PfSense 2.4.3-RELEASE hang/crash reboots - "Fatal trap 9:"
Hardware: -As a virtual machine on top of VMWare ESXi 5.5 Update 2 (build 5230635).
- Machine has 4 vCPU and 1 GB RAM assigned to it along with a 20 GB thin provisioned vDisk
- VM Version “vmx-09”.
- 2 x NICs (em0 and em1)
- em0 is used for WAN connectivity
- em1 is used as a VLAN trunk (I have about 10 VLANs for various networks)
What this machine does:
- Internet Connectivity
- All our remote sites VPN back to this pfSense over IPSEC
- Acme (Lets Encrypt Cert renewal)
The problem: pfSense reboots at random times (sometimes during heavy load, sometimes when there is no load). This virtual machine has run without issues for several years now only to start exhibiting random reboots from time to time in the past month or so. It happens once every 2-3 days on average (sometimes more frequently). If I log into pfSense just after the reboot, there is a “crash dump available” message at the top of the screen.
At the very bottom of the crash dump (which is thousands of lines long of course)
Fatal trap 9: general protection fault while in kernel mode cpuid = 2; apic id = 02 instruction pointer = 0x20:0xffffffff80e81172 stack pointer = 0x28:0xfffffe004ce829b0 frame pointer = 0x28:0xfffffe004ce82a40 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 = 12 (swi4: clock (0)) version.txt06000027413265503336 7624 ustarrootwheelFreeBSD 11.1-RELEASE-p7 #10 r313908+986837ba7e9(RELENG_2_4): Mon Mar 26 18:08:25 CDT 2018 email@example.com:/builder/ce-243/tmp/obj/builder/ce-243/tmp/FreeBSD-src/sys/pfSense
So far, I have done a reinstall of pfSense using the latest 2.4.3 media and by using the “rescue config.xml” option in the shell. Obviously this has not helped. What else can I do to figure out why this is happening? As far as I am aware, there were no changes made to our infrastructure when this issue started happening so not too sure what could be the root cause.
That’s not enough of the crash dump to say for sure what happened. Need to see the backtrace near the start of the dump. It should be keeping a copy of that crash dump and showing it to you in the GUI after it reboots.
If not then perhaps your VM doesn’t have any swap space configured.
jimp, am I able to PM the crash dump to you? I’d rather not post that stuff on a public forum.
Yes, that’s fine.
Thanks jimp, much appreciated. I have pm’ed the logs to you.
OK, nothing private in the backtrace so that should be OK here (from the archive you sent):
db:0:kdb.enter.default> bt Tracing pid 0 tid 100048 td 0xfffff8000365d000 key_addref() at key_addref+0x4/frame 0xfffffe004cf96640 ipsec_getpcbpolicy() at ipsec_getpcbpolicy+0x51/frame 0xfffffe004cf96680 ipsec4_getpolicy() at ipsec4_getpolicy+0x25/frame 0xfffffe004cf96720 ipsec4_in_reject() at ipsec4_in_reject+0x1d/frame 0xfffffe004cf96750 udp_append() at udp_append+0xaa/frame 0xfffffe004cf967c0 udp_input() at udp_input+0x49b/frame 0xfffffe004cf96880 ip_input() at ip_input+0x135/frame 0xfffffe004cf968e0 netisr_dispatch_src() at netisr_dispatch_src+0xa0/frame 0xfffffe004cf96930 ether_demux() at ether_demux+0x16d/frame 0xfffffe004cf96960 ether_nh_input() at ether_nh_input+0x337/frame 0xfffffe004cf969c0 netisr_dispatch_src() at netisr_dispatch_src+0xa0/frame 0xfffffe004cf96a10 ether_input() at ether_input+0x26/frame 0xfffffe004cf96a30 if_input() at if_input+0xa/frame 0xfffffe004cf96a40 lem_rxeof() at lem_rxeof+0x3ef/frame 0xfffffe004cf96ae0 lem_handle_rxtx() at lem_handle_rxtx+0x32/frame 0xfffffe004cf96b20 taskqueue_run_locked() at taskqueue_run_locked+0x147/frame 0xfffffe004cf96b80 taskqueue_thread_loop() at taskqueue_thread_loop+0xb8/frame 0xfffffe004cf96bb0 fork_exit() at fork_exit+0x85/frame 0xfffffe004cf96bf0 fork_trampoline() at fork_trampoline+0xe/frame 0xfffffe004cf96bf0
That appears to be a crash in IPsec key management, though I can’t recall seeing a crash ever happen there.
Is there any way you can try it on a current ESX like a fully patched 6.5/6.5U1/6.7?
pfSense 2.4.3 is based on FreeBSD 11.1 and VMWare claims that FreeBSD 11 is only compatible with 6.5 or newer. Though I know some people run it on 5.5, there is always a chance for compatibility issues.
Here’s another crash. The one I sent you was from yesterday AM. This one is from Wednesday last week (18th)
db:0:kdb.enter.default> bt Tracing pid 12 tid 100008 td 0xfffff800032eb5c0 key_timehandler() at key_timehandler+0x732/frame 0xfffffe004ce82a40 softclock_call_cc() at softclock_call_cc+0x13b/frame 0xfffffe004ce82af0 softclock() at softclock+0xb9/frame 0xfffffe004ce82b20 intr_event_execute_handlers() at intr_event_execute_handlers+0xec/frame 0xfffffe004ce82b60 ithread_loop() at ithread_loop+0xd6/frame 0xfffffe004ce82bb0 fork_exit() at fork_exit+0x85/frame 0xfffffe004ce82bf0 fork_trampoline() at fork_trampoline+0xe/frame 0xfffffe004ce82bf0 --- trap 0, rip = 0, rsp = 0, rbp = 0 ---
Does this look ipsec related too? I don’t see any mention of ipsec in there.
Anyway, I am getting desperate to resolve this so last night I rebuilt the VM with vmxnet3 interfaces rather than e1000.
Changing VMWare versions isn’t relaly an option at this stage - we are part way through a migration to prodmox.
Ok, unfortunately it crashed again just now. I’m not convinced this is an issue with pfsense/esxi interop issue. I’ve got ~17 pfSense 2.4.2-RELEASE, almost all of them of them running on VMWare ESXi 5.5 U3 without issue. That’s BSD 11.1 as well - none of those crash.
db:0:kdb.enter.default> bt Tracing pid 35907 tid 100123 td 0xfffff8000e19a5c0 in_pcbfree() at in_pcbfree+0x143/frame 0xfffffe0000208940 udp_detach() at udp_detach+0xa2/frame 0xfffffe0000208970 sofree() at sofree+0x101/frame 0xfffffe00002089a0 soclose() at soclose+0x366/frame 0xfffffe00002089f0 closef() at closef+0x264/frame 0xfffffe0000208a80 closefp() at closefp+0x9d/frame 0xfffffe0000208ac0 amd64_syscall() at amd64_syscall+0xa4c/frame 0xfffffe0000208bf0 fast_syscall_common() at fast_syscall_common+0x106/frame 0x7fffffffe810
Those last two do not appear to be related to IPsec. The fact that every panic is different would usually make me lean toward hardware though, but since it’s virtual that gets trickier.
They are vaguely like crashes we used to see a long time ago when the NIC queues had to be reduced to 1. You might try that, see if it helps. https://doc.pfsense.org/index.php/Tuning_and_Troubleshooting_Network_Cards#Intel_igb.284.29_and_em.284.29_Cards
That most recent crash - that happened with a vmxnet adapter (VMXNET 3 VMWare Paravirtual Adapter).
Do you still think it is worthwhile reducing the queues to one? The page you linked does not reference vmxnet network adapters at all. What is the line to put into /boot/loader.conf to reduce queues on the vmxnet adapter to 1?
I’m not sure there is actually a tunable for that, but you can try.
We only saw those particular crashes with igb, not vmxnet* or em.
This is still happening. It was fine for 4 days then restarted twice in an hour. Each time the stack trace shows some different faulting component.
I have set up a brand new VWMare ESXi host running VMWare ESXi 6.0 U3 (Released Feb 2018). I have moved our router VM to this host, edited virtual machine to latest (vm version 11) that is supported on ESXi 6.0 U3… lets see how this pans out.
One odd thing I’ve noticed is a message in the Web Client which says
The configured guest OS (FreeBSD (64-bit)) for this virtual machine does not match the guest that is currently running (FreeBSD 11.1-RELEASE-p7). You should specify the correct guest OS to allow for guest-specific optimizations.
Not sure what that’s all about…
That last bit is fixed in ESX 6.7. It’s harmless, though.
6.0 U3 still isn’t technically compatible with FreeBSD 11.x. It may work but ESX only claims support for 6.5 and later.
Ok, I am going to consider this “fixed” now. The system has been up for over 5 days. This is the longest the system has stayed up since the rebooting issue started (very unusual to make it past 3-4 days).
Must be some sort of interop issue with FreeBSD 11.x and VMWare 5.5 U3.
Note that I actually neglected to check the VM version. It was actually running VM Version 8 which is the default hardware version for ESXi 5.0 which is when this deployment was put in. Simply updating this to VMX version 10 which is the latest supported for ESXi 5.5 could potentially have solved the problem, I just totallly missed that this could be the cause of the issue.
Spoke too soon, it just crashed and rebooted.
db:0:kdb.enter.default> bt Tracing pid 7 tid 100044 td 0xfffff8000342a5c0 kdb_enter() at kdb_enter+0x3b/frame 0xfffffe004cf361e0 vpanic() at vpanic+0x1a3/frame 0xfffffe004cf36260 panic() at panic+0x43/frame 0xfffffe004cf362c0 complete_jsegs() at complete_jsegs+0x854/frame 0xfffffe004cf36310 softdep_disk_write_complete() at softdep_disk_write_complete+0x42c/frame 0xfffffe004cf36370 bufdone_finish() at bufdone_finish+0x34/frame 0xfffffe004cf363e0 bufdone() at bufdone+0x87/frame 0xfffffe004cf36400 g_io_deliver() at g_io_deliver+0x205/frame 0xfffffe004cf36460 g_io_deliver() at g_io_deliver+0x205/frame 0xfffffe004cf364c0 g_io_deliver() at g_io_deliver+0x205/frame 0xfffffe004cf36520 g_disk_done() at g_disk_done+0x129/frame 0xfffffe004cf36570 dadone() at dadone+0x1826/frame 0xfffffe004cf36b20 xpt_done_process() at xpt_done_process+0x677/frame 0xfffffe004cf36b60 xpt_done_td() at xpt_done_td+0x196/frame 0xfffffe004cf36bb0 fork_exit() at fork_exit+0x85/frame 0xfffffe004cf36bf0 fork_trampoline() at fork_trampoline+0xe/frame 0xfffffe004cf36bf0 --- trap 0, rip = 0, rsp = 0, rbp = 0 ---
Tonight, I am going to create a brand new VM with brand new config file/disks, install pfSense and restore config to see if that stops this.
If that doesn’t work I am out of ideas - I have other deployment that run this exact same setup (ESXi 5.5 U3, pfSense, IPSEC, pfBlocker NG) that don’t have this problem.
I didn’t end up rebuilding the router. That’s way too much work. I instead decided to disable AES-NI on the system and switch all IPSEC tunnels from AES-GCM to Blowfish to see if that will help.
It has now been over 7 days, no reboots. I have never made it this long without a reboot before.
Ok I am calling this fixed. I’ve got an uptime of 14 days after disabling AES/NI on this machine. Previously I couldn’t make it past 4-5 days.