path: root/fs/bcachefs/journal_reclaim.c

// SPDX-License-Identifier: GPL-2.0

#include "bcachefs.h"
#include "btree_key_cache.h"
#include "btree_update.h"
#include "btree_write_buffer.h"
#include "buckets.h"
#include "errcode.h"
#include "error.h"
#include "journal.h"
#include "journal_io.h"
#include "journal_reclaim.h"
#include "replicas.h"
#include "sb-members.h"
#include "trace.h"

#include <linux/kthread.h>
#include <linux/sched/mm.h>

/* Free space calculations: */

static unsigned journal_space_from(struct journal_device *ja,
				   enum journal_space_from from)
{
	switch (from) {
	case journal_space_discarded:
		return ja->discard_idx;
	case journal_space_clean_ondisk:
		return ja->dirty_idx_ondisk;
	case journal_space_clean:
		return ja->dirty_idx;
	default:
		BUG();
	}
}

unsigned bch2_journal_dev_buckets_available(struct journal *j,
					    struct journal_device *ja,
					    enum journal_space_from from)
{
	if (!ja->nr)
		return 0;

	unsigned available = (journal_space_from(ja, from) -
			      ja->cur_idx - 1 + ja->nr) % ja->nr;

	/*
	 * Don't use the last bucket unless writing the new last_seq
	 * will make another bucket available:
	 */
	if (available && ja->dirty_idx_ondisk == ja->dirty_idx)
		--available;

	return available;
}

void bch2_journal_set_watermark(struct journal *j)
{
	struct bch_fs *c = container_of(j, struct bch_fs, journal);
	bool low_on_space = j->space[journal_space_clean].total * 4 <=
		j->space[journal_space_total].total;
	bool low_on_pin = fifo_free(&j->pin) < j->pin.size / 4;
	bool low_on_wb = bch2_btree_write_buffer_must_wait(c);
	unsigned watermark = low_on_space || low_on_pin || low_on_wb
		? BCH_WATERMARK_reclaim
		: BCH_WATERMARK_stripe;

	if (track_event_change(&c->times[BCH_TIME_blocked_journal_low_on_space], low_on_space) ||
	    track_event_change(&c->times[BCH_TIME_blocked_journal_low_on_pin], low_on_pin) ||
	    track_event_change(&c->times[BCH_TIME_blocked_write_buffer_full], low_on_wb))
		trace_and_count(c, journal_full, c);

	mod_bit(JOURNAL_space_low, &j->flags, low_on_space || low_on_pin);

	swap(watermark, j->watermark);
	if (watermark > j->watermark)
		journal_wake(j);
}

static struct journal_space
journal_dev_space_available(struct journal *j, struct bch_dev *ca,
			    enum journal_space_from from)
{
	struct journal_device *ja = &ca->journal;
	unsigned sectors, buckets, unwritten;
	u64 seq;

	if (from == journal_space_total)
		return (struct journal_space) {
			.next_entry	= ca->mi.bucket_size,
			.total		= ca->mi.bucket_size * ja->nr,
		};

	buckets = bch2_journal_dev_buckets_available(j, ja, from);
	sectors = ja->sectors_free;

	/*
	 * We that we don't allocate the space for a journal entry
	 * until we write it out - thus, account for it here:
	 */
	for (seq = journal_last_unwritten_seq(j);
	     seq <= journal_cur_seq(j);
	     seq++) {
		unwritten = j->buf[seq & JOURNAL_BUF_MASK].sectors;

		if (!unwritten)
			continue;

		/* entry won't fit on this device, skip: */
		if (unwritten > ca->mi.bucket_size)
			continue;

		if (unwritten >= sectors) {
			if (!buckets) {
				sectors = 0;
				break;
			}

			buckets--;
			sectors = ca->mi.bucket_size;
		}

		sectors -= unwritten;
	}

	if (sectors < ca->mi.bucket_size && buckets) {
		buckets--;
		sectors = ca->mi.bucket_size;
	}

	return (struct journal_space) {
		.next_entry	= sectors,
		.total		= sectors + buckets * ca->mi.bucket_size,
	};
}

static struct journal_space __journal_space_available(struct journal *j, unsigned nr_devs_want,
			    enum journal_space_from from)
{
	struct bch_fs *c = container_of(j, struct bch_fs, journal);
	unsigned pos, nr_devs = 0;
	struct journal_space space, dev_space[BCH_SB_MEMBERS_MAX];
	unsigned min_bucket_size = U32_MAX;

	BUG_ON(nr_devs_want > ARRAY_SIZE(dev_space));

	rcu_read_lock();
	for_each_member_device_rcu(c, ca, &c->rw_devs[BCH_DATA_journal]) {
		if (!ca->journal.nr ||
		    !ca->mi.durability)
			continue;

		min_bucket_size = min(min_bucket_size, ca->mi.bucket_size);

		space = journal_dev_space_available(j, ca, from);
		if (!space.next_entry)
			continue;

		for (pos = 0; pos < nr_devs; pos++)
			if (space.total > dev_space[pos].total)
				break;

		array_insert_item(dev_space, nr_devs, pos, space);
	}
	rcu_read_unlock();

	if (nr_devs < nr_devs_want)
		return (struct journal_space) { 0, 0 };

	/*