2 * Add configfs and memory store: Kyungchan Koh <kkc6196@fb.com> and
3 * Shaohua Li <shli@fb.com>
5 #include <linux/module.h>
7 #include <linux/moduleparam.h>
8 #include <linux/sched.h>
10 #include <linux/blkdev.h>
11 #include <linux/init.h>
12 #include <linux/slab.h>
13 #include <linux/blk-mq.h>
14 #include <linux/hrtimer.h>
15 #include <linux/configfs.h>
16 #include <linux/badblocks.h>
17 #include <linux/fault-inject.h>
19 #define SECTOR_SHIFT 9
20 #define PAGE_SECTORS_SHIFT (PAGE_SHIFT - SECTOR_SHIFT)
21 #define PAGE_SECTORS (1 << PAGE_SECTORS_SHIFT)
22 #define SECTOR_SIZE (1 << SECTOR_SHIFT)
23 #define SECTOR_MASK (PAGE_SECTORS - 1)
27 #define TICKS_PER_SEC 50ULL
28 #define TIMER_INTERVAL (NSEC_PER_SEC / TICKS_PER_SEC)
30 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
31 static DECLARE_FAULT_ATTR(null_timeout_attr
);
32 static DECLARE_FAULT_ATTR(null_requeue_attr
);
35 static inline u64
mb_per_tick(int mbps
)
37 return (1 << 20) / TICKS_PER_SEC
* ((u64
) mbps
);
41 struct list_head list
;
42 struct llist_node ll_list
;
43 struct __call_single_data csd
;
48 struct nullb_queue
*nq
;
53 unsigned long *tag_map
;
54 wait_queue_head_t wait
;
55 unsigned int queue_depth
;
56 struct nullb_device
*dev
;
57 unsigned int requeue_selection
;
59 struct nullb_cmd
*cmds
;
63 * Status flags for nullb_device.
65 * CONFIGURED: Device has been configured and turned on. Cannot reconfigure.
66 * UP: Device is currently on and visible in userspace.
67 * THROTTLED: Device is being throttled.
68 * CACHE: Device is using a write-back cache.
70 enum nullb_device_flags
{
71 NULLB_DEV_FL_CONFIGURED
= 0,
73 NULLB_DEV_FL_THROTTLED
= 2,
74 NULLB_DEV_FL_CACHE
= 3,
78 * nullb_page is a page in memory for nullb devices.
80 * @page: The page holding the data.
81 * @bitmap: The bitmap represents which sector in the page has data.
82 * Each bit represents one block size. For example, sector 8
83 * will use the 7th bit
84 * The highest 2 bits of bitmap are for special purpose. LOCK means the cache
85 * page is being flushing to storage. FREE means the cache page is freed and
86 * should be skipped from flushing to storage. Please see
87 * null_make_cache_space
93 #define NULLB_PAGE_LOCK (sizeof(unsigned long) * 8 - 1)
94 #define NULLB_PAGE_FREE (sizeof(unsigned long) * 8 - 2)
98 struct config_item item
;
99 struct radix_tree_root data
; /* data stored in the disk */
100 struct radix_tree_root cache
; /* disk cache data */
101 unsigned long flags
; /* device flags */
102 unsigned int curr_cache
;
103 struct badblocks badblocks
;
105 unsigned long size
; /* device size in MB */
106 unsigned long completion_nsec
; /* time in ns to complete a request */
107 unsigned long cache_size
; /* disk cache size in MB */
108 unsigned int submit_queues
; /* number of submission queues */
109 unsigned int home_node
; /* home node for the device */
110 unsigned int queue_mode
; /* block interface */
111 unsigned int blocksize
; /* block size */
112 unsigned int irqmode
; /* IRQ completion handler */
113 unsigned int hw_queue_depth
; /* queue depth */
114 unsigned int index
; /* index of the disk, only valid with a disk */
115 unsigned int mbps
; /* Bandwidth throttle cap (in MB/s) */
116 bool blocking
; /* blocking blk-mq device */
117 bool use_per_node_hctx
; /* use per-node allocation for hardware context */
118 bool power
; /* power on/off the device */
119 bool memory_backed
; /* if data is stored in memory */
120 bool discard
; /* if support discard */
124 struct nullb_device
*dev
;
125 struct list_head list
;
127 struct request_queue
*q
;
128 struct gendisk
*disk
;
129 struct blk_mq_tag_set
*tag_set
;
130 struct blk_mq_tag_set __tag_set
;
131 unsigned int queue_depth
;
132 atomic_long_t cur_bytes
;
133 struct hrtimer bw_timer
;
134 unsigned long cache_flush_pos
;
137 struct nullb_queue
*queues
;
138 unsigned int nr_queues
;
139 char disk_name
[DISK_NAME_LEN
];
142 static LIST_HEAD(nullb_list
);
143 static struct mutex lock
;
144 static int null_major
;
145 static DEFINE_IDA(nullb_indexes
);
146 static struct blk_mq_tag_set tag_set
;
150 NULL_IRQ_SOFTIRQ
= 1,
160 static int g_no_sched
;
161 module_param_named(no_sched
, g_no_sched
, int, S_IRUGO
);
162 MODULE_PARM_DESC(no_sched
, "No io scheduler");
164 static int g_submit_queues
= 1;
165 module_param_named(submit_queues
, g_submit_queues
, int, S_IRUGO
);
166 MODULE_PARM_DESC(submit_queues
, "Number of submission queues");
168 static int g_home_node
= NUMA_NO_NODE
;
169 module_param_named(home_node
, g_home_node
, int, S_IRUGO
);
170 MODULE_PARM_DESC(home_node
, "Home node for the device");
172 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
173 static char g_timeout_str
[80];
174 module_param_string(timeout
, g_timeout_str
, sizeof(g_timeout_str
), S_IRUGO
);
176 static char g_requeue_str
[80];
177 module_param_string(requeue
, g_requeue_str
, sizeof(g_requeue_str
), S_IRUGO
);
180 static int g_queue_mode
= NULL_Q_MQ
;
182 static int null_param_store_val(const char *str
, int *val
, int min
, int max
)
186 ret
= kstrtoint(str
, 10, &new_val
);
190 if (new_val
< min
|| new_val
> max
)
197 static int null_set_queue_mode(const char *str
, const struct kernel_param
*kp
)
199 return null_param_store_val(str
, &g_queue_mode
, NULL_Q_BIO
, NULL_Q_MQ
);
202 static const struct kernel_param_ops null_queue_mode_param_ops
= {
203 .set
= null_set_queue_mode
,
204 .get
= param_get_int
,
207 device_param_cb(queue_mode
, &null_queue_mode_param_ops
, &g_queue_mode
, S_IRUGO
);
208 MODULE_PARM_DESC(queue_mode
, "Block interface to use (0=bio,1=rq,2=multiqueue)");
210 static int g_gb
= 250;
211 module_param_named(gb
, g_gb
, int, S_IRUGO
);
212 MODULE_PARM_DESC(gb
, "Size in GB");
214 static int g_bs
= 512;
215 module_param_named(bs
, g_bs
, int, S_IRUGO
);
216 MODULE_PARM_DESC(bs
, "Block size (in bytes)");
218 static int nr_devices
= 1;
219 module_param(nr_devices
, int, S_IRUGO
);
220 MODULE_PARM_DESC(nr_devices
, "Number of devices to register");
222 static bool g_blocking
;
223 module_param_named(blocking
, g_blocking
, bool, S_IRUGO
);
224 MODULE_PARM_DESC(blocking
, "Register as a blocking blk-mq driver device");
226 static bool shared_tags
;
227 module_param(shared_tags
, bool, S_IRUGO
);
228 MODULE_PARM_DESC(shared_tags
, "Share tag set between devices for blk-mq");
230 static int g_irqmode
= NULL_IRQ_SOFTIRQ
;
232 static int null_set_irqmode(const char *str
, const struct kernel_param
*kp
)
234 return null_param_store_val(str
, &g_irqmode
, NULL_IRQ_NONE
,
238 static const struct kernel_param_ops null_irqmode_param_ops
= {
239 .set
= null_set_irqmode
,
240 .get
= param_get_int
,
243 device_param_cb(irqmode
, &null_irqmode_param_ops
, &g_irqmode
, S_IRUGO
);
244 MODULE_PARM_DESC(irqmode
, "IRQ completion handler. 0-none, 1-softirq, 2-timer");
246 static unsigned long g_completion_nsec
= 10000;
247 module_param_named(completion_nsec
, g_completion_nsec
, ulong
, S_IRUGO
);
248 MODULE_PARM_DESC(completion_nsec
, "Time in ns to complete a request in hardware. Default: 10,000ns");
250 static int g_hw_queue_depth
= 64;
251 module_param_named(hw_queue_depth
, g_hw_queue_depth
, int, S_IRUGO
);
252 MODULE_PARM_DESC(hw_queue_depth
, "Queue depth for each hardware queue. Default: 64");
254 static bool g_use_per_node_hctx
;
255 module_param_named(use_per_node_hctx
, g_use_per_node_hctx
, bool, S_IRUGO
);
256 MODULE_PARM_DESC(use_per_node_hctx
, "Use per-node allocation for hardware context queues. Default: false");
258 static struct nullb_device
*null_alloc_dev(void);
259 static void null_free_dev(struct nullb_device
*dev
);
260 static void null_del_dev(struct nullb
*nullb
);
261 static int null_add_dev(struct nullb_device
*dev
);
262 static void null_free_device_storage(struct nullb_device
*dev
, bool is_cache
);
264 static inline struct nullb_device
*to_nullb_device(struct config_item
*item
)
266 return item
? container_of(item
, struct nullb_device
, item
) : NULL
;
269 static inline ssize_t
nullb_device_uint_attr_show(unsigned int val
, char *page
)
271 return snprintf(page
, PAGE_SIZE
, "%u\n", val
);
274 static inline ssize_t
nullb_device_ulong_attr_show(unsigned long val
,
277 return snprintf(page
, PAGE_SIZE
, "%lu\n", val
);
280 static inline ssize_t
nullb_device_bool_attr_show(bool val
, char *page
)
282 return snprintf(page
, PAGE_SIZE
, "%u\n", val
);
285 static ssize_t
nullb_device_uint_attr_store(unsigned int *val
,
286 const char *page
, size_t count
)
291 result
= kstrtouint(page
, 0, &tmp
);
299 static ssize_t
nullb_device_ulong_attr_store(unsigned long *val
,
300 const char *page
, size_t count
)
305 result
= kstrtoul(page
, 0, &tmp
);
313 static ssize_t
nullb_device_bool_attr_store(bool *val
, const char *page
,
319 result
= kstrtobool(page
, &tmp
);
327 /* The following macro should only be used with TYPE = {uint, ulong, bool}. */
328 #define NULLB_DEVICE_ATTR(NAME, TYPE) \
330 nullb_device_##NAME##_show(struct config_item *item, char *page) \
332 return nullb_device_##TYPE##_attr_show( \
333 to_nullb_device(item)->NAME, page); \
336 nullb_device_##NAME##_store(struct config_item *item, const char *page, \
339 if (test_bit(NULLB_DEV_FL_CONFIGURED, &to_nullb_device(item)->flags)) \
341 return nullb_device_##TYPE##_attr_store( \
342 &to_nullb_device(item)->NAME, page, count); \
344 CONFIGFS_ATTR(nullb_device_, NAME);
346 NULLB_DEVICE_ATTR(size
, ulong
);
347 NULLB_DEVICE_ATTR(completion_nsec
, ulong
);
348 NULLB_DEVICE_ATTR(submit_queues
, uint
);
349 NULLB_DEVICE_ATTR(home_node
, uint
);
350 NULLB_DEVICE_ATTR(queue_mode
, uint
);
351 NULLB_DEVICE_ATTR(blocksize
, uint
);
352 NULLB_DEVICE_ATTR(irqmode
, uint
);
353 NULLB_DEVICE_ATTR(hw_queue_depth
, uint
);
354 NULLB_DEVICE_ATTR(index
, uint
);
355 NULLB_DEVICE_ATTR(blocking
, bool);
356 NULLB_DEVICE_ATTR(use_per_node_hctx
, bool);
357 NULLB_DEVICE_ATTR(memory_backed
, bool);
358 NULLB_DEVICE_ATTR(discard
, bool);
359 NULLB_DEVICE_ATTR(mbps
, uint
);
360 NULLB_DEVICE_ATTR(cache_size
, ulong
);
362 static ssize_t
nullb_device_power_show(struct config_item
*item
, char *page
)
364 return nullb_device_bool_attr_show(to_nullb_device(item
)->power
, page
);
367 static ssize_t
nullb_device_power_store(struct config_item
*item
,
368 const char *page
, size_t count
)
370 struct nullb_device
*dev
= to_nullb_device(item
);
374 ret
= nullb_device_bool_attr_store(&newp
, page
, count
);
378 if (!dev
->power
&& newp
) {
379 if (test_and_set_bit(NULLB_DEV_FL_UP
, &dev
->flags
))
381 if (null_add_dev(dev
)) {
382 clear_bit(NULLB_DEV_FL_UP
, &dev
->flags
);
386 set_bit(NULLB_DEV_FL_CONFIGURED
, &dev
->flags
);
388 } else if (dev
->power
&& !newp
) {
391 null_del_dev(dev
->nullb
);
393 clear_bit(NULLB_DEV_FL_UP
, &dev
->flags
);
399 CONFIGFS_ATTR(nullb_device_
, power
);
401 static ssize_t
nullb_device_badblocks_show(struct config_item
*item
, char *page
)
403 struct nullb_device
*t_dev
= to_nullb_device(item
);
405 return badblocks_show(&t_dev
->badblocks
, page
, 0);
408 static ssize_t
nullb_device_badblocks_store(struct config_item
*item
,
409 const char *page
, size_t count
)
411 struct nullb_device
*t_dev
= to_nullb_device(item
);
412 char *orig
, *buf
, *tmp
;
416 orig
= kstrndup(page
, count
, GFP_KERNEL
);
420 buf
= strstrip(orig
);
423 if (buf
[0] != '+' && buf
[0] != '-')
425 tmp
= strchr(&buf
[1], '-');
429 ret
= kstrtoull(buf
+ 1, 0, &start
);
432 ret
= kstrtoull(tmp
+ 1, 0, &end
);
438 /* enable badblocks */
439 cmpxchg(&t_dev
->badblocks
.shift
, -1, 0);
441 ret
= badblocks_set(&t_dev
->badblocks
, start
,
444 ret
= badblocks_clear(&t_dev
->badblocks
, start
,
452 CONFIGFS_ATTR(nullb_device_
, badblocks
);
454 static struct configfs_attribute
*nullb_device_attrs
[] = {
455 &nullb_device_attr_size
,
456 &nullb_device_attr_completion_nsec
,
457 &nullb_device_attr_submit_queues
,
458 &nullb_device_attr_home_node
,
459 &nullb_device_attr_queue_mode
,
460 &nullb_device_attr_blocksize
,
461 &nullb_device_attr_irqmode
,
462 &nullb_device_attr_hw_queue_depth
,
463 &nullb_device_attr_index
,
464 &nullb_device_attr_blocking
,
465 &nullb_device_attr_use_per_node_hctx
,
466 &nullb_device_attr_power
,
467 &nullb_device_attr_memory_backed
,
468 &nullb_device_attr_discard
,
469 &nullb_device_attr_mbps
,
470 &nullb_device_attr_cache_size
,
471 &nullb_device_attr_badblocks
,
475 static void nullb_device_release(struct config_item
*item
)
477 struct nullb_device
*dev
= to_nullb_device(item
);
479 null_free_device_storage(dev
, false);
483 static struct configfs_item_operations nullb_device_ops
= {
484 .release
= nullb_device_release
,
487 static const struct config_item_type nullb_device_type
= {
488 .ct_item_ops
= &nullb_device_ops
,
489 .ct_attrs
= nullb_device_attrs
,
490 .ct_owner
= THIS_MODULE
,
494 config_item
*nullb_group_make_item(struct config_group
*group
, const char *name
)
496 struct nullb_device
*dev
;
498 dev
= null_alloc_dev();
500 return ERR_PTR(-ENOMEM
);
502 config_item_init_type_name(&dev
->item
, name
, &nullb_device_type
);
508 nullb_group_drop_item(struct config_group
*group
, struct config_item
*item
)
510 struct nullb_device
*dev
= to_nullb_device(item
);
512 if (test_and_clear_bit(NULLB_DEV_FL_UP
, &dev
->flags
)) {
515 null_del_dev(dev
->nullb
);
519 config_item_put(item
);
522 static ssize_t
memb_group_features_show(struct config_item
*item
, char *page
)
524 return snprintf(page
, PAGE_SIZE
, "memory_backed,discard,bandwidth,cache,badblocks\n");
527 CONFIGFS_ATTR_RO(memb_group_
, features
);
529 static struct configfs_attribute
*nullb_group_attrs
[] = {
530 &memb_group_attr_features
,
534 static struct configfs_group_operations nullb_group_ops
= {
535 .make_item
= nullb_group_make_item
,
536 .drop_item
= nullb_group_drop_item
,
539 static const struct config_item_type nullb_group_type
= {
540 .ct_group_ops
= &nullb_group_ops
,
541 .ct_attrs
= nullb_group_attrs
,
542 .ct_owner
= THIS_MODULE
,
545 static struct configfs_subsystem nullb_subsys
= {
548 .ci_namebuf
= "nullb",
549 .ci_type
= &nullb_group_type
,
554 static inline int null_cache_active(struct nullb
*nullb
)
556 return test_bit(NULLB_DEV_FL_CACHE
, &nullb
->dev
->flags
);
559 static struct nullb_device
*null_alloc_dev(void)
561 struct nullb_device
*dev
;
563 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
566 INIT_RADIX_TREE(&dev
->data
, GFP_ATOMIC
);
567 INIT_RADIX_TREE(&dev
->cache
, GFP_ATOMIC
);
568 if (badblocks_init(&dev
->badblocks
, 0)) {
573 dev
->size
= g_gb
* 1024;
574 dev
->completion_nsec
= g_completion_nsec
;
575 dev
->submit_queues
= g_submit_queues
;
576 dev
->home_node
= g_home_node
;
577 dev
->queue_mode
= g_queue_mode
;
578 dev
->blocksize
= g_bs
;
579 dev
->irqmode
= g_irqmode
;
580 dev
->hw_queue_depth
= g_hw_queue_depth
;
581 dev
->blocking
= g_blocking
;
582 dev
->use_per_node_hctx
= g_use_per_node_hctx
;
586 static void null_free_dev(struct nullb_device
*dev
)
591 badblocks_exit(&dev
->badblocks
);
595 static void put_tag(struct nullb_queue
*nq
, unsigned int tag
)
597 clear_bit_unlock(tag
, nq
->tag_map
);
599 if (waitqueue_active(&nq
->wait
))
603 static unsigned int get_tag(struct nullb_queue
*nq
)
608 tag
= find_first_zero_bit(nq
->tag_map
, nq
->queue_depth
);
609 if (tag
>= nq
->queue_depth
)
611 } while (test_and_set_bit_lock(tag
, nq
->tag_map
));
616 static void free_cmd(struct nullb_cmd
*cmd
)
618 put_tag(cmd
->nq
, cmd
->tag
);
621 static enum hrtimer_restart
null_cmd_timer_expired(struct hrtimer
*timer
);
623 static struct nullb_cmd
*__alloc_cmd(struct nullb_queue
*nq
)
625 struct nullb_cmd
*cmd
;
630 cmd
= &nq
->cmds
[tag
];
633 if (nq
->dev
->irqmode
== NULL_IRQ_TIMER
) {
634 hrtimer_init(&cmd
->timer
, CLOCK_MONOTONIC
,
636 cmd
->timer
.function
= null_cmd_timer_expired
;
644 static struct nullb_cmd
*alloc_cmd(struct nullb_queue
*nq
, int can_wait
)
646 struct nullb_cmd
*cmd
;
649 cmd
= __alloc_cmd(nq
);
650 if (cmd
|| !can_wait
)
654 prepare_to_wait(&nq
->wait
, &wait
, TASK_UNINTERRUPTIBLE
);
655 cmd
= __alloc_cmd(nq
);
662 finish_wait(&nq
->wait
, &wait
);
666 static void end_cmd(struct nullb_cmd
*cmd
)
668 struct request_queue
*q
= NULL
;
669 int queue_mode
= cmd
->nq
->dev
->queue_mode
;
674 switch (queue_mode
) {
676 blk_mq_end_request(cmd
->rq
, cmd
->error
);
679 INIT_LIST_HEAD(&cmd
->rq
->queuelist
);
680 blk_end_request_all(cmd
->rq
, cmd
->error
);
683 cmd
->bio
->bi_status
= cmd
->error
;
690 /* Restart queue if needed, as we are freeing a tag */
691 if (queue_mode
== NULL_Q_RQ
&& blk_queue_stopped(q
)) {
694 spin_lock_irqsave(q
->queue_lock
, flags
);
695 blk_start_queue_async(q
);
696 spin_unlock_irqrestore(q
->queue_lock
, flags
);
700 static enum hrtimer_restart
null_cmd_timer_expired(struct hrtimer
*timer
)
702 end_cmd(container_of(timer
, struct nullb_cmd
, timer
));
704 return HRTIMER_NORESTART
;
707 static void null_cmd_end_timer(struct nullb_cmd
*cmd
)
709 ktime_t kt
= cmd
->nq
->dev
->completion_nsec
;
711 hrtimer_start(&cmd
->timer
, kt
, HRTIMER_MODE_REL
);
714 static void null_softirq_done_fn(struct request
*rq
)
716 struct nullb
*nullb
= rq
->q
->queuedata
;
718 if (nullb
->dev
->queue_mode
== NULL_Q_MQ
)
719 end_cmd(blk_mq_rq_to_pdu(rq
));
721 end_cmd(rq
->special
);
724 static struct nullb_page
*null_alloc_page(gfp_t gfp_flags
)
726 struct nullb_page
*t_page
;
728 t_page
= kmalloc(sizeof(struct nullb_page
), gfp_flags
);
732 t_page
->page
= alloc_pages(gfp_flags
, 0);
744 static void null_free_page(struct nullb_page
*t_page
)
746 __set_bit(NULLB_PAGE_FREE
, &t_page
->bitmap
);
747 if (test_bit(NULLB_PAGE_LOCK
, &t_page
->bitmap
))
749 __free_page(t_page
->page
);
753 static void null_free_sector(struct nullb
*nullb
, sector_t sector
,
756 unsigned int sector_bit
;
758 struct nullb_page
*t_page
, *ret
;
759 struct radix_tree_root
*root
;
761 root
= is_cache
? &nullb
->dev
->cache
: &nullb
->dev
->data
;
762 idx
= sector
>> PAGE_SECTORS_SHIFT
;
763 sector_bit
= (sector
& SECTOR_MASK
);
765 t_page
= radix_tree_lookup(root
, idx
);
767 __clear_bit(sector_bit
, &t_page
->bitmap
);
769 if (!t_page
->bitmap
) {
770 ret
= radix_tree_delete_item(root
, idx
, t_page
);
771 WARN_ON(ret
!= t_page
);
774 nullb
->dev
->curr_cache
-= PAGE_SIZE
;
779 static struct nullb_page
*null_radix_tree_insert(struct nullb
*nullb
, u64 idx
,
780 struct nullb_page
*t_page
, bool is_cache
)
782 struct radix_tree_root
*root
;
784 root
= is_cache
? &nullb
->dev
->cache
: &nullb
->dev
->data
;
786 if (radix_tree_insert(root
, idx
, t_page
)) {
787 null_free_page(t_page
);
788 t_page
= radix_tree_lookup(root
, idx
);
789 WARN_ON(!t_page
|| t_page
->page
->index
!= idx
);
791 nullb
->dev
->curr_cache
+= PAGE_SIZE
;
796 static void null_free_device_storage(struct nullb_device
*dev
, bool is_cache
)
798 unsigned long pos
= 0;
800 struct nullb_page
*ret
, *t_pages
[FREE_BATCH
];
801 struct radix_tree_root
*root
;
803 root
= is_cache
? &dev
->cache
: &dev
->data
;
808 nr_pages
= radix_tree_gang_lookup(root
,
809 (void **)t_pages
, pos
, FREE_BATCH
);
811 for (i
= 0; i
< nr_pages
; i
++) {
812 pos
= t_pages
[i
]->page
->index
;
813 ret
= radix_tree_delete_item(root
, pos
, t_pages
[i
]);
814 WARN_ON(ret
!= t_pages
[i
]);
819 } while (nr_pages
== FREE_BATCH
);
825 static struct nullb_page
*__null_lookup_page(struct nullb
*nullb
,
826 sector_t sector
, bool for_write
, bool is_cache
)
828 unsigned int sector_bit
;
830 struct nullb_page
*t_page
;
831 struct radix_tree_root
*root
;
833 idx
= sector
>> PAGE_SECTORS_SHIFT
;
834 sector_bit
= (sector
& SECTOR_MASK
);
836 root
= is_cache
? &nullb
->dev
->cache
: &nullb
->dev
->data
;
837 t_page
= radix_tree_lookup(root
, idx
);
838 WARN_ON(t_page
&& t_page
->page
->index
!= idx
);
840 if (t_page
&& (for_write
|| test_bit(sector_bit
, &t_page
->bitmap
)))
846 static struct nullb_page
*null_lookup_page(struct nullb
*nullb
,
847 sector_t sector
, bool for_write
, bool ignore_cache
)
849 struct nullb_page
*page
= NULL
;
852 page
= __null_lookup_page(nullb
, sector
, for_write
, true);
855 return __null_lookup_page(nullb
, sector
, for_write
, false);
858 static struct nullb_page
*null_insert_page(struct nullb
*nullb
,
859 sector_t sector
, bool ignore_cache
)
862 struct nullb_page
*t_page
;
864 t_page
= null_lookup_page(nullb
, sector
, true, ignore_cache
);
868 spin_unlock_irq(&nullb
->lock
);
870 t_page
= null_alloc_page(GFP_NOIO
);
874 if (radix_tree_preload(GFP_NOIO
))
877 spin_lock_irq(&nullb
->lock
);
878 idx
= sector
>> PAGE_SECTORS_SHIFT
;
879 t_page
->page
->index
= idx
;
880 t_page
= null_radix_tree_insert(nullb
, idx
, t_page
, !ignore_cache
);
881 radix_tree_preload_end();
885 null_free_page(t_page
);
887 spin_lock_irq(&nullb
->lock
);
888 return null_lookup_page(nullb
, sector
, true, ignore_cache
);
891 static int null_flush_cache_page(struct nullb
*nullb
, struct nullb_page
*c_page
)
896 struct nullb_page
*t_page
, *ret
;
899 idx
= c_page
->page
->index
;
901 t_page
= null_insert_page(nullb
, idx
<< PAGE_SECTORS_SHIFT
, true);
903 __clear_bit(NULLB_PAGE_LOCK
, &c_page
->bitmap
);
904 if (test_bit(NULLB_PAGE_FREE
, &c_page
->bitmap
)) {
905 null_free_page(c_page
);
906 if (t_page
&& t_page
->bitmap
== 0) {
907 ret
= radix_tree_delete_item(&nullb
->dev
->data
,
909 null_free_page(t_page
);
917 src
= kmap_atomic(c_page
->page
);
918 dst
= kmap_atomic(t_page
->page
);
920 for (i
= 0; i
< PAGE_SECTORS
;
921 i
+= (nullb
->dev
->blocksize
>> SECTOR_SHIFT
)) {
922 if (test_bit(i
, &c_page
->bitmap
)) {
923 offset
= (i
<< SECTOR_SHIFT
);
924 memcpy(dst
+ offset
, src
+ offset
,
925 nullb
->dev
->blocksize
);
926 __set_bit(i
, &t_page
->bitmap
);
933 ret
= radix_tree_delete_item(&nullb
->dev
->cache
, idx
, c_page
);
935 nullb
->dev
->curr_cache
-= PAGE_SIZE
;
940 static int null_make_cache_space(struct nullb
*nullb
, unsigned long n
)
942 int i
, err
, nr_pages
;
943 struct nullb_page
*c_pages
[FREE_BATCH
];
944 unsigned long flushed
= 0, one_round
;
947 if ((nullb
->dev
->cache_size
* 1024 * 1024) >
948 nullb
->dev
->curr_cache
+ n
|| nullb
->dev
->curr_cache
== 0)
951 nr_pages
= radix_tree_gang_lookup(&nullb
->dev
->cache
,
952 (void **)c_pages
, nullb
->cache_flush_pos
, FREE_BATCH
);
954 * nullb_flush_cache_page could unlock before using the c_pages. To
955 * avoid race, we don't allow page free
957 for (i
= 0; i
< nr_pages
; i
++) {
958 nullb
->cache_flush_pos
= c_pages
[i
]->page
->index
;
960 * We found the page which is being flushed to disk by other
963 if (test_bit(NULLB_PAGE_LOCK
, &c_pages
[i
]->bitmap
))
966 __set_bit(NULLB_PAGE_LOCK
, &c_pages
[i
]->bitmap
);
970 for (i
= 0; i
< nr_pages
; i
++) {
971 if (c_pages
[i
] == NULL
)
973 err
= null_flush_cache_page(nullb
, c_pages
[i
]);
978 flushed
+= one_round
<< PAGE_SHIFT
;
982 nullb
->cache_flush_pos
= 0;
983 if (one_round
== 0) {
984 /* give other threads a chance */
985 spin_unlock_irq(&nullb
->lock
);
986 spin_lock_irq(&nullb
->lock
);
993 static int copy_to_nullb(struct nullb
*nullb
, struct page
*source
,
994 unsigned int off
, sector_t sector
, size_t n
, bool is_fua
)
996 size_t temp
, count
= 0;
998 struct nullb_page
*t_page
;
1002 temp
= min_t(size_t, nullb
->dev
->blocksize
, n
- count
);
1004 if (null_cache_active(nullb
) && !is_fua
)
1005 null_make_cache_space(nullb
, PAGE_SIZE
);
1007 offset
= (sector
& SECTOR_MASK
) << SECTOR_SHIFT
;
1008 t_page
= null_insert_page(nullb
, sector
,
1009 !null_cache_active(nullb
) || is_fua
);
1013 src
= kmap_atomic(source
);
1014 dst
= kmap_atomic(t_page
->page
);
1015 memcpy(dst
+ offset
, src
+ off
+ count
, temp
);
1019 __set_bit(sector
& SECTOR_MASK
, &t_page
->bitmap
);
1022 null_free_sector(nullb
, sector
, true);
1025 sector
+= temp
>> SECTOR_SHIFT
;
1030 static int copy_from_nullb(struct nullb
*nullb
, struct page
*dest
,
1031 unsigned int off
, sector_t sector
, size_t n
)
1033 size_t temp
, count
= 0;
1034 unsigned int offset
;
1035 struct nullb_page
*t_page
;
1039 temp
= min_t(size_t, nullb
->dev
->blocksize
, n
- count
);
1041 offset
= (sector
& SECTOR_MASK
) << SECTOR_SHIFT
;
1042 t_page
= null_lookup_page(nullb
, sector
, false,
1043 !null_cache_active(nullb
));
1045 dst
= kmap_atomic(dest
);
1047 memset(dst
+ off
+ count
, 0, temp
);
1050 src
= kmap_atomic(t_page
->page
);
1051 memcpy(dst
+ off
+ count
, src
+ offset
, temp
);
1057 sector
+= temp
>> SECTOR_SHIFT
;
1062 static void null_handle_discard(struct nullb
*nullb
, sector_t sector
, size_t n
)
1066 spin_lock_irq(&nullb
->lock
);
1068 temp
= min_t(size_t, n
, nullb
->dev
->blocksize
);
1069 null_free_sector(nullb
, sector
, false);
1070 if (null_cache_active(nullb
))
1071 null_free_sector(nullb
, sector
, true);
1072 sector
+= temp
>> SECTOR_SHIFT
;
1075 spin_unlock_irq(&nullb
->lock
);
1078 static int null_handle_flush(struct nullb
*nullb
)
1082 if (!null_cache_active(nullb
))
1085 spin_lock_irq(&nullb
->lock
);
1087 err
= null_make_cache_space(nullb
,
1088 nullb
->dev
->cache_size
* 1024 * 1024);
1089 if (err
|| nullb
->dev
->curr_cache
== 0)
1093 WARN_ON(!radix_tree_empty(&nullb
->dev
->cache
));
1094 spin_unlock_irq(&nullb
->lock
);
1098 static int null_transfer(struct nullb
*nullb
, struct page
*page
,
1099 unsigned int len
, unsigned int off
, bool is_write
, sector_t sector
,
1105 err
= copy_from_nullb(nullb
, page
, off
, sector
, len
);
1106 flush_dcache_page(page
);
1108 flush_dcache_page(page
);
1109 err
= copy_to_nullb(nullb
, page
, off
, sector
, len
, is_fua
);
1115 static int null_handle_rq(struct nullb_cmd
*cmd
)
1117 struct request
*rq
= cmd
->rq
;
1118 struct nullb
*nullb
= cmd
->nq
->dev
->nullb
;
1122 struct req_iterator iter
;
1123 struct bio_vec bvec
;
1125 sector
= blk_rq_pos(rq
);
1127 if (req_op(rq
) == REQ_OP_DISCARD
) {
1128 null_handle_discard(nullb
, sector
, blk_rq_bytes(rq
));
1132 spin_lock_irq(&nullb
->lock
);
1133 rq_for_each_segment(bvec
, rq
, iter
) {
1135 err
= null_transfer(nullb
, bvec
.bv_page
, len
, bvec
.bv_offset
,
1136 op_is_write(req_op(rq
)), sector
,
1137 req_op(rq
) & REQ_FUA
);
1139 spin_unlock_irq(&nullb
->lock
);
1142 sector
+= len
>> SECTOR_SHIFT
;
1144 spin_unlock_irq(&nullb
->lock
);
1149 static int null_handle_bio(struct nullb_cmd
*cmd
)
1151 struct bio
*bio
= cmd
->bio
;
1152 struct nullb
*nullb
= cmd
->nq
->dev
->nullb
;
1156 struct bio_vec bvec
;
1157 struct bvec_iter iter
;
1159 sector
= bio
->bi_iter
.bi_sector
;
1161 if (bio_op(bio
) == REQ_OP_DISCARD
) {
1162 null_handle_discard(nullb
, sector
,
1163 bio_sectors(bio
) << SECTOR_SHIFT
);
1167 spin_lock_irq(&nullb
->lock
);
1168 bio_for_each_segment(bvec
, bio
, iter
) {
1170 err
= null_transfer(nullb
, bvec
.bv_page
, len
, bvec
.bv_offset
,
1171 op_is_write(bio_op(bio
)), sector
,
1172 bio_op(bio
) & REQ_FUA
);
1174 spin_unlock_irq(&nullb
->lock
);
1177 sector
+= len
>> SECTOR_SHIFT
;
1179 spin_unlock_irq(&nullb
->lock
);
1183 static void null_stop_queue(struct nullb
*nullb
)
1185 struct request_queue
*q
= nullb
->q
;
1187 if (nullb
->dev
->queue_mode
== NULL_Q_MQ
)
1188 blk_mq_stop_hw_queues(q
);
1190 spin_lock_irq(q
->queue_lock
);
1192 spin_unlock_irq(q
->queue_lock
);
1196 static void null_restart_queue_async(struct nullb
*nullb
)
1198 struct request_queue
*q
= nullb
->q
;
1199 unsigned long flags
;
1201 if (nullb
->dev
->queue_mode
== NULL_Q_MQ
)
1202 blk_mq_start_stopped_hw_queues(q
, true);
1204 spin_lock_irqsave(q
->queue_lock
, flags
);
1205 blk_start_queue_async(q
);
1206 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1210 static blk_status_t
null_handle_cmd(struct nullb_cmd
*cmd
)
1212 struct nullb_device
*dev
= cmd
->nq
->dev
;
1213 struct nullb
*nullb
= dev
->nullb
;
1216 if (test_bit(NULLB_DEV_FL_THROTTLED
, &dev
->flags
)) {
1217 struct request
*rq
= cmd
->rq
;
1219 if (!hrtimer_active(&nullb
->bw_timer
))
1220 hrtimer_restart(&nullb
->bw_timer
);
1222 if (atomic_long_sub_return(blk_rq_bytes(rq
),
1223 &nullb
->cur_bytes
) < 0) {
1224 null_stop_queue(nullb
);
1225 /* race with timer */
1226 if (atomic_long_read(&nullb
->cur_bytes
) > 0)
1227 null_restart_queue_async(nullb
);
1228 if (dev
->queue_mode
== NULL_Q_RQ
) {
1229 struct request_queue
*q
= nullb
->q
;
1231 spin_lock_irq(q
->queue_lock
);
1232 rq
->rq_flags
|= RQF_DONTPREP
;
1233 blk_requeue_request(q
, rq
);
1234 spin_unlock_irq(q
->queue_lock
);
1237 /* requeue request */
1238 return BLK_STS_DEV_RESOURCE
;
1242 if (nullb
->dev
->badblocks
.shift
!= -1) {
1244 sector_t sector
, size
, first_bad
;
1245 bool is_flush
= true;
1247 if (dev
->queue_mode
== NULL_Q_BIO
&&
1248 bio_op(cmd
->bio
) != REQ_OP_FLUSH
) {
1250 sector
= cmd
->bio
->bi_iter
.bi_sector
;
1251 size
= bio_sectors(cmd
->bio
);
1253 if (dev
->queue_mode
!= NULL_Q_BIO
&&
1254 req_op(cmd
->rq
) != REQ_OP_FLUSH
) {
1256 sector
= blk_rq_pos(cmd
->rq
);
1257 size
= blk_rq_sectors(cmd
->rq
);
1259 if (!is_flush
&& badblocks_check(&nullb
->dev
->badblocks
, sector
,
1260 size
, &first_bad
, &bad_sectors
)) {
1261 cmd
->error
= BLK_STS_IOERR
;
1266 if (dev
->memory_backed
) {
1267 if (dev
->queue_mode
== NULL_Q_BIO
) {
1268 if (bio_op(cmd
->bio
) == REQ_OP_FLUSH
)
1269 err
= null_handle_flush(nullb
);
1271 err
= null_handle_bio(cmd
);
1273 if (req_op(cmd
->rq
) == REQ_OP_FLUSH
)
1274 err
= null_handle_flush(nullb
);
1276 err
= null_handle_rq(cmd
);
1279 cmd
->error
= errno_to_blk_status(err
);
1281 /* Complete IO by inline, softirq or timer */
1282 switch (dev
->irqmode
) {
1283 case NULL_IRQ_SOFTIRQ
:
1284 switch (dev
->queue_mode
) {
1286 blk_mq_complete_request(cmd
->rq
);
1289 blk_complete_request(cmd
->rq
);
1293 * XXX: no proper submitting cpu information available.
1302 case NULL_IRQ_TIMER
:
1303 null_cmd_end_timer(cmd
);
1309 static enum hrtimer_restart
nullb_bwtimer_fn(struct hrtimer
*timer
)
1311 struct nullb
*nullb
= container_of(timer
, struct nullb
, bw_timer
);
1312 ktime_t timer_interval
= ktime_set(0, TIMER_INTERVAL
);
1313 unsigned int mbps
= nullb
->dev
->mbps
;
1315 if (atomic_long_read(&nullb
->cur_bytes
) == mb_per_tick(mbps
))
1316 return HRTIMER_NORESTART
;
1318 atomic_long_set(&nullb
->cur_bytes
, mb_per_tick(mbps
));
1319 null_restart_queue_async(nullb
);
1321 hrtimer_forward_now(&nullb
->bw_timer
, timer_interval
);
1323 return HRTIMER_RESTART
;
1326 static void nullb_setup_bwtimer(struct nullb
*nullb
)
1328 ktime_t timer_interval
= ktime_set(0, TIMER_INTERVAL
);
1330 hrtimer_init(&nullb
->bw_timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL
);
1331 nullb
->bw_timer
.function
= nullb_bwtimer_fn
;
1332 atomic_long_set(&nullb
->cur_bytes
, mb_per_tick(nullb
->dev
->mbps
));
1333 hrtimer_start(&nullb
->bw_timer
, timer_interval
, HRTIMER_MODE_REL
);
1336 static struct nullb_queue
*nullb_to_queue(struct nullb
*nullb
)
1340 if (nullb
->nr_queues
!= 1)
1341 index
= raw_smp_processor_id() / ((nr_cpu_ids
+ nullb
->nr_queues
- 1) / nullb
->nr_queues
);
1343 return &nullb
->queues
[index
];
1346 static blk_qc_t
null_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1348 struct nullb
*nullb
= q
->queuedata
;
1349 struct nullb_queue
*nq
= nullb_to_queue(nullb
);
1350 struct nullb_cmd
*cmd
;
1352 cmd
= alloc_cmd(nq
, 1);
1355 null_handle_cmd(cmd
);
1356 return BLK_QC_T_NONE
;
1359 static enum blk_eh_timer_return
null_rq_timed_out_fn(struct request
*rq
)
1361 pr_info("null: rq %p timed out\n", rq
);
1362 return BLK_EH_HANDLED
;
1365 static int null_rq_prep_fn(struct request_queue
*q
, struct request
*req
)
1367 struct nullb
*nullb
= q
->queuedata
;
1368 struct nullb_queue
*nq
= nullb_to_queue(nullb
);
1369 struct nullb_cmd
*cmd
;
1371 cmd
= alloc_cmd(nq
, 0);
1379 return BLKPREP_DEFER
;
1382 static bool should_timeout_request(struct request
*rq
)
1384 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1385 if (g_timeout_str
[0])
1386 return should_fail(&null_timeout_attr
, 1);
1391 static bool should_requeue_request(struct request
*rq
)
1393 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1394 if (g_requeue_str
[0])
1395 return should_fail(&null_requeue_attr
, 1);
1400 static void null_request_fn(struct request_queue
*q
)
1404 while ((rq
= blk_fetch_request(q
)) != NULL
) {
1405 struct nullb_cmd
*cmd
= rq
->special
;
1407 /* just ignore the request */
1408 if (should_timeout_request(rq
))
1410 if (should_requeue_request(rq
)) {
1411 blk_requeue_request(q
, rq
);
1415 spin_unlock_irq(q
->queue_lock
);
1416 null_handle_cmd(cmd
);
1417 spin_lock_irq(q
->queue_lock
);
1421 static enum blk_eh_timer_return
null_timeout_rq(struct request
*rq
, bool res
)
1423 pr_info("null: rq %p timed out\n", rq
);
1424 return BLK_EH_HANDLED
;
1427 static blk_status_t
null_queue_rq(struct blk_mq_hw_ctx
*hctx
,
1428 const struct blk_mq_queue_data
*bd
)
1430 struct nullb_cmd
*cmd
= blk_mq_rq_to_pdu(bd
->rq
);
1431 struct nullb_queue
*nq
= hctx
->driver_data
;
1433 might_sleep_if(hctx
->flags
& BLK_MQ_F_BLOCKING
);
1435 if (nq
->dev
->irqmode
== NULL_IRQ_TIMER
) {
1436 hrtimer_init(&cmd
->timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL
);
1437 cmd
->timer
.function
= null_cmd_timer_expired
;
1442 blk_mq_start_request(bd
->rq
);
1444 if (should_requeue_request(bd
->rq
)) {
1446 * Alternate between hitting the core BUSY path, and the
1447 * driver driven requeue path
1449 nq
->requeue_selection
++;
1450 if (nq
->requeue_selection
& 1)
1451 return BLK_STS_RESOURCE
;
1453 blk_mq_requeue_request(bd
->rq
, true);
1457 if (should_timeout_request(bd
->rq
))
1460 return null_handle_cmd(cmd
);
1463 static const struct blk_mq_ops null_mq_ops
= {
1464 .queue_rq
= null_queue_rq
,
1465 .complete
= null_softirq_done_fn
,
1466 .timeout
= null_timeout_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_del_dev(struct nullb
*nullb
)
1487 struct nullb_device
*dev
= nullb
->dev
;
1489 ida_simple_remove(&nullb_indexes
, nullb
->index
);
1491 list_del_init(&nullb
->list
);
1493 del_gendisk(nullb
->disk
);
1495 if (test_bit(NULLB_DEV_FL_THROTTLED
, &nullb
->dev
->flags
)) {
1496 hrtimer_cancel(&nullb
->bw_timer
);
1497 atomic_long_set(&nullb
->cur_bytes
, LONG_MAX
);
1498 null_restart_queue_async(nullb
);
1501 blk_cleanup_queue(nullb
->q
);
1502 if (dev
->queue_mode
== NULL_Q_MQ
&&
1503 nullb
->tag_set
== &nullb
->__tag_set
)
1504 blk_mq_free_tag_set(nullb
->tag_set
);
1505 put_disk(nullb
->disk
);
1506 cleanup_queues(nullb
);
1507 if (null_cache_active(nullb
))
1508 null_free_device_storage(nullb
->dev
, true);
1513 static void null_config_discard(struct nullb
*nullb
)
1515 if (nullb
->dev
->discard
== false)
1517 nullb
->q
->limits
.discard_granularity
= nullb
->dev
->blocksize
;
1518 nullb
->q
->limits
.discard_alignment
= nullb
->dev
->blocksize
;
1519 blk_queue_max_discard_sectors(nullb
->q
, UINT_MAX
>> 9);
1520 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, nullb
->q
);
1523 static int null_open(struct block_device
*bdev
, fmode_t mode
)
1528 static void null_release(struct gendisk
*disk
, fmode_t mode
)
1532 static const struct block_device_operations null_fops
= {
1533 .owner
= THIS_MODULE
,
1535 .release
= null_release
,
1538 static void null_init_queue(struct nullb
*nullb
, struct nullb_queue
*nq
)
1543 init_waitqueue_head(&nq
->wait
);
1544 nq
->queue_depth
= nullb
->queue_depth
;
1545 nq
->dev
= nullb
->dev
;
1548 static void null_init_queues(struct nullb
*nullb
)
1550 struct request_queue
*q
= nullb
->q
;
1551 struct blk_mq_hw_ctx
*hctx
;
1552 struct nullb_queue
*nq
;
1555 queue_for_each_hw_ctx(q
, hctx
, i
) {
1556 if (!hctx
->nr_ctx
|| !hctx
->tags
)
1558 nq
= &nullb
->queues
[i
];
1559 hctx
->driver_data
= nq
;
1560 null_init_queue(nullb
, nq
);
1565 static int setup_commands(struct nullb_queue
*nq
)
1567 struct nullb_cmd
*cmd
;
1570 nq
->cmds
= kzalloc(nq
->queue_depth
* sizeof(*cmd
), GFP_KERNEL
);
1574 tag_size
= ALIGN(nq
->queue_depth
, BITS_PER_LONG
) / BITS_PER_LONG
;
1575 nq
->tag_map
= kzalloc(tag_size
* sizeof(unsigned long), GFP_KERNEL
);
1581 for (i
= 0; i
< nq
->queue_depth
; i
++) {
1583 INIT_LIST_HEAD(&cmd
->list
);
1584 cmd
->ll_list
.next
= NULL
;
1591 static int setup_queues(struct nullb
*nullb
)
1593 nullb
->queues
= kzalloc(nullb
->dev
->submit_queues
*
1594 sizeof(struct nullb_queue
), GFP_KERNEL
);
1598 nullb
->nr_queues
= 0;
1599 nullb
->queue_depth
= nullb
->dev
->hw_queue_depth
;
1604 static int init_driver_queues(struct nullb
*nullb
)
1606 struct nullb_queue
*nq
;
1609 for (i
= 0; i
< nullb
->dev
->submit_queues
; i
++) {
1610 nq
= &nullb
->queues
[i
];
1612 null_init_queue(nullb
, nq
);
1614 ret
= setup_commands(nq
);
1622 static int null_gendisk_register(struct nullb
*nullb
)
1624 struct gendisk
*disk
;
1627 disk
= nullb
->disk
= alloc_disk_node(1, nullb
->dev
->home_node
);
1630 size
= (sector_t
)nullb
->dev
->size
* 1024 * 1024ULL;
1631 set_capacity(disk
, size
>> 9);
1633 disk
->flags
|= GENHD_FL_EXT_DEVT
| GENHD_FL_SUPPRESS_PARTITION_INFO
;
1634 disk
->major
= null_major
;
1635 disk
->first_minor
= nullb
->index
;
1636 disk
->fops
= &null_fops
;
1637 disk
->private_data
= nullb
;
1638 disk
->queue
= nullb
->q
;
1639 strncpy(disk
->disk_name
, nullb
->disk_name
, DISK_NAME_LEN
);
1645 static int null_init_tag_set(struct nullb
*nullb
, struct blk_mq_tag_set
*set
)
1647 set
->ops
= &null_mq_ops
;
1648 set
->nr_hw_queues
= nullb
? nullb
->dev
->submit_queues
:
1650 set
->queue_depth
= nullb
? nullb
->dev
->hw_queue_depth
:
1652 set
->numa_node
= nullb
? nullb
->dev
->home_node
: g_home_node
;
1653 set
->cmd_size
= sizeof(struct nullb_cmd
);
1654 set
->flags
= BLK_MQ_F_SHOULD_MERGE
;
1656 set
->flags
|= BLK_MQ_F_NO_SCHED
;
1657 set
->driver_data
= NULL
;
1659 if ((nullb
&& nullb
->dev
->blocking
) || g_blocking
)
1660 set
->flags
|= BLK_MQ_F_BLOCKING
;
1662 return blk_mq_alloc_tag_set(set
);
1665 static void null_validate_conf(struct nullb_device
*dev
)
1667 dev
->blocksize
= round_down(dev
->blocksize
, 512);
1668 dev
->blocksize
= clamp_t(unsigned int, dev
->blocksize
, 512, 4096);
1670 if (dev
->queue_mode
== NULL_Q_MQ
&& dev
->use_per_node_hctx
) {
1671 if (dev
->submit_queues
!= nr_online_nodes
)
1672 dev
->submit_queues
= nr_online_nodes
;
1673 } else if (dev
->submit_queues
> nr_cpu_ids
)
1674 dev
->submit_queues
= nr_cpu_ids
;
1675 else if (dev
->submit_queues
== 0)
1676 dev
->submit_queues
= 1;
1678 dev
->queue_mode
= min_t(unsigned int, dev
->queue_mode
, NULL_Q_MQ
);
1679 dev
->irqmode
= min_t(unsigned int, dev
->irqmode
, NULL_IRQ_TIMER
);
1681 /* Do memory allocation, so set blocking */
1682 if (dev
->memory_backed
)
1683 dev
->blocking
= true;
1684 else /* cache is meaningless */
1685 dev
->cache_size
= 0;
1686 dev
->cache_size
= min_t(unsigned long, ULONG_MAX
/ 1024 / 1024,
1688 dev
->mbps
= min_t(unsigned int, 1024 * 40, dev
->mbps
);
1689 /* can not stop a queue */
1690 if (dev
->queue_mode
== NULL_Q_BIO
)
1694 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1695 static bool __null_setup_fault(struct fault_attr
*attr
, char *str
)
1700 if (!setup_fault_attr(attr
, str
))
1708 static bool null_setup_fault(void)
1710 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1711 if (!__null_setup_fault(&null_timeout_attr
, g_timeout_str
))
1713 if (!__null_setup_fault(&null_requeue_attr
, g_requeue_str
))
1719 static int null_add_dev(struct nullb_device
*dev
)
1721 struct nullb
*nullb
;
1724 null_validate_conf(dev
);
1726 nullb
= kzalloc_node(sizeof(*nullb
), GFP_KERNEL
, dev
->home_node
);
1734 spin_lock_init(&nullb
->lock
);
1736 rv
= setup_queues(nullb
);
1738 goto out_free_nullb
;
1740 if (dev
->queue_mode
== NULL_Q_MQ
) {
1742 nullb
->tag_set
= &tag_set
;
1745 nullb
->tag_set
= &nullb
->__tag_set
;
1746 rv
= null_init_tag_set(nullb
, nullb
->tag_set
);
1750 goto out_cleanup_queues
;
1752 if (!null_setup_fault())
1753 goto out_cleanup_queues
;
1755 nullb
->tag_set
->timeout
= 5 * HZ
;
1756 nullb
->q
= blk_mq_init_queue(nullb
->tag_set
);
1757 if (IS_ERR(nullb
->q
)) {
1759 goto out_cleanup_tags
;
1761 null_init_queues(nullb
);
1762 } else if (dev
->queue_mode
== NULL_Q_BIO
) {
1763 nullb
->q
= blk_alloc_queue_node(GFP_KERNEL
, dev
->home_node
,
1767 goto out_cleanup_queues
;
1769 blk_queue_make_request(nullb
->q
, null_queue_bio
);
1770 rv
= init_driver_queues(nullb
);
1772 goto out_cleanup_blk_queue
;
1774 nullb
->q
= blk_init_queue_node(null_request_fn
, &nullb
->lock
,
1778 goto out_cleanup_queues
;
1781 if (!null_setup_fault())
1782 goto out_cleanup_blk_queue
;
1784 blk_queue_prep_rq(nullb
->q
, null_rq_prep_fn
);
1785 blk_queue_softirq_done(nullb
->q
, null_softirq_done_fn
);
1786 blk_queue_rq_timed_out(nullb
->q
, null_rq_timed_out_fn
);
1787 nullb
->q
->rq_timeout
= 5 * HZ
;
1788 rv
= init_driver_queues(nullb
);
1790 goto out_cleanup_blk_queue
;
1794 set_bit(NULLB_DEV_FL_THROTTLED
, &dev
->flags
);
1795 nullb_setup_bwtimer(nullb
);
1798 if (dev
->cache_size
> 0) {
1799 set_bit(NULLB_DEV_FL_CACHE
, &nullb
->dev
->flags
);
1800 blk_queue_write_cache(nullb
->q
, true, true);
1801 blk_queue_flush_queueable(nullb
->q
, true);
1804 nullb
->q
->queuedata
= nullb
;
1805 queue_flag_set_unlocked(QUEUE_FLAG_NONROT
, nullb
->q
);
1806 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM
, nullb
->q
);
1809 nullb
->index
= ida_simple_get(&nullb_indexes
, 0, 0, GFP_KERNEL
);
1810 dev
->index
= nullb
->index
;
1811 mutex_unlock(&lock
);
1813 blk_queue_logical_block_size(nullb
->q
, dev
->blocksize
);
1814 blk_queue_physical_block_size(nullb
->q
, dev
->blocksize
);
1816 null_config_discard(nullb
);
1818 sprintf(nullb
->disk_name
, "nullb%d", nullb
->index
);
1820 rv
= null_gendisk_register(nullb
);
1822 goto out_cleanup_blk_queue
;
1825 list_add_tail(&nullb
->list
, &nullb_list
);
1826 mutex_unlock(&lock
);
1829 out_cleanup_blk_queue
:
1830 blk_cleanup_queue(nullb
->q
);
1832 if (dev
->queue_mode
== NULL_Q_MQ
&& nullb
->tag_set
== &nullb
->__tag_set
)
1833 blk_mq_free_tag_set(nullb
->tag_set
);
1835 cleanup_queues(nullb
);
1842 static int __init
null_init(void)
1846 struct nullb
*nullb
;
1847 struct nullb_device
*dev
;
1849 /* check for nullb_page.bitmap */
1850 if (sizeof(unsigned long) * 8 - 2 < (PAGE_SIZE
>> SECTOR_SHIFT
))
1853 if (g_bs
> PAGE_SIZE
) {
1854 pr_warn("null_blk: invalid block size\n");
1855 pr_warn("null_blk: defaults block size to %lu\n", PAGE_SIZE
);
1859 if (g_queue_mode
== NULL_Q_MQ
&& g_use_per_node_hctx
) {
1860 if (g_submit_queues
!= nr_online_nodes
) {
1861 pr_warn("null_blk: submit_queues param is set to %u.\n",
1863 g_submit_queues
= nr_online_nodes
;
1865 } else if (g_submit_queues
> nr_cpu_ids
)
1866 g_submit_queues
= nr_cpu_ids
;
1867 else if (g_submit_queues
<= 0)
1868 g_submit_queues
= 1;
1870 if (g_queue_mode
== NULL_Q_MQ
&& shared_tags
) {
1871 ret
= null_init_tag_set(NULL
, &tag_set
);
1876 config_group_init(&nullb_subsys
.su_group
);
1877 mutex_init(&nullb_subsys
.su_mutex
);
1879 ret
= configfs_register_subsystem(&nullb_subsys
);
1885 null_major
= register_blkdev(0, "nullb");
1886 if (null_major
< 0) {
1891 for (i
= 0; i
< nr_devices
; i
++) {
1892 dev
= null_alloc_dev();
1897 ret
= null_add_dev(dev
);
1904 pr_info("null: module loaded\n");
1908 while (!list_empty(&nullb_list
)) {
1909 nullb
= list_entry(nullb_list
.next
, struct nullb
, list
);
1911 null_del_dev(nullb
);
1914 unregister_blkdev(null_major
, "nullb");
1916 configfs_unregister_subsystem(&nullb_subsys
);
1918 if (g_queue_mode
== NULL_Q_MQ
&& shared_tags
)
1919 blk_mq_free_tag_set(&tag_set
);
1923 static void __exit
null_exit(void)
1925 struct nullb
*nullb
;
1927 configfs_unregister_subsystem(&nullb_subsys
);
1929 unregister_blkdev(null_major
, "nullb");
1932 while (!list_empty(&nullb_list
)) {
1933 struct nullb_device
*dev
;
1935 nullb
= list_entry(nullb_list
.next
, struct nullb
, list
);
1937 null_del_dev(nullb
);
1940 mutex_unlock(&lock
);
1942 if (g_queue_mode
== NULL_Q_MQ
&& shared_tags
)
1943 blk_mq_free_tag_set(&tag_set
);
1946 module_init(null_init
);
1947 module_exit(null_exit
);
1949 MODULE_AUTHOR("Jens Axboe <axboe@kernel.dk>");
1950 MODULE_LICENSE("GPL");