[ruby-core:113347] [Ruby master Bug#19606] addr2line.c broken on Fedora 38
From:
"kjtsanaktsidis (KJ Tsanaktsidis) via ruby-core" <ruby-core@...>
Date:
2023-04-26 07:31:30 UTC
List:
ruby-core #113347
Issue #19606 has been updated by kjtsanaktsidis (KJ Tsanaktsidis).
This is also causing test failures on my machine, because there are tests on the output of the bug reporter - e.g.
```
[17591/23407] TestBugReporter#test_bug_reporter_add = 0.15 s
4) Failure:
TestBugReporter#test_bug_reporter_add [/home/kj/ruby/test/-ext-/bug_reporter/test_bug_reporter.rb:31]:
pid 169551 exit 1
| -:1: [BUG] Segmentation fault at 0x000003e80002964f
| ruby 3.3.0dev (2023-04-24T03:48:15Z master 886986b3ef) [x86_64-linux]
|
| -- Control frame information -----------------------------------------------
| c:0003 p:---- s:0012 e:000011 CFUNC :kill
| c:0002 p:0022 s:0006 e:000005 EVAL -:1 [FINISH]
| c:0001 p:0000 s:0003 E:001100 DUMMY [FINISH]
|
| -- Ruby level backtrace information ----------------------------------------
| -:1:in `<main>'
| -:1:in `kill'
|
| -- Threading information ---------------------------------------------------
| Total ractor count: 1
| Ruby thread count for this ractor: 1
|
| -- Machine register context ------------------------------------------------
| RIP: 0x00007f0793a6adab RBP: 0x000000000000000b RSP: 0x00007ffe339ef5b8
| RAX: 0x0000000000000000 RBX: 0x0000000000000001 RCX: 0x00007f0793a6adab
| RDX: 0x000000000002964f RDI: 0x000000000002964f RSI: 0x000000000000000b
| R8: 0x0000000000000000 R9: 0x0000000000000000 R10: 0x00007f0793a434e8
| R11: 0x0000000000000206 R12: 0x0000000000000002 R13: 0x00007f0793927048
| R14: 0x000000000002964f R15: 0x0000000000000001 EFL: 0x0000000000000206
|
| -- C level backtrace information -------------------------------------------
| 1344: Abbrev Number 27547 not found
..
1. [2/2] Assertion for "stderr"
| Expected /Sample bug reporter: 12345/
| to match
| "-- Control frame information -----------------------------------------------\n"+
| "c:0003 p:---- s:0012 e:000011 CFUNC :kill\n"+
| "c:0002 p:0022 s:0006 e:000005 EVAL -:1 [FINISH]\n"+
| "c:0001 p:0000 s:0003 E:001100 DUMMY [FINISH]\n\n"+
| "-- Ruby level backtrace information ----------------------------------------\n"+
| "-:1:in `<main>'\n"+
| "-:1:in `kill'\n\n"+
| "-- Threading information ---------------------------------------------------\n"+
| "Total ractor count: 1\n"+
| "Ruby thread count for this ractor: 1\n\n"+
| "-- Machine register context ------------------------------------------------\n"+
| " RIP: 0x00007f0793a6adab RBP: 0x000000000000000b RSP: 0x00007ffe339ef5b8\n"+
| " RAX: 0x0000000000000000 RBX: 0x0000000000000001 RCX: 0x00007f0793a6adab\n"+
| " RDX: 0x000000000002964f RDI: 0x000000000002964f RSI: 0x000000000000000b\n"+
| " R8: 0x0000000000000000 R9: 0x0000000000000000 R10: 0x00007f0793a434e8\n"+
| " R11: 0x0000000000000206 R12: 0x0000000000000002 R13: 0x00007f0793927048\n"+
| " R14: 0x000000000002964f R15: 0x0000000000000001 EFL: 0x0000000000000206\n\n"+
| "-- C level backtrace information -------------------------------------------\n"+
| "1344: Abbrev Number 27547 not found\n"
| after 4 patterns with 123 characters.
````
----------------------------------------
Bug #19606: addr2line.c broken on Fedora 38
https://bugs.ruby-lang.org/issues/19606#change-102907
* Author: kjtsanaktsidis (KJ Tsanaktsidis)
* Status: Open
* Priority: Normal
* Backport: 3.0: UNKNOWN, 3.1: UNKNOWN, 3.2: UNKNOWN
----------------------------------------
I'm running the Fedora 38 beta on my machine, and the Ruby crash reporter is itself crashing while trying to print C-level backtraces, with an error like this:
```
-- C level backtrace information -------------------------------------------
1344: Abbrev Number 27547 not found
```
This seems to happen because the debuginfo provided by Fedora 38 (for e.g. the libc frames) is compressed with dwz. Some debginfo is shared between multiple files, and a reference to the shared file is put in the `.gnu_debugaltlink` attribute of the debug object. Then, the main debug object contains forms of type `DW_FORM_GNU_ref_alt` and `DW_FORM_GNU_strp_alt`, whose values will refer to offsets inside the debugaltlink file.
For example, libc symbols contain a `.gnu_debugaltlink` section
```
% readelf -x .gnu_debugaltlink /usr/lib/debug/lib64/libc.so.6-2.37-1.fc38.x86_64.debug
Hex dump of section '.gnu_debugaltlink':
0x00000000 2e2e2f2e 64777a2f 676c6962 632d322e ../.dwz/glibc-2.
0x00000010 33372d31 2e666333 382e7838 365f3634 37-1.fc38.x86_64
0x00000020 002a30a8 c71ff2fd aa58a637 226c0f56 .*0......X.7"l.V
0x00000030 3ffd674a de ?.gJ.
```
Some DWARF abbrev's contain references to this shared data:
```
% readelf --debug-dump /usr/lib/debug/lib64/libc.so.6-2.37-1.fc38.x86_64.debug
...
Contents of the .debug_abbrev section (loaded from /usr/lib/debug/lib64/libc.so.6-2.37-1.fc38.x86_64.debug):
...
50 DW_TAG_typedef [no children]
DW_AT_name DW_FORM_GNU_strp_alt
DW_AT_decl_file DW_FORM_data1
DW_AT_decl_line DW_FORM_data1
DW_AT_decl_column DW_FORM_data1
DW_AT_type DW_FORM_GNU_ref_alt
DW_AT value: 0 DW_FORM value: 0
...
```
Because addr2line.c doesn't know how to read these `DW_FORM_GNU_*` forms in `debug_info_reader_read_value`, it doesn't advance the reader at all and so the DWARF parsing becomes confused. This can lead to a garbage abbrev number being read, or even to just straight up segfaults.
I have a patch which simply skips over the right number of bytes for these forms, without actually reading any of the data: https://github.com/ruby/ruby/pull/7731. This was sufficient to make the crash reporter work again properly on my machine.
_Technically_ speaking, perhaps we should _actually_ open the `.gnu_debugaltlink` file and dig out the referenced attribute value - this would be required if filename strings were put into the shared dwz info with `DW_FORM_GNU_strp_apt`. If you think this is necessary I can try and put together a follow up patch to do this. However I've not actually seen a debuginfo file that does this, yet.
--
https://bugs.ruby-lang.org/
______________________________________________
ruby-core mailing list -- [email protected]
To unsubscribe send an email to [email protected]
ruby-core info -- https://ml.ruby-lang.org/mailman3/postorius/lists/ruby-core.ml.ruby-lang.org/