xref: /openbmc/linux/block/ioctl.c (revision 97e6ea6d) 
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 #include "blk.h"
15 
16 static int blkpg_do_ioctl(struct block_device *bdev,
17 			  struct blkpg_partition __user *upart, int op)
18 {
19 	struct gendisk *disk = bdev->bd_disk;
20 	struct blkpg_partition p;
21 	long long start, length;
22 
23 	if (!capable(CAP_SYS_ADMIN))
24 		return -EACCES;
25 	if (copy_from_user(&p, upart, sizeof(struct blkpg_partition)))
26 		return -EFAULT;
27 	if (bdev_is_partition(bdev))
28 		return -EINVAL;
29 
30 	if (p.pno <= 0)
31 		return -EINVAL;
32 
33 	if (op == BLKPG_DEL_PARTITION)
34 		return bdev_del_partition(disk, p.pno);
35 
36 	start = p.start >> SECTOR_SHIFT;
37 	length = p.length >> SECTOR_SHIFT;
38 
39 	switch (op) {
40 	case BLKPG_ADD_PARTITION:
41 		/* check if partition is aligned to blocksize */
42 		if (p.start & (bdev_logical_block_size(bdev) - 1))
43 			return -EINVAL;
44 		return bdev_add_partition(disk, p.pno, start, length);
45 	case BLKPG_RESIZE_PARTITION:
46 		return bdev_resize_partition(disk, p.pno, start, length);
47 	default:
48 		return -EINVAL;
49 	}
50 }
51 
52 static int blkpg_ioctl(struct block_device *bdev,
53 		       struct blkpg_ioctl_arg __user *arg)
54 {
55 	struct blkpg_partition __user *udata;
56 	int op;
57 
58 	if (get_user(op, &arg->op) || get_user(udata, &arg->data))
59 		return -EFAULT;
60 
61 	return blkpg_do_ioctl(bdev, udata, op);
62 }
63 
64 #ifdef CONFIG_COMPAT
65 struct compat_blkpg_ioctl_arg {
66 	compat_int_t op;
67 	compat_int_t flags;
68 	compat_int_t datalen;
69 	compat_caddr_t data;
70 };
71 
72 static int compat_blkpg_ioctl(struct block_device *bdev,
73 			      struct compat_blkpg_ioctl_arg __user *arg)
74 {
75 	compat_caddr_t udata;
76 	int op;
77 
78 	if (get_user(op, &arg->op) || get_user(udata, &arg->data))
79 		return -EFAULT;
80 
81 	return blkpg_do_ioctl(bdev, compat_ptr(udata), op);
82 }
83 #endif
84 
85 static int blkdev_reread_part(struct block_device *bdev, fmode_t mode)
86 {
87 	struct block_device *tmp;
88 
89 	if (!disk_part_scan_enabled(bdev->bd_disk) || bdev_is_partition(bdev))
90 		return -EINVAL;
91 	if (!capable(CAP_SYS_ADMIN))
92 		return -EACCES;
93 	if (bdev->bd_disk->open_partitions)
94 		return -EBUSY;
95 
96 	/*
97 	 * Reopen the device to revalidate the driver state and force a
98 	 * partition rescan.
99 	 */
100 	mode &= ~FMODE_EXCL;
101 	set_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
102 
103 	tmp = blkdev_get_by_dev(bdev->bd_dev, mode, NULL);
104 	if (IS_ERR(tmp))
105 		return PTR_ERR(tmp);
106 	blkdev_put(tmp, mode);
107 	return 0;
108 }
109 
110 static int blk_ioctl_discard(struct block_device *bdev, fmode_t mode,
111 		unsigned long arg, unsigned long flags)
112 {
113 	uint64_t range[2];
114 	uint64_t start, len;
115 	struct request_queue *q = bdev_get_queue(bdev);
116 	int err;
117 
118 	if (!(mode & FMODE_WRITE))
119 		return -EBADF;
120 
121 	if (!blk_queue_discard(q))
122 		return -EOPNOTSUPP;
123 
124 	if (copy_from_user(range, (void __user *)arg, sizeof(range)))
125 		return -EFAULT;
126 
127 	start = range[0];
128 	len = range[1];
129 
130 	if (start & 511)
131 		return -EINVAL;
132 	if (len & 511)
133 		return -EINVAL;
134 
135 	if (start + len > i_size_read(bdev->bd_inode))
136 		return -EINVAL;
137 
138 	err = truncate_bdev_range(bdev, mode, start, start + len - 1);
139 	if (err)
140 		return err;
141 
142 	return blkdev_issue_discard(bdev, start >> 9, len >> 9,
143 				    GFP_KERNEL, flags);
144 }
145 
146 static int blk_ioctl_zeroout(struct block_device *bdev, fmode_t mode,
147 		unsigned long arg)
148 {
149 	uint64_t range[2];
150 	uint64_t start, end, len;
151 	int err;
152 
153 	if (!(mode & FMODE_WRITE))
154 		return -EBADF;
155 
156 	if (copy_from_user(range, (void __user *)arg, sizeof(range)))
157 		return -EFAULT;
158 
159 	start = range[0];
160 	len = range[1];
161 	end = start + len - 1;
162 
163 	if (start & 511)
164 		return -EINVAL;
165 	if (len & 511)
166 		return -EINVAL;
167 	if (end >= (uint64_t)i_size_read(bdev->bd_inode))
168 		return -EINVAL;
169 	if (end < start)
170 		return -EINVAL;
171 
172 	/* Invalidate the page cache, including dirty pages */
173 	err = truncate_bdev_range(bdev, mode, start, end);
174 	if (err)
175 		return err;
176 
177 	return blkdev_issue_zeroout(bdev, start >> 9, len >> 9, GFP_KERNEL,
178 			BLKDEV_ZERO_NOUNMAP);
179 }
180 
181 static int put_ushort(unsigned short __user *argp, unsigned short val)
182 {
183 	return put_user(val, argp);
184 }
185 
186 static int put_int(int __user *argp, int val)
187 {
188 	return put_user(val, argp);
189 }
190 
191 static int put_uint(unsigned int __user *argp, unsigned int val)
192 {
193 	return put_user(val, argp);
194 }
195 
196 static int put_long(long __user *argp, long val)
197 {
198 	return put_user(val, argp);
199 }
200 
201 static int put_ulong(unsigned long __user *argp, unsigned long val)
202 {
203 	return put_user(val, argp);
204 }
205 
206 static int put_u64(u64 __user *argp, u64 val)
207 {
208 	return put_user(val, argp);
209 }
210 
211 #ifdef CONFIG_COMPAT
212 static int compat_put_long(compat_long_t __user *argp, long val)
213 {
214 	return put_user(val, argp);
215 }
216 
217 static int compat_put_ulong(compat_ulong_t __user *argp, compat_ulong_t val)
218 {
219 	return put_user(val, argp);
220 }
221 #endif
222 
223 #ifdef CONFIG_COMPAT
224 /*
225  * This is the equivalent of compat_ptr_ioctl(), to be used by block
226  * drivers that implement only commands that are completely compatible
227  * between 32-bit and 64-bit user space
228  */
229 int blkdev_compat_ptr_ioctl(struct block_device *bdev, fmode_t mode,
230 			unsigned cmd, unsigned long arg)
231 {
232 	struct gendisk *disk = bdev->bd_disk;
233 
234 	if (disk->fops->ioctl)
235 		return disk->fops->ioctl(bdev, mode, cmd,
236 					 (unsigned long)compat_ptr(arg));
237 
238 	return -ENOIOCTLCMD;
239 }
240 EXPORT_SYMBOL(blkdev_compat_ptr_ioctl);
241 #endif
242 
243 static int blkdev_pr_register(struct block_device *bdev,
244 		struct pr_registration __user *arg)
245 {
246 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
247 	struct pr_registration reg;
248 
249 	if (!capable(CAP_SYS_ADMIN))
250 		return -EPERM;
251 	if (!ops || !ops->pr_register)
252 		return -EOPNOTSUPP;
253 	if (copy_from_user(&reg, arg, sizeof(reg)))
254 		return -EFAULT;
255 
256 	if (reg.flags & ~PR_FL_IGNORE_KEY)
257 		return -EOPNOTSUPP;
258 	return ops->pr_register(bdev, reg.old_key, reg.new_key, reg.flags);
259 }
260 
261 static int blkdev_pr_reserve(struct block_device *bdev,
262 		struct pr_reservation __user *arg)
263 {
264 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
265 	struct pr_reservation rsv;
266 
267 	if (!capable(CAP_SYS_ADMIN))
268 		return -EPERM;
269 	if (!ops || !ops->pr_reserve)
270 		return -EOPNOTSUPP;
271 	if (copy_from_user(&rsv, arg, sizeof(rsv)))
272 		return -EFAULT;
273 
274 	if (rsv.flags & ~PR_FL_IGNORE_KEY)
275 		return -EOPNOTSUPP;
276 	return ops->pr_reserve(bdev, rsv.key, rsv.type, rsv.flags);
277 }
278 
279 static int blkdev_pr_release(struct block_device *bdev,
280 		struct pr_reservation __user *arg)
281 {
282 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
283 	struct pr_reservation rsv;
284 
285 	if (!capable(CAP_SYS_ADMIN))
286 		return -EPERM;
287 	if (!ops || !ops->pr_release)
288 		return -EOPNOTSUPP;
289 	if (copy_from_user(&rsv, arg, sizeof(rsv)))
290 		return -EFAULT;
291 
292 	if (rsv.flags)
293 		return -EOPNOTSUPP;
294 	return ops->pr_release(bdev, rsv.key, rsv.type);
295 }
296 
297 static int blkdev_pr_preempt(struct block_device *bdev,
298 		struct pr_preempt __user *arg, bool abort)
299 {
300 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
301 	struct pr_preempt p;
302 
303 	if (!capable(CAP_SYS_ADMIN))
304 		return -EPERM;
305 	if (!ops || !ops->pr_preempt)
306 		return -EOPNOTSUPP;
307 	if (copy_from_user(&p, arg, sizeof(p)))
308 		return -EFAULT;
309 
310 	if (p.flags)
311 		return -EOPNOTSUPP;
312 	return ops->pr_preempt(bdev, p.old_key, p.new_key, p.type, abort);
313 }
314 
315 static int blkdev_pr_clear(struct block_device *bdev,
316 		struct pr_clear __user *arg)
317 {
318 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
319 	struct pr_clear c;
320 
321 	if (!capable(CAP_SYS_ADMIN))
322 		return -EPERM;
323 	if (!ops || !ops->pr_clear)
324 		return -EOPNOTSUPP;
325 	if (copy_from_user(&c, arg, sizeof(c)))
326 		return -EFAULT;
327 
328 	if (c.flags)
329 		return -EOPNOTSUPP;
330 	return ops->pr_clear(bdev, c.key);
331 }
332 
333 static int blkdev_flushbuf(struct block_device *bdev, fmode_t mode,
334 		unsigned cmd, unsigned long arg)
335 {
336 	if (!capable(CAP_SYS_ADMIN))
337 		return -EACCES;
338 	fsync_bdev(bdev);
339 	invalidate_bdev(bdev);
340 	return 0;
341 }
342 
343 static int blkdev_roset(struct block_device *bdev, fmode_t mode,
344 		unsigned cmd, unsigned long arg)
345 {
346 	int ret, n;
347 
348 	if (!capable(CAP_SYS_ADMIN))
349 		return -EACCES;
350 
351 	if (get_user(n, (int __user *)arg))
352 		return -EFAULT;
353 	if (bdev->bd_disk->fops->set_read_only) {
354 		ret = bdev->bd_disk->fops->set_read_only(bdev, n);
355 		if (ret)
356 			return ret;
357 	}
358 	bdev->bd_read_only = n;
359 	return 0;
360 }
361 
362 static int blkdev_getgeo(struct block_device *bdev,
363 		struct hd_geometry __user *argp)
364 {
365 	struct gendisk *disk = bdev->bd_disk;
366 	struct hd_geometry geo;
367 	int ret;
368 
369 	if (!argp)
370 		return -EINVAL;
371 	if (!disk->fops->getgeo)
372 		return -ENOTTY;
373 
374 	/*
375 	 * We need to set the startsect first, the driver may
376 	 * want to override it.
377 	 */
378 	memset(&geo, 0, sizeof(geo));
379 	geo.start = get_start_sect(bdev);
380 	ret = disk->fops->getgeo(bdev, &geo);
381 	if (ret)
382 		return ret;
383 	if (copy_to_user(argp, &geo, sizeof(geo)))
384 		return -EFAULT;
385 	return 0;
386 }
387 
388 #ifdef CONFIG_COMPAT
389 struct compat_hd_geometry {
390 	unsigned char heads;
391 	unsigned char sectors;
392 	unsigned short cylinders;
393 	u32 start;
394 };
395 
396 static int compat_hdio_getgeo(struct block_device *bdev,
397 			      struct compat_hd_geometry __user *ugeo)
398 {
399 	struct gendisk *disk = bdev->bd_disk;
400 	struct hd_geometry geo;
401 	int ret;
402 
403 	if (!ugeo)
404 		return -EINVAL;
405 	if (!disk->fops->getgeo)
406 		return -ENOTTY;
407 
408 	memset(&geo, 0, sizeof(geo));
409 	/*
410 	 * We need to set the startsect first, the driver may
411 	 * want to override it.
412 	 */
413 	geo.start = get_start_sect(bdev);
414 	ret = disk->fops->getgeo(bdev, &geo);
415 	if (ret)
416 		return ret;
417 
418 	ret = copy_to_user(ugeo, &geo, 4);
419 	ret |= put_user(geo.start, &ugeo->start);
420 	if (ret)
421 		ret = -EFAULT;
422 
423 	return ret;
424 }
425 #endif
426 
427 /* set the logical block size */
428 static int blkdev_bszset(struct block_device *bdev, fmode_t mode,
429 		int __user *argp)
430 {
431 	int ret, n;
432 
433 	if (!capable(CAP_SYS_ADMIN))
434 		return -EACCES;
435 	if (!argp)
436 		return -EINVAL;
437 	if (get_user(n, argp))
438 		return -EFAULT;
439 
440 	if (mode & FMODE_EXCL)
441 		return set_blocksize(bdev, n);
442 
443 	if (IS_ERR(blkdev_get_by_dev(bdev->bd_dev, mode | FMODE_EXCL, &bdev)))
444 		return -EBUSY;
445 	ret = set_blocksize(bdev, n);
446 	blkdev_put(bdev, mode | FMODE_EXCL);
447 
448 	return ret;
449 }
450 
451 /*
452  * Common commands that are handled the same way on native and compat
453  * user space. Note the separate arg/argp parameters that are needed
454  * to deal with the compat_ptr() conversion.
455  */
456 static int blkdev_common_ioctl(struct block_device *bdev, fmode_t mode,
457 				unsigned cmd, unsigned long arg, void __user *argp)
458 {
459 	unsigned int max_sectors;
460 
461 	switch (cmd) {
462 	case BLKFLSBUF:
463 		return blkdev_flushbuf(bdev, mode, cmd, arg);
464 	case BLKROSET:
465 		return blkdev_roset(bdev, mode, cmd, arg);
466 	case BLKDISCARD:
467 		return blk_ioctl_discard(bdev, mode, arg, 0);
468 	case BLKSECDISCARD:
469 		return blk_ioctl_discard(bdev, mode, arg,
470 				BLKDEV_DISCARD_SECURE);
471 	case BLKZEROOUT:
472 		return blk_ioctl_zeroout(bdev, mode, arg);
473 	case BLKGETDISKSEQ:
474 		return put_u64(argp, bdev->bd_disk->diskseq);
475 	case BLKREPORTZONE:
476 		return blkdev_report_zones_ioctl(bdev, mode, cmd, arg);
477 	case BLKRESETZONE:
478 	case BLKOPENZONE:
479 	case BLKCLOSEZONE:
480 	case BLKFINISHZONE:
481 		return blkdev_zone_mgmt_ioctl(bdev, mode, cmd, arg);
482 	case BLKGETZONESZ:
483 		return put_uint(argp, bdev_zone_sectors(bdev));
484 	case BLKGETNRZONES:
485 		return put_uint(argp, blkdev_nr_zones(bdev->bd_disk));
486 	case BLKROGET:
487 		return put_int(argp, bdev_read_only(bdev) != 0);
488 	case BLKSSZGET: /* get block device logical block size */
489 		return put_int(argp, bdev_logical_block_size(bdev));
490 	case BLKPBSZGET: /* get block device physical block size */
491 		return put_uint(argp, bdev_physical_block_size(bdev));
492 	case BLKIOMIN:
493 		return put_uint(argp, bdev_io_min(bdev));
494 	case BLKIOOPT:
495 		return put_uint(argp, bdev_io_opt(bdev));
496 	case BLKALIGNOFF:
497 		return put_int(argp, bdev_alignment_offset(bdev));
498 	case BLKDISCARDZEROES:
499 		return put_uint(argp, 0);
500 	case BLKSECTGET:
501 		max_sectors = min_t(unsigned int, USHRT_MAX,
502 				    queue_max_sectors(bdev_get_queue(bdev)));
503 		return put_ushort(argp, max_sectors);
504 	case BLKROTATIONAL:
505 		return put_ushort(argp, !blk_queue_nonrot(bdev_get_queue(bdev)));
506 	case BLKRASET:
507 	case BLKFRASET:
508 		if(!capable(CAP_SYS_ADMIN))
509 			return -EACCES;
510 		bdev->bd_disk->bdi->ra_pages = (arg * 512) / PAGE_SIZE;
511 		return 0;
512 	case BLKRRPART:
513 		return blkdev_reread_part(bdev, mode);
514 	case BLKTRACESTART:
515 	case BLKTRACESTOP:
516 	case BLKTRACETEARDOWN:
517 		return blk_trace_ioctl(bdev, cmd, argp);
518 	case IOC_PR_REGISTER:
519 		return blkdev_pr_register(bdev, argp);
520 	case IOC_PR_RESERVE:
521 		return blkdev_pr_reserve(bdev, argp);
522 	case IOC_PR_RELEASE:
523 		return blkdev_pr_release(bdev, argp);
524 	case IOC_PR_PREEMPT:
525 		return blkdev_pr_preempt(bdev, argp, false);
526 	case IOC_PR_PREEMPT_ABORT:
527 		return blkdev_pr_preempt(bdev, argp, true);
528 	case IOC_PR_CLEAR:
529 		return blkdev_pr_clear(bdev, argp);
530 	default:
531 		return -ENOIOCTLCMD;
532 	}
533 }
534 
535 /*
536  * Always keep this in sync with compat_blkdev_ioctl()
537  * to handle all incompatible commands in both functions.
538  *
539  * New commands must be compatible and go into blkdev_common_ioctl
540  */
541 int blkdev_ioctl(struct block_device *bdev, fmode_t mode, unsigned cmd,
542 			unsigned long arg)
543 {
544 	int ret;
545 	loff_t size;
546 	void __user *argp = (void __user *)arg;
547 
548 	switch (cmd) {
549 	/* These need separate implementations for the data structure */
550 	case HDIO_GETGEO:
551 		return blkdev_getgeo(bdev, argp);
552 	case BLKPG:
553 		return blkpg_ioctl(bdev, argp);
554 
555 	/* Compat mode returns 32-bit data instead of 'long' */
556 	case BLKRAGET:
557 	case BLKFRAGET:
558 		if (!argp)
559 			return -EINVAL;
560 		return put_long(argp,
561 			(bdev->bd_disk->bdi->ra_pages * PAGE_SIZE) / 512);
562 	case BLKGETSIZE:
563 		size = i_size_read(bdev->bd_inode);
564 		if ((size >> 9) > ~0UL)
565 			return -EFBIG;
566 		return put_ulong(argp, size >> 9);
567 
568 	/* The data is compatible, but the command number is different */
569 	case BLKBSZGET: /* get block device soft block size (cf. BLKSSZGET) */
570 		return put_int(argp, block_size(bdev));
571 	case BLKBSZSET:
572 		return blkdev_bszset(bdev, mode, argp);
573 	case BLKGETSIZE64:
574 		return put_u64(argp, i_size_read(bdev->bd_inode));
575 
576 	/* Incompatible alignment on i386 */
577 	case BLKTRACESETUP:
578 		return blk_trace_ioctl(bdev, cmd, argp);
579 	default:
580 		break;
581 	}
582 
583 	ret = blkdev_common_ioctl(bdev, mode, cmd, arg, argp);
584 	if (ret != -ENOIOCTLCMD)
585 		return ret;
586 
587 	if (!bdev->bd_disk->fops->ioctl)
588 		return -ENOTTY;
589 	return bdev->bd_disk->fops->ioctl(bdev, mode, cmd, arg);
590 }
591 EXPORT_SYMBOL_GPL(blkdev_ioctl); /* for /dev/raw */
592 
593 #ifdef CONFIG_COMPAT
594 
595 #define BLKBSZGET_32		_IOR(0x12, 112, int)
596 #define BLKBSZSET_32		_IOW(0x12, 113, int)
597 #define BLKGETSIZE64_32		_IOR(0x12, 114, int)
598 
599 /* Most of the generic ioctls are handled in the normal fallback path.
600    This assumes the blkdev's low level compat_ioctl always returns
601    ENOIOCTLCMD for unknown ioctls. */
602 long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg)
603 {
604 	int ret;
605 	void __user *argp = compat_ptr(arg);
606 	struct block_device *bdev = I_BDEV(file->f_mapping->host);
607 	struct gendisk *disk = bdev->bd_disk;
608 	fmode_t mode = file->f_mode;
609 	loff_t size;
610 
611 	/*
612 	 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
613 	 * to updated it before every ioctl.
614 	 */
615 	if (file->f_flags & O_NDELAY)
616 		mode |= FMODE_NDELAY;
617 	else
618 		mode &= ~FMODE_NDELAY;
619 
620 	switch (cmd) {
621 	/* These need separate implementations for the data structure */
622 	case HDIO_GETGEO:
623 		return compat_hdio_getgeo(bdev, argp);
624 	case BLKPG:
625 		return compat_blkpg_ioctl(bdev, argp);
626 
627 	/* Compat mode returns 32-bit data instead of 'long' */
628 	case BLKRAGET:
629 	case BLKFRAGET:
630 		if (!argp)
631 			return -EINVAL;
632 		return compat_put_long(argp,
633 			(bdev->bd_disk->bdi->ra_pages * PAGE_SIZE) / 512);
634 	case BLKGETSIZE:
635 		size = i_size_read(bdev->bd_inode);
636 		if ((size >> 9) > ~0UL)
637 			return -EFBIG;
638 		return compat_put_ulong(argp, size >> 9);
639 
640 	/* The data is compatible, but the command number is different */
641 	case BLKBSZGET_32: /* get the logical block size (cf. BLKSSZGET) */
642 		return put_int(argp, bdev_logical_block_size(bdev));
643 	case BLKBSZSET_32:
644 		return blkdev_bszset(bdev, mode, argp);
645 	case BLKGETSIZE64_32:
646 		return put_u64(argp, i_size_read(bdev->bd_inode));
647 
648 	/* Incompatible alignment on i386 */
649 	case BLKTRACESETUP32:
650 		return blk_trace_ioctl(bdev, cmd, argp);
651 	default:
652 		break;
653 	}
654 
655 	ret = blkdev_common_ioctl(bdev, mode, cmd, arg, argp);
656 	if (ret == -ENOIOCTLCMD && disk->fops->compat_ioctl)
657 		ret = disk->fops->compat_ioctl(bdev, mode, cmd, arg);
658 
659 	return ret;
660 }
661 #endif
662