| Message ID | 20211111133752.159379-1-mlombard@redhat.com (mailing list archive) |
|---|---|
| State | Deferred |
| Headers | show |
| Series | [V2] target: iscsi: simplify the connection closing mechanism | expand |
On 11/11/21 7:37 AM, Maurizio Lombardi wrote: > When the connection reinstatement is performed, the target driver > executes a complex scheme of complete()/wait_for_completion() that is not > really needed. > > Considering that: > > 1) The callers of iscsit_connection_reinstatement_rcfr() and > iscsit_cause_connection_reinstatement() hold a reference > to the conn structure.> > 2) iscsit_close_connection() will sleep when calling > iscsit_check_conn_usage_count() until the conn structure's refcount > reaches zero. > > we can optimize the driver the following way: > > * The threads that must sleep until the connection is closed > will all wait for the "conn_wait_comp" completion, > iscsit_close_connection() will then call complete_all() to wake them up. > No need to have multiple completion structures. > > * The conn_post_wait_comp completion is not necessary and can be removed > because iscsit_close_connection() sleeps until all the other threads > release the conn structure. > (see the iscsit_check_conn_usage_count() function) > > V2: do not set connection_reinstatement to 1 in iscsit_close_connection(), > leave iscsit_cause_connection_reinstatement() deal with reentrancy. > What was the issue with setting it to 1?
On Tue, Dec 07, 2021 at 01:59:08PM -0600, Mike Christie wrote: > On 11/11/21 7:37 AM, Maurizio Lombardi wrote: > > When the connection reinstatement is performed, the target driver > > executes a complex scheme of complete()/wait_for_completion() that is not > > really needed. > > > > Considering that: > > > > 1) The callers of iscsit_connection_reinstatement_rcfr() and > > iscsit_cause_connection_reinstatement() hold a reference > > to the conn structure.> > 2) iscsit_close_connection() will sleep when calling > > iscsit_check_conn_usage_count() until the conn structure's refcount > > reaches zero. > > > > we can optimize the driver the following way: > > > > * The threads that must sleep until the connection is closed > > will all wait for the "conn_wait_comp" completion, > > iscsit_close_connection() will then call complete_all() to wake them up. > > No need to have multiple completion structures. > > > > * The conn_post_wait_comp completion is not necessary and can be removed > > because iscsit_close_connection() sleeps until all the other threads > > release the conn structure. > > (see the iscsit_check_conn_usage_count() function) > > > > V2: do not set connection_reinstatement to 1 in iscsit_close_connection(), > > leave iscsit_cause_connection_reinstatement() deal with reentrancy. > > > > What was the issue with setting it to 1? > At first I thought that setting it to 1 could introduce a race condition, then I realized that it's not the case, setting it to 1 simply has no effect and therefore the assignment can be removed. Maurizio
diff --git a/drivers/target/iscsi/iscsi_target.c b/drivers/target/iscsi/iscsi_target.c index 2c54c5d8412d..d558c4c064cb 100644 --- a/drivers/target/iscsi/iscsi_target.c +++ b/drivers/target/iscsi/iscsi_target.c @@ -4226,31 +4226,10 @@ int iscsit_close_connection( /* * If connection reinstatement is being performed on this connection, * up the connection reinstatement semaphore that is being blocked on - * in iscsit_cause_connection_reinstatement(). + * in iscsit_cause_connection_reinstatement() or + * in iscsit_connection_reinstatement_rcfr() */ - spin_lock_bh(&conn->state_lock); - if (atomic_read(&conn->sleep_on_conn_wait_comp)) { - spin_unlock_bh(&conn->state_lock); - complete(&conn->conn_wait_comp); - wait_for_completion(&conn->conn_post_wait_comp); - spin_lock_bh(&conn->state_lock); - } - - /* - * If connection reinstatement is being performed on this connection - * by receiving a REMOVECONNFORRECOVERY logout request, up the - * connection wait rcfr semaphore that is being blocked on - * an iscsit_connection_reinstatement_rcfr(). - */ - if (atomic_read(&conn->connection_wait_rcfr)) { - spin_unlock_bh(&conn->state_lock); - complete(&conn->conn_wait_rcfr_comp); - wait_for_completion(&conn->conn_post_wait_comp); - spin_lock_bh(&conn->state_lock); - } - atomic_set(&conn->connection_reinstatement, 1); - spin_unlock_bh(&conn->state_lock); - + complete_all(&conn->conn_wait_comp); /* * If any other processes are accessing this connection pointer we * must wait until they have completed. diff --git a/drivers/target/iscsi/iscsi_target_erl0.c b/drivers/target/iscsi/iscsi_target_erl0.c index 102c9cbf59f3..584e0a0b517d 100644 --- a/drivers/target/iscsi/iscsi_target_erl0.c +++ b/drivers/target/iscsi/iscsi_target_erl0.c @@ -839,8 +839,7 @@ void iscsit_connection_reinstatement_rcfr(struct iscsi_conn *conn) send_sig(SIGINT, conn->rx_thread, 1); sleep: - wait_for_completion(&conn->conn_wait_rcfr_comp); - complete(&conn->conn_post_wait_comp); + wait_for_completion(&conn->conn_wait_comp); } void iscsit_cause_connection_reinstatement(struct iscsi_conn *conn, int sleep) @@ -871,12 +870,9 @@ void iscsit_cause_connection_reinstatement(struct iscsi_conn *conn, int sleep) spin_unlock_bh(&conn->state_lock); return; } - - atomic_set(&conn->sleep_on_conn_wait_comp, 1); spin_unlock_bh(&conn->state_lock); wait_for_completion(&conn->conn_wait_comp); - complete(&conn->conn_post_wait_comp); } EXPORT_SYMBOL(iscsit_cause_connection_reinstatement); diff --git a/drivers/target/iscsi/iscsi_target_login.c b/drivers/target/iscsi/iscsi_target_login.c index 1a9c50401bdb..982c23459272 100644 --- a/drivers/target/iscsi/iscsi_target_login.c +++ b/drivers/target/iscsi/iscsi_target_login.c @@ -1096,9 +1096,7 @@ static struct iscsi_conn *iscsit_alloc_conn(struct iscsi_np *np) INIT_LIST_HEAD(&conn->conn_cmd_list); INIT_LIST_HEAD(&conn->immed_queue_list); INIT_LIST_HEAD(&conn->response_queue_list); - init_completion(&conn->conn_post_wait_comp); init_completion(&conn->conn_wait_comp); - init_completion(&conn->conn_wait_rcfr_comp); init_completion(&conn->conn_waiting_on_uc_comp); init_completion(&conn->conn_logout_comp); init_completion(&conn->rx_half_close_comp); diff --git a/drivers/target/iscsi/iscsi_target_util.c b/drivers/target/iscsi/iscsi_target_util.c index 6dd5810e2af1..d7b1f9110d49 100644 --- a/drivers/target/iscsi/iscsi_target_util.c +++ b/drivers/target/iscsi/iscsi_target_util.c @@ -824,9 +824,6 @@ struct iscsi_conn *iscsit_get_conn_from_cid_rcfr(struct iscsi_session *sess, u16 list_for_each_entry(conn, &sess->sess_conn_list, conn_list) { if (conn->cid == cid) { iscsit_inc_conn_usage_count(conn); - spin_lock(&conn->state_lock); - atomic_set(&conn->connection_wait_rcfr, 1); - spin_unlock(&conn->state_lock); spin_unlock_bh(&sess->conn_lock); return conn; } diff --git a/include/target/iscsi/iscsi_target_core.h b/include/target/iscsi/iscsi_target_core.h index 1eccb2ac7d02..aeb8932507c2 100644 --- a/include/target/iscsi/iscsi_target_core.h +++ b/include/target/iscsi/iscsi_target_core.h @@ -542,12 +542,8 @@ struct iscsi_conn { atomic_t connection_exit; atomic_t connection_recovery; atomic_t connection_reinstatement; - atomic_t connection_wait_rcfr; - atomic_t sleep_on_conn_wait_comp; atomic_t transport_failed; - struct completion conn_post_wait_comp; struct completion conn_wait_comp; - struct completion conn_wait_rcfr_comp; struct completion conn_waiting_on_uc_comp; struct completion conn_logout_comp; struct completion tx_half_close_comp;
When the connection reinstatement is performed, the target driver executes a complex scheme of complete()/wait_for_completion() that is not really needed. Considering that: 1) The callers of iscsit_connection_reinstatement_rcfr() and iscsit_cause_connection_reinstatement() hold a reference to the conn structure. 2) iscsit_close_connection() will sleep when calling iscsit_check_conn_usage_count() until the conn structure's refcount reaches zero. we can optimize the driver the following way: * The threads that must sleep until the connection is closed will all wait for the "conn_wait_comp" completion, iscsit_close_connection() will then call complete_all() to wake them up. No need to have multiple completion structures. * The conn_post_wait_comp completion is not necessary and can be removed because iscsit_close_connection() sleeps until all the other threads release the conn structure. (see the iscsit_check_conn_usage_count() function) V2: do not set connection_reinstatement to 1 in iscsit_close_connection(), leave iscsit_cause_connection_reinstatement() deal with reentrancy. Signed-off-by: Maurizio Lombardi <mlombard@redhat.com> --- drivers/target/iscsi/iscsi_target.c | 27 +++-------------------- drivers/target/iscsi/iscsi_target_erl0.c | 6 +---- drivers/target/iscsi/iscsi_target_login.c | 2 -- drivers/target/iscsi/iscsi_target_util.c | 3 --- include/target/iscsi/iscsi_target_core.h | 4 ---- 5 files changed, 4 insertions(+), 38 deletions(-)