From: Nicholas Bellinger <nab@xxxxxxxxxxxxxxx> Hi all, Here are two patches to address issues encountered over the last month while stress testing with ESX hosts. This first is hopefully the last regression around iscsi-target changes over the last releases to create kthreads on-demand during login negotiation. It addresses a case that would end up leaving left-over iscsi_target_rx_thread() in uninterruptible sleep, if the failure occured in iscsi_target_do_tx_login_io() attempting to send the last login response PDU. The second is a COMPARE_AND_WRITE use-after-free bug, that is difficult to hit for normal backends, but with just the right scheduling delays will result in OOPsen. The problem centers around the use of SCF_COMPARE_AND_WRITE_POST flag checking in target_complete_ok_work() to determine the first or second phase processing of COMPARE_AND_WRITE. That is, there is nothing that prevents the CAW callbacks in target_complete_ok_work() from completing in reverse order, so the dependency on checking cmd->se_cmd_flags is incorrect. To address this, allow cmd->transport_complete_callback() to propigate up 'post_ret' to target_complete_ok_work(), and avoid se_cmd dereference after ->transport_complete_callback(). Both patches are straight-forward fixes, and have been verified extensively on Linux + ESX hosts the last weeks. --nab Nicholas Bellinger (2): iscsi-target: Fix rx_login_comp hang after login failure target: Fix race for SCF_COMPARE_AND_WRITE_POST checking drivers/target/iscsi/iscsi_target.c | 13 ++++++++++++- drivers/target/iscsi/iscsi_target_nego.c | 1 + drivers/target/target_core_sbc.c | 13 +++++++++---- drivers/target/target_core_transport.c | 14 ++++++++------ include/target/target_core_base.h | 2 +- 5 files changed, 31 insertions(+), 12 deletions(-) -- 1.9.1 -- To unsubscribe from this list: send the line "unsubscribe linux-scsi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
