xref: /linux/drivers/block/null_blk/main.c (revision 001821b0e79716c4e17c71d8e053a23599a7a508)
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 /*
119  * Historic queue modes.
120  *
121  * These days nothing but NULL_Q_MQ is actually supported, but we keep it the
122  * enum for error reporting.
123  */
124 enum {
125 	NULL_Q_BIO	= 0,
126 	NULL_Q_RQ	= 1,
127 	NULL_Q_MQ	= 2,
128 };
129 
130 static int g_queue_mode = NULL_Q_MQ;
131 
132 static int null_param_store_val(const char *str, int *val, int min, int max)
133 {
134 	int ret, new_val;
135 
136 	ret = kstrtoint(str, 10, &new_val);
137 	if (ret)
138 		return -EINVAL;
139 
140 	if (new_val < min || new_val > max)
141 		return -EINVAL;
142 
143 	*val = new_val;
144 	return 0;
145 }
146 
147 static int null_set_queue_mode(const char *str, const struct kernel_param *kp)
148 {
149 	return null_param_store_val(str, &g_queue_mode, NULL_Q_BIO, NULL_Q_MQ);
150 }
151 
152 static const struct kernel_param_ops null_queue_mode_param_ops = {
153 	.set	= null_set_queue_mode,
154 	.get	= param_get_int,
155 };
156 
157 device_param_cb(queue_mode, &null_queue_mode_param_ops, &g_queue_mode, 0444);
158 MODULE_PARM_DESC(queue_mode, "Block interface to use (0=bio,1=rq,2=multiqueue)");
159 
160 static int g_gb = 250;
161 module_param_named(gb, g_gb, int, 0444);
162 MODULE_PARM_DESC(gb, "Size in GB");
163 
164 static int g_bs = 512;
165 module_param_named(bs, g_bs, int, 0444);
166 MODULE_PARM_DESC(bs, "Block size (in bytes)");
167 
168 static int g_max_sectors;
169 module_param_named(max_sectors, g_max_sectors, int, 0444);
170 MODULE_PARM_DESC(max_sectors, "Maximum size of a command (in 512B sectors)");
171 
172 static unsigned int nr_devices = 1;
173 module_param(nr_devices, uint, 0444);
174 MODULE_PARM_DESC(nr_devices, "Number of devices to register");
175 
176 static bool g_blocking;
177 module_param_named(blocking, g_blocking, bool, 0444);
178 MODULE_PARM_DESC(blocking, "Register as a blocking blk-mq driver device");
179 
180 static bool g_shared_tags;
181 module_param_named(shared_tags, g_shared_tags, bool, 0444);
182 MODULE_PARM_DESC(shared_tags, "Share tag set between devices for blk-mq");
183 
184 static bool g_shared_tag_bitmap;
185 module_param_named(shared_tag_bitmap, g_shared_tag_bitmap, bool, 0444);
186 MODULE_PARM_DESC(shared_tag_bitmap, "Use shared tag bitmap for all submission queues for blk-mq");
187 
188 static int g_irqmode = NULL_IRQ_SOFTIRQ;
189 
190 static int null_set_irqmode(const char *str, const struct kernel_param *kp)
191 {
192 	return null_param_store_val(str, &g_irqmode, NULL_IRQ_NONE,
193 					NULL_IRQ_TIMER);
194 }
195 
196 static const struct kernel_param_ops null_irqmode_param_ops = {
197 	.set	= null_set_irqmode,
198 	.get	= param_get_int,
199 };
200 
201 device_param_cb(irqmode, &null_irqmode_param_ops, &g_irqmode, 0444);
202 MODULE_PARM_DESC(irqmode, "IRQ completion handler. 0-none, 1-softirq, 2-timer");
203 
204 static unsigned long g_completion_nsec = 10000;
205 module_param_named(completion_nsec, g_completion_nsec, ulong, 0444);
206 MODULE_PARM_DESC(completion_nsec, "Time in ns to complete a request in hardware. Default: 10,000ns");
207 
208 static int g_hw_queue_depth = 64;
209 module_param_named(hw_queue_depth, g_hw_queue_depth, int, 0444);
210 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 64");
211 
212 static bool g_use_per_node_hctx;
213 module_param_named(use_per_node_hctx, g_use_per_node_hctx, bool, 0444);
214 MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware context queues. Default: false");
215 
216 static bool g_memory_backed;
217 module_param_named(memory_backed, g_memory_backed, bool, 0444);
218 MODULE_PARM_DESC(memory_backed, "Create a memory-backed block device. Default: false");
219 
220 static bool g_discard;
221 module_param_named(discard, g_discard, bool, 0444);
222 MODULE_PARM_DESC(discard, "Support discard operations (requires memory-backed null_blk device). Default: false");
223 
224 static unsigned long g_cache_size;
225 module_param_named(cache_size, g_cache_size, ulong, 0444);
226 MODULE_PARM_DESC(mbps, "Cache size in MiB for memory-backed device. Default: 0 (none)");
227 
228 static bool g_fua = true;
229 module_param_named(fua, g_fua, bool, 0444);
230 MODULE_PARM_DESC(zoned, "Enable/disable FUA support when cache_size is used. Default: true");
231 
232 static unsigned int g_mbps;
233 module_param_named(mbps, g_mbps, uint, 0444);
234 MODULE_PARM_DESC(mbps, "Limit maximum bandwidth (in MiB/s). Default: 0 (no limit)");
235 
236 static bool g_zoned;
237 module_param_named(zoned, g_zoned, bool, S_IRUGO);
238 MODULE_PARM_DESC(zoned, "Make device as a host-managed zoned block device. Default: false");
239 
240 static unsigned long g_zone_size = 256;
241 module_param_named(zone_size, g_zone_size, ulong, S_IRUGO);
242 MODULE_PARM_DESC(zone_size, "Zone size in MB when block device is zoned. Must be power-of-two: Default: 256");
243 
244 static unsigned long g_zone_capacity;
245 module_param_named(zone_capacity, g_zone_capacity, ulong, 0444);
246 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");
247 
248 static unsigned int g_zone_nr_conv;
249 module_param_named(zone_nr_conv, g_zone_nr_conv, uint, 0444);
250 MODULE_PARM_DESC(zone_nr_conv, "Number of conventional zones when block device is zoned. Default: 0");
251 
252 static unsigned int g_zone_max_open;
253 module_param_named(zone_max_open, g_zone_max_open, uint, 0444);
254 MODULE_PARM_DESC(zone_max_open, "Maximum number of open zones when block device is zoned. Default: 0 (no limit)");
255 
256 static unsigned int g_zone_max_active;
257 module_param_named(zone_max_active, g_zone_max_active, uint, 0444);
258 MODULE_PARM_DESC(zone_max_active, "Maximum number of active zones when block device is zoned. Default: 0 (no limit)");
259 
260 static int g_zone_append_max_sectors = INT_MAX;
261 module_param_named(zone_append_max_sectors, g_zone_append_max_sectors, int, 0444);
262 MODULE_PARM_DESC(zone_append_max_sectors,
263 		 "Maximum size of a zone append command (in 512B sectors). Specify 0 for zone append emulation");
264 
265 static struct nullb_device *null_alloc_dev(void);
266 static void null_free_dev(struct nullb_device *dev);
267 static void null_del_dev(struct nullb *nullb);
268 static int null_add_dev(struct nullb_device *dev);
269 static struct nullb *null_find_dev_by_name(const char *name);
270 static void null_free_device_storage(struct nullb_device *dev, bool is_cache);
271 
272 static inline struct nullb_device *to_nullb_device(struct config_item *item)
273 {
274 	return item ? container_of(to_config_group(item), struct nullb_device, group) : NULL;
275 }
276 
277 static inline ssize_t nullb_device_uint_attr_show(unsigned int val, char *page)
278 {
279 	return snprintf(page, PAGE_SIZE, "%u\n", val);
280 }
281 
282 static inline ssize_t nullb_device_ulong_attr_show(unsigned long val,
283 	char *page)
284 {
285 	return snprintf(page, PAGE_SIZE, "%lu\n", val);
286 }
287 
288 static inline ssize_t nullb_device_bool_attr_show(bool val, char *page)
289 {
290 	return snprintf(page, PAGE_SIZE, "%u\n", val);
291 }
292 
293 static ssize_t nullb_device_uint_attr_store(unsigned int *val,
294 	const char *page, size_t count)
295 {
296 	unsigned int tmp;
297 	int result;
298 
299 	result = kstrtouint(page, 0, &tmp);
300 	if (result < 0)
301 		return result;
302 
303 	*val = tmp;
304 	return count;
305 }
306 
307 static ssize_t nullb_device_ulong_attr_store(unsigned long *val,
308 	const char *page, size_t count)
309 {
310 	int result;
311 	unsigned long tmp;
312 
313 	result = kstrtoul(page, 0, &tmp);
314 	if (result < 0)
315 		return result;
316 
317 	*val = tmp;
318 	return count;
319 }
320 
321 static ssize_t nullb_device_bool_attr_store(bool *val, const char *page,
322 	size_t count)
323 {
324 	bool tmp;
325 	int result;
326 
327 	result = kstrtobool(page,  &tmp);
328 	if (result < 0)
329 		return result;
330 
331 	*val = tmp;
332 	return count;
333 }
334 
335 /* The following macro should only be used with TYPE = {uint, ulong, bool}. */
336 #define NULLB_DEVICE_ATTR(NAME, TYPE, APPLY)				\
337 static ssize_t								\
338 nullb_device_##NAME##_show(struct config_item *item, char *page)	\
339 {									\
340 	return nullb_device_##TYPE##_attr_show(				\
341 				to_nullb_device(item)->NAME, page);	\
342 }									\
343 static ssize_t								\
344 nullb_device_##NAME##_store(struct config_item *item, const char *page,	\
345 			    size_t count)				\
346 {									\
347 	int (*apply_fn)(struct nullb_device *dev, TYPE new_value) = APPLY;\
348 	struct nullb_device *dev = to_nullb_device(item);		\
349 	TYPE new_value = 0;						\
350 	int ret;							\
351 									\
352 	ret = nullb_device_##TYPE##_attr_store(&new_value, page, count);\
353 	if (ret < 0)							\
354 		return ret;						\
355 	if (apply_fn)							\
356 		ret = apply_fn(dev, new_value);				\
357 	else if (test_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags)) 	\
358 		ret = -EBUSY;						\
359 	if (ret < 0)							\
360 		return ret;						\
361 	dev->NAME = new_value;						\
362 	return count;							\
363 }									\
364 CONFIGFS_ATTR(nullb_device_, NAME);
365 
366 static int nullb_update_nr_hw_queues(struct nullb_device *dev,
367 				     unsigned int submit_queues,
368 				     unsigned int poll_queues)
369 
370 {
371 	struct blk_mq_tag_set *set;
372 	int ret, nr_hw_queues;
373 
374 	if (!dev->nullb)
375 		return 0;
376 
377 	/*
378 	 * Make sure at least one submit queue exists.
379 	 */
380 	if (!submit_queues)
381 		return -EINVAL;
382 
383 	/*
384 	 * Make sure that null_init_hctx() does not access nullb->queues[] past
385 	 * the end of that array.
386 	 */
387 	if (submit_queues > nr_cpu_ids || poll_queues > g_poll_queues)
388 		return -EINVAL;
389 
390 	/*
391 	 * Keep previous and new queue numbers in nullb_device for reference in
392 	 * the call back function null_map_queues().
393 	 */
394 	dev->prev_submit_queues = dev->submit_queues;
395 	dev->prev_poll_queues = dev->poll_queues;
396 	dev->submit_queues = submit_queues;
397 	dev->poll_queues = poll_queues;
398 
399 	set = dev->nullb->tag_set;
400 	nr_hw_queues = submit_queues + poll_queues;
401 	blk_mq_update_nr_hw_queues(set, nr_hw_queues);
402 	ret = set->nr_hw_queues == nr_hw_queues ? 0 : -ENOMEM;
403 
404 	if (ret) {
405 		/* on error, revert the queue numbers */
406 		dev->submit_queues = dev->prev_submit_queues;
407 		dev->poll_queues = dev->prev_poll_queues;
408 	}
409 
410 	return ret;
411 }
412 
413 static int nullb_apply_submit_queues(struct nullb_device *dev,
414 				     unsigned int submit_queues)
415 {
416 	int ret;
417 
418 	mutex_lock(&lock);
419 	ret = nullb_update_nr_hw_queues(dev, submit_queues, dev->poll_queues);
420 	mutex_unlock(&lock);
421 
422 	return ret;
423 }
424 
425 static int nullb_apply_poll_queues(struct nullb_device *dev,
426 				   unsigned int poll_queues)
427 {
428 	int ret;
429 
430 	mutex_lock(&lock);
431 	ret = nullb_update_nr_hw_queues(dev, dev->submit_queues, poll_queues);
432 	mutex_unlock(&lock);
433 
434 	return ret;
435 }
436 
437 NULLB_DEVICE_ATTR(size, ulong, NULL);
438 NULLB_DEVICE_ATTR(completion_nsec, ulong, NULL);
439 NULLB_DEVICE_ATTR(submit_queues, uint, nullb_apply_submit_queues);
440 NULLB_DEVICE_ATTR(poll_queues, uint, nullb_apply_poll_queues);
441 NULLB_DEVICE_ATTR(home_node, uint, NULL);
442 NULLB_DEVICE_ATTR(queue_mode, uint, NULL);
443 NULLB_DEVICE_ATTR(blocksize, uint, NULL);
444 NULLB_DEVICE_ATTR(max_sectors, uint, NULL);
445 NULLB_DEVICE_ATTR(irqmode, uint, NULL);
446 NULLB_DEVICE_ATTR(hw_queue_depth, uint, NULL);
447 NULLB_DEVICE_ATTR(index, uint, NULL);
448 NULLB_DEVICE_ATTR(blocking, bool, NULL);
449 NULLB_DEVICE_ATTR(use_per_node_hctx, bool, NULL);
450 NULLB_DEVICE_ATTR(memory_backed, bool, NULL);
451 NULLB_DEVICE_ATTR(discard, bool, NULL);
452 NULLB_DEVICE_ATTR(mbps, uint, NULL);
453 NULLB_DEVICE_ATTR(cache_size, ulong, NULL);
454 NULLB_DEVICE_ATTR(zoned, bool, NULL);
455 NULLB_DEVICE_ATTR(zone_size, ulong, NULL);
456 NULLB_DEVICE_ATTR(zone_capacity, ulong, NULL);
457 NULLB_DEVICE_ATTR(zone_nr_conv, uint, NULL);
458 NULLB_DEVICE_ATTR(zone_max_open, uint, NULL);
459 NULLB_DEVICE_ATTR(zone_max_active, uint, NULL);
460 NULLB_DEVICE_ATTR(zone_append_max_sectors, uint, NULL);
461 NULLB_DEVICE_ATTR(virt_boundary, bool, NULL);
462 NULLB_DEVICE_ATTR(no_sched, bool, NULL);
463 NULLB_DEVICE_ATTR(shared_tags, bool, NULL);
464 NULLB_DEVICE_ATTR(shared_tag_bitmap, bool, NULL);
465 NULLB_DEVICE_ATTR(fua, bool, NULL);
466 
467 static ssize_t nullb_device_power_show(struct config_item *item, char *page)
468 {
469 	return nullb_device_bool_attr_show(to_nullb_device(item)->power, page);
470 }
471 
472 static ssize_t nullb_device_power_store(struct config_item *item,
473 				     const char *page, size_t count)
474 {
475 	struct nullb_device *dev = to_nullb_device(item);
476 	bool newp = false;
477 	ssize_t ret;
478 
479 	ret = nullb_device_bool_attr_store(&newp, page, count);
480 	if (ret < 0)
481 		return ret;
482 
483 	ret = count;
484 	mutex_lock(&lock);
485 	if (!dev->power && newp) {
486 		if (test_and_set_bit(NULLB_DEV_FL_UP, &dev->flags))
487 			goto out;
488 
489 		ret = null_add_dev(dev);
490 		if (ret) {
491 			clear_bit(NULLB_DEV_FL_UP, &dev->flags);
492 			goto out;
493 		}
494 
495 		set_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
496 		dev->power = newp;
497 	} else if (dev->power && !newp) {
498 		if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
499 			dev->power = newp;
500 			null_del_dev(dev->nullb);
501 		}
502 		clear_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
503 	}
504 
505 out:
506 	mutex_unlock(&lock);
507 	return ret;
508 }
509 
510 CONFIGFS_ATTR(nullb_device_, power);
511 
512 static ssize_t nullb_device_badblocks_show(struct config_item *item, char *page)
513 {
514 	struct nullb_device *t_dev = to_nullb_device(item);
515 
516 	return badblocks_show(&t_dev->badblocks, page, 0);
517 }
518 
519 static ssize_t nullb_device_badblocks_store(struct config_item *item,
520 				     const char *page, size_t count)
521 {
522 	struct nullb_device *t_dev = to_nullb_device(item);
523 	char *orig, *buf, *tmp;
524 	u64 start, end;
525 	int ret;
526 
527 	orig = kstrndup(page, count, GFP_KERNEL);
528 	if (!orig)
529 		return -ENOMEM;
530 
531 	buf = strstrip(orig);
532 
533 	ret = -EINVAL;
534 	if (buf[0] != '+' && buf[0] != '-')
535 		goto out;
536 	tmp = strchr(&buf[1], '-');
537 	if (!tmp)
538 		goto out;
539 	*tmp = '\0';
540 	ret = kstrtoull(buf + 1, 0, &start);
541 	if (ret)
542 		goto out;
543 	ret = kstrtoull(tmp + 1, 0, &end);
544 	if (ret)
545 		goto out;
546 	ret = -EINVAL;
547 	if (start > end)
548 		goto out;
549 	/* enable badblocks */
550 	cmpxchg(&t_dev->badblocks.shift, -1, 0);
551 	if (buf[0] == '+')
552 		ret = badblocks_set(&t_dev->badblocks, start,
553 			end - start + 1, 1);
554 	else
555 		ret = badblocks_clear(&t_dev->badblocks, start,
556 			end - start + 1);
557 	if (ret == 0)
558 		ret = count;
559 out:
560 	kfree(orig);
561 	return ret;
562 }
563 CONFIGFS_ATTR(nullb_device_, badblocks);
564 
565 static ssize_t nullb_device_zone_readonly_store(struct config_item *item,
566 						const char *page, size_t count)
567 {
568 	struct nullb_device *dev = to_nullb_device(item);
569 
570 	return zone_cond_store(dev, page, count, BLK_ZONE_COND_READONLY);
571 }
572 CONFIGFS_ATTR_WO(nullb_device_, zone_readonly);
573 
574 static ssize_t nullb_device_zone_offline_store(struct config_item *item,
575 					       const char *page, size_t count)
576 {
577 	struct nullb_device *dev = to_nullb_device(item);
578 
579 	return zone_cond_store(dev, page, count, BLK_ZONE_COND_OFFLINE);
580 }
581 CONFIGFS_ATTR_WO(nullb_device_, zone_offline);
582 
583 static struct configfs_attribute *nullb_device_attrs[] = {
584 	&nullb_device_attr_size,
585 	&nullb_device_attr_completion_nsec,
586 	&nullb_device_attr_submit_queues,
587 	&nullb_device_attr_poll_queues,
588 	&nullb_device_attr_home_node,
589 	&nullb_device_attr_queue_mode,
590 	&nullb_device_attr_blocksize,
591 	&nullb_device_attr_max_sectors,
592 	&nullb_device_attr_irqmode,
593 	&nullb_device_attr_hw_queue_depth,
594 	&nullb_device_attr_index,
595 	&nullb_device_attr_blocking,
596 	&nullb_device_attr_use_per_node_hctx,
597 	&nullb_device_attr_power,
598 	&nullb_device_attr_memory_backed,
599 	&nullb_device_attr_discard,
600 	&nullb_device_attr_mbps,
601 	&nullb_device_attr_cache_size,
602 	&nullb_device_attr_badblocks,
603 	&nullb_device_attr_zoned,
604 	&nullb_device_attr_zone_size,
605 	&nullb_device_attr_zone_capacity,
606 	&nullb_device_attr_zone_nr_conv,
607 	&nullb_device_attr_zone_max_open,
608 	&nullb_device_attr_zone_max_active,
609 	&nullb_device_attr_zone_append_max_sectors,
610 	&nullb_device_attr_zone_readonly,
611 	&nullb_device_attr_zone_offline,
612 	&nullb_device_attr_virt_boundary,
613 	&nullb_device_attr_no_sched,
614 	&nullb_device_attr_shared_tags,
615 	&nullb_device_attr_shared_tag_bitmap,
616 	&nullb_device_attr_fua,
617 	NULL,
618 };
619 
620 static void nullb_device_release(struct config_item *item)
621 {
622 	struct nullb_device *dev = to_nullb_device(item);
623 
624 	null_free_device_storage(dev, false);
625 	null_free_dev(dev);
626 }
627 
628 static struct configfs_item_operations nullb_device_ops = {
629 	.release	= nullb_device_release,
630 };
631 
632 static const struct config_item_type nullb_device_type = {
633 	.ct_item_ops	= &nullb_device_ops,
634 	.ct_attrs	= nullb_device_attrs,
635 	.ct_owner	= THIS_MODULE,
636 };
637 
638 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
639 
640 static void nullb_add_fault_config(struct nullb_device *dev)
641 {
642 	fault_config_init(&dev->timeout_config, "timeout_inject");
643 	fault_config_init(&dev->requeue_config, "requeue_inject");
644 	fault_config_init(&dev->init_hctx_fault_config, "init_hctx_fault_inject");
645 
646 	configfs_add_default_group(&dev->timeout_config.group, &dev->group);
647 	configfs_add_default_group(&dev->requeue_config.group, &dev->group);
648 	configfs_add_default_group(&dev->init_hctx_fault_config.group, &dev->group);
649 }
650 
651 #else
652 
653 static void nullb_add_fault_config(struct nullb_device *dev)
654 {
655 }
656 
657 #endif
658 
659 static struct
660 config_group *nullb_group_make_group(struct config_group *group, const char *name)
661 {
662 	struct nullb_device *dev;
663 
664 	if (null_find_dev_by_name(name))
665 		return ERR_PTR(-EEXIST);
666 
667 	dev = null_alloc_dev();
668 	if (!dev)
669 		return ERR_PTR(-ENOMEM);
670 
671 	config_group_init_type_name(&dev->group, name, &nullb_device_type);
672 	nullb_add_fault_config(dev);
673 
674 	return &dev->group;
675 }
676 
677 static void
678 nullb_group_drop_item(struct config_group *group, struct config_item *item)
679 {
680 	struct nullb_device *dev = to_nullb_device(item);
681 
682 	if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
683 		mutex_lock(&lock);
684 		dev->power = false;
685 		null_del_dev(dev->nullb);
686 		mutex_unlock(&lock);
687 	}
688 
689 	config_item_put(item);
690 }
691 
692 static ssize_t memb_group_features_show(struct config_item *item, char *page)
693 {
694 	return snprintf(page, PAGE_SIZE,
695 			"badblocks,blocking,blocksize,cache_size,fua,"
696 			"completion_nsec,discard,home_node,hw_queue_depth,"
697 			"irqmode,max_sectors,mbps,memory_backed,no_sched,"
698 			"poll_queues,power,queue_mode,shared_tag_bitmap,"
699 			"shared_tags,size,submit_queues,use_per_node_hctx,"
700 			"virt_boundary,zoned,zone_capacity,zone_max_active,"
701 			"zone_max_open,zone_nr_conv,zone_offline,zone_readonly,"
702 			"zone_size,zone_append_max_sectors\n");
703 }
704 
705 CONFIGFS_ATTR_RO(memb_group_, features);
706 
707 static struct configfs_attribute *nullb_group_attrs[] = {
708 	&memb_group_attr_features,
709 	NULL,
710 };
711 
712 static struct configfs_group_operations nullb_group_ops = {
713 	.make_group	= nullb_group_make_group,
714 	.drop_item	= nullb_group_drop_item,
715 };
716 
717 static const struct config_item_type nullb_group_type = {
718 	.ct_group_ops	= &nullb_group_ops,
719 	.ct_attrs	= nullb_group_attrs,
720 	.ct_owner	= THIS_MODULE,
721 };
722 
723 static struct configfs_subsystem nullb_subsys = {
724 	.su_group = {
725 		.cg_item = {
726 			.ci_namebuf = "nullb",
727 			.ci_type = &nullb_group_type,
728 		},
729 	},
730 };
731 
732 static inline int null_cache_active(struct nullb *nullb)
733 {
734 	return test_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
735 }
736 
737 static struct nullb_device *null_alloc_dev(void)
738 {
739 	struct nullb_device *dev;
740 
741 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
742 	if (!dev)
743 		return NULL;
744 
745 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
746 	dev->timeout_config.attr = null_timeout_attr;
747 	dev->requeue_config.attr = null_requeue_attr;
748 	dev->init_hctx_fault_config.attr = null_init_hctx_attr;
749 #endif
750 
751 	INIT_RADIX_TREE(&dev->data, GFP_ATOMIC);
752 	INIT_RADIX_TREE(&dev->cache, GFP_ATOMIC);
753 	if (badblocks_init(&dev->badblocks, 0)) {
754 		kfree(dev);
755 		return NULL;
756 	}
757 
758 	dev->size = g_gb * 1024;
759 	dev->completion_nsec = g_completion_nsec;
760 	dev->submit_queues = g_submit_queues;
761 	dev->prev_submit_queues = g_submit_queues;
762 	dev->poll_queues = g_poll_queues;
763 	dev->prev_poll_queues = g_poll_queues;
764 	dev->home_node = g_home_node;
765 	dev->queue_mode = g_queue_mode;
766 	dev->blocksize = g_bs;
767 	dev->max_sectors = g_max_sectors;
768 	dev->irqmode = g_irqmode;
769 	dev->hw_queue_depth = g_hw_queue_depth;
770 	dev->blocking = g_blocking;
771 	dev->memory_backed = g_memory_backed;
772 	dev->discard = g_discard;
773 	dev->cache_size = g_cache_size;
774 	dev->mbps = g_mbps;
775 	dev->use_per_node_hctx = g_use_per_node_hctx;
776 	dev->zoned = g_zoned;
777 	dev->zone_size = g_zone_size;
778 	dev->zone_capacity = g_zone_capacity;
779 	dev->zone_nr_conv = g_zone_nr_conv;
780 	dev->zone_max_open = g_zone_max_open;
781 	dev->zone_max_active = g_zone_max_active;
782 	dev->zone_append_max_sectors = g_zone_append_max_sectors;
783 	dev->virt_boundary = g_virt_boundary;
784 	dev->no_sched = g_no_sched;
785 	dev->shared_tags = g_shared_tags;
786 	dev->shared_tag_bitmap = g_shared_tag_bitmap;
787 	dev->fua = g_fua;
788 
789 	return dev;
790 }
791 
792 static void null_free_dev(struct nullb_device *dev)
793 {
794 	if (!dev)
795 		return;
796 
797 	null_free_zoned_dev(dev);
798 	badblocks_exit(&dev->badblocks);
799 	kfree(dev);
800 }
801 
802 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
803 {
804 	struct nullb_cmd *cmd = container_of(timer, struct nullb_cmd, timer);
805 
806 	blk_mq_end_request(blk_mq_rq_from_pdu(cmd), cmd->error);
807 	return HRTIMER_NORESTART;
808 }
809 
810 static void null_cmd_end_timer(struct nullb_cmd *cmd)
811 {
812 	ktime_t kt = cmd->nq->dev->completion_nsec;
813 
814 	hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL);
815 }
816 
817 static void null_complete_rq(struct request *rq)
818 {
819 	struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq);
820 
821 	blk_mq_end_request(rq, cmd->error);
822 }
823 
824 static struct nullb_page *null_alloc_page(void)
825 {
826 	struct nullb_page *t_page;
827 
828 	t_page = kmalloc(sizeof(struct nullb_page), GFP_NOIO);
829 	if (!t_page)
830 		return NULL;
831 
832 	t_page->page = alloc_pages(GFP_NOIO, 0);
833 	if (!t_page->page) {
834 		kfree(t_page);
835 		return NULL;
836 	}
837 
838 	memset(t_page->bitmap, 0, sizeof(t_page->bitmap));
839 	return t_page;
840 }
841 
842 static void null_free_page(struct nullb_page *t_page)
843 {
844 	__set_bit(NULLB_PAGE_FREE, t_page->bitmap);
845 	if (test_bit(NULLB_PAGE_LOCK, t_page->bitmap))
846 		return;
847 	__free_page(t_page->page);
848 	kfree(t_page);
849 }
850 
851 static bool null_page_empty(struct nullb_page *page)
852 {
853 	int size = MAP_SZ - 2;
854 
855 	return find_first_bit(page->bitmap, size) == size;
856 }
857 
858 static void null_free_sector(struct nullb *nullb, sector_t sector,
859 	bool is_cache)
860 {
861 	unsigned int sector_bit;
862 	u64 idx;
863 	struct nullb_page *t_page, *ret;
864 	struct radix_tree_root *root;
865 
866 	root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
867 	idx = sector >> PAGE_SECTORS_SHIFT;
868 	sector_bit = (sector & SECTOR_MASK);
869 
870 	t_page = radix_tree_lookup(root, idx);
871 	if (t_page) {
872 		__clear_bit(sector_bit, t_page->bitmap);
873 
874 		if (null_page_empty(t_page)) {
875 			ret = radix_tree_delete_item(root, idx, t_page);
876 			WARN_ON(ret != t_page);
877 			null_free_page(ret);
878 			if (is_cache)
879 				nullb->dev->curr_cache -= PAGE_SIZE;
880 		}
881 	}
882 }
883 
884 static struct nullb_page *null_radix_tree_insert(struct nullb *nullb, u64 idx,
885 	struct nullb_page *t_page, bool is_cache)
886 {
887 	struct radix_tree_root *root;
888 
889 	root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
890 
891 	if (radix_tree_insert(root, idx, t_page)) {
892 		null_free_page(t_page);
893 		t_page = radix_tree_lookup(root, idx);
894 		WARN_ON(!t_page || t_page->page->index != idx);
895 	} else if (is_cache)
896 		nullb->dev->curr_cache += PAGE_SIZE;
897 
898 	return t_page;
899 }
900 
901 static void null_free_device_storage(struct nullb_device *dev, bool is_cache)
902 {
903 	unsigned long pos = 0;
904 	int nr_pages;
905 	struct nullb_page *ret, *t_pages[FREE_BATCH];
906 	struct radix_tree_root *root;
907 
908 	root = is_cache ? &dev->cache : &dev->data;
909 
910 	do {
911 		int i;
912 
913 		nr_pages = radix_tree_gang_lookup(root,
914 				(void **)t_pages, pos, FREE_BATCH);
915 
916 		for (i = 0; i < nr_pages; i++) {
917 			pos = t_pages[i]->page->index;
918 			ret = radix_tree_delete_item(root, pos, t_pages[i]);
919 			WARN_ON(ret != t_pages[i]);
920 			null_free_page(ret);
921 		}
922 
923 		pos++;
924 	} while (nr_pages == FREE_BATCH);
925 
926 	if (is_cache)
927 		dev->curr_cache = 0;
928 }
929 
930 static struct nullb_page *__null_lookup_page(struct nullb *nullb,
931 	sector_t sector, bool for_write, bool is_cache)
932 {
933 	unsigned int sector_bit;
934 	u64 idx;
935 	struct nullb_page *t_page;
936 	struct radix_tree_root *root;
937 
938 	idx = sector >> PAGE_SECTORS_SHIFT;
939 	sector_bit = (sector & SECTOR_MASK);
940 
941 	root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
942 	t_page = radix_tree_lookup(root, idx);
943 	WARN_ON(t_page && t_page->page->index != idx);
944 
945 	if (t_page && (for_write || test_bit(sector_bit, t_page->bitmap)))
946 		return t_page;
947 
948 	return NULL;
949 }
950 
951 static struct nullb_page *null_lookup_page(struct nullb *nullb,
952 	sector_t sector, bool for_write, bool ignore_cache)
953 {
954 	struct nullb_page *page = NULL;
955 
956 	if (!ignore_cache)
957 		page = __null_lookup_page(nullb, sector, for_write, true);
958 	if (page)
959 		return page;
960 	return __null_lookup_page(nullb, sector, for_write, false);
961 }
962 
963 static struct nullb_page *null_insert_page(struct nullb *nullb,
964 					   sector_t sector, bool ignore_cache)
965 	__releases(&nullb->lock)
966 	__acquires(&nullb->lock)
967 {
968 	u64 idx;
969 	struct nullb_page *t_page;
970 
971 	t_page = null_lookup_page(nullb, sector, true, ignore_cache);
972 	if (t_page)
973 		return t_page;
974 
975 	spin_unlock_irq(&nullb->lock);
976 
977 	t_page = null_alloc_page();
978 	if (!t_page)
979 		goto out_lock;
980 
981 	if (radix_tree_preload(GFP_NOIO))
982 		goto out_freepage;
983 
984 	spin_lock_irq(&nullb->lock);
985 	idx = sector >> PAGE_SECTORS_SHIFT;
986 	t_page->page->index = idx;
987 	t_page = null_radix_tree_insert(nullb, idx, t_page, !ignore_cache);
988 	radix_tree_preload_end();
989 
990 	return t_page;
991 out_freepage:
992 	null_free_page(t_page);
993 out_lock:
994 	spin_lock_irq(&nullb->lock);
995 	return null_lookup_page(nullb, sector, true, ignore_cache);
996 }
997 
998 static int null_flush_cache_page(struct nullb *nullb, struct nullb_page *c_page)
999 {
1000 	int i;
1001 	unsigned int offset;
1002 	u64 idx;
1003 	struct nullb_page *t_page, *ret;
1004 	void *dst, *src;
1005 
1006 	idx = c_page->page->index;
1007 
1008 	t_page = null_insert_page(nullb, idx << PAGE_SECTORS_SHIFT, true);
1009 
1010 	__clear_bit(NULLB_PAGE_LOCK, c_page->bitmap);
1011 	if (test_bit(NULLB_PAGE_FREE, c_page->bitmap)) {
1012 		null_free_page(c_page);
1013 		if (t_page && null_page_empty(t_page)) {
1014 			ret = radix_tree_delete_item(&nullb->dev->data,
1015 				idx, t_page);
1016 			null_free_page(t_page);
1017 		}
1018 		return 0;
1019 	}
1020 
1021 	if (!t_page)
1022 		return -ENOMEM;
1023 
1024 	src = kmap_local_page(c_page->page);
1025 	dst = kmap_local_page(t_page->page);
1026 
1027 	for (i = 0; i < PAGE_SECTORS;
1028 			i += (nullb->dev->blocksize >> SECTOR_SHIFT)) {
1029 		if (test_bit(i, c_page->bitmap)) {
1030 			offset = (i << SECTOR_SHIFT);
1031 			memcpy(dst + offset, src + offset,
1032 				nullb->dev->blocksize);
1033 			__set_bit(i, t_page->bitmap);
1034 		}
1035 	}
1036 
1037 	kunmap_local(dst);
1038 	kunmap_local(src);
1039 
1040 	ret = radix_tree_delete_item(&nullb->dev->cache, idx, c_page);
1041 	null_free_page(ret);
1042 	nullb->dev->curr_cache -= PAGE_SIZE;
1043 
1044 	return 0;
1045 }
1046 
1047 static int null_make_cache_space(struct nullb *nullb, unsigned long n)
1048 {
1049 	int i, err, nr_pages;
1050 	struct nullb_page *c_pages[FREE_BATCH];
1051 	unsigned long flushed = 0, one_round;
1052 
1053 again:
1054 	if ((nullb->dev->cache_size * 1024 * 1024) >
1055 	     nullb->dev->curr_cache + n || nullb->dev->curr_cache == 0)
1056 		return 0;
1057 
1058 	nr_pages = radix_tree_gang_lookup(&nullb->dev->cache,
1059 			(void **)c_pages, nullb->cache_flush_pos, FREE_BATCH);
1060 	/*
1061 	 * nullb_flush_cache_page could unlock before using the c_pages. To
1062 	 * avoid race, we don't allow page free
1063 	 */
1064 	for (i = 0; i < nr_pages; i++) {
1065 		nullb->cache_flush_pos = c_pages[i]->page->index;
1066 		/*
1067 		 * We found the page which is being flushed to disk by other
1068 		 * threads
1069 		 */
1070 		if (test_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap))
1071 			c_pages[i] = NULL;
1072 		else
1073 			__set_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap);
1074 	}
1075 
1076 	one_round = 0;
1077 	for (i = 0; i < nr_pages; i++) {
1078 		if (c_pages[i] == NULL)
1079 			continue;
1080 		err = null_flush_cache_page(nullb, c_pages[i]);
1081 		if (err)
1082 			return err;
1083 		one_round++;
1084 	}
1085 	flushed += one_round << PAGE_SHIFT;
1086 
1087 	if (n > flushed) {
1088 		if (nr_pages == 0)
1089 			nullb->cache_flush_pos = 0;
1090 		if (one_round == 0) {
1091 			/* give other threads a chance */
1092 			spin_unlock_irq(&nullb->lock);
1093 			spin_lock_irq(&nullb->lock);
1094 		}
1095 		goto again;
1096 	}
1097 	return 0;
1098 }
1099 
1100 static int copy_to_nullb(struct nullb *nullb, struct page *source,
1101 	unsigned int off, sector_t sector, size_t n, bool is_fua)
1102 {
1103 	size_t temp, count = 0;
1104 	unsigned int offset;
1105 	struct nullb_page *t_page;
1106 
1107 	while (count < n) {
1108 		temp = min_t(size_t, nullb->dev->blocksize, n - count);
1109 
1110 		if (null_cache_active(nullb) && !is_fua)
1111 			null_make_cache_space(nullb, PAGE_SIZE);
1112 
1113 		offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
1114 		t_page = null_insert_page(nullb, sector,
1115 			!null_cache_active(nullb) || is_fua);
1116 		if (!t_page)
1117 			return -ENOSPC;
1118 
1119 		memcpy_page(t_page->page, offset, source, off + count, temp);
1120 
1121 		__set_bit(sector & SECTOR_MASK, t_page->bitmap);
1122 
1123 		if (is_fua)
1124 			null_free_sector(nullb, sector, true);
1125 
1126 		count += temp;
1127 		sector += temp >> SECTOR_SHIFT;
1128 	}
1129 	return 0;
1130 }
1131 
1132 static int copy_from_nullb(struct nullb *nullb, struct page *dest,
1133 	unsigned int off, sector_t sector, size_t n)
1134 {
1135 	size_t temp, count = 0;
1136 	unsigned int offset;
1137 	struct nullb_page *t_page;
1138 
1139 	while (count < n) {
1140 		temp = min_t(size_t, nullb->dev->blocksize, n - count);
1141 
1142 		offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
1143 		t_page = null_lookup_page(nullb, sector, false,
1144 			!null_cache_active(nullb));
1145 
1146 		if (t_page)
1147 			memcpy_page(dest, off + count, t_page->page, offset,
1148 				    temp);
1149 		else
1150 			zero_user(dest, off + count, temp);
1151 
1152 		count += temp;
1153 		sector += temp >> SECTOR_SHIFT;
1154 	}
1155 	return 0;
1156 }
1157 
1158 static void nullb_fill_pattern(struct nullb *nullb, struct page *page,
1159 			       unsigned int len, unsigned int off)
1160 {
1161 	memset_page(page, off, 0xff, len);
1162 }
1163 
1164 blk_status_t null_handle_discard(struct nullb_device *dev,
1165 				 sector_t sector, sector_t nr_sectors)
1166 {
1167 	struct nullb *nullb = dev->nullb;
1168 	size_t n = nr_sectors << SECTOR_SHIFT;
1169 	size_t temp;
1170 
1171 	spin_lock_irq(&nullb->lock);
1172 	while (n > 0) {
1173 		temp = min_t(size_t, n, dev->blocksize);
1174 		null_free_sector(nullb, sector, false);
1175 		if (null_cache_active(nullb))
1176 			null_free_sector(nullb, sector, true);
1177 		sector += temp >> SECTOR_SHIFT;
1178 		n -= temp;
1179 	}
1180 	spin_unlock_irq(&nullb->lock);
1181 
1182 	return BLK_STS_OK;
1183 }
1184 
1185 static blk_status_t null_handle_flush(struct nullb *nullb)
1186 {
1187 	int err;
1188 
1189 	if (!null_cache_active(nullb))
1190 		return 0;
1191 
1192 	spin_lock_irq(&nullb->lock);
1193 	while (true) {
1194 		err = null_make_cache_space(nullb,
1195 			nullb->dev->cache_size * 1024 * 1024);
1196 		if (err || nullb->dev->curr_cache == 0)
1197 			break;
1198 	}
1199 
1200 	WARN_ON(!radix_tree_empty(&nullb->dev->cache));
1201 	spin_unlock_irq(&nullb->lock);
1202 	return errno_to_blk_status(err);
1203 }
1204 
1205 static int null_transfer(struct nullb *nullb, struct page *page,
1206 	unsigned int len, unsigned int off, bool is_write, sector_t sector,
1207 	bool is_fua)
1208 {
1209 	struct nullb_device *dev = nullb->dev;
1210 	unsigned int valid_len = len;
1211 	int err = 0;
1212 
1213 	if (!is_write) {
1214 		if (dev->zoned)
1215 			valid_len = null_zone_valid_read_len(nullb,
1216 				sector, len);
1217 
1218 		if (valid_len) {
1219 			err = copy_from_nullb(nullb, page, off,
1220 				sector, valid_len);
1221 			off += valid_len;
1222 			len -= valid_len;
1223 		}
1224 
1225 		if (len)
1226 			nullb_fill_pattern(nullb, page, len, off);
1227 		flush_dcache_page(page);
1228 	} else {
1229 		flush_dcache_page(page);
1230 		err = copy_to_nullb(nullb, page, off, sector, len, is_fua);
1231 	}
1232 
1233 	return err;
1234 }
1235 
1236 static blk_status_t null_handle_rq(struct nullb_cmd *cmd)
1237 {
1238 	struct request *rq = blk_mq_rq_from_pdu(cmd);
1239 	struct nullb *nullb = cmd->nq->dev->nullb;
1240 	int err = 0;
1241 	unsigned int len;
1242 	sector_t sector = blk_rq_pos(rq);
1243 	struct req_iterator iter;
1244 	struct bio_vec bvec;
1245 
1246 	spin_lock_irq(&nullb->lock);
1247 	rq_for_each_segment(bvec, rq, iter) {
1248 		len = bvec.bv_len;
1249 		err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1250 				     op_is_write(req_op(rq)), sector,
1251 				     rq->cmd_flags & REQ_FUA);
1252 		if (err)
1253 			break;
1254 		sector += len >> SECTOR_SHIFT;
1255 	}
1256 	spin_unlock_irq(&nullb->lock);
1257 
1258 	return errno_to_blk_status(err);
1259 }
1260 
1261 static inline blk_status_t null_handle_throttled(struct nullb_cmd *cmd)
1262 {
1263 	struct nullb_device *dev = cmd->nq->dev;
1264 	struct nullb *nullb = dev->nullb;
1265 	blk_status_t sts = BLK_STS_OK;
1266 	struct request *rq = blk_mq_rq_from_pdu(cmd);
1267 
1268 	if (!hrtimer_active(&nullb->bw_timer))
1269 		hrtimer_restart(&nullb->bw_timer);
1270 
1271 	if (atomic_long_sub_return(blk_rq_bytes(rq), &nullb->cur_bytes) < 0) {
1272 		blk_mq_stop_hw_queues(nullb->q);
1273 		/* race with timer */
1274 		if (atomic_long_read(&nullb->cur_bytes) > 0)
1275 			blk_mq_start_stopped_hw_queues(nullb->q, true);
1276 		/* requeue request */
1277 		sts = BLK_STS_DEV_RESOURCE;
1278 	}
1279 	return sts;
1280 }
1281 
1282 static inline blk_status_t null_handle_badblocks(struct nullb_cmd *cmd,
1283 						 sector_t sector,
1284 						 sector_t nr_sectors)
1285 {
1286 	struct badblocks *bb = &cmd->nq->dev->badblocks;
1287 	sector_t first_bad;
1288 	int bad_sectors;
1289 
1290 	if (badblocks_check(bb, sector, nr_sectors, &first_bad, &bad_sectors))
1291 		return BLK_STS_IOERR;
1292 
1293 	return BLK_STS_OK;
1294 }
1295 
1296 static inline blk_status_t null_handle_memory_backed(struct nullb_cmd *cmd,
1297 						     enum req_op op,
1298 						     sector_t sector,
1299 						     sector_t nr_sectors)
1300 {
1301 	struct nullb_device *dev = cmd->nq->dev;
1302 
1303 	if (op == REQ_OP_DISCARD)
1304 		return null_handle_discard(dev, sector, nr_sectors);
1305 
1306 	return null_handle_rq(cmd);
1307 }
1308 
1309 static void nullb_zero_read_cmd_buffer(struct nullb_cmd *cmd)
1310 {
1311 	struct request *rq = blk_mq_rq_from_pdu(cmd);
1312 	struct nullb_device *dev = cmd->nq->dev;
1313 	struct bio *bio;
1314 
1315 	if (!dev->memory_backed && req_op(rq) == REQ_OP_READ) {
1316 		__rq_for_each_bio(bio, rq)
1317 			zero_fill_bio(bio);
1318 	}
1319 }
1320 
1321 static inline void nullb_complete_cmd(struct nullb_cmd *cmd)
1322 {
1323 	struct request *rq = blk_mq_rq_from_pdu(cmd);
1324 
1325 	/*
1326 	 * Since root privileges are required to configure the null_blk
1327 	 * driver, it is fine that this driver does not initialize the
1328 	 * data buffers of read commands. Zero-initialize these buffers
1329 	 * anyway if KMSAN is enabled to prevent that KMSAN complains
1330 	 * about null_blk not initializing read data buffers.
1331 	 */
1332 	if (IS_ENABLED(CONFIG_KMSAN))
1333 		nullb_zero_read_cmd_buffer(cmd);
1334 
1335 	/* Complete IO by inline, softirq or timer */
1336 	switch (cmd->nq->dev->irqmode) {
1337 	case NULL_IRQ_SOFTIRQ:
1338 		blk_mq_complete_request(rq);
1339 		break;
1340 	case NULL_IRQ_NONE:
1341 		blk_mq_end_request(rq, cmd->error);
1342 		break;
1343 	case NULL_IRQ_TIMER:
1344 		null_cmd_end_timer(cmd);
1345 		break;
1346 	}
1347 }
1348 
1349 blk_status_t null_process_cmd(struct nullb_cmd *cmd, enum req_op op,
1350 			      sector_t sector, unsigned int nr_sectors)
1351 {
1352 	struct nullb_device *dev = cmd->nq->dev;
1353 	blk_status_t ret;
1354 
1355 	if (dev->badblocks.shift != -1) {
1356 		ret = null_handle_badblocks(cmd, sector, nr_sectors);
1357 		if (ret != BLK_STS_OK)
1358 			return ret;
1359 	}
1360 
1361 	if (dev->memory_backed)
1362 		return null_handle_memory_backed(cmd, op, sector, nr_sectors);
1363 
1364 	return BLK_STS_OK;
1365 }
1366 
1367 static void null_handle_cmd(struct nullb_cmd *cmd, sector_t sector,
1368 			    sector_t nr_sectors, enum req_op op)
1369 {
1370 	struct nullb_device *dev = cmd->nq->dev;
1371 	struct nullb *nullb = dev->nullb;
1372 	blk_status_t sts;
1373 
1374 	if (op == REQ_OP_FLUSH) {
1375 		cmd->error = null_handle_flush(nullb);
1376 		goto out;
1377 	}
1378 
1379 	if (dev->zoned)
1380 		sts = null_process_zoned_cmd(cmd, op, sector, nr_sectors);
1381 	else
1382 		sts = null_process_cmd(cmd, op, sector, nr_sectors);
1383 
1384 	/* Do not overwrite errors (e.g. timeout errors) */
1385 	if (cmd->error == BLK_STS_OK)
1386 		cmd->error = sts;
1387 
1388 out:
1389 	nullb_complete_cmd(cmd);
1390 }
1391 
1392 static enum hrtimer_restart nullb_bwtimer_fn(struct hrtimer *timer)
1393 {
1394 	struct nullb *nullb = container_of(timer, struct nullb, bw_timer);
1395 	ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1396 	unsigned int mbps = nullb->dev->mbps;
1397 
1398 	if (atomic_long_read(&nullb->cur_bytes) == mb_per_tick(mbps))
1399 		return HRTIMER_NORESTART;
1400 
1401 	atomic_long_set(&nullb->cur_bytes, mb_per_tick(mbps));
1402 	blk_mq_start_stopped_hw_queues(nullb->q, true);
1403 
1404 	hrtimer_forward_now(&nullb->bw_timer, timer_interval);
1405 
1406 	return HRTIMER_RESTART;
1407 }
1408 
1409 static void nullb_setup_bwtimer(struct nullb *nullb)
1410 {
1411 	ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1412 
1413 	hrtimer_init(&nullb->bw_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1414 	nullb->bw_timer.function = nullb_bwtimer_fn;
1415 	atomic_long_set(&nullb->cur_bytes, mb_per_tick(nullb->dev->mbps));
1416 	hrtimer_start(&nullb->bw_timer, timer_interval, HRTIMER_MODE_REL);
1417 }
1418 
1419 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1420 
1421 static bool should_timeout_request(struct request *rq)
1422 {
1423 	struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq);
1424 	struct nullb_device *dev = cmd->nq->dev;
1425 
1426 	return should_fail(&dev->timeout_config.attr, 1);
1427 }
1428 
1429 static bool should_requeue_request(struct request *rq)
1430 {
1431 	struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq);
1432 	struct nullb_device *dev = cmd->nq->dev;
1433 
1434 	return should_fail(&dev->requeue_config.attr, 1);
1435 }
1436 
1437 static bool should_init_hctx_fail(struct nullb_device *dev)
1438 {
1439 	return should_fail(&dev->init_hctx_fault_config.attr, 1);
1440 }
1441 
1442 #else
1443 
1444 static bool should_timeout_request(struct request *rq)
1445 {
1446 	return false;
1447 }
1448 
1449 static bool should_requeue_request(struct request *rq)
1450 {
1451 	return false;
1452 }
1453 
1454 static bool should_init_hctx_fail(struct nullb_device *dev)
1455 {
1456 	return false;
1457 }
1458 
1459 #endif
1460 
1461 static void null_map_queues(struct blk_mq_tag_set *set)
1462 {
1463 	struct nullb *nullb = set->driver_data;
1464 	int i, qoff;
1465 	unsigned int submit_queues = g_submit_queues;
1466 	unsigned int poll_queues = g_poll_queues;
1467 
1468 	if (nullb) {
1469 		struct nullb_device *dev = nullb->dev;
1470 
1471 		/*
1472 		 * Refer nr_hw_queues of the tag set to check if the expected
1473 		 * number of hardware queues are prepared. If block layer failed
1474 		 * to prepare them, use previous numbers of submit queues and
1475 		 * poll queues to map queues.
1476 		 */
1477 		if (set->nr_hw_queues ==
1478 		    dev->submit_queues + dev->poll_queues) {
1479 			submit_queues = dev->submit_queues;
1480 			poll_queues = dev->poll_queues;
1481 		} else if (set->nr_hw_queues ==
1482 			   dev->prev_submit_queues + dev->prev_poll_queues) {
1483 			submit_queues = dev->prev_submit_queues;
1484 			poll_queues = dev->prev_poll_queues;
1485 		} else {
1486 			pr_warn("tag set has unexpected nr_hw_queues: %d\n",
1487 				set->nr_hw_queues);
1488 			WARN_ON_ONCE(true);
1489 			submit_queues = 1;
1490 			poll_queues = 0;
1491 		}
1492 	}
1493 
1494 	for (i = 0, qoff = 0; i < set->nr_maps; i++) {
1495 		struct blk_mq_queue_map *map = &set->map[i];
1496 
1497 		switch (i) {
1498 		case HCTX_TYPE_DEFAULT:
1499 			map->nr_queues = submit_queues;
1500 			break;
1501 		case HCTX_TYPE_READ:
1502 			map->nr_queues = 0;
1503 			continue;
1504 		case HCTX_TYPE_POLL:
1505 			map->nr_queues = poll_queues;
1506 			break;
1507 		}
1508 		map->queue_offset = qoff;
1509 		qoff += map->nr_queues;
1510 		blk_mq_map_queues(map);
1511 	}
1512 }
1513 
1514 static int null_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1515 {
1516 	struct nullb_queue *nq = hctx->driver_data;
1517 	LIST_HEAD(list);
1518 	int nr = 0;
1519 	struct request *rq;
1520 
1521 	spin_lock(&nq->poll_lock);
1522 	list_splice_init(&nq->poll_list, &list);
1523 	list_for_each_entry(rq, &list, queuelist)
1524 		blk_mq_set_request_complete(rq);
1525 	spin_unlock(&nq->poll_lock);
1526 
1527 	while (!list_empty(&list)) {
1528 		struct nullb_cmd *cmd;
1529 		struct request *req;
1530 
1531 		req = list_first_entry(&list, struct request, queuelist);
1532 		list_del_init(&req->queuelist);
1533 		cmd = blk_mq_rq_to_pdu(req);
1534 		cmd->error = null_process_cmd(cmd, req_op(req), blk_rq_pos(req),
1535 						blk_rq_sectors(req));
1536 		if (!blk_mq_add_to_batch(req, iob, (__force int) cmd->error,
1537 					blk_mq_end_request_batch))
1538 			blk_mq_end_request(req, cmd->error);
1539 		nr++;
1540 	}
1541 
1542 	return nr;
1543 }
1544 
1545 static enum blk_eh_timer_return null_timeout_rq(struct request *rq)
1546 {
1547 	struct blk_mq_hw_ctx *hctx = rq->mq_hctx;
1548 	struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq);
1549 
1550 	if (hctx->type == HCTX_TYPE_POLL) {
1551 		struct nullb_queue *nq = hctx->driver_data;
1552 
1553 		spin_lock(&nq->poll_lock);
1554 		/* The request may have completed meanwhile. */
1555 		if (blk_mq_request_completed(rq)) {
1556 			spin_unlock(&nq->poll_lock);
1557 			return BLK_EH_DONE;
1558 		}
1559 		list_del_init(&rq->queuelist);
1560 		spin_unlock(&nq->poll_lock);
1561 	}
1562 
1563 	pr_info("rq %p timed out\n", rq);
1564 
1565 	/*
1566 	 * If the device is marked as blocking (i.e. memory backed or zoned
1567 	 * device), the submission path may be blocked waiting for resources
1568 	 * and cause real timeouts. For these real timeouts, the submission
1569 	 * path will complete the request using blk_mq_complete_request().
1570 	 * Only fake timeouts need to execute blk_mq_complete_request() here.
1571 	 */
1572 	cmd->error = BLK_STS_TIMEOUT;
1573 	if (cmd->fake_timeout || hctx->type == HCTX_TYPE_POLL)
1574 		blk_mq_complete_request(rq);
1575 	return BLK_EH_DONE;
1576 }
1577 
1578 static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx,
1579 				  const struct blk_mq_queue_data *bd)
1580 {
1581 	struct request *rq = bd->rq;
1582 	struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq);
1583 	struct nullb_queue *nq = hctx->driver_data;
1584 	sector_t nr_sectors = blk_rq_sectors(rq);
1585 	sector_t sector = blk_rq_pos(rq);
1586 	const bool is_poll = hctx->type == HCTX_TYPE_POLL;
1587 
1588 	might_sleep_if(hctx->flags & BLK_MQ_F_BLOCKING);
1589 
1590 	if (!is_poll && nq->dev->irqmode == NULL_IRQ_TIMER) {
1591 		hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1592 		cmd->timer.function = null_cmd_timer_expired;
1593 	}
1594 	cmd->error = BLK_STS_OK;
1595 	cmd->nq = nq;
1596 	cmd->fake_timeout = should_timeout_request(rq) ||
1597 		blk_should_fake_timeout(rq->q);
1598 
1599 	if (should_requeue_request(rq)) {
1600 		/*
1601 		 * Alternate between hitting the core BUSY path, and the
1602 		 * driver driven requeue path
1603 		 */
1604 		nq->requeue_selection++;
1605 		if (nq->requeue_selection & 1)
1606 			return BLK_STS_RESOURCE;
1607 		blk_mq_requeue_request(rq, true);
1608 		return BLK_STS_OK;
1609 	}
1610 
1611 	if (test_bit(NULLB_DEV_FL_THROTTLED, &nq->dev->flags)) {
1612 		blk_status_t sts = null_handle_throttled(cmd);
1613 
1614 		if (sts != BLK_STS_OK)
1615 			return sts;
1616 	}
1617 
1618 	blk_mq_start_request(rq);
1619 
1620 	if (is_poll) {
1621 		spin_lock(&nq->poll_lock);
1622 		list_add_tail(&rq->queuelist, &nq->poll_list);
1623 		spin_unlock(&nq->poll_lock);
1624 		return BLK_STS_OK;
1625 	}
1626 	if (cmd->fake_timeout)
1627 		return BLK_STS_OK;
1628 
1629 	null_handle_cmd(cmd, sector, nr_sectors, req_op(rq));
1630 	return BLK_STS_OK;
1631 }
1632 
1633 static void null_queue_rqs(struct request **rqlist)
1634 {
1635 	struct request *requeue_list = NULL;
1636 	struct request **requeue_lastp = &requeue_list;
1637 	struct blk_mq_queue_data bd = { };
1638 	blk_status_t ret;
1639 
1640 	do {
1641 		struct request *rq = rq_list_pop(rqlist);
1642 
1643 		bd.rq = rq;
1644 		ret = null_queue_rq(rq->mq_hctx, &bd);
1645 		if (ret != BLK_STS_OK)
1646 			rq_list_add_tail(&requeue_lastp, rq);
1647 	} while (!rq_list_empty(*rqlist));
1648 
1649 	*rqlist = requeue_list;
1650 }
1651 
1652 static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq)
1653 {
1654 	nq->dev = nullb->dev;
1655 	INIT_LIST_HEAD(&nq->poll_list);
1656 	spin_lock_init(&nq->poll_lock);
1657 }
1658 
1659 static int null_init_hctx(struct blk_mq_hw_ctx *hctx, void *driver_data,
1660 			  unsigned int hctx_idx)
1661 {
1662 	struct nullb *nullb = hctx->queue->queuedata;
1663 	struct nullb_queue *nq;
1664 
1665 	if (should_init_hctx_fail(nullb->dev))
1666 		return -EFAULT;
1667 
1668 	nq = &nullb->queues[hctx_idx];
1669 	hctx->driver_data = nq;
1670 	null_init_queue(nullb, nq);
1671 
1672 	return 0;
1673 }
1674 
1675 static const struct blk_mq_ops null_mq_ops = {
1676 	.queue_rq       = null_queue_rq,
1677 	.queue_rqs	= null_queue_rqs,
1678 	.complete	= null_complete_rq,
1679 	.timeout	= null_timeout_rq,
1680 	.poll		= null_poll,
1681 	.map_queues	= null_map_queues,
1682 	.init_hctx	= null_init_hctx,
1683 };
1684 
1685 static void null_del_dev(struct nullb *nullb)
1686 {
1687 	struct nullb_device *dev;
1688 
1689 	if (!nullb)
1690 		return;
1691 
1692 	dev = nullb->dev;
1693 
1694 	ida_free(&nullb_indexes, nullb->index);
1695 
1696 	list_del_init(&nullb->list);
1697 
1698 	del_gendisk(nullb->disk);
1699 
1700 	if (test_bit(NULLB_DEV_FL_THROTTLED, &nullb->dev->flags)) {
1701 		hrtimer_cancel(&nullb->bw_timer);
1702 		atomic_long_set(&nullb->cur_bytes, LONG_MAX);
1703 		blk_mq_start_stopped_hw_queues(nullb->q, true);
1704 	}
1705 
1706 	put_disk(nullb->disk);
1707 	if (nullb->tag_set == &nullb->__tag_set)
1708 		blk_mq_free_tag_set(nullb->tag_set);
1709 	kfree(nullb->queues);
1710 	if (null_cache_active(nullb))
1711 		null_free_device_storage(nullb->dev, true);
1712 	kfree(nullb);
1713 	dev->nullb = NULL;
1714 }
1715 
1716 static void null_config_discard(struct nullb *nullb, struct queue_limits *lim)
1717 {
1718 	if (nullb->dev->discard == false)
1719 		return;
1720 
1721 	if (!nullb->dev->memory_backed) {
1722 		nullb->dev->discard = false;
1723 		pr_info("discard option is ignored without memory backing\n");
1724 		return;
1725 	}
1726 
1727 	if (nullb->dev->zoned) {
1728 		nullb->dev->discard = false;
1729 		pr_info("discard option is ignored in zoned mode\n");
1730 		return;
1731 	}
1732 
1733 	lim->max_hw_discard_sectors = UINT_MAX >> 9;
1734 }
1735 
1736 static const struct block_device_operations null_ops = {
1737 	.owner		= THIS_MODULE,
1738 	.report_zones	= null_report_zones,
1739 };
1740 
1741 static int setup_queues(struct nullb *nullb)
1742 {
1743 	int nqueues = nr_cpu_ids;
1744 
1745 	if (g_poll_queues)
1746 		nqueues += g_poll_queues;
1747 
1748 	nullb->queues = kcalloc(nqueues, sizeof(struct nullb_queue),
1749 				GFP_KERNEL);
1750 	if (!nullb->queues)
1751 		return -ENOMEM;
1752 
1753 	return 0;
1754 }
1755 
1756 static int null_init_tag_set(struct blk_mq_tag_set *set, int poll_queues)
1757 {
1758 	set->ops = &null_mq_ops;
1759 	set->cmd_size = sizeof(struct nullb_cmd);
1760 	set->timeout = 5 * HZ;
1761 	set->nr_maps = 1;
1762 	if (poll_queues) {
1763 		set->nr_hw_queues += poll_queues;
1764 		set->nr_maps += 2;
1765 	}
1766 	return blk_mq_alloc_tag_set(set);
1767 }
1768 
1769 static int null_init_global_tag_set(void)
1770 {
1771 	int error;
1772 
1773 	if (tag_set.ops)
1774 		return 0;
1775 
1776 	tag_set.nr_hw_queues = g_submit_queues;
1777 	tag_set.queue_depth = g_hw_queue_depth;
1778 	tag_set.numa_node = g_home_node;
1779 	tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1780 	if (g_no_sched)
1781 		tag_set.flags |= BLK_MQ_F_NO_SCHED;
1782 	if (g_shared_tag_bitmap)
1783 		tag_set.flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1784 	if (g_blocking)
1785 		tag_set.flags |= BLK_MQ_F_BLOCKING;
1786 
1787 	error = null_init_tag_set(&tag_set, g_poll_queues);
1788 	if (error)
1789 		tag_set.ops = NULL;
1790 	return error;
1791 }
1792 
1793 static int null_setup_tagset(struct nullb *nullb)
1794 {
1795 	if (nullb->dev->shared_tags) {
1796 		nullb->tag_set = &tag_set;
1797 		return null_init_global_tag_set();
1798 	}
1799 
1800 	nullb->tag_set = &nullb->__tag_set;
1801 	nullb->tag_set->driver_data = nullb;
1802 	nullb->tag_set->nr_hw_queues = nullb->dev->submit_queues;
1803 	nullb->tag_set->queue_depth = nullb->dev->hw_queue_depth;
1804 	nullb->tag_set->numa_node = nullb->dev->home_node;
1805 	nullb->tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
1806 	if (nullb->dev->no_sched)
1807 		nullb->tag_set->flags |= BLK_MQ_F_NO_SCHED;
1808 	if (nullb->dev->shared_tag_bitmap)
1809 		nullb->tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1810 	if (nullb->dev->blocking)
1811 		nullb->tag_set->flags |= BLK_MQ_F_BLOCKING;
1812 	return null_init_tag_set(nullb->tag_set, nullb->dev->poll_queues);
1813 }
1814 
1815 static int null_validate_conf(struct nullb_device *dev)
1816 {
1817 	if (dev->queue_mode == NULL_Q_RQ) {
1818 		pr_err("legacy IO path is no longer available\n");
1819 		return -EINVAL;
1820 	}
1821 	if (dev->queue_mode == NULL_Q_BIO) {
1822 		pr_err("BIO-based IO path is no longer available, using blk-mq instead.\n");
1823 		dev->queue_mode = NULL_Q_MQ;
1824 	}
1825 
1826 	dev->blocksize = round_down(dev->blocksize, 512);
1827 	dev->blocksize = clamp_t(unsigned int, dev->blocksize, 512, 4096);
1828 
1829 	if (dev->use_per_node_hctx) {
1830 		if (dev->submit_queues != nr_online_nodes)
1831 			dev->submit_queues = nr_online_nodes;
1832 	} else if (dev->submit_queues > nr_cpu_ids)
1833 		dev->submit_queues = nr_cpu_ids;
1834 	else if (dev->submit_queues == 0)
1835 		dev->submit_queues = 1;
1836 	dev->prev_submit_queues = dev->submit_queues;
1837 
1838 	if (dev->poll_queues > g_poll_queues)
1839 		dev->poll_queues = g_poll_queues;
1840 	dev->prev_poll_queues = dev->poll_queues;
1841 	dev->irqmode = min_t(unsigned int, dev->irqmode, NULL_IRQ_TIMER);
1842 
1843 	/* Do memory allocation, so set blocking */
1844 	if (dev->memory_backed)
1845 		dev->blocking = true;
1846 	else /* cache is meaningless */
1847 		dev->cache_size = 0;
1848 	dev->cache_size = min_t(unsigned long, ULONG_MAX / 1024 / 1024,
1849 						dev->cache_size);
1850 	dev->mbps = min_t(unsigned int, 1024 * 40, dev->mbps);
1851 
1852 	if (dev->zoned &&
1853 	    (!dev->zone_size || !is_power_of_2(dev->zone_size))) {
1854 		pr_err("zone_size must be power-of-two\n");
1855 		return -EINVAL;
1856 	}
1857 
1858 	return 0;
1859 }
1860 
1861 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1862 static bool __null_setup_fault(struct fault_attr *attr, char *str)
1863 {
1864 	if (!str[0])
1865 		return true;
1866 
1867 	if (!setup_fault_attr(attr, str))
1868 		return false;
1869 
1870 	attr->verbose = 0;
1871 	return true;
1872 }
1873 #endif
1874 
1875 static bool null_setup_fault(void)
1876 {
1877 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1878 	if (!__null_setup_fault(&null_timeout_attr, g_timeout_str))
1879 		return false;
1880 	if (!__null_setup_fault(&null_requeue_attr, g_requeue_str))
1881 		return false;
1882 	if (!__null_setup_fault(&null_init_hctx_attr, g_init_hctx_str))
1883 		return false;
1884 #endif
1885 	return true;
1886 }
1887 
1888 static int null_add_dev(struct nullb_device *dev)
1889 {
1890 	struct queue_limits lim = {
1891 		.logical_block_size	= dev->blocksize,
1892 		.physical_block_size	= dev->blocksize,
1893 		.max_hw_sectors		= dev->max_sectors,
1894 	};
1895 
1896 	struct nullb *nullb;
1897 	int rv;
1898 
1899 	rv = null_validate_conf(dev);
1900 	if (rv)
1901 		return rv;
1902 
1903 	nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, dev->home_node);
1904 	if (!nullb) {
1905 		rv = -ENOMEM;
1906 		goto out;
1907 	}
1908 	nullb->dev = dev;
1909 	dev->nullb = nullb;
1910 
1911 	spin_lock_init(&nullb->lock);
1912 
1913 	rv = setup_queues(nullb);
1914 	if (rv)
1915 		goto out_free_nullb;
1916 
1917 	rv = null_setup_tagset(nullb);
1918 	if (rv)
1919 		goto out_cleanup_queues;
1920 
1921 	if (dev->virt_boundary)
1922 		lim.virt_boundary_mask = PAGE_SIZE - 1;
1923 	null_config_discard(nullb, &lim);
1924 	if (dev->zoned) {
1925 		rv = null_init_zoned_dev(dev, &lim);
1926 		if (rv)
1927 			goto out_cleanup_tags;
1928 	}
1929 
1930 	nullb->disk = blk_mq_alloc_disk(nullb->tag_set, &lim, nullb);
1931 	if (IS_ERR(nullb->disk)) {
1932 		rv = PTR_ERR(nullb->disk);
1933 		goto out_cleanup_zone;
1934 	}
1935 	nullb->q = nullb->disk->queue;
1936 
1937 	if (dev->mbps) {
1938 		set_bit(NULLB_DEV_FL_THROTTLED, &dev->flags);
1939 		nullb_setup_bwtimer(nullb);
1940 	}
1941 
1942 	if (dev->cache_size > 0) {
1943 		set_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
1944 		blk_queue_write_cache(nullb->q, true, dev->fua);
1945 	}
1946 
1947 	nullb->q->queuedata = nullb;
1948 	blk_queue_flag_set(QUEUE_FLAG_NONROT, nullb->q);
1949 
1950 	rv = ida_alloc(&nullb_indexes, GFP_KERNEL);
1951 	if (rv < 0)
1952 		goto out_cleanup_disk;
1953 
1954 	nullb->index = rv;
1955 	dev->index = rv;
1956 
1957 	if (config_item_name(&dev->group.cg_item)) {
1958 		/* Use configfs dir name as the device name */
1959 		snprintf(nullb->disk_name, sizeof(nullb->disk_name),
1960 			 "%s", config_item_name(&dev->group.cg_item));
1961 	} else {
1962 		sprintf(nullb->disk_name, "nullb%d", nullb->index);
1963 	}
1964 
1965 	set_capacity(nullb->disk,
1966 		((sector_t)nullb->dev->size * SZ_1M) >> SECTOR_SHIFT);
1967 	nullb->disk->major = null_major;
1968 	nullb->disk->first_minor = nullb->index;
1969 	nullb->disk->minors = 1;
1970 	nullb->disk->fops = &null_ops;
1971 	nullb->disk->private_data = nullb;
1972 	strscpy_pad(nullb->disk->disk_name, nullb->disk_name, DISK_NAME_LEN);
1973 
1974 	if (nullb->dev->zoned) {
1975 		rv = null_register_zoned_dev(nullb);
1976 		if (rv)
1977 			goto out_ida_free;
1978 	}
1979 
1980 	rv = add_disk(nullb->disk);
1981 	if (rv)
1982 		goto out_ida_free;
1983 
1984 	list_add_tail(&nullb->list, &nullb_list);
1985 
1986 	pr_info("disk %s created\n", nullb->disk_name);
1987 
1988 	return 0;
1989 
1990 out_ida_free:
1991 	ida_free(&nullb_indexes, nullb->index);
1992 out_cleanup_disk:
1993 	put_disk(nullb->disk);
1994 out_cleanup_zone:
1995 	null_free_zoned_dev(dev);
1996 out_cleanup_tags:
1997 	if (nullb->tag_set == &nullb->__tag_set)
1998 		blk_mq_free_tag_set(nullb->tag_set);
1999 out_cleanup_queues:
2000 	kfree(nullb->queues);
2001 out_free_nullb:
2002 	kfree(nullb);
2003 	dev->nullb = NULL;
2004 out:
2005 	return rv;
2006 }
2007 
2008 static struct nullb *null_find_dev_by_name(const char *name)
2009 {
2010 	struct nullb *nullb = NULL, *nb;
2011 
2012 	mutex_lock(&lock);
2013 	list_for_each_entry(nb, &nullb_list, list) {
2014 		if (strcmp(nb->disk_name, name) == 0) {
2015 			nullb = nb;
2016 			break;
2017 		}
2018 	}
2019 	mutex_unlock(&lock);
2020 
2021 	return nullb;
2022 }
2023 
2024 static int null_create_dev(void)
2025 {
2026 	struct nullb_device *dev;
2027 	int ret;
2028 
2029 	dev = null_alloc_dev();
2030 	if (!dev)
2031 		return -ENOMEM;
2032 
2033 	mutex_lock(&lock);
2034 	ret = null_add_dev(dev);
2035 	mutex_unlock(&lock);
2036 	if (ret) {
2037 		null_free_dev(dev);
2038 		return ret;
2039 	}
2040 
2041 	return 0;
2042 }
2043 
2044 static void null_destroy_dev(struct nullb *nullb)
2045 {
2046 	struct nullb_device *dev = nullb->dev;
2047 
2048 	null_del_dev(nullb);
2049 	null_free_device_storage(dev, false);
2050 	null_free_dev(dev);
2051 }
2052 
2053 static int __init null_init(void)
2054 {
2055 	int ret = 0;
2056 	unsigned int i;
2057 	struct nullb *nullb;
2058 
2059 	if (g_bs > PAGE_SIZE) {
2060 		pr_warn("invalid block size\n");
2061 		pr_warn("defaults block size to %lu\n", PAGE_SIZE);
2062 		g_bs = PAGE_SIZE;
2063 	}
2064 
2065 	if (g_home_node != NUMA_NO_NODE && g_home_node >= nr_online_nodes) {
2066 		pr_err("invalid home_node value\n");
2067 		g_home_node = NUMA_NO_NODE;
2068 	}
2069 
2070 	if (!null_setup_fault())
2071 		return -EINVAL;
2072 
2073 	if (g_queue_mode == NULL_Q_RQ) {
2074 		pr_err("legacy IO path is no longer available\n");
2075 		return -EINVAL;
2076 	}
2077 
2078 	if (g_use_per_node_hctx) {
2079 		if (g_submit_queues != nr_online_nodes) {
2080 			pr_warn("submit_queues param is set to %u.\n",
2081 				nr_online_nodes);
2082 			g_submit_queues = nr_online_nodes;
2083 		}
2084 	} else if (g_submit_queues > nr_cpu_ids) {
2085 		g_submit_queues = nr_cpu_ids;
2086 	} else if (g_submit_queues <= 0) {
2087 		g_submit_queues = 1;
2088 	}
2089 
2090 	config_group_init(&nullb_subsys.su_group);
2091 	mutex_init(&nullb_subsys.su_mutex);
2092 
2093 	ret = configfs_register_subsystem(&nullb_subsys);
2094 	if (ret)
2095 		return ret;
2096 
2097 	mutex_init(&lock);
2098 
2099 	null_major = register_blkdev(0, "nullb");
2100 	if (null_major < 0) {
2101 		ret = null_major;
2102 		goto err_conf;
2103 	}
2104 
2105 	for (i = 0; i < nr_devices; i++) {
2106 		ret = null_create_dev();
2107 		if (ret)
2108 			goto err_dev;
2109 	}
2110 
2111 	pr_info("module loaded\n");
2112 	return 0;
2113 
2114 err_dev:
2115 	while (!list_empty(&nullb_list)) {
2116 		nullb = list_entry(nullb_list.next, struct nullb, list);
2117 		null_destroy_dev(nullb);
2118 	}
2119 	unregister_blkdev(null_major, "nullb");
2120 err_conf:
2121 	configfs_unregister_subsystem(&nullb_subsys);
2122 	return ret;
2123 }
2124 
2125 static void __exit null_exit(void)
2126 {
2127 	struct nullb *nullb;
2128 
2129 	configfs_unregister_subsystem(&nullb_subsys);
2130 
2131 	unregister_blkdev(null_major, "nullb");
2132 
2133 	mutex_lock(&lock);
2134 	while (!list_empty(&nullb_list)) {
2135 		nullb = list_entry(nullb_list.next, struct nullb, list);
2136 		null_destroy_dev(nullb);
2137 	}
2138 	mutex_unlock(&lock);
2139 
2140 	if (tag_set.ops)
2141 		blk_mq_free_tag_set(&tag_set);
2142 
2143 	mutex_destroy(&lock);
2144 }
2145 
2146 module_init(null_init);
2147 module_exit(null_exit);
2148 
2149 MODULE_AUTHOR("Jens Axboe <axboe@kernel.dk>");
2150 MODULE_DESCRIPTION("multi queue aware block test driver");
2151 MODULE_LICENSE("GPL");
2152