Puppet » Facter

it seems there was a change in util/virtual.rb:

return true if txt =~ /^{(s_context|VxID)::blank:*[1-9]/}

should be

return true if txt =~ /^{(s_context|VxID)::blank:*[0-9]/}

i am using vserver v 2.3

regards, Jean.

the logic changed:

in 1.5.6 it search /proc/virtual to see if this is a host, if it find a “/proc/self/status” with starting vxid 1-9 then this is a guest.

in 1.5.7 it search a vxid in /proc/self/status and if it find one it search for /proc/virtual BUT the issue is that host have xid of 0 on vserver 2.3 and with vserver 2.1 you do NOT have vixd in proc/self status on the host at all…so the logic fails completly.

here is for 2.3 :

FROM HOST: Name: cat State: R (running) Tgid: 21074 Pid: 21074 PPid: 21020 TracerPid: 0 Uid: 47000 47000 47000 47000 Gid: 4733 4733 4733 4733 FDSize: 64 Groups: 0 4733 VmPeak: 3800 kB VmSize: 3800 kB VmLck: 0 kB VmHWM: 468 kB VmRSS: 468 kB VmData: 176 kB VmStk: 84 kB VmExe: 32 kB VmLib: 1432 kB VmPTE: 28 kB Threads: 1 SigQ: 0/71680 SigPnd: 0000000000000000 ShdPnd: 0000000000000000 SigBlk: 0000000000000000 SigIgn: 0000000000000000 SigCgt: 0000000000000000 CapInh: 0000000000000000 CapPrm: 0000000000000000 CapEff: 0000000000000000 CapBnd: fffffffffffffeff Cpus_allowed: ff Cpus_allowed_list: 0-7 Mems_allowed: 1 Mems_allowed_list: 0 VxID: 0 NxID: 0 voluntary_ctxt_switches: 2 nonvoluntary_ctxt_switches: 0

FROM GUEST: Name: cat State: R (running) Tgid: 21149 Pid: 21149 PPid: 21142 TracerPid: 0 Uid: 0 0 0 0 Gid: 0 0 0 0 FDSize: 64 Groups: 0 VmPeak: 1564 kB VmSize: 1564 kB VmLck: 0 kB VmHWM: 384 kB VmRSS: 384 kB VmData: 160 kB VmStk: 84 kB VmExe: 16 kB VmLib: 1284 kB VmPTE: 20 kB Threads: 1 SigQ: 0/71680 SigPnd: 0000000000000000 ShdPnd: 0000000000000000 SigBlk: 0000000000000000 SigIgn: 0000000000000000 SigCgt: 0000000000000000 CapInh: 0000000000000000 CapPrm: fffffffffffffeff CapEff: fffffffffffffeff CapBnd: fffffffffffffeff Cpus_allowed: ff Cpus_allowed_list: 0-7 Mems_allowed: 1 Mems_allowed_list: 0 VxID: 40128 NxID: 40128 voluntary_ctxt_switches: 1 nonvoluntary_ctxt_switches: 0

from 2.1 now:

FROM HOST: Name: cat State: R (running) SleepAVG: 88% Tgid: 24625 Pid: 24625 PPid: 24618 TracerPid: 0 Uid: 47000 47000 47000 47000 Gid: 4733 4733 4733 4733 FDSize: 32 Groups: 0 4733 VmPeak: 1768 kB VmSize: 1768 kB VmLck: 0 kB VmHWM: 396 kB VmRSS: 396 kB VmData: 160 kB VmStk: 88 kB VmExe: 28 kB VmLib: 1468 kB VmPTE: 12 kB Threads: 1 SigQ: 0/4294967295 SigPnd: 0000000000000000 ShdPnd: 0000000000000000 SigBlk: 0000000000000000 SigIgn: 0000000000000000 SigCgt: 0000000000000000 CapInh: 0000000000000000 CapPrm: 0000000000000000 CapEff: 0000000000000000 s_context: 0 ctxflags: none initpid: none ipv4root: 0 ipv4root_bcast: 0

FROM GUEST: Name: cat State: R (running) SleepAVG: 58% Tgid: 24671 Pid: 24671 PPid: 24670 TracerPid: 0 Uid: 0 0 0 0 Gid: 0 0 0 0 FDSize: 32 Groups: 0 VmPeak: 1580 kB VmSize: 1580 kB VmLck: 0 kB VmHWM: 372 kB VmRSS: 372 kB VmData: 152 kB VmStk: 88 kB VmExe: 16 kB VmLib: 1280 kB VmPTE: 12 kB Threads: 1 SigQ: 0/4294967295 SigPnd: 0000000000000000 ShdPnd: 0000000000000000 SigBlk: 0000000000000000 SigIgn: 0000000000000000 SigCgt: 0000000000000000 CapInh: 0000000000000000 CapPrm: 00000000344c04ff CapEff: 00000000344c04ff s_context: 40074 ctxflags: 1602020010 initpid: 0 ipv4root: 4a00007f/ffffffff 4a24f6d5/00ffffff ipv4root_bcast: 00000000

Actually, the vserver_host should always have s_context: 0. s_context is the security context of the vserver, 0 should be the host.

Rein Henrichs wrote:

So far, it is my understanding that:

1. /proc/self/status for vserver 2.1 contains a VxID field while 2.3 contains a s_context field.
2. In each version, these fields can be used to determine whether we are a vserver

Here’s where my confusion starts.

Jean’s suggested change from:

return true if txt =~ /(s_context|VxID):[[:blank:]]*[1-9]/

to:

return true if txt =~ /(s_context|VxID):[[:blank:]]*[0-9]/

would imply that vserver? should be true if we are a guest or a host. Currently only the guest (non-zero VxID or s_context) returns true for vserver?. Is this correct? My assumption is that the suggested 0-9 behavior is correct (after all, why else would this ticket exist?) but I would like confirmation.

I have a testing harness in place using Jean’s sample data and can submit a patch quickly once I figure out the correct behavior. I’ve pushed my (work in progress) branch to my github at wip/2865 if anyone wants to check my work.

Looks good and seems to match my understanding of the issue too.

hi,

yes the test is there to see it this is a vserver enabled kernel, a second test “vserver_type” is there to see if it is the host or the guest. So this firts test should include the 0 as host are obviously vserver enabled .

def self.vserver?
    return false unless FileTest.exists?("/proc/self/status")
    txt = File.read("/proc/self/status")
    return true if txt =~ /^(s_context|VxID):[[:blank:]]*[1-9]/ <=========== SHOULD INCLUDE 0 TOO
    return false
end
def self.vserver_type
    if self.vserver?
        if FileTest.exists?("/proc/virtual")
            "vserver_host"
        else
            "vserver"
        end
     end
end

virtual => vserver virtual => vserver_host

the second test “vserver_type” is done only if the first return true, the current code just forbid the test to work on host as their vxid is 0 therefor host are considered normal plain linux server and not vserver enabled ones breaking previous behavior of facter (and my recipe).

regards, Jean