search

dtaht / sch_cake

Out of tree build for the new cake qdisc
100 stars 35 forks source link

potential bug filed over at openwrt's tracker #119

Open moeller0 opened 5 years ago

moeller0 commented 5 years ago

https://bugs.openwrt.org/index.php?do=details&task_id=2346 Has a report for issues with cake and conntrack which might be interesting.

chromi commented 5 years ago

It looks like a build problem to me, involving a heavily patched "vendor kernel". Probably Kevin knows the most about the nuances of this.

tohojo commented 5 years ago

Jonathan Morton notifications@github.com writes:

It looks like a build problem to me, involving a heavily patched "vendor kernel". Probably Kevin knows the most about the nuances of this.

Yeah, it's definitely something weird in the build system; question is whether this is an upstream openwrt issue or if it's patched...

ghost commented 4 years ago

Does it require some particular symbol setting enabled in the kernel build conf?

syslog prints

sch_cake: Unknown symbol nf_conntrack_find_get (err -2) sch_cake: Unknown symbol nf_ct_get_tuplepr (err -2)

and errno 2 hints

ENOENT 2 No such file or directory

tohojo commented 4 years ago

n8v8R notifications@github.com writes:

Does it require some particular symbol setting enabled in the kernel build conf?

syslog prints

sch_cake: Unknown symbol nf_conntrack_find_get (err -2) sch_cake: Unknown symbol nf_ct_get_tuplepr (err -2)

and errno 2 hints

ENOENT 2 No such file or directory

Those look like missing conntrack functions; they should be ifdef'ed out if conntrack is not enabled, though:

if IS_REACHABLE(CONFIG_NF_CONNTRACK)

ghost commented 4 years ago

lsmod | grep sch_cake

nf_conntrack           77824 48 sch_cake,nf_nat_pptp,nf_conntrack_pptp,ipt_MASQUERADE,xt_state,xt_nat,xt_helper,xt_conntrack,xt_connmark,xt_connlimit,xt_connbytes,xt_REDIRECT,xt_CT,nft_redir_ipv6,nft_redir_ipv4,nft_redir,nft_nat,nft_masq_ipv6,nft_masq_ipv4,nft_masq,nft_flow_offload,nft_ct,nf_nat_tftp,nf_nat_snmp_basic,nf_nat_sip,nf_nat_masquerade_ipv6,nf_nat_masquerade_ipv4,nf_nat_irc,nf_conntrack_ipv6,nf_nat_ipv6,nf_conntrack_ipv4,nf_nat_ipv4,nf_nat_h323,nf_nat_ftp,nf_nat_amanda,nf_nat,nf_flow_table,nf_conntrack_tftp,nf_conntrack_snmp,nf_conntrack_sip,nf_conntrack_rtcache,nf_conntrack_proto_gre,nf_conntrack_netlink,nf_conntrack_irc,nf_conntrack_h323,nf_conntrack_ftp,nf_conntrack_broadcast,nf_conntrack_amanda
sch_cake               40960  0
CONFIG_NF_CONNTRACK=m
CONFIG_NF_CONNTRACK_MARK=y
CONFIG_NF_CONNTRACK_ZONES=y
CONFIG_NF_CONNTRACK_PROCFS=y
CONFIG_NF_CONNTRACK_EVENTS=y
CONFIG_NF_CONNTRACK_RTCACHE=m
# CONFIG_NF_CONNTRACK_TIMEOUT is not set
# CONFIG_NF_CONNTRACK_TIMESTAMP is not set
CONFIG_NF_CONNTRACK_LABELS=y
CONFIG_NF_CONNTRACK_AMANDA=m
CONFIG_NF_CONNTRACK_FTP=m
CONFIG_NF_CONNTRACK_H323=m
CONFIG_NF_CONNTRACK_IRC=m
CONFIG_NF_CONNTRACK_BROADCAST=m
# CONFIG_NF_CONNTRACK_NETBIOS_NS is not set
CONFIG_NF_CONNTRACK_SNMP=m
CONFIG_NF_CONNTRACK_PPTP=m
# CONFIG_NF_CONNTRACK_SANE is not set
CONFIG_NF_CONNTRACK_SIP=m
CONFIG_NF_CONNTRACK_TFTP=m
CONFIG_NETFILTER_XT_MATCH_CONNTRACK=m
CONFIG_NF_CONNTRACK_IPV4=m
CONFIG_NF_CONNTRACK_IPV6=m

What is potentially missing and where?

tohojo commented 4 years ago

n8v8R notifications@github.com writes:

lsmod | grep sch_cake

nf_conntrack           77824 48 sch_cake,nf_nat_pptp,nf_conntrack_pptp,ipt_MASQUERADE,xt_state,xt_nat,xt_helper,xt_conntrack,xt_connmark,xt_connlimit,xt_connbytes,xt_REDIRECT,xt_CT,nft_redir_ipv6,nft_redir_ipv4,nft_redir,nft_nat,nft_masq_ipv6,nft_masq_ipv4,nft_masq,nft_flow_offload,nft_ct,nf_nat_tftp,nf_nat_snmp_basic,nf_nat_sip,nf_nat_masquerade_ipv6,nf_nat_masquerade_ipv4,nf_nat_irc,nf_conntrack_ipv6,nf_nat_ipv6,nf_conntrack_ipv4,nf_nat_ipv4,nf_nat_h323,nf_nat_ftp,nf_nat_amanda,nf_nat,nf_flow_table,nf_conntrack_tftp,nf_conntrack_snmp,nf_conntrack_sip,nf_conntrack_rtcache,nf_conntrack_proto_gre,nf_conntrack_netlink,nf_conntrack_irc,nf_conntrack_h323,nf_conntrack_ftp,nf_conntrack_broadcast,nf_conntrack_amanda
sch_cake               40960  0
CONFIG_NF_CONNTRACK=m
CONFIG_NF_CONNTRACK_MARK=y
CONFIG_NF_CONNTRACK_ZONES=y
CONFIG_NF_CONNTRACK_PROCFS=y
CONFIG_NF_CONNTRACK_EVENTS=y
CONFIG_NF_CONNTRACK_RTCACHE=m
# CONFIG_NF_CONNTRACK_TIMEOUT is not set
# CONFIG_NF_CONNTRACK_TIMESTAMP is not set
CONFIG_NF_CONNTRACK_LABELS=y
CONFIG_NF_CONNTRACK_AMANDA=m
CONFIG_NF_CONNTRACK_FTP=m
CONFIG_NF_CONNTRACK_H323=m
CONFIG_NF_CONNTRACK_IRC=m
CONFIG_NF_CONNTRACK_BROADCAST=m
# CONFIG_NF_CONNTRACK_NETBIOS_NS is not set
CONFIG_NF_CONNTRACK_SNMP=m
CONFIG_NF_CONNTRACK_PPTP=m
# CONFIG_NF_CONNTRACK_SANE is not set
CONFIG_NF_CONNTRACK_SIP=m
CONFIG_NF_CONNTRACK_TFTP=m
CONFIG_NETFILTER_XT_MATCH_CONNTRACK=m
CONFIG_NF_CONNTRACK_IPV4=m
CONFIG_NF_CONNTRACK_IPV6=m

What is potentially missing and where?

Nothing, it looks like?

The thing is also that the error you posted should make the module load fail. So since sch_cake is in fact loaded, I'm wondering if the error is a false positive? Maybe the module load order is off? I.e., if it tries to load the qdisc before conntrack at boot, that will probably fail, but then it tries again later (when the qdisc is applied), and it succeeds?

ghost commented 4 years ago

Fine if it is a ghost, just seeing such error reports in the logs is a bit confusing and led me to a quest for the cause of it and potential remedy.

tohojo commented 4 years ago

n8v8R notifications@github.com writes:

Fine if it is a ghost, just seeing such error reports in the logs is a bit confusing and led me to a quest for the cause of it and potential remedy.

I think it is; normally this would be fixed by module dependency resolution. So maybe that is borked? Not sure how that is done on openwrt, actually...

ghost commented 4 years ago

If am not mistaken interdependency of module load order (e.g. load B only after A is loaded) is not feasible in OpenWrt. Instead the order seems being statically set via

ghost commented 4 years ago

Suppose the issue can be closed since the conclusion appears to be a negligible ghost with no inclement impact (least I hope so)

tohojo commented 4 years ago

n8v8R notifications@github.com writes:

If am not mistaken interdependency of module load order (e.g. load B only after A is loaded) is not feasible in OpenWrt. Instead the order seems being statically set via

  • /etc/modules-boot.d/
  • /etc/modules.d/

Right, well in that case maybe the fix for openwrt is just to drop a file for sch_cake in there somewhere?

ghost commented 4 years ago

There is

ls -af /etc/modules.d/ | grep sch

70-sched-core 75-sched-cake 72-sched-bpf

but I am not sure how that fairs with regard to

Maybe the module load order is off? I.e., if it tries to load the qdisc before conntrack at boot, that will probably fail, but then it tries again later (when the qdisc is applied), and it succeeds?

tohojo commented 4 years ago

n8v8R notifications@github.com writes:

There is

ls -af /etc/modules.d/ | grep sch

70-sched-core 75-sched-cake 72-sched-bpf

but I am not sure how that fairs with regard to

Maybe the module load order is off? I.e., if it tries to load the qdisc before conntrack at boot, that will probably fail, but then it tries again later (when the qdisc is applied), and it succeeds?

I assume they're loaded in sort order; and the nf_conntrack file in that dir isn't numbered, so it would come after. I think we may have found the source of the error :)

ghost commented 4 years ago

To my understanding it is an alphanumerical order:

tohojo commented 4 years ago

n8v8R notifications@github.com writes:

To my understanding is is a alphanumerical order:

  • numericals first, lowest first and ascending
  • alphabetical next, start with a and ascending

Yup, which puts '75-sched-cake' before 'nf-conntrack' :)

moeller0 commented 4 years ago

On Jan 10, 2020, at 12:43, Toke Høiland-Jørgensen notifications@github.com wrote:

n8v8R notifications@github.com writes:

To my understanding is is a alphanumerical order:

  • numericals first, lowest first and ascending
  • alphabetical next, start with a and ascending

Yup, which puts '75-sched-cake' before 'nf-conntrack' :) — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.

Why is 75-sched-cake in there at all? sqm-scripts will try to load cake's module on start, so his might not be needed in there at all?

ghost commented 4 years ago

Why is 75-sched-cake in there at all?

It is provided by kmod-sched-cake

tohojo commented 4 years ago

n8v8R notifications@github.com writes:

Why is 75-sched-cake in there at all?

It is provided by kmod-sched-cake

I think it's standard to provide these files for all kernel modules on openwrt. Since there's no module autoloading, if we get rid of it people won't be able to use the qdisc outside of sqm-scripts, which is probably not what we want...

ghost commented 4 years ago

Having changed

/etc/modules.d/71-nf-conntrack /etc/modules.d/71-nf-conntrack-netlink

and rebooted and the errors are not exhibiting any more

tohojo commented 4 years ago

n8v8R notifications@github.com writes:

Having changed

/etc/modules.d/71-nf-conntrack /etc/modules.d/71-nf-conntrack-netlink

and rebooted and the errors are not exhibiting any more

Great, there we go, then! :)

ghost commented 4 years ago

@tohojo would you be kind enough to sponsor a PR at OpenWrt?

tohojo commented 4 years ago

n8v8R notifications@github.com writes:

@tohojo would be kind enough to sponsor a PR at OpenWrt?

Don't think I have permissions for that in the core openwrt repo...

dtaht commented 2 years ago

not sure if this is still an issue?