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 <asm/uaccess.h>
11
12static int blkpg_ioctl(struct block_device *bdev, struct blkpg_ioctl_arg __user *arg)
13{
14 struct block_device *bdevp;
15 struct gendisk *disk;
16 struct hd_struct *part, *lpart;
17 struct blkpg_ioctl_arg a;
18 struct blkpg_partition p;
19 struct disk_part_iter piter;
20 long long start, length;
21 int partno;
22
23 if (!capable(CAP_SYS_ADMIN))
24 return -EACCES;
25 if (copy_from_user(&a, arg, sizeof(struct blkpg_ioctl_arg)))
26 return -EFAULT;
27 if (copy_from_user(&p, a.data, sizeof(struct blkpg_partition)))
28 return -EFAULT;
29 disk = bdev->bd_disk;
30 if (bdev != bdev->bd_contains)
31 return -EINVAL;
32 partno = p.pno;
33 if (partno <= 0)
34 return -EINVAL;
35 switch (a.op) {
36 case BLKPG_ADD_PARTITION:
37 start = p.start >> 9;
38 length = p.length >> 9;
39 /* check for fit in a hd_struct */
40 if (sizeof(sector_t) == sizeof(long) &&
41 sizeof(long long) > sizeof(long)) {
42 long pstart = start, plength = length;
43 if (pstart != start || plength != length
44 || pstart < 0 || plength < 0 || partno > 65535)
45 return -EINVAL;
46 }
47
48 mutex_lock(&bdev->bd_mutex);
49
50 /* overlap? */
51 disk_part_iter_init(&piter, disk,
52 DISK_PITER_INCL_EMPTY);
53 while ((part = disk_part_iter_next(&piter))) {
54 if (!(start + length <= part->start_sect ||
55 start >= part->start_sect + part->nr_sects)) {
56 disk_part_iter_exit(&piter);
57 mutex_unlock(&bdev->bd_mutex);
58 return -EBUSY;
59 }
60 }
61 disk_part_iter_exit(&piter);
62
63 /* all seems OK */
64 part = add_partition(disk, partno, start, length,
65 ADDPART_FLAG_NONE, NULL);
66 mutex_unlock(&bdev->bd_mutex);
67 return PTR_ERR_OR_ZERO(part);
68 case BLKPG_DEL_PARTITION:
69 part = disk_get_part(disk, partno);
70 if (!part)
71 return -ENXIO;
72
73 bdevp = bdget(part_devt(part));
74 disk_put_part(part);
75 if (!bdevp)
76 return -ENOMEM;
77
78 mutex_lock(&bdevp->bd_mutex);
79 if (bdevp->bd_openers) {
80 mutex_unlock(&bdevp->bd_mutex);
81 bdput(bdevp);
82 return -EBUSY;
83 }
84 /* all seems OK */
85 fsync_bdev(bdevp);
86 invalidate_bdev(bdevp);
87
88 mutex_lock_nested(&bdev->bd_mutex, 1);
89 delete_partition(disk, partno);
90 mutex_unlock(&bdev->bd_mutex);
91 mutex_unlock(&bdevp->bd_mutex);
92 bdput(bdevp);
93
94 return 0;
95 case BLKPG_RESIZE_PARTITION:
96 start = p.start >> 9;
97 /* new length of partition in bytes */
98 length = p.length >> 9;
99 /* check for fit in a hd_struct */
100 if (sizeof(sector_t) == sizeof(long) &&
101 sizeof(long long) > sizeof(long)) {
102 long pstart = start, plength = length;
103 if (pstart != start || plength != length
104 || pstart < 0 || plength < 0)
105 return -EINVAL;
106 }
107 part = disk_get_part(disk, partno);
108 if (!part)
109 return -ENXIO;
110 bdevp = bdget(part_devt(part));
111 if (!bdevp) {
112 disk_put_part(part);
113 return -ENOMEM;
114 }
115 mutex_lock(&bdevp->bd_mutex);
116 mutex_lock_nested(&bdev->bd_mutex, 1);
117 if (start != part->start_sect) {
118 mutex_unlock(&bdevp->bd_mutex);
119 mutex_unlock(&bdev->bd_mutex);
120 bdput(bdevp);
121 disk_put_part(part);
122 return -EINVAL;
123 }
124 /* overlap? */
125 disk_part_iter_init(&piter, disk,
126 DISK_PITER_INCL_EMPTY);
127 while ((lpart = disk_part_iter_next(&piter))) {
128 if (lpart->partno != partno &&
129 !(start + length <= lpart->start_sect ||
130 start >= lpart->start_sect + lpart->nr_sects)
131 ) {
132 disk_part_iter_exit(&piter);
133 mutex_unlock(&bdevp->bd_mutex);
134 mutex_unlock(&bdev->bd_mutex);
135 bdput(bdevp);
136 disk_put_part(part);
137 return -EBUSY;
138 }
139 }
140 disk_part_iter_exit(&piter);
141 part_nr_sects_write(part, (sector_t)length);
142 i_size_write(bdevp->bd_inode, p.length);
143 mutex_unlock(&bdevp->bd_mutex);
144 mutex_unlock(&bdev->bd_mutex);
145 bdput(bdevp);
146 disk_put_part(part);
147 return 0;
148 default:
149 return -EINVAL;
150 }
151}
152
153static int blkdev_reread_part(struct block_device *bdev)
154{
155 struct gendisk *disk = bdev->bd_disk;
156 int res;
157
158 if (!disk_part_scan_enabled(disk) || bdev != bdev->bd_contains)
159 return -EINVAL;
160 if (!capable(CAP_SYS_ADMIN))
161 return -EACCES;
162 if (!mutex_trylock(&bdev->bd_mutex))
163 return -EBUSY;
164 res = rescan_partitions(disk, bdev);
165 mutex_unlock(&bdev->bd_mutex);
166 return res;
167}
168
169static int blk_ioctl_discard(struct block_device *bdev, uint64_t start,
170 uint64_t len, int secure)
171{
172 unsigned long flags = 0;
173
174 if (start & 511)
175 return -EINVAL;
176 if (len & 511)
177 return -EINVAL;
178 start >>= 9;
179 len >>= 9;
180
181 if (start + len > (i_size_read(bdev->bd_inode) >> 9))
182 return -EINVAL;
183 if (secure)
184 flags |= BLKDEV_DISCARD_SECURE;
185 return blkdev_issue_discard(bdev, start, len, GFP_KERNEL, flags);
186}
187
188static int blk_ioctl_zeroout(struct block_device *bdev, uint64_t start,
189 uint64_t len)
190{
191 if (start & 511)
192 return -EINVAL;
193 if (len & 511)
194 return -EINVAL;
195 start >>= 9;
196 len >>= 9;
197
198 if (start + len > (i_size_read(bdev->bd_inode) >> 9))
199 return -EINVAL;
200
201 return blkdev_issue_zeroout(bdev, start, len, GFP_KERNEL);
202}
203
204static int put_ushort(unsigned long arg, unsigned short val)
205{
206 return put_user(val, (unsigned short __user *)arg);
207}
208
209static int put_int(unsigned long arg, int val)
210{
211 return put_user(val, (int __user *)arg);
212}
213
214static int put_uint(unsigned long arg, unsigned int val)
215{
216 return put_user(val, (unsigned int __user *)arg);
217}
218
219static int put_long(unsigned long arg, long val)
220{
221 return put_user(val, (long __user *)arg);
222}
223
224static int put_ulong(unsigned long arg, unsigned long val)
225{
226 return put_user(val, (unsigned long __user *)arg);
227}
228
229static int put_u64(unsigned long arg, u64 val)
230{
231 return put_user(val, (u64 __user *)arg);
232}
233
234int __blkdev_driver_ioctl(struct block_device *bdev, fmode_t mode,
235 unsigned cmd, unsigned long arg)
236{
237 struct gendisk *disk = bdev->bd_disk;
238
239 if (disk->fops->ioctl)
240 return disk->fops->ioctl(bdev, mode, cmd, arg);
241
242 return -ENOTTY;
243}
244/*
245 * For the record: _GPL here is only because somebody decided to slap it
246 * on the previous export. Sheer idiocy, since it wasn't copyrightable
247 * at all and could be open-coded without any exports by anybody who cares.
248 */
249EXPORT_SYMBOL_GPL(__blkdev_driver_ioctl);
250
251/*
252 * Is it an unrecognized ioctl? The correct returns are either
253 * ENOTTY (final) or ENOIOCTLCMD ("I don't know this one, try a
254 * fallback"). ENOIOCTLCMD gets turned into ENOTTY by the ioctl
255 * code before returning.
256 *
257 * Confused drivers sometimes return EINVAL, which is wrong. It
258 * means "I understood the ioctl command, but the parameters to
259 * it were wrong".
260 *
261 * We should aim to just fix the broken drivers, the EINVAL case
262 * should go away.
263 */
264static inline int is_unrecognized_ioctl(int ret)
265{
266 return ret == -EINVAL ||
267 ret == -ENOTTY ||
268 ret == -ENOIOCTLCMD;
269}
270
271/*
272 * always keep this in sync with compat_blkdev_ioctl()
273 */
274int blkdev_ioctl(struct block_device *bdev, fmode_t mode, unsigned cmd,
275 unsigned long arg)
276{
277 struct gendisk *disk = bdev->bd_disk;
278 struct backing_dev_info *bdi;
279 loff_t size;
280 int ret, n;
281
282 switch(cmd) {
283 case BLKFLSBUF:
284 if (!capable(CAP_SYS_ADMIN))
285 return -EACCES;
286
287 ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
288 if (!is_unrecognized_ioctl(ret))
289 return ret;
290
291 fsync_bdev(bdev);
292 invalidate_bdev(bdev);
293 return 0;
294
295 case BLKROSET:
296 ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
297 if (!is_unrecognized_ioctl(ret))
298 return ret;
299 if (!capable(CAP_SYS_ADMIN))
300 return -EACCES;
301 if (get_user(n, (int __user *)(arg)))
302 return -EFAULT;
303 set_device_ro(bdev, n);
304 return 0;
305
306 case BLKDISCARD:
307 case BLKSECDISCARD: {
308 uint64_t range[2];
309
310 if (!(mode & FMODE_WRITE))
311 return -EBADF;
312
313 if (copy_from_user(range, (void __user *)arg, sizeof(range)))
314 return -EFAULT;
315
316 return blk_ioctl_discard(bdev, range[0], range[1],
317 cmd == BLKSECDISCARD);
318 }
319 case BLKZEROOUT: {
320 uint64_t range[2];
321
322 if (!(mode & FMODE_WRITE))
323 return -EBADF;
324
325 if (copy_from_user(range, (void __user *)arg, sizeof(range)))
326 return -EFAULT;
327
328 return blk_ioctl_zeroout(bdev, range[0], range[1]);
329 }
330
331 case HDIO_GETGEO: {
332 struct hd_geometry geo;
333
334 if (!arg)
335 return -EINVAL;
336 if (!disk->fops->getgeo)
337 return -ENOTTY;
338
339 /*
340 * We need to set the startsect first, the driver may
341 * want to override it.
342 */
343 memset(&geo, 0, sizeof(geo));
344 geo.start = get_start_sect(bdev);
345 ret = disk->fops->getgeo(bdev, &geo);
346 if (ret)
347 return ret;
348 if (copy_to_user((struct hd_geometry __user *)arg, &geo,
349 sizeof(geo)))
350 return -EFAULT;
351 return 0;
352 }
353 case BLKRAGET:
354 case BLKFRAGET:
355 if (!arg)
356 return -EINVAL;
357 bdi = blk_get_backing_dev_info(bdev);
358 if (bdi == NULL)
359 return -ENOTTY;
360 return put_long(arg, (bdi->ra_pages * PAGE_CACHE_SIZE) / 512);
361 case BLKROGET:
362 return put_int(arg, bdev_read_only(bdev) != 0);
363 case BLKBSZGET: /* get block device soft block size (cf. BLKSSZGET) */
364 return put_int(arg, block_size(bdev));
365 case BLKSSZGET: /* get block device logical block size */
366 return put_int(arg, bdev_logical_block_size(bdev));
367 case BLKPBSZGET: /* get block device physical block size */
368 return put_uint(arg, bdev_physical_block_size(bdev));
369 case BLKIOMIN:
370 return put_uint(arg, bdev_io_min(bdev));
371 case BLKIOOPT:
372 return put_uint(arg, bdev_io_opt(bdev));
373 case BLKALIGNOFF:
374 return put_int(arg, bdev_alignment_offset(bdev));
375 case BLKDISCARDZEROES:
376 return put_uint(arg, bdev_discard_zeroes_data(bdev));
377 case BLKSECTGET:
378 return put_ushort(arg, queue_max_sectors(bdev_get_queue(bdev)));
379 case BLKROTATIONAL:
380 return put_ushort(arg, !blk_queue_nonrot(bdev_get_queue(bdev)));
381 case BLKRASET:
382 case BLKFRASET:
383 if(!capable(CAP_SYS_ADMIN))
384 return -EACCES;
385 bdi = blk_get_backing_dev_info(bdev);
386 if (bdi == NULL)
387 return -ENOTTY;
388 bdi->ra_pages = (arg * 512) / PAGE_CACHE_SIZE;
389 return 0;
390 case BLKBSZSET:
391 /* set the logical block size */
392 if (!capable(CAP_SYS_ADMIN))
393 return -EACCES;
394 if (!arg)
395 return -EINVAL;
396 if (get_user(n, (int __user *) arg))
397 return -EFAULT;
398 if (!(mode & FMODE_EXCL)) {
399 bdgrab(bdev);
400 if (blkdev_get(bdev, mode | FMODE_EXCL, &bdev) < 0)
401 return -EBUSY;
402 }
403 ret = set_blocksize(bdev, n);
404 if (!(mode & FMODE_EXCL))
405 blkdev_put(bdev, mode | FMODE_EXCL);
406 return ret;
407 case BLKPG:
408 ret = blkpg_ioctl(bdev, (struct blkpg_ioctl_arg __user *) arg);
409 break;
410 case BLKRRPART:
411 ret = blkdev_reread_part(bdev);
412 break;
413 case BLKGETSIZE:
414 size = i_size_read(bdev->bd_inode);
415 if ((size >> 9) > ~0UL)
416 return -EFBIG;
417 return put_ulong(arg, size >> 9);
418 case BLKGETSIZE64:
419 return put_u64(arg, i_size_read(bdev->bd_inode));
420 case BLKTRACESTART:
421 case BLKTRACESTOP:
422 case BLKTRACESETUP:
423 case BLKTRACETEARDOWN:
424 ret = blk_trace_ioctl(bdev, cmd, (char __user *) arg);
425 break;
426 default:
427 ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
428 }
429 return ret;
430}
431EXPORT_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 <linux/pagemap.h>
15#include <linux/io_uring/cmd.h>
16#include <uapi/linux/blkdev.h>
17#include "blk.h"
18
19static int blkpg_do_ioctl(struct block_device *bdev,
20 struct blkpg_partition __user *upart, int op)
21{
22 struct gendisk *disk = bdev->bd_disk;
23 struct blkpg_partition p;
24 sector_t start, length, capacity, end;
25
26 if (!capable(CAP_SYS_ADMIN))
27 return -EACCES;
28 if (copy_from_user(&p, upart, sizeof(struct blkpg_partition)))
29 return -EFAULT;
30 if (bdev_is_partition(bdev))
31 return -EINVAL;
32
33 if (p.pno <= 0)
34 return -EINVAL;
35
36 if (op == BLKPG_DEL_PARTITION)
37 return bdev_del_partition(disk, p.pno);
38
39 if (p.start < 0 || p.length <= 0 || LLONG_MAX - p.length < p.start)
40 return -EINVAL;
41 /* Check that the partition is aligned to the block size */
42 if (!IS_ALIGNED(p.start | p.length, bdev_logical_block_size(bdev)))
43 return -EINVAL;
44
45 start = p.start >> SECTOR_SHIFT;
46 length = p.length >> SECTOR_SHIFT;
47 capacity = get_capacity(disk);
48
49 if (check_add_overflow(start, length, &end))
50 return -EINVAL;
51
52 if (start >= capacity || end > capacity)
53 return -EINVAL;
54
55 switch (op) {
56 case BLKPG_ADD_PARTITION:
57 return bdev_add_partition(disk, p.pno, start, length);
58 case BLKPG_RESIZE_PARTITION:
59 return bdev_resize_partition(disk, p.pno, start, length);
60 default:
61 return -EINVAL;
62 }
63}
64
65static int blkpg_ioctl(struct block_device *bdev,
66 struct blkpg_ioctl_arg __user *arg)
67{
68 struct blkpg_partition __user *udata;
69 int op;
70
71 if (get_user(op, &arg->op) || get_user(udata, &arg->data))
72 return -EFAULT;
73
74 return blkpg_do_ioctl(bdev, udata, op);
75}
76
77#ifdef CONFIG_COMPAT
78struct compat_blkpg_ioctl_arg {
79 compat_int_t op;
80 compat_int_t flags;
81 compat_int_t datalen;
82 compat_caddr_t data;
83};
84
85static int compat_blkpg_ioctl(struct block_device *bdev,
86 struct compat_blkpg_ioctl_arg __user *arg)
87{
88 compat_caddr_t udata;
89 int op;
90
91 if (get_user(op, &arg->op) || get_user(udata, &arg->data))
92 return -EFAULT;
93
94 return blkpg_do_ioctl(bdev, compat_ptr(udata), op);
95}
96#endif
97
98/*
99 * Check that [start, start + len) is a valid range from the block device's
100 * perspective, including verifying that it can be correctly translated into
101 * logical block addresses.
102 */
103static int blk_validate_byte_range(struct block_device *bdev,
104 uint64_t start, uint64_t len)
105{
106 unsigned int bs_mask = bdev_logical_block_size(bdev) - 1;
107 uint64_t end;
108
109 if ((start | len) & bs_mask)
110 return -EINVAL;
111 if (!len)
112 return -EINVAL;
113 if (check_add_overflow(start, len, &end) || end > bdev_nr_bytes(bdev))
114 return -EINVAL;
115
116 return 0;
117}
118
119static int blk_ioctl_discard(struct block_device *bdev, blk_mode_t mode,
120 unsigned long arg)
121{
122 uint64_t range[2], start, len;
123 struct bio *prev = NULL, *bio;
124 sector_t sector, nr_sects;
125 struct blk_plug plug;
126 int err;
127
128 if (copy_from_user(range, (void __user *)arg, sizeof(range)))
129 return -EFAULT;
130 start = range[0];
131 len = range[1];
132
133 if (!bdev_max_discard_sectors(bdev))
134 return -EOPNOTSUPP;
135
136 if (!(mode & BLK_OPEN_WRITE))
137 return -EBADF;
138 if (bdev_read_only(bdev))
139 return -EPERM;
140 err = blk_validate_byte_range(bdev, start, len);
141 if (err)
142 return err;
143
144 filemap_invalidate_lock(bdev->bd_mapping);
145 err = truncate_bdev_range(bdev, mode, start, start + len - 1);
146 if (err)
147 goto fail;
148
149 sector = start >> SECTOR_SHIFT;
150 nr_sects = len >> SECTOR_SHIFT;
151
152 blk_start_plug(&plug);
153 while (1) {
154 if (fatal_signal_pending(current)) {
155 if (prev)
156 bio_await_chain(prev);
157 err = -EINTR;
158 goto out_unplug;
159 }
160 bio = blk_alloc_discard_bio(bdev, §or, &nr_sects,
161 GFP_KERNEL);
162 if (!bio)
163 break;
164 prev = bio_chain_and_submit(prev, bio);
165 }
166 if (prev) {
167 err = submit_bio_wait(prev);
168 if (err == -EOPNOTSUPP)
169 err = 0;
170 bio_put(prev);
171 }
172out_unplug:
173 blk_finish_plug(&plug);
174fail:
175 filemap_invalidate_unlock(bdev->bd_mapping);
176 return err;
177}
178
179static int blk_ioctl_secure_erase(struct block_device *bdev, blk_mode_t mode,
180 void __user *argp)
181{
182 uint64_t start, len, end;
183 uint64_t range[2];
184 int err;
185
186 if (!(mode & BLK_OPEN_WRITE))
187 return -EBADF;
188 if (!bdev_max_secure_erase_sectors(bdev))
189 return -EOPNOTSUPP;
190 if (copy_from_user(range, argp, sizeof(range)))
191 return -EFAULT;
192
193 start = range[0];
194 len = range[1];
195 if ((start & 511) || (len & 511))
196 return -EINVAL;
197 if (check_add_overflow(start, len, &end) ||
198 end > bdev_nr_bytes(bdev))
199 return -EINVAL;
200
201 filemap_invalidate_lock(bdev->bd_mapping);
202 err = truncate_bdev_range(bdev, mode, start, end - 1);
203 if (!err)
204 err = blkdev_issue_secure_erase(bdev, start >> 9, len >> 9,
205 GFP_KERNEL);
206 filemap_invalidate_unlock(bdev->bd_mapping);
207 return err;
208}
209
210
211static int blk_ioctl_zeroout(struct block_device *bdev, blk_mode_t mode,
212 unsigned long arg)
213{
214 uint64_t range[2];
215 uint64_t start, end, len;
216 int err;
217
218 if (!(mode & BLK_OPEN_WRITE))
219 return -EBADF;
220
221 if (copy_from_user(range, (void __user *)arg, sizeof(range)))
222 return -EFAULT;
223
224 start = range[0];
225 len = range[1];
226 end = start + len - 1;
227
228 if (start & 511)
229 return -EINVAL;
230 if (len & 511)
231 return -EINVAL;
232 if (end >= (uint64_t)bdev_nr_bytes(bdev))
233 return -EINVAL;
234 if (end < start)
235 return -EINVAL;
236
237 /* Invalidate the page cache, including dirty pages */
238 filemap_invalidate_lock(bdev->bd_mapping);
239 err = truncate_bdev_range(bdev, mode, start, end);
240 if (err)
241 goto fail;
242
243 err = blkdev_issue_zeroout(bdev, start >> 9, len >> 9, GFP_KERNEL,
244 BLKDEV_ZERO_NOUNMAP | BLKDEV_ZERO_KILLABLE);
245
246fail:
247 filemap_invalidate_unlock(bdev->bd_mapping);
248 return err;
249}
250
251static int put_ushort(unsigned short __user *argp, unsigned short val)
252{
253 return put_user(val, argp);
254}
255
256static int put_int(int __user *argp, int val)
257{
258 return put_user(val, argp);
259}
260
261static int put_uint(unsigned int __user *argp, unsigned int val)
262{
263 return put_user(val, argp);
264}
265
266static int put_long(long __user *argp, long val)
267{
268 return put_user(val, argp);
269}
270
271static int put_ulong(unsigned long __user *argp, unsigned long val)
272{
273 return put_user(val, argp);
274}
275
276static int put_u64(u64 __user *argp, u64 val)
277{
278 return put_user(val, argp);
279}
280
281#ifdef CONFIG_COMPAT
282static int compat_put_long(compat_long_t __user *argp, long val)
283{
284 return put_user(val, argp);
285}
286
287static int compat_put_ulong(compat_ulong_t __user *argp, compat_ulong_t val)
288{
289 return put_user(val, argp);
290}
291#endif
292
293#ifdef CONFIG_COMPAT
294/*
295 * This is the equivalent of compat_ptr_ioctl(), to be used by block
296 * drivers that implement only commands that are completely compatible
297 * between 32-bit and 64-bit user space
298 */
299int blkdev_compat_ptr_ioctl(struct block_device *bdev, blk_mode_t mode,
300 unsigned cmd, unsigned long arg)
301{
302 struct gendisk *disk = bdev->bd_disk;
303
304 if (disk->fops->ioctl)
305 return disk->fops->ioctl(bdev, mode, cmd,
306 (unsigned long)compat_ptr(arg));
307
308 return -ENOIOCTLCMD;
309}
310EXPORT_SYMBOL(blkdev_compat_ptr_ioctl);
311#endif
312
313static bool blkdev_pr_allowed(struct block_device *bdev, blk_mode_t mode)
314{
315 /* no sense to make reservations for partitions */
316 if (bdev_is_partition(bdev))
317 return false;
318
319 if (capable(CAP_SYS_ADMIN))
320 return true;
321 /*
322 * Only allow unprivileged reservations if the file descriptor is open
323 * for writing.
324 */
325 return mode & BLK_OPEN_WRITE;
326}
327
328static int blkdev_pr_register(struct block_device *bdev, blk_mode_t mode,
329 struct pr_registration __user *arg)
330{
331 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
332 struct pr_registration reg;
333
334 if (!blkdev_pr_allowed(bdev, mode))
335 return -EPERM;
336 if (!ops || !ops->pr_register)
337 return -EOPNOTSUPP;
338 if (copy_from_user(®, arg, sizeof(reg)))
339 return -EFAULT;
340
341 if (reg.flags & ~PR_FL_IGNORE_KEY)
342 return -EOPNOTSUPP;
343 return ops->pr_register(bdev, reg.old_key, reg.new_key, reg.flags);
344}
345
346static int blkdev_pr_reserve(struct block_device *bdev, blk_mode_t mode,
347 struct pr_reservation __user *arg)
348{
349 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
350 struct pr_reservation rsv;
351
352 if (!blkdev_pr_allowed(bdev, mode))
353 return -EPERM;
354 if (!ops || !ops->pr_reserve)
355 return -EOPNOTSUPP;
356 if (copy_from_user(&rsv, arg, sizeof(rsv)))
357 return -EFAULT;
358
359 if (rsv.flags & ~PR_FL_IGNORE_KEY)
360 return -EOPNOTSUPP;
361 return ops->pr_reserve(bdev, rsv.key, rsv.type, rsv.flags);
362}
363
364static int blkdev_pr_release(struct block_device *bdev, blk_mode_t mode,
365 struct pr_reservation __user *arg)
366{
367 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
368 struct pr_reservation rsv;
369
370 if (!blkdev_pr_allowed(bdev, mode))
371 return -EPERM;
372 if (!ops || !ops->pr_release)
373 return -EOPNOTSUPP;
374 if (copy_from_user(&rsv, arg, sizeof(rsv)))
375 return -EFAULT;
376
377 if (rsv.flags)
378 return -EOPNOTSUPP;
379 return ops->pr_release(bdev, rsv.key, rsv.type);
380}
381
382static int blkdev_pr_preempt(struct block_device *bdev, blk_mode_t mode,
383 struct pr_preempt __user *arg, bool abort)
384{
385 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
386 struct pr_preempt p;
387
388 if (!blkdev_pr_allowed(bdev, mode))
389 return -EPERM;
390 if (!ops || !ops->pr_preempt)
391 return -EOPNOTSUPP;
392 if (copy_from_user(&p, arg, sizeof(p)))
393 return -EFAULT;
394
395 if (p.flags)
396 return -EOPNOTSUPP;
397 return ops->pr_preempt(bdev, p.old_key, p.new_key, p.type, abort);
398}
399
400static int blkdev_pr_clear(struct block_device *bdev, blk_mode_t mode,
401 struct pr_clear __user *arg)
402{
403 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
404 struct pr_clear c;
405
406 if (!blkdev_pr_allowed(bdev, mode))
407 return -EPERM;
408 if (!ops || !ops->pr_clear)
409 return -EOPNOTSUPP;
410 if (copy_from_user(&c, arg, sizeof(c)))
411 return -EFAULT;
412
413 if (c.flags)
414 return -EOPNOTSUPP;
415 return ops->pr_clear(bdev, c.key);
416}
417
418static int blkdev_flushbuf(struct block_device *bdev, unsigned cmd,
419 unsigned long arg)
420{
421 if (!capable(CAP_SYS_ADMIN))
422 return -EACCES;
423
424 mutex_lock(&bdev->bd_holder_lock);
425 if (bdev->bd_holder_ops && bdev->bd_holder_ops->sync)
426 bdev->bd_holder_ops->sync(bdev);
427 else {
428 mutex_unlock(&bdev->bd_holder_lock);
429 sync_blockdev(bdev);
430 }
431
432 invalidate_bdev(bdev);
433 return 0;
434}
435
436static int blkdev_roset(struct block_device *bdev, unsigned cmd,
437 unsigned long arg)
438{
439 int ret, n;
440
441 if (!capable(CAP_SYS_ADMIN))
442 return -EACCES;
443
444 if (get_user(n, (int __user *)arg))
445 return -EFAULT;
446 if (bdev->bd_disk->fops->set_read_only) {
447 ret = bdev->bd_disk->fops->set_read_only(bdev, n);
448 if (ret)
449 return ret;
450 }
451 if (n)
452 bdev_set_flag(bdev, BD_READ_ONLY);
453 else
454 bdev_clear_flag(bdev, BD_READ_ONLY);
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#ifdef CONFIG_COMPAT
485struct compat_hd_geometry {
486 unsigned char heads;
487 unsigned char sectors;
488 unsigned short cylinders;
489 u32 start;
490};
491
492static int compat_hdio_getgeo(struct block_device *bdev,
493 struct compat_hd_geometry __user *ugeo)
494{
495 struct gendisk *disk = bdev->bd_disk;
496 struct hd_geometry geo;
497 int ret;
498
499 if (!ugeo)
500 return -EINVAL;
501 if (!disk->fops->getgeo)
502 return -ENOTTY;
503
504 memset(&geo, 0, sizeof(geo));
505 /*
506 * We need to set the startsect first, the driver may
507 * want to override it.
508 */
509 geo.start = get_start_sect(bdev);
510 ret = disk->fops->getgeo(bdev, &geo);
511 if (ret)
512 return ret;
513
514 ret = copy_to_user(ugeo, &geo, 4);
515 ret |= put_user(geo.start, &ugeo->start);
516 if (ret)
517 ret = -EFAULT;
518
519 return ret;
520}
521#endif
522
523/* set the logical block size */
524static int blkdev_bszset(struct file *file, blk_mode_t mode,
525 int __user *argp)
526{
527 // this one might be file_inode(file)->i_rdev - a rare valid
528 // use of file_inode() for those.
529 dev_t dev = I_BDEV(file->f_mapping->host)->bd_dev;
530 struct file *excl_file;
531 int ret, n;
532
533 if (!capable(CAP_SYS_ADMIN))
534 return -EACCES;
535 if (!argp)
536 return -EINVAL;
537 if (get_user(n, argp))
538 return -EFAULT;
539
540 if (mode & BLK_OPEN_EXCL)
541 return set_blocksize(file, n);
542
543 excl_file = bdev_file_open_by_dev(dev, mode, &dev, NULL);
544 if (IS_ERR(excl_file))
545 return -EBUSY;
546 ret = set_blocksize(excl_file, n);
547 fput(excl_file);
548 return ret;
549}
550
551/*
552 * Common commands that are handled the same way on native and compat
553 * user space. Note the separate arg/argp parameters that are needed
554 * to deal with the compat_ptr() conversion.
555 */
556static int blkdev_common_ioctl(struct block_device *bdev, blk_mode_t mode,
557 unsigned int cmd, unsigned long arg,
558 void __user *argp)
559{
560 unsigned int max_sectors;
561
562 switch (cmd) {
563 case BLKFLSBUF:
564 return blkdev_flushbuf(bdev, cmd, arg);
565 case BLKROSET:
566 return blkdev_roset(bdev, cmd, arg);
567 case BLKDISCARD:
568 return blk_ioctl_discard(bdev, mode, arg);
569 case BLKSECDISCARD:
570 return blk_ioctl_secure_erase(bdev, mode, argp);
571 case BLKZEROOUT:
572 return blk_ioctl_zeroout(bdev, mode, arg);
573 case BLKGETDISKSEQ:
574 return put_u64(argp, bdev->bd_disk->diskseq);
575 case BLKREPORTZONE:
576 return blkdev_report_zones_ioctl(bdev, cmd, arg);
577 case BLKRESETZONE:
578 case BLKOPENZONE:
579 case BLKCLOSEZONE:
580 case BLKFINISHZONE:
581 return blkdev_zone_mgmt_ioctl(bdev, mode, cmd, arg);
582 case BLKGETZONESZ:
583 return put_uint(argp, bdev_zone_sectors(bdev));
584 case BLKGETNRZONES:
585 return put_uint(argp, bdev_nr_zones(bdev));
586 case BLKROGET:
587 return put_int(argp, bdev_read_only(bdev) != 0);
588 case BLKSSZGET: /* get block device logical block size */
589 return put_int(argp, bdev_logical_block_size(bdev));
590 case BLKPBSZGET: /* get block device physical block size */
591 return put_uint(argp, bdev_physical_block_size(bdev));
592 case BLKIOMIN:
593 return put_uint(argp, bdev_io_min(bdev));
594 case BLKIOOPT:
595 return put_uint(argp, bdev_io_opt(bdev));
596 case BLKALIGNOFF:
597 return put_int(argp, bdev_alignment_offset(bdev));
598 case BLKDISCARDZEROES:
599 return put_uint(argp, 0);
600 case BLKSECTGET:
601 max_sectors = min_t(unsigned int, USHRT_MAX,
602 queue_max_sectors(bdev_get_queue(bdev)));
603 return put_ushort(argp, max_sectors);
604 case BLKROTATIONAL:
605 return put_ushort(argp, !bdev_nonrot(bdev));
606 case BLKRASET:
607 case BLKFRASET:
608 if(!capable(CAP_SYS_ADMIN))
609 return -EACCES;
610 bdev->bd_disk->bdi->ra_pages = (arg * 512) / PAGE_SIZE;
611 return 0;
612 case BLKRRPART:
613 if (!capable(CAP_SYS_ADMIN))
614 return -EACCES;
615 if (bdev_is_partition(bdev))
616 return -EINVAL;
617 return disk_scan_partitions(bdev->bd_disk,
618 mode | BLK_OPEN_STRICT_SCAN);
619 case BLKTRACESTART:
620 case BLKTRACESTOP:
621 case BLKTRACETEARDOWN:
622 return blk_trace_ioctl(bdev, cmd, argp);
623 case IOC_PR_REGISTER:
624 return blkdev_pr_register(bdev, mode, argp);
625 case IOC_PR_RESERVE:
626 return blkdev_pr_reserve(bdev, mode, argp);
627 case IOC_PR_RELEASE:
628 return blkdev_pr_release(bdev, mode, argp);
629 case IOC_PR_PREEMPT:
630 return blkdev_pr_preempt(bdev, mode, argp, false);
631 case IOC_PR_PREEMPT_ABORT:
632 return blkdev_pr_preempt(bdev, mode, argp, true);
633 case IOC_PR_CLEAR:
634 return blkdev_pr_clear(bdev, mode, argp);
635 default:
636 return -ENOIOCTLCMD;
637 }
638}
639
640/*
641 * Always keep this in sync with compat_blkdev_ioctl()
642 * to handle all incompatible commands in both functions.
643 *
644 * New commands must be compatible and go into blkdev_common_ioctl
645 */
646long blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg)
647{
648 struct block_device *bdev = I_BDEV(file->f_mapping->host);
649 void __user *argp = (void __user *)arg;
650 blk_mode_t mode = file_to_blk_mode(file);
651 int ret;
652
653 switch (cmd) {
654 /* These need separate implementations for the data structure */
655 case HDIO_GETGEO:
656 return blkdev_getgeo(bdev, argp);
657 case BLKPG:
658 return blkpg_ioctl(bdev, argp);
659
660 /* Compat mode returns 32-bit data instead of 'long' */
661 case BLKRAGET:
662 case BLKFRAGET:
663 if (!argp)
664 return -EINVAL;
665 return put_long(argp,
666 (bdev->bd_disk->bdi->ra_pages * PAGE_SIZE) / 512);
667 case BLKGETSIZE:
668 if (bdev_nr_sectors(bdev) > ~0UL)
669 return -EFBIG;
670 return put_ulong(argp, bdev_nr_sectors(bdev));
671
672 /* The data is compatible, but the command number is different */
673 case BLKBSZGET: /* get block device soft block size (cf. BLKSSZGET) */
674 return put_int(argp, block_size(bdev));
675 case BLKBSZSET:
676 return blkdev_bszset(file, mode, argp);
677 case BLKGETSIZE64:
678 return put_u64(argp, bdev_nr_bytes(bdev));
679
680 /* Incompatible alignment on i386 */
681 case BLKTRACESETUP:
682 return blk_trace_ioctl(bdev, cmd, argp);
683 default:
684 break;
685 }
686
687 ret = blkdev_common_ioctl(bdev, mode, cmd, arg, argp);
688 if (ret != -ENOIOCTLCMD)
689 return ret;
690
691 if (!bdev->bd_disk->fops->ioctl)
692 return -ENOTTY;
693 return bdev->bd_disk->fops->ioctl(bdev, mode, cmd, arg);
694}
695
696#ifdef CONFIG_COMPAT
697
698#define BLKBSZGET_32 _IOR(0x12, 112, int)
699#define BLKBSZSET_32 _IOW(0x12, 113, int)
700#define BLKGETSIZE64_32 _IOR(0x12, 114, int)
701
702/* Most of the generic ioctls are handled in the normal fallback path.
703 This assumes the blkdev's low level compat_ioctl always returns
704 ENOIOCTLCMD for unknown ioctls. */
705long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg)
706{
707 int ret;
708 void __user *argp = compat_ptr(arg);
709 struct block_device *bdev = I_BDEV(file->f_mapping->host);
710 struct gendisk *disk = bdev->bd_disk;
711 blk_mode_t mode = file_to_blk_mode(file);
712
713 switch (cmd) {
714 /* These need separate implementations for the data structure */
715 case HDIO_GETGEO:
716 return compat_hdio_getgeo(bdev, argp);
717 case BLKPG:
718 return compat_blkpg_ioctl(bdev, argp);
719
720 /* Compat mode returns 32-bit data instead of 'long' */
721 case BLKRAGET:
722 case BLKFRAGET:
723 if (!argp)
724 return -EINVAL;
725 return compat_put_long(argp,
726 (bdev->bd_disk->bdi->ra_pages * PAGE_SIZE) / 512);
727 case BLKGETSIZE:
728 if (bdev_nr_sectors(bdev) > ~(compat_ulong_t)0)
729 return -EFBIG;
730 return compat_put_ulong(argp, bdev_nr_sectors(bdev));
731
732 /* The data is compatible, but the command number is different */
733 case BLKBSZGET_32: /* get the logical block size (cf. BLKSSZGET) */
734 return put_int(argp, bdev_logical_block_size(bdev));
735 case BLKBSZSET_32:
736 return blkdev_bszset(file, mode, argp);
737 case BLKGETSIZE64_32:
738 return put_u64(argp, bdev_nr_bytes(bdev));
739
740 /* Incompatible alignment on i386 */
741 case BLKTRACESETUP32:
742 return blk_trace_ioctl(bdev, cmd, argp);
743 default:
744 break;
745 }
746
747 ret = blkdev_common_ioctl(bdev, mode, cmd, arg, argp);
748 if (ret == -ENOIOCTLCMD && disk->fops->compat_ioctl)
749 ret = disk->fops->compat_ioctl(bdev, mode, cmd, arg);
750
751 return ret;
752}
753#endif
754
755struct blk_iou_cmd {
756 int res;
757 bool nowait;
758};
759
760static void blk_cmd_complete(struct io_uring_cmd *cmd, unsigned int issue_flags)
761{
762 struct blk_iou_cmd *bic = io_uring_cmd_to_pdu(cmd, struct blk_iou_cmd);
763
764 if (bic->res == -EAGAIN && bic->nowait)
765 io_uring_cmd_issue_blocking(cmd);
766 else
767 io_uring_cmd_done(cmd, bic->res, 0, issue_flags);
768}
769
770static void bio_cmd_bio_end_io(struct bio *bio)
771{
772 struct io_uring_cmd *cmd = bio->bi_private;
773 struct blk_iou_cmd *bic = io_uring_cmd_to_pdu(cmd, struct blk_iou_cmd);
774
775 if (unlikely(bio->bi_status) && !bic->res)
776 bic->res = blk_status_to_errno(bio->bi_status);
777
778 io_uring_cmd_do_in_task_lazy(cmd, blk_cmd_complete);
779 bio_put(bio);
780}
781
782static int blkdev_cmd_discard(struct io_uring_cmd *cmd,
783 struct block_device *bdev,
784 uint64_t start, uint64_t len, bool nowait)
785{
786 struct blk_iou_cmd *bic = io_uring_cmd_to_pdu(cmd, struct blk_iou_cmd);
787 gfp_t gfp = nowait ? GFP_NOWAIT : GFP_KERNEL;
788 sector_t sector = start >> SECTOR_SHIFT;
789 sector_t nr_sects = len >> SECTOR_SHIFT;
790 struct bio *prev = NULL, *bio;
791 int err;
792
793 if (!bdev_max_discard_sectors(bdev))
794 return -EOPNOTSUPP;
795 if (!(file_to_blk_mode(cmd->file) & BLK_OPEN_WRITE))
796 return -EBADF;
797 if (bdev_read_only(bdev))
798 return -EPERM;
799 err = blk_validate_byte_range(bdev, start, len);
800 if (err)
801 return err;
802
803 err = filemap_invalidate_pages(bdev->bd_mapping, start,
804 start + len - 1, nowait);
805 if (err)
806 return err;
807
808 while (true) {
809 bio = blk_alloc_discard_bio(bdev, §or, &nr_sects, gfp);
810 if (!bio)
811 break;
812 if (nowait) {
813 /*
814 * Don't allow multi-bio non-blocking submissions as
815 * subsequent bios may fail but we won't get a direct
816 * indication of that. Normally, the caller should
817 * retry from a blocking context.
818 */
819 if (unlikely(nr_sects)) {
820 bio_put(bio);
821 return -EAGAIN;
822 }
823 bio->bi_opf |= REQ_NOWAIT;
824 }
825
826 prev = bio_chain_and_submit(prev, bio);
827 }
828 if (unlikely(!prev))
829 return -EAGAIN;
830 if (unlikely(nr_sects))
831 bic->res = -EAGAIN;
832
833 prev->bi_private = cmd;
834 prev->bi_end_io = bio_cmd_bio_end_io;
835 submit_bio(prev);
836 return -EIOCBQUEUED;
837}
838
839int blkdev_uring_cmd(struct io_uring_cmd *cmd, unsigned int issue_flags)
840{
841 struct block_device *bdev = I_BDEV(cmd->file->f_mapping->host);
842 struct blk_iou_cmd *bic = io_uring_cmd_to_pdu(cmd, struct blk_iou_cmd);
843 const struct io_uring_sqe *sqe = cmd->sqe;
844 u32 cmd_op = cmd->cmd_op;
845 uint64_t start, len;
846
847 if (unlikely(sqe->ioprio || sqe->__pad1 || sqe->len ||
848 sqe->rw_flags || sqe->file_index))
849 return -EINVAL;
850
851 bic->res = 0;
852 bic->nowait = issue_flags & IO_URING_F_NONBLOCK;
853
854 start = READ_ONCE(sqe->addr);
855 len = READ_ONCE(sqe->addr3);
856
857 switch (cmd_op) {
858 case BLOCK_URING_CMD_DISCARD:
859 return blkdev_cmd_discard(cmd, bdev, start, len, bic->nowait);
860 }
861 return -EINVAL;
862}