Reconnect: If we are logged in and we experience a session failure, then

try to re-connect and redrive all I/O

Signed-off-by: Ronnie Sahlberg <ronniesahlberg@gmail.com>

include/iscsi-private.h

@@ -87,6 +87,7 @@ struct iscsi_context {
 	int secneg_phase;
 	int login_attempts;
 	int is_loggedin;
+	int is_reconnecting;
 
 	int chap_a;
 	int chap_i;
@@ -109,6 +110,9 @@ struct iscsi_context {
 	enum iscsi_initial_r2t use_initial_r2t;
 	enum iscsi_immediate_data want_immediate_data;
 	enum iscsi_immediate_data use_immediate_data;
+
+	int lun;
+	const char *portal;
 };
 
 #define ISCSI_PDU_IMMEDIATE		       0x40
@@ -261,5 +265,7 @@ unsigned long crc32c(char *buf, int len);
 
 struct scsi_task *iscsi_scsi_get_task_from_pdu(struct iscsi_pdu *pdu);
 
+int iscsi_reconnect(struct iscsi_context *iscsi);
+
 #endif /* __iscsi_private_h__ */
 

lib/connect.c

@@ -15,8 +15,11 @@
    along with this program; if not, see <http://www.gnu.org/licenses/>.
 */
 #include <stdio.h>
+#include <string.h>
 #include <stdlib.h>
+#include <unistd.h>
 #include <errno.h>
+#include "slist.h"
 #include "iscsi.h"
 #include "iscsi-private.h"
 #include "scsi-lowlevel.h"
@@ -109,6 +112,9 @@ iscsi_full_connect_async(struct iscsi_context *iscsi, const char *portal,
 {
 	struct connect_task *ct;
 
+	iscsi->lun = lun;
+	iscsi->portal = strdup(portal);
+
 	ct = malloc(sizeof(struct connect_task));
 	if (ct == NULL) {
 		iscsi_set_error(iscsi, "Out-of-memory. Failed to allocate "
@@ -124,3 +130,106 @@ iscsi_full_connect_async(struct iscsi_context *iscsi, const char *portal,
 	}
 	return 0;
 }
+
+int iscsi_reconnect(struct iscsi_context *old_iscsi)
+{
+	struct iscsi_context *iscsi;
+
+try_again:
+
+	iscsi = iscsi_create_context(old_iscsi->initiator_name);
+	iscsi->is_reconnecting = 1;
+
+	iscsi_set_targetname(iscsi, old_iscsi->target_name);
+
+	iscsi_set_header_digest(iscsi, old_iscsi->want_header_digest);
+
+	if (old_iscsi->user != NULL) {
+		iscsi_set_initiator_username_pwd(iscsi, old_iscsi->user, old_iscsi->passwd);
+	}
+
+	iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL);
+
+	iscsi->lun = old_iscsi->lun;
+
+	iscsi->portal = strdup(old_iscsi->portal);
+
+	if (iscsi_full_connect_sync(iscsi, iscsi->portal, iscsi->lun) != 0) {
+		iscsi_destroy_context(iscsi);
+		sleep(1);
+		goto try_again;
+	}
+
+	while (old_iscsi->waitpdu) {
+		struct iscsi_pdu *pdu = old_iscsi->waitpdu;
+
+		SLIST_REMOVE(&old_iscsi->waitpdu, pdu);
+		if (pdu->itt == 0xffffffff) {
+			continue;
+		}
+
+		pdu->itt   = iscsi->itt++;
+		iscsi_pdu_set_itt(pdu, pdu->itt);
+
+		pdu->cmdsn = iscsi->cmdsn++;
+		iscsi_pdu_set_cmdsn(pdu, pdu->cmdsn);
+
+		iscsi_pdu_set_expstatsn(pdu, iscsi->statsn);
+		iscsi->statsn++;
+
+		pdu->written = 0;
+		SLIST_ADD_END(&iscsi->outqueue, pdu);
+	}
+
+	while (old_iscsi->outqueue) {
+		struct iscsi_pdu *pdu = old_iscsi->outqueue;
+
+		SLIST_REMOVE(&old_iscsi->outqueue, pdu);
+		if (pdu->itt == 0xffffffff) {
+			continue;
+		}
+
+		pdu->itt   = iscsi->itt++;
+		iscsi_pdu_set_itt(pdu, pdu->itt);
+
+		pdu->cmdsn = iscsi->cmdsn++;
+		iscsi_pdu_set_cmdsn(pdu, pdu->cmdsn);
+
+		iscsi_pdu_set_expstatsn(pdu, iscsi->statsn);
+		iscsi->statsn++;
+
+		pdu->written = 0;
+		SLIST_ADD_END(&iscsi->outqueue, pdu);
+	}
+
+	if (dup2(iscsi->fd, old_iscsi->fd) == -1) {
+		iscsi_destroy_context(iscsi);
+		goto try_again;
+	}
+	
+
+	free(discard_const(old_iscsi->initiator_name));
+	free(discard_const(old_iscsi->target_name));
+	free(discard_const(old_iscsi->target_address));
+	free(discard_const(old_iscsi->alias));
+	free(discard_const(old_iscsi->portal));
+	if (old_iscsi->incoming != NULL) {
+		iscsi_free_iscsi_in_pdu(old_iscsi->incoming);
+	}
+	if (old_iscsi->inqueue != NULL) {
+		iscsi_free_iscsi_inqueue(old_iscsi->inqueue);
+	}
+	free(old_iscsi->error_string);
+	free(discard_const(old_iscsi->user));
+	free(discard_const(old_iscsi->passwd));
+	free(discard_const(old_iscsi->chap_c));
+
+	close(iscsi->fd);
+	iscsi->fd = old_iscsi->fd;
+	memcpy(old_iscsi, iscsi, sizeof(struct iscsi_context));
+	free(iscsi);
+
+	old_iscsi->is_reconnecting = 0;
+
+	return 0;
+}

lib/init.c

@@ -188,15 +188,25 @@ iscsi_destroy_context(struct iscsi_context *iscsi)
 	while ((pdu = iscsi->outqueue)) {
 		SLIST_REMOVE(&iscsi->outqueue, pdu);
 		if ( !(pdu->flags & ISCSI_PDU_NO_CALLBACK)) {
-			pdu->callback(iscsi, SCSI_STATUS_CANCELLED, NULL,
-				      pdu->private_data);
+			/* If an error happened during connect/login, we dont want to
+			   call any of the callbacks.
+			 */
+			if (iscsi->is_loggedin) {
+				pdu->callback(iscsi, SCSI_STATUS_CANCELLED, NULL,
+						pdu->private_data);
+			}		      
 		}
 		iscsi_free_pdu(iscsi, pdu);
 	}
 	while ((pdu = iscsi->waitpdu)) {
 		SLIST_REMOVE(&iscsi->waitpdu, pdu);
-		pdu->callback(iscsi, SCSI_STATUS_CANCELLED, NULL,
-				     pdu->private_data);
+		/* If an error happened during connect/login, we dont want to
+		   call any of the callbacks.
+		 */
+		if (iscsi->is_loggedin) {
+			pdu->callback(iscsi, SCSI_STATUS_CANCELLED, NULL,
+					pdu->private_data);
+		}
 		iscsi_free_pdu(iscsi, pdu);
 	}
 
@@ -212,6 +222,9 @@ iscsi_destroy_context(struct iscsi_context *iscsi)
 	free(discard_const(iscsi->alias));
 	iscsi->alias = NULL;
 
+	free(discard_const(iscsi->portal));
+	iscsi->portal = NULL;
+
 	if (iscsi->incoming != NULL) {
 		iscsi_free_iscsi_in_pdu(iscsi->incoming);
 	}
@@ -228,7 +241,6 @@ iscsi_destroy_context(struct iscsi_context *iscsi)
 	free(discard_const(iscsi->passwd));
 	iscsi->passwd = NULL;
 
-
 	free(discard_const(iscsi->chap_c));
 	iscsi->chap_c = NULL;
 

lib/socket.c

@@ -410,6 +410,11 @@ iscsi_service(struct iscsi_context *iscsi, int revents)
 		}
 		iscsi->socket_status_cb(iscsi, SCSI_STATUS_ERROR, NULL,
 					iscsi->connect_data);
+		if (iscsi->is_loggedin) {
+			if (iscsi_reconnect(iscsi) == 0) {
+				return 0;
+			}
+		}
 		return -1;
 	}
 	if (revents & POLLHUP) {
@@ -417,6 +422,11 @@ iscsi_service(struct iscsi_context *iscsi, int revents)
 				"socket error.");
 		iscsi->socket_status_cb(iscsi, SCSI_STATUS_ERROR, NULL,
 					iscsi->connect_data);
+		if (iscsi->is_loggedin) {
+			if (iscsi_reconnect(iscsi) == 0) {
+				return 0;
+			}
+		}
 		return -1;
 	}
 
@@ -434,6 +444,11 @@ iscsi_service(struct iscsi_context *iscsi, int revents)
 					strerror(err), err);
 			iscsi->socket_status_cb(iscsi, SCSI_STATUS_ERROR,
 						NULL, iscsi->connect_data);
+			if (iscsi->is_loggedin) {
+				if (iscsi_reconnect(iscsi) == 0) {
+					return 0;
+				}
+			}
 			return -1;
 		}
 
@@ -448,12 +463,23 @@ iscsi_service(struct iscsi_context *iscsi, int revents)
 
 	if (revents & POLLOUT && iscsi->outqueue != NULL) {
 		if (iscsi_write_to_socket(iscsi) != 0) {
+			if (iscsi->is_loggedin) {
+				if (iscsi_reconnect(iscsi) == 0) {
+					return 0;
+				}
+			}
 			return -1;
 		}
 	}
 	if (revents & POLLIN) {
-		if (iscsi_read_from_socket(iscsi) != 0)
+		if (iscsi_read_from_socket(iscsi) != 0) {
+			if (iscsi->is_loggedin) {
+				if (iscsi_reconnect(iscsi) == 0) {
+					return 0;
+				}
+			}
 			return -1;
+		}
 	}
 
 	return 0;

lib/sync.c

@@ -38,19 +38,29 @@ static void
 event_loop(struct iscsi_context *iscsi, struct iscsi_sync_state *state)
 {
 	struct pollfd pfd;
+	int ret;
 
 	while (state->finished == 0) {
 		pfd.fd = iscsi_get_fd(iscsi);
 		pfd.events = iscsi_which_events(iscsi);
 
-		if (poll(&pfd, 1, -1) < 0) {
+		if ((ret = poll(&pfd, 1, 1000)) < 0) {
 			iscsi_set_error(iscsi, "Poll failed");
 			return;
 		}
+		if (ret == 0) {
+			/* poll timedout, try again */
+			continue;
+		}
 		if (iscsi_service(iscsi, pfd.revents) < 0) {
+			/* if we are reconnecting we just try and try again */
+			if (iscsi->is_reconnecting) {
+				continue;
+			}
+
 			iscsi_set_error(iscsi,
-					"iscsi_service failed with : %s",
-					iscsi_get_error(iscsi));
+				"iscsi_service failed with : %s",
+				iscsi_get_error(iscsi));
 			return;
 		}
 	}