[mle] use increasing timeout for child role restoration (#11895)

This change updates the `PrevRoleRestorer` logic to use an increasing
timeout when a non-sleepy device sends `Child Update Request`
messages to restore its previous child role.

The timeout starts at 4 seconds and doubles with each subsequent
retransmission. This strategy is designed to handle scenarios where
the parent may also be restarting, such as after a network-wide power
outage, by allowing more time for the parent to recover. Over four
attempts, the device waits a total of 29 seconds (4 + 8 + 16 + 1)
before abandoning the restoration process.

Sleepy devices continue to use a short and fixed 1-second timeout
between retransmissions.

Additionally, if the restoring child receives a Child Update Request
from its former parent, it switches back to the shorter 1-second
timeout to expedite the restoration process and allow at least
two more Child Update attempts.

src/core/thread/mle.cpp

@@ -2231,6 +2231,7 @@ void Mle::HandleChildUpdateRequestOnChild(RxInfo &aRxInfo)
         // message (`canTrustMessage` will be `false`).
 
         VerifyOrExit(mPrevRoleRestorer.IsRestoringChildRole());
+        mPrevRoleRestorer.HandleChildUpdateRequest(aRxInfo);
     }
 
     SuccessOrExit(error = Tlv::Find<SourceAddressTlv>(aRxInfo.mMessage, sourceAddress));
@@ -4203,6 +4204,8 @@ Mle::PrevRoleRestorer::PrevRoleRestorer(Instance &aInstance)
     : InstanceLocator(aInstance)
     , mState(kIdle)
     , mAttempts(0)
+    , mUseIncreasingTimeout(false)
+    , mCurTimeout(0)
     , mTimer(aInstance)
 {
 }
@@ -4242,9 +4245,38 @@ Error Mle::PrevRoleRestorer::Start(void)
 
     VerifyOrExit(Get<Mle>().mLastSavedRole == kRoleChild);
     VerifyOrExit(Get<Mle>().mParent.IsStateValidOrRestoring());
+
+    // Try to restore the previous child role by sending up to
+    // `kChildUpdateAttempts` Child Update Requests to the former
+    // parent.
+    //
+    // A non-sleepy child uses an increasing timeout (starting at
+    // `kChildUpdateStartTimeout` and doubling) to give a potentially
+    // restarting parent more time to recover. The total wait time is
+    // 29 seconds (4+8+16+1). A sleepy child uses a fixed short
+    // timeout of one seconds(`kChildUpdateMinTimeout`).
+    //
+    // Receiving an MLE message from the parent triggers a switch to
+    // the short timeout and guarantees at least two more attempts
+    // (`kExtraChildUpdatesAfterRxFromParent`).
+
     SetState(kRestoringChildRole);
+
     GenerateRandomChallenge();
-    mAttempts = kMaxChildUpdatesToRestoreRole;
+
+    mAttempts = kChildUpdateAttempts;
+
+    if (!Get<Mle>().IsRxOnWhenIdle())
+    {
+        mUseIncreasingTimeout = false;
+        mCurTimeout           = kChildUpdateMinTimeout;
+    }
+    else
+    {
+        mUseIncreasingTimeout = true;
+        mCurTimeout           = kChildUpdateStartTimeout;
+    }
+
     mTimer.Start(Get<Mle>().GenerateRandomDelay(kMaxStartDelay));
     error = kErrorNone;
 
@@ -4305,12 +4337,48 @@ exit:
 
 void Mle::PrevRoleRestorer::SendChildUpdate(void)
 {
-    mTimer.Start(Random::NonCrypto::AddJitter(kChildUpdateRetxDelay, kRetxJitter));
+    if (mAttempts == 0)
+    {
+        mTimer.Start(kChildUpdateMinTimeout);
+    }
+    else
+    {
+        mTimer.Start(Random::NonCrypto::AddJitter(mCurTimeout, kChildUpdateRetxJitter));
+
+        if (mUseIncreasingTimeout)
+        {
+            mCurTimeout += mCurTimeout;
+        }
+    }
 
     LogDebg("Sending Child Update Request to restore child role, remaining attempts: %u", mAttempts);
     IgnoreError(Get<Mle>().SendChildUpdateRequestToParent(kToRestoreChildRole));
 }
 
+void Mle::PrevRoleRestorer::CheckIfMessageIsFromParent(RxInfo &aRxInfo)
+{
+    VerifyOrExit(IsRestoringChildRole());
+    VerifyOrExit(aRxInfo.mNeighbor == &Get<Mle>().GetParent());
+
+    VerifyOrExit(mUseIncreasingTimeout);
+
+    LogInfo("Received msg from former parent, speeding up child role restoration");
+
+    mUseIncreasingTimeout = false;
+    mCurTimeout           = kChildUpdateMinTimeout;
+
+    if (mAttempts <= kExtraChildUpdatesAfterRxFromParent)
+    {
+        mAttempts = kExtraChildUpdatesAfterRxFromParent;
+        LogInfo("Allow extra Child Update attempts %u", mAttempts);
+    }
+
+    mTimer.FireAtIfEarlier(TimerMilli::GetNow() + Random::NonCrypto::AddJitter(mCurTimeout, kChildUpdateRetxJitter));
+
+exit:
+    return;
+}
+
 #if OPENTHREAD_FTD
 
 void Mle::PrevRoleRestorer::DetermineMaxLinkRequestAttempts(void)

src/core/thread/mle.hpp

@@ -1843,15 +1843,20 @@ private:
         bool  IsRestoringChildRole(void) const { return mState == kRestoringChildRole; }
         bool  IsRestoringRouterOrLeaderRole(void) const { return mState == kRestoringRouterOrLeaderRole; }
         void  HandleTimer(void);
+        void  HandleChildUpdateRequest(RxInfo &aRxInfo) { CheckIfMessageIsFromParent(aRxInfo); }
 
         void               GenerateRandomChallenge(void) { mChallenge.GenerateRandom(); }
         const TxChallenge &GetChallenge(void) const { return mChallenge; }
 
     private:
-        static constexpr uint32_t kMaxStartDelay                = 25;
-        static constexpr uint8_t  kMaxChildUpdatesToRestoreRole = kMaxChildKeepAliveAttempts;
-        static constexpr uint32_t kChildUpdateRetxDelay         = kUnicastRetxDelay; /// 1000 msec
-        static constexpr uint16_t kRetxJitter                   = 5;
+        static constexpr uint32_t kMaxStartDelay = 25; // in msec
+
+        // Restoring Child Role (sending "child update request").
+        static constexpr uint8_t  kChildUpdateAttempts                = 4;
+        static constexpr uint8_t  kExtraChildUpdatesAfterRxFromParent = 2;
+        static constexpr uint16_t kChildUpdateMinTimeout              = 1000; // in ms
+        static constexpr uint16_t kChildUpdateStartTimeout            = 4000; // in ms
+        static constexpr uint16_t kChildUpdateRetxJitter              = 25;   // in ms
 
         enum State : uint8_t
         {
@@ -1862,6 +1867,7 @@ private:
 
         void SetState(State aState);
         void SendChildUpdate(void);
+        void CheckIfMessageIsFromParent(RxInfo &aRxInfo);
 #if OPENTHREAD_FTD
         void DetermineMaxLinkRequestAttempts(void);
         void SendMulticastLinkRequest(void);
@@ -1870,7 +1876,9 @@ private:
         using DelayTimer = TimerMilliIn<Mle, &Mle::HandleRoleRestorerTimer>;
 
         State       mState;
-        uint8_t     mAttempts;
+        uint8_t     mAttempts : 7;
+        bool        mUseIncreasingTimeout : 1;
+        uint16_t    mCurTimeout;
         DelayTimer  mTimer;
         TxChallenge mChallenge;
     };