]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - block/blk-sysfs.c
blkcg: implement per-blkg request allocation
[mirror_ubuntu-zesty-kernel.git] / block / blk-sysfs.c
1 /*
2 * Functions related to sysfs handling
3 */
4 #include <linux/kernel.h>
5 #include <linux/slab.h>
6 #include <linux/module.h>
7 #include <linux/bio.h>
8 #include <linux/blkdev.h>
9 #include <linux/blktrace_api.h>
10
11 #include "blk.h"
12 #include "blk-cgroup.h"
13
14 struct queue_sysfs_entry {
15 struct attribute attr;
16 ssize_t (*show)(struct request_queue *, char *);
17 ssize_t (*store)(struct request_queue *, const char *, size_t);
18 };
19
20 static ssize_t
21 queue_var_show(unsigned long var, char *page)
22 {
23 return sprintf(page, "%lu\n", var);
24 }
25
26 static ssize_t
27 queue_var_store(unsigned long *var, const char *page, size_t count)
28 {
29 char *p = (char *) page;
30
31 *var = simple_strtoul(p, &p, 10);
32 return count;
33 }
34
35 static ssize_t queue_requests_show(struct request_queue *q, char *page)
36 {
37 return queue_var_show(q->nr_requests, (page));
38 }
39
40 static ssize_t
41 queue_requests_store(struct request_queue *q, const char *page, size_t count)
42 {
43 struct request_list *rl;
44 unsigned long nr;
45 int ret;
46
47 if (!q->request_fn)
48 return -EINVAL;
49
50 ret = queue_var_store(&nr, page, count);
51 if (nr < BLKDEV_MIN_RQ)
52 nr = BLKDEV_MIN_RQ;
53
54 spin_lock_irq(q->queue_lock);
55 q->nr_requests = nr;
56 blk_queue_congestion_threshold(q);
57
58 /* congestion isn't cgroup aware and follows root blkcg for now */
59 rl = &q->root_rl;
60
61 if (rl->count[BLK_RW_SYNC] >= queue_congestion_on_threshold(q))
62 blk_set_queue_congested(q, BLK_RW_SYNC);
63 else if (rl->count[BLK_RW_SYNC] < queue_congestion_off_threshold(q))
64 blk_clear_queue_congested(q, BLK_RW_SYNC);
65
66 if (rl->count[BLK_RW_ASYNC] >= queue_congestion_on_threshold(q))
67 blk_set_queue_congested(q, BLK_RW_ASYNC);
68 else if (rl->count[BLK_RW_ASYNC] < queue_congestion_off_threshold(q))
69 blk_clear_queue_congested(q, BLK_RW_ASYNC);
70
71 blk_queue_for_each_rl(rl, q) {
72 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
73 blk_set_rl_full(rl, BLK_RW_SYNC);
74 } else {
75 blk_clear_rl_full(rl, BLK_RW_SYNC);
76 wake_up(&rl->wait[BLK_RW_SYNC]);
77 }
78
79 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
80 blk_set_rl_full(rl, BLK_RW_ASYNC);
81 } else {
82 blk_clear_rl_full(rl, BLK_RW_ASYNC);
83 wake_up(&rl->wait[BLK_RW_ASYNC]);
84 }
85 }
86
87 spin_unlock_irq(q->queue_lock);
88 return ret;
89 }
90
91 static ssize_t queue_ra_show(struct request_queue *q, char *page)
92 {
93 unsigned long ra_kb = q->backing_dev_info.ra_pages <<
94 (PAGE_CACHE_SHIFT - 10);
95
96 return queue_var_show(ra_kb, (page));
97 }
98
99 static ssize_t
100 queue_ra_store(struct request_queue *q, const char *page, size_t count)
101 {
102 unsigned long ra_kb;
103 ssize_t ret = queue_var_store(&ra_kb, page, count);
104
105 q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
106
107 return ret;
108 }
109
110 static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
111 {
112 int max_sectors_kb = queue_max_sectors(q) >> 1;
113
114 return queue_var_show(max_sectors_kb, (page));
115 }
116
117 static ssize_t queue_max_segments_show(struct request_queue *q, char *page)
118 {
119 return queue_var_show(queue_max_segments(q), (page));
120 }
121
122 static ssize_t queue_max_integrity_segments_show(struct request_queue *q, char *page)
123 {
124 return queue_var_show(q->limits.max_integrity_segments, (page));
125 }
126
127 static ssize_t queue_max_segment_size_show(struct request_queue *q, char *page)
128 {
129 if (blk_queue_cluster(q))
130 return queue_var_show(queue_max_segment_size(q), (page));
131
132 return queue_var_show(PAGE_CACHE_SIZE, (page));
133 }
134
135 static ssize_t queue_logical_block_size_show(struct request_queue *q, char *page)
136 {
137 return queue_var_show(queue_logical_block_size(q), page);
138 }
139
140 static ssize_t queue_physical_block_size_show(struct request_queue *q, char *page)
141 {
142 return queue_var_show(queue_physical_block_size(q), page);
143 }
144
145 static ssize_t queue_io_min_show(struct request_queue *q, char *page)
146 {
147 return queue_var_show(queue_io_min(q), page);
148 }
149
150 static ssize_t queue_io_opt_show(struct request_queue *q, char *page)
151 {
152 return queue_var_show(queue_io_opt(q), page);
153 }
154
155 static ssize_t queue_discard_granularity_show(struct request_queue *q, char *page)
156 {
157 return queue_var_show(q->limits.discard_granularity, page);
158 }
159
160 static ssize_t queue_discard_max_show(struct request_queue *q, char *page)
161 {
162 return sprintf(page, "%llu\n",
163 (unsigned long long)q->limits.max_discard_sectors << 9);
164 }
165
166 static ssize_t queue_discard_zeroes_data_show(struct request_queue *q, char *page)
167 {
168 return queue_var_show(queue_discard_zeroes_data(q), page);
169 }
170
171 static ssize_t
172 queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
173 {
174 unsigned long max_sectors_kb,
175 max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1,
176 page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
177 ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
178
179 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
180 return -EINVAL;
181
182 spin_lock_irq(q->queue_lock);
183 q->limits.max_sectors = max_sectors_kb << 1;
184 spin_unlock_irq(q->queue_lock);
185
186 return ret;
187 }
188
189 static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
190 {
191 int max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1;
192
193 return queue_var_show(max_hw_sectors_kb, (page));
194 }
195
196 #define QUEUE_SYSFS_BIT_FNS(name, flag, neg) \
197 static ssize_t \
198 queue_show_##name(struct request_queue *q, char *page) \
199 { \
200 int bit; \
201 bit = test_bit(QUEUE_FLAG_##flag, &q->queue_flags); \
202 return queue_var_show(neg ? !bit : bit, page); \
203 } \
204 static ssize_t \
205 queue_store_##name(struct request_queue *q, const char *page, size_t count) \
206 { \
207 unsigned long val; \
208 ssize_t ret; \
209 ret = queue_var_store(&val, page, count); \
210 if (neg) \
211 val = !val; \
212 \
213 spin_lock_irq(q->queue_lock); \
214 if (val) \
215 queue_flag_set(QUEUE_FLAG_##flag, q); \
216 else \
217 queue_flag_clear(QUEUE_FLAG_##flag, q); \
218 spin_unlock_irq(q->queue_lock); \
219 return ret; \
220 }
221
222 QUEUE_SYSFS_BIT_FNS(nonrot, NONROT, 1);
223 QUEUE_SYSFS_BIT_FNS(random, ADD_RANDOM, 0);
224 QUEUE_SYSFS_BIT_FNS(iostats, IO_STAT, 0);
225 #undef QUEUE_SYSFS_BIT_FNS
226
227 static ssize_t queue_nomerges_show(struct request_queue *q, char *page)
228 {
229 return queue_var_show((blk_queue_nomerges(q) << 1) |
230 blk_queue_noxmerges(q), page);
231 }
232
233 static ssize_t queue_nomerges_store(struct request_queue *q, const char *page,
234 size_t count)
235 {
236 unsigned long nm;
237 ssize_t ret = queue_var_store(&nm, page, count);
238
239 spin_lock_irq(q->queue_lock);
240 queue_flag_clear(QUEUE_FLAG_NOMERGES, q);
241 queue_flag_clear(QUEUE_FLAG_NOXMERGES, q);
242 if (nm == 2)
243 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
244 else if (nm)
245 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
246 spin_unlock_irq(q->queue_lock);
247
248 return ret;
249 }
250
251 static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page)
252 {
253 bool set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags);
254 bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &q->queue_flags);
255
256 return queue_var_show(set << force, page);
257 }
258
259 static ssize_t
260 queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count)
261 {
262 ssize_t ret = -EINVAL;
263 #if defined(CONFIG_USE_GENERIC_SMP_HELPERS)
264 unsigned long val;
265
266 ret = queue_var_store(&val, page, count);
267 spin_lock_irq(q->queue_lock);
268 if (val == 2) {
269 queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
270 queue_flag_set(QUEUE_FLAG_SAME_FORCE, q);
271 } else if (val == 1) {
272 queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
273 queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
274 } else if (val == 0) {
275 queue_flag_clear(QUEUE_FLAG_SAME_COMP, q);
276 queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
277 }
278 spin_unlock_irq(q->queue_lock);
279 #endif
280 return ret;
281 }
282
283 static struct queue_sysfs_entry queue_requests_entry = {
284 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
285 .show = queue_requests_show,
286 .store = queue_requests_store,
287 };
288
289 static struct queue_sysfs_entry queue_ra_entry = {
290 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
291 .show = queue_ra_show,
292 .store = queue_ra_store,
293 };
294
295 static struct queue_sysfs_entry queue_max_sectors_entry = {
296 .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
297 .show = queue_max_sectors_show,
298 .store = queue_max_sectors_store,
299 };
300
301 static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
302 .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
303 .show = queue_max_hw_sectors_show,
304 };
305
306 static struct queue_sysfs_entry queue_max_segments_entry = {
307 .attr = {.name = "max_segments", .mode = S_IRUGO },
308 .show = queue_max_segments_show,
309 };
310
311 static struct queue_sysfs_entry queue_max_integrity_segments_entry = {
312 .attr = {.name = "max_integrity_segments", .mode = S_IRUGO },
313 .show = queue_max_integrity_segments_show,
314 };
315
316 static struct queue_sysfs_entry queue_max_segment_size_entry = {
317 .attr = {.name = "max_segment_size", .mode = S_IRUGO },
318 .show = queue_max_segment_size_show,
319 };
320
321 static struct queue_sysfs_entry queue_iosched_entry = {
322 .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
323 .show = elv_iosched_show,
324 .store = elv_iosched_store,
325 };
326
327 static struct queue_sysfs_entry queue_hw_sector_size_entry = {
328 .attr = {.name = "hw_sector_size", .mode = S_IRUGO },
329 .show = queue_logical_block_size_show,
330 };
331
332 static struct queue_sysfs_entry queue_logical_block_size_entry = {
333 .attr = {.name = "logical_block_size", .mode = S_IRUGO },
334 .show = queue_logical_block_size_show,
335 };
336
337 static struct queue_sysfs_entry queue_physical_block_size_entry = {
338 .attr = {.name = "physical_block_size", .mode = S_IRUGO },
339 .show = queue_physical_block_size_show,
340 };
341
342 static struct queue_sysfs_entry queue_io_min_entry = {
343 .attr = {.name = "minimum_io_size", .mode = S_IRUGO },
344 .show = queue_io_min_show,
345 };
346
347 static struct queue_sysfs_entry queue_io_opt_entry = {
348 .attr = {.name = "optimal_io_size", .mode = S_IRUGO },
349 .show = queue_io_opt_show,
350 };
351
352 static struct queue_sysfs_entry queue_discard_granularity_entry = {
353 .attr = {.name = "discard_granularity", .mode = S_IRUGO },
354 .show = queue_discard_granularity_show,
355 };
356
357 static struct queue_sysfs_entry queue_discard_max_entry = {
358 .attr = {.name = "discard_max_bytes", .mode = S_IRUGO },
359 .show = queue_discard_max_show,
360 };
361
362 static struct queue_sysfs_entry queue_discard_zeroes_data_entry = {
363 .attr = {.name = "discard_zeroes_data", .mode = S_IRUGO },
364 .show = queue_discard_zeroes_data_show,
365 };
366
367 static struct queue_sysfs_entry queue_nonrot_entry = {
368 .attr = {.name = "rotational", .mode = S_IRUGO | S_IWUSR },
369 .show = queue_show_nonrot,
370 .store = queue_store_nonrot,
371 };
372
373 static struct queue_sysfs_entry queue_nomerges_entry = {
374 .attr = {.name = "nomerges", .mode = S_IRUGO | S_IWUSR },
375 .show = queue_nomerges_show,
376 .store = queue_nomerges_store,
377 };
378
379 static struct queue_sysfs_entry queue_rq_affinity_entry = {
380 .attr = {.name = "rq_affinity", .mode = S_IRUGO | S_IWUSR },
381 .show = queue_rq_affinity_show,
382 .store = queue_rq_affinity_store,
383 };
384
385 static struct queue_sysfs_entry queue_iostats_entry = {
386 .attr = {.name = "iostats", .mode = S_IRUGO | S_IWUSR },
387 .show = queue_show_iostats,
388 .store = queue_store_iostats,
389 };
390
391 static struct queue_sysfs_entry queue_random_entry = {
392 .attr = {.name = "add_random", .mode = S_IRUGO | S_IWUSR },
393 .show = queue_show_random,
394 .store = queue_store_random,
395 };
396
397 static struct attribute *default_attrs[] = {
398 &queue_requests_entry.attr,
399 &queue_ra_entry.attr,
400 &queue_max_hw_sectors_entry.attr,
401 &queue_max_sectors_entry.attr,
402 &queue_max_segments_entry.attr,
403 &queue_max_integrity_segments_entry.attr,
404 &queue_max_segment_size_entry.attr,
405 &queue_iosched_entry.attr,
406 &queue_hw_sector_size_entry.attr,
407 &queue_logical_block_size_entry.attr,
408 &queue_physical_block_size_entry.attr,
409 &queue_io_min_entry.attr,
410 &queue_io_opt_entry.attr,
411 &queue_discard_granularity_entry.attr,
412 &queue_discard_max_entry.attr,
413 &queue_discard_zeroes_data_entry.attr,
414 &queue_nonrot_entry.attr,
415 &queue_nomerges_entry.attr,
416 &queue_rq_affinity_entry.attr,
417 &queue_iostats_entry.attr,
418 &queue_random_entry.attr,
419 NULL,
420 };
421
422 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
423
424 static ssize_t
425 queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
426 {
427 struct queue_sysfs_entry *entry = to_queue(attr);
428 struct request_queue *q =
429 container_of(kobj, struct request_queue, kobj);
430 ssize_t res;
431
432 if (!entry->show)
433 return -EIO;
434 mutex_lock(&q->sysfs_lock);
435 if (blk_queue_dead(q)) {
436 mutex_unlock(&q->sysfs_lock);
437 return -ENOENT;
438 }
439 res = entry->show(q, page);
440 mutex_unlock(&q->sysfs_lock);
441 return res;
442 }
443
444 static ssize_t
445 queue_attr_store(struct kobject *kobj, struct attribute *attr,
446 const char *page, size_t length)
447 {
448 struct queue_sysfs_entry *entry = to_queue(attr);
449 struct request_queue *q;
450 ssize_t res;
451
452 if (!entry->store)
453 return -EIO;
454
455 q = container_of(kobj, struct request_queue, kobj);
456 mutex_lock(&q->sysfs_lock);
457 if (blk_queue_dead(q)) {
458 mutex_unlock(&q->sysfs_lock);
459 return -ENOENT;
460 }
461 res = entry->store(q, page, length);
462 mutex_unlock(&q->sysfs_lock);
463 return res;
464 }
465
466 /**
467 * blk_release_queue: - release a &struct request_queue when it is no longer needed
468 * @kobj: the kobj belonging to the request queue to be released
469 *
470 * Description:
471 * blk_release_queue is the pair to blk_init_queue() or
472 * blk_queue_make_request(). It should be called when a request queue is
473 * being released; typically when a block device is being de-registered.
474 * Currently, its primary task it to free all the &struct request
475 * structures that were allocated to the queue and the queue itself.
476 *
477 * Caveat:
478 * Hopefully the low level driver will have finished any
479 * outstanding requests first...
480 **/
481 static void blk_release_queue(struct kobject *kobj)
482 {
483 struct request_queue *q =
484 container_of(kobj, struct request_queue, kobj);
485
486 blk_sync_queue(q);
487
488 blkcg_exit_queue(q);
489
490 if (q->elevator) {
491 spin_lock_irq(q->queue_lock);
492 ioc_clear_queue(q);
493 spin_unlock_irq(q->queue_lock);
494 elevator_exit(q->elevator);
495 }
496
497 blk_exit_rl(&q->root_rl);
498
499 if (q->queue_tags)
500 __blk_queue_free_tags(q);
501
502 blk_trace_shutdown(q);
503
504 bdi_destroy(&q->backing_dev_info);
505
506 ida_simple_remove(&blk_queue_ida, q->id);
507 kmem_cache_free(blk_requestq_cachep, q);
508 }
509
510 static const struct sysfs_ops queue_sysfs_ops = {
511 .show = queue_attr_show,
512 .store = queue_attr_store,
513 };
514
515 struct kobj_type blk_queue_ktype = {
516 .sysfs_ops = &queue_sysfs_ops,
517 .default_attrs = default_attrs,
518 .release = blk_release_queue,
519 };
520
521 int blk_register_queue(struct gendisk *disk)
522 {
523 int ret;
524 struct device *dev = disk_to_dev(disk);
525 struct request_queue *q = disk->queue;
526
527 if (WARN_ON(!q))
528 return -ENXIO;
529
530 ret = blk_trace_init_sysfs(dev);
531 if (ret)
532 return ret;
533
534 ret = kobject_add(&q->kobj, kobject_get(&dev->kobj), "%s", "queue");
535 if (ret < 0) {
536 blk_trace_remove_sysfs(dev);
537 return ret;
538 }
539
540 kobject_uevent(&q->kobj, KOBJ_ADD);
541
542 if (!q->request_fn)
543 return 0;
544
545 ret = elv_register_queue(q);
546 if (ret) {
547 kobject_uevent(&q->kobj, KOBJ_REMOVE);
548 kobject_del(&q->kobj);
549 blk_trace_remove_sysfs(dev);
550 kobject_put(&dev->kobj);
551 return ret;
552 }
553
554 return 0;
555 }
556
557 void blk_unregister_queue(struct gendisk *disk)
558 {
559 struct request_queue *q = disk->queue;
560
561 if (WARN_ON(!q))
562 return;
563
564 if (q->request_fn)
565 elv_unregister_queue(q);
566
567 kobject_uevent(&q->kobj, KOBJ_REMOVE);
568 kobject_del(&q->kobj);
569 blk_trace_remove_sysfs(disk_to_dev(disk));
570 kobject_put(&disk_to_dev(disk)->kobj);
571 }