xref: /openbmc/linux/block/ioctl.c (revision f1770e3c)
1 #include <linux/capability.h>
2 #include <linux/blkdev.h>
3 #include <linux/export.h>
4 #include <linux/gfp.h>
5 #include <linux/blkpg.h>
6 #include <linux/hdreg.h>
7 #include <linux/backing-dev.h>
8 #include <linux/fs.h>
9 #include <linux/blktrace_api.h>
10 #include <linux/pr.h>
11 #include <linux/uaccess.h>
12 
13 static int blkpg_ioctl(struct block_device *bdev, struct blkpg_ioctl_arg __user *arg)
14 {
15 	struct block_device *bdevp;
16 	struct gendisk *disk;
17 	struct hd_struct *part, *lpart;
18 	struct blkpg_ioctl_arg a;
19 	struct blkpg_partition p;
20 	struct disk_part_iter piter;
21 	long long start, length;
22 	int partno;
23 
24 	if (!capable(CAP_SYS_ADMIN))
25 		return -EACCES;
26 	if (copy_from_user(&a, arg, sizeof(struct blkpg_ioctl_arg)))
27 		return -EFAULT;
28 	if (copy_from_user(&p, a.data, 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 (a.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 /*
158  * This is an exported API for the block driver, and will not
159  * acquire bd_mutex. This API should be used in case that
160  * caller has held bd_mutex already.
161  */
162 int __blkdev_reread_part(struct block_device *bdev)
163 {
164 	struct gendisk *disk = bdev->bd_disk;
165 
166 	if (!disk_part_scan_enabled(disk) || bdev != bdev->bd_contains)
167 		return -EINVAL;
168 	if (!capable(CAP_SYS_ADMIN))
169 		return -EACCES;
170 
171 	lockdep_assert_held(&bdev->bd_mutex);
172 
173 	return rescan_partitions(disk, bdev);
174 }
175 EXPORT_SYMBOL(__blkdev_reread_part);
176 
177 /*
178  * This is an exported API for the block driver, and will
179  * try to acquire bd_mutex. If bd_mutex has been held already
180  * in current context, please call __blkdev_reread_part().
181  *
182  * Make sure the held locks in current context aren't required
183  * in open()/close() handler and I/O path for avoiding ABBA deadlock:
184  * - bd_mutex is held before calling block driver's open/close
185  *   handler
186  * - reading partition table may submit I/O to the block device
187  */
188 int blkdev_reread_part(struct block_device *bdev)
189 {
190 	int res;
191 
192 	mutex_lock(&bdev->bd_mutex);
193 	res = __blkdev_reread_part(bdev);
194 	mutex_unlock(&bdev->bd_mutex);
195 
196 	return res;
197 }
198 EXPORT_SYMBOL(blkdev_reread_part);
199 
200 static int blk_ioctl_discard(struct block_device *bdev, fmode_t mode,
201 		unsigned long arg, unsigned long flags)
202 {
203 	uint64_t range[2];
204 	uint64_t start, len;
205 	struct request_queue *q = bdev_get_queue(bdev);
206 	struct address_space *mapping = bdev->bd_inode->i_mapping;
207 
208 
209 	if (!(mode & FMODE_WRITE))
210 		return -EBADF;
211 
212 	if (!blk_queue_discard(q))
213 		return -EOPNOTSUPP;
214 
215 	if (copy_from_user(range, (void __user *)arg, sizeof(range)))
216 		return -EFAULT;
217 
218 	start = range[0];
219 	len = range[1];
220 
221 	if (start & 511)
222 		return -EINVAL;
223 	if (len & 511)
224 		return -EINVAL;
225 
226 	if (start + len > i_size_read(bdev->bd_inode))
227 		return -EINVAL;
228 	truncate_inode_pages_range(mapping, start, start + len - 1);
229 	return blkdev_issue_discard(bdev, start >> 9, len >> 9,
230 				    GFP_KERNEL, flags);
231 }
232 
233 static int blk_ioctl_zeroout(struct block_device *bdev, fmode_t mode,
234 		unsigned long arg)
235 {
236 	uint64_t range[2];
237 	struct address_space *mapping;
238 	uint64_t start, end, len;
239 
240 	if (!(mode & FMODE_WRITE))
241 		return -EBADF;
242 
243 	if (copy_from_user(range, (void __user *)arg, sizeof(range)))
244 		return -EFAULT;
245 
246 	start = range[0];
247 	len = range[1];
248 	end = start + len - 1;
249 
250 	if (start & 511)
251 		return -EINVAL;
252 	if (len & 511)
253 		return -EINVAL;
254 	if (end >= (uint64_t)i_size_read(bdev->bd_inode))
255 		return -EINVAL;
256 	if (end < start)
257 		return -EINVAL;
258 
259 	/* Invalidate the page cache, including dirty pages */
260 	mapping = bdev->bd_inode->i_mapping;
261 	truncate_inode_pages_range(mapping, start, end);
262 
263 	return blkdev_issue_zeroout(bdev, start >> 9, len >> 9, GFP_KERNEL,
264 			BLKDEV_ZERO_NOUNMAP);
265 }
266 
267 static int put_ushort(unsigned long arg, unsigned short val)
268 {
269 	return put_user(val, (unsigned short __user *)arg);
270 }
271 
272 static int put_int(unsigned long arg, int val)
273 {
274 	return put_user(val, (int __user *)arg);
275 }
276 
277 static int put_uint(unsigned long arg, unsigned int val)
278 {
279 	return put_user(val, (unsigned int __user *)arg);
280 }
281 
282 static int put_long(unsigned long arg, long val)
283 {
284 	return put_user(val, (long __user *)arg);
285 }
286 
287 static int put_ulong(unsigned long arg, unsigned long val)
288 {
289 	return put_user(val, (unsigned long __user *)arg);
290 }
291 
292 static int put_u64(unsigned long arg, u64 val)
293 {
294 	return put_user(val, (u64 __user *)arg);
295 }
296 
297 int __blkdev_driver_ioctl(struct block_device *bdev, fmode_t mode,
298 			unsigned cmd, unsigned long arg)
299 {
300 	struct gendisk *disk = bdev->bd_disk;
301 
302 	if (disk->fops->ioctl)
303 		return disk->fops->ioctl(bdev, mode, cmd, arg);
304 
305 	return -ENOTTY;
306 }
307 /*
308  * For the record: _GPL here is only because somebody decided to slap it
309  * on the previous export.  Sheer idiocy, since it wasn't copyrightable
310  * at all and could be open-coded without any exports by anybody who cares.
311  */
312 EXPORT_SYMBOL_GPL(__blkdev_driver_ioctl);
313 
314 static int blkdev_pr_register(struct block_device *bdev,
315 		struct pr_registration __user *arg)
316 {
317 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
318 	struct pr_registration reg;
319 
320 	if (!capable(CAP_SYS_ADMIN))
321 		return -EPERM;
322 	if (!ops || !ops->pr_register)
323 		return -EOPNOTSUPP;
324 	if (copy_from_user(&reg, arg, sizeof(reg)))
325 		return -EFAULT;
326 
327 	if (reg.flags & ~PR_FL_IGNORE_KEY)
328 		return -EOPNOTSUPP;
329 	return ops->pr_register(bdev, reg.old_key, reg.new_key, reg.flags);
330 }
331 
332 static int blkdev_pr_reserve(struct block_device *bdev,
333 		struct pr_reservation __user *arg)
334 {
335 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
336 	struct pr_reservation rsv;
337 
338 	if (!capable(CAP_SYS_ADMIN))
339 		return -EPERM;
340 	if (!ops || !ops->pr_reserve)
341 		return -EOPNOTSUPP;
342 	if (copy_from_user(&rsv, arg, sizeof(rsv)))
343 		return -EFAULT;
344 
345 	if (rsv.flags & ~PR_FL_IGNORE_KEY)
346 		return -EOPNOTSUPP;
347 	return ops->pr_reserve(bdev, rsv.key, rsv.type, rsv.flags);
348 }
349 
350 static int blkdev_pr_release(struct block_device *bdev,
351 		struct pr_reservation __user *arg)
352 {
353 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
354 	struct pr_reservation rsv;
355 
356 	if (!capable(CAP_SYS_ADMIN))
357 		return -EPERM;
358 	if (!ops || !ops->pr_release)
359 		return -EOPNOTSUPP;
360 	if (copy_from_user(&rsv, arg, sizeof(rsv)))
361 		return -EFAULT;
362 
363 	if (rsv.flags)
364 		return -EOPNOTSUPP;
365 	return ops->pr_release(bdev, rsv.key, rsv.type);
366 }
367 
368 static int blkdev_pr_preempt(struct block_device *bdev,
369 		struct pr_preempt __user *arg, bool abort)
370 {
371 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
372 	struct pr_preempt p;
373 
374 	if (!capable(CAP_SYS_ADMIN))
375 		return -EPERM;
376 	if (!ops || !ops->pr_preempt)
377 		return -EOPNOTSUPP;
378 	if (copy_from_user(&p, arg, sizeof(p)))
379 		return -EFAULT;
380 
381 	if (p.flags)
382 		return -EOPNOTSUPP;
383 	return ops->pr_preempt(bdev, p.old_key, p.new_key, p.type, abort);
384 }
385 
386 static int blkdev_pr_clear(struct block_device *bdev,
387 		struct pr_clear __user *arg)
388 {
389 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
390 	struct pr_clear c;
391 
392 	if (!capable(CAP_SYS_ADMIN))
393 		return -EPERM;
394 	if (!ops || !ops->pr_clear)
395 		return -EOPNOTSUPP;
396 	if (copy_from_user(&c, arg, sizeof(c)))
397 		return -EFAULT;
398 
399 	if (c.flags)
400 		return -EOPNOTSUPP;
401 	return ops->pr_clear(bdev, c.key);
402 }
403 
404 /*
405  * Is it an unrecognized ioctl? The correct returns are either
406  * ENOTTY (final) or ENOIOCTLCMD ("I don't know this one, try a
407  * fallback"). ENOIOCTLCMD gets turned into ENOTTY by the ioctl
408  * code before returning.
409  *
410  * Confused drivers sometimes return EINVAL, which is wrong. It
411  * means "I understood the ioctl command, but the parameters to
412  * it were wrong".
413  *
414  * We should aim to just fix the broken drivers, the EINVAL case
415  * should go away.
416  */
417 static inline int is_unrecognized_ioctl(int ret)
418 {
419 	return	ret == -EINVAL ||
420 		ret == -ENOTTY ||
421 		ret == -ENOIOCTLCMD;
422 }
423 
424 static int blkdev_flushbuf(struct block_device *bdev, fmode_t mode,
425 		unsigned cmd, unsigned long arg)
426 {
427 	int ret;
428 
429 	if (!capable(CAP_SYS_ADMIN))
430 		return -EACCES;
431 
432 	ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
433 	if (!is_unrecognized_ioctl(ret))
434 		return ret;
435 
436 	fsync_bdev(bdev);
437 	invalidate_bdev(bdev);
438 	return 0;
439 }
440 
441 static int blkdev_roset(struct block_device *bdev, fmode_t mode,
442 		unsigned cmd, unsigned long arg)
443 {
444 	int ret, n;
445 
446 	if (!capable(CAP_SYS_ADMIN))
447 		return -EACCES;
448 
449 	ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
450 	if (!is_unrecognized_ioctl(ret))
451 		return ret;
452 	if (get_user(n, (int __user *)arg))
453 		return -EFAULT;
454 	set_device_ro(bdev, n);
455 	return 0;
456 }
457 
458 static int blkdev_getgeo(struct block_device *bdev,
459 		struct hd_geometry __user *argp)
460 {
461 	struct gendisk *disk = bdev->bd_disk;
462 	struct hd_geometry geo;
463 	int ret;
464 
465 	if (!argp)
466 		return -EINVAL;
467 	if (!disk->fops->getgeo)
468 		return -ENOTTY;
469 
470 	/*
471 	 * We need to set the startsect first, the driver may
472 	 * want to override it.
473 	 */
474 	memset(&geo, 0, sizeof(geo));
475 	geo.start = get_start_sect(bdev);
476 	ret = disk->fops->getgeo(bdev, &geo);
477 	if (ret)
478 		return ret;
479 	if (copy_to_user(argp, &geo, sizeof(geo)))
480 		return -EFAULT;
481 	return 0;
482 }
483 
484 /* set the logical block size */
485 static int blkdev_bszset(struct block_device *bdev, fmode_t mode,
486 		int __user *argp)
487 {
488 	int ret, n;
489 
490 	if (!capable(CAP_SYS_ADMIN))
491 		return -EACCES;
492 	if (!argp)
493 		return -EINVAL;
494 	if (get_user(n, argp))
495 		return -EFAULT;
496 
497 	if (!(mode & FMODE_EXCL)) {
498 		bdgrab(bdev);
499 		if (blkdev_get(bdev, mode | FMODE_EXCL, &bdev) < 0)
500 			return -EBUSY;
501 	}
502 
503 	ret = set_blocksize(bdev, n);
504 	if (!(mode & FMODE_EXCL))
505 		blkdev_put(bdev, mode | FMODE_EXCL);
506 	return ret;
507 }
508 
509 /*
510  * always keep this in sync with compat_blkdev_ioctl()
511  */
512 int blkdev_ioctl(struct block_device *bdev, fmode_t mode, unsigned cmd,
513 			unsigned long arg)
514 {
515 	void __user *argp = (void __user *)arg;
516 	loff_t size;
517 	unsigned int max_sectors;
518 
519 	switch (cmd) {
520 	case BLKFLSBUF:
521 		return blkdev_flushbuf(bdev, mode, cmd, arg);
522 	case BLKROSET:
523 		return blkdev_roset(bdev, mode, cmd, arg);
524 	case BLKDISCARD:
525 		return blk_ioctl_discard(bdev, mode, arg, 0);
526 	case BLKSECDISCARD:
527 		return blk_ioctl_discard(bdev, mode, arg,
528 				BLKDEV_DISCARD_SECURE);
529 	case BLKZEROOUT:
530 		return blk_ioctl_zeroout(bdev, mode, arg);
531 	case BLKREPORTZONE:
532 		return blkdev_report_zones_ioctl(bdev, mode, cmd, arg);
533 	case BLKRESETZONE:
534 		return blkdev_reset_zones_ioctl(bdev, mode, cmd, arg);
535 	case BLKGETZONESZ:
536 		return put_uint(arg, bdev_zone_sectors(bdev));
537 	case BLKGETNRZONES:
538 		return put_uint(arg, blkdev_nr_zones(bdev));
539 	case HDIO_GETGEO:
540 		return blkdev_getgeo(bdev, argp);
541 	case BLKRAGET:
542 	case BLKFRAGET:
543 		if (!arg)
544 			return -EINVAL;
545 		return put_long(arg, (bdev->bd_bdi->ra_pages*PAGE_SIZE) / 512);
546 	case BLKROGET:
547 		return put_int(arg, bdev_read_only(bdev) != 0);
548 	case BLKBSZGET: /* get block device soft block size (cf. BLKSSZGET) */
549 		return put_int(arg, block_size(bdev));
550 	case BLKSSZGET: /* get block device logical block size */
551 		return put_int(arg, bdev_logical_block_size(bdev));
552 	case BLKPBSZGET: /* get block device physical block size */
553 		return put_uint(arg, bdev_physical_block_size(bdev));
554 	case BLKIOMIN:
555 		return put_uint(arg, bdev_io_min(bdev));
556 	case BLKIOOPT:
557 		return put_uint(arg, bdev_io_opt(bdev));
558 	case BLKALIGNOFF:
559 		return put_int(arg, bdev_alignment_offset(bdev));
560 	case BLKDISCARDZEROES:
561 		return put_uint(arg, 0);
562 	case BLKSECTGET:
563 		max_sectors = min_t(unsigned int, USHRT_MAX,
564 				    queue_max_sectors(bdev_get_queue(bdev)));
565 		return put_ushort(arg, max_sectors);
566 	case BLKROTATIONAL:
567 		return put_ushort(arg, !blk_queue_nonrot(bdev_get_queue(bdev)));
568 	case BLKRASET:
569 	case BLKFRASET:
570 		if(!capable(CAP_SYS_ADMIN))
571 			return -EACCES;
572 		bdev->bd_bdi->ra_pages = (arg * 512) / PAGE_SIZE;
573 		return 0;
574 	case BLKBSZSET:
575 		return blkdev_bszset(bdev, mode, argp);
576 	case BLKPG:
577 		return blkpg_ioctl(bdev, argp);
578 	case BLKRRPART:
579 		return blkdev_reread_part(bdev);
580 	case BLKGETSIZE:
581 		size = i_size_read(bdev->bd_inode);
582 		if ((size >> 9) > ~0UL)
583 			return -EFBIG;
584 		return put_ulong(arg, size >> 9);
585 	case BLKGETSIZE64:
586 		return put_u64(arg, i_size_read(bdev->bd_inode));
587 	case BLKTRACESTART:
588 	case BLKTRACESTOP:
589 	case BLKTRACESETUP:
590 	case BLKTRACETEARDOWN:
591 		return blk_trace_ioctl(bdev, cmd, argp);
592 	case IOC_PR_REGISTER:
593 		return blkdev_pr_register(bdev, argp);
594 	case IOC_PR_RESERVE:
595 		return blkdev_pr_reserve(bdev, argp);
596 	case IOC_PR_RELEASE:
597 		return blkdev_pr_release(bdev, argp);
598 	case IOC_PR_PREEMPT:
599 		return blkdev_pr_preempt(bdev, argp, false);
600 	case IOC_PR_PREEMPT_ABORT:
601 		return blkdev_pr_preempt(bdev, argp, true);
602 	case IOC_PR_CLEAR:
603 		return blkdev_pr_clear(bdev, argp);
604 	default:
605 		return __blkdev_driver_ioctl(bdev, mode, cmd, arg);
606 	}
607 }
608 EXPORT_SYMBOL_GPL(blkdev_ioctl);
609