tcpdump - dump traffic on a network
tcpdump [ -deflnNOpqStvx ] [ -c count ] [ -F file ]
[ -i interface ] [ -r file ] [ -s snaplen ]
[ -w file ] expression
Tcpdump prints out the headers of packets on a network interface that match the boolean expression. Under SunOS: You must be root to invoke tcpdump or it must be installed setuid to root. Under Ultrix: Any user can invoke tcpdump once the super-user has enabled promiscuous-mode operation using pfconfig(8). Under BSD: Access is controlled by the permissions on /dev/bpf0, etc.
Print `foreign' internet addresses numerically rather than symbolically (this option is intended to get around serious brain damage in Sun's yp server -- usually it hangs forever translating non-local internet numbers).
The expression consists of one or more primitives. Primitives usually consist of an id (name or number) preceded by one or more qualifiers. There are three different kinds of qualifier:
dir qualifiers specify a particular tranfer direction to and/or from id. Possible directions are src, dst, src or dst and src and dst. E.g., `src foo', `dst net 128.3', `src or dst port ftp-data'. If there is no dir qualifier, src or dst is assumed.
proto qualifiers restrict the match to a particular protocol. Possible protos are: ether, fddi, ip, arp, rarp, decnet, lat, moprc, mopdl, tcp and udp. E.g., `ether src foo', `arp net 128.3', `tcp port 21'. If there is no proto qualifier, all protocols consistent with the type are assumed. E.g., `src foo' means `(ip or arp or rarp) src foo' (except the latter is not legal syntax), `net bar' means `(ip or arp or rarp) net bar' and `port 53' means `(tcp or udp) port 53'.
[`fddi' is actually an alias for `ether'; the parser treats them identically as meaning ``the data link level used on the specified network interface.'' FDDI headers contain Ethernet-like source and destination addresses, and often contain Ethernet-like packet types, so you can filter on these FDDI fields just as with the analogous Ethernet fields. FDDI headers also contain other fields, but you cannot name them explicitly in a filter expression.]
In addition to the above, there are some special `primitive' keywords that don't follow the pattern: gateway, broadcast, less, greater and arithmetic expressions. All of these are described below.
More complex filter expressions are built up by using the words and, or and not to combine primitives. E.g., `host foo and not port ftp and not port ftp-data'. To save typing, identical qualifier lists can be omitted. E.g., `tcp dst port ftp or ftp-data or domain' is exactly the same as `tcp dst port ftp or tcp dst port ftp-data or tcp dst port domain'.
Allowable primitives are:
For example, `ether[0] & 1 != 0' catches all multicast traffic. The expression `ip[0] & 0xf != 5' catches all IP packets with options. The expression `ip[6:2] & 0x1fff = 0' catches only unfragmented datagrams and frag zero of fragmented datagrams. This check is implicitly applied to the tcp and udp index operations. For instance, tcp[0] always means the first byte of the TCP header, and never means the first byte of an intervening fragment.
Primitives may be combined using:
Negation (`!' or `not').
Concatenation (`&&' or `and').
Alternation (`||' or `or').
Negation has highest precedence. Alternation and concatenation have equal precedence and associate left to right. Note that explicit and tokens, not juxtaposition, are now required for concatenation.
If an identifier is given without a keyword, the most recent keyword is assumed. For example, not host vs and ace is short for not host vs and host ace which should not be confused with not ( host vs or ace )
Expression arguments can be passed to tcpdump as either a single argument or as multiple arguments, whichever is more convenient. Generally, if the expression contains Shell metacharacters, it is easier to pass it as a single, quoted argument. Multiple arguments are concatenated with spaces before being parsed.
To print all packets arriving at or departing from sundown:
tcpdump host sundownTo print traffic between helios and either hot or ace:
tcpdump host helios and \( hot or ace \)To print all IP packets between ace and any host except helios:
tcpdump ip host ace and not heliosTo print all traffic between local hosts and hosts at Berkeley:
tcpdump net ucb-etherTo print all ftp traffic through internet gateway snup: (note that the expression is quoted to prevent the shell from (mis-)interpreting the parentheses):
tcpdump 'gateway snup and (port ftp or ftp-data)'To print traffic neither sourced from nor destined for local hosts (if you gateway to one other net, this stuff should never make it onto your local net).
tcpdump ip and not net localnetTo print the start and end packets (the SYN and FIN packets) of each TCP conversation that involves a non-local host.
tcpdump 'tcp[13] & 3 != 0 and not src and dst net localnet'To print IP packets longer than 576 bytes sent through gateway snup:
tcpdump 'gateway snup and ip[2:2] > 576'To print IP broadcast or multicast packets that were not sent via ethernet broadcast or multicast:
tcpdump 'ether[0] & 1 = 0 and ip[16] >= 224'To print all ICMP packets that are not echo requests/replies (i.e., not ping packets):
tcpdump 'icmp[0] != 8 and icmp[0] != 0"
The output of tcpdump is protocol dependent. The following gives a brief description and examples of most of the formats.
If the '-e' option is given, the link level header is printed out. On ethernets, the source and destination addresses, protocol, and packet length are printed.
On FDDI networks, the '-e' option causes tcpdump to print the `frame control' field, the source and destination addresses, and the packet length. (The `frame control' field governs the interpretation of the rest of the packet. Normal packets (such as those containing IP datagrams) are `async' packets, with a priority value between 0 and 7; for example, `async4'. Such packets are assumed to contain an 802.2 Logical Link Control (LLC) packet; the LLC header is printed if it is not an ISO datagram or a so-called SNAP packet.
(N.B.: The following description assumes familiarity with the SLIP compression algorithm described in RFC-1144.)
On SLIP links, a direction indicator (``I'' for inbound, ``O'' for outbound), packet type, and compression information are printed out. The packet type is printed first. The three types are ip, utcp, and ctcp. No further link information is printed for ip packets. For TCP packets, the connection identifier is printed following the type. If the packet is compressed, its encoded header is printed out. The special cases are printed out as *S+n and *SA+n, where n is the amount by which the sequence number (or sequence number and ack) has changed. If it is not a special case, zero or more changes are printed. A change is indicated by U (urgent pointer), W (window), A (ack), S (sequence number), and I (packet ID), followed by a delta (+n or -n), or a new value (=n). Finally, the amount of data in the packet and compressed header length are printed.
For example, the following line shows an outbound compressed TCP packet, with an implicit connection identifier; the ack has changed by 6, the sequence number by 49, and the packet ID by 6; there are 3 bytes of data and 6 bytes of compressed header:
O ctcp * A+6 S+49 I+6 3 (6)
Arp/rarp output shows the type of request and its arguments. The format is intended to be self explanatory. Here is a short sample taken from the start of an `rlogin' from host rtsg to host csam:
arp who-has csam tell rtsg arp reply csam is-at CSAMThe first line says that rtsg sent an arp packet asking for the ethernet address of internet host csam. Csam replies with its ethernet address (in this example, ethernet addresses are in caps and internet addresses in lower case).
This would look less redundant if we had done tcpdump -n:
arp who-has 128.3.254.6 tell 128.3.254.68 arp reply 128.3.254.6 is-at 02:07:01:00:01:c4If we had done tcpdump -e, the fact that the first packet is broadcast and the second is point-to-point would be visible:
RTSG Broadcast 0806 64: arp who-has csam tell rtsg CSAM RTSG 0806 64: arp reply csam is-at CSAMFor the first packet this says the ethernet source address is RTSG, the destination is the broadcast address, the type field contained hex 0806 (type ETHER_ARP) and the total length was 64 bytes.
(N.B.:The following description assumes familiarity with the TCP protocol described in RFC-793. If you are not familiar with the protocol, neither this description nor tcpdump will be of much use to you.)
The general format of a tcp protocol line is:
src > dst: flags data-seqno ack window urgent optionsSrc and dst are the source and destination IP addresses and ports. Flags are some combination of S (SYN), F (FIN), P (PUSH) or R (RST) or a single `.' (no flags). Data-seqno describes the portion of sequence space covered by the data in this packet (see example below). Ack is sequence number of the next data expected the other direction on this connection. Window is the number of bytes of receive buffer space available the other direction on this connection. Urg indicates there is `urgent' data in the packet. Options are tcp options enclosed in angle brackets (e.g., <mss 1024>).
Src, dst and flags are always present. The other fields depend on the contents of the packet's tcp protocol header and are output only if appropriate.
Here is the opening portion of an rlogin from host rtsg to host csam.
rtsg.1023 > csam.login: S 768512:768512(0) win 4096 <mss 1024> csam.login > rtsg.1023: S 947648:947648(0) ack 768513 win 4096 <mss 1024> rtsg.1023 > csam.login: . ack 1 win 4096 rtsg.1023 > csam.login: P 1:2(1) ack 1 win 4096 csam.login > rtsg.1023: . ack 2 win 4096 rtsg.1023 > csam.login: P 2:21(19) ack 1 win 4096 csam.login > rtsg.1023: P 1:2(1) ack 21 win 4077 csam.login > rtsg.1023: P 2:3(1) ack 21 win 4077 urg 1 csam.login > rtsg.1023: P 3:4(1) ack 21 win 4077 urg 1The first line says that tcp port 1023 on rtsg sent a packet to port login on csam. The S indicates that the SYN flag was set. The packet sequence number was 768512 and it contained no data. (The notation is `first:last(nbytes)' which means `sequence numbers first up to but not including last which is nbytes bytes of user data'.) There was no piggy-backed ack, the available receive window was 4096 bytes and there was a max-segment-size option requesting an mss of 1024 bytes.
Csam replies with a similar packet except it includes a piggy-backed ack for rtsg's SYN. Rtsg then acks csam's SYN. The `.' means no flags were set. The packet contained no data so there is no data sequence number. Note that the ack sequence number is a small integer (1). The first time tcpdump sees a tcp `conversation', it prints the sequence number from the packet. On subsequent packets of the conversation, the difference between the current packet's sequence number and this initial sequence number is printed. This means that sequence numbers after the first can be interpreted as relative byte positions in the conversation's data stream (with the first data byte each direction being `1'). `-S' will override this feature, causing the original sequence numbers to be output.
On the 6th line, rtsg sends csam 19 bytes of data (bytes 2 through 20 in the rtsg -> csam side of the conversation). The PUSH flag is set in the packet. On the 7th line, csam says it's received data sent by rtsg up to but not including byte 21. Most of this data is apparently sitting in the socket buffer since csam's receive window has gotten 19 bytes smaller. Csam also sends one byte of data to rtsg in this packet. On the 8th and 9th lines, csam sends two bytes of urgent, pushed data to rtsg.
UDP format is illustrated by this rwho packet:
actinide.who > broadcast.who: udp 84This says that port who on host actinide sent a udp datagram to port who on host broadcast, the Internet broadcast address. The packet contained 84 bytes of user data.
Some UDP services are recognized (from the source or destination port number) and the higher level protocol information printed. In particular, Domain Name service requests (RFC-1034/1035) and Sun RPC calls (RFC-1050) to NFS.
(N.B.:The following description assumes familiarity with the Domain Service protocol described in RFC-1035. If you are not familiar with the protocol, the following description will appear to be written in greek.)
Name server requests are formatted as
src > dst: id op? flags qtype qclass name (len) h2opolo.1538 > helios.domain: 3+ A? ucbvax.berkeley.edu. (37)Host h2opolo asked the domain server on helios for an address record (qtype=A) associated with the name ucb-vax.berkeley.edu. The query id was `3'. The `+' indicates the recursion desired flag was set. The query length was 37 bytes, not including the UDP and IP protocol headers. The query operation was the normal one, Query, so the op field was omitted. If the op had been anything else, it would have been printed between the `3' and the `+'. Similarly, the qclass was the normal one, C_IN, and omitted. Any other qclass would have been printed immediately after the `A'.
A few anomalies are checked and may result in extra fields enclosed in square brackets: If a query contains an answer, name server or authority section, ancount, nscount, or arcount are printed as `[na]', `[nn]' or `[nau]' where n is the appropriate count. If any of the response bits are set (AA, RA or rcode) or any of the `must be zero' bits are set in bytes two and three, `[b2&3=x]' is printed, where x is the hex value of header bytes two and three.
Name server responses are formatted as
src > dst: id op rcode flags a/n/au type class data (len) helios.domain > h2opolo.1538: 3 3/3/7 A 128.32.137.3 (273) helios.domain > h2opolo.1537: 2 NXDomain* 0/1/0 (97)In the first example, helios responds to query id 3 from h2opolo with 3 answer records, 3 name server records and 7 authority records. The first answer record is type A (address) and its data is internet address 128.32.137.3. The total size of the response was 273 bytes, excluding UDP and IP headers. The op (Query) and response code (NoError) were omitted, as was the class (C_IN) of the A record.
In the second example, helios responds to query 2 with a response code of non-existent domain (NXDomain) with no answers, one name server and no authority records. The `*' indicates that the authoritative answer bit was set. Since there were no answers, no type, class or data were printed.
Other flag characters that might appear are `-' (recursion available, RA, not set) and `|' (truncated message, TC, set). If the `question' section doesn't contain exactly one entry, `[nq]' is printed.
Note that name server requests and responses tend to be large and the default snaplen of 96 bytes may not capture enough of the packet to print. Use the -s flag to increase the snaplen if you need to seriously investigate name server traffic. `-s 128' has worked well for me.
Sun NFS (Network File System) requests and replies are printed as:
src.xid > dst.nfs: len op args
src.nfs > dst.xid: reply stat len op results
sushi.6709 > wrl.nfs: 112 readlink fh 21,24/10.73165
wrl.nfs > sushi.6709: reply ok 40 readlink "../var"
sushi.201b > wrl.nfs:
144 lookup fh 9,74/4096.6878 "xcolors"
wrl.nfs > sushi.201b:
reply ok 128 lookup fh 9,74/4134.3150
In the first line, host sushi sends a transaction with id
6709 to wrl (note that the number following the src host
is a transaction id, not the source port). The request
was 112 bytes, excluding the UDP and IP headers. The
operation was a readlink (read symbolic link) on file handle
(fh) 21,24/10.731657119. (If one is lucky, as in this
case, the file handle can be interpreted as a major,minor
device number pair, followed by the inode number and
generation number.) Wrl replies `ok' with the contents of
the link.
In the third line, sushi asks wrl to lookup the name `xcolors' in directory file 9,74/4096.6878. Note that the data printed depends on the operation type. The format is intended to be self explanatory if read in conjunction with an NFS protocol spec.
If the -v (verbose) flag is given, additional information is printed. For example:
sushi.1372a > wrl.nfs:
148 read fh 21,11/12.195 8192 bytes @ 24576
wrl.nfs > sushi.1372a:
reply ok 1472 read REG 100664 ids 417/0 sz 29388
(-v also prints the IP header TTL, ID, and fragmentation
fields, which have been omitted from this example.) In
the first line, sushi asks wrl to read 8192 bytes from
file 21,11/12.195, at byte offset 24576. Wrl replies
`ok'; the packet shown on the second line is the first
fragment of the reply, and hence is only 1472 bytes long
(the other bytes will follow in subsequent fragments, but
these fragments do not have NFS or even UDP headers and so
might not be printed, depending on the filter expression
used). Because the -v flag is given, some of the file
attributes (which are returned in addition to the file
data) are printed: the file type (``REG'', for regular
file), the file mode (in octal), the uid and gid, and the
file size.
If the -v flag is given more than once, even more details are printed.
Note that NFS requests are very large and much of the detail won't be printed unless snaplen is increased. Try using `-s 192' to watch NFS traffic.
NFS reply packets do not explicitly identify the RPC operation. Instead, tcpdump keeps track of ``recent'' requests, and matches them to the replies using the transaction ID. If a reply does not closely follow the corresponding request, it might not be parseble.
Appletalk DDP packets encapsulated in UDP datagrams are de-encapsulated and dumped as DDP packets (i.e., all the UDP header information is discarded). The file /etc/atalk.names is used to translate appletalk net and node numbers to names. Lines in this file have the form
number name 1.254 ether 16.1 icsd-net 1.254.110 aceThe first two lines give the names of appletalk networks. The third line gives the name of a particular host (a host is distinguished from a net by the 3rd octet in the number - a net number must have two octets and a host number must have three octets.) The number and name should be separated by whitespace (blanks or tabs). The /etc/atalk.names file may contain blank lines or comment lines (lines starting with a `#').
Appletalk addresses are printed in the form
net.host.port 144.1.209.2 > icsd-net.112.220 office.2 > icsd-net.112.220 jssmag.149.235 > icsd-net.2(If the /etc/atalk.names doesn't exist or doesn't contain an entry for some appletalk host/net number, addresses are printed in numeric form.) In the first example, NBP (DDP port 2) on net 144.1 node 209 is sending to whatever is listening on port 220 of net icsd node 112. The second line is the same except the full name of the source node is known (`office'). The third line is a send from port 235 on net jssmag node 149 to broadcast on the icsd-net NBP port (note that the broadcast address (255) is indicatedn by a net name with no host number - for this reason it's a good idea to keep node names and net names distinct in /etc/atalk.names).
NBP (name binding protocol) and ATP (Appletalk transaction protocol) packets have their contents interpreted. Other protocols just dump the protocol name (or number if no name is registered for the protocol) and packet size.
NBP packets are formatted like the following examples:
icsd-net.112.220 > jssmag.2: nbp-lkup 190: "=:LaserWriter@*" jssmag.209.2 > icsd-net.112.220: nbp-reply 190: "RM1140:LaserWriter@*" 250 techpit.2 > icsd-net.112.220: nbp-reply 190: "techpit:LaserWriter@*" 186The first line is a name lookup request for laserwriters sent by net icsd host 112 and broadcast on net jssmag. The nbp id for the lookup is 190. The second line shows a reply for this request (note that it has the same id) from host jssmag.209 saying that it has a laserwriter resource named "RM1140" registered on port 250. The third line is another reply to the same request saying host techpit has laserwriter "techpit" registered on port 186.
ATP packet formatting is demonstrated by the following example:
jssmag.209.165 > helios.132: atp-req 12266<0-7> 0xae030001 helios.132 > jssmag.209.165: atp-resp 12266:0 (512) 0xae040000 helios.132 > jssmag.209.165: atp-resp 12266:1 (512) 0xae040000 helios.132 > jssmag.209.165: atp-resp 12266:2 (512) 0xae040000 helios.132 > jssmag.209.165: atp-resp 12266:3 (512) 0xae040000 helios.132 > jssmag.209.165: atp-resp 12266:4 (512) 0xae040000 helios.132 > jssmag.209.165: atp-resp 12266:5 (512) 0xae040000 helios.132 > jssmag.209.165: atp-resp 12266:6 (512) 0xae040000 helios.132 > jssmag.209.165: atp-resp*12266:7 (512) 0xae040000 jssmag.209.165 > helios.132: atp-req 12266<3,5> 0xae030001 helios.132 > jssmag.209.165: atp-resp 12266:3 (512) 0xae040000 helios.132 > jssmag.209.165: atp-resp 12266:5 (512) 0xae040000 jssmag.209.165 > helios.132: atp-rel 12266<0-7> 0xae030001 jssmag.209.133 > helios.132: atp-req* 12267<0-7> 0xae030002Jssmag.209 initiates transaction id 12266 with host helios by requesting up to 8 packets (the `<0-7>'). The hex number at the end of the line is the value of the `userdata' field in the request.
Helios responds with 8 512-byte packets. The `:digit' following the transaction id gives the packet sequence number in the transaction and the number in parens is the amount of data in the packet, excluding the atp header. The `*' on packet 7 indicates that the EOM bit was set.
Jssmag.209 then requests that packets 3 & 5 be retransmitted. Helios resends them then jssmag.209 releases the transaction. Finally, jssmag.209 initiates the next request. The `*' on the request indicates that XO (`exactly once') was not set.
Fragmented Internet datagrams are printed as
(frag id:size@offset+) (frag id:size@offset)(The first form indicates there are more fragments. The second indicates this is the last fragment.)
Id is the fragment id. Size is the fragment size (in bytes) excluding the IP header. Offset is this fragment's offset (in bytes) in the original datagram.
The fragment information is output for each fragment. The first fragment contains the higher level protocol header and the frag info is printed after the protocol info. Fragments after the first contain no higher level protocol header and the frag info is printed after the source and destination addresses. For example, here is part of an ftp from arizona.edu to lbl-rtsg.arpa over a CSNET connection that doesn't appear to handle 576 byte datagrams:
arizona.ftp-data > rtsg.1170:
. 1024:1332(308) ack 1 win 4096 (frag 595a:328@0+)
arizona > rtsg: (frag 595a:204@328)
rtsg.1170 > arizona.ftp-data: . ack 1536 win 2560
There are a couple of things to note here: First,
addresses in the 2nd line don't include port numbers.
This is because the TCP protocol information is all in the
first fragment and we have no idea what the port or
sequence numbers are when we print the later fragments.
Second, the tcp sequence information in the first line is
printed as if there were 308 bytes of user data when, in
fact, there are 512 bytes (308 in the first frag and 204
in the second). If you are looking for holes in the
sequence space or trying to match up acks with packets,
this can fool you.
A packet with the IP don't fragment flag is marked with a trailing (DF).
By default, all output lines are preceded by a timestamp. The timestamp is the current clock time in the form
hh:mm:ss.fracand is as accurate as the kernel's clock (e.g., +-10ms on a Sun-3). The timestamp reflects the time the kernel first saw the packet. No attempt is made to account for the time lag between when the ethernet interface removed the packet from the wire and when the kernel serviced the `new packet' interrupt (of course, with Sun's lousy clock resolution this time lag is negligible.)
traffic(1C), nit(4P), bpf(4)
Van Jacobson (van@helios.ee.lbl.gov), Craig Leres (leres@helios.ee.lbl.gov) and Steven McCanne (mccanne@helios.ee.lbl.gov), all of Lawrence Berkeley Laboratory, University of California, Berkeley, CA.
NIT doesn't let you watch your own outbound traffic, BPF will. We recommend that you use the latter.
tcpdump for Ultrix requires Ultrix version 4.0 or later; the kernel has to have been built with the packetfilter pseudo-device driver (see packetfilter(4)). In order to watch either your own outbound or inbound traffic, you will need to use Ultrix version 4.2 or later, and you will have to have used the pfconfig(8) command to enable ``copyall'' mode.
Under SunOS 4.1, the packet capture code (or Streams NIT) is not what you'd call efficient. Don't plan on doing much with your Sun while you're monitoring a busy network.
On Sun systems prior to release 3.2, NIT is very buggy. If run on an old system, tcpdump may crash the machine.
Some attempt should be made to reassemble IP fragments or, at least to compute the right length for the higher level protocol.
Name server inverse queries are not dumped correctly: The (empty) question section is printed rather than real query in the answer section. Some believe that inverse queries are themselves a bug and prefer to fix the program generating them rather than tcpdump.
Apple Ethertalk DDP packets could be dumped as easily as KIP DDP packets but aren't. Even if we were inclined to do anything to promote the use of Ethertalk (we aren't), LBL doesn't allow Ethertalk on any of its networks so we'd would have no way of testing this code.
A packet trace that crosses a daylight savings time change will give skewed time stamps (the time change is ignored).
Filters expressions that manipulate FDDI headers assume that all FDDI packets are encapsulated Ethernet packets. This is true for IP, ARP, and DECNET Phase IV, but is not true for protocols such as ISO CLNS. Therefore, the filter may inadvertently accept certain packets that do not properly match the filter expression.