extmod,rp2: Keep LWIP timer running if lwip poll_sockets() is called.

Fixes rp2 issue where socket.getaddrinfo() could block indefinitely if an
interface goes down and still has a DNS server configured, as the LWIP
timer stops running and can't time out the DNS query.

Adds a regression test under multi_wlan that times out on rp2 without this
fix.

Fixes issue micropython#18797.

This work was funded through GitHub Sponsors.

Signed-off-by: Angus Gratton <angus@redyak.com.au>

Author: projectgus

extmod/modlwip.c

@@ -95,6 +95,11 @@
 #define MICROPY_PY_LWIP_EXIT
 #endif
 
+#ifndef MICROPY_PY_LWIP_POLL_HOOK
+// Optional port-level hook called if/when LWIP is being polled
+#define MICROPY_PY_LWIP_POLL_HOOK
+#endif
+
 #ifdef MICROPY_PY_LWIP_SLIP
 #include "netif/slipif.h"
 #include "lwip/sio.h"
@@ -360,6 +365,7 @@ static inline bool socket_is_timedout(lwip_socket_obj_t *socket, mp_uint_t ticks
 }
 
 static inline void poll_sockets(void) {
+    MICROPY_PY_LWIP_POLL_HOOK
     mp_event_wait_ms(1);
 }
 

ports/rp2/mpconfigport.h

@@ -294,6 +294,7 @@ typedef intptr_t mp_off_t;
 #include "pico/rand.h"
 extern void lwip_lock_acquire(void);
 extern void lwip_lock_release(void);
+extern void lwip_poll_hook(void);
 
 #if MICROPY_PY_BLUETOOTH || MICROPY_PY_BLUETOOTH_CYW43
 // Bluetooth code only runs in the scheduler, no locking/mutex required.

ports/rp2/mphalport.h

@@ -51,6 +51,7 @@
 #define MICROPY_PY_LWIP_ENTER   lwip_lock_acquire();
 #define MICROPY_PY_LWIP_REENTER lwip_lock_acquire();
 #define MICROPY_PY_LWIP_EXIT    lwip_lock_release();
+#define MICROPY_PY_LWIP_POLL_HOOK lwip_poll_hook();
 
 // Port level Wait-for-Event macro
 //

ports/rp2/mpnetworkport.c

@@ -134,6 +134,15 @@ void lwip_lock_release(void) {
     pendsv_resume();
 }
 
+void lwip_poll_hook(void) {
+    // Start the network soft timer if necessary (it may still only run once,
+    // but ensure timeouts will complete inside any loop that's polling lwip)
+    if (mp_network_soft_timer.mode == SOFT_TIMER_MODE_ONE_SHOT) {
+        mp_network_soft_timer.mode = SOFT_TIMER_MODE_PERIODIC;
+        soft_timer_reinsert(&mp_network_soft_timer, LWIP_TICK_RATE_MS);
+    }
+}
+
 // This is called by soft_timer and executes at PendSV level.
 static void mp_network_soft_timer_callback(soft_timer_entry_t *self) {
     // Run the lwIP internal updates.
@@ -179,10 +188,8 @@ void mod_network_lwip_init(void) {
 static void mp_network_netif_status_cb(struct netif *netif, netif_nsc_reason_t reason, const netif_ext_callback_args_t *args) {
     // Start the network soft timer any time an interface comes up, unless
     // it's already running
-    if (reason == LWIP_NSC_LINK_CHANGED && args->link_changed.state
-        && mp_network_soft_timer.mode == SOFT_TIMER_MODE_ONE_SHOT) {
-        mp_network_soft_timer.mode = SOFT_TIMER_MODE_PERIODIC;
-        soft_timer_reinsert(&mp_network_soft_timer, LWIP_TICK_RATE_MS);
+    if (reason == LWIP_NSC_LINK_CHANGED && args->link_changed.state) {
+        lwip_poll_hook();
     }
 
     if (reason == LWIP_NSC_NETIF_REMOVED) {

tests/multi_wlan/getaddrinfo.py

@@ -0,0 +1,39 @@
+# This is a regression test to ensure getaddrinfo() fails (after a timeout)
+# when Wi-Fi is disconnected.
+#
+# It doesn't require multiple instances, but it does require Wi-Fi to be present
+# but not active, and for no other network interface to be configured. The only
+# tests which already meet these conditions is here in multi_wlan tests.
+try:
+    from network import WLAN
+except (ImportError, NameError):
+    print("SKIP")
+    raise SystemExit
+
+import socket
+
+
+def instance0():
+    WLAN(WLAN.IF_AP).active(0)
+    wlan = WLAN(WLAN.IF_STA)
+    wlan.active(0)
+
+    multitest.next()
+
+    try:
+        socket.getaddrinfo("micropython.org", 80)
+    except OSError as er:
+        print(
+            "active(0) failed"
+        )  # This may fail with code -6 or -2 depending on whether WLAN has been active since reset
+
+    wlan.active(1)
+
+    try:
+        socket.getaddrinfo("micropython.org", 80)
+    except OSError as er:
+        print(
+            "active(1) failed", er.errno in (-2, -202)
+        )  # This one should always be -2 or -202 depending on port
+
+    wlan.active(0)

tests/multi_wlan/getaddrinfo.py.exp

@@ -0,0 +1,3 @@
+--- instance0 ---
+active(0) failed
+active(1) failed True