UCT/TCP: Filtered out bridge devices.

openucx · Nov 7, 2023 · b6e1145 · b6e1145
1 parent a459ad9
commit b6e1145
Show file tree

Hide file tree

Showing 3 changed files with 31 additions and 6 deletions.
diff --git a/src/uct/tcp/tcp.h b/src/uct/tcp/tcp.h
@@ -465,6 +465,7 @@ typedef struct uct_tcp_md {
     struct {
         int         af_prio_count;
         sa_family_t af_prio_list[2];
+        int         bridge_devices;
     } config;
 } uct_tcp_md_t;
 
@@ -475,6 +476,7 @@ typedef struct uct_tcp_md {
 typedef struct uct_tcp_md_config {
     uct_md_config_t                   super;
     UCS_CONFIG_STRING_ARRAY_FIELD(af) af_prio;
+    int                               bridge_devices;
 } uct_tcp_md_config_t;
 
 

diff --git a/src/uct/tcp/tcp_iface.c b/src/uct/tcp/tcp_iface.c
@@ -862,6 +862,23 @@ static UCS_CLASS_DEFINE_NEW_FUNC(uct_tcp_iface_t, uct_iface_t, uct_md_h,
                                  uct_worker_h, const uct_iface_params_t*,
                                  const uct_iface_config_t*);
 
+static int uct_tcp_filter_bridges(const struct dirent *entry)
+{
+    char path[PATH_MAX];
+    struct stat st;
+    int is_bridge;
+
+    ucs_snprintf_safe(path, PATH_MAX, UCT_TCP_IFACE_NETDEV_DIR "/%s/bridge",
+                      entry->d_name);
+
+    is_bridge = (stat(path, &st) == 0) && S_ISDIR(st.st_mode);
+    if (is_bridge) {
+        ucs_debug("filtered out bridge device %s", entry->d_name);
+    }
+
+    return !is_bridge;
+}
+
 ucs_status_t uct_tcp_query_devices(uct_md_h md,
                                    uct_tl_device_resource_t **devices_p,
                                    unsigned *num_devices_p)
@@ -877,7 +894,9 @@ ucs_status_t uct_tcp_query_devices(uct_md_h md,
     char path_buffer[PATH_MAX];
     ucs_sys_device_t sys_dev;
 
-    n = scandir(UCT_TCP_IFACE_NETDEV_DIR, &entries, NULL, alphasort);
+    n = scandir(UCT_TCP_IFACE_NETDEV_DIR, &entries,
+                tcp_md->config.bridge_devices ? NULL : uct_tcp_filter_bridges,
+                alphasort);
     if (n == -1) {
         ucs_error("scandir(%s) failed: %m", UCT_TCP_IFACE_NETDEV_DIR);
         status = UCS_ERR_IO_ERROR;

diff --git a/src/uct/tcp/tcp_md.c b/src/uct/tcp/tcp_md.c
@@ -14,13 +14,16 @@
 
 
 static ucs_config_field_t uct_tcp_md_config_table[] = {
-    {"", "", NULL,
-     ucs_offsetof(uct_tcp_md_config_t, super), UCS_CONFIG_TYPE_TABLE(uct_md_config_table)},
+    {"", "", NULL, ucs_offsetof(uct_tcp_md_config_t, super),
+     UCS_CONFIG_TYPE_TABLE(uct_md_config_table)},
 
     {"AF_PRIO", "inet,inet6",
      "Priority of address families used for socket connections",
      ucs_offsetof(uct_tcp_md_config_t, af_prio), UCS_CONFIG_TYPE_STRING_ARRAY},
 
+    {"BRIDGE_DEVICES", "n", "Use bridge devices",
+     ucs_offsetof(uct_tcp_md_config_t, bridge_devices), UCS_CONFIG_TYPE_BOOL},
+
     {NULL}
 };
 
@@ -76,9 +79,10 @@ uct_tcp_md_open(uct_component_t *component, const char *md_name,
         goto err;
     }
 
-    tcp_md->super.ops            = &uct_tcp_md_ops;
-    tcp_md->super.component      = &uct_tcp_component;
-    tcp_md->config.af_prio_count = ucs_min(md_config->af_prio.count, 2);
+    tcp_md->super.ops             = &uct_tcp_md_ops;
+    tcp_md->super.component       = &uct_tcp_component;
+    tcp_md->config.af_prio_count  = ucs_min(md_config->af_prio.count, 2);
+    tcp_md->config.bridge_devices = md_config->bridge_devices;
 
     for (i = 0; i < tcp_md->config.af_prio_count; i++) {
         if (!strcasecmp(md_config->af_prio.af[i], "inet")) {