xref: /linux/block/blk-sysfs.c (revision b85d45947951d23cb22d90caecf4c1eb81342c96)
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/backing-dev.h>
10 #include <linux/blktrace_api.h>
11 #include <linux/blk-mq.h>
12 #include <linux/blk-cgroup.h>
13 
14 #include "blk.h"
15 #include "blk-mq.h"
16 
17 struct queue_sysfs_entry {
18 	struct attribute attr;
19 	ssize_t (*show)(struct request_queue *, char *);
20 	ssize_t (*store)(struct request_queue *, const char *, size_t);
21 };
22 
23 static ssize_t
24 queue_var_show(unsigned long var, char *page)
25 {
26 	return sprintf(page, "%lu\n", var);
27 }
28 
29 static ssize_t
30 queue_var_store(unsigned long *var, const char *page, size_t count)
31 {
32 	int err;
33 	unsigned long v;
34 
35 	err = kstrtoul(page, 10, &v);
36 	if (err || v > UINT_MAX)
37 		return -EINVAL;
38 
39 	*var = v;
40 
41 	return count;
42 }
43 
44 static ssize_t queue_requests_show(struct request_queue *q, char *page)
45 {
46 	return queue_var_show(q->nr_requests, (page));
47 }
48 
49 static ssize_t
50 queue_requests_store(struct request_queue *q, const char *page, size_t count)
51 {
52 	unsigned long nr;
53 	int ret, err;
54 
55 	if (!q->request_fn && !q->mq_ops)
56 		return -EINVAL;
57 
58 	ret = queue_var_store(&nr, page, count);
59 	if (ret < 0)
60 		return ret;
61 
62 	if (nr < BLKDEV_MIN_RQ)
63 		nr = BLKDEV_MIN_RQ;
64 
65 	if (q->request_fn)
66 		err = blk_update_nr_requests(q, nr);
67 	else
68 		err = blk_mq_update_nr_requests(q, nr);
69 
70 	if (err)
71 		return err;
72 
73 	return ret;
74 }
75 
76 static ssize_t queue_ra_show(struct request_queue *q, char *page)
77 {
78 	unsigned long ra_kb = q->backing_dev_info.ra_pages <<
79 					(PAGE_CACHE_SHIFT - 10);
80 
81 	return queue_var_show(ra_kb, (page));
82 }
83 
84 static ssize_t
85 queue_ra_store(struct request_queue *q, const char *page, size_t count)
86 {
87 	unsigned long ra_kb;
88 	ssize_t ret = queue_var_store(&ra_kb, page, count);
89 
90 	if (ret < 0)
91 		return ret;
92 
93 	q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
94 
95 	return ret;
96 }
97 
98 static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
99 {
100 	int max_sectors_kb = queue_max_sectors(q) >> 1;
101 
102 	return queue_var_show(max_sectors_kb, (page));
103 }
104 
105 static ssize_t queue_max_segments_show(struct request_queue *q, char *page)
106 {
107 	return queue_var_show(queue_max_segments(q), (page));
108 }
109 
110 static ssize_t queue_max_integrity_segments_show(struct request_queue *q, char *page)
111 {
112 	return queue_var_show(q->limits.max_integrity_segments, (page));
113 }
114 
115 static ssize_t queue_max_segment_size_show(struct request_queue *q, char *page)
116 {
117 	if (blk_queue_cluster(q))
118 		return queue_var_show(queue_max_segment_size(q), (page));
119 
120 	return queue_var_show(PAGE_CACHE_SIZE, (page));
121 }
122 
123 static ssize_t queue_logical_block_size_show(struct request_queue *q, char *page)
124 {
125 	return queue_var_show(queue_logical_block_size(q), page);
126 }
127 
128 static ssize_t queue_physical_block_size_show(struct request_queue *q, char *page)
129 {
130 	return queue_var_show(queue_physical_block_size(q), page);
131 }
132 
133 static ssize_t queue_io_min_show(struct request_queue *q, char *page)
134 {
135 	return queue_var_show(queue_io_min(q), page);
136 }
137 
138 static ssize_t queue_io_opt_show(struct request_queue *q, char *page)
139 {
140 	return queue_var_show(queue_io_opt(q), page);
141 }
142 
143 static ssize_t queue_discard_granularity_show(struct request_queue *q, char *page)
144 {
145 	return queue_var_show(q->limits.discard_granularity, page);
146 }
147 
148 static ssize_t queue_discard_max_hw_show(struct request_queue *q, char *page)
149 {
150 	unsigned long long val;
151 
152 	val = q->limits.max_hw_discard_sectors << 9;
153 	return sprintf(page, "%llu\n", val);
154 }
155 
156 static ssize_t queue_discard_max_show(struct request_queue *q, char *page)
157 {
158 	return sprintf(page, "%llu\n",
159 		       (unsigned long long)q->limits.max_discard_sectors << 9);
160 }
161 
162 static ssize_t queue_discard_max_store(struct request_queue *q,
163 				       const char *page, size_t count)
164 {
165 	unsigned long max_discard;
166 	ssize_t ret = queue_var_store(&max_discard, page, count);
167 
168 	if (ret < 0)
169 		return ret;
170 
171 	if (max_discard & (q->limits.discard_granularity - 1))
172 		return -EINVAL;
173 
174 	max_discard >>= 9;
175 	if (max_discard > UINT_MAX)
176 		return -EINVAL;
177 
178 	if (max_discard > q->limits.max_hw_discard_sectors)
179 		max_discard = q->limits.max_hw_discard_sectors;
180 
181 	q->limits.max_discard_sectors = max_discard;
182 	return ret;
183 }
184 
185 static ssize_t queue_discard_zeroes_data_show(struct request_queue *q, char *page)
186 {
187 	return queue_var_show(queue_discard_zeroes_data(q), page);
188 }
189 
190 static ssize_t queue_write_same_max_show(struct request_queue *q, char *page)
191 {
192 	return sprintf(page, "%llu\n",
193 		(unsigned long long)q->limits.max_write_same_sectors << 9);
194 }
195 
196 
197 static ssize_t
198 queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
199 {
200 	unsigned long max_sectors_kb,
201 		max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1,
202 			page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
203 	ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
204 
205 	if (ret < 0)
206 		return ret;
207 
208 	if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
209 		return -EINVAL;
210 
211 	spin_lock_irq(q->queue_lock);
212 	q->limits.max_sectors = max_sectors_kb << 1;
213 	spin_unlock_irq(q->queue_lock);
214 
215 	return ret;
216 }
217 
218 static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
219 {
220 	int max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1;
221 
222 	return queue_var_show(max_hw_sectors_kb, (page));
223 }
224 
225 #define QUEUE_SYSFS_BIT_FNS(name, flag, neg)				\
226 static ssize_t								\
227 queue_show_##name(struct request_queue *q, char *page)			\
228 {									\
229 	int bit;							\
230 	bit = test_bit(QUEUE_FLAG_##flag, &q->queue_flags);		\
231 	return queue_var_show(neg ? !bit : bit, page);			\
232 }									\
233 static ssize_t								\
234 queue_store_##name(struct request_queue *q, const char *page, size_t count) \
235 {									\
236 	unsigned long val;						\
237 	ssize_t ret;							\
238 	ret = queue_var_store(&val, page, count);			\
239 	if (ret < 0)							\
240 		 return ret;						\
241 	if (neg)							\
242 		val = !val;						\
243 									\
244 	spin_lock_irq(q->queue_lock);					\
245 	if (val)							\
246 		queue_flag_set(QUEUE_FLAG_##flag, q);			\
247 	else								\
248 		queue_flag_clear(QUEUE_FLAG_##flag, q);			\
249 	spin_unlock_irq(q->queue_lock);					\
250 	return ret;							\
251 }
252 
253 QUEUE_SYSFS_BIT_FNS(nonrot, NONROT, 1);
254 QUEUE_SYSFS_BIT_FNS(random, ADD_RANDOM, 0);
255 QUEUE_SYSFS_BIT_FNS(iostats, IO_STAT, 0);
256 #undef QUEUE_SYSFS_BIT_FNS
257 
258 static ssize_t queue_nomerges_show(struct request_queue *q, char *page)
259 {
260 	return queue_var_show((blk_queue_nomerges(q) << 1) |
261 			       blk_queue_noxmerges(q), page);
262 }
263 
264 static ssize_t queue_nomerges_store(struct request_queue *q, const char *page,
265 				    size_t count)
266 {
267 	unsigned long nm;
268 	ssize_t ret = queue_var_store(&nm, page, count);
269 
270 	if (ret < 0)
271 		return ret;
272 
273 	spin_lock_irq(q->queue_lock);
274 	queue_flag_clear(QUEUE_FLAG_NOMERGES, q);
275 	queue_flag_clear(QUEUE_FLAG_NOXMERGES, q);
276 	if (nm == 2)
277 		queue_flag_set(QUEUE_FLAG_NOMERGES, q);
278 	else if (nm)
279 		queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
280 	spin_unlock_irq(q->queue_lock);
281 
282 	return ret;
283 }
284 
285 static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page)
286 {
287 	bool set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags);
288 	bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &q->queue_flags);
289 
290 	return queue_var_show(set << force, page);
291 }
292 
293 static ssize_t
294 queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count)
295 {
296 	ssize_t ret = -EINVAL;
297 #ifdef CONFIG_SMP
298 	unsigned long val;
299 
300 	ret = queue_var_store(&val, page, count);
301 	if (ret < 0)
302 		return ret;
303 
304 	spin_lock_irq(q->queue_lock);
305 	if (val == 2) {
306 		queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
307 		queue_flag_set(QUEUE_FLAG_SAME_FORCE, q);
308 	} else if (val == 1) {
309 		queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
310 		queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
311 	} else if (val == 0) {
312 		queue_flag_clear(QUEUE_FLAG_SAME_COMP, q);
313 		queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
314 	}
315 	spin_unlock_irq(q->queue_lock);
316 #endif
317 	return ret;
318 }
319 
320 static struct queue_sysfs_entry queue_requests_entry = {
321 	.attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
322 	.show = queue_requests_show,
323 	.store = queue_requests_store,
324 };
325 
326 static struct queue_sysfs_entry queue_ra_entry = {
327 	.attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
328 	.show = queue_ra_show,
329 	.store = queue_ra_store,
330 };
331 
332 static struct queue_sysfs_entry queue_max_sectors_entry = {
333 	.attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
334 	.show = queue_max_sectors_show,
335 	.store = queue_max_sectors_store,
336 };
337 
338 static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
339 	.attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
340 	.show = queue_max_hw_sectors_show,
341 };
342 
343 static struct queue_sysfs_entry queue_max_segments_entry = {
344 	.attr = {.name = "max_segments", .mode = S_IRUGO },
345 	.show = queue_max_segments_show,
346 };
347 
348 static struct queue_sysfs_entry queue_max_integrity_segments_entry = {
349 	.attr = {.name = "max_integrity_segments", .mode = S_IRUGO },
350 	.show = queue_max_integrity_segments_show,
351 };
352 
353 static struct queue_sysfs_entry queue_max_segment_size_entry = {
354 	.attr = {.name = "max_segment_size", .mode = S_IRUGO },
355 	.show = queue_max_segment_size_show,
356 };
357 
358 static struct queue_sysfs_entry queue_iosched_entry = {
359 	.attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
360 	.show = elv_iosched_show,
361 	.store = elv_iosched_store,
362 };
363 
364 static struct queue_sysfs_entry queue_hw_sector_size_entry = {
365 	.attr = {.name = "hw_sector_size", .mode = S_IRUGO },
366 	.show = queue_logical_block_size_show,
367 };
368 
369 static struct queue_sysfs_entry queue_logical_block_size_entry = {
370 	.attr = {.name = "logical_block_size", .mode = S_IRUGO },
371 	.show = queue_logical_block_size_show,
372 };
373 
374 static struct queue_sysfs_entry queue_physical_block_size_entry = {
375 	.attr = {.name = "physical_block_size", .mode = S_IRUGO },
376 	.show = queue_physical_block_size_show,
377 };
378 
379 static struct queue_sysfs_entry queue_io_min_entry = {
380 	.attr = {.name = "minimum_io_size", .mode = S_IRUGO },
381 	.show = queue_io_min_show,
382 };
383 
384 static struct queue_sysfs_entry queue_io_opt_entry = {
385 	.attr = {.name = "optimal_io_size", .mode = S_IRUGO },
386 	.show = queue_io_opt_show,
387 };
388 
389 static struct queue_sysfs_entry queue_discard_granularity_entry = {
390 	.attr = {.name = "discard_granularity", .mode = S_IRUGO },
391 	.show = queue_discard_granularity_show,
392 };
393 
394 static struct queue_sysfs_entry queue_discard_max_hw_entry = {
395 	.attr = {.name = "discard_max_hw_bytes", .mode = S_IRUGO },
396 	.show = queue_discard_max_hw_show,
397 };
398 
399 static struct queue_sysfs_entry queue_discard_max_entry = {
400 	.attr = {.name = "discard_max_bytes", .mode = S_IRUGO | S_IWUSR },
401 	.show = queue_discard_max_show,
402 	.store = queue_discard_max_store,
403 };
404 
405 static struct queue_sysfs_entry queue_discard_zeroes_data_entry = {
406 	.attr = {.name = "discard_zeroes_data", .mode = S_IRUGO },
407 	.show = queue_discard_zeroes_data_show,
408 };
409 
410 static struct queue_sysfs_entry queue_write_same_max_entry = {
411 	.attr = {.name = "write_same_max_bytes", .mode = S_IRUGO },
412 	.show = queue_write_same_max_show,
413 };
414 
415 static struct queue_sysfs_entry queue_nonrot_entry = {
416 	.attr = {.name = "rotational", .mode = S_IRUGO | S_IWUSR },
417 	.show = queue_show_nonrot,
418 	.store = queue_store_nonrot,
419 };
420 
421 static struct queue_sysfs_entry queue_nomerges_entry = {
422 	.attr = {.name = "nomerges", .mode = S_IRUGO | S_IWUSR },
423 	.show = queue_nomerges_show,
424 	.store = queue_nomerges_store,
425 };
426 
427 static struct queue_sysfs_entry queue_rq_affinity_entry = {
428 	.attr = {.name = "rq_affinity", .mode = S_IRUGO | S_IWUSR },
429 	.show = queue_rq_affinity_show,
430 	.store = queue_rq_affinity_store,
431 };
432 
433 static struct queue_sysfs_entry queue_iostats_entry = {
434 	.attr = {.name = "iostats", .mode = S_IRUGO | S_IWUSR },
435 	.show = queue_show_iostats,
436 	.store = queue_store_iostats,
437 };
438 
439 static struct queue_sysfs_entry queue_random_entry = {
440 	.attr = {.name = "add_random", .mode = S_IRUGO | S_IWUSR },
441 	.show = queue_show_random,
442 	.store = queue_store_random,
443 };
444 
445 static struct attribute *default_attrs[] = {
446 	&queue_requests_entry.attr,
447 	&queue_ra_entry.attr,
448 	&queue_max_hw_sectors_entry.attr,
449 	&queue_max_sectors_entry.attr,
450 	&queue_max_segments_entry.attr,
451 	&queue_max_integrity_segments_entry.attr,
452 	&queue_max_segment_size_entry.attr,
453 	&queue_iosched_entry.attr,
454 	&queue_hw_sector_size_entry.attr,
455 	&queue_logical_block_size_entry.attr,
456 	&queue_physical_block_size_entry.attr,
457 	&queue_io_min_entry.attr,
458 	&queue_io_opt_entry.attr,
459 	&queue_discard_granularity_entry.attr,
460 	&queue_discard_max_entry.attr,
461 	&queue_discard_max_hw_entry.attr,
462 	&queue_discard_zeroes_data_entry.attr,
463 	&queue_write_same_max_entry.attr,
464 	&queue_nonrot_entry.attr,
465 	&queue_nomerges_entry.attr,
466 	&queue_rq_affinity_entry.attr,
467 	&queue_iostats_entry.attr,
468 	&queue_random_entry.attr,
469 	NULL,
470 };
471 
472 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
473 
474 static ssize_t
475 queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
476 {
477 	struct queue_sysfs_entry *entry = to_queue(attr);
478 	struct request_queue *q =
479 		container_of(kobj, struct request_queue, kobj);
480 	ssize_t res;
481 
482 	if (!entry->show)
483 		return -EIO;
484 	mutex_lock(&q->sysfs_lock);
485 	if (blk_queue_dying(q)) {
486 		mutex_unlock(&q->sysfs_lock);
487 		return -ENOENT;
488 	}
489 	res = entry->show(q, page);
490 	mutex_unlock(&q->sysfs_lock);
491 	return res;
492 }
493 
494 static ssize_t
495 queue_attr_store(struct kobject *kobj, struct attribute *attr,
496 		    const char *page, size_t length)
497 {
498 	struct queue_sysfs_entry *entry = to_queue(attr);
499 	struct request_queue *q;
500 	ssize_t res;
501 
502 	if (!entry->store)
503 		return -EIO;
504 
505 	q = container_of(kobj, struct request_queue, kobj);
506 	mutex_lock(&q->sysfs_lock);
507 	if (blk_queue_dying(q)) {
508 		mutex_unlock(&q->sysfs_lock);
509 		return -ENOENT;
510 	}
511 	res = entry->store(q, page, length);
512 	mutex_unlock(&q->sysfs_lock);
513 	return res;
514 }
515 
516 static void blk_free_queue_rcu(struct rcu_head *rcu_head)
517 {
518 	struct request_queue *q = container_of(rcu_head, struct request_queue,
519 					       rcu_head);
520 	kmem_cache_free(blk_requestq_cachep, q);
521 }
522 
523 /**
524  * blk_release_queue: - release a &struct request_queue when it is no longer needed
525  * @kobj:    the kobj belonging to the request queue to be released
526  *
527  * Description:
528  *     blk_release_queue is the pair to blk_init_queue() or
529  *     blk_queue_make_request().  It should be called when a request queue is
530  *     being released; typically when a block device is being de-registered.
531  *     Currently, its primary task it to free all the &struct request
532  *     structures that were allocated to the queue and the queue itself.
533  *
534  * Note:
535  *     The low level driver must have finished any outstanding requests first
536  *     via blk_cleanup_queue().
537  **/
538 static void blk_release_queue(struct kobject *kobj)
539 {
540 	struct request_queue *q =
541 		container_of(kobj, struct request_queue, kobj);
542 
543 	bdi_exit(&q->backing_dev_info);
544 	blkcg_exit_queue(q);
545 
546 	if (q->elevator) {
547 		spin_lock_irq(q->queue_lock);
548 		ioc_clear_queue(q);
549 		spin_unlock_irq(q->queue_lock);
550 		elevator_exit(q->elevator);
551 	}
552 
553 	blk_exit_rl(&q->root_rl);
554 
555 	if (q->queue_tags)
556 		__blk_queue_free_tags(q);
557 
558 	if (!q->mq_ops)
559 		blk_free_flush_queue(q->fq);
560 	else
561 		blk_mq_release(q);
562 
563 	blk_trace_shutdown(q);
564 
565 	if (q->bio_split)
566 		bioset_free(q->bio_split);
567 
568 	ida_simple_remove(&blk_queue_ida, q->id);
569 	call_rcu(&q->rcu_head, blk_free_queue_rcu);
570 }
571 
572 static const struct sysfs_ops queue_sysfs_ops = {
573 	.show	= queue_attr_show,
574 	.store	= queue_attr_store,
575 };
576 
577 struct kobj_type blk_queue_ktype = {
578 	.sysfs_ops	= &queue_sysfs_ops,
579 	.default_attrs	= default_attrs,
580 	.release	= blk_release_queue,
581 };
582 
583 int blk_register_queue(struct gendisk *disk)
584 {
585 	int ret;
586 	struct device *dev = disk_to_dev(disk);
587 	struct request_queue *q = disk->queue;
588 
589 	if (WARN_ON(!q))
590 		return -ENXIO;
591 
592 	/*
593 	 * SCSI probing may synchronously create and destroy a lot of
594 	 * request_queues for non-existent devices.  Shutting down a fully
595 	 * functional queue takes measureable wallclock time as RCU grace
596 	 * periods are involved.  To avoid excessive latency in these
597 	 * cases, a request_queue starts out in a degraded mode which is
598 	 * faster to shut down and is made fully functional here as
599 	 * request_queues for non-existent devices never get registered.
600 	 */
601 	if (!blk_queue_init_done(q)) {
602 		queue_flag_set_unlocked(QUEUE_FLAG_INIT_DONE, q);
603 		blk_queue_bypass_end(q);
604 		if (q->mq_ops)
605 			blk_mq_finish_init(q);
606 	}
607 
608 	ret = blk_trace_init_sysfs(dev);
609 	if (ret)
610 		return ret;
611 
612 	ret = kobject_add(&q->kobj, kobject_get(&dev->kobj), "%s", "queue");
613 	if (ret < 0) {
614 		blk_trace_remove_sysfs(dev);
615 		return ret;
616 	}
617 
618 	kobject_uevent(&q->kobj, KOBJ_ADD);
619 
620 	if (q->mq_ops)
621 		blk_mq_register_disk(disk);
622 
623 	if (!q->request_fn)
624 		return 0;
625 
626 	ret = elv_register_queue(q);
627 	if (ret) {
628 		kobject_uevent(&q->kobj, KOBJ_REMOVE);
629 		kobject_del(&q->kobj);
630 		blk_trace_remove_sysfs(dev);
631 		kobject_put(&dev->kobj);
632 		return ret;
633 	}
634 
635 	return 0;
636 }
637 
638 void blk_unregister_queue(struct gendisk *disk)
639 {
640 	struct request_queue *q = disk->queue;
641 
642 	if (WARN_ON(!q))
643 		return;
644 
645 	if (q->mq_ops)
646 		blk_mq_unregister_disk(disk);
647 
648 	if (q->request_fn)
649 		elv_unregister_queue(q);
650 
651 	kobject_uevent(&q->kobj, KOBJ_REMOVE);
652 	kobject_del(&q->kobj);
653 	blk_trace_remove_sysfs(disk_to_dev(disk));
654 	kobject_put(&disk_to_dev(disk)->kobj);
655 }
656