xref: /linux/block/ioctl.c (revision 7ff836f064e2c814a7504c91a4464eea45d475bd)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/capability.h>
3 #include <linux/blkdev.h>
4 #include <linux/export.h>
5 #include <linux/gfp.h>
6 #include <linux/blkpg.h>
7 #include <linux/hdreg.h>
8 #include <linux/backing-dev.h>
9 #include <linux/fs.h>
10 #include <linux/blktrace_api.h>
11 #include <linux/pr.h>
12 #include <linux/uaccess.h>
13 
14 static int blkpg_ioctl(struct block_device *bdev, struct blkpg_ioctl_arg __user *arg)
15 {
16 	struct block_device *bdevp;
17 	struct gendisk *disk;
18 	struct hd_struct *part, *lpart;
19 	struct blkpg_ioctl_arg a;
20 	struct blkpg_partition p;
21 	struct disk_part_iter piter;
22 	long long start, length;
23 	int partno;
24 
25 	if (!capable(CAP_SYS_ADMIN))
26 		return -EACCES;
27 	if (copy_from_user(&a, arg, sizeof(struct blkpg_ioctl_arg)))
28 		return -EFAULT;
29 	if (copy_from_user(&p, a.data, sizeof(struct blkpg_partition)))
30 		return -EFAULT;
31 	disk = bdev->bd_disk;
32 	if (bdev != bdev->bd_contains)
33 		return -EINVAL;
34 	partno = p.pno;
35 	if (partno <= 0)
36 		return -EINVAL;
37 	switch (a.op) {
38 		case BLKPG_ADD_PARTITION:
39 			start = p.start >> 9;
40 			length = p.length >> 9;
41 			/* check for fit in a hd_struct */
42 			if (sizeof(sector_t) == sizeof(long) &&
43 			    sizeof(long long) > sizeof(long)) {
44 				long pstart = start, plength = length;
45 				if (pstart != start || plength != length
46 				    || pstart < 0 || plength < 0 || partno > 65535)
47 					return -EINVAL;
48 			}
49 			/* check if partition is aligned to blocksize */
50 			if (p.start & (bdev_logical_block_size(bdev) - 1))
51 				return -EINVAL;
52 
53 			mutex_lock(&bdev->bd_mutex);
54 
55 			/* overlap? */
56 			disk_part_iter_init(&piter, disk,
57 					    DISK_PITER_INCL_EMPTY);
58 			while ((part = disk_part_iter_next(&piter))) {
59 				if (!(start + length <= part->start_sect ||
60 				      start >= part->start_sect + part->nr_sects)) {
61 					disk_part_iter_exit(&piter);
62 					mutex_unlock(&bdev->bd_mutex);
63 					return -EBUSY;
64 				}
65 			}
66 			disk_part_iter_exit(&piter);
67 
68 			/* all seems OK */
69 			part = add_partition(disk, partno, start, length,
70 					     ADDPART_FLAG_NONE, NULL);
71 			mutex_unlock(&bdev->bd_mutex);
72 			return PTR_ERR_OR_ZERO(part);
73 		case BLKPG_DEL_PARTITION:
74 			part = disk_get_part(disk, partno);
75 			if (!part)
76 				return -ENXIO;
77 
78 			bdevp = bdget(part_devt(part));
79 			disk_put_part(part);
80 			if (!bdevp)
81 				return -ENOMEM;
82 
83 			mutex_lock(&bdevp->bd_mutex);
84 			if (bdevp->bd_openers) {
85 				mutex_unlock(&bdevp->bd_mutex);
86 				bdput(bdevp);
87 				return -EBUSY;
88 			}
89 			/* all seems OK */
90 			fsync_bdev(bdevp);
91 			invalidate_bdev(bdevp);
92 
93 			mutex_lock_nested(&bdev->bd_mutex, 1);
94 			delete_partition(disk, partno);
95 			mutex_unlock(&bdev->bd_mutex);
96 			mutex_unlock(&bdevp->bd_mutex);
97 			bdput(bdevp);
98 
99 			return 0;
100 		case BLKPG_RESIZE_PARTITION:
101 			start = p.start >> 9;
102 			/* new length of partition in bytes */
103 			length = p.length >> 9;
104 			/* check for fit in a hd_struct */
105 			if (sizeof(sector_t) == sizeof(long) &&
106 			    sizeof(long long) > sizeof(long)) {
107 				long pstart = start, plength = length;
108 				if (pstart != start || plength != length
109 				    || pstart < 0 || plength < 0)
110 					return -EINVAL;
111 			}
112 			part = disk_get_part(disk, partno);
113 			if (!part)
114 				return -ENXIO;
115 			bdevp = bdget(part_devt(part));
116 			if (!bdevp) {
117 				disk_put_part(part);
118 				return -ENOMEM;
119 			}
120 			mutex_lock(&bdevp->bd_mutex);
121 			mutex_lock_nested(&bdev->bd_mutex, 1);
122 			if (start != part->start_sect) {
123 				mutex_unlock(&bdevp->bd_mutex);
124 				mutex_unlock(&bdev->bd_mutex);
125 				bdput(bdevp);
126 				disk_put_part(part);
127 				return -EINVAL;
128 			}
129 			/* overlap? */
130 			disk_part_iter_init(&piter, disk,
131 					    DISK_PITER_INCL_EMPTY);
132 			while ((lpart = disk_part_iter_next(&piter))) {
133 				if (lpart->partno != partno &&
134 				   !(start + length <= lpart->start_sect ||
135 				   start >= lpart->start_sect + lpart->nr_sects)
136 				   ) {
137 					disk_part_iter_exit(&piter);
138 					mutex_unlock(&bdevp->bd_mutex);
139 					mutex_unlock(&bdev->bd_mutex);
140 					bdput(bdevp);
141 					disk_put_part(part);
142 					return -EBUSY;
143 				}
144 			}
145 			disk_part_iter_exit(&piter);
146 			part_nr_sects_write(part, (sector_t)length);
147 			i_size_write(bdevp->bd_inode, p.length);
148 			mutex_unlock(&bdevp->bd_mutex);
149 			mutex_unlock(&bdev->bd_mutex);
150 			bdput(bdevp);
151 			disk_put_part(part);
152 			return 0;
153 		default:
154 			return -EINVAL;
155 	}
156 }
157 
158 /*
159  * This is an exported API for the block driver, and will not
160  * acquire bd_mutex. This API should be used in case that
161  * caller has held bd_mutex already.
162  */
163 int __blkdev_reread_part(struct block_device *bdev)
164 {
165 	struct gendisk *disk = bdev->bd_disk;
166 
167 	if (!disk_part_scan_enabled(disk) || bdev != bdev->bd_contains)
168 		return -EINVAL;
169 	if (!capable(CAP_SYS_ADMIN))
170 		return -EACCES;
171 
172 	lockdep_assert_held(&bdev->bd_mutex);
173 
174 	return rescan_partitions(disk, bdev);
175 }
176 EXPORT_SYMBOL(__blkdev_reread_part);
177 
178 /*
179  * This is an exported API for the block driver, and will
180  * try to acquire bd_mutex. If bd_mutex has been held already
181  * in current context, please call __blkdev_reread_part().
182  *
183  * Make sure the held locks in current context aren't required
184  * in open()/close() handler and I/O path for avoiding ABBA deadlock:
185  * - bd_mutex is held before calling block driver's open/close
186  *   handler
187  * - reading partition table may submit I/O to the block device
188  */
189 int blkdev_reread_part(struct block_device *bdev)
190 {
191 	int res;
192 
193 	mutex_lock(&bdev->bd_mutex);
194 	res = __blkdev_reread_part(bdev);
195 	mutex_unlock(&bdev->bd_mutex);
196 
197 	return res;
198 }
199 EXPORT_SYMBOL(blkdev_reread_part);
200 
201 static int blk_ioctl_discard(struct block_device *bdev, fmode_t mode,
202 		unsigned long arg, unsigned long flags)
203 {
204 	uint64_t range[2];
205 	uint64_t start, len;
206 	struct request_queue *q = bdev_get_queue(bdev);
207 	struct address_space *mapping = bdev->bd_inode->i_mapping;
208 
209 
210 	if (!(mode & FMODE_WRITE))
211 		return -EBADF;
212 
213 	if (!blk_queue_discard(q))
214 		return -EOPNOTSUPP;
215 
216 	if (copy_from_user(range, (void __user *)arg, sizeof(range)))
217 		return -EFAULT;
218 
219 	start = range[0];
220 	len = range[1];
221 
222 	if (start & 511)
223 		return -EINVAL;
224 	if (len & 511)
225 		return -EINVAL;
226 
227 	if (start + len > i_size_read(bdev->bd_inode))
228 		return -EINVAL;
229 	truncate_inode_pages_range(mapping, start, start + len - 1);
230 	return blkdev_issue_discard(bdev, start >> 9, len >> 9,
231 				    GFP_KERNEL, flags);
232 }
233 
234 static int blk_ioctl_zeroout(struct block_device *bdev, fmode_t mode,
235 		unsigned long arg)
236 {
237 	uint64_t range[2];
238 	struct address_space *mapping;
239 	uint64_t start, end, len;
240 
241 	if (!(mode & FMODE_WRITE))
242 		return -EBADF;
243 
244 	if (copy_from_user(range, (void __user *)arg, sizeof(range)))
245 		return -EFAULT;
246 
247 	start = range[0];
248 	len = range[1];
249 	end = start + len - 1;
250 
251 	if (start & 511)
252 		return -EINVAL;
253 	if (len & 511)
254 		return -EINVAL;
255 	if (end >= (uint64_t)i_size_read(bdev->bd_inode))
256 		return -EINVAL;
257 	if (end < start)
258 		return -EINVAL;
259 
260 	/* Invalidate the page cache, including dirty pages */
261 	mapping = bdev->bd_inode->i_mapping;
262 	truncate_inode_pages_range(mapping, start, end);
263 
264 	return blkdev_issue_zeroout(bdev, start >> 9, len >> 9, GFP_KERNEL,
265 			BLKDEV_ZERO_NOUNMAP);
266 }
267 
268 static int put_ushort(unsigned long arg, unsigned short val)
269 {
270 	return put_user(val, (unsigned short __user *)arg);
271 }
272 
273 static int put_int(unsigned long arg, int val)
274 {
275 	return put_user(val, (int __user *)arg);
276 }
277 
278 static int put_uint(unsigned long arg, unsigned int val)
279 {
280 	return put_user(val, (unsigned int __user *)arg);
281 }
282 
283 static int put_long(unsigned long arg, long val)
284 {
285 	return put_user(val, (long __user *)arg);
286 }
287 
288 static int put_ulong(unsigned long arg, unsigned long val)
289 {
290 	return put_user(val, (unsigned long __user *)arg);
291 }
292 
293 static int put_u64(unsigned long arg, u64 val)
294 {
295 	return put_user(val, (u64 __user *)arg);
296 }
297 
298 int __blkdev_driver_ioctl(struct block_device *bdev, fmode_t mode,
299 			unsigned cmd, unsigned long arg)
300 {
301 	struct gendisk *disk = bdev->bd_disk;
302 
303 	if (disk->fops->ioctl)
304 		return disk->fops->ioctl(bdev, mode, cmd, arg);
305 
306 	return -ENOTTY;
307 }
308 /*
309  * For the record: _GPL here is only because somebody decided to slap it
310  * on the previous export.  Sheer idiocy, since it wasn't copyrightable
311  * at all and could be open-coded without any exports by anybody who cares.
312  */
313 EXPORT_SYMBOL_GPL(__blkdev_driver_ioctl);
314 
315 static int blkdev_pr_register(struct block_device *bdev,
316 		struct pr_registration __user *arg)
317 {
318 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
319 	struct pr_registration reg;
320 
321 	if (!capable(CAP_SYS_ADMIN))
322 		return -EPERM;
323 	if (!ops || !ops->pr_register)
324 		return -EOPNOTSUPP;
325 	if (copy_from_user(&reg, arg, sizeof(reg)))
326 		return -EFAULT;
327 
328 	if (reg.flags & ~PR_FL_IGNORE_KEY)
329 		return -EOPNOTSUPP;
330 	return ops->pr_register(bdev, reg.old_key, reg.new_key, reg.flags);
331 }
332 
333 static int blkdev_pr_reserve(struct block_device *bdev,
334 		struct pr_reservation __user *arg)
335 {
336 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
337 	struct pr_reservation rsv;
338 
339 	if (!capable(CAP_SYS_ADMIN))
340 		return -EPERM;
341 	if (!ops || !ops->pr_reserve)
342 		return -EOPNOTSUPP;
343 	if (copy_from_user(&rsv, arg, sizeof(rsv)))
344 		return -EFAULT;
345 
346 	if (rsv.flags & ~PR_FL_IGNORE_KEY)
347 		return -EOPNOTSUPP;
348 	return ops->pr_reserve(bdev, rsv.key, rsv.type, rsv.flags);
349 }
350 
351 static int blkdev_pr_release(struct block_device *bdev,
352 		struct pr_reservation __user *arg)
353 {
354 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
355 	struct pr_reservation rsv;
356 
357 	if (!capable(CAP_SYS_ADMIN))
358 		return -EPERM;
359 	if (!ops || !ops->pr_release)
360 		return -EOPNOTSUPP;
361 	if (copy_from_user(&rsv, arg, sizeof(rsv)))
362 		return -EFAULT;
363 
364 	if (rsv.flags)
365 		return -EOPNOTSUPP;
366 	return ops->pr_release(bdev, rsv.key, rsv.type);
367 }
368 
369 static int blkdev_pr_preempt(struct block_device *bdev,
370 		struct pr_preempt __user *arg, bool abort)
371 {
372 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
373 	struct pr_preempt p;
374 
375 	if (!capable(CAP_SYS_ADMIN))
376 		return -EPERM;
377 	if (!ops || !ops->pr_preempt)
378 		return -EOPNOTSUPP;
379 	if (copy_from_user(&p, arg, sizeof(p)))
380 		return -EFAULT;
381 
382 	if (p.flags)
383 		return -EOPNOTSUPP;
384 	return ops->pr_preempt(bdev, p.old_key, p.new_key, p.type, abort);
385 }
386 
387 static int blkdev_pr_clear(struct block_device *bdev,
388 		struct pr_clear __user *arg)
389 {
390 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
391 	struct pr_clear c;
392 
393 	if (!capable(CAP_SYS_ADMIN))
394 		return -EPERM;
395 	if (!ops || !ops->pr_clear)
396 		return -EOPNOTSUPP;
397 	if (copy_from_user(&c, arg, sizeof(c)))
398 		return -EFAULT;
399 
400 	if (c.flags)
401 		return -EOPNOTSUPP;
402 	return ops->pr_clear(bdev, c.key);
403 }
404 
405 /*
406  * Is it an unrecognized ioctl? The correct returns are either
407  * ENOTTY (final) or ENOIOCTLCMD ("I don't know this one, try a
408  * fallback"). ENOIOCTLCMD gets turned into ENOTTY by the ioctl
409  * code before returning.
410  *
411  * Confused drivers sometimes return EINVAL, which is wrong. It
412  * means "I understood the ioctl command, but the parameters to
413  * it were wrong".
414  *
415  * We should aim to just fix the broken drivers, the EINVAL case
416  * should go away.
417  */
418 static inline int is_unrecognized_ioctl(int ret)
419 {
420 	return	ret == -EINVAL ||
421 		ret == -ENOTTY ||
422 		ret == -ENOIOCTLCMD;
423 }
424 
425 static int blkdev_flushbuf(struct block_device *bdev, fmode_t mode,
426 		unsigned cmd, unsigned long arg)
427 {
428 	int ret;
429 
430 	if (!capable(CAP_SYS_ADMIN))
431 		return -EACCES;
432 
433 	ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
434 	if (!is_unrecognized_ioctl(ret))
435 		return ret;
436 
437 	fsync_bdev(bdev);
438 	invalidate_bdev(bdev);
439 	return 0;
440 }
441 
442 static int blkdev_roset(struct block_device *bdev, fmode_t mode,
443 		unsigned cmd, unsigned long arg)
444 {
445 	int ret, n;
446 
447 	if (!capable(CAP_SYS_ADMIN))
448 		return -EACCES;
449 
450 	ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
451 	if (!is_unrecognized_ioctl(ret))
452 		return ret;
453 	if (get_user(n, (int __user *)arg))
454 		return -EFAULT;
455 	set_device_ro(bdev, n);
456 	return 0;
457 }
458 
459 static int blkdev_getgeo(struct block_device *bdev,
460 		struct hd_geometry __user *argp)
461 {
462 	struct gendisk *disk = bdev->bd_disk;
463 	struct hd_geometry geo;
464 	int ret;
465 
466 	if (!argp)
467 		return -EINVAL;
468 	if (!disk->fops->getgeo)
469 		return -ENOTTY;
470 
471 	/*
472 	 * We need to set the startsect first, the driver may
473 	 * want to override it.
474 	 */
475 	memset(&geo, 0, sizeof(geo));
476 	geo.start = get_start_sect(bdev);
477 	ret = disk->fops->getgeo(bdev, &geo);
478 	if (ret)
479 		return ret;
480 	if (copy_to_user(argp, &geo, sizeof(geo)))
481 		return -EFAULT;
482 	return 0;
483 }
484 
485 /* set the logical block size */
486 static int blkdev_bszset(struct block_device *bdev, fmode_t mode,
487 		int __user *argp)
488 {
489 	int ret, n;
490 
491 	if (!capable(CAP_SYS_ADMIN))
492 		return -EACCES;
493 	if (!argp)
494 		return -EINVAL;
495 	if (get_user(n, argp))
496 		return -EFAULT;
497 
498 	if (!(mode & FMODE_EXCL)) {
499 		bdgrab(bdev);
500 		if (blkdev_get(bdev, mode | FMODE_EXCL, &bdev) < 0)
501 			return -EBUSY;
502 	}
503 
504 	ret = set_blocksize(bdev, n);
505 	if (!(mode & FMODE_EXCL))
506 		blkdev_put(bdev, mode | FMODE_EXCL);
507 	return ret;
508 }
509 
510 /*
511  * always keep this in sync with compat_blkdev_ioctl()
512  */
513 int blkdev_ioctl(struct block_device *bdev, fmode_t mode, unsigned cmd,
514 			unsigned long arg)
515 {
516 	void __user *argp = (void __user *)arg;
517 	loff_t size;
518 	unsigned int max_sectors;
519 
520 	switch (cmd) {
521 	case BLKFLSBUF:
522 		return blkdev_flushbuf(bdev, mode, cmd, arg);
523 	case BLKROSET:
524 		return blkdev_roset(bdev, mode, cmd, arg);
525 	case BLKDISCARD:
526 		return blk_ioctl_discard(bdev, mode, arg, 0);
527 	case BLKSECDISCARD:
528 		return blk_ioctl_discard(bdev, mode, arg,
529 				BLKDEV_DISCARD_SECURE);
530 	case BLKZEROOUT:
531 		return blk_ioctl_zeroout(bdev, mode, arg);
532 	case BLKREPORTZONE:
533 		return blkdev_report_zones_ioctl(bdev, mode, cmd, arg);
534 	case BLKRESETZONE:
535 		return blkdev_reset_zones_ioctl(bdev, mode, cmd, arg);
536 	case BLKGETZONESZ:
537 		return put_uint(arg, bdev_zone_sectors(bdev));
538 	case BLKGETNRZONES:
539 		return put_uint(arg, blkdev_nr_zones(bdev));
540 	case HDIO_GETGEO:
541 		return blkdev_getgeo(bdev, argp);
542 	case BLKRAGET:
543 	case BLKFRAGET:
544 		if (!arg)
545 			return -EINVAL;
546 		return put_long(arg, (bdev->bd_bdi->ra_pages*PAGE_SIZE) / 512);
547 	case BLKROGET:
548 		return put_int(arg, bdev_read_only(bdev) != 0);
549 	case BLKBSZGET: /* get block device soft block size (cf. BLKSSZGET) */
550 		return put_int(arg, block_size(bdev));
551 	case BLKSSZGET: /* get block device logical block size */
552 		return put_int(arg, bdev_logical_block_size(bdev));
553 	case BLKPBSZGET: /* get block device physical block size */
554 		return put_uint(arg, bdev_physical_block_size(bdev));
555 	case BLKIOMIN:
556 		return put_uint(arg, bdev_io_min(bdev));
557 	case BLKIOOPT:
558 		return put_uint(arg, bdev_io_opt(bdev));
559 	case BLKALIGNOFF:
560 		return put_int(arg, bdev_alignment_offset(bdev));
561 	case BLKDISCARDZEROES:
562 		return put_uint(arg, 0);
563 	case BLKSECTGET:
564 		max_sectors = min_t(unsigned int, USHRT_MAX,
565 				    queue_max_sectors(bdev_get_queue(bdev)));
566 		return put_ushort(arg, max_sectors);
567 	case BLKROTATIONAL:
568 		return put_ushort(arg, !blk_queue_nonrot(bdev_get_queue(bdev)));
569 	case BLKRASET:
570 	case BLKFRASET:
571 		if(!capable(CAP_SYS_ADMIN))
572 			return -EACCES;
573 		bdev->bd_bdi->ra_pages = (arg * 512) / PAGE_SIZE;
574 		return 0;
575 	case BLKBSZSET:
576 		return blkdev_bszset(bdev, mode, argp);
577 	case BLKPG:
578 		return blkpg_ioctl(bdev, argp);
579 	case BLKRRPART:
580 		return blkdev_reread_part(bdev);
581 	case BLKGETSIZE:
582 		size = i_size_read(bdev->bd_inode);
583 		if ((size >> 9) > ~0UL)
584 			return -EFBIG;
585 		return put_ulong(arg, size >> 9);
586 	case BLKGETSIZE64:
587 		return put_u64(arg, i_size_read(bdev->bd_inode));
588 	case BLKTRACESTART:
589 	case BLKTRACESTOP:
590 	case BLKTRACESETUP:
591 	case BLKTRACETEARDOWN:
592 		return blk_trace_ioctl(bdev, cmd, argp);
593 	case IOC_PR_REGISTER:
594 		return blkdev_pr_register(bdev, argp);
595 	case IOC_PR_RESERVE:
596 		return blkdev_pr_reserve(bdev, argp);
597 	case IOC_PR_RELEASE:
598 		return blkdev_pr_release(bdev, argp);
599 	case IOC_PR_PREEMPT:
600 		return blkdev_pr_preempt(bdev, argp, false);
601 	case IOC_PR_PREEMPT_ABORT:
602 		return blkdev_pr_preempt(bdev, argp, true);
603 	case IOC_PR_CLEAR:
604 		return blkdev_pr_clear(bdev, argp);
605 	default:
606 		return __blkdev_driver_ioctl(bdev, mode, cmd, arg);
607 	}
608 }
609 EXPORT_SYMBOL_GPL(blkdev_ioctl);
610