tcpreplay - Man Page

Replay network traffic stored in pcap files

Examples (TL;DR)

List available network interfaces: tcpreplay --listnics
Replay traffic to interface: tcpreplay [-i|--intf1] eth0 traffic.pcap
Replay traffic to interface and stdout: tcpreplay [-i|--intf1] eth0 [-v|--verbose] traffic.pcap
Replay traffic to interface as fast as possible: tcpreplay [-i|--intf1] eth0 [-t|--topspeed] traffic.pcap
Replay traffic to interface at given Mbps: tcpreplay [-i|--intf1] eth0 [-M|--mbps] 10 traffic.pcap
Replay traffic to interface several times: tcpreplay [-i|--intf1] eth0 [-l|--loop] num_times traffic.pcap

Synopsis

tcpreplay [OPTION]... <pcap_file(s)> | <pcap_dir(s)>

Description

tcpreplay is a tool for replaying network traffic from files saved with tcpdump or other tools which write pcap(3) files.

The basic operation of tcpreplay is to resend all packets from the input file(s) at the speed at which they were recorded, or a specified data rate, up to as fast as the hardware is capable.

Optionally, the traffic can be split between two interfaces, written to files, filtered and edited in various ways, providing the means to test firewalls, NIDS and other network devices.

On Linux, raw IP (layer 3 only) interfaces such as WireGuard and tun are supported automatically: any layer 2 header in the input file (Ethernet including VLAN tags, Linux SLL, loopback, ...) is stripped and packets are sent as bare IPv4/IPv6. Non-IP packets such as ARP cannot be sent on these interfaces and are reported as failed.

For more details, please see the Tcpreplay Manual at: http://tcpreplay.appneta.com

Options

-d, --dbug=number

Enable debugging output. Range: 0 to 5. Default: 0.

If configured with --enable-debug, then you can specify a verbosity level for debugging output. Higher numbers increase verbosity.

-q, --quiet

Quiet mode.

Print nothing except the statistics at the end of the run

-T, --timer=string

Select packet timing mode: select, ioport, gtod, nano. Default: gtod.

Allows you to select the packet timing method to use:

nano: Use nanosleep() API
select: Use select() API
ioport: Write to the i386 IO Port 0x80
gtod [default]: Use a gettimeofday() loop

--maxsleep=number

Sleep for no more then X milliseconds between packets. Default: 0.

Set a limit for the maximum number of milliseconds that tcpreplay will sleep between packets. Effectively prevents long delays between packets without effecting the majority of packets. Default is disabled.

-v, --verbose

Print decoded packets via tcpdump to STDOUT.

-A, --decode=string

Arguments passed to tcpdump decoder. Requires: --verbose.

When enabling verbose mode (-v) you may also specify one or more additional arguments to pass to tcpdump to modify the way packets are decoded. By default, -n and -l are used. Be sure to quote the arguments like: -A "-axxx" so that they are not interpreted by tcpreplay. Please see the tcpdump(1) man page for a complete list of options.

-K, --preload_pcap

Preloads packets into RAM before sending.

This option loads the specified pcap(s) into RAM before starting to send in order to improve replay performance while introducing a startup performance hit. Preloading can be used with or without --loop. This option also suppresses flow statistics collection for every iteration, which can significantly reduce memory usage. Flow statistics are predicted based on options supplied and statistics collected from the first loop iteration.

-c, --cachefile=string

Split traffic via a tcpprep cache file. Requires: --intf2. Cannot be combined with: --dualfile.

If you have a pcap file you would like to use to send bi-directional traffic through a device (firewall, router, IDS, etc) then using tcpprep you can create a cachefile which tcpreplay will use to split the traffic across two network interfaces.

-2, --dualfile

Replay two files at a time from a network tap. Requires: --intf2. Cannot be combined with: --cachefile.

If you captured network traffic using a network tap, then you can end up with two pcap files- one for each direction. This option will replay these two files at the same time, one on each interface and inter-mix them using the timestamps in each.

-i, --intf1=string

Client to server/RX/primary traffic output interface. This option is required.

Required network interface used to send either all traffic or traffic which is marked as 'primary' via tcpprep. Primary traffic is usually client-to-server or inbound (RX) on khial virtual interfaces.

-I, --intf2=string

Server to client/TX/secondary traffic output interface.

Optional network interface used to send traffic which is marked as 'secondary' via tcpprep. Secondary traffic is usually server-to-client or outbound (TX) on khial virtual interfaces. Generally, it only makes sense to use this option with --cachefile.

-w, --write=string

Pcap file to receive traffic outputs. Cannot be combined with: --intf2.

Optional pcap file name used to receive traffic.

--include=string

Send only selected packet numbers. Cannot be combined with: --exclude.

Override default of processing all packets stored in the capture file and only send packets that are part of a supplied list of packet numbers.

-x P:1-5,9,15,72-

would skip packets 1 through 5, the 9th and 15th packet, and packets 72 until the end of the file

--exclude=string

Send all but selected packet numbers. Cannot be combined with: --include.

Override default of processing all packets stored in the capture file and only send packets that are NOT part of a supplied list of packet numbers.

-x P:1-5,9,15,72-

would skip packets 1 through 5, the 9th and 15th packet, and packets 72 until the end of the file

--listnics

List available network interfaces and exit.

-l, --loop=number

Loop through the capture file X times. Range: 0 to any. Default: 1.

--loopdelay-ms=number

Delay between loops in milliseconds. Range: 0 to any. Default: 0. Requires: --loop. Cannot be combined with: --loopdelay-ns.

--loopdelay-ns=number

Delay between loops in nanoseconds. Range: 0 to any. Default: 0. Requires: --loop. Cannot be combined with: --loopdelay-ms.

By default, tcpreplay will use loop delay with microsecond accuracy (loopdelay-ms). In order to use loop delay with nanosecond accuracy you need to use nano packet timing mode.

--pktlen

Override the snaplen and use the actual packet len.

By default, tcpreplay will send packets based on the size of the "snaplen" stored in the pcap file which is usually the correct thing to do. However, occasionally, tools will store more bytes then told to. By specifying this option, tcpreplay will ignore the snaplen field and instead try to send packets based on the original packet length. Bad things may happen if you specify this option.

-L, --limit=number

Limit the number of packets to send. Range: 1 to any. Default: -1.

By default, tcpreplay will send all the packets. Alternatively, you can specify a maximum number of packets to send.

--duration=number

Limit the number of seconds to send. Range: 1 to any. Default: -1.

By default, tcpreplay will send all the packets. Alternatively, you can specify a maximum number of seconds to transmit.

-x, --multiplier=string

Modify replay speed to a given multiple. Cannot be combined with: --pps, --mbps, --oneatatime, --topspeed.

Specify a value to modify the packet replay speed. Examples:

    2.0 will replay traffic at twice the speed captured
    0.7 will replay traffic at 70% the speed captured

-p, --pps=string

Replay packets at a given packets/sec. Cannot be combined with: --multiplier, --mbps, --oneatatime, --topspeed.

Specify a value to regulate the packet replay to a specific packet-per-second rate. Examples:

    200 will replay traffic at 200 packets per second
    0.25 will replay traffic at 15 packets per minute

-M, --mbps=string

Replay packets at a given Mbps. Cannot be combined with: --multiplier, --pps, --oneatatime, --topspeed.

Specify a floating point value for the Mbps rate that tcpreplay should send packets at.

-t, --topspeed

Replay packets as fast as possible. Cannot be combined with: --mbps, --multiplier, --pps, --oneatatime.

-o, --oneatatime

Replay one packet at a time for each user input. Cannot be combined with: --mbps, --pps, --multiplier, --topspeed.

Allows you to step through one or more packets at a time.

--pps-multi=number

Number of packets to send for each time interval. Range: 1 to any. Default: 1. Requires: --pps.

When trying to send packets at very high rates, the time between each packet can be so short that it is impossible to accurately sleep for the required period of time. This option allows you to send multiple packets at a time, thus allowing for longer sleep times which can be more accurately implemented.

--loss=string

Simulate random percent packet loss (0-100).

Set to a percentage between 0 and 100 to simulate packet loss at send. Loss is random: each packet is independently dropped with the given probability, not on a fixed pattern (e.g. not every Nth packet), so the actual number dropped will vary slightly run to run. Dropped packets are not sent and are not counted as failed.

Examples:

    25 will drop approximately one quarter of packets, chosen at random
    100 will drop every packet (nothing will be sent)

--unique-ip

Modify IP addresses each loop iteration to generate unique flows. Requires: --loop.

Ensure IPv4 and IPv6 packets will be unique for each --loop iteration. This is done in a way that will not alter packet CRC, and therefore will generally not affect performance. This option will significantly increase the flows/sec over generated over multiple loop iterations.

--unique-ip-loops=string

Number of times to loop before assigning new unique ip. Requires: --unique-ip.

Number of --loop iterations before a new unique IP is assigned. Default is 1. Assumes both --loop and --unique-ip.

--netmap

Write packets directly to netmap enabled network adapter.

This feature will detect netmap capable network drivers on Linux and BSD systems. If detected, the network driver is bypassed for the execution duration, and network buffers will be written to directly. This will allow you to achieve full line rates on commodity network adapters, similar to rates achieved by commercial network traffic generators. Note that bypassing the network driver will disrupt other applications connected through the test interface. See INSTALL for more information.

This feature can also be enabled by specifying an interface as 'netmap:<intf>' or 'vale:<intf>. For example 'netmap:eth0' specifies netmap over interface eth0.

--nm-delay=number

Netmap startup delay. Default: 10.

Number of seconds to delay after netmap is loaded. Required to ensure interfaces are fully up before netmap transmit. Requires netmap option. Default is 10 seconds.

--no-flow-stats

Suppress printing and tracking flow count, rates and expirations.

Suppress the collection and printing of flow statistics. This option may improve performance when not using --preload-pcap option, otherwise its only function is to suppress printing.

The flow feature will track and print statistics of the flows being sent. A flow is loosely defined as a unique combination of a 5-tuple, i.e. source IP, destination IP, source port, destination port and protocol.

If --loop is specified, the flows from one iteration to the next will not be unique, unless the packets are altered. Use --unique-ip or tcpreplay-edit to alter packets between iterations.

--flow-expiry=number

Number of inactive seconds before a flow is considered expired. Range: 0 to any. Default: 0. Cannot be combined with: --no-flow-stats.

This option will track and report flow expirations based on the flow idle times. The timestamps within the pcap file are used to determine the expiry, not the actual timestamp of the packets are replayed. For example, a value of 30 suggests that if no traffic is seen on a flow for 30 seconds, any subsequent traffic would be considered a new flow, and thereby will increment the flows and flows per second (fps) statistics.

This option can be used to optimize flow timeout settings for flow products. Setting the timeout low may lead to flows being dropped when in fact the flow is simply slow to respond. Configuring your flow timeouts too high may increase resources required by your flow product.

Note that using this option while replaying at higher than original speeds can lead to inflated flows and fps counts.

Default is 0 (no expiry) and a typical value is 30-120 seconds.

-P, --pid

Print the PID of tcpreplay at startup.

--stats=number

Print statistics every X seconds, or every loop if '0'. Range: 0 to any.

Note that timed delays are a "best effort" and long delays between sending packets may cause equally long delays between printing statistics.

-W, --suppress-warnings

suppress printing warning messages.

--xdp

Write packets directly to AF_XDP enabled network adapter.

This feature will detect AF_XDP capable network drivers on Linux systems that have 'libxdp-dev' and 'libbpf-dev' installed. If detected, the network stack is bypassed and packets are sent directly to an eBPF enabled driver directly. This will allow you to achieve full line rates on commodity network adapters, similar to rates achieved by commercial network traffic generators.

--xdp-batch-size=number

The maximum number of packets that can be submitted to the AF_XDP TX ring at once. Range: 1 to 4096. Default: 1. Requires: --xdp, --topspeed. Cannot be combined with: --oneatatime, --pps-multi.

Only allowed if sending at top speed. AF_XDP sends are pipelined across the whole UMEM regardless of this setting, so a batch of 1 already reaches line rate on common adapters - this should not normally need changing. It exists for very high speed links (100GigE and up), especially with small packet sizes, where a deeper batch may still help amortize per-packet overhead.

--xdp-queue=number

Which of the adapter’s queues to bind the AF_XDP socket to. Range: 0 to 4095. Default: 0. Requires: --xdp.

An AF_XDP socket is bound to a single hardware queue. Queue 0 is used by default; on a multi-queue adapter you may need a different one, since the queue must exist on the adapter and traffic steering may not use queue 0. If the queue does not exist the socket cannot be created.

--xdp-no-fallback

Fail instead of falling back when AF_XDP is unavailable. Requires: --xdp.

By default, if an AF_XDP socket cannot be set up on the interface - the driver has no XDP support at all, or the requested queue does not exist - tcpreplay warns and replays using its default injection method instead of failing. Use this to make that case a hard error, which is what you want when benchmarking: a silent change of injection method would change what is being measured.

--io-uring

Send packets asynchronously via Linux io_uring.

This feature is available on Linux systems with a kernel that supports io_uring and 'liburing-dev' installed. Packets are still sent through a PF_PACKET raw socket, but sends are submitted asynchronously through an io_uring submission queue, which reduces per-packet syscall overhead and lets the kernel process transmissions while tcpreplay prepares the next packet.

--raw

Send packets via a PF_INET raw IP socket instead of PF_PACKET.

By default, tcpreplay injects packets with PF_PACKET, which writes directly to the network driver below the IP stack. This means locally-generated netfilter/iptables rules (and anything else hooked into the normal Linux IP stack) never see the traffic, even when replaying it on the same host that has those rules configured.

This option instead sends each packet’s IP header and payload through a PF_INET raw IP (SOCK_RAW) socket bound to the interface given by --intf1/--intf2, so the kernel’s normal IP stack — including netfilter/iptables — processes it like any other locally-generated packet.

The trade-off is L2 fidelity: the kernel builds its own Ethernet framing for the outgoing packet, so the source/destination MAC addresses from the capture are not reproduced, and only IPv4 packets are currently supported. This option is Linux-only and requires appropriate privileges to open a raw socket (typically root or CAP_NET_RAW).

-V, --version