bcache: Fix moving_gc deadlocking with a foreground write
Deadlock happened because a foreground write slept, waiting for a bucket
to be allocated. Normally the gc would mark buckets available for invalidation.
But the moving_gc was stuck waiting for outstanding writes to complete.
These writes used the bcache_wq, the same queue foreground writes used.
This fix gives moving_gc its own work queue, so it was still finish moving
even if foreground writes are stuck waiting for allocation. It also makes
work queue a parameter to the data_insert path, so moving_gc can use its
workqueue for writes.
Signed-off-by: Nicholas Swenson <nks@daterainc.com>
Signed-off-by: Kent Overstreet <kmo@daterainc.com>
diff --git a/drivers/md/bcache/movinggc.c b/drivers/md/bcache/movinggc.c
index 9eb60d1..8c72051 100644
--- a/drivers/md/bcache/movinggc.c
+++ b/drivers/md/bcache/movinggc.c
@@ -115,7 +115,7 @@
closure_call(&op->cl, bch_data_insert, NULL, cl);
}
- continue_at(cl, write_moving_finish, system_wq);
+ continue_at(cl, write_moving_finish, op->wq);
}
static void read_moving_submit(struct closure *cl)
@@ -125,7 +125,7 @@
bch_submit_bbio(bio, io->op.c, &io->w->key, 0);
- continue_at(cl, write_moving, system_wq);
+ continue_at(cl, write_moving, io->op.wq);
}
static void read_moving(struct cache_set *c)
@@ -160,6 +160,7 @@
io->w = w;
io->op.inode = KEY_INODE(&w->key);
io->op.c = c;
+ io->op.wq = c->moving_gc_wq;
moving_init(io);
bio = &io->bio.bio;