On Jun 8, 2025, at 08:28, Kent Overstreet <[email protected]> wrote: > > Include a comment this time - we can't call iter_init() in a transaction > restart, and write bufer flush doesn't guarantee that we aren't in a > restart when it returns.
Why would it related to iter_init(), bch2_trans_begin(trans) is at the top of for_each_btree_key_max > > Signed-off-by: Kent Overstreet <[email protected]> > --- > fs/bcachefs/movinggc.c | 6 ++++++ > 1 file changed, 6 insertions(+) > > diff --git a/fs/bcachefs/movinggc.c b/fs/bcachefs/movinggc.c > index 6d7b1d5f7697..697d58b0bf91 100644 > --- a/fs/bcachefs/movinggc.c > +++ b/fs/bcachefs/movinggc.c > @@ -154,6 +154,12 @@ static int bch2_copygc_get_buckets(struct moving_context > *ctxt, > if (bch2_fs_fatal_err_on(ret, c, "%s: from > bch2_btree_write_buffer_tryflush()", bch2_err_str(ret))) > return ret; > > + /* > + * we might be in a transaction restart from write buffer flush, start a > + * new transaction before iter_init -> path_get > + */ > + bch2_trans_begin(trans); > + > ret = for_each_btree_key_max(trans, iter, BTREE_ID_lru, > lru_pos(BCH_LRU_BUCKET_FRAGMENTATION, 0, 0), > lru_pos(BCH_LRU_BUCKET_FRAGMENTATION, U64_MAX, LRU_TIME_MAX), > -- > 2.49.0 >
