When afs_update_cell() replaces the cell->vl_servers list, it uses RCU
protocol so that proc is protected, but doesn't take ->vl_servers_lock to
protect afs_start_vl_iteration() (which does actually take a shared lock).

Fix this by making afs_update_cell() take an exclusive lock when replacing

Fixes: 0a5143f2f89c ("afs: Implement VL server rotation")
Signed-off-by: David Howells <dhowe...@redhat.com>

 fs/afs/cell.c |    5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/fs/afs/cell.c b/fs/afs/cell.c
index 9de46116c749..9ca075e11239 100644
--- a/fs/afs/cell.c
+++ b/fs/afs/cell.c
@@ -404,12 +404,11 @@ static void afs_update_cell(struct afs_cell *cell)
                clear_bit(AFS_CELL_FL_DNS_FAIL, &cell->flags);
                clear_bit(AFS_CELL_FL_NOT_FOUND, &cell->flags);
-               /* Exclusion on changing vl_addrs is achieved by a
-                * non-reentrant work item.
-                */
+               write_lock(&cell->vl_servers_lock);
                old = rcu_dereference_protected(cell->vl_servers, true);
                rcu_assign_pointer(cell->vl_servers, vllist);
                cell->dns_expiry = expiry;
+               write_unlock(&cell->vl_servers_lock);
                if (old)
                        afs_put_vlserverlist(cell->net, old);

Reply via email to