projects/linux·blob master · refs

2225 lines · 56.9 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Add configfs and memory store: Kyungchan Koh <[email protected]> and
4 * Shaohua Li <[email protected]>
5 */
6#include <linux/module.h>
7
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"
13
14#undef pr_fmt
15#define pr_fmt(fmt) "null_blk: " fmt
16
17#define FREE_BATCH 16
18
19#define TICKS_PER_SEC 50ULL
20#define TIMER_INTERVAL (NSEC_PER_SEC / TICKS_PER_SEC)
21
22#ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
23static DECLARE_FAULT_ATTR(null_timeout_attr);
24static DECLARE_FAULT_ATTR(null_requeue_attr);
25static DECLARE_FAULT_ATTR(null_init_hctx_attr);
26#endif
27
28static inline u64 mb_per_tick(int mbps)
29{
30 return (1 << 20) / TICKS_PER_SEC * ((u64) mbps);
31}
32
33/*
34 * Status flags for nullb_device.
35 *
36 * CONFIGURED: Device has been configured and turned on. Cannot reconfigure.
37 * UP: Device is currently on and visible in userspace.
38 * THROTTLED: Device is being throttled.
39 * CACHE: Device is using a write-back cache.
40 */
41enum nullb_device_flags {
42 NULLB_DEV_FL_CONFIGURED = 0,
43 NULLB_DEV_FL_UP = 1,
44 NULLB_DEV_FL_THROTTLED = 2,
45 NULLB_DEV_FL_CACHE = 3,
46};
47
48#define MAP_SZ ((PAGE_SIZE >> SECTOR_SHIFT) + 2)
49/*
50 * nullb_page is a page in memory for nullb devices.
51 *
52 * @page: The page holding the data.
53 * @bitmap: The bitmap represents which sector in the page has data.
54 * Each bit represents one block size. For example, sector 8
55 * will use the 7th bit
56 * The highest 2 bits of bitmap are for special purpose. LOCK means the cache
57 * page is being flushing to storage. FREE means the cache page is freed and
58 * should be skipped from flushing to storage. Please see
59 * null_make_cache_space
60 */
61struct nullb_page {
62 struct page *page;
63 DECLARE_BITMAP(bitmap, MAP_SZ);
64};
65#define NULLB_PAGE_LOCK (MAP_SZ - 1)
66#define NULLB_PAGE_FREE (MAP_SZ - 2)
67
68static LIST_HEAD(nullb_list);
69static DEFINE_MUTEX(lock);
70static int null_major;
71static DEFINE_IDA(nullb_indexes);
72static struct blk_mq_tag_set tag_set;
73
74enum {
75 NULL_IRQ_NONE = 0,
76 NULL_IRQ_SOFTIRQ = 1,
77 NULL_IRQ_TIMER = 2,
78};
79
80static bool g_virt_boundary;
81module_param_named(virt_boundary, g_virt_boundary, bool, 0444);
82MODULE_PARM_DESC(virt_boundary, "Require a virtual boundary for the device. Default: False");
83
84static int g_no_sched;
85module_param_named(no_sched, g_no_sched, int, 0444);
86MODULE_PARM_DESC(no_sched, "No io scheduler");
87
88static int g_submit_queues = 1;
89module_param_named(submit_queues, g_submit_queues, int, 0444);
90MODULE_PARM_DESC(submit_queues, "Number of submission queues");
91
92static int g_poll_queues = 1;
93module_param_named(poll_queues, g_poll_queues, int, 0444);
94MODULE_PARM_DESC(poll_queues, "Number of IOPOLL submission queues");
95
96static int g_home_node = NUMA_NO_NODE;
97module_param_named(home_node, g_home_node, int, 0444);
98MODULE_PARM_DESC(home_node, "Home node for the device");
99
100#ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
101/*
102 * For more details about fault injection, please refer to
103 * Documentation/fault-injection/fault-injection.rst.
104 */
105static char g_timeout_str[80];
106module_param_string(timeout, g_timeout_str, sizeof(g_timeout_str), 0444);
107MODULE_PARM_DESC(timeout, "Fault injection. timeout=<interval>,<probability>,<space>,<times>");
108
109static char g_requeue_str[80];
110module_param_string(requeue, g_requeue_str, sizeof(g_requeue_str), 0444);
111MODULE_PARM_DESC(requeue, "Fault injection. requeue=<interval>,<probability>,<space>,<times>");
112
113static char g_init_hctx_str[80];
114module_param_string(init_hctx, g_init_hctx_str, sizeof(g_init_hctx_str), 0444);
115MODULE_PARM_DESC(init_hctx, "Fault injection to fail hctx init. init_hctx=<interval>,<probability>,<space>,<times>");
116#endif
117
118/*
119 * Historic queue modes.
120 *
121 * These days nothing but NULL_Q_MQ is actually supported, but we keep it the
122 * enum for error reporting.
123 */
124enum {
125 NULL_Q_BIO = 0,
126 NULL_Q_RQ = 1,
127 NULL_Q_MQ = 2,
128};
129
130static int g_queue_mode = NULL_Q_MQ;
131
132static int null_param_store_val(const char *str, int *val, int min, int max)
133{
134 int ret, new_val;
135
136 ret = kstrtoint(str, 10, &new_val);
137 if (ret)
138 return -EINVAL;
139
140 if (new_val < min || new_val > max)
141 return -EINVAL;
142
143 *val = new_val;
144 return 0;
145}
146
147static int null_set_queue_mode(const char *str, const struct kernel_param *kp)
148{
149 return null_param_store_val(str, &g_queue_mode, NULL_Q_BIO, NULL_Q_MQ);
150}
151
152static const struct kernel_param_ops null_queue_mode_param_ops = {
153 .set = null_set_queue_mode,
154 .get = param_get_int,
155};
156
157device_param_cb(queue_mode, &null_queue_mode_param_ops, &g_queue_mode, 0444);
158MODULE_PARM_DESC(queue_mode, "Block interface to use (0=bio,1=rq,2=multiqueue)");
159
160static int g_gb = 250;
161module_param_named(gb, g_gb, int, 0444);
162MODULE_PARM_DESC(gb, "Size in GB");
163
164static int g_bs = 512;
165module_param_named(bs, g_bs, int, 0444);
166MODULE_PARM_DESC(bs, "Block size (in bytes)");
167
168static int g_max_sectors;
169module_param_named(max_sectors, g_max_sectors, int, 0444);
170MODULE_PARM_DESC(max_sectors, "Maximum size of a command (in 512B sectors)");
171
172static unsigned int nr_devices = 1;
173module_param(nr_devices, uint, 0444);
174MODULE_PARM_DESC(nr_devices, "Number of devices to register");
175
176static bool g_blocking;
177module_param_named(blocking, g_blocking, bool, 0444);
178MODULE_PARM_DESC(blocking, "Register as a blocking blk-mq driver device");
179
180static bool g_shared_tags;
181module_param_named(shared_tags, g_shared_tags, bool, 0444);
182MODULE_PARM_DESC(shared_tags, "Share tag set between devices for blk-mq");
183
184static bool g_shared_tag_bitmap;
185module_param_named(shared_tag_bitmap, g_shared_tag_bitmap, bool, 0444);
186MODULE_PARM_DESC(shared_tag_bitmap, "Use shared tag bitmap for all submission queues for blk-mq");
187
188static int g_irqmode = NULL_IRQ_SOFTIRQ;
189
190static int null_set_irqmode(const char *str, const struct kernel_param *kp)
191{
192 return null_param_store_val(str, &g_irqmode, NULL_IRQ_NONE,
193 NULL_IRQ_TIMER);
194}
195
196static const struct kernel_param_ops null_irqmode_param_ops = {
197 .set = null_set_irqmode,
198 .get = param_get_int,
199};
200
201device_param_cb(irqmode, &null_irqmode_param_ops, &g_irqmode, 0444);
202MODULE_PARM_DESC(irqmode, "IRQ completion handler. 0-none, 1-softirq, 2-timer");
203
204static unsigned long g_completion_nsec = 10000;
205module_param_named(completion_nsec, g_completion_nsec, ulong, 0444);
206MODULE_PARM_DESC(completion_nsec, "Time in ns to complete a request in hardware. Default: 10,000ns");
207
208static int g_hw_queue_depth = 64;
209module_param_named(hw_queue_depth, g_hw_queue_depth, int, 0444);
210MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 64");
211
212static bool g_use_per_node_hctx;
213module_param_named(use_per_node_hctx, g_use_per_node_hctx, bool, 0444);
214MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware context queues. Default: false");
215
216static bool g_memory_backed;
217module_param_named(memory_backed, g_memory_backed, bool, 0444);
218MODULE_PARM_DESC(memory_backed, "Create a memory-backed block device. Default: false");
219
220static bool g_discard;
221module_param_named(discard, g_discard, bool, 0444);
222MODULE_PARM_DESC(discard, "Support discard operations (requires memory-backed null_blk device). Default: false");
223
224static unsigned long g_cache_size;
225module_param_named(cache_size, g_cache_size, ulong, 0444);
226MODULE_PARM_DESC(cache_size, "Cache size in MiB for memory-backed device. Default: 0 (none)");
227
228static bool g_fua = true;
229module_param_named(fua, g_fua, bool, 0444);
230MODULE_PARM_DESC(fua, "Enable/disable FUA support when cache_size is used. Default: true");
231
232static unsigned int g_mbps;
233module_param_named(mbps, g_mbps, uint, 0444);
234MODULE_PARM_DESC(mbps, "Limit maximum bandwidth (in MiB/s). Default: 0 (no limit)");
235
236static bool g_zoned;
237module_param_named(zoned, g_zoned, bool, S_IRUGO);
238MODULE_PARM_DESC(zoned, "Make device as a host-managed zoned block device. Default: false");
239
240static unsigned long g_zone_size = 256;
241module_param_named(zone_size, g_zone_size, ulong, S_IRUGO);
242MODULE_PARM_DESC(zone_size, "Zone size in MB when block device is zoned. Must be power-of-two: Default: 256");
243
244static unsigned long g_zone_capacity;
245module_param_named(zone_capacity, g_zone_capacity, ulong, 0444);
246MODULE_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");
247
248static unsigned int g_zone_nr_conv;
249module_param_named(zone_nr_conv, g_zone_nr_conv, uint, 0444);
250MODULE_PARM_DESC(zone_nr_conv, "Number of conventional zones when block device is zoned. Default: 0");
251
252static unsigned int g_zone_max_open;
253module_param_named(zone_max_open, g_zone_max_open, uint, 0444);
254MODULE_PARM_DESC(zone_max_open, "Maximum number of open zones when block device is zoned. Default: 0 (no limit)");
255
256static unsigned int g_zone_max_active;
257module_param_named(zone_max_active, g_zone_max_active, uint, 0444);
258MODULE_PARM_DESC(zone_max_active, "Maximum number of active zones when block device is zoned. Default: 0 (no limit)");
259
260static int g_zone_append_max_sectors = INT_MAX;
261module_param_named(zone_append_max_sectors, g_zone_append_max_sectors, int, 0444);
262MODULE_PARM_DESC(zone_append_max_sectors,
263 "Maximum size of a zone append command (in 512B sectors). Specify 0 for zone append emulation");
264
265static bool g_zone_full;
266module_param_named(zone_full, g_zone_full, bool, S_IRUGO);
267MODULE_PARM_DESC(zone_full, "Initialize the sequential write required zones of a zoned device to be full. Default: false");
268
269static bool g_rotational;
270module_param_named(rotational, g_rotational, bool, S_IRUGO);
271MODULE_PARM_DESC(rotational, "Set the rotational feature for the device. Default: false");
272
273static struct nullb_device *null_alloc_dev(void);
274static void null_free_dev(struct nullb_device *dev);
275static void null_del_dev(struct nullb *nullb);
276static int null_add_dev(struct nullb_device *dev);
277static struct nullb *null_find_dev_by_name(const char *name);
278static void null_free_device_storage(struct nullb_device *dev, bool is_cache);
279
280static inline struct nullb_device *to_nullb_device(struct config_item *item)
281{
282 return item ? container_of(to_config_group(item), struct nullb_device, group) : NULL;
283}
284
285static inline ssize_t nullb_device_uint_attr_show(unsigned int val, char *page)
286{
287 return snprintf(page, PAGE_SIZE, "%u\n", val);
288}
289
290static inline ssize_t nullb_device_ulong_attr_show(unsigned long val,
291 char *page)
292{
293 return snprintf(page, PAGE_SIZE, "%lu\n", val);
294}
295
296static inline ssize_t nullb_device_bool_attr_show(bool val, char *page)
297{
298 return snprintf(page, PAGE_SIZE, "%u\n", val);
299}
300
301static ssize_t nullb_device_uint_attr_store(unsigned int *val,
302 const char *page, size_t count)
303{
304 unsigned int tmp;
305 int result;
306
307 result = kstrtouint(page, 0, &tmp);
308 if (result < 0)
309 return result;
310
311 *val = tmp;
312 return count;
313}
314
315static ssize_t nullb_device_ulong_attr_store(unsigned long *val,
316 const char *page, size_t count)
317{
318 int result;
319 unsigned long tmp;
320
321 result = kstrtoul(page, 0, &tmp);
322 if (result < 0)
323 return result;
324
325 *val = tmp;
326 return count;
327}
328
329static ssize_t nullb_device_bool_attr_store(bool *val, const char *page,
330 size_t count)
331{
332 bool tmp;
333 int result;
334
335 result = kstrtobool(page, &tmp);
336 if (result < 0)
337 return result;
338
339 *val = tmp;
340 return count;
341}
342
343/*
344 * The following macro should only be used with TYPE = {uint, ulong, bool}.
345 *
346 * The device configuration is modified under the global lock to serialize
347 * attribute changes against null_add_dev() and null_del_dev(): without this,
348 * an attribute could be changed while null_add_dev() is running, that is,
349 * before NULLB_DEV_FL_CONFIGURED is set, which would let null_add_dev()
350 * observe inconsistent values for the device configuration.
351 */
352#define NULLB_DEVICE_ATTR(NAME, TYPE, APPLY) \
353static ssize_t \
354nullb_device_##NAME##_show(struct config_item *item, char *page) \
355{ \
356 guard(mutex)(&lock); \
357 return nullb_device_##TYPE##_attr_show( \
358 to_nullb_device(item)->NAME, page); \
359} \
360static ssize_t \
361nullb_device_##NAME##_store(struct config_item *item, const char *page, \
362 size_t count) \
363{ \
364 int (*apply_fn)(struct nullb_device *dev, TYPE new_value) = APPLY;\
365 struct nullb_device *dev = to_nullb_device(item); \
366 TYPE new_value = 0; \
367 int ret; \
368 \
369 ret = nullb_device_##TYPE##_attr_store(&new_value, page, count);\
370 if (ret < 0) \
371 return ret; \
372 guard(mutex)(&lock); \
373 if (apply_fn) \
374 ret = apply_fn(dev, new_value); \
375 else if (test_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags)) \
376 ret = -EBUSY; \
377 if (ret < 0) \
378 return ret; \
379 dev->NAME = new_value; \
380 return count; \
381} \
382CONFIGFS_ATTR(nullb_device_, NAME);
383
384static int nullb_update_nr_hw_queues(struct nullb_device *dev,
385 unsigned int submit_queues,
386 unsigned int poll_queues)
387
388{
389 struct blk_mq_tag_set *set;
390 int ret, nr_hw_queues;
391
392 lockdep_assert_held(&lock);
393
394 if (!dev->nullb)
395 return 0;
396
397 /*
398 * A shared tag_set is mapped via the module-wide queue counts, so a
399 * per-device resize is meaningless. On shrink it would also leave
400 * mq_map[] pointing at NULLed hctx slots, causing a NULL deref in
401 * blk_mq_map_swqueue(). Reject it.
402 */
403 if (dev->shared_tags)
404 return -EINVAL;
405
406 /*
407 * Make sure at least one submit queue exists.
408 */
409 if (!submit_queues)
410 return -EINVAL;
411
412 /*
413 * Make sure that null_init_hctx() does not access nullb->queues[] past
414 * the end of that array.
415 */
416 if (submit_queues > nr_cpu_ids || poll_queues > g_poll_queues)
417 return -EINVAL;
418
419 /*
420 * Keep previous and new queue numbers in nullb_device for reference in
421 * the call back function null_map_queues().
422 */
423 dev->prev_submit_queues = dev->submit_queues;
424 dev->prev_poll_queues = dev->poll_queues;
425 dev->submit_queues = submit_queues;
426 dev->poll_queues = poll_queues;
427
428 set = dev->nullb->tag_set;
429 nr_hw_queues = submit_queues + poll_queues;
430 blk_mq_update_nr_hw_queues(set, nr_hw_queues);
431 ret = set->nr_hw_queues == nr_hw_queues ? 0 : -ENOMEM;
432
433 if (ret) {
434 /* on error, revert the queue numbers */
435 dev->submit_queues = dev->prev_submit_queues;
436 dev->poll_queues = dev->prev_poll_queues;
437 }
438
439 return ret;
440}
441
442static int nullb_apply_submit_queues(struct nullb_device *dev,
443 unsigned int submit_queues)
444{
445 return nullb_update_nr_hw_queues(dev, submit_queues, dev->poll_queues);
446}
447
448static int nullb_apply_poll_queues(struct nullb_device *dev,
449 unsigned int poll_queues)
450{
451 return nullb_update_nr_hw_queues(dev, dev->submit_queues, poll_queues);
452}
453
454NULLB_DEVICE_ATTR(size, ulong, NULL);
455NULLB_DEVICE_ATTR(completion_nsec, ulong, NULL);
456NULLB_DEVICE_ATTR(submit_queues, uint, nullb_apply_submit_queues);
457NULLB_DEVICE_ATTR(poll_queues, uint, nullb_apply_poll_queues);
458NULLB_DEVICE_ATTR(home_node, uint, NULL);
459NULLB_DEVICE_ATTR(queue_mode, uint, NULL);
460NULLB_DEVICE_ATTR(blocksize, uint, NULL);
461NULLB_DEVICE_ATTR(max_sectors, uint, NULL);
462NULLB_DEVICE_ATTR(irqmode, uint, NULL);
463NULLB_DEVICE_ATTR(hw_queue_depth, uint, NULL);
464NULLB_DEVICE_ATTR(index, uint, NULL);
465NULLB_DEVICE_ATTR(blocking, bool, NULL);
466NULLB_DEVICE_ATTR(use_per_node_hctx, bool, NULL);
467NULLB_DEVICE_ATTR(memory_backed, bool, NULL);
468NULLB_DEVICE_ATTR(discard, bool, NULL);
469NULLB_DEVICE_ATTR(mbps, uint, NULL);
470NULLB_DEVICE_ATTR(cache_size, ulong, NULL);
471NULLB_DEVICE_ATTR(zoned, bool, NULL);
472NULLB_DEVICE_ATTR(zone_size, ulong, NULL);
473NULLB_DEVICE_ATTR(zone_capacity, ulong, NULL);
474NULLB_DEVICE_ATTR(zone_nr_conv, uint, NULL);
475NULLB_DEVICE_ATTR(zone_max_open, uint, NULL);
476NULLB_DEVICE_ATTR(zone_max_active, uint, NULL);
477NULLB_DEVICE_ATTR(zone_append_max_sectors, uint, NULL);
478NULLB_DEVICE_ATTR(zone_full, bool, NULL);
479NULLB_DEVICE_ATTR(virt_boundary, bool, NULL);
480NULLB_DEVICE_ATTR(no_sched, bool, NULL);
481NULLB_DEVICE_ATTR(shared_tags, bool, NULL);
482NULLB_DEVICE_ATTR(shared_tag_bitmap, bool, NULL);
483NULLB_DEVICE_ATTR(fua, bool, NULL);
484NULLB_DEVICE_ATTR(rotational, bool, NULL);
485NULLB_DEVICE_ATTR(badblocks_once, bool, NULL);
486NULLB_DEVICE_ATTR(badblocks_partial_io, bool, NULL);
487
488static ssize_t nullb_device_power_show(struct config_item *item, char *page)
489{
490 guard(mutex)(&lock);
491 return nullb_device_bool_attr_show(to_nullb_device(item)->power, page);
492}
493
494static ssize_t nullb_device_power_store(struct config_item *item,
495 const char *page, size_t count)
496{
497 struct nullb_device *dev = to_nullb_device(item);
498 bool newp = false;
499 ssize_t ret;
500
501 ret = nullb_device_bool_attr_store(&newp, page, count);
502 if (ret < 0)
503 return ret;
504
505 ret = count;
506 guard(mutex)(&lock);
507 if (!dev->power && newp) {
508 if (test_and_set_bit(NULLB_DEV_FL_UP, &dev->flags))
509 return ret;
510
511 ret = null_add_dev(dev);
512 if (ret) {
513 clear_bit(NULLB_DEV_FL_UP, &dev->flags);
514 return ret;
515 }
516
517 set_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
518 dev->power = newp;
519 ret = count;
520 } else if (dev->power && !newp) {
521 if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
522 dev->power = newp;
523 null_del_dev(dev->nullb);
524 }
525 clear_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
526 }
527
528 return ret;
529}
530
531CONFIGFS_ATTR(nullb_device_, power);
532
533static ssize_t nullb_device_badblocks_show(struct config_item *item, char *page)
534{
535 struct nullb_device *t_dev = to_nullb_device(item);
536
537 return badblocks_show(&t_dev->badblocks, page, 0);
538}
539
540static ssize_t nullb_device_badblocks_store(struct config_item *item,
541 const char *page, size_t count)
542{
543 struct nullb_device *t_dev = to_nullb_device(item);
544 char *orig, *buf, *tmp;
545 u64 start, end;
546 int ret;
547
548 orig = kstrndup(page, count, GFP_KERNEL);
549 if (!orig)
550 return -ENOMEM;
551
552 buf = strstrip(orig);
553
554 ret = -EINVAL;
555 if (buf[0] != '+' && buf[0] != '-')
556 goto out;
557 tmp = strchr(&buf[1], '-');
558 if (!tmp)
559 goto out;
560 *tmp = '\0';
561 ret = kstrtoull(buf + 1, 0, &start);
562 if (ret)
563 goto out;
564 ret = kstrtoull(tmp + 1, 0, &end);
565 if (ret)
566 goto out;
567 ret = -EINVAL;
568 if (start > end)
569 goto out;
570 /* enable badblocks */
571 cmpxchg(&t_dev->badblocks.shift, -1, 0);
572 if (buf[0] == '+') {
573 if (badblocks_set(&t_dev->badblocks, start,
574 end - start + 1, 1))
575 ret = count;
576 } else if (badblocks_clear(&t_dev->badblocks, start,
577 end - start + 1)) {
578 ret = count;
579 }
580out:
581 kfree(orig);
582 return ret;
583}
584CONFIGFS_ATTR(nullb_device_, badblocks);
585
586static ssize_t nullb_device_zone_readonly_store(struct config_item *item,
587 const char *page, size_t count)
588{
589 struct nullb_device *dev = to_nullb_device(item);
590
591 guard(mutex)(&lock);
592 return zone_cond_store(dev, page, count, BLK_ZONE_COND_READONLY);
593}
594CONFIGFS_ATTR_WO(nullb_device_, zone_readonly);
595
596static ssize_t nullb_device_zone_offline_store(struct config_item *item,
597 const char *page, size_t count)
598{
599 struct nullb_device *dev = to_nullb_device(item);
600
601 guard(mutex)(&lock);
602 return zone_cond_store(dev, page, count, BLK_ZONE_COND_OFFLINE);
603}
604CONFIGFS_ATTR_WO(nullb_device_, zone_offline);
605
606static struct configfs_attribute *nullb_device_attrs[] = {
607 &nullb_device_attr_badblocks,
608 &nullb_device_attr_badblocks_once,
609 &nullb_device_attr_badblocks_partial_io,
610 &nullb_device_attr_blocking,
611 &nullb_device_attr_blocksize,
612 &nullb_device_attr_cache_size,
613 &nullb_device_attr_completion_nsec,
614 &nullb_device_attr_discard,
615 &nullb_device_attr_fua,
616 &nullb_device_attr_home_node,
617 &nullb_device_attr_hw_queue_depth,
618 &nullb_device_attr_index,
619 &nullb_device_attr_irqmode,
620 &nullb_device_attr_max_sectors,
621 &nullb_device_attr_mbps,
622 &nullb_device_attr_memory_backed,
623 &nullb_device_attr_no_sched,
624 &nullb_device_attr_poll_queues,
625 &nullb_device_attr_power,
626 &nullb_device_attr_queue_mode,
627 &nullb_device_attr_rotational,
628 &nullb_device_attr_shared_tag_bitmap,
629 &nullb_device_attr_shared_tags,
630 &nullb_device_attr_size,
631 &nullb_device_attr_submit_queues,
632 &nullb_device_attr_use_per_node_hctx,
633 &nullb_device_attr_virt_boundary,
634 &nullb_device_attr_zone_append_max_sectors,
635 &nullb_device_attr_zone_capacity,
636 &nullb_device_attr_zone_full,
637 &nullb_device_attr_zone_max_active,
638 &nullb_device_attr_zone_max_open,
639 &nullb_device_attr_zone_nr_conv,
640 &nullb_device_attr_zone_offline,
641 &nullb_device_attr_zone_readonly,
642 &nullb_device_attr_zone_size,
643 &nullb_device_attr_zoned,
644 NULL,
645};
646
647static void nullb_device_release(struct config_item *item)
648{
649 struct nullb_device *dev = to_nullb_device(item);
650
651 null_free_device_storage(dev, false);
652 null_free_dev(dev);
653}
654
655static const struct configfs_item_operations nullb_device_ops = {
656 .release = nullb_device_release,
657};
658
659static const struct config_item_type nullb_device_type = {
660 .ct_item_ops = &nullb_device_ops,
661 .ct_attrs = nullb_device_attrs,
662 .ct_owner = THIS_MODULE,
663};
664
665#ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
666
667static void nullb_add_fault_config(struct nullb_device *dev)
668{
669 fault_config_init(&dev->timeout_config, "timeout_inject");
670 fault_config_init(&dev->requeue_config, "requeue_inject");
671 fault_config_init(&dev->init_hctx_fault_config, "init_hctx_fault_inject");
672
673 configfs_add_default_group(&dev->timeout_config.group, &dev->group);
674 configfs_add_default_group(&dev->requeue_config.group, &dev->group);
675 configfs_add_default_group(&dev->init_hctx_fault_config.group, &dev->group);
676}
677
678static void nullb_del_fault_config(struct nullb_device *dev)
679{
680 config_item_put(&dev->init_hctx_fault_config.group.cg_item);
681 config_item_put(&dev->requeue_config.group.cg_item);
682 config_item_put(&dev->timeout_config.group.cg_item);
683}
684
685#else
686
687static void nullb_add_fault_config(struct nullb_device *dev)
688{
689}
690
691static void nullb_del_fault_config(struct nullb_device *dev)
692{
693}
694#endif
695
696static struct
697config_group *nullb_group_make_group(struct config_group *group, const char *name)
698{
699 struct nullb_device *dev;
700
701 if (null_find_dev_by_name(name))
702 return ERR_PTR(-EEXIST);
703
704 dev = null_alloc_dev();
705 if (!dev)
706 return ERR_PTR(-ENOMEM);
707
708 config_group_init_type_name(&dev->group, name, &nullb_device_type);
709 nullb_add_fault_config(dev);
710
711 return &dev->group;
712}
713
714static void
715nullb_group_drop_item(struct config_group *group, struct config_item *item)
716{
717 struct nullb_device *dev = to_nullb_device(item);
718
719 if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
720 guard(mutex)(&lock);
721 dev->power = false;
722 null_del_dev(dev->nullb);
723 }
724 nullb_del_fault_config(dev);
725 config_item_put(item);
726}
727
728static ssize_t memb_group_features_show(struct config_item *item, char *page)
729{
730
731 struct configfs_attribute **entry;
732 char delimiter = ',';
733 size_t left = PAGE_SIZE;
734 size_t written = 0;
735 int ret;
736
737 for (entry = &nullb_device_attrs[0]; *entry && left > 0; entry++) {
738 if (!*(entry + 1))
739 delimiter = '\n';
740 ret = snprintf(page + written, left, "%s%c", (*entry)->ca_name,
741 delimiter);
742 if (ret >= left) {
743 WARN_ONCE(1, "Too many null_blk features to print\n");
744 memzero_explicit(page, PAGE_SIZE);
745 return -ENOBUFS;
746 }
747 left -= ret;
748 written += ret;
749 }
750
751 return written;
752}
753
754CONFIGFS_ATTR_RO(memb_group_, features);
755
756static struct configfs_attribute *nullb_group_attrs[] = {
757 &memb_group_attr_features,
758 NULL,
759};
760
761static const struct configfs_group_operations nullb_group_ops = {
762 .make_group = nullb_group_make_group,
763 .drop_item = nullb_group_drop_item,
764};
765
766static const struct config_item_type nullb_group_type = {
767 .ct_group_ops = &nullb_group_ops,
768 .ct_attrs = nullb_group_attrs,
769 .ct_owner = THIS_MODULE,
770};
771
772static struct configfs_subsystem nullb_subsys = {
773 .su_group = {
774 .cg_item = {
775 .ci_namebuf = "nullb",
776 .ci_type = &nullb_group_type,
777 },
778 },
779};
780
781static inline int null_cache_active(struct nullb *nullb)
782{
783 return test_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
784}
785
786static struct nullb_device *null_alloc_dev(void)
787{
788 struct nullb_device *dev;
789
790 dev = kzalloc_obj(*dev);
791 if (!dev)
792 return NULL;
793
794#ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
795 dev->timeout_config.attr = null_timeout_attr;
796 dev->requeue_config.attr = null_requeue_attr;
797 dev->init_hctx_fault_config.attr = null_init_hctx_attr;
798#endif
799
800 INIT_RADIX_TREE(&dev->data, GFP_ATOMIC);
801 INIT_RADIX_TREE(&dev->cache, GFP_ATOMIC);
802 if (badblocks_init(&dev->badblocks, 0)) {
803 kfree(dev);
804 return NULL;
805 }
806
807 dev->size = g_gb * 1024;
808 dev->completion_nsec = g_completion_nsec;
809 dev->submit_queues = g_submit_queues;
810 dev->prev_submit_queues = g_submit_queues;
811 dev->poll_queues = g_poll_queues;
812 dev->prev_poll_queues = g_poll_queues;
813 dev->home_node = g_home_node;
814 dev->queue_mode = g_queue_mode;
815 dev->blocksize = g_bs;
816 dev->max_sectors = g_max_sectors;
817 dev->irqmode = g_irqmode;
818 dev->hw_queue_depth = g_hw_queue_depth;
819 dev->blocking = g_blocking;
820 dev->memory_backed = g_memory_backed;
821 dev->discard = g_discard;
822 dev->cache_size = g_cache_size;
823 dev->mbps = g_mbps;
824 dev->use_per_node_hctx = g_use_per_node_hctx;
825 dev->zoned = g_zoned;
826 dev->zone_size = g_zone_size;
827 dev->zone_capacity = g_zone_capacity;
828 dev->zone_nr_conv = g_zone_nr_conv;
829 dev->zone_max_open = g_zone_max_open;
830 dev->zone_max_active = g_zone_max_active;
831 dev->zone_append_max_sectors = g_zone_append_max_sectors;
832 dev->zone_full = g_zone_full;
833 dev->virt_boundary = g_virt_boundary;
834 dev->no_sched = g_no_sched;
835 dev->shared_tags = g_shared_tags;
836 dev->shared_tag_bitmap = g_shared_tag_bitmap;
837 dev->fua = g_fua;
838 dev->rotational = g_rotational;
839
840 return dev;
841}
842
843static void null_free_dev(struct nullb_device *dev)
844{
845 if (!dev)
846 return;
847
848 badblocks_exit(&dev->badblocks);
849 kfree(dev);
850}
851
852static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
853{
854 struct nullb_cmd *cmd = container_of(timer, struct nullb_cmd, timer);
855
856 blk_mq_end_request(blk_mq_rq_from_pdu(cmd), cmd->error);
857 return HRTIMER_NORESTART;
858}
859
860static void null_cmd_end_timer(struct nullb_cmd *cmd)
861{
862 ktime_t kt = cmd->nq->dev->completion_nsec;
863
864 hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL);
865}
866
867static void null_complete_rq(struct request *rq)
868{
869 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq);
870
871 blk_mq_end_request(rq, cmd->error);
872}
873
874static struct nullb_page *null_alloc_page(void)
875{
876 struct nullb_page *t_page;
877
878 t_page = kmalloc_obj(struct nullb_page, GFP_NOIO);
879 if (!t_page)
880 return NULL;
881
882 t_page->page = alloc_pages(GFP_NOIO, 0);
883 if (!t_page->page) {
884 kfree(t_page);
885 return NULL;
886 }
887
888 memset(t_page->bitmap, 0, sizeof(t_page->bitmap));
889 return t_page;
890}
891
892static void null_free_page(struct nullb_page *t_page)
893{
894 __set_bit(NULLB_PAGE_FREE, t_page->bitmap);
895 if (test_bit(NULLB_PAGE_LOCK, t_page->bitmap))
896 return;
897 __free_page(t_page->page);
898 kfree(t_page);
899}
900
901static bool null_page_empty(struct nullb_page *page)
902{
903 int size = MAP_SZ - 2;
904
905 return find_first_bit(page->bitmap, size) == size;
906}
907
908static void null_free_sector(struct nullb *nullb, sector_t sector,
909 bool is_cache)
910{
911 unsigned int sector_bit;
912 u64 idx;
913 struct nullb_page *t_page, *ret;
914 struct radix_tree_root *root;
915
916 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
917 idx = sector >> PAGE_SECTORS_SHIFT;
918 sector_bit = (sector & SECTOR_MASK);
919
920 t_page = radix_tree_lookup(root, idx);
921 if (t_page) {
922 __clear_bit(sector_bit, t_page->bitmap);
923
924 if (null_page_empty(t_page)) {
925 ret = radix_tree_delete_item(root, idx, t_page);
926 WARN_ON(ret != t_page);
927 null_free_page(ret);
928 if (is_cache)
929 nullb->dev->curr_cache -= PAGE_SIZE;
930 }
931 }
932}
933
934static struct nullb_page *null_radix_tree_insert(struct nullb *nullb, u64 idx,
935 struct nullb_page *t_page, bool is_cache)
936{
937 struct radix_tree_root *root;
938
939 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
940
941 if (radix_tree_insert(root, idx, t_page)) {
942 null_free_page(t_page);
943 t_page = radix_tree_lookup(root, idx);
944 WARN_ON(!t_page || t_page->page->private != idx);
945 } else if (is_cache)
946 nullb->dev->curr_cache += PAGE_SIZE;
947
948 return t_page;
949}
950
951static void null_free_device_storage(struct nullb_device *dev, bool is_cache)
952{
953 unsigned long pos = 0;
954 int nr_pages;
955 struct nullb_page *ret, *t_pages[FREE_BATCH];
956 struct radix_tree_root *root;
957
958 root = is_cache ? &dev->cache : &dev->data;
959
960 do {
961 int i;
962
963 nr_pages = radix_tree_gang_lookup(root,
964 (void **)t_pages, pos, FREE_BATCH);
965
966 for (i = 0; i < nr_pages; i++) {
967 pos = t_pages[i]->page->private;
968 ret = radix_tree_delete_item(root, pos, t_pages[i]);
969 WARN_ON(ret != t_pages[i]);
970 null_free_page(ret);
971 }
972
973 pos++;
974 } while (nr_pages == FREE_BATCH);
975
976 if (is_cache)
977 dev->curr_cache = 0;
978}
979
980static struct nullb_page *__null_lookup_page(struct nullb *nullb,
981 sector_t sector, bool for_write, bool is_cache)
982{
983 unsigned int sector_bit;
984 u64 idx;
985 struct nullb_page *t_page;
986 struct radix_tree_root *root;
987
988 idx = sector >> PAGE_SECTORS_SHIFT;
989 sector_bit = (sector & SECTOR_MASK);
990
991 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
992 t_page = radix_tree_lookup(root, idx);
993 WARN_ON(t_page && t_page->page->private != idx);
994
995 if (t_page && (for_write || test_bit(sector_bit, t_page->bitmap)))
996 return t_page;
997
998 return NULL;
999}
1000
1001static struct nullb_page *null_lookup_page(struct nullb *nullb,
1002 sector_t sector, bool for_write, bool ignore_cache)
1003{
1004 struct nullb_page *page = NULL;
1005
1006 if (!ignore_cache)
1007 page = __null_lookup_page(nullb, sector, for_write, true);
1008 if (page)
1009 return page;
1010 return __null_lookup_page(nullb, sector, for_write, false);
1011}
1012
1013static struct nullb_page *null_insert_page(struct nullb *nullb,
1014 sector_t sector, bool ignore_cache)
1015 __releases(&nullb->lock)
1016 __acquires(&nullb->lock)
1017{
1018 u64 idx;
1019 struct nullb_page *t_page;
1020
1021 t_page = null_lookup_page(nullb, sector, true, ignore_cache);
1022 if (t_page)
1023 return t_page;
1024
1025 spin_unlock_irq(&nullb->lock);
1026
1027 t_page = null_alloc_page();
1028 if (!t_page)
1029 goto out_lock;
1030
1031 if (radix_tree_preload(GFP_NOIO))
1032 goto out_freepage;
1033
1034 spin_lock_irq(&nullb->lock);
1035 idx = sector >> PAGE_SECTORS_SHIFT;
1036 t_page->page->private = idx;
1037 t_page = null_radix_tree_insert(nullb, idx, t_page, !ignore_cache);
1038 radix_tree_preload_end();
1039
1040 return t_page;
1041out_freepage:
1042 null_free_page(t_page);
1043out_lock:
1044 spin_lock_irq(&nullb->lock);
1045 return null_lookup_page(nullb, sector, true, ignore_cache);
1046}
1047
1048static int null_flush_cache_page(struct nullb *nullb, struct nullb_page *c_page)
1049{
1050 int i;
1051 unsigned int offset;
1052 u64 idx;
1053 struct nullb_page *t_page, *ret;
1054 void *dst, *src;
1055
1056 idx = c_page->page->private;
1057
1058 t_page = null_insert_page(nullb, idx << PAGE_SECTORS_SHIFT, true);
1059
1060 __clear_bit(NULLB_PAGE_LOCK, c_page->bitmap);
1061 if (test_bit(NULLB_PAGE_FREE, c_page->bitmap)) {
1062 null_free_page(c_page);
1063 if (t_page && null_page_empty(t_page)) {
1064 ret = radix_tree_delete_item(&nullb->dev->data,
1065 idx, t_page);
1066 null_free_page(t_page);
1067 }
1068 return 0;
1069 }
1070
1071 if (!t_page)
1072 return -ENOMEM;
1073
1074 src = kmap_local_page(c_page->page);
1075 dst = kmap_local_page(t_page->page);
1076
1077 for (i = 0; i < PAGE_SECTORS;
1078 i += (nullb->dev->blocksize >> SECTOR_SHIFT)) {
1079 if (test_bit(i, c_page->bitmap)) {
1080 offset = (i << SECTOR_SHIFT);
1081 memcpy(dst + offset, src + offset,
1082 nullb->dev->blocksize);
1083 __set_bit(i, t_page->bitmap);
1084 }
1085 }
1086
1087 kunmap_local(dst);
1088 kunmap_local(src);
1089
1090 ret = radix_tree_delete_item(&nullb->dev->cache, idx, c_page);
1091 null_free_page(ret);
1092 nullb->dev->curr_cache -= PAGE_SIZE;
1093
1094 return 0;
1095}
1096
1097static int null_make_cache_space(struct nullb *nullb, unsigned long n)
1098{
1099 int i, err, nr_pages;
1100 struct nullb_page *c_pages[FREE_BATCH];
1101 unsigned long flushed = 0, one_round;
1102
1103again:
1104 if ((nullb->dev->cache_size * 1024 * 1024) >
1105 nullb->dev->curr_cache + n || nullb->dev->curr_cache == 0)
1106 return 0;
1107
1108 nr_pages = radix_tree_gang_lookup(&nullb->dev->cache,
1109 (void **)c_pages, nullb->cache_flush_pos, FREE_BATCH);
1110 /*
1111 * nullb_flush_cache_page could unlock before using the c_pages. To
1112 * avoid race, we don't allow page free
1113 */
1114 for (i = 0; i < nr_pages; i++) {
1115 nullb->cache_flush_pos = c_pages[i]->page->private;
1116 /*
1117 * We found the page which is being flushed to disk by other
1118 * threads
1119 */
1120 if (test_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap))
1121 c_pages[i] = NULL;
1122 else
1123 __set_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap);
1124 }
1125
1126 one_round = 0;
1127 for (i = 0; i < nr_pages; i++) {
1128 if (c_pages[i] == NULL)
1129 continue;
1130 err = null_flush_cache_page(nullb, c_pages[i]);
1131 if (err)
1132 return err;
1133 one_round++;
1134 }
1135 flushed += one_round << PAGE_SHIFT;
1136
1137 if (n > flushed) {
1138 if (nr_pages == 0)
1139 nullb->cache_flush_pos = 0;
1140 if (one_round == 0) {
1141 /* give other threads a chance */
1142 spin_unlock_irq(&nullb->lock);
1143 spin_lock_irq(&nullb->lock);
1144 }
1145 goto again;
1146 }
1147 return 0;
1148}
1149
1150static blk_status_t copy_to_nullb(struct nullb *nullb, void *source,
1151 loff_t pos, size_t n, bool is_fua)
1152{
1153 size_t temp, count = 0;
1154 struct nullb_page *t_page;
1155 sector_t sector;
1156
1157 while (count < n) {
1158 temp = min3(nullb->dev->blocksize, n - count,
1159 PAGE_SIZE - offset_in_page(pos));
1160 sector = pos >> SECTOR_SHIFT;
1161
1162 if (null_cache_active(nullb) && !is_fua)
1163 null_make_cache_space(nullb, PAGE_SIZE);
1164
1165 t_page = null_insert_page(nullb, sector,
1166 !null_cache_active(nullb) || is_fua);
1167 if (!t_page)
1168 return BLK_STS_NOSPC;
1169
1170 memcpy_to_page(t_page->page, offset_in_page(pos),
1171 source + count, temp);
1172
1173 __set_bit(sector & SECTOR_MASK, t_page->bitmap);
1174
1175 if (is_fua)
1176 null_free_sector(nullb, sector, true);
1177
1178 count += temp;
1179 pos += temp;
1180 }
1181 return BLK_STS_OK;
1182}
1183
1184static void copy_from_nullb(struct nullb *nullb, void *dest, loff_t pos,
1185 size_t n)
1186{
1187 size_t temp, count = 0;
1188 struct nullb_page *t_page;
1189 sector_t sector;
1190
1191 while (count < n) {
1192 temp = min3(nullb->dev->blocksize, n - count,
1193 PAGE_SIZE - offset_in_page(pos));
1194 sector = pos >> SECTOR_SHIFT;
1195
1196 t_page = null_lookup_page(nullb, sector, false,
1197 !null_cache_active(nullb));
1198 if (t_page)
1199 memcpy_from_page(dest + count, t_page->page,
1200 offset_in_page(pos), temp);
1201 else
1202 memset(dest + count, 0, temp);
1203
1204 count += temp;
1205 pos += temp;
1206 }
1207}
1208
1209blk_status_t null_handle_discard(struct nullb_device *dev,
1210 sector_t sector, sector_t nr_sectors)
1211{
1212 struct nullb *nullb = dev->nullb;
1213 size_t n = nr_sectors << SECTOR_SHIFT;
1214 size_t temp;
1215
1216 spin_lock_irq(&nullb->lock);
1217 while (n > 0) {
1218 temp = min_t(size_t, n, dev->blocksize);
1219 null_free_sector(nullb, sector, false);
1220 if (null_cache_active(nullb))
1221 null_free_sector(nullb, sector, true);
1222 sector += temp >> SECTOR_SHIFT;
1223 n -= temp;
1224 }
1225 spin_unlock_irq(&nullb->lock);
1226
1227 return BLK_STS_OK;
1228}
1229
1230static blk_status_t null_handle_flush(struct nullb *nullb)
1231{
1232 int err;
1233
1234 if (!null_cache_active(nullb))
1235 return 0;
1236
1237 spin_lock_irq(&nullb->lock);
1238 while (true) {
1239 err = null_make_cache_space(nullb,
1240 nullb->dev->cache_size * 1024 * 1024);
1241 if (err || nullb->dev->curr_cache == 0)
1242 break;
1243 }
1244
1245 WARN_ON(!radix_tree_empty(&nullb->dev->cache));
1246 spin_unlock_irq(&nullb->lock);
1247 return errno_to_blk_status(err);
1248}
1249
1250static blk_status_t null_transfer(struct nullb *nullb, struct page *page,
1251 unsigned int len, unsigned int off, bool is_write, loff_t pos,
1252 bool is_fua)
1253{
1254 struct nullb_device *dev = nullb->dev;
1255 blk_status_t err = BLK_STS_OK;
1256 unsigned int valid_len = len;
1257 void *p;
1258
1259 p = kmap_local_page(page) + off;
1260 if (!is_write) {
1261 if (dev->zoned) {
1262 valid_len = null_zone_valid_read_len(nullb,
1263 pos >> SECTOR_SHIFT, len);
1264 if (valid_len && valid_len != len)
1265 valid_len -= pos & (SECTOR_SIZE - 1);
1266 }
1267
1268 if (valid_len) {
1269 copy_from_nullb(nullb, p, pos, valid_len);
1270 off += valid_len;
1271 len -= valid_len;
1272 }
1273
1274 if (len)
1275 memset(p + valid_len, 0xff, len);
1276 flush_dcache_page(page);
1277 } else {
1278 flush_dcache_page(page);
1279 err = copy_to_nullb(nullb, p, pos, len, is_fua);
1280 }
1281
1282 kunmap_local(p);
1283 return err;
1284}
1285
1286/*
1287 * Transfer data for the given request. The transfer size is capped with the
1288 * nr_sectors argument.
1289 */
1290static blk_status_t null_handle_data_transfer(struct nullb_cmd *cmd,
1291 sector_t nr_sectors)
1292{
1293 struct request *rq = blk_mq_rq_from_pdu(cmd);
1294 struct nullb *nullb = cmd->nq->dev->nullb;
1295 blk_status_t err = BLK_STS_OK;
1296 unsigned int len;
1297 loff_t pos = blk_rq_pos(rq) << SECTOR_SHIFT;
1298 unsigned int max_bytes = nr_sectors << SECTOR_SHIFT;
1299 unsigned int transferred_bytes = 0;
1300 struct req_iterator iter;
1301 struct bio_vec bvec;
1302
1303 spin_lock_irq(&nullb->lock);
1304 rq_for_each_segment(bvec, rq, iter) {
1305 len = bvec.bv_len;
1306 if (transferred_bytes + len > max_bytes)
1307 len = max_bytes - transferred_bytes;
1308 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1309 op_is_write(req_op(rq)), pos,
1310 rq->cmd_flags & REQ_FUA);
1311 if (err)
1312 break;
1313 pos += len;
1314 transferred_bytes += len;
1315 if (transferred_bytes >= max_bytes)
1316 break;
1317 }
1318 spin_unlock_irq(&nullb->lock);
1319
1320 return err;
1321}
1322
1323static inline blk_status_t null_handle_throttled(struct nullb_cmd *cmd)
1324{
1325 struct nullb_device *dev = cmd->nq->dev;
1326 struct nullb *nullb = dev->nullb;
1327 blk_status_t sts = BLK_STS_OK;
1328 struct request *rq = blk_mq_rq_from_pdu(cmd);
1329
1330 if (!hrtimer_active(&nullb->bw_timer))
1331 hrtimer_restart(&nullb->bw_timer);
1332
1333 if (atomic_long_sub_return(blk_rq_bytes(rq), &nullb->cur_bytes) < 0) {
1334 blk_mq_stop_hw_queues(nullb->q);
1335 /* race with timer */
1336 if (atomic_long_read(&nullb->cur_bytes) > 0)
1337 blk_mq_start_stopped_hw_queues(nullb->q, true);
1338 /* requeue request */
1339 sts = BLK_STS_DEV_RESOURCE;
1340 }
1341 return sts;
1342}
1343
1344/*
1345 * Check if the command should fail for the badblocks. If so, return
1346 * BLK_STS_IOERR and return number of partial I/O sectors to be written or read,
1347 * which may be less than the requested number of sectors.
1348 *
1349 * @cmd: The command to handle.
1350 * @sector: The start sector for I/O.
1351 * @nr_sectors: Specifies number of sectors to write or read, and returns the
1352 * number of sectors to be written or read.
1353 */
1354blk_status_t null_handle_badblocks(struct nullb_cmd *cmd, sector_t sector,
1355 unsigned int *nr_sectors)
1356{
1357 struct badblocks *bb = &cmd->nq->dev->badblocks;
1358 struct nullb_device *dev = cmd->nq->dev;
1359 unsigned int block_sectors = dev->blocksize >> SECTOR_SHIFT;
1360 sector_t first_bad, bad_sectors;
1361 unsigned int partial_io_sectors = 0;
1362
1363 if (!badblocks_check(bb, sector, *nr_sectors, &first_bad, &bad_sectors))
1364 return BLK_STS_OK;
1365
1366 if (cmd->nq->dev->badblocks_once)
1367 badblocks_clear(bb, first_bad, bad_sectors);
1368
1369 if (cmd->nq->dev->badblocks_partial_io) {
1370 if (!IS_ALIGNED(first_bad, block_sectors))
1371 first_bad = ALIGN_DOWN(first_bad, block_sectors);
1372 if (sector < first_bad)
1373 partial_io_sectors = first_bad - sector;
1374 }
1375 *nr_sectors = partial_io_sectors;
1376
1377 return BLK_STS_IOERR;
1378}
1379
1380blk_status_t null_handle_memory_backed(struct nullb_cmd *cmd, enum req_op op,
1381 sector_t sector, sector_t nr_sectors)
1382{
1383 struct nullb_device *dev = cmd->nq->dev;
1384
1385 if (op == REQ_OP_DISCARD)
1386 return null_handle_discard(dev, sector, nr_sectors);
1387
1388 return null_handle_data_transfer(cmd, nr_sectors);
1389}
1390
1391static void nullb_zero_read_cmd_buffer(struct nullb_cmd *cmd)
1392{
1393 struct request *rq = blk_mq_rq_from_pdu(cmd);
1394 struct nullb_device *dev = cmd->nq->dev;
1395 struct bio *bio;
1396
1397 if (!dev->memory_backed && req_op(rq) == REQ_OP_READ) {
1398 __rq_for_each_bio(bio, rq)
1399 zero_fill_bio(bio);
1400 }
1401}
1402
1403static inline void nullb_complete_cmd(struct nullb_cmd *cmd)
1404{
1405 struct request *rq = blk_mq_rq_from_pdu(cmd);
1406
1407 /*
1408 * Since root privileges are required to configure the null_blk
1409 * driver, it is fine that this driver does not initialize the
1410 * data buffers of read commands. Zero-initialize these buffers
1411 * anyway if KMSAN is enabled to prevent that KMSAN complains
1412 * about null_blk not initializing read data buffers.
1413 */
1414 if (IS_ENABLED(CONFIG_KMSAN))
1415 nullb_zero_read_cmd_buffer(cmd);
1416
1417 /* Complete IO by inline, softirq or timer */
1418 switch (cmd->nq->dev->irqmode) {
1419 case NULL_IRQ_SOFTIRQ:
1420 blk_mq_complete_request(rq);
1421 break;
1422 case NULL_IRQ_NONE:
1423 blk_mq_end_request(rq, cmd->error);
1424 break;
1425 case NULL_IRQ_TIMER:
1426 null_cmd_end_timer(cmd);
1427 break;
1428 }
1429}
1430
1431blk_status_t null_process_cmd(struct nullb_cmd *cmd, enum req_op op,
1432 sector_t sector, unsigned int nr_sectors)
1433{
1434 struct nullb_device *dev = cmd->nq->dev;
1435 blk_status_t badblocks_ret = BLK_STS_OK;
1436 blk_status_t ret;
1437
1438 if (dev->badblocks.shift != -1)
1439 badblocks_ret = null_handle_badblocks(cmd, sector, &nr_sectors);
1440
1441 if (dev->memory_backed && nr_sectors) {
1442 ret = null_handle_memory_backed(cmd, op, sector, nr_sectors);
1443 if (ret != BLK_STS_OK)
1444 return ret;
1445 }
1446
1447 return badblocks_ret;
1448}
1449
1450static void null_handle_cmd(struct nullb_cmd *cmd, sector_t sector,
1451 sector_t nr_sectors, enum req_op op)
1452{
1453 struct nullb_device *dev = cmd->nq->dev;
1454 struct nullb *nullb = dev->nullb;
1455 blk_status_t sts;
1456
1457 if (op == REQ_OP_FLUSH) {
1458 cmd->error = null_handle_flush(nullb);
1459 goto out;
1460 }
1461
1462 if (dev->zoned)
1463 sts = null_process_zoned_cmd(cmd, op, sector, nr_sectors);
1464 else
1465 sts = null_process_cmd(cmd, op, sector, nr_sectors);
1466
1467 /* Do not overwrite errors (e.g. timeout errors) */
1468 if (cmd->error == BLK_STS_OK)
1469 cmd->error = sts;
1470
1471out:
1472 nullb_complete_cmd(cmd);
1473}
1474
1475static enum hrtimer_restart nullb_bwtimer_fn(struct hrtimer *timer)
1476{
1477 struct nullb *nullb = container_of(timer, struct nullb, bw_timer);
1478 ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1479 unsigned int mbps = nullb->dev->mbps;
1480
1481 if (atomic_long_read(&nullb->cur_bytes) == mb_per_tick(mbps))
1482 return HRTIMER_NORESTART;
1483
1484 atomic_long_set(&nullb->cur_bytes, mb_per_tick(mbps));
1485 blk_mq_start_stopped_hw_queues(nullb->q, true);
1486
1487 hrtimer_forward_now(&nullb->bw_timer, timer_interval);
1488
1489 return HRTIMER_RESTART;
1490}
1491
1492static void nullb_setup_bwtimer(struct nullb *nullb)
1493{
1494 ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1495
1496 hrtimer_setup(&nullb->bw_timer, nullb_bwtimer_fn, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1497 atomic_long_set(&nullb->cur_bytes, mb_per_tick(nullb->dev->mbps));
1498 hrtimer_start(&nullb->bw_timer, timer_interval, HRTIMER_MODE_REL);
1499}
1500
1501#ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1502
1503static bool should_timeout_request(struct request *rq)
1504{
1505 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq);
1506 struct nullb_device *dev = cmd->nq->dev;
1507
1508 return should_fail(&dev->timeout_config.attr, 1);
1509}
1510
1511static bool should_requeue_request(struct request *rq)
1512{
1513 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq);
1514 struct nullb_device *dev = cmd->nq->dev;
1515
1516 return should_fail(&dev->requeue_config.attr, 1);
1517}
1518
1519static bool should_init_hctx_fail(struct nullb_device *dev)
1520{
1521 return should_fail(&dev->init_hctx_fault_config.attr, 1);
1522}
1523
1524#else
1525
1526static bool should_timeout_request(struct request *rq)
1527{
1528 return false;
1529}
1530
1531static bool should_requeue_request(struct request *rq)
1532{
1533 return false;
1534}
1535
1536static bool should_init_hctx_fail(struct nullb_device *dev)
1537{
1538 return false;
1539}
1540
1541#endif
1542
1543static void null_map_queues(struct blk_mq_tag_set *set)
1544{
1545 struct nullb *nullb = set->driver_data;
1546 int i, qoff;
1547 unsigned int submit_queues = g_submit_queues;
1548 unsigned int poll_queues = g_poll_queues;
1549
1550 if (nullb) {
1551 struct nullb_device *dev = nullb->dev;
1552
1553 /*
1554 * Refer nr_hw_queues of the tag set to check if the expected
1555 * number of hardware queues are prepared. If block layer failed
1556 * to prepare them, use previous numbers of submit queues and
1557 * poll queues to map queues.
1558 */
1559 if (set->nr_hw_queues ==
1560 dev->submit_queues + dev->poll_queues) {
1561 submit_queues = dev->submit_queues;
1562 poll_queues = dev->poll_queues;
1563 } else if (set->nr_hw_queues ==
1564 dev->prev_submit_queues + dev->prev_poll_queues) {
1565 submit_queues = dev->prev_submit_queues;
1566 poll_queues = dev->prev_poll_queues;
1567 } else {
1568 pr_warn("tag set has unexpected nr_hw_queues: %d\n",
1569 set->nr_hw_queues);
1570 WARN_ON_ONCE(true);
1571 submit_queues = 1;
1572 poll_queues = 0;
1573 }
1574 }
1575
1576 for (i = 0, qoff = 0; i < set->nr_maps; i++) {
1577 struct blk_mq_queue_map *map = &set->map[i];
1578
1579 switch (i) {
1580 case HCTX_TYPE_DEFAULT:
1581 map->nr_queues = submit_queues;
1582 break;
1583 case HCTX_TYPE_READ:
1584 map->nr_queues = 0;
1585 continue;
1586 case HCTX_TYPE_POLL:
1587 map->nr_queues = poll_queues;
1588 break;
1589 }
1590 map->queue_offset = qoff;
1591 qoff += map->nr_queues;
1592 blk_mq_map_queues(map);
1593 }
1594}
1595
1596static int null_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1597{
1598 struct nullb_queue *nq = hctx->driver_data;
1599 LIST_HEAD(list);
1600 int nr = 0;
1601 struct request *rq;
1602
1603 spin_lock(&nq->poll_lock);
1604 list_splice_init(&nq->poll_list, &list);
1605 list_for_each_entry(rq, &list, queuelist)
1606 blk_mq_set_request_complete(rq);
1607 spin_unlock(&nq->poll_lock);
1608
1609 while (!list_empty(&list)) {
1610 struct nullb_cmd *cmd;
1611 struct request *req;
1612
1613 req = list_first_entry(&list, struct request, queuelist);
1614 list_del_init(&req->queuelist);
1615 cmd = blk_mq_rq_to_pdu(req);
1616 cmd->error = null_process_cmd(cmd, req_op(req), blk_rq_pos(req),
1617 blk_rq_sectors(req));
1618 if (!blk_mq_add_to_batch(req, iob, cmd->error != BLK_STS_OK,
1619 blk_mq_end_request_batch))
1620 blk_mq_end_request(req, cmd->error);
1621 nr++;
1622 }
1623
1624 return nr;
1625}
1626
1627static enum blk_eh_timer_return null_timeout_rq(struct request *rq)
1628{
1629 struct blk_mq_hw_ctx *hctx = rq->mq_hctx;
1630 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq);
1631
1632 if (hctx->type == HCTX_TYPE_POLL) {
1633 struct nullb_queue *nq = hctx->driver_data;
1634
1635 spin_lock(&nq->poll_lock);
1636 /* The request may have completed meanwhile. */
1637 if (blk_mq_request_completed(rq)) {
1638 spin_unlock(&nq->poll_lock);
1639 return BLK_EH_DONE;
1640 }
1641 list_del_init(&rq->queuelist);
1642 spin_unlock(&nq->poll_lock);
1643 }
1644
1645 pr_info("rq %p timed out\n", rq);
1646
1647 /*
1648 * If the device is marked as blocking (i.e. memory backed or zoned
1649 * device), the submission path may be blocked waiting for resources
1650 * and cause real timeouts. For these real timeouts, the submission
1651 * path will complete the request using blk_mq_complete_request().
1652 * Only fake timeouts need to execute blk_mq_complete_request() here.
1653 */
1654 cmd->error = BLK_STS_TIMEOUT;
1655 if (cmd->fake_timeout || hctx->type == HCTX_TYPE_POLL)
1656 blk_mq_complete_request(rq);
1657 return BLK_EH_DONE;
1658}
1659
1660static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx,
1661 const struct blk_mq_queue_data *bd)
1662{
1663 struct request *rq = bd->rq;
1664 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq);
1665 struct nullb_queue *nq = hctx->driver_data;
1666 sector_t nr_sectors = blk_rq_sectors(rq);
1667 sector_t sector = blk_rq_pos(rq);
1668 const bool is_poll = hctx->type == HCTX_TYPE_POLL;
1669
1670 might_sleep_if(hctx->flags & BLK_MQ_F_BLOCKING);
1671
1672 if (!is_poll && nq->dev->irqmode == NULL_IRQ_TIMER) {
1673 hrtimer_setup(&cmd->timer, null_cmd_timer_expired, CLOCK_MONOTONIC,
1674 HRTIMER_MODE_REL);
1675 }
1676 cmd->error = BLK_STS_OK;
1677 cmd->nq = nq;
1678 cmd->fake_timeout = should_timeout_request(rq) ||
1679 blk_should_fake_timeout(rq->q);
1680
1681 if (should_requeue_request(rq)) {
1682 /*
1683 * Alternate between hitting the core BUSY path, and the
1684 * driver driven requeue path
1685 */
1686 nq->requeue_selection++;
1687 if (nq->requeue_selection & 1)
1688 return BLK_STS_RESOURCE;
1689 blk_mq_requeue_request(rq, true);
1690 return BLK_STS_OK;
1691 }
1692
1693 if (test_bit(NULLB_DEV_FL_THROTTLED, &nq->dev->flags)) {
1694 blk_status_t sts = null_handle_throttled(cmd);
1695
1696 if (sts != BLK_STS_OK)
1697 return sts;
1698 }
1699
1700 blk_mq_start_request(rq);
1701
1702 if (is_poll) {
1703 spin_lock(&nq->poll_lock);
1704 list_add_tail(&rq->queuelist, &nq->poll_list);
1705 spin_unlock(&nq->poll_lock);
1706 return BLK_STS_OK;
1707 }
1708 if (cmd->fake_timeout)
1709 return BLK_STS_OK;
1710
1711 null_handle_cmd(cmd, sector, nr_sectors, req_op(rq));
1712 return BLK_STS_OK;
1713}
1714
1715static void null_queue_rqs(struct rq_list *rqlist)
1716{
1717 struct rq_list requeue_list = {};
1718 struct blk_mq_queue_data bd = { };
1719 blk_status_t ret;
1720
1721 do {
1722 struct request *rq = rq_list_pop(rqlist);
1723
1724 bd.rq = rq;
1725 ret = null_queue_rq(rq->mq_hctx, &bd);
1726 if (ret != BLK_STS_OK)
1727 rq_list_add_tail(&requeue_list, rq);
1728 } while (!rq_list_empty(rqlist));
1729
1730 *rqlist = requeue_list;
1731}
1732
1733static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq)
1734{
1735 nq->dev = nullb->dev;
1736 INIT_LIST_HEAD(&nq->poll_list);
1737 spin_lock_init(&nq->poll_lock);
1738}
1739
1740static int null_init_hctx(struct blk_mq_hw_ctx *hctx, void *driver_data,
1741 unsigned int hctx_idx)
1742{
1743 struct nullb *nullb = hctx->queue->queuedata;
1744 struct nullb_queue *nq;
1745
1746 if (should_init_hctx_fail(nullb->dev))
1747 return -EFAULT;
1748
1749 nq = &nullb->queues[hctx_idx];
1750 hctx->driver_data = nq;
1751 null_init_queue(nullb, nq);
1752
1753 return 0;
1754}
1755
1756static const struct blk_mq_ops null_mq_ops = {
1757 .queue_rq = null_queue_rq,
1758 .queue_rqs = null_queue_rqs,
1759 .complete = null_complete_rq,
1760 .timeout = null_timeout_rq,
1761 .poll = null_poll,
1762 .map_queues = null_map_queues,
1763 .init_hctx = null_init_hctx,
1764};
1765
1766static void null_del_dev(struct nullb *nullb)
1767{
1768 struct nullb_device *dev;
1769
1770 if (!nullb)
1771 return;
1772
1773 dev = nullb->dev;
1774
1775 ida_free(&nullb_indexes, nullb->index);
1776
1777 list_del_init(&nullb->list);
1778
1779 del_gendisk(nullb->disk);
1780
1781 if (test_bit(NULLB_DEV_FL_THROTTLED, &dev->flags)) {
1782 hrtimer_cancel(&nullb->bw_timer);
1783 atomic_long_set(&nullb->cur_bytes, LONG_MAX);
1784 blk_mq_start_stopped_hw_queues(nullb->q, true);
1785 }
1786
1787 put_disk(nullb->disk);
1788 null_free_zoned_dev(dev);
1789 if (nullb->tag_set == &nullb->__tag_set)
1790 blk_mq_free_tag_set(nullb->tag_set);
1791 kfree(nullb->queues);
1792 if (null_cache_active(nullb))
1793 null_free_device_storage(dev, true);
1794 kfree(nullb);
1795 dev->nullb = NULL;
1796}
1797
1798static void null_config_discard(struct nullb *nullb, struct queue_limits *lim)
1799{
1800 if (nullb->dev->discard == false)
1801 return;
1802
1803 if (!nullb->dev->memory_backed) {
1804 nullb->dev->discard = false;
1805 pr_info("discard option is ignored without memory backing\n");
1806 return;
1807 }
1808
1809 if (nullb->dev->zoned) {
1810 nullb->dev->discard = false;
1811 pr_info("discard option is ignored in zoned mode\n");
1812 return;
1813 }
1814
1815 lim->max_hw_discard_sectors = UINT_MAX >> 9;
1816}
1817
1818static const struct block_device_operations null_ops = {
1819 .owner = THIS_MODULE,
1820 .report_zones = null_report_zones,
1821};
1822
1823static int setup_queues(struct nullb *nullb)
1824{
1825 int nqueues = nr_cpu_ids;
1826
1827 if (g_poll_queues)
1828 nqueues += g_poll_queues;
1829
1830 nullb->queues = kzalloc_objs(struct nullb_queue, nqueues);
1831 if (!nullb->queues)
1832 return -ENOMEM;
1833
1834 return 0;
1835}
1836
1837static int null_init_tag_set(struct blk_mq_tag_set *set, int poll_queues)
1838{
1839 set->ops = &null_mq_ops;
1840 set->cmd_size = sizeof(struct nullb_cmd);
1841 set->timeout = 5 * HZ;
1842 set->nr_maps = 1;
1843 if (poll_queues) {
1844 set->nr_hw_queues += poll_queues;
1845 set->nr_maps += 2;
1846 }
1847 return blk_mq_alloc_tag_set(set);
1848}
1849
1850static int null_init_global_tag_set(void)
1851{
1852 int error;
1853
1854 if (tag_set.ops)
1855 return 0;
1856
1857 tag_set.nr_hw_queues = g_submit_queues;
1858 tag_set.queue_depth = g_hw_queue_depth;
1859 tag_set.numa_node = g_home_node;
1860 if (g_no_sched)
1861 tag_set.flags |= BLK_MQ_F_NO_SCHED_BY_DEFAULT;
1862 if (g_shared_tag_bitmap)
1863 tag_set.flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1864 if (g_blocking)
1865 tag_set.flags |= BLK_MQ_F_BLOCKING;
1866
1867 error = null_init_tag_set(&tag_set, g_poll_queues);
1868 if (error)
1869 tag_set.ops = NULL;
1870 return error;
1871}
1872
1873static int null_setup_tagset(struct nullb *nullb)
1874{
1875 if (nullb->dev->shared_tags) {
1876 nullb->tag_set = &tag_set;
1877 return null_init_global_tag_set();
1878 }
1879
1880 nullb->tag_set = &nullb->__tag_set;
1881 nullb->tag_set->driver_data = nullb;
1882 nullb->tag_set->nr_hw_queues = nullb->dev->submit_queues;
1883 nullb->tag_set->queue_depth = nullb->dev->hw_queue_depth;
1884 nullb->tag_set->numa_node = nullb->dev->home_node;
1885 if (nullb->dev->no_sched)
1886 nullb->tag_set->flags |= BLK_MQ_F_NO_SCHED_BY_DEFAULT;
1887 if (nullb->dev->shared_tag_bitmap)
1888 nullb->tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1889 if (nullb->dev->blocking)
1890 nullb->tag_set->flags |= BLK_MQ_F_BLOCKING;
1891 return null_init_tag_set(nullb->tag_set, nullb->dev->poll_queues);
1892}
1893
1894static int null_validate_conf(struct nullb_device *dev)
1895{
1896 if (dev->queue_mode == NULL_Q_RQ) {
1897 pr_err("legacy IO path is no longer available\n");
1898 return -EINVAL;
1899 }
1900 if (dev->queue_mode == NULL_Q_BIO) {
1901 pr_err("BIO-based IO path is no longer available, using blk-mq instead.\n");
1902 dev->queue_mode = NULL_Q_MQ;
1903 }
1904
1905 if (dev->use_per_node_hctx) {
1906 if (dev->submit_queues != nr_online_nodes)
1907 dev->submit_queues = nr_online_nodes;
1908 } else if (dev->submit_queues > nr_cpu_ids)
1909 dev->submit_queues = nr_cpu_ids;
1910 else if (dev->submit_queues == 0)
1911 dev->submit_queues = 1;
1912 dev->prev_submit_queues = dev->submit_queues;
1913
1914 if (dev->poll_queues > g_poll_queues)
1915 dev->poll_queues = g_poll_queues;
1916 dev->prev_poll_queues = dev->poll_queues;
1917 dev->irqmode = min_t(unsigned int, dev->irqmode, NULL_IRQ_TIMER);
1918
1919 /* Do memory allocation, so set blocking */
1920 if (dev->memory_backed)
1921 dev->blocking = true;
1922 else /* cache is meaningless */
1923 dev->cache_size = 0;
1924 dev->cache_size = min_t(unsigned long, ULONG_MAX / 1024 / 1024,
1925 dev->cache_size);
1926 dev->mbps = min_t(unsigned int, 1024 * 40, dev->mbps);
1927
1928 if (dev->zoned &&
1929 (!dev->zone_size || !is_power_of_2(dev->zone_size))) {
1930 pr_err("zone_size must be power-of-two\n");
1931 return -EINVAL;
1932 }
1933
1934 return 0;
1935}
1936
1937#ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1938static bool __null_setup_fault(struct fault_attr *attr, char *str)
1939{
1940 if (!str[0])
1941 return true;
1942
1943 if (!setup_fault_attr(attr, str))
1944 return false;
1945
1946 attr->verbose = 0;
1947 return true;
1948}
1949#endif
1950
1951static bool null_setup_fault(void)
1952{
1953#ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1954 if (!__null_setup_fault(&null_timeout_attr, g_timeout_str))
1955 return false;
1956 if (!__null_setup_fault(&null_requeue_attr, g_requeue_str))
1957 return false;
1958 if (!__null_setup_fault(&null_init_hctx_attr, g_init_hctx_str))
1959 return false;
1960#endif
1961 return true;
1962}
1963
1964static int null_add_dev(struct nullb_device *dev)
1965{
1966 struct queue_limits lim = {
1967 .logical_block_size = dev->blocksize,
1968 .physical_block_size = dev->blocksize,
1969 .max_hw_sectors = dev->max_sectors,
1970 .dma_alignment = 1,
1971 };
1972
1973 struct nullb *nullb;
1974 int rv;
1975
1976 rv = null_validate_conf(dev);
1977 if (rv)
1978 return rv;
1979
1980 nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, dev->home_node);
1981 if (!nullb) {
1982 rv = -ENOMEM;
1983 goto out;
1984 }
1985 nullb->dev = dev;
1986 dev->nullb = nullb;
1987
1988 spin_lock_init(&nullb->lock);
1989
1990 rv = setup_queues(nullb);
1991 if (rv)
1992 goto out_free_nullb;
1993
1994 rv = null_setup_tagset(nullb);
1995 if (rv)
1996 goto out_cleanup_queues;
1997
1998 if (dev->virt_boundary)
1999 lim.virt_boundary_mask = PAGE_SIZE - 1;
2000 null_config_discard(nullb, &lim);
2001 if (dev->zoned) {
2002 rv = null_init_zoned_dev(dev, &lim);
2003 if (rv)
2004 goto out_cleanup_tags;
2005 }
2006
2007 if (dev->cache_size > 0) {
2008 set_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
2009 lim.features |= BLK_FEAT_WRITE_CACHE;
2010 if (dev->fua)
2011 lim.features |= BLK_FEAT_FUA;
2012 }
2013
2014 if (dev->rotational)
2015 lim.features |= BLK_FEAT_ROTATIONAL;
2016
2017 nullb->disk = blk_mq_alloc_disk(nullb->tag_set, &lim, nullb);
2018 if (IS_ERR(nullb->disk)) {
2019 rv = PTR_ERR(nullb->disk);
2020 goto out_cleanup_zone;
2021 }
2022 nullb->q = nullb->disk->queue;
2023
2024 if (dev->mbps) {
2025 set_bit(NULLB_DEV_FL_THROTTLED, &dev->flags);
2026 nullb_setup_bwtimer(nullb);
2027 }
2028
2029 nullb->q->queuedata = nullb;
2030
2031 rv = ida_alloc(&nullb_indexes, GFP_KERNEL);
2032 if (rv < 0)
2033 goto out_cleanup_disk;
2034
2035 nullb->index = rv;
2036 dev->index = rv;
2037
2038 if (config_item_name(&dev->group.cg_item)) {
2039 /* Use configfs dir name as the device name */
2040 snprintf(nullb->disk_name, sizeof(nullb->disk_name),
2041 "%s", config_item_name(&dev->group.cg_item));
2042 } else {
2043 sprintf(nullb->disk_name, "nullb%d", nullb->index);
2044 }
2045
2046 set_capacity(nullb->disk,
2047 ((sector_t)nullb->dev->size * SZ_1M) >> SECTOR_SHIFT);
2048 nullb->disk->major = null_major;
2049 nullb->disk->first_minor = nullb->index;
2050 nullb->disk->minors = 1;
2051 nullb->disk->fops = &null_ops;
2052 nullb->disk->private_data = nullb;
2053 strscpy(nullb->disk->disk_name, nullb->disk_name);
2054
2055 if (nullb->dev->zoned) {
2056 rv = null_register_zoned_dev(nullb);
2057 if (rv)
2058 goto out_ida_free;
2059 }
2060
2061 rv = add_disk(nullb->disk);
2062 if (rv)
2063 goto out_ida_free;
2064
2065 list_add_tail(&nullb->list, &nullb_list);
2066
2067 pr_info("disk %s created\n", nullb->disk_name);
2068
2069 return 0;
2070
2071out_ida_free:
2072 ida_free(&nullb_indexes, nullb->index);
2073out_cleanup_disk:
2074 put_disk(nullb->disk);
2075out_cleanup_zone:
2076 null_free_zoned_dev(dev);
2077out_cleanup_tags:
2078 if (nullb->tag_set == &nullb->__tag_set)
2079 blk_mq_free_tag_set(nullb->tag_set);
2080out_cleanup_queues:
2081 kfree(nullb->queues);
2082out_free_nullb:
2083 kfree(nullb);
2084 dev->nullb = NULL;
2085out:
2086 return rv;
2087}
2088
2089static struct nullb *null_find_dev_by_name(const char *name)
2090{
2091 struct nullb *nullb = NULL, *nb;
2092
2093 guard(mutex)(&lock);
2094 list_for_each_entry(nb, &nullb_list, list) {
2095 if (strcmp(nb->disk_name, name) == 0) {
2096 nullb = nb;
2097 break;
2098 }
2099 }
2100
2101 return nullb;
2102}
2103
2104static int null_create_dev(void)
2105{
2106 struct nullb_device *dev;
2107 int ret;
2108
2109 dev = null_alloc_dev();
2110 if (!dev)
2111 return -ENOMEM;
2112
2113 scoped_guard(mutex, &lock) {
2114 ret = null_add_dev(dev);
2115 }
2116 if (ret) {
2117 null_free_dev(dev);
2118 return ret;
2119 }
2120
2121 return 0;
2122}
2123
2124static void null_destroy_dev(struct nullb *nullb)
2125{
2126 struct nullb_device *dev = nullb->dev;
2127
2128 null_del_dev(nullb);
2129 null_free_device_storage(dev, false);
2130 null_free_dev(dev);
2131}
2132
2133static int __init null_init(void)
2134{
2135 int ret = 0;
2136 unsigned int i;
2137 struct nullb *nullb;
2138
2139 if (g_bs > PAGE_SIZE) {
2140 pr_warn("invalid block size\n");
2141 pr_warn("defaults block size to %lu\n", PAGE_SIZE);
2142 g_bs = PAGE_SIZE;
2143 }
2144
2145 if (g_home_node != NUMA_NO_NODE && g_home_node >= nr_online_nodes) {
2146 pr_err("invalid home_node value\n");
2147 g_home_node = NUMA_NO_NODE;
2148 }
2149
2150 if (!null_setup_fault())
2151 return -EINVAL;
2152
2153 if (g_queue_mode == NULL_Q_RQ) {
2154 pr_err("legacy IO path is no longer available\n");
2155 return -EINVAL;
2156 }
2157
2158 if (g_use_per_node_hctx) {
2159 if (g_submit_queues != nr_online_nodes) {
2160 pr_warn("submit_queues param is set to %u.\n",
2161 nr_online_nodes);
2162 g_submit_queues = nr_online_nodes;
2163 }
2164 } else if (g_submit_queues > nr_cpu_ids) {
2165 g_submit_queues = nr_cpu_ids;
2166 } else if (g_submit_queues <= 0) {
2167 g_submit_queues = 1;
2168 }
2169
2170 config_group_init(&nullb_subsys.su_group);
2171 mutex_init(&nullb_subsys.su_mutex);
2172
2173 null_major = register_blkdev(0, "nullb");
2174 if (null_major < 0)
2175 return null_major;
2176
2177 for (i = 0; i < nr_devices; i++) {
2178 ret = null_create_dev();
2179 if (ret)
2180 goto err_dev;
2181 }
2182
2183 ret = configfs_register_subsystem(&nullb_subsys);
2184 if (ret)
2185 goto err_dev;
2186
2187 pr_info("module loaded\n");
2188 return 0;
2189
2190err_dev:
2191 while (!list_empty(&nullb_list)) {
2192 nullb = list_entry(nullb_list.next, struct nullb, list);
2193 null_destroy_dev(nullb);
2194 }
2195 unregister_blkdev(null_major, "nullb");
2196 if (tag_set.ops)
2197 blk_mq_free_tag_set(&tag_set);
2198 return ret;
2199}
2200
2201static void __exit null_exit(void)
2202{
2203 struct nullb *nullb;
2204
2205 configfs_unregister_subsystem(&nullb_subsys);
2206
2207 scoped_guard(mutex, &lock) {
2208 while (!list_empty(&nullb_list)) {
2209 nullb = list_entry(nullb_list.next, struct nullb, list);
2210 null_destroy_dev(nullb);
2211 }
2212 }
2213
2214 unregister_blkdev(null_major, "nullb");
2215
2216 if (tag_set.ops)
2217 blk_mq_free_tag_set(&tag_set);
2218}
2219
2220module_init(null_init);
2221module_exit(null_exit);
2222
2223MODULE_AUTHOR("Jens Axboe <[email protected]>");
2224MODULE_DESCRIPTION("multi queue aware block test driver");
2225MODULE_LICENSE("GPL");