bcachefs: Don't fail mounts due to devices that are marked as failed
authorKent Overstreet <kent.overstreet@gmail.com>
Sat, 3 Apr 2021 03:41:10 +0000 (23:41 -0400)
committerKent Overstreet <kent.overstreet@linux.dev>
Sun, 22 Oct 2023 21:08:54 +0000 (17:08 -0400)
If a given set of replicas is entirely on failed devices, don't fail the
mount: we will still fail the mount if we have some copies on non failed
devices.

Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
fs/bcachefs/replicas.c

index 8003973..15ff0d3 100644 (file)
@@ -975,11 +975,18 @@ bool bch2_have_enough_devs(struct bch_fs *c, struct bch_devs_mask devs,
 
        percpu_down_read(&c->mark_lock);
        for_each_cpu_replicas_entry(&c->replicas, e) {
-               unsigned i, nr_online = 0, dflags = 0;
+               unsigned i, nr_online = 0, nr_failed = 0, dflags = 0;
                bool metadata = e->data_type < BCH_DATA_user;
 
-               for (i = 0; i < e->nr_devs; i++)
+               for (i = 0; i < e->nr_devs; i++) {
+                       struct bch_dev *ca = bch_dev_bkey_exists(c, e->devs[i]);
+
                        nr_online += test_bit(e->devs[i], devs.d);
+                       nr_failed += ca->mi.state == BCH_MEMBER_STATE_FAILED;
+               }
+
+               if (nr_failed == e->nr_devs)
+                       continue;
 
                if (nr_online < e->nr_required)
                        dflags |= metadata