Time out an abandoned admin edit transaction (#11254)

begin_edit_settings sets a bool that only the matching commit ever
clears. If the commit never arrives -- the client dropped its link
partway through a bulk config import, or went away entirely -- the
transaction stays open indefinitely, and every later config write from
any client is applied to RAM, acknowledged, and then never saved. The
writes look successful and are visible in get_config, but the node
reverts all of them at the next boot, and only a reboot clears it.

Give the transaction a one minute idle timeout. Each deferred write
restarts the clock, so the window bounds the gap between writes rather
than the length of the edit; a bulk import sends them milliseconds
apart. The next admin message after it lapses retires the transaction,
persisting the segments it had deferred and flushing the warnings it was
holding. The check runs after the auth gates and before the switch, so
every case sees consistent state and the recovery's flash write happens
in the main loop rather than a disconnect callback.

It saves rather than rolls back because there is nothing to roll back
to: each write already took effect in RAM and was acknowledged, so the
transaction only ever deferred the save. That also makes an early expiry
cheap -- the worst case for a client that really was still going is that
its remaining writes are saved individually instead of batched. Closing
on disconnect instead was tempting, but the reporter's captures show iOS
dropping and reconnecting within half a second several times per
session, which would abort imports that currently survive the blip.

Also drops the file-scope hasOpenEditTransaction, shadowed by the class
member at every use site and referenced nowhere else in the tree.

Reported in #11245

Author: thebentern

src/modules/AdminModule.cpp

@@ -69,7 +69,6 @@
 #endif
 
 AdminModule *adminModule;
-bool hasOpenEditTransaction;
 
 #if !(MESHTASTIC_EXCLUDE_PKI_KEYGEN || MESHTASTIC_EXCLUDE_PKI)
 static bool licensedIdentityWillMigrate()
@@ -240,6 +239,9 @@ bool AdminModule::handleReceivedProtobuf(const meshtastic_MeshPacket &mp, meshta
             return handled;
         }
     }
+    // Before the switch, so every case below sees consistent transaction state.
+    expireStaleEditTransaction();
+
     switch (r->which_payload_variant) {
 
 #ifdef MESHTASTIC_ENCRYPTED_STORAGE
@@ -481,12 +483,14 @@ bool AdminModule::handleReceivedProtobuf(const meshtastic_MeshPacket &mp, meshta
     case meshtastic_AdminMessage_begin_edit_settings_tag: {
         LOG_INFO("Begin transaction for editing settings");
         hasOpenEditTransaction = true;
+        editTransactionActivityMs = millis();
         break;
     }
     case meshtastic_AdminMessage_commit_edit_settings_tag: {
         disableBluetooth();
         LOG_INFO("Commit transaction for edited settings");
         hasOpenEditTransaction = false;
+        deferredEditSegments = 0;
         saveChanges(SEGMENT_CONFIG | SEGMENT_MODULECONFIG | SEGMENT_DEVICESTATE | SEGMENT_CHANNELS | SEGMENT_NODEDATABASE);
         flushChannelWarnings(); // one coalesced message for everything edited in this transaction
         break;
@@ -1875,6 +1879,23 @@ void AdminModule::reboot(int32_t seconds)
     rebootAtMsec = (seconds < 0) ? 0 : (millis() + seconds * 1000);
 }
 
+// Without this, a commit that never arrives leaves the transaction open forever and every later
+// config write from any client is applied, acknowledged, and then never saved.
+void AdminModule::expireStaleEditTransaction()
+{
+    if (!hasOpenEditTransaction || Throttle::isWithinTimespanMs(editTransactionActivityMs, EDIT_TRANSACTION_IDLE_MS))
+        return;
+
+    LOG_WARN("Edit transaction abandoned for %us; committing what it applied", EDIT_TRANSACTION_IDLE_MS / 1000);
+    hasOpenEditTransaction = false;
+    int segments = deferredEditSegments;
+    deferredEditSegments = 0;
+    // No reboot: the settings are already live in RAM and the client that would expect one is gone.
+    if (segments)
+        saveChanges(segments, false);
+    flushChannelWarnings();
+}
+
 void AdminModule::saveChanges(int saveWhat, bool shouldReboot)
 {
 #ifdef PIO_UNIT_TESTING
@@ -1885,6 +1906,8 @@ void AdminModule::saveChanges(int saveWhat, bool shouldReboot)
         service->reloadConfig(saveWhat); // Calls saveToDisk among other things
     } else {
         LOG_INFO("Delay save of changes to disk until the open transaction is committed");
+        editTransactionActivityMs = millis(); // still in use, so not the abandoned kind we time out
+        deferredEditSegments |= saveWhat;
     }
     if (shouldReboot && !hasOpenEditTransaction) {
         reboot(DEFAULT_REBOOT_SECONDS);

src/modules/AdminModule.h

@@ -40,6 +40,13 @@ class AdminModule : public ProtobufModule<meshtastic_AdminMessage>, public Obser
 
   private:
     bool hasOpenEditTransaction = false;
+    // Each deferred write restarts the clock, so this bounds the gap between writes, not the length
+    // of the edit; a bulk import sends them milliseconds apart.
+    static constexpr uint32_t EDIT_TRANSACTION_IDLE_MS = 60 * 1000;
+    uint32_t editTransactionActivityMs = 0; // millis() of the last save this transaction deferred
+    int deferredEditSegments = 0;           // segments that transaction has touched but not yet saved
+    /// Retire an open edit transaction whose client stopped talking, persisting what it applied.
+    void expireStaleEditTransaction();
 #ifdef PIO_UNIT_TESTING
     int lastSaveWhatForTest = 0;
 #endif

test/support/AdminModuleTestShim.h

@@ -21,9 +21,18 @@ class AdminModuleTestShim : public AdminModule
     meshtastic_MeshPacket *reply() { return myReply; }
 
     // With an "open edit transaction" saveChanges() is a pure no-op: no reloadConfig/saveToDisk/reboot.
-    void deferSaves() { hasOpenEditTransaction = true; }
+    // Stamps the clock like begin_edit_settings, so a slow suite can't age the transaction into expiry.
+    void deferSaves()
+    {
+        hasOpenEditTransaction = true;
+        editTransactionActivityMs = millis();
+    }
     int savedSegments() const { return lastSaveWhatForTest; }
 
+    bool editTransactionOpen() const { return hasOpenEditTransaction; }
+    // Backdate past the idle window so a test sees an abandoned transaction without waiting it out.
+    void ageEditTransaction() { editTransactionActivityMs = millis() - EDIT_TRANSACTION_IDLE_MS - 1; }
+
     // Setters may allocate an error reply from packetPool; drain it each iteration or the pool leaks.
     void drainReply()
     {

test/test_admin_radio/test_main.cpp

@@ -1574,6 +1574,16 @@ static void sendCommitEdit()
     sendAdmin(m);
 }
 
+// An admin message that changes nothing. It answers, so drain the reply or the packet pool leaks.
+static void sendGetDeviceMetadata()
+{
+    meshtastic_AdminMessage m = meshtastic_AdminMessage_init_zero;
+    m.which_payload_variant = meshtastic_AdminMessage_get_device_metadata_request_tag;
+    m.get_device_metadata_request = true;
+    sendAdmin(m);
+    testAdmin->drainReply();
+}
+
 // Preset = LongFast on US, unlicensed owner. "LongFast" is the display name we compare against.
 static void usePresetLongFast()
 {
@@ -1640,6 +1650,53 @@ static void test_warn_transaction_singleChannel_keepsSpecificMessage()
     TEST_ASSERT_EQUAL_INT(0, warningsContaining("on channels"));
 }
 
+// An idle transaction is retired by the next admin message, flushing the warnings it held.
+static void test_editTransaction_abandoned_isRetiredOnNextAdminMessage()
+{
+    usePresetLongFast();
+    sendBeginEdit();
+    sendSetChannel(makeChannel(0, meshtastic_Channel_Role_PRIMARY, "long fast", DEFAULT_KEY, 1));
+    // Deferred, exactly as before: nothing emitted while the transaction looks alive.
+    TEST_ASSERT_EQUAL_INT(0, (int)capturedWarnings.size());
+    TEST_ASSERT_TRUE(testAdmin->editTransactionOpen());
+
+    testAdmin->ageEditTransaction();
+    sendGetDeviceMetadata(); // any later admin message, from any client
+
+    TEST_ASSERT_FALSE(testAdmin->editTransactionOpen());
+    TEST_ASSERT_EQUAL_INT(1, warningsContaining("looks like a mistype of 'LongFast'"));
+}
+
+// A write arriving after abandonment is saved, not deferred to a commit that never comes.
+static void test_editTransaction_abandoned_laterWriteIsNoLongerDeferred()
+{
+    usePresetLongFast();
+    sendBeginEdit();
+    testAdmin->ageEditTransaction();
+
+    sendSetChannel(makeChannel(0, meshtastic_Channel_Role_PRIMARY, "long fast", DEFAULT_KEY, 1));
+
+    // The write itself retired the stale transaction, so its own warning is emitted immediately.
+    TEST_ASSERT_FALSE(testAdmin->editTransactionOpen());
+    TEST_ASSERT_EQUAL_INT(1, warningsContaining("looks like a mistype of 'LongFast'"));
+}
+
+// A transaction still in use is left alone: each write refreshes the window.
+static void test_editTransaction_active_isNotRetired()
+{
+    usePresetLongFast();
+    sendBeginEdit();
+    sendSetChannel(makeChannel(0, meshtastic_Channel_Role_PRIMARY, "long fast", DEFAULT_KEY, 1));
+    sendSetChannel(makeChannel(1, meshtastic_Channel_Role_SECONDARY, "long fast", DEFAULT_KEY, 1));
+
+    TEST_ASSERT_TRUE(testAdmin->editTransactionOpen());
+    TEST_ASSERT_EQUAL_INT(0, (int)capturedWarnings.size());
+
+    sendCommitEdit();
+    TEST_ASSERT_FALSE(testAdmin->editTransactionOpen());
+    TEST_ASSERT_EQUAL_INT(1, warningsContaining("There may be name issues on channels 0, 1"));
+}
+
 static void test_warn_license_noTransaction_emittedImmediately()
 {
     usePresetLongFast();
@@ -1801,6 +1858,9 @@ void setup()
     RUN_TEST(test_warn_cleanChannel_noMessage);
     RUN_TEST(test_warn_transaction_multipleChannels_singleCoalescedMessage);
     RUN_TEST(test_warn_transaction_singleChannel_keepsSpecificMessage);
+    RUN_TEST(test_editTransaction_abandoned_isRetiredOnNextAdminMessage);
+    RUN_TEST(test_editTransaction_abandoned_laterWriteIsNoLongerDeferred);
+    RUN_TEST(test_editTransaction_active_isNotRetired);
     RUN_TEST(test_warn_license_noTransaction_emittedImmediately);
     RUN_TEST(test_warn_license_transaction_coalescedToSingleMessage);