Message-ID: <wf7rcischdpj6vbmchen6g7qynemfhyhgfx5uiezyun65qm5oi@3icznxytufar>
Date: Tue, 18 Nov 2025 15:31:09 +0100
From: Alejandro Colomar <alx@...nel.org>
To: Jakub Głogowski <not@...dz.net>
Cc: linux-man@...r.kernel.org, LKML <linux-kernel@...r.kernel.org>, 
	Linux API <linux-api@...r.kernel.org>, ej@...i.de
Subject: Re: [PATCH 1/2] man/man7/ip.7: Clarify PKTINFO's semantics depending
 on packet direction

Hi Jakub,

On Fri, Nov 14, 2025 at 03:29:30PM +0100, Jakub Głogowski wrote:
> For recvmsg(2), ipi_spec_dst is set by ipv4_pktinfo_prepare() to the
> result of fib_compute_sec_dst().  The latter was introduced in
> 	linux.git 35ebf65e851c6d97 ("ipv4: Create and use fib_compute_spec_dst() helper.").
> 
> Quoting its commit message:
> 
> > The specific destination is the host we direct unicast replies to.
> > Usually this is the original packet source address, but if we are
> > responding to a multicast or broadcast packet we have to use something
> > different.
> >
> > Specifically we must use the source address we would use if we were to
> > send a packet to the unicast source of the original packet.
> 
> Experimentation seems to confirm that behavior.
> 
> As for the note about ipi_spec_dst being on a different interface:
> - For unicast packets (for which ipi_spec_dst is the original
>   destination address), I believe this is trivially true because Linux
>   uses the weak host model (unless there's some interaction with
>   RTCF_LOCAL that I'm missing).
> - For multicast/broadcast packets, fib_compute_sec_dst() only passes the
>   original interface to the lookup in the context of L3M.  In
>   particular, the original implementation (cited above) set iif and oof
>   to 0. Also, citing
> 	linux.git e7372197e15856ec ("net/ipv4: Set oif in fib_compute_spec_dst"),
>   > If the device is not enslaved, oif is still 0 so no affect.
> 
> It doesn't seem like using an address specifically from the interface
> the packet was received on was ever the intention.  I've also confirmed
> this behavior (sending a multicast packet from another machine, whose IP
> I've routed to a dummy interface).
> 
> I'm focusing on this because that's a misconception I've had before
> digging into the code - the sendmsg behavior explained in the same
> paragraph made me think ipi_spec_dst was the (primary?) address of
> ipi_ifindex.  I think this is worth clarifying.
> 
> I've made it explicit that ipi_addr isn't used by sendmsg because that's
> another possible misconception.
> 
> The (first) extra comma in sendmsg's ipi_spec_dst's description is meant
> to emphasize that it's used as the local source address _and_ for the
> routing table lookup, as opposed to just affecting the routing table
> lookup.
> Stylistically it might be a bit weird but idk how to convey this better.
> 
> Apart from the cited commits I was referencing the linux-6.17.7 tarball.
> 
> __fib_validate_source (and the comment near it) might also be of
> interest to people trying to figure out what "specific destinations"
> are, exactly.
> 
> Signed-off-by: Jakub Głogowski <not@...dz.net>

Thanks!  I've applied the patch.  I've added CC tags (please, copy those
yourself in future patches).

I've also s/PKTINFO/IP_PKTINFO/ in the subject.

And I've applied minor wording and source improvements.

I've pushed here:
<https://www.alejandro-colomar.es/src/alx/linux/man-pages/man-pages.git/commit/?h=contrib&id=b8f472450f6607e2d5bd68a1b60615a91ed3d111>
(use port 80).

> ---
>  man/man7/ip.7 | 16 +++++++++++++---
>  1 file changed, 13 insertions(+), 3 deletions(-)
> 
> diff --git a/man/man7/ip.7 b/man/man7/ip.7
> index a92939cd0..a7f118b42 100644
> --- a/man/man7/ip.7
> +++ b/man/man7/ip.7
> @@ -809,12 +809,20 @@ .SS Socket options
>  .EE
>  .in
>  .IP
> +When returned by
> +.BR recvmsg (2) ,
>  .I ipi_ifindex
>  is the unique index of the interface the packet was received on.
>  .I ipi_spec_dst
> -is the local address of the packet and
> +is the preferred source address for replies to the given packet, and
>  .I ipi_addr
>  is the destination address in the packet header.
> +These addresses are usually the same,
> +but can differ for broadcast or multicast packets.
> +Note that, depending on the configured routes,

I've removed 'Note that,'.  It's redundant.  Everything in a manual page
should be noteworthy.


Have a lovely day!
Alex

> +.I ipi_spec_dst
> +might belong to a different interface from the one that received the packet.
> +.IP
>  If
>  .B IP_PKTINFO
>  is passed to
> @@ -822,14 +830,16 @@ .SS Socket options
>  and
>  .\" This field is grossly misnamed
>  .I ipi_spec_dst
> -is not zero, then it is used as the local source address for the routing
> -table lookup and for setting up IP source route options.
> +is not zero, then it is used as the local source address, for the routing
> +table lookup, and for setting up IP source route options.
>  When
>  .I ipi_ifindex
>  is not zero, the primary local address of the interface specified by the
>  index overwrites
>  .I ipi_spec_dst
>  for the routing table lookup.
> +.I ipi_addr
> +is ignored.
>  .IP
>  Not supported for
>  .B SOCK_STREAM
> -- 
> 2.47.3
> 

-- 
<https://www.alejandro-colomar.es>
Use port 80 (that is, <...:80/>).

Download attachment "signature.asc" of type "application/pgp-signature" (834 bytes)

Powered by blists - more mailing lists