Oops on shutdown

[krizhanovsky]

tempesta.sh stop leads to Oops in dmesg:

    [tempesta] Shutdown Tempesta
    [tempesta] Close 1 listening sockets
    [tempesta] Close 1 backend sockets
    [tempesta]   Destroy server socket ffff88003d8e2740
    kmem_cache_destroy tfw_conn_cache: Slab cache still has objects
    CPU: 1 PID: 4475 Comm: rmmod Tainted: G           O 3.10.10+ #7
    Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Bochs 01/01/2011
     0000000000000000 ffff88003d573ea8 ffffffff824d2d9c ffff88003d573ec8
     ffffffff8209b75f ffff88003d718200 ffffffffa003de90 ffff88003d573ed8
     ffffffffa0023ae3 ffff88003d573ee8 ffffffffa00377c7 ffff88003d573f78
    Call Trace:
     [<ffffffff824d2d9c>] dump_stack+0x19/0x1b
     [<ffffffff8209b75f>] kmem_cache_destroy+0xd8/0xea
     [<ffffffffa0023ae3>] tfw_connection_exit+0x2b/0x30 [tempesta_fw]
     [<ffffffffa00377c7>] tfw_exit+0x20/0x3b [tempesta_fw]
     [<ffffffff82065cd6>] SyS_delete_module+0x19b/0x215
     [<ffffffff8202103f>] ? do_page_fault+0x9/0xe
     [<ffffffff824d7f12>] system_call_fastpath+0x16/0x1b

[krizhanovsky]

Should be easy to debug - the Oops is clear enough.

[vdmit11]

The fix turns out to be not so easy:
To empty the cache we need to iterate over all existing connections and close them.
But we don't maintain the list of opened connections.
Alexander, you told that there is a way to do that in the linux kernel, and there is a similar task in ss_drain_accept_queue(), so I'm re-assigning the issue to you.

[krizhanovsky]

The code evolved significantly since that time and we don't see the bug any more.