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 int g_irqmode = NULL_IRQ_SOFTIRQ;
169 static int null_set_irqmode(const char *str, const struct kernel_param *kp)
171 return null_param_store_val(str, &g_irqmode, NULL_IRQ_NONE,
175 static const struct kernel_param_ops null_irqmode_param_ops = {
176 .set = null_set_irqmode,
177 .get = param_get_int,
180 device_param_cb(irqmode, &null_irqmode_param_ops, &g_irqmode, 0444);
181 MODULE_PARM_DESC(irqmode, "IRQ completion handler. 0-none, 1-softirq, 2-timer");
183 static unsigned long g_completion_nsec = 10000;
184 module_param_named(completion_nsec, g_completion_nsec, ulong, 0444);
185 MODULE_PARM_DESC(completion_nsec, "Time in ns to complete a request in hardware. Default: 10,000ns");
187 static int g_hw_queue_depth = 64;
188 module_param_named(hw_queue_depth, g_hw_queue_depth, int, 0444);
189 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 64");
191 static bool g_use_per_node_hctx;
192 module_param_named(use_per_node_hctx, g_use_per_node_hctx, bool, 0444);
193 MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware context queues. Default: false");
196 module_param_named(zoned, g_zoned, bool, S_IRUGO);
197 MODULE_PARM_DESC(zoned, "Make device as a host-managed zoned block device. Default: false");
199 static unsigned long g_zone_size = 256;
200 module_param_named(zone_size, g_zone_size, ulong, S_IRUGO);
201 MODULE_PARM_DESC(zone_size, "Zone size in MB when block device is zoned. Must be power-of-two: Default: 256");
203 static unsigned int g_zone_nr_conv;
204 module_param_named(zone_nr_conv, g_zone_nr_conv, uint, 0444);
205 MODULE_PARM_DESC(zone_nr_conv, "Number of conventional zones when block device is zoned. Default: 0");
207 static struct nullb_device *null_alloc_dev(void);
208 static void null_free_dev(struct nullb_device *dev);
209 static void null_del_dev(struct nullb *nullb);
210 static int null_add_dev(struct nullb_device *dev);
211 static void null_free_device_storage(struct nullb_device *dev, bool is_cache);
213 static inline struct nullb_device *to_nullb_device(struct config_item *item)
215 return item ? container_of(item, struct nullb_device, item) : NULL;
218 static inline ssize_t nullb_device_uint_attr_show(unsigned int val, char *page)
220 return snprintf(page, PAGE_SIZE, "%u\n", val);
223 static inline ssize_t nullb_device_ulong_attr_show(unsigned long val,
226 return snprintf(page, PAGE_SIZE, "%lu\n", val);
229 static inline ssize_t nullb_device_bool_attr_show(bool val, char *page)
231 return snprintf(page, PAGE_SIZE, "%u\n", val);
234 static ssize_t nullb_device_uint_attr_store(unsigned int *val,
235 const char *page, size_t count)
240 result = kstrtouint(page, 0, &tmp);
248 static ssize_t nullb_device_ulong_attr_store(unsigned long *val,
249 const char *page, size_t count)
254 result = kstrtoul(page, 0, &tmp);
262 static ssize_t nullb_device_bool_attr_store(bool *val, const char *page,
268 result = kstrtobool(page, &tmp);
276 /* The following macro should only be used with TYPE = {uint, ulong, bool}. */
277 #define NULLB_DEVICE_ATTR(NAME, TYPE, APPLY) \
279 nullb_device_##NAME##_show(struct config_item *item, char *page) \
281 return nullb_device_##TYPE##_attr_show( \
282 to_nullb_device(item)->NAME, page); \
285 nullb_device_##NAME##_store(struct config_item *item, const char *page, \
288 int (*apply_fn)(struct nullb_device *dev, TYPE new_value) = APPLY;\
289 struct nullb_device *dev = to_nullb_device(item); \
290 TYPE new_value = 0; \
293 ret = nullb_device_##TYPE##_attr_store(&new_value, page, count);\
297 ret = apply_fn(dev, new_value); \
298 else if (test_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags)) \
302 dev->NAME = new_value; \
305 CONFIGFS_ATTR(nullb_device_, NAME);
307 static int nullb_apply_submit_queues(struct nullb_device *dev,
308 unsigned int submit_queues)
310 struct nullb *nullb = dev->nullb;
311 struct blk_mq_tag_set *set;
317 * Make sure that null_init_hctx() does not access nullb->queues[] past
318 * the end of that array.
320 if (submit_queues > nr_cpu_ids)
322 set = nullb->tag_set;
323 blk_mq_update_nr_hw_queues(set, submit_queues);
324 return set->nr_hw_queues == submit_queues ? 0 : -ENOMEM;
327 NULLB_DEVICE_ATTR(size, ulong, NULL);
328 NULLB_DEVICE_ATTR(completion_nsec, ulong, NULL);
329 NULLB_DEVICE_ATTR(submit_queues, uint, nullb_apply_submit_queues);
330 NULLB_DEVICE_ATTR(home_node, uint, NULL);
331 NULLB_DEVICE_ATTR(queue_mode, uint, NULL);
332 NULLB_DEVICE_ATTR(blocksize, uint, NULL);
333 NULLB_DEVICE_ATTR(irqmode, uint, NULL);
334 NULLB_DEVICE_ATTR(hw_queue_depth, uint, NULL);
335 NULLB_DEVICE_ATTR(index, uint, NULL);
336 NULLB_DEVICE_ATTR(blocking, bool, NULL);
337 NULLB_DEVICE_ATTR(use_per_node_hctx, bool, NULL);
338 NULLB_DEVICE_ATTR(memory_backed, bool, NULL);
339 NULLB_DEVICE_ATTR(discard, bool, NULL);
340 NULLB_DEVICE_ATTR(mbps, uint, NULL);
341 NULLB_DEVICE_ATTR(cache_size, ulong, NULL);
342 NULLB_DEVICE_ATTR(zoned, bool, NULL);
343 NULLB_DEVICE_ATTR(zone_size, ulong, NULL);
344 NULLB_DEVICE_ATTR(zone_nr_conv, uint, NULL);
346 static ssize_t nullb_device_power_show(struct config_item *item, char *page)
348 return nullb_device_bool_attr_show(to_nullb_device(item)->power, page);
351 static ssize_t nullb_device_power_store(struct config_item *item,
352 const char *page, size_t count)
354 struct nullb_device *dev = to_nullb_device(item);
358 ret = nullb_device_bool_attr_store(&newp, page, count);
362 if (!dev->power && newp) {
363 if (test_and_set_bit(NULLB_DEV_FL_UP, &dev->flags))
365 if (null_add_dev(dev)) {
366 clear_bit(NULLB_DEV_FL_UP, &dev->flags);
370 set_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
372 } else if (dev->power && !newp) {
373 if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
376 null_del_dev(dev->nullb);
379 clear_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
385 CONFIGFS_ATTR(nullb_device_, power);
387 static ssize_t nullb_device_badblocks_show(struct config_item *item, char *page)
389 struct nullb_device *t_dev = to_nullb_device(item);
391 return badblocks_show(&t_dev->badblocks, page, 0);
394 static ssize_t nullb_device_badblocks_store(struct config_item *item,
395 const char *page, size_t count)
397 struct nullb_device *t_dev = to_nullb_device(item);
398 char *orig, *buf, *tmp;
402 orig = kstrndup(page, count, GFP_KERNEL);
406 buf = strstrip(orig);
409 if (buf[0] != '+' && buf[0] != '-')
411 tmp = strchr(&buf[1], '-');
415 ret = kstrtoull(buf + 1, 0, &start);
418 ret = kstrtoull(tmp + 1, 0, &end);
424 /* enable badblocks */
425 cmpxchg(&t_dev->badblocks.shift, -1, 0);
427 ret = badblocks_set(&t_dev->badblocks, start,
430 ret = badblocks_clear(&t_dev->badblocks, start,
438 CONFIGFS_ATTR(nullb_device_, badblocks);
440 static struct configfs_attribute *nullb_device_attrs[] = {
441 &nullb_device_attr_size,
442 &nullb_device_attr_completion_nsec,
443 &nullb_device_attr_submit_queues,
444 &nullb_device_attr_home_node,
445 &nullb_device_attr_queue_mode,
446 &nullb_device_attr_blocksize,
447 &nullb_device_attr_irqmode,
448 &nullb_device_attr_hw_queue_depth,
449 &nullb_device_attr_index,
450 &nullb_device_attr_blocking,
451 &nullb_device_attr_use_per_node_hctx,
452 &nullb_device_attr_power,
453 &nullb_device_attr_memory_backed,
454 &nullb_device_attr_discard,
455 &nullb_device_attr_mbps,
456 &nullb_device_attr_cache_size,
457 &nullb_device_attr_badblocks,
458 &nullb_device_attr_zoned,
459 &nullb_device_attr_zone_size,
460 &nullb_device_attr_zone_nr_conv,
464 static void nullb_device_release(struct config_item *item)
466 struct nullb_device *dev = to_nullb_device(item);
468 null_free_device_storage(dev, false);
472 static struct configfs_item_operations nullb_device_ops = {
473 .release = nullb_device_release,
476 static const struct config_item_type nullb_device_type = {
477 .ct_item_ops = &nullb_device_ops,
478 .ct_attrs = nullb_device_attrs,
479 .ct_owner = THIS_MODULE,
483 config_item *nullb_group_make_item(struct config_group *group, const char *name)
485 struct nullb_device *dev;
487 dev = null_alloc_dev();
489 return ERR_PTR(-ENOMEM);
491 config_item_init_type_name(&dev->item, name, &nullb_device_type);
497 nullb_group_drop_item(struct config_group *group, struct config_item *item)
499 struct nullb_device *dev = to_nullb_device(item);
501 if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
504 null_del_dev(dev->nullb);
508 config_item_put(item);
511 static ssize_t memb_group_features_show(struct config_item *item, char *page)
513 return snprintf(page, PAGE_SIZE, "memory_backed,discard,bandwidth,cache,badblocks,zoned,zone_size,zone_nr_conv\n");
516 CONFIGFS_ATTR_RO(memb_group_, features);
518 static struct configfs_attribute *nullb_group_attrs[] = {
519 &memb_group_attr_features,
523 static struct configfs_group_operations nullb_group_ops = {
524 .make_item = nullb_group_make_item,
525 .drop_item = nullb_group_drop_item,
528 static const struct config_item_type nullb_group_type = {
529 .ct_group_ops = &nullb_group_ops,
530 .ct_attrs = nullb_group_attrs,
531 .ct_owner = THIS_MODULE,
534 static struct configfs_subsystem nullb_subsys = {
537 .ci_namebuf = "nullb",
538 .ci_type = &nullb_group_type,
543 static inline int null_cache_active(struct nullb *nullb)
545 return test_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
548 static struct nullb_device *null_alloc_dev(void)
550 struct nullb_device *dev;
552 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
555 INIT_RADIX_TREE(&dev->data, GFP_ATOMIC);
556 INIT_RADIX_TREE(&dev->cache, GFP_ATOMIC);
557 if (badblocks_init(&dev->badblocks, 0)) {
562 dev->size = g_gb * 1024;
563 dev->completion_nsec = g_completion_nsec;
564 dev->submit_queues = g_submit_queues;
565 dev->home_node = g_home_node;
566 dev->queue_mode = g_queue_mode;
567 dev->blocksize = g_bs;
568 dev->irqmode = g_irqmode;
569 dev->hw_queue_depth = g_hw_queue_depth;
570 dev->blocking = g_blocking;
571 dev->use_per_node_hctx = g_use_per_node_hctx;
572 dev->zoned = g_zoned;
573 dev->zone_size = g_zone_size;
574 dev->zone_nr_conv = g_zone_nr_conv;
578 static void null_free_dev(struct nullb_device *dev)
583 null_free_zoned_dev(dev);
584 badblocks_exit(&dev->badblocks);
588 static void put_tag(struct nullb_queue *nq, unsigned int tag)
590 clear_bit_unlock(tag, nq->tag_map);
592 if (waitqueue_active(&nq->wait))
596 static unsigned int get_tag(struct nullb_queue *nq)
601 tag = find_first_zero_bit(nq->tag_map, nq->queue_depth);
602 if (tag >= nq->queue_depth)
604 } while (test_and_set_bit_lock(tag, nq->tag_map));
609 static void free_cmd(struct nullb_cmd *cmd)
611 put_tag(cmd->nq, cmd->tag);
614 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer);
616 static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq)
618 struct nullb_cmd *cmd;
623 cmd = &nq->cmds[tag];
625 cmd->error = BLK_STS_OK;
627 if (nq->dev->irqmode == NULL_IRQ_TIMER) {
628 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC,
630 cmd->timer.function = null_cmd_timer_expired;
638 static struct nullb_cmd *alloc_cmd(struct nullb_queue *nq, int can_wait)
640 struct nullb_cmd *cmd;
643 cmd = __alloc_cmd(nq);
644 if (cmd || !can_wait)
648 prepare_to_wait(&nq->wait, &wait, TASK_UNINTERRUPTIBLE);
649 cmd = __alloc_cmd(nq);
656 finish_wait(&nq->wait, &wait);
660 static void end_cmd(struct nullb_cmd *cmd)
662 int queue_mode = cmd->nq->dev->queue_mode;
664 switch (queue_mode) {
666 blk_mq_end_request(cmd->rq, cmd->error);
669 cmd->bio->bi_status = cmd->error;
677 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
679 end_cmd(container_of(timer, struct nullb_cmd, timer));
681 return HRTIMER_NORESTART;
684 static void null_cmd_end_timer(struct nullb_cmd *cmd)
686 ktime_t kt = cmd->nq->dev->completion_nsec;
688 hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL);
691 static void null_complete_rq(struct request *rq)
693 end_cmd(blk_mq_rq_to_pdu(rq));
696 static struct nullb_page *null_alloc_page(gfp_t gfp_flags)
698 struct nullb_page *t_page;
700 t_page = kmalloc(sizeof(struct nullb_page), gfp_flags);
704 t_page->page = alloc_pages(gfp_flags, 0);
708 memset(t_page->bitmap, 0, sizeof(t_page->bitmap));
716 static void null_free_page(struct nullb_page *t_page)
718 __set_bit(NULLB_PAGE_FREE, t_page->bitmap);
719 if (test_bit(NULLB_PAGE_LOCK, t_page->bitmap))
721 __free_page(t_page->page);
725 static bool null_page_empty(struct nullb_page *page)
727 int size = MAP_SZ - 2;
729 return find_first_bit(page->bitmap, size) == size;
732 static void null_free_sector(struct nullb *nullb, sector_t sector,
735 unsigned int sector_bit;
737 struct nullb_page *t_page, *ret;
738 struct radix_tree_root *root;
740 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
741 idx = sector >> PAGE_SECTORS_SHIFT;
742 sector_bit = (sector & SECTOR_MASK);
744 t_page = radix_tree_lookup(root, idx);
746 __clear_bit(sector_bit, t_page->bitmap);
748 if (null_page_empty(t_page)) {
749 ret = radix_tree_delete_item(root, idx, t_page);
750 WARN_ON(ret != t_page);
753 nullb->dev->curr_cache -= PAGE_SIZE;
758 static struct nullb_page *null_radix_tree_insert(struct nullb *nullb, u64 idx,
759 struct nullb_page *t_page, bool is_cache)
761 struct radix_tree_root *root;
763 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
765 if (radix_tree_insert(root, idx, t_page)) {
766 null_free_page(t_page);
767 t_page = radix_tree_lookup(root, idx);
768 WARN_ON(!t_page || t_page->page->index != idx);
770 nullb->dev->curr_cache += PAGE_SIZE;
775 static void null_free_device_storage(struct nullb_device *dev, bool is_cache)
777 unsigned long pos = 0;
779 struct nullb_page *ret, *t_pages[FREE_BATCH];
780 struct radix_tree_root *root;
782 root = is_cache ? &dev->cache : &dev->data;
787 nr_pages = radix_tree_gang_lookup(root,
788 (void **)t_pages, pos, FREE_BATCH);
790 for (i = 0; i < nr_pages; i++) {
791 pos = t_pages[i]->page->index;
792 ret = radix_tree_delete_item(root, pos, t_pages[i]);
793 WARN_ON(ret != t_pages[i]);
798 } while (nr_pages == FREE_BATCH);
804 static struct nullb_page *__null_lookup_page(struct nullb *nullb,
805 sector_t sector, bool for_write, bool is_cache)
807 unsigned int sector_bit;
809 struct nullb_page *t_page;
810 struct radix_tree_root *root;
812 idx = sector >> PAGE_SECTORS_SHIFT;
813 sector_bit = (sector & SECTOR_MASK);
815 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
816 t_page = radix_tree_lookup(root, idx);
817 WARN_ON(t_page && t_page->page->index != idx);
819 if (t_page && (for_write || test_bit(sector_bit, t_page->bitmap)))
825 static struct nullb_page *null_lookup_page(struct nullb *nullb,
826 sector_t sector, bool for_write, bool ignore_cache)
828 struct nullb_page *page = NULL;
831 page = __null_lookup_page(nullb, sector, for_write, true);
834 return __null_lookup_page(nullb, sector, for_write, false);
837 static struct nullb_page *null_insert_page(struct nullb *nullb,
838 sector_t sector, bool ignore_cache)
839 __releases(&nullb->lock)
840 __acquires(&nullb->lock)
843 struct nullb_page *t_page;
845 t_page = null_lookup_page(nullb, sector, true, ignore_cache);
849 spin_unlock_irq(&nullb->lock);
851 t_page = null_alloc_page(GFP_NOIO);
855 if (radix_tree_preload(GFP_NOIO))
858 spin_lock_irq(&nullb->lock);
859 idx = sector >> PAGE_SECTORS_SHIFT;
860 t_page->page->index = idx;
861 t_page = null_radix_tree_insert(nullb, idx, t_page, !ignore_cache);
862 radix_tree_preload_end();
866 null_free_page(t_page);
868 spin_lock_irq(&nullb->lock);
869 return null_lookup_page(nullb, sector, true, ignore_cache);
872 static int null_flush_cache_page(struct nullb *nullb, struct nullb_page *c_page)
877 struct nullb_page *t_page, *ret;
880 idx = c_page->page->index;
882 t_page = null_insert_page(nullb, idx << PAGE_SECTORS_SHIFT, true);
884 __clear_bit(NULLB_PAGE_LOCK, c_page->bitmap);
885 if (test_bit(NULLB_PAGE_FREE, c_page->bitmap)) {
886 null_free_page(c_page);
887 if (t_page && null_page_empty(t_page)) {
888 ret = radix_tree_delete_item(&nullb->dev->data,
890 null_free_page(t_page);
898 src = kmap_atomic(c_page->page);
899 dst = kmap_atomic(t_page->page);
901 for (i = 0; i < PAGE_SECTORS;
902 i += (nullb->dev->blocksize >> SECTOR_SHIFT)) {
903 if (test_bit(i, c_page->bitmap)) {
904 offset = (i << SECTOR_SHIFT);
905 memcpy(dst + offset, src + offset,
906 nullb->dev->blocksize);
907 __set_bit(i, t_page->bitmap);
914 ret = radix_tree_delete_item(&nullb->dev->cache, idx, c_page);
916 nullb->dev->curr_cache -= PAGE_SIZE;
921 static int null_make_cache_space(struct nullb *nullb, unsigned long n)
923 int i, err, nr_pages;
924 struct nullb_page *c_pages[FREE_BATCH];
925 unsigned long flushed = 0, one_round;
928 if ((nullb->dev->cache_size * 1024 * 1024) >
929 nullb->dev->curr_cache + n || nullb->dev->curr_cache == 0)
932 nr_pages = radix_tree_gang_lookup(&nullb->dev->cache,
933 (void **)c_pages, nullb->cache_flush_pos, FREE_BATCH);
935 * nullb_flush_cache_page could unlock before using the c_pages. To
936 * avoid race, we don't allow page free
938 for (i = 0; i < nr_pages; i++) {
939 nullb->cache_flush_pos = c_pages[i]->page->index;
941 * We found the page which is being flushed to disk by other
944 if (test_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap))
947 __set_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap);
951 for (i = 0; i < nr_pages; i++) {
952 if (c_pages[i] == NULL)
954 err = null_flush_cache_page(nullb, c_pages[i]);
959 flushed += one_round << PAGE_SHIFT;
963 nullb->cache_flush_pos = 0;
964 if (one_round == 0) {
965 /* give other threads a chance */
966 spin_unlock_irq(&nullb->lock);
967 spin_lock_irq(&nullb->lock);
974 static int copy_to_nullb(struct nullb *nullb, struct page *source,
975 unsigned int off, sector_t sector, size_t n, bool is_fua)
977 size_t temp, count = 0;
979 struct nullb_page *t_page;
983 temp = min_t(size_t, nullb->dev->blocksize, n - count);
985 if (null_cache_active(nullb) && !is_fua)
986 null_make_cache_space(nullb, PAGE_SIZE);
988 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
989 t_page = null_insert_page(nullb, sector,
990 !null_cache_active(nullb) || is_fua);
994 src = kmap_atomic(source);
995 dst = kmap_atomic(t_page->page);
996 memcpy(dst + offset, src + off + count, temp);
1000 __set_bit(sector & SECTOR_MASK, t_page->bitmap);
1003 null_free_sector(nullb, sector, true);
1006 sector += temp >> SECTOR_SHIFT;
1011 static int copy_from_nullb(struct nullb *nullb, struct page *dest,
1012 unsigned int off, sector_t sector, size_t n)
1014 size_t temp, count = 0;
1015 unsigned int offset;
1016 struct nullb_page *t_page;
1020 temp = min_t(size_t, nullb->dev->blocksize, n - count);
1022 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
1023 t_page = null_lookup_page(nullb, sector, false,
1024 !null_cache_active(nullb));
1026 dst = kmap_atomic(dest);
1028 memset(dst + off + count, 0, temp);
1031 src = kmap_atomic(t_page->page);
1032 memcpy(dst + off + count, src + offset, temp);
1038 sector += temp >> SECTOR_SHIFT;
1043 static void nullb_fill_pattern(struct nullb *nullb, struct page *page,
1044 unsigned int len, unsigned int off)
1048 dst = kmap_atomic(page);
1049 memset(dst + off, 0xFF, len);
1053 static void null_handle_discard(struct nullb *nullb, sector_t sector, size_t n)
1057 spin_lock_irq(&nullb->lock);
1059 temp = min_t(size_t, n, nullb->dev->blocksize);
1060 null_free_sector(nullb, sector, false);
1061 if (null_cache_active(nullb))
1062 null_free_sector(nullb, sector, true);
1063 sector += temp >> SECTOR_SHIFT;
1066 spin_unlock_irq(&nullb->lock);
1069 static int null_handle_flush(struct nullb *nullb)
1073 if (!null_cache_active(nullb))
1076 spin_lock_irq(&nullb->lock);
1078 err = null_make_cache_space(nullb,
1079 nullb->dev->cache_size * 1024 * 1024);
1080 if (err || nullb->dev->curr_cache == 0)
1084 WARN_ON(!radix_tree_empty(&nullb->dev->cache));
1085 spin_unlock_irq(&nullb->lock);
1089 static int null_transfer(struct nullb *nullb, struct page *page,
1090 unsigned int len, unsigned int off, bool is_write, sector_t sector,
1093 struct nullb_device *dev = nullb->dev;
1094 unsigned int valid_len = len;
1099 valid_len = null_zone_valid_read_len(nullb,
1103 err = copy_from_nullb(nullb, page, off,
1110 nullb_fill_pattern(nullb, page, len, off);
1111 flush_dcache_page(page);
1113 flush_dcache_page(page);
1114 err = copy_to_nullb(nullb, page, off, sector, len, is_fua);
1120 static int null_handle_rq(struct nullb_cmd *cmd)
1122 struct request *rq = cmd->rq;
1123 struct nullb *nullb = cmd->nq->dev->nullb;
1127 struct req_iterator iter;
1128 struct bio_vec bvec;
1130 sector = blk_rq_pos(rq);
1132 if (req_op(rq) == REQ_OP_DISCARD) {
1133 null_handle_discard(nullb, sector, blk_rq_bytes(rq));
1137 spin_lock_irq(&nullb->lock);
1138 rq_for_each_segment(bvec, rq, iter) {
1140 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1141 op_is_write(req_op(rq)), sector,
1142 req_op(rq) & REQ_FUA);
1144 spin_unlock_irq(&nullb->lock);
1147 sector += len >> SECTOR_SHIFT;
1149 spin_unlock_irq(&nullb->lock);
1154 static int null_handle_bio(struct nullb_cmd *cmd)
1156 struct bio *bio = cmd->bio;
1157 struct nullb *nullb = cmd->nq->dev->nullb;
1161 struct bio_vec bvec;
1162 struct bvec_iter iter;
1164 sector = bio->bi_iter.bi_sector;
1166 if (bio_op(bio) == REQ_OP_DISCARD) {
1167 null_handle_discard(nullb, sector,
1168 bio_sectors(bio) << SECTOR_SHIFT);
1172 spin_lock_irq(&nullb->lock);
1173 bio_for_each_segment(bvec, bio, iter) {
1175 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1176 op_is_write(bio_op(bio)), sector,
1177 bio->bi_opf & REQ_FUA);
1179 spin_unlock_irq(&nullb->lock);
1182 sector += len >> SECTOR_SHIFT;
1184 spin_unlock_irq(&nullb->lock);
1188 static void null_stop_queue(struct nullb *nullb)
1190 struct request_queue *q = nullb->q;
1192 if (nullb->dev->queue_mode == NULL_Q_MQ)
1193 blk_mq_stop_hw_queues(q);
1196 static void null_restart_queue_async(struct nullb *nullb)
1198 struct request_queue *q = nullb->q;
1200 if (nullb->dev->queue_mode == NULL_Q_MQ)
1201 blk_mq_start_stopped_hw_queues(q, true);
1204 static inline blk_status_t null_handle_throttled(struct nullb_cmd *cmd)
1206 struct nullb_device *dev = cmd->nq->dev;
1207 struct nullb *nullb = dev->nullb;
1208 blk_status_t sts = BLK_STS_OK;
1209 struct request *rq = cmd->rq;
1211 if (!hrtimer_active(&nullb->bw_timer))
1212 hrtimer_restart(&nullb->bw_timer);
1214 if (atomic_long_sub_return(blk_rq_bytes(rq), &nullb->cur_bytes) < 0) {
1215 null_stop_queue(nullb);
1216 /* race with timer */
1217 if (atomic_long_read(&nullb->cur_bytes) > 0)
1218 null_restart_queue_async(nullb);
1219 /* requeue request */
1220 sts = BLK_STS_DEV_RESOURCE;
1225 static inline blk_status_t null_handle_badblocks(struct nullb_cmd *cmd,
1227 sector_t nr_sectors)
1229 struct badblocks *bb = &cmd->nq->dev->badblocks;
1233 if (badblocks_check(bb, sector, nr_sectors, &first_bad, &bad_sectors))
1234 return BLK_STS_IOERR;
1239 static inline blk_status_t null_handle_memory_backed(struct nullb_cmd *cmd,
1242 struct nullb_device *dev = cmd->nq->dev;
1245 if (dev->queue_mode == NULL_Q_BIO)
1246 err = null_handle_bio(cmd);
1248 err = null_handle_rq(cmd);
1250 return errno_to_blk_status(err);
1253 static void nullb_zero_read_cmd_buffer(struct nullb_cmd *cmd)
1255 struct nullb_device *dev = cmd->nq->dev;
1258 if (dev->memory_backed)
1261 if (dev->queue_mode == NULL_Q_BIO && bio_op(cmd->bio) == REQ_OP_READ) {
1262 zero_fill_bio(cmd->bio);
1263 } else if (req_op(cmd->rq) == REQ_OP_READ) {
1264 __rq_for_each_bio(bio, cmd->rq)
1269 static inline void nullb_complete_cmd(struct nullb_cmd *cmd)
1272 * Since root privileges are required to configure the null_blk
1273 * driver, it is fine that this driver does not initialize the
1274 * data buffers of read commands. Zero-initialize these buffers
1275 * anyway if KMSAN is enabled to prevent that KMSAN complains
1276 * about null_blk not initializing read data buffers.
1278 if (IS_ENABLED(CONFIG_KMSAN))
1279 nullb_zero_read_cmd_buffer(cmd);
1281 /* Complete IO by inline, softirq or timer */
1282 switch (cmd->nq->dev->irqmode) {
1283 case NULL_IRQ_SOFTIRQ:
1284 switch (cmd->nq->dev->queue_mode) {
1286 blk_mq_complete_request(cmd->rq);
1290 * XXX: no proper submitting cpu information available.
1299 case NULL_IRQ_TIMER:
1300 null_cmd_end_timer(cmd);
1305 blk_status_t null_process_cmd(struct nullb_cmd *cmd,
1306 enum req_opf op, sector_t sector,
1307 unsigned int nr_sectors)
1309 struct nullb_device *dev = cmd->nq->dev;
1312 if (dev->badblocks.shift != -1) {
1313 ret = null_handle_badblocks(cmd, sector, nr_sectors);
1314 if (ret != BLK_STS_OK)
1318 if (dev->memory_backed)
1319 return null_handle_memory_backed(cmd, op);
1324 static blk_status_t null_handle_cmd(struct nullb_cmd *cmd, sector_t sector,
1325 sector_t nr_sectors, enum req_opf op)
1327 struct nullb_device *dev = cmd->nq->dev;
1328 struct nullb *nullb = dev->nullb;
1331 if (test_bit(NULLB_DEV_FL_THROTTLED, &dev->flags)) {
1332 sts = null_handle_throttled(cmd);
1333 if (sts != BLK_STS_OK)
1337 if (op == REQ_OP_FLUSH) {
1338 cmd->error = errno_to_blk_status(null_handle_flush(nullb));
1343 cmd->error = null_process_zoned_cmd(cmd, op,
1344 sector, nr_sectors);
1346 cmd->error = null_process_cmd(cmd, op, sector, nr_sectors);
1349 nullb_complete_cmd(cmd);
1353 static enum hrtimer_restart nullb_bwtimer_fn(struct hrtimer *timer)
1355 struct nullb *nullb = container_of(timer, struct nullb, bw_timer);
1356 ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1357 unsigned int mbps = nullb->dev->mbps;
1359 if (atomic_long_read(&nullb->cur_bytes) == mb_per_tick(mbps))
1360 return HRTIMER_NORESTART;
1362 atomic_long_set(&nullb->cur_bytes, mb_per_tick(mbps));
1363 null_restart_queue_async(nullb);
1365 hrtimer_forward_now(&nullb->bw_timer, timer_interval);
1367 return HRTIMER_RESTART;
1370 static void nullb_setup_bwtimer(struct nullb *nullb)
1372 ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1374 hrtimer_init(&nullb->bw_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1375 nullb->bw_timer.function = nullb_bwtimer_fn;
1376 atomic_long_set(&nullb->cur_bytes, mb_per_tick(nullb->dev->mbps));
1377 hrtimer_start(&nullb->bw_timer, timer_interval, HRTIMER_MODE_REL);
1380 static struct nullb_queue *nullb_to_queue(struct nullb *nullb)
1384 if (nullb->nr_queues != 1)
1385 index = raw_smp_processor_id() / ((nr_cpu_ids + nullb->nr_queues - 1) / nullb->nr_queues);
1387 return &nullb->queues[index];
1390 static blk_qc_t null_queue_bio(struct request_queue *q, struct bio *bio)
1392 sector_t sector = bio->bi_iter.bi_sector;
1393 sector_t nr_sectors = bio_sectors(bio);
1394 struct nullb *nullb = q->queuedata;
1395 struct nullb_queue *nq = nullb_to_queue(nullb);
1396 struct nullb_cmd *cmd;
1398 cmd = alloc_cmd(nq, 1);
1401 null_handle_cmd(cmd, sector, nr_sectors, bio_op(bio));
1402 return BLK_QC_T_NONE;
1405 static bool should_timeout_request(struct request *rq)
1407 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1408 if (g_timeout_str[0])
1409 return should_fail(&null_timeout_attr, 1);
1414 static bool should_requeue_request(struct request *rq)
1416 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1417 if (g_requeue_str[0])
1418 return should_fail(&null_requeue_attr, 1);
1423 static enum blk_eh_timer_return null_timeout_rq(struct request *rq, bool res)
1425 pr_info("rq %p timed out\n", rq);
1426 blk_mq_force_complete_rq(rq);
1430 static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx,
1431 const struct blk_mq_queue_data *bd)
1433 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
1434 struct nullb_queue *nq = hctx->driver_data;
1435 sector_t nr_sectors = blk_rq_sectors(bd->rq);
1436 sector_t sector = blk_rq_pos(bd->rq);
1438 might_sleep_if(hctx->flags & BLK_MQ_F_BLOCKING);
1440 if (nq->dev->irqmode == NULL_IRQ_TIMER) {
1441 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1442 cmd->timer.function = null_cmd_timer_expired;
1445 cmd->error = BLK_STS_OK;
1448 blk_mq_start_request(bd->rq);
1450 if (should_requeue_request(bd->rq)) {
1452 * Alternate between hitting the core BUSY path, and the
1453 * driver driven requeue path
1455 nq->requeue_selection++;
1456 if (nq->requeue_selection & 1)
1457 return BLK_STS_RESOURCE;
1459 blk_mq_requeue_request(bd->rq, true);
1463 if (should_timeout_request(bd->rq))
1466 return null_handle_cmd(cmd, sector, nr_sectors, req_op(bd->rq));
1469 static void cleanup_queue(struct nullb_queue *nq)
1475 static void cleanup_queues(struct nullb *nullb)
1479 for (i = 0; i < nullb->nr_queues; i++)
1480 cleanup_queue(&nullb->queues[i]);
1482 kfree(nullb->queues);
1485 static void null_exit_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx)
1487 struct nullb_queue *nq = hctx->driver_data;
1488 struct nullb *nullb = nq->dev->nullb;
1493 static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq)
1495 init_waitqueue_head(&nq->wait);
1496 nq->queue_depth = nullb->queue_depth;
1497 nq->dev = nullb->dev;
1500 static int null_init_hctx(struct blk_mq_hw_ctx *hctx, void *driver_data,
1501 unsigned int hctx_idx)
1503 struct nullb *nullb = hctx->queue->queuedata;
1504 struct nullb_queue *nq;
1506 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1507 if (g_init_hctx_str[0] && should_fail(&null_init_hctx_attr, 1))
1511 nq = &nullb->queues[hctx_idx];
1512 hctx->driver_data = nq;
1513 null_init_queue(nullb, nq);
1519 static const struct blk_mq_ops null_mq_ops = {
1520 .queue_rq = null_queue_rq,
1521 .complete = null_complete_rq,
1522 .timeout = null_timeout_rq,
1523 .init_hctx = null_init_hctx,
1524 .exit_hctx = null_exit_hctx,
1527 static void null_del_dev(struct nullb *nullb)
1529 struct nullb_device *dev;
1536 ida_simple_remove(&nullb_indexes, nullb->index);
1538 list_del_init(&nullb->list);
1540 del_gendisk(nullb->disk);
1542 if (test_bit(NULLB_DEV_FL_THROTTLED, &nullb->dev->flags)) {
1543 hrtimer_cancel(&nullb->bw_timer);
1544 atomic_long_set(&nullb->cur_bytes, LONG_MAX);
1545 null_restart_queue_async(nullb);
1548 blk_cleanup_queue(nullb->q);
1549 if (dev->queue_mode == NULL_Q_MQ &&
1550 nullb->tag_set == &nullb->__tag_set)
1551 blk_mq_free_tag_set(nullb->tag_set);
1552 put_disk(nullb->disk);
1553 cleanup_queues(nullb);
1554 if (null_cache_active(nullb))
1555 null_free_device_storage(nullb->dev, true);
1560 static void null_config_discard(struct nullb *nullb)
1562 if (nullb->dev->discard == false)
1565 if (nullb->dev->zoned) {
1566 nullb->dev->discard = false;
1567 pr_info("discard option is ignored in zoned mode\n");
1571 nullb->q->limits.discard_granularity = nullb->dev->blocksize;
1572 nullb->q->limits.discard_alignment = nullb->dev->blocksize;
1573 blk_queue_max_discard_sectors(nullb->q, UINT_MAX >> 9);
1574 blk_queue_flag_set(QUEUE_FLAG_DISCARD, nullb->q);
1577 static const struct block_device_operations null_ops = {
1578 .owner = THIS_MODULE,
1579 .report_zones = null_report_zones,
1582 static int setup_commands(struct nullb_queue *nq)
1584 struct nullb_cmd *cmd;
1587 nq->cmds = kcalloc(nq->queue_depth, sizeof(*cmd), GFP_KERNEL);
1591 tag_size = ALIGN(nq->queue_depth, BITS_PER_LONG) / BITS_PER_LONG;
1592 nq->tag_map = kcalloc(tag_size, sizeof(unsigned long), GFP_KERNEL);
1598 for (i = 0; i < nq->queue_depth; i++) {
1606 static int setup_queues(struct nullb *nullb)
1608 nullb->queues = kcalloc(nr_cpu_ids, sizeof(struct nullb_queue),
1613 nullb->queue_depth = nullb->dev->hw_queue_depth;
1618 static int init_driver_queues(struct nullb *nullb)
1620 struct nullb_queue *nq;
1623 for (i = 0; i < nullb->dev->submit_queues; i++) {
1624 nq = &nullb->queues[i];
1626 null_init_queue(nullb, nq);
1628 ret = setup_commands(nq);
1636 static int null_gendisk_register(struct nullb *nullb)
1638 sector_t size = ((sector_t)nullb->dev->size * SZ_1M) >> SECTOR_SHIFT;
1639 struct gendisk *disk;
1641 disk = nullb->disk = alloc_disk_node(1, nullb->dev->home_node);
1644 set_capacity(disk, size);
1646 disk->flags |= GENHD_FL_EXT_DEVT | GENHD_FL_SUPPRESS_PARTITION_INFO;
1647 disk->major = null_major;
1648 disk->first_minor = nullb->index;
1649 disk->fops = &null_ops;
1650 disk->private_data = nullb;
1651 disk->queue = nullb->q;
1652 strncpy(disk->disk_name, nullb->disk_name, DISK_NAME_LEN);
1654 if (nullb->dev->zoned) {
1655 int ret = null_register_zoned_dev(nullb);
1665 static int null_init_tag_set(struct nullb *nullb, struct blk_mq_tag_set *set)
1667 set->ops = &null_mq_ops;
1668 set->nr_hw_queues = nullb ? nullb->dev->submit_queues :
1670 set->queue_depth = nullb ? nullb->dev->hw_queue_depth :
1672 set->numa_node = nullb ? nullb->dev->home_node : g_home_node;
1673 set->cmd_size = sizeof(struct nullb_cmd);
1674 set->flags = BLK_MQ_F_SHOULD_MERGE;
1676 set->flags |= BLK_MQ_F_NO_SCHED;
1677 set->driver_data = NULL;
1679 if ((nullb && nullb->dev->blocking) || g_blocking)
1680 set->flags |= BLK_MQ_F_BLOCKING;
1682 return blk_mq_alloc_tag_set(set);
1685 static int null_validate_conf(struct nullb_device *dev)
1687 dev->blocksize = round_down(dev->blocksize, 512);
1688 dev->blocksize = clamp_t(unsigned int, dev->blocksize, 512, 4096);
1690 if (dev->queue_mode == NULL_Q_MQ && dev->use_per_node_hctx) {
1691 if (dev->submit_queues != nr_online_nodes)
1692 dev->submit_queues = nr_online_nodes;
1693 } else if (dev->submit_queues > nr_cpu_ids)
1694 dev->submit_queues = nr_cpu_ids;
1695 else if (dev->submit_queues == 0)
1696 dev->submit_queues = 1;
1698 dev->queue_mode = min_t(unsigned int, dev->queue_mode, NULL_Q_MQ);
1699 dev->irqmode = min_t(unsigned int, dev->irqmode, NULL_IRQ_TIMER);
1701 /* Do memory allocation, so set blocking */
1702 if (dev->memory_backed)
1703 dev->blocking = true;
1704 else /* cache is meaningless */
1705 dev->cache_size = 0;
1706 dev->cache_size = min_t(unsigned long, ULONG_MAX / 1024 / 1024,
1708 dev->mbps = min_t(unsigned int, 1024 * 40, dev->mbps);
1709 /* can not stop a queue */
1710 if (dev->queue_mode == NULL_Q_BIO)
1714 (!dev->zone_size || !is_power_of_2(dev->zone_size))) {
1715 pr_err("zone_size must be power-of-two\n");
1722 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1723 static bool __null_setup_fault(struct fault_attr *attr, char *str)
1728 if (!setup_fault_attr(attr, str))
1736 static bool null_setup_fault(void)
1738 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1739 if (!__null_setup_fault(&null_timeout_attr, g_timeout_str))
1741 if (!__null_setup_fault(&null_requeue_attr, g_requeue_str))
1743 if (!__null_setup_fault(&null_init_hctx_attr, g_init_hctx_str))
1749 static int null_add_dev(struct nullb_device *dev)
1751 struct nullb *nullb;
1754 rv = null_validate_conf(dev);
1758 nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, dev->home_node);
1766 spin_lock_init(&nullb->lock);
1768 rv = setup_queues(nullb);
1770 goto out_free_nullb;
1772 if (dev->queue_mode == NULL_Q_MQ) {
1774 nullb->tag_set = &tag_set;
1777 nullb->tag_set = &nullb->__tag_set;
1778 rv = null_init_tag_set(nullb, nullb->tag_set);
1782 goto out_cleanup_queues;
1784 if (!null_setup_fault())
1785 goto out_cleanup_queues;
1787 nullb->tag_set->timeout = 5 * HZ;
1788 nullb->q = blk_mq_init_queue_data(nullb->tag_set, nullb);
1789 if (IS_ERR(nullb->q)) {
1791 goto out_cleanup_tags;
1793 } else if (dev->queue_mode == NULL_Q_BIO) {
1794 nullb->q = blk_alloc_queue(null_queue_bio, dev->home_node);
1797 goto out_cleanup_queues;
1799 rv = init_driver_queues(nullb);
1801 goto out_cleanup_blk_queue;
1805 set_bit(NULLB_DEV_FL_THROTTLED, &dev->flags);
1806 nullb_setup_bwtimer(nullb);
1809 if (dev->cache_size > 0) {
1810 set_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
1811 blk_queue_write_cache(nullb->q, true, true);
1815 rv = null_init_zoned_dev(dev, nullb->q);
1817 goto out_cleanup_blk_queue;
1820 nullb->q->queuedata = nullb;
1821 blk_queue_flag_set(QUEUE_FLAG_NONROT, nullb->q);
1822 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, nullb->q);
1825 nullb->index = ida_simple_get(&nullb_indexes, 0, 0, GFP_KERNEL);
1826 dev->index = nullb->index;
1827 mutex_unlock(&lock);
1829 blk_queue_logical_block_size(nullb->q, dev->blocksize);
1830 blk_queue_physical_block_size(nullb->q, dev->blocksize);
1832 null_config_discard(nullb);
1834 sprintf(nullb->disk_name, "nullb%d", nullb->index);
1836 rv = null_gendisk_register(nullb);
1838 goto out_cleanup_zone;
1841 list_add_tail(&nullb->list, &nullb_list);
1842 mutex_unlock(&lock);
1846 null_free_zoned_dev(dev);
1847 out_cleanup_blk_queue:
1848 blk_cleanup_queue(nullb->q);
1850 if (dev->queue_mode == NULL_Q_MQ && nullb->tag_set == &nullb->__tag_set)
1851 blk_mq_free_tag_set(nullb->tag_set);
1853 cleanup_queues(nullb);
1861 static int __init null_init(void)
1865 struct nullb *nullb;
1866 struct nullb_device *dev;
1868 if (g_bs > PAGE_SIZE) {
1869 pr_warn("invalid block size\n");
1870 pr_warn("defaults block size to %lu\n", PAGE_SIZE);
1874 if (g_home_node != NUMA_NO_NODE && g_home_node >= nr_online_nodes) {
1875 pr_err("invalid home_node value\n");
1876 g_home_node = NUMA_NO_NODE;
1879 if (g_queue_mode == NULL_Q_RQ) {
1880 pr_err("legacy IO path no longer available\n");
1883 if (g_queue_mode == NULL_Q_MQ && g_use_per_node_hctx) {
1884 if (g_submit_queues != nr_online_nodes) {
1885 pr_warn("submit_queues param is set to %u.\n",
1887 g_submit_queues = nr_online_nodes;
1889 } else if (g_submit_queues > nr_cpu_ids)
1890 g_submit_queues = nr_cpu_ids;
1891 else if (g_submit_queues <= 0)
1892 g_submit_queues = 1;
1894 if (g_queue_mode == NULL_Q_MQ && shared_tags) {
1895 ret = null_init_tag_set(NULL, &tag_set);
1900 config_group_init(&nullb_subsys.su_group);
1901 mutex_init(&nullb_subsys.su_mutex);
1903 ret = configfs_register_subsystem(&nullb_subsys);
1909 null_major = register_blkdev(0, "nullb");
1910 if (null_major < 0) {
1915 for (i = 0; i < nr_devices; i++) {
1916 dev = null_alloc_dev();
1921 ret = null_add_dev(dev);
1928 pr_info("module loaded\n");
1932 while (!list_empty(&nullb_list)) {
1933 nullb = list_entry(nullb_list.next, struct nullb, list);
1935 null_del_dev(nullb);
1938 unregister_blkdev(null_major, "nullb");
1940 configfs_unregister_subsystem(&nullb_subsys);
1942 if (g_queue_mode == NULL_Q_MQ && shared_tags)
1943 blk_mq_free_tag_set(&tag_set);
1947 static void __exit null_exit(void)
1949 struct nullb *nullb;
1951 configfs_unregister_subsystem(&nullb_subsys);
1953 unregister_blkdev(null_major, "nullb");
1956 while (!list_empty(&nullb_list)) {
1957 struct nullb_device *dev;
1959 nullb = list_entry(nullb_list.next, struct nullb, list);
1961 null_del_dev(nullb);
1964 mutex_unlock(&lock);
1966 if (g_queue_mode == NULL_Q_MQ && shared_tags)
1967 blk_mq_free_tag_set(&tag_set);
1970 module_init(null_init);
1971 module_exit(null_exit);
1973 MODULE_AUTHOR("Jens Axboe <axboe@kernel.dk>");
1974 MODULE_LICENSE("GPL");