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 FREE_BATCH              16
  15
  16 #define TICKS_PER_SEC           50ULL
  17 #define TIMER_INTERVAL          (NSEC_PER_SEC / TICKS_PER_SEC)
  18
  19 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
  20 static DECLARE_FAULT_ATTR(null_timeout_attr);
  21 static DECLARE_FAULT_ATTR(null_requeue_attr);
  22 static DECLARE_FAULT_ATTR(null_init_hctx_attr);
  23 #endif
  24
  25 static inline u64 mb_per_tick(int mbps)
  26 {
  27         return (1 << 20) / TICKS_PER_SEC * ((u64) mbps);
  28 }
  29
  30 /*
  31  * Status flags for nullb_device.
  32  *
  33  * CONFIGURED:  Device has been configured and turned on. Cannot reconfigure.
  34  * UP:          Device is currently on and visible in userspace.
  35  * THROTTLED:   Device is being throttled.
  36  * CACHE:       Device is using a write-back cache.
  37  */
  38 enum nullb_device_flags {
  39         NULLB_DEV_FL_CONFIGURED = 0,
  40         NULLB_DEV_FL_UP         = 1,
  41         NULLB_DEV_FL_THROTTLED  = 2,
  42         NULLB_DEV_FL_CACHE      = 3,
  43 };
  44
  45 #define MAP_SZ          ((PAGE_SIZE >> SECTOR_SHIFT) + 2)
  46 /*
  47  * nullb_page is a page in memory for nullb devices.
  48  *
  49  * @page:       The page holding the data.
  50  * @bitmap:     The bitmap represents which sector in the page has data.
  51  *              Each bit represents one block size. For example, sector 8
  52  *              will use the 7th bit
  53  * The highest 2 bits of bitmap are for special purpose. LOCK means the cache
  54  * page is being flushing to storage. FREE means the cache page is freed and
  55  * should be skipped from flushing to storage. Please see
  56  * null_make_cache_space
  57  */
  58 struct nullb_page {
  59         struct page *page;
  60         DECLARE_BITMAP(bitmap, MAP_SZ);
  61 };
  62 #define NULLB_PAGE_LOCK (MAP_SZ - 1)
  63 #define NULLB_PAGE_FREE (MAP_SZ - 2)
  64
  65 static LIST_HEAD(nullb_list);
  66 static struct mutex lock;
  67 static int null_major;
  68 static DEFINE_IDA(nullb_indexes);
  69 static struct blk_mq_tag_set tag_set;
  70
  71 enum {
  72         NULL_IRQ_NONE           = 0,
  73         NULL_IRQ_SOFTIRQ        = 1,
  74         NULL_IRQ_TIMER          = 2,
  75 };
  76
  77 enum {
  78         NULL_Q_BIO              = 0,
  79         NULL_Q_RQ               = 1,
  80         NULL_Q_MQ               = 2,
  81 };
  82
  83 static bool g_virt_boundary = false;
  84 module_param_named(virt_boundary, g_virt_boundary, bool, 0444);
  85 MODULE_PARM_DESC(virt_boundary, "Require a virtual boundary for the device. Default: False");
  86
  87 static int g_no_sched;
  88 module_param_named(no_sched, g_no_sched, int, 0444);
  89 MODULE_PARM_DESC(no_sched, "No io scheduler");
  90
  91 static int g_submit_queues = 1;
  92 module_param_named(submit_queues, g_submit_queues, int, 0444);
  93 MODULE_PARM_DESC(submit_queues, "Number of submission queues");
  94
  95 static int g_home_node = NUMA_NO_NODE;
  96 module_param_named(home_node, g_home_node, int, 0444);
  97 MODULE_PARM_DESC(home_node, "Home node for the device");
  98
  99 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
 100 /*
 101  * For more details about fault injection, please refer to
 102  * Documentation/fault-injection/fault-injection.rst.
 103  */
 104 static char g_timeout_str[80];
 105 module_param_string(timeout, g_timeout_str, sizeof(g_timeout_str), 0444);
 106 MODULE_PARM_DESC(timeout, "Fault injection. timeout=<interval>,<probability>,<space>,<times>");
 107
 108 static char g_requeue_str[80];
 109 module_param_string(requeue, g_requeue_str, sizeof(g_requeue_str), 0444);
 110 MODULE_PARM_DESC(requeue, "Fault injection. requeue=<interval>,<probability>,<space>,<times>");
 111
 112 static char g_init_hctx_str[80];
 113 module_param_string(init_hctx, g_init_hctx_str, sizeof(g_init_hctx_str), 0444);
 114 MODULE_PARM_DESC(init_hctx, "Fault injection to fail hctx init. init_hctx=<interval>,<probability>,<space>,<times>");
 115 #endif
 116
 117 static int g_queue_mode = NULL_Q_MQ;
 118
 119 static int null_param_store_val(const char *str, int *val, int min, int max)
 120 {
 121         int ret, new_val;
 122
 123         ret = kstrtoint(str, 10, &new_val);
 124         if (ret)
 125                 return -EINVAL;
 126
 127         if (new_val < min || new_val > max)
 128                 return -EINVAL;
 129
 130         *val = new_val;
 131         return 0;
 132 }
 133
 134 static int null_set_queue_mode(const char *str, const struct kernel_param *kp)
 135 {
 136         return null_param_store_val(str, &g_queue_mode, NULL_Q_BIO, NULL_Q_MQ);
 137 }
 138
 139 static const struct kernel_param_ops null_queue_mode_param_ops = {
 140         .set    = null_set_queue_mode,
 141         .get    = param_get_int,
 142 };
 143
 144 device_param_cb(queue_mode, &null_queue_mode_param_ops, &g_queue_mode, 0444);
 145 MODULE_PARM_DESC(queue_mode, "Block interface to use (0=bio,1=rq,2=multiqueue)");
 146
 147 static int g_gb = 250;
 148 module_param_named(gb, g_gb, int, 0444);
 149 MODULE_PARM_DESC(gb, "Size in GB");
 150
 151 static int g_bs = 512;
 152 module_param_named(bs, g_bs, int, 0444);
 153 MODULE_PARM_DESC(bs, "Block size (in bytes)");
 154
 155 static int g_max_sectors;
 156 module_param_named(max_sectors, g_max_sectors, int, 0444);
 157 MODULE_PARM_DESC(max_sectors, "Maximum size of a command (in 512B sectors)");
 158
 159 static unsigned int nr_devices = 1;
 160 module_param(nr_devices, uint, 0444);
 161 MODULE_PARM_DESC(nr_devices, "Number of devices to register");
 162
 163 static bool g_blocking;
 164 module_param_named(blocking, g_blocking, bool, 0444);
 165 MODULE_PARM_DESC(blocking, "Register as a blocking blk-mq driver device");
 166
 167 static bool shared_tags;
 168 module_param(shared_tags, bool, 0444);
 169 MODULE_PARM_DESC(shared_tags, "Share tag set between devices for blk-mq");
 170
 171 static bool g_shared_tag_bitmap;
 172 module_param_named(shared_tag_bitmap, g_shared_tag_bitmap, bool, 0444);
 173 MODULE_PARM_DESC(shared_tag_bitmap, "Use shared tag bitmap for all submission queues for blk-mq");
 174
 175 static int g_irqmode = NULL_IRQ_SOFTIRQ;
 176
 177 static int null_set_irqmode(const char *str, const struct kernel_param *kp)
 178 {
 179         return null_param_store_val(str, &g_irqmode, NULL_IRQ_NONE,
 180                                         NULL_IRQ_TIMER);
 181 }
 182
 183 static const struct kernel_param_ops null_irqmode_param_ops = {
 184         .set    = null_set_irqmode,
 185         .get    = param_get_int,
 186 };
 187
 188 device_param_cb(irqmode, &null_irqmode_param_ops, &g_irqmode, 0444);
 189 MODULE_PARM_DESC(irqmode, "IRQ completion handler. 0-none, 1-softirq, 2-timer");
 190
 191 static unsigned long g_completion_nsec = 10000;
 192 module_param_named(completion_nsec, g_completion_nsec, ulong, 0444);
 193 MODULE_PARM_DESC(completion_nsec, "Time in ns to complete a request in hardware. Default: 10,000ns");
 194
 195 static int g_hw_queue_depth = 64;
 196 module_param_named(hw_queue_depth, g_hw_queue_depth, int, 0444);
 197 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 64");
 198
 199 static bool g_use_per_node_hctx;
 200 module_param_named(use_per_node_hctx, g_use_per_node_hctx, bool, 0444);
 201 MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware context queues. Default: false");
 202
 203 static bool g_zoned;
 204 module_param_named(zoned, g_zoned, bool, S_IRUGO);
 205 MODULE_PARM_DESC(zoned, "Make device as a host-managed zoned block device. Default: false");
 206
 207 static unsigned long g_zone_size = 256;
 208 module_param_named(zone_size, g_zone_size, ulong, S_IRUGO);
 209 MODULE_PARM_DESC(zone_size, "Zone size in MB when block device is zoned. Must be power-of-two: Default: 256");
 210
 211 static unsigned long g_zone_capacity;
 212 module_param_named(zone_capacity, g_zone_capacity, ulong, 0444);
 213 MODULE_PARM_DESC(zone_capacity, "Zone capacity in MB when block device is zoned. Can be less than or equal to zone size. Default: Zone size");
 214
 215 static unsigned int g_zone_nr_conv;
 216 module_param_named(zone_nr_conv, g_zone_nr_conv, uint, 0444);
 217 MODULE_PARM_DESC(zone_nr_conv, "Number of conventional zones when block device is zoned. Default: 0");
 218
 219 static unsigned int g_zone_max_open;
 220 module_param_named(zone_max_open, g_zone_max_open, uint, 0444);
 221 MODULE_PARM_DESC(zone_max_open, "Maximum number of open zones when block device is zoned. Default: 0 (no limit)");
 222
 223 static unsigned int g_zone_max_active;
 224 module_param_named(zone_max_active, g_zone_max_active, uint, 0444);
 225 MODULE_PARM_DESC(zone_max_active, "Maximum number of active zones when block device is zoned. Default: 0 (no limit)");
 226
 227 static struct nullb_device *null_alloc_dev(void);
 228 static void null_free_dev(struct nullb_device *dev);
 229 static void null_del_dev(struct nullb *nullb);
 230 static int null_add_dev(struct nullb_device *dev);
 231 static void null_free_device_storage(struct nullb_device *dev, bool is_cache);
 232
 233 static inline struct nullb_device *to_nullb_device(struct config_item *item)
 234 {
 235         return item ? container_of(item, struct nullb_device, item) : NULL;
 236 }
 237
 238 static inline ssize_t nullb_device_uint_attr_show(unsigned int val, char *page)
 239 {
 240         return snprintf(page, PAGE_SIZE, "%u\n", val);
 241 }
 242
 243 static inline ssize_t nullb_device_ulong_attr_show(unsigned long val,
 244         char *page)
 245 {
 246         return snprintf(page, PAGE_SIZE, "%lu\n", val);
 247 }
 248
 249 static inline ssize_t nullb_device_bool_attr_show(bool val, char *page)
 250 {
 251         return snprintf(page, PAGE_SIZE, "%u\n", val);
 252 }
 253
 254 static ssize_t nullb_device_uint_attr_store(unsigned int *val,
 255         const char *page, size_t count)
 256 {
 257         unsigned int tmp;
 258         int result;
 259
 260         result = kstrtouint(page, 0, &tmp);
 261         if (result < 0)
 262                 return result;
 263
 264         *val = tmp;
 265         return count;
 266 }
 267
 268 static ssize_t nullb_device_ulong_attr_store(unsigned long *val,
 269         const char *page, size_t count)
 270 {
 271         int result;
 272         unsigned long tmp;
 273
 274         result = kstrtoul(page, 0, &tmp);
 275         if (result < 0)
 276                 return result;
 277
 278         *val = tmp;
 279         return count;
 280 }
 281
 282 static ssize_t nullb_device_bool_attr_store(bool *val, const char *page,
 283         size_t count)
 284 {
 285         bool tmp;
 286         int result;
 287
 288         result = kstrtobool(page,  &tmp);
 289         if (result < 0)
 290                 return result;
 291
 292         *val = tmp;
 293         return count;
 294 }
 295
 296 /* The following macro should only be used with TYPE = {uint, ulong, bool}. */
 297 #define NULLB_DEVICE_ATTR(NAME, TYPE, APPLY)                            \
 298 static ssize_t                                                          \
 299 nullb_device_##NAME##_show(struct config_item *item, char *page)        \
 300 {                                                                       \
 301         return nullb_device_##TYPE##_attr_show(                         \
 302                                 to_nullb_device(item)->NAME, page);     \
 303 }                                                                       \
 304 static ssize_t                                                          \
 305 nullb_device_##NAME##_store(struct config_item *item, const char *page, \
 306                             size_t count)                               \
 307 {                                                                       \
 308         int (*apply_fn)(struct nullb_device *dev, TYPE new_value) = APPLY;\
 309         struct nullb_device *dev = to_nullb_device(item);               \
 310         TYPE new_value = 0;                                             \
 311         int ret;                                                        \
 312                                                                         \
 313         ret = nullb_device_##TYPE##_attr_store(&new_value, page, count);\
 314         if (ret < 0)                                                    \
 315                 return ret;                                             \
 316         if (apply_fn)                                                   \
 317                 ret = apply_fn(dev, new_value);                         \
 318         else if (test_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags))        \
 319                 ret = -EBUSY;                                           \
 320         if (ret < 0)                                                    \
 321                 return ret;                                             \
 322         dev->NAME = new_value;                                          \
 323         return count;                                                   \
 324 }                                                                       \
 325 CONFIGFS_ATTR(nullb_device_, NAME);
 326
 327 static int nullb_apply_submit_queues(struct nullb_device *dev,
 328                                      unsigned int submit_queues)
 329 {
 330         struct nullb *nullb = dev->nullb;
 331         struct blk_mq_tag_set *set;
 332
 333         if (!nullb)
 334                 return 0;
 335
 336         /*
 337          * Make sure that null_init_hctx() does not access nullb->queues[] past
 338          * the end of that array.
 339          */
 340         if (submit_queues > nr_cpu_ids)
 341                 return -EINVAL;
 342         set = nullb->tag_set;
 343         blk_mq_update_nr_hw_queues(set, submit_queues);
 344         return set->nr_hw_queues == submit_queues ? 0 : -ENOMEM;
 345 }
 346
 347 NULLB_DEVICE_ATTR(size, ulong, NULL);
 348 NULLB_DEVICE_ATTR(completion_nsec, ulong, NULL);
 349 NULLB_DEVICE_ATTR(submit_queues, uint, nullb_apply_submit_queues);
 350 NULLB_DEVICE_ATTR(home_node, uint, NULL);
 351 NULLB_DEVICE_ATTR(queue_mode, uint, NULL);
 352 NULLB_DEVICE_ATTR(blocksize, uint, NULL);
 353 NULLB_DEVICE_ATTR(max_sectors, uint, NULL);
 354 NULLB_DEVICE_ATTR(irqmode, uint, NULL);
 355 NULLB_DEVICE_ATTR(hw_queue_depth, uint, NULL);
 356 NULLB_DEVICE_ATTR(index, uint, NULL);
 357 NULLB_DEVICE_ATTR(blocking, bool, NULL);
 358 NULLB_DEVICE_ATTR(use_per_node_hctx, bool, NULL);
 359 NULLB_DEVICE_ATTR(memory_backed, bool, NULL);
 360 NULLB_DEVICE_ATTR(discard, bool, NULL);
 361 NULLB_DEVICE_ATTR(mbps, uint, NULL);
 362 NULLB_DEVICE_ATTR(cache_size, ulong, NULL);
 363 NULLB_DEVICE_ATTR(zoned, bool, NULL);
 364 NULLB_DEVICE_ATTR(zone_size, ulong, NULL);
 365 NULLB_DEVICE_ATTR(zone_capacity, ulong, NULL);
 366 NULLB_DEVICE_ATTR(zone_nr_conv, uint, NULL);
 367 NULLB_DEVICE_ATTR(zone_max_open, uint, NULL);
 368 NULLB_DEVICE_ATTR(zone_max_active, uint, NULL);
 369 NULLB_DEVICE_ATTR(virt_boundary, bool, NULL);
 370
 371 static ssize_t nullb_device_power_show(struct config_item *item, char *page)
 372 {
 373         return nullb_device_bool_attr_show(to_nullb_device(item)->power, page);
 374 }
 375
 376 static ssize_t nullb_device_power_store(struct config_item *item,
 377                                      const char *page, size_t count)
 378 {
 379         struct nullb_device *dev = to_nullb_device(item);
 380         bool newp = false;
 381         ssize_t ret;
 382
 383         ret = nullb_device_bool_attr_store(&newp, page, count);
 384         if (ret < 0)
 385                 return ret;
 386
 387         if (!dev->power && newp) {
 388                 if (test_and_set_bit(NULLB_DEV_FL_UP, &dev->flags))
 389                         return count;
 390                 if (null_add_dev(dev)) {
 391                         clear_bit(NULLB_DEV_FL_UP, &dev->flags);
 392                         return -ENOMEM;
 393                 }
 394
 395                 set_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
 396                 dev->power = newp;
 397         } else if (dev->power && !newp) {
 398                 if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
 399                         mutex_lock(&lock);
 400                         dev->power = newp;
 401                         null_del_dev(dev->nullb);
 402                         mutex_unlock(&lock);
 403                 }
 404                 clear_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
 405         }
 406
 407         return count;
 408 }
 409
 410 CONFIGFS_ATTR(nullb_device_, power);
 411
 412 static ssize_t nullb_device_badblocks_show(struct config_item *item, char *page)
 413 {
 414         struct nullb_device *t_dev = to_nullb_device(item);
 415
 416         return badblocks_show(&t_dev->badblocks, page, 0);
 417 }
 418
 419 static ssize_t nullb_device_badblocks_store(struct config_item *item,
 420                                      const char *page, size_t count)
 421 {
 422         struct nullb_device *t_dev = to_nullb_device(item);
 423         char *orig, *buf, *tmp;
 424         u64 start, end;
 425         int ret;
 426
 427         orig = kstrndup(page, count, GFP_KERNEL);
 428         if (!orig)
 429                 return -ENOMEM;
 430
 431         buf = strstrip(orig);
 432
 433         ret = -EINVAL;
 434         if (buf[0] != '+' && buf[0] != '-')
 435                 goto out;
 436         tmp = strchr(&buf[1], '-');
 437         if (!tmp)
 438                 goto out;
 439         *tmp = '\0';
 440         ret = kstrtoull(buf + 1, 0, &start);
 441         if (ret)
 442                 goto out;
 443         ret = kstrtoull(tmp + 1, 0, &end);
 444         if (ret)
 445                 goto out;
 446         ret = -EINVAL;
 447         if (start > end)
 448                 goto out;
 449         /* enable badblocks */
 450         cmpxchg(&t_dev->badblocks.shift, -1, 0);
 451         if (buf[0] == '+')
 452                 ret = badblocks_set(&t_dev->badblocks, start,
 453                         end - start + 1, 1);
 454         else
 455                 ret = badblocks_clear(&t_dev->badblocks, start,
 456                         end - start + 1);
 457         if (ret == 0)
 458                 ret = count;
 459 out:
 460         kfree(orig);
 461         return ret;
 462 }
 463 CONFIGFS_ATTR(nullb_device_, badblocks);
 464
 465 static struct configfs_attribute *nullb_device_attrs[] = {
 466         &nullb_device_attr_size,
 467         &nullb_device_attr_completion_nsec,
 468         &nullb_device_attr_submit_queues,
 469         &nullb_device_attr_home_node,
 470         &nullb_device_attr_queue_mode,
 471         &nullb_device_attr_blocksize,
 472         &nullb_device_attr_max_sectors,
 473         &nullb_device_attr_irqmode,
 474         &nullb_device_attr_hw_queue_depth,
 475         &nullb_device_attr_index,
 476         &nullb_device_attr_blocking,
 477         &nullb_device_attr_use_per_node_hctx,
 478         &nullb_device_attr_power,
 479         &nullb_device_attr_memory_backed,
 480         &nullb_device_attr_discard,
 481         &nullb_device_attr_mbps,
 482         &nullb_device_attr_cache_size,
 483         &nullb_device_attr_badblocks,
 484         &nullb_device_attr_zoned,
 485         &nullb_device_attr_zone_size,
 486         &nullb_device_attr_zone_capacity,
 487         &nullb_device_attr_zone_nr_conv,
 488         &nullb_device_attr_zone_max_open,
 489         &nullb_device_attr_zone_max_active,
 490         &nullb_device_attr_virt_boundary,
 491         NULL,
 492 };
 493
 494 static void nullb_device_release(struct config_item *item)
 495 {
 496         struct nullb_device *dev = to_nullb_device(item);
 497
 498         null_free_device_storage(dev, false);
 499         null_free_dev(dev);
 500 }
 501
 502 static struct configfs_item_operations nullb_device_ops = {
 503         .release        = nullb_device_release,
 504 };
 505
 506 static const struct config_item_type nullb_device_type = {
 507         .ct_item_ops    = &nullb_device_ops,
 508         .ct_attrs       = nullb_device_attrs,
 509         .ct_owner       = THIS_MODULE,
 510 };
 511
 512 static struct
 513 config_item *nullb_group_make_item(struct config_group *group, const char *name)
 514 {
 515         struct nullb_device *dev;
 516
 517         dev = null_alloc_dev();
 518         if (!dev)
 519                 return ERR_PTR(-ENOMEM);
 520
 521         config_item_init_type_name(&dev->item, name, &nullb_device_type);
 522
 523         return &dev->item;
 524 }
 525
 526 static void
 527 nullb_group_drop_item(struct config_group *group, struct config_item *item)
 528 {
 529         struct nullb_device *dev = to_nullb_device(item);
 530
 531         if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
 532                 mutex_lock(&lock);
 533                 dev->power = false;
 534                 null_del_dev(dev->nullb);
 535                 mutex_unlock(&lock);
 536         }
 537
 538         config_item_put(item);
 539 }
 540
 541 static ssize_t memb_group_features_show(struct config_item *item, char *page)
 542 {
 543         return snprintf(page, PAGE_SIZE,
 544                         "memory_backed,discard,bandwidth,cache,badblocks,zoned,zone_size,zone_capacity,zone_nr_conv,zone_max_open,zone_max_active,blocksize,max_sectors,virt_boundary\n");
 545 }
 546
 547 CONFIGFS_ATTR_RO(memb_group_, features);
 548
 549 static struct configfs_attribute *nullb_group_attrs[] = {
 550         &memb_group_attr_features,
 551         NULL,
 552 };
 553
 554 static struct configfs_group_operations nullb_group_ops = {
 555         .make_item      = nullb_group_make_item,
 556         .drop_item      = nullb_group_drop_item,
 557 };
 558
 559 static const struct config_item_type nullb_group_type = {
 560         .ct_group_ops   = &nullb_group_ops,
 561         .ct_attrs       = nullb_group_attrs,
 562         .ct_owner       = THIS_MODULE,
 563 };
 564
 565 static struct configfs_subsystem nullb_subsys = {
 566         .su_group = {
 567                 .cg_item = {
 568                         .ci_namebuf = "nullb",
 569                         .ci_type = &nullb_group_type,
 570                 },
 571         },
 572 };
 573
 574 static inline int null_cache_active(struct nullb *nullb)
 575 {
 576         return test_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
 577 }
 578
 579 static struct nullb_device *null_alloc_dev(void)
 580 {
 581         struct nullb_device *dev;
 582
 583         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
 584         if (!dev)
 585                 return NULL;
 586         INIT_RADIX_TREE(&dev->data, GFP_ATOMIC);
 587         INIT_RADIX_TREE(&dev->cache, GFP_ATOMIC);
 588         if (badblocks_init(&dev->badblocks, 0)) {
 589                 kfree(dev);
 590                 return NULL;
 591         }
 592
 593         dev->size = g_gb * 1024;
 594         dev->completion_nsec = g_completion_nsec;
 595         dev->submit_queues = g_submit_queues;
 596         dev->home_node = g_home_node;
 597         dev->queue_mode = g_queue_mode;
 598         dev->blocksize = g_bs;
 599         dev->max_sectors = g_max_sectors;
 600         dev->irqmode = g_irqmode;
 601         dev->hw_queue_depth = g_hw_queue_depth;
 602         dev->blocking = g_blocking;
 603         dev->use_per_node_hctx = g_use_per_node_hctx;
 604         dev->zoned = g_zoned;
 605         dev->zone_size = g_zone_size;
 606         dev->zone_capacity = g_zone_capacity;
 607         dev->zone_nr_conv = g_zone_nr_conv;
 608         dev->zone_max_open = g_zone_max_open;
 609         dev->zone_max_active = g_zone_max_active;
 610         dev->virt_boundary = g_virt_boundary;
 611         return dev;
 612 }
 613
 614 static void null_free_dev(struct nullb_device *dev)
 615 {
 616         if (!dev)
 617                 return;
 618
 619         null_free_zoned_dev(dev);
 620         badblocks_exit(&dev->badblocks);
 621         kfree(dev);
 622 }
 623
 624 static void put_tag(struct nullb_queue *nq, unsigned int tag)
 625 {
 626         clear_bit_unlock(tag, nq->tag_map);
 627
 628         if (waitqueue_active(&nq->wait))
 629                 wake_up(&nq->wait);
 630 }
 631
 632 static unsigned int get_tag(struct nullb_queue *nq)
 633 {
 634         unsigned int tag;
 635
 636         do {
 637                 tag = find_first_zero_bit(nq->tag_map, nq->queue_depth);
 638                 if (tag >= nq->queue_depth)
 639                         return -1U;
 640         } while (test_and_set_bit_lock(tag, nq->tag_map));
 641
 642         return tag;
 643 }
 644
 645 static void free_cmd(struct nullb_cmd *cmd)
 646 {
 647         put_tag(cmd->nq, cmd->tag);
 648 }
 649
 650 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer);
 651
 652 static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq)
 653 {
 654         struct nullb_cmd *cmd;
 655         unsigned int tag;
 656
 657         tag = get_tag(nq);
 658         if (tag != -1U) {
 659                 cmd = &nq->cmds[tag];
 660                 cmd->tag = tag;
 661                 cmd->error = BLK_STS_OK;
 662                 cmd->nq = nq;
 663                 if (nq->dev->irqmode == NULL_IRQ_TIMER) {
 664                         hrtimer_init(&cmd->timer, CLOCK_MONOTONIC,
 665                                      HRTIMER_MODE_REL);
 666                         cmd->timer.function = null_cmd_timer_expired;
 667                 }
 668                 return cmd;
 669         }
 670
 671         return NULL;
 672 }
 673
 674 static struct nullb_cmd *alloc_cmd(struct nullb_queue *nq, int can_wait)
 675 {
 676         struct nullb_cmd *cmd;
 677         DEFINE_WAIT(wait);
 678
 679         cmd = __alloc_cmd(nq);
 680         if (cmd || !can_wait)
 681                 return cmd;
 682
 683         do {
 684                 prepare_to_wait(&nq->wait, &wait, TASK_UNINTERRUPTIBLE);
 685                 cmd = __alloc_cmd(nq);
 686                 if (cmd)
 687                         break;
 688
 689                 io_schedule();
 690         } while (1);
 691
 692         finish_wait(&nq->wait, &wait);
 693         return cmd;
 694 }
 695
 696 static void end_cmd(struct nullb_cmd *cmd)
 697 {
 698         int queue_mode = cmd->nq->dev->queue_mode;
 699
 700         switch (queue_mode)  {
 701         case NULL_Q_MQ:
 702                 blk_mq_end_request(cmd->rq, cmd->error);
 703                 return;
 704         case NULL_Q_BIO:
 705                 cmd->bio->bi_status = cmd->error;
 706                 bio_endio(cmd->bio);
 707                 break;
 708         }
 709
 710         free_cmd(cmd);
 711 }
 712
 713 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
 714 {
 715         end_cmd(container_of(timer, struct nullb_cmd, timer));
 716
 717         return HRTIMER_NORESTART;
 718 }
 719
 720 static void null_cmd_end_timer(struct nullb_cmd *cmd)
 721 {
 722         ktime_t kt = cmd->nq->dev->completion_nsec;
 723
 724         hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL);
 725 }
 726
 727 static void null_complete_rq(struct request *rq)
 728 {
 729         end_cmd(blk_mq_rq_to_pdu(rq));
 730 }
 731
 732 static struct nullb_page *null_alloc_page(gfp_t gfp_flags)
 733 {
 734         struct nullb_page *t_page;
 735
 736         t_page = kmalloc(sizeof(struct nullb_page), gfp_flags);
 737         if (!t_page)
 738                 goto out;
 739
 740         t_page->page = alloc_pages(gfp_flags, 0);
 741         if (!t_page->page)
 742                 goto out_freepage;
 743
 744         memset(t_page->bitmap, 0, sizeof(t_page->bitmap));
 745         return t_page;
 746 out_freepage:
 747         kfree(t_page);
 748 out:
 749         return NULL;
 750 }
 751
 752 static void null_free_page(struct nullb_page *t_page)
 753 {
 754         __set_bit(NULLB_PAGE_FREE, t_page->bitmap);
 755         if (test_bit(NULLB_PAGE_LOCK, t_page->bitmap))
 756                 return;
 757         __free_page(t_page->page);
 758         kfree(t_page);
 759 }
 760
 761 static bool null_page_empty(struct nullb_page *page)
 762 {
 763         int size = MAP_SZ - 2;
 764
 765         return find_first_bit(page->bitmap, size) == size;
 766 }
 767
 768 static void null_free_sector(struct nullb *nullb, sector_t sector,
 769         bool is_cache)
 770 {
 771         unsigned int sector_bit;
 772         u64 idx;
 773         struct nullb_page *t_page, *ret;
 774         struct radix_tree_root *root;
 775
 776         root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
 777         idx = sector >> PAGE_SECTORS_SHIFT;
 778         sector_bit = (sector & SECTOR_MASK);
 779
 780         t_page = radix_tree_lookup(root, idx);
 781         if (t_page) {
 782                 __clear_bit(sector_bit, t_page->bitmap);
 783
 784                 if (null_page_empty(t_page)) {
 785                         ret = radix_tree_delete_item(root, idx, t_page);
 786                         WARN_ON(ret != t_page);
 787                         null_free_page(ret);
 788                         if (is_cache)
 789                                 nullb->dev->curr_cache -= PAGE_SIZE;
 790                 }
 791         }
 792 }
 793
 794 static struct nullb_page *null_radix_tree_insert(struct nullb *nullb, u64 idx,
 795         struct nullb_page *t_page, bool is_cache)
 796 {
 797         struct radix_tree_root *root;
 798
 799         root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
 800
 801         if (radix_tree_insert(root, idx, t_page)) {
 802                 null_free_page(t_page);
 803                 t_page = radix_tree_lookup(root, idx);
 804                 WARN_ON(!t_page || t_page->page->index != idx);
 805         } else if (is_cache)
 806                 nullb->dev->curr_cache += PAGE_SIZE;
 807
 808         return t_page;
 809 }
 810
 811 static void null_free_device_storage(struct nullb_device *dev, bool is_cache)
 812 {
 813         unsigned long pos = 0;
 814         int nr_pages;
 815         struct nullb_page *ret, *t_pages[FREE_BATCH];
 816         struct radix_tree_root *root;
 817
 818         root = is_cache ? &dev->cache : &dev->data;
 819
 820         do {
 821                 int i;
 822
 823                 nr_pages = radix_tree_gang_lookup(root,
 824                                 (void **)t_pages, pos, FREE_BATCH);
 825
 826                 for (i = 0; i < nr_pages; i++) {
 827                         pos = t_pages[i]->page->index;
 828                         ret = radix_tree_delete_item(root, pos, t_pages[i]);
 829                         WARN_ON(ret != t_pages[i]);
 830                         null_free_page(ret);
 831                 }
 832
 833                 pos++;
 834         } while (nr_pages == FREE_BATCH);
 835
 836         if (is_cache)
 837                 dev->curr_cache = 0;
 838 }
 839
 840 static struct nullb_page *__null_lookup_page(struct nullb *nullb,
 841         sector_t sector, bool for_write, bool is_cache)
 842 {
 843         unsigned int sector_bit;
 844         u64 idx;
 845         struct nullb_page *t_page;
 846         struct radix_tree_root *root;
 847
 848         idx = sector >> PAGE_SECTORS_SHIFT;
 849         sector_bit = (sector & SECTOR_MASK);
 850
 851         root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
 852         t_page = radix_tree_lookup(root, idx);
 853         WARN_ON(t_page && t_page->page->index != idx);
 854
 855         if (t_page && (for_write || test_bit(sector_bit, t_page->bitmap)))
 856                 return t_page;
 857
 858         return NULL;
 859 }
 860
 861 static struct nullb_page *null_lookup_page(struct nullb *nullb,
 862         sector_t sector, bool for_write, bool ignore_cache)
 863 {
 864         struct nullb_page *page = NULL;
 865
 866         if (!ignore_cache)
 867                 page = __null_lookup_page(nullb, sector, for_write, true);
 868         if (page)
 869                 return page;
 870         return __null_lookup_page(nullb, sector, for_write, false);
 871 }
 872
 873 static struct nullb_page *null_insert_page(struct nullb *nullb,
 874                                            sector_t sector, bool ignore_cache)
 875         __releases(&nullb->lock)
 876         __acquires(&nullb->lock)
 877 {
 878         u64 idx;
 879         struct nullb_page *t_page;
 880
 881         t_page = null_lookup_page(nullb, sector, true, ignore_cache);
 882         if (t_page)
 883                 return t_page;
 884
 885         spin_unlock_irq(&nullb->lock);
 886
 887         t_page = null_alloc_page(GFP_NOIO);
 888         if (!t_page)
 889                 goto out_lock;
 890
 891         if (radix_tree_preload(GFP_NOIO))
 892                 goto out_freepage;
 893
 894         spin_lock_irq(&nullb->lock);
 895         idx = sector >> PAGE_SECTORS_SHIFT;
 896         t_page->page->index = idx;
 897         t_page = null_radix_tree_insert(nullb, idx, t_page, !ignore_cache);
 898         radix_tree_preload_end();
 899
 900         return t_page;
 901 out_freepage:
 902         null_free_page(t_page);
 903 out_lock:
 904         spin_lock_irq(&nullb->lock);
 905         return null_lookup_page(nullb, sector, true, ignore_cache);
 906 }
 907
 908 static int null_flush_cache_page(struct nullb *nullb, struct nullb_page *c_page)
 909 {
 910         int i;
 911         unsigned int offset;
 912         u64 idx;
 913         struct nullb_page *t_page, *ret;
 914         void *dst, *src;
 915
 916         idx = c_page->page->index;
 917
 918         t_page = null_insert_page(nullb, idx << PAGE_SECTORS_SHIFT, true);
 919
 920         __clear_bit(NULLB_PAGE_LOCK, c_page->bitmap);
 921         if (test_bit(NULLB_PAGE_FREE, c_page->bitmap)) {
 922                 null_free_page(c_page);
 923                 if (t_page && null_page_empty(t_page)) {
 924                         ret = radix_tree_delete_item(&nullb->dev->data,
 925                                 idx, t_page);
 926                         null_free_page(t_page);
 927                 }
 928                 return 0;
 929         }
 930
 931         if (!t_page)
 932                 return -ENOMEM;
 933
 934         src = kmap_atomic(c_page->page);
 935         dst = kmap_atomic(t_page->page);
 936
 937         for (i = 0; i < PAGE_SECTORS;
 938                         i += (nullb->dev->blocksize >> SECTOR_SHIFT)) {
 939                 if (test_bit(i, c_page->bitmap)) {
 940                         offset = (i << SECTOR_SHIFT);
 941                         memcpy(dst + offset, src + offset,
 942                                 nullb->dev->blocksize);
 943                         __set_bit(i, t_page->bitmap);
 944                 }
 945         }
 946
 947         kunmap_atomic(dst);
 948         kunmap_atomic(src);
 949
 950         ret = radix_tree_delete_item(&nullb->dev->cache, idx, c_page);
 951         null_free_page(ret);
 952         nullb->dev->curr_cache -= PAGE_SIZE;
 953
 954         return 0;
 955 }
 956
 957 static int null_make_cache_space(struct nullb *nullb, unsigned long n)
 958 {
 959         int i, err, nr_pages;
 960         struct nullb_page *c_pages[FREE_BATCH];
 961         unsigned long flushed = 0, one_round;
 962
 963 again:
 964         if ((nullb->dev->cache_size * 1024 * 1024) >
 965              nullb->dev->curr_cache + n || nullb->dev->curr_cache == 0)
 966                 return 0;
 967
 968         nr_pages = radix_tree_gang_lookup(&nullb->dev->cache,
 969                         (void **)c_pages, nullb->cache_flush_pos, FREE_BATCH);
 970         /*
 971          * nullb_flush_cache_page could unlock before using the c_pages. To
 972          * avoid race, we don't allow page free
 973          */
 974         for (i = 0; i < nr_pages; i++) {
 975                 nullb->cache_flush_pos = c_pages[i]->page->index;
 976                 /*
 977                  * We found the page which is being flushed to disk by other
 978                  * threads
 979                  */
 980                 if (test_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap))
 981                         c_pages[i] = NULL;
 982                 else
 983                         __set_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap);
 984         }
 985
 986         one_round = 0;
 987         for (i = 0; i < nr_pages; i++) {
 988                 if (c_pages[i] == NULL)
 989                         continue;
 990                 err = null_flush_cache_page(nullb, c_pages[i]);
 991                 if (err)
 992                         return err;
 993                 one_round++;
 994         }
 995         flushed += one_round << PAGE_SHIFT;
 996
 997         if (n > flushed) {
 998                 if (nr_pages == 0)
 999                         nullb->cache_flush_pos = 0;
1000                 if (one_round == 0) {
1001                         /* give other threads a chance */
1002                         spin_unlock_irq(&nullb->lock);
1003                         spin_lock_irq(&nullb->lock);
1004                 }
1005                 goto again;
1006         }
1007         return 0;
1008 }
1009
1010 static int copy_to_nullb(struct nullb *nullb, struct page *source,
1011         unsigned int off, sector_t sector, size_t n, bool is_fua)
1012 {
1013         size_t temp, count = 0;
1014         unsigned int offset;
1015         struct nullb_page *t_page;
1016         void *dst, *src;
1017
1018         while (count < n) {
1019                 temp = min_t(size_t, nullb->dev->blocksize, n - count);
1020
1021                 if (null_cache_active(nullb) && !is_fua)
1022                         null_make_cache_space(nullb, PAGE_SIZE);
1023
1024                 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
1025                 t_page = null_insert_page(nullb, sector,
1026                         !null_cache_active(nullb) || is_fua);
1027                 if (!t_page)
1028                         return -ENOSPC;
1029
1030                 src = kmap_atomic(source);
1031                 dst = kmap_atomic(t_page->page);
1032                 memcpy(dst + offset, src + off + count, temp);
1033                 kunmap_atomic(dst);
1034                 kunmap_atomic(src);
1035
1036                 __set_bit(sector & SECTOR_MASK, t_page->bitmap);
1037
1038                 if (is_fua)
1039                         null_free_sector(nullb, sector, true);
1040
1041                 count += temp;
1042                 sector += temp >> SECTOR_SHIFT;
1043         }
1044         return 0;
1045 }
1046
1047 static int copy_from_nullb(struct nullb *nullb, struct page *dest,
1048         unsigned int off, sector_t sector, size_t n)
1049 {
1050         size_t temp, count = 0;
1051         unsigned int offset;
1052         struct nullb_page *t_page;
1053         void *dst, *src;
1054
1055         while (count < n) {
1056                 temp = min_t(size_t, nullb->dev->blocksize, n - count);
1057
1058                 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
1059                 t_page = null_lookup_page(nullb, sector, false,
1060                         !null_cache_active(nullb));
1061
1062                 dst = kmap_atomic(dest);
1063                 if (!t_page) {
1064                         memset(dst + off + count, 0, temp);
1065                         goto next;
1066                 }
1067                 src = kmap_atomic(t_page->page);
1068                 memcpy(dst + off + count, src + offset, temp);
1069                 kunmap_atomic(src);
1070 next:
1071                 kunmap_atomic(dst);
1072
1073                 count += temp;
1074                 sector += temp >> SECTOR_SHIFT;
1075         }
1076         return 0;
1077 }
1078
1079 static void nullb_fill_pattern(struct nullb *nullb, struct page *page,
1080                                unsigned int len, unsigned int off)
1081 {
1082         void *dst;
1083
1084         dst = kmap_atomic(page);
1085         memset(dst + off, 0xFF, len);
1086         kunmap_atomic(dst);
1087 }
1088
1089 blk_status_t null_handle_discard(struct nullb_device *dev,
1090                                  sector_t sector, sector_t nr_sectors)
1091 {
1092         struct nullb *nullb = dev->nullb;
1093         size_t n = nr_sectors << SECTOR_SHIFT;
1094         size_t temp;
1095
1096         spin_lock_irq(&nullb->lock);
1097         while (n > 0) {
1098                 temp = min_t(size_t, n, dev->blocksize);
1099                 null_free_sector(nullb, sector, false);
1100                 if (null_cache_active(nullb))
1101                         null_free_sector(nullb, sector, true);
1102                 sector += temp >> SECTOR_SHIFT;
1103                 n -= temp;
1104         }
1105         spin_unlock_irq(&nullb->lock);
1106
1107         return BLK_STS_OK;
1108 }
1109
1110 static int null_handle_flush(struct nullb *nullb)
1111 {
1112         int err;
1113
1114         if (!null_cache_active(nullb))
1115                 return 0;
1116
1117         spin_lock_irq(&nullb->lock);
1118         while (true) {
1119                 err = null_make_cache_space(nullb,
1120                         nullb->dev->cache_size * 1024 * 1024);
1121                 if (err || nullb->dev->curr_cache == 0)
1122                         break;
1123         }
1124
1125         WARN_ON(!radix_tree_empty(&nullb->dev->cache));
1126         spin_unlock_irq(&nullb->lock);
1127         return err;
1128 }
1129
1130 static int null_transfer(struct nullb *nullb, struct page *page,
1131         unsigned int len, unsigned int off, bool is_write, sector_t sector,
1132         bool is_fua)
1133 {
1134         struct nullb_device *dev = nullb->dev;
1135         unsigned int valid_len = len;
1136         int err = 0;
1137
1138         if (!is_write) {
1139                 if (dev->zoned)
1140                         valid_len = null_zone_valid_read_len(nullb,
1141                                 sector, len);
1142
1143                 if (valid_len) {
1144                         err = copy_from_nullb(nullb, page, off,
1145                                 sector, valid_len);
1146                         off += valid_len;
1147                         len -= valid_len;
1148                 }
1149
1150                 if (len)
1151                         nullb_fill_pattern(nullb, page, len, off);
1152                 flush_dcache_page(page);
1153         } else {
1154                 flush_dcache_page(page);
1155                 err = copy_to_nullb(nullb, page, off, sector, len, is_fua);
1156         }
1157
1158         return err;
1159 }
1160
1161 static int null_handle_rq(struct nullb_cmd *cmd)
1162 {
1163         struct request *rq = cmd->rq;
1164         struct nullb *nullb = cmd->nq->dev->nullb;
1165         int err;
1166         unsigned int len;
1167         sector_t sector = blk_rq_pos(rq);
1168         struct req_iterator iter;
1169         struct bio_vec bvec;
1170
1171         spin_lock_irq(&nullb->lock);
1172         rq_for_each_segment(bvec, rq, iter) {
1173                 len = bvec.bv_len;
1174                 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1175                                      op_is_write(req_op(rq)), sector,
1176                                      rq->cmd_flags & REQ_FUA);
1177                 if (err) {
1178                         spin_unlock_irq(&nullb->lock);
1179                         return err;
1180                 }
1181                 sector += len >> SECTOR_SHIFT;
1182         }
1183         spin_unlock_irq(&nullb->lock);
1184
1185         return 0;
1186 }
1187
1188 static int null_handle_bio(struct nullb_cmd *cmd)
1189 {
1190         struct bio *bio = cmd->bio;
1191         struct nullb *nullb = cmd->nq->dev->nullb;
1192         int err;
1193         unsigned int len;
1194         sector_t sector = bio->bi_iter.bi_sector;
1195         struct bio_vec bvec;
1196         struct bvec_iter iter;
1197
1198         spin_lock_irq(&nullb->lock);
1199         bio_for_each_segment(bvec, bio, iter) {
1200                 len = bvec.bv_len;
1201                 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1202                                      op_is_write(bio_op(bio)), sector,
1203                                      bio->bi_opf & REQ_FUA);
1204                 if (err) {
1205                         spin_unlock_irq(&nullb->lock);
1206                         return err;
1207                 }
1208                 sector += len >> SECTOR_SHIFT;
1209         }
1210         spin_unlock_irq(&nullb->lock);
1211         return 0;
1212 }
1213
1214 static void null_stop_queue(struct nullb *nullb)
1215 {
1216         struct request_queue *q = nullb->q;
1217
1218         if (nullb->dev->queue_mode == NULL_Q_MQ)
1219                 blk_mq_stop_hw_queues(q);
1220 }
1221
1222 static void null_restart_queue_async(struct nullb *nullb)
1223 {
1224         struct request_queue *q = nullb->q;
1225
1226         if (nullb->dev->queue_mode == NULL_Q_MQ)
1227                 blk_mq_start_stopped_hw_queues(q, true);
1228 }
1229
1230 static inline blk_status_t null_handle_throttled(struct nullb_cmd *cmd)
1231 {
1232         struct nullb_device *dev = cmd->nq->dev;
1233         struct nullb *nullb = dev->nullb;
1234         blk_status_t sts = BLK_STS_OK;
1235         struct request *rq = cmd->rq;
1236
1237         if (!hrtimer_active(&nullb->bw_timer))
1238                 hrtimer_restart(&nullb->bw_timer);
1239
1240         if (atomic_long_sub_return(blk_rq_bytes(rq), &nullb->cur_bytes) < 0) {
1241                 null_stop_queue(nullb);
1242                 /* race with timer */
1243                 if (atomic_long_read(&nullb->cur_bytes) > 0)
1244                         null_restart_queue_async(nullb);
1245                 /* requeue request */
1246                 sts = BLK_STS_DEV_RESOURCE;
1247         }
1248         return sts;
1249 }
1250
1251 static inline blk_status_t null_handle_badblocks(struct nullb_cmd *cmd,
1252                                                  sector_t sector,
1253                                                  sector_t nr_sectors)
1254 {
1255         struct badblocks *bb = &cmd->nq->dev->badblocks;
1256         sector_t first_bad;
1257         int bad_sectors;
1258
1259         if (badblocks_check(bb, sector, nr_sectors, &first_bad, &bad_sectors))
1260                 return BLK_STS_IOERR;
1261
1262         return BLK_STS_OK;
1263 }
1264
1265 static inline blk_status_t null_handle_memory_backed(struct nullb_cmd *cmd,
1266                                                      enum req_opf op,
1267                                                      sector_t sector,
1268                                                      sector_t nr_sectors)
1269 {
1270         struct nullb_device *dev = cmd->nq->dev;
1271         int err;
1272
1273         if (op == REQ_OP_DISCARD)
1274                 return null_handle_discard(dev, sector, nr_sectors);
1275
1276         if (dev->queue_mode == NULL_Q_BIO)
1277                 err = null_handle_bio(cmd);
1278         else
1279                 err = null_handle_rq(cmd);
1280
1281         return errno_to_blk_status(err);
1282 }
1283
1284 static void nullb_zero_read_cmd_buffer(struct nullb_cmd *cmd)
1285 {
1286         struct nullb_device *dev = cmd->nq->dev;
1287         struct bio *bio;
1288
1289         if (dev->memory_backed)
1290                 return;
1291
1292         if (dev->queue_mode == NULL_Q_BIO && bio_op(cmd->bio) == REQ_OP_READ) {
1293                 zero_fill_bio(cmd->bio);
1294         } else if (req_op(cmd->rq) == REQ_OP_READ) {
1295                 __rq_for_each_bio(bio, cmd->rq)
1296                         zero_fill_bio(bio);
1297         }
1298 }
1299
1300 static inline void nullb_complete_cmd(struct nullb_cmd *cmd)
1301 {
1302         /*
1303          * Since root privileges are required to configure the null_blk
1304          * driver, it is fine that this driver does not initialize the
1305          * data buffers of read commands. Zero-initialize these buffers
1306          * anyway if KMSAN is enabled to prevent that KMSAN complains
1307          * about null_blk not initializing read data buffers.
1308          */
1309         if (IS_ENABLED(CONFIG_KMSAN))
1310                 nullb_zero_read_cmd_buffer(cmd);
1311
1312         /* Complete IO by inline, softirq or timer */
1313         switch (cmd->nq->dev->irqmode) {
1314         case NULL_IRQ_SOFTIRQ:
1315                 switch (cmd->nq->dev->queue_mode) {
1316                 case NULL_Q_MQ:
1317                         if (likely(!blk_should_fake_timeout(cmd->rq->q)))
1318                                 blk_mq_complete_request(cmd->rq);
1319                         break;
1320                 case NULL_Q_BIO:
1321                         /*
1322                          * XXX: no proper submitting cpu information available.
1323                          */
1324                         end_cmd(cmd);
1325                         break;
1326                 }
1327                 break;
1328         case NULL_IRQ_NONE:
1329                 end_cmd(cmd);
1330                 break;
1331         case NULL_IRQ_TIMER:
1332                 null_cmd_end_timer(cmd);
1333                 break;
1334         }
1335 }
1336
1337 blk_status_t null_process_cmd(struct nullb_cmd *cmd,
1338                               enum req_opf op, sector_t sector,
1339                               unsigned int nr_sectors)
1340 {
1341         struct nullb_device *dev = cmd->nq->dev;
1342         blk_status_t ret;
1343
1344         if (dev->badblocks.shift != -1) {
1345                 ret = null_handle_badblocks(cmd, sector, nr_sectors);
1346                 if (ret != BLK_STS_OK)
1347                         return ret;
1348         }
1349
1350         if (dev->memory_backed)
1351                 return null_handle_memory_backed(cmd, op, sector, nr_sectors);
1352
1353         return BLK_STS_OK;
1354 }
1355
1356 static blk_status_t null_handle_cmd(struct nullb_cmd *cmd, sector_t sector,
1357                                     sector_t nr_sectors, enum req_opf op)
1358 {
1359         struct nullb_device *dev = cmd->nq->dev;
1360         struct nullb *nullb = dev->nullb;
1361         blk_status_t sts;
1362
1363         if (test_bit(NULLB_DEV_FL_THROTTLED, &dev->flags)) {
1364                 sts = null_handle_throttled(cmd);
1365                 if (sts != BLK_STS_OK)
1366                         return sts;
1367         }
1368
1369         if (op == REQ_OP_FLUSH) {
1370                 cmd->error = errno_to_blk_status(null_handle_flush(nullb));
1371                 goto out;
1372         }
1373
1374         if (dev->zoned)
1375                 sts = null_process_zoned_cmd(cmd, op, sector, nr_sectors);
1376         else
1377                 sts = null_process_cmd(cmd, op, sector, nr_sectors);
1378
1379         /* Do not overwrite errors (e.g. timeout errors) */
1380         if (cmd->error == BLK_STS_OK)
1381                 cmd->error = sts;
1382
1383 out:
1384         nullb_complete_cmd(cmd);
1385         return BLK_STS_OK;
1386 }
1387
1388 static enum hrtimer_restart nullb_bwtimer_fn(struct hrtimer *timer)
1389 {
1390         struct nullb *nullb = container_of(timer, struct nullb, bw_timer);
1391         ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1392         unsigned int mbps = nullb->dev->mbps;
1393
1394         if (atomic_long_read(&nullb->cur_bytes) == mb_per_tick(mbps))
1395                 return HRTIMER_NORESTART;
1396
1397         atomic_long_set(&nullb->cur_bytes, mb_per_tick(mbps));
1398         null_restart_queue_async(nullb);
1399
1400         hrtimer_forward_now(&nullb->bw_timer, timer_interval);
1401
1402         return HRTIMER_RESTART;
1403 }
1404
1405 static void nullb_setup_bwtimer(struct nullb *nullb)
1406 {
1407         ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1408
1409         hrtimer_init(&nullb->bw_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1410         nullb->bw_timer.function = nullb_bwtimer_fn;
1411         atomic_long_set(&nullb->cur_bytes, mb_per_tick(nullb->dev->mbps));
1412         hrtimer_start(&nullb->bw_timer, timer_interval, HRTIMER_MODE_REL);
1413 }
1414
1415 static struct nullb_queue *nullb_to_queue(struct nullb *nullb)
1416 {
1417         int index = 0;
1418
1419         if (nullb->nr_queues != 1)
1420                 index = raw_smp_processor_id() / ((nr_cpu_ids + nullb->nr_queues - 1) / nullb->nr_queues);
1421
1422         return &nullb->queues[index];
1423 }
1424
1425 static blk_qc_t null_submit_bio(struct bio *bio)
1426 {
1427         sector_t sector = bio->bi_iter.bi_sector;
1428         sector_t nr_sectors = bio_sectors(bio);
1429         struct nullb *nullb = bio->bi_bdev->bd_disk->private_data;
1430         struct nullb_queue *nq = nullb_to_queue(nullb);
1431         struct nullb_cmd *cmd;
1432
1433         cmd = alloc_cmd(nq, 1);
1434         cmd->bio = bio;
1435
1436         null_handle_cmd(cmd, sector, nr_sectors, bio_op(bio));
1437         return BLK_QC_T_NONE;
1438 }
1439
1440 static bool should_timeout_request(struct request *rq)
1441 {
1442 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1443         if (g_timeout_str[0])
1444                 return should_fail(&null_timeout_attr, 1);
1445 #endif
1446         return false;
1447 }
1448
1449 static bool should_requeue_request(struct request *rq)
1450 {
1451 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1452         if (g_requeue_str[0])
1453                 return should_fail(&null_requeue_attr, 1);
1454 #endif
1455         return false;
1456 }
1457
1458 static enum blk_eh_timer_return null_timeout_rq(struct request *rq, bool res)
1459 {
1460         struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq);
1461
1462         pr_info("rq %p timed out\n", rq);
1463
1464         /*
1465          * If the device is marked as blocking (i.e. memory backed or zoned
1466          * device), the submission path may be blocked waiting for resources
1467          * and cause real timeouts. For these real timeouts, the submission
1468          * path will complete the request using blk_mq_complete_request().
1469          * Only fake timeouts need to execute blk_mq_complete_request() here.
1470          */
1471         cmd->error = BLK_STS_TIMEOUT;
1472         if (cmd->fake_timeout)
1473                 blk_mq_complete_request(rq);
1474         return BLK_EH_DONE;
1475 }
1476
1477 static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx,
1478                          const struct blk_mq_queue_data *bd)
1479 {
1480         struct nullb_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
1481         struct nullb_queue *nq = hctx->driver_data;
1482         sector_t nr_sectors = blk_rq_sectors(bd->rq);
1483         sector_t sector = blk_rq_pos(bd->rq);
1484
1485         might_sleep_if(hctx->flags & BLK_MQ_F_BLOCKING);
1486
1487         if (nq->dev->irqmode == NULL_IRQ_TIMER) {
1488                 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1489                 cmd->timer.function = null_cmd_timer_expired;
1490         }
1491         cmd->rq = bd->rq;
1492         cmd->error = BLK_STS_OK;
1493         cmd->nq = nq;
1494         cmd->fake_timeout = should_timeout_request(bd->rq);
1495
1496         blk_mq_start_request(bd->rq);
1497
1498         if (should_requeue_request(bd->rq)) {
1499                 /*
1500                  * Alternate between hitting the core BUSY path, and the
1501                  * driver driven requeue path
1502                  */
1503                 nq->requeue_selection++;
1504                 if (nq->requeue_selection & 1)
1505                         return BLK_STS_RESOURCE;
1506                 else {
1507                         blk_mq_requeue_request(bd->rq, true);
1508                         return BLK_STS_OK;
1509                 }
1510         }
1511         if (cmd->fake_timeout)
1512                 return BLK_STS_OK;
1513
1514         return null_handle_cmd(cmd, sector, nr_sectors, req_op(bd->rq));
1515 }
1516
1517 static void cleanup_queue(struct nullb_queue *nq)
1518 {
1519         kfree(nq->tag_map);
1520         kfree(nq->cmds);
1521 }
1522
1523 static void cleanup_queues(struct nullb *nullb)
1524 {
1525         int i;
1526
1527         for (i = 0; i < nullb->nr_queues; i++)
1528                 cleanup_queue(&nullb->queues[i]);
1529
1530         kfree(nullb->queues);
1531 }
1532
1533 static void null_exit_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx)
1534 {
1535         struct nullb_queue *nq = hctx->driver_data;
1536         struct nullb *nullb = nq->dev->nullb;
1537
1538         nullb->nr_queues--;
1539 }
1540
1541 static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq)
1542 {
1543         init_waitqueue_head(&nq->wait);
1544         nq->queue_depth = nullb->queue_depth;
1545         nq->dev = nullb->dev;
1546 }
1547
1548 static int null_init_hctx(struct blk_mq_hw_ctx *hctx, void *driver_data,
1549                           unsigned int hctx_idx)
1550 {
1551         struct nullb *nullb = hctx->queue->queuedata;
1552         struct nullb_queue *nq;
1553
1554 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1555         if (g_init_hctx_str[0] && should_fail(&null_init_hctx_attr, 1))
1556                 return -EFAULT;
1557 #endif
1558
1559         nq = &nullb->queues[hctx_idx];
1560         hctx->driver_data = nq;
1561         null_init_queue(nullb, nq);
1562         nullb->nr_queues++;
1563
1564         return 0;
1565 }
1566
1567 static const struct blk_mq_ops null_mq_ops = {
1568         .queue_rq       = null_queue_rq,
1569         .complete       = null_complete_rq,
1570         .timeout        = null_timeout_rq,
1571         .init_hctx      = null_init_hctx,
1572         .exit_hctx      = null_exit_hctx,
1573 };
1574
1575 static void null_del_dev(struct nullb *nullb)
1576 {
1577         struct nullb_device *dev;
1578
1579         if (!nullb)
1580                 return;
1581
1582         dev = nullb->dev;
1583
1584         ida_simple_remove(&nullb_indexes, nullb->index);
1585
1586         list_del_init(&nullb->list);
1587
1588         del_gendisk(nullb->disk);
1589
1590         if (test_bit(NULLB_DEV_FL_THROTTLED, &nullb->dev->flags)) {
1591                 hrtimer_cancel(&nullb->bw_timer);
1592                 atomic_long_set(&nullb->cur_bytes, LONG_MAX);
1593                 null_restart_queue_async(nullb);
1594         }
1595
1596         blk_cleanup_disk(nullb->disk);
1597         if (dev->queue_mode == NULL_Q_MQ &&
1598             nullb->tag_set == &nullb->__tag_set)
1599                 blk_mq_free_tag_set(nullb->tag_set);
1600         cleanup_queues(nullb);
1601         if (null_cache_active(nullb))
1602                 null_free_device_storage(nullb->dev, true);
1603         kfree(nullb);
1604         dev->nullb = NULL;
1605 }
1606
1607 static void null_config_discard(struct nullb *nullb)
1608 {
1609         if (nullb->dev->discard == false)
1610                 return;
1611
1612         if (!nullb->dev->memory_backed) {
1613                 nullb->dev->discard = false;
1614                 pr_info("discard option is ignored without memory backing\n");
1615                 return;
1616         }
1617
1618         if (nullb->dev->zoned) {
1619                 nullb->dev->discard = false;
1620                 pr_info("discard option is ignored in zoned mode\n");
1621                 return;
1622         }
1623
1624         nullb->q->limits.discard_granularity = nullb->dev->blocksize;
1625         nullb->q->limits.discard_alignment = nullb->dev->blocksize;
1626         blk_queue_max_discard_sectors(nullb->q, UINT_MAX >> 9);
1627         blk_queue_flag_set(QUEUE_FLAG_DISCARD, nullb->q);
1628 }
1629
1630 static const struct block_device_operations null_bio_ops = {
1631         .owner          = THIS_MODULE,
1632         .submit_bio     = null_submit_bio,
1633         .report_zones   = null_report_zones,
1634 };
1635
1636 static const struct block_device_operations null_rq_ops = {
1637         .owner          = THIS_MODULE,
1638         .report_zones   = null_report_zones,
1639 };
1640
1641 static int setup_commands(struct nullb_queue *nq)
1642 {
1643         struct nullb_cmd *cmd;
1644         int i, tag_size;
1645
1646         nq->cmds = kcalloc(nq->queue_depth, sizeof(*cmd), GFP_KERNEL);
1647         if (!nq->cmds)
1648                 return -ENOMEM;
1649
1650         tag_size = ALIGN(nq->queue_depth, BITS_PER_LONG) / BITS_PER_LONG;
1651         nq->tag_map = kcalloc(tag_size, sizeof(unsigned long), GFP_KERNEL);
1652         if (!nq->tag_map) {
1653                 kfree(nq->cmds);
1654                 return -ENOMEM;
1655         }
1656
1657         for (i = 0; i < nq->queue_depth; i++) {
1658                 cmd = &nq->cmds[i];
1659                 cmd->tag = -1U;
1660         }
1661
1662         return 0;
1663 }
1664
1665 static int setup_queues(struct nullb *nullb)
1666 {
1667         nullb->queues = kcalloc(nr_cpu_ids, sizeof(struct nullb_queue),
1668                                 GFP_KERNEL);
1669         if (!nullb->queues)
1670                 return -ENOMEM;
1671
1672         nullb->queue_depth = nullb->dev->hw_queue_depth;
1673
1674         return 0;
1675 }
1676
1677 static int init_driver_queues(struct nullb *nullb)
1678 {
1679         struct nullb_queue *nq;
1680         int i, ret = 0;
1681
1682         for (i = 0; i < nullb->dev->submit_queues; i++) {
1683                 nq = &nullb->queues[i];
1684
1685                 null_init_queue(nullb, nq);
1686
1687                 ret = setup_commands(nq);
1688                 if (ret)
1689                         return ret;
1690                 nullb->nr_queues++;
1691         }
1692         return 0;
1693 }
1694
1695 static int null_gendisk_register(struct nullb *nullb)
1696 {
1697         sector_t size = ((sector_t)nullb->dev->size * SZ_1M) >> SECTOR_SHIFT;
1698         struct gendisk *disk = nullb->disk;
1699
1700         set_capacity(disk, size);
1701
1702         disk->flags |= GENHD_FL_EXT_DEVT | GENHD_FL_SUPPRESS_PARTITION_INFO;
1703         disk->major             = null_major;
1704         disk->first_minor       = nullb->index;
1705         disk->minors            = 1;
1706         if (queue_is_mq(nullb->q))
1707                 disk->fops              = &null_rq_ops;
1708         else
1709                 disk->fops              = &null_bio_ops;
1710         disk->private_data      = nullb;
1711         strncpy(disk->disk_name, nullb->disk_name, DISK_NAME_LEN);
1712
1713         if (nullb->dev->zoned) {
1714                 int ret = null_register_zoned_dev(nullb);
1715
1716                 if (ret)
1717                         return ret;
1718         }
1719
1720         return add_disk(disk);
1721 }
1722
1723 static int null_init_tag_set(struct nullb *nullb, struct blk_mq_tag_set *set)
1724 {
1725         set->ops = &null_mq_ops;
1726         set->nr_hw_queues = nullb ? nullb->dev->submit_queues :
1727                                                 g_submit_queues;
1728         set->queue_depth = nullb ? nullb->dev->hw_queue_depth :
1729                                                 g_hw_queue_depth;
1730         set->numa_node = nullb ? nullb->dev->home_node : g_home_node;
1731         set->cmd_size   = sizeof(struct nullb_cmd);
1732         set->flags = BLK_MQ_F_SHOULD_MERGE;
1733         if (g_no_sched)
1734                 set->flags |= BLK_MQ_F_NO_SCHED;
1735         if (g_shared_tag_bitmap)
1736                 set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1737         set->driver_data = NULL;
1738
1739         if ((nullb && nullb->dev->blocking) || g_blocking)
1740                 set->flags |= BLK_MQ_F_BLOCKING;
1741
1742         return blk_mq_alloc_tag_set(set);
1743 }
1744
1745 static int null_validate_conf(struct nullb_device *dev)
1746 {
1747         dev->blocksize = round_down(dev->blocksize, 512);
1748         dev->blocksize = clamp_t(unsigned int, dev->blocksize, 512, 4096);
1749
1750         if (dev->queue_mode == NULL_Q_MQ && dev->use_per_node_hctx) {
1751                 if (dev->submit_queues != nr_online_nodes)
1752                         dev->submit_queues = nr_online_nodes;
1753         } else if (dev->submit_queues > nr_cpu_ids)
1754                 dev->submit_queues = nr_cpu_ids;
1755         else if (dev->submit_queues == 0)
1756                 dev->submit_queues = 1;
1757
1758         dev->queue_mode = min_t(unsigned int, dev->queue_mode, NULL_Q_MQ);
1759         dev->irqmode = min_t(unsigned int, dev->irqmode, NULL_IRQ_TIMER);
1760
1761         /* Do memory allocation, so set blocking */
1762         if (dev->memory_backed)
1763                 dev->blocking = true;
1764         else /* cache is meaningless */
1765                 dev->cache_size = 0;
1766         dev->cache_size = min_t(unsigned long, ULONG_MAX / 1024 / 1024,
1767                                                 dev->cache_size);
1768         dev->mbps = min_t(unsigned int, 1024 * 40, dev->mbps);
1769         /* can not stop a queue */
1770         if (dev->queue_mode == NULL_Q_BIO)
1771                 dev->mbps = 0;
1772
1773         if (dev->zoned &&
1774             (!dev->zone_size || !is_power_of_2(dev->zone_size))) {
1775                 pr_err("zone_size must be power-of-two\n");
1776                 return -EINVAL;
1777         }
1778
1779         return 0;
1780 }
1781
1782 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1783 static bool __null_setup_fault(struct fault_attr *attr, char *str)
1784 {
1785         if (!str[0])
1786                 return true;
1787
1788         if (!setup_fault_attr(attr, str))
1789                 return false;
1790
1791         attr->verbose = 0;
1792         return true;
1793 }
1794 #endif
1795
1796 static bool null_setup_fault(void)
1797 {
1798 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1799         if (!__null_setup_fault(&null_timeout_attr, g_timeout_str))
1800                 return false;
1801         if (!__null_setup_fault(&null_requeue_attr, g_requeue_str))
1802                 return false;
1803         if (!__null_setup_fault(&null_init_hctx_attr, g_init_hctx_str))
1804                 return false;
1805 #endif
1806         return true;
1807 }
1808
1809 static int null_add_dev(struct nullb_device *dev)
1810 {
1811         struct nullb *nullb;
1812         int rv;
1813
1814         rv = null_validate_conf(dev);
1815         if (rv)
1816                 return rv;
1817
1818         nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, dev->home_node);
1819         if (!nullb) {
1820                 rv = -ENOMEM;
1821                 goto out;
1822         }
1823         nullb->dev = dev;
1824         dev->nullb = nullb;
1825
1826         spin_lock_init(&nullb->lock);
1827
1828         rv = setup_queues(nullb);
1829         if (rv)
1830                 goto out_free_nullb;
1831
1832         if (dev->queue_mode == NULL_Q_MQ) {
1833                 if (shared_tags) {
1834                         nullb->tag_set = &tag_set;
1835                         rv = 0;
1836                 } else {
1837                         nullb->tag_set = &nullb->__tag_set;
1838                         rv = null_init_tag_set(nullb, nullb->tag_set);
1839                 }
1840
1841                 if (rv)
1842                         goto out_cleanup_queues;
1843
1844                 if (!null_setup_fault())
1845                         goto out_cleanup_tags;
1846
1847                 nullb->tag_set->timeout = 5 * HZ;
1848                 nullb->disk = blk_mq_alloc_disk(nullb->tag_set, nullb);
1849                 if (IS_ERR(nullb->disk)) {
1850                         rv = PTR_ERR(nullb->disk);
1851                         goto out_cleanup_tags;
1852                 }
1853                 nullb->q = nullb->disk->queue;
1854         } else if (dev->queue_mode == NULL_Q_BIO) {
1855                 rv = -ENOMEM;
1856                 nullb->disk = blk_alloc_disk(nullb->dev->home_node);
1857                 if (!nullb->disk)
1858                         goto out_cleanup_queues;
1859
1860                 nullb->q = nullb->disk->queue;
1861                 rv = init_driver_queues(nullb);
1862                 if (rv)
1863                         goto out_cleanup_disk;
1864         }
1865
1866         if (dev->mbps) {
1867                 set_bit(NULLB_DEV_FL_THROTTLED, &dev->flags);
1868                 nullb_setup_bwtimer(nullb);
1869         }
1870
1871         if (dev->cache_size > 0) {
1872                 set_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
1873                 blk_queue_write_cache(nullb->q, true, true);
1874         }
1875
1876         if (dev->zoned) {
1877                 rv = null_init_zoned_dev(dev, nullb->q);
1878                 if (rv)
1879                         goto out_cleanup_disk;
1880         }
1881
1882         nullb->q->queuedata = nullb;
1883         blk_queue_flag_set(QUEUE_FLAG_NONROT, nullb->q);
1884         blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, nullb->q);
1885
1886         mutex_lock(&lock);
1887         nullb->index = ida_simple_get(&nullb_indexes, 0, 0, GFP_KERNEL);
1888         dev->index = nullb->index;
1889         mutex_unlock(&lock);
1890
1891         blk_queue_logical_block_size(nullb->q, dev->blocksize);
1892         blk_queue_physical_block_size(nullb->q, dev->blocksize);
1893         if (!dev->max_sectors)
1894                 dev->max_sectors = queue_max_hw_sectors(nullb->q);
1895         dev->max_sectors = min_t(unsigned int, dev->max_sectors,
1896                                  BLK_DEF_MAX_SECTORS);
1897         blk_queue_max_hw_sectors(nullb->q, dev->max_sectors);
1898
1899         if (dev->virt_boundary)
1900                 blk_queue_virt_boundary(nullb->q, PAGE_SIZE - 1);
1901
1902         null_config_discard(nullb);
1903
1904         sprintf(nullb->disk_name, "nullb%d", nullb->index);
1905
1906         rv = null_gendisk_register(nullb);
1907         if (rv)
1908                 goto out_cleanup_zone;
1909
1910         mutex_lock(&lock);
1911         list_add_tail(&nullb->list, &nullb_list);
1912         mutex_unlock(&lock);
1913
1914         return 0;
1915 out_cleanup_zone:
1916         null_free_zoned_dev(dev);
1917 out_cleanup_disk:
1918         blk_cleanup_disk(nullb->disk);
1919 out_cleanup_tags:
1920         if (dev->queue_mode == NULL_Q_MQ && nullb->tag_set == &nullb->__tag_set)
1921                 blk_mq_free_tag_set(nullb->tag_set);
1922 out_cleanup_queues:
1923         cleanup_queues(nullb);
1924 out_free_nullb:
1925         kfree(nullb);
1926         dev->nullb = NULL;
1927 out:
1928         return rv;
1929 }
1930
1931 static int __init null_init(void)
1932 {
1933         int ret = 0;
1934         unsigned int i;
1935         struct nullb *nullb;
1936         struct nullb_device *dev;
1937
1938         if (g_bs > PAGE_SIZE) {
1939                 pr_warn("invalid block size\n");
1940                 pr_warn("defaults block size to %lu\n", PAGE_SIZE);
1941                 g_bs = PAGE_SIZE;
1942         }
1943
1944         if (g_max_sectors > BLK_DEF_MAX_SECTORS) {
1945                 pr_warn("invalid max sectors\n");
1946                 pr_warn("defaults max sectors to %u\n", BLK_DEF_MAX_SECTORS);
1947                 g_max_sectors = BLK_DEF_MAX_SECTORS;
1948         }
1949
1950         if (g_home_node != NUMA_NO_NODE && g_home_node >= nr_online_nodes) {
1951                 pr_err("invalid home_node value\n");
1952                 g_home_node = NUMA_NO_NODE;
1953         }
1954
1955         if (g_queue_mode == NULL_Q_RQ) {
1956                 pr_err("legacy IO path no longer available\n");
1957                 return -EINVAL;
1958         }
1959         if (g_queue_mode == NULL_Q_MQ && g_use_per_node_hctx) {
1960                 if (g_submit_queues != nr_online_nodes) {
1961                         pr_warn("submit_queues param is set to %u.\n",
1962                                                         nr_online_nodes);
1963                         g_submit_queues = nr_online_nodes;
1964                 }
1965         } else if (g_submit_queues > nr_cpu_ids)
1966                 g_submit_queues = nr_cpu_ids;
1967         else if (g_submit_queues <= 0)
1968                 g_submit_queues = 1;
1969
1970         if (g_queue_mode == NULL_Q_MQ && shared_tags) {
1971                 ret = null_init_tag_set(NULL, &tag_set);
1972                 if (ret)
1973                         return ret;
1974         }
1975
1976         config_group_init(&nullb_subsys.su_group);
1977         mutex_init(&nullb_subsys.su_mutex);
1978
1979         ret = configfs_register_subsystem(&nullb_subsys);
1980         if (ret)
1981                 goto err_tagset;
1982
1983         mutex_init(&lock);
1984
1985         null_major = register_blkdev(0, "nullb");
1986         if (null_major < 0) {
1987                 ret = null_major;
1988                 goto err_conf;
1989         }
1990
1991         for (i = 0; i < nr_devices; i++) {
1992                 dev = null_alloc_dev();
1993                 if (!dev) {
1994                         ret = -ENOMEM;
1995                         goto err_dev;
1996                 }
1997                 ret = null_add_dev(dev);
1998                 if (ret) {
1999                         null_free_dev(dev);
2000                         goto err_dev;
2001                 }
2002         }
2003
2004         pr_info("module loaded\n");
2005         return 0;
2006
2007 err_dev:
2008         while (!list_empty(&nullb_list)) {
2009                 nullb = list_entry(nullb_list.next, struct nullb, list);
2010                 dev = nullb->dev;
2011                 null_del_dev(nullb);
2012                 null_free_dev(dev);
2013         }
2014         unregister_blkdev(null_major, "nullb");
2015 err_conf:
2016         configfs_unregister_subsystem(&nullb_subsys);
2017 err_tagset:
2018         if (g_queue_mode == NULL_Q_MQ && shared_tags)
2019                 blk_mq_free_tag_set(&tag_set);
2020         return ret;
2021 }
2022
2023 static void __exit null_exit(void)
2024 {
2025         struct nullb *nullb;
2026
2027         configfs_unregister_subsystem(&nullb_subsys);
2028
2029         unregister_blkdev(null_major, "nullb");
2030
2031         mutex_lock(&lock);
2032         while (!list_empty(&nullb_list)) {
2033                 struct nullb_device *dev;
2034
2035                 nullb = list_entry(nullb_list.next, struct nullb, list);
2036                 dev = nullb->dev;
2037                 null_del_dev(nullb);
2038                 null_free_dev(dev);
2039         }
2040         mutex_unlock(&lock);
2041
2042         if (g_queue_mode == NULL_Q_MQ && shared_tags)
2043                 blk_mq_free_tag_set(&tag_set);
2044 }
2045
2046 module_init(null_init);
2047 module_exit(null_exit);
2048
2049 MODULE_AUTHOR("Jens Axboe <axboe@kernel.dk>");
2050 MODULE_LICENSE("GPL");