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