History

block: move the bdi from the request_queue to the gendisk

The backing device information only makes sense for file system I/O, and thus belongs into the gendisk and not the lower level request_queue structure. Move it there.

Christoph Hellwig committed 5 years ago

Commit edb08721 parent 100816214 files · +58 −63Browse files

Unified · Split

14 files
  1. block/bfq-iosched.c +2 −2
  2. block/blk-cgroup.c +3 −4
  3. block/blk-core.c +3 −10
  4. block/blk-mq.c +1 −1
  5. block/blk-settings.c +9 −5
  6. block/blk-sysfs.c +12 −14
  7. block/blk-wbt.c +5 −5
  8. block/genhd.c +14 −9
  9. drivers/block/drbd/drbd_req.c +2 −3
  10. drivers/block/pktcdvd.c +3 −5
  11. fs/block_dev.c +2 −2
  12. fs/fat/fatent.c +1 −0
  13. include/linux/blkdev.h +0 −3
  14. include/linux/genhd.h +1 −0
block/bfq-iosched.c+2−2
@@ -5266,8 +5266,8 @@ bfq_set_next_ioprio_data(struct bfq_queue *bfqq, struct bfq_io_cq *bic) · 5265 unmodified lines
52665266 switch (ioprio_class) {
52675267 default:
52685268 pr_err("bdi %s: bfq: bad prio class %d\n",
5269− bdi_dev_name(bfqq->bfqd->queue->backing_dev_info),
5270− ioprio_class);
5269+ bdi_dev_name(queue_to_disk(bfqq->bfqd->queue)->bdi),
5270+ ioprio_class);
52715271 fallthrough;
52725272 case IOPRIO_CLASS_NONE:
52735273 /*
block/blk-cgroup.c+3−4
@@ -489,10 +489,9 @@ static int blkcg_reset_stats(struct cgroup_subsys_state *css, · 488 unmodified lines
489489
490490 const char *blkg_dev_name(struct blkcg_gq *blkg)
491491 {
492− /* some drivers (floppy) instantiate a queue w/o disk registered */
493− if (blkg->q->backing_dev_info->dev)
494− return bdi_dev_name(blkg->q->backing_dev_info);
495− return NULL;
492+ if (!queue_has_disk(blkg->q) || !queue_to_disk(blkg->q)->bdi->dev)
493+ return NULL;
494+ return bdi_dev_name(queue_to_disk(blkg->q)->bdi);
496495 }
497496
498497 /**
block/blk-core.c+3−10
@@ -14,7 +14,6 @@ · 13 unmodified lines
1414 */
1515 #include <linux/kernel.h>
1616 #include <linux/module.h>
17−#include <linux/backing-dev.h>
1817 #include <linux/bio.h>
1918 #include <linux/blkdev.h>
2019 #include <linux/blk-mq.h>
@@ -531,13 +530,9 @@ struct request_queue *blk_alloc_queue(int node_id) · 510 unmodified lines
531530 if (ret)
532531 goto fail_id;
533532
534− q->backing_dev_info = bdi_alloc(node_id);
535− if (!q->backing_dev_info)
536− goto fail_split;
537−
538533 q->stats = blk_alloc_queue_stats();
539534 if (!q->stats)
540− goto fail_stats;
535+ goto fail_split;
541536
542537 q->node = node_id;
543538
@@ -567,7 +562,7 @@ struct request_queue *blk_alloc_queue(int node_id) · 23 unmodified lines
567562 if (percpu_ref_init(&q->q_usage_counter,
568563 blk_queue_usage_counter_release,
569564 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
570− goto fail_bdi;
565+ goto fail_stats;
571566
572567 if (blkcg_init_queue(q))
573568 goto fail_ref;
@@ -580,10 +575,8 @@ struct request_queue *blk_alloc_queue(int node_id) · 6 unmodified lines
580575
581576 fail_ref:
582577 percpu_ref_exit(&q->q_usage_counter);
583−fail_bdi:
584− blk_free_queue_stats(q->stats);
585578 fail_stats:
586− bdi_put(q->backing_dev_info);
579+ blk_free_queue_stats(q->stats);
587580 fail_split:
588581 bioset_exit(&q->bio_split);
589582 fail_id:
block/blk-mq.c+1−1
@@ -525,7 +525,7 @@ void blk_mq_free_request(struct request *rq) · 524 unmodified lines
525525 __blk_mq_dec_active_requests(hctx);
526526
527527 if (unlikely(laptop_mode && !blk_rq_is_passthrough(rq)))
528− laptop_io_completion(q->backing_dev_info);
528+ laptop_io_completion(queue_to_disk(q)->bdi);
529529
530530 rq_qos_done(q, rq);
531531
block/blk-settings.c+9−5
@@ -8,6 +8,7 @@ · 7 unmodified lines
88 #include <linux/bio.h>
99 #include <linux/blkdev.h>
1010 #include <linux/pagemap.h>
11+#include <linux/backing-dev-defs.h>
1112 #include <linux/gcd.h>
1213 #include <linux/lcm.h>
1314 #include <linux/jiffies.h>
@@ -140,7 +141,9 @@ void blk_queue_max_hw_sectors(struct request_queue *q, unsigned int max_hw_secto · 126 unmodified lines
140141 limits->logical_block_size >> SECTOR_SHIFT);
141142 limits->max_sectors = max_sectors;
142143
143− q->backing_dev_info->io_pages = max_sectors >> (PAGE_SHIFT - 9);
144+ if (!queue_has_disk(q))
145+ return;
146+ queue_to_disk(q)->bdi->io_pages = max_sectors >> (PAGE_SHIFT - 9);
144147 }
145148 EXPORT_SYMBOL(blk_queue_max_hw_sectors);
146149
@@ -388,10 +391,9 @@ void disk_update_readahead(struct gendisk *disk) · 241 unmodified lines
388391 * For read-ahead of large files to be effective, we need to read ahead
389392 * at least twice the optimal I/O size.
390393 */
391− q->backing_dev_info->ra_pages =
394+ disk->bdi->ra_pages =
392395 max(queue_io_opt(q) * 2 / PAGE_SIZE, VM_READAHEAD_PAGES);
393− q->backing_dev_info->io_pages =
394− queue_max_sectors(q) >> (PAGE_SHIFT - 9);
396+ disk->bdi->io_pages = queue_max_sectors(q) >> (PAGE_SHIFT - 9);
395397 }
396398 EXPORT_SYMBOL_GPL(disk_update_readahead);
397399
@@ -473,7 +475,9 @@ EXPORT_SYMBOL(blk_limits_io_opt); · 75 unmodified lines
473475 void blk_queue_io_opt(struct request_queue *q, unsigned int opt)
474476 {
475477 blk_limits_io_opt(&q->limits, opt);
476− q->backing_dev_info->ra_pages =
478+ if (!queue_has_disk(q))
479+ return;
480+ queue_to_disk(q)->bdi->ra_pages =
477481 max(queue_io_opt(q) * 2 / PAGE_SIZE, VM_READAHEAD_PAGES);
478482 }
479483 EXPORT_SYMBOL(blk_queue_io_opt);
block/blk-sysfs.c+12−14
@@ -88,9 +88,11 @@ queue_requests_store(struct request_queue *q, const char *page, size_t count) · 87 unmodified lines
8888
8989 static ssize_t queue_ra_show(struct request_queue *q, char *page)
9090 {
91− unsigned long ra_kb = q->backing_dev_info->ra_pages <<
92− (PAGE_SHIFT - 10);
91+ unsigned long ra_kb;
9392
93+ if (!queue_has_disk(q))
94+ return -EINVAL;
95+ ra_kb = queue_to_disk(q)->bdi->ra_pages << (PAGE_SHIFT - 10);
9496 return queue_var_show(ra_kb, page);
9597 }
9698
@@ -98,13 +100,14 @@ static ssize_t · 1 unmodified line
98100 queue_ra_store(struct request_queue *q, const char *page, size_t count)
99101 {
100102 unsigned long ra_kb;
101− ssize_t ret = queue_var_store(&ra_kb, page, count);
103+ ssize_t ret;
102104
105+ if (!queue_has_disk(q))
106+ return -EINVAL;
107+ ret = queue_var_store(&ra_kb, page, count);
103108 if (ret < 0)
104109 return ret;
105−
106− q->backing_dev_info->ra_pages = ra_kb >> (PAGE_SHIFT - 10);
107−
110+ queue_to_disk(q)->bdi->ra_pages = ra_kb >> (PAGE_SHIFT - 10);
108111 return ret;
109112 }
110113
@@ -251,7 +254,9 @@ queue_max_sectors_store(struct request_queue *q, const char *page, size_t count) · 140 unmodified lines
251254
252255 spin_lock_irq(&q->queue_lock);
253256 q->limits.max_sectors = max_sectors_kb << 1;
254− q->backing_dev_info->io_pages = max_sectors_kb >> (PAGE_SHIFT - 10);
257+ if (queue_has_disk(q))
258+ queue_to_disk(q)->bdi->io_pages =
259+ max_sectors_kb >> (PAGE_SHIFT - 10);
255260 spin_unlock_irq(&q->queue_lock);
256261
257262 return ret;
@@ -766,13 +771,6 @@ static void blk_exit_queue(struct request_queue *q) · 508 unmodified lines
766771 * e.g. blkcg_print_blkgs() to crash.
767772 */
768773 blkcg_exit_queue(q);
769−
770− /*
771− * Since the cgroup code may dereference the @q->backing_dev_info
772− * pointer, only decrease its reference count after having removed the
773− * association with the block cgroup controller.
774− */
775− bdi_put(q->backing_dev_info);
776774 }
777775
778776 /**
block/blk-wbt.c+5−5
@@ -97,7 +97,7 @@ static void wb_timestamp(struct rq_wb *rwb, unsigned long *var) · 96 unmodified lines
9797 */
9898 static bool wb_recent_wait(struct rq_wb *rwb)
9999 {
100− struct bdi_writeback *wb = &rwb->rqos.q->backing_dev_info->wb;
100+ struct bdi_writeback *wb = &queue_to_disk(rwb->rqos.q)->bdi->wb;
101101
102102 return time_before(jiffies, wb->dirty_sleep + HZ);
103103 }
@@ -234,7 +234,7 @@ enum { · 130 unmodified lines
234234
235235 static int latency_exceeded(struct rq_wb *rwb, struct blk_rq_stat *stat)
236236 {
237− struct backing_dev_info *bdi = rwb->rqos.q->backing_dev_info;
237+ struct backing_dev_info *bdi = queue_to_disk(rwb->rqos.q)->bdi;
238238 struct rq_depth *rqd = &rwb->rq_depth;
239239 u64 thislat;
240240
@@ -287,7 +287,7 @@ static int latency_exceeded(struct rq_wb *rwb, struct blk_rq_stat *stat) · 46 unmodified lines
287287
288288 static void rwb_trace_step(struct rq_wb *rwb, const char *msg)
289289 {
290− struct backing_dev_info *bdi = rwb->rqos.q->backing_dev_info;
290+ struct backing_dev_info *bdi = queue_to_disk(rwb->rqos.q)->bdi;
291291 struct rq_depth *rqd = &rwb->rq_depth;
292292
293293 trace_wbt_step(bdi, msg, rqd->scale_step, rwb->cur_win_nsec,
@@ -359,8 +359,8 @@ static void wb_timer_fn(struct blk_stat_callback *cb) · 65 unmodified lines
359359
360360 status = latency_exceeded(rwb, cb->stat);
361361
362− trace_wbt_timer(rwb->rqos.q->backing_dev_info, status, rqd->scale_step,
363− inflight);
362+ trace_wbt_timer(queue_to_disk(rwb->rqos.q)->bdi, status,
363+ rqd->scale_step, inflight);
364364
365365 /*
366366 * If we exceeded the latency target, step down. If we did not,
block/genhd.c+14−9
@@ -466,10 +466,9 @@ static void register_disk(struct device *parent, struct gendisk *disk, · 465 unmodified lines
466466 dev_set_uevent_suppress(ddev, 0);
467467 disk_uevent(disk, KOBJ_ADD);
468468
469− if (disk->queue->backing_dev_info->dev) {
470− err = sysfs_create_link(&ddev->kobj,
471− &disk->queue->backing_dev_info->dev->kobj,
472− "bdi");
469+ if (disk->bdi->dev) {
470+ err = sysfs_create_link(&ddev->kobj, &disk->bdi->dev->kobj,
471+ "bdi");
473472 WARN_ON(err);
474473 }
475474 }
@@ -540,15 +539,14 @@ void device_add_disk(struct device *parent, struct gendisk *disk, · 64 unmodified lines
540539 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
541540 disk->flags |= GENHD_FL_NO_PART_SCAN;
542541 } else {
543− struct backing_dev_info *bdi = disk->queue->backing_dev_info;
544542 struct device *dev = disk_to_dev(disk);
545543
546544 /* Register BDI before referencing it from bdev */
547545 dev->devt = MKDEV(disk->major, disk->first_minor);
548− ret = bdi_register(bdi, "%u:%u",
546+ ret = bdi_register(disk->bdi, "%u:%u",
549547 disk->major, disk->first_minor);
550548 WARN_ON(ret);
551− bdi_set_owner(bdi, dev);
549+ bdi_set_owner(disk->bdi, dev);
552550 bdev_add(disk->part0, dev->devt);
553551 }
554552 register_disk(parent, disk, groups);
@@ -615,7 +613,7 @@ void del_gendisk(struct gendisk *disk) · 60 unmodified lines
615613 * Unregister bdi before releasing device numbers (as they can
616614 * get reused and we'd get clashes in sysfs).
617615 */
618− bdi_unregister(disk->queue->backing_dev_info);
616+ bdi_unregister(disk->bdi);
619617 }
620618
621619 blk_unregister_queue(disk);
@@ -1088,6 +1086,7 @@ static void disk_release(struct device *dev) · 466 unmodified lines
10881086
10891087 might_sleep();
10901088
1089+ bdi_put(disk->bdi);
10911090 if (MAJOR(dev->devt) == BLOCK_EXT_MAJOR)
10921091 blk_free_ext_minor(MINOR(dev->devt));
10931092 disk_release_events(disk);
@@ -1268,9 +1267,13 @@ struct gendisk *__alloc_disk_node(int minors, int node_id) · 174 unmodified lines
12681267 if (!disk)
12691268 return NULL;
12701269
1270+ disk->bdi = bdi_alloc(node_id);
1271+ if (!disk->bdi)
1272+ goto out_free_disk;
1273+
12711274 disk->part0 = bdev_alloc(disk, 0);
12721275 if (!disk->part0)
1273− goto out_free_disk;
1276+ goto out_free_bdi;
12741277
12751278 disk->node_id = node_id;
12761279 mutex_init(&disk->open_mutex);
@@ -1292,6 +1295,8 @@ struct gendisk *__alloc_disk_node(int minors, int node_id) · 15 unmodified lines
12921295 out_destroy_part_tbl:
12931296 xa_destroy(&disk->part_tbl);
12941297 iput(disk->part0->bd_inode);
1298+out_free_bdi:
1299+ bdi_put(disk->bdi);
12951300 out_free_disk:
12961301 kfree(disk);
12971302 return NULL;
drivers/block/drbd/drbd_req.c+2−3
@@ -905,13 +905,12 @@ static bool drbd_may_do_local_read(struct drbd_device *device, sector_t sector, · 904 unmodified lines
905905 static bool remote_due_to_read_balancing(struct drbd_device *device, sector_t sector,
906906 enum drbd_read_balancing rbm)
907907 {
908− struct backing_dev_info *bdi;
909908 int stripe_shift;
910909
911910 switch (rbm) {
912911 case RB_CONGESTED_REMOTE:
913− bdi = device->ldev->backing_bdev->bd_disk->queue->backing_dev_info;
914− return bdi_read_congested(bdi);
912+ return bdi_read_congested(
913+ device->ldev->backing_bdev->bd_disk->bdi);
915914 case RB_LEAST_PENDING:
916915 return atomic_read(&device->local_cnt) >
917916 atomic_read(&device->ap_pending_cnt) + atomic_read(&device->rs_pending_cnt);
drivers/block/pktcdvd.c+3−5
@@ -1183,10 +1183,8 @@ try_next_bio: · 1182 unmodified lines
11831183 wakeup = (pd->write_congestion_on > 0
11841184 && pd->bio_queue_size <= pd->write_congestion_off);
11851185 spin_unlock(&pd->lock);
1186− if (wakeup) {
1187− clear_bdi_congested(pd->disk->queue->backing_dev_info,
1188− BLK_RW_ASYNC);
1189− }
1186+ if (wakeup)
1187+ clear_bdi_congested(pd->disk->bdi, BLK_RW_ASYNC);
11901188
11911189 pkt->sleep_time = max(PACKET_WAIT_TIME, 1);
11921190 pkt_set_state(pkt, PACKET_WAITING_STATE);
@@ -2366,7 +2364,7 @@ static void pkt_make_request_write(struct request_queue *q, struct bio *bio) · 1173 unmodified lines
23662364 spin_lock(&pd->lock);
23672365 if (pd->write_congestion_on > 0
23682366 && pd->bio_queue_size >= pd->write_congestion_on) {
2369− set_bdi_congested(q->backing_dev_info, BLK_RW_ASYNC);
2367+ set_bdi_congested(bio->bi_bdev->bd_disk->bdi, BLK_RW_ASYNC);
23702368 do {
23712369 spin_unlock(&pd->lock);
23722370 congestion_wait(BLK_RW_ASYNC, HZ);
fs/block_dev.c+2−2
@@ -1087,7 +1087,7 @@ static int blkdev_get_whole(struct block_device *bdev, fmode_t mode) · 1086 unmodified lines
10871087 if (!bdev->bd_openers) {
10881088 set_init_blocksize(bdev);
10891089 if (bdev->bd_bdi == &noop_backing_dev_info)
1090− bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1090+ bdev->bd_bdi = bdi_get(disk->bdi);
10911091 }
10921092 if (test_bit(GD_NEED_PART_SCAN, &disk->state))
10931093 bdev_disk_changed(disk, false);
@@ -1122,7 +1122,7 @@ static int blkdev_get_part(struct block_device *part, fmode_t mode) · 28 unmodified lines
11221122 disk->open_partitions++;
11231123 set_init_blocksize(part);
11241124 if (part->bd_bdi == &noop_backing_dev_info)
1125− part->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1125+ part->bd_bdi = bdi_get(disk->bdi);
11261126 done:
11271127 part->bd_openers++;
11281128 return 0;
fs/fat/fatent.c+1−0
@@ -5,6 +5,7 @@ · 4 unmodified lines
55
66 #include <linux/blkdev.h>
77 #include <linux/sched/signal.h>
8+#include <linux/backing-dev-defs.h>
89 #include "fat.h"
910
1011 struct fatent_operations {
include/linux/blkdev.h+0−3
@@ -11,7 +11,6 @@ · 10 unmodified lines
1111 #include <linux/minmax.h>
1212 #include <linux/timer.h>
1313 #include <linux/workqueue.h>
14−#include <linux/backing-dev-defs.h>
1514 #include <linux/wait.h>
1615 #include <linux/mempool.h>
1716 #include <linux/pfn.h>
@@ -398,8 +397,6 @@ struct request_queue { · 380 unmodified lines
398397 struct blk_mq_hw_ctx **queue_hw_ctx;
399398 unsigned int nr_hw_queues;
400399
401− struct backing_dev_info *backing_dev_info;
402−
403400 /*
404401 * The queue owner gets to use this for whatever they like.
405402 * ll_rw_blk doesn't touch it.
include/linux/genhd.h+1−0
@@ -158,6 +158,7 @@ struct gendisk { · 157 unmodified lines
158158 struct mutex open_mutex; /* open/close mutex */
159159 unsigned open_partitions; /* number of open partitions */
160160
161+ struct backing_dev_info *bdi;
161162 struct kobject *slave_dir;
162163 #ifdef CONFIG_BLOCK_HOLDER_DEPRECATED
163164 struct list_head slave_bdevs;