SS: properly implement ss_drain_accept_queue()

[krizhanovsky]

Rework ss_drain_accept_queue(), see TODO and FIXME comments in the function.

Somewhat linked with #100 in sense that Linux socket kernel API must be patched and/or partially reimplemented in Tempesta SS.

[krizhanovsky]

This becomes a bug after fixing #116 and #254. To reproduce the issue upstream server should be shutdown to generate multiple errors sent to client (I used wrk -c 1000 -t 20 -d 300s http://172.16.0.5/ to generate client load). This leads to many clients reconnection and with enough load we get following crash on Tempesta restart:

    [10888.819577] [tempesta] Stopping all modules...
    [10888.821087] BUG: unable to handle kernel paging request at 0000010000000198
    [10888.822001] IP: [<ffffffff810c072c>] __lock_acquire+0x9c/0x1620
    [10888.822001] PGD 0 
    [10888.822001] Oops: 0002 [#1] SMP
    [10888.822001] Modules linked in: tfw_sched_rr(O) tfw_sched_http(O) tfw_sched_hash(O) tempesta_fw(O) tempesta_db(O) tempesta_tls(O) nf_conntrack_ipv4 nf_defrag_ipv4 xt_conntrack nf_conntrack ipt_REJECT nf_reject_ipv4 xt_CHECKSUM iptable_mangle bridge stp llc ip6table_filter ip6_tables iptable_filter ip_tables ebtable_nat ebtables xfs libcrc32c crc32c_intel ghash_clmulni_intel aesni_intel ppdev aes_x86_64 glue_helper lrw gf128mul ablk_helper cryptd parport_pc parport i2c_piix4 input_leds led_class pcspkr dm_mirror dm_region_hash acpi_cpufreq i2c_core serio_raw dm_log dm_mod uinput btrfs xor raid6_pq ata_generic pata_acpi e1000 ata_piix floppy ipv6 crc_ccitt autofs4 [last unloaded: tempesta_tls]
    [10888.822001] CPU: 2 PID: 15123 Comm: sysctl Tainted: G           O    4.8.15-tfw #16
    [10888.822001] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.9.1-1.fc24 04/01/2014
    [10888.822001] task: ffff88011824e440 task.stack: ffff8800b39a8000
    [10888.822001] RIP: 0010:[<ffffffff810c072c>]  [<ffffffff810c072c>] __lock_acquire+0x9c/0x1620
    [10888.822001] RSP: 0018:ffff8800b39abb70  EFLAGS: 00010006
    [10888.822001] RAX: 0000010000000000 RBX: ffff88010454f560 RCX: 0000000000000000
    [10888.822001] RDX: 0000000000000000 RSI: 0000000000000000 RDI: 0000000000000000
    [10888.822001] RBP: ffff8800b39abc08 R08: 0000000000000001 R09: 0000000000000000
    [10888.822001] R10: ffff88011824e440 R11: 0000000000000001 R12: 0000000000000000
    [10888.822001] R13: 0000000000000000 R14: 0000000000000001 R15: 0000000000000000
    [10888.822001] FS:  00007ff5c3827740(0000) GS:ffff88013fd00000(0000) knlGS:0000000000000000
    [10888.822001] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
    [10888.822001] CR2: 0000010000000198 CR3: 00000000ab3e1000 CR4: 00000000003406e0
    [10888.822001] Stack:
    [10888.822001]  ffffffff82df7150 0000000000000003 ffff88011824e440 0000000000000003
    [10888.822001]  ffff88011824e440 ffff880118247000 ffff8800b39abbd0 0000000000000003
    [10888.822001]  ffff88011824e440 0000000000000003 ffff88011824e440 0000000000000003
    [10888.822001] Call Trace:
    [10888.822001]  [<ffffffff810c20ae>] lock_acquire+0xbe/0x1f0
    [10888.822001]  [<ffffffff815bae78>] ? inet_csk_listen_stop+0xe8/0x2a0
    [10888.822001]  [<ffffffff81652e08>] _raw_spin_lock+0x38/0x50
    [10888.822001]  [<ffffffff815bae78>] ? inet_csk_listen_stop+0xe8/0x2a0
    [10888.822001]  [<ffffffff815bae78>] inet_csk_listen_stop+0xe8/0x2a0
    [10888.822001]  [<ffffffff815bdf45>] tcp_close+0x45/0x460
    [10888.822001]  [<ffffffffa076fbc9>] ss_release+0x19/0x20 [tempesta_fw]
    [10888.822001]  [<ffffffffa0771c77>] tfw_sock_clnt_stop_all+0x47/0xd0 [tempesta_fw]
    [10888.822001]  [<ffffffffa0759d02>] tfw_cfg_stop+0x32/0x90 [tempesta_fw]
    [10888.822001]  [<ffffffffa076e015>] handle_sysctl_state_io+0x185/0x1b0 [tempesta_fw]
    [10888.822001]  [<ffffffffa076dec2>] ? handle_sysctl_state_io+0x32/0x1b0 [tempesta_fw]
    [10888.822001]  [<ffffffff8127a5e5>] proc_sys_call_handler+0xc5/0xe0
    [10888.822001]  [<ffffffff8127a614>] proc_sys_write+0x14/0x20
    [10888.822001]  [<ffffffff811f9f08>] __vfs_write+0x28/0x120
    [10888.822001]  [<ffffffff810bcbdc>] ? percpu_down_read+0x5c/0x90
    [10888.822001]  [<ffffffff811fd8d1>] ? __sb_start_write+0xd1/0xf0
    [10888.822001]  [<ffffffff811fd8d1>] ? __sb_start_write+0xd1/0xf0
    [10888.822001]  [<ffffffff811fa602>] vfs_write+0xb2/0x1b0
    [10888.822001]  [<ffffffff811fb959>] SyS_write+0x49/0xa0
    [10888.822001]  [<ffffffff810029c8>] do_syscall_64+0x58/0x110
    [10888.822001]  [<ffffffff8165385a>] entry_SYSCALL64_slow_path+0x25/0x25
    [10888.822001] Code: 10 44 89 44 24 1c 89 4c 24 20 e8 c0 da ff ff 48 85 c0 8b 4c 24 20 44 8b 44 24 1c 44 8b 4c 24 10 4c 8b 54 24 08 0f 84 d7 07 00 00 <f0> ff 80 98 01 00 00 8b 35 3f 19 21 02 45 8b aa 18 08 00 00 85 
    [10888.822001] RIP  [<ffffffff810c072c>] __lock_acquire+0x9c/0x1620
    [10888.822001]  RSP <ffff8800b39abb70>
    [10888.822001] CR2: 0000010000000198
    [10888.822001] ---[ end trace fb09fa3d4b91c406 ]---
    [10888.822001] Kernel panic - not syncing: Fatal exception in interrupt
    [10888.822001] Kernel Offset: disabled
    [10888.822001] Rebooting in 10 seconds..

The reason for the crash is that request socket which inet_csk_listen_stop() tries to lock is already closed and freed, because we removed wrong socket at ss_drain_accept_queue().