1 // SPDX-License-Identifier: GPL-2.0-only
3 * Add configfs and memory store: Kyungchan Koh <kkc6196@fb.com> and
4 * Shaohua Li <shli@fb.com>
6 #include <linux/module.h>
8 #include <linux/moduleparam.h>
9 #include <linux/sched.h>
11 #include <linux/init.h>
15 #define pr_fmt(fmt) "null_blk: " fmt
19 #define TICKS_PER_SEC 50ULL
20 #define TIMER_INTERVAL (NSEC_PER_SEC / TICKS_PER_SEC)
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);
28 static inline u64 mb_per_tick(int mbps)
30 return (1 << 20) / TICKS_PER_SEC * ((u64) mbps);
34 * Status flags for nullb_device.
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.
41 enum nullb_device_flags {
42 NULLB_DEV_FL_CONFIGURED = 0,
44 NULLB_DEV_FL_THROTTLED = 2,
45 NULLB_DEV_FL_CACHE = 3,
48 #define MAP_SZ ((PAGE_SIZE >> SECTOR_SHIFT) + 2)
50 * nullb_page is a page in memory for nullb devices.
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
63 DECLARE_BITMAP(bitmap, MAP_SZ);
65 #define NULLB_PAGE_LOCK (MAP_SZ - 1)
66 #define NULLB_PAGE_FREE (MAP_SZ - 2)
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;
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");
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");
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");
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");
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");
100 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
102 * For more details about fault injection, please refer to
103 * Documentation/fault-injection/fault-injection.rst.
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>");
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>");
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>");
119 * Historic queue modes.
121 * These days nothing but NULL_Q_MQ is actually supported, but we keep it the
122 * enum for error reporting.
130 static int g_queue_mode = NULL_Q_MQ;
132 static int null_param_store_val(const char *str, int *val, int min, int max)
136 ret = kstrtoint(str, 10, &new_val);
140 if (new_val < min || new_val > max)
147 static int null_set_queue_mode(const char *str, const struct kernel_param *kp)
149 return null_param_store_val(str, &g_queue_mode, NULL_Q_BIO, NULL_Q_MQ);
152 static const struct kernel_param_ops null_queue_mode_param_ops = {
153 .set = null_set_queue_mode,
154 .get = param_get_int,
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)");
160 static int g_gb = 250;
161 module_param_named(gb, g_gb, int, 0444);
162 MODULE_PARM_DESC(gb, "Size in GB");
164 static int g_bs = 512;
165 module_param_named(bs, g_bs, int, 0444);
166 MODULE_PARM_DESC(bs, "Block size (in bytes)");
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)");
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");
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");
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");
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");
188 static int g_irqmode = NULL_IRQ_SOFTIRQ;
190 static int null_set_irqmode(const char *str, const struct kernel_param *kp)
192 return null_param_store_val(str, &g_irqmode, NULL_IRQ_NONE,
196 static const struct kernel_param_ops null_irqmode_param_ops = {
197 .set = null_set_irqmode,
198 .get = param_get_int,
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");
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");
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");
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");
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");
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");
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)");
228 static unsigned int g_mbps;
229 module_param_named(mbps, g_mbps, uint, 0444);
230 MODULE_PARM_DESC(mbps, "Limit maximum bandwidth (in MiB/s). Default: 0 (no limit)");
233 module_param_named(zoned, g_zoned, bool, S_IRUGO);
234 MODULE_PARM_DESC(zoned, "Make device as a host-managed zoned block device. Default: false");
236 static unsigned long g_zone_size = 256;
237 module_param_named(zone_size, g_zone_size, ulong, S_IRUGO);
238 MODULE_PARM_DESC(zone_size, "Zone size in MB when block device is zoned. Must be power-of-two: Default: 256");
240 static unsigned long g_zone_capacity;
241 module_param_named(zone_capacity, g_zone_capacity, ulong, 0444);
242 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");
244 static unsigned int g_zone_nr_conv;
245 module_param_named(zone_nr_conv, g_zone_nr_conv, uint, 0444);
246 MODULE_PARM_DESC(zone_nr_conv, "Number of conventional zones when block device is zoned. Default: 0");
248 static unsigned int g_zone_max_open;
249 module_param_named(zone_max_open, g_zone_max_open, uint, 0444);
250 MODULE_PARM_DESC(zone_max_open, "Maximum number of open zones when block device is zoned. Default: 0 (no limit)");
252 static unsigned int g_zone_max_active;
253 module_param_named(zone_max_active, g_zone_max_active, uint, 0444);
254 MODULE_PARM_DESC(zone_max_active, "Maximum number of active zones when block device is zoned. Default: 0 (no limit)");
256 static struct nullb_device *null_alloc_dev(void);
257 static void null_free_dev(struct nullb_device *dev);
258 static void null_del_dev(struct nullb *nullb);
259 static int null_add_dev(struct nullb_device *dev);
260 static struct nullb *null_find_dev_by_name(const char *name);
261 static void null_free_device_storage(struct nullb_device *dev, bool is_cache);
263 static inline struct nullb_device *to_nullb_device(struct config_item *item)
265 return item ? container_of(to_config_group(item), struct nullb_device, group) : NULL;
268 static inline ssize_t nullb_device_uint_attr_show(unsigned int val, char *page)
270 return snprintf(page, PAGE_SIZE, "%u\n", val);
273 static inline ssize_t nullb_device_ulong_attr_show(unsigned long val,
276 return snprintf(page, PAGE_SIZE, "%lu\n", val);
279 static inline ssize_t nullb_device_bool_attr_show(bool val, char *page)
281 return snprintf(page, PAGE_SIZE, "%u\n", val);
284 static ssize_t nullb_device_uint_attr_store(unsigned int *val,
285 const char *page, size_t count)
290 result = kstrtouint(page, 0, &tmp);
298 static ssize_t nullb_device_ulong_attr_store(unsigned long *val,
299 const char *page, size_t count)
304 result = kstrtoul(page, 0, &tmp);
312 static ssize_t nullb_device_bool_attr_store(bool *val, const char *page,
318 result = kstrtobool(page, &tmp);
326 /* The following macro should only be used with TYPE = {uint, ulong, bool}. */
327 #define NULLB_DEVICE_ATTR(NAME, TYPE, APPLY) \
329 nullb_device_##NAME##_show(struct config_item *item, char *page) \
331 return nullb_device_##TYPE##_attr_show( \
332 to_nullb_device(item)->NAME, page); \
335 nullb_device_##NAME##_store(struct config_item *item, const char *page, \
338 int (*apply_fn)(struct nullb_device *dev, TYPE new_value) = APPLY;\
339 struct nullb_device *dev = to_nullb_device(item); \
340 TYPE new_value = 0; \
343 ret = nullb_device_##TYPE##_attr_store(&new_value, page, count);\
347 ret = apply_fn(dev, new_value); \
348 else if (test_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags)) \
352 dev->NAME = new_value; \
355 CONFIGFS_ATTR(nullb_device_, NAME);
357 static int nullb_update_nr_hw_queues(struct nullb_device *dev,
358 unsigned int submit_queues,
359 unsigned int poll_queues)
362 struct blk_mq_tag_set *set;
363 int ret, nr_hw_queues;
369 * Make sure at least one submit queue exists.
375 * Make sure that null_init_hctx() does not access nullb->queues[] past
376 * the end of that array.
378 if (submit_queues > nr_cpu_ids || poll_queues > g_poll_queues)
382 * Keep previous and new queue numbers in nullb_device for reference in
383 * the call back function null_map_queues().
385 dev->prev_submit_queues = dev->submit_queues;
386 dev->prev_poll_queues = dev->poll_queues;
387 dev->submit_queues = submit_queues;
388 dev->poll_queues = poll_queues;
390 set = dev->nullb->tag_set;
391 nr_hw_queues = submit_queues + poll_queues;
392 blk_mq_update_nr_hw_queues(set, nr_hw_queues);
393 ret = set->nr_hw_queues == nr_hw_queues ? 0 : -ENOMEM;
396 /* on error, revert the queue numbers */
397 dev->submit_queues = dev->prev_submit_queues;
398 dev->poll_queues = dev->prev_poll_queues;
404 static int nullb_apply_submit_queues(struct nullb_device *dev,
405 unsigned int submit_queues)
407 return nullb_update_nr_hw_queues(dev, submit_queues, dev->poll_queues);
410 static int nullb_apply_poll_queues(struct nullb_device *dev,
411 unsigned int poll_queues)
413 return nullb_update_nr_hw_queues(dev, dev->submit_queues, poll_queues);
416 NULLB_DEVICE_ATTR(size, ulong, NULL);
417 NULLB_DEVICE_ATTR(completion_nsec, ulong, NULL);
418 NULLB_DEVICE_ATTR(submit_queues, uint, nullb_apply_submit_queues);
419 NULLB_DEVICE_ATTR(poll_queues, uint, nullb_apply_poll_queues);
420 NULLB_DEVICE_ATTR(home_node, uint, NULL);
421 NULLB_DEVICE_ATTR(queue_mode, uint, NULL);
422 NULLB_DEVICE_ATTR(blocksize, uint, NULL);
423 NULLB_DEVICE_ATTR(max_sectors, uint, NULL);
424 NULLB_DEVICE_ATTR(irqmode, uint, NULL);
425 NULLB_DEVICE_ATTR(hw_queue_depth, uint, NULL);
426 NULLB_DEVICE_ATTR(index, uint, NULL);
427 NULLB_DEVICE_ATTR(blocking, bool, NULL);
428 NULLB_DEVICE_ATTR(use_per_node_hctx, bool, NULL);
429 NULLB_DEVICE_ATTR(memory_backed, bool, NULL);
430 NULLB_DEVICE_ATTR(discard, bool, NULL);
431 NULLB_DEVICE_ATTR(mbps, uint, NULL);
432 NULLB_DEVICE_ATTR(cache_size, ulong, NULL);
433 NULLB_DEVICE_ATTR(zoned, bool, NULL);
434 NULLB_DEVICE_ATTR(zone_size, ulong, NULL);
435 NULLB_DEVICE_ATTR(zone_capacity, ulong, NULL);
436 NULLB_DEVICE_ATTR(zone_nr_conv, uint, NULL);
437 NULLB_DEVICE_ATTR(zone_max_open, uint, NULL);
438 NULLB_DEVICE_ATTR(zone_max_active, uint, NULL);
439 NULLB_DEVICE_ATTR(virt_boundary, bool, NULL);
440 NULLB_DEVICE_ATTR(no_sched, bool, NULL);
441 NULLB_DEVICE_ATTR(shared_tags, bool, NULL);
442 NULLB_DEVICE_ATTR(shared_tag_bitmap, bool, NULL);
444 static ssize_t nullb_device_power_show(struct config_item *item, char *page)
446 return nullb_device_bool_attr_show(to_nullb_device(item)->power, page);
449 static ssize_t nullb_device_power_store(struct config_item *item,
450 const char *page, size_t count)
452 struct nullb_device *dev = to_nullb_device(item);
456 ret = nullb_device_bool_attr_store(&newp, page, count);
460 if (!dev->power && newp) {
461 if (test_and_set_bit(NULLB_DEV_FL_UP, &dev->flags))
463 ret = null_add_dev(dev);
465 clear_bit(NULLB_DEV_FL_UP, &dev->flags);
469 set_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
471 } else if (dev->power && !newp) {
472 if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
475 null_del_dev(dev->nullb);
478 clear_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
484 CONFIGFS_ATTR(nullb_device_, power);
486 static ssize_t nullb_device_badblocks_show(struct config_item *item, char *page)
488 struct nullb_device *t_dev = to_nullb_device(item);
490 return badblocks_show(&t_dev->badblocks, page, 0);
493 static ssize_t nullb_device_badblocks_store(struct config_item *item,
494 const char *page, size_t count)
496 struct nullb_device *t_dev = to_nullb_device(item);
497 char *orig, *buf, *tmp;
501 orig = kstrndup(page, count, GFP_KERNEL);
505 buf = strstrip(orig);
508 if (buf[0] != '+' && buf[0] != '-')
510 tmp = strchr(&buf[1], '-');
514 ret = kstrtoull(buf + 1, 0, &start);
517 ret = kstrtoull(tmp + 1, 0, &end);
523 /* enable badblocks */
524 cmpxchg(&t_dev->badblocks.shift, -1, 0);
526 ret = badblocks_set(&t_dev->badblocks, start,
529 ret = badblocks_clear(&t_dev->badblocks, start,
537 CONFIGFS_ATTR(nullb_device_, badblocks);
539 static ssize_t nullb_device_zone_readonly_store(struct config_item *item,
540 const char *page, size_t count)
542 struct nullb_device *dev = to_nullb_device(item);
544 return zone_cond_store(dev, page, count, BLK_ZONE_COND_READONLY);
546 CONFIGFS_ATTR_WO(nullb_device_, zone_readonly);
548 static ssize_t nullb_device_zone_offline_store(struct config_item *item,
549 const char *page, size_t count)
551 struct nullb_device *dev = to_nullb_device(item);
553 return zone_cond_store(dev, page, count, BLK_ZONE_COND_OFFLINE);
555 CONFIGFS_ATTR_WO(nullb_device_, zone_offline);
557 static struct configfs_attribute *nullb_device_attrs[] = {
558 &nullb_device_attr_size,
559 &nullb_device_attr_completion_nsec,
560 &nullb_device_attr_submit_queues,
561 &nullb_device_attr_poll_queues,
562 &nullb_device_attr_home_node,
563 &nullb_device_attr_queue_mode,
564 &nullb_device_attr_blocksize,
565 &nullb_device_attr_max_sectors,
566 &nullb_device_attr_irqmode,
567 &nullb_device_attr_hw_queue_depth,
568 &nullb_device_attr_index,
569 &nullb_device_attr_blocking,
570 &nullb_device_attr_use_per_node_hctx,
571 &nullb_device_attr_power,
572 &nullb_device_attr_memory_backed,
573 &nullb_device_attr_discard,
574 &nullb_device_attr_mbps,
575 &nullb_device_attr_cache_size,
576 &nullb_device_attr_badblocks,
577 &nullb_device_attr_zoned,
578 &nullb_device_attr_zone_size,
579 &nullb_device_attr_zone_capacity,
580 &nullb_device_attr_zone_nr_conv,
581 &nullb_device_attr_zone_max_open,
582 &nullb_device_attr_zone_max_active,
583 &nullb_device_attr_zone_readonly,
584 &nullb_device_attr_zone_offline,
585 &nullb_device_attr_virt_boundary,
586 &nullb_device_attr_no_sched,
587 &nullb_device_attr_shared_tags,
588 &nullb_device_attr_shared_tag_bitmap,
592 static void nullb_device_release(struct config_item *item)
594 struct nullb_device *dev = to_nullb_device(item);
596 null_free_device_storage(dev, false);
600 static struct configfs_item_operations nullb_device_ops = {
601 .release = nullb_device_release,
604 static const struct config_item_type nullb_device_type = {
605 .ct_item_ops = &nullb_device_ops,
606 .ct_attrs = nullb_device_attrs,
607 .ct_owner = THIS_MODULE,
610 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
612 static void nullb_add_fault_config(struct nullb_device *dev)
614 fault_config_init(&dev->timeout_config, "timeout_inject");
615 fault_config_init(&dev->requeue_config, "requeue_inject");
616 fault_config_init(&dev->init_hctx_fault_config, "init_hctx_fault_inject");
618 configfs_add_default_group(&dev->timeout_config.group, &dev->group);
619 configfs_add_default_group(&dev->requeue_config.group, &dev->group);
620 configfs_add_default_group(&dev->init_hctx_fault_config.group, &dev->group);
625 static void nullb_add_fault_config(struct nullb_device *dev)
632 config_group *nullb_group_make_group(struct config_group *group, const char *name)
634 struct nullb_device *dev;
636 if (null_find_dev_by_name(name))
637 return ERR_PTR(-EEXIST);
639 dev = null_alloc_dev();
641 return ERR_PTR(-ENOMEM);
643 config_group_init_type_name(&dev->group, name, &nullb_device_type);
644 nullb_add_fault_config(dev);
650 nullb_group_drop_item(struct config_group *group, struct config_item *item)
652 struct nullb_device *dev = to_nullb_device(item);
654 if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
657 null_del_dev(dev->nullb);
661 config_item_put(item);
664 static ssize_t memb_group_features_show(struct config_item *item, char *page)
666 return snprintf(page, PAGE_SIZE,
667 "badblocks,blocking,blocksize,cache_size,"
668 "completion_nsec,discard,home_node,hw_queue_depth,"
669 "irqmode,max_sectors,mbps,memory_backed,no_sched,"
670 "poll_queues,power,queue_mode,shared_tag_bitmap,"
671 "shared_tags,size,submit_queues,use_per_node_hctx,"
672 "virt_boundary,zoned,zone_capacity,zone_max_active,"
673 "zone_max_open,zone_nr_conv,zone_offline,zone_readonly,"
677 CONFIGFS_ATTR_RO(memb_group_, features);
679 static struct configfs_attribute *nullb_group_attrs[] = {
680 &memb_group_attr_features,
684 static struct configfs_group_operations nullb_group_ops = {
685 .make_group = nullb_group_make_group,
686 .drop_item = nullb_group_drop_item,
689 static const struct config_item_type nullb_group_type = {
690 .ct_group_ops = &nullb_group_ops,
691 .ct_attrs = nullb_group_attrs,
692 .ct_owner = THIS_MODULE,
695 static struct configfs_subsystem nullb_subsys = {
698 .ci_namebuf = "nullb",
699 .ci_type = &nullb_group_type,
704 static inline int null_cache_active(struct nullb *nullb)
706 return test_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
709 static struct nullb_device *null_alloc_dev(void)
711 struct nullb_device *dev;
713 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
717 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
718 dev->timeout_config.attr = null_timeout_attr;
719 dev->requeue_config.attr = null_requeue_attr;
720 dev->init_hctx_fault_config.attr = null_init_hctx_attr;
723 INIT_RADIX_TREE(&dev->data, GFP_ATOMIC);
724 INIT_RADIX_TREE(&dev->cache, GFP_ATOMIC);
725 if (badblocks_init(&dev->badblocks, 0)) {
730 dev->size = g_gb * 1024;
731 dev->completion_nsec = g_completion_nsec;
732 dev->submit_queues = g_submit_queues;
733 dev->prev_submit_queues = g_submit_queues;
734 dev->poll_queues = g_poll_queues;
735 dev->prev_poll_queues = g_poll_queues;
736 dev->home_node = g_home_node;
737 dev->queue_mode = g_queue_mode;
738 dev->blocksize = g_bs;
739 dev->max_sectors = g_max_sectors;
740 dev->irqmode = g_irqmode;
741 dev->hw_queue_depth = g_hw_queue_depth;
742 dev->blocking = g_blocking;
743 dev->memory_backed = g_memory_backed;
744 dev->discard = g_discard;
745 dev->cache_size = g_cache_size;
747 dev->use_per_node_hctx = g_use_per_node_hctx;
748 dev->zoned = g_zoned;
749 dev->zone_size = g_zone_size;
750 dev->zone_capacity = g_zone_capacity;
751 dev->zone_nr_conv = g_zone_nr_conv;
752 dev->zone_max_open = g_zone_max_open;
753 dev->zone_max_active = g_zone_max_active;
754 dev->virt_boundary = g_virt_boundary;
755 dev->no_sched = g_no_sched;
756 dev->shared_tags = g_shared_tags;
757 dev->shared_tag_bitmap = g_shared_tag_bitmap;
761 static void null_free_dev(struct nullb_device *dev)
766 null_free_zoned_dev(dev);
767 badblocks_exit(&dev->badblocks);
771 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
773 struct nullb_cmd *cmd = container_of(timer, struct nullb_cmd, timer);
775 blk_mq_end_request(blk_mq_rq_from_pdu(cmd), cmd->error);
776 return HRTIMER_NORESTART;
779 static void null_cmd_end_timer(struct nullb_cmd *cmd)
781 ktime_t kt = cmd->nq->dev->completion_nsec;
783 hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL);
786 static void null_complete_rq(struct request *rq)
788 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq);
790 blk_mq_end_request(rq, cmd->error);
793 static struct nullb_page *null_alloc_page(void)
795 struct nullb_page *t_page;
797 t_page = kmalloc(sizeof(struct nullb_page), GFP_NOIO);
801 t_page->page = alloc_pages(GFP_NOIO, 0);
807 memset(t_page->bitmap, 0, sizeof(t_page->bitmap));
811 static void null_free_page(struct nullb_page *t_page)
813 __set_bit(NULLB_PAGE_FREE, t_page->bitmap);
814 if (test_bit(NULLB_PAGE_LOCK, t_page->bitmap))
816 __free_page(t_page->page);
820 static bool null_page_empty(struct nullb_page *page)
822 int size = MAP_SZ - 2;
824 return find_first_bit(page->bitmap, size) == size;
827 static void null_free_sector(struct nullb *nullb, sector_t sector,
830 unsigned int sector_bit;
832 struct nullb_page *t_page, *ret;
833 struct radix_tree_root *root;
835 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
836 idx = sector >> PAGE_SECTORS_SHIFT;
837 sector_bit = (sector & SECTOR_MASK);
839 t_page = radix_tree_lookup(root, idx);
841 __clear_bit(sector_bit, t_page->bitmap);
843 if (null_page_empty(t_page)) {
844 ret = radix_tree_delete_item(root, idx, t_page);
845 WARN_ON(ret != t_page);
848 nullb->dev->curr_cache -= PAGE_SIZE;
853 static struct nullb_page *null_radix_tree_insert(struct nullb *nullb, u64 idx,
854 struct nullb_page *t_page, bool is_cache)
856 struct radix_tree_root *root;
858 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
860 if (radix_tree_insert(root, idx, t_page)) {
861 null_free_page(t_page);
862 t_page = radix_tree_lookup(root, idx);
863 WARN_ON(!t_page || t_page->page->index != idx);
865 nullb->dev->curr_cache += PAGE_SIZE;
870 static void null_free_device_storage(struct nullb_device *dev, bool is_cache)
872 unsigned long pos = 0;
874 struct nullb_page *ret, *t_pages[FREE_BATCH];
875 struct radix_tree_root *root;
877 root = is_cache ? &dev->cache : &dev->data;
882 nr_pages = radix_tree_gang_lookup(root,
883 (void **)t_pages, pos, FREE_BATCH);
885 for (i = 0; i < nr_pages; i++) {
886 pos = t_pages[i]->page->index;
887 ret = radix_tree_delete_item(root, pos, t_pages[i]);
888 WARN_ON(ret != t_pages[i]);
893 } while (nr_pages == FREE_BATCH);
899 static struct nullb_page *__null_lookup_page(struct nullb *nullb,
900 sector_t sector, bool for_write, bool is_cache)
902 unsigned int sector_bit;
904 struct nullb_page *t_page;
905 struct radix_tree_root *root;
907 idx = sector >> PAGE_SECTORS_SHIFT;
908 sector_bit = (sector & SECTOR_MASK);
910 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
911 t_page = radix_tree_lookup(root, idx);
912 WARN_ON(t_page && t_page->page->index != idx);
914 if (t_page && (for_write || test_bit(sector_bit, t_page->bitmap)))
920 static struct nullb_page *null_lookup_page(struct nullb *nullb,
921 sector_t sector, bool for_write, bool ignore_cache)
923 struct nullb_page *page = NULL;
926 page = __null_lookup_page(nullb, sector, for_write, true);
929 return __null_lookup_page(nullb, sector, for_write, false);
932 static struct nullb_page *null_insert_page(struct nullb *nullb,
933 sector_t sector, bool ignore_cache)
934 __releases(&nullb->lock)
935 __acquires(&nullb->lock)
938 struct nullb_page *t_page;
940 t_page = null_lookup_page(nullb, sector, true, ignore_cache);
944 spin_unlock_irq(&nullb->lock);
946 t_page = null_alloc_page();
950 if (radix_tree_preload(GFP_NOIO))
953 spin_lock_irq(&nullb->lock);
954 idx = sector >> PAGE_SECTORS_SHIFT;
955 t_page->page->index = idx;
956 t_page = null_radix_tree_insert(nullb, idx, t_page, !ignore_cache);
957 radix_tree_preload_end();
961 null_free_page(t_page);
963 spin_lock_irq(&nullb->lock);
964 return null_lookup_page(nullb, sector, true, ignore_cache);
967 static int null_flush_cache_page(struct nullb *nullb, struct nullb_page *c_page)
972 struct nullb_page *t_page, *ret;
975 idx = c_page->page->index;
977 t_page = null_insert_page(nullb, idx << PAGE_SECTORS_SHIFT, true);
979 __clear_bit(NULLB_PAGE_LOCK, c_page->bitmap);
980 if (test_bit(NULLB_PAGE_FREE, c_page->bitmap)) {
981 null_free_page(c_page);
982 if (t_page && null_page_empty(t_page)) {
983 ret = radix_tree_delete_item(&nullb->dev->data,
985 null_free_page(t_page);
993 src = kmap_local_page(c_page->page);
994 dst = kmap_local_page(t_page->page);
996 for (i = 0; i < PAGE_SECTORS;
997 i += (nullb->dev->blocksize >> SECTOR_SHIFT)) {
998 if (test_bit(i, c_page->bitmap)) {
999 offset = (i << SECTOR_SHIFT);
1000 memcpy(dst + offset, src + offset,
1001 nullb->dev->blocksize);
1002 __set_bit(i, t_page->bitmap);
1009 ret = radix_tree_delete_item(&nullb->dev->cache, idx, c_page);
1010 null_free_page(ret);
1011 nullb->dev->curr_cache -= PAGE_SIZE;
1016 static int null_make_cache_space(struct nullb *nullb, unsigned long n)
1018 int i, err, nr_pages;
1019 struct nullb_page *c_pages[FREE_BATCH];
1020 unsigned long flushed = 0, one_round;
1023 if ((nullb->dev->cache_size * 1024 * 1024) >
1024 nullb->dev->curr_cache + n || nullb->dev->curr_cache == 0)
1027 nr_pages = radix_tree_gang_lookup(&nullb->dev->cache,
1028 (void **)c_pages, nullb->cache_flush_pos, FREE_BATCH);
1030 * nullb_flush_cache_page could unlock before using the c_pages. To
1031 * avoid race, we don't allow page free
1033 for (i = 0; i < nr_pages; i++) {
1034 nullb->cache_flush_pos = c_pages[i]->page->index;
1036 * We found the page which is being flushed to disk by other
1039 if (test_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap))
1042 __set_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap);
1046 for (i = 0; i < nr_pages; i++) {
1047 if (c_pages[i] == NULL)
1049 err = null_flush_cache_page(nullb, c_pages[i]);
1054 flushed += one_round << PAGE_SHIFT;
1058 nullb->cache_flush_pos = 0;
1059 if (one_round == 0) {
1060 /* give other threads a chance */
1061 spin_unlock_irq(&nullb->lock);
1062 spin_lock_irq(&nullb->lock);
1069 static int copy_to_nullb(struct nullb *nullb, struct page *source,
1070 unsigned int off, sector_t sector, size_t n, bool is_fua)
1072 size_t temp, count = 0;
1073 unsigned int offset;
1074 struct nullb_page *t_page;
1077 temp = min_t(size_t, nullb->dev->blocksize, n - count);
1079 if (null_cache_active(nullb) && !is_fua)
1080 null_make_cache_space(nullb, PAGE_SIZE);
1082 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
1083 t_page = null_insert_page(nullb, sector,
1084 !null_cache_active(nullb) || is_fua);
1088 memcpy_page(t_page->page, offset, source, off + count, temp);
1090 __set_bit(sector & SECTOR_MASK, t_page->bitmap);
1093 null_free_sector(nullb, sector, true);
1096 sector += temp >> SECTOR_SHIFT;
1101 static int copy_from_nullb(struct nullb *nullb, struct page *dest,
1102 unsigned int off, sector_t sector, size_t n)
1104 size_t temp, count = 0;
1105 unsigned int offset;
1106 struct nullb_page *t_page;
1109 temp = min_t(size_t, nullb->dev->blocksize, n - count);
1111 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
1112 t_page = null_lookup_page(nullb, sector, false,
1113 !null_cache_active(nullb));
1116 memcpy_page(dest, off + count, t_page->page, offset,
1119 zero_user(dest, off + count, temp);
1122 sector += temp >> SECTOR_SHIFT;
1127 static void nullb_fill_pattern(struct nullb *nullb, struct page *page,
1128 unsigned int len, unsigned int off)
1130 memset_page(page, off, 0xff, len);
1133 blk_status_t null_handle_discard(struct nullb_device *dev,
1134 sector_t sector, sector_t nr_sectors)
1136 struct nullb *nullb = dev->nullb;
1137 size_t n = nr_sectors << SECTOR_SHIFT;
1140 spin_lock_irq(&nullb->lock);
1142 temp = min_t(size_t, n, dev->blocksize);
1143 null_free_sector(nullb, sector, false);
1144 if (null_cache_active(nullb))
1145 null_free_sector(nullb, sector, true);
1146 sector += temp >> SECTOR_SHIFT;
1149 spin_unlock_irq(&nullb->lock);
1154 static int null_handle_flush(struct nullb *nullb)
1158 if (!null_cache_active(nullb))
1161 spin_lock_irq(&nullb->lock);
1163 err = null_make_cache_space(nullb,
1164 nullb->dev->cache_size * 1024 * 1024);
1165 if (err || nullb->dev->curr_cache == 0)
1169 WARN_ON(!radix_tree_empty(&nullb->dev->cache));
1170 spin_unlock_irq(&nullb->lock);
1174 static int null_transfer(struct nullb *nullb, struct page *page,
1175 unsigned int len, unsigned int off, bool is_write, sector_t sector,
1178 struct nullb_device *dev = nullb->dev;
1179 unsigned int valid_len = len;
1184 valid_len = null_zone_valid_read_len(nullb,
1188 err = copy_from_nullb(nullb, page, off,
1195 nullb_fill_pattern(nullb, page, len, off);
1196 flush_dcache_page(page);
1198 flush_dcache_page(page);
1199 err = copy_to_nullb(nullb, page, off, sector, len, is_fua);
1205 static int null_handle_rq(struct nullb_cmd *cmd)
1207 struct request *rq = blk_mq_rq_from_pdu(cmd);
1208 struct nullb *nullb = cmd->nq->dev->nullb;
1211 sector_t sector = blk_rq_pos(rq);
1212 struct req_iterator iter;
1213 struct bio_vec bvec;
1215 spin_lock_irq(&nullb->lock);
1216 rq_for_each_segment(bvec, rq, iter) {
1218 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1219 op_is_write(req_op(rq)), sector,
1220 rq->cmd_flags & REQ_FUA);
1222 spin_unlock_irq(&nullb->lock);
1225 sector += len >> SECTOR_SHIFT;
1227 spin_unlock_irq(&nullb->lock);
1232 static inline blk_status_t null_handle_throttled(struct nullb_cmd *cmd)
1234 struct nullb_device *dev = cmd->nq->dev;
1235 struct nullb *nullb = dev->nullb;
1236 blk_status_t sts = BLK_STS_OK;
1237 struct request *rq = blk_mq_rq_from_pdu(cmd);
1239 if (!hrtimer_active(&nullb->bw_timer))
1240 hrtimer_restart(&nullb->bw_timer);
1242 if (atomic_long_sub_return(blk_rq_bytes(rq), &nullb->cur_bytes) < 0) {
1243 blk_mq_stop_hw_queues(nullb->q);
1244 /* race with timer */
1245 if (atomic_long_read(&nullb->cur_bytes) > 0)
1246 blk_mq_start_stopped_hw_queues(nullb->q, true);
1247 /* requeue request */
1248 sts = BLK_STS_DEV_RESOURCE;
1253 static inline blk_status_t null_handle_badblocks(struct nullb_cmd *cmd,
1255 sector_t nr_sectors)
1257 struct badblocks *bb = &cmd->nq->dev->badblocks;
1261 if (badblocks_check(bb, sector, nr_sectors, &first_bad, &bad_sectors))
1262 return BLK_STS_IOERR;
1267 static inline blk_status_t null_handle_memory_backed(struct nullb_cmd *cmd,
1270 sector_t nr_sectors)
1272 struct nullb_device *dev = cmd->nq->dev;
1274 if (op == REQ_OP_DISCARD)
1275 return null_handle_discard(dev, sector, nr_sectors);
1276 return errno_to_blk_status(null_handle_rq(cmd));
1280 static void nullb_zero_read_cmd_buffer(struct nullb_cmd *cmd)
1282 struct request *rq = blk_mq_rq_from_pdu(cmd);
1283 struct nullb_device *dev = cmd->nq->dev;
1286 if (!dev->memory_backed && req_op(rq) == REQ_OP_READ) {
1287 __rq_for_each_bio(bio, rq)
1292 static inline void nullb_complete_cmd(struct nullb_cmd *cmd)
1294 struct request *rq = blk_mq_rq_from_pdu(cmd);
1297 * Since root privileges are required to configure the null_blk
1298 * driver, it is fine that this driver does not initialize the
1299 * data buffers of read commands. Zero-initialize these buffers
1300 * anyway if KMSAN is enabled to prevent that KMSAN complains
1301 * about null_blk not initializing read data buffers.
1303 if (IS_ENABLED(CONFIG_KMSAN))
1304 nullb_zero_read_cmd_buffer(cmd);
1306 /* Complete IO by inline, softirq or timer */
1307 switch (cmd->nq->dev->irqmode) {
1308 case NULL_IRQ_SOFTIRQ:
1309 blk_mq_complete_request(rq);
1312 blk_mq_end_request(rq, cmd->error);
1314 case NULL_IRQ_TIMER:
1315 null_cmd_end_timer(cmd);
1320 blk_status_t null_process_cmd(struct nullb_cmd *cmd, enum req_op op,
1321 sector_t sector, unsigned int nr_sectors)
1323 struct nullb_device *dev = cmd->nq->dev;
1326 if (dev->badblocks.shift != -1) {
1327 ret = null_handle_badblocks(cmd, sector, nr_sectors);
1328 if (ret != BLK_STS_OK)
1332 if (dev->memory_backed)
1333 return null_handle_memory_backed(cmd, op, sector, nr_sectors);
1338 static void null_handle_cmd(struct nullb_cmd *cmd, sector_t sector,
1339 sector_t nr_sectors, enum req_op op)
1341 struct nullb_device *dev = cmd->nq->dev;
1342 struct nullb *nullb = dev->nullb;
1345 if (op == REQ_OP_FLUSH) {
1346 cmd->error = errno_to_blk_status(null_handle_flush(nullb));
1351 sts = null_process_zoned_cmd(cmd, op, sector, nr_sectors);
1353 sts = null_process_cmd(cmd, op, sector, nr_sectors);
1355 /* Do not overwrite errors (e.g. timeout errors) */
1356 if (cmd->error == BLK_STS_OK)
1360 nullb_complete_cmd(cmd);
1363 static enum hrtimer_restart nullb_bwtimer_fn(struct hrtimer *timer)
1365 struct nullb *nullb = container_of(timer, struct nullb, bw_timer);
1366 ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1367 unsigned int mbps = nullb->dev->mbps;
1369 if (atomic_long_read(&nullb->cur_bytes) == mb_per_tick(mbps))
1370 return HRTIMER_NORESTART;
1372 atomic_long_set(&nullb->cur_bytes, mb_per_tick(mbps));
1373 blk_mq_start_stopped_hw_queues(nullb->q, true);
1375 hrtimer_forward_now(&nullb->bw_timer, timer_interval);
1377 return HRTIMER_RESTART;
1380 static void nullb_setup_bwtimer(struct nullb *nullb)
1382 ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1384 hrtimer_init(&nullb->bw_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1385 nullb->bw_timer.function = nullb_bwtimer_fn;
1386 atomic_long_set(&nullb->cur_bytes, mb_per_tick(nullb->dev->mbps));
1387 hrtimer_start(&nullb->bw_timer, timer_interval, HRTIMER_MODE_REL);
1390 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1392 static bool should_timeout_request(struct request *rq)
1394 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq);
1395 struct nullb_device *dev = cmd->nq->dev;
1397 return should_fail(&dev->timeout_config.attr, 1);
1400 static bool should_requeue_request(struct request *rq)
1402 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq);
1403 struct nullb_device *dev = cmd->nq->dev;
1405 return should_fail(&dev->requeue_config.attr, 1);
1408 static bool should_init_hctx_fail(struct nullb_device *dev)
1410 return should_fail(&dev->init_hctx_fault_config.attr, 1);
1415 static bool should_timeout_request(struct request *rq)
1420 static bool should_requeue_request(struct request *rq)
1425 static bool should_init_hctx_fail(struct nullb_device *dev)
1432 static void null_map_queues(struct blk_mq_tag_set *set)
1434 struct nullb *nullb = set->driver_data;
1436 unsigned int submit_queues = g_submit_queues;
1437 unsigned int poll_queues = g_poll_queues;
1440 struct nullb_device *dev = nullb->dev;
1443 * Refer nr_hw_queues of the tag set to check if the expected
1444 * number of hardware queues are prepared. If block layer failed
1445 * to prepare them, use previous numbers of submit queues and
1446 * poll queues to map queues.
1448 if (set->nr_hw_queues ==
1449 dev->submit_queues + dev->poll_queues) {
1450 submit_queues = dev->submit_queues;
1451 poll_queues = dev->poll_queues;
1452 } else if (set->nr_hw_queues ==
1453 dev->prev_submit_queues + dev->prev_poll_queues) {
1454 submit_queues = dev->prev_submit_queues;
1455 poll_queues = dev->prev_poll_queues;
1457 pr_warn("tag set has unexpected nr_hw_queues: %d\n",
1465 for (i = 0, qoff = 0; i < set->nr_maps; i++) {
1466 struct blk_mq_queue_map *map = &set->map[i];
1469 case HCTX_TYPE_DEFAULT:
1470 map->nr_queues = submit_queues;
1472 case HCTX_TYPE_READ:
1475 case HCTX_TYPE_POLL:
1476 map->nr_queues = poll_queues;
1479 map->queue_offset = qoff;
1480 qoff += map->nr_queues;
1481 blk_mq_map_queues(map);
1485 static int null_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1487 struct nullb_queue *nq = hctx->driver_data;
1492 spin_lock(&nq->poll_lock);
1493 list_splice_init(&nq->poll_list, &list);
1494 list_for_each_entry(rq, &list, queuelist)
1495 blk_mq_set_request_complete(rq);
1496 spin_unlock(&nq->poll_lock);
1498 while (!list_empty(&list)) {
1499 struct nullb_cmd *cmd;
1500 struct request *req;
1502 req = list_first_entry(&list, struct request, queuelist);
1503 list_del_init(&req->queuelist);
1504 cmd = blk_mq_rq_to_pdu(req);
1505 cmd->error = null_process_cmd(cmd, req_op(req), blk_rq_pos(req),
1506 blk_rq_sectors(req));
1507 if (!blk_mq_add_to_batch(req, iob, (__force int) cmd->error,
1508 blk_mq_end_request_batch))
1509 blk_mq_end_request(req, cmd->error);
1516 static enum blk_eh_timer_return null_timeout_rq(struct request *rq)
1518 struct blk_mq_hw_ctx *hctx = rq->mq_hctx;
1519 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq);
1521 if (hctx->type == HCTX_TYPE_POLL) {
1522 struct nullb_queue *nq = hctx->driver_data;
1524 spin_lock(&nq->poll_lock);
1525 /* The request may have completed meanwhile. */
1526 if (blk_mq_request_completed(rq)) {
1527 spin_unlock(&nq->poll_lock);
1530 list_del_init(&rq->queuelist);
1531 spin_unlock(&nq->poll_lock);
1534 pr_info("rq %p timed out\n", rq);
1537 * If the device is marked as blocking (i.e. memory backed or zoned
1538 * device), the submission path may be blocked waiting for resources
1539 * and cause real timeouts. For these real timeouts, the submission
1540 * path will complete the request using blk_mq_complete_request().
1541 * Only fake timeouts need to execute blk_mq_complete_request() here.
1543 cmd->error = BLK_STS_TIMEOUT;
1544 if (cmd->fake_timeout || hctx->type == HCTX_TYPE_POLL)
1545 blk_mq_complete_request(rq);
1549 static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx,
1550 const struct blk_mq_queue_data *bd)
1552 struct request *rq = bd->rq;
1553 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq);
1554 struct nullb_queue *nq = hctx->driver_data;
1555 sector_t nr_sectors = blk_rq_sectors(rq);
1556 sector_t sector = blk_rq_pos(rq);
1557 const bool is_poll = hctx->type == HCTX_TYPE_POLL;
1559 might_sleep_if(hctx->flags & BLK_MQ_F_BLOCKING);
1561 if (!is_poll && nq->dev->irqmode == NULL_IRQ_TIMER) {
1562 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1563 cmd->timer.function = null_cmd_timer_expired;
1565 cmd->error = BLK_STS_OK;
1567 cmd->fake_timeout = should_timeout_request(rq) ||
1568 blk_should_fake_timeout(rq->q);
1570 if (should_requeue_request(rq)) {
1572 * Alternate between hitting the core BUSY path, and the
1573 * driver driven requeue path
1575 nq->requeue_selection++;
1576 if (nq->requeue_selection & 1)
1577 return BLK_STS_RESOURCE;
1578 blk_mq_requeue_request(rq, true);
1582 if (test_bit(NULLB_DEV_FL_THROTTLED, &nq->dev->flags)) {
1583 blk_status_t sts = null_handle_throttled(cmd);
1585 if (sts != BLK_STS_OK)
1589 blk_mq_start_request(rq);
1592 spin_lock(&nq->poll_lock);
1593 list_add_tail(&rq->queuelist, &nq->poll_list);
1594 spin_unlock(&nq->poll_lock);
1597 if (cmd->fake_timeout)
1600 null_handle_cmd(cmd, sector, nr_sectors, req_op(rq));
1604 static void null_queue_rqs(struct request **rqlist)
1606 struct request *requeue_list = NULL;
1607 struct request **requeue_lastp = &requeue_list;
1608 struct blk_mq_queue_data bd = { };
1612 struct request *rq = rq_list_pop(rqlist);
1615 ret = null_queue_rq(rq->mq_hctx, &bd);
1616 if (ret != BLK_STS_OK)
1617 rq_list_add_tail(&requeue_lastp, rq);
1618 } while (!rq_list_empty(*rqlist));
1620 *rqlist = requeue_list;
1623 static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq)
1625 nq->dev = nullb->dev;
1626 INIT_LIST_HEAD(&nq->poll_list);
1627 spin_lock_init(&nq->poll_lock);
1630 static int null_init_hctx(struct blk_mq_hw_ctx *hctx, void *driver_data,
1631 unsigned int hctx_idx)
1633 struct nullb *nullb = hctx->queue->queuedata;
1634 struct nullb_queue *nq;
1636 if (should_init_hctx_fail(nullb->dev))
1639 nq = &nullb->queues[hctx_idx];
1640 hctx->driver_data = nq;
1641 null_init_queue(nullb, nq);
1646 static const struct blk_mq_ops null_mq_ops = {
1647 .queue_rq = null_queue_rq,
1648 .queue_rqs = null_queue_rqs,
1649 .complete = null_complete_rq,
1650 .timeout = null_timeout_rq,
1652 .map_queues = null_map_queues,
1653 .init_hctx = null_init_hctx,
1656 static void null_del_dev(struct nullb *nullb)
1658 struct nullb_device *dev;
1665 ida_free(&nullb_indexes, nullb->index);
1667 list_del_init(&nullb->list);
1669 del_gendisk(nullb->disk);
1671 if (test_bit(NULLB_DEV_FL_THROTTLED, &nullb->dev->flags)) {
1672 hrtimer_cancel(&nullb->bw_timer);
1673 atomic_long_set(&nullb->cur_bytes, LONG_MAX);
1674 blk_mq_start_stopped_hw_queues(nullb->q, true);
1677 put_disk(nullb->disk);
1678 if (nullb->tag_set == &nullb->__tag_set)
1679 blk_mq_free_tag_set(nullb->tag_set);
1680 kfree(nullb->queues);
1681 if (null_cache_active(nullb))
1682 null_free_device_storage(nullb->dev, true);
1687 static void null_config_discard(struct nullb *nullb, struct queue_limits *lim)
1689 if (nullb->dev->discard == false)
1692 if (!nullb->dev->memory_backed) {
1693 nullb->dev->discard = false;
1694 pr_info("discard option is ignored without memory backing\n");
1698 if (nullb->dev->zoned) {
1699 nullb->dev->discard = false;
1700 pr_info("discard option is ignored in zoned mode\n");
1704 lim->max_hw_discard_sectors = UINT_MAX >> 9;
1707 static const struct block_device_operations null_ops = {
1708 .owner = THIS_MODULE,
1709 .report_zones = null_report_zones,
1712 static int setup_queues(struct nullb *nullb)
1714 int nqueues = nr_cpu_ids;
1717 nqueues += g_poll_queues;
1719 nullb->queues = kcalloc(nqueues, sizeof(struct nullb_queue),
1727 static int null_init_tag_set(struct blk_mq_tag_set *set, int poll_queues)
1729 set->ops = &null_mq_ops;
1730 set->cmd_size = sizeof(struct nullb_cmd);
1731 set->timeout = 5 * HZ;
1734 set->nr_hw_queues += poll_queues;
1737 return blk_mq_alloc_tag_set(set);
1740 static int null_init_global_tag_set(void)
1747 tag_set.nr_hw_queues = g_submit_queues;
1748 tag_set.queue_depth = g_hw_queue_depth;
1749 tag_set.numa_node = g_home_node;
1750 tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1752 tag_set.flags |= BLK_MQ_F_NO_SCHED;
1753 if (g_shared_tag_bitmap)
1754 tag_set.flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1756 tag_set.flags |= BLK_MQ_F_BLOCKING;
1758 error = null_init_tag_set(&tag_set, g_poll_queues);
1764 static int null_setup_tagset(struct nullb *nullb)
1766 if (nullb->dev->shared_tags) {
1767 nullb->tag_set = &tag_set;
1768 return null_init_global_tag_set();
1771 nullb->tag_set = &nullb->__tag_set;
1772 nullb->tag_set->driver_data = nullb;
1773 nullb->tag_set->nr_hw_queues = nullb->dev->submit_queues;
1774 nullb->tag_set->queue_depth = nullb->dev->hw_queue_depth;
1775 nullb->tag_set->numa_node = nullb->dev->home_node;
1776 nullb->tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
1777 if (nullb->dev->no_sched)
1778 nullb->tag_set->flags |= BLK_MQ_F_NO_SCHED;
1779 if (nullb->dev->shared_tag_bitmap)
1780 nullb->tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1781 if (nullb->dev->blocking)
1782 nullb->tag_set->flags |= BLK_MQ_F_BLOCKING;
1783 return null_init_tag_set(nullb->tag_set, nullb->dev->poll_queues);
1786 static int null_validate_conf(struct nullb_device *dev)
1788 if (dev->queue_mode == NULL_Q_RQ) {
1789 pr_err("legacy IO path is no longer available\n");
1792 if (dev->queue_mode == NULL_Q_BIO) {
1793 pr_err("BIO-based IO path is no longer available, using blk-mq instead.\n");
1794 dev->queue_mode = NULL_Q_MQ;
1797 dev->blocksize = round_down(dev->blocksize, 512);
1798 dev->blocksize = clamp_t(unsigned int, dev->blocksize, 512, 4096);
1800 if (dev->use_per_node_hctx) {
1801 if (dev->submit_queues != nr_online_nodes)
1802 dev->submit_queues = nr_online_nodes;
1803 } else if (dev->submit_queues > nr_cpu_ids)
1804 dev->submit_queues = nr_cpu_ids;
1805 else if (dev->submit_queues == 0)
1806 dev->submit_queues = 1;
1807 dev->prev_submit_queues = dev->submit_queues;
1809 if (dev->poll_queues > g_poll_queues)
1810 dev->poll_queues = g_poll_queues;
1811 dev->prev_poll_queues = dev->poll_queues;
1812 dev->irqmode = min_t(unsigned int, dev->irqmode, NULL_IRQ_TIMER);
1814 /* Do memory allocation, so set blocking */
1815 if (dev->memory_backed)
1816 dev->blocking = true;
1817 else /* cache is meaningless */
1818 dev->cache_size = 0;
1819 dev->cache_size = min_t(unsigned long, ULONG_MAX / 1024 / 1024,
1821 dev->mbps = min_t(unsigned int, 1024 * 40, dev->mbps);
1824 (!dev->zone_size || !is_power_of_2(dev->zone_size))) {
1825 pr_err("zone_size must be power-of-two\n");
1832 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1833 static bool __null_setup_fault(struct fault_attr *attr, char *str)
1838 if (!setup_fault_attr(attr, str))
1846 static bool null_setup_fault(void)
1848 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1849 if (!__null_setup_fault(&null_timeout_attr, g_timeout_str))
1851 if (!__null_setup_fault(&null_requeue_attr, g_requeue_str))
1853 if (!__null_setup_fault(&null_init_hctx_attr, g_init_hctx_str))
1859 static int null_add_dev(struct nullb_device *dev)
1861 struct queue_limits lim = {
1862 .logical_block_size = dev->blocksize,
1863 .physical_block_size = dev->blocksize,
1864 .max_hw_sectors = dev->max_sectors,
1867 struct nullb *nullb;
1870 rv = null_validate_conf(dev);
1874 nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, dev->home_node);
1882 spin_lock_init(&nullb->lock);
1884 rv = setup_queues(nullb);
1886 goto out_free_nullb;
1888 rv = null_setup_tagset(nullb);
1890 goto out_cleanup_queues;
1892 if (dev->virt_boundary)
1893 lim.virt_boundary_mask = PAGE_SIZE - 1;
1894 null_config_discard(nullb, &lim);
1896 rv = null_init_zoned_dev(dev, &lim);
1898 goto out_cleanup_tags;
1901 nullb->disk = blk_mq_alloc_disk(nullb->tag_set, &lim, nullb);
1902 if (IS_ERR(nullb->disk)) {
1903 rv = PTR_ERR(nullb->disk);
1904 goto out_cleanup_zone;
1906 nullb->q = nullb->disk->queue;
1909 set_bit(NULLB_DEV_FL_THROTTLED, &dev->flags);
1910 nullb_setup_bwtimer(nullb);
1913 if (dev->cache_size > 0) {
1914 set_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
1915 blk_queue_write_cache(nullb->q, true, true);
1918 nullb->q->queuedata = nullb;
1919 blk_queue_flag_set(QUEUE_FLAG_NONROT, nullb->q);
1922 rv = ida_alloc(&nullb_indexes, GFP_KERNEL);
1924 mutex_unlock(&lock);
1925 goto out_cleanup_disk;
1929 mutex_unlock(&lock);
1931 if (config_item_name(&dev->group.cg_item)) {
1932 /* Use configfs dir name as the device name */
1933 snprintf(nullb->disk_name, sizeof(nullb->disk_name),
1934 "%s", config_item_name(&dev->group.cg_item));
1936 sprintf(nullb->disk_name, "nullb%d", nullb->index);
1939 set_capacity(nullb->disk,
1940 ((sector_t)nullb->dev->size * SZ_1M) >> SECTOR_SHIFT);
1941 nullb->disk->major = null_major;
1942 nullb->disk->first_minor = nullb->index;
1943 nullb->disk->minors = 1;
1944 nullb->disk->fops = &null_ops;
1945 nullb->disk->private_data = nullb;
1946 strscpy_pad(nullb->disk->disk_name, nullb->disk_name, DISK_NAME_LEN);
1948 if (nullb->dev->zoned) {
1949 rv = null_register_zoned_dev(nullb);
1954 rv = add_disk(nullb->disk);
1959 list_add_tail(&nullb->list, &nullb_list);
1960 mutex_unlock(&lock);
1962 pr_info("disk %s created\n", nullb->disk_name);
1967 ida_free(&nullb_indexes, nullb->index);
1969 put_disk(nullb->disk);
1971 null_free_zoned_dev(dev);
1973 if (nullb->tag_set == &nullb->__tag_set)
1974 blk_mq_free_tag_set(nullb->tag_set);
1976 kfree(nullb->queues);
1984 static struct nullb *null_find_dev_by_name(const char *name)
1986 struct nullb *nullb = NULL, *nb;
1989 list_for_each_entry(nb, &nullb_list, list) {
1990 if (strcmp(nb->disk_name, name) == 0) {
1995 mutex_unlock(&lock);
2000 static int null_create_dev(void)
2002 struct nullb_device *dev;
2005 dev = null_alloc_dev();
2009 ret = null_add_dev(dev);
2018 static void null_destroy_dev(struct nullb *nullb)
2020 struct nullb_device *dev = nullb->dev;
2022 null_del_dev(nullb);
2023 null_free_device_storage(dev, false);
2027 static int __init null_init(void)
2031 struct nullb *nullb;
2033 if (g_bs > PAGE_SIZE) {
2034 pr_warn("invalid block size\n");
2035 pr_warn("defaults block size to %lu\n", PAGE_SIZE);
2039 if (g_home_node != NUMA_NO_NODE && g_home_node >= nr_online_nodes) {
2040 pr_err("invalid home_node value\n");
2041 g_home_node = NUMA_NO_NODE;
2044 if (!null_setup_fault())
2047 if (g_queue_mode == NULL_Q_RQ) {
2048 pr_err("legacy IO path is no longer available\n");
2052 if (g_use_per_node_hctx) {
2053 if (g_submit_queues != nr_online_nodes) {
2054 pr_warn("submit_queues param is set to %u.\n",
2056 g_submit_queues = nr_online_nodes;
2058 } else if (g_submit_queues > nr_cpu_ids) {
2059 g_submit_queues = nr_cpu_ids;
2060 } else if (g_submit_queues <= 0) {
2061 g_submit_queues = 1;
2064 config_group_init(&nullb_subsys.su_group);
2065 mutex_init(&nullb_subsys.su_mutex);
2067 ret = configfs_register_subsystem(&nullb_subsys);
2073 null_major = register_blkdev(0, "nullb");
2074 if (null_major < 0) {
2079 for (i = 0; i < nr_devices; i++) {
2080 ret = null_create_dev();
2085 pr_info("module loaded\n");
2089 while (!list_empty(&nullb_list)) {
2090 nullb = list_entry(nullb_list.next, struct nullb, list);
2091 null_destroy_dev(nullb);
2093 unregister_blkdev(null_major, "nullb");
2095 configfs_unregister_subsystem(&nullb_subsys);
2099 static void __exit null_exit(void)
2101 struct nullb *nullb;
2103 configfs_unregister_subsystem(&nullb_subsys);
2105 unregister_blkdev(null_major, "nullb");
2108 while (!list_empty(&nullb_list)) {
2109 nullb = list_entry(nullb_list.next, struct nullb, list);
2110 null_destroy_dev(nullb);
2112 mutex_unlock(&lock);
2115 blk_mq_free_tag_set(&tag_set);
2118 module_init(null_init);
2119 module_exit(null_exit);
2121 MODULE_AUTHOR("Jens Axboe <axboe@kernel.dk>");
2122 MODULE_LICENSE("GPL");