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>
14 #define PAGE_SECTORS_SHIFT (PAGE_SHIFT - SECTOR_SHIFT)
15 #define PAGE_SECTORS (1 << PAGE_SECTORS_SHIFT)
16 #define SECTOR_MASK (PAGE_SECTORS - 1)
20 #define TICKS_PER_SEC 50ULL
21 #define TIMER_INTERVAL (NSEC_PER_SEC / TICKS_PER_SEC)
23 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
24 static DECLARE_FAULT_ATTR(null_timeout_attr);
25 static DECLARE_FAULT_ATTR(null_requeue_attr);
26 static DECLARE_FAULT_ATTR(null_init_hctx_attr);
29 static inline u64 mb_per_tick(int mbps)
31 return (1 << 20) / TICKS_PER_SEC * ((u64) mbps);
35 * Status flags for nullb_device.
37 * CONFIGURED: Device has been configured and turned on. Cannot reconfigure.
38 * UP: Device is currently on and visible in userspace.
39 * THROTTLED: Device is being throttled.
40 * CACHE: Device is using a write-back cache.
42 enum nullb_device_flags {
43 NULLB_DEV_FL_CONFIGURED = 0,
45 NULLB_DEV_FL_THROTTLED = 2,
46 NULLB_DEV_FL_CACHE = 3,
49 #define MAP_SZ ((PAGE_SIZE >> SECTOR_SHIFT) + 2)
51 * nullb_page is a page in memory for nullb devices.
53 * @page: The page holding the data.
54 * @bitmap: The bitmap represents which sector in the page has data.
55 * Each bit represents one block size. For example, sector 8
56 * will use the 7th bit
57 * The highest 2 bits of bitmap are for special purpose. LOCK means the cache
58 * page is being flushing to storage. FREE means the cache page is freed and
59 * should be skipped from flushing to storage. Please see
60 * null_make_cache_space
64 DECLARE_BITMAP(bitmap, MAP_SZ);
66 #define NULLB_PAGE_LOCK (MAP_SZ - 1)
67 #define NULLB_PAGE_FREE (MAP_SZ - 2)
69 static LIST_HEAD(nullb_list);
70 static struct mutex lock;
71 static int null_major;
72 static DEFINE_IDA(nullb_indexes);
73 static struct blk_mq_tag_set tag_set;
87 static int g_no_sched;
88 module_param_named(no_sched, g_no_sched, int, 0444);
89 MODULE_PARM_DESC(no_sched, "No io scheduler");
91 static int g_submit_queues = 1;
92 module_param_named(submit_queues, g_submit_queues, int, 0444);
93 MODULE_PARM_DESC(submit_queues, "Number of submission queues");
95 static int g_home_node = NUMA_NO_NODE;
96 module_param_named(home_node, g_home_node, int, 0444);
97 MODULE_PARM_DESC(home_node, "Home node for the device");
99 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
101 * For more details about fault injection, please refer to
102 * Documentation/fault-injection/fault-injection.rst.
104 static char g_timeout_str[80];
105 module_param_string(timeout, g_timeout_str, sizeof(g_timeout_str), 0444);
106 MODULE_PARM_DESC(timeout, "Fault injection. timeout=<interval>,<probability>,<space>,<times>");
108 static char g_requeue_str[80];
109 module_param_string(requeue, g_requeue_str, sizeof(g_requeue_str), 0444);
110 MODULE_PARM_DESC(requeue, "Fault injection. requeue=<interval>,<probability>,<space>,<times>");
112 static char g_init_hctx_str[80];
113 module_param_string(init_hctx, g_init_hctx_str, sizeof(g_init_hctx_str), 0444);
114 MODULE_PARM_DESC(init_hctx, "Fault injection to fail hctx init. init_hctx=<interval>,<probability>,<space>,<times>");
117 static int g_queue_mode = NULL_Q_MQ;
119 static int null_param_store_val(const char *str, int *val, int min, int max)
123 ret = kstrtoint(str, 10, &new_val);
127 if (new_val < min || new_val > max)
134 static int null_set_queue_mode(const char *str, const struct kernel_param *kp)
136 return null_param_store_val(str, &g_queue_mode, NULL_Q_BIO, NULL_Q_MQ);
139 static const struct kernel_param_ops null_queue_mode_param_ops = {
140 .set = null_set_queue_mode,
141 .get = param_get_int,
144 device_param_cb(queue_mode, &null_queue_mode_param_ops, &g_queue_mode, 0444);
145 MODULE_PARM_DESC(queue_mode, "Block interface to use (0=bio,1=rq,2=multiqueue)");
147 static int g_gb = 250;
148 module_param_named(gb, g_gb, int, 0444);
149 MODULE_PARM_DESC(gb, "Size in GB");
151 static int g_bs = 512;
152 module_param_named(bs, g_bs, int, 0444);
153 MODULE_PARM_DESC(bs, "Block size (in bytes)");
155 static unsigned int nr_devices = 1;
156 module_param(nr_devices, uint, 0444);
157 MODULE_PARM_DESC(nr_devices, "Number of devices to register");
159 static bool g_blocking;
160 module_param_named(blocking, g_blocking, bool, 0444);
161 MODULE_PARM_DESC(blocking, "Register as a blocking blk-mq driver device");
163 static bool shared_tags;
164 module_param(shared_tags, bool, 0444);
165 MODULE_PARM_DESC(shared_tags, "Share tag set between devices for blk-mq");
167 static bool g_shared_tag_bitmap;
168 module_param_named(shared_tag_bitmap, g_shared_tag_bitmap, bool, 0444);
169 MODULE_PARM_DESC(shared_tag_bitmap, "Use shared tag bitmap for all submission queues for blk-mq");
171 static int g_irqmode = NULL_IRQ_SOFTIRQ;
173 static int null_set_irqmode(const char *str, const struct kernel_param *kp)
175 return null_param_store_val(str, &g_irqmode, NULL_IRQ_NONE,
179 static const struct kernel_param_ops null_irqmode_param_ops = {
180 .set = null_set_irqmode,
181 .get = param_get_int,
184 device_param_cb(irqmode, &null_irqmode_param_ops, &g_irqmode, 0444);
185 MODULE_PARM_DESC(irqmode, "IRQ completion handler. 0-none, 1-softirq, 2-timer");
187 static unsigned long g_completion_nsec = 10000;
188 module_param_named(completion_nsec, g_completion_nsec, ulong, 0444);
189 MODULE_PARM_DESC(completion_nsec, "Time in ns to complete a request in hardware. Default: 10,000ns");
191 static int g_hw_queue_depth = 64;
192 module_param_named(hw_queue_depth, g_hw_queue_depth, int, 0444);
193 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 64");
195 static bool g_use_per_node_hctx;
196 module_param_named(use_per_node_hctx, g_use_per_node_hctx, bool, 0444);
197 MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware context queues. Default: false");
200 module_param_named(zoned, g_zoned, bool, S_IRUGO);
201 MODULE_PARM_DESC(zoned, "Make device as a host-managed zoned block device. Default: false");
203 static unsigned long g_zone_size = 256;
204 module_param_named(zone_size, g_zone_size, ulong, S_IRUGO);
205 MODULE_PARM_DESC(zone_size, "Zone size in MB when block device is zoned. Must be power-of-two: Default: 256");
207 static unsigned long g_zone_capacity;
208 module_param_named(zone_capacity, g_zone_capacity, ulong, 0444);
209 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");
211 static unsigned int g_zone_nr_conv;
212 module_param_named(zone_nr_conv, g_zone_nr_conv, uint, 0444);
213 MODULE_PARM_DESC(zone_nr_conv, "Number of conventional zones when block device is zoned. Default: 0");
215 static unsigned int g_zone_max_open;
216 module_param_named(zone_max_open, g_zone_max_open, uint, 0444);
217 MODULE_PARM_DESC(zone_max_open, "Maximum number of open zones when block device is zoned. Default: 0 (no limit)");
219 static unsigned int g_zone_max_active;
220 module_param_named(zone_max_active, g_zone_max_active, uint, 0444);
221 MODULE_PARM_DESC(zone_max_active, "Maximum number of active zones when block device is zoned. Default: 0 (no limit)");
223 static struct nullb_device *null_alloc_dev(void);
224 static void null_free_dev(struct nullb_device *dev);
225 static void null_del_dev(struct nullb *nullb);
226 static int null_add_dev(struct nullb_device *dev);
227 static void null_free_device_storage(struct nullb_device *dev, bool is_cache);
229 static inline struct nullb_device *to_nullb_device(struct config_item *item)
231 return item ? container_of(item, struct nullb_device, item) : NULL;
234 static inline ssize_t nullb_device_uint_attr_show(unsigned int val, char *page)
236 return snprintf(page, PAGE_SIZE, "%u\n", val);
239 static inline ssize_t nullb_device_ulong_attr_show(unsigned long val,
242 return snprintf(page, PAGE_SIZE, "%lu\n", val);
245 static inline ssize_t nullb_device_bool_attr_show(bool val, char *page)
247 return snprintf(page, PAGE_SIZE, "%u\n", val);
250 static ssize_t nullb_device_uint_attr_store(unsigned int *val,
251 const char *page, size_t count)
256 result = kstrtouint(page, 0, &tmp);
264 static ssize_t nullb_device_ulong_attr_store(unsigned long *val,
265 const char *page, size_t count)
270 result = kstrtoul(page, 0, &tmp);
278 static ssize_t nullb_device_bool_attr_store(bool *val, const char *page,
284 result = kstrtobool(page, &tmp);
292 /* The following macro should only be used with TYPE = {uint, ulong, bool}. */
293 #define NULLB_DEVICE_ATTR(NAME, TYPE, APPLY) \
295 nullb_device_##NAME##_show(struct config_item *item, char *page) \
297 return nullb_device_##TYPE##_attr_show( \
298 to_nullb_device(item)->NAME, page); \
301 nullb_device_##NAME##_store(struct config_item *item, const char *page, \
304 int (*apply_fn)(struct nullb_device *dev, TYPE new_value) = APPLY;\
305 struct nullb_device *dev = to_nullb_device(item); \
306 TYPE new_value = 0; \
309 ret = nullb_device_##TYPE##_attr_store(&new_value, page, count);\
313 ret = apply_fn(dev, new_value); \
314 else if (test_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags)) \
318 dev->NAME = new_value; \
321 CONFIGFS_ATTR(nullb_device_, NAME);
323 static int nullb_apply_submit_queues(struct nullb_device *dev,
324 unsigned int submit_queues)
326 struct nullb *nullb = dev->nullb;
327 struct blk_mq_tag_set *set;
333 * Make sure that null_init_hctx() does not access nullb->queues[] past
334 * the end of that array.
336 if (submit_queues > nr_cpu_ids)
338 set = nullb->tag_set;
339 blk_mq_update_nr_hw_queues(set, submit_queues);
340 return set->nr_hw_queues == submit_queues ? 0 : -ENOMEM;
343 NULLB_DEVICE_ATTR(size, ulong, NULL);
344 NULLB_DEVICE_ATTR(completion_nsec, ulong, NULL);
345 NULLB_DEVICE_ATTR(submit_queues, uint, nullb_apply_submit_queues);
346 NULLB_DEVICE_ATTR(home_node, uint, NULL);
347 NULLB_DEVICE_ATTR(queue_mode, uint, NULL);
348 NULLB_DEVICE_ATTR(blocksize, uint, NULL);
349 NULLB_DEVICE_ATTR(irqmode, uint, NULL);
350 NULLB_DEVICE_ATTR(hw_queue_depth, uint, NULL);
351 NULLB_DEVICE_ATTR(index, uint, NULL);
352 NULLB_DEVICE_ATTR(blocking, bool, NULL);
353 NULLB_DEVICE_ATTR(use_per_node_hctx, bool, NULL);
354 NULLB_DEVICE_ATTR(memory_backed, bool, NULL);
355 NULLB_DEVICE_ATTR(discard, bool, NULL);
356 NULLB_DEVICE_ATTR(mbps, uint, NULL);
357 NULLB_DEVICE_ATTR(cache_size, ulong, NULL);
358 NULLB_DEVICE_ATTR(zoned, bool, NULL);
359 NULLB_DEVICE_ATTR(zone_size, ulong, NULL);
360 NULLB_DEVICE_ATTR(zone_capacity, ulong, NULL);
361 NULLB_DEVICE_ATTR(zone_nr_conv, uint, NULL);
362 NULLB_DEVICE_ATTR(zone_max_open, uint, NULL);
363 NULLB_DEVICE_ATTR(zone_max_active, uint, NULL);
365 static ssize_t nullb_device_power_show(struct config_item *item, char *page)
367 return nullb_device_bool_attr_show(to_nullb_device(item)->power, page);
370 static ssize_t nullb_device_power_store(struct config_item *item,
371 const char *page, size_t count)
373 struct nullb_device *dev = to_nullb_device(item);
377 ret = nullb_device_bool_attr_store(&newp, page, count);
381 if (!dev->power && newp) {
382 if (test_and_set_bit(NULLB_DEV_FL_UP, &dev->flags))
384 if (null_add_dev(dev)) {
385 clear_bit(NULLB_DEV_FL_UP, &dev->flags);
389 set_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
391 } else if (dev->power && !newp) {
392 if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
395 null_del_dev(dev->nullb);
398 clear_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
404 CONFIGFS_ATTR(nullb_device_, power);
406 static ssize_t nullb_device_badblocks_show(struct config_item *item, char *page)
408 struct nullb_device *t_dev = to_nullb_device(item);
410 return badblocks_show(&t_dev->badblocks, page, 0);
413 static ssize_t nullb_device_badblocks_store(struct config_item *item,
414 const char *page, size_t count)
416 struct nullb_device *t_dev = to_nullb_device(item);
417 char *orig, *buf, *tmp;
421 orig = kstrndup(page, count, GFP_KERNEL);
425 buf = strstrip(orig);
428 if (buf[0] != '+' && buf[0] != '-')
430 tmp = strchr(&buf[1], '-');
434 ret = kstrtoull(buf + 1, 0, &start);
437 ret = kstrtoull(tmp + 1, 0, &end);
443 /* enable badblocks */
444 cmpxchg(&t_dev->badblocks.shift, -1, 0);
446 ret = badblocks_set(&t_dev->badblocks, start,
449 ret = badblocks_clear(&t_dev->badblocks, start,
457 CONFIGFS_ATTR(nullb_device_, badblocks);
459 static struct configfs_attribute *nullb_device_attrs[] = {
460 &nullb_device_attr_size,
461 &nullb_device_attr_completion_nsec,
462 &nullb_device_attr_submit_queues,
463 &nullb_device_attr_home_node,
464 &nullb_device_attr_queue_mode,
465 &nullb_device_attr_blocksize,
466 &nullb_device_attr_irqmode,
467 &nullb_device_attr_hw_queue_depth,
468 &nullb_device_attr_index,
469 &nullb_device_attr_blocking,
470 &nullb_device_attr_use_per_node_hctx,
471 &nullb_device_attr_power,
472 &nullb_device_attr_memory_backed,
473 &nullb_device_attr_discard,
474 &nullb_device_attr_mbps,
475 &nullb_device_attr_cache_size,
476 &nullb_device_attr_badblocks,
477 &nullb_device_attr_zoned,
478 &nullb_device_attr_zone_size,
479 &nullb_device_attr_zone_capacity,
480 &nullb_device_attr_zone_nr_conv,
481 &nullb_device_attr_zone_max_open,
482 &nullb_device_attr_zone_max_active,
486 static void nullb_device_release(struct config_item *item)
488 struct nullb_device *dev = to_nullb_device(item);
490 null_free_device_storage(dev, false);
494 static struct configfs_item_operations nullb_device_ops = {
495 .release = nullb_device_release,
498 static const struct config_item_type nullb_device_type = {
499 .ct_item_ops = &nullb_device_ops,
500 .ct_attrs = nullb_device_attrs,
501 .ct_owner = THIS_MODULE,
505 config_item *nullb_group_make_item(struct config_group *group, const char *name)
507 struct nullb_device *dev;
509 dev = null_alloc_dev();
511 return ERR_PTR(-ENOMEM);
513 config_item_init_type_name(&dev->item, name, &nullb_device_type);
519 nullb_group_drop_item(struct config_group *group, struct config_item *item)
521 struct nullb_device *dev = to_nullb_device(item);
523 if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
526 null_del_dev(dev->nullb);
530 config_item_put(item);
533 static ssize_t memb_group_features_show(struct config_item *item, char *page)
535 return snprintf(page, PAGE_SIZE,
536 "memory_backed,discard,bandwidth,cache,badblocks,zoned,zone_size,zone_capacity,zone_nr_conv,zone_max_open,zone_max_active\n");
539 CONFIGFS_ATTR_RO(memb_group_, features);
541 static struct configfs_attribute *nullb_group_attrs[] = {
542 &memb_group_attr_features,
546 static struct configfs_group_operations nullb_group_ops = {
547 .make_item = nullb_group_make_item,
548 .drop_item = nullb_group_drop_item,
551 static const struct config_item_type nullb_group_type = {
552 .ct_group_ops = &nullb_group_ops,
553 .ct_attrs = nullb_group_attrs,
554 .ct_owner = THIS_MODULE,
557 static struct configfs_subsystem nullb_subsys = {
560 .ci_namebuf = "nullb",
561 .ci_type = &nullb_group_type,
566 static inline int null_cache_active(struct nullb *nullb)
568 return test_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
571 static struct nullb_device *null_alloc_dev(void)
573 struct nullb_device *dev;
575 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
578 INIT_RADIX_TREE(&dev->data, GFP_ATOMIC);
579 INIT_RADIX_TREE(&dev->cache, GFP_ATOMIC);
580 if (badblocks_init(&dev->badblocks, 0)) {
585 dev->size = g_gb * 1024;
586 dev->completion_nsec = g_completion_nsec;
587 dev->submit_queues = g_submit_queues;
588 dev->home_node = g_home_node;
589 dev->queue_mode = g_queue_mode;
590 dev->blocksize = g_bs;
591 dev->irqmode = g_irqmode;
592 dev->hw_queue_depth = g_hw_queue_depth;
593 dev->blocking = g_blocking;
594 dev->use_per_node_hctx = g_use_per_node_hctx;
595 dev->zoned = g_zoned;
596 dev->zone_size = g_zone_size;
597 dev->zone_capacity = g_zone_capacity;
598 dev->zone_nr_conv = g_zone_nr_conv;
599 dev->zone_max_open = g_zone_max_open;
600 dev->zone_max_active = g_zone_max_active;
604 static void null_free_dev(struct nullb_device *dev)
609 null_free_zoned_dev(dev);
610 badblocks_exit(&dev->badblocks);
614 static void put_tag(struct nullb_queue *nq, unsigned int tag)
616 clear_bit_unlock(tag, nq->tag_map);
618 if (waitqueue_active(&nq->wait))
622 static unsigned int get_tag(struct nullb_queue *nq)
627 tag = find_first_zero_bit(nq->tag_map, nq->queue_depth);
628 if (tag >= nq->queue_depth)
630 } while (test_and_set_bit_lock(tag, nq->tag_map));
635 static void free_cmd(struct nullb_cmd *cmd)
637 put_tag(cmd->nq, cmd->tag);
640 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer);
642 static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq)
644 struct nullb_cmd *cmd;
649 cmd = &nq->cmds[tag];
651 cmd->error = BLK_STS_OK;
653 if (nq->dev->irqmode == NULL_IRQ_TIMER) {
654 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC,
656 cmd->timer.function = null_cmd_timer_expired;
664 static struct nullb_cmd *alloc_cmd(struct nullb_queue *nq, int can_wait)
666 struct nullb_cmd *cmd;
669 cmd = __alloc_cmd(nq);
670 if (cmd || !can_wait)
674 prepare_to_wait(&nq->wait, &wait, TASK_UNINTERRUPTIBLE);
675 cmd = __alloc_cmd(nq);
682 finish_wait(&nq->wait, &wait);
686 static void end_cmd(struct nullb_cmd *cmd)
688 int queue_mode = cmd->nq->dev->queue_mode;
690 switch (queue_mode) {
692 blk_mq_end_request(cmd->rq, cmd->error);
695 cmd->bio->bi_status = cmd->error;
703 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
705 end_cmd(container_of(timer, struct nullb_cmd, timer));
707 return HRTIMER_NORESTART;
710 static void null_cmd_end_timer(struct nullb_cmd *cmd)
712 ktime_t kt = cmd->nq->dev->completion_nsec;
714 hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL);
717 static void null_complete_rq(struct request *rq)
719 end_cmd(blk_mq_rq_to_pdu(rq));
722 static struct nullb_page *null_alloc_page(gfp_t gfp_flags)
724 struct nullb_page *t_page;
726 t_page = kmalloc(sizeof(struct nullb_page), gfp_flags);
730 t_page->page = alloc_pages(gfp_flags, 0);
734 memset(t_page->bitmap, 0, sizeof(t_page->bitmap));
742 static void null_free_page(struct nullb_page *t_page)
744 __set_bit(NULLB_PAGE_FREE, t_page->bitmap);
745 if (test_bit(NULLB_PAGE_LOCK, t_page->bitmap))
747 __free_page(t_page->page);
751 static bool null_page_empty(struct nullb_page *page)
753 int size = MAP_SZ - 2;
755 return find_first_bit(page->bitmap, size) == size;
758 static void null_free_sector(struct nullb *nullb, sector_t sector,
761 unsigned int sector_bit;
763 struct nullb_page *t_page, *ret;
764 struct radix_tree_root *root;
766 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
767 idx = sector >> PAGE_SECTORS_SHIFT;
768 sector_bit = (sector & SECTOR_MASK);
770 t_page = radix_tree_lookup(root, idx);
772 __clear_bit(sector_bit, t_page->bitmap);
774 if (null_page_empty(t_page)) {
775 ret = radix_tree_delete_item(root, idx, t_page);
776 WARN_ON(ret != t_page);
779 nullb->dev->curr_cache -= PAGE_SIZE;
784 static struct nullb_page *null_radix_tree_insert(struct nullb *nullb, u64 idx,
785 struct nullb_page *t_page, bool is_cache)
787 struct radix_tree_root *root;
789 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
791 if (radix_tree_insert(root, idx, t_page)) {
792 null_free_page(t_page);
793 t_page = radix_tree_lookup(root, idx);
794 WARN_ON(!t_page || t_page->page->index != idx);
796 nullb->dev->curr_cache += PAGE_SIZE;
801 static void null_free_device_storage(struct nullb_device *dev, bool is_cache)
803 unsigned long pos = 0;
805 struct nullb_page *ret, *t_pages[FREE_BATCH];
806 struct radix_tree_root *root;
808 root = is_cache ? &dev->cache : &dev->data;
813 nr_pages = radix_tree_gang_lookup(root,
814 (void **)t_pages, pos, FREE_BATCH);
816 for (i = 0; i < nr_pages; i++) {
817 pos = t_pages[i]->page->index;
818 ret = radix_tree_delete_item(root, pos, t_pages[i]);
819 WARN_ON(ret != t_pages[i]);
824 } while (nr_pages == FREE_BATCH);
830 static struct nullb_page *__null_lookup_page(struct nullb *nullb,
831 sector_t sector, bool for_write, bool is_cache)
833 unsigned int sector_bit;
835 struct nullb_page *t_page;
836 struct radix_tree_root *root;
838 idx = sector >> PAGE_SECTORS_SHIFT;
839 sector_bit = (sector & SECTOR_MASK);
841 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
842 t_page = radix_tree_lookup(root, idx);
843 WARN_ON(t_page && t_page->page->index != idx);
845 if (t_page && (for_write || test_bit(sector_bit, t_page->bitmap)))
851 static struct nullb_page *null_lookup_page(struct nullb *nullb,
852 sector_t sector, bool for_write, bool ignore_cache)
854 struct nullb_page *page = NULL;
857 page = __null_lookup_page(nullb, sector, for_write, true);
860 return __null_lookup_page(nullb, sector, for_write, false);
863 static struct nullb_page *null_insert_page(struct nullb *nullb,
864 sector_t sector, bool ignore_cache)
865 __releases(&nullb->lock)
866 __acquires(&nullb->lock)
869 struct nullb_page *t_page;
871 t_page = null_lookup_page(nullb, sector, true, ignore_cache);
875 spin_unlock_irq(&nullb->lock);
877 t_page = null_alloc_page(GFP_NOIO);
881 if (radix_tree_preload(GFP_NOIO))
884 spin_lock_irq(&nullb->lock);
885 idx = sector >> PAGE_SECTORS_SHIFT;
886 t_page->page->index = idx;
887 t_page = null_radix_tree_insert(nullb, idx, t_page, !ignore_cache);
888 radix_tree_preload_end();
892 null_free_page(t_page);
894 spin_lock_irq(&nullb->lock);
895 return null_lookup_page(nullb, sector, true, ignore_cache);
898 static int null_flush_cache_page(struct nullb *nullb, struct nullb_page *c_page)
903 struct nullb_page *t_page, *ret;
906 idx = c_page->page->index;
908 t_page = null_insert_page(nullb, idx << PAGE_SECTORS_SHIFT, true);
910 __clear_bit(NULLB_PAGE_LOCK, c_page->bitmap);
911 if (test_bit(NULLB_PAGE_FREE, c_page->bitmap)) {
912 null_free_page(c_page);
913 if (t_page && null_page_empty(t_page)) {
914 ret = radix_tree_delete_item(&nullb->dev->data,
916 null_free_page(t_page);
924 src = kmap_atomic(c_page->page);
925 dst = kmap_atomic(t_page->page);
927 for (i = 0; i < PAGE_SECTORS;
928 i += (nullb->dev->blocksize >> SECTOR_SHIFT)) {
929 if (test_bit(i, c_page->bitmap)) {
930 offset = (i << SECTOR_SHIFT);
931 memcpy(dst + offset, src + offset,
932 nullb->dev->blocksize);
933 __set_bit(i, t_page->bitmap);
940 ret = radix_tree_delete_item(&nullb->dev->cache, idx, c_page);
942 nullb->dev->curr_cache -= PAGE_SIZE;
947 static int null_make_cache_space(struct nullb *nullb, unsigned long n)
949 int i, err, nr_pages;
950 struct nullb_page *c_pages[FREE_BATCH];
951 unsigned long flushed = 0, one_round;
954 if ((nullb->dev->cache_size * 1024 * 1024) >
955 nullb->dev->curr_cache + n || nullb->dev->curr_cache == 0)
958 nr_pages = radix_tree_gang_lookup(&nullb->dev->cache,
959 (void **)c_pages, nullb->cache_flush_pos, FREE_BATCH);
961 * nullb_flush_cache_page could unlock before using the c_pages. To
962 * avoid race, we don't allow page free
964 for (i = 0; i < nr_pages; i++) {
965 nullb->cache_flush_pos = c_pages[i]->page->index;
967 * We found the page which is being flushed to disk by other
970 if (test_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap))
973 __set_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap);
977 for (i = 0; i < nr_pages; i++) {
978 if (c_pages[i] == NULL)
980 err = null_flush_cache_page(nullb, c_pages[i]);
985 flushed += one_round << PAGE_SHIFT;
989 nullb->cache_flush_pos = 0;
990 if (one_round == 0) {
991 /* give other threads a chance */
992 spin_unlock_irq(&nullb->lock);
993 spin_lock_irq(&nullb->lock);
1000 static int copy_to_nullb(struct nullb *nullb, struct page *source,
1001 unsigned int off, sector_t sector, size_t n, bool is_fua)
1003 size_t temp, count = 0;
1004 unsigned int offset;
1005 struct nullb_page *t_page;
1009 temp = min_t(size_t, nullb->dev->blocksize, n - count);
1011 if (null_cache_active(nullb) && !is_fua)
1012 null_make_cache_space(nullb, PAGE_SIZE);
1014 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
1015 t_page = null_insert_page(nullb, sector,
1016 !null_cache_active(nullb) || is_fua);
1020 src = kmap_atomic(source);
1021 dst = kmap_atomic(t_page->page);
1022 memcpy(dst + offset, src + off + count, temp);
1026 __set_bit(sector & SECTOR_MASK, t_page->bitmap);
1029 null_free_sector(nullb, sector, true);
1032 sector += temp >> SECTOR_SHIFT;
1037 static int copy_from_nullb(struct nullb *nullb, struct page *dest,
1038 unsigned int off, sector_t sector, size_t n)
1040 size_t temp, count = 0;
1041 unsigned int offset;
1042 struct nullb_page *t_page;
1046 temp = min_t(size_t, nullb->dev->blocksize, n - count);
1048 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
1049 t_page = null_lookup_page(nullb, sector, false,
1050 !null_cache_active(nullb));
1052 dst = kmap_atomic(dest);
1054 memset(dst + off + count, 0, temp);
1057 src = kmap_atomic(t_page->page);
1058 memcpy(dst + off + count, src + offset, temp);
1064 sector += temp >> SECTOR_SHIFT;
1069 static void nullb_fill_pattern(struct nullb *nullb, struct page *page,
1070 unsigned int len, unsigned int off)
1074 dst = kmap_atomic(page);
1075 memset(dst + off, 0xFF, len);
1079 static void null_handle_discard(struct nullb *nullb, sector_t sector, size_t n)
1083 spin_lock_irq(&nullb->lock);
1085 temp = min_t(size_t, n, nullb->dev->blocksize);
1086 null_free_sector(nullb, sector, false);
1087 if (null_cache_active(nullb))
1088 null_free_sector(nullb, sector, true);
1089 sector += temp >> SECTOR_SHIFT;
1092 spin_unlock_irq(&nullb->lock);
1095 static int null_handle_flush(struct nullb *nullb)
1099 if (!null_cache_active(nullb))
1102 spin_lock_irq(&nullb->lock);
1104 err = null_make_cache_space(nullb,
1105 nullb->dev->cache_size * 1024 * 1024);
1106 if (err || nullb->dev->curr_cache == 0)
1110 WARN_ON(!radix_tree_empty(&nullb->dev->cache));
1111 spin_unlock_irq(&nullb->lock);
1115 static int null_transfer(struct nullb *nullb, struct page *page,
1116 unsigned int len, unsigned int off, bool is_write, sector_t sector,
1119 struct nullb_device *dev = nullb->dev;
1120 unsigned int valid_len = len;
1125 valid_len = null_zone_valid_read_len(nullb,
1129 err = copy_from_nullb(nullb, page, off,
1136 nullb_fill_pattern(nullb, page, len, off);
1137 flush_dcache_page(page);
1139 flush_dcache_page(page);
1140 err = copy_to_nullb(nullb, page, off, sector, len, is_fua);
1146 static int null_handle_rq(struct nullb_cmd *cmd)
1148 struct request *rq = cmd->rq;
1149 struct nullb *nullb = cmd->nq->dev->nullb;
1153 struct req_iterator iter;
1154 struct bio_vec bvec;
1156 sector = blk_rq_pos(rq);
1158 if (req_op(rq) == REQ_OP_DISCARD) {
1159 null_handle_discard(nullb, sector, blk_rq_bytes(rq));
1163 spin_lock_irq(&nullb->lock);
1164 rq_for_each_segment(bvec, rq, iter) {
1166 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1167 op_is_write(req_op(rq)), sector,
1168 rq->cmd_flags & REQ_FUA);
1170 spin_unlock_irq(&nullb->lock);
1173 sector += len >> SECTOR_SHIFT;
1175 spin_unlock_irq(&nullb->lock);
1180 static int null_handle_bio(struct nullb_cmd *cmd)
1182 struct bio *bio = cmd->bio;
1183 struct nullb *nullb = cmd->nq->dev->nullb;
1187 struct bio_vec bvec;
1188 struct bvec_iter iter;
1190 sector = bio->bi_iter.bi_sector;
1192 if (bio_op(bio) == REQ_OP_DISCARD) {
1193 null_handle_discard(nullb, sector,
1194 bio_sectors(bio) << SECTOR_SHIFT);
1198 spin_lock_irq(&nullb->lock);
1199 bio_for_each_segment(bvec, bio, iter) {
1201 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1202 op_is_write(bio_op(bio)), sector,
1203 bio->bi_opf & REQ_FUA);
1205 spin_unlock_irq(&nullb->lock);
1208 sector += len >> SECTOR_SHIFT;
1210 spin_unlock_irq(&nullb->lock);
1214 static void null_stop_queue(struct nullb *nullb)
1216 struct request_queue *q = nullb->q;
1218 if (nullb->dev->queue_mode == NULL_Q_MQ)
1219 blk_mq_stop_hw_queues(q);
1222 static void null_restart_queue_async(struct nullb *nullb)
1224 struct request_queue *q = nullb->q;
1226 if (nullb->dev->queue_mode == NULL_Q_MQ)
1227 blk_mq_start_stopped_hw_queues(q, true);
1230 static inline blk_status_t null_handle_throttled(struct nullb_cmd *cmd)
1232 struct nullb_device *dev = cmd->nq->dev;
1233 struct nullb *nullb = dev->nullb;
1234 blk_status_t sts = BLK_STS_OK;
1235 struct request *rq = cmd->rq;
1237 if (!hrtimer_active(&nullb->bw_timer))
1238 hrtimer_restart(&nullb->bw_timer);
1240 if (atomic_long_sub_return(blk_rq_bytes(rq), &nullb->cur_bytes) < 0) {
1241 null_stop_queue(nullb);
1242 /* race with timer */
1243 if (atomic_long_read(&nullb->cur_bytes) > 0)
1244 null_restart_queue_async(nullb);
1245 /* requeue request */
1246 sts = BLK_STS_DEV_RESOURCE;
1251 static inline blk_status_t null_handle_badblocks(struct nullb_cmd *cmd,
1253 sector_t nr_sectors)
1255 struct badblocks *bb = &cmd->nq->dev->badblocks;
1259 if (badblocks_check(bb, sector, nr_sectors, &first_bad, &bad_sectors))
1260 return BLK_STS_IOERR;
1265 static inline blk_status_t null_handle_memory_backed(struct nullb_cmd *cmd,
1268 struct nullb_device *dev = cmd->nq->dev;
1271 if (dev->queue_mode == NULL_Q_BIO)
1272 err = null_handle_bio(cmd);
1274 err = null_handle_rq(cmd);
1276 return errno_to_blk_status(err);
1279 static void nullb_zero_read_cmd_buffer(struct nullb_cmd *cmd)
1281 struct nullb_device *dev = cmd->nq->dev;
1284 if (dev->memory_backed)
1287 if (dev->queue_mode == NULL_Q_BIO && bio_op(cmd->bio) == REQ_OP_READ) {
1288 zero_fill_bio(cmd->bio);
1289 } else if (req_op(cmd->rq) == REQ_OP_READ) {
1290 __rq_for_each_bio(bio, cmd->rq)
1295 static inline void nullb_complete_cmd(struct nullb_cmd *cmd)
1298 * Since root privileges are required to configure the null_blk
1299 * driver, it is fine that this driver does not initialize the
1300 * data buffers of read commands. Zero-initialize these buffers
1301 * anyway if KMSAN is enabled to prevent that KMSAN complains
1302 * about null_blk not initializing read data buffers.
1304 if (IS_ENABLED(CONFIG_KMSAN))
1305 nullb_zero_read_cmd_buffer(cmd);
1307 /* Complete IO by inline, softirq or timer */
1308 switch (cmd->nq->dev->irqmode) {
1309 case NULL_IRQ_SOFTIRQ:
1310 switch (cmd->nq->dev->queue_mode) {
1312 if (likely(!blk_should_fake_timeout(cmd->rq->q)))
1313 blk_mq_complete_request(cmd->rq);
1317 * XXX: no proper submitting cpu information available.
1326 case NULL_IRQ_TIMER:
1327 null_cmd_end_timer(cmd);
1332 blk_status_t null_process_cmd(struct nullb_cmd *cmd,
1333 enum req_opf op, sector_t sector,
1334 unsigned int nr_sectors)
1336 struct nullb_device *dev = cmd->nq->dev;
1339 if (dev->badblocks.shift != -1) {
1340 ret = null_handle_badblocks(cmd, sector, nr_sectors);
1341 if (ret != BLK_STS_OK)
1345 if (dev->memory_backed)
1346 return null_handle_memory_backed(cmd, op);
1351 static blk_status_t null_handle_cmd(struct nullb_cmd *cmd, sector_t sector,
1352 sector_t nr_sectors, enum req_opf op)
1354 struct nullb_device *dev = cmd->nq->dev;
1355 struct nullb *nullb = dev->nullb;
1358 if (test_bit(NULLB_DEV_FL_THROTTLED, &dev->flags)) {
1359 sts = null_handle_throttled(cmd);
1360 if (sts != BLK_STS_OK)
1364 if (op == REQ_OP_FLUSH) {
1365 cmd->error = errno_to_blk_status(null_handle_flush(nullb));
1370 sts = null_process_zoned_cmd(cmd, op, sector, nr_sectors);
1372 sts = null_process_cmd(cmd, op, sector, nr_sectors);
1374 /* Do not overwrite errors (e.g. timeout errors) */
1375 if (cmd->error == BLK_STS_OK)
1379 nullb_complete_cmd(cmd);
1383 static enum hrtimer_restart nullb_bwtimer_fn(struct hrtimer *timer)
1385 struct nullb *nullb = container_of(timer, struct nullb, bw_timer);
1386 ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1387 unsigned int mbps = nullb->dev->mbps;
1389 if (atomic_long_read(&nullb->cur_bytes) == mb_per_tick(mbps))
1390 return HRTIMER_NORESTART;
1392 atomic_long_set(&nullb->cur_bytes, mb_per_tick(mbps));
1393 null_restart_queue_async(nullb);
1395 hrtimer_forward_now(&nullb->bw_timer, timer_interval);
1397 return HRTIMER_RESTART;
1400 static void nullb_setup_bwtimer(struct nullb *nullb)
1402 ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1404 hrtimer_init(&nullb->bw_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1405 nullb->bw_timer.function = nullb_bwtimer_fn;
1406 atomic_long_set(&nullb->cur_bytes, mb_per_tick(nullb->dev->mbps));
1407 hrtimer_start(&nullb->bw_timer, timer_interval, HRTIMER_MODE_REL);
1410 static struct nullb_queue *nullb_to_queue(struct nullb *nullb)
1414 if (nullb->nr_queues != 1)
1415 index = raw_smp_processor_id() / ((nr_cpu_ids + nullb->nr_queues - 1) / nullb->nr_queues);
1417 return &nullb->queues[index];
1420 static blk_qc_t null_submit_bio(struct bio *bio)
1422 sector_t sector = bio->bi_iter.bi_sector;
1423 sector_t nr_sectors = bio_sectors(bio);
1424 struct nullb *nullb = bio->bi_disk->private_data;
1425 struct nullb_queue *nq = nullb_to_queue(nullb);
1426 struct nullb_cmd *cmd;
1428 cmd = alloc_cmd(nq, 1);
1431 null_handle_cmd(cmd, sector, nr_sectors, bio_op(bio));
1432 return BLK_QC_T_NONE;
1435 static bool should_timeout_request(struct request *rq)
1437 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1438 if (g_timeout_str[0])
1439 return should_fail(&null_timeout_attr, 1);
1444 static bool should_requeue_request(struct request *rq)
1446 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1447 if (g_requeue_str[0])
1448 return should_fail(&null_requeue_attr, 1);
1453 static enum blk_eh_timer_return null_timeout_rq(struct request *rq, bool res)
1455 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq);
1457 pr_info("rq %p timed out\n", rq);
1460 * If the device is marked as blocking (i.e. memory backed or zoned
1461 * device), the submission path may be blocked waiting for resources
1462 * and cause real timeouts. For these real timeouts, the submission
1463 * path will complete the request using blk_mq_complete_request().
1464 * Only fake timeouts need to execute blk_mq_complete_request() here.
1466 cmd->error = BLK_STS_TIMEOUT;
1467 if (cmd->fake_timeout)
1468 blk_mq_complete_request(rq);
1472 static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx,
1473 const struct blk_mq_queue_data *bd)
1475 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
1476 struct nullb_queue *nq = hctx->driver_data;
1477 sector_t nr_sectors = blk_rq_sectors(bd->rq);
1478 sector_t sector = blk_rq_pos(bd->rq);
1480 might_sleep_if(hctx->flags & BLK_MQ_F_BLOCKING);
1482 if (nq->dev->irqmode == NULL_IRQ_TIMER) {
1483 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1484 cmd->timer.function = null_cmd_timer_expired;
1487 cmd->error = BLK_STS_OK;
1489 cmd->fake_timeout = should_timeout_request(bd->rq);
1491 blk_mq_start_request(bd->rq);
1493 if (should_requeue_request(bd->rq)) {
1495 * Alternate between hitting the core BUSY path, and the
1496 * driver driven requeue path
1498 nq->requeue_selection++;
1499 if (nq->requeue_selection & 1)
1500 return BLK_STS_RESOURCE;
1502 blk_mq_requeue_request(bd->rq, true);
1506 if (cmd->fake_timeout)
1509 return null_handle_cmd(cmd, sector, nr_sectors, req_op(bd->rq));
1512 static void cleanup_queue(struct nullb_queue *nq)
1518 static void cleanup_queues(struct nullb *nullb)
1522 for (i = 0; i < nullb->nr_queues; i++)
1523 cleanup_queue(&nullb->queues[i]);
1525 kfree(nullb->queues);
1528 static void null_exit_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx)
1530 struct nullb_queue *nq = hctx->driver_data;
1531 struct nullb *nullb = nq->dev->nullb;
1536 static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq)
1538 init_waitqueue_head(&nq->wait);
1539 nq->queue_depth = nullb->queue_depth;
1540 nq->dev = nullb->dev;
1543 static int null_init_hctx(struct blk_mq_hw_ctx *hctx, void *driver_data,
1544 unsigned int hctx_idx)
1546 struct nullb *nullb = hctx->queue->queuedata;
1547 struct nullb_queue *nq;
1549 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1550 if (g_init_hctx_str[0] && should_fail(&null_init_hctx_attr, 1))
1554 nq = &nullb->queues[hctx_idx];
1555 hctx->driver_data = nq;
1556 null_init_queue(nullb, nq);
1562 static const struct blk_mq_ops null_mq_ops = {
1563 .queue_rq = null_queue_rq,
1564 .complete = null_complete_rq,
1565 .timeout = null_timeout_rq,
1566 .init_hctx = null_init_hctx,
1567 .exit_hctx = null_exit_hctx,
1570 static void null_del_dev(struct nullb *nullb)
1572 struct nullb_device *dev;
1579 ida_simple_remove(&nullb_indexes, nullb->index);
1581 list_del_init(&nullb->list);
1583 del_gendisk(nullb->disk);
1585 if (test_bit(NULLB_DEV_FL_THROTTLED, &nullb->dev->flags)) {
1586 hrtimer_cancel(&nullb->bw_timer);
1587 atomic_long_set(&nullb->cur_bytes, LONG_MAX);
1588 null_restart_queue_async(nullb);
1591 blk_cleanup_queue(nullb->q);
1592 if (dev->queue_mode == NULL_Q_MQ &&
1593 nullb->tag_set == &nullb->__tag_set)
1594 blk_mq_free_tag_set(nullb->tag_set);
1595 put_disk(nullb->disk);
1596 cleanup_queues(nullb);
1597 if (null_cache_active(nullb))
1598 null_free_device_storage(nullb->dev, true);
1603 static void null_config_discard(struct nullb *nullb)
1605 if (nullb->dev->discard == false)
1608 if (nullb->dev->zoned) {
1609 nullb->dev->discard = false;
1610 pr_info("discard option is ignored in zoned mode\n");
1614 nullb->q->limits.discard_granularity = nullb->dev->blocksize;
1615 nullb->q->limits.discard_alignment = nullb->dev->blocksize;
1616 blk_queue_max_discard_sectors(nullb->q, UINT_MAX >> 9);
1617 blk_queue_flag_set(QUEUE_FLAG_DISCARD, nullb->q);
1620 static const struct block_device_operations null_bio_ops = {
1621 .owner = THIS_MODULE,
1622 .submit_bio = null_submit_bio,
1623 .report_zones = null_report_zones,
1626 static const struct block_device_operations null_rq_ops = {
1627 .owner = THIS_MODULE,
1628 .report_zones = null_report_zones,
1631 static int setup_commands(struct nullb_queue *nq)
1633 struct nullb_cmd *cmd;
1636 nq->cmds = kcalloc(nq->queue_depth, sizeof(*cmd), GFP_KERNEL);
1640 tag_size = ALIGN(nq->queue_depth, BITS_PER_LONG) / BITS_PER_LONG;
1641 nq->tag_map = kcalloc(tag_size, sizeof(unsigned long), GFP_KERNEL);
1647 for (i = 0; i < nq->queue_depth; i++) {
1655 static int setup_queues(struct nullb *nullb)
1657 nullb->queues = kcalloc(nr_cpu_ids, sizeof(struct nullb_queue),
1662 nullb->queue_depth = nullb->dev->hw_queue_depth;
1667 static int init_driver_queues(struct nullb *nullb)
1669 struct nullb_queue *nq;
1672 for (i = 0; i < nullb->dev->submit_queues; i++) {
1673 nq = &nullb->queues[i];
1675 null_init_queue(nullb, nq);
1677 ret = setup_commands(nq);
1685 static int null_gendisk_register(struct nullb *nullb)
1687 sector_t size = ((sector_t)nullb->dev->size * SZ_1M) >> SECTOR_SHIFT;
1688 struct gendisk *disk;
1690 disk = nullb->disk = alloc_disk_node(1, nullb->dev->home_node);
1693 set_capacity(disk, size);
1695 disk->flags |= GENHD_FL_EXT_DEVT | GENHD_FL_SUPPRESS_PARTITION_INFO;
1696 disk->major = null_major;
1697 disk->first_minor = nullb->index;
1698 if (queue_is_mq(nullb->q))
1699 disk->fops = &null_rq_ops;
1701 disk->fops = &null_bio_ops;
1702 disk->private_data = nullb;
1703 disk->queue = nullb->q;
1704 strncpy(disk->disk_name, nullb->disk_name, DISK_NAME_LEN);
1706 if (nullb->dev->zoned) {
1707 int ret = null_register_zoned_dev(nullb);
1717 static int null_init_tag_set(struct nullb *nullb, struct blk_mq_tag_set *set)
1719 set->ops = &null_mq_ops;
1720 set->nr_hw_queues = nullb ? nullb->dev->submit_queues :
1722 set->queue_depth = nullb ? nullb->dev->hw_queue_depth :
1724 set->numa_node = nullb ? nullb->dev->home_node : g_home_node;
1725 set->cmd_size = sizeof(struct nullb_cmd);
1726 set->flags = BLK_MQ_F_SHOULD_MERGE;
1728 set->flags |= BLK_MQ_F_NO_SCHED;
1729 if (g_shared_tag_bitmap)
1730 set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1731 set->driver_data = NULL;
1733 if ((nullb && nullb->dev->blocking) || g_blocking)
1734 set->flags |= BLK_MQ_F_BLOCKING;
1736 return blk_mq_alloc_tag_set(set);
1739 static int null_validate_conf(struct nullb_device *dev)
1741 dev->blocksize = round_down(dev->blocksize, 512);
1742 dev->blocksize = clamp_t(unsigned int, dev->blocksize, 512, 4096);
1744 if (dev->queue_mode == NULL_Q_MQ && dev->use_per_node_hctx) {
1745 if (dev->submit_queues != nr_online_nodes)
1746 dev->submit_queues = nr_online_nodes;
1747 } else if (dev->submit_queues > nr_cpu_ids)
1748 dev->submit_queues = nr_cpu_ids;
1749 else if (dev->submit_queues == 0)
1750 dev->submit_queues = 1;
1752 dev->queue_mode = min_t(unsigned int, dev->queue_mode, NULL_Q_MQ);
1753 dev->irqmode = min_t(unsigned int, dev->irqmode, NULL_IRQ_TIMER);
1755 /* Do memory allocation, so set blocking */
1756 if (dev->memory_backed)
1757 dev->blocking = true;
1758 else /* cache is meaningless */
1759 dev->cache_size = 0;
1760 dev->cache_size = min_t(unsigned long, ULONG_MAX / 1024 / 1024,
1762 dev->mbps = min_t(unsigned int, 1024 * 40, dev->mbps);
1763 /* can not stop a queue */
1764 if (dev->queue_mode == NULL_Q_BIO)
1768 (!dev->zone_size || !is_power_of_2(dev->zone_size))) {
1769 pr_err("zone_size must be power-of-two\n");
1776 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1777 static bool __null_setup_fault(struct fault_attr *attr, char *str)
1782 if (!setup_fault_attr(attr, str))
1790 static bool null_setup_fault(void)
1792 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1793 if (!__null_setup_fault(&null_timeout_attr, g_timeout_str))
1795 if (!__null_setup_fault(&null_requeue_attr, g_requeue_str))
1797 if (!__null_setup_fault(&null_init_hctx_attr, g_init_hctx_str))
1803 static int null_add_dev(struct nullb_device *dev)
1805 struct nullb *nullb;
1808 rv = null_validate_conf(dev);
1812 nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, dev->home_node);
1820 spin_lock_init(&nullb->lock);
1822 rv = setup_queues(nullb);
1824 goto out_free_nullb;
1826 if (dev->queue_mode == NULL_Q_MQ) {
1828 nullb->tag_set = &tag_set;
1831 nullb->tag_set = &nullb->__tag_set;
1832 rv = null_init_tag_set(nullb, nullb->tag_set);
1836 goto out_cleanup_queues;
1838 if (!null_setup_fault())
1839 goto out_cleanup_queues;
1841 nullb->tag_set->timeout = 5 * HZ;
1842 nullb->q = blk_mq_init_queue_data(nullb->tag_set, nullb);
1843 if (IS_ERR(nullb->q)) {
1845 goto out_cleanup_tags;
1847 } else if (dev->queue_mode == NULL_Q_BIO) {
1848 nullb->q = blk_alloc_queue(dev->home_node);
1851 goto out_cleanup_queues;
1853 rv = init_driver_queues(nullb);
1855 goto out_cleanup_blk_queue;
1859 set_bit(NULLB_DEV_FL_THROTTLED, &dev->flags);
1860 nullb_setup_bwtimer(nullb);
1863 if (dev->cache_size > 0) {
1864 set_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
1865 blk_queue_write_cache(nullb->q, true, true);
1869 rv = null_init_zoned_dev(dev, nullb->q);
1871 goto out_cleanup_blk_queue;
1874 nullb->q->queuedata = nullb;
1875 blk_queue_flag_set(QUEUE_FLAG_NONROT, nullb->q);
1876 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, nullb->q);
1879 rv = ida_simple_get(&nullb_indexes, 0, 0, GFP_KERNEL);
1881 mutex_unlock(&lock);
1882 goto out_cleanup_zone;
1886 mutex_unlock(&lock);
1888 blk_queue_logical_block_size(nullb->q, dev->blocksize);
1889 blk_queue_physical_block_size(nullb->q, dev->blocksize);
1891 null_config_discard(nullb);
1893 sprintf(nullb->disk_name, "nullb%d", nullb->index);
1895 rv = null_gendisk_register(nullb);
1900 list_add_tail(&nullb->list, &nullb_list);
1901 mutex_unlock(&lock);
1906 ida_free(&nullb_indexes, nullb->index);
1908 null_free_zoned_dev(dev);
1909 out_cleanup_blk_queue:
1910 blk_cleanup_queue(nullb->q);
1912 if (dev->queue_mode == NULL_Q_MQ && nullb->tag_set == &nullb->__tag_set)
1913 blk_mq_free_tag_set(nullb->tag_set);
1915 cleanup_queues(nullb);
1923 static int __init null_init(void)
1927 struct nullb *nullb;
1928 struct nullb_device *dev;
1930 if (g_bs > PAGE_SIZE) {
1931 pr_warn("invalid block size\n");
1932 pr_warn("defaults block size to %lu\n", PAGE_SIZE);
1936 if (g_home_node != NUMA_NO_NODE && g_home_node >= nr_online_nodes) {
1937 pr_err("invalid home_node value\n");
1938 g_home_node = NUMA_NO_NODE;
1941 if (g_queue_mode == NULL_Q_RQ) {
1942 pr_err("legacy IO path no longer available\n");
1945 if (g_queue_mode == NULL_Q_MQ && g_use_per_node_hctx) {
1946 if (g_submit_queues != nr_online_nodes) {
1947 pr_warn("submit_queues param is set to %u.\n",
1949 g_submit_queues = nr_online_nodes;
1951 } else if (g_submit_queues > nr_cpu_ids)
1952 g_submit_queues = nr_cpu_ids;
1953 else if (g_submit_queues <= 0)
1954 g_submit_queues = 1;
1956 if (g_queue_mode == NULL_Q_MQ && shared_tags) {
1957 ret = null_init_tag_set(NULL, &tag_set);
1962 config_group_init(&nullb_subsys.su_group);
1963 mutex_init(&nullb_subsys.su_mutex);
1965 ret = configfs_register_subsystem(&nullb_subsys);
1971 null_major = register_blkdev(0, "nullb");
1972 if (null_major < 0) {
1977 for (i = 0; i < nr_devices; i++) {
1978 dev = null_alloc_dev();
1983 ret = null_add_dev(dev);
1990 pr_info("module loaded\n");
1994 while (!list_empty(&nullb_list)) {
1995 nullb = list_entry(nullb_list.next, struct nullb, list);
1997 null_del_dev(nullb);
2000 unregister_blkdev(null_major, "nullb");
2002 configfs_unregister_subsystem(&nullb_subsys);
2004 if (g_queue_mode == NULL_Q_MQ && shared_tags)
2005 blk_mq_free_tag_set(&tag_set);
2009 static void __exit null_exit(void)
2011 struct nullb *nullb;
2013 configfs_unregister_subsystem(&nullb_subsys);
2015 unregister_blkdev(null_major, "nullb");
2018 while (!list_empty(&nullb_list)) {
2019 struct nullb_device *dev;
2021 nullb = list_entry(nullb_list.next, struct nullb, list);
2023 null_del_dev(nullb);
2026 mutex_unlock(&lock);
2028 if (g_queue_mode == NULL_Q_MQ && shared_tags)
2029 blk_mq_free_tag_set(&tag_set);
2032 module_init(null_init);
2033 module_exit(null_exit);
2035 MODULE_AUTHOR("Jens Axboe <axboe@kernel.dk>");
2036 MODULE_LICENSE("GPL");