RH9代理服务器又垮了，提示内核stack overflow? 盼高手指点!

jayson2004

我的代理服务器(RH9平台，IPTABLE作NAT）又宕机了，一天几次，无法忍受了！在线客户机大概100个左右。宕机的时候键盘、网络无响应。服务器硬件配置1个志强2.4/512M/40GIDE ，性能应该足够了。是否内核(网卡)中断的原因？还是内存的问题?
盼高手指点!

/var/log/messages提示:

Dec 18 10:05:14 proxy kernel: do_IRQ: stack overflow: 812
Dec 18 10:05:14 proxy kernel: dc9a89dc 0000032c 00000000 dd634840 c020b490 dc9a94c0 dc9a8a38 c010d768
Dec 18 10:05:14 proxy kernel: dd634840 00000170 00000000 c020b490 dc9a94c0 dc9a8a38 00000000 00000068
Dec 18 10:05:14 proxy kernel: 00000068 ffffff0b e0a1e7a0 00000060 00010206 00000000 c03d81a0 dc9a9a98
Dec 18 10:05:14 proxy kernel: Call Trace: [<c020b490>] ip_rcv_finish [kernel] 0x0 (0xdc9a89ec))
Dec 18 10:05:14 proxy kernel: [<c010d768>] call_do_IRQ [kernel] 0x5 (0xdc9a89f8))
Dec 18 10:05:14 proxy kernel: [<c020b490>] ip_rcv_finish [kernel] 0x0 (0xdc9a8a08))
Dec 18 10:05:14 proxy kernel: [<e0a1e7a0>] __constant_c_and_count_memset [ipfilter] 0x56 (0xdc9a8a24))
Dec 18 10:05:14 proxy kernel: [<e0a1e34e>] linuxmag_hook [ipfilter] 0x89 (0xdc9a8a3c))
Dec 18 10:05:14 proxy kernel: [<e09f9a95>] ip_nat_cheat_check_R1e4e73a8 [iptable_nat] 0x35 (0xdc9a99ec))
Dec 18 10:05:14 proxy kernel: [<e09fa81b>] manip_pkt [iptable_nat] 0x9b (0xdc9a9a04))
Dec 18 10:05:14 proxy kernel: [<e09fac45>] icmp_reply_translation [iptable_nat] 0xd5 (0xdc9a9a24))
Dec 18 10:05:14 proxy kernel: [<e09f9240>] ip_nat_fn [iptable_nat] 0x1e0 (0xdc9a9a64))
Dec 18 10:05:14 proxy kernel: [<c020b490>] ip_rcv_finish [kernel] 0x0 (0xdc9a9a90))
Dec 18 10:05:14 proxy kernel: [<c01fc264>] nf_iterate [kernel] 0x54 (0xdc9a9a9c))
Dec 18 10:05:14 proxy kernel: [<c020b490>] ip_rcv_finish [kernel] 0x0 (0xdc9a9ab0))
Dec 18 10:05:14 proxy kernel: [<c01fc532>] nf_hook_slow [kernel] 0x72 (0xdc9a9ac4))
Dec 18 10:05:14 proxy kernel: [<c020b490>] ip_rcv_finish [kernel] 0x0 (0xdc9a9ae0))
Dec 18 10:05:14 proxy kernel: [<e0a1edc8>] linuxmag_ops [ipfilter] 0x0 (0xdc9a9ae4))
Dec 18 10:05:14 proxy kernel: [<c020b267>] ip_rcv [kernel] 0x197 (0xdc9a9af8))
Dec 18 10:05:14 proxy kernel: [<c020b490>] ip_rcv_finish [kernel] 0x0 (0xdc9a9b10))
Dec 18 10:05:15 proxy kernel: [<c01f5dae>] netif_receive_skb [kernel] 0x13e (0xdc9a9b28))
Dec 18 10:05:15 proxy kernel: [<c01f5eed>] process_backlog [kernel] 0x6d (0xdc9a9b48))
Dec 18 10:05:15 proxy kernel: [<c01f5ffa>] net_rx_action [kernel] 0x6a (0xdc9a9b60))
Dec 18 10:05:15 proxy kernel: [<c0121685>] do_softirq [kernel] 0x95 (0xdc9a9b7c))
Dec 18 10:05:15 proxy kernel: [<c010ac8e>] do_IRQ [kernel] 0xbe (0xdc9a9b94))
Dec 18 10:05:15 proxy kernel: [<c010d768>] call_do_IRQ [kernel] 0x5 (0xdc9a9bb4))
Dec 18 10:05:15 proxy kernel: [<c0150068>] do_execve [kernel] 0x18 (0xdc9a9bd4))
Dec 18 10:05:15 proxy kernel: [<c0144013>] page_remove_rmap [kernel] 0xd3 (0xdc9a9be0))
Dec 18 10:05:15 proxy kernel: [<c012eabd>] vm_pte_clear [kernel] 0x2d (0xdc9a9bec))
Dec 18 10:05:15 proxy kernel: [<c01311aa>] zap_pte_range [kernel] 0x15a (0xdc9a9c04))
Dec 18 10:05:15 proxy kernel: [<c01340be>] do_generic_file_read [kernel] 0x24e (0xdc9a9c14))
Dec 18 10:05:15 proxy kernel: [<c012f18b>] zap_page_range [kernel] 0xcb (0xdc9a9c2c))
Dec 18 10:05:15 proxy kernel: [<c013270f>] exit_mmap [kernel] 0xaf (0xdc9a9c6c))
Dec 18 10:05:15 proxy kernel: [<c014f971>] exec_mmap [kernel] 0x161 (0xdc9a9c90))
Dec 18 10:05:15 proxy kernel: [<c014f994>] flush_old_exec [kernel] 0x14 (0xdc9a9cac))
Dec 18 10:05:15 proxy kernel: [<c0167474>] load_elf_binary [kernel] 0x2b4 (0xdc9a9cc0))
Dec 18 10:05:15 proxy kernel: [<c011ca1e>] ll_copy_from_user [kernel] 0x4e (0xdc9a9de0))
Dec 18 10:05:15 proxy kernel: [<c01671c0>] load_elf_binary [kernel] 0x0 (0xdc9a9df0))
Dec 18 10:05:15 proxy kernel: [<c014ffc2>] search_binary_handler [kernel] 0x102 (0xdc9a9dfc))
Dec 18 10:05:15 proxy kernel: [<c01501cb>] do_execve [kernel] 0x17b (0xdc9a9e44))
Dec 18 10:05:15 proxy kernel: [<c0107c00>] sys_execve [kernel] 0x50 (0xdc9a9fa4))
Dec 18 10:05:15 proxy kernel: [<c0109537>] system_call [kernel] 0x33 (0xdc9a9fc0))

xiaohaomei

将linux以nat方式作为代理服务器时，它所能支持的用户是很有限的。因为一般进来的网络请求包比较小，而出去的应答包都比较大，就拿web服务来说吧！因此，此时该代理服务器很容易成为整个系统的瓶颈，也就是说它的扩展性很有限。我建议你可以先尝试只允许20个客户，然后30个，这样去测试系统在可以稳定运行的情况下可以支持的客户数。