Help Understanding a Crash [kernel panic]

[bmeeks]

@none-0 said in Help Understanding a Crash [kernel panic]:

Thanks @stephenw10 and @bmeeks, I'm indeed using Suricata with Inline mode.
Other than IP's conflicting, should I keep an eye for other problems that might "overwhelm" netmap in this case?

It's not so much netmap being "overwhelmed" as it is an issue with the "non-thread safe nature" of the host stack ring pair exposed by netmap with FreeBSD. Currently the Suricata binary runs in autofp runmode. That means it uses multiple capture threads, but there is only a single pair of host stack netmap rings that have to be shared among those capture threads (the threads are capturing data going to and from the NIC and kernel host stack). The Suricata binary is not doing any kind of locking with those netmap threads accessing the single host stack ring pair. That is likely the root cause of the netmap errors. The threads are probably modifying some memory pointers concurrently, and that is confusing netmap. It detects the problem, and attempts to fix it by calling the netmap_ring_reinit() function. I'm working with an upstream developer to hopefully fix that in Suricata 6.0.4, due out later this year. That fix will include opening multiple host rings and including some mutex locking to keep things synchronized.

[None 0]

Really appreciate that you explained it, @bmeeks!
I'll take that in consideration if another crash happens. Gonna switch to Legacy for a while, tho.

@bmeeks said in Help Understanding a Crash [kernel panic]:

I'm working with an upstream developer to hopefully fix that in Suricata 6.0.4, due out later this year. That fix will include opening multiple host rings and including some mutex locking to keep things synchronized.

Concurrency control must be difficult to implement. Keep up the good work!

[None 0]

So, I turned Inline mode ON again, and after a day or so:

Spoiler

<118>pfSense 2.5.2-RELEASE amd64 Fri Jul 02 15:33:00 EDT 2021
<118>Bootup complete
<6>pid 88136 (suricata), jid 0, uid 0: exited on signal 11 (core dumped)
<6>arp: 192.168.0.84 moved from 00:15:5d:XX:XX:XX to 00:15:5d:XX:XX:XX on igb1
<6>igb1: link state changed to DOWN
<6>igb1: link state changed to UP
787.976946 [1684] nm_txsync_prologue igb1 TX0: fail 'head > kring->rtail && head < kring->rhead' h 975 c 975 t 512 rh 976 rc 976 rt 512 hc 976 ht 512
787.976965 [1787] netmap_ring_reinit called for igb1 TX0
730.812187 [1684] nm_txsync_prologue igb1 TX0: fail 'head > kring->rtail && head < kring->rhead' h 642 c 642 t 512 rh 643 rc 643 rt 512 hc 643 ht 512
730.812205 [1787] netmap_ring_reinit called for igb1 TX0
<6>igb1: link state changed to DOWN
<6>igb1: link state changed to UP
<6>igb1: link state changed to DOWN
<6>igb1: link state changed to UP
<6>igb1: link state changed to DOWN
<6>igb1: link state changed to UP
<6>igb1: link state changed to DOWN
<6>igb1: link state changed to UP
443.587062 [1684] nm_txsync_prologue igb1 TX0: fail 'head > kring->rtail && head < kring->rhead' h 804 c 804 t 512 rh 805 rc 805 rt 512 hc 805 ht 512
443.587089 [1787] netmap_ring_reinit called for igb1 TX0
<6>igb1: link state changed to DOWN
<6>igb1: link state changed to UP
<6>igb1: link state changed to DOWN
<6>igb1: link state changed to UP
<6>arp: 192.168.0.43 moved from 00:1f:c3:XX:XX:XX to 24:f5:ab:XX:XX:XX on igb1
<6>arp: 192.168.0.43 moved from 24:f5:ab:XX:XX:XX to 00:1f:c3:XX:XX:XX on igb1
<6>igb1: link state changed to DOWN
<6>igb1: link state changed to UP
<6>igb1: link state changed to DOWN
<6>igb1: link state changed to UP

Fatal trap 9: general protection fault while in kernel mode

Since inline + Suricata is causing crashes, you guys suggest to use Snort instead? Would single thread affect performance on a giga network with 100+ devices (generally speaking)?

[bmeeks]

Performance would probably be about the same with Snort. That's because even though Suricata itself is multithreaded, the actual interface between the host OS stack and the NIC mediated by netmap is currently constrained by the fact only a single pair of host stack TX/RX rings is available. That in turn constrains Suricata to single-threaded operation for packet acquisition (same as Snort).

But this is something I am currently working on improving along with the Suricata development team. Perhaps this improvement is ready by the release of 6.0.4 Suricata. We will see.

In the meantime, why not just switch over to Legacy Blocking Mode for a while? Only if that really is a problem for you, would I switch over to Snort. Switching would mean reconfiguring a lot of stuff.

[None 0]

@bmeeks Right. Thanks.

In legacy, do I need to enable and config SID Mgmt to ET rules to work, or just set IPS Policy + Ruleset?

[bmeeks]

@none-0 said in Help Understanding a Crash [kernel panic]:

@bmeeks Right. Thanks.

In legacy, do I need to enable and config SID Mgmt to ET rules to work, or just set IPS Policy + Ruleset?

IPS Policies are only available for Snort Subscriber Rules. That's because that feature depends on a special metadata tag that the Snort rules authors include with their rules package. The ET rules do not have that metadata tag. Each Snort rule is tagged by the authors with one or more IPS Policy tags, and the IPS Policy feature in the Snort and Suricata packages keys off that tag to select rules. Since the ET rules don't have the tag, the feature can't work for them.

So when using ET rules, you will need to manually enable the ET categories and/or the individual rules you want. You can do that either via SID MGMT (the best way, in my opinion), or on the CATEGORIES and RULES tabs by checking boxes and clicking icons.

You can use both techniques at the same time. So use IPS Policy to let it auto-select the Snort Subscriber rules, and then use either SID MGMT or the manual process to choose your ET rules.

[None 0]

Hello guys,

I had another crash, and apparently for the same reason: IPs bouncing between two MACs, and then calling the netmap_ring_reinit. Strangely I was using the Legacy mode this time (so it may be another type of problem), and it's the first one since 13 days ago.

Spoiler

ddb.txt
msgbuf.txt

[bmeeks]

@none-0 said in Help Understanding a Crash [kernel panic]:

Hello guys,

I had another crash, and apparently for the same reason: IPs bouncing between two MACs, and then calling the netmap_ring_reinit. Strangely I was using the Legacy mode this time (so it may be another type of problem), and it's the first one since 13 days ago.

Spoiler

ddb.txt
msgbuf.txt

Using Legacy Mode would take netmap 100% out of the equation, and thus the netmap_ring_init() function could not be called. That is a netmap-specific function, and it will only be called by netmap itself. So you had something running using netmap operation, and Snort and Suricata are the only two packages I am aware of on pfSense that use netmap. Perhaps you had a duplicate zombie process running on a Suricata interface (two instances of Suricata running on the same interface). One old instance might have been running netmap mode and a newer one Legacy Mode, but both on the same interface.

You are running 2.5.2 RELEASE, so you don't have access to the latest Suricata updates with the new 6.0.3 binary. It will be interesting to see how that update performs for you when it becomes available.

[None 0]

oh man
I knew if this was the case, the other process shouldn't still be running after the restart, but anyway:
/usr/local/bin/suricata -i igb1 -D -c /usr/local/etc/suricata/suricata_25400_igb1/suricata.yaml --pidfile /var/run/suricata_igb125400.pid
Only this one now.

You are running 2.5.2 RELEASE, so you don't have access to the latest Suricata updates with the new 6.0.3 binary. It will be interesting to see how that update performs for you when it becomes available.

I'm looking forward to the release! Unfortunatly I can't move to devel and try it right now, but I will update this topic (if possible) when I do.

Thanks again, @bmeeks!

[None 0]

Had another crash today :/
Same reason. Should I uninstall n install Suricata again?

[bmeeks]

@none-0 said in Help Understanding a Crash [kernel panic]:

Had another crash today :/
Same reason. Should I uninstall n install Suricata again?

No, uninstalling and then reinstalling the exact same binary will not likely have any effect on that error.

It is likely a problem within the compiled binary code, and the exact same code will get installed again if you remove and reinstall the package. The only way to have a meaningful test would be to move to the 2.6.0 snapshot branch and install the newer Suricata from there. That package has a completely different binary in it.

In my experience with that error, I have not seen it cause a kernel panic, but that does not mean it couldn't. Maybe it is tickling something in your system just right such that it triggers the crash. My suspicion is multithreaded access to the single pair of netmap RX/TX rings exposed by the host stack in the older netmap API used by Suricata in pfSense CE and pfSense+ RELEASE versions today. There is new code to address that in the 6.0.3 Suricata package currently available in the 2.6.0 Snapshots branch.

[None 0]

@bmeeks Just thought it could be a problem with my installation, but it doesn't seem to make sense...
Alright, disabled blocking, and gonna try the DEV branch when I have time.

Thanks!

[bmeeks]

@none-0 said in Help Understanding a Crash [kernel panic]:

@bmeeks Just thought it could be a problem with my installation, but it doesn't seem to make sense...
Alright, disabled blocking, and gonna try the DEV branch when I have time.

Thanks!

If you are using Legacy Blocking Mode, then netmap is 100% completely and totally out of the picture. You should never see a netmap_ring_reinit() error in Legacy Mode. The only way to see that error is if something is still running with the netmap kernel device. If you are still getting kernel crashes and not using Inline IPS Mode on any interface, then Suricata is not the root cause of the problem.

Everything I mentioned above about the new Suricata in the Snapshots branch only applies when using Inline IPS Mode.

[None 0]

@bmeeks
My mistake, there's no netmap_ring_reinit this time:

<6>arp: 192.168.0.30 moved from 00:1d:60:7d:8c:61 to 00:05:4b:04:5e:7c on igb1
<6>arp: 192.168.0.39 moved from 00:22:15:6c:eb:96 to 00:e0:53:0b:40:f8 on igb1
<6>pid 18785 (grep), jid 0, uid 0: exited on signal 10 (core dumped)
<6>igb1: link state changed to DOWN
<6>igb1: link state changed to UP
<6>igb1: link state changed to DOWN
<6>igb1: link state changed to UP
<6>igb1: link state changed to DOWN
<6>igb1: link state changed to UP
<6>igb1: link state changed to DOWN
<6>igb1: link state changed to UP
<6>arp: 192.168.0.30 moved from f4:6d:04:e4:84:54 to 00:05:4b:04:5e:7c on igb1
<6>igb1: promiscuous mode disabled
<6>igb1: promiscuous mode enabled
<6>igb1: link state changed to DOWN
<6>igb1: link state changed to UP
<6>igb1: link state changed to DOWN
<6>igb1: link state changed to UP


Fatal trap 12: page fault while in kernel mode
cpuid = 2; apic id = 02
fault virtual address	= 0xffff
fault code		= supervisor read data, page not present
instruction pointer	= 0x20:0xffffffff83cf7396
stack pointer	        = 0x28:0xfffffe0089906ac0
frame pointer	        = 0x28:0xfffffe0089906ac0
code segment		= base 0x0, limit 0xfffff, type 0x1b
			= DPL 0, pres 1, long 1, def32 0, gran 1
processor eflags	= interrupt enabled, resume, IOPL = 0
current process		= 19 (dp_zil_clean_taskq_)
trap number		= 12
panic: page fault
cpuid = 2
time = 1630329038
KDB: enter: panic

So as you said, this is not related at all with Suricata. netmap is out.

Ok, so in this log there'r "promiscuous mode" changes, and even if that is normal, I uninstalled darkstat anyway.

Now the only other pkg installed is pfBlockerNG-devel. Let's see if my system will crash again in the next days....

Thanks.

[stephenw10]

You have the backtrace from the crash report?

[None 0]

@stephenw10 Hi! Yes, here: ddb.txt

[stephenw10]

Mmm, well very similar but not identical:

db:0:kdb.enter.default>  bt
Tracing pid 40766 tid 100593 td 0xfffff8013f829740
kdb_enter() at kdb_enter+0x37/frame 0xfffffe00910f4530
vpanic() at vpanic+0x197/frame 0xfffffe00910f4580
panic() at panic+0x43/frame 0xfffffe00910f45e0
trap_fatal() at trap_fatal+0x391/frame 0xfffffe00910f4640
trap_pfault() at trap_pfault+0x4f/frame 0xfffffe00910f4690
trap() at trap+0x286/frame 0xfffffe00910f47a0
calltrap() at calltrap+0x8/frame 0xfffffe00910f47a0
--- trap 0xc, rip = 0xffffffff8120874b, rsp = 0xfffffe00910f4870, rbp = 0xfffffe00910f4880 ---
vm_radix_remove() at vm_radix_remove+0x1b/frame 0xfffffe00910f4880
vm_page_free_prep() at vm_page_free_prep+0x55/frame 0xfffffe00910f48a0
vm_page_free_toq() at vm_page_free_toq+0x12/frame 0xfffffe00910f48d0
vm_object_page_remove() at vm_object_page_remove+0x61/frame 0xfffffe00910f4930
vm_map_entry_delete() at vm_map_entry_delete+0x104/frame 0xfffffe00910f4980
vm_map_delete() at vm_map_delete+0x184/frame 0xfffffe00910f49e0
vm_map_remove() at vm_map_remove+0xab/frame 0xfffffe00910f4a10
vmspace_exit() at vmspace_exit+0xcb/frame 0xfffffe00910f4a50
exit1() at exit1+0x55b/frame 0xfffffe00910f4ab0
sys_sys_exit() at sys_sys_exit+0xd/frame 0xfffffe00910f4ac0
amd64_syscall() at amd64_syscall+0x387/frame 0xfffffe00910f4bf0
fast_syscall_common() at fast_syscall_common+0xf8/frame 0xfffffe00910f4bf0
--- syscall (1, FreeBSD ELF64, sys_sys_exit), rip = 0x800c2a00a, rsp = 0x7fffffffec38, rbp = 0x7fffffffec50 ---

What is that running on?

[None 0]

@stephenw10

What is that running on?

Sorry, what you mean?

[bmeeks]

@none-0 said in Help Understanding a Crash [kernel panic]:

@stephenw10

What is that running on?

Sorry, what you mean?

I suspect he means what type of hardware -- Netgate appliance (and if so, which model, as different models have different CPU families) or generic Intel/AMD hardware.

[None 0]

@bmeeks @stephenw10

It's a generic Intel box:
CPU: i3-4170
RAM: Kingston 2x8GB
Mobo: Asus (don't remember the model, but I can check)
Network adp: Intel I350-T4V2
login-to-view