drivers/block/null_blk_main.c

   1 /*
   2  * Add configfs and memory store: Kyungchan Koh <kkc6196@fb.com> and
   3  * Shaohua Li <shli@fb.com>
   4  */
   5 #include <linux/module.h>
   6
   7 #include <linux/moduleparam.h>
   8 #include <linux/sched.h>
   9 #include <linux/fs.h>
  10 #include <linux/init.h>
  11 #include "null_blk.h"
  12
  13 #define PAGE_SECTORS_SHIFT      (PAGE_SHIFT - SECTOR_SHIFT)
  14 #define PAGE_SECTORS            (1 << PAGE_SECTORS_SHIFT)
  15 #define SECTOR_MASK             (PAGE_SECTORS - 1)
  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 #endif
  26
  27 static inline u64 mb_per_tick(int mbps)
  28 {
  29         return (1 << 20) / TICKS_PER_SEC * ((u64) mbps);
  30 }
  31
  32 /*
  33  * Status flags for nullb_device.
  34  *
  35  * CONFIGURED:  Device has been configured and turned on. Cannot reconfigure.
  36  * UP:          Device is currently on and visible in userspace.
  37  * THROTTLED:   Device is being throttled.
  38  * CACHE:       Device is using a write-back cache.
  39  */
  40 enum nullb_device_flags {
  41         NULLB_DEV_FL_CONFIGURED = 0,
  42         NULLB_DEV_FL_UP         = 1,
  43         NULLB_DEV_FL_THROTTLED  = 2,
  44         NULLB_DEV_FL_CACHE      = 3,
  45 };
  46
  47 #define MAP_SZ          ((PAGE_SIZE >> SECTOR_SHIFT) + 2)
  48 /*
  49  * nullb_page is a page in memory for nullb devices.
  50  *
  51  * @page:       The page holding the data.
  52  * @bitmap:     The bitmap represents which sector in the page has data.
  53  *              Each bit represents one block size. For example, sector 8
  54  *              will use the 7th bit
  55  * The highest 2 bits of bitmap are for special purpose. LOCK means the cache
  56  * page is being flushing to storage. FREE means the cache page is freed and
  57  * should be skipped from flushing to storage. Please see
  58  * null_make_cache_space
  59  */
  60 struct nullb_page {
  61         struct page *page;
  62         DECLARE_BITMAP(bitmap, MAP_SZ);
  63 };
  64 #define NULLB_PAGE_LOCK (MAP_SZ - 1)
  65 #define NULLB_PAGE_FREE (MAP_SZ - 2)
  66
  67 static LIST_HEAD(nullb_list);
  68 static struct mutex lock;
  69 static int null_major;
  70 static DEFINE_IDA(nullb_indexes);
  71 static struct blk_mq_tag_set tag_set;
  72
  73 enum {
  74         NULL_IRQ_NONE           = 0,
  75         NULL_IRQ_SOFTIRQ        = 1,
  76         NULL_IRQ_TIMER          = 2,
  77 };
  78
  79 enum {
  80         NULL_Q_BIO              = 0,
  81         NULL_Q_RQ               = 1,
  82         NULL_Q_MQ               = 2,
  83 };
  84
  85 static int g_no_sched;
  86 module_param_named(no_sched, g_no_sched, int, 0444);
  87 MODULE_PARM_DESC(no_sched, "No io scheduler");
  88
  89 static int g_submit_queues = 1;
  90 module_param_named(submit_queues, g_submit_queues, int, 0444);
  91 MODULE_PARM_DESC(submit_queues, "Number of submission queues");
  92
  93 static int g_home_node = NUMA_NO_NODE;
  94 module_param_named(home_node, g_home_node, int, 0444);
  95 MODULE_PARM_DESC(home_node, "Home node for the device");
  96
  97 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
  98 static char g_timeout_str[80];
  99 module_param_string(timeout, g_timeout_str, sizeof(g_timeout_str), 0444);
 100
 101 static char g_requeue_str[80];
 102 module_param_string(requeue, g_requeue_str, sizeof(g_requeue_str), 0444);
 103 #endif
 104
 105 static int g_queue_mode = NULL_Q_MQ;
 106
 107 static int null_param_store_val(const char *str, int *val, int min, int max)
 108 {
 109         int ret, new_val;
 110
 111         ret = kstrtoint(str, 10, &new_val);
 112         if (ret)
 113                 return -EINVAL;
 114
 115         if (new_val < min || new_val > max)
 116                 return -EINVAL;
 117
 118         *val = new_val;
 119         return 0;
 120 }
 121
 122 static int null_set_queue_mode(const char *str, const struct kernel_param *kp)
 123 {
 124         return null_param_store_val(str, &g_queue_mode, NULL_Q_BIO, NULL_Q_MQ);
 125 }
 126
 127 static const struct kernel_param_ops null_queue_mode_param_ops = {
 128         .set    = null_set_queue_mode,
 129         .get    = param_get_int,
 130 };
 131
 132 device_param_cb(queue_mode, &null_queue_mode_param_ops, &g_queue_mode, 0444);
 133 MODULE_PARM_DESC(queue_mode, "Block interface to use (0=bio,1=rq,2=multiqueue)");
 134
 135 static int g_gb = 250;
 136 module_param_named(gb, g_gb, int, 0444);
 137 MODULE_PARM_DESC(gb, "Size in GB");
 138
 139 static int g_bs = 512;
 140 module_param_named(bs, g_bs, int, 0444);
 141 MODULE_PARM_DESC(bs, "Block size (in bytes)");
 142
 143 static int nr_devices = 1;
 144 module_param(nr_devices, int, 0444);
 145 MODULE_PARM_DESC(nr_devices, "Number of devices to register");
 146
 147 static bool g_blocking;
 148 module_param_named(blocking, g_blocking, bool, 0444);
 149 MODULE_PARM_DESC(blocking, "Register as a blocking blk-mq driver device");
 150
 151 static bool shared_tags;
 152 module_param(shared_tags, bool, 0444);
 153 MODULE_PARM_DESC(shared_tags, "Share tag set between devices for blk-mq");
 154
 155 static int g_irqmode = NULL_IRQ_SOFTIRQ;
 156
 157 static int null_set_irqmode(const char *str, const struct kernel_param *kp)
 158 {
 159         return null_param_store_val(str, &g_irqmode, NULL_IRQ_NONE,
 160                                         NULL_IRQ_TIMER);
 161 }
 162
 163 static const struct kernel_param_ops null_irqmode_param_ops = {
 164         .set    = null_set_irqmode,
 165         .get    = param_get_int,
 166 };
 167
 168 device_param_cb(irqmode, &null_irqmode_param_ops, &g_irqmode, 0444);
 169 MODULE_PARM_DESC(irqmode, "IRQ completion handler. 0-none, 1-softirq, 2-timer");
 170
 171 static unsigned long g_completion_nsec = 10000;
 172 module_param_named(completion_nsec, g_completion_nsec, ulong, 0444);
 173 MODULE_PARM_DESC(completion_nsec, "Time in ns to complete a request in hardware. Default: 10,000ns");
 174
 175 static int g_hw_queue_depth = 64;
 176 module_param_named(hw_queue_depth, g_hw_queue_depth, int, 0444);
 177 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 64");
 178
 179 static bool g_use_per_node_hctx;
 180 module_param_named(use_per_node_hctx, g_use_per_node_hctx, bool, 0444);
 181 MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware context queues. Default: false");
 182
 183 static bool g_zoned;
 184 module_param_named(zoned, g_zoned, bool, S_IRUGO);
 185 MODULE_PARM_DESC(zoned, "Make device as a host-managed zoned block device. Default: false");
 186
 187 static unsigned long g_zone_size = 256;
 188 module_param_named(zone_size, g_zone_size, ulong, S_IRUGO);
 189 MODULE_PARM_DESC(zone_size, "Zone size in MB when block device is zoned. Must be power-of-two: Default: 256");
 190
 191 static struct nullb_device *null_alloc_dev(void);
 192 static void null_free_dev(struct nullb_device *dev);
 193 static void null_del_dev(struct nullb *nullb);
 194 static int null_add_dev(struct nullb_device *dev);
 195 static void null_free_device_storage(struct nullb_device *dev, bool is_cache);
 196
 197 static inline struct nullb_device *to_nullb_device(struct config_item *item)
 198 {
 199         return item ? container_of(item, struct nullb_device, item) : NULL;
 200 }
 201
 202 static inline ssize_t nullb_device_uint_attr_show(unsigned int val, char *page)
 203 {
 204         return snprintf(page, PAGE_SIZE, "%u\n", val);
 205 }
 206
 207 static inline ssize_t nullb_device_ulong_attr_show(unsigned long val,
 208         char *page)
 209 {
 210         return snprintf(page, PAGE_SIZE, "%lu\n", val);
 211 }
 212
 213 static inline ssize_t nullb_device_bool_attr_show(bool val, char *page)
 214 {
 215         return snprintf(page, PAGE_SIZE, "%u\n", val);
 216 }
 217
 218 static ssize_t nullb_device_uint_attr_store(unsigned int *val,
 219         const char *page, size_t count)
 220 {
 221         unsigned int tmp;
 222         int result;
 223
 224         result = kstrtouint(page, 0, &tmp);
 225         if (result)
 226                 return result;
 227
 228         *val = tmp;
 229         return count;
 230 }
 231
 232 static ssize_t nullb_device_ulong_attr_store(unsigned long *val,
 233         const char *page, size_t count)
 234 {
 235         int result;
 236         unsigned long tmp;
 237
 238         result = kstrtoul(page, 0, &tmp);
 239         if (result)
 240                 return result;
 241
 242         *val = tmp;
 243         return count;
 244 }
 245
 246 static ssize_t nullb_device_bool_attr_store(bool *val, const char *page,
 247         size_t count)
 248 {
 249         bool tmp;
 250         int result;
 251
 252         result = kstrtobool(page,  &tmp);
 253         if (result)
 254                 return result;
 255
 256         *val = tmp;
 257         return count;
 258 }
 259
 260 /* The following macro should only be used with TYPE = {uint, ulong, bool}. */
 261 #define NULLB_DEVICE_ATTR(NAME, TYPE)                                           \
 262 static ssize_t                                                                  \
 263 nullb_device_##NAME##_show(struct config_item *item, char *page)                \
 264 {                                                                               \
 265         return nullb_device_##TYPE##_attr_show(                                 \
 266                                 to_nullb_device(item)->NAME, page);             \
 267 }                                                                               \
 268 static ssize_t                                                                  \
 269 nullb_device_##NAME##_store(struct config_item *item, const char *page,         \
 270                             size_t count)                                       \
 271 {                                                                               \
 272         if (test_bit(NULLB_DEV_FL_CONFIGURED, &to_nullb_device(item)->flags))   \
 273                 return -EBUSY;                                                  \
 274         return nullb_device_##TYPE##_attr_store(                                \
 275                         &to_nullb_device(item)->NAME, page, count);             \
 276 }                                                                               \
 277 CONFIGFS_ATTR(nullb_device_, NAME);
 278
 279 NULLB_DEVICE_ATTR(size, ulong);
 280 NULLB_DEVICE_ATTR(completion_nsec, ulong);
 281 NULLB_DEVICE_ATTR(submit_queues, uint);
 282 NULLB_DEVICE_ATTR(home_node, uint);
 283 NULLB_DEVICE_ATTR(queue_mode, uint);
 284 NULLB_DEVICE_ATTR(blocksize, uint);
 285 NULLB_DEVICE_ATTR(irqmode, uint);
 286 NULLB_DEVICE_ATTR(hw_queue_depth, uint);
 287 NULLB_DEVICE_ATTR(index, uint);
 288 NULLB_DEVICE_ATTR(blocking, bool);
 289 NULLB_DEVICE_ATTR(use_per_node_hctx, bool);
 290 NULLB_DEVICE_ATTR(memory_backed, bool);
 291 NULLB_DEVICE_ATTR(discard, bool);
 292 NULLB_DEVICE_ATTR(mbps, uint);
 293 NULLB_DEVICE_ATTR(cache_size, ulong);
 294 NULLB_DEVICE_ATTR(zoned, bool);
 295 NULLB_DEVICE_ATTR(zone_size, ulong);
 296
 297 static ssize_t nullb_device_power_show(struct config_item *item, char *page)
 298 {
 299         return nullb_device_bool_attr_show(to_nullb_device(item)->power, page);
 300 }
 301
 302 static ssize_t nullb_device_power_store(struct config_item *item,
 303                                      const char *page, size_t count)
 304 {
 305         struct nullb_device *dev = to_nullb_device(item);
 306         bool newp = false;
 307         ssize_t ret;
 308
 309         ret = nullb_device_bool_attr_store(&newp, page, count);
 310         if (ret < 0)
 311                 return ret;
 312
 313         if (!dev->power && newp) {
 314                 if (test_and_set_bit(NULLB_DEV_FL_UP, &dev->flags))
 315                         return count;
 316                 if (null_add_dev(dev)) {
 317                         clear_bit(NULLB_DEV_FL_UP, &dev->flags);
 318                         return -ENOMEM;
 319                 }
 320
 321                 set_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
 322                 dev->power = newp;
 323         } else if (dev->power && !newp) {
 324                 mutex_lock(&lock);
 325                 dev->power = newp;
 326                 null_del_dev(dev->nullb);
 327                 mutex_unlock(&lock);
 328                 clear_bit(NULLB_DEV_FL_UP, &dev->flags);
 329                 clear_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
 330         }
 331
 332         return count;
 333 }
 334
 335 CONFIGFS_ATTR(nullb_device_, power);
 336
 337 static ssize_t nullb_device_badblocks_show(struct config_item *item, char *page)
 338 {
 339         struct nullb_device *t_dev = to_nullb_device(item);
 340
 341         return badblocks_show(&t_dev->badblocks, page, 0);
 342 }
 343
 344 static ssize_t nullb_device_badblocks_store(struct config_item *item,
 345                                      const char *page, size_t count)
 346 {
 347         struct nullb_device *t_dev = to_nullb_device(item);
 348         char *orig, *buf, *tmp;
 349         u64 start, end;
 350         int ret;
 351
 352         orig = kstrndup(page, count, GFP_KERNEL);
 353         if (!orig)
 354                 return -ENOMEM;
 355
 356         buf = strstrip(orig);
 357
 358         ret = -EINVAL;
 359         if (buf[0] != '+' && buf[0] != '-')
 360                 goto out;
 361         tmp = strchr(&buf[1], '-');
 362         if (!tmp)
 363                 goto out;
 364         *tmp = '\0';
 365         ret = kstrtoull(buf + 1, 0, &start);
 366         if (ret)
 367                 goto out;
 368         ret = kstrtoull(tmp + 1, 0, &end);
 369         if (ret)
 370                 goto out;
 371         ret = -EINVAL;
 372         if (start > end)
 373                 goto out;
 374         /* enable badblocks */
 375         cmpxchg(&t_dev->badblocks.shift, -1, 0);
 376         if (buf[0] == '+')
 377                 ret = badblocks_set(&t_dev->badblocks, start,
 378                         end - start + 1, 1);
 379         else
 380                 ret = badblocks_clear(&t_dev->badblocks, start,
 381                         end - start + 1);
 382         if (ret == 0)
 383                 ret = count;
 384 out:
 385         kfree(orig);
 386         return ret;
 387 }
 388 CONFIGFS_ATTR(nullb_device_, badblocks);
 389
 390 static struct configfs_attribute *nullb_device_attrs[] = {
 391         &nullb_device_attr_size,
 392         &nullb_device_attr_completion_nsec,
 393         &nullb_device_attr_submit_queues,
 394         &nullb_device_attr_home_node,
 395         &nullb_device_attr_queue_mode,
 396         &nullb_device_attr_blocksize,
 397         &nullb_device_attr_irqmode,
 398         &nullb_device_attr_hw_queue_depth,
 399         &nullb_device_attr_index,
 400         &nullb_device_attr_blocking,
 401         &nullb_device_attr_use_per_node_hctx,
 402         &nullb_device_attr_power,
 403         &nullb_device_attr_memory_backed,
 404         &nullb_device_attr_discard,
 405         &nullb_device_attr_mbps,
 406         &nullb_device_attr_cache_size,
 407         &nullb_device_attr_badblocks,
 408         &nullb_device_attr_zoned,
 409         &nullb_device_attr_zone_size,
 410         NULL,
 411 };
 412
 413 static void nullb_device_release(struct config_item *item)
 414 {
 415         struct nullb_device *dev = to_nullb_device(item);
 416
 417         null_free_device_storage(dev, false);
 418         null_free_dev(dev);
 419 }
 420
 421 static struct configfs_item_operations nullb_device_ops = {
 422         .release        = nullb_device_release,
 423 };
 424
 425 static const struct config_item_type nullb_device_type = {
 426         .ct_item_ops    = &nullb_device_ops,
 427         .ct_attrs       = nullb_device_attrs,
 428         .ct_owner       = THIS_MODULE,
 429 };
 430
 431 static struct
 432 config_item *nullb_group_make_item(struct config_group *group, const char *name)
 433 {
 434         struct nullb_device *dev;
 435
 436         dev = null_alloc_dev();
 437         if (!dev)
 438                 return ERR_PTR(-ENOMEM);
 439
 440         config_item_init_type_name(&dev->item, name, &nullb_device_type);
 441
 442         return &dev->item;
 443 }
 444
 445 static void
 446 nullb_group_drop_item(struct config_group *group, struct config_item *item)
 447 {
 448         struct nullb_device *dev = to_nullb_device(item);
 449
 450         if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
 451                 mutex_lock(&lock);
 452                 dev->power = false;
 453                 null_del_dev(dev->nullb);
 454                 mutex_unlock(&lock);
 455         }
 456
 457         config_item_put(item);
 458 }
 459
 460 static ssize_t memb_group_features_show(struct config_item *item, char *page)
 461 {
 462         return snprintf(page, PAGE_SIZE, "memory_backed,discard,bandwidth,cache,badblocks,zoned,zone_size\n");
 463 }
 464
 465 CONFIGFS_ATTR_RO(memb_group_, features);
 466
 467 static struct configfs_attribute *nullb_group_attrs[] = {
 468         &memb_group_attr_features,
 469         NULL,
 470 };
 471
 472 static struct configfs_group_operations nullb_group_ops = {
 473         .make_item      = nullb_group_make_item,
 474         .drop_item      = nullb_group_drop_item,
 475 };
 476
 477 static const struct config_item_type nullb_group_type = {
 478         .ct_group_ops   = &nullb_group_ops,
 479         .ct_attrs       = nullb_group_attrs,
 480         .ct_owner       = THIS_MODULE,
 481 };
 482
 483 static struct configfs_subsystem nullb_subsys = {
 484         .su_group = {
 485                 .cg_item = {
 486                         .ci_namebuf = "nullb",
 487                         .ci_type = &nullb_group_type,
 488                 },
 489         },
 490 };
 491
 492 static inline int null_cache_active(struct nullb *nullb)
 493 {
 494         return test_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
 495 }
 496
 497 static struct nullb_device *null_alloc_dev(void)
 498 {
 499         struct nullb_device *dev;
 500
 501         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
 502         if (!dev)
 503                 return NULL;
 504         INIT_RADIX_TREE(&dev->data, GFP_ATOMIC);
 505         INIT_RADIX_TREE(&dev->cache, GFP_ATOMIC);
 506         if (badblocks_init(&dev->badblocks, 0)) {
 507                 kfree(dev);
 508                 return NULL;
 509         }
 510
 511         dev->size = g_gb * 1024;
 512         dev->completion_nsec = g_completion_nsec;
 513         dev->submit_queues = g_submit_queues;
 514         dev->home_node = g_home_node;
 515         dev->queue_mode = g_queue_mode;
 516         dev->blocksize = g_bs;
 517         dev->irqmode = g_irqmode;
 518         dev->hw_queue_depth = g_hw_queue_depth;
 519         dev->blocking = g_blocking;
 520         dev->use_per_node_hctx = g_use_per_node_hctx;
 521         dev->zoned = g_zoned;
 522         dev->zone_size = g_zone_size;
 523         return dev;
 524 }
 525
 526 static void null_free_dev(struct nullb_device *dev)
 527 {
 528         if (!dev)
 529                 return;
 530
 531         null_zone_exit(dev);
 532         badblocks_exit(&dev->badblocks);
 533         kfree(dev);
 534 }
 535
 536 static void put_tag(struct nullb_queue *nq, unsigned int tag)
 537 {
 538         clear_bit_unlock(tag, nq->tag_map);
 539
 540         if (waitqueue_active(&nq->wait))
 541                 wake_up(&nq->wait);
 542 }
 543
 544 static unsigned int get_tag(struct nullb_queue *nq)
 545 {
 546         unsigned int tag;
 547
 548         do {
 549                 tag = find_first_zero_bit(nq->tag_map, nq->queue_depth);
 550                 if (tag >= nq->queue_depth)
 551                         return -1U;
 552         } while (test_and_set_bit_lock(tag, nq->tag_map));
 553
 554         return tag;
 555 }
 556
 557 static void free_cmd(struct nullb_cmd *cmd)
 558 {
 559         put_tag(cmd->nq, cmd->tag);
 560 }
 561
 562 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer);
 563
 564 static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq)
 565 {
 566         struct nullb_cmd *cmd;
 567         unsigned int tag;
 568
 569         tag = get_tag(nq);
 570         if (tag != -1U) {
 571                 cmd = &nq->cmds[tag];
 572                 cmd->tag = tag;
 573                 cmd->nq = nq;
 574                 if (nq->dev->irqmode == NULL_IRQ_TIMER) {
 575                         hrtimer_init(&cmd->timer, CLOCK_MONOTONIC,
 576                                      HRTIMER_MODE_REL);
 577                         cmd->timer.function = null_cmd_timer_expired;
 578                 }
 579                 return cmd;
 580         }
 581
 582         return NULL;
 583 }
 584
 585 static struct nullb_cmd *alloc_cmd(struct nullb_queue *nq, int can_wait)
 586 {
 587         struct nullb_cmd *cmd;
 588         DEFINE_WAIT(wait);
 589
 590         cmd = __alloc_cmd(nq);
 591         if (cmd || !can_wait)
 592                 return cmd;
 593
 594         do {
 595                 prepare_to_wait(&nq->wait, &wait, TASK_UNINTERRUPTIBLE);
 596                 cmd = __alloc_cmd(nq);
 597                 if (cmd)
 598                         break;
 599
 600                 io_schedule();
 601         } while (1);
 602
 603         finish_wait(&nq->wait, &wait);
 604         return cmd;
 605 }
 606
 607 static void end_cmd(struct nullb_cmd *cmd)
 608 {
 609         int queue_mode = cmd->nq->dev->queue_mode;
 610
 611         switch (queue_mode)  {
 612         case NULL_Q_MQ:
 613                 blk_mq_end_request(cmd->rq, cmd->error);
 614                 return;
 615         case NULL_Q_BIO:
 616                 cmd->bio->bi_status = cmd->error;
 617                 bio_endio(cmd->bio);
 618                 break;
 619         }
 620
 621         free_cmd(cmd);
 622 }
 623
 624 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
 625 {
 626         end_cmd(container_of(timer, struct nullb_cmd, timer));
 627
 628         return HRTIMER_NORESTART;
 629 }
 630
 631 static void null_cmd_end_timer(struct nullb_cmd *cmd)
 632 {
 633         ktime_t kt = cmd->nq->dev->completion_nsec;
 634
 635         hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL);
 636 }
 637
 638 static void null_softirq_done_fn(struct request *rq)
 639 {
 640         struct nullb *nullb = rq->q->queuedata;
 641
 642         if (nullb->dev->queue_mode == NULL_Q_MQ)
 643                 end_cmd(blk_mq_rq_to_pdu(rq));
 644         else
 645                 end_cmd(rq->special);
 646 }
 647
 648 static struct nullb_page *null_alloc_page(gfp_t gfp_flags)
 649 {
 650         struct nullb_page *t_page;
 651
 652         t_page = kmalloc(sizeof(struct nullb_page), gfp_flags);
 653         if (!t_page)
 654                 goto out;
 655
 656         t_page->page = alloc_pages(gfp_flags, 0);
 657         if (!t_page->page)
 658                 goto out_freepage;
 659
 660         memset(t_page->bitmap, 0, sizeof(t_page->bitmap));
 661         return t_page;
 662 out_freepage:
 663         kfree(t_page);
 664 out:
 665         return NULL;
 666 }
 667
 668 static void null_free_page(struct nullb_page *t_page)
 669 {
 670         __set_bit(NULLB_PAGE_FREE, t_page->bitmap);
 671         if (test_bit(NULLB_PAGE_LOCK, t_page->bitmap))
 672                 return;
 673         __free_page(t_page->page);
 674         kfree(t_page);
 675 }
 676
 677 static bool null_page_empty(struct nullb_page *page)
 678 {
 679         int size = MAP_SZ - 2;
 680
 681         return find_first_bit(page->bitmap, size) == size;
 682 }
 683
 684 static void null_free_sector(struct nullb *nullb, sector_t sector,
 685         bool is_cache)
 686 {
 687         unsigned int sector_bit;
 688         u64 idx;
 689         struct nullb_page *t_page, *ret;
 690         struct radix_tree_root *root;
 691
 692         root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
 693         idx = sector >> PAGE_SECTORS_SHIFT;
 694         sector_bit = (sector & SECTOR_MASK);
 695
 696         t_page = radix_tree_lookup(root, idx);
 697         if (t_page) {
 698                 __clear_bit(sector_bit, t_page->bitmap);
 699
 700                 if (null_page_empty(t_page)) {
 701                         ret = radix_tree_delete_item(root, idx, t_page);
 702                         WARN_ON(ret != t_page);
 703                         null_free_page(ret);
 704                         if (is_cache)
 705                                 nullb->dev->curr_cache -= PAGE_SIZE;
 706                 }
 707         }
 708 }
 709
 710 static struct nullb_page *null_radix_tree_insert(struct nullb *nullb, u64 idx,
 711         struct nullb_page *t_page, bool is_cache)
 712 {
 713         struct radix_tree_root *root;
 714
 715         root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
 716
 717         if (radix_tree_insert(root, idx, t_page)) {
 718                 null_free_page(t_page);
 719                 t_page = radix_tree_lookup(root, idx);
 720                 WARN_ON(!t_page || t_page->page->index != idx);
 721         } else if (is_cache)
 722                 nullb->dev->curr_cache += PAGE_SIZE;
 723
 724         return t_page;
 725 }
 726
 727 static void null_free_device_storage(struct nullb_device *dev, bool is_cache)
 728 {
 729         unsigned long pos = 0;
 730         int nr_pages;
 731         struct nullb_page *ret, *t_pages[FREE_BATCH];
 732         struct radix_tree_root *root;
 733
 734         root = is_cache ? &dev->cache : &dev->data;
 735
 736         do {
 737                 int i;
 738
 739                 nr_pages = radix_tree_gang_lookup(root,
 740                                 (void **)t_pages, pos, FREE_BATCH);
 741
 742                 for (i = 0; i < nr_pages; i++) {
 743                         pos = t_pages[i]->page->index;
 744                         ret = radix_tree_delete_item(root, pos, t_pages[i]);
 745                         WARN_ON(ret != t_pages[i]);
 746                         null_free_page(ret);
 747                 }
 748
 749                 pos++;
 750         } while (nr_pages == FREE_BATCH);
 751
 752         if (is_cache)
 753                 dev->curr_cache = 0;
 754 }
 755
 756 static struct nullb_page *__null_lookup_page(struct nullb *nullb,
 757         sector_t sector, bool for_write, bool is_cache)
 758 {
 759         unsigned int sector_bit;
 760         u64 idx;
 761         struct nullb_page *t_page;
 762         struct radix_tree_root *root;
 763
 764         idx = sector >> PAGE_SECTORS_SHIFT;
 765         sector_bit = (sector & SECTOR_MASK);
 766
 767         root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
 768         t_page = radix_tree_lookup(root, idx);
 769         WARN_ON(t_page && t_page->page->index != idx);
 770
 771         if (t_page && (for_write || test_bit(sector_bit, t_page->bitmap)))
 772                 return t_page;
 773
 774         return NULL;
 775 }
 776
 777 static struct nullb_page *null_lookup_page(struct nullb *nullb,
 778         sector_t sector, bool for_write, bool ignore_cache)
 779 {
 780         struct nullb_page *page = NULL;
 781
 782         if (!ignore_cache)
 783                 page = __null_lookup_page(nullb, sector, for_write, true);
 784         if (page)
 785                 return page;
 786         return __null_lookup_page(nullb, sector, for_write, false);
 787 }
 788
 789 static struct nullb_page *null_insert_page(struct nullb *nullb,
 790                                            sector_t sector, bool ignore_cache)
 791         __releases(&nullb->lock)
 792         __acquires(&nullb->lock)
 793 {
 794         u64 idx;
 795         struct nullb_page *t_page;
 796
 797         t_page = null_lookup_page(nullb, sector, true, ignore_cache);
 798         if (t_page)
 799                 return t_page;
 800
 801         spin_unlock_irq(&nullb->lock);
 802
 803         t_page = null_alloc_page(GFP_NOIO);
 804         if (!t_page)
 805                 goto out_lock;
 806
 807         if (radix_tree_preload(GFP_NOIO))
 808                 goto out_freepage;
 809
 810         spin_lock_irq(&nullb->lock);
 811         idx = sector >> PAGE_SECTORS_SHIFT;
 812         t_page->page->index = idx;
 813         t_page = null_radix_tree_insert(nullb, idx, t_page, !ignore_cache);
 814         radix_tree_preload_end();
 815
 816         return t_page;
 817 out_freepage:
 818         null_free_page(t_page);
 819 out_lock:
 820         spin_lock_irq(&nullb->lock);
 821         return null_lookup_page(nullb, sector, true, ignore_cache);
 822 }
 823
 824 static int null_flush_cache_page(struct nullb *nullb, struct nullb_page *c_page)
 825 {
 826         int i;
 827         unsigned int offset;
 828         u64 idx;
 829         struct nullb_page *t_page, *ret;
 830         void *dst, *src;
 831
 832         idx = c_page->page->index;
 833
 834         t_page = null_insert_page(nullb, idx << PAGE_SECTORS_SHIFT, true);
 835
 836         __clear_bit(NULLB_PAGE_LOCK, c_page->bitmap);
 837         if (test_bit(NULLB_PAGE_FREE, c_page->bitmap)) {
 838                 null_free_page(c_page);
 839                 if (t_page && null_page_empty(t_page)) {
 840                         ret = radix_tree_delete_item(&nullb->dev->data,
 841                                 idx, t_page);
 842                         null_free_page(t_page);
 843                 }
 844                 return 0;
 845         }
 846
 847         if (!t_page)
 848                 return -ENOMEM;
 849
 850         src = kmap_atomic(c_page->page);
 851         dst = kmap_atomic(t_page->page);
 852
 853         for (i = 0; i < PAGE_SECTORS;
 854                         i += (nullb->dev->blocksize >> SECTOR_SHIFT)) {
 855                 if (test_bit(i, c_page->bitmap)) {
 856                         offset = (i << SECTOR_SHIFT);
 857                         memcpy(dst + offset, src + offset,
 858                                 nullb->dev->blocksize);
 859                         __set_bit(i, t_page->bitmap);
 860                 }
 861         }
 862
 863         kunmap_atomic(dst);
 864         kunmap_atomic(src);
 865
 866         ret = radix_tree_delete_item(&nullb->dev->cache, idx, c_page);
 867         null_free_page(ret);
 868         nullb->dev->curr_cache -= PAGE_SIZE;
 869
 870         return 0;
 871 }
 872
 873 static int null_make_cache_space(struct nullb *nullb, unsigned long n)
 874 {
 875         int i, err, nr_pages;
 876         struct nullb_page *c_pages[FREE_BATCH];
 877         unsigned long flushed = 0, one_round;
 878
 879 again:
 880         if ((nullb->dev->cache_size * 1024 * 1024) >
 881              nullb->dev->curr_cache + n || nullb->dev->curr_cache == 0)
 882                 return 0;
 883
 884         nr_pages = radix_tree_gang_lookup(&nullb->dev->cache,
 885                         (void **)c_pages, nullb->cache_flush_pos, FREE_BATCH);
 886         /*
 887          * nullb_flush_cache_page could unlock before using the c_pages. To
 888          * avoid race, we don't allow page free
 889          */
 890         for (i = 0; i < nr_pages; i++) {
 891                 nullb->cache_flush_pos = c_pages[i]->page->index;
 892                 /*
 893                  * We found the page which is being flushed to disk by other
 894                  * threads
 895                  */
 896                 if (test_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap))
 897                         c_pages[i] = NULL;
 898                 else
 899                         __set_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap);
 900         }
 901
 902         one_round = 0;
 903         for (i = 0; i < nr_pages; i++) {
 904                 if (c_pages[i] == NULL)
 905                         continue;
 906                 err = null_flush_cache_page(nullb, c_pages[i]);
 907                 if (err)
 908                         return err;
 909                 one_round++;
 910         }
 911         flushed += one_round << PAGE_SHIFT;
 912
 913         if (n > flushed) {
 914                 if (nr_pages == 0)
 915                         nullb->cache_flush_pos = 0;
 916                 if (one_round == 0) {
 917                         /* give other threads a chance */
 918                         spin_unlock_irq(&nullb->lock);
 919                         spin_lock_irq(&nullb->lock);
 920                 }
 921                 goto again;
 922         }
 923         return 0;
 924 }
 925
 926 static int copy_to_nullb(struct nullb *nullb, struct page *source,
 927         unsigned int off, sector_t sector, size_t n, bool is_fua)
 928 {
 929         size_t temp, count = 0;
 930         unsigned int offset;
 931         struct nullb_page *t_page;
 932         void *dst, *src;
 933
 934         while (count < n) {
 935                 temp = min_t(size_t, nullb->dev->blocksize, n - count);
 936
 937                 if (null_cache_active(nullb) && !is_fua)
 938                         null_make_cache_space(nullb, PAGE_SIZE);
 939
 940                 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
 941                 t_page = null_insert_page(nullb, sector,
 942                         !null_cache_active(nullb) || is_fua);
 943                 if (!t_page)
 944                         return -ENOSPC;
 945
 946                 src = kmap_atomic(source);
 947                 dst = kmap_atomic(t_page->page);
 948                 memcpy(dst + offset, src + off + count, temp);
 949                 kunmap_atomic(dst);
 950                 kunmap_atomic(src);
 951
 952                 __set_bit(sector & SECTOR_MASK, t_page->bitmap);
 953
 954                 if (is_fua)
 955                         null_free_sector(nullb, sector, true);
 956
 957                 count += temp;
 958                 sector += temp >> SECTOR_SHIFT;
 959         }
 960         return 0;
 961 }
 962
 963 static int copy_from_nullb(struct nullb *nullb, struct page *dest,
 964         unsigned int off, sector_t sector, size_t n)
 965 {
 966         size_t temp, count = 0;
 967         unsigned int offset;
 968         struct nullb_page *t_page;
 969         void *dst, *src;
 970
 971         while (count < n) {
 972                 temp = min_t(size_t, nullb->dev->blocksize, n - count);
 973
 974                 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
 975                 t_page = null_lookup_page(nullb, sector, false,
 976                         !null_cache_active(nullb));
 977
 978                 dst = kmap_atomic(dest);
 979                 if (!t_page) {
 980                         memset(dst + off + count, 0, temp);
 981                         goto next;
 982                 }
 983                 src = kmap_atomic(t_page->page);
 984                 memcpy(dst + off + count, src + offset, temp);
 985                 kunmap_atomic(src);
 986 next:
 987                 kunmap_atomic(dst);
 988
 989                 count += temp;
 990                 sector += temp >> SECTOR_SHIFT;
 991         }
 992         return 0;
 993 }
 994
 995 static void null_handle_discard(struct nullb *nullb, sector_t sector, size_t n)
 996 {
 997         size_t temp;
 998
 999         spin_lock_irq(&nullb->lock);
1000         while (n > 0) {
1001                 temp = min_t(size_t, n, nullb->dev->blocksize);
1002                 null_free_sector(nullb, sector, false);
1003                 if (null_cache_active(nullb))
1004                         null_free_sector(nullb, sector, true);
1005                 sector += temp >> SECTOR_SHIFT;
1006                 n -= temp;
1007         }
1008         spin_unlock_irq(&nullb->lock);
1009 }
1010
1011 static int null_handle_flush(struct nullb *nullb)
1012 {
1013         int err;
1014
1015         if (!null_cache_active(nullb))
1016                 return 0;
1017
1018         spin_lock_irq(&nullb->lock);
1019         while (true) {
1020                 err = null_make_cache_space(nullb,
1021                         nullb->dev->cache_size * 1024 * 1024);
1022                 if (err || nullb->dev->curr_cache == 0)
1023                         break;
1024         }
1025
1026         WARN_ON(!radix_tree_empty(&nullb->dev->cache));
1027         spin_unlock_irq(&nullb->lock);
1028         return err;
1029 }
1030
1031 static int null_transfer(struct nullb *nullb, struct page *page,
1032         unsigned int len, unsigned int off, bool is_write, sector_t sector,
1033         bool is_fua)
1034 {
1035         int err = 0;
1036
1037         if (!is_write) {
1038                 err = copy_from_nullb(nullb, page, off, sector, len);
1039                 flush_dcache_page(page);
1040         } else {
1041                 flush_dcache_page(page);
1042                 err = copy_to_nullb(nullb, page, off, sector, len, is_fua);
1043         }
1044
1045         return err;
1046 }
1047
1048 static int null_handle_rq(struct nullb_cmd *cmd)
1049 {
1050         struct request *rq = cmd->rq;
1051         struct nullb *nullb = cmd->nq->dev->nullb;
1052         int err;
1053         unsigned int len;
1054         sector_t sector;
1055         struct req_iterator iter;
1056         struct bio_vec bvec;
1057
1058         sector = blk_rq_pos(rq);
1059
1060         if (req_op(rq) == REQ_OP_DISCARD) {
1061                 null_handle_discard(nullb, sector, blk_rq_bytes(rq));
1062                 return 0;
1063         }
1064
1065         spin_lock_irq(&nullb->lock);
1066         rq_for_each_segment(bvec, rq, iter) {
1067                 len = bvec.bv_len;
1068                 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1069                                      op_is_write(req_op(rq)), sector,
1070                                      req_op(rq) & REQ_FUA);
1071                 if (err) {
1072                         spin_unlock_irq(&nullb->lock);
1073                         return err;
1074                 }
1075                 sector += len >> SECTOR_SHIFT;
1076         }
1077         spin_unlock_irq(&nullb->lock);
1078
1079         return 0;
1080 }
1081
1082 static int null_handle_bio(struct nullb_cmd *cmd)
1083 {
1084         struct bio *bio = cmd->bio;
1085         struct nullb *nullb = cmd->nq->dev->nullb;
1086         int err;
1087         unsigned int len;
1088         sector_t sector;
1089         struct bio_vec bvec;
1090         struct bvec_iter iter;
1091
1092         sector = bio->bi_iter.bi_sector;
1093
1094         if (bio_op(bio) == REQ_OP_DISCARD) {
1095                 null_handle_discard(nullb, sector,
1096                         bio_sectors(bio) << SECTOR_SHIFT);
1097                 return 0;
1098         }
1099
1100         spin_lock_irq(&nullb->lock);
1101         bio_for_each_segment(bvec, bio, iter) {
1102                 len = bvec.bv_len;
1103                 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1104                                      op_is_write(bio_op(bio)), sector,
1105                                      bio_op(bio) & REQ_FUA);
1106                 if (err) {
1107                         spin_unlock_irq(&nullb->lock);
1108                         return err;
1109                 }
1110                 sector += len >> SECTOR_SHIFT;
1111         }
1112         spin_unlock_irq(&nullb->lock);
1113         return 0;
1114 }
1115
1116 static void null_stop_queue(struct nullb *nullb)
1117 {
1118         struct request_queue *q = nullb->q;
1119
1120         if (nullb->dev->queue_mode == NULL_Q_MQ)
1121                 blk_mq_stop_hw_queues(q);
1122 }
1123
1124 static void null_restart_queue_async(struct nullb *nullb)
1125 {
1126         struct request_queue *q = nullb->q;
1127
1128         if (nullb->dev->queue_mode == NULL_Q_MQ)
1129                 blk_mq_start_stopped_hw_queues(q, true);
1130 }
1131
1132 static blk_status_t null_handle_cmd(struct nullb_cmd *cmd)
1133 {
1134         struct nullb_device *dev = cmd->nq->dev;
1135         struct nullb *nullb = dev->nullb;
1136         int err = 0;
1137
1138         if (test_bit(NULLB_DEV_FL_THROTTLED, &dev->flags)) {
1139                 struct request *rq = cmd->rq;
1140
1141                 if (!hrtimer_active(&nullb->bw_timer))
1142                         hrtimer_restart(&nullb->bw_timer);
1143
1144                 if (atomic_long_sub_return(blk_rq_bytes(rq),
1145                                 &nullb->cur_bytes) < 0) {
1146                         null_stop_queue(nullb);
1147                         /* race with timer */
1148                         if (atomic_long_read(&nullb->cur_bytes) > 0)
1149                                 null_restart_queue_async(nullb);
1150                         /* requeue request */
1151                         return BLK_STS_DEV_RESOURCE;
1152                 }
1153         }
1154
1155         if (nullb->dev->badblocks.shift != -1) {
1156                 int bad_sectors;
1157                 sector_t sector, size, first_bad;
1158                 bool is_flush = true;
1159
1160                 if (dev->queue_mode == NULL_Q_BIO &&
1161                                 bio_op(cmd->bio) != REQ_OP_FLUSH) {
1162                         is_flush = false;
1163                         sector = cmd->bio->bi_iter.bi_sector;
1164                         size = bio_sectors(cmd->bio);
1165                 }
1166                 if (dev->queue_mode != NULL_Q_BIO &&
1167                                 req_op(cmd->rq) != REQ_OP_FLUSH) {
1168                         is_flush = false;
1169                         sector = blk_rq_pos(cmd->rq);
1170                         size = blk_rq_sectors(cmd->rq);
1171                 }
1172                 if (!is_flush && badblocks_check(&nullb->dev->badblocks, sector,
1173                                 size, &first_bad, &bad_sectors)) {
1174                         cmd->error = BLK_STS_IOERR;
1175                         goto out;
1176                 }
1177         }
1178
1179         if (dev->memory_backed) {
1180                 if (dev->queue_mode == NULL_Q_BIO) {
1181                         if (bio_op(cmd->bio) == REQ_OP_FLUSH)
1182                                 err = null_handle_flush(nullb);
1183                         else
1184                                 err = null_handle_bio(cmd);
1185                 } else {
1186                         if (req_op(cmd->rq) == REQ_OP_FLUSH)
1187                                 err = null_handle_flush(nullb);
1188                         else
1189                                 err = null_handle_rq(cmd);
1190                 }
1191         }
1192         cmd->error = errno_to_blk_status(err);
1193
1194         if (!cmd->error && dev->zoned) {
1195                 sector_t sector;
1196                 unsigned int nr_sectors;
1197                 int op;
1198
1199                 if (dev->queue_mode == NULL_Q_BIO) {
1200                         op = bio_op(cmd->bio);
1201                         sector = cmd->bio->bi_iter.bi_sector;
1202                         nr_sectors = cmd->bio->bi_iter.bi_size >> 9;
1203                 } else {
1204                         op = req_op(cmd->rq);
1205                         sector = blk_rq_pos(cmd->rq);
1206                         nr_sectors = blk_rq_sectors(cmd->rq);
1207                 }
1208
1209                 if (op == REQ_OP_WRITE)
1210                         null_zone_write(cmd, sector, nr_sectors);
1211                 else if (op == REQ_OP_ZONE_RESET)
1212                         null_zone_reset(cmd, sector);
1213         }
1214 out:
1215         /* Complete IO by inline, softirq or timer */
1216         switch (dev->irqmode) {
1217         case NULL_IRQ_SOFTIRQ:
1218                 switch (dev->queue_mode)  {
1219                 case NULL_Q_MQ:
1220                         blk_mq_complete_request(cmd->rq);
1221                         break;
1222                 case NULL_Q_BIO:
1223                         /*
1224                          * XXX: no proper submitting cpu information available.
1225                          */
1226                         end_cmd(cmd);
1227                         break;
1228                 }
1229                 break;
1230         case NULL_IRQ_NONE:
1231                 end_cmd(cmd);
1232                 break;
1233         case NULL_IRQ_TIMER:
1234                 null_cmd_end_timer(cmd);
1235                 break;
1236         }
1237         return BLK_STS_OK;
1238 }
1239
1240 static enum hrtimer_restart nullb_bwtimer_fn(struct hrtimer *timer)
1241 {
1242         struct nullb *nullb = container_of(timer, struct nullb, bw_timer);
1243         ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1244         unsigned int mbps = nullb->dev->mbps;
1245
1246         if (atomic_long_read(&nullb->cur_bytes) == mb_per_tick(mbps))
1247                 return HRTIMER_NORESTART;
1248
1249         atomic_long_set(&nullb->cur_bytes, mb_per_tick(mbps));
1250         null_restart_queue_async(nullb);
1251
1252         hrtimer_forward_now(&nullb->bw_timer, timer_interval);
1253
1254         return HRTIMER_RESTART;
1255 }
1256
1257 static void nullb_setup_bwtimer(struct nullb *nullb)
1258 {
1259         ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1260
1261         hrtimer_init(&nullb->bw_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1262         nullb->bw_timer.function = nullb_bwtimer_fn;
1263         atomic_long_set(&nullb->cur_bytes, mb_per_tick(nullb->dev->mbps));
1264         hrtimer_start(&nullb->bw_timer, timer_interval, HRTIMER_MODE_REL);
1265 }
1266
1267 static struct nullb_queue *nullb_to_queue(struct nullb *nullb)
1268 {
1269         int index = 0;
1270
1271         if (nullb->nr_queues != 1)
1272                 index = raw_smp_processor_id() / ((nr_cpu_ids + nullb->nr_queues - 1) / nullb->nr_queues);
1273
1274         return &nullb->queues[index];
1275 }
1276
1277 static blk_qc_t null_queue_bio(struct request_queue *q, struct bio *bio)
1278 {
1279         struct nullb *nullb = q->queuedata;
1280         struct nullb_queue *nq = nullb_to_queue(nullb);
1281         struct nullb_cmd *cmd;
1282
1283         cmd = alloc_cmd(nq, 1);
1284         cmd->bio = bio;
1285
1286         null_handle_cmd(cmd);
1287         return BLK_QC_T_NONE;
1288 }
1289
1290 static bool should_timeout_request(struct request *rq)
1291 {
1292 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1293         if (g_timeout_str[0])
1294                 return should_fail(&null_timeout_attr, 1);
1295 #endif
1296         return false;
1297 }
1298
1299 static bool should_requeue_request(struct request *rq)
1300 {
1301 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1302         if (g_requeue_str[0])
1303                 return should_fail(&null_requeue_attr, 1);
1304 #endif
1305         return false;
1306 }
1307
1308 static enum blk_eh_timer_return null_timeout_rq(struct request *rq, bool res)
1309 {
1310         pr_info("null: rq %p timed out\n", rq);
1311         blk_mq_complete_request(rq);
1312         return BLK_EH_DONE;
1313 }
1314
1315 static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx,
1316                          const struct blk_mq_queue_data *bd)
1317 {
1318         struct nullb_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
1319         struct nullb_queue *nq = hctx->driver_data;
1320
1321         might_sleep_if(hctx->flags & BLK_MQ_F_BLOCKING);
1322
1323         if (nq->dev->irqmode == NULL_IRQ_TIMER) {
1324                 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1325                 cmd->timer.function = null_cmd_timer_expired;
1326         }
1327         cmd->rq = bd->rq;
1328         cmd->nq = nq;
1329
1330         blk_mq_start_request(bd->rq);
1331
1332         if (should_requeue_request(bd->rq)) {
1333                 /*
1334                  * Alternate between hitting the core BUSY path, and the
1335                  * driver driven requeue path
1336                  */
1337                 nq->requeue_selection++;
1338                 if (nq->requeue_selection & 1)
1339                         return BLK_STS_RESOURCE;
1340                 else {
1341                         blk_mq_requeue_request(bd->rq, true);
1342                         return BLK_STS_OK;
1343                 }
1344         }
1345         if (should_timeout_request(bd->rq))
1346                 return BLK_STS_OK;
1347
1348         return null_handle_cmd(cmd);
1349 }
1350
1351 static const struct blk_mq_ops null_mq_ops = {
1352         .queue_rq       = null_queue_rq,
1353         .complete       = null_softirq_done_fn,
1354         .timeout        = null_timeout_rq,
1355 };
1356
1357 static void cleanup_queue(struct nullb_queue *nq)
1358 {
1359         kfree(nq->tag_map);
1360         kfree(nq->cmds);
1361 }
1362
1363 static void cleanup_queues(struct nullb *nullb)
1364 {
1365         int i;
1366
1367         for (i = 0; i < nullb->nr_queues; i++)
1368                 cleanup_queue(&nullb->queues[i]);
1369
1370         kfree(nullb->queues);
1371 }
1372
1373 static void null_del_dev(struct nullb *nullb)
1374 {
1375         struct nullb_device *dev = nullb->dev;
1376
1377         ida_simple_remove(&nullb_indexes, nullb->index);
1378
1379         list_del_init(&nullb->list);
1380
1381         del_gendisk(nullb->disk);
1382
1383         if (test_bit(NULLB_DEV_FL_THROTTLED, &nullb->dev->flags)) {
1384                 hrtimer_cancel(&nullb->bw_timer);
1385                 atomic_long_set(&nullb->cur_bytes, LONG_MAX);
1386                 null_restart_queue_async(nullb);
1387         }
1388
1389         blk_cleanup_queue(nullb->q);
1390         if (dev->queue_mode == NULL_Q_MQ &&
1391             nullb->tag_set == &nullb->__tag_set)
1392                 blk_mq_free_tag_set(nullb->tag_set);
1393         put_disk(nullb->disk);
1394         cleanup_queues(nullb);
1395         if (null_cache_active(nullb))
1396                 null_free_device_storage(nullb->dev, true);
1397         kfree(nullb);
1398         dev->nullb = NULL;
1399 }
1400
1401 static void null_config_discard(struct nullb *nullb)
1402 {
1403         if (nullb->dev->discard == false)
1404                 return;
1405         nullb->q->limits.discard_granularity = nullb->dev->blocksize;
1406         nullb->q->limits.discard_alignment = nullb->dev->blocksize;
1407         blk_queue_max_discard_sectors(nullb->q, UINT_MAX >> 9);
1408         blk_queue_flag_set(QUEUE_FLAG_DISCARD, nullb->q);
1409 }
1410
1411 static int null_open(struct block_device *bdev, fmode_t mode)
1412 {
1413         return 0;
1414 }
1415
1416 static void null_release(struct gendisk *disk, fmode_t mode)
1417 {
1418 }
1419
1420 static const struct block_device_operations null_fops = {
1421         .owner =        THIS_MODULE,
1422         .open =         null_open,
1423         .release =      null_release,
1424         .report_zones = null_zone_report,
1425 };
1426
1427 static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq)
1428 {
1429         BUG_ON(!nullb);
1430         BUG_ON(!nq);
1431
1432         init_waitqueue_head(&nq->wait);
1433         nq->queue_depth = nullb->queue_depth;
1434         nq->dev = nullb->dev;
1435 }
1436
1437 static void null_init_queues(struct nullb *nullb)
1438 {
1439         struct request_queue *q = nullb->q;
1440         struct blk_mq_hw_ctx *hctx;
1441         struct nullb_queue *nq;
1442         int i;
1443
1444         queue_for_each_hw_ctx(q, hctx, i) {
1445                 if (!hctx->nr_ctx || !hctx->tags)
1446                         continue;
1447                 nq = &nullb->queues[i];
1448                 hctx->driver_data = nq;
1449                 null_init_queue(nullb, nq);
1450                 nullb->nr_queues++;
1451         }
1452 }
1453
1454 static int setup_commands(struct nullb_queue *nq)
1455 {
1456         struct nullb_cmd *cmd;
1457         int i, tag_size;
1458
1459         nq->cmds = kcalloc(nq->queue_depth, sizeof(*cmd), GFP_KERNEL);
1460         if (!nq->cmds)
1461                 return -ENOMEM;
1462
1463         tag_size = ALIGN(nq->queue_depth, BITS_PER_LONG) / BITS_PER_LONG;
1464         nq->tag_map = kcalloc(tag_size, sizeof(unsigned long), GFP_KERNEL);
1465         if (!nq->tag_map) {
1466                 kfree(nq->cmds);
1467                 return -ENOMEM;
1468         }
1469
1470         for (i = 0; i < nq->queue_depth; i++) {
1471                 cmd = &nq->cmds[i];
1472                 INIT_LIST_HEAD(&cmd->list);
1473                 cmd->ll_list.next = NULL;
1474                 cmd->tag = -1U;
1475         }
1476
1477         return 0;
1478 }
1479
1480 static int setup_queues(struct nullb *nullb)
1481 {
1482         nullb->queues = kcalloc(nullb->dev->submit_queues,
1483                                 sizeof(struct nullb_queue),
1484                                 GFP_KERNEL);
1485         if (!nullb->queues)
1486                 return -ENOMEM;
1487
1488         nullb->nr_queues = 0;
1489         nullb->queue_depth = nullb->dev->hw_queue_depth;
1490
1491         return 0;
1492 }
1493
1494 static int init_driver_queues(struct nullb *nullb)
1495 {
1496         struct nullb_queue *nq;
1497         int i, ret = 0;
1498
1499         for (i = 0; i < nullb->dev->submit_queues; i++) {
1500                 nq = &nullb->queues[i];
1501
1502                 null_init_queue(nullb, nq);
1503
1504                 ret = setup_commands(nq);
1505                 if (ret)
1506                         return ret;
1507                 nullb->nr_queues++;
1508         }
1509         return 0;
1510 }
1511
1512 static int null_gendisk_register(struct nullb *nullb)
1513 {
1514         struct gendisk *disk;
1515         sector_t size;
1516
1517         disk = nullb->disk = alloc_disk_node(1, nullb->dev->home_node);
1518         if (!disk)
1519                 return -ENOMEM;
1520         size = (sector_t)nullb->dev->size * 1024 * 1024ULL;
1521         set_capacity(disk, size >> 9);
1522
1523         disk->flags |= GENHD_FL_EXT_DEVT | GENHD_FL_SUPPRESS_PARTITION_INFO;
1524         disk->major             = null_major;
1525         disk->first_minor       = nullb->index;
1526         disk->fops              = &null_fops;
1527         disk->private_data      = nullb;
1528         disk->queue             = nullb->q;
1529         strncpy(disk->disk_name, nullb->disk_name, DISK_NAME_LEN);
1530
1531         if (nullb->dev->zoned) {
1532                 int ret = blk_revalidate_disk_zones(disk);
1533
1534                 if (ret != 0)
1535                         return ret;
1536         }
1537
1538         add_disk(disk);
1539         return 0;
1540 }
1541
1542 static int null_init_tag_set(struct nullb *nullb, struct blk_mq_tag_set *set)
1543 {
1544         set->ops = &null_mq_ops;
1545         set->nr_hw_queues = nullb ? nullb->dev->submit_queues :
1546                                                 g_submit_queues;
1547         set->queue_depth = nullb ? nullb->dev->hw_queue_depth :
1548                                                 g_hw_queue_depth;
1549         set->numa_node = nullb ? nullb->dev->home_node : g_home_node;
1550         set->cmd_size   = sizeof(struct nullb_cmd);
1551         set->flags = BLK_MQ_F_SHOULD_MERGE;
1552         if (g_no_sched)
1553                 set->flags |= BLK_MQ_F_NO_SCHED;
1554         set->driver_data = NULL;
1555
1556         if ((nullb && nullb->dev->blocking) || g_blocking)
1557                 set->flags |= BLK_MQ_F_BLOCKING;
1558
1559         return blk_mq_alloc_tag_set(set);
1560 }
1561
1562 static void null_validate_conf(struct nullb_device *dev)
1563 {
1564         dev->blocksize = round_down(dev->blocksize, 512);
1565         dev->blocksize = clamp_t(unsigned int, dev->blocksize, 512, 4096);
1566
1567         if (dev->queue_mode == NULL_Q_MQ && dev->use_per_node_hctx) {
1568                 if (dev->submit_queues != nr_online_nodes)
1569                         dev->submit_queues = nr_online_nodes;
1570         } else if (dev->submit_queues > nr_cpu_ids)
1571                 dev->submit_queues = nr_cpu_ids;
1572         else if (dev->submit_queues == 0)
1573                 dev->submit_queues = 1;
1574
1575         dev->queue_mode = min_t(unsigned int, dev->queue_mode, NULL_Q_MQ);
1576         dev->irqmode = min_t(unsigned int, dev->irqmode, NULL_IRQ_TIMER);
1577
1578         /* Do memory allocation, so set blocking */
1579         if (dev->memory_backed)
1580                 dev->blocking = true;
1581         else /* cache is meaningless */
1582                 dev->cache_size = 0;
1583         dev->cache_size = min_t(unsigned long, ULONG_MAX / 1024 / 1024,
1584                                                 dev->cache_size);
1585         dev->mbps = min_t(unsigned int, 1024 * 40, dev->mbps);
1586         /* can not stop a queue */
1587         if (dev->queue_mode == NULL_Q_BIO)
1588                 dev->mbps = 0;
1589 }
1590
1591 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1592 static bool __null_setup_fault(struct fault_attr *attr, char *str)
1593 {
1594         if (!str[0])
1595                 return true;
1596
1597         if (!setup_fault_attr(attr, str))
1598                 return false;
1599
1600         attr->verbose = 0;
1601         return true;
1602 }
1603 #endif
1604
1605 static bool null_setup_fault(void)
1606 {
1607 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1608         if (!__null_setup_fault(&null_timeout_attr, g_timeout_str))
1609                 return false;
1610         if (!__null_setup_fault(&null_requeue_attr, g_requeue_str))
1611                 return false;
1612 #endif
1613         return true;
1614 }
1615
1616 static int null_add_dev(struct nullb_device *dev)
1617 {
1618         struct nullb *nullb;
1619         int rv;
1620
1621         null_validate_conf(dev);
1622
1623         nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, dev->home_node);
1624         if (!nullb) {
1625                 rv = -ENOMEM;
1626                 goto out;
1627         }
1628         nullb->dev = dev;
1629         dev->nullb = nullb;
1630
1631         spin_lock_init(&nullb->lock);
1632
1633         rv = setup_queues(nullb);
1634         if (rv)
1635                 goto out_free_nullb;
1636
1637         if (dev->queue_mode == NULL_Q_MQ) {
1638                 if (shared_tags) {
1639                         nullb->tag_set = &tag_set;
1640                         rv = 0;
1641                 } else {
1642                         nullb->tag_set = &nullb->__tag_set;
1643                         rv = null_init_tag_set(nullb, nullb->tag_set);
1644                 }
1645
1646                 if (rv)
1647                         goto out_cleanup_queues;
1648
1649                 if (!null_setup_fault())
1650                         goto out_cleanup_queues;
1651
1652                 nullb->tag_set->timeout = 5 * HZ;
1653                 nullb->q = blk_mq_init_queue(nullb->tag_set);
1654                 if (IS_ERR(nullb->q)) {
1655                         rv = -ENOMEM;
1656                         goto out_cleanup_tags;
1657                 }
1658                 null_init_queues(nullb);
1659         } else if (dev->queue_mode == NULL_Q_BIO) {
1660                 nullb->q = blk_alloc_queue_node(GFP_KERNEL, dev->home_node,
1661                                                 NULL);
1662                 if (!nullb->q) {
1663                         rv = -ENOMEM;
1664                         goto out_cleanup_queues;
1665                 }
1666                 blk_queue_make_request(nullb->q, null_queue_bio);
1667                 rv = init_driver_queues(nullb);
1668                 if (rv)
1669                         goto out_cleanup_blk_queue;
1670         }
1671
1672         if (dev->mbps) {
1673                 set_bit(NULLB_DEV_FL_THROTTLED, &dev->flags);
1674                 nullb_setup_bwtimer(nullb);
1675         }
1676
1677         if (dev->cache_size > 0) {
1678                 set_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
1679                 blk_queue_write_cache(nullb->q, true, true);
1680                 blk_queue_flush_queueable(nullb->q, true);
1681         }
1682
1683         if (dev->zoned) {
1684                 rv = null_zone_init(dev);
1685                 if (rv)
1686                         goto out_cleanup_blk_queue;
1687
1688                 blk_queue_chunk_sectors(nullb->q, dev->zone_size_sects);
1689                 nullb->q->limits.zoned = BLK_ZONED_HM;
1690         }
1691
1692         nullb->q->queuedata = nullb;
1693         blk_queue_flag_set(QUEUE_FLAG_NONROT, nullb->q);
1694         blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, nullb->q);
1695
1696         mutex_lock(&lock);
1697         nullb->index = ida_simple_get(&nullb_indexes, 0, 0, GFP_KERNEL);
1698         dev->index = nullb->index;
1699         mutex_unlock(&lock);
1700
1701         blk_queue_logical_block_size(nullb->q, dev->blocksize);
1702         blk_queue_physical_block_size(nullb->q, dev->blocksize);
1703
1704         null_config_discard(nullb);
1705
1706         sprintf(nullb->disk_name, "nullb%d", nullb->index);
1707
1708         rv = null_gendisk_register(nullb);
1709         if (rv)
1710                 goto out_cleanup_zone;
1711
1712         mutex_lock(&lock);
1713         list_add_tail(&nullb->list, &nullb_list);
1714         mutex_unlock(&lock);
1715
1716         return 0;
1717 out_cleanup_zone:
1718         if (dev->zoned)
1719                 null_zone_exit(dev);
1720 out_cleanup_blk_queue:
1721         blk_cleanup_queue(nullb->q);
1722 out_cleanup_tags:
1723         if (dev->queue_mode == NULL_Q_MQ && nullb->tag_set == &nullb->__tag_set)
1724                 blk_mq_free_tag_set(nullb->tag_set);
1725 out_cleanup_queues:
1726         cleanup_queues(nullb);
1727 out_free_nullb:
1728         kfree(nullb);
1729 out:
1730         return rv;
1731 }
1732
1733 static int __init null_init(void)
1734 {
1735         int ret = 0;
1736         unsigned int i;
1737         struct nullb *nullb;
1738         struct nullb_device *dev;
1739
1740         if (g_bs > PAGE_SIZE) {
1741                 pr_warn("null_blk: invalid block size\n");
1742                 pr_warn("null_blk: defaults block size to %lu\n", PAGE_SIZE);
1743                 g_bs = PAGE_SIZE;
1744         }
1745
1746         if (!is_power_of_2(g_zone_size)) {
1747                 pr_err("null_blk: zone_size must be power-of-two\n");
1748                 return -EINVAL;
1749         }
1750
1751         if (g_queue_mode == NULL_Q_RQ) {
1752                 pr_err("null_blk: legacy IO path no longer available\n");
1753                 return -EINVAL;
1754         }
1755         if (g_queue_mode == NULL_Q_MQ && g_use_per_node_hctx) {
1756                 if (g_submit_queues != nr_online_nodes) {
1757                         pr_warn("null_blk: submit_queues param is set to %u.\n",
1758                                                         nr_online_nodes);
1759                         g_submit_queues = nr_online_nodes;
1760                 }
1761         } else if (g_submit_queues > nr_cpu_ids)
1762                 g_submit_queues = nr_cpu_ids;
1763         else if (g_submit_queues <= 0)
1764                 g_submit_queues = 1;
1765
1766         if (g_queue_mode == NULL_Q_MQ && shared_tags) {
1767                 ret = null_init_tag_set(NULL, &tag_set);
1768                 if (ret)
1769                         return ret;
1770         }
1771
1772         config_group_init(&nullb_subsys.su_group);
1773         mutex_init(&nullb_subsys.su_mutex);
1774
1775         ret = configfs_register_subsystem(&nullb_subsys);
1776         if (ret)
1777                 goto err_tagset;
1778
1779         mutex_init(&lock);
1780
1781         null_major = register_blkdev(0, "nullb");
1782         if (null_major < 0) {
1783                 ret = null_major;
1784                 goto err_conf;
1785         }
1786
1787         for (i = 0; i < nr_devices; i++) {
1788                 dev = null_alloc_dev();
1789                 if (!dev) {
1790                         ret = -ENOMEM;
1791                         goto err_dev;
1792                 }
1793                 ret = null_add_dev(dev);
1794                 if (ret) {
1795                         null_free_dev(dev);
1796                         goto err_dev;
1797                 }
1798         }
1799
1800         pr_info("null: module loaded\n");
1801         return 0;
1802
1803 err_dev:
1804         while (!list_empty(&nullb_list)) {
1805                 nullb = list_entry(nullb_list.next, struct nullb, list);
1806                 dev = nullb->dev;
1807                 null_del_dev(nullb);
1808                 null_free_dev(dev);
1809         }
1810         unregister_blkdev(null_major, "nullb");
1811 err_conf:
1812         configfs_unregister_subsystem(&nullb_subsys);
1813 err_tagset:
1814         if (g_queue_mode == NULL_Q_MQ && shared_tags)
1815                 blk_mq_free_tag_set(&tag_set);
1816         return ret;
1817 }
1818
1819 static void __exit null_exit(void)
1820 {
1821         struct nullb *nullb;
1822
1823         configfs_unregister_subsystem(&nullb_subsys);
1824
1825         unregister_blkdev(null_major, "nullb");
1826
1827         mutex_lock(&lock);
1828         while (!list_empty(&nullb_list)) {
1829                 struct nullb_device *dev;
1830
1831                 nullb = list_entry(nullb_list.next, struct nullb, list);
1832                 dev = nullb->dev;
1833                 null_del_dev(nullb);
1834                 null_free_dev(dev);
1835         }
1836         mutex_unlock(&lock);
1837
1838         if (g_queue_mode == NULL_Q_MQ && shared_tags)
1839                 blk_mq_free_tag_set(&tag_set);
1840 }
1841
1842 module_init(null_init);
1843 module_exit(null_exit);
1844
1845 MODULE_AUTHOR("Jens Axboe <axboe@kernel.dk>");
1846 MODULE_LICENSE("GPL");