mirror of
https://github.com/zephyrproject-rtos/zephyr
synced 2025-09-02 10:12:51 +00:00
The documentation says that the API will automatically append the net_buf fragment to the end of network packet fragment chain. This was not the case and current only user for this API in echo-server sample appended the fragment itself. The fix is to automatically append the fragment to the end of fragment chain. Signed-off-by: Jukka Rissanen <jukka.rissanen@linux.intel.com>
1929 lines
42 KiB
C
1929 lines
42 KiB
C
/* net_app.c */
|
|
|
|
/*
|
|
* Copyright (c) 2017 Intel Corporation.
|
|
*
|
|
* SPDX-License-Identifier: Apache-2.0
|
|
*/
|
|
|
|
#if defined(CONFIG_NET_DEBUG_APP)
|
|
#define SYS_LOG_DOMAIN "net/app"
|
|
#define NET_SYS_LOG_LEVEL SYS_LOG_LEVEL_DEBUG
|
|
#define NET_LOG_ENABLED 1
|
|
#endif
|
|
|
|
#include <zephyr.h>
|
|
#include <string.h>
|
|
#include <errno.h>
|
|
#include <stdlib.h>
|
|
|
|
#include <net/net_core.h>
|
|
#include <net/net_ip.h>
|
|
#include <net/net_if.h>
|
|
#include <net/dhcpv4.h>
|
|
#include <net/net_mgmt.h>
|
|
#include <net/udp.h>
|
|
|
|
#include <net/net_app.h>
|
|
|
|
#include "../../ip/udp_internal.h"
|
|
|
|
#include "net_app_private.h"
|
|
|
|
#if defined(CONFIG_NET_APP_DTLS_TIMEOUT)
|
|
#define DTLS_TIMEOUT K_SECONDS(CONFIG_NET_APP_DTLS_TIMEOUT)
|
|
#else
|
|
#define DTLS_TIMEOUT K_SECONDS(15)
|
|
#endif
|
|
|
|
#if defined(CONFIG_NET_DEBUG_APP)
|
|
static sys_slist_t _net_app_instances;
|
|
|
|
void _net_app_register(struct net_app_ctx *ctx)
|
|
{
|
|
sys_slist_prepend(&_net_app_instances, &ctx->node);
|
|
}
|
|
|
|
void _net_app_unregister(struct net_app_ctx *ctx)
|
|
{
|
|
sys_slist_find_and_remove(&_net_app_instances, &ctx->node);
|
|
}
|
|
|
|
static void net_app_foreach(net_app_ctx_cb_t cb, enum net_app_type type,
|
|
void *user_data)
|
|
{
|
|
struct net_app_ctx *ctx;
|
|
|
|
SYS_SLIST_FOR_EACH_CONTAINER(&_net_app_instances, ctx, node) {
|
|
if (ctx->is_init && ctx->app_type == type) {
|
|
cb(ctx, user_data);
|
|
}
|
|
}
|
|
}
|
|
|
|
void net_app_server_foreach(net_app_ctx_cb_t cb, void *user_data)
|
|
{
|
|
net_app_foreach(cb, NET_APP_SERVER, user_data);
|
|
}
|
|
|
|
void net_app_client_foreach(net_app_ctx_cb_t cb, void *user_data)
|
|
{
|
|
net_app_foreach(cb, NET_APP_CLIENT, user_data);
|
|
}
|
|
#endif /* CONFIG_NET_DEBUG_APP */
|
|
|
|
#if defined(CONFIG_NET_CONTEXT_NET_PKT_POOL)
|
|
int net_app_set_net_pkt_pool(struct net_app_ctx *ctx,
|
|
net_pkt_get_slab_func_t tx_slab,
|
|
net_pkt_get_pool_func_t data_pool)
|
|
{
|
|
ctx->tx_slab = tx_slab;
|
|
ctx->data_pool = data_pool;
|
|
|
|
return 0;
|
|
}
|
|
#endif /* CONFIG_NET_CONTEXT_NET_PKT_POOL */
|
|
|
|
#if defined(CONFIG_NET_DEBUG_APP)
|
|
char *_net_app_sprint_ipaddr(char *buf, int buflen,
|
|
const struct sockaddr *addr)
|
|
{
|
|
if (addr->family == AF_INET6) {
|
|
#if defined(CONFIG_NET_IPV6)
|
|
char ipaddr[NET_IPV6_ADDR_LEN];
|
|
|
|
net_addr_ntop(addr->family,
|
|
&net_sin6(addr)->sin6_addr,
|
|
ipaddr, sizeof(ipaddr));
|
|
snprintk(buf, buflen, "[%s]:%u", ipaddr,
|
|
ntohs(net_sin6(addr)->sin6_port));
|
|
#endif
|
|
} else if (addr->family == AF_INET) {
|
|
#if defined(CONFIG_NET_IPV4)
|
|
char ipaddr[NET_IPV4_ADDR_LEN];
|
|
|
|
net_addr_ntop(addr->family,
|
|
&net_sin(addr)->sin_addr,
|
|
ipaddr, sizeof(ipaddr));
|
|
snprintk(buf, buflen, "%s:%u", ipaddr,
|
|
ntohs(net_sin(addr)->sin_port));
|
|
#endif
|
|
}
|
|
|
|
return buf;
|
|
}
|
|
|
|
void _net_app_print_info(struct net_app_ctx *ctx)
|
|
{
|
|
#define PORT_STR_LEN sizeof("[]:xxxxx")
|
|
char local[NET_IPV6_ADDR_LEN + PORT_STR_LEN];
|
|
char remote[NET_IPV6_ADDR_LEN + PORT_STR_LEN];
|
|
|
|
_net_app_sprint_ipaddr(local, sizeof(local), &ctx->default_ctx->local);
|
|
_net_app_sprint_ipaddr(remote, sizeof(remote),
|
|
&ctx->default_ctx->remote);
|
|
|
|
NET_DBG("net app connect %s %s %s",
|
|
local,
|
|
ctx->app_type == NET_APP_CLIENT ? "->" : "<-",
|
|
remote);
|
|
}
|
|
#endif /* CONFIG_NET_DEBUG_APP */
|
|
|
|
#if defined(CONFIG_NET_APP_SERVER) || defined(CONFIG_NET_APP_CLIENT)
|
|
void _net_app_received(struct net_context *net_ctx,
|
|
struct net_pkt *pkt,
|
|
int status,
|
|
void *user_data)
|
|
{
|
|
struct net_app_ctx *ctx = user_data;
|
|
|
|
#if defined(CONFIG_NET_APP_CLIENT)
|
|
if (ctx->app_type == NET_APP_CLIENT) {
|
|
if (!pkt) {
|
|
if (ctx->cb.close) {
|
|
ctx->cb.close(ctx, status, ctx->user_data);
|
|
}
|
|
|
|
return;
|
|
}
|
|
|
|
if (ctx->cb.recv) {
|
|
ctx->cb.recv(ctx, pkt, status, ctx->user_data);
|
|
}
|
|
}
|
|
#endif
|
|
|
|
#if defined(CONFIG_NET_APP_SERVER)
|
|
if (ctx->app_type == NET_APP_SERVER) {
|
|
if (!pkt) {
|
|
if (ctx->cb.close) {
|
|
ctx->cb.close(ctx, status, ctx->user_data);
|
|
}
|
|
|
|
#if defined(CONFIG_NET_TCP)
|
|
if (ctx->proto == IPPROTO_TCP) {
|
|
net_context_put(ctx->server.net_ctx);
|
|
ctx->server.net_ctx = NULL;
|
|
}
|
|
#endif
|
|
|
|
return;
|
|
}
|
|
|
|
if (ctx->cb.recv) {
|
|
ctx->cb.recv(ctx, pkt, status, ctx->user_data);
|
|
}
|
|
}
|
|
#endif
|
|
}
|
|
#endif /* CONFIG_NET_APP_SERVER || CONFIG_NET_APP_CLIENT */
|
|
|
|
#if defined(CONFIG_NET_APP_SERVER) || defined(CONFIG_NET_APP_CLIENT)
|
|
int _net_app_set_net_ctx(struct net_app_ctx *ctx,
|
|
struct net_context *net_ctx,
|
|
struct sockaddr *addr,
|
|
socklen_t socklen,
|
|
enum net_ip_protocol proto)
|
|
{
|
|
int ret;
|
|
|
|
ret = net_context_bind(net_ctx, addr, socklen);
|
|
if (ret < 0) {
|
|
NET_ERR("Cannot bind context (%d)", ret);
|
|
goto out;
|
|
}
|
|
|
|
#if defined(CONFIG_NET_APP_SERVER) && defined(CONFIG_NET_TCP)
|
|
if (ctx->app_type == NET_APP_SERVER && proto == IPPROTO_TCP) {
|
|
ret = net_context_listen(net_ctx, 0);
|
|
if (ret < 0) {
|
|
NET_ERR("Cannot listen context (%d)", ret);
|
|
goto out;
|
|
}
|
|
|
|
ret = net_context_accept(net_ctx, _net_app_accept_cb,
|
|
K_NO_WAIT, ctx);
|
|
if (ret < 0) {
|
|
NET_ERR("Cannot accept context (%d)", ret);
|
|
goto out;
|
|
}
|
|
|
|
/* TCP recv callback is set after we have accepted the
|
|
* connection.
|
|
*/
|
|
}
|
|
#endif /* CONFIG_NET_APP_SERVER && CONFIG_NET_TCP */
|
|
|
|
#if defined(CONFIG_NET_APP_SERVER) && defined(CONFIG_NET_UDP)
|
|
if (ctx->app_type == NET_APP_SERVER && proto == IPPROTO_UDP) {
|
|
net_context_recv(net_ctx, _net_app_received, K_NO_WAIT, ctx);
|
|
}
|
|
#endif /* CONFIG_NET_APP_SERVER && CONFIG_NET_UDP */
|
|
|
|
out:
|
|
return ret;
|
|
}
|
|
|
|
int _net_app_set_local_addr(struct sockaddr *addr, const char *myaddr,
|
|
u16_t port)
|
|
{
|
|
if (myaddr) {
|
|
void *inaddr;
|
|
|
|
if (addr->family == AF_INET) {
|
|
#if defined(CONFIG_NET_IPV4)
|
|
inaddr = &net_sin(addr)->sin_addr;
|
|
net_sin(addr)->sin_port = htons(port);
|
|
#else
|
|
return -EPFNOSUPPORT;
|
|
#endif
|
|
} else if (addr->family == AF_INET6) {
|
|
#if defined(CONFIG_NET_IPV6)
|
|
inaddr = &net_sin6(addr)->sin6_addr;
|
|
net_sin6(addr)->sin6_port = htons(port);
|
|
#else
|
|
return -EPFNOSUPPORT;
|
|
#endif
|
|
} else {
|
|
return -EAFNOSUPPORT;
|
|
}
|
|
|
|
return net_addr_pton(addr->family, myaddr, inaddr);
|
|
}
|
|
|
|
/* If the caller did not supply the address where to bind, then
|
|
* try to figure it out ourselves.
|
|
*/
|
|
if (addr->family == AF_INET6) {
|
|
#if defined(CONFIG_NET_IPV6)
|
|
net_ipaddr_copy(&net_sin6(addr)->sin6_addr,
|
|
net_if_ipv6_select_src_addr(NULL,
|
|
(struct in6_addr *)
|
|
net_ipv6_unspecified_address()));
|
|
#else
|
|
return -EPFNOSUPPORT;
|
|
#endif
|
|
} else if (addr->family == AF_INET) {
|
|
#if defined(CONFIG_NET_IPV4)
|
|
struct net_if *iface = net_if_get_default();
|
|
|
|
/* For IPv4 we take the first address in the interface */
|
|
net_ipaddr_copy(&net_sin(addr)->sin_addr,
|
|
&iface->ipv4.unicast[0].address.in_addr);
|
|
#else
|
|
return -EPFNOSUPPORT;
|
|
#endif
|
|
}
|
|
|
|
return 0;
|
|
}
|
|
#endif /* CONFIG_NET_APP_SERVER || CONFIG_NET_APP_CLIENT */
|
|
|
|
#if defined(CONFIG_NET_APP_SERVER)
|
|
#endif /* CONFIG_NET_APP_SERVER */
|
|
|
|
#if defined(CONFIG_NET_IPV4) && (defined(CONFIG_NET_APP_SERVER) || \
|
|
defined(CONFIG_NET_APP_CLIENT))
|
|
static int setup_ipv4_ctx(struct net_app_ctx *ctx,
|
|
enum net_sock_type sock_type,
|
|
enum net_ip_protocol proto)
|
|
{
|
|
int ret;
|
|
|
|
ret = net_context_get(AF_INET, sock_type, proto, &ctx->ipv4.ctx);
|
|
if (ret < 0) {
|
|
NET_ERR("Cannot get network context (%d)", ret);
|
|
ctx->ipv4.ctx = NULL;
|
|
return ret;
|
|
}
|
|
|
|
net_context_setup_pools(ctx->ipv4.ctx, ctx->tx_slab,
|
|
ctx->data_pool);
|
|
|
|
return ret;
|
|
}
|
|
#endif /* CONFIG_NET_IPV4 */
|
|
|
|
#if defined(CONFIG_NET_IPV6) && (defined(CONFIG_NET_APP_SERVER) || \
|
|
defined(CONFIG_NET_APP_CLIENT))
|
|
static int setup_ipv6_ctx(struct net_app_ctx *ctx,
|
|
enum net_sock_type sock_type,
|
|
enum net_ip_protocol proto)
|
|
{
|
|
int ret;
|
|
|
|
ret = net_context_get(AF_INET6, sock_type, proto, &ctx->ipv6.ctx);
|
|
if (ret < 0) {
|
|
NET_ERR("Cannot get network context (%d)", ret);
|
|
ctx->ipv6.ctx = NULL;
|
|
return ret;
|
|
}
|
|
|
|
net_context_setup_pools(ctx->ipv6.ctx, ctx->tx_slab,
|
|
ctx->data_pool);
|
|
|
|
return ret;
|
|
}
|
|
#endif /* CONFIG_NET_IPV6 */
|
|
|
|
#if defined(CONFIG_NET_APP_SERVER) || defined(CONFIG_NET_APP_CLIENT)
|
|
static void select_default_ctx(struct net_app_ctx *ctx)
|
|
{
|
|
#if defined(CONFIG_NET_IPV6)
|
|
ctx->default_ctx = &ctx->ipv6;
|
|
#elif defined(CONFIG_NET_IPV4)
|
|
ctx->default_ctx = &ctx->ipv4;
|
|
#endif
|
|
}
|
|
|
|
int _net_app_config_local_ctx(struct net_app_ctx *ctx,
|
|
enum net_sock_type sock_type,
|
|
enum net_ip_protocol proto,
|
|
struct sockaddr *addr)
|
|
{
|
|
int ret;
|
|
|
|
if (!addr) {
|
|
#if defined(CONFIG_NET_IPV6)
|
|
if (ctx->ipv6.local.family == AF_INET6 ||
|
|
ctx->ipv6.local.family == AF_UNSPEC) {
|
|
ret = setup_ipv6_ctx(ctx, sock_type, proto);
|
|
} else {
|
|
ret = -EPFNOSUPPORT;
|
|
goto fail;
|
|
}
|
|
#endif
|
|
|
|
#if defined(CONFIG_NET_IPV4)
|
|
if (ctx->ipv4.local.family == AF_INET ||
|
|
ctx->ipv4.local.family == AF_UNSPEC) {
|
|
ret = setup_ipv4_ctx(ctx, sock_type, proto);
|
|
} else {
|
|
ret = -EPFNOSUPPORT;
|
|
goto fail;
|
|
}
|
|
#endif
|
|
|
|
select_default_ctx(ctx);
|
|
} else {
|
|
if (addr->family == AF_INET6) {
|
|
#if defined(CONFIG_NET_IPV6)
|
|
ret = setup_ipv6_ctx(ctx, sock_type, proto);
|
|
ctx->default_ctx = &ctx->ipv6;
|
|
#else
|
|
ret = -EPFNOSUPPORT;
|
|
goto fail;
|
|
#endif
|
|
} else if (addr->family == AF_INET) {
|
|
#if defined(CONFIG_NET_IPV4)
|
|
ret = setup_ipv4_ctx(ctx, sock_type, proto);
|
|
ctx->default_ctx = &ctx->ipv4;
|
|
#else
|
|
ret = -EPFNOSUPPORT;
|
|
goto fail;
|
|
#endif
|
|
} else if (addr->family == AF_UNSPEC) {
|
|
#if defined(CONFIG_NET_IPV4)
|
|
ret = setup_ipv4_ctx(ctx, sock_type, proto);
|
|
ctx->default_ctx = &ctx->ipv4;
|
|
#endif
|
|
/* We ignore the IPv4 error if IPv6 is enabled */
|
|
#if defined(CONFIG_NET_IPV6)
|
|
ret = setup_ipv6_ctx(ctx, sock_type, proto);
|
|
ctx->default_ctx = &ctx->ipv6;
|
|
#endif
|
|
} else {
|
|
ret = -EINVAL;
|
|
goto fail;
|
|
}
|
|
}
|
|
|
|
fail:
|
|
return ret;
|
|
}
|
|
#endif /* CONFIG_NET_APP_SERVER || CONFIG_NET_APP_CLIENT */
|
|
|
|
int net_app_release(struct net_app_ctx *ctx)
|
|
{
|
|
if (!ctx) {
|
|
return -EINVAL;
|
|
}
|
|
|
|
if (!ctx->is_init) {
|
|
return -ENOENT;
|
|
}
|
|
|
|
#if defined(CONFIG_NET_IPV6)
|
|
if (ctx->ipv6.ctx) {
|
|
net_context_put(ctx->ipv6.ctx);
|
|
ctx->ipv6.ctx = NULL;
|
|
}
|
|
#endif /* CONFIG_NET_IPV6 */
|
|
|
|
#if defined(CONFIG_NET_IPV4)
|
|
if (ctx->ipv4.ctx) {
|
|
net_context_put(ctx->ipv4.ctx);
|
|
ctx->ipv4.ctx = NULL;
|
|
}
|
|
#endif /* CONFIG_NET_IPV4 */
|
|
|
|
ctx->is_init = false;
|
|
|
|
_net_app_unregister(ctx);
|
|
|
|
return 0;
|
|
}
|
|
|
|
#if defined(CONFIG_NET_APP_CLIENT)
|
|
static inline
|
|
struct net_context *select_client_ctx(struct net_app_ctx *ctx,
|
|
const struct sockaddr *dst)
|
|
{
|
|
if (ctx->proto == IPPROTO_UDP) {
|
|
if (!dst) {
|
|
if (ctx->is_tls) {
|
|
#if defined(CONFIG_NET_APP_DTLS)
|
|
if (ctx->dtls.ctx) {
|
|
return ctx->dtls.ctx;
|
|
} else {
|
|
return ctx->default_ctx->ctx;
|
|
}
|
|
#else
|
|
return NULL;
|
|
#endif
|
|
} else {
|
|
return ctx->default_ctx->ctx;
|
|
}
|
|
} else {
|
|
if (ctx->is_tls) {
|
|
#if defined(CONFIG_NET_APP_DTLS)
|
|
if (ctx->dtls.ctx) {
|
|
return ctx->dtls.ctx;
|
|
}
|
|
#else
|
|
return NULL;
|
|
#endif
|
|
}
|
|
|
|
goto common_checks;
|
|
}
|
|
} else {
|
|
if (!dst) {
|
|
return ctx->default_ctx->ctx;
|
|
} else {
|
|
common_checks:
|
|
if (dst->family == AF_INET) {
|
|
#if defined(CONFIG_NET_IPV4)
|
|
return ctx->ipv4.ctx;
|
|
#else
|
|
return NULL;
|
|
#endif
|
|
}
|
|
|
|
if (dst->family == AF_INET6) {
|
|
#if defined(CONFIG_NET_IPV6)
|
|
return ctx->ipv6.ctx;
|
|
#else
|
|
return NULL;
|
|
#endif
|
|
}
|
|
|
|
if (dst->family == AF_UNSPEC) {
|
|
return ctx->default_ctx->ctx;
|
|
}
|
|
}
|
|
}
|
|
|
|
return NULL;
|
|
}
|
|
#else
|
|
#define select_client_ctx(...) NULL
|
|
#endif /* CONFIG_NET_APP_CLIENT */
|
|
|
|
#if defined(CONFIG_NET_APP_SERVER)
|
|
static inline
|
|
struct net_context *select_server_ctx(struct net_app_ctx *ctx,
|
|
const struct sockaddr *dst)
|
|
{
|
|
if (ctx->proto == IPPROTO_TCP) {
|
|
#if defined(CONFIG_NET_TCP)
|
|
return ctx->server.net_ctx;
|
|
#else
|
|
return NULL;
|
|
#endif
|
|
} else if (ctx->proto == IPPROTO_UDP) {
|
|
if (!dst) {
|
|
if (ctx->is_tls) {
|
|
#if defined(CONFIG_NET_APP_DTLS)
|
|
return ctx->dtls.ctx;
|
|
#else
|
|
return NULL;
|
|
#endif
|
|
} else {
|
|
return ctx->default_ctx->ctx;
|
|
}
|
|
} else {
|
|
if (ctx->is_tls) {
|
|
#if defined(CONFIG_NET_APP_DTLS)
|
|
return ctx->dtls.ctx;
|
|
#else
|
|
return NULL;
|
|
#endif
|
|
}
|
|
|
|
if (dst->family == AF_INET) {
|
|
#if defined(CONFIG_NET_IPV4)
|
|
return ctx->ipv4.ctx;
|
|
#else
|
|
return NULL;
|
|
#endif
|
|
}
|
|
|
|
if (dst->family == AF_INET6) {
|
|
#if defined(CONFIG_NET_IPV6)
|
|
return ctx->ipv6.ctx;
|
|
#else
|
|
return NULL;
|
|
#endif
|
|
}
|
|
|
|
if (dst->family == AF_UNSPEC) {
|
|
return ctx->default_ctx->ctx;
|
|
}
|
|
}
|
|
}
|
|
|
|
return NULL;
|
|
}
|
|
#else
|
|
#define select_server_ctx(...) NULL
|
|
#endif /* CONFIG_NET_APP_SERVER */
|
|
|
|
struct net_context *_net_app_select_net_ctx(struct net_app_ctx *ctx,
|
|
const struct sockaddr *dst)
|
|
{
|
|
if (ctx->app_type == NET_APP_CLIENT) {
|
|
return select_client_ctx(ctx, dst);
|
|
}
|
|
|
|
if (ctx->app_type == NET_APP_SERVER) {
|
|
return select_server_ctx(ctx, dst);
|
|
}
|
|
|
|
return NULL;
|
|
}
|
|
|
|
int net_app_set_cb(struct net_app_ctx *ctx,
|
|
net_app_connect_cb_t connect_cb,
|
|
net_app_recv_cb_t recv_cb,
|
|
net_app_send_cb_t send_cb,
|
|
net_app_close_cb_t close_cb)
|
|
{
|
|
if (!ctx) {
|
|
return -EINVAL;
|
|
}
|
|
|
|
if (!ctx->is_init) {
|
|
return -ENOENT;
|
|
}
|
|
|
|
ctx->cb.connect = connect_cb;
|
|
ctx->cb.recv = recv_cb;
|
|
ctx->cb.send = send_cb;
|
|
ctx->cb.close = close_cb;
|
|
|
|
return 0;
|
|
}
|
|
|
|
static void _app_send(struct net_context *net_ctx,
|
|
int status,
|
|
void *token,
|
|
void *user_data)
|
|
{
|
|
struct net_app_ctx *ctx = user_data;
|
|
|
|
ARG_UNUSED(ctx);
|
|
|
|
#if defined(CONFIG_NET_APP_CLIENT)
|
|
if (ctx->app_type == NET_APP_CLIENT && ctx->cb.send) {
|
|
ctx->cb.send(ctx, status, token, ctx->user_data);
|
|
}
|
|
#endif
|
|
|
|
#if defined(CONFIG_NET_APP_SERVER)
|
|
if (ctx->app_type == NET_APP_SERVER && ctx->cb.send) {
|
|
ctx->cb.send(ctx, status, token, ctx->user_data);
|
|
}
|
|
#endif
|
|
}
|
|
|
|
int net_app_send_pkt(struct net_app_ctx *ctx,
|
|
struct net_pkt *pkt,
|
|
const struct sockaddr *dst,
|
|
socklen_t dst_len,
|
|
s32_t timeout,
|
|
void *user_data_send)
|
|
{
|
|
int ret;
|
|
|
|
if (!ctx) {
|
|
return -EINVAL;
|
|
}
|
|
|
|
if (!ctx->is_init) {
|
|
return -ENOENT;
|
|
}
|
|
|
|
net_pkt_set_appdatalen(pkt, net_buf_frags_len(pkt->frags));
|
|
|
|
if (ctx->proto == IPPROTO_UDP) {
|
|
if (!dst) {
|
|
if (net_pkt_family(pkt) == AF_INET) {
|
|
#if defined(CONFIG_NET_IPV4)
|
|
dst = &ctx->ipv4.remote;
|
|
dst_len = sizeof(struct sockaddr_in);
|
|
#else
|
|
return -EPFNOSUPPORT;
|
|
#endif
|
|
} else {
|
|
if (net_pkt_family(pkt) == AF_INET6) {
|
|
#if defined(CONFIG_NET_IPV6)
|
|
dst = &ctx->ipv6.remote;
|
|
dst_len = sizeof(struct sockaddr_in6);
|
|
#else
|
|
return -EPFNOSUPPORT;
|
|
#endif
|
|
} else {
|
|
return -EPFNOSUPPORT;
|
|
}
|
|
}
|
|
} else {
|
|
if (net_pkt_family(pkt) == AF_INET) {
|
|
#if defined(CONFIG_NET_IPV4)
|
|
net_ipaddr_copy(net_sin(&ctx->ipv4.remote),
|
|
net_sin(dst));
|
|
dst_len = sizeof(struct sockaddr_in);
|
|
#else
|
|
return -EPFNOSUPPORT;
|
|
#endif
|
|
} else {
|
|
if (net_pkt_family(pkt) == AF_INET6) {
|
|
#if defined(CONFIG_NET_IPV6)
|
|
net_ipaddr_copy(
|
|
net_sin6(&ctx->ipv6.remote),
|
|
net_sin6(dst));
|
|
dst_len = sizeof(struct sockaddr_in6);
|
|
#else
|
|
return -EPFNOSUPPORT;
|
|
#endif
|
|
} else {
|
|
return -EPFNOSUPPORT;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
ret = ctx->send_data(pkt, dst, dst_len, _app_send, timeout,
|
|
user_data_send, ctx);
|
|
if (ret < 0) {
|
|
NET_DBG("Cannot send to peer (%d)", ret);
|
|
}
|
|
|
|
return ret;
|
|
}
|
|
|
|
int net_app_send_buf(struct net_app_ctx *ctx,
|
|
u8_t *buf,
|
|
size_t buf_len,
|
|
const struct sockaddr *dst,
|
|
socklen_t dst_len,
|
|
s32_t timeout,
|
|
void *user_data_send)
|
|
{
|
|
struct net_context *net_ctx;
|
|
struct net_pkt *pkt;
|
|
struct net_buf *frag;
|
|
size_t len, pos = 0;
|
|
int ret;
|
|
|
|
if (!ctx) {
|
|
return -EINVAL;
|
|
}
|
|
|
|
if (!ctx->is_init) {
|
|
return -ENOENT;
|
|
}
|
|
|
|
if (!buf_len) {
|
|
return -EMSGSIZE;
|
|
}
|
|
|
|
net_ctx = _net_app_select_net_ctx(ctx, dst);
|
|
if (!net_ctx) {
|
|
return -ENOENT;
|
|
}
|
|
|
|
pkt = net_pkt_get_tx(net_ctx, timeout);
|
|
if (!pkt) {
|
|
return -ENOMEM;
|
|
}
|
|
|
|
net_pkt_set_appdatalen(pkt, buf_len);
|
|
|
|
while (buf_len) {
|
|
frag = net_pkt_get_data(net_ctx, timeout);
|
|
if (!frag) {
|
|
net_pkt_unref(pkt);
|
|
return -ENOMEM;
|
|
}
|
|
|
|
len = net_buf_tailroom(frag);
|
|
if (len >= buf_len) {
|
|
net_buf_add_mem(frag, buf + pos, buf_len);
|
|
net_pkt_frag_add(pkt, frag);
|
|
goto send;
|
|
}
|
|
|
|
net_buf_add_mem(frag, buf + pos, len);
|
|
net_pkt_frag_add(pkt, frag);
|
|
|
|
pos += len;
|
|
buf_len -= len;
|
|
}
|
|
|
|
send:
|
|
ret = ctx->send_data(pkt, dst, dst_len, _app_send, timeout,
|
|
user_data_send, ctx);
|
|
if (ret < 0) {
|
|
NET_DBG("Cannot send to peer (%d)", ret);
|
|
net_pkt_unref(pkt);
|
|
}
|
|
|
|
return ret;
|
|
}
|
|
|
|
struct net_pkt *net_app_get_net_pkt(struct net_app_ctx *ctx,
|
|
sa_family_t family,
|
|
s32_t timeout)
|
|
{
|
|
struct net_context *net_ctx;
|
|
struct sockaddr dst;
|
|
|
|
if (!ctx) {
|
|
return NULL;
|
|
}
|
|
|
|
if (!ctx->is_init) {
|
|
return NULL;
|
|
}
|
|
|
|
dst.family = family;
|
|
|
|
net_ctx = _net_app_select_net_ctx(ctx, &dst);
|
|
if (!net_ctx) {
|
|
return NULL;
|
|
}
|
|
|
|
return net_pkt_get_tx(net_ctx, timeout);
|
|
}
|
|
|
|
struct net_buf *net_app_get_net_buf(struct net_app_ctx *ctx,
|
|
struct net_pkt *pkt,
|
|
s32_t timeout)
|
|
{
|
|
struct net_buf *frag;
|
|
|
|
if (!ctx || !pkt) {
|
|
return NULL;
|
|
}
|
|
|
|
if (!ctx->is_init) {
|
|
return NULL;
|
|
}
|
|
|
|
frag = net_pkt_get_frag(pkt, timeout);
|
|
if (!frag) {
|
|
return NULL;
|
|
}
|
|
|
|
net_pkt_frag_add(pkt, frag);
|
|
|
|
return frag;
|
|
}
|
|
|
|
int net_app_close(struct net_app_ctx *ctx)
|
|
{
|
|
struct net_context *net_ctx;
|
|
|
|
if (!ctx) {
|
|
return -EINVAL;
|
|
}
|
|
|
|
if (!ctx->is_init) {
|
|
return -ENOENT;
|
|
}
|
|
|
|
net_ctx = _net_app_select_net_ctx(ctx, NULL);
|
|
if (!net_ctx) {
|
|
return -EAFNOSUPPORT;
|
|
}
|
|
|
|
if (ctx->cb.close) {
|
|
ctx->cb.close(ctx, 0, ctx->user_data);
|
|
}
|
|
|
|
#if defined(CONFIG_NET_APP_SERVER) && defined(CONFIG_NET_TCP)
|
|
if (ctx->app_type == NET_APP_SERVER) {
|
|
ctx->server.net_ctx = NULL;
|
|
}
|
|
#endif
|
|
|
|
net_context_put(net_ctx);
|
|
|
|
return 0;
|
|
}
|
|
|
|
#if defined(CONFIG_NET_APP_TLS)
|
|
#if defined(MBEDTLS_DEBUG_C) && defined(CONFIG_NET_DEBUG_APP)
|
|
static void my_debug(void *ctx, int level,
|
|
const char *file, int line, const char *str)
|
|
{
|
|
const char *p, *basename;
|
|
int len;
|
|
|
|
ARG_UNUSED(ctx);
|
|
|
|
/* Extract basename from file */
|
|
for (p = basename = file; *p != '\0'; p++) {
|
|
if (*p == '/' || *p == '\\') {
|
|
basename = p + 1;
|
|
}
|
|
|
|
}
|
|
|
|
/* Avoid printing double newlines */
|
|
len = strlen(str);
|
|
if (str[len - 1] == '\n') {
|
|
((char *)str)[len - 1] = '\0';
|
|
}
|
|
|
|
NET_DBG("%s:%04d: |%d| %s", basename, line, level, str);
|
|
}
|
|
#endif /* MBEDTLS_DEBUG_C && CONFIG_NET_DEBUG_APP */
|
|
|
|
static void ssl_sent(struct net_context *context,
|
|
int status, void *token, void *user_data)
|
|
{
|
|
struct net_app_ctx *ctx = user_data;
|
|
|
|
k_sem_give(&ctx->tls.mbedtls.ssl_ctx.tx_sem);
|
|
}
|
|
|
|
/* Send encrypted data */
|
|
int _net_app_ssl_tx(void *context, const unsigned char *buf, size_t size)
|
|
{
|
|
struct net_app_ctx *ctx = context;
|
|
struct net_pkt *send_buf;
|
|
int ret, len;
|
|
|
|
send_buf = net_app_get_net_pkt(ctx, AF_UNSPEC, BUF_ALLOC_TIMEOUT);
|
|
if (!send_buf) {
|
|
return MBEDTLS_ERR_SSL_ALLOC_FAILED;
|
|
}
|
|
|
|
ret = net_pkt_append_all(send_buf, size, (u8_t *)buf,
|
|
BUF_ALLOC_TIMEOUT);
|
|
if (!ret) {
|
|
/* Cannot append data */
|
|
net_pkt_unref(send_buf);
|
|
return 0;
|
|
}
|
|
|
|
len = size;
|
|
|
|
if (ctx->proto == IPPROTO_UDP) {
|
|
#if defined(CONFIG_NET_APP_DTLS)
|
|
if (!ctx->dtls.ctx) {
|
|
net_pkt_unref(send_buf);
|
|
return MBEDTLS_ERR_SSL_INTERNAL_ERROR;
|
|
}
|
|
|
|
ret = net_context_sendto(send_buf,
|
|
&ctx->dtls.ctx->remote,
|
|
sizeof(ctx->dtls.ctx->remote),
|
|
ssl_sent, K_NO_WAIT, NULL, ctx);
|
|
#else
|
|
ret = -EPROTONOSUPPORT;
|
|
#endif
|
|
} else {
|
|
ret = net_context_send(send_buf, ssl_sent, K_NO_WAIT, NULL,
|
|
ctx);
|
|
}
|
|
|
|
if (ret < 0) {
|
|
net_pkt_unref(send_buf);
|
|
return MBEDTLS_ERR_SSL_INTERNAL_ERROR;
|
|
}
|
|
|
|
k_sem_take(&ctx->tls.mbedtls.ssl_ctx.tx_sem, K_FOREVER);
|
|
|
|
return len;
|
|
}
|
|
|
|
/* This gets plain data and then it passes it to TLS handler thread to be
|
|
* encrypted and transmitted to peer. Note that we do not send the data
|
|
* directly here because of the mbedtls stack requirements which are quite
|
|
* high. So no calls to mbedtls from this processing context.
|
|
*/
|
|
int _net_app_tls_sendto(struct net_pkt *pkt,
|
|
const struct sockaddr *dst_addr,
|
|
socklen_t addrlen,
|
|
net_context_send_cb_t cb,
|
|
s32_t timeout,
|
|
void *token,
|
|
void *user_data)
|
|
{
|
|
struct net_app_ctx *ctx = user_data;
|
|
struct net_app_fifo_block *tx_data;
|
|
struct k_mem_block block;
|
|
int ret;
|
|
|
|
ARG_UNUSED(dst_addr);
|
|
ARG_UNUSED(addrlen);
|
|
|
|
if (pkt && !net_pkt_appdatalen(pkt)) {
|
|
return -EINVAL;
|
|
}
|
|
|
|
ret = k_mem_pool_alloc(ctx->tls.pool, &block,
|
|
sizeof(struct net_app_fifo_block),
|
|
BUF_ALLOC_TIMEOUT);
|
|
if (ret < 0) {
|
|
return -ENOMEM;
|
|
}
|
|
|
|
tx_data = block.data;
|
|
tx_data->pkt = pkt;
|
|
tx_data->dir = NET_APP_PKT_TX;
|
|
tx_data->token = token;
|
|
tx_data->cb = cb;
|
|
|
|
/* For freeing memory later */
|
|
memcpy(&tx_data->block, &block, sizeof(struct k_mem_block));
|
|
|
|
k_fifo_put(&ctx->tls.mbedtls.ssl_ctx.tx_rx_fifo, (void *)tx_data);
|
|
|
|
return 0;
|
|
}
|
|
|
|
#if defined(CONFIG_NET_APP_DTLS)
|
|
#if defined(CONFIG_NET_CONTEXT_NET_PKT_POOL)
|
|
static inline void copy_pool_vars(struct net_context *new_context,
|
|
struct net_context *listen_context)
|
|
{
|
|
new_context->tx_slab = listen_context->tx_slab;
|
|
new_context->data_pool = listen_context->data_pool;
|
|
}
|
|
#else
|
|
#define copy_pool_vars(...)
|
|
#endif /* CONFIG_NET_CONTEXT_NET_PKT_POOL */
|
|
|
|
static void dtls_timing_set_delay(void *data, uint32_t int_ms, uint32_t fin_ms)
|
|
{
|
|
struct dtls_timing_context *ctx = data;
|
|
|
|
#if DTLS_EXTRA_DEBUG == 1
|
|
NET_DBG("Setting DTLS delays for %p, int %ums fin %ums",
|
|
ctx, int_ms, fin_ms);
|
|
#endif
|
|
|
|
ctx->int_ms = int_ms;
|
|
ctx->fin_ms = fin_ms;
|
|
|
|
if (fin_ms != 0) {
|
|
ctx->snapshot = k_uptime_get_32();
|
|
}
|
|
}
|
|
|
|
static int dtls_timing_get_delay(void *data)
|
|
{
|
|
struct dtls_timing_context *timing = data;
|
|
unsigned long elapsed_ms;
|
|
|
|
NET_ASSERT(timing);
|
|
|
|
#if DTLS_EXTRA_DEBUG == 1
|
|
NET_DBG("Get DTLS delays for %p, int %ums fin %ums snapshot %d",
|
|
timing, timing->int_ms, timing->fin_ms, timing->snapshot);
|
|
#endif
|
|
|
|
if (timing->fin_ms == 0) {
|
|
return -1;
|
|
}
|
|
|
|
elapsed_ms = k_uptime_get_32() - timing->snapshot;
|
|
|
|
if (elapsed_ms >= timing->fin_ms) {
|
|
return 2;
|
|
}
|
|
|
|
if (elapsed_ms >= timing->int_ms) {
|
|
return 1;
|
|
}
|
|
|
|
return 0;
|
|
}
|
|
|
|
static void dtls_cleanup(struct net_app_ctx *ctx, bool cancel_timer)
|
|
{
|
|
ctx->dtls.fin_timer_cancelled = true;
|
|
|
|
if (cancel_timer) {
|
|
k_delayed_work_cancel(&ctx->dtls.fin_timer);
|
|
}
|
|
|
|
/* It might be that ctx is already cleared so check it here */
|
|
if (ctx->dtls.ctx) {
|
|
net_udp_unregister(ctx->dtls.ctx->conn_handler);
|
|
net_context_put(ctx->dtls.ctx);
|
|
ctx->dtls.ctx = NULL;
|
|
}
|
|
}
|
|
|
|
static void dtls_timeout(struct k_work *work)
|
|
{
|
|
struct net_app_ctx *ctx =
|
|
CONTAINER_OF(work, struct net_app_ctx, dtls.fin_timer);
|
|
|
|
if (!ctx->dtls.fin_timer_cancelled) {
|
|
NET_DBG("Did not receive DTLS traffic in %dms", DTLS_TIMEOUT);
|
|
|
|
dtls_cleanup(ctx, false);
|
|
}
|
|
}
|
|
|
|
enum net_verdict _net_app_dtls_established(struct net_conn *conn,
|
|
struct net_pkt *pkt,
|
|
void *user_data)
|
|
{
|
|
struct net_app_ctx *ctx = user_data;
|
|
struct net_app_fifo_block *rx_data = NULL;
|
|
struct k_mem_block block;
|
|
struct net_buf *frag;
|
|
u16_t offset;
|
|
int ret, len;
|
|
|
|
if (!pkt) {
|
|
return NET_DROP;
|
|
}
|
|
|
|
len = net_pkt_get_len(pkt) - net_pkt_ip_hdr_len(pkt) -
|
|
net_pkt_ipv6_ext_len(pkt) - sizeof(struct net_udp_hdr);
|
|
if (len <= 0) {
|
|
return NET_DROP;
|
|
}
|
|
|
|
net_pkt_set_appdatalen(pkt, len);
|
|
|
|
frag = net_frag_get_pos(pkt, net_pkt_ip_hdr_len(pkt) +
|
|
net_pkt_ipv6_ext_len(pkt) +
|
|
sizeof(struct net_udp_hdr),
|
|
&offset);
|
|
if (frag) {
|
|
net_pkt_set_appdata(pkt, frag->data + offset);
|
|
}
|
|
|
|
ret = k_mem_pool_alloc(ctx->tls.pool, &block,
|
|
sizeof(struct net_app_fifo_block),
|
|
BUF_ALLOC_TIMEOUT);
|
|
if (ret < 0) {
|
|
NET_DBG("Not enough space in DTLS mem pool");
|
|
return NET_DROP;
|
|
}
|
|
|
|
rx_data = block.data;
|
|
rx_data->pkt = pkt;
|
|
rx_data->dir = NET_APP_PKT_RX;
|
|
|
|
/* For freeing memory later */
|
|
memcpy(&rx_data->block, &block, sizeof(struct k_mem_block));
|
|
|
|
NET_DBG("Encrypted DTLS data received in pkt %p", pkt);
|
|
|
|
k_fifo_put(&ctx->tls.mbedtls.ssl_ctx.tx_rx_fifo, (void *)rx_data);
|
|
|
|
ctx->dtls.fin_timer_cancelled = true;
|
|
k_delayed_work_cancel(&ctx->dtls.fin_timer);
|
|
|
|
k_yield();
|
|
|
|
ctx->dtls.fin_timer_cancelled = false;
|
|
k_delayed_work_submit(&ctx->dtls.fin_timer, DTLS_TIMEOUT);
|
|
|
|
return NET_OK;
|
|
}
|
|
|
|
static int accept_dtls(struct net_app_ctx *ctx,
|
|
struct net_context *context,
|
|
struct net_pkt *pkt)
|
|
{
|
|
struct net_context *dtls_context;
|
|
struct net_udp_hdr hdr, *udp_hdr;
|
|
struct sockaddr remote_addr;
|
|
struct sockaddr local_addr;
|
|
socklen_t addrlen;
|
|
int ret;
|
|
|
|
udp_hdr = net_udp_get_hdr(pkt, &hdr);
|
|
if (!udp_hdr) {
|
|
NET_DBG("Dropping invalid pkt %p", pkt);
|
|
goto pkt_drop;
|
|
}
|
|
|
|
/* We create a new context that starts to wait data. */
|
|
ret = net_context_get(net_pkt_family(pkt), SOCK_DGRAM, IPPROTO_UDP,
|
|
&dtls_context);
|
|
if (ret < 0) {
|
|
NET_DBG("Cannot get accepted context, pkt %p dropped", pkt);
|
|
goto pkt_drop;
|
|
}
|
|
|
|
#if defined(CONFIG_NET_IPV6)
|
|
if (net_context_get_family(context) == AF_INET6) {
|
|
struct sockaddr_in6 *local_addr6 = net_sin6(&local_addr);
|
|
struct sockaddr_in6 *remote_addr6 = net_sin6(&remote_addr);
|
|
|
|
remote_addr6->sin6_family = AF_INET6;
|
|
local_addr6->sin6_family = AF_INET6;
|
|
|
|
local_addr6->sin6_port = udp_hdr->dst_port;
|
|
remote_addr6->sin6_port = udp_hdr->src_port;
|
|
|
|
net_ipaddr_copy(&local_addr6->sin6_addr,
|
|
&NET_IPV6_HDR(pkt)->dst);
|
|
net_ipaddr_copy(&remote_addr6->sin6_addr,
|
|
&NET_IPV6_HDR(pkt)->src);
|
|
addrlen = sizeof(struct sockaddr_in6);
|
|
} else
|
|
#endif /* CONFIG_NET_IPV6 */
|
|
|
|
#if defined(CONFIG_NET_IPV4)
|
|
if (net_context_get_family(context) == AF_INET) {
|
|
struct sockaddr_in *local_addr4 = net_sin(&local_addr);
|
|
struct sockaddr_in *remote_addr4 = net_sin(&remote_addr);
|
|
|
|
remote_addr4->sin_family = AF_INET;
|
|
local_addr4->sin_family = AF_INET;
|
|
|
|
local_addr4->sin_port = udp_hdr->dst_port;
|
|
remote_addr4->sin_port = udp_hdr->src_port;
|
|
|
|
net_ipaddr_copy(&local_addr4->sin_addr,
|
|
&NET_IPV4_HDR(pkt)->dst);
|
|
net_ipaddr_copy(&remote_addr4->sin_addr,
|
|
&NET_IPV4_HDR(pkt)->src);
|
|
addrlen = sizeof(struct sockaddr_in);
|
|
} else
|
|
#endif /* CONFIG_NET_IPV4 */
|
|
{
|
|
NET_ASSERT_INFO(false, "Invalid protocol family %d",
|
|
net_context_get_family(context));
|
|
goto ctx_drop;
|
|
}
|
|
|
|
ret = net_context_bind(dtls_context, &local_addr, sizeof(local_addr));
|
|
if (ret < 0) {
|
|
NET_DBG("Cannot bind accepted DTLS context");
|
|
goto ctx_drop;
|
|
}
|
|
|
|
dtls_context->flags |= NET_CONTEXT_REMOTE_ADDR_SET;
|
|
|
|
memcpy(&dtls_context->remote, &remote_addr, sizeof(remote_addr));
|
|
|
|
ret = net_udp_register(&dtls_context->remote,
|
|
&local_addr,
|
|
ntohs(net_sin(&dtls_context->remote)->sin_port),
|
|
ntohs(net_sin(&local_addr)->sin_port),
|
|
_net_app_dtls_established,
|
|
ctx,
|
|
&dtls_context->conn_handler);
|
|
if (ret < 0) {
|
|
NET_DBG("Cannot register accepted DTLS handler (%d)", ret);
|
|
goto ctx_drop;
|
|
}
|
|
|
|
copy_pool_vars(dtls_context, context);
|
|
|
|
net_context_set_state(dtls_context, NET_CONTEXT_CONNECTED);
|
|
|
|
NET_DBG("New DTLS connection %p accepted", dtls_context);
|
|
|
|
ctx->dtls.ctx = dtls_context;
|
|
|
|
ctx->dtls.fin_timer_cancelled = false;
|
|
k_delayed_work_submit(&ctx->dtls.fin_timer, DTLS_TIMEOUT);
|
|
|
|
return 0;
|
|
|
|
ctx_drop:
|
|
net_context_unref(dtls_context);
|
|
|
|
pkt_drop:
|
|
net_pkt_unref(pkt);
|
|
|
|
return -ECONNABORTED;
|
|
}
|
|
#else /* CONFIG_NET_APP_DTLS */
|
|
#define dtls_cleanup(...)
|
|
#endif /* CONFIG_NET_APP_DTLS */
|
|
|
|
/* Receive encrypted data from network. Put that data into fifo
|
|
* that will be read by tls thread.
|
|
*/
|
|
void _net_app_tls_received(struct net_context *context,
|
|
struct net_pkt *pkt,
|
|
int status,
|
|
void *user_data)
|
|
{
|
|
struct net_app_ctx *ctx = user_data;
|
|
struct net_app_fifo_block *rx_data = NULL;
|
|
struct k_mem_block block;
|
|
int ret;
|
|
|
|
ARG_UNUSED(context);
|
|
ARG_UNUSED(status);
|
|
|
|
if (pkt && !net_pkt_appdatalen(pkt)) {
|
|
net_pkt_unref(pkt);
|
|
return;
|
|
}
|
|
|
|
#if defined(CONFIG_NET_APP_DTLS)
|
|
/* Client connections that are initiated by us, are passed through
|
|
* as is.
|
|
*/
|
|
if (ctx->proto == IPPROTO_UDP && ctx->app_type == NET_APP_SERVER) {
|
|
if (ctx->dtls.ctx) {
|
|
/* There will be a separate handler for these DTLS
|
|
* packets so if they are arriving here, then that is
|
|
* an error.
|
|
*/
|
|
NET_DBG("DTLS context already created, pkt %p dropped",
|
|
pkt);
|
|
net_pkt_unref(pkt);
|
|
return;
|
|
} else {
|
|
if (ctx->dtls.fin_timer_cancelled) {
|
|
if (pkt) {
|
|
net_pkt_unref(pkt);
|
|
pkt = NULL;
|
|
}
|
|
|
|
ctx->dtls.fin_timer_cancelled = false;
|
|
|
|
goto dtls_disconnect;
|
|
}
|
|
|
|
ret = accept_dtls(ctx, context, pkt);
|
|
if (ret < 0) {
|
|
NET_DBG("Cannot accept new DTLS "
|
|
"connection (%d)", ret);
|
|
net_pkt_unref(pkt);
|
|
return;
|
|
}
|
|
|
|
/* The first packet is passed as is in below code,
|
|
* subsequent packets are handled by dtls_established()
|
|
*/
|
|
}
|
|
}
|
|
dtls_disconnect:
|
|
#endif /* CONFIG_NET_APP_DTLS */
|
|
|
|
ret = k_mem_pool_alloc(ctx->tls.pool, &block,
|
|
sizeof(struct net_app_fifo_block),
|
|
BUF_ALLOC_TIMEOUT);
|
|
if (ret < 0) {
|
|
if (pkt) {
|
|
net_pkt_unref(pkt);
|
|
}
|
|
|
|
NET_DBG("Not enough space in TLS mem pool");
|
|
return;
|
|
}
|
|
|
|
rx_data = block.data;
|
|
rx_data->pkt = pkt;
|
|
rx_data->dir = NET_APP_PKT_RX;
|
|
|
|
/* For freeing memory later */
|
|
memcpy(&rx_data->block, &block, sizeof(struct k_mem_block));
|
|
|
|
NET_DBG("Encrypted data received in pkt %p", pkt);
|
|
|
|
k_fifo_put(&ctx->tls.mbedtls.ssl_ctx.tx_rx_fifo, (void *)rx_data);
|
|
}
|
|
|
|
static int tls_sendto(struct net_app_ctx *ctx,
|
|
struct net_app_fifo_block *tx_data)
|
|
{
|
|
u16_t len;
|
|
int ret;
|
|
|
|
len = net_pkt_appdatalen(tx_data->pkt);
|
|
if (len == 0) {
|
|
ret = -EINVAL;
|
|
goto out;
|
|
}
|
|
|
|
ret = net_frag_linearize(ctx->tls.request_buf,
|
|
ctx->tls.request_buf_len,
|
|
tx_data->pkt,
|
|
net_pkt_ip_hdr_len(tx_data->pkt),
|
|
len);
|
|
if (ret < 0) {
|
|
NET_DBG("Cannot linearize send data (%d)", ret);
|
|
goto out;
|
|
}
|
|
|
|
if (ret != len) {
|
|
NET_DBG("Linear copy error (%u vs %d)", len, ret);
|
|
ret = -EINVAL;
|
|
goto out;
|
|
}
|
|
|
|
do {
|
|
ret = mbedtls_ssl_write(&ctx->tls.mbedtls.ssl,
|
|
ctx->tls.request_buf, len);
|
|
if (ret == MBEDTLS_ERR_NET_CONN_RESET) {
|
|
_net_app_print_error(
|
|
"peer closed the connection -0x%x", ret);
|
|
goto out;
|
|
}
|
|
|
|
if (ret != MBEDTLS_ERR_SSL_WANT_READ &&
|
|
ret != MBEDTLS_ERR_SSL_WANT_WRITE) {
|
|
if (ret < 0) {
|
|
_net_app_print_error(
|
|
"mbedtls_ssl_write returned -0x%x",
|
|
ret);
|
|
goto out;
|
|
}
|
|
}
|
|
} while (ret <= 0);
|
|
|
|
out:
|
|
if (tx_data->cb) {
|
|
tx_data->cb(net_pkt_context(tx_data->pkt), ret,
|
|
tx_data->token, ctx);
|
|
}
|
|
|
|
net_pkt_unref(tx_data->pkt);
|
|
|
|
return ret;
|
|
}
|
|
|
|
#if defined(CONFIG_NET_APP_DTLS)
|
|
static inline void set_remote_endpoint(struct net_app_ctx *ctx,
|
|
struct net_pkt *pkt)
|
|
{
|
|
struct net_udp_hdr hdr, *udp_hdr;
|
|
|
|
udp_hdr = net_udp_get_hdr(pkt, &hdr);
|
|
if (!udp_hdr) {
|
|
return;
|
|
}
|
|
|
|
if (net_pkt_family(pkt) == AF_INET) {
|
|
#if defined(CONFIG_NET_IPV4)
|
|
net_sin(&ctx->ipv4.remote)->sin_port = udp_hdr->src_port;
|
|
|
|
net_ipaddr_copy(&net_sin(&ctx->ipv4.remote)->sin_addr,
|
|
&NET_IPV4_HDR(pkt)->src);
|
|
#endif
|
|
return;
|
|
}
|
|
|
|
if (net_pkt_family(pkt) == AF_INET6) {
|
|
#if defined(CONFIG_NET_IPV6)
|
|
net_sin6(&ctx->ipv6.remote)->sin6_port = udp_hdr->src_port;
|
|
|
|
net_ipaddr_copy(&net_sin6(&ctx->ipv6.remote)->sin6_addr,
|
|
&NET_IPV6_HDR(pkt)->src);
|
|
#endif
|
|
return;
|
|
}
|
|
}
|
|
#endif /* CONFIG_NET_APP_DTLS */
|
|
|
|
/* This will copy data from received net_pkt buf into mbedtls internal buffers.
|
|
*/
|
|
int _net_app_ssl_mux(void *context, unsigned char *buf, size_t size)
|
|
{
|
|
struct net_app_ctx *ctx = context;
|
|
struct net_app_fifo_block *rx_data;
|
|
u16_t read_bytes;
|
|
u8_t *ptr;
|
|
int pos;
|
|
int len;
|
|
int ret = 0;
|
|
|
|
if (!ctx->tls.mbedtls.ssl_ctx.frag) {
|
|
again:
|
|
rx_data = k_fifo_get(&ctx->tls.mbedtls.ssl_ctx.tx_rx_fifo,
|
|
K_FOREVER);
|
|
if (!rx_data->pkt) {
|
|
NET_DBG("Closing %p connection", ctx);
|
|
k_mem_pool_free(&rx_data->block);
|
|
return -EIO;
|
|
}
|
|
|
|
NET_DBG("%s encrypted data in pkt %p",
|
|
rx_data->dir == NET_APP_PKT_TX ? "Sending" :
|
|
"Receiving",
|
|
rx_data->pkt);
|
|
|
|
/* If the fifo contains something we need to send, then try
|
|
* to send it here and then go back waiting more data.
|
|
*/
|
|
if (rx_data->dir == NET_APP_PKT_TX) {
|
|
tls_sendto(ctx, rx_data);
|
|
k_mem_pool_free(&rx_data->block);
|
|
goto again;
|
|
}
|
|
|
|
ctx->tls.mbedtls.ssl_ctx.rx_pkt = rx_data->pkt;
|
|
|
|
k_mem_pool_free(&rx_data->block);
|
|
|
|
read_bytes = net_pkt_appdatalen(
|
|
ctx->tls.mbedtls.ssl_ctx.rx_pkt);
|
|
|
|
ctx->tls.mbedtls.ssl_ctx.remaining = read_bytes;
|
|
ctx->tls.mbedtls.ssl_ctx.frag =
|
|
ctx->tls.mbedtls.ssl_ctx.rx_pkt->frags;
|
|
|
|
ptr = net_pkt_appdata(ctx->tls.mbedtls.ssl_ctx.rx_pkt);
|
|
len = ptr - ctx->tls.mbedtls.ssl_ctx.frag->data;
|
|
|
|
if (len > ctx->tls.mbedtls.ssl_ctx.frag->size) {
|
|
NET_ERR("Buf overflow (%d > %u)", len,
|
|
ctx->tls.mbedtls.ssl_ctx.frag->size);
|
|
return -EINVAL;
|
|
}
|
|
|
|
/* This will get rid of IP header */
|
|
net_buf_pull(ctx->tls.mbedtls.ssl_ctx.frag, len);
|
|
} else {
|
|
read_bytes = ctx->tls.mbedtls.ssl_ctx.remaining;
|
|
ptr = ctx->tls.mbedtls.ssl_ctx.frag->data;
|
|
}
|
|
|
|
len = ctx->tls.mbedtls.ssl_ctx.frag->len;
|
|
pos = 0;
|
|
if (read_bytes > size) {
|
|
while (ctx->tls.mbedtls.ssl_ctx.frag) {
|
|
read_bytes = len < (size - pos) ? len : (size - pos);
|
|
|
|
#if RX_EXTRA_DEBUG == 1
|
|
NET_DBG("Copying %d bytes", read_bytes);
|
|
#endif
|
|
|
|
memcpy(buf + pos, ptr, read_bytes);
|
|
|
|
pos += read_bytes;
|
|
if (pos < size) {
|
|
ctx->tls.mbedtls.ssl_ctx.frag =
|
|
ctx->tls.mbedtls.ssl_ctx.frag->frags;
|
|
ptr = ctx->tls.mbedtls.ssl_ctx.frag->data;
|
|
len = ctx->tls.mbedtls.ssl_ctx.frag->len;
|
|
} else {
|
|
if (read_bytes == len) {
|
|
ctx->tls.mbedtls.ssl_ctx.frag =
|
|
ctx->tls.mbedtls.ssl_ctx.frag->frags;
|
|
} else {
|
|
net_buf_pull(
|
|
ctx->tls.mbedtls.ssl_ctx.frag,
|
|
read_bytes);
|
|
}
|
|
|
|
ctx->tls.mbedtls.ssl_ctx.remaining -= size;
|
|
return size;
|
|
}
|
|
}
|
|
} else {
|
|
while (ctx->tls.mbedtls.ssl_ctx.frag) {
|
|
#if RX_EXTRA_DEBUG == 1
|
|
NET_DBG("Copying all %d bytes", len);
|
|
#endif
|
|
|
|
memcpy(buf + pos, ptr, len);
|
|
|
|
pos += len;
|
|
ctx->tls.mbedtls.ssl_ctx.frag =
|
|
ctx->tls.mbedtls.ssl_ctx.frag->frags;
|
|
if (!ctx->tls.mbedtls.ssl_ctx.frag) {
|
|
break;
|
|
}
|
|
|
|
ptr = ctx->tls.mbedtls.ssl_ctx.frag->data;
|
|
len = ctx->tls.mbedtls.ssl_ctx.frag->len;
|
|
}
|
|
|
|
#if defined(CONFIG_NET_APP_DTLS)
|
|
if (ctx->proto == IPPROTO_UDP) {
|
|
set_remote_endpoint(ctx,
|
|
ctx->tls.mbedtls.ssl_ctx.rx_pkt);
|
|
}
|
|
#endif /* CONFIG_NET_APP_DTLS */
|
|
|
|
net_pkt_unref(ctx->tls.mbedtls.ssl_ctx.rx_pkt);
|
|
ctx->tls.mbedtls.ssl_ctx.rx_pkt = NULL;
|
|
ctx->tls.mbedtls.ssl_ctx.frag = NULL;
|
|
ctx->tls.mbedtls.ssl_ctx.remaining = 0;
|
|
|
|
if (read_bytes != pos) {
|
|
return -EIO;
|
|
}
|
|
|
|
ret = read_bytes;
|
|
}
|
|
|
|
return ret;
|
|
}
|
|
|
|
int _net_app_entropy_source(void *data, unsigned char *output, size_t len,
|
|
size_t *olen)
|
|
{
|
|
u32_t seed;
|
|
|
|
ARG_UNUSED(data);
|
|
|
|
seed = sys_rand32_get();
|
|
|
|
if (len > sizeof(seed)) {
|
|
len = sizeof(seed);
|
|
}
|
|
|
|
memcpy(output, &seed, len);
|
|
|
|
*olen = len;
|
|
return 0;
|
|
}
|
|
|
|
int _net_app_ssl_mainloop(struct net_app_ctx *ctx)
|
|
{
|
|
size_t len;
|
|
int ret;
|
|
|
|
reset:
|
|
mbedtls_ssl_session_reset(&ctx->tls.mbedtls.ssl);
|
|
|
|
#if defined(CONFIG_NET_APP_DTLS)
|
|
mbedtls_ssl_set_timer_cb(&ctx->tls.mbedtls.ssl,
|
|
&ctx->tls.mbedtls.timing_ctx,
|
|
dtls_timing_set_delay,
|
|
dtls_timing_get_delay);
|
|
|
|
#if defined(CONFIG_NET_APP_SERVER)
|
|
if (ctx->app_type == NET_APP_SERVER) {
|
|
ctx->tls.mbedtls.ssl_ctx.client_id =
|
|
ctx->tls.mbedtls.ssl_ctx.remaining;
|
|
|
|
ret = mbedtls_ssl_set_client_transport_id(
|
|
&ctx->tls.mbedtls.ssl,
|
|
&ctx->tls.mbedtls.ssl_ctx.client_id,
|
|
sizeof(char));
|
|
if (ret != 0) {
|
|
_net_app_print_error(
|
|
"mbedtls_ssl_set_client_transport_id "
|
|
" returned -0x%x\n\n", ret);
|
|
goto close;
|
|
}
|
|
}
|
|
#endif /* CONFIG_NET_APP_SERVER */
|
|
#endif /* CONFIG_NET_APP_DTLS */
|
|
|
|
mbedtls_ssl_set_bio(&ctx->tls.mbedtls.ssl, ctx,
|
|
_net_app_ssl_tx, _net_app_ssl_mux, NULL);
|
|
|
|
/* SSL handshake. The ssl_rx() function will be called next by
|
|
* mbedtls library. The ssl_rx() will block and wait that data is
|
|
* received by ssl_received() and passed to it via fifo. After
|
|
* receiving the data, this function will then proceed with secure
|
|
* connection establishment.
|
|
*/
|
|
/* Waiting SSL handshake */
|
|
do {
|
|
ret = mbedtls_ssl_handshake(&ctx->tls.mbedtls.ssl);
|
|
if (ret != MBEDTLS_ERR_SSL_WANT_READ &&
|
|
ret != MBEDTLS_ERR_SSL_WANT_WRITE) {
|
|
if (ret < 0) {
|
|
goto close;
|
|
}
|
|
}
|
|
} while (ret != 0);
|
|
|
|
/* We call the connect cb only once for each connection. The TLS
|
|
* might require new handshakes etc, but application does not need
|
|
* to care about that.
|
|
*/
|
|
if (!ctx->tls.connect_cb_called && ctx->cb.connect) {
|
|
NET_DBG("Calling connect cb for ctx %p", ctx);
|
|
ctx->cb.connect(ctx, 0, ctx->user_data);
|
|
ctx->tls.connect_cb_called = true;
|
|
}
|
|
|
|
do {
|
|
again:
|
|
len = ctx->tls.request_buf_len - 1;
|
|
memset(ctx->tls.request_buf, 0, ctx->tls.request_buf_len);
|
|
|
|
ret = mbedtls_ssl_read(&ctx->tls.mbedtls.ssl,
|
|
ctx->tls.request_buf, len);
|
|
if (ret == MBEDTLS_ERR_SSL_WANT_READ ||
|
|
ret == MBEDTLS_ERR_SSL_WANT_WRITE) {
|
|
continue;
|
|
}
|
|
|
|
if (ret <= 0) {
|
|
switch (ret) {
|
|
case MBEDTLS_ERR_SSL_PEER_CLOSE_NOTIFY:
|
|
NET_DBG("Connection was closed gracefully");
|
|
goto close;
|
|
|
|
case MBEDTLS_ERR_NET_CONN_RESET:
|
|
NET_DBG("Connection was reset by peer");
|
|
break;
|
|
|
|
case -EIO:
|
|
break;
|
|
|
|
default:
|
|
_net_app_print_error(
|
|
"mbedtls_ssl_read returned -0x%x",
|
|
ret);
|
|
break;
|
|
}
|
|
|
|
goto close;
|
|
}
|
|
|
|
if (ctx->cb.recv) {
|
|
struct net_pkt *pkt;
|
|
int len = ret;
|
|
|
|
pkt = net_pkt_get_rx(_net_app_select_net_ctx(ctx, NULL),
|
|
BUF_ALLOC_TIMEOUT);
|
|
if (!pkt) {
|
|
ret = -ENOMEM;
|
|
goto close;
|
|
}
|
|
|
|
ret = net_pkt_append_all(pkt, len,
|
|
ctx->tls.request_buf,
|
|
BUF_ALLOC_TIMEOUT);
|
|
if (!ret) {
|
|
/* Not all data was appended */
|
|
net_pkt_unref(pkt);
|
|
ret = -ENOMEM;
|
|
goto close;
|
|
}
|
|
|
|
net_pkt_set_appdatalen(pkt, len);
|
|
net_pkt_set_appdata(pkt, pkt->frags->data);
|
|
|
|
ctx->cb.recv(ctx, pkt, 0, ctx->user_data);
|
|
|
|
goto again;
|
|
}
|
|
|
|
} while (ret < 0);
|
|
|
|
/* Read another message */
|
|
goto reset;
|
|
|
|
close:
|
|
/* The -EIO code means that the connection was closed. The error
|
|
* value is not known by mbedtls so do not print info about it.
|
|
*/
|
|
if (ret != -EIO) {
|
|
_net_app_print_error("Closing connection -0x%x", ret);
|
|
}
|
|
|
|
#if defined(CONFIG_NET_APP_DTLS)
|
|
if (ctx->proto == IPPROTO_UDP && ctx->dtls.ctx) {
|
|
NET_DBG("Releasing DTLS context %p", ctx->dtls.ctx);
|
|
|
|
dtls_cleanup(ctx, true);
|
|
}
|
|
#endif /* CONFIG_NET_APP_DTLS */
|
|
|
|
return ret;
|
|
}
|
|
|
|
int _net_app_tls_init(struct net_app_ctx *ctx, int client_or_server)
|
|
{
|
|
enum net_sock_type sock_type;
|
|
int ret;
|
|
|
|
k_fifo_init(&ctx->tls.mbedtls.ssl_ctx.tx_rx_fifo);
|
|
k_sem_init(&ctx->tls.mbedtls.ssl_ctx.tx_sem, 0, UINT_MAX);
|
|
|
|
mbedtls_platform_set_printf(printk);
|
|
|
|
#if defined(MBEDTLS_X509_CRT_PARSE_C)
|
|
if (client_or_server == MBEDTLS_SSL_IS_SERVER) {
|
|
#if defined(CONFIG_NET_APP_SERVER)
|
|
mbedtls_x509_crt_init(&ctx->tls.mbedtls.srvcert);
|
|
#endif
|
|
} else {
|
|
#if defined(CONFIG_NET_APP_CLIENT)
|
|
mbedtls_x509_crt_init(&ctx->tls.mbedtls.ca_cert);
|
|
#endif
|
|
}
|
|
#endif /* MBEDTLS_X509_CRT_PARSE_C */
|
|
|
|
#if defined(CONFIG_NET_APP_SERVER)
|
|
if (client_or_server == MBEDTLS_SSL_IS_SERVER) {
|
|
mbedtls_pk_init(&ctx->tls.mbedtls.pkey);
|
|
}
|
|
#endif
|
|
|
|
mbedtls_ssl_init(&ctx->tls.mbedtls.ssl);
|
|
mbedtls_ssl_config_init(&ctx->tls.mbedtls.conf);
|
|
mbedtls_entropy_init(&ctx->tls.mbedtls.entropy);
|
|
mbedtls_ctr_drbg_init(&ctx->tls.mbedtls.ctr_drbg);
|
|
|
|
#if defined(MBEDTLS_DEBUG_C) && defined(CONFIG_NET_DEBUG_APP)
|
|
mbedtls_debug_set_threshold(DEBUG_THRESHOLD);
|
|
mbedtls_ssl_conf_dbg(&ctx->tls.mbedtls.conf, my_debug, NULL);
|
|
#endif
|
|
|
|
/* Seed the RNG */
|
|
mbedtls_entropy_add_source(&ctx->tls.mbedtls.entropy,
|
|
ctx->tls.mbedtls.entropy_src_cb,
|
|
NULL,
|
|
MBEDTLS_ENTROPY_MAX_GATHER,
|
|
MBEDTLS_ENTROPY_SOURCE_STRONG);
|
|
|
|
ret = mbedtls_ctr_drbg_seed(
|
|
&ctx->tls.mbedtls.ctr_drbg,
|
|
mbedtls_entropy_func,
|
|
&ctx->tls.mbedtls.entropy,
|
|
(const unsigned char *)ctx->tls.mbedtls.personalization_data,
|
|
ctx->tls.mbedtls.personalization_data_len);
|
|
if (ret != 0) {
|
|
_net_app_print_error("mbedtls_ctr_drbg_seed returned -0x%x",
|
|
ret);
|
|
goto exit;
|
|
}
|
|
|
|
if (ctx->sock_type == SOCK_DGRAM) {
|
|
sock_type = MBEDTLS_SSL_TRANSPORT_DATAGRAM;
|
|
} else {
|
|
sock_type = MBEDTLS_SSL_TRANSPORT_STREAM;
|
|
}
|
|
|
|
/* Setup SSL defaults etc. */
|
|
ret = mbedtls_ssl_config_defaults(&ctx->tls.mbedtls.conf,
|
|
client_or_server,
|
|
sock_type,
|
|
MBEDTLS_SSL_PRESET_DEFAULT);
|
|
if (ret != 0) {
|
|
_net_app_print_error("mbedtls_ssl_config_defaults "
|
|
"returned -0x%x", ret);
|
|
goto exit;
|
|
}
|
|
|
|
mbedtls_ssl_conf_rng(&ctx->tls.mbedtls.conf,
|
|
mbedtls_ctr_drbg_random,
|
|
&ctx->tls.mbedtls.ctr_drbg);
|
|
|
|
#if defined(CONFIG_NET_APP_DTLS)
|
|
if (sock_type == MBEDTLS_SSL_TRANSPORT_DATAGRAM) {
|
|
ret = mbedtls_ssl_cookie_setup(&ctx->tls.mbedtls.cookie_ctx,
|
|
mbedtls_ctr_drbg_random,
|
|
&ctx->tls.mbedtls.ctr_drbg);
|
|
if (ret != 0) {
|
|
_net_app_print_error("mbedtls_ssl_cookie_setup "
|
|
"returned -0x%x", ret);
|
|
goto exit;
|
|
}
|
|
|
|
mbedtls_ssl_conf_dtls_cookies(&ctx->tls.mbedtls.conf,
|
|
mbedtls_ssl_cookie_write,
|
|
mbedtls_ssl_cookie_check,
|
|
&ctx->tls.mbedtls.cookie_ctx);
|
|
|
|
k_delayed_work_init(&ctx->dtls.fin_timer, dtls_timeout);
|
|
}
|
|
#endif /* CONFIG_NET_APP_DTLS */
|
|
|
|
if (client_or_server == MBEDTLS_SSL_IS_SERVER) {
|
|
/* Load the certificates and private RSA key. This needs to be
|
|
* done by the user so we call a callback that user must have
|
|
* provided.
|
|
*/
|
|
#if defined(CONFIG_NET_APP_SERVER)
|
|
ret = ctx->tls.mbedtls.cert_cb(ctx, &ctx->tls.mbedtls.srvcert,
|
|
&ctx->tls.mbedtls.pkey);
|
|
if (ret != 0) {
|
|
goto exit;
|
|
}
|
|
#endif
|
|
} else {
|
|
#if defined(CONFIG_NET_APP_CLIENT)
|
|
ret = ctx->tls.mbedtls.ca_cert_cb(ctx,
|
|
&ctx->tls.mbedtls.ca_cert);
|
|
if (ret != 0) {
|
|
goto exit;
|
|
}
|
|
#endif
|
|
}
|
|
|
|
#if defined(MBEDTLS_X509_CRT_PARSE_C) && defined(CONFIG_NET_APP_SERVER)
|
|
if (client_or_server == MBEDTLS_SSL_IS_SERVER) {
|
|
mbedtls_ssl_conf_ca_chain(&ctx->tls.mbedtls.conf,
|
|
ctx->tls.mbedtls.srvcert.next,
|
|
NULL);
|
|
|
|
ret = mbedtls_ssl_conf_own_cert(&ctx->tls.mbedtls.conf,
|
|
&ctx->tls.mbedtls.srvcert,
|
|
&ctx->tls.mbedtls.pkey);
|
|
if (ret != 0) {
|
|
_net_app_print_error("mbedtls_ssl_conf_own_cert "
|
|
"returned -0x%x", ret);
|
|
goto exit;
|
|
}
|
|
}
|
|
#endif /* MBEDTLS_X509_CRT_PARSE_C */
|
|
|
|
ret = mbedtls_ssl_setup(&ctx->tls.mbedtls.ssl,
|
|
&ctx->tls.mbedtls.conf);
|
|
if (ret != 0) {
|
|
_net_app_print_error("mbedtls_ssl_setup returned -0x%x", ret);
|
|
goto exit;
|
|
}
|
|
|
|
#if defined(MBEDTLS_X509_CRT_PARSE_C) && defined(CONFIG_NET_APP_CLIENT)
|
|
if (client_or_server == MBEDTLS_SSL_IS_CLIENT &&
|
|
ctx->tls.cert_host) {
|
|
ret = mbedtls_ssl_set_hostname(&ctx->tls.mbedtls.ssl,
|
|
ctx->tls.cert_host);
|
|
if (ret != 0) {
|
|
_net_app_print_error(
|
|
"mbedtls_ssl_set_hostname returned -0x%x",
|
|
ret);
|
|
goto exit;
|
|
}
|
|
}
|
|
#endif
|
|
|
|
NET_DBG("SSL %s setup done",
|
|
client_or_server == MBEDTLS_SSL_IS_CLIENT ? "client" :
|
|
"server");
|
|
|
|
exit:
|
|
/* The mbedtls resources are freed by _net_app_tls_handler_stop()
|
|
* which is called if this routine returns < 0
|
|
*/
|
|
return ret;
|
|
}
|
|
|
|
void _net_app_tls_handler_stop(struct net_app_ctx *ctx)
|
|
{
|
|
mbedtls_ssl_free(&ctx->tls.mbedtls.ssl);
|
|
mbedtls_ssl_config_free(&ctx->tls.mbedtls.conf);
|
|
mbedtls_ctr_drbg_free(&ctx->tls.mbedtls.ctr_drbg);
|
|
mbedtls_entropy_free(&ctx->tls.mbedtls.entropy);
|
|
|
|
/* Empty the fifo just in case there is any received packets
|
|
* still there.
|
|
*/
|
|
while (1) {
|
|
struct net_app_fifo_block *tx_rx_data;
|
|
|
|
tx_rx_data = k_fifo_get(&ctx->tls.mbedtls.ssl_ctx.tx_rx_fifo,
|
|
K_NO_WAIT);
|
|
if (!tx_rx_data) {
|
|
break;
|
|
}
|
|
|
|
net_pkt_unref(tx_rx_data->pkt);
|
|
|
|
k_mem_pool_free(&tx_rx_data->block);
|
|
}
|
|
|
|
dtls_cleanup(ctx, true);
|
|
|
|
NET_DBG("TLS thread %p stopped", ctx->tls.tid);
|
|
|
|
k_thread_abort(ctx->tls.tid);
|
|
ctx->tls.tid = 0;
|
|
}
|
|
#endif /* CONFIG_NET_APP_TLS */
|
|
|