[rds-devel] [PATCH] net/rds: fix unaligned memory access

[David Miller]

From: David Laight <David.Laight at ACULAB.COM>
Date: Thu, 30 Apr 2015 15:23:39 +0000

> From: David Ahern
>> Sent: 30 April 2015 15:28
>> rdma_conn_param private data is copied using memcpy after headers such
>> as cma_hdr (see cma_resolve_ib_udp as example). so the start of the
>> private data is aligned to the end of the structure that come before. if
>> this structure end with u32 the meaning is that the start of the private
>> data will be 4 bytes aligned. structures that use u8/u16/u32/u64 are
>> naturally aligned but in case the structure start is not 8 bytes aligned,
>> all u64 members of this structure will not be aligned. to solve this issue
>> we must use special macros that allow unaligned access to those
>> unaligned members.
>> 
>> Addresses the following kernel log seen when attempting to use RDMA:
>> 
>> Kernel unaligned access at TPC[10507a88] rds_ib_cm_connect_complete+0x1bc/0x1e0 [rds_rdma]
> ...
>> diff --git a/net/rds/ib_cm.c b/net/rds/ib_cm.c
>> index 31b74f5..e0bbbac 100644
>> --- a/net/rds/ib_cm.c
>> +++ b/net/rds/ib_cm.c
>> @@ -183,8 +183,17 @@ void rds_ib_cm_connect_complete(struct rds_connection *conn, struct rdma_cm_even
>> 
>>  	/* If the peer gave us the last packet it saw, process this as if
>>  	 * we had received a regular ACK. */
>> -	if (dp && dp->dp_ack_seq)
>> -		rds_send_drop_acked(conn, be64_to_cpu(dp->dp_ack_seq), NULL);
> 
> Or just mark 'dp_ack_seq' with __attribute((aligned(4))).
> Then the compiler will generate two 32bit loads.
> (provided be64_to_cpu() doesn't do anything stupid.)

It's a __be64 which already requires 8 byte alignment.

That's not what is causing the problem here.  It's the opaque area that
this structure can end up being placed afterwards.