libceph: a per-osdc crush scratch buffer
With the addition of erasure coding support in the future, scratch
variable-length array in crush_do_rule_ary() is going to grow to at
least 200 bytes on average, on top of another 128 bytes consumed by
rawosd/osd arrays in the call chain. Replace it with a buffer inside
struct osdmap and a mutex. This shouldn't result in any contention,
because all osd requests were already serialized by request_mutex at
that point; the only unlocked caller was ceph_ioctl_get_dataloc().
Signed-off-by: Ilya Dryomov <ilya.dryomov@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
diff --git a/net/ceph/osdmap.c b/net/ceph/osdmap.c
index aade4a5..9d1aaa2 100644
--- a/net/ceph/osdmap.c
+++ b/net/ceph/osdmap.c
@@ -698,7 +698,9 @@
map = kzalloc(sizeof(*map), GFP_NOFS);
if (map == NULL)
return ERR_PTR(-ENOMEM);
+
map->pg_temp = RB_ROOT;
+ mutex_init(&map->crush_scratch_mutex);
ceph_decode_16_safe(p, end, version, bad);
if (version > 6) {
@@ -1142,14 +1144,20 @@
}
EXPORT_SYMBOL(ceph_oloc_oid_to_pg);
-static int crush_do_rule_ary(const struct crush_map *map, int ruleno, int x,
- int *result, int result_max,
- const __u32 *weight, int weight_max)
+static int do_crush(struct ceph_osdmap *map, int ruleno, int x,
+ int *result, int result_max,
+ const __u32 *weight, int weight_max)
{
- int scratch[result_max * 3];
+ int r;
- return crush_do_rule(map, ruleno, x, result, result_max,
- weight, weight_max, scratch);
+ BUG_ON(result_max > CEPH_PG_MAX_SIZE);
+
+ mutex_lock(&map->crush_scratch_mutex);
+ r = crush_do_rule(map->crush, ruleno, x, result, result_max,
+ weight, weight_max, map->crush_scratch_ary);
+ mutex_unlock(&map->crush_scratch_mutex);
+
+ return r;
}
/*
@@ -1205,9 +1213,8 @@
pool->pgp_num_mask) +
(unsigned)pgid.pool;
}
- r = crush_do_rule_ary(osdmap->crush, ruleno, pps,
- osds, min_t(int, pool->size, *num),
- osdmap->osd_weight, osdmap->max_osd);
+ r = do_crush(osdmap, ruleno, pps, osds, min_t(int, pool->size, *num),
+ osdmap->osd_weight, osdmap->max_osd);
if (r < 0) {
pr_err("error %d from crush rule: pool %lld ruleset %d type %d"
" size %d\n", r, pgid.pool, pool->crush_ruleset,