Merge pull request #30268 from yuwata/network-fix-too-many-waiting-replies

network: fix issue caused by too many waiting replies

Author: keszybz

src/libsystemd/sd-netlink/netlink-util.h

@@ -107,5 +107,7 @@ int rtattr_read_nexthop(const struct rtnexthop *rtnh, size_t size, int family, O
 
 void netlink_seal_message(sd_netlink *nl, sd_netlink_message *m);
 
+size_t netlink_get_reply_callback_count(sd_netlink *nl);
+
 /* TODO: to be exported later */
 int sd_netlink_sendv(sd_netlink *nl, sd_netlink_message **messages, size_t msgcnt, uint32_t **ret_serial);

src/libsystemd/sd-netlink/sd-netlink.c

@@ -458,6 +458,12 @@ static int timeout_compare(const void *a, const void *b) {
         return CMP(x->timeout, y->timeout);
 }
 
+size_t netlink_get_reply_callback_count(sd_netlink *nl) {
+        assert(nl);
+
+        return hashmap_size(nl->reply_callbacks);
+}
+
 int sd_netlink_call_async(
                 sd_netlink *nl,
                 sd_netlink_slot **ret_slot,
@@ -477,7 +483,7 @@ int sd_netlink_call_async(
         assert_return(!netlink_pid_changed(nl), -ECHILD);
 
         if (hashmap_size(nl->reply_callbacks) >= REPLY_CALLBACKS_MAX)
-                return -ERANGE;
+                return -EXFULL;
 
         r = hashmap_ensure_allocated(&nl->reply_callbacks, &trivial_hash_ops);
         if (r < 0)

src/network/networkd-queue.c

@@ -7,6 +7,8 @@
 #include "networkd-queue.h"
 #include "string-table.h"
 
+#define REPLY_CALLBACK_COUNT_THRESHOLD 128
+
 static Request *request_free(Request *req) {
         if (!req)
                 return NULL;
@@ -220,6 +222,13 @@ int manager_process_requests(sd_event_source *s, void *userdata) {
                         if (req->waiting_reply)
                                 continue; /* Waiting for netlink reply. */
 
+                        /* Typically, requests send netlink message asynchronously. If there are many requests
+                         * queued, then this event may make reply callback queue in sd-netlink full. */
+                        if (netlink_get_reply_callback_count(manager->rtnl) >= REPLY_CALLBACK_COUNT_THRESHOLD ||
+                            netlink_get_reply_callback_count(manager->genl) >= REPLY_CALLBACK_COUNT_THRESHOLD ||
+                            fw_ctx_get_reply_callback_count(manager->fw_ctx) >= REPLY_CALLBACK_COUNT_THRESHOLD)
+                                return 0;
+
                         r = req->process(req, link, req->userdata);
                         if (r == 0)
                                 continue;

src/shared/firewall-util.c

@@ -8,6 +8,7 @@
 #include "firewall-util.h"
 #include "firewall-util-private.h"
 #include "log.h"
+#include "netlink-util.h"
 #include "string-table.h"
 
 static const char * const firewall_backend_table[_FW_BACKEND_MAX] = {
@@ -90,6 +91,13 @@ FirewallContext *fw_ctx_free(FirewallContext *ctx) {
         return mfree(ctx);
 }
 
+size_t fw_ctx_get_reply_callback_count(FirewallContext *ctx) {
+        if (!ctx || !ctx->nfnl)
+                return 0;
+
+        return netlink_get_reply_callback_count(ctx->nfnl);
+}
+
 int fw_add_masquerade(
                 FirewallContext **ctx,
                 bool add,

src/shared/firewall-util.h

@@ -15,6 +15,8 @@ FirewallContext *fw_ctx_free(FirewallContext *ctx);
 
 DEFINE_TRIVIAL_CLEANUP_FUNC(FirewallContext *, fw_ctx_free);
 
+size_t fw_ctx_get_reply_callback_count(FirewallContext *ctx);
+
 int fw_add_masquerade(
                 FirewallContext **ctx,
                 bool add,