[SR-Users] possible TCP deadlock (tls again?) // pike module not releasing IPs
miconda at gmail.com
Mon Dec 16 17:23:26 CET 2019
I pinged Victor to see if he can figure out what happens within the deb
building process that makes the libssl mutex fix not enabled.
The extra .so preload object should be still installed, try to see if it
On 16.12.19 12:09, Aymeric Moizard wrote:
> Good catch!
> As I said in my first mail, I also add the issue with latest 5.2.X so
> I suppose the deb package has the same issue for 52X.
> Is the extra binary to load still there? I will check that as soon as
> I'm online...
> Tks a lot!
> Le lun. 16 déc. 2019 à 11:16, Daniel-Constantin Mierla
> <miconda at gmail.com <mailto:miconda at gmail.com>> a écrit :
> for some reason the binary doesn't seem to have the libssl mutex
> fix, in my system with the libssl 1.1 gives:
> # kamailio -I
> Print out of kamailio internals
> Version: kamailio 5.3.1 (x86_64/linux) f36ac2
> Default config: /tmp/kamailio-5.3/etc/kamailio/kamailio.cfg
> Default paths to modules: /tmp/kamailio-5.3/lib64/kamailio/modules
> Compile flags: USE_TCP, USE_TLS, USE_SCTP, TLS_HOOKS,
> USE_RAW_SOCKS, DISABLE_NAGLE, USE_MCAST, DNS_IP_HACK, SHM_MMAP,
> PKG_MALLOC, Q_MALLOC, F_MALLOC, TLSF_MALLOC, DBG_SR_MEMORY,
> USE_FUTEX, FAST_LOCK-ADAPTIVE_WAIT, USE_DNS_CACHE,
> USE_DNS_FAILOVER, USE_NAPTR, USE_DST_BLACKLIST, HAVE_RESOLV_RES,
> DEFAULT PKG_SIZE=8MB
> DEFAULT SHM_SIZE=64MB
> TCP poll methods: poll, epoll_lt, epoll_et, sigio_rt, select
> Source code revision ID: f36ac2
> Compiled with: gcc 9.2.1
> Compiled architecture: x86_64
> Compiled on: 11:11:20 Dec 16 2019
> Thank you for flying kamailio!
> The important part above is the presence of
> TLS_PTHREAD_MUTEX_SHARED compile time flag in the output.
> Needs to be investigated why the dep packages have the kamailio
> binary without the libssl mutex fix enabled.
> On 16.12.19 09:22, Aymeric Moizard wrote:
>> Hi Daniel,
>> Tks a lot for lookint at it.
>> $ ldd /usr/lib/x86_64-linux-gnu/kamailio/modules/tls.so
>> linux-vdso.so.1 (0x00007fff997dd000)
>> libssl.so.1.1 => /usr/lib/x86_64-linux-gnu/libssl.so.1.1
>> libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6
>> libcrypto.so.1.1 =>
>> /usr/lib/x86_64-linux-gnu/libcrypto.so.1.1 (0x00007fe40ad03000)
>> libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2
>> libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0
>> /lib64/ld-linux-x86-64.so.2 (0x00007fe40ba4a000)
>> $ /usr/sbin/kamailio -I
>> Print out of kamailio internals
>> Version: kamailio 5.3.1 (x86_64/linux)
>> Default config: /etc/kamailio/kamailio.cfg
>> Default paths to modules:
>> Compile flags: USE_TCP, USE_TLS, USE_SCTP, TLS_HOOKS,
>> USE_RAW_SOCKS, DISABLE_NAGLE, USE_MCAST, DNS_IP_HACK, SHM_MMAP,
>> PKG_MALLOC, Q_MALLOC, F_MALLOC, TLSF_MALLOC, DBG_SR_MEMORY,
>> USE_FUTEX, FAST_LOCK-ADAPTIVE_WAIT, USE_DNS_CACHE,
>> USE_DNS_FAILOVER, USE_NAPTR, USE_DST_BLACKLIST, HAVE_RESOLV_RES
>> DEFAULT PKG_SIZE=8MB
>> DEFAULT SHM_SIZE=64MB
>> TCP poll methods: poll, epoll_lt, epoll_et, sigio_rt, select
>> Source code revision ID: unknown
>> Compiled with: gcc 6.3.0
>> Compiled architecture: x86_64
>> Compiled on:
>> Thank you for flying kamailio!
>> Additional note:
>> I have tried to better understand the pike module and after
>> reading the "end" of the module documentation,
>> I do better understand the "Tree of IP" and settings.
>> The pike documentation, for each settins and description, should
>> refer to the section "Chapter 3. Developer Guide",
>> otherwise, the parameters cannot be understood. Also, it's not
>> possible to understand, according to me, the real time
>> for removing an IP from the tree (removing it 100% or only last
>> node of IP)
>> Looking again at my statistics, I feel the first graph is
>> definitly showing an issue. This graph is showing
>> "$stat(location-users)" and "$stat(location-contacts)". During
>> the 10 hours, many users are banned, unregistred, etc..
>> so it is really not expected that the number of registred users
>> is maintained. From what I understand, the fact
>> that the stats went down when deadlock dissapeared obviouly means
>> kamailio threads was in a bad state for the
>> last 10 hours...
>> If you need more information, let me know...
>> Le lun. 16 déc. 2019 à 08:22, Daniel-Constantin Mierla
>> <miconda at gmail.com <mailto:miconda at gmail.com>> a écrit :
>> can you provide output of ldd for tls.so and output of
>> "kamailio -I" (that's an uppercase i)?
>> On 13.12.19 16:39, Aymeric Moizard wrote:
>>> Hi List,
>>> * In the past, I had deadlock which was, most probably,
>>> related to ssl1.1.
>>> We have discussed this issue, and a fix is supposed to
>>> workaround the issue that was detected.
>>> * With latest 5.2.X, I have experienced ONCE a similar
>>> behavior with TCP and TLS being mostly stuck. I have not
>>> been using this version much, but the fix was supposed to be
>>> in the core of kamailio.
>>> The status of the server this night:
>>> * I'm today running version: kamailio 5.3.1 (x86_64/linux),
>>> * Installed on stretch using
>>> http://deb.kamailio.org/kamailio53 repository.
>>> * This versions use libssl1.1
>>> * A user reported that he can't connect with TCP
>>> * An average of 5000 IPs per 10 minutes are being banned by
>>> the pike module
>>> (could be twice the same)
>>> * at the end of the outage, I had 2479 IP in my ipban
>>> htable. (which is equivalent to my statistics showing 2
>>> bans/IP every 10 minutes = 5000)
>>> * looking at my logs, it appears that most (ALL?) ip being
>>> banned... are my regular users.
>>> * looking at my logs, I can't understand why pike would
>>> block them.
>>> This is a graph for statistics on my service for the last 24
>>> Yesterday, at 22:18:39, kamailio started to BAN some IPs. 52
>>> IPs were banned in a period of 10 minutes. I can confirm
>>> this from my logs.
>>> My pike configuration is this one:
>>> modparam("pike", "sampling_time_unit", 2)
>>> modparam("pike", "reqs_density_per_unit", 64)
>>> modparam("pike", "remove_latency", 4)
>>> When detecting the issue, this morning, I typed:
>>> $> sudo kamctl stats
>>> $> sudo kamcmd htable.dump ipban
>>> //FAILURE (answer too large...)
>>> $> sudo kamctl trap
>>> Then, I started an agent with TCP and it worked...???
>>> Then, a few seconds, may be a minute after:
>>> $> sudo kamcmd htable.dump ipban
>>> //SUCCESS and shows 2479 banned ip.
>>> and... everything is back to normal in a few minutes.
>>> I haven't restarted kamailio, and all statistics are as
>>> expected, as usual.
>>> Thus, it looks that " sudo kamctl trap" has triggered
>>> something. I already
>>> experienced a similar behavior -when testing my ssl1.1
>>> deadlock last year-.
>>> 2 questions:
>>> 1/ I beleive my "pike" configuration should not ban users.
>>> Is my pike configuration wrong?
>>> As an example, pike has banned an IP sending one
>>> message/second. I believe my configuration should accept that?
>>> 2/ Could there still be a TLS issue with libssl1.1?
>>> This is the result of the "kamctl trap":
>>> Sorry for the long story & hoping to find a long term
>>> solution or at least a workaround!
>>> Antisip - http://www.antisip.com
>>> Kamailio (SER) - Users Mailing List
>>> sr-users at lists.kamailio.org <mailto:sr-users at lists.kamailio.org>
>> Daniel-Constantin Mierla -- www.asipto.com <http://www.asipto.com>
>> www.twitter.com/miconda <http://www.twitter.com/miconda> -- www.linkedin.com/in/miconda <http://www.linkedin.com/in/miconda>
>> Kamailio World Conference - April 27-29, 2020, in Berlin -- www.kamailioworld.com <http://www.kamailioworld.com>
>> Antisip - http://www.antisip.com
> Daniel-Constantin Mierla -- www.asipto.com <http://www.asipto.com>
> www.twitter.com/miconda <http://www.twitter.com/miconda> -- www.linkedin.com/in/miconda <http://www.linkedin.com/in/miconda>
> Kamailio World Conference - April 27-29, 2020, in Berlin -- www.kamailioworld.com <http://www.kamailioworld.com>
Daniel-Constantin Mierla -- www.asipto.com
www.twitter.com/miconda -- www.linkedin.com/in/miconda
Kamailio World Conference - April 27-29, 2020, in Berlin -- www.kamailioworld.com
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the sr-users