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 <linux/blk-integrity.h>
17 #include <uapi/linux/blkdev.h>
18 #include "blk.h"
19 #include "blk-crypto-internal.h"
20
blkpg_do_ioctl(struct block_device * bdev,struct blkpg_partition __user * upart,int op)21 static int blkpg_do_ioctl(struct block_device *bdev,
22 struct blkpg_partition __user *upart, int op)
23 {
24 struct gendisk *disk = bdev->bd_disk;
25 struct blkpg_partition p;
26 sector_t start, length, capacity, end;
27
28 if (!capable(CAP_SYS_ADMIN))
29 return -EACCES;
30 if (copy_from_user(&p, upart, sizeof(struct blkpg_partition)))
31 return -EFAULT;
32 if (bdev_is_partition(bdev))
33 return -EINVAL;
34
35 if (p.pno <= 0)
36 return -EINVAL;
37
38 if (op == BLKPG_DEL_PARTITION)
39 return bdev_del_partition(disk, p.pno);
40
41 if (p.start < 0 || p.length <= 0 || LLONG_MAX - p.length < p.start)
42 return -EINVAL;
43 /* Check that the partition is aligned to the block size */
44 if (!IS_ALIGNED(p.start | p.length, bdev_logical_block_size(bdev)))
45 return -EINVAL;
46
47 start = p.start >> SECTOR_SHIFT;
48 length = p.length >> SECTOR_SHIFT;
49 capacity = get_capacity(disk);
50
51 if (check_add_overflow(start, length, &end))
52 return -EINVAL;
53
54 if (start >= capacity || end > capacity)
55 return -EINVAL;
56
57 switch (op) {
58 case BLKPG_ADD_PARTITION:
59 return bdev_add_partition(disk, p.pno, start, length);
60 case BLKPG_RESIZE_PARTITION:
61 return bdev_resize_partition(disk, p.pno, start, length);
62 default:
63 return -EINVAL;
64 }
65 }
66
blkpg_ioctl(struct block_device * bdev,struct blkpg_ioctl_arg __user * arg)67 static int blkpg_ioctl(struct block_device *bdev,
68 struct blkpg_ioctl_arg __user *arg)
69 {
70 struct blkpg_partition __user *udata;
71 int op;
72
73 if (get_user(op, &arg->op) || get_user(udata, &arg->data))
74 return -EFAULT;
75
76 return blkpg_do_ioctl(bdev, udata, op);
77 }
78
79 #ifdef CONFIG_COMPAT
80 struct compat_blkpg_ioctl_arg {
81 compat_int_t op;
82 compat_int_t flags;
83 compat_int_t datalen;
84 compat_caddr_t data;
85 };
86
compat_blkpg_ioctl(struct block_device * bdev,struct compat_blkpg_ioctl_arg __user * arg)87 static int compat_blkpg_ioctl(struct block_device *bdev,
88 struct compat_blkpg_ioctl_arg __user *arg)
89 {
90 compat_caddr_t udata;
91 int op;
92
93 if (get_user(op, &arg->op) || get_user(udata, &arg->data))
94 return -EFAULT;
95
96 return blkpg_do_ioctl(bdev, compat_ptr(udata), op);
97 }
98 #endif
99
100 /*
101 * Check that [start, start + len) is a valid range from the block device's
102 * perspective, including verifying that it can be correctly translated into
103 * logical block addresses.
104 */
blk_validate_byte_range(struct block_device * bdev,uint64_t start,uint64_t len)105 static int blk_validate_byte_range(struct block_device *bdev,
106 uint64_t start, uint64_t len)
107 {
108 unsigned int bs_mask = bdev_logical_block_size(bdev) - 1;
109 uint64_t end;
110
111 if ((start | len) & bs_mask)
112 return -EINVAL;
113 if (!len)
114 return -EINVAL;
115 if (check_add_overflow(start, len, &end) || end > bdev_nr_bytes(bdev))
116 return -EINVAL;
117
118 return 0;
119 }
120
blk_ioctl_discard(struct block_device * bdev,blk_mode_t mode,unsigned long arg)121 static int blk_ioctl_discard(struct block_device *bdev, blk_mode_t mode,
122 unsigned long arg)
123 {
124 uint64_t range[2], start, len;
125 struct bio *prev = NULL, *bio;
126 sector_t sector, nr_sects;
127 struct blk_plug plug;
128 int err;
129
130 if (copy_from_user(range, (void __user *)arg, sizeof(range)))
131 return -EFAULT;
132 start = range[0];
133 len = range[1];
134
135 if (!bdev_max_discard_sectors(bdev))
136 return -EOPNOTSUPP;
137
138 if (!(mode & BLK_OPEN_WRITE))
139 return -EBADF;
140 if (bdev_read_only(bdev))
141 return -EPERM;
142 err = blk_validate_byte_range(bdev, start, len);
143 if (err)
144 return err;
145
146 inode_lock(bdev->bd_mapping->host);
147 filemap_invalidate_lock(bdev->bd_mapping);
148 err = truncate_bdev_range(bdev, mode, start, start + len - 1);
149 if (err)
150 goto fail;
151
152 sector = start >> SECTOR_SHIFT;
153 nr_sects = len >> SECTOR_SHIFT;
154
155 blk_start_plug(&plug);
156 while (!fatal_signal_pending(current)) {
157 bio = blk_alloc_discard_bio(bdev, §or, &nr_sects,
158 GFP_KERNEL);
159 if (!bio)
160 break;
161 prev = bio_chain_and_submit(prev, bio);
162 }
163 if (prev) {
164 err = bio_submit_or_kill(prev, BLKDEV_ZERO_KILLABLE);
165 if (err == -EOPNOTSUPP)
166 err = 0;
167 bio_put(prev);
168 }
169 blk_finish_plug(&plug);
170 fail:
171 filemap_invalidate_unlock(bdev->bd_mapping);
172 inode_unlock(bdev->bd_mapping->host);
173 return err;
174 }
175
blk_ioctl_secure_erase(struct block_device * bdev,blk_mode_t mode,void __user * argp)176 static int blk_ioctl_secure_erase(struct block_device *bdev, blk_mode_t mode,
177 void __user *argp)
178 {
179 uint64_t start, len, end;
180 uint64_t range[2];
181 int err;
182
183 if (!(mode & BLK_OPEN_WRITE))
184 return -EBADF;
185 if (!bdev_max_secure_erase_sectors(bdev))
186 return -EOPNOTSUPP;
187 if (copy_from_user(range, argp, sizeof(range)))
188 return -EFAULT;
189
190 start = range[0];
191 len = range[1];
192 if ((start & 511) || (len & 511))
193 return -EINVAL;
194 if (check_add_overflow(start, len, &end) ||
195 end > bdev_nr_bytes(bdev))
196 return -EINVAL;
197
198 inode_lock(bdev->bd_mapping->host);
199 filemap_invalidate_lock(bdev->bd_mapping);
200 err = truncate_bdev_range(bdev, mode, start, end - 1);
201 if (!err)
202 err = blkdev_issue_secure_erase(bdev, start >> 9, len >> 9,
203 GFP_KERNEL);
204 filemap_invalidate_unlock(bdev->bd_mapping);
205 inode_unlock(bdev->bd_mapping->host);
206 return err;
207 }
208
209
blk_ioctl_zeroout(struct block_device * bdev,blk_mode_t mode,unsigned long arg)210 static int blk_ioctl_zeroout(struct block_device *bdev, blk_mode_t mode,
211 unsigned long arg)
212 {
213 uint64_t range[2];
214 uint64_t start, end, len;
215 int err;
216
217 if (!(mode & BLK_OPEN_WRITE))
218 return -EBADF;
219
220 if (copy_from_user(range, (void __user *)arg, sizeof(range)))
221 return -EFAULT;
222
223 start = range[0];
224 len = range[1];
225 end = start + len - 1;
226
227 if (start & 511)
228 return -EINVAL;
229 if (len & 511)
230 return -EINVAL;
231 if (end >= (uint64_t)bdev_nr_bytes(bdev))
232 return -EINVAL;
233 if (end < start)
234 return -EINVAL;
235
236 /* Invalidate the page cache, including dirty pages */
237 inode_lock(bdev->bd_mapping->host);
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 inode_unlock(bdev->bd_mapping->host);
249 return err;
250 }
251
put_ushort(unsigned short __user * argp,unsigned short val)252 static int put_ushort(unsigned short __user *argp, unsigned short val)
253 {
254 return put_user(val, argp);
255 }
256
put_int(int __user * argp,int val)257 static int put_int(int __user *argp, int val)
258 {
259 return put_user(val, argp);
260 }
261
put_uint(unsigned int __user * argp,unsigned int val)262 static int put_uint(unsigned int __user *argp, unsigned int val)
263 {
264 return put_user(val, argp);
265 }
266
put_long(long __user * argp,long val)267 static int put_long(long __user *argp, long val)
268 {
269 return put_user(val, argp);
270 }
271
put_ulong(unsigned long __user * argp,unsigned long val)272 static int put_ulong(unsigned long __user *argp, unsigned long val)
273 {
274 return put_user(val, argp);
275 }
276
put_u64(u64 __user * argp,u64 val)277 static int put_u64(u64 __user *argp, u64 val)
278 {
279 return put_user(val, argp);
280 }
281
282 #ifdef CONFIG_COMPAT
compat_put_long(compat_long_t __user * argp,long val)283 static int compat_put_long(compat_long_t __user *argp, long val)
284 {
285 return put_user(val, argp);
286 }
287
compat_put_ulong(compat_ulong_t __user * argp,compat_ulong_t val)288 static int compat_put_ulong(compat_ulong_t __user *argp, compat_ulong_t val)
289 {
290 return put_user(val, argp);
291 }
292 #endif
293
294 #ifdef CONFIG_COMPAT
295 /*
296 * This is the equivalent of compat_ptr_ioctl(), to be used by block
297 * drivers that implement only commands that are completely compatible
298 * between 32-bit and 64-bit user space
299 */
blkdev_compat_ptr_ioctl(struct block_device * bdev,blk_mode_t mode,unsigned cmd,unsigned long arg)300 int blkdev_compat_ptr_ioctl(struct block_device *bdev, blk_mode_t mode,
301 unsigned cmd, unsigned long arg)
302 {
303 struct gendisk *disk = bdev->bd_disk;
304
305 if (disk->fops->ioctl)
306 return disk->fops->ioctl(bdev, mode, cmd,
307 (unsigned long)compat_ptr(arg));
308
309 return -ENOIOCTLCMD;
310 }
311 EXPORT_SYMBOL(blkdev_compat_ptr_ioctl);
312 #endif
313
314 enum pr_direction {
315 PR_IN, /* read from device */
316 PR_OUT, /* write to device */
317 };
318
blkdev_pr_allowed(struct block_device * bdev,blk_mode_t mode,enum pr_direction dir)319 static bool blkdev_pr_allowed(struct block_device *bdev, blk_mode_t mode,
320 enum pr_direction dir)
321 {
322 /* no sense to make reservations for partitions */
323 if (bdev_is_partition(bdev))
324 return false;
325
326 if (capable(CAP_SYS_ADMIN))
327 return true;
328
329 /*
330 * Only allow unprivileged reservation _out_ commands if the file
331 * descriptor is open for writing. Allow reservation _in_ commands if
332 * the file descriptor is open for reading since they do not modify the
333 * device.
334 */
335 if (dir == PR_IN)
336 return mode & BLK_OPEN_READ;
337 else
338 return mode & BLK_OPEN_WRITE;
339 }
340
blkdev_pr_register(struct block_device * bdev,blk_mode_t mode,struct pr_registration __user * arg)341 static int blkdev_pr_register(struct block_device *bdev, blk_mode_t mode,
342 struct pr_registration __user *arg)
343 {
344 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
345 struct pr_registration reg;
346
347 if (!blkdev_pr_allowed(bdev, mode, PR_OUT))
348 return -EPERM;
349 if (!ops || !ops->pr_register)
350 return -EOPNOTSUPP;
351 if (copy_from_user(®, arg, sizeof(reg)))
352 return -EFAULT;
353
354 if (reg.flags & ~PR_FL_IGNORE_KEY)
355 return -EOPNOTSUPP;
356 return ops->pr_register(bdev, reg.old_key, reg.new_key, reg.flags);
357 }
358
blkdev_pr_reserve(struct block_device * bdev,blk_mode_t mode,struct pr_reservation __user * arg)359 static int blkdev_pr_reserve(struct block_device *bdev, blk_mode_t mode,
360 struct pr_reservation __user *arg)
361 {
362 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
363 struct pr_reservation rsv;
364
365 if (!blkdev_pr_allowed(bdev, mode, PR_OUT))
366 return -EPERM;
367 if (!ops || !ops->pr_reserve)
368 return -EOPNOTSUPP;
369 if (copy_from_user(&rsv, arg, sizeof(rsv)))
370 return -EFAULT;
371
372 if (rsv.flags & ~PR_FL_IGNORE_KEY)
373 return -EOPNOTSUPP;
374 return ops->pr_reserve(bdev, rsv.key, rsv.type, rsv.flags);
375 }
376
blkdev_pr_release(struct block_device * bdev,blk_mode_t mode,struct pr_reservation __user * arg)377 static int blkdev_pr_release(struct block_device *bdev, blk_mode_t mode,
378 struct pr_reservation __user *arg)
379 {
380 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
381 struct pr_reservation rsv;
382
383 if (!blkdev_pr_allowed(bdev, mode, PR_OUT))
384 return -EPERM;
385 if (!ops || !ops->pr_release)
386 return -EOPNOTSUPP;
387 if (copy_from_user(&rsv, arg, sizeof(rsv)))
388 return -EFAULT;
389
390 if (rsv.flags)
391 return -EOPNOTSUPP;
392 return ops->pr_release(bdev, rsv.key, rsv.type);
393 }
394
blkdev_pr_preempt(struct block_device * bdev,blk_mode_t mode,struct pr_preempt __user * arg,bool abort)395 static int blkdev_pr_preempt(struct block_device *bdev, blk_mode_t mode,
396 struct pr_preempt __user *arg, bool abort)
397 {
398 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
399 struct pr_preempt p;
400
401 if (!blkdev_pr_allowed(bdev, mode, PR_OUT))
402 return -EPERM;
403 if (!ops || !ops->pr_preempt)
404 return -EOPNOTSUPP;
405 if (copy_from_user(&p, arg, sizeof(p)))
406 return -EFAULT;
407
408 if (p.flags)
409 return -EOPNOTSUPP;
410 return ops->pr_preempt(bdev, p.old_key, p.new_key, p.type, abort);
411 }
412
blkdev_pr_clear(struct block_device * bdev,blk_mode_t mode,struct pr_clear __user * arg)413 static int blkdev_pr_clear(struct block_device *bdev, blk_mode_t mode,
414 struct pr_clear __user *arg)
415 {
416 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
417 struct pr_clear c;
418
419 if (!blkdev_pr_allowed(bdev, mode, PR_OUT))
420 return -EPERM;
421 if (!ops || !ops->pr_clear)
422 return -EOPNOTSUPP;
423 if (copy_from_user(&c, arg, sizeof(c)))
424 return -EFAULT;
425
426 if (c.flags)
427 return -EOPNOTSUPP;
428 return ops->pr_clear(bdev, c.key);
429 }
430
blkdev_pr_read_keys(struct block_device * bdev,blk_mode_t mode,struct pr_read_keys __user * arg)431 static int blkdev_pr_read_keys(struct block_device *bdev, blk_mode_t mode,
432 struct pr_read_keys __user *arg)
433 {
434 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
435 struct pr_keys *keys_info;
436 struct pr_read_keys read_keys;
437 u64 __user *keys_ptr;
438 size_t keys_info_len;
439 size_t keys_copy_len;
440 int ret;
441
442 if (!blkdev_pr_allowed(bdev, mode, PR_IN))
443 return -EPERM;
444 if (!ops || !ops->pr_read_keys)
445 return -EOPNOTSUPP;
446
447 if (copy_from_user(&read_keys, arg, sizeof(read_keys)))
448 return -EFAULT;
449
450 if (read_keys.num_keys > PR_KEYS_MAX)
451 return -EINVAL;
452
453 keys_info_len = struct_size(keys_info, keys, read_keys.num_keys);
454
455 keys_info = kvzalloc(keys_info_len, GFP_KERNEL);
456 if (!keys_info)
457 return -ENOMEM;
458
459 keys_info->num_keys = read_keys.num_keys;
460
461 ret = ops->pr_read_keys(bdev, keys_info);
462 if (ret)
463 goto out;
464
465 /* Copy out individual keys */
466 keys_ptr = u64_to_user_ptr(read_keys.keys_ptr);
467 keys_copy_len = min(read_keys.num_keys, keys_info->num_keys) *
468 sizeof(keys_info->keys[0]);
469
470 if (copy_to_user(keys_ptr, keys_info->keys, keys_copy_len)) {
471 ret = -EFAULT;
472 goto out;
473 }
474
475 /* Copy out the arg struct */
476 read_keys.generation = keys_info->generation;
477 read_keys.num_keys = keys_info->num_keys;
478
479 if (copy_to_user(arg, &read_keys, sizeof(read_keys)))
480 ret = -EFAULT;
481 out:
482 kvfree(keys_info);
483 return ret;
484 }
485
blkdev_pr_read_reservation(struct block_device * bdev,blk_mode_t mode,struct pr_read_reservation __user * arg)486 static int blkdev_pr_read_reservation(struct block_device *bdev,
487 blk_mode_t mode, struct pr_read_reservation __user *arg)
488 {
489 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
490 struct pr_held_reservation rsv = {};
491 struct pr_read_reservation out = {};
492 int ret;
493
494 if (!blkdev_pr_allowed(bdev, mode, PR_IN))
495 return -EPERM;
496 if (!ops || !ops->pr_read_reservation)
497 return -EOPNOTSUPP;
498
499 ret = ops->pr_read_reservation(bdev, &rsv);
500 if (ret)
501 return ret;
502
503 out.key = rsv.key;
504 out.generation = rsv.generation;
505 out.type = rsv.type;
506
507 if (copy_to_user(arg, &out, sizeof(out)))
508 return -EFAULT;
509 return 0;
510 }
511
blkdev_flushbuf(struct block_device * bdev,unsigned cmd,unsigned long arg)512 static int blkdev_flushbuf(struct block_device *bdev, unsigned cmd,
513 unsigned long arg)
514 {
515 if (!capable(CAP_SYS_ADMIN))
516 return -EACCES;
517
518 mutex_lock(&bdev->bd_holder_lock);
519 if (bdev->bd_holder_ops && bdev->bd_holder_ops->sync)
520 bdev->bd_holder_ops->sync(bdev);
521 else {
522 mutex_unlock(&bdev->bd_holder_lock);
523 sync_blockdev(bdev);
524 }
525
526 invalidate_bdev(bdev);
527 return 0;
528 }
529
blkdev_roset(struct block_device * bdev,unsigned cmd,unsigned long arg)530 static int blkdev_roset(struct block_device *bdev, unsigned cmd,
531 unsigned long arg)
532 {
533 int ret, n;
534
535 if (!capable(CAP_SYS_ADMIN))
536 return -EACCES;
537
538 if (get_user(n, (int __user *)arg))
539 return -EFAULT;
540 if (bdev->bd_disk->fops->set_read_only) {
541 ret = bdev->bd_disk->fops->set_read_only(bdev, n);
542 if (ret)
543 return ret;
544 }
545 if (n)
546 bdev_set_flag(bdev, BD_READ_ONLY);
547 else
548 bdev_clear_flag(bdev, BD_READ_ONLY);
549 return 0;
550 }
551
blkdev_getgeo(struct block_device * bdev,struct hd_geometry __user * argp)552 static int blkdev_getgeo(struct block_device *bdev,
553 struct hd_geometry __user *argp)
554 {
555 struct gendisk *disk = bdev->bd_disk;
556 struct hd_geometry geo;
557 int ret;
558
559 if (!argp)
560 return -EINVAL;
561 if (!disk->fops->getgeo)
562 return -ENOTTY;
563
564 /*
565 * We need to set the startsect first, the driver may
566 * want to override it.
567 */
568 memset(&geo, 0, sizeof(geo));
569 geo.start = get_start_sect(bdev);
570 ret = disk->fops->getgeo(disk, &geo);
571 if (ret)
572 return ret;
573 if (copy_to_user(argp, &geo, sizeof(geo)))
574 return -EFAULT;
575 return 0;
576 }
577
578 #ifdef CONFIG_COMPAT
579 struct compat_hd_geometry {
580 unsigned char heads;
581 unsigned char sectors;
582 unsigned short cylinders;
583 u32 start;
584 };
585
compat_hdio_getgeo(struct block_device * bdev,struct compat_hd_geometry __user * ugeo)586 static int compat_hdio_getgeo(struct block_device *bdev,
587 struct compat_hd_geometry __user *ugeo)
588 {
589 struct gendisk *disk = bdev->bd_disk;
590 struct hd_geometry geo;
591 int ret;
592
593 if (!ugeo)
594 return -EINVAL;
595 if (!disk->fops->getgeo)
596 return -ENOTTY;
597
598 memset(&geo, 0, sizeof(geo));
599 /*
600 * We need to set the startsect first, the driver may
601 * want to override it.
602 */
603 geo.start = get_start_sect(bdev);
604 ret = disk->fops->getgeo(disk, &geo);
605 if (ret)
606 return ret;
607
608 ret = copy_to_user(ugeo, &geo, 4);
609 ret |= put_user(geo.start, &ugeo->start);
610 if (ret)
611 ret = -EFAULT;
612
613 return ret;
614 }
615 #endif
616
617 /* set the logical block size */
blkdev_bszset(struct file * file,blk_mode_t mode,int __user * argp)618 static int blkdev_bszset(struct file *file, blk_mode_t mode,
619 int __user *argp)
620 {
621 // this one might be file_inode(file)->i_rdev - a rare valid
622 // use of file_inode() for those.
623 dev_t dev = I_BDEV(file->f_mapping->host)->bd_dev;
624 struct file *excl_file;
625 int ret, n;
626
627 if (!capable(CAP_SYS_ADMIN))
628 return -EACCES;
629 if (!argp)
630 return -EINVAL;
631 if (get_user(n, argp))
632 return -EFAULT;
633
634 if (mode & BLK_OPEN_EXCL)
635 return set_blocksize(file, n);
636
637 excl_file = bdev_file_open_by_dev(dev, mode, &dev, NULL);
638 if (IS_ERR(excl_file))
639 return -EBUSY;
640 ret = set_blocksize(excl_file, n);
641 fput(excl_file);
642 return ret;
643 }
644
645 /*
646 * Common commands that are handled the same way on native and compat
647 * user space. Note the separate arg/argp parameters that are needed
648 * to deal with the compat_ptr() conversion.
649 */
blkdev_common_ioctl(struct block_device * bdev,blk_mode_t mode,unsigned int cmd,unsigned long arg,void __user * argp)650 static int blkdev_common_ioctl(struct block_device *bdev, blk_mode_t mode,
651 unsigned int cmd, unsigned long arg,
652 void __user *argp)
653 {
654 unsigned int max_sectors;
655
656 switch (cmd) {
657 case BLKFLSBUF:
658 return blkdev_flushbuf(bdev, cmd, arg);
659 case BLKROSET:
660 return blkdev_roset(bdev, cmd, arg);
661 case BLKDISCARD:
662 return blk_ioctl_discard(bdev, mode, arg);
663 case BLKSECDISCARD:
664 return blk_ioctl_secure_erase(bdev, mode, argp);
665 case BLKZEROOUT:
666 return blk_ioctl_zeroout(bdev, mode, arg);
667 case BLKGETDISKSEQ:
668 return put_u64(argp, bdev->bd_disk->diskseq);
669 case BLKREPORTZONE:
670 case BLKREPORTZONEV2:
671 return blkdev_report_zones_ioctl(bdev, cmd, arg);
672 case BLKRESETZONE:
673 case BLKOPENZONE:
674 case BLKCLOSEZONE:
675 case BLKFINISHZONE:
676 return blkdev_zone_mgmt_ioctl(bdev, mode, cmd, arg);
677 case BLKGETZONESZ:
678 return put_uint(argp, bdev_zone_sectors(bdev));
679 case BLKGETNRZONES:
680 return put_uint(argp, bdev_nr_zones(bdev));
681 case BLKROGET:
682 return put_int(argp, bdev_read_only(bdev) != 0);
683 case BLKSSZGET: /* get block device logical block size */
684 return put_int(argp, bdev_logical_block_size(bdev));
685 case BLKPBSZGET: /* get block device physical block size */
686 return put_uint(argp, bdev_physical_block_size(bdev));
687 case BLKIOMIN:
688 return put_uint(argp, bdev_io_min(bdev));
689 case BLKIOOPT:
690 return put_uint(argp, bdev_io_opt(bdev));
691 case BLKALIGNOFF:
692 return put_int(argp, bdev_alignment_offset(bdev));
693 case BLKDISCARDZEROES:
694 return put_uint(argp, 0);
695 case BLKSECTGET:
696 max_sectors = min_t(unsigned int, USHRT_MAX,
697 queue_max_sectors(bdev_get_queue(bdev)));
698 return put_ushort(argp, max_sectors);
699 case BLKROTATIONAL:
700 return put_ushort(argp, bdev_rot(bdev));
701 case BLKRASET:
702 case BLKFRASET:
703 if(!capable(CAP_SYS_ADMIN))
704 return -EACCES;
705 bdev->bd_disk->bdi->ra_pages = (arg * 512) / PAGE_SIZE;
706 return 0;
707 case BLKRRPART:
708 if (!capable(CAP_SYS_ADMIN))
709 return -EACCES;
710 if (bdev_is_partition(bdev))
711 return -EINVAL;
712 return disk_scan_partitions(bdev->bd_disk,
713 mode | BLK_OPEN_STRICT_SCAN);
714 case BLKTRACESTART:
715 case BLKTRACESTOP:
716 case BLKTRACETEARDOWN:
717 return blk_trace_ioctl(bdev, cmd, argp);
718 case BLKCRYPTOIMPORTKEY:
719 case BLKCRYPTOGENERATEKEY:
720 case BLKCRYPTOPREPAREKEY:
721 return blk_crypto_ioctl(bdev, cmd, argp);
722 case IOC_PR_REGISTER:
723 return blkdev_pr_register(bdev, mode, argp);
724 case IOC_PR_RESERVE:
725 return blkdev_pr_reserve(bdev, mode, argp);
726 case IOC_PR_RELEASE:
727 return blkdev_pr_release(bdev, mode, argp);
728 case IOC_PR_PREEMPT:
729 return blkdev_pr_preempt(bdev, mode, argp, false);
730 case IOC_PR_PREEMPT_ABORT:
731 return blkdev_pr_preempt(bdev, mode, argp, true);
732 case IOC_PR_CLEAR:
733 return blkdev_pr_clear(bdev, mode, argp);
734 case IOC_PR_READ_KEYS:
735 return blkdev_pr_read_keys(bdev, mode, argp);
736 case IOC_PR_READ_RESERVATION:
737 return blkdev_pr_read_reservation(bdev, mode, argp);
738 default:
739 return blk_get_meta_cap(bdev, cmd, argp);
740 }
741 }
742
743 /*
744 * Always keep this in sync with compat_blkdev_ioctl()
745 * to handle all incompatible commands in both functions.
746 *
747 * New commands must be compatible and go into blkdev_common_ioctl
748 */
blkdev_ioctl(struct file * file,unsigned cmd,unsigned long arg)749 long blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg)
750 {
751 struct block_device *bdev = I_BDEV(file->f_mapping->host);
752 void __user *argp = (void __user *)arg;
753 blk_mode_t mode = file_to_blk_mode(file);
754 int ret;
755
756 switch (cmd) {
757 /* These need separate implementations for the data structure */
758 case HDIO_GETGEO:
759 return blkdev_getgeo(bdev, argp);
760 case BLKPG:
761 return blkpg_ioctl(bdev, argp);
762
763 /* Compat mode returns 32-bit data instead of 'long' */
764 case BLKRAGET:
765 case BLKFRAGET:
766 if (!argp)
767 return -EINVAL;
768 return put_long(argp,
769 (bdev->bd_disk->bdi->ra_pages * PAGE_SIZE) / 512);
770 case BLKGETSIZE:
771 if (bdev_nr_sectors(bdev) > ~0UL)
772 return -EFBIG;
773 return put_ulong(argp, bdev_nr_sectors(bdev));
774
775 /* The data is compatible, but the command number is different */
776 case BLKBSZGET: /* get block device soft block size (cf. BLKSSZGET) */
777 return put_int(argp, block_size(bdev));
778 case BLKBSZSET:
779 return blkdev_bszset(file, mode, argp);
780 case BLKGETSIZE64:
781 return put_u64(argp, bdev_nr_bytes(bdev));
782
783 /* Incompatible alignment on i386 */
784 case BLKTRACESETUP:
785 case BLKTRACESETUP2:
786 return blk_trace_ioctl(bdev, cmd, argp);
787 default:
788 break;
789 }
790
791 ret = blkdev_common_ioctl(bdev, mode, cmd, arg, argp);
792 if (ret != -ENOIOCTLCMD)
793 return ret;
794
795 if (!bdev->bd_disk->fops->ioctl)
796 return -ENOTTY;
797 return bdev->bd_disk->fops->ioctl(bdev, mode, cmd, arg);
798 }
799
800 #ifdef CONFIG_COMPAT
801
802 #define BLKBSZGET_32 _IOR(0x12, 112, int)
803 #define BLKBSZSET_32 _IOW(0x12, 113, int)
804 #define BLKGETSIZE64_32 _IOR(0x12, 114, int)
805
806 /* Most of the generic ioctls are handled in the normal fallback path.
807 This assumes the blkdev's low level compat_ioctl always returns
808 ENOIOCTLCMD for unknown ioctls. */
compat_blkdev_ioctl(struct file * file,unsigned cmd,unsigned long arg)809 long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg)
810 {
811 int ret;
812 void __user *argp = compat_ptr(arg);
813 struct block_device *bdev = I_BDEV(file->f_mapping->host);
814 struct gendisk *disk = bdev->bd_disk;
815 blk_mode_t mode = file_to_blk_mode(file);
816
817 switch (cmd) {
818 /* These need separate implementations for the data structure */
819 case HDIO_GETGEO:
820 return compat_hdio_getgeo(bdev, argp);
821 case BLKPG:
822 return compat_blkpg_ioctl(bdev, argp);
823
824 /* Compat mode returns 32-bit data instead of 'long' */
825 case BLKRAGET:
826 case BLKFRAGET:
827 if (!argp)
828 return -EINVAL;
829 return compat_put_long(argp,
830 (bdev->bd_disk->bdi->ra_pages * PAGE_SIZE) / 512);
831 case BLKGETSIZE:
832 if (bdev_nr_sectors(bdev) > ~(compat_ulong_t)0)
833 return -EFBIG;
834 return compat_put_ulong(argp, bdev_nr_sectors(bdev));
835
836 /* The data is compatible, but the command number is different */
837 case BLKBSZGET_32: /* get the logical block size (cf. BLKSSZGET) */
838 return put_int(argp, bdev_logical_block_size(bdev));
839 case BLKBSZSET_32:
840 return blkdev_bszset(file, mode, argp);
841 case BLKGETSIZE64_32:
842 return put_u64(argp, bdev_nr_bytes(bdev));
843
844 /* Incompatible alignment on i386 */
845 case BLKTRACESETUP32:
846 return blk_trace_ioctl(bdev, cmd, argp);
847 default:
848 break;
849 }
850
851 ret = blkdev_common_ioctl(bdev, mode, cmd, arg, argp);
852 if (ret == -ENOIOCTLCMD && disk->fops->compat_ioctl)
853 ret = disk->fops->compat_ioctl(bdev, mode, cmd, arg);
854
855 return ret;
856 }
857 #endif
858
859 struct blk_iou_cmd {
860 int res;
861 bool nowait;
862 };
863
blk_cmd_complete(struct io_tw_req tw_req,io_tw_token_t tw)864 static void blk_cmd_complete(struct io_tw_req tw_req, io_tw_token_t tw)
865 {
866 struct io_uring_cmd *cmd = io_uring_cmd_from_tw(tw_req);
867 struct blk_iou_cmd *bic = io_uring_cmd_to_pdu(cmd, struct blk_iou_cmd);
868
869 if (bic->res == -EAGAIN && bic->nowait)
870 io_uring_cmd_issue_blocking(cmd);
871 else
872 io_uring_cmd_done(cmd, bic->res,
873 IO_URING_CMD_TASK_WORK_ISSUE_FLAGS);
874 }
875
bio_cmd_bio_end_io(struct bio * bio)876 static void bio_cmd_bio_end_io(struct bio *bio)
877 {
878 struct io_uring_cmd *cmd = bio->bi_private;
879 struct blk_iou_cmd *bic = io_uring_cmd_to_pdu(cmd, struct blk_iou_cmd);
880
881 if (unlikely(bio->bi_status) && !bic->res)
882 bic->res = blk_status_to_errno(bio->bi_status);
883
884 io_uring_cmd_do_in_task_lazy(cmd, blk_cmd_complete);
885 bio_put(bio);
886 }
887
blkdev_cmd_discard(struct io_uring_cmd * cmd,struct block_device * bdev,uint64_t start,uint64_t len,bool nowait)888 static int blkdev_cmd_discard(struct io_uring_cmd *cmd,
889 struct block_device *bdev,
890 uint64_t start, uint64_t len, bool nowait)
891 {
892 struct blk_iou_cmd *bic = io_uring_cmd_to_pdu(cmd, struct blk_iou_cmd);
893 gfp_t gfp = nowait ? GFP_NOWAIT : GFP_KERNEL;
894 sector_t sector = start >> SECTOR_SHIFT;
895 sector_t nr_sects = len >> SECTOR_SHIFT;
896 struct bio *prev = NULL, *bio;
897 int err;
898
899 if (!bdev_max_discard_sectors(bdev))
900 return -EOPNOTSUPP;
901 if (!(file_to_blk_mode(cmd->file) & BLK_OPEN_WRITE))
902 return -EBADF;
903 if (bdev_read_only(bdev))
904 return -EPERM;
905 err = blk_validate_byte_range(bdev, start, len);
906 if (err)
907 return err;
908
909 err = filemap_invalidate_pages(bdev->bd_mapping, start,
910 start + len - 1, nowait);
911 if (err)
912 return err;
913
914 while (true) {
915 bio = blk_alloc_discard_bio(bdev, §or, &nr_sects, gfp);
916 if (!bio)
917 break;
918 if (nowait) {
919 /*
920 * Don't allow multi-bio non-blocking submissions as
921 * subsequent bios may fail but we won't get a direct
922 * indication of that. Normally, the caller should
923 * retry from a blocking context.
924 */
925 if (unlikely(nr_sects)) {
926 bio_put(bio);
927 return -EAGAIN;
928 }
929 bio->bi_opf |= REQ_NOWAIT;
930 }
931
932 prev = bio_chain_and_submit(prev, bio);
933 }
934 if (unlikely(!prev))
935 return -EAGAIN;
936 if (unlikely(nr_sects))
937 bic->res = -EAGAIN;
938
939 prev->bi_private = cmd;
940 prev->bi_end_io = bio_cmd_bio_end_io;
941 submit_bio(prev);
942 return -EIOCBQUEUED;
943 }
944
blkdev_uring_cmd(struct io_uring_cmd * cmd,unsigned int issue_flags)945 int blkdev_uring_cmd(struct io_uring_cmd *cmd, unsigned int issue_flags)
946 {
947 struct block_device *bdev = I_BDEV(cmd->file->f_mapping->host);
948 struct blk_iou_cmd *bic = io_uring_cmd_to_pdu(cmd, struct blk_iou_cmd);
949 const struct io_uring_sqe *sqe = cmd->sqe;
950 u32 cmd_op = cmd->cmd_op;
951 uint64_t start, len;
952
953 if (unlikely(sqe->ioprio || sqe->__pad1 || sqe->len ||
954 sqe->rw_flags || sqe->file_index))
955 return -EINVAL;
956
957 bic->res = 0;
958 bic->nowait = issue_flags & IO_URING_F_NONBLOCK;
959
960 start = READ_ONCE(sqe->addr);
961 len = READ_ONCE(sqe->addr3);
962
963 switch (cmd_op) {
964 case BLOCK_URING_CMD_DISCARD:
965 return blkdev_cmd_discard(cmd, bdev, start, len, bic->nowait);
966 }
967 return -EINVAL;
968 }
969