PfSense crashed on Alix
-
Unfortunately with the snapshot server out of commission until the new one is put in place there isn't much to do or try except keep track of the traces.
-
Yeah well but this doesn't answer my question what the real problem is. Or do you think that these problems were silently fixed in a new snapshot?
-
It's hard to say with any certainty until someone with more in-depth knowledge of the freebsd kernel (such as ermal) can have a look and see if he can tell what is going on.
-
Yep. That's what I'm waiting for ;)
-
I've looked at a lot of FreeBSD dumps. This sort of problem is sometimes fairly straight forward to find but can also be very difficult to find. It can have a variety of causes including passing the wrong type of data structure to a function and freeing a data structure then reusing it while its being used for another purpose.
If I was looking at this problem I expect the most useful items of information to me would be
-
a precise identification of the build on the which the problem was observed
-
a way of making it happen, even if it makes it happen only one in four times
One of the back traces shows:
ucom_attach(c0d56e6d,c0cd10c0,c2378cb0,c2378c98,0,…) at ucom_attach+0x542b
ucom_attach(c24ab000,0,109,cd9a2d5d,38ea,...) at ucom_attach+0x89d7
The offsets can be misleadng in that static functions don't appear in the symbol table available to the crash time debugger. Since 0x1000 is 4k, 0x89d7 is at least 32k and its pretty unlikely that an attach function would have anything like that amount of code. This offset likely is in some static function whose code starts at a higher address than the code for ucom_attach.Another of the reports shows:
Fatal trap 12: page fault while in kernel mode
fault virtual address = 0x72636524
fault code       = supervisor write, page not present
instruction pointer  = 0x20:0xc096993c
stack pointer     = 0x28:0xc2378b10
frame pointer     = 0x28:0xc2378b64If you look at the virtual address you might notice that it could be considered to be printable text: "?ecr" (the ? is for the character who binary representation is 0x24; I don't have the mapping from 0x24 to printable character in my head). From the reported code it would appear that a data structure referenced by rn_match has a text string where rn_match is expecting it to hold the address of another data structure. The challenge is to find out how that happened.
-
-
What I see now is that this happens every 3-4 days. So I guess I will do a reboot now every night via cron to see if this then stops until I have better builds…
-
With the daily reboot in place I am not seeing this problem anymore. So what is the status of these problems? Has anyone (ermal) taken a look at the bt's? Is this "problem" fixed in newer snaps?
-
Today this happened again. So I cannot use this workaround :(
Here is the bt:
rn_match(c0cd504c,c283fd00,0,c2981718,e2992850,…) at rn_match+0x17
pfr_match_addr(c288b9b0,c2741034,2,e299283c,e2992838,...) at pfr_match_addr+0x63
pf_test_tcp(e2992938,e2992934,1,c26c4600,c272bd00,...) at pf_test_tcp+0x4cb
pf_test(1,c2610400,e2992afc,0,0,...) at pf_test+0x8d2
init_pf_mutex(0,e2992afc,c2610400,1,0,...) at init_pf_mutex+0x5e6
pfil_run_hooks(c0cfd1c0,e2992b4c,c2610400,1,0,...) at pfil_run_hooks+0x7e
ip_input(c272bd00,246,c24d38c0,e2992b74,c06fd9a1,...) at ip_input+0x278
netisr_dispatch_src(1,0,c272bd00,e2992bac,c08e3f0f,...) at netisr_dispatch_src+0x89
netisr_dispatch(1,c272bd00,c2610400,c2610400,c274101a,...) at netisr_dispatch+0x20
ether_demux(c2610400,c272bd00,3,0,3,...) at ether_demux+0x16f
ether_vlanencap(c2610400,c272bd00,ece0,18,c272bd00,...) at ether_vlanencap+0x43f
ieee80211_hostap_detach(c2700000,c315a000,c272bd00,c2532480,c2438d80,...) at ieee80211_hostap_detach+0x362
ieee80211_hostap_detach(c315a000,c272bd00,17,ffffffa0,0,...) at ieee80211_hostap_detach+0x29a7
ath_suspend(c2514000,1,0,c0ca937c,0,...) at ath_suspend+0x1f67
taskqueue_run(c251d100,c251d118,0,c0b53f14,0,...) at taskqueue_run+0x132
taskqueue_thread_loop(c2514270,e2992d38,0,0,0,...) at taskqueue_thread_loop+0x88
fork_exit(c086b060,c2514270,e2992d38) at fork_exit+0x90
fork_trampoline() at fork_trampoline+0x8
--- trap 0, eip = 0, esp = 0xe2992d70, ebp = 0 ---Please guys. Give me any info. What else do you need? Does nobody use 2.0-beta1 on Alix boards? Can't be...
-
I use 2.0-beta1 on my ALIX but has not crashed on me yet. I haven't passed much traffic through it though as it's just been used for light testing and such.
-
i use 22nd snapshot on alix, hasnt crashed for me till now so might be some hardware issue or something like that
-
I don't think it is hardware related since 1.2.3 is running fine on this box. This just happened now with 2.0-beta1…
-
We're looking into it.
-
Thanks! Is there anyway I can tell pfSense to reboot automatically when it panics? But I guess no :-(
I think I will go back to 1.2.3 because my girlfriend hates me everytime the internet connection dies and now it is almost daily ;)Downgrade is only working by re-flashing? I have an old 1.2.3 conf…
-
I can confirm now that it is definitely not a hardware issue. Now my box is running fine again with 1.2.3. I will use 2.0 again when it has RC status the earliest…
Thanks
-
I just had what is possibly the same problem on an old Dell OptiPlex GX200 with a dual Intel gigabit card installed.
LAN IP was completely dead, and I did not have a keyboard so no backtrace:
em1: watchdog timeout – resetting
Fatal trap 12: page fault while in kernel mode
cpuid = 0; apic id = 00
fault virtual address = 0xe0500a4
fault code = supervisor read, page not present
instruction pointer - 0x20:0xc0a63aa7
stack pointer = 0x28:0xe2c547c4
frame pointer = 0x28:0xe2c547f0
code segment = base 0x0, limit 0xfffff, type 0x1b
  DPL 0, pres 1, def32 1, gran 1
processor eflags = interrupt enabled, resume, IOPL = 0
current process = 0 (em0 taskq) -
Stopped at rn_match+0x17: movl 0xc(%eax),%ebx
It happened again some 20 hours later.
LAN dead again, but stopped at exactly the same instruction.
Since writing down the bactrace was too painful, I took a picture instead.
-
I just had what is possibly the same problem on an old Dell OptiPlex GX200 with a dual Intel gigabit card installed.
LAN IP was completely dead, and I did not have a keyboard so no backtrace:
em1: watchdog timeout – resetting
Fatal trap 12: page fault while in kernel mode
cpuid = 0; apic id = 00
fault virtual address = 0xe0500a4
fault code = supervisor read, page not present
instruction pointer - 0x20:0xc0a63aa7
stack pointer = 0x28:0xe2c547c4
frame pointer = 0x28:0xe2c547f0
code segment = base 0x0, limit 0xfffff, type 0x1b
   DPL 0, pres 1, def32 1, gran 1
processor eflags = interrupt enabled, resume, IOPL = 0
current process = 0 (em0 taskq) -
I've had multiple similar problems. Unfortunately I haven't been able to grab the log since the problems have occured in production environments. Network just stops suddenly working. It can run well for days or just for an hour or so.
My config includes use of limiters. Is it possible that it causes problems?
That's just my hunch, since I've got older snapshots running fine without limiters.My config is fairly simple.
WAN and another WAN on OPT.
A couple NAT/FW rules inbound.
Nothing else. -
my alix with 20th feb snapshot works perfect and older versions also have been running stable enough for me for as much as 15 days then its no crash but i usually endup trying newer snapshots.
-
Stopped at rn_match+0x17: movl 0xc(%eax),%ebx
It happened again some 20 hours later.
LAN dead again, but stopped at exactly the same instruction.
Since writing down the bactrace was too painful, I took a picture instead.For the past 5 days I have been running completely stable on 1.2.3 using the same hardware and configuration (recreated configuration since downgrading is not possible).
Something bad is happening in 2.0 for sure…
