From ea1ddd8800c82fb8ae7ac2f54f4217a35d6f463d Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Gustavsson?= <bjorn@erlang.org>
Date: Wed, 11 Oct 2017 14:45:11 +0200
Subject: Dump literals separately to avoid incomplete heap data
When a literal was used from several processes, the literal would
be dumped in only one of the processes. The other processes
that referenced the literals would have incomplete heap data.
---
lib/observer/src/crashdump_viewer.erl | 52 ++++++++++++++++++++++++++--
lib/observer/test/crashdump_viewer_SUITE.erl | 5 ++-
2 files changed, 53 insertions(+), 4 deletions(-)
(limited to 'lib/observer')
diff --git a/lib/observer/src/crashdump_viewer.erl b/lib/observer/src/crashdump_viewer.erl
index bc3ce5d547..ddf2879a97 100644
--- a/lib/observer/src/crashdump_viewer.erl
+++ b/lib/observer/src/crashdump_viewer.erl
@@ -104,6 +104,7 @@
-define(index_table,index_table).
-define(instr_data,instr_data).
-define(internal_ets,internal_ets).
+-define(literals,literals).
-define(loaded_modules,loaded_modules).
-define(memory,memory).
-define(memory_map,memory_map).
@@ -785,6 +786,9 @@ parse_vsn_str(Str,WS) ->
%%% Progress is reported during the time and MUST be checked with
%%% crashdump_viewer:get_progress/0 until it returns {ok,done}.
do_read_file(File) ->
+ erase(?literals), %Clear literal cache.
+ put(truncated,false), %Not truncated (yet).
+ erase(truncated_reason), %Not truncated (yet).
case file:read_file_info(File) of
{ok,#file_info{type=regular,
access=FileA,
@@ -856,6 +860,19 @@ indexify(Fd,AddrAdj,Bin,N) ->
{?proc_heap,LastId} ->
[{_,LastPos}] = lookup_index(?proc_heap,LastId),
ets:insert(cdv_heap_file_chars,{LastId,N+Start+1-LastPos});
+ {?literals,[]} ->
+ case get(truncated_reason) of
+ undefined ->
+ [{_,LastPos}] = lookup_index(?literals,[]),
+ ets:insert(cdv_heap_file_chars,
+ {literals,N+Start+1-LastPos});
+ _ ->
+ %% Literals are truncated. Make sure we never
+ %% attempt to read in the literals. (Heaps that
+ %% references literals will show markers for
+ %% incomplete heaps, but will otherwise work.)
+ delete_index(?literals, [])
+ end;
_ -> ok
end,
indexify(Fd,AddrAdj,Rest,N1);
@@ -908,6 +925,7 @@ check_if_truncated() ->
find_truncated_proc({Tag,_Id}) when Tag==?atoms;
Tag==?binary;
Tag==?instr_data;
+ Tag==?literals;
Tag==?memory_status;
Tag==?memory_map ->
put(truncated_proc,false);
@@ -1386,13 +1404,33 @@ maybe_other_node2(Channel) ->
expand_memory(Fd,Pid,DumpVsn) ->
BinAddrAdj = get_bin_addr_adj(DumpVsn),
put(fd,Fd),
- Dict = read_heap(Fd,Pid,BinAddrAdj,gb_trees:empty()),
+ Dict0 = case get(?literals) of
+ undefined ->
+ Literals = read_literals(Fd),
+ put(?literals,Literals),
+ put(fd,Fd),
+ Literals;
+ Literals ->
+ Literals
+ end,
+ Dict = read_heap(Fd,Pid,BinAddrAdj,Dict0),
Expanded = {read_stack_dump(Fd,Pid,BinAddrAdj,Dict),
read_messages(Fd,Pid,BinAddrAdj,Dict),
read_dictionary(Fd,Pid,BinAddrAdj,Dict)},
erase(fd),
Expanded.
+read_literals(Fd) ->
+ case lookup_index(?literals,[]) of
+ [{_,Start}] ->
+ [{_,Chars}] = ets:lookup(cdv_heap_file_chars,literals),
+ init_progress("Reading literals",Chars),
+ pos_bof(Fd,Start),
+ read_heap(0,gb_trees:empty());
+ [] ->
+ gb_trees:empty()
+ end.
+
%%%-----------------------------------------------------------------
%%% This is a workaround for a bug in dump versions prior to 0.3:
%%% Addresses were truncated to 32 bits. This could cause binaries to
@@ -2829,6 +2867,10 @@ reset_tables() ->
insert_index(Tag,Id,Pos) ->
ets:insert(cdv_dump_index_table,{{Tag,Pos},Id}).
+delete_index(Tag,Id) ->
+ Ms = [{{{Tag,'$1'},Id},[],[true]}],
+ ets:select_delete(cdv_dump_index_table, Ms).
+
lookup_index({Tag,Id}) ->
lookup_index(Tag,Id);
lookup_index(Tag) ->
@@ -2845,6 +2887,7 @@ insert_binary_index(Addr,Pos) ->
lookup_binary_index(Addr) ->
ets:lookup(cdv_binary_index_table,Addr).
+
%%-----------------------------------------------------------------
%% Convert tags read from crashdump to atoms used as first part of key
%% in cdv_dump_index_table
@@ -2862,6 +2905,7 @@ tag_to_atom("hidden_node") -> ?hidden_node;
tag_to_atom("index_table") -> ?index_table;
tag_to_atom("instr_data") -> ?instr_data;
tag_to_atom("internal_ets") -> ?internal_ets;
+tag_to_atom("literals") -> ?literals;
tag_to_atom("loaded_modules") -> ?loaded_modules;
tag_to_atom("memory") -> ?memory;
tag_to_atom("mod") -> ?mod;
@@ -2885,8 +2929,10 @@ tag_to_atom(UnknownTag) ->
%%%-----------------------------------------------------------------
%%% Store last tag for use when truncated, and reason if aborted
put_last_tag(?abort,Reason) ->
- %% Don't overwrite the real last tag
- put(truncated_reason,Reason);
+ %% Don't overwrite the real last tag, and make sure to return
+ %% the previous last tag.
+ put(truncated_reason,Reason),
+ get(last_tag);
put_last_tag(Tag,Id) ->
put(last_tag,{Tag,Id}).
diff --git a/lib/observer/test/crashdump_viewer_SUITE.erl b/lib/observer/test/crashdump_viewer_SUITE.erl
index 0ec6d8dcfc..6ac9d7d3fb 100644
--- a/lib/observer/test/crashdump_viewer_SUITE.erl
+++ b/lib/observer/test/crashdump_viewer_SUITE.erl
@@ -613,7 +613,10 @@ truncate_dump(File) ->
{win32,_} -> <<"\r\n">>;
_ -> <<"\n">>
end,
- [StartBin,AfterTag] = binary:split(Bin,BinTag),
+ %% Split after "our binary" created by crashdump_helper
+ %% (it may not be the first binary).
+ RE = <<"\n=binary:(?=[0-9A-Z]+",NewLine/binary,"FF:010203)">>,
+ [StartBin,AfterTag] = re:split(Bin,RE,[{parts,2}]),
[AddrAndSize,BinaryAndRest] = binary:split(AfterTag,Colon),
[Binary,_Rest] = binary:split(BinaryAndRest,NewLine),
TruncSize = byte_size(Binary) - 2,
--
cgit v1.2.3