Commit 9e903352 authored by Kent Overstreet's avatar Kent Overstreet
Browse files

bcachefs: Fix discard path journal flushing 



The discard path is supposed to issue journal flushes when there's too
many buckets empty buckets that need a journal commit before they can be
written to again, but at some point this code seems to have been lost.

Bring it back with a new optimization to make sure we don't issue too
many journal flushes: the journal now tracks the sequence number of the
most recent flush in progress, which the discard path uses when deciding
which buckets need a journal flush.

Signed-off-by: default avatarKent Overstreet <kent.overstreet@linux.dev>
parent 2ef995df

fs/bcachefs/alloc_background.c

+25 −22
Original line number Diff line number Diff line
@@ -1803,7 +1803,6 @@ struct discard_buckets_state { 
	u64		open;

	u64		need_journal_commit;

	u64		discarded;

	u64		need_journal_commit_this_dev;

};



static int bch2_discard_one_bucket(struct btree_trans *trans,
@@ -1827,11 +1826,11 @@ static int bch2_discard_one_bucket(struct btree_trans *trans, 
		goto out;

	}



	if (bch2_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,

			c->journal.flushed_seq_ondisk,

			pos.inode, pos.offset)) {

	u64 seq_ready = bch2_bucket_journal_seq_ready(&c->buckets_waiting_for_journal,

						      pos.inode, pos.offset);

	if (seq_ready > c->journal.flushed_seq_ondisk) {

		if (seq_ready > c->journal.flushing_seq)

			s->need_journal_commit++;

		s->need_journal_commit_this_dev++;

		goto out;

	}
@@ -1865,8 +1864,11 @@ static int bch2_discard_one_bucket(struct btree_trans *trans, 
		discard_locked = true;

	}



	if (!bkey_eq(*discard_pos_done, iter.pos) &&

	    ca->mi.discard && !c->opts.nochanges) {

	if (!bkey_eq(*discard_pos_done, iter.pos)) {

		s->discarded++;

		*discard_pos_done = iter.pos;



		if (ca->mi.discard && !c->opts.nochanges) {

			/*

			 * This works without any other locks because this is the only

			 * thread that removes items from the need_discard tree
@@ -1876,13 +1878,11 @@ static int bch2_discard_one_bucket(struct btree_trans *trans, 
					     k.k->p.offset * ca->mi.bucket_size,

					     ca->mi.bucket_size,

					     GFP_KERNEL);

		*discard_pos_done = iter.pos;

		s->discarded++;



			ret = bch2_trans_relock_notrace(trans);

			if (ret)

				goto out;

		}

	}



	SET_BCH_ALLOC_V4_NEED_DISCARD(&a->v, false);

	alloc_data_type_set(&a->v, a->v.data_type);
@@ -1929,6 +1929,9 @@ static void bch2_do_discards_work(struct work_struct *work) 
				   POS(ca->dev_idx, U64_MAX), 0, k,

			bch2_discard_one_bucket(trans, ca, &iter, &discard_pos_done, &s, false)));



	if (s.need_journal_commit > dev_buckets_available(ca, BCH_WATERMARK_normal))

		bch2_journal_flush_async(&c->journal, NULL);



	trace_discard_buckets(c, s.seen, s.open, s.need_journal_commit, s.discarded,

			      bch2_err_str(ret));
@@ -2024,7 +2027,7 @@ static void bch2_do_discards_fast_work(struct work_struct *work) 
			break;

	}



	trace_discard_buckets(c, s.seen, s.open, s.need_journal_commit, s.discarded, bch2_err_str(ret));

	trace_discard_buckets_fast(c, s.seen, s.open, s.need_journal_commit, s.discarded, bch2_err_str(ret));



	bch2_trans_put(trans);

	percpu_ref_put(&ca->io_ref);

fs/bcachefs/alloc_foreground.c

+7 −3
Original line number Diff line number Diff line
@@ -205,8 +205,12 @@ static inline bool may_alloc_bucket(struct bch_fs *c, 
		return false;

	}



	if (bch2_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,

			c->journal.flushed_seq_ondisk, bucket.inode, bucket.offset)) {

	u64 journal_seq_ready =

		bch2_bucket_journal_seq_ready(&c->buckets_waiting_for_journal,

					      bucket.inode, bucket.offset);

	if (journal_seq_ready > c->journal.flushed_seq_ondisk) {

		if (journal_seq_ready > c->journal.flushing_seq)

			s->need_journal_commit++;

		s->skipped_need_journal_commit++;

		return false;

	}
@@ -570,7 +574,7 @@ static struct open_bucket *bch2_bucket_alloc_trans(struct btree_trans *trans, 
		? bch2_bucket_alloc_freelist(trans, ca, watermark, &s, cl)

		: bch2_bucket_alloc_early(trans, ca, watermark, &s, cl);



	if (s.skipped_need_journal_commit * 2 > avail)

	if (s.need_journal_commit * 2 > avail)

		bch2_journal_flush_async(&c->journal, NULL);



	if (!ob && s.btree_bitmap != BTREE_BITMAP_ANY) {

fs/bcachefs/alloc_types.h

+1 −0
Original line number Diff line number Diff line
@@ -18,6 +18,7 @@ struct bucket_alloc_state { 
	u64	buckets_seen;

	u64	skipped_open;

	u64	skipped_need_journal_commit;

	u64	need_journal_commit;

	u64	skipped_nocow;

	u64	skipped_nouse;

	u64	skipped_mi_btree_bitmap;

fs/bcachefs/buckets_waiting_for_journal.c

+5 −7
Original line number Diff line number Diff line
@@ -22,23 +22,21 @@ static void bucket_table_init(struct buckets_waiting_for_journal_table *t, size_ 
	memset(t->d, 0, sizeof(t->d[0]) << t->bits);

}



bool bch2_bucket_needs_journal_commit(struct buckets_waiting_for_journal *b,

				      u64 flushed_seq,

u64 bch2_bucket_journal_seq_ready(struct buckets_waiting_for_journal *b,

				  unsigned dev, u64 bucket)

{

	struct buckets_waiting_for_journal_table *t;

	u64 dev_bucket = (u64) dev << 56 | bucket;

	bool ret = false;

	unsigned i;

	u64 ret = 0;



	mutex_lock(&b->lock);

	t = b->t;



	for (i = 0; i < ARRAY_SIZE(t->hash_seeds); i++) {

	for (unsigned i = 0; i < ARRAY_SIZE(t->hash_seeds); i++) {

		struct bucket_hashed *h = bucket_hash(t, i, dev_bucket);



		if (h->dev_bucket == dev_bucket) {

			ret = h->journal_seq > flushed_seq;

			ret = h->journal_seq;

			break;

		}

	}

fs/bcachefs/buckets_waiting_for_journal.h

+2 −2
Original line number Diff line number Diff line
@@ -4,8 +4,8 @@ 


#include "buckets_waiting_for_journal_types.h"



bool bch2_bucket_needs_journal_commit(struct buckets_waiting_for_journal *,

				      u64, unsigned, u64);

u64 bch2_bucket_journal_seq_ready(struct buckets_waiting_for_journal *,

				  unsigned, u64);

int bch2_set_bucket_needs_journal_commit(struct buckets_waiting_for_journal *,

					 u64, unsigned, u64, u64);