xref: /openbmc/linux/block/ioctl.c (revision 806b5228) 
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 blk_ioctl_discard(struct block_device *bdev, fmode_t mode,
86 		unsigned long arg, unsigned long flags)
87 {
88 	uint64_t range[2];
89 	uint64_t start, len;
90 	struct request_queue *q = bdev_get_queue(bdev);
91 	struct inode *inode = bdev->bd_inode;
92 	int err;
93 
94 	if (!(mode & FMODE_WRITE))
95 		return -EBADF;
96 
97 	if (!blk_queue_discard(q))
98 		return -EOPNOTSUPP;
99 
100 	if (copy_from_user(range, (void __user *)arg, sizeof(range)))
101 		return -EFAULT;
102 
103 	start = range[0];
104 	len = range[1];
105 
106 	if (start & 511)
107 		return -EINVAL;
108 	if (len & 511)
109 		return -EINVAL;
110 
111 	if (start + len > bdev_nr_bytes(bdev))
112 		return -EINVAL;
113 
114 	filemap_invalidate_lock(inode->i_mapping);
115 	err = truncate_bdev_range(bdev, mode, start, start + len - 1);
116 	if (err)
117 		goto fail;
118 
119 	err = blkdev_issue_discard(bdev, start >> 9, len >> 9,
120 				   GFP_KERNEL, flags);
121 
122 fail:
123 	filemap_invalidate_unlock(inode->i_mapping);
124 	return err;
125 }
126 
127 static int blk_ioctl_zeroout(struct block_device *bdev, fmode_t mode,
128 		unsigned long arg)
129 {
130 	uint64_t range[2];
131 	uint64_t start, end, len;
132 	struct inode *inode = bdev->bd_inode;
133 	int err;
134 
135 	if (!(mode & FMODE_WRITE))
136 		return -EBADF;
137 
138 	if (copy_from_user(range, (void __user *)arg, sizeof(range)))
139 		return -EFAULT;
140 
141 	start = range[0];
142 	len = range[1];
143 	end = start + len - 1;
144 
145 	if (start & 511)
146 		return -EINVAL;
147 	if (len & 511)
148 		return -EINVAL;
149 	if (end >= (uint64_t)bdev_nr_bytes(bdev))
150 		return -EINVAL;
151 	if (end < start)
152 		return -EINVAL;
153 
154 	/* Invalidate the page cache, including dirty pages */
155 	filemap_invalidate_lock(inode->i_mapping);
156 	err = truncate_bdev_range(bdev, mode, start, end);
157 	if (err)
158 		goto fail;
159 
160 	err = blkdev_issue_zeroout(bdev, start >> 9, len >> 9, GFP_KERNEL,
161 				   BLKDEV_ZERO_NOUNMAP);
162 
163 fail:
164 	filemap_invalidate_unlock(inode->i_mapping);
165 	return err;
166 }
167 
168 static int put_ushort(unsigned short __user *argp, unsigned short val)
169 {
170 	return put_user(val, argp);
171 }
172 
173 static int put_int(int __user *argp, int val)
174 {
175 	return put_user(val, argp);
176 }
177 
178 static int put_uint(unsigned int __user *argp, unsigned int val)
179 {
180 	return put_user(val, argp);
181 }
182 
183 static int put_long(long __user *argp, long val)
184 {
185 	return put_user(val, argp);
186 }
187 
188 static int put_ulong(unsigned long __user *argp, unsigned long val)
189 {
190 	return put_user(val, argp);
191 }
192 
193 static int put_u64(u64 __user *argp, u64 val)
194 {
195 	return put_user(val, argp);
196 }
197 
198 #ifdef CONFIG_COMPAT
199 static int compat_put_long(compat_long_t __user *argp, long val)
200 {
201 	return put_user(val, argp);
202 }
203 
204 static int compat_put_ulong(compat_ulong_t __user *argp, compat_ulong_t val)
205 {
206 	return put_user(val, argp);
207 }
208 #endif
209 
210 #ifdef CONFIG_COMPAT
211 /*
212  * This is the equivalent of compat_ptr_ioctl(), to be used by block
213  * drivers that implement only commands that are completely compatible
214  * between 32-bit and 64-bit user space
215  */
216 int blkdev_compat_ptr_ioctl(struct block_device *bdev, fmode_t mode,
217 			unsigned cmd, unsigned long arg)
218 {
219 	struct gendisk *disk = bdev->bd_disk;
220 
221 	if (disk->fops->ioctl)
222 		return disk->fops->ioctl(bdev, mode, cmd,
223 					 (unsigned long)compat_ptr(arg));
224 
225 	return -ENOIOCTLCMD;
226 }
227 EXPORT_SYMBOL(blkdev_compat_ptr_ioctl);
228 #endif
229 
230 static int blkdev_pr_register(struct block_device *bdev,
231 		struct pr_registration __user *arg)
232 {
233 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
234 	struct pr_registration reg;
235 
236 	if (!capable(CAP_SYS_ADMIN))
237 		return -EPERM;
238 	if (!ops || !ops->pr_register)
239 		return -EOPNOTSUPP;
240 	if (copy_from_user(&reg, arg, sizeof(reg)))
241 		return -EFAULT;
242 
243 	if (reg.flags & ~PR_FL_IGNORE_KEY)
244 		return -EOPNOTSUPP;
245 	return ops->pr_register(bdev, reg.old_key, reg.new_key, reg.flags);
246 }
247 
248 static int blkdev_pr_reserve(struct block_device *bdev,
249 		struct pr_reservation __user *arg)
250 {
251 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
252 	struct pr_reservation rsv;
253 
254 	if (!capable(CAP_SYS_ADMIN))
255 		return -EPERM;
256 	if (!ops || !ops->pr_reserve)
257 		return -EOPNOTSUPP;
258 	if (copy_from_user(&rsv, arg, sizeof(rsv)))
259 		return -EFAULT;
260 
261 	if (rsv.flags & ~PR_FL_IGNORE_KEY)
262 		return -EOPNOTSUPP;
263 	return ops->pr_reserve(bdev, rsv.key, rsv.type, rsv.flags);
264 }
265 
266 static int blkdev_pr_release(struct block_device *bdev,
267 		struct pr_reservation __user *arg)
268 {
269 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
270 	struct pr_reservation rsv;
271 
272 	if (!capable(CAP_SYS_ADMIN))
273 		return -EPERM;
274 	if (!ops || !ops->pr_release)
275 		return -EOPNOTSUPP;
276 	if (copy_from_user(&rsv, arg, sizeof(rsv)))
277 		return -EFAULT;
278 
279 	if (rsv.flags)
280 		return -EOPNOTSUPP;
281 	return ops->pr_release(bdev, rsv.key, rsv.type);
282 }
283 
284 static int blkdev_pr_preempt(struct block_device *bdev,
285 		struct pr_preempt __user *arg, bool abort)
286 {
287 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
288 	struct pr_preempt p;
289 
290 	if (!capable(CAP_SYS_ADMIN))
291 		return -EPERM;
292 	if (!ops || !ops->pr_preempt)
293 		return -EOPNOTSUPP;
294 	if (copy_from_user(&p, arg, sizeof(p)))
295 		return -EFAULT;
296 
297 	if (p.flags)
298 		return -EOPNOTSUPP;
299 	return ops->pr_preempt(bdev, p.old_key, p.new_key, p.type, abort);
300 }
301 
302 static int blkdev_pr_clear(struct block_device *bdev,
303 		struct pr_clear __user *arg)
304 {
305 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
306 	struct pr_clear c;
307 
308 	if (!capable(CAP_SYS_ADMIN))
309 		return -EPERM;
310 	if (!ops || !ops->pr_clear)
311 		return -EOPNOTSUPP;
312 	if (copy_from_user(&c, arg, sizeof(c)))
313 		return -EFAULT;
314 
315 	if (c.flags)
316 		return -EOPNOTSUPP;
317 	return ops->pr_clear(bdev, c.key);
318 }
319 
320 static int blkdev_flushbuf(struct block_device *bdev, fmode_t mode,
321 		unsigned cmd, unsigned long arg)
322 {
323 	if (!capable(CAP_SYS_ADMIN))
324 		return -EACCES;
325 	fsync_bdev(bdev);
326 	invalidate_bdev(bdev);
327 	return 0;
328 }
329 
330 static int blkdev_roset(struct block_device *bdev, fmode_t mode,
331 		unsigned cmd, unsigned long arg)
332 {
333 	int ret, n;
334 
335 	if (!capable(CAP_SYS_ADMIN))
336 		return -EACCES;
337 
338 	if (get_user(n, (int __user *)arg))
339 		return -EFAULT;
340 	if (bdev->bd_disk->fops->set_read_only) {
341 		ret = bdev->bd_disk->fops->set_read_only(bdev, n);
342 		if (ret)
343 			return ret;
344 	}
345 	bdev->bd_read_only = n;
346 	return 0;
347 }
348 
349 static int blkdev_getgeo(struct block_device *bdev,
350 		struct hd_geometry __user *argp)
351 {
352 	struct gendisk *disk = bdev->bd_disk;
353 	struct hd_geometry geo;
354 	int ret;
355 
356 	if (!argp)
357 		return -EINVAL;
358 	if (!disk->fops->getgeo)
359 		return -ENOTTY;
360 
361 	/*
362 	 * We need to set the startsect first, the driver may
363 	 * want to override it.
364 	 */
365 	memset(&geo, 0, sizeof(geo));
366 	geo.start = get_start_sect(bdev);
367 	ret = disk->fops->getgeo(bdev, &geo);
368 	if (ret)
369 		return ret;
370 	if (copy_to_user(argp, &geo, sizeof(geo)))
371 		return -EFAULT;
372 	return 0;
373 }
374 
375 #ifdef CONFIG_COMPAT
376 struct compat_hd_geometry {
377 	unsigned char heads;
378 	unsigned char sectors;
379 	unsigned short cylinders;
380 	u32 start;
381 };
382 
383 static int compat_hdio_getgeo(struct block_device *bdev,
384 			      struct compat_hd_geometry __user *ugeo)
385 {
386 	struct gendisk *disk = bdev->bd_disk;
387 	struct hd_geometry geo;
388 	int ret;
389 
390 	if (!ugeo)
391 		return -EINVAL;
392 	if (!disk->fops->getgeo)
393 		return -ENOTTY;
394 
395 	memset(&geo, 0, sizeof(geo));
396 	/*
397 	 * We need to set the startsect first, the driver may
398 	 * want to override it.
399 	 */
400 	geo.start = get_start_sect(bdev);
401 	ret = disk->fops->getgeo(bdev, &geo);
402 	if (ret)
403 		return ret;
404 
405 	ret = copy_to_user(ugeo, &geo, 4);
406 	ret |= put_user(geo.start, &ugeo->start);
407 	if (ret)
408 		ret = -EFAULT;
409 
410 	return ret;
411 }
412 #endif
413 
414 /* set the logical block size */
415 static int blkdev_bszset(struct block_device *bdev, fmode_t mode,
416 		int __user *argp)
417 {
418 	int ret, n;
419 
420 	if (!capable(CAP_SYS_ADMIN))
421 		return -EACCES;
422 	if (!argp)
423 		return -EINVAL;
424 	if (get_user(n, argp))
425 		return -EFAULT;
426 
427 	if (mode & FMODE_EXCL)
428 		return set_blocksize(bdev, n);
429 
430 	if (IS_ERR(blkdev_get_by_dev(bdev->bd_dev, mode | FMODE_EXCL, &bdev)))
431 		return -EBUSY;
432 	ret = set_blocksize(bdev, n);
433 	blkdev_put(bdev, mode | FMODE_EXCL);
434 
435 	return ret;
436 }
437 
438 /*
439  * Common commands that are handled the same way on native and compat
440  * user space. Note the separate arg/argp parameters that are needed
441  * to deal with the compat_ptr() conversion.
442  */
443 static int blkdev_common_ioctl(struct block_device *bdev, fmode_t mode,
444 				unsigned cmd, unsigned long arg, void __user *argp)
445 {
446 	unsigned int max_sectors;
447 
448 	switch (cmd) {
449 	case BLKFLSBUF:
450 		return blkdev_flushbuf(bdev, mode, cmd, arg);
451 	case BLKROSET:
452 		return blkdev_roset(bdev, mode, cmd, arg);
453 	case BLKDISCARD:
454 		return blk_ioctl_discard(bdev, mode, arg, 0);
455 	case BLKSECDISCARD:
456 		return blk_ioctl_discard(bdev, mode, arg,
457 				BLKDEV_DISCARD_SECURE);
458 	case BLKZEROOUT:
459 		return blk_ioctl_zeroout(bdev, mode, arg);
460 	case BLKGETDISKSEQ:
461 		return put_u64(argp, bdev->bd_disk->diskseq);
462 	case BLKREPORTZONE:
463 		return blkdev_report_zones_ioctl(bdev, mode, cmd, arg);
464 	case BLKRESETZONE:
465 	case BLKOPENZONE:
466 	case BLKCLOSEZONE:
467 	case BLKFINISHZONE:
468 		return blkdev_zone_mgmt_ioctl(bdev, mode, cmd, arg);
469 	case BLKGETZONESZ:
470 		return put_uint(argp, bdev_zone_sectors(bdev));
471 	case BLKGETNRZONES:
472 		return put_uint(argp, blkdev_nr_zones(bdev->bd_disk));
473 	case BLKROGET:
474 		return put_int(argp, bdev_read_only(bdev) != 0);
475 	case BLKSSZGET: /* get block device logical block size */
476 		return put_int(argp, bdev_logical_block_size(bdev));
477 	case BLKPBSZGET: /* get block device physical block size */
478 		return put_uint(argp, bdev_physical_block_size(bdev));
479 	case BLKIOMIN:
480 		return put_uint(argp, bdev_io_min(bdev));
481 	case BLKIOOPT:
482 		return put_uint(argp, bdev_io_opt(bdev));
483 	case BLKALIGNOFF:
484 		return put_int(argp, bdev_alignment_offset(bdev));
485 	case BLKDISCARDZEROES:
486 		return put_uint(argp, 0);
487 	case BLKSECTGET:
488 		max_sectors = min_t(unsigned int, USHRT_MAX,
489 				    queue_max_sectors(bdev_get_queue(bdev)));
490 		return put_ushort(argp, max_sectors);
491 	case BLKROTATIONAL:
492 		return put_ushort(argp, !blk_queue_nonrot(bdev_get_queue(bdev)));
493 	case BLKRASET:
494 	case BLKFRASET:
495 		if(!capable(CAP_SYS_ADMIN))
496 			return -EACCES;
497 		bdev->bd_disk->bdi->ra_pages = (arg * 512) / PAGE_SIZE;
498 		return 0;
499 	case BLKRRPART:
500 		if (!capable(CAP_SYS_ADMIN))
501 			return -EACCES;
502 		if (bdev_is_partition(bdev))
503 			return -EINVAL;
504 		return disk_scan_partitions(bdev->bd_disk, mode & ~FMODE_EXCL);
505 	case BLKTRACESTART:
506 	case BLKTRACESTOP:
507 	case BLKTRACETEARDOWN:
508 		return blk_trace_ioctl(bdev, cmd, argp);
509 	case IOC_PR_REGISTER:
510 		return blkdev_pr_register(bdev, argp);
511 	case IOC_PR_RESERVE:
512 		return blkdev_pr_reserve(bdev, argp);
513 	case IOC_PR_RELEASE:
514 		return blkdev_pr_release(bdev, argp);
515 	case IOC_PR_PREEMPT:
516 		return blkdev_pr_preempt(bdev, argp, false);
517 	case IOC_PR_PREEMPT_ABORT:
518 		return blkdev_pr_preempt(bdev, argp, true);
519 	case IOC_PR_CLEAR:
520 		return blkdev_pr_clear(bdev, argp);
521 	default:
522 		return -ENOIOCTLCMD;
523 	}
524 }
525 
526 /*
527  * Always keep this in sync with compat_blkdev_ioctl()
528  * to handle all incompatible commands in both functions.
529  *
530  * New commands must be compatible and go into blkdev_common_ioctl
531  */
532 long blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg)
533 {
534 	struct block_device *bdev = I_BDEV(file->f_mapping->host);
535 	void __user *argp = (void __user *)arg;
536 	fmode_t mode = file->f_mode;
537 	int ret;
538 
539 	/*
540 	 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
541 	 * to updated it before every ioctl.
542 	 */
543 	if (file->f_flags & O_NDELAY)
544 		mode |= FMODE_NDELAY;
545 	else
546 		mode &= ~FMODE_NDELAY;
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 		if (bdev_nr_sectors(bdev) > ~0UL)
564 			return -EFBIG;
565 		return put_ulong(argp, bdev_nr_sectors(bdev));
566 
567 	/* The data is compatible, but the command number is different */
568 	case BLKBSZGET: /* get block device soft block size (cf. BLKSSZGET) */
569 		return put_int(argp, block_size(bdev));
570 	case BLKBSZSET:
571 		return blkdev_bszset(bdev, mode, argp);
572 	case BLKGETSIZE64:
573 		return put_u64(argp, bdev_nr_bytes(bdev));
574 
575 	/* Incompatible alignment on i386 */
576 	case BLKTRACESETUP:
577 		return blk_trace_ioctl(bdev, cmd, argp);
578 	default:
579 		break;
580 	}
581 
582 	ret = blkdev_common_ioctl(bdev, mode, cmd, arg, argp);
583 	if (ret != -ENOIOCTLCMD)
584 		return ret;
585 
586 	if (!bdev->bd_disk->fops->ioctl)
587 		return -ENOTTY;
588 	return bdev->bd_disk->fops->ioctl(bdev, mode, cmd, arg);
589 }
590 
591 #ifdef CONFIG_COMPAT
592 
593 #define BLKBSZGET_32		_IOR(0x12, 112, int)
594 #define BLKBSZSET_32		_IOW(0x12, 113, int)
595 #define BLKGETSIZE64_32		_IOR(0x12, 114, int)
596 
597 /* Most of the generic ioctls are handled in the normal fallback path.
598    This assumes the blkdev's low level compat_ioctl always returns
599    ENOIOCTLCMD for unknown ioctls. */
600 long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg)
601 {
602 	int ret;
603 	void __user *argp = compat_ptr(arg);
604 	struct block_device *bdev = I_BDEV(file->f_mapping->host);
605 	struct gendisk *disk = bdev->bd_disk;
606 	fmode_t mode = file->f_mode;
607 
608 	/*
609 	 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
610 	 * to updated it before every ioctl.
611 	 */
612 	if (file->f_flags & O_NDELAY)
613 		mode |= FMODE_NDELAY;
614 	else
615 		mode &= ~FMODE_NDELAY;
616 
617 	switch (cmd) {
618 	/* These need separate implementations for the data structure */
619 	case HDIO_GETGEO:
620 		return compat_hdio_getgeo(bdev, argp);
621 	case BLKPG:
622 		return compat_blkpg_ioctl(bdev, argp);
623 
624 	/* Compat mode returns 32-bit data instead of 'long' */
625 	case BLKRAGET:
626 	case BLKFRAGET:
627 		if (!argp)
628 			return -EINVAL;
629 		return compat_put_long(argp,
630 			(bdev->bd_disk->bdi->ra_pages * PAGE_SIZE) / 512);
631 	case BLKGETSIZE:
632 		if (bdev_nr_sectors(bdev) > ~(compat_ulong_t)0)
633 			return -EFBIG;
634 		return compat_put_ulong(argp, bdev_nr_sectors(bdev));
635 
636 	/* The data is compatible, but the command number is different */
637 	case BLKBSZGET_32: /* get the logical block size (cf. BLKSSZGET) */
638 		return put_int(argp, bdev_logical_block_size(bdev));
639 	case BLKBSZSET_32:
640 		return blkdev_bszset(bdev, mode, argp);
641 	case BLKGETSIZE64_32:
642 		return put_u64(argp, bdev_nr_bytes(bdev));
643 
644 	/* Incompatible alignment on i386 */
645 	case BLKTRACESETUP32:
646 		return blk_trace_ioctl(bdev, cmd, argp);
647 	default:
648 		break;
649 	}
650 
651 	ret = blkdev_common_ioctl(bdev, mode, cmd, arg, argp);
652 	if (ret == -ENOIOCTLCMD && disk->fops->compat_ioctl)
653 		ret = disk->fops->compat_ioctl(bdev, mode, cmd, arg);
654 
655 	return ret;
656 }
657 #endif
658