drivers/block/null_blk_main.c

   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 #define PAGE_SECTORS_SHIFT      (PAGE_SHIFT - SECTOR_SHIFT)
  15 #define PAGE_SECTORS            (1 << PAGE_SECTORS_SHIFT)
  16 #define SECTOR_MASK             (PAGE_SECTORS - 1)
  17
  18 #define FREE_BATCH              16
  19
  20 #define TICKS_PER_SEC           50ULL
  21 #define TIMER_INTERVAL          (NSEC_PER_SEC / TICKS_PER_SEC)
  22
  23 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
  24 static DECLARE_FAULT_ATTR(null_timeout_attr);
  25 static DECLARE_FAULT_ATTR(null_requeue_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 enum {
  81         NULL_Q_BIO              = 0,
  82         NULL_Q_RQ               = 1,
  83         NULL_Q_MQ               = 2,
  84 };
  85
  86 static int g_no_sched;
  87 module_param_named(no_sched, g_no_sched, int, 0444);
  88 MODULE_PARM_DESC(no_sched, "No io scheduler");
  89
  90 static int g_submit_queues = 1;
  91 module_param_named(submit_queues, g_submit_queues, int, 0444);
  92 MODULE_PARM_DESC(submit_queues, "Number of submission queues");
  93
  94 static int g_home_node = NUMA_NO_NODE;
  95 module_param_named(home_node, g_home_node, int, 0444);
  96 MODULE_PARM_DESC(home_node, "Home node for the device");
  97
  98 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
  99 static char g_timeout_str[80];
 100 module_param_string(timeout, g_timeout_str, sizeof(g_timeout_str), 0444);
 101
 102 static char g_requeue_str[80];
 103 module_param_string(requeue, g_requeue_str, sizeof(g_requeue_str), 0444);
 104 #endif
 105
 106 static int g_queue_mode = NULL_Q_MQ;
 107
 108 static int null_param_store_val(const char *str, int *val, int min, int max)
 109 {
 110         int ret, new_val;
 111
 112         ret = kstrtoint(str, 10, &new_val);
 113         if (ret)
 114                 return -EINVAL;
 115
 116         if (new_val < min || new_val > max)
 117                 return -EINVAL;
 118
 119         *val = new_val;
 120         return 0;
 121 }
 122
 123 static int null_set_queue_mode(const char *str, const struct kernel_param *kp)
 124 {
 125         return null_param_store_val(str, &g_queue_mode, NULL_Q_BIO, NULL_Q_MQ);
 126 }
 127
 128 static const struct kernel_param_ops null_queue_mode_param_ops = {
 129         .set    = null_set_queue_mode,
 130         .get    = param_get_int,
 131 };
 132
 133 device_param_cb(queue_mode, &null_queue_mode_param_ops, &g_queue_mode, 0444);
 134 MODULE_PARM_DESC(queue_mode, "Block interface to use (0=bio,1=rq,2=multiqueue)");
 135
 136 static int g_gb = 250;
 137 module_param_named(gb, g_gb, int, 0444);
 138 MODULE_PARM_DESC(gb, "Size in GB");
 139
 140 static int g_bs = 512;
 141 module_param_named(bs, g_bs, int, 0444);
 142 MODULE_PARM_DESC(bs, "Block size (in bytes)");
 143
 144 static unsigned int nr_devices = 1;
 145 module_param(nr_devices, uint, 0444);
 146 MODULE_PARM_DESC(nr_devices, "Number of devices to register");
 147
 148 static bool g_blocking;
 149 module_param_named(blocking, g_blocking, bool, 0444);
 150 MODULE_PARM_DESC(blocking, "Register as a blocking blk-mq driver device");
 151
 152 static bool shared_tags;
 153 module_param(shared_tags, bool, 0444);
 154 MODULE_PARM_DESC(shared_tags, "Share tag set between devices for blk-mq");
 155
 156 static int g_irqmode = NULL_IRQ_SOFTIRQ;
 157
 158 static int null_set_irqmode(const char *str, const struct kernel_param *kp)
 159 {
 160         return null_param_store_val(str, &g_irqmode, NULL_IRQ_NONE,
 161                                         NULL_IRQ_TIMER);
 162 }
 163
 164 static const struct kernel_param_ops null_irqmode_param_ops = {
 165         .set    = null_set_irqmode,
 166         .get    = param_get_int,
 167 };
 168
 169 device_param_cb(irqmode, &null_irqmode_param_ops, &g_irqmode, 0444);
 170 MODULE_PARM_DESC(irqmode, "IRQ completion handler. 0-none, 1-softirq, 2-timer");
 171
 172 static unsigned long g_completion_nsec = 10000;
 173 module_param_named(completion_nsec, g_completion_nsec, ulong, 0444);
 174 MODULE_PARM_DESC(completion_nsec, "Time in ns to complete a request in hardware. Default: 10,000ns");
 175
 176 static int g_hw_queue_depth = 64;
 177 module_param_named(hw_queue_depth, g_hw_queue_depth, int, 0444);
 178 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 64");
 179
 180 static bool g_use_per_node_hctx;
 181 module_param_named(use_per_node_hctx, g_use_per_node_hctx, bool, 0444);
 182 MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware context queues. Default: false");
 183
 184 static bool g_zoned;
 185 module_param_named(zoned, g_zoned, bool, S_IRUGO);
 186 MODULE_PARM_DESC(zoned, "Make device as a host-managed zoned block device. Default: false");
 187
 188 static unsigned long g_zone_size = 256;
 189 module_param_named(zone_size, g_zone_size, ulong, S_IRUGO);
 190 MODULE_PARM_DESC(zone_size, "Zone size in MB when block device is zoned. Must be power-of-two: Default: 256");
 191
 192 static unsigned int g_zone_nr_conv;
 193 module_param_named(zone_nr_conv, g_zone_nr_conv, uint, 0444);
 194 MODULE_PARM_DESC(zone_nr_conv, "Number of conventional zones when block device is zoned. Default: 0");
 195
 196 static struct nullb_device *null_alloc_dev(void);
 197 static void null_free_dev(struct nullb_device *dev);
 198 static void null_del_dev(struct nullb *nullb);
 199 static int null_add_dev(struct nullb_device *dev);
 200 static void null_free_device_storage(struct nullb_device *dev, bool is_cache);
 201
 202 static inline struct nullb_device *to_nullb_device(struct config_item *item)
 203 {
 204         return item ? container_of(item, struct nullb_device, item) : NULL;
 205 }
 206
 207 static inline ssize_t nullb_device_uint_attr_show(unsigned int val, char *page)
 208 {
 209         return snprintf(page, PAGE_SIZE, "%u\n", val);
 210 }
 211
 212 static inline ssize_t nullb_device_ulong_attr_show(unsigned long val,
 213         char *page)
 214 {
 215         return snprintf(page, PAGE_SIZE, "%lu\n", val);
 216 }
 217
 218 static inline ssize_t nullb_device_bool_attr_show(bool val, char *page)
 219 {
 220         return snprintf(page, PAGE_SIZE, "%u\n", val);
 221 }
 222
 223 static ssize_t nullb_device_uint_attr_store(unsigned int *val,
 224         const char *page, size_t count)
 225 {
 226         unsigned int tmp;
 227         int result;
 228
 229         result = kstrtouint(page, 0, &tmp);
 230         if (result)
 231                 return result;
 232
 233         *val = tmp;
 234         return count;
 235 }
 236
 237 static ssize_t nullb_device_ulong_attr_store(unsigned long *val,
 238         const char *page, size_t count)
 239 {
 240         int result;
 241         unsigned long tmp;
 242
 243         result = kstrtoul(page, 0, &tmp);
 244         if (result)
 245                 return result;
 246
 247         *val = tmp;
 248         return count;
 249 }
 250
 251 static ssize_t nullb_device_bool_attr_store(bool *val, const char *page,
 252         size_t count)
 253 {
 254         bool tmp;
 255         int result;
 256
 257         result = kstrtobool(page,  &tmp);
 258         if (result)
 259                 return result;
 260
 261         *val = tmp;
 262         return count;
 263 }
 264
 265 /* The following macro should only be used with TYPE = {uint, ulong, bool}. */
 266 #define NULLB_DEVICE_ATTR(NAME, TYPE)                                           \
 267 static ssize_t                                                                  \
 268 nullb_device_##NAME##_show(struct config_item *item, char *page)                \
 269 {                                                                               \
 270         return nullb_device_##TYPE##_attr_show(                                 \
 271                                 to_nullb_device(item)->NAME, page);             \
 272 }                                                                               \
 273 static ssize_t                                                                  \
 274 nullb_device_##NAME##_store(struct config_item *item, const char *page,         \
 275                             size_t count)                                       \
 276 {                                                                               \
 277         if (test_bit(NULLB_DEV_FL_CONFIGURED, &to_nullb_device(item)->flags))   \
 278                 return -EBUSY;                                                  \
 279         return nullb_device_##TYPE##_attr_store(                                \
 280                         &to_nullb_device(item)->NAME, page, count);             \
 281 }                                                                               \
 282 CONFIGFS_ATTR(nullb_device_, NAME);
 283
 284 NULLB_DEVICE_ATTR(size, ulong);
 285 NULLB_DEVICE_ATTR(completion_nsec, ulong);
 286 NULLB_DEVICE_ATTR(submit_queues, uint);
 287 NULLB_DEVICE_ATTR(home_node, uint);
 288 NULLB_DEVICE_ATTR(queue_mode, uint);
 289 NULLB_DEVICE_ATTR(blocksize, uint);
 290 NULLB_DEVICE_ATTR(irqmode, uint);
 291 NULLB_DEVICE_ATTR(hw_queue_depth, uint);
 292 NULLB_DEVICE_ATTR(index, uint);
 293 NULLB_DEVICE_ATTR(blocking, bool);
 294 NULLB_DEVICE_ATTR(use_per_node_hctx, bool);
 295 NULLB_DEVICE_ATTR(memory_backed, bool);
 296 NULLB_DEVICE_ATTR(discard, bool);
 297 NULLB_DEVICE_ATTR(mbps, uint);
 298 NULLB_DEVICE_ATTR(cache_size, ulong);
 299 NULLB_DEVICE_ATTR(zoned, bool);
 300 NULLB_DEVICE_ATTR(zone_size, ulong);
 301 NULLB_DEVICE_ATTR(zone_nr_conv, uint);
 302
 303 static ssize_t nullb_device_power_show(struct config_item *item, char *page)
 304 {
 305         return nullb_device_bool_attr_show(to_nullb_device(item)->power, page);
 306 }
 307
 308 static ssize_t nullb_device_power_store(struct config_item *item,
 309                                      const char *page, size_t count)
 310 {
 311         struct nullb_device *dev = to_nullb_device(item);
 312         bool newp = false;
 313         ssize_t ret;
 314
 315         ret = nullb_device_bool_attr_store(&newp, page, count);
 316         if (ret < 0)
 317                 return ret;
 318
 319         if (!dev->power && newp) {
 320                 if (test_and_set_bit(NULLB_DEV_FL_UP, &dev->flags))
 321                         return count;
 322                 if (null_add_dev(dev)) {
 323                         clear_bit(NULLB_DEV_FL_UP, &dev->flags);
 324                         return -ENOMEM;
 325                 }
 326
 327                 set_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
 328                 dev->power = newp;
 329         } else if (dev->power && !newp) {
 330                 if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
 331                         mutex_lock(&lock);
 332                         dev->power = newp;
 333                         null_del_dev(dev->nullb);
 334                         mutex_unlock(&lock);
 335                 }
 336                 clear_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
 337         }
 338
 339         return count;
 340 }
 341
 342 CONFIGFS_ATTR(nullb_device_, power);
 343
 344 static ssize_t nullb_device_badblocks_show(struct config_item *item, char *page)
 345 {
 346         struct nullb_device *t_dev = to_nullb_device(item);
 347
 348         return badblocks_show(&t_dev->badblocks, page, 0);
 349 }
 350
 351 static ssize_t nullb_device_badblocks_store(struct config_item *item,
 352                                      const char *page, size_t count)
 353 {
 354         struct nullb_device *t_dev = to_nullb_device(item);
 355         char *orig, *buf, *tmp;
 356         u64 start, end;
 357         int ret;
 358
 359         orig = kstrndup(page, count, GFP_KERNEL);
 360         if (!orig)
 361                 return -ENOMEM;
 362
 363         buf = strstrip(orig);
 364
 365         ret = -EINVAL;
 366         if (buf[0] != '+' && buf[0] != '-')
 367                 goto out;
 368         tmp = strchr(&buf[1], '-');
 369         if (!tmp)
 370                 goto out;
 371         *tmp = '\0';
 372         ret = kstrtoull(buf + 1, 0, &start);
 373         if (ret)
 374                 goto out;
 375         ret = kstrtoull(tmp + 1, 0, &end);
 376         if (ret)
 377                 goto out;
 378         ret = -EINVAL;
 379         if (start > end)
 380                 goto out;
 381         /* enable badblocks */
 382         cmpxchg(&t_dev->badblocks.shift, -1, 0);
 383         if (buf[0] == '+')
 384                 ret = badblocks_set(&t_dev->badblocks, start,
 385                         end - start + 1, 1);
 386         else
 387                 ret = badblocks_clear(&t_dev->badblocks, start,
 388                         end - start + 1);
 389         if (ret == 0)
 390                 ret = count;
 391 out:
 392         kfree(orig);
 393         return ret;
 394 }
 395 CONFIGFS_ATTR(nullb_device_, badblocks);
 396
 397 static struct configfs_attribute *nullb_device_attrs[] = {
 398         &nullb_device_attr_size,
 399         &nullb_device_attr_completion_nsec,
 400         &nullb_device_attr_submit_queues,
 401         &nullb_device_attr_home_node,
 402         &nullb_device_attr_queue_mode,
 403         &nullb_device_attr_blocksize,
 404         &nullb_device_attr_irqmode,
 405         &nullb_device_attr_hw_queue_depth,
 406         &nullb_device_attr_index,
 407         &nullb_device_attr_blocking,
 408         &nullb_device_attr_use_per_node_hctx,
 409         &nullb_device_attr_power,
 410         &nullb_device_attr_memory_backed,
 411         &nullb_device_attr_discard,
 412         &nullb_device_attr_mbps,
 413         &nullb_device_attr_cache_size,
 414         &nullb_device_attr_badblocks,
 415         &nullb_device_attr_zoned,
 416         &nullb_device_attr_zone_size,
 417         &nullb_device_attr_zone_nr_conv,
 418         NULL,
 419 };
 420
 421 static void nullb_device_release(struct config_item *item)
 422 {
 423         struct nullb_device *dev = to_nullb_device(item);
 424
 425         null_free_device_storage(dev, false);
 426         null_free_dev(dev);
 427 }
 428
 429 static struct configfs_item_operations nullb_device_ops = {
 430         .release        = nullb_device_release,
 431 };
 432
 433 static const struct config_item_type nullb_device_type = {
 434         .ct_item_ops    = &nullb_device_ops,
 435         .ct_attrs       = nullb_device_attrs,
 436         .ct_owner       = THIS_MODULE,
 437 };
 438
 439 static struct
 440 config_item *nullb_group_make_item(struct config_group *group, const char *name)
 441 {
 442         struct nullb_device *dev;
 443
 444         dev = null_alloc_dev();
 445         if (!dev)
 446                 return ERR_PTR(-ENOMEM);
 447
 448         config_item_init_type_name(&dev->item, name, &nullb_device_type);
 449
 450         return &dev->item;
 451 }
 452
 453 static void
 454 nullb_group_drop_item(struct config_group *group, struct config_item *item)
 455 {
 456         struct nullb_device *dev = to_nullb_device(item);
 457
 458         if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
 459                 mutex_lock(&lock);
 460                 dev->power = false;
 461                 null_del_dev(dev->nullb);
 462                 mutex_unlock(&lock);
 463         }
 464
 465         config_item_put(item);
 466 }
 467
 468 static ssize_t memb_group_features_show(struct config_item *item, char *page)
 469 {
 470         return snprintf(page, PAGE_SIZE, "memory_backed,discard,bandwidth,cache,badblocks,zoned,zone_size\n");
 471 }
 472
 473 CONFIGFS_ATTR_RO(memb_group_, features);
 474
 475 static struct configfs_attribute *nullb_group_attrs[] = {
 476         &memb_group_attr_features,
 477         NULL,
 478 };
 479
 480 static struct configfs_group_operations nullb_group_ops = {
 481         .make_item      = nullb_group_make_item,
 482         .drop_item      = nullb_group_drop_item,
 483 };
 484
 485 static const struct config_item_type nullb_group_type = {
 486         .ct_group_ops   = &nullb_group_ops,
 487         .ct_attrs       = nullb_group_attrs,
 488         .ct_owner       = THIS_MODULE,
 489 };
 490
 491 static struct configfs_subsystem nullb_subsys = {
 492         .su_group = {
 493                 .cg_item = {
 494                         .ci_namebuf = "nullb",
 495                         .ci_type = &nullb_group_type,
 496                 },
 497         },
 498 };
 499
 500 static inline int null_cache_active(struct nullb *nullb)
 501 {
 502         return test_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
 503 }
 504
 505 static struct nullb_device *null_alloc_dev(void)
 506 {
 507         struct nullb_device *dev;
 508
 509         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
 510         if (!dev)
 511                 return NULL;
 512         INIT_RADIX_TREE(&dev->data, GFP_ATOMIC);
 513         INIT_RADIX_TREE(&dev->cache, GFP_ATOMIC);
 514         if (badblocks_init(&dev->badblocks, 0)) {
 515                 kfree(dev);
 516                 return NULL;
 517         }
 518
 519         dev->size = g_gb * 1024;
 520         dev->completion_nsec = g_completion_nsec;
 521         dev->submit_queues = g_submit_queues;
 522         dev->home_node = g_home_node;
 523         dev->queue_mode = g_queue_mode;
 524         dev->blocksize = g_bs;
 525         dev->irqmode = g_irqmode;
 526         dev->hw_queue_depth = g_hw_queue_depth;
 527         dev->blocking = g_blocking;
 528         dev->use_per_node_hctx = g_use_per_node_hctx;
 529         dev->zoned = g_zoned;
 530         dev->zone_size = g_zone_size;
 531         dev->zone_nr_conv = g_zone_nr_conv;
 532         return dev;
 533 }
 534
 535 static void null_free_dev(struct nullb_device *dev)
 536 {
 537         if (!dev)
 538                 return;
 539
 540         null_zone_exit(dev);
 541         badblocks_exit(&dev->badblocks);
 542         kfree(dev);
 543 }
 544
 545 static void put_tag(struct nullb_queue *nq, unsigned int tag)
 546 {
 547         clear_bit_unlock(tag, nq->tag_map);
 548
 549         if (waitqueue_active(&nq->wait))
 550                 wake_up(&nq->wait);
 551 }
 552
 553 static unsigned int get_tag(struct nullb_queue *nq)
 554 {
 555         unsigned int tag;
 556
 557         do {
 558                 tag = find_first_zero_bit(nq->tag_map, nq->queue_depth);
 559                 if (tag >= nq->queue_depth)
 560                         return -1U;
 561         } while (test_and_set_bit_lock(tag, nq->tag_map));
 562
 563         return tag;
 564 }
 565
 566 static void free_cmd(struct nullb_cmd *cmd)
 567 {
 568         put_tag(cmd->nq, cmd->tag);
 569 }
 570
 571 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer);
 572
 573 static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq)
 574 {
 575         struct nullb_cmd *cmd;
 576         unsigned int tag;
 577
 578         tag = get_tag(nq);
 579         if (tag != -1U) {
 580                 cmd = &nq->cmds[tag];
 581                 cmd->tag = tag;
 582                 cmd->nq = nq;
 583                 if (nq->dev->irqmode == NULL_IRQ_TIMER) {
 584                         hrtimer_init(&cmd->timer, CLOCK_MONOTONIC,
 585                                      HRTIMER_MODE_REL);
 586                         cmd->timer.function = null_cmd_timer_expired;
 587                 }
 588                 return cmd;
 589         }
 590
 591         return NULL;
 592 }
 593
 594 static struct nullb_cmd *alloc_cmd(struct nullb_queue *nq, int can_wait)
 595 {
 596         struct nullb_cmd *cmd;
 597         DEFINE_WAIT(wait);
 598
 599         cmd = __alloc_cmd(nq);
 600         if (cmd || !can_wait)
 601                 return cmd;
 602
 603         do {
 604                 prepare_to_wait(&nq->wait, &wait, TASK_UNINTERRUPTIBLE);
 605                 cmd = __alloc_cmd(nq);
 606                 if (cmd)
 607                         break;
 608
 609                 io_schedule();
 610         } while (1);
 611
 612         finish_wait(&nq->wait, &wait);
 613         return cmd;
 614 }
 615
 616 static void end_cmd(struct nullb_cmd *cmd)
 617 {
 618         int queue_mode = cmd->nq->dev->queue_mode;
 619
 620         switch (queue_mode)  {
 621         case NULL_Q_MQ:
 622                 blk_mq_end_request(cmd->rq, cmd->error);
 623                 return;
 624         case NULL_Q_BIO:
 625                 cmd->bio->bi_status = cmd->error;
 626                 bio_endio(cmd->bio);
 627                 break;
 628         }
 629
 630         free_cmd(cmd);
 631 }
 632
 633 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
 634 {
 635         end_cmd(container_of(timer, struct nullb_cmd, timer));
 636
 637         return HRTIMER_NORESTART;
 638 }
 639
 640 static void null_cmd_end_timer(struct nullb_cmd *cmd)
 641 {
 642         ktime_t kt = cmd->nq->dev->completion_nsec;
 643
 644         hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL);
 645 }
 646
 647 static void null_complete_rq(struct request *rq)
 648 {
 649         end_cmd(blk_mq_rq_to_pdu(rq));
 650 }
 651
 652 static struct nullb_page *null_alloc_page(gfp_t gfp_flags)
 653 {
 654         struct nullb_page *t_page;
 655
 656         t_page = kmalloc(sizeof(struct nullb_page), gfp_flags);
 657         if (!t_page)
 658                 goto out;
 659
 660         t_page->page = alloc_pages(gfp_flags, 0);
 661         if (!t_page->page)
 662                 goto out_freepage;
 663
 664         memset(t_page->bitmap, 0, sizeof(t_page->bitmap));
 665         return t_page;
 666 out_freepage:
 667         kfree(t_page);
 668 out:
 669         return NULL;
 670 }
 671
 672 static void null_free_page(struct nullb_page *t_page)
 673 {
 674         __set_bit(NULLB_PAGE_FREE, t_page->bitmap);
 675         if (test_bit(NULLB_PAGE_LOCK, t_page->bitmap))
 676                 return;
 677         __free_page(t_page->page);
 678         kfree(t_page);
 679 }
 680
 681 static bool null_page_empty(struct nullb_page *page)
 682 {
 683         int size = MAP_SZ - 2;
 684
 685         return find_first_bit(page->bitmap, size) == size;
 686 }
 687
 688 static void null_free_sector(struct nullb *nullb, sector_t sector,
 689         bool is_cache)
 690 {
 691         unsigned int sector_bit;
 692         u64 idx;
 693         struct nullb_page *t_page, *ret;
 694         struct radix_tree_root *root;
 695
 696         root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
 697         idx = sector >> PAGE_SECTORS_SHIFT;
 698         sector_bit = (sector & SECTOR_MASK);
 699
 700         t_page = radix_tree_lookup(root, idx);
 701         if (t_page) {
 702                 __clear_bit(sector_bit, t_page->bitmap);
 703
 704                 if (null_page_empty(t_page)) {
 705                         ret = radix_tree_delete_item(root, idx, t_page);
 706                         WARN_ON(ret != t_page);
 707                         null_free_page(ret);
 708                         if (is_cache)
 709                                 nullb->dev->curr_cache -= PAGE_SIZE;
 710                 }
 711         }
 712 }
 713
 714 static struct nullb_page *null_radix_tree_insert(struct nullb *nullb, u64 idx,
 715         struct nullb_page *t_page, bool is_cache)
 716 {
 717         struct radix_tree_root *root;
 718
 719         root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
 720
 721         if (radix_tree_insert(root, idx, t_page)) {
 722                 null_free_page(t_page);
 723                 t_page = radix_tree_lookup(root, idx);
 724                 WARN_ON(!t_page || t_page->page->index != idx);
 725         } else if (is_cache)
 726                 nullb->dev->curr_cache += PAGE_SIZE;
 727
 728         return t_page;
 729 }
 730
 731 static void null_free_device_storage(struct nullb_device *dev, bool is_cache)
 732 {
 733         unsigned long pos = 0;
 734         int nr_pages;
 735         struct nullb_page *ret, *t_pages[FREE_BATCH];
 736         struct radix_tree_root *root;
 737
 738         root = is_cache ? &dev->cache : &dev->data;
 739
 740         do {
 741                 int i;
 742
 743                 nr_pages = radix_tree_gang_lookup(root,
 744                                 (void **)t_pages, pos, FREE_BATCH);
 745
 746                 for (i = 0; i < nr_pages; i++) {
 747                         pos = t_pages[i]->page->index;
 748                         ret = radix_tree_delete_item(root, pos, t_pages[i]);
 749                         WARN_ON(ret != t_pages[i]);
 750                         null_free_page(ret);
 751                 }
 752
 753                 pos++;
 754         } while (nr_pages == FREE_BATCH);
 755
 756         if (is_cache)
 757                 dev->curr_cache = 0;
 758 }
 759
 760 static struct nullb_page *__null_lookup_page(struct nullb *nullb,
 761         sector_t sector, bool for_write, bool is_cache)
 762 {
 763         unsigned int sector_bit;
 764         u64 idx;
 765         struct nullb_page *t_page;
 766         struct radix_tree_root *root;
 767
 768         idx = sector >> PAGE_SECTORS_SHIFT;
 769         sector_bit = (sector & SECTOR_MASK);
 770
 771         root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
 772         t_page = radix_tree_lookup(root, idx);
 773         WARN_ON(t_page && t_page->page->index != idx);
 774
 775         if (t_page && (for_write || test_bit(sector_bit, t_page->bitmap)))
 776                 return t_page;
 777
 778         return NULL;
 779 }
 780
 781 static struct nullb_page *null_lookup_page(struct nullb *nullb,
 782         sector_t sector, bool for_write, bool ignore_cache)
 783 {
 784         struct nullb_page *page = NULL;
 785
 786         if (!ignore_cache)
 787                 page = __null_lookup_page(nullb, sector, for_write, true);
 788         if (page)
 789                 return page;
 790         return __null_lookup_page(nullb, sector, for_write, false);
 791 }
 792
 793 static struct nullb_page *null_insert_page(struct nullb *nullb,
 794                                            sector_t sector, bool ignore_cache)
 795         __releases(&nullb->lock)
 796         __acquires(&nullb->lock)
 797 {
 798         u64 idx;
 799         struct nullb_page *t_page;
 800
 801         t_page = null_lookup_page(nullb, sector, true, ignore_cache);
 802         if (t_page)
 803                 return t_page;
 804
 805         spin_unlock_irq(&nullb->lock);
 806
 807         t_page = null_alloc_page(GFP_NOIO);
 808         if (!t_page)
 809                 goto out_lock;
 810
 811         if (radix_tree_preload(GFP_NOIO))
 812                 goto out_freepage;
 813
 814         spin_lock_irq(&nullb->lock);
 815         idx = sector >> PAGE_SECTORS_SHIFT;
 816         t_page->page->index = idx;
 817         t_page = null_radix_tree_insert(nullb, idx, t_page, !ignore_cache);
 818         radix_tree_preload_end();
 819
 820         return t_page;
 821 out_freepage:
 822         null_free_page(t_page);
 823 out_lock:
 824         spin_lock_irq(&nullb->lock);
 825         return null_lookup_page(nullb, sector, true, ignore_cache);
 826 }
 827
 828 static int null_flush_cache_page(struct nullb *nullb, struct nullb_page *c_page)
 829 {
 830         int i;
 831         unsigned int offset;
 832         u64 idx;
 833         struct nullb_page *t_page, *ret;
 834         void *dst, *src;
 835
 836         idx = c_page->page->index;
 837
 838         t_page = null_insert_page(nullb, idx << PAGE_SECTORS_SHIFT, true);
 839
 840         __clear_bit(NULLB_PAGE_LOCK, c_page->bitmap);
 841         if (test_bit(NULLB_PAGE_FREE, c_page->bitmap)) {
 842                 null_free_page(c_page);
 843                 if (t_page && null_page_empty(t_page)) {
 844                         ret = radix_tree_delete_item(&nullb->dev->data,
 845                                 idx, t_page);
 846                         null_free_page(t_page);
 847                 }
 848                 return 0;
 849         }
 850
 851         if (!t_page)
 852                 return -ENOMEM;
 853
 854         src = kmap_atomic(c_page->page);
 855         dst = kmap_atomic(t_page->page);
 856
 857         for (i = 0; i < PAGE_SECTORS;
 858                         i += (nullb->dev->blocksize >> SECTOR_SHIFT)) {
 859                 if (test_bit(i, c_page->bitmap)) {
 860                         offset = (i << SECTOR_SHIFT);
 861                         memcpy(dst + offset, src + offset,
 862                                 nullb->dev->blocksize);
 863                         __set_bit(i, t_page->bitmap);
 864                 }
 865         }
 866
 867         kunmap_atomic(dst);
 868         kunmap_atomic(src);
 869
 870         ret = radix_tree_delete_item(&nullb->dev->cache, idx, c_page);
 871         null_free_page(ret);
 872         nullb->dev->curr_cache -= PAGE_SIZE;
 873
 874         return 0;
 875 }
 876
 877 static int null_make_cache_space(struct nullb *nullb, unsigned long n)
 878 {
 879         int i, err, nr_pages;
 880         struct nullb_page *c_pages[FREE_BATCH];
 881         unsigned long flushed = 0, one_round;
 882
 883 again:
 884         if ((nullb->dev->cache_size * 1024 * 1024) >
 885              nullb->dev->curr_cache + n || nullb->dev->curr_cache == 0)
 886                 return 0;
 887
 888         nr_pages = radix_tree_gang_lookup(&nullb->dev->cache,
 889                         (void **)c_pages, nullb->cache_flush_pos, FREE_BATCH);
 890         /*
 891          * nullb_flush_cache_page could unlock before using the c_pages. To
 892          * avoid race, we don't allow page free
 893          */
 894         for (i = 0; i < nr_pages; i++) {
 895                 nullb->cache_flush_pos = c_pages[i]->page->index;
 896                 /*
 897                  * We found the page which is being flushed to disk by other
 898                  * threads
 899                  */
 900                 if (test_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap))
 901                         c_pages[i] = NULL;
 902                 else
 903                         __set_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap);
 904         }
 905
 906         one_round = 0;
 907         for (i = 0; i < nr_pages; i++) {
 908                 if (c_pages[i] == NULL)
 909                         continue;
 910                 err = null_flush_cache_page(nullb, c_pages[i]);
 911                 if (err)
 912                         return err;
 913                 one_round++;
 914         }
 915         flushed += one_round << PAGE_SHIFT;
 916
 917         if (n > flushed) {
 918                 if (nr_pages == 0)
 919                         nullb->cache_flush_pos = 0;
 920                 if (one_round == 0) {
 921                         /* give other threads a chance */
 922                         spin_unlock_irq(&nullb->lock);
 923                         spin_lock_irq(&nullb->lock);
 924                 }
 925                 goto again;
 926         }
 927         return 0;
 928 }
 929
 930 static int copy_to_nullb(struct nullb *nullb, struct page *source,
 931         unsigned int off, sector_t sector, size_t n, bool is_fua)
 932 {
 933         size_t temp, count = 0;
 934         unsigned int offset;
 935         struct nullb_page *t_page;
 936         void *dst, *src;
 937
 938         while (count < n) {
 939                 temp = min_t(size_t, nullb->dev->blocksize, n - count);
 940
 941                 if (null_cache_active(nullb) && !is_fua)
 942                         null_make_cache_space(nullb, PAGE_SIZE);
 943
 944                 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
 945                 t_page = null_insert_page(nullb, sector,
 946                         !null_cache_active(nullb) || is_fua);
 947                 if (!t_page)
 948                         return -ENOSPC;
 949
 950                 src = kmap_atomic(source);
 951                 dst = kmap_atomic(t_page->page);
 952                 memcpy(dst + offset, src + off + count, temp);
 953                 kunmap_atomic(dst);
 954                 kunmap_atomic(src);
 955
 956                 __set_bit(sector & SECTOR_MASK, t_page->bitmap);
 957
 958                 if (is_fua)
 959                         null_free_sector(nullb, sector, true);
 960
 961                 count += temp;
 962                 sector += temp >> SECTOR_SHIFT;
 963         }
 964         return 0;
 965 }
 966
 967 static int copy_from_nullb(struct nullb *nullb, struct page *dest,
 968         unsigned int off, sector_t sector, size_t n)
 969 {
 970         size_t temp, count = 0;
 971         unsigned int offset;
 972         struct nullb_page *t_page;
 973         void *dst, *src;
 974
 975         while (count < n) {
 976                 temp = min_t(size_t, nullb->dev->blocksize, n - count);
 977
 978                 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
 979                 t_page = null_lookup_page(nullb, sector, false,
 980                         !null_cache_active(nullb));
 981
 982                 dst = kmap_atomic(dest);
 983                 if (!t_page) {
 984                         memset(dst + off + count, 0, temp);
 985                         goto next;
 986                 }
 987                 src = kmap_atomic(t_page->page);
 988                 memcpy(dst + off + count, src + offset, temp);
 989                 kunmap_atomic(src);
 990 next:
 991                 kunmap_atomic(dst);
 992
 993                 count += temp;
 994                 sector += temp >> SECTOR_SHIFT;
 995         }
 996         return 0;
 997 }
 998
 999 static void null_handle_discard(struct nullb *nullb, sector_t sector, size_t n)
1000 {
1001         size_t temp;
1002
1003         spin_lock_irq(&nullb->lock);
1004         while (n > 0) {
1005                 temp = min_t(size_t, n, nullb->dev->blocksize);
1006                 null_free_sector(nullb, sector, false);
1007                 if (null_cache_active(nullb))
1008                         null_free_sector(nullb, sector, true);
1009                 sector += temp >> SECTOR_SHIFT;
1010                 n -= temp;
1011         }
1012         spin_unlock_irq(&nullb->lock);
1013 }
1014
1015 static int null_handle_flush(struct nullb *nullb)
1016 {
1017         int err;
1018
1019         if (!null_cache_active(nullb))
1020                 return 0;
1021
1022         spin_lock_irq(&nullb->lock);
1023         while (true) {
1024                 err = null_make_cache_space(nullb,
1025                         nullb->dev->cache_size * 1024 * 1024);
1026                 if (err || nullb->dev->curr_cache == 0)
1027                         break;
1028         }
1029
1030         WARN_ON(!radix_tree_empty(&nullb->dev->cache));
1031         spin_unlock_irq(&nullb->lock);
1032         return err;
1033 }
1034
1035 static int null_transfer(struct nullb *nullb, struct page *page,
1036         unsigned int len, unsigned int off, bool is_write, sector_t sector,
1037         bool is_fua)
1038 {
1039         int err = 0;
1040
1041         if (!is_write) {
1042                 err = copy_from_nullb(nullb, page, off, sector, len);
1043                 flush_dcache_page(page);
1044         } else {
1045                 flush_dcache_page(page);
1046                 err = copy_to_nullb(nullb, page, off, sector, len, is_fua);
1047         }
1048
1049         return err;
1050 }
1051
1052 static int null_handle_rq(struct nullb_cmd *cmd)
1053 {
1054         struct request *rq = cmd->rq;
1055         struct nullb *nullb = cmd->nq->dev->nullb;
1056         int err;
1057         unsigned int len;
1058         sector_t sector;
1059         struct req_iterator iter;
1060         struct bio_vec bvec;
1061
1062         sector = blk_rq_pos(rq);
1063
1064         if (req_op(rq) == REQ_OP_DISCARD) {
1065                 null_handle_discard(nullb, sector, blk_rq_bytes(rq));
1066                 return 0;
1067         }
1068
1069         spin_lock_irq(&nullb->lock);
1070         rq_for_each_segment(bvec, rq, iter) {
1071                 len = bvec.bv_len;
1072                 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1073                                      op_is_write(req_op(rq)), sector,
1074                                      req_op(rq) & REQ_FUA);
1075                 if (err) {
1076                         spin_unlock_irq(&nullb->lock);
1077                         return err;
1078                 }
1079                 sector += len >> SECTOR_SHIFT;
1080         }
1081         spin_unlock_irq(&nullb->lock);
1082
1083         return 0;
1084 }
1085
1086 static int null_handle_bio(struct nullb_cmd *cmd)
1087 {
1088         struct bio *bio = cmd->bio;
1089         struct nullb *nullb = cmd->nq->dev->nullb;
1090         int err;
1091         unsigned int len;
1092         sector_t sector;
1093         struct bio_vec bvec;
1094         struct bvec_iter iter;
1095
1096         sector = bio->bi_iter.bi_sector;
1097
1098         if (bio_op(bio) == REQ_OP_DISCARD) {
1099                 null_handle_discard(nullb, sector,
1100                         bio_sectors(bio) << SECTOR_SHIFT);
1101                 return 0;
1102         }
1103
1104         spin_lock_irq(&nullb->lock);
1105         bio_for_each_segment(bvec, bio, iter) {
1106                 len = bvec.bv_len;
1107                 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1108                                      op_is_write(bio_op(bio)), sector,
1109                                      bio->bi_opf & REQ_FUA);
1110                 if (err) {
1111                         spin_unlock_irq(&nullb->lock);
1112                         return err;
1113                 }
1114                 sector += len >> SECTOR_SHIFT;
1115         }
1116         spin_unlock_irq(&nullb->lock);
1117         return 0;
1118 }
1119
1120 static void null_stop_queue(struct nullb *nullb)
1121 {
1122         struct request_queue *q = nullb->q;
1123
1124         if (nullb->dev->queue_mode == NULL_Q_MQ)
1125                 blk_mq_stop_hw_queues(q);
1126 }
1127
1128 static void null_restart_queue_async(struct nullb *nullb)
1129 {
1130         struct request_queue *q = nullb->q;
1131
1132         if (nullb->dev->queue_mode == NULL_Q_MQ)
1133                 blk_mq_start_stopped_hw_queues(q, true);
1134 }
1135
1136 static inline blk_status_t null_handle_throttled(struct nullb_cmd *cmd)
1137 {
1138         struct nullb_device *dev = cmd->nq->dev;
1139         struct nullb *nullb = dev->nullb;
1140         blk_status_t sts = BLK_STS_OK;
1141         struct request *rq = cmd->rq;
1142
1143         if (!hrtimer_active(&nullb->bw_timer))
1144                 hrtimer_restart(&nullb->bw_timer);
1145
1146         if (atomic_long_sub_return(blk_rq_bytes(rq), &nullb->cur_bytes) < 0) {
1147                 null_stop_queue(nullb);
1148                 /* race with timer */
1149                 if (atomic_long_read(&nullb->cur_bytes) > 0)
1150                         null_restart_queue_async(nullb);
1151                 /* requeue request */
1152                 sts = BLK_STS_DEV_RESOURCE;
1153         }
1154         return sts;
1155 }
1156
1157 static inline blk_status_t null_handle_badblocks(struct nullb_cmd *cmd,
1158                                                  sector_t sector,
1159                                                  sector_t nr_sectors)
1160 {
1161         struct badblocks *bb = &cmd->nq->dev->badblocks;
1162         sector_t first_bad;
1163         int bad_sectors;
1164
1165         if (badblocks_check(bb, sector, nr_sectors, &first_bad, &bad_sectors))
1166                 return BLK_STS_IOERR;
1167
1168         return BLK_STS_OK;
1169 }
1170
1171 static inline blk_status_t null_handle_memory_backed(struct nullb_cmd *cmd,
1172                                                      enum req_opf op)
1173 {
1174         struct nullb_device *dev = cmd->nq->dev;
1175         int err;
1176
1177         if (dev->queue_mode == NULL_Q_BIO)
1178                 err = null_handle_bio(cmd);
1179         else
1180                 err = null_handle_rq(cmd);
1181
1182         return errno_to_blk_status(err);
1183 }
1184
1185 static inline void nullb_complete_cmd(struct nullb_cmd *cmd)
1186 {
1187         /* Complete IO by inline, softirq or timer */
1188         switch (cmd->nq->dev->irqmode) {
1189         case NULL_IRQ_SOFTIRQ:
1190                 switch (cmd->nq->dev->queue_mode) {
1191                 case NULL_Q_MQ:
1192                         blk_mq_complete_request(cmd->rq);
1193                         break;
1194                 case NULL_Q_BIO:
1195                         /*
1196                          * XXX: no proper submitting cpu information available.
1197                          */
1198                         end_cmd(cmd);
1199                         break;
1200                 }
1201                 break;
1202         case NULL_IRQ_NONE:
1203                 end_cmd(cmd);
1204                 break;
1205         case NULL_IRQ_TIMER:
1206                 null_cmd_end_timer(cmd);
1207                 break;
1208         }
1209 }
1210
1211 static blk_status_t null_handle_cmd(struct nullb_cmd *cmd, sector_t sector,
1212                                     sector_t nr_sectors, enum req_opf op)
1213 {
1214         struct nullb_device *dev = cmd->nq->dev;
1215         struct nullb *nullb = dev->nullb;
1216         blk_status_t sts;
1217
1218         if (test_bit(NULLB_DEV_FL_THROTTLED, &dev->flags)) {
1219                 sts = null_handle_throttled(cmd);
1220                 if (sts != BLK_STS_OK)
1221                         return sts;
1222         }
1223
1224         if (op == REQ_OP_FLUSH) {
1225                 cmd->error = errno_to_blk_status(null_handle_flush(nullb));
1226                 goto out;
1227         }
1228
1229         if (nullb->dev->badblocks.shift != -1) {
1230                 cmd->error = null_handle_badblocks(cmd, sector, nr_sectors);
1231                 if (cmd->error != BLK_STS_OK)
1232                         goto out;
1233         }
1234
1235         if (dev->memory_backed)
1236                 cmd->error = null_handle_memory_backed(cmd, op);
1237
1238         if (!cmd->error && dev->zoned)
1239                 cmd->error = null_handle_zoned(cmd, op, sector, nr_sectors);
1240
1241 out:
1242         nullb_complete_cmd(cmd);
1243         return BLK_STS_OK;
1244 }
1245
1246 static enum hrtimer_restart nullb_bwtimer_fn(struct hrtimer *timer)
1247 {
1248         struct nullb *nullb = container_of(timer, struct nullb, bw_timer);
1249         ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1250         unsigned int mbps = nullb->dev->mbps;
1251
1252         if (atomic_long_read(&nullb->cur_bytes) == mb_per_tick(mbps))
1253                 return HRTIMER_NORESTART;
1254
1255         atomic_long_set(&nullb->cur_bytes, mb_per_tick(mbps));
1256         null_restart_queue_async(nullb);
1257
1258         hrtimer_forward_now(&nullb->bw_timer, timer_interval);
1259
1260         return HRTIMER_RESTART;
1261 }
1262
1263 static void nullb_setup_bwtimer(struct nullb *nullb)
1264 {
1265         ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1266
1267         hrtimer_init(&nullb->bw_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1268         nullb->bw_timer.function = nullb_bwtimer_fn;
1269         atomic_long_set(&nullb->cur_bytes, mb_per_tick(nullb->dev->mbps));
1270         hrtimer_start(&nullb->bw_timer, timer_interval, HRTIMER_MODE_REL);
1271 }
1272
1273 static struct nullb_queue *nullb_to_queue(struct nullb *nullb)
1274 {
1275         int index = 0;
1276
1277         if (nullb->nr_queues != 1)
1278                 index = raw_smp_processor_id() / ((nr_cpu_ids + nullb->nr_queues - 1) / nullb->nr_queues);
1279
1280         return &nullb->queues[index];
1281 }
1282
1283 static blk_qc_t null_queue_bio(struct request_queue *q, struct bio *bio)
1284 {
1285         sector_t sector = bio->bi_iter.bi_sector;
1286         sector_t nr_sectors = bio_sectors(bio);
1287         struct nullb *nullb = q->queuedata;
1288         struct nullb_queue *nq = nullb_to_queue(nullb);
1289         struct nullb_cmd *cmd;
1290
1291         cmd = alloc_cmd(nq, 1);
1292         cmd->bio = bio;
1293
1294         null_handle_cmd(cmd, sector, nr_sectors, bio_op(bio));
1295         return BLK_QC_T_NONE;
1296 }
1297
1298 static bool should_timeout_request(struct request *rq)
1299 {
1300 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1301         if (g_timeout_str[0])
1302                 return should_fail(&null_timeout_attr, 1);
1303 #endif
1304         return false;
1305 }
1306
1307 static bool should_requeue_request(struct request *rq)
1308 {
1309 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1310         if (g_requeue_str[0])
1311                 return should_fail(&null_requeue_attr, 1);
1312 #endif
1313         return false;
1314 }
1315
1316 static enum blk_eh_timer_return null_timeout_rq(struct request *rq, bool res)
1317 {
1318         pr_info("rq %p timed out\n", rq);
1319         blk_mq_complete_request(rq);
1320         return BLK_EH_DONE;
1321 }
1322
1323 static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx,
1324                          const struct blk_mq_queue_data *bd)
1325 {
1326         struct nullb_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
1327         struct nullb_queue *nq = hctx->driver_data;
1328         sector_t nr_sectors = blk_rq_sectors(bd->rq);
1329         sector_t sector = blk_rq_pos(bd->rq);
1330
1331         might_sleep_if(hctx->flags & BLK_MQ_F_BLOCKING);
1332
1333         if (nq->dev->irqmode == NULL_IRQ_TIMER) {
1334                 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1335                 cmd->timer.function = null_cmd_timer_expired;
1336         }
1337         cmd->rq = bd->rq;
1338         cmd->nq = nq;
1339
1340         blk_mq_start_request(bd->rq);
1341
1342         if (should_requeue_request(bd->rq)) {
1343                 /*
1344                  * Alternate between hitting the core BUSY path, and the
1345                  * driver driven requeue path
1346                  */
1347                 nq->requeue_selection++;
1348                 if (nq->requeue_selection & 1)
1349                         return BLK_STS_RESOURCE;
1350                 else {
1351                         blk_mq_requeue_request(bd->rq, true);
1352                         return BLK_STS_OK;
1353                 }
1354         }
1355         if (should_timeout_request(bd->rq))
1356                 return BLK_STS_OK;
1357
1358         return null_handle_cmd(cmd, sector, nr_sectors, req_op(bd->rq));
1359 }
1360
1361 static const struct blk_mq_ops null_mq_ops = {
1362         .queue_rq       = null_queue_rq,
1363         .complete       = null_complete_rq,
1364         .timeout        = null_timeout_rq,
1365 };
1366
1367 static void cleanup_queue(struct nullb_queue *nq)
1368 {
1369         kfree(nq->tag_map);
1370         kfree(nq->cmds);
1371 }
1372
1373 static void cleanup_queues(struct nullb *nullb)
1374 {
1375         int i;
1376
1377         for (i = 0; i < nullb->nr_queues; i++)
1378                 cleanup_queue(&nullb->queues[i]);
1379
1380         kfree(nullb->queues);
1381 }
1382
1383 static void null_del_dev(struct nullb *nullb)
1384 {
1385         struct nullb_device *dev = nullb->dev;
1386
1387         ida_simple_remove(&nullb_indexes, nullb->index);
1388
1389         list_del_init(&nullb->list);
1390
1391         del_gendisk(nullb->disk);
1392
1393         if (test_bit(NULLB_DEV_FL_THROTTLED, &nullb->dev->flags)) {
1394                 hrtimer_cancel(&nullb->bw_timer);
1395                 atomic_long_set(&nullb->cur_bytes, LONG_MAX);
1396                 null_restart_queue_async(nullb);
1397         }
1398
1399         blk_cleanup_queue(nullb->q);
1400         if (dev->queue_mode == NULL_Q_MQ &&
1401             nullb->tag_set == &nullb->__tag_set)
1402                 blk_mq_free_tag_set(nullb->tag_set);
1403         put_disk(nullb->disk);
1404         cleanup_queues(nullb);
1405         if (null_cache_active(nullb))
1406                 null_free_device_storage(nullb->dev, true);
1407         kfree(nullb);
1408         dev->nullb = NULL;
1409 }
1410
1411 static void null_config_discard(struct nullb *nullb)
1412 {
1413         if (nullb->dev->discard == false)
1414                 return;
1415         nullb->q->limits.discard_granularity = nullb->dev->blocksize;
1416         nullb->q->limits.discard_alignment = nullb->dev->blocksize;
1417         blk_queue_max_discard_sectors(nullb->q, UINT_MAX >> 9);
1418         blk_queue_flag_set(QUEUE_FLAG_DISCARD, nullb->q);
1419 }
1420
1421 static int null_open(struct block_device *bdev, fmode_t mode)
1422 {
1423         return 0;
1424 }
1425
1426 static void null_release(struct gendisk *disk, fmode_t mode)
1427 {
1428 }
1429
1430 static const struct block_device_operations null_fops = {
1431         .owner =        THIS_MODULE,
1432         .open =         null_open,
1433         .release =      null_release,
1434         .report_zones = null_zone_report,
1435 };
1436
1437 static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq)
1438 {
1439         BUG_ON(!nullb);
1440         BUG_ON(!nq);
1441
1442         init_waitqueue_head(&nq->wait);
1443         nq->queue_depth = nullb->queue_depth;
1444         nq->dev = nullb->dev;
1445 }
1446
1447 static void null_init_queues(struct nullb *nullb)
1448 {
1449         struct request_queue *q = nullb->q;
1450         struct blk_mq_hw_ctx *hctx;
1451         struct nullb_queue *nq;
1452         int i;
1453
1454         queue_for_each_hw_ctx(q, hctx, i) {
1455                 if (!hctx->nr_ctx || !hctx->tags)
1456                         continue;
1457                 nq = &nullb->queues[i];
1458                 hctx->driver_data = nq;
1459                 null_init_queue(nullb, nq);
1460                 nullb->nr_queues++;
1461         }
1462 }
1463
1464 static int setup_commands(struct nullb_queue *nq)
1465 {
1466         struct nullb_cmd *cmd;
1467         int i, tag_size;
1468
1469         nq->cmds = kcalloc(nq->queue_depth, sizeof(*cmd), GFP_KERNEL);
1470         if (!nq->cmds)
1471                 return -ENOMEM;
1472
1473         tag_size = ALIGN(nq->queue_depth, BITS_PER_LONG) / BITS_PER_LONG;
1474         nq->tag_map = kcalloc(tag_size, sizeof(unsigned long), GFP_KERNEL);
1475         if (!nq->tag_map) {
1476                 kfree(nq->cmds);
1477                 return -ENOMEM;
1478         }
1479
1480         for (i = 0; i < nq->queue_depth; i++) {
1481                 cmd = &nq->cmds[i];
1482                 INIT_LIST_HEAD(&cmd->list);
1483                 cmd->ll_list.next = NULL;
1484                 cmd->tag = -1U;
1485         }
1486
1487         return 0;
1488 }
1489
1490 static int setup_queues(struct nullb *nullb)
1491 {
1492         nullb->queues = kcalloc(nullb->dev->submit_queues,
1493                                 sizeof(struct nullb_queue),
1494                                 GFP_KERNEL);
1495         if (!nullb->queues)
1496                 return -ENOMEM;
1497
1498         nullb->queue_depth = nullb->dev->hw_queue_depth;
1499
1500         return 0;
1501 }
1502
1503 static int init_driver_queues(struct nullb *nullb)
1504 {
1505         struct nullb_queue *nq;
1506         int i, ret = 0;
1507
1508         for (i = 0; i < nullb->dev->submit_queues; i++) {
1509                 nq = &nullb->queues[i];
1510
1511                 null_init_queue(nullb, nq);
1512
1513                 ret = setup_commands(nq);
1514                 if (ret)
1515                         return ret;
1516                 nullb->nr_queues++;
1517         }
1518         return 0;
1519 }
1520
1521 static int null_gendisk_register(struct nullb *nullb)
1522 {
1523         struct gendisk *disk;
1524         sector_t size;
1525
1526         disk = nullb->disk = alloc_disk_node(1, nullb->dev->home_node);
1527         if (!disk)
1528                 return -ENOMEM;
1529         size = (sector_t)nullb->dev->size * 1024 * 1024ULL;
1530         set_capacity(disk, size >> 9);
1531
1532         disk->flags |= GENHD_FL_EXT_DEVT | GENHD_FL_SUPPRESS_PARTITION_INFO;
1533         disk->major             = null_major;
1534         disk->first_minor       = nullb->index;
1535         disk->fops              = &null_fops;
1536         disk->private_data      = nullb;
1537         disk->queue             = nullb->q;
1538         strncpy(disk->disk_name, nullb->disk_name, DISK_NAME_LEN);
1539
1540         if (nullb->dev->zoned) {
1541                 int ret = blk_revalidate_disk_zones(disk);
1542
1543                 if (ret != 0)
1544                         return ret;
1545         }
1546
1547         add_disk(disk);
1548         return 0;
1549 }
1550
1551 static int null_init_tag_set(struct nullb *nullb, struct blk_mq_tag_set *set)
1552 {
1553         set->ops = &null_mq_ops;
1554         set->nr_hw_queues = nullb ? nullb->dev->submit_queues :
1555                                                 g_submit_queues;
1556         set->queue_depth = nullb ? nullb->dev->hw_queue_depth :
1557                                                 g_hw_queue_depth;
1558         set->numa_node = nullb ? nullb->dev->home_node : g_home_node;
1559         set->cmd_size   = sizeof(struct nullb_cmd);
1560         set->flags = BLK_MQ_F_SHOULD_MERGE;
1561         if (g_no_sched)
1562                 set->flags |= BLK_MQ_F_NO_SCHED;
1563         set->driver_data = NULL;
1564
1565         if ((nullb && nullb->dev->blocking) || g_blocking)
1566                 set->flags |= BLK_MQ_F_BLOCKING;
1567
1568         return blk_mq_alloc_tag_set(set);
1569 }
1570
1571 static void null_validate_conf(struct nullb_device *dev)
1572 {
1573         dev->blocksize = round_down(dev->blocksize, 512);
1574         dev->blocksize = clamp_t(unsigned int, dev->blocksize, 512, 4096);
1575
1576         if (dev->queue_mode == NULL_Q_MQ && dev->use_per_node_hctx) {
1577                 if (dev->submit_queues != nr_online_nodes)
1578                         dev->submit_queues = nr_online_nodes;
1579         } else if (dev->submit_queues > nr_cpu_ids)
1580                 dev->submit_queues = nr_cpu_ids;
1581         else if (dev->submit_queues == 0)
1582                 dev->submit_queues = 1;
1583
1584         dev->queue_mode = min_t(unsigned int, dev->queue_mode, NULL_Q_MQ);
1585         dev->irqmode = min_t(unsigned int, dev->irqmode, NULL_IRQ_TIMER);
1586
1587         /* Do memory allocation, so set blocking */
1588         if (dev->memory_backed)
1589                 dev->blocking = true;
1590         else /* cache is meaningless */
1591                 dev->cache_size = 0;
1592         dev->cache_size = min_t(unsigned long, ULONG_MAX / 1024 / 1024,
1593                                                 dev->cache_size);
1594         dev->mbps = min_t(unsigned int, 1024 * 40, dev->mbps);
1595         /* can not stop a queue */
1596         if (dev->queue_mode == NULL_Q_BIO)
1597                 dev->mbps = 0;
1598 }
1599
1600 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1601 static bool __null_setup_fault(struct fault_attr *attr, char *str)
1602 {
1603         if (!str[0])
1604                 return true;
1605
1606         if (!setup_fault_attr(attr, str))
1607                 return false;
1608
1609         attr->verbose = 0;
1610         return true;
1611 }
1612 #endif
1613
1614 static bool null_setup_fault(void)
1615 {
1616 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1617         if (!__null_setup_fault(&null_timeout_attr, g_timeout_str))
1618                 return false;
1619         if (!__null_setup_fault(&null_requeue_attr, g_requeue_str))
1620                 return false;
1621 #endif
1622         return true;
1623 }
1624
1625 static int null_add_dev(struct nullb_device *dev)
1626 {
1627         struct nullb *nullb;
1628         int rv;
1629
1630         null_validate_conf(dev);
1631
1632         nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, dev->home_node);
1633         if (!nullb) {
1634                 rv = -ENOMEM;
1635                 goto out;
1636         }
1637         nullb->dev = dev;
1638         dev->nullb = nullb;
1639
1640         spin_lock_init(&nullb->lock);
1641
1642         rv = setup_queues(nullb);
1643         if (rv)
1644                 goto out_free_nullb;
1645
1646         if (dev->queue_mode == NULL_Q_MQ) {
1647                 if (shared_tags) {
1648                         nullb->tag_set = &tag_set;
1649                         rv = 0;
1650                 } else {
1651                         nullb->tag_set = &nullb->__tag_set;
1652                         rv = null_init_tag_set(nullb, nullb->tag_set);
1653                 }
1654
1655                 if (rv)
1656                         goto out_cleanup_queues;
1657
1658                 if (!null_setup_fault())
1659                         goto out_cleanup_queues;
1660
1661                 nullb->tag_set->timeout = 5 * HZ;
1662                 nullb->q = blk_mq_init_queue(nullb->tag_set);
1663                 if (IS_ERR(nullb->q)) {
1664                         rv = -ENOMEM;
1665                         goto out_cleanup_tags;
1666                 }
1667                 null_init_queues(nullb);
1668         } else if (dev->queue_mode == NULL_Q_BIO) {
1669                 nullb->q = blk_alloc_queue_node(GFP_KERNEL, dev->home_node);
1670                 if (!nullb->q) {
1671                         rv = -ENOMEM;
1672                         goto out_cleanup_queues;
1673                 }
1674                 blk_queue_make_request(nullb->q, null_queue_bio);
1675                 rv = init_driver_queues(nullb);
1676                 if (rv)
1677                         goto out_cleanup_blk_queue;
1678         }
1679
1680         if (dev->mbps) {
1681                 set_bit(NULLB_DEV_FL_THROTTLED, &dev->flags);
1682                 nullb_setup_bwtimer(nullb);
1683         }
1684
1685         if (dev->cache_size > 0) {
1686                 set_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
1687                 blk_queue_write_cache(nullb->q, true, true);
1688         }
1689
1690         if (dev->zoned) {
1691                 rv = null_zone_init(dev);
1692                 if (rv)
1693                         goto out_cleanup_blk_queue;
1694
1695                 blk_queue_chunk_sectors(nullb->q, dev->zone_size_sects);
1696                 nullb->q->limits.zoned = BLK_ZONED_HM;
1697                 blk_queue_flag_set(QUEUE_FLAG_ZONE_RESETALL, nullb->q);
1698                 blk_queue_required_elevator_features(nullb->q,
1699                                                 ELEVATOR_F_ZBD_SEQ_WRITE);
1700         }
1701
1702         nullb->q->queuedata = nullb;
1703         blk_queue_flag_set(QUEUE_FLAG_NONROT, nullb->q);
1704         blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, nullb->q);
1705
1706         mutex_lock(&lock);
1707         nullb->index = ida_simple_get(&nullb_indexes, 0, 0, GFP_KERNEL);
1708         dev->index = nullb->index;
1709         mutex_unlock(&lock);
1710
1711         blk_queue_logical_block_size(nullb->q, dev->blocksize);
1712         blk_queue_physical_block_size(nullb->q, dev->blocksize);
1713
1714         null_config_discard(nullb);
1715
1716         sprintf(nullb->disk_name, "nullb%d", nullb->index);
1717
1718         rv = null_gendisk_register(nullb);
1719         if (rv)
1720                 goto out_cleanup_zone;
1721
1722         mutex_lock(&lock);
1723         list_add_tail(&nullb->list, &nullb_list);
1724         mutex_unlock(&lock);
1725
1726         return 0;
1727 out_cleanup_zone:
1728         if (dev->zoned)
1729                 null_zone_exit(dev);
1730 out_cleanup_blk_queue:
1731         blk_cleanup_queue(nullb->q);
1732 out_cleanup_tags:
1733         if (dev->queue_mode == NULL_Q_MQ && nullb->tag_set == &nullb->__tag_set)
1734                 blk_mq_free_tag_set(nullb->tag_set);
1735 out_cleanup_queues:
1736         cleanup_queues(nullb);
1737 out_free_nullb:
1738         kfree(nullb);
1739 out:
1740         return rv;
1741 }
1742
1743 static int __init null_init(void)
1744 {
1745         int ret = 0;
1746         unsigned int i;
1747         struct nullb *nullb;
1748         struct nullb_device *dev;
1749
1750         if (g_bs > PAGE_SIZE) {
1751                 pr_warn("invalid block size\n");
1752                 pr_warn("defaults block size to %lu\n", PAGE_SIZE);
1753                 g_bs = PAGE_SIZE;
1754         }
1755
1756         if (!is_power_of_2(g_zone_size)) {
1757                 pr_err("zone_size must be power-of-two\n");
1758                 return -EINVAL;
1759         }
1760
1761         if (g_home_node != NUMA_NO_NODE && g_home_node >= nr_online_nodes) {
1762                 pr_err("invalid home_node value\n");
1763                 g_home_node = NUMA_NO_NODE;
1764         }
1765
1766         if (g_queue_mode == NULL_Q_RQ) {
1767                 pr_err("legacy IO path no longer available\n");
1768                 return -EINVAL;
1769         }
1770         if (g_queue_mode == NULL_Q_MQ && g_use_per_node_hctx) {
1771                 if (g_submit_queues != nr_online_nodes) {
1772                         pr_warn("submit_queues param is set to %u.\n",
1773                                                         nr_online_nodes);
1774                         g_submit_queues = nr_online_nodes;
1775                 }
1776         } else if (g_submit_queues > nr_cpu_ids)
1777                 g_submit_queues = nr_cpu_ids;
1778         else if (g_submit_queues <= 0)
1779                 g_submit_queues = 1;
1780
1781         if (g_queue_mode == NULL_Q_MQ && shared_tags) {
1782                 ret = null_init_tag_set(NULL, &tag_set);
1783                 if (ret)
1784                         return ret;
1785         }
1786
1787         config_group_init(&nullb_subsys.su_group);
1788         mutex_init(&nullb_subsys.su_mutex);
1789
1790         ret = configfs_register_subsystem(&nullb_subsys);
1791         if (ret)
1792                 goto err_tagset;
1793
1794         mutex_init(&lock);
1795
1796         null_major = register_blkdev(0, "nullb");
1797         if (null_major < 0) {
1798                 ret = null_major;
1799                 goto err_conf;
1800         }
1801
1802         for (i = 0; i < nr_devices; i++) {
1803                 dev = null_alloc_dev();
1804                 if (!dev) {
1805                         ret = -ENOMEM;
1806                         goto err_dev;
1807                 }
1808                 ret = null_add_dev(dev);
1809                 if (ret) {
1810                         null_free_dev(dev);
1811                         goto err_dev;
1812                 }
1813         }
1814
1815         pr_info("module loaded\n");
1816         return 0;
1817
1818 err_dev:
1819         while (!list_empty(&nullb_list)) {
1820                 nullb = list_entry(nullb_list.next, struct nullb, list);
1821                 dev = nullb->dev;
1822                 null_del_dev(nullb);
1823                 null_free_dev(dev);
1824         }
1825         unregister_blkdev(null_major, "nullb");
1826 err_conf:
1827         configfs_unregister_subsystem(&nullb_subsys);
1828 err_tagset:
1829         if (g_queue_mode == NULL_Q_MQ && shared_tags)
1830                 blk_mq_free_tag_set(&tag_set);
1831         return ret;
1832 }
1833
1834 static void __exit null_exit(void)
1835 {
1836         struct nullb *nullb;
1837
1838         configfs_unregister_subsystem(&nullb_subsys);
1839
1840         unregister_blkdev(null_major, "nullb");
1841
1842         mutex_lock(&lock);
1843         while (!list_empty(&nullb_list)) {
1844                 struct nullb_device *dev;
1845
1846                 nullb = list_entry(nullb_list.next, struct nullb, list);
1847                 dev = nullb->dev;
1848                 null_del_dev(nullb);
1849                 null_free_dev(dev);
1850         }
1851         mutex_unlock(&lock);
1852
1853         if (g_queue_mode == NULL_Q_MQ && shared_tags)
1854                 blk_mq_free_tag_set(&tag_set);
1855 }
1856
1857 module_init(null_init);
1858 module_exit(null_exit);
1859
1860 MODULE_AUTHOR("Jens Axboe <axboe@kernel.dk>");
1861 MODULE_LICENSE("GPL");