Loading...
1#include <linux/capability.h>
2#include <linux/blkdev.h>
3#include <linux/export.h>
4#include <linux/gfp.h>
5#include <linux/blkpg.h>
6#include <linux/hdreg.h>
7#include <linux/badblocks.h>
8#include <linux/backing-dev.h>
9#include <linux/fs.h>
10#include <linux/blktrace_api.h>
11#include <linux/pr.h>
12#include <asm/uaccess.h>
13
14static int blkpg_ioctl(struct block_device *bdev, struct blkpg_ioctl_arg __user *arg)
15{
16 struct block_device *bdevp;
17 struct gendisk *disk;
18 struct hd_struct *part, *lpart;
19 struct blkpg_ioctl_arg a;
20 struct blkpg_partition p;
21 struct disk_part_iter piter;
22 long long start, length;
23 int partno;
24
25 if (!capable(CAP_SYS_ADMIN))
26 return -EACCES;
27 if (copy_from_user(&a, arg, sizeof(struct blkpg_ioctl_arg)))
28 return -EFAULT;
29 if (copy_from_user(&p, a.data, sizeof(struct blkpg_partition)))
30 return -EFAULT;
31 disk = bdev->bd_disk;
32 if (bdev != bdev->bd_contains)
33 return -EINVAL;
34 partno = p.pno;
35 if (partno <= 0)
36 return -EINVAL;
37 switch (a.op) {
38 case BLKPG_ADD_PARTITION:
39 start = p.start >> 9;
40 length = p.length >> 9;
41 /* check for fit in a hd_struct */
42 if (sizeof(sector_t) == sizeof(long) &&
43 sizeof(long long) > sizeof(long)) {
44 long pstart = start, plength = length;
45 if (pstart != start || plength != length
46 || pstart < 0 || plength < 0 || partno > 65535)
47 return -EINVAL;
48 }
49
50 mutex_lock(&bdev->bd_mutex);
51
52 /* overlap? */
53 disk_part_iter_init(&piter, disk,
54 DISK_PITER_INCL_EMPTY);
55 while ((part = disk_part_iter_next(&piter))) {
56 if (!(start + length <= part->start_sect ||
57 start >= part->start_sect + part->nr_sects)) {
58 disk_part_iter_exit(&piter);
59 mutex_unlock(&bdev->bd_mutex);
60 return -EBUSY;
61 }
62 }
63 disk_part_iter_exit(&piter);
64
65 /* all seems OK */
66 part = add_partition(disk, partno, start, length,
67 ADDPART_FLAG_NONE, NULL);
68 mutex_unlock(&bdev->bd_mutex);
69 return PTR_ERR_OR_ZERO(part);
70 case BLKPG_DEL_PARTITION:
71 part = disk_get_part(disk, partno);
72 if (!part)
73 return -ENXIO;
74
75 bdevp = bdget(part_devt(part));
76 disk_put_part(part);
77 if (!bdevp)
78 return -ENOMEM;
79
80 mutex_lock(&bdevp->bd_mutex);
81 if (bdevp->bd_openers) {
82 mutex_unlock(&bdevp->bd_mutex);
83 bdput(bdevp);
84 return -EBUSY;
85 }
86 /* all seems OK */
87 fsync_bdev(bdevp);
88 invalidate_bdev(bdevp);
89
90 mutex_lock_nested(&bdev->bd_mutex, 1);
91 delete_partition(disk, partno);
92 mutex_unlock(&bdev->bd_mutex);
93 mutex_unlock(&bdevp->bd_mutex);
94 bdput(bdevp);
95
96 return 0;
97 case BLKPG_RESIZE_PARTITION:
98 start = p.start >> 9;
99 /* new length of partition in bytes */
100 length = p.length >> 9;
101 /* check for fit in a hd_struct */
102 if (sizeof(sector_t) == sizeof(long) &&
103 sizeof(long long) > sizeof(long)) {
104 long pstart = start, plength = length;
105 if (pstart != start || plength != length
106 || pstart < 0 || plength < 0)
107 return -EINVAL;
108 }
109 part = disk_get_part(disk, partno);
110 if (!part)
111 return -ENXIO;
112 bdevp = bdget(part_devt(part));
113 if (!bdevp) {
114 disk_put_part(part);
115 return -ENOMEM;
116 }
117 mutex_lock(&bdevp->bd_mutex);
118 mutex_lock_nested(&bdev->bd_mutex, 1);
119 if (start != part->start_sect) {
120 mutex_unlock(&bdevp->bd_mutex);
121 mutex_unlock(&bdev->bd_mutex);
122 bdput(bdevp);
123 disk_put_part(part);
124 return -EINVAL;
125 }
126 /* overlap? */
127 disk_part_iter_init(&piter, disk,
128 DISK_PITER_INCL_EMPTY);
129 while ((lpart = disk_part_iter_next(&piter))) {
130 if (lpart->partno != partno &&
131 !(start + length <= lpart->start_sect ||
132 start >= lpart->start_sect + lpart->nr_sects)
133 ) {
134 disk_part_iter_exit(&piter);
135 mutex_unlock(&bdevp->bd_mutex);
136 mutex_unlock(&bdev->bd_mutex);
137 bdput(bdevp);
138 disk_put_part(part);
139 return -EBUSY;
140 }
141 }
142 disk_part_iter_exit(&piter);
143 part_nr_sects_write(part, (sector_t)length);
144 i_size_write(bdevp->bd_inode, p.length);
145 mutex_unlock(&bdevp->bd_mutex);
146 mutex_unlock(&bdev->bd_mutex);
147 bdput(bdevp);
148 disk_put_part(part);
149 return 0;
150 default:
151 return -EINVAL;
152 }
153}
154
155/*
156 * This is an exported API for the block driver, and will not
157 * acquire bd_mutex. This API should be used in case that
158 * caller has held bd_mutex already.
159 */
160int __blkdev_reread_part(struct block_device *bdev)
161{
162 struct gendisk *disk = bdev->bd_disk;
163
164 if (!disk_part_scan_enabled(disk) || bdev != bdev->bd_contains)
165 return -EINVAL;
166 if (!capable(CAP_SYS_ADMIN))
167 return -EACCES;
168
169 lockdep_assert_held(&bdev->bd_mutex);
170
171 return rescan_partitions(disk, bdev);
172}
173EXPORT_SYMBOL(__blkdev_reread_part);
174
175/*
176 * This is an exported API for the block driver, and will
177 * try to acquire bd_mutex. If bd_mutex has been held already
178 * in current context, please call __blkdev_reread_part().
179 *
180 * Make sure the held locks in current context aren't required
181 * in open()/close() handler and I/O path for avoiding ABBA deadlock:
182 * - bd_mutex is held before calling block driver's open/close
183 * handler
184 * - reading partition table may submit I/O to the block device
185 */
186int blkdev_reread_part(struct block_device *bdev)
187{
188 int res;
189
190 mutex_lock(&bdev->bd_mutex);
191 res = __blkdev_reread_part(bdev);
192 mutex_unlock(&bdev->bd_mutex);
193
194 return res;
195}
196EXPORT_SYMBOL(blkdev_reread_part);
197
198static int blk_ioctl_discard(struct block_device *bdev, fmode_t mode,
199 unsigned long arg, unsigned long flags)
200{
201 uint64_t range[2];
202 uint64_t start, len;
203
204 if (!(mode & FMODE_WRITE))
205 return -EBADF;
206
207 if (copy_from_user(range, (void __user *)arg, sizeof(range)))
208 return -EFAULT;
209
210 start = range[0];
211 len = range[1];
212
213 if (start & 511)
214 return -EINVAL;
215 if (len & 511)
216 return -EINVAL;
217 start >>= 9;
218 len >>= 9;
219
220 if (start + len > (i_size_read(bdev->bd_inode) >> 9))
221 return -EINVAL;
222 return blkdev_issue_discard(bdev, start, len, GFP_KERNEL, flags);
223}
224
225static int blk_ioctl_zeroout(struct block_device *bdev, fmode_t mode,
226 unsigned long arg)
227{
228 uint64_t range[2];
229 uint64_t start, len;
230
231 if (!(mode & FMODE_WRITE))
232 return -EBADF;
233
234 if (copy_from_user(range, (void __user *)arg, sizeof(range)))
235 return -EFAULT;
236
237 start = range[0];
238 len = range[1];
239
240 if (start & 511)
241 return -EINVAL;
242 if (len & 511)
243 return -EINVAL;
244 start >>= 9;
245 len >>= 9;
246
247 if (start + len > (i_size_read(bdev->bd_inode) >> 9))
248 return -EINVAL;
249
250 return blkdev_issue_zeroout(bdev, start, len, GFP_KERNEL, false);
251}
252
253static int put_ushort(unsigned long arg, unsigned short val)
254{
255 return put_user(val, (unsigned short __user *)arg);
256}
257
258static int put_int(unsigned long arg, int val)
259{
260 return put_user(val, (int __user *)arg);
261}
262
263static int put_uint(unsigned long arg, unsigned int val)
264{
265 return put_user(val, (unsigned int __user *)arg);
266}
267
268static int put_long(unsigned long arg, long val)
269{
270 return put_user(val, (long __user *)arg);
271}
272
273static int put_ulong(unsigned long arg, unsigned long val)
274{
275 return put_user(val, (unsigned long __user *)arg);
276}
277
278static int put_u64(unsigned long arg, u64 val)
279{
280 return put_user(val, (u64 __user *)arg);
281}
282
283int __blkdev_driver_ioctl(struct block_device *bdev, fmode_t mode,
284 unsigned cmd, unsigned long arg)
285{
286 struct gendisk *disk = bdev->bd_disk;
287
288 if (disk->fops->ioctl)
289 return disk->fops->ioctl(bdev, mode, cmd, arg);
290
291 return -ENOTTY;
292}
293/*
294 * For the record: _GPL here is only because somebody decided to slap it
295 * on the previous export. Sheer idiocy, since it wasn't copyrightable
296 * at all and could be open-coded without any exports by anybody who cares.
297 */
298EXPORT_SYMBOL_GPL(__blkdev_driver_ioctl);
299
300static int blkdev_pr_register(struct block_device *bdev,
301 struct pr_registration __user *arg)
302{
303 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
304 struct pr_registration reg;
305
306 if (!capable(CAP_SYS_ADMIN))
307 return -EPERM;
308 if (!ops || !ops->pr_register)
309 return -EOPNOTSUPP;
310 if (copy_from_user(®, arg, sizeof(reg)))
311 return -EFAULT;
312
313 if (reg.flags & ~PR_FL_IGNORE_KEY)
314 return -EOPNOTSUPP;
315 return ops->pr_register(bdev, reg.old_key, reg.new_key, reg.flags);
316}
317
318static int blkdev_pr_reserve(struct block_device *bdev,
319 struct pr_reservation __user *arg)
320{
321 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
322 struct pr_reservation rsv;
323
324 if (!capable(CAP_SYS_ADMIN))
325 return -EPERM;
326 if (!ops || !ops->pr_reserve)
327 return -EOPNOTSUPP;
328 if (copy_from_user(&rsv, arg, sizeof(rsv)))
329 return -EFAULT;
330
331 if (rsv.flags & ~PR_FL_IGNORE_KEY)
332 return -EOPNOTSUPP;
333 return ops->pr_reserve(bdev, rsv.key, rsv.type, rsv.flags);
334}
335
336static int blkdev_pr_release(struct block_device *bdev,
337 struct pr_reservation __user *arg)
338{
339 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
340 struct pr_reservation rsv;
341
342 if (!capable(CAP_SYS_ADMIN))
343 return -EPERM;
344 if (!ops || !ops->pr_release)
345 return -EOPNOTSUPP;
346 if (copy_from_user(&rsv, arg, sizeof(rsv)))
347 return -EFAULT;
348
349 if (rsv.flags)
350 return -EOPNOTSUPP;
351 return ops->pr_release(bdev, rsv.key, rsv.type);
352}
353
354static int blkdev_pr_preempt(struct block_device *bdev,
355 struct pr_preempt __user *arg, bool abort)
356{
357 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
358 struct pr_preempt p;
359
360 if (!capable(CAP_SYS_ADMIN))
361 return -EPERM;
362 if (!ops || !ops->pr_preempt)
363 return -EOPNOTSUPP;
364 if (copy_from_user(&p, arg, sizeof(p)))
365 return -EFAULT;
366
367 if (p.flags)
368 return -EOPNOTSUPP;
369 return ops->pr_preempt(bdev, p.old_key, p.new_key, p.type, abort);
370}
371
372static int blkdev_pr_clear(struct block_device *bdev,
373 struct pr_clear __user *arg)
374{
375 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
376 struct pr_clear c;
377
378 if (!capable(CAP_SYS_ADMIN))
379 return -EPERM;
380 if (!ops || !ops->pr_clear)
381 return -EOPNOTSUPP;
382 if (copy_from_user(&c, arg, sizeof(c)))
383 return -EFAULT;
384
385 if (c.flags)
386 return -EOPNOTSUPP;
387 return ops->pr_clear(bdev, c.key);
388}
389
390/*
391 * Is it an unrecognized ioctl? The correct returns are either
392 * ENOTTY (final) or ENOIOCTLCMD ("I don't know this one, try a
393 * fallback"). ENOIOCTLCMD gets turned into ENOTTY by the ioctl
394 * code before returning.
395 *
396 * Confused drivers sometimes return EINVAL, which is wrong. It
397 * means "I understood the ioctl command, but the parameters to
398 * it were wrong".
399 *
400 * We should aim to just fix the broken drivers, the EINVAL case
401 * should go away.
402 */
403static inline int is_unrecognized_ioctl(int ret)
404{
405 return ret == -EINVAL ||
406 ret == -ENOTTY ||
407 ret == -ENOIOCTLCMD;
408}
409
410#ifdef CONFIG_FS_DAX
411bool blkdev_dax_capable(struct block_device *bdev)
412{
413 struct gendisk *disk = bdev->bd_disk;
414
415 if (!disk->fops->direct_access)
416 return false;
417
418 /*
419 * If the partition is not aligned on a page boundary, we can't
420 * do dax I/O to it.
421 */
422 if ((bdev->bd_part->start_sect % (PAGE_SIZE / 512))
423 || (bdev->bd_part->nr_sects % (PAGE_SIZE / 512)))
424 return false;
425
426 /*
427 * If the device has known bad blocks, force all I/O through the
428 * driver / page cache.
429 *
430 * TODO: support finer grained dax error handling
431 */
432 if (disk->bb && disk->bb->count)
433 return false;
434
435 return true;
436}
437#endif
438
439static int blkdev_flushbuf(struct block_device *bdev, fmode_t mode,
440 unsigned cmd, unsigned long arg)
441{
442 int ret;
443
444 if (!capable(CAP_SYS_ADMIN))
445 return -EACCES;
446
447 ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
448 if (!is_unrecognized_ioctl(ret))
449 return ret;
450
451 fsync_bdev(bdev);
452 invalidate_bdev(bdev);
453 return 0;
454}
455
456static int blkdev_roset(struct block_device *bdev, fmode_t mode,
457 unsigned cmd, unsigned long arg)
458{
459 int ret, n;
460
461 ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
462 if (!is_unrecognized_ioctl(ret))
463 return ret;
464 if (!capable(CAP_SYS_ADMIN))
465 return -EACCES;
466 if (get_user(n, (int __user *)arg))
467 return -EFAULT;
468 set_device_ro(bdev, n);
469 return 0;
470}
471
472static int blkdev_getgeo(struct block_device *bdev,
473 struct hd_geometry __user *argp)
474{
475 struct gendisk *disk = bdev->bd_disk;
476 struct hd_geometry geo;
477 int ret;
478
479 if (!argp)
480 return -EINVAL;
481 if (!disk->fops->getgeo)
482 return -ENOTTY;
483
484 /*
485 * We need to set the startsect first, the driver may
486 * want to override it.
487 */
488 memset(&geo, 0, sizeof(geo));
489 geo.start = get_start_sect(bdev);
490 ret = disk->fops->getgeo(bdev, &geo);
491 if (ret)
492 return ret;
493 if (copy_to_user(argp, &geo, sizeof(geo)))
494 return -EFAULT;
495 return 0;
496}
497
498/* set the logical block size */
499static int blkdev_bszset(struct block_device *bdev, fmode_t mode,
500 int __user *argp)
501{
502 int ret, n;
503
504 if (!capable(CAP_SYS_ADMIN))
505 return -EACCES;
506 if (!argp)
507 return -EINVAL;
508 if (get_user(n, argp))
509 return -EFAULT;
510
511 if (!(mode & FMODE_EXCL)) {
512 bdgrab(bdev);
513 if (blkdev_get(bdev, mode | FMODE_EXCL, &bdev) < 0)
514 return -EBUSY;
515 }
516
517 ret = set_blocksize(bdev, n);
518 if (!(mode & FMODE_EXCL))
519 blkdev_put(bdev, mode | FMODE_EXCL);
520 return ret;
521}
522
523/*
524 * always keep this in sync with compat_blkdev_ioctl()
525 */
526int blkdev_ioctl(struct block_device *bdev, fmode_t mode, unsigned cmd,
527 unsigned long arg)
528{
529 struct backing_dev_info *bdi;
530 void __user *argp = (void __user *)arg;
531 loff_t size;
532 unsigned int max_sectors;
533
534 switch (cmd) {
535 case BLKFLSBUF:
536 return blkdev_flushbuf(bdev, mode, cmd, arg);
537 case BLKROSET:
538 return blkdev_roset(bdev, mode, cmd, arg);
539 case BLKDISCARD:
540 return blk_ioctl_discard(bdev, mode, arg, 0);
541 case BLKSECDISCARD:
542 return blk_ioctl_discard(bdev, mode, arg,
543 BLKDEV_DISCARD_SECURE);
544 case BLKZEROOUT:
545 return blk_ioctl_zeroout(bdev, mode, arg);
546 case HDIO_GETGEO:
547 return blkdev_getgeo(bdev, argp);
548 case BLKRAGET:
549 case BLKFRAGET:
550 if (!arg)
551 return -EINVAL;
552 bdi = blk_get_backing_dev_info(bdev);
553 return put_long(arg, (bdi->ra_pages * PAGE_SIZE) / 512);
554 case BLKROGET:
555 return put_int(arg, bdev_read_only(bdev) != 0);
556 case BLKBSZGET: /* get block device soft block size (cf. BLKSSZGET) */
557 return put_int(arg, block_size(bdev));
558 case BLKSSZGET: /* get block device logical block size */
559 return put_int(arg, bdev_logical_block_size(bdev));
560 case BLKPBSZGET: /* get block device physical block size */
561 return put_uint(arg, bdev_physical_block_size(bdev));
562 case BLKIOMIN:
563 return put_uint(arg, bdev_io_min(bdev));
564 case BLKIOOPT:
565 return put_uint(arg, bdev_io_opt(bdev));
566 case BLKALIGNOFF:
567 return put_int(arg, bdev_alignment_offset(bdev));
568 case BLKDISCARDZEROES:
569 return put_uint(arg, bdev_discard_zeroes_data(bdev));
570 case BLKSECTGET:
571 max_sectors = min_t(unsigned int, USHRT_MAX,
572 queue_max_sectors(bdev_get_queue(bdev)));
573 return put_ushort(arg, max_sectors);
574 case BLKROTATIONAL:
575 return put_ushort(arg, !blk_queue_nonrot(bdev_get_queue(bdev)));
576 case BLKRASET:
577 case BLKFRASET:
578 if(!capable(CAP_SYS_ADMIN))
579 return -EACCES;
580 bdi = blk_get_backing_dev_info(bdev);
581 bdi->ra_pages = (arg * 512) / PAGE_SIZE;
582 return 0;
583 case BLKBSZSET:
584 return blkdev_bszset(bdev, mode, argp);
585 case BLKPG:
586 return blkpg_ioctl(bdev, argp);
587 case BLKRRPART:
588 return blkdev_reread_part(bdev);
589 case BLKGETSIZE:
590 size = i_size_read(bdev->bd_inode);
591 if ((size >> 9) > ~0UL)
592 return -EFBIG;
593 return put_ulong(arg, size >> 9);
594 case BLKGETSIZE64:
595 return put_u64(arg, i_size_read(bdev->bd_inode));
596 case BLKTRACESTART:
597 case BLKTRACESTOP:
598 case BLKTRACESETUP:
599 case BLKTRACETEARDOWN:
600 return blk_trace_ioctl(bdev, cmd, argp);
601 case BLKDAXGET:
602 return put_int(arg, !!(bdev->bd_inode->i_flags & S_DAX));
603 break;
604 case IOC_PR_REGISTER:
605 return blkdev_pr_register(bdev, argp);
606 case IOC_PR_RESERVE:
607 return blkdev_pr_reserve(bdev, argp);
608 case IOC_PR_RELEASE:
609 return blkdev_pr_release(bdev, argp);
610 case IOC_PR_PREEMPT:
611 return blkdev_pr_preempt(bdev, argp, false);
612 case IOC_PR_PREEMPT_ABORT:
613 return blkdev_pr_preempt(bdev, argp, true);
614 case IOC_PR_CLEAR:
615 return blkdev_pr_clear(bdev, argp);
616 default:
617 return __blkdev_driver_ioctl(bdev, mode, cmd, arg);
618 }
619}
620EXPORT_SYMBOL_GPL(blkdev_ioctl);
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
16static 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 sector_t start, length, capacity, end;
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 if (p.start < 0 || p.length <= 0 || p.start + p.length < 0)
37 return -EINVAL;
38 /* Check that the partition is aligned to the block size */
39 if (!IS_ALIGNED(p.start | p.length, bdev_logical_block_size(bdev)))
40 return -EINVAL;
41
42 start = p.start >> SECTOR_SHIFT;
43 length = p.length >> SECTOR_SHIFT;
44 capacity = get_capacity(disk);
45
46 if (check_add_overflow(start, length, &end))
47 return -EINVAL;
48
49 if (start >= capacity || end > capacity)
50 return -EINVAL;
51
52 switch (op) {
53 case BLKPG_ADD_PARTITION:
54 return bdev_add_partition(disk, p.pno, start, length);
55 case BLKPG_RESIZE_PARTITION:
56 return bdev_resize_partition(disk, p.pno, start, length);
57 default:
58 return -EINVAL;
59 }
60}
61
62static int blkpg_ioctl(struct block_device *bdev,
63 struct blkpg_ioctl_arg __user *arg)
64{
65 struct blkpg_partition __user *udata;
66 int op;
67
68 if (get_user(op, &arg->op) || get_user(udata, &arg->data))
69 return -EFAULT;
70
71 return blkpg_do_ioctl(bdev, udata, op);
72}
73
74#ifdef CONFIG_COMPAT
75struct compat_blkpg_ioctl_arg {
76 compat_int_t op;
77 compat_int_t flags;
78 compat_int_t datalen;
79 compat_caddr_t data;
80};
81
82static int compat_blkpg_ioctl(struct block_device *bdev,
83 struct compat_blkpg_ioctl_arg __user *arg)
84{
85 compat_caddr_t udata;
86 int op;
87
88 if (get_user(op, &arg->op) || get_user(udata, &arg->data))
89 return -EFAULT;
90
91 return blkpg_do_ioctl(bdev, compat_ptr(udata), op);
92}
93#endif
94
95static int blk_ioctl_discard(struct block_device *bdev, blk_mode_t mode,
96 unsigned long arg)
97{
98 uint64_t range[2];
99 uint64_t start, len, end;
100 struct inode *inode = bdev->bd_inode;
101 int err;
102
103 if (!(mode & BLK_OPEN_WRITE))
104 return -EBADF;
105
106 if (!bdev_max_discard_sectors(bdev))
107 return -EOPNOTSUPP;
108
109 if (copy_from_user(range, (void __user *)arg, sizeof(range)))
110 return -EFAULT;
111
112 start = range[0];
113 len = range[1];
114
115 if (start & 511)
116 return -EINVAL;
117 if (len & 511)
118 return -EINVAL;
119
120 if (check_add_overflow(start, len, &end) ||
121 end > bdev_nr_bytes(bdev))
122 return -EINVAL;
123
124 filemap_invalidate_lock(inode->i_mapping);
125 err = truncate_bdev_range(bdev, mode, start, start + len - 1);
126 if (err)
127 goto fail;
128 err = blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
129fail:
130 filemap_invalidate_unlock(inode->i_mapping);
131 return err;
132}
133
134static int blk_ioctl_secure_erase(struct block_device *bdev, blk_mode_t mode,
135 void __user *argp)
136{
137 uint64_t start, len;
138 uint64_t range[2];
139 int err;
140
141 if (!(mode & BLK_OPEN_WRITE))
142 return -EBADF;
143 if (!bdev_max_secure_erase_sectors(bdev))
144 return -EOPNOTSUPP;
145 if (copy_from_user(range, argp, sizeof(range)))
146 return -EFAULT;
147
148 start = range[0];
149 len = range[1];
150 if ((start & 511) || (len & 511))
151 return -EINVAL;
152 if (start + len > bdev_nr_bytes(bdev))
153 return -EINVAL;
154
155 filemap_invalidate_lock(bdev->bd_inode->i_mapping);
156 err = truncate_bdev_range(bdev, mode, start, start + len - 1);
157 if (!err)
158 err = blkdev_issue_secure_erase(bdev, start >> 9, len >> 9,
159 GFP_KERNEL);
160 filemap_invalidate_unlock(bdev->bd_inode->i_mapping);
161 return err;
162}
163
164
165static int blk_ioctl_zeroout(struct block_device *bdev, blk_mode_t mode,
166 unsigned long arg)
167{
168 uint64_t range[2];
169 uint64_t start, end, len;
170 struct inode *inode = bdev->bd_inode;
171 int err;
172
173 if (!(mode & BLK_OPEN_WRITE))
174 return -EBADF;
175
176 if (copy_from_user(range, (void __user *)arg, sizeof(range)))
177 return -EFAULT;
178
179 start = range[0];
180 len = range[1];
181 end = start + len - 1;
182
183 if (start & 511)
184 return -EINVAL;
185 if (len & 511)
186 return -EINVAL;
187 if (end >= (uint64_t)bdev_nr_bytes(bdev))
188 return -EINVAL;
189 if (end < start)
190 return -EINVAL;
191
192 /* Invalidate the page cache, including dirty pages */
193 filemap_invalidate_lock(inode->i_mapping);
194 err = truncate_bdev_range(bdev, mode, start, end);
195 if (err)
196 goto fail;
197
198 err = blkdev_issue_zeroout(bdev, start >> 9, len >> 9, GFP_KERNEL,
199 BLKDEV_ZERO_NOUNMAP);
200
201fail:
202 filemap_invalidate_unlock(inode->i_mapping);
203 return err;
204}
205
206static int put_ushort(unsigned short __user *argp, unsigned short val)
207{
208 return put_user(val, argp);
209}
210
211static int put_int(int __user *argp, int val)
212{
213 return put_user(val, argp);
214}
215
216static int put_uint(unsigned int __user *argp, unsigned int val)
217{
218 return put_user(val, argp);
219}
220
221static int put_long(long __user *argp, long val)
222{
223 return put_user(val, argp);
224}
225
226static int put_ulong(unsigned long __user *argp, unsigned long val)
227{
228 return put_user(val, argp);
229}
230
231static int put_u64(u64 __user *argp, u64 val)
232{
233 return put_user(val, argp);
234}
235
236#ifdef CONFIG_COMPAT
237static int compat_put_long(compat_long_t __user *argp, long val)
238{
239 return put_user(val, argp);
240}
241
242static int compat_put_ulong(compat_ulong_t __user *argp, compat_ulong_t val)
243{
244 return put_user(val, argp);
245}
246#endif
247
248#ifdef CONFIG_COMPAT
249/*
250 * This is the equivalent of compat_ptr_ioctl(), to be used by block
251 * drivers that implement only commands that are completely compatible
252 * between 32-bit and 64-bit user space
253 */
254int blkdev_compat_ptr_ioctl(struct block_device *bdev, blk_mode_t mode,
255 unsigned cmd, unsigned long arg)
256{
257 struct gendisk *disk = bdev->bd_disk;
258
259 if (disk->fops->ioctl)
260 return disk->fops->ioctl(bdev, mode, cmd,
261 (unsigned long)compat_ptr(arg));
262
263 return -ENOIOCTLCMD;
264}
265EXPORT_SYMBOL(blkdev_compat_ptr_ioctl);
266#endif
267
268static bool blkdev_pr_allowed(struct block_device *bdev, blk_mode_t mode)
269{
270 /* no sense to make reservations for partitions */
271 if (bdev_is_partition(bdev))
272 return false;
273
274 if (capable(CAP_SYS_ADMIN))
275 return true;
276 /*
277 * Only allow unprivileged reservations if the file descriptor is open
278 * for writing.
279 */
280 return mode & BLK_OPEN_WRITE;
281}
282
283static int blkdev_pr_register(struct block_device *bdev, blk_mode_t mode,
284 struct pr_registration __user *arg)
285{
286 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
287 struct pr_registration reg;
288
289 if (!blkdev_pr_allowed(bdev, mode))
290 return -EPERM;
291 if (!ops || !ops->pr_register)
292 return -EOPNOTSUPP;
293 if (copy_from_user(®, arg, sizeof(reg)))
294 return -EFAULT;
295
296 if (reg.flags & ~PR_FL_IGNORE_KEY)
297 return -EOPNOTSUPP;
298 return ops->pr_register(bdev, reg.old_key, reg.new_key, reg.flags);
299}
300
301static int blkdev_pr_reserve(struct block_device *bdev, blk_mode_t mode,
302 struct pr_reservation __user *arg)
303{
304 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
305 struct pr_reservation rsv;
306
307 if (!blkdev_pr_allowed(bdev, mode))
308 return -EPERM;
309 if (!ops || !ops->pr_reserve)
310 return -EOPNOTSUPP;
311 if (copy_from_user(&rsv, arg, sizeof(rsv)))
312 return -EFAULT;
313
314 if (rsv.flags & ~PR_FL_IGNORE_KEY)
315 return -EOPNOTSUPP;
316 return ops->pr_reserve(bdev, rsv.key, rsv.type, rsv.flags);
317}
318
319static int blkdev_pr_release(struct block_device *bdev, blk_mode_t mode,
320 struct pr_reservation __user *arg)
321{
322 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
323 struct pr_reservation rsv;
324
325 if (!blkdev_pr_allowed(bdev, mode))
326 return -EPERM;
327 if (!ops || !ops->pr_release)
328 return -EOPNOTSUPP;
329 if (copy_from_user(&rsv, arg, sizeof(rsv)))
330 return -EFAULT;
331
332 if (rsv.flags)
333 return -EOPNOTSUPP;
334 return ops->pr_release(bdev, rsv.key, rsv.type);
335}
336
337static int blkdev_pr_preempt(struct block_device *bdev, blk_mode_t mode,
338 struct pr_preempt __user *arg, bool abort)
339{
340 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
341 struct pr_preempt p;
342
343 if (!blkdev_pr_allowed(bdev, mode))
344 return -EPERM;
345 if (!ops || !ops->pr_preempt)
346 return -EOPNOTSUPP;
347 if (copy_from_user(&p, arg, sizeof(p)))
348 return -EFAULT;
349
350 if (p.flags)
351 return -EOPNOTSUPP;
352 return ops->pr_preempt(bdev, p.old_key, p.new_key, p.type, abort);
353}
354
355static int blkdev_pr_clear(struct block_device *bdev, blk_mode_t mode,
356 struct pr_clear __user *arg)
357{
358 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
359 struct pr_clear c;
360
361 if (!blkdev_pr_allowed(bdev, mode))
362 return -EPERM;
363 if (!ops || !ops->pr_clear)
364 return -EOPNOTSUPP;
365 if (copy_from_user(&c, arg, sizeof(c)))
366 return -EFAULT;
367
368 if (c.flags)
369 return -EOPNOTSUPP;
370 return ops->pr_clear(bdev, c.key);
371}
372
373static int blkdev_flushbuf(struct block_device *bdev, unsigned cmd,
374 unsigned long arg)
375{
376 if (!capable(CAP_SYS_ADMIN))
377 return -EACCES;
378
379 mutex_lock(&bdev->bd_holder_lock);
380 if (bdev->bd_holder_ops && bdev->bd_holder_ops->sync)
381 bdev->bd_holder_ops->sync(bdev);
382 else {
383 mutex_unlock(&bdev->bd_holder_lock);
384 sync_blockdev(bdev);
385 }
386
387 invalidate_bdev(bdev);
388 return 0;
389}
390
391static int blkdev_roset(struct block_device *bdev, unsigned cmd,
392 unsigned long arg)
393{
394 int ret, n;
395
396 if (!capable(CAP_SYS_ADMIN))
397 return -EACCES;
398
399 if (get_user(n, (int __user *)arg))
400 return -EFAULT;
401 if (bdev->bd_disk->fops->set_read_only) {
402 ret = bdev->bd_disk->fops->set_read_only(bdev, n);
403 if (ret)
404 return ret;
405 }
406 bdev->bd_read_only = n;
407 return 0;
408}
409
410static int blkdev_getgeo(struct block_device *bdev,
411 struct hd_geometry __user *argp)
412{
413 struct gendisk *disk = bdev->bd_disk;
414 struct hd_geometry geo;
415 int ret;
416
417 if (!argp)
418 return -EINVAL;
419 if (!disk->fops->getgeo)
420 return -ENOTTY;
421
422 /*
423 * We need to set the startsect first, the driver may
424 * want to override it.
425 */
426 memset(&geo, 0, sizeof(geo));
427 geo.start = get_start_sect(bdev);
428 ret = disk->fops->getgeo(bdev, &geo);
429 if (ret)
430 return ret;
431 if (copy_to_user(argp, &geo, sizeof(geo)))
432 return -EFAULT;
433 return 0;
434}
435
436#ifdef CONFIG_COMPAT
437struct compat_hd_geometry {
438 unsigned char heads;
439 unsigned char sectors;
440 unsigned short cylinders;
441 u32 start;
442};
443
444static int compat_hdio_getgeo(struct block_device *bdev,
445 struct compat_hd_geometry __user *ugeo)
446{
447 struct gendisk *disk = bdev->bd_disk;
448 struct hd_geometry geo;
449 int ret;
450
451 if (!ugeo)
452 return -EINVAL;
453 if (!disk->fops->getgeo)
454 return -ENOTTY;
455
456 memset(&geo, 0, sizeof(geo));
457 /*
458 * We need to set the startsect first, the driver may
459 * want to override it.
460 */
461 geo.start = get_start_sect(bdev);
462 ret = disk->fops->getgeo(bdev, &geo);
463 if (ret)
464 return ret;
465
466 ret = copy_to_user(ugeo, &geo, 4);
467 ret |= put_user(geo.start, &ugeo->start);
468 if (ret)
469 ret = -EFAULT;
470
471 return ret;
472}
473#endif
474
475/* set the logical block size */
476static int blkdev_bszset(struct block_device *bdev, blk_mode_t mode,
477 int __user *argp)
478{
479 int ret, n;
480 struct file *file;
481
482 if (!capable(CAP_SYS_ADMIN))
483 return -EACCES;
484 if (!argp)
485 return -EINVAL;
486 if (get_user(n, argp))
487 return -EFAULT;
488
489 if (mode & BLK_OPEN_EXCL)
490 return set_blocksize(bdev, n);
491
492 file = bdev_file_open_by_dev(bdev->bd_dev, mode, &bdev, NULL);
493 if (IS_ERR(file))
494 return -EBUSY;
495 ret = set_blocksize(bdev, n);
496 fput(file);
497 return ret;
498}
499
500/*
501 * Common commands that are handled the same way on native and compat
502 * user space. Note the separate arg/argp parameters that are needed
503 * to deal with the compat_ptr() conversion.
504 */
505static int blkdev_common_ioctl(struct block_device *bdev, blk_mode_t mode,
506 unsigned int cmd, unsigned long arg,
507 void __user *argp)
508{
509 unsigned int max_sectors;
510
511 switch (cmd) {
512 case BLKFLSBUF:
513 return blkdev_flushbuf(bdev, cmd, arg);
514 case BLKROSET:
515 return blkdev_roset(bdev, cmd, arg);
516 case BLKDISCARD:
517 return blk_ioctl_discard(bdev, mode, arg);
518 case BLKSECDISCARD:
519 return blk_ioctl_secure_erase(bdev, mode, argp);
520 case BLKZEROOUT:
521 return blk_ioctl_zeroout(bdev, mode, arg);
522 case BLKGETDISKSEQ:
523 return put_u64(argp, bdev->bd_disk->diskseq);
524 case BLKREPORTZONE:
525 return blkdev_report_zones_ioctl(bdev, cmd, arg);
526 case BLKRESETZONE:
527 case BLKOPENZONE:
528 case BLKCLOSEZONE:
529 case BLKFINISHZONE:
530 return blkdev_zone_mgmt_ioctl(bdev, mode, cmd, arg);
531 case BLKGETZONESZ:
532 return put_uint(argp, bdev_zone_sectors(bdev));
533 case BLKGETNRZONES:
534 return put_uint(argp, bdev_nr_zones(bdev));
535 case BLKROGET:
536 return put_int(argp, bdev_read_only(bdev) != 0);
537 case BLKSSZGET: /* get block device logical block size */
538 return put_int(argp, bdev_logical_block_size(bdev));
539 case BLKPBSZGET: /* get block device physical block size */
540 return put_uint(argp, bdev_physical_block_size(bdev));
541 case BLKIOMIN:
542 return put_uint(argp, bdev_io_min(bdev));
543 case BLKIOOPT:
544 return put_uint(argp, bdev_io_opt(bdev));
545 case BLKALIGNOFF:
546 return put_int(argp, bdev_alignment_offset(bdev));
547 case BLKDISCARDZEROES:
548 return put_uint(argp, 0);
549 case BLKSECTGET:
550 max_sectors = min_t(unsigned int, USHRT_MAX,
551 queue_max_sectors(bdev_get_queue(bdev)));
552 return put_ushort(argp, max_sectors);
553 case BLKROTATIONAL:
554 return put_ushort(argp, !bdev_nonrot(bdev));
555 case BLKRASET:
556 case BLKFRASET:
557 if(!capable(CAP_SYS_ADMIN))
558 return -EACCES;
559 bdev->bd_disk->bdi->ra_pages = (arg * 512) / PAGE_SIZE;
560 return 0;
561 case BLKRRPART:
562 if (!capable(CAP_SYS_ADMIN))
563 return -EACCES;
564 if (bdev_is_partition(bdev))
565 return -EINVAL;
566 return disk_scan_partitions(bdev->bd_disk,
567 mode | BLK_OPEN_STRICT_SCAN);
568 case BLKTRACESTART:
569 case BLKTRACESTOP:
570 case BLKTRACETEARDOWN:
571 return blk_trace_ioctl(bdev, cmd, argp);
572 case IOC_PR_REGISTER:
573 return blkdev_pr_register(bdev, mode, argp);
574 case IOC_PR_RESERVE:
575 return blkdev_pr_reserve(bdev, mode, argp);
576 case IOC_PR_RELEASE:
577 return blkdev_pr_release(bdev, mode, argp);
578 case IOC_PR_PREEMPT:
579 return blkdev_pr_preempt(bdev, mode, argp, false);
580 case IOC_PR_PREEMPT_ABORT:
581 return blkdev_pr_preempt(bdev, mode, argp, true);
582 case IOC_PR_CLEAR:
583 return blkdev_pr_clear(bdev, mode, argp);
584 default:
585 return -ENOIOCTLCMD;
586 }
587}
588
589/*
590 * Always keep this in sync with compat_blkdev_ioctl()
591 * to handle all incompatible commands in both functions.
592 *
593 * New commands must be compatible and go into blkdev_common_ioctl
594 */
595long blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg)
596{
597 struct block_device *bdev = I_BDEV(file->f_mapping->host);
598 void __user *argp = (void __user *)arg;
599 blk_mode_t mode = file_to_blk_mode(file);
600 int ret;
601
602 switch (cmd) {
603 /* These need separate implementations for the data structure */
604 case HDIO_GETGEO:
605 return blkdev_getgeo(bdev, argp);
606 case BLKPG:
607 return blkpg_ioctl(bdev, argp);
608
609 /* Compat mode returns 32-bit data instead of 'long' */
610 case BLKRAGET:
611 case BLKFRAGET:
612 if (!argp)
613 return -EINVAL;
614 return put_long(argp,
615 (bdev->bd_disk->bdi->ra_pages * PAGE_SIZE) / 512);
616 case BLKGETSIZE:
617 if (bdev_nr_sectors(bdev) > ~0UL)
618 return -EFBIG;
619 return put_ulong(argp, bdev_nr_sectors(bdev));
620
621 /* The data is compatible, but the command number is different */
622 case BLKBSZGET: /* get block device soft block size (cf. BLKSSZGET) */
623 return put_int(argp, block_size(bdev));
624 case BLKBSZSET:
625 return blkdev_bszset(bdev, mode, argp);
626 case BLKGETSIZE64:
627 return put_u64(argp, bdev_nr_bytes(bdev));
628
629 /* Incompatible alignment on i386 */
630 case BLKTRACESETUP:
631 return blk_trace_ioctl(bdev, cmd, argp);
632 default:
633 break;
634 }
635
636 ret = blkdev_common_ioctl(bdev, mode, cmd, arg, argp);
637 if (ret != -ENOIOCTLCMD)
638 return ret;
639
640 if (!bdev->bd_disk->fops->ioctl)
641 return -ENOTTY;
642 return bdev->bd_disk->fops->ioctl(bdev, mode, cmd, arg);
643}
644
645#ifdef CONFIG_COMPAT
646
647#define BLKBSZGET_32 _IOR(0x12, 112, int)
648#define BLKBSZSET_32 _IOW(0x12, 113, int)
649#define BLKGETSIZE64_32 _IOR(0x12, 114, int)
650
651/* Most of the generic ioctls are handled in the normal fallback path.
652 This assumes the blkdev's low level compat_ioctl always returns
653 ENOIOCTLCMD for unknown ioctls. */
654long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg)
655{
656 int ret;
657 void __user *argp = compat_ptr(arg);
658 struct block_device *bdev = I_BDEV(file->f_mapping->host);
659 struct gendisk *disk = bdev->bd_disk;
660 blk_mode_t mode = file_to_blk_mode(file);
661
662 switch (cmd) {
663 /* These need separate implementations for the data structure */
664 case HDIO_GETGEO:
665 return compat_hdio_getgeo(bdev, argp);
666 case BLKPG:
667 return compat_blkpg_ioctl(bdev, argp);
668
669 /* Compat mode returns 32-bit data instead of 'long' */
670 case BLKRAGET:
671 case BLKFRAGET:
672 if (!argp)
673 return -EINVAL;
674 return compat_put_long(argp,
675 (bdev->bd_disk->bdi->ra_pages * PAGE_SIZE) / 512);
676 case BLKGETSIZE:
677 if (bdev_nr_sectors(bdev) > ~(compat_ulong_t)0)
678 return -EFBIG;
679 return compat_put_ulong(argp, bdev_nr_sectors(bdev));
680
681 /* The data is compatible, but the command number is different */
682 case BLKBSZGET_32: /* get the logical block size (cf. BLKSSZGET) */
683 return put_int(argp, bdev_logical_block_size(bdev));
684 case BLKBSZSET_32:
685 return blkdev_bszset(bdev, mode, argp);
686 case BLKGETSIZE64_32:
687 return put_u64(argp, bdev_nr_bytes(bdev));
688
689 /* Incompatible alignment on i386 */
690 case BLKTRACESETUP32:
691 return blk_trace_ioctl(bdev, cmd, argp);
692 default:
693 break;
694 }
695
696 ret = blkdev_common_ioctl(bdev, mode, cmd, arg, argp);
697 if (ret == -ENOIOCTLCMD && disk->fops->compat_ioctl)
698 ret = disk->fops->compat_ioctl(bdev, mode, cmd, arg);
699
700 return ret;
701}
702#endif