erts: Add crash dumping of EXITING and FREE processes

5 files changed, 145 insertions, 19 deletions

diff --git a/erts/emulator/beam/break.c b/erts/emulator/beam/break.c
index 27bf2187c2..303ab9d9b7 100644
--- a/erts/emulator/beam/break.c
+++ b/erts/emulator/beam/break.c
@@ -39,6 +39,7 @@
 #include "erl_hl_timer.h"
 #include "erl_thr_progress.h"
 #include "erl_proc_sig_queue.h"
+#include "dist.h"
 
 /* Forward declarations -- should really appear somewhere else */
 static void process_killer(void);
@@ -81,11 +82,32 @@ process_info(fmtfn_t to, void *to_arg)
 	    /* Do not include processes with no heap,
 	     * they are most likely just created and has invalid data
 	     */
-	    if (!ERTS_PROC_IS_EXITING(p) && p->heap != NULL)
+	    if (p->heap != NULL)
 		print_process_info(to, to_arg, p, 0);
 	}
     }
 
+    /* Look for FREE processes in the run-queues and dist entries.
+       These have been removed from the ptab but we still want them
+       in the crash dump for debugging. */
+
+    /* First loop through all run-queues */
+    for (i = 0; i < erts_no_schedulers + ERTS_NUM_DIRTY_RUNQS; i++) {
+        ErtsRunQueue *rq = ERTS_RUNQ_IX(i);
+        int j;
+        for (j = 0; j < ERTS_NO_PROC_PRIO_QUEUES; j++) {
+            Process *p = rq->procs.prio[j].first;
+            while (p) {
+                if (ERTS_PSFLG_FREE & erts_atomic32_read_acqb(&p->state))
+                    print_process_info(to, to_arg, p, 0);
+                p = p->next;
+            }
+        }
+    }
+
+    /* Then check all dist entries */
+    erts_dist_print_procs_suspended_on_de(to, to_arg);
+
     port_info(to, to_arg);
 }
 
@@ -206,6 +228,7 @@ print_process_info(fmtfn_t to, void *to_arg, Process *p, ErtsProcLocks orig_lock
 {
     int garbing = 0;
     int running = 0;
+    int exiting = 0;
     Sint len;
     struct saved_calls *scb;
     erts_aint32_t state;
@@ -226,6 +249,9 @@ print_process_info(fmtfn_t to, void *to_arg, Process *p, ErtsProcLocks orig_lock
 			| ERTS_PSFLG_DIRTY_RUNNING))
         running = 1;
 
+    if (state & ERTS_PSFLG_EXITING)
+        exiting = 1;
+
     if (!(locks & ERTS_PROC_LOCK_MAIN)) {
         locks |= ERTS_PROC_LOCK_MAIN;
         if (ERTS_IS_CRASH_DUMPING && running) {
@@ -246,7 +272,7 @@ print_process_info(fmtfn_t to, void *to_arg, Process *p, ErtsProcLocks orig_lock
      * If the process is registered as a global process, display the
      * registered name
      */
-    if (p->common.u.alive.reg)
+    if (!ERTS_PROC_IS_EXITING(p) && p->common.u.alive.reg)
 	erts_print(to, to_arg, "Name: %T\n", p->common.u.alive.reg->name);
 
     /*
@@ -332,7 +358,7 @@ print_process_info(fmtfn_t to, void *to_arg, Process *p, ErtsProcLocks orig_lock
     }
 
     /* display the links only if there are any*/
-    if (ERTS_P_LINKS(p) || ERTS_P_MONITORS(p) || ERTS_P_LT_MONITORS(p)) {
+    if (!exiting && (ERTS_P_LINKS(p) || ERTS_P_MONITORS(p) || ERTS_P_LT_MONITORS(p))) {
 	PrintMonitorContext context = {1, to, to_arg};
 	erts_print(to, to_arg,"Link list: [");
 	erts_link_tree_foreach(ERTS_P_LINKS(p), doit_print_link, &context);	
diff --git a/erts/emulator/beam/dist.c b/erts/emulator/beam/dist.c
index 2c2af4bdd3..4c200ecc83 100644
--- a/erts/emulator/beam/dist.c
+++ b/erts/emulator/beam/dist.c
@@ -5046,3 +5046,22 @@ erts_processes_monitoring_nodes(Process *c_p)
 
     return ctxt.res;
 }
+
+static void
+print_suspended_on_de(fmtfn_t to, void *to_arg, DistEntry *dep)
+{
+    for (; dep; dep = dep->next) {
+        ErtsProcList *curr = erts_proclist_peek_first(dep->suspended);
+        while (curr) {
+            if (!is_internal_pid(curr->u.pid))
+                print_process_info(to, to_arg, curr->u.p, 0);
+            curr = erts_proclist_peek_next(dep->suspended, curr);
+        }
+    }
+}
+
+void
+erts_dist_print_procs_suspended_on_de(fmtfn_t to, void *to_arg) {
+    print_suspended_on_de(to, to_arg, erts_hidden_dist_entries);
+    print_suspended_on_de(to, to_arg, erts_visible_dist_entries);
+}
diff --git a/erts/emulator/beam/dist.h b/erts/emulator/beam/dist.h
index 9b5e62ab7e..f953a2ab8c 100644
--- a/erts/emulator/beam/dist.h
+++ b/erts/emulator/beam/dist.h
@@ -319,5 +319,6 @@ extern int erts_dsig_prepare(ErtsDSigSendContext *,
                              int,
                              int);
 
+void erts_dist_print_procs_suspended_on_de(fmtfn_t to, void *to_arg);
 int erts_auto_connect(DistEntry* dep, Process *proc, ErtsProcLocks proc_locks);
 #endif
diff --git a/erts/emulator/beam/erl_process_dump.c b/erts/emulator/beam/erl_process_dump.c
index a164ed543e..71262061dd 100644
--- a/erts/emulator/beam/erl_process_dump.c
+++ b/erts/emulator/beam/erl_process_dump.c
@@ -121,12 +121,14 @@ Uint erts_process_memory(Process *p, int include_sigs_in_transit)
 
     size += sizeof(Process);
 
-    erts_link_tree_foreach(ERTS_P_LINKS(p),
-                           link_size, (void *) &size);
-    erts_monitor_tree_foreach(ERTS_P_MONITORS(p),
-                              monitor_size, (void *) &size);
-    erts_monitor_list_foreach(ERTS_P_LT_MONITORS(p),
-                              monitor_size, (void *) &size);
+    if ((erts_atomic32_read_nob(&p->state) & ERTS_PSFLG_EXITING) == 0) {
+        erts_link_tree_foreach(ERTS_P_LINKS(p),
+                               link_size, (void *) &size);
+        erts_monitor_tree_foreach(ERTS_P_MONITORS(p),
+                                  monitor_size, (void *) &size);
+        erts_monitor_list_foreach(ERTS_P_LT_MONITORS(p),
+                                  monitor_size, (void *) &size);
+    }
     size += (p->heap_sz + p->mbuf_sz) * sizeof(Eterm);
     if (p->abandoned_heap)
         size += (p->hend - p->heap) * sizeof(Eterm);
diff --git a/erts/emulator/test/dump_SUITE.erl b/erts/emulator/test/dump_SUITE.erl
index d0237b78cc..3b860ebdf6 100644
--- a/erts/emulator/test/dump_SUITE.erl
+++ b/erts/emulator/test/dump_SUITE.erl
@@ -24,7 +24,7 @@
 
 -export([all/0, suite/0, init_per_testcase/2, end_per_testcase/2]).
 
--export([signal_abort/1]).
+-export([signal_abort/1, exiting_dump/1, free_dump/1]).
 
 -export([load/0]).
 
@@ -35,7 +35,7 @@ suite() ->
      {timetrap, {minutes, 2}}].
 
 all() ->
-    [signal_abort].
+    [signal_abort, exiting_dump, free_dump].
 
 init_per_testcase(signal_abort, Config) ->
     SO = erlang:system_info(schedulers_online),
@@ -48,7 +48,10 @@ init_per_testcase(signal_abort, Config) ->
             {skip, "the platform does not support scheduler dump"};
        Dump ->
             Config
-    end.
+    end;
+init_per_testcase(_, Config) ->
+    Config.
+
 
 end_per_testcase(_, Config) ->
     Config.
@@ -79,8 +82,6 @@ signal_abort(Config) ->
 
     {ok, Bin} = get_dump_when_done(Dump),
 
-    ct:log("~s",[Bin]),
-
     {match, Matches} = re:run(Bin,"Current Process: <",[global]),
 
     ct:log("Found ~p",[Matches]),
@@ -91,6 +92,85 @@ signal_abort(Config) ->
 
     ok.
 
+load() ->
+    lists:seq(1,10000),
+    load().
+
+
+%% Test that crash dumping when a process is in the state EXITING works
+exiting_dump(Config) when is_list(Config) ->
+    Dump = filename:join(proplists:get_value(priv_dir, Config),"signal_abort.dump"),
+
+    {ok, Node} = start_node(Config),
+
+    Self = self(),
+
+    Pid = spawn_link(Node,
+                     fun() ->
+                             [begin
+                                  T = ets:new(hej,[]),
+                                  [ets:insert(T,{I,I}) || I <- lists:seq(1,1000)]
+                              end || _ <- lists:seq(1,1000)],
+                             Self ! ready,
+                             receive ok -> ok end
+                     end),
+
+    true = rpc:call(Node, os, putenv, ["ERL_CRASH_DUMP",Dump]),
+
+    receive ready -> unlink(Pid), Pid ! ok end,
+
+    rpc:call(Node, erlang, halt, ["dump"]),
+
+    {ok, Bin} = get_dump_when_done(Dump),
+
+    {match, Matches} = re:run(Bin,"^State: Exiting", [global, multiline]),
+
+    ct:log("Found ~p",[Matches]),
+
+    true = length(Matches) == 1,
+
+    file:delete(Dump),
+
+    ok.
+
+%% Test that crash dumping when a process is in the state FREE works
+free_dump(Config) when is_list(Config) ->
+    Dump = filename:join(proplists:get_value(priv_dir, Config),"signal_abort.dump"),
+
+    {ok, Node} = start_node(Config),
+
+    Self = self(),
+
+    Pid = spawn_link(Node,
+                     fun() ->
+                             Self ! ready,
+                             receive
+                                 ok ->
+                                     unlink(Self),
+                                     exit(lists:duplicate(1000,1000))
+                             end
+                     end),
+
+    true = rpc:call(Node, os, putenv, ["ERL_CRASH_DUMP",Dump]),
+
+    [erlang:monitor(process, Pid) || _ <- lists:seq(1,10000)],
+    receive ready -> unlink(Pid), Pid ! ok end,
+
+    rpc:call(Node, erlang, halt, ["dump"]),
+
+    {ok, Bin} = get_dump_when_done(Dump),
+
+    {match, Matches} = re:run(Bin,"^State: Non Existing", [global, multiline]),
+
+    ct:log("Found ~p",[Matches]),
+
+    true = length(Matches) == 1,
+
+    file:delete(Dump),
+
+    ok.
+
+
 get_dump_when_done(Dump) ->
     case file:read_file_info(Dump) of
         {ok, #file_info{ size = Sz }} ->
@@ -104,15 +184,13 @@ get_dump_when_done(Dump, Sz) ->
     timer:sleep(1000),
     case file:read_file_info(Dump) of
         {ok, #file_info{ size = Sz }} ->
-            file:read_file(Dump);
+            {ok, Bin} = file:read_file(Dump),
+            ct:log("~s",[Bin]),
+            {ok, Bin};
         {ok, #file_info{ size = NewSz }} ->
             get_dump_when_done(Dump, NewSz)
     end.
 
-load() ->
-    lists:seq(1,10000),
-    load().
-
 start_node(Config) when is_list(Config) ->
     Pa = filename:dirname(code:which(?MODULE)),
     Name = list_to_atom(atom_to_list(?MODULE)