?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'work.regset' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
[linux/fpc-iii.git] / drivers / block / null_blk_main.c
blob47a9dad880af2aac70567889f13dc42680be9a0c
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>
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 #define PAGE_SECTORS_SHIFT (PAGE_SHIFT - SECTOR_SHIFT)
15 #define PAGE_SECTORS (1 << PAGE_SECTORS_SHIFT)
16 #define SECTOR_MASK (PAGE_SECTORS - 1)
17
18 #define FREE_BATCH 16
19
20 #define TICKS_PER_SEC 50ULL
21 #define TIMER_INTERVAL (NSEC_PER_SEC / TICKS_PER_SEC)
22
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
28
29 static inline u64 mb_per_tick(int mbps)
30 {
31 return (1 << 20) / TICKS_PER_SEC * ((u64) mbps);
32 }
33
34 /*
35 * Status flags for nullb_device.
36 *
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.
41 */
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,
47 };
48
49 #define MAP_SZ ((PAGE_SIZE >> SECTOR_SHIFT) + 2)
50 /*
51 * nullb_page is a page in memory for nullb devices.
52 *
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
61 */
62 struct nullb_page {
63 struct page *page;
64 DECLARE_BITMAP(bitmap, MAP_SZ);
65 };
66 #define NULLB_PAGE_LOCK (MAP_SZ - 1)
67 #define NULLB_PAGE_FREE (MAP_SZ - 2)
68
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;
74
75 enum {
76 NULL_IRQ_NONE = 0,
77 NULL_IRQ_SOFTIRQ = 1,
78 NULL_IRQ_TIMER = 2,
79 };
80
81 enum {
82 NULL_Q_BIO = 0,
83 NULL_Q_RQ = 1,
84 NULL_Q_MQ = 2,
85 };
86
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");
90
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");
94
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");
98
99 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
100 /*
101 * For more details about fault injection, please refer to
102 * Documentation/fault-injection/fault-injection.rst.
103 */
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>");
107
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>");
111
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
116
117 static int g_queue_mode = NULL_Q_MQ;
118
119 static int null_param_store_val(const char *str, int *val, int min, int max)
120 {
121 int ret, new_val;
122
123 ret = kstrtoint(str, 10, &new_val);
124 if (ret)
125 return -EINVAL;
126
127 if (new_val < min || new_val > max)
128 return -EINVAL;
129
130 *val = new_val;
131 return 0;
132 }
133
134 static int null_set_queue_mode(const char *str, const struct kernel_param *kp)
135 {
136 return null_param_store_val(str, &g_queue_mode, NULL_Q_BIO, NULL_Q_MQ);
137 }
138
139 static const struct kernel_param_ops null_queue_mode_param_ops = {
140 .set = null_set_queue_mode,
141 .get = param_get_int,
142 };
143
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)");
146
147 static int g_gb = 250;
148 module_param_named(gb, g_gb, int, 0444);
149 MODULE_PARM_DESC(gb, "Size in GB");
150
151 static int g_bs = 512;
152 module_param_named(bs, g_bs, int, 0444);
153 MODULE_PARM_DESC(bs, "Block size (in bytes)");
154
155 static unsigned int nr_devices = 1;
156 module_param(nr_devices, uint, 0444);
157 MODULE_PARM_DESC(nr_devices, "Number of devices to register");
158
159 static bool g_blocking;
160 module_param_named(blocking, g_blocking, bool, 0444);
161 MODULE_PARM_DESC(blocking, "Register as a blocking blk-mq driver device");
162
163 static bool shared_tags;
164 module_param(shared_tags, bool, 0444);
165 MODULE_PARM_DESC(shared_tags, "Share tag set between devices for blk-mq");
166
167 static int g_irqmode = NULL_IRQ_SOFTIRQ;
168
169 static int null_set_irqmode(const char *str, const struct kernel_param *kp)
170 {
171 return null_param_store_val(str, &g_irqmode, NULL_IRQ_NONE,
172 NULL_IRQ_TIMER);
173 }
174
175 static const struct kernel_param_ops null_irqmode_param_ops = {
176 .set = null_set_irqmode,
177 .get = param_get_int,
178 };
179
180 device_param_cb(irqmode, &null_irqmode_param_ops, &g_irqmode, 0444);
181 MODULE_PARM_DESC(irqmode, "IRQ completion handler. 0-none, 1-softirq, 2-timer");
182
183 static unsigned long g_completion_nsec = 10000;
184 module_param_named(completion_nsec, g_completion_nsec, ulong, 0444);
185 MODULE_PARM_DESC(completion_nsec, "Time in ns to complete a request in hardware. Default: 10,000ns");
186
187 static int g_hw_queue_depth = 64;
188 module_param_named(hw_queue_depth, g_hw_queue_depth, int, 0444);
189 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 64");
190
191 static bool g_use_per_node_hctx;
192 module_param_named(use_per_node_hctx, g_use_per_node_hctx, bool, 0444);
193 MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware context queues. Default: false");
194
195 static bool g_zoned;
196 module_param_named(zoned, g_zoned, bool, S_IRUGO);
197 MODULE_PARM_DESC(zoned, "Make device as a host-managed zoned block device. Default: false");
198
199 static unsigned long g_zone_size = 256;
200 module_param_named(zone_size, g_zone_size, ulong, S_IRUGO);
201 MODULE_PARM_DESC(zone_size, "Zone size in MB when block device is zoned. Must be power-of-two: Default: 256");
202
203 static unsigned long g_zone_capacity;
204 module_param_named(zone_capacity, g_zone_capacity, ulong, 0444);
205 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");
206
207 static unsigned int g_zone_nr_conv;
208 module_param_named(zone_nr_conv, g_zone_nr_conv, uint, 0444);
209 MODULE_PARM_DESC(zone_nr_conv, "Number of conventional zones when block device is zoned. Default: 0");
210
211 static struct nullb_device *null_alloc_dev(void);
212 static void null_free_dev(struct nullb_device *dev);
213 static void null_del_dev(struct nullb *nullb);
214 static int null_add_dev(struct nullb_device *dev);
215 static void null_free_device_storage(struct nullb_device *dev, bool is_cache);
216
217 static inline struct nullb_device *to_nullb_device(struct config_item *item)
218 {
219 return item ? container_of(item, struct nullb_device, item) : NULL;
220 }
221
222 static inline ssize_t nullb_device_uint_attr_show(unsigned int val, char *page)
223 {
224 return snprintf(page, PAGE_SIZE, "%u\n", val);
225 }
226
227 static inline ssize_t nullb_device_ulong_attr_show(unsigned long val,
228 char *page)
229 {
230 return snprintf(page, PAGE_SIZE, "%lu\n", val);
231 }
232
233 static inline ssize_t nullb_device_bool_attr_show(bool val, char *page)
234 {
235 return snprintf(page, PAGE_SIZE, "%u\n", val);
236 }
237
238 static ssize_t nullb_device_uint_attr_store(unsigned int *val,
239 const char *page, size_t count)
240 {
241 unsigned int tmp;
242 int result;
243
244 result = kstrtouint(page, 0, &tmp);
245 if (result < 0)
246 return result;
247
248 *val = tmp;
249 return count;
250 }
251
252 static ssize_t nullb_device_ulong_attr_store(unsigned long *val,
253 const char *page, size_t count)
254 {
255 int result;
256 unsigned long tmp;
257
258 result = kstrtoul(page, 0, &tmp);
259 if (result < 0)
260 return result;
261
262 *val = tmp;
263 return count;
264 }
265
266 static ssize_t nullb_device_bool_attr_store(bool *val, const char *page,
267 size_t count)
268 {
269 bool tmp;
270 int result;
271
272 result = kstrtobool(page, &tmp);
273 if (result < 0)
274 return result;
275
276 *val = tmp;
277 return count;
278 }
279
280 /* The following macro should only be used with TYPE = {uint, ulong, bool}. */
281 #define NULLB_DEVICE_ATTR(NAME, TYPE, APPLY) \
282 static ssize_t \
283 nullb_device_##NAME##_show(struct config_item *item, char *page) \
284 { \
285 return nullb_device_##TYPE##_attr_show( \
286 to_nullb_device(item)->NAME, page); \
287 } \
288 static ssize_t \
289 nullb_device_##NAME##_store(struct config_item *item, const char *page, \
290 size_t count) \
291 { \
292 int (*apply_fn)(struct nullb_device *dev, TYPE new_value) = APPLY;\
293 struct nullb_device *dev = to_nullb_device(item); \
294 TYPE new_value = 0; \
295 int ret; \
296 \
297 ret = nullb_device_##TYPE##_attr_store(&new_value, page, count);\
298 if (ret < 0) \
299 return ret; \
300 if (apply_fn) \
301 ret = apply_fn(dev, new_value); \
302 else if (test_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags)) \
303 ret = -EBUSY; \
304 if (ret < 0) \
305 return ret; \
306 dev->NAME = new_value; \
307 return count; \
308 } \
309 CONFIGFS_ATTR(nullb_device_, NAME);
310
311 static int nullb_apply_submit_queues(struct nullb_device *dev,
312 unsigned int submit_queues)
313 {
314 struct nullb *nullb = dev->nullb;
315 struct blk_mq_tag_set *set;
316
317 if (!nullb)
318 return 0;
319
320 /*
321 * Make sure that null_init_hctx() does not access nullb->queues[] past
322 * the end of that array.
323 */
324 if (submit_queues > nr_cpu_ids)
325 return -EINVAL;
326 set = nullb->tag_set;
327 blk_mq_update_nr_hw_queues(set, submit_queues);
328 return set->nr_hw_queues == submit_queues ? 0 : -ENOMEM;
329 }
330
331 NULLB_DEVICE_ATTR(size, ulong, NULL);
332 NULLB_DEVICE_ATTR(completion_nsec, ulong, NULL);
333 NULLB_DEVICE_ATTR(submit_queues, uint, nullb_apply_submit_queues);
334 NULLB_DEVICE_ATTR(home_node, uint, NULL);
335 NULLB_DEVICE_ATTR(queue_mode, uint, NULL);
336 NULLB_DEVICE_ATTR(blocksize, uint, NULL);
337 NULLB_DEVICE_ATTR(irqmode, uint, NULL);
338 NULLB_DEVICE_ATTR(hw_queue_depth, uint, NULL);
339 NULLB_DEVICE_ATTR(index, uint, NULL);
340 NULLB_DEVICE_ATTR(blocking, bool, NULL);
341 NULLB_DEVICE_ATTR(use_per_node_hctx, bool, NULL);
342 NULLB_DEVICE_ATTR(memory_backed, bool, NULL);
343 NULLB_DEVICE_ATTR(discard, bool, NULL);
344 NULLB_DEVICE_ATTR(mbps, uint, NULL);
345 NULLB_DEVICE_ATTR(cache_size, ulong, NULL);
346 NULLB_DEVICE_ATTR(zoned, bool, NULL);
347 NULLB_DEVICE_ATTR(zone_size, ulong, NULL);
348 NULLB_DEVICE_ATTR(zone_capacity, ulong, NULL);
349 NULLB_DEVICE_ATTR(zone_nr_conv, uint, NULL);
350
351 static ssize_t nullb_device_power_show(struct config_item *item, char *page)
352 {
353 return nullb_device_bool_attr_show(to_nullb_device(item)->power, page);
354 }
355
356 static ssize_t nullb_device_power_store(struct config_item *item,
357 const char *page, size_t count)
358 {
359 struct nullb_device *dev = to_nullb_device(item);
360 bool newp = false;
361 ssize_t ret;
362
363 ret = nullb_device_bool_attr_store(&newp, page, count);
364 if (ret < 0)
365 return ret;
366
367 if (!dev->power && newp) {
368 if (test_and_set_bit(NULLB_DEV_FL_UP, &dev->flags))
369 return count;
370 if (null_add_dev(dev)) {
371 clear_bit(NULLB_DEV_FL_UP, &dev->flags);
372 return -ENOMEM;
373 }
374
375 set_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
376 dev->power = newp;
377 } else if (dev->power && !newp) {
378 if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
379 mutex_lock(&lock);
380 dev->power = newp;
381 null_del_dev(dev->nullb);
382 mutex_unlock(&lock);
383 }
384 clear_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
385 }
386
387 return count;
388 }
389
390 CONFIGFS_ATTR(nullb_device_, power);
391
392 static ssize_t nullb_device_badblocks_show(struct config_item *item, char *page)
393 {
394 struct nullb_device *t_dev = to_nullb_device(item);
395
396 return badblocks_show(&t_dev->badblocks, page, 0);
397 }
398
399 static ssize_t nullb_device_badblocks_store(struct config_item *item,
400 const char *page, size_t count)
401 {
402 struct nullb_device *t_dev = to_nullb_device(item);
403 char *orig, *buf, *tmp;
404 u64 start, end;
405 int ret;
406
407 orig = kstrndup(page, count, GFP_KERNEL);
408 if (!orig)
409 return -ENOMEM;
410
411 buf = strstrip(orig);
412
413 ret = -EINVAL;
414 if (buf[0] != '+' && buf[0] != '-')
415 goto out;
416 tmp = strchr(&buf[1], '-');
417 if (!tmp)
418 goto out;
419 *tmp = '\0';
420 ret = kstrtoull(buf + 1, 0, &start);
421 if (ret)
422 goto out;
423 ret = kstrtoull(tmp + 1, 0, &end);
424 if (ret)
425 goto out;
426 ret = -EINVAL;
427 if (start > end)
428 goto out;
429 /* enable badblocks */
430 cmpxchg(&t_dev->badblocks.shift, -1, 0);
431 if (buf[0] == '+')
432 ret = badblocks_set(&t_dev->badblocks, start,
433 end - start + 1, 1);
434 else
435 ret = badblocks_clear(&t_dev->badblocks, start,
436 end - start + 1);
437 if (ret == 0)
438 ret = count;
439 out:
440 kfree(orig);
441 return ret;
442 }
443 CONFIGFS_ATTR(nullb_device_, badblocks);
444
445 static struct configfs_attribute *nullb_device_attrs[] = {
446 &nullb_device_attr_size,
447 &nullb_device_attr_completion_nsec,
448 &nullb_device_attr_submit_queues,
449 &nullb_device_attr_home_node,
450 &nullb_device_attr_queue_mode,
451 &nullb_device_attr_blocksize,
452 &nullb_device_attr_irqmode,
453 &nullb_device_attr_hw_queue_depth,
454 &nullb_device_attr_index,
455 &nullb_device_attr_blocking,
456 &nullb_device_attr_use_per_node_hctx,
457 &nullb_device_attr_power,
458 &nullb_device_attr_memory_backed,
459 &nullb_device_attr_discard,
460 &nullb_device_attr_mbps,
461 &nullb_device_attr_cache_size,
462 &nullb_device_attr_badblocks,
463 &nullb_device_attr_zoned,
464 &nullb_device_attr_zone_size,
465 &nullb_device_attr_zone_capacity,
466 &nullb_device_attr_zone_nr_conv,
467 NULL,
468 };
469
470 static void nullb_device_release(struct config_item *item)
471 {
472 struct nullb_device *dev = to_nullb_device(item);
473
474 null_free_device_storage(dev, false);
475 null_free_dev(dev);
476 }
477
478 static struct configfs_item_operations nullb_device_ops = {
479 .release = nullb_device_release,
480 };
481
482 static const struct config_item_type nullb_device_type = {
483 .ct_item_ops = &nullb_device_ops,
484 .ct_attrs = nullb_device_attrs,
485 .ct_owner = THIS_MODULE,
486 };
487
488 static struct
489 config_item *nullb_group_make_item(struct config_group *group, const char *name)
490 {
491 struct nullb_device *dev;
492
493 dev = null_alloc_dev();
494 if (!dev)
495 return ERR_PTR(-ENOMEM);
496
497 config_item_init_type_name(&dev->item, name, &nullb_device_type);
498
499 return &dev->item;
500 }
501
502 static void
503 nullb_group_drop_item(struct config_group *group, struct config_item *item)
504 {
505 struct nullb_device *dev = to_nullb_device(item);
506
507 if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
508 mutex_lock(&lock);
509 dev->power = false;
510 null_del_dev(dev->nullb);
511 mutex_unlock(&lock);
512 }
513
514 config_item_put(item);
515 }
516
517 static ssize_t memb_group_features_show(struct config_item *item, char *page)
518 {
519 return snprintf(page, PAGE_SIZE,
520 "memory_backed,discard,bandwidth,cache,badblocks,zoned,zone_size,zone_capacity,zone_nr_conv\n");
521 }
522
523 CONFIGFS_ATTR_RO(memb_group_, features);
524
525 static struct configfs_attribute *nullb_group_attrs[] = {
526 &memb_group_attr_features,
527 NULL,
528 };
529
530 static struct configfs_group_operations nullb_group_ops = {
531 .make_item = nullb_group_make_item,
532 .drop_item = nullb_group_drop_item,
533 };
534
535 static const struct config_item_type nullb_group_type = {
536 .ct_group_ops = &nullb_group_ops,
537 .ct_attrs = nullb_group_attrs,
538 .ct_owner = THIS_MODULE,
539 };
540
541 static struct configfs_subsystem nullb_subsys = {
542 .su_group = {
543 .cg_item = {
544 .ci_namebuf = "nullb",
545 .ci_type = &nullb_group_type,
546 },
547 },
548 };
549
550 static inline int null_cache_active(struct nullb *nullb)
551 {
552 return test_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
553 }
554
555 static struct nullb_device *null_alloc_dev(void)
556 {
557 struct nullb_device *dev;
558
559 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
560 if (!dev)
561 return NULL;
562 INIT_RADIX_TREE(&dev->data, GFP_ATOMIC);
563 INIT_RADIX_TREE(&dev->cache, GFP_ATOMIC);
564 if (badblocks_init(&dev->badblocks, 0)) {
565 kfree(dev);
566 return NULL;
567 }
568
569 dev->size = g_gb * 1024;
570 dev->completion_nsec = g_completion_nsec;
571 dev->submit_queues = g_submit_queues;
572 dev->home_node = g_home_node;
573 dev->queue_mode = g_queue_mode;
574 dev->blocksize = g_bs;
575 dev->irqmode = g_irqmode;
576 dev->hw_queue_depth = g_hw_queue_depth;
577 dev->blocking = g_blocking;
578 dev->use_per_node_hctx = g_use_per_node_hctx;
579 dev->zoned = g_zoned;
580 dev->zone_size = g_zone_size;
581 dev->zone_capacity = g_zone_capacity;
582 dev->zone_nr_conv = g_zone_nr_conv;
583 return dev;
584 }
585
586 static void null_free_dev(struct nullb_device *dev)
587 {
588 if (!dev)
589 return;
590
591 null_free_zoned_dev(dev);
592 badblocks_exit(&dev->badblocks);
593 kfree(dev);
594 }
595
596 static void put_tag(struct nullb_queue *nq, unsigned int tag)
597 {
598 clear_bit_unlock(tag, nq->tag_map);
599
600 if (waitqueue_active(&nq->wait))
601 wake_up(&nq->wait);
602 }
603
604 static unsigned int get_tag(struct nullb_queue *nq)
605 {
606 unsigned int tag;
607
608 do {
609 tag = find_first_zero_bit(nq->tag_map, nq->queue_depth);
610 if (tag >= nq->queue_depth)
611 return -1U;
612 } while (test_and_set_bit_lock(tag, nq->tag_map));
613
614 return tag;
615 }
616
617 static void free_cmd(struct nullb_cmd *cmd)
618 {
619 put_tag(cmd->nq, cmd->tag);
620 }
621
622 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer);
623
624 static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq)
625 {
626 struct nullb_cmd *cmd;
627 unsigned int tag;
628
629 tag = get_tag(nq);
630 if (tag != -1U) {
631 cmd = &nq->cmds[tag];
632 cmd->tag = tag;
633 cmd->error = BLK_STS_OK;
634 cmd->nq = nq;
635 if (nq->dev->irqmode == NULL_IRQ_TIMER) {
636 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC,
637 HRTIMER_MODE_REL);
638 cmd->timer.function = null_cmd_timer_expired;
639 }
640 return cmd;
641 }
642
643 return NULL;
644 }
645
646 static struct nullb_cmd *alloc_cmd(struct nullb_queue *nq, int can_wait)
647 {
648 struct nullb_cmd *cmd;
649 DEFINE_WAIT(wait);
650
651 cmd = __alloc_cmd(nq);
652 if (cmd || !can_wait)
653 return cmd;
654
655 do {
656 prepare_to_wait(&nq->wait, &wait, TASK_UNINTERRUPTIBLE);
657 cmd = __alloc_cmd(nq);
658 if (cmd)
659 break;
660
661 io_schedule();
662 } while (1);
663
664 finish_wait(&nq->wait, &wait);
665 return cmd;
666 }
667
668 static void end_cmd(struct nullb_cmd *cmd)
669 {
670 int queue_mode = cmd->nq->dev->queue_mode;
671
672 switch (queue_mode) {
673 case NULL_Q_MQ:
674 blk_mq_end_request(cmd->rq, cmd->error);
675 return;
676 case NULL_Q_BIO:
677 cmd->bio->bi_status = cmd->error;
678 bio_endio(cmd->bio);
679 break;
680 }
681
682 free_cmd(cmd);
683 }
684
685 static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer)
686 {
687 end_cmd(container_of(timer, struct nullb_cmd, timer));
688
689 return HRTIMER_NORESTART;
690 }
691
692 static void null_cmd_end_timer(struct nullb_cmd *cmd)
693 {
694 ktime_t kt = cmd->nq->dev->completion_nsec;
695
696 hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL);
697 }
698
699 static void null_complete_rq(struct request *rq)
700 {
701 end_cmd(blk_mq_rq_to_pdu(rq));
702 }
703
704 static struct nullb_page *null_alloc_page(gfp_t gfp_flags)
705 {
706 struct nullb_page *t_page;
707
708 t_page = kmalloc(sizeof(struct nullb_page), gfp_flags);
709 if (!t_page)
710 goto out;
711
712 t_page->page = alloc_pages(gfp_flags, 0);
713 if (!t_page->page)
714 goto out_freepage;
715
716 memset(t_page->bitmap, 0, sizeof(t_page->bitmap));
717 return t_page;
718 out_freepage:
719 kfree(t_page);
720 out:
721 return NULL;
722 }
723
724 static void null_free_page(struct nullb_page *t_page)
725 {
726 __set_bit(NULLB_PAGE_FREE, t_page->bitmap);
727 if (test_bit(NULLB_PAGE_LOCK, t_page->bitmap))
728 return;
729 __free_page(t_page->page);
730 kfree(t_page);
731 }
732
733 static bool null_page_empty(struct nullb_page *page)
734 {
735 int size = MAP_SZ - 2;
736
737 return find_first_bit(page->bitmap, size) == size;
738 }
739
740 static void null_free_sector(struct nullb *nullb, sector_t sector,
741 bool is_cache)
742 {
743 unsigned int sector_bit;
744 u64 idx;
745 struct nullb_page *t_page, *ret;
746 struct radix_tree_root *root;
747
748 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
749 idx = sector >> PAGE_SECTORS_SHIFT;
750 sector_bit = (sector & SECTOR_MASK);
751
752 t_page = radix_tree_lookup(root, idx);
753 if (t_page) {
754 __clear_bit(sector_bit, t_page->bitmap);
755
756 if (null_page_empty(t_page)) {
757 ret = radix_tree_delete_item(root, idx, t_page);
758 WARN_ON(ret != t_page);
759 null_free_page(ret);
760 if (is_cache)
761 nullb->dev->curr_cache -= PAGE_SIZE;
762 }
763 }
764 }
765
766 static struct nullb_page *null_radix_tree_insert(struct nullb *nullb, u64 idx,
767 struct nullb_page *t_page, bool is_cache)
768 {
769 struct radix_tree_root *root;
770
771 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
772
773 if (radix_tree_insert(root, idx, t_page)) {
774 null_free_page(t_page);
775 t_page = radix_tree_lookup(root, idx);
776 WARN_ON(!t_page || t_page->page->index != idx);
777 } else if (is_cache)
778 nullb->dev->curr_cache += PAGE_SIZE;
779
780 return t_page;
781 }
782
783 static void null_free_device_storage(struct nullb_device *dev, bool is_cache)
784 {
785 unsigned long pos = 0;
786 int nr_pages;
787 struct nullb_page *ret, *t_pages[FREE_BATCH];
788 struct radix_tree_root *root;
789
790 root = is_cache ? &dev->cache : &dev->data;
791
792 do {
793 int i;
794
795 nr_pages = radix_tree_gang_lookup(root,
796 (void **)t_pages, pos, FREE_BATCH);
797
798 for (i = 0; i < nr_pages; i++) {
799 pos = t_pages[i]->page->index;
800 ret = radix_tree_delete_item(root, pos, t_pages[i]);
801 WARN_ON(ret != t_pages[i]);
802 null_free_page(ret);
803 }
804
805 pos++;
806 } while (nr_pages == FREE_BATCH);
807
808 if (is_cache)
809 dev->curr_cache = 0;
810 }
811
812 static struct nullb_page *__null_lookup_page(struct nullb *nullb,
813 sector_t sector, bool for_write, bool is_cache)
814 {
815 unsigned int sector_bit;
816 u64 idx;
817 struct nullb_page *t_page;
818 struct radix_tree_root *root;
819
820 idx = sector >> PAGE_SECTORS_SHIFT;
821 sector_bit = (sector & SECTOR_MASK);
822
823 root = is_cache ? &nullb->dev->cache : &nullb->dev->data;
824 t_page = radix_tree_lookup(root, idx);
825 WARN_ON(t_page && t_page->page->index != idx);
826
827 if (t_page && (for_write || test_bit(sector_bit, t_page->bitmap)))
828 return t_page;
829
830 return NULL;
831 }
832
833 static struct nullb_page *null_lookup_page(struct nullb *nullb,
834 sector_t sector, bool for_write, bool ignore_cache)
835 {
836 struct nullb_page *page = NULL;
837
838 if (!ignore_cache)
839 page = __null_lookup_page(nullb, sector, for_write, true);
840 if (page)
841 return page;
842 return __null_lookup_page(nullb, sector, for_write, false);
843 }
844
845 static struct nullb_page *null_insert_page(struct nullb *nullb,
846 sector_t sector, bool ignore_cache)
847 __releases(&nullb->lock)
848 __acquires(&nullb->lock)
849 {
850 u64 idx;
851 struct nullb_page *t_page;
852
853 t_page = null_lookup_page(nullb, sector, true, ignore_cache);
854 if (t_page)
855 return t_page;
856
857 spin_unlock_irq(&nullb->lock);
858
859 t_page = null_alloc_page(GFP_NOIO);
860 if (!t_page)
861 goto out_lock;
862
863 if (radix_tree_preload(GFP_NOIO))
864 goto out_freepage;
865
866 spin_lock_irq(&nullb->lock);
867 idx = sector >> PAGE_SECTORS_SHIFT;
868 t_page->page->index = idx;
869 t_page = null_radix_tree_insert(nullb, idx, t_page, !ignore_cache);
870 radix_tree_preload_end();
871
872 return t_page;
873 out_freepage:
874 null_free_page(t_page);
875 out_lock:
876 spin_lock_irq(&nullb->lock);
877 return null_lookup_page(nullb, sector, true, ignore_cache);
878 }
879
880 static int null_flush_cache_page(struct nullb *nullb, struct nullb_page *c_page)
881 {
882 int i;
883 unsigned int offset;
884 u64 idx;
885 struct nullb_page *t_page, *ret;
886 void *dst, *src;
887
888 idx = c_page->page->index;
889
890 t_page = null_insert_page(nullb, idx << PAGE_SECTORS_SHIFT, true);
891
892 __clear_bit(NULLB_PAGE_LOCK, c_page->bitmap);
893 if (test_bit(NULLB_PAGE_FREE, c_page->bitmap)) {
894 null_free_page(c_page);
895 if (t_page && null_page_empty(t_page)) {
896 ret = radix_tree_delete_item(&nullb->dev->data,
897 idx, t_page);
898 null_free_page(t_page);
899 }
900 return 0;
901 }
902
903 if (!t_page)
904 return -ENOMEM;
905
906 src = kmap_atomic(c_page->page);
907 dst = kmap_atomic(t_page->page);
908
909 for (i = 0; i < PAGE_SECTORS;
910 i += (nullb->dev->blocksize >> SECTOR_SHIFT)) {
911 if (test_bit(i, c_page->bitmap)) {
912 offset = (i << SECTOR_SHIFT);
913 memcpy(dst + offset, src + offset,
914 nullb->dev->blocksize);
915 __set_bit(i, t_page->bitmap);
916 }
917 }
918
919 kunmap_atomic(dst);
920 kunmap_atomic(src);
921
922 ret = radix_tree_delete_item(&nullb->dev->cache, idx, c_page);
923 null_free_page(ret);
924 nullb->dev->curr_cache -= PAGE_SIZE;
925
926 return 0;
927 }
928
929 static int null_make_cache_space(struct nullb *nullb, unsigned long n)
930 {
931 int i, err, nr_pages;
932 struct nullb_page *c_pages[FREE_BATCH];
933 unsigned long flushed = 0, one_round;
934
935 again:
936 if ((nullb->dev->cache_size * 1024 * 1024) >
937 nullb->dev->curr_cache + n || nullb->dev->curr_cache == 0)
938 return 0;
939
940 nr_pages = radix_tree_gang_lookup(&nullb->dev->cache,
941 (void **)c_pages, nullb->cache_flush_pos, FREE_BATCH);
942 /*
943 * nullb_flush_cache_page could unlock before using the c_pages. To
944 * avoid race, we don't allow page free
945 */
946 for (i = 0; i < nr_pages; i++) {
947 nullb->cache_flush_pos = c_pages[i]->page->index;
948 /*
949 * We found the page which is being flushed to disk by other
950 * threads
951 */
952 if (test_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap))
953 c_pages[i] = NULL;
954 else
955 __set_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap);
956 }
957
958 one_round = 0;
959 for (i = 0; i < nr_pages; i++) {
960 if (c_pages[i] == NULL)
961 continue;
962 err = null_flush_cache_page(nullb, c_pages[i]);
963 if (err)
964 return err;
965 one_round++;
966 }
967 flushed += one_round << PAGE_SHIFT;
968
969 if (n > flushed) {
970 if (nr_pages == 0)
971 nullb->cache_flush_pos = 0;
972 if (one_round == 0) {
973 /* give other threads a chance */
974 spin_unlock_irq(&nullb->lock);
975 spin_lock_irq(&nullb->lock);
976 }
977 goto again;
978 }
979 return 0;
980 }
981
982 static int copy_to_nullb(struct nullb *nullb, struct page *source,
983 unsigned int off, sector_t sector, size_t n, bool is_fua)
984 {
985 size_t temp, count = 0;
986 unsigned int offset;
987 struct nullb_page *t_page;
988 void *dst, *src;
989
990 while (count < n) {
991 temp = min_t(size_t, nullb->dev->blocksize, n - count);
992
993 if (null_cache_active(nullb) && !is_fua)
994 null_make_cache_space(nullb, PAGE_SIZE);
995
996 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
997 t_page = null_insert_page(nullb, sector,
998 !null_cache_active(nullb) || is_fua);
999 if (!t_page)
1000 return -ENOSPC;
1001
1002 src = kmap_atomic(source);
1003 dst = kmap_atomic(t_page->page);
1004 memcpy(dst + offset, src + off + count, temp);
1005 kunmap_atomic(dst);
1006 kunmap_atomic(src);
1007
1008 __set_bit(sector & SECTOR_MASK, t_page->bitmap);
1009
1010 if (is_fua)
1011 null_free_sector(nullb, sector, true);
1012
1013 count += temp;
1014 sector += temp >> SECTOR_SHIFT;
1015 }
1016 return 0;
1017 }
1018
1019 static int copy_from_nullb(struct nullb *nullb, struct page *dest,
1020 unsigned int off, sector_t sector, size_t n)
1021 {
1022 size_t temp, count = 0;
1023 unsigned int offset;
1024 struct nullb_page *t_page;
1025 void *dst, *src;
1026
1027 while (count < n) {
1028 temp = min_t(size_t, nullb->dev->blocksize, n - count);
1029
1030 offset = (sector & SECTOR_MASK) << SECTOR_SHIFT;
1031 t_page = null_lookup_page(nullb, sector, false,
1032 !null_cache_active(nullb));
1033
1034 dst = kmap_atomic(dest);
1035 if (!t_page) {
1036 memset(dst + off + count, 0, temp);
1037 goto next;
1038 }
1039 src = kmap_atomic(t_page->page);
1040 memcpy(dst + off + count, src + offset, temp);
1041 kunmap_atomic(src);
1042 next:
1043 kunmap_atomic(dst);
1044
1045 count += temp;
1046 sector += temp >> SECTOR_SHIFT;
1047 }
1048 return 0;
1049 }
1050
1051 static void nullb_fill_pattern(struct nullb *nullb, struct page *page,
1052 unsigned int len, unsigned int off)
1053 {
1054 void *dst;
1055
1056 dst = kmap_atomic(page);
1057 memset(dst + off, 0xFF, len);
1058 kunmap_atomic(dst);
1059 }
1060
1061 static void null_handle_discard(struct nullb *nullb, sector_t sector, size_t n)
1062 {
1063 size_t temp;
1064
1065 spin_lock_irq(&nullb->lock);
1066 while (n > 0) {
1067 temp = min_t(size_t, n, nullb->dev->blocksize);
1068 null_free_sector(nullb, sector, false);
1069 if (null_cache_active(nullb))
1070 null_free_sector(nullb, sector, true);
1071 sector += temp >> SECTOR_SHIFT;
1072 n -= temp;
1073 }
1074 spin_unlock_irq(&nullb->lock);
1075 }
1076
1077 static int null_handle_flush(struct nullb *nullb)
1078 {
1079 int err;
1080
1081 if (!null_cache_active(nullb))
1082 return 0;
1083
1084 spin_lock_irq(&nullb->lock);
1085 while (true) {
1086 err = null_make_cache_space(nullb,
1087 nullb->dev->cache_size * 1024 * 1024);
1088 if (err || nullb->dev->curr_cache == 0)
1089 break;
1090 }
1091
1092 WARN_ON(!radix_tree_empty(&nullb->dev->cache));
1093 spin_unlock_irq(&nullb->lock);
1094 return err;
1095 }
1096
1097 static int null_transfer(struct nullb *nullb, struct page *page,
1098 unsigned int len, unsigned int off, bool is_write, sector_t sector,
1099 bool is_fua)
1100 {
1101 struct nullb_device *dev = nullb->dev;
1102 unsigned int valid_len = len;
1103 int err = 0;
1104
1105 if (!is_write) {
1106 if (dev->zoned)
1107 valid_len = null_zone_valid_read_len(nullb,
1108 sector, len);
1109
1110 if (valid_len) {
1111 err = copy_from_nullb(nullb, page, off,
1112 sector, valid_len);
1113 off += valid_len;
1114 len -= valid_len;
1115 }
1116
1117 if (len)
1118 nullb_fill_pattern(nullb, page, len, off);
1119 flush_dcache_page(page);
1120 } else {
1121 flush_dcache_page(page);
1122 err = copy_to_nullb(nullb, page, off, sector, len, is_fua);
1123 }
1124
1125 return err;
1126 }
1127
1128 static int null_handle_rq(struct nullb_cmd *cmd)
1129 {
1130 struct request *rq = cmd->rq;
1131 struct nullb *nullb = cmd->nq->dev->nullb;
1132 int err;
1133 unsigned int len;
1134 sector_t sector;
1135 struct req_iterator iter;
1136 struct bio_vec bvec;
1137
1138 sector = blk_rq_pos(rq);
1139
1140 if (req_op(rq) == REQ_OP_DISCARD) {
1141 null_handle_discard(nullb, sector, blk_rq_bytes(rq));
1142 return 0;
1143 }
1144
1145 spin_lock_irq(&nullb->lock);
1146 rq_for_each_segment(bvec, rq, iter) {
1147 len = bvec.bv_len;
1148 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1149 op_is_write(req_op(rq)), sector,
1150 req_op(rq) & REQ_FUA);
1151 if (err) {
1152 spin_unlock_irq(&nullb->lock);
1153 return err;
1154 }
1155 sector += len >> SECTOR_SHIFT;
1156 }
1157 spin_unlock_irq(&nullb->lock);
1158
1159 return 0;
1160 }
1161
1162 static int null_handle_bio(struct nullb_cmd *cmd)
1163 {
1164 struct bio *bio = cmd->bio;
1165 struct nullb *nullb = cmd->nq->dev->nullb;
1166 int err;
1167 unsigned int len;
1168 sector_t sector;
1169 struct bio_vec bvec;
1170 struct bvec_iter iter;
1171
1172 sector = bio->bi_iter.bi_sector;
1173
1174 if (bio_op(bio) == REQ_OP_DISCARD) {
1175 null_handle_discard(nullb, sector,
1176 bio_sectors(bio) << SECTOR_SHIFT);
1177 return 0;
1178 }
1179
1180 spin_lock_irq(&nullb->lock);
1181 bio_for_each_segment(bvec, bio, iter) {
1182 len = bvec.bv_len;
1183 err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset,
1184 op_is_write(bio_op(bio)), sector,
1185 bio->bi_opf & REQ_FUA);
1186 if (err) {
1187 spin_unlock_irq(&nullb->lock);
1188 return err;
1189 }
1190 sector += len >> SECTOR_SHIFT;
1191 }
1192 spin_unlock_irq(&nullb->lock);
1193 return 0;
1194 }
1195
1196 static void null_stop_queue(struct nullb *nullb)
1197 {
1198 struct request_queue *q = nullb->q;
1199
1200 if (nullb->dev->queue_mode == NULL_Q_MQ)
1201 blk_mq_stop_hw_queues(q);
1202 }
1203
1204 static void null_restart_queue_async(struct nullb *nullb)
1205 {
1206 struct request_queue *q = nullb->q;
1207
1208 if (nullb->dev->queue_mode == NULL_Q_MQ)
1209 blk_mq_start_stopped_hw_queues(q, true);
1210 }
1211
1212 static inline blk_status_t null_handle_throttled(struct nullb_cmd *cmd)
1213 {
1214 struct nullb_device *dev = cmd->nq->dev;
1215 struct nullb *nullb = dev->nullb;
1216 blk_status_t sts = BLK_STS_OK;
1217 struct request *rq = cmd->rq;
1218
1219 if (!hrtimer_active(&nullb->bw_timer))
1220 hrtimer_restart(&nullb->bw_timer);
1221
1222 if (atomic_long_sub_return(blk_rq_bytes(rq), &nullb->cur_bytes) < 0) {
1223 null_stop_queue(nullb);
1224 /* race with timer */
1225 if (atomic_long_read(&nullb->cur_bytes) > 0)
1226 null_restart_queue_async(nullb);
1227 /* requeue request */
1228 sts = BLK_STS_DEV_RESOURCE;
1229 }
1230 return sts;
1231 }
1232
1233 static inline blk_status_t null_handle_badblocks(struct nullb_cmd *cmd,
1234 sector_t sector,
1235 sector_t nr_sectors)
1236 {
1237 struct badblocks *bb = &cmd->nq->dev->badblocks;
1238 sector_t first_bad;
1239 int bad_sectors;
1240
1241 if (badblocks_check(bb, sector, nr_sectors, &first_bad, &bad_sectors))
1242 return BLK_STS_IOERR;
1243
1244 return BLK_STS_OK;
1245 }
1246
1247 static inline blk_status_t null_handle_memory_backed(struct nullb_cmd *cmd,
1248 enum req_opf op)
1249 {
1250 struct nullb_device *dev = cmd->nq->dev;
1251 int err;
1252
1253 if (dev->queue_mode == NULL_Q_BIO)
1254 err = null_handle_bio(cmd);
1255 else
1256 err = null_handle_rq(cmd);
1257
1258 return errno_to_blk_status(err);
1259 }
1260
1261 static void nullb_zero_read_cmd_buffer(struct nullb_cmd *cmd)
1262 {
1263 struct nullb_device *dev = cmd->nq->dev;
1264 struct bio *bio;
1265
1266 if (dev->memory_backed)
1267 return;
1268
1269 if (dev->queue_mode == NULL_Q_BIO && bio_op(cmd->bio) == REQ_OP_READ) {
1270 zero_fill_bio(cmd->bio);
1271 } else if (req_op(cmd->rq) == REQ_OP_READ) {
1272 __rq_for_each_bio(bio, cmd->rq)
1273 zero_fill_bio(bio);
1274 }
1275 }
1276
1277 static inline void nullb_complete_cmd(struct nullb_cmd *cmd)
1278 {
1279 /*
1280 * Since root privileges are required to configure the null_blk
1281 * driver, it is fine that this driver does not initialize the
1282 * data buffers of read commands. Zero-initialize these buffers
1283 * anyway if KMSAN is enabled to prevent that KMSAN complains
1284 * about null_blk not initializing read data buffers.
1285 */
1286 if (IS_ENABLED(CONFIG_KMSAN))
1287 nullb_zero_read_cmd_buffer(cmd);
1288
1289 /* Complete IO by inline, softirq or timer */
1290 switch (cmd->nq->dev->irqmode) {
1291 case NULL_IRQ_SOFTIRQ:
1292 switch (cmd->nq->dev->queue_mode) {
1293 case NULL_Q_MQ:
1294 if (likely(!blk_should_fake_timeout(cmd->rq->q)))
1295 blk_mq_complete_request(cmd->rq);
1296 break;
1297 case NULL_Q_BIO:
1298 /*
1299 * XXX: no proper submitting cpu information available.
1300 */
1301 end_cmd(cmd);
1302 break;
1303 }
1304 break;
1305 case NULL_IRQ_NONE:
1306 end_cmd(cmd);
1307 break;
1308 case NULL_IRQ_TIMER:
1309 null_cmd_end_timer(cmd);
1310 break;
1311 }
1312 }
1313
1314 blk_status_t null_process_cmd(struct nullb_cmd *cmd,
1315 enum req_opf op, sector_t sector,
1316 unsigned int nr_sectors)
1317 {
1318 struct nullb_device *dev = cmd->nq->dev;
1319 blk_status_t ret;
1320
1321 if (dev->badblocks.shift != -1) {
1322 ret = null_handle_badblocks(cmd, sector, nr_sectors);
1323 if (ret != BLK_STS_OK)
1324 return ret;
1325 }
1326
1327 if (dev->memory_backed)
1328 return null_handle_memory_backed(cmd, op);
1329
1330 return BLK_STS_OK;
1331 }
1332
1333 static blk_status_t null_handle_cmd(struct nullb_cmd *cmd, sector_t sector,
1334 sector_t nr_sectors, enum req_opf op)
1335 {
1336 struct nullb_device *dev = cmd->nq->dev;
1337 struct nullb *nullb = dev->nullb;
1338 blk_status_t sts;
1339
1340 if (test_bit(NULLB_DEV_FL_THROTTLED, &dev->flags)) {
1341 sts = null_handle_throttled(cmd);
1342 if (sts != BLK_STS_OK)
1343 return sts;
1344 }
1345
1346 if (op == REQ_OP_FLUSH) {
1347 cmd->error = errno_to_blk_status(null_handle_flush(nullb));
1348 goto out;
1349 }
1350
1351 if (dev->zoned)
1352 cmd->error = null_process_zoned_cmd(cmd, op,
1353 sector, nr_sectors);
1354 else
1355 cmd->error = null_process_cmd(cmd, op, sector, nr_sectors);
1356
1357 out:
1358 nullb_complete_cmd(cmd);
1359 return BLK_STS_OK;
1360 }
1361
1362 static enum hrtimer_restart nullb_bwtimer_fn(struct hrtimer *timer)
1363 {
1364 struct nullb *nullb = container_of(timer, struct nullb, bw_timer);
1365 ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1366 unsigned int mbps = nullb->dev->mbps;
1367
1368 if (atomic_long_read(&nullb->cur_bytes) == mb_per_tick(mbps))
1369 return HRTIMER_NORESTART;
1370
1371 atomic_long_set(&nullb->cur_bytes, mb_per_tick(mbps));
1372 null_restart_queue_async(nullb);
1373
1374 hrtimer_forward_now(&nullb->bw_timer, timer_interval);
1375
1376 return HRTIMER_RESTART;
1377 }
1378
1379 static void nullb_setup_bwtimer(struct nullb *nullb)
1380 {
1381 ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL);
1382
1383 hrtimer_init(&nullb->bw_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1384 nullb->bw_timer.function = nullb_bwtimer_fn;
1385 atomic_long_set(&nullb->cur_bytes, mb_per_tick(nullb->dev->mbps));
1386 hrtimer_start(&nullb->bw_timer, timer_interval, HRTIMER_MODE_REL);
1387 }
1388
1389 static struct nullb_queue *nullb_to_queue(struct nullb *nullb)
1390 {
1391 int index = 0;
1392
1393 if (nullb->nr_queues != 1)
1394 index = raw_smp_processor_id() / ((nr_cpu_ids + nullb->nr_queues - 1) / nullb->nr_queues);
1395
1396 return &nullb->queues[index];
1397 }
1398
1399 static blk_qc_t null_submit_bio(struct bio *bio)
1400 {
1401 sector_t sector = bio->bi_iter.bi_sector;
1402 sector_t nr_sectors = bio_sectors(bio);
1403 struct nullb *nullb = bio->bi_disk->private_data;
1404 struct nullb_queue *nq = nullb_to_queue(nullb);
1405 struct nullb_cmd *cmd;
1406
1407 cmd = alloc_cmd(nq, 1);
1408 cmd->bio = bio;
1409
1410 null_handle_cmd(cmd, sector, nr_sectors, bio_op(bio));
1411 return BLK_QC_T_NONE;
1412 }
1413
1414 static bool should_timeout_request(struct request *rq)
1415 {
1416 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1417 if (g_timeout_str[0])
1418 return should_fail(&null_timeout_attr, 1);
1419 #endif
1420 return false;
1421 }
1422
1423 static bool should_requeue_request(struct request *rq)
1424 {
1425 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1426 if (g_requeue_str[0])
1427 return should_fail(&null_requeue_attr, 1);
1428 #endif
1429 return false;
1430 }
1431
1432 static enum blk_eh_timer_return null_timeout_rq(struct request *rq, bool res)
1433 {
1434 pr_info("rq %p timed out\n", rq);
1435 blk_mq_complete_request(rq);
1436 return BLK_EH_DONE;
1437 }
1438
1439 static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx,
1440 const struct blk_mq_queue_data *bd)
1441 {
1442 struct nullb_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
1443 struct nullb_queue *nq = hctx->driver_data;
1444 sector_t nr_sectors = blk_rq_sectors(bd->rq);
1445 sector_t sector = blk_rq_pos(bd->rq);
1446
1447 might_sleep_if(hctx->flags & BLK_MQ_F_BLOCKING);
1448
1449 if (nq->dev->irqmode == NULL_IRQ_TIMER) {
1450 hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1451 cmd->timer.function = null_cmd_timer_expired;
1452 }
1453 cmd->rq = bd->rq;
1454 cmd->error = BLK_STS_OK;
1455 cmd->nq = nq;
1456
1457 blk_mq_start_request(bd->rq);
1458
1459 if (should_requeue_request(bd->rq)) {
1460 /*
1461 * Alternate between hitting the core BUSY path, and the
1462 * driver driven requeue path
1463 */
1464 nq->requeue_selection++;
1465 if (nq->requeue_selection & 1)
1466 return BLK_STS_RESOURCE;
1467 else {
1468 blk_mq_requeue_request(bd->rq, true);
1469 return BLK_STS_OK;
1470 }
1471 }
1472 if (should_timeout_request(bd->rq))
1473 return BLK_STS_OK;
1474
1475 return null_handle_cmd(cmd, sector, nr_sectors, req_op(bd->rq));
1476 }
1477
1478 static void cleanup_queue(struct nullb_queue *nq)
1479 {
1480 kfree(nq->tag_map);
1481 kfree(nq->cmds);
1482 }
1483
1484 static void cleanup_queues(struct nullb *nullb)
1485 {
1486 int i;
1487
1488 for (i = 0; i < nullb->nr_queues; i++)
1489 cleanup_queue(&nullb->queues[i]);
1490
1491 kfree(nullb->queues);
1492 }
1493
1494 static void null_exit_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx)
1495 {
1496 struct nullb_queue *nq = hctx->driver_data;
1497 struct nullb *nullb = nq->dev->nullb;
1498
1499 nullb->nr_queues--;
1500 }
1501
1502 static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq)
1503 {
1504 init_waitqueue_head(&nq->wait);
1505 nq->queue_depth = nullb->queue_depth;
1506 nq->dev = nullb->dev;
1507 }
1508
1509 static int null_init_hctx(struct blk_mq_hw_ctx *hctx, void *driver_data,
1510 unsigned int hctx_idx)
1511 {
1512 struct nullb *nullb = hctx->queue->queuedata;
1513 struct nullb_queue *nq;
1514
1515 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1516 if (g_init_hctx_str[0] && should_fail(&null_init_hctx_attr, 1))
1517 return -EFAULT;
1518 #endif
1519
1520 nq = &nullb->queues[hctx_idx];
1521 hctx->driver_data = nq;
1522 null_init_queue(nullb, nq);
1523 nullb->nr_queues++;
1524
1525 return 0;
1526 }
1527
1528 static const struct blk_mq_ops null_mq_ops = {
1529 .queue_rq = null_queue_rq,
1530 .complete = null_complete_rq,
1531 .timeout = null_timeout_rq,
1532 .init_hctx = null_init_hctx,
1533 .exit_hctx = null_exit_hctx,
1534 };
1535
1536 static void null_del_dev(struct nullb *nullb)
1537 {
1538 struct nullb_device *dev;
1539
1540 if (!nullb)
1541 return;
1542
1543 dev = nullb->dev;
1544
1545 ida_simple_remove(&nullb_indexes, nullb->index);
1546
1547 list_del_init(&nullb->list);
1548
1549 del_gendisk(nullb->disk);
1550
1551 if (test_bit(NULLB_DEV_FL_THROTTLED, &nullb->dev->flags)) {
1552 hrtimer_cancel(&nullb->bw_timer);
1553 atomic_long_set(&nullb->cur_bytes, LONG_MAX);
1554 null_restart_queue_async(nullb);
1555 }
1556
1557 blk_cleanup_queue(nullb->q);
1558 if (dev->queue_mode == NULL_Q_MQ &&
1559 nullb->tag_set == &nullb->__tag_set)
1560 blk_mq_free_tag_set(nullb->tag_set);
1561 put_disk(nullb->disk);
1562 cleanup_queues(nullb);
1563 if (null_cache_active(nullb))
1564 null_free_device_storage(nullb->dev, true);
1565 kfree(nullb);
1566 dev->nullb = NULL;
1567 }
1568
1569 static void null_config_discard(struct nullb *nullb)
1570 {
1571 if (nullb->dev->discard == false)
1572 return;
1573
1574 if (nullb->dev->zoned) {
1575 nullb->dev->discard = false;
1576 pr_info("discard option is ignored in zoned mode\n");
1577 return;
1578 }
1579
1580 nullb->q->limits.discard_granularity = nullb->dev->blocksize;
1581 nullb->q->limits.discard_alignment = nullb->dev->blocksize;
1582 blk_queue_max_discard_sectors(nullb->q, UINT_MAX >> 9);
1583 blk_queue_flag_set(QUEUE_FLAG_DISCARD, nullb->q);
1584 }
1585
1586 static const struct block_device_operations null_bio_ops = {
1587 .owner = THIS_MODULE,
1588 .submit_bio = null_submit_bio,
1589 .report_zones = null_report_zones,
1590 };
1591
1592 static const struct block_device_operations null_rq_ops = {
1593 .owner = THIS_MODULE,
1594 .report_zones = null_report_zones,
1595 };
1596
1597 static int setup_commands(struct nullb_queue *nq)
1598 {
1599 struct nullb_cmd *cmd;
1600 int i, tag_size;
1601
1602 nq->cmds = kcalloc(nq->queue_depth, sizeof(*cmd), GFP_KERNEL);
1603 if (!nq->cmds)
1604 return -ENOMEM;
1605
1606 tag_size = ALIGN(nq->queue_depth, BITS_PER_LONG) / BITS_PER_LONG;
1607 nq->tag_map = kcalloc(tag_size, sizeof(unsigned long), GFP_KERNEL);
1608 if (!nq->tag_map) {
1609 kfree(nq->cmds);
1610 return -ENOMEM;
1611 }
1612
1613 for (i = 0; i < nq->queue_depth; i++) {
1614 cmd = &nq->cmds[i];
1615 cmd->tag = -1U;
1616 }
1617
1618 return 0;
1619 }
1620
1621 static int setup_queues(struct nullb *nullb)
1622 {
1623 nullb->queues = kcalloc(nr_cpu_ids, sizeof(struct nullb_queue),
1624 GFP_KERNEL);
1625 if (!nullb->queues)
1626 return -ENOMEM;
1627
1628 nullb->queue_depth = nullb->dev->hw_queue_depth;
1629
1630 return 0;
1631 }
1632
1633 static int init_driver_queues(struct nullb *nullb)
1634 {
1635 struct nullb_queue *nq;
1636 int i, ret = 0;
1637
1638 for (i = 0; i < nullb->dev->submit_queues; i++) {
1639 nq = &nullb->queues[i];
1640
1641 null_init_queue(nullb, nq);
1642
1643 ret = setup_commands(nq);
1644 if (ret)
1645 return ret;
1646 nullb->nr_queues++;
1647 }
1648 return 0;
1649 }
1650
1651 static int null_gendisk_register(struct nullb *nullb)
1652 {
1653 sector_t size = ((sector_t)nullb->dev->size * SZ_1M) >> SECTOR_SHIFT;
1654 struct gendisk *disk;
1655
1656 disk = nullb->disk = alloc_disk_node(1, nullb->dev->home_node);
1657 if (!disk)
1658 return -ENOMEM;
1659 set_capacity(disk, size);
1660
1661 disk->flags |= GENHD_FL_EXT_DEVT | GENHD_FL_SUPPRESS_PARTITION_INFO;
1662 disk->major = null_major;
1663 disk->first_minor = nullb->index;
1664 if (queue_is_mq(nullb->q))
1665 disk->fops = &null_rq_ops;
1666 else
1667 disk->fops = &null_bio_ops;
1668 disk->private_data = nullb;
1669 disk->queue = nullb->q;
1670 strncpy(disk->disk_name, nullb->disk_name, DISK_NAME_LEN);
1671
1672 if (nullb->dev->zoned) {
1673 int ret = null_register_zoned_dev(nullb);
1674
1675 if (ret)
1676 return ret;
1677 }
1678
1679 add_disk(disk);
1680 return 0;
1681 }
1682
1683 static int null_init_tag_set(struct nullb *nullb, struct blk_mq_tag_set *set)
1684 {
1685 set->ops = &null_mq_ops;
1686 set->nr_hw_queues = nullb ? nullb->dev->submit_queues :
1687 g_submit_queues;
1688 set->queue_depth = nullb ? nullb->dev->hw_queue_depth :
1689 g_hw_queue_depth;
1690 set->numa_node = nullb ? nullb->dev->home_node : g_home_node;
1691 set->cmd_size = sizeof(struct nullb_cmd);
1692 set->flags = BLK_MQ_F_SHOULD_MERGE;
1693 if (g_no_sched)
1694 set->flags |= BLK_MQ_F_NO_SCHED;
1695 set->driver_data = NULL;
1696
1697 if ((nullb && nullb->dev->blocking) || g_blocking)
1698 set->flags |= BLK_MQ_F_BLOCKING;
1699
1700 return blk_mq_alloc_tag_set(set);
1701 }
1702
1703 static int null_validate_conf(struct nullb_device *dev)
1704 {
1705 dev->blocksize = round_down(dev->blocksize, 512);
1706 dev->blocksize = clamp_t(unsigned int, dev->blocksize, 512, 4096);
1707
1708 if (dev->queue_mode == NULL_Q_MQ && dev->use_per_node_hctx) {
1709 if (dev->submit_queues != nr_online_nodes)
1710 dev->submit_queues = nr_online_nodes;
1711 } else if (dev->submit_queues > nr_cpu_ids)
1712 dev->submit_queues = nr_cpu_ids;
1713 else if (dev->submit_queues == 0)
1714 dev->submit_queues = 1;
1715
1716 dev->queue_mode = min_t(unsigned int, dev->queue_mode, NULL_Q_MQ);
1717 dev->irqmode = min_t(unsigned int, dev->irqmode, NULL_IRQ_TIMER);
1718
1719 /* Do memory allocation, so set blocking */
1720 if (dev->memory_backed)
1721 dev->blocking = true;
1722 else /* cache is meaningless */
1723 dev->cache_size = 0;
1724 dev->cache_size = min_t(unsigned long, ULONG_MAX / 1024 / 1024,
1725 dev->cache_size);
1726 dev->mbps = min_t(unsigned int, 1024 * 40, dev->mbps);
1727 /* can not stop a queue */
1728 if (dev->queue_mode == NULL_Q_BIO)
1729 dev->mbps = 0;
1730
1731 if (dev->zoned &&
1732 (!dev->zone_size || !is_power_of_2(dev->zone_size))) {
1733 pr_err("zone_size must be power-of-two\n");
1734 return -EINVAL;
1735 }
1736
1737 return 0;
1738 }
1739
1740 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1741 static bool __null_setup_fault(struct fault_attr *attr, char *str)
1742 {
1743 if (!str[0])
1744 return true;
1745
1746 if (!setup_fault_attr(attr, str))
1747 return false;
1748
1749 attr->verbose = 0;
1750 return true;
1751 }
1752 #endif
1753
1754 static bool null_setup_fault(void)
1755 {
1756 #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION
1757 if (!__null_setup_fault(&null_timeout_attr, g_timeout_str))
1758 return false;
1759 if (!__null_setup_fault(&null_requeue_attr, g_requeue_str))
1760 return false;
1761 if (!__null_setup_fault(&null_init_hctx_attr, g_init_hctx_str))
1762 return false;
1763 #endif
1764 return true;
1765 }
1766
1767 static int null_add_dev(struct nullb_device *dev)
1768 {
1769 struct nullb *nullb;
1770 int rv;
1771
1772 rv = null_validate_conf(dev);
1773 if (rv)
1774 return rv;
1775
1776 nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, dev->home_node);
1777 if (!nullb) {
1778 rv = -ENOMEM;
1779 goto out;
1780 }
1781 nullb->dev = dev;
1782 dev->nullb = nullb;
1783
1784 spin_lock_init(&nullb->lock);
1785
1786 rv = setup_queues(nullb);
1787 if (rv)
1788 goto out_free_nullb;
1789
1790 if (dev->queue_mode == NULL_Q_MQ) {
1791 if (shared_tags) {
1792 nullb->tag_set = &tag_set;
1793 rv = 0;
1794 } else {
1795 nullb->tag_set = &nullb->__tag_set;
1796 rv = null_init_tag_set(nullb, nullb->tag_set);
1797 }
1798
1799 if (rv)
1800 goto out_cleanup_queues;
1801
1802 if (!null_setup_fault())
1803 goto out_cleanup_queues;
1804
1805 nullb->tag_set->timeout = 5 * HZ;
1806 nullb->q = blk_mq_init_queue_data(nullb->tag_set, nullb);
1807 if (IS_ERR(nullb->q)) {
1808 rv = -ENOMEM;
1809 goto out_cleanup_tags;
1810 }
1811 } else if (dev->queue_mode == NULL_Q_BIO) {
1812 nullb->q = blk_alloc_queue(dev->home_node);
1813 if (!nullb->q) {
1814 rv = -ENOMEM;
1815 goto out_cleanup_queues;
1816 }
1817 rv = init_driver_queues(nullb);
1818 if (rv)
1819 goto out_cleanup_blk_queue;
1820 }
1821
1822 if (dev->mbps) {
1823 set_bit(NULLB_DEV_FL_THROTTLED, &dev->flags);
1824 nullb_setup_bwtimer(nullb);
1825 }
1826
1827 if (dev->cache_size > 0) {
1828 set_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags);
1829 blk_queue_write_cache(nullb->q, true, true);
1830 }
1831
1832 if (dev->zoned) {
1833 rv = null_init_zoned_dev(dev, nullb->q);
1834 if (rv)
1835 goto out_cleanup_blk_queue;
1836 }
1837
1838 nullb->q->queuedata = nullb;
1839 blk_queue_flag_set(QUEUE_FLAG_NONROT, nullb->q);
1840 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, nullb->q);
1841
1842 mutex_lock(&lock);
1843 nullb->index = ida_simple_get(&nullb_indexes, 0, 0, GFP_KERNEL);
1844 dev->index = nullb->index;
1845 mutex_unlock(&lock);
1846
1847 blk_queue_logical_block_size(nullb->q, dev->blocksize);
1848 blk_queue_physical_block_size(nullb->q, dev->blocksize);
1849
1850 null_config_discard(nullb);
1851
1852 sprintf(nullb->disk_name, "nullb%d", nullb->index);
1853
1854 rv = null_gendisk_register(nullb);
1855 if (rv)
1856 goto out_cleanup_zone;
1857
1858 mutex_lock(&lock);
1859 list_add_tail(&nullb->list, &nullb_list);
1860 mutex_unlock(&lock);
1861
1862 return 0;
1863 out_cleanup_zone:
1864 null_free_zoned_dev(dev);
1865 out_cleanup_blk_queue:
1866 blk_cleanup_queue(nullb->q);
1867 out_cleanup_tags:
1868 if (dev->queue_mode == NULL_Q_MQ && nullb->tag_set == &nullb->__tag_set)
1869 blk_mq_free_tag_set(nullb->tag_set);
1870 out_cleanup_queues:
1871 cleanup_queues(nullb);
1872 out_free_nullb:
1873 kfree(nullb);
1874 dev->nullb = NULL;
1875 out:
1876 return rv;
1877 }
1878
1879 static int __init null_init(void)
1880 {
1881 int ret = 0;
1882 unsigned int i;
1883 struct nullb *nullb;
1884 struct nullb_device *dev;
1885
1886 if (g_bs > PAGE_SIZE) {
1887 pr_warn("invalid block size\n");
1888 pr_warn("defaults block size to %lu\n", PAGE_SIZE);
1889 g_bs = PAGE_SIZE;
1890 }
1891
1892 if (g_home_node != NUMA_NO_NODE && g_home_node >= nr_online_nodes) {
1893 pr_err("invalid home_node value\n");
1894 g_home_node = NUMA_NO_NODE;
1895 }
1896
1897 if (g_queue_mode == NULL_Q_RQ) {
1898 pr_err("legacy IO path no longer available\n");
1899 return -EINVAL;
1900 }
1901 if (g_queue_mode == NULL_Q_MQ && g_use_per_node_hctx) {
1902 if (g_submit_queues != nr_online_nodes) {
1903 pr_warn("submit_queues param is set to %u.\n",
1904 nr_online_nodes);
1905 g_submit_queues = nr_online_nodes;
1906 }
1907 } else if (g_submit_queues > nr_cpu_ids)
1908 g_submit_queues = nr_cpu_ids;
1909 else if (g_submit_queues <= 0)
1910 g_submit_queues = 1;
1911
1912 if (g_queue_mode == NULL_Q_MQ && shared_tags) {
1913 ret = null_init_tag_set(NULL, &tag_set);
1914 if (ret)
1915 return ret;
1916 }
1917
1918 config_group_init(&nullb_subsys.su_group);
1919 mutex_init(&nullb_subsys.su_mutex);
1920
1921 ret = configfs_register_subsystem(&nullb_subsys);
1922 if (ret)
1923 goto err_tagset;
1924
1925 mutex_init(&lock);
1926
1927 null_major = register_blkdev(0, "nullb");
1928 if (null_major < 0) {
1929 ret = null_major;
1930 goto err_conf;
1931 }
1932
1933 for (i = 0; i < nr_devices; i++) {
1934 dev = null_alloc_dev();
1935 if (!dev) {
1936 ret = -ENOMEM;
1937 goto err_dev;
1938 }
1939 ret = null_add_dev(dev);
1940 if (ret) {
1941 null_free_dev(dev);
1942 goto err_dev;
1943 }
1944 }
1945
1946 pr_info("module loaded\n");
1947 return 0;
1948
1949 err_dev:
1950 while (!list_empty(&nullb_list)) {
1951 nullb = list_entry(nullb_list.next, struct nullb, list);
1952 dev = nullb->dev;
1953 null_del_dev(nullb);
1954 null_free_dev(dev);
1955 }
1956 unregister_blkdev(null_major, "nullb");
1957 err_conf:
1958 configfs_unregister_subsystem(&nullb_subsys);
1959 err_tagset:
1960 if (g_queue_mode == NULL_Q_MQ && shared_tags)
1961 blk_mq_free_tag_set(&tag_set);
1962 return ret;
1963 }
1964
1965 static void __exit null_exit(void)
1966 {
1967 struct nullb *nullb;
1968
1969 configfs_unregister_subsystem(&nullb_subsys);
1970
1971 unregister_blkdev(null_major, "nullb");
1972
1973 mutex_lock(&lock);
1974 while (!list_empty(&nullb_list)) {
1975 struct nullb_device *dev;
1976
1977 nullb = list_entry(nullb_list.next, struct nullb, list);
1978 dev = nullb->dev;
1979 null_del_dev(nullb);
1980 null_free_dev(dev);
1981 }
1982 mutex_unlock(&lock);
1983
1984 if (g_queue_mode == NULL_Q_MQ && shared_tags)
1985 blk_mq_free_tag_set(&tag_set);
1986 }
1987
1988 module_init(null_init);
1989 module_exit(null_exit);
1990
1991 MODULE_AUTHOR("Jens Axboe <axboe@kernel.dk>");
1992 MODULE_LICENSE("GPL");
Linux with added FPC-III support