xref: /linux/block/ioctl.c (revision 3932b9ca55b0be314a36d3e84faff3e823c081f5)
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 
12 static 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 
153 static 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 
169 static 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 
188 static 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 
204 static int put_ushort(unsigned long arg, unsigned short val)
205 {
206 	return put_user(val, (unsigned short __user *)arg);
207 }
208 
209 static int put_int(unsigned long arg, int val)
210 {
211 	return put_user(val, (int __user *)arg);
212 }
213 
214 static int put_uint(unsigned long arg, unsigned int val)
215 {
216 	return put_user(val, (unsigned int __user *)arg);
217 }
218 
219 static int put_long(unsigned long arg, long val)
220 {
221 	return put_user(val, (long __user *)arg);
222 }
223 
224 static int put_ulong(unsigned long arg, unsigned long val)
225 {
226 	return put_user(val, (unsigned long __user *)arg);
227 }
228 
229 static int put_u64(unsigned long arg, u64 val)
230 {
231 	return put_user(val, (u64 __user *)arg);
232 }
233 
234 int __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  */
249 EXPORT_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  */
264 static 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  */
274 int 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 	unsigned int max_sectors;
282 
283 	switch(cmd) {
284 	case BLKFLSBUF:
285 		if (!capable(CAP_SYS_ADMIN))
286 			return -EACCES;
287 
288 		ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
289 		if (!is_unrecognized_ioctl(ret))
290 			return ret;
291 
292 		fsync_bdev(bdev);
293 		invalidate_bdev(bdev);
294 		return 0;
295 
296 	case BLKROSET:
297 		ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
298 		if (!is_unrecognized_ioctl(ret))
299 			return ret;
300 		if (!capable(CAP_SYS_ADMIN))
301 			return -EACCES;
302 		if (get_user(n, (int __user *)(arg)))
303 			return -EFAULT;
304 		set_device_ro(bdev, n);
305 		return 0;
306 
307 	case BLKDISCARD:
308 	case BLKSECDISCARD: {
309 		uint64_t range[2];
310 
311 		if (!(mode & FMODE_WRITE))
312 			return -EBADF;
313 
314 		if (copy_from_user(range, (void __user *)arg, sizeof(range)))
315 			return -EFAULT;
316 
317 		return blk_ioctl_discard(bdev, range[0], range[1],
318 					 cmd == BLKSECDISCARD);
319 	}
320 	case BLKZEROOUT: {
321 		uint64_t range[2];
322 
323 		if (!(mode & FMODE_WRITE))
324 			return -EBADF;
325 
326 		if (copy_from_user(range, (void __user *)arg, sizeof(range)))
327 			return -EFAULT;
328 
329 		return blk_ioctl_zeroout(bdev, range[0], range[1]);
330 	}
331 
332 	case HDIO_GETGEO: {
333 		struct hd_geometry geo;
334 
335 		if (!arg)
336 			return -EINVAL;
337 		if (!disk->fops->getgeo)
338 			return -ENOTTY;
339 
340 		/*
341 		 * We need to set the startsect first, the driver may
342 		 * want to override it.
343 		 */
344 		memset(&geo, 0, sizeof(geo));
345 		geo.start = get_start_sect(bdev);
346 		ret = disk->fops->getgeo(bdev, &geo);
347 		if (ret)
348 			return ret;
349 		if (copy_to_user((struct hd_geometry __user *)arg, &geo,
350 					sizeof(geo)))
351 			return -EFAULT;
352 		return 0;
353 	}
354 	case BLKRAGET:
355 	case BLKFRAGET:
356 		if (!arg)
357 			return -EINVAL;
358 		bdi = blk_get_backing_dev_info(bdev);
359 		if (bdi == NULL)
360 			return -ENOTTY;
361 		return put_long(arg, (bdi->ra_pages * PAGE_CACHE_SIZE) / 512);
362 	case BLKROGET:
363 		return put_int(arg, bdev_read_only(bdev) != 0);
364 	case BLKBSZGET: /* get block device soft block size (cf. BLKSSZGET) */
365 		return put_int(arg, block_size(bdev));
366 	case BLKSSZGET: /* get block device logical block size */
367 		return put_int(arg, bdev_logical_block_size(bdev));
368 	case BLKPBSZGET: /* get block device physical block size */
369 		return put_uint(arg, bdev_physical_block_size(bdev));
370 	case BLKIOMIN:
371 		return put_uint(arg, bdev_io_min(bdev));
372 	case BLKIOOPT:
373 		return put_uint(arg, bdev_io_opt(bdev));
374 	case BLKALIGNOFF:
375 		return put_int(arg, bdev_alignment_offset(bdev));
376 	case BLKDISCARDZEROES:
377 		return put_uint(arg, bdev_discard_zeroes_data(bdev));
378 	case BLKSECTGET:
379 		max_sectors = min_t(unsigned int, USHRT_MAX,
380 				    queue_max_sectors(bdev_get_queue(bdev)));
381 		return put_ushort(arg, max_sectors);
382 	case BLKROTATIONAL:
383 		return put_ushort(arg, !blk_queue_nonrot(bdev_get_queue(bdev)));
384 	case BLKRASET:
385 	case BLKFRASET:
386 		if(!capable(CAP_SYS_ADMIN))
387 			return -EACCES;
388 		bdi = blk_get_backing_dev_info(bdev);
389 		if (bdi == NULL)
390 			return -ENOTTY;
391 		bdi->ra_pages = (arg * 512) / PAGE_CACHE_SIZE;
392 		return 0;
393 	case BLKBSZSET:
394 		/* set the logical block size */
395 		if (!capable(CAP_SYS_ADMIN))
396 			return -EACCES;
397 		if (!arg)
398 			return -EINVAL;
399 		if (get_user(n, (int __user *) arg))
400 			return -EFAULT;
401 		if (!(mode & FMODE_EXCL)) {
402 			bdgrab(bdev);
403 			if (blkdev_get(bdev, mode | FMODE_EXCL, &bdev) < 0)
404 				return -EBUSY;
405 		}
406 		ret = set_blocksize(bdev, n);
407 		if (!(mode & FMODE_EXCL))
408 			blkdev_put(bdev, mode | FMODE_EXCL);
409 		return ret;
410 	case BLKPG:
411 		ret = blkpg_ioctl(bdev, (struct blkpg_ioctl_arg __user *) arg);
412 		break;
413 	case BLKRRPART:
414 		ret = blkdev_reread_part(bdev);
415 		break;
416 	case BLKGETSIZE:
417 		size = i_size_read(bdev->bd_inode);
418 		if ((size >> 9) > ~0UL)
419 			return -EFBIG;
420 		return put_ulong(arg, size >> 9);
421 	case BLKGETSIZE64:
422 		return put_u64(arg, i_size_read(bdev->bd_inode));
423 	case BLKTRACESTART:
424 	case BLKTRACESTOP:
425 	case BLKTRACESETUP:
426 	case BLKTRACETEARDOWN:
427 		ret = blk_trace_ioctl(bdev, cmd, (char __user *) arg);
428 		break;
429 	default:
430 		ret = __blkdev_driver_ioctl(bdev, mode, cmd, arg);
431 	}
432 	return ret;
433 }
434 EXPORT_SYMBOL_GPL(blkdev_ioctl);
435