Awesome
ewpcap is a native Erlang interface to PCAP that can be used for reading and writing packets from the network.
ewpcap is meant to be a portable raw socket interface to all the platforms that support Erlang and libpcap.
WARNING
ewpcap was written and tested under Linux. But if you are using a Unix system, you may want to use one of these projects instead:
-
procket : https://github.com/msantos/procket
-
epcap : https://github.com/msantos/epcap
REQUIREMENTS
-
libpcap/winpcap
On Ubuntu: sudo apt-get install libpcap-dev
These libraries are not required but can be used with ewpcap:
-
pkt: https://github.com/msantos/pkt.git
Use pkt to decode/encode packets read from the network.
-
privileges
ewpcap requires beam to be running with root privileges:
* using sudo
sudo erl -pa ebin
* using capabilities
setcap cap_net_raw=ep /path/to/beam.smp
COMPILING
rebar3 do clean, compile, ct
DATA TYPES
ewpcap_resource()
A record returned by open/0,1,2.
The record contains 2 fields:
* res: an NIF resource associated with the pcap socket. The
pcap process terminates when this resource is garbage
collected.
* ref: reference to socket handle in packet tuple
EXPORTS
open() -> {ok, Socket} | {error, Error}
open(Dev) -> {ok, Socket} | {error, Error}
open(Dev, Options) -> {ok, Socket} | {error, Error}
Types Dev = binary() | string()
Error = enomem | string()
Socket = resource()
Options = [ Option ]
Option = {promisc, boolean()}
| {snaplen, integer()}
| {timeout, immediate | infinity | non_neg_integer()}
| {buffer, integer()}
| {monitor, boolean()}
| {filter, iolist()}
| {time_unit, timestamp | microsecond}
| FilterOpts
Open a network interface and begin receiving packets.
The returned Socket in the 'ok' tuple must be kept by the
process. When the socket goes out of scope, the pcap filter will
be shut down and all resources associated with the socket will
be freed. See also close/1.
Dev is the name of the network device. If an empty binary (<<>>)
is passed in, pcap will select a default interface.
If an error occurs, the PCAP string describing the error is
returned to the caller.
open/1 and open/2 default to:
* promiscuous mode disabled
* a snaplen (packet length) of 65535 bytes
* uses immediate mode
* the time unit is an erlang timestamp in the same format
as now/0
* no filter (all packets are received)
The `timeout` option sets `pcap_set_timeout(3PCAP)` and
`pcap_set_immediate_mode(3PCAP)`. By default, ewpcap uses
`immediate` mode and returns packets as they are received.
Setting timeout to an integer value disables `immediate` mode
and buffers any packets until either the timeout is reached or
the buffer is filled.
If ewpcap is dropping packets (see stats/1), the PCAP buffer
size can be increased (should be some multiple of the snaplen).
Wireless devices can be set to use monitor mode (rfmon) by
passing in the 'monitor' option.
The timestamp in the message can be formatted either as a now/0
tuple or returned in microseconds.
For filter options, see filter/3.
Packets are returned as messages to the caller:
{ewpcap, Ref, DatalinkType, Time, Length, Packet}
Ref is a reference identifying the socket handle.
The DataLinkType is an integer representing the link layer,
e.g., ethernet, Linux cooked socket.
The Time is a tuple in the same format as erlang:now/0, {MegaSecs,
Secs, MicroSecs} or microseconds.
The Length corresponds to the actual packet length on the
wire. The captured packet may have been truncated. To get the
captured packet length, use byte_size(Packet).
The Packet is a binary holding the captured data.
Errors will be sent to the caller and the pcap filter will
be terminated:
{ewpcap_error, Ref, Error}
close(Socket) -> ok
Closes the pcap descriptor. See "SCHEDULER LATENCY".
filter(Socket, Filter) -> ok | {error, Error}
filter(Socket, Filter, Options) -> ok | {error, Error}
Types Socket = resource()
Filter = iolist()
Error = enomem | string()
Options = [ Option ]
Option = {optimize, boolean()}
| {netmask, integer()}
| {limit, integer()}
Compile a PCAP filter and apply it to the PCAP descriptor.
Since the library passes the filter string to pcap_compile(3PCAP)
directly, any bugs in pcap_compile() may cause the Erlang VM
to crash. Do not use filters from untrusted sources.
Filters are limited to 8192 bytes by default since it may be
possible for very large filters to cause a stack overflow. For
example:
ewpcap:open(<<>>, [{filter, string:copies("ip and ", 50000) ++ "ip"}, {limit, -1}])
read(Socket) -> {ok, Packet}
read(Socket, Timeout) -> {ok, Packet} | {error, Error}
Types Socket = resource()
Timeout = uint() | infinity
Packet = binary()
Error = eagain | string()
Convenience function wrapping receive, returning the packet
contents.
write(Socket, Packet) -> ok | {error, string()}
Types Socket = resource()
Packet = iodata()
Write the packet to the network. See pcap_sendpacket(3PCAP).
dev() -> {ok, string()} | {error, string()}
Returns the default device used by PCAP.
getifaddrs() -> {ok, Iflist} | {error, posix()}
Types Iflist = [{Ifname, [Ifopt]}]
Ifname = string()
Ifopt = {flag, [Flag]}
| {addr, Addr}
| {netmask, Netmask}
| {broadaddr, Broadaddr}
| {dstaddr, Dstaddr}
| {description, string()}
Flag = loopback
Addr = Netmask = Broadaddr = Dstaddr = ip_address()
Returns a list of interfaces. Ifname can be used as the first
parameter to open/1 and open/2.
This function is modelled on inet:getifaddrs/0 but uses
pcap_findalldevs(3PCAP) to look up the interface attributes:
* getifaddrs/0 may return pseudo devices, such as the "any"
device on Linux
* getifaddrs/0 will only return the list of devices that
can be used with open/1 and open/2. An empty list ({ok,
[]}) may be returned if the user does not have permission
to open any of the system interfaces
stats(Socket) -> {ok, #ewpcap_stat{}} | {error, string()}
Types Socket = resource()
To use the return value as a record, include the header:
-include_lib("ewpcap/include/ewpcap.hrl").
stats/1 returns statistics about dropped packets. See
pcap_stats(3PCAP) for details.
The ewpcap_stat records contains these fields:
recv : number of packets received
drop : number of packets dropped due to insufficient buffer
ifdrop : number of packets dropped by the network interface
capt : always 0 (was number of packets received by the application (Win32 only))
SCHEDULER LATENCY
In normal usage, ewpcap does not perform any blocking operations that could interfere with the scheduler. For example, spawning one or more long running pcap processes is scheduler friendly.
To confirm, run:
erlang:system_monitor(self(), [{long_schedule, 10}]).
ewpcap may block when stopping the pcap process. This situation might occur if rapidly spawning and garbage collecting pcap processes:
% Don't do this: ewpcap resource freed when spawned processes exit
N = 100,
[ spawn(fun() -> ewpcap:open(<<>>, [{filter, "tcp and port 9876"}]), ok end)
|| X <- lists:seq(1,N) ].
However, if you need to do it, there are some workarounds:
- decrease the pcap timeout
% Set timeout to 1 ms
ewpcap:open(<<>>, [{filter, "tcp"}, {timeout, 1}]).
- explicitly do resource cleanup on a dirty scheduler
{ok, Socket} = ewpcap:open(<<>>, [{filter, "tcp"}]),
ewpcap:close(Socket)
DISABLING DIRTY SCHEDULER SUPPORT
Use of the dirty scheduler can be disabled by setting an environment variable:
EWPCAP_DISABLE_DIRTY_SCHEDULER=1 rebar3 do clean, compile
It is safe to disable for normal operation (but see "SCHEDULER LATENCY").
EXAMPLES
-module(icmp_resend).
-export([start/1]).
% icmp_resend:start("eth0").
start(Dev) ->
{ok, Socket} = ewpcap:open(Dev, [{filter, "icmp"}]),
resend(Socket).
resend(Socket) ->
{ok, Packet} = ewpcap:read(Socket),
ok = ewpcap:write(Socket, Packet),
resend(Socket).