Btw, here's the updated patch: >From 731b9981fc49d9b11e93b5c2f0a56e27862b4335 Mon Sep 17 00:00:00 2001 From: Nicholas Bellinger <[email protected]> Date: Tue, 19 Jan 2016 15:23:02 -0800 Subject: [PATCH] target: Fix remote-port TMR ABORT + se_cmd fabric stop
To address the bug where fabric driver level shutdown of se_cmd occurs at the same time when TMR CMD_T_ABORTED is happening resulting in a -1 ->cmd_kref, this patch adds a target_tmr_put_cmd() wrapper + CMD_T_FABRIC_STOP bit that is used to determine when TMR + driver I_T nexus shutdown is happening concurrently. It changes target_sess_cmd_list_set_waiting() to obtain se_cmd->cmd_kref + set CMD_T_FABRIC_STOP, and drop local reference in target_wait_for_sess_cmds() and invoke extra target_put_sess_cmd() during Task Aborted Status (TAS) when necessary. Also, it adds a new target_wait_free_cmd() wrapper around transport_wait_for_tasks() for the special case within transport_generic_free_cmd() to set CMD_T_FABRIC_STOP, and is now aware of CMD_T_ABORTED + CMD_T_TAS status bits to know when an extra transport_put_cmd() during TAS is required. Note transport_generic_free_cmd() is expected to block on cmd->cmd_wait_comp in order to follow what iscsi-target expects during iscsi_conn context se_cmd shutdown. Cc: Quinn Tran <[email protected]> Cc: Himanshu Madhani <[email protected]> Cc: Sagi Grimberg <[email protected]> Cc: Christoph Hellwig <[email protected]> Cc: Hannes Reinecke <[email protected]> Cc: Andy Grover <[email protected]> Cc: Mike Christie <[email protected]> Cc: [email protected] # 3.10+ Signed-off-by: Nicholas Bellinger <[email protected]> --- drivers/target/target_core_tmr.c | 39 +++++++++- drivers/target/target_core_transport.c | 132 ++++++++++++++++++++++++--------- include/target/target_core_base.h | 2 + 3 files changed, 136 insertions(+), 37 deletions(-) diff --git a/drivers/target/target_core_tmr.c b/drivers/target/target_core_tmr.c index 9d67d16..2ac3228 100644 --- a/drivers/target/target_core_tmr.c +++ b/drivers/target/target_core_tmr.c @@ -110,6 +110,7 @@ static int target_check_cdb_and_preempt(struct list_head *list, static bool __target_check_io_state(struct se_cmd *se_cmd) { struct se_session *sess = se_cmd->se_sess; + bool ret; assert_spin_locked(&sess->sess_cmd_lock); WARN_ON_ONCE(!irqs_disabled()); @@ -129,10 +130,36 @@ static bool __target_check_io_state(struct se_cmd *se_cmd) spin_unlock(&se_cmd->t_state_lock); return false; } + if (sess->sess_tearing_down || se_cmd->cmd_wait_set) { + pr_debug("Attempted to abort io tag: %llu already shutdown," + " skipping\n", se_cmd->tag); + spin_unlock(&se_cmd->t_state_lock); + return false; + } se_cmd->transport_state |= CMD_T_ABORTED; spin_unlock(&se_cmd->t_state_lock); - return kref_get_unless_zero(&se_cmd->cmd_kref); + ret = kref_get_unless_zero(&se_cmd->cmd_kref); + if (ret) + se_cmd->cmd_wait_set = 1; + return ret; +} + +static void target_tmr_put_cmd(struct se_cmd *se_cmd) +{ + unsigned long flags; + bool fabric_stop; + + spin_lock_irqsave(&se_cmd->t_state_lock, flags); + fabric_stop = (se_cmd->transport_state & CMD_T_FABRIC_STOP); + spin_unlock_irqrestore(&se_cmd->t_state_lock, flags); + + target_put_sess_cmd(se_cmd); + + if (!fabric_stop) { + wait_for_completion(&se_cmd->cmd_wait_comp); + se_cmd->se_tfo->release_cmd(se_cmd); + } } void core_tmr_abort_task( @@ -178,7 +205,7 @@ void core_tmr_abort_task( transport_wait_for_tasks(se_cmd); transport_cmd_finish_abort(se_cmd, true); - target_put_sess_cmd(se_cmd); + target_tmr_put_cmd(se_cmd); printk("ABORT_TASK: Sending TMR_FUNCTION_COMPLETE for" " ref_tag: %llu\n", ref_tag); @@ -252,8 +279,12 @@ static void core_tmr_drain_tmr_list( spin_unlock(&sess->sess_cmd_lock); if (!rc) { printk("LUN_RESET TMR: non-zero kref_get_unless_zero\n"); + spin_unlock(&sess->sess_cmd_lock); continue; } + cmd->cmd_wait_set = true; + spin_unlock(&sess->sess_cmd_lock); + list_move_tail(&tmr_p->tmr_list, &drain_tmr_list); } spin_unlock_irqrestore(&dev->se_tmr_lock, flags); @@ -271,7 +302,7 @@ static void core_tmr_drain_tmr_list( transport_wait_for_tasks(cmd); transport_cmd_finish_abort(cmd, 1); - target_put_sess_cmd(cmd); + target_tmr_put_cmd(cmd); } } @@ -370,7 +401,7 @@ static void core_tmr_drain_state_list( transport_wait_for_tasks(cmd); core_tmr_handle_tas_abort(tmr_sess, cmd, tas); - target_put_sess_cmd(cmd); + target_tmr_put_cmd(cmd); } } diff --git a/drivers/target/target_core_transport.c b/drivers/target/target_core_transport.c index 94e372a..a0c6e2e 100644 --- a/drivers/target/target_core_transport.c +++ b/drivers/target/target_core_transport.c @@ -2431,18 +2431,33 @@ static void transport_write_pending_qf(struct se_cmd *cmd) } } +static bool +__transport_wait_for_tasks(struct se_cmd *, bool, bool *, bool *, + unsigned long *flags); + +static void target_wait_free_cmd(struct se_cmd *cmd, bool *aborted, bool *tas) +{ + unsigned long flags; + + spin_lock_irqsave(&cmd->t_state_lock, flags); + __transport_wait_for_tasks(cmd, true, aborted, tas, &flags); + spin_unlock_irqrestore(&cmd->t_state_lock, flags); +} + int transport_generic_free_cmd(struct se_cmd *cmd, int wait_for_tasks) { int ret = 0; + bool aborted = false, tas = false; if (!(cmd->se_cmd_flags & SCF_SE_LUN_CMD)) { if (wait_for_tasks && (cmd->se_cmd_flags & SCF_SCSI_TMR_CDB)) - transport_wait_for_tasks(cmd); + target_wait_free_cmd(cmd, &aborted, &tas); - ret = transport_put_cmd(cmd); + if (!aborted || tas) + ret = transport_put_cmd(cmd); } else { if (wait_for_tasks) - transport_wait_for_tasks(cmd); + target_wait_free_cmd(cmd, &aborted, &tas); /* * Handle WRITE failure case where transport_generic_new_cmd() * has already added se_cmd to state_list, but fabric has @@ -2454,7 +2469,20 @@ int transport_generic_free_cmd(struct se_cmd *cmd, int wait_for_tasks) if (cmd->se_lun) transport_lun_remove_cmd(cmd); - ret = transport_put_cmd(cmd); + if (!aborted || tas) + ret = transport_put_cmd(cmd); + } + /* + * If the task has been internally aborted due to TMR ABORT_TASK + * or LUN_RESET, target_core_tmr.c is responsible for performing + * the remaining calls to target_put_sess_cmd(), and not the + * callers of this function. + */ + if (aborted) { + pr_debug("Detected CMD_T_ABORTED for ITT: %llu\n", cmd->tag); + wait_for_completion(&cmd->cmd_wait_comp); + cmd->se_tfo->release_cmd(cmd); + ret = 1; } return ret; } @@ -2517,7 +2545,7 @@ static void target_release_cmd_kref(struct kref *kref) se_cmd->se_tfo->release_cmd(se_cmd); return; } - if (se_sess->sess_tearing_down && se_cmd->cmd_wait_set) { + if (se_cmd->cmd_wait_set) { spin_unlock_irqrestore(&se_sess->sess_cmd_lock, flags); target_free_cmd_mem(se_cmd); complete(&se_cmd->cmd_wait_comp); @@ -2555,6 +2583,7 @@ void target_sess_cmd_list_set_waiting(struct se_session *se_sess) { struct se_cmd *se_cmd; unsigned long flags; + int rc; spin_lock_irqsave(&se_sess->sess_cmd_lock, flags); if (se_sess->sess_tearing_down) { @@ -2564,8 +2593,15 @@ void target_sess_cmd_list_set_waiting(struct se_session *se_sess) se_sess->sess_tearing_down = 1; list_splice_init(&se_sess->sess_cmd_list, &se_sess->sess_wait_list); - list_for_each_entry(se_cmd, &se_sess->sess_wait_list, se_cmd_list) - se_cmd->cmd_wait_set = 1; + list_for_each_entry(se_cmd, &se_sess->sess_wait_list, se_cmd_list) { + rc = kref_get_unless_zero(&se_cmd->cmd_kref); + if (rc) { + se_cmd->cmd_wait_set = 1; + spin_lock(&se_cmd->t_state_lock); + se_cmd->transport_state |= CMD_T_FABRIC_STOP; + spin_unlock(&se_cmd->t_state_lock); + } + } spin_unlock_irqrestore(&se_sess->sess_cmd_lock, flags); } @@ -2578,6 +2614,7 @@ void target_wait_for_sess_cmds(struct se_session *se_sess) { struct se_cmd *se_cmd, *tmp_cmd; unsigned long flags; + bool tas; list_for_each_entry_safe(se_cmd, tmp_cmd, &se_sess->sess_wait_list, se_cmd_list) { @@ -2587,6 +2624,15 @@ void target_wait_for_sess_cmds(struct se_session *se_sess) " %d\n", se_cmd, se_cmd->t_state, se_cmd->se_tfo->get_cmd_state(se_cmd)); + spin_lock_irqsave(&se_cmd->t_state_lock, flags); + tas = (se_cmd->transport_state & CMD_T_TAS); + spin_unlock_irqrestore(&se_cmd->t_state_lock, flags); + + if (!target_put_sess_cmd(se_cmd)) { + if (tas) + target_put_sess_cmd(se_cmd); + } + wait_for_completion(&se_cmd->cmd_wait_comp); pr_debug("After cmd_wait_comp: se_cmd: %p t_state: %d" " fabric state: %d\n", se_cmd, se_cmd->t_state, @@ -2608,53 +2654,72 @@ void transport_clear_lun_ref(struct se_lun *lun) wait_for_completion(&lun->lun_ref_comp); } -/** - * transport_wait_for_tasks - wait for completion to occur - * @cmd: command to wait - * - * Called from frontend fabric context to wait for storage engine - * to pause and/or release frontend generated struct se_cmd. - */ -bool transport_wait_for_tasks(struct se_cmd *cmd) +static bool +__transport_wait_for_tasks(struct se_cmd *cmd, bool fabric_stop, + bool *aborted, bool *tas, unsigned long *flags) + __releases(&cmd->t_state_lock) + __acquires(&cmd->t_state_lock) { - unsigned long flags; - spin_lock_irqsave(&cmd->t_state_lock, flags); + assert_spin_locked(&cmd->t_state_lock); + WARN_ON_ONCE(!irqs_disabled()); + + if (fabric_stop) + cmd->transport_state |= CMD_T_FABRIC_STOP; + + if (cmd->transport_state & CMD_T_ABORTED) + *aborted = true; + + if (cmd->transport_state & CMD_T_TAS) + *tas = true; + if (!(cmd->se_cmd_flags & SCF_SE_LUN_CMD) && - !(cmd->se_cmd_flags & SCF_SCSI_TMR_CDB)) { - spin_unlock_irqrestore(&cmd->t_state_lock, flags); + !(cmd->se_cmd_flags & SCF_SCSI_TMR_CDB)) return false; - } if (!(cmd->se_cmd_flags & SCF_SUPPORTED_SAM_OPCODE) && - !(cmd->se_cmd_flags & SCF_SCSI_TMR_CDB)) { - spin_unlock_irqrestore(&cmd->t_state_lock, flags); + !(cmd->se_cmd_flags & SCF_SCSI_TMR_CDB)) return false; - } - if (!(cmd->transport_state & CMD_T_ACTIVE)) { - spin_unlock_irqrestore(&cmd->t_state_lock, flags); + if (!(cmd->transport_state & CMD_T_ACTIVE)) return false; - } cmd->transport_state |= CMD_T_STOP; - pr_debug("wait_for_tasks: Stopping %p ITT: 0x%08llx i_state: %d, t_state: %d, CMD_T_STOP\n", - cmd, cmd->tag, cmd->se_tfo->get_cmd_state(cmd), cmd->t_state); + pr_debug("wait_for_tasks: Stopping %p ITT: 0x%08llx i_state: %d," + " t_state: %d, CMD_T_STOP\n", cmd, cmd->tag, + cmd->se_tfo->get_cmd_state(cmd), cmd->t_state); - spin_unlock_irqrestore(&cmd->t_state_lock, flags); + spin_unlock_irqrestore(&cmd->t_state_lock, *flags); wait_for_completion(&cmd->t_transport_stop_comp); - spin_lock_irqsave(&cmd->t_state_lock, flags); + spin_lock_irqsave(&cmd->t_state_lock, *flags); cmd->transport_state &= ~(CMD_T_ACTIVE | CMD_T_STOP); - pr_debug("wait_for_tasks: Stopped wait_for_completion(&cmd->t_transport_stop_comp) for ITT: 0x%08llx\n", - cmd->tag); + pr_debug("wait_for_tasks: Stopped wait_for_completion(&cmd->" + "t_transport_stop_comp) for ITT: 0x%08llx\n", cmd->tag); + + return true; +} +/** + * transport_wait_for_tasks - wait for completion to occur + * @cmd: command to wait + * + * Called from frontend fabric context to wait for storage engine + * to pause and/or release frontend generated struct se_cmd. + */ +bool transport_wait_for_tasks(struct se_cmd *cmd) +{ + unsigned long flags; + bool ret, aborted = false, tas = false; + + spin_lock_irqsave(&cmd->t_state_lock, flags); + ret = __transport_wait_for_tasks(cmd, false, &aborted, &tas, &flags); spin_unlock_irqrestore(&cmd->t_state_lock, flags); - return true; + return ret; } EXPORT_SYMBOL(transport_wait_for_tasks); @@ -2869,6 +2934,7 @@ void transport_send_task_abort(struct se_cmd *cmd) spin_unlock_irqrestore(&cmd->t_state_lock, flags); return; } + cmd->transport_state |= CMD_T_TAS; spin_unlock_irqrestore(&cmd->t_state_lock, flags); /* diff --git a/include/target/target_core_base.h b/include/target/target_core_base.h index 1a76726..1579539e 100644 --- a/include/target/target_core_base.h +++ b/include/target/target_core_base.h @@ -493,6 +493,8 @@ struct se_cmd { #define CMD_T_DEV_ACTIVE (1 << 7) #define CMD_T_REQUEST_STOP (1 << 8) #define CMD_T_BUSY (1 << 9) +#define CMD_T_TAS (1 << 10) +#define CMD_T_FABRIC_STOP (1 << 11) spinlock_t t_state_lock; struct kref cmd_kref; struct completion t_transport_stop_comp; -- To unsubscribe from this list: send the line "unsubscribe linux-scsi" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html
