Skip to contentSkip to conversation

ucc_geth: nf_conntrack: table full, dropping packet.

PATCH Applied Joakim Tjernlund 4 participants 44 messages Mar 23, 2009 – Mar 25, 2009 3 messages not in the archive

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

doing a "ping -f -l 3" on my host towards my board on linus tree as of Friday results in lots of:
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
__ratelimit: 11 callbacks suppressed
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.

for ucc_geth on a MPC832x.
This really looks strange to me, ideas?

Jocke

#1 lore raw
Quoted 10 lines · Joakim Tjernlund

doing a "ping -f -l 3" on my host towards my board on linus tree as of Friday results in lots of:
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
__ratelimit: 11 callbacks suppressed
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.

Quoted 2 lines

for ucc_geth on a MPC832x.
This really looks strange to me, ideas?

What does /proc/net/netfilter/nf_conntrack show?

#2 lore raw

Patrick McHardy <[email protected]> wrote on 23/03/2009 13:15:45:

Quoted 14 lines
Quoted 12 lines · Joakim Tjernlund

doing a "ping -f -l 3" on my host towards my board on linus tree as of

Friday results in lots of:
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet. __ratelimit: 11 callbacks suppressed
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.

Quoted 4 lines
Quoted 2 lines

for ucc_geth on a MPC832x.
This really looks strange to me, ideas?

What does /proc/net/netfilter/nf_conntrack show?

There is no /proc/net/netfilter/nf_conntrack. There is a /proc/net/nf_conntrack though and it is empty. If I telnet to the board I see:
ipv4 2 tcp 6 431990 ESTABLISHED src=192.168.1.15 dst=192.168.1.17
sport=56445 dport=23 src=192.168.1.17 dst=192.168.1.15 sport=23 dport=56445 [ASSURED] use=1
ipv4 2 udp 17 7 src=192.168.1.228 dst=192.168.1.255 sport=138
dport=138 [UNREPLIED] src=192.168.1.255 dst=192.168.1.228 sport=138 dport=138 use=1
ipv4 2 udp 17 20 src=127.0.0.1 dst=127.0.0.1 sport=34261 dport=53
[UNREPLIED] src=127.0.0.1 dst=127.0.0.1 sport=53 dport=34261 use=1
ipv4 2 udp 17 2 src=192.168.1.199 dst=192.168.1.255 sport=138
dport=138 [UNREPLIED] src=192.168.1.255 dst=192.168.1.199 sport=138 dport=138 use=1
ipv4 2 udp 17 20 src=127.0.0.1 dst=127.0.0.1 sport=40417 dport=53
[UNREPLIED] src=127.0.0.1 dst=127.0.0.1 sport=53 dport=40417 use=1

Jocke

#3 lore raw
Quoted 14 lines · Joakim Tjernlund

Patrick McHardy <[email protected]> wrote on 23/03/2009 13:15:45:

Quoted 13 lines
Quoted 12 lines · Joakim Tjernlund

doing a "ping -f -l 3" on my host towards my board on linus tree as of

Friday results in lots of:
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet. __ratelimit: 11 callbacks suppressed
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.

Quoted 7 lines
Quoted 3 lines
Quoted 2 lines

for ucc_geth on a MPC832x.
This really looks strange to me, ideas?

What does /proc/net/netfilter/nf_conntrack show?

There is no /proc/net/netfilter/nf_conntrack. There is a /proc/net/nf_conntrack though and it is empty. If I telnet to the board I see:

That means that something is leaking conntrack references, most likely by leaking skbs. Since I haven't seen any other reports, my guess would be the ucc_geth driver.

#4 lore raw

Patrick McHardy <[email protected]> wrote on 23/03/2009 13:29:33:

Quoted 4 lines
Quoted 3 lines · Joakim Tjernlund

Patrick McHardy <[email protected]> wrote on 23/03/2009 13:15:45:

Quoted 2 lines
Quoted 1 line · Joakim Tjernlund

doing a "ping -f -l 3" on my host towards my board on linus tree as

of

Quoted 15 lines
Quoted 15 lines
Quoted 14 lines
Quoted 11 lines

Friday results in lots of:
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet. __ratelimit: 11 callbacks suppressed
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.

Quoted 2 lines

for ucc_geth on a MPC832x.
This really looks strange to me, ideas?

What does /proc/net/netfilter/nf_conntrack show?

Quoted 7 lines
Quoted 3 lines

There is no /proc/net/netfilter/nf_conntrack. There is a /proc/net/nf_conntrack though and it is empty. If I telnet to the board I see:

That means that something is leaking conntrack references, most likely by leaking skbs. Since I haven't seen any other reports, my guess would be the ucc_geth driver.

hmm, I cannot see what in the ucc_geth driver is possibly "leaking". One thing
I do notice is that the board becomes almost unresponsive during the ping flood.
Perhaps it is building up a backlog of conntracks during the ping flood?

Jocke

A message that is not in the archive (sent off-list or not yet received): OF387EC803.F810F72A-ONC1257582.00468C6E-C1257582.00475783@LocalDomain

#6 lore raw

Joakim Tjernlund/Transmode wrote on 23/03/2009 13:59:15:

Quoted 6 lines

Patrick McHardy <[email protected]> wrote on 23/03/2009 13:29:33:

Quoted 4 lines
Quoted 3 lines · Joakim Tjernlund

Patrick McHardy <[email protected]> wrote on 23/03/2009 13:15:45:

Quoted 2 lines
Quoted 1 line · Joakim Tjernlund

doing a "ping -f -l 3" on my host towards my board on linus tree

as of

Quoted 13 lines
Quoted 13 lines
Quoted 13 lines
Quoted 13 lines
Quoted 13 lines

Friday results in lots of:
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet. __ratelimit: 11 callbacks suppressed
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.

for ucc_geth on a MPC832x.
This really looks strange to me, ideas?

Quoted 8 lines
Quoted 8 lines
Quoted 5 lines
Quoted 1 line

What does /proc/net/netfilter/nf_conntrack show?

There is no /proc/net/netfilter/nf_conntrack. There is a /proc/net/nf_conntrack though and it is empty. If I telnet to the board I see:

That means that something is leaking conntrack references, most likely by leaking skbs. Since I haven't seen any other reports, my guess

would

Quoted 3 lines
Quoted 1 line

be the ucc_geth driver.

hmm, I cannot see what in the ucc_geth driver is possibly "leaking". One

thing

Quoted 1 line

I do notice is that the board becomes almost unresponsive during the

ping flood.

Quoted 3 lines

Perhaps it is building up a backlog of conntracks during the ping flood?

Jocke

next hmm, ethtool -S eth0 shows:
...
rx-mismatch-drop-frames: 13
...

13 matches the number of lost frames after I stop the ping flood.

The MPC832x manual says this(if i found the correct counter):

REBASE+2C MisMatchDrop 32 Counts number of frames dropped due to MAC filtering process, (e.g.
Address Mismatch, Type mismatch) and that would
otherwise
considered good frame that would be transferred
to upper layers

Why would a ping flood result in Address Mismatch or Type mismatch?

Leo?

Jocke

#7 lore raw
In reply to Patrick McHardy

Patrick McHardy <[email protected]> wrote on 23/03/2009 13:29:33:

Quoted 5 lines
Quoted 3 lines · Joakim Tjernlund

Patrick McHardy <[email protected]> wrote on 23/03/2009 13:15:45:

Quoted 2 lines
Quoted 1 line · Joakim Tjernlund

doing a "ping -f -l 3" on my host towards my board on linus tree as

of

Quoted 15 lines
Quoted 15 lines
Quoted 14 lines
Quoted 11 lines

Friday results in lots of:
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet. __ratelimit: 11 callbacks suppressed
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.

Quoted 2 lines

for ucc_geth on a MPC832x.
This really looks strange to me, ideas?

What does /proc/net/netfilter/nf_conntrack show?

Quoted 8 lines
Quoted 3 lines

There is no /proc/net/netfilter/nf_conntrack. There is a /proc/net/nf_conntrack though and it is empty. If I telnet to the board I see:

That means that something is leaking conntrack references, most likely by leaking skbs. Since I haven't seen any other reports, my guess would be the ucc_geth driver.

Mucking around with the ucc_geth driver I found that if I:
- Move TX from IRQ to NAPI context
- double the weight.
- after booting up, wait a few mins until the JFFS2 GC kernel thread has
stopped
scanning the FS

Then the "nf_conntrack: table full, dropping packet." msgs stops. Does this seem right to you guys?

Jocke

#8 lore raw
Quoted 19 lines · Joakim Tjernlund

Patrick McHardy <[email protected]> wrote on 23/03/2009 13:29:33:

Quoted 8 lines
Quoted 4 lines

There is no /proc/net/netfilter/nf_conntrack. There is a /proc/net/nf_conntrack though and it is empty. If I telnet to the board I see:

That means that something is leaking conntrack references, most likely by leaking skbs. Since I haven't seen any other reports, my guess would be the ucc_geth driver.

Mucking around with the ucc_geth driver I found that if I:
- Move TX from IRQ to NAPI context
- double the weight.
- after booting up, wait a few mins until the JFFS2 GC kernel thread has
stopped
scanning the FS

Quoted 2 lines

Then the "nf_conntrack: table full, dropping packet." msgs stops. Does this seem right to you guys?

No. As I said, something seems to be leaking packets. You should be able to confirm that by checking the sk_buff slabs in /proc/slabinfo. If that *doesn't* show any signs of a leak, please run "conntrack -E" to capture the conntrack events before the "table full" message appears and post the output.

#9 lore raw

Patrick McHardy <[email protected]> wrote on 23/03/2009 18:49:15:

Quoted 10 lines
Quoted 8 lines · Joakim Tjernlund

Patrick McHardy <[email protected]> wrote on 23/03/2009 13:29:33:

Quoted 5 lines
Quoted 4 lines

There is no /proc/net/netfilter/nf_conntrack. There is a /proc/net/nf_conntrack though and it is empty. If I telnet to the board I see:

That means that something is leaking conntrack references, most

likely

Quoted 1 line
Quoted 1 line
Quoted 1 line

by leaking skbs. Since I haven't seen any other reports, my guess

would

Quoted 7 lines
Quoted 7 lines
Quoted 2 lines

be the ucc_geth driver.

Mucking around with the ucc_geth driver I found that if I:
- Move TX from IRQ to NAPI context
- double the weight.
- after booting up, wait a few mins until the JFFS2 GC kernel thread

has

Quoted 11 lines
Quoted 5 lines

stopped
scanning the FS

Then the "nf_conntrack: table full, dropping packet." msgs stops. Does this seem right to you guys?

No. As I said, something seems to be leaking packets. You should be able to confirm that by checking the sk_buff slabs in /proc/slabinfo. If that *doesn't* show any signs of a leak, please run "conntrack -E" to capture the conntrack events before the "table full" message appears and post the output.

skbuff does not differ much, but others do

Before ping:
skbuff_fclone_cache 0 0 352 11 1 : tunables 54 27 0
: slabdata 0 0 0
skbuff_head_cache 20 20 192 20 1 : tunables 120 60 0

: slabdata      1      1      0
  size-64              731    767     64   59    1 : tunables  120   60 0 
: slabdata     13     13      0

nf_conntrack 10 19 208 19 1 : tunables 120 60 0
: slabdata 1 1 0

During ping:
skbuff_fclone_cache 0 0 352 11 1 : tunables 54 27 0
: slabdata 0 0 0
skbuff_head_cache 40 40 192 20 1 : tunables 120 60 0
: slabdata 2 2 0
size-64 8909 8909 64 59 1 : tunables 120 60 0
: slabdata 151 151 0
nf_conntrack 5111 5111 208 19 1 : tunables 120 60 0
: slabdata 269 269 0

This feels more like the freeing of conntrack objects are delayed and builds up when ping flooding.

Don't have "conntrack -E" for my embedded board so that will have to wait a bit longer.

Jocke

#10 lore raw
Quoted 53 lines · Joakim Tjernlund

Patrick McHardy <[email protected]> wrote on 23/03/2009 18:49:15:

Quoted 9 lines
Quoted 8 lines · Joakim Tjernlund

Patrick McHardy <[email protected]> wrote on 23/03/2009 13:29:33:

Quoted 5 lines
Quoted 4 lines

There is no /proc/net/netfilter/nf_conntrack. There is a /proc/net/nf_conntrack though and it is empty. If I telnet to the board I see:

That means that something is leaking conntrack references, most

likely

Quoted 1 line
Quoted 1 line
Quoted 1 line

by leaking skbs. Since I haven't seen any other reports, my guess

would

Quoted 6 lines
Quoted 6 lines
Quoted 2 lines

be the ucc_geth driver.

Mucking around with the ucc_geth driver I found that if I:
- Move TX from IRQ to NAPI context
- double the weight.
- after booting up, wait a few mins until the JFFS2 GC kernel thread

has

Quoted 10 lines
Quoted 5 lines

stopped
scanning the FS

Then the "nf_conntrack: table full, dropping packet." msgs stops. Does this seem right to you guys?

No. As I said, something seems to be leaking packets. You should be able to confirm that by checking the sk_buff slabs in /proc/slabinfo. If that *doesn't* show any signs of a leak, please run "conntrack -E" to capture the conntrack events before the "table full" message appears and post the output.

skbuff does not differ much, but others do

Before ping:
skbuff_fclone_cache 0 0 352 11 1 : tunables 54 27 0
: slabdata 0 0 0
skbuff_head_cache 20 20 192 20 1 : tunables 120 60 0

: slabdata      1      1      0
  size-64              731    767     64   59    1 : tunables  120   60 0 
: slabdata     13     13      0

nf_conntrack 10 19 208 19 1 : tunables 120 60 0
: slabdata 1 1 0

During ping:
skbuff_fclone_cache 0 0 352 11 1 : tunables 54 27 0
: slabdata 0 0 0
skbuff_head_cache 40 40 192 20 1 : tunables 120 60 0
: slabdata 2 2 0
size-64 8909 8909 64 59 1 : tunables 120 60 0
: slabdata 151 151 0
nf_conntrack 5111 5111 208 19 1 : tunables 120 60 0
: slabdata 269 269 0

Quoted 5 lines

This feels more like the freeing of conntrack objects are delayed and builds up when ping flooding.

Don't have "conntrack -E" for my embedded board so that will have to wait a bit longer.

I dont understand how your ping can use so many conntrack entries...

Then, as I said yesterday, I believe you have a RCU delay, because of a misbehaving driver or something...

grep RCU .config
grep CONFIG_SMP .config

You could change qhimark from 10000 to 1000 in kernel/rcuclassic.c (line 80) as a workaround. It should force a quiescent state after 1000 freed conntracks.

#11 lore raw

Eric Dumazet <[email protected]> wrote on 24/03/2009 10:12:53:

Quoted 13 lines
Quoted 11 lines · Joakim Tjernlund

Patrick McHardy <[email protected]> wrote on 23/03/2009 18:49:15:

Quoted 9 lines
Quoted 8 lines · Joakim Tjernlund

Patrick McHardy <[email protected]> wrote on 23/03/2009 13:29:33:

Quoted 5 lines
Quoted 4 lines

There is no /proc/net/netfilter/nf_conntrack. There is a /proc/net/nf_conntrack though and it is empty. If I telnet to the board I see:

That means that something is leaking conntrack references, most

likely

Quoted 8 lines
Quoted 8 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line

by leaking skbs. Since I haven't seen any other reports, my guess

would

Quoted 6 lines
Quoted 6 lines
Quoted 2 lines

be the ucc_geth driver.

Mucking around with the ucc_geth driver I found that if I:
- Move TX from IRQ to NAPI context
- double the weight.
- after booting up, wait a few mins until the JFFS2 GC kernel

thread

Quoted 12 lines
Quoted 12 lines

has

Quoted 10 lines
Quoted 5 lines

stopped
scanning the FS

Then the "nf_conntrack: table full, dropping packet." msgs stops. Does this seem right to you guys?

No. As I said, something seems to be leaking packets. You should be able to confirm that by checking the sk_buff slabs in /proc/slabinfo. If that *doesn't* show any signs of a leak, please run "conntrack -E" to capture the conntrack events before the "table full" message appears and post the output.

Quoted 4 lines
Quoted 4 lines

skbuff does not differ much, but others do

Before ping:
skbuff_fclone_cache 0 0 352 11 1 : tunables 54 27

0

Quoted 2 lines
Quoted 2 lines

: slabdata 0 0 0
skbuff_head_cache 20 20 192 20 1 : tunables 120 60

0

Quoted 2 lines
Quoted 2 lines
: slabdata      1      1      0
  size-64              731    767     64   59    1 : tunables  120 60 

0

Quoted 2 lines
Quoted 2 lines

: slabdata 13 13 0
nf_conntrack 10 19 208 19 1 : tunables 120 60

0

Quoted 4 lines
Quoted 4 lines

: slabdata 1 1 0

During ping:
skbuff_fclone_cache 0 0 352 11 1 : tunables 54 27

0

Quoted 2 lines
Quoted 2 lines

: slabdata 0 0 0
skbuff_head_cache 40 40 192 20 1 : tunables 120 60

0

Quoted 2 lines
Quoted 2 lines

: slabdata 2 2 0
size-64 8909 8909 64 59 1 : tunables 120 60

0

Quoted 2 lines
Quoted 2 lines

: slabdata 151 151 0
nf_conntrack 5111 5111 208 19 1 : tunables 120 60

0

Quoted 6 lines
Quoted 6 lines

: slabdata 269 269 0

This feels more like the freeing of conntrack objects are delayed and builds up when ping flooding.

Don't have "conntrack -E" for my embedded board so that will have to

wait

Quoted 8 lines
Quoted 1 line

a bit longer.

I dont understand how your ping can use so many conntrack entries...

Then, as I said yesterday, I believe you have a RCU delay, because of a misbehaving driver or something...

grep RCU .config

grep RCU .config
# RCU Subsystem
CONFIG_CLASSIC_RCU=y
# CONFIG_TREE_RCU is not set
# CONFIG_PREEMPT_RCU is not set
# CONFIG_TREE_RCU_TRACE is not set
# CONFIG_PREEMPT_RCU_TRACE is not set
# CONFIG_RCU_TORTURE_TEST is not set
# CONFIG_RCU_CPU_STALL_DETECTOR is not set

Quoted 1 line

grep CONFIG_SMP .config

grep CONFIG_SMP .config
# CONFIG_SMP is not set

Quoted 2 lines

You could change qhimark from 10000 to 1000 in kernel/rcuclassic.c (line

80)

Quoted 1 line

as a workaround. It should force a quiescent state after 1000 freed

conntracks.

right, doing this almost killed all conntrack messages, had to stress it pretty
hard before I saw handful "nf_conntrack: table full, dropping packet"

RCU is not my cup of tea, do you have any ideas were to look?

Jocke

#12 lore raw

PATCH 1/1 conntrack: Reduce conntrack count in nf_conntrack_free()

Quoted 91 lines · Joakim Tjernlund

Eric Dumazet <[email protected]> wrote on 24/03/2009 10:12:53:

Quoted 12 lines
Quoted 11 lines · Joakim Tjernlund

Patrick McHardy <[email protected]> wrote on 23/03/2009 18:49:15:

Quoted 9 lines
Quoted 8 lines · Joakim Tjernlund

Patrick McHardy <[email protected]> wrote on 23/03/2009 13:29:33:

Quoted 5 lines
Quoted 4 lines

There is no /proc/net/netfilter/nf_conntrack. There is a /proc/net/nf_conntrack though and it is empty. If I telnet to the board I see:

That means that something is leaking conntrack references, most

likely

Quoted 8 lines
Quoted 8 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line

by leaking skbs. Since I haven't seen any other reports, my guess

would

Quoted 6 lines
Quoted 6 lines
Quoted 2 lines

be the ucc_geth driver.

Mucking around with the ucc_geth driver I found that if I:
- Move TX from IRQ to NAPI context
- double the weight.
- after booting up, wait a few mins until the JFFS2 GC kernel

thread

Quoted 11 lines
Quoted 11 lines

has

Quoted 10 lines
Quoted 5 lines

stopped
scanning the FS

Then the "nf_conntrack: table full, dropping packet." msgs stops. Does this seem right to you guys?

No. As I said, something seems to be leaking packets. You should be able to confirm that by checking the sk_buff slabs in /proc/slabinfo. If that *doesn't* show any signs of a leak, please run "conntrack -E" to capture the conntrack events before the "table full" message appears and post the output.

Quoted 4 lines
Quoted 4 lines

skbuff does not differ much, but others do

Before ping:
skbuff_fclone_cache 0 0 352 11 1 : tunables 54 27

0

Quoted 2 lines
Quoted 2 lines

: slabdata 0 0 0
skbuff_head_cache 20 20 192 20 1 : tunables 120 60

0

Quoted 2 lines
Quoted 2 lines
: slabdata      1      1      0
  size-64              731    767     64   59    1 : tunables  120 60 

0

Quoted 2 lines
Quoted 2 lines

: slabdata 13 13 0
nf_conntrack 10 19 208 19 1 : tunables 120 60

0

Quoted 4 lines
Quoted 4 lines

: slabdata 1 1 0

During ping:
skbuff_fclone_cache 0 0 352 11 1 : tunables 54 27

0

Quoted 2 lines
Quoted 2 lines

: slabdata 0 0 0
skbuff_head_cache 40 40 192 20 1 : tunables 120 60

0

Quoted 2 lines
Quoted 2 lines

: slabdata 2 2 0
size-64 8909 8909 64 59 1 : tunables 120 60

0

Quoted 2 lines
Quoted 2 lines

: slabdata 151 151 0
nf_conntrack 5111 5111 208 19 1 : tunables 120 60

0

Quoted 6 lines
Quoted 6 lines

: slabdata 269 269 0

This feels more like the freeing of conntrack objects are delayed and builds up when ping flooding.

Don't have "conntrack -E" for my embedded board so that will have to

wait

Quoted 7 lines
Quoted 1 line

a bit longer.

I dont understand how your ping can use so many conntrack entries...

Then, as I said yesterday, I believe you have a RCU delay, because of a misbehaving driver or something...

grep RCU .config

grep RCU .config
# RCU Subsystem
CONFIG_CLASSIC_RCU=y
# CONFIG_TREE_RCU is not set
# CONFIG_PREEMPT_RCU is not set
# CONFIG_TREE_RCU_TRACE is not set
# CONFIG_PREEMPT_RCU_TRACE is not set
# CONFIG_RCU_TORTURE_TEST is not set
# CONFIG_RCU_CPU_STALL_DETECTOR is not set

Quoted 1 line

grep CONFIG_SMP .config

grep CONFIG_SMP .config
# CONFIG_SMP is not set

Quoted 1 line

You could change qhimark from 10000 to 1000 in kernel/rcuclassic.c (line

80)

Quoted 8 lines
Quoted 1 line

as a workaround. It should force a quiescent state after 1000 freed

conntracks.

right, doing this almost killed all conntrack messages, had to stress it pretty
hard before I saw handful "nf_conntrack: table full, dropping packet"

RCU is not my cup of tea, do you have any ideas were to look?

In a stress situation, you feed more deleted conntracks to call_rcu() than the blimit (10 real freeing per RCU softirq invocation).

So with default qhimark being 10000, this means about 10000 conntracks can sit in RCU (per CPU) before being really freed.

Only when hitting 10000, RCU enters a special mode to free all queued items, instead of a small batch of 10

To solve your problem we can :

1) reduce qhimark from 10000 to 1000 (for example)
Probably should be done to reduce some spikes in RCU code when freeing whole 10000 elements...
OR
2) change conntrack tunable (max conntrack entries on your machine) OR
3) change net/netfilter/nf_conntrack_core.c to decrement net->ct.count in nf_conntrack_free() instead of callback.

[PATCH] conntrack: Reduce conntrack count in nf_conntrack_free()

We use RCU to defer freeing of conntrack structures. In DOS situation, RCU might accumulate about 10.000 elements per CPU in its internal queues. To get accurate conntrack counts (at the expense of slightly more RAM used), we might consider conntrack counter not taking into account "about to be freed elements, waiting in RCU queues". We thus decrement it in nf_conntrack_free(), not in the RCU callback.

net/netfilter/nf_conntrack_core.c +3 -2
index f4935e3..6478dc7 100644
@@ -516,16 +516,17 @@ EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 static void nf_conntrack_free_rcu(struct rcu_head *head)
 {
 	struct nf_conn *ct = container_of(head, struct nf_conn, rcu);
-	struct net *net = nf_ct_net(ct);
 
 	nf_ct_ext_free(ct);
 	kmem_cache_free(nf_conntrack_cachep, ct);
-	atomic_dec(&net->ct.count);
 }
 
 void nf_conntrack_free(struct nf_conn *ct)
 {
+	struct net *net = nf_ct_net(ct);
+
 	nf_ct_ext_destroy(ct);
+	atomic_dec(&net->ct.count);
 	call_rcu(&ct->rcu, nf_conntrack_free_rcu);
 }
 EXPORT_SYMBOL_GPL(nf_conntrack_free);
#13 lore raw

Re: [PATCH] conntrack: Reduce conntrack count in nf_conntrack_free()

Quoted 147 lines · Eric Dumazet
Quoted 91 lines · Joakim Tjernlund

Eric Dumazet <[email protected]> wrote on 24/03/2009 10:12:53:

Quoted 12 lines
Quoted 11 lines · Joakim Tjernlund

Patrick McHardy <[email protected]> wrote on 23/03/2009 18:49:15:

Quoted 9 lines
Quoted 8 lines · Joakim Tjernlund

Patrick McHardy <[email protected]> wrote on 23/03/2009 13:29:33:

Quoted 5 lines
Quoted 4 lines

There is no /proc/net/netfilter/nf_conntrack. There is a /proc/net/nf_conntrack though and it is empty. If I telnet to the board I see:

That means that something is leaking conntrack references, most

likely

Quoted 8 lines
Quoted 8 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line

by leaking skbs. Since I haven't seen any other reports, my guess

would

Quoted 6 lines
Quoted 6 lines
Quoted 2 lines

be the ucc_geth driver.

Mucking around with the ucc_geth driver I found that if I:
- Move TX from IRQ to NAPI context
- double the weight.
- after booting up, wait a few mins until the JFFS2 GC kernel

thread

Quoted 11 lines
Quoted 11 lines

has

Quoted 10 lines
Quoted 5 lines

stopped
scanning the FS

Then the "nf_conntrack: table full, dropping packet." msgs stops. Does this seem right to you guys?

No. As I said, something seems to be leaking packets. You should be able to confirm that by checking the sk_buff slabs in /proc/slabinfo. If that *doesn't* show any signs of a leak, please run "conntrack -E" to capture the conntrack events before the "table full" message appears and post the output.

Quoted 4 lines
Quoted 4 lines

skbuff does not differ much, but others do

Before ping:
skbuff_fclone_cache 0 0 352 11 1 : tunables 54 27

0

Quoted 2 lines
Quoted 2 lines

: slabdata 0 0 0
skbuff_head_cache 20 20 192 20 1 : tunables 120 60

0

Quoted 2 lines
Quoted 2 lines
: slabdata      1      1      0
  size-64              731    767     64   59    1 : tunables  120 60 

0

Quoted 2 lines
Quoted 2 lines

: slabdata 13 13 0
nf_conntrack 10 19 208 19 1 : tunables 120 60

0

Quoted 4 lines
Quoted 4 lines

: slabdata 1 1 0

During ping:
skbuff_fclone_cache 0 0 352 11 1 : tunables 54 27

0

Quoted 2 lines
Quoted 2 lines

: slabdata 0 0 0
skbuff_head_cache 40 40 192 20 1 : tunables 120 60

0

Quoted 2 lines
Quoted 2 lines

: slabdata 2 2 0
size-64 8909 8909 64 59 1 : tunables 120 60

0

Quoted 2 lines
Quoted 2 lines

: slabdata 151 151 0
nf_conntrack 5111 5111 208 19 1 : tunables 120 60

0

Quoted 6 lines
Quoted 6 lines

: slabdata 269 269 0

This feels more like the freeing of conntrack objects are delayed and builds up when ping flooding.

Don't have "conntrack -E" for my embedded board so that will have to

wait

Quoted 7 lines
Quoted 1 line

a bit longer.

I dont understand how your ping can use so many conntrack entries...

Then, as I said yesterday, I believe you have a RCU delay, because of a misbehaving driver or something...

grep RCU .config

grep RCU .config
# RCU Subsystem
CONFIG_CLASSIC_RCU=y
# CONFIG_TREE_RCU is not set
# CONFIG_PREEMPT_RCU is not set
# CONFIG_TREE_RCU_TRACE is not set
# CONFIG_PREEMPT_RCU_TRACE is not set
# CONFIG_RCU_TORTURE_TEST is not set
# CONFIG_RCU_CPU_STALL_DETECTOR is not set

Quoted 1 line

grep CONFIG_SMP .config

grep CONFIG_SMP .config
# CONFIG_SMP is not set

Quoted 1 line

You could change qhimark from 10000 to 1000 in kernel/rcuclassic.c (line

80)

Quoted 8 lines
Quoted 1 line

as a workaround. It should force a quiescent state after 1000 freed

conntracks.

right, doing this almost killed all conntrack messages, had to stress it pretty
hard before I saw handful "nf_conntrack: table full, dropping packet"

RCU is not my cup of tea, do you have any ideas were to look?

In a stress situation, you feed more deleted conntracks to call_rcu() than the blimit (10 real freeing per RCU softirq invocation).

So with default qhimark being 10000, this means about 10000 conntracks can sit in RCU (per CPU) before being really freed.

Only when hitting 10000, RCU enters a special mode to free all queued items, instead of a small batch of 10

To solve your problem we can :

1) reduce qhimark from 10000 to 1000 (for example)
Probably should be done to reduce some spikes in RCU code when freeing whole 10000 elements...
OR
2) change conntrack tunable (max conntrack entries on your machine) OR
3) change net/netfilter/nf_conntrack_core.c to decrement net->ct.count in nf_conntrack_free() instead of callback.

[PATCH] conntrack: Reduce conntrack count in nf_conntrack_free()

We use RCU to defer freeing of conntrack structures. In DOS situation, RCU might accumulate about 10.000 elements per CPU in its internal queues. To get accurate conntrack counts (at the expense of slightly more RAM used), we might consider conntrack counter not taking into account "about to be freed elements, waiting in RCU queues". We thus decrement it in nf_conntrack_free(), not in the RCU callback.

net/netfilter/nf_conntrack_core.c +0 -2
index f4935e3..6478dc7 100644
@@ -516,16 +516,17 @@ EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 static void nf_conntrack_free_rcu(struct rcu_head *head)
 {
 	struct nf_conn *ct = container_of(head, struct nf_conn, rcu);
-	struct net *net = nf_ct_net(ct);
 
 	nf_ct_ext_free(ct);
 	kmem_cache_free(nf_conntrack_cachep, ct);
-	atomic_dec(&net->ct.count);
 }

net/netfilter/nf_conntrack_core.c EXPORT_SYMBOL_GPL(nf_conntrack_alloc); · line 532

Quoted 10 lines
net/netfilter/nf_conntrack_core.c +3 -0
 
 void nf_conntrack_free(struct nf_conn *ct)
 {
+	struct net *net = nf_ct_net(ct);
+
 	nf_ct_ext_destroy(ct);
+	atomic_dec(&net->ct.count);
 	call_rcu(&ct->rcu, nf_conntrack_free_rcu);
 }
 EXPORT_SYMBOL_GPL(nf_conntrack_free);

I forgot to say this is what we do for 'struct file' freeing as well. We decrement nr_files in file_free(), not in file_free_rcu()

static inline void file_free_rcu(struct rcu_head *head)

{
        struct file *f = container_of(head, struct file, f_u.fu_rcuhead);

        put_cred(f->f_cred);
        kmem_cache_free(filp_cachep, f);
}

static inline void file_free(struct file *f)

{
        percpu_counter_dec(&nr_files);      <<<< HERE >>>>
        file_check_state(f);
        call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
}
#14 lore raw

Re: [PATCH] conntrack: Reduce conntrack count in nf_conntrack_free()

Quoted 47 lines · Eric Dumazet
Quoted 47 lines

In a stress situation, you feed more deleted conntracks to call_rcu() than the blimit (10 real freeing per RCU softirq invocation).

So with default qhimark being 10000, this means about 10000 conntracks can sit in RCU (per CPU) before being really freed.

Only when hitting 10000, RCU enters a special mode to free all queued items, instead of a small batch of 10

To solve your problem we can :

1) reduce qhimark from 10000 to 1000 (for example)
Probably should be done to reduce some spikes in RCU code when freeing whole 10000 elements...
OR
2) change conntrack tunable (max conntrack entries on your machine) OR
3) change net/netfilter/nf_conntrack_core.c to decrement net->ct.count in nf_conntrack_free() instead of callback.

[PATCH] conntrack: Reduce conntrack count in nf_conntrack_free()

We use RCU to defer freeing of conntrack structures. In DOS situation, RCU might accumulate about 10.000 elements per CPU in its internal queues. To get accurate conntrack counts (at the expense of slightly more RAM used), we might consider conntrack counter not taking into account "about to be freed elements, waiting in RCU queues". We thus decrement it in nf_conntrack_free(), not in the RCU callback.

net/netfilter/nf_conntrack_core.c +0 -2
index f4935e3..6478dc7 100644
@@ -516,16 +516,17 @@ EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 static void nf_conntrack_free_rcu(struct rcu_head *head)
 {
 	struct nf_conn *ct = container_of(head, struct nf_conn, rcu);
-	struct net *net = nf_ct_net(ct);
 
 	nf_ct_ext_free(ct);
 	kmem_cache_free(nf_conntrack_cachep, ct);
-	atomic_dec(&net->ct.count);
 }
 
Quoted 12 lines

net/netfilter/nf_conntrack_core.c EXPORT_SYMBOL_GPL(nf_conntrack_alloc); · line 532

Quoted 9 lines
net/netfilter/nf_conntrack_core.c +3 -0
 void nf_conntrack_free(struct nf_conn *ct)
 {
+	struct net *net = nf_ct_net(ct);
+
 	nf_ct_ext_destroy(ct);
+	atomic_dec(&net->ct.count);
 	call_rcu(&ct->rcu, nf_conntrack_free_rcu);
 }
 EXPORT_SYMBOL_GPL(nf_conntrack_free);

I forgot to say this is what we do for 'struct file' freeing as well. We decrement nr_files in file_free(), not in file_free_rcu()

While temporarily exceeding the limit by up to 10000 entries is quite a lot, I guess the important thing is that it can't grow unbounded, so I think this patch is fine.

#15 lore raw

Re: [PATCH] conntrack: Reduce conntrack count in nf_conntrack_free()

Quoted 70 lines · Patrick McHardy
Quoted 55 lines · Eric Dumazet
Quoted 55 lines

In a stress situation, you feed more deleted conntracks to call_rcu() than
the blimit (10 real freeing per RCU softirq invocation).
So with default qhimark being 10000, this means about 10000 conntracks can sit in RCU (per CPU) before being really freed.

Only when hitting 10000, RCU enters a special mode to free all queued items, instead
of a small batch of 10

To solve your problem we can :

1) reduce qhimark from 10000 to 1000 (for example)
Probably should be done to reduce some spikes in RCU code when
freeing
whole 10000 elements...
OR
2) change conntrack tunable (max conntrack entries on your machine) OR
3) change net/netfilter/nf_conntrack_core.c to decrement net->ct.count in nf_conntrack_free() instead of callback.

[PATCH] conntrack: Reduce conntrack count in nf_conntrack_free()

We use RCU to defer freeing of conntrack structures. In DOS situation, RCU might
accumulate about 10.000 elements per CPU in its internal queues. To get accurate
conntrack counts (at the expense of slightly more RAM used), we might consider
conntrack counter not taking into account "about to be freed elements, waiting
in RCU queues". We thus decrement it in nf_conntrack_free(), not in the RCU
callback.

net/netfilter/nf_conntrack_core.c +0 -2
index f4935e3..6478dc7 100644
@@ -516,16 +516,17 @@ EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 static void nf_conntrack_free_rcu(struct rcu_head *head)
 {
     struct nf_conn *ct = container_of(head, struct nf_conn, rcu);
-    struct net *net = nf_ct_net(ct);
 
     nf_ct_ext_free(ct);
     kmem_cache_free(nf_conntrack_cachep, ct);
-    atomic_dec(&net->ct.count);
 }
 
Quoted 12 lines

net/netfilter/nf_conntrack_core.c EXPORT_SYMBOL_GPL(nf_conntrack_alloc); · line 532

Quoted 9 lines
net/netfilter/nf_conntrack_core.c +3 -0
 void nf_conntrack_free(struct nf_conn *ct)
 {
+    struct net *net = nf_ct_net(ct);
+
     nf_ct_ext_destroy(ct);
+    atomic_dec(&net->ct.count);
     call_rcu(&ct->rcu, nf_conntrack_free_rcu);
 }
 EXPORT_SYMBOL_GPL(nf_conntrack_free);

I forgot to say this is what we do for 'struct file' freeing as well. We decrement nr_files in file_free(), not in file_free_rcu()

Quoted 5 lines

While temporarily exceeding the limit by up to 10000 entries is quite a lot, I guess the important thing is that it can't grow unbounded, so I think this patch is fine.

Maybe we could use SLAB_DESTROY_BY_RCU thing and no more call_rcu() queueing problem. That would better use CPU caches as well...

#16 lore raw

Re: [PATCH] conntrack: Reduce conntrack count in nf_conntrack_free()

Quoted 11 lines · Eric Dumazet
Quoted 7 lines · Patrick McHardy
Quoted 2 lines

I forgot to say this is what we do for 'struct file' freeing as well. We decrement nr_files in file_free(), not in file_free_rcu()

While temporarily exceeding the limit by up to 10000 entries is quite a lot, I guess the important thing is that it can't grow unbounded, so I think this patch is fine.

Maybe we could use SLAB_DESTROY_BY_RCU thing and no more call_rcu() queueing problem. That would better use CPU caches as well...

I'm not sure I understand the rules correctly, but we'd still have to wait for the grace period before an object can be reused, no?

#17 lore raw

Re: [PATCH] conntrack: Reduce conntrack count in nf_conntrack_free()

Quoted 11 lines · Patrick McHardy
Quoted 10 lines · Eric Dumazet
Quoted 8 lines · Patrick McHardy
Quoted 3 lines

I forgot to say this is what we do for 'struct file' freeing as well. We
decrement nr_files in file_free(), not in file_free_rcu()

While temporarily exceeding the limit by up to 10000 entries is quite a lot, I guess the important thing is that it can't grow unbounded, so I think this patch is fine.

Quoted 7 lines
Quoted 3 lines

Maybe we could use SLAB_DESTROY_BY_RCU thing and no more call_rcu() queueing
problem. That would better use CPU caches as well...

I'm not sure I understand the rules correctly, but we'd still have to wait for the grace period before an object can be reused, no?

No we dont have to, but we must do additionnal checks after getting a reference on object found on lookup.
(We must re-check the keys used during search)

This re-check is not very expensive since everything is hot in cpu cache.

Check Documentation/RCU/rculist_nulls.txt for some documentation.

1) Lookup algo
--------------

rcu_read_lock()
begin:
obj = lockless_lookup(key);
if (obj) {
if (!try_get_ref(obj)) // might fail for free objects

    goto begin;
  /*
   * Because a writer could delete object, and a writer could
   * reuse these object before the RCU grace period, we
   * must check key after geting the reference on object
   */
  if (obj->key != key) { // not the object we expected
     put_ref(obj);
     goto begin;

}
}
rcu_read_unlock();

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): [email protected]

#20 lore raw

Re: [PATCH] conntrack: Reduce conntrack count in nf_conntrack_free()

Quoted 18 lines · Patrick McHardy
Quoted 17 lines · Eric Dumazet
Quoted 16 lines · Eric Dumazet
Quoted 8 lines · Patrick McHardy
Quoted 3 lines · Eric Dumazet

Maybe we could use SLAB_DESTROY_BY_RCU thing and no more call_rcu() queueing
problem. That would better use CPU caches as well...

I'm not sure I understand the rules correctly, but we'd still have to wait for the grace period before an object can be reused, no?

No we dont have to, but we must do additionnal checks after getting a reference on object found on lookup.
(We must re-check the keys used during search)

This re-check is not very expensive since everything is hot in cpu cache.

Quoted 10 lines
Quoted 5 lines
Quoted 2 lines

Check Documentation/RCU/rculist_nulls.txt for some documentation.

Patrick, I can work on this if you want, since this stuff is fresh in my head, let me know if you already challenged it or not :)

I'm still looking at the details, if you want to take care of this, that would be great .)

I have a litle problem on __nf_conntrack_find() being exported.

Problem is that with SLAB_DESTROY_BY_RCU we must take a reference on object to recheck it. So ideally only nf_conntrack_find_get() should be used, or callers of __nf_conntrack_find() should lock nf_conntrack_lock (as properly done for example in net/netfilter/nf_conntrack_netlink.c, line 1292)

Here is preliminary patch for review (not tested at all, its 4h50 am here :) )

Could you help me, by checking __nf_conntrack_find() use in net/netfilter/xt_connlimit.c ? and line 1246 of net/netfilter/nf_conntrack_netlink.c

This part is a litle bit gray for me. :)

Thank you

(Patch against nf-next-2.6 tree of course)

Eric

 include/net/netfilter/nf_conntrack.h                  |   14 -
 include/net/netfilter/nf_conntrack_tuple.h            |    6
 include/net/netns/conntrack.h                         |    5
 net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c |   16 -
 net/ipv4/netfilter/nf_nat_core.c                      |    2
 net/netfilter/nf_conntrack_core.c                     |  121 +++++-----
 net/netfilter/nf_conntrack_expect.c                   |    2
 net/netfilter/nf_conntrack_helper.c                   |    7
 net/netfilter/nf_conntrack_netlink.c                  |    7
 net/netfilter/nf_conntrack_standalone.c               |   16 -
 10 files changed, 109 insertions(+), 87 deletions(-)
include/net/netfilter/nf_conntrack.h +8 -6
index 4dfb793..6c3f964 100644
@@ -91,8 +91,7 @@ struct nf_conn_help {
 #include <net/netfilter/ipv4/nf_conntrack_ipv4.h>
 #include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
 
-struct nf_conn
-{
+struct nf_conn {
 	/* Usage count in here is 1 for hash table/destruct timer, 1 per skb,
            plus 1 for any connection(s) we are `master' for */
 	struct nf_conntrack ct_general;
@@ -126,7 +125,6 @@ struct nf_conn
 #ifdef CONFIG_NET_NS
 	struct net *ct_net;
 #endif
-	struct rcu_head rcu;
 };
 
 static inline struct nf_conn *
@@ -190,9 +188,13 @@ static inline void nf_ct_put(struct nf_conn *ct)
 extern int nf_ct_l3proto_try_module_get(unsigned short l3proto);
 extern void nf_ct_l3proto_module_put(unsigned short l3proto);
 
-extern struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced);
-extern void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced,
-				 unsigned int size);
+/*
+ * Allocate a hashtable of hlist_head (if nulls == 0),
+ * or hlist_nulls_head (if nulls == 1)
+ */
+extern void *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced, int nulls);
+
+extern void nf_ct_free_hashtable(void *hash, int vmalloced, unsigned int size);
 
 extern struct nf_conntrack_tuple_hash *
 __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple);
include/net/netfilter/nf_conntrack_tuple.h +3 -3
index f2f6aa7..2628c15 100644
@@ -12,6 +12,7 @@
 
 #include <linux/netfilter/x_tables.h>
 #include <linux/netfilter/nf_conntrack_tuple_common.h>
+#include <linux/list_nulls.h>
 
 /* A `tuple' is a structure containing the information to uniquely
   identify a connection.  ie. if two packets have the same tuple, they
@@ -146,9 +147,8 @@ static inline void nf_ct_dump_tuple(const struct nf_conntrack_tuple *t)
 	((enum ip_conntrack_dir)(h)->tuple.dst.dir)
 
 /* Connections have two entries in the hash table: one for each way */
-struct nf_conntrack_tuple_hash
-{
-	struct hlist_node hnode;
+struct nf_conntrack_tuple_hash {
+	struct hlist_nulls_node hnnode;
 	struct nf_conntrack_tuple tuple;
 };
 
include/net/netns/conntrack.h +3 -2
index f4498a6..9dc5840 100644
@@ -2,6 +2,7 @@
 #define __NETNS_CONNTRACK_H
 
 #include <linux/list.h>
+#include <linux/list_nulls.h>
 #include <asm/atomic.h>
 
 struct ctl_table_header;
@@ -10,9 +11,9 @@ struct nf_conntrack_ecache;
 struct netns_ct {
 	atomic_t		count;
 	unsigned int		expect_count;
-	struct hlist_head	*hash;
+	struct hlist_nulls_head	*hash;
 	struct hlist_head	*expect_hash;
-	struct hlist_head	unconfirmed;
+	struct hlist_nulls_head	unconfirmed;
 	struct ip_conntrack_stat *stat;
 #ifdef CONFIG_NF_CONNTRACK_EVENTS
 	struct nf_conntrack_ecache *ecache;
net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c +8 -8
index 6ba5c55..fcbcf62 100644
@@ -25,30 +25,30 @@ struct ct_iter_state {
 	unsigned int bucket;
 };
 
-static struct hlist_node *ct_get_first(struct seq_file *seq)
+static struct hlist_nulls_node *ct_get_first(struct seq_file *seq)
 {
 	struct net *net = seq_file_net(seq);
 	struct ct_iter_state *st = seq->private;
-	struct hlist_node *n;
+	struct hlist_nulls_node *n;
 
 	for (st->bucket = 0;
 	     st->bucket < nf_conntrack_htable_size;
 	     st->bucket++) {
 		n = rcu_dereference(net->ct.hash[st->bucket].first);
-		if (n)
+		if (!is_a_nulls(n))
 			return n;
 	}
 	return NULL;
 }
 
-static struct hlist_node *ct_get_next(struct seq_file *seq,
-				      struct hlist_node *head)
+static struct hlist_nulls_node *ct_get_next(struct seq_file *seq,
+				      struct hlist_nulls_node *head)
 {
 	struct net *net = seq_file_net(seq);
 	struct ct_iter_state *st = seq->private;
 
 	head = rcu_dereference(head->next);
-	while (head == NULL) {
+	while (is_a_nulls(head)) {
 		if (++st->bucket >= nf_conntrack_htable_size)
 			return NULL;
 		head = rcu_dereference(net->ct.hash[st->bucket].first);
@@ -56,9 +56,9 @@ static struct hlist_node *ct_get_next(struct seq_file *seq,
 	return head;
 }
 
-static struct hlist_node *ct_get_idx(struct seq_file *seq, loff_t pos)
+static struct hlist_nulls_node *ct_get_idx(struct seq_file *seq, loff_t pos)
 {
-	struct hlist_node *head = ct_get_first(seq);
+	struct hlist_nulls_node *head = ct_get_first(seq);
 
 	if (head)
 		while (pos && (head = ct_get_next(seq, head)))
net/ipv4/netfilter/nf_nat_core.c +1 -1
index a65cf69..fe65187 100644
@@ -679,7 +679,7 @@ nfnetlink_parse_nat_setup(struct nf_conn *ct,
 static int __net_init nf_nat_net_init(struct net *net)
 {
 	net->ipv4.nat_bysource = nf_ct_alloc_hashtable(&nf_nat_htable_size,
-						      &net->ipv4.nat_vmalloced);
+						      &net->ipv4.nat_vmalloced, 0);
 	if (!net->ipv4.nat_bysource)
 		return -ENOMEM;
 	return 0;
net/netfilter/nf_conntrack_core.c +69 -52

311 lines not on this page · raw message

net/netfilter/nf_conntrack_expect.c +1 -1
index 357ba39..3940f99 100644
@@ -604,7 +604,7 @@ int nf_conntrack_expect_init(struct net *net)
 
 	net->ct.expect_count = 0;
 	net->ct.expect_hash = nf_ct_alloc_hashtable(&nf_ct_expect_hsize,
-						  &net->ct.expect_vmalloc);
+						  &net->ct.expect_vmalloc, 0);
 	if (net->ct.expect_hash == NULL)
 		goto err1;
 
net/netfilter/nf_conntrack_helper.c +4 -3
index a51bdac..6066144 100644
@@ -158,6 +158,7 @@ static void __nf_conntrack_helper_unregister(struct nf_conntrack_helper *me,
 	struct nf_conntrack_tuple_hash *h;
 	struct nf_conntrack_expect *exp;
 	const struct hlist_node *n, *next;
+	const struct hlist_nulls_node *nn;
 	unsigned int i;
 
 	/* Get rid of expectations */
@@ -174,10 +175,10 @@ static void __nf_conntrack_helper_unregister(struct nf_conntrack_helper *me,
 	}
 
 	/* Get rid of expecteds, set helpers to NULL. */
-	hlist_for_each_entry(h, n, &net->ct.unconfirmed, hnode)
+	hlist_for_each_entry(h, nn, &net->ct.unconfirmed, hnnode)
 		unhelp(h, me);
 	for (i = 0; i < nf_conntrack_htable_size; i++) {
-		hlist_for_each_entry(h, n, &net->ct.hash[i], hnode)
+		hlist_nulls_for_each_entry(h, nn, &net->ct.hash[i], hnnode)
 			unhelp(h, me);
 	}
 }
@@ -217,7 +218,7 @@ int nf_conntrack_helper_init(void)
 
 	nf_ct_helper_hsize = 1; /* gets rounded up to use one page */
 	nf_ct_helper_hash = nf_ct_alloc_hashtable(&nf_ct_helper_hsize,
-						  &nf_ct_helper_vmalloc);
+						  &nf_ct_helper_vmalloc, 0);
 	if (!nf_ct_helper_hash)
 		return -ENOMEM;
 
net/netfilter/nf_conntrack_netlink.c +4 -3
index 1b75c9e..6520c94 100644
@@ -19,6 +19,7 @@
 #include <linux/module.h>
 #include <linux/kernel.h>
 #include <linux/rculist.h>
+#include <linux/rculist_nulls.h>
 #include <linux/types.h>
 #include <linux/timer.h>
 #include <linux/skbuff.h>
@@ -536,7 +537,7 @@ ctnetlink_dump_table(struct sk_buff *skb, struct netlink_callback *cb)
 {
 	struct nf_conn *ct, *last;
 	struct nf_conntrack_tuple_hash *h;
-	struct hlist_node *n;
+	struct hlist_nulls_node *n;
 	struct nfgenmsg *nfmsg = NLMSG_DATA(cb->nlh);
 	u_int8_t l3proto = nfmsg->nfgen_family;
 
@@ -544,8 +545,8 @@ ctnetlink_dump_table(struct sk_buff *skb, struct netlink_callback *cb)
 	last = (struct nf_conn *)cb->args[1];
 	for (; cb->args[0] < nf_conntrack_htable_size; cb->args[0]++) {
 restart:
-		hlist_for_each_entry_rcu(h, n, &init_net.ct.hash[cb->args[0]],
-					 hnode) {
+		hlist_nulls_for_each_entry_rcu(h, n, &init_net.ct.hash[cb->args[0]],
+					 hnnode) {
 			if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
 				continue;
 			ct = nf_ct_tuplehash_to_ctrack(h);
net/netfilter/nf_conntrack_standalone.c +8 -8
index 4da54b0..f768368 100644
@@ -44,30 +44,30 @@ struct ct_iter_state {
 	unsigned int bucket;
 };
 
-static struct hlist_node *ct_get_first(struct seq_file *seq)
+static struct hlist_nulls_node *ct_get_first(struct seq_file *seq)
 {
 	struct net *net = seq_file_net(seq);
 	struct ct_iter_state *st = seq->private;
-	struct hlist_node *n;
+	struct hlist_nulls_node *n;
 
 	for (st->bucket = 0;
 	     st->bucket < nf_conntrack_htable_size;
 	     st->bucket++) {
 		n = rcu_dereference(net->ct.hash[st->bucket].first);
-		if (n)
+		if (!is_a_nulls(n))
 			return n;
 	}
 	return NULL;
 }
 
-static struct hlist_node *ct_get_next(struct seq_file *seq,
-				      struct hlist_node *head)
+static struct hlist_nulls_node *ct_get_next(struct seq_file *seq,
+				      struct hlist_nulls_node *head)
 {
 	struct net *net = seq_file_net(seq);
 	struct ct_iter_state *st = seq->private;
 
 	head = rcu_dereference(head->next);
-	while (head == NULL) {
+	while (is_a_nulls(head)) {
 		if (++st->bucket >= nf_conntrack_htable_size)
 			return NULL;
 		head = rcu_dereference(net->ct.hash[st->bucket].first);
@@ -75,9 +75,9 @@ static struct hlist_node *ct_get_next(struct seq_file *seq,
 	return head;
 }
 
-static struct hlist_node *ct_get_idx(struct seq_file *seq, loff_t pos)
+static struct hlist_nulls_node *ct_get_idx(struct seq_file *seq, loff_t pos)
 {
-	struct hlist_node *head = ct_get_first(seq);
+	struct hlist_nulls_node *head = ct_get_first(seq);
 
 	if (head)
 		while (pos && (head = ct_get_next(seq, head)))

--
To unsubscribe from this list: send the line "unsubscribe netfilter-devel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#21 lore raw

Re: [PATCH] conntrack: Reduce conntrack count in nf_conntrack_free()

Quoted 7 lines · Eric Dumazet

I have a litle problem on __nf_conntrack_find() being exported.

Problem is that with SLAB_DESTROY_BY_RCU we must take a reference on object to recheck it. So ideally only nf_conntrack_find_get() should be used, or callers of __nf_conntrack_find() should lock nf_conntrack_lock (as properly done for example in net/netfilter/nf_conntrack_netlink.c, line 1292)

Quoted 6 lines

Here is preliminary patch for review (not tested at all, its 4h50 am here :) )

Could you help me, by checking __nf_conntrack_find() use in net/netfilter/xt_connlimit.c ? and line 1246 of net/netfilter/nf_conntrack_netlink.c

This part is a litle bit gray for me. :)

In case of xt_connlimit, it seems fine to just take a reference. In case of ctnetlink, keeping the unreferenced lookup under the lock seems fine. We unfortunately have to export some internals like nf_conntrack lock for ctnetlink anyways, so I don't think it would be worth to change it to take references and unexport the lookup function.

Quoted 26 lines
net/netfilter/nf_conntrack_core.c +10 -2
+/*
+ * Warning :
+ * - Caller must take a reference on returned object
+ *   and recheck nf_ct_tuple_equal(tuple, &h->tuple)
+ * OR
+ * - Caller must lock nf_conntrack_lock before calling this function
+ */
 struct nf_conntrack_tuple_hash *
 __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple)
 {
 	struct nf_conntrack_tuple_hash *h;
-	struct hlist_node *n;
+	struct hlist_nulls_node *n;
 	unsigned int hash = hash_conntrack(tuple);
 
 	/* Disable BHs the entire time since we normally need to disable them
 	 * at least once for the stats anyway.
 	 */
 	local_bh_disable();
-	hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnode) {
+begin:
+	hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
 		if (nf_ct_tuple_equal(tuple, &h->tuple)) {
 			NF_CT_STAT_INC(net, found);
 			local_bh_enable();

net/netfilter/nf_conntrack_core.c __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple) · line 279

Quoted 10 lines
net/netfilter/nf_conntrack_core.c +7 -0
@@ -261,6 +270,13 @@ __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple)
 		}
 		NF_CT_STAT_INC(net, searched);
 	}
+	/*
+	 * if the nulls value we got at the end of this lookup is
+	 * not the expected one, we must restart lookup.
+	 * We probably met an item that was moved to another chain.
+	 */
+	if (get_nulls_value(n) != hash)
+		goto begin;

Are you sure this is enough? An entry might have been reused and added to the same chain I think, so I think we need to recheck the tuple.

Quoted 7 lines
net/netfilter/nf_conntrack_core.c +0 -0
 	local_bh_enable();
 
 	return NULL;
@@ -275,11 +291,18 @@ nf_conntrack_find_get(struct net *net, const struct nf_conntrack_tuple *tuple)
 	struct nf_conn *ct;
 
 	rcu_read_lock();

net/netfilter/nf_conntrack_core.c nf_conntrack_find_get(struct net *net, const struct nf_conntrack_tuple *tuple) · line 303

Quoted 10 lines
net/netfilter/nf_conntrack_core.c +5 -0
+begin:
 	h = __nf_conntrack_find(net, tuple);
 	if (h) {
 		ct = nf_ct_tuplehash_to_ctrack(h);
 		if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
 			h = NULL;
+		else {
+			if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple))) {
+				nf_ct_put(ct);
+				goto begin;

Ah I see, the hash comparison above is only an optimization?

Quoted 5 lines
net/netfilter/nf_conntrack_core.c +2 -0
+			}
+		}
 	}
 	rcu_read_unlock();
 
#22 lore raw

Re: [PATCH] conntrack: Reduce conntrack count in nf_conntrack_free()

Quoted 60 lines · Patrick McHardy
Quoted 9 lines · Eric Dumazet

I have a litle problem on __nf_conntrack_find() being exported.

Problem is that with SLAB_DESTROY_BY_RCU we must take a reference on object
to recheck it. So ideally only nf_conntrack_find_get() should be used, or callers of __nf_conntrack_find() should lock nf_conntrack_lock (as properly done for example in net/netfilter/nf_conntrack_netlink.c, line 1292)

Quoted 8 lines

Here is preliminary patch for review (not tested at all, its 4h50 am here :) )

Could you help me, by checking __nf_conntrack_find() use in net/netfilter/xt_connlimit.c ?
and line 1246 of net/netfilter/nf_conntrack_netlink.c

This part is a litle bit gray for me. :)

In case of xt_connlimit, it seems fine to just take a reference. In case of ctnetlink, keeping the unreferenced lookup under the lock seems fine. We unfortunately have to export some internals like nf_conntrack lock for ctnetlink anyways, so I don't think it would be worth to change it to take references and unexport the lookup function.

Quoted 34 lines

+/*
+ * Warning :
+ * - Caller must take a reference on returned object
+ * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
+ * OR
+ * - Caller must lock nf_conntrack_lock before calling this function
+ */
struct nf_conntrack_tuple_hash * __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple
*tuple)

 {
     struct nf_conntrack_tuple_hash *h;
net/netfilter/nf_conntrack_core.c +1 -1
-    struct hlist_node *n;
+    struct hlist_nulls_node *n;
     unsigned int hash = hash_conntrack(tuple);
 
     /* Disable BHs the entire time since we normally need to disable

them

      * at least once for the stats anyway.
      */
     local_bh_disable();
net/netfilter/nf_conntrack_core.c +4 -1
-    hlist_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnode) {
+begin:
+    hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
         if (nf_ct_tuple_equal(tuple, &h->tuple)) {
             NF_CT_STAT_INC(net, found);
             local_bh_enable();
@@ -261,6 +270,13 @@ __nf_conntrack_find(struct net *net, const struct
 nf_conntrack_tuple *tuple)
         }
         NF_CT_STAT_INC(net, searched);
     }
+    /*
+     * if the nulls value we got at the end of this lookup is
Quoted 8 lines

net/netfilter/nf_conntrack_core.c __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple) · line 279

Quoted 5 lines
net/netfilter/nf_conntrack_core.c +5 -0
+     * not the expected one, we must restart lookup.
+     * We probably met an item that was moved to another chain.
+     */
+    if (get_nulls_value(n) != hash)
+        goto begin;

Are you sure this is enough? An entry might have been reused and added to the same chain I think, so I think we need to recheck the tuple.

Yes, done in caller

Quoted 8 lines
Quoted 8 lines
net/netfilter/nf_conntrack_core.c +0 -0
     local_bh_enable();
 
     return NULL;
net/netfilter/nf_conntrack_core.c +0 -0
@@ -275,11 +291,18 @@ nf_conntrack_find_get(struct net *net, const
 struct nf_conntrack_tuple *tuple)
     struct nf_conn *ct;
 
Quoted 21 lines

net/netfilter/nf_conntrack_core.c nf_conntrack_find_get(struct net *net, const struct nf_conntrack_tuple *tuple) · line 303

Quoted 11 lines
net/netfilter/nf_conntrack_core.c +5 -0
     rcu_read_lock();
+begin:
     h = __nf_conntrack_find(net, tuple);
     if (h) {
         ct = nf_ct_tuplehash_to_ctrack(h);
         if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
             h = NULL;
+        else {
+            if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple))) {
+                nf_ct_put(ct);
+                goto begin;

Ah I see, the hash comparison above is only an optimization?

Quoted 5 lines
net/netfilter/nf_conntrack_core.c +2 -0
+            }
+        }
     }
     rcu_read_unlock();

check net/ipv4/udp.c for an example (__udp4_lib_lookup())

In case of UDP, key check is not returning true/false, but a score. So UDP case is a litle bit more complex than conntrack case.

        rcu_read_lock();
begin:
        result = NULL;
        badness = -1;
        sk_nulls_for_each_rcu(sk, node, &hslot->head) {
                score = compute_score(sk, net, saddr, hnum, sport,
                                      daddr, dport, dif);
                if (score > badness) {
                        result = sk;
                        badness = score;
                }
        }
        /*
         * if the nulls value we got at the end of this lookup is
         * not the expected one, we must restart lookup.
         * We probably met an item that was moved to another chain.
         */
        if (get_nulls_value(node) != hash)
                goto begin;

        if (result) {
                if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
                        result = NULL;
                else if (unlikely(compute_score(result, net, saddr, hnum, sport,
                                  daddr, dport, dif) < badness)) {
                        sock_put(result);
                        goto begin;
                }
        }
        rcu_read_unlock();
        return result;
#23 lore raw

Re: [PATCH] conntrack: Reduce conntrack count in nf_conntrack_free()

In reply to Eric Dumazet

Eric Dumazet <[email protected]> wrote on 24/03/2009 13:07:16:

Quoted 25 lines
Quoted 23 lines · Joakim Tjernlund

Eric Dumazet <[email protected]> wrote on 24/03/2009 10:12:53:

Quoted 14 lines
Quoted 12 lines · Joakim Tjernlund

Patrick McHardy <[email protected]> wrote on 23/03/2009 18:49:15:

Quoted 9 lines
Quoted 8 lines · Joakim Tjernlund

Patrick McHardy <[email protected]> wrote on 23/03/2009 13:29:33:

Quoted 5 lines
Quoted 4 lines

There is no /proc/net/netfilter/nf_conntrack. There is a /proc/net/nf_conntrack though and it is empty. If I telnet to the board I see:

That means that something is leaking conntrack references, most

likely

Quoted 1 line
Quoted 1 line
Quoted 1 line

by leaking skbs. Since I haven't seen any other reports, my guess

Quoted 7 lines
Quoted 7 lines

would

Quoted 6 lines
Quoted 6 lines
Quoted 2 lines

be the ucc_geth driver.

Mucking around with the ucc_geth driver I found that if I:
- Move TX from IRQ to NAPI context
- double the weight.
- after booting up, wait a few mins until the JFFS2 GC kernel

thread

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

has

Quoted 7 lines
Quoted 5 lines

stopped
scanning the FS

Then the "nf_conntrack: table full, dropping packet." msgs stops. Does this seem right to you guys?

No. As I said, something seems to be leaking packets. You should be able to confirm that by checking the sk_buff slabs in

/proc/slabinfo.

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

If that *doesn't* show any signs of a leak, please run "conntrack

-E"

Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 2 lines

to capture the conntrack events before the "table full" message appears and post the output.

skbuff does not differ much, but others do

Before ping:
skbuff_fclone_cache 0 0 352 11 1 : tunables 54

27

Quoted 3 lines
Quoted 3 lines

0

Quoted 2 lines
Quoted 2 lines

: slabdata 0 0 0
skbuff_head_cache 20 20 192 20 1 : tunables 120

60

Quoted 3 lines
Quoted 3 lines

0

Quoted 2 lines
Quoted 2 lines
: slabdata      1      1      0
  size-64              731    767     64   59    1 : tunables  120 

60

Quoted 3 lines
Quoted 3 lines

0

Quoted 2 lines
Quoted 2 lines

: slabdata 13 13 0
nf_conntrack 10 19 208 19 1 : tunables 120

60

Quoted 5 lines
Quoted 5 lines

0

Quoted 4 lines
Quoted 4 lines

: slabdata 1 1 0

During ping:
skbuff_fclone_cache 0 0 352 11 1 : tunables 54

27

Quoted 3 lines
Quoted 3 lines

0

Quoted 2 lines
Quoted 2 lines

: slabdata 0 0 0
skbuff_head_cache 40 40 192 20 1 : tunables 120

60

Quoted 3 lines
Quoted 3 lines

0

Quoted 2 lines
Quoted 2 lines

: slabdata 2 2 0
size-64 8909 8909 64 59 1 : tunables 120

60

Quoted 3 lines
Quoted 3 lines

0

Quoted 2 lines
Quoted 2 lines

: slabdata 151 151 0
nf_conntrack 5111 5111 208 19 1 : tunables 120

60

Quoted 4 lines
Quoted 4 lines

0

Quoted 3 lines
Quoted 3 lines

: slabdata 269 269 0

This feels more like the freeing of conntrack objects are delayed

and

Quoted 22 lines
Quoted 22 lines
Quoted 3 lines
Quoted 3 lines

builds up when ping flooding.

Don't have "conntrack -E" for my embedded board so that will have to

wait

Quoted 7 lines
Quoted 1 line

a bit longer.

I dont understand how your ping can use so many conntrack entries...

Then, as I said yesterday, I believe you have a RCU delay, because of a misbehaving driver or something...

grep RCU .config

grep RCU .config
# RCU Subsystem
CONFIG_CLASSIC_RCU=y
# CONFIG_TREE_RCU is not set
# CONFIG_PREEMPT_RCU is not set
# CONFIG_TREE_RCU_TRACE is not set
# CONFIG_PREEMPT_RCU_TRACE is not set
# CONFIG_RCU_TORTURE_TEST is not set
# CONFIG_RCU_CPU_STALL_DETECTOR is not set

Quoted 5 lines
Quoted 5 lines
Quoted 1 line

grep CONFIG_SMP .config

grep CONFIG_SMP .config
# CONFIG_SMP is not set

Quoted 1 line

You could change qhimark from 10000 to 1000 in kernel/rcuclassic.c

(line

Quoted 5 lines
Quoted 5 lines

80)

Quoted 1 line

as a workaround. It should force a quiescent state after 1000 freed

conntracks.

right, doing this almost killed all conntrack messages, had to stress

it

Quoted 6 lines
Quoted 4 lines

pretty
hard before I saw handful "nf_conntrack: table full, dropping packet"

RCU is not my cup of tea, do you have any ideas were to look?

In a stress situation, you feed more deleted conntracks to call_rcu()

than

Quoted 6 lines

the blimit (10 real freeing per RCU softirq invocation).

So with default qhimark being 10000, this means about 10000 conntracks can sit in RCU (per CPU) before being really freed.

Only when hitting 10000, RCU enters a special mode to free all queued

items, instead

Quoted 6 lines

of a small batch of 10

To solve your problem we can :

1) reduce qhimark from 10000 to 1000 (for example) Probably should be done to reduce some spikes in RCU code when

freeing

Quoted 8 lines
   whole 10000 elements...

OR
2) change conntrack tunable (max conntrack entries on your machine) OR
3) change net/netfilter/nf_conntrack_core.c to decrement net->ct.count in nf_conntrack_free() instead of callback.

[PATCH] conntrack: Reduce conntrack count in nf_conntrack_free()

The patch fixes the problem and the system feels a bit more responsive too, thanks.
I guess I should probably do both 1) and 3) as my board is pretty slow too.

Been trying to figure out a good value for NAPI weigth too. Currently my HW RX and TX queues are 16 pkgs deep and weigth is 16 too. If I move TX processing
to NAPI context AND increase weigth to 32, the system is a lot more responsive during
ping flooding. Does weigth 32 make sense when the HW TX and RX queues are 16?

Jocke

#24 lore raw

Re: [PATCH] conntrack: Reduce conntrack count in nf_conntrack_free()

Quoted 4 lines · Joakim Tjernlund

Eric Dumazet <[email protected]> wrote on 24/03/2009 13:07:16:

Quoted 1 line

[PATCH] conntrack: Reduce conntrack count in nf_conntrack_free()

The patch fixes the problem and the system feels a bit more responsive too, thanks.

Applied, thanks everyone.

#25 lore raw

Re: [PATCH] conntrack: Reduce conntrack count in nf_conntrack_free()

In reply to Joakim Tjernlund
Quoted 146 lines · Joakim Tjernlund

Eric Dumazet <[email protected]> wrote on 24/03/2009 13:07:16:

Quoted 24 lines
Quoted 23 lines · Joakim Tjernlund

Eric Dumazet <[email protected]> wrote on 24/03/2009 10:12:53:

Quoted 14 lines
Quoted 12 lines · Joakim Tjernlund

Patrick McHardy <[email protected]> wrote on 23/03/2009 18:49:15:

Quoted 9 lines
Quoted 8 lines · Joakim Tjernlund

Patrick McHardy <[email protected]> wrote on 23/03/2009 13:29:33:

Quoted 5 lines
Quoted 4 lines

There is no /proc/net/netfilter/nf_conntrack. There is a /proc/net/nf_conntrack though and it is empty. If I telnet to the board I see:

That means that something is leaking conntrack references, most

likely

Quoted 1 line
Quoted 1 line
Quoted 1 line

by leaking skbs. Since I haven't seen any other reports, my guess

Quoted 7 lines
Quoted 7 lines

would

Quoted 6 lines
Quoted 6 lines
Quoted 2 lines

be the ucc_geth driver.

Mucking around with the ucc_geth driver I found that if I:
- Move TX from IRQ to NAPI context
- double the weight.
- after booting up, wait a few mins until the JFFS2 GC kernel

thread

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

has

Quoted 7 lines
Quoted 5 lines

stopped
scanning the FS

Then the "nf_conntrack: table full, dropping packet." msgs stops. Does this seem right to you guys?

No. As I said, something seems to be leaking packets. You should be able to confirm that by checking the sk_buff slabs in

/proc/slabinfo.

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

If that *doesn't* show any signs of a leak, please run "conntrack

-E"

Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 2 lines

to capture the conntrack events before the "table full" message appears and post the output.

skbuff does not differ much, but others do

Before ping:
skbuff_fclone_cache 0 0 352 11 1 : tunables 54

27

Quoted 3 lines
Quoted 3 lines

0

Quoted 2 lines
Quoted 2 lines

: slabdata 0 0 0
skbuff_head_cache 20 20 192 20 1 : tunables 120

60

Quoted 3 lines
Quoted 3 lines

0

Quoted 2 lines
Quoted 2 lines
: slabdata      1      1      0
  size-64              731    767     64   59    1 : tunables  120 

60

Quoted 3 lines
Quoted 3 lines

0

Quoted 2 lines
Quoted 2 lines

: slabdata 13 13 0
nf_conntrack 10 19 208 19 1 : tunables 120

60

Quoted 5 lines
Quoted 5 lines

0

Quoted 4 lines
Quoted 4 lines

: slabdata 1 1 0

During ping:
skbuff_fclone_cache 0 0 352 11 1 : tunables 54

27

Quoted 3 lines
Quoted 3 lines

0

Quoted 2 lines
Quoted 2 lines

: slabdata 0 0 0
skbuff_head_cache 40 40 192 20 1 : tunables 120

60

Quoted 3 lines
Quoted 3 lines

0

Quoted 2 lines
Quoted 2 lines

: slabdata 2 2 0
size-64 8909 8909 64 59 1 : tunables 120

60

Quoted 3 lines
Quoted 3 lines

0

Quoted 2 lines
Quoted 2 lines

: slabdata 151 151 0
nf_conntrack 5111 5111 208 19 1 : tunables 120

60

Quoted 4 lines
Quoted 4 lines

0

Quoted 3 lines
Quoted 3 lines

: slabdata 269 269 0

This feels more like the freeing of conntrack objects are delayed

and

Quoted 22 lines
Quoted 22 lines
Quoted 3 lines
Quoted 3 lines

builds up when ping flooding.

Don't have "conntrack -E" for my embedded board so that will have to

wait

Quoted 7 lines
Quoted 1 line

a bit longer.

I dont understand how your ping can use so many conntrack entries...

Then, as I said yesterday, I believe you have a RCU delay, because of a misbehaving driver or something...

grep RCU .config

grep RCU .config
# RCU Subsystem
CONFIG_CLASSIC_RCU=y
# CONFIG_TREE_RCU is not set
# CONFIG_PREEMPT_RCU is not set
# CONFIG_TREE_RCU_TRACE is not set
# CONFIG_PREEMPT_RCU_TRACE is not set
# CONFIG_RCU_TORTURE_TEST is not set
# CONFIG_RCU_CPU_STALL_DETECTOR is not set

Quoted 5 lines
Quoted 5 lines
Quoted 1 line

grep CONFIG_SMP .config

grep CONFIG_SMP .config
# CONFIG_SMP is not set

Quoted 1 line

You could change qhimark from 10000 to 1000 in kernel/rcuclassic.c

(line

Quoted 5 lines
Quoted 5 lines

80)

Quoted 1 line

as a workaround. It should force a quiescent state after 1000 freed

conntracks.

right, doing this almost killed all conntrack messages, had to stress

it

Quoted 5 lines
Quoted 4 lines

pretty
hard before I saw handful "nf_conntrack: table full, dropping packet"

RCU is not my cup of tea, do you have any ideas were to look?

In a stress situation, you feed more deleted conntracks to call_rcu()

than

Quoted 6 lines

the blimit (10 real freeing per RCU softirq invocation).

So with default qhimark being 10000, this means about 10000 conntracks can sit in RCU (per CPU) before being really freed.

Only when hitting 10000, RCU enters a special mode to free all queued

items, instead

Quoted 6 lines

of a small batch of 10

To solve your problem we can :

1) reduce qhimark from 10000 to 1000 (for example) Probably should be done to reduce some spikes in RCU code when

freeing

Quoted 8 lines
   whole 10000 elements...

OR
2) change conntrack tunable (max conntrack entries on your machine) OR
3) change net/netfilter/nf_conntrack_core.c to decrement net->ct.count in nf_conntrack_free() instead of callback.

[PATCH] conntrack: Reduce conntrack count in nf_conntrack_free()

The patch fixes the problem and the system feels a bit more responsive too, thanks.
I guess I should probably do both 1) and 3) as my board is pretty slow too.

Quoted 7 lines

Been trying to figure out a good value for NAPI weigth too. Currently my HW RX and TX queues are 16 pkgs deep and weigth is 16 too. If I move TX processing
to NAPI context AND increase weigth to 32, the system is a lot more responsive during
ping flooding. Does weigth 32 make sense when the HW TX and RX queues are 16?

If you only have one NIC, I dont understand why changing weight should make a difference. Are you referring to dev_weight or netdev_budget ?

# cat /proc/sys/net/core/dev_weight
64
# cat /proc/sys/net/core/netdev_budget
300
#26 lore raw

Re: [PATCH] conntrack: Reduce conntrack count in nf_conntrack_free()

Eric Dumazet <[email protected]> wrote on 24/03/2009 14:29:29:

[SNIP]

Quoted 8 lines
Quoted 8 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line

You could change qhimark from 10000 to 1000 in kernel/rcuclassic.c

(line

Quoted 6 lines
Quoted 6 lines

80)

Quoted 1 line

as a workaround. It should force a quiescent state after 1000 freed

conntracks.

right, doing this almost killed all conntrack messages, had to

stress

Quoted 3 lines
Quoted 3 lines

it

Quoted 2 lines
Quoted 2 lines

pretty
hard before I saw handful "nf_conntrack: table full, dropping

packet"

Quoted 8 lines
Quoted 8 lines
Quoted 3 lines
Quoted 2 lines

RCU is not my cup of tea, do you have any ideas were to look?

In a stress situation, you feed more deleted conntracks to call_rcu()

than

Quoted 3 lines

the blimit (10 real freeing per RCU softirq invocation).

So with default qhimark being 10000, this means about 10000

conntracks

Quoted 9 lines
Quoted 9 lines
Quoted 3 lines

can sit in RCU (per CPU) before being really freed.

Only when hitting 10000, RCU enters a special mode to free all queued

items, instead

Quoted 4 lines

of a small batch of 10

To solve your problem we can :

Quoted 8 lines
Quoted 8 lines
Quoted 2 lines

1) reduce qhimark from 10000 to 1000 (for example) Probably should be done to reduce some spikes in RCU code when

freeing

Quoted 5 lines
   whole 10000 elements...

OR
2) change conntrack tunable (max conntrack entries on your machine) OR
3) change net/netfilter/nf_conntrack_core.c to decrement

net->ct.count

Quoted 4 lines
Quoted 4 lines
Quoted 3 lines

in nf_conntrack_free() instead of callback.

[PATCH] conntrack: Reduce conntrack count in nf_conntrack_free()

Quoted 8 lines
Quoted 8 lines

The patch fixes the problem and the system feels a bit more responsive

too, thanks.
I guess I should probably do both 1) and 3) as my board is pretty slow

too.

Been trying to figure out a good value for NAPI weigth too. Currently

my

Quoted 1 line
Quoted 1 line

HW RX and TX queues are 16 pkgs deep and weigth is 16 too. If I move

TX

Quoted 4 lines
Quoted 4 lines

processing
to NAPI context AND increase weigth to 32, the system is a lot more responsive during
ping flooding. Does weigth 32 make sense when the HW TX and RX queues

are

Quoted 3 lines
Quoted 1 line

16?

If you only have one NIC, I dont understand why changing weight should

make

Quoted 6 lines

a difference. Are you referring to dev_weight or netdev_budget ?

# cat /proc/sys/net/core/dev_weight
64
# cat /proc/sys/net/core/netdev_budget
300

I mean this call in ucc_geth:
netif_napi_add(dev, &ugeth->napi, ucc_geth_poll, UCC_GETH_DEV_WEIGHT);
UCC_GETH_DEV_WEIGHT is 16

Noticed that rcuclassic.c has a
module_param(qhimark, int, 0);
But I can't figure out hot to set this qhimark from the cmdline.
rcuclassic.c is not a module(I don't use modules at all)

Jocke
Jocke

#27 lore raw

Re: [PATCH] conntrack: Reduce conntrack count in nf_conntrack_free()

In reply to Eric Dumazet

On Tue, 2009-03-24 at 13:07 +0100, Eric Dumazet wrote:

Hi Eric,

Quoted 6 lines

We use RCU to defer freeing of conntrack structures. In DOS situation, RCU might accumulate about 10.000 elements per CPU in its internal queues. To get accurate conntrack counts (at the expense of slightly more RAM used), we might consider conntrack counter not taking into account "about to be freed elements, waiting in RCU queues". We thus decrement it in nf_conntrack_free(), not in the RCU callback.

Your patch fixes the problem on my board too (embedded mips router 250Mhz), thanks.

Yet I'm concerned about what you said concerning RAM usage. I have a very small amount on memory left on my board (less than 4M), and I tuned ip route cache size and nf_conntrack_max to make sure I won't go OOM.

With your patch, does it mean 10000 conntrack entries can be allocated while nf_conntrack_max is say only 2048 ?

Regards,

Signature · 1 line
Maxime
#28 lore raw

Re: [PATCH] conntrack: Reduce conntrack count in nf_conntrack_free()

Quoted 12 lines · Maxime Bizon
Quoted 6 lines · Eric Dumazet

We use RCU to defer freeing of conntrack structures. In DOS situation, RCU might accumulate about 10.000 elements per CPU in its internal queues. To get accurate conntrack counts (at the expense of slightly more RAM used), we might consider conntrack counter not taking into account "about to be freed elements, waiting in RCU queues". We thus decrement it in nf_conntrack_free(), not in the RCU callback.

Your patch fixes the problem on my board too (embedded mips router 250Mhz), thanks.

Quoted 6 lines

Yet I'm concerned about what you said concerning RAM usage. I have a very small amount on memory left on my board (less than 4M), and I tuned ip route cache size and nf_conntrack_max to make sure I won't go OOM.

With your patch, does it mean 10000 conntrack entries can be allocated while nf_conntrack_max is say only 2048 ?

Temporarily under worst-case circumstances, yes. Eric is already working on his proposed improvement though :)

#29 lore raw

Re: [PATCH] conntrack: Reduce conntrack count in nf_conntrack_free()

In reply to Maxime Bizon
Quoted 14 lines · Maxime Bizon

On Tue, 2009-03-24 at 13:07 +0100, Eric Dumazet wrote:

Hi Eric,

Quoted 6 lines

We use RCU to defer freeing of conntrack structures. In DOS situation, RCU might accumulate about 10.000 elements per CPU in its internal queues. To get accurate conntrack counts (at the expense of slightly more RAM used), we might consider conntrack counter not taking into account "about to be freed elements, waiting in RCU queues". We thus decrement it in nf_conntrack_free(), not in the RCU callback.

Your patch fixes the problem on my board too (embedded mips router 250Mhz), thanks.

Quoted 6 lines

Yet I'm concerned about what you said concerning RAM usage. I have a very small amount on memory left on my board (less than 4M), and I tuned ip route cache size and nf_conntrack_max to make sure I won't go OOM.

With your patch, does it mean 10000 conntrack entries can be allocated while nf_conntrack_max is say only 2048 ?

Well... yes, RCU can have this 'interesting' OOM property.

For small machines, you really want to lower RCU parameters, because as you said, we also push route cache entries in RCU queue, my patch being applied or not (But using call_rcu_bh(), so we have lower latencies I think)

We are working on a SLAB_DESTROY_BY_RCU implementation so that conntrack wont use call_rcu() anymore, give us a couple of days :)

Paul, could we have /sys knobs to be able to tune qhimark, blimit & qlowmark ?

Thanks

#30 lore raw

PATCH 1/1 netfilter: Use hlist_add_head_rcu() in nf_conntrack_set_hashsize()

Quoted 4 lines · Eric Dumazet

We are working on a SLAB_DESTROY_BY_RCU implementation so that conntrack wont use call_rcu() anymore, give us a couple of days :)

While working on this stuff, I found one suspect use of hlist_add_head()

Its not a hot path, I believe following patch would make sure nothing wrong happens.

If a chain contains element A and B, then we might build a new table with a new chain containing B and A (in this reverse order), and a cpu could see A->next = B (new pointer), B->next = A (old pointer)

Thanks

[PATCH] netfilter: Use hlist_add_head_rcu() in nf_conntrack_set_hashsize()

Using hlist_add_head() in nf_conntrack_set_hashsize() is quite dangerous. Without any barrier, one CPU could see a loop while doing its lookup. Its true new table cannot be seen by another cpu, but previous table is still readable.

net/netfilter/nf_conntrack_core.c +1 -1
index 55befe5..54e983f 100644
@@ -1121,7 +1121,7 @@ int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
 					struct nf_conntrack_tuple_hash, hnode);
 			hlist_del_rcu(&h->hnode);
 			bucket = __hash_conntrack(&h->tuple, hashsize, rnd);
-			hlist_add_head(&h->hnode, &hash[bucket]);
+			hlist_add_head_rcu(&h->hnode, &hash[bucket]);
 		}
 	}
 	old_size = nf_conntrack_htable_size;
#31 lore raw

Re: [PATCH] netfilter: Use hlist_add_head_rcu() in nf_conntrack_set_hashsize()

Quoted 8 lines · Eric Dumazet

While working on this stuff, I found one suspect use of hlist_add_head()

Its not a hot path, I believe following patch would make sure nothing wrong happens.

If a chain contains element A and B, then we might build a new table with a new chain containing B and A (in this reverse order), and a cpu could see A->next = B (new pointer), B->next = A (old pointer)

I think you're right.

Quoted 6 lines

[PATCH] netfilter: Use hlist_add_head_rcu() in nf_conntrack_set_hashsize()

Using hlist_add_head() in nf_conntrack_set_hashsize() is quite dangerous. Without any barrier, one CPU could see a loop while doing its lookup. Its true new table cannot be seen by another cpu, but previous table is still readable.

Applied, thanks Eric.

#32 lore raw

PATCH 1/1 conntrack: use SLAB_DESTROY_BY_RCU for nf_conn structs

In reply to Eric Dumazet

Hi Patrick

Here is the patch I had the time to test this time... No problem so far on my machine.
I did a UDP flood stress.

Thank you

[PATCH] conntrack: use SLAB_DESTROY_BY_RCU and get rid of call_rcu()

Use "hlist_nulls" infrastructure we added in 2.6.29 for RCUification of UDP & TCP.

This permits an easy conversion from call_rcu() based hash lists to a SLAB_DESTROY_BY_RCU one.

Avoiding call_rcu() delay at nf_conn freeing time has numerous gains.

First, it doesnt fill RCU queues (up to 10000 elements per cpu). This reduces OOM possibility, if queued elements are not taken into account This reduces latency problems when RCU queue size hits hilimit and triggers emergency mode.

- It allows fast reuse of just freed elements, permitting better use of CPU cache.

- We delete rcu_head from "struct nf_conn", shrinking size of this structure by 8 or 16 bytes.

This patch only takes care of "struct nf_conn". call_rcu() is still used for less critical conntrack parts, that may be converted later if necessary.

---
 include/net/netfilter/nf_conntrack.h                  |   14 -
 include/net/netfilter/nf_conntrack_tuple.h            |    6
 include/net/netns/conntrack.h                         |    5
 net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c |   16 -
 net/ipv4/netfilter/nf_nat_core.c                      |    2
 net/netfilter/nf_conntrack_core.c                     |  123 +++++-----
 net/netfilter/nf_conntrack_expect.c                   |    2
 net/netfilter/nf_conntrack_helper.c                   |    7
 net/netfilter/nf_conntrack_netlink.c                  |   10
 net/netfilter/nf_conntrack_standalone.c               |   16 -
 net/netfilter/xt_connlimit.c                          |    4
 11 files changed, 114 insertions(+), 91 deletions(-)
include/net/netfilter/nf_conntrack.h +8 -6
index 4dfb793..6c3f964 100644
@@ -91,8 +91,7 @@ struct nf_conn_help {
 #include <net/netfilter/ipv4/nf_conntrack_ipv4.h>
 #include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
 
-struct nf_conn
-{
+struct nf_conn {
 	/* Usage count in here is 1 for hash table/destruct timer, 1 per skb,
            plus 1 for any connection(s) we are `master' for */
 	struct nf_conntrack ct_general;
@@ -126,7 +125,6 @@ struct nf_conn
 #ifdef CONFIG_NET_NS
 	struct net *ct_net;
 #endif
-	struct rcu_head rcu;
 };
 
 static inline struct nf_conn *
@@ -190,9 +188,13 @@ static inline void nf_ct_put(struct nf_conn *ct)
 extern int nf_ct_l3proto_try_module_get(unsigned short l3proto);
 extern void nf_ct_l3proto_module_put(unsigned short l3proto);
 
-extern struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced);
-extern void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced,
-				 unsigned int size);
+/*
+ * Allocate a hashtable of hlist_head (if nulls == 0),
+ * or hlist_nulls_head (if nulls == 1)
+ */
+extern void *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced, int nulls);
+
+extern void nf_ct_free_hashtable(void *hash, int vmalloced, unsigned int size);
 
 extern struct nf_conntrack_tuple_hash *
 __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple);
include/net/netfilter/nf_conntrack_tuple.h +3 -3
index f2f6aa7..2628c15 100644
@@ -12,6 +12,7 @@
 
 #include <linux/netfilter/x_tables.h>
 #include <linux/netfilter/nf_conntrack_tuple_common.h>
+#include <linux/list_nulls.h>
 
 /* A `tuple' is a structure containing the information to uniquely
   identify a connection.  ie. if two packets have the same tuple, they
@@ -146,9 +147,8 @@ static inline void nf_ct_dump_tuple(const struct nf_conntrack_tuple *t)
 	((enum ip_conntrack_dir)(h)->tuple.dst.dir)
 
 /* Connections have two entries in the hash table: one for each way */
-struct nf_conntrack_tuple_hash
-{
-	struct hlist_node hnode;
+struct nf_conntrack_tuple_hash {
+	struct hlist_nulls_node hnnode;
 	struct nf_conntrack_tuple tuple;
 };
 
include/net/netns/conntrack.h +3 -2
index f4498a6..9dc5840 100644
@@ -2,6 +2,7 @@
 #define __NETNS_CONNTRACK_H
 
 #include <linux/list.h>
+#include <linux/list_nulls.h>
 #include <asm/atomic.h>
 
 struct ctl_table_header;
@@ -10,9 +11,9 @@ struct nf_conntrack_ecache;
 struct netns_ct {
 	atomic_t		count;
 	unsigned int		expect_count;
-	struct hlist_head	*hash;
+	struct hlist_nulls_head	*hash;
 	struct hlist_head	*expect_hash;
-	struct hlist_head	unconfirmed;
+	struct hlist_nulls_head	unconfirmed;
 	struct ip_conntrack_stat *stat;
 #ifdef CONFIG_NF_CONNTRACK_EVENTS
 	struct nf_conntrack_ecache *ecache;
net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c +8 -8
index 6ba5c55..fcbcf62 100644
@@ -25,30 +25,30 @@ struct ct_iter_state {
 	unsigned int bucket;
 };
 
-static struct hlist_node *ct_get_first(struct seq_file *seq)
+static struct hlist_nulls_node *ct_get_first(struct seq_file *seq)
 {
 	struct net *net = seq_file_net(seq);
 	struct ct_iter_state *st = seq->private;
-	struct hlist_node *n;
+	struct hlist_nulls_node *n;
 
 	for (st->bucket = 0;
 	     st->bucket < nf_conntrack_htable_size;
 	     st->bucket++) {
 		n = rcu_dereference(net->ct.hash[st->bucket].first);
-		if (n)
+		if (!is_a_nulls(n))
 			return n;
 	}
 	return NULL;
 }
 
-static struct hlist_node *ct_get_next(struct seq_file *seq,
-				      struct hlist_node *head)
+static struct hlist_nulls_node *ct_get_next(struct seq_file *seq,
+				      struct hlist_nulls_node *head)
 {
 	struct net *net = seq_file_net(seq);
 	struct ct_iter_state *st = seq->private;
 
 	head = rcu_dereference(head->next);
-	while (head == NULL) {
+	while (is_a_nulls(head)) {
 		if (++st->bucket >= nf_conntrack_htable_size)
 			return NULL;
 		head = rcu_dereference(net->ct.hash[st->bucket].first);
@@ -56,9 +56,9 @@ static struct hlist_node *ct_get_next(struct seq_file *seq,
 	return head;
 }
 
-static struct hlist_node *ct_get_idx(struct seq_file *seq, loff_t pos)
+static struct hlist_nulls_node *ct_get_idx(struct seq_file *seq, loff_t pos)
 {
-	struct hlist_node *head = ct_get_first(seq);
+	struct hlist_nulls_node *head = ct_get_first(seq);
 
 	if (head)
 		while (pos && (head = ct_get_next(seq, head)))
net/ipv4/netfilter/nf_nat_core.c +1 -1
index a65cf69..fe65187 100644
@@ -679,7 +679,7 @@ nfnetlink_parse_nat_setup(struct nf_conn *ct,
 static int __net_init nf_nat_net_init(struct net *net)
 {
 	net->ipv4.nat_bysource = nf_ct_alloc_hashtable(&nf_nat_htable_size,
-						      &net->ipv4.nat_vmalloced);
+						      &net->ipv4.nat_vmalloced, 0);
 	if (!net->ipv4.nat_bysource)
 		return -ENOMEM;
 	return 0;
net/netfilter/nf_conntrack_core.c +70 -53

319 lines not on this page · raw message

net/netfilter/nf_conntrack_expect.c +1 -1
index 357ba39..3940f99 100644
@@ -604,7 +604,7 @@ int nf_conntrack_expect_init(struct net *net)
 
 	net->ct.expect_count = 0;
 	net->ct.expect_hash = nf_ct_alloc_hashtable(&nf_ct_expect_hsize,
-						  &net->ct.expect_vmalloc);
+						  &net->ct.expect_vmalloc, 0);
 	if (net->ct.expect_hash == NULL)
 		goto err1;
 
net/netfilter/nf_conntrack_helper.c +4 -3
index a51bdac..6066144 100644
@@ -158,6 +158,7 @@ static void __nf_conntrack_helper_unregister(struct nf_conntrack_helper *me,
 	struct nf_conntrack_tuple_hash *h;
 	struct nf_conntrack_expect *exp;
 	const struct hlist_node *n, *next;
+	const struct hlist_nulls_node *nn;
 	unsigned int i;
 
 	/* Get rid of expectations */
@@ -174,10 +175,10 @@ static void __nf_conntrack_helper_unregister(struct nf_conntrack_helper *me,
 	}
 
 	/* Get rid of expecteds, set helpers to NULL. */
-	hlist_for_each_entry(h, n, &net->ct.unconfirmed, hnode)
+	hlist_for_each_entry(h, nn, &net->ct.unconfirmed, hnnode)
 		unhelp(h, me);
 	for (i = 0; i < nf_conntrack_htable_size; i++) {
-		hlist_for_each_entry(h, n, &net->ct.hash[i], hnode)
+		hlist_nulls_for_each_entry(h, nn, &net->ct.hash[i], hnnode)
 			unhelp(h, me);
 	}
 }
@@ -217,7 +218,7 @@ int nf_conntrack_helper_init(void)
 
 	nf_ct_helper_hsize = 1; /* gets rounded up to use one page */
 	nf_ct_helper_hash = nf_ct_alloc_hashtable(&nf_ct_helper_hsize,
-						  &nf_ct_helper_vmalloc);
+						  &nf_ct_helper_vmalloc, 0);
 	if (!nf_ct_helper_hash)
 		return -ENOMEM;
 
net/netfilter/nf_conntrack_netlink.c +5 -5
index 1b75c9e..e73272b 100644
@@ -19,6 +19,7 @@
 #include <linux/module.h>
 #include <linux/kernel.h>
 #include <linux/rculist.h>
+#include <linux/rculist_nulls.h>
 #include <linux/types.h>
 #include <linux/timer.h>
 #include <linux/skbuff.h>
@@ -536,7 +537,7 @@ ctnetlink_dump_table(struct sk_buff *skb, struct netlink_callback *cb)
 {
 	struct nf_conn *ct, *last;
 	struct nf_conntrack_tuple_hash *h;
-	struct hlist_node *n;
+	struct hlist_nulls_node *n;
 	struct nfgenmsg *nfmsg = NLMSG_DATA(cb->nlh);
 	u_int8_t l3proto = nfmsg->nfgen_family;
 
@@ -544,8 +545,8 @@ ctnetlink_dump_table(struct sk_buff *skb, struct netlink_callback *cb)
 	last = (struct nf_conn *)cb->args[1];
 	for (; cb->args[0] < nf_conntrack_htable_size; cb->args[0]++) {
 restart:
-		hlist_for_each_entry_rcu(h, n, &init_net.ct.hash[cb->args[0]],
-					 hnode) {
+		hlist_nulls_for_each_entry_rcu(h, n, &init_net.ct.hash[cb->args[0]],
+					 hnnode) {
 			if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
 				continue;
 			ct = nf_ct_tuplehash_to_ctrack(h);
@@ -1242,13 +1243,12 @@ ctnetlink_create_conntrack(struct nlattr *cda[],
 		if (err < 0)
 			goto err2;
 
-		master_h = __nf_conntrack_find(&init_net, &master);
+		master_h = nf_conntrack_find_get(&init_net, &master);
 		if (master_h == NULL) {
 			err = -ENOENT;
 			goto err2;
 		}
 		master_ct = nf_ct_tuplehash_to_ctrack(master_h);
-		nf_conntrack_get(&master_ct->ct_general);
 		__set_bit(IPS_EXPECTED_BIT, &ct->status);
 		ct->master = master_ct;
 	}
net/netfilter/nf_conntrack_standalone.c +8 -8
index 4da54b0..f768368 100644
@@ -44,30 +44,30 @@ struct ct_iter_state {
 	unsigned int bucket;
 };
 
-static struct hlist_node *ct_get_first(struct seq_file *seq)
+static struct hlist_nulls_node *ct_get_first(struct seq_file *seq)
 {
 	struct net *net = seq_file_net(seq);
 	struct ct_iter_state *st = seq->private;
-	struct hlist_node *n;
+	struct hlist_nulls_node *n;
 
 	for (st->bucket = 0;
 	     st->bucket < nf_conntrack_htable_size;
 	     st->bucket++) {
 		n = rcu_dereference(net->ct.hash[st->bucket].first);
-		if (n)
+		if (!is_a_nulls(n))
 			return n;
 	}
 	return NULL;
 }
 
-static struct hlist_node *ct_get_next(struct seq_file *seq,
-				      struct hlist_node *head)
+static struct hlist_nulls_node *ct_get_next(struct seq_file *seq,
+				      struct hlist_nulls_node *head)
 {
 	struct net *net = seq_file_net(seq);
 	struct ct_iter_state *st = seq->private;
 
 	head = rcu_dereference(head->next);
-	while (head == NULL) {
+	while (is_a_nulls(head)) {
 		if (++st->bucket >= nf_conntrack_htable_size)
 			return NULL;
 		head = rcu_dereference(net->ct.hash[st->bucket].first);
@@ -75,9 +75,9 @@ static struct hlist_node *ct_get_next(struct seq_file *seq,
 	return head;
 }
 
-static struct hlist_node *ct_get_idx(struct seq_file *seq, loff_t pos)
+static struct hlist_nulls_node *ct_get_idx(struct seq_file *seq, loff_t pos)
 {
-	struct hlist_node *head = ct_get_first(seq);
+	struct hlist_nulls_node *head = ct_get_first(seq);
 
 	if (head)
 		while (pos && (head = ct_get_next(seq, head)))
net/netfilter/xt_connlimit.c +3 -1
index 7f404cc..3bf47d8 100644
@@ -123,7 +123,7 @@ static int count_them(struct xt_connlimit_data *data,
 
 	/* check the saved connections */
 	list_for_each_entry_safe(conn, tmp, hash, list) {
-		found    = __nf_conntrack_find(&init_net, &conn->tuple);
+		found    = nf_conntrack_find_get(&init_net, &conn->tuple);
 		found_ct = NULL;
 
 		if (found != NULL)
@@ -151,6 +151,7 @@ static int count_them(struct xt_connlimit_data *data,
 			 * we do not care about connections which are
 			 * closed already -> ditch it
 			 */
+			nf_ct_put(found_ct);
 			list_del(&conn->list);
 			kfree(conn);
 			continue;
@@ -160,6 +161,7 @@ static int count_them(struct xt_connlimit_data *data,
 		    match->family))
 			/* same source network -> be counted! */
 			++matches;
+		nf_ct_put(found_ct);
 	}
 
 	rcu_read_unlock();
#33 lore raw

Re: [PATCH] conntrack: use SLAB_DESTROY_BY_RCU for nf_conn structs

Quoted 5 lines · Eric Dumazet

Hi Patrick

Here is the patch I had the time to test this time... No problem so far on my machine.
I did a UDP flood stress.

Thanks Eric. Most parts looks good, just two questions below:

Quoted 16 lines
net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c +2 -2
index 6ba5c55..fcbcf62 100644
@@ -25,30 +25,30 @@ struct ct_iter_state {
 	unsigned int bucket;
 };
 
-static struct hlist_node *ct_get_first(struct seq_file *seq)
+static struct hlist_nulls_node *ct_get_first(struct seq_file *seq)
 {
 	struct net *net = seq_file_net(seq);
 	struct ct_iter_state *st = seq->private;
-	struct hlist_node *n;
+	struct hlist_nulls_node *n;
 

net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c struct ct_iter_state { · line 42

Quoted 10 lines
net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c +1 -1
 	for (st->bucket = 0;
 	     st->bucket < nf_conntrack_htable_size;
 	     st->bucket++) {
 		n = rcu_dereference(net->ct.hash[st->bucket].first);
-		if (n)
+		if (!is_a_nulls(n))
 			return n;
 	}
 	return NULL;
 }

Don't we need to make sure the entry is not reused while dumping it?

Quoted 26 lines
net/netfilter/nf_conntrack_helper.c +3 -2
index a51bdac..6066144 100644
@@ -158,6 +158,7 @@ static void __nf_conntrack_helper_unregister(struct nf_conntrack_helper *me,
 	struct nf_conntrack_tuple_hash *h;
 	struct nf_conntrack_expect *exp;
 	const struct hlist_node *n, *next;
+	const struct hlist_nulls_node *nn;
 	unsigned int i;
 
 	/* Get rid of expectations */
@@ -174,10 +175,10 @@ static void __nf_conntrack_helper_unregister(struct nf_conntrack_helper *me,
 	}
 
 	/* Get rid of expecteds, set helpers to NULL. */
-	hlist_for_each_entry(h, n, &net->ct.unconfirmed, hnode)
+	hlist_for_each_entry(h, nn, &net->ct.unconfirmed, hnnode)
 		unhelp(h, me);
 	for (i = 0; i < nf_conntrack_htable_size; i++) {
-		hlist_for_each_entry(h, n, &net->ct.hash[i], hnode)
+		hlist_nulls_for_each_entry(h, nn, &net->ct.hash[i], hnnode)
 			unhelp(h, me);
 	}
 }

net/netfilter/nf_conntrack_helper.c int nf_conntrack_helper_init(void) · line 221

Quoted 5 lines
net/netfilter/nf_conntrack_helper.c +1 -1
@@ -217,7 +218,7 @@ int nf_conntrack_helper_init(void)
 
 	nf_ct_helper_hsize = 1; /* gets rounded up to use one page */
 	nf_ct_helper_hash = nf_ct_alloc_hashtable(&nf_ct_helper_hsize,
-						  &nf_ct_helper_vmalloc);
+						  &nf_ct_helper_vmalloc, 0);

This should be "1" I think since it wants a hlist_nulls hash.

#34 lore raw

Re: [PATCH] conntrack: use SLAB_DESTROY_BY_RCU for nf_conn structs

Quoted 8 lines · Patrick McHardy

net/netfilter/nf_conntrack_helper.c int nf_conntrack_helper_init(void) · line 221

Quoted 6 lines
net/netfilter/nf_conntrack_helper.c +1 -1
@@ -217,7 +218,7 @@ int nf_conntrack_helper_init(void)
 
     nf_ct_helper_hsize = 1; /* gets rounded up to use one page */
     nf_ct_helper_hash = nf_ct_alloc_hashtable(&nf_ct_helper_hsize,
-                          &nf_ct_helper_vmalloc);
+                          &nf_ct_helper_vmalloc, 0);

This should be "1" I think since it wants a hlist_nulls hash.

OK I just realized my mistake, please ignore :)

#35 lore raw

Re: [PATCH] conntrack: use SLAB_DESTROY_BY_RCU for nf_conn structs

In reply to Patrick McHardy
Quoted 26 lines · Patrick McHardy
Quoted 5 lines · Eric Dumazet

Hi Patrick

Here is the patch I had the time to test this time... No problem so far on my machine.
I did a UDP flood stress.

Thanks Eric. Most parts looks good, just two questions below:

Quoted 17 lines
net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c +2 -2
index 6ba5c55..fcbcf62 100644
@@ -25,30 +25,30 @@ struct ct_iter_state {
     unsigned int bucket;
 };
 
-static struct hlist_node *ct_get_first(struct seq_file *seq)
+static struct hlist_nulls_node *ct_get_first(struct seq_file *seq)
 {
     struct net *net = seq_file_net(seq);
     struct ct_iter_state *st = seq->private;
-    struct hlist_node *n;
+    struct hlist_nulls_node *n;
 
Quoted 14 lines

net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c struct ct_iter_state { · line 42

Quoted 10 lines
net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c +1 -1
     for (st->bucket = 0;
          st->bucket < nf_conntrack_htable_size;
          st->bucket++) {
         n = rcu_dereference(net->ct.hash[st->bucket].first);
-        if (n)
+        if (!is_a_nulls(n))
             return n;
     }
     return NULL;
 }

Don't we need to make sure the entry is not reused while dumping it?

Ah yes, I forgot that for UDP/TCP I had to change locking on this part. Because messing with reference count was crazy...
But in UDP/TCP we have different spinlock for each chain, so hold time was small enough.

So I guess that with central conntrack lock, we need to take references on entries while dumping them.

#36 lore raw

Re: [PATCH] conntrack: use SLAB_DESTROY_BY_RCU for nf_conn structs

Quoted 22 lines · Eric Dumazet
Quoted 17 lines · Patrick McHardy
Quoted 7 lines
net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c +2 -1
+static struct hlist_nulls_node *ct_get_first(struct seq_file *seq)
 {
     struct net *net = seq_file_net(seq);
     struct ct_iter_state *st = seq->private;
-    struct hlist_node *n;
+    struct hlist_nulls_node *n;
 

net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c struct ct_iter_state { · line 42

Quoted 10 lines
net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c +1 -1
     for (st->bucket = 0;
          st->bucket < nf_conntrack_htable_size;
          st->bucket++) {
         n = rcu_dereference(net->ct.hash[st->bucket].first);
-        if (n)
+        if (!is_a_nulls(n))
             return n;
     }
     return NULL;
 }
Quoted 3 lines

Don't we need to make sure the entry is not reused while dumping it?

Quoted 7 lines

Ah yes, I forgot that for UDP/TCP I had to change locking on this part. Because messing with reference count was crazy...
But in UDP/TCP we have different spinlock for each chain, so hold time was small enough.

So I guess that with central conntrack lock, we need to take references on entries while dumping them.

Yes, I think so too.
--
To unsubscribe from this list: send the line "unsubscribe netfilter-devel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#37 lore raw

Re: [PATCH] conntrack: use SLAB_DESTROY_BY_RCU for nf_conn structs

Quoted 28 lines · Patrick McHardy
Quoted 27 lines · Eric Dumazet
Quoted 17 lines · Patrick McHardy
Quoted 7 lines
(quoted diff) +2 -1
+static struct hlist_nulls_node *ct_get_first(struct seq_file *seq)
 {
     struct net *net = seq_file_net(seq);
     struct ct_iter_state *st = seq->private;
-    struct hlist_node *n;
+    struct hlist_nulls_node *n;
 

Review

Quoted 10 lines
(quoted diff) +1 -1
     for (st->bucket = 0;
          st->bucket < nf_conntrack_htable_size;
          st->bucket++) {
         n = rcu_dereference(net->ct.hash[st->bucket].first);
-        if (n)
+        if (!is_a_nulls(n))
             return n;
     }
     return NULL;
 }
Quoted 3 lines

Don't we need to make sure the entry is not reused while dumping it?

Ah yes, I forgot that for UDP/TCP I had to change locking on this part. Because messing with reference count was crazy...
But in UDP/TCP we have different spinlock for each chain, so hold time was small enough.

Quoted 5 lines
Quoted 3 lines

So I guess that with central conntrack lock, we need to take references on entries
while dumping them.

Yes, I think so too.

Here is take 2 of the patch with proper ref counting on dumping.

Thank you

[PATCH] conntrack: use SLAB_DESTROY_BY_RCU and get rid of call_rcu()

Use "hlist_nulls" infrastructure we added in 2.6.29 for RCUification of UDP & TCP.

This permits an easy conversion from call_rcu() based hash lists to a SLAB_DESTROY_BY_RCU one.

Avoiding call_rcu() delay at nf_conn freeing time has numerous gains.

First, it doesnt fill RCU queues (up to 10000 elements per cpu). This reduces OOM possibility, if queued elements are not taken into account This reduces latency problems when RCU queue size hits hilimit and triggers emergency mode.

- It allows fast reuse of just freed elements, permitting better use of CPU cache.

- We delete rcu_head from "struct nf_conn", shrinking size of this structure by 8 or 16 bytes.

This patch only takes care of "struct nf_conn". call_rcu() is still used for less critical conntrack parts, that may be converted later if necessary.

---
 include/net/netfilter/nf_conntrack.h                  |   14 -
 include/net/netfilter/nf_conntrack_tuple.h            |    6
 include/net/netns/conntrack.h                         |    5
 net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c |   57 ++--
 net/ipv4/netfilter/nf_nat_core.c                      |    2
 net/netfilter/nf_conntrack_core.c                     |  123 +++++-----
 net/netfilter/nf_conntrack_expect.c                   |    2
 net/netfilter/nf_conntrack_helper.c                   |    7
 net/netfilter/nf_conntrack_netlink.c                  |   20 -
 net/netfilter/nf_conntrack_standalone.c               |   51 ++--
 net/netfilter/xt_connlimit.c                          |    6
 11 files changed, 166 insertions(+), 127 deletions(-)
include/net/netfilter/nf_conntrack.h +8 -6
index 4dfb793..6c3f964 100644
@@ -91,8 +91,7 @@ struct nf_conn_help {
 #include <net/netfilter/ipv4/nf_conntrack_ipv4.h>
 #include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
 
-struct nf_conn
-{
+struct nf_conn {
 	/* Usage count in here is 1 for hash table/destruct timer, 1 per skb,
            plus 1 for any connection(s) we are `master' for */
 	struct nf_conntrack ct_general;
@@ -126,7 +125,6 @@ struct nf_conn
 #ifdef CONFIG_NET_NS
 	struct net *ct_net;
 #endif
-	struct rcu_head rcu;
 };
 
 static inline struct nf_conn *
@@ -190,9 +188,13 @@ static inline void nf_ct_put(struct nf_conn *ct)
 extern int nf_ct_l3proto_try_module_get(unsigned short l3proto);
 extern void nf_ct_l3proto_module_put(unsigned short l3proto);
 
-extern struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced);
-extern void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced,
-				 unsigned int size);
+/*
+ * Allocate a hashtable of hlist_head (if nulls == 0),
+ * or hlist_nulls_head (if nulls == 1)
+ */
+extern void *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced, int nulls);
+
+extern void nf_ct_free_hashtable(void *hash, int vmalloced, unsigned int size);
 
 extern struct nf_conntrack_tuple_hash *
 __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple);
include/net/netfilter/nf_conntrack_tuple.h +3 -3
index f2f6aa7..2628c15 100644
@@ -12,6 +12,7 @@
 
 #include <linux/netfilter/x_tables.h>
 #include <linux/netfilter/nf_conntrack_tuple_common.h>
+#include <linux/list_nulls.h>
 
 /* A `tuple' is a structure containing the information to uniquely
   identify a connection.  ie. if two packets have the same tuple, they
@@ -146,9 +147,8 @@ static inline void nf_ct_dump_tuple(const struct nf_conntrack_tuple *t)
 	((enum ip_conntrack_dir)(h)->tuple.dst.dir)
 
 /* Connections have two entries in the hash table: one for each way */
-struct nf_conntrack_tuple_hash
-{
-	struct hlist_node hnode;
+struct nf_conntrack_tuple_hash {
+	struct hlist_nulls_node hnnode;
 	struct nf_conntrack_tuple tuple;
 };
 
include/net/netns/conntrack.h +3 -2
index f4498a6..9dc5840 100644
@@ -2,6 +2,7 @@
 #define __NETNS_CONNTRACK_H
 
 #include <linux/list.h>
+#include <linux/list_nulls.h>
 #include <asm/atomic.h>
 
 struct ctl_table_header;
@@ -10,9 +11,9 @@ struct nf_conntrack_ecache;
 struct netns_ct {
 	atomic_t		count;
 	unsigned int		expect_count;
-	struct hlist_head	*hash;
+	struct hlist_nulls_head	*hash;
 	struct hlist_head	*expect_hash;
-	struct hlist_head	unconfirmed;
+	struct hlist_nulls_head	unconfirmed;
 	struct ip_conntrack_stat *stat;
 #ifdef CONFIG_NF_CONNTRACK_EVENTS
 	struct nf_conntrack_ecache *ecache;
net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c +32 -25

140 lines not on this page · raw message

net/ipv4/netfilter/nf_nat_core.c +1 -1
index a65cf69..fe65187 100644
@@ -679,7 +679,7 @@ nfnetlink_parse_nat_setup(struct nf_conn *ct,
 static int __net_init nf_nat_net_init(struct net *net)
 {
 	net->ipv4.nat_bysource = nf_ct_alloc_hashtable(&nf_nat_htable_size,
-						      &net->ipv4.nat_vmalloced);
+						      &net->ipv4.nat_vmalloced, 0);
 	if (!net->ipv4.nat_bysource)
 		return -ENOMEM;
 	return 0;
net/netfilter/nf_conntrack_core.c +70 -53

319 lines not on this page · raw message

net/netfilter/nf_conntrack_expect.c +1 -1
index 357ba39..3940f99 100644
@@ -604,7 +604,7 @@ int nf_conntrack_expect_init(struct net *net)
 
 	net->ct.expect_count = 0;
 	net->ct.expect_hash = nf_ct_alloc_hashtable(&nf_ct_expect_hsize,
-						  &net->ct.expect_vmalloc);
+						  &net->ct.expect_vmalloc, 0);
 	if (net->ct.expect_hash == NULL)
 		goto err1;
 
net/netfilter/nf_conntrack_helper.c +4 -3
index a51bdac..6066144 100644
@@ -158,6 +158,7 @@ static void __nf_conntrack_helper_unregister(struct nf_conntrack_helper *me,
 	struct nf_conntrack_tuple_hash *h;
 	struct nf_conntrack_expect *exp;
 	const struct hlist_node *n, *next;
+	const struct hlist_nulls_node *nn;
 	unsigned int i;
 
 	/* Get rid of expectations */
@@ -174,10 +175,10 @@ static void __nf_conntrack_helper_unregister(struct nf_conntrack_helper *me,
 	}
 
 	/* Get rid of expecteds, set helpers to NULL. */
-	hlist_for_each_entry(h, n, &net->ct.unconfirmed, hnode)
+	hlist_for_each_entry(h, nn, &net->ct.unconfirmed, hnnode)
 		unhelp(h, me);
 	for (i = 0; i < nf_conntrack_htable_size; i++) {
-		hlist_for_each_entry(h, n, &net->ct.hash[i], hnode)
+		hlist_nulls_for_each_entry(h, nn, &net->ct.hash[i], hnnode)
 			unhelp(h, me);
 	}
 }
@@ -217,7 +218,7 @@ int nf_conntrack_helper_init(void)
 
 	nf_ct_helper_hsize = 1; /* gets rounded up to use one page */
 	nf_ct_helper_hash = nf_ct_alloc_hashtable(&nf_ct_helper_hsize,
-						  &nf_ct_helper_vmalloc);
+						  &nf_ct_helper_vmalloc, 0);
 	if (!nf_ct_helper_hash)
 		return -ENOMEM;
 
net/netfilter/nf_conntrack_netlink.c +11 -9
index 1b75c9e..349bbef 100644
@@ -19,6 +19,7 @@
 #include <linux/module.h>
 #include <linux/kernel.h>
 #include <linux/rculist.h>
+#include <linux/rculist_nulls.h>
 #include <linux/types.h>
 #include <linux/timer.h>
 #include <linux/skbuff.h>
@@ -536,7 +537,7 @@ ctnetlink_dump_table(struct sk_buff *skb, struct netlink_callback *cb)
 {
 	struct nf_conn *ct, *last;
 	struct nf_conntrack_tuple_hash *h;
-	struct hlist_node *n;
+	struct hlist_nulls_node *n;
 	struct nfgenmsg *nfmsg = NLMSG_DATA(cb->nlh);
 	u_int8_t l3proto = nfmsg->nfgen_family;
 
@@ -544,27 +545,27 @@ ctnetlink_dump_table(struct sk_buff *skb, struct netlink_callback *cb)
 	last = (struct nf_conn *)cb->args[1];
 	for (; cb->args[0] < nf_conntrack_htable_size; cb->args[0]++) {
 restart:
-		hlist_for_each_entry_rcu(h, n, &init_net.ct.hash[cb->args[0]],
-					 hnode) {
+		hlist_nulls_for_each_entry_rcu(h, n, &init_net.ct.hash[cb->args[0]],
+					 hnnode) {
 			if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
 				continue;
 			ct = nf_ct_tuplehash_to_ctrack(h);
+			if (!atomic_inc_not_zero(&ct->ct_general.use))
+				continue;
 			/* Dump entries of a given L3 protocol number.
 			 * If it is not specified, ie. l3proto == 0,
 			 * then dump everything. */
 			if (l3proto && nf_ct_l3num(ct) != l3proto)
-				continue;
+				goto releasect;
 			if (cb->args[1]) {
 				if (ct != last)
-					continue;
+					goto releasect;
 				cb->args[1] = 0;
 			}
 			if (ctnetlink_fill_info(skb, NETLINK_CB(cb->skb).pid,
 						cb->nlh->nlmsg_seq,
 						IPCTNL_MSG_CT_NEW,
 						1, ct) < 0) {
-				if (!atomic_inc_not_zero(&ct->ct_general.use))
-					continue;
 				cb->args[1] = (unsigned long)ct;
 				goto out;
 			}
@@ -577,6 +578,8 @@ restart:
 				if (acct)
 					memset(acct, 0, sizeof(struct nf_conn_counter[IP_CT_DIR_MAX]));
 			}
+releasect:
+		nf_ct_put(ct);
 		}
 		if (cb->args[1]) {
 			cb->args[1] = 0;
@@ -1242,13 +1245,12 @@ ctnetlink_create_conntrack(struct nlattr *cda[],
 		if (err < 0)
 			goto err2;
 
-		master_h = __nf_conntrack_find(&init_net, &master);
+		master_h = nf_conntrack_find_get(&init_net, &master);
 		if (master_h == NULL) {
 			err = -ENOENT;
 			goto err2;
 		}
 		master_ct = nf_ct_tuplehash_to_ctrack(master_h);
-		nf_conntrack_get(&master_ct->ct_general);
 		__set_bit(IPS_EXPECTED_BIT, &ct->status);
 		ct->master = master_ct;
 	}
net/netfilter/nf_conntrack_standalone.c +29 -22

135 lines not on this page · raw message

net/netfilter/xt_connlimit.c +4 -2
index 7f404cc..6809809 100644
@@ -108,7 +108,7 @@ static int count_them(struct xt_connlimit_data *data,
 	const struct nf_conntrack_tuple_hash *found;
 	struct xt_connlimit_conn *conn;
 	struct xt_connlimit_conn *tmp;
-	const struct nf_conn *found_ct;
+	struct nf_conn *found_ct;
 	struct list_head *hash;
 	bool addit = true;
 	int matches = 0;
@@ -123,7 +123,7 @@ static int count_them(struct xt_connlimit_data *data,
 
 	/* check the saved connections */
 	list_for_each_entry_safe(conn, tmp, hash, list) {
-		found    = __nf_conntrack_find(&init_net, &conn->tuple);
+		found    = nf_conntrack_find_get(&init_net, &conn->tuple);
 		found_ct = NULL;
 
 		if (found != NULL)
@@ -151,6 +151,7 @@ static int count_them(struct xt_connlimit_data *data,
 			 * we do not care about connections which are
 			 * closed already -> ditch it
 			 */
+			nf_ct_put(found_ct);
 			list_del(&conn->list);
 			kfree(conn);
 			continue;
@@ -160,6 +161,7 @@ static int count_them(struct xt_connlimit_data *data,
 		    match->family))
 			/* same source network -> be counted! */
 			++matches;
+		nf_ct_put(found_ct);
 	}
 
 	rcu_read_unlock();

--
To unsubscribe from this list: send the line "unsubscribe netfilter-devel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#38 lore raw

Re: [PATCH] conntrack: use SLAB_DESTROY_BY_RCU for nf_conn structs

Quoted 1 line · Eric Dumazet

Here is take 2 of the patch with proper ref counting on dumping.

Thanks, one final question about the seq-file handling:

Quoted 58 lines
net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c +8 -9
index 6ba5c55..0b870b9 100644
@@ -25,30 +25,30 @@ struct ct_iter_state {
 	unsigned int bucket;
 };
 
-static struct hlist_node *ct_get_first(struct seq_file *seq)
+static struct hlist_nulls_node *ct_get_first(struct seq_file *seq)
 {
 	struct net *net = seq_file_net(seq);
 	struct ct_iter_state *st = seq->private;
-	struct hlist_node *n;
+	struct hlist_nulls_node *n;
 
 	for (st->bucket = 0;
 	     st->bucket < nf_conntrack_htable_size;
 	     st->bucket++) {
 		n = rcu_dereference(net->ct.hash[st->bucket].first);
-		if (n)
+		if (!is_a_nulls(n))
 			return n;
 	}
 	return NULL;
 }
 
-static struct hlist_node *ct_get_next(struct seq_file *seq,
-				      struct hlist_node *head)
+static struct hlist_nulls_node *ct_get_next(struct seq_file *seq,
+				      struct hlist_nulls_node *head)
 {
 	struct net *net = seq_file_net(seq);
 	struct ct_iter_state *st = seq->private;
 
 	head = rcu_dereference(head->next);
-	while (head == NULL) {
+	while (is_a_nulls(head)) {
 		if (++st->bucket >= nf_conntrack_htable_size)
 			return NULL;
 		head = rcu_dereference(net->ct.hash[st->bucket].first);
@@ -56,9 +56,9 @@ static struct hlist_node *ct_get_next(struct seq_file *seq,
 	return head;
 }
 
-static struct hlist_node *ct_get_idx(struct seq_file *seq, loff_t pos)
+static struct hlist_nulls_node *ct_get_idx(struct seq_file *seq, loff_t pos)
 {
-	struct hlist_node *head = ct_get_first(seq);
+	struct hlist_nulls_node *head = ct_get_first(seq);
 
 	if (head)
 		while (pos && (head = ct_get_next(seq, head)))
@@ -87,69 +87,76 @@ static void ct_seq_stop(struct seq_file *s, void *v)
 
 static int ct_seq_show(struct seq_file *s, void *v)
 {
-	const struct nf_conntrack_tuple_hash *hash = v;

net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c static void ct_seq_stop(struct seq_file *s, void *v) · line 98

Quoted 10 lines
net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c +5 -1
-	const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(hash);
+	struct nf_conntrack_tuple_hash *hash = v;
+	struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(hash);
 	const struct nf_conntrack_l3proto *l3proto;
 	const struct nf_conntrack_l4proto *l4proto;
+	int ret = 0;
 
 	NF_CT_ASSERT(ct);
+	if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
+		return 0;

Can we assume the next pointer still points to the next entry in the same chain after the refcount dropped to zero?

#39 lore raw

Re: [PATCH] conntrack: use SLAB_DESTROY_BY_RCU for nf_conn structs

Quoted 73 lines · Patrick McHardy
Quoted 1 line · Eric Dumazet

Here is take 2 of the patch with proper ref counting on dumping.

Thanks, one final question about the seq-file handling:

Quoted 68 lines
net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c +8 -8
index 6ba5c55..0b870b9 100644
@@ -25,30 +25,30 @@ struct ct_iter_state {
     unsigned int bucket;
 };
 
-static struct hlist_node *ct_get_first(struct seq_file *seq)
+static struct hlist_nulls_node *ct_get_first(struct seq_file *seq)
 {
     struct net *net = seq_file_net(seq);
     struct ct_iter_state *st = seq->private;
-    struct hlist_node *n;
+    struct hlist_nulls_node *n;
 
     for (st->bucket = 0;
          st->bucket < nf_conntrack_htable_size;
          st->bucket++) {
         n = rcu_dereference(net->ct.hash[st->bucket].first);
-        if (n)
+        if (!is_a_nulls(n))
             return n;
     }
     return NULL;
 }
 
-static struct hlist_node *ct_get_next(struct seq_file *seq,
-                      struct hlist_node *head)
+static struct hlist_nulls_node *ct_get_next(struct seq_file *seq,
+                      struct hlist_nulls_node *head)
 {
     struct net *net = seq_file_net(seq);
     struct ct_iter_state *st = seq->private;
 
     head = rcu_dereference(head->next);
-    while (head == NULL) {
+    while (is_a_nulls(head)) {
         if (++st->bucket >= nf_conntrack_htable_size)
             return NULL;
         head = rcu_dereference(net->ct.hash[st->bucket].first);
@@ -56,9 +56,9 @@ static struct hlist_node *ct_get_next(struct
 seq_file *seq,
     return head;
 }
 
-static struct hlist_node *ct_get_idx(struct seq_file *seq, loff_t pos)
+static struct hlist_nulls_node *ct_get_idx(struct seq_file *seq,
 loff_t pos)
 {
-    struct hlist_node *head = ct_get_first(seq);
+    struct hlist_nulls_node *head = ct_get_first(seq);
 
     if (head)
         while (pos && (head = ct_get_next(seq, head)))
net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c +3 -2
@@ -87,69 +87,76 @@ static void ct_seq_stop(struct seq_file *s, void *v)
 
 static int ct_seq_show(struct seq_file *s, void *v)
 {
-    const struct nf_conntrack_tuple_hash *hash = v;
-    const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(hash);
+    struct nf_conntrack_tuple_hash *hash = v;
+    struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(hash);
     const struct nf_conntrack_l3proto *l3proto;
     const struct nf_conntrack_l4proto *l4proto;
+    int ret = 0;
 
Quoted 9 lines

net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c static void ct_seq_stop(struct seq_file *s, void *v) · line 98

Quoted 3 lines
net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c +2 -0
     NF_CT_ASSERT(ct);
+    if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
+        return 0;

Can we assume the next pointer still points to the next entry in the same chain after the refcount dropped to zero?

We are looking chain N.
If we cannot atomic_inc() refcount, we got some deleted entry.
If we could atomic_inc, we can meet an entry that just moved to another chain X

When hitting its end, we continue the search to the N+1 chain so we only skip the end of previous chain (N). We can 'forget' some entries, we can print several time one given entry.

We could solve this by :

1) Checking hash value : if not one expected ->
Going back to head of chain N, (potentially re-printing already handled entries) So it is not a *perfect* solution.

2) Use a locking to forbid writers (as done in UDP/TCP), but it is expensive and wont solve other problem :

We wont avoid emitting same entry several time anyway (this is a flaw of current seq_file handling, since we 'count' entries to be skiped, and this is wrong if some entries were deleted or inserted meanwhile)

We have same problem on /proc/net/udp & /proc/net/tcp, I am not sure we should care...

Also, current resizing code can give to a /proc/net/ip_conntrack reader a problem, since hash table can switch while its doing its dumping : many entries might be lost or regiven...

#40 lore raw

Re: [PATCH] conntrack: use SLAB_DESTROY_BY_RCU for nf_conn structs

Quoted 22 lines · Eric Dumazet
Quoted 6 lines · Patrick McHardy

Review

Quoted 3 lines
     NF_CT_ASSERT(ct);
(quoted diff) +2 -0
+    if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
+        return 0;

Can we assume the next pointer still points to the next entry in the same chain after the refcount dropped to zero?

We are looking chain N.
If we cannot atomic_inc() refcount, we got some deleted entry.
If we could atomic_inc, we can meet an entry that just moved to another chain X

When hitting its end, we continue the search to the N+1 chain so we only skip the end of previous chain (N). We can 'forget' some entries, we can print several time one given entry.

We could solve this by :

1) Checking hash value : if not one expected ->
Going back to head of chain N, (potentially re-printing already handled entries) So it is not a *perfect* solution.

Quoted 7 lines

2) Use a locking to forbid writers (as done in UDP/TCP), but it is expensive and wont solve other problem :

We wont avoid emitting same entry several time anyway (this is a flaw of current seq_file handling, since we 'count' entries to be skiped, and this is wrong if some entries were deleted or inserted meanwhile)

We have same problem on /proc/net/udp & /proc/net/tcp, I am not sure we should care...

I think double entries are not a problem, as you say, there are already other cases where this can happen. But I think we should try our best that every entry present at the start and still present at the end of a dump is also contained in the dump, otherwise the guantees seem to weak to still be useful. Your first proposal would do exactly that, right?

Quoted 2 lines

Also, current resizing code can give to a /proc/net/ip_conntrack reader a problem, since hash table can switch while its doing its dumping : many entries might be lost or regiven...

Thats true. But its a very rare operation, so I think its mainly a documentation issue.
--
To unsubscribe from this list: send the line "unsubscribe netfilter-devel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#41 lore raw

Re: [PATCH] conntrack: use SLAB_DESTROY_BY_RCU for nf_conn structs

Quoted 38 lines · Patrick McHardy
Quoted 29 lines · Eric Dumazet
Quoted 6 lines · Patrick McHardy

Review

Quoted 3 lines
     NF_CT_ASSERT(ct);
(quoted diff) +2 -0
+    if (unlikely(!atomic_inc_not_zero(&ct->ct_general.use)))
+        return 0;

Can we assume the next pointer still points to the next entry in the same chain after the refcount dropped to zero?

We are looking chain N.
If we cannot atomic_inc() refcount, we got some deleted entry. If we could atomic_inc, we can meet an entry that just moved to another chain X

When hitting its end, we continue the search to the N+1 chain so we only skip the end of previous chain (N). We can 'forget' some entries, we can print
several time one given entry.

We could solve this by :

1) Checking hash value : if not one expected -> Going back to head
of chain N, (potentially re-printing already handled entries)
So it is not a *perfect* solution.

2) Use a locking to forbid writers (as done in UDP/TCP), but it is expensive and
wont solve other problem :

Quoted 7 lines

We wont avoid emitting same entry several time anyway (this is a flaw of current seq_file handling, since we 'count' entries to be skiped, and this is
wrong if some entries were deleted or inserted meanwhile)

We have same problem on /proc/net/udp & /proc/net/tcp, I am not sure we should care...

Quoted 6 lines

I think double entries are not a problem, as you say, there are already other cases where this can happen. But I think we should try our best that every entry present at the start and still present at the end of a dump is also contained in the dump, otherwise the guantees seem to weak to still be useful. Your first proposal would do exactly that, right?

If your concern is to not forget entries, and we are allowed to print some entries several times, then we can just check the final "nulls" value, and if we find a different value than expected for chain N, go back to begining of chain N.

No need to check hash value (this could help not print several time same entry, we dont care that much)

+ while (is_a_nulls(head)) {
+ if (likely(get_nulls_value(head) == st->bucket)) {
+ if (++st->bucket >= nf_conntrack_htable_size)
+ return NULL;
+ }

Thank you

[PATCH] conntrack: use SLAB_DESTROY_BY_RCU and get rid of call_rcu()

Use "hlist_nulls" infrastructure we added in 2.6.29 for RCUification of UDP & TCP.

This permits an easy conversion from call_rcu() based hash lists to a SLAB_DESTROY_BY_RCU one.

Avoiding call_rcu() delay at nf_conn freeing time has numerous gains.

First, it doesnt fill RCU queues (up to 10000 elements per cpu). This reduces OOM possibility, if queued elements are not taken into account This reduces latency problems when RCU queue size hits hilimit and triggers emergency mode.

- It allows fast reuse of just freed elements, permitting better use of CPU cache.

- We delete rcu_head from "struct nf_conn", shrinking size of this structure by 8 or 16 bytes.

This patch only takes care of "struct nf_conn". call_rcu() is still used for less critical conntrack parts, that may be converted later if necessary.

---
 include/net/netfilter/nf_conntrack.h                  |   14 -
 include/net/netfilter/nf_conntrack_tuple.h            |    6
 include/net/netns/conntrack.h                         |    5
 net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c |   63 ++---
 net/ipv4/netfilter/nf_nat_core.c                      |    2
 net/netfilter/nf_conntrack_core.c                     |  123 +++++-----
 net/netfilter/nf_conntrack_expect.c                   |    2
 net/netfilter/nf_conntrack_helper.c                   |    7
 net/netfilter/nf_conntrack_netlink.c                  |   20 -
 net/netfilter/nf_conntrack_standalone.c               |   57 ++--
 net/netfilter/xt_connlimit.c                          |    6
 11 files changed, 174 insertions(+), 131 deletions(-)
include/net/netfilter/nf_conntrack.h +8 -6
index 4dfb793..6c3f964 100644
@@ -91,8 +91,7 @@ struct nf_conn_help {
 #include <net/netfilter/ipv4/nf_conntrack_ipv4.h>
 #include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
 
-struct nf_conn
-{
+struct nf_conn {
 	/* Usage count in here is 1 for hash table/destruct timer, 1 per skb,
            plus 1 for any connection(s) we are `master' for */
 	struct nf_conntrack ct_general;
@@ -126,7 +125,6 @@ struct nf_conn
 #ifdef CONFIG_NET_NS
 	struct net *ct_net;
 #endif
-	struct rcu_head rcu;
 };
 
 static inline struct nf_conn *
@@ -190,9 +188,13 @@ static inline void nf_ct_put(struct nf_conn *ct)
 extern int nf_ct_l3proto_try_module_get(unsigned short l3proto);
 extern void nf_ct_l3proto_module_put(unsigned short l3proto);
 
-extern struct hlist_head *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced);
-extern void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced,
-				 unsigned int size);
+/*
+ * Allocate a hashtable of hlist_head (if nulls == 0),
+ * or hlist_nulls_head (if nulls == 1)
+ */
+extern void *nf_ct_alloc_hashtable(unsigned int *sizep, int *vmalloced, int nulls);
+
+extern void nf_ct_free_hashtable(void *hash, int vmalloced, unsigned int size);
 
 extern struct nf_conntrack_tuple_hash *
 __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple);
include/net/netfilter/nf_conntrack_tuple.h +3 -3
index f2f6aa7..2628c15 100644
@@ -12,6 +12,7 @@
 
 #include <linux/netfilter/x_tables.h>
 #include <linux/netfilter/nf_conntrack_tuple_common.h>
+#include <linux/list_nulls.h>
 
 /* A `tuple' is a structure containing the information to uniquely
   identify a connection.  ie. if two packets have the same tuple, they
@@ -146,9 +147,8 @@ static inline void nf_ct_dump_tuple(const struct nf_conntrack_tuple *t)
 	((enum ip_conntrack_dir)(h)->tuple.dst.dir)
 
 /* Connections have two entries in the hash table: one for each way */
-struct nf_conntrack_tuple_hash
-{
-	struct hlist_node hnode;
+struct nf_conntrack_tuple_hash {
+	struct hlist_nulls_node hnnode;
 	struct nf_conntrack_tuple tuple;
 };
 
include/net/netns/conntrack.h +3 -2
index f4498a6..9dc5840 100644
@@ -2,6 +2,7 @@
 #define __NETNS_CONNTRACK_H
 
 #include <linux/list.h>
+#include <linux/list_nulls.h>
 #include <asm/atomic.h>
 
 struct ctl_table_header;
@@ -10,9 +11,9 @@ struct nf_conntrack_ecache;
 struct netns_ct {
 	atomic_t		count;
 	unsigned int		expect_count;
-	struct hlist_head	*hash;
+	struct hlist_nulls_head	*hash;
 	struct hlist_head	*expect_hash;
-	struct hlist_head	unconfirmed;
+	struct hlist_nulls_head	unconfirmed;
 	struct ip_conntrack_stat *stat;
 #ifdef CONFIG_NF_CONNTRACK_EVENTS
 	struct nf_conntrack_ecache *ecache;
net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c +36 -27

145 lines not on this page · raw message

net/ipv4/netfilter/nf_nat_core.c +1 -1
index a65cf69..fe65187 100644
@@ -679,7 +679,7 @@ nfnetlink_parse_nat_setup(struct nf_conn *ct,
 static int __net_init nf_nat_net_init(struct net *net)
 {
 	net->ipv4.nat_bysource = nf_ct_alloc_hashtable(&nf_nat_htable_size,
-						      &net->ipv4.nat_vmalloced);
+						      &net->ipv4.nat_vmalloced, 0);
 	if (!net->ipv4.nat_bysource)
 		return -ENOMEM;
 	return 0;
net/netfilter/nf_conntrack_core.c +70 -53

319 lines not on this page · raw message

net/netfilter/nf_conntrack_expect.c +1 -1
index 357ba39..3940f99 100644
@@ -604,7 +604,7 @@ int nf_conntrack_expect_init(struct net *net)
 
 	net->ct.expect_count = 0;
 	net->ct.expect_hash = nf_ct_alloc_hashtable(&nf_ct_expect_hsize,
-						  &net->ct.expect_vmalloc);
+						  &net->ct.expect_vmalloc, 0);
 	if (net->ct.expect_hash == NULL)
 		goto err1;
 
net/netfilter/nf_conntrack_helper.c +4 -3
index a51bdac..6066144 100644
@@ -158,6 +158,7 @@ static void __nf_conntrack_helper_unregister(struct nf_conntrack_helper *me,
 	struct nf_conntrack_tuple_hash *h;
 	struct nf_conntrack_expect *exp;
 	const struct hlist_node *n, *next;
+	const struct hlist_nulls_node *nn;
 	unsigned int i;
 
 	/* Get rid of expectations */
@@ -174,10 +175,10 @@ static void __nf_conntrack_helper_unregister(struct nf_conntrack_helper *me,
 	}
 
 	/* Get rid of expecteds, set helpers to NULL. */
-	hlist_for_each_entry(h, n, &net->ct.unconfirmed, hnode)
+	hlist_for_each_entry(h, nn, &net->ct.unconfirmed, hnnode)
 		unhelp(h, me);
 	for (i = 0; i < nf_conntrack_htable_size; i++) {
-		hlist_for_each_entry(h, n, &net->ct.hash[i], hnode)
+		hlist_nulls_for_each_entry(h, nn, &net->ct.hash[i], hnnode)
 			unhelp(h, me);
 	}
 }
@@ -217,7 +218,7 @@ int nf_conntrack_helper_init(void)
 
 	nf_ct_helper_hsize = 1; /* gets rounded up to use one page */
 	nf_ct_helper_hash = nf_ct_alloc_hashtable(&nf_ct_helper_hsize,
-						  &nf_ct_helper_vmalloc);
+						  &nf_ct_helper_vmalloc, 0);
 	if (!nf_ct_helper_hash)
 		return -ENOMEM;
 
net/netfilter/nf_conntrack_netlink.c +11 -9
index 1b75c9e..349bbef 100644
@@ -19,6 +19,7 @@
 #include <linux/module.h>
 #include <linux/kernel.h>
 #include <linux/rculist.h>
+#include <linux/rculist_nulls.h>
 #include <linux/types.h>
 #include <linux/timer.h>
 #include <linux/skbuff.h>
@@ -536,7 +537,7 @@ ctnetlink_dump_table(struct sk_buff *skb, struct netlink_callback *cb)
 {
 	struct nf_conn *ct, *last;
 	struct nf_conntrack_tuple_hash *h;
-	struct hlist_node *n;
+	struct hlist_nulls_node *n;
 	struct nfgenmsg *nfmsg = NLMSG_DATA(cb->nlh);
 	u_int8_t l3proto = nfmsg->nfgen_family;
 
@@ -544,27 +545,27 @@ ctnetlink_dump_table(struct sk_buff *skb, struct netlink_callback *cb)
 	last = (struct nf_conn *)cb->args[1];
 	for (; cb->args[0] < nf_conntrack_htable_size; cb->args[0]++) {
 restart:
-		hlist_for_each_entry_rcu(h, n, &init_net.ct.hash[cb->args[0]],
-					 hnode) {
+		hlist_nulls_for_each_entry_rcu(h, n, &init_net.ct.hash[cb->args[0]],
+					 hnnode) {
 			if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
 				continue;
 			ct = nf_ct_tuplehash_to_ctrack(h);
+			if (!atomic_inc_not_zero(&ct->ct_general.use))
+				continue;
 			/* Dump entries of a given L3 protocol number.
 			 * If it is not specified, ie. l3proto == 0,
 			 * then dump everything. */
 			if (l3proto && nf_ct_l3num(ct) != l3proto)
-				continue;
+				goto releasect;
 			if (cb->args[1]) {
 				if (ct != last)
-					continue;
+					goto releasect;
 				cb->args[1] = 0;
 			}
 			if (ctnetlink_fill_info(skb, NETLINK_CB(cb->skb).pid,
 						cb->nlh->nlmsg_seq,
 						IPCTNL_MSG_CT_NEW,
 						1, ct) < 0) {
-				if (!atomic_inc_not_zero(&ct->ct_general.use))
-					continue;
 				cb->args[1] = (unsigned long)ct;
 				goto out;
 			}
@@ -577,6 +578,8 @@ restart:
 				if (acct)
 					memset(acct, 0, sizeof(struct nf_conn_counter[IP_CT_DIR_MAX]));
 			}
+releasect:
+		nf_ct_put(ct);
 		}
 		if (cb->args[1]) {
 			cb->args[1] = 0;
@@ -1242,13 +1245,12 @@ ctnetlink_create_conntrack(struct nlattr *cda[],
 		if (err < 0)
 			goto err2;
 
-		master_h = __nf_conntrack_find(&init_net, &master);
+		master_h = nf_conntrack_find_get(&init_net, &master);
 		if (master_h == NULL) {
 			err = -ENOENT;
 			goto err2;
 		}
 		master_ct = nf_ct_tuplehash_to_ctrack(master_h);
-		nf_conntrack_get(&master_ct->ct_general);
 		__set_bit(IPS_EXPECTED_BIT, &ct->status);
 		ct->master = master_ct;
 	}
net/netfilter/nf_conntrack_standalone.c +33 -24

140 lines not on this page · raw message

net/netfilter/xt_connlimit.c +4 -2
index 7f404cc..6809809 100644
@@ -108,7 +108,7 @@ static int count_them(struct xt_connlimit_data *data,
 	const struct nf_conntrack_tuple_hash *found;
 	struct xt_connlimit_conn *conn;
 	struct xt_connlimit_conn *tmp;
-	const struct nf_conn *found_ct;
+	struct nf_conn *found_ct;
 	struct list_head *hash;
 	bool addit = true;
 	int matches = 0;
@@ -123,7 +123,7 @@ static int count_them(struct xt_connlimit_data *data,
 
 	/* check the saved connections */
 	list_for_each_entry_safe(conn, tmp, hash, list) {
-		found    = __nf_conntrack_find(&init_net, &conn->tuple);
+		found    = nf_conntrack_find_get(&init_net, &conn->tuple);
 		found_ct = NULL;
 
 		if (found != NULL)
@@ -151,6 +151,7 @@ static int count_them(struct xt_connlimit_data *data,
 			 * we do not care about connections which are
 			 * closed already -> ditch it
 			 */
+			nf_ct_put(found_ct);
 			list_del(&conn->list);
 			kfree(conn);
 			continue;
@@ -160,6 +161,7 @@ static int count_them(struct xt_connlimit_data *data,
 		    match->family))
 			/* same source network -> be counted! */
 			++matches;
+		nf_ct_put(found_ct);
 	}
 
 	rcu_read_unlock();

--
To unsubscribe from this list: send the line "unsubscribe netfilter-devel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#42 lore raw

Re: [PATCH] conntrack: use SLAB_DESTROY_BY_RCU for nf_conn structs

Quoted 14 lines · Eric Dumazet
Quoted 6 lines · Patrick McHardy

I think double entries are not a problem, as you say, there are already other cases where this can happen. But I think we should try our best that every entry present at the start and still present at the end of a dump is also contained in the dump, otherwise the guantees seem to weak to still be useful. Your first proposal would do exactly that, right?

If your concern is to not forget entries, and we are allowed to print some entries several times, then we can just check the final "nulls" value, and if we find a different value than expected for chain N, go back to begining of chain N.

No need to check hash value (this could help not print several time same entry, we dont care that much)

net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c struct ct_iter_state { · line 55

Quoted 5 lines
net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c +5 -0
+       while (is_a_nulls(head)) {
+               if (likely(get_nulls_value(head) == st->bucket)) {
+                       if (++st->bucket >= nf_conntrack_htable_size)
+                               return NULL;
+               }

Looks perfect :)

Quoted 17 lines

[PATCH] conntrack: use SLAB_DESTROY_BY_RCU and get rid of call_rcu()

Use "hlist_nulls" infrastructure we added in 2.6.29 for RCUification of UDP & TCP.

This permits an easy conversion from call_rcu() based hash lists to a SLAB_DESTROY_BY_RCU one.

Avoiding call_rcu() delay at nf_conn freeing time has numerous gains.

First, it doesnt fill RCU queues (up to 10000 elements per cpu). This reduces OOM possibility, if queued elements are not taken into account This reduces latency problems when RCU queue size hits hilimit and triggers emergency mode.

- It allows fast reuse of just freed elements, permitting better use of CPU cache.

Quoted 6 lines

- We delete rcu_head from "struct nf_conn", shrinking size of this structure by 8 or 16 bytes.

This patch only takes care of "struct nf_conn". call_rcu() is still used for less critical conntrack parts, that may be converted later if necessary.

Applied, thanks a lot.

#43 lore raw

Re: [PATCH] conntrack: Reduce conntrack count in nf_conntrack_free()

In reply to Maxime Bizon

Maxime Bizon <[email protected]> wrote on 24/03/2009 16:17:30:

Quoted 16 lines · Eric Dumazet

Hi Eric,

Quoted 6 lines

We use RCU to defer freeing of conntrack structures. In DOS situation, RCU might accumulate about 10.000 elements per CPU in its internal queues. To get accurate conntrack counts (at the expense of slightly more RAM used), we might consider conntrack counter not taking into account "about to be freed elements, waiting in RCU queues". We thus decrement it in nf_conntrack_free(), not in the RCU callback.

Your patch fixes the problem on my board too (embedded mips router 250Mhz), thanks.

Quoted 6 lines

Yet I'm concerned about what you said concerning RAM usage. I have a very small amount on memory left on my board (less than 4M), and I tuned ip route cache size and nf_conntrack_max to make sure I won't go OOM.

With your patch, does it mean 10000 conntrack entries can be allocated while nf_conntrack_max is say only 2048 ?

Just add "rcuclassic.qhimark=2048" to your cmdline.

Jocke

#44 lore raw
In reply to Joakim Tjernlund
Quoted 35 lines · Joakim Tjernlund

Patrick McHardy <[email protected]> wrote on 23/03/2009 13:29:33:

Quoted 4 lines
Quoted 3 lines · Joakim Tjernlund

Patrick McHardy <[email protected]> wrote on 23/03/2009 13:15:45:

Quoted 2 lines
Quoted 1 line · Joakim Tjernlund

doing a "ping -f -l 3" on my host towards my board on linus tree as

of

Quoted 12 lines
Quoted 12 lines
Quoted 12 lines
Quoted 12 lines

Friday results in lots of:
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet. __ratelimit: 11 callbacks suppressed
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.

for ucc_geth on a MPC832x.
This really looks strange to me, ideas?

Quoted 8 lines
Quoted 4 lines
Quoted 1 line

What does /proc/net/netfilter/nf_conntrack show?

There is no /proc/net/netfilter/nf_conntrack. There is a /proc/net/nf_conntrack though and it is empty. If I telnet to the board I see:

That means that something is leaking conntrack references, most likely by leaking skbs. Since I haven't seen any other reports, my guess would be the ucc_geth driver.

Mucking around with the ucc_geth driver I found that if I:
- Move TX from IRQ to NAPI context
- double the weight.
- after booting up, wait a few mins until the JFFS2 GC kernel thread has
stopped
scanning the FS

Quoted 3 lines

Then the "nf_conntrack: table full, dropping packet." msgs stops. Does this seem right to you guys?

How many cpus do you have ?

What kernel version do you use ?

#45 lore raw

Eric Dumazet <[email protected]> wrote on 23/03/2009 18:49:38:

Quoted 8 lines
Quoted 6 lines · Joakim Tjernlund

Patrick McHardy <[email protected]> wrote on 23/03/2009 13:29:33:

Quoted 4 lines
Quoted 3 lines · Joakim Tjernlund

Patrick McHardy <[email protected]> wrote on 23/03/2009 13:15:45:

Quoted 2 lines
Quoted 1 line · Joakim Tjernlund

doing a "ping -f -l 3" on my host towards my board on linus tree

as

Quoted 11 lines
Quoted 11 lines

of

Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines

Friday results in lots of:
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet. __ratelimit: 11 callbacks suppressed
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 6 lines
Quoted 3 lines
Quoted 2 lines

for ucc_geth on a MPC832x.
This really looks strange to me, ideas?

What does /proc/net/netfilter/nf_conntrack show?

There is no /proc/net/netfilter/nf_conntrack. There is a /proc/net/nf_conntrack though and it is empty. If I telnet to the board I see:

That means that something is leaking conntrack references, most

likely

Quoted 1 line
Quoted 1 line
Quoted 1 line

by leaking skbs. Since I haven't seen any other reports, my guess

would

Quoted 7 lines
Quoted 7 lines
Quoted 2 lines

be the ucc_geth driver.

Mucking around with the ucc_geth driver I found that if I:
- Move TX from IRQ to NAPI context
- double the weight.
- after booting up, wait a few mins until the JFFS2 GC kernel thread

has

Quoted 8 lines
Quoted 6 lines

stopped
scanning the FS

Then the "nf_conntrack: table full, dropping packet." msgs stops. Does this seem right to you guys?

How many cpus do you have ?

Just one, it is an embedded board running at 266 MHz

Quoted 2 lines

What kernel version do you use ?

Linus tree as of Friday

#46 lore raw
Quoted 44 lines · Joakim Tjernlund

Eric Dumazet <[email protected]> wrote on 23/03/2009 18:49:38:

Quoted 7 lines
Quoted 6 lines · Joakim Tjernlund

Patrick McHardy <[email protected]> wrote on 23/03/2009 13:29:33:

Quoted 4 lines
Quoted 3 lines · Joakim Tjernlund

Patrick McHardy <[email protected]> wrote on 23/03/2009 13:15:45:

Quoted 2 lines
Quoted 1 line · Joakim Tjernlund

doing a "ping -f -l 3" on my host towards my board on linus tree

as

Quoted 11 lines
Quoted 11 lines

of

Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines

Friday results in lots of:
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet. __ratelimit: 11 callbacks suppressed
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.
nf_conntrack: table full, dropping packet.

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 6 lines
Quoted 3 lines
Quoted 2 lines

for ucc_geth on a MPC832x.
This really looks strange to me, ideas?

What does /proc/net/netfilter/nf_conntrack show?

There is no /proc/net/netfilter/nf_conntrack. There is a /proc/net/nf_conntrack though and it is empty. If I telnet to the board I see:

That means that something is leaking conntrack references, most

likely

Quoted 1 line
Quoted 1 line
Quoted 1 line

by leaking skbs. Since I haven't seen any other reports, my guess

would

Quoted 6 lines
Quoted 6 lines
Quoted 2 lines

be the ucc_geth driver.

Mucking around with the ucc_geth driver I found that if I:
- Move TX from IRQ to NAPI context
- double the weight.
- after booting up, wait a few mins until the JFFS2 GC kernel thread

has

Quoted 6 lines
Quoted 6 lines

stopped
scanning the FS

Then the "nf_conntrack: table full, dropping packet." msgs stops. Does this seem right to you guys?

Quoted 9 lines
Quoted 1 line

How many cpus do you have ?

Just one, it is an embedded board running at 266 MHz

Quoted 1 line

What kernel version do you use ?

Linus tree as of Friday

I suspect RCU problem. Maybe the GC kernel threads blocks synchronize_rcu() ?