Kernel/linux-stable
1
0
Fork 0
mirror of https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git synced 2025-01-01 10:45:49 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

bcachefs: Btree splits now only take the locks they need

Browse Source 
Previously, bch2_btree_update_start() would always take all intent
locks, all the way up to the root.

We've finally got data from users where this became a scalability issue
- so, this patch fixes bch2_btree_update_start() to only take the locks
we need.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This commit is contained in:
Kent Overstreet 2022-09-27 18:57:34 -04:00
parent 969576ecae
commit 1f0f731ffe
2 changed files with 26 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

42
fs/bcachefs/btree_update_interior.c
View File

@ -1070,24 +1070,24 @@ bch2_btree_update_start(struct btree_trans *trans, struct btree_path *path,
nr_nodes[!!update_level] += 1 + split;
update_level++;
if (!btree_path_node(path, update_level))
ret = bch2_btree_path_upgrade(trans, path, update_level + 1);
if (ret)
return ERR_PTR(ret);
if (!btree_path_node(path, update_level)) {
/* Allocating new root? */
nr_nodes[1] += split;
update_level = BTREE_MAX_DEPTH;
break;
}
if (bch2_btree_node_insert_fits(c, path->l[update_level].b,
BKEY_BTREE_PTR_U64s_MAX * (1 + split)))
break;
/*
* XXX: figure out how far we might need to split,
* instead of locking/reserving all the way to the root:
*/
split = update_level + 1 < BTREE_MAX_DEPTH;
split = true;
}
/* Might have to allocate a new root: */
if (update_level < BTREE_MAX_DEPTH)
nr_nodes[1] += 1;
ret = bch2_btree_path_upgrade(trans, path, U8_MAX);
if (ret)
return ERR_PTR(ret);
if (flags & BTREE_INSERT_GC_LOCK_HELD)
lockdep_assert_held(&c->gc_lock);
else if (!down_read_trylock(&c->gc_lock)) {
@ -1108,6 +1108,7 @@ bch2_btree_update_start(struct btree_trans *trans, struct btree_path *path,
as->mode = BTREE_INTERIOR_NO_UPDATE;
as->took_gc_lock = !(flags & BTREE_INSERT_GC_LOCK_HELD);
as->btree_id = path->btree_id;
as->update_level = update_level;
INIT_LIST_HEAD(&as->list);
INIT_LIST_HEAD(&as->unwritten_list);
INIT_LIST_HEAD(&as->write_blocked_list);
@ -1511,7 +1512,7 @@ static int btree_split(struct btree_update *as, struct btree_trans *trans,
int ret = 0;
BUG_ON(!parent && (b != btree_node_root(c, b)));
BUG_ON(!btree_node_intent_locked(path, btree_node_root(c, b)->c.level));
BUG_ON(parent && !btree_node_intent_locked(path, b->c.level + 1));
bch2_btree_interior_update_will_free_node(as, b);
@ -1698,7 +1699,7 @@ static int bch2_btree_insert_node(struct btree_update *as, struct btree_trans *t
int ret;
lockdep_assert_held(&c->gc_lock);
BUG_ON(!btree_node_intent_locked(path, btree_node_root(c, b)->c.level));
BUG_ON(!btree_node_intent_locked(path, b->c.level));
BUG_ON(!b->c.level);
BUG_ON(!as || as->b);
bch2_verify_keylist_sorted(keys);
@ -1738,6 +1739,13 @@ static int bch2_btree_insert_node(struct btree_update *as, struct btree_trans *t
btree_node_interior_verify(c, b);
return 0;
split:
/*
* We could attempt to avoid the transaction restart, by calling
* bch2_btree_path_upgrade() and allocating more nodes:
*/
if (b->c.level >= as->update_level)
return btree_trans_restart(trans, BCH_ERR_transaction_restart_split_race);
return btree_split(as, trans, path, b, keys, flags);
}
@ -1763,7 +1771,7 @@ int bch2_btree_split_leaf(struct btree_trans *trans,
bch2_btree_update_done(as, trans);
for (l = path->level + 1; btree_path_node(path, l) && !ret; l++)
for (l = path->level + 1; btree_node_intent_locked(path, l) && !ret; l++)
ret = bch2_foreground_maybe_merge(trans, path, l, flags);
return ret;

1
fs/bcachefs/btree_update_interior.h
View File

@ -52,6 +52,7 @@ struct btree_update {
unsigned took_gc_lock:1;
enum btree_id btree_id;
unsigned update_level;
struct disk_reservation disk_res;
struct journal_preres journal_preres;