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