Re: How to investigate ENOBUFS?

[mlelstv%serpens.de@localhost (Michael van Elst)]

manu%NetBSD.org@localhost (Emmanuel Dreyfus) writes:

>The filesystem will never recover. As I understand, this happens because
>some socket has pending data that is never readen. Do I have a way to
>discover what socket is holding buffer space? And is there a way to see how
>much buffer space is allocated and how much remain, by the way?

netstat -m will show you how many buffers are allocated.
vmstat -m will show you the parameters of the mbpool_cache ("mbpl").

A kernel with option MBUFTRACE lets netstat -mssv report finer
details on mbuf usage. Unfortunately not on a per-socket level.

Each socket will allocate net.inet.{udp,tcp}.{sendspace,recvspace} unless
the application asks for a specific size. In -current we have buffer
autosizing specified by net.inet.{udp,tcp}.{sendbuf,recvbuf}_inc and
net.inet.{udp,tcp}.{sendbuf,recvbuf}_max. But I don't know a way to
query the buffer size from the outside short of gdb.

ENOBUFS is also reported when the sending buffer of a (SOCK_DGRAM only?)
socket is full. netstat shows send-q and recv-q, not the buffer sizes
but the corresponding buffer usages that give a hint what socket is
sending.

-- 
-- 
                                Michael van Elst
Internet: mlelstv%serpens.de@localhost
                                "A potential Snark may lurk in every tree."