Bug #5744
closednetmap: 6.0.9 v14 backport causes known packet stalls from v14 implementation in "legacy" mode too
Added by Franco Fichtner almost 2 years ago. Updated 10 months ago.
Description
Hi,
We have shipped Suricata 6.0.9 to OPNsense 22.7.9 and got a significant number of reports that packet transmission is stalling completely since the update. We don't have the new API enabled. The same problems existed when we briefly shipped the v14 version in an earlier version of OPNsense this year and can also be seen on the 7 beta.
There are a number of changes in https://github.com/OISF/suricata/pull/7468/files that are not protected by USE_NEW_NETMAP_API ifdef logic and the change around removal of nm_dispatch() seems to hint at the possibility that the receive side introduces a problem via commonly used new function NetmapReadPackets().
Cheers,
Franco
Files
source-netmap.c.diff (2.46 KB) source-netmap.c.diff | Reverts a unnecessary change from Suricata 6.0.8 | Jeff Lucovsky, 12/07/2022 01:43 PM | |
source-netmap.c.v2.diff (5.48 KB) source-netmap.c.v2.diff | Jeff Lucovsky, 12/12/2022 03:18 PM | ||
netmap-patch.v3.diff (7.7 KB) netmap-patch.v3.diff | Jeff Lucovsky, 12/14/2022 02:29 PM | ||
gdb1.txt (6.49 KB) gdb1.txt | GDB output, no symbols | Clyne Snowtail, 12/16/2022 10:35 PM | |
netmap-patch.v4.diff (23.6 KB) netmap-patch.v4.diff | Bill Meeks, 01/24/2023 12:33 AM | ||
suricata.yaml (68.2 KB) suricata.yaml | Suricata Config | Clyne Snowtail, 02/06/2023 05:01 AM | |
netmap-patch-v5.diff (596 Bytes) netmap-patch-v5.diff | NETMAP_V14 patch for 6.0.10 branch | Bill Meeks, 02/14/2023 04:44 AM |
Updated by Victor Julien almost 2 years ago
- Status changed from New to Assigned
- Assignee changed from OISF Dev to Jeff Lucovsky
- Priority changed from Normal to High
- Target version changed from TBD to 6.0.10
We need to decide if this can be fixed or otherwise revert the whole change.
Updated by Bill Meeks almost 2 years ago
As a point of reference:
I have essentially the same Netmap v14 patch in Suricata used on pfSense since August of last year, but have not seen reports logged of stalling traffic by users there. There are several users on pfSense that have netmap inline IPS mode running. Currently the release branch of pfSense is on FreeBSD 12.3-STABLE. The new development branch is FreeBSD 14, but I'm not sure how many of the development snapshot users are running netmap and inline IPS mode there. Don't know if the newer FreeBSD version may be in play here ???
I have not audited the patch we are using in pfSense against the latest updates to the Netmap v14 API in Suricata 6.x or 7.x, so it's possible there is a subtle difference in the patch that is impacting the OPNsense users. The current patch used on pfSense is here: https://github.com/pfsense/FreeBSD-ports/blob/devel/security/suricata/files/patch-netmap.diff. The core code changes should be the same because it is the code Jeff and I collaborated on last year. There have been some additions by Jeff to wrap more sections of code with the USE_NEW_NETMAP_API logic that we are not currently using within pfSense. Perhaps a bug snuck in that way ??
Updated by Jeff Lucovsky almost 2 years ago
What's your host operating system?
Can you share your Suricata netmap configuration? Is the host stack being used?
Updated by Bill Meeks almost 2 years ago
For additional info: the configuration for IPS mode in pfSense uses the host stack mode where one netmap endpoint is a hardware NIC, and the other endpoint is the host stack.
Updated by Franco Fichtner almost 2 years ago
We are using OPNsense 22.x, which is based on FreeBSD 13.1. Our config is simple:
- interface: igb0
copy-iface: igb0^
- interface: igb0^
copy-iface: igb0
Reproducibility differs largely, the users reporting issues say it happens after a day or so:
https://www.reddit.com/r/opnsense/comments/zbt3il/after_2279_update_the_gateway_suddenly_dies_after/
https://forum.opnsense.org/index.php?topic=31338.0
The FreeBSD port hasn't been updated yet.. it's either risk break by update or pull the v14 patch in its entirety. The latter we will have to do for OPNsense anyway to hotfix the situation.
Cheers,
Franco
Updated by Jeff Lucovsky almost 2 years ago
Franco, are you able to get any stack information from the Suricata worker threads when the stall occurs? Are you able to reproduce the issue in-house?
Bill, to what Suricata version does the patch apply to?
Updated by Franco Fichtner almost 2 years ago
How would I generate stack information in order to send to you? I have on instance here but so far still working. Users might give a helping hand.
Cheers,
Franco
Updated by Jeff Lucovsky almost 2 years ago
Is it possible to attach to the Suricata process with gdb -p <pid>
and then bt
?
Updated by Franco Fichtner almost 2 years ago
Perhaps adding the following might help:
We seem to be using...
runmode: workers
And this is the default?
runmode: autofp
So far I could not reproduce, but will keep it running overnight.
Cheers,
Franco
Updated by Bill Meeks almost 2 years ago
Jeff Lucovsky wrote in #note-7:
Bill, to what Suricata version does the patch apply to?
As I mentioned in the offline emails we exchanged, I am using my original version of the patch against Suricata 6.0.8. Also have used it successfully with 6.0.6 and older.
The core changes in the patch I'm using versus what is in the 6.0.9 codebase are the same so far as I know. The differences have to do with some of the #ifdef/#endif code section wrappers to accommodate both the older netmap API and the v14 API.
I am now compiling a binary package for pfSense testing using the base 6.0.9 code base WITHOUT my patch file and using the "netmap_v14" configure directive. Will test that in my virtual machine to see if the bug surfaces there. This will be on FreeBSD 14-CURRENT (which is what pfSense development is using).
Updated by Bill Meeks almost 2 years ago
I have a native Suricata 6.0.9 package with the NETMAP_V14 API enabled running in IPS mode on a pfSense virtual machine for testing. It has been running for a few hours now without a traffic interruption, but I will continue to monitor it for at least a full day. It is running with "autofp" mode. If I do not reproduce the stall with the current configuration, I will swap to "workers" mode and repeat the test in case the threading model is the key to triggering the stall.
Updated by Clyne Snowtail almost 2 years ago
Franco Fichtner wrote in #note-10:
Perhaps adding the following might help:
We seem to be using...
runmode: workers
And this is the default?
runmode: autofp
So far I could not reproduce, but will keep it running overnight.
Cheers,
Franco
If no one else can reproduce, I can attempt to after I return home Friday. I dont want to kill my WAN while Im remote.
For me, it was easy to trigger it with reasonably high throughput. Running speedtest.net on my 250/250mb WAN connection very reliably trips the stall somewhere during the upload test. Im not sure if its outgoing specific or just due to it hitting some limit that happens to occur during the uplink since speedtest.net always does down first then up.
Updated by Franco Fichtner almost 2 years ago
Thanks, so far no luck on my end.
Don't forget to install gdb first as it's no longer installed by default:
- pkg install gdb
Might be difficult to download when the traffic stopped ;)
Cheers,
Franco
Updated by Bill Meeks almost 2 years ago
Is anything logged by either Suricata or the system when the stall occurs?
The symptoms here seem suspiciously very similar to the old flow manager bug that was fixed last year (back in late June of 2021 as I recall). The symptoms with that bug were a sudden stoppage of traffic on a netmap interface. And just like described for this bug, the problem took some time to surface (many times more than 12 hours of normal operation and then the stall). Absolutely nothing was logged when that stall occurred. The network traffic on the interface just stopped.
Investigating further showed a type of thread deadlock. The problem only surfaced on FreeBSD because FreeBSD is very strict about allowing only the thread that obtained a lock to release that lock. The old flow manager code, in an effort to improve efficiency, was recycling queues (or something similar, sorry but my memory is foggy now) by passing them to other threads for use. The receiving thread would then attempt to unlock the queue prior to using it. This worked on Linux systems and most others, but failed on FreeBSD because it is undefined behavior for a non-owner thread to release a lock in FreeBSD.
Could this possibly be related to some new variation of this old bug?
P.S. -- my test virtual machine with 6.0.9 in netmap IPS mode is still operating normally. I will let it run a full 24 hours, then switch the threading model to "workers" mode and test again. Testing now with "autofp" mode.
Updated by Jeff Lucovsky almost 2 years ago
- File source-netmap.c.diff source-netmap.c.diff added
I've attached a patch that removes an unnecessary (but perhaps harmless) difference from the netmap packet source in 6.0.8.
franco and/or @bmeeks Would you be able to apply this patch to the Suricata 6.0.9 version of @src/source-netmap.c
and give things a try?
Updated by Bill Meeks almost 2 years ago
Jeff Lucovsky wrote in #note-16:
I've attached a patch that removes an unnecessary (but perhaps harmless) difference from the netmap packet source in 6.0.8.
franco and/or @bmeeks Would you be able to apply this patch to the Suricata 6.0.9 version of @src/source-netmap.c
and give things a try?
Looking over your patch, I think the changes at lines 900-905 are significant, and should have a positive impact on the bug reported by the OPNsense users.
I am using only the Netmap v14 API in my pfSense package, so never would have encountered any odd behavior with the legacy Netmap API. It should be noted that the legacy API and the v14 API use completely different netmap descriptor structures and also a different support library. Exclusive use of the v14 API may be why pfSense users have not reported the stall. In the patch file I am currently using for 6.0.8 and below on pfSense, all of the #ifdef/#endif conditionals around the Netmap v14 API were removed (along with the legacy code).
Updated by Franco Fichtner almost 2 years ago
Now there is three versions to try and reproduce:
1. Initial release of 6.0.9: # pkg add -f https://pkg.opnsense.org/FreeBSD:13:amd64/snapshots/misc/suricata-6.0.9.pkg
2. Full revert of v14: # pkg add -f https://pkg.opnsense.org/FreeBSD:13:amd64/snapshots/misc/suricata-6.0.9_1.pkg
3. Jeff's proposal: # pkg add -f https://pkg.opnsense.org/FreeBSD:13:amd64/snapshots/misc/suricata-6.0.9_2.pkg
I'm still unable to trigger the stall. We are trying to collect more info about the conditions this happens in (system, NIC, how many interfaces).
Cheers,
Franco
Updated by Bill Meeks almost 2 years ago
I, likewise, have thus far been unable to reproduce the stall condition in my testing with 6.0.9 using netmap in IPS mode.
Updated by Clyne Snowtail almost 2 years ago
Backtrace, failed on second speedtest
Hopefully its useful.
(gdb) bt
#0 0x0000000801b3e3ba in _nanosleep () from /lib/libc.so.7
#1 0x0000000800a4364c in ?? () from /lib/libthr.so.3
#2 0x0000000801ac1e46 in usleep () from /lib/libc.so.7
#3 0x00000000004edbaa in ?? ()
#4 0x00000000004ed501 in SuricataMain ()
#5 0x00000000003f9b90 in _start ()
Edit:
I somewhat suspect that suricata is tripping a bug someplace else, like in a driver. I think this due to the fact I restarted suricata after the above so I could run it again to see if the bt was different. The WAN came back as expected. I triggered the issue again and the WAN dropped once more. However I was unable to get it back after that second time. Even just stopping suricata and not running it, the WAN remained down. I was forced to reboot the VM.
Updated by Franco Fichtner almost 2 years ago
Sort of expected this without debug information compiled...
The bigger question for now is:
Is the issue reproducible in the _1 and _2 variants?
From another side we have confirmation about a deadlock happening inside emulated TX netmap kernel code, but I'm still surprised that 6.0.9 triggers a behavioural change that can be observed by multiple people.
Thanks,
Franco
Updated by Bill Meeks almost 2 years ago
Clyne Snowtail wrote in #note-20:
Backtrace, failed on second speedtest
Hopefully its useful.
(gdb) bt
#0 0x0000000801b3e3ba in _nanosleep () from /lib/libc.so.7
#1 0x0000000800a4364c in ?? () from /lib/libthr.so.3
#2 0x0000000801ac1e46 in usleep () from /lib/libc.so.7
#3 0x00000000004edbaa in ?? ()
#4 0x00000000004ed501 in SuricataMain ()
#5 0x00000000003f9b90 in _start ()
Edit:
I somewhat suspect that suricata is tripping a bug someplace else, like in a driver. I think this due to the fact I restarted suricata after the above so I could run it again to see if the bt was different. The WAN came back as expected. I triggered the issue again and the WAN dropped once more. However I was unable to get it back after that second time. Even just stopping suricata and not running it, the WAN remained down. I was forced to reboot the VM.
Set up a reliable console session with the test system (meaning one that is not dependent on the interface running Suricata). If you are testing a VM, then directly opening the VM's console will work. Next, determine the Process ID of the running Suricata process. Start a download or otherwise reproduce the stall condition. After the stall occurs, execute this command:
procstat -t <pid> (where <pid> is the Suricata process ID)
Look at the list of active threads and see if any are condition umtxn . This was the thread deadlock condition caused by that previous Flow Manager bug. Just want to see if a similar lock is happening. That was the state a thread locked up in when a non-owner thread attempted to release a lock previously obtained by some other thread. Because the symptoms here are so similar to the old bug, I'm wondering if another similar "lock/unlock" issue has surfaced ?? And as I stated previously, this is a FreeBSD-specific issue as Linux and most other OS variants are fine with non-owner threads releasing a lock generated by another thread.
Updated by Clyne Snowtail almost 2 years ago
PID TID COMM TDNAME CPU PRI STATE WCHAN
42798 150916 suricata - -1 120 sleep nanslp
42798 210817 suricata W#01-xn1 -1 120 sleep select
42798 210824 suricata W#01-xn1^ -1 120 sleep select
42798 210825 suricata FM#01 -1 120 sleep uwait
42798 210826 suricata FR#01 -1 120 sleep uwait
42798 210827 suricata CW -1 120 sleep uwait
42798 210828 suricata CS -1 120 sleep uwait
The above was true for both 6.0.9 and 6.0.9_2. 6.0.9_1 I have not had fail yet, but I will leave it in place for the time being.
_2 variant Im going to say is not the issue. It immediately failed just like vanilla.
Updated by Bill Meeks almost 2 years ago
Clyne Snowtail wrote in #note-23:
PID TID COMM TDNAME CPU PRI STATE WCHAN
42798 150916 suricata - -1 120 sleep nanslp
42798 210817 suricata W#01-xn1 -1 120 sleep select
42798 210824 suricata W#01-xn1^ -1 120 sleep select
42798 210825 suricata FM#01 -1 120 sleep uwait
42798 210826 suricata FR#01 -1 120 sleep uwait
42798 210827 suricata CW -1 120 sleep uwait
42798 210828 suricata CS -1 120 sleep uwaitThe above was true for both 6.0.9 and 6.0.9_2. 6.0.9_1 I have not had fail yet, but I will leave it in place for the time being.
_2 variant Im going to say is not the issue. It immediately failed just like vanilla.
Thank you for posting these results. Nothing wrong there, so it appears a new variation of the old bug is not at play.
Updated by Franco Fichtner almost 2 years ago
Clyne, thanks for the input!
I found a typo in the ifdef of the original patch:
https://github.com/OISF/suricata/pull/8274
I'm still looking through the differences... The changes in thread selection through NetmapGetRSSCount() / GetIfaceRSSQueuesNum() might still change this in unwanted direction for threads: auto mode.
Updated by Jeff Lucovsky almost 2 years ago
@Franco Fichtner Thanks for finding the typo -- i'll make sure that change is committed to the master-6.0.x branch when we resolve the current issue.
Updated by Jeff Lucovsky almost 2 years ago
@Franco Fichtner Have you tried the patch I prepared and attached to this issue?
Updated by Clyne Snowtail almost 2 years ago
Jeff Lucovsky wrote in #note-27:
@Franco Fichtner Have you tried the patch I prepared and attached to this issue?
He built Suricata with your patch as 6.0.9_2 and I installed that version. It failed as quick as the standard release.
Also Franco's 6.0.9_1 with the v14 changes backed out is what Im running right now and it has not had the same issues. Over 24 hours with no problem.
Updated by Bill Meeks almost 2 years ago
I am now leaning toward it being some area of the code where we forgot to provide the proper #ifdef/#endif conditional wrapper for some section of needed legacy netmap API code. That typo identified by Franco is one area, but there may be more.
I use the pure v14 API in the pfSense package with no issues. We have not used the legacy netmap API in pfSense since August of 2021. That is on both FreeBSD 12.3 STABLE and 14 CURRENT.
Updated by Franco Fichtner almost 2 years ago
The typo would likely only affect V14 operability if at all.
And, yes, Clyne said the full revert of the commits on top of 6.0.9 is the only one that is working normally (_1 variant).
The _2 with Jeff's proposed patch doesn't work either. I've also compared netmap code between netmap_user.c and libnetmap and the code seems to be equivalent indeed (minus idiosyncrasies that netmap authors built into their code for things that are not really being used in the header struct).
I've built a _3 variant here fixing the typo and backing out the configure.ac changes to see if the problem is actually with the code or not.
https://pkg.opnsense.org/FreeBSD:13:amd64/snapshots/misc/suricata-6.0.9_3.pkg
I've also compared debug log output between unmodified versions 6.0.8 and 6.0.9 but couldn't see a change in threads or attach behaviour for the ports being given in the configuration.
Cheers,
Franco
Updated by Jeff Lucovsky almost 2 years ago
ClyneSnowtail and @franco Can you tell me what Suricata runmode -- @autofp
or workers
your deployment uses?
Updated by Franco Fichtner almost 2 years ago
It's workers
, attached full yaml here...
# cat /usr/local/etc/suricata/suricata.yaml %YAML 1.1 --- # Suricata configuration file. In addition to the comments describing all # options in this file, full documentation can be found at: # https://suricata.readthedocs.io/en/latest/configuration/suricata-yaml.html ## ## Step 1: inform Suricata about your network ## vars: # more specific is better for alert accuracy and performance address-groups: HOME_NET: "[10.0.0.0/8]" EXTERNAL_NET: "!$HOME_NET" HTTP_SERVERS: "$HOME_NET" SMTP_SERVERS: "$HOME_NET" SQL_SERVERS: "$HOME_NET" DNS_SERVERS: "$HOME_NET" TELNET_SERVERS: "$HOME_NET" AIM_SERVERS: "$EXTERNAL_NET" DC_SERVERS: "$HOME_NET" DNP3_SERVER: "$HOME_NET" DNP3_CLIENT: "$HOME_NET" MODBUS_CLIENT: "$HOME_NET" MODBUS_SERVER: "$HOME_NET" ENIP_CLIENT: "$HOME_NET" ENIP_SERVER: "$HOME_NET" port-groups: HTTP_PORTS: "80" SHELLCODE_PORTS: "!80" ORACLE_PORTS: 1521 SSH_PORTS: 22 DNP3_PORTS: 20000 MODBUS_PORTS: 502 FILE_DATA_PORTS: "[$HTTP_PORTS,110,143]" FTP_PORTS: 21 VXLAN_PORTS: 4789 ## ## Step 2: select outputs to enable ## # The default logging directory. Any log or output file will be # placed here if its not specified with a full path name. This can be # overridden with the -l command line parameter. default-log-dir: /var/log/suricata/ # global stats configuration stats: enabled: yes # The interval field (in seconds) controls at what interval # the loggers are invoked. interval: 8 # Add decode events as stats. #decoder-events: true # Decoder event prefix in stats. Has been 'decoder' before, but that leads # to missing events in the eve.stats records. See issue #2225. #decoder-events-prefix: "decoder.event" # Add stream events as stats. #stream-events: false # Configure the type of alert (and other) logging you would like. outputs: # a line based alerts log similar to Snort's fast.log - fast: enabled: no filename: fast.log append: yes #filetype: regular # 'regular', 'unix_stream' or 'unix_dgram' # Extensible Event Format (nicknamed EVE) event log in JSON format - eve-log: enabled: yes filetype: regular #regular|syslog|unix_dgram|unix_stream|redis filename: eve.json #prefix: "@cee: " # prefix to prepend to each log entry # the following are valid when type: syslog above #identity: "suricata" #facility: local5 #level: Info ## possible levels: Emergency, Alert, Critical, ## Error, Warning, Notice, Info, Debug #redis: # server: 127.0.0.1 # port: 6379 # async: true ## if redis replies are read asynchronously # mode: list ## possible values: list|lpush (default), rpush, channel|publish # ## lpush and rpush are using a Redis list. "list" is an alias for lpush # ## publish is using a Redis channel. "channel" is an alias for publish # key: suricata ## key or channel to use (default to suricata) # Redis pipelining set up. This will enable to only do a query every # 'batch-size' events. This should lower the latency induced by network # connection at the cost of some memory. There is no flushing implemented # so this setting as to be reserved to high traffic suricata. # pipelining: # enabled: yes ## set enable to yes to enable query pipelining # batch-size: 10 ## number of entry to keep in buffer # Include top level metadata. Default yes. #metadata: no # include the name of the input pcap file in pcap file processing mode pcap-file: false # Community Flow ID # Adds a 'community_id' field to EVE records. These are meant to give # a records a predictable flow id that can be used to match records to # output of other tools such as Bro. # # Takes a 'seed' that needs to be same across sensors and tools # to make the id less predictable. # enable/disable the community id feature. community-id: false # Seed value for the ID output. Valid values are 0-65535. community-id-seed: 0 # HTTP X-Forwarded-For support by adding an extra field or overwriting # the source or destination IP address (depending on flow direction) # with the one reported in the X-Forwarded-For HTTP header. This is # helpful when reviewing alerts for traffic that is being reverse # or forward proxied. xff: enabled: no # Two operation modes are available, "extra-data" and "overwrite". mode: extra-data # Two proxy deployments are supported, "reverse" and "forward". In # a "reverse" deployment the IP address used is the last one, in a # "forward" deployment the first IP address is used. deployment: reverse # Header name where the actual IP address will be reported, if more # than one IP address is present, the last IP address will be the # one taken into consideration. header: X-Forwarded-For types: - alert: # packet: yes # enable dumping of packet (without stream segments) # metadata: no # enable inclusion of app layer metadata with alert. Default yes # http-body: yes # Requires metadata; enable dumping of http body in Base64 # http-body-printable: yes # Requires metadata; enable dumping of http body in printable format # Enable the logging of tagged packets for rules using the # "tag" keyword. tagged-packets: yes http: yes tls: yes - anomaly: # Anomaly log records describe unexpected conditions such # as truncated packets, packets with invalid IP/UDP/TCP # length values, and other events that render the packet # invalid for further processing or describe unexpected # behavior on an established stream. Networks which # experience high occurrences of anomalies may experience # packet processing degradation. # # Anomalies are reported for the following: # 1. Decode: Values and conditions that are detected while # decoding individual packets. This includes invalid or # unexpected values for low-level protocol lengths as well # as stream related events (TCP 3-way handshake issues, # unexpected sequence number, etc). # 2. Stream: This includes stream related events (TCP # 3-way handshake issues, unexpected sequence number, # etc). # 3. Application layer: These denote application layer # specific conditions that are unexpected, invalid or are # unexpected given the application monitoring state. # # By default, anomaly logging is disabled. When anomaly # logging is enabled, applayer anomaly reporting is # enabled. # enabled: yes # # Choose one or more types of anomaly logging and whether to enable # logging of the packet header for packet anomalies. types: # decode: no # stream: no # applayer: yes #packethdr: no # - http: # #extended: yes # enable this for extended logging information # # custom allows additional http fields to be included in eve-log # # the example below adds three additional fields when uncommented # #custom: [Accept-Encoding, Accept-Language, Authorization] # # set this value to one and only one among {both, request, response} # # to dump all http headers for every http request and/or response # # dump-all-headers: none # - dns: # # This configuration uses the new DNS logging format, # # the old configuration is still available: # # https://suricata.readthedocs.io/en/latest/output/eve/eve-json-output.html#dns-v1-format # # # As of Suricata 5.0, version 2 of the eve dns output # # format is the default. # #version: 2 # # # Enable/disable this logger. Default: enabled. # #enabled: yes # # # Control logging of requests and responses: # # - requests: enable logging of DNS queries # # - responses: enable logging of DNS answers # # By default both requests and responses are logged. # #requests: no # #responses: no # # # Format of answer logging: # # - detailed: array item per answer # # - grouped: answers aggregated by type # # Default: all # #formats: [detailed, grouped] # # # Types to log, based on the query type. # # Default: all. # #types: [a, aaaa, cname, mx, ns, ptr, txt] # - tls: # extended: yes # enable this for extended logging information # # output TLS transaction where the session is resumed using a # # session id # #session-resumption: no # # custom allows to control which tls fields that are included # # in eve-log # #custom: [subject, issuer, session_resumed, serial, fingerprint, sni, version, not_before, not_after, certificate, chain, ja3, ja3s] # - files: # force-magic: no # force logging magic on all logged files # # force logging of checksums, available hash functions are md5, # # sha1 and sha256 # #force-hash: [md5] - drop: alerts: yes # log alerts that caused drops flows: start # start or all: 'start' logs only a single drop # per flow direction. All logs each dropped pkt. # - smtp: # #extended: yes # enable this for extended logging information # # this includes: bcc, message-id, subject, x_mailer, user-agent # # custom fields logging from the list: # # reply-to, bcc, message-id, subject, x-mailer, user-agent, received, # # x-originating-ip, in-reply-to, references, importance, priority, # # sensitivity, organization, content-md5, date # #custom: [received, x-mailer, x-originating-ip, relays, reply-to, bcc] # # output md5 of fields: body, subject # # for the body you need to set app-layer.protocols.smtp.mime.body-md5 # # to yes # #md5: [body, subject] #- dnp3 # - ftp #- rdp # - nfs # - smb # - tftp # - ikev2 # - krb5 # - snmp #- sip # - dhcp: # enabled: yes # # When extended mode is on, all DHCP messages are logged # # with full detail. When extended mode is off (the # # default), just enough information to map a MAC address # # to an IP address is logged. # extended: no # - ssh # # - stats: # # totals: yes # stats for all threads merged together # # threads: no # per thread stats # # deltas: no # include delta values # # bi-directional flows # - flow # # uni-directional flows # #- netflow # Metadata event type. Triggered whenever a pktvar is saved # and will include the pktvars, flowvars, flowbits and # flowints. #- metadata # deprecated - unified2 alert format for use with Barnyard2 - unified2-alert: enabled: no # for further options see: # https://suricata.readthedocs.io/en/suricata-5.0.0/configuration/suricata-yaml.html#alert-output-for-use-with-barnyard2-unified2-alert # a line based log of HTTP requests (no alerts) - http-log: enabled: no filename: http.log append: yes #extended: yes # enable this for extended logging information #custom: yes # enabled the custom logging format (defined by customformat) #customformat: "%%D-%H:%M:%St.%z %X-Forwarded-Fori %H %m %h %u %s %B %a:%p -> %A:%P" #filetype: regular # 'regular', 'unix_stream' or 'unix_dgram' # a line based log of TLS handshake parameters (no alerts) - tls-log: enabled: no # Log TLS connections. filename: tls.log # File to store TLS logs. append: yes #extended: yes # Log extended information like fingerprint #custom: yes # enabled the custom logging format (defined by customformat) #customformat: "%%D-%H:%M:%St.%z %a:%p -> %A:%P %v %n %d %D" #filetype: regular # 'regular', 'unix_stream' or 'unix_dgram' # output TLS transaction where the session is resumed using a # session id #session-resumption: no # output module to store certificates chain to disk - tls-store: enabled: no #certs-log-dir: certs # directory to store the certificates files # Packet log... log packets in pcap format. 3 modes of operation: "normal" # "multi" and "sguil". # # In normal mode a pcap file "filename" is created in the default-log-dir, # or are as specified by "dir". # In multi mode, a file is created per thread. This will perform much # better, but will create multiple files where 'normal' would create one. # In multi mode the filename takes a few special variables: # - %n -- thread number # - %i -- thread id # - %t -- timestamp (secs or secs.usecs based on 'ts-format' # E.g. filename: pcap.%n.%t # # Note that it's possible to use directories, but the directories are not # created by Suricata. E.g. filename: pcaps/%n/log.%s will log into the # per thread directory. # # Also note that the limit and max-files settings are enforced per thread. # So the size limit when using 8 threads with 1000mb files and 2000 files # is: 8*1000*2000 ~ 16TiB. # # In Sguil mode "dir" indicates the base directory. In this base dir the # pcaps are created in th directory structure Sguil expects: # # $sguil-base-dir/YYYY-MM-DD/$filename.<timestamp> # # By default all packets are logged except: # - TCP streams beyond stream.reassembly.depth # - encrypted streams after the key exchange # - pcap-log: enabled: no filename: log.pcap # File size limit. Can be specified in kb, mb, gb. Just a number # is parsed as bytes. limit: 1000mb # If set to a value will enable ring buffer mode. Will keep Maximum of "max-files" of size "limit" max-files: 2000 # Compression algorithm for pcap files. Possible values: none, lz4. # Enabling compression is incompatible with the sguil mode. Note also # that on Windows, enabling compression will *increase* disk I/O. compression: none # Further options for lz4 compression. The compression level can be set # to a value between 0 and 16, where higher values result in higher # compression. #lz4-checksum: no #lz4-level: 0 mode: normal # normal, multi or sguil. # Directory to place pcap files. If not provided the default log # directory will be used. Required for "sguil" mode. #dir: /nsm_data/ #ts-format: usec # sec or usec second format (default) is filename.sec usec is filename.sec.usec use-stream-depth: no #If set to "yes" packets seen after reaching stream inspection depth are ignored. "no" logs all packets honor-pass-rules: no # If set to "yes", flows in which a pass rule matched will stopped being logged. # a full alerts log containing much information for signature writers # or for investigating suspected false positives. - alert-debug: enabled: no filename: alert-debug.log append: yes #filetype: regular # 'regular', 'unix_stream' or 'unix_dgram' # alert output to prelude (https://www.prelude-siem.org/) only # available if Suricata has been compiled with --enable-prelude - alert-prelude: enabled: no profile: suricata log-packet-content: no log-packet-header: yes # Stats.log contains data from various counters of the suricata engine. - stats: enabled: yes filename: stats.log append: yes # append to file (yes) or overwrite it (no) totals: yes # stats for all threads merged together threads: no # per thread stats #null-values: yes # print counters that have value 0 # a line based alerts log similar to fast.log into syslog - syslog: enabled: yes # reported identity to syslog. If omitted the program name (usually # suricata) will be used. #identity: "suricata" facility: local5 level: Notice ## possible levels: Emergency, Alert, Critical, ## Error, Warning, Notice, Info, Debug # deprecated a line based information for dropped packets in IPS mode - drop: enabled: no # further options documented at: # https://suricata.readthedocs.io/en/suricata-5.0.0/configuration/suricata-yaml.html#drop-log-a-line-based-information-for-dropped-packets # Output module for storing files on disk. Files are stored in a # directory names consisting of the first 2 characters of the # SHA256 of the file. Each file is given its SHA256 as a filename. # # When a duplicate file is found, the existing file is touched to # have its timestamps updated. # # Unlike the older filestore, metadata is not written out by default # as each file should already have a "fileinfo" record in the # eve.log. If write-fileinfo is set to yes, the each file will have # one more associated .json files that consists of the fileinfo # record. A fileinfo file will be written for each occurrence of the # file seen using a filename suffix to ensure uniqueness. # # To prune the filestore directory see the "suricatactl filestore # prune" command which can delete files over a certain age. - file-store: version: 2 enabled: no # Set the directory for the filestore. If the path is not # absolute will be be relative to the default-log-dir. #dir: filestore # Write out a fileinfo record for each occurrence of a # file. Disabled by default as each occurrence is already logged # as a fileinfo record to the main eve-log. #write-fileinfo: yes # Force storing of all files. Default: no. #force-filestore: yes # Override the global stream-depth for sessions in which we want # to perform file extraction. Set to 0 for unlimited. #stream-depth: 0 # Uncomment the following variable to define how many files can # remain open for filestore by Suricata. Default value is 0 which # means files get closed after each write #max-open-files: 1000 # Force logging of checksums, available hash functions are md5, # sha1 and sha256. Note that SHA256 is automatically forced by # the use of this output module as it uses the SHA256 as the # file naming scheme. #force-hash: [sha1, md5] # NOTE: X-Forwarded configuration is ignored if write-fileinfo is disabled # HTTP X-Forwarded-For support by adding an extra field or overwriting # the source or destination IP address (depending on flow direction) # with the one reported in the X-Forwarded-For HTTP header. This is # helpful when reviewing alerts for traffic that is being reverse # or forward proxied. xff: enabled: no # Two operation modes are available, "extra-data" and "overwrite". mode: extra-data # Two proxy deployments are supported, "reverse" and "forward". In # a "reverse" deployment the IP address used is the last one, in a # "forward" deployment the first IP address is used. deployment: reverse # Header name where the actual IP address will be reported, if more # than one IP address is present, the last IP address will be the # one taken into consideration. header: X-Forwarded-For # deprecated - file-store v1 - file-store: enabled: no # further options documented at: # https://suricata.readthedocs.io/en/suricata-5.0.0/file-extraction/file-extraction.html#file-store-version-1 # Log TCP data after stream normalization # 2 types: file or dir. File logs into a single logfile. Dir creates # 2 files per TCP session and stores the raw TCP data into them. # Using 'both' will enable both file and dir modes. # # Note: limited by stream.reassembly.depth - tcp-data: enabled: no type: file filename: tcp-data.log # Log HTTP body data after normalization, dechunking and unzipping. # 2 types: file or dir. File logs into a single logfile. Dir creates # 2 files per HTTP session and stores the normalized data into them. # Using 'both' will enable both file and dir modes. # # Note: limited by the body limit settings - http-body-data: enabled: no type: file filename: http-data.log # Lua Output Support - execute lua script to generate alert and event # output. # Documented at: # https://suricata.readthedocs.io/en/latest/output/lua-output.html - lua: enabled: no #scripts-dir: /etc/suricata/lua-output/ scripts: # - script1.lua # Logging configuration. This is not about logging IDS alerts/events, but # output about what Suricata is doing, like startup messages, errors, etc. logging: # The default log level, can be overridden in an output section. # Note that debug level logging will only be emitted if Suricata was # compiled with the --enable-debug configure option. # # This value is overridden by the SC_LOG_LEVEL env var. default-log-level: notice # The default output format. Optional parameter, should default to # something reasonable if not provided. Can be overridden in an # output section. You can leave this out to get the default. # # This value is overridden by the SC_LOG_FORMAT env var. #default-log-format: "[%i] %t - (%f:%l) <%d> (%n) -- " # A regex to filter output. Can be overridden in an output section. # Defaults to empty (no filter). # # This value is overridden by the SC_LOG_OP_FILTER env var. default-output-filter: # Define your logging outputs. If none are defined, or they are all # disabled you will get the default - console output. outputs: - console: enabled: no - syslog: enabled: yes facility: local5 format: "[%i] <%d> -- " ## ## Step 4: configure common capture settings ## ## See "Advanced Capture Options" below for more options, including NETMAP ## and PF_RING. ## # Linux high speed capture support af-packet: - interface: eth0 # Number of receive threads. "auto" uses the number of cores #threads: auto # Default clusterid. AF_PACKET will load balance packets based on flow. cluster-id: 99 # Default AF_PACKET cluster type. AF_PACKET can load balance per flow or per hash. # This is only supported for Linux kernel > 3.1 # possible value are: # * cluster_flow: all packets of a given flow are send to the same socket # * cluster_cpu: all packets treated in kernel by a CPU are send to the same socket # * cluster_qm: all packets linked by network card to a RSS queue are sent to the same # socket. Requires at least Linux 3.14. # * cluster_ebpf: eBPF file load balancing. See doc/userguide/capture-hardware/ebpf-xdp.rst for # more info. # Recommended modes are cluster_flow on most boxes and cluster_cpu or cluster_qm on system # with capture card using RSS (require cpu affinity tuning and system irq tuning) cluster-type: cluster_flow # In some fragmentation case, the hash can not be computed. If "defrag" is set # to yes, the kernel will do the needed defragmentation before sending the packets. defrag: yes # To use the ring feature of AF_PACKET, set 'use-mmap' to yes #use-mmap: yes # Lock memory map to avoid it goes to swap. Be careful that over subscribing could lock # your system #mmap-locked: yes # Use tpacket_v3 capture mode, only active if use-mmap is true # Don't use it in IPS or TAP mode as it causes severe latency #tpacket-v3: yes # Ring size will be computed with respect to max_pending_packets and number # of threads. You can set manually the ring size in number of packets by setting # the following value. If you are using flow cluster-type and have really network # intensive single-flow you could want to set the ring-size independently of the number # of threads: #ring-size: 2048 # Block size is used by tpacket_v3 only. It should set to a value high enough to contain # a decent number of packets. Size is in bytes so please consider your MTU. It should be # a power of 2 and it must be multiple of page size (usually 4096). #block-size: 32768 # tpacket_v3 block timeout: an open block is passed to userspace if it is not # filled after block-timeout milliseconds. #block-timeout: 10 # On busy system, this could help to set it to yes to recover from a packet drop # phase. This will result in some packets (at max a ring flush) being non treated. #use-emergency-flush: yes # recv buffer size, increase value could improve performance # buffer-size: 32768 # Set to yes to disable promiscuous mode # disable-promisc: no # Choose checksum verification mode for the interface. At the moment # of the capture, some packets may be with an invalid checksum due to # offloading to the network card of the checksum computation. # Possible values are: # - kernel: use indication sent by kernel for each packet (default) # - yes: checksum validation is forced # - no: checksum validation is disabled # - auto: suricata uses a statistical approach to detect when # checksum off-loading is used. # Warning: 'checksum-validation' must be set to yes to have any validation #checksum-checks: kernel # BPF filter to apply to this interface. The pcap filter syntax apply here. #bpf-filter: port 80 or udp # You can use the following variables to activate AF_PACKET tap or IPS mode. # If copy-mode is set to ips or tap, the traffic coming to the current # interface will be copied to the copy-iface interface. If 'tap' is set, the # copy is complete. If 'ips' is set, the packet matching a 'drop' action # will not be copied. #copy-mode: ips #copy-iface: eth1 # For eBPF and XDP setup including bypass, filter and load balancing, please # see doc/userguide/capture-hardware/ebpf-xdp.rst for more info. # Put default values here. These will be used for an interface that is not # in the list above. - interface: default #threads: auto #use-mmap: no #tpacket-v3: yes # Cross platform libpcap capture support pcap: - interface: default # On Linux, pcap will try to use mmaped capture and will use buffer-size # as total of memory used by the ring. So set this to something bigger # than 1% of your bandwidth. #buffer-size: 16777216 #bpf-filter: "tcp and port 25" # Choose checksum verification mode for the interface. At the moment # of the capture, some packets may be with an invalid checksum due to # offloading to the network card of the checksum computation. # Possible values are: # - yes: checksum validation is forced # - no: checksum validation is disabled # - auto: Suricata uses a statistical approach to detect when # checksum off-loading is used. (default) # Warning: 'checksum-validation' must be set to yes to have any validation #checksum-checks: auto # With some accelerator cards using a modified libpcap (like myricom), you # may want to have the same number of capture threads as the number of capture # rings. In this case, set up the threads variable to N to start N threads # listening on the same interface. #threads: 16 # set to no to disable promiscuous mode: promisc: no # promiscuous mode # set snaplen, if not set it defaults to MTU if MTU can be known # via ioctl call and to full capture if not. #snaplen: 1518 pcap-file: # Possible values are: # - yes: checksum validation is forced # - no: checksum validation is disabled # - auto: Suricata uses a statistical approach to detect when # checksum off-loading is used. (default) # Warning: 'checksum-validation' must be set to yes to have checksum tested checksum-checks: auto # See "Advanced Capture Options" below for more options, including NETMAP # and PF_RING. ## ## Step 5: App Layer Protocol Configuration ## # Configure the app-layer parsers. The protocols section details each # protocol. # # The option "enabled" takes 3 values - "yes", "no", "detection-only". # "yes" enables both detection and the parser, "no" disables both, and # "detection-only" enables protocol detection only (parser disabled). app-layer: protocols: krb5: enabled: yes snmp: enabled: yes ikev2: enabled: yes tls: enabled: yes detection-ports: dp: 443 # Generate JA3 fingerprint from client hello. If not specified it # will be disabled by default, but enabled if rules require it. ja3-fingerprints: yes # What to do when the encrypted communications start: # - default: keep tracking TLS session, check for protocol anomalies, # inspect tls_* keywords. Disables inspection of unmodified # 'content' signatures. # - bypass: stop processing this flow as much as possible. No further # TLS parsing and inspection. Offload flow bypass to kernel # or hardware if possible. # - full: keep tracking and inspection as normal. Unmodified content # keyword signatures are inspected as well. # # For best performance, select 'bypass'. # #encryption-handling: default dcerpc: enabled: yes ftp: enabled: yes # memcap: 64mb # RDP, disabled by default. rdp: #enabled: no ssh: enabled: yes smtp: enabled: yes raw-extraction: no # Configure SMTP-MIME Decoder mime: # Decode MIME messages from SMTP transactions # (may be resource intensive) # This field supercedes all others because it turns the entire # process on or off decode-mime: yes # Decode MIME entity bodies (ie. base64, quoted-printable, etc.) decode-base64: yes decode-quoted-printable: yes # Maximum bytes per header data value stored in the data structure # (default is 2000) header-value-depth: 2000 # Extract URLs and save in state data structure extract-urls: yes # Set to yes to compute the md5 of the mail body. You will then # be able to journalize it. body-md5: no # Configure inspected-tracker for file_data keyword inspected-tracker: content-limit: 100000 content-inspect-min-size: 32768 content-inspect-window: 4096 imap: enabled: detection-only smb: enabled: yes detection-ports: dp: 139, 445 # Stream reassembly size for SMB streams. By default track it completely. #stream-depth: 0 nfs: enabled: yes tftp: enabled: yes dns: # memcaps. Globally and per flow/state. #global-memcap: 16mb #state-memcap: 512kb # How many unreplied DNS requests are considered a flood. # If the limit is reached, app-layer-event:dns.flooded; will match. #request-flood: 500 tcp: enabled: yes detection-ports: dp: 53 udp: enabled: yes detection-ports: dp: 53 http: enabled: yes # memcap: Maximum memory capacity for http # Default is unlimited, value can be such as 64mb # default-config: Used when no server-config matches # personality: List of personalities used by default # request-body-limit: Limit reassembly of request body for inspection # by http_client_body & pcre /P option. # response-body-limit: Limit reassembly of response body for inspection # by file_data, http_server_body & pcre /Q option. # # For advanced options, see the user guide # server-config: List of server configurations to use if address matches # address: List of IP addresses or networks for this block # personality: List of personalities used by this block # # Then, all the fields from default-config can be overloaded # # Currently Available Personalities: # Minimal, Generic, IDS (default), IIS_4_0, IIS_5_0, IIS_5_1, IIS_6_0, # IIS_7_0, IIS_7_5, Apache_2 libhtp: default-config: personality: IDS # Can be specified in kb, mb, gb. Just a number indicates # it's in bytes. request-body-limit: 100kb response-body-limit: 100kb # inspection limits request-body-minimal-inspect-size: 32kb request-body-inspect-window: 4kb response-body-minimal-inspect-size: 40kb response-body-inspect-window: 16kb # response body decompression (0 disables) response-body-decompress-layer-limit: 2 # auto will use http-body-inline mode in IPS mode, yes or no set it statically http-body-inline: auto # Decompress SWF files. # 2 types: 'deflate', 'lzma', 'both' will decompress deflate and lzma # compress-depth: # Specifies the maximum amount of data to decompress, # set 0 for unlimited. # decompress-depth: # Specifies the maximum amount of decompressed data to obtain, # set 0 for unlimited. swf-decompression: enabled: yes type: both compress-depth: 0 decompress-depth: 0 # Take a random value for inspection sizes around the specified value. # This lower the risk of some evasion technics but could lead # detection change between runs. It is set to 'yes' by default. #randomize-inspection-sizes: yes # If randomize-inspection-sizes is active, the value of various # inspection size will be chosen in the [1 - range%, 1 + range%] # range # Default value of randomize-inspection-range is 10. #randomize-inspection-range: 10 # decoding double-decode-path: no double-decode-query: no # Can disable LZMA decompression #lzma-enabled: yes # Memory limit usage for LZMA decompression dictionary # Data is decompressed until dictionary reaches this size #lzma-memlimit: 1mb # Maximum decompressed size with a compression ratio # above 2048 (only LZMA can reach this ratio, deflate cannot) #compression-bomb-limit: 1mb server-config: #- apache: # address: [192.168.1.0/24, 127.0.0.0/8, "::1"] # personality: Apache_2 # # Can be specified in kb, mb, gb. Just a number indicates # # it's in bytes. # request-body-limit: 4096 # response-body-limit: 4096 # double-decode-path: no # double-decode-query: no #- iis7: # address: # - 192.168.0.0/24 # - 192.168.10.0/24 # personality: IIS_7_0 # # Can be specified in kb, mb, gb. Just a number indicates # # it's in bytes. # request-body-limit: 4096 # response-body-limit: 4096 # double-decode-path: no # double-decode-query: no # Note: Modbus probe parser is minimalist due to the poor significant field # Only Modbus message length (greater than Modbus header length) # And Protocol ID (equal to 0) are checked in probing parser # It is important to enable detection port and define Modbus port # to avoid false positive modbus: # How many unreplied Modbus requests are considered a flood. # If the limit is reached, app-layer-event:modbus.flooded; will match. #request-flood: 500 enabled: yes detection-ports: dp: 502 # According to MODBUS Messaging on TCP/IP Implementation Guide V1.0b, it # is recommended to keep the TCP connection opened with a remote device # and not to open and close it for each MODBUS/TCP transaction. In that # case, it is important to set the depth of the stream reassembling as # unlimited (stream.reassembly.depth: 0) # Stream reassembly size for modbus. By default track it completely. stream-depth: 0 # DNP3 dnp3: enabled: no detection-ports: dp: 20000 # SCADA EtherNet/IP and CIP protocol support enip: enabled: no detection-ports: dp: 44818 sp: 44818 ntp: enabled: yes dhcp: enabled: yes # SIP, disabled by default. sip: #enabled: no # Limit for the maximum number of asn1 frames to decode (default 256) asn1-max-frames: 256 ############################################################################## ## ## Advanced settings below ## ############################################################################## ## ## Run Options ## # Run suricata as user and group. #run-as: # user: suri # group: suri # Some logging module will use that name in event as identifier. The default # value is the hostname #sensor-name: suricata # Default location of the pid file. The pid file is only used in # daemon mode (start Suricata with -D). If not running in daemon mode # the --pidfile command line option must be used to create a pid file. #pid-file: /var/run/suricata.pid # Daemon working directory # Suricata will change directory to this one if provided # Default: "/" #daemon-directory: "/" # Umask. # Suricata will use this umask if it is provided. By default it will use the # umask passed on by the shell. #umask: 022 # Suricata core dump configuration. Limits the size of the core dump file to # approximately max-dump. The actual core dump size will be a multiple of the # page size. Core dumps that would be larger than max-dump are truncated. On # Linux, the actual core dump size may be a few pages larger than max-dump. # Setting max-dump to 0 disables core dumping. # Setting max-dump to 'unlimited' will give the full core dump file. # On 32-bit Linux, a max-dump value >= ULONG_MAX may cause the core dump size # to be 'unlimited'. coredump: max-dump: unlimited # If Suricata box is a router for the sniffed networks, set it to 'router'. If # it is a pure sniffing setup, set it to 'sniffer-only'. # If set to auto, the variable is internally switch to 'router' in IPS mode # and 'sniffer-only' in IDS mode. # This feature is currently only used by the reject* keywords. host-mode: auto # Number of packets preallocated per thread. The default is 1024. A higher number # will make sure each CPU will be more easily kept busy, but may negatively # impact caching. #max-pending-packets: 1024 # Runmode the engine should use. Please check --list-runmodes to get the available # runmodes for each packet acquisition method. Default depends on selected capture # method. 'workers' generally gives best performance. runmode: workers # Specifies the kind of flow load balancer used by the flow pinned autofp mode. # # Supported schedulers are: # # hash - Flow assigned to threads using the 5-7 tuple hash. # ippair - Flow assigned to threads using addresses only. # #autofp-scheduler: hash # Preallocated size for packet. Default is 1514 which is the classical # size for pcap on ethernet. You should adjust this value to the highest # packet size (MTU + hardware header) on your system. # Unix command socket can be used to pass commands to Suricata. # An external tool can then connect to get information from Suricata # or trigger some modifications of the engine. Set enabled to yes # to activate the feature. In auto mode, the feature will only be # activated in live capture mode. You can use the filename variable to set # the file name of the socket. unix-command: enabled: no #filename: custom.socket # Magic file. The extension .mgc is added to the value here. #magic-file: /usr/share/file/magic magic-file: /usr/share/misc/magic # GeoIP2 database file. Specify path and filename of GeoIP2 database # if using rules with "geoip" rule option. #geoip-database: /usr/local/share/GeoLite2/GeoLite2-Country.mmdb legacy: uricontent: enabled ## ## Detection settings ## # Set the order of alerts based on actions # The default order is pass, drop, reject, alert # action-order: # - pass # - drop # - reject # - alert # IP Reputation #reputation-categories-file: /usr/local/etc/suricata/iprep/categories.txt #default-reputation-path: /usr/local/etc/suricata/iprep #reputation-files: # - reputation.list # When run with the option --engine-analysis, the engine will read each of # the parameters below, and print reports for each of the enabled sections # and exit. The reports are printed to a file in the default log dir # given by the parameter "default-log-dir", with engine reporting # subsection below printing reports in its own report file. engine-analysis: # enables printing reports for fast-pattern for every rule. rules-fast-pattern: yes # enables printing reports for each rule rules: yes #recursion and match limits for PCRE where supported pcre: match-limit: 3500 match-limit-recursion: 1500 ## ## Advanced Traffic Tracking and Reconstruction Settings ## # Host specific policies for defragmentation and TCP stream # reassembly. The host OS lookup is done using a radix tree, just # like a routing table so the most specific entry matches. host-os-policy: # Make the default policy windows. windows: [0.0.0.0/0] bsd: [] bsd-right: [] old-linux: [] linux: [] old-solaris: [] solaris: [] hpux10: [] hpux11: [] irix: [] macos: [] vista: [] windows2k3: [] # Defrag settings: defrag: memcap: 32mb hash-size: 65536 trackers: 65535 # number of defragmented flows to follow max-frags: 65535 # number of fragments to keep (higher than trackers) prealloc: yes timeout: 60 # Enable defrag per host settings # host-config: # # - dmz: # timeout: 30 # address: [192.168.1.0/24, 127.0.0.0/8, 1.1.1.0/24, 2.2.2.0/24, "1.1.1.1", "2.2.2.2", "::1"] # # - lan: # timeout: 45 # address: # - 192.168.0.0/24 # - 192.168.10.0/24 # - 172.16.14.0/24 # Flow settings: # By default, the reserved memory (memcap) for flows is 32MB. This is the limit # for flow allocation inside the engine. You can change this value to allow # more memory usage for flows. # The hash-size determine the size of the hash used to identify flows inside # the engine, and by default the value is 65536. # At the startup, the engine can preallocate a number of flows, to get a better # performance. The number of flows preallocated is 10000 by default. # emergency-recovery is the percentage of flows that the engine need to # prune before unsetting the emergency state. The emergency state is activated # when the memcap limit is reached, allowing to create new flows, but # pruning them with the emergency timeouts (they are defined below). # If the memcap is reached, the engine will try to prune flows # with the default timeouts. If it doesn't find a flow to prune, it will set # the emergency bit and it will try again with more aggressive timeouts. # If that doesn't work, then it will try to kill the last time seen flows # not in use. # The memcap can be specified in kb, mb, gb. Just a number indicates it's # in bytes. flow: memcap: 128mb hash-size: 65536 prealloc: 10000 emergency-recovery: 30 #managers: 1 # default to one flow manager #recyclers: 1 # default to one flow recycler thread # This option controls the use of vlan ids in the flow (and defrag) # hashing. Normally this should be enabled, but in some (broken) # setups where both sides of a flow are not tagged with the same vlan # tag, we can ignore the vlan id's in the flow hashing. vlan: use-for-tracking: true # Specific timeouts for flows. Here you can specify the timeouts that the # active flows will wait to transit from the current state to another, on each # protocol. The value of "new" determine the seconds to wait after a handshake or # stream startup before the engine free the data of that flow it doesn't # change the state to established (usually if we don't receive more packets # of that flow). The value of "established" is the amount of # seconds that the engine will wait to free the flow if it spend that amount # without receiving new packets or closing the connection. "closed" is the # amount of time to wait after a flow is closed (usually zero). "bypassed" # timeout controls locally bypassed flows. For these flows we don't do any other # tracking. If no packets have been seen after this timeout, the flow is discarded. # # There's an emergency mode that will become active under attack circumstances, # making the engine to check flow status faster. This configuration variables # use the prefix "emergency-" and work similar as the normal ones. # Some timeouts doesn't apply to all the protocols, like "closed", for udp and # icmp. flow-timeouts: default: new: 30 established: 300 closed: 0 bypassed: 100 emergency-new: 10 emergency-established: 100 emergency-closed: 0 emergency-bypassed: 50 tcp: new: 60 established: 600 closed: 60 bypassed: 100 emergency-new: 5 emergency-established: 100 emergency-closed: 10 emergency-bypassed: 50 udp: new: 30 established: 300 bypassed: 100 emergency-new: 10 emergency-established: 100 emergency-bypassed: 50 icmp: new: 30 established: 300 bypassed: 100 emergency-new: 10 emergency-established: 100 emergency-bypassed: 50 # Stream engine settings. Here the TCP stream tracking and reassembly # engine is configured. # # stream: # memcap: 32mb # Can be specified in kb, mb, gb. Just a # # number indicates it's in bytes. # checksum-validation: yes # To validate the checksum of received # # packet. If csum validation is specified as # # "yes", then packet with invalid csum will not # # be processed by the engine stream/app layer. # # Warning: locally generated traffic can be # # generated without checksum due to hardware offload # # of checksum. You can control the handling of checksum # # on a per-interface basis via the 'checksum-checks' # # option # prealloc-sessions: 2k # 2k sessions prealloc'd per stream thread # midstream: false # don't allow midstream session pickups # async-oneside: false # don't enable async stream handling # inline: no # stream inline mode # drop-invalid: yes # in inline mode, drop packets that are invalid with regards to streaming engine # max-synack-queued: 5 # Max different SYN/ACKs to queue # bypass: no # Bypass packets when stream.reassembly.depth is reached. # # Warning: first side to reach this triggers # # the bypass. # # reassembly: # memcap: 64mb # Can be specified in kb, mb, gb. Just a number # # indicates it's in bytes. # depth: 1mb # Can be specified in kb, mb, gb. Just a number # # indicates it's in bytes. # toserver-chunk-size: 2560 # inspect raw stream in chunks of at least # # this size. Can be specified in kb, mb, # # gb. Just a number indicates it's in bytes. # toclient-chunk-size: 2560 # inspect raw stream in chunks of at least # # this size. Can be specified in kb, mb, # # gb. Just a number indicates it's in bytes. # randomize-chunk-size: yes # Take a random value for chunk size around the specified value. # # This lower the risk of some evasion technics but could lead # # detection change between runs. It is set to 'yes' by default. # randomize-chunk-range: 10 # If randomize-chunk-size is active, the value of chunk-size is # # a random value between (1 - randomize-chunk-range/100)*toserver-chunk-size # # and (1 + randomize-chunk-range/100)*toserver-chunk-size and the same # # calculation for toclient-chunk-size. # # Default value of randomize-chunk-range is 10. # # raw: yes # 'Raw' reassembly enabled or disabled. # # raw is for content inspection by detection # # engine. # # segment-prealloc: 2048 # number of segments preallocated per thread # # check-overlap-different-data: true|false # # check if a segment contains different data # # than what we've already seen for that # # position in the stream. # # This is enabled automatically if inline mode # # is used or when stream-event:reassembly_overlap_different_data; # # is used in a rule. # stream: memcap: 64mb checksum-validation: yes # reject wrong csums inline: true reassembly: memcap: 256mb depth: 1mb # reassemble 1mb into a stream toserver-chunk-size: 2560 toclient-chunk-size: 2560 randomize-chunk-size: yes #randomize-chunk-range: 10 #raw: yes #chunk-prealloc: 2048 #check-overlap-different-data: true # Host table: # # Host table is used by tagging and per host thresholding subsystems. # host: hash-size: 4096 prealloc: 1000 memcap: 32mb # IP Pair table: # # Used by xbits 'ippair' tracking. # #ippair: # hash-size: 4096 # prealloc: 1000 # memcap: 32mb # Decoder settings decoder: # Teredo decoder is known to not be completely accurate # as it will sometimes detect non-teredo as teredo. teredo: enabled: true # VXLAN decoder is assigned to up to 4 UDP ports. By default only the # IANA assigned port 4789 is enabled. vxlan: enabled: true ports: $VXLAN_PORTS # syntax: '8472, 4789' ## ## Performance tuning and profiling ## # The detection engine builds internal groups of signatures. The engine # allow us to specify the profile to use for them, to manage memory on an # efficient way keeping a good performance. For the profile keyword you # can use the words "low", "medium", "high" or "custom". If you use custom # make sure to define the values at "- custom-values" as your convenience. # Usually you would prefer medium/high/low. # # "sgh mpm-context", indicates how the staging should allot mpm contexts for # the signature groups. "single" indicates the use of a single context for # all the signature group heads. "full" indicates a mpm-context for each # group head. "auto" lets the engine decide the distribution of contexts # based on the information the engine gathers on the patterns from each # group head. # # The option inspection-recursion-limit is used to limit the recursive calls # in the content inspection code. For certain payload-sig combinations, we # might end up taking too much time in the content inspection code. # If the argument specified is 0, the engine uses an internally defined # default limit. On not specifying a value, we use no limits on the recursion. detect: profile: medium custom-values: toclient-groups: 3 toserver-groups: 25 sgh-mpm-context: auto inspection-recursion-limit: 3000 # If set to yes, the loading of signatures will be made after the capture # is started. This will limit the downtime in IPS mode. #delayed-detect: yes prefilter: # default prefiltering setting. "mpm" only creates MPM/fast_pattern # engines. "auto" also sets up prefilter engines for other keywords. # Use --list-keywords=all to see which keywords support prefiltering. default: mpm # the grouping values above control how many groups are created per # direction. Port whitelisting forces that port to get it's own group. # Very common ports will benefit, as well as ports with many expensive # rules. grouping: #tcp-whitelist: 53, 80, 139, 443, 445, 1433, 3306, 3389, 6666, 6667, 8080 #udp-whitelist: 53, 135, 5060 profiling: # Log the rules that made it past the prefilter stage, per packet # default is off. The threshold setting determines how many rules # must have made it past pre-filter for that rule to trigger the # logging. #inspect-logging-threshold: 200 grouping: dump-to-disk: false include-rules: false # very verbose include-mpm-stats: false # Select the multi pattern algorithm you want to run for scan/search the # in the engine. # # The supported algorithms are: # "ac" - Aho-Corasick, default implementation # "ac-bs" - Aho-Corasick, reduced memory implementation # "ac-ks" - Aho-Corasick, "Ken Steele" variant # "hs" - Hyperscan, available when built with Hyperscan support # # The default mpm-algo value of "auto" will use "hs" if Hyperscan is # available, "ac" otherwise. # # The mpm you choose also decides the distribution of mpm contexts for # signature groups, specified by the conf - "detect.sgh-mpm-context". # Selecting "ac" as the mpm would require "detect.sgh-mpm-context" # to be set to "single", because of ac's memory requirements, unless the # ruleset is small enough to fit in one's memory, in which case one can # use "full" with "ac". Rest of the mpms can be run in "full" mode. mpm-algo: hs # Select the matching algorithm you want to use for single-pattern searches. # # Supported algorithms are "bm" (Boyer-Moore) and "hs" (Hyperscan, only # available if Suricata has been built with Hyperscan support). # # The default of "auto" will use "hs" if available, otherwise "bm". spm-algo: auto # Suricata is multi-threaded. Here the threading can be influenced. threading: set-cpu-affinity: no # Tune cpu affinity of threads. Each family of threads can be bound # on specific CPUs. # # These 2 apply to the all runmodes: # management-cpu-set is used for flow timeout handling, counters # worker-cpu-set is used for 'worker' threads # # Additionally, for autofp these apply: # receive-cpu-set is used for capture threads # verdict-cpu-set is used for IPS verdict threads # cpu-affinity: - management-cpu-set: cpu: [ 0 ] # include only these CPUs in affinity settings - receive-cpu-set: cpu: [ 0 ] # include only these CPUs in affinity settings - worker-cpu-set: cpu: [ "all" ] mode: "exclusive" # Use explicitly 3 threads and don't compute number by using # detect-thread-ratio variable: # threads: 3 prio: low: [ 0 ] medium: [ "1-2" ] high: [ 3 ] default: "medium" #- verdict-cpu-set: # cpu: [ 0 ] # prio: # default: "high" # # By default Suricata creates one "detect" thread per available CPU/CPU core. # This setting allows controlling this behaviour. A ratio setting of 2 will # create 2 detect threads for each CPU/CPU core. So for a dual core CPU this # will result in 4 detect threads. If values below 1 are used, less threads # are created. So on a dual core CPU a setting of 0.5 results in 1 detect # thread being created. Regardless of the setting at a minimum 1 detect # thread will always be created. # detect-thread-ratio: 1.0 # Luajit has a strange memory requirement, it's 'states' need to be in the # first 2G of the process' memory. # # 'luajit.states' is used to control how many states are preallocated. # State use: per detect script: 1 per detect thread. Per output script: 1 per # script. luajit: states: 128 # Profiling settings. Only effective if Suricata has been built with the # the --enable-profiling configure flag. # profiling: # Run profiling for every xth packet. The default is 1, which means we # profile every packet. If set to 1000, one packet is profiled for every # 1000 received. #sample-rate: 1000 # rule profiling rules: # Profiling can be disabled here, but it will still have a # performance impact if compiled in. enabled: yes filename: rule_perf.log append: yes # Sort options: ticks, avgticks, checks, matches, maxticks # If commented out all the sort options will be used. #sort: avgticks # Limit the number of sids for which stats are shown at exit (per sort). limit: 10 # output to json #json: @e_enable_evelog@ # per keyword profiling keywords: enabled: yes filename: keyword_perf.log append: yes prefilter: enabled: yes filename: prefilter_perf.log append: yes # per rulegroup profiling rulegroups: enabled: yes filename: rule_group_perf.log append: yes # packet profiling packets: # Profiling can be disabled here, but it will still have a # performance impact if compiled in. enabled: yes filename: packet_stats.log append: yes # per packet csv output csv: # Output can be disabled here, but it will still have a # performance impact if compiled in. enabled: no filename: packet_stats.csv # profiling of locking. Only available when Suricata was built with # --enable-profiling-locks. locks: enabled: no filename: lock_stats.log append: yes pcap-log: enabled: no filename: pcaplog_stats.log append: yes ## ## Netfilter integration ## # When running in NFQ inline mode, it is possible to use a simulated # non-terminal NFQUEUE verdict. # This permit to do send all needed packet to Suricata via this a rule: # iptables -I FORWARD -m mark ! --mark $MARK/$MASK -j NFQUEUE # And below, you can have your standard filtering ruleset. To activate # this mode, you need to set mode to 'repeat' # If you want packet to be sent to another queue after an ACCEPT decision # set mode to 'route' and set next-queue value. # On linux >= 3.1, you can set batchcount to a value > 1 to improve performance # by processing several packets before sending a verdict (worker runmode only). # On linux >= 3.6, you can set the fail-open option to yes to have the kernel # accept the packet if Suricata is not able to keep pace. # bypass mark and mask can be used to implement NFQ bypass. If bypass mark is # set then the NFQ bypass is activated. Suricata will set the bypass mark/mask # on packet of a flow that need to be bypassed. The Netfilter ruleset has to # directly accept all packets of a flow once a packet has been marked. nfq: # mode: accept # repeat-mark: 1 # repeat-mask: 1 # bypass-mark: 1 # bypass-mask: 1 # route-queue: 2 # batchcount: 20 # fail-open: yes #nflog support nflog: # netlink multicast group # (the same as the iptables --nflog-group param) # Group 0 is used by the kernel, so you can't use it - group: 2 # netlink buffer size buffer-size: 18432 # put default value here - group: default # set number of packet to queue inside kernel qthreshold: 1 # set the delay before flushing packet in the queue inside kernel qtimeout: 100 # netlink max buffer size max-size: 20000 ## ## Advanced Capture Options ## # general settings affecting packet capture capture: # disable NIC offloading. It's restored when Suricata exits. # Enabled by default. #disable-offloading: false # # disable checksum validation. Same as setting '-k none' on the # commandline. #checksum-validation: none # Netmap support # # Netmap operates with NIC directly in driver, so you need FreeBSD 11+ which have # built-in netmap support or compile and install netmap module and appropriate # NIC driver on your Linux system. # To reach maximum throughput disable all receive-, segmentation-, # checksum- offloadings on NIC. # Disabling Tx checksum offloading is *required* for connecting OS endpoint # with NIC endpoint. # You can find more information at https://github.com/luigirizzo/netmap # netmap: - interface: default # Number of capture threads. "auto" uses number of RSS queues on interface. # Warning: unless the RSS hashing is symmetrical, this will lead to # accuracy issues. threads: auto # You can use the following variables to activate netmap tap or IPS mode. # If copy-mode is set to ips or tap, the traffic coming to the current # interface will be copied to the copy-iface interface. If 'tap' is set, the # copy is complete. If 'ips' is set, the packet matching a 'drop' action # will not be copied. # To specify the OS as the copy-iface (so the OS can route packets, or forward # to a service running on the same machine) add a plus sign at the end # (e.g. "copy-iface: eth0+"). Don't forget to set up a symmetrical eth0+ -> eth0 # for return packets. Hardware checksumming must be *off* on the interface if # using an OS endpoint (e.g. 'ifconfig eth0 -rxcsum -txcsum -rxcsum6 -txcsum6' for FreeBSD # or 'ethtool -K eth0 tx off rx off' for Linux). copy-mode: ips #copy-iface: eth3 # Set to yes to disable promiscuous mode disable-promisc: yes # promiscuous mode # Choose checksum verification mode for the interface. At the moment # of the capture, some packets may be with an invalid checksum due to # offloading to the network card of the checksum computation. # Possible values are: # - yes: checksum validation is forced # - no: checksum validation is disabled # - auto: Suricata uses a statistical approach to detect when # checksum off-loading is used. # Warning: 'checksum-validation' must be set to yes to have any validation checksum-checks: auto # BPF filter to apply to this interface. The pcap filter syntax apply here. #bpf-filter: port 80 or udp - interface: igb2_vlan42 copy-iface: igb2_vlan42^ - interface: igb2_vlan42^ copy-iface: igb2_vlan42 - interface: igb1 copy-iface: igb1^ - interface: igb1^ copy-iface: igb1 # For FreeBSD ipfw(8) divert(4) support. # Please make sure you have ipfw_load="YES" and ipdivert_load="YES" # in /etc/loader.conf or kldload'ing the appropriate kernel modules. # Additionally, you need to have an ipfw rule for the engine to see # the packets from ipfw. For Example: # # ipfw add 100 divert 8000 ip from any to any # # The 8000 above should be the same number you passed on the command # line, i.e. -d 8000 # ipfw: # Reinject packets at the specified ipfw rule number. This config # option is the ipfw rule number AT WHICH rule processing continues # in the ipfw processing system after the engine has finished # inspecting the packet for acceptance. If no rule number is specified, # accepted packets are reinjected at the divert rule which they entered # and IPFW rule processing continues. No check is done to verify # this will rule makes sense so care must be taken to avoid loops in ipfw. # ## The following example tells the engine to reinject packets # back into the ipfw firewall AT rule number 5500: # # ipfw-reinjection-rule-number: 5500 napatech: # The Host Buffer Allowance for all streams # (-1 = OFF, 1 - 100 = percentage of the host buffer that can be held back) # This may be enabled when sharing streams with another application. # Otherwise, it should be turned off. #hba: -1 # When use_all_streams is set to "yes" the initialization code will query # the Napatech service for all configured streams and listen on all of them. # When set to "no" the streams config array will be used. # # This option necessitates running the appropriate NTPL commands to create # the desired streams prior to running suricata. #use-all-streams: no # The streams to listen on when auto-config is disabled or when and threading # cpu-affinity is disabled. This can be either: # an individual stream (e.g. streams: [0]) # or # a range of streams (e.g. streams: ["0-3"]) # streams: ["0-3"] # When auto-config is enabled the streams will be created and assigned # automatically to the NUMA node where the thread resides. If cpu-affinity # is enabled in the threading section. Then the streams will be created # according to the number of worker threads specified in the worker cpu set. # Otherwise, the streams array is used to define the streams. # # This option cannot be used simultaneous with "use-all-streams". # auto-config: yes # Ports indicates which napatech ports are to be used in auto-config mode. # these are the port ID's of the ports that will be merged prior to the # traffic being distributed to the streams. # # This can be specified in any of the following ways: # # a list of individual ports (e.g. ports: [0,1,2,3]) # # a range of ports (e.g. ports: [0-3]) # # "all" to indicate that all ports are to be merged together # (e.g. ports: [all]) # # This has no effect if auto-config is disabled. # ports: [all] # When auto-config is enabled the hashmode specifies the algorithm for # determining to which stream a given packet is to be delivered. # This can be any valid Napatech NTPL hashmode command. # # The most common hashmode commands are: hash2tuple, hash2tuplesorted, # hash5tuple, hash5tuplesorted and roundrobin. # # See Napatech NTPL documentation other hashmodes and details on their use. # # This has no effect if auto-config is disabled. # hashmode: hash5tuplesorted ## ## Configure Suricata to load Suricata-Update managed rules. ## ## If this section is completely commented out move down to the "Advanced rule ## file configuration". ## default-rule-path: /usr/local/etc/suricata/opnsense.rules rule-files: - suricata.rules ## ## Auxiliary configuration files. ## classification-file: /usr/local/etc/suricata/classification.config reference-config-file: /usr/local/etc/suricata/reference.config # threshold-file: /usr/local/etc/suricata/threshold.config ## ## Include other configs ## # Includes. Files included here will be handled as if they were # inlined in this configuration file. # include installed rules list (generated by OPNsense install rules script) include: installed_rules.yaml # include custom file (may be persistently modified) include: custom.yaml
Updated by Jeff Lucovsky almost 2 years ago
- File source-netmap.c.v2.diff source-netmap.c.v2.diff added
Here's an updated patch which eliminates the rest of the changes between 6.0.8 and 6.0.9 when not using the V14 api
If you're able to try the patch, please let me know if this helps/hurts or there's no behavior change from the original issue.
Updated by Bill Meeks almost 2 years ago
Does the v14 Netmap API also produce the stall condition in OPNsense? Seems the discussion thus far has centered on using the Legacy (or prior to v14) API. Just curious whether the v14 API produces the same condition.
Updated by Franco Fichtner almost 2 years ago
Does the v14 Netmap API also produce the stall condition in OPNsense? Seems the discussion thus far has centered on using the Legacy (or prior to v14) API. Just curious whether the v14 API produces the same condition.
Yes, V14 changes always behaved worse and we even had one failed rollout of V14 into production. The issue surfaced here in the legacy mode is very likely the same bad behaviour.
Jeff: thanks I will prepare another build tomorrow morning.
Cheers,
Franco
Updated by Bill Meeks almost 2 years ago
Jeff and Franco:
Don't forget about the changes in runmode-netmap.c . In my version of the original patches, there are some changes in runmode-netmap.c and util-runmodes.c. The source-netmap.c file is not the only one changed as a result of the Netmap v14 implementation. At least the patch I am using on the pfSense side has changes in these other files still.
Updated by Clyne Snowtail almost 2 years ago
Franco Fichtner wrote in #note-30:
The typo would likely only affect V14 operability if at all.
And, yes, Clyne said the full revert of the commits on top of 6.0.9 is the only one that is working normally (_1 variant).
The _2 with Jeff's proposed patch doesn't work either. I've also compared netmap code between netmap_user.c and libnetmap and the code seems to be equivalent indeed (minus idiosyncrasies that netmap authors built into their code for things that are not really being used in the header struct).
I've built a _3 variant here fixing the typo and backing out the configure.ac changes to see if the problem is actually with the code or not.
https://pkg.opnsense.org/FreeBSD:13:amd64/snapshots/misc/suricata-6.0.9_3.pkg
I've also compared debug log output between unmodified versions 6.0.8 and 6.0.9 but couldn't see a change in threads or attach behaviour for the ports being given in the configuration.
Cheers,
Franco
This _3 variant did not fix the problem.
I will standby for _4 and give that a shot
Updated by Franco Fichtner almost 2 years ago
Here is the _4 https://pkg.opnsense.org/FreeBSD:13:amd64/snapshots/misc/suricata-6.0.9_4.pkg based on Jeff's source-netmap.c.v2.diff
Updated by Clyne Snowtail almost 2 years ago
Franco Fichtner wrote in #note-38:
Here is the _4 https://pkg.opnsense.org/FreeBSD:13:amd64/snapshots/misc/suricata-6.0.9_4.pkg based on Jeff's source-netmap.c.v2.diff
Fail. No change from vanilla 6.0.9
Updated by Jeff Lucovsky almost 2 years ago
- File netmap-patch.v3.diff netmap-patch.v3.diff added
I found an issue where a per-device lock was not released following an error from nm_inject
The error would manifest as a "stall" -- but only when the runmode was autofp
I've attached the patch.
Updated by Franco Fichtner almost 2 years ago
Updated by Clyne Snowtail almost 2 years ago
Franco Fichtner wrote in #note-41:
v3/_5 : https://pkg.opnsense.org/FreeBSD:13:amd64/snapshots/misc/suricata-6.0.9_5.pkg
Nope. Since we are running workers I guess thats expected though.
Updated by Jeff Lucovsky almost 2 years ago
The latest patch eliminates nearly all differences between 6.0.8 and 6.0.9 when not using the v14 api.
I think the best way forward is to get a stack trace from the worker thread(s) that are stalled.
Updated by Jeff Lucovsky almost 2 years ago
Franco Fichtner wrote in #note-44:
Did you forget the attachment?
No; I was referring to the "v3" attachment posted yesterday.
Updated by Clyne Snowtail almost 2 years ago
Tell me what to do and Ill get you the info you need.
Updated by Jeff Lucovsky almost 2 years ago
Clyne Snowtail wrote in #note-46:
Tell me what to do and Ill get you the info you need.
When Suricata is stalled, attach to Suricata using Suricata's PID
$ sudo gdb -p $(pidof suricata) -ex 'set pagination off' -ex 'thread apply all bt' -ex detach -ex quit > output-file 2>&1This command:
- Invokes gdb and attaches to the Suricata process
- Disables pagination so output is delivered without pausing
- Gets a stack trace of all Suricata threads
- Detaches from the Suricata process allowing it to continue
- Exits the gdb session
Save the output and post here.
Updated by Clyne Snowtail almost 2 years ago
Franco Fichtner wrote in #note-44:
Did you forget the attachment?
Franco can you get me packages with debugging symbols or tell me how to rebuild the system with symbols enabled? It looks like I need all the suricata deps and subdeps so probably most/all of the base system with symbols.
Updated by Jeff Lucovsky almost 2 years ago
@Clyne Snowtail @Franco Fichtner Have either or you been able to get a stack trace with better information?
Updated by Clyne Snowtail almost 2 years ago
Jeff Lucovsky wrote in #note-49:
@Clyne Snowtail @Franco Fichtner Have either or you been able to get a stack trace with better information?
No.
I attempted to build my own release of OPNSense with debugging symbols enabled during compilation, however it would fail before completion. I was trying to send a message to Franco for help but havent heard back.
Updated by Jeff Lucovsky almost 2 years ago
- Status changed from Assigned to In Review
Updated by Bill Meeks almost 2 years ago
- File netmap-patch.v4.diff netmap-patch.v4.diff added
I have attached a slightly modified version of Jeff's earlier patches. This version, when the new NETMAP_V14 option knob is NOT enabled, should produce a netmap API in Suricata that mimics precisely the older netmap API in Suricata 6.0.8 and earlier. This is the API version with only a single host stack ring pair.
When the new NETMAP_V14 option knob is enabled, this patch produces a netmap API version that creates multiple host stack rings to match the number of rings exposed by the physical NIC in the netmap interface pair (source interface and destination or copy-to interface).
Updated by Jeff Lucovsky almost 2 years ago
- Status changed from In Review to Closed
Reworked ifdef
code merged in https://github.com/OISF/suricata/pull/8458
Updated by Franco Fichtner almost 2 years ago
Geez guys why always with the rush? We're in the middle of a major release and I really cannot verify any of this if we start rushing towards 6.0.10 again.
Updated by Franco Fichtner almost 2 years ago
Here is a package for 6.0.10 to test: https://pkg.opnsense.org/FreeBSD:13:amd64/snapshots/misc/suricata-6.0.10.pkg
If it doesn't work as expected I'll give Bill's suggestion a try.
Updated by Clyne Snowtail almost 2 years ago
Franco Fichtner wrote in #note-57:
Here is a package for 6.0.10 to test: https://pkg.opnsense.org/FreeBSD:13:amd64/snapshots/misc/suricata-6.0.10.pkg
If it doesn't work as expected I'll give Bill's suggestion a try.
Nope still doesnt work. It passed my initial speed test then locked up a few minutes later.
Updated by Bill Meeks almost 2 years ago
Clyne Snowtail wrote in #note-58:
Nope still doesnt work. It passed my initial speed test then locked up a few minutes later.
Would you share your Suricata setup? What NIC driver is in use and do you have any VLANs on the Suricata interfaces?
Updated by Clyne Snowtail almost 2 years ago
- File suricata.yaml suricata.yaml added
Heres the config file.
Its sniffing on my WAN; no VLAN present there.
NIC driver is xn, for a Xen VNI
Updated by Bill Meeks almost 2 years ago
Clyne Snowtail wrote in #note-60:
Heres the config file.
Its sniffing on my WAN; no VLAN present there.
NIC driver is xn, for a Xen VNI
Ah...so you are using a virtual NIC and thus the emulated netmap adapter. The xn
adapter is not supported for native netmap operation. The stall problem is something Franco and others are working on in the emulated netmap adapter code on FreeBSD. Some bug there could be the root cause. But I don't know why this patch tickles the defect, though.
Updated by Clyne Snowtail almost 2 years ago
Yes...I have another physical NIC in the machine and my intention had been to do PCI Passthrough on it and use it for the WAN interface, but my system seems to have some issue with the IOMMU that it isnt responding correctly when I attempt.
Ill just stick with the modified 6.0.9 and prevent it from being upgraded until they can hopefully solve the root problem, or I can solve my PCI Passthrough.
Updated by Franco Fichtner almost 2 years ago
This version is Bill's latest proposal: https://pkg.opnsense.org/FreeBSD:13:amd64/snapshots/misc/suricata-6.0.9_6.pkg
Updated by Bill Meeks almost 2 years ago
Clyne Snowtail wrote in #note-62:
Yes...I have another physical NIC in the machine and my intention had been to do PCI Passthrough on it and use it for the WAN interface, but my system seems to have some issue with the IOMMU that it isnt responding correctly when I attempt.
Ill just stick with the modified 6.0.9 and prevent it from being upgraded until they can hopefully solve the root problem, or I can solve my PCI Passthrough.
Please don't give up on us yet <grin>. You seem to have the most reliable system for reproducing the problem. Please test the latest Suricata package version just posted by Franco. We are trying to determine why this new netmap patch is giving you problems when it should, based on the configure options provided during compilation, produce a binary whose netmap parts are identical to that of 6.0.8 and thus should work the same.
Currently OPNsense is not using the new multiple host rings API (what we call the Netmap v14 API). So we are truly puzzled why the patch is causing problems because it has #ifdef compiler conditionals in it that skip including the new v14 API code unless it is called for in the Makefile configure options.
Updated by Clyne Snowtail almost 2 years ago
Im game for whatever you are!
That one doesnt work either.
Updated by Bill Meeks almost 2 years ago
Clyne Snowtail wrote in #note-65:
Im game for whatever you are!
That one doesnt work either.
Shoot! Now I am totally flummoxed. The only possibility is something perhaps in the runmode-netmap.c
file. I will do a line-by-line compare and see if I can find something that is not covered by the #ifdef conditionals. That's the only thing I can imagine - something I'm missing that is not covered by an #ifdef conditional.
Updated by Bill Meeks almost 2 years ago
Just to add some new data to the record of this bug --
I have a pfSense user that is experiencing pretty much the same stalling issue when using runmode "workers". On pfSense, we are using the NETMAP_v14 API and have been since August of 2021. I had one isolated report of issues after the NETMAP_v14 rollout, but that was it. But, on pfSense, the default Suricata configuration for all modes uses runmode "autofp". Users have an option to change that, but I believe very few, if any, make the change to "workers".
The user I just mentioned has no stability issues with runmode "autofp", but gets reduced speeds (high 600s up to about 700 megabits/sec on a Gigabit connection) using that runmode on a new SG-6100 firewall appliance he had just purchased. I suggested he switch to runmode "workers", and he immediately was able to achieve link saturation speed (930+ megabits/sec), but then after a short period of repeated speed tests the box would hang and need a reboot to recover. This was repeatable for him. The firewall is a Netgate SG-6100 appliance. Previously the user had Protectli hardware that he says ran Suricata in IPS mode at full link speeds without issue. The Protectli used the em
NIC driver while the Netgate SG-6100 uses both the ix
and igc
NIC drivers on its available ports.
The throughput increase using runmode "workers" is expected, but the instability (namely the stall/hang condition requiring a reboot to clear) is certainly not expected. But perhaps this is a clue. What differences in "workers" runmode might could lead to the lockup? We know "workers" uses a single thread per netmap ring for the entire processing chain on a packet, and also uses no device locks while in the netmap code since the same thread reads the packet, analyzes it, and then either drops it or writes it back to netmap. The "autofp" mode uses a single thread per netmap ring for reading packets, but uses a shared "pool" of threads for analyzing packets and dispatching them via netmap. Because of the shared pool of "netmap output threads", netmap ring/device locking is used to control concurrent access to the same ring.
Updated by Bill Meeks almost 2 years ago
- File netmap-patch-v5.diff netmap-patch-v5.diff added
I think I may have found the one remaining "difference" between the latest netmap V14 patch and the netmap code that existed in Suricata 6.0.8 prior to the introduction of the NETMAP_V14 option.
I've attached a patch file that can be applied to the current Suricata 6.0.10 branch. Because I don't have the hardware nor environment to reproduce the stall bug, if the OPNsense team would like to create a test Suricata package using 6.0.10 base with the supplied netmap-patch-v5.diff
file applied, I would be appreciative.
Here is what I changed in this latest patch. I'll start with the pertinent section of source-netmap.c
code from the older 6.0.8 branch --
if (nm_inject(ntv->ifdst->nmd, GET_PKT_DATA(p), GET_PKT_LEN(p)) == 0) {
SCLogDebug("failed to send %s -> %s",
ntv->ifsrc->ifname, ntv->ifdst->ifname);
ntv->drops++;
}
SCLogDebug("sent succesfully: %s(%d)->%s(%d) (%u)",
ntv->ifsrc->ifname, ntv->ifsrc->ring,
ntv->ifdst->ifname, ntv->ifdst->ring, GET_PKT_LEN(p));
ioctl(ntv->ifdst->nmd->fd, NIOCTXSYNC, 0);
return TM_ECODE_OK;
Notice that in the 6.0.8 netmap code above, even if the call to nm_inject()
fails (by returning 0) the code still falls through and eventually calls ioctl(ntv->ifdst->nmd->fd, NIOCTXSYNC, 0)
to have the kernel sync the TX netmap ring. When the kernel code syncs the netmap TX ring, it sends the packets placed on the ring on to their intended destination and then updates the various ring pointers (cur, head, and tail) to indicate open and available transmission slots for the next nm_inject()
call to utilize. But note that ioctl(NIOCTXSYNC)
is a non-blocking call. It will always return immediately whether the kernel was actually able to make new space available or not. As a side note, the code above will also log a false debug message saying the packet was sent successfully when it actually was not.
Now, here is the current code in the 6.0.10 branch (this is also the same code as was in the original 6.0.9 Netmap V14 release) --
/* attempt to write the packet into the netmap ring buffer(s) */
#if USE_NEW_NETMAP_API
if (nmport_inject(ntv->ifdst->nmd, GET_PKT_DATA(p), GET_PKT_LEN(p)) == 0) {
#else
if (nm_inject(ntv->ifdst->nmd, GET_PKT_DATA(p), GET_PKT_LEN(p)) == 0) {
#endif
if (ntv->flags & NETMAP_FLAG_EXCL_RING_ACCESS) {
SCMutexUnlock(&ntv->ifdst->netmap_dev_lock);
}
SCLogDebug("failed to send %s -> %s", ntv->ifsrc->ifname, ntv->ifdst->ifname);
ntv->drops++;
return TM_ECODE_FAILED;
}
SCLogDebug("sent successfully: %s(%d)->%s(%d) (%u)", ntv->ifsrc->ifname, ntv->ifsrc->ring,
ntv->ifdst->ifname, ntv->ifdst->ring, GET_PKT_LEN(p));
ioctl(ntv->ifdst->nmd->fd, NIOCTXSYNC, 0);
if (ntv->flags & NETMAP_FLAG_EXCL_RING_ACCESS) {
SCMutexUnlock(&ntv->ifdst->netmap_dev_lock);
}
return TM_ECODE_OK;
Notice that in the NETMAP_V14 API code above, if the nm_inject()
or nmport_inject()
function fails to write the packet, the code counts a dropped packet and bails without calling ioctl(ntv->ifdst->nmd->fd, NIOCTXSYNC, 0)
. This means the kernel does not sync the TX ring, and I think this leads to the lockup/stall condition. I can see this being particularly problematic under high traffic rates as that is when the nm_inject()
or nmport_inject()
functions are most likely to return without writing the packet because the thread may need to wait for the kernel to process a potentially full TX ring and open up space for new packets to be added to the ring. I believe this bug might result in a "stuck thread" because the netmap ring is never TX sync'd and thus never emptied. Eventually, all active threads could wind up in this state and packet transmission stalls on the interface.
After thinking about this some more, I think this entire section of code needs to be refactored a little. It should attempt the packet write by calling nm_inject()
or nmport_inject()
, but if the call fails it should enter either a POLLOUT poll() wait with a timeout, or else start a short timed loop periodically retrying the write with ioctl(NIOCTXSYNC) calls in between checks until it completes the write or eventually times out and logs a dropped packet and associated error.
This logic error is mine. I inadvertently introduced it when I originally created the NETMAP_V14 multiple host stack rings patch that Jeff based his work on.
Updated by Franco Fichtner almost 2 years ago
Hi Bill,
That looks like a probable cause indeed. Generic netmap mode is more prone to stalls when TX sync is not called as TX sync is also used to free buffers. If we are full and don't call TX sync the likelihood of stalling forever increases.
Test version for Bill's v5 patch: https://pkg.opnsense.org/FreeBSD:13:amd64/snapshots/misc/suricata-6.0.10_1.pkg
It would probably make sense to reopen this ticket or at least go through the motions to consider patching this in the upcoming version 7 as well as 6.0.11.
Cheers,
Franco
Updated by Bill Meeks almost 2 years ago
Hi Franco:
I am hoping this fix addresses and corrects the underlying issue with the stalls.
As I mentioned, my feeling is there are more improvements that can be made to the logic in this section of the Suricata netmap code. This is the NetmapWritePacket()
function in source-netmap.c
. More robustness and better error handling are needed. I welcome any input you or Jeff have in this area. I suspect you have some valuable insights from your recent and continuing work on the netmap generic adapter kernel code. Below is a quick description of how the netmap processing works in Suricata.
The netmap code in Suricata operates using three basic functions: all contained in source-netmap.c
. The first is ReceiveNetmapLoop()
. This is a simple looping function that uses poll()
with the POLLIN event to wait for the kernel to signal that data packets are available in the netmap device ring pointed to by the file descriptor supplied to poll()
. Once the poll()
call returns signaling data is available, either function nm_dispatch()
or function NetmapReadPackets()
(the latter when the new NETMAP_V14 option is enabled) is called to read the data and write it into a Suricata Packet
structure for processing by the rules engine. The third major function in the netmap processing loop is NetmapWritePacket()
. This last function gets called as part of the ReleasePacket() callback Suricata uses once a Packet has been fully processed by the rules engine and a verdict has been rendered. In the netmap code, the ReleasePacket() callback is set for NetmapReleasePacket()
which then winds up calling NetmapWritePacket()
when using IPS mode. This last function call is where data is either dropped and not written out, or pulled back out of the Suricata Packet structure and placed onto the netmap destination port ring to send it to the original intended destination.
Updated by Jeff Lucovsky almost 2 years ago
- Related to Bug #5862: netmap: packet stalls added
Updated by Jeff Lucovsky almost 2 years ago
Updated by Bill Meeks almost 2 years ago
I have some very encouraging news to report about my latest patch. It appears to correct the stall issue. I encountered a former OPNsense user on the pfSense forum, and we were discussing the stall issue. He agreed to test Franco's latest Suricata test package for OPNsense. He was routinely experiencing the stall condition previously with the new Suricata netmap code (6.0.9) shortly after it was released last December. He just resurrected his OPNsense VM and tested again with the latest Suricata netmap patch applied (my v5 patch) and had zero stalls and was able to saturate his Gigabit link.
So, I think the v5 patch is the solution. I will cross-post these results in the newly opened tickets.
Updated by Esko Harhio almost 2 years ago
Bill Meeks wrote in #note-73:
......
So, I think the v5 patch is the solution. I will cross-post these results in the newly opened tickets.
i can too confirm the v5 patch did fix the stalling issue in my OPNsense test vm.
Updated by Franco Fichtner almost 2 years ago
Updated by Victor Julien almost 2 years ago
- Related to Bug #5863: netmap: packet stalls (6.0.x backport) added
Updated by Clyne Snowtail almost 2 years ago
I was out of town all week and couldnt test as I wouldnt be able to reverse it.
I have tested it now and it is fixing it for me as well.
Updated by Bill Meeks almost 2 years ago
Clyne Snowtail wrote in #note-79:
I was out of town all week and couldnt test as I wouldnt be able to reverse it.
I have tested it now and it is fixing it for me as well.
Thank you for testing and posting the results. This fix should get into the next 6.0.11 release of Suricata. I have already incorporated the patch into the Suricata package we use on pfSense.
Updated by Franco Fichtner 10 months ago
I know this is closed but the issue appears to be present in 7.0.2 now that we moved to Suricata 7 in OPNsense 24.1. I'll make a new ticket with the proper info attached tomorrow. ;)
Updated by Jeff Lucovsky 10 months ago
Franco Fichtner wrote in #note-81:
I know this is closed but the issue appears to be present in 7.0.2 now that we moved to Suricata 7 in OPNsense 24.1. I'll make a new ticket with the proper info attached tomorrow. ;)
I've done a first-order evaluation of the Netmap code and the changes fixing the stalling issue. Both 6.0.x and 7.0.x (and master) have the fixes for the stall. Any additional contextual information that you have with the 7.0.2 systems experiencing that issue will help.
Updated by Franco Fichtner 10 months ago
Hi Jeff,
Thanks for taking a look. You are right. Netmap appears fine and packets show as passed, but we get a lot of stalled connections (but not ping). It might have something to do with midstream handling, but midstream-policy is not making it much better and the stalls appear even on newly established TCP connections. Maybe the flow detection isn't working for NIC-host combinations, but I won't be able to provide the necessary information this week. For now we had to roll back to version 6.
Cheers,
Franco
Updated by Franco Fichtner 10 months ago
At first glance that's not it, but I'll double-check tomorrow morning.
My colleague says the beta1 worked and then somewhere it stopped working but still very early in the bisect.
Thanks so far to you both!
Updated by Bill Meeks 10 months ago
If it helps as an additional data point --
We have been running Suricata 7.0.2 in pfSense for quite a while with no similar stalling issue reported. We do restrict the user configurations to physical interfaces only, though. No support for Bridge Interfaces, and for VLANs, we map to the parent physical interface. Also have a check for the physical NIC driver and restrict Suricata netmap configurations to only those known to have native netmap support.
Currently the underlying FreeBSD version is 14.0. So far as I recall, the Suricata netmap code is identical for both 7.x and 6.0.x now after the fix this original ticket describes.
Updated by Franco Fichtner 10 months ago
It needs both config tweaks to get it into working order, which is what made testing for a singular cause difficult:
livedev.use-for-tracking: false
stream.midstream-policy: ignore
Should I add a ticket to do some sort of validation or config error on these settings in combination with netmap? The defaults are ok I guess, but the side effects are more or less fatal.
Thanks,
Franco
Updated by Bill Meeks 10 months ago
Franco Fichtner wrote in #note-88:
It needs both config tweaks to get it into working order, which is what made testing for a singular cause difficult:
livedev.use-for-tracking: false
stream.midstream-policy: ignore
I defaulted the new stream.midstream-policy
to ignore
when I rolled out Suricata 7.0.2 in the pfSense package. That's probably why I had not seen a package user complaining of netmap stalls on the pfSense side. I currently have no override of the livedev.use-for-tracking
parameter, though. So, I guess it is defaulting to true
.
Updated by Jeff Lucovsky 10 months ago
franco @bmeeks See redmine issue 6755. I'll backport it to 7.0.x once it's been merged into @master
.