From: Matthias Schiffer mschiffer@universe-factory.net
With this patch the maximum fragment size is reduced from 1400 to 1280 bytes.
Fragmentation v2 correctly uses the smaller of 1400 and the interface MTU, thus generally supporting interfaces with an MTU < 1400 bytes, too.
However, currently "Fragmentation v2" does not support re-fragmentation. Which means that once a packet is split into two packets of 1400 + x bytes for instance and the next hop provides an interface with an even smaller MTU of 1280 bytes, then the larger fragment is lost.
A maximum fragment size of 1280 bytes is a safer option as this is the minimum MTU required by IPv6, making interfaces with an MTU < 1280 rather exotic.
Regarding performance, this should have no negative impact on unicast traffic: Having some more bytes in the smaller and some less in the larger does not change the sum of both fragments.
Concerning TT, choosing 1280 bytes fragments might result in more TT messages than necessary when a large network is bridged into batman-adv. However, the TT overhead in general is marginal due to its reactive nature, therefore such a performance impact on TT should not be noticeable for a user.
Cc: Matthias Schiffer mschiffer@universe-factory.net [linus.luessing@c0d3.blue: Added commit message] Signed-off-by: Linus Lüssing linus.luessing@c0d3.blue ---
This patch is currently used in Gluon [1] because we stumbled over this issue in setups where batman-adv first fragmented to 1400 bytes for a wired, 1500B MTU connection and the larger fragment then being lost on the next hop's VPN interface with an MTU of 1280 bytes.
[1]: https://github.com/freifunk-gluon/gluon/tree/master/patches/packages/routing
net/batman-adv/main.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/net/batman-adv/main.h b/net/batman-adv/main.h index 57a8103..81c6a38 100644 --- a/net/batman-adv/main.h +++ b/net/batman-adv/main.h @@ -168,7 +168,7 @@ enum batadv_uev_type { /* Maximum number of fragments for one packet */ #define BATADV_FRAG_MAX_FRAGMENTS 16 /* Maxumim size of each fragment */ -#define BATADV_FRAG_MAX_FRAG_SIZE 1400 +#define BATADV_FRAG_MAX_FRAG_SIZE 1280 /* Time to keep fragments while waiting for rest of the fragments */ #define BATADV_FRAG_TIMEOUT 10000
On Sonntag, 29. Januar 2017 06:14:28 CET Linus Lüssing wrote: [...]
However, currently "Fragmentation v2" does not support re-fragmentation. Which means that once a packet is split into two packets of 1400 + x bytes for instance and the next hop provides an interface with an even smaller MTU of 1280 bytes, then the larger fragment is lost.
[...]
Beside this change, did you try to make the fragments more equally sized? Here is just some really ugly and untested pseudo code:
@@ -454,7 +454,7 @@ int batadv_frag_send_packet(struct sk_buff *skb, struct sk_buff *skb_fragment; unsigned int mtu = neigh_node->if_incoming->net_dev->mtu; unsigned int header_size = sizeof(frag_header); - unsigned int max_fragment_size, max_packet_size; + unsigned int max_fragment_size, num_fragments; int ret;
/* To avoid merge and refragmentation at next-hops we never send @@ -462,10 +462,17 @@ int batadv_frag_send_packet(struct sk_buff *skb, */ mtu = min_t(unsigned int, mtu, BATADV_FRAG_MAX_FRAG_SIZE); max_fragment_size = mtu - header_size; - max_packet_size = max_fragment_size * BATADV_FRAG_MAX_FRAGMENTS; + + if (skb->len == 0) + return -EINVAL; + +#define ceil_div(x, y) (1 + ((x - 1) / y)) + num_fragments = ceil_div(skb->len, max_fragment_size); + max_fragment_size = ceil_div(skb->len, num_fragments); +#undef ceil_div
/* Don't even try to fragment, if we need more than 16 fragments */ - if (skb->len > max_packet_size) { + if (num_fragments > BATADV_FRAG_MAX_FRAGMENTS) { ret = -EAGAIN; goto free_skb; }
Kind regards, Sven
On Tue, Jan 31, 2017 at 12:07:24PM +0100, Sven Eckelmann wrote:
On Sonntag, 29. Januar 2017 06:14:28 CET Linus Lüssing wrote: [...]
However, currently "Fragmentation v2" does not support re-fragmentation. Which means that once a packet is split into two packets of 1400 + x bytes for instance and the next hop provides an interface with an even smaller MTU of 1280 bytes, then the larger fragment is lost.
[...]
Beside this change, did you try to make the fragments more equally sized? Here is just some really ugly and untested pseudo code:
Have thought about it initially but pushed it asided when I noticed it would only help regarding the initial issue for the unicast packets but not potentially large TTs.
But generally, I like the idea of making fragments more equally sized :-).
b.a.t.m.a.n@lists.open-mesh.org
-
Linus Lüssing -
Sven Eckelmann