btrfs: send: index backref cache by node number instead of by sector number

We now have a nodesize_bits member in fs_info so we can index an extent
buffer in the backref cache by node number instead of by sector number.
While this allows for a denser index space with the possibility of using
less maple tree nodes, in practice it's unlikely to hit such benefits
since we currently limit the maximum number of keys in the cache to 128,
so unless all extent buffers are contiguous we are unlikely to see a
memory usage reduction in the backing maple tree due to fewer nodes.
Nevertheless it doesn't cost anything to index by node number and it's
more logical.

Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
This commit is contained in:
Filipe Manana 2025-09-08 12:19:29 +01:00 committed by David Sterba
parent 2753e49176
commit 0dc93e4652

View File

@ -1388,7 +1388,7 @@ static bool lookup_backref_cache(u64 leaf_bytenr, void *ctx,
struct backref_ctx *bctx = ctx;
struct send_ctx *sctx = bctx->sctx;
struct btrfs_fs_info *fs_info = sctx->send_root->fs_info;
const u64 key = leaf_bytenr >> fs_info->sectorsize_bits;
const u64 key = leaf_bytenr >> fs_info->nodesize_bits;
struct btrfs_lru_cache_entry *raw_entry;
struct backref_cache_entry *entry;
@ -1443,7 +1443,7 @@ static void store_backref_cache(u64 leaf_bytenr, const struct ulist *root_ids,
if (!new_entry)
return;
new_entry->entry.key = leaf_bytenr >> fs_info->sectorsize_bits;
new_entry->entry.key = leaf_bytenr >> fs_info->nodesize_bits;
new_entry->entry.gen = 0;
new_entry->num_roots = 0;
ULIST_ITER_INIT(&uiter);