git/list[1] front-page[2] threads[3] people[4] search[5] about
 

Re: [PATCH 00/10] Advertise trace2 SID in protocol capabilities

From
Josh Steadmon <steadmon@google.com>
Date
Nov 2, 2020, 22:20 UTC
Message-ID
<20201102222020.GA1904687@google.com>
In-Reply-To
<4f1a1bab-7ac7-b8dd-acb2-6aeb04be3171@jeffhostetler.com>
On 2020.10.30 11:54, Jeff Hostetler wrote:
Show 34 quoted lines
> 
> 
> On 10/29/20 5:32 PM, Josh Steadmon wrote:
> > In order to more easily debug remote operations, it is useful to be able
> > to inspect both client-side and server-side traces. This series allows
> > clients to record the server's trace2 session ID, and vice versa, by
> > advertising the SID in a new "trace2-sid" protocol capability.
> > 
> 
> Very nice!  This should be very helpful when matching up client and
> server commands.
> 
> 
> > Two questions in particular for reviewers:
> > 
> > 1) Is trace2/tr2_sid.h intended to be visible to the rest of the code,
> >    or is the trace2/ directory supposed to be opaque implementation
> >    detail? If the latter, would it be acceptable to move tr2_sid_get()
> >    into trace2.h?
> 
> I put all the trace2-private stuff in that sub-directory and gave
> everything tr2_ prefixes to hide the details as much as I could
> (and as an alternative to the usual tendency of having everything
> be static within a massive .c file).
> 
> So, yeah, my intent was to make all of it opaque.
> So that just trace2.h contains the official API.
> 
> Perhaps in trace2.c you could create:
> 
> const char *trace2_session_id(void)
> {
>     return tr2_sid_get();
> }
Done in V2, thanks.
Show 25 quoted lines
> > 2) upload-pack generally takes configuration via flags rather than
> >    gitconfig. From offline discussions, it sounds like this is an
> >    intentional choice to limit potential vulnerability from malicious
> >    configs in local repositories accessed via the file:// URL scheme. Is
> >    it reasonable to load the trace2.announceSID option from config files
> >    in upload-pack, or should this be changed to a flag?
> 
> I don't have the history to comment on this.
> 
> One thing to consider is that the SID for a Git process is built up
> from the SID of the parent and the unique data for the current process.
> So for example, if `git fetch` has SID `<sid1>` and it spawns
> `git upload-pack`, the child will have SID `<sid1>/<sid2>` and if that
> spawns `git index-pack`, that child will have `<sid1>/<sid2>/<sid3>`.
> This is very helpful when tracking down parent/child relationships
> and perf hunting.
> 
> This SID inheritance is passed via an environment variable to
> the child, which extends it and passes the longer version to its
> children.
> 
> So the value being passed between client and server over the
> protocol may look like `<sid1>/<sid2>/<sid3>` rather than just a
> single `<sid_x>` term.  For your purposes, do you want or care if
> you get the single term or the full SID ?

I'm not sure we care too much one way or the other. A single component of the SID should be enough to join client & server logs, but it's easier to just send the whole thing.

Show 14 quoted lines
> Also, there's nothing to stop someone from seeding that environment
> variable in their shell with some mischief before launching the
> top-level Git command.  So the above example might see the SID as
> `<mischief>/<sid1>/<sid2>/<sid3>`.  I'm not sure if this could be
> abused when inserted into the V0/V1/V2 protocol or your logging
> facility.
> 
>     $ GIT_TRACE2_EVENT=1 GIT_TRACE2_PARENT_SID=hello git version
>     {"event":"version","sid":"hello/20201030T154143.660608Z-H86606a97-
> P00001d30",...}
>     ...
> 
> So maybe we want to have a public API to return a pointer to just
> the final `<sid_x>` term ?  (Then again, maybe I worry too much.)

Yeah, it's certainly possible to muck with the SID as you describe, but I'm not sure I see any big problems that could be caused. If someone points out an issue I've missed, I'll be happy to change this, though.

> Thanks,
> Jeff
Thanks for the review!
Show 25 quoted lines
> > Josh Steadmon (10):
> >    docs: new capability to advertise trace2 SIDs
> >    docs: new trace2.advertiseSID option
> >    upload-pack: advertise trace2 SID in v0 capabilities
> >    receive-pack: advertise trace2 SID in v0 capabilities
> >    serve: advertise trace2 SID in v2 capabilities
> >    transport: log received server trace2 SID
> >    fetch-pack: advertise trace2 SID in capabilities
> >    upload-pack, serve: log received client trace2 SID
> >    send-pack: advertise trace2 SID in capabilities
> >    receive-pack: log received client trace2 SID
> > 
> >   Documentation/config/trace2.txt               |  4 +
> >   .../technical/protocol-capabilities.txt       | 13 ++-
> >   Documentation/technical/protocol-v2.txt       |  9 +++
> >   builtin/receive-pack.c                        | 16 ++++
> >   fetch-pack.c                                  | 11 +++
> >   send-pack.c                                   |  9 +++
> >   serve.c                                       | 19 +++++
> >   t/t5705-trace2-sid-in-capabilities.sh         | 79 +++++++++++++++++++
> >   transport.c                                   | 10 +++
> >   upload-pack.c                                 | 23 +++++-
> >   10 files changed, 190 insertions(+), 3 deletions(-)
> >   create mode 100755 t/t5705-trace2-sid-in-capabilities.sh
> > 
Previous: Jeff HostetlerNext: Junio C Hamano
Message 13 of 66 in “Advertise trace2 SID in protocol capabilities”
  1. 00/10 Advertise trace2 SID in protocol capabilitiesJosh Steadmon, Oct 29, 2020
  2. 01/10 docs: new capability to advertise trace2 SIDsJosh Steadmon, Oct 29, 2020
  3. 02/10 docs: new trace2.advertiseSID optionJosh Steadmon, Oct 29, 2020
  4. 03/10 upload-pack: advertise trace2 SID in v0 capabilitiesJosh Steadmon, Oct 29, 2020
  5. 05/10 serve: advertise trace2 SID in v2 capabilitiesJosh Steadmon, Oct 29, 2020
  6. 04/10 receive-pack: advertise trace2 SID in v0 capabilitiesJosh Steadmon, Oct 29, 2020
  7. 06/10 transport: log received server trace2 SIDJosh Steadmon, Oct 29, 2020
  8. 07/10 fetch-pack: advertise trace2 SID in capabilitiesJosh Steadmon, Oct 29, 2020
  9. 08/10 upload-pack, serve: log received client trace2 SIDJosh Steadmon, Oct 29, 2020
  10. 09/10 send-pack: advertise trace2 SID in capabilitiesJosh Steadmon, Oct 29, 2020
  11. 10/10 receive-pack: log received client trace2 SIDJosh Steadmon, Oct 29, 2020
  12. Jeff HostetlerOct 30, 2020
  13. Josh SteadmonNov 2, 2020
  14. Junio C HamanoNov 3, 2020
  15. Jeff HostetlerNov 5, 2020
  16. Josh SteadmonNov 10, 2020
  17. Junio C HamanoOct 30, 2020
  18. 00/11 Advertise trace2 SID in protocol capabilitiesJosh Steadmon, Nov 2, 2020
  19. 02/11 docs: new trace2.advertiseSID optionJosh Steadmon, Nov 2, 2020
  20. Junio C HamanoNov 3, 2020
  21. Josh SteadmonNov 5, 2020
  22. Junio C HamanoNov 5, 2020
  23. Johannes SchindelinNov 6, 2020
  24. 01/11 docs: new capability to advertise trace2 SIDsJosh Steadmon, Nov 2, 2020
  25. Junio C HamanoNov 3, 2020
  26. Jeff HostetlerNov 5, 2020
  27. Ævar Arnfjörð BjarmasonNov 12, 2020
  28. Junio C HamanoNov 12, 2020
  29. Jeff HostetlerNov 12, 2020
  30. Josh SteadmonNov 11, 2020
  31. 03/11 trace2: add a public function for getting the SIDJosh Steadmon, Nov 2, 2020
  32. 05/11 receive-pack: advertise trace2 SID in v0 capabilitiesJosh Steadmon, Nov 2, 2020
  33. 04/11 upload-pack: advertise trace2 SID in v0 capabilitiesJosh Steadmon, Nov 2, 2020
  34. Junio C HamanoNov 3, 2020
  35. Josh SteadmonNov 5, 2020
  36. 06/11 serve: advertise trace2 SID in v2 capabilitiesJosh Steadmon, Nov 2, 2020
  37. Junio C HamanoNov 4, 2020
  38. 07/11 transport: log received server trace2 SIDJosh Steadmon, Nov 2, 2020
  39. Junio C HamanoNov 4, 2020
  40. Josh SteadmonNov 11, 2020
  41. Jeff HostetlerNov 12, 2020
  42. 08/11 fetch-pack: advertise trace2 SID in capabilitiesJosh Steadmon, Nov 2, 2020
  43. Junio C HamanoNov 4, 2020
  44. Josh SteadmonNov 5, 2020
  45. Junio C HamanoNov 5, 2020
  46. Josh SteadmonNov 11, 2020
  47. 09/11 upload-pack, serve: log received client trace2 SIDJosh Steadmon, Nov 2, 2020
  48. Junio C HamanoNov 4, 2020
  49. Josh SteadmonNov 5, 2020
  50. 10/11 send-pack: advertise trace2 SID in capabilitiesJosh Steadmon, Nov 2, 2020
  51. 11/11 receive-pack: log received client trace2 SIDJosh Steadmon, Nov 2, 2020
  52. Junio C HamanoNov 3, 2020
  53. 00/11 Advertise session ID in protocol capabilitiesJosh Steadmon, Nov 11, 2020
  54. 06/11 serve: advertise session ID in v2 capabilitiesJosh Steadmon, Nov 11, 2020
  55. 08/11 fetch-pack: advertise session ID in capabilitiesJosh Steadmon, Nov 11, 2020
  56. 07/11 transport: log received server session IDJosh Steadmon, Nov 11, 2020
  57. 10/11 send-pack: advertise session ID in capabilitiesJosh Steadmon, Nov 11, 2020
  58. 09/11 upload-pack, serve: log received client session IDJosh Steadmon, Nov 11, 2020
  59. 11/11 receive-pack: log received client session IDJosh Steadmon, Nov 11, 2020
  60. 05/11 receive-pack: advertise session ID in v0 capabilitiesJosh Steadmon, Nov 11, 2020
  61. 04/11 upload-pack: advertise session ID in v0 capabilitiesJosh Steadmon, Nov 11, 2020
  62. 03/11 trace2: add a public function for getting the SIDJosh Steadmon, Nov 11, 2020
  63. 02/11 docs: new transfer.advertiseSID optionJosh Steadmon, Nov 11, 2020
  64. 01/11 docs: new capability to advertise session IDsJosh Steadmon, Nov 11, 2020
  65. Junio C HamanoNov 25, 2020
  66. Josh SteadmonNov 25, 2020

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.