File 1427-erts-Don-t-expose-timeout-primitives-to-hu... of Package erlang

Overview Repositories Revisions Requests Users Attributes Meta

File 1427-erts-Don-t-expose-timeout-primitives-to-huge-timeout.patch of Package erlang

From 114466c12919f97ea7c63c35c939b9dac5a087af Mon Sep 17 00:00:00 2001
From: Rickard Green <rickard@erlang.org>
Date: Sat, 23 Dec 2023 22:34:22 +0100
Subject: [PATCH 1/3] [erts] Don't expose timeout primitives to huge timeouts

---
 erts/emulator/beam/time.c              | 17 ++++++++-
 erts/emulator/test/timer_bif_SUITE.erl | 53 ++++++++++++++++++++++++--
 2 files changed, 65 insertions(+), 5 deletions(-)

diff --git a/erts/emulator/beam/time.c b/erts/emulator/beam/time.c
index ce57b041d6..9584c6b600 100644
--- a/erts/emulator/beam/time.c
+++ b/erts/emulator/beam/time.c
@@ -613,7 +613,22 @@ find_next_timeout(ErtsSchedulerData *esdp, ErtsTimerWheel *tiw)
     }
 
 done: {
-        ErtsMonotonicTime min_timeout;
+        ErtsMonotonicTime min_timeout, timeout_pos_limit;
+
+        timeout_pos_limit = tiw->pos + ERTS_CLKTCKS_WEEK;
+        if (min_timeout_pos > timeout_pos_limit) {
+            /*
+             * We never expose a timeout larger than a week in order to avoid
+             * issues with primitives that have a limited maximum timeout
+             * time (for example, poll() with a timeout in milliseconds passed
+             * in a variable of type 'int' which is around 3,5 weeks). The
+             * overhead, in case all timers are very far in the future, will
+             * be that the scheduler once a week will have to check if we got
+             * any timeouts closer in time than a week...
+             */
+            min_timeout_pos = timeout_pos_limit;
+            true_min_timeout = 0;
+        }
 
         min_timeout = ERTS_CLKTCKS_TO_MONOTONIC(min_timeout_pos);
         tiw->next_timeout_pos = min_timeout_pos;
diff --git a/erts/emulator/test/timer_bif_SUITE.erl b/erts/emulator/test/timer_bif_SUITE.erl
index c02d2c767f..9586d4b23f 100644
--- a/erts/emulator/test/timer_bif_SUITE.erl
+++ b/erts/emulator/test/timer_bif_SUITE.erl
@@ -32,7 +32,8 @@
 %	 same_time_yielding_with_cancel_other_accessor/1,
 	 auto_cancel_yielding/1,
          suspended_scheduler_timeout/1,
-         multizero_timeout_in_timeout/1]).
+         multizero_timeout_in_timeout/1,
+         huge_timeout/1]).
 
 -include_lib("common_test/include/ct.hrl").
 
@@ -72,7 +73,8 @@ all() ->
 %     same_time_yielding_with_cancel_other_accessor,
      auto_cancel_yielding,
      suspended_scheduler_timeout,
-     multizero_timeout_in_timeout].
+     multizero_timeout_in_timeout,
+     huge_timeout].
 
 
 %% Basic start_timer/3 functionality
@@ -771,8 +773,51 @@ multizero_timeout_in_timeout(Config) when is_list(Config) ->
     io:format("Time=~p~n", [Time]),
     true = Time < Timeout + MaxTimeoutDiff,
     ok.
-            
-        
+
+huge_timeout(Config) when is_list(Config) ->
+    %% More than 2^31 seconds...
+    huge_timeout_test((1 bsl 31)*1000 + 1000),
+    %% More than 2^32 seconds...
+    huge_timeout_test((1 bsl 32)*1000 + 1000),
+    %% More than 2^31 milliseconds...
+    huge_timeout_test((1 bsl 31) + 1000),
+    %% More than 2^32 milliseconds...
+    huge_timeout_test((1 bsl 32) + 1000),
+    ok.
+
+huge_timeout_test(HugeTmo) ->
+    SOnln = erlang:system_info(schedulers_online),
+    process_flag(trap_exit, true),
+    %% Likely to hit the bug if we set a huge timeout in an
+    %% empty timer wheel, then set a small timeout
+    %% and let the small timeout trigger. The huge timeout
+    %% will then be found as the next timeout in the wheel.
+    %% Just setting a huge timeout wont trigger the bug
+    %% since an empty wheel have a fake timeout of one week.
+    Ps = lists:map(
+           fun (N) ->
+                   spawn_opt(
+                     fun () ->
+                             erlang:send_after(HugeTmo, self(), hej),
+                             erlang:send_after(2, self(), hej),
+                             receive after infinity -> ok end
+                     end, [{scheduler,N}, link])
+           end, lists:seq(1, erlang:system_info(schedulers))),
+    %% If we have schedulers offline, those timer wheels are likely
+    %% empty, so set them online increasing the chanse of hitting
+    %% a bug.
+    try
+        erlang:system_flag(schedulers_online, erlang:system_info(schedulers)),
+        receive after 1000 -> ok end
+    after
+        lists:foreach(fun (P) ->
+                              unlink(P),
+                              exit(P, kill),
+                              false = is_process_alive(P)
+                      end, Ps),
+        erlang:system_flag(schedulers_online, SOnln),
+        process_flag(trap_exit, false)
+    end.
 
 process_is_cleaned_up(P) when is_pid(P) ->
     undefined == erts_debug:get_internal_state({process_status, P}).
-- 
2.35.3

Places

File 1427-erts-Don-t-expose-timeout-primitives-to-huge-timeout.patch of Package erlang

Places