u64 search;
u64 target;
u64 nread = 0;
+ u64 nread_max;
struct extent_buffer *eb;
u32 nr;
u32 blocksize;
u32 nscan = 0;
- if (level != 1)
+ if (level != 1 && path->reada != READA_FORWARD_ALWAYS)
return;
if (!path->nodes[level])
node = path->nodes[level];
+ /*
+ * Since the time between visiting leaves is much shorter than the time
+ * between visiting nodes, limit read ahead of nodes to 1, to avoid too
+ * much IO at once (possibly random).
+ */
+ if (path->reada == READA_FORWARD_ALWAYS) {
+ if (level > 1)
+ nread_max = node->fs_info->nodesize;
+ else
+ nread_max = SZ_128K;
+ } else {
+ nread_max = SZ_64K;
+ }
+
search = btrfs_node_blockptr(node, slot);
blocksize = fs_info->nodesize;
eb = find_extent_buffer(fs_info, search);
if (nr == 0)
break;
nr--;
- } else if (path->reada == READA_FORWARD) {
+ } else if (path->reada == READA_FORWARD ||
+ path->reada == READA_FORWARD_ALWAYS) {
nr++;
if (nr >= nritems)
break;
break;
}
search = btrfs_node_blockptr(node, nr);
- if ((search <= target && target - search <= 65536) ||
+ if (path->reada == READA_FORWARD_ALWAYS ||
+ (search <= target && target - search <= 65536) ||
(search > target && search - target <= 65536)) {
btrfs_readahead_node_child(node, nr);
nread += blocksize;
}
nscan++;
- if ((nread > 65536 || nscan > 32))
+ if (nread > nread_max || nscan > 32)
break;
}
}
tmp = find_extent_buffer(fs_info, blocknr);
if (tmp) {
+ if (p->reada == READA_FORWARD_ALWAYS)
+ reada_for_search(fs_info, p, level, slot, key->objectid);
+
/* first we do an atomic uptodate check */
if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
/*
struct btrfs_key_ptr ptrs[];
} __attribute__ ((__packed__));
+/* Read ahead values for struct btrfs_path.reada */
+enum {
+ READA_NONE,
+ READA_BACK,
+ READA_FORWARD,
+ /*
+ * Similar to READA_FORWARD but unlike it:
+ *
+ * 1) It will trigger readahead even for leaves that are not close to
+ * each other on disk;
+ * 2) It also triggers readahead for nodes;
+ * 3) During a search, even when a node or leaf is already in memory, it
+ * will still trigger readahead for other nodes and leaves that follow
+ * it.
+ *
+ * This is meant to be used only when we know we are iterating over the
+ * entire tree or a very large part of it.
+ */
+ READA_FORWARD_ALWAYS,
+};
+
/*
* btrfs_paths remember the path taken from the root down to the leaf.
* level 0 is always the leaf, and nodes[1...BTRFS_MAX_LEVEL] will point
* The slots array records the index of the item or block pointer
* used while walking the tree.
*/
-enum { READA_NONE, READA_BACK, READA_FORWARD };
struct btrfs_path {
struct extent_buffer *nodes[BTRFS_MAX_LEVEL];
int slots[BTRFS_MAX_LEVEL];