real interface MTU different than seen in the output of "ip link show" command

[Martin T <m4rtntns@gmail.com>]

Hi,

today I noticed an interesting behavior where the PC NIC does not
discard a received Ethernet frame, although it is larger than MTU on
this NIC. For example, I made PC1[eth0] <-> [eth0]PC2 and PC1[eth0]
<-> [eth0]PC3 connections and set the MTU on PC1 eth0 interface to
9000 bytes and MTU on PC2 and PC3 eth0 interfaces to 1500
bytes(confirmed with "ip link show" command). I expected PC2 and PC3
to drop frames >1500 bytes, but instead PC2 processed frames up to
4210 bytes and PC3 up to 8166 bytes. eth0 in PC2 uses e1000e driver
and eth0 in PC3 uses r8169 driver included with jessie.
On the other hand, Cisco switch works as expected. For example, if its
port Fa0/30 MTU is 1998 bytes and I send >1998 byte frames, then those
are dropped and counted as giants:

c2960#sh int Fa0/30 counters errors

Port        Align-Err     FCS-Err    Xmit-Err     Rcv-Err  UnderSize
OutDiscards
Fa0/30              0           0           0           0          0
         0

Port      Single-Col  Multi-Col   Late-Col  Excess-Col  Carri-Sen
Runts     Giants
Fa0/30             0          0          0           0          0
    0          4
c2960#


How to explain this behavior?


thanks,
Martin