1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Add configfs and memory store: Kyungchan Koh <kkc6196@fb.com> and 4 * Shaohua Li <shli@fb.com> 5 */ 6 #include <linux/module.h> 7 8 #include <linux/moduleparam.h> 9 #include <linux/sched.h> 10 #include <linux/fs.h> 11 #include <linux/init.h> 12 #include "null_blk.h" 13 14 #undef pr_fmt 15 #define pr_fmt(fmt) "null_blk: " fmt 16 17 #define FREE_BATCH 16 18 19 #define TICKS_PER_SEC 50ULL 20 #define TIMER_INTERVAL (NSEC_PER_SEC / TICKS_PER_SEC) 21 22 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION 23 static DECLARE_FAULT_ATTR(null_timeout_attr); 24 static DECLARE_FAULT_ATTR(null_requeue_attr); 25 static DECLARE_FAULT_ATTR(null_init_hctx_attr); 26 #endif 27 28 static inline u64 mb_per_tick(int mbps) 29 { 30 return (1 << 20) / TICKS_PER_SEC * ((u64) mbps); 31 } 32 33 /* 34 * Status flags for nullb_device. 35 * 36 * CONFIGURED: Device has been configured and turned on. Cannot reconfigure. 37 * UP: Device is currently on and visible in userspace. 38 * THROTTLED: Device is being throttled. 39 * CACHE: Device is using a write-back cache. 40 */ 41 enum nullb_device_flags { 42 NULLB_DEV_FL_CONFIGURED = 0, 43 NULLB_DEV_FL_UP = 1, 44 NULLB_DEV_FL_THROTTLED = 2, 45 NULLB_DEV_FL_CACHE = 3, 46 }; 47 48 #define MAP_SZ ((PAGE_SIZE >> SECTOR_SHIFT) + 2) 49 /* 50 * nullb_page is a page in memory for nullb devices. 51 * 52 * @page: The page holding the data. 53 * @bitmap: The bitmap represents which sector in the page has data. 54 * Each bit represents one block size. For example, sector 8 55 * will use the 7th bit 56 * The highest 2 bits of bitmap are for special purpose. LOCK means the cache 57 * page is being flushing to storage. FREE means the cache page is freed and 58 * should be skipped from flushing to storage. Please see 59 * null_make_cache_space 60 */ 61 struct nullb_page { 62 struct page *page; 63 DECLARE_BITMAP(bitmap, MAP_SZ); 64 }; 65 #define NULLB_PAGE_LOCK (MAP_SZ - 1) 66 #define NULLB_PAGE_FREE (MAP_SZ - 2) 67 68 static LIST_HEAD(nullb_list); 69 static struct mutex lock; 70 static int null_major; 71 static DEFINE_IDA(nullb_indexes); 72 static struct blk_mq_tag_set tag_set; 73 74 enum { 75 NULL_IRQ_NONE = 0, 76 NULL_IRQ_SOFTIRQ = 1, 77 NULL_IRQ_TIMER = 2, 78 }; 79 80 static bool g_virt_boundary = false; 81 module_param_named(virt_boundary, g_virt_boundary, bool, 0444); 82 MODULE_PARM_DESC(virt_boundary, "Require a virtual boundary for the device. Default: False"); 83 84 static int g_no_sched; 85 module_param_named(no_sched, g_no_sched, int, 0444); 86 MODULE_PARM_DESC(no_sched, "No io scheduler"); 87 88 static int g_submit_queues = 1; 89 module_param_named(submit_queues, g_submit_queues, int, 0444); 90 MODULE_PARM_DESC(submit_queues, "Number of submission queues"); 91 92 static int g_poll_queues = 1; 93 module_param_named(poll_queues, g_poll_queues, int, 0444); 94 MODULE_PARM_DESC(poll_queues, "Number of IOPOLL submission queues"); 95 96 static int g_home_node = NUMA_NO_NODE; 97 module_param_named(home_node, g_home_node, int, 0444); 98 MODULE_PARM_DESC(home_node, "Home node for the device"); 99 100 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION 101 /* 102 * For more details about fault injection, please refer to 103 * Documentation/fault-injection/fault-injection.rst. 104 */ 105 static char g_timeout_str[80]; 106 module_param_string(timeout, g_timeout_str, sizeof(g_timeout_str), 0444); 107 MODULE_PARM_DESC(timeout, "Fault injection. timeout=<interval>,<probability>,<space>,<times>"); 108 109 static char g_requeue_str[80]; 110 module_param_string(requeue, g_requeue_str, sizeof(g_requeue_str), 0444); 111 MODULE_PARM_DESC(requeue, "Fault injection. requeue=<interval>,<probability>,<space>,<times>"); 112 113 static char g_init_hctx_str[80]; 114 module_param_string(init_hctx, g_init_hctx_str, sizeof(g_init_hctx_str), 0444); 115 MODULE_PARM_DESC(init_hctx, "Fault injection to fail hctx init. init_hctx=<interval>,<probability>,<space>,<times>"); 116 #endif 117 118 /* 119 * Historic queue modes. 120 * 121 * These days nothing but NULL_Q_MQ is actually supported, but we keep it the 122 * enum for error reporting. 123 */ 124 enum { 125 NULL_Q_BIO = 0, 126 NULL_Q_RQ = 1, 127 NULL_Q_MQ = 2, 128 }; 129 130 static int g_queue_mode = NULL_Q_MQ; 131 132 static int null_param_store_val(const char *str, int *val, int min, int max) 133 { 134 int ret, new_val; 135 136 ret = kstrtoint(str, 10, &new_val); 137 if (ret) 138 return -EINVAL; 139 140 if (new_val < min || new_val > max) 141 return -EINVAL; 142 143 *val = new_val; 144 return 0; 145 } 146 147 static int null_set_queue_mode(const char *str, const struct kernel_param *kp) 148 { 149 return null_param_store_val(str, &g_queue_mode, NULL_Q_BIO, NULL_Q_MQ); 150 } 151 152 static const struct kernel_param_ops null_queue_mode_param_ops = { 153 .set = null_set_queue_mode, 154 .get = param_get_int, 155 }; 156 157 device_param_cb(queue_mode, &null_queue_mode_param_ops, &g_queue_mode, 0444); 158 MODULE_PARM_DESC(queue_mode, "Block interface to use (0=bio,1=rq,2=multiqueue)"); 159 160 static int g_gb = 250; 161 module_param_named(gb, g_gb, int, 0444); 162 MODULE_PARM_DESC(gb, "Size in GB"); 163 164 static int g_bs = 512; 165 module_param_named(bs, g_bs, int, 0444); 166 MODULE_PARM_DESC(bs, "Block size (in bytes)"); 167 168 static int g_max_sectors; 169 module_param_named(max_sectors, g_max_sectors, int, 0444); 170 MODULE_PARM_DESC(max_sectors, "Maximum size of a command (in 512B sectors)"); 171 172 static unsigned int nr_devices = 1; 173 module_param(nr_devices, uint, 0444); 174 MODULE_PARM_DESC(nr_devices, "Number of devices to register"); 175 176 static bool g_blocking; 177 module_param_named(blocking, g_blocking, bool, 0444); 178 MODULE_PARM_DESC(blocking, "Register as a blocking blk-mq driver device"); 179 180 static bool g_shared_tags; 181 module_param_named(shared_tags, g_shared_tags, bool, 0444); 182 MODULE_PARM_DESC(shared_tags, "Share tag set between devices for blk-mq"); 183 184 static bool g_shared_tag_bitmap; 185 module_param_named(shared_tag_bitmap, g_shared_tag_bitmap, bool, 0444); 186 MODULE_PARM_DESC(shared_tag_bitmap, "Use shared tag bitmap for all submission queues for blk-mq"); 187 188 static int g_irqmode = NULL_IRQ_SOFTIRQ; 189 190 static int null_set_irqmode(const char *str, const struct kernel_param *kp) 191 { 192 return null_param_store_val(str, &g_irqmode, NULL_IRQ_NONE, 193 NULL_IRQ_TIMER); 194 } 195 196 static const struct kernel_param_ops null_irqmode_param_ops = { 197 .set = null_set_irqmode, 198 .get = param_get_int, 199 }; 200 201 device_param_cb(irqmode, &null_irqmode_param_ops, &g_irqmode, 0444); 202 MODULE_PARM_DESC(irqmode, "IRQ completion handler. 0-none, 1-softirq, 2-timer"); 203 204 static unsigned long g_completion_nsec = 10000; 205 module_param_named(completion_nsec, g_completion_nsec, ulong, 0444); 206 MODULE_PARM_DESC(completion_nsec, "Time in ns to complete a request in hardware. Default: 10,000ns"); 207 208 static int g_hw_queue_depth = 64; 209 module_param_named(hw_queue_depth, g_hw_queue_depth, int, 0444); 210 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 64"); 211 212 static bool g_use_per_node_hctx; 213 module_param_named(use_per_node_hctx, g_use_per_node_hctx, bool, 0444); 214 MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware context queues. Default: false"); 215 216 static bool g_memory_backed; 217 module_param_named(memory_backed, g_memory_backed, bool, 0444); 218 MODULE_PARM_DESC(memory_backed, "Create a memory-backed block device. Default: false"); 219 220 static bool g_discard; 221 module_param_named(discard, g_discard, bool, 0444); 222 MODULE_PARM_DESC(discard, "Support discard operations (requires memory-backed null_blk device). Default: false"); 223 224 static unsigned long g_cache_size; 225 module_param_named(cache_size, g_cache_size, ulong, 0444); 226 MODULE_PARM_DESC(mbps, "Cache size in MiB for memory-backed device. Default: 0 (none)"); 227 228 static bool g_fua = true; 229 module_param_named(fua, g_fua, bool, 0444); 230 MODULE_PARM_DESC(zoned, "Enable/disable FUA support when cache_size is used. Default: true"); 231 232 static unsigned int g_mbps; 233 module_param_named(mbps, g_mbps, uint, 0444); 234 MODULE_PARM_DESC(mbps, "Limit maximum bandwidth (in MiB/s). Default: 0 (no limit)"); 235 236 static bool g_zoned; 237 module_param_named(zoned, g_zoned, bool, S_IRUGO); 238 MODULE_PARM_DESC(zoned, "Make device as a host-managed zoned block device. Default: false"); 239 240 static unsigned long g_zone_size = 256; 241 module_param_named(zone_size, g_zone_size, ulong, S_IRUGO); 242 MODULE_PARM_DESC(zone_size, "Zone size in MB when block device is zoned. Must be power-of-two: Default: 256"); 243 244 static unsigned long g_zone_capacity; 245 module_param_named(zone_capacity, g_zone_capacity, ulong, 0444); 246 MODULE_PARM_DESC(zone_capacity, "Zone capacity in MB when block device is zoned. Can be less than or equal to zone size. Default: Zone size"); 247 248 static unsigned int g_zone_nr_conv; 249 module_param_named(zone_nr_conv, g_zone_nr_conv, uint, 0444); 250 MODULE_PARM_DESC(zone_nr_conv, "Number of conventional zones when block device is zoned. Default: 0"); 251 252 static unsigned int g_zone_max_open; 253 module_param_named(zone_max_open, g_zone_max_open, uint, 0444); 254 MODULE_PARM_DESC(zone_max_open, "Maximum number of open zones when block device is zoned. Default: 0 (no limit)"); 255 256 static unsigned int g_zone_max_active; 257 module_param_named(zone_max_active, g_zone_max_active, uint, 0444); 258 MODULE_PARM_DESC(zone_max_active, "Maximum number of active zones when block device is zoned. Default: 0 (no limit)"); 259 260 static int g_zone_append_max_sectors = INT_MAX; 261 module_param_named(zone_append_max_sectors, g_zone_append_max_sectors, int, 0444); 262 MODULE_PARM_DESC(zone_append_max_sectors, 263 "Maximum size of a zone append command (in 512B sectors). Specify 0 for zone append emulation"); 264 265 static struct nullb_device *null_alloc_dev(void); 266 static void null_free_dev(struct nullb_device *dev); 267 static void null_del_dev(struct nullb *nullb); 268 static int null_add_dev(struct nullb_device *dev); 269 static struct nullb *null_find_dev_by_name(const char *name); 270 static void null_free_device_storage(struct nullb_device *dev, bool is_cache); 271 272 static inline struct nullb_device *to_nullb_device(struct config_item *item) 273 { 274 return item ? container_of(to_config_group(item), struct nullb_device, group) : NULL; 275 } 276 277 static inline ssize_t nullb_device_uint_attr_show(unsigned int val, char *page) 278 { 279 return snprintf(page, PAGE_SIZE, "%u\n", val); 280 } 281 282 static inline ssize_t nullb_device_ulong_attr_show(unsigned long val, 283 char *page) 284 { 285 return snprintf(page, PAGE_SIZE, "%lu\n", val); 286 } 287 288 static inline ssize_t nullb_device_bool_attr_show(bool val, char *page) 289 { 290 return snprintf(page, PAGE_SIZE, "%u\n", val); 291 } 292 293 static ssize_t nullb_device_uint_attr_store(unsigned int *val, 294 const char *page, size_t count) 295 { 296 unsigned int tmp; 297 int result; 298 299 result = kstrtouint(page, 0, &tmp); 300 if (result < 0) 301 return result; 302 303 *val = tmp; 304 return count; 305 } 306 307 static ssize_t nullb_device_ulong_attr_store(unsigned long *val, 308 const char *page, size_t count) 309 { 310 int result; 311 unsigned long tmp; 312 313 result = kstrtoul(page, 0, &tmp); 314 if (result < 0) 315 return result; 316 317 *val = tmp; 318 return count; 319 } 320 321 static ssize_t nullb_device_bool_attr_store(bool *val, const char *page, 322 size_t count) 323 { 324 bool tmp; 325 int result; 326 327 result = kstrtobool(page, &tmp); 328 if (result < 0) 329 return result; 330 331 *val = tmp; 332 return count; 333 } 334 335 /* The following macro should only be used with TYPE = {uint, ulong, bool}. */ 336 #define NULLB_DEVICE_ATTR(NAME, TYPE, APPLY) \ 337 static ssize_t \ 338 nullb_device_##NAME##_show(struct config_item *item, char *page) \ 339 { \ 340 return nullb_device_##TYPE##_attr_show( \ 341 to_nullb_device(item)->NAME, page); \ 342 } \ 343 static ssize_t \ 344 nullb_device_##NAME##_store(struct config_item *item, const char *page, \ 345 size_t count) \ 346 { \ 347 int (*apply_fn)(struct nullb_device *dev, TYPE new_value) = APPLY;\ 348 struct nullb_device *dev = to_nullb_device(item); \ 349 TYPE new_value = 0; \ 350 int ret; \ 351 \ 352 ret = nullb_device_##TYPE##_attr_store(&new_value, page, count);\ 353 if (ret < 0) \ 354 return ret; \ 355 if (apply_fn) \ 356 ret = apply_fn(dev, new_value); \ 357 else if (test_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags)) \ 358 ret = -EBUSY; \ 359 if (ret < 0) \ 360 return ret; \ 361 dev->NAME = new_value; \ 362 return count; \ 363 } \ 364 CONFIGFS_ATTR(nullb_device_, NAME); 365 366 static int nullb_update_nr_hw_queues(struct nullb_device *dev, 367 unsigned int submit_queues, 368 unsigned int poll_queues) 369 370 { 371 struct blk_mq_tag_set *set; 372 int ret, nr_hw_queues; 373 374 if (!dev->nullb) 375 return 0; 376 377 /* 378 * Make sure at least one submit queue exists. 379 */ 380 if (!submit_queues) 381 return -EINVAL; 382 383 /* 384 * Make sure that null_init_hctx() does not access nullb->queues[] past 385 * the end of that array. 386 */ 387 if (submit_queues > nr_cpu_ids || poll_queues > g_poll_queues) 388 return -EINVAL; 389 390 /* 391 * Keep previous and new queue numbers in nullb_device for reference in 392 * the call back function null_map_queues(). 393 */ 394 dev->prev_submit_queues = dev->submit_queues; 395 dev->prev_poll_queues = dev->poll_queues; 396 dev->submit_queues = submit_queues; 397 dev->poll_queues = poll_queues; 398 399 set = dev->nullb->tag_set; 400 nr_hw_queues = submit_queues + poll_queues; 401 blk_mq_update_nr_hw_queues(set, nr_hw_queues); 402 ret = set->nr_hw_queues == nr_hw_queues ? 0 : -ENOMEM; 403 404 if (ret) { 405 /* on error, revert the queue numbers */ 406 dev->submit_queues = dev->prev_submit_queues; 407 dev->poll_queues = dev->prev_poll_queues; 408 } 409 410 return ret; 411 } 412 413 static int nullb_apply_submit_queues(struct nullb_device *dev, 414 unsigned int submit_queues) 415 { 416 int ret; 417 418 mutex_lock(&lock); 419 ret = nullb_update_nr_hw_queues(dev, submit_queues, dev->poll_queues); 420 mutex_unlock(&lock); 421 422 return ret; 423 } 424 425 static int nullb_apply_poll_queues(struct nullb_device *dev, 426 unsigned int poll_queues) 427 { 428 int ret; 429 430 mutex_lock(&lock); 431 ret = nullb_update_nr_hw_queues(dev, dev->submit_queues, poll_queues); 432 mutex_unlock(&lock); 433 434 return ret; 435 } 436 437 NULLB_DEVICE_ATTR(size, ulong, NULL); 438 NULLB_DEVICE_ATTR(completion_nsec, ulong, NULL); 439 NULLB_DEVICE_ATTR(submit_queues, uint, nullb_apply_submit_queues); 440 NULLB_DEVICE_ATTR(poll_queues, uint, nullb_apply_poll_queues); 441 NULLB_DEVICE_ATTR(home_node, uint, NULL); 442 NULLB_DEVICE_ATTR(queue_mode, uint, NULL); 443 NULLB_DEVICE_ATTR(blocksize, uint, NULL); 444 NULLB_DEVICE_ATTR(max_sectors, uint, NULL); 445 NULLB_DEVICE_ATTR(irqmode, uint, NULL); 446 NULLB_DEVICE_ATTR(hw_queue_depth, uint, NULL); 447 NULLB_DEVICE_ATTR(index, uint, NULL); 448 NULLB_DEVICE_ATTR(blocking, bool, NULL); 449 NULLB_DEVICE_ATTR(use_per_node_hctx, bool, NULL); 450 NULLB_DEVICE_ATTR(memory_backed, bool, NULL); 451 NULLB_DEVICE_ATTR(discard, bool, NULL); 452 NULLB_DEVICE_ATTR(mbps, uint, NULL); 453 NULLB_DEVICE_ATTR(cache_size, ulong, NULL); 454 NULLB_DEVICE_ATTR(zoned, bool, NULL); 455 NULLB_DEVICE_ATTR(zone_size, ulong, NULL); 456 NULLB_DEVICE_ATTR(zone_capacity, ulong, NULL); 457 NULLB_DEVICE_ATTR(zone_nr_conv, uint, NULL); 458 NULLB_DEVICE_ATTR(zone_max_open, uint, NULL); 459 NULLB_DEVICE_ATTR(zone_max_active, uint, NULL); 460 NULLB_DEVICE_ATTR(zone_append_max_sectors, uint, NULL); 461 NULLB_DEVICE_ATTR(virt_boundary, bool, NULL); 462 NULLB_DEVICE_ATTR(no_sched, bool, NULL); 463 NULLB_DEVICE_ATTR(shared_tags, bool, NULL); 464 NULLB_DEVICE_ATTR(shared_tag_bitmap, bool, NULL); 465 NULLB_DEVICE_ATTR(fua, bool, NULL); 466 467 static ssize_t nullb_device_power_show(struct config_item *item, char *page) 468 { 469 return nullb_device_bool_attr_show(to_nullb_device(item)->power, page); 470 } 471 472 static ssize_t nullb_device_power_store(struct config_item *item, 473 const char *page, size_t count) 474 { 475 struct nullb_device *dev = to_nullb_device(item); 476 bool newp = false; 477 ssize_t ret; 478 479 ret = nullb_device_bool_attr_store(&newp, page, count); 480 if (ret < 0) 481 return ret; 482 483 ret = count; 484 mutex_lock(&lock); 485 if (!dev->power && newp) { 486 if (test_and_set_bit(NULLB_DEV_FL_UP, &dev->flags)) 487 goto out; 488 489 ret = null_add_dev(dev); 490 if (ret) { 491 clear_bit(NULLB_DEV_FL_UP, &dev->flags); 492 goto out; 493 } 494 495 set_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags); 496 dev->power = newp; 497 ret = count; 498 } else if (dev->power && !newp) { 499 if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) { 500 dev->power = newp; 501 null_del_dev(dev->nullb); 502 } 503 clear_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags); 504 } 505 506 out: 507 mutex_unlock(&lock); 508 return ret; 509 } 510 511 CONFIGFS_ATTR(nullb_device_, power); 512 513 static ssize_t nullb_device_badblocks_show(struct config_item *item, char *page) 514 { 515 struct nullb_device *t_dev = to_nullb_device(item); 516 517 return badblocks_show(&t_dev->badblocks, page, 0); 518 } 519 520 static ssize_t nullb_device_badblocks_store(struct config_item *item, 521 const char *page, size_t count) 522 { 523 struct nullb_device *t_dev = to_nullb_device(item); 524 char *orig, *buf, *tmp; 525 u64 start, end; 526 int ret; 527 528 orig = kstrndup(page, count, GFP_KERNEL); 529 if (!orig) 530 return -ENOMEM; 531 532 buf = strstrip(orig); 533 534 ret = -EINVAL; 535 if (buf[0] != '+' && buf[0] != '-') 536 goto out; 537 tmp = strchr(&buf[1], '-'); 538 if (!tmp) 539 goto out; 540 *tmp = '\0'; 541 ret = kstrtoull(buf + 1, 0, &start); 542 if (ret) 543 goto out; 544 ret = kstrtoull(tmp + 1, 0, &end); 545 if (ret) 546 goto out; 547 ret = -EINVAL; 548 if (start > end) 549 goto out; 550 /* enable badblocks */ 551 cmpxchg(&t_dev->badblocks.shift, -1, 0); 552 if (buf[0] == '+') 553 ret = badblocks_set(&t_dev->badblocks, start, 554 end - start + 1, 1); 555 else 556 ret = badblocks_clear(&t_dev->badblocks, start, 557 end - start + 1); 558 if (ret == 0) 559 ret = count; 560 out: 561 kfree(orig); 562 return ret; 563 } 564 CONFIGFS_ATTR(nullb_device_, badblocks); 565 566 static ssize_t nullb_device_zone_readonly_store(struct config_item *item, 567 const char *page, size_t count) 568 { 569 struct nullb_device *dev = to_nullb_device(item); 570 571 return zone_cond_store(dev, page, count, BLK_ZONE_COND_READONLY); 572 } 573 CONFIGFS_ATTR_WO(nullb_device_, zone_readonly); 574 575 static ssize_t nullb_device_zone_offline_store(struct config_item *item, 576 const char *page, size_t count) 577 { 578 struct nullb_device *dev = to_nullb_device(item); 579 580 return zone_cond_store(dev, page, count, BLK_ZONE_COND_OFFLINE); 581 } 582 CONFIGFS_ATTR_WO(nullb_device_, zone_offline); 583 584 static struct configfs_attribute *nullb_device_attrs[] = { 585 &nullb_device_attr_size, 586 &nullb_device_attr_completion_nsec, 587 &nullb_device_attr_submit_queues, 588 &nullb_device_attr_poll_queues, 589 &nullb_device_attr_home_node, 590 &nullb_device_attr_queue_mode, 591 &nullb_device_attr_blocksize, 592 &nullb_device_attr_max_sectors, 593 &nullb_device_attr_irqmode, 594 &nullb_device_attr_hw_queue_depth, 595 &nullb_device_attr_index, 596 &nullb_device_attr_blocking, 597 &nullb_device_attr_use_per_node_hctx, 598 &nullb_device_attr_power, 599 &nullb_device_attr_memory_backed, 600 &nullb_device_attr_discard, 601 &nullb_device_attr_mbps, 602 &nullb_device_attr_cache_size, 603 &nullb_device_attr_badblocks, 604 &nullb_device_attr_zoned, 605 &nullb_device_attr_zone_size, 606 &nullb_device_attr_zone_capacity, 607 &nullb_device_attr_zone_nr_conv, 608 &nullb_device_attr_zone_max_open, 609 &nullb_device_attr_zone_max_active, 610 &nullb_device_attr_zone_append_max_sectors, 611 &nullb_device_attr_zone_readonly, 612 &nullb_device_attr_zone_offline, 613 &nullb_device_attr_virt_boundary, 614 &nullb_device_attr_no_sched, 615 &nullb_device_attr_shared_tags, 616 &nullb_device_attr_shared_tag_bitmap, 617 &nullb_device_attr_fua, 618 NULL, 619 }; 620 621 static void nullb_device_release(struct config_item *item) 622 { 623 struct nullb_device *dev = to_nullb_device(item); 624 625 null_free_device_storage(dev, false); 626 null_free_dev(dev); 627 } 628 629 static struct configfs_item_operations nullb_device_ops = { 630 .release = nullb_device_release, 631 }; 632 633 static const struct config_item_type nullb_device_type = { 634 .ct_item_ops = &nullb_device_ops, 635 .ct_attrs = nullb_device_attrs, 636 .ct_owner = THIS_MODULE, 637 }; 638 639 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION 640 641 static void nullb_add_fault_config(struct nullb_device *dev) 642 { 643 fault_config_init(&dev->timeout_config, "timeout_inject"); 644 fault_config_init(&dev->requeue_config, "requeue_inject"); 645 fault_config_init(&dev->init_hctx_fault_config, "init_hctx_fault_inject"); 646 647 configfs_add_default_group(&dev->timeout_config.group, &dev->group); 648 configfs_add_default_group(&dev->requeue_config.group, &dev->group); 649 configfs_add_default_group(&dev->init_hctx_fault_config.group, &dev->group); 650 } 651 652 #else 653 654 static void nullb_add_fault_config(struct nullb_device *dev) 655 { 656 } 657 658 #endif 659 660 static struct 661 config_group *nullb_group_make_group(struct config_group *group, const char *name) 662 { 663 struct nullb_device *dev; 664 665 if (null_find_dev_by_name(name)) 666 return ERR_PTR(-EEXIST); 667 668 dev = null_alloc_dev(); 669 if (!dev) 670 return ERR_PTR(-ENOMEM); 671 672 config_group_init_type_name(&dev->group, name, &nullb_device_type); 673 nullb_add_fault_config(dev); 674 675 return &dev->group; 676 } 677 678 static void 679 nullb_group_drop_item(struct config_group *group, struct config_item *item) 680 { 681 struct nullb_device *dev = to_nullb_device(item); 682 683 if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) { 684 mutex_lock(&lock); 685 dev->power = false; 686 null_del_dev(dev->nullb); 687 mutex_unlock(&lock); 688 } 689 690 config_item_put(item); 691 } 692 693 static ssize_t memb_group_features_show(struct config_item *item, char *page) 694 { 695 return snprintf(page, PAGE_SIZE, 696 "badblocks,blocking,blocksize,cache_size,fua," 697 "completion_nsec,discard,home_node,hw_queue_depth," 698 "irqmode,max_sectors,mbps,memory_backed,no_sched," 699 "poll_queues,power,queue_mode,shared_tag_bitmap," 700 "shared_tags,size,submit_queues,use_per_node_hctx," 701 "virt_boundary,zoned,zone_capacity,zone_max_active," 702 "zone_max_open,zone_nr_conv,zone_offline,zone_readonly," 703 "zone_size,zone_append_max_sectors\n"); 704 } 705 706 CONFIGFS_ATTR_RO(memb_group_, features); 707 708 static struct configfs_attribute *nullb_group_attrs[] = { 709 &memb_group_attr_features, 710 NULL, 711 }; 712 713 static struct configfs_group_operations nullb_group_ops = { 714 .make_group = nullb_group_make_group, 715 .drop_item = nullb_group_drop_item, 716 }; 717 718 static const struct config_item_type nullb_group_type = { 719 .ct_group_ops = &nullb_group_ops, 720 .ct_attrs = nullb_group_attrs, 721 .ct_owner = THIS_MODULE, 722 }; 723 724 static struct configfs_subsystem nullb_subsys = { 725 .su_group = { 726 .cg_item = { 727 .ci_namebuf = "nullb", 728 .ci_type = &nullb_group_type, 729 }, 730 }, 731 }; 732 733 static inline int null_cache_active(struct nullb *nullb) 734 { 735 return test_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags); 736 } 737 738 static struct nullb_device *null_alloc_dev(void) 739 { 740 struct nullb_device *dev; 741 742 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 743 if (!dev) 744 return NULL; 745 746 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION 747 dev->timeout_config.attr = null_timeout_attr; 748 dev->requeue_config.attr = null_requeue_attr; 749 dev->init_hctx_fault_config.attr = null_init_hctx_attr; 750 #endif 751 752 INIT_RADIX_TREE(&dev->data, GFP_ATOMIC); 753 INIT_RADIX_TREE(&dev->cache, GFP_ATOMIC); 754 if (badblocks_init(&dev->badblocks, 0)) { 755 kfree(dev); 756 return NULL; 757 } 758 759 dev->size = g_gb * 1024; 760 dev->completion_nsec = g_completion_nsec; 761 dev->submit_queues = g_submit_queues; 762 dev->prev_submit_queues = g_submit_queues; 763 dev->poll_queues = g_poll_queues; 764 dev->prev_poll_queues = g_poll_queues; 765 dev->home_node = g_home_node; 766 dev->queue_mode = g_queue_mode; 767 dev->blocksize = g_bs; 768 dev->max_sectors = g_max_sectors; 769 dev->irqmode = g_irqmode; 770 dev->hw_queue_depth = g_hw_queue_depth; 771 dev->blocking = g_blocking; 772 dev->memory_backed = g_memory_backed; 773 dev->discard = g_discard; 774 dev->cache_size = g_cache_size; 775 dev->mbps = g_mbps; 776 dev->use_per_node_hctx = g_use_per_node_hctx; 777 dev->zoned = g_zoned; 778 dev->zone_size = g_zone_size; 779 dev->zone_capacity = g_zone_capacity; 780 dev->zone_nr_conv = g_zone_nr_conv; 781 dev->zone_max_open = g_zone_max_open; 782 dev->zone_max_active = g_zone_max_active; 783 dev->zone_append_max_sectors = g_zone_append_max_sectors; 784 dev->virt_boundary = g_virt_boundary; 785 dev->no_sched = g_no_sched; 786 dev->shared_tags = g_shared_tags; 787 dev->shared_tag_bitmap = g_shared_tag_bitmap; 788 dev->fua = g_fua; 789 790 return dev; 791 } 792 793 static void null_free_dev(struct nullb_device *dev) 794 { 795 if (!dev) 796 return; 797 798 null_free_zoned_dev(dev); 799 badblocks_exit(&dev->badblocks); 800 kfree(dev); 801 } 802 803 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer) 804 { 805 struct nullb_cmd *cmd = container_of(timer, struct nullb_cmd, timer); 806 807 blk_mq_end_request(blk_mq_rq_from_pdu(cmd), cmd->error); 808 return HRTIMER_NORESTART; 809 } 810 811 static void null_cmd_end_timer(struct nullb_cmd *cmd) 812 { 813 ktime_t kt = cmd->nq->dev->completion_nsec; 814 815 hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL); 816 } 817 818 static void null_complete_rq(struct request *rq) 819 { 820 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq); 821 822 blk_mq_end_request(rq, cmd->error); 823 } 824 825 static struct nullb_page *null_alloc_page(void) 826 { 827 struct nullb_page *t_page; 828 829 t_page = kmalloc(sizeof(struct nullb_page), GFP_NOIO); 830 if (!t_page) 831 return NULL; 832 833 t_page->page = alloc_pages(GFP_NOIO, 0); 834 if (!t_page->page) { 835 kfree(t_page); 836 return NULL; 837 } 838 839 memset(t_page->bitmap, 0, sizeof(t_page->bitmap)); 840 return t_page; 841 } 842 843 static void null_free_page(struct nullb_page *t_page) 844 { 845 __set_bit(NULLB_PAGE_FREE, t_page->bitmap); 846 if (test_bit(NULLB_PAGE_LOCK, t_page->bitmap)) 847 return; 848 __free_page(t_page->page); 849 kfree(t_page); 850 } 851 852 static bool null_page_empty(struct nullb_page *page) 853 { 854 int size = MAP_SZ - 2; 855 856 return find_first_bit(page->bitmap, size) == size; 857 } 858 859 static void null_free_sector(struct nullb *nullb, sector_t sector, 860 bool is_cache) 861 { 862 unsigned int sector_bit; 863 u64 idx; 864 struct nullb_page *t_page, *ret; 865 struct radix_tree_root *root; 866 867 root = is_cache ? &nullb->dev->cache : &nullb->dev->data; 868 idx = sector >> PAGE_SECTORS_SHIFT; 869 sector_bit = (sector & SECTOR_MASK); 870 871 t_page = radix_tree_lookup(root, idx); 872 if (t_page) { 873 __clear_bit(sector_bit, t_page->bitmap); 874 875 if (null_page_empty(t_page)) { 876 ret = radix_tree_delete_item(root, idx, t_page); 877 WARN_ON(ret != t_page); 878 null_free_page(ret); 879 if (is_cache) 880 nullb->dev->curr_cache -= PAGE_SIZE; 881 } 882 } 883 } 884 885 static struct nullb_page *null_radix_tree_insert(struct nullb *nullb, u64 idx, 886 struct nullb_page *t_page, bool is_cache) 887 { 888 struct radix_tree_root *root; 889 890 root = is_cache ? &nullb->dev->cache : &nullb->dev->data; 891 892 if (radix_tree_insert(root, idx, t_page)) { 893 null_free_page(t_page); 894 t_page = radix_tree_lookup(root, idx); 895 WARN_ON(!t_page || t_page->page->index != idx); 896 } else if (is_cache) 897 nullb->dev->curr_cache += PAGE_SIZE; 898 899 return t_page; 900 } 901 902 static void null_free_device_storage(struct nullb_device *dev, bool is_cache) 903 { 904 unsigned long pos = 0; 905 int nr_pages; 906 struct nullb_page *ret, *t_pages[FREE_BATCH]; 907 struct radix_tree_root *root; 908 909 root = is_cache ? &dev->cache : &dev->data; 910 911 do { 912 int i; 913 914 nr_pages = radix_tree_gang_lookup(root, 915 (void **)t_pages, pos, FREE_BATCH); 916 917 for (i = 0; i < nr_pages; i++) { 918 pos = t_pages[i]->page->index; 919 ret = radix_tree_delete_item(root, pos, t_pages[i]); 920 WARN_ON(ret != t_pages[i]); 921 null_free_page(ret); 922 } 923 924 pos++; 925 } while (nr_pages == FREE_BATCH); 926 927 if (is_cache) 928 dev->curr_cache = 0; 929 } 930 931 static struct nullb_page *__null_lookup_page(struct nullb *nullb, 932 sector_t sector, bool for_write, bool is_cache) 933 { 934 unsigned int sector_bit; 935 u64 idx; 936 struct nullb_page *t_page; 937 struct radix_tree_root *root; 938 939 idx = sector >> PAGE_SECTORS_SHIFT; 940 sector_bit = (sector & SECTOR_MASK); 941 942 root = is_cache ? &nullb->dev->cache : &nullb->dev->data; 943 t_page = radix_tree_lookup(root, idx); 944 WARN_ON(t_page && t_page->page->index != idx); 945 946 if (t_page && (for_write || test_bit(sector_bit, t_page->bitmap))) 947 return t_page; 948 949 return NULL; 950 } 951 952 static struct nullb_page *null_lookup_page(struct nullb *nullb, 953 sector_t sector, bool for_write, bool ignore_cache) 954 { 955 struct nullb_page *page = NULL; 956 957 if (!ignore_cache) 958 page = __null_lookup_page(nullb, sector, for_write, true); 959 if (page) 960 return page; 961 return __null_lookup_page(nullb, sector, for_write, false); 962 } 963 964 static struct nullb_page *null_insert_page(struct nullb *nullb, 965 sector_t sector, bool ignore_cache) 966 __releases(&nullb->lock) 967 __acquires(&nullb->lock) 968 { 969 u64 idx; 970 struct nullb_page *t_page; 971 972 t_page = null_lookup_page(nullb, sector, true, ignore_cache); 973 if (t_page) 974 return t_page; 975 976 spin_unlock_irq(&nullb->lock); 977 978 t_page = null_alloc_page(); 979 if (!t_page) 980 goto out_lock; 981 982 if (radix_tree_preload(GFP_NOIO)) 983 goto out_freepage; 984 985 spin_lock_irq(&nullb->lock); 986 idx = sector >> PAGE_SECTORS_SHIFT; 987 t_page->page->index = idx; 988 t_page = null_radix_tree_insert(nullb, idx, t_page, !ignore_cache); 989 radix_tree_preload_end(); 990 991 return t_page; 992 out_freepage: 993 null_free_page(t_page); 994 out_lock: 995 spin_lock_irq(&nullb->lock); 996 return null_lookup_page(nullb, sector, true, ignore_cache); 997 } 998 999 static int null_flush_cache_page(struct nullb *nullb, struct nullb_page *c_page) 1000 { 1001 int i; 1002 unsigned int offset; 1003 u64 idx; 1004 struct nullb_page *t_page, *ret; 1005 void *dst, *src; 1006 1007 idx = c_page->page->index; 1008 1009 t_page = null_insert_page(nullb, idx << PAGE_SECTORS_SHIFT, true); 1010 1011 __clear_bit(NULLB_PAGE_LOCK, c_page->bitmap); 1012 if (test_bit(NULLB_PAGE_FREE, c_page->bitmap)) { 1013 null_free_page(c_page); 1014 if (t_page && null_page_empty(t_page)) { 1015 ret = radix_tree_delete_item(&nullb->dev->data, 1016 idx, t_page); 1017 null_free_page(t_page); 1018 } 1019 return 0; 1020 } 1021 1022 if (!t_page) 1023 return -ENOMEM; 1024 1025 src = kmap_local_page(c_page->page); 1026 dst = kmap_local_page(t_page->page); 1027 1028 for (i = 0; i < PAGE_SECTORS; 1029 i += (nullb->dev->blocksize >> SECTOR_SHIFT)) { 1030 if (test_bit(i, c_page->bitmap)) { 1031 offset = (i << SECTOR_SHIFT); 1032 memcpy(dst + offset, src + offset, 1033 nullb->dev->blocksize); 1034 __set_bit(i, t_page->bitmap); 1035 } 1036 } 1037 1038 kunmap_local(dst); 1039 kunmap_local(src); 1040 1041 ret = radix_tree_delete_item(&nullb->dev->cache, idx, c_page); 1042 null_free_page(ret); 1043 nullb->dev->curr_cache -= PAGE_SIZE; 1044 1045 return 0; 1046 } 1047 1048 static int null_make_cache_space(struct nullb *nullb, unsigned long n) 1049 { 1050 int i, err, nr_pages; 1051 struct nullb_page *c_pages[FREE_BATCH]; 1052 unsigned long flushed = 0, one_round; 1053 1054 again: 1055 if ((nullb->dev->cache_size * 1024 * 1024) > 1056 nullb->dev->curr_cache + n || nullb->dev->curr_cache == 0) 1057 return 0; 1058 1059 nr_pages = radix_tree_gang_lookup(&nullb->dev->cache, 1060 (void **)c_pages, nullb->cache_flush_pos, FREE_BATCH); 1061 /* 1062 * nullb_flush_cache_page could unlock before using the c_pages. To 1063 * avoid race, we don't allow page free 1064 */ 1065 for (i = 0; i < nr_pages; i++) { 1066 nullb->cache_flush_pos = c_pages[i]->page->index; 1067 /* 1068 * We found the page which is being flushed to disk by other 1069 * threads 1070 */ 1071 if (test_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap)) 1072 c_pages[i] = NULL; 1073 else 1074 __set_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap); 1075 } 1076 1077 one_round = 0; 1078 for (i = 0; i < nr_pages; i++) { 1079 if (c_pages[i] == NULL) 1080 continue; 1081 err = null_flush_cache_page(nullb, c_pages[i]); 1082 if (err) 1083 return err; 1084 one_round++; 1085 } 1086 flushed += one_round << PAGE_SHIFT; 1087 1088 if (n > flushed) { 1089 if (nr_pages == 0) 1090 nullb->cache_flush_pos = 0; 1091 if (one_round == 0) { 1092 /* give other threads a chance */ 1093 spin_unlock_irq(&nullb->lock); 1094 spin_lock_irq(&nullb->lock); 1095 } 1096 goto again; 1097 } 1098 return 0; 1099 } 1100 1101 static int copy_to_nullb(struct nullb *nullb, struct page *source, 1102 unsigned int off, sector_t sector, size_t n, bool is_fua) 1103 { 1104 size_t temp, count = 0; 1105 unsigned int offset; 1106 struct nullb_page *t_page; 1107 1108 while (count < n) { 1109 temp = min_t(size_t, nullb->dev->blocksize, n - count); 1110 1111 if (null_cache_active(nullb) && !is_fua) 1112 null_make_cache_space(nullb, PAGE_SIZE); 1113 1114 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT; 1115 t_page = null_insert_page(nullb, sector, 1116 !null_cache_active(nullb) || is_fua); 1117 if (!t_page) 1118 return -ENOSPC; 1119 1120 memcpy_page(t_page->page, offset, source, off + count, temp); 1121 1122 __set_bit(sector & SECTOR_MASK, t_page->bitmap); 1123 1124 if (is_fua) 1125 null_free_sector(nullb, sector, true); 1126 1127 count += temp; 1128 sector += temp >> SECTOR_SHIFT; 1129 } 1130 return 0; 1131 } 1132 1133 static int copy_from_nullb(struct nullb *nullb, struct page *dest, 1134 unsigned int off, sector_t sector, size_t n) 1135 { 1136 size_t temp, count = 0; 1137 unsigned int offset; 1138 struct nullb_page *t_page; 1139 1140 while (count < n) { 1141 temp = min_t(size_t, nullb->dev->blocksize, n - count); 1142 1143 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT; 1144 t_page = null_lookup_page(nullb, sector, false, 1145 !null_cache_active(nullb)); 1146 1147 if (t_page) 1148 memcpy_page(dest, off + count, t_page->page, offset, 1149 temp); 1150 else 1151 zero_user(dest, off + count, temp); 1152 1153 count += temp; 1154 sector += temp >> SECTOR_SHIFT; 1155 } 1156 return 0; 1157 } 1158 1159 static void nullb_fill_pattern(struct nullb *nullb, struct page *page, 1160 unsigned int len, unsigned int off) 1161 { 1162 memset_page(page, off, 0xff, len); 1163 } 1164 1165 blk_status_t null_handle_discard(struct nullb_device *dev, 1166 sector_t sector, sector_t nr_sectors) 1167 { 1168 struct nullb *nullb = dev->nullb; 1169 size_t n = nr_sectors << SECTOR_SHIFT; 1170 size_t temp; 1171 1172 spin_lock_irq(&nullb->lock); 1173 while (n > 0) { 1174 temp = min_t(size_t, n, dev->blocksize); 1175 null_free_sector(nullb, sector, false); 1176 if (null_cache_active(nullb)) 1177 null_free_sector(nullb, sector, true); 1178 sector += temp >> SECTOR_SHIFT; 1179 n -= temp; 1180 } 1181 spin_unlock_irq(&nullb->lock); 1182 1183 return BLK_STS_OK; 1184 } 1185 1186 static blk_status_t null_handle_flush(struct nullb *nullb) 1187 { 1188 int err; 1189 1190 if (!null_cache_active(nullb)) 1191 return 0; 1192 1193 spin_lock_irq(&nullb->lock); 1194 while (true) { 1195 err = null_make_cache_space(nullb, 1196 nullb->dev->cache_size * 1024 * 1024); 1197 if (err || nullb->dev->curr_cache == 0) 1198 break; 1199 } 1200 1201 WARN_ON(!radix_tree_empty(&nullb->dev->cache)); 1202 spin_unlock_irq(&nullb->lock); 1203 return errno_to_blk_status(err); 1204 } 1205 1206 static int null_transfer(struct nullb *nullb, struct page *page, 1207 unsigned int len, unsigned int off, bool is_write, sector_t sector, 1208 bool is_fua) 1209 { 1210 struct nullb_device *dev = nullb->dev; 1211 unsigned int valid_len = len; 1212 int err = 0; 1213 1214 if (!is_write) { 1215 if (dev->zoned) 1216 valid_len = null_zone_valid_read_len(nullb, 1217 sector, len); 1218 1219 if (valid_len) { 1220 err = copy_from_nullb(nullb, page, off, 1221 sector, valid_len); 1222 off += valid_len; 1223 len -= valid_len; 1224 } 1225 1226 if (len) 1227 nullb_fill_pattern(nullb, page, len, off); 1228 flush_dcache_page(page); 1229 } else { 1230 flush_dcache_page(page); 1231 err = copy_to_nullb(nullb, page, off, sector, len, is_fua); 1232 } 1233 1234 return err; 1235 } 1236 1237 static blk_status_t null_handle_rq(struct nullb_cmd *cmd) 1238 { 1239 struct request *rq = blk_mq_rq_from_pdu(cmd); 1240 struct nullb *nullb = cmd->nq->dev->nullb; 1241 int err = 0; 1242 unsigned int len; 1243 sector_t sector = blk_rq_pos(rq); 1244 struct req_iterator iter; 1245 struct bio_vec bvec; 1246 1247 spin_lock_irq(&nullb->lock); 1248 rq_for_each_segment(bvec, rq, iter) { 1249 len = bvec.bv_len; 1250 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset, 1251 op_is_write(req_op(rq)), sector, 1252 rq->cmd_flags & REQ_FUA); 1253 if (err) 1254 break; 1255 sector += len >> SECTOR_SHIFT; 1256 } 1257 spin_unlock_irq(&nullb->lock); 1258 1259 return errno_to_blk_status(err); 1260 } 1261 1262 static inline blk_status_t null_handle_throttled(struct nullb_cmd *cmd) 1263 { 1264 struct nullb_device *dev = cmd->nq->dev; 1265 struct nullb *nullb = dev->nullb; 1266 blk_status_t sts = BLK_STS_OK; 1267 struct request *rq = blk_mq_rq_from_pdu(cmd); 1268 1269 if (!hrtimer_active(&nullb->bw_timer)) 1270 hrtimer_restart(&nullb->bw_timer); 1271 1272 if (atomic_long_sub_return(blk_rq_bytes(rq), &nullb->cur_bytes) < 0) { 1273 blk_mq_stop_hw_queues(nullb->q); 1274 /* race with timer */ 1275 if (atomic_long_read(&nullb->cur_bytes) > 0) 1276 blk_mq_start_stopped_hw_queues(nullb->q, true); 1277 /* requeue request */ 1278 sts = BLK_STS_DEV_RESOURCE; 1279 } 1280 return sts; 1281 } 1282 1283 static inline blk_status_t null_handle_badblocks(struct nullb_cmd *cmd, 1284 sector_t sector, 1285 sector_t nr_sectors) 1286 { 1287 struct badblocks *bb = &cmd->nq->dev->badblocks; 1288 sector_t first_bad; 1289 int bad_sectors; 1290 1291 if (badblocks_check(bb, sector, nr_sectors, &first_bad, &bad_sectors)) 1292 return BLK_STS_IOERR; 1293 1294 return BLK_STS_OK; 1295 } 1296 1297 static inline blk_status_t null_handle_memory_backed(struct nullb_cmd *cmd, 1298 enum req_op op, 1299 sector_t sector, 1300 sector_t nr_sectors) 1301 { 1302 struct nullb_device *dev = cmd->nq->dev; 1303 1304 if (op == REQ_OP_DISCARD) 1305 return null_handle_discard(dev, sector, nr_sectors); 1306 1307 return null_handle_rq(cmd); 1308 } 1309 1310 static void nullb_zero_read_cmd_buffer(struct nullb_cmd *cmd) 1311 { 1312 struct request *rq = blk_mq_rq_from_pdu(cmd); 1313 struct nullb_device *dev = cmd->nq->dev; 1314 struct bio *bio; 1315 1316 if (!dev->memory_backed && req_op(rq) == REQ_OP_READ) { 1317 __rq_for_each_bio(bio, rq) 1318 zero_fill_bio(bio); 1319 } 1320 } 1321 1322 static inline void nullb_complete_cmd(struct nullb_cmd *cmd) 1323 { 1324 struct request *rq = blk_mq_rq_from_pdu(cmd); 1325 1326 /* 1327 * Since root privileges are required to configure the null_blk 1328 * driver, it is fine that this driver does not initialize the 1329 * data buffers of read commands. Zero-initialize these buffers 1330 * anyway if KMSAN is enabled to prevent that KMSAN complains 1331 * about null_blk not initializing read data buffers. 1332 */ 1333 if (IS_ENABLED(CONFIG_KMSAN)) 1334 nullb_zero_read_cmd_buffer(cmd); 1335 1336 /* Complete IO by inline, softirq or timer */ 1337 switch (cmd->nq->dev->irqmode) { 1338 case NULL_IRQ_SOFTIRQ: 1339 blk_mq_complete_request(rq); 1340 break; 1341 case NULL_IRQ_NONE: 1342 blk_mq_end_request(rq, cmd->error); 1343 break; 1344 case NULL_IRQ_TIMER: 1345 null_cmd_end_timer(cmd); 1346 break; 1347 } 1348 } 1349 1350 blk_status_t null_process_cmd(struct nullb_cmd *cmd, enum req_op op, 1351 sector_t sector, unsigned int nr_sectors) 1352 { 1353 struct nullb_device *dev = cmd->nq->dev; 1354 blk_status_t ret; 1355 1356 if (dev->badblocks.shift != -1) { 1357 ret = null_handle_badblocks(cmd, sector, nr_sectors); 1358 if (ret != BLK_STS_OK) 1359 return ret; 1360 } 1361 1362 if (dev->memory_backed) 1363 return null_handle_memory_backed(cmd, op, sector, nr_sectors); 1364 1365 return BLK_STS_OK; 1366 } 1367 1368 static void null_handle_cmd(struct nullb_cmd *cmd, sector_t sector, 1369 sector_t nr_sectors, enum req_op op) 1370 { 1371 struct nullb_device *dev = cmd->nq->dev; 1372 struct nullb *nullb = dev->nullb; 1373 blk_status_t sts; 1374 1375 if (op == REQ_OP_FLUSH) { 1376 cmd->error = null_handle_flush(nullb); 1377 goto out; 1378 } 1379 1380 if (dev->zoned) 1381 sts = null_process_zoned_cmd(cmd, op, sector, nr_sectors); 1382 else 1383 sts = null_process_cmd(cmd, op, sector, nr_sectors); 1384 1385 /* Do not overwrite errors (e.g. timeout errors) */ 1386 if (cmd->error == BLK_STS_OK) 1387 cmd->error = sts; 1388 1389 out: 1390 nullb_complete_cmd(cmd); 1391 } 1392 1393 static enum hrtimer_restart nullb_bwtimer_fn(struct hrtimer *timer) 1394 { 1395 struct nullb *nullb = container_of(timer, struct nullb, bw_timer); 1396 ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL); 1397 unsigned int mbps = nullb->dev->mbps; 1398 1399 if (atomic_long_read(&nullb->cur_bytes) == mb_per_tick(mbps)) 1400 return HRTIMER_NORESTART; 1401 1402 atomic_long_set(&nullb->cur_bytes, mb_per_tick(mbps)); 1403 blk_mq_start_stopped_hw_queues(nullb->q, true); 1404 1405 hrtimer_forward_now(&nullb->bw_timer, timer_interval); 1406 1407 return HRTIMER_RESTART; 1408 } 1409 1410 static void nullb_setup_bwtimer(struct nullb *nullb) 1411 { 1412 ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL); 1413 1414 hrtimer_init(&nullb->bw_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); 1415 nullb->bw_timer.function = nullb_bwtimer_fn; 1416 atomic_long_set(&nullb->cur_bytes, mb_per_tick(nullb->dev->mbps)); 1417 hrtimer_start(&nullb->bw_timer, timer_interval, HRTIMER_MODE_REL); 1418 } 1419 1420 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION 1421 1422 static bool should_timeout_request(struct request *rq) 1423 { 1424 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq); 1425 struct nullb_device *dev = cmd->nq->dev; 1426 1427 return should_fail(&dev->timeout_config.attr, 1); 1428 } 1429 1430 static bool should_requeue_request(struct request *rq) 1431 { 1432 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq); 1433 struct nullb_device *dev = cmd->nq->dev; 1434 1435 return should_fail(&dev->requeue_config.attr, 1); 1436 } 1437 1438 static bool should_init_hctx_fail(struct nullb_device *dev) 1439 { 1440 return should_fail(&dev->init_hctx_fault_config.attr, 1); 1441 } 1442 1443 #else 1444 1445 static bool should_timeout_request(struct request *rq) 1446 { 1447 return false; 1448 } 1449 1450 static bool should_requeue_request(struct request *rq) 1451 { 1452 return false; 1453 } 1454 1455 static bool should_init_hctx_fail(struct nullb_device *dev) 1456 { 1457 return false; 1458 } 1459 1460 #endif 1461 1462 static void null_map_queues(struct blk_mq_tag_set *set) 1463 { 1464 struct nullb *nullb = set->driver_data; 1465 int i, qoff; 1466 unsigned int submit_queues = g_submit_queues; 1467 unsigned int poll_queues = g_poll_queues; 1468 1469 if (nullb) { 1470 struct nullb_device *dev = nullb->dev; 1471 1472 /* 1473 * Refer nr_hw_queues of the tag set to check if the expected 1474 * number of hardware queues are prepared. If block layer failed 1475 * to prepare them, use previous numbers of submit queues and 1476 * poll queues to map queues. 1477 */ 1478 if (set->nr_hw_queues == 1479 dev->submit_queues + dev->poll_queues) { 1480 submit_queues = dev->submit_queues; 1481 poll_queues = dev->poll_queues; 1482 } else if (set->nr_hw_queues == 1483 dev->prev_submit_queues + dev->prev_poll_queues) { 1484 submit_queues = dev->prev_submit_queues; 1485 poll_queues = dev->prev_poll_queues; 1486 } else { 1487 pr_warn("tag set has unexpected nr_hw_queues: %d\n", 1488 set->nr_hw_queues); 1489 WARN_ON_ONCE(true); 1490 submit_queues = 1; 1491 poll_queues = 0; 1492 } 1493 } 1494 1495 for (i = 0, qoff = 0; i < set->nr_maps; i++) { 1496 struct blk_mq_queue_map *map = &set->map[i]; 1497 1498 switch (i) { 1499 case HCTX_TYPE_DEFAULT: 1500 map->nr_queues = submit_queues; 1501 break; 1502 case HCTX_TYPE_READ: 1503 map->nr_queues = 0; 1504 continue; 1505 case HCTX_TYPE_POLL: 1506 map->nr_queues = poll_queues; 1507 break; 1508 } 1509 map->queue_offset = qoff; 1510 qoff += map->nr_queues; 1511 blk_mq_map_queues(map); 1512 } 1513 } 1514 1515 static int null_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob) 1516 { 1517 struct nullb_queue *nq = hctx->driver_data; 1518 LIST_HEAD(list); 1519 int nr = 0; 1520 struct request *rq; 1521 1522 spin_lock(&nq->poll_lock); 1523 list_splice_init(&nq->poll_list, &list); 1524 list_for_each_entry(rq, &list, queuelist) 1525 blk_mq_set_request_complete(rq); 1526 spin_unlock(&nq->poll_lock); 1527 1528 while (!list_empty(&list)) { 1529 struct nullb_cmd *cmd; 1530 struct request *req; 1531 1532 req = list_first_entry(&list, struct request, queuelist); 1533 list_del_init(&req->queuelist); 1534 cmd = blk_mq_rq_to_pdu(req); 1535 cmd->error = null_process_cmd(cmd, req_op(req), blk_rq_pos(req), 1536 blk_rq_sectors(req)); 1537 if (!blk_mq_add_to_batch(req, iob, (__force int) cmd->error, 1538 blk_mq_end_request_batch)) 1539 blk_mq_end_request(req, cmd->error); 1540 nr++; 1541 } 1542 1543 return nr; 1544 } 1545 1546 static enum blk_eh_timer_return null_timeout_rq(struct request *rq) 1547 { 1548 struct blk_mq_hw_ctx *hctx = rq->mq_hctx; 1549 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq); 1550 1551 if (hctx->type == HCTX_TYPE_POLL) { 1552 struct nullb_queue *nq = hctx->driver_data; 1553 1554 spin_lock(&nq->poll_lock); 1555 /* The request may have completed meanwhile. */ 1556 if (blk_mq_request_completed(rq)) { 1557 spin_unlock(&nq->poll_lock); 1558 return BLK_EH_DONE; 1559 } 1560 list_del_init(&rq->queuelist); 1561 spin_unlock(&nq->poll_lock); 1562 } 1563 1564 pr_info("rq %p timed out\n", rq); 1565 1566 /* 1567 * If the device is marked as blocking (i.e. memory backed or zoned 1568 * device), the submission path may be blocked waiting for resources 1569 * and cause real timeouts. For these real timeouts, the submission 1570 * path will complete the request using blk_mq_complete_request(). 1571 * Only fake timeouts need to execute blk_mq_complete_request() here. 1572 */ 1573 cmd->error = BLK_STS_TIMEOUT; 1574 if (cmd->fake_timeout || hctx->type == HCTX_TYPE_POLL) 1575 blk_mq_complete_request(rq); 1576 return BLK_EH_DONE; 1577 } 1578 1579 static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx, 1580 const struct blk_mq_queue_data *bd) 1581 { 1582 struct request *rq = bd->rq; 1583 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq); 1584 struct nullb_queue *nq = hctx->driver_data; 1585 sector_t nr_sectors = blk_rq_sectors(rq); 1586 sector_t sector = blk_rq_pos(rq); 1587 const bool is_poll = hctx->type == HCTX_TYPE_POLL; 1588 1589 might_sleep_if(hctx->flags & BLK_MQ_F_BLOCKING); 1590 1591 if (!is_poll && nq->dev->irqmode == NULL_IRQ_TIMER) { 1592 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); 1593 cmd->timer.function = null_cmd_timer_expired; 1594 } 1595 cmd->error = BLK_STS_OK; 1596 cmd->nq = nq; 1597 cmd->fake_timeout = should_timeout_request(rq) || 1598 blk_should_fake_timeout(rq->q); 1599 1600 if (should_requeue_request(rq)) { 1601 /* 1602 * Alternate between hitting the core BUSY path, and the 1603 * driver driven requeue path 1604 */ 1605 nq->requeue_selection++; 1606 if (nq->requeue_selection & 1) 1607 return BLK_STS_RESOURCE; 1608 blk_mq_requeue_request(rq, true); 1609 return BLK_STS_OK; 1610 } 1611 1612 if (test_bit(NULLB_DEV_FL_THROTTLED, &nq->dev->flags)) { 1613 blk_status_t sts = null_handle_throttled(cmd); 1614 1615 if (sts != BLK_STS_OK) 1616 return sts; 1617 } 1618 1619 blk_mq_start_request(rq); 1620 1621 if (is_poll) { 1622 spin_lock(&nq->poll_lock); 1623 list_add_tail(&rq->queuelist, &nq->poll_list); 1624 spin_unlock(&nq->poll_lock); 1625 return BLK_STS_OK; 1626 } 1627 if (cmd->fake_timeout) 1628 return BLK_STS_OK; 1629 1630 null_handle_cmd(cmd, sector, nr_sectors, req_op(rq)); 1631 return BLK_STS_OK; 1632 } 1633 1634 static void null_queue_rqs(struct request **rqlist) 1635 { 1636 struct request *requeue_list = NULL; 1637 struct request **requeue_lastp = &requeue_list; 1638 struct blk_mq_queue_data bd = { }; 1639 blk_status_t ret; 1640 1641 do { 1642 struct request *rq = rq_list_pop(rqlist); 1643 1644 bd.rq = rq; 1645 ret = null_queue_rq(rq->mq_hctx, &bd); 1646 if (ret != BLK_STS_OK) 1647 rq_list_add_tail(&requeue_lastp, rq); 1648 } while (!rq_list_empty(*rqlist)); 1649 1650 *rqlist = requeue_list; 1651 } 1652 1653 static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq) 1654 { 1655 nq->dev = nullb->dev; 1656 INIT_LIST_HEAD(&nq->poll_list); 1657 spin_lock_init(&nq->poll_lock); 1658 } 1659 1660 static int null_init_hctx(struct blk_mq_hw_ctx *hctx, void *driver_data, 1661 unsigned int hctx_idx) 1662 { 1663 struct nullb *nullb = hctx->queue->queuedata; 1664 struct nullb_queue *nq; 1665 1666 if (should_init_hctx_fail(nullb->dev)) 1667 return -EFAULT; 1668 1669 nq = &nullb->queues[hctx_idx]; 1670 hctx->driver_data = nq; 1671 null_init_queue(nullb, nq); 1672 1673 return 0; 1674 } 1675 1676 static const struct blk_mq_ops null_mq_ops = { 1677 .queue_rq = null_queue_rq, 1678 .queue_rqs = null_queue_rqs, 1679 .complete = null_complete_rq, 1680 .timeout = null_timeout_rq, 1681 .poll = null_poll, 1682 .map_queues = null_map_queues, 1683 .init_hctx = null_init_hctx, 1684 }; 1685 1686 static void null_del_dev(struct nullb *nullb) 1687 { 1688 struct nullb_device *dev; 1689 1690 if (!nullb) 1691 return; 1692 1693 dev = nullb->dev; 1694 1695 ida_free(&nullb_indexes, nullb->index); 1696 1697 list_del_init(&nullb->list); 1698 1699 del_gendisk(nullb->disk); 1700 1701 if (test_bit(NULLB_DEV_FL_THROTTLED, &nullb->dev->flags)) { 1702 hrtimer_cancel(&nullb->bw_timer); 1703 atomic_long_set(&nullb->cur_bytes, LONG_MAX); 1704 blk_mq_start_stopped_hw_queues(nullb->q, true); 1705 } 1706 1707 put_disk(nullb->disk); 1708 if (nullb->tag_set == &nullb->__tag_set) 1709 blk_mq_free_tag_set(nullb->tag_set); 1710 kfree(nullb->queues); 1711 if (null_cache_active(nullb)) 1712 null_free_device_storage(nullb->dev, true); 1713 kfree(nullb); 1714 dev->nullb = NULL; 1715 } 1716 1717 static void null_config_discard(struct nullb *nullb, struct queue_limits *lim) 1718 { 1719 if (nullb->dev->discard == false) 1720 return; 1721 1722 if (!nullb->dev->memory_backed) { 1723 nullb->dev->discard = false; 1724 pr_info("discard option is ignored without memory backing\n"); 1725 return; 1726 } 1727 1728 if (nullb->dev->zoned) { 1729 nullb->dev->discard = false; 1730 pr_info("discard option is ignored in zoned mode\n"); 1731 return; 1732 } 1733 1734 lim->max_hw_discard_sectors = UINT_MAX >> 9; 1735 } 1736 1737 static const struct block_device_operations null_ops = { 1738 .owner = THIS_MODULE, 1739 .report_zones = null_report_zones, 1740 }; 1741 1742 static int setup_queues(struct nullb *nullb) 1743 { 1744 int nqueues = nr_cpu_ids; 1745 1746 if (g_poll_queues) 1747 nqueues += g_poll_queues; 1748 1749 nullb->queues = kcalloc(nqueues, sizeof(struct nullb_queue), 1750 GFP_KERNEL); 1751 if (!nullb->queues) 1752 return -ENOMEM; 1753 1754 return 0; 1755 } 1756 1757 static int null_init_tag_set(struct blk_mq_tag_set *set, int poll_queues) 1758 { 1759 set->ops = &null_mq_ops; 1760 set->cmd_size = sizeof(struct nullb_cmd); 1761 set->timeout = 5 * HZ; 1762 set->nr_maps = 1; 1763 if (poll_queues) { 1764 set->nr_hw_queues += poll_queues; 1765 set->nr_maps += 2; 1766 } 1767 return blk_mq_alloc_tag_set(set); 1768 } 1769 1770 static int null_init_global_tag_set(void) 1771 { 1772 int error; 1773 1774 if (tag_set.ops) 1775 return 0; 1776 1777 tag_set.nr_hw_queues = g_submit_queues; 1778 tag_set.queue_depth = g_hw_queue_depth; 1779 tag_set.numa_node = g_home_node; 1780 tag_set.flags = BLK_MQ_F_SHOULD_MERGE; 1781 if (g_no_sched) 1782 tag_set.flags |= BLK_MQ_F_NO_SCHED; 1783 if (g_shared_tag_bitmap) 1784 tag_set.flags |= BLK_MQ_F_TAG_HCTX_SHARED; 1785 if (g_blocking) 1786 tag_set.flags |= BLK_MQ_F_BLOCKING; 1787 1788 error = null_init_tag_set(&tag_set, g_poll_queues); 1789 if (error) 1790 tag_set.ops = NULL; 1791 return error; 1792 } 1793 1794 static int null_setup_tagset(struct nullb *nullb) 1795 { 1796 if (nullb->dev->shared_tags) { 1797 nullb->tag_set = &tag_set; 1798 return null_init_global_tag_set(); 1799 } 1800 1801 nullb->tag_set = &nullb->__tag_set; 1802 nullb->tag_set->driver_data = nullb; 1803 nullb->tag_set->nr_hw_queues = nullb->dev->submit_queues; 1804 nullb->tag_set->queue_depth = nullb->dev->hw_queue_depth; 1805 nullb->tag_set->numa_node = nullb->dev->home_node; 1806 nullb->tag_set->flags = BLK_MQ_F_SHOULD_MERGE; 1807 if (nullb->dev->no_sched) 1808 nullb->tag_set->flags |= BLK_MQ_F_NO_SCHED; 1809 if (nullb->dev->shared_tag_bitmap) 1810 nullb->tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED; 1811 if (nullb->dev->blocking) 1812 nullb->tag_set->flags |= BLK_MQ_F_BLOCKING; 1813 return null_init_tag_set(nullb->tag_set, nullb->dev->poll_queues); 1814 } 1815 1816 static int null_validate_conf(struct nullb_device *dev) 1817 { 1818 if (dev->queue_mode == NULL_Q_RQ) { 1819 pr_err("legacy IO path is no longer available\n"); 1820 return -EINVAL; 1821 } 1822 if (dev->queue_mode == NULL_Q_BIO) { 1823 pr_err("BIO-based IO path is no longer available, using blk-mq instead.\n"); 1824 dev->queue_mode = NULL_Q_MQ; 1825 } 1826 1827 if (blk_validate_block_size(dev->blocksize)) 1828 return -EINVAL; 1829 1830 if (dev->use_per_node_hctx) { 1831 if (dev->submit_queues != nr_online_nodes) 1832 dev->submit_queues = nr_online_nodes; 1833 } else if (dev->submit_queues > nr_cpu_ids) 1834 dev->submit_queues = nr_cpu_ids; 1835 else if (dev->submit_queues == 0) 1836 dev->submit_queues = 1; 1837 dev->prev_submit_queues = dev->submit_queues; 1838 1839 if (dev->poll_queues > g_poll_queues) 1840 dev->poll_queues = g_poll_queues; 1841 dev->prev_poll_queues = dev->poll_queues; 1842 dev->irqmode = min_t(unsigned int, dev->irqmode, NULL_IRQ_TIMER); 1843 1844 /* Do memory allocation, so set blocking */ 1845 if (dev->memory_backed) 1846 dev->blocking = true; 1847 else /* cache is meaningless */ 1848 dev->cache_size = 0; 1849 dev->cache_size = min_t(unsigned long, ULONG_MAX / 1024 / 1024, 1850 dev->cache_size); 1851 dev->mbps = min_t(unsigned int, 1024 * 40, dev->mbps); 1852 1853 if (dev->zoned && 1854 (!dev->zone_size || !is_power_of_2(dev->zone_size))) { 1855 pr_err("zone_size must be power-of-two\n"); 1856 return -EINVAL; 1857 } 1858 1859 return 0; 1860 } 1861 1862 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION 1863 static bool __null_setup_fault(struct fault_attr *attr, char *str) 1864 { 1865 if (!str[0]) 1866 return true; 1867 1868 if (!setup_fault_attr(attr, str)) 1869 return false; 1870 1871 attr->verbose = 0; 1872 return true; 1873 } 1874 #endif 1875 1876 static bool null_setup_fault(void) 1877 { 1878 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION 1879 if (!__null_setup_fault(&null_timeout_attr, g_timeout_str)) 1880 return false; 1881 if (!__null_setup_fault(&null_requeue_attr, g_requeue_str)) 1882 return false; 1883 if (!__null_setup_fault(&null_init_hctx_attr, g_init_hctx_str)) 1884 return false; 1885 #endif 1886 return true; 1887 } 1888 1889 static int null_add_dev(struct nullb_device *dev) 1890 { 1891 struct queue_limits lim = { 1892 .logical_block_size = dev->blocksize, 1893 .physical_block_size = dev->blocksize, 1894 .max_hw_sectors = dev->max_sectors, 1895 }; 1896 1897 struct nullb *nullb; 1898 int rv; 1899 1900 rv = null_validate_conf(dev); 1901 if (rv) 1902 return rv; 1903 1904 nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, dev->home_node); 1905 if (!nullb) { 1906 rv = -ENOMEM; 1907 goto out; 1908 } 1909 nullb->dev = dev; 1910 dev->nullb = nullb; 1911 1912 spin_lock_init(&nullb->lock); 1913 1914 rv = setup_queues(nullb); 1915 if (rv) 1916 goto out_free_nullb; 1917 1918 rv = null_setup_tagset(nullb); 1919 if (rv) 1920 goto out_cleanup_queues; 1921 1922 if (dev->virt_boundary) 1923 lim.virt_boundary_mask = PAGE_SIZE - 1; 1924 null_config_discard(nullb, &lim); 1925 if (dev->zoned) { 1926 rv = null_init_zoned_dev(dev, &lim); 1927 if (rv) 1928 goto out_cleanup_tags; 1929 } 1930 1931 nullb->disk = blk_mq_alloc_disk(nullb->tag_set, &lim, nullb); 1932 if (IS_ERR(nullb->disk)) { 1933 rv = PTR_ERR(nullb->disk); 1934 goto out_cleanup_zone; 1935 } 1936 nullb->q = nullb->disk->queue; 1937 1938 if (dev->mbps) { 1939 set_bit(NULLB_DEV_FL_THROTTLED, &dev->flags); 1940 nullb_setup_bwtimer(nullb); 1941 } 1942 1943 if (dev->cache_size > 0) { 1944 set_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags); 1945 blk_queue_write_cache(nullb->q, true, dev->fua); 1946 } 1947 1948 nullb->q->queuedata = nullb; 1949 blk_queue_flag_set(QUEUE_FLAG_NONROT, nullb->q); 1950 1951 rv = ida_alloc(&nullb_indexes, GFP_KERNEL); 1952 if (rv < 0) 1953 goto out_cleanup_disk; 1954 1955 nullb->index = rv; 1956 dev->index = rv; 1957 1958 if (config_item_name(&dev->group.cg_item)) { 1959 /* Use configfs dir name as the device name */ 1960 snprintf(nullb->disk_name, sizeof(nullb->disk_name), 1961 "%s", config_item_name(&dev->group.cg_item)); 1962 } else { 1963 sprintf(nullb->disk_name, "nullb%d", nullb->index); 1964 } 1965 1966 set_capacity(nullb->disk, 1967 ((sector_t)nullb->dev->size * SZ_1M) >> SECTOR_SHIFT); 1968 nullb->disk->major = null_major; 1969 nullb->disk->first_minor = nullb->index; 1970 nullb->disk->minors = 1; 1971 nullb->disk->fops = &null_ops; 1972 nullb->disk->private_data = nullb; 1973 strscpy_pad(nullb->disk->disk_name, nullb->disk_name, DISK_NAME_LEN); 1974 1975 if (nullb->dev->zoned) { 1976 rv = null_register_zoned_dev(nullb); 1977 if (rv) 1978 goto out_ida_free; 1979 } 1980 1981 rv = add_disk(nullb->disk); 1982 if (rv) 1983 goto out_ida_free; 1984 1985 list_add_tail(&nullb->list, &nullb_list); 1986 1987 pr_info("disk %s created\n", nullb->disk_name); 1988 1989 return 0; 1990 1991 out_ida_free: 1992 ida_free(&nullb_indexes, nullb->index); 1993 out_cleanup_disk: 1994 put_disk(nullb->disk); 1995 out_cleanup_zone: 1996 null_free_zoned_dev(dev); 1997 out_cleanup_tags: 1998 if (nullb->tag_set == &nullb->__tag_set) 1999 blk_mq_free_tag_set(nullb->tag_set); 2000 out_cleanup_queues: 2001 kfree(nullb->queues); 2002 out_free_nullb: 2003 kfree(nullb); 2004 dev->nullb = NULL; 2005 out: 2006 return rv; 2007 } 2008 2009 static struct nullb *null_find_dev_by_name(const char *name) 2010 { 2011 struct nullb *nullb = NULL, *nb; 2012 2013 mutex_lock(&lock); 2014 list_for_each_entry(nb, &nullb_list, list) { 2015 if (strcmp(nb->disk_name, name) == 0) { 2016 nullb = nb; 2017 break; 2018 } 2019 } 2020 mutex_unlock(&lock); 2021 2022 return nullb; 2023 } 2024 2025 static int null_create_dev(void) 2026 { 2027 struct nullb_device *dev; 2028 int ret; 2029 2030 dev = null_alloc_dev(); 2031 if (!dev) 2032 return -ENOMEM; 2033 2034 mutex_lock(&lock); 2035 ret = null_add_dev(dev); 2036 mutex_unlock(&lock); 2037 if (ret) { 2038 null_free_dev(dev); 2039 return ret; 2040 } 2041 2042 return 0; 2043 } 2044 2045 static void null_destroy_dev(struct nullb *nullb) 2046 { 2047 struct nullb_device *dev = nullb->dev; 2048 2049 null_del_dev(nullb); 2050 null_free_device_storage(dev, false); 2051 null_free_dev(dev); 2052 } 2053 2054 static int __init null_init(void) 2055 { 2056 int ret = 0; 2057 unsigned int i; 2058 struct nullb *nullb; 2059 2060 if (g_bs > PAGE_SIZE) { 2061 pr_warn("invalid block size\n"); 2062 pr_warn("defaults block size to %lu\n", PAGE_SIZE); 2063 g_bs = PAGE_SIZE; 2064 } 2065 2066 if (g_home_node != NUMA_NO_NODE && g_home_node >= nr_online_nodes) { 2067 pr_err("invalid home_node value\n"); 2068 g_home_node = NUMA_NO_NODE; 2069 } 2070 2071 if (!null_setup_fault()) 2072 return -EINVAL; 2073 2074 if (g_queue_mode == NULL_Q_RQ) { 2075 pr_err("legacy IO path is no longer available\n"); 2076 return -EINVAL; 2077 } 2078 2079 if (g_use_per_node_hctx) { 2080 if (g_submit_queues != nr_online_nodes) { 2081 pr_warn("submit_queues param is set to %u.\n", 2082 nr_online_nodes); 2083 g_submit_queues = nr_online_nodes; 2084 } 2085 } else if (g_submit_queues > nr_cpu_ids) { 2086 g_submit_queues = nr_cpu_ids; 2087 } else if (g_submit_queues <= 0) { 2088 g_submit_queues = 1; 2089 } 2090 2091 config_group_init(&nullb_subsys.su_group); 2092 mutex_init(&nullb_subsys.su_mutex); 2093 2094 ret = configfs_register_subsystem(&nullb_subsys); 2095 if (ret) 2096 return ret; 2097 2098 mutex_init(&lock); 2099 2100 null_major = register_blkdev(0, "nullb"); 2101 if (null_major < 0) { 2102 ret = null_major; 2103 goto err_conf; 2104 } 2105 2106 for (i = 0; i < nr_devices; i++) { 2107 ret = null_create_dev(); 2108 if (ret) 2109 goto err_dev; 2110 } 2111 2112 pr_info("module loaded\n"); 2113 return 0; 2114 2115 err_dev: 2116 while (!list_empty(&nullb_list)) { 2117 nullb = list_entry(nullb_list.next, struct nullb, list); 2118 null_destroy_dev(nullb); 2119 } 2120 unregister_blkdev(null_major, "nullb"); 2121 err_conf: 2122 configfs_unregister_subsystem(&nullb_subsys); 2123 return ret; 2124 } 2125 2126 static void __exit null_exit(void) 2127 { 2128 struct nullb *nullb; 2129 2130 configfs_unregister_subsystem(&nullb_subsys); 2131 2132 unregister_blkdev(null_major, "nullb"); 2133 2134 mutex_lock(&lock); 2135 while (!list_empty(&nullb_list)) { 2136 nullb = list_entry(nullb_list.next, struct nullb, list); 2137 null_destroy_dev(nullb); 2138 } 2139 mutex_unlock(&lock); 2140 2141 if (tag_set.ops) 2142 blk_mq_free_tag_set(&tag_set); 2143 2144 mutex_destroy(&lock); 2145 } 2146 2147 module_init(null_init); 2148 module_exit(null_exit); 2149 2150 MODULE_AUTHOR("Jens Axboe <axboe@kernel.dk>"); 2151 MODULE_DESCRIPTION("multi queue aware block test driver"); 2152 MODULE_LICENSE("GPL"); 2153