xref: /openbmc/linux/drivers/block/null_blk/main.c (revision 2ae1beb3)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Add configfs and memory store: Kyungchan Koh <kkc6196@fb.com> and
4  * Shaohua Li <shli@fb.com>
5  */
6 #include <linux/module.h>
7 
8 #include <linux/moduleparam.h>
9 #include <linux/sched.h>
10 #include <linux/fs.h>
11 #include <linux/init.h>
12 #include "null_blk.h"
13 
14 #undef pr_fmt
15 #define pr_fmt(fmt)	"null_blk: " fmt
16 
17 #define FREE_BATCH		16
18 
19 #define TICKS_PER_SEC		50ULL
20 #define TIMER_INTERVAL		(NSEC_PER_SEC / TICKS_PER_SEC)
21 
22 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
23 static DECLARE_FAULT_ATTR(null_timeout_attr);
24 static DECLARE_FAULT_ATTR(null_requeue_attr);
25 static DECLARE_FAULT_ATTR(null_init_hctx_attr);
26 #endif
27 
28 static inline u64 mb_per_tick(int mbps)
29 {
30 	return (1 << 20) / TICKS_PER_SEC * ((u64) mbps);
31 }
32 
33 /*
34  * Status flags for nullb_device.
35  *
36  * CONFIGURED:	Device has been configured and turned on. Cannot reconfigure.
37  * UP:		Device is currently on and visible in userspace.
38  * THROTTLED:	Device is being throttled.
39  * CACHE:	Device is using a write-back cache.
40  */
41 enum nullb_device_flags {
42 	NULLB_DEV_FL_CONFIGURED	= 0,
43 	NULLB_DEV_FL_UP		= 1,
44 	NULLB_DEV_FL_THROTTLED	= 2,
45 	NULLB_DEV_FL_CACHE	= 3,
46 };
47 
48 #define MAP_SZ		((PAGE_SIZE >> SECTOR_SHIFT) + 2)
49 /*
50  * nullb_page is a page in memory for nullb devices.
51  *
52  * @page:	The page holding the data.
53  * @bitmap:	The bitmap represents which sector in the page has data.
54  *		Each bit represents one block size. For example, sector 8
55  *		will use the 7th bit
56  * The highest 2 bits of bitmap are for special purpose. LOCK means the cache
57  * page is being flushing to storage. FREE means the cache page is freed and
58  * should be skipped from flushing to storage. Please see
59  * null_make_cache_space
60  */
61 struct nullb_page {
62 	struct page *page;
63 	DECLARE_BITMAP(bitmap, MAP_SZ);
64 };
65 #define NULLB_PAGE_LOCK (MAP_SZ - 1)
66 #define NULLB_PAGE_FREE (MAP_SZ - 2)
67 
68 static LIST_HEAD(nullb_list);
69 static struct mutex lock;
70 static int null_major;
71 static DEFINE_IDA(nullb_indexes);
72 static struct blk_mq_tag_set tag_set;
73 
74 enum {
75 	NULL_IRQ_NONE		= 0,
76 	NULL_IRQ_SOFTIRQ	= 1,
77 	NULL_IRQ_TIMER		= 2,
78 };
79 
80 static bool g_virt_boundary = false;
81 module_param_named(virt_boundary, g_virt_boundary, bool, 0444);
82 MODULE_PARM_DESC(virt_boundary, "Require a virtual boundary for the device. Default: False");
83 
84 static int g_no_sched;
85 module_param_named(no_sched, g_no_sched, int, 0444);
86 MODULE_PARM_DESC(no_sched, "No io scheduler");
87 
88 static int g_submit_queues = 1;
89 module_param_named(submit_queues, g_submit_queues, int, 0444);
90 MODULE_PARM_DESC(submit_queues, "Number of submission queues");
91 
92 static int g_poll_queues = 1;
93 module_param_named(poll_queues, g_poll_queues, int, 0444);
94 MODULE_PARM_DESC(poll_queues, "Number of IOPOLL submission queues");
95 
96 static int g_home_node = NUMA_NO_NODE;
97 module_param_named(home_node, g_home_node, int, 0444);
98 MODULE_PARM_DESC(home_node, "Home node for the device");
99 
100 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
101 /*
102  * For more details about fault injection, please refer to
103  * Documentation/fault-injection/fault-injection.rst.
104  */
105 static char g_timeout_str[80];
106 module_param_string(timeout, g_timeout_str, sizeof(g_timeout_str), 0444);
107 MODULE_PARM_DESC(timeout, "Fault injection. timeout=<interval>,<probability>,<space>,<times>");
108 
109 static char g_requeue_str[80];
110 module_param_string(requeue, g_requeue_str, sizeof(g_requeue_str), 0444);
111 MODULE_PARM_DESC(requeue, "Fault injection. requeue=<interval>,<probability>,<space>,<times>");
112 
113 static char g_init_hctx_str[80];
114 module_param_string(init_hctx, g_init_hctx_str, sizeof(g_init_hctx_str), 0444);
115 MODULE_PARM_DESC(init_hctx, "Fault injection to fail hctx init. init_hctx=<interval>,<probability>,<space>,<times>");
116 #endif
117 
118 static int g_queue_mode = NULL_Q_MQ;
119 
120 static int null_param_store_val(const char *str, int *val, int min, int max)
121 {
122 	int ret, new_val;
123 
124 	ret = kstrtoint(str, 10, &new_val);
125 	if (ret)
126 		return -EINVAL;
127 
128 	if (new_val < min || new_val > max)
129 		return -EINVAL;
130 
131 	*val = new_val;
132 	return 0;
133 }
134 
135 static int null_set_queue_mode(const char *str, const struct kernel_param *kp)
136 {
137 	return null_param_store_val(str, &g_queue_mode, NULL_Q_BIO, NULL_Q_MQ);
138 }
139 
140 static const struct kernel_param_ops null_queue_mode_param_ops = {
141 	.set	= null_set_queue_mode,
142 	.get	= param_get_int,
143 };
144 
145 device_param_cb(queue_mode, &null_queue_mode_param_ops, &g_queue_mode, 0444);
146 MODULE_PARM_DESC(queue_mode, "Block interface to use (0=bio,1=rq,2=multiqueue)");
147 
148 static int g_gb = 250;
149 module_param_named(gb, g_gb, int, 0444);
150 MODULE_PARM_DESC(gb, "Size in GB");
151 
152 static int g_bs = 512;
153 module_param_named(bs, g_bs, int, 0444);
154 MODULE_PARM_DESC(bs, "Block size (in bytes)");
155 
156 static int g_max_sectors;
157 module_param_named(max_sectors, g_max_sectors, int, 0444);
158 MODULE_PARM_DESC(max_sectors, "Maximum size of a command (in 512B sectors)");
159 
160 static unsigned int nr_devices = 1;
161 module_param(nr_devices, uint, 0444);
162 MODULE_PARM_DESC(nr_devices, "Number of devices to register");
163 
164 static bool g_blocking;
165 module_param_named(blocking, g_blocking, bool, 0444);
166 MODULE_PARM_DESC(blocking, "Register as a blocking blk-mq driver device");
167 
168 static bool shared_tags;
169 module_param(shared_tags, bool, 0444);
170 MODULE_PARM_DESC(shared_tags, "Share tag set between devices for blk-mq");
171 
172 static bool g_shared_tag_bitmap;
173 module_param_named(shared_tag_bitmap, g_shared_tag_bitmap, bool, 0444);
174 MODULE_PARM_DESC(shared_tag_bitmap, "Use shared tag bitmap for all submission queues for blk-mq");
175 
176 static int g_irqmode = NULL_IRQ_SOFTIRQ;
177 
178 static int null_set_irqmode(const char *str, const struct kernel_param *kp)
179 {
180 	return null_param_store_val(str, &g_irqmode, NULL_IRQ_NONE,
181 					NULL_IRQ_TIMER);
182 }
183 
184 static const struct kernel_param_ops null_irqmode_param_ops = {
185 	.set	= null_set_irqmode,
186 	.get	= param_get_int,
187 };
188 
189 device_param_cb(irqmode, &null_irqmode_param_ops, &g_irqmode, 0444);
190 MODULE_PARM_DESC(irqmode, "IRQ completion handler. 0-none, 1-softirq, 2-timer");
191 
192 static unsigned long g_completion_nsec = 10000;
193 module_param_named(completion_nsec, g_completion_nsec, ulong, 0444);
194 MODULE_PARM_DESC(completion_nsec, "Time in ns to complete a request in hardware. Default: 10,000ns");
195 
196 static int g_hw_queue_depth = 64;
197 module_param_named(hw_queue_depth, g_hw_queue_depth, int, 0444);
198 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 64");
199 
200 static bool g_use_per_node_hctx;
201 module_param_named(use_per_node_hctx, g_use_per_node_hctx, bool, 0444);
202 MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware context queues. Default: false");
203 
204 static bool g_memory_backed;
205 module_param_named(memory_backed, g_memory_backed, bool, 0444);
206 MODULE_PARM_DESC(memory_backed, "Create a memory-backed block device. Default: false");
207 
208 static bool g_discard;
209 module_param_named(discard, g_discard, bool, 0444);
210 MODULE_PARM_DESC(discard, "Support discard operations (requires memory-backed null_blk device). Default: false");
211 
212 static unsigned long g_cache_size;
213 module_param_named(cache_size, g_cache_size, ulong, 0444);
214 MODULE_PARM_DESC(mbps, "Cache size in MiB for memory-backed device. Default: 0 (none)");
215 
216 static unsigned int g_mbps;
217 module_param_named(mbps, g_mbps, uint, 0444);
218 MODULE_PARM_DESC(mbps, "Limit maximum bandwidth (in MiB/s). Default: 0 (no limit)");
219 
220 static bool g_zoned;
221 module_param_named(zoned, g_zoned, bool, S_IRUGO);
222 MODULE_PARM_DESC(zoned, "Make device as a host-managed zoned block device. Default: false");
223 
224 static unsigned long g_zone_size = 256;
225 module_param_named(zone_size, g_zone_size, ulong, S_IRUGO);
226 MODULE_PARM_DESC(zone_size, "Zone size in MB when block device is zoned. Must be power-of-two: Default: 256");
227 
228 static unsigned long g_zone_capacity;
229 module_param_named(zone_capacity, g_zone_capacity, ulong, 0444);
230 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");
231 
232 static unsigned int g_zone_nr_conv;
233 module_param_named(zone_nr_conv, g_zone_nr_conv, uint, 0444);
234 MODULE_PARM_DESC(zone_nr_conv, "Number of conventional zones when block device is zoned. Default: 0");
235 
236 static unsigned int g_zone_max_open;
237 module_param_named(zone_max_open, g_zone_max_open, uint, 0444);
238 MODULE_PARM_DESC(zone_max_open, "Maximum number of open zones when block device is zoned. Default: 0 (no limit)");
239 
240 static unsigned int g_zone_max_active;
241 module_param_named(zone_max_active, g_zone_max_active, uint, 0444);
242 MODULE_PARM_DESC(zone_max_active, "Maximum number of active zones when block device is zoned. Default: 0 (no limit)");
243 
244 static struct nullb_device *null_alloc_dev(void);
245 static void null_free_dev(struct nullb_device *dev);
246 static void null_del_dev(struct nullb *nullb);
247 static int null_add_dev(struct nullb_device *dev);
248 static struct nullb *null_find_dev_by_name(const char *name);
249 static void null_free_device_storage(struct nullb_device *dev, bool is_cache);
250 
251 static inline struct nullb_device *to_nullb_device(struct config_item *item)
252 {
253 	return item ? container_of(to_config_group(item), struct nullb_device, group) : NULL;
254 }
255 
256 static 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 
261 static 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 
267 static inline ssize_t nullb_device_bool_attr_show(bool val, char *page)
268 {
269 	return snprintf(page, PAGE_SIZE, "%u\n", val);
270 }
271 
272 static 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);
279 	if (result < 0)
280 		return result;
281 
282 	*val = tmp;
283 	return count;
284 }
285 
286 static 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);
293 	if (result < 0)
294 		return result;
295 
296 	*val = tmp;
297 	return count;
298 }
299 
300 static 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);
307 	if (result < 0)
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}. */
315 #define NULLB_DEVICE_ATTR(NAME, TYPE, APPLY)				\
316 static ssize_t								\
317 nullb_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 }									\
322 static ssize_t								\
323 nullb_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);		\
328 	TYPE new_value = 0;						\
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 }									\
343 CONFIGFS_ATTR(nullb_device_, NAME);
344 
345 static int nullb_update_nr_hw_queues(struct nullb_device *dev,
346 				     unsigned int submit_queues,
347 				     unsigned int poll_queues)
348 
349 {
350 	struct blk_mq_tag_set *set;
351 	int ret, nr_hw_queues;
352 
353 	if (!dev->nullb)
354 		return 0;
355 
356 	/*
357 	 * Make sure at least one submit queue exists.
358 	 */
359 	if (!submit_queues)
360 		return -EINVAL;
361 
362 	/*
363 	 * Make sure that null_init_hctx() does not access nullb->queues[] past
364 	 * the end of that array.
365 	 */
366 	if (submit_queues > nr_cpu_ids || poll_queues > g_poll_queues)
367 		return -EINVAL;
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 
392 static 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 
398 static 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);
402 }
403 
404 NULLB_DEVICE_ATTR(size, ulong, NULL);
405 NULLB_DEVICE_ATTR(completion_nsec, ulong, NULL);
406 NULLB_DEVICE_ATTR(submit_queues, uint, nullb_apply_submit_queues);
407 NULLB_DEVICE_ATTR(poll_queues, uint, nullb_apply_poll_queues);
408 NULLB_DEVICE_ATTR(home_node, uint, NULL);
409 NULLB_DEVICE_ATTR(queue_mode, uint, NULL);
410 NULLB_DEVICE_ATTR(blocksize, uint, NULL);
411 NULLB_DEVICE_ATTR(max_sectors, uint, NULL);
412 NULLB_DEVICE_ATTR(irqmode, uint, NULL);
413 NULLB_DEVICE_ATTR(hw_queue_depth, uint, NULL);
414 NULLB_DEVICE_ATTR(index, uint, NULL);
415 NULLB_DEVICE_ATTR(blocking, bool, NULL);
416 NULLB_DEVICE_ATTR(use_per_node_hctx, bool, NULL);
417 NULLB_DEVICE_ATTR(memory_backed, bool, NULL);
418 NULLB_DEVICE_ATTR(discard, bool, NULL);
419 NULLB_DEVICE_ATTR(mbps, uint, NULL);
420 NULLB_DEVICE_ATTR(cache_size, ulong, NULL);
421 NULLB_DEVICE_ATTR(zoned, bool, NULL);
422 NULLB_DEVICE_ATTR(zone_size, ulong, NULL);
423 NULLB_DEVICE_ATTR(zone_capacity, ulong, NULL);
424 NULLB_DEVICE_ATTR(zone_nr_conv, uint, NULL);
425 NULLB_DEVICE_ATTR(zone_max_open, uint, NULL);
426 NULLB_DEVICE_ATTR(zone_max_active, uint, NULL);
427 NULLB_DEVICE_ATTR(virt_boundary, bool, NULL);
428 NULLB_DEVICE_ATTR(no_sched, bool, NULL);
429 NULLB_DEVICE_ATTR(shared_tag_bitmap, bool, NULL);
430 
431 static 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 
436 static 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;
450 		ret = null_add_dev(dev);
451 		if (ret) {
452 			clear_bit(NULLB_DEV_FL_UP, &dev->flags);
453 			return ret;
454 		}
455 
456 		set_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
457 		dev->power = newp;
458 	} else if (dev->power && !newp) {
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 		}
465 		clear_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
466 	}
467 
468 	return count;
469 }
470 
471 CONFIGFS_ATTR(nullb_device_, power);
472 
473 static 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 
480 static 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;
520 out:
521 	kfree(orig);
522 	return ret;
523 }
524 CONFIGFS_ATTR(nullb_device_, badblocks);
525 
526 static 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 }
533 CONFIGFS_ATTR_WO(nullb_device_, zone_readonly);
534 
535 static 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 }
542 CONFIGFS_ATTR_WO(nullb_device_, zone_offline);
543 
544 static struct configfs_attribute *nullb_device_attrs[] = {
545 	&nullb_device_attr_size,
546 	&nullb_device_attr_completion_nsec,
547 	&nullb_device_attr_submit_queues,
548 	&nullb_device_attr_poll_queues,
549 	&nullb_device_attr_home_node,
550 	&nullb_device_attr_queue_mode,
551 	&nullb_device_attr_blocksize,
552 	&nullb_device_attr_max_sectors,
553 	&nullb_device_attr_irqmode,
554 	&nullb_device_attr_hw_queue_depth,
555 	&nullb_device_attr_index,
556 	&nullb_device_attr_blocking,
557 	&nullb_device_attr_use_per_node_hctx,
558 	&nullb_device_attr_power,
559 	&nullb_device_attr_memory_backed,
560 	&nullb_device_attr_discard,
561 	&nullb_device_attr_mbps,
562 	&nullb_device_attr_cache_size,
563 	&nullb_device_attr_badblocks,
564 	&nullb_device_attr_zoned,
565 	&nullb_device_attr_zone_size,
566 	&nullb_device_attr_zone_capacity,
567 	&nullb_device_attr_zone_nr_conv,
568 	&nullb_device_attr_zone_max_open,
569 	&nullb_device_attr_zone_max_active,
570 	&nullb_device_attr_zone_readonly,
571 	&nullb_device_attr_zone_offline,
572 	&nullb_device_attr_virt_boundary,
573 	&nullb_device_attr_no_sched,
574 	&nullb_device_attr_shared_tag_bitmap,
575 	NULL,
576 };
577 
578 static void nullb_device_release(struct config_item *item)
579 {
580 	struct nullb_device *dev = to_nullb_device(item);
581 
582 	null_free_device_storage(dev, false);
583 	null_free_dev(dev);
584 }
585 
586 static struct configfs_item_operations nullb_device_ops = {
587 	.release	= nullb_device_release,
588 };
589 
590 static const struct config_item_type nullb_device_type = {
591 	.ct_item_ops	= &nullb_device_ops,
592 	.ct_attrs	= nullb_device_attrs,
593 	.ct_owner	= THIS_MODULE,
594 };
595 
596 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
597 
598 static void nullb_add_fault_config(struct nullb_device *dev)
599 {
600 	fault_config_init(&dev->timeout_config, "timeout_inject");
601 	fault_config_init(&dev->requeue_config, "requeue_inject");
602 	fault_config_init(&dev->init_hctx_fault_config, "init_hctx_fault_inject");
603 
604 	configfs_add_default_group(&dev->timeout_config.group, &dev->group);
605 	configfs_add_default_group(&dev->requeue_config.group, &dev->group);
606 	configfs_add_default_group(&dev->init_hctx_fault_config.group, &dev->group);
607 }
608 
609 #else
610 
611 static void nullb_add_fault_config(struct nullb_device *dev)
612 {
613 }
614 
615 #endif
616 
617 static struct
618 config_group *nullb_group_make_group(struct config_group *group, const char *name)
619 {
620 	struct nullb_device *dev;
621 
622 	if (null_find_dev_by_name(name))
623 		return ERR_PTR(-EEXIST);
624 
625 	dev = null_alloc_dev();
626 	if (!dev)
627 		return ERR_PTR(-ENOMEM);
628 
629 	config_group_init_type_name(&dev->group, name, &nullb_device_type);
630 	nullb_add_fault_config(dev);
631 
632 	return &dev->group;
633 }
634 
635 static void
636 nullb_group_drop_item(struct config_group *group, struct config_item *item)
637 {
638 	struct nullb_device *dev = to_nullb_device(item);
639 
640 	if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
641 		mutex_lock(&lock);
642 		dev->power = false;
643 		null_del_dev(dev->nullb);
644 		mutex_unlock(&lock);
645 	}
646 
647 	config_item_put(item);
648 }
649 
650 static ssize_t memb_group_features_show(struct config_item *item, char *page)
651 {
652 	return snprintf(page, PAGE_SIZE,
653 			"badblocks,blocking,blocksize,cache_size,"
654 			"completion_nsec,discard,home_node,hw_queue_depth,"
655 			"irqmode,max_sectors,mbps,memory_backed,no_sched,"
656 			"poll_queues,power,queue_mode,shared_tag_bitmap,size,"
657 			"submit_queues,use_per_node_hctx,virt_boundary,zoned,"
658 			"zone_capacity,zone_max_active,zone_max_open,"
659 			"zone_nr_conv,zone_offline,zone_readonly,zone_size\n");
660 }
661 
662 CONFIGFS_ATTR_RO(memb_group_, features);
663 
664 static struct configfs_attribute *nullb_group_attrs[] = {
665 	&memb_group_attr_features,
666 	NULL,
667 };
668 
669 static struct configfs_group_operations nullb_group_ops = {
670 	.make_group	= nullb_group_make_group,
671 	.drop_item	= nullb_group_drop_item,
672 };
673 
674 static const struct config_item_type nullb_group_type = {
675 	.ct_group_ops	= &nullb_group_ops,
676 	.ct_attrs	= nullb_group_attrs,
677 	.ct_owner	= THIS_MODULE,
678 };
679 
680 static struct configfs_subsystem nullb_subsys = {
681 	.su_group = {
682 		.cg_item = {
683 			.ci_namebuf = "nullb",
684 			.ci_type = &nullb_group_type,
685 		},
686 	},
687 };
688 
689 static inline int null_cache_active(struct nullb *nullb)
690 {
691 	return test_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
692 }
693 
694 static struct nullb_device *null_alloc_dev(void)
695 {
696 	struct nullb_device *dev;
697 
698 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
699 	if (!dev)
700 		return NULL;
701 
702 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
703 	dev->timeout_config.attr = null_timeout_attr;
704 	dev->requeue_config.attr = null_requeue_attr;
705 	dev->init_hctx_fault_config.attr = null_init_hctx_attr;
706 #endif
707 
708 	INIT_RADIX_TREE(&dev->data, GFP_ATOMIC);
709 	INIT_RADIX_TREE(&dev->cache, GFP_ATOMIC);
710 	if (badblocks_init(&dev->badblocks, 0)) {
711 		kfree(dev);
712 		return NULL;
713 	}
714 
715 	dev->size = g_gb * 1024;
716 	dev->completion_nsec = g_completion_nsec;
717 	dev->submit_queues = g_submit_queues;
718 	dev->prev_submit_queues = g_submit_queues;
719 	dev->poll_queues = g_poll_queues;
720 	dev->prev_poll_queues = g_poll_queues;
721 	dev->home_node = g_home_node;
722 	dev->queue_mode = g_queue_mode;
723 	dev->blocksize = g_bs;
724 	dev->max_sectors = g_max_sectors;
725 	dev->irqmode = g_irqmode;
726 	dev->hw_queue_depth = g_hw_queue_depth;
727 	dev->blocking = g_blocking;
728 	dev->memory_backed = g_memory_backed;
729 	dev->discard = g_discard;
730 	dev->cache_size = g_cache_size;
731 	dev->mbps = g_mbps;
732 	dev->use_per_node_hctx = g_use_per_node_hctx;
733 	dev->zoned = g_zoned;
734 	dev->zone_size = g_zone_size;
735 	dev->zone_capacity = g_zone_capacity;
736 	dev->zone_nr_conv = g_zone_nr_conv;
737 	dev->zone_max_open = g_zone_max_open;
738 	dev->zone_max_active = g_zone_max_active;
739 	dev->virt_boundary = g_virt_boundary;
740 	dev->no_sched = g_no_sched;
741 	dev->shared_tag_bitmap = g_shared_tag_bitmap;
742 	return dev;
743 }
744 
745 static void null_free_dev(struct nullb_device *dev)
746 {
747 	if (!dev)
748 		return;
749 
750 	null_free_zoned_dev(dev);
751 	badblocks_exit(&dev->badblocks);
752 	kfree(dev);
753 }
754 
755 static void put_tag(struct nullb_queue *nq, unsigned int tag)
756 {
757 	clear_bit_unlock(tag, nq->tag_map);
758 
759 	if (waitqueue_active(&nq->wait))
760 		wake_up(&nq->wait);
761 }
762 
763 static unsigned int get_tag(struct nullb_queue *nq)
764 {
765 	unsigned int tag;
766 
767 	do {
768 		tag = find_first_zero_bit(nq->tag_map, nq->queue_depth);
769 		if (tag >= nq->queue_depth)
770 			return -1U;
771 	} while (test_and_set_bit_lock(tag, nq->tag_map));
772 
773 	return tag;
774 }
775 
776 static void free_cmd(struct nullb_cmd *cmd)
777 {
778 	put_tag(cmd->nq, cmd->tag);
779 }
780 
781 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer);
782 
783 static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq)
784 {
785 	struct nullb_cmd *cmd;
786 	unsigned int tag;
787 
788 	tag = get_tag(nq);
789 	if (tag != -1U) {
790 		cmd = &nq->cmds[tag];
791 		cmd->tag = tag;
792 		cmd->error = BLK_STS_OK;
793 		cmd->nq = nq;
794 		if (nq->dev->irqmode == NULL_IRQ_TIMER) {
795 			hrtimer_init(&cmd->timer, CLOCK_MONOTONIC,
796 				     HRTIMER_MODE_REL);
797 			cmd->timer.function = null_cmd_timer_expired;
798 		}
799 		return cmd;
800 	}
801 
802 	return NULL;
803 }
804 
805 static struct nullb_cmd *alloc_cmd(struct nullb_queue *nq, struct bio *bio)
806 {
807 	struct nullb_cmd *cmd;
808 	DEFINE_WAIT(wait);
809 
810 	do {
811 		/*
812 		 * This avoids multiple return statements, multiple calls to
813 		 * __alloc_cmd() and a fast path call to prepare_to_wait().
814 		 */
815 		cmd = __alloc_cmd(nq);
816 		if (cmd) {
817 			cmd->bio = bio;
818 			return cmd;
819 		}
820 		prepare_to_wait(&nq->wait, &wait, TASK_UNINTERRUPTIBLE);
821 		io_schedule();
822 		finish_wait(&nq->wait, &wait);
823 	} while (1);
824 }
825 
826 static void end_cmd(struct nullb_cmd *cmd)
827 {
828 	int queue_mode = cmd->nq->dev->queue_mode;
829 
830 	switch (queue_mode)  {
831 	case NULL_Q_MQ:
832 		blk_mq_end_request(cmd->rq, cmd->error);
833 		return;
834 	case NULL_Q_BIO:
835 		cmd->bio->bi_status = cmd->error;
836 		bio_endio(cmd->bio);
837 		break;
838 	}
839 
840 	free_cmd(cmd);
841 }
842 
843 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
844 {
845 	end_cmd(container_of(timer, struct nullb_cmd, timer));
846 
847 	return HRTIMER_NORESTART;
848 }
849 
850 static void null_cmd_end_timer(struct nullb_cmd *cmd)
851 {
852 	ktime_t kt = cmd->nq->dev->completion_nsec;
853 
854 	hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL);
855 }
856 
857 static void null_complete_rq(struct request *rq)
858 {
859 	end_cmd(blk_mq_rq_to_pdu(rq));
860 }
861 
862 static struct nullb_page *null_alloc_page(void)
863 {
864 	struct nullb_page *t_page;
865 
866 	t_page = kmalloc(sizeof(struct nullb_page), GFP_NOIO);
867 	if (!t_page)
868 		return NULL;
869 
870 	t_page->page = alloc_pages(GFP_NOIO, 0);
871 	if (!t_page->page) {
872 		kfree(t_page);
873 		return NULL;
874 	}
875 
876 	memset(t_page->bitmap, 0, sizeof(t_page->bitmap));
877 	return t_page;
878 }
879 
880 static void null_free_page(struct nullb_page *t_page)
881 {
882 	__set_bit(NULLB_PAGE_FREE, t_page->bitmap);
883 	if (test_bit(NULLB_PAGE_LOCK, t_page->bitmap))
884 		return;
885 	__free_page(t_page->page);
886 	kfree(t_page);
887 }
888 
889 static bool null_page_empty(struct nullb_page *page)
890 {
891 	int size = MAP_SZ - 2;
892 
893 	return find_first_bit(page->bitmap, size) == size;
894 }
895 
896 static void null_free_sector(struct nullb *nullb, sector_t sector,
897 	bool is_cache)
898 {
899 	unsigned int sector_bit;
900 	u64 idx;
901 	struct nullb_page *t_page, *ret;
902 	struct radix_tree_root *root;
903 
904 	root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
905 	idx = sector >> PAGE_SECTORS_SHIFT;
906 	sector_bit = (sector & SECTOR_MASK);
907 
908 	t_page = radix_tree_lookup(root, idx);
909 	if (t_page) {
910 		__clear_bit(sector_bit, t_page->bitmap);
911 
912 		if (null_page_empty(t_page)) {
913 			ret = radix_tree_delete_item(root, idx, t_page);
914 			WARN_ON(ret != t_page);
915 			null_free_page(ret);
916 			if (is_cache)
917 				nullb->dev->curr_cache -= PAGE_SIZE;
918 		}
919 	}
920 }
921 
922 static struct nullb_page *null_radix_tree_insert(struct nullb *nullb, u64 idx,
923 	struct nullb_page *t_page, bool is_cache)
924 {
925 	struct radix_tree_root *root;
926 
927 	root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
928 
929 	if (radix_tree_insert(root, idx, t_page)) {
930 		null_free_page(t_page);
931 		t_page = radix_tree_lookup(root, idx);
932 		WARN_ON(!t_page || t_page->page->index != idx);
933 	} else if (is_cache)
934 		nullb->dev->curr_cache += PAGE_SIZE;
935 
936 	return t_page;
937 }
938 
939 static void null_free_device_storage(struct nullb_device *dev, bool is_cache)
940 {
941 	unsigned long pos = 0;
942 	int nr_pages;
943 	struct nullb_page *ret, *t_pages[FREE_BATCH];
944 	struct radix_tree_root *root;
945 
946 	root = is_cache ? &dev->cache : &dev->data;
947 
948 	do {
949 		int i;
950 
951 		nr_pages = radix_tree_gang_lookup(root,
952 				(void **)t_pages, pos, FREE_BATCH);
953 
954 		for (i = 0; i < nr_pages; i++) {
955 			pos = t_pages[i]->page->index;
956 			ret = radix_tree_delete_item(root, pos, t_pages[i]);
957 			WARN_ON(ret != t_pages[i]);
958 			null_free_page(ret);
959 		}
960 
961 		pos++;
962 	} while (nr_pages == FREE_BATCH);
963 
964 	if (is_cache)
965 		dev->curr_cache = 0;
966 }
967 
968 static struct nullb_page *__null_lookup_page(struct nullb *nullb,
969 	sector_t sector, bool for_write, bool is_cache)
970 {
971 	unsigned int sector_bit;
972 	u64 idx;
973 	struct nullb_page *t_page;
974 	struct radix_tree_root *root;
975 
976 	idx = sector >> PAGE_SECTORS_SHIFT;
977 	sector_bit = (sector & SECTOR_MASK);
978 
979 	root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
980 	t_page = radix_tree_lookup(root, idx);
981 	WARN_ON(t_page && t_page->page->index != idx);
982 
983 	if (t_page && (for_write || test_bit(sector_bit, t_page->bitmap)))
984 		return t_page;
985 
986 	return NULL;
987 }
988 
989 static struct nullb_page *null_lookup_page(struct nullb *nullb,
990 	sector_t sector, bool for_write, bool ignore_cache)
991 {
992 	struct nullb_page *page = NULL;
993 
994 	if (!ignore_cache)
995 		page = __null_lookup_page(nullb, sector, for_write, true);
996 	if (page)
997 		return page;
998 	return __null_lookup_page(nullb, sector, for_write, false);
999 }
1000 
1001 static struct nullb_page *null_insert_page(struct nullb *nullb,
1002 					   sector_t sector, bool ignore_cache)
1003 	__releases(&nullb->lock)
1004 	__acquires(&nullb->lock)
1005 {
1006 	u64 idx;
1007 	struct nullb_page *t_page;
1008 
1009 	t_page = null_lookup_page(nullb, sector, true, ignore_cache);
1010 	if (t_page)
1011 		return t_page;
1012 
1013 	spin_unlock_irq(&nullb->lock);
1014 
1015 	t_page = null_alloc_page();
1016 	if (!t_page)
1017 		goto out_lock;
1018 
1019 	if (radix_tree_preload(GFP_NOIO))
1020 		goto out_freepage;
1021 
1022 	spin_lock_irq(&nullb->lock);
1023 	idx = sector >> PAGE_SECTORS_SHIFT;
1024 	t_page->page->index = idx;
1025 	t_page = null_radix_tree_insert(nullb, idx, t_page, !ignore_cache);
1026 	radix_tree_preload_end();
1027 
1028 	return t_page;
1029 out_freepage:
1030 	null_free_page(t_page);
1031 out_lock:
1032 	spin_lock_irq(&nullb->lock);
1033 	return null_lookup_page(nullb, sector, true, ignore_cache);
1034 }
1035 
1036 static int null_flush_cache_page(struct nullb *nullb, struct nullb_page *c_page)
1037 {
1038 	int i;
1039 	unsigned int offset;
1040 	u64 idx;
1041 	struct nullb_page *t_page, *ret;
1042 	void *dst, *src;
1043 
1044 	idx = c_page->page->index;
1045 
1046 	t_page = null_insert_page(nullb, idx << PAGE_SECTORS_SHIFT, true);
1047 
1048 	__clear_bit(NULLB_PAGE_LOCK, c_page->bitmap);
1049 	if (test_bit(NULLB_PAGE_FREE, c_page->bitmap)) {
1050 		null_free_page(c_page);
1051 		if (t_page && null_page_empty(t_page)) {
1052 			ret = radix_tree_delete_item(&nullb->dev->data,
1053 				idx, t_page);
1054 			null_free_page(t_page);
1055 		}
1056 		return 0;
1057 	}
1058 
1059 	if (!t_page)
1060 		return -ENOMEM;
1061 
1062 	src = kmap_local_page(c_page->page);
1063 	dst = kmap_local_page(t_page->page);
1064 
1065 	for (i = 0; i < PAGE_SECTORS;
1066 			i += (nullb->dev->blocksize >> SECTOR_SHIFT)) {
1067 		if (test_bit(i, c_page->bitmap)) {
1068 			offset = (i << SECTOR_SHIFT);
1069 			memcpy(dst + offset, src + offset,
1070 				nullb->dev->blocksize);
1071 			__set_bit(i, t_page->bitmap);
1072 		}
1073 	}
1074 
1075 	kunmap_local(dst);
1076 	kunmap_local(src);
1077 
1078 	ret = radix_tree_delete_item(&nullb->dev->cache, idx, c_page);
1079 	null_free_page(ret);
1080 	nullb->dev->curr_cache -= PAGE_SIZE;
1081 
1082 	return 0;
1083 }
1084 
1085 static int null_make_cache_space(struct nullb *nullb, unsigned long n)
1086 {
1087 	int i, err, nr_pages;
1088 	struct nullb_page *c_pages[FREE_BATCH];
1089 	unsigned long flushed = 0, one_round;
1090 
1091 again:
1092 	if ((nullb->dev->cache_size * 1024 * 1024) >
1093 	     nullb->dev->curr_cache + n || nullb->dev->curr_cache == 0)
1094 		return 0;
1095 
1096 	nr_pages = radix_tree_gang_lookup(&nullb->dev->cache,
1097 			(void **)c_pages, nullb->cache_flush_pos, FREE_BATCH);
1098 	/*
1099 	 * nullb_flush_cache_page could unlock before using the c_pages. To
1100 	 * avoid race, we don't allow page free
1101 	 */
1102 	for (i = 0; i < nr_pages; i++) {
1103 		nullb->cache_flush_pos = c_pages[i]->page->index;
1104 		/*
1105 		 * We found the page which is being flushed to disk by other
1106 		 * threads
1107 		 */
1108 		if (test_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap))
1109 			c_pages[i] = NULL;
1110 		else
1111 			__set_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap);
1112 	}
1113 
1114 	one_round = 0;
1115 	for (i = 0; i < nr_pages; i++) {
1116 		if (c_pages[i] == NULL)
1117 			continue;
1118 		err = null_flush_cache_page(nullb, c_pages[i]);
1119 		if (err)
1120 			return err;
1121 		one_round++;
1122 	}
1123 	flushed += one_round << PAGE_SHIFT;
1124 
1125 	if (n > flushed) {
1126 		if (nr_pages == 0)
1127 			nullb->cache_flush_pos = 0;
1128 		if (one_round == 0) {
1129 			/* give other threads a chance */
1130 			spin_unlock_irq(&nullb->lock);
1131 			spin_lock_irq(&nullb->lock);
1132 		}
1133 		goto again;
1134 	}
1135 	return 0;
1136 }
1137 
1138 static int copy_to_nullb(struct nullb *nullb, struct page *source,
1139 	unsigned int off, sector_t sector, size_t n, bool is_fua)
1140 {
1141 	size_t temp, count = 0;
1142 	unsigned int offset;
1143 	struct nullb_page *t_page;
1144 
1145 	while (count < n) {
1146 		temp = min_t(size_t, nullb->dev->blocksize, n - count);
1147 
1148 		if (null_cache_active(nullb) && !is_fua)
1149 			null_make_cache_space(nullb, PAGE_SIZE);
1150 
1151 		offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
1152 		t_page = null_insert_page(nullb, sector,
1153 			!null_cache_active(nullb) || is_fua);
1154 		if (!t_page)
1155 			return -ENOSPC;
1156 
1157 		memcpy_page(t_page->page, offset, source, off + count, temp);
1158 
1159 		__set_bit(sector & SECTOR_MASK, t_page->bitmap);
1160 
1161 		if (is_fua)
1162 			null_free_sector(nullb, sector, true);
1163 
1164 		count += temp;
1165 		sector += temp >> SECTOR_SHIFT;
1166 	}
1167 	return 0;
1168 }
1169 
1170 static int copy_from_nullb(struct nullb *nullb, struct page *dest,
1171 	unsigned int off, sector_t sector, size_t n)
1172 {
1173 	size_t temp, count = 0;
1174 	unsigned int offset;
1175 	struct nullb_page *t_page;
1176 
1177 	while (count < n) {
1178 		temp = min_t(size_t, nullb->dev->blocksize, n - count);
1179 
1180 		offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
1181 		t_page = null_lookup_page(nullb, sector, false,
1182 			!null_cache_active(nullb));
1183 
1184 		if (t_page)
1185 			memcpy_page(dest, off + count, t_page->page, offset,
1186 				    temp);
1187 		else
1188 			zero_user(dest, off + count, temp);
1189 
1190 		count += temp;
1191 		sector += temp >> SECTOR_SHIFT;
1192 	}
1193 	return 0;
1194 }
1195 
1196 static void nullb_fill_pattern(struct nullb *nullb, struct page *page,
1197 			       unsigned int len, unsigned int off)
1198 {
1199 	memset_page(page, off, 0xff, len);
1200 }
1201 
1202 blk_status_t null_handle_discard(struct nullb_device *dev,
1203 				 sector_t sector, sector_t nr_sectors)
1204 {
1205 	struct nullb *nullb = dev->nullb;
1206 	size_t n = nr_sectors << SECTOR_SHIFT;
1207 	size_t temp;
1208 
1209 	spin_lock_irq(&nullb->lock);
1210 	while (n > 0) {
1211 		temp = min_t(size_t, n, dev->blocksize);
1212 		null_free_sector(nullb, sector, false);
1213 		if (null_cache_active(nullb))
1214 			null_free_sector(nullb, sector, true);
1215 		sector += temp >> SECTOR_SHIFT;
1216 		n -= temp;
1217 	}
1218 	spin_unlock_irq(&nullb->lock);
1219 
1220 	return BLK_STS_OK;
1221 }
1222 
1223 static int null_handle_flush(struct nullb *nullb)
1224 {
1225 	int err;
1226 
1227 	if (!null_cache_active(nullb))
1228 		return 0;
1229 
1230 	spin_lock_irq(&nullb->lock);
1231 	while (true) {
1232 		err = null_make_cache_space(nullb,
1233 			nullb->dev->cache_size * 1024 * 1024);
1234 		if (err || nullb->dev->curr_cache == 0)
1235 			break;
1236 	}
1237 
1238 	WARN_ON(!radix_tree_empty(&nullb->dev->cache));
1239 	spin_unlock_irq(&nullb->lock);
1240 	return err;
1241 }
1242 
1243 static int null_transfer(struct nullb *nullb, struct page *page,
1244 	unsigned int len, unsigned int off, bool is_write, sector_t sector,
1245 	bool is_fua)
1246 {
1247 	struct nullb_device *dev = nullb->dev;
1248 	unsigned int valid_len = len;
1249 	int err = 0;
1250 
1251 	if (!is_write) {
1252 		if (dev->zoned)
1253 			valid_len = null_zone_valid_read_len(nullb,
1254 				sector, len);
1255 
1256 		if (valid_len) {
1257 			err = copy_from_nullb(nullb, page, off,
1258 				sector, valid_len);
1259 			off += valid_len;
1260 			len -= valid_len;
1261 		}
1262 
1263 		if (len)
1264 			nullb_fill_pattern(nullb, page, len, off);
1265 		flush_dcache_page(page);
1266 	} else {
1267 		flush_dcache_page(page);
1268 		err = copy_to_nullb(nullb, page, off, sector, len, is_fua);
1269 	}
1270 
1271 	return err;
1272 }
1273 
1274 static int null_handle_rq(struct nullb_cmd *cmd)
1275 {
1276 	struct request *rq = cmd->rq;
1277 	struct nullb *nullb = cmd->nq->dev->nullb;
1278 	int err;
1279 	unsigned int len;
1280 	sector_t sector = blk_rq_pos(rq);
1281 	struct req_iterator iter;
1282 	struct bio_vec bvec;
1283 
1284 	spin_lock_irq(&nullb->lock);
1285 	rq_for_each_segment(bvec, rq, iter) {
1286 		len = bvec.bv_len;
1287 		err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1288 				     op_is_write(req_op(rq)), sector,
1289 				     rq->cmd_flags & REQ_FUA);
1290 		if (err) {
1291 			spin_unlock_irq(&nullb->lock);
1292 			return err;
1293 		}
1294 		sector += len >> SECTOR_SHIFT;
1295 	}
1296 	spin_unlock_irq(&nullb->lock);
1297 
1298 	return 0;
1299 }
1300 
1301 static int null_handle_bio(struct nullb_cmd *cmd)
1302 {
1303 	struct bio *bio = cmd->bio;
1304 	struct nullb *nullb = cmd->nq->dev->nullb;
1305 	int err;
1306 	unsigned int len;
1307 	sector_t sector = bio->bi_iter.bi_sector;
1308 	struct bio_vec bvec;
1309 	struct bvec_iter iter;
1310 
1311 	spin_lock_irq(&nullb->lock);
1312 	bio_for_each_segment(bvec, bio, iter) {
1313 		len = bvec.bv_len;
1314 		err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1315 				     op_is_write(bio_op(bio)), sector,
1316 				     bio->bi_opf & REQ_FUA);
1317 		if (err) {
1318 			spin_unlock_irq(&nullb->lock);
1319 			return err;
1320 		}
1321 		sector += len >> SECTOR_SHIFT;
1322 	}
1323 	spin_unlock_irq(&nullb->lock);
1324 	return 0;
1325 }
1326 
1327 static void null_stop_queue(struct nullb *nullb)
1328 {
1329 	struct request_queue *q = nullb->q;
1330 
1331 	if (nullb->dev->queue_mode == NULL_Q_MQ)
1332 		blk_mq_stop_hw_queues(q);
1333 }
1334 
1335 static void null_restart_queue_async(struct nullb *nullb)
1336 {
1337 	struct request_queue *q = nullb->q;
1338 
1339 	if (nullb->dev->queue_mode == NULL_Q_MQ)
1340 		blk_mq_start_stopped_hw_queues(q, true);
1341 }
1342 
1343 static inline blk_status_t null_handle_throttled(struct nullb_cmd *cmd)
1344 {
1345 	struct nullb_device *dev = cmd->nq->dev;
1346 	struct nullb *nullb = dev->nullb;
1347 	blk_status_t sts = BLK_STS_OK;
1348 	struct request *rq = cmd->rq;
1349 
1350 	if (!hrtimer_active(&nullb->bw_timer))
1351 		hrtimer_restart(&nullb->bw_timer);
1352 
1353 	if (atomic_long_sub_return(blk_rq_bytes(rq), &nullb->cur_bytes) < 0) {
1354 		null_stop_queue(nullb);
1355 		/* race with timer */
1356 		if (atomic_long_read(&nullb->cur_bytes) > 0)
1357 			null_restart_queue_async(nullb);
1358 		/* requeue request */
1359 		sts = BLK_STS_DEV_RESOURCE;
1360 	}
1361 	return sts;
1362 }
1363 
1364 static inline blk_status_t null_handle_badblocks(struct nullb_cmd *cmd,
1365 						 sector_t sector,
1366 						 sector_t nr_sectors)
1367 {
1368 	struct badblocks *bb = &cmd->nq->dev->badblocks;
1369 	sector_t first_bad;
1370 	int bad_sectors;
1371 
1372 	if (badblocks_check(bb, sector, nr_sectors, &first_bad, &bad_sectors))
1373 		return BLK_STS_IOERR;
1374 
1375 	return BLK_STS_OK;
1376 }
1377 
1378 static inline blk_status_t null_handle_memory_backed(struct nullb_cmd *cmd,
1379 						     enum req_op op,
1380 						     sector_t sector,
1381 						     sector_t nr_sectors)
1382 {
1383 	struct nullb_device *dev = cmd->nq->dev;
1384 	int err;
1385 
1386 	if (op == REQ_OP_DISCARD)
1387 		return null_handle_discard(dev, sector, nr_sectors);
1388 
1389 	if (dev->queue_mode == NULL_Q_BIO)
1390 		err = null_handle_bio(cmd);
1391 	else
1392 		err = null_handle_rq(cmd);
1393 
1394 	return errno_to_blk_status(err);
1395 }
1396 
1397 static void nullb_zero_read_cmd_buffer(struct nullb_cmd *cmd)
1398 {
1399 	struct nullb_device *dev = cmd->nq->dev;
1400 	struct bio *bio;
1401 
1402 	if (dev->memory_backed)
1403 		return;
1404 
1405 	if (dev->queue_mode == NULL_Q_BIO && bio_op(cmd->bio) == REQ_OP_READ) {
1406 		zero_fill_bio(cmd->bio);
1407 	} else if (req_op(cmd->rq) == REQ_OP_READ) {
1408 		__rq_for_each_bio(bio, cmd->rq)
1409 			zero_fill_bio(bio);
1410 	}
1411 }
1412 
1413 static inline void nullb_complete_cmd(struct nullb_cmd *cmd)
1414 {
1415 	/*
1416 	 * Since root privileges are required to configure the null_blk
1417 	 * driver, it is fine that this driver does not initialize the
1418 	 * data buffers of read commands. Zero-initialize these buffers
1419 	 * anyway if KMSAN is enabled to prevent that KMSAN complains
1420 	 * about null_blk not initializing read data buffers.
1421 	 */
1422 	if (IS_ENABLED(CONFIG_KMSAN))
1423 		nullb_zero_read_cmd_buffer(cmd);
1424 
1425 	/* Complete IO by inline, softirq or timer */
1426 	switch (cmd->nq->dev->irqmode) {
1427 	case NULL_IRQ_SOFTIRQ:
1428 		switch (cmd->nq->dev->queue_mode) {
1429 		case NULL_Q_MQ:
1430 			blk_mq_complete_request(cmd->rq);
1431 			break;
1432 		case NULL_Q_BIO:
1433 			/*
1434 			 * XXX: no proper submitting cpu information available.
1435 			 */
1436 			end_cmd(cmd);
1437 			break;
1438 		}
1439 		break;
1440 	case NULL_IRQ_NONE:
1441 		end_cmd(cmd);
1442 		break;
1443 	case NULL_IRQ_TIMER:
1444 		null_cmd_end_timer(cmd);
1445 		break;
1446 	}
1447 }
1448 
1449 blk_status_t null_process_cmd(struct nullb_cmd *cmd, enum req_op op,
1450 			      sector_t sector, unsigned int nr_sectors)
1451 {
1452 	struct nullb_device *dev = cmd->nq->dev;
1453 	blk_status_t ret;
1454 
1455 	if (dev->badblocks.shift != -1) {
1456 		ret = null_handle_badblocks(cmd, sector, nr_sectors);
1457 		if (ret != BLK_STS_OK)
1458 			return ret;
1459 	}
1460 
1461 	if (dev->memory_backed)
1462 		return null_handle_memory_backed(cmd, op, sector, nr_sectors);
1463 
1464 	return BLK_STS_OK;
1465 }
1466 
1467 static blk_status_t null_handle_cmd(struct nullb_cmd *cmd, sector_t sector,
1468 				    sector_t nr_sectors, enum req_op op)
1469 {
1470 	struct nullb_device *dev = cmd->nq->dev;
1471 	struct nullb *nullb = dev->nullb;
1472 	blk_status_t sts;
1473 
1474 	if (test_bit(NULLB_DEV_FL_THROTTLED, &dev->flags)) {
1475 		sts = null_handle_throttled(cmd);
1476 		if (sts != BLK_STS_OK)
1477 			return sts;
1478 	}
1479 
1480 	if (op == REQ_OP_FLUSH) {
1481 		cmd->error = errno_to_blk_status(null_handle_flush(nullb));
1482 		goto out;
1483 	}
1484 
1485 	if (dev->zoned)
1486 		sts = null_process_zoned_cmd(cmd, op, sector, nr_sectors);
1487 	else
1488 		sts = null_process_cmd(cmd, op, sector, nr_sectors);
1489 
1490 	/* Do not overwrite errors (e.g. timeout errors) */
1491 	if (cmd->error == BLK_STS_OK)
1492 		cmd->error = sts;
1493 
1494 out:
1495 	nullb_complete_cmd(cmd);
1496 	return BLK_STS_OK;
1497 }
1498 
1499 static enum hrtimer_restart nullb_bwtimer_fn(struct hrtimer *timer)
1500 {
1501 	struct nullb *nullb = container_of(timer, struct nullb, bw_timer);
1502 	ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1503 	unsigned int mbps = nullb->dev->mbps;
1504 
1505 	if (atomic_long_read(&nullb->cur_bytes) == mb_per_tick(mbps))
1506 		return HRTIMER_NORESTART;
1507 
1508 	atomic_long_set(&nullb->cur_bytes, mb_per_tick(mbps));
1509 	null_restart_queue_async(nullb);
1510 
1511 	hrtimer_forward_now(&nullb->bw_timer, timer_interval);
1512 
1513 	return HRTIMER_RESTART;
1514 }
1515 
1516 static void nullb_setup_bwtimer(struct nullb *nullb)
1517 {
1518 	ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1519 
1520 	hrtimer_init(&nullb->bw_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1521 	nullb->bw_timer.function = nullb_bwtimer_fn;
1522 	atomic_long_set(&nullb->cur_bytes, mb_per_tick(nullb->dev->mbps));
1523 	hrtimer_start(&nullb->bw_timer, timer_interval, HRTIMER_MODE_REL);
1524 }
1525 
1526 static struct nullb_queue *nullb_to_queue(struct nullb *nullb)
1527 {
1528 	int index = 0;
1529 
1530 	if (nullb->nr_queues != 1)
1531 		index = raw_smp_processor_id() / ((nr_cpu_ids + nullb->nr_queues - 1) / nullb->nr_queues);
1532 
1533 	return &nullb->queues[index];
1534 }
1535 
1536 static void null_submit_bio(struct bio *bio)
1537 {
1538 	sector_t sector = bio->bi_iter.bi_sector;
1539 	sector_t nr_sectors = bio_sectors(bio);
1540 	struct nullb *nullb = bio->bi_bdev->bd_disk->private_data;
1541 	struct nullb_queue *nq = nullb_to_queue(nullb);
1542 
1543 	null_handle_cmd(alloc_cmd(nq, bio), sector, nr_sectors, bio_op(bio));
1544 }
1545 
1546 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1547 
1548 static bool should_timeout_request(struct request *rq)
1549 {
1550 	struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq);
1551 	struct nullb_device *dev = cmd->nq->dev;
1552 
1553 	return should_fail(&dev->timeout_config.attr, 1);
1554 }
1555 
1556 static bool should_requeue_request(struct request *rq)
1557 {
1558 	struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq);
1559 	struct nullb_device *dev = cmd->nq->dev;
1560 
1561 	return should_fail(&dev->requeue_config.attr, 1);
1562 }
1563 
1564 static bool should_init_hctx_fail(struct nullb_device *dev)
1565 {
1566 	return should_fail(&dev->init_hctx_fault_config.attr, 1);
1567 }
1568 
1569 #else
1570 
1571 static bool should_timeout_request(struct request *rq)
1572 {
1573 	return false;
1574 }
1575 
1576 static bool should_requeue_request(struct request *rq)
1577 {
1578 	return false;
1579 }
1580 
1581 static bool should_init_hctx_fail(struct nullb_device *dev)
1582 {
1583 	return false;
1584 }
1585 
1586 #endif
1587 
1588 static void null_map_queues(struct blk_mq_tag_set *set)
1589 {
1590 	struct nullb *nullb = set->driver_data;
1591 	int i, qoff;
1592 	unsigned int submit_queues = g_submit_queues;
1593 	unsigned int poll_queues = g_poll_queues;
1594 
1595 	if (nullb) {
1596 		struct nullb_device *dev = nullb->dev;
1597 
1598 		/*
1599 		 * Refer nr_hw_queues of the tag set to check if the expected
1600 		 * number of hardware queues are prepared. If block layer failed
1601 		 * to prepare them, use previous numbers of submit queues and
1602 		 * poll queues to map queues.
1603 		 */
1604 		if (set->nr_hw_queues ==
1605 		    dev->submit_queues + dev->poll_queues) {
1606 			submit_queues = dev->submit_queues;
1607 			poll_queues = dev->poll_queues;
1608 		} else if (set->nr_hw_queues ==
1609 			   dev->prev_submit_queues + dev->prev_poll_queues) {
1610 			submit_queues = dev->prev_submit_queues;
1611 			poll_queues = dev->prev_poll_queues;
1612 		} else {
1613 			pr_warn("tag set has unexpected nr_hw_queues: %d\n",
1614 				set->nr_hw_queues);
1615 			WARN_ON_ONCE(true);
1616 			submit_queues = 1;
1617 			poll_queues = 0;
1618 		}
1619 	}
1620 
1621 	for (i = 0, qoff = 0; i < set->nr_maps; i++) {
1622 		struct blk_mq_queue_map *map = &set->map[i];
1623 
1624 		switch (i) {
1625 		case HCTX_TYPE_DEFAULT:
1626 			map->nr_queues = submit_queues;
1627 			break;
1628 		case HCTX_TYPE_READ:
1629 			map->nr_queues = 0;
1630 			continue;
1631 		case HCTX_TYPE_POLL:
1632 			map->nr_queues = poll_queues;
1633 			break;
1634 		}
1635 		map->queue_offset = qoff;
1636 		qoff += map->nr_queues;
1637 		blk_mq_map_queues(map);
1638 	}
1639 }
1640 
1641 static int null_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1642 {
1643 	struct nullb_queue *nq = hctx->driver_data;
1644 	LIST_HEAD(list);
1645 	int nr = 0;
1646 	struct request *rq;
1647 
1648 	spin_lock(&nq->poll_lock);
1649 	list_splice_init(&nq->poll_list, &list);
1650 	list_for_each_entry(rq, &list, queuelist)
1651 		blk_mq_set_request_complete(rq);
1652 	spin_unlock(&nq->poll_lock);
1653 
1654 	while (!list_empty(&list)) {
1655 		struct nullb_cmd *cmd;
1656 		struct request *req;
1657 
1658 		req = list_first_entry(&list, struct request, queuelist);
1659 		list_del_init(&req->queuelist);
1660 		cmd = blk_mq_rq_to_pdu(req);
1661 		cmd->error = null_process_cmd(cmd, req_op(req), blk_rq_pos(req),
1662 						blk_rq_sectors(req));
1663 		if (!blk_mq_add_to_batch(req, iob, (__force int) cmd->error,
1664 					blk_mq_end_request_batch))
1665 			end_cmd(cmd);
1666 		nr++;
1667 	}
1668 
1669 	return nr;
1670 }
1671 
1672 static enum blk_eh_timer_return null_timeout_rq(struct request *rq)
1673 {
1674 	struct blk_mq_hw_ctx *hctx = rq->mq_hctx;
1675 	struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq);
1676 
1677 	if (hctx->type == HCTX_TYPE_POLL) {
1678 		struct nullb_queue *nq = hctx->driver_data;
1679 
1680 		spin_lock(&nq->poll_lock);
1681 		/* The request may have completed meanwhile. */
1682 		if (blk_mq_request_completed(rq)) {
1683 			spin_unlock(&nq->poll_lock);
1684 			return BLK_EH_DONE;
1685 		}
1686 		list_del_init(&rq->queuelist);
1687 		spin_unlock(&nq->poll_lock);
1688 	}
1689 
1690 	pr_info("rq %p timed out\n", rq);
1691 
1692 	/*
1693 	 * If the device is marked as blocking (i.e. memory backed or zoned
1694 	 * device), the submission path may be blocked waiting for resources
1695 	 * and cause real timeouts. For these real timeouts, the submission
1696 	 * path will complete the request using blk_mq_complete_request().
1697 	 * Only fake timeouts need to execute blk_mq_complete_request() here.
1698 	 */
1699 	cmd->error = BLK_STS_TIMEOUT;
1700 	if (cmd->fake_timeout || hctx->type == HCTX_TYPE_POLL)
1701 		blk_mq_complete_request(rq);
1702 	return BLK_EH_DONE;
1703 }
1704 
1705 static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx,
1706 				  const struct blk_mq_queue_data *bd)
1707 {
1708 	struct request *rq = bd->rq;
1709 	struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq);
1710 	struct nullb_queue *nq = hctx->driver_data;
1711 	sector_t nr_sectors = blk_rq_sectors(rq);
1712 	sector_t sector = blk_rq_pos(rq);
1713 	const bool is_poll = hctx->type == HCTX_TYPE_POLL;
1714 
1715 	might_sleep_if(hctx->flags & BLK_MQ_F_BLOCKING);
1716 
1717 	if (!is_poll && nq->dev->irqmode == NULL_IRQ_TIMER) {
1718 		hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1719 		cmd->timer.function = null_cmd_timer_expired;
1720 	}
1721 	cmd->rq = rq;
1722 	cmd->error = BLK_STS_OK;
1723 	cmd->nq = nq;
1724 	cmd->fake_timeout = should_timeout_request(rq) ||
1725 		blk_should_fake_timeout(rq->q);
1726 
1727 	blk_mq_start_request(rq);
1728 
1729 	if (should_requeue_request(rq)) {
1730 		/*
1731 		 * Alternate between hitting the core BUSY path, and the
1732 		 * driver driven requeue path
1733 		 */
1734 		nq->requeue_selection++;
1735 		if (nq->requeue_selection & 1)
1736 			return BLK_STS_RESOURCE;
1737 		blk_mq_requeue_request(rq, true);
1738 		return BLK_STS_OK;
1739 	}
1740 
1741 	if (is_poll) {
1742 		spin_lock(&nq->poll_lock);
1743 		list_add_tail(&rq->queuelist, &nq->poll_list);
1744 		spin_unlock(&nq->poll_lock);
1745 		return BLK_STS_OK;
1746 	}
1747 	if (cmd->fake_timeout)
1748 		return BLK_STS_OK;
1749 
1750 	return null_handle_cmd(cmd, sector, nr_sectors, req_op(rq));
1751 }
1752 
1753 static void cleanup_queue(struct nullb_queue *nq)
1754 {
1755 	bitmap_free(nq->tag_map);
1756 	kfree(nq->cmds);
1757 }
1758 
1759 static void cleanup_queues(struct nullb *nullb)
1760 {
1761 	int i;
1762 
1763 	for (i = 0; i < nullb->nr_queues; i++)
1764 		cleanup_queue(&nullb->queues[i]);
1765 
1766 	kfree(nullb->queues);
1767 }
1768 
1769 static void null_exit_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx)
1770 {
1771 	struct nullb_queue *nq = hctx->driver_data;
1772 	struct nullb *nullb = nq->dev->nullb;
1773 
1774 	nullb->nr_queues--;
1775 }
1776 
1777 static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq)
1778 {
1779 	init_waitqueue_head(&nq->wait);
1780 	nq->queue_depth = nullb->queue_depth;
1781 	nq->dev = nullb->dev;
1782 	INIT_LIST_HEAD(&nq->poll_list);
1783 	spin_lock_init(&nq->poll_lock);
1784 }
1785 
1786 static int null_init_hctx(struct blk_mq_hw_ctx *hctx, void *driver_data,
1787 			  unsigned int hctx_idx)
1788 {
1789 	struct nullb *nullb = hctx->queue->queuedata;
1790 	struct nullb_queue *nq;
1791 
1792 	if (should_init_hctx_fail(nullb->dev))
1793 		return -EFAULT;
1794 
1795 	nq = &nullb->queues[hctx_idx];
1796 	hctx->driver_data = nq;
1797 	null_init_queue(nullb, nq);
1798 	nullb->nr_queues++;
1799 
1800 	return 0;
1801 }
1802 
1803 static const struct blk_mq_ops null_mq_ops = {
1804 	.queue_rq       = null_queue_rq,
1805 	.complete	= null_complete_rq,
1806 	.timeout	= null_timeout_rq,
1807 	.poll		= null_poll,
1808 	.map_queues	= null_map_queues,
1809 	.init_hctx	= null_init_hctx,
1810 	.exit_hctx	= null_exit_hctx,
1811 };
1812 
1813 static void null_del_dev(struct nullb *nullb)
1814 {
1815 	struct nullb_device *dev;
1816 
1817 	if (!nullb)
1818 		return;
1819 
1820 	dev = nullb->dev;
1821 
1822 	ida_simple_remove(&nullb_indexes, nullb->index);
1823 
1824 	list_del_init(&nullb->list);
1825 
1826 	del_gendisk(nullb->disk);
1827 
1828 	if (test_bit(NULLB_DEV_FL_THROTTLED, &nullb->dev->flags)) {
1829 		hrtimer_cancel(&nullb->bw_timer);
1830 		atomic_long_set(&nullb->cur_bytes, LONG_MAX);
1831 		null_restart_queue_async(nullb);
1832 	}
1833 
1834 	put_disk(nullb->disk);
1835 	if (dev->queue_mode == NULL_Q_MQ &&
1836 	    nullb->tag_set == &nullb->__tag_set)
1837 		blk_mq_free_tag_set(nullb->tag_set);
1838 	cleanup_queues(nullb);
1839 	if (null_cache_active(nullb))
1840 		null_free_device_storage(nullb->dev, true);
1841 	kfree(nullb);
1842 	dev->nullb = NULL;
1843 }
1844 
1845 static void null_config_discard(struct nullb *nullb)
1846 {
1847 	if (nullb->dev->discard == false)
1848 		return;
1849 
1850 	if (!nullb->dev->memory_backed) {
1851 		nullb->dev->discard = false;
1852 		pr_info("discard option is ignored without memory backing\n");
1853 		return;
1854 	}
1855 
1856 	if (nullb->dev->zoned) {
1857 		nullb->dev->discard = false;
1858 		pr_info("discard option is ignored in zoned mode\n");
1859 		return;
1860 	}
1861 
1862 	nullb->q->limits.discard_granularity = nullb->dev->blocksize;
1863 	blk_queue_max_discard_sectors(nullb->q, UINT_MAX >> 9);
1864 }
1865 
1866 static const struct block_device_operations null_bio_ops = {
1867 	.owner		= THIS_MODULE,
1868 	.submit_bio	= null_submit_bio,
1869 	.report_zones	= null_report_zones,
1870 };
1871 
1872 static const struct block_device_operations null_rq_ops = {
1873 	.owner		= THIS_MODULE,
1874 	.report_zones	= null_report_zones,
1875 };
1876 
1877 static int setup_commands(struct nullb_queue *nq)
1878 {
1879 	struct nullb_cmd *cmd;
1880 	int i;
1881 
1882 	nq->cmds = kcalloc(nq->queue_depth, sizeof(*cmd), GFP_KERNEL);
1883 	if (!nq->cmds)
1884 		return -ENOMEM;
1885 
1886 	nq->tag_map = bitmap_zalloc(nq->queue_depth, GFP_KERNEL);
1887 	if (!nq->tag_map) {
1888 		kfree(nq->cmds);
1889 		return -ENOMEM;
1890 	}
1891 
1892 	for (i = 0; i < nq->queue_depth; i++) {
1893 		cmd = &nq->cmds[i];
1894 		cmd->tag = -1U;
1895 	}
1896 
1897 	return 0;
1898 }
1899 
1900 static int setup_queues(struct nullb *nullb)
1901 {
1902 	int nqueues = nr_cpu_ids;
1903 
1904 	if (g_poll_queues)
1905 		nqueues += g_poll_queues;
1906 
1907 	nullb->queues = kcalloc(nqueues, sizeof(struct nullb_queue),
1908 				GFP_KERNEL);
1909 	if (!nullb->queues)
1910 		return -ENOMEM;
1911 
1912 	nullb->queue_depth = nullb->dev->hw_queue_depth;
1913 	return 0;
1914 }
1915 
1916 static int init_driver_queues(struct nullb *nullb)
1917 {
1918 	struct nullb_queue *nq;
1919 	int i, ret = 0;
1920 
1921 	for (i = 0; i < nullb->dev->submit_queues; i++) {
1922 		nq = &nullb->queues[i];
1923 
1924 		null_init_queue(nullb, nq);
1925 
1926 		ret = setup_commands(nq);
1927 		if (ret)
1928 			return ret;
1929 		nullb->nr_queues++;
1930 	}
1931 	return 0;
1932 }
1933 
1934 static int null_gendisk_register(struct nullb *nullb)
1935 {
1936 	sector_t size = ((sector_t)nullb->dev->size * SZ_1M) >> SECTOR_SHIFT;
1937 	struct gendisk *disk = nullb->disk;
1938 
1939 	set_capacity(disk, size);
1940 
1941 	disk->major		= null_major;
1942 	disk->first_minor	= nullb->index;
1943 	disk->minors		= 1;
1944 	if (queue_is_mq(nullb->q))
1945 		disk->fops		= &null_rq_ops;
1946 	else
1947 		disk->fops		= &null_bio_ops;
1948 	disk->private_data	= nullb;
1949 	strncpy(disk->disk_name, nullb->disk_name, DISK_NAME_LEN);
1950 
1951 	if (nullb->dev->zoned) {
1952 		int ret = null_register_zoned_dev(nullb);
1953 
1954 		if (ret)
1955 			return ret;
1956 	}
1957 
1958 	return add_disk(disk);
1959 }
1960 
1961 static int null_init_tag_set(struct nullb *nullb, struct blk_mq_tag_set *set)
1962 {
1963 	unsigned int flags = BLK_MQ_F_SHOULD_MERGE;
1964 	int hw_queues, numa_node;
1965 	unsigned int queue_depth;
1966 	int poll_queues;
1967 
1968 	if (nullb) {
1969 		hw_queues = nullb->dev->submit_queues;
1970 		poll_queues = nullb->dev->poll_queues;
1971 		queue_depth = nullb->dev->hw_queue_depth;
1972 		numa_node = nullb->dev->home_node;
1973 		if (nullb->dev->no_sched)
1974 			flags |= BLK_MQ_F_NO_SCHED;
1975 		if (nullb->dev->shared_tag_bitmap)
1976 			flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1977 		if (nullb->dev->blocking)
1978 			flags |= BLK_MQ_F_BLOCKING;
1979 	} else {
1980 		hw_queues = g_submit_queues;
1981 		poll_queues = g_poll_queues;
1982 		queue_depth = g_hw_queue_depth;
1983 		numa_node = g_home_node;
1984 		if (g_no_sched)
1985 			flags |= BLK_MQ_F_NO_SCHED;
1986 		if (g_shared_tag_bitmap)
1987 			flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1988 		if (g_blocking)
1989 			flags |= BLK_MQ_F_BLOCKING;
1990 	}
1991 
1992 	set->ops = &null_mq_ops;
1993 	set->cmd_size	= sizeof(struct nullb_cmd);
1994 	set->flags = flags;
1995 	set->driver_data = nullb;
1996 	set->nr_hw_queues = hw_queues;
1997 	set->queue_depth = queue_depth;
1998 	set->numa_node = numa_node;
1999 	if (poll_queues) {
2000 		set->nr_hw_queues += poll_queues;
2001 		set->nr_maps = 3;
2002 	} else {
2003 		set->nr_maps = 1;
2004 	}
2005 
2006 	return blk_mq_alloc_tag_set(set);
2007 }
2008 
2009 static int null_validate_conf(struct nullb_device *dev)
2010 {
2011 	if (dev->queue_mode == NULL_Q_RQ) {
2012 		pr_err("legacy IO path is no longer available\n");
2013 		return -EINVAL;
2014 	}
2015 
2016 	if (blk_validate_block_size(dev->blocksize))
2017 		return -EINVAL;
2018 
2019 	if (dev->queue_mode == NULL_Q_MQ && dev->use_per_node_hctx) {
2020 		if (dev->submit_queues != nr_online_nodes)
2021 			dev->submit_queues = nr_online_nodes;
2022 	} else if (dev->submit_queues > nr_cpu_ids)
2023 		dev->submit_queues = nr_cpu_ids;
2024 	else if (dev->submit_queues == 0)
2025 		dev->submit_queues = 1;
2026 	dev->prev_submit_queues = dev->submit_queues;
2027 
2028 	if (dev->poll_queues > g_poll_queues)
2029 		dev->poll_queues = g_poll_queues;
2030 	dev->prev_poll_queues = dev->poll_queues;
2031 
2032 	dev->queue_mode = min_t(unsigned int, dev->queue_mode, NULL_Q_MQ);
2033 	dev->irqmode = min_t(unsigned int, dev->irqmode, NULL_IRQ_TIMER);
2034 
2035 	/* Do memory allocation, so set blocking */
2036 	if (dev->memory_backed)
2037 		dev->blocking = true;
2038 	else /* cache is meaningless */
2039 		dev->cache_size = 0;
2040 	dev->cache_size = min_t(unsigned long, ULONG_MAX / 1024 / 1024,
2041 						dev->cache_size);
2042 	dev->mbps = min_t(unsigned int, 1024 * 40, dev->mbps);
2043 	/* can not stop a queue */
2044 	if (dev->queue_mode == NULL_Q_BIO)
2045 		dev->mbps = 0;
2046 
2047 	if (dev->zoned &&
2048 	    (!dev->zone_size || !is_power_of_2(dev->zone_size))) {
2049 		pr_err("zone_size must be power-of-two\n");
2050 		return -EINVAL;
2051 	}
2052 
2053 	return 0;
2054 }
2055 
2056 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
2057 static bool __null_setup_fault(struct fault_attr *attr, char *str)
2058 {
2059 	if (!str[0])
2060 		return true;
2061 
2062 	if (!setup_fault_attr(attr, str))
2063 		return false;
2064 
2065 	attr->verbose = 0;
2066 	return true;
2067 }
2068 #endif
2069 
2070 static bool null_setup_fault(void)
2071 {
2072 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
2073 	if (!__null_setup_fault(&null_timeout_attr, g_timeout_str))
2074 		return false;
2075 	if (!__null_setup_fault(&null_requeue_attr, g_requeue_str))
2076 		return false;
2077 	if (!__null_setup_fault(&null_init_hctx_attr, g_init_hctx_str))
2078 		return false;
2079 #endif
2080 	return true;
2081 }
2082 
2083 static int null_add_dev(struct nullb_device *dev)
2084 {
2085 	struct nullb *nullb;
2086 	int rv;
2087 
2088 	rv = null_validate_conf(dev);
2089 	if (rv)
2090 		return rv;
2091 
2092 	nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, dev->home_node);
2093 	if (!nullb) {
2094 		rv = -ENOMEM;
2095 		goto out;
2096 	}
2097 	nullb->dev = dev;
2098 	dev->nullb = nullb;
2099 
2100 	spin_lock_init(&nullb->lock);
2101 
2102 	rv = setup_queues(nullb);
2103 	if (rv)
2104 		goto out_free_nullb;
2105 
2106 	if (dev->queue_mode == NULL_Q_MQ) {
2107 		if (shared_tags) {
2108 			nullb->tag_set = &tag_set;
2109 			rv = 0;
2110 		} else {
2111 			nullb->tag_set = &nullb->__tag_set;
2112 			rv = null_init_tag_set(nullb, nullb->tag_set);
2113 		}
2114 
2115 		if (rv)
2116 			goto out_cleanup_queues;
2117 
2118 		nullb->tag_set->timeout = 5 * HZ;
2119 		nullb->disk = blk_mq_alloc_disk(nullb->tag_set, nullb);
2120 		if (IS_ERR(nullb->disk)) {
2121 			rv = PTR_ERR(nullb->disk);
2122 			goto out_cleanup_tags;
2123 		}
2124 		nullb->q = nullb->disk->queue;
2125 	} else if (dev->queue_mode == NULL_Q_BIO) {
2126 		rv = -ENOMEM;
2127 		nullb->disk = blk_alloc_disk(nullb->dev->home_node);
2128 		if (!nullb->disk)
2129 			goto out_cleanup_queues;
2130 
2131 		nullb->q = nullb->disk->queue;
2132 		rv = init_driver_queues(nullb);
2133 		if (rv)
2134 			goto out_cleanup_disk;
2135 	}
2136 
2137 	if (dev->mbps) {
2138 		set_bit(NULLB_DEV_FL_THROTTLED, &dev->flags);
2139 		nullb_setup_bwtimer(nullb);
2140 	}
2141 
2142 	if (dev->cache_size > 0) {
2143 		set_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
2144 		blk_queue_write_cache(nullb->q, true, true);
2145 	}
2146 
2147 	if (dev->zoned) {
2148 		rv = null_init_zoned_dev(dev, nullb->q);
2149 		if (rv)
2150 			goto out_cleanup_disk;
2151 	}
2152 
2153 	nullb->q->queuedata = nullb;
2154 	blk_queue_flag_set(QUEUE_FLAG_NONROT, nullb->q);
2155 
2156 	mutex_lock(&lock);
2157 	rv = ida_simple_get(&nullb_indexes, 0, 0, GFP_KERNEL);
2158 	if (rv < 0) {
2159 		mutex_unlock(&lock);
2160 		goto out_cleanup_zone;
2161 	}
2162 	nullb->index = rv;
2163 	dev->index = rv;
2164 	mutex_unlock(&lock);
2165 
2166 	blk_queue_logical_block_size(nullb->q, dev->blocksize);
2167 	blk_queue_physical_block_size(nullb->q, dev->blocksize);
2168 	if (dev->max_sectors)
2169 		blk_queue_max_hw_sectors(nullb->q, dev->max_sectors);
2170 
2171 	if (dev->virt_boundary)
2172 		blk_queue_virt_boundary(nullb->q, PAGE_SIZE - 1);
2173 
2174 	null_config_discard(nullb);
2175 
2176 	if (config_item_name(&dev->group.cg_item)) {
2177 		/* Use configfs dir name as the device name */
2178 		snprintf(nullb->disk_name, sizeof(nullb->disk_name),
2179 			 "%s", config_item_name(&dev->group.cg_item));
2180 	} else {
2181 		sprintf(nullb->disk_name, "nullb%d", nullb->index);
2182 	}
2183 
2184 	rv = null_gendisk_register(nullb);
2185 	if (rv)
2186 		goto out_ida_free;
2187 
2188 	mutex_lock(&lock);
2189 	list_add_tail(&nullb->list, &nullb_list);
2190 	mutex_unlock(&lock);
2191 
2192 	pr_info("disk %s created\n", nullb->disk_name);
2193 
2194 	return 0;
2195 
2196 out_ida_free:
2197 	ida_free(&nullb_indexes, nullb->index);
2198 out_cleanup_zone:
2199 	null_free_zoned_dev(dev);
2200 out_cleanup_disk:
2201 	put_disk(nullb->disk);
2202 out_cleanup_tags:
2203 	if (dev->queue_mode == NULL_Q_MQ && nullb->tag_set == &nullb->__tag_set)
2204 		blk_mq_free_tag_set(nullb->tag_set);
2205 out_cleanup_queues:
2206 	cleanup_queues(nullb);
2207 out_free_nullb:
2208 	kfree(nullb);
2209 	dev->nullb = NULL;
2210 out:
2211 	return rv;
2212 }
2213 
2214 static struct nullb *null_find_dev_by_name(const char *name)
2215 {
2216 	struct nullb *nullb = NULL, *nb;
2217 
2218 	mutex_lock(&lock);
2219 	list_for_each_entry(nb, &nullb_list, list) {
2220 		if (strcmp(nb->disk_name, name) == 0) {
2221 			nullb = nb;
2222 			break;
2223 		}
2224 	}
2225 	mutex_unlock(&lock);
2226 
2227 	return nullb;
2228 }
2229 
2230 static int null_create_dev(void)
2231 {
2232 	struct nullb_device *dev;
2233 	int ret;
2234 
2235 	dev = null_alloc_dev();
2236 	if (!dev)
2237 		return -ENOMEM;
2238 
2239 	ret = null_add_dev(dev);
2240 	if (ret) {
2241 		null_free_dev(dev);
2242 		return ret;
2243 	}
2244 
2245 	return 0;
2246 }
2247 
2248 static void null_destroy_dev(struct nullb *nullb)
2249 {
2250 	struct nullb_device *dev = nullb->dev;
2251 
2252 	null_del_dev(nullb);
2253 	null_free_device_storage(dev, false);
2254 	null_free_dev(dev);
2255 }
2256 
2257 static int __init null_init(void)
2258 {
2259 	int ret = 0;
2260 	unsigned int i;
2261 	struct nullb *nullb;
2262 
2263 	if (g_bs > PAGE_SIZE) {
2264 		pr_warn("invalid block size\n");
2265 		pr_warn("defaults block size to %lu\n", PAGE_SIZE);
2266 		g_bs = PAGE_SIZE;
2267 	}
2268 
2269 	if (g_home_node != NUMA_NO_NODE && g_home_node >= nr_online_nodes) {
2270 		pr_err("invalid home_node value\n");
2271 		g_home_node = NUMA_NO_NODE;
2272 	}
2273 
2274 	if (!null_setup_fault())
2275 		return -EINVAL;
2276 
2277 	if (g_queue_mode == NULL_Q_RQ) {
2278 		pr_err("legacy IO path is no longer available\n");
2279 		return -EINVAL;
2280 	}
2281 
2282 	if (g_queue_mode == NULL_Q_MQ && g_use_per_node_hctx) {
2283 		if (g_submit_queues != nr_online_nodes) {
2284 			pr_warn("submit_queues param is set to %u.\n",
2285 				nr_online_nodes);
2286 			g_submit_queues = nr_online_nodes;
2287 		}
2288 	} else if (g_submit_queues > nr_cpu_ids) {
2289 		g_submit_queues = nr_cpu_ids;
2290 	} else if (g_submit_queues <= 0) {
2291 		g_submit_queues = 1;
2292 	}
2293 
2294 	if (g_queue_mode == NULL_Q_MQ && shared_tags) {
2295 		ret = null_init_tag_set(NULL, &tag_set);
2296 		if (ret)
2297 			return ret;
2298 	}
2299 
2300 	config_group_init(&nullb_subsys.su_group);
2301 	mutex_init(&nullb_subsys.su_mutex);
2302 
2303 	ret = configfs_register_subsystem(&nullb_subsys);
2304 	if (ret)
2305 		goto err_tagset;
2306 
2307 	mutex_init(&lock);
2308 
2309 	null_major = register_blkdev(0, "nullb");
2310 	if (null_major < 0) {
2311 		ret = null_major;
2312 		goto err_conf;
2313 	}
2314 
2315 	for (i = 0; i < nr_devices; i++) {
2316 		ret = null_create_dev();
2317 		if (ret)
2318 			goto err_dev;
2319 	}
2320 
2321 	pr_info("module loaded\n");
2322 	return 0;
2323 
2324 err_dev:
2325 	while (!list_empty(&nullb_list)) {
2326 		nullb = list_entry(nullb_list.next, struct nullb, list);
2327 		null_destroy_dev(nullb);
2328 	}
2329 	unregister_blkdev(null_major, "nullb");
2330 err_conf:
2331 	configfs_unregister_subsystem(&nullb_subsys);
2332 err_tagset:
2333 	if (g_queue_mode == NULL_Q_MQ && shared_tags)
2334 		blk_mq_free_tag_set(&tag_set);
2335 	return ret;
2336 }
2337 
2338 static void __exit null_exit(void)
2339 {
2340 	struct nullb *nullb;
2341 
2342 	configfs_unregister_subsystem(&nullb_subsys);
2343 
2344 	unregister_blkdev(null_major, "nullb");
2345 
2346 	mutex_lock(&lock);
2347 	while (!list_empty(&nullb_list)) {
2348 		nullb = list_entry(nullb_list.next, struct nullb, list);
2349 		null_destroy_dev(nullb);
2350 	}
2351 	mutex_unlock(&lock);
2352 
2353 	if (g_queue_mode == NULL_Q_MQ && shared_tags)
2354 		blk_mq_free_tag_set(&tag_set);
2355 
2356 	mutex_destroy(&lock);
2357 }
2358 
2359 module_init(null_init);
2360 module_exit(null_exit);
2361 
2362 MODULE_AUTHOR("Jens Axboe <axboe@kernel.dk>");
2363 MODULE_DESCRIPTION("multi queue aware block test driver");
2364 MODULE_LICENSE("GPL");
2365