2006-01-10 22:03:57

by Ulrich Drepper

[permalink] [raw]
Subject: ntohs/ntohl and bitops

I just saw this in a patch:

+ if (ntohs(ih->frag_off) & IP_OFFSET)
+ return EBT_NOMATCH;

This isn't optimal, it requires a byte switch little endian machines.
The compiler isn't smart enough. It would be better to use

if (ih->frag_off & ntohs(IP_OFFSET))

where the byte-swap can be done at compile time. This is kind of ugly,
I guess, so maybe a dedicate macro

net_host_bit_p(ih->frag_off, IP_OFFSET)

??

--
➧ Ulrich Drepper ➧ Red Hat, Inc. ➧ 444 Castro St ➧ Mountain View, CA ❖


Attachments:
signature.asc (251.00 B)
OpenPGP digital signature

2006-01-11 08:00:18

by David Miller

[permalink] [raw]
Subject: Re: ntohs/ntohl and bitops

From: Ulrich Drepper <[email protected]>
Date: Tue, 10 Jan 2006 14:02:52 -0800

> I just saw this in a patch:
>
> + if (ntohs(ih->frag_off) & IP_OFFSET)
> + return EBT_NOMATCH;
>
> This isn't optimal, it requires a byte switch little endian machines.
> The compiler isn't smart enough. It would be better to use
>
> if (ih->frag_off & ntohs(IP_OFFSET))
>
> where the byte-swap can be done at compile time. This is kind of ugly,
> I guess, so maybe a dedicate macro
>
> net_host_bit_p(ih->frag_off, IP_OFFSET)

The first suggestion isn't considered ugly, and the best form is:

if (ih->frag_off & __constant_htons(IP_OFFSET))

I'll fix that up when I get a chance, thanks for catching it Uli.

2006-01-11 08:13:23

by Arjan van de Ven

[permalink] [raw]
Subject: Re: ntohs/ntohl and bitops

On Wed, 2006-01-11 at 00:00 -0800, David S. Miller wrote:
> From: Ulrich Drepper <[email protected]>
> Date: Tue, 10 Jan 2006 14:02:52 -0800
>
> > I just saw this in a patch:
> >
> > + if (ntohs(ih->frag_off) & IP_OFFSET)
> > + return EBT_NOMATCH;
> >
> > This isn't optimal, it requires a byte switch little endian machines.
> > The compiler isn't smart enough. It would be better to use
> >
> > if (ih->frag_off & ntohs(IP_OFFSET))
> >
> > where the byte-swap can be done at compile time. This is kind of ugly,
> > I guess, so maybe a dedicate macro
> >
> > net_host_bit_p(ih->frag_off, IP_OFFSET)
>
> The first suggestion isn't considered ugly, and the best form is:
>
> if (ih->frag_off & __constant_htons(IP_OFFSET))

why this __constant_htons and not just plain htons ??
htons() gets auto-remapped to that anyway via the builtin "is this a
constant" thing...... and to be honest htons() is more readable.


2006-01-11 08:36:14

by David Miller

[permalink] [raw]
Subject: Re: ntohs/ntohl and bitops

From: Arjan van de Ven <[email protected]>
Date: Wed, 11 Jan 2006 09:13:11 +0100

> > The first suggestion isn't considered ugly, and the best form is:
> >
> > if (ih->frag_off & __constant_htons(IP_OFFSET))
>
> why this __constant_htons and not just plain htons ??
> htons() gets auto-remapped to that anyway via the builtin "is this a
> constant" thing...... and to be honest htons() is more readable.

You're right.

We use the __constant_*() things for structure initialization
which needs to be compile time.

2006-01-11 08:36:04

by Ulrich Drepper

[permalink] [raw]
Subject: Re: ntohs/ntohl and bitops

Arjan van de Ven wrote:
> why this __constant_htons and not just plain htons ??
> htons() gets auto-remapped to that anyway via the builtin "is this a
> constant" thing...... and to be honest htons() is more readable.

Indeed, why the need for the uglification?

Anyway, candidates for this kind of transformation:

net/ipx/af_ipx.c:1450: if (ntohs(addr->sipx_port) <
IPX_MIN_EPHEMERAL_SOCKET &&
net/atm/br2684.c:308: if (ntohs(eth->h_proto) >= 1536)
net/ipv6/netfilter/ip6t_LOG.c:114: if
(ntohs(fh->frag_off) & 0xFFF8)
net/ipv6/exthdrs_core.c:89: if (ntohs(*fp) & ~0x7)
net/ipv4/ipmr.c:1182: if (skb->len+encap > dst_mtu(&rt->u.dst) &&
(ntohs(iph->frag_off) & IP_DF)) {
net/ipv4/netfilter/ip_conntrack_helper_pptp.c:710: if
(ntohs(pptph->packetType) != PPTP_PACKET_CONTROL ||
net/ipv4/netfilter/ipt_LOG.c:70: if (ntohs(ih->frag_off) & IP_CE)
net/ipv4/netfilter/ipt_LOG.c:72: if (ntohs(ih->frag_off) & IP_DF)
net/ipv4/netfilter/ipt_LOG.c:74: if (ntohs(ih->frag_off) & IP_MF)
net/ipv4/netfilter/ipt_LOG.c:78: if (ntohs(ih->frag_off) & IP_OFFSET)
net/ipv4/netfilter/ipt_LOG.c:108: if (ntohs(ih->frag_off)
& IP_OFFSET)
net/ipv4/netfilter/ipt_LOG.c:180: if (ntohs(ih->frag_off)
& IP_OFFSET)
net/ipv4/netfilter/ipt_LOG.c:221: if (ntohs(ih->frag_off)
& IP_OFFSET)
net/ipv4/netfilter/ipt_LOG.c:286: if (ntohs(ih->frag_off)
& IP_OFFSET)
net/ipv4/netfilter/ipt_LOG.c:311: if (ntohs(ih->frag_off)
& IP_OFFSET)
net/bridge/netfilter/ebt_ip.c:55: if (ntohs(ih->frag_off)
& IP_OFFSET)
net/sunrpc/auth_gss/svcauth_gss.c:793: if (ntohl(svc_getu32(argv)) !=
RPC_GSS_VERSION)
net/sunrpc/auth_gss/svcauth_gss.c:823: if
(ntohl(svc_getu32(argv)) != RPC_AUTH_NULL)
net/sunrpc/auth_gss/svcauth_gss.c:825: if
(ntohl(svc_getu32(argv)) != 0)
net/bluetooth/bnep/core.c:343: if (ntohs(s->eh.h_proto) == 0x8100) {


(svcauth_gss.c:825 is a good one).



For consistency reasons, ntohs should be changed to htons here:

net/ipv6/netfilter/ip6t_eui64.c:43: if (eth_hdr(skb)->h_proto ==
ntohs(ETH_P_IPV6)) {


Around net/ipv4/ipconfig.c:376 you might want to store the transformed
ic_myaddr in a variable instead of repeating the ntohl call.

--
➧ Ulrich Drepper ➧ Red Hat, Inc. ➧ 444 Castro St ➧ Mountain View, CA ❖


Attachments:
signature.asc (251.00 B)
OpenPGP digital signature

2006-01-11 08:44:17

by David Miller

[permalink] [raw]
Subject: Re: ntohs/ntohl and bitops

From: Ulrich Drepper <[email protected]>
Date: Wed, 11 Jan 2006 00:36:11 -0800

> Anyway, candidates for this kind of transformation:
>
> net/ipx/af_ipx.c:1450: if (ntohs(addr->sipx_port) <
> IPX_MIN_EPHEMERAL_SOCKET &&

Does it work for comparisons? F.e.:

if (val < ntohs(VAL))



2006-01-11 09:49:12

by Jakub Jelinek

[permalink] [raw]
Subject: Re: ntohs/ntohl and bitops

On Wed, Jan 11, 2006 at 12:44:18AM -0800, David S. Miller wrote:
> From: Ulrich Drepper <[email protected]>
> Date: Wed, 11 Jan 2006 00:36:11 -0800
>
> > Anyway, candidates for this kind of transformation:
> >
> > net/ipx/af_ipx.c:1450: if (ntohs(addr->sipx_port) <
> > IPX_MIN_EPHEMERAL_SOCKET &&
>
> Does it work for comparisons? F.e.:
>
> if (val < ntohs(VAL))

No, only for ==, !=, &, |, ~.

Jakub

2006-01-11 18:35:55

by Jan Engelhardt

[permalink] [raw]
Subject: Re: ntohs/ntohl and bitops

>> Does it work for comparisons? F.e.:
>>
>> if (val < ntohs(VAL))
>
>No, only for ==, !=, &, |, ~.

And ^?


Jan Engelhardt
--

2006-01-11 18:45:19

by Ulrich Drepper

[permalink] [raw]
Subject: Re: ntohs/ntohl and bitops

Jan Engelhardt wrote:
> And ^?

^ is fine.

And just to be complete: I sent DaveM the correct line for that file in
question, I just C&Ped the wrong line my scripts produced. The mail
seem not to make it to lkml. In fact, none of my mail originated from
the gmail account ever made it. I don't know what filter doesn't like me.

--
➧ Ulrich Drepper ➧ Red Hat, Inc. ➧ 444 Castro St ➧ Mountain View, CA ❖


Attachments:
signature.asc (251.00 B)
OpenPGP digital signature

2006-01-11 20:19:15

by David Miller

[permalink] [raw]
Subject: Re: ntohs/ntohl and bitops

From: Ulrich Drepper <[email protected]>
Date: Wed, 11 Jan 2006 10:45:00 -0800

> And just to be complete: I sent DaveM the correct line for that file in
> question, I just C&Ped the wrong line my scripts produced. The mail
> seem not to make it to lkml. In fact, none of my mail originated from
> the gmail account ever made it. I don't know what filter doesn't like me.

This one:

m/^Content-Transfer-Encoding:\s*base64/i

2006-01-14 09:33:54

by Pavel Machek

[permalink] [raw]
Subject: Re: ntohs/ntohl and bitops


On Wed 11-01-06 00:00:20, David S. Miller wrote:
> From: Ulrich Drepper <[email protected]>
> Date: Tue, 10 Jan 2006 14:02:52 -0800
>
> > I just saw this in a patch:
> >
> > + if (ntohs(ih->frag_off) & IP_OFFSET)
> > + return EBT_NOMATCH;
> >
> > This isn't optimal, it requires a byte switch little endian machines.
> > The compiler isn't smart enough. It would be better to use
> >
> > if (ih->frag_off & ntohs(IP_OFFSET))
> >
> > where the byte-swap can be done at compile time. This is kind of ugly,
> > I guess, so maybe a dedicate macro
> >
> > net_host_bit_p(ih->frag_off, IP_OFFSET)
>
> The first suggestion isn't considered ugly, and the best form is:
>
> if (ih->frag_off & __constant_htons(IP_OFFSET))
>
> I'll fix that up when I get a chance, thanks for catching it Uli.

Could you possibly
#define IP_OFFSET htons(1234)
?

Noone should need it in native endianity, no?

--
Thanks, Sharp!

2006-01-14 09:51:30

by YOSHIFUJI Hideaki

[permalink] [raw]
Subject: Re: ntohs/ntohl and bitops

In article <[email protected]> (at Thu, 12 Jan 2006 01:04:06 +0000), Pavel Machek <[email protected]> says:

>
> On Wed 11-01-06 00:00:20, David S. Miller wrote:
> > From: Ulrich Drepper <[email protected]>
> > Date: Tue, 10 Jan 2006 14:02:52 -0800
> >
> > > I just saw this in a patch:
> > >
> > > + if (ntohs(ih->frag_off) & IP_OFFSET)
> > > + return EBT_NOMATCH;
> > >
> > > This isn't optimal, it requires a byte switch little endian machines.
> > > The compiler isn't smart enough. It would be better to use
> > >
> > > if (ih->frag_off & ntohs(IP_OFFSET))
:
> Could you possibly
> #define IP_OFFSET htons(1234)
> ?

In this case, you should use __constant_htons().
I still prefer:
if (ih->frag_off & htons(IP_OFFSET))
though.

--yoshfuji

2006-01-15 01:50:06

by David Miller

[permalink] [raw]
Subject: Re: ntohs/ntohl and bitops

From: Pavel Machek <[email protected]>
Date: Thu, 12 Jan 2006 01:04:06 +0000

> Could you possibly
> #define IP_OFFSET htons(1234)
> ?
>
> Noone should need it in native endianity, no?

That's definitely going to be error prone.
And I bet the BSD headers define it in cpu endainness
as well.