splitting m_flags to pkthdr.flags + m_flags

[Andre Oppermann]

On 02.11.2012 13:38, Gleb Smirnoff wrote:
>    Hello networkers,
>
>    some (most) of m_flags bits are describing features that can
> be present on a packet only, not on a single mbuf. Since we are
> close to exhaustion of available bits, and many subsystems prefer
> to use one of M_PROTO bits instead of grabbing a flag, I propose

The use of M_PROTO for protocol and layer specific use is very much
its intended purpose and should be kept that way with overlays.  The
available range of M_PROTO can be extended with your proposed change.
M_PROTO could be split into mbuf and mbuf header flags.  Most of the
time they are also only relevant on packets.

> to split m_flags to two parts:

I fully agree on the split.  The previous mixing of mbuf and mbuf header
flags was irritating and confusing.

> In the m_flags remain:
> #define M_EXT           0x00000001 /* has associated external storage */
> #define M_PKTHDR        0x00000002 /* start of record */
> #define M_EOR           0x00000004 /* end of record */
> #define M_RDONLY        0x00000008 /* associated data is marked read-only */
> and all M_PROTO flags.
>
> struct pkthdr grows by one word and its new flag word now
> posesses:
>
> #define M_BCAST         0x00000200 /* send/received as link-level broadcast */
> #define M_MCAST         0x00000400 /* send/received as link-level multicast */
> #define M_FRAG          0x00000800 /* packet is a fragment of a larger packet */
> #define M_FIRSTFRAG     0x00001000 /* packet is first fragment */
> #define M_LASTFRAG      0x00002000 /* packet is last fragment */

I wonder if actually have any users of the M_*FRAG flags.  And if yes,
if it can be solved in a different way with less flags.

> #define M_SKIP_FIREWALL 0x00004000 /* skip firewall processing */

This one should become an M_PROTO overlay.  It is only relevant within
a protocol layer.

> #define M_VLANTAG       0x00010000 /* ether_vtag is valid */
> #define M_PROMISC       0x00020000 /* packet was not for us */
> #define M_FLOWID        0x00400000 /* deprecated: flowid is valid */
> #define M_HASHTYPEBITS  0x0F000000 /* mask of bits holding flowid hash type */
 >
> Some M_PROTO abusements like M_FASTFWD_OURS and recently added M_IP_NEXTHOP
> also go to the pkthdr mbuf flags.

On the contrary.  This isn't abuse of M_PROTO.  It's the exact pupose
of it.

> P.S.
> An attentive reader may have noticed that I missed M_NOFREE and M_FREELIST.
> Yep, these flags coming from historical mbuf allocator from FreeBSD 4.x times
> are about to be deleted, we carefully examine them, but never set. Patch
> for review attached.

Looks good.  Go ahead from me.

-- 
Andre