xref: /openbmc/linux/block/ioctl.c (revision aac5987a)
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 
206 	if (!(mode & FMODE_WRITE))
207 		return -EBADF;
208 
209 	if (copy_from_user(range, (void __user *)arg, sizeof(range)))
210 		return -EFAULT;
211 
212 	start = range[0];
213 	len = range[1];
214 
215 	if (start & 511)
216 		return -EINVAL;
217 	if (len & 511)
218 		return -EINVAL;
219 	start >>= 9;
220 	len >>= 9;
221 
222 	if (start + len > (i_size_read(bdev->bd_inode) >> 9))
223 		return -EINVAL;
224 	return blkdev_issue_discard(bdev, start, len, GFP_KERNEL, flags);
225 }
226 
227 static int blk_ioctl_zeroout(struct block_device *bdev, fmode_t mode,
228 		unsigned long arg)
229 {
230 	uint64_t range[2];
231 	struct address_space *mapping;
232 	uint64_t start, end, len;
233 
234 	if (!(mode & FMODE_WRITE))
235 		return -EBADF;
236 
237 	if (copy_from_user(range, (void __user *)arg, sizeof(range)))
238 		return -EFAULT;
239 
240 	start = range[0];
241 	len = range[1];
242 	end = start + len - 1;
243 
244 	if (start & 511)
245 		return -EINVAL;
246 	if (len & 511)
247 		return -EINVAL;
248 	if (end >= (uint64_t)i_size_read(bdev->bd_inode))
249 		return -EINVAL;
250 	if (end < start)
251 		return -EINVAL;
252 
253 	/* Invalidate the page cache, including dirty pages */
254 	mapping = bdev->bd_inode->i_mapping;
255 	truncate_inode_pages_range(mapping, start, end);
256 
257 	return blkdev_issue_zeroout(bdev, start >> 9, len >> 9, GFP_KERNEL,
258 				    false);
259 }
260 
261 static int put_ushort(unsigned long arg, unsigned short val)
262 {
263 	return put_user(val, (unsigned short __user *)arg);
264 }
265 
266 static int put_int(unsigned long arg, int val)
267 {
268 	return put_user(val, (int __user *)arg);
269 }
270 
271 static int put_uint(unsigned long arg, unsigned int val)
272 {
273 	return put_user(val, (unsigned int __user *)arg);
274 }
275 
276 static int put_long(unsigned long arg, long val)
277 {
278 	return put_user(val, (long __user *)arg);
279 }
280 
281 static int put_ulong(unsigned long arg, unsigned long val)
282 {
283 	return put_user(val, (unsigned long __user *)arg);
284 }
285 
286 static int put_u64(unsigned long arg, u64 val)
287 {
288 	return put_user(val, (u64 __user *)arg);
289 }
290 
291 int __blkdev_driver_ioctl(struct block_device *bdev, fmode_t mode,
292 			unsigned cmd, unsigned long arg)
293 {
294 	struct gendisk *disk = bdev->bd_disk;
295 
296 	if (disk->fops->ioctl)
297 		return disk->fops->ioctl(bdev, mode, cmd, arg);
298 
299 	return -ENOTTY;
300 }
301 /*
302  * For the record: _GPL here is only because somebody decided to slap it
303  * on the previous export.  Sheer idiocy, since it wasn't copyrightable
304  * at all and could be open-coded without any exports by anybody who cares.
305  */
306 EXPORT_SYMBOL_GPL(__blkdev_driver_ioctl);
307 
308 static int blkdev_pr_register(struct block_device *bdev,
309 		struct pr_registration __user *arg)
310 {
311 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
312 	struct pr_registration reg;
313 
314 	if (!capable(CAP_SYS_ADMIN))
315 		return -EPERM;
316 	if (!ops || !ops->pr_register)
317 		return -EOPNOTSUPP;
318 	if (copy_from_user(&reg, arg, sizeof(reg)))
319 		return -EFAULT;
320 
321 	if (reg.flags & ~PR_FL_IGNORE_KEY)
322 		return -EOPNOTSUPP;
323 	return ops->pr_register(bdev, reg.old_key, reg.new_key, reg.flags);
324 }
325 
326 static int blkdev_pr_reserve(struct block_device *bdev,
327 		struct pr_reservation __user *arg)
328 {
329 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
330 	struct pr_reservation rsv;
331 
332 	if (!capable(CAP_SYS_ADMIN))
333 		return -EPERM;
334 	if (!ops || !ops->pr_reserve)
335 		return -EOPNOTSUPP;
336 	if (copy_from_user(&rsv, arg, sizeof(rsv)))
337 		return -EFAULT;
338 
339 	if (rsv.flags & ~PR_FL_IGNORE_KEY)
340 		return -EOPNOTSUPP;
341 	return ops->pr_reserve(bdev, rsv.key, rsv.type, rsv.flags);
342 }
343 
344 static int blkdev_pr_release(struct block_device *bdev,
345 		struct pr_reservation __user *arg)
346 {
347 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
348 	struct pr_reservation rsv;
349 
350 	if (!capable(CAP_SYS_ADMIN))
351 		return -EPERM;
352 	if (!ops || !ops->pr_release)
353 		return -EOPNOTSUPP;
354 	if (copy_from_user(&rsv, arg, sizeof(rsv)))
355 		return -EFAULT;
356 
357 	if (rsv.flags)
358 		return -EOPNOTSUPP;
359 	return ops->pr_release(bdev, rsv.key, rsv.type);
360 }
361 
362 static int blkdev_pr_preempt(struct block_device *bdev,
363 		struct pr_preempt __user *arg, bool abort)
364 {
365 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
366 	struct pr_preempt p;
367 
368 	if (!capable(CAP_SYS_ADMIN))
369 		return -EPERM;
370 	if (!ops || !ops->pr_preempt)
371 		return -EOPNOTSUPP;
372 	if (copy_from_user(&p, arg, sizeof(p)))
373 		return -EFAULT;
374 
375 	if (p.flags)
376 		return -EOPNOTSUPP;
377 	return ops->pr_preempt(bdev, p.old_key, p.new_key, p.type, abort);
378 }
379 
380 static int blkdev_pr_clear(struct block_device *bdev,
381 		struct pr_clear __user *arg)
382 {
383 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
384 	struct pr_clear c;
385 
386 	if (!capable(CAP_SYS_ADMIN))
387 		return -EPERM;
388 	if (!ops || !ops->pr_clear)
389 		return -EOPNOTSUPP;
390 	if (copy_from_user(&c, arg, sizeof(c)))
391 		return -EFAULT;
392 
393 	if (c.flags)
394 		return -EOPNOTSUPP;
395 	return ops->pr_clear(bdev, c.key);
396 }
397 
398 /*
399  * Is it an unrecognized ioctl? The correct returns are either
400  * ENOTTY (final) or ENOIOCTLCMD ("I don't know this one, try a
401  * fallback"). ENOIOCTLCMD gets turned into ENOTTY by the ioctl
402  * code before returning.
403  *
404  * Confused drivers sometimes return EINVAL, which is wrong. It
405  * means "I understood the ioctl command, but the parameters to
406  * it were wrong".
407  *
408  * We should aim to just fix the broken drivers, the EINVAL case
409  * should go away.
410  */
411 static inline int is_unrecognized_ioctl(int ret)
412 {
413 	return	ret == -EINVAL ||
414 		ret == -ENOTTY ||
415 		ret == -ENOIOCTLCMD;
416 }
417 
418 static int blkdev_flushbuf(struct block_device *bdev, fmode_t mode,
419 		unsigned cmd, unsigned long arg)
420 {
421 	int ret;
422 
423 	if (!capable(CAP_SYS_ADMIN))
424 		return -EACCES;
425 
426 	ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
427 	if (!is_unrecognized_ioctl(ret))
428 		return ret;
429 
430 	fsync_bdev(bdev);
431 	invalidate_bdev(bdev);
432 	return 0;
433 }
434 
435 static int blkdev_roset(struct block_device *bdev, fmode_t mode,
436 		unsigned cmd, unsigned long arg)
437 {
438 	int ret, n;
439 
440 	ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
441 	if (!is_unrecognized_ioctl(ret))
442 		return ret;
443 	if (!capable(CAP_SYS_ADMIN))
444 		return -EACCES;
445 	if (get_user(n, (int __user *)arg))
446 		return -EFAULT;
447 	set_device_ro(bdev, n);
448 	return 0;
449 }
450 
451 static int blkdev_getgeo(struct block_device *bdev,
452 		struct hd_geometry __user *argp)
453 {
454 	struct gendisk *disk = bdev->bd_disk;
455 	struct hd_geometry geo;
456 	int ret;
457 
458 	if (!argp)
459 		return -EINVAL;
460 	if (!disk->fops->getgeo)
461 		return -ENOTTY;
462 
463 	/*
464 	 * We need to set the startsect first, the driver may
465 	 * want to override it.
466 	 */
467 	memset(&geo, 0, sizeof(geo));
468 	geo.start = get_start_sect(bdev);
469 	ret = disk->fops->getgeo(bdev, &geo);
470 	if (ret)
471 		return ret;
472 	if (copy_to_user(argp, &geo, sizeof(geo)))
473 		return -EFAULT;
474 	return 0;
475 }
476 
477 /* set the logical block size */
478 static int blkdev_bszset(struct block_device *bdev, fmode_t mode,
479 		int __user *argp)
480 {
481 	int ret, n;
482 
483 	if (!capable(CAP_SYS_ADMIN))
484 		return -EACCES;
485 	if (!argp)
486 		return -EINVAL;
487 	if (get_user(n, argp))
488 		return -EFAULT;
489 
490 	if (!(mode & FMODE_EXCL)) {
491 		bdgrab(bdev);
492 		if (blkdev_get(bdev, mode | FMODE_EXCL, &bdev) < 0)
493 			return -EBUSY;
494 	}
495 
496 	ret = set_blocksize(bdev, n);
497 	if (!(mode & FMODE_EXCL))
498 		blkdev_put(bdev, mode | FMODE_EXCL);
499 	return ret;
500 }
501 
502 /*
503  * always keep this in sync with compat_blkdev_ioctl()
504  */
505 int blkdev_ioctl(struct block_device *bdev, fmode_t mode, unsigned cmd,
506 			unsigned long arg)
507 {
508 	void __user *argp = (void __user *)arg;
509 	loff_t size;
510 	unsigned int max_sectors;
511 
512 	switch (cmd) {
513 	case BLKFLSBUF:
514 		return blkdev_flushbuf(bdev, mode, cmd, arg);
515 	case BLKROSET:
516 		return blkdev_roset(bdev, mode, cmd, arg);
517 	case BLKDISCARD:
518 		return blk_ioctl_discard(bdev, mode, arg, 0);
519 	case BLKSECDISCARD:
520 		return blk_ioctl_discard(bdev, mode, arg,
521 				BLKDEV_DISCARD_SECURE);
522 	case BLKZEROOUT:
523 		return blk_ioctl_zeroout(bdev, mode, arg);
524 	case BLKREPORTZONE:
525 		return blkdev_report_zones_ioctl(bdev, mode, cmd, arg);
526 	case BLKRESETZONE:
527 		return blkdev_reset_zones_ioctl(bdev, mode, cmd, arg);
528 	case HDIO_GETGEO:
529 		return blkdev_getgeo(bdev, argp);
530 	case BLKRAGET:
531 	case BLKFRAGET:
532 		if (!arg)
533 			return -EINVAL;
534 		return put_long(arg, (bdev->bd_bdi->ra_pages*PAGE_SIZE) / 512);
535 	case BLKROGET:
536 		return put_int(arg, bdev_read_only(bdev) != 0);
537 	case BLKBSZGET: /* get block device soft block size (cf. BLKSSZGET) */
538 		return put_int(arg, block_size(bdev));
539 	case BLKSSZGET: /* get block device logical block size */
540 		return put_int(arg, bdev_logical_block_size(bdev));
541 	case BLKPBSZGET: /* get block device physical block size */
542 		return put_uint(arg, bdev_physical_block_size(bdev));
543 	case BLKIOMIN:
544 		return put_uint(arg, bdev_io_min(bdev));
545 	case BLKIOOPT:
546 		return put_uint(arg, bdev_io_opt(bdev));
547 	case BLKALIGNOFF:
548 		return put_int(arg, bdev_alignment_offset(bdev));
549 	case BLKDISCARDZEROES:
550 		return put_uint(arg, bdev_discard_zeroes_data(bdev));
551 	case BLKSECTGET:
552 		max_sectors = min_t(unsigned int, USHRT_MAX,
553 				    queue_max_sectors(bdev_get_queue(bdev)));
554 		return put_ushort(arg, max_sectors);
555 	case BLKROTATIONAL:
556 		return put_ushort(arg, !blk_queue_nonrot(bdev_get_queue(bdev)));
557 	case BLKRASET:
558 	case BLKFRASET:
559 		if(!capable(CAP_SYS_ADMIN))
560 			return -EACCES;
561 		bdev->bd_bdi->ra_pages = (arg * 512) / PAGE_SIZE;
562 		return 0;
563 	case BLKBSZSET:
564 		return blkdev_bszset(bdev, mode, argp);
565 	case BLKPG:
566 		return blkpg_ioctl(bdev, argp);
567 	case BLKRRPART:
568 		return blkdev_reread_part(bdev);
569 	case BLKGETSIZE:
570 		size = i_size_read(bdev->bd_inode);
571 		if ((size >> 9) > ~0UL)
572 			return -EFBIG;
573 		return put_ulong(arg, size >> 9);
574 	case BLKGETSIZE64:
575 		return put_u64(arg, i_size_read(bdev->bd_inode));
576 	case BLKTRACESTART:
577 	case BLKTRACESTOP:
578 	case BLKTRACESETUP:
579 	case BLKTRACETEARDOWN:
580 		return blk_trace_ioctl(bdev, cmd, argp);
581 	case IOC_PR_REGISTER:
582 		return blkdev_pr_register(bdev, argp);
583 	case IOC_PR_RESERVE:
584 		return blkdev_pr_reserve(bdev, argp);
585 	case IOC_PR_RELEASE:
586 		return blkdev_pr_release(bdev, argp);
587 	case IOC_PR_PREEMPT:
588 		return blkdev_pr_preempt(bdev, argp, false);
589 	case IOC_PR_PREEMPT_ABORT:
590 		return blkdev_pr_preempt(bdev, argp, true);
591 	case IOC_PR_CLEAR:
592 		return blkdev_pr_clear(bdev, argp);
593 	default:
594 		return __blkdev_driver_ioctl(bdev, mode, cmd, arg);
595 	}
596 }
597 EXPORT_SYMBOL_GPL(blkdev_ioctl);
598