| 1 | |
| 2 | #include <linux/module.h>
|
| 3 | #include <linux/blkdev.h>
|
| 4 | #include <linux/bio.h>
|
| 5 |
|
| 6 | #include "../dm-core.h"
|
| 7 | #include "cache_dev.h"
|
| 8 | #include "backing_dev.h"
|
| 9 | #include "cache.h"
|
| 10 | #include "dm_pcache.h"
|
| 11 |
|
| 12 | void pcache_defer_reqs_kick(struct dm_pcache *pcache)
|
| 13 | {
|
| 14 | struct pcache_cache *cache = &pcache->cache;
|
| 15 |
|
| 16 | spin_lock(&cache->seg_map_lock);
|
| 17 | if (!cache->cache_full)
|
| 18 | queue_work(pcache->task_wq, &pcache->defered_req_work);
|
| 19 | spin_unlock(&cache->seg_map_lock);
|
| 20 | }
|
| 21 |
|
| 22 | static void defer_req(struct pcache_request *pcache_req)
|
| 23 | {
|
| 24 | struct dm_pcache *pcache = pcache_req->pcache;
|
| 25 |
|
| 26 | BUG_ON(!list_empty(&pcache_req->list_node));
|
| 27 |
|
| 28 | spin_lock(&pcache->defered_req_list_lock);
|
| 29 | list_add(&pcache_req->list_node, &pcache->defered_req_list);
|
| 30 | pcache_defer_reqs_kick(pcache);
|
| 31 | spin_unlock(&pcache->defered_req_list_lock);
|
| 32 | }
|
| 33 |
|
| 34 | static void defered_req_fn(struct work_struct *work)
|
| 35 | {
|
| 36 | struct dm_pcache *pcache = container_of(work, struct dm_pcache, defered_req_work);
|
| 37 | struct pcache_request *pcache_req;
|
| 38 | LIST_HEAD(tmp_list);
|
| 39 | int ret;
|
| 40 |
|
| 41 | if (pcache_is_stopping(pcache))
|
| 42 | return;
|
| 43 |
|
| 44 | spin_lock(&pcache->defered_req_list_lock);
|
| 45 | list_splice_init(&pcache->defered_req_list, &tmp_list);
|
| 46 | spin_unlock(&pcache->defered_req_list_lock);
|
| 47 |
|
| 48 | while (!list_empty(&tmp_list)) {
|
| 49 | pcache_req = list_first_entry(&tmp_list,
|
| 50 | struct pcache_request, list_node);
|
| 51 | list_del_init(&pcache_req->list_node);
|
| 52 | pcache_req->ret = 0;
|
| 53 | ret = pcache_cache_handle_req(&pcache->cache, pcache_req);
|
| 54 | if (ret == -EBUSY)
|
| 55 | defer_req(pcache_req);
|
| 56 | else
|
| 57 | pcache_req_put(pcache_req, ret);
|
| 58 | }
|
| 59 | }
|
| 60 |
|
| 61 | void pcache_req_get(struct pcache_request *pcache_req)
|
| 62 | {
|
| 63 | kref_get(&pcache_req->ref);
|
| 64 | }
|
| 65 |
|
| 66 | static void end_req(struct kref *ref)
|
| 67 | {
|
| 68 | struct pcache_request *pcache_req = container_of(ref, struct pcache_request, ref);
|
| 69 | struct dm_pcache *pcache = pcache_req->pcache;
|
| 70 | struct bio *bio = pcache_req->bio;
|
| 71 | int ret = pcache_req->ret;
|
| 72 |
|
| 73 | if (ret == -EBUSY) {
|
| 74 | pcache_req_get(pcache_req);
|
| 75 | defer_req(pcache_req);
|
| 76 | } else {
|
| 77 | bio->bi_status = errno_to_blk_status(ret);
|
| 78 | bio_endio(bio);
|
| 79 |
|
| 80 | if (atomic_dec_and_test(&pcache->inflight_reqs))
|
| 81 | wake_up(&pcache->inflight_wq);
|
| 82 | }
|
| 83 | }
|
| 84 |
|
| 85 | void pcache_req_put(struct pcache_request *pcache_req, int ret)
|
| 86 | {
|
| 87 |
|
| 88 | if (ret && !pcache_req->ret)
|
| 89 | pcache_req->ret = ret;
|
| 90 |
|
| 91 | kref_put(&pcache_req->ref, end_req);
|
| 92 | }
|
| 93 |
|
| 94 | static bool at_least_one_arg(struct dm_arg_set *as, char **error)
|
| 95 | {
|
| 96 | if (!as->argc) {
|
| 97 | *error = "Insufficient args";
|
| 98 | return false;
|
| 99 | }
|
| 100 |
|
| 101 | return true;
|
| 102 | }
|
| 103 |
|
| 104 | static int parse_cache_dev(struct dm_pcache *pcache, struct dm_arg_set *as,
|
| 105 | char **error)
|
| 106 | {
|
| 107 | int ret;
|
| 108 |
|
| 109 | if (!at_least_one_arg(as, error))
|
| 110 | return -EINVAL;
|
| 111 | ret = dm_get_device(pcache->ti, dm_shift_arg(as),
|
| 112 | BLK_OPEN_READ | BLK_OPEN_WRITE,
|
| 113 | &pcache->cache_dev.dm_dev);
|
| 114 | if (ret) {
|
| 115 | *error = "Error opening cache device";
|
| 116 | return ret;
|
| 117 | }
|
| 118 |
|
| 119 | return 0;
|
| 120 | }
|
| 121 |
|
| 122 | static int parse_backing_dev(struct dm_pcache *pcache, struct dm_arg_set *as,
|
| 123 | char **error)
|
| 124 | {
|
| 125 | int ret;
|
| 126 |
|
| 127 | if (!at_least_one_arg(as, error))
|
| 128 | return -EINVAL;
|
| 129 |
|
| 130 | ret = dm_get_device(pcache->ti, dm_shift_arg(as),
|
| 131 | BLK_OPEN_READ | BLK_OPEN_WRITE,
|
| 132 | &pcache->backing_dev.dm_dev);
|
| 133 | if (ret) {
|
| 134 | *error = "Error opening backing device";
|
| 135 | return ret;
|
| 136 | }
|
| 137 |
|
| 138 | return 0;
|
| 139 | }
|
| 140 |
|
| 141 | static void pcache_init_opts(struct pcache_cache_options *opts)
|
| 142 | {
|
| 143 | opts->cache_mode = PCACHE_CACHE_MODE_WRITEBACK;
|
| 144 | opts->data_crc = false;
|
| 145 | }
|
| 146 |
|
| 147 | static int parse_cache_opts(struct dm_pcache *pcache, struct dm_arg_set *as,
|
| 148 | char **error)
|
| 149 | {
|
| 150 | struct pcache_cache_options *opts = &pcache->opts;
|
| 151 | static const struct dm_arg _args[] = {
|
| 152 | {0, 4, "Invalid number of cache option arguments"},
|
| 153 | };
|
| 154 | unsigned int argc;
|
| 155 | const char *arg;
|
| 156 | int ret;
|
| 157 |
|
| 158 | pcache_init_opts(opts);
|
| 159 | if (!as->argc)
|
| 160 | return 0;
|
| 161 |
|
| 162 | ret = dm_read_arg_group(_args, as, &argc, error);
|
| 163 | if (ret)
|
| 164 | return -EINVAL;
|
| 165 |
|
| 166 | while (argc) {
|
| 167 | arg = dm_shift_arg(as);
|
| 168 | argc--;
|
| 169 |
|
| 170 | if (!strcmp(arg, "cache_mode")) {
|
| 171 | if (!argc) {
|
| 172 | *error = "Missing value for cache_mode";
|
| 173 | return -EINVAL;
|
| 174 | }
|
| 175 | arg = dm_shift_arg(as);
|
| 176 | if (!strcmp(arg, "writeback")) {
|
| 177 | opts->cache_mode = PCACHE_CACHE_MODE_WRITEBACK;
|
| 178 | } else {
|
| 179 | *error = "Invalid cache mode parameter";
|
| 180 | return -EINVAL;
|
| 181 | }
|
| 182 | argc--;
|
| 183 | } else if (!strcmp(arg, "data_crc")) {
|
| 184 | if (!argc) {
|
| 185 | *error = "Missing value for data_crc";
|
| 186 | return -EINVAL;
|
| 187 | }
|
| 188 | arg = dm_shift_arg(as);
|
| 189 | if (!strcmp(arg, "true")) {
|
| 190 | opts->data_crc = true;
|
| 191 | } else if (!strcmp(arg, "false")) {
|
| 192 | opts->data_crc = false;
|
| 193 | } else {
|
| 194 | *error = "Invalid data crc parameter";
|
| 195 | return -EINVAL;
|
| 196 | }
|
| 197 | argc--;
|
| 198 | } else {
|
| 199 | *error = "Unrecognised cache option requested";
|
| 200 | return -EINVAL;
|
| 201 | }
|
| 202 | }
|
| 203 |
|
| 204 | return 0;
|
| 205 | }
|
| 206 |
|
| 207 | static int pcache_start(struct dm_pcache *pcache, char **error)
|
| 208 | {
|
| 209 | int ret;
|
| 210 |
|
| 211 | ret = cache_dev_start(pcache);
|
| 212 | if (ret) {
|
| 213 | *error = "Failed to start cache dev";
|
| 214 | return ret;
|
| 215 | }
|
| 216 |
|
| 217 | ret = backing_dev_start(pcache);
|
| 218 | if (ret) {
|
| 219 | *error = "Failed to start backing dev";
|
| 220 | goto stop_cache;
|
| 221 | }
|
| 222 |
|
| 223 | ret = pcache_cache_start(pcache);
|
| 224 | if (ret) {
|
| 225 | *error = "Failed to start pcache";
|
| 226 | goto stop_backing;
|
| 227 | }
|
| 228 |
|
| 229 | return 0;
|
| 230 | stop_backing:
|
| 231 | backing_dev_stop(pcache);
|
| 232 | stop_cache:
|
| 233 | cache_dev_stop(pcache);
|
| 234 |
|
| 235 | return ret;
|
| 236 | }
|
| 237 |
|
| 238 | static void pcache_destroy_args(struct dm_pcache *pcache)
|
| 239 | {
|
| 240 | if (pcache->cache_dev.dm_dev)
|
| 241 | dm_put_device(pcache->ti, pcache->cache_dev.dm_dev);
|
| 242 | if (pcache->backing_dev.dm_dev)
|
| 243 | dm_put_device(pcache->ti, pcache->backing_dev.dm_dev);
|
| 244 | }
|
| 245 |
|
| 246 | static int pcache_parse_args(struct dm_pcache *pcache, unsigned int argc, char **argv,
|
| 247 | char **error)
|
| 248 | {
|
| 249 | struct dm_arg_set as;
|
| 250 | int ret;
|
| 251 |
|
| 252 | as.argc = argc;
|
| 253 | as.argv = argv;
|
| 254 |
|
| 255 | |
| 256 | * Parse cache device
|
| 257 |
|
| 258 | ret = parse_cache_dev(pcache, &as, error);
|
| 259 | if (ret)
|
| 260 | return ret;
|
| 261 | |
| 262 | * Parse backing device
|
| 263 |
|
| 264 | ret = parse_backing_dev(pcache, &as, error);
|
| 265 | if (ret)
|
| 266 | goto out;
|
| 267 | |
| 268 | * Parse optional arguments
|
| 269 |
|
| 270 | ret = parse_cache_opts(pcache, &as, error);
|
| 271 | if (ret)
|
| 272 | goto out;
|
| 273 |
|
| 274 | return 0;
|
| 275 | out:
|
| 276 | pcache_destroy_args(pcache);
|
| 277 | return ret;
|
| 278 | }
|
| 279 |
|
| 280 | static int dm_pcache_ctr(struct dm_target *ti, unsigned int argc, char **argv)
|
| 281 | {
|
| 282 | struct mapped_device *md = ti->table->md;
|
| 283 | struct dm_pcache *pcache;
|
| 284 | int ret;
|
| 285 |
|
| 286 | if (md->map) {
|
| 287 | ti->error = "Don't support table loading for live md";
|
| 288 | return -EOPNOTSUPP;
|
| 289 | }
|
| 290 |
|
| 291 |
|
| 292 | pcache = kzalloc_obj(struct dm_pcache);
|
| 293 | if (!pcache)
|
| 294 | return -ENOMEM;
|
| 295 |
|
| 296 | pcache->task_wq = alloc_workqueue("pcache-%s-wq", WQ_UNBOUND | WQ_MEM_RECLAIM,
|
| 297 | 0, md->name);
|
| 298 | if (!pcache->task_wq) {
|
| 299 | ret = -ENOMEM;
|
| 300 | goto free_pcache;
|
| 301 | }
|
| 302 |
|
| 303 | spin_lock_init(&pcache->defered_req_list_lock);
|
| 304 | INIT_LIST_HEAD(&pcache->defered_req_list);
|
| 305 | INIT_WORK(&pcache->defered_req_work, defered_req_fn);
|
| 306 | pcache->ti = ti;
|
| 307 |
|
| 308 | ret = pcache_parse_args(pcache, argc, argv, &ti->error);
|
| 309 | if (ret)
|
| 310 | goto destroy_wq;
|
| 311 |
|
| 312 | ret = pcache_start(pcache, &ti->error);
|
| 313 | if (ret)
|
| 314 | goto destroy_args;
|
| 315 |
|
| 316 | ti->num_flush_bios = 1;
|
| 317 | ti->flush_supported = true;
|
| 318 | ti->per_io_data_size = sizeof(struct pcache_request);
|
| 319 | ti->private = pcache;
|
| 320 | atomic_set(&pcache->inflight_reqs, 0);
|
| 321 | atomic_set(&pcache->state, PCACHE_STATE_RUNNING);
|
| 322 | init_waitqueue_head(&pcache->inflight_wq);
|
| 323 |
|
| 324 | return 0;
|
| 325 | destroy_args:
|
| 326 | pcache_destroy_args(pcache);
|
| 327 | destroy_wq:
|
| 328 | destroy_workqueue(pcache->task_wq);
|
| 329 | free_pcache:
|
| 330 | kfree(pcache);
|
| 331 |
|
| 332 | return ret;
|
| 333 | }
|
| 334 |
|
| 335 | static void defer_req_stop(struct dm_pcache *pcache)
|
| 336 | {
|
| 337 | struct pcache_request *pcache_req;
|
| 338 | LIST_HEAD(tmp_list);
|
| 339 |
|
| 340 | flush_work(&pcache->defered_req_work);
|
| 341 |
|
| 342 | spin_lock(&pcache->defered_req_list_lock);
|
| 343 | list_splice_init(&pcache->defered_req_list, &tmp_list);
|
| 344 | spin_unlock(&pcache->defered_req_list_lock);
|
| 345 |
|
| 346 | while (!list_empty(&tmp_list)) {
|
| 347 | pcache_req = list_first_entry(&tmp_list,
|
| 348 | struct pcache_request, list_node);
|
| 349 | list_del_init(&pcache_req->list_node);
|
| 350 | pcache_req_put(pcache_req, -EIO);
|
| 351 | }
|
| 352 | }
|
| 353 |
|
| 354 | static void dm_pcache_dtr(struct dm_target *ti)
|
| 355 | {
|
| 356 | struct dm_pcache *pcache;
|
| 357 |
|
| 358 | pcache = ti->private;
|
| 359 | atomic_set(&pcache->state, PCACHE_STATE_STOPPING);
|
| 360 | defer_req_stop(pcache);
|
| 361 |
|
| 362 | wait_event(pcache->inflight_wq,
|
| 363 | atomic_read(&pcache->inflight_reqs) == 0);
|
| 364 |
|
| 365 | pcache_cache_stop(pcache);
|
| 366 | backing_dev_stop(pcache);
|
| 367 | cache_dev_stop(pcache);
|
| 368 |
|
| 369 | pcache_destroy_args(pcache);
|
| 370 | drain_workqueue(pcache->task_wq);
|
| 371 | destroy_workqueue(pcache->task_wq);
|
| 372 |
|
| 373 | kfree(pcache);
|
| 374 | }
|
| 375 |
|
| 376 | static int dm_pcache_map_bio(struct dm_target *ti, struct bio *bio)
|
| 377 | {
|
| 378 | struct pcache_request *pcache_req = dm_per_bio_data(bio, sizeof(struct pcache_request));
|
| 379 | struct dm_pcache *pcache = ti->private;
|
| 380 | int ret;
|
| 381 |
|
| 382 | pcache_req->pcache = pcache;
|
| 383 | kref_init(&pcache_req->ref);
|
| 384 | pcache_req->ret = 0;
|
| 385 | pcache_req->bio = bio;
|
| 386 | pcache_req->off = (u64)bio->bi_iter.bi_sector << SECTOR_SHIFT;
|
| 387 | pcache_req->data_len = bio->bi_iter.bi_size;
|
| 388 | INIT_LIST_HEAD(&pcache_req->list_node);
|
| 389 | atomic_inc(&pcache->inflight_reqs);
|
| 390 |
|
| 391 | ret = pcache_cache_handle_req(&pcache->cache, pcache_req);
|
| 392 | if (ret == -EBUSY)
|
| 393 | defer_req(pcache_req);
|
| 394 | else
|
| 395 | pcache_req_put(pcache_req, ret);
|
| 396 |
|
| 397 | return DM_MAPIO_SUBMITTED;
|
| 398 | }
|
| 399 |
|
| 400 | static void dm_pcache_status(struct dm_target *ti, status_type_t type,
|
| 401 | unsigned int status_flags, char *result,
|
| 402 | unsigned int maxlen)
|
| 403 | {
|
| 404 | struct dm_pcache *pcache = ti->private;
|
| 405 | struct pcache_cache_dev *cache_dev = &pcache->cache_dev;
|
| 406 | struct pcache_backing_dev *backing_dev = &pcache->backing_dev;
|
| 407 | struct pcache_cache *cache = &pcache->cache;
|
| 408 | unsigned int sz = 0;
|
| 409 |
|
| 410 | switch (type) {
|
| 411 | case STATUSTYPE_INFO:
|
| 412 | DMEMIT("%x %u %u %u %u %x %u:%u %u:%u %u:%u",
|
| 413 | cache_dev->sb_flags,
|
| 414 | cache_dev->seg_num,
|
| 415 | cache->n_segs,
|
| 416 | bitmap_weight(cache->seg_map, cache->n_segs),
|
| 417 | pcache_cache_get_gc_percent(cache),
|
| 418 | cache->cache_info.flags,
|
| 419 | cache->key_head.cache_seg->cache_seg_id,
|
| 420 | cache->key_head.seg_off,
|
| 421 | cache->dirty_tail.cache_seg->cache_seg_id,
|
| 422 | cache->dirty_tail.seg_off,
|
| 423 | cache->key_tail.cache_seg->cache_seg_id,
|
| 424 | cache->key_tail.seg_off);
|
| 425 | break;
|
| 426 | case STATUSTYPE_TABLE:
|
| 427 | DMEMIT("%s %s 4 cache_mode writeback crc %s",
|
| 428 | cache_dev->dm_dev->name,
|
| 429 | backing_dev->dm_dev->name,
|
| 430 | cache_data_crc_on(cache) ? "true" : "false");
|
| 431 | break;
|
| 432 | case STATUSTYPE_IMA:
|
| 433 | *result = '\0';
|
| 434 | break;
|
| 435 | }
|
| 436 | }
|
| 437 |
|
| 438 | static int dm_pcache_message(struct dm_target *ti, unsigned int argc,
|
| 439 | char **argv, char *result, unsigned int maxlen)
|
| 440 | {
|
| 441 | struct dm_pcache *pcache = ti->private;
|
| 442 | u8 val;
|
| 443 |
|
| 444 | if (argc != 2)
|
| 445 | goto err;
|
| 446 |
|
| 447 | if (!strcasecmp(argv[0], "gc_percent")) {
|
| 448 | if (kstrtou8(argv[1], 10, &val))
|
| 449 | goto err;
|
| 450 |
|
| 451 | return pcache_cache_set_gc_percent(&pcache->cache, val);
|
| 452 | }
|
| 453 | err:
|
| 454 | return -EINVAL;
|
| 455 | }
|
| 456 |
|
| 457 | static struct target_type dm_pcache_target = {
|
| 458 | .name = "pcache",
|
| 459 | .version = {0, 1, 0},
|
| 460 | .module = THIS_MODULE,
|
| 461 | .features = DM_TARGET_SINGLETON,
|
| 462 | .ctr = dm_pcache_ctr,
|
| 463 | .dtr = dm_pcache_dtr,
|
| 464 | .map = dm_pcache_map_bio,
|
| 465 | .status = dm_pcache_status,
|
| 466 | .message = dm_pcache_message,
|
| 467 | };
|
| 468 |
|
| 469 | static int __init dm_pcache_init(void)
|
| 470 | {
|
| 471 | int ret;
|
| 472 |
|
| 473 | ret = pcache_backing_init();
|
| 474 | if (ret)
|
| 475 | goto err;
|
| 476 |
|
| 477 | ret = pcache_cache_init();
|
| 478 | if (ret)
|
| 479 | goto backing_exit;
|
| 480 |
|
| 481 | ret = dm_register_target(&dm_pcache_target);
|
| 482 | if (ret)
|
| 483 | goto cache_exit;
|
| 484 | return 0;
|
| 485 |
|
| 486 | cache_exit:
|
| 487 | pcache_cache_exit();
|
| 488 | backing_exit:
|
| 489 | pcache_backing_exit();
|
| 490 | err:
|
| 491 | return ret;
|
| 492 | }
|
| 493 | module_init(dm_pcache_init);
|
| 494 |
|
| 495 | static void __exit dm_pcache_exit(void)
|
| 496 | {
|
| 497 | dm_unregister_target(&dm_pcache_target);
|
| 498 | pcache_cache_exit();
|
| 499 | pcache_backing_exit();
|
| 500 | }
|
| 501 | module_exit(dm_pcache_exit);
|
| 502 |
|
| 503 | MODULE_DESCRIPTION("dm-pcache Persistent Cache for block device");
|
| 504 | MODULE_AUTHOR("Dongsheng Yang <[email protected]>");
|
| 505 | MODULE_LICENSE("GPL");
|