[PATCH] Fix for hang of Ordered task in TCM
From: Michael Cyr <hidden>
Date: 2016-05-13 22:16:12
Also in:
linux-scsi, lkml
Subsystem:
scsi target subsystem, the rest · Maintainers:
"Martin K. Petersen", Linus Torvalds
If a command with a Simple task attribute is failed due to a Unit Attention, then a subsequent command with an Ordered task attribute will hang forever. The reason for this is that the Unit Attention status is checked for in target_setup_cmd_from_cdb, before the call to target_execute_cmd, which calls target_handle_task_attr, which in turn increments dev->simple_cmds. However, transport_generic_request_failure still calls transport_complete_task_attr, which will decrement dev->simple_cmds. In this case, simple_cmds is now -1. So when a command with the Ordered task attribute is sent, target_handle_task_attr sees that dev->simple_cmds is not 0, so it decides it can't execute the command until all the (nonexistent) Simple commands have completed. The solution I've implemented is to move target_scsi3_ua_check, as well as target_alua_state_check and target_check_reservation, into target_execute_cmd, after the call to target_handle_task_attr. I believe this is actually the correct way this should be handled. According to SAM-4 r14, under section 5.14: "h) if a command other than INQUIRY, REPORT LUNS, REQUEST SENSE, or NOTIFY DATA TRANSFER DEVICE enters the enabled command state while a unit attention condition exists for the SCSI initiator port associated with the I_T nexus on which the command was received, the device server shall terminate the command with a CHECK CONDITION status. The device server shall provide sense data that reports a unit attention condition for the SCSI initiator port that sent the command on the I_T nexus." But according to section 8.5 and 8.6, a command which is not yet executed because of the presence of other tasks in the task set (i.e., one for which target_handle_task_attr returns true) would not enter the enabled command state; it would be in the dormant command state. target_execute_cmd would get called when a command entered the enabled command state, and thus that is the appropriate place to check for Unit Attenion. Similarly, though not quite as explicit, section 5.3.3 tells us that a Reservation Conflict status has a lower precedence than a Unit Attention, and so this would also seem to be the appropriate place to call target_check_reservation. I'm less sure about target_alua_state_check, since I'm not very familiar with ALUA. Signed-off-by: Michael Cyr <redacted> --- drivers/target/target_core_transport.c | 41 ++++++++++++++++++++-------------- 1 file changed, 24 insertions(+), 17 deletions(-)
diff --git a/drivers/target/target_core_transport.c b/drivers/target/target_core_transport.c
index 6c089af..2ee5502 100644
--- a/drivers/target/target_core_transport.c
+++ b/drivers/target/target_core_transport.c@@ -1303,23 +1303,6 @@ target_setup_cmd_from_cdb(struct se_cmd *cmd, unsigned char *cdb) trace_target_sequencer_start(cmd); - /* - * Check for an existing UNIT ATTENTION condition - */ - ret = target_scsi3_ua_check(cmd); - if (ret) - return ret; - - ret = target_alua_state_check(cmd); - if (ret) - return ret; - - ret = target_check_reservation(cmd); - if (ret) { - cmd->scsi_status = SAM_STAT_RESERVATION_CONFLICT; - return ret; - } - ret = dev->transport->parse_cdb(cmd); if (ret == TCM_UNSUPPORTED_SCSI_OPCODE) pr_warn_ratelimited("%s/%s: Unsupported SCSI Opcode 0x%02x, sending CHECK_CONDITION.\n",
@@ -1865,6 +1848,8 @@ static int __transport_check_aborted_status(struct se_cmd *, int); void target_execute_cmd(struct se_cmd *cmd) { + sense_reason_t ret; + /* * Determine if frontend context caller is requesting the stopping of * this command for frontend exceptions.
@@ -1899,6 +1884,28 @@ void target_execute_cmd(struct se_cmd *cmd) return; } + /* + * Check for an existing UNIT ATTENTION condition + */ + ret = target_scsi3_ua_check(cmd); + if (ret) { + transport_generic_request_failure(cmd, ret); + return; + } + + ret = target_alua_state_check(cmd); + if (ret) { + transport_generic_request_failure(cmd, ret); + return; + } + + ret = target_check_reservation(cmd); + if (ret) { + cmd->scsi_status = SAM_STAT_RESERVATION_CONFLICT; + transport_generic_request_failure(cmd, ret); + return; + } + __target_execute_cmd(cmd); } EXPORT_SYMBOL(target_execute_cmd);
--
2.5.0