Merge pull request #14600 from rabbitmq/mergify/bp/v4.1.x/pr-14599

CQ shared: Fix off-by-nine error leading to lost messages (backport #14576) (backport #14599)

Author: lhoguin

deps/rabbit/src/rabbit_msg_store.erl

@@ -1599,7 +1599,7 @@ scan_data(<<Size:64, MsgIdInt:128, _Rest/bits>> = Data, Fd, Fun, Offset, FileSiz
     end;
 %% This might be the start of a message.
 scan_data(<<Size:64, Rest/bits>> = Data, Fd, Fun, Offset, FileSize, MsgIdsFound, Acc)
-          when byte_size(Rest) < Size + 1, Size < FileSize - Offset ->
+          when byte_size(Rest) < Size + 1, Size + 9 =< FileSize - Offset ->
     scan(Data, Fd, Fun, Offset, FileSize, MsgIdsFound, Acc);
 scan_data(Data, Fd, Fun, Offset, FileSize, MsgIdsFound, Acc)
           when byte_size(Data) < 8 ->

deps/rabbit/test/backing_queue_SUITE.erl

@@ -710,6 +710,17 @@ msg_store_file_scan1(Config) ->
         [{bin, <<0, 0:48, 17, 17, "idididididididid", 255, 0:4352/unit:8, 255>>}],
         {ok, [{<<"idididididididid">>, 4378, 1}]},
         fun(Obj = {<<"idididididididid">>, 4378, 1}) -> {valid, Obj}; (_) -> invalid end),
+    %% Off-by-nine regression testing. The file scanning could miss
+    %% some messages if previous data looked like a message but its
+    %% size went past the end of the file.
+    lists:foreach(fun(N) ->
+        ok = Scan([
+            {bin, <<(4194304 + N):64, 0:(4194304 - 8 - 25 - 10)/unit:8>>},
+            {msg, gen_id(), <<>>},
+            %% Padding ensures there's no 255 at the end of the size indicated by 'bin'.
+            {pad, 10}
+        ])
+    end, lists:seq(-9, -1)),
     %% All good!!
     passed.