IB/iser: Avoid recv buffer exhaustion caused by unexpected PDUs
authorDavid Disseldorp <ddiss@sgi.com>
Sun, 21 Dec 2008 21:56:50 +0000 (13:56 -0800)
committerRoland Dreier <rolandd@cisco.com>
Sun, 21 Dec 2008 21:56:50 +0000 (13:56 -0800)
commitbba7ebba3b17f4fe8c5907a32e16d9bd3fcf5192
tree8f4773e3a504bf17a8e0dfc0eaa3f3efa55941c0
parent061e41fdb5047b1fb161e89664057835935ca1d2
IB/iser: Avoid recv buffer exhaustion caused by unexpected PDUs

iSCSI/iSER targets may send PDUs without a prior request from the
initiator.  RFC 5046 refers to these PDUs as "unexpected".  NOP-In PDUs
with itt=RESERVED and Asynchronous Message PDUs occupy this category.

The amount of active "unexpected" PDU's an iSER target may have at any
time is governed by the MaxOutstandingUnexpectedPDUs key, which is not
yet supported.

Currently when an iSER target sends an "unexpected" PDU, the
initiators recv buffer consumed by the PDU is not replaced.  If over
initial_post_recv_bufs_num "unexpected" PDUs are received then the
receive queue will run out of receive work requests entirely.

This patch ensures recv buffers consumed by "unexpected" PDUs are
replaced in the next iser_post_receive_control() call.

Signed-off-by: David Disseldorp <ddiss@sgi.com>
Signed-off-by: Ken Sandars <ksandars@sgi.com>
Acked-by: Or Gerlitz <ogerlitz@voltaire.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
drivers/infiniband/ulp/iser/iscsi_iser.h
drivers/infiniband/ulp/iser/iser_initiator.c
drivers/infiniband/ulp/iser/iser_verbs.c