net/rds: Fix MR reference counting problem

In rds_free_mr(), it calls rds_destroy_mr(mr) directly.  But this
defeats the purpose of reference counting and makes MR free handling
impossible.  It means that holding a reference does not guarantee that
it is safe to access some fields.  For example, In
rds_cmsg_rdma_dest(), it increases the ref count, unlocks and then
calls mr->r_trans->sync_mr().  But if rds_free_mr() (and
rds_destroy_mr()) is called in between (there is no lock preventing
this to happen), r_trans_private is set to NULL, causing a panic.
Similar issue is in rds_rdma_unuse().

Reported-by: zerons <sironhide0null@gmail.com>
Signed-off-by: Ka-Cheong Poon <ka-cheong.poon@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This commit is contained in:
Ka-Cheong Poon
2020-04-08 03:21:02 -07:00
committed by David S. Miller
parent e228a5d05e
commit 2fabef4f65
2 changed files with 12 additions and 21 deletions

View File

@@ -299,19 +299,11 @@ struct rds_mr {
unsigned int r_invalidate:1;
unsigned int r_write:1;
/* This is for RDS_MR_DEAD.
* It would be nice & consistent to make this part of the above
* bit field here, but we need to use test_and_set_bit.
*/
unsigned long r_state;
struct rds_sock *r_sock; /* back pointer to the socket that owns us */
struct rds_transport *r_trans;
void *r_trans_private;
};
/* Flags for mr->r_state */
#define RDS_MR_DEAD 0
static inline rds_rdma_cookie_t rds_rdma_make_cookie(u32 r_key, u32 offset)
{
return r_key | (((u64) offset) << 32);