[squid-users] client->Squid: TCP [RST] and [RST,ACK]

Maciej Leks maciej.leks at gmail.com
Tue Feb 28 20:34:45 UTC 2023 

wt., 28 lut 2023 o 20:06 Alex Rousskov
<rousskov at measurement-factory.com> napisał(a):
>
> On 2/28/23 13:07, Maciej Leks wrote:
> >>Does child Squid bump the TLS client connection, tunnel it, or
> >>terminates it (i.e. the child works as a reverse proxy)?
>
> > * client connects to child Squid's http_port without "ssl-bump".
> > * client sends a plain text CONNECT request to child Squid.
> > * child Squid connects to parent using cache_peer with mTLS.
>
> I assume that, by mTLS, you mean a cache_peer directive with "tls"
> option _and_ with "sslcert" (or equivalent) option to specify the TLS
> client certificate to be sent to the cache peer.
>
> I assume that the packets you pasted earlier were between a child Squid
> and its parent.

Exactly.

> >> What are those TLS alerts?
> > Code 21 - decryption_failed_RESERVED(21).
>
> I would focus on this error and ignore RST packets for now. Most likely,
> this error is what triggers those packets.
>
>
> HTH,
>
> Alex.

According to your experience - who sends this TLS "Encrypted Alert"
and RST - either the server (Salesforce) or parent squid?

Rgrds,
Maciek

> >>If those TLS alerts are close_notify alerts, then the RST packets you
> >>are seeing are probably triggered by the other side doing clean TLS
> >>shutdown (i.e. sending a TLS close_notify alert before closing the
> >>connection), either after receiving a FIN packet (unlikely with
> >>half_closed_clients off?) or while that FIN packet is being generated or
> >>is in-flight. When Such RST packets would be sent by the TCP stack
> >>rather than Squid code itself. They may be an indication of a benign
> >>race condition, a bug, or some other deficiency.
> >
> > half_closed_clients is off
> >
> >>Higher-level information (who initiates closure and why) may be needed
> >>to figure this out. I recommend sharing a link to a compressed ALL,9
> >>cache.log reproducing the problem with a single transaction _combined_
> >>with a matching packet trace file in libpcap format.
> >
> > There is no other way than turning debug on.
> >
> > Maciek
> >
> > wt., 28 lut 2023 o 15:15 Alex Rousskov <rousskov at measurement-factory.com
> > <mailto:rousskov at measurement-factory.com>> napisał(a):
> >
> >     On 2/28/23 08:35, Maciej Leks wrote:
> >      > Am I right in saying that RST it's a design intent of squid to end
> >      > connections quickly? I've started digging into the squid code and see
> >      > SO_LINGER and timeout set to 0, which means that it's done on purpose
> >      > not to hang on connections in TIME_WAIT state?
> >
> >     Does child Squid bump the TLS client connection, tunnel it, or
> >     terminates it (i.e. the child works as a reverse proxy)?
> >
> >     What are those TLS alerts?
> >
> >     If those TLS alerts are close_notify alerts, then the RST packets you
> >     are seeing are probably triggered by the other side doing clean TLS
> >     shutdown (i.e. sending a TLS close_notify alert before closing the
> >     connection), either after receiving a FIN packet (unlikely with
> >     half_closed_clients off?) or while that FIN packet is being
> >     generated or
> >     is in-flight. When Such RST packets would be sent by the TCP stack
> >     rather than Squid code itself. They may be an indication of a benign
> >     race condition, a bug, or some other deficiency.
> >
> >     Higher-level information (who initiates closure and why) may be needed
> >     to figure this out. I recommend sharing a link to a compressed ALL,9
> >     cache.log reproducing the problem with a single transaction _combined_
> >     with a matching packet trace file in libpcap format.
> >
> >     https://wiki.squid-cache.org/SquidFaq/BugReporting#debugging-a-single-transaction <https://wiki.squid-cache.org/SquidFaq/BugReporting#debugging-a-single-transaction>
> >
> >     HTH,
> >
> >     Alex.
> >
> >
> >      > wt., 28 lut 2023 o 08:12 Maciej Leks <maciej.leks at gmail.com
> >     <mailto:maciej.leks at gmail.com>> napisał(a):
> >      >>
> >      >> For a couple of days we've  been encountering many ECONNRESET error
> >      >> messages in our nodejs client to Salesforce. Even if access.log
> >     shows
> >      >> only /200 tshark shows a messy communication between client->squid
> >      >> (child squid).
> >      >>
> >      >> The architecture: nodejs client->3..* child squid->2*parent
> >      >> squids->cloud Salesforce
> >      >>
> >      >> Some examples ([RST]]: 46 1.015513576 100.121.10.169 → 100.113.27.73
> >      >> TCP 66 46098 → 3128 [ACK] Seq=1721 Ack=140135 Win=214656 Len=0
> >      >> TSval=2672547296 TSecr=1443424287
> >      >>
> >      >>     47 1.016152326 100.113.27.73 → 100.121.10.169 TCP 66 3128 →
> >     46098
> >      >> [FIN, ACK] Seq=140135 Ack=1721 Win=42368 Len=0 TSval=1443424288
> >      >> TSecr=2672547296
> >      >>     48 1.017856001 100.121.10.169 → 100.113.27.73 TLSv1.2 97
> >     Encrypted Alert
> >      >>     49 1.017893411 100.121.10.169 → 100.113.27.73 TCP 66 46098 →
> >     3128
> >      >> [FIN, ACK] Seq=1752 Ack=140136 Win=214656 Len=0 TSval=2672547298
> >      >> TSecr=1443424288
> >      >>     50 1.018002285 100.113.27.73 → 100.121.10.169 TCP 54 3128 →
> >     46098
> >      >> [RST] Seq=140136 Win=0 Len=0
> >      >>     51 1.018019806 100.113.27.73 → 100.121.10.169 TCP 54 3128 →
> >     46098
> >      >> [RST] Seq=140136 Win=0 Len=0
> >      >>
> >      >> [RST,ACK]:
> >      >> 592 67.664585034 100.121.10.169 → 100.113.27.73 TLSv1.2 97
> >     Encrypted Alert
> >      >>    593 67.664737552 100.113.27.73 → 100.121.10.169 TCP 66 3128 →
> >     52202
> >      >> [ACK] Seq=7973 Ack=1129 Win=42752 Len=0 TSval=1443490937
> >      >> TSecr=2672613945
> >      >>    594 67.664841613 100.121.10.169 → 100.113.27.73 TCP 66 52202
> >     → 3128
> >      >> [FIN, ACK] Seq=1129 Ack=7973 Win=42368 Len=0 TSval=2672613945
> >      >> TSecr=1443490937
> >      >>    595 67.664895660 100.113.27.73 → 100.121.10.169 TCP 66 3128 →
> >     52202
> >      >> [RST, ACK] Seq=7973 Ack=1129 Win=42752 Len=0 TSval=1443490937
> >      >> TSecr=2672613945
> >      >>    596 67.664936264 100.113.27.73 → 100.121.10.169 TCP 54 3128 →
> >     52202
> >      >> [RST] Seq=7973 Win=0 Len=0
> >      >>
> >      >> I'm wondering how to debug it (what exactly) and whether the reason
> >      >> may be on the squid side (specific configuration?).
> >      >>
> >      >> env: GKE/k8s
> >      >> client container: alpine linux
> >      >> child squid container: alpine linux
> >      >> version: 5.7
> >      >>
> >      >> Cheers,
> >      >> Maciek Leks
> >      > _______________________________________________
> >      > squid-users mailing list
> >      > squid-users at lists.squid-cache.org
> >     <mailto:squid-users at lists.squid-cache.org>
> >      > http://lists.squid-cache.org/listinfo/squid-users
> >     <http://lists.squid-cache.org/listinfo/squid-users>
> >
> >     _______________________________________________
> >     squid-users mailing list
> >     squid-users at lists.squid-cache.org
> >     <mailto:squid-users at lists.squid-cache.org>
> >     http://lists.squid-cache.org/listinfo/squid-users
> >     <http://lists.squid-cache.org/listinfo/squid-users>
> >
> >
> > _______________________________________________
> > squid-users mailing list
> > squid-users at lists.squid-cache.org
> > http://lists.squid-cache.org/listinfo/squid-users
>
> _______________________________________________
> squid-users mailing list
> squid-users at lists.squid-cache.org
> http://lists.squid-cache.org/listinfo/squid-users

More information about the squid-users mailing list