xref: /openbmc/linux/drivers/block/null_blk/main.c (revision 7a2f6f61)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Add configfs and memory store: Kyungchan Koh <kkc6196@fb.com> and
4  * Shaohua Li <shli@fb.com>
5  */
6 #include <linux/module.h>
7 
8 #include <linux/moduleparam.h>
9 #include <linux/sched.h>
10 #include <linux/fs.h>
11 #include <linux/init.h>
12 #include "null_blk.h"
13 
14 #define FREE_BATCH		16
15 
16 #define TICKS_PER_SEC		50ULL
17 #define TIMER_INTERVAL		(NSEC_PER_SEC / TICKS_PER_SEC)
18 
19 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
20 static DECLARE_FAULT_ATTR(null_timeout_attr);
21 static DECLARE_FAULT_ATTR(null_requeue_attr);
22 static DECLARE_FAULT_ATTR(null_init_hctx_attr);
23 #endif
24 
25 static inline u64 mb_per_tick(int mbps)
26 {
27 	return (1 << 20) / TICKS_PER_SEC * ((u64) mbps);
28 }
29 
30 /*
31  * Status flags for nullb_device.
32  *
33  * CONFIGURED:	Device has been configured and turned on. Cannot reconfigure.
34  * UP:		Device is currently on and visible in userspace.
35  * THROTTLED:	Device is being throttled.
36  * CACHE:	Device is using a write-back cache.
37  */
38 enum nullb_device_flags {
39 	NULLB_DEV_FL_CONFIGURED	= 0,
40 	NULLB_DEV_FL_UP		= 1,
41 	NULLB_DEV_FL_THROTTLED	= 2,
42 	NULLB_DEV_FL_CACHE	= 3,
43 };
44 
45 #define MAP_SZ		((PAGE_SIZE >> SECTOR_SHIFT) + 2)
46 /*
47  * nullb_page is a page in memory for nullb devices.
48  *
49  * @page:	The page holding the data.
50  * @bitmap:	The bitmap represents which sector in the page has data.
51  *		Each bit represents one block size. For example, sector 8
52  *		will use the 7th bit
53  * The highest 2 bits of bitmap are for special purpose. LOCK means the cache
54  * page is being flushing to storage. FREE means the cache page is freed and
55  * should be skipped from flushing to storage. Please see
56  * null_make_cache_space
57  */
58 struct nullb_page {
59 	struct page *page;
60 	DECLARE_BITMAP(bitmap, MAP_SZ);
61 };
62 #define NULLB_PAGE_LOCK (MAP_SZ - 1)
63 #define NULLB_PAGE_FREE (MAP_SZ - 2)
64 
65 static LIST_HEAD(nullb_list);
66 static struct mutex lock;
67 static int null_major;
68 static DEFINE_IDA(nullb_indexes);
69 static struct blk_mq_tag_set tag_set;
70 
71 enum {
72 	NULL_IRQ_NONE		= 0,
73 	NULL_IRQ_SOFTIRQ	= 1,
74 	NULL_IRQ_TIMER		= 2,
75 };
76 
77 enum {
78 	NULL_Q_BIO		= 0,
79 	NULL_Q_RQ		= 1,
80 	NULL_Q_MQ		= 2,
81 };
82 
83 static bool g_virt_boundary = false;
84 module_param_named(virt_boundary, g_virt_boundary, bool, 0444);
85 MODULE_PARM_DESC(virt_boundary, "Require a virtual boundary for the device. Default: False");
86 
87 static int g_no_sched;
88 module_param_named(no_sched, g_no_sched, int, 0444);
89 MODULE_PARM_DESC(no_sched, "No io scheduler");
90 
91 static int g_submit_queues = 1;
92 module_param_named(submit_queues, g_submit_queues, int, 0444);
93 MODULE_PARM_DESC(submit_queues, "Number of submission queues");
94 
95 static int g_poll_queues = 1;
96 module_param_named(poll_queues, g_poll_queues, int, 0444);
97 MODULE_PARM_DESC(poll_queues, "Number of IOPOLL submission queues");
98 
99 static int g_home_node = NUMA_NO_NODE;
100 module_param_named(home_node, g_home_node, int, 0444);
101 MODULE_PARM_DESC(home_node, "Home node for the device");
102 
103 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
104 /*
105  * For more details about fault injection, please refer to
106  * Documentation/fault-injection/fault-injection.rst.
107  */
108 static char g_timeout_str[80];
109 module_param_string(timeout, g_timeout_str, sizeof(g_timeout_str), 0444);
110 MODULE_PARM_DESC(timeout, "Fault injection. timeout=<interval>,<probability>,<space>,<times>");
111 
112 static char g_requeue_str[80];
113 module_param_string(requeue, g_requeue_str, sizeof(g_requeue_str), 0444);
114 MODULE_PARM_DESC(requeue, "Fault injection. requeue=<interval>,<probability>,<space>,<times>");
115 
116 static char g_init_hctx_str[80];
117 module_param_string(init_hctx, g_init_hctx_str, sizeof(g_init_hctx_str), 0444);
118 MODULE_PARM_DESC(init_hctx, "Fault injection to fail hctx init. init_hctx=<interval>,<probability>,<space>,<times>");
119 #endif
120 
121 static int g_queue_mode = NULL_Q_MQ;
122 
123 static int null_param_store_val(const char *str, int *val, int min, int max)
124 {
125 	int ret, new_val;
126 
127 	ret = kstrtoint(str, 10, &new_val);
128 	if (ret)
129 		return -EINVAL;
130 
131 	if (new_val < min || new_val > max)
132 		return -EINVAL;
133 
134 	*val = new_val;
135 	return 0;
136 }
137 
138 static int null_set_queue_mode(const char *str, const struct kernel_param *kp)
139 {
140 	return null_param_store_val(str, &g_queue_mode, NULL_Q_BIO, NULL_Q_MQ);
141 }
142 
143 static const struct kernel_param_ops null_queue_mode_param_ops = {
144 	.set	= null_set_queue_mode,
145 	.get	= param_get_int,
146 };
147 
148 device_param_cb(queue_mode, &null_queue_mode_param_ops, &g_queue_mode, 0444);
149 MODULE_PARM_DESC(queue_mode, "Block interface to use (0=bio,1=rq,2=multiqueue)");
150 
151 static int g_gb = 250;
152 module_param_named(gb, g_gb, int, 0444);
153 MODULE_PARM_DESC(gb, "Size in GB");
154 
155 static int g_bs = 512;
156 module_param_named(bs, g_bs, int, 0444);
157 MODULE_PARM_DESC(bs, "Block size (in bytes)");
158 
159 static int g_max_sectors;
160 module_param_named(max_sectors, g_max_sectors, int, 0444);
161 MODULE_PARM_DESC(max_sectors, "Maximum size of a command (in 512B sectors)");
162 
163 static unsigned int nr_devices = 1;
164 module_param(nr_devices, uint, 0444);
165 MODULE_PARM_DESC(nr_devices, "Number of devices to register");
166 
167 static bool g_blocking;
168 module_param_named(blocking, g_blocking, bool, 0444);
169 MODULE_PARM_DESC(blocking, "Register as a blocking blk-mq driver device");
170 
171 static bool shared_tags;
172 module_param(shared_tags, bool, 0444);
173 MODULE_PARM_DESC(shared_tags, "Share tag set between devices for blk-mq");
174 
175 static bool g_shared_tag_bitmap;
176 module_param_named(shared_tag_bitmap, g_shared_tag_bitmap, bool, 0444);
177 MODULE_PARM_DESC(shared_tag_bitmap, "Use shared tag bitmap for all submission queues for blk-mq");
178 
179 static int g_irqmode = NULL_IRQ_SOFTIRQ;
180 
181 static int null_set_irqmode(const char *str, const struct kernel_param *kp)
182 {
183 	return null_param_store_val(str, &g_irqmode, NULL_IRQ_NONE,
184 					NULL_IRQ_TIMER);
185 }
186 
187 static const struct kernel_param_ops null_irqmode_param_ops = {
188 	.set	= null_set_irqmode,
189 	.get	= param_get_int,
190 };
191 
192 device_param_cb(irqmode, &null_irqmode_param_ops, &g_irqmode, 0444);
193 MODULE_PARM_DESC(irqmode, "IRQ completion handler. 0-none, 1-softirq, 2-timer");
194 
195 static unsigned long g_completion_nsec = 10000;
196 module_param_named(completion_nsec, g_completion_nsec, ulong, 0444);
197 MODULE_PARM_DESC(completion_nsec, "Time in ns to complete a request in hardware. Default: 10,000ns");
198 
199 static int g_hw_queue_depth = 64;
200 module_param_named(hw_queue_depth, g_hw_queue_depth, int, 0444);
201 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 64");
202 
203 static bool g_use_per_node_hctx;
204 module_param_named(use_per_node_hctx, g_use_per_node_hctx, bool, 0444);
205 MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware context queues. Default: false");
206 
207 static bool g_zoned;
208 module_param_named(zoned, g_zoned, bool, S_IRUGO);
209 MODULE_PARM_DESC(zoned, "Make device as a host-managed zoned block device. Default: false");
210 
211 static unsigned long g_zone_size = 256;
212 module_param_named(zone_size, g_zone_size, ulong, S_IRUGO);
213 MODULE_PARM_DESC(zone_size, "Zone size in MB when block device is zoned. Must be power-of-two: Default: 256");
214 
215 static unsigned long g_zone_capacity;
216 module_param_named(zone_capacity, g_zone_capacity, ulong, 0444);
217 MODULE_PARM_DESC(zone_capacity, "Zone capacity in MB when block device is zoned. Can be less than or equal to zone size. Default: Zone size");
218 
219 static unsigned int g_zone_nr_conv;
220 module_param_named(zone_nr_conv, g_zone_nr_conv, uint, 0444);
221 MODULE_PARM_DESC(zone_nr_conv, "Number of conventional zones when block device is zoned. Default: 0");
222 
223 static unsigned int g_zone_max_open;
224 module_param_named(zone_max_open, g_zone_max_open, uint, 0444);
225 MODULE_PARM_DESC(zone_max_open, "Maximum number of open zones when block device is zoned. Default: 0 (no limit)");
226 
227 static unsigned int g_zone_max_active;
228 module_param_named(zone_max_active, g_zone_max_active, uint, 0444);
229 MODULE_PARM_DESC(zone_max_active, "Maximum number of active zones when block device is zoned. Default: 0 (no limit)");
230 
231 static struct nullb_device *null_alloc_dev(void);
232 static void null_free_dev(struct nullb_device *dev);
233 static void null_del_dev(struct nullb *nullb);
234 static int null_add_dev(struct nullb_device *dev);
235 static void null_free_device_storage(struct nullb_device *dev, bool is_cache);
236 
237 static inline struct nullb_device *to_nullb_device(struct config_item *item)
238 {
239 	return item ? container_of(item, struct nullb_device, item) : NULL;
240 }
241 
242 static inline ssize_t nullb_device_uint_attr_show(unsigned int val, char *page)
243 {
244 	return snprintf(page, PAGE_SIZE, "%u\n", val);
245 }
246 
247 static inline ssize_t nullb_device_ulong_attr_show(unsigned long val,
248 	char *page)
249 {
250 	return snprintf(page, PAGE_SIZE, "%lu\n", val);
251 }
252 
253 static inline ssize_t nullb_device_bool_attr_show(bool val, char *page)
254 {
255 	return snprintf(page, PAGE_SIZE, "%u\n", val);
256 }
257 
258 static ssize_t nullb_device_uint_attr_store(unsigned int *val,
259 	const char *page, size_t count)
260 {
261 	unsigned int tmp;
262 	int result;
263 
264 	result = kstrtouint(page, 0, &tmp);
265 	if (result < 0)
266 		return result;
267 
268 	*val = tmp;
269 	return count;
270 }
271 
272 static ssize_t nullb_device_ulong_attr_store(unsigned long *val,
273 	const char *page, size_t count)
274 {
275 	int result;
276 	unsigned long tmp;
277 
278 	result = kstrtoul(page, 0, &tmp);
279 	if (result < 0)
280 		return result;
281 
282 	*val = tmp;
283 	return count;
284 }
285 
286 static ssize_t nullb_device_bool_attr_store(bool *val, const char *page,
287 	size_t count)
288 {
289 	bool tmp;
290 	int result;
291 
292 	result = kstrtobool(page,  &tmp);
293 	if (result < 0)
294 		return result;
295 
296 	*val = tmp;
297 	return count;
298 }
299 
300 /* The following macro should only be used with TYPE = {uint, ulong, bool}. */
301 #define NULLB_DEVICE_ATTR(NAME, TYPE, APPLY)				\
302 static ssize_t								\
303 nullb_device_##NAME##_show(struct config_item *item, char *page)	\
304 {									\
305 	return nullb_device_##TYPE##_attr_show(				\
306 				to_nullb_device(item)->NAME, page);	\
307 }									\
308 static ssize_t								\
309 nullb_device_##NAME##_store(struct config_item *item, const char *page,	\
310 			    size_t count)				\
311 {									\
312 	int (*apply_fn)(struct nullb_device *dev, TYPE new_value) = APPLY;\
313 	struct nullb_device *dev = to_nullb_device(item);		\
314 	TYPE new_value = 0;						\
315 	int ret;							\
316 									\
317 	ret = nullb_device_##TYPE##_attr_store(&new_value, page, count);\
318 	if (ret < 0)							\
319 		return ret;						\
320 	if (apply_fn)							\
321 		ret = apply_fn(dev, new_value);				\
322 	else if (test_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags)) 	\
323 		ret = -EBUSY;						\
324 	if (ret < 0)							\
325 		return ret;						\
326 	dev->NAME = new_value;						\
327 	return count;							\
328 }									\
329 CONFIGFS_ATTR(nullb_device_, NAME);
330 
331 static int nullb_update_nr_hw_queues(struct nullb_device *dev,
332 				     unsigned int submit_queues,
333 				     unsigned int poll_queues)
334 
335 {
336 	struct blk_mq_tag_set *set;
337 	int ret, nr_hw_queues;
338 
339 	if (!dev->nullb)
340 		return 0;
341 
342 	/*
343 	 * Make sure at least one submit queue exists.
344 	 */
345 	if (!submit_queues)
346 		return -EINVAL;
347 
348 	/*
349 	 * Make sure that null_init_hctx() does not access nullb->queues[] past
350 	 * the end of that array.
351 	 */
352 	if (submit_queues > nr_cpu_ids || poll_queues > g_poll_queues)
353 		return -EINVAL;
354 
355 	/*
356 	 * Keep previous and new queue numbers in nullb_device for reference in
357 	 * the call back function null_map_queues().
358 	 */
359 	dev->prev_submit_queues = dev->submit_queues;
360 	dev->prev_poll_queues = dev->poll_queues;
361 	dev->submit_queues = submit_queues;
362 	dev->poll_queues = poll_queues;
363 
364 	set = dev->nullb->tag_set;
365 	nr_hw_queues = submit_queues + poll_queues;
366 	blk_mq_update_nr_hw_queues(set, nr_hw_queues);
367 	ret = set->nr_hw_queues == nr_hw_queues ? 0 : -ENOMEM;
368 
369 	if (ret) {
370 		/* on error, revert the queue numbers */
371 		dev->submit_queues = dev->prev_submit_queues;
372 		dev->poll_queues = dev->prev_poll_queues;
373 	}
374 
375 	return ret;
376 }
377 
378 static int nullb_apply_submit_queues(struct nullb_device *dev,
379 				     unsigned int submit_queues)
380 {
381 	return nullb_update_nr_hw_queues(dev, submit_queues, dev->poll_queues);
382 }
383 
384 static int nullb_apply_poll_queues(struct nullb_device *dev,
385 				   unsigned int poll_queues)
386 {
387 	return nullb_update_nr_hw_queues(dev, dev->submit_queues, poll_queues);
388 }
389 
390 NULLB_DEVICE_ATTR(size, ulong, NULL);
391 NULLB_DEVICE_ATTR(completion_nsec, ulong, NULL);
392 NULLB_DEVICE_ATTR(submit_queues, uint, nullb_apply_submit_queues);
393 NULLB_DEVICE_ATTR(poll_queues, uint, nullb_apply_poll_queues);
394 NULLB_DEVICE_ATTR(home_node, uint, NULL);
395 NULLB_DEVICE_ATTR(queue_mode, uint, NULL);
396 NULLB_DEVICE_ATTR(blocksize, uint, NULL);
397 NULLB_DEVICE_ATTR(max_sectors, uint, NULL);
398 NULLB_DEVICE_ATTR(irqmode, uint, NULL);
399 NULLB_DEVICE_ATTR(hw_queue_depth, uint, NULL);
400 NULLB_DEVICE_ATTR(index, uint, NULL);
401 NULLB_DEVICE_ATTR(blocking, bool, NULL);
402 NULLB_DEVICE_ATTR(use_per_node_hctx, bool, NULL);
403 NULLB_DEVICE_ATTR(memory_backed, bool, NULL);
404 NULLB_DEVICE_ATTR(discard, bool, NULL);
405 NULLB_DEVICE_ATTR(mbps, uint, NULL);
406 NULLB_DEVICE_ATTR(cache_size, ulong, NULL);
407 NULLB_DEVICE_ATTR(zoned, bool, NULL);
408 NULLB_DEVICE_ATTR(zone_size, ulong, NULL);
409 NULLB_DEVICE_ATTR(zone_capacity, ulong, NULL);
410 NULLB_DEVICE_ATTR(zone_nr_conv, uint, NULL);
411 NULLB_DEVICE_ATTR(zone_max_open, uint, NULL);
412 NULLB_DEVICE_ATTR(zone_max_active, uint, NULL);
413 NULLB_DEVICE_ATTR(virt_boundary, bool, NULL);
414 
415 static ssize_t nullb_device_power_show(struct config_item *item, char *page)
416 {
417 	return nullb_device_bool_attr_show(to_nullb_device(item)->power, page);
418 }
419 
420 static ssize_t nullb_device_power_store(struct config_item *item,
421 				     const char *page, size_t count)
422 {
423 	struct nullb_device *dev = to_nullb_device(item);
424 	bool newp = false;
425 	ssize_t ret;
426 
427 	ret = nullb_device_bool_attr_store(&newp, page, count);
428 	if (ret < 0)
429 		return ret;
430 
431 	if (!dev->power && newp) {
432 		if (test_and_set_bit(NULLB_DEV_FL_UP, &dev->flags))
433 			return count;
434 		ret = null_add_dev(dev);
435 		if (ret) {
436 			clear_bit(NULLB_DEV_FL_UP, &dev->flags);
437 			return ret;
438 		}
439 
440 		set_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
441 		dev->power = newp;
442 	} else if (dev->power && !newp) {
443 		if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
444 			mutex_lock(&lock);
445 			dev->power = newp;
446 			null_del_dev(dev->nullb);
447 			mutex_unlock(&lock);
448 		}
449 		clear_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
450 	}
451 
452 	return count;
453 }
454 
455 CONFIGFS_ATTR(nullb_device_, power);
456 
457 static ssize_t nullb_device_badblocks_show(struct config_item *item, char *page)
458 {
459 	struct nullb_device *t_dev = to_nullb_device(item);
460 
461 	return badblocks_show(&t_dev->badblocks, page, 0);
462 }
463 
464 static ssize_t nullb_device_badblocks_store(struct config_item *item,
465 				     const char *page, size_t count)
466 {
467 	struct nullb_device *t_dev = to_nullb_device(item);
468 	char *orig, *buf, *tmp;
469 	u64 start, end;
470 	int ret;
471 
472 	orig = kstrndup(page, count, GFP_KERNEL);
473 	if (!orig)
474 		return -ENOMEM;
475 
476 	buf = strstrip(orig);
477 
478 	ret = -EINVAL;
479 	if (buf[0] != '+' && buf[0] != '-')
480 		goto out;
481 	tmp = strchr(&buf[1], '-');
482 	if (!tmp)
483 		goto out;
484 	*tmp = '\0';
485 	ret = kstrtoull(buf + 1, 0, &start);
486 	if (ret)
487 		goto out;
488 	ret = kstrtoull(tmp + 1, 0, &end);
489 	if (ret)
490 		goto out;
491 	ret = -EINVAL;
492 	if (start > end)
493 		goto out;
494 	/* enable badblocks */
495 	cmpxchg(&t_dev->badblocks.shift, -1, 0);
496 	if (buf[0] == '+')
497 		ret = badblocks_set(&t_dev->badblocks, start,
498 			end - start + 1, 1);
499 	else
500 		ret = badblocks_clear(&t_dev->badblocks, start,
501 			end - start + 1);
502 	if (ret == 0)
503 		ret = count;
504 out:
505 	kfree(orig);
506 	return ret;
507 }
508 CONFIGFS_ATTR(nullb_device_, badblocks);
509 
510 static struct configfs_attribute *nullb_device_attrs[] = {
511 	&nullb_device_attr_size,
512 	&nullb_device_attr_completion_nsec,
513 	&nullb_device_attr_submit_queues,
514 	&nullb_device_attr_poll_queues,
515 	&nullb_device_attr_home_node,
516 	&nullb_device_attr_queue_mode,
517 	&nullb_device_attr_blocksize,
518 	&nullb_device_attr_max_sectors,
519 	&nullb_device_attr_irqmode,
520 	&nullb_device_attr_hw_queue_depth,
521 	&nullb_device_attr_index,
522 	&nullb_device_attr_blocking,
523 	&nullb_device_attr_use_per_node_hctx,
524 	&nullb_device_attr_power,
525 	&nullb_device_attr_memory_backed,
526 	&nullb_device_attr_discard,
527 	&nullb_device_attr_mbps,
528 	&nullb_device_attr_cache_size,
529 	&nullb_device_attr_badblocks,
530 	&nullb_device_attr_zoned,
531 	&nullb_device_attr_zone_size,
532 	&nullb_device_attr_zone_capacity,
533 	&nullb_device_attr_zone_nr_conv,
534 	&nullb_device_attr_zone_max_open,
535 	&nullb_device_attr_zone_max_active,
536 	&nullb_device_attr_virt_boundary,
537 	NULL,
538 };
539 
540 static void nullb_device_release(struct config_item *item)
541 {
542 	struct nullb_device *dev = to_nullb_device(item);
543 
544 	null_free_device_storage(dev, false);
545 	null_free_dev(dev);
546 }
547 
548 static struct configfs_item_operations nullb_device_ops = {
549 	.release	= nullb_device_release,
550 };
551 
552 static const struct config_item_type nullb_device_type = {
553 	.ct_item_ops	= &nullb_device_ops,
554 	.ct_attrs	= nullb_device_attrs,
555 	.ct_owner	= THIS_MODULE,
556 };
557 
558 static struct
559 config_item *nullb_group_make_item(struct config_group *group, const char *name)
560 {
561 	struct nullb_device *dev;
562 
563 	dev = null_alloc_dev();
564 	if (!dev)
565 		return ERR_PTR(-ENOMEM);
566 
567 	config_item_init_type_name(&dev->item, name, &nullb_device_type);
568 
569 	return &dev->item;
570 }
571 
572 static void
573 nullb_group_drop_item(struct config_group *group, struct config_item *item)
574 {
575 	struct nullb_device *dev = to_nullb_device(item);
576 
577 	if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
578 		mutex_lock(&lock);
579 		dev->power = false;
580 		null_del_dev(dev->nullb);
581 		mutex_unlock(&lock);
582 	}
583 
584 	config_item_put(item);
585 }
586 
587 static ssize_t memb_group_features_show(struct config_item *item, char *page)
588 {
589 	return snprintf(page, PAGE_SIZE,
590 			"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");
591 }
592 
593 CONFIGFS_ATTR_RO(memb_group_, features);
594 
595 static struct configfs_attribute *nullb_group_attrs[] = {
596 	&memb_group_attr_features,
597 	NULL,
598 };
599 
600 static struct configfs_group_operations nullb_group_ops = {
601 	.make_item	= nullb_group_make_item,
602 	.drop_item	= nullb_group_drop_item,
603 };
604 
605 static const struct config_item_type nullb_group_type = {
606 	.ct_group_ops	= &nullb_group_ops,
607 	.ct_attrs	= nullb_group_attrs,
608 	.ct_owner	= THIS_MODULE,
609 };
610 
611 static struct configfs_subsystem nullb_subsys = {
612 	.su_group = {
613 		.cg_item = {
614 			.ci_namebuf = "nullb",
615 			.ci_type = &nullb_group_type,
616 		},
617 	},
618 };
619 
620 static inline int null_cache_active(struct nullb *nullb)
621 {
622 	return test_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
623 }
624 
625 static struct nullb_device *null_alloc_dev(void)
626 {
627 	struct nullb_device *dev;
628 
629 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
630 	if (!dev)
631 		return NULL;
632 	INIT_RADIX_TREE(&dev->data, GFP_ATOMIC);
633 	INIT_RADIX_TREE(&dev->cache, GFP_ATOMIC);
634 	if (badblocks_init(&dev->badblocks, 0)) {
635 		kfree(dev);
636 		return NULL;
637 	}
638 
639 	dev->size = g_gb * 1024;
640 	dev->completion_nsec = g_completion_nsec;
641 	dev->submit_queues = g_submit_queues;
642 	dev->prev_submit_queues = g_submit_queues;
643 	dev->poll_queues = g_poll_queues;
644 	dev->prev_poll_queues = g_poll_queues;
645 	dev->home_node = g_home_node;
646 	dev->queue_mode = g_queue_mode;
647 	dev->blocksize = g_bs;
648 	dev->max_sectors = g_max_sectors;
649 	dev->irqmode = g_irqmode;
650 	dev->hw_queue_depth = g_hw_queue_depth;
651 	dev->blocking = g_blocking;
652 	dev->use_per_node_hctx = g_use_per_node_hctx;
653 	dev->zoned = g_zoned;
654 	dev->zone_size = g_zone_size;
655 	dev->zone_capacity = g_zone_capacity;
656 	dev->zone_nr_conv = g_zone_nr_conv;
657 	dev->zone_max_open = g_zone_max_open;
658 	dev->zone_max_active = g_zone_max_active;
659 	dev->virt_boundary = g_virt_boundary;
660 	return dev;
661 }
662 
663 static void null_free_dev(struct nullb_device *dev)
664 {
665 	if (!dev)
666 		return;
667 
668 	null_free_zoned_dev(dev);
669 	badblocks_exit(&dev->badblocks);
670 	kfree(dev);
671 }
672 
673 static void put_tag(struct nullb_queue *nq, unsigned int tag)
674 {
675 	clear_bit_unlock(tag, nq->tag_map);
676 
677 	if (waitqueue_active(&nq->wait))
678 		wake_up(&nq->wait);
679 }
680 
681 static unsigned int get_tag(struct nullb_queue *nq)
682 {
683 	unsigned int tag;
684 
685 	do {
686 		tag = find_first_zero_bit(nq->tag_map, nq->queue_depth);
687 		if (tag >= nq->queue_depth)
688 			return -1U;
689 	} while (test_and_set_bit_lock(tag, nq->tag_map));
690 
691 	return tag;
692 }
693 
694 static void free_cmd(struct nullb_cmd *cmd)
695 {
696 	put_tag(cmd->nq, cmd->tag);
697 }
698 
699 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer);
700 
701 static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq)
702 {
703 	struct nullb_cmd *cmd;
704 	unsigned int tag;
705 
706 	tag = get_tag(nq);
707 	if (tag != -1U) {
708 		cmd = &nq->cmds[tag];
709 		cmd->tag = tag;
710 		cmd->error = BLK_STS_OK;
711 		cmd->nq = nq;
712 		if (nq->dev->irqmode == NULL_IRQ_TIMER) {
713 			hrtimer_init(&cmd->timer, CLOCK_MONOTONIC,
714 				     HRTIMER_MODE_REL);
715 			cmd->timer.function = null_cmd_timer_expired;
716 		}
717 		return cmd;
718 	}
719 
720 	return NULL;
721 }
722 
723 static struct nullb_cmd *alloc_cmd(struct nullb_queue *nq, struct bio *bio)
724 {
725 	struct nullb_cmd *cmd;
726 	DEFINE_WAIT(wait);
727 
728 	do {
729 		/*
730 		 * This avoids multiple return statements, multiple calls to
731 		 * __alloc_cmd() and a fast path call to prepare_to_wait().
732 		 */
733 		cmd = __alloc_cmd(nq);
734 		if (cmd) {
735 			cmd->bio = bio;
736 			return cmd;
737 		}
738 		prepare_to_wait(&nq->wait, &wait, TASK_UNINTERRUPTIBLE);
739 		io_schedule();
740 		finish_wait(&nq->wait, &wait);
741 	} while (1);
742 }
743 
744 static void end_cmd(struct nullb_cmd *cmd)
745 {
746 	int queue_mode = cmd->nq->dev->queue_mode;
747 
748 	switch (queue_mode)  {
749 	case NULL_Q_MQ:
750 		blk_mq_end_request(cmd->rq, cmd->error);
751 		return;
752 	case NULL_Q_BIO:
753 		cmd->bio->bi_status = cmd->error;
754 		bio_endio(cmd->bio);
755 		break;
756 	}
757 
758 	free_cmd(cmd);
759 }
760 
761 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
762 {
763 	end_cmd(container_of(timer, struct nullb_cmd, timer));
764 
765 	return HRTIMER_NORESTART;
766 }
767 
768 static void null_cmd_end_timer(struct nullb_cmd *cmd)
769 {
770 	ktime_t kt = cmd->nq->dev->completion_nsec;
771 
772 	hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL);
773 }
774 
775 static void null_complete_rq(struct request *rq)
776 {
777 	end_cmd(blk_mq_rq_to_pdu(rq));
778 }
779 
780 static struct nullb_page *null_alloc_page(void)
781 {
782 	struct nullb_page *t_page;
783 
784 	t_page = kmalloc(sizeof(struct nullb_page), GFP_NOIO);
785 	if (!t_page)
786 		return NULL;
787 
788 	t_page->page = alloc_pages(GFP_NOIO, 0);
789 	if (!t_page->page) {
790 		kfree(t_page);
791 		return NULL;
792 	}
793 
794 	memset(t_page->bitmap, 0, sizeof(t_page->bitmap));
795 	return t_page;
796 }
797 
798 static void null_free_page(struct nullb_page *t_page)
799 {
800 	__set_bit(NULLB_PAGE_FREE, t_page->bitmap);
801 	if (test_bit(NULLB_PAGE_LOCK, t_page->bitmap))
802 		return;
803 	__free_page(t_page->page);
804 	kfree(t_page);
805 }
806 
807 static bool null_page_empty(struct nullb_page *page)
808 {
809 	int size = MAP_SZ - 2;
810 
811 	return find_first_bit(page->bitmap, size) == size;
812 }
813 
814 static void null_free_sector(struct nullb *nullb, sector_t sector,
815 	bool is_cache)
816 {
817 	unsigned int sector_bit;
818 	u64 idx;
819 	struct nullb_page *t_page, *ret;
820 	struct radix_tree_root *root;
821 
822 	root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
823 	idx = sector >> PAGE_SECTORS_SHIFT;
824 	sector_bit = (sector & SECTOR_MASK);
825 
826 	t_page = radix_tree_lookup(root, idx);
827 	if (t_page) {
828 		__clear_bit(sector_bit, t_page->bitmap);
829 
830 		if (null_page_empty(t_page)) {
831 			ret = radix_tree_delete_item(root, idx, t_page);
832 			WARN_ON(ret != t_page);
833 			null_free_page(ret);
834 			if (is_cache)
835 				nullb->dev->curr_cache -= PAGE_SIZE;
836 		}
837 	}
838 }
839 
840 static struct nullb_page *null_radix_tree_insert(struct nullb *nullb, u64 idx,
841 	struct nullb_page *t_page, bool is_cache)
842 {
843 	struct radix_tree_root *root;
844 
845 	root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
846 
847 	if (radix_tree_insert(root, idx, t_page)) {
848 		null_free_page(t_page);
849 		t_page = radix_tree_lookup(root, idx);
850 		WARN_ON(!t_page || t_page->page->index != idx);
851 	} else if (is_cache)
852 		nullb->dev->curr_cache += PAGE_SIZE;
853 
854 	return t_page;
855 }
856 
857 static void null_free_device_storage(struct nullb_device *dev, bool is_cache)
858 {
859 	unsigned long pos = 0;
860 	int nr_pages;
861 	struct nullb_page *ret, *t_pages[FREE_BATCH];
862 	struct radix_tree_root *root;
863 
864 	root = is_cache ? &dev->cache : &dev->data;
865 
866 	do {
867 		int i;
868 
869 		nr_pages = radix_tree_gang_lookup(root,
870 				(void **)t_pages, pos, FREE_BATCH);
871 
872 		for (i = 0; i < nr_pages; i++) {
873 			pos = t_pages[i]->page->index;
874 			ret = radix_tree_delete_item(root, pos, t_pages[i]);
875 			WARN_ON(ret != t_pages[i]);
876 			null_free_page(ret);
877 		}
878 
879 		pos++;
880 	} while (nr_pages == FREE_BATCH);
881 
882 	if (is_cache)
883 		dev->curr_cache = 0;
884 }
885 
886 static struct nullb_page *__null_lookup_page(struct nullb *nullb,
887 	sector_t sector, bool for_write, bool is_cache)
888 {
889 	unsigned int sector_bit;
890 	u64 idx;
891 	struct nullb_page *t_page;
892 	struct radix_tree_root *root;
893 
894 	idx = sector >> PAGE_SECTORS_SHIFT;
895 	sector_bit = (sector & SECTOR_MASK);
896 
897 	root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
898 	t_page = radix_tree_lookup(root, idx);
899 	WARN_ON(t_page && t_page->page->index != idx);
900 
901 	if (t_page && (for_write || test_bit(sector_bit, t_page->bitmap)))
902 		return t_page;
903 
904 	return NULL;
905 }
906 
907 static struct nullb_page *null_lookup_page(struct nullb *nullb,
908 	sector_t sector, bool for_write, bool ignore_cache)
909 {
910 	struct nullb_page *page = NULL;
911 
912 	if (!ignore_cache)
913 		page = __null_lookup_page(nullb, sector, for_write, true);
914 	if (page)
915 		return page;
916 	return __null_lookup_page(nullb, sector, for_write, false);
917 }
918 
919 static struct nullb_page *null_insert_page(struct nullb *nullb,
920 					   sector_t sector, bool ignore_cache)
921 	__releases(&nullb->lock)
922 	__acquires(&nullb->lock)
923 {
924 	u64 idx;
925 	struct nullb_page *t_page;
926 
927 	t_page = null_lookup_page(nullb, sector, true, ignore_cache);
928 	if (t_page)
929 		return t_page;
930 
931 	spin_unlock_irq(&nullb->lock);
932 
933 	t_page = null_alloc_page();
934 	if (!t_page)
935 		goto out_lock;
936 
937 	if (radix_tree_preload(GFP_NOIO))
938 		goto out_freepage;
939 
940 	spin_lock_irq(&nullb->lock);
941 	idx = sector >> PAGE_SECTORS_SHIFT;
942 	t_page->page->index = idx;
943 	t_page = null_radix_tree_insert(nullb, idx, t_page, !ignore_cache);
944 	radix_tree_preload_end();
945 
946 	return t_page;
947 out_freepage:
948 	null_free_page(t_page);
949 out_lock:
950 	spin_lock_irq(&nullb->lock);
951 	return null_lookup_page(nullb, sector, true, ignore_cache);
952 }
953 
954 static int null_flush_cache_page(struct nullb *nullb, struct nullb_page *c_page)
955 {
956 	int i;
957 	unsigned int offset;
958 	u64 idx;
959 	struct nullb_page *t_page, *ret;
960 	void *dst, *src;
961 
962 	idx = c_page->page->index;
963 
964 	t_page = null_insert_page(nullb, idx << PAGE_SECTORS_SHIFT, true);
965 
966 	__clear_bit(NULLB_PAGE_LOCK, c_page->bitmap);
967 	if (test_bit(NULLB_PAGE_FREE, c_page->bitmap)) {
968 		null_free_page(c_page);
969 		if (t_page && null_page_empty(t_page)) {
970 			ret = radix_tree_delete_item(&nullb->dev->data,
971 				idx, t_page);
972 			null_free_page(t_page);
973 		}
974 		return 0;
975 	}
976 
977 	if (!t_page)
978 		return -ENOMEM;
979 
980 	src = kmap_atomic(c_page->page);
981 	dst = kmap_atomic(t_page->page);
982 
983 	for (i = 0; i < PAGE_SECTORS;
984 			i += (nullb->dev->blocksize >> SECTOR_SHIFT)) {
985 		if (test_bit(i, c_page->bitmap)) {
986 			offset = (i << SECTOR_SHIFT);
987 			memcpy(dst + offset, src + offset,
988 				nullb->dev->blocksize);
989 			__set_bit(i, t_page->bitmap);
990 		}
991 	}
992 
993 	kunmap_atomic(dst);
994 	kunmap_atomic(src);
995 
996 	ret = radix_tree_delete_item(&nullb->dev->cache, idx, c_page);
997 	null_free_page(ret);
998 	nullb->dev->curr_cache -= PAGE_SIZE;
999 
1000 	return 0;
1001 }
1002 
1003 static int null_make_cache_space(struct nullb *nullb, unsigned long n)
1004 {
1005 	int i, err, nr_pages;
1006 	struct nullb_page *c_pages[FREE_BATCH];
1007 	unsigned long flushed = 0, one_round;
1008 
1009 again:
1010 	if ((nullb->dev->cache_size * 1024 * 1024) >
1011 	     nullb->dev->curr_cache + n || nullb->dev->curr_cache == 0)
1012 		return 0;
1013 
1014 	nr_pages = radix_tree_gang_lookup(&nullb->dev->cache,
1015 			(void **)c_pages, nullb->cache_flush_pos, FREE_BATCH);
1016 	/*
1017 	 * nullb_flush_cache_page could unlock before using the c_pages. To
1018 	 * avoid race, we don't allow page free
1019 	 */
1020 	for (i = 0; i < nr_pages; i++) {
1021 		nullb->cache_flush_pos = c_pages[i]->page->index;
1022 		/*
1023 		 * We found the page which is being flushed to disk by other
1024 		 * threads
1025 		 */
1026 		if (test_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap))
1027 			c_pages[i] = NULL;
1028 		else
1029 			__set_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap);
1030 	}
1031 
1032 	one_round = 0;
1033 	for (i = 0; i < nr_pages; i++) {
1034 		if (c_pages[i] == NULL)
1035 			continue;
1036 		err = null_flush_cache_page(nullb, c_pages[i]);
1037 		if (err)
1038 			return err;
1039 		one_round++;
1040 	}
1041 	flushed += one_round << PAGE_SHIFT;
1042 
1043 	if (n > flushed) {
1044 		if (nr_pages == 0)
1045 			nullb->cache_flush_pos = 0;
1046 		if (one_round == 0) {
1047 			/* give other threads a chance */
1048 			spin_unlock_irq(&nullb->lock);
1049 			spin_lock_irq(&nullb->lock);
1050 		}
1051 		goto again;
1052 	}
1053 	return 0;
1054 }
1055 
1056 static int copy_to_nullb(struct nullb *nullb, struct page *source,
1057 	unsigned int off, sector_t sector, size_t n, bool is_fua)
1058 {
1059 	size_t temp, count = 0;
1060 	unsigned int offset;
1061 	struct nullb_page *t_page;
1062 	void *dst, *src;
1063 
1064 	while (count < n) {
1065 		temp = min_t(size_t, nullb->dev->blocksize, n - count);
1066 
1067 		if (null_cache_active(nullb) && !is_fua)
1068 			null_make_cache_space(nullb, PAGE_SIZE);
1069 
1070 		offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
1071 		t_page = null_insert_page(nullb, sector,
1072 			!null_cache_active(nullb) || is_fua);
1073 		if (!t_page)
1074 			return -ENOSPC;
1075 
1076 		src = kmap_atomic(source);
1077 		dst = kmap_atomic(t_page->page);
1078 		memcpy(dst + offset, src + off + count, temp);
1079 		kunmap_atomic(dst);
1080 		kunmap_atomic(src);
1081 
1082 		__set_bit(sector & SECTOR_MASK, t_page->bitmap);
1083 
1084 		if (is_fua)
1085 			null_free_sector(nullb, sector, true);
1086 
1087 		count += temp;
1088 		sector += temp >> SECTOR_SHIFT;
1089 	}
1090 	return 0;
1091 }
1092 
1093 static int copy_from_nullb(struct nullb *nullb, struct page *dest,
1094 	unsigned int off, sector_t sector, size_t n)
1095 {
1096 	size_t temp, count = 0;
1097 	unsigned int offset;
1098 	struct nullb_page *t_page;
1099 	void *dst, *src;
1100 
1101 	while (count < n) {
1102 		temp = min_t(size_t, nullb->dev->blocksize, n - count);
1103 
1104 		offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
1105 		t_page = null_lookup_page(nullb, sector, false,
1106 			!null_cache_active(nullb));
1107 
1108 		dst = kmap_atomic(dest);
1109 		if (!t_page) {
1110 			memset(dst + off + count, 0, temp);
1111 			goto next;
1112 		}
1113 		src = kmap_atomic(t_page->page);
1114 		memcpy(dst + off + count, src + offset, temp);
1115 		kunmap_atomic(src);
1116 next:
1117 		kunmap_atomic(dst);
1118 
1119 		count += temp;
1120 		sector += temp >> SECTOR_SHIFT;
1121 	}
1122 	return 0;
1123 }
1124 
1125 static void nullb_fill_pattern(struct nullb *nullb, struct page *page,
1126 			       unsigned int len, unsigned int off)
1127 {
1128 	void *dst;
1129 
1130 	dst = kmap_atomic(page);
1131 	memset(dst + off, 0xFF, len);
1132 	kunmap_atomic(dst);
1133 }
1134 
1135 blk_status_t null_handle_discard(struct nullb_device *dev,
1136 				 sector_t sector, sector_t nr_sectors)
1137 {
1138 	struct nullb *nullb = dev->nullb;
1139 	size_t n = nr_sectors << SECTOR_SHIFT;
1140 	size_t temp;
1141 
1142 	spin_lock_irq(&nullb->lock);
1143 	while (n > 0) {
1144 		temp = min_t(size_t, n, dev->blocksize);
1145 		null_free_sector(nullb, sector, false);
1146 		if (null_cache_active(nullb))
1147 			null_free_sector(nullb, sector, true);
1148 		sector += temp >> SECTOR_SHIFT;
1149 		n -= temp;
1150 	}
1151 	spin_unlock_irq(&nullb->lock);
1152 
1153 	return BLK_STS_OK;
1154 }
1155 
1156 static int null_handle_flush(struct nullb *nullb)
1157 {
1158 	int err;
1159 
1160 	if (!null_cache_active(nullb))
1161 		return 0;
1162 
1163 	spin_lock_irq(&nullb->lock);
1164 	while (true) {
1165 		err = null_make_cache_space(nullb,
1166 			nullb->dev->cache_size * 1024 * 1024);
1167 		if (err || nullb->dev->curr_cache == 0)
1168 			break;
1169 	}
1170 
1171 	WARN_ON(!radix_tree_empty(&nullb->dev->cache));
1172 	spin_unlock_irq(&nullb->lock);
1173 	return err;
1174 }
1175 
1176 static int null_transfer(struct nullb *nullb, struct page *page,
1177 	unsigned int len, unsigned int off, bool is_write, sector_t sector,
1178 	bool is_fua)
1179 {
1180 	struct nullb_device *dev = nullb->dev;
1181 	unsigned int valid_len = len;
1182 	int err = 0;
1183 
1184 	if (!is_write) {
1185 		if (dev->zoned)
1186 			valid_len = null_zone_valid_read_len(nullb,
1187 				sector, len);
1188 
1189 		if (valid_len) {
1190 			err = copy_from_nullb(nullb, page, off,
1191 				sector, valid_len);
1192 			off += valid_len;
1193 			len -= valid_len;
1194 		}
1195 
1196 		if (len)
1197 			nullb_fill_pattern(nullb, page, len, off);
1198 		flush_dcache_page(page);
1199 	} else {
1200 		flush_dcache_page(page);
1201 		err = copy_to_nullb(nullb, page, off, sector, len, is_fua);
1202 	}
1203 
1204 	return err;
1205 }
1206 
1207 static int null_handle_rq(struct nullb_cmd *cmd)
1208 {
1209 	struct request *rq = cmd->rq;
1210 	struct nullb *nullb = cmd->nq->dev->nullb;
1211 	int err;
1212 	unsigned int len;
1213 	sector_t sector = blk_rq_pos(rq);
1214 	struct req_iterator iter;
1215 	struct bio_vec bvec;
1216 
1217 	spin_lock_irq(&nullb->lock);
1218 	rq_for_each_segment(bvec, rq, iter) {
1219 		len = bvec.bv_len;
1220 		err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1221 				     op_is_write(req_op(rq)), sector,
1222 				     rq->cmd_flags & REQ_FUA);
1223 		if (err) {
1224 			spin_unlock_irq(&nullb->lock);
1225 			return err;
1226 		}
1227 		sector += len >> SECTOR_SHIFT;
1228 	}
1229 	spin_unlock_irq(&nullb->lock);
1230 
1231 	return 0;
1232 }
1233 
1234 static int null_handle_bio(struct nullb_cmd *cmd)
1235 {
1236 	struct bio *bio = cmd->bio;
1237 	struct nullb *nullb = cmd->nq->dev->nullb;
1238 	int err;
1239 	unsigned int len;
1240 	sector_t sector = bio->bi_iter.bi_sector;
1241 	struct bio_vec bvec;
1242 	struct bvec_iter iter;
1243 
1244 	spin_lock_irq(&nullb->lock);
1245 	bio_for_each_segment(bvec, bio, iter) {
1246 		len = bvec.bv_len;
1247 		err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1248 				     op_is_write(bio_op(bio)), sector,
1249 				     bio->bi_opf & REQ_FUA);
1250 		if (err) {
1251 			spin_unlock_irq(&nullb->lock);
1252 			return err;
1253 		}
1254 		sector += len >> SECTOR_SHIFT;
1255 	}
1256 	spin_unlock_irq(&nullb->lock);
1257 	return 0;
1258 }
1259 
1260 static void null_stop_queue(struct nullb *nullb)
1261 {
1262 	struct request_queue *q = nullb->q;
1263 
1264 	if (nullb->dev->queue_mode == NULL_Q_MQ)
1265 		blk_mq_stop_hw_queues(q);
1266 }
1267 
1268 static void null_restart_queue_async(struct nullb *nullb)
1269 {
1270 	struct request_queue *q = nullb->q;
1271 
1272 	if (nullb->dev->queue_mode == NULL_Q_MQ)
1273 		blk_mq_start_stopped_hw_queues(q, true);
1274 }
1275 
1276 static inline blk_status_t null_handle_throttled(struct nullb_cmd *cmd)
1277 {
1278 	struct nullb_device *dev = cmd->nq->dev;
1279 	struct nullb *nullb = dev->nullb;
1280 	blk_status_t sts = BLK_STS_OK;
1281 	struct request *rq = cmd->rq;
1282 
1283 	if (!hrtimer_active(&nullb->bw_timer))
1284 		hrtimer_restart(&nullb->bw_timer);
1285 
1286 	if (atomic_long_sub_return(blk_rq_bytes(rq), &nullb->cur_bytes) < 0) {
1287 		null_stop_queue(nullb);
1288 		/* race with timer */
1289 		if (atomic_long_read(&nullb->cur_bytes) > 0)
1290 			null_restart_queue_async(nullb);
1291 		/* requeue request */
1292 		sts = BLK_STS_DEV_RESOURCE;
1293 	}
1294 	return sts;
1295 }
1296 
1297 static inline blk_status_t null_handle_badblocks(struct nullb_cmd *cmd,
1298 						 sector_t sector,
1299 						 sector_t nr_sectors)
1300 {
1301 	struct badblocks *bb = &cmd->nq->dev->badblocks;
1302 	sector_t first_bad;
1303 	int bad_sectors;
1304 
1305 	if (badblocks_check(bb, sector, nr_sectors, &first_bad, &bad_sectors))
1306 		return BLK_STS_IOERR;
1307 
1308 	return BLK_STS_OK;
1309 }
1310 
1311 static inline blk_status_t null_handle_memory_backed(struct nullb_cmd *cmd,
1312 						     enum req_opf op,
1313 						     sector_t sector,
1314 						     sector_t nr_sectors)
1315 {
1316 	struct nullb_device *dev = cmd->nq->dev;
1317 	int err;
1318 
1319 	if (op == REQ_OP_DISCARD)
1320 		return null_handle_discard(dev, sector, nr_sectors);
1321 
1322 	if (dev->queue_mode == NULL_Q_BIO)
1323 		err = null_handle_bio(cmd);
1324 	else
1325 		err = null_handle_rq(cmd);
1326 
1327 	return errno_to_blk_status(err);
1328 }
1329 
1330 static void nullb_zero_read_cmd_buffer(struct nullb_cmd *cmd)
1331 {
1332 	struct nullb_device *dev = cmd->nq->dev;
1333 	struct bio *bio;
1334 
1335 	if (dev->memory_backed)
1336 		return;
1337 
1338 	if (dev->queue_mode == NULL_Q_BIO && bio_op(cmd->bio) == REQ_OP_READ) {
1339 		zero_fill_bio(cmd->bio);
1340 	} else if (req_op(cmd->rq) == REQ_OP_READ) {
1341 		__rq_for_each_bio(bio, cmd->rq)
1342 			zero_fill_bio(bio);
1343 	}
1344 }
1345 
1346 static inline void nullb_complete_cmd(struct nullb_cmd *cmd)
1347 {
1348 	/*
1349 	 * Since root privileges are required to configure the null_blk
1350 	 * driver, it is fine that this driver does not initialize the
1351 	 * data buffers of read commands. Zero-initialize these buffers
1352 	 * anyway if KMSAN is enabled to prevent that KMSAN complains
1353 	 * about null_blk not initializing read data buffers.
1354 	 */
1355 	if (IS_ENABLED(CONFIG_KMSAN))
1356 		nullb_zero_read_cmd_buffer(cmd);
1357 
1358 	/* Complete IO by inline, softirq or timer */
1359 	switch (cmd->nq->dev->irqmode) {
1360 	case NULL_IRQ_SOFTIRQ:
1361 		switch (cmd->nq->dev->queue_mode) {
1362 		case NULL_Q_MQ:
1363 			if (likely(!blk_should_fake_timeout(cmd->rq->q)))
1364 				blk_mq_complete_request(cmd->rq);
1365 			break;
1366 		case NULL_Q_BIO:
1367 			/*
1368 			 * XXX: no proper submitting cpu information available.
1369 			 */
1370 			end_cmd(cmd);
1371 			break;
1372 		}
1373 		break;
1374 	case NULL_IRQ_NONE:
1375 		end_cmd(cmd);
1376 		break;
1377 	case NULL_IRQ_TIMER:
1378 		null_cmd_end_timer(cmd);
1379 		break;
1380 	}
1381 }
1382 
1383 blk_status_t null_process_cmd(struct nullb_cmd *cmd,
1384 			      enum req_opf op, sector_t sector,
1385 			      unsigned int nr_sectors)
1386 {
1387 	struct nullb_device *dev = cmd->nq->dev;
1388 	blk_status_t ret;
1389 
1390 	if (dev->badblocks.shift != -1) {
1391 		ret = null_handle_badblocks(cmd, sector, nr_sectors);
1392 		if (ret != BLK_STS_OK)
1393 			return ret;
1394 	}
1395 
1396 	if (dev->memory_backed)
1397 		return null_handle_memory_backed(cmd, op, sector, nr_sectors);
1398 
1399 	return BLK_STS_OK;
1400 }
1401 
1402 static blk_status_t null_handle_cmd(struct nullb_cmd *cmd, sector_t sector,
1403 				    sector_t nr_sectors, enum req_opf op)
1404 {
1405 	struct nullb_device *dev = cmd->nq->dev;
1406 	struct nullb *nullb = dev->nullb;
1407 	blk_status_t sts;
1408 
1409 	if (test_bit(NULLB_DEV_FL_THROTTLED, &dev->flags)) {
1410 		sts = null_handle_throttled(cmd);
1411 		if (sts != BLK_STS_OK)
1412 			return sts;
1413 	}
1414 
1415 	if (op == REQ_OP_FLUSH) {
1416 		cmd->error = errno_to_blk_status(null_handle_flush(nullb));
1417 		goto out;
1418 	}
1419 
1420 	if (dev->zoned)
1421 		sts = null_process_zoned_cmd(cmd, op, sector, nr_sectors);
1422 	else
1423 		sts = null_process_cmd(cmd, op, sector, nr_sectors);
1424 
1425 	/* Do not overwrite errors (e.g. timeout errors) */
1426 	if (cmd->error == BLK_STS_OK)
1427 		cmd->error = sts;
1428 
1429 out:
1430 	nullb_complete_cmd(cmd);
1431 	return BLK_STS_OK;
1432 }
1433 
1434 static enum hrtimer_restart nullb_bwtimer_fn(struct hrtimer *timer)
1435 {
1436 	struct nullb *nullb = container_of(timer, struct nullb, bw_timer);
1437 	ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1438 	unsigned int mbps = nullb->dev->mbps;
1439 
1440 	if (atomic_long_read(&nullb->cur_bytes) == mb_per_tick(mbps))
1441 		return HRTIMER_NORESTART;
1442 
1443 	atomic_long_set(&nullb->cur_bytes, mb_per_tick(mbps));
1444 	null_restart_queue_async(nullb);
1445 
1446 	hrtimer_forward_now(&nullb->bw_timer, timer_interval);
1447 
1448 	return HRTIMER_RESTART;
1449 }
1450 
1451 static void nullb_setup_bwtimer(struct nullb *nullb)
1452 {
1453 	ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1454 
1455 	hrtimer_init(&nullb->bw_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1456 	nullb->bw_timer.function = nullb_bwtimer_fn;
1457 	atomic_long_set(&nullb->cur_bytes, mb_per_tick(nullb->dev->mbps));
1458 	hrtimer_start(&nullb->bw_timer, timer_interval, HRTIMER_MODE_REL);
1459 }
1460 
1461 static struct nullb_queue *nullb_to_queue(struct nullb *nullb)
1462 {
1463 	int index = 0;
1464 
1465 	if (nullb->nr_queues != 1)
1466 		index = raw_smp_processor_id() / ((nr_cpu_ids + nullb->nr_queues - 1) / nullb->nr_queues);
1467 
1468 	return &nullb->queues[index];
1469 }
1470 
1471 static void null_submit_bio(struct bio *bio)
1472 {
1473 	sector_t sector = bio->bi_iter.bi_sector;
1474 	sector_t nr_sectors = bio_sectors(bio);
1475 	struct nullb *nullb = bio->bi_bdev->bd_disk->private_data;
1476 	struct nullb_queue *nq = nullb_to_queue(nullb);
1477 
1478 	null_handle_cmd(alloc_cmd(nq, bio), sector, nr_sectors, bio_op(bio));
1479 }
1480 
1481 static bool should_timeout_request(struct request *rq)
1482 {
1483 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1484 	if (g_timeout_str[0])
1485 		return should_fail(&null_timeout_attr, 1);
1486 #endif
1487 	return false;
1488 }
1489 
1490 static bool should_requeue_request(struct request *rq)
1491 {
1492 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1493 	if (g_requeue_str[0])
1494 		return should_fail(&null_requeue_attr, 1);
1495 #endif
1496 	return false;
1497 }
1498 
1499 static int null_map_queues(struct blk_mq_tag_set *set)
1500 {
1501 	struct nullb *nullb = set->driver_data;
1502 	int i, qoff;
1503 	unsigned int submit_queues = g_submit_queues;
1504 	unsigned int poll_queues = g_poll_queues;
1505 
1506 	if (nullb) {
1507 		struct nullb_device *dev = nullb->dev;
1508 
1509 		/*
1510 		 * Refer nr_hw_queues of the tag set to check if the expected
1511 		 * number of hardware queues are prepared. If block layer failed
1512 		 * to prepare them, use previous numbers of submit queues and
1513 		 * poll queues to map queues.
1514 		 */
1515 		if (set->nr_hw_queues ==
1516 		    dev->submit_queues + dev->poll_queues) {
1517 			submit_queues = dev->submit_queues;
1518 			poll_queues = dev->poll_queues;
1519 		} else if (set->nr_hw_queues ==
1520 			   dev->prev_submit_queues + dev->prev_poll_queues) {
1521 			submit_queues = dev->prev_submit_queues;
1522 			poll_queues = dev->prev_poll_queues;
1523 		} else {
1524 			pr_warn("tag set has unexpected nr_hw_queues: %d\n",
1525 				set->nr_hw_queues);
1526 			return -EINVAL;
1527 		}
1528 	}
1529 
1530 	for (i = 0, qoff = 0; i < set->nr_maps; i++) {
1531 		struct blk_mq_queue_map *map = &set->map[i];
1532 
1533 		switch (i) {
1534 		case HCTX_TYPE_DEFAULT:
1535 			map->nr_queues = submit_queues;
1536 			break;
1537 		case HCTX_TYPE_READ:
1538 			map->nr_queues = 0;
1539 			continue;
1540 		case HCTX_TYPE_POLL:
1541 			map->nr_queues = poll_queues;
1542 			break;
1543 		}
1544 		map->queue_offset = qoff;
1545 		qoff += map->nr_queues;
1546 		blk_mq_map_queues(map);
1547 	}
1548 
1549 	return 0;
1550 }
1551 
1552 static int null_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1553 {
1554 	struct nullb_queue *nq = hctx->driver_data;
1555 	LIST_HEAD(list);
1556 	int nr = 0;
1557 
1558 	spin_lock(&nq->poll_lock);
1559 	list_splice_init(&nq->poll_list, &list);
1560 	spin_unlock(&nq->poll_lock);
1561 
1562 	while (!list_empty(&list)) {
1563 		struct nullb_cmd *cmd;
1564 		struct request *req;
1565 
1566 		req = list_first_entry(&list, struct request, queuelist);
1567 		list_del_init(&req->queuelist);
1568 		cmd = blk_mq_rq_to_pdu(req);
1569 		cmd->error = null_process_cmd(cmd, req_op(req), blk_rq_pos(req),
1570 						blk_rq_sectors(req));
1571 		if (!blk_mq_add_to_batch(req, iob, (__force int) cmd->error,
1572 					blk_mq_end_request_batch))
1573 			end_cmd(cmd);
1574 		nr++;
1575 	}
1576 
1577 	return nr;
1578 }
1579 
1580 static enum blk_eh_timer_return null_timeout_rq(struct request *rq, bool res)
1581 {
1582 	struct blk_mq_hw_ctx *hctx = rq->mq_hctx;
1583 	struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq);
1584 
1585 	pr_info("rq %p timed out\n", rq);
1586 
1587 	if (hctx->type == HCTX_TYPE_POLL) {
1588 		struct nullb_queue *nq = hctx->driver_data;
1589 
1590 		spin_lock(&nq->poll_lock);
1591 		list_del_init(&rq->queuelist);
1592 		spin_unlock(&nq->poll_lock);
1593 	}
1594 
1595 	/*
1596 	 * If the device is marked as blocking (i.e. memory backed or zoned
1597 	 * device), the submission path may be blocked waiting for resources
1598 	 * and cause real timeouts. For these real timeouts, the submission
1599 	 * path will complete the request using blk_mq_complete_request().
1600 	 * Only fake timeouts need to execute blk_mq_complete_request() here.
1601 	 */
1602 	cmd->error = BLK_STS_TIMEOUT;
1603 	if (cmd->fake_timeout || hctx->type == HCTX_TYPE_POLL)
1604 		blk_mq_complete_request(rq);
1605 	return BLK_EH_DONE;
1606 }
1607 
1608 static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx,
1609 			 const struct blk_mq_queue_data *bd)
1610 {
1611 	struct nullb_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
1612 	struct nullb_queue *nq = hctx->driver_data;
1613 	sector_t nr_sectors = blk_rq_sectors(bd->rq);
1614 	sector_t sector = blk_rq_pos(bd->rq);
1615 	const bool is_poll = hctx->type == HCTX_TYPE_POLL;
1616 
1617 	might_sleep_if(hctx->flags & BLK_MQ_F_BLOCKING);
1618 
1619 	if (!is_poll && nq->dev->irqmode == NULL_IRQ_TIMER) {
1620 		hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1621 		cmd->timer.function = null_cmd_timer_expired;
1622 	}
1623 	cmd->rq = bd->rq;
1624 	cmd->error = BLK_STS_OK;
1625 	cmd->nq = nq;
1626 	cmd->fake_timeout = should_timeout_request(bd->rq);
1627 
1628 	blk_mq_start_request(bd->rq);
1629 
1630 	if (should_requeue_request(bd->rq)) {
1631 		/*
1632 		 * Alternate between hitting the core BUSY path, and the
1633 		 * driver driven requeue path
1634 		 */
1635 		nq->requeue_selection++;
1636 		if (nq->requeue_selection & 1)
1637 			return BLK_STS_RESOURCE;
1638 		else {
1639 			blk_mq_requeue_request(bd->rq, true);
1640 			return BLK_STS_OK;
1641 		}
1642 	}
1643 
1644 	if (is_poll) {
1645 		spin_lock(&nq->poll_lock);
1646 		list_add_tail(&bd->rq->queuelist, &nq->poll_list);
1647 		spin_unlock(&nq->poll_lock);
1648 		return BLK_STS_OK;
1649 	}
1650 	if (cmd->fake_timeout)
1651 		return BLK_STS_OK;
1652 
1653 	return null_handle_cmd(cmd, sector, nr_sectors, req_op(bd->rq));
1654 }
1655 
1656 static void cleanup_queue(struct nullb_queue *nq)
1657 {
1658 	kfree(nq->tag_map);
1659 	kfree(nq->cmds);
1660 }
1661 
1662 static void cleanup_queues(struct nullb *nullb)
1663 {
1664 	int i;
1665 
1666 	for (i = 0; i < nullb->nr_queues; i++)
1667 		cleanup_queue(&nullb->queues[i]);
1668 
1669 	kfree(nullb->queues);
1670 }
1671 
1672 static void null_exit_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx)
1673 {
1674 	struct nullb_queue *nq = hctx->driver_data;
1675 	struct nullb *nullb = nq->dev->nullb;
1676 
1677 	nullb->nr_queues--;
1678 }
1679 
1680 static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq)
1681 {
1682 	init_waitqueue_head(&nq->wait);
1683 	nq->queue_depth = nullb->queue_depth;
1684 	nq->dev = nullb->dev;
1685 	INIT_LIST_HEAD(&nq->poll_list);
1686 	spin_lock_init(&nq->poll_lock);
1687 }
1688 
1689 static int null_init_hctx(struct blk_mq_hw_ctx *hctx, void *driver_data,
1690 			  unsigned int hctx_idx)
1691 {
1692 	struct nullb *nullb = hctx->queue->queuedata;
1693 	struct nullb_queue *nq;
1694 
1695 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1696 	if (g_init_hctx_str[0] && should_fail(&null_init_hctx_attr, 1))
1697 		return -EFAULT;
1698 #endif
1699 
1700 	nq = &nullb->queues[hctx_idx];
1701 	hctx->driver_data = nq;
1702 	null_init_queue(nullb, nq);
1703 	nullb->nr_queues++;
1704 
1705 	return 0;
1706 }
1707 
1708 static const struct blk_mq_ops null_mq_ops = {
1709 	.queue_rq       = null_queue_rq,
1710 	.complete	= null_complete_rq,
1711 	.timeout	= null_timeout_rq,
1712 	.poll		= null_poll,
1713 	.map_queues	= null_map_queues,
1714 	.init_hctx	= null_init_hctx,
1715 	.exit_hctx	= null_exit_hctx,
1716 };
1717 
1718 static void null_del_dev(struct nullb *nullb)
1719 {
1720 	struct nullb_device *dev;
1721 
1722 	if (!nullb)
1723 		return;
1724 
1725 	dev = nullb->dev;
1726 
1727 	ida_simple_remove(&nullb_indexes, nullb->index);
1728 
1729 	list_del_init(&nullb->list);
1730 
1731 	del_gendisk(nullb->disk);
1732 
1733 	if (test_bit(NULLB_DEV_FL_THROTTLED, &nullb->dev->flags)) {
1734 		hrtimer_cancel(&nullb->bw_timer);
1735 		atomic_long_set(&nullb->cur_bytes, LONG_MAX);
1736 		null_restart_queue_async(nullb);
1737 	}
1738 
1739 	blk_cleanup_disk(nullb->disk);
1740 	if (dev->queue_mode == NULL_Q_MQ &&
1741 	    nullb->tag_set == &nullb->__tag_set)
1742 		blk_mq_free_tag_set(nullb->tag_set);
1743 	cleanup_queues(nullb);
1744 	if (null_cache_active(nullb))
1745 		null_free_device_storage(nullb->dev, true);
1746 	kfree(nullb);
1747 	dev->nullb = NULL;
1748 }
1749 
1750 static void null_config_discard(struct nullb *nullb)
1751 {
1752 	if (nullb->dev->discard == false)
1753 		return;
1754 
1755 	if (!nullb->dev->memory_backed) {
1756 		nullb->dev->discard = false;
1757 		pr_info("discard option is ignored without memory backing\n");
1758 		return;
1759 	}
1760 
1761 	if (nullb->dev->zoned) {
1762 		nullb->dev->discard = false;
1763 		pr_info("discard option is ignored in zoned mode\n");
1764 		return;
1765 	}
1766 
1767 	nullb->q->limits.discard_granularity = nullb->dev->blocksize;
1768 	nullb->q->limits.discard_alignment = nullb->dev->blocksize;
1769 	blk_queue_max_discard_sectors(nullb->q, UINT_MAX >> 9);
1770 	blk_queue_flag_set(QUEUE_FLAG_DISCARD, nullb->q);
1771 }
1772 
1773 static const struct block_device_operations null_bio_ops = {
1774 	.owner		= THIS_MODULE,
1775 	.submit_bio	= null_submit_bio,
1776 	.report_zones	= null_report_zones,
1777 };
1778 
1779 static const struct block_device_operations null_rq_ops = {
1780 	.owner		= THIS_MODULE,
1781 	.report_zones	= null_report_zones,
1782 };
1783 
1784 static int setup_commands(struct nullb_queue *nq)
1785 {
1786 	struct nullb_cmd *cmd;
1787 	int i, tag_size;
1788 
1789 	nq->cmds = kcalloc(nq->queue_depth, sizeof(*cmd), GFP_KERNEL);
1790 	if (!nq->cmds)
1791 		return -ENOMEM;
1792 
1793 	tag_size = ALIGN(nq->queue_depth, BITS_PER_LONG) / BITS_PER_LONG;
1794 	nq->tag_map = kcalloc(tag_size, sizeof(unsigned long), GFP_KERNEL);
1795 	if (!nq->tag_map) {
1796 		kfree(nq->cmds);
1797 		return -ENOMEM;
1798 	}
1799 
1800 	for (i = 0; i < nq->queue_depth; i++) {
1801 		cmd = &nq->cmds[i];
1802 		cmd->tag = -1U;
1803 	}
1804 
1805 	return 0;
1806 }
1807 
1808 static int setup_queues(struct nullb *nullb)
1809 {
1810 	int nqueues = nr_cpu_ids;
1811 
1812 	if (g_poll_queues)
1813 		nqueues += g_poll_queues;
1814 
1815 	nullb->queues = kcalloc(nqueues, sizeof(struct nullb_queue),
1816 				GFP_KERNEL);
1817 	if (!nullb->queues)
1818 		return -ENOMEM;
1819 
1820 	nullb->queue_depth = nullb->dev->hw_queue_depth;
1821 	return 0;
1822 }
1823 
1824 static int init_driver_queues(struct nullb *nullb)
1825 {
1826 	struct nullb_queue *nq;
1827 	int i, ret = 0;
1828 
1829 	for (i = 0; i < nullb->dev->submit_queues; i++) {
1830 		nq = &nullb->queues[i];
1831 
1832 		null_init_queue(nullb, nq);
1833 
1834 		ret = setup_commands(nq);
1835 		if (ret)
1836 			return ret;
1837 		nullb->nr_queues++;
1838 	}
1839 	return 0;
1840 }
1841 
1842 static int null_gendisk_register(struct nullb *nullb)
1843 {
1844 	sector_t size = ((sector_t)nullb->dev->size * SZ_1M) >> SECTOR_SHIFT;
1845 	struct gendisk *disk = nullb->disk;
1846 
1847 	set_capacity(disk, size);
1848 
1849 	disk->major		= null_major;
1850 	disk->first_minor	= nullb->index;
1851 	disk->minors		= 1;
1852 	if (queue_is_mq(nullb->q))
1853 		disk->fops		= &null_rq_ops;
1854 	else
1855 		disk->fops		= &null_bio_ops;
1856 	disk->private_data	= nullb;
1857 	strncpy(disk->disk_name, nullb->disk_name, DISK_NAME_LEN);
1858 
1859 	if (nullb->dev->zoned) {
1860 		int ret = null_register_zoned_dev(nullb);
1861 
1862 		if (ret)
1863 			return ret;
1864 	}
1865 
1866 	return add_disk(disk);
1867 }
1868 
1869 static int null_init_tag_set(struct nullb *nullb, struct blk_mq_tag_set *set)
1870 {
1871 	int poll_queues;
1872 
1873 	set->ops = &null_mq_ops;
1874 	set->nr_hw_queues = nullb ? nullb->dev->submit_queues :
1875 						g_submit_queues;
1876 	poll_queues = nullb ? nullb->dev->poll_queues : g_poll_queues;
1877 	if (poll_queues)
1878 		set->nr_hw_queues += poll_queues;
1879 	set->queue_depth = nullb ? nullb->dev->hw_queue_depth :
1880 						g_hw_queue_depth;
1881 	set->numa_node = nullb ? nullb->dev->home_node : g_home_node;
1882 	set->cmd_size	= sizeof(struct nullb_cmd);
1883 	set->flags = BLK_MQ_F_SHOULD_MERGE;
1884 	if (g_no_sched)
1885 		set->flags |= BLK_MQ_F_NO_SCHED;
1886 	if (g_shared_tag_bitmap)
1887 		set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1888 	set->driver_data = nullb;
1889 	if (poll_queues)
1890 		set->nr_maps = 3;
1891 	else
1892 		set->nr_maps = 1;
1893 
1894 	if ((nullb && nullb->dev->blocking) || g_blocking)
1895 		set->flags |= BLK_MQ_F_BLOCKING;
1896 
1897 	return blk_mq_alloc_tag_set(set);
1898 }
1899 
1900 static int null_validate_conf(struct nullb_device *dev)
1901 {
1902 	dev->blocksize = round_down(dev->blocksize, 512);
1903 	dev->blocksize = clamp_t(unsigned int, dev->blocksize, 512, 4096);
1904 
1905 	if (dev->queue_mode == NULL_Q_MQ && dev->use_per_node_hctx) {
1906 		if (dev->submit_queues != nr_online_nodes)
1907 			dev->submit_queues = nr_online_nodes;
1908 	} else if (dev->submit_queues > nr_cpu_ids)
1909 		dev->submit_queues = nr_cpu_ids;
1910 	else if (dev->submit_queues == 0)
1911 		dev->submit_queues = 1;
1912 	dev->prev_submit_queues = dev->submit_queues;
1913 
1914 	if (dev->poll_queues > g_poll_queues)
1915 		dev->poll_queues = g_poll_queues;
1916 	dev->prev_poll_queues = dev->poll_queues;
1917 
1918 	dev->queue_mode = min_t(unsigned int, dev->queue_mode, NULL_Q_MQ);
1919 	dev->irqmode = min_t(unsigned int, dev->irqmode, NULL_IRQ_TIMER);
1920 
1921 	/* Do memory allocation, so set blocking */
1922 	if (dev->memory_backed)
1923 		dev->blocking = true;
1924 	else /* cache is meaningless */
1925 		dev->cache_size = 0;
1926 	dev->cache_size = min_t(unsigned long, ULONG_MAX / 1024 / 1024,
1927 						dev->cache_size);
1928 	dev->mbps = min_t(unsigned int, 1024 * 40, dev->mbps);
1929 	/* can not stop a queue */
1930 	if (dev->queue_mode == NULL_Q_BIO)
1931 		dev->mbps = 0;
1932 
1933 	if (dev->zoned &&
1934 	    (!dev->zone_size || !is_power_of_2(dev->zone_size))) {
1935 		pr_err("zone_size must be power-of-two\n");
1936 		return -EINVAL;
1937 	}
1938 
1939 	return 0;
1940 }
1941 
1942 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1943 static bool __null_setup_fault(struct fault_attr *attr, char *str)
1944 {
1945 	if (!str[0])
1946 		return true;
1947 
1948 	if (!setup_fault_attr(attr, str))
1949 		return false;
1950 
1951 	attr->verbose = 0;
1952 	return true;
1953 }
1954 #endif
1955 
1956 static bool null_setup_fault(void)
1957 {
1958 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1959 	if (!__null_setup_fault(&null_timeout_attr, g_timeout_str))
1960 		return false;
1961 	if (!__null_setup_fault(&null_requeue_attr, g_requeue_str))
1962 		return false;
1963 	if (!__null_setup_fault(&null_init_hctx_attr, g_init_hctx_str))
1964 		return false;
1965 #endif
1966 	return true;
1967 }
1968 
1969 static int null_add_dev(struct nullb_device *dev)
1970 {
1971 	struct nullb *nullb;
1972 	int rv;
1973 
1974 	rv = null_validate_conf(dev);
1975 	if (rv)
1976 		return rv;
1977 
1978 	nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, dev->home_node);
1979 	if (!nullb) {
1980 		rv = -ENOMEM;
1981 		goto out;
1982 	}
1983 	nullb->dev = dev;
1984 	dev->nullb = nullb;
1985 
1986 	spin_lock_init(&nullb->lock);
1987 
1988 	rv = setup_queues(nullb);
1989 	if (rv)
1990 		goto out_free_nullb;
1991 
1992 	if (dev->queue_mode == NULL_Q_MQ) {
1993 		if (shared_tags) {
1994 			nullb->tag_set = &tag_set;
1995 			rv = 0;
1996 		} else {
1997 			nullb->tag_set = &nullb->__tag_set;
1998 			rv = null_init_tag_set(nullb, nullb->tag_set);
1999 		}
2000 
2001 		if (rv)
2002 			goto out_cleanup_queues;
2003 
2004 		if (!null_setup_fault())
2005 			goto out_cleanup_tags;
2006 
2007 		nullb->tag_set->timeout = 5 * HZ;
2008 		nullb->disk = blk_mq_alloc_disk(nullb->tag_set, nullb);
2009 		if (IS_ERR(nullb->disk)) {
2010 			rv = PTR_ERR(nullb->disk);
2011 			goto out_cleanup_tags;
2012 		}
2013 		nullb->q = nullb->disk->queue;
2014 	} else if (dev->queue_mode == NULL_Q_BIO) {
2015 		rv = -ENOMEM;
2016 		nullb->disk = blk_alloc_disk(nullb->dev->home_node);
2017 		if (!nullb->disk)
2018 			goto out_cleanup_queues;
2019 
2020 		nullb->q = nullb->disk->queue;
2021 		rv = init_driver_queues(nullb);
2022 		if (rv)
2023 			goto out_cleanup_disk;
2024 	}
2025 
2026 	if (dev->mbps) {
2027 		set_bit(NULLB_DEV_FL_THROTTLED, &dev->flags);
2028 		nullb_setup_bwtimer(nullb);
2029 	}
2030 
2031 	if (dev->cache_size > 0) {
2032 		set_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
2033 		blk_queue_write_cache(nullb->q, true, true);
2034 	}
2035 
2036 	if (dev->zoned) {
2037 		rv = null_init_zoned_dev(dev, nullb->q);
2038 		if (rv)
2039 			goto out_cleanup_disk;
2040 	}
2041 
2042 	nullb->q->queuedata = nullb;
2043 	blk_queue_flag_set(QUEUE_FLAG_NONROT, nullb->q);
2044 	blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, nullb->q);
2045 
2046 	mutex_lock(&lock);
2047 	nullb->index = ida_simple_get(&nullb_indexes, 0, 0, GFP_KERNEL);
2048 	dev->index = nullb->index;
2049 	mutex_unlock(&lock);
2050 
2051 	blk_queue_logical_block_size(nullb->q, dev->blocksize);
2052 	blk_queue_physical_block_size(nullb->q, dev->blocksize);
2053 	if (!dev->max_sectors)
2054 		dev->max_sectors = queue_max_hw_sectors(nullb->q);
2055 	dev->max_sectors = min_t(unsigned int, dev->max_sectors,
2056 				 BLK_DEF_MAX_SECTORS);
2057 	blk_queue_max_hw_sectors(nullb->q, dev->max_sectors);
2058 
2059 	if (dev->virt_boundary)
2060 		blk_queue_virt_boundary(nullb->q, PAGE_SIZE - 1);
2061 
2062 	null_config_discard(nullb);
2063 
2064 	sprintf(nullb->disk_name, "nullb%d", nullb->index);
2065 
2066 	rv = null_gendisk_register(nullb);
2067 	if (rv)
2068 		goto out_cleanup_zone;
2069 
2070 	mutex_lock(&lock);
2071 	list_add_tail(&nullb->list, &nullb_list);
2072 	mutex_unlock(&lock);
2073 
2074 	return 0;
2075 out_cleanup_zone:
2076 	null_free_zoned_dev(dev);
2077 out_cleanup_disk:
2078 	blk_cleanup_disk(nullb->disk);
2079 out_cleanup_tags:
2080 	if (dev->queue_mode == NULL_Q_MQ && nullb->tag_set == &nullb->__tag_set)
2081 		blk_mq_free_tag_set(nullb->tag_set);
2082 out_cleanup_queues:
2083 	cleanup_queues(nullb);
2084 out_free_nullb:
2085 	kfree(nullb);
2086 	dev->nullb = NULL;
2087 out:
2088 	return rv;
2089 }
2090 
2091 static int __init null_init(void)
2092 {
2093 	int ret = 0;
2094 	unsigned int i;
2095 	struct nullb *nullb;
2096 	struct nullb_device *dev;
2097 
2098 	if (g_bs > PAGE_SIZE) {
2099 		pr_warn("invalid block size\n");
2100 		pr_warn("defaults block size to %lu\n", PAGE_SIZE);
2101 		g_bs = PAGE_SIZE;
2102 	}
2103 
2104 	if (g_max_sectors > BLK_DEF_MAX_SECTORS) {
2105 		pr_warn("invalid max sectors\n");
2106 		pr_warn("defaults max sectors to %u\n", BLK_DEF_MAX_SECTORS);
2107 		g_max_sectors = BLK_DEF_MAX_SECTORS;
2108 	}
2109 
2110 	if (g_home_node != NUMA_NO_NODE && g_home_node >= nr_online_nodes) {
2111 		pr_err("invalid home_node value\n");
2112 		g_home_node = NUMA_NO_NODE;
2113 	}
2114 
2115 	if (g_queue_mode == NULL_Q_RQ) {
2116 		pr_err("legacy IO path no longer available\n");
2117 		return -EINVAL;
2118 	}
2119 	if (g_queue_mode == NULL_Q_MQ && g_use_per_node_hctx) {
2120 		if (g_submit_queues != nr_online_nodes) {
2121 			pr_warn("submit_queues param is set to %u.\n",
2122 							nr_online_nodes);
2123 			g_submit_queues = nr_online_nodes;
2124 		}
2125 	} else if (g_submit_queues > nr_cpu_ids)
2126 		g_submit_queues = nr_cpu_ids;
2127 	else if (g_submit_queues <= 0)
2128 		g_submit_queues = 1;
2129 
2130 	if (g_queue_mode == NULL_Q_MQ && shared_tags) {
2131 		ret = null_init_tag_set(NULL, &tag_set);
2132 		if (ret)
2133 			return ret;
2134 	}
2135 
2136 	config_group_init(&nullb_subsys.su_group);
2137 	mutex_init(&nullb_subsys.su_mutex);
2138 
2139 	ret = configfs_register_subsystem(&nullb_subsys);
2140 	if (ret)
2141 		goto err_tagset;
2142 
2143 	mutex_init(&lock);
2144 
2145 	null_major = register_blkdev(0, "nullb");
2146 	if (null_major < 0) {
2147 		ret = null_major;
2148 		goto err_conf;
2149 	}
2150 
2151 	for (i = 0; i < nr_devices; i++) {
2152 		dev = null_alloc_dev();
2153 		if (!dev) {
2154 			ret = -ENOMEM;
2155 			goto err_dev;
2156 		}
2157 		ret = null_add_dev(dev);
2158 		if (ret) {
2159 			null_free_dev(dev);
2160 			goto err_dev;
2161 		}
2162 	}
2163 
2164 	pr_info("module loaded\n");
2165 	return 0;
2166 
2167 err_dev:
2168 	while (!list_empty(&nullb_list)) {
2169 		nullb = list_entry(nullb_list.next, struct nullb, list);
2170 		dev = nullb->dev;
2171 		null_del_dev(nullb);
2172 		null_free_dev(dev);
2173 	}
2174 	unregister_blkdev(null_major, "nullb");
2175 err_conf:
2176 	configfs_unregister_subsystem(&nullb_subsys);
2177 err_tagset:
2178 	if (g_queue_mode == NULL_Q_MQ && shared_tags)
2179 		blk_mq_free_tag_set(&tag_set);
2180 	return ret;
2181 }
2182 
2183 static void __exit null_exit(void)
2184 {
2185 	struct nullb *nullb;
2186 
2187 	configfs_unregister_subsystem(&nullb_subsys);
2188 
2189 	unregister_blkdev(null_major, "nullb");
2190 
2191 	mutex_lock(&lock);
2192 	while (!list_empty(&nullb_list)) {
2193 		struct nullb_device *dev;
2194 
2195 		nullb = list_entry(nullb_list.next, struct nullb, list);
2196 		dev = nullb->dev;
2197 		null_del_dev(nullb);
2198 		null_free_dev(dev);
2199 	}
2200 	mutex_unlock(&lock);
2201 
2202 	if (g_queue_mode == NULL_Q_MQ && shared_tags)
2203 		blk_mq_free_tag_set(&tag_set);
2204 }
2205 
2206 module_init(null_init);
2207 module_exit(null_exit);
2208 
2209 MODULE_AUTHOR("Jens Axboe <axboe@kernel.dk>");
2210 MODULE_LICENSE("GPL");
2211