Spdk/module/event/subsystems/nvmf/nvmf_tgt.c

496 lines
13 KiB
C
Raw Normal View History

/*-
* BSD LICENSE
*
* 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 "event_nvmf.h"
#include "spdk/bdev.h"
#include "spdk/event.h"
#include "spdk/thread.h"
#include "spdk/log.h"
#include "spdk/nvme.h"
#include "spdk/nvmf_cmd.h"
#include "spdk/util.h"
enum nvmf_tgt_state {
NVMF_TGT_INIT_NONE = 0,
NVMF_TGT_INIT_PARSE_CONFIG,
NVMF_TGT_INIT_CREATE_POLL_GROUPS,
NVMF_TGT_INIT_START_SUBSYSTEMS,
NVMF_TGT_INIT_START_ACCEPTOR,
NVMF_TGT_RUNNING,
NVMF_TGT_FINI_STOP_SUBSYSTEMS,
NVMF_TGT_FINI_DESTROY_POLL_GROUPS,
NVMF_TGT_FINI_STOP_ACCEPTOR,
NVMF_TGT_FINI_FREE_RESOURCES,
NVMF_TGT_STOPPED,
NVMF_TGT_ERROR,
};
struct nvmf_tgt_poll_group {
struct spdk_nvmf_poll_group *group;
struct spdk_thread *thread;
TAILQ_ENTRY(nvmf_tgt_poll_group) link;
};
struct spdk_nvmf_tgt *g_spdk_nvmf_tgt = NULL;
static enum nvmf_tgt_state g_tgt_state;
static struct spdk_thread *g_tgt_init_thread = NULL;
nvmf_tgt: Detect completion of poll group destroy at shutdown Previously NVMe-oF target had requested each thread to destroy its poll group, but had moved to the next state without waiting for completion. To create and destroy poll group threads dynamically, NVMe-oF target have to know the completion of poll group destroy. The reason is as follows. spdk_thread_exit() requires that each poll group thread to unregister its pollers, but poller unregistration is done asynchronously, and releaes its I/O channels, but I/O channel release is done asynchronously. This patch does the following. To avoid using mutex, add the init thread. spdk_for_each_thread() requires completion callback but the completion callback is usable only if each message handler is synchronous. spdk_nvmf_poll_group_destroy() is asynchronous now. So replace spdk_for_each_thread() by parsing g_poll_groups and sending message to each corresonding thread. Then add comletion callback nvmf_tgt_destroy_poll_group_done() to spdk_nvmf_poll_group_destroy(). nvmf_tgt_destroy_poll_group_done() sends message to the init thread. The init thread executes _nvmf_tgt_destroy_poll_group_done(). _nvmf_tgt_destroy_poll_group_done() decrements g_num_poll_groups and move to the next state if it becomes zero. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: I97eb90b9dbff29d7702b20cf7ac233d39d7216b6 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/496 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
2020-02-03 01:10:34 +00:00
static struct spdk_thread *g_tgt_fini_thread = NULL;
static TAILQ_HEAD(, nvmf_tgt_poll_group) g_poll_groups = TAILQ_HEAD_INITIALIZER(g_poll_groups);
static size_t g_num_poll_groups = 0;
static struct spdk_poller *g_acceptor_poller = NULL;
static void nvmf_tgt_advance_state(void);
static void
nvmf_shutdown_cb(void *arg1)
{
/* Still in initialization state, defer shutdown operation */
if (g_tgt_state < NVMF_TGT_RUNNING) {
spdk_thread_send_msg(spdk_get_thread(), nvmf_shutdown_cb, NULL);
return;
} else if (g_tgt_state != NVMF_TGT_RUNNING && g_tgt_state != NVMF_TGT_ERROR) {
/* Already in Shutdown status, ignore the signal */
return;
}
if (g_tgt_state == NVMF_TGT_ERROR) {
/* Parse configuration error */
g_tgt_state = NVMF_TGT_FINI_FREE_RESOURCES;
} else {
g_tgt_state = NVMF_TGT_FINI_STOP_SUBSYSTEMS;
}
nvmf_tgt_advance_state();
}
static void
nvmf_subsystem_fini(void)
{
nvmf_shutdown_cb(NULL);
}
static int
acceptor_poll(void *arg)
{
struct spdk_nvmf_tgt *tgt = arg;
uint32_t count;
count = spdk_nvmf_tgt_accept(tgt);
if (count > 0) {
return SPDK_POLLER_BUSY;
} else {
return SPDK_POLLER_IDLE;
}
}
static void
nvmf_tgt: Detect completion of poll group destroy at shutdown Previously NVMe-oF target had requested each thread to destroy its poll group, but had moved to the next state without waiting for completion. To create and destroy poll group threads dynamically, NVMe-oF target have to know the completion of poll group destroy. The reason is as follows. spdk_thread_exit() requires that each poll group thread to unregister its pollers, but poller unregistration is done asynchronously, and releaes its I/O channels, but I/O channel release is done asynchronously. This patch does the following. To avoid using mutex, add the init thread. spdk_for_each_thread() requires completion callback but the completion callback is usable only if each message handler is synchronous. spdk_nvmf_poll_group_destroy() is asynchronous now. So replace spdk_for_each_thread() by parsing g_poll_groups and sending message to each corresonding thread. Then add comletion callback nvmf_tgt_destroy_poll_group_done() to spdk_nvmf_poll_group_destroy(). nvmf_tgt_destroy_poll_group_done() sends message to the init thread. The init thread executes _nvmf_tgt_destroy_poll_group_done(). _nvmf_tgt_destroy_poll_group_done() decrements g_num_poll_groups and move to the next state if it becomes zero. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: I97eb90b9dbff29d7702b20cf7ac233d39d7216b6 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/496 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
2020-02-03 01:10:34 +00:00
_nvmf_tgt_destroy_poll_group_done(void *ctx)
{
nvmf_tgt: Detect completion of poll group destroy at shutdown Previously NVMe-oF target had requested each thread to destroy its poll group, but had moved to the next state without waiting for completion. To create and destroy poll group threads dynamically, NVMe-oF target have to know the completion of poll group destroy. The reason is as follows. spdk_thread_exit() requires that each poll group thread to unregister its pollers, but poller unregistration is done asynchronously, and releaes its I/O channels, but I/O channel release is done asynchronously. This patch does the following. To avoid using mutex, add the init thread. spdk_for_each_thread() requires completion callback but the completion callback is usable only if each message handler is synchronous. spdk_nvmf_poll_group_destroy() is asynchronous now. So replace spdk_for_each_thread() by parsing g_poll_groups and sending message to each corresonding thread. Then add comletion callback nvmf_tgt_destroy_poll_group_done() to spdk_nvmf_poll_group_destroy(). nvmf_tgt_destroy_poll_group_done() sends message to the init thread. The init thread executes _nvmf_tgt_destroy_poll_group_done(). _nvmf_tgt_destroy_poll_group_done() decrements g_num_poll_groups and move to the next state if it becomes zero. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: I97eb90b9dbff29d7702b20cf7ac233d39d7216b6 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/496 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
2020-02-03 01:10:34 +00:00
assert(g_num_poll_groups > 0);
if (--g_num_poll_groups == 0) {
g_tgt_state = NVMF_TGT_FINI_STOP_ACCEPTOR;
nvmf_tgt_advance_state();
}
}
static void
nvmf_tgt_destroy_poll_group_done(void *cb_arg, int status)
{
struct nvmf_tgt_poll_group *pg = cb_arg;
free(pg);
spdk_thread_send_msg(g_tgt_fini_thread, _nvmf_tgt_destroy_poll_group_done, NULL);
spdk_thread_exit(spdk_get_thread());
}
static void
nvmf_tgt_destroy_poll_group(void *ctx)
nvmf_tgt: Detect completion of poll group destroy at shutdown Previously NVMe-oF target had requested each thread to destroy its poll group, but had moved to the next state without waiting for completion. To create and destroy poll group threads dynamically, NVMe-oF target have to know the completion of poll group destroy. The reason is as follows. spdk_thread_exit() requires that each poll group thread to unregister its pollers, but poller unregistration is done asynchronously, and releaes its I/O channels, but I/O channel release is done asynchronously. This patch does the following. To avoid using mutex, add the init thread. spdk_for_each_thread() requires completion callback but the completion callback is usable only if each message handler is synchronous. spdk_nvmf_poll_group_destroy() is asynchronous now. So replace spdk_for_each_thread() by parsing g_poll_groups and sending message to each corresonding thread. Then add comletion callback nvmf_tgt_destroy_poll_group_done() to spdk_nvmf_poll_group_destroy(). nvmf_tgt_destroy_poll_group_done() sends message to the init thread. The init thread executes _nvmf_tgt_destroy_poll_group_done(). _nvmf_tgt_destroy_poll_group_done() decrements g_num_poll_groups and move to the next state if it becomes zero. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: I97eb90b9dbff29d7702b20cf7ac233d39d7216b6 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/496 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
2020-02-03 01:10:34 +00:00
{
struct nvmf_tgt_poll_group *pg = ctx;
spdk_nvmf_poll_group_destroy(pg->group, nvmf_tgt_destroy_poll_group_done, pg);
}
static void
nvmf_tgt_destroy_poll_groups(void)
{
struct nvmf_tgt_poll_group *pg, *tpg;
nvmf_tgt: Detect completion of poll group destroy at shutdown Previously NVMe-oF target had requested each thread to destroy its poll group, but had moved to the next state without waiting for completion. To create and destroy poll group threads dynamically, NVMe-oF target have to know the completion of poll group destroy. The reason is as follows. spdk_thread_exit() requires that each poll group thread to unregister its pollers, but poller unregistration is done asynchronously, and releaes its I/O channels, but I/O channel release is done asynchronously. This patch does the following. To avoid using mutex, add the init thread. spdk_for_each_thread() requires completion callback but the completion callback is usable only if each message handler is synchronous. spdk_nvmf_poll_group_destroy() is asynchronous now. So replace spdk_for_each_thread() by parsing g_poll_groups and sending message to each corresonding thread. Then add comletion callback nvmf_tgt_destroy_poll_group_done() to spdk_nvmf_poll_group_destroy(). nvmf_tgt_destroy_poll_group_done() sends message to the init thread. The init thread executes _nvmf_tgt_destroy_poll_group_done(). _nvmf_tgt_destroy_poll_group_done() decrements g_num_poll_groups and move to the next state if it becomes zero. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: I97eb90b9dbff29d7702b20cf7ac233d39d7216b6 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/496 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
2020-02-03 01:10:34 +00:00
g_tgt_fini_thread = spdk_get_thread();
assert(g_tgt_fini_thread != NULL);
TAILQ_FOREACH_SAFE(pg, &g_poll_groups, link, tpg) {
nvmf_tgt: Detect completion of poll group destroy at shutdown Previously NVMe-oF target had requested each thread to destroy its poll group, but had moved to the next state without waiting for completion. To create and destroy poll group threads dynamically, NVMe-oF target have to know the completion of poll group destroy. The reason is as follows. spdk_thread_exit() requires that each poll group thread to unregister its pollers, but poller unregistration is done asynchronously, and releaes its I/O channels, but I/O channel release is done asynchronously. This patch does the following. To avoid using mutex, add the init thread. spdk_for_each_thread() requires completion callback but the completion callback is usable only if each message handler is synchronous. spdk_nvmf_poll_group_destroy() is asynchronous now. So replace spdk_for_each_thread() by parsing g_poll_groups and sending message to each corresonding thread. Then add comletion callback nvmf_tgt_destroy_poll_group_done() to spdk_nvmf_poll_group_destroy(). nvmf_tgt_destroy_poll_group_done() sends message to the init thread. The init thread executes _nvmf_tgt_destroy_poll_group_done(). _nvmf_tgt_destroy_poll_group_done() decrements g_num_poll_groups and move to the next state if it becomes zero. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: I97eb90b9dbff29d7702b20cf7ac233d39d7216b6 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/496 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
2020-02-03 01:10:34 +00:00
TAILQ_REMOVE(&g_poll_groups, pg, link);
spdk_thread_send_msg(pg->thread, nvmf_tgt_destroy_poll_group, pg);
}
}
static void
nvmf_tgt_create_poll_group_done(void *ctx)
{
struct nvmf_tgt_poll_group *pg = ctx;
TAILQ_INSERT_TAIL(&g_poll_groups, pg, link);
assert(g_num_poll_groups < spdk_env_get_core_count());
if (++g_num_poll_groups == spdk_env_get_core_count()) {
g_tgt_state = NVMF_TGT_INIT_START_SUBSYSTEMS;
nvmf_tgt_advance_state();
}
}
static void
nvmf_tgt_create_poll_group(void *ctx)
{
struct nvmf_tgt_poll_group *pg;
pg = calloc(1, sizeof(*pg));
if (!pg) {
SPDK_ERRLOG("Not enough memory to allocate poll groups\n");
spdk_app_stop(-ENOMEM);
return;
}
pg->thread = spdk_get_thread();
pg->group = spdk_nvmf_poll_group_create(g_spdk_nvmf_tgt);
spdk_thread_send_msg(g_tgt_init_thread, nvmf_tgt_create_poll_group_done, pg);
}
static void
nvmf_tgt_create_poll_groups(void)
{
struct spdk_cpuset tmp_cpumask = {};
uint32_t i;
char thread_name[32];
struct spdk_thread *thread;
g_tgt_init_thread = spdk_get_thread();
assert(g_tgt_init_thread != NULL);
SPDK_ENV_FOREACH_CORE(i) {
spdk_cpuset_zero(&tmp_cpumask);
spdk_cpuset_set_cpu(&tmp_cpumask, i, true);
snprintf(thread_name, sizeof(thread_name), "nvmf_tgt_poll_group_%u", i);
thread = spdk_thread_create(thread_name, &tmp_cpumask);
assert(thread != NULL);
spdk_thread_send_msg(thread, nvmf_tgt_create_poll_group, NULL);
}
}
static void
nvmf_tgt_subsystem_started(struct spdk_nvmf_subsystem *subsystem,
void *cb_arg, int status)
{
subsystem = spdk_nvmf_subsystem_get_next(subsystem);
int rc;
if (subsystem) {
rc = spdk_nvmf_subsystem_start(subsystem, nvmf_tgt_subsystem_started, NULL);
if (rc) {
g_tgt_state = NVMF_TGT_FINI_STOP_SUBSYSTEMS;
SPDK_ERRLOG("Unable to start NVMe-oF subsystem. Stopping app.\n");
nvmf_tgt_advance_state();
}
return;
}
g_tgt_state = NVMF_TGT_INIT_START_ACCEPTOR;
nvmf_tgt_advance_state();
}
static void
nvmf_tgt_subsystem_stopped(struct spdk_nvmf_subsystem *subsystem,
void *cb_arg, int status)
{
subsystem = spdk_nvmf_subsystem_get_next(subsystem);
int rc;
if (subsystem) {
rc = spdk_nvmf_subsystem_stop(subsystem, nvmf_tgt_subsystem_stopped, NULL);
if (rc) {
SPDK_ERRLOG("Unable to stop NVMe-oF subsystem. Trying others.\n");
nvmf_tgt_subsystem_stopped(subsystem, NULL, 0);
}
return;
}
g_tgt_state = NVMF_TGT_FINI_DESTROY_POLL_GROUPS;
nvmf_tgt_advance_state();
}
static void
nvmf_tgt_destroy_done(void *ctx, int status)
{
g_tgt_state = NVMF_TGT_STOPPED;
free(g_spdk_nvmf_tgt_conf);
g_spdk_nvmf_tgt_conf = NULL;
nvmf_tgt_advance_state();
}
static void
nvmf_tgt_parse_conf_done(int status)
{
g_tgt_state = (status == 0) ? NVMF_TGT_INIT_CREATE_POLL_GROUPS : NVMF_TGT_ERROR;
nvmf_tgt_advance_state();
}
static void
nvmf_tgt_parse_conf_start(void *ctx)
{
if (nvmf_parse_conf(nvmf_tgt_parse_conf_done)) {
SPDK_ERRLOG("nvmf_parse_conf() failed\n");
g_tgt_state = NVMF_TGT_ERROR;
nvmf_tgt_advance_state();
}
}
static void
fixup_identify_ctrlr(struct spdk_nvmf_request *req)
{
uint32_t length;
int rc;
struct spdk_nvme_ctrlr_data *nvme_cdata;
struct spdk_nvme_ctrlr_data nvmf_cdata = {};
struct spdk_nvmf_ctrlr *ctrlr = spdk_nvmf_request_get_ctrlr(req);
struct spdk_nvme_cpl *rsp = spdk_nvmf_request_get_response(req);
/* This is the identify data from the NVMe drive */
spdk_nvmf_request_get_data(req, (void **)&nvme_cdata, &length);
/* Get the NVMF identify data */
rc = spdk_nvmf_ctrlr_identify_ctrlr(ctrlr, &nvmf_cdata);
if (rc != SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) {
rsp->status.sct = SPDK_NVME_SCT_GENERIC;
rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
return;
}
/* Fixup NVMF identify data with NVMe identify data */
/* Serial Number (SN) */
memcpy(&nvmf_cdata.sn[0], &nvme_cdata->sn[0], sizeof(nvmf_cdata.sn));
/* Model Number (MN) */
memcpy(&nvmf_cdata.mn[0], &nvme_cdata->mn[0], sizeof(nvmf_cdata.mn));
/* Firmware Revision (FR) */
memcpy(&nvmf_cdata.fr[0], &nvme_cdata->fr[0], sizeof(nvmf_cdata.fr));
/* IEEE OUI Identifier (IEEE) */
memcpy(&nvmf_cdata.ieee[0], &nvme_cdata->ieee[0], sizeof(nvmf_cdata.ieee));
/* FRU Globally Unique Identifier (FGUID) */
/* Copy the fixed up data back to the response */
memcpy(nvme_cdata, &nvmf_cdata, length);
}
static int
nvmf_custom_identify_hdlr(struct spdk_nvmf_request *req)
{
struct spdk_nvme_cmd *cmd = spdk_nvmf_request_get_cmd(req);
struct spdk_bdev *bdev;
struct spdk_bdev_desc *desc;
struct spdk_io_channel *ch;
struct spdk_nvmf_subsystem *subsys;
int rc;
if (cmd->cdw10_bits.identify.cns != SPDK_NVME_IDENTIFY_CTRLR) {
return -1; /* continue */
}
subsys = spdk_nvmf_request_get_subsystem(req);
if (subsys == NULL) {
return -1;
}
/* Only procss this request if it has exactly one namespace */
if (spdk_nvmf_subsystem_get_max_nsid(subsys) != 1) {
return -1;
}
/* Forward to first namespace if it supports NVME admin commands */
rc = spdk_nvmf_request_get_bdev(1, req, &bdev, &desc, &ch);
if (rc) {
/* No bdev found for this namespace. Continue. */
return -1;
}
if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_NVME_ADMIN)) {
return -1;
}
return spdk_nvmf_bdev_ctrlr_nvme_passthru_admin(bdev, desc, ch, req, fixup_identify_ctrlr);
}
static void
nvmf_tgt_advance_state(void)
{
enum nvmf_tgt_state prev_state;
int rc = -1;
int ret;
do {
prev_state = g_tgt_state;
switch (g_tgt_state) {
case NVMF_TGT_INIT_NONE: {
g_tgt_state = NVMF_TGT_INIT_PARSE_CONFIG;
break;
}
case NVMF_TGT_INIT_PARSE_CONFIG:
/* Send message to self to call parse conf func.
* Prevents it from possibly performing cb before getting
* out of this function, which causes problems. */
spdk_thread_send_msg(spdk_get_thread(), nvmf_tgt_parse_conf_start, NULL);
break;
case NVMF_TGT_INIT_CREATE_POLL_GROUPS:
/* Config parsed */
if (g_spdk_nvmf_tgt_conf->admin_passthru.identify_ctrlr) {
SPDK_NOTICELOG("Custom identify ctrlr handler enabled\n");
spdk_nvmf_set_custom_admin_cmd_hdlr(SPDK_NVME_OPC_IDENTIFY, nvmf_custom_identify_hdlr);
}
/* Create poll group threads, and send a message to each thread
* and create a poll group.
*/
nvmf_tgt_create_poll_groups();
break;
case NVMF_TGT_INIT_START_SUBSYSTEMS: {
struct spdk_nvmf_subsystem *subsystem;
subsystem = spdk_nvmf_subsystem_get_first(g_spdk_nvmf_tgt);
if (subsystem) {
ret = spdk_nvmf_subsystem_start(subsystem, nvmf_tgt_subsystem_started, NULL);
if (ret) {
SPDK_ERRLOG("Unable to start NVMe-oF subsystem. Stopping app.\n");
g_tgt_state = NVMF_TGT_FINI_STOP_SUBSYSTEMS;
}
} else {
g_tgt_state = NVMF_TGT_INIT_START_ACCEPTOR;
}
break;
}
case NVMF_TGT_INIT_START_ACCEPTOR:
g_acceptor_poller = SPDK_POLLER_REGISTER(acceptor_poll, g_spdk_nvmf_tgt,
g_spdk_nvmf_tgt_conf->acceptor_poll_rate);
g_tgt_state = NVMF_TGT_RUNNING;
break;
case NVMF_TGT_RUNNING:
spdk_subsystem_init_next(0);
break;
case NVMF_TGT_FINI_STOP_SUBSYSTEMS: {
struct spdk_nvmf_subsystem *subsystem;
subsystem = spdk_nvmf_subsystem_get_first(g_spdk_nvmf_tgt);
if (subsystem) {
ret = spdk_nvmf_subsystem_stop(subsystem, nvmf_tgt_subsystem_stopped, NULL);
if (ret) {
nvmf_tgt_subsystem_stopped(subsystem, NULL, 0);
}
} else {
g_tgt_state = NVMF_TGT_FINI_DESTROY_POLL_GROUPS;
}
break;
}
case NVMF_TGT_FINI_DESTROY_POLL_GROUPS:
/* Send a message to each poll group thread, and terminate the thread */
nvmf_tgt: Detect completion of poll group destroy at shutdown Previously NVMe-oF target had requested each thread to destroy its poll group, but had moved to the next state without waiting for completion. To create and destroy poll group threads dynamically, NVMe-oF target have to know the completion of poll group destroy. The reason is as follows. spdk_thread_exit() requires that each poll group thread to unregister its pollers, but poller unregistration is done asynchronously, and releaes its I/O channels, but I/O channel release is done asynchronously. This patch does the following. To avoid using mutex, add the init thread. spdk_for_each_thread() requires completion callback but the completion callback is usable only if each message handler is synchronous. spdk_nvmf_poll_group_destroy() is asynchronous now. So replace spdk_for_each_thread() by parsing g_poll_groups and sending message to each corresonding thread. Then add comletion callback nvmf_tgt_destroy_poll_group_done() to spdk_nvmf_poll_group_destroy(). nvmf_tgt_destroy_poll_group_done() sends message to the init thread. The init thread executes _nvmf_tgt_destroy_poll_group_done(). _nvmf_tgt_destroy_poll_group_done() decrements g_num_poll_groups and move to the next state if it becomes zero. Signed-off-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Change-Id: I97eb90b9dbff29d7702b20cf7ac233d39d7216b6 Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/496 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Aleksey Marchuk <alexeymar@mellanox.com>
2020-02-03 01:10:34 +00:00
nvmf_tgt_destroy_poll_groups();
break;
case NVMF_TGT_FINI_STOP_ACCEPTOR:
spdk_poller_unregister(&g_acceptor_poller);
g_tgt_state = NVMF_TGT_FINI_FREE_RESOURCES;
break;
case NVMF_TGT_FINI_FREE_RESOURCES:
spdk_nvmf_tgt_destroy(g_spdk_nvmf_tgt, nvmf_tgt_destroy_done, NULL);
break;
case NVMF_TGT_STOPPED:
spdk_subsystem_fini_next();
return;
case NVMF_TGT_ERROR:
spdk_subsystem_init_next(rc);
return;
}
} while (g_tgt_state != prev_state);
}
static void
nvmf_subsystem_init(void)
{
g_tgt_state = NVMF_TGT_INIT_NONE;
nvmf_tgt_advance_state();
}
static void
nvmf_subsystem_write_config_json(struct spdk_json_write_ctx *w)
{
spdk_json_write_array_begin(w);
spdk_json_write_object_begin(w);
spdk_json_write_named_string(w, "method", "nvmf_set_config");
spdk_json_write_named_object_begin(w, "params");
spdk_json_write_named_uint32(w, "acceptor_poll_rate", g_spdk_nvmf_tgt_conf->acceptor_poll_rate);
spdk_json_write_named_object_begin(w, "admin_cmd_passthru");
spdk_json_write_named_bool(w, "identify_ctrlr",
g_spdk_nvmf_tgt_conf->admin_passthru.identify_ctrlr);
spdk_json_write_object_end(w);
spdk_json_write_object_end(w);
spdk_json_write_object_end(w);
spdk_nvmf_tgt_write_config_json(w, g_spdk_nvmf_tgt);
spdk_json_write_array_end(w);
}
static struct spdk_subsystem g_spdk_subsystem_nvmf = {
.name = "nvmf",
.init = nvmf_subsystem_init,
.fini = nvmf_subsystem_fini,
.write_config_json = nvmf_subsystem_write_config_json,
};
SPDK_SUBSYSTEM_REGISTER(g_spdk_subsystem_nvmf)
SPDK_SUBSYSTEM_DEPEND(nvmf, bdev)
SPDK_SUBSYSTEM_DEPEND(nvmf, sock)