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#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);