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 static DECLARE_FAULT_ATTR(null_init_hctx_attr);
  27 #endif
  28
  29 static inline u64 mb_per_tick(int mbps)
  30 {
  31         return (1 << 20) / TICKS_PER_SEC * ((u64) mbps);
  32 }
  33
  34 /*
  35  * Status flags for nullb_device.
  36  *
  37  * CONFIGURED:  Device has been configured and turned on. Cannot reconfigure.
  38  * UP:          Device is currently on and visible in userspace.
  39  * THROTTLED:   Device is being throttled.
  40  * CACHE:       Device is using a write-back cache.
  41  */
  42 enum nullb_device_flags {
  43         NULLB_DEV_FL_CONFIGURED = 0,
  44         NULLB_DEV_FL_UP         = 1,
  45         NULLB_DEV_FL_THROTTLED  = 2,
  46         NULLB_DEV_FL_CACHE      = 3,
  47 };
  48
  49 #define MAP_SZ          ((PAGE_SIZE >> SECTOR_SHIFT) + 2)
  50 /*
  51  * nullb_page is a page in memory for nullb devices.
  52  *
  53  * @page:       The page holding the data.
  54  * @bitmap:     The bitmap represents which sector in the page has data.
  55  *              Each bit represents one block size. For example, sector 8
  56  *              will use the 7th bit
  57  * The highest 2 bits of bitmap are for special purpose. LOCK means the cache
  58  * page is being flushing to storage. FREE means the cache page is freed and
  59  * should be skipped from flushing to storage. Please see
  60  * null_make_cache_space
  61  */
  62 struct nullb_page {
  63         struct page *page;
  64         DECLARE_BITMAP(bitmap, MAP_SZ);
  65 };
  66 #define NULLB_PAGE_LOCK (MAP_SZ - 1)
  67 #define NULLB_PAGE_FREE (MAP_SZ - 2)
  68
  69 static LIST_HEAD(nullb_list);
  70 static struct mutex lock;
  71 static int null_major;
  72 static DEFINE_IDA(nullb_indexes);
  73 static struct blk_mq_tag_set tag_set;
  74
  75 enum {
  76         NULL_IRQ_NONE           = 0,
  77         NULL_IRQ_SOFTIRQ        = 1,
  78         NULL_IRQ_TIMER          = 2,
  79 };
  80
  81 enum {
  82         NULL_Q_BIO              = 0,
  83         NULL_Q_RQ               = 1,
  84         NULL_Q_MQ               = 2,
  85 };
  86
  87 static bool g_virt_boundary = false;
  88 module_param_named(virt_boundary, g_virt_boundary, bool, 0444);
  89 MODULE_PARM_DESC(virt_boundary, "Require a virtual boundary for the device. Default: False");
  90
  91 static int g_no_sched;
  92 module_param_named(no_sched, g_no_sched, int, 0444);
  93 MODULE_PARM_DESC(no_sched, "No io scheduler");
  94
  95 static int g_submit_queues = 1;
  96 module_param_named(submit_queues, g_submit_queues, int, 0444);
  97 MODULE_PARM_DESC(submit_queues, "Number of submission queues");
  98
  99 static int g_home_node = NUMA_NO_NODE;
 100 module_param_named(home_node, g_home_node, int, 0444);
 101 MODULE_PARM_DESC(home_node, "Home node for the device");
 102
 103 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
 104 /*
 105  * For more details about fault injection, please refer to
 106  * Documentation/fault-injection/fault-injection.rst.
 107  */
 108 static char g_timeout_str[80];
 109 module_param_string(timeout, g_timeout_str, sizeof(g_timeout_str), 0444);
 110 MODULE_PARM_DESC(timeout, "Fault injection. timeout=<interval>,<probability>,<space>,<times>");
 111
 112 static char g_requeue_str[80];
 113 module_param_string(requeue, g_requeue_str, sizeof(g_requeue_str), 0444);
 114 MODULE_PARM_DESC(requeue, "Fault injection. requeue=<interval>,<probability>,<space>,<times>");
 115
 116 static char g_init_hctx_str[80];
 117 module_param_string(init_hctx, g_init_hctx_str, sizeof(g_init_hctx_str), 0444);
 118 MODULE_PARM_DESC(init_hctx, "Fault injection to fail hctx init. init_hctx=<interval>,<probability>,<space>,<times>");
 119 #endif
 120
 121 static int g_queue_mode = NULL_Q_MQ;
 122
 123 static int null_param_store_val(const char *str, int *val, int min, int max)
 124 {
 125         int ret, new_val;
 126
 127         ret = kstrtoint(str, 10, &new_val);
 128         if (ret)
 129                 return -EINVAL;
 130
 131         if (new_val < min || new_val > max)
 132                 return -EINVAL;
 133
 134         *val = new_val;
 135         return 0;
 136 }
 137
 138 static int null_set_queue_mode(const char *str, const struct kernel_param *kp)
 139 {
 140         return null_param_store_val(str, &g_queue_mode, NULL_Q_BIO, NULL_Q_MQ);
 141 }
 142
 143 static const struct kernel_param_ops null_queue_mode_param_ops = {
 144         .set    = null_set_queue_mode,
 145         .get    = param_get_int,
 146 };
 147
 148 device_param_cb(queue_mode, &null_queue_mode_param_ops, &g_queue_mode, 0444);
 149 MODULE_PARM_DESC(queue_mode, "Block interface to use (0=bio,1=rq,2=multiqueue)");
 150
 151 static int g_gb = 250;
 152 module_param_named(gb, g_gb, int, 0444);
 153 MODULE_PARM_DESC(gb, "Size in GB");
 154
 155 static int g_bs = 512;
 156 module_param_named(bs, g_bs, int, 0444);
 157 MODULE_PARM_DESC(bs, "Block size (in bytes)");
 158
 159 static int g_max_sectors;
 160 module_param_named(max_sectors, g_max_sectors, int, 0444);
 161 MODULE_PARM_DESC(max_sectors, "Maximum size of a command (in 512B sectors)");
 162
 163 static unsigned int nr_devices = 1;
 164 module_param(nr_devices, uint, 0444);
 165 MODULE_PARM_DESC(nr_devices, "Number of devices to register");
 166
 167 static bool g_blocking;
 168 module_param_named(blocking, g_blocking, bool, 0444);
 169 MODULE_PARM_DESC(blocking, "Register as a blocking blk-mq driver device");
 170
 171 static bool shared_tags;
 172 module_param(shared_tags, bool, 0444);
 173 MODULE_PARM_DESC(shared_tags, "Share tag set between devices for blk-mq");
 174
 175 static bool g_shared_tag_bitmap;
 176 module_param_named(shared_tag_bitmap, g_shared_tag_bitmap, bool, 0444);
 177 MODULE_PARM_DESC(shared_tag_bitmap, "Use shared tag bitmap for all submission queues for blk-mq");
 178
 179 static int g_irqmode = NULL_IRQ_SOFTIRQ;
 180
 181 static int null_set_irqmode(const char *str, const struct kernel_param *kp)
 182 {
 183         return null_param_store_val(str, &g_irqmode, NULL_IRQ_NONE,
 184                                         NULL_IRQ_TIMER);
 185 }
 186
 187 static const struct kernel_param_ops null_irqmode_param_ops = {
 188         .set    = null_set_irqmode,
 189         .get    = param_get_int,
 190 };
 191
 192 device_param_cb(irqmode, &null_irqmode_param_ops, &g_irqmode, 0444);
 193 MODULE_PARM_DESC(irqmode, "IRQ completion handler. 0-none, 1-softirq, 2-timer");
 194
 195 static unsigned long g_completion_nsec = 10000;
 196 module_param_named(completion_nsec, g_completion_nsec, ulong, 0444);
 197 MODULE_PARM_DESC(completion_nsec, "Time in ns to complete a request in hardware. Default: 10,000ns");
 198
 199 static int g_hw_queue_depth = 64;
 200 module_param_named(hw_queue_depth, g_hw_queue_depth, int, 0444);
 201 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 64");
 202
 203 static bool g_use_per_node_hctx;
 204 module_param_named(use_per_node_hctx, g_use_per_node_hctx, bool, 0444);
 205 MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware context queues. Default: false");
 206
 207 static bool g_zoned;
 208 module_param_named(zoned, g_zoned, bool, S_IRUGO);
 209 MODULE_PARM_DESC(zoned, "Make device as a host-managed zoned block device. Default: false");
 210
 211 static unsigned long g_zone_size = 256;
 212 module_param_named(zone_size, g_zone_size, ulong, S_IRUGO);
 213 MODULE_PARM_DESC(zone_size, "Zone size in MB when block device is zoned. Must be power-of-two: Default: 256");
 214
 215 static unsigned long g_zone_capacity;
 216 module_param_named(zone_capacity, g_zone_capacity, ulong, 0444);
 217 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");
 218
 219 static unsigned int g_zone_nr_conv;
 220 module_param_named(zone_nr_conv, g_zone_nr_conv, uint, 0444);
 221 MODULE_PARM_DESC(zone_nr_conv, "Number of conventional zones when block device is zoned. Default: 0");
 222
 223 static unsigned int g_zone_max_open;
 224 module_param_named(zone_max_open, g_zone_max_open, uint, 0444);
 225 MODULE_PARM_DESC(zone_max_open, "Maximum number of open zones when block device is zoned. Default: 0 (no limit)");
 226
 227 static unsigned int g_zone_max_active;
 228 module_param_named(zone_max_active, g_zone_max_active, uint, 0444);
 229 MODULE_PARM_DESC(zone_max_active, "Maximum number of active zones when block device is zoned. Default: 0 (no limit)");
 230
 231 static struct nullb_device *null_alloc_dev(void);
 232 static void null_free_dev(struct nullb_device *dev);
 233 static void null_del_dev(struct nullb *nullb);
 234 static int null_add_dev(struct nullb_device *dev);
 235 static void null_free_device_storage(struct nullb_device *dev, bool is_cache);
 236
 237 static inline struct nullb_device *to_nullb_device(struct config_item *item)
 238 {
 239         return item ? container_of(item, struct nullb_device, item) : NULL;
 240 }
 241
 242 static inline ssize_t nullb_device_uint_attr_show(unsigned int val, char *page)
 243 {
 244         return snprintf(page, PAGE_SIZE, "%u\n", val);
 245 }
 246
 247 static inline ssize_t nullb_device_ulong_attr_show(unsigned long val,
 248         char *page)
 249 {
 250         return snprintf(page, PAGE_SIZE, "%lu\n", val);
 251 }
 252
 253 static inline ssize_t nullb_device_bool_attr_show(bool val, char *page)
 254 {
 255         return snprintf(page, PAGE_SIZE, "%u\n", val);
 256 }
 257
 258 static ssize_t nullb_device_uint_attr_store(unsigned int *val,
 259         const char *page, size_t count)
 260 {
 261         unsigned int tmp;
 262         int result;
 263
 264         result = kstrtouint(page, 0, &tmp);
 265         if (result < 0)
 266                 return result;
 267
 268         *val = tmp;
 269         return count;
 270 }
 271
 272 static ssize_t nullb_device_ulong_attr_store(unsigned long *val,
 273         const char *page, size_t count)
 274 {
 275         int result;
 276         unsigned long tmp;
 277
 278         result = kstrtoul(page, 0, &tmp);
 279         if (result < 0)
 280                 return result;
 281
 282         *val = tmp;
 283         return count;
 284 }
 285
 286 static ssize_t nullb_device_bool_attr_store(bool *val, const char *page,
 287         size_t count)
 288 {
 289         bool tmp;
 290         int result;
 291
 292         result = kstrtobool(page,  &tmp);
 293         if (result < 0)
 294                 return result;
 295
 296         *val = tmp;
 297         return count;
 298 }
 299
 300 /* The following macro should only be used with TYPE = {uint, ulong, bool}. */
 301 #define NULLB_DEVICE_ATTR(NAME, TYPE, APPLY)                            \
 302 static ssize_t                                                          \
 303 nullb_device_##NAME##_show(struct config_item *item, char *page)        \
 304 {                                                                       \
 305         return nullb_device_##TYPE##_attr_show(                         \
 306                                 to_nullb_device(item)->NAME, page);     \
 307 }                                                                       \
 308 static ssize_t                                                          \
 309 nullb_device_##NAME##_store(struct config_item *item, const char *page, \
 310                             size_t count)                               \
 311 {                                                                       \
 312         int (*apply_fn)(struct nullb_device *dev, TYPE new_value) = APPLY;\
 313         struct nullb_device *dev = to_nullb_device(item);               \
 314         TYPE new_value = 0;                                             \
 315         int ret;                                                        \
 316                                                                         \
 317         ret = nullb_device_##TYPE##_attr_store(&new_value, page, count);\
 318         if (ret < 0)                                                    \
 319                 return ret;                                             \
 320         if (apply_fn)                                                   \
 321                 ret = apply_fn(dev, new_value);                         \
 322         else if (test_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags))        \
 323                 ret = -EBUSY;                                           \
 324         if (ret < 0)                                                    \
 325                 return ret;                                             \
 326         dev->NAME = new_value;                                          \
 327         return count;                                                   \
 328 }                                                                       \
 329 CONFIGFS_ATTR(nullb_device_, NAME);
 330
 331 static int nullb_apply_submit_queues(struct nullb_device *dev,
 332                                      unsigned int submit_queues)
 333 {
 334         struct nullb *nullb = dev->nullb;
 335         struct blk_mq_tag_set *set;
 336
 337         if (!nullb)
 338                 return 0;
 339
 340         /*
 341          * Make sure that null_init_hctx() does not access nullb->queues[] past
 342          * the end of that array.
 343          */
 344         if (submit_queues > nr_cpu_ids)
 345                 return -EINVAL;
 346         set = nullb->tag_set;
 347         blk_mq_update_nr_hw_queues(set, submit_queues);
 348         return set->nr_hw_queues == submit_queues ? 0 : -ENOMEM;
 349 }
 350
 351 NULLB_DEVICE_ATTR(size, ulong, NULL);
 352 NULLB_DEVICE_ATTR(completion_nsec, ulong, NULL);
 353 NULLB_DEVICE_ATTR(submit_queues, uint, nullb_apply_submit_queues);
 354 NULLB_DEVICE_ATTR(home_node, uint, NULL);
 355 NULLB_DEVICE_ATTR(queue_mode, uint, NULL);
 356 NULLB_DEVICE_ATTR(blocksize, uint, NULL);
 357 NULLB_DEVICE_ATTR(max_sectors, uint, NULL);
 358 NULLB_DEVICE_ATTR(irqmode, uint, NULL);
 359 NULLB_DEVICE_ATTR(hw_queue_depth, uint, NULL);
 360 NULLB_DEVICE_ATTR(index, uint, NULL);
 361 NULLB_DEVICE_ATTR(blocking, bool, NULL);
 362 NULLB_DEVICE_ATTR(use_per_node_hctx, bool, NULL);
 363 NULLB_DEVICE_ATTR(memory_backed, bool, NULL);
 364 NULLB_DEVICE_ATTR(discard, bool, NULL);
 365 NULLB_DEVICE_ATTR(mbps, uint, NULL);
 366 NULLB_DEVICE_ATTR(cache_size, ulong, NULL);
 367 NULLB_DEVICE_ATTR(zoned, bool, NULL);
 368 NULLB_DEVICE_ATTR(zone_size, ulong, NULL);
 369 NULLB_DEVICE_ATTR(zone_capacity, ulong, NULL);
 370 NULLB_DEVICE_ATTR(zone_nr_conv, uint, NULL);
 371 NULLB_DEVICE_ATTR(zone_max_open, uint, NULL);
 372 NULLB_DEVICE_ATTR(zone_max_active, uint, NULL);
 373 NULLB_DEVICE_ATTR(virt_boundary, bool, NULL);
 374
 375 static ssize_t nullb_device_power_show(struct config_item *item, char *page)
 376 {
 377         return nullb_device_bool_attr_show(to_nullb_device(item)->power, page);
 378 }
 379
 380 static ssize_t nullb_device_power_store(struct config_item *item,
 381                                      const char *page, size_t count)
 382 {
 383         struct nullb_device *dev = to_nullb_device(item);
 384         bool newp = false;
 385         ssize_t ret;
 386
 387         ret = nullb_device_bool_attr_store(&newp, page, count);
 388         if (ret < 0)
 389                 return ret;
 390
 391         if (!dev->power && newp) {
 392                 if (test_and_set_bit(NULLB_DEV_FL_UP, &dev->flags))
 393                         return count;
 394                 if (null_add_dev(dev)) {
 395                         clear_bit(NULLB_DEV_FL_UP, &dev->flags);
 396                         return -ENOMEM;
 397                 }
 398
 399                 set_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
 400                 dev->power = newp;
 401         } else if (dev->power && !newp) {
 402                 if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
 403                         mutex_lock(&lock);
 404                         dev->power = newp;
 405                         null_del_dev(dev->nullb);
 406                         mutex_unlock(&lock);
 407                 }
 408                 clear_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
 409         }
 410
 411         return count;
 412 }
 413
 414 CONFIGFS_ATTR(nullb_device_, power);
 415
 416 static ssize_t nullb_device_badblocks_show(struct config_item *item, char *page)
 417 {
 418         struct nullb_device *t_dev = to_nullb_device(item);
 419
 420         return badblocks_show(&t_dev->badblocks, page, 0);
 421 }
 422
 423 static ssize_t nullb_device_badblocks_store(struct config_item *item,
 424                                      const char *page, size_t count)
 425 {
 426         struct nullb_device *t_dev = to_nullb_device(item);
 427         char *orig, *buf, *tmp;
 428         u64 start, end;
 429         int ret;
 430
 431         orig = kstrndup(page, count, GFP_KERNEL);
 432         if (!orig)
 433                 return -ENOMEM;
 434
 435         buf = strstrip(orig);
 436
 437         ret = -EINVAL;
 438         if (buf[0] != '+' && buf[0] != '-')
 439                 goto out;
 440         tmp = strchr(&buf[1], '-');
 441         if (!tmp)
 442                 goto out;
 443         *tmp = '\0';
 444         ret = kstrtoull(buf + 1, 0, &start);
 445         if (ret)
 446                 goto out;
 447         ret = kstrtoull(tmp + 1, 0, &end);
 448         if (ret)
 449                 goto out;
 450         ret = -EINVAL;
 451         if (start > end)
 452                 goto out;
 453         /* enable badblocks */
 454         cmpxchg(&t_dev->badblocks.shift, -1, 0);
 455         if (buf[0] == '+')
 456                 ret = badblocks_set(&t_dev->badblocks, start,
 457                         end - start + 1, 1);
 458         else
 459                 ret = badblocks_clear(&t_dev->badblocks, start,
 460                         end - start + 1);
 461         if (ret == 0)
 462                 ret = count;
 463 out:
 464         kfree(orig);
 465         return ret;
 466 }
 467 CONFIGFS_ATTR(nullb_device_, badblocks);
 468
 469 static struct configfs_attribute *nullb_device_attrs[] = {
 470         &nullb_device_attr_size,
 471         &nullb_device_attr_completion_nsec,
 472         &nullb_device_attr_submit_queues,
 473         &nullb_device_attr_home_node,
 474         &nullb_device_attr_queue_mode,
 475         &nullb_device_attr_blocksize,
 476         &nullb_device_attr_max_sectors,
 477         &nullb_device_attr_irqmode,
 478         &nullb_device_attr_hw_queue_depth,
 479         &nullb_device_attr_index,
 480         &nullb_device_attr_blocking,
 481         &nullb_device_attr_use_per_node_hctx,
 482         &nullb_device_attr_power,
 483         &nullb_device_attr_memory_backed,
 484         &nullb_device_attr_discard,
 485         &nullb_device_attr_mbps,
 486         &nullb_device_attr_cache_size,
 487         &nullb_device_attr_badblocks,
 488         &nullb_device_attr_zoned,
 489         &nullb_device_attr_zone_size,
 490         &nullb_device_attr_zone_capacity,
 491         &nullb_device_attr_zone_nr_conv,
 492         &nullb_device_attr_zone_max_open,
 493         &nullb_device_attr_zone_max_active,
 494         &nullb_device_attr_virt_boundary,
 495         NULL,
 496 };
 497
 498 static void nullb_device_release(struct config_item *item)
 499 {
 500         struct nullb_device *dev = to_nullb_device(item);
 501
 502         null_free_device_storage(dev, false);
 503         null_free_dev(dev);
 504 }
 505
 506 static struct configfs_item_operations nullb_device_ops = {
 507         .release        = nullb_device_release,
 508 };
 509
 510 static const struct config_item_type nullb_device_type = {
 511         .ct_item_ops    = &nullb_device_ops,
 512         .ct_attrs       = nullb_device_attrs,
 513         .ct_owner       = THIS_MODULE,
 514 };
 515
 516 static struct
 517 config_item *nullb_group_make_item(struct config_group *group, const char *name)
 518 {
 519         struct nullb_device *dev;
 520
 521         dev = null_alloc_dev();
 522         if (!dev)
 523                 return ERR_PTR(-ENOMEM);
 524
 525         config_item_init_type_name(&dev->item, name, &nullb_device_type);
 526
 527         return &dev->item;
 528 }
 529
 530 static void
 531 nullb_group_drop_item(struct config_group *group, struct config_item *item)
 532 {
 533         struct nullb_device *dev = to_nullb_device(item);
 534
 535         if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
 536                 mutex_lock(&lock);
 537                 dev->power = false;
 538                 null_del_dev(dev->nullb);
 539                 mutex_unlock(&lock);
 540         }
 541
 542         config_item_put(item);
 543 }
 544
 545 static ssize_t memb_group_features_show(struct config_item *item, char *page)
 546 {
 547         return snprintf(page, PAGE_SIZE,
 548                         "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");
 549 }
 550
 551 CONFIGFS_ATTR_RO(memb_group_, features);
 552
 553 static struct configfs_attribute *nullb_group_attrs[] = {
 554         &memb_group_attr_features,
 555         NULL,
 556 };
 557
 558 static struct configfs_group_operations nullb_group_ops = {
 559         .make_item      = nullb_group_make_item,
 560         .drop_item      = nullb_group_drop_item,
 561 };
 562
 563 static const struct config_item_type nullb_group_type = {
 564         .ct_group_ops   = &nullb_group_ops,
 565         .ct_attrs       = nullb_group_attrs,
 566         .ct_owner       = THIS_MODULE,
 567 };
 568
 569 static struct configfs_subsystem nullb_subsys = {
 570         .su_group = {
 571                 .cg_item = {
 572                         .ci_namebuf = "nullb",
 573                         .ci_type = &nullb_group_type,
 574                 },
 575         },
 576 };
 577
 578 static inline int null_cache_active(struct nullb *nullb)
 579 {
 580         return test_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
 581 }
 582
 583 static struct nullb_device *null_alloc_dev(void)
 584 {
 585         struct nullb_device *dev;
 586
 587         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
 588         if (!dev)
 589                 return NULL;
 590         INIT_RADIX_TREE(&dev->data, GFP_ATOMIC);
 591         INIT_RADIX_TREE(&dev->cache, GFP_ATOMIC);
 592         if (badblocks_init(&dev->badblocks, 0)) {
 593                 kfree(dev);
 594                 return NULL;
 595         }
 596
 597         dev->size = g_gb * 1024;
 598         dev->completion_nsec = g_completion_nsec;
 599         dev->submit_queues = g_submit_queues;
 600         dev->home_node = g_home_node;
 601         dev->queue_mode = g_queue_mode;
 602         dev->blocksize = g_bs;
 603         dev->max_sectors = g_max_sectors;
 604         dev->irqmode = g_irqmode;
 605         dev->hw_queue_depth = g_hw_queue_depth;
 606         dev->blocking = g_blocking;
 607         dev->use_per_node_hctx = g_use_per_node_hctx;
 608         dev->zoned = g_zoned;
 609         dev->zone_size = g_zone_size;
 610         dev->zone_capacity = g_zone_capacity;
 611         dev->zone_nr_conv = g_zone_nr_conv;
 612         dev->zone_max_open = g_zone_max_open;
 613         dev->zone_max_active = g_zone_max_active;
 614         dev->virt_boundary = g_virt_boundary;
 615         return dev;
 616 }
 617
 618 static void null_free_dev(struct nullb_device *dev)
 619 {
 620         if (!dev)
 621                 return;
 622
 623         null_free_zoned_dev(dev);
 624         badblocks_exit(&dev->badblocks);
 625         kfree(dev);
 626 }
 627
 628 static void put_tag(struct nullb_queue *nq, unsigned int tag)
 629 {
 630         clear_bit_unlock(tag, nq->tag_map);
 631
 632         if (waitqueue_active(&nq->wait))
 633                 wake_up(&nq->wait);
 634 }
 635
 636 static unsigned int get_tag(struct nullb_queue *nq)
 637 {
 638         unsigned int tag;
 639
 640         do {
 641                 tag = find_first_zero_bit(nq->tag_map, nq->queue_depth);
 642                 if (tag >= nq->queue_depth)
 643                         return -1U;
 644         } while (test_and_set_bit_lock(tag, nq->tag_map));
 645
 646         return tag;
 647 }
 648
 649 static void free_cmd(struct nullb_cmd *cmd)
 650 {
 651         put_tag(cmd->nq, cmd->tag);
 652 }
 653
 654 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer);
 655
 656 static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq)
 657 {
 658         struct nullb_cmd *cmd;
 659         unsigned int tag;
 660
 661         tag = get_tag(nq);
 662         if (tag != -1U) {
 663                 cmd = &nq->cmds[tag];
 664                 cmd->tag = tag;
 665                 cmd->error = BLK_STS_OK;
 666                 cmd->nq = nq;
 667                 if (nq->dev->irqmode == NULL_IRQ_TIMER) {
 668                         hrtimer_init(&cmd->timer, CLOCK_MONOTONIC,
 669                                      HRTIMER_MODE_REL);
 670                         cmd->timer.function = null_cmd_timer_expired;
 671                 }
 672                 return cmd;
 673         }
 674
 675         return NULL;
 676 }
 677
 678 static struct nullb_cmd *alloc_cmd(struct nullb_queue *nq, int can_wait)
 679 {
 680         struct nullb_cmd *cmd;
 681         DEFINE_WAIT(wait);
 682
 683         cmd = __alloc_cmd(nq);
 684         if (cmd || !can_wait)
 685                 return cmd;
 686
 687         do {
 688                 prepare_to_wait(&nq->wait, &wait, TASK_UNINTERRUPTIBLE);
 689                 cmd = __alloc_cmd(nq);
 690                 if (cmd)
 691                         break;
 692
 693                 io_schedule();
 694         } while (1);
 695
 696         finish_wait(&nq->wait, &wait);
 697         return cmd;
 698 }
 699
 700 static void end_cmd(struct nullb_cmd *cmd)
 701 {
 702         int queue_mode = cmd->nq->dev->queue_mode;
 703
 704         switch (queue_mode)  {
 705         case NULL_Q_MQ:
 706                 blk_mq_end_request(cmd->rq, cmd->error);
 707                 return;
 708         case NULL_Q_BIO:
 709                 cmd->bio->bi_status = cmd->error;
 710                 bio_endio(cmd->bio);
 711                 break;
 712         }
 713
 714         free_cmd(cmd);
 715 }
 716
 717 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
 718 {
 719         end_cmd(container_of(timer, struct nullb_cmd, timer));
 720
 721         return HRTIMER_NORESTART;
 722 }
 723
 724 static void null_cmd_end_timer(struct nullb_cmd *cmd)
 725 {
 726         ktime_t kt = cmd->nq->dev->completion_nsec;
 727
 728         hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL);
 729 }
 730
 731 static void null_complete_rq(struct request *rq)
 732 {
 733         end_cmd(blk_mq_rq_to_pdu(rq));
 734 }
 735
 736 static struct nullb_page *null_alloc_page(gfp_t gfp_flags)
 737 {
 738         struct nullb_page *t_page;
 739
 740         t_page = kmalloc(sizeof(struct nullb_page), gfp_flags);
 741         if (!t_page)
 742                 goto out;
 743
 744         t_page->page = alloc_pages(gfp_flags, 0);
 745         if (!t_page->page)
 746                 goto out_freepage;
 747
 748         memset(t_page->bitmap, 0, sizeof(t_page->bitmap));
 749         return t_page;
 750 out_freepage:
 751         kfree(t_page);
 752 out:
 753         return NULL;
 754 }
 755
 756 static void null_free_page(struct nullb_page *t_page)
 757 {
 758         __set_bit(NULLB_PAGE_FREE, t_page->bitmap);
 759         if (test_bit(NULLB_PAGE_LOCK, t_page->bitmap))
 760                 return;
 761         __free_page(t_page->page);
 762         kfree(t_page);
 763 }
 764
 765 static bool null_page_empty(struct nullb_page *page)
 766 {
 767         int size = MAP_SZ - 2;
 768
 769         return find_first_bit(page->bitmap, size) == size;
 770 }
 771
 772 static void null_free_sector(struct nullb *nullb, sector_t sector,
 773         bool is_cache)
 774 {
 775         unsigned int sector_bit;
 776         u64 idx;
 777         struct nullb_page *t_page, *ret;
 778         struct radix_tree_root *root;
 779
 780         root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
 781         idx = sector >> PAGE_SECTORS_SHIFT;
 782         sector_bit = (sector & SECTOR_MASK);
 783
 784         t_page = radix_tree_lookup(root, idx);
 785         if (t_page) {
 786                 __clear_bit(sector_bit, t_page->bitmap);
 787
 788                 if (null_page_empty(t_page)) {
 789                         ret = radix_tree_delete_item(root, idx, t_page);
 790                         WARN_ON(ret != t_page);
 791                         null_free_page(ret);
 792                         if (is_cache)
 793                                 nullb->dev->curr_cache -= PAGE_SIZE;
 794                 }
 795         }
 796 }
 797
 798 static struct nullb_page *null_radix_tree_insert(struct nullb *nullb, u64 idx,
 799         struct nullb_page *t_page, bool is_cache)
 800 {
 801         struct radix_tree_root *root;
 802
 803         root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
 804
 805         if (radix_tree_insert(root, idx, t_page)) {
 806                 null_free_page(t_page);
 807                 t_page = radix_tree_lookup(root, idx);
 808                 WARN_ON(!t_page || t_page->page->index != idx);
 809         } else if (is_cache)
 810                 nullb->dev->curr_cache += PAGE_SIZE;
 811
 812         return t_page;
 813 }
 814
 815 static void null_free_device_storage(struct nullb_device *dev, bool is_cache)
 816 {
 817         unsigned long pos = 0;
 818         int nr_pages;
 819         struct nullb_page *ret, *t_pages[FREE_BATCH];
 820         struct radix_tree_root *root;
 821
 822         root = is_cache ? &dev->cache : &dev->data;
 823
 824         do {
 825                 int i;
 826
 827                 nr_pages = radix_tree_gang_lookup(root,
 828                                 (void **)t_pages, pos, FREE_BATCH);
 829
 830                 for (i = 0; i < nr_pages; i++) {
 831                         pos = t_pages[i]->page->index;
 832                         ret = radix_tree_delete_item(root, pos, t_pages[i]);
 833                         WARN_ON(ret != t_pages[i]);
 834                         null_free_page(ret);
 835                 }
 836
 837                 pos++;
 838         } while (nr_pages == FREE_BATCH);
 839
 840         if (is_cache)
 841                 dev->curr_cache = 0;
 842 }
 843
 844 static struct nullb_page *__null_lookup_page(struct nullb *nullb,
 845         sector_t sector, bool for_write, bool is_cache)
 846 {
 847         unsigned int sector_bit;
 848         u64 idx;
 849         struct nullb_page *t_page;
 850         struct radix_tree_root *root;
 851
 852         idx = sector >> PAGE_SECTORS_SHIFT;
 853         sector_bit = (sector & SECTOR_MASK);
 854
 855         root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
 856         t_page = radix_tree_lookup(root, idx);
 857         WARN_ON(t_page && t_page->page->index != idx);
 858
 859         if (t_page && (for_write || test_bit(sector_bit, t_page->bitmap)))
 860                 return t_page;
 861
 862         return NULL;
 863 }
 864
 865 static struct nullb_page *null_lookup_page(struct nullb *nullb,
 866         sector_t sector, bool for_write, bool ignore_cache)
 867 {
 868         struct nullb_page *page = NULL;
 869
 870         if (!ignore_cache)
 871                 page = __null_lookup_page(nullb, sector, for_write, true);
 872         if (page)
 873                 return page;
 874         return __null_lookup_page(nullb, sector, for_write, false);
 875 }
 876
 877 static struct nullb_page *null_insert_page(struct nullb *nullb,
 878                                            sector_t sector, bool ignore_cache)
 879         __releases(&nullb->lock)
 880         __acquires(&nullb->lock)
 881 {
 882         u64 idx;
 883         struct nullb_page *t_page;
 884
 885         t_page = null_lookup_page(nullb, sector, true, ignore_cache);
 886         if (t_page)
 887                 return t_page;
 888
 889         spin_unlock_irq(&nullb->lock);
 890
 891         t_page = null_alloc_page(GFP_NOIO);
 892         if (!t_page)
 893                 goto out_lock;
 894
 895         if (radix_tree_preload(GFP_NOIO))
 896                 goto out_freepage;
 897
 898         spin_lock_irq(&nullb->lock);
 899         idx = sector >> PAGE_SECTORS_SHIFT;
 900         t_page->page->index = idx;
 901         t_page = null_radix_tree_insert(nullb, idx, t_page, !ignore_cache);
 902         radix_tree_preload_end();
 903
 904         return t_page;
 905 out_freepage:
 906         null_free_page(t_page);
 907 out_lock:
 908         spin_lock_irq(&nullb->lock);
 909         return null_lookup_page(nullb, sector, true, ignore_cache);
 910 }
 911
 912 static int null_flush_cache_page(struct nullb *nullb, struct nullb_page *c_page)
 913 {
 914         int i;
 915         unsigned int offset;
 916         u64 idx;
 917         struct nullb_page *t_page, *ret;
 918         void *dst, *src;
 919
 920         idx = c_page->page->index;
 921
 922         t_page = null_insert_page(nullb, idx << PAGE_SECTORS_SHIFT, true);
 923
 924         __clear_bit(NULLB_PAGE_LOCK, c_page->bitmap);
 925         if (test_bit(NULLB_PAGE_FREE, c_page->bitmap)) {
 926                 null_free_page(c_page);
 927                 if (t_page && null_page_empty(t_page)) {
 928                         ret = radix_tree_delete_item(&nullb->dev->data,
 929                                 idx, t_page);
 930                         null_free_page(t_page);
 931                 }
 932                 return 0;
 933         }
 934
 935         if (!t_page)
 936                 return -ENOMEM;
 937
 938         src = kmap_atomic(c_page->page);
 939         dst = kmap_atomic(t_page->page);
 940
 941         for (i = 0; i < PAGE_SECTORS;
 942                         i += (nullb->dev->blocksize >> SECTOR_SHIFT)) {
 943                 if (test_bit(i, c_page->bitmap)) {
 944                         offset = (i << SECTOR_SHIFT);
 945                         memcpy(dst + offset, src + offset,
 946                                 nullb->dev->blocksize);
 947                         __set_bit(i, t_page->bitmap);
 948                 }
 949         }
 950
 951         kunmap_atomic(dst);
 952         kunmap_atomic(src);
 953
 954         ret = radix_tree_delete_item(&nullb->dev->cache, idx, c_page);
 955         null_free_page(ret);
 956         nullb->dev->curr_cache -= PAGE_SIZE;
 957
 958         return 0;
 959 }
 960
 961 static int null_make_cache_space(struct nullb *nullb, unsigned long n)
 962 {
 963         int i, err, nr_pages;
 964         struct nullb_page *c_pages[FREE_BATCH];
 965         unsigned long flushed = 0, one_round;
 966
 967 again:
 968         if ((nullb->dev->cache_size * 1024 * 1024) >
 969              nullb->dev->curr_cache + n || nullb->dev->curr_cache == 0)
 970                 return 0;
 971
 972         nr_pages = radix_tree_gang_lookup(&nullb->dev->cache,
 973                         (void **)c_pages, nullb->cache_flush_pos, FREE_BATCH);
 974         /*
 975          * nullb_flush_cache_page could unlock before using the c_pages. To
 976          * avoid race, we don't allow page free
 977          */
 978         for (i = 0; i < nr_pages; i++) {
 979                 nullb->cache_flush_pos = c_pages[i]->page->index;
 980                 /*
 981                  * We found the page which is being flushed to disk by other
 982                  * threads
 983                  */
 984                 if (test_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap))
 985                         c_pages[i] = NULL;
 986                 else
 987                         __set_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap);
 988         }
 989
 990         one_round = 0;
 991         for (i = 0; i < nr_pages; i++) {
 992                 if (c_pages[i] == NULL)
 993                         continue;
 994                 err = null_flush_cache_page(nullb, c_pages[i]);
 995                 if (err)
 996                         return err;
 997                 one_round++;
 998         }
 999         flushed += one_round << PAGE_SHIFT;
1000
1001         if (n > flushed) {
1002                 if (nr_pages == 0)
1003                         nullb->cache_flush_pos = 0;
1004                 if (one_round == 0) {
1005                         /* give other threads a chance */
1006                         spin_unlock_irq(&nullb->lock);
1007                         spin_lock_irq(&nullb->lock);
1008                 }
1009                 goto again;
1010         }
1011         return 0;
1012 }
1013
1014 static int copy_to_nullb(struct nullb *nullb, struct page *source,
1015         unsigned int off, sector_t sector, size_t n, bool is_fua)
1016 {
1017         size_t temp, count = 0;
1018         unsigned int offset;
1019         struct nullb_page *t_page;
1020         void *dst, *src;
1021
1022         while (count < n) {
1023                 temp = min_t(size_t, nullb->dev->blocksize, n - count);
1024
1025                 if (null_cache_active(nullb) && !is_fua)
1026                         null_make_cache_space(nullb, PAGE_SIZE);
1027
1028                 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
1029                 t_page = null_insert_page(nullb, sector,
1030                         !null_cache_active(nullb) || is_fua);
1031                 if (!t_page)
1032                         return -ENOSPC;
1033
1034                 src = kmap_atomic(source);
1035                 dst = kmap_atomic(t_page->page);
1036                 memcpy(dst + offset, src + off + count, temp);
1037                 kunmap_atomic(dst);
1038                 kunmap_atomic(src);
1039
1040                 __set_bit(sector & SECTOR_MASK, t_page->bitmap);
1041
1042                 if (is_fua)
1043                         null_free_sector(nullb, sector, true);
1044
1045                 count += temp;
1046                 sector += temp >> SECTOR_SHIFT;
1047         }
1048         return 0;
1049 }
1050
1051 static int copy_from_nullb(struct nullb *nullb, struct page *dest,
1052         unsigned int off, sector_t sector, size_t n)
1053 {
1054         size_t temp, count = 0;
1055         unsigned int offset;
1056         struct nullb_page *t_page;
1057         void *dst, *src;
1058
1059         while (count < n) {
1060                 temp = min_t(size_t, nullb->dev->blocksize, n - count);
1061
1062                 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
1063                 t_page = null_lookup_page(nullb, sector, false,
1064                         !null_cache_active(nullb));
1065
1066                 dst = kmap_atomic(dest);
1067                 if (!t_page) {
1068                         memset(dst + off + count, 0, temp);
1069                         goto next;
1070                 }
1071                 src = kmap_atomic(t_page->page);
1072                 memcpy(dst + off + count, src + offset, temp);
1073                 kunmap_atomic(src);
1074 next:
1075                 kunmap_atomic(dst);
1076
1077                 count += temp;
1078                 sector += temp >> SECTOR_SHIFT;
1079         }
1080         return 0;
1081 }
1082
1083 static void nullb_fill_pattern(struct nullb *nullb, struct page *page,
1084                                unsigned int len, unsigned int off)
1085 {
1086         void *dst;
1087
1088         dst = kmap_atomic(page);
1089         memset(dst + off, 0xFF, len);
1090         kunmap_atomic(dst);
1091 }
1092
1093 blk_status_t null_handle_discard(struct nullb_device *dev,
1094                                  sector_t sector, sector_t nr_sectors)
1095 {
1096         struct nullb *nullb = dev->nullb;
1097         size_t n = nr_sectors << SECTOR_SHIFT;
1098         size_t temp;
1099
1100         spin_lock_irq(&nullb->lock);
1101         while (n > 0) {
1102                 temp = min_t(size_t, n, dev->blocksize);
1103                 null_free_sector(nullb, sector, false);
1104                 if (null_cache_active(nullb))
1105                         null_free_sector(nullb, sector, true);
1106                 sector += temp >> SECTOR_SHIFT;
1107                 n -= temp;
1108         }
1109         spin_unlock_irq(&nullb->lock);
1110
1111         return BLK_STS_OK;
1112 }
1113
1114 static int null_handle_flush(struct nullb *nullb)
1115 {
1116         int err;
1117
1118         if (!null_cache_active(nullb))
1119                 return 0;
1120
1121         spin_lock_irq(&nullb->lock);
1122         while (true) {
1123                 err = null_make_cache_space(nullb,
1124                         nullb->dev->cache_size * 1024 * 1024);
1125                 if (err || nullb->dev->curr_cache == 0)
1126                         break;
1127         }
1128
1129         WARN_ON(!radix_tree_empty(&nullb->dev->cache));
1130         spin_unlock_irq(&nullb->lock);
1131         return err;
1132 }
1133
1134 static int null_transfer(struct nullb *nullb, struct page *page,
1135         unsigned int len, unsigned int off, bool is_write, sector_t sector,
1136         bool is_fua)
1137 {
1138         struct nullb_device *dev = nullb->dev;
1139         unsigned int valid_len = len;
1140         int err = 0;
1141
1142         if (!is_write) {
1143                 if (dev->zoned)
1144                         valid_len = null_zone_valid_read_len(nullb,
1145                                 sector, len);
1146
1147                 if (valid_len) {
1148                         err = copy_from_nullb(nullb, page, off,
1149                                 sector, valid_len);
1150                         off += valid_len;
1151                         len -= valid_len;
1152                 }
1153
1154                 if (len)
1155                         nullb_fill_pattern(nullb, page, len, off);
1156                 flush_dcache_page(page);
1157         } else {
1158                 flush_dcache_page(page);
1159                 err = copy_to_nullb(nullb, page, off, sector, len, is_fua);
1160         }
1161
1162         return err;
1163 }
1164
1165 static int null_handle_rq(struct nullb_cmd *cmd)
1166 {
1167         struct request *rq = cmd->rq;
1168         struct nullb *nullb = cmd->nq->dev->nullb;
1169         int err;
1170         unsigned int len;
1171         sector_t sector = blk_rq_pos(rq);
1172         struct req_iterator iter;
1173         struct bio_vec bvec;
1174
1175         spin_lock_irq(&nullb->lock);
1176         rq_for_each_segment(bvec, rq, iter) {
1177                 len = bvec.bv_len;
1178                 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1179                                      op_is_write(req_op(rq)), sector,
1180                                      rq->cmd_flags & REQ_FUA);
1181                 if (err) {
1182                         spin_unlock_irq(&nullb->lock);
1183                         return err;
1184                 }
1185                 sector += len >> SECTOR_SHIFT;
1186         }
1187         spin_unlock_irq(&nullb->lock);
1188
1189         return 0;
1190 }
1191
1192 static int null_handle_bio(struct nullb_cmd *cmd)
1193 {
1194         struct bio *bio = cmd->bio;
1195         struct nullb *nullb = cmd->nq->dev->nullb;
1196         int err;
1197         unsigned int len;
1198         sector_t sector = bio->bi_iter.bi_sector;
1199         struct bio_vec bvec;
1200         struct bvec_iter iter;
1201
1202         spin_lock_irq(&nullb->lock);
1203         bio_for_each_segment(bvec, bio, iter) {
1204                 len = bvec.bv_len;
1205                 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1206                                      op_is_write(bio_op(bio)), sector,
1207                                      bio->bi_opf & REQ_FUA);
1208                 if (err) {
1209                         spin_unlock_irq(&nullb->lock);
1210                         return err;
1211                 }
1212                 sector += len >> SECTOR_SHIFT;
1213         }
1214         spin_unlock_irq(&nullb->lock);
1215         return 0;
1216 }
1217
1218 static void null_stop_queue(struct nullb *nullb)
1219 {
1220         struct request_queue *q = nullb->q;
1221
1222         if (nullb->dev->queue_mode == NULL_Q_MQ)
1223                 blk_mq_stop_hw_queues(q);
1224 }
1225
1226 static void null_restart_queue_async(struct nullb *nullb)
1227 {
1228         struct request_queue *q = nullb->q;
1229
1230         if (nullb->dev->queue_mode == NULL_Q_MQ)
1231                 blk_mq_start_stopped_hw_queues(q, true);
1232 }
1233
1234 static inline blk_status_t null_handle_throttled(struct nullb_cmd *cmd)
1235 {
1236         struct nullb_device *dev = cmd->nq->dev;
1237         struct nullb *nullb = dev->nullb;
1238         blk_status_t sts = BLK_STS_OK;
1239         struct request *rq = cmd->rq;
1240
1241         if (!hrtimer_active(&nullb->bw_timer))
1242                 hrtimer_restart(&nullb->bw_timer);
1243
1244         if (atomic_long_sub_return(blk_rq_bytes(rq), &nullb->cur_bytes) < 0) {
1245                 null_stop_queue(nullb);
1246                 /* race with timer */
1247                 if (atomic_long_read(&nullb->cur_bytes) > 0)
1248                         null_restart_queue_async(nullb);
1249                 /* requeue request */
1250                 sts = BLK_STS_DEV_RESOURCE;
1251         }
1252         return sts;
1253 }
1254
1255 static inline blk_status_t null_handle_badblocks(struct nullb_cmd *cmd,
1256                                                  sector_t sector,
1257                                                  sector_t nr_sectors)
1258 {
1259         struct badblocks *bb = &cmd->nq->dev->badblocks;
1260         sector_t first_bad;
1261         int bad_sectors;
1262
1263         if (badblocks_check(bb, sector, nr_sectors, &first_bad, &bad_sectors))
1264                 return BLK_STS_IOERR;
1265
1266         return BLK_STS_OK;
1267 }
1268
1269 static inline blk_status_t null_handle_memory_backed(struct nullb_cmd *cmd,
1270                                                      enum req_opf op,
1271                                                      sector_t sector,
1272                                                      sector_t nr_sectors)
1273 {
1274         struct nullb_device *dev = cmd->nq->dev;
1275         int err;
1276
1277         if (op == REQ_OP_DISCARD)
1278                 return null_handle_discard(dev, sector, nr_sectors);
1279
1280         if (dev->queue_mode == NULL_Q_BIO)
1281                 err = null_handle_bio(cmd);
1282         else
1283                 err = null_handle_rq(cmd);
1284
1285         return errno_to_blk_status(err);
1286 }
1287
1288 static void nullb_zero_read_cmd_buffer(struct nullb_cmd *cmd)
1289 {
1290         struct nullb_device *dev = cmd->nq->dev;
1291         struct bio *bio;
1292
1293         if (dev->memory_backed)
1294                 return;
1295
1296         if (dev->queue_mode == NULL_Q_BIO && bio_op(cmd->bio) == REQ_OP_READ) {
1297                 zero_fill_bio(cmd->bio);
1298         } else if (req_op(cmd->rq) == REQ_OP_READ) {
1299                 __rq_for_each_bio(bio, cmd->rq)
1300                         zero_fill_bio(bio);
1301         }
1302 }
1303
1304 static inline void nullb_complete_cmd(struct nullb_cmd *cmd)
1305 {
1306         /*
1307          * Since root privileges are required to configure the null_blk
1308          * driver, it is fine that this driver does not initialize the
1309          * data buffers of read commands. Zero-initialize these buffers
1310          * anyway if KMSAN is enabled to prevent that KMSAN complains
1311          * about null_blk not initializing read data buffers.
1312          */
1313         if (IS_ENABLED(CONFIG_KMSAN))
1314                 nullb_zero_read_cmd_buffer(cmd);
1315
1316         /* Complete IO by inline, softirq or timer */
1317         switch (cmd->nq->dev->irqmode) {
1318         case NULL_IRQ_SOFTIRQ:
1319                 switch (cmd->nq->dev->queue_mode) {
1320                 case NULL_Q_MQ:
1321                         if (likely(!blk_should_fake_timeout(cmd->rq->q)))
1322                                 blk_mq_complete_request(cmd->rq);
1323                         break;
1324                 case NULL_Q_BIO:
1325                         /*
1326                          * XXX: no proper submitting cpu information available.
1327                          */
1328                         end_cmd(cmd);
1329                         break;
1330                 }
1331                 break;
1332         case NULL_IRQ_NONE:
1333                 end_cmd(cmd);
1334                 break;
1335         case NULL_IRQ_TIMER:
1336                 null_cmd_end_timer(cmd);
1337                 break;
1338         }
1339 }
1340
1341 blk_status_t null_process_cmd(struct nullb_cmd *cmd,
1342                               enum req_opf op, sector_t sector,
1343                               unsigned int nr_sectors)
1344 {
1345         struct nullb_device *dev = cmd->nq->dev;
1346         blk_status_t ret;
1347
1348         if (dev->badblocks.shift != -1) {
1349                 ret = null_handle_badblocks(cmd, sector, nr_sectors);
1350                 if (ret != BLK_STS_OK)
1351                         return ret;
1352         }
1353
1354         if (dev->memory_backed)
1355                 return null_handle_memory_backed(cmd, op, sector, nr_sectors);
1356
1357         return BLK_STS_OK;
1358 }
1359
1360 static blk_status_t null_handle_cmd(struct nullb_cmd *cmd, sector_t sector,
1361                                     sector_t nr_sectors, enum req_opf op)
1362 {
1363         struct nullb_device *dev = cmd->nq->dev;
1364         struct nullb *nullb = dev->nullb;
1365         blk_status_t sts;
1366
1367         if (test_bit(NULLB_DEV_FL_THROTTLED, &dev->flags)) {
1368                 sts = null_handle_throttled(cmd);
1369                 if (sts != BLK_STS_OK)
1370                         return sts;
1371         }
1372
1373         if (op == REQ_OP_FLUSH) {
1374                 cmd->error = errno_to_blk_status(null_handle_flush(nullb));
1375                 goto out;
1376         }
1377
1378         if (dev->zoned)
1379                 sts = null_process_zoned_cmd(cmd, op, sector, nr_sectors);
1380         else
1381                 sts = null_process_cmd(cmd, op, sector, nr_sectors);
1382
1383         /* Do not overwrite errors (e.g. timeout errors) */
1384         if (cmd->error == BLK_STS_OK)
1385                 cmd->error = sts;
1386
1387 out:
1388         nullb_complete_cmd(cmd);
1389         return BLK_STS_OK;
1390 }
1391
1392 static enum hrtimer_restart nullb_bwtimer_fn(struct hrtimer *timer)
1393 {
1394         struct nullb *nullb = container_of(timer, struct nullb, bw_timer);
1395         ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1396         unsigned int mbps = nullb->dev->mbps;
1397
1398         if (atomic_long_read(&nullb->cur_bytes) == mb_per_tick(mbps))
1399                 return HRTIMER_NORESTART;
1400
1401         atomic_long_set(&nullb->cur_bytes, mb_per_tick(mbps));
1402         null_restart_queue_async(nullb);
1403
1404         hrtimer_forward_now(&nullb->bw_timer, timer_interval);
1405
1406         return HRTIMER_RESTART;
1407 }
1408
1409 static void nullb_setup_bwtimer(struct nullb *nullb)
1410 {
1411         ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1412
1413         hrtimer_init(&nullb->bw_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1414         nullb->bw_timer.function = nullb_bwtimer_fn;
1415         atomic_long_set(&nullb->cur_bytes, mb_per_tick(nullb->dev->mbps));
1416         hrtimer_start(&nullb->bw_timer, timer_interval, HRTIMER_MODE_REL);
1417 }
1418
1419 static struct nullb_queue *nullb_to_queue(struct nullb *nullb)
1420 {
1421         int index = 0;
1422
1423         if (nullb->nr_queues != 1)
1424                 index = raw_smp_processor_id() / ((nr_cpu_ids + nullb->nr_queues - 1) / nullb->nr_queues);
1425
1426         return &nullb->queues[index];
1427 }
1428
1429 static blk_qc_t null_submit_bio(struct bio *bio)
1430 {
1431         sector_t sector = bio->bi_iter.bi_sector;
1432         sector_t nr_sectors = bio_sectors(bio);
1433         struct nullb *nullb = bio->bi_bdev->bd_disk->private_data;
1434         struct nullb_queue *nq = nullb_to_queue(nullb);
1435         struct nullb_cmd *cmd;
1436
1437         cmd = alloc_cmd(nq, 1);
1438         cmd->bio = bio;
1439
1440         null_handle_cmd(cmd, sector, nr_sectors, bio_op(bio));
1441         return BLK_QC_T_NONE;
1442 }
1443
1444 static bool should_timeout_request(struct request *rq)
1445 {
1446 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1447         if (g_timeout_str[0])
1448                 return should_fail(&null_timeout_attr, 1);
1449 #endif
1450         return false;
1451 }
1452
1453 static bool should_requeue_request(struct request *rq)
1454 {
1455 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1456         if (g_requeue_str[0])
1457                 return should_fail(&null_requeue_attr, 1);
1458 #endif
1459         return false;
1460 }
1461
1462 static enum blk_eh_timer_return null_timeout_rq(struct request *rq, bool res)
1463 {
1464         struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq);
1465
1466         pr_info("rq %p timed out\n", rq);
1467
1468         /*
1469          * If the device is marked as blocking (i.e. memory backed or zoned
1470          * device), the submission path may be blocked waiting for resources
1471          * and cause real timeouts. For these real timeouts, the submission
1472          * path will complete the request using blk_mq_complete_request().
1473          * Only fake timeouts need to execute blk_mq_complete_request() here.
1474          */
1475         cmd->error = BLK_STS_TIMEOUT;
1476         if (cmd->fake_timeout)
1477                 blk_mq_complete_request(rq);
1478         return BLK_EH_DONE;
1479 }
1480
1481 static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx,
1482                          const struct blk_mq_queue_data *bd)
1483 {
1484         struct nullb_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
1485         struct nullb_queue *nq = hctx->driver_data;
1486         sector_t nr_sectors = blk_rq_sectors(bd->rq);
1487         sector_t sector = blk_rq_pos(bd->rq);
1488
1489         might_sleep_if(hctx->flags & BLK_MQ_F_BLOCKING);
1490
1491         if (nq->dev->irqmode == NULL_IRQ_TIMER) {
1492                 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1493                 cmd->timer.function = null_cmd_timer_expired;
1494         }
1495         cmd->rq = bd->rq;
1496         cmd->error = BLK_STS_OK;
1497         cmd->nq = nq;
1498         cmd->fake_timeout = should_timeout_request(bd->rq);
1499
1500         blk_mq_start_request(bd->rq);
1501
1502         if (should_requeue_request(bd->rq)) {
1503                 /*
1504                  * Alternate between hitting the core BUSY path, and the
1505                  * driver driven requeue path
1506                  */
1507                 nq->requeue_selection++;
1508                 if (nq->requeue_selection & 1)
1509                         return BLK_STS_RESOURCE;
1510                 else {
1511                         blk_mq_requeue_request(bd->rq, true);
1512                         return BLK_STS_OK;
1513                 }
1514         }
1515         if (cmd->fake_timeout)
1516                 return BLK_STS_OK;
1517
1518         return null_handle_cmd(cmd, sector, nr_sectors, req_op(bd->rq));
1519 }
1520
1521 static void cleanup_queue(struct nullb_queue *nq)
1522 {
1523         kfree(nq->tag_map);
1524         kfree(nq->cmds);
1525 }
1526
1527 static void cleanup_queues(struct nullb *nullb)
1528 {
1529         int i;
1530
1531         for (i = 0; i < nullb->nr_queues; i++)
1532                 cleanup_queue(&nullb->queues[i]);
1533
1534         kfree(nullb->queues);
1535 }
1536
1537 static void null_exit_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx)
1538 {
1539         struct nullb_queue *nq = hctx->driver_data;
1540         struct nullb *nullb = nq->dev->nullb;
1541
1542         nullb->nr_queues--;
1543 }
1544
1545 static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq)
1546 {
1547         init_waitqueue_head(&nq->wait);
1548         nq->queue_depth = nullb->queue_depth;
1549         nq->dev = nullb->dev;
1550 }
1551
1552 static int null_init_hctx(struct blk_mq_hw_ctx *hctx, void *driver_data,
1553                           unsigned int hctx_idx)
1554 {
1555         struct nullb *nullb = hctx->queue->queuedata;
1556         struct nullb_queue *nq;
1557
1558 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1559         if (g_init_hctx_str[0] && should_fail(&null_init_hctx_attr, 1))
1560                 return -EFAULT;
1561 #endif
1562
1563         nq = &nullb->queues[hctx_idx];
1564         hctx->driver_data = nq;
1565         null_init_queue(nullb, nq);
1566         nullb->nr_queues++;
1567
1568         return 0;
1569 }
1570
1571 static const struct blk_mq_ops null_mq_ops = {
1572         .queue_rq       = null_queue_rq,
1573         .complete       = null_complete_rq,
1574         .timeout        = null_timeout_rq,
1575         .init_hctx      = null_init_hctx,
1576         .exit_hctx      = null_exit_hctx,
1577 };
1578
1579 static void null_del_dev(struct nullb *nullb)
1580 {
1581         struct nullb_device *dev;
1582
1583         if (!nullb)
1584                 return;
1585
1586         dev = nullb->dev;
1587
1588         ida_simple_remove(&nullb_indexes, nullb->index);
1589
1590         list_del_init(&nullb->list);
1591
1592         del_gendisk(nullb->disk);
1593
1594         if (test_bit(NULLB_DEV_FL_THROTTLED, &nullb->dev->flags)) {
1595                 hrtimer_cancel(&nullb->bw_timer);
1596                 atomic_long_set(&nullb->cur_bytes, LONG_MAX);
1597                 null_restart_queue_async(nullb);
1598         }
1599
1600         blk_cleanup_disk(nullb->disk);
1601         if (dev->queue_mode == NULL_Q_MQ &&
1602             nullb->tag_set == &nullb->__tag_set)
1603                 blk_mq_free_tag_set(nullb->tag_set);
1604         cleanup_queues(nullb);
1605         if (null_cache_active(nullb))
1606                 null_free_device_storage(nullb->dev, true);
1607         kfree(nullb);
1608         dev->nullb = NULL;
1609 }
1610
1611 static void null_config_discard(struct nullb *nullb)
1612 {
1613         if (nullb->dev->discard == false)
1614                 return;
1615
1616         if (!nullb->dev->memory_backed) {
1617                 nullb->dev->discard = false;
1618                 pr_info("discard option is ignored without memory backing\n");
1619                 return;
1620         }
1621
1622         if (nullb->dev->zoned) {
1623                 nullb->dev->discard = false;
1624                 pr_info("discard option is ignored in zoned mode\n");
1625                 return;
1626         }
1627
1628         nullb->q->limits.discard_granularity = nullb->dev->blocksize;
1629         nullb->q->limits.discard_alignment = nullb->dev->blocksize;
1630         blk_queue_max_discard_sectors(nullb->q, UINT_MAX >> 9);
1631         blk_queue_flag_set(QUEUE_FLAG_DISCARD, nullb->q);
1632 }
1633
1634 static const struct block_device_operations null_bio_ops = {
1635         .owner          = THIS_MODULE,
1636         .submit_bio     = null_submit_bio,
1637         .report_zones   = null_report_zones,
1638 };
1639
1640 static const struct block_device_operations null_rq_ops = {
1641         .owner          = THIS_MODULE,
1642         .report_zones   = null_report_zones,
1643 };
1644
1645 static int setup_commands(struct nullb_queue *nq)
1646 {
1647         struct nullb_cmd *cmd;
1648         int i, tag_size;
1649
1650         nq->cmds = kcalloc(nq->queue_depth, sizeof(*cmd), GFP_KERNEL);
1651         if (!nq->cmds)
1652                 return -ENOMEM;
1653
1654         tag_size = ALIGN(nq->queue_depth, BITS_PER_LONG) / BITS_PER_LONG;
1655         nq->tag_map = kcalloc(tag_size, sizeof(unsigned long), GFP_KERNEL);
1656         if (!nq->tag_map) {
1657                 kfree(nq->cmds);
1658                 return -ENOMEM;
1659         }
1660
1661         for (i = 0; i < nq->queue_depth; i++) {
1662                 cmd = &nq->cmds[i];
1663                 cmd->tag = -1U;
1664         }
1665
1666         return 0;
1667 }
1668
1669 static int setup_queues(struct nullb *nullb)
1670 {
1671         nullb->queues = kcalloc(nr_cpu_ids, sizeof(struct nullb_queue),
1672                                 GFP_KERNEL);
1673         if (!nullb->queues)
1674                 return -ENOMEM;
1675
1676         nullb->queue_depth = nullb->dev->hw_queue_depth;
1677
1678         return 0;
1679 }
1680
1681 static int init_driver_queues(struct nullb *nullb)
1682 {
1683         struct nullb_queue *nq;
1684         int i, ret = 0;
1685
1686         for (i = 0; i < nullb->dev->submit_queues; i++) {
1687                 nq = &nullb->queues[i];
1688
1689                 null_init_queue(nullb, nq);
1690
1691                 ret = setup_commands(nq);
1692                 if (ret)
1693                         return ret;
1694                 nullb->nr_queues++;
1695         }
1696         return 0;
1697 }
1698
1699 static int null_gendisk_register(struct nullb *nullb)
1700 {
1701         sector_t size = ((sector_t)nullb->dev->size * SZ_1M) >> SECTOR_SHIFT;
1702         struct gendisk *disk = nullb->disk;
1703
1704         set_capacity(disk, size);
1705
1706         disk->flags |= GENHD_FL_EXT_DEVT | GENHD_FL_SUPPRESS_PARTITION_INFO;
1707         disk->major             = null_major;
1708         disk->first_minor       = nullb->index;
1709         disk->minors            = 1;
1710         if (queue_is_mq(nullb->q))
1711                 disk->fops              = &null_rq_ops;
1712         else
1713                 disk->fops              = &null_bio_ops;
1714         disk->private_data      = nullb;
1715         strncpy(disk->disk_name, nullb->disk_name, DISK_NAME_LEN);
1716
1717         if (nullb->dev->zoned) {
1718                 int ret = null_register_zoned_dev(nullb);
1719
1720                 if (ret)
1721                         return ret;
1722         }
1723
1724         add_disk(disk);
1725         return 0;
1726 }
1727
1728 static int null_init_tag_set(struct nullb *nullb, struct blk_mq_tag_set *set)
1729 {
1730         set->ops = &null_mq_ops;
1731         set->nr_hw_queues = nullb ? nullb->dev->submit_queues :
1732                                                 g_submit_queues;
1733         set->queue_depth = nullb ? nullb->dev->hw_queue_depth :
1734                                                 g_hw_queue_depth;
1735         set->numa_node = nullb ? nullb->dev->home_node : g_home_node;
1736         set->cmd_size   = sizeof(struct nullb_cmd);
1737         set->flags = BLK_MQ_F_SHOULD_MERGE;
1738         if (g_no_sched)
1739                 set->flags |= BLK_MQ_F_NO_SCHED;
1740         if (g_shared_tag_bitmap)
1741                 set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1742         set->driver_data = NULL;
1743
1744         if ((nullb && nullb->dev->blocking) || g_blocking)
1745                 set->flags |= BLK_MQ_F_BLOCKING;
1746
1747         return blk_mq_alloc_tag_set(set);
1748 }
1749
1750 static int null_validate_conf(struct nullb_device *dev)
1751 {
1752         dev->blocksize = round_down(dev->blocksize, 512);
1753         dev->blocksize = clamp_t(unsigned int, dev->blocksize, 512, 4096);
1754
1755         if (dev->queue_mode == NULL_Q_MQ && dev->use_per_node_hctx) {
1756                 if (dev->submit_queues != nr_online_nodes)
1757                         dev->submit_queues = nr_online_nodes;
1758         } else if (dev->submit_queues > nr_cpu_ids)
1759                 dev->submit_queues = nr_cpu_ids;
1760         else if (dev->submit_queues == 0)
1761                 dev->submit_queues = 1;
1762
1763         dev->queue_mode = min_t(unsigned int, dev->queue_mode, NULL_Q_MQ);
1764         dev->irqmode = min_t(unsigned int, dev->irqmode, NULL_IRQ_TIMER);
1765
1766         /* Do memory allocation, so set blocking */
1767         if (dev->memory_backed)
1768                 dev->blocking = true;
1769         else /* cache is meaningless */
1770                 dev->cache_size = 0;
1771         dev->cache_size = min_t(unsigned long, ULONG_MAX / 1024 / 1024,
1772                                                 dev->cache_size);
1773         dev->mbps = min_t(unsigned int, 1024 * 40, dev->mbps);
1774         /* can not stop a queue */
1775         if (dev->queue_mode == NULL_Q_BIO)
1776                 dev->mbps = 0;
1777
1778         if (dev->zoned &&
1779             (!dev->zone_size || !is_power_of_2(dev->zone_size))) {
1780                 pr_err("zone_size must be power-of-two\n");
1781                 return -EINVAL;
1782         }
1783
1784         return 0;
1785 }
1786
1787 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1788 static bool __null_setup_fault(struct fault_attr *attr, char *str)
1789 {
1790         if (!str[0])
1791                 return true;
1792
1793         if (!setup_fault_attr(attr, str))
1794                 return false;
1795
1796         attr->verbose = 0;
1797         return true;
1798 }
1799 #endif
1800
1801 static bool null_setup_fault(void)
1802 {
1803 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1804         if (!__null_setup_fault(&null_timeout_attr, g_timeout_str))
1805                 return false;
1806         if (!__null_setup_fault(&null_requeue_attr, g_requeue_str))
1807                 return false;
1808         if (!__null_setup_fault(&null_init_hctx_attr, g_init_hctx_str))
1809                 return false;
1810 #endif
1811         return true;
1812 }
1813
1814 static int null_add_dev(struct nullb_device *dev)
1815 {
1816         struct nullb *nullb;
1817         int rv;
1818
1819         rv = null_validate_conf(dev);
1820         if (rv)
1821                 return rv;
1822
1823         nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, dev->home_node);
1824         if (!nullb) {
1825                 rv = -ENOMEM;
1826                 goto out;
1827         }
1828         nullb->dev = dev;
1829         dev->nullb = nullb;
1830
1831         spin_lock_init(&nullb->lock);
1832
1833         rv = setup_queues(nullb);
1834         if (rv)
1835                 goto out_free_nullb;
1836
1837         if (dev->queue_mode == NULL_Q_MQ) {
1838                 if (shared_tags) {
1839                         nullb->tag_set = &tag_set;
1840                         rv = 0;
1841                 } else {
1842                         nullb->tag_set = &nullb->__tag_set;
1843                         rv = null_init_tag_set(nullb, nullb->tag_set);
1844                 }
1845
1846                 if (rv)
1847                         goto out_cleanup_queues;
1848
1849                 if (!null_setup_fault())
1850                         goto out_cleanup_tags;
1851
1852                 nullb->tag_set->timeout = 5 * HZ;
1853                 nullb->disk = blk_mq_alloc_disk(nullb->tag_set, nullb);
1854                 if (IS_ERR(nullb->disk)) {
1855                         rv = PTR_ERR(nullb->disk);
1856                         goto out_cleanup_tags;
1857                 }
1858                 nullb->q = nullb->disk->queue;
1859         } else if (dev->queue_mode == NULL_Q_BIO) {
1860                 rv = -ENOMEM;
1861                 nullb->disk = blk_alloc_disk(nullb->dev->home_node);
1862                 if (!nullb->disk)
1863                         goto out_cleanup_queues;
1864
1865                 nullb->q = nullb->disk->queue;
1866                 rv = init_driver_queues(nullb);
1867                 if (rv)
1868                         goto out_cleanup_disk;
1869         }
1870
1871         if (dev->mbps) {
1872                 set_bit(NULLB_DEV_FL_THROTTLED, &dev->flags);
1873                 nullb_setup_bwtimer(nullb);
1874         }
1875
1876         if (dev->cache_size > 0) {
1877                 set_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
1878                 blk_queue_write_cache(nullb->q, true, true);
1879         }
1880
1881         if (dev->zoned) {
1882                 rv = null_init_zoned_dev(dev, nullb->q);
1883                 if (rv)
1884                         goto out_cleanup_disk;
1885         }
1886
1887         nullb->q->queuedata = nullb;
1888         blk_queue_flag_set(QUEUE_FLAG_NONROT, nullb->q);
1889         blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, nullb->q);
1890
1891         mutex_lock(&lock);
1892         nullb->index = ida_simple_get(&nullb_indexes, 0, 0, GFP_KERNEL);
1893         dev->index = nullb->index;
1894         mutex_unlock(&lock);
1895
1896         blk_queue_logical_block_size(nullb->q, dev->blocksize);
1897         blk_queue_physical_block_size(nullb->q, dev->blocksize);
1898         if (!dev->max_sectors)
1899                 dev->max_sectors = queue_max_hw_sectors(nullb->q);
1900         dev->max_sectors = min_t(unsigned int, dev->max_sectors,
1901                                  BLK_DEF_MAX_SECTORS);
1902         blk_queue_max_hw_sectors(nullb->q, dev->max_sectors);
1903
1904         if (dev->virt_boundary)
1905                 blk_queue_virt_boundary(nullb->q, PAGE_SIZE - 1);
1906
1907         null_config_discard(nullb);
1908
1909         sprintf(nullb->disk_name, "nullb%d", nullb->index);
1910
1911         rv = null_gendisk_register(nullb);
1912         if (rv)
1913                 goto out_cleanup_zone;
1914
1915         mutex_lock(&lock);
1916         list_add_tail(&nullb->list, &nullb_list);
1917         mutex_unlock(&lock);
1918
1919         return 0;
1920 out_cleanup_zone:
1921         null_free_zoned_dev(dev);
1922 out_cleanup_disk:
1923         blk_cleanup_disk(nullb->disk);
1924 out_cleanup_tags:
1925         if (dev->queue_mode == NULL_Q_MQ && nullb->tag_set == &nullb->__tag_set)
1926                 blk_mq_free_tag_set(nullb->tag_set);
1927 out_cleanup_queues:
1928         cleanup_queues(nullb);
1929 out_free_nullb:
1930         kfree(nullb);
1931         dev->nullb = NULL;
1932 out:
1933         return rv;
1934 }
1935
1936 static int __init null_init(void)
1937 {
1938         int ret = 0;
1939         unsigned int i;
1940         struct nullb *nullb;
1941         struct nullb_device *dev;
1942
1943         if (g_bs > PAGE_SIZE) {
1944                 pr_warn("invalid block size\n");
1945                 pr_warn("defaults block size to %lu\n", PAGE_SIZE);
1946                 g_bs = PAGE_SIZE;
1947         }
1948
1949         if (g_max_sectors > BLK_DEF_MAX_SECTORS) {
1950                 pr_warn("invalid max sectors\n");
1951                 pr_warn("defaults max sectors to %u\n", BLK_DEF_MAX_SECTORS);
1952                 g_max_sectors = BLK_DEF_MAX_SECTORS;
1953         }
1954
1955         if (g_home_node != NUMA_NO_NODE && g_home_node >= nr_online_nodes) {
1956                 pr_err("invalid home_node value\n");
1957                 g_home_node = NUMA_NO_NODE;
1958         }
1959
1960         if (g_queue_mode == NULL_Q_RQ) {
1961                 pr_err("legacy IO path no longer available\n");
1962                 return -EINVAL;
1963         }
1964         if (g_queue_mode == NULL_Q_MQ && g_use_per_node_hctx) {
1965                 if (g_submit_queues != nr_online_nodes) {
1966                         pr_warn("submit_queues param is set to %u.\n",
1967                                                         nr_online_nodes);
1968                         g_submit_queues = nr_online_nodes;
1969                 }
1970         } else if (g_submit_queues > nr_cpu_ids)
1971                 g_submit_queues = nr_cpu_ids;
1972         else if (g_submit_queues <= 0)
1973                 g_submit_queues = 1;
1974
1975         if (g_queue_mode == NULL_Q_MQ && shared_tags) {
1976                 ret = null_init_tag_set(NULL, &tag_set);
1977                 if (ret)
1978                         return ret;
1979         }
1980
1981         config_group_init(&nullb_subsys.su_group);
1982         mutex_init(&nullb_subsys.su_mutex);
1983
1984         ret = configfs_register_subsystem(&nullb_subsys);
1985         if (ret)
1986                 goto err_tagset;
1987
1988         mutex_init(&lock);
1989
1990         null_major = register_blkdev(0, "nullb");
1991         if (null_major < 0) {
1992                 ret = null_major;
1993                 goto err_conf;
1994         }
1995
1996         for (i = 0; i < nr_devices; i++) {
1997                 dev = null_alloc_dev();
1998                 if (!dev) {
1999                         ret = -ENOMEM;
2000                         goto err_dev;
2001                 }
2002                 ret = null_add_dev(dev);
2003                 if (ret) {
2004                         null_free_dev(dev);
2005                         goto err_dev;
2006                 }
2007         }
2008
2009         pr_info("module loaded\n");
2010         return 0;
2011
2012 err_dev:
2013         while (!list_empty(&nullb_list)) {
2014                 nullb = list_entry(nullb_list.next, struct nullb, list);
2015                 dev = nullb->dev;
2016                 null_del_dev(nullb);
2017                 null_free_dev(dev);
2018         }
2019         unregister_blkdev(null_major, "nullb");
2020 err_conf:
2021         configfs_unregister_subsystem(&nullb_subsys);
2022 err_tagset:
2023         if (g_queue_mode == NULL_Q_MQ && shared_tags)
2024                 blk_mq_free_tag_set(&tag_set);
2025         return ret;
2026 }
2027
2028 static void __exit null_exit(void)
2029 {
2030         struct nullb *nullb;
2031
2032         configfs_unregister_subsystem(&nullb_subsys);
2033
2034         unregister_blkdev(null_major, "nullb");
2035
2036         mutex_lock(&lock);
2037         while (!list_empty(&nullb_list)) {
2038                 struct nullb_device *dev;
2039
2040                 nullb = list_entry(nullb_list.next, struct nullb, list);
2041                 dev = nullb->dev;
2042                 null_del_dev(nullb);
2043                 null_free_dev(dev);
2044         }
2045         mutex_unlock(&lock);
2046
2047         if (g_queue_mode == NULL_Q_MQ && shared_tags)
2048                 blk_mq_free_tag_set(&tag_set);
2049 }
2050
2051 module_init(null_init);
2052 module_exit(null_exit);
2053
2054 MODULE_AUTHOR("Jens Axboe <axboe@kernel.dk>");
2055 MODULE_LICENSE("GPL");