1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (c) 2018 Red Hat, Inc.
4 *
5 * This is a test "dust" device, which fails reads on specified
6 * sectors, emulating the behavior of a hard disk drive sending
7 * a "Read Medium Error" sense.
8 *
9 */
10
11#include <linux/device-mapper.h>
12#include <linux/module.h>
13#include <linux/rbtree.h>
14
15#define DM_MSG_PREFIX "dust"
16
17struct badblock {
18 struct rb_node node;
19 sector_t bb;
20 unsigned char wr_fail_cnt;
21};
22
23struct dust_device {
24 struct dm_dev *dev;
25 struct rb_root badblocklist;
26 unsigned long long badblock_count;
27 spinlock_t dust_lock;
28 unsigned int blksz;
29 int sect_per_block_shift;
30 unsigned int sect_per_block;
31 sector_t start;
32 bool fail_read_on_bb:1;
33 bool quiet_mode:1;
34};
35
36static struct badblock *dust_rb_search(struct rb_root *root, sector_t blk)
37{
38 struct rb_node *node = root->rb_node;
39
40 while (node) {
41 struct badblock *bblk = rb_entry(node, struct badblock, node);
42
43 if (bblk->bb > blk)
44 node = node->rb_left;
45 else if (bblk->bb < blk)
46 node = node->rb_right;
47 else
48 return bblk;
49 }
50
51 return NULL;
52}
53
54static bool dust_rb_insert(struct rb_root *root, struct badblock *new)
55{
56 struct badblock *bblk;
57 struct rb_node **link = &root->rb_node, *parent = NULL;
58 sector_t value = new->bb;
59
60 while (*link) {
61 parent = *link;
62 bblk = rb_entry(parent, struct badblock, node);
63
64 if (bblk->bb > value)
65 link = &(*link)->rb_left;
66 else if (bblk->bb < value)
67 link = &(*link)->rb_right;
68 else
69 return false;
70 }
71
72 rb_link_node(node: &new->node, parent, rb_link: link);
73 rb_insert_color(&new->node, root);
74
75 return true;
76}
77
78static int dust_remove_block(struct dust_device *dd, unsigned long long block)
79{
80 struct badblock *bblock;
81 unsigned long flags;
82
83 spin_lock_irqsave(&dd->dust_lock, flags);
84 bblock = dust_rb_search(root: &dd->badblocklist, blk: block);
85
86 if (bblock == NULL) {
87 if (!dd->quiet_mode) {
88 DMERR("%s: block %llu not found in badblocklist",
89 __func__, block);
90 }
91 spin_unlock_irqrestore(lock: &dd->dust_lock, flags);
92 return -EINVAL;
93 }
94
95 rb_erase(&bblock->node, &dd->badblocklist);
96 dd->badblock_count--;
97 if (!dd->quiet_mode)
98 DMINFO("%s: badblock removed at block %llu", __func__, block);
99 kfree(objp: bblock);
100 spin_unlock_irqrestore(lock: &dd->dust_lock, flags);
101
102 return 0;
103}
104
105static int dust_add_block(struct dust_device *dd, unsigned long long block,
106 unsigned char wr_fail_cnt)
107{
108 struct badblock *bblock;
109 unsigned long flags;
110
111 bblock = kmalloc(size: sizeof(*bblock), GFP_KERNEL);
112 if (bblock == NULL) {
113 if (!dd->quiet_mode)
114 DMERR("%s: badblock allocation failed", __func__);
115 return -ENOMEM;
116 }
117
118 spin_lock_irqsave(&dd->dust_lock, flags);
119 bblock->bb = block;
120 bblock->wr_fail_cnt = wr_fail_cnt;
121 if (!dust_rb_insert(root: &dd->badblocklist, new: bblock)) {
122 if (!dd->quiet_mode) {
123 DMERR("%s: block %llu already in badblocklist",
124 __func__, block);
125 }
126 spin_unlock_irqrestore(lock: &dd->dust_lock, flags);
127 kfree(objp: bblock);
128 return -EINVAL;
129 }
130
131 dd->badblock_count++;
132 if (!dd->quiet_mode) {
133 DMINFO("%s: badblock added at block %llu with write fail count %u",
134 __func__, block, wr_fail_cnt);
135 }
136 spin_unlock_irqrestore(lock: &dd->dust_lock, flags);
137
138 return 0;
139}
140
141static int dust_query_block(struct dust_device *dd, unsigned long long block, char *result,
142 unsigned int maxlen, unsigned int *sz_ptr)
143{
144 struct badblock *bblock;
145 unsigned long flags;
146 unsigned int sz = *sz_ptr;
147
148 spin_lock_irqsave(&dd->dust_lock, flags);
149 bblock = dust_rb_search(root: &dd->badblocklist, blk: block);
150 if (bblock != NULL)
151 DMEMIT("%s: block %llu found in badblocklist", __func__, block);
152 else
153 DMEMIT("%s: block %llu not found in badblocklist", __func__, block);
154 spin_unlock_irqrestore(lock: &dd->dust_lock, flags);
155
156 return 1;
157}
158
159static int __dust_map_read(struct dust_device *dd, sector_t thisblock)
160{
161 struct badblock *bblk = dust_rb_search(root: &dd->badblocklist, blk: thisblock);
162
163 if (bblk)
164 return DM_MAPIO_KILL;
165
166 return DM_MAPIO_REMAPPED;
167}
168
169static int dust_map_read(struct dust_device *dd, sector_t thisblock,
170 bool fail_read_on_bb)
171{
172 unsigned long flags;
173 int r = DM_MAPIO_REMAPPED;
174
175 if (fail_read_on_bb) {
176 thisblock >>= dd->sect_per_block_shift;
177 spin_lock_irqsave(&dd->dust_lock, flags);
178 r = __dust_map_read(dd, thisblock);
179 spin_unlock_irqrestore(lock: &dd->dust_lock, flags);
180 }
181
182 return r;
183}
184
185static int __dust_map_write(struct dust_device *dd, sector_t thisblock)
186{
187 struct badblock *bblk = dust_rb_search(root: &dd->badblocklist, blk: thisblock);
188
189 if (bblk && bblk->wr_fail_cnt > 0) {
190 bblk->wr_fail_cnt--;
191 return DM_MAPIO_KILL;
192 }
193
194 if (bblk) {
195 rb_erase(&bblk->node, &dd->badblocklist);
196 dd->badblock_count--;
197 kfree(objp: bblk);
198 if (!dd->quiet_mode) {
199 sector_div(thisblock, dd->sect_per_block);
200 DMINFO("block %llu removed from badblocklist by write",
201 (unsigned long long)thisblock);
202 }
203 }
204
205 return DM_MAPIO_REMAPPED;
206}
207
208static int dust_map_write(struct dust_device *dd, sector_t thisblock,
209 bool fail_read_on_bb)
210{
211 unsigned long flags;
212 int r = DM_MAPIO_REMAPPED;
213
214 if (fail_read_on_bb) {
215 thisblock >>= dd->sect_per_block_shift;
216 spin_lock_irqsave(&dd->dust_lock, flags);
217 r = __dust_map_write(dd, thisblock);
218 spin_unlock_irqrestore(lock: &dd->dust_lock, flags);
219 }
220
221 return r;
222}
223
224static int dust_map(struct dm_target *ti, struct bio *bio)
225{
226 struct dust_device *dd = ti->private;
227 int r;
228
229 bio_set_dev(bio, bdev: dd->dev->bdev);
230 bio->bi_iter.bi_sector = dd->start + dm_target_offset(ti, bio->bi_iter.bi_sector);
231
232 if (bio_data_dir(bio) == READ)
233 r = dust_map_read(dd, thisblock: bio->bi_iter.bi_sector, fail_read_on_bb: dd->fail_read_on_bb);
234 else
235 r = dust_map_write(dd, thisblock: bio->bi_iter.bi_sector, fail_read_on_bb: dd->fail_read_on_bb);
236
237 return r;
238}
239
240static bool __dust_clear_badblocks(struct rb_root *tree,
241 unsigned long long count)
242{
243 struct rb_node *node = NULL, *nnode = NULL;
244
245 nnode = rb_first(tree);
246 if (nnode == NULL) {
247 BUG_ON(count != 0);
248 return false;
249 }
250
251 while (nnode) {
252 node = nnode;
253 nnode = rb_next(node);
254 rb_erase(node, tree);
255 count--;
256 kfree(objp: node);
257 }
258 BUG_ON(count != 0);
259 BUG_ON(tree->rb_node != NULL);
260
261 return true;
262}
263
264static int dust_clear_badblocks(struct dust_device *dd, char *result, unsigned int maxlen,
265 unsigned int *sz_ptr)
266{
267 unsigned long flags;
268 struct rb_root badblocklist;
269 unsigned long long badblock_count;
270 unsigned int sz = *sz_ptr;
271
272 spin_lock_irqsave(&dd->dust_lock, flags);
273 badblocklist = dd->badblocklist;
274 badblock_count = dd->badblock_count;
275 dd->badblocklist = RB_ROOT;
276 dd->badblock_count = 0;
277 spin_unlock_irqrestore(lock: &dd->dust_lock, flags);
278
279 if (!__dust_clear_badblocks(tree: &badblocklist, count: badblock_count))
280 DMEMIT("%s: no badblocks found", __func__);
281 else
282 DMEMIT("%s: badblocks cleared", __func__);
283
284 return 1;
285}
286
287static int dust_list_badblocks(struct dust_device *dd, char *result, unsigned int maxlen,
288 unsigned int *sz_ptr)
289{
290 unsigned long flags;
291 struct rb_root badblocklist;
292 struct rb_node *node;
293 struct badblock *bblk;
294 unsigned int sz = *sz_ptr;
295 unsigned long long num = 0;
296
297 spin_lock_irqsave(&dd->dust_lock, flags);
298 badblocklist = dd->badblocklist;
299 for (node = rb_first(&badblocklist); node; node = rb_next(node)) {
300 bblk = rb_entry(node, struct badblock, node);
301 DMEMIT("%llu\n", bblk->bb);
302 num++;
303 }
304
305 spin_unlock_irqrestore(lock: &dd->dust_lock, flags);
306 if (!num)
307 DMEMIT("No blocks in badblocklist");
308
309 return 1;
310}
311
312/*
313 * Target parameters:
314 *
315 * <device_path> <offset> <blksz>
316 *
317 * device_path: path to the block device
318 * offset: offset to data area from start of device_path
319 * blksz: block size (minimum 512, maximum 1073741824, must be a power of 2)
320 */
321static int dust_ctr(struct dm_target *ti, unsigned int argc, char **argv)
322{
323 struct dust_device *dd;
324 unsigned long long tmp;
325 char dummy;
326 unsigned int blksz;
327 unsigned int sect_per_block;
328 sector_t DUST_MAX_BLKSZ_SECTORS = 2097152;
329 sector_t max_block_sectors = min(ti->len, DUST_MAX_BLKSZ_SECTORS);
330
331 if (argc != 3) {
332 ti->error = "Invalid argument count";
333 return -EINVAL;
334 }
335
336 if (kstrtouint(s: argv[2], base: 10, res: &blksz) || !blksz) {
337 ti->error = "Invalid block size parameter";
338 return -EINVAL;
339 }
340
341 if (blksz < 512) {
342 ti->error = "Block size must be at least 512";
343 return -EINVAL;
344 }
345
346 if (!is_power_of_2(n: blksz)) {
347 ti->error = "Block size must be a power of 2";
348 return -EINVAL;
349 }
350
351 if (to_sector(n: blksz) > max_block_sectors) {
352 ti->error = "Block size is too large";
353 return -EINVAL;
354 }
355
356 sect_per_block = (blksz >> SECTOR_SHIFT);
357
358 if (sscanf(argv[1], "%llu%c", &tmp, &dummy) != 1 || tmp != (sector_t)tmp) {
359 ti->error = "Invalid device offset sector";
360 return -EINVAL;
361 }
362
363 dd = kzalloc(size: sizeof(struct dust_device), GFP_KERNEL);
364 if (dd == NULL) {
365 ti->error = "Cannot allocate context";
366 return -ENOMEM;
367 }
368
369 if (dm_get_device(ti, path: argv[0], mode: dm_table_get_mode(t: ti->table), result: &dd->dev)) {
370 ti->error = "Device lookup failed";
371 kfree(objp: dd);
372 return -EINVAL;
373 }
374
375 dd->sect_per_block = sect_per_block;
376 dd->blksz = blksz;
377 dd->start = tmp;
378
379 dd->sect_per_block_shift = __ffs(sect_per_block);
380
381 /*
382 * Whether to fail a read on a "bad" block.
383 * Defaults to false; enabled later by message.
384 */
385 dd->fail_read_on_bb = false;
386
387 /*
388 * Initialize bad block list rbtree.
389 */
390 dd->badblocklist = RB_ROOT;
391 dd->badblock_count = 0;
392 spin_lock_init(&dd->dust_lock);
393
394 dd->quiet_mode = false;
395
396 BUG_ON(dm_set_target_max_io_len(ti, dd->sect_per_block) != 0);
397
398 ti->num_discard_bios = 1;
399 ti->num_flush_bios = 1;
400 ti->private = dd;
401
402 return 0;
403}
404
405static void dust_dtr(struct dm_target *ti)
406{
407 struct dust_device *dd = ti->private;
408
409 __dust_clear_badblocks(tree: &dd->badblocklist, count: dd->badblock_count);
410 dm_put_device(ti, d: dd->dev);
411 kfree(objp: dd);
412}
413
414static int dust_message(struct dm_target *ti, unsigned int argc, char **argv,
415 char *result, unsigned int maxlen)
416{
417 struct dust_device *dd = ti->private;
418 sector_t size = bdev_nr_sectors(bdev: dd->dev->bdev);
419 bool invalid_msg = false;
420 int r = -EINVAL;
421 unsigned long long tmp, block;
422 unsigned char wr_fail_cnt;
423 unsigned int tmp_ui;
424 unsigned long flags;
425 unsigned int sz = 0;
426 char dummy;
427
428 if (argc == 1) {
429 if (!strcasecmp(s1: argv[0], s2: "addbadblock") ||
430 !strcasecmp(s1: argv[0], s2: "removebadblock") ||
431 !strcasecmp(s1: argv[0], s2: "queryblock")) {
432 DMERR("%s requires an additional argument", argv[0]);
433 } else if (!strcasecmp(s1: argv[0], s2: "disable")) {
434 DMINFO("disabling read failures on bad sectors");
435 dd->fail_read_on_bb = false;
436 r = 0;
437 } else if (!strcasecmp(s1: argv[0], s2: "enable")) {
438 DMINFO("enabling read failures on bad sectors");
439 dd->fail_read_on_bb = true;
440 r = 0;
441 } else if (!strcasecmp(s1: argv[0], s2: "countbadblocks")) {
442 spin_lock_irqsave(&dd->dust_lock, flags);
443 DMEMIT("countbadblocks: %llu badblock(s) found",
444 dd->badblock_count);
445 spin_unlock_irqrestore(lock: &dd->dust_lock, flags);
446 r = 1;
447 } else if (!strcasecmp(s1: argv[0], s2: "clearbadblocks")) {
448 r = dust_clear_badblocks(dd, result, maxlen, sz_ptr: &sz);
449 } else if (!strcasecmp(s1: argv[0], s2: "quiet")) {
450 if (!dd->quiet_mode)
451 dd->quiet_mode = true;
452 else
453 dd->quiet_mode = false;
454 r = 0;
455 } else if (!strcasecmp(s1: argv[0], s2: "listbadblocks")) {
456 r = dust_list_badblocks(dd, result, maxlen, sz_ptr: &sz);
457 } else {
458 invalid_msg = true;
459 }
460 } else if (argc == 2) {
461 if (sscanf(argv[1], "%llu%c", &tmp, &dummy) != 1)
462 return r;
463
464 block = tmp;
465 sector_div(size, dd->sect_per_block);
466 if (block > size) {
467 DMERR("selected block value out of range");
468 return r;
469 }
470
471 if (!strcasecmp(s1: argv[0], s2: "addbadblock"))
472 r = dust_add_block(dd, block, wr_fail_cnt: 0);
473 else if (!strcasecmp(s1: argv[0], s2: "removebadblock"))
474 r = dust_remove_block(dd, block);
475 else if (!strcasecmp(s1: argv[0], s2: "queryblock"))
476 r = dust_query_block(dd, block, result, maxlen, sz_ptr: &sz);
477 else
478 invalid_msg = true;
479
480 } else if (argc == 3) {
481 if (sscanf(argv[1], "%llu%c", &tmp, &dummy) != 1)
482 return r;
483
484 if (sscanf(argv[2], "%u%c", &tmp_ui, &dummy) != 1)
485 return r;
486
487 block = tmp;
488 if (tmp_ui > 255) {
489 DMERR("selected write fail count out of range");
490 return r;
491 }
492 wr_fail_cnt = tmp_ui;
493 sector_div(size, dd->sect_per_block);
494 if (block > size) {
495 DMERR("selected block value out of range");
496 return r;
497 }
498
499 if (!strcasecmp(s1: argv[0], s2: "addbadblock"))
500 r = dust_add_block(dd, block, wr_fail_cnt);
501 else
502 invalid_msg = true;
503
504 } else
505 DMERR("invalid number of arguments '%d'", argc);
506
507 if (invalid_msg)
508 DMERR("unrecognized message '%s' received", argv[0]);
509
510 return r;
511}
512
513static void dust_status(struct dm_target *ti, status_type_t type,
514 unsigned int status_flags, char *result, unsigned int maxlen)
515{
516 struct dust_device *dd = ti->private;
517 unsigned int sz = 0;
518
519 switch (type) {
520 case STATUSTYPE_INFO:
521 DMEMIT("%s %s %s", dd->dev->name,
522 dd->fail_read_on_bb ? "fail_read_on_bad_block" : "bypass",
523 dd->quiet_mode ? "quiet" : "verbose");
524 break;
525
526 case STATUSTYPE_TABLE:
527 DMEMIT("%s %llu %u", dd->dev->name,
528 (unsigned long long)dd->start, dd->blksz);
529 break;
530
531 case STATUSTYPE_IMA:
532 *result = '\0';
533 break;
534 }
535}
536
537static int dust_prepare_ioctl(struct dm_target *ti, struct block_device **bdev)
538{
539 struct dust_device *dd = ti->private;
540 struct dm_dev *dev = dd->dev;
541
542 *bdev = dev->bdev;
543
544 /*
545 * Only pass ioctls through if the device sizes match exactly.
546 */
547 if (dd->start || ti->len != bdev_nr_sectors(bdev: dev->bdev))
548 return 1;
549
550 return 0;
551}
552
553static int dust_iterate_devices(struct dm_target *ti, iterate_devices_callout_fn fn,
554 void *data)
555{
556 struct dust_device *dd = ti->private;
557
558 return fn(ti, dd->dev, dd->start, ti->len, data);
559}
560
561static struct target_type dust_target = {
562 .name = "dust",
563 .version = {1, 0, 0},
564 .module = THIS_MODULE,
565 .ctr = dust_ctr,
566 .dtr = dust_dtr,
567 .iterate_devices = dust_iterate_devices,
568 .map = dust_map,
569 .message = dust_message,
570 .status = dust_status,
571 .prepare_ioctl = dust_prepare_ioctl,
572};
573module_dm(dust);
574
575MODULE_DESCRIPTION(DM_NAME " dust test target");
576MODULE_AUTHOR("Bryan Gurney <dm-devel@redhat.com>");
577MODULE_LICENSE("GPL");
578

source code of linux/drivers/md/dm-dust.c