143 			}
144 			disk_part_iter_exit(&piter);
145 			part_nr_sects_write(part, (sector_t)length);
146 			i_size_write(bdevp->bd_inode, p.length);
147 			mutex_unlock(&bdevp->bd_mutex);
148 			mutex_unlock(&bdev->bd_mutex);
149 			bdput(bdevp);
150 			disk_put_part(part);
151 			return 0;
152 		default:
153 			return -EINVAL;
154 	}
155 }
156 
157 static int blkpg_ioctl(struct block_device *bdev,
158 		       struct blkpg_ioctl_arg __user *arg)
159 {
160 	struct blkpg_partition __user *udata;
161 	int op;
162 
163 	if (get_user(op, &arg->op) || get_user(udata, &arg->data))
164 		return -EFAULT;
165 
166 	return blkpg_do_ioctl(bdev, udata, op);
167 }
168 
169 #ifdef CONFIG_COMPAT
170 struct compat_blkpg_ioctl_arg {
171 	compat_int_t op;
172 	compat_int_t flags;
173 	compat_int_t datalen;
174 	compat_caddr_t data;
175 };
176 
177 static int compat_blkpg_ioctl(struct block_device *bdev,
178 			      struct compat_blkpg_ioctl_arg __user *arg)
179 {
180 	compat_caddr_t udata;
181 	int op;
182 
183 	if (get_user(op, &arg->op) || get_user(udata, &arg->data))
184 		return -EFAULT;
185 
186 	return blkpg_do_ioctl(bdev, compat_ptr(udata), op);
187 }
188 #endif
189 
190 static int blkdev_reread_part(struct block_device *bdev)
191 {
192 	int ret;
193 
194 	if (!disk_part_scan_enabled(bdev->bd_disk) || bdev != bdev->bd_contains)
195 		return -EINVAL;
196 	if (!capable(CAP_SYS_ADMIN))
197 		return -EACCES;
198 
199 	mutex_lock(&bdev->bd_mutex);
200 	ret = bdev_disk_changed(bdev, false);
201 	mutex_unlock(&bdev->bd_mutex);
202 
203 	return ret;
204 }
205 
206 static int blk_ioctl_discard(struct block_device *bdev, fmode_t mode,
207 		unsigned long arg, unsigned long flags)
208 {
209 	uint64_t range[2];
210 	uint64_t start, len;
211 	struct request_queue *q = bdev_get_queue(bdev);
212 	struct address_space *mapping = bdev->bd_inode->i_mapping;
213 
214 
215 	if (!(mode & FMODE_WRITE))
216 		return -EBADF;
217 
218 	if (!blk_queue_discard(q))
219 		return -EOPNOTSUPP;
220 
221 	if (copy_from_user(range, (void __user *)arg, sizeof(range)))
222 		return -EFAULT;
223 
224 	start = range[0];
225 	len = range[1];
226 
227 	if (start & 511)
228 		return -EINVAL;
229 	if (len & 511)
230 		return -EINVAL;
231 
232 	if (start + len > i_size_read(bdev->bd_inode))
233 		return -EINVAL;
234 	truncate_inode_pages_range(mapping, start, start + len - 1);
235 	return blkdev_issue_discard(bdev, start >> 9, len >> 9,
236 				    GFP_KERNEL, flags);
237 }
238 
239 static int blk_ioctl_zeroout(struct block_device *bdev, fmode_t mode,
240 		unsigned long arg)
241 {
242 	uint64_t range[2];
243 	struct address_space *mapping;
244 	uint64_t start, end, len;
245 
246 	if (!(mode & FMODE_WRITE))
247 		return -EBADF;
248 
249 	if (copy_from_user(range, (void __user *)arg, sizeof(range)))
250 		return -EFAULT;
251 
252 	start = range[0];
253 	len = range[1];
254 	end = start + len - 1;
255 
256 	if (start & 511)
257 		return -EINVAL;
258 	if (len & 511)
259 		return -EINVAL;
260 	if (end >= (uint64_t)i_size_read(bdev->bd_inode))
261 		return -EINVAL;
262 	if (end < start)
263 		return -EINVAL;
264 
265 	/* Invalidate the page cache, including dirty pages */
266 	mapping = bdev->bd_inode->i_mapping;
267 	truncate_inode_pages_range(mapping, start, end);
268 
269 	return blkdev_issue_zeroout(bdev, start >> 9, len >> 9, GFP_KERNEL,
270 			BLKDEV_ZERO_NOUNMAP);
271 }
272 
273 static int put_ushort(unsigned short __user *argp, unsigned short val)
274 {
275 	return put_user(val, argp);
276 }
277 
278 static int put_int(int __user *argp, int val)
279 {
280 	return put_user(val, argp);
281 }
282 
283 static int put_uint(unsigned int __user *argp, unsigned int val)
284 {
285 	return put_user(val, argp);
286 }
287 
288 static int put_long(long __user *argp, long val)
289 {
290 	return put_user(val, argp);
291 }
292 
293 static int put_ulong(unsigned long __user *argp, unsigned long val)
294 {
295 	return put_user(val, argp);
296 }
297 
298 static int put_u64(u64 __user *argp, u64 val)
299 {
300 	return put_user(val, argp);
301 }
302 
303 #ifdef CONFIG_COMPAT
304 static int compat_put_long(compat_long_t *argp, long val)
305 {
306 	return put_user(val, argp);
307 }
308 
309 static int compat_put_ulong(compat_ulong_t *argp, compat_ulong_t val)
310 {
311 	return put_user(val, argp);
312 }
313 #endif
314 
315 int __blkdev_driver_ioctl(struct block_device *bdev, fmode_t mode,
316 			unsigned cmd, unsigned long arg)
317 {
318 	struct gendisk *disk = bdev->bd_disk;
319 
320 	if (disk->fops->ioctl)
321 		return disk->fops->ioctl(bdev, mode, cmd, arg);
322 
323 	return -ENOTTY;
324 }
325 /*
326  * For the record: _GPL here is only because somebody decided to slap it
327  * on the previous export.  Sheer idiocy, since it wasn't copyrightable
328  * at all and could be open-coded without any exports by anybody who cares.
329  */
330 EXPORT_SYMBOL_GPL(__blkdev_driver_ioctl);
331 
332 #ifdef CONFIG_COMPAT
333 /*
334  * This is the equivalent of compat_ptr_ioctl(), to be used by block
335  * drivers that implement only commands that are completely compatible
336  * between 32-bit and 64-bit user space
337  */
338 int blkdev_compat_ptr_ioctl(struct block_device *bdev, fmode_t mode,
339 			unsigned cmd, unsigned long arg)
340 {
341 	struct gendisk *disk = bdev->bd_disk;
342 
343 	if (disk->fops->ioctl)
344 		return disk->fops->ioctl(bdev, mode, cmd,
345 					 (unsigned long)compat_ptr(arg));
346 
347 	return -ENOIOCTLCMD;
348 }
349 EXPORT_SYMBOL(blkdev_compat_ptr_ioctl);
350 #endif
351 
352 static int blkdev_pr_register(struct block_device *bdev,
353 		struct pr_registration __user *arg)
354 {
355 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
356 	struct pr_registration reg;
357 
358 	if (!capable(CAP_SYS_ADMIN))
359 		return -EPERM;
360 	if (!ops || !ops->pr_register)
361 		return -EOPNOTSUPP;
362 	if (copy_from_user(&reg, arg, sizeof(reg)))
363 		return -EFAULT;
364 
365 	if (reg.flags & ~PR_FL_IGNORE_KEY)
366 		return -EOPNOTSUPP;
367 	return ops->pr_register(bdev, reg.old_key, reg.new_key, reg.flags);
368 }
369 
370 static int blkdev_pr_reserve(struct block_device *bdev,
371 		struct pr_reservation __user *arg)
372 {
373 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
374 	struct pr_reservation rsv;
375 
376 	if (!capable(CAP_SYS_ADMIN))
377 		return -EPERM;
378 	if (!ops || !ops->pr_reserve)
379 		return -EOPNOTSUPP;
380 	if (copy_from_user(&rsv, arg, sizeof(rsv)))
381 		return -EFAULT;
382 
383 	if (rsv.flags & ~PR_FL_IGNORE_KEY)
384 		return -EOPNOTSUPP;
385 	return ops->pr_reserve(bdev, rsv.key, rsv.type, rsv.flags);
386 }
387 
388 static int blkdev_pr_release(struct block_device *bdev,
389 		struct pr_reservation __user *arg)
390 {
391 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
392 	struct pr_reservation rsv;
393 
394 	if (!capable(CAP_SYS_ADMIN))
395 		return -EPERM;
396 	if (!ops || !ops->pr_release)
397 		return -EOPNOTSUPP;
398 	if (copy_from_user(&rsv, arg, sizeof(rsv)))
399 		return -EFAULT;
400 
401 	if (rsv.flags)
402 		return -EOPNOTSUPP;
403 	return ops->pr_release(bdev, rsv.key, rsv.type);
404 }
405 
406 static int blkdev_pr_preempt(struct block_device *bdev,
407 		struct pr_preempt __user *arg, bool abort)
408 {
409 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
410 	struct pr_preempt p;
411 
412 	if (!capable(CAP_SYS_ADMIN))
413 		return -EPERM;
414 	if (!ops || !ops->pr_preempt)
415 		return -EOPNOTSUPP;
416 	if (copy_from_user(&p, arg, sizeof(p)))
417 		return -EFAULT;
418 
419 	if (p.flags)
420 		return -EOPNOTSUPP;
421 	return ops->pr_preempt(bdev, p.old_key, p.new_key, p.type, abort);
422 }
423 
424 static int blkdev_pr_clear(struct block_device *bdev,
425 		struct pr_clear __user *arg)
426 {
427 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
428 	struct pr_clear c;
429 
430 	if (!capable(CAP_SYS_ADMIN))
431 		return -EPERM;
432 	if (!ops || !ops->pr_clear)
433 		return -EOPNOTSUPP;
434 	if (copy_from_user(&c, arg, sizeof(c)))
435 		return -EFAULT;
436 
437 	if (c.flags)
438 		return -EOPNOTSUPP;
439 	return ops->pr_clear(bdev, c.key);
440 }
441 
442 /*
443  * Is it an unrecognized ioctl? The correct returns are either
444  * ENOTTY (final) or ENOIOCTLCMD ("I don't know this one, try a
445  * fallback"). ENOIOCTLCMD gets turned into ENOTTY by the ioctl
446  * code before returning.
447  *
448  * Confused drivers sometimes return EINVAL, which is wrong. It
449  * means "I understood the ioctl command, but the parameters to
450  * it were wrong".
451  *
452  * We should aim to just fix the broken drivers, the EINVAL case
453  * should go away.
454  */
455 static inline int is_unrecognized_ioctl(int ret)
456 {
457 	return	ret == -EINVAL ||
458 		ret == -ENOTTY ||
459 		ret == -ENOIOCTLCMD;
460 }
461 
462 static int blkdev_flushbuf(struct block_device *bdev, fmode_t mode,
463 		unsigned cmd, unsigned long arg)
464 {
465 	int ret;
466 
467 	if (!capable(CAP_SYS_ADMIN))
468 		return -EACCES;
469 
470 	ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
471 	if (!is_unrecognized_ioctl(ret))
472 		return ret;
473 
474 	fsync_bdev(bdev);
475 	invalidate_bdev(bdev);
476 	return 0;
477 }
478 
479 static int blkdev_roset(struct block_device *bdev, fmode_t mode,
480 		unsigned cmd, unsigned long arg)
481 {
482 	int ret, n;
483 
484 	if (!capable(CAP_SYS_ADMIN))
485 		return -EACCES;
486 
487 	ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
488 	if (!is_unrecognized_ioctl(ret))
489 		return ret;
490 	if (get_user(n, (int __user *)arg))
491 		return -EFAULT;
492 	set_device_ro(bdev, n);
493 	return 0;
494 }
495 
496 static int blkdev_getgeo(struct block_device *bdev,
497 		struct hd_geometry __user *argp)
498 {
499 	struct gendisk *disk = bdev->bd_disk;
500 	struct hd_geometry geo;
501 	int ret;
502 
503 	if (!argp)
504 		return -EINVAL;
505 	if (!disk->fops->getgeo)
506 		return -ENOTTY;
507 
508 	/*
509 	 * We need to set the startsect first, the driver may
510 	 * want to override it.
511 	 */
512 	memset(&geo, 0, sizeof(geo));
513 	geo.start = get_start_sect(bdev);
514 	ret = disk->fops->getgeo(bdev, &geo);
515 	if (ret)
516 		return ret;
517 	if (copy_to_user(argp, &geo, sizeof(geo)))
518 		return -EFAULT;
519 	return 0;
520 }
521 
522 #ifdef CONFIG_COMPAT
523 struct compat_hd_geometry {
524 	unsigned char heads;
525 	unsigned char sectors;
526 	unsigned short cylinders;
527 	u32 start;
528 };
529 
530 static int compat_hdio_getgeo(struct block_device *bdev,
531 			      struct compat_hd_geometry __user *ugeo)
532 {
533 	struct gendisk *disk = bdev->bd_disk;
534 	struct hd_geometry geo;
535 	int ret;
536 
537 	if (!ugeo)
538 		return -EINVAL;
539 	if (!disk->fops->getgeo)
540 		return -ENOTTY;
541 
542 	memset(&geo, 0, sizeof(geo));
543 	/*
544 	 * We need to set the startsect first, the driver may
545 	 * want to override it.
546 	 */
547 	geo.start = get_start_sect(bdev);
548 	ret = disk->fops->getgeo(bdev, &geo);
549 	if (ret)
550 		return ret;
551 
552 	ret = copy_to_user(ugeo, &geo, 4);
553 	ret |= put_user(geo.start, &ugeo->start);
554 	if (ret)
555 		ret = -EFAULT;
556 
557 	return ret;
558 }
559 #endif
560 
561 /* set the logical block size */
562 static int blkdev_bszset(struct block_device *bdev, fmode_t mode,
563 		int __user *argp)
564 {
565 	int ret, n;
566 
567 	if (!capable(CAP_SYS_ADMIN))
568 		return -EACCES;
569 	if (!argp)
570 		return -EINVAL;
571 	if (get_user(n, argp))
572 		return -EFAULT;
573 
574 	if (!(mode & FMODE_EXCL)) {
575 		bdgrab(bdev);
576 		if (blkdev_get(bdev, mode | FMODE_EXCL, &bdev) < 0)
577 			return -EBUSY;
578 	}
579 
580 	ret = set_blocksize(bdev, n);
581 	if (!(mode & FMODE_EXCL))
582 		blkdev_put(bdev, mode | FMODE_EXCL);
583 	return ret;
584 }
585 
586 /*
587  * Common commands that are handled the same way on native and compat
588  * user space. Note the separate arg/argp parameters that are needed
589  * to deal with the compat_ptr() conversion.
590  */
591 static int blkdev_common_ioctl(struct block_device *bdev, fmode_t mode,
592 				unsigned cmd, unsigned long arg, void __user *argp)
593 {
594 	unsigned int max_sectors;
595 
596 	switch (cmd) {
597 	case BLKFLSBUF:
598 		return blkdev_flushbuf(bdev, mode, cmd, arg);
599 	case BLKROSET:
600 		return blkdev_roset(bdev, mode, cmd, arg);
601 	case BLKDISCARD:
602 		return blk_ioctl_discard(bdev, mode, arg, 0);
603 	case BLKSECDISCARD:
604 		return blk_ioctl_discard(bdev, mode, arg,
605 				BLKDEV_DISCARD_SECURE);
606 	case BLKZEROOUT:
607 		return blk_ioctl_zeroout(bdev, mode, arg);
608 	case BLKREPORTZONE:
609 		return blkdev_report_zones_ioctl(bdev, mode, cmd, arg);
610 	case BLKRESETZONE:
611 	case BLKOPENZONE:
612 	case BLKCLOSEZONE:
613 	case BLKFINISHZONE:
614 		return blkdev_zone_mgmt_ioctl(bdev, mode, cmd, arg);
615 	case BLKGETZONESZ:
616 		return put_uint(argp, bdev_zone_sectors(bdev));
617 	case BLKGETNRZONES:
618 		return put_uint(argp, blkdev_nr_zones(bdev->bd_disk));
619 	case BLKROGET:
620 		return put_int(argp, bdev_read_only(bdev) != 0);
621 	case BLKSSZGET: /* get block device logical block size */
622 		return put_int(argp, bdev_logical_block_size(bdev));
623 	case BLKPBSZGET: /* get block device physical block size */
624 		return put_uint(argp, bdev_physical_block_size(bdev));
625 	case BLKIOMIN:
626 		return put_uint(argp, bdev_io_min(bdev));
627 	case BLKIOOPT:
628 		return put_uint(argp, bdev_io_opt(bdev));
629 	case BLKALIGNOFF:
630 		return put_int(argp, bdev_alignment_offset(bdev));
631 	case BLKDISCARDZEROES:
632 		return put_uint(argp, 0);
633 	case BLKSECTGET:
634 		max_sectors = min_t(unsigned int, USHRT_MAX,
635 				    queue_max_sectors(bdev_get_queue(bdev)));
636 		return put_ushort(argp, max_sectors);
637 	case BLKROTATIONAL:
638 		return put_ushort(argp, !blk_queue_nonrot(bdev_get_queue(bdev)));
639 	case BLKRASET:
640 	case BLKFRASET:
641 		if(!capable(CAP_SYS_ADMIN))
642 			return -EACCES;
643 		bdev->bd_bdi->ra_pages = (arg * 512) / PAGE_SIZE;
644 		return 0;
645 	case BLKRRPART:
646 		return blkdev_reread_part(bdev);
647 	case BLKTRACESTART:
648 	case BLKTRACESTOP:
649 	case BLKTRACETEARDOWN:
650 		return blk_trace_ioctl(bdev, cmd, argp);
651 	case IOC_PR_REGISTER:
652 		return blkdev_pr_register(bdev, argp);
653 	case IOC_PR_RESERVE:
654 		return blkdev_pr_reserve(bdev, argp);
655 	case IOC_PR_RELEASE:
656 		return blkdev_pr_release(bdev, argp);
657 	case IOC_PR_PREEMPT:
658 		return blkdev_pr_preempt(bdev, argp, false);
659 	case IOC_PR_PREEMPT_ABORT:
660 		return blkdev_pr_preempt(bdev, argp, true);
661 	case IOC_PR_CLEAR:
662 		return blkdev_pr_clear(bdev, argp);
663 	default:
664 		return -ENOIOCTLCMD;
665 	}
666 }
667 
668 /*
669  * Always keep this in sync with compat_blkdev_ioctl()
670  * to handle all incompatible commands in both functions.
671  *
672  * New commands must be compatible and go into blkdev_common_ioctl
673  */
674 int blkdev_ioctl(struct block_device *bdev, fmode_t mode, unsigned cmd,
675 			unsigned long arg)
676 {
677 	int ret;
678 	loff_t size;
679 	void __user *argp = (void __user *)arg;
680 
681 	switch (cmd) {
682 	/* These need separate implementations for the data structure */
683 	case HDIO_GETGEO:
684 		return blkdev_getgeo(bdev, argp);
685 	case BLKPG:
686 		return blkpg_ioctl(bdev, argp);
687 
688 	/* Compat mode returns 32-bit data instead of 'long' */
689 	case BLKRAGET:
690 	case BLKFRAGET:
691 		if (!argp)
692 			return -EINVAL;
693 		return put_long(argp, (bdev->bd_bdi->ra_pages*PAGE_SIZE) / 512);
694 	case BLKGETSIZE:
695 		size = i_size_read(bdev->bd_inode);
696 		if ((size >> 9) > ~0UL)
697 			return -EFBIG;
698 		return put_ulong(argp, size >> 9);
699 
700 	/* The data is compatible, but the command number is different */
701 	case BLKBSZGET: /* get block device soft block size (cf. BLKSSZGET) */
702 		return put_int(argp, block_size(bdev));
703 	case BLKBSZSET:
704 		return blkdev_bszset(bdev, mode, argp);
705 	case BLKGETSIZE64:
706 		return put_u64(argp, i_size_read(bdev->bd_inode));
707 
708 	/* Incompatible alignment on i386 */
709 	case BLKTRACESETUP:
710 		return blk_trace_ioctl(bdev, cmd, argp);
711 	default:
712 		break;
713 	}
714 
715 	ret = blkdev_common_ioctl(bdev, mode, cmd, arg, argp);
716 	if (ret == -ENOIOCTLCMD)
717 		return __blkdev_driver_ioctl(bdev, mode, cmd, arg);
718 
719 	return ret;
720 }
721 EXPORT_SYMBOL_GPL(blkdev_ioctl); /* for /dev/raw */
722 
723 #ifdef CONFIG_COMPAT
724 
725 #define BLKBSZGET_32		_IOR(0x12, 112, int)
726 #define BLKBSZSET_32		_IOW(0x12, 113, int)
727 #define BLKGETSIZE64_32		_IOR(0x12, 114, int)
728 
729 /* Most of the generic ioctls are handled in the normal fallback path.
730    This assumes the blkdev's low level compat_ioctl always returns
731    ENOIOCTLCMD for unknown ioctls. */
732 long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg)
733 {
734 	int ret;
735 	void __user *argp = compat_ptr(arg);
736 	struct inode *inode = file->f_mapping->host;
737 	struct block_device *bdev = inode->i_bdev;
738 	struct gendisk *disk = bdev->bd_disk;
739 	fmode_t mode = file->f_mode;
740 	loff_t size;
741 
742 	/*
743 	 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
744 	 * to updated it before every ioctl.
745 	 */
746 	if (file->f_flags & O_NDELAY)
747 		mode |= FMODE_NDELAY;
748 	else
749 		mode &= ~FMODE_NDELAY;
750 
751 	switch (cmd) {
752 	/* These need separate implementations for the data structure */
753 	case HDIO_GETGEO:
754 		return compat_hdio_getgeo(bdev, argp);
755 	case BLKPG:
756 		return compat_blkpg_ioctl(bdev, argp);
757 
758 	/* Compat mode returns 32-bit data instead of 'long' */
759 	case BLKRAGET:
760 	case BLKFRAGET:
761 		if (!argp)
762 			return -EINVAL;
763 		return compat_put_long(argp,
764 			       (bdev->bd_bdi->ra_pages * PAGE_SIZE) / 512);
765 	case BLKGETSIZE:
766 		size = i_size_read(bdev->bd_inode);
767 		if ((size >> 9) > ~0UL)
768 			return -EFBIG;
769 		return compat_put_ulong(argp, size >> 9);
770 
771 	/* The data is compatible, but the command number is different */
772 	case BLKBSZGET_32: /* get the logical block size (cf. BLKSSZGET) */
773 		return put_int(argp, bdev_logical_block_size(bdev));
774 	case BLKBSZSET_32:
775 		return blkdev_bszset(bdev, mode, argp);
776 	case BLKGETSIZE64_32:
777 		return put_u64(argp, i_size_read(bdev->bd_inode));
778 
779 	/* Incompatible alignment on i386 */
780 	case BLKTRACESETUP32:
781 		return blk_trace_ioctl(bdev, cmd, argp);
782 	default:
783 		break;
784 	}
785 
786 	ret = blkdev_common_ioctl(bdev, mode, cmd, arg, argp);
787 	if (ret == -ENOIOCTLCMD && disk->fops->compat_ioctl)
788 		ret = disk->fops->compat_ioctl(bdev, mode, cmd, arg);
789 
790 	return ret;
791 }
792 #endif
793