mod_ssl, block-on-read problem?

am 25.01.2005 06:53:03 von Bob Tanner

Have the folling:

apache-1.3.33
libapache-mod-ssl-2.8.22
kernel-2.4.26-1-686-smp

Having a problem where https connections just won't die. Over time the
process table files and box crawls or falls to its knees.

Installed debugging version of apache, here is gdb's backtrace showing the
block on read() called from mod_ssl's ssl_io_unregister() function.

(gdb) bt
#0 0x40048a98 in read () from /lib/libpthread.so.0
#1 0x40357a30 in ?? () from /usr/lib/apache/1.3/mod_ssl.so
#2 0x40340a95 in ssl_io_unregister () from /usr/lib/apache/1.3/mod_ssl.so
#3 0x080730ae in ap_hook_call_func (ap=0x0, he=0x3, hf=0x809cdac)
at ap_hook.c:721
#4 0x08072d00 in ap_hook_call (
hook=0xfffffe00

) at ap_hook.c:382
^^^^^^^^^^^^^^^^^^^^^^^ problem here too?

#5 0x080536b7 in ap_read (fb=0x809cd64, buf=0x809cdac, nbyte=4096)
at buff.c:254
#6 0x08053bf3 in read_with_errors (fb=0x809cd64, buf=0x809cdac, nbyte=4096)
at buff.c:298
#7 0x08053ec5 in ap_bgets (buff=0xbfffc240 "", n=8192, fb=0x809cd64)
at buff.c:882
#8 0x08063423 in ap_getline (s=0xbfffc240 "", n=8192, in=0x809cd64, fold=0)
at http_protocol.c:867
#9 0x0806367e in read_request_line (r=0x8095074) at http_protocol.c:991
#10 0x08063d13 in ap_read_request (conn=0x8336f3c) at http_protocol.c:1187
#11 0x08060a80 in child_main (child_num_arg=-512) at http_main.c:4862
#12 0x08060de7 in make_child (s=0xfffffe00, slot=5, now=-512)
at http_main.c:5052
#13 0x080610c7 in perform_idle_server_maintenance () at http_main.c:5237
#14 0x08061a48 in standalone_main (argc=1, argv=0xbfffe514) at
http_main.c:5500
#15 0x08061ff8 in main (argc=1, argv=0xbfffe514) at http_main.c:5768

Using strace -p I get:
read(3,

So, I'm pretty sure the block is happening on the read of file description 3.

Using lsof -n -p I get and looking for FD 3, I see this:

apache.db 21547 www-data 3u IPv4 16364769 TCP
xx.xxx.xxx.xxx:www->xxx.xxx.xxx.xxx:51923 (ESTABLISHED)

So I know fd 3 is an IPv4 client and the socket is established.

After collecting several days of data, the client IPs seem to be cache
servers.

Anyone know of a reason why the socket can't/won't close?

That's all the info I can think of reporting.

Anyone seen this problem before?

Since the problem doesn't happen with http, I -assume- this is a mod_ssl
problem, but if not, I'll try the apache mailing list.

Thanks.

--
Bob Tanner | Phone : (952)943-8700
http://www.mn-linux.org, Minnesota, Linux | Fax : (952)943-8500
Key fingerprint = AB15 0BDF BCDE 4369 5B42 1973 7CF1 A709 2CC1 B288
____________________________________________________________ __________
Apache Interface to OpenSSL (mod_ssl) www.modssl.org
User Support Mailing List modssl-users@modssl.org
Automated List Manager majordomo@modssl.org

Re: mod_ssl, block-on-read problem?

am 25.01.2005 07:02:16 von Bob Tanner

On Monday 24 January 2005 11:53 pm, Bob Tanner wrote:
> Have the folling:
>
> apache-1.3.33
> libapache-mod-ssl-2.8.22
> kernel-2.4.26-1-686-smp
>
> Having a problem where https connections just won't die. Over time the
> process table files and box crawls or falls to its knees.
>
> Installed debugging version of apache, here is gdb's backtrace showing the
> block on read() called from mod_ssl's ssl_io_unregister() function.

This url sounds like the problem I'm having.

http://www.issociate.de/board/post/44974/

Any solution to the above?

--
Bob Tanner | Phone : (952)943-8700
http://www.mn-linux.org, Minnesota, Linux | Fax : (952)943-8500
Key fingerprint = AB15 0BDF BCDE 4369 5B42 1973 7CF1 A709 2CC1 B288
____________________________________________________________ __________
Apache Interface to OpenSSL (mod_ssl) www.modssl.org
User Support Mailing List modssl-users@modssl.org
Automated List Manager majordomo@modssl.org

Re: mod_ssl, block-on-read problem?

am 25.01.2005 10:44:35 von Ringaby Anders

Hello Bob.

> #4 0x08072d00 in ap_hook_call (
> hook=0xfffffe00

) at ap_hook.c:382
> ^^^^^^^^^^^^^^^^^^^^^^^ problem here too?

I don't know about this one ....

> Using strace -p I get:
> read(3,
>
> Using lsof -n -p I get and looking for FD 3, I see this:
>
> apache.db 21547 www-data 3u IPv4 16364769 TCP
> xx.xxx.xxx.xxx:www->xxx.xxx.xxx.xxx:51923 (ESTABLISHED)

..... but when it comes to this one, I would say that the main reason
why a read() keeps hanging on a TCP socket is that the client or peer
has not shut down the connection, that is, the client or peer has not
performed a shutdown(), close() or exit().

Of course, you could argue that mod_ssl should have implemented a
timeout for conditions like that.

Maybe there is some kind of protocol problem here, in that mod_ssl is
still expecting data, while the client thinks there is no more data
to send, or maybe the client too is hanging in a read().

Regards

Anders

____________________________________________________________ __________
Apache Interface to OpenSSL (mod_ssl) www.modssl.org
User Support Mailing List modssl-users@modssl.org
Automated List Manager majordomo@modssl.org