Re: Help pinpointing cause of tgtd instability

[ronnie sahlberg <ronniesahlberg@xxxxxxxxx>]

I have tried to write a small tool to generate a huge amount of READ10
tasks, then using both
ABORT_TASK as well as ABORT_TASK_SET
but I can not get it to crash here.


I have attached the ABORT_TASK/ABORT_TASK_SET torture program here in
case someone wants to
try to tweak it to reproduce the crashes.


It links with libiscsi.

regards
ronnie sahlberg


On Sat, Dec 10, 2011 at 8:34 PM, FUJITA Tomonori
<fujita.tomonori@xxxxxxxxxxxxx> wrote:
> On Thu, 8 Dec 2011 13:33:04 -0500
> John Pletka <jpletka@xxxxxxxxxxx> wrote:
>
>> What is the side effect of disabling the task manager?  Any important
>> functionality lost?  Will it need to be periodically restarted to
>
> I think that if the task management fails, a Linux initiator closes
> the connection and try to reconnect then if it works, the initiator
> sends the same command again. So the file system doesn't detect
> failure. If reconnecting fails, the command fails, then the file
> system hits the I/O error.
>
>
>> reset whatever the task manager was doing? This is going into a
>> low-priority production server, so anything except data corruption can
>> be dealt with.
>
> --
> To unsubscribe from this list: send the line "unsubscribe stgt" in
> the body of a message to majordomo@xxxxxxxxxxxxxxx
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

#define PORTAL "127.0.0.1:3260"
#define LUN 1
#define TARGET "iqn.ronnie.test"

#define NUM_ABORTS 1000000
#define MAX_CONCURRENT 16


#include <stdio.h>
#include <stdlib.h>
#include <stdint.h>
#include <string.h>
#include <poll.h>
#include <time.h>
#include <iscsi/iscsi.h>
#include <iscsi/scsi-lowlevel.h>

struct client {
       int finished;
       int num_aborts;
};

static void read10_cb(struct iscsi_context *iscsi, int status, void *command_data, void *private_data)
{
}

static void tm_at_cb(struct iscsi_context *iscsi, int status, void *command_data, void *private_data)
{
}

static void tm_ats_cb(struct iscsi_context *iscsi, int status, void *command_data, void *private_data)
{
	struct client *clnt = (struct client *)private_data;
	int i, num;

	if (clnt->num_aborts++ > NUM_ABORTS) {
		clnt->finished = 1;
	} else {
		struct scsi_task *task;

		num = random()%MAX_CONCURRENT + 1;
		for ( i = 0; i < num; i++) {
			if ((task = iscsi_read10_task(iscsi, LUN, 0, 512, 512, read10_cb, NULL)) == NULL) {
				printf("failed to send read10 command\n");
				exit(10);
			}
			/* abort some of the tasks immeidately and sometimes send two of them */
			if (num & 0x01) {			
				if (iscsi_task_mgmt_abort_task_async(iscsi, task, tm_at_cb, NULL) != 0) {
					printf("failed to send task management to abort the task\n");
					exit(10);
				}
			}
			if (num & 0x02) {			
				if (iscsi_task_mgmt_abort_task_async(iscsi, task, tm_at_cb, NULL) != 0) {
					printf("failed to send task management to abort the task\n");
					exit(10);
				}
			}
		}
		if (iscsi_task_mgmt_abort_task_set_async(iscsi, LUN, tm_ats_cb, clnt) != 0) {
			printf("failed to send task management to abort the task set\n");
			exit(10);
		}
	}
}

static void connect_cb(struct iscsi_context *iscsi, int status, void *command_data, void *private_data)
{
	struct client *clnt = (struct client *)private_data;

	printf("Connected to iscsi socket\n");

	if (status != 0) {
		printf("connect_cb: connection  failed status:%d\n", status);
		exit(10);
	}

	tm_ats_cb(iscsi, 0, command_data, private_data);
}

int main(int argc, char *argv[])
{
	struct iscsi_context *iscsi;
	struct pollfd pfd;
	struct client clnt;
	time_t t;

	memset(&clnt, 0, sizeof(struct client));

	iscsi = iscsi_create_context("iqn.2002-10.com.ronnie:tmgm-torture");
	if (iscsi == NULL) {
		printf("Failed to create context\n");
		exit(10);
	}

	iscsi_set_targetname(iscsi, TARGET);
	iscsi_set_session_type(iscsi, ISCSI_SESSION_NORMAL);
	iscsi_set_header_digest(iscsi, ISCSI_HEADER_DIGEST_NONE_CRC32C);

	if (iscsi_full_connect_async(iscsi, PORTAL, LUN, connect_cb, &clnt) != 0) {
		printf("iscsi_connect failed. %s\n", iscsi_get_error(iscsi));
		exit(10);
	}

	while (clnt.finished == 0) {
		time_t new_t = time(NULL);

		if (t != new_t) {
			printf("num tasks set aborts %d\n", clnt.num_aborts);
		}
		t = new_t;

		pfd.fd = iscsi_get_fd(iscsi);
		pfd.events = iscsi_which_events(iscsi);

		if (poll(&pfd, 1, -1) < 0) {
			printf("Poll failed");
			exit(10);
		}
		if (iscsi_service(iscsi, pfd.revents) < 0) {
			printf("iscsi_service failed with : %s\n", iscsi_get_error(iscsi));
			break;
		}
	}


	iscsi_destroy_context(iscsi);

	printf("finished\n");
	return 0;
}