RE: [PATCH 0/1] Fix hang in t5562, introduced in v2.21.0-rc1 (stack traces inside)

From: Randall S. Becker <rsbecker@nexbridge.com>
Date: Feb 15, 2019, 12:59 UTC
Message-ID: <000501d4c52e$4d485320$e7d8f960$@nexbridge.com>
In-Reply-To: <20190215034735.GF3064@jessie.local>

On February 14, 2019 22:48, Max Kirillov wrote:

> To: Randall S. Becker <rsbecker@nexbridge.com>
> Cc: 'Max Kirillov' <max@max630.net>; 'Johannes Schindelin via

GitGitGadget'

> <gitgitgadget@gmail.com>; git@vger.kernel.org; 'Junio C Hamano'
> <gitster@pobox.com>
> Subject: Re: [PATCH 0/1] Fix hang in t5562, introduced in v2.21.0-rc1

(stack

Show 86 quoted lines

> traces inside)
> 
> On Thu, Feb 14, 2019 at 06:33:56PM -0500, Randall S. Becker wrote:
> > Here is the full set of traces (from subtest 6, which just hung).
> > There are no I/O errors reported on any pipe or file descriptor. There
> > is one git process waiting for a read to occur but no one is doing any
> > writing. Most processes are sitting in waitpid, except for the
> > initiating git, which is waiting on a read that never receives data, so
> everyone is asleep and hung.
> > The git process sitting in read is reading from a PIPE, not a file.
> >
> > There are no other processes involved in the test that I can see.
> >
> > Perl (waiting for output to be read):
> >   waitpid + 0x130 (SLr)
> >   $n_EnterPriv + 0x280 (Milli)
> >   Perl_wait4pid + 0x130 (UCr)
> >   Perl_my_pclose + 0x4C0 (UCr)
> >   Perl_io_close + 0x180 (UCr)
> >   Perl_do_close + 0x620 (UCr)
> >   Perl_pp_close + 0xA70 (UCr)
> >   Perl_runops_standard + 0xF0 (UCr)
> >   S_run_body + 0x870 (UCr)
> >   perl_run + 0x2D0 (UCr)
> >   main + 0x3D0 (UCr)
> >
> > git-http-backend:
> >   waitpid + 0x320 (SLr)
> >   $n_EnterPriv + 0x280 (Milli)
> >   cleanup_children + 0x5D0 (UCr)
> >   cleanup_children_on_exit + 0x70 (UCr)
> >   git_atexit_dispatch + 0x200 (UCr)
> >   __process_atexit_functions + 0xA0 (DLL zcredll)
> >   CRE_TERMINATOR_ + 0xB50 (DLL zcredll)
> >   exit + 0x2A0 (DLL zcrtldll)
> >   die_webcgi + 0x240 (UCr)
> >   die_errno + 0x360 (UCr)
> >   write_or_die + 0x1C0 (UCr)
> >   end_headers + 0x1A0 (UCr)
> >   die_webcgi + 0x220 (UCr)
> >   die + 0x320 (UCr)
> >   inflate_request + 0x520 (UCr)
> >   run_service + 0xC20 (UCr)
> >   service_rpc + 0x530 (UCr)
> >   cmd_main + 0xD00 (UCr)
> >   main + 0x190 (UCr)
> >
> > git (one of them):
> >   read64_ + 0x140 (SLr)
> >   $n_EnterPriv + 0x280 (Milli)
> >   xread + 0x130 (UCr)
> >   read_in_full + 0x130 (UCr)
> >   get_packet_data + 0x4B0 (UCr)
> >   packet_read_with_status + 0x230 (UCr)
> >   packet_reader_read + 0x310 (UCr)
> >   receive_needs + 0x300 (UCr)
> >   upload_pack + 0x680 (UCr)
> >   cmd_upload_pack + 0x830 (UCr)
> >   run_builtin + 0x980 (UCr)
> >   handle_builtin + 0x570 (UCr)
> >   run_argv + 0x210 (UCr)
> >   cmd_main + 0x710 (UCr)
> >   main + 0x190 (UCr)
> >
> > bash:
> >   waitpid + 0x130 (SLr)
> >   $n_EnterPriv + 0x280 (Milli)
> >   waitchld + 0x1F0 (UCr)
> >   wait_for + 0xFD0 (UCr)
> >   execute_command_internal + 0x1990 (UCr)
> >   execute_command + 0xC0 (UCr)
> >   reader_loop + 0x4F0 (UCr)
> >   main + 0x1140 (UCr)
> >
> > git (the other one):
> >   waitpid + 0x130 (SLr)
> >   $n_EnterPriv + 0x280 (Milli)
> >   wait_or_whine + 0xE0 (UCr)
> >   finish_command + 0x100 (UCr)
> >   run_command + 0x1F0 (UCr)
> >   execv_dashed_external + 0x800 (UCr)
> >   run_argv + 0x250 (UCr)
> >   cmd_main + 0x710 (UCr)
> >   main + 0x190 (UCr)
> 
> This list does not say which process whose child but it seems like #3 is

child of

> #2 and #2 waits for #3, but #3 does not exit. Which is strange because it
> should have send SIGTERM to it. Could the git-upload-pack somehow be
> masking SIGTERM?

SIGTERM cannot be masked or easily caught on the platform, so I don't think that's it. waitpid fails if the supplied pid is invalid, so it's not that either. However, I am not certain SIGPIPE is raised if a waited read is posted on a different pipe.