[pkg-lxc-devel] apt http backend hangs in LXC container on ppc64el

[Thadeu Lima de Souza Cascardo]

On January 22, 2017 2:34:09 PM GMT-02:00, Antonio Terceiro <terceiro at debian.org> wrote:
>Hi,
>
>[sorry for the large cross-post, but I don't really know where the
>problem could be; please keep me in Cc: since I'm not debian-powerpc@
>not on deity@]
>
>The Campinas IBM team has generously made a few VMs available to run
>the
>Debian CI (https://ci.debian.net) on ppc64el. I am having some trouble
>to get it working properly due to an issue with LXC. I have done this
>countless times on amd64 and arm64, and never had any issues like this
>one, so that is why I suspect it's ppc64el-specific.
>
>`apt update` inside a LXC container hangs forever like this:
>
># apt update
>0% [Waiting for headers]^C
>
>Running `apt update` under strace shows this infinite loop at the end:
>
>_newselect(6, [5], [], NULL, {tv_sec=0, tv_usec=500000}) = 0 (Timeout)
>rt_sigprocmask(SIG_BLOCK, [WINCH], [], 8) = 0
>rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0
>0% [Waiting for headers]) = 25aders]", 25
>_newselect(6, [5], [], NULL, {tv_sec=0, tv_usec=500000}) = 0 (Timeout)
>rt_sigprocmask(SIG_BLOCK, [WINCH], [], 8) = 0
>rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0
>0% [Waiting for headers]) = 25aders]", 25
>_newselect(6, [5], [], NULL, {tv_sec=0, tv_usec=500000}) = 0 (Timeout)
>rt_sigprocmask(SIG_BLOCK, [WINCH], [], 8) = 0
>rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0
>0% [Waiting for headers]) = 25aders]", 25
>_newselect(6, [5], [], NULL, {tv_sec=0, tv_usec=500000}) = 0 (Timeout)
>rt_sigprocmask(SIG_BLOCK, [WINCH], [], 8) = 0
>rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0
>0% [Waiting for headers]) = 25aders]", 25
>_newselect(6, [5], [], NULL, {tv_sec=0, tv_usec=500000}) = 0 (Timeout)
>rt_sigprocmask(SIG_BLOCK, [WINCH], [], 8) = 0
>rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0
>0% [Waiting for headers]) = 25aders]", 25
>_newselect(6, [5], [], NULL, {tv_sec=0, tv_usec=500000}^Cstrace:
>Process 127 detached
> <detached ...>
>
>The host is a ppc64el VM running jessie. I tried LXC guests running
>both
>jessie and sid, and I get the same result on both.
>
>I have also tried all of the following:
>
>- running `apt update` from the host, when chroot'ed to the container
>  rootfs; just works normally
>
>- upgrading the kernel on the host to the latest one in
>jessie-backports
>  (4.8); did not help
>
>- upgrading LXC to the latest version from jessie-backports (2.0), and
>  starting the container with it; did not help

What is file descriptor 5? I would assume it's the http connection. Does wget/curl work fine inside the container? How is the network behavior in general? Are you using veth plus NAT? What if you set a namespace with unshare -n, setup a veth pair and set the same network setup? Does it work?
-- 
Sent from my Android device with K-9 Mail. Please excuse my brevity.