[stable regression] tcp: make connect() mem charging friendly

[stable regression] tcp: make connect() mem charging friendly

[Ben Hutchings]

[-- Attachment #1: Type: text/plain, Size: 1318 bytes --]

Commit 355a901e6cf1b2b763ec85caa2a9f04fbcc4ab4a ("tcp: make connect()
mem charging friendly") was backported to various stable branches:

v3.10.73: e64a85197b3f tcp: make connect() mem charging friendly
v3.12.40: d06381e8aac5 tcp: make connect() mem charging friendly
v3.14.37: 5a8e8f482b4a tcp: make connect() mem charging friendly
v3.18.11: e8f117f002ca tcp: make connect() mem charging friendly
v3.13.11-ckt19: de023863df9d tcp: make connect() mem charging friendly
v3.16.7-ckt9: bea5f6ef9fcb tcp: make connect() mem charging friendly

On the 3.16 branch, this has resulted in a regression for TCP Fast Open:
<https://bugs.debian.org/782515>.  The BUG() at the top of
tcp_transmit_skb() fires as tcp_skb_pcount(skb) == 0.

tcp_send_syn_data() does:

	memcpy(syn_data->cb, syn->cb, sizeof(syn->cb));

Since commit cd7d8498c9a5 ("tcp: change tcp_skb_pcount() location") this
is sufficient to set the GSO segment count correctly.  But in older
branches (< 3.18) the GSO segment count in skb_shared_info is used and
is no longer copied by tcp_send_syn_data().

All the versions listed above, except v3.18.11, appear to have suffered
this regression.

Ben.

-- 
Ben Hutchings
Editing code like this is akin to sticking plasters on the bleeding stump
of a severed limb. - me, 29 June 1999

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 811 bytes --]

Re: [stable regression] tcp: make connect() mem charging friendly

[Eric Dumazet]

On Tue, 2015-04-14 at 08:32 +0100, Ben Hutchings wrote:
> Commit 355a901e6cf1b2b763ec85caa2a9f04fbcc4ab4a ("tcp: make connect()
> mem charging friendly") was backported to various stable branches:
> 
> v3.10.73: e64a85197b3f tcp: make connect() mem charging friendly
> v3.12.40: d06381e8aac5 tcp: make connect() mem charging friendly
> v3.14.37: 5a8e8f482b4a tcp: make connect() mem charging friendly
> v3.18.11: e8f117f002ca tcp: make connect() mem charging friendly
> v3.13.11-ckt19: de023863df9d tcp: make connect() mem charging friendly
> v3.16.7-ckt9: bea5f6ef9fcb tcp: make connect() mem charging friendly
> 
> On the 3.16 branch, this has resulted in a regression for TCP Fast Open:
> <https://bugs.debian.org/782515>.  The BUG() at the top of
> tcp_transmit_skb() fires as tcp_skb_pcount(skb) == 0.
> 
> tcp_send_syn_data() does:
> 
> 	memcpy(syn_data->cb, syn->cb, sizeof(syn->cb));
> 
> Since commit cd7d8498c9a5 ("tcp: change tcp_skb_pcount() location") this
> is sufficient to set the GSO segment count correctly.  But in older
> branches (< 3.18) the GSO segment count in skb_shared_info is used and
> is no longer copied by tcp_send_syn_data().
> 
> All the versions listed above, except v3.18.11, appear to have suffered
> this regression.
> 
> Ben.
> 

Hi Ben

Sorry to ear that. Have you already fixed this, or are you asking for
help ?

Re: [stable regression] tcp: make connect() mem charging friendly

[Ben Hutchings]

[-- Attachment #1: Type: text/plain, Size: 1761 bytes --]

On Tue, 2015-04-14 at 05:19 -0700, Eric Dumazet wrote:
> On Tue, 2015-04-14 at 08:32 +0100, Ben Hutchings wrote:
> > Commit 355a901e6cf1b2b763ec85caa2a9f04fbcc4ab4a ("tcp: make connect()
> > mem charging friendly") was backported to various stable branches:
> > 
> > v3.10.73: e64a85197b3f tcp: make connect() mem charging friendly
> > v3.12.40: d06381e8aac5 tcp: make connect() mem charging friendly
> > v3.14.37: 5a8e8f482b4a tcp: make connect() mem charging friendly
> > v3.18.11: e8f117f002ca tcp: make connect() mem charging friendly
> > v3.13.11-ckt19: de023863df9d tcp: make connect() mem charging friendly
> > v3.16.7-ckt9: bea5f6ef9fcb tcp: make connect() mem charging friendly
> > 
> > On the 3.16 branch, this has resulted in a regression for TCP Fast Open:
> > <https://bugs.debian.org/782515>.  The BUG() at the top of
> > tcp_transmit_skb() fires as tcp_skb_pcount(skb) == 0.
> > 
> > tcp_send_syn_data() does:
> > 
> > 	memcpy(syn_data->cb, syn->cb, sizeof(syn->cb));
> > 
> > Since commit cd7d8498c9a5 ("tcp: change tcp_skb_pcount() location") this
> > is sufficient to set the GSO segment count correctly.  But in older
> > branches (< 3.18) the GSO segment count in skb_shared_info is used and
> > is no longer copied by tcp_send_syn_data().
> > 
> > All the versions listed above, except v3.18.11, appear to have suffered
> > this regression.
> > 
> > Ben.
> > 
> 
> Hi Ben
> 
> Sorry to ear that. Have you already fixed this, or are you asking for
> help ?

That was as far as I got this morning.  I can probably come up with a
fix later if you don't sooner.

Ben.

-- 
Ben Hutchings
Editing code like this is akin to sticking plasters on the bleeding stump
of a severed limb. - me, 29 June 1999

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 811 bytes --]