projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
fault-inject: allow configuration via configfs
[linux-block.git] / drivers / block / null_blk / main.c
CommitLineData
09c434b8 1// SPDX-License-Identifier: GPL-2.0-only
3bf2bd20
SL		 2/*
3 * Add configfs and memory store: Kyungchan Koh <kkc6196@fb.com> and
4 * Shaohua Li <shli@fb.com>
5 */
f2298c04 6#include <linux/module.h>
fc1bc354 7
f2298c04
JA		 8#include <linux/moduleparam.h>
9#include <linux/sched.h>
10#include <linux/fs.h>
f2298c04 11#include <linux/init.h>
6dad38d3 12#include "null_blk.h"
f2298c04 13
db060f54
DLM		 14#undef pr_fmt
15#define pr_fmt(fmt) "null_blk: " fmt
16
5bcd0e0c
SL		 17#define FREE_BATCH 16
18
eff2c4f1
SL		 19#define TICKS_PER_SEC 50ULL
20#define TIMER_INTERVAL (NSEC_PER_SEC / TICKS_PER_SEC)
21
33f782c4 22#ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
93b57046 23static DECLARE_FAULT_ATTR(null_timeout_attr);
24941b90 24static DECLARE_FAULT_ATTR(null_requeue_attr);
596444e7 25static DECLARE_FAULT_ATTR(null_init_hctx_attr);
33f782c4 26#endif
93b57046 27
eff2c4f1
SL		 28static inline u64 mb_per_tick(int mbps)
29{
30 return (1 << 20) / TICKS_PER_SEC * ((u64) mbps);
31}
f2298c04 32
3bf2bd20
SL		 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.
eff2c4f1 38 * THROTTLED: Device is being throttled.
deb78b41 39 * CACHE: Device is using a write-back cache.
3bf2bd20
SL		 40 */
41enum nullb_device_flags {
42 NULLB_DEV_FL_CONFIGURED = 0,
43 NULLB_DEV_FL_UP = 1,
eff2c4f1 44 NULLB_DEV_FL_THROTTLED = 2,
deb78b41 45 NULLB_DEV_FL_CACHE = 3,
3bf2bd20
SL		 46};
47
66231ad3 48#define MAP_SZ ((PAGE_SIZE >> SECTOR_SHIFT) + 2)
5bcd0e0c
SL		 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
deb78b41
SL		 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
5bcd0e0c
SL		 60 */
61struct nullb_page {
62 struct page *page;
66231ad3 63 DECLARE_BITMAP(bitmap, MAP_SZ);
5bcd0e0c 64};
66231ad3
ML		 65#define NULLB_PAGE_LOCK (MAP_SZ - 1)
66#define NULLB_PAGE_FREE (MAP_SZ - 2)
5bcd0e0c 67
f2298c04
JA		 68static LIST_HEAD(nullb_list);
69static struct mutex lock;
70static int null_major;
94bc02e3 71static DEFINE_IDA(nullb_indexes);
82f402fe 72static struct blk_mq_tag_set tag_set;
f2298c04 73
f2298c04
JA		 74enum {
75 NULL_IRQ_NONE = 0,
76 NULL_IRQ_SOFTIRQ = 1,
77 NULL_IRQ_TIMER = 2,
ce2c350b 78};
f2298c04 79
cee1b215
MG		 80static bool g_virt_boundary = false;
81module_param_named(virt_boundary, g_virt_boundary, bool, 0444);
82MODULE_PARM_DESC(virt_boundary, "Require a virtual boundary for the device. Default: False");
83
b3cffc38 84static int g_no_sched;
5657a819 85module_param_named(no_sched, g_no_sched, int, 0444);
b3cffc38 86MODULE_PARM_DESC(no_sched, "No io scheduler");
87
2984c868 88static int g_submit_queues = 1;
5657a819 89module_param_named(submit_queues, g_submit_queues, int, 0444);
f2298c04
JA		 90MODULE_PARM_DESC(submit_queues, "Number of submission queues");
91
0a593fbb
JA		 92static int g_poll_queues = 1;
93module_param_named(poll_queues, g_poll_queues, int, 0444);
94MODULE_PARM_DESC(poll_queues, "Number of IOPOLL submission queues");
95
2984c868 96static int g_home_node = NUMA_NO_NODE;
5657a819 97module_param_named(home_node, g_home_node, int, 0444);
f2298c04
JA		 98MODULE_PARM_DESC(home_node, "Home node for the device");
99
33f782c4 100#ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
290df92a
DZ		 101/*
102 * For more details about fault injection, please refer to
103 * Documentation/fault-injection/fault-injection.rst.
104 */
93b57046 105static char g_timeout_str[80];
5657a819 106module_param_string(timeout, g_timeout_str, sizeof(g_timeout_str), 0444);
290df92a 107MODULE_PARM_DESC(timeout, "Fault injection. timeout=<interval>,<probability>,<space>,<times>");
24941b90
JA		 108
109static char g_requeue_str[80];
5657a819 110module_param_string(requeue, g_requeue_str, sizeof(g_requeue_str), 0444);
290df92a 111MODULE_PARM_DESC(requeue, "Fault injection. requeue=<interval>,<probability>,<space>,<times>");
596444e7
BVA		 112
113static char g_init_hctx_str[80];
114module_param_string(init_hctx, g_init_hctx_str, sizeof(g_init_hctx_str), 0444);
290df92a 115MODULE_PARM_DESC(init_hctx, "Fault injection to fail hctx init. init_hctx=<interval>,<probability>,<space>,<times>");
33f782c4 116#endif
93b57046 117
2984c868 118static int g_queue_mode = NULL_Q_MQ;
709c8667
MB		 119
120static 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
135static int null_set_queue_mode(const char *str, const struct kernel_param *kp)
136{
2984c868 137 return null_param_store_val(str, &g_queue_mode, NULL_Q_BIO, NULL_Q_MQ);
709c8667
MB		 138}
139
9c27847d 140static const struct kernel_param_ops null_queue_mode_param_ops = {
709c8667
MB		 141 .set = null_set_queue_mode,
142 .get = param_get_int,
143};
144
5657a819 145device_param_cb(queue_mode, &null_queue_mode_param_ops, &g_queue_mode, 0444);
54ae81cd 146MODULE_PARM_DESC(queue_mode, "Block interface to use (0=bio,1=rq,2=multiqueue)");
f2298c04 147
2984c868 148static int g_gb = 250;
5657a819 149module_param_named(gb, g_gb, int, 0444);
f2298c04
JA		 150MODULE_PARM_DESC(gb, "Size in GB");
151
2984c868 152static int g_bs = 512;
5657a819 153module_param_named(bs, g_bs, int, 0444);
f2298c04
JA		 154MODULE_PARM_DESC(bs, "Block size (in bytes)");
155
ea17fd35
DLM		 156static int g_max_sectors;
157module_param_named(max_sectors, g_max_sectors, int, 0444);
158MODULE_PARM_DESC(max_sectors, "Maximum size of a command (in 512B sectors)");
159
f7c4ce89 160static unsigned int nr_devices = 1;
701dfc42 161module_param(nr_devices, uint, 0444);
f2298c04
JA		 162MODULE_PARM_DESC(nr_devices, "Number of devices to register");
163
2984c868 164static bool g_blocking;
5657a819 165module_param_named(blocking, g_blocking, bool, 0444);
db5bcf87
JA		 166MODULE_PARM_DESC(blocking, "Register as a blocking blk-mq driver device");
167
82f402fe 168static bool shared_tags;
5657a819 169module_param(shared_tags, bool, 0444);
82f402fe
JA		 170MODULE_PARM_DESC(shared_tags, "Share tag set between devices for blk-mq");
171
0905053b
JG		 172static bool g_shared_tag_bitmap;
173module_param_named(shared_tag_bitmap, g_shared_tag_bitmap, bool, 0444);
174MODULE_PARM_DESC(shared_tag_bitmap, "Use shared tag bitmap for all submission queues for blk-mq");
175
2984c868 176static int g_irqmode = NULL_IRQ_SOFTIRQ;
709c8667
MB		 177
178static int null_set_irqmode(const char *str, const struct kernel_param *kp)
179{
2984c868 180 return null_param_store_val(str, &g_irqmode, NULL_IRQ_NONE,
709c8667
MB		 181 NULL_IRQ_TIMER);
182}
183
9c27847d 184static const struct kernel_param_ops null_irqmode_param_ops = {
709c8667
MB		 185 .set = null_set_irqmode,
186 .get = param_get_int,
187};
188
5657a819 189device_param_cb(irqmode, &null_irqmode_param_ops, &g_irqmode, 0444);
f2298c04
JA		 190MODULE_PARM_DESC(irqmode, "IRQ completion handler. 0-none, 1-softirq, 2-timer");
191
2984c868 192static unsigned long g_completion_nsec = 10000;
5657a819 193module_param_named(completion_nsec, g_completion_nsec, ulong, 0444);
f2298c04
JA		 194MODULE_PARM_DESC(completion_nsec, "Time in ns to complete a request in hardware. Default: 10,000ns");
195
2984c868 196static int g_hw_queue_depth = 64;
5657a819 197module_param_named(hw_queue_depth, g_hw_queue_depth, int, 0444);
f2298c04
JA		 198MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 64");
199
2984c868 200static bool g_use_per_node_hctx;
5657a819 201module_param_named(use_per_node_hctx, g_use_per_node_hctx, bool, 0444);
20005244 202MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware context queues. Default: false");
f2298c04 203
058efe00
VF		 204static bool g_memory_backed;
205module_param_named(memory_backed, g_memory_backed, bool, 0444);
206MODULE_PARM_DESC(memory_backed, "Create a memory-backed block device. Default: false");
207
208static bool g_discard;
209module_param_named(discard, g_discard, bool, 0444);
210MODULE_PARM_DESC(discard, "Support discard operations (requires memory-backed null_blk device). Default: false");
211
212static unsigned long g_cache_size;
213module_param_named(cache_size, g_cache_size, ulong, 0444);
214MODULE_PARM_DESC(mbps, "Cache size in MiB for memory-backed device. Default: 0 (none)");
215
216static unsigned int g_mbps;
217module_param_named(mbps, g_mbps, uint, 0444);
218MODULE_PARM_DESC(mbps, "Limit maximum bandwidth (in MiB/s). Default: 0 (no limit)");
219
ca4b2a01
MB		 220static bool g_zoned;
221module_param_named(zoned, g_zoned, bool, S_IRUGO);
222MODULE_PARM_DESC(zoned, "Make device as a host-managed zoned block device. Default: false");
223
224static unsigned long g_zone_size = 256;
225module_param_named(zone_size, g_zone_size, ulong, S_IRUGO);
226MODULE_PARM_DESC(zone_size, "Zone size in MB when block device is zoned. Must be power-of-two: Default: 256");
227
089565fb
AR		 228static unsigned long g_zone_capacity;
229module_param_named(zone_capacity, g_zone_capacity, ulong, 0444);
230MODULE_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");
231
ea2c18e1
MS		 232static unsigned int g_zone_nr_conv;
233module_param_named(zone_nr_conv, g_zone_nr_conv, uint, 0444);
234MODULE_PARM_DESC(zone_nr_conv, "Number of conventional zones when block device is zoned. Default: 0");
235
dc4d137e
NC		 236static unsigned int g_zone_max_open;
237module_param_named(zone_max_open, g_zone_max_open, uint, 0444);
238MODULE_PARM_DESC(zone_max_open, "Maximum number of open zones when block device is zoned. Default: 0 (no limit)");
239
240static unsigned int g_zone_max_active;
241module_param_named(zone_max_active, g_zone_max_active, uint, 0444);
242MODULE_PARM_DESC(zone_max_active, "Maximum number of active zones when block device is zoned. Default: 0 (no limit)");
243
3bf2bd20
SL		 244static struct nullb_device *null_alloc_dev(void);
245static void null_free_dev(struct nullb_device *dev);
cedcafad
SL		 246static void null_del_dev(struct nullb *nullb);
247static int null_add_dev(struct nullb_device *dev);
49c3b926 248static struct nullb *null_find_dev_by_name(const char *name);
deb78b41 249static void null_free_device_storage(struct nullb_device *dev, bool is_cache);
3bf2bd20
SL		 250
251static inline struct nullb_device *to_nullb_device(struct config_item *item)
252{
253 return item ? container_of(item, struct nullb_device, item) : NULL;
254}
255
256static inline ssize_t nullb_device_uint_attr_show(unsigned int val, char *page)
257{
258 return snprintf(page, PAGE_SIZE, "%u\n", val);
259}
260
261static inline ssize_t nullb_device_ulong_attr_show(unsigned long val,
262 char *page)
263{
264 return snprintf(page, PAGE_SIZE, "%lu\n", val);
265}
266
267static inline ssize_t nullb_device_bool_attr_show(bool val, char *page)
268{
269 return snprintf(page, PAGE_SIZE, "%u\n", val);
270}
271
272static ssize_t nullb_device_uint_attr_store(unsigned int *val,
273 const char *page, size_t count)
274{
275 unsigned int tmp;
276 int result;
277
278 result = kstrtouint(page, 0, &tmp);
45919fbf 279 if (result < 0)
3bf2bd20
SL		 280 return result;
281
282 *val = tmp;
283 return count;
284}
285
286static ssize_t nullb_device_ulong_attr_store(unsigned long *val,
287 const char *page, size_t count)
288{
289 int result;
290 unsigned long tmp;
291
292 result = kstrtoul(page, 0, &tmp);
45919fbf 293 if (result < 0)
3bf2bd20
SL		 294 return result;
295
296 *val = tmp;
297 return count;
298}
299
300static ssize_t nullb_device_bool_attr_store(bool *val, const char *page,
301 size_t count)
302{
303 bool tmp;
304 int result;
305
306 result = kstrtobool(page, &tmp);
45919fbf 307 if (result < 0)
3bf2bd20
SL		 308 return result;
309
310 *val = tmp;
311 return count;
312}
313
314/* The following macro should only be used with TYPE = {uint, ulong, bool}. */
ca0a95a6
AM		 315#define NULLB_DEVICE_ATTR(NAME, TYPE, APPLY) \
316static ssize_t \
317nullb_device_##NAME##_show(struct config_item *item, char *page) \
318{ \
319 return nullb_device_##TYPE##_attr_show( \
320 to_nullb_device(item)->NAME, page); \
321} \
322static ssize_t \
323nullb_device_##NAME##_store(struct config_item *item, const char *page, \
324 size_t count) \
325{ \
326 int (*apply_fn)(struct nullb_device *dev, TYPE new_value) = APPLY;\
327 struct nullb_device *dev = to_nullb_device(item); \
b9853b4d 328 TYPE new_value = 0; \
ca0a95a6
AM		 329 int ret; \
330 \
331 ret = nullb_device_##TYPE##_attr_store(&new_value, page, count);\
332 if (ret < 0) \
333 return ret; \
334 if (apply_fn) \
335 ret = apply_fn(dev, new_value); \
336 else if (test_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags)) \
337 ret = -EBUSY; \
338 if (ret < 0) \
339 return ret; \
340 dev->NAME = new_value; \
341 return count; \
342} \
3bf2bd20
SL		 343CONFIGFS_ATTR(nullb_device_, NAME);
344
15dfc662
SK		 345static int nullb_update_nr_hw_queues(struct nullb_device *dev,
346 unsigned int submit_queues,
347 unsigned int poll_queues)
348
45919fbf 349{
45919fbf 350 struct blk_mq_tag_set *set;
15dfc662 351 int ret, nr_hw_queues;
45919fbf 352
15dfc662 353 if (!dev->nullb)
45919fbf
BVA		 354 return 0;
355
15dfc662 356 /*
2bfdbe8b 357 * Make sure at least one submit queue exists.
15dfc662 358 */
2bfdbe8b 359 if (!submit_queues)
15dfc662
SK		 360 return -EINVAL;
361
78b10be2
BVA		 362 /*
363 * Make sure that null_init_hctx() does not access nullb->queues[] past
364 * the end of that array.
365 */
15dfc662 366 if (submit_queues > nr_cpu_ids || poll_queues > g_poll_queues)
78b10be2 367 return -EINVAL;
15dfc662
SK		 368
369 /*
370 * Keep previous and new queue numbers in nullb_device for reference in
371 * the call back function null_map_queues().
372 */
373 dev->prev_submit_queues = dev->submit_queues;
374 dev->prev_poll_queues = dev->poll_queues;
375 dev->submit_queues = submit_queues;
376 dev->poll_queues = poll_queues;
377
378 set = dev->nullb->tag_set;
379 nr_hw_queues = submit_queues + poll_queues;
380 blk_mq_update_nr_hw_queues(set, nr_hw_queues);
381 ret = set->nr_hw_queues == nr_hw_queues ? 0 : -ENOMEM;
382
383 if (ret) {
384 /* on error, revert the queue numbers */
385 dev->submit_queues = dev->prev_submit_queues;
386 dev->poll_queues = dev->prev_poll_queues;
387 }
388
389 return ret;
390}
391
392static int nullb_apply_submit_queues(struct nullb_device *dev,
393 unsigned int submit_queues)
394{
395 return nullb_update_nr_hw_queues(dev, submit_queues, dev->poll_queues);
396}
397
398static int nullb_apply_poll_queues(struct nullb_device *dev,
399 unsigned int poll_queues)
400{
401 return nullb_update_nr_hw_queues(dev, dev->submit_queues, poll_queues);
45919fbf
BVA		 402}
403
404NULLB_DEVICE_ATTR(size, ulong, NULL);
405NULLB_DEVICE_ATTR(completion_nsec, ulong, NULL);
406NULLB_DEVICE_ATTR(submit_queues, uint, nullb_apply_submit_queues);
15dfc662 407NULLB_DEVICE_ATTR(poll_queues, uint, nullb_apply_poll_queues);
45919fbf
BVA		 408NULLB_DEVICE_ATTR(home_node, uint, NULL);
409NULLB_DEVICE_ATTR(queue_mode, uint, NULL);
410NULLB_DEVICE_ATTR(blocksize, uint, NULL);
ea17fd35 411NULLB_DEVICE_ATTR(max_sectors, uint, NULL);
45919fbf
BVA		 412NULLB_DEVICE_ATTR(irqmode, uint, NULL);
413NULLB_DEVICE_ATTR(hw_queue_depth, uint, NULL);
414NULLB_DEVICE_ATTR(index, uint, NULL);
415NULLB_DEVICE_ATTR(blocking, bool, NULL);
416NULLB_DEVICE_ATTR(use_per_node_hctx, bool, NULL);
417NULLB_DEVICE_ATTR(memory_backed, bool, NULL);
418NULLB_DEVICE_ATTR(discard, bool, NULL);
419NULLB_DEVICE_ATTR(mbps, uint, NULL);
420NULLB_DEVICE_ATTR(cache_size, ulong, NULL);
421NULLB_DEVICE_ATTR(zoned, bool, NULL);
422NULLB_DEVICE_ATTR(zone_size, ulong, NULL);
089565fb 423NULLB_DEVICE_ATTR(zone_capacity, ulong, NULL);
45919fbf 424NULLB_DEVICE_ATTR(zone_nr_conv, uint, NULL);
dc4d137e
NC		 425NULLB_DEVICE_ATTR(zone_max_open, uint, NULL);
426NULLB_DEVICE_ATTR(zone_max_active, uint, NULL);
cee1b215 427NULLB_DEVICE_ATTR(virt_boundary, bool, NULL);
7012eef5
VF		 428NULLB_DEVICE_ATTR(no_sched, bool, NULL);
429NULLB_DEVICE_ATTR(shared_tag_bitmap, bool, NULL);
3bf2bd20 430
cedcafad
SL		 431static ssize_t nullb_device_power_show(struct config_item *item, char *page)
432{
433 return nullb_device_bool_attr_show(to_nullb_device(item)->power, page);
434}
435
436static ssize_t nullb_device_power_store(struct config_item *item,
437 const char *page, size_t count)
438{
439 struct nullb_device *dev = to_nullb_device(item);
440 bool newp = false;
441 ssize_t ret;
442
443 ret = nullb_device_bool_attr_store(&newp, page, count);
444 if (ret < 0)
445 return ret;
446
447 if (!dev->power && newp) {
448 if (test_and_set_bit(NULLB_DEV_FL_UP, &dev->flags))
449 return count;
a75110c3
CK		 450 ret = null_add_dev(dev);
451 if (ret) {
cedcafad 452 clear_bit(NULLB_DEV_FL_UP, &dev->flags);
a75110c3 453 return ret;
cedcafad
SL		 454 }
455
456 set_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
457 dev->power = newp;
b3c30512 458 } else if (dev->power && !newp) {
7602843f
BL		 459 if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
460 mutex_lock(&lock);
461 dev->power = newp;
462 null_del_dev(dev->nullb);
463 mutex_unlock(&lock);
464 }
00a8cdb8 465 clear_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
cedcafad
SL		 466 }
467
468 return count;
469}
470
471CONFIGFS_ATTR(nullb_device_, power);
472
2f54a613
SL		 473static ssize_t nullb_device_badblocks_show(struct config_item *item, char *page)
474{
475 struct nullb_device *t_dev = to_nullb_device(item);
476
477 return badblocks_show(&t_dev->badblocks, page, 0);
478}
479
480static ssize_t nullb_device_badblocks_store(struct config_item *item,
481 const char *page, size_t count)
482{
483 struct nullb_device *t_dev = to_nullb_device(item);
484 char *orig, *buf, *tmp;
485 u64 start, end;
486 int ret;
487
488 orig = kstrndup(page, count, GFP_KERNEL);
489 if (!orig)
490 return -ENOMEM;
491
492 buf = strstrip(orig);
493
494 ret = -EINVAL;
495 if (buf[0] != '+' && buf[0] != '-')
496 goto out;
497 tmp = strchr(&buf[1], '-');
498 if (!tmp)
499 goto out;
500 *tmp = '\0';
501 ret = kstrtoull(buf + 1, 0, &start);
502 if (ret)
503 goto out;
504 ret = kstrtoull(tmp + 1, 0, &end);
505 if (ret)
506 goto out;
507 ret = -EINVAL;
508 if (start > end)
509 goto out;
510 /* enable badblocks */
511 cmpxchg(&t_dev->badblocks.shift, -1, 0);
512 if (buf[0] == '+')
513 ret = badblocks_set(&t_dev->badblocks, start,
514 end - start + 1, 1);
515 else
516 ret = badblocks_clear(&t_dev->badblocks, start,
517 end - start + 1);
518 if (ret == 0)
519 ret = count;
520out:
521 kfree(orig);
522 return ret;
523}
524CONFIGFS_ATTR(nullb_device_, badblocks);
525
d3a57388
SK		 526static ssize_t nullb_device_zone_readonly_store(struct config_item *item,
527 const char *page, size_t count)
528{
529 struct nullb_device *dev = to_nullb_device(item);
530
531 return zone_cond_store(dev, page, count, BLK_ZONE_COND_READONLY);
532}
533CONFIGFS_ATTR_WO(nullb_device_, zone_readonly);
534
535static ssize_t nullb_device_zone_offline_store(struct config_item *item,
536 const char *page, size_t count)
537{
538 struct nullb_device *dev = to_nullb_device(item);
539
540 return zone_cond_store(dev, page, count, BLK_ZONE_COND_OFFLINE);
541}
542CONFIGFS_ATTR_WO(nullb_device_, zone_offline);
543
3bf2bd20
SL		 544static struct configfs_attribute *nullb_device_attrs[] = {
545 &nullb_device_attr_size,
546 &nullb_device_attr_completion_nsec,
547 &nullb_device_attr_submit_queues,
0a593fbb 548 &nullb_device_attr_poll_queues,
3bf2bd20
SL		 549 &nullb_device_attr_home_node,
550 &nullb_device_attr_queue_mode,
551 &nullb_device_attr_blocksize,
ea17fd35 552 &nullb_device_attr_max_sectors,
3bf2bd20
SL		 553 &nullb_device_attr_irqmode,
554 &nullb_device_attr_hw_queue_depth,
cedcafad 555 &nullb_device_attr_index,
3bf2bd20
SL		 556 &nullb_device_attr_blocking,
557 &nullb_device_attr_use_per_node_hctx,
cedcafad 558 &nullb_device_attr_power,
5bcd0e0c 559 &nullb_device_attr_memory_backed,
306eb6b4 560 &nullb_device_attr_discard,
eff2c4f1 561 &nullb_device_attr_mbps,
deb78b41 562 &nullb_device_attr_cache_size,
2f54a613 563 &nullb_device_attr_badblocks,
ca4b2a01
MB		 564 &nullb_device_attr_zoned,
565 &nullb_device_attr_zone_size,
089565fb 566 &nullb_device_attr_zone_capacity,
ea2c18e1 567 &nullb_device_attr_zone_nr_conv,
dc4d137e
NC		 568 &nullb_device_attr_zone_max_open,
569 &nullb_device_attr_zone_max_active,
d3a57388
SK		 570 &nullb_device_attr_zone_readonly,
571 &nullb_device_attr_zone_offline,
cee1b215 572 &nullb_device_attr_virt_boundary,
7012eef5
VF		 573 &nullb_device_attr_no_sched,
574 &nullb_device_attr_shared_tag_bitmap,
3bf2bd20
SL		 575 NULL,
576};
577
578static void nullb_device_release(struct config_item *item)
579{
5bcd0e0c
SL		 580 struct nullb_device *dev = to_nullb_device(item);
581
deb78b41 582 null_free_device_storage(dev, false);
5bcd0e0c 583 null_free_dev(dev);
3bf2bd20
SL		 584}
585
586static struct configfs_item_operations nullb_device_ops = {
587 .release = nullb_device_release,
588};
589
e1919dff 590static const struct config_item_type nullb_device_type = {
3bf2bd20
SL		 591 .ct_item_ops = &nullb_device_ops,
592 .ct_attrs = nullb_device_attrs,
593 .ct_owner = THIS_MODULE,
594};
595
596static struct
597config_item *nullb_group_make_item(struct config_group *group, const char *name)
598{
599 struct nullb_device *dev;
600
49c3b926
DLM		 601 if (null_find_dev_by_name(name))
602 return ERR_PTR(-EEXIST);
603
3bf2bd20
SL		 604 dev = null_alloc_dev();
605 if (!dev)
606 return ERR_PTR(-ENOMEM);
607
608 config_item_init_type_name(&dev->item, name, &nullb_device_type);
609
610 return &dev->item;
611}
612
613static void
614nullb_group_drop_item(struct config_group *group, struct config_item *item)
615{
cedcafad
SL		 616 struct nullb_device *dev = to_nullb_device(item);
617
618 if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
619 mutex_lock(&lock);
620 dev->power = false;
621 null_del_dev(dev->nullb);
622 mutex_unlock(&lock);
623 }
624
3bf2bd20
SL		 625 config_item_put(item);
626}
627
628static ssize_t memb_group_features_show(struct config_item *item, char *page)
629{
089565fb 630 return snprintf(page, PAGE_SIZE,
7012eef5
VF		 631 "badblocks,blocking,blocksize,cache_size,"
632 "completion_nsec,discard,home_node,hw_queue_depth,"
633 "irqmode,max_sectors,mbps,memory_backed,no_sched,"
634 "poll_queues,power,queue_mode,shared_tag_bitmap,size,"
635 "submit_queues,use_per_node_hctx,virt_boundary,zoned,"
636 "zone_capacity,zone_max_active,zone_max_open,"
d3a57388 637 "zone_nr_conv,zone_offline,zone_readonly,zone_size\n");
3bf2bd20
SL		 638}
639
640CONFIGFS_ATTR_RO(memb_group_, features);
641
642static struct configfs_attribute *nullb_group_attrs[] = {
643 &memb_group_attr_features,
644 NULL,
645};
646
647static struct configfs_group_operations nullb_group_ops = {
648 .make_item = nullb_group_make_item,
649 .drop_item = nullb_group_drop_item,
650};
651
e1919dff 652static const struct config_item_type nullb_group_type = {
3bf2bd20
SL		 653 .ct_group_ops = &nullb_group_ops,
654 .ct_attrs = nullb_group_attrs,
655 .ct_owner = THIS_MODULE,
656};
657
658static struct configfs_subsystem nullb_subsys = {
659 .su_group = {
660 .cg_item = {
661 .ci_namebuf = "nullb",
662 .ci_type = &nullb_group_type,
663 },
664 },
665};
666
deb78b41
SL		 667static inline int null_cache_active(struct nullb *nullb)
668{
669 return test_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
670}
671
2984c868
SL		 672static struct nullb_device *null_alloc_dev(void)
673{
674 struct nullb_device *dev;
675
676 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
677 if (!dev)
678 return NULL;
5bcd0e0c 679 INIT_RADIX_TREE(&dev->data, GFP_ATOMIC);
deb78b41 680 INIT_RADIX_TREE(&dev->cache, GFP_ATOMIC);
2f54a613
SL		 681 if (badblocks_init(&dev->badblocks, 0)) {
682 kfree(dev);
683 return NULL;
684 }
685
2984c868
SL		 686 dev->size = g_gb * 1024;
687 dev->completion_nsec = g_completion_nsec;
688 dev->submit_queues = g_submit_queues;
15dfc662 689 dev->prev_submit_queues = g_submit_queues;
0a593fbb 690 dev->poll_queues = g_poll_queues;
15dfc662 691 dev->prev_poll_queues = g_poll_queues;
2984c868
SL		 692 dev->home_node = g_home_node;
693 dev->queue_mode = g_queue_mode;
694 dev->blocksize = g_bs;
ea17fd35 695 dev->max_sectors = g_max_sectors;
2984c868
SL		 696 dev->irqmode = g_irqmode;
697 dev->hw_queue_depth = g_hw_queue_depth;
2984c868 698 dev->blocking = g_blocking;
058efe00
VF		 699 dev->memory_backed = g_memory_backed;
700 dev->discard = g_discard;
701 dev->cache_size = g_cache_size;
702 dev->mbps = g_mbps;
2984c868 703 dev->use_per_node_hctx = g_use_per_node_hctx;
ca4b2a01
MB		 704 dev->zoned = g_zoned;
705 dev->zone_size = g_zone_size;
089565fb 706 dev->zone_capacity = g_zone_capacity;
ea2c18e1 707 dev->zone_nr_conv = g_zone_nr_conv;
dc4d137e
NC		 708 dev->zone_max_open = g_zone_max_open;
709 dev->zone_max_active = g_zone_max_active;
cee1b215 710 dev->virt_boundary = g_virt_boundary;
7012eef5
VF		 711 dev->no_sched = g_no_sched;
712 dev->shared_tag_bitmap = g_shared_tag_bitmap;
2984c868
SL		 713 return dev;
714}
715
716static void null_free_dev(struct nullb_device *dev)
717{
1addb798
DD		 718 if (!dev)
719 return;
720
d205bde7 721 null_free_zoned_dev(dev);
1addb798 722 badblocks_exit(&dev->badblocks);
2984c868
SL		 723 kfree(dev);
724}
725
f2298c04
JA		 726static void put_tag(struct nullb_queue *nq, unsigned int tag)
727{
728 clear_bit_unlock(tag, nq->tag_map);
729
730 if (waitqueue_active(&nq->wait))
731 wake_up(&nq->wait);
732}
733
734static unsigned int get_tag(struct nullb_queue *nq)
735{
736 unsigned int tag;
737
738 do {
739 tag = find_first_zero_bit(nq->tag_map, nq->queue_depth);
740 if (tag >= nq->queue_depth)
741 return -1U;
742 } while (test_and_set_bit_lock(tag, nq->tag_map));
743
744 return tag;
745}
746
747static void free_cmd(struct nullb_cmd *cmd)
748{
749 put_tag(cmd->nq, cmd->tag);
750}
751
3c395a96
PV		 752static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer);
753
f2298c04
JA		 754static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq)
755{
756 struct nullb_cmd *cmd;
757 unsigned int tag;
758
759 tag = get_tag(nq);
760 if (tag != -1U) {
761 cmd = &nq->cmds[tag];
762 cmd->tag = tag;
ff770422 763 cmd->error = BLK_STS_OK;
f2298c04 764 cmd->nq = nq;
2984c868 765 if (nq->dev->irqmode == NULL_IRQ_TIMER) {
3c395a96
PV		 766 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC,
767 HRTIMER_MODE_REL);
768 cmd->timer.function = null_cmd_timer_expired;
769 }
f2298c04
JA		 770 return cmd;
771 }
772
773 return NULL;
774}
775
3d3472f3 776static struct nullb_cmd *alloc_cmd(struct nullb_queue *nq, struct bio *bio)
f2298c04
JA		 777{
778 struct nullb_cmd *cmd;
779 DEFINE_WAIT(wait);
780
f2298c04 781 do {
3d3472f3
CK		 782 /*
783 * This avoids multiple return statements, multiple calls to
784 * __alloc_cmd() and a fast path call to prepare_to_wait().
785 */
f2298c04 786 cmd = __alloc_cmd(nq);
3d3472f3
CK		 787 if (cmd) {
788 cmd->bio = bio;
789 return cmd;
790 }
791 prepare_to_wait(&nq->wait, &wait, TASK_UNINTERRUPTIBLE);
f2298c04 792 io_schedule();
3d3472f3 793 finish_wait(&nq->wait, &wait);
f2298c04 794 } while (1);
f2298c04
JA		 795}
796
797static void end_cmd(struct nullb_cmd *cmd)
798{
2984c868 799 int queue_mode = cmd->nq->dev->queue_mode;
cf8ecc5a 800
ce2c350b
CH		 801 switch (queue_mode) {
802 case NULL_Q_MQ:
5bcd0e0c 803 blk_mq_end_request(cmd->rq, cmd->error);
ce2c350b 804 return;
ce2c350b 805 case NULL_Q_BIO:
5bcd0e0c 806 cmd->bio->bi_status = cmd->error;
4246a0b6 807 bio_endio(cmd->bio);
48cc661e 808 break;
ce2c350b 809 }
f2298c04 810
48cc661e 811 free_cmd(cmd);
cf8ecc5a
AA		 812}
813
814static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
815{
816 end_cmd(container_of(timer, struct nullb_cmd, timer));
f2298c04
JA		 817
818 return HRTIMER_NORESTART;
819}
820
821static void null_cmd_end_timer(struct nullb_cmd *cmd)
822{
2984c868 823 ktime_t kt = cmd->nq->dev->completion_nsec;
f2298c04 824
3c395a96 825 hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL);
f2298c04
JA		 826}
827
49f66136 828static void null_complete_rq(struct request *rq)
f2298c04 829{
49f66136 830 end_cmd(blk_mq_rq_to_pdu(rq));
f2298c04
JA		 831}
832
c90b6b50 833static struct nullb_page *null_alloc_page(void)
5bcd0e0c
SL		 834{
835 struct nullb_page *t_page;
836
c90b6b50 837 t_page = kmalloc(sizeof(struct nullb_page), GFP_NOIO);
5bcd0e0c 838 if (!t_page)
df00b1d2 839 return NULL;
5bcd0e0c 840
c90b6b50 841 t_page->page = alloc_pages(GFP_NOIO, 0);
df00b1d2
CK		 842 if (!t_page->page) {
843 kfree(t_page);
844 return NULL;
845 }
5bcd0e0c 846
66231ad3 847 memset(t_page->bitmap, 0, sizeof(t_page->bitmap));
5bcd0e0c 848 return t_page;
5bcd0e0c
SL		 849}
850
851static void null_free_page(struct nullb_page *t_page)
852{
66231ad3
ML		 853 __set_bit(NULLB_PAGE_FREE, t_page->bitmap);
854 if (test_bit(NULLB_PAGE_LOCK, t_page->bitmap))
deb78b41 855 return;
5bcd0e0c
SL		 856 __free_page(t_page->page);
857 kfree(t_page);
858}
859
66231ad3
ML		 860static bool null_page_empty(struct nullb_page *page)
861{
862 int size = MAP_SZ - 2;
863
864 return find_first_bit(page->bitmap, size) == size;
865}
866
deb78b41
SL		 867static void null_free_sector(struct nullb *nullb, sector_t sector,
868 bool is_cache)
5bcd0e0c
SL		 869{
870 unsigned int sector_bit;
871 u64 idx;
872 struct nullb_page *t_page, *ret;
873 struct radix_tree_root *root;
874
deb78b41 875 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
5bcd0e0c
SL		 876 idx = sector >> PAGE_SECTORS_SHIFT;
877 sector_bit = (sector & SECTOR_MASK);
878
879 t_page = radix_tree_lookup(root, idx);
880 if (t_page) {
66231ad3 881 __clear_bit(sector_bit, t_page->bitmap);
5bcd0e0c 882
66231ad3 883 if (null_page_empty(t_page)) {
5bcd0e0c
SL		 884 ret = radix_tree_delete_item(root, idx, t_page);
885 WARN_ON(ret != t_page);
886 null_free_page(ret);
deb78b41
SL		 887 if (is_cache)
888 nullb->dev->curr_cache -= PAGE_SIZE;
5bcd0e0c
SL		 889 }
890 }
891}
892
893static struct nullb_page *null_radix_tree_insert(struct nullb *nullb, u64 idx,
deb78b41 894 struct nullb_page *t_page, bool is_cache)
5bcd0e0c
SL		 895{
896 struct radix_tree_root *root;
897
deb78b41 898 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
5bcd0e0c
SL		 899
900 if (radix_tree_insert(root, idx, t_page)) {
901 null_free_page(t_page);
902 t_page = radix_tree_lookup(root, idx);
903 WARN_ON(!t_page || t_page->page->index != idx);
deb78b41
SL		 904 } else if (is_cache)
905 nullb->dev->curr_cache += PAGE_SIZE;
5bcd0e0c
SL		 906
907 return t_page;
908}
909
deb78b41 910static void null_free_device_storage(struct nullb_device *dev, bool is_cache)
5bcd0e0c
SL		 911{
912 unsigned long pos = 0;
913 int nr_pages;
914 struct nullb_page *ret, *t_pages[FREE_BATCH];
915 struct radix_tree_root *root;
916
deb78b41 917 root = is_cache ? &dev->cache : &dev->data;
5bcd0e0c
SL		 918
919 do {
920 int i;
921
922 nr_pages = radix_tree_gang_lookup(root,
923 (void **)t_pages, pos, FREE_BATCH);
924
925 for (i = 0; i < nr_pages; i++) {
926 pos = t_pages[i]->page->index;
927 ret = radix_tree_delete_item(root, pos, t_pages[i]);
928 WARN_ON(ret != t_pages[i]);
929 null_free_page(ret);
930 }
931
932 pos++;
933 } while (nr_pages == FREE_BATCH);
deb78b41
SL		 934
935 if (is_cache)
936 dev->curr_cache = 0;
5bcd0e0c
SL		 937}
938
deb78b41
SL		 939static struct nullb_page *__null_lookup_page(struct nullb *nullb,
940 sector_t sector, bool for_write, bool is_cache)
5bcd0e0c
SL		 941{
942 unsigned int sector_bit;
943 u64 idx;
944 struct nullb_page *t_page;
deb78b41 945 struct radix_tree_root *root;
5bcd0e0c
SL		 946
947 idx = sector >> PAGE_SECTORS_SHIFT;
948 sector_bit = (sector & SECTOR_MASK);
949
deb78b41
SL		 950 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
951 t_page = radix_tree_lookup(root, idx);
5bcd0e0c
SL		 952 WARN_ON(t_page && t_page->page->index != idx);
953
66231ad3 954 if (t_page && (for_write || test_bit(sector_bit, t_page->bitmap)))
5bcd0e0c
SL		 955 return t_page;
956
957 return NULL;
958}
959
deb78b41
SL		 960static struct nullb_page *null_lookup_page(struct nullb *nullb,
961 sector_t sector, bool for_write, bool ignore_cache)
962{
963 struct nullb_page *page = NULL;
964
965 if (!ignore_cache)
966 page = __null_lookup_page(nullb, sector, for_write, true);
967 if (page)
968 return page;
969 return __null_lookup_page(nullb, sector, for_write, false);
970}
971
5bcd0e0c 972static struct nullb_page *null_insert_page(struct nullb *nullb,
61884de0
JA		 973 sector_t sector, bool ignore_cache)
974 __releases(&nullb->lock)
975 __acquires(&nullb->lock)
5bcd0e0c
SL		 976{
977 u64 idx;
978 struct nullb_page *t_page;
979
deb78b41 980 t_page = null_lookup_page(nullb, sector, true, ignore_cache);
5bcd0e0c
SL		 981 if (t_page)
982 return t_page;
983
984 spin_unlock_irq(&nullb->lock);
985
c90b6b50 986 t_page = null_alloc_page();
5bcd0e0c
SL		 987 if (!t_page)
988 goto out_lock;
989
990 if (radix_tree_preload(GFP_NOIO))
991 goto out_freepage;
992
993 spin_lock_irq(&nullb->lock);
994 idx = sector >> PAGE_SECTORS_SHIFT;
995 t_page->page->index = idx;
deb78b41 996 t_page = null_radix_tree_insert(nullb, idx, t_page, !ignore_cache);
5bcd0e0c
SL		 997 radix_tree_preload_end();
998
999 return t_page;
1000out_freepage:
1001 null_free_page(t_page);
1002out_lock:
1003 spin_lock_irq(&nullb->lock);
deb78b41
SL		 1004 return null_lookup_page(nullb, sector, true, ignore_cache);
1005}
1006
1007static int null_flush_cache_page(struct nullb *nullb, struct nullb_page *c_page)
1008{
1009 int i;
1010 unsigned int offset;
1011 u64 idx;
1012 struct nullb_page *t_page, *ret;
1013 void *dst, *src;
1014
1015 idx = c_page->page->index;
1016
1017 t_page = null_insert_page(nullb, idx << PAGE_SECTORS_SHIFT, true);
1018
66231ad3
ML		 1019 __clear_bit(NULLB_PAGE_LOCK, c_page->bitmap);
1020 if (test_bit(NULLB_PAGE_FREE, c_page->bitmap)) {
deb78b41 1021 null_free_page(c_page);
66231ad3 1022 if (t_page && null_page_empty(t_page)) {
deb78b41
SL		 1023 ret = radix_tree_delete_item(&nullb->dev->data,
1024 idx, t_page);
1025 null_free_page(t_page);
1026 }
1027 return 0;
1028 }
1029
1030 if (!t_page)
1031 return -ENOMEM;
1032
acc3c879
CK		 1033 src = kmap_local_page(c_page->page);
1034 dst = kmap_local_page(t_page->page);
deb78b41
SL		 1035
1036 for (i = 0; i < PAGE_SECTORS;
1037 i += (nullb->dev->blocksize >> SECTOR_SHIFT)) {
66231ad3 1038 if (test_bit(i, c_page->bitmap)) {
deb78b41
SL		 1039 offset = (i << SECTOR_SHIFT);
1040 memcpy(dst + offset, src + offset,
1041 nullb->dev->blocksize);
66231ad3 1042 __set_bit(i, t_page->bitmap);
deb78b41
SL		 1043 }
1044 }
1045
acc3c879
CK		 1046 kunmap_local(dst);
1047 kunmap_local(src);
deb78b41
SL		 1048
1049 ret = radix_tree_delete_item(&nullb->dev->cache, idx, c_page);
1050 null_free_page(ret);
1051 nullb->dev->curr_cache -= PAGE_SIZE;
1052
1053 return 0;
1054}
1055
1056static int null_make_cache_space(struct nullb *nullb, unsigned long n)
f2298c04 1057{
deb78b41
SL		 1058 int i, err, nr_pages;
1059 struct nullb_page *c_pages[FREE_BATCH];
1060 unsigned long flushed = 0, one_round;
1061
1062again:
1063 if ((nullb->dev->cache_size * 1024 * 1024) >
1064 nullb->dev->curr_cache + n || nullb->dev->curr_cache == 0)
1065 return 0;
1066
1067 nr_pages = radix_tree_gang_lookup(&nullb->dev->cache,
1068 (void **)c_pages, nullb->cache_flush_pos, FREE_BATCH);
1069 /*
1070 * nullb_flush_cache_page could unlock before using the c_pages. To
1071 * avoid race, we don't allow page free
1072 */
1073 for (i = 0; i < nr_pages; i++) {
1074 nullb->cache_flush_pos = c_pages[i]->page->index;
1075 /*
1076 * We found the page which is being flushed to disk by other
1077 * threads
1078 */
66231ad3 1079 if (test_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap))
deb78b41
SL		 1080 c_pages[i] = NULL;
1081 else
66231ad3 1082 __set_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap);
deb78b41
SL		 1083 }
1084
1085 one_round = 0;
1086 for (i = 0; i < nr_pages; i++) {
1087 if (c_pages[i] == NULL)
1088 continue;
1089 err = null_flush_cache_page(nullb, c_pages[i]);
1090 if (err)
1091 return err;
1092 one_round++;
1093 }
1094 flushed += one_round << PAGE_SHIFT;
1095
1096 if (n > flushed) {
1097 if (nr_pages == 0)
1098 nullb->cache_flush_pos = 0;
1099 if (one_round == 0) {
1100 /* give other threads a chance */
1101 spin_unlock_irq(&nullb->lock);
1102 spin_lock_irq(&nullb->lock);
1103 }
1104 goto again;
1105 }
1106 return 0;
5bcd0e0c
SL		 1107}
1108
1109static int copy_to_nullb(struct nullb *nullb, struct page *source,
deb78b41 1110 unsigned int off, sector_t sector, size_t n, bool is_fua)
5bcd0e0c
SL		 1111{
1112 size_t temp, count = 0;
1113 unsigned int offset;
1114 struct nullb_page *t_page;
5bcd0e0c
SL		 1115
1116 while (count < n) {
1117 temp = min_t(size_t, nullb->dev->blocksize, n - count);
1118
deb78b41
SL		 1119 if (null_cache_active(nullb) && !is_fua)
1120 null_make_cache_space(nullb, PAGE_SIZE);
1121
5bcd0e0c 1122 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
deb78b41
SL		 1123 t_page = null_insert_page(nullb, sector,
1124 !null_cache_active(nullb) || is_fua);
5bcd0e0c
SL		 1125 if (!t_page)
1126 return -ENOSPC;
1127
fbb5615f 1128 memcpy_page(t_page->page, offset, source, off + count, temp);
5bcd0e0c 1129
66231ad3 1130 __set_bit(sector & SECTOR_MASK, t_page->bitmap);
5bcd0e0c 1131
deb78b41
SL		 1132 if (is_fua)
1133 null_free_sector(nullb, sector, true);
1134
5bcd0e0c
SL		 1135 count += temp;
1136 sector += temp >> SECTOR_SHIFT;
1137 }
1138 return 0;
1139}
1140
1141static int copy_from_nullb(struct nullb *nullb, struct page *dest,
1142 unsigned int off, sector_t sector, size_t n)
1143{
1144 size_t temp, count = 0;
1145 unsigned int offset;
1146 struct nullb_page *t_page;
5bcd0e0c
SL		 1147
1148 while (count < n) {
1149 temp = min_t(size_t, nullb->dev->blocksize, n - count);
1150
1151 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
deb78b41
SL		 1152 t_page = null_lookup_page(nullb, sector, false,
1153 !null_cache_active(nullb));
5bcd0e0c 1154
fbb5615f
CK		 1155 if (t_page)
1156 memcpy_page(dest, off + count, t_page->page, offset,
1157 temp);
1158 else
1159 zero_user(dest, off + count, temp);
5bcd0e0c
SL		 1160
1161 count += temp;
1162 sector += temp >> SECTOR_SHIFT;
1163 }
1164 return 0;
1165}
1166
dd85b492
AJ		 1167static void nullb_fill_pattern(struct nullb *nullb, struct page *page,
1168 unsigned int len, unsigned int off)
1169{
fbb5615f 1170 memset_page(page, off, 0xff, len);
dd85b492
AJ		 1171}
1172
0ec4d913
DLM		 1173blk_status_t null_handle_discard(struct nullb_device *dev,
1174 sector_t sector, sector_t nr_sectors)
306eb6b4 1175{
49c7089f
DLM		 1176 struct nullb *nullb = dev->nullb;
1177 size_t n = nr_sectors << SECTOR_SHIFT;
306eb6b4
SL		 1178 size_t temp;
1179
1180 spin_lock_irq(&nullb->lock);
1181 while (n > 0) {
49c7089f 1182 temp = min_t(size_t, n, dev->blocksize);
deb78b41
SL		 1183 null_free_sector(nullb, sector, false);
1184 if (null_cache_active(nullb))
1185 null_free_sector(nullb, sector, true);
306eb6b4
SL		 1186 sector += temp >> SECTOR_SHIFT;
1187 n -= temp;
1188 }
1189 spin_unlock_irq(&nullb->lock);
49c7089f
DLM		 1190
1191 return BLK_STS_OK;
306eb6b4
SL		 1192}
1193
deb78b41
SL		 1194static int null_handle_flush(struct nullb *nullb)
1195{
1196 int err;
1197
1198 if (!null_cache_active(nullb))
1199 return 0;
1200
1201 spin_lock_irq(&nullb->lock);
1202 while (true) {
1203 err = null_make_cache_space(nullb,
1204 nullb->dev->cache_size * 1024 * 1024);
1205 if (err || nullb->dev->curr_cache == 0)
1206 break;
1207 }
1208
1209 WARN_ON(!radix_tree_empty(&nullb->dev->cache));
1210 spin_unlock_irq(&nullb->lock);
1211 return err;
1212}
1213
5bcd0e0c 1214static int null_transfer(struct nullb *nullb, struct page *page,
deb78b41
SL		 1215 unsigned int len, unsigned int off, bool is_write, sector_t sector,
1216 bool is_fua)
f2298c04 1217{
dd85b492
AJ		 1218 struct nullb_device *dev = nullb->dev;
1219 unsigned int valid_len = len;
5bcd0e0c
SL		 1220 int err = 0;
1221
1222 if (!is_write) {
dd85b492
AJ		 1223 if (dev->zoned)
1224 valid_len = null_zone_valid_read_len(nullb,
1225 sector, len);
1226
1227 if (valid_len) {
1228 err = copy_from_nullb(nullb, page, off,
1229 sector, valid_len);
1230 off += valid_len;
1231 len -= valid_len;
1232 }
1233
1234 if (len)
1235 nullb_fill_pattern(nullb, page, len, off);
5bcd0e0c
SL		 1236 flush_dcache_page(page);
1237 } else {
1238 flush_dcache_page(page);
deb78b41 1239 err = copy_to_nullb(nullb, page, off, sector, len, is_fua);
5bcd0e0c
SL		 1240 }
1241
1242 return err;
1243}
1244
1245static int null_handle_rq(struct nullb_cmd *cmd)
1246{
1247 struct request *rq = cmd->rq;
1248 struct nullb *nullb = cmd->nq->dev->nullb;
1249 int err;
1250 unsigned int len;
49c7089f 1251 sector_t sector = blk_rq_pos(rq);
5bcd0e0c
SL		 1252 struct req_iterator iter;
1253 struct bio_vec bvec;
1254
5bcd0e0c
SL		 1255 spin_lock_irq(&nullb->lock);
1256 rq_for_each_segment(bvec, rq, iter) {
1257 len = bvec.bv_len;
1258 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
deb78b41 1259 op_is_write(req_op(rq)), sector,
2d62e6b0 1260 rq->cmd_flags & REQ_FUA);
5bcd0e0c
SL		 1261 if (err) {
1262 spin_unlock_irq(&nullb->lock);
1263 return err;
1264 }
1265 sector += len >> SECTOR_SHIFT;
1266 }
1267 spin_unlock_irq(&nullb->lock);
1268
1269 return 0;
1270}
1271
1272static int null_handle_bio(struct nullb_cmd *cmd)
1273{
1274 struct bio *bio = cmd->bio;
1275 struct nullb *nullb = cmd->nq->dev->nullb;
1276 int err;
1277 unsigned int len;
49c7089f 1278 sector_t sector = bio->bi_iter.bi_sector;
5bcd0e0c
SL		 1279 struct bio_vec bvec;
1280 struct bvec_iter iter;
1281
5bcd0e0c
SL		 1282 spin_lock_irq(&nullb->lock);
1283 bio_for_each_segment(bvec, bio, iter) {
1284 len = bvec.bv_len;
1285 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
deb78b41 1286 op_is_write(bio_op(bio)), sector,
bf7c7a04 1287 bio->bi_opf & REQ_FUA);
5bcd0e0c
SL		 1288 if (err) {
1289 spin_unlock_irq(&nullb->lock);
1290 return err;
1291 }
1292 sector += len >> SECTOR_SHIFT;
1293 }
1294 spin_unlock_irq(&nullb->lock);
1295 return 0;
1296}
1297
eff2c4f1
SL		 1298static void null_stop_queue(struct nullb *nullb)
1299{
1300 struct request_queue *q = nullb->q;
1301
1302 if (nullb->dev->queue_mode == NULL_Q_MQ)
1303 blk_mq_stop_hw_queues(q);
eff2c4f1
SL		 1304}
1305
1306static void null_restart_queue_async(struct nullb *nullb)
1307{
1308 struct request_queue *q = nullb->q;
eff2c4f1
SL		 1309
1310 if (nullb->dev->queue_mode == NULL_Q_MQ)
1311 blk_mq_start_stopped_hw_queues(q, true);
eff2c4f1
SL		 1312}
1313
adb84284
CK		 1314static inline blk_status_t null_handle_throttled(struct nullb_cmd *cmd)
1315{
1316 struct nullb_device *dev = cmd->nq->dev;
1317 struct nullb *nullb = dev->nullb;
1318 blk_status_t sts = BLK_STS_OK;
1319 struct request *rq = cmd->rq;
1320
1321 if (!hrtimer_active(&nullb->bw_timer))
1322 hrtimer_restart(&nullb->bw_timer);
1323
1324 if (atomic_long_sub_return(blk_rq_bytes(rq), &nullb->cur_bytes) < 0) {
1325 null_stop_queue(nullb);
1326 /* race with timer */
1327 if (atomic_long_read(&nullb->cur_bytes) > 0)
1328 null_restart_queue_async(nullb);
1329 /* requeue request */
1330 sts = BLK_STS_DEV_RESOURCE;
1331 }
1332 return sts;
1333}
1334
8f94d1c1
CK		 1335static inline blk_status_t null_handle_badblocks(struct nullb_cmd *cmd,
1336 sector_t sector,
1337 sector_t nr_sectors)
1338{
1339 struct badblocks *bb = &cmd->nq->dev->badblocks;
1340 sector_t first_bad;
1341 int bad_sectors;
1342
1343 if (badblocks_check(bb, sector, nr_sectors, &first_bad, &bad_sectors))
1344 return BLK_STS_IOERR;
1345
1346 return BLK_STS_OK;
1347}
1348
7ea88e22 1349static inline blk_status_t null_handle_memory_backed(struct nullb_cmd *cmd,
ff07a02e 1350 enum req_op op,
49c7089f
DLM		 1351 sector_t sector,
1352 sector_t nr_sectors)
7ea88e22
CK		 1353{
1354 struct nullb_device *dev = cmd->nq->dev;
1355 int err;
1356
49c7089f
DLM		 1357 if (op == REQ_OP_DISCARD)
1358 return null_handle_discard(dev, sector, nr_sectors);
1359
7ea88e22
CK		 1360 if (dev->queue_mode == NULL_Q_BIO)
1361 err = null_handle_bio(cmd);
1362 else
1363 err = null_handle_rq(cmd);
1364
1365 return errno_to_blk_status(err);
1366}
1367
cecbc9ce
BVA		 1368static void nullb_zero_read_cmd_buffer(struct nullb_cmd *cmd)
1369{
1370 struct nullb_device *dev = cmd->nq->dev;
1371 struct bio *bio;
1372
1373 if (dev->memory_backed)
1374 return;
1375
1376 if (dev->queue_mode == NULL_Q_BIO && bio_op(cmd->bio) == REQ_OP_READ) {
1377 zero_fill_bio(cmd->bio);
1378 } else if (req_op(cmd->rq) == REQ_OP_READ) {
1379 __rq_for_each_bio(bio, cmd->rq)
1380 zero_fill_bio(bio);
1381 }
1382}
1383
a3d7d674
CK		 1384static inline void nullb_complete_cmd(struct nullb_cmd *cmd)
1385{
cecbc9ce
BVA		 1386 /*
1387 * Since root privileges are required to configure the null_blk
1388 * driver, it is fine that this driver does not initialize the
1389 * data buffers of read commands. Zero-initialize these buffers
1390 * anyway if KMSAN is enabled to prevent that KMSAN complains
1391 * about null_blk not initializing read data buffers.
1392 */
1393 if (IS_ENABLED(CONFIG_KMSAN))
1394 nullb_zero_read_cmd_buffer(cmd);
1395
a3d7d674
CK		 1396 /* Complete IO by inline, softirq or timer */
1397 switch (cmd->nq->dev->irqmode) {
1398 case NULL_IRQ_SOFTIRQ:
1399 switch (cmd->nq->dev->queue_mode) {
1400 case NULL_Q_MQ:
15f73f5b
CH		 1401 if (likely(!blk_should_fake_timeout(cmd->rq->q)))
1402 blk_mq_complete_request(cmd->rq);
a3d7d674
CK		 1403 break;
1404 case NULL_Q_BIO:
1405 /*
1406 * XXX: no proper submitting cpu information available.
1407 */
1408 end_cmd(cmd);
1409 break;
1410 }
1411 break;
1412 case NULL_IRQ_NONE:
1413 end_cmd(cmd);
1414 break;
1415 case NULL_IRQ_TIMER:
1416 null_cmd_end_timer(cmd);
1417 break;
1418 }
1419}
1420
ff07a02e
BVA		 1421blk_status_t null_process_cmd(struct nullb_cmd *cmd, enum req_op op,
1422 sector_t sector, unsigned int nr_sectors)
9dd44c7e
DLM		 1423{
1424 struct nullb_device *dev = cmd->nq->dev;
1425 blk_status_t ret;
1426
1427 if (dev->badblocks.shift != -1) {
1428 ret = null_handle_badblocks(cmd, sector, nr_sectors);
1429 if (ret != BLK_STS_OK)
1430 return ret;
1431 }
1432
1433 if (dev->memory_backed)
49c7089f 1434 return null_handle_memory_backed(cmd, op, sector, nr_sectors);
9dd44c7e
DLM		 1435
1436 return BLK_STS_OK;
1437}
1438
d4b186ed 1439static blk_status_t null_handle_cmd(struct nullb_cmd *cmd, sector_t sector,
ff07a02e 1440 sector_t nr_sectors, enum req_op op)
5bcd0e0c
SL		 1441{
1442 struct nullb_device *dev = cmd->nq->dev;
eff2c4f1 1443 struct nullb *nullb = dev->nullb;
adb84284 1444 blk_status_t sts;
5bcd0e0c 1445
eff2c4f1 1446 if (test_bit(NULLB_DEV_FL_THROTTLED, &dev->flags)) {
adb84284
CK		 1447 sts = null_handle_throttled(cmd);
1448 if (sts != BLK_STS_OK)
1449 return sts;
eff2c4f1
SL		 1450 }
1451
d4b186ed
CK		 1452 if (op == REQ_OP_FLUSH) {
1453 cmd->error = errno_to_blk_status(null_handle_flush(nullb));
1454 goto out;
1455 }
d4b186ed 1456
9dd44c7e 1457 if (dev->zoned)
de3510e5 1458 sts = null_process_zoned_cmd(cmd, op, sector, nr_sectors);
9dd44c7e 1459 else
de3510e5
DLM		 1460 sts = null_process_cmd(cmd, op, sector, nr_sectors);
1461
1462 /* Do not overwrite errors (e.g. timeout errors) */
1463 if (cmd->error == BLK_STS_OK)
1464 cmd->error = sts;
fceb5d1b 1465
2f54a613 1466out:
a3d7d674 1467 nullb_complete_cmd(cmd);
5bcd0e0c 1468 return BLK_STS_OK;
f2298c04
JA		 1469}
1470
eff2c4f1
SL		 1471static enum hrtimer_restart nullb_bwtimer_fn(struct hrtimer *timer)
1472{
1473 struct nullb *nullb = container_of(timer, struct nullb, bw_timer);
1474 ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1475 unsigned int mbps = nullb->dev->mbps;
1476
1477 if (atomic_long_read(&nullb->cur_bytes) == mb_per_tick(mbps))
1478 return HRTIMER_NORESTART;
1479
1480 atomic_long_set(&nullb->cur_bytes, mb_per_tick(mbps));
1481 null_restart_queue_async(nullb);
1482
1483 hrtimer_forward_now(&nullb->bw_timer, timer_interval);
1484
1485 return HRTIMER_RESTART;
1486}
1487
1488static void nullb_setup_bwtimer(struct nullb *nullb)
1489{
1490 ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1491
1492 hrtimer_init(&nullb->bw_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1493 nullb->bw_timer.function = nullb_bwtimer_fn;
1494 atomic_long_set(&nullb->cur_bytes, mb_per_tick(nullb->dev->mbps));
1495 hrtimer_start(&nullb->bw_timer, timer_interval, HRTIMER_MODE_REL);
f2298c04
JA		 1496}
1497
1498static struct nullb_queue *nullb_to_queue(struct nullb *nullb)
1499{
1500 int index = 0;
1501
1502 if (nullb->nr_queues != 1)
1503 index = raw_smp_processor_id() / ((nr_cpu_ids + nullb->nr_queues - 1) / nullb->nr_queues);
1504
1505 return &nullb->queues[index];
1506}
1507
3e08773c 1508static void null_submit_bio(struct bio *bio)
f2298c04 1509{
d4b186ed
CK		 1510 sector_t sector = bio->bi_iter.bi_sector;
1511 sector_t nr_sectors = bio_sectors(bio);
309dca30 1512 struct nullb *nullb = bio->bi_bdev->bd_disk->private_data;
f2298c04 1513 struct nullb_queue *nq = nullb_to_queue(nullb);
f2298c04 1514
3d3472f3 1515 null_handle_cmd(alloc_cmd(nq, bio), sector, nr_sectors, bio_op(bio));
f2298c04
JA		 1516}
1517
93b57046
JA		 1518static bool should_timeout_request(struct request *rq)
1519{
33f782c4 1520#ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
93b57046
JA		 1521 if (g_timeout_str[0])
1522 return should_fail(&null_timeout_attr, 1);
33f782c4 1523#endif
24941b90
JA		 1524 return false;
1525}
93b57046 1526
24941b90
JA		 1527static bool should_requeue_request(struct request *rq)
1528{
1529#ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1530 if (g_requeue_str[0])
1531 return should_fail(&null_requeue_attr, 1);
1532#endif
93b57046
JA		 1533 return false;
1534}
1535
a4e1d0b7 1536static void null_map_queues(struct blk_mq_tag_set *set)
0a593fbb
JA		 1537{
1538 struct nullb *nullb = set->driver_data;
1539 int i, qoff;
15dfc662
SK		 1540 unsigned int submit_queues = g_submit_queues;
1541 unsigned int poll_queues = g_poll_queues;
1542
1543 if (nullb) {
1544 struct nullb_device *dev = nullb->dev;
1545
1546 /*
1547 * Refer nr_hw_queues of the tag set to check if the expected
1548 * number of hardware queues are prepared. If block layer failed
1549 * to prepare them, use previous numbers of submit queues and
1550 * poll queues to map queues.
1551 */
1552 if (set->nr_hw_queues ==
1553 dev->submit_queues + dev->poll_queues) {
1554 submit_queues = dev->submit_queues;
1555 poll_queues = dev->poll_queues;
1556 } else if (set->nr_hw_queues ==
1557 dev->prev_submit_queues + dev->prev_poll_queues) {
1558 submit_queues = dev->prev_submit_queues;
1559 poll_queues = dev->prev_poll_queues;
1560 } else {
1561 pr_warn("tag set has unexpected nr_hw_queues: %d\n",
1562 set->nr_hw_queues);
10b41ea1
BVA		 1563 WARN_ON_ONCE(true);
1564 submit_queues = 1;
1565 poll_queues = 0;
15dfc662
SK		 1566 }
1567 }
0a593fbb
JA		 1568
1569 for (i = 0, qoff = 0; i < set->nr_maps; i++) {
1570 struct blk_mq_queue_map *map = &set->map[i];
1571
1572 switch (i) {
1573 case HCTX_TYPE_DEFAULT:
15dfc662 1574 map->nr_queues = submit_queues;
0a593fbb
JA		 1575 break;
1576 case HCTX_TYPE_READ:
1577 map->nr_queues = 0;
1578 continue;
1579 case HCTX_TYPE_POLL:
15dfc662 1580 map->nr_queues = poll_queues;
0a593fbb
JA		 1581 break;
1582 }
1583 map->queue_offset = qoff;
1584 qoff += map->nr_queues;
1585 blk_mq_map_queues(map);
1586 }
0a593fbb
JA		 1587}
1588
1589static int null_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1590{
1591 struct nullb_queue *nq = hctx->driver_data;
1592 LIST_HEAD(list);
1593 int nr = 0;
1594
1595 spin_lock(&nq->poll_lock);
1596 list_splice_init(&nq->poll_list, &list);
1597 spin_unlock(&nq->poll_lock);
1598
1599 while (!list_empty(&list)) {
1600 struct nullb_cmd *cmd;
1601 struct request *req;
1602
1603 req = list_first_entry(&list, struct request, queuelist);
1604 list_del_init(&req->queuelist);
1605 cmd = blk_mq_rq_to_pdu(req);
1606 cmd->error = null_process_cmd(cmd, req_op(req), blk_rq_pos(req),
1607 blk_rq_sectors(req));
c5eafd79 1608 if (!blk_mq_add_to_batch(req, iob, (__force int) cmd->error,
2385ebf3
ML		 1609 blk_mq_end_request_batch))
1610 end_cmd(cmd);
0a593fbb
JA		 1611 nr++;
1612 }
1613
1614 return nr;
1615}
1616
9bdb4833 1617static enum blk_eh_timer_return null_timeout_rq(struct request *rq)
5448aca4 1618{
0a593fbb 1619 struct blk_mq_hw_ctx *hctx = rq->mq_hctx;
de3510e5
DLM		 1620 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq);
1621
9c7eddf1 1622 pr_info("rq %p timed out\n", rq);
de3510e5 1623
0a593fbb
JA		 1624 if (hctx->type == HCTX_TYPE_POLL) {
1625 struct nullb_queue *nq = hctx->driver_data;
1626
1627 spin_lock(&nq->poll_lock);
1628 list_del_init(&rq->queuelist);
1629 spin_unlock(&nq->poll_lock);
1630 }
1631
de3510e5
DLM		 1632 /*
1633 * If the device is marked as blocking (i.e. memory backed or zoned
1634 * device), the submission path may be blocked waiting for resources
1635 * and cause real timeouts. For these real timeouts, the submission
1636 * path will complete the request using blk_mq_complete_request().
1637 * Only fake timeouts need to execute blk_mq_complete_request() here.
1638 */
1639 cmd->error = BLK_STS_TIMEOUT;
3e3876d3 1640 if (cmd->fake_timeout || hctx->type == HCTX_TYPE_POLL)
de3510e5 1641 blk_mq_complete_request(rq);
0df0bb08 1642 return BLK_EH_DONE;
5448aca4
JA		 1643}
1644
fc17b653 1645static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx,
74c45052 1646 const struct blk_mq_queue_data *bd)
f2298c04 1647{
74c45052 1648 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
2984c868 1649 struct nullb_queue *nq = hctx->driver_data;
d4b186ed
CK		 1650 sector_t nr_sectors = blk_rq_sectors(bd->rq);
1651 sector_t sector = blk_rq_pos(bd->rq);
0a593fbb 1652 const bool is_poll = hctx->type == HCTX_TYPE_POLL;
f2298c04 1653
db5bcf87
JA		 1654 might_sleep_if(hctx->flags & BLK_MQ_F_BLOCKING);
1655
0a593fbb 1656 if (!is_poll && nq->dev->irqmode == NULL_IRQ_TIMER) {
3c395a96
PV		 1657 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1658 cmd->timer.function = null_cmd_timer_expired;
1659 }
74c45052 1660 cmd->rq = bd->rq;
ff770422 1661 cmd->error = BLK_STS_OK;
2984c868 1662 cmd->nq = nq;
de3510e5 1663 cmd->fake_timeout = should_timeout_request(bd->rq);
f2298c04 1664
74c45052 1665 blk_mq_start_request(bd->rq);
e2490073 1666
24941b90
JA		 1667 if (should_requeue_request(bd->rq)) {
1668 /*
1669 * Alternate between hitting the core BUSY path, and the
1670 * driver driven requeue path
1671 */
1672 nq->requeue_selection++;
1673 if (nq->requeue_selection & 1)
1674 return BLK_STS_RESOURCE;
1675 else {
1676 blk_mq_requeue_request(bd->rq, true);
1677 return BLK_STS_OK;
1678 }
1679 }
0a593fbb
JA		 1680
1681 if (is_poll) {
1682 spin_lock(&nq->poll_lock);
1683 list_add_tail(&bd->rq->queuelist, &nq->poll_list);
1684 spin_unlock(&nq->poll_lock);
1685 return BLK_STS_OK;
1686 }
de3510e5 1687 if (cmd->fake_timeout)
24941b90 1688 return BLK_STS_OK;
93b57046 1689
d4b186ed 1690 return null_handle_cmd(cmd, sector, nr_sectors, req_op(bd->rq));
f2298c04
JA		 1691}
1692
de65d2d2
MB		 1693static void cleanup_queue(struct nullb_queue *nq)
1694{
eb25ad80 1695 bitmap_free(nq->tag_map);
de65d2d2
MB		 1696 kfree(nq->cmds);
1697}
1698
1699static void cleanup_queues(struct nullb *nullb)
1700{
1701 int i;
1702
1703 for (i = 0; i < nullb->nr_queues; i++)
1704 cleanup_queue(&nullb->queues[i]);
1705
1706 kfree(nullb->queues);
1707}
1708
78b10be2
BVA		 1709static void null_exit_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx)
1710{
1711 struct nullb_queue *nq = hctx->driver_data;
1712 struct nullb *nullb = nq->dev->nullb;
1713
1714 nullb->nr_queues--;
1715}
1716
1717static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq)
1718{
1719 init_waitqueue_head(&nq->wait);
1720 nq->queue_depth = nullb->queue_depth;
1721 nq->dev = nullb->dev;
0a593fbb
JA		 1722 INIT_LIST_HEAD(&nq->poll_list);
1723 spin_lock_init(&nq->poll_lock);
78b10be2
BVA		 1724}
1725
1726static int null_init_hctx(struct blk_mq_hw_ctx *hctx, void *driver_data,
1727 unsigned int hctx_idx)
1728{
1729 struct nullb *nullb = hctx->queue->queuedata;
1730 struct nullb_queue *nq;
1731
596444e7
BVA		 1732#ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1733 if (g_init_hctx_str[0] && should_fail(&null_init_hctx_attr, 1))
1734 return -EFAULT;
1735#endif
1736
78b10be2
BVA		 1737 nq = &nullb->queues[hctx_idx];
1738 hctx->driver_data = nq;
1739 null_init_queue(nullb, nq);
1740 nullb->nr_queues++;
1741
1742 return 0;
1743}
1744
1745static const struct blk_mq_ops null_mq_ops = {
1746 .queue_rq = null_queue_rq,
1747 .complete = null_complete_rq,
1748 .timeout = null_timeout_rq,
0a593fbb
JA		 1749 .poll = null_poll,
1750 .map_queues = null_map_queues,
78b10be2
BVA		 1751 .init_hctx = null_init_hctx,
1752 .exit_hctx = null_exit_hctx,
1753};
1754
9ae2d0aa
MB		 1755static void null_del_dev(struct nullb *nullb)
1756{
9b03b713
BVA		 1757 struct nullb_device *dev;
1758
1759 if (!nullb)
1760 return;
1761
1762 dev = nullb->dev;
2984c868 1763
94bc02e3
SL		 1764 ida_simple_remove(&nullb_indexes, nullb->index);
1765
9ae2d0aa
MB		 1766 list_del_init(&nullb->list);
1767
74ede5af 1768 del_gendisk(nullb->disk);
eff2c4f1
SL		 1769
1770 if (test_bit(NULLB_DEV_FL_THROTTLED, &nullb->dev->flags)) {
1771 hrtimer_cancel(&nullb->bw_timer);
1772 atomic_long_set(&nullb->cur_bytes, LONG_MAX);
1773 null_restart_queue_async(nullb);
1774 }
1775
8b9ab626 1776 put_disk(nullb->disk);
2984c868
SL		 1777 if (dev->queue_mode == NULL_Q_MQ &&
1778 nullb->tag_set == &nullb->__tag_set)
82f402fe 1779 blk_mq_free_tag_set(nullb->tag_set);
9ae2d0aa 1780 cleanup_queues(nullb);
deb78b41
SL		 1781 if (null_cache_active(nullb))
1782 null_free_device_storage(nullb->dev, true);
9ae2d0aa 1783 kfree(nullb);
2984c868 1784 dev->nullb = NULL;
9ae2d0aa
MB		 1785}
1786
306eb6b4
SL		 1787static void null_config_discard(struct nullb *nullb)
1788{
1789 if (nullb->dev->discard == false)
1790 return;
1592cd15 1791
49c7089f
DLM		 1792 if (!nullb->dev->memory_backed) {
1793 nullb->dev->discard = false;
1794 pr_info("discard option is ignored without memory backing\n");
1795 return;
1796 }
1797
1592cd15
CK		 1798 if (nullb->dev->zoned) {
1799 nullb->dev->discard = false;
1800 pr_info("discard option is ignored in zoned mode\n");
1801 return;
1802 }
1803
306eb6b4 1804 nullb->q->limits.discard_granularity = nullb->dev->blocksize;
306eb6b4 1805 blk_queue_max_discard_sectors(nullb->q, UINT_MAX >> 9);
9ae2d0aa
MB		 1806}
1807
c62b37d9
CH		 1808static const struct block_device_operations null_bio_ops = {
1809 .owner = THIS_MODULE,
1810 .submit_bio = null_submit_bio,
1811 .report_zones = null_report_zones,
1812};
1813
1814static const struct block_device_operations null_rq_ops = {
e3f89564 1815 .owner = THIS_MODULE,
7fc8fb51 1816 .report_zones = null_report_zones,
f2298c04
JA		 1817};
1818
1819static int setup_commands(struct nullb_queue *nq)
1820{
1821 struct nullb_cmd *cmd;
eb25ad80 1822 int i;
f2298c04 1823
6396bb22 1824 nq->cmds = kcalloc(nq->queue_depth, sizeof(*cmd), GFP_KERNEL);
f2298c04 1825 if (!nq->cmds)
2d263a78 1826 return -ENOMEM;
f2298c04 1827
eb25ad80 1828 nq->tag_map = bitmap_zalloc(nq->queue_depth, GFP_KERNEL);
f2298c04
JA		 1829 if (!nq->tag_map) {
1830 kfree(nq->cmds);
2d263a78 1831 return -ENOMEM;
f2298c04
JA		 1832 }
1833
1834 for (i = 0; i < nq->queue_depth; i++) {
1835 cmd = &nq->cmds[i];
f2298c04
JA		 1836 cmd->tag = -1U;
1837 }
1838
1839 return 0;
1840}
1841
f2298c04
JA		 1842static int setup_queues(struct nullb *nullb)
1843{
0a593fbb
JA		 1844 int nqueues = nr_cpu_ids;
1845
1846 if (g_poll_queues)
1847 nqueues += g_poll_queues;
1848
1849 nullb->queues = kcalloc(nqueues, sizeof(struct nullb_queue),
6396bb22 1850 GFP_KERNEL);
f2298c04 1851 if (!nullb->queues)
2d263a78 1852 return -ENOMEM;
f2298c04 1853
2984c868 1854 nullb->queue_depth = nullb->dev->hw_queue_depth;
2d263a78
MB		 1855 return 0;
1856}
1857
1858static int init_driver_queues(struct nullb *nullb)
1859{
1860 struct nullb_queue *nq;
1861 int i, ret = 0;
f2298c04 1862
2984c868 1863 for (i = 0; i < nullb->dev->submit_queues; i++) {
f2298c04 1864 nq = &nullb->queues[i];
2d263a78
MB		 1865
1866 null_init_queue(nullb, nq);
1867
1868 ret = setup_commands(nq);
1869 if (ret)
31f9690e 1870 return ret;
f2298c04
JA		 1871 nullb->nr_queues++;
1872 }
2d263a78 1873 return 0;
f2298c04
JA		 1874}
1875
9ae2d0aa 1876static int null_gendisk_register(struct nullb *nullb)
f2298c04 1877{
979d5447 1878 sector_t size = ((sector_t)nullb->dev->size * SZ_1M) >> SECTOR_SHIFT;
132226b3 1879 struct gendisk *disk = nullb->disk;
9ae2d0aa 1880
979d5447 1881 set_capacity(disk, size);
9ae2d0aa 1882
9ae2d0aa
MB		 1883 disk->major = null_major;
1884 disk->first_minor = nullb->index;
132226b3 1885 disk->minors = 1;
c62b37d9
CH		 1886 if (queue_is_mq(nullb->q))
1887 disk->fops = &null_rq_ops;
1888 else
1889 disk->fops = &null_bio_ops;
9ae2d0aa 1890 disk->private_data = nullb;
9ae2d0aa
MB		 1891 strncpy(disk->disk_name, nullb->disk_name, DISK_NAME_LEN);
1892
bf505456 1893 if (nullb->dev->zoned) {
d205bde7
DLM		 1894 int ret = null_register_zoned_dev(nullb);
1895
1896 if (ret)
1897 return ret;
bf505456
DLM		 1898 }
1899
10e7123d 1900 return add_disk(disk);
9ae2d0aa
MB		 1901}
1902
2984c868 1903static int null_init_tag_set(struct nullb *nullb, struct blk_mq_tag_set *set)
82f402fe 1904{
7012eef5
VF		 1905 unsigned int flags = BLK_MQ_F_SHOULD_MERGE;
1906 int hw_queues, numa_node;
1907 unsigned int queue_depth;
0a593fbb
JA		 1908 int poll_queues;
1909
7012eef5
VF		 1910 if (nullb) {
1911 hw_queues = nullb->dev->submit_queues;
1912 poll_queues = nullb->dev->poll_queues;
1913 queue_depth = nullb->dev->hw_queue_depth;
1914 numa_node = nullb->dev->home_node;
1915 if (nullb->dev->no_sched)
1916 flags |= BLK_MQ_F_NO_SCHED;
1917 if (nullb->dev->shared_tag_bitmap)
1918 flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1919 if (nullb->dev->blocking)
1920 flags |= BLK_MQ_F_BLOCKING;
1921 } else {
1922 hw_queues = g_submit_queues;
1923 poll_queues = g_poll_queues;
1924 queue_depth = g_hw_queue_depth;
1925 numa_node = g_home_node;
1926 if (g_no_sched)
1927 flags |= BLK_MQ_F_NO_SCHED;
1928 if (g_shared_tag_bitmap)
1929 flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1930 if (g_blocking)
1931 flags |= BLK_MQ_F_BLOCKING;
1932 }
1933
82f402fe 1934 set->ops = &null_mq_ops;
82f402fe 1935 set->cmd_size = sizeof(struct nullb_cmd);
7012eef5 1936 set->flags = flags;
0a593fbb 1937 set->driver_data = nullb;
7012eef5
VF		 1938 set->nr_hw_queues = hw_queues;
1939 set->queue_depth = queue_depth;
1940 set->numa_node = numa_node;
1941 if (poll_queues) {
1942 set->nr_hw_queues += poll_queues;
0a593fbb 1943 set->nr_maps = 3;
7012eef5 1944 } else {
0a593fbb 1945 set->nr_maps = 1;
7012eef5 1946 }
82f402fe
JA		 1947
1948 return blk_mq_alloc_tag_set(set);
1949}
1950
5c4bd1f4 1951static int null_validate_conf(struct nullb_device *dev)
cedcafad
SL		 1952{
1953 dev->blocksize = round_down(dev->blocksize, 512);
1954 dev->blocksize = clamp_t(unsigned int, dev->blocksize, 512, 4096);
cedcafad
SL		 1955
1956 if (dev->queue_mode == NULL_Q_MQ && dev->use_per_node_hctx) {
1957 if (dev->submit_queues != nr_online_nodes)
1958 dev->submit_queues = nr_online_nodes;
1959 } else if (dev->submit_queues > nr_cpu_ids)
1960 dev->submit_queues = nr_cpu_ids;
1961 else if (dev->submit_queues == 0)
1962 dev->submit_queues = 1;
15dfc662
SK		 1963 dev->prev_submit_queues = dev->submit_queues;
1964
1965 if (dev->poll_queues > g_poll_queues)
1966 dev->poll_queues = g_poll_queues;
15dfc662 1967 dev->prev_poll_queues = dev->poll_queues;
cedcafad
SL		 1968
1969 dev->queue_mode = min_t(unsigned int, dev->queue_mode, NULL_Q_MQ);
1970 dev->irqmode = min_t(unsigned int, dev->irqmode, NULL_IRQ_TIMER);
5bcd0e0c
SL		 1971
1972 /* Do memory allocation, so set blocking */
1973 if (dev->memory_backed)
1974 dev->blocking = true;
deb78b41
SL		 1975 else /* cache is meaningless */
1976 dev->cache_size = 0;
1977 dev->cache_size = min_t(unsigned long, ULONG_MAX / 1024 / 1024,
1978 dev->cache_size);
eff2c4f1
SL		 1979 dev->mbps = min_t(unsigned int, 1024 * 40, dev->mbps);
1980 /* can not stop a queue */
1981 if (dev->queue_mode == NULL_Q_BIO)
1982 dev->mbps = 0;
5c4bd1f4
DLM		 1983
1984 if (dev->zoned &&
1985 (!dev->zone_size || !is_power_of_2(dev->zone_size))) {
1986 pr_err("zone_size must be power-of-two\n");
1987 return -EINVAL;
1988 }
1989
1990 return 0;
cedcafad
SL		 1991}
1992
33f782c4 1993#ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
24941b90
JA		 1994static bool __null_setup_fault(struct fault_attr *attr, char *str)
1995{
1996 if (!str[0])
93b57046
JA		 1997 return true;
1998
24941b90 1999 if (!setup_fault_attr(attr, str))
93b57046
JA		 2000 return false;
2001
24941b90
JA		 2002 attr->verbose = 0;
2003 return true;
2004}
2005#endif
2006
2007static bool null_setup_fault(void)
2008{
2009#ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
2010 if (!__null_setup_fault(&null_timeout_attr, g_timeout_str))
2011 return false;
2012 if (!__null_setup_fault(&null_requeue_attr, g_requeue_str))
2013 return false;
596444e7
BVA		 2014 if (!__null_setup_fault(&null_init_hctx_attr, g_init_hctx_str))
2015 return false;
33f782c4 2016#endif
93b57046
JA		 2017 return true;
2018}
2019
2984c868 2020static int null_add_dev(struct nullb_device *dev)
9ae2d0aa
MB		 2021{
2022 struct nullb *nullb;
dc501dc0 2023 int rv;
f2298c04 2024
5c4bd1f4
DLM		 2025 rv = null_validate_conf(dev);
2026 if (rv)
2027 return rv;
cedcafad 2028
2984c868 2029 nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, dev->home_node);
dc501dc0
RE		 2030 if (!nullb) {
2031 rv = -ENOMEM;
24d2f903 2032 goto out;
dc501dc0 2033 }
2984c868
SL		 2034 nullb->dev = dev;
2035 dev->nullb = nullb;
f2298c04
JA		 2036
2037 spin_lock_init(&nullb->lock);
2038
dc501dc0
RE		 2039 rv = setup_queues(nullb);
2040 if (rv)
24d2f903 2041 goto out_free_nullb;
f2298c04 2042
2984c868 2043 if (dev->queue_mode == NULL_Q_MQ) {
82f402fe
JA		 2044 if (shared_tags) {
2045 nullb->tag_set = &tag_set;
2046 rv = 0;
2047 } else {
2048 nullb->tag_set = &nullb->__tag_set;
2984c868 2049 rv = null_init_tag_set(nullb, nullb->tag_set);
82f402fe
JA		 2050 }
2051
dc501dc0 2052 if (rv)
24d2f903
CH		 2053 goto out_cleanup_queues;
2054
93b57046 2055 if (!null_setup_fault())
132226b3 2056 goto out_cleanup_tags;
93b57046 2057
5448aca4 2058 nullb->tag_set->timeout = 5 * HZ;
6759b1a2
CH		 2059 nullb->disk = blk_mq_alloc_disk(nullb->tag_set, nullb);
2060 if (IS_ERR(nullb->disk)) {
2061 rv = PTR_ERR(nullb->disk);
24d2f903 2062 goto out_cleanup_tags;
6759b1a2
CH		 2063 }
2064 nullb->q = nullb->disk->queue;
2984c868 2065 } else if (dev->queue_mode == NULL_Q_BIO) {
132226b3
CH		 2066 rv = -ENOMEM;
2067 nullb->disk = blk_alloc_disk(nullb->dev->home_node);
2068 if (!nullb->disk)
24d2f903 2069 goto out_cleanup_queues;
132226b3
CH		 2070
2071 nullb->q = nullb->disk->queue;
31f9690e
JK		 2072 rv = init_driver_queues(nullb);
2073 if (rv)
132226b3 2074 goto out_cleanup_disk;
f2298c04
JA		 2075 }
2076
eff2c4f1
SL		 2077 if (dev->mbps) {
2078 set_bit(NULLB_DEV_FL_THROTTLED, &dev->flags);
2079 nullb_setup_bwtimer(nullb);
2080 }
2081
deb78b41
SL		 2082 if (dev->cache_size > 0) {
2083 set_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
2084 blk_queue_write_cache(nullb->q, true, true);
deb78b41
SL		 2085 }
2086
ca4b2a01 2087 if (dev->zoned) {
d205bde7 2088 rv = null_init_zoned_dev(dev, nullb->q);
ca4b2a01 2089 if (rv)
132226b3 2090 goto out_cleanup_disk;
ca4b2a01
MB		 2091 }
2092
f2298c04 2093 nullb->q->queuedata = nullb;
8b904b5b
BVA		 2094 blk_queue_flag_set(QUEUE_FLAG_NONROT, nullb->q);
2095 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, nullb->q);
f2298c04 2096
f2298c04 2097 mutex_lock(&lock);
ee452a8d
DC		 2098 rv = ida_simple_get(&nullb_indexes, 0, 0, GFP_KERNEL);
2099 if (rv < 0) {
2100 mutex_unlock(&lock);
2101 goto out_cleanup_zone;
2102 }
2103 nullb->index = rv;
2104 dev->index = rv;
f2298c04
JA		 2105 mutex_unlock(&lock);
2106
2984c868
SL		 2107 blk_queue_logical_block_size(nullb->q, dev->blocksize);
2108 blk_queue_physical_block_size(nullb->q, dev->blocksize);
ea17fd35
DLM		 2109 if (!dev->max_sectors)
2110 dev->max_sectors = queue_max_hw_sectors(nullb->q);
0a26f327 2111 dev->max_sectors = min(dev->max_sectors, BLK_DEF_MAX_SECTORS);
ea17fd35 2112 blk_queue_max_hw_sectors(nullb->q, dev->max_sectors);
f2298c04 2113
cee1b215
MG		 2114 if (dev->virt_boundary)
2115 blk_queue_virt_boundary(nullb->q, PAGE_SIZE - 1);
2116
306eb6b4 2117 null_config_discard(nullb);
f2298c04 2118
49c3b926
DLM		 2119 if (config_item_name(&dev->item)) {
2120 /* Use configfs dir name as the device name */
2121 snprintf(nullb->disk_name, sizeof(nullb->disk_name),
2122 "%s", config_item_name(&dev->item));
2123 } else {
2124 sprintf(nullb->disk_name, "nullb%d", nullb->index);
2125 }
b2b7e001 2126
74ede5af 2127 rv = null_gendisk_register(nullb);
9ae2d0aa 2128 if (rv)
ee452a8d 2129 goto out_ida_free;
a514379b
MB		 2130
2131 mutex_lock(&lock);
2132 list_add_tail(&nullb->list, &nullb_list);
2133 mutex_unlock(&lock);
3681c85d 2134
db060f54
DLM		 2135 pr_info("disk %s created\n", nullb->disk_name);
2136
f2298c04 2137 return 0;
ee452a8d
DC		 2138
2139out_ida_free:
2140 ida_free(&nullb_indexes, nullb->index);
ca4b2a01 2141out_cleanup_zone:
d205bde7 2142 null_free_zoned_dev(dev);
132226b3 2143out_cleanup_disk:
8b9ab626 2144 put_disk(nullb->disk);
24d2f903 2145out_cleanup_tags:
2984c868 2146 if (dev->queue_mode == NULL_Q_MQ && nullb->tag_set == &nullb->__tag_set)
82f402fe 2147 blk_mq_free_tag_set(nullb->tag_set);
24d2f903
CH		 2148out_cleanup_queues:
2149 cleanup_queues(nullb);
2150out_free_nullb:
2151 kfree(nullb);
2004bfde 2152 dev->nullb = NULL;
24d2f903 2153out:
dc501dc0 2154 return rv;
f2298c04
JA		 2155}
2156
49c3b926
DLM		 2157static struct nullb *null_find_dev_by_name(const char *name)
2158{
2159 struct nullb *nullb = NULL, *nb;
2160
2161 mutex_lock(&lock);
2162 list_for_each_entry(nb, &nullb_list, list) {
2163 if (strcmp(nb->disk_name, name) == 0) {
2164 nullb = nb;
2165 break;
2166 }
2167 }
2168 mutex_unlock(&lock);
2169
2170 return nullb;
2171}
2172
b3a0a73e
DLM		 2173static int null_create_dev(void)
2174{
2175 struct nullb_device *dev;
2176 int ret;
2177
2178 dev = null_alloc_dev();
2179 if (!dev)
2180 return -ENOMEM;
2181
2182 ret = null_add_dev(dev);
2183 if (ret) {
2184 null_free_dev(dev);
2185 return ret;
2186 }
2187
2188 return 0;
2189}
2190
2191static void null_destroy_dev(struct nullb *nullb)
2192{
2193 struct nullb_device *dev = nullb->dev;
2194
2195 null_del_dev(nullb);
2196 null_free_dev(dev);
2197}
2198
f2298c04
JA		 2199static int __init null_init(void)
2200{
af096e22 2201 int ret = 0;
f2298c04 2202 unsigned int i;
af096e22 2203 struct nullb *nullb;
f2298c04 2204
2984c868 2205 if (g_bs > PAGE_SIZE) {
9c7eddf1
AA		 2206 pr_warn("invalid block size\n");
2207 pr_warn("defaults block size to %lu\n", PAGE_SIZE);
2984c868 2208 g_bs = PAGE_SIZE;
9967d8ac 2209 }
f2298c04 2210
ea17fd35
DLM		 2211 if (g_max_sectors > BLK_DEF_MAX_SECTORS) {
2212 pr_warn("invalid max sectors\n");
2213 pr_warn("defaults max sectors to %u\n", BLK_DEF_MAX_SECTORS);
2214 g_max_sectors = BLK_DEF_MAX_SECTORS;
2215 }
2216
7ff684a6 2217 if (g_home_node != NUMA_NO_NODE && g_home_node >= nr_online_nodes) {
9c7eddf1 2218 pr_err("invalid home_node value\n");
7ff684a6
JP		 2219 g_home_node = NUMA_NO_NODE;
2220 }
2221
e50b1e32 2222 if (g_queue_mode == NULL_Q_RQ) {
525323d2 2223 pr_err("legacy IO path is no longer available\n");
e50b1e32
JA		 2224 return -EINVAL;
2225 }
525323d2 2226
2984c868
SL		 2227 if (g_queue_mode == NULL_Q_MQ && g_use_per_node_hctx) {
2228 if (g_submit_queues != nr_online_nodes) {
9c7eddf1 2229 pr_warn("submit_queues param is set to %u.\n",
525323d2 2230 nr_online_nodes);
2984c868 2231 g_submit_queues = nr_online_nodes;
fc1bc354 2232 }
525323d2 2233 } else if (g_submit_queues > nr_cpu_ids) {
2984c868 2234 g_submit_queues = nr_cpu_ids;
525323d2 2235 } else if (g_submit_queues <= 0) {
2984c868 2236 g_submit_queues = 1;
525323d2 2237 }
f2298c04 2238
2984c868
SL		 2239 if (g_queue_mode == NULL_Q_MQ && shared_tags) {
2240 ret = null_init_tag_set(NULL, &tag_set);
db2d153d
MG		 2241 if (ret)
2242 return ret;
2243 }
2244
3bf2bd20
SL		 2245 config_group_init(&nullb_subsys.su_group);
2246 mutex_init(&nullb_subsys.su_mutex);
2247
2248 ret = configfs_register_subsystem(&nullb_subsys);
2249 if (ret)
2250 goto err_tagset;
2251
f2298c04
JA		 2252 mutex_init(&lock);
2253
f2298c04 2254 null_major = register_blkdev(0, "nullb");
db2d153d
MG		 2255 if (null_major < 0) {
2256 ret = null_major;
3bf2bd20 2257 goto err_conf;
db2d153d 2258 }
f2298c04
JA		 2259
2260 for (i = 0; i < nr_devices; i++) {
b3a0a73e
DLM		 2261 ret = null_create_dev();
2262 if (ret)
af096e22 2263 goto err_dev;
f2298c04
JA		 2264 }
2265
9c7eddf1 2266 pr_info("module loaded\n");
f2298c04 2267 return 0;
af096e22
MH		 2268
2269err_dev:
2270 while (!list_empty(&nullb_list)) {
2271 nullb = list_entry(nullb_list.next, struct nullb, list);
b3a0a73e 2272 null_destroy_dev(nullb);
af096e22 2273 }
af096e22 2274 unregister_blkdev(null_major, "nullb");
3bf2bd20
SL		 2275err_conf:
2276 configfs_unregister_subsystem(&nullb_subsys);
db2d153d 2277err_tagset:
2984c868 2278 if (g_queue_mode == NULL_Q_MQ && shared_tags)
db2d153d 2279 blk_mq_free_tag_set(&tag_set);
af096e22 2280 return ret;
f2298c04
JA		 2281}
2282
2283static void __exit null_exit(void)
2284{
2285 struct nullb *nullb;
2286
3bf2bd20
SL		 2287 configfs_unregister_subsystem(&nullb_subsys);
2288
f2298c04
JA		 2289 unregister_blkdev(null_major, "nullb");
2290
2291 mutex_lock(&lock);
2292 while (!list_empty(&nullb_list)) {
2293 nullb = list_entry(nullb_list.next, struct nullb, list);
b3a0a73e 2294 null_destroy_dev(nullb);
f2298c04
JA		 2295 }
2296 mutex_unlock(&lock);
6bb9535b 2297
2984c868 2298 if (g_queue_mode == NULL_Q_MQ && shared_tags)
82f402fe 2299 blk_mq_free_tag_set(&tag_set);
f2298c04
JA		 2300}
2301
2302module_init(null_init);
2303module_exit(null_exit);
2304
231b3db1 2305MODULE_AUTHOR("Jens Axboe <axboe@kernel.dk>");
f2298c04 2306MODULE_LICENSE("GPL");
Linux 6.x block layer and io_uring tree(s)