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 #undef pr_fmt
  15 #define pr_fmt(fmt)     "null_blk: " fmt
  16
  17 #define FREE_BATCH              16
  18
  19 #define TICKS_PER_SEC           50ULL
  20 #define TIMER_INTERVAL          (NSEC_PER_SEC / TICKS_PER_SEC)
  21
  22 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
  23 static DECLARE_FAULT_ATTR(null_timeout_attr);
  24 static DECLARE_FAULT_ATTR(null_requeue_attr);
  25 static DECLARE_FAULT_ATTR(null_init_hctx_attr);
  26 #endif
  27
  28 static inline u64 mb_per_tick(int mbps)
  29 {
  30         return (1 << 20) / TICKS_PER_SEC * ((u64) mbps);
  31 }
  32
  33 /*
  34  * Status flags for nullb_device.
  35  *
  36  * CONFIGURED:  Device has been configured and turned on. Cannot reconfigure.
  37  * UP:          Device is currently on and visible in userspace.
  38  * THROTTLED:   Device is being throttled.
  39  * CACHE:       Device is using a write-back cache.
  40  */
  41 enum nullb_device_flags {
  42         NULLB_DEV_FL_CONFIGURED = 0,
  43         NULLB_DEV_FL_UP         = 1,
  44         NULLB_DEV_FL_THROTTLED  = 2,
  45         NULLB_DEV_FL_CACHE      = 3,
  46 };
  47
  48 #define MAP_SZ          ((PAGE_SIZE >> SECTOR_SHIFT) + 2)
  49 /*
  50  * nullb_page is a page in memory for nullb devices.
  51  *
  52  * @page:       The page holding the data.
  53  * @bitmap:     The bitmap represents which sector in the page has data.
  54  *              Each bit represents one block size. For example, sector 8
  55  *              will use the 7th bit
  56  * The highest 2 bits of bitmap are for special purpose. LOCK means the cache
  57  * page is being flushing to storage. FREE means the cache page is freed and
  58  * should be skipped from flushing to storage. Please see
  59  * null_make_cache_space
  60  */
  61 struct nullb_page {
  62         struct page *page;
  63         DECLARE_BITMAP(bitmap, MAP_SZ);
  64 };
  65 #define NULLB_PAGE_LOCK (MAP_SZ - 1)
  66 #define NULLB_PAGE_FREE (MAP_SZ - 2)
  67
  68 static LIST_HEAD(nullb_list);
  69 static struct mutex lock;
  70 static int null_major;
  71 static DEFINE_IDA(nullb_indexes);
  72 static struct blk_mq_tag_set tag_set;
  73
  74 enum {
  75         NULL_IRQ_NONE           = 0,
  76         NULL_IRQ_SOFTIRQ        = 1,
  77         NULL_IRQ_TIMER          = 2,
  78 };
  79
  80 static bool g_virt_boundary = false;
  81 module_param_named(virt_boundary, g_virt_boundary, bool, 0444);
  82 MODULE_PARM_DESC(virt_boundary, "Require a virtual boundary for the device. Default: False");
  83
  84 static int g_no_sched;
  85 module_param_named(no_sched, g_no_sched, int, 0444);
  86 MODULE_PARM_DESC(no_sched, "No io scheduler");
  87
  88 static int g_submit_queues = 1;
  89 module_param_named(submit_queues, g_submit_queues, int, 0444);
  90 MODULE_PARM_DESC(submit_queues, "Number of submission queues");
  91
  92 static int g_poll_queues = 1;
  93 module_param_named(poll_queues, g_poll_queues, int, 0444);
  94 MODULE_PARM_DESC(poll_queues, "Number of IOPOLL submission queues");
  95
  96 static int g_home_node = NUMA_NO_NODE;
  97 module_param_named(home_node, g_home_node, int, 0444);
  98 MODULE_PARM_DESC(home_node, "Home node for the device");
  99
 100 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
 101 /*
 102  * For more details about fault injection, please refer to
 103  * Documentation/fault-injection/fault-injection.rst.
 104  */
 105 static char g_timeout_str[80];
 106 module_param_string(timeout, g_timeout_str, sizeof(g_timeout_str), 0444);
 107 MODULE_PARM_DESC(timeout, "Fault injection. timeout=<interval>,<probability>,<space>,<times>");
 108
 109 static char g_requeue_str[80];
 110 module_param_string(requeue, g_requeue_str, sizeof(g_requeue_str), 0444);
 111 MODULE_PARM_DESC(requeue, "Fault injection. requeue=<interval>,<probability>,<space>,<times>");
 112
 113 static char g_init_hctx_str[80];
 114 module_param_string(init_hctx, g_init_hctx_str, sizeof(g_init_hctx_str), 0444);
 115 MODULE_PARM_DESC(init_hctx, "Fault injection to fail hctx init. init_hctx=<interval>,<probability>,<space>,<times>");
 116 #endif
 117
 118 static int g_queue_mode = NULL_Q_MQ;
 119
 120 static int null_param_store_val(const char *str, int *val, int min, int max)
 121 {
 122         int ret, new_val;
 123
 124         ret = kstrtoint(str, 10, &new_val);
 125         if (ret)
 126                 return -EINVAL;
 127
 128         if (new_val < min || new_val > max)
 129                 return -EINVAL;
 130
 131         *val = new_val;
 132         return 0;
 133 }
 134
 135 static int null_set_queue_mode(const char *str, const struct kernel_param *kp)
 136 {
 137         return null_param_store_val(str, &g_queue_mode, NULL_Q_BIO, NULL_Q_MQ);
 138 }
 139
 140 static const struct kernel_param_ops null_queue_mode_param_ops = {
 141         .set    = null_set_queue_mode,
 142         .get    = param_get_int,
 143 };
 144
 145 device_param_cb(queue_mode, &null_queue_mode_param_ops, &g_queue_mode, 0444);
 146 MODULE_PARM_DESC(queue_mode, "Block interface to use (0=bio,1=rq,2=multiqueue)");
 147
 148 static int g_gb = 250;
 149 module_param_named(gb, g_gb, int, 0444);
 150 MODULE_PARM_DESC(gb, "Size in GB");
 151
 152 static int g_bs = 512;
 153 module_param_named(bs, g_bs, int, 0444);
 154 MODULE_PARM_DESC(bs, "Block size (in bytes)");
 155
 156 static int g_max_sectors;
 157 module_param_named(max_sectors, g_max_sectors, int, 0444);
 158 MODULE_PARM_DESC(max_sectors, "Maximum size of a command (in 512B sectors)");
 159
 160 static unsigned int nr_devices = 1;
 161 module_param(nr_devices, uint, 0444);
 162 MODULE_PARM_DESC(nr_devices, "Number of devices to register");
 163
 164 static bool g_blocking;
 165 module_param_named(blocking, g_blocking, bool, 0444);
 166 MODULE_PARM_DESC(blocking, "Register as a blocking blk-mq driver device");
 167
 168 static bool shared_tags;
 169 module_param(shared_tags, bool, 0444);
 170 MODULE_PARM_DESC(shared_tags, "Share tag set between devices for blk-mq");
 171
 172 static bool g_shared_tag_bitmap;
 173 module_param_named(shared_tag_bitmap, g_shared_tag_bitmap, bool, 0444);
 174 MODULE_PARM_DESC(shared_tag_bitmap, "Use shared tag bitmap for all submission queues for blk-mq");
 175
 176 static int g_irqmode = NULL_IRQ_SOFTIRQ;
 177
 178 static int null_set_irqmode(const char *str, const struct kernel_param *kp)
 179 {
 180         return null_param_store_val(str, &g_irqmode, NULL_IRQ_NONE,
 181                                         NULL_IRQ_TIMER);
 182 }
 183
 184 static const struct kernel_param_ops null_irqmode_param_ops = {
 185         .set    = null_set_irqmode,
 186         .get    = param_get_int,
 187 };
 188
 189 device_param_cb(irqmode, &null_irqmode_param_ops, &g_irqmode, 0444);
 190 MODULE_PARM_DESC(irqmode, "IRQ completion handler. 0-none, 1-softirq, 2-timer");
 191
 192 static unsigned long g_completion_nsec = 10000;
 193 module_param_named(completion_nsec, g_completion_nsec, ulong, 0444);
 194 MODULE_PARM_DESC(completion_nsec, "Time in ns to complete a request in hardware. Default: 10,000ns");
 195
 196 static int g_hw_queue_depth = 64;
 197 module_param_named(hw_queue_depth, g_hw_queue_depth, int, 0444);
 198 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 64");
 199
 200 static bool g_use_per_node_hctx;
 201 module_param_named(use_per_node_hctx, g_use_per_node_hctx, bool, 0444);
 202 MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware context queues. Default: false");
 203
 204 static bool g_zoned;
 205 module_param_named(zoned, g_zoned, bool, S_IRUGO);
 206 MODULE_PARM_DESC(zoned, "Make device as a host-managed zoned block device. Default: false");
 207
 208 static unsigned long g_zone_size = 256;
 209 module_param_named(zone_size, g_zone_size, ulong, S_IRUGO);
 210 MODULE_PARM_DESC(zone_size, "Zone size in MB when block device is zoned. Must be power-of-two: Default: 256");
 211
 212 static unsigned long g_zone_capacity;
 213 module_param_named(zone_capacity, g_zone_capacity, ulong, 0444);
 214 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");
 215
 216 static unsigned int g_zone_nr_conv;
 217 module_param_named(zone_nr_conv, g_zone_nr_conv, uint, 0444);
 218 MODULE_PARM_DESC(zone_nr_conv, "Number of conventional zones when block device is zoned. Default: 0");
 219
 220 static unsigned int g_zone_max_open;
 221 module_param_named(zone_max_open, g_zone_max_open, uint, 0444);
 222 MODULE_PARM_DESC(zone_max_open, "Maximum number of open zones when block device is zoned. Default: 0 (no limit)");
 223
 224 static unsigned int g_zone_max_active;
 225 module_param_named(zone_max_active, g_zone_max_active, uint, 0444);
 226 MODULE_PARM_DESC(zone_max_active, "Maximum number of active zones when block device is zoned. Default: 0 (no limit)");
 227
 228 static struct nullb_device *null_alloc_dev(void);
 229 static void null_free_dev(struct nullb_device *dev);
 230 static void null_del_dev(struct nullb *nullb);
 231 static int null_add_dev(struct nullb_device *dev);
 232 static struct nullb *null_find_dev_by_name(const char *name);
 233 static void null_free_device_storage(struct nullb_device *dev, bool is_cache);
 234
 235 static inline struct nullb_device *to_nullb_device(struct config_item *item)
 236 {
 237         return item ? container_of(item, struct nullb_device, item) : NULL;
 238 }
 239
 240 static inline ssize_t nullb_device_uint_attr_show(unsigned int val, char *page)
 241 {
 242         return snprintf(page, PAGE_SIZE, "%u\n", val);
 243 }
 244
 245 static inline ssize_t nullb_device_ulong_attr_show(unsigned long val,
 246         char *page)
 247 {
 248         return snprintf(page, PAGE_SIZE, "%lu\n", val);
 249 }
 250
 251 static inline ssize_t nullb_device_bool_attr_show(bool val, char *page)
 252 {
 253         return snprintf(page, PAGE_SIZE, "%u\n", val);
 254 }
 255
 256 static ssize_t nullb_device_uint_attr_store(unsigned int *val,
 257         const char *page, size_t count)
 258 {
 259         unsigned int tmp;
 260         int result;
 261
 262         result = kstrtouint(page, 0, &tmp);
 263         if (result < 0)
 264                 return result;
 265
 266         *val = tmp;
 267         return count;
 268 }
 269
 270 static ssize_t nullb_device_ulong_attr_store(unsigned long *val,
 271         const char *page, size_t count)
 272 {
 273         int result;
 274         unsigned long tmp;
 275
 276         result = kstrtoul(page, 0, &tmp);
 277         if (result < 0)
 278                 return result;
 279
 280         *val = tmp;
 281         return count;
 282 }
 283
 284 static ssize_t nullb_device_bool_attr_store(bool *val, const char *page,
 285         size_t count)
 286 {
 287         bool tmp;
 288         int result;
 289
 290         result = kstrtobool(page,  &tmp);
 291         if (result < 0)
 292                 return result;
 293
 294         *val = tmp;
 295         return count;
 296 }
 297
 298 /* The following macro should only be used with TYPE = {uint, ulong, bool}. */
 299 #define NULLB_DEVICE_ATTR(NAME, TYPE, APPLY)                            \
 300 static ssize_t                                                          \
 301 nullb_device_##NAME##_show(struct config_item *item, char *page)        \
 302 {                                                                       \
 303         return nullb_device_##TYPE##_attr_show(                         \
 304                                 to_nullb_device(item)->NAME, page);     \
 305 }                                                                       \
 306 static ssize_t                                                          \
 307 nullb_device_##NAME##_store(struct config_item *item, const char *page, \
 308                             size_t count)                               \
 309 {                                                                       \
 310         int (*apply_fn)(struct nullb_device *dev, TYPE new_value) = APPLY;\
 311         struct nullb_device *dev = to_nullb_device(item);               \
 312         TYPE new_value = 0;                                             \
 313         int ret;                                                        \
 314                                                                         \
 315         ret = nullb_device_##TYPE##_attr_store(&new_value, page, count);\
 316         if (ret < 0)                                                    \
 317                 return ret;                                             \
 318         if (apply_fn)                                                   \
 319                 ret = apply_fn(dev, new_value);                         \
 320         else if (test_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags))        \
 321                 ret = -EBUSY;                                           \
 322         if (ret < 0)                                                    \
 323                 return ret;                                             \
 324         dev->NAME = new_value;                                          \
 325         return count;                                                   \
 326 }                                                                       \
 327 CONFIGFS_ATTR(nullb_device_, NAME);
 328
 329 static int nullb_update_nr_hw_queues(struct nullb_device *dev,
 330                                      unsigned int submit_queues,
 331                                      unsigned int poll_queues)
 332
 333 {
 334         struct blk_mq_tag_set *set;
 335         int ret, nr_hw_queues;
 336
 337         if (!dev->nullb)
 338                 return 0;
 339
 340         /*
 341          * Make sure at least one submit queue exists.
 342          */
 343         if (!submit_queues)
 344                 return -EINVAL;
 345
 346         /*
 347          * Make sure that null_init_hctx() does not access nullb->queues[] past
 348          * the end of that array.
 349          */
 350         if (submit_queues > nr_cpu_ids || poll_queues > g_poll_queues)
 351                 return -EINVAL;
 352
 353         /*
 354          * Keep previous and new queue numbers in nullb_device for reference in
 355          * the call back function null_map_queues().
 356          */
 357         dev->prev_submit_queues = dev->submit_queues;
 358         dev->prev_poll_queues = dev->poll_queues;
 359         dev->submit_queues = submit_queues;
 360         dev->poll_queues = poll_queues;
 361
 362         set = dev->nullb->tag_set;
 363         nr_hw_queues = submit_queues + poll_queues;
 364         blk_mq_update_nr_hw_queues(set, nr_hw_queues);
 365         ret = set->nr_hw_queues == nr_hw_queues ? 0 : -ENOMEM;
 366
 367         if (ret) {
 368                 /* on error, revert the queue numbers */
 369                 dev->submit_queues = dev->prev_submit_queues;
 370                 dev->poll_queues = dev->prev_poll_queues;
 371         }
 372
 373         return ret;
 374 }
 375
 376 static int nullb_apply_submit_queues(struct nullb_device *dev,
 377                                      unsigned int submit_queues)
 378 {
 379         return nullb_update_nr_hw_queues(dev, submit_queues, dev->poll_queues);
 380 }
 381
 382 static int nullb_apply_poll_queues(struct nullb_device *dev,
 383                                    unsigned int poll_queues)
 384 {
 385         return nullb_update_nr_hw_queues(dev, dev->submit_queues, poll_queues);
 386 }
 387
 388 NULLB_DEVICE_ATTR(size, ulong, NULL);
 389 NULLB_DEVICE_ATTR(completion_nsec, ulong, NULL);
 390 NULLB_DEVICE_ATTR(submit_queues, uint, nullb_apply_submit_queues);
 391 NULLB_DEVICE_ATTR(poll_queues, uint, nullb_apply_poll_queues);
 392 NULLB_DEVICE_ATTR(home_node, uint, NULL);
 393 NULLB_DEVICE_ATTR(queue_mode, uint, NULL);
 394 NULLB_DEVICE_ATTR(blocksize, uint, NULL);
 395 NULLB_DEVICE_ATTR(max_sectors, uint, NULL);
 396 NULLB_DEVICE_ATTR(irqmode, uint, NULL);
 397 NULLB_DEVICE_ATTR(hw_queue_depth, uint, NULL);
 398 NULLB_DEVICE_ATTR(index, uint, NULL);
 399 NULLB_DEVICE_ATTR(blocking, bool, NULL);
 400 NULLB_DEVICE_ATTR(use_per_node_hctx, bool, NULL);
 401 NULLB_DEVICE_ATTR(memory_backed, bool, NULL);
 402 NULLB_DEVICE_ATTR(discard, bool, NULL);
 403 NULLB_DEVICE_ATTR(mbps, uint, NULL);
 404 NULLB_DEVICE_ATTR(cache_size, ulong, NULL);
 405 NULLB_DEVICE_ATTR(zoned, bool, NULL);
 406 NULLB_DEVICE_ATTR(zone_size, ulong, NULL);
 407 NULLB_DEVICE_ATTR(zone_capacity, ulong, NULL);
 408 NULLB_DEVICE_ATTR(zone_nr_conv, uint, NULL);
 409 NULLB_DEVICE_ATTR(zone_max_open, uint, NULL);
 410 NULLB_DEVICE_ATTR(zone_max_active, uint, NULL);
 411 NULLB_DEVICE_ATTR(virt_boundary, bool, NULL);
 412
 413 static ssize_t nullb_device_power_show(struct config_item *item, char *page)
 414 {
 415         return nullb_device_bool_attr_show(to_nullb_device(item)->power, page);
 416 }
 417
 418 static ssize_t nullb_device_power_store(struct config_item *item,
 419                                      const char *page, size_t count)
 420 {
 421         struct nullb_device *dev = to_nullb_device(item);
 422         bool newp = false;
 423         ssize_t ret;
 424
 425         ret = nullb_device_bool_attr_store(&newp, page, count);
 426         if (ret < 0)
 427                 return ret;
 428
 429         if (!dev->power && newp) {
 430                 if (test_and_set_bit(NULLB_DEV_FL_UP, &dev->flags))
 431                         return count;
 432                 ret = null_add_dev(dev);
 433                 if (ret) {
 434                         clear_bit(NULLB_DEV_FL_UP, &dev->flags);
 435                         return ret;
 436                 }
 437
 438                 set_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
 439                 dev->power = newp;
 440         } else if (dev->power && !newp) {
 441                 if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
 442                         mutex_lock(&lock);
 443                         dev->power = newp;
 444                         null_del_dev(dev->nullb);
 445                         mutex_unlock(&lock);
 446                 }
 447                 clear_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
 448         }
 449
 450         return count;
 451 }
 452
 453 CONFIGFS_ATTR(nullb_device_, power);
 454
 455 static ssize_t nullb_device_badblocks_show(struct config_item *item, char *page)
 456 {
 457         struct nullb_device *t_dev = to_nullb_device(item);
 458
 459         return badblocks_show(&t_dev->badblocks, page, 0);
 460 }
 461
 462 static ssize_t nullb_device_badblocks_store(struct config_item *item,
 463                                      const char *page, size_t count)
 464 {
 465         struct nullb_device *t_dev = to_nullb_device(item);
 466         char *orig, *buf, *tmp;
 467         u64 start, end;
 468         int ret;
 469
 470         orig = kstrndup(page, count, GFP_KERNEL);
 471         if (!orig)
 472                 return -ENOMEM;
 473
 474         buf = strstrip(orig);
 475
 476         ret = -EINVAL;
 477         if (buf[0] != '+' && buf[0] != '-')
 478                 goto out;
 479         tmp = strchr(&buf[1], '-');
 480         if (!tmp)
 481                 goto out;
 482         *tmp = '\0';
 483         ret = kstrtoull(buf + 1, 0, &start);
 484         if (ret)
 485                 goto out;
 486         ret = kstrtoull(tmp + 1, 0, &end);
 487         if (ret)
 488                 goto out;
 489         ret = -EINVAL;
 490         if (start > end)
 491                 goto out;
 492         /* enable badblocks */
 493         cmpxchg(&t_dev->badblocks.shift, -1, 0);
 494         if (buf[0] == '+')
 495                 ret = badblocks_set(&t_dev->badblocks, start,
 496                         end - start + 1, 1);
 497         else
 498                 ret = badblocks_clear(&t_dev->badblocks, start,
 499                         end - start + 1);
 500         if (ret == 0)
 501                 ret = count;
 502 out:
 503         kfree(orig);
 504         return ret;
 505 }
 506 CONFIGFS_ATTR(nullb_device_, badblocks);
 507
 508 static struct configfs_attribute *nullb_device_attrs[] = {
 509         &nullb_device_attr_size,
 510         &nullb_device_attr_completion_nsec,
 511         &nullb_device_attr_submit_queues,
 512         &nullb_device_attr_poll_queues,
 513         &nullb_device_attr_home_node,
 514         &nullb_device_attr_queue_mode,
 515         &nullb_device_attr_blocksize,
 516         &nullb_device_attr_max_sectors,
 517         &nullb_device_attr_irqmode,
 518         &nullb_device_attr_hw_queue_depth,
 519         &nullb_device_attr_index,
 520         &nullb_device_attr_blocking,
 521         &nullb_device_attr_use_per_node_hctx,
 522         &nullb_device_attr_power,
 523         &nullb_device_attr_memory_backed,
 524         &nullb_device_attr_discard,
 525         &nullb_device_attr_mbps,
 526         &nullb_device_attr_cache_size,
 527         &nullb_device_attr_badblocks,
 528         &nullb_device_attr_zoned,
 529         &nullb_device_attr_zone_size,
 530         &nullb_device_attr_zone_capacity,
 531         &nullb_device_attr_zone_nr_conv,
 532         &nullb_device_attr_zone_max_open,
 533         &nullb_device_attr_zone_max_active,
 534         &nullb_device_attr_virt_boundary,
 535         NULL,
 536 };
 537
 538 static void nullb_device_release(struct config_item *item)
 539 {
 540         struct nullb_device *dev = to_nullb_device(item);
 541
 542         null_free_device_storage(dev, false);
 543         null_free_dev(dev);
 544 }
 545
 546 static struct configfs_item_operations nullb_device_ops = {
 547         .release        = nullb_device_release,
 548 };
 549
 550 static const struct config_item_type nullb_device_type = {
 551         .ct_item_ops    = &nullb_device_ops,
 552         .ct_attrs       = nullb_device_attrs,
 553         .ct_owner       = THIS_MODULE,
 554 };
 555
 556 static struct
 557 config_item *nullb_group_make_item(struct config_group *group, const char *name)
 558 {
 559         struct nullb_device *dev;
 560
 561         if (null_find_dev_by_name(name))
 562                 return ERR_PTR(-EEXIST);
 563
 564         dev = null_alloc_dev();
 565         if (!dev)
 566                 return ERR_PTR(-ENOMEM);
 567
 568         config_item_init_type_name(&dev->item, name, &nullb_device_type);
 569
 570         return &dev->item;
 571 }
 572
 573 static void
 574 nullb_group_drop_item(struct config_group *group, struct config_item *item)
 575 {
 576         struct nullb_device *dev = to_nullb_device(item);
 577
 578         if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
 579                 mutex_lock(&lock);
 580                 dev->power = false;
 581                 null_del_dev(dev->nullb);
 582                 mutex_unlock(&lock);
 583         }
 584
 585         config_item_put(item);
 586 }
 587
 588 static ssize_t memb_group_features_show(struct config_item *item, char *page)
 589 {
 590         return snprintf(page, PAGE_SIZE,
 591                         "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");
 592 }
 593
 594 CONFIGFS_ATTR_RO(memb_group_, features);
 595
 596 static struct configfs_attribute *nullb_group_attrs[] = {
 597         &memb_group_attr_features,
 598         NULL,
 599 };
 600
 601 static struct configfs_group_operations nullb_group_ops = {
 602         .make_item      = nullb_group_make_item,
 603         .drop_item      = nullb_group_drop_item,
 604 };
 605
 606 static const struct config_item_type nullb_group_type = {
 607         .ct_group_ops   = &nullb_group_ops,
 608         .ct_attrs       = nullb_group_attrs,
 609         .ct_owner       = THIS_MODULE,
 610 };
 611
 612 static struct configfs_subsystem nullb_subsys = {
 613         .su_group = {
 614                 .cg_item = {
 615                         .ci_namebuf = "nullb",
 616                         .ci_type = &nullb_group_type,
 617                 },
 618         },
 619 };
 620
 621 static inline int null_cache_active(struct nullb *nullb)
 622 {
 623         return test_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
 624 }
 625
 626 static struct nullb_device *null_alloc_dev(void)
 627 {
 628         struct nullb_device *dev;
 629
 630         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
 631         if (!dev)
 632                 return NULL;
 633         INIT_RADIX_TREE(&dev->data, GFP_ATOMIC);
 634         INIT_RADIX_TREE(&dev->cache, GFP_ATOMIC);
 635         if (badblocks_init(&dev->badblocks, 0)) {
 636                 kfree(dev);
 637                 return NULL;
 638         }
 639
 640         dev->size = g_gb * 1024;
 641         dev->completion_nsec = g_completion_nsec;
 642         dev->submit_queues = g_submit_queues;
 643         dev->prev_submit_queues = g_submit_queues;
 644         dev->poll_queues = g_poll_queues;
 645         dev->prev_poll_queues = g_poll_queues;
 646         dev->home_node = g_home_node;
 647         dev->queue_mode = g_queue_mode;
 648         dev->blocksize = g_bs;
 649         dev->max_sectors = g_max_sectors;
 650         dev->irqmode = g_irqmode;
 651         dev->hw_queue_depth = g_hw_queue_depth;
 652         dev->blocking = g_blocking;
 653         dev->use_per_node_hctx = g_use_per_node_hctx;
 654         dev->zoned = g_zoned;
 655         dev->zone_size = g_zone_size;
 656         dev->zone_capacity = g_zone_capacity;
 657         dev->zone_nr_conv = g_zone_nr_conv;
 658         dev->zone_max_open = g_zone_max_open;
 659         dev->zone_max_active = g_zone_max_active;
 660         dev->virt_boundary = g_virt_boundary;
 661         return dev;
 662 }
 663
 664 static void null_free_dev(struct nullb_device *dev)
 665 {
 666         if (!dev)
 667                 return;
 668
 669         null_free_zoned_dev(dev);
 670         badblocks_exit(&dev->badblocks);
 671         kfree(dev);
 672 }
 673
 674 static void put_tag(struct nullb_queue *nq, unsigned int tag)
 675 {
 676         clear_bit_unlock(tag, nq->tag_map);
 677
 678         if (waitqueue_active(&nq->wait))
 679                 wake_up(&nq->wait);
 680 }
 681
 682 static unsigned int get_tag(struct nullb_queue *nq)
 683 {
 684         unsigned int tag;
 685
 686         do {
 687                 tag = find_first_zero_bit(nq->tag_map, nq->queue_depth);
 688                 if (tag >= nq->queue_depth)
 689                         return -1U;
 690         } while (test_and_set_bit_lock(tag, nq->tag_map));
 691
 692         return tag;
 693 }
 694
 695 static void free_cmd(struct nullb_cmd *cmd)
 696 {
 697         put_tag(cmd->nq, cmd->tag);
 698 }
 699
 700 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer);
 701
 702 static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq)
 703 {
 704         struct nullb_cmd *cmd;
 705         unsigned int tag;
 706
 707         tag = get_tag(nq);
 708         if (tag != -1U) {
 709                 cmd = &nq->cmds[tag];
 710                 cmd->tag = tag;
 711                 cmd->error = BLK_STS_OK;
 712                 cmd->nq = nq;
 713                 if (nq->dev->irqmode == NULL_IRQ_TIMER) {
 714                         hrtimer_init(&cmd->timer, CLOCK_MONOTONIC,
 715                                      HRTIMER_MODE_REL);
 716                         cmd->timer.function = null_cmd_timer_expired;
 717                 }
 718                 return cmd;
 719         }
 720
 721         return NULL;
 722 }
 723
 724 static struct nullb_cmd *alloc_cmd(struct nullb_queue *nq, struct bio *bio)
 725 {
 726         struct nullb_cmd *cmd;
 727         DEFINE_WAIT(wait);
 728
 729         do {
 730                 /*
 731                  * This avoids multiple return statements, multiple calls to
 732                  * __alloc_cmd() and a fast path call to prepare_to_wait().
 733                  */
 734                 cmd = __alloc_cmd(nq);
 735                 if (cmd) {
 736                         cmd->bio = bio;
 737                         return cmd;
 738                 }
 739                 prepare_to_wait(&nq->wait, &wait, TASK_UNINTERRUPTIBLE);
 740                 io_schedule();
 741                 finish_wait(&nq->wait, &wait);
 742         } while (1);
 743 }
 744
 745 static void end_cmd(struct nullb_cmd *cmd)
 746 {
 747         int queue_mode = cmd->nq->dev->queue_mode;
 748
 749         switch (queue_mode)  {
 750         case NULL_Q_MQ:
 751                 blk_mq_end_request(cmd->rq, cmd->error);
 752                 return;
 753         case NULL_Q_BIO:
 754                 cmd->bio->bi_status = cmd->error;
 755                 bio_endio(cmd->bio);
 756                 break;
 757         }
 758
 759         free_cmd(cmd);
 760 }
 761
 762 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
 763 {
 764         end_cmd(container_of(timer, struct nullb_cmd, timer));
 765
 766         return HRTIMER_NORESTART;
 767 }
 768
 769 static void null_cmd_end_timer(struct nullb_cmd *cmd)
 770 {
 771         ktime_t kt = cmd->nq->dev->completion_nsec;
 772
 773         hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL);
 774 }
 775
 776 static void null_complete_rq(struct request *rq)
 777 {
 778         end_cmd(blk_mq_rq_to_pdu(rq));
 779 }
 780
 781 static struct nullb_page *null_alloc_page(void)
 782 {
 783         struct nullb_page *t_page;
 784
 785         t_page = kmalloc(sizeof(struct nullb_page), GFP_NOIO);
 786         if (!t_page)
 787                 return NULL;
 788
 789         t_page->page = alloc_pages(GFP_NOIO, 0);
 790         if (!t_page->page) {
 791                 kfree(t_page);
 792                 return NULL;
 793         }
 794
 795         memset(t_page->bitmap, 0, sizeof(t_page->bitmap));
 796         return t_page;
 797 }
 798
 799 static void null_free_page(struct nullb_page *t_page)
 800 {
 801         __set_bit(NULLB_PAGE_FREE, t_page->bitmap);
 802         if (test_bit(NULLB_PAGE_LOCK, t_page->bitmap))
 803                 return;
 804         __free_page(t_page->page);
 805         kfree(t_page);
 806 }
 807
 808 static bool null_page_empty(struct nullb_page *page)
 809 {
 810         int size = MAP_SZ - 2;
 811
 812         return find_first_bit(page->bitmap, size) == size;
 813 }
 814
 815 static void null_free_sector(struct nullb *nullb, sector_t sector,
 816         bool is_cache)
 817 {
 818         unsigned int sector_bit;
 819         u64 idx;
 820         struct nullb_page *t_page, *ret;
 821         struct radix_tree_root *root;
 822
 823         root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
 824         idx = sector >> PAGE_SECTORS_SHIFT;
 825         sector_bit = (sector & SECTOR_MASK);
 826
 827         t_page = radix_tree_lookup(root, idx);
 828         if (t_page) {
 829                 __clear_bit(sector_bit, t_page->bitmap);
 830
 831                 if (null_page_empty(t_page)) {
 832                         ret = radix_tree_delete_item(root, idx, t_page);
 833                         WARN_ON(ret != t_page);
 834                         null_free_page(ret);
 835                         if (is_cache)
 836                                 nullb->dev->curr_cache -= PAGE_SIZE;
 837                 }
 838         }
 839 }
 840
 841 static struct nullb_page *null_radix_tree_insert(struct nullb *nullb, u64 idx,
 842         struct nullb_page *t_page, bool is_cache)
 843 {
 844         struct radix_tree_root *root;
 845
 846         root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
 847
 848         if (radix_tree_insert(root, idx, t_page)) {
 849                 null_free_page(t_page);
 850                 t_page = radix_tree_lookup(root, idx);
 851                 WARN_ON(!t_page || t_page->page->index != idx);
 852         } else if (is_cache)
 853                 nullb->dev->curr_cache += PAGE_SIZE;
 854
 855         return t_page;
 856 }
 857
 858 static void null_free_device_storage(struct nullb_device *dev, bool is_cache)
 859 {
 860         unsigned long pos = 0;
 861         int nr_pages;
 862         struct nullb_page *ret, *t_pages[FREE_BATCH];
 863         struct radix_tree_root *root;
 864
 865         root = is_cache ? &dev->cache : &dev->data;
 866
 867         do {
 868                 int i;
 869
 870                 nr_pages = radix_tree_gang_lookup(root,
 871                                 (void **)t_pages, pos, FREE_BATCH);
 872
 873                 for (i = 0; i < nr_pages; i++) {
 874                         pos = t_pages[i]->page->index;
 875                         ret = radix_tree_delete_item(root, pos, t_pages[i]);
 876                         WARN_ON(ret != t_pages[i]);
 877                         null_free_page(ret);
 878                 }
 879
 880                 pos++;
 881         } while (nr_pages == FREE_BATCH);
 882
 883         if (is_cache)
 884                 dev->curr_cache = 0;
 885 }
 886
 887 static struct nullb_page *__null_lookup_page(struct nullb *nullb,
 888         sector_t sector, bool for_write, bool is_cache)
 889 {
 890         unsigned int sector_bit;
 891         u64 idx;
 892         struct nullb_page *t_page;
 893         struct radix_tree_root *root;
 894
 895         idx = sector >> PAGE_SECTORS_SHIFT;
 896         sector_bit = (sector & SECTOR_MASK);
 897
 898         root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
 899         t_page = radix_tree_lookup(root, idx);
 900         WARN_ON(t_page && t_page->page->index != idx);
 901
 902         if (t_page && (for_write || test_bit(sector_bit, t_page->bitmap)))
 903                 return t_page;
 904
 905         return NULL;
 906 }
 907
 908 static struct nullb_page *null_lookup_page(struct nullb *nullb,
 909         sector_t sector, bool for_write, bool ignore_cache)
 910 {
 911         struct nullb_page *page = NULL;
 912
 913         if (!ignore_cache)
 914                 page = __null_lookup_page(nullb, sector, for_write, true);
 915         if (page)
 916                 return page;
 917         return __null_lookup_page(nullb, sector, for_write, false);
 918 }
 919
 920 static struct nullb_page *null_insert_page(struct nullb *nullb,
 921                                            sector_t sector, bool ignore_cache)
 922         __releases(&nullb->lock)
 923         __acquires(&nullb->lock)
 924 {
 925         u64 idx;
 926         struct nullb_page *t_page;
 927
 928         t_page = null_lookup_page(nullb, sector, true, ignore_cache);
 929         if (t_page)
 930                 return t_page;
 931
 932         spin_unlock_irq(&nullb->lock);
 933
 934         t_page = null_alloc_page();
 935         if (!t_page)
 936                 goto out_lock;
 937
 938         if (radix_tree_preload(GFP_NOIO))
 939                 goto out_freepage;
 940
 941         spin_lock_irq(&nullb->lock);
 942         idx = sector >> PAGE_SECTORS_SHIFT;
 943         t_page->page->index = idx;
 944         t_page = null_radix_tree_insert(nullb, idx, t_page, !ignore_cache);
 945         radix_tree_preload_end();
 946
 947         return t_page;
 948 out_freepage:
 949         null_free_page(t_page);
 950 out_lock:
 951         spin_lock_irq(&nullb->lock);
 952         return null_lookup_page(nullb, sector, true, ignore_cache);
 953 }
 954
 955 static int null_flush_cache_page(struct nullb *nullb, struct nullb_page *c_page)
 956 {
 957         int i;
 958         unsigned int offset;
 959         u64 idx;
 960         struct nullb_page *t_page, *ret;
 961         void *dst, *src;
 962
 963         idx = c_page->page->index;
 964
 965         t_page = null_insert_page(nullb, idx << PAGE_SECTORS_SHIFT, true);
 966
 967         __clear_bit(NULLB_PAGE_LOCK, c_page->bitmap);
 968         if (test_bit(NULLB_PAGE_FREE, c_page->bitmap)) {
 969                 null_free_page(c_page);
 970                 if (t_page && null_page_empty(t_page)) {
 971                         ret = radix_tree_delete_item(&nullb->dev->data,
 972                                 idx, t_page);
 973                         null_free_page(t_page);
 974                 }
 975                 return 0;
 976         }
 977
 978         if (!t_page)
 979                 return -ENOMEM;
 980
 981         src = kmap_atomic(c_page->page);
 982         dst = kmap_atomic(t_page->page);
 983
 984         for (i = 0; i < PAGE_SECTORS;
 985                         i += (nullb->dev->blocksize >> SECTOR_SHIFT)) {
 986                 if (test_bit(i, c_page->bitmap)) {
 987                         offset = (i << SECTOR_SHIFT);
 988                         memcpy(dst + offset, src + offset,
 989                                 nullb->dev->blocksize);
 990                         __set_bit(i, t_page->bitmap);
 991                 }
 992         }
 993
 994         kunmap_atomic(dst);
 995         kunmap_atomic(src);
 996
 997         ret = radix_tree_delete_item(&nullb->dev->cache, idx, c_page);
 998         null_free_page(ret);
 999         nullb->dev->curr_cache -= PAGE_SIZE;
1000
1001         return 0;
1002 }
1003
1004 static int null_make_cache_space(struct nullb *nullb, unsigned long n)
1005 {
1006         int i, err, nr_pages;
1007         struct nullb_page *c_pages[FREE_BATCH];
1008         unsigned long flushed = 0, one_round;
1009
1010 again:
1011         if ((nullb->dev->cache_size * 1024 * 1024) >
1012              nullb->dev->curr_cache + n || nullb->dev->curr_cache == 0)
1013                 return 0;
1014
1015         nr_pages = radix_tree_gang_lookup(&nullb->dev->cache,
1016                         (void **)c_pages, nullb->cache_flush_pos, FREE_BATCH);
1017         /*
1018          * nullb_flush_cache_page could unlock before using the c_pages. To
1019          * avoid race, we don't allow page free
1020          */
1021         for (i = 0; i < nr_pages; i++) {
1022                 nullb->cache_flush_pos = c_pages[i]->page->index;
1023                 /*
1024                  * We found the page which is being flushed to disk by other
1025                  * threads
1026                  */
1027                 if (test_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap))
1028                         c_pages[i] = NULL;
1029                 else
1030                         __set_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap);
1031         }
1032
1033         one_round = 0;
1034         for (i = 0; i < nr_pages; i++) {
1035                 if (c_pages[i] == NULL)
1036                         continue;
1037                 err = null_flush_cache_page(nullb, c_pages[i]);
1038                 if (err)
1039                         return err;
1040                 one_round++;
1041         }
1042         flushed += one_round << PAGE_SHIFT;
1043
1044         if (n > flushed) {
1045                 if (nr_pages == 0)
1046                         nullb->cache_flush_pos = 0;
1047                 if (one_round == 0) {
1048                         /* give other threads a chance */
1049                         spin_unlock_irq(&nullb->lock);
1050                         spin_lock_irq(&nullb->lock);
1051                 }
1052                 goto again;
1053         }
1054         return 0;
1055 }
1056
1057 static int copy_to_nullb(struct nullb *nullb, struct page *source,
1058         unsigned int off, sector_t sector, size_t n, bool is_fua)
1059 {
1060         size_t temp, count = 0;
1061         unsigned int offset;
1062         struct nullb_page *t_page;
1063         void *dst, *src;
1064
1065         while (count < n) {
1066                 temp = min_t(size_t, nullb->dev->blocksize, n - count);
1067
1068                 if (null_cache_active(nullb) && !is_fua)
1069                         null_make_cache_space(nullb, PAGE_SIZE);
1070
1071                 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
1072                 t_page = null_insert_page(nullb, sector,
1073                         !null_cache_active(nullb) || is_fua);
1074                 if (!t_page)
1075                         return -ENOSPC;
1076
1077                 src = kmap_atomic(source);
1078                 dst = kmap_atomic(t_page->page);
1079                 memcpy(dst + offset, src + off + count, temp);
1080                 kunmap_atomic(dst);
1081                 kunmap_atomic(src);
1082
1083                 __set_bit(sector & SECTOR_MASK, t_page->bitmap);
1084
1085                 if (is_fua)
1086                         null_free_sector(nullb, sector, true);
1087
1088                 count += temp;
1089                 sector += temp >> SECTOR_SHIFT;
1090         }
1091         return 0;
1092 }
1093
1094 static int copy_from_nullb(struct nullb *nullb, struct page *dest,
1095         unsigned int off, sector_t sector, size_t n)
1096 {
1097         size_t temp, count = 0;
1098         unsigned int offset;
1099         struct nullb_page *t_page;
1100         void *dst, *src;
1101
1102         while (count < n) {
1103                 temp = min_t(size_t, nullb->dev->blocksize, n - count);
1104
1105                 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
1106                 t_page = null_lookup_page(nullb, sector, false,
1107                         !null_cache_active(nullb));
1108
1109                 dst = kmap_atomic(dest);
1110                 if (!t_page) {
1111                         memset(dst + off + count, 0, temp);
1112                         goto next;
1113                 }
1114                 src = kmap_atomic(t_page->page);
1115                 memcpy(dst + off + count, src + offset, temp);
1116                 kunmap_atomic(src);
1117 next:
1118                 kunmap_atomic(dst);
1119
1120                 count += temp;
1121                 sector += temp >> SECTOR_SHIFT;
1122         }
1123         return 0;
1124 }
1125
1126 static void nullb_fill_pattern(struct nullb *nullb, struct page *page,
1127                                unsigned int len, unsigned int off)
1128 {
1129         void *dst;
1130
1131         dst = kmap_atomic(page);
1132         memset(dst + off, 0xFF, len);
1133         kunmap_atomic(dst);
1134 }
1135
1136 blk_status_t null_handle_discard(struct nullb_device *dev,
1137                                  sector_t sector, sector_t nr_sectors)
1138 {
1139         struct nullb *nullb = dev->nullb;
1140         size_t n = nr_sectors << SECTOR_SHIFT;
1141         size_t temp;
1142
1143         spin_lock_irq(&nullb->lock);
1144         while (n > 0) {
1145                 temp = min_t(size_t, n, dev->blocksize);
1146                 null_free_sector(nullb, sector, false);
1147                 if (null_cache_active(nullb))
1148                         null_free_sector(nullb, sector, true);
1149                 sector += temp >> SECTOR_SHIFT;
1150                 n -= temp;
1151         }
1152         spin_unlock_irq(&nullb->lock);
1153
1154         return BLK_STS_OK;
1155 }
1156
1157 static int null_handle_flush(struct nullb *nullb)
1158 {
1159         int err;
1160
1161         if (!null_cache_active(nullb))
1162                 return 0;
1163
1164         spin_lock_irq(&nullb->lock);
1165         while (true) {
1166                 err = null_make_cache_space(nullb,
1167                         nullb->dev->cache_size * 1024 * 1024);
1168                 if (err || nullb->dev->curr_cache == 0)
1169                         break;
1170         }
1171
1172         WARN_ON(!radix_tree_empty(&nullb->dev->cache));
1173         spin_unlock_irq(&nullb->lock);
1174         return err;
1175 }
1176
1177 static int null_transfer(struct nullb *nullb, struct page *page,
1178         unsigned int len, unsigned int off, bool is_write, sector_t sector,
1179         bool is_fua)
1180 {
1181         struct nullb_device *dev = nullb->dev;
1182         unsigned int valid_len = len;
1183         int err = 0;
1184
1185         if (!is_write) {
1186                 if (dev->zoned)
1187                         valid_len = null_zone_valid_read_len(nullb,
1188                                 sector, len);
1189
1190                 if (valid_len) {
1191                         err = copy_from_nullb(nullb, page, off,
1192                                 sector, valid_len);
1193                         off += valid_len;
1194                         len -= valid_len;
1195                 }
1196
1197                 if (len)
1198                         nullb_fill_pattern(nullb, page, len, off);
1199                 flush_dcache_page(page);
1200         } else {
1201                 flush_dcache_page(page);
1202                 err = copy_to_nullb(nullb, page, off, sector, len, is_fua);
1203         }
1204
1205         return err;
1206 }
1207
1208 static int null_handle_rq(struct nullb_cmd *cmd)
1209 {
1210         struct request *rq = cmd->rq;
1211         struct nullb *nullb = cmd->nq->dev->nullb;
1212         int err;
1213         unsigned int len;
1214         sector_t sector = blk_rq_pos(rq);
1215         struct req_iterator iter;
1216         struct bio_vec bvec;
1217
1218         spin_lock_irq(&nullb->lock);
1219         rq_for_each_segment(bvec, rq, iter) {
1220                 len = bvec.bv_len;
1221                 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1222                                      op_is_write(req_op(rq)), sector,
1223                                      rq->cmd_flags & REQ_FUA);
1224                 if (err) {
1225                         spin_unlock_irq(&nullb->lock);
1226                         return err;
1227                 }
1228                 sector += len >> SECTOR_SHIFT;
1229         }
1230         spin_unlock_irq(&nullb->lock);
1231
1232         return 0;
1233 }
1234
1235 static int null_handle_bio(struct nullb_cmd *cmd)
1236 {
1237         struct bio *bio = cmd->bio;
1238         struct nullb *nullb = cmd->nq->dev->nullb;
1239         int err;
1240         unsigned int len;
1241         sector_t sector = bio->bi_iter.bi_sector;
1242         struct bio_vec bvec;
1243         struct bvec_iter iter;
1244
1245         spin_lock_irq(&nullb->lock);
1246         bio_for_each_segment(bvec, bio, iter) {
1247                 len = bvec.bv_len;
1248                 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1249                                      op_is_write(bio_op(bio)), sector,
1250                                      bio->bi_opf & REQ_FUA);
1251                 if (err) {
1252                         spin_unlock_irq(&nullb->lock);
1253                         return err;
1254                 }
1255                 sector += len >> SECTOR_SHIFT;
1256         }
1257         spin_unlock_irq(&nullb->lock);
1258         return 0;
1259 }
1260
1261 static void null_stop_queue(struct nullb *nullb)
1262 {
1263         struct request_queue *q = nullb->q;
1264
1265         if (nullb->dev->queue_mode == NULL_Q_MQ)
1266                 blk_mq_stop_hw_queues(q);
1267 }
1268
1269 static void null_restart_queue_async(struct nullb *nullb)
1270 {
1271         struct request_queue *q = nullb->q;
1272
1273         if (nullb->dev->queue_mode == NULL_Q_MQ)
1274                 blk_mq_start_stopped_hw_queues(q, true);
1275 }
1276
1277 static inline blk_status_t null_handle_throttled(struct nullb_cmd *cmd)
1278 {
1279         struct nullb_device *dev = cmd->nq->dev;
1280         struct nullb *nullb = dev->nullb;
1281         blk_status_t sts = BLK_STS_OK;
1282         struct request *rq = cmd->rq;
1283
1284         if (!hrtimer_active(&nullb->bw_timer))
1285                 hrtimer_restart(&nullb->bw_timer);
1286
1287         if (atomic_long_sub_return(blk_rq_bytes(rq), &nullb->cur_bytes) < 0) {
1288                 null_stop_queue(nullb);
1289                 /* race with timer */
1290                 if (atomic_long_read(&nullb->cur_bytes) > 0)
1291                         null_restart_queue_async(nullb);
1292                 /* requeue request */
1293                 sts = BLK_STS_DEV_RESOURCE;
1294         }
1295         return sts;
1296 }
1297
1298 static inline blk_status_t null_handle_badblocks(struct nullb_cmd *cmd,
1299                                                  sector_t sector,
1300                                                  sector_t nr_sectors)
1301 {
1302         struct badblocks *bb = &cmd->nq->dev->badblocks;
1303         sector_t first_bad;
1304         int bad_sectors;
1305
1306         if (badblocks_check(bb, sector, nr_sectors, &first_bad, &bad_sectors))
1307                 return BLK_STS_IOERR;
1308
1309         return BLK_STS_OK;
1310 }
1311
1312 static inline blk_status_t null_handle_memory_backed(struct nullb_cmd *cmd,
1313                                                      enum req_opf op,
1314                                                      sector_t sector,
1315                                                      sector_t nr_sectors)
1316 {
1317         struct nullb_device *dev = cmd->nq->dev;
1318         int err;
1319
1320         if (op == REQ_OP_DISCARD)
1321                 return null_handle_discard(dev, sector, nr_sectors);
1322
1323         if (dev->queue_mode == NULL_Q_BIO)
1324                 err = null_handle_bio(cmd);
1325         else
1326                 err = null_handle_rq(cmd);
1327
1328         return errno_to_blk_status(err);
1329 }
1330
1331 static void nullb_zero_read_cmd_buffer(struct nullb_cmd *cmd)
1332 {
1333         struct nullb_device *dev = cmd->nq->dev;
1334         struct bio *bio;
1335
1336         if (dev->memory_backed)
1337                 return;
1338
1339         if (dev->queue_mode == NULL_Q_BIO && bio_op(cmd->bio) == REQ_OP_READ) {
1340                 zero_fill_bio(cmd->bio);
1341         } else if (req_op(cmd->rq) == REQ_OP_READ) {
1342                 __rq_for_each_bio(bio, cmd->rq)
1343                         zero_fill_bio(bio);
1344         }
1345 }
1346
1347 static inline void nullb_complete_cmd(struct nullb_cmd *cmd)
1348 {
1349         /*
1350          * Since root privileges are required to configure the null_blk
1351          * driver, it is fine that this driver does not initialize the
1352          * data buffers of read commands. Zero-initialize these buffers
1353          * anyway if KMSAN is enabled to prevent that KMSAN complains
1354          * about null_blk not initializing read data buffers.
1355          */
1356         if (IS_ENABLED(CONFIG_KMSAN))
1357                 nullb_zero_read_cmd_buffer(cmd);
1358
1359         /* Complete IO by inline, softirq or timer */
1360         switch (cmd->nq->dev->irqmode) {
1361         case NULL_IRQ_SOFTIRQ:
1362                 switch (cmd->nq->dev->queue_mode) {
1363                 case NULL_Q_MQ:
1364                         if (likely(!blk_should_fake_timeout(cmd->rq->q)))
1365                                 blk_mq_complete_request(cmd->rq);
1366                         break;
1367                 case NULL_Q_BIO:
1368                         /*
1369                          * XXX: no proper submitting cpu information available.
1370                          */
1371                         end_cmd(cmd);
1372                         break;
1373                 }
1374                 break;
1375         case NULL_IRQ_NONE:
1376                 end_cmd(cmd);
1377                 break;
1378         case NULL_IRQ_TIMER:
1379                 null_cmd_end_timer(cmd);
1380                 break;
1381         }
1382 }
1383
1384 blk_status_t null_process_cmd(struct nullb_cmd *cmd,
1385                               enum req_opf op, sector_t sector,
1386                               unsigned int nr_sectors)
1387 {
1388         struct nullb_device *dev = cmd->nq->dev;
1389         blk_status_t ret;
1390
1391         if (dev->badblocks.shift != -1) {
1392                 ret = null_handle_badblocks(cmd, sector, nr_sectors);
1393                 if (ret != BLK_STS_OK)
1394                         return ret;
1395         }
1396
1397         if (dev->memory_backed)
1398                 return null_handle_memory_backed(cmd, op, sector, nr_sectors);
1399
1400         return BLK_STS_OK;
1401 }
1402
1403 static blk_status_t null_handle_cmd(struct nullb_cmd *cmd, sector_t sector,
1404                                     sector_t nr_sectors, enum req_opf op)
1405 {
1406         struct nullb_device *dev = cmd->nq->dev;
1407         struct nullb *nullb = dev->nullb;
1408         blk_status_t sts;
1409
1410         if (test_bit(NULLB_DEV_FL_THROTTLED, &dev->flags)) {
1411                 sts = null_handle_throttled(cmd);
1412                 if (sts != BLK_STS_OK)
1413                         return sts;
1414         }
1415
1416         if (op == REQ_OP_FLUSH) {
1417                 cmd->error = errno_to_blk_status(null_handle_flush(nullb));
1418                 goto out;
1419         }
1420
1421         if (dev->zoned)
1422                 sts = null_process_zoned_cmd(cmd, op, sector, nr_sectors);
1423         else
1424                 sts = null_process_cmd(cmd, op, sector, nr_sectors);
1425
1426         /* Do not overwrite errors (e.g. timeout errors) */
1427         if (cmd->error == BLK_STS_OK)
1428                 cmd->error = sts;
1429
1430 out:
1431         nullb_complete_cmd(cmd);
1432         return BLK_STS_OK;
1433 }
1434
1435 static enum hrtimer_restart nullb_bwtimer_fn(struct hrtimer *timer)
1436 {
1437         struct nullb *nullb = container_of(timer, struct nullb, bw_timer);
1438         ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1439         unsigned int mbps = nullb->dev->mbps;
1440
1441         if (atomic_long_read(&nullb->cur_bytes) == mb_per_tick(mbps))
1442                 return HRTIMER_NORESTART;
1443
1444         atomic_long_set(&nullb->cur_bytes, mb_per_tick(mbps));
1445         null_restart_queue_async(nullb);
1446
1447         hrtimer_forward_now(&nullb->bw_timer, timer_interval);
1448
1449         return HRTIMER_RESTART;
1450 }
1451
1452 static void nullb_setup_bwtimer(struct nullb *nullb)
1453 {
1454         ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1455
1456         hrtimer_init(&nullb->bw_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1457         nullb->bw_timer.function = nullb_bwtimer_fn;
1458         atomic_long_set(&nullb->cur_bytes, mb_per_tick(nullb->dev->mbps));
1459         hrtimer_start(&nullb->bw_timer, timer_interval, HRTIMER_MODE_REL);
1460 }
1461
1462 static struct nullb_queue *nullb_to_queue(struct nullb *nullb)
1463 {
1464         int index = 0;
1465
1466         if (nullb->nr_queues != 1)
1467                 index = raw_smp_processor_id() / ((nr_cpu_ids + nullb->nr_queues - 1) / nullb->nr_queues);
1468
1469         return &nullb->queues[index];
1470 }
1471
1472 static void null_submit_bio(struct bio *bio)
1473 {
1474         sector_t sector = bio->bi_iter.bi_sector;
1475         sector_t nr_sectors = bio_sectors(bio);
1476         struct nullb *nullb = bio->bi_bdev->bd_disk->private_data;
1477         struct nullb_queue *nq = nullb_to_queue(nullb);
1478
1479         null_handle_cmd(alloc_cmd(nq, bio), sector, nr_sectors, bio_op(bio));
1480 }
1481
1482 static bool should_timeout_request(struct request *rq)
1483 {
1484 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1485         if (g_timeout_str[0])
1486                 return should_fail(&null_timeout_attr, 1);
1487 #endif
1488         return false;
1489 }
1490
1491 static bool should_requeue_request(struct request *rq)
1492 {
1493 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1494         if (g_requeue_str[0])
1495                 return should_fail(&null_requeue_attr, 1);
1496 #endif
1497         return false;
1498 }
1499
1500 static int null_map_queues(struct blk_mq_tag_set *set)
1501 {
1502         struct nullb *nullb = set->driver_data;
1503         int i, qoff;
1504         unsigned int submit_queues = g_submit_queues;
1505         unsigned int poll_queues = g_poll_queues;
1506
1507         if (nullb) {
1508                 struct nullb_device *dev = nullb->dev;
1509
1510                 /*
1511                  * Refer nr_hw_queues of the tag set to check if the expected
1512                  * number of hardware queues are prepared. If block layer failed
1513                  * to prepare them, use previous numbers of submit queues and
1514                  * poll queues to map queues.
1515                  */
1516                 if (set->nr_hw_queues ==
1517                     dev->submit_queues + dev->poll_queues) {
1518                         submit_queues = dev->submit_queues;
1519                         poll_queues = dev->poll_queues;
1520                 } else if (set->nr_hw_queues ==
1521                            dev->prev_submit_queues + dev->prev_poll_queues) {
1522                         submit_queues = dev->prev_submit_queues;
1523                         poll_queues = dev->prev_poll_queues;
1524                 } else {
1525                         pr_warn("tag set has unexpected nr_hw_queues: %d\n",
1526                                 set->nr_hw_queues);
1527                         return -EINVAL;
1528                 }
1529         }
1530
1531         for (i = 0, qoff = 0; i < set->nr_maps; i++) {
1532                 struct blk_mq_queue_map *map = &set->map[i];
1533
1534                 switch (i) {
1535                 case HCTX_TYPE_DEFAULT:
1536                         map->nr_queues = submit_queues;
1537                         break;
1538                 case HCTX_TYPE_READ:
1539                         map->nr_queues = 0;
1540                         continue;
1541                 case HCTX_TYPE_POLL:
1542                         map->nr_queues = poll_queues;
1543                         break;
1544                 }
1545                 map->queue_offset = qoff;
1546                 qoff += map->nr_queues;
1547                 blk_mq_map_queues(map);
1548         }
1549
1550         return 0;
1551 }
1552
1553 static int null_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1554 {
1555         struct nullb_queue *nq = hctx->driver_data;
1556         LIST_HEAD(list);
1557         int nr = 0;
1558
1559         spin_lock(&nq->poll_lock);
1560         list_splice_init(&nq->poll_list, &list);
1561         spin_unlock(&nq->poll_lock);
1562
1563         while (!list_empty(&list)) {
1564                 struct nullb_cmd *cmd;
1565                 struct request *req;
1566
1567                 req = list_first_entry(&list, struct request, queuelist);
1568                 list_del_init(&req->queuelist);
1569                 cmd = blk_mq_rq_to_pdu(req);
1570                 cmd->error = null_process_cmd(cmd, req_op(req), blk_rq_pos(req),
1571                                                 blk_rq_sectors(req));
1572                 if (!blk_mq_add_to_batch(req, iob, (__force int) cmd->error,
1573                                         blk_mq_end_request_batch))
1574                         end_cmd(cmd);
1575                 nr++;
1576         }
1577
1578         return nr;
1579 }
1580
1581 static enum blk_eh_timer_return null_timeout_rq(struct request *rq, bool res)
1582 {
1583         struct blk_mq_hw_ctx *hctx = rq->mq_hctx;
1584         struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq);
1585
1586         pr_info("rq %p timed out\n", rq);
1587
1588         if (hctx->type == HCTX_TYPE_POLL) {
1589                 struct nullb_queue *nq = hctx->driver_data;
1590
1591                 spin_lock(&nq->poll_lock);
1592                 list_del_init(&rq->queuelist);
1593                 spin_unlock(&nq->poll_lock);
1594         }
1595
1596         /*
1597          * If the device is marked as blocking (i.e. memory backed or zoned
1598          * device), the submission path may be blocked waiting for resources
1599          * and cause real timeouts. For these real timeouts, the submission
1600          * path will complete the request using blk_mq_complete_request().
1601          * Only fake timeouts need to execute blk_mq_complete_request() here.
1602          */
1603         cmd->error = BLK_STS_TIMEOUT;
1604         if (cmd->fake_timeout || hctx->type == HCTX_TYPE_POLL)
1605                 blk_mq_complete_request(rq);
1606         return BLK_EH_DONE;
1607 }
1608
1609 static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx,
1610                          const struct blk_mq_queue_data *bd)
1611 {
1612         struct nullb_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
1613         struct nullb_queue *nq = hctx->driver_data;
1614         sector_t nr_sectors = blk_rq_sectors(bd->rq);
1615         sector_t sector = blk_rq_pos(bd->rq);
1616         const bool is_poll = hctx->type == HCTX_TYPE_POLL;
1617
1618         might_sleep_if(hctx->flags & BLK_MQ_F_BLOCKING);
1619
1620         if (!is_poll && nq->dev->irqmode == NULL_IRQ_TIMER) {
1621                 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1622                 cmd->timer.function = null_cmd_timer_expired;
1623         }
1624         cmd->rq = bd->rq;
1625         cmd->error = BLK_STS_OK;
1626         cmd->nq = nq;
1627         cmd->fake_timeout = should_timeout_request(bd->rq);
1628
1629         blk_mq_start_request(bd->rq);
1630
1631         if (should_requeue_request(bd->rq)) {
1632                 /*
1633                  * Alternate between hitting the core BUSY path, and the
1634                  * driver driven requeue path
1635                  */
1636                 nq->requeue_selection++;
1637                 if (nq->requeue_selection & 1)
1638                         return BLK_STS_RESOURCE;
1639                 else {
1640                         blk_mq_requeue_request(bd->rq, true);
1641                         return BLK_STS_OK;
1642                 }
1643         }
1644
1645         if (is_poll) {
1646                 spin_lock(&nq->poll_lock);
1647                 list_add_tail(&bd->rq->queuelist, &nq->poll_list);
1648                 spin_unlock(&nq->poll_lock);
1649                 return BLK_STS_OK;
1650         }
1651         if (cmd->fake_timeout)
1652                 return BLK_STS_OK;
1653
1654         return null_handle_cmd(cmd, sector, nr_sectors, req_op(bd->rq));
1655 }
1656
1657 static void cleanup_queue(struct nullb_queue *nq)
1658 {
1659         kfree(nq->tag_map);
1660         kfree(nq->cmds);
1661 }
1662
1663 static void cleanup_queues(struct nullb *nullb)
1664 {
1665         int i;
1666
1667         for (i = 0; i < nullb->nr_queues; i++)
1668                 cleanup_queue(&nullb->queues[i]);
1669
1670         kfree(nullb->queues);
1671 }
1672
1673 static void null_exit_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx)
1674 {
1675         struct nullb_queue *nq = hctx->driver_data;
1676         struct nullb *nullb = nq->dev->nullb;
1677
1678         nullb->nr_queues--;
1679 }
1680
1681 static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq)
1682 {
1683         init_waitqueue_head(&nq->wait);
1684         nq->queue_depth = nullb->queue_depth;
1685         nq->dev = nullb->dev;
1686         INIT_LIST_HEAD(&nq->poll_list);
1687         spin_lock_init(&nq->poll_lock);
1688 }
1689
1690 static int null_init_hctx(struct blk_mq_hw_ctx *hctx, void *driver_data,
1691                           unsigned int hctx_idx)
1692 {
1693         struct nullb *nullb = hctx->queue->queuedata;
1694         struct nullb_queue *nq;
1695
1696 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1697         if (g_init_hctx_str[0] && should_fail(&null_init_hctx_attr, 1))
1698                 return -EFAULT;
1699 #endif
1700
1701         nq = &nullb->queues[hctx_idx];
1702         hctx->driver_data = nq;
1703         null_init_queue(nullb, nq);
1704         nullb->nr_queues++;
1705
1706         return 0;
1707 }
1708
1709 static const struct blk_mq_ops null_mq_ops = {
1710         .queue_rq       = null_queue_rq,
1711         .complete       = null_complete_rq,
1712         .timeout        = null_timeout_rq,
1713         .poll           = null_poll,
1714         .map_queues     = null_map_queues,
1715         .init_hctx      = null_init_hctx,
1716         .exit_hctx      = null_exit_hctx,
1717 };
1718
1719 static void null_del_dev(struct nullb *nullb)
1720 {
1721         struct nullb_device *dev;
1722
1723         if (!nullb)
1724                 return;
1725
1726         dev = nullb->dev;
1727
1728         ida_simple_remove(&nullb_indexes, nullb->index);
1729
1730         list_del_init(&nullb->list);
1731
1732         del_gendisk(nullb->disk);
1733
1734         if (test_bit(NULLB_DEV_FL_THROTTLED, &nullb->dev->flags)) {
1735                 hrtimer_cancel(&nullb->bw_timer);
1736                 atomic_long_set(&nullb->cur_bytes, LONG_MAX);
1737                 null_restart_queue_async(nullb);
1738         }
1739
1740         blk_cleanup_disk(nullb->disk);
1741         if (dev->queue_mode == NULL_Q_MQ &&
1742             nullb->tag_set == &nullb->__tag_set)
1743                 blk_mq_free_tag_set(nullb->tag_set);
1744         cleanup_queues(nullb);
1745         if (null_cache_active(nullb))
1746                 null_free_device_storage(nullb->dev, true);
1747         kfree(nullb);
1748         dev->nullb = NULL;
1749 }
1750
1751 static void null_config_discard(struct nullb *nullb)
1752 {
1753         if (nullb->dev->discard == false)
1754                 return;
1755
1756         if (!nullb->dev->memory_backed) {
1757                 nullb->dev->discard = false;
1758                 pr_info("discard option is ignored without memory backing\n");
1759                 return;
1760         }
1761
1762         if (nullb->dev->zoned) {
1763                 nullb->dev->discard = false;
1764                 pr_info("discard option is ignored in zoned mode\n");
1765                 return;
1766         }
1767
1768         nullb->q->limits.discard_granularity = nullb->dev->blocksize;
1769         blk_queue_max_discard_sectors(nullb->q, UINT_MAX >> 9);
1770 }
1771
1772 static const struct block_device_operations null_bio_ops = {
1773         .owner          = THIS_MODULE,
1774         .submit_bio     = null_submit_bio,
1775         .report_zones   = null_report_zones,
1776 };
1777
1778 static const struct block_device_operations null_rq_ops = {
1779         .owner          = THIS_MODULE,
1780         .report_zones   = null_report_zones,
1781 };
1782
1783 static int setup_commands(struct nullb_queue *nq)
1784 {
1785         struct nullb_cmd *cmd;
1786         int i, tag_size;
1787
1788         nq->cmds = kcalloc(nq->queue_depth, sizeof(*cmd), GFP_KERNEL);
1789         if (!nq->cmds)
1790                 return -ENOMEM;
1791
1792         tag_size = ALIGN(nq->queue_depth, BITS_PER_LONG) / BITS_PER_LONG;
1793         nq->tag_map = kcalloc(tag_size, sizeof(unsigned long), GFP_KERNEL);
1794         if (!nq->tag_map) {
1795                 kfree(nq->cmds);
1796                 return -ENOMEM;
1797         }
1798
1799         for (i = 0; i < nq->queue_depth; i++) {
1800                 cmd = &nq->cmds[i];
1801                 cmd->tag = -1U;
1802         }
1803
1804         return 0;
1805 }
1806
1807 static int setup_queues(struct nullb *nullb)
1808 {
1809         int nqueues = nr_cpu_ids;
1810
1811         if (g_poll_queues)
1812                 nqueues += g_poll_queues;
1813
1814         nullb->queues = kcalloc(nqueues, sizeof(struct nullb_queue),
1815                                 GFP_KERNEL);
1816         if (!nullb->queues)
1817                 return -ENOMEM;
1818
1819         nullb->queue_depth = nullb->dev->hw_queue_depth;
1820         return 0;
1821 }
1822
1823 static int init_driver_queues(struct nullb *nullb)
1824 {
1825         struct nullb_queue *nq;
1826         int i, ret = 0;
1827
1828         for (i = 0; i < nullb->dev->submit_queues; i++) {
1829                 nq = &nullb->queues[i];
1830
1831                 null_init_queue(nullb, nq);
1832
1833                 ret = setup_commands(nq);
1834                 if (ret)
1835                         return ret;
1836                 nullb->nr_queues++;
1837         }
1838         return 0;
1839 }
1840
1841 static int null_gendisk_register(struct nullb *nullb)
1842 {
1843         sector_t size = ((sector_t)nullb->dev->size * SZ_1M) >> SECTOR_SHIFT;
1844         struct gendisk *disk = nullb->disk;
1845
1846         set_capacity(disk, size);
1847
1848         disk->major             = null_major;
1849         disk->first_minor       = nullb->index;
1850         disk->minors            = 1;
1851         if (queue_is_mq(nullb->q))
1852                 disk->fops              = &null_rq_ops;
1853         else
1854                 disk->fops              = &null_bio_ops;
1855         disk->private_data      = nullb;
1856         strncpy(disk->disk_name, nullb->disk_name, DISK_NAME_LEN);
1857
1858         if (nullb->dev->zoned) {
1859                 int ret = null_register_zoned_dev(nullb);
1860
1861                 if (ret)
1862                         return ret;
1863         }
1864
1865         return add_disk(disk);
1866 }
1867
1868 static int null_init_tag_set(struct nullb *nullb, struct blk_mq_tag_set *set)
1869 {
1870         int poll_queues;
1871
1872         set->ops = &null_mq_ops;
1873         set->nr_hw_queues = nullb ? nullb->dev->submit_queues :
1874                                                 g_submit_queues;
1875         poll_queues = nullb ? nullb->dev->poll_queues : g_poll_queues;
1876         if (poll_queues)
1877                 set->nr_hw_queues += poll_queues;
1878         set->queue_depth = nullb ? nullb->dev->hw_queue_depth :
1879                                                 g_hw_queue_depth;
1880         set->numa_node = nullb ? nullb->dev->home_node : g_home_node;
1881         set->cmd_size   = sizeof(struct nullb_cmd);
1882         set->flags = BLK_MQ_F_SHOULD_MERGE;
1883         if (g_no_sched)
1884                 set->flags |= BLK_MQ_F_NO_SCHED;
1885         if (g_shared_tag_bitmap)
1886                 set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1887         set->driver_data = nullb;
1888         if (poll_queues)
1889                 set->nr_maps = 3;
1890         else
1891                 set->nr_maps = 1;
1892
1893         if ((nullb && nullb->dev->blocking) || g_blocking)
1894                 set->flags |= BLK_MQ_F_BLOCKING;
1895
1896         return blk_mq_alloc_tag_set(set);
1897 }
1898
1899 static int null_validate_conf(struct nullb_device *dev)
1900 {
1901         dev->blocksize = round_down(dev->blocksize, 512);
1902         dev->blocksize = clamp_t(unsigned int, dev->blocksize, 512, 4096);
1903
1904         if (dev->queue_mode == NULL_Q_MQ && dev->use_per_node_hctx) {
1905                 if (dev->submit_queues != nr_online_nodes)
1906                         dev->submit_queues = nr_online_nodes;
1907         } else if (dev->submit_queues > nr_cpu_ids)
1908                 dev->submit_queues = nr_cpu_ids;
1909         else if (dev->submit_queues == 0)
1910                 dev->submit_queues = 1;
1911         dev->prev_submit_queues = dev->submit_queues;
1912
1913         if (dev->poll_queues > g_poll_queues)
1914                 dev->poll_queues = g_poll_queues;
1915         dev->prev_poll_queues = dev->poll_queues;
1916
1917         dev->queue_mode = min_t(unsigned int, dev->queue_mode, NULL_Q_MQ);
1918         dev->irqmode = min_t(unsigned int, dev->irqmode, NULL_IRQ_TIMER);
1919
1920         /* Do memory allocation, so set blocking */
1921         if (dev->memory_backed)
1922                 dev->blocking = true;
1923         else /* cache is meaningless */
1924                 dev->cache_size = 0;
1925         dev->cache_size = min_t(unsigned long, ULONG_MAX / 1024 / 1024,
1926                                                 dev->cache_size);
1927         dev->mbps = min_t(unsigned int, 1024 * 40, dev->mbps);
1928         /* can not stop a queue */
1929         if (dev->queue_mode == NULL_Q_BIO)
1930                 dev->mbps = 0;
1931
1932         if (dev->zoned &&
1933             (!dev->zone_size || !is_power_of_2(dev->zone_size))) {
1934                 pr_err("zone_size must be power-of-two\n");
1935                 return -EINVAL;
1936         }
1937
1938         return 0;
1939 }
1940
1941 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1942 static bool __null_setup_fault(struct fault_attr *attr, char *str)
1943 {
1944         if (!str[0])
1945                 return true;
1946
1947         if (!setup_fault_attr(attr, str))
1948                 return false;
1949
1950         attr->verbose = 0;
1951         return true;
1952 }
1953 #endif
1954
1955 static bool null_setup_fault(void)
1956 {
1957 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1958         if (!__null_setup_fault(&null_timeout_attr, g_timeout_str))
1959                 return false;
1960         if (!__null_setup_fault(&null_requeue_attr, g_requeue_str))
1961                 return false;
1962         if (!__null_setup_fault(&null_init_hctx_attr, g_init_hctx_str))
1963                 return false;
1964 #endif
1965         return true;
1966 }
1967
1968 static int null_add_dev(struct nullb_device *dev)
1969 {
1970         struct nullb *nullb;
1971         int rv;
1972
1973         rv = null_validate_conf(dev);
1974         if (rv)
1975                 return rv;
1976
1977         nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, dev->home_node);
1978         if (!nullb) {
1979                 rv = -ENOMEM;
1980                 goto out;
1981         }
1982         nullb->dev = dev;
1983         dev->nullb = nullb;
1984
1985         spin_lock_init(&nullb->lock);
1986
1987         rv = setup_queues(nullb);
1988         if (rv)
1989                 goto out_free_nullb;
1990
1991         if (dev->queue_mode == NULL_Q_MQ) {
1992                 if (shared_tags) {
1993                         nullb->tag_set = &tag_set;
1994                         rv = 0;
1995                 } else {
1996                         nullb->tag_set = &nullb->__tag_set;
1997                         rv = null_init_tag_set(nullb, nullb->tag_set);
1998                 }
1999
2000                 if (rv)
2001                         goto out_cleanup_queues;
2002
2003                 if (!null_setup_fault())
2004                         goto out_cleanup_tags;
2005
2006                 nullb->tag_set->timeout = 5 * HZ;
2007                 nullb->disk = blk_mq_alloc_disk(nullb->tag_set, nullb);
2008                 if (IS_ERR(nullb->disk)) {
2009                         rv = PTR_ERR(nullb->disk);
2010                         goto out_cleanup_tags;
2011                 }
2012                 nullb->q = nullb->disk->queue;
2013         } else if (dev->queue_mode == NULL_Q_BIO) {
2014                 rv = -ENOMEM;
2015                 nullb->disk = blk_alloc_disk(nullb->dev->home_node);
2016                 if (!nullb->disk)
2017                         goto out_cleanup_queues;
2018
2019                 nullb->q = nullb->disk->queue;
2020                 rv = init_driver_queues(nullb);
2021                 if (rv)
2022                         goto out_cleanup_disk;
2023         }
2024
2025         if (dev->mbps) {
2026                 set_bit(NULLB_DEV_FL_THROTTLED, &dev->flags);
2027                 nullb_setup_bwtimer(nullb);
2028         }
2029
2030         if (dev->cache_size > 0) {
2031                 set_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
2032                 blk_queue_write_cache(nullb->q, true, true);
2033         }
2034
2035         if (dev->zoned) {
2036                 rv = null_init_zoned_dev(dev, nullb->q);
2037                 if (rv)
2038                         goto out_cleanup_disk;
2039         }
2040
2041         nullb->q->queuedata = nullb;
2042         blk_queue_flag_set(QUEUE_FLAG_NONROT, nullb->q);
2043         blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, nullb->q);
2044
2045         mutex_lock(&lock);
2046         nullb->index = ida_simple_get(&nullb_indexes, 0, 0, GFP_KERNEL);
2047         dev->index = nullb->index;
2048         mutex_unlock(&lock);
2049
2050         blk_queue_logical_block_size(nullb->q, dev->blocksize);
2051         blk_queue_physical_block_size(nullb->q, dev->blocksize);
2052         if (!dev->max_sectors)
2053                 dev->max_sectors = queue_max_hw_sectors(nullb->q);
2054         dev->max_sectors = min_t(unsigned int, dev->max_sectors,
2055                                  BLK_DEF_MAX_SECTORS);
2056         blk_queue_max_hw_sectors(nullb->q, dev->max_sectors);
2057
2058         if (dev->virt_boundary)
2059                 blk_queue_virt_boundary(nullb->q, PAGE_SIZE - 1);
2060
2061         null_config_discard(nullb);
2062
2063         if (config_item_name(&dev->item)) {
2064                 /* Use configfs dir name as the device name */
2065                 snprintf(nullb->disk_name, sizeof(nullb->disk_name),
2066                          "%s", config_item_name(&dev->item));
2067         } else {
2068                 sprintf(nullb->disk_name, "nullb%d", nullb->index);
2069         }
2070
2071         rv = null_gendisk_register(nullb);
2072         if (rv)
2073                 goto out_cleanup_zone;
2074
2075         mutex_lock(&lock);
2076         list_add_tail(&nullb->list, &nullb_list);
2077         mutex_unlock(&lock);
2078
2079         pr_info("disk %s created\n", nullb->disk_name);
2080
2081         return 0;
2082 out_cleanup_zone:
2083         null_free_zoned_dev(dev);
2084 out_cleanup_disk:
2085         blk_cleanup_disk(nullb->disk);
2086 out_cleanup_tags:
2087         if (dev->queue_mode == NULL_Q_MQ && nullb->tag_set == &nullb->__tag_set)
2088                 blk_mq_free_tag_set(nullb->tag_set);
2089 out_cleanup_queues:
2090         cleanup_queues(nullb);
2091 out_free_nullb:
2092         kfree(nullb);
2093         dev->nullb = NULL;
2094 out:
2095         return rv;
2096 }
2097
2098 static struct nullb *null_find_dev_by_name(const char *name)
2099 {
2100         struct nullb *nullb = NULL, *nb;
2101
2102         mutex_lock(&lock);
2103         list_for_each_entry(nb, &nullb_list, list) {
2104                 if (strcmp(nb->disk_name, name) == 0) {
2105                         nullb = nb;
2106                         break;
2107                 }
2108         }
2109         mutex_unlock(&lock);
2110
2111         return nullb;
2112 }
2113
2114 static int null_create_dev(void)
2115 {
2116         struct nullb_device *dev;
2117         int ret;
2118
2119         dev = null_alloc_dev();
2120         if (!dev)
2121                 return -ENOMEM;
2122
2123         ret = null_add_dev(dev);
2124         if (ret) {
2125                 null_free_dev(dev);
2126                 return ret;
2127         }
2128
2129         return 0;
2130 }
2131
2132 static void null_destroy_dev(struct nullb *nullb)
2133 {
2134         struct nullb_device *dev = nullb->dev;
2135
2136         null_del_dev(nullb);
2137         null_free_dev(dev);
2138 }
2139
2140 static int __init null_init(void)
2141 {
2142         int ret = 0;
2143         unsigned int i;
2144         struct nullb *nullb;
2145
2146         if (g_bs > PAGE_SIZE) {
2147                 pr_warn("invalid block size\n");
2148                 pr_warn("defaults block size to %lu\n", PAGE_SIZE);
2149                 g_bs = PAGE_SIZE;
2150         }
2151
2152         if (g_max_sectors > BLK_DEF_MAX_SECTORS) {
2153                 pr_warn("invalid max sectors\n");
2154                 pr_warn("defaults max sectors to %u\n", BLK_DEF_MAX_SECTORS);
2155                 g_max_sectors = BLK_DEF_MAX_SECTORS;
2156         }
2157
2158         if (g_home_node != NUMA_NO_NODE && g_home_node >= nr_online_nodes) {
2159                 pr_err("invalid home_node value\n");
2160                 g_home_node = NUMA_NO_NODE;
2161         }
2162
2163         if (g_queue_mode == NULL_Q_RQ) {
2164                 pr_err("legacy IO path is no longer available\n");
2165                 return -EINVAL;
2166         }
2167
2168         if (g_queue_mode == NULL_Q_MQ && g_use_per_node_hctx) {
2169                 if (g_submit_queues != nr_online_nodes) {
2170                         pr_warn("submit_queues param is set to %u.\n",
2171                                 nr_online_nodes);
2172                         g_submit_queues = nr_online_nodes;
2173                 }
2174         } else if (g_submit_queues > nr_cpu_ids) {
2175                 g_submit_queues = nr_cpu_ids;
2176         } else if (g_submit_queues <= 0) {
2177                 g_submit_queues = 1;
2178         }
2179
2180         if (g_queue_mode == NULL_Q_MQ && shared_tags) {
2181                 ret = null_init_tag_set(NULL, &tag_set);
2182                 if (ret)
2183                         return ret;
2184         }
2185
2186         config_group_init(&nullb_subsys.su_group);
2187         mutex_init(&nullb_subsys.su_mutex);
2188
2189         ret = configfs_register_subsystem(&nullb_subsys);
2190         if (ret)
2191                 goto err_tagset;
2192
2193         mutex_init(&lock);
2194
2195         null_major = register_blkdev(0, "nullb");
2196         if (null_major < 0) {
2197                 ret = null_major;
2198                 goto err_conf;
2199         }
2200
2201         for (i = 0; i < nr_devices; i++) {
2202                 ret = null_create_dev();
2203                 if (ret)
2204                         goto err_dev;
2205         }
2206
2207         pr_info("module loaded\n");
2208         return 0;
2209
2210 err_dev:
2211         while (!list_empty(&nullb_list)) {
2212                 nullb = list_entry(nullb_list.next, struct nullb, list);
2213                 null_destroy_dev(nullb);
2214         }
2215         unregister_blkdev(null_major, "nullb");
2216 err_conf:
2217         configfs_unregister_subsystem(&nullb_subsys);
2218 err_tagset:
2219         if (g_queue_mode == NULL_Q_MQ && shared_tags)
2220                 blk_mq_free_tag_set(&tag_set);
2221         return ret;
2222 }
2223
2224 static void __exit null_exit(void)
2225 {
2226         struct nullb *nullb;
2227
2228         configfs_unregister_subsystem(&nullb_subsys);
2229
2230         unregister_blkdev(null_major, "nullb");
2231
2232         mutex_lock(&lock);
2233         while (!list_empty(&nullb_list)) {
2234                 nullb = list_entry(nullb_list.next, struct nullb, list);
2235                 null_destroy_dev(nullb);
2236         }
2237         mutex_unlock(&lock);
2238
2239         if (g_queue_mode == NULL_Q_MQ && shared_tags)
2240                 blk_mq_free_tag_set(&tag_set);
2241 }
2242
2243 module_init(null_init);
2244 module_exit(null_exit);
2245
2246 MODULE_AUTHOR("Jens Axboe <axboe@kernel.dk>");
2247 MODULE_LICENSE("GPL");