Commit | Line | Data |
---|---|---|
09c434b8 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
3bf2bd20 SL |
2 | /* |
3 | * Add configfs and memory store: Kyungchan Koh <kkc6196@fb.com> and | |
4 | * Shaohua Li <shli@fb.com> | |
5 | */ | |
f2298c04 | 6 | #include <linux/module.h> |
fc1bc354 | 7 | |
f2298c04 JA |
8 | #include <linux/moduleparam.h> |
9 | #include <linux/sched.h> | |
10 | #include <linux/fs.h> | |
f2298c04 | 11 | #include <linux/init.h> |
6dad38d3 | 12 | #include "null_blk.h" |
f2298c04 | 13 | |
db060f54 DLM |
14 | #undef pr_fmt |
15 | #define pr_fmt(fmt) "null_blk: " fmt | |
16 | ||
5bcd0e0c SL |
17 | #define FREE_BATCH 16 |
18 | ||
eff2c4f1 SL |
19 | #define TICKS_PER_SEC 50ULL |
20 | #define TIMER_INTERVAL (NSEC_PER_SEC / TICKS_PER_SEC) | |
21 | ||
33f782c4 | 22 | #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION |
93b57046 | 23 | static DECLARE_FAULT_ATTR(null_timeout_attr); |
24941b90 | 24 | static DECLARE_FAULT_ATTR(null_requeue_attr); |
596444e7 | 25 | static DECLARE_FAULT_ATTR(null_init_hctx_attr); |
33f782c4 | 26 | #endif |
93b57046 | 27 | |
eff2c4f1 SL |
28 | static inline u64 mb_per_tick(int mbps) |
29 | { | |
30 | return (1 << 20) / TICKS_PER_SEC * ((u64) mbps); | |
31 | } | |
f2298c04 | 32 | |
3bf2bd20 SL |
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. | |
eff2c4f1 | 38 | * THROTTLED: Device is being throttled. |
deb78b41 | 39 | * CACHE: Device is using a write-back cache. |
3bf2bd20 SL |
40 | */ |
41 | enum nullb_device_flags { | |
42 | NULLB_DEV_FL_CONFIGURED = 0, | |
43 | NULLB_DEV_FL_UP = 1, | |
eff2c4f1 | 44 | NULLB_DEV_FL_THROTTLED = 2, |
deb78b41 | 45 | NULLB_DEV_FL_CACHE = 3, |
3bf2bd20 SL |
46 | }; |
47 | ||
66231ad3 | 48 | #define MAP_SZ ((PAGE_SIZE >> SECTOR_SHIFT) + 2) |
5bcd0e0c SL |
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 | |
deb78b41 SL |
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 | |
5bcd0e0c SL |
60 | */ |
61 | struct nullb_page { | |
62 | struct page *page; | |
66231ad3 | 63 | DECLARE_BITMAP(bitmap, MAP_SZ); |
5bcd0e0c | 64 | }; |
66231ad3 ML |
65 | #define NULLB_PAGE_LOCK (MAP_SZ - 1) |
66 | #define NULLB_PAGE_FREE (MAP_SZ - 2) | |
5bcd0e0c | 67 | |
f2298c04 JA |
68 | static LIST_HEAD(nullb_list); |
69 | static struct mutex lock; | |
70 | static int null_major; | |
94bc02e3 | 71 | static DEFINE_IDA(nullb_indexes); |
82f402fe | 72 | static struct blk_mq_tag_set tag_set; |
f2298c04 | 73 | |
f2298c04 JA |
74 | enum { |
75 | NULL_IRQ_NONE = 0, | |
76 | NULL_IRQ_SOFTIRQ = 1, | |
77 | NULL_IRQ_TIMER = 2, | |
ce2c350b | 78 | }; |
f2298c04 | 79 | |
cee1b215 MG |
80 | static bool g_virt_boundary = false; |
81 | module_param_named(virt_boundary, g_virt_boundary, bool, 0444); | |
82 | MODULE_PARM_DESC(virt_boundary, "Require a virtual boundary for the device. Default: False"); | |
83 | ||
b3cffc38 | 84 | static int g_no_sched; |
5657a819 | 85 | module_param_named(no_sched, g_no_sched, int, 0444); |
b3cffc38 | 86 | MODULE_PARM_DESC(no_sched, "No io scheduler"); |
87 | ||
2984c868 | 88 | static int g_submit_queues = 1; |
5657a819 | 89 | module_param_named(submit_queues, g_submit_queues, int, 0444); |
f2298c04 JA |
90 | MODULE_PARM_DESC(submit_queues, "Number of submission queues"); |
91 | ||
0a593fbb JA |
92 | static int g_poll_queues = 1; |
93 | module_param_named(poll_queues, g_poll_queues, int, 0444); | |
94 | MODULE_PARM_DESC(poll_queues, "Number of IOPOLL submission queues"); | |
95 | ||
2984c868 | 96 | static int g_home_node = NUMA_NO_NODE; |
5657a819 | 97 | module_param_named(home_node, g_home_node, int, 0444); |
f2298c04 JA |
98 | MODULE_PARM_DESC(home_node, "Home node for the device"); |
99 | ||
33f782c4 | 100 | #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION |
290df92a DZ |
101 | /* |
102 | * For more details about fault injection, please refer to | |
103 | * Documentation/fault-injection/fault-injection.rst. | |
104 | */ | |
93b57046 | 105 | static char g_timeout_str[80]; |
5657a819 | 106 | module_param_string(timeout, g_timeout_str, sizeof(g_timeout_str), 0444); |
290df92a | 107 | MODULE_PARM_DESC(timeout, "Fault injection. timeout=<interval>,<probability>,<space>,<times>"); |
24941b90 JA |
108 | |
109 | static char g_requeue_str[80]; | |
5657a819 | 110 | module_param_string(requeue, g_requeue_str, sizeof(g_requeue_str), 0444); |
290df92a | 111 | MODULE_PARM_DESC(requeue, "Fault injection. requeue=<interval>,<probability>,<space>,<times>"); |
596444e7 BVA |
112 | |
113 | static char g_init_hctx_str[80]; | |
114 | module_param_string(init_hctx, g_init_hctx_str, sizeof(g_init_hctx_str), 0444); | |
290df92a | 115 | MODULE_PARM_DESC(init_hctx, "Fault injection to fail hctx init. init_hctx=<interval>,<probability>,<space>,<times>"); |
33f782c4 | 116 | #endif |
93b57046 | 117 | |
2984c868 | 118 | static int g_queue_mode = NULL_Q_MQ; |
709c8667 MB |
119 | |
120 | static int null_param_store_val(const char *str, int *val, int min, int max) | |
121 | { | |
122 | int ret, new_val; | |
123 | ||
124 | ret = kstrtoint(str, 10, &new_val); | |
125 | if (ret) | |
126 | return -EINVAL; | |
127 | ||
128 | if (new_val < min || new_val > max) | |
129 | return -EINVAL; | |
130 | ||
131 | *val = new_val; | |
132 | return 0; | |
133 | } | |
134 | ||
135 | static int null_set_queue_mode(const char *str, const struct kernel_param *kp) | |
136 | { | |
2984c868 | 137 | return null_param_store_val(str, &g_queue_mode, NULL_Q_BIO, NULL_Q_MQ); |
709c8667 MB |
138 | } |
139 | ||
9c27847d | 140 | static const struct kernel_param_ops null_queue_mode_param_ops = { |
709c8667 MB |
141 | .set = null_set_queue_mode, |
142 | .get = param_get_int, | |
143 | }; | |
144 | ||
5657a819 | 145 | device_param_cb(queue_mode, &null_queue_mode_param_ops, &g_queue_mode, 0444); |
54ae81cd | 146 | MODULE_PARM_DESC(queue_mode, "Block interface to use (0=bio,1=rq,2=multiqueue)"); |
f2298c04 | 147 | |
2984c868 | 148 | static int g_gb = 250; |
5657a819 | 149 | module_param_named(gb, g_gb, int, 0444); |
f2298c04 JA |
150 | MODULE_PARM_DESC(gb, "Size in GB"); |
151 | ||
2984c868 | 152 | static int g_bs = 512; |
5657a819 | 153 | module_param_named(bs, g_bs, int, 0444); |
f2298c04 JA |
154 | MODULE_PARM_DESC(bs, "Block size (in bytes)"); |
155 | ||
ea17fd35 DLM |
156 | static int g_max_sectors; |
157 | module_param_named(max_sectors, g_max_sectors, int, 0444); | |
158 | MODULE_PARM_DESC(max_sectors, "Maximum size of a command (in 512B sectors)"); | |
159 | ||
f7c4ce89 | 160 | static unsigned int nr_devices = 1; |
701dfc42 | 161 | module_param(nr_devices, uint, 0444); |
f2298c04 JA |
162 | MODULE_PARM_DESC(nr_devices, "Number of devices to register"); |
163 | ||
2984c868 | 164 | static bool g_blocking; |
5657a819 | 165 | module_param_named(blocking, g_blocking, bool, 0444); |
db5bcf87 JA |
166 | MODULE_PARM_DESC(blocking, "Register as a blocking blk-mq driver device"); |
167 | ||
82f402fe | 168 | static bool shared_tags; |
5657a819 | 169 | module_param(shared_tags, bool, 0444); |
82f402fe JA |
170 | MODULE_PARM_DESC(shared_tags, "Share tag set between devices for blk-mq"); |
171 | ||
0905053b JG |
172 | static bool g_shared_tag_bitmap; |
173 | module_param_named(shared_tag_bitmap, g_shared_tag_bitmap, bool, 0444); | |
174 | MODULE_PARM_DESC(shared_tag_bitmap, "Use shared tag bitmap for all submission queues for blk-mq"); | |
175 | ||
2984c868 | 176 | static int g_irqmode = NULL_IRQ_SOFTIRQ; |
709c8667 MB |
177 | |
178 | static int null_set_irqmode(const char *str, const struct kernel_param *kp) | |
179 | { | |
2984c868 | 180 | return null_param_store_val(str, &g_irqmode, NULL_IRQ_NONE, |
709c8667 MB |
181 | NULL_IRQ_TIMER); |
182 | } | |
183 | ||
9c27847d | 184 | static const struct kernel_param_ops null_irqmode_param_ops = { |
709c8667 MB |
185 | .set = null_set_irqmode, |
186 | .get = param_get_int, | |
187 | }; | |
188 | ||
5657a819 | 189 | device_param_cb(irqmode, &null_irqmode_param_ops, &g_irqmode, 0444); |
f2298c04 JA |
190 | MODULE_PARM_DESC(irqmode, "IRQ completion handler. 0-none, 1-softirq, 2-timer"); |
191 | ||
2984c868 | 192 | static unsigned long g_completion_nsec = 10000; |
5657a819 | 193 | module_param_named(completion_nsec, g_completion_nsec, ulong, 0444); |
f2298c04 JA |
194 | MODULE_PARM_DESC(completion_nsec, "Time in ns to complete a request in hardware. Default: 10,000ns"); |
195 | ||
2984c868 | 196 | static int g_hw_queue_depth = 64; |
5657a819 | 197 | module_param_named(hw_queue_depth, g_hw_queue_depth, int, 0444); |
f2298c04 JA |
198 | MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 64"); |
199 | ||
2984c868 | 200 | static bool g_use_per_node_hctx; |
5657a819 | 201 | module_param_named(use_per_node_hctx, g_use_per_node_hctx, bool, 0444); |
20005244 | 202 | MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware context queues. Default: false"); |
f2298c04 | 203 | |
058efe00 VF |
204 | static bool g_memory_backed; |
205 | module_param_named(memory_backed, g_memory_backed, bool, 0444); | |
206 | MODULE_PARM_DESC(memory_backed, "Create a memory-backed block device. Default: false"); | |
207 | ||
208 | static bool g_discard; | |
209 | module_param_named(discard, g_discard, bool, 0444); | |
210 | MODULE_PARM_DESC(discard, "Support discard operations (requires memory-backed null_blk device). Default: false"); | |
211 | ||
212 | static unsigned long g_cache_size; | |
213 | module_param_named(cache_size, g_cache_size, ulong, 0444); | |
214 | MODULE_PARM_DESC(mbps, "Cache size in MiB for memory-backed device. Default: 0 (none)"); | |
215 | ||
216 | static unsigned int g_mbps; | |
217 | module_param_named(mbps, g_mbps, uint, 0444); | |
218 | MODULE_PARM_DESC(mbps, "Limit maximum bandwidth (in MiB/s). Default: 0 (no limit)"); | |
219 | ||
ca4b2a01 MB |
220 | static bool g_zoned; |
221 | module_param_named(zoned, g_zoned, bool, S_IRUGO); | |
222 | MODULE_PARM_DESC(zoned, "Make device as a host-managed zoned block device. Default: false"); | |
223 | ||
224 | static unsigned long g_zone_size = 256; | |
225 | module_param_named(zone_size, g_zone_size, ulong, S_IRUGO); | |
226 | MODULE_PARM_DESC(zone_size, "Zone size in MB when block device is zoned. Must be power-of-two: Default: 256"); | |
227 | ||
089565fb AR |
228 | static unsigned long g_zone_capacity; |
229 | module_param_named(zone_capacity, g_zone_capacity, ulong, 0444); | |
230 | 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"); | |
231 | ||
ea2c18e1 MS |
232 | static unsigned int g_zone_nr_conv; |
233 | module_param_named(zone_nr_conv, g_zone_nr_conv, uint, 0444); | |
234 | MODULE_PARM_DESC(zone_nr_conv, "Number of conventional zones when block device is zoned. Default: 0"); | |
235 | ||
dc4d137e NC |
236 | static unsigned int g_zone_max_open; |
237 | module_param_named(zone_max_open, g_zone_max_open, uint, 0444); | |
238 | MODULE_PARM_DESC(zone_max_open, "Maximum number of open zones when block device is zoned. Default: 0 (no limit)"); | |
239 | ||
240 | static unsigned int g_zone_max_active; | |
241 | module_param_named(zone_max_active, g_zone_max_active, uint, 0444); | |
242 | MODULE_PARM_DESC(zone_max_active, "Maximum number of active zones when block device is zoned. Default: 0 (no limit)"); | |
243 | ||
3bf2bd20 SL |
244 | static struct nullb_device *null_alloc_dev(void); |
245 | static void null_free_dev(struct nullb_device *dev); | |
cedcafad SL |
246 | static void null_del_dev(struct nullb *nullb); |
247 | static int null_add_dev(struct nullb_device *dev); | |
49c3b926 | 248 | static struct nullb *null_find_dev_by_name(const char *name); |
deb78b41 | 249 | static void null_free_device_storage(struct nullb_device *dev, bool is_cache); |
3bf2bd20 SL |
250 | |
251 | static inline struct nullb_device *to_nullb_device(struct config_item *item) | |
252 | { | |
bb4c19e0 | 253 | return item ? container_of(to_config_group(item), struct nullb_device, group) : NULL; |
3bf2bd20 SL |
254 | } |
255 | ||
256 | static inline ssize_t nullb_device_uint_attr_show(unsigned int val, char *page) | |
257 | { | |
258 | return snprintf(page, PAGE_SIZE, "%u\n", val); | |
259 | } | |
260 | ||
261 | static inline ssize_t nullb_device_ulong_attr_show(unsigned long val, | |
262 | char *page) | |
263 | { | |
264 | return snprintf(page, PAGE_SIZE, "%lu\n", val); | |
265 | } | |
266 | ||
267 | static inline ssize_t nullb_device_bool_attr_show(bool val, char *page) | |
268 | { | |
269 | return snprintf(page, PAGE_SIZE, "%u\n", val); | |
270 | } | |
271 | ||
272 | static ssize_t nullb_device_uint_attr_store(unsigned int *val, | |
273 | const char *page, size_t count) | |
274 | { | |
275 | unsigned int tmp; | |
276 | int result; | |
277 | ||
278 | result = kstrtouint(page, 0, &tmp); | |
45919fbf | 279 | if (result < 0) |
3bf2bd20 SL |
280 | return result; |
281 | ||
282 | *val = tmp; | |
283 | return count; | |
284 | } | |
285 | ||
286 | static ssize_t nullb_device_ulong_attr_store(unsigned long *val, | |
287 | const char *page, size_t count) | |
288 | { | |
289 | int result; | |
290 | unsigned long tmp; | |
291 | ||
292 | result = kstrtoul(page, 0, &tmp); | |
45919fbf | 293 | if (result < 0) |
3bf2bd20 SL |
294 | return result; |
295 | ||
296 | *val = tmp; | |
297 | return count; | |
298 | } | |
299 | ||
300 | static ssize_t nullb_device_bool_attr_store(bool *val, const char *page, | |
301 | size_t count) | |
302 | { | |
303 | bool tmp; | |
304 | int result; | |
305 | ||
306 | result = kstrtobool(page, &tmp); | |
45919fbf | 307 | if (result < 0) |
3bf2bd20 SL |
308 | return result; |
309 | ||
310 | *val = tmp; | |
311 | return count; | |
312 | } | |
313 | ||
314 | /* The following macro should only be used with TYPE = {uint, ulong, bool}. */ | |
ca0a95a6 AM |
315 | #define NULLB_DEVICE_ATTR(NAME, TYPE, APPLY) \ |
316 | static ssize_t \ | |
317 | nullb_device_##NAME##_show(struct config_item *item, char *page) \ | |
318 | { \ | |
319 | return nullb_device_##TYPE##_attr_show( \ | |
320 | to_nullb_device(item)->NAME, page); \ | |
321 | } \ | |
322 | static ssize_t \ | |
323 | nullb_device_##NAME##_store(struct config_item *item, const char *page, \ | |
324 | size_t count) \ | |
325 | { \ | |
326 | int (*apply_fn)(struct nullb_device *dev, TYPE new_value) = APPLY;\ | |
327 | struct nullb_device *dev = to_nullb_device(item); \ | |
b9853b4d | 328 | TYPE new_value = 0; \ |
ca0a95a6 AM |
329 | int ret; \ |
330 | \ | |
331 | ret = nullb_device_##TYPE##_attr_store(&new_value, page, count);\ | |
332 | if (ret < 0) \ | |
333 | return ret; \ | |
334 | if (apply_fn) \ | |
335 | ret = apply_fn(dev, new_value); \ | |
336 | else if (test_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags)) \ | |
337 | ret = -EBUSY; \ | |
338 | if (ret < 0) \ | |
339 | return ret; \ | |
340 | dev->NAME = new_value; \ | |
341 | return count; \ | |
342 | } \ | |
3bf2bd20 SL |
343 | CONFIGFS_ATTR(nullb_device_, NAME); |
344 | ||
15dfc662 SK |
345 | static int nullb_update_nr_hw_queues(struct nullb_device *dev, |
346 | unsigned int submit_queues, | |
347 | unsigned int poll_queues) | |
348 | ||
45919fbf | 349 | { |
45919fbf | 350 | struct blk_mq_tag_set *set; |
15dfc662 | 351 | int ret, nr_hw_queues; |
45919fbf | 352 | |
15dfc662 | 353 | if (!dev->nullb) |
45919fbf BVA |
354 | return 0; |
355 | ||
15dfc662 | 356 | /* |
2bfdbe8b | 357 | * Make sure at least one submit queue exists. |
15dfc662 | 358 | */ |
2bfdbe8b | 359 | if (!submit_queues) |
15dfc662 SK |
360 | return -EINVAL; |
361 | ||
78b10be2 BVA |
362 | /* |
363 | * Make sure that null_init_hctx() does not access nullb->queues[] past | |
364 | * the end of that array. | |
365 | */ | |
15dfc662 | 366 | if (submit_queues > nr_cpu_ids || poll_queues > g_poll_queues) |
78b10be2 | 367 | return -EINVAL; |
15dfc662 SK |
368 | |
369 | /* | |
370 | * Keep previous and new queue numbers in nullb_device for reference in | |
371 | * the call back function null_map_queues(). | |
372 | */ | |
373 | dev->prev_submit_queues = dev->submit_queues; | |
374 | dev->prev_poll_queues = dev->poll_queues; | |
375 | dev->submit_queues = submit_queues; | |
376 | dev->poll_queues = poll_queues; | |
377 | ||
378 | set = dev->nullb->tag_set; | |
379 | nr_hw_queues = submit_queues + poll_queues; | |
380 | blk_mq_update_nr_hw_queues(set, nr_hw_queues); | |
381 | ret = set->nr_hw_queues == nr_hw_queues ? 0 : -ENOMEM; | |
382 | ||
383 | if (ret) { | |
384 | /* on error, revert the queue numbers */ | |
385 | dev->submit_queues = dev->prev_submit_queues; | |
386 | dev->poll_queues = dev->prev_poll_queues; | |
387 | } | |
388 | ||
389 | return ret; | |
390 | } | |
391 | ||
392 | static int nullb_apply_submit_queues(struct nullb_device *dev, | |
393 | unsigned int submit_queues) | |
394 | { | |
395 | return nullb_update_nr_hw_queues(dev, submit_queues, dev->poll_queues); | |
396 | } | |
397 | ||
398 | static int nullb_apply_poll_queues(struct nullb_device *dev, | |
399 | unsigned int poll_queues) | |
400 | { | |
401 | return nullb_update_nr_hw_queues(dev, dev->submit_queues, poll_queues); | |
45919fbf BVA |
402 | } |
403 | ||
404 | NULLB_DEVICE_ATTR(size, ulong, NULL); | |
405 | NULLB_DEVICE_ATTR(completion_nsec, ulong, NULL); | |
406 | NULLB_DEVICE_ATTR(submit_queues, uint, nullb_apply_submit_queues); | |
15dfc662 | 407 | NULLB_DEVICE_ATTR(poll_queues, uint, nullb_apply_poll_queues); |
45919fbf BVA |
408 | NULLB_DEVICE_ATTR(home_node, uint, NULL); |
409 | NULLB_DEVICE_ATTR(queue_mode, uint, NULL); | |
410 | NULLB_DEVICE_ATTR(blocksize, uint, NULL); | |
ea17fd35 | 411 | NULLB_DEVICE_ATTR(max_sectors, uint, NULL); |
45919fbf BVA |
412 | NULLB_DEVICE_ATTR(irqmode, uint, NULL); |
413 | NULLB_DEVICE_ATTR(hw_queue_depth, uint, NULL); | |
414 | NULLB_DEVICE_ATTR(index, uint, NULL); | |
415 | NULLB_DEVICE_ATTR(blocking, bool, NULL); | |
416 | NULLB_DEVICE_ATTR(use_per_node_hctx, bool, NULL); | |
417 | NULLB_DEVICE_ATTR(memory_backed, bool, NULL); | |
418 | NULLB_DEVICE_ATTR(discard, bool, NULL); | |
419 | NULLB_DEVICE_ATTR(mbps, uint, NULL); | |
420 | NULLB_DEVICE_ATTR(cache_size, ulong, NULL); | |
421 | NULLB_DEVICE_ATTR(zoned, bool, NULL); | |
422 | NULLB_DEVICE_ATTR(zone_size, ulong, NULL); | |
089565fb | 423 | NULLB_DEVICE_ATTR(zone_capacity, ulong, NULL); |
45919fbf | 424 | NULLB_DEVICE_ATTR(zone_nr_conv, uint, NULL); |
dc4d137e NC |
425 | NULLB_DEVICE_ATTR(zone_max_open, uint, NULL); |
426 | NULLB_DEVICE_ATTR(zone_max_active, uint, NULL); | |
cee1b215 | 427 | NULLB_DEVICE_ATTR(virt_boundary, bool, NULL); |
7012eef5 VF |
428 | NULLB_DEVICE_ATTR(no_sched, bool, NULL); |
429 | NULLB_DEVICE_ATTR(shared_tag_bitmap, bool, NULL); | |
3bf2bd20 | 430 | |
cedcafad SL |
431 | static ssize_t nullb_device_power_show(struct config_item *item, char *page) |
432 | { | |
433 | return nullb_device_bool_attr_show(to_nullb_device(item)->power, page); | |
434 | } | |
435 | ||
436 | static ssize_t nullb_device_power_store(struct config_item *item, | |
437 | const char *page, size_t count) | |
438 | { | |
439 | struct nullb_device *dev = to_nullb_device(item); | |
440 | bool newp = false; | |
441 | ssize_t ret; | |
442 | ||
443 | ret = nullb_device_bool_attr_store(&newp, page, count); | |
444 | if (ret < 0) | |
445 | return ret; | |
446 | ||
447 | if (!dev->power && newp) { | |
448 | if (test_and_set_bit(NULLB_DEV_FL_UP, &dev->flags)) | |
449 | return count; | |
a75110c3 CK |
450 | ret = null_add_dev(dev); |
451 | if (ret) { | |
cedcafad | 452 | clear_bit(NULLB_DEV_FL_UP, &dev->flags); |
a75110c3 | 453 | return ret; |
cedcafad SL |
454 | } |
455 | ||
456 | set_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags); | |
457 | dev->power = newp; | |
b3c30512 | 458 | } else if (dev->power && !newp) { |
7602843f BL |
459 | if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) { |
460 | mutex_lock(&lock); | |
461 | dev->power = newp; | |
462 | null_del_dev(dev->nullb); | |
463 | mutex_unlock(&lock); | |
464 | } | |
00a8cdb8 | 465 | clear_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags); |
cedcafad SL |
466 | } |
467 | ||
468 | return count; | |
469 | } | |
470 | ||
471 | CONFIGFS_ATTR(nullb_device_, power); | |
472 | ||
2f54a613 SL |
473 | static ssize_t nullb_device_badblocks_show(struct config_item *item, char *page) |
474 | { | |
475 | struct nullb_device *t_dev = to_nullb_device(item); | |
476 | ||
477 | return badblocks_show(&t_dev->badblocks, page, 0); | |
478 | } | |
479 | ||
480 | static ssize_t nullb_device_badblocks_store(struct config_item *item, | |
481 | const char *page, size_t count) | |
482 | { | |
483 | struct nullb_device *t_dev = to_nullb_device(item); | |
484 | char *orig, *buf, *tmp; | |
485 | u64 start, end; | |
486 | int ret; | |
487 | ||
488 | orig = kstrndup(page, count, GFP_KERNEL); | |
489 | if (!orig) | |
490 | return -ENOMEM; | |
491 | ||
492 | buf = strstrip(orig); | |
493 | ||
494 | ret = -EINVAL; | |
495 | if (buf[0] != '+' && buf[0] != '-') | |
496 | goto out; | |
497 | tmp = strchr(&buf[1], '-'); | |
498 | if (!tmp) | |
499 | goto out; | |
500 | *tmp = '\0'; | |
501 | ret = kstrtoull(buf + 1, 0, &start); | |
502 | if (ret) | |
503 | goto out; | |
504 | ret = kstrtoull(tmp + 1, 0, &end); | |
505 | if (ret) | |
506 | goto out; | |
507 | ret = -EINVAL; | |
508 | if (start > end) | |
509 | goto out; | |
510 | /* enable badblocks */ | |
511 | cmpxchg(&t_dev->badblocks.shift, -1, 0); | |
512 | if (buf[0] == '+') | |
513 | ret = badblocks_set(&t_dev->badblocks, start, | |
514 | end - start + 1, 1); | |
515 | else | |
516 | ret = badblocks_clear(&t_dev->badblocks, start, | |
517 | end - start + 1); | |
518 | if (ret == 0) | |
519 | ret = count; | |
520 | out: | |
521 | kfree(orig); | |
522 | return ret; | |
523 | } | |
524 | CONFIGFS_ATTR(nullb_device_, badblocks); | |
525 | ||
d3a57388 SK |
526 | static ssize_t nullb_device_zone_readonly_store(struct config_item *item, |
527 | const char *page, size_t count) | |
528 | { | |
529 | struct nullb_device *dev = to_nullb_device(item); | |
530 | ||
531 | return zone_cond_store(dev, page, count, BLK_ZONE_COND_READONLY); | |
532 | } | |
533 | CONFIGFS_ATTR_WO(nullb_device_, zone_readonly); | |
534 | ||
535 | static ssize_t nullb_device_zone_offline_store(struct config_item *item, | |
536 | const char *page, size_t count) | |
537 | { | |
538 | struct nullb_device *dev = to_nullb_device(item); | |
539 | ||
540 | return zone_cond_store(dev, page, count, BLK_ZONE_COND_OFFLINE); | |
541 | } | |
542 | CONFIGFS_ATTR_WO(nullb_device_, zone_offline); | |
543 | ||
3bf2bd20 SL |
544 | static struct configfs_attribute *nullb_device_attrs[] = { |
545 | &nullb_device_attr_size, | |
546 | &nullb_device_attr_completion_nsec, | |
547 | &nullb_device_attr_submit_queues, | |
0a593fbb | 548 | &nullb_device_attr_poll_queues, |
3bf2bd20 SL |
549 | &nullb_device_attr_home_node, |
550 | &nullb_device_attr_queue_mode, | |
551 | &nullb_device_attr_blocksize, | |
ea17fd35 | 552 | &nullb_device_attr_max_sectors, |
3bf2bd20 SL |
553 | &nullb_device_attr_irqmode, |
554 | &nullb_device_attr_hw_queue_depth, | |
cedcafad | 555 | &nullb_device_attr_index, |
3bf2bd20 SL |
556 | &nullb_device_attr_blocking, |
557 | &nullb_device_attr_use_per_node_hctx, | |
cedcafad | 558 | &nullb_device_attr_power, |
5bcd0e0c | 559 | &nullb_device_attr_memory_backed, |
306eb6b4 | 560 | &nullb_device_attr_discard, |
eff2c4f1 | 561 | &nullb_device_attr_mbps, |
deb78b41 | 562 | &nullb_device_attr_cache_size, |
2f54a613 | 563 | &nullb_device_attr_badblocks, |
ca4b2a01 MB |
564 | &nullb_device_attr_zoned, |
565 | &nullb_device_attr_zone_size, | |
089565fb | 566 | &nullb_device_attr_zone_capacity, |
ea2c18e1 | 567 | &nullb_device_attr_zone_nr_conv, |
dc4d137e NC |
568 | &nullb_device_attr_zone_max_open, |
569 | &nullb_device_attr_zone_max_active, | |
d3a57388 SK |
570 | &nullb_device_attr_zone_readonly, |
571 | &nullb_device_attr_zone_offline, | |
cee1b215 | 572 | &nullb_device_attr_virt_boundary, |
7012eef5 VF |
573 | &nullb_device_attr_no_sched, |
574 | &nullb_device_attr_shared_tag_bitmap, | |
3bf2bd20 SL |
575 | NULL, |
576 | }; | |
577 | ||
578 | static void nullb_device_release(struct config_item *item) | |
579 | { | |
5bcd0e0c SL |
580 | struct nullb_device *dev = to_nullb_device(item); |
581 | ||
deb78b41 | 582 | null_free_device_storage(dev, false); |
5bcd0e0c | 583 | null_free_dev(dev); |
3bf2bd20 SL |
584 | } |
585 | ||
586 | static struct configfs_item_operations nullb_device_ops = { | |
587 | .release = nullb_device_release, | |
588 | }; | |
589 | ||
e1919dff | 590 | static const struct config_item_type nullb_device_type = { |
3bf2bd20 SL |
591 | .ct_item_ops = &nullb_device_ops, |
592 | .ct_attrs = nullb_device_attrs, | |
593 | .ct_owner = THIS_MODULE, | |
594 | }; | |
595 | ||
bb4c19e0 AM |
596 | #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION |
597 | ||
598 | static void nullb_add_fault_config(struct nullb_device *dev) | |
599 | { | |
600 | fault_config_init(&dev->timeout_config, "timeout_inject"); | |
601 | fault_config_init(&dev->requeue_config, "requeue_inject"); | |
602 | fault_config_init(&dev->init_hctx_fault_config, "init_hctx_fault_inject"); | |
603 | ||
604 | configfs_add_default_group(&dev->timeout_config.group, &dev->group); | |
605 | configfs_add_default_group(&dev->requeue_config.group, &dev->group); | |
606 | configfs_add_default_group(&dev->init_hctx_fault_config.group, &dev->group); | |
607 | } | |
608 | ||
609 | #else | |
610 | ||
611 | static void nullb_add_fault_config(struct nullb_device *dev) | |
612 | { | |
613 | } | |
614 | ||
615 | #endif | |
616 | ||
3bf2bd20 | 617 | static struct |
bb4c19e0 | 618 | config_group *nullb_group_make_group(struct config_group *group, const char *name) |
3bf2bd20 SL |
619 | { |
620 | struct nullb_device *dev; | |
621 | ||
49c3b926 DLM |
622 | if (null_find_dev_by_name(name)) |
623 | return ERR_PTR(-EEXIST); | |
624 | ||
3bf2bd20 SL |
625 | dev = null_alloc_dev(); |
626 | if (!dev) | |
627 | return ERR_PTR(-ENOMEM); | |
628 | ||
bb4c19e0 AM |
629 | config_group_init_type_name(&dev->group, name, &nullb_device_type); |
630 | nullb_add_fault_config(dev); | |
3bf2bd20 | 631 | |
bb4c19e0 | 632 | return &dev->group; |
3bf2bd20 SL |
633 | } |
634 | ||
635 | static void | |
636 | nullb_group_drop_item(struct config_group *group, struct config_item *item) | |
637 | { | |
cedcafad SL |
638 | struct nullb_device *dev = to_nullb_device(item); |
639 | ||
640 | if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) { | |
641 | mutex_lock(&lock); | |
642 | dev->power = false; | |
643 | null_del_dev(dev->nullb); | |
644 | mutex_unlock(&lock); | |
645 | } | |
646 | ||
3bf2bd20 SL |
647 | config_item_put(item); |
648 | } | |
649 | ||
650 | static ssize_t memb_group_features_show(struct config_item *item, char *page) | |
651 | { | |
089565fb | 652 | return snprintf(page, PAGE_SIZE, |
7012eef5 VF |
653 | "badblocks,blocking,blocksize,cache_size," |
654 | "completion_nsec,discard,home_node,hw_queue_depth," | |
655 | "irqmode,max_sectors,mbps,memory_backed,no_sched," | |
656 | "poll_queues,power,queue_mode,shared_tag_bitmap,size," | |
657 | "submit_queues,use_per_node_hctx,virt_boundary,zoned," | |
658 | "zone_capacity,zone_max_active,zone_max_open," | |
d3a57388 | 659 | "zone_nr_conv,zone_offline,zone_readonly,zone_size\n"); |
3bf2bd20 SL |
660 | } |
661 | ||
662 | CONFIGFS_ATTR_RO(memb_group_, features); | |
663 | ||
664 | static struct configfs_attribute *nullb_group_attrs[] = { | |
665 | &memb_group_attr_features, | |
666 | NULL, | |
667 | }; | |
668 | ||
669 | static struct configfs_group_operations nullb_group_ops = { | |
bb4c19e0 | 670 | .make_group = nullb_group_make_group, |
3bf2bd20 SL |
671 | .drop_item = nullb_group_drop_item, |
672 | }; | |
673 | ||
e1919dff | 674 | static const struct config_item_type nullb_group_type = { |
3bf2bd20 SL |
675 | .ct_group_ops = &nullb_group_ops, |
676 | .ct_attrs = nullb_group_attrs, | |
677 | .ct_owner = THIS_MODULE, | |
678 | }; | |
679 | ||
680 | static struct configfs_subsystem nullb_subsys = { | |
681 | .su_group = { | |
682 | .cg_item = { | |
683 | .ci_namebuf = "nullb", | |
684 | .ci_type = &nullb_group_type, | |
685 | }, | |
686 | }, | |
687 | }; | |
688 | ||
deb78b41 SL |
689 | static inline int null_cache_active(struct nullb *nullb) |
690 | { | |
691 | return test_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags); | |
692 | } | |
693 | ||
2984c868 SL |
694 | static struct nullb_device *null_alloc_dev(void) |
695 | { | |
696 | struct nullb_device *dev; | |
697 | ||
698 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); | |
699 | if (!dev) | |
700 | return NULL; | |
bb4c19e0 AM |
701 | |
702 | #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION | |
703 | dev->timeout_config.attr = null_timeout_attr; | |
704 | dev->requeue_config.attr = null_requeue_attr; | |
705 | dev->init_hctx_fault_config.attr = null_init_hctx_attr; | |
706 | #endif | |
707 | ||
5bcd0e0c | 708 | INIT_RADIX_TREE(&dev->data, GFP_ATOMIC); |
deb78b41 | 709 | INIT_RADIX_TREE(&dev->cache, GFP_ATOMIC); |
2f54a613 SL |
710 | if (badblocks_init(&dev->badblocks, 0)) { |
711 | kfree(dev); | |
712 | return NULL; | |
713 | } | |
714 | ||
2984c868 SL |
715 | dev->size = g_gb * 1024; |
716 | dev->completion_nsec = g_completion_nsec; | |
717 | dev->submit_queues = g_submit_queues; | |
15dfc662 | 718 | dev->prev_submit_queues = g_submit_queues; |
0a593fbb | 719 | dev->poll_queues = g_poll_queues; |
15dfc662 | 720 | dev->prev_poll_queues = g_poll_queues; |
2984c868 SL |
721 | dev->home_node = g_home_node; |
722 | dev->queue_mode = g_queue_mode; | |
723 | dev->blocksize = g_bs; | |
ea17fd35 | 724 | dev->max_sectors = g_max_sectors; |
2984c868 SL |
725 | dev->irqmode = g_irqmode; |
726 | dev->hw_queue_depth = g_hw_queue_depth; | |
2984c868 | 727 | dev->blocking = g_blocking; |
058efe00 VF |
728 | dev->memory_backed = g_memory_backed; |
729 | dev->discard = g_discard; | |
730 | dev->cache_size = g_cache_size; | |
731 | dev->mbps = g_mbps; | |
2984c868 | 732 | dev->use_per_node_hctx = g_use_per_node_hctx; |
ca4b2a01 MB |
733 | dev->zoned = g_zoned; |
734 | dev->zone_size = g_zone_size; | |
089565fb | 735 | dev->zone_capacity = g_zone_capacity; |
ea2c18e1 | 736 | dev->zone_nr_conv = g_zone_nr_conv; |
dc4d137e NC |
737 | dev->zone_max_open = g_zone_max_open; |
738 | dev->zone_max_active = g_zone_max_active; | |
cee1b215 | 739 | dev->virt_boundary = g_virt_boundary; |
7012eef5 VF |
740 | dev->no_sched = g_no_sched; |
741 | dev->shared_tag_bitmap = g_shared_tag_bitmap; | |
2984c868 SL |
742 | return dev; |
743 | } | |
744 | ||
745 | static void null_free_dev(struct nullb_device *dev) | |
746 | { | |
1addb798 DD |
747 | if (!dev) |
748 | return; | |
749 | ||
d205bde7 | 750 | null_free_zoned_dev(dev); |
1addb798 | 751 | badblocks_exit(&dev->badblocks); |
2984c868 SL |
752 | kfree(dev); |
753 | } | |
754 | ||
f2298c04 JA |
755 | static void put_tag(struct nullb_queue *nq, unsigned int tag) |
756 | { | |
757 | clear_bit_unlock(tag, nq->tag_map); | |
758 | ||
759 | if (waitqueue_active(&nq->wait)) | |
760 | wake_up(&nq->wait); | |
761 | } | |
762 | ||
763 | static unsigned int get_tag(struct nullb_queue *nq) | |
764 | { | |
765 | unsigned int tag; | |
766 | ||
767 | do { | |
768 | tag = find_first_zero_bit(nq->tag_map, nq->queue_depth); | |
769 | if (tag >= nq->queue_depth) | |
770 | return -1U; | |
771 | } while (test_and_set_bit_lock(tag, nq->tag_map)); | |
772 | ||
773 | return tag; | |
774 | } | |
775 | ||
776 | static void free_cmd(struct nullb_cmd *cmd) | |
777 | { | |
778 | put_tag(cmd->nq, cmd->tag); | |
779 | } | |
780 | ||
3c395a96 PV |
781 | static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer); |
782 | ||
f2298c04 JA |
783 | static struct nullb_cmd *__alloc_cmd(struct nullb_queue *nq) |
784 | { | |
785 | struct nullb_cmd *cmd; | |
786 | unsigned int tag; | |
787 | ||
788 | tag = get_tag(nq); | |
789 | if (tag != -1U) { | |
790 | cmd = &nq->cmds[tag]; | |
791 | cmd->tag = tag; | |
ff770422 | 792 | cmd->error = BLK_STS_OK; |
f2298c04 | 793 | cmd->nq = nq; |
2984c868 | 794 | if (nq->dev->irqmode == NULL_IRQ_TIMER) { |
3c395a96 PV |
795 | hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, |
796 | HRTIMER_MODE_REL); | |
797 | cmd->timer.function = null_cmd_timer_expired; | |
798 | } | |
f2298c04 JA |
799 | return cmd; |
800 | } | |
801 | ||
802 | return NULL; | |
803 | } | |
804 | ||
3d3472f3 | 805 | static struct nullb_cmd *alloc_cmd(struct nullb_queue *nq, struct bio *bio) |
f2298c04 JA |
806 | { |
807 | struct nullb_cmd *cmd; | |
808 | DEFINE_WAIT(wait); | |
809 | ||
f2298c04 | 810 | do { |
3d3472f3 CK |
811 | /* |
812 | * This avoids multiple return statements, multiple calls to | |
813 | * __alloc_cmd() and a fast path call to prepare_to_wait(). | |
814 | */ | |
f2298c04 | 815 | cmd = __alloc_cmd(nq); |
3d3472f3 CK |
816 | if (cmd) { |
817 | cmd->bio = bio; | |
818 | return cmd; | |
819 | } | |
820 | prepare_to_wait(&nq->wait, &wait, TASK_UNINTERRUPTIBLE); | |
f2298c04 | 821 | io_schedule(); |
3d3472f3 | 822 | finish_wait(&nq->wait, &wait); |
f2298c04 | 823 | } while (1); |
f2298c04 JA |
824 | } |
825 | ||
826 | static void end_cmd(struct nullb_cmd *cmd) | |
827 | { | |
2984c868 | 828 | int queue_mode = cmd->nq->dev->queue_mode; |
cf8ecc5a | 829 | |
ce2c350b CH |
830 | switch (queue_mode) { |
831 | case NULL_Q_MQ: | |
5bcd0e0c | 832 | blk_mq_end_request(cmd->rq, cmd->error); |
ce2c350b | 833 | return; |
ce2c350b | 834 | case NULL_Q_BIO: |
5bcd0e0c | 835 | cmd->bio->bi_status = cmd->error; |
4246a0b6 | 836 | bio_endio(cmd->bio); |
48cc661e | 837 | break; |
ce2c350b | 838 | } |
f2298c04 | 839 | |
48cc661e | 840 | free_cmd(cmd); |
cf8ecc5a AA |
841 | } |
842 | ||
843 | static enum hrtimer_restart null_cmd_timer_expired(struct hrtimer *timer) | |
844 | { | |
845 | end_cmd(container_of(timer, struct nullb_cmd, timer)); | |
f2298c04 JA |
846 | |
847 | return HRTIMER_NORESTART; | |
848 | } | |
849 | ||
850 | static void null_cmd_end_timer(struct nullb_cmd *cmd) | |
851 | { | |
2984c868 | 852 | ktime_t kt = cmd->nq->dev->completion_nsec; |
f2298c04 | 853 | |
3c395a96 | 854 | hrtimer_start(&cmd->timer, kt, HRTIMER_MODE_REL); |
f2298c04 JA |
855 | } |
856 | ||
49f66136 | 857 | static void null_complete_rq(struct request *rq) |
f2298c04 | 858 | { |
49f66136 | 859 | end_cmd(blk_mq_rq_to_pdu(rq)); |
f2298c04 JA |
860 | } |
861 | ||
c90b6b50 | 862 | static struct nullb_page *null_alloc_page(void) |
5bcd0e0c SL |
863 | { |
864 | struct nullb_page *t_page; | |
865 | ||
c90b6b50 | 866 | t_page = kmalloc(sizeof(struct nullb_page), GFP_NOIO); |
5bcd0e0c | 867 | if (!t_page) |
df00b1d2 | 868 | return NULL; |
5bcd0e0c | 869 | |
c90b6b50 | 870 | t_page->page = alloc_pages(GFP_NOIO, 0); |
df00b1d2 CK |
871 | if (!t_page->page) { |
872 | kfree(t_page); | |
873 | return NULL; | |
874 | } | |
5bcd0e0c | 875 | |
66231ad3 | 876 | memset(t_page->bitmap, 0, sizeof(t_page->bitmap)); |
5bcd0e0c | 877 | return t_page; |
5bcd0e0c SL |
878 | } |
879 | ||
880 | static void null_free_page(struct nullb_page *t_page) | |
881 | { | |
66231ad3 ML |
882 | __set_bit(NULLB_PAGE_FREE, t_page->bitmap); |
883 | if (test_bit(NULLB_PAGE_LOCK, t_page->bitmap)) | |
deb78b41 | 884 | return; |
5bcd0e0c SL |
885 | __free_page(t_page->page); |
886 | kfree(t_page); | |
887 | } | |
888 | ||
66231ad3 ML |
889 | static bool null_page_empty(struct nullb_page *page) |
890 | { | |
891 | int size = MAP_SZ - 2; | |
892 | ||
893 | return find_first_bit(page->bitmap, size) == size; | |
894 | } | |
895 | ||
deb78b41 SL |
896 | static void null_free_sector(struct nullb *nullb, sector_t sector, |
897 | bool is_cache) | |
5bcd0e0c SL |
898 | { |
899 | unsigned int sector_bit; | |
900 | u64 idx; | |
901 | struct nullb_page *t_page, *ret; | |
902 | struct radix_tree_root *root; | |
903 | ||
deb78b41 | 904 | root = is_cache ? &nullb->dev->cache : &nullb->dev->data; |
5bcd0e0c SL |
905 | idx = sector >> PAGE_SECTORS_SHIFT; |
906 | sector_bit = (sector & SECTOR_MASK); | |
907 | ||
908 | t_page = radix_tree_lookup(root, idx); | |
909 | if (t_page) { | |
66231ad3 | 910 | __clear_bit(sector_bit, t_page->bitmap); |
5bcd0e0c | 911 | |
66231ad3 | 912 | if (null_page_empty(t_page)) { |
5bcd0e0c SL |
913 | ret = radix_tree_delete_item(root, idx, t_page); |
914 | WARN_ON(ret != t_page); | |
915 | null_free_page(ret); | |
deb78b41 SL |
916 | if (is_cache) |
917 | nullb->dev->curr_cache -= PAGE_SIZE; | |
5bcd0e0c SL |
918 | } |
919 | } | |
920 | } | |
921 | ||
922 | static struct nullb_page *null_radix_tree_insert(struct nullb *nullb, u64 idx, | |
deb78b41 | 923 | struct nullb_page *t_page, bool is_cache) |
5bcd0e0c SL |
924 | { |
925 | struct radix_tree_root *root; | |
926 | ||
deb78b41 | 927 | root = is_cache ? &nullb->dev->cache : &nullb->dev->data; |
5bcd0e0c SL |
928 | |
929 | if (radix_tree_insert(root, idx, t_page)) { | |
930 | null_free_page(t_page); | |
931 | t_page = radix_tree_lookup(root, idx); | |
932 | WARN_ON(!t_page || t_page->page->index != idx); | |
deb78b41 SL |
933 | } else if (is_cache) |
934 | nullb->dev->curr_cache += PAGE_SIZE; | |
5bcd0e0c SL |
935 | |
936 | return t_page; | |
937 | } | |
938 | ||
deb78b41 | 939 | static void null_free_device_storage(struct nullb_device *dev, bool is_cache) |
5bcd0e0c SL |
940 | { |
941 | unsigned long pos = 0; | |
942 | int nr_pages; | |
943 | struct nullb_page *ret, *t_pages[FREE_BATCH]; | |
944 | struct radix_tree_root *root; | |
945 | ||
deb78b41 | 946 | root = is_cache ? &dev->cache : &dev->data; |
5bcd0e0c SL |
947 | |
948 | do { | |
949 | int i; | |
950 | ||
951 | nr_pages = radix_tree_gang_lookup(root, | |
952 | (void **)t_pages, pos, FREE_BATCH); | |
953 | ||
954 | for (i = 0; i < nr_pages; i++) { | |
955 | pos = t_pages[i]->page->index; | |
956 | ret = radix_tree_delete_item(root, pos, t_pages[i]); | |
957 | WARN_ON(ret != t_pages[i]); | |
958 | null_free_page(ret); | |
959 | } | |
960 | ||
961 | pos++; | |
962 | } while (nr_pages == FREE_BATCH); | |
deb78b41 SL |
963 | |
964 | if (is_cache) | |
965 | dev->curr_cache = 0; | |
5bcd0e0c SL |
966 | } |
967 | ||
deb78b41 SL |
968 | static struct nullb_page *__null_lookup_page(struct nullb *nullb, |
969 | sector_t sector, bool for_write, bool is_cache) | |
5bcd0e0c SL |
970 | { |
971 | unsigned int sector_bit; | |
972 | u64 idx; | |
973 | struct nullb_page *t_page; | |
deb78b41 | 974 | struct radix_tree_root *root; |
5bcd0e0c SL |
975 | |
976 | idx = sector >> PAGE_SECTORS_SHIFT; | |
977 | sector_bit = (sector & SECTOR_MASK); | |
978 | ||
deb78b41 SL |
979 | root = is_cache ? &nullb->dev->cache : &nullb->dev->data; |
980 | t_page = radix_tree_lookup(root, idx); | |
5bcd0e0c SL |
981 | WARN_ON(t_page && t_page->page->index != idx); |
982 | ||
66231ad3 | 983 | if (t_page && (for_write || test_bit(sector_bit, t_page->bitmap))) |
5bcd0e0c SL |
984 | return t_page; |
985 | ||
986 | return NULL; | |
987 | } | |
988 | ||
deb78b41 SL |
989 | static struct nullb_page *null_lookup_page(struct nullb *nullb, |
990 | sector_t sector, bool for_write, bool ignore_cache) | |
991 | { | |
992 | struct nullb_page *page = NULL; | |
993 | ||
994 | if (!ignore_cache) | |
995 | page = __null_lookup_page(nullb, sector, for_write, true); | |
996 | if (page) | |
997 | return page; | |
998 | return __null_lookup_page(nullb, sector, for_write, false); | |
999 | } | |
1000 | ||
5bcd0e0c | 1001 | static struct nullb_page *null_insert_page(struct nullb *nullb, |
61884de0 JA |
1002 | sector_t sector, bool ignore_cache) |
1003 | __releases(&nullb->lock) | |
1004 | __acquires(&nullb->lock) | |
5bcd0e0c SL |
1005 | { |
1006 | u64 idx; | |
1007 | struct nullb_page *t_page; | |
1008 | ||
deb78b41 | 1009 | t_page = null_lookup_page(nullb, sector, true, ignore_cache); |
5bcd0e0c SL |
1010 | if (t_page) |
1011 | return t_page; | |
1012 | ||
1013 | spin_unlock_irq(&nullb->lock); | |
1014 | ||
c90b6b50 | 1015 | t_page = null_alloc_page(); |
5bcd0e0c SL |
1016 | if (!t_page) |
1017 | goto out_lock; | |
1018 | ||
1019 | if (radix_tree_preload(GFP_NOIO)) | |
1020 | goto out_freepage; | |
1021 | ||
1022 | spin_lock_irq(&nullb->lock); | |
1023 | idx = sector >> PAGE_SECTORS_SHIFT; | |
1024 | t_page->page->index = idx; | |
deb78b41 | 1025 | t_page = null_radix_tree_insert(nullb, idx, t_page, !ignore_cache); |
5bcd0e0c SL |
1026 | radix_tree_preload_end(); |
1027 | ||
1028 | return t_page; | |
1029 | out_freepage: | |
1030 | null_free_page(t_page); | |
1031 | out_lock: | |
1032 | spin_lock_irq(&nullb->lock); | |
deb78b41 SL |
1033 | return null_lookup_page(nullb, sector, true, ignore_cache); |
1034 | } | |
1035 | ||
1036 | static int null_flush_cache_page(struct nullb *nullb, struct nullb_page *c_page) | |
1037 | { | |
1038 | int i; | |
1039 | unsigned int offset; | |
1040 | u64 idx; | |
1041 | struct nullb_page *t_page, *ret; | |
1042 | void *dst, *src; | |
1043 | ||
1044 | idx = c_page->page->index; | |
1045 | ||
1046 | t_page = null_insert_page(nullb, idx << PAGE_SECTORS_SHIFT, true); | |
1047 | ||
66231ad3 ML |
1048 | __clear_bit(NULLB_PAGE_LOCK, c_page->bitmap); |
1049 | if (test_bit(NULLB_PAGE_FREE, c_page->bitmap)) { | |
deb78b41 | 1050 | null_free_page(c_page); |
66231ad3 | 1051 | if (t_page && null_page_empty(t_page)) { |
deb78b41 SL |
1052 | ret = radix_tree_delete_item(&nullb->dev->data, |
1053 | idx, t_page); | |
1054 | null_free_page(t_page); | |
1055 | } | |
1056 | return 0; | |
1057 | } | |
1058 | ||
1059 | if (!t_page) | |
1060 | return -ENOMEM; | |
1061 | ||
acc3c879 CK |
1062 | src = kmap_local_page(c_page->page); |
1063 | dst = kmap_local_page(t_page->page); | |
deb78b41 SL |
1064 | |
1065 | for (i = 0; i < PAGE_SECTORS; | |
1066 | i += (nullb->dev->blocksize >> SECTOR_SHIFT)) { | |
66231ad3 | 1067 | if (test_bit(i, c_page->bitmap)) { |
deb78b41 SL |
1068 | offset = (i << SECTOR_SHIFT); |
1069 | memcpy(dst + offset, src + offset, | |
1070 | nullb->dev->blocksize); | |
66231ad3 | 1071 | __set_bit(i, t_page->bitmap); |
deb78b41 SL |
1072 | } |
1073 | } | |
1074 | ||
acc3c879 CK |
1075 | kunmap_local(dst); |
1076 | kunmap_local(src); | |
deb78b41 SL |
1077 | |
1078 | ret = radix_tree_delete_item(&nullb->dev->cache, idx, c_page); | |
1079 | null_free_page(ret); | |
1080 | nullb->dev->curr_cache -= PAGE_SIZE; | |
1081 | ||
1082 | return 0; | |
1083 | } | |
1084 | ||
1085 | static int null_make_cache_space(struct nullb *nullb, unsigned long n) | |
f2298c04 | 1086 | { |
deb78b41 SL |
1087 | int i, err, nr_pages; |
1088 | struct nullb_page *c_pages[FREE_BATCH]; | |
1089 | unsigned long flushed = 0, one_round; | |
1090 | ||
1091 | again: | |
1092 | if ((nullb->dev->cache_size * 1024 * 1024) > | |
1093 | nullb->dev->curr_cache + n || nullb->dev->curr_cache == 0) | |
1094 | return 0; | |
1095 | ||
1096 | nr_pages = radix_tree_gang_lookup(&nullb->dev->cache, | |
1097 | (void **)c_pages, nullb->cache_flush_pos, FREE_BATCH); | |
1098 | /* | |
1099 | * nullb_flush_cache_page could unlock before using the c_pages. To | |
1100 | * avoid race, we don't allow page free | |
1101 | */ | |
1102 | for (i = 0; i < nr_pages; i++) { | |
1103 | nullb->cache_flush_pos = c_pages[i]->page->index; | |
1104 | /* | |
1105 | * We found the page which is being flushed to disk by other | |
1106 | * threads | |
1107 | */ | |
66231ad3 | 1108 | if (test_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap)) |
deb78b41 SL |
1109 | c_pages[i] = NULL; |
1110 | else | |
66231ad3 | 1111 | __set_bit(NULLB_PAGE_LOCK, c_pages[i]->bitmap); |
deb78b41 SL |
1112 | } |
1113 | ||
1114 | one_round = 0; | |
1115 | for (i = 0; i < nr_pages; i++) { | |
1116 | if (c_pages[i] == NULL) | |
1117 | continue; | |
1118 | err = null_flush_cache_page(nullb, c_pages[i]); | |
1119 | if (err) | |
1120 | return err; | |
1121 | one_round++; | |
1122 | } | |
1123 | flushed += one_round << PAGE_SHIFT; | |
1124 | ||
1125 | if (n > flushed) { | |
1126 | if (nr_pages == 0) | |
1127 | nullb->cache_flush_pos = 0; | |
1128 | if (one_round == 0) { | |
1129 | /* give other threads a chance */ | |
1130 | spin_unlock_irq(&nullb->lock); | |
1131 | spin_lock_irq(&nullb->lock); | |
1132 | } | |
1133 | goto again; | |
1134 | } | |
1135 | return 0; | |
5bcd0e0c SL |
1136 | } |
1137 | ||
1138 | static int copy_to_nullb(struct nullb *nullb, struct page *source, | |
deb78b41 | 1139 | unsigned int off, sector_t sector, size_t n, bool is_fua) |
5bcd0e0c SL |
1140 | { |
1141 | size_t temp, count = 0; | |
1142 | unsigned int offset; | |
1143 | struct nullb_page *t_page; | |
5bcd0e0c SL |
1144 | |
1145 | while (count < n) { | |
1146 | temp = min_t(size_t, nullb->dev->blocksize, n - count); | |
1147 | ||
deb78b41 SL |
1148 | if (null_cache_active(nullb) && !is_fua) |
1149 | null_make_cache_space(nullb, PAGE_SIZE); | |
1150 | ||
5bcd0e0c | 1151 | offset = (sector & SECTOR_MASK) << SECTOR_SHIFT; |
deb78b41 SL |
1152 | t_page = null_insert_page(nullb, sector, |
1153 | !null_cache_active(nullb) || is_fua); | |
5bcd0e0c SL |
1154 | if (!t_page) |
1155 | return -ENOSPC; | |
1156 | ||
fbb5615f | 1157 | memcpy_page(t_page->page, offset, source, off + count, temp); |
5bcd0e0c | 1158 | |
66231ad3 | 1159 | __set_bit(sector & SECTOR_MASK, t_page->bitmap); |
5bcd0e0c | 1160 | |
deb78b41 SL |
1161 | if (is_fua) |
1162 | null_free_sector(nullb, sector, true); | |
1163 | ||
5bcd0e0c SL |
1164 | count += temp; |
1165 | sector += temp >> SECTOR_SHIFT; | |
1166 | } | |
1167 | return 0; | |
1168 | } | |
1169 | ||
1170 | static int copy_from_nullb(struct nullb *nullb, struct page *dest, | |
1171 | unsigned int off, sector_t sector, size_t n) | |
1172 | { | |
1173 | size_t temp, count = 0; | |
1174 | unsigned int offset; | |
1175 | struct nullb_page *t_page; | |
5bcd0e0c SL |
1176 | |
1177 | while (count < n) { | |
1178 | temp = min_t(size_t, nullb->dev->blocksize, n - count); | |
1179 | ||
1180 | offset = (sector & SECTOR_MASK) << SECTOR_SHIFT; | |
deb78b41 SL |
1181 | t_page = null_lookup_page(nullb, sector, false, |
1182 | !null_cache_active(nullb)); | |
5bcd0e0c | 1183 | |
fbb5615f CK |
1184 | if (t_page) |
1185 | memcpy_page(dest, off + count, t_page->page, offset, | |
1186 | temp); | |
1187 | else | |
1188 | zero_user(dest, off + count, temp); | |
5bcd0e0c SL |
1189 | |
1190 | count += temp; | |
1191 | sector += temp >> SECTOR_SHIFT; | |
1192 | } | |
1193 | return 0; | |
1194 | } | |
1195 | ||
dd85b492 AJ |
1196 | static void nullb_fill_pattern(struct nullb *nullb, struct page *page, |
1197 | unsigned int len, unsigned int off) | |
1198 | { | |
fbb5615f | 1199 | memset_page(page, off, 0xff, len); |
dd85b492 AJ |
1200 | } |
1201 | ||
0ec4d913 DLM |
1202 | blk_status_t null_handle_discard(struct nullb_device *dev, |
1203 | sector_t sector, sector_t nr_sectors) | |
306eb6b4 | 1204 | { |
49c7089f DLM |
1205 | struct nullb *nullb = dev->nullb; |
1206 | size_t n = nr_sectors << SECTOR_SHIFT; | |
306eb6b4 SL |
1207 | size_t temp; |
1208 | ||
1209 | spin_lock_irq(&nullb->lock); | |
1210 | while (n > 0) { | |
49c7089f | 1211 | temp = min_t(size_t, n, dev->blocksize); |
deb78b41 SL |
1212 | null_free_sector(nullb, sector, false); |
1213 | if (null_cache_active(nullb)) | |
1214 | null_free_sector(nullb, sector, true); | |
306eb6b4 SL |
1215 | sector += temp >> SECTOR_SHIFT; |
1216 | n -= temp; | |
1217 | } | |
1218 | spin_unlock_irq(&nullb->lock); | |
49c7089f DLM |
1219 | |
1220 | return BLK_STS_OK; | |
306eb6b4 SL |
1221 | } |
1222 | ||
deb78b41 SL |
1223 | static int null_handle_flush(struct nullb *nullb) |
1224 | { | |
1225 | int err; | |
1226 | ||
1227 | if (!null_cache_active(nullb)) | |
1228 | return 0; | |
1229 | ||
1230 | spin_lock_irq(&nullb->lock); | |
1231 | while (true) { | |
1232 | err = null_make_cache_space(nullb, | |
1233 | nullb->dev->cache_size * 1024 * 1024); | |
1234 | if (err || nullb->dev->curr_cache == 0) | |
1235 | break; | |
1236 | } | |
1237 | ||
1238 | WARN_ON(!radix_tree_empty(&nullb->dev->cache)); | |
1239 | spin_unlock_irq(&nullb->lock); | |
1240 | return err; | |
1241 | } | |
1242 | ||
5bcd0e0c | 1243 | static int null_transfer(struct nullb *nullb, struct page *page, |
deb78b41 SL |
1244 | unsigned int len, unsigned int off, bool is_write, sector_t sector, |
1245 | bool is_fua) | |
f2298c04 | 1246 | { |
dd85b492 AJ |
1247 | struct nullb_device *dev = nullb->dev; |
1248 | unsigned int valid_len = len; | |
5bcd0e0c SL |
1249 | int err = 0; |
1250 | ||
1251 | if (!is_write) { | |
dd85b492 AJ |
1252 | if (dev->zoned) |
1253 | valid_len = null_zone_valid_read_len(nullb, | |
1254 | sector, len); | |
1255 | ||
1256 | if (valid_len) { | |
1257 | err = copy_from_nullb(nullb, page, off, | |
1258 | sector, valid_len); | |
1259 | off += valid_len; | |
1260 | len -= valid_len; | |
1261 | } | |
1262 | ||
1263 | if (len) | |
1264 | nullb_fill_pattern(nullb, page, len, off); | |
5bcd0e0c SL |
1265 | flush_dcache_page(page); |
1266 | } else { | |
1267 | flush_dcache_page(page); | |
deb78b41 | 1268 | err = copy_to_nullb(nullb, page, off, sector, len, is_fua); |
5bcd0e0c SL |
1269 | } |
1270 | ||
1271 | return err; | |
1272 | } | |
1273 | ||
1274 | static int null_handle_rq(struct nullb_cmd *cmd) | |
1275 | { | |
1276 | struct request *rq = cmd->rq; | |
1277 | struct nullb *nullb = cmd->nq->dev->nullb; | |
1278 | int err; | |
1279 | unsigned int len; | |
49c7089f | 1280 | sector_t sector = blk_rq_pos(rq); |
5bcd0e0c SL |
1281 | struct req_iterator iter; |
1282 | struct bio_vec bvec; | |
1283 | ||
5bcd0e0c SL |
1284 | spin_lock_irq(&nullb->lock); |
1285 | rq_for_each_segment(bvec, rq, iter) { | |
1286 | len = bvec.bv_len; | |
1287 | err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset, | |
deb78b41 | 1288 | op_is_write(req_op(rq)), sector, |
2d62e6b0 | 1289 | rq->cmd_flags & REQ_FUA); |
5bcd0e0c SL |
1290 | if (err) { |
1291 | spin_unlock_irq(&nullb->lock); | |
1292 | return err; | |
1293 | } | |
1294 | sector += len >> SECTOR_SHIFT; | |
1295 | } | |
1296 | spin_unlock_irq(&nullb->lock); | |
1297 | ||
1298 | return 0; | |
1299 | } | |
1300 | ||
1301 | static int null_handle_bio(struct nullb_cmd *cmd) | |
1302 | { | |
1303 | struct bio *bio = cmd->bio; | |
1304 | struct nullb *nullb = cmd->nq->dev->nullb; | |
1305 | int err; | |
1306 | unsigned int len; | |
49c7089f | 1307 | sector_t sector = bio->bi_iter.bi_sector; |
5bcd0e0c SL |
1308 | struct bio_vec bvec; |
1309 | struct bvec_iter iter; | |
1310 | ||
5bcd0e0c SL |
1311 | spin_lock_irq(&nullb->lock); |
1312 | bio_for_each_segment(bvec, bio, iter) { | |
1313 | len = bvec.bv_len; | |
1314 | err = null_transfer(nullb, bvec.bv_page, len, bvec.bv_offset, | |
deb78b41 | 1315 | op_is_write(bio_op(bio)), sector, |
bf7c7a04 | 1316 | bio->bi_opf & REQ_FUA); |
5bcd0e0c SL |
1317 | if (err) { |
1318 | spin_unlock_irq(&nullb->lock); | |
1319 | return err; | |
1320 | } | |
1321 | sector += len >> SECTOR_SHIFT; | |
1322 | } | |
1323 | spin_unlock_irq(&nullb->lock); | |
1324 | return 0; | |
1325 | } | |
1326 | ||
eff2c4f1 SL |
1327 | static void null_stop_queue(struct nullb *nullb) |
1328 | { | |
1329 | struct request_queue *q = nullb->q; | |
1330 | ||
1331 | if (nullb->dev->queue_mode == NULL_Q_MQ) | |
1332 | blk_mq_stop_hw_queues(q); | |
eff2c4f1 SL |
1333 | } |
1334 | ||
1335 | static void null_restart_queue_async(struct nullb *nullb) | |
1336 | { | |
1337 | struct request_queue *q = nullb->q; | |
eff2c4f1 SL |
1338 | |
1339 | if (nullb->dev->queue_mode == NULL_Q_MQ) | |
1340 | blk_mq_start_stopped_hw_queues(q, true); | |
eff2c4f1 SL |
1341 | } |
1342 | ||
adb84284 CK |
1343 | static inline blk_status_t null_handle_throttled(struct nullb_cmd *cmd) |
1344 | { | |
1345 | struct nullb_device *dev = cmd->nq->dev; | |
1346 | struct nullb *nullb = dev->nullb; | |
1347 | blk_status_t sts = BLK_STS_OK; | |
1348 | struct request *rq = cmd->rq; | |
1349 | ||
1350 | if (!hrtimer_active(&nullb->bw_timer)) | |
1351 | hrtimer_restart(&nullb->bw_timer); | |
1352 | ||
1353 | if (atomic_long_sub_return(blk_rq_bytes(rq), &nullb->cur_bytes) < 0) { | |
1354 | null_stop_queue(nullb); | |
1355 | /* race with timer */ | |
1356 | if (atomic_long_read(&nullb->cur_bytes) > 0) | |
1357 | null_restart_queue_async(nullb); | |
1358 | /* requeue request */ | |
1359 | sts = BLK_STS_DEV_RESOURCE; | |
1360 | } | |
1361 | return sts; | |
1362 | } | |
1363 | ||
8f94d1c1 CK |
1364 | static inline blk_status_t null_handle_badblocks(struct nullb_cmd *cmd, |
1365 | sector_t sector, | |
1366 | sector_t nr_sectors) | |
1367 | { | |
1368 | struct badblocks *bb = &cmd->nq->dev->badblocks; | |
1369 | sector_t first_bad; | |
1370 | int bad_sectors; | |
1371 | ||
1372 | if (badblocks_check(bb, sector, nr_sectors, &first_bad, &bad_sectors)) | |
1373 | return BLK_STS_IOERR; | |
1374 | ||
1375 | return BLK_STS_OK; | |
1376 | } | |
1377 | ||
7ea88e22 | 1378 | static inline blk_status_t null_handle_memory_backed(struct nullb_cmd *cmd, |
ff07a02e | 1379 | enum req_op op, |
49c7089f DLM |
1380 | sector_t sector, |
1381 | sector_t nr_sectors) | |
7ea88e22 CK |
1382 | { |
1383 | struct nullb_device *dev = cmd->nq->dev; | |
1384 | int err; | |
1385 | ||
49c7089f DLM |
1386 | if (op == REQ_OP_DISCARD) |
1387 | return null_handle_discard(dev, sector, nr_sectors); | |
1388 | ||
7ea88e22 CK |
1389 | if (dev->queue_mode == NULL_Q_BIO) |
1390 | err = null_handle_bio(cmd); | |
1391 | else | |
1392 | err = null_handle_rq(cmd); | |
1393 | ||
1394 | return errno_to_blk_status(err); | |
1395 | } | |
1396 | ||
cecbc9ce BVA |
1397 | static void nullb_zero_read_cmd_buffer(struct nullb_cmd *cmd) |
1398 | { | |
1399 | struct nullb_device *dev = cmd->nq->dev; | |
1400 | struct bio *bio; | |
1401 | ||
1402 | if (dev->memory_backed) | |
1403 | return; | |
1404 | ||
1405 | if (dev->queue_mode == NULL_Q_BIO && bio_op(cmd->bio) == REQ_OP_READ) { | |
1406 | zero_fill_bio(cmd->bio); | |
1407 | } else if (req_op(cmd->rq) == REQ_OP_READ) { | |
1408 | __rq_for_each_bio(bio, cmd->rq) | |
1409 | zero_fill_bio(bio); | |
1410 | } | |
1411 | } | |
1412 | ||
a3d7d674 CK |
1413 | static inline void nullb_complete_cmd(struct nullb_cmd *cmd) |
1414 | { | |
cecbc9ce BVA |
1415 | /* |
1416 | * Since root privileges are required to configure the null_blk | |
1417 | * driver, it is fine that this driver does not initialize the | |
1418 | * data buffers of read commands. Zero-initialize these buffers | |
1419 | * anyway if KMSAN is enabled to prevent that KMSAN complains | |
1420 | * about null_blk not initializing read data buffers. | |
1421 | */ | |
1422 | if (IS_ENABLED(CONFIG_KMSAN)) | |
1423 | nullb_zero_read_cmd_buffer(cmd); | |
1424 | ||
a3d7d674 CK |
1425 | /* Complete IO by inline, softirq or timer */ |
1426 | switch (cmd->nq->dev->irqmode) { | |
1427 | case NULL_IRQ_SOFTIRQ: | |
1428 | switch (cmd->nq->dev->queue_mode) { | |
1429 | case NULL_Q_MQ: | |
63f88659 | 1430 | blk_mq_complete_request(cmd->rq); |
a3d7d674 CK |
1431 | break; |
1432 | case NULL_Q_BIO: | |
1433 | /* | |
1434 | * XXX: no proper submitting cpu information available. | |
1435 | */ | |
1436 | end_cmd(cmd); | |
1437 | break; | |
1438 | } | |
1439 | break; | |
1440 | case NULL_IRQ_NONE: | |
1441 | end_cmd(cmd); | |
1442 | break; | |
1443 | case NULL_IRQ_TIMER: | |
1444 | null_cmd_end_timer(cmd); | |
1445 | break; | |
1446 | } | |
1447 | } | |
1448 | ||
ff07a02e BVA |
1449 | blk_status_t null_process_cmd(struct nullb_cmd *cmd, enum req_op op, |
1450 | sector_t sector, unsigned int nr_sectors) | |
9dd44c7e DLM |
1451 | { |
1452 | struct nullb_device *dev = cmd->nq->dev; | |
1453 | blk_status_t ret; | |
1454 | ||
1455 | if (dev->badblocks.shift != -1) { | |
1456 | ret = null_handle_badblocks(cmd, sector, nr_sectors); | |
1457 | if (ret != BLK_STS_OK) | |
1458 | return ret; | |
1459 | } | |
1460 | ||
1461 | if (dev->memory_backed) | |
49c7089f | 1462 | return null_handle_memory_backed(cmd, op, sector, nr_sectors); |
9dd44c7e DLM |
1463 | |
1464 | return BLK_STS_OK; | |
1465 | } | |
1466 | ||
d4b186ed | 1467 | static blk_status_t null_handle_cmd(struct nullb_cmd *cmd, sector_t sector, |
ff07a02e | 1468 | sector_t nr_sectors, enum req_op op) |
5bcd0e0c SL |
1469 | { |
1470 | struct nullb_device *dev = cmd->nq->dev; | |
eff2c4f1 | 1471 | struct nullb *nullb = dev->nullb; |
adb84284 | 1472 | blk_status_t sts; |
5bcd0e0c | 1473 | |
eff2c4f1 | 1474 | if (test_bit(NULLB_DEV_FL_THROTTLED, &dev->flags)) { |
adb84284 CK |
1475 | sts = null_handle_throttled(cmd); |
1476 | if (sts != BLK_STS_OK) | |
1477 | return sts; | |
eff2c4f1 SL |
1478 | } |
1479 | ||
d4b186ed CK |
1480 | if (op == REQ_OP_FLUSH) { |
1481 | cmd->error = errno_to_blk_status(null_handle_flush(nullb)); | |
1482 | goto out; | |
1483 | } | |
d4b186ed | 1484 | |
9dd44c7e | 1485 | if (dev->zoned) |
de3510e5 | 1486 | sts = null_process_zoned_cmd(cmd, op, sector, nr_sectors); |
9dd44c7e | 1487 | else |
de3510e5 DLM |
1488 | sts = null_process_cmd(cmd, op, sector, nr_sectors); |
1489 | ||
1490 | /* Do not overwrite errors (e.g. timeout errors) */ | |
1491 | if (cmd->error == BLK_STS_OK) | |
1492 | cmd->error = sts; | |
fceb5d1b | 1493 | |
2f54a613 | 1494 | out: |
a3d7d674 | 1495 | nullb_complete_cmd(cmd); |
5bcd0e0c | 1496 | return BLK_STS_OK; |
f2298c04 JA |
1497 | } |
1498 | ||
eff2c4f1 SL |
1499 | static enum hrtimer_restart nullb_bwtimer_fn(struct hrtimer *timer) |
1500 | { | |
1501 | struct nullb *nullb = container_of(timer, struct nullb, bw_timer); | |
1502 | ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL); | |
1503 | unsigned int mbps = nullb->dev->mbps; | |
1504 | ||
1505 | if (atomic_long_read(&nullb->cur_bytes) == mb_per_tick(mbps)) | |
1506 | return HRTIMER_NORESTART; | |
1507 | ||
1508 | atomic_long_set(&nullb->cur_bytes, mb_per_tick(mbps)); | |
1509 | null_restart_queue_async(nullb); | |
1510 | ||
1511 | hrtimer_forward_now(&nullb->bw_timer, timer_interval); | |
1512 | ||
1513 | return HRTIMER_RESTART; | |
1514 | } | |
1515 | ||
1516 | static void nullb_setup_bwtimer(struct nullb *nullb) | |
1517 | { | |
1518 | ktime_t timer_interval = ktime_set(0, TIMER_INTERVAL); | |
1519 | ||
1520 | hrtimer_init(&nullb->bw_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
1521 | nullb->bw_timer.function = nullb_bwtimer_fn; | |
1522 | atomic_long_set(&nullb->cur_bytes, mb_per_tick(nullb->dev->mbps)); | |
1523 | hrtimer_start(&nullb->bw_timer, timer_interval, HRTIMER_MODE_REL); | |
f2298c04 JA |
1524 | } |
1525 | ||
1526 | static struct nullb_queue *nullb_to_queue(struct nullb *nullb) | |
1527 | { | |
1528 | int index = 0; | |
1529 | ||
1530 | if (nullb->nr_queues != 1) | |
1531 | index = raw_smp_processor_id() / ((nr_cpu_ids + nullb->nr_queues - 1) / nullb->nr_queues); | |
1532 | ||
1533 | return &nullb->queues[index]; | |
1534 | } | |
1535 | ||
3e08773c | 1536 | static void null_submit_bio(struct bio *bio) |
f2298c04 | 1537 | { |
d4b186ed CK |
1538 | sector_t sector = bio->bi_iter.bi_sector; |
1539 | sector_t nr_sectors = bio_sectors(bio); | |
309dca30 | 1540 | struct nullb *nullb = bio->bi_bdev->bd_disk->private_data; |
f2298c04 | 1541 | struct nullb_queue *nq = nullb_to_queue(nullb); |
f2298c04 | 1542 | |
3d3472f3 | 1543 | null_handle_cmd(alloc_cmd(nq, bio), sector, nr_sectors, bio_op(bio)); |
f2298c04 JA |
1544 | } |
1545 | ||
bb4c19e0 AM |
1546 | #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION |
1547 | ||
1548 | static bool should_timeout_request(struct request *rq) | |
1549 | { | |
1550 | struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
1551 | struct nullb_device *dev = cmd->nq->dev; | |
1552 | ||
1553 | return should_fail(&dev->timeout_config.attr, 1); | |
1554 | } | |
1555 | ||
1556 | static bool should_requeue_request(struct request *rq) | |
1557 | { | |
1558 | struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
1559 | struct nullb_device *dev = cmd->nq->dev; | |
1560 | ||
1561 | return should_fail(&dev->requeue_config.attr, 1); | |
1562 | } | |
1563 | ||
1564 | static bool should_init_hctx_fail(struct nullb_device *dev) | |
1565 | { | |
1566 | return should_fail(&dev->init_hctx_fault_config.attr, 1); | |
1567 | } | |
1568 | ||
1569 | #else | |
1570 | ||
93b57046 JA |
1571 | static bool should_timeout_request(struct request *rq) |
1572 | { | |
24941b90 JA |
1573 | return false; |
1574 | } | |
93b57046 | 1575 | |
24941b90 JA |
1576 | static bool should_requeue_request(struct request *rq) |
1577 | { | |
93b57046 JA |
1578 | return false; |
1579 | } | |
1580 | ||
bb4c19e0 AM |
1581 | static bool should_init_hctx_fail(struct nullb_device *dev) |
1582 | { | |
1583 | return false; | |
1584 | } | |
1585 | ||
1586 | #endif | |
1587 | ||
a4e1d0b7 | 1588 | static void null_map_queues(struct blk_mq_tag_set *set) |
0a593fbb JA |
1589 | { |
1590 | struct nullb *nullb = set->driver_data; | |
1591 | int i, qoff; | |
15dfc662 SK |
1592 | unsigned int submit_queues = g_submit_queues; |
1593 | unsigned int poll_queues = g_poll_queues; | |
1594 | ||
1595 | if (nullb) { | |
1596 | struct nullb_device *dev = nullb->dev; | |
1597 | ||
1598 | /* | |
1599 | * Refer nr_hw_queues of the tag set to check if the expected | |
1600 | * number of hardware queues are prepared. If block layer failed | |
1601 | * to prepare them, use previous numbers of submit queues and | |
1602 | * poll queues to map queues. | |
1603 | */ | |
1604 | if (set->nr_hw_queues == | |
1605 | dev->submit_queues + dev->poll_queues) { | |
1606 | submit_queues = dev->submit_queues; | |
1607 | poll_queues = dev->poll_queues; | |
1608 | } else if (set->nr_hw_queues == | |
1609 | dev->prev_submit_queues + dev->prev_poll_queues) { | |
1610 | submit_queues = dev->prev_submit_queues; | |
1611 | poll_queues = dev->prev_poll_queues; | |
1612 | } else { | |
1613 | pr_warn("tag set has unexpected nr_hw_queues: %d\n", | |
1614 | set->nr_hw_queues); | |
10b41ea1 BVA |
1615 | WARN_ON_ONCE(true); |
1616 | submit_queues = 1; | |
1617 | poll_queues = 0; | |
15dfc662 SK |
1618 | } |
1619 | } | |
0a593fbb JA |
1620 | |
1621 | for (i = 0, qoff = 0; i < set->nr_maps; i++) { | |
1622 | struct blk_mq_queue_map *map = &set->map[i]; | |
1623 | ||
1624 | switch (i) { | |
1625 | case HCTX_TYPE_DEFAULT: | |
15dfc662 | 1626 | map->nr_queues = submit_queues; |
0a593fbb JA |
1627 | break; |
1628 | case HCTX_TYPE_READ: | |
1629 | map->nr_queues = 0; | |
1630 | continue; | |
1631 | case HCTX_TYPE_POLL: | |
15dfc662 | 1632 | map->nr_queues = poll_queues; |
0a593fbb JA |
1633 | break; |
1634 | } | |
1635 | map->queue_offset = qoff; | |
1636 | qoff += map->nr_queues; | |
1637 | blk_mq_map_queues(map); | |
1638 | } | |
0a593fbb JA |
1639 | } |
1640 | ||
1641 | static int null_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob) | |
1642 | { | |
1643 | struct nullb_queue *nq = hctx->driver_data; | |
1644 | LIST_HEAD(list); | |
1645 | int nr = 0; | |
1646 | ||
1647 | spin_lock(&nq->poll_lock); | |
1648 | list_splice_init(&nq->poll_list, &list); | |
1649 | spin_unlock(&nq->poll_lock); | |
1650 | ||
1651 | while (!list_empty(&list)) { | |
1652 | struct nullb_cmd *cmd; | |
1653 | struct request *req; | |
1654 | ||
1655 | req = list_first_entry(&list, struct request, queuelist); | |
1656 | list_del_init(&req->queuelist); | |
1657 | cmd = blk_mq_rq_to_pdu(req); | |
1658 | cmd->error = null_process_cmd(cmd, req_op(req), blk_rq_pos(req), | |
1659 | blk_rq_sectors(req)); | |
c5eafd79 | 1660 | if (!blk_mq_add_to_batch(req, iob, (__force int) cmd->error, |
2385ebf3 ML |
1661 | blk_mq_end_request_batch)) |
1662 | end_cmd(cmd); | |
0a593fbb JA |
1663 | nr++; |
1664 | } | |
1665 | ||
1666 | return nr; | |
1667 | } | |
1668 | ||
9bdb4833 | 1669 | static enum blk_eh_timer_return null_timeout_rq(struct request *rq) |
5448aca4 | 1670 | { |
0a593fbb | 1671 | struct blk_mq_hw_ctx *hctx = rq->mq_hctx; |
de3510e5 DLM |
1672 | struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq); |
1673 | ||
9c7eddf1 | 1674 | pr_info("rq %p timed out\n", rq); |
de3510e5 | 1675 | |
0a593fbb JA |
1676 | if (hctx->type == HCTX_TYPE_POLL) { |
1677 | struct nullb_queue *nq = hctx->driver_data; | |
1678 | ||
1679 | spin_lock(&nq->poll_lock); | |
1680 | list_del_init(&rq->queuelist); | |
1681 | spin_unlock(&nq->poll_lock); | |
1682 | } | |
1683 | ||
de3510e5 DLM |
1684 | /* |
1685 | * If the device is marked as blocking (i.e. memory backed or zoned | |
1686 | * device), the submission path may be blocked waiting for resources | |
1687 | * and cause real timeouts. For these real timeouts, the submission | |
1688 | * path will complete the request using blk_mq_complete_request(). | |
1689 | * Only fake timeouts need to execute blk_mq_complete_request() here. | |
1690 | */ | |
1691 | cmd->error = BLK_STS_TIMEOUT; | |
3e3876d3 | 1692 | if (cmd->fake_timeout || hctx->type == HCTX_TYPE_POLL) |
de3510e5 | 1693 | blk_mq_complete_request(rq); |
0df0bb08 | 1694 | return BLK_EH_DONE; |
5448aca4 JA |
1695 | } |
1696 | ||
fc17b653 | 1697 | static blk_status_t null_queue_rq(struct blk_mq_hw_ctx *hctx, |
b6402014 | 1698 | const struct blk_mq_queue_data *bd) |
f2298c04 | 1699 | { |
b6402014 DLM |
1700 | struct request *rq = bd->rq; |
1701 | struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
2984c868 | 1702 | struct nullb_queue *nq = hctx->driver_data; |
b6402014 DLM |
1703 | sector_t nr_sectors = blk_rq_sectors(rq); |
1704 | sector_t sector = blk_rq_pos(rq); | |
0a593fbb | 1705 | const bool is_poll = hctx->type == HCTX_TYPE_POLL; |
f2298c04 | 1706 | |
db5bcf87 JA |
1707 | might_sleep_if(hctx->flags & BLK_MQ_F_BLOCKING); |
1708 | ||
0a593fbb | 1709 | if (!is_poll && nq->dev->irqmode == NULL_IRQ_TIMER) { |
3c395a96 PV |
1710 | hrtimer_init(&cmd->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1711 | cmd->timer.function = null_cmd_timer_expired; | |
1712 | } | |
b6402014 | 1713 | cmd->rq = rq; |
ff770422 | 1714 | cmd->error = BLK_STS_OK; |
2984c868 | 1715 | cmd->nq = nq; |
b6402014 DLM |
1716 | cmd->fake_timeout = should_timeout_request(rq) || |
1717 | blk_should_fake_timeout(rq->q); | |
f2298c04 | 1718 | |
b6402014 | 1719 | blk_mq_start_request(rq); |
e2490073 | 1720 | |
b6402014 | 1721 | if (should_requeue_request(rq)) { |
24941b90 JA |
1722 | /* |
1723 | * Alternate between hitting the core BUSY path, and the | |
1724 | * driver driven requeue path | |
1725 | */ | |
1726 | nq->requeue_selection++; | |
1727 | if (nq->requeue_selection & 1) | |
1728 | return BLK_STS_RESOURCE; | |
b6402014 DLM |
1729 | blk_mq_requeue_request(rq, true); |
1730 | return BLK_STS_OK; | |
24941b90 | 1731 | } |
0a593fbb JA |
1732 | |
1733 | if (is_poll) { | |
1734 | spin_lock(&nq->poll_lock); | |
b6402014 | 1735 | list_add_tail(&rq->queuelist, &nq->poll_list); |
0a593fbb JA |
1736 | spin_unlock(&nq->poll_lock); |
1737 | return BLK_STS_OK; | |
1738 | } | |
de3510e5 | 1739 | if (cmd->fake_timeout) |
24941b90 | 1740 | return BLK_STS_OK; |
93b57046 | 1741 | |
b6402014 | 1742 | return null_handle_cmd(cmd, sector, nr_sectors, req_op(rq)); |
f2298c04 JA |
1743 | } |
1744 | ||
de65d2d2 MB |
1745 | static void cleanup_queue(struct nullb_queue *nq) |
1746 | { | |
eb25ad80 | 1747 | bitmap_free(nq->tag_map); |
de65d2d2 MB |
1748 | kfree(nq->cmds); |
1749 | } | |
1750 | ||
1751 | static void cleanup_queues(struct nullb *nullb) | |
1752 | { | |
1753 | int i; | |
1754 | ||
1755 | for (i = 0; i < nullb->nr_queues; i++) | |
1756 | cleanup_queue(&nullb->queues[i]); | |
1757 | ||
1758 | kfree(nullb->queues); | |
1759 | } | |
1760 | ||
78b10be2 BVA |
1761 | static void null_exit_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx) |
1762 | { | |
1763 | struct nullb_queue *nq = hctx->driver_data; | |
1764 | struct nullb *nullb = nq->dev->nullb; | |
1765 | ||
1766 | nullb->nr_queues--; | |
1767 | } | |
1768 | ||
1769 | static void null_init_queue(struct nullb *nullb, struct nullb_queue *nq) | |
1770 | { | |
1771 | init_waitqueue_head(&nq->wait); | |
1772 | nq->queue_depth = nullb->queue_depth; | |
1773 | nq->dev = nullb->dev; | |
0a593fbb JA |
1774 | INIT_LIST_HEAD(&nq->poll_list); |
1775 | spin_lock_init(&nq->poll_lock); | |
78b10be2 BVA |
1776 | } |
1777 | ||
1778 | static int null_init_hctx(struct blk_mq_hw_ctx *hctx, void *driver_data, | |
1779 | unsigned int hctx_idx) | |
1780 | { | |
1781 | struct nullb *nullb = hctx->queue->queuedata; | |
1782 | struct nullb_queue *nq; | |
1783 | ||
bb4c19e0 | 1784 | if (should_init_hctx_fail(nullb->dev)) |
596444e7 | 1785 | return -EFAULT; |
596444e7 | 1786 | |
78b10be2 BVA |
1787 | nq = &nullb->queues[hctx_idx]; |
1788 | hctx->driver_data = nq; | |
1789 | null_init_queue(nullb, nq); | |
1790 | nullb->nr_queues++; | |
1791 | ||
1792 | return 0; | |
1793 | } | |
1794 | ||
1795 | static const struct blk_mq_ops null_mq_ops = { | |
1796 | .queue_rq = null_queue_rq, | |
1797 | .complete = null_complete_rq, | |
1798 | .timeout = null_timeout_rq, | |
0a593fbb JA |
1799 | .poll = null_poll, |
1800 | .map_queues = null_map_queues, | |
78b10be2 BVA |
1801 | .init_hctx = null_init_hctx, |
1802 | .exit_hctx = null_exit_hctx, | |
1803 | }; | |
1804 | ||
9ae2d0aa MB |
1805 | static void null_del_dev(struct nullb *nullb) |
1806 | { | |
9b03b713 BVA |
1807 | struct nullb_device *dev; |
1808 | ||
1809 | if (!nullb) | |
1810 | return; | |
1811 | ||
1812 | dev = nullb->dev; | |
2984c868 | 1813 | |
94bc02e3 SL |
1814 | ida_simple_remove(&nullb_indexes, nullb->index); |
1815 | ||
9ae2d0aa MB |
1816 | list_del_init(&nullb->list); |
1817 | ||
74ede5af | 1818 | del_gendisk(nullb->disk); |
eff2c4f1 SL |
1819 | |
1820 | if (test_bit(NULLB_DEV_FL_THROTTLED, &nullb->dev->flags)) { | |
1821 | hrtimer_cancel(&nullb->bw_timer); | |
1822 | atomic_long_set(&nullb->cur_bytes, LONG_MAX); | |
1823 | null_restart_queue_async(nullb); | |
1824 | } | |
1825 | ||
8b9ab626 | 1826 | put_disk(nullb->disk); |
2984c868 SL |
1827 | if (dev->queue_mode == NULL_Q_MQ && |
1828 | nullb->tag_set == &nullb->__tag_set) | |
82f402fe | 1829 | blk_mq_free_tag_set(nullb->tag_set); |
9ae2d0aa | 1830 | cleanup_queues(nullb); |
deb78b41 SL |
1831 | if (null_cache_active(nullb)) |
1832 | null_free_device_storage(nullb->dev, true); | |
9ae2d0aa | 1833 | kfree(nullb); |
2984c868 | 1834 | dev->nullb = NULL; |
9ae2d0aa MB |
1835 | } |
1836 | ||
306eb6b4 SL |
1837 | static void null_config_discard(struct nullb *nullb) |
1838 | { | |
1839 | if (nullb->dev->discard == false) | |
1840 | return; | |
1592cd15 | 1841 | |
49c7089f DLM |
1842 | if (!nullb->dev->memory_backed) { |
1843 | nullb->dev->discard = false; | |
1844 | pr_info("discard option is ignored without memory backing\n"); | |
1845 | return; | |
1846 | } | |
1847 | ||
1592cd15 CK |
1848 | if (nullb->dev->zoned) { |
1849 | nullb->dev->discard = false; | |
1850 | pr_info("discard option is ignored in zoned mode\n"); | |
1851 | return; | |
1852 | } | |
1853 | ||
306eb6b4 | 1854 | nullb->q->limits.discard_granularity = nullb->dev->blocksize; |
306eb6b4 | 1855 | blk_queue_max_discard_sectors(nullb->q, UINT_MAX >> 9); |
9ae2d0aa MB |
1856 | } |
1857 | ||
c62b37d9 CH |
1858 | static const struct block_device_operations null_bio_ops = { |
1859 | .owner = THIS_MODULE, | |
1860 | .submit_bio = null_submit_bio, | |
1861 | .report_zones = null_report_zones, | |
1862 | }; | |
1863 | ||
1864 | static const struct block_device_operations null_rq_ops = { | |
e3f89564 | 1865 | .owner = THIS_MODULE, |
7fc8fb51 | 1866 | .report_zones = null_report_zones, |
f2298c04 JA |
1867 | }; |
1868 | ||
1869 | static int setup_commands(struct nullb_queue *nq) | |
1870 | { | |
1871 | struct nullb_cmd *cmd; | |
eb25ad80 | 1872 | int i; |
f2298c04 | 1873 | |
6396bb22 | 1874 | nq->cmds = kcalloc(nq->queue_depth, sizeof(*cmd), GFP_KERNEL); |
f2298c04 | 1875 | if (!nq->cmds) |
2d263a78 | 1876 | return -ENOMEM; |
f2298c04 | 1877 | |
eb25ad80 | 1878 | nq->tag_map = bitmap_zalloc(nq->queue_depth, GFP_KERNEL); |
f2298c04 JA |
1879 | if (!nq->tag_map) { |
1880 | kfree(nq->cmds); | |
2d263a78 | 1881 | return -ENOMEM; |
f2298c04 JA |
1882 | } |
1883 | ||
1884 | for (i = 0; i < nq->queue_depth; i++) { | |
1885 | cmd = &nq->cmds[i]; | |
f2298c04 JA |
1886 | cmd->tag = -1U; |
1887 | } | |
1888 | ||
1889 | return 0; | |
1890 | } | |
1891 | ||
f2298c04 JA |
1892 | static int setup_queues(struct nullb *nullb) |
1893 | { | |
0a593fbb JA |
1894 | int nqueues = nr_cpu_ids; |
1895 | ||
1896 | if (g_poll_queues) | |
1897 | nqueues += g_poll_queues; | |
1898 | ||
1899 | nullb->queues = kcalloc(nqueues, sizeof(struct nullb_queue), | |
6396bb22 | 1900 | GFP_KERNEL); |
f2298c04 | 1901 | if (!nullb->queues) |
2d263a78 | 1902 | return -ENOMEM; |
f2298c04 | 1903 | |
2984c868 | 1904 | nullb->queue_depth = nullb->dev->hw_queue_depth; |
2d263a78 MB |
1905 | return 0; |
1906 | } | |
1907 | ||
1908 | static int init_driver_queues(struct nullb *nullb) | |
1909 | { | |
1910 | struct nullb_queue *nq; | |
1911 | int i, ret = 0; | |
f2298c04 | 1912 | |
2984c868 | 1913 | for (i = 0; i < nullb->dev->submit_queues; i++) { |
f2298c04 | 1914 | nq = &nullb->queues[i]; |
2d263a78 MB |
1915 | |
1916 | null_init_queue(nullb, nq); | |
1917 | ||
1918 | ret = setup_commands(nq); | |
1919 | if (ret) | |
31f9690e | 1920 | return ret; |
f2298c04 JA |
1921 | nullb->nr_queues++; |
1922 | } | |
2d263a78 | 1923 | return 0; |
f2298c04 JA |
1924 | } |
1925 | ||
9ae2d0aa | 1926 | static int null_gendisk_register(struct nullb *nullb) |
f2298c04 | 1927 | { |
979d5447 | 1928 | sector_t size = ((sector_t)nullb->dev->size * SZ_1M) >> SECTOR_SHIFT; |
132226b3 | 1929 | struct gendisk *disk = nullb->disk; |
9ae2d0aa | 1930 | |
979d5447 | 1931 | set_capacity(disk, size); |
9ae2d0aa | 1932 | |
9ae2d0aa MB |
1933 | disk->major = null_major; |
1934 | disk->first_minor = nullb->index; | |
132226b3 | 1935 | disk->minors = 1; |
c62b37d9 CH |
1936 | if (queue_is_mq(nullb->q)) |
1937 | disk->fops = &null_rq_ops; | |
1938 | else | |
1939 | disk->fops = &null_bio_ops; | |
9ae2d0aa | 1940 | disk->private_data = nullb; |
9ae2d0aa MB |
1941 | strncpy(disk->disk_name, nullb->disk_name, DISK_NAME_LEN); |
1942 | ||
bf505456 | 1943 | if (nullb->dev->zoned) { |
d205bde7 DLM |
1944 | int ret = null_register_zoned_dev(nullb); |
1945 | ||
1946 | if (ret) | |
1947 | return ret; | |
bf505456 DLM |
1948 | } |
1949 | ||
10e7123d | 1950 | return add_disk(disk); |
9ae2d0aa MB |
1951 | } |
1952 | ||
2984c868 | 1953 | static int null_init_tag_set(struct nullb *nullb, struct blk_mq_tag_set *set) |
82f402fe | 1954 | { |
7012eef5 VF |
1955 | unsigned int flags = BLK_MQ_F_SHOULD_MERGE; |
1956 | int hw_queues, numa_node; | |
1957 | unsigned int queue_depth; | |
0a593fbb JA |
1958 | int poll_queues; |
1959 | ||
7012eef5 VF |
1960 | if (nullb) { |
1961 | hw_queues = nullb->dev->submit_queues; | |
1962 | poll_queues = nullb->dev->poll_queues; | |
1963 | queue_depth = nullb->dev->hw_queue_depth; | |
1964 | numa_node = nullb->dev->home_node; | |
1965 | if (nullb->dev->no_sched) | |
1966 | flags |= BLK_MQ_F_NO_SCHED; | |
1967 | if (nullb->dev->shared_tag_bitmap) | |
1968 | flags |= BLK_MQ_F_TAG_HCTX_SHARED; | |
1969 | if (nullb->dev->blocking) | |
1970 | flags |= BLK_MQ_F_BLOCKING; | |
1971 | } else { | |
1972 | hw_queues = g_submit_queues; | |
1973 | poll_queues = g_poll_queues; | |
1974 | queue_depth = g_hw_queue_depth; | |
1975 | numa_node = g_home_node; | |
1976 | if (g_no_sched) | |
1977 | flags |= BLK_MQ_F_NO_SCHED; | |
1978 | if (g_shared_tag_bitmap) | |
1979 | flags |= BLK_MQ_F_TAG_HCTX_SHARED; | |
1980 | if (g_blocking) | |
1981 | flags |= BLK_MQ_F_BLOCKING; | |
1982 | } | |
1983 | ||
82f402fe | 1984 | set->ops = &null_mq_ops; |
82f402fe | 1985 | set->cmd_size = sizeof(struct nullb_cmd); |
7012eef5 | 1986 | set->flags = flags; |
0a593fbb | 1987 | set->driver_data = nullb; |
7012eef5 VF |
1988 | set->nr_hw_queues = hw_queues; |
1989 | set->queue_depth = queue_depth; | |
1990 | set->numa_node = numa_node; | |
1991 | if (poll_queues) { | |
1992 | set->nr_hw_queues += poll_queues; | |
0a593fbb | 1993 | set->nr_maps = 3; |
7012eef5 | 1994 | } else { |
0a593fbb | 1995 | set->nr_maps = 1; |
7012eef5 | 1996 | } |
82f402fe JA |
1997 | |
1998 | return blk_mq_alloc_tag_set(set); | |
1999 | } | |
2000 | ||
5c4bd1f4 | 2001 | static int null_validate_conf(struct nullb_device *dev) |
cedcafad | 2002 | { |
63f8793e CK |
2003 | if (dev->queue_mode == NULL_Q_RQ) { |
2004 | pr_err("legacy IO path is no longer available\n"); | |
2005 | return -EINVAL; | |
2006 | } | |
2007 | ||
cedcafad SL |
2008 | dev->blocksize = round_down(dev->blocksize, 512); |
2009 | dev->blocksize = clamp_t(unsigned int, dev->blocksize, 512, 4096); | |
cedcafad SL |
2010 | |
2011 | if (dev->queue_mode == NULL_Q_MQ && dev->use_per_node_hctx) { | |
2012 | if (dev->submit_queues != nr_online_nodes) | |
2013 | dev->submit_queues = nr_online_nodes; | |
2014 | } else if (dev->submit_queues > nr_cpu_ids) | |
2015 | dev->submit_queues = nr_cpu_ids; | |
2016 | else if (dev->submit_queues == 0) | |
2017 | dev->submit_queues = 1; | |
15dfc662 SK |
2018 | dev->prev_submit_queues = dev->submit_queues; |
2019 | ||
2020 | if (dev->poll_queues > g_poll_queues) | |
2021 | dev->poll_queues = g_poll_queues; | |
15dfc662 | 2022 | dev->prev_poll_queues = dev->poll_queues; |
cedcafad SL |
2023 | |
2024 | dev->queue_mode = min_t(unsigned int, dev->queue_mode, NULL_Q_MQ); | |
2025 | dev->irqmode = min_t(unsigned int, dev->irqmode, NULL_IRQ_TIMER); | |
5bcd0e0c SL |
2026 | |
2027 | /* Do memory allocation, so set blocking */ | |
2028 | if (dev->memory_backed) | |
2029 | dev->blocking = true; | |
deb78b41 SL |
2030 | else /* cache is meaningless */ |
2031 | dev->cache_size = 0; | |
2032 | dev->cache_size = min_t(unsigned long, ULONG_MAX / 1024 / 1024, | |
2033 | dev->cache_size); | |
eff2c4f1 SL |
2034 | dev->mbps = min_t(unsigned int, 1024 * 40, dev->mbps); |
2035 | /* can not stop a queue */ | |
2036 | if (dev->queue_mode == NULL_Q_BIO) | |
2037 | dev->mbps = 0; | |
5c4bd1f4 DLM |
2038 | |
2039 | if (dev->zoned && | |
2040 | (!dev->zone_size || !is_power_of_2(dev->zone_size))) { | |
2041 | pr_err("zone_size must be power-of-two\n"); | |
2042 | return -EINVAL; | |
2043 | } | |
2044 | ||
2045 | return 0; | |
cedcafad SL |
2046 | } |
2047 | ||
33f782c4 | 2048 | #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION |
24941b90 JA |
2049 | static bool __null_setup_fault(struct fault_attr *attr, char *str) |
2050 | { | |
2051 | if (!str[0]) | |
93b57046 JA |
2052 | return true; |
2053 | ||
24941b90 | 2054 | if (!setup_fault_attr(attr, str)) |
93b57046 JA |
2055 | return false; |
2056 | ||
24941b90 JA |
2057 | attr->verbose = 0; |
2058 | return true; | |
2059 | } | |
2060 | #endif | |
2061 | ||
2062 | static bool null_setup_fault(void) | |
2063 | { | |
2064 | #ifdef CONFIG_BLK_DEV_NULL_BLK_FAULT_INJECTION | |
2065 | if (!__null_setup_fault(&null_timeout_attr, g_timeout_str)) | |
2066 | return false; | |
2067 | if (!__null_setup_fault(&null_requeue_attr, g_requeue_str)) | |
2068 | return false; | |
596444e7 BVA |
2069 | if (!__null_setup_fault(&null_init_hctx_attr, g_init_hctx_str)) |
2070 | return false; | |
33f782c4 | 2071 | #endif |
93b57046 JA |
2072 | return true; |
2073 | } | |
2074 | ||
2984c868 | 2075 | static int null_add_dev(struct nullb_device *dev) |
9ae2d0aa MB |
2076 | { |
2077 | struct nullb *nullb; | |
dc501dc0 | 2078 | int rv; |
f2298c04 | 2079 | |
5c4bd1f4 DLM |
2080 | rv = null_validate_conf(dev); |
2081 | if (rv) | |
2082 | return rv; | |
cedcafad | 2083 | |
2984c868 | 2084 | nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, dev->home_node); |
dc501dc0 RE |
2085 | if (!nullb) { |
2086 | rv = -ENOMEM; | |
24d2f903 | 2087 | goto out; |
dc501dc0 | 2088 | } |
2984c868 SL |
2089 | nullb->dev = dev; |
2090 | dev->nullb = nullb; | |
f2298c04 JA |
2091 | |
2092 | spin_lock_init(&nullb->lock); | |
2093 | ||
dc501dc0 RE |
2094 | rv = setup_queues(nullb); |
2095 | if (rv) | |
24d2f903 | 2096 | goto out_free_nullb; |
f2298c04 | 2097 | |
2984c868 | 2098 | if (dev->queue_mode == NULL_Q_MQ) { |
82f402fe JA |
2099 | if (shared_tags) { |
2100 | nullb->tag_set = &tag_set; | |
2101 | rv = 0; | |
2102 | } else { | |
2103 | nullb->tag_set = &nullb->__tag_set; | |
2984c868 | 2104 | rv = null_init_tag_set(nullb, nullb->tag_set); |
82f402fe JA |
2105 | } |
2106 | ||
dc501dc0 | 2107 | if (rv) |
24d2f903 CH |
2108 | goto out_cleanup_queues; |
2109 | ||
5448aca4 | 2110 | nullb->tag_set->timeout = 5 * HZ; |
6759b1a2 CH |
2111 | nullb->disk = blk_mq_alloc_disk(nullb->tag_set, nullb); |
2112 | if (IS_ERR(nullb->disk)) { | |
2113 | rv = PTR_ERR(nullb->disk); | |
24d2f903 | 2114 | goto out_cleanup_tags; |
6759b1a2 CH |
2115 | } |
2116 | nullb->q = nullb->disk->queue; | |
2984c868 | 2117 | } else if (dev->queue_mode == NULL_Q_BIO) { |
132226b3 CH |
2118 | rv = -ENOMEM; |
2119 | nullb->disk = blk_alloc_disk(nullb->dev->home_node); | |
2120 | if (!nullb->disk) | |
24d2f903 | 2121 | goto out_cleanup_queues; |
132226b3 CH |
2122 | |
2123 | nullb->q = nullb->disk->queue; | |
31f9690e JK |
2124 | rv = init_driver_queues(nullb); |
2125 | if (rv) | |
132226b3 | 2126 | goto out_cleanup_disk; |
f2298c04 JA |
2127 | } |
2128 | ||
eff2c4f1 SL |
2129 | if (dev->mbps) { |
2130 | set_bit(NULLB_DEV_FL_THROTTLED, &dev->flags); | |
2131 | nullb_setup_bwtimer(nullb); | |
2132 | } | |
2133 | ||
deb78b41 SL |
2134 | if (dev->cache_size > 0) { |
2135 | set_bit(NULLB_DEV_FL_CACHE, &nullb->dev->flags); | |
2136 | blk_queue_write_cache(nullb->q, true, true); | |
deb78b41 SL |
2137 | } |
2138 | ||
ca4b2a01 | 2139 | if (dev->zoned) { |
d205bde7 | 2140 | rv = null_init_zoned_dev(dev, nullb->q); |
ca4b2a01 | 2141 | if (rv) |
132226b3 | 2142 | goto out_cleanup_disk; |
ca4b2a01 MB |
2143 | } |
2144 | ||
f2298c04 | 2145 | nullb->q->queuedata = nullb; |
8b904b5b | 2146 | blk_queue_flag_set(QUEUE_FLAG_NONROT, nullb->q); |
f2298c04 | 2147 | |
f2298c04 | 2148 | mutex_lock(&lock); |
ee452a8d DC |
2149 | rv = ida_simple_get(&nullb_indexes, 0, 0, GFP_KERNEL); |
2150 | if (rv < 0) { | |
2151 | mutex_unlock(&lock); | |
2152 | goto out_cleanup_zone; | |
2153 | } | |
2154 | nullb->index = rv; | |
2155 | dev->index = rv; | |
f2298c04 JA |
2156 | mutex_unlock(&lock); |
2157 | ||
2984c868 SL |
2158 | blk_queue_logical_block_size(nullb->q, dev->blocksize); |
2159 | blk_queue_physical_block_size(nullb->q, dev->blocksize); | |
ea17fd35 DLM |
2160 | if (!dev->max_sectors) |
2161 | dev->max_sectors = queue_max_hw_sectors(nullb->q); | |
0a26f327 | 2162 | dev->max_sectors = min(dev->max_sectors, BLK_DEF_MAX_SECTORS); |
ea17fd35 | 2163 | blk_queue_max_hw_sectors(nullb->q, dev->max_sectors); |
f2298c04 | 2164 | |
cee1b215 MG |
2165 | if (dev->virt_boundary) |
2166 | blk_queue_virt_boundary(nullb->q, PAGE_SIZE - 1); | |
2167 | ||
306eb6b4 | 2168 | null_config_discard(nullb); |
f2298c04 | 2169 | |
bb4c19e0 | 2170 | if (config_item_name(&dev->group.cg_item)) { |
49c3b926 DLM |
2171 | /* Use configfs dir name as the device name */ |
2172 | snprintf(nullb->disk_name, sizeof(nullb->disk_name), | |
bb4c19e0 | 2173 | "%s", config_item_name(&dev->group.cg_item)); |
49c3b926 DLM |
2174 | } else { |
2175 | sprintf(nullb->disk_name, "nullb%d", nullb->index); | |
2176 | } | |
b2b7e001 | 2177 | |
74ede5af | 2178 | rv = null_gendisk_register(nullb); |
9ae2d0aa | 2179 | if (rv) |
ee452a8d | 2180 | goto out_ida_free; |
a514379b MB |
2181 | |
2182 | mutex_lock(&lock); | |
2183 | list_add_tail(&nullb->list, &nullb_list); | |
2184 | mutex_unlock(&lock); | |
3681c85d | 2185 | |
db060f54 DLM |
2186 | pr_info("disk %s created\n", nullb->disk_name); |
2187 | ||
f2298c04 | 2188 | return 0; |
ee452a8d DC |
2189 | |
2190 | out_ida_free: | |
2191 | ida_free(&nullb_indexes, nullb->index); | |
ca4b2a01 | 2192 | out_cleanup_zone: |
d205bde7 | 2193 | null_free_zoned_dev(dev); |
132226b3 | 2194 | out_cleanup_disk: |
8b9ab626 | 2195 | put_disk(nullb->disk); |
24d2f903 | 2196 | out_cleanup_tags: |
2984c868 | 2197 | if (dev->queue_mode == NULL_Q_MQ && nullb->tag_set == &nullb->__tag_set) |
82f402fe | 2198 | blk_mq_free_tag_set(nullb->tag_set); |
24d2f903 CH |
2199 | out_cleanup_queues: |
2200 | cleanup_queues(nullb); | |
2201 | out_free_nullb: | |
2202 | kfree(nullb); | |
2004bfde | 2203 | dev->nullb = NULL; |
24d2f903 | 2204 | out: |
dc501dc0 | 2205 | return rv; |
f2298c04 JA |
2206 | } |
2207 | ||
49c3b926 DLM |
2208 | static struct nullb *null_find_dev_by_name(const char *name) |
2209 | { | |
2210 | struct nullb *nullb = NULL, *nb; | |
2211 | ||
2212 | mutex_lock(&lock); | |
2213 | list_for_each_entry(nb, &nullb_list, list) { | |
2214 | if (strcmp(nb->disk_name, name) == 0) { | |
2215 | nullb = nb; | |
2216 | break; | |
2217 | } | |
2218 | } | |
2219 | mutex_unlock(&lock); | |
2220 | ||
2221 | return nullb; | |
2222 | } | |
2223 | ||
b3a0a73e DLM |
2224 | static int null_create_dev(void) |
2225 | { | |
2226 | struct nullb_device *dev; | |
2227 | int ret; | |
2228 | ||
2229 | dev = null_alloc_dev(); | |
2230 | if (!dev) | |
2231 | return -ENOMEM; | |
2232 | ||
2233 | ret = null_add_dev(dev); | |
2234 | if (ret) { | |
2235 | null_free_dev(dev); | |
2236 | return ret; | |
2237 | } | |
2238 | ||
2239 | return 0; | |
2240 | } | |
2241 | ||
2242 | static void null_destroy_dev(struct nullb *nullb) | |
2243 | { | |
2244 | struct nullb_device *dev = nullb->dev; | |
2245 | ||
2246 | null_del_dev(nullb); | |
2247 | null_free_dev(dev); | |
2248 | } | |
2249 | ||
f2298c04 JA |
2250 | static int __init null_init(void) |
2251 | { | |
af096e22 | 2252 | int ret = 0; |
f2298c04 | 2253 | unsigned int i; |
af096e22 | 2254 | struct nullb *nullb; |
f2298c04 | 2255 | |
2984c868 | 2256 | if (g_bs > PAGE_SIZE) { |
9c7eddf1 AA |
2257 | pr_warn("invalid block size\n"); |
2258 | pr_warn("defaults block size to %lu\n", PAGE_SIZE); | |
2984c868 | 2259 | g_bs = PAGE_SIZE; |
9967d8ac | 2260 | } |
f2298c04 | 2261 | |
ea17fd35 DLM |
2262 | if (g_max_sectors > BLK_DEF_MAX_SECTORS) { |
2263 | pr_warn("invalid max sectors\n"); | |
2264 | pr_warn("defaults max sectors to %u\n", BLK_DEF_MAX_SECTORS); | |
2265 | g_max_sectors = BLK_DEF_MAX_SECTORS; | |
2266 | } | |
2267 | ||
7ff684a6 | 2268 | if (g_home_node != NUMA_NO_NODE && g_home_node >= nr_online_nodes) { |
9c7eddf1 | 2269 | pr_err("invalid home_node value\n"); |
7ff684a6 JP |
2270 | g_home_node = NUMA_NO_NODE; |
2271 | } | |
2272 | ||
bb4c19e0 AM |
2273 | if (!null_setup_fault()) |
2274 | return -EINVAL; | |
2275 | ||
e50b1e32 | 2276 | if (g_queue_mode == NULL_Q_RQ) { |
525323d2 | 2277 | pr_err("legacy IO path is no longer available\n"); |
e50b1e32 JA |
2278 | return -EINVAL; |
2279 | } | |
525323d2 | 2280 | |
2984c868 SL |
2281 | if (g_queue_mode == NULL_Q_MQ && g_use_per_node_hctx) { |
2282 | if (g_submit_queues != nr_online_nodes) { | |
9c7eddf1 | 2283 | pr_warn("submit_queues param is set to %u.\n", |
525323d2 | 2284 | nr_online_nodes); |
2984c868 | 2285 | g_submit_queues = nr_online_nodes; |
fc1bc354 | 2286 | } |
525323d2 | 2287 | } else if (g_submit_queues > nr_cpu_ids) { |
2984c868 | 2288 | g_submit_queues = nr_cpu_ids; |
525323d2 | 2289 | } else if (g_submit_queues <= 0) { |
2984c868 | 2290 | g_submit_queues = 1; |
525323d2 | 2291 | } |
f2298c04 | 2292 | |
2984c868 SL |
2293 | if (g_queue_mode == NULL_Q_MQ && shared_tags) { |
2294 | ret = null_init_tag_set(NULL, &tag_set); | |
db2d153d MG |
2295 | if (ret) |
2296 | return ret; | |
2297 | } | |
2298 | ||
3bf2bd20 SL |
2299 | config_group_init(&nullb_subsys.su_group); |
2300 | mutex_init(&nullb_subsys.su_mutex); | |
2301 | ||
2302 | ret = configfs_register_subsystem(&nullb_subsys); | |
2303 | if (ret) | |
2304 | goto err_tagset; | |
2305 | ||
f2298c04 JA |
2306 | mutex_init(&lock); |
2307 | ||
f2298c04 | 2308 | null_major = register_blkdev(0, "nullb"); |
db2d153d MG |
2309 | if (null_major < 0) { |
2310 | ret = null_major; | |
3bf2bd20 | 2311 | goto err_conf; |
db2d153d | 2312 | } |
f2298c04 JA |
2313 | |
2314 | for (i = 0; i < nr_devices; i++) { | |
b3a0a73e DLM |
2315 | ret = null_create_dev(); |
2316 | if (ret) | |
af096e22 | 2317 | goto err_dev; |
f2298c04 JA |
2318 | } |
2319 | ||
9c7eddf1 | 2320 | pr_info("module loaded\n"); |
f2298c04 | 2321 | return 0; |
af096e22 MH |
2322 | |
2323 | err_dev: | |
2324 | while (!list_empty(&nullb_list)) { | |
2325 | nullb = list_entry(nullb_list.next, struct nullb, list); | |
b3a0a73e | 2326 | null_destroy_dev(nullb); |
af096e22 | 2327 | } |
af096e22 | 2328 | unregister_blkdev(null_major, "nullb"); |
3bf2bd20 SL |
2329 | err_conf: |
2330 | configfs_unregister_subsystem(&nullb_subsys); | |
db2d153d | 2331 | err_tagset: |
2984c868 | 2332 | if (g_queue_mode == NULL_Q_MQ && shared_tags) |
db2d153d | 2333 | blk_mq_free_tag_set(&tag_set); |
af096e22 | 2334 | return ret; |
f2298c04 JA |
2335 | } |
2336 | ||
2337 | static void __exit null_exit(void) | |
2338 | { | |
2339 | struct nullb *nullb; | |
2340 | ||
3bf2bd20 SL |
2341 | configfs_unregister_subsystem(&nullb_subsys); |
2342 | ||
f2298c04 JA |
2343 | unregister_blkdev(null_major, "nullb"); |
2344 | ||
2345 | mutex_lock(&lock); | |
2346 | while (!list_empty(&nullb_list)) { | |
2347 | nullb = list_entry(nullb_list.next, struct nullb, list); | |
b3a0a73e | 2348 | null_destroy_dev(nullb); |
f2298c04 JA |
2349 | } |
2350 | mutex_unlock(&lock); | |
6bb9535b | 2351 | |
2984c868 | 2352 | if (g_queue_mode == NULL_Q_MQ && shared_tags) |
82f402fe | 2353 | blk_mq_free_tag_set(&tag_set); |
f2298c04 JA |
2354 | } |
2355 | ||
2356 | module_init(null_init); | |
2357 | module_exit(null_exit); | |
2358 | ||
231b3db1 | 2359 | MODULE_AUTHOR("Jens Axboe <axboe@kernel.dk>"); |
f2298c04 | 2360 | MODULE_LICENSE("GPL"); |