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 blkpg_partition p;
20 long long start, length;
21
22 if (!capable(CAP_SYS_ADMIN))
23 return -EACCES;
24 if (copy_from_user(&p, upart, sizeof(struct blkpg_partition)))
25 return -EFAULT;
26 if (bdev_is_partition(bdev))
27 return -EINVAL;
28
29 if (p.pno <= 0)
30 return -EINVAL;
31
32 if (op == BLKPG_DEL_PARTITION)
33 return bdev_del_partition(bdev, p.pno);
34
35 start = p.start >> SECTOR_SHIFT;
36 length = p.length >> SECTOR_SHIFT;
37
38 switch (op) {
39 case BLKPG_ADD_PARTITION:
40 /* check if partition is aligned to blocksize */
41 if (p.start & (bdev_logical_block_size(bdev) - 1))
42 return -EINVAL;
43 return bdev_add_partition(bdev, p.pno, start, length);
44 case BLKPG_RESIZE_PARTITION:
45 return bdev_resize_partition(bdev, p.pno, start, length);
46 default:
47 return -EINVAL;
48 }
49}
50
51static int blkpg_ioctl(struct block_device *bdev,
52 struct blkpg_ioctl_arg __user *arg)
53{
54 struct blkpg_partition __user *udata;
55 int op;
56
57 if (get_user(op, &arg->op) || get_user(udata, &arg->data))
58 return -EFAULT;
59
60 return blkpg_do_ioctl(bdev, udata, op);
61}
62
63#ifdef CONFIG_COMPAT
64struct compat_blkpg_ioctl_arg {
65 compat_int_t op;
66 compat_int_t flags;
67 compat_int_t datalen;
68 compat_caddr_t data;
69};
70
71static int compat_blkpg_ioctl(struct block_device *bdev,
72 struct compat_blkpg_ioctl_arg __user *arg)
73{
74 compat_caddr_t udata;
75 int op;
76
77 if (get_user(op, &arg->op) || get_user(udata, &arg->data))
78 return -EFAULT;
79
80 return blkpg_do_ioctl(bdev, compat_ptr(udata), op);
81}
82#endif
83
84static int blkdev_reread_part(struct block_device *bdev, fmode_t mode)
85{
86 struct block_device *tmp;
87
88 if (!disk_part_scan_enabled(bdev->bd_disk) || bdev_is_partition(bdev))
89 return -EINVAL;
90 if (!capable(CAP_SYS_ADMIN))
91 return -EACCES;
92 if (bdev->bd_disk->open_partitions)
93 return -EBUSY;
94
95 /*
96 * Reopen the device to revalidate the driver state and force a
97 * partition rescan.
98 */
99 mode &= ~FMODE_EXCL;
100 set_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
101
102 tmp = blkdev_get_by_dev(bdev->bd_dev, mode, NULL);
103 if (IS_ERR(tmp))
104 return PTR_ERR(tmp);
105 blkdev_put(tmp, mode);
106 return 0;
107}
108
109static int blk_ioctl_discard(struct block_device *bdev, fmode_t mode,
110 unsigned long arg, unsigned long flags)
111{
112 uint64_t range[2];
113 uint64_t start, len;
114 struct request_queue *q = bdev_get_queue(bdev);
115 int err;
116
117 if (!(mode & FMODE_WRITE))
118 return -EBADF;
119
120 if (!blk_queue_discard(q))
121 return -EOPNOTSUPP;
122
123 if (copy_from_user(range, (void __user *)arg, sizeof(range)))
124 return -EFAULT;
125
126 start = range[0];
127 len = range[1];
128
129 if (start & 511)
130 return -EINVAL;
131 if (len & 511)
132 return -EINVAL;
133
134 if (start + len > i_size_read(bdev->bd_inode))
135 return -EINVAL;
136
137 err = truncate_bdev_range(bdev, mode, start, start + len - 1);
138 if (err)
139 return err;
140
141 return blkdev_issue_discard(bdev, start >> 9, len >> 9,
142 GFP_KERNEL, flags);
143}
144
145static int blk_ioctl_zeroout(struct block_device *bdev, fmode_t mode,
146 unsigned long arg)
147{
148 uint64_t range[2];
149 uint64_t start, end, len;
150 int err;
151
152 if (!(mode & FMODE_WRITE))
153 return -EBADF;
154
155 if (copy_from_user(range, (void __user *)arg, sizeof(range)))
156 return -EFAULT;
157
158 start = range[0];
159 len = range[1];
160 end = start + len - 1;
161
162 if (start & 511)
163 return -EINVAL;
164 if (len & 511)
165 return -EINVAL;
166 if (end >= (uint64_t)i_size_read(bdev->bd_inode))
167 return -EINVAL;
168 if (end < start)
169 return -EINVAL;
170
171 /* Invalidate the page cache, including dirty pages */
172 err = truncate_bdev_range(bdev, mode, start, end);
173 if (err)
174 return err;
175
176 return blkdev_issue_zeroout(bdev, start >> 9, len >> 9, GFP_KERNEL,
177 BLKDEV_ZERO_NOUNMAP);
178}
179
180static int put_ushort(unsigned short __user *argp, unsigned short val)
181{
182 return put_user(val, argp);
183}
184
185static int put_int(int __user *argp, int val)
186{
187 return put_user(val, argp);
188}
189
190static int put_uint(unsigned int __user *argp, unsigned int val)
191{
192 return put_user(val, argp);
193}
194
195static int put_long(long __user *argp, long val)
196{
197 return put_user(val, argp);
198}
199
200static int put_ulong(unsigned long __user *argp, unsigned long val)
201{
202 return put_user(val, argp);
203}
204
205static int put_u64(u64 __user *argp, u64 val)
206{
207 return put_user(val, argp);
208}
209
210#ifdef CONFIG_COMPAT
211static int compat_put_long(compat_long_t __user *argp, long val)
212{
213 return put_user(val, argp);
214}
215
216static int compat_put_ulong(compat_ulong_t __user *argp, compat_ulong_t val)
217{
218 return put_user(val, argp);
219}
220#endif
221
222#ifdef CONFIG_COMPAT
223/*
224 * This is the equivalent of compat_ptr_ioctl(), to be used by block
225 * drivers that implement only commands that are completely compatible
226 * between 32-bit and 64-bit user space
227 */
228int blkdev_compat_ptr_ioctl(struct block_device *bdev, fmode_t mode,
229 unsigned cmd, unsigned long arg)
230{
231 struct gendisk *disk = bdev->bd_disk;
232
233 if (disk->fops->ioctl)
234 return disk->fops->ioctl(bdev, mode, cmd,
235 (unsigned long)compat_ptr(arg));
236
237 return -ENOIOCTLCMD;
238}
239EXPORT_SYMBOL(blkdev_compat_ptr_ioctl);
240#endif
241
242static int blkdev_pr_register(struct block_device *bdev,
243 struct pr_registration __user *arg)
244{
245 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
246 struct pr_registration reg;
247
248 if (!capable(CAP_SYS_ADMIN))
249 return -EPERM;
250 if (!ops || !ops->pr_register)
251 return -EOPNOTSUPP;
252 if (copy_from_user(®, arg, sizeof(reg)))
253 return -EFAULT;
254
255 if (reg.flags & ~PR_FL_IGNORE_KEY)
256 return -EOPNOTSUPP;
257 return ops->pr_register(bdev, reg.old_key, reg.new_key, reg.flags);
258}
259
260static int blkdev_pr_reserve(struct block_device *bdev,
261 struct pr_reservation __user *arg)
262{
263 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
264 struct pr_reservation rsv;
265
266 if (!capable(CAP_SYS_ADMIN))
267 return -EPERM;
268 if (!ops || !ops->pr_reserve)
269 return -EOPNOTSUPP;
270 if (copy_from_user(&rsv, arg, sizeof(rsv)))
271 return -EFAULT;
272
273 if (rsv.flags & ~PR_FL_IGNORE_KEY)
274 return -EOPNOTSUPP;
275 return ops->pr_reserve(bdev, rsv.key, rsv.type, rsv.flags);
276}
277
278static int blkdev_pr_release(struct block_device *bdev,
279 struct pr_reservation __user *arg)
280{
281 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
282 struct pr_reservation rsv;
283
284 if (!capable(CAP_SYS_ADMIN))
285 return -EPERM;
286 if (!ops || !ops->pr_release)
287 return -EOPNOTSUPP;
288 if (copy_from_user(&rsv, arg, sizeof(rsv)))
289 return -EFAULT;
290
291 if (rsv.flags)
292 return -EOPNOTSUPP;
293 return ops->pr_release(bdev, rsv.key, rsv.type);
294}
295
296static int blkdev_pr_preempt(struct block_device *bdev,
297 struct pr_preempt __user *arg, bool abort)
298{
299 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
300 struct pr_preempt p;
301
302 if (!capable(CAP_SYS_ADMIN))
303 return -EPERM;
304 if (!ops || !ops->pr_preempt)
305 return -EOPNOTSUPP;
306 if (copy_from_user(&p, arg, sizeof(p)))
307 return -EFAULT;
308
309 if (p.flags)
310 return -EOPNOTSUPP;
311 return ops->pr_preempt(bdev, p.old_key, p.new_key, p.type, abort);
312}
313
314static int blkdev_pr_clear(struct block_device *bdev,
315 struct pr_clear __user *arg)
316{
317 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
318 struct pr_clear c;
319
320 if (!capable(CAP_SYS_ADMIN))
321 return -EPERM;
322 if (!ops || !ops->pr_clear)
323 return -EOPNOTSUPP;
324 if (copy_from_user(&c, arg, sizeof(c)))
325 return -EFAULT;
326
327 if (c.flags)
328 return -EOPNOTSUPP;
329 return ops->pr_clear(bdev, c.key);
330}
331
332static int blkdev_flushbuf(struct block_device *bdev, fmode_t mode,
333 unsigned cmd, unsigned long arg)
334{
335 if (!capable(CAP_SYS_ADMIN))
336 return -EACCES;
337 fsync_bdev(bdev);
338 invalidate_bdev(bdev);
339 return 0;
340}
341
342static int blkdev_roset(struct block_device *bdev, fmode_t mode,
343 unsigned cmd, unsigned long arg)
344{
345 int ret, n;
346
347 if (!capable(CAP_SYS_ADMIN))
348 return -EACCES;
349
350 if (get_user(n, (int __user *)arg))
351 return -EFAULT;
352 if (bdev->bd_disk->fops->set_read_only) {
353 ret = bdev->bd_disk->fops->set_read_only(bdev, n);
354 if (ret)
355 return ret;
356 }
357 bdev->bd_read_only = n;
358 return 0;
359}
360
361static int blkdev_getgeo(struct block_device *bdev,
362 struct hd_geometry __user *argp)
363{
364 struct gendisk *disk = bdev->bd_disk;
365 struct hd_geometry geo;
366 int ret;
367
368 if (!argp)
369 return -EINVAL;
370 if (!disk->fops->getgeo)
371 return -ENOTTY;
372
373 /*
374 * We need to set the startsect first, the driver may
375 * want to override it.
376 */
377 memset(&geo, 0, sizeof(geo));
378 geo.start = get_start_sect(bdev);
379 ret = disk->fops->getgeo(bdev, &geo);
380 if (ret)
381 return ret;
382 if (copy_to_user(argp, &geo, sizeof(geo)))
383 return -EFAULT;
384 return 0;
385}
386
387#ifdef CONFIG_COMPAT
388struct compat_hd_geometry {
389 unsigned char heads;
390 unsigned char sectors;
391 unsigned short cylinders;
392 u32 start;
393};
394
395static int compat_hdio_getgeo(struct block_device *bdev,
396 struct compat_hd_geometry __user *ugeo)
397{
398 struct gendisk *disk = bdev->bd_disk;
399 struct hd_geometry geo;
400 int ret;
401
402 if (!ugeo)
403 return -EINVAL;
404 if (!disk->fops->getgeo)
405 return -ENOTTY;
406
407 memset(&geo, 0, sizeof(geo));
408 /*
409 * We need to set the startsect first, the driver may
410 * want to override it.
411 */
412 geo.start = get_start_sect(bdev);
413 ret = disk->fops->getgeo(bdev, &geo);
414 if (ret)
415 return ret;
416
417 ret = copy_to_user(ugeo, &geo, 4);
418 ret |= put_user(geo.start, &ugeo->start);
419 if (ret)
420 ret = -EFAULT;
421
422 return ret;
423}
424#endif
425
426/* set the logical block size */
427static int blkdev_bszset(struct block_device *bdev, fmode_t mode,
428 int __user *argp)
429{
430 int ret, n;
431
432 if (!capable(CAP_SYS_ADMIN))
433 return -EACCES;
434 if (!argp)
435 return -EINVAL;
436 if (get_user(n, argp))
437 return -EFAULT;
438
439 if (mode & FMODE_EXCL)
440 return set_blocksize(bdev, n);
441
442 if (IS_ERR(blkdev_get_by_dev(bdev->bd_dev, mode | FMODE_EXCL, &bdev)))
443 return -EBUSY;
444 ret = set_blocksize(bdev, n);
445 blkdev_put(bdev, mode | FMODE_EXCL);
446
447 return ret;
448}
449
450/*
451 * Common commands that are handled the same way on native and compat
452 * user space. Note the separate arg/argp parameters that are needed
453 * to deal with the compat_ptr() conversion.
454 */
455static int blkdev_common_ioctl(struct block_device *bdev, fmode_t mode,
456 unsigned cmd, unsigned long arg, void __user *argp)
457{
458 unsigned int max_sectors;
459
460 switch (cmd) {
461 case BLKFLSBUF:
462 return blkdev_flushbuf(bdev, mode, cmd, arg);
463 case BLKROSET:
464 return blkdev_roset(bdev, mode, cmd, arg);
465 case BLKDISCARD:
466 return blk_ioctl_discard(bdev, mode, arg, 0);
467 case BLKSECDISCARD:
468 return blk_ioctl_discard(bdev, mode, arg,
469 BLKDEV_DISCARD_SECURE);
470 case BLKZEROOUT:
471 return blk_ioctl_zeroout(bdev, mode, arg);
472 case BLKREPORTZONE:
473 return blkdev_report_zones_ioctl(bdev, mode, cmd, arg);
474 case BLKRESETZONE:
475 case BLKOPENZONE:
476 case BLKCLOSEZONE:
477 case BLKFINISHZONE:
478 return blkdev_zone_mgmt_ioctl(bdev, mode, cmd, arg);
479 case BLKGETZONESZ:
480 return put_uint(argp, bdev_zone_sectors(bdev));
481 case BLKGETNRZONES:
482 return put_uint(argp, blkdev_nr_zones(bdev->bd_disk));
483 case BLKROGET:
484 return put_int(argp, bdev_read_only(bdev) != 0);
485 case BLKSSZGET: /* get block device logical block size */
486 return put_int(argp, bdev_logical_block_size(bdev));
487 case BLKPBSZGET: /* get block device physical block size */
488 return put_uint(argp, bdev_physical_block_size(bdev));
489 case BLKIOMIN:
490 return put_uint(argp, bdev_io_min(bdev));
491 case BLKIOOPT:
492 return put_uint(argp, bdev_io_opt(bdev));
493 case BLKALIGNOFF:
494 return put_int(argp, bdev_alignment_offset(bdev));
495 case BLKDISCARDZEROES:
496 return put_uint(argp, 0);
497 case BLKSECTGET:
498 max_sectors = min_t(unsigned int, USHRT_MAX,
499 queue_max_sectors(bdev_get_queue(bdev)));
500 return put_ushort(argp, max_sectors);
501 case BLKROTATIONAL:
502 return put_ushort(argp, !blk_queue_nonrot(bdev_get_queue(bdev)));
503 case BLKRASET:
504 case BLKFRASET:
505 if(!capable(CAP_SYS_ADMIN))
506 return -EACCES;
507 bdev->bd_bdi->ra_pages = (arg * 512) / PAGE_SIZE;
508 return 0;
509 case BLKRRPART:
510 return blkdev_reread_part(bdev, mode);
511 case BLKTRACESTART:
512 case BLKTRACESTOP:
513 case BLKTRACETEARDOWN:
514 return blk_trace_ioctl(bdev, cmd, argp);
515 case IOC_PR_REGISTER:
516 return blkdev_pr_register(bdev, argp);
517 case IOC_PR_RESERVE:
518 return blkdev_pr_reserve(bdev, argp);
519 case IOC_PR_RELEASE:
520 return blkdev_pr_release(bdev, argp);
521 case IOC_PR_PREEMPT:
522 return blkdev_pr_preempt(bdev, argp, false);
523 case IOC_PR_PREEMPT_ABORT:
524 return blkdev_pr_preempt(bdev, argp, true);
525 case IOC_PR_CLEAR:
526 return blkdev_pr_clear(bdev, argp);
527 default:
528 return -ENOIOCTLCMD;
529 }
530}
531
532/*
533 * Always keep this in sync with compat_blkdev_ioctl()
534 * to handle all incompatible commands in both functions.
535 *
536 * New commands must be compatible and go into blkdev_common_ioctl
537 */
538int blkdev_ioctl(struct block_device *bdev, fmode_t mode, unsigned cmd,
539 unsigned long arg)
540{
541 int ret;
542 loff_t size;
543 void __user *argp = (void __user *)arg;
544
545 switch (cmd) {
546 /* These need separate implementations for the data structure */
547 case HDIO_GETGEO:
548 return blkdev_getgeo(bdev, argp);
549 case BLKPG:
550 return blkpg_ioctl(bdev, argp);
551
552 /* Compat mode returns 32-bit data instead of 'long' */
553 case BLKRAGET:
554 case BLKFRAGET:
555 if (!argp)
556 return -EINVAL;
557 return put_long(argp, (bdev->bd_bdi->ra_pages*PAGE_SIZE) / 512);
558 case BLKGETSIZE:
559 size = i_size_read(bdev->bd_inode);
560 if ((size >> 9) > ~0UL)
561 return -EFBIG;
562 return put_ulong(argp, size >> 9);
563
564 /* The data is compatible, but the command number is different */
565 case BLKBSZGET: /* get block device soft block size (cf. BLKSSZGET) */
566 return put_int(argp, block_size(bdev));
567 case BLKBSZSET:
568 return blkdev_bszset(bdev, mode, argp);
569 case BLKGETSIZE64:
570 return put_u64(argp, i_size_read(bdev->bd_inode));
571
572 /* Incompatible alignment on i386 */
573 case BLKTRACESETUP:
574 return blk_trace_ioctl(bdev, cmd, argp);
575 default:
576 break;
577 }
578
579 ret = blkdev_common_ioctl(bdev, mode, cmd, arg, argp);
580 if (ret != -ENOIOCTLCMD)
581 return ret;
582
583 if (!bdev->bd_disk->fops->ioctl)
584 return -ENOTTY;
585 return bdev->bd_disk->fops->ioctl(bdev, mode, cmd, arg);
586}
587EXPORT_SYMBOL_GPL(blkdev_ioctl); /* for /dev/raw */
588
589#ifdef CONFIG_COMPAT
590
591#define BLKBSZGET_32 _IOR(0x12, 112, int)
592#define BLKBSZSET_32 _IOW(0x12, 113, int)
593#define BLKGETSIZE64_32 _IOR(0x12, 114, int)
594
595/* Most of the generic ioctls are handled in the normal fallback path.
596 This assumes the blkdev's low level compat_ioctl always returns
597 ENOIOCTLCMD for unknown ioctls. */
598long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg)
599{
600 int ret;
601 void __user *argp = compat_ptr(arg);
602 struct block_device *bdev = I_BDEV(file->f_mapping->host);
603 struct gendisk *disk = bdev->bd_disk;
604 fmode_t mode = file->f_mode;
605 loff_t size;
606
607 /*
608 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
609 * to updated it before every ioctl.
610 */
611 if (file->f_flags & O_NDELAY)
612 mode |= FMODE_NDELAY;
613 else
614 mode &= ~FMODE_NDELAY;
615
616 switch (cmd) {
617 /* These need separate implementations for the data structure */
618 case HDIO_GETGEO:
619 return compat_hdio_getgeo(bdev, argp);
620 case BLKPG:
621 return compat_blkpg_ioctl(bdev, argp);
622
623 /* Compat mode returns 32-bit data instead of 'long' */
624 case BLKRAGET:
625 case BLKFRAGET:
626 if (!argp)
627 return -EINVAL;
628 return compat_put_long(argp,
629 (bdev->bd_bdi->ra_pages * PAGE_SIZE) / 512);
630 case BLKGETSIZE:
631 size = i_size_read(bdev->bd_inode);
632 if ((size >> 9) > ~0UL)
633 return -EFBIG;
634 return compat_put_ulong(argp, size >> 9);
635
636 /* The data is compatible, but the command number is different */
637 case BLKBSZGET_32: /* get the logical block size (cf. BLKSSZGET) */
638 return put_int(argp, bdev_logical_block_size(bdev));
639 case BLKBSZSET_32:
640 return blkdev_bszset(bdev, mode, argp);
641 case BLKGETSIZE64_32:
642 return put_u64(argp, i_size_read(bdev->bd_inode));
643
644 /* Incompatible alignment on i386 */
645 case BLKTRACESETUP32:
646 return blk_trace_ioctl(bdev, cmd, argp);
647 default:
648 break;
649 }
650
651 ret = blkdev_common_ioctl(bdev, mode, cmd, arg, argp);
652 if (ret == -ENOIOCTLCMD && disk->fops->compat_ioctl)
653 ret = disk->fops->compat_ioctl(bdev, mode, cmd, arg);
654
655 return ret;
656}
657#endif