mirrors/scoutfs
1
0
Fork 0
mirror of https://github.com/versity/scoutfs.git synced 2026-01-09 13:23:14 +00:00
Code Issues Packages Projects Releases Wiki Activity

Avoid deadlock from block reclaim in rht resize

Browse Source 
The RCU hash table uses deferred work to resize the hash table.  There's
a time during resize when hash table iteration will return EAGAIN until
resize makes more progress.  During this time resize can perform
GFP_KERNEL allocations.

Our shrinker tries to iterate over its RCU hash table to find blocks to
reclaim.  It tries to restart iteration if it gets EAGAIN on the
assumption that it will be usable again soon.

Combine the two and our shrinker can get stuck retrying iteration
indefinitely because it's shrinking on behalf of the hash table resizing
that is trying to allocate the next table before making iteration work
again.  We have to stop shrinking in this case so that the resizing
caller can proceed.

Signed-off-by: Zach Brown <zab@versity.com>
This commit is contained in:
Zach Brown
2023-06-15 14:45:26 -07:00
parent 5a1e5639c2
commit acafb869e7
2 changed files with 16 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
kmod/src/block.c
View File

@@ -1096,6 +1096,7 @@ static int block_shrink(struct shrinker *shrink, struct shrink_control *sc)
struct super_block *sb = binf->sb;
struct rhashtable_iter iter;
struct block_private *bp;
bool stop = false;
unsigned long nr;
u64 recently;
@@ -1107,7 +1108,6 @@ static int block_shrink(struct shrinker *shrink, struct shrink_control *sc)
nr = DIV_ROUND_UP(nr, SCOUTFS_BLOCK_LG_PAGES_PER);
restart:
recently = accessed_recently(binf);
rhashtable_walk_enter(&binf->ht, &iter);
rhashtable_walk_start(&iter);
@@ -1129,12 +1129,15 @@ restart:
if (bp == NULL)
break;
if (bp == ERR_PTR(-EAGAIN)) {
/* hard exit to wait for rcu rebalance to finish */
rhashtable_walk_stop(&iter);
rhashtable_walk_exit(&iter);
scoutfs_inc_counter(sb, block_cache_shrink_restart);
synchronize_rcu();
goto restart;
/*
* We can be called from reclaim in the allocation
* to resize the hash table itself. We have to
* return so that the caller can proceed and
* enable hash table iteration again.
*/
scoutfs_inc_counter(sb, block_cache_shrink_stop);
stop = true;
break;
}
scoutfs_inc_counter(sb, block_cache_shrink_next);
@@ -1157,8 +1160,11 @@ restart:
rhashtable_walk_stop(&iter);
rhashtable_walk_exit(&iter);
out:
return min_t(u64, (u64)atomic_read(&binf->total_inserted) * SCOUTFS_BLOCK_LG_PAGES_PER,
INT_MAX);
if (stop)
return -1;
else
return min_t(u64, INT_MAX,
(u64)atomic_read(&binf->total_inserted) * SCOUTFS_BLOCK_LG_PAGES_PER);
}
struct sm_block_completion {

2
kmod/src/counters.h
View File

@@ -34,7 +34,7 @@
EXPAND_COUNTER(block_cache_shrink_next) \
EXPAND_COUNTER(block_cache_shrink_recent) \
EXPAND_COUNTER(block_cache_shrink_remove) \
EXPAND_COUNTER(block_cache_shrink_restart) \
EXPAND_COUNTER(block_cache_shrink_stop) \
EXPAND_COUNTER(btree_compact_values) \
EXPAND_COUNTER(btree_compact_values_enomem) \
EXPAND_COUNTER(btree_delete) \