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