On Tue, Mar 18, 2008 at 01:38:16AM +0000, Simon Raven wrote:
> Package: lighttpd
> Version: 1.4.19-1
> Severity: normal
>
> hi,
>
> i'm filing this from an x86 box, but the bug is in PPC, i don't see this
> behaviour
> in x86. i've run the daemon through strace, and right when it spawns off a
> child with
> clone(), the parent SEGVs, making the pid file for s-s-d, and other issues,
> go buggery.
>
>
> a bit before, and a bit after what i said:
>
> =====
> 12428 rt_sigaction(SIGTTOU, {SIG_IGN}, {SIG_DFL}, 8) = 0
> 12428 rt_sigaction(SIGTTIN, {SIG_IGN}, {SIG_DFL}, 8) = 0
> 12428 rt_sigaction(SIGTSTP, {SIG_IGN}, {SIG_DFL}, 8) = 0
> 12428 clone(child_stack=0,
> flags=CLONE_CHILD_CLEARTID|CLONE_CHILD_SETTID|SIGCHLD,
> child_tidptr=0x4801c3e8) = 12429
> 12428 --- SIGSEGV (Segmentation fault) @ 0 (0) ---
> 12429 rt_sigaction(SIGPIPE, {SIG_IGN}, <unfinished ...>
> 12428 +++ killed by SIGSEGV +++
> 12429 <... rt_sigaction resumed> {SIG_DFL}, 8) = 0
> 12429 getsockname(11, {sa_family=AF_FILE, [EMAIL PROTECTED], [2]) = 0
> 12429 getpeername(11, {sa_family=AF_FILE, path="/var/run/slapd/ldapi"}, [23])
> = 0
> 12429 fcntl64(11, F_GETFD) = 0x1 (flags FD_CLOEXEC)
> =====
>
> the bug has existed since a few package versions ago. i've tried backports,
> same problem.
> as a consequnce of this bug, the package configuration fails.
>
> thanks
> simonWell, strace is not the proper tool to understand segfaults, and a backtrace would really help. Though, you will need to rebuild lighttpd keeping debug symbols, and run it again with ulimit -c unlimited, and setting your /proc/sys/kernel/core_pattern to e.g. /tmp/%e.%p.core so that lighttpd cores go to /tmp/lighttpd.$PID.core. Then use gdb to read the backtrace. Running lighttpd under valgrind may help. -- ·O· Pierre Habouzit ··O [EMAIL PROTECTED] OOO http://www.madism.org
pgplDCck9j2cS.pgp
Description: PGP signature
