[PATCH 3/5] inet_diag: Extend the meminfo report

[PATCH 3/5] inet_diag: Extend the meminfo report

[Pavel Emelyanov]

Add the backlog length and wmem and rmem buffers limits.
With this the TCP memory info is full and can be analyzed.

Few words about extending the existing structure. The nla engine stores lengths of
fields and the nla parser will parse the new format successfully (checked on the ss
tool). But new apps will have to take this field's length into account. Hopefully
this is acceptable price for not introducing another ext bit and structure.

Singed-off-by: Pavel Emelyanov <xemul@parallels.com>

---
 include/linux/inet_diag.h |    3 +++
 net/ipv4/inet_diag.c      |    3 +++
 2 files changed, 6 insertions(+), 0 deletions(-)

diff --git a/include/linux/inet_diag.h b/include/linux/inet_diag.h
index afa5d5c..d274712 100644
--- a/include/linux/inet_diag.h
+++ b/include/linux/inet_diag.h
@@ -120,6 +120,9 @@ struct inet_diag_meminfo {
 	__u32	idiag_wmem;
 	__u32	idiag_fmem;
 	__u32	idiag_tmem;
+	__u32	idiag_backlog;
+	__u32	idiag_rcvbuf;
+	__u32	idiag_sndbuf;
 };
 
 /* INET_DIAG_VEGASINFO */
diff --git a/net/ipv4/inet_diag.c b/net/ipv4/inet_diag.c
index fb2e47f..91f9092 100644
--- a/net/ipv4/inet_diag.c
+++ b/net/ipv4/inet_diag.c
@@ -134,6 +134,9 @@ int inet_sk_diag_fill(struct sock *sk, struct inet_connection_sock *icsk,
 		minfo->idiag_wmem = sk->sk_wmem_queued;
 		minfo->idiag_fmem = sk->sk_forward_alloc;
 		minfo->idiag_tmem = sk_wmem_alloc_get(sk);
+		minfo->idiag_backlog = sk->sk_backlog.len;
+		minfo->idiag_rcvbuf = sk->sk_rcvbuf;
+		minfo->idiag_sndbuf = sk->sk_sndbuf;
 	}
 
 	if (icsk == NULL) {
-- 
1.5.5.6

Re: [PATCH 3/5] inet_diag: Extend the meminfo report

[David Miller]

From: Pavel Emelyanov <xemul@parallels.com>
Date: Wed, 28 Dec 2011 17:38:40 +0400

> Add the backlog length and wmem and rmem buffers limits.
> With this the TCP memory info is full and can be analyzed.
> 
> Few words about extending the existing structure. The nla engine stores lengths of
> fields and the nla parser will parse the new format successfully (checked on the ss
> tool). But new apps will have to take this field's length into account. Hopefully
> this is acceptable price for not introducing another ext bit and structure.
> 
> Singed-off-by: Pavel Emelyanov <xemul@parallels.com>

I think you'll need to add a new extension, rather than create this incompatability.

Re: [PATCH 3/5] inet_diag: Extend the meminfo report

[Pavel Emelyanov]

On 12/28/2011 10:45 PM, David Miller wrote:
> From: Pavel Emelyanov <xemul@parallels.com>
> Date: Wed, 28 Dec 2011 17:38:40 +0400
> 
>> Add the backlog length and wmem and rmem buffers limits.
>> With this the TCP memory info is full and can be analyzed.
>>
>> Few words about extending the existing structure. The nla engine stores lengths of
>> fields and the nla parser will parse the new format successfully (checked on the ss
>> tool). But new apps will have to take this field's length into account. Hopefully
>> this is acceptable price for not introducing another ext bit and structure.
>>
>> Singed-off-by: Pavel Emelyanov <xemul@parallels.com>
> 
> I think you'll need to add a new extension, rather than create this incompatability.
> 

OK, I will, but I seem to misunderstand the NLA's length value then. My impression is
that it was introduced to make it possible to put the payload of different length,
probably adding more stuff at the end.

If I'm wrong about it, what is the real purpose of this field? _Only_ to make nla_parse
jump on the "generic" offsets?

Thanks,
Pavel

Re: [PATCH 3/5] inet_diag: Extend the meminfo report

[David Miller]

From: Pavel Emelyanov <xemul@parallels.com>
Date: Wed, 28 Dec 2011 22:58:57 +0400

> On 12/28/2011 10:45 PM, David Miller wrote:
>> From: Pavel Emelyanov <xemul@parallels.com>
>> Date: Wed, 28 Dec 2011 17:38:40 +0400
>> 
>>> Add the backlog length and wmem and rmem buffers limits.
>>> With this the TCP memory info is full and can be analyzed.
>>>
>>> Few words about extending the existing structure. The nla engine stores lengths of
>>> fields and the nla parser will parse the new format successfully (checked on the ss
>>> tool). But new apps will have to take this field's length into account. Hopefully
>>> this is acceptable price for not introducing another ext bit and structure.
>>>
>>> Singed-off-by: Pavel Emelyanov <xemul@parallels.com>
>> 
>> I think you'll need to add a new extension, rather than create this incompatability.
>> 
> 
> OK, I will, but I seem to misunderstand the NLA's length value then. My impression is
> that it was introduced to make it possible to put the payload of different length,
> probably adding more stuff at the end.

It's to validate that the object inside the attribute has the expected length.