Merge pull request #5177 from Alexays/fix/recent-issue-bugs

fix: backlight I/O flooding (#5020) + network IP recovery after link flap (#5122)

src/modules/network.cpp

@@ -183,6 +183,12 @@ void waybar::modules::Network::createEventSocket() {
   if (nl_socket_set_nonblocking(ev_sock_)) {
     throw std::runtime_error("Can't set non-blocking on network socket");
   }
+  // Enlarge the socket receive buffer so that a burst of link/address/route
+  // change notifications (e.g. a router reboot or a PPPoE redial) is less likely
+  // to overflow it and make the kernel drop messages (ENOBUFS). Overruns are
+  // still handled in worker() by resynchronising the state, but a larger buffer
+  // avoids most of them. The kernel caps the request at net.core.rmem_max.
+  nl_socket_set_buffer_size(ev_sock_, 1024 * 1024, 0);
   nl_socket_add_memberships(ev_sock_, RTNLGRP_LINK, RTNLGRP_IPV4_IFADDR, RTNLGRP_IPV6_IFADDR, 0);
   if (!config_["interface"].isString()) {
     nl_socket_add_memberships(ev_sock_, RTNLGRP_IPV4_ROUTE, RTNLGRP_IPV6_ROUTE, 0);
@@ -277,6 +283,27 @@ void waybar::modules::Network::worker() {
               rc = 0;
               break;
             }
+            if (rc == -NLE_NOMEM || errno == ENOBUFS) {
+              // The kernel dropped multicast notifications because our receive
+              // buffer overflowed. This happens during a burst of
+              // link/address/route changes such as a router reboot or a PPPoE
+              // redial. We have lost track of the current state -- in
+              // particular the RTM_NEWADDR carrying the interface's new IP
+              // address may have been dropped -- so request a fresh dump to
+              // resynchronise. Without this the address (cleared by the
+              // preceding RTM_DELADDR) would stay blank until Waybar is
+              // restarted, because nothing else re-queries it (#5122).
+              spdlog::warn("network: netlink receive buffer overrun, resyncing state");
+              want_link_dump_ = true;
+              want_addr_dump_ = true;
+              if (!config_["interface"].isString()) {
+                want_route_dump_ = true;
+              }
+              askForStateDump();
+              // Keep draining; the next recv proceeds normally now that the
+              // overrun has been reported.
+              continue;
+            }
           }
           if (rc < 0) {
             spdlog::error("nl_recvmsgs_default error: {}", nl_geterror(-rc));

src/util/backlight_backend.cpp

@@ -230,9 +230,21 @@ BacklightBackend::BacklightBackend(std::chrono::milliseconds interval,
         upsert_device(devices, dev.get());
       }
 
-      // Refresh state if timed out
+      // Refresh state if timed out. Only re-read the sysfs attributes of the
+      // devices we already track instead of re-enumerating the whole udev tree
+      // (udev_enumerate_scan_devices), which walks all of /sys/class/backlight
+      // and /sys/class/leds and floods the filesystem with open/close syscalls
+      // on every polling tick (#5020). Device add/remove is already delivered
+      // by the udev monitor above, so a periodic full re-scan is redundant.
       if (event_count == 0) {
-        enumerate_devices(devices, udev.get());
+        for (const auto& device : devices) {
+          std::unique_ptr<udev_device, UdevDeviceDeleter> dev{
+              udev_device_new_from_subsystem_sysname(udev.get(), device.subsystem().c_str(),
+                                                     device.name().c_str())};
+          if (dev) {
+            upsert_device(devices, dev.get());
+          }
+        }
       }
       {
         std::scoped_lock<std::mutex> lock(udev_thread_mutex_);