diff --git a/include/grpc/impl/codegen/grpc_types.h b/include/grpc/impl/codegen/grpc_types.h
index af3d0c219165d15747afeb92905dc41421f42c32..cd3cae71b6a67ea086d298134f481411ea621d43 100644
--- a/include/grpc/impl/codegen/grpc_types.h
+++ b/include/grpc/impl/codegen/grpc_types.h
@@ -154,6 +154,8 @@ typedef struct {
#define GRPC_SSL_TARGET_NAME_OVERRIDE_ARG "grpc.ssl_target_name_override"
/* Maximum metadata size */
#define GRPC_ARG_MAX_METADATA_SIZE "grpc.max_metadata_size"
+/** If non-zero, allow the use of SO_REUSEPORT if it's available (default 1) */
+#define GRPC_ARG_ALLOW_REUSEPORT "grpc.so_reuseport"
/** Result of a grpc call. If the caller satisfies the prerequisites of a
particular operation, the grpc_call_error returned will be GRPC_CALL_OK.
diff --git a/src/core/ext/transport/chttp2/server/insecure/server_chttp2.c b/src/core/ext/transport/chttp2/server/insecure/server_chttp2.c
index d1a58b66211177cc8909f5051cf4324c7d546565..08a09a933a67ab86fffc6403c185cfb18a397e89 100644
--- a/src/core/ext/transport/chttp2/server/insecure/server_chttp2.c
+++ b/src/core/ext/transport/chttp2/server/insecure/server_chttp2.c
@@ -102,7 +102,8 @@ int grpc_server_add_insecure_http2_port(grpc_server *server, const char *addr) {
goto error;
}
- err = grpc_tcp_server_create(NULL, &tcp);
+ err =
+ grpc_tcp_server_create(NULL, grpc_server_get_channel_args(server), &tcp);
if (err != GRPC_ERROR_NONE) {
goto error;
}
diff --git a/src/core/ext/transport/chttp2/server/secure/server_secure_chttp2.c b/src/core/ext/transport/chttp2/server/secure/server_secure_chttp2.c
index 629959b7c03d0c41298d002c4f5f7080ffa2e0d5..64162cc785775d899b2c0fadb0ec64e237b8fcf9 100644
--- a/src/core/ext/transport/chttp2/server/secure/server_secure_chttp2.c
+++ b/src/core/ext/transport/chttp2/server/secure/server_secure_chttp2.c
@@ -208,7 +208,8 @@ int grpc_server_add_secure_http2_port(grpc_server *server, const char *addr,
state = gpr_malloc(sizeof(*state));
memset(state, 0, sizeof(*state));
grpc_closure_init(&state->destroy_closure, destroy_done, state);
- err = grpc_tcp_server_create(&state->destroy_closure, &tcp);
+ err = grpc_tcp_server_create(&state->destroy_closure,
+ grpc_server_get_channel_args(server), &tcp);
if (err != GRPC_ERROR_NONE) {
goto error;
}
diff --git a/src/core/lib/iomgr/socket_utils_common_posix.c b/src/core/lib/iomgr/socket_utils_common_posix.c
index d1721c910c41af3c543c372f63a38e72c27d207e..bd8ef00c859a83c75eda765175256ac3b7b6e42a 100644
--- a/src/core/lib/iomgr/socket_utils_common_posix.c
+++ b/src/core/lib/iomgr/socket_utils_common_posix.c
@@ -154,6 +154,28 @@ grpc_error *grpc_set_socket_reuse_addr(int fd, int reuse) {
return GRPC_ERROR_NONE;
}
+/* set a socket to reuse old addresses */
+grpc_error *grpc_set_socket_reuse_port(int fd, int reuse) {
+#ifndef SO_REUSEPORT
+ return GRPC_ERROR_CREATE("SO_REUSEPORT unavailable on compiling system");
+#else
+ int val = (reuse != 0);
+ int newval;
+ socklen_t intlen = sizeof(newval);
+ if (0 != setsockopt(fd, SOL_SOCKET, SO_REUSEPORT, &val, sizeof(val))) {
+ return GRPC_OS_ERROR(errno, "setsockopt(SO_REUSEPORT)");
+ }
+ if (0 != getsockopt(fd, SOL_SOCKET, SO_REUSEPORT, &newval, &intlen)) {
+ return GRPC_OS_ERROR(errno, "getsockopt(SO_REUSEPORT)");
+ }
+ if ((newval != 0) != val) {
+ return GRPC_ERROR_CREATE("Failed to set SO_REUSEPORT");
+ }
+
+ return GRPC_ERROR_NONE;
+#endif
+}
+
/* disable nagle */
grpc_error *grpc_set_socket_low_latency(int fd, int low_latency) {
int val = (low_latency != 0);
diff --git a/src/core/lib/iomgr/socket_utils_posix.h b/src/core/lib/iomgr/socket_utils_posix.h
index 4e66cbc0c58cb8da2fd93b5dddc02884e28f6ce0..607fbc7a791672df7c20799304fb48d1cce86614 100644
--- a/src/core/lib/iomgr/socket_utils_posix.h
+++ b/src/core/lib/iomgr/socket_utils_posix.h
@@ -55,6 +55,9 @@ grpc_error *grpc_set_socket_reuse_addr(int fd, int reuse);
/* disable nagle */
grpc_error *grpc_set_socket_low_latency(int fd, int low_latency);
+/* set SO_REUSEPORT */
+grpc_error *grpc_set_socket_reuse_port(int fd, int reuse);
+
/* Returns true if this system can create AF_INET6 sockets bound to ::1.
The value is probed once, and cached for the life of the process.
diff --git a/src/core/lib/iomgr/tcp_server.h b/src/core/lib/iomgr/tcp_server.h
index 924121f0c79e335f22be9baadccbe7e713c57138..79f0242b26d90e0d00751da98a75271bfb411866 100644
--- a/src/core/lib/iomgr/tcp_server.h
+++ b/src/core/lib/iomgr/tcp_server.h
@@ -34,6 +34,8 @@
#ifndef GRPC_CORE_LIB_IOMGR_TCP_SERVER_H
#define GRPC_CORE_LIB_IOMGR_TCP_SERVER_H
+#include <grpc/grpc.h>
+
#include "src/core/lib/iomgr/closure.h"
#include "src/core/lib/iomgr/endpoint.h"
@@ -58,6 +60,7 @@ typedef void (*grpc_tcp_server_cb)(grpc_exec_ctx *exec_ctx, void *arg,
If shutdown_complete is not NULL, it will be used by
grpc_tcp_server_unref() when the ref count reaches zero. */
grpc_error *grpc_tcp_server_create(grpc_closure *shutdown_complete,
+ const grpc_channel_args *args,
grpc_tcp_server **server);
/* Start listening to bound ports */
diff --git a/src/core/lib/iomgr/tcp_server_posix.c b/src/core/lib/iomgr/tcp_server_posix.c
index db70fada96aa5cb32670214cc322a10f4d16e663..7047934780b3a3f20bd4f26aae1ae920afffdf49 100644
--- a/src/core/lib/iomgr/tcp_server_posix.c
+++ b/src/core/lib/iomgr/tcp_server_posix.c
@@ -112,8 +112,10 @@ struct grpc_tcp_server {
/* destroyed port count: how many ports are completely destroyed */
size_t destroyed_ports;
- /* is this server shutting down? (boolean) */
- int shutdown;
+ /* is this server shutting down? */
+ bool shutdown;
+ /* use SO_REUSEPORT */
+ bool so_reuseport;
/* linked list of server ports */
grpc_tcp_listener *head;
@@ -132,14 +134,42 @@ struct grpc_tcp_server {
size_t pollset_count;
};
+static gpr_once check_init = GPR_ONCE_INIT;
+static bool has_so_reuseport;
+
+static void init(void) {
+ int s = socket(AF_INET, SOCK_STREAM, 0);
+ if (s >= 0) {
+ has_so_reuseport = GRPC_LOG_IF_ERROR("check for SO_REUSEPORT",
+ grpc_set_socket_reuse_port(s, 1));
+ close(s);
+ }
+}
+
grpc_error *grpc_tcp_server_create(grpc_closure *shutdown_complete,
+ const grpc_channel_args *args,
grpc_tcp_server **server) {
+ gpr_once_init(&check_init, init);
+
grpc_tcp_server *s = gpr_malloc(sizeof(grpc_tcp_server));
+ s->so_reuseport = has_so_reuseport;
+ for (size_t i = 0; i < (args == NULL ? 0 : args->num_args); i++) {
+ if (0 == strcmp(GRPC_ARG_ALLOW_REUSEPORT, args->args[i].key)) {
+ if (args->args[i].type == GRPC_ARG_INTEGER) {
+ s->so_reuseport =
+ has_so_reuseport && (args->args[i].value.integer != 0);
+ } else {
+ gpr_free(s);
+ return GRPC_ERROR_CREATE(GRPC_ARG_ALLOW_REUSEPORT
+ " must be an integer");
+ }
+ }
+ }
gpr_ref_init(&s->refs, 1);
gpr_mu_init(&s->mu);
s->active_ports = 0;
s->destroyed_ports = 0;
- s->shutdown = 0;
+ s->shutdown = false;
s->shutdown_starting.head = NULL;
s->shutdown_starting.tail = NULL;
s->shutdown_complete = shutdown_complete;
@@ -214,7 +244,7 @@ static void tcp_server_destroy(grpc_exec_ctx *exec_ctx, grpc_tcp_server *s) {
gpr_mu_lock(&s->mu);
GPR_ASSERT(!s->shutdown);
- s->shutdown = 1;
+ s->shutdown = true;
/* shutdown all fd's */
if (s->active_ports) {
@@ -264,13 +294,19 @@ static int get_max_accept_queue_size(void) {
/* Prepare a recently-created socket for listening. */
static grpc_error *prepare_socket(int fd, const struct sockaddr *addr,
- size_t addr_len, int *port) {
+ size_t addr_len, bool so_reuseport,
+ int *port) {
struct sockaddr_storage sockname_temp;
socklen_t sockname_len;
grpc_error *err = GRPC_ERROR_NONE;
GPR_ASSERT(fd >= 0);
+ if (so_reuseport) {
+ err = grpc_set_socket_reuse_port(fd, 1);
+ if (err != GRPC_ERROR_NONE) goto error;
+ }
+
err = grpc_set_socket_nonblocking(fd, 1);
if (err != GRPC_ERROR_NONE) goto error;
err = grpc_set_socket_cloexec(fd, 1);
@@ -397,7 +433,7 @@ static grpc_error *add_socket_to_server(grpc_tcp_server *s, int fd,
char *addr_str;
char *name;
- grpc_error *err = prepare_socket(fd, addr, addr_len, &port);
+ grpc_error *err = prepare_socket(fd, addr, addr_len, s->so_reuseport, &port);
if (err == GRPC_ERROR_NONE) {
GPR_ASSERT(port > 0);
grpc_sockaddr_to_string(&addr_str, (struct sockaddr *)&addr, 1);
@@ -433,6 +469,51 @@ static grpc_error *add_socket_to_server(grpc_tcp_server *s, int fd,
return err;
}
+static grpc_error *clone_port(grpc_tcp_listener *listener, unsigned count) {
+ grpc_tcp_listener *sp = NULL;
+ char *addr_str;
+ char *name;
+ grpc_error *err;
+
+ for (grpc_tcp_listener *l = listener->next; l && l->is_sibling; l = l->next) {
+ l->fd_index += count;
+ }
+
+ for (unsigned i = 0; i < count; i++) {
+ int fd, port;
+ grpc_dualstack_mode dsmode;
+ err = grpc_create_dualstack_socket(&listener->addr.sockaddr, SOCK_STREAM, 0,
+ &dsmode, &fd);
+ if (err != GRPC_ERROR_NONE) return err;
+ err = prepare_socket(fd, &listener->addr.sockaddr, listener->addr_len, true,
+ &port);
+ if (err != GRPC_ERROR_NONE) return err;
+ grpc_sockaddr_to_string(&addr_str, &listener->addr.sockaddr, 1);
+ gpr_asprintf(&name, "tcp-server-listener:%s/clone-%d", addr_str, i);
+ sp = gpr_malloc(sizeof(grpc_tcp_listener));
+ sp->next = listener->next;
+ listener->next = sp;
+ sp->server = listener->server;
+ sp->fd = fd;
+ sp->emfd = grpc_fd_create(fd, name);
+ memcpy(sp->addr.untyped, listener->addr.untyped, listener->addr_len);
+ sp->addr_len = listener->addr_len;
+ sp->port = port;
+ sp->port_index = listener->port_index;
+ sp->fd_index = listener->fd_index + count - i;
+ sp->is_sibling = 1;
+ sp->sibling = listener->is_sibling ? listener->sibling : listener;
+ GPR_ASSERT(sp->emfd);
+ while (listener->server->tail->next != NULL) {
+ listener->server->tail = listener->server->tail->next;
+ }
+ gpr_free(addr_str);
+ gpr_free(name);
+ }
+
+ return GRPC_ERROR_NONE;
+}
+
grpc_error *grpc_tcp_server_add_port(grpc_tcp_server *s, const void *addr,
size_t addr_len, int *out_port) {
grpc_tcp_listener *sp;
@@ -589,14 +670,29 @@ void grpc_tcp_server_start(grpc_exec_ctx *exec_ctx, grpc_tcp_server *s,
s->on_accept_cb_arg = on_accept_cb_arg;
s->pollsets = pollsets;
s->pollset_count = pollset_count;
- for (sp = s->head; sp; sp = sp->next) {
- for (i = 0; i < pollset_count; i++) {
- grpc_pollset_add_fd(exec_ctx, pollsets[i], sp->emfd);
+ sp = s->head;
+ while (sp != NULL) {
+ if (s->so_reuseport && pollset_count > 1) {
+ GPR_ASSERT(GRPC_LOG_IF_ERROR(
+ "clone_port", clone_port(sp, (unsigned)(pollset_count - 1))));
+ for (i = 0; i < pollset_count; i++) {
+ grpc_pollset_add_fd(exec_ctx, pollsets[i], sp->emfd);
+ sp->read_closure.cb = on_read;
+ sp->read_closure.cb_arg = sp;
+ grpc_fd_notify_on_read(exec_ctx, sp->emfd, &sp->read_closure);
+ s->active_ports++;
+ sp = sp->next;
+ }
+ } else {
+ for (i = 0; i < pollset_count; i++) {
+ grpc_pollset_add_fd(exec_ctx, pollsets[i], sp->emfd);
+ }
+ sp->read_closure.cb = on_read;
+ sp->read_closure.cb_arg = sp;
+ grpc_fd_notify_on_read(exec_ctx, sp->emfd, &sp->read_closure);
+ s->active_ports++;
+ sp = sp->next;
}
- sp->read_closure.cb = on_read;
- sp->read_closure.cb_arg = sp;
- grpc_fd_notify_on_read(exec_ctx, sp->emfd, &sp->read_closure);
- s->active_ports++;
}
gpr_mu_unlock(&s->mu);
}