xref: /linux/block/ioctl.c (revision 46f2dd5ce5723a2c07051d332f8f1c4c4ce548f3)
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 <linux/pagemap.h>
15 #include <linux/io_uring/cmd.h>
16 #include <uapi/linux/blkdev.h>
17 #include "blk.h"
18 
19 static int blkpg_do_ioctl(struct block_device *bdev,
20 			  struct blkpg_partition __user *upart, int op)
21 {
22 	struct gendisk *disk = bdev->bd_disk;
23 	struct blkpg_partition p;
24 	sector_t start, length, capacity, end;
25 
26 	if (!capable(CAP_SYS_ADMIN))
27 		return -EACCES;
28 	if (copy_from_user(&p, upart, sizeof(struct blkpg_partition)))
29 		return -EFAULT;
30 	if (bdev_is_partition(bdev))
31 		return -EINVAL;
32 
33 	if (p.pno <= 0)
34 		return -EINVAL;
35 
36 	if (op == BLKPG_DEL_PARTITION)
37 		return bdev_del_partition(disk, p.pno);
38 
39 	if (p.start < 0 || p.length <= 0 || LLONG_MAX - p.length < p.start)
40 		return -EINVAL;
41 	/* Check that the partition is aligned to the block size */
42 	if (!IS_ALIGNED(p.start | p.length, bdev_logical_block_size(bdev)))
43 		return -EINVAL;
44 
45 	start = p.start >> SECTOR_SHIFT;
46 	length = p.length >> SECTOR_SHIFT;
47 	capacity = get_capacity(disk);
48 
49 	if (check_add_overflow(start, length, &end))
50 		return -EINVAL;
51 
52 	if (start >= capacity || end > capacity)
53 		return -EINVAL;
54 
55 	switch (op) {
56 	case BLKPG_ADD_PARTITION:
57 		return bdev_add_partition(disk, p.pno, start, length);
58 	case BLKPG_RESIZE_PARTITION:
59 		return bdev_resize_partition(disk, p.pno, start, length);
60 	default:
61 		return -EINVAL;
62 	}
63 }
64 
65 static int blkpg_ioctl(struct block_device *bdev,
66 		       struct blkpg_ioctl_arg __user *arg)
67 {
68 	struct blkpg_partition __user *udata;
69 	int op;
70 
71 	if (get_user(op, &arg->op) || get_user(udata, &arg->data))
72 		return -EFAULT;
73 
74 	return blkpg_do_ioctl(bdev, udata, op);
75 }
76 
77 #ifdef CONFIG_COMPAT
78 struct compat_blkpg_ioctl_arg {
79 	compat_int_t op;
80 	compat_int_t flags;
81 	compat_int_t datalen;
82 	compat_caddr_t data;
83 };
84 
85 static int compat_blkpg_ioctl(struct block_device *bdev,
86 			      struct compat_blkpg_ioctl_arg __user *arg)
87 {
88 	compat_caddr_t udata;
89 	int op;
90 
91 	if (get_user(op, &arg->op) || get_user(udata, &arg->data))
92 		return -EFAULT;
93 
94 	return blkpg_do_ioctl(bdev, compat_ptr(udata), op);
95 }
96 #endif
97 
98 /*
99  * Check that [start, start + len) is a valid range from the block device's
100  * perspective, including verifying that it can be correctly translated into
101  * logical block addresses.
102  */
103 static int blk_validate_byte_range(struct block_device *bdev,
104 				   uint64_t start, uint64_t len)
105 {
106 	unsigned int bs_mask = bdev_logical_block_size(bdev) - 1;
107 	uint64_t end;
108 
109 	if ((start | len) & bs_mask)
110 		return -EINVAL;
111 	if (!len)
112 		return -EINVAL;
113 	if (check_add_overflow(start, len, &end) || end > bdev_nr_bytes(bdev))
114 		return -EINVAL;
115 
116 	return 0;
117 }
118 
119 static int blk_ioctl_discard(struct block_device *bdev, blk_mode_t mode,
120 		unsigned long arg)
121 {
122 	uint64_t range[2], start, len;
123 	struct bio *prev = NULL, *bio;
124 	sector_t sector, nr_sects;
125 	struct blk_plug plug;
126 	int err;
127 
128 	if (copy_from_user(range, (void __user *)arg, sizeof(range)))
129 		return -EFAULT;
130 	start = range[0];
131 	len = range[1];
132 
133 	if (!bdev_max_discard_sectors(bdev))
134 		return -EOPNOTSUPP;
135 
136 	if (!(mode & BLK_OPEN_WRITE))
137 		return -EBADF;
138 	if (bdev_read_only(bdev))
139 		return -EPERM;
140 	err = blk_validate_byte_range(bdev, start, len);
141 	if (err)
142 		return err;
143 
144 	filemap_invalidate_lock(bdev->bd_mapping);
145 	err = truncate_bdev_range(bdev, mode, start, start + len - 1);
146 	if (err)
147 		goto fail;
148 
149 	sector = start >> SECTOR_SHIFT;
150 	nr_sects = len >> SECTOR_SHIFT;
151 
152 	blk_start_plug(&plug);
153 	while (1) {
154 		if (fatal_signal_pending(current)) {
155 			if (prev)
156 				bio_await_chain(prev);
157 			err = -EINTR;
158 			goto out_unplug;
159 		}
160 		bio = blk_alloc_discard_bio(bdev, &sector, &nr_sects,
161 				GFP_KERNEL);
162 		if (!bio)
163 			break;
164 		prev = bio_chain_and_submit(prev, bio);
165 	}
166 	if (prev) {
167 		err = submit_bio_wait(prev);
168 		if (err == -EOPNOTSUPP)
169 			err = 0;
170 		bio_put(prev);
171 	}
172 out_unplug:
173 	blk_finish_plug(&plug);
174 fail:
175 	filemap_invalidate_unlock(bdev->bd_mapping);
176 	return err;
177 }
178 
179 static int blk_ioctl_secure_erase(struct block_device *bdev, blk_mode_t mode,
180 		void __user *argp)
181 {
182 	uint64_t start, len, end;
183 	uint64_t range[2];
184 	int err;
185 
186 	if (!(mode & BLK_OPEN_WRITE))
187 		return -EBADF;
188 	if (!bdev_max_secure_erase_sectors(bdev))
189 		return -EOPNOTSUPP;
190 	if (copy_from_user(range, argp, sizeof(range)))
191 		return -EFAULT;
192 
193 	start = range[0];
194 	len = range[1];
195 	if ((start & 511) || (len & 511))
196 		return -EINVAL;
197 	if (check_add_overflow(start, len, &end) ||
198 	    end > bdev_nr_bytes(bdev))
199 		return -EINVAL;
200 
201 	filemap_invalidate_lock(bdev->bd_mapping);
202 	err = truncate_bdev_range(bdev, mode, start, end - 1);
203 	if (!err)
204 		err = blkdev_issue_secure_erase(bdev, start >> 9, len >> 9,
205 						GFP_KERNEL);
206 	filemap_invalidate_unlock(bdev->bd_mapping);
207 	return err;
208 }
209 
210 
211 static int blk_ioctl_zeroout(struct block_device *bdev, blk_mode_t mode,
212 		unsigned long arg)
213 {
214 	uint64_t range[2];
215 	uint64_t start, end, len;
216 	int err;
217 
218 	if (!(mode & BLK_OPEN_WRITE))
219 		return -EBADF;
220 
221 	if (copy_from_user(range, (void __user *)arg, sizeof(range)))
222 		return -EFAULT;
223 
224 	start = range[0];
225 	len = range[1];
226 	end = start + len - 1;
227 
228 	if (start & 511)
229 		return -EINVAL;
230 	if (len & 511)
231 		return -EINVAL;
232 	if (end >= (uint64_t)bdev_nr_bytes(bdev))
233 		return -EINVAL;
234 	if (end < start)
235 		return -EINVAL;
236 
237 	/* Invalidate the page cache, including dirty pages */
238 	filemap_invalidate_lock(bdev->bd_mapping);
239 	err = truncate_bdev_range(bdev, mode, start, end);
240 	if (err)
241 		goto fail;
242 
243 	err = blkdev_issue_zeroout(bdev, start >> 9, len >> 9, GFP_KERNEL,
244 				   BLKDEV_ZERO_NOUNMAP | BLKDEV_ZERO_KILLABLE);
245 
246 fail:
247 	filemap_invalidate_unlock(bdev->bd_mapping);
248 	return err;
249 }
250 
251 static int put_ushort(unsigned short __user *argp, unsigned short val)
252 {
253 	return put_user(val, argp);
254 }
255 
256 static int put_int(int __user *argp, int val)
257 {
258 	return put_user(val, argp);
259 }
260 
261 static int put_uint(unsigned int __user *argp, unsigned int val)
262 {
263 	return put_user(val, argp);
264 }
265 
266 static int put_long(long __user *argp, long val)
267 {
268 	return put_user(val, argp);
269 }
270 
271 static int put_ulong(unsigned long __user *argp, unsigned long val)
272 {
273 	return put_user(val, argp);
274 }
275 
276 static int put_u64(u64 __user *argp, u64 val)
277 {
278 	return put_user(val, argp);
279 }
280 
281 #ifdef CONFIG_COMPAT
282 static int compat_put_long(compat_long_t __user *argp, long val)
283 {
284 	return put_user(val, argp);
285 }
286 
287 static int compat_put_ulong(compat_ulong_t __user *argp, compat_ulong_t val)
288 {
289 	return put_user(val, argp);
290 }
291 #endif
292 
293 #ifdef CONFIG_COMPAT
294 /*
295  * This is the equivalent of compat_ptr_ioctl(), to be used by block
296  * drivers that implement only commands that are completely compatible
297  * between 32-bit and 64-bit user space
298  */
299 int blkdev_compat_ptr_ioctl(struct block_device *bdev, blk_mode_t mode,
300 			unsigned cmd, unsigned long arg)
301 {
302 	struct gendisk *disk = bdev->bd_disk;
303 
304 	if (disk->fops->ioctl)
305 		return disk->fops->ioctl(bdev, mode, cmd,
306 					 (unsigned long)compat_ptr(arg));
307 
308 	return -ENOIOCTLCMD;
309 }
310 EXPORT_SYMBOL(blkdev_compat_ptr_ioctl);
311 #endif
312 
313 static bool blkdev_pr_allowed(struct block_device *bdev, blk_mode_t mode)
314 {
315 	/* no sense to make reservations for partitions */
316 	if (bdev_is_partition(bdev))
317 		return false;
318 
319 	if (capable(CAP_SYS_ADMIN))
320 		return true;
321 	/*
322 	 * Only allow unprivileged reservations if the file descriptor is open
323 	 * for writing.
324 	 */
325 	return mode & BLK_OPEN_WRITE;
326 }
327 
328 static int blkdev_pr_register(struct block_device *bdev, blk_mode_t mode,
329 		struct pr_registration __user *arg)
330 {
331 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
332 	struct pr_registration reg;
333 
334 	if (!blkdev_pr_allowed(bdev, mode))
335 		return -EPERM;
336 	if (!ops || !ops->pr_register)
337 		return -EOPNOTSUPP;
338 	if (copy_from_user(&reg, arg, sizeof(reg)))
339 		return -EFAULT;
340 
341 	if (reg.flags & ~PR_FL_IGNORE_KEY)
342 		return -EOPNOTSUPP;
343 	return ops->pr_register(bdev, reg.old_key, reg.new_key, reg.flags);
344 }
345 
346 static int blkdev_pr_reserve(struct block_device *bdev, blk_mode_t mode,
347 		struct pr_reservation __user *arg)
348 {
349 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
350 	struct pr_reservation rsv;
351 
352 	if (!blkdev_pr_allowed(bdev, mode))
353 		return -EPERM;
354 	if (!ops || !ops->pr_reserve)
355 		return -EOPNOTSUPP;
356 	if (copy_from_user(&rsv, arg, sizeof(rsv)))
357 		return -EFAULT;
358 
359 	if (rsv.flags & ~PR_FL_IGNORE_KEY)
360 		return -EOPNOTSUPP;
361 	return ops->pr_reserve(bdev, rsv.key, rsv.type, rsv.flags);
362 }
363 
364 static int blkdev_pr_release(struct block_device *bdev, blk_mode_t mode,
365 		struct pr_reservation __user *arg)
366 {
367 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
368 	struct pr_reservation rsv;
369 
370 	if (!blkdev_pr_allowed(bdev, mode))
371 		return -EPERM;
372 	if (!ops || !ops->pr_release)
373 		return -EOPNOTSUPP;
374 	if (copy_from_user(&rsv, arg, sizeof(rsv)))
375 		return -EFAULT;
376 
377 	if (rsv.flags)
378 		return -EOPNOTSUPP;
379 	return ops->pr_release(bdev, rsv.key, rsv.type);
380 }
381 
382 static int blkdev_pr_preempt(struct block_device *bdev, blk_mode_t mode,
383 		struct pr_preempt __user *arg, bool abort)
384 {
385 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
386 	struct pr_preempt p;
387 
388 	if (!blkdev_pr_allowed(bdev, mode))
389 		return -EPERM;
390 	if (!ops || !ops->pr_preempt)
391 		return -EOPNOTSUPP;
392 	if (copy_from_user(&p, arg, sizeof(p)))
393 		return -EFAULT;
394 
395 	if (p.flags)
396 		return -EOPNOTSUPP;
397 	return ops->pr_preempt(bdev, p.old_key, p.new_key, p.type, abort);
398 }
399 
400 static int blkdev_pr_clear(struct block_device *bdev, blk_mode_t mode,
401 		struct pr_clear __user *arg)
402 {
403 	const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
404 	struct pr_clear c;
405 
406 	if (!blkdev_pr_allowed(bdev, mode))
407 		return -EPERM;
408 	if (!ops || !ops->pr_clear)
409 		return -EOPNOTSUPP;
410 	if (copy_from_user(&c, arg, sizeof(c)))
411 		return -EFAULT;
412 
413 	if (c.flags)
414 		return -EOPNOTSUPP;
415 	return ops->pr_clear(bdev, c.key);
416 }
417 
418 static int blkdev_flushbuf(struct block_device *bdev, unsigned cmd,
419 		unsigned long arg)
420 {
421 	if (!capable(CAP_SYS_ADMIN))
422 		return -EACCES;
423 
424 	mutex_lock(&bdev->bd_holder_lock);
425 	if (bdev->bd_holder_ops && bdev->bd_holder_ops->sync)
426 		bdev->bd_holder_ops->sync(bdev);
427 	else {
428 		mutex_unlock(&bdev->bd_holder_lock);
429 		sync_blockdev(bdev);
430 	}
431 
432 	invalidate_bdev(bdev);
433 	return 0;
434 }
435 
436 static int blkdev_roset(struct block_device *bdev, unsigned cmd,
437 		unsigned long arg)
438 {
439 	int ret, n;
440 
441 	if (!capable(CAP_SYS_ADMIN))
442 		return -EACCES;
443 
444 	if (get_user(n, (int __user *)arg))
445 		return -EFAULT;
446 	if (bdev->bd_disk->fops->set_read_only) {
447 		ret = bdev->bd_disk->fops->set_read_only(bdev, n);
448 		if (ret)
449 			return ret;
450 	}
451 	if (n)
452 		bdev_set_flag(bdev, BD_READ_ONLY);
453 	else
454 		bdev_clear_flag(bdev, BD_READ_ONLY);
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 #ifdef CONFIG_COMPAT
485 struct compat_hd_geometry {
486 	unsigned char heads;
487 	unsigned char sectors;
488 	unsigned short cylinders;
489 	u32 start;
490 };
491 
492 static int compat_hdio_getgeo(struct block_device *bdev,
493 			      struct compat_hd_geometry __user *ugeo)
494 {
495 	struct gendisk *disk = bdev->bd_disk;
496 	struct hd_geometry geo;
497 	int ret;
498 
499 	if (!ugeo)
500 		return -EINVAL;
501 	if (!disk->fops->getgeo)
502 		return -ENOTTY;
503 
504 	memset(&geo, 0, sizeof(geo));
505 	/*
506 	 * We need to set the startsect first, the driver may
507 	 * want to override it.
508 	 */
509 	geo.start = get_start_sect(bdev);
510 	ret = disk->fops->getgeo(bdev, &geo);
511 	if (ret)
512 		return ret;
513 
514 	ret = copy_to_user(ugeo, &geo, 4);
515 	ret |= put_user(geo.start, &ugeo->start);
516 	if (ret)
517 		ret = -EFAULT;
518 
519 	return ret;
520 }
521 #endif
522 
523 /* set the logical block size */
524 static int blkdev_bszset(struct file *file, blk_mode_t mode,
525 		int __user *argp)
526 {
527 	// this one might be file_inode(file)->i_rdev - a rare valid
528 	// use of file_inode() for those.
529 	dev_t dev = I_BDEV(file->f_mapping->host)->bd_dev;
530 	struct file *excl_file;
531 	int ret, n;
532 
533 	if (!capable(CAP_SYS_ADMIN))
534 		return -EACCES;
535 	if (!argp)
536 		return -EINVAL;
537 	if (get_user(n, argp))
538 		return -EFAULT;
539 
540 	if (mode & BLK_OPEN_EXCL)
541 		return set_blocksize(file, n);
542 
543 	excl_file = bdev_file_open_by_dev(dev, mode, &dev, NULL);
544 	if (IS_ERR(excl_file))
545 		return -EBUSY;
546 	ret = set_blocksize(excl_file, n);
547 	fput(excl_file);
548 	return ret;
549 }
550 
551 /*
552  * Common commands that are handled the same way on native and compat
553  * user space. Note the separate arg/argp parameters that are needed
554  * to deal with the compat_ptr() conversion.
555  */
556 static int blkdev_common_ioctl(struct block_device *bdev, blk_mode_t mode,
557 			       unsigned int cmd, unsigned long arg,
558 			       void __user *argp)
559 {
560 	unsigned int max_sectors;
561 
562 	switch (cmd) {
563 	case BLKFLSBUF:
564 		return blkdev_flushbuf(bdev, cmd, arg);
565 	case BLKROSET:
566 		return blkdev_roset(bdev, cmd, arg);
567 	case BLKDISCARD:
568 		return blk_ioctl_discard(bdev, mode, arg);
569 	case BLKSECDISCARD:
570 		return blk_ioctl_secure_erase(bdev, mode, argp);
571 	case BLKZEROOUT:
572 		return blk_ioctl_zeroout(bdev, mode, arg);
573 	case BLKGETDISKSEQ:
574 		return put_u64(argp, bdev->bd_disk->diskseq);
575 	case BLKREPORTZONE:
576 		return blkdev_report_zones_ioctl(bdev, cmd, arg);
577 	case BLKRESETZONE:
578 	case BLKOPENZONE:
579 	case BLKCLOSEZONE:
580 	case BLKFINISHZONE:
581 		return blkdev_zone_mgmt_ioctl(bdev, mode, cmd, arg);
582 	case BLKGETZONESZ:
583 		return put_uint(argp, bdev_zone_sectors(bdev));
584 	case BLKGETNRZONES:
585 		return put_uint(argp, bdev_nr_zones(bdev));
586 	case BLKROGET:
587 		return put_int(argp, bdev_read_only(bdev) != 0);
588 	case BLKSSZGET: /* get block device logical block size */
589 		return put_int(argp, bdev_logical_block_size(bdev));
590 	case BLKPBSZGET: /* get block device physical block size */
591 		return put_uint(argp, bdev_physical_block_size(bdev));
592 	case BLKIOMIN:
593 		return put_uint(argp, bdev_io_min(bdev));
594 	case BLKIOOPT:
595 		return put_uint(argp, bdev_io_opt(bdev));
596 	case BLKALIGNOFF:
597 		return put_int(argp, bdev_alignment_offset(bdev));
598 	case BLKDISCARDZEROES:
599 		return put_uint(argp, 0);
600 	case BLKSECTGET:
601 		max_sectors = min_t(unsigned int, USHRT_MAX,
602 				    queue_max_sectors(bdev_get_queue(bdev)));
603 		return put_ushort(argp, max_sectors);
604 	case BLKROTATIONAL:
605 		return put_ushort(argp, !bdev_nonrot(bdev));
606 	case BLKRASET:
607 	case BLKFRASET:
608 		if(!capable(CAP_SYS_ADMIN))
609 			return -EACCES;
610 		bdev->bd_disk->bdi->ra_pages = (arg * 512) / PAGE_SIZE;
611 		return 0;
612 	case BLKRRPART:
613 		if (!capable(CAP_SYS_ADMIN))
614 			return -EACCES;
615 		if (bdev_is_partition(bdev))
616 			return -EINVAL;
617 		return disk_scan_partitions(bdev->bd_disk,
618 				mode | BLK_OPEN_STRICT_SCAN);
619 	case BLKTRACESTART:
620 	case BLKTRACESTOP:
621 	case BLKTRACETEARDOWN:
622 		return blk_trace_ioctl(bdev, cmd, argp);
623 	case IOC_PR_REGISTER:
624 		return blkdev_pr_register(bdev, mode, argp);
625 	case IOC_PR_RESERVE:
626 		return blkdev_pr_reserve(bdev, mode, argp);
627 	case IOC_PR_RELEASE:
628 		return blkdev_pr_release(bdev, mode, argp);
629 	case IOC_PR_PREEMPT:
630 		return blkdev_pr_preempt(bdev, mode, argp, false);
631 	case IOC_PR_PREEMPT_ABORT:
632 		return blkdev_pr_preempt(bdev, mode, argp, true);
633 	case IOC_PR_CLEAR:
634 		return blkdev_pr_clear(bdev, mode, argp);
635 	default:
636 		return -ENOIOCTLCMD;
637 	}
638 }
639 
640 /*
641  * Always keep this in sync with compat_blkdev_ioctl()
642  * to handle all incompatible commands in both functions.
643  *
644  * New commands must be compatible and go into blkdev_common_ioctl
645  */
646 long blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg)
647 {
648 	struct block_device *bdev = I_BDEV(file->f_mapping->host);
649 	void __user *argp = (void __user *)arg;
650 	blk_mode_t mode = file_to_blk_mode(file);
651 	int ret;
652 
653 	switch (cmd) {
654 	/* These need separate implementations for the data structure */
655 	case HDIO_GETGEO:
656 		return blkdev_getgeo(bdev, argp);
657 	case BLKPG:
658 		return blkpg_ioctl(bdev, argp);
659 
660 	/* Compat mode returns 32-bit data instead of 'long' */
661 	case BLKRAGET:
662 	case BLKFRAGET:
663 		if (!argp)
664 			return -EINVAL;
665 		return put_long(argp,
666 			(bdev->bd_disk->bdi->ra_pages * PAGE_SIZE) / 512);
667 	case BLKGETSIZE:
668 		if (bdev_nr_sectors(bdev) > ~0UL)
669 			return -EFBIG;
670 		return put_ulong(argp, bdev_nr_sectors(bdev));
671 
672 	/* The data is compatible, but the command number is different */
673 	case BLKBSZGET: /* get block device soft block size (cf. BLKSSZGET) */
674 		return put_int(argp, block_size(bdev));
675 	case BLKBSZSET:
676 		return blkdev_bszset(file, mode, argp);
677 	case BLKGETSIZE64:
678 		return put_u64(argp, bdev_nr_bytes(bdev));
679 
680 	/* Incompatible alignment on i386 */
681 	case BLKTRACESETUP:
682 		return blk_trace_ioctl(bdev, cmd, argp);
683 	default:
684 		break;
685 	}
686 
687 	ret = blkdev_common_ioctl(bdev, mode, cmd, arg, argp);
688 	if (ret != -ENOIOCTLCMD)
689 		return ret;
690 
691 	if (!bdev->bd_disk->fops->ioctl)
692 		return -ENOTTY;
693 	return bdev->bd_disk->fops->ioctl(bdev, mode, cmd, arg);
694 }
695 
696 #ifdef CONFIG_COMPAT
697 
698 #define BLKBSZGET_32		_IOR(0x12, 112, int)
699 #define BLKBSZSET_32		_IOW(0x12, 113, int)
700 #define BLKGETSIZE64_32		_IOR(0x12, 114, int)
701 
702 /* Most of the generic ioctls are handled in the normal fallback path.
703    This assumes the blkdev's low level compat_ioctl always returns
704    ENOIOCTLCMD for unknown ioctls. */
705 long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg)
706 {
707 	int ret;
708 	void __user *argp = compat_ptr(arg);
709 	struct block_device *bdev = I_BDEV(file->f_mapping->host);
710 	struct gendisk *disk = bdev->bd_disk;
711 	blk_mode_t mode = file_to_blk_mode(file);
712 
713 	switch (cmd) {
714 	/* These need separate implementations for the data structure */
715 	case HDIO_GETGEO:
716 		return compat_hdio_getgeo(bdev, argp);
717 	case BLKPG:
718 		return compat_blkpg_ioctl(bdev, argp);
719 
720 	/* Compat mode returns 32-bit data instead of 'long' */
721 	case BLKRAGET:
722 	case BLKFRAGET:
723 		if (!argp)
724 			return -EINVAL;
725 		return compat_put_long(argp,
726 			(bdev->bd_disk->bdi->ra_pages * PAGE_SIZE) / 512);
727 	case BLKGETSIZE:
728 		if (bdev_nr_sectors(bdev) > ~(compat_ulong_t)0)
729 			return -EFBIG;
730 		return compat_put_ulong(argp, bdev_nr_sectors(bdev));
731 
732 	/* The data is compatible, but the command number is different */
733 	case BLKBSZGET_32: /* get the logical block size (cf. BLKSSZGET) */
734 		return put_int(argp, bdev_logical_block_size(bdev));
735 	case BLKBSZSET_32:
736 		return blkdev_bszset(file, mode, argp);
737 	case BLKGETSIZE64_32:
738 		return put_u64(argp, bdev_nr_bytes(bdev));
739 
740 	/* Incompatible alignment on i386 */
741 	case BLKTRACESETUP32:
742 		return blk_trace_ioctl(bdev, cmd, argp);
743 	default:
744 		break;
745 	}
746 
747 	ret = blkdev_common_ioctl(bdev, mode, cmd, arg, argp);
748 	if (ret == -ENOIOCTLCMD && disk->fops->compat_ioctl)
749 		ret = disk->fops->compat_ioctl(bdev, mode, cmd, arg);
750 
751 	return ret;
752 }
753 #endif
754 
755 struct blk_iou_cmd {
756 	int res;
757 	bool nowait;
758 };
759 
760 static void blk_cmd_complete(struct io_uring_cmd *cmd, unsigned int issue_flags)
761 {
762 	struct blk_iou_cmd *bic = io_uring_cmd_to_pdu(cmd, struct blk_iou_cmd);
763 
764 	if (bic->res == -EAGAIN && bic->nowait)
765 		io_uring_cmd_issue_blocking(cmd);
766 	else
767 		io_uring_cmd_done(cmd, bic->res, 0, issue_flags);
768 }
769 
770 static void bio_cmd_bio_end_io(struct bio *bio)
771 {
772 	struct io_uring_cmd *cmd = bio->bi_private;
773 	struct blk_iou_cmd *bic = io_uring_cmd_to_pdu(cmd, struct blk_iou_cmd);
774 
775 	if (unlikely(bio->bi_status) && !bic->res)
776 		bic->res = blk_status_to_errno(bio->bi_status);
777 
778 	io_uring_cmd_do_in_task_lazy(cmd, blk_cmd_complete);
779 	bio_put(bio);
780 }
781 
782 static int blkdev_cmd_discard(struct io_uring_cmd *cmd,
783 			      struct block_device *bdev,
784 			      uint64_t start, uint64_t len, bool nowait)
785 {
786 	struct blk_iou_cmd *bic = io_uring_cmd_to_pdu(cmd, struct blk_iou_cmd);
787 	gfp_t gfp = nowait ? GFP_NOWAIT : GFP_KERNEL;
788 	sector_t sector = start >> SECTOR_SHIFT;
789 	sector_t nr_sects = len >> SECTOR_SHIFT;
790 	struct bio *prev = NULL, *bio;
791 	int err;
792 
793 	if (!bdev_max_discard_sectors(bdev))
794 		return -EOPNOTSUPP;
795 	if (!(file_to_blk_mode(cmd->file) & BLK_OPEN_WRITE))
796 		return -EBADF;
797 	if (bdev_read_only(bdev))
798 		return -EPERM;
799 	err = blk_validate_byte_range(bdev, start, len);
800 	if (err)
801 		return err;
802 
803 	err = filemap_invalidate_pages(bdev->bd_mapping, start,
804 					start + len - 1, nowait);
805 	if (err)
806 		return err;
807 
808 	while (true) {
809 		bio = blk_alloc_discard_bio(bdev, &sector, &nr_sects, gfp);
810 		if (!bio)
811 			break;
812 		if (nowait) {
813 			/*
814 			 * Don't allow multi-bio non-blocking submissions as
815 			 * subsequent bios may fail but we won't get a direct
816 			 * indication of that. Normally, the caller should
817 			 * retry from a blocking context.
818 			 */
819 			if (unlikely(nr_sects)) {
820 				bio_put(bio);
821 				return -EAGAIN;
822 			}
823 			bio->bi_opf |= REQ_NOWAIT;
824 		}
825 
826 		prev = bio_chain_and_submit(prev, bio);
827 	}
828 	if (unlikely(!prev))
829 		return -EAGAIN;
830 	if (unlikely(nr_sects))
831 		bic->res = -EAGAIN;
832 
833 	prev->bi_private = cmd;
834 	prev->bi_end_io = bio_cmd_bio_end_io;
835 	submit_bio(prev);
836 	return -EIOCBQUEUED;
837 }
838 
839 int blkdev_uring_cmd(struct io_uring_cmd *cmd, unsigned int issue_flags)
840 {
841 	struct block_device *bdev = I_BDEV(cmd->file->f_mapping->host);
842 	struct blk_iou_cmd *bic = io_uring_cmd_to_pdu(cmd, struct blk_iou_cmd);
843 	const struct io_uring_sqe *sqe = cmd->sqe;
844 	u32 cmd_op = cmd->cmd_op;
845 	uint64_t start, len;
846 
847 	if (unlikely(sqe->ioprio || sqe->__pad1 || sqe->len ||
848 		     sqe->rw_flags || sqe->file_index))
849 		return -EINVAL;
850 
851 	bic->res = 0;
852 	bic->nowait = issue_flags & IO_URING_F_NONBLOCK;
853 
854 	start = READ_ONCE(sqe->addr);
855 	len = READ_ONCE(sqe->addr3);
856 
857 	switch (cmd_op) {
858 	case BLOCK_URING_CMD_DISCARD:
859 		return blkdev_cmd_discard(cmd, bdev, start, len, bic->nowait);
860 	}
861 	return -EINVAL;
862 }
863