WIP FPC-III support
[linux/fpc-iii.git] / drivers / block / null_blk / main.c
blob5357c3a4a36fc39c6e0986cb3af8b5516c0785f3
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Add configfs and memory store: Kyungchan Koh <kkc6196@fb.com> and
4 * Shaohua Li <shli@fb.com>
5 */
6 #include <linux/module.h>
8 #include <linux/moduleparam.h>
9 #include <linux/sched.h>
10 #include <linux/fs.h>
11 #include <linux/init.h>
12 #include "null_blk.h"
14 #define PAGE_SECTORS_SHIFT (PAGE_SHIFT - SECTOR_SHIFT)
15 #define PAGE_SECTORS (1 << PAGE_SECTORS_SHIFT)
16 #define SECTOR_MASK (PAGE_SECTORS - 1)
18 #define FREE_BATCH 16
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);
27 #endif
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,
44 NULLB_DEV_FL_UP = 1,
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
62 struct nullb_page {
63 struct page *page;
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;
75 enum {
76 NULL_IRQ_NONE = 0,
77 NULL_IRQ_SOFTIRQ = 1,
78 NULL_IRQ_TIMER = 2,
81 enum {
82 NULL_Q_BIO = 0,
83 NULL_Q_RQ = 1,
84 NULL_Q_MQ = 2,
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>");
115 #endif
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)
121 int ret, new_val;
123 ret = kstrtoint(str, 10, &new_val);
124 if (ret)
125 return -EINVAL;
127 if (new_val < min || new_val > max)
128 return -EINVAL;
130 *val = new_val;
131 return 0;
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 int g_max_sectors;
156 module_param_named(max_sectors, g_max_sectors, int, 0444);
157 MODULE_PARM_DESC(max_sectors, "Maximum size of a command (in 512B sectors)");
159 static unsigned int nr_devices = 1;
160 module_param(nr_devices, uint, 0444);
161 MODULE_PARM_DESC(nr_devices, "Number of devices to register");
163 static bool g_blocking;
164 module_param_named(blocking, g_blocking, bool, 0444);
165 MODULE_PARM_DESC(blocking, "Register as a blocking blk-mq driver device");
167 static bool shared_tags;
168 module_param(shared_tags, bool, 0444);
169 MODULE_PARM_DESC(shared_tags, "Share tag set between devices for blk-mq");
171 static bool g_shared_tag_bitmap;
172 module_param_named(shared_tag_bitmap, g_shared_tag_bitmap, bool, 0444);
173 MODULE_PARM_DESC(shared_tag_bitmap, "Use shared tag bitmap for all submission queues for blk-mq");
175 static int g_irqmode = NULL_IRQ_SOFTIRQ;
177 static int null_set_irqmode(const char *str, const struct kernel_param *kp)
179 return null_param_store_val(str, &g_irqmode, NULL_IRQ_NONE,
180 NULL_IRQ_TIMER);
183 static const struct kernel_param_ops null_irqmode_param_ops = {
184 .set = null_set_irqmode,
185 .get = param_get_int,
188 device_param_cb(irqmode, &null_irqmode_param_ops, &g_irqmode, 0444);
189 MODULE_PARM_DESC(irqmode, "IRQ completion handler. 0-none, 1-softirq, 2-timer");
191 static unsigned long g_completion_nsec = 10000;
192 module_param_named(completion_nsec, g_completion_nsec, ulong, 0444);
193 MODULE_PARM_DESC(completion_nsec, "Time in ns to complete a request in hardware. Default: 10,000ns");
195 static int g_hw_queue_depth = 64;
196 module_param_named(hw_queue_depth, g_hw_queue_depth, int, 0444);
197 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 64");
199 static bool g_use_per_node_hctx;
200 module_param_named(use_per_node_hctx, g_use_per_node_hctx, bool, 0444);
201 MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware context queues. Default: false");
203 static bool g_zoned;
204 module_param_named(zoned, g_zoned, bool, S_IRUGO);
205 MODULE_PARM_DESC(zoned, "Make device as a host-managed zoned block device. Default: false");
207 static unsigned long g_zone_size = 256;
208 module_param_named(zone_size, g_zone_size, ulong, S_IRUGO);
209 MODULE_PARM_DESC(zone_size, "Zone size in MB when block device is zoned. Must be power-of-two: Default: 256");
211 static unsigned long g_zone_capacity;
212 module_param_named(zone_capacity, g_zone_capacity, ulong, 0444);
213 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");
215 static unsigned int g_zone_nr_conv;
216 module_param_named(zone_nr_conv, g_zone_nr_conv, uint, 0444);
217 MODULE_PARM_DESC(zone_nr_conv, "Number of conventional zones when block device is zoned. Default: 0");
219 static unsigned int g_zone_max_open;
220 module_param_named(zone_max_open, g_zone_max_open, uint, 0444);
221 MODULE_PARM_DESC(zone_max_open, "Maximum number of open zones when block device is zoned. Default: 0 (no limit)");
223 static unsigned int g_zone_max_active;
224 module_param_named(zone_max_active, g_zone_max_active, uint, 0444);
225 MODULE_PARM_DESC(zone_max_active, "Maximum number of active zones when block device is zoned. Default: 0 (no limit)");
227 static struct nullb_device *null_alloc_dev(void);
228 static void null_free_dev(struct nullb_device *dev);
229 static void null_del_dev(struct nullb *nullb);
230 static int null_add_dev(struct nullb_device *dev);
231 static void null_free_device_storage(struct nullb_device *dev, bool is_cache);
233 static inline struct nullb_device *to_nullb_device(struct config_item *item)
235 return item ? container_of(item, struct nullb_device, item) : NULL;
238 static inline ssize_t nullb_device_uint_attr_show(unsigned int val, char *page)
240 return snprintf(page, PAGE_SIZE, "%u\n", val);
243 static inline ssize_t nullb_device_ulong_attr_show(unsigned long val,
244 char *page)
246 return snprintf(page, PAGE_SIZE, "%lu\n", val);
249 static inline ssize_t nullb_device_bool_attr_show(bool val, char *page)
251 return snprintf(page, PAGE_SIZE, "%u\n", val);
254 static ssize_t nullb_device_uint_attr_store(unsigned int *val,
255 const char *page, size_t count)
257 unsigned int tmp;
258 int result;
260 result = kstrtouint(page, 0, &tmp);
261 if (result < 0)
262 return result;
264 *val = tmp;
265 return count;
268 static ssize_t nullb_device_ulong_attr_store(unsigned long *val,
269 const char *page, size_t count)
271 int result;
272 unsigned long tmp;
274 result = kstrtoul(page, 0, &tmp);
275 if (result < 0)
276 return result;
278 *val = tmp;
279 return count;
282 static ssize_t nullb_device_bool_attr_store(bool *val, const char *page,
283 size_t count)
285 bool tmp;
286 int result;
288 result = kstrtobool(page, &tmp);
289 if (result < 0)
290 return result;
292 *val = tmp;
293 return count;
296 /* The following macro should only be used with TYPE = {uint, ulong, bool}. */
297 #define NULLB_DEVICE_ATTR(NAME, TYPE, APPLY) \
298 static ssize_t \
299 nullb_device_##NAME##_show(struct config_item *item, char *page) \
301 return nullb_device_##TYPE##_attr_show( \
302 to_nullb_device(item)->NAME, page); \
304 static ssize_t \
305 nullb_device_##NAME##_store(struct config_item *item, const char *page, \
306 size_t count) \
308 int (*apply_fn)(struct nullb_device *dev, TYPE new_value) = APPLY;\
309 struct nullb_device *dev = to_nullb_device(item); \
310 TYPE new_value = 0; \
311 int ret; \
313 ret = nullb_device_##TYPE##_attr_store(&new_value, page, count);\
314 if (ret < 0) \
315 return ret; \
316 if (apply_fn) \
317 ret = apply_fn(dev, new_value); \
318 else if (test_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags)) \
319 ret = -EBUSY; \
320 if (ret < 0) \
321 return ret; \
322 dev->NAME = new_value; \
323 return count; \
325 CONFIGFS_ATTR(nullb_device_, NAME);
327 static int nullb_apply_submit_queues(struct nullb_device *dev,
328 unsigned int submit_queues)
330 struct nullb *nullb = dev->nullb;
331 struct blk_mq_tag_set *set;
333 if (!nullb)
334 return 0;
337 * Make sure that null_init_hctx() does not access nullb->queues[] past
338 * the end of that array.
340 if (submit_queues > nr_cpu_ids)
341 return -EINVAL;
342 set = nullb->tag_set;
343 blk_mq_update_nr_hw_queues(set, submit_queues);
344 return set->nr_hw_queues == submit_queues ? 0 : -ENOMEM;
347 NULLB_DEVICE_ATTR(size, ulong, NULL);
348 NULLB_DEVICE_ATTR(completion_nsec, ulong, NULL);
349 NULLB_DEVICE_ATTR(submit_queues, uint, nullb_apply_submit_queues);
350 NULLB_DEVICE_ATTR(home_node, uint, NULL);
351 NULLB_DEVICE_ATTR(queue_mode, uint, NULL);
352 NULLB_DEVICE_ATTR(blocksize, uint, NULL);
353 NULLB_DEVICE_ATTR(max_sectors, uint, NULL);
354 NULLB_DEVICE_ATTR(irqmode, uint, NULL);
355 NULLB_DEVICE_ATTR(hw_queue_depth, uint, NULL);
356 NULLB_DEVICE_ATTR(index, uint, NULL);
357 NULLB_DEVICE_ATTR(blocking, bool, NULL);
358 NULLB_DEVICE_ATTR(use_per_node_hctx, bool, NULL);
359 NULLB_DEVICE_ATTR(memory_backed, bool, NULL);
360 NULLB_DEVICE_ATTR(discard, bool, NULL);
361 NULLB_DEVICE_ATTR(mbps, uint, NULL);
362 NULLB_DEVICE_ATTR(cache_size, ulong, NULL);
363 NULLB_DEVICE_ATTR(zoned, bool, NULL);
364 NULLB_DEVICE_ATTR(zone_size, ulong, NULL);
365 NULLB_DEVICE_ATTR(zone_capacity, ulong, NULL);
366 NULLB_DEVICE_ATTR(zone_nr_conv, uint, NULL);
367 NULLB_DEVICE_ATTR(zone_max_open, uint, NULL);
368 NULLB_DEVICE_ATTR(zone_max_active, uint, NULL);
370 static ssize_t nullb_device_power_show(struct config_item *item, char *page)
372 return nullb_device_bool_attr_show(to_nullb_device(item)->power, page);
375 static ssize_t nullb_device_power_store(struct config_item *item,
376 const char *page, size_t count)
378 struct nullb_device *dev = to_nullb_device(item);
379 bool newp = false;
380 ssize_t ret;
382 ret = nullb_device_bool_attr_store(&newp, page, count);
383 if (ret < 0)
384 return ret;
386 if (!dev->power && newp) {
387 if (test_and_set_bit(NULLB_DEV_FL_UP, &dev->flags))
388 return count;
389 if (null_add_dev(dev)) {
390 clear_bit(NULLB_DEV_FL_UP, &dev->flags);
391 return -ENOMEM;
394 set_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
395 dev->power = newp;
396 } else if (dev->power && !newp) {
397 if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
398 mutex_lock(&lock);
399 dev->power = newp;
400 null_del_dev(dev->nullb);
401 mutex_unlock(&lock);
403 clear_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
406 return count;
409 CONFIGFS_ATTR(nullb_device_, power);
411 static ssize_t nullb_device_badblocks_show(struct config_item *item, char *page)
413 struct nullb_device *t_dev = to_nullb_device(item);
415 return badblocks_show(&t_dev->badblocks, page, 0);
418 static ssize_t nullb_device_badblocks_store(struct config_item *item,
419 const char *page, size_t count)
421 struct nullb_device *t_dev = to_nullb_device(item);
422 char *orig, *buf, *tmp;
423 u64 start, end;
424 int ret;
426 orig = kstrndup(page, count, GFP_KERNEL);
427 if (!orig)
428 return -ENOMEM;
430 buf = strstrip(orig);
432 ret = -EINVAL;
433 if (buf[0] != '+' && buf[0] != '-')
434 goto out;
435 tmp = strchr(&buf[1], '-');
436 if (!tmp)
437 goto out;
438 *tmp = '\0';
439 ret = kstrtoull(buf + 1, 0, &start);
440 if (ret)
441 goto out;
442 ret = kstrtoull(tmp + 1, 0, &end);
443 if (ret)
444 goto out;
445 ret = -EINVAL;
446 if (start > end)
447 goto out;
448 /* enable badblocks */
449 cmpxchg(&t_dev->badblocks.shift, -1, 0);
450 if (buf[0] == '+')
451 ret = badblocks_set(&t_dev->badblocks, start,
452 end - start + 1, 1);
453 else
454 ret = badblocks_clear(&t_dev->badblocks, start,
455 end - start + 1);
456 if (ret == 0)
457 ret = count;
458 out:
459 kfree(orig);
460 return ret;
462 CONFIGFS_ATTR(nullb_device_, badblocks);
464 static struct configfs_attribute *nullb_device_attrs[] = {
465 &nullb_device_attr_size,
466 &nullb_device_attr_completion_nsec,
467 &nullb_device_attr_submit_queues,
468 &nullb_device_attr_home_node,
469 &nullb_device_attr_queue_mode,
470 &nullb_device_attr_blocksize,
471 &nullb_device_attr_max_sectors,
472 &nullb_device_attr_irqmode,
473 &nullb_device_attr_hw_queue_depth,
474 &nullb_device_attr_index,
475 &nullb_device_attr_blocking,
476 &nullb_device_attr_use_per_node_hctx,
477 &nullb_device_attr_power,
478 &nullb_device_attr_memory_backed,
479 &nullb_device_attr_discard,
480 &nullb_device_attr_mbps,
481 &nullb_device_attr_cache_size,
482 &nullb_device_attr_badblocks,
483 &nullb_device_attr_zoned,
484 &nullb_device_attr_zone_size,
485 &nullb_device_attr_zone_capacity,
486 &nullb_device_attr_zone_nr_conv,
487 &nullb_device_attr_zone_max_open,
488 &nullb_device_attr_zone_max_active,
489 NULL,
492 static void nullb_device_release(struct config_item *item)
494 struct nullb_device *dev = to_nullb_device(item);
496 null_free_device_storage(dev, false);
497 null_free_dev(dev);
500 static struct configfs_item_operations nullb_device_ops = {
501 .release = nullb_device_release,
504 static const struct config_item_type nullb_device_type = {
505 .ct_item_ops = &nullb_device_ops,
506 .ct_attrs = nullb_device_attrs,
507 .ct_owner = THIS_MODULE,
510 static struct
511 config_item *nullb_group_make_item(struct config_group *group, const char *name)
513 struct nullb_device *dev;
515 dev = null_alloc_dev();
516 if (!dev)
517 return ERR_PTR(-ENOMEM);
519 config_item_init_type_name(&dev->item, name, &nullb_device_type);
521 return &dev->item;
524 static void
525 nullb_group_drop_item(struct config_group *group, struct config_item *item)
527 struct nullb_device *dev = to_nullb_device(item);
529 if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
530 mutex_lock(&lock);
531 dev->power = false;
532 null_del_dev(dev->nullb);
533 mutex_unlock(&lock);
536 config_item_put(item);
539 static ssize_t memb_group_features_show(struct config_item *item, char *page)
541 return snprintf(page, PAGE_SIZE,
542 "memory_backed,discard,bandwidth,cache,badblocks,zoned,zone_size,zone_capacity,zone_nr_conv,zone_max_open,zone_max_active,blocksize,max_sectors\n");
545 CONFIGFS_ATTR_RO(memb_group_, features);
547 static struct configfs_attribute *nullb_group_attrs[] = {
548 &memb_group_attr_features,
549 NULL,
552 static struct configfs_group_operations nullb_group_ops = {
553 .make_item = nullb_group_make_item,
554 .drop_item = nullb_group_drop_item,
557 static const struct config_item_type nullb_group_type = {
558 .ct_group_ops = &nullb_group_ops,
559 .ct_attrs = nullb_group_attrs,
560 .ct_owner = THIS_MODULE,
563 static struct configfs_subsystem nullb_subsys = {
564 .su_group = {
565 .cg_item = {
566 .ci_namebuf = "nullb",
567 .ci_type = &nullb_group_type,
572 static inline int null_cache_active(struct nullb *nullb)
574 return test_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
577 static struct nullb_device *null_alloc_dev(void)
579 struct nullb_device *dev;
581 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
582 if (!dev)
583 return NULL;
584 INIT_RADIX_TREE(&dev->data, GFP_ATOMIC);
585 INIT_RADIX_TREE(&dev->cache, GFP_ATOMIC);
586 if (badblocks_init(&dev->badblocks, 0)) {
587 kfree(dev);
588 return NULL;
591 dev->size = g_gb * 1024;
592 dev->completion_nsec = g_completion_nsec;
593 dev->submit_queues = g_submit_queues;
594 dev->home_node = g_home_node;
595 dev->queue_mode = g_queue_mode;
596 dev->blocksize = g_bs;
597 dev->max_sectors = g_max_sectors;
598 dev->irqmode = g_irqmode;
599 dev->hw_queue_depth = g_hw_queue_depth;
600 dev->blocking = g_blocking;
601 dev->use_per_node_hctx = g_use_per_node_hctx;
602 dev->zoned = g_zoned;
603 dev->zone_size = g_zone_size;
604 dev->zone_capacity = g_zone_capacity;
605 dev->zone_nr_conv = g_zone_nr_conv;
606 dev->zone_max_open = g_zone_max_open;
607 dev->zone_max_active = g_zone_max_active;
608 return dev;
611 static void null_free_dev(struct nullb_device *dev)
613 if (!dev)
614 return;
616 null_free_zoned_dev(dev);
617 badblocks_exit(&dev->badblocks);
618 kfree(dev);
621 static void put_tag(struct nullb_queue *nq, unsigned int tag)
623 clear_bit_unlock(tag, nq->tag_map);
625 if (waitqueue_active(&nq->wait))
626 wake_up(&nq->wait);
629 static unsigned int get_tag(struct nullb_queue *nq)
631 unsigned int tag;
633 do {
634 tag = find_first_zero_bit(nq->tag_map, nq->queue_depth);
635 if (tag >= nq->queue_depth)
636 return -1U;
637 } while (test_and_set_bit_lock(tag, nq->tag_map));
639 return tag;
642 static void free_cmd(struct nullb_cmd *cmd)
644 put_tag(cmd->nq, cmd->tag);
647 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer);
649 static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq)
651 struct nullb_cmd *cmd;
652 unsigned int tag;
654 tag = get_tag(nq);
655 if (tag != -1U) {
656 cmd = &nq->cmds[tag];
657 cmd->tag = tag;
658 cmd->error = BLK_STS_OK;
659 cmd->nq = nq;
660 if (nq->dev->irqmode == NULL_IRQ_TIMER) {
661 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC,
662 HRTIMER_MODE_REL);
663 cmd->timer.function = null_cmd_timer_expired;
665 return cmd;
668 return NULL;
671 static struct nullb_cmd *alloc_cmd(struct nullb_queue *nq, int can_wait)
673 struct nullb_cmd *cmd;
674 DEFINE_WAIT(wait);
676 cmd = __alloc_cmd(nq);
677 if (cmd || !can_wait)
678 return cmd;
680 do {
681 prepare_to_wait(&nq->wait, &wait, TASK_UNINTERRUPTIBLE);
682 cmd = __alloc_cmd(nq);
683 if (cmd)
684 break;
686 io_schedule();
687 } while (1);
689 finish_wait(&nq->wait, &wait);
690 return cmd;
693 static void end_cmd(struct nullb_cmd *cmd)
695 int queue_mode = cmd->nq->dev->queue_mode;
697 switch (queue_mode) {
698 case NULL_Q_MQ:
699 blk_mq_end_request(cmd->rq, cmd->error);
700 return;
701 case NULL_Q_BIO:
702 cmd->bio->bi_status = cmd->error;
703 bio_endio(cmd->bio);
704 break;
707 free_cmd(cmd);
710 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
712 end_cmd(container_of(timer, struct nullb_cmd, timer));
714 return HRTIMER_NORESTART;
717 static void null_cmd_end_timer(struct nullb_cmd *cmd)
719 ktime_t kt = cmd->nq->dev->completion_nsec;
721 hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL);
724 static void null_complete_rq(struct request *rq)
726 end_cmd(blk_mq_rq_to_pdu(rq));
729 static struct nullb_page *null_alloc_page(gfp_t gfp_flags)
731 struct nullb_page *t_page;
733 t_page = kmalloc(sizeof(struct nullb_page), gfp_flags);
734 if (!t_page)
735 goto out;
737 t_page->page = alloc_pages(gfp_flags, 0);
738 if (!t_page->page)
739 goto out_freepage;
741 memset(t_page->bitmap, 0, sizeof(t_page->bitmap));
742 return t_page;
743 out_freepage:
744 kfree(t_page);
745 out:
746 return NULL;
749 static void null_free_page(struct nullb_page *t_page)
751 __set_bit(NULLB_PAGE_FREE, t_page->bitmap);
752 if (test_bit(NULLB_PAGE_LOCK, t_page->bitmap))
753 return;
754 __free_page(t_page->page);
755 kfree(t_page);
758 static bool null_page_empty(struct nullb_page *page)
760 int size = MAP_SZ - 2;
762 return find_first_bit(page->bitmap, size) == size;
765 static void null_free_sector(struct nullb *nullb, sector_t sector,
766 bool is_cache)
768 unsigned int sector_bit;
769 u64 idx;
770 struct nullb_page *t_page, *ret;
771 struct radix_tree_root *root;
773 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
774 idx = sector >> PAGE_SECTORS_SHIFT;
775 sector_bit = (sector & SECTOR_MASK);
777 t_page = radix_tree_lookup(root, idx);
778 if (t_page) {
779 __clear_bit(sector_bit, t_page->bitmap);
781 if (null_page_empty(t_page)) {
782 ret = radix_tree_delete_item(root, idx, t_page);
783 WARN_ON(ret != t_page);
784 null_free_page(ret);
785 if (is_cache)
786 nullb->dev->curr_cache -= PAGE_SIZE;
791 static struct nullb_page *null_radix_tree_insert(struct nullb *nullb, u64 idx,
792 struct nullb_page *t_page, bool is_cache)
794 struct radix_tree_root *root;
796 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
798 if (radix_tree_insert(root, idx, t_page)) {
799 null_free_page(t_page);
800 t_page = radix_tree_lookup(root, idx);
801 WARN_ON(!t_page || t_page->page->index != idx);
802 } else if (is_cache)
803 nullb->dev->curr_cache += PAGE_SIZE;
805 return t_page;
808 static void null_free_device_storage(struct nullb_device *dev, bool is_cache)
810 unsigned long pos = 0;
811 int nr_pages;
812 struct nullb_page *ret, *t_pages[FREE_BATCH];
813 struct radix_tree_root *root;
815 root = is_cache ? &dev->cache : &dev->data;
817 do {
818 int i;
820 nr_pages = radix_tree_gang_lookup(root,
821 (void **)t_pages, pos, FREE_BATCH);
823 for (i = 0; i < nr_pages; i++) {
824 pos = t_pages[i]->page->index;
825 ret = radix_tree_delete_item(root, pos, t_pages[i]);
826 WARN_ON(ret != t_pages[i]);
827 null_free_page(ret);
830 pos++;
831 } while (nr_pages == FREE_BATCH);
833 if (is_cache)
834 dev->curr_cache = 0;
837 static struct nullb_page *__null_lookup_page(struct nullb *nullb,
838 sector_t sector, bool for_write, bool is_cache)
840 unsigned int sector_bit;
841 u64 idx;
842 struct nullb_page *t_page;
843 struct radix_tree_root *root;
845 idx = sector >> PAGE_SECTORS_SHIFT;
846 sector_bit = (sector & SECTOR_MASK);
848 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
849 t_page = radix_tree_lookup(root, idx);
850 WARN_ON(t_page && t_page->page->index != idx);
852 if (t_page && (for_write || test_bit(sector_bit, t_page->bitmap)))
853 return t_page;
855 return NULL;
858 static struct nullb_page *null_lookup_page(struct nullb *nullb,
859 sector_t sector, bool for_write, bool ignore_cache)
861 struct nullb_page *page = NULL;
863 if (!ignore_cache)
864 page = __null_lookup_page(nullb, sector, for_write, true);
865 if (page)
866 return page;
867 return __null_lookup_page(nullb, sector, for_write, false);
870 static struct nullb_page *null_insert_page(struct nullb *nullb,
871 sector_t sector, bool ignore_cache)
872 __releases(&nullb->lock)
873 __acquires(&nullb->lock)
875 u64 idx;
876 struct nullb_page *t_page;
878 t_page = null_lookup_page(nullb, sector, true, ignore_cache);
879 if (t_page)
880 return t_page;
882 spin_unlock_irq(&nullb->lock);
884 t_page = null_alloc_page(GFP_NOIO);
885 if (!t_page)
886 goto out_lock;
888 if (radix_tree_preload(GFP_NOIO))
889 goto out_freepage;
891 spin_lock_irq(&nullb->lock);
892 idx = sector >> PAGE_SECTORS_SHIFT;
893 t_page->page->index = idx;
894 t_page = null_radix_tree_insert(nullb, idx, t_page, !ignore_cache);
895 radix_tree_preload_end();
897 return t_page;
898 out_freepage:
899 null_free_page(t_page);
900 out_lock:
901 spin_lock_irq(&nullb->lock);
902 return null_lookup_page(nullb, sector, true, ignore_cache);
905 static int null_flush_cache_page(struct nullb *nullb, struct nullb_page *c_page)
907 int i;
908 unsigned int offset;
909 u64 idx;
910 struct nullb_page *t_page, *ret;
911 void *dst, *src;
913 idx = c_page->page->index;
915 t_page = null_insert_page(nullb, idx << PAGE_SECTORS_SHIFT, true);
917 __clear_bit(NULLB_PAGE_LOCK, c_page->bitmap);
918 if (test_bit(NULLB_PAGE_FREE, c_page->bitmap)) {
919 null_free_page(c_page);
920 if (t_page && null_page_empty(t_page)) {
921 ret = radix_tree_delete_item(&nullb->dev->data,
922 idx, t_page);
923 null_free_page(t_page);
925 return 0;
928 if (!t_page)
929 return -ENOMEM;
931 src = kmap_atomic(c_page->page);
932 dst = kmap_atomic(t_page->page);
934 for (i = 0; i < PAGE_SECTORS;
935 i += (nullb->dev->blocksize >> SECTOR_SHIFT)) {
936 if (test_bit(i, c_page->bitmap)) {
937 offset = (i << SECTOR_SHIFT);
938 memcpy(dst + offset, src + offset,
939 nullb->dev->blocksize);
940 __set_bit(i, t_page->bitmap);
944 kunmap_atomic(dst);
945 kunmap_atomic(src);
947 ret = radix_tree_delete_item(&nullb->dev->cache, idx, c_page);
948 null_free_page(ret);
949 nullb->dev->curr_cache -= PAGE_SIZE;
951 return 0;
954 static int null_make_cache_space(struct nullb *nullb, unsigned long n)
956 int i, err, nr_pages;
957 struct nullb_page *c_pages[FREE_BATCH];
958 unsigned long flushed = 0, one_round;
960 again:
961 if ((nullb->dev->cache_size * 1024 * 1024) >
962 nullb->dev->curr_cache + n || nullb->dev->curr_cache == 0)
963 return 0;
965 nr_pages = radix_tree_gang_lookup(&nullb->dev->cache,
966 (void **)c_pages, nullb->cache_flush_pos, FREE_BATCH);
968 * nullb_flush_cache_page could unlock before using the c_pages. To
969 * avoid race, we don't allow page free
971 for (i = 0; i < nr_pages; i++) {
972 nullb->cache_flush_pos = c_pages[i]->page->index;
974 * We found the page which is being flushed to disk by other
975 * threads
977 if (test_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap))
978 c_pages[i] = NULL;
979 else
980 __set_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap);
983 one_round = 0;
984 for (i = 0; i < nr_pages; i++) {
985 if (c_pages[i] == NULL)
986 continue;
987 err = null_flush_cache_page(nullb, c_pages[i]);
988 if (err)
989 return err;
990 one_round++;
992 flushed += one_round << PAGE_SHIFT;
994 if (n > flushed) {
995 if (nr_pages == 0)
996 nullb->cache_flush_pos = 0;
997 if (one_round == 0) {
998 /* give other threads a chance */
999 spin_unlock_irq(&nullb->lock);
1000 spin_lock_irq(&nullb->lock);
1002 goto again;
1004 return 0;
1007 static int copy_to_nullb(struct nullb *nullb, struct page *source,
1008 unsigned int off, sector_t sector, size_t n, bool is_fua)
1010 size_t temp, count = 0;
1011 unsigned int offset;
1012 struct nullb_page *t_page;
1013 void *dst, *src;
1015 while (count < n) {
1016 temp = min_t(size_t, nullb->dev->blocksize, n - count);
1018 if (null_cache_active(nullb) && !is_fua)
1019 null_make_cache_space(nullb, PAGE_SIZE);
1021 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
1022 t_page = null_insert_page(nullb, sector,
1023 !null_cache_active(nullb) || is_fua);
1024 if (!t_page)
1025 return -ENOSPC;
1027 src = kmap_atomic(source);
1028 dst = kmap_atomic(t_page->page);
1029 memcpy(dst + offset, src + off + count, temp);
1030 kunmap_atomic(dst);
1031 kunmap_atomic(src);
1033 __set_bit(sector & SECTOR_MASK, t_page->bitmap);
1035 if (is_fua)
1036 null_free_sector(nullb, sector, true);
1038 count += temp;
1039 sector += temp >> SECTOR_SHIFT;
1041 return 0;
1044 static int copy_from_nullb(struct nullb *nullb, struct page *dest,
1045 unsigned int off, sector_t sector, size_t n)
1047 size_t temp, count = 0;
1048 unsigned int offset;
1049 struct nullb_page *t_page;
1050 void *dst, *src;
1052 while (count < n) {
1053 temp = min_t(size_t, nullb->dev->blocksize, n - count);
1055 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
1056 t_page = null_lookup_page(nullb, sector, false,
1057 !null_cache_active(nullb));
1059 dst = kmap_atomic(dest);
1060 if (!t_page) {
1061 memset(dst + off + count, 0, temp);
1062 goto next;
1064 src = kmap_atomic(t_page->page);
1065 memcpy(dst + off + count, src + offset, temp);
1066 kunmap_atomic(src);
1067 next:
1068 kunmap_atomic(dst);
1070 count += temp;
1071 sector += temp >> SECTOR_SHIFT;
1073 return 0;
1076 static void nullb_fill_pattern(struct nullb *nullb, struct page *page,
1077 unsigned int len, unsigned int off)
1079 void *dst;
1081 dst = kmap_atomic(page);
1082 memset(dst + off, 0xFF, len);
1083 kunmap_atomic(dst);
1086 blk_status_t null_handle_discard(struct nullb_device *dev,
1087 sector_t sector, sector_t nr_sectors)
1089 struct nullb *nullb = dev->nullb;
1090 size_t n = nr_sectors << SECTOR_SHIFT;
1091 size_t temp;
1093 spin_lock_irq(&nullb->lock);
1094 while (n > 0) {
1095 temp = min_t(size_t, n, dev->blocksize);
1096 null_free_sector(nullb, sector, false);
1097 if (null_cache_active(nullb))
1098 null_free_sector(nullb, sector, true);
1099 sector += temp >> SECTOR_SHIFT;
1100 n -= temp;
1102 spin_unlock_irq(&nullb->lock);
1104 return BLK_STS_OK;
1107 static int null_handle_flush(struct nullb *nullb)
1109 int err;
1111 if (!null_cache_active(nullb))
1112 return 0;
1114 spin_lock_irq(&nullb->lock);
1115 while (true) {
1116 err = null_make_cache_space(nullb,
1117 nullb->dev->cache_size * 1024 * 1024);
1118 if (err || nullb->dev->curr_cache == 0)
1119 break;
1122 WARN_ON(!radix_tree_empty(&nullb->dev->cache));
1123 spin_unlock_irq(&nullb->lock);
1124 return err;
1127 static int null_transfer(struct nullb *nullb, struct page *page,
1128 unsigned int len, unsigned int off, bool is_write, sector_t sector,
1129 bool is_fua)
1131 struct nullb_device *dev = nullb->dev;
1132 unsigned int valid_len = len;
1133 int err = 0;
1135 if (!is_write) {
1136 if (dev->zoned)
1137 valid_len = null_zone_valid_read_len(nullb,
1138 sector, len);
1140 if (valid_len) {
1141 err = copy_from_nullb(nullb, page, off,
1142 sector, valid_len);
1143 off += valid_len;
1144 len -= valid_len;
1147 if (len)
1148 nullb_fill_pattern(nullb, page, len, off);
1149 flush_dcache_page(page);
1150 } else {
1151 flush_dcache_page(page);
1152 err = copy_to_nullb(nullb, page, off, sector, len, is_fua);
1155 return err;
1158 static int null_handle_rq(struct nullb_cmd *cmd)
1160 struct request *rq = cmd->rq;
1161 struct nullb *nullb = cmd->nq->dev->nullb;
1162 int err;
1163 unsigned int len;
1164 sector_t sector = blk_rq_pos(rq);
1165 struct req_iterator iter;
1166 struct bio_vec bvec;
1168 spin_lock_irq(&nullb->lock);
1169 rq_for_each_segment(bvec, rq, iter) {
1170 len = bvec.bv_len;
1171 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1172 op_is_write(req_op(rq)), sector,
1173 rq->cmd_flags & REQ_FUA);
1174 if (err) {
1175 spin_unlock_irq(&nullb->lock);
1176 return err;
1178 sector += len >> SECTOR_SHIFT;
1180 spin_unlock_irq(&nullb->lock);
1182 return 0;
1185 static int null_handle_bio(struct nullb_cmd *cmd)
1187 struct bio *bio = cmd->bio;
1188 struct nullb *nullb = cmd->nq->dev->nullb;
1189 int err;
1190 unsigned int len;
1191 sector_t sector = bio->bi_iter.bi_sector;
1192 struct bio_vec bvec;
1193 struct bvec_iter iter;
1195 spin_lock_irq(&nullb->lock);
1196 bio_for_each_segment(bvec, bio, iter) {
1197 len = bvec.bv_len;
1198 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1199 op_is_write(bio_op(bio)), sector,
1200 bio->bi_opf & REQ_FUA);
1201 if (err) {
1202 spin_unlock_irq(&nullb->lock);
1203 return err;
1205 sector += len >> SECTOR_SHIFT;
1207 spin_unlock_irq(&nullb->lock);
1208 return 0;
1211 static void null_stop_queue(struct nullb *nullb)
1213 struct request_queue *q = nullb->q;
1215 if (nullb->dev->queue_mode == NULL_Q_MQ)
1216 blk_mq_stop_hw_queues(q);
1219 static void null_restart_queue_async(struct nullb *nullb)
1221 struct request_queue *q = nullb->q;
1223 if (nullb->dev->queue_mode == NULL_Q_MQ)
1224 blk_mq_start_stopped_hw_queues(q, true);
1227 static inline blk_status_t null_handle_throttled(struct nullb_cmd *cmd)
1229 struct nullb_device *dev = cmd->nq->dev;
1230 struct nullb *nullb = dev->nullb;
1231 blk_status_t sts = BLK_STS_OK;
1232 struct request *rq = cmd->rq;
1234 if (!hrtimer_active(&nullb->bw_timer))
1235 hrtimer_restart(&nullb->bw_timer);
1237 if (atomic_long_sub_return(blk_rq_bytes(rq), &nullb->cur_bytes) < 0) {
1238 null_stop_queue(nullb);
1239 /* race with timer */
1240 if (atomic_long_read(&nullb->cur_bytes) > 0)
1241 null_restart_queue_async(nullb);
1242 /* requeue request */
1243 sts = BLK_STS_DEV_RESOURCE;
1245 return sts;
1248 static inline blk_status_t null_handle_badblocks(struct nullb_cmd *cmd,
1249 sector_t sector,
1250 sector_t nr_sectors)
1252 struct badblocks *bb = &cmd->nq->dev->badblocks;
1253 sector_t first_bad;
1254 int bad_sectors;
1256 if (badblocks_check(bb, sector, nr_sectors, &first_bad, &bad_sectors))
1257 return BLK_STS_IOERR;
1259 return BLK_STS_OK;
1262 static inline blk_status_t null_handle_memory_backed(struct nullb_cmd *cmd,
1263 enum req_opf op,
1264 sector_t sector,
1265 sector_t nr_sectors)
1267 struct nullb_device *dev = cmd->nq->dev;
1268 int err;
1270 if (op == REQ_OP_DISCARD)
1271 return null_handle_discard(dev, sector, nr_sectors);
1273 if (dev->queue_mode == NULL_Q_BIO)
1274 err = null_handle_bio(cmd);
1275 else
1276 err = null_handle_rq(cmd);
1278 return errno_to_blk_status(err);
1281 static void nullb_zero_read_cmd_buffer(struct nullb_cmd *cmd)
1283 struct nullb_device *dev = cmd->nq->dev;
1284 struct bio *bio;
1286 if (dev->memory_backed)
1287 return;
1289 if (dev->queue_mode == NULL_Q_BIO && bio_op(cmd->bio) == REQ_OP_READ) {
1290 zero_fill_bio(cmd->bio);
1291 } else if (req_op(cmd->rq) == REQ_OP_READ) {
1292 __rq_for_each_bio(bio, cmd->rq)
1293 zero_fill_bio(bio);
1297 static inline void nullb_complete_cmd(struct nullb_cmd *cmd)
1300 * Since root privileges are required to configure the null_blk
1301 * driver, it is fine that this driver does not initialize the
1302 * data buffers of read commands. Zero-initialize these buffers
1303 * anyway if KMSAN is enabled to prevent that KMSAN complains
1304 * about null_blk not initializing read data buffers.
1306 if (IS_ENABLED(CONFIG_KMSAN))
1307 nullb_zero_read_cmd_buffer(cmd);
1309 /* Complete IO by inline, softirq or timer */
1310 switch (cmd->nq->dev->irqmode) {
1311 case NULL_IRQ_SOFTIRQ:
1312 switch (cmd->nq->dev->queue_mode) {
1313 case NULL_Q_MQ:
1314 if (likely(!blk_should_fake_timeout(cmd->rq->q)))
1315 blk_mq_complete_request(cmd->rq);
1316 break;
1317 case NULL_Q_BIO:
1319 * XXX: no proper submitting cpu information available.
1321 end_cmd(cmd);
1322 break;
1324 break;
1325 case NULL_IRQ_NONE:
1326 end_cmd(cmd);
1327 break;
1328 case NULL_IRQ_TIMER:
1329 null_cmd_end_timer(cmd);
1330 break;
1334 blk_status_t null_process_cmd(struct nullb_cmd *cmd,
1335 enum req_opf op, sector_t sector,
1336 unsigned int nr_sectors)
1338 struct nullb_device *dev = cmd->nq->dev;
1339 blk_status_t ret;
1341 if (dev->badblocks.shift != -1) {
1342 ret = null_handle_badblocks(cmd, sector, nr_sectors);
1343 if (ret != BLK_STS_OK)
1344 return ret;
1347 if (dev->memory_backed)
1348 return null_handle_memory_backed(cmd, op, sector, nr_sectors);
1350 return BLK_STS_OK;
1353 static blk_status_t null_handle_cmd(struct nullb_cmd *cmd, sector_t sector,
1354 sector_t nr_sectors, enum req_opf op)
1356 struct nullb_device *dev = cmd->nq->dev;
1357 struct nullb *nullb = dev->nullb;
1358 blk_status_t sts;
1360 if (test_bit(NULLB_DEV_FL_THROTTLED, &dev->flags)) {
1361 sts = null_handle_throttled(cmd);
1362 if (sts != BLK_STS_OK)
1363 return sts;
1366 if (op == REQ_OP_FLUSH) {
1367 cmd->error = errno_to_blk_status(null_handle_flush(nullb));
1368 goto out;
1371 if (dev->zoned)
1372 cmd->error = null_process_zoned_cmd(cmd, op,
1373 sector, nr_sectors);
1374 else
1375 cmd->error = null_process_cmd(cmd, op, sector, nr_sectors);
1377 out:
1378 nullb_complete_cmd(cmd);
1379 return BLK_STS_OK;
1382 static enum hrtimer_restart nullb_bwtimer_fn(struct hrtimer *timer)
1384 struct nullb *nullb = container_of(timer, struct nullb, bw_timer);
1385 ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1386 unsigned int mbps = nullb->dev->mbps;
1388 if (atomic_long_read(&nullb->cur_bytes) == mb_per_tick(mbps))
1389 return HRTIMER_NORESTART;
1391 atomic_long_set(&nullb->cur_bytes, mb_per_tick(mbps));
1392 null_restart_queue_async(nullb);
1394 hrtimer_forward_now(&nullb->bw_timer, timer_interval);
1396 return HRTIMER_RESTART;
1399 static void nullb_setup_bwtimer(struct nullb *nullb)
1401 ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1403 hrtimer_init(&nullb->bw_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1404 nullb->bw_timer.function = nullb_bwtimer_fn;
1405 atomic_long_set(&nullb->cur_bytes, mb_per_tick(nullb->dev->mbps));
1406 hrtimer_start(&nullb->bw_timer, timer_interval, HRTIMER_MODE_REL);
1409 static struct nullb_queue *nullb_to_queue(struct nullb *nullb)
1411 int index = 0;
1413 if (nullb->nr_queues != 1)
1414 index = raw_smp_processor_id() / ((nr_cpu_ids + nullb->nr_queues - 1) / nullb->nr_queues);
1416 return &nullb->queues[index];
1419 static blk_qc_t null_submit_bio(struct bio *bio)
1421 sector_t sector = bio->bi_iter.bi_sector;
1422 sector_t nr_sectors = bio_sectors(bio);
1423 struct nullb *nullb = bio->bi_disk->private_data;
1424 struct nullb_queue *nq = nullb_to_queue(nullb);
1425 struct nullb_cmd *cmd;
1427 cmd = alloc_cmd(nq, 1);
1428 cmd->bio = bio;
1430 null_handle_cmd(cmd, sector, nr_sectors, bio_op(bio));
1431 return BLK_QC_T_NONE;
1434 static bool should_timeout_request(struct request *rq)
1436 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1437 if (g_timeout_str[0])
1438 return should_fail(&null_timeout_attr, 1);
1439 #endif
1440 return false;
1443 static bool should_requeue_request(struct request *rq)
1445 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1446 if (g_requeue_str[0])
1447 return should_fail(&null_requeue_attr, 1);
1448 #endif
1449 return false;
1452 static enum blk_eh_timer_return null_timeout_rq(struct request *rq, bool res)
1454 pr_info("rq %p timed out\n", rq);
1455 blk_mq_complete_request(rq);
1456 return BLK_EH_DONE;
1459 static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx,
1460 const struct blk_mq_queue_data *bd)
1462 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
1463 struct nullb_queue *nq = hctx->driver_data;
1464 sector_t nr_sectors = blk_rq_sectors(bd->rq);
1465 sector_t sector = blk_rq_pos(bd->rq);
1467 might_sleep_if(hctx->flags & BLK_MQ_F_BLOCKING);
1469 if (nq->dev->irqmode == NULL_IRQ_TIMER) {
1470 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1471 cmd->timer.function = null_cmd_timer_expired;
1473 cmd->rq = bd->rq;
1474 cmd->error = BLK_STS_OK;
1475 cmd->nq = nq;
1477 blk_mq_start_request(bd->rq);
1479 if (should_requeue_request(bd->rq)) {
1481 * Alternate between hitting the core BUSY path, and the
1482 * driver driven requeue path
1484 nq->requeue_selection++;
1485 if (nq->requeue_selection & 1)
1486 return BLK_STS_RESOURCE;
1487 else {
1488 blk_mq_requeue_request(bd->rq, true);
1489 return BLK_STS_OK;
1492 if (should_timeout_request(bd->rq))
1493 return BLK_STS_OK;
1495 return null_handle_cmd(cmd, sector, nr_sectors, req_op(bd->rq));
1498 static void cleanup_queue(struct nullb_queue *nq)
1500 kfree(nq->tag_map);
1501 kfree(nq->cmds);
1504 static void cleanup_queues(struct nullb *nullb)
1506 int i;
1508 for (i = 0; i < nullb->nr_queues; i++)
1509 cleanup_queue(&nullb->queues[i]);
1511 kfree(nullb->queues);
1514 static void null_exit_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx)
1516 struct nullb_queue *nq = hctx->driver_data;
1517 struct nullb *nullb = nq->dev->nullb;
1519 nullb->nr_queues--;
1522 static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq)
1524 init_waitqueue_head(&nq->wait);
1525 nq->queue_depth = nullb->queue_depth;
1526 nq->dev = nullb->dev;
1529 static int null_init_hctx(struct blk_mq_hw_ctx *hctx, void *driver_data,
1530 unsigned int hctx_idx)
1532 struct nullb *nullb = hctx->queue->queuedata;
1533 struct nullb_queue *nq;
1535 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1536 if (g_init_hctx_str[0] && should_fail(&null_init_hctx_attr, 1))
1537 return -EFAULT;
1538 #endif
1540 nq = &nullb->queues[hctx_idx];
1541 hctx->driver_data = nq;
1542 null_init_queue(nullb, nq);
1543 nullb->nr_queues++;
1545 return 0;
1548 static const struct blk_mq_ops null_mq_ops = {
1549 .queue_rq = null_queue_rq,
1550 .complete = null_complete_rq,
1551 .timeout = null_timeout_rq,
1552 .init_hctx = null_init_hctx,
1553 .exit_hctx = null_exit_hctx,
1556 static void null_del_dev(struct nullb *nullb)
1558 struct nullb_device *dev;
1560 if (!nullb)
1561 return;
1563 dev = nullb->dev;
1565 ida_simple_remove(&nullb_indexes, nullb->index);
1567 list_del_init(&nullb->list);
1569 del_gendisk(nullb->disk);
1571 if (test_bit(NULLB_DEV_FL_THROTTLED, &nullb->dev->flags)) {
1572 hrtimer_cancel(&nullb->bw_timer);
1573 atomic_long_set(&nullb->cur_bytes, LONG_MAX);
1574 null_restart_queue_async(nullb);
1577 blk_cleanup_queue(nullb->q);
1578 if (dev->queue_mode == NULL_Q_MQ &&
1579 nullb->tag_set == &nullb->__tag_set)
1580 blk_mq_free_tag_set(nullb->tag_set);
1581 put_disk(nullb->disk);
1582 cleanup_queues(nullb);
1583 if (null_cache_active(nullb))
1584 null_free_device_storage(nullb->dev, true);
1585 kfree(nullb);
1586 dev->nullb = NULL;
1589 static void null_config_discard(struct nullb *nullb)
1591 if (nullb->dev->discard == false)
1592 return;
1594 if (!nullb->dev->memory_backed) {
1595 nullb->dev->discard = false;
1596 pr_info("discard option is ignored without memory backing\n");
1597 return;
1600 if (nullb->dev->zoned) {
1601 nullb->dev->discard = false;
1602 pr_info("discard option is ignored in zoned mode\n");
1603 return;
1606 nullb->q->limits.discard_granularity = nullb->dev->blocksize;
1607 nullb->q->limits.discard_alignment = nullb->dev->blocksize;
1608 blk_queue_max_discard_sectors(nullb->q, UINT_MAX >> 9);
1609 blk_queue_flag_set(QUEUE_FLAG_DISCARD, nullb->q);
1612 static const struct block_device_operations null_bio_ops = {
1613 .owner = THIS_MODULE,
1614 .submit_bio = null_submit_bio,
1615 .report_zones = null_report_zones,
1618 static const struct block_device_operations null_rq_ops = {
1619 .owner = THIS_MODULE,
1620 .report_zones = null_report_zones,
1623 static int setup_commands(struct nullb_queue *nq)
1625 struct nullb_cmd *cmd;
1626 int i, tag_size;
1628 nq->cmds = kcalloc(nq->queue_depth, sizeof(*cmd), GFP_KERNEL);
1629 if (!nq->cmds)
1630 return -ENOMEM;
1632 tag_size = ALIGN(nq->queue_depth, BITS_PER_LONG) / BITS_PER_LONG;
1633 nq->tag_map = kcalloc(tag_size, sizeof(unsigned long), GFP_KERNEL);
1634 if (!nq->tag_map) {
1635 kfree(nq->cmds);
1636 return -ENOMEM;
1639 for (i = 0; i < nq->queue_depth; i++) {
1640 cmd = &nq->cmds[i];
1641 cmd->tag = -1U;
1644 return 0;
1647 static int setup_queues(struct nullb *nullb)
1649 nullb->queues = kcalloc(nr_cpu_ids, sizeof(struct nullb_queue),
1650 GFP_KERNEL);
1651 if (!nullb->queues)
1652 return -ENOMEM;
1654 nullb->queue_depth = nullb->dev->hw_queue_depth;
1656 return 0;
1659 static int init_driver_queues(struct nullb *nullb)
1661 struct nullb_queue *nq;
1662 int i, ret = 0;
1664 for (i = 0; i < nullb->dev->submit_queues; i++) {
1665 nq = &nullb->queues[i];
1667 null_init_queue(nullb, nq);
1669 ret = setup_commands(nq);
1670 if (ret)
1671 return ret;
1672 nullb->nr_queues++;
1674 return 0;
1677 static int null_gendisk_register(struct nullb *nullb)
1679 sector_t size = ((sector_t)nullb->dev->size * SZ_1M) >> SECTOR_SHIFT;
1680 struct gendisk *disk;
1682 disk = nullb->disk = alloc_disk_node(1, nullb->dev->home_node);
1683 if (!disk)
1684 return -ENOMEM;
1685 set_capacity(disk, size);
1687 disk->flags |= GENHD_FL_EXT_DEVT | GENHD_FL_SUPPRESS_PARTITION_INFO;
1688 disk->major = null_major;
1689 disk->first_minor = nullb->index;
1690 if (queue_is_mq(nullb->q))
1691 disk->fops = &null_rq_ops;
1692 else
1693 disk->fops = &null_bio_ops;
1694 disk->private_data = nullb;
1695 disk->queue = nullb->q;
1696 strncpy(disk->disk_name, nullb->disk_name, DISK_NAME_LEN);
1698 if (nullb->dev->zoned) {
1699 int ret = null_register_zoned_dev(nullb);
1701 if (ret)
1702 return ret;
1705 add_disk(disk);
1706 return 0;
1709 static int null_init_tag_set(struct nullb *nullb, struct blk_mq_tag_set *set)
1711 set->ops = &null_mq_ops;
1712 set->nr_hw_queues = nullb ? nullb->dev->submit_queues :
1713 g_submit_queues;
1714 set->queue_depth = nullb ? nullb->dev->hw_queue_depth :
1715 g_hw_queue_depth;
1716 set->numa_node = nullb ? nullb->dev->home_node : g_home_node;
1717 set->cmd_size = sizeof(struct nullb_cmd);
1718 set->flags = BLK_MQ_F_SHOULD_MERGE;
1719 if (g_no_sched)
1720 set->flags |= BLK_MQ_F_NO_SCHED;
1721 if (g_shared_tag_bitmap)
1722 set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1723 set->driver_data = NULL;
1725 if ((nullb && nullb->dev->blocking) || g_blocking)
1726 set->flags |= BLK_MQ_F_BLOCKING;
1728 return blk_mq_alloc_tag_set(set);
1731 static int null_validate_conf(struct nullb_device *dev)
1733 dev->blocksize = round_down(dev->blocksize, 512);
1734 dev->blocksize = clamp_t(unsigned int, dev->blocksize, 512, 4096);
1736 if (dev->queue_mode == NULL_Q_MQ && dev->use_per_node_hctx) {
1737 if (dev->submit_queues != nr_online_nodes)
1738 dev->submit_queues = nr_online_nodes;
1739 } else if (dev->submit_queues > nr_cpu_ids)
1740 dev->submit_queues = nr_cpu_ids;
1741 else if (dev->submit_queues == 0)
1742 dev->submit_queues = 1;
1744 dev->queue_mode = min_t(unsigned int, dev->queue_mode, NULL_Q_MQ);
1745 dev->irqmode = min_t(unsigned int, dev->irqmode, NULL_IRQ_TIMER);
1747 /* Do memory allocation, so set blocking */
1748 if (dev->memory_backed)
1749 dev->blocking = true;
1750 else /* cache is meaningless */
1751 dev->cache_size = 0;
1752 dev->cache_size = min_t(unsigned long, ULONG_MAX / 1024 / 1024,
1753 dev->cache_size);
1754 dev->mbps = min_t(unsigned int, 1024 * 40, dev->mbps);
1755 /* can not stop a queue */
1756 if (dev->queue_mode == NULL_Q_BIO)
1757 dev->mbps = 0;
1759 if (dev->zoned &&
1760 (!dev->zone_size || !is_power_of_2(dev->zone_size))) {
1761 pr_err("zone_size must be power-of-two\n");
1762 return -EINVAL;
1765 return 0;
1768 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1769 static bool __null_setup_fault(struct fault_attr *attr, char *str)
1771 if (!str[0])
1772 return true;
1774 if (!setup_fault_attr(attr, str))
1775 return false;
1777 attr->verbose = 0;
1778 return true;
1780 #endif
1782 static bool null_setup_fault(void)
1784 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1785 if (!__null_setup_fault(&null_timeout_attr, g_timeout_str))
1786 return false;
1787 if (!__null_setup_fault(&null_requeue_attr, g_requeue_str))
1788 return false;
1789 if (!__null_setup_fault(&null_init_hctx_attr, g_init_hctx_str))
1790 return false;
1791 #endif
1792 return true;
1795 static int null_add_dev(struct nullb_device *dev)
1797 struct nullb *nullb;
1798 int rv;
1800 rv = null_validate_conf(dev);
1801 if (rv)
1802 return rv;
1804 nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, dev->home_node);
1805 if (!nullb) {
1806 rv = -ENOMEM;
1807 goto out;
1809 nullb->dev = dev;
1810 dev->nullb = nullb;
1812 spin_lock_init(&nullb->lock);
1814 rv = setup_queues(nullb);
1815 if (rv)
1816 goto out_free_nullb;
1818 if (dev->queue_mode == NULL_Q_MQ) {
1819 if (shared_tags) {
1820 nullb->tag_set = &tag_set;
1821 rv = 0;
1822 } else {
1823 nullb->tag_set = &nullb->__tag_set;
1824 rv = null_init_tag_set(nullb, nullb->tag_set);
1827 if (rv)
1828 goto out_cleanup_queues;
1830 if (!null_setup_fault())
1831 goto out_cleanup_queues;
1833 nullb->tag_set->timeout = 5 * HZ;
1834 nullb->q = blk_mq_init_queue_data(nullb->tag_set, nullb);
1835 if (IS_ERR(nullb->q)) {
1836 rv = -ENOMEM;
1837 goto out_cleanup_tags;
1839 } else if (dev->queue_mode == NULL_Q_BIO) {
1840 nullb->q = blk_alloc_queue(dev->home_node);
1841 if (!nullb->q) {
1842 rv = -ENOMEM;
1843 goto out_cleanup_queues;
1845 rv = init_driver_queues(nullb);
1846 if (rv)
1847 goto out_cleanup_blk_queue;
1850 if (dev->mbps) {
1851 set_bit(NULLB_DEV_FL_THROTTLED, &dev->flags);
1852 nullb_setup_bwtimer(nullb);
1855 if (dev->cache_size > 0) {
1856 set_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
1857 blk_queue_write_cache(nullb->q, true, true);
1860 if (dev->zoned) {
1861 rv = null_init_zoned_dev(dev, nullb->q);
1862 if (rv)
1863 goto out_cleanup_blk_queue;
1866 nullb->q->queuedata = nullb;
1867 blk_queue_flag_set(QUEUE_FLAG_NONROT, nullb->q);
1868 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, nullb->q);
1870 mutex_lock(&lock);
1871 nullb->index = ida_simple_get(&nullb_indexes, 0, 0, GFP_KERNEL);
1872 dev->index = nullb->index;
1873 mutex_unlock(&lock);
1875 blk_queue_logical_block_size(nullb->q, dev->blocksize);
1876 blk_queue_physical_block_size(nullb->q, dev->blocksize);
1877 if (!dev->max_sectors)
1878 dev->max_sectors = queue_max_hw_sectors(nullb->q);
1879 dev->max_sectors = min_t(unsigned int, dev->max_sectors,
1880 BLK_DEF_MAX_SECTORS);
1881 blk_queue_max_hw_sectors(nullb->q, dev->max_sectors);
1883 null_config_discard(nullb);
1885 sprintf(nullb->disk_name, "nullb%d", nullb->index);
1887 rv = null_gendisk_register(nullb);
1888 if (rv)
1889 goto out_cleanup_zone;
1891 mutex_lock(&lock);
1892 list_add_tail(&nullb->list, &nullb_list);
1893 mutex_unlock(&lock);
1895 return 0;
1896 out_cleanup_zone:
1897 null_free_zoned_dev(dev);
1898 out_cleanup_blk_queue:
1899 blk_cleanup_queue(nullb->q);
1900 out_cleanup_tags:
1901 if (dev->queue_mode == NULL_Q_MQ && nullb->tag_set == &nullb->__tag_set)
1902 blk_mq_free_tag_set(nullb->tag_set);
1903 out_cleanup_queues:
1904 cleanup_queues(nullb);
1905 out_free_nullb:
1906 kfree(nullb);
1907 dev->nullb = NULL;
1908 out:
1909 return rv;
1912 static int __init null_init(void)
1914 int ret = 0;
1915 unsigned int i;
1916 struct nullb *nullb;
1917 struct nullb_device *dev;
1919 if (g_bs > PAGE_SIZE) {
1920 pr_warn("invalid block size\n");
1921 pr_warn("defaults block size to %lu\n", PAGE_SIZE);
1922 g_bs = PAGE_SIZE;
1925 if (g_max_sectors > BLK_DEF_MAX_SECTORS) {
1926 pr_warn("invalid max sectors\n");
1927 pr_warn("defaults max sectors to %u\n", BLK_DEF_MAX_SECTORS);
1928 g_max_sectors = BLK_DEF_MAX_SECTORS;
1931 if (g_home_node != NUMA_NO_NODE && g_home_node >= nr_online_nodes) {
1932 pr_err("invalid home_node value\n");
1933 g_home_node = NUMA_NO_NODE;
1936 if (g_queue_mode == NULL_Q_RQ) {
1937 pr_err("legacy IO path no longer available\n");
1938 return -EINVAL;
1940 if (g_queue_mode == NULL_Q_MQ && g_use_per_node_hctx) {
1941 if (g_submit_queues != nr_online_nodes) {
1942 pr_warn("submit_queues param is set to %u.\n",
1943 nr_online_nodes);
1944 g_submit_queues = nr_online_nodes;
1946 } else if (g_submit_queues > nr_cpu_ids)
1947 g_submit_queues = nr_cpu_ids;
1948 else if (g_submit_queues <= 0)
1949 g_submit_queues = 1;
1951 if (g_queue_mode == NULL_Q_MQ && shared_tags) {
1952 ret = null_init_tag_set(NULL, &tag_set);
1953 if (ret)
1954 return ret;
1957 config_group_init(&nullb_subsys.su_group);
1958 mutex_init(&nullb_subsys.su_mutex);
1960 ret = configfs_register_subsystem(&nullb_subsys);
1961 if (ret)
1962 goto err_tagset;
1964 mutex_init(&lock);
1966 null_major = register_blkdev(0, "nullb");
1967 if (null_major < 0) {
1968 ret = null_major;
1969 goto err_conf;
1972 for (i = 0; i < nr_devices; i++) {
1973 dev = null_alloc_dev();
1974 if (!dev) {
1975 ret = -ENOMEM;
1976 goto err_dev;
1978 ret = null_add_dev(dev);
1979 if (ret) {
1980 null_free_dev(dev);
1981 goto err_dev;
1985 pr_info("module loaded\n");
1986 return 0;
1988 err_dev:
1989 while (!list_empty(&nullb_list)) {
1990 nullb = list_entry(nullb_list.next, struct nullb, list);
1991 dev = nullb->dev;
1992 null_del_dev(nullb);
1993 null_free_dev(dev);
1995 unregister_blkdev(null_major, "nullb");
1996 err_conf:
1997 configfs_unregister_subsystem(&nullb_subsys);
1998 err_tagset:
1999 if (g_queue_mode == NULL_Q_MQ && shared_tags)
2000 blk_mq_free_tag_set(&tag_set);
2001 return ret;
2004 static void __exit null_exit(void)
2006 struct nullb *nullb;
2008 configfs_unregister_subsystem(&nullb_subsys);
2010 unregister_blkdev(null_major, "nullb");
2012 mutex_lock(&lock);
2013 while (!list_empty(&nullb_list)) {
2014 struct nullb_device *dev;
2016 nullb = list_entry(nullb_list.next, struct nullb, list);
2017 dev = nullb->dev;
2018 null_del_dev(nullb);
2019 null_free_dev(dev);
2021 mutex_unlock(&lock);
2023 if (g_queue_mode == NULL_Q_MQ && shared_tags)
2024 blk_mq_free_tag_set(&tag_set);
2027 module_init(null_init);
2028 module_exit(null_exit);
2030 MODULE_AUTHOR("Jens Axboe <axboe@kernel.dk>");
2031 MODULE_LICENSE("GPL");