nvmap: page pools: replace background allocator with background zeroer
[linux-3.10.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         int err;
30         unsigned long v;
31
32         err = strict_strtoul(page, 10, &v);
33         if (err || v > UINT_MAX)
34                 return -EINVAL;
35
36         *var = v;
37
38         return count;
39 }
40
41 static ssize_t queue_requests_show(struct request_queue *q, char *page)
42 {
43         return queue_var_show(q->nr_requests, (page));
44 }
45
46 static ssize_t
47 queue_requests_store(struct request_queue *q, const char *page, size_t count)
48 {
49         struct request_list *rl;
50         unsigned long nr;
51         int ret;
52
53         if (!q->request_fn)
54                 return -EINVAL;
55
56         ret = queue_var_store(&nr, page, count);
57         if (ret < 0)
58                 return ret;
59
60         if (nr < BLKDEV_MIN_RQ)
61                 nr = BLKDEV_MIN_RQ;
62
63         spin_lock_irq(q->queue_lock);
64         q->nr_requests = nr;
65         blk_queue_congestion_threshold(q);
66
67         /* congestion isn't cgroup aware and follows root blkcg for now */
68         rl = &q->root_rl;
69
70         if (rl->count[BLK_RW_SYNC] >= queue_congestion_on_threshold(q))
71                 blk_set_queue_congested(q, BLK_RW_SYNC);
72         else if (rl->count[BLK_RW_SYNC] < queue_congestion_off_threshold(q))
73                 blk_clear_queue_congested(q, BLK_RW_SYNC);
74
75         if (rl->count[BLK_RW_ASYNC] >= queue_congestion_on_threshold(q))
76                 blk_set_queue_congested(q, BLK_RW_ASYNC);
77         else if (rl->count[BLK_RW_ASYNC] < queue_congestion_off_threshold(q))
78                 blk_clear_queue_congested(q, BLK_RW_ASYNC);
79
80         blk_queue_for_each_rl(rl, q) {
81                 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
82                         blk_set_rl_full(rl, BLK_RW_SYNC);
83                 } else {
84                         blk_clear_rl_full(rl, BLK_RW_SYNC);
85                         wake_up(&rl->wait[BLK_RW_SYNC]);
86                 }
87
88                 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
89                         blk_set_rl_full(rl, BLK_RW_ASYNC);
90                 } else {
91                         blk_clear_rl_full(rl, BLK_RW_ASYNC);
92                         wake_up(&rl->wait[BLK_RW_ASYNC]);
93                 }
94         }
95
96         spin_unlock_irq(q->queue_lock);
97         return ret;
98 }
99
100 static ssize_t queue_ra_show(struct request_queue *q, char *page)
101 {
102         unsigned long ra_kb = q->backing_dev_info.ra_pages <<
103                                         (PAGE_CACHE_SHIFT - 10);
104
105         return queue_var_show(ra_kb, (page));
106 }
107
108 static ssize_t
109 queue_ra_store(struct request_queue *q, const char *page, size_t count)
110 {
111         unsigned long ra_kb;
112         ssize_t ret = queue_var_store(&ra_kb, page, count);
113
114         if (ret < 0)
115                 return ret;
116
117         q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
118
119         return ret;
120 }
121
122 static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
123 {
124         int max_sectors_kb = queue_max_sectors(q) >> 1;
125
126         return queue_var_show(max_sectors_kb, (page));
127 }
128
129 static ssize_t queue_max_segments_show(struct request_queue *q, char *page)
130 {
131         return queue_var_show(queue_max_segments(q), (page));
132 }
133
134 static ssize_t queue_max_integrity_segments_show(struct request_queue *q, char *page)
135 {
136         return queue_var_show(q->limits.max_integrity_segments, (page));
137 }
138
139 static ssize_t queue_max_segment_size_show(struct request_queue *q, char *page)
140 {
141         if (blk_queue_cluster(q))
142                 return queue_var_show(queue_max_segment_size(q), (page));
143
144         return queue_var_show(PAGE_CACHE_SIZE, (page));
145 }
146
147 static ssize_t queue_logical_block_size_show(struct request_queue *q, char *page)
148 {
149         return queue_var_show(queue_logical_block_size(q), page);
150 }
151
152 static ssize_t queue_physical_block_size_show(struct request_queue *q, char *page)
153 {
154         return queue_var_show(queue_physical_block_size(q), page);
155 }
156
157 static ssize_t queue_io_min_show(struct request_queue *q, char *page)
158 {
159         return queue_var_show(queue_io_min(q), page);
160 }
161
162 static ssize_t queue_io_opt_show(struct request_queue *q, char *page)
163 {
164         return queue_var_show(queue_io_opt(q), page);
165 }
166
167 static ssize_t queue_discard_granularity_show(struct request_queue *q, char *page)
168 {
169         return queue_var_show(q->limits.discard_granularity, page);
170 }
171
172 static ssize_t queue_discard_max_show(struct request_queue *q, char *page)
173 {
174         return sprintf(page, "%llu\n",
175                        (unsigned long long)q->limits.max_discard_sectors << 9);
176 }
177
178 static ssize_t queue_discard_zeroes_data_show(struct request_queue *q, char *page)
179 {
180         return queue_var_show(queue_discard_zeroes_data(q), page);
181 }
182
183 static ssize_t queue_write_same_max_show(struct request_queue *q, char *page)
184 {
185         return sprintf(page, "%llu\n",
186                 (unsigned long long)q->limits.max_write_same_sectors << 9);
187 }
188
189
190 static ssize_t
191 queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
192 {
193         unsigned long max_sectors_kb,
194                 max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1,
195                         page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
196         ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
197
198         if (ret < 0)
199                 return ret;
200
201         if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
202                 return -EINVAL;
203
204         spin_lock_irq(q->queue_lock);
205         q->limits.max_sectors = max_sectors_kb << 1;
206         spin_unlock_irq(q->queue_lock);
207
208         return ret;
209 }
210
211 static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
212 {
213         int max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1;
214
215         return queue_var_show(max_hw_sectors_kb, (page));
216 }
217
218 #define QUEUE_SYSFS_BIT_FNS(name, flag, neg)                            \
219 static ssize_t                                                          \
220 queue_show_##name(struct request_queue *q, char *page)                  \
221 {                                                                       \
222         int bit;                                                        \
223         bit = test_bit(QUEUE_FLAG_##flag, &q->queue_flags);             \
224         return queue_var_show(neg ? !bit : bit, page);                  \
225 }                                                                       \
226 static ssize_t                                                          \
227 queue_store_##name(struct request_queue *q, const char *page, size_t count) \
228 {                                                                       \
229         unsigned long val;                                              \
230         ssize_t ret;                                                    \
231         ret = queue_var_store(&val, page, count);                       \
232         if (ret < 0)                                                    \
233                  return ret;                                            \
234         if (neg)                                                        \
235                 val = !val;                                             \
236                                                                         \
237         spin_lock_irq(q->queue_lock);                                   \
238         if (val)                                                        \
239                 queue_flag_set(QUEUE_FLAG_##flag, q);                   \
240         else                                                            \
241                 queue_flag_clear(QUEUE_FLAG_##flag, q);                 \
242         spin_unlock_irq(q->queue_lock);                                 \
243         return ret;                                                     \
244 }
245
246 QUEUE_SYSFS_BIT_FNS(nonrot, NONROT, 1);
247 QUEUE_SYSFS_BIT_FNS(random, ADD_RANDOM, 0);
248 QUEUE_SYSFS_BIT_FNS(iostats, IO_STAT, 0);
249 #undef QUEUE_SYSFS_BIT_FNS
250
251 static ssize_t queue_nomerges_show(struct request_queue *q, char *page)
252 {
253         return queue_var_show((blk_queue_nomerges(q) << 1) |
254                                blk_queue_noxmerges(q), page);
255 }
256
257 static ssize_t queue_nomerges_store(struct request_queue *q, const char *page,
258                                     size_t count)
259 {
260         unsigned long nm;
261         ssize_t ret = queue_var_store(&nm, page, count);
262
263         if (ret < 0)
264                 return ret;
265
266         spin_lock_irq(q->queue_lock);
267         queue_flag_clear(QUEUE_FLAG_NOMERGES, q);
268         queue_flag_clear(QUEUE_FLAG_NOXMERGES, q);
269         if (nm == 2)
270                 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
271         else if (nm)
272                 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
273         spin_unlock_irq(q->queue_lock);
274
275         return ret;
276 }
277
278 static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page)
279 {
280         bool set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags);
281         bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &q->queue_flags);
282
283         return queue_var_show(set << force, page);
284 }
285
286 static ssize_t
287 queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count)
288 {
289         ssize_t ret = -EINVAL;
290 #ifdef CONFIG_SMP
291         unsigned long val;
292
293         ret = queue_var_store(&val, page, count);
294         if (ret < 0)
295                 return ret;
296
297         spin_lock_irq(q->queue_lock);
298         if (val == 2) {
299                 queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
300                 queue_flag_set(QUEUE_FLAG_SAME_FORCE, q);
301         } else if (val == 1) {
302                 queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
303                 queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
304         } else if (val == 0) {
305                 queue_flag_clear(QUEUE_FLAG_SAME_COMP, q);
306                 queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
307         }
308         spin_unlock_irq(q->queue_lock);
309 #endif
310         return ret;
311 }
312
313 static struct queue_sysfs_entry queue_requests_entry = {
314         .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
315         .show = queue_requests_show,
316         .store = queue_requests_store,
317 };
318
319 static struct queue_sysfs_entry queue_ra_entry = {
320         .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
321         .show = queue_ra_show,
322         .store = queue_ra_store,
323 };
324
325 static struct queue_sysfs_entry queue_max_sectors_entry = {
326         .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
327         .show = queue_max_sectors_show,
328         .store = queue_max_sectors_store,
329 };
330
331 static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
332         .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
333         .show = queue_max_hw_sectors_show,
334 };
335
336 static struct queue_sysfs_entry queue_max_segments_entry = {
337         .attr = {.name = "max_segments", .mode = S_IRUGO },
338         .show = queue_max_segments_show,
339 };
340
341 static struct queue_sysfs_entry queue_max_integrity_segments_entry = {
342         .attr = {.name = "max_integrity_segments", .mode = S_IRUGO },
343         .show = queue_max_integrity_segments_show,
344 };
345
346 static struct queue_sysfs_entry queue_max_segment_size_entry = {
347         .attr = {.name = "max_segment_size", .mode = S_IRUGO },
348         .show = queue_max_segment_size_show,
349 };
350
351 static struct queue_sysfs_entry queue_iosched_entry = {
352         .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
353         .show = elv_iosched_show,
354         .store = elv_iosched_store,
355 };
356
357 static struct queue_sysfs_entry queue_hw_sector_size_entry = {
358         .attr = {.name = "hw_sector_size", .mode = S_IRUGO },
359         .show = queue_logical_block_size_show,
360 };
361
362 static struct queue_sysfs_entry queue_logical_block_size_entry = {
363         .attr = {.name = "logical_block_size", .mode = S_IRUGO },
364         .show = queue_logical_block_size_show,
365 };
366
367 static struct queue_sysfs_entry queue_physical_block_size_entry = {
368         .attr = {.name = "physical_block_size", .mode = S_IRUGO },
369         .show = queue_physical_block_size_show,
370 };
371
372 static struct queue_sysfs_entry queue_io_min_entry = {
373         .attr = {.name = "minimum_io_size", .mode = S_IRUGO },
374         .show = queue_io_min_show,
375 };
376
377 static struct queue_sysfs_entry queue_io_opt_entry = {
378         .attr = {.name = "optimal_io_size", .mode = S_IRUGO },
379         .show = queue_io_opt_show,
380 };
381
382 static struct queue_sysfs_entry queue_discard_granularity_entry = {
383         .attr = {.name = "discard_granularity", .mode = S_IRUGO },
384         .show = queue_discard_granularity_show,
385 };
386
387 static struct queue_sysfs_entry queue_discard_max_entry = {
388         .attr = {.name = "discard_max_bytes", .mode = S_IRUGO },
389         .show = queue_discard_max_show,
390 };
391
392 static struct queue_sysfs_entry queue_discard_zeroes_data_entry = {
393         .attr = {.name = "discard_zeroes_data", .mode = S_IRUGO },
394         .show = queue_discard_zeroes_data_show,
395 };
396
397 static struct queue_sysfs_entry queue_write_same_max_entry = {
398         .attr = {.name = "write_same_max_bytes", .mode = S_IRUGO },
399         .show = queue_write_same_max_show,
400 };
401
402 static struct queue_sysfs_entry queue_nonrot_entry = {
403         .attr = {.name = "rotational", .mode = S_IRUGO | S_IWUSR },
404         .show = queue_show_nonrot,
405         .store = queue_store_nonrot,
406 };
407
408 static struct queue_sysfs_entry queue_nomerges_entry = {
409         .attr = {.name = "nomerges", .mode = S_IRUGO | S_IWUSR },
410         .show = queue_nomerges_show,
411         .store = queue_nomerges_store,
412 };
413
414 static struct queue_sysfs_entry queue_rq_affinity_entry = {
415         .attr = {.name = "rq_affinity", .mode = S_IRUGO | S_IWUSR },
416         .show = queue_rq_affinity_show,
417         .store = queue_rq_affinity_store,
418 };
419
420 static struct queue_sysfs_entry queue_iostats_entry = {
421         .attr = {.name = "iostats", .mode = S_IRUGO | S_IWUSR },
422         .show = queue_show_iostats,
423         .store = queue_store_iostats,
424 };
425
426 static struct queue_sysfs_entry queue_random_entry = {
427         .attr = {.name = "add_random", .mode = S_IRUGO | S_IWUSR },
428         .show = queue_show_random,
429         .store = queue_store_random,
430 };
431
432 static struct attribute *default_attrs[] = {
433         &queue_requests_entry.attr,
434         &queue_ra_entry.attr,
435         &queue_max_hw_sectors_entry.attr,
436         &queue_max_sectors_entry.attr,
437         &queue_max_segments_entry.attr,
438         &queue_max_integrity_segments_entry.attr,
439         &queue_max_segment_size_entry.attr,
440         &queue_iosched_entry.attr,
441         &queue_hw_sector_size_entry.attr,
442         &queue_logical_block_size_entry.attr,
443         &queue_physical_block_size_entry.attr,
444         &queue_io_min_entry.attr,
445         &queue_io_opt_entry.attr,
446         &queue_discard_granularity_entry.attr,
447         &queue_discard_max_entry.attr,
448         &queue_discard_zeroes_data_entry.attr,
449         &queue_write_same_max_entry.attr,
450         &queue_nonrot_entry.attr,
451         &queue_nomerges_entry.attr,
452         &queue_rq_affinity_entry.attr,
453         &queue_iostats_entry.attr,
454         &queue_random_entry.attr,
455         NULL,
456 };
457
458 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
459
460 static ssize_t
461 queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
462 {
463         struct queue_sysfs_entry *entry = to_queue(attr);
464         struct request_queue *q =
465                 container_of(kobj, struct request_queue, kobj);
466         ssize_t res;
467
468         if (!entry->show)
469                 return -EIO;
470         mutex_lock(&q->sysfs_lock);
471         if (blk_queue_dying(q)) {
472                 mutex_unlock(&q->sysfs_lock);
473                 return -ENOENT;
474         }
475         res = entry->show(q, page);
476         mutex_unlock(&q->sysfs_lock);
477         return res;
478 }
479
480 static ssize_t
481 queue_attr_store(struct kobject *kobj, struct attribute *attr,
482                     const char *page, size_t length)
483 {
484         struct queue_sysfs_entry *entry = to_queue(attr);
485         struct request_queue *q;
486         ssize_t res;
487
488         if (!entry->store)
489                 return -EIO;
490
491         q = container_of(kobj, struct request_queue, kobj);
492         mutex_lock(&q->sysfs_lock);
493         if (blk_queue_dying(q)) {
494                 mutex_unlock(&q->sysfs_lock);
495                 return -ENOENT;
496         }
497         res = entry->store(q, page, length);
498         mutex_unlock(&q->sysfs_lock);
499         return res;
500 }
501
502 static void blk_free_queue_rcu(struct rcu_head *rcu_head)
503 {
504         struct request_queue *q = container_of(rcu_head, struct request_queue,
505                                                rcu_head);
506         kmem_cache_free(blk_requestq_cachep, q);
507 }
508
509 /**
510  * blk_release_queue: - release a &struct request_queue when it is no longer needed
511  * @kobj:    the kobj belonging to the request queue to be released
512  *
513  * Description:
514  *     blk_release_queue is the pair to blk_init_queue() or
515  *     blk_queue_make_request().  It should be called when a request queue is
516  *     being released; typically when a block device is being de-registered.
517  *     Currently, its primary task it to free all the &struct request
518  *     structures that were allocated to the queue and the queue itself.
519  *
520  * Caveat:
521  *     Hopefully the low level driver will have finished any
522  *     outstanding requests first...
523  **/
524 static void blk_release_queue(struct kobject *kobj)
525 {
526         struct request_queue *q =
527                 container_of(kobj, struct request_queue, kobj);
528
529         blk_sync_queue(q);
530
531         blkcg_exit_queue(q);
532
533         if (q->elevator) {
534                 spin_lock_irq(q->queue_lock);
535                 ioc_clear_queue(q);
536                 spin_unlock_irq(q->queue_lock);
537                 elevator_exit(q->elevator);
538         }
539
540         blk_exit_rl(&q->root_rl);
541
542         if (q->queue_tags)
543                 __blk_queue_free_tags(q);
544
545         blk_trace_shutdown(q);
546
547         bdi_destroy(&q->backing_dev_info);
548
549         ida_simple_remove(&blk_queue_ida, q->id);
550         call_rcu(&q->rcu_head, blk_free_queue_rcu);
551 }
552
553 static const struct sysfs_ops queue_sysfs_ops = {
554         .show   = queue_attr_show,
555         .store  = queue_attr_store,
556 };
557
558 struct kobj_type blk_queue_ktype = {
559         .sysfs_ops      = &queue_sysfs_ops,
560         .default_attrs  = default_attrs,
561         .release        = blk_release_queue,
562 };
563
564 int blk_register_queue(struct gendisk *disk)
565 {
566         int ret;
567         struct device *dev = disk_to_dev(disk);
568         struct request_queue *q = disk->queue;
569
570         if (WARN_ON(!q))
571                 return -ENXIO;
572
573         /*
574          * Initialization must be complete by now.  Finish the initial
575          * bypass from queue allocation.
576          */
577         blk_queue_bypass_end(q);
578
579         ret = blk_trace_init_sysfs(dev);
580         if (ret)
581                 return ret;
582
583         ret = kobject_add(&q->kobj, kobject_get(&dev->kobj), "%s", "queue");
584         if (ret < 0) {
585                 blk_trace_remove_sysfs(dev);
586                 return ret;
587         }
588
589         kobject_uevent(&q->kobj, KOBJ_ADD);
590
591         if (!q->request_fn)
592                 return 0;
593
594         ret = elv_register_queue(q);
595         if (ret) {
596                 kobject_uevent(&q->kobj, KOBJ_REMOVE);
597                 kobject_del(&q->kobj);
598                 blk_trace_remove_sysfs(dev);
599                 kobject_put(&dev->kobj);
600                 return ret;
601         }
602
603         return 0;
604 }
605
606 void blk_unregister_queue(struct gendisk *disk)
607 {
608         struct request_queue *q = disk->queue;
609
610         if (WARN_ON(!q))
611                 return;
612
613         if (q->request_fn)
614                 elv_unregister_queue(q);
615
616         kobject_uevent(&q->kobj, KOBJ_REMOVE);
617         kobject_del(&q->kobj);
618         blk_trace_remove_sysfs(disk_to_dev(disk));
619         kobject_put(&disk_to_dev(disk)->kobj);
620 }