Spdk/lib/iscsi/conn.c

1681 lines
44 KiB
C
Raw Normal View History

/*-
* BSD LICENSE
*
* Copyright (C) 2008-2012 Daisuke Aoyama <aoyama@peach.ne.jp>.
* Copyright (c) Intel Corporation.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
*
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in
* the documentation and/or other materials provided with the
* distribution.
* * Neither the name of Intel Corporation nor the names of its
* contributors may be used to endorse or promote products derived
* from this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
* A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
* OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
* LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#include "spdk/stdinc.h"
#include "spdk/endian.h"
#include "spdk/env.h"
#include "spdk/likely.h"
#include "spdk/thread.h"
#include "spdk/queue.h"
#include "spdk/trace.h"
#include "spdk/net.h"
#include "spdk/sock.h"
#include "spdk/string.h"
#include "spdk/log.h"
#include "iscsi/task.h"
#include "iscsi/conn.h"
#include "iscsi/tgt_node.h"
#include "iscsi/portal_grp.h"
#define MAKE_DIGEST_WORD(BUF, CRC32C) \
( ((*((uint8_t *)(BUF)+0)) = (uint8_t)((uint32_t)(CRC32C) >> 0)), \
((*((uint8_t *)(BUF)+1)) = (uint8_t)((uint32_t)(CRC32C) >> 8)), \
((*((uint8_t *)(BUF)+2)) = (uint8_t)((uint32_t)(CRC32C) >> 16)), \
((*((uint8_t *)(BUF)+3)) = (uint8_t)((uint32_t)(CRC32C) >> 24)))
#define SPDK_ISCSI_CONNECTION_MEMSET(conn) \
memset(&(conn)->portal, 0, sizeof(*(conn)) - \
offsetof(struct spdk_iscsi_conn, portal));
static struct spdk_iscsi_conn *g_conns_array = NULL;
static TAILQ_HEAD(, spdk_iscsi_conn) g_free_conns = TAILQ_HEAD_INITIALIZER(g_free_conns);
static TAILQ_HEAD(, spdk_iscsi_conn) g_active_conns = TAILQ_HEAD_INITIALIZER(g_active_conns);
static pthread_mutex_t g_conns_mutex = PTHREAD_MUTEX_INITIALIZER;
static struct spdk_poller *g_shutdown_timer = NULL;
static void iscsi_conn_sock_cb(void *arg, struct spdk_sock_group *group,
struct spdk_sock *sock);
static struct spdk_iscsi_conn *
allocate_conn(void)
{
struct spdk_iscsi_conn *conn;
pthread_mutex_lock(&g_conns_mutex);
conn = TAILQ_FIRST(&g_free_conns);
if (conn != NULL) {
assert(!conn->is_valid);
TAILQ_REMOVE(&g_free_conns, conn, conn_link);
SPDK_ISCSI_CONNECTION_MEMSET(conn);
conn->is_valid = 1;
TAILQ_INSERT_TAIL(&g_active_conns, conn, conn_link);
}
pthread_mutex_unlock(&g_conns_mutex);
return conn;
}
static void
_free_conn(struct spdk_iscsi_conn *conn)
{
TAILQ_REMOVE(&g_active_conns, conn, conn_link);
memset(conn->portal_host, 0, sizeof(conn->portal_host));
memset(conn->portal_port, 0, sizeof(conn->portal_port));
conn->is_valid = 0;
TAILQ_INSERT_TAIL(&g_free_conns, conn, conn_link);
}
static void
free_conn(struct spdk_iscsi_conn *conn)
{
pthread_mutex_lock(&g_conns_mutex);
_free_conn(conn);
pthread_mutex_unlock(&g_conns_mutex);
}
static void
_iscsi_conns_cleanup(void)
{
free(g_conns_array);
}
int initialize_iscsi_conns(void)
{
uint32_t i;
SPDK_DEBUGLOG(iscsi, "spdk_iscsi_init\n");
g_conns_array = calloc(MAX_ISCSI_CONNECTIONS, sizeof(struct spdk_iscsi_conn));
if (g_conns_array == NULL) {
return -ENOMEM;
}
for (i = 0; i < MAX_ISCSI_CONNECTIONS; i++) {
g_conns_array[i].id = i;
TAILQ_INSERT_TAIL(&g_free_conns, &g_conns_array[i], conn_link);
}
return 0;
}
static void
iscsi_poll_group_add_conn(struct spdk_iscsi_poll_group *pg, struct spdk_iscsi_conn *conn)
{
int rc;
rc = spdk_sock_group_add_sock(pg->sock_group, conn->sock, iscsi_conn_sock_cb, conn);
if (rc < 0) {
SPDK_ERRLOG("Failed to add sock=%p of conn=%p\n", conn->sock, conn);
return;
}
conn->is_stopped = false;
STAILQ_INSERT_TAIL(&pg->connections, conn, pg_link);
}
static void
iscsi_poll_group_remove_conn(struct spdk_iscsi_poll_group *pg, struct spdk_iscsi_conn *conn)
{
int rc;
assert(conn->sock != NULL);
rc = spdk_sock_group_remove_sock(pg->sock_group, conn->sock);
if (rc < 0) {
SPDK_ERRLOG("Failed to remove sock=%p of conn=%p\n", conn->sock, conn);
}
conn->is_stopped = true;
STAILQ_REMOVE(&pg->connections, conn, spdk_iscsi_conn, pg_link);
}
static void
iscsi_conn_start(void *ctx)
{
struct spdk_iscsi_conn *conn = ctx;
iscsi_poll_group_add_conn(conn->pg, conn);
}
int
iscsi_conn_construct(struct spdk_iscsi_portal *portal,
struct spdk_sock *sock)
{
struct spdk_iscsi_poll_group *pg;
struct spdk_iscsi_conn *conn;
int i, rc;
conn = allocate_conn();
if (conn == NULL) {
SPDK_ERRLOG("Could not allocate connection.\n");
return -1;
}
pthread_mutex_lock(&g_iscsi.mutex);
conn->timeout = g_iscsi.timeout * spdk_get_ticks_hz(); /* seconds to TSC */
conn->nopininterval = g_iscsi.nopininterval;
conn->nopininterval *= spdk_get_ticks_hz(); /* seconds to TSC */
conn->last_nopin = spdk_get_ticks();
conn->nop_outstanding = false;
conn->data_out_cnt = 0;
conn->data_in_cnt = 0;
conn->disable_chap = portal->group->disable_chap;
conn->require_chap = portal->group->require_chap;
conn->mutual_chap = portal->group->mutual_chap;
conn->chap_group = portal->group->chap_group;
pthread_mutex_unlock(&g_iscsi.mutex);
conn->MaxRecvDataSegmentLength = 8192; /* RFC3720(12.12) */
conn->portal = portal;
conn->pg_tag = portal->group->tag;
memcpy(conn->portal_host, portal->host, strlen(portal->host));
memcpy(conn->portal_port, portal->port, strlen(portal->port));
conn->sock = sock;
conn->state = ISCSI_CONN_STATE_INVALID;
conn->login_phase = ISCSI_SECURITY_NEGOTIATION_PHASE;
conn->ttt = 0;
conn->partial_text_parameter = NULL;
for (i = 0; i < MAX_CONNECTION_PARAMS; i++) {
conn->conn_param_state_negotiated[i] = false;
}
for (i = 0; i < MAX_SESSION_PARAMS; i++) {
conn->sess_param_state_negotiated[i] = false;
}
conn->pdu_recv_state = ISCSI_PDU_RECV_STATE_AWAIT_PDU_READY;
TAILQ_INIT(&conn->write_pdu_list);
TAILQ_INIT(&conn->snack_pdu_list);
TAILQ_INIT(&conn->queued_r2t_tasks);
TAILQ_INIT(&conn->active_r2t_tasks);
TAILQ_INIT(&conn->queued_datain_tasks);
memset(&conn->luns, 0, sizeof(conn->luns));
rc = spdk_sock_getaddr(sock, conn->target_addr, sizeof conn->target_addr, NULL,
conn->initiator_addr, sizeof conn->initiator_addr, NULL);
if (rc < 0) {
SPDK_ERRLOG("spdk_sock_getaddr() failed\n");
goto error_return;
}
/* set low water mark */
rc = spdk_sock_set_recvlowat(conn->sock, 1);
if (rc != 0) {
SPDK_ERRLOG("spdk_sock_set_recvlowat() failed\n");
goto error_return;
}
/* set default params */
rc = iscsi_conn_params_init(&conn->params);
if (rc < 0) {
SPDK_ERRLOG("iscsi_conn_params_init() failed\n");
goto error_return;
}
conn->logout_request_timer = NULL;
conn->logout_timer = NULL;
conn->shutdown_timer = NULL;
SPDK_DEBUGLOG(iscsi, "Launching connection on acceptor thread\n");
conn->pending_task_cnt = 0;
/* Get the first poll group. */
pg = TAILQ_FIRST(&g_iscsi.poll_group_head);
if (pg == NULL) {
SPDK_ERRLOG("There is no poll group.\n");
assert(false);
goto error_return;
}
iscsi: Assign connections to poll groups instead of lcores. This patch binds poll groups to SPDK thread through IO channel and assigns connections to poll groups instead of cores. iSCSI subsystem registers iSCSI global object as an IO device, and create poll groups as context of IO channels of the IO device. Each portal get and hold portal group on which the corresponding acceptor is running. When a connection is constructed, iSCSI subsystem assigns a poll group to the connection by getting it from the corresponding portal. When a connection enters full-feature phase, iSCSI subsystem schedules the connection to a poll group by round-robin. Then, each connection can know its running SPDK thread directly and can use SPDK message passing infrastructure instead of SPDK event framework. By this change, iSCSI connections are binded to SPDK thread, and not binded to processor core anymore. Some other changes in this patch are - core ID is removed from the output of get_iscsi_connections RPC. The upcoming patches will change the RPC to use spdk_for_each_channel and can access SPDK thread safely, and add SPDK thread ID instead. - utilize UT multithread framework added by the last patch to test iSCSI poll groups by UT. Change-Id: Iec73c778aa413bcabdb63141cc41d4160911ea0e Signed-off-by: Ben Walker <benjamin.walker@intel.com> Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/463359 Reviewed-by: Broadcom SPDK FC-NVMe CI <spdk-ci.pdl@broadcom.com> Reviewed-by: Jim Harris <james.r.harris@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2019-07-30 03:00:15 +00:00
conn->pg = pg;
spdk_thread_send_msg(spdk_io_channel_get_thread(spdk_io_channel_from_ctx(pg)),
iscsi_conn_start, conn);
return 0;
error_return:
iscsi_param_free(conn->params);
free_conn(conn);
return -1;
}
void
iscsi_conn_free_pdu(struct spdk_iscsi_conn *conn, struct spdk_iscsi_pdu *pdu)
{
iscsi_conn_xfer_complete_cb cb_fn;
void *cb_arg;
cb_fn = pdu->cb_fn;
cb_arg = pdu->cb_arg;
assert(cb_fn != NULL);
pdu->cb_fn = NULL;
if (pdu->task) {
iscsi_task_put(pdu->task);
}
iscsi_put_pdu(pdu);
cb_fn(cb_arg);
}
static int
iscsi_conn_free_tasks(struct spdk_iscsi_conn *conn)
{
struct spdk_iscsi_pdu *pdu, *tmp_pdu;
struct spdk_iscsi_task *iscsi_task, *tmp_iscsi_task;
TAILQ_FOREACH_SAFE(pdu, &conn->snack_pdu_list, tailq, tmp_pdu) {
TAILQ_REMOVE(&conn->snack_pdu_list, pdu, tailq);
iscsi_conn_free_pdu(conn, pdu);
}
TAILQ_FOREACH_SAFE(iscsi_task, &conn->queued_datain_tasks, link, tmp_iscsi_task) {
if (!iscsi_task->is_queued) {
TAILQ_REMOVE(&conn->queued_datain_tasks, iscsi_task, link);
iscsi_task_put(iscsi_task);
}
}
/* We have to parse conn->write_pdu_list in the end. In iscsi_conn_free_pdu(),
* iscsi_conn_handle_queued_datain_tasks() may be called, and
* iscsi_conn_handle_queued_datain_tasks() will parse conn->queued_datain_tasks
* and may stack some PDUs to conn->write_pdu_list. Hence when we come here, we
* have to ensure there is no associated task in conn->queued_datain_tasks.
*/
TAILQ_FOREACH_SAFE(pdu, &conn->write_pdu_list, tailq, tmp_pdu) {
TAILQ_REMOVE(&conn->write_pdu_list, pdu, tailq);
iscsi_conn_free_pdu(conn, pdu);
}
if (conn->pending_task_cnt) {
return -1;
}
return 0;
}
static void
iscsi_conn_cleanup_backend(struct spdk_iscsi_conn *conn)
{
int rc;
struct spdk_iscsi_tgt_node *target;
if (conn->sess->connections > 1) {
/* connection specific cleanup */
} else if (!g_iscsi.AllowDuplicateIsid) {
/* clean up all tasks to all LUNs for session */
target = conn->sess->target;
if (target != NULL) {
rc = iscsi_tgt_node_cleanup_luns(conn, target);
if (rc < 0) {
SPDK_ERRLOG("target abort failed\n");
}
}
}
}
static void
iscsi_conn_free(struct spdk_iscsi_conn *conn)
{
struct spdk_iscsi_sess *sess;
int idx;
uint32_t i;
pthread_mutex_lock(&g_conns_mutex);
if (conn->sess == NULL) {
goto end;
}
idx = -1;
sess = conn->sess;
conn->sess = NULL;
for (i = 0; i < sess->connections; i++) {
if (sess->conns[i] == conn) {
idx = i;
break;
}
}
if (idx < 0) {
SPDK_ERRLOG("remove conn not found\n");
} else {
for (i = idx; i < sess->connections - 1; i++) {
sess->conns[i] = sess->conns[i + 1];
}
sess->conns[sess->connections - 1] = NULL;
sess->connections--;
if (sess->connections == 0) {
/* cleanup last connection */
SPDK_DEBUGLOG(iscsi,
"cleanup last conn free sess\n");
iscsi_free_sess(sess);
}
}
SPDK_DEBUGLOG(iscsi, "Terminating connections(tsih %d): %d\n",
sess->tsih, sess->connections);
end:
SPDK_DEBUGLOG(iscsi, "cleanup free conn\n");
iscsi_param_free(conn->params);
_free_conn(conn);
pthread_mutex_unlock(&g_conns_mutex);
}
static void
iscsi_conn_close_lun(struct spdk_iscsi_conn *conn, int lun_id)
{
struct spdk_iscsi_lun *iscsi_lun;
iscsi_lun = conn->luns[lun_id];
if (iscsi_lun == NULL) {
return;
}
spdk_scsi_lun_free_io_channel(iscsi_lun->desc);
spdk_scsi_lun_close(iscsi_lun->desc);
spdk_poller_unregister(&iscsi_lun->remove_poller);
free(iscsi_lun);
conn->luns[lun_id] = NULL;
}
static void
iscsi_conn_close_luns(struct spdk_iscsi_conn *conn)
{
int i;
for (i = 0; i < SPDK_SCSI_DEV_MAX_LUN; i++) {
iscsi_conn_close_lun(conn, i);
}
}
static bool
iscsi_conn_check_tasks_for_lun(struct spdk_iscsi_conn *conn,
struct spdk_scsi_lun *lun)
{
struct spdk_iscsi_pdu *pdu, *tmp_pdu;
struct spdk_iscsi_task *task;
assert(lun != NULL);
/* We can remove deferred PDUs safely because they are already flushed. */
TAILQ_FOREACH_SAFE(pdu, &conn->snack_pdu_list, tailq, tmp_pdu) {
if (lun == pdu->task->scsi.lun) {
TAILQ_REMOVE(&conn->snack_pdu_list, pdu, tailq);
iscsi_conn_free_pdu(conn, pdu);
}
}
TAILQ_FOREACH(task, &conn->queued_datain_tasks, link) {
if (lun == task->scsi.lun) {
return false;
}
}
/* This check loop works even when connection exits in the middle of LUN hotplug
* because all PDUs in write_pdu_list are removed in iscsi_conn_free_tasks().
*/
TAILQ_FOREACH(pdu, &conn->write_pdu_list, tailq) {
if (pdu->task && lun == pdu->task->scsi.lun) {
return false;
}
}
return true;
}
static int
iscsi_conn_remove_lun(void *ctx)
{
struct spdk_iscsi_lun *iscsi_lun = ctx;
struct spdk_iscsi_conn *conn = iscsi_lun->conn;
struct spdk_scsi_lun *lun = iscsi_lun->lun;
int lun_id = spdk_scsi_lun_get_id(lun);
if (!iscsi_conn_check_tasks_for_lun(conn, lun)) {
return SPDK_POLLER_BUSY;
}
iscsi_conn_close_lun(conn, lun_id);
return SPDK_POLLER_BUSY;
}
static void
_iscsi_conn_hotremove_lun(void *ctx)
{
struct spdk_iscsi_lun *iscsi_lun = ctx;
struct spdk_iscsi_conn *conn = iscsi_lun->conn;
struct spdk_scsi_lun *lun = iscsi_lun->lun;
iscsi: Assign connections to poll groups instead of lcores. This patch binds poll groups to SPDK thread through IO channel and assigns connections to poll groups instead of cores. iSCSI subsystem registers iSCSI global object as an IO device, and create poll groups as context of IO channels of the IO device. Each portal get and hold portal group on which the corresponding acceptor is running. When a connection is constructed, iSCSI subsystem assigns a poll group to the connection by getting it from the corresponding portal. When a connection enters full-feature phase, iSCSI subsystem schedules the connection to a poll group by round-robin. Then, each connection can know its running SPDK thread directly and can use SPDK message passing infrastructure instead of SPDK event framework. By this change, iSCSI connections are binded to SPDK thread, and not binded to processor core anymore. Some other changes in this patch are - core ID is removed from the output of get_iscsi_connections RPC. The upcoming patches will change the RPC to use spdk_for_each_channel and can access SPDK thread safely, and add SPDK thread ID instead. - utilize UT multithread framework added by the last patch to test iSCSI poll groups by UT. Change-Id: Iec73c778aa413bcabdb63141cc41d4160911ea0e Signed-off-by: Ben Walker <benjamin.walker@intel.com> Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/463359 Reviewed-by: Broadcom SPDK FC-NVMe CI <spdk-ci.pdl@broadcom.com> Reviewed-by: Jim Harris <james.r.harris@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2019-07-30 03:00:15 +00:00
assert(spdk_io_channel_get_thread(spdk_io_channel_from_ctx(conn->pg)) ==
spdk_get_thread());
/* If a connection is already in stating status, just return */
if (conn->state >= ISCSI_CONN_STATE_EXITING) {
return;
}
iscsi_clear_all_transfer_task(conn, lun, NULL);
iscsi_lun->remove_poller = SPDK_POLLER_REGISTER(iscsi_conn_remove_lun, iscsi_lun,
1000);
}
static void
iscsi_conn_hotremove_lun(struct spdk_scsi_lun *lun, void *remove_ctx)
{
struct spdk_iscsi_conn *conn = remove_ctx;
int lun_id = spdk_scsi_lun_get_id(lun);
struct spdk_iscsi_lun *iscsi_lun;
iscsi_lun = conn->luns[lun_id];
if (iscsi_lun == NULL) {
SPDK_ERRLOG("LUN hotplug was notified to the unallocated LUN %d.\n", lun_id);
return;
}
spdk_thread_send_msg(spdk_io_channel_get_thread(spdk_io_channel_from_ctx(conn->pg)),
_iscsi_conn_hotremove_lun, iscsi_lun);
}
static int
iscsi_conn_open_lun(struct spdk_iscsi_conn *conn, int lun_id,
struct spdk_scsi_lun *lun)
{
int rc;
struct spdk_iscsi_lun *iscsi_lun;
iscsi_lun = calloc(1, sizeof(*iscsi_lun));
if (iscsi_lun == NULL) {
return -ENOMEM;
}
iscsi_lun->conn = conn;
iscsi_lun->lun = lun;
rc = spdk_scsi_lun_open(lun, iscsi_conn_hotremove_lun, conn, &iscsi_lun->desc);
if (rc != 0) {
free(iscsi_lun);
return rc;
}
rc = spdk_scsi_lun_allocate_io_channel(iscsi_lun->desc);
if (rc != 0) {
spdk_scsi_lun_close(iscsi_lun->desc);
free(iscsi_lun);
return rc;
}
conn->luns[lun_id] = iscsi_lun;
return 0;
}
static void
iscsi_conn_open_luns(struct spdk_iscsi_conn *conn)
{
int i, rc;
struct spdk_scsi_lun *lun;
for (i = 0; i < SPDK_SCSI_DEV_MAX_LUN; i++) {
lun = spdk_scsi_dev_get_lun(conn->dev, i);
if (lun == NULL) {
continue;
}
rc = iscsi_conn_open_lun(conn, i, lun);
if (rc != 0) {
goto error;
}
}
return;
error:
iscsi_conn_close_luns(conn);
}
/**
* This function will stop executing the specified connection.
*/
static void
iscsi_conn_stop(struct spdk_iscsi_conn *conn)
{
struct spdk_iscsi_tgt_node *target;
assert(conn->state == ISCSI_CONN_STATE_EXITED);
assert(conn->data_in_cnt == 0);
assert(conn->data_out_cnt == 0);
if (conn->sess != NULL &&
conn->sess->session_type == SESSION_TYPE_NORMAL &&
conn->full_feature) {
target = conn->sess->target;
pthread_mutex_lock(&target->mutex);
target->num_active_conns--;
pthread_mutex_unlock(&target->mutex);
iscsi_conn_close_luns(conn);
}
iscsi: Assign connections to poll groups instead of lcores. This patch binds poll groups to SPDK thread through IO channel and assigns connections to poll groups instead of cores. iSCSI subsystem registers iSCSI global object as an IO device, and create poll groups as context of IO channels of the IO device. Each portal get and hold portal group on which the corresponding acceptor is running. When a connection is constructed, iSCSI subsystem assigns a poll group to the connection by getting it from the corresponding portal. When a connection enters full-feature phase, iSCSI subsystem schedules the connection to a poll group by round-robin. Then, each connection can know its running SPDK thread directly and can use SPDK message passing infrastructure instead of SPDK event framework. By this change, iSCSI connections are binded to SPDK thread, and not binded to processor core anymore. Some other changes in this patch are - core ID is removed from the output of get_iscsi_connections RPC. The upcoming patches will change the RPC to use spdk_for_each_channel and can access SPDK thread safely, and add SPDK thread ID instead. - utilize UT multithread framework added by the last patch to test iSCSI poll groups by UT. Change-Id: Iec73c778aa413bcabdb63141cc41d4160911ea0e Signed-off-by: Ben Walker <benjamin.walker@intel.com> Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/463359 Reviewed-by: Broadcom SPDK FC-NVMe CI <spdk-ci.pdl@broadcom.com> Reviewed-by: Jim Harris <james.r.harris@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2019-07-30 03:00:15 +00:00
assert(spdk_io_channel_get_thread(spdk_io_channel_from_ctx(conn->pg)) ==
spdk_get_thread());
}
static int
_iscsi_conn_check_shutdown(void *arg)
{
struct spdk_iscsi_conn *conn = arg;
int rc;
rc = iscsi_conn_free_tasks(conn);
if (rc < 0) {
return SPDK_POLLER_BUSY;
}
spdk_poller_unregister(&conn->shutdown_timer);
iscsi_conn_stop(conn);
iscsi_conn_free(conn);
return SPDK_POLLER_BUSY;
}
static void
_iscsi_conn_destruct(struct spdk_iscsi_conn *conn)
{
int rc;
iscsi_poll_group_remove_conn(conn->pg, conn);
spdk_sock_close(&conn->sock);
iscsi_clear_all_transfer_task(conn, NULL, NULL);
spdk_poller_unregister(&conn->logout_request_timer);
spdk_poller_unregister(&conn->logout_timer);
rc = iscsi_conn_free_tasks(conn);
if (rc < 0) {
/* The connection cannot be freed yet. Check back later. */
conn->shutdown_timer = SPDK_POLLER_REGISTER(_iscsi_conn_check_shutdown, conn, 1000);
} else {
iscsi_conn_stop(conn);
iscsi_conn_free(conn);
}
}
static int
_iscsi_conn_check_pending_tasks(void *arg)
{
struct spdk_iscsi_conn *conn = arg;
if (conn->dev != NULL &&
spdk_scsi_dev_has_pending_tasks(conn->dev, conn->initiator_port)) {
return SPDK_POLLER_BUSY;
}
spdk_poller_unregister(&conn->shutdown_timer);
_iscsi_conn_destruct(conn);
return SPDK_POLLER_BUSY;
}
void
iscsi_conn_destruct(struct spdk_iscsi_conn *conn)
{
2019-10-31 03:00:44 +00:00
struct spdk_iscsi_pdu *pdu;
struct spdk_iscsi_task *task;
int opcode;
/* If a connection is already in exited status, just return */
if (conn->state >= ISCSI_CONN_STATE_EXITED) {
return;
}
conn->state = ISCSI_CONN_STATE_EXITED;
2019-10-31 03:00:44 +00:00
/*
* Each connection pre-allocates its next PDU - make sure these get
* freed here.
*/
pdu = conn->pdu_in_progress;
if (pdu) {
/* remove the task left in the PDU too. */
task = pdu->task;
if (task) {
opcode = pdu->bhs.opcode;
switch (opcode) {
case ISCSI_OP_SCSI:
case ISCSI_OP_SCSI_DATAOUT:
spdk_scsi_task_process_abort(&task->scsi);
iscsi_task_cpl(&task->scsi);
2019-10-31 03:00:44 +00:00
break;
default:
SPDK_ERRLOG("unexpected opcode %x\n", opcode);
iscsi_task_put(task);
2019-10-31 03:00:44 +00:00
break;
}
}
iscsi_put_pdu(pdu);
2019-10-31 03:00:44 +00:00
conn->pdu_in_progress = NULL;
}
if (conn->sess != NULL && conn->pending_task_cnt > 0) {
iscsi_conn_cleanup_backend(conn);
}
if (conn->dev != NULL &&
spdk_scsi_dev_has_pending_tasks(conn->dev, conn->initiator_port)) {
conn->shutdown_timer = SPDK_POLLER_REGISTER(_iscsi_conn_check_pending_tasks, conn, 1000);
} else {
_iscsi_conn_destruct(conn);
}
}
int
iscsi_get_active_conns(struct spdk_iscsi_tgt_node *target)
{
struct spdk_iscsi_conn *conn;
int num = 0;
if (g_conns_array == MAP_FAILED) {
return 0;
}
pthread_mutex_lock(&g_conns_mutex);
TAILQ_FOREACH(conn, &g_active_conns, conn_link) {
if (target == NULL || conn->target == target) {
num++;
}
}
pthread_mutex_unlock(&g_conns_mutex);
return num;
}
static void
iscsi_conn_check_shutdown_cb(void *arg1)
{
_iscsi_conns_cleanup();
shutdown_iscsi_conns_done();
}
static int
iscsi_conn_check_shutdown(void *arg)
{
if (iscsi_get_active_conns(NULL) != 0) {
return SPDK_POLLER_BUSY;
}
spdk_poller_unregister(&g_shutdown_timer);
spdk_thread_send_msg(spdk_get_thread(), iscsi_conn_check_shutdown_cb, NULL);
return SPDK_POLLER_BUSY;
}
static void
iscsi_send_logout_request(struct spdk_iscsi_conn *conn)
{
struct spdk_iscsi_pdu *rsp_pdu;
struct iscsi_bhs_async *rsph;
rsp_pdu = iscsi_get_pdu(conn);
assert(rsp_pdu != NULL);
rsph = (struct iscsi_bhs_async *)&rsp_pdu->bhs;
rsp_pdu->data = NULL;
rsph->opcode = ISCSI_OP_ASYNC;
to_be32(&rsph->ffffffff, 0xFFFFFFFF);
rsph->async_event = 1;
to_be16(&rsph->param3, ISCSI_LOGOUT_REQUEST_TIMEOUT);
to_be32(&rsph->stat_sn, conn->StatSN);
conn->StatSN++;
to_be32(&rsph->exp_cmd_sn, conn->sess->ExpCmdSN);
to_be32(&rsph->max_cmd_sn, conn->sess->MaxCmdSN);
iscsi_conn_write_pdu(conn, rsp_pdu, iscsi_conn_pdu_generic_complete, NULL);
}
static int
logout_request_timeout(void *arg)
{
struct spdk_iscsi_conn *conn = arg;
if (conn->state < ISCSI_CONN_STATE_EXITING) {
conn->state = ISCSI_CONN_STATE_EXITING;
}
return SPDK_POLLER_BUSY;
}
/* If the connection is running and logout is not requested yet, request logout
* to initiator and wait for the logout process to start.
*/
static void
_iscsi_conn_request_logout(void *ctx)
{
struct spdk_iscsi_conn *conn = ctx;
if (conn->state > ISCSI_CONN_STATE_RUNNING ||
conn->logout_request_timer != NULL) {
return;
}
iscsi_send_logout_request(conn);
conn->logout_request_timer = SPDK_POLLER_REGISTER(logout_request_timeout,
conn, ISCSI_LOGOUT_REQUEST_TIMEOUT * 1000000);
}
static void
iscsi_conn_request_logout(struct spdk_iscsi_conn *conn)
{
struct spdk_thread *thread;
if (conn->state == ISCSI_CONN_STATE_INVALID) {
/* Move it to EXITING state if the connection is in login. */
conn->state = ISCSI_CONN_STATE_EXITING;
} else if (conn->state == ISCSI_CONN_STATE_RUNNING &&
conn->logout_request_timer == NULL) {
thread = spdk_io_channel_get_thread(spdk_io_channel_from_ctx(conn->pg));
spdk_thread_send_msg(thread, _iscsi_conn_request_logout, conn);
}
}
void
iscsi_conns_request_logout(struct spdk_iscsi_tgt_node *target, int pg_tag)
{
struct spdk_iscsi_conn *conn;
if (g_conns_array == MAP_FAILED) {
return;
}
pthread_mutex_lock(&g_conns_mutex);
TAILQ_FOREACH(conn, &g_active_conns, conn_link) {
if ((target == NULL) ||
(conn->target == target && (pg_tag < 0 || conn->pg_tag == pg_tag))) {
iscsi_conn_request_logout(conn);
}
}
pthread_mutex_unlock(&g_conns_mutex);
}
void
shutdown_iscsi_conns(void)
{
iscsi_conns_request_logout(NULL, -1);
g_shutdown_timer = SPDK_POLLER_REGISTER(iscsi_conn_check_shutdown, NULL, 1000);
}
/* Do not set conn->state if the connection has already started exiting.
* This ensures we do not move a connection from EXITED state back to EXITING.
*/
static void
_iscsi_conn_drop(void *ctx)
{
struct spdk_iscsi_conn *conn = ctx;
if (conn->state < ISCSI_CONN_STATE_EXITING) {
conn->state = ISCSI_CONN_STATE_EXITING;
}
}
int
iscsi_drop_conns(struct spdk_iscsi_conn *conn, const char *conn_match,
int drop_all)
{
struct spdk_iscsi_conn *xconn;
const char *xconn_match;
struct spdk_thread *thread;
int num;
SPDK_DEBUGLOG(iscsi, "iscsi_drop_conns\n");
num = 0;
pthread_mutex_lock(&g_conns_mutex);
if (g_conns_array == MAP_FAILED) {
goto exit;
}
TAILQ_FOREACH(xconn, &g_active_conns, conn_link) {
if (xconn == conn) {
continue;
}
if (!drop_all && xconn->initiator_port == NULL) {
continue;
}
xconn_match =
drop_all ? xconn->initiator_name : spdk_scsi_port_get_name(xconn->initiator_port);
if (!strcasecmp(conn_match, xconn_match) &&
conn->target == xconn->target) {
if (num == 0) {
/*
* Only print this message before we report the
* first dropped connection.
*/
SPDK_ERRLOG("drop old connections %s by %s\n",
conn->target->name, conn_match);
}
SPDK_ERRLOG("exiting conn by %s (%s)\n",
xconn_match, xconn->initiator_addr);
if (xconn->sess != NULL) {
SPDK_DEBUGLOG(iscsi, "TSIH=%u\n", xconn->sess->tsih);
} else {
SPDK_DEBUGLOG(iscsi, "TSIH=xx\n");
}
SPDK_DEBUGLOG(iscsi, "CID=%u\n", xconn->cid);
thread = spdk_io_channel_get_thread(spdk_io_channel_from_ctx(xconn->pg));
spdk_thread_send_msg(thread, _iscsi_conn_drop, xconn);
num++;
}
}
exit:
pthread_mutex_unlock(&g_conns_mutex);
if (num != 0) {
SPDK_ERRLOG("exiting %d conns\n", num);
}
return 0;
}
static int
_iscsi_conn_abort_queued_datain_task(struct spdk_iscsi_conn *conn,
struct spdk_iscsi_task *task)
{
struct spdk_iscsi_task *subtask;
uint32_t remaining_size;
lib/iscsi: Make the max number of read subtasks for large read I/O configurable For some use case that there is heavy large read I/O, the performance bottleneck due to MAX_LARGE_DATAIN_PER_CONNECTION was reported. The following assumes that all I/Os are large read. Large read primary task whose I/O size is more than SPDK_BDEV_LARGE_BUF_MAX_SIZE (=64KB) is split into multiple read subtasks. spdk_iscsi_globals::MaxQueueDepth limits maximum number of outstanding read primary tasks, and MAX_LARGE_DATAIN_PER_CONNECTION (=64) limits maximum number of outstanding read subtasks. MAX_LARGE_DATAIN_PER_CONNECTION is also used to calculate PDU pool. To remove the performance bottleneck, change the macro constant MAX_LARGE_DATAIN_PER_CONNECTION to a global variable spdk_iscsi_globals::MaxLargeDataInPerConnection. We don't see any negative side effect if we set spdk_iscsi_globals::MaxLargeDataInPerConnection to 64. The use case that reported the performance issue will change the value of spdk_iscsi_globals::MaxLargeDataInPerConnection by its own responsibility. The next patch will add the value of spdk_iscsi_globals::MaxLargeDataInPerConnection to iSCSI options, and make it configurable by JSON RPC. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: Ifc30cdb8e00d50f4d3755ff399263cf5d0b681b6 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/3755 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Changpeng Liu <changpeng.liu@intel.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2020-08-11 15:29:25 +00:00
if (conn->data_in_cnt >= g_iscsi.MaxLargeDataInPerConnection) {
return -1;
}
assert(task->current_datain_offset <= task->scsi.transfer_len);
/* Stop split and abort read I/O for remaining data. */
if (task->current_datain_offset < task->scsi.transfer_len) {
remaining_size = task->scsi.transfer_len - task->current_datain_offset;
subtask = iscsi_task_get(conn, task, iscsi_task_cpl);
assert(subtask != NULL);
subtask->scsi.offset = task->current_datain_offset;
subtask->scsi.length = remaining_size;
spdk_scsi_task_set_data(&subtask->scsi, NULL, 0);
task->current_datain_offset += subtask->scsi.length;
subtask->scsi.transfer_len = subtask->scsi.length;
spdk_scsi_task_process_abort(&subtask->scsi);
iscsi_task_cpl(&subtask->scsi);
}
/* Remove the primary task from the list because all subtasks are submitted
* or aborted.
*/
assert(task->current_datain_offset == task->scsi.transfer_len);
TAILQ_REMOVE(&conn->queued_datain_tasks, task, link);
return 0;
}
int
iscsi_conn_abort_queued_datain_task(struct spdk_iscsi_conn *conn,
uint32_t ref_task_tag)
{
struct spdk_iscsi_task *task;
TAILQ_FOREACH(task, &conn->queued_datain_tasks, link) {
if (task->tag == ref_task_tag) {
return _iscsi_conn_abort_queued_datain_task(conn, task);
}
}
return 0;
}
int
iscsi_conn_abort_queued_datain_tasks(struct spdk_iscsi_conn *conn,
struct spdk_scsi_lun *lun,
struct spdk_iscsi_pdu *pdu)
{
struct spdk_iscsi_task *task, *task_tmp;
struct spdk_iscsi_pdu *pdu_tmp;
int rc;
TAILQ_FOREACH_SAFE(task, &conn->queued_datain_tasks, link, task_tmp) {
pdu_tmp = iscsi_task_get_pdu(task);
if ((lun == NULL || lun == task->scsi.lun) &&
(pdu == NULL || (spdk_sn32_lt(pdu_tmp->cmd_sn, pdu->cmd_sn)))) {
rc = _iscsi_conn_abort_queued_datain_task(conn, task);
if (rc != 0) {
return rc;
}
}
}
return 0;
}
int
iscsi_conn_handle_queued_datain_tasks(struct spdk_iscsi_conn *conn)
{
struct spdk_iscsi_task *task;
while (!TAILQ_EMPTY(&conn->queued_datain_tasks) &&
lib/iscsi: Make the max number of read subtasks for large read I/O configurable For some use case that there is heavy large read I/O, the performance bottleneck due to MAX_LARGE_DATAIN_PER_CONNECTION was reported. The following assumes that all I/Os are large read. Large read primary task whose I/O size is more than SPDK_BDEV_LARGE_BUF_MAX_SIZE (=64KB) is split into multiple read subtasks. spdk_iscsi_globals::MaxQueueDepth limits maximum number of outstanding read primary tasks, and MAX_LARGE_DATAIN_PER_CONNECTION (=64) limits maximum number of outstanding read subtasks. MAX_LARGE_DATAIN_PER_CONNECTION is also used to calculate PDU pool. To remove the performance bottleneck, change the macro constant MAX_LARGE_DATAIN_PER_CONNECTION to a global variable spdk_iscsi_globals::MaxLargeDataInPerConnection. We don't see any negative side effect if we set spdk_iscsi_globals::MaxLargeDataInPerConnection to 64. The use case that reported the performance issue will change the value of spdk_iscsi_globals::MaxLargeDataInPerConnection by its own responsibility. The next patch will add the value of spdk_iscsi_globals::MaxLargeDataInPerConnection to iSCSI options, and make it configurable by JSON RPC. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: Ifc30cdb8e00d50f4d3755ff399263cf5d0b681b6 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/3755 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Changpeng Liu <changpeng.liu@intel.com> Reviewed-by: Jim Harris <james.r.harris@intel.com>
2020-08-11 15:29:25 +00:00
conn->data_in_cnt < g_iscsi.MaxLargeDataInPerConnection) {
task = TAILQ_FIRST(&conn->queued_datain_tasks);
assert(task->current_datain_offset <= task->scsi.transfer_len);
if (task->current_datain_offset < task->scsi.transfer_len) {
struct spdk_iscsi_task *subtask;
uint32_t remaining_size = 0;
remaining_size = task->scsi.transfer_len - task->current_datain_offset;
subtask = iscsi_task_get(conn, task, iscsi_task_cpl);
assert(subtask != NULL);
subtask->scsi.offset = task->current_datain_offset;
spdk_scsi_task_set_data(&subtask->scsi, NULL, 0);
if (spdk_scsi_dev_get_lun(conn->dev, task->lun_id) == NULL) {
/* Stop submitting split read I/Os for remaining data. */
TAILQ_REMOVE(&conn->queued_datain_tasks, task, link);
task->current_datain_offset += remaining_size;
assert(task->current_datain_offset == task->scsi.transfer_len);
subtask->scsi.transfer_len = remaining_size;
spdk_scsi_task_process_null_lun(&subtask->scsi);
iscsi_task_cpl(&subtask->scsi);
return 0;
}
subtask->scsi.length = spdk_min(SPDK_BDEV_LARGE_BUF_MAX_SIZE, remaining_size);
task->current_datain_offset += subtask->scsi.length;
iscsi_queue_task(conn, subtask);
}
if (task->current_datain_offset == task->scsi.transfer_len) {
TAILQ_REMOVE(&conn->queued_datain_tasks, task, link);
}
}
return 0;
}
void
iscsi_task_mgmt_cpl(struct spdk_scsi_task *scsi_task)
{
struct spdk_iscsi_task *task = iscsi_task_from_scsi_task(scsi_task);
iscsi_task_mgmt_response(task->conn, task);
iscsi_task_put(task);
}
static void
iscsi_task_copy_to_rsp_scsi_status(struct spdk_iscsi_task *primary,
struct spdk_scsi_task *task)
{
memcpy(primary->rsp_sense_data, task->sense_data, task->sense_data_len);
primary->rsp_sense_data_len = task->sense_data_len;
primary->rsp_scsi_status = task->status;
}
static void
iscsi_task_copy_from_rsp_scsi_status(struct spdk_scsi_task *task,
struct spdk_iscsi_task *primary)
{
memcpy(task->sense_data, primary->rsp_sense_data,
primary->rsp_sense_data_len);
task->sense_data_len = primary->rsp_sense_data_len;
task->status = primary->rsp_scsi_status;
}
static void
process_completed_read_subtask_list(struct spdk_iscsi_conn *conn,
struct spdk_iscsi_task *primary)
{
struct spdk_iscsi_task *subtask, *tmp;
TAILQ_FOREACH_SAFE(subtask, &primary->subtask_list, subtask_link, tmp) {
if (subtask->scsi.offset == primary->bytes_completed) {
TAILQ_REMOVE(&primary->subtask_list, subtask, subtask_link);
primary->bytes_completed += subtask->scsi.length;
if (primary->bytes_completed == primary->scsi.transfer_len) {
iscsi_task_put(primary);
}
iscsi_task_response(conn, subtask);
iscsi_task_put(subtask);
} else {
break;
}
}
}
static void
process_read_task_completion(struct spdk_iscsi_conn *conn,
struct spdk_iscsi_task *task,
struct spdk_iscsi_task *primary)
{
struct spdk_iscsi_task *tmp;
/* If the status of the completed subtask is the first failure,
* copy it to out-of-order subtasks and remember it as the status
* of the command,
*
* Even if the status of the completed task is success,
* there are any failed subtask ever, copy the first failed status
* to it.
*/
if (task->scsi.status != SPDK_SCSI_STATUS_GOOD) {
if (primary->rsp_scsi_status == SPDK_SCSI_STATUS_GOOD) {
TAILQ_FOREACH(tmp, &primary->subtask_list, subtask_link) {
spdk_scsi_task_copy_status(&tmp->scsi, &task->scsi);
}
iscsi_task_copy_to_rsp_scsi_status(primary, &task->scsi);
}
} else if (primary->rsp_scsi_status != SPDK_SCSI_STATUS_GOOD) {
iscsi_task_copy_from_rsp_scsi_status(&task->scsi, primary);
}
if (task == primary) {
primary->bytes_completed = task->scsi.length;
/* For non split read I/O */
assert(primary->bytes_completed == task->scsi.transfer_len);
iscsi_task_response(conn, task);
iscsi_task_put(task);
} else {
if (task->scsi.offset != primary->bytes_completed) {
TAILQ_FOREACH(tmp, &primary->subtask_list, subtask_link) {
if (task->scsi.offset < tmp->scsi.offset) {
TAILQ_INSERT_BEFORE(tmp, task, subtask_link);
return;
}
}
TAILQ_INSERT_TAIL(&primary->subtask_list, task, subtask_link);
} else {
TAILQ_INSERT_HEAD(&primary->subtask_list, task, subtask_link);
process_completed_read_subtask_list(conn, primary);
}
}
}
static void
process_non_read_task_completion(struct spdk_iscsi_conn *conn,
struct spdk_iscsi_task *task,
struct spdk_iscsi_task *primary)
{
primary->bytes_completed += task->scsi.length;
/* If the status of the subtask is the first failure, remember it as
* the status of the command and set it to the status of the primary
* task later.
*
* If the first failed task is the primary, two copies can be avoided
* but code simplicity is prioritized.
*/
if (task->scsi.status == SPDK_SCSI_STATUS_GOOD) {
if (task != primary) {
primary->scsi.data_transferred += task->scsi.data_transferred;
}
} else if (primary->rsp_scsi_status == SPDK_SCSI_STATUS_GOOD) {
iscsi_task_copy_to_rsp_scsi_status(primary, &task->scsi);
}
if (primary->bytes_completed == primary->scsi.transfer_len) {
/*
* Check if this is the last task completed for an iSCSI write
* that required child subtasks. If task != primary, we know
* for sure that it was part of an iSCSI write with child subtasks.
* The trickier case is when the last task completed was the initial
* task - in this case the task will have a smaller length than
* the overall transfer length.
*/
if (task != primary || task->scsi.length != task->scsi.transfer_len) {
/* If LUN is removed in the middle of the iSCSI write sequence,
* primary might complete the write to the initiator because it is not
* ensured that the initiator will send all data requested by R2Ts.
*
* We check it and skip the following if primary is completed. (see
* iscsi_clear_all_transfer_task() in iscsi.c.)
*/
if (primary->is_r2t_active) {
if (primary->rsp_scsi_status != SPDK_SCSI_STATUS_GOOD) {
iscsi_task_copy_from_rsp_scsi_status(&primary->scsi, primary);
}
iscsi_task_response(conn, primary);
iscsi_del_transfer_task(conn, primary->tag);
}
} else {
iscsi_task_response(conn, task);
}
}
iscsi_task_put(task);
}
void
iscsi_task_cpl(struct spdk_scsi_task *scsi_task)
{
struct spdk_iscsi_task *primary;
struct spdk_iscsi_task *task = iscsi_task_from_scsi_task(scsi_task);
struct spdk_iscsi_conn *conn = task->conn;
struct spdk_iscsi_pdu *pdu = task->pdu;
spdk_trace_record(TRACE_ISCSI_TASK_DONE, conn->id, 0, (uintptr_t)task, 0);
task->is_queued = false;
primary = iscsi_task_get_primary(task);
if (iscsi_task_is_read(primary)) {
process_read_task_completion(conn, task, primary);
} else {
process_non_read_task_completion(conn, task, primary);
}
if (!task->parent) {
spdk_trace_record(TRACE_ISCSI_PDU_COMPLETED, 0, 0, (uintptr_t)pdu, 0);
}
}
static void
iscsi_conn_send_nopin(struct spdk_iscsi_conn *conn)
{
struct spdk_iscsi_pdu *rsp_pdu;
struct iscsi_bhs_nop_in *rsp;
/* Only send nopin if we have logged in and are in a normal session. */
if (conn->sess == NULL ||
!conn->full_feature ||
!iscsi_param_eq_val(conn->sess->params, "SessionType", "Normal")) {
return;
}
SPDK_DEBUGLOG(iscsi, "send NOPIN isid=%"PRIx64", tsih=%u, cid=%u\n",
conn->sess->isid, conn->sess->tsih, conn->cid);
SPDK_DEBUGLOG(iscsi, "StatSN=%u, ExpCmdSN=%u, MaxCmdSN=%u\n",
conn->StatSN, conn->sess->ExpCmdSN,
conn->sess->MaxCmdSN);
rsp_pdu = iscsi_get_pdu(conn);
rsp = (struct iscsi_bhs_nop_in *) &rsp_pdu->bhs;
rsp_pdu->data = NULL;
/*
* iscsi_get_pdu() memset's the PDU for us, so only fill out the needed
* fields.
*/
rsp->opcode = ISCSI_OP_NOPIN;
rsp->flags = 0x80;
/*
* Technically the to_be32() is not needed here, since
* to_be32(0xFFFFFFFU) returns 0xFFFFFFFFU.
*/
to_be32(&rsp->itt, 0xFFFFFFFFU);
to_be32(&rsp->ttt, conn->id);
to_be32(&rsp->stat_sn, conn->StatSN);
to_be32(&rsp->exp_cmd_sn, conn->sess->ExpCmdSN);
to_be32(&rsp->max_cmd_sn, conn->sess->MaxCmdSN);
iscsi_conn_write_pdu(conn, rsp_pdu, iscsi_conn_pdu_generic_complete, NULL);
conn->last_nopin = spdk_get_ticks();
conn->nop_outstanding = true;
}
void
iscsi_conn_handle_nop(struct spdk_iscsi_conn *conn)
{
uint64_t tsc;
/**
* This function will be executed by nop_poller of iSCSI polling group, so
* we need to check the connection state first, then do the nop interval
* expiration check work.
*/
if ((conn->state == ISCSI_CONN_STATE_EXITED) ||
(conn->state == ISCSI_CONN_STATE_EXITING)) {
return;
}
/* Check for nop interval expiration */
tsc = spdk_get_ticks();
if (conn->nop_outstanding) {
if ((tsc - conn->last_nopin) > conn->timeout) {
SPDK_ERRLOG("Timed out waiting for NOP-Out response from initiator\n");
SPDK_ERRLOG(" tsc=0x%lx, last_nopin=0x%lx\n", tsc, conn->last_nopin);
SPDK_ERRLOG(" initiator=%s, target=%s\n", conn->initiator_name,
conn->target_short_name);
conn->state = ISCSI_CONN_STATE_EXITING;
}
} else if (tsc - conn->last_nopin > conn->nopininterval) {
iscsi_conn_send_nopin(conn);
}
}
/**
* \brief Reads data for the specified iSCSI connection from its TCP socket.
*
* The TCP socket is marked as non-blocking, so this function may not read
* all data requested.
*
* Returns SPDK_ISCSI_CONNECTION_FATAL if the recv() operation indicates a fatal
* error with the TCP connection (including if the TCP connection was closed
* unexpectedly.
*
* Otherwise returns the number of bytes successfully read.
*/
int
iscsi_conn_read_data(struct spdk_iscsi_conn *conn, int bytes,
void *buf)
{
int ret;
if (bytes == 0) {
return 0;
}
ret = spdk_sock_recv(conn->sock, buf, bytes);
if (ret > 0) {
spdk_trace_record(TRACE_ISCSI_READ_FROM_SOCKET_DONE, conn->id, ret, 0, 0);
return ret;
}
if (ret < 0) {
if (errno == EAGAIN || errno == EWOULDBLOCK) {
return 0;
}
/* For connect reset issue, do not output error log */
if (errno == ECONNRESET) {
SPDK_DEBUGLOG(iscsi, "spdk_sock_recv() failed, errno %d: %s\n",
errno, spdk_strerror(errno));
} else {
SPDK_ERRLOG("spdk_sock_recv() failed, errno %d: %s\n",
errno, spdk_strerror(errno));
}
}
/* connection closed */
return SPDK_ISCSI_CONNECTION_FATAL;
}
int
iscsi_conn_readv_data(struct spdk_iscsi_conn *conn,
struct iovec *iov, int iovcnt)
{
int ret;
if (iov == NULL || iovcnt == 0) {
return 0;
}
if (iovcnt == 1) {
return iscsi_conn_read_data(conn, iov[0].iov_len,
iov[0].iov_base);
}
ret = spdk_sock_readv(conn->sock, iov, iovcnt);
if (ret > 0) {
spdk_trace_record(TRACE_ISCSI_READ_FROM_SOCKET_DONE, conn->id, ret, 0, 0);
return ret;
}
if (ret < 0) {
if (errno == EAGAIN || errno == EWOULDBLOCK) {
return 0;
}
/* For connect reset issue, do not output error log */
if (errno == ECONNRESET) {
SPDK_DEBUGLOG(iscsi, "spdk_sock_readv() failed, errno %d: %s\n",
errno, spdk_strerror(errno));
} else {
SPDK_ERRLOG("spdk_sock_readv() failed, errno %d: %s\n",
errno, spdk_strerror(errno));
}
}
/* connection closed */
return SPDK_ISCSI_CONNECTION_FATAL;
}
static bool
iscsi_is_free_pdu_deferred(struct spdk_iscsi_pdu *pdu)
{
if (pdu == NULL) {
return false;
}
if (pdu->bhs.opcode == ISCSI_OP_R2T ||
pdu->bhs.opcode == ISCSI_OP_SCSI_DATAIN) {
return true;
}
return false;
}
static int
iscsi_dif_verify(struct spdk_iscsi_pdu *pdu, struct spdk_dif_ctx *dif_ctx)
{
struct iovec iov;
struct spdk_dif_error err_blk = {};
uint32_t num_blocks;
int rc;
iov.iov_base = pdu->data;
iov.iov_len = pdu->data_buf_len;
num_blocks = pdu->data_buf_len / dif_ctx->block_size;
rc = spdk_dif_verify(&iov, 1, num_blocks, dif_ctx, &err_blk);
if (rc != 0) {
SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n",
err_blk.err_type, err_blk.err_offset);
}
return rc;
}
static void
_iscsi_conn_pdu_write_done(void *cb_arg, int err)
{
struct spdk_iscsi_pdu *pdu = cb_arg;
struct spdk_iscsi_conn *conn = pdu->conn;
assert(conn != NULL);
if (spdk_unlikely(conn->state >= ISCSI_CONN_STATE_EXITING)) {
/* The other policy will recycle the resource */
return;
}
TAILQ_REMOVE(&conn->write_pdu_list, pdu, tailq);
if (err != 0) {
conn->state = ISCSI_CONN_STATE_EXITING;
} else {
spdk_trace_record(TRACE_ISCSI_FLUSH_WRITEBUF_DONE, conn->id, pdu->mapped_length, (uintptr_t)pdu, 0);
}
if ((conn->full_feature) &&
(conn->sess->ErrorRecoveryLevel >= 1) &&
iscsi_is_free_pdu_deferred(pdu)) {
SPDK_DEBUGLOG(iscsi, "stat_sn=%d\n",
from_be32(&pdu->bhs.stat_sn));
TAILQ_INSERT_TAIL(&conn->snack_pdu_list, pdu,
tailq);
} else {
iscsi_conn_free_pdu(conn, pdu);
}
}
void
iscsi_conn_pdu_generic_complete(void *cb_arg)
{
}
void
iscsi_conn_write_pdu(struct spdk_iscsi_conn *conn, struct spdk_iscsi_pdu *pdu,
iscsi_conn_xfer_complete_cb cb_fn,
void *cb_arg)
{
uint32_t crc32c;
ssize_t rc;
if (spdk_unlikely(pdu->dif_insert_or_strip)) {
rc = iscsi_dif_verify(pdu, &pdu->dif_ctx);
if (rc != 0) {
iscsi_conn_free_pdu(conn, pdu);
conn->state = ISCSI_CONN_STATE_EXITING;
return;
}
}
if (pdu->bhs.opcode != ISCSI_OP_LOGIN_RSP) {
/* Header Digest */
if (conn->header_digest) {
crc32c = iscsi_pdu_calc_header_digest(pdu);
MAKE_DIGEST_WORD(pdu->header_digest, crc32c);
}
/* Data Digest */
if (conn->data_digest && DGET24(pdu->bhs.data_segment_len) != 0) {
crc32c = iscsi_pdu_calc_data_digest(pdu);
MAKE_DIGEST_WORD(pdu->data_digest, crc32c);
}
}
pdu->cb_fn = cb_fn;
pdu->cb_arg = cb_arg;
TAILQ_INSERT_TAIL(&conn->write_pdu_list, pdu, tailq);
if (spdk_unlikely(conn->state >= ISCSI_CONN_STATE_EXITING)) {
return;
}
pdu->sock_req.iovcnt = iscsi_build_iovs(conn, pdu->iov, SPDK_COUNTOF(pdu->iov), pdu,
&pdu->mapped_length);
pdu->sock_req.cb_fn = _iscsi_conn_pdu_write_done;
pdu->sock_req.cb_arg = pdu;
spdk_trace_record(TRACE_ISCSI_FLUSH_WRITEBUF_START, conn->id, pdu->mapped_length, (uintptr_t)pdu,
pdu->sock_req.iovcnt);
spdk_sock_writev_async(conn->sock, &pdu->sock_req);
}
static void
iscsi_conn_sock_cb(void *arg, struct spdk_sock_group *group, struct spdk_sock *sock)
{
struct spdk_iscsi_conn *conn = arg;
int rc;
assert(conn != NULL);
if ((conn->state == ISCSI_CONN_STATE_EXITED) ||
(conn->state == ISCSI_CONN_STATE_EXITING)) {
return;
}
/* Handle incoming PDUs */
rc = iscsi_handle_incoming_pdus(conn);
if (rc < 0) {
conn->state = ISCSI_CONN_STATE_EXITING;
}
}
static void
iscsi: Assign connections to poll groups instead of lcores. This patch binds poll groups to SPDK thread through IO channel and assigns connections to poll groups instead of cores. iSCSI subsystem registers iSCSI global object as an IO device, and create poll groups as context of IO channels of the IO device. Each portal get and hold portal group on which the corresponding acceptor is running. When a connection is constructed, iSCSI subsystem assigns a poll group to the connection by getting it from the corresponding portal. When a connection enters full-feature phase, iSCSI subsystem schedules the connection to a poll group by round-robin. Then, each connection can know its running SPDK thread directly and can use SPDK message passing infrastructure instead of SPDK event framework. By this change, iSCSI connections are binded to SPDK thread, and not binded to processor core anymore. Some other changes in this patch are - core ID is removed from the output of get_iscsi_connections RPC. The upcoming patches will change the RPC to use spdk_for_each_channel and can access SPDK thread safely, and add SPDK thread ID instead. - utilize UT multithread framework added by the last patch to test iSCSI poll groups by UT. Change-Id: Iec73c778aa413bcabdb63141cc41d4160911ea0e Signed-off-by: Ben Walker <benjamin.walker@intel.com> Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/463359 Reviewed-by: Broadcom SPDK FC-NVMe CI <spdk-ci.pdl@broadcom.com> Reviewed-by: Jim Harris <james.r.harris@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2019-07-30 03:00:15 +00:00
iscsi_conn_full_feature_migrate(void *arg)
{
iscsi: Assign connections to poll groups instead of lcores. This patch binds poll groups to SPDK thread through IO channel and assigns connections to poll groups instead of cores. iSCSI subsystem registers iSCSI global object as an IO device, and create poll groups as context of IO channels of the IO device. Each portal get and hold portal group on which the corresponding acceptor is running. When a connection is constructed, iSCSI subsystem assigns a poll group to the connection by getting it from the corresponding portal. When a connection enters full-feature phase, iSCSI subsystem schedules the connection to a poll group by round-robin. Then, each connection can know its running SPDK thread directly and can use SPDK message passing infrastructure instead of SPDK event framework. By this change, iSCSI connections are binded to SPDK thread, and not binded to processor core anymore. Some other changes in this patch are - core ID is removed from the output of get_iscsi_connections RPC. The upcoming patches will change the RPC to use spdk_for_each_channel and can access SPDK thread safely, and add SPDK thread ID instead. - utilize UT multithread framework added by the last patch to test iSCSI poll groups by UT. Change-Id: Iec73c778aa413bcabdb63141cc41d4160911ea0e Signed-off-by: Ben Walker <benjamin.walker@intel.com> Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/463359 Reviewed-by: Broadcom SPDK FC-NVMe CI <spdk-ci.pdl@broadcom.com> Reviewed-by: Jim Harris <james.r.harris@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2019-07-30 03:00:15 +00:00
struct spdk_iscsi_conn *conn = arg;
if (conn->state >= ISCSI_CONN_STATE_EXITING) {
/* Connection is being exited before this callback is executed. */
SPDK_DEBUGLOG(iscsi, "Connection is already exited.\n");
return;
}
if (conn->sess->session_type == SESSION_TYPE_NORMAL) {
iscsi_conn_open_luns(conn);
}
iscsi: Assign connections to poll groups instead of lcores. This patch binds poll groups to SPDK thread through IO channel and assigns connections to poll groups instead of cores. iSCSI subsystem registers iSCSI global object as an IO device, and create poll groups as context of IO channels of the IO device. Each portal get and hold portal group on which the corresponding acceptor is running. When a connection is constructed, iSCSI subsystem assigns a poll group to the connection by getting it from the corresponding portal. When a connection enters full-feature phase, iSCSI subsystem schedules the connection to a poll group by round-robin. Then, each connection can know its running SPDK thread directly and can use SPDK message passing infrastructure instead of SPDK event framework. By this change, iSCSI connections are binded to SPDK thread, and not binded to processor core anymore. Some other changes in this patch are - core ID is removed from the output of get_iscsi_connections RPC. The upcoming patches will change the RPC to use spdk_for_each_channel and can access SPDK thread safely, and add SPDK thread ID instead. - utilize UT multithread framework added by the last patch to test iSCSI poll groups by UT. Change-Id: Iec73c778aa413bcabdb63141cc41d4160911ea0e Signed-off-by: Ben Walker <benjamin.walker@intel.com> Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/463359 Reviewed-by: Broadcom SPDK FC-NVMe CI <spdk-ci.pdl@broadcom.com> Reviewed-by: Jim Harris <james.r.harris@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2019-07-30 03:00:15 +00:00
/* Add this connection to the assigned poll group. */
iscsi_poll_group_add_conn(conn->pg, conn);
}
iscsi: Assign connections to poll groups instead of lcores. This patch binds poll groups to SPDK thread through IO channel and assigns connections to poll groups instead of cores. iSCSI subsystem registers iSCSI global object as an IO device, and create poll groups as context of IO channels of the IO device. Each portal get and hold portal group on which the corresponding acceptor is running. When a connection is constructed, iSCSI subsystem assigns a poll group to the connection by getting it from the corresponding portal. When a connection enters full-feature phase, iSCSI subsystem schedules the connection to a poll group by round-robin. Then, each connection can know its running SPDK thread directly and can use SPDK message passing infrastructure instead of SPDK event framework. By this change, iSCSI connections are binded to SPDK thread, and not binded to processor core anymore. Some other changes in this patch are - core ID is removed from the output of get_iscsi_connections RPC. The upcoming patches will change the RPC to use spdk_for_each_channel and can access SPDK thread safely, and add SPDK thread ID instead. - utilize UT multithread framework added by the last patch to test iSCSI poll groups by UT. Change-Id: Iec73c778aa413bcabdb63141cc41d4160911ea0e Signed-off-by: Ben Walker <benjamin.walker@intel.com> Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/463359 Reviewed-by: Broadcom SPDK FC-NVMe CI <spdk-ci.pdl@broadcom.com> Reviewed-by: Jim Harris <james.r.harris@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2019-07-30 03:00:15 +00:00
static struct spdk_iscsi_poll_group *g_next_pg = NULL;
void
iscsi_conn_schedule(struct spdk_iscsi_conn *conn)
{
struct spdk_iscsi_poll_group *pg;
struct spdk_iscsi_tgt_node *target;
if (conn->sess->session_type != SESSION_TYPE_NORMAL) {
/* Leave all non-normal sessions on the acceptor
* thread. */
return;
}
pthread_mutex_lock(&g_iscsi.mutex);
target = conn->sess->target;
pthread_mutex_lock(&target->mutex);
target->num_active_conns++;
if (target->num_active_conns == 1) {
/**
* This is the only active connection for this target node.
iscsi: Assign connections to poll groups instead of lcores. This patch binds poll groups to SPDK thread through IO channel and assigns connections to poll groups instead of cores. iSCSI subsystem registers iSCSI global object as an IO device, and create poll groups as context of IO channels of the IO device. Each portal get and hold portal group on which the corresponding acceptor is running. When a connection is constructed, iSCSI subsystem assigns a poll group to the connection by getting it from the corresponding portal. When a connection enters full-feature phase, iSCSI subsystem schedules the connection to a poll group by round-robin. Then, each connection can know its running SPDK thread directly and can use SPDK message passing infrastructure instead of SPDK event framework. By this change, iSCSI connections are binded to SPDK thread, and not binded to processor core anymore. Some other changes in this patch are - core ID is removed from the output of get_iscsi_connections RPC. The upcoming patches will change the RPC to use spdk_for_each_channel and can access SPDK thread safely, and add SPDK thread ID instead. - utilize UT multithread framework added by the last patch to test iSCSI poll groups by UT. Change-Id: Iec73c778aa413bcabdb63141cc41d4160911ea0e Signed-off-by: Ben Walker <benjamin.walker@intel.com> Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/463359 Reviewed-by: Broadcom SPDK FC-NVMe CI <spdk-ci.pdl@broadcom.com> Reviewed-by: Jim Harris <james.r.harris@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2019-07-30 03:00:15 +00:00
* Pick a poll group using round-robin.
*/
iscsi: Assign connections to poll groups instead of lcores. This patch binds poll groups to SPDK thread through IO channel and assigns connections to poll groups instead of cores. iSCSI subsystem registers iSCSI global object as an IO device, and create poll groups as context of IO channels of the IO device. Each portal get and hold portal group on which the corresponding acceptor is running. When a connection is constructed, iSCSI subsystem assigns a poll group to the connection by getting it from the corresponding portal. When a connection enters full-feature phase, iSCSI subsystem schedules the connection to a poll group by round-robin. Then, each connection can know its running SPDK thread directly and can use SPDK message passing infrastructure instead of SPDK event framework. By this change, iSCSI connections are binded to SPDK thread, and not binded to processor core anymore. Some other changes in this patch are - core ID is removed from the output of get_iscsi_connections RPC. The upcoming patches will change the RPC to use spdk_for_each_channel and can access SPDK thread safely, and add SPDK thread ID instead. - utilize UT multithread framework added by the last patch to test iSCSI poll groups by UT. Change-Id: Iec73c778aa413bcabdb63141cc41d4160911ea0e Signed-off-by: Ben Walker <benjamin.walker@intel.com> Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/463359 Reviewed-by: Broadcom SPDK FC-NVMe CI <spdk-ci.pdl@broadcom.com> Reviewed-by: Jim Harris <james.r.harris@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2019-07-30 03:00:15 +00:00
if (g_next_pg == NULL) {
g_next_pg = TAILQ_FIRST(&g_iscsi.poll_group_head);
iscsi: Assign connections to poll groups instead of lcores. This patch binds poll groups to SPDK thread through IO channel and assigns connections to poll groups instead of cores. iSCSI subsystem registers iSCSI global object as an IO device, and create poll groups as context of IO channels of the IO device. Each portal get and hold portal group on which the corresponding acceptor is running. When a connection is constructed, iSCSI subsystem assigns a poll group to the connection by getting it from the corresponding portal. When a connection enters full-feature phase, iSCSI subsystem schedules the connection to a poll group by round-robin. Then, each connection can know its running SPDK thread directly and can use SPDK message passing infrastructure instead of SPDK event framework. By this change, iSCSI connections are binded to SPDK thread, and not binded to processor core anymore. Some other changes in this patch are - core ID is removed from the output of get_iscsi_connections RPC. The upcoming patches will change the RPC to use spdk_for_each_channel and can access SPDK thread safely, and add SPDK thread ID instead. - utilize UT multithread framework added by the last patch to test iSCSI poll groups by UT. Change-Id: Iec73c778aa413bcabdb63141cc41d4160911ea0e Signed-off-by: Ben Walker <benjamin.walker@intel.com> Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/463359 Reviewed-by: Broadcom SPDK FC-NVMe CI <spdk-ci.pdl@broadcom.com> Reviewed-by: Jim Harris <james.r.harris@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2019-07-30 03:00:15 +00:00
assert(g_next_pg != NULL);
}
pg = g_next_pg;
g_next_pg = TAILQ_NEXT(g_next_pg, link);
/* Save the pg in the target node so it can be used for any other connections to this target node. */
target->pg = pg;
} else {
/**
* There are other active connections for this target node.
*/
iscsi: Assign connections to poll groups instead of lcores. This patch binds poll groups to SPDK thread through IO channel and assigns connections to poll groups instead of cores. iSCSI subsystem registers iSCSI global object as an IO device, and create poll groups as context of IO channels of the IO device. Each portal get and hold portal group on which the corresponding acceptor is running. When a connection is constructed, iSCSI subsystem assigns a poll group to the connection by getting it from the corresponding portal. When a connection enters full-feature phase, iSCSI subsystem schedules the connection to a poll group by round-robin. Then, each connection can know its running SPDK thread directly and can use SPDK message passing infrastructure instead of SPDK event framework. By this change, iSCSI connections are binded to SPDK thread, and not binded to processor core anymore. Some other changes in this patch are - core ID is removed from the output of get_iscsi_connections RPC. The upcoming patches will change the RPC to use spdk_for_each_channel and can access SPDK thread safely, and add SPDK thread ID instead. - utilize UT multithread framework added by the last patch to test iSCSI poll groups by UT. Change-Id: Iec73c778aa413bcabdb63141cc41d4160911ea0e Signed-off-by: Ben Walker <benjamin.walker@intel.com> Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/463359 Reviewed-by: Broadcom SPDK FC-NVMe CI <spdk-ci.pdl@broadcom.com> Reviewed-by: Jim Harris <james.r.harris@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2019-07-30 03:00:15 +00:00
pg = target->pg;
}
iscsi: Assign connections to poll groups instead of lcores. This patch binds poll groups to SPDK thread through IO channel and assigns connections to poll groups instead of cores. iSCSI subsystem registers iSCSI global object as an IO device, and create poll groups as context of IO channels of the IO device. Each portal get and hold portal group on which the corresponding acceptor is running. When a connection is constructed, iSCSI subsystem assigns a poll group to the connection by getting it from the corresponding portal. When a connection enters full-feature phase, iSCSI subsystem schedules the connection to a poll group by round-robin. Then, each connection can know its running SPDK thread directly and can use SPDK message passing infrastructure instead of SPDK event framework. By this change, iSCSI connections are binded to SPDK thread, and not binded to processor core anymore. Some other changes in this patch are - core ID is removed from the output of get_iscsi_connections RPC. The upcoming patches will change the RPC to use spdk_for_each_channel and can access SPDK thread safely, and add SPDK thread ID instead. - utilize UT multithread framework added by the last patch to test iSCSI poll groups by UT. Change-Id: Iec73c778aa413bcabdb63141cc41d4160911ea0e Signed-off-by: Ben Walker <benjamin.walker@intel.com> Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/463359 Reviewed-by: Broadcom SPDK FC-NVMe CI <spdk-ci.pdl@broadcom.com> Reviewed-by: Jim Harris <james.r.harris@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2019-07-30 03:00:15 +00:00
pthread_mutex_unlock(&target->mutex);
pthread_mutex_unlock(&g_iscsi.mutex);
iscsi: Assign connections to poll groups instead of lcores. This patch binds poll groups to SPDK thread through IO channel and assigns connections to poll groups instead of cores. iSCSI subsystem registers iSCSI global object as an IO device, and create poll groups as context of IO channels of the IO device. Each portal get and hold portal group on which the corresponding acceptor is running. When a connection is constructed, iSCSI subsystem assigns a poll group to the connection by getting it from the corresponding portal. When a connection enters full-feature phase, iSCSI subsystem schedules the connection to a poll group by round-robin. Then, each connection can know its running SPDK thread directly and can use SPDK message passing infrastructure instead of SPDK event framework. By this change, iSCSI connections are binded to SPDK thread, and not binded to processor core anymore. Some other changes in this patch are - core ID is removed from the output of get_iscsi_connections RPC. The upcoming patches will change the RPC to use spdk_for_each_channel and can access SPDK thread safely, and add SPDK thread ID instead. - utilize UT multithread framework added by the last patch to test iSCSI poll groups by UT. Change-Id: Iec73c778aa413bcabdb63141cc41d4160911ea0e Signed-off-by: Ben Walker <benjamin.walker@intel.com> Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/463359 Reviewed-by: Broadcom SPDK FC-NVMe CI <spdk-ci.pdl@broadcom.com> Reviewed-by: Jim Harris <james.r.harris@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2019-07-30 03:00:15 +00:00
assert(spdk_io_channel_get_thread(spdk_io_channel_from_ctx(conn->pg)) ==
spdk_get_thread());
/* Remove this connection from the previous poll group */
iscsi_poll_group_remove_conn(conn->pg, conn);
iscsi: Assign connections to poll groups instead of lcores. This patch binds poll groups to SPDK thread through IO channel and assigns connections to poll groups instead of cores. iSCSI subsystem registers iSCSI global object as an IO device, and create poll groups as context of IO channels of the IO device. Each portal get and hold portal group on which the corresponding acceptor is running. When a connection is constructed, iSCSI subsystem assigns a poll group to the connection by getting it from the corresponding portal. When a connection enters full-feature phase, iSCSI subsystem schedules the connection to a poll group by round-robin. Then, each connection can know its running SPDK thread directly and can use SPDK message passing infrastructure instead of SPDK event framework. By this change, iSCSI connections are binded to SPDK thread, and not binded to processor core anymore. Some other changes in this patch are - core ID is removed from the output of get_iscsi_connections RPC. The upcoming patches will change the RPC to use spdk_for_each_channel and can access SPDK thread safely, and add SPDK thread ID instead. - utilize UT multithread framework added by the last patch to test iSCSI poll groups by UT. Change-Id: Iec73c778aa413bcabdb63141cc41d4160911ea0e Signed-off-by: Ben Walker <benjamin.walker@intel.com> Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/463359 Reviewed-by: Broadcom SPDK FC-NVMe CI <spdk-ci.pdl@broadcom.com> Reviewed-by: Jim Harris <james.r.harris@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2019-07-30 03:00:15 +00:00
conn->pg = pg;
spdk_thread_send_msg(spdk_io_channel_get_thread(spdk_io_channel_from_ctx(pg)),
iscsi_conn_full_feature_migrate, conn);
}
static int
logout_timeout(void *arg)
{
struct spdk_iscsi_conn *conn = arg;
if (conn->state < ISCSI_CONN_STATE_EXITING) {
conn->state = ISCSI_CONN_STATE_EXITING;
}
return SPDK_POLLER_BUSY;
}
void
iscsi_conn_logout(struct spdk_iscsi_conn *conn)
{
conn->is_logged_out = true;
conn->logout_timer = SPDK_POLLER_REGISTER(logout_timeout, conn, ISCSI_LOGOUT_TIMEOUT * 1000000);
}
SPDK_TRACE_REGISTER_FN(iscsi_conn_trace, "iscsi_conn", TRACE_GROUP_ISCSI)
{
spdk_trace_register_owner(OWNER_ISCSI_CONN, 'c');
spdk_trace_register_object(OBJECT_ISCSI_PDU, 'p');
spdk_trace_register_description("ISCSI_READ_DONE", TRACE_ISCSI_READ_FROM_SOCKET_DONE,
OWNER_ISCSI_CONN, OBJECT_NONE, 0, 0, "");
spdk_trace_register_description("ISCSI_WRITE_START", TRACE_ISCSI_FLUSH_WRITEBUF_START,
OWNER_ISCSI_CONN, OBJECT_NONE, 0, 0, "iovec: ");
spdk_trace_register_description("ISCSI_WRITE_DONE", TRACE_ISCSI_FLUSH_WRITEBUF_DONE,
OWNER_ISCSI_CONN, OBJECT_NONE, 0, 0, "");
spdk_trace_register_description("ISCSI_READ_PDU", TRACE_ISCSI_READ_PDU,
OWNER_ISCSI_CONN, OBJECT_ISCSI_PDU, 1, 0, "opc: ");
spdk_trace_register_description("ISCSI_TASK_DONE", TRACE_ISCSI_TASK_DONE,
OWNER_ISCSI_CONN, OBJECT_SCSI_TASK, 0, 0, "");
spdk_trace_register_description("ISCSI_TASK_QUEUE", TRACE_ISCSI_TASK_QUEUE,
OWNER_ISCSI_CONN, OBJECT_SCSI_TASK, 1, 1, "pdu: ");
spdk_trace_register_description("ISCSI_TASK_EXECUTED", TRACE_ISCSI_TASK_EXECUTED,
OWNER_ISCSI_CONN, OBJECT_ISCSI_PDU, 0, 0, "");
spdk_trace_register_description("ISCSI_PDU_COMPLETED", TRACE_ISCSI_PDU_COMPLETED,
OWNER_ISCSI_CONN, OBJECT_ISCSI_PDU, 0, 0, "");
}
void
iscsi_conn_info_json(struct spdk_json_write_ctx *w, struct spdk_iscsi_conn *conn)
{
uint16_t tsih;
if (!conn->is_valid) {
return;
}
spdk_json_write_object_begin(w);
spdk_json_write_named_int32(w, "id", conn->id);
spdk_json_write_named_int32(w, "cid", conn->cid);
/*
* If we try to return data for a connection that has not
* logged in yet, the session will not be set. So in this
* case, return -1 for the tsih rather than segfaulting
* on the null conn->sess.
*/
if (conn->sess == NULL) {
tsih = -1;
} else {
tsih = conn->sess->tsih;
}
spdk_json_write_named_int32(w, "tsih", tsih);
spdk_json_write_named_string(w, "initiator_addr", conn->initiator_addr);
spdk_json_write_named_string(w, "target_addr", conn->target_addr);
spdk_json_write_named_string(w, "target_node_name", conn->target_short_name);
spdk_json_write_named_string(w, "thread_name",
spdk_thread_get_name(spdk_get_thread()));
spdk_json_write_object_end(w);
}