tcpdump/CONTRIBUTING.md

Some Information for Contributors

Thank you for considering to make a contribution to tcpdump! Please use the guidelines below to achieve the best results and experience for everyone.

How to report bugs and other problems

To report a security issue (segfault, buffer overflow, infinite loop, arbitrary code execution etc) please send an e-mail to security@tcpdump.org, do not use the bug tracker!

To report a non-security problem (failure to compile, incorrect output in the protocol printout, missing support for a particular protocol etc) please check first that it reproduces with the latest stable release of tcpdump and the latest stable release of libpcap. If it does, please check that the problem reproduces with the current git master branch of tcpdump and the current git master branch of libpcap. If it does (and it is not a security-related problem, otherwise see above), please navigate to the bug tracker and check if the problem has already been reported. If it has not, please open a new issue and provide the following details:

• tcpdump and libpcap version (tcpdump --version)
• operating system name and version and any other details that may be relevant (uname -a, compiler name and version, CPU type etc.)
• custom configure/cmake flags, if any
• statement of the problem
• steps to reproduce

Please note that if you know exactly how to solve the problem and the solution would not be too intrusive, it would be best to contribute some development time and to open a pull request instead as discussed below.

Still not sure how to do? Feel free to subscribe to the mailing list and ask!

How to add new code and to update existing code

0. Check that there isn't a pull request already opened for the changes you intend to make.

1. Fork the Tcpdump repository.

2. The easiest way to test your changes on multiple operating systems and architectures is to let the upstream CI test your pull request (more on this below).

3. Setup your git working copy

   git clone https://github.com/<username>/tcpdump.git
   cd tcpdump
   git remote add upstream https://github.com/the-tcpdump-group/tcpdump
   git fetch upstream
   

4. Do a touch .devel in your working directory. Currently, the effect is

   • add (via configure, in Makefile) some warnings options (-Wall, -Wmissing-prototypes, -Wstrict-prototypes, ...) to the compiler if it supports these options,
   • have the Makefile support make depend and the configure script run it.

5. Configure and build

   ./configure && make -s && make check
   

6. Add/update tests The tests directory contains regression tests of the dissection of captured packets. Those captured packets were saved running tcpdump with option -w sample.pcap. Additional options, such as -n, are used to create relevant and reproducible output; -# is used to indicate which particular packets have output that differs. The tests are run with the TZ environment variable set to GMT0, so that UTC, rather than the local time where the tests are being run, is used when "local time" values are printed. The actual test compares the current text output with the expected result (sample.out) saved from a previous version.

   Any new/updated fields in a dissector must be present in a sample.pcap file and the corresponding output file.

   Configuration is set in tests/TESTLIST. Each line in this file has the following format:

   test-name   sample.pcap   sample.out   tcpdump-options
   

   The sample.out file can be produced as follows:

   (cd tests && TZ=GMT0 ../tcpdump -# -n -r sample.pcap tcpdump-options > sample.out)
   

   Or, for convenience, use ./update-test.sh test-name

   It is often useful to have test outputs with different verbosity levels (none, -v, -vv, -vvv, etc.) depending on the code.

7. Test using make check (current build options) and ./build_matrix.sh (a multitude of build options, build systems and compilers). If you can, test on more than one operating system. Don't send a pull request until all tests pass.

8. Try to rebase your commits to keep the history simple.

   git fetch upstream
   git rebase upstream/master
   

   (If the rebase fails and you cannot resolve, issue git rebase --abort and ask for help in the pull request comment.)

9. Once 100% happy, put your work into your forked repository using git push.

10. Initiate and send a pull request. This will trigger the upstream repository CI tests.

Code style and generic remarks

• A thorough reading of some other printers code is useful.

• To help learn how tcpdump works or to help debugging: You can configure and build tcpdump with the instrumentation of functions:

  $ ./configure --enable-instrument-functions
  $ make -s clean all
  

  This generates instrumentation calls for entry and exit to functions. Just after function entry and just before function exit, these profiling functions are called and print the function names with indentation and call level.

  If entering in a function, it prints also the calling function name with file name and line number. There may be a small shift in the line number.

  In some cases, with Clang 11, the file number is unknown (printed '??') or the line number is unknown (printed '?'). In this case, use GCC.

  If the environment variable INSTRUMENT is

  • unset or set to an empty string, print nothing, like with no instrumentation
  • set to "all" or "a", print all the functions names
  • set to "global" or "g", print only the global functions names

  This allows to run:

  $ INSTRUMENT=a ./tcpdump ...
  $ INSTRUMENT=g ./tcpdump ...
  $ INSTRUMENT= ./tcpdump ...
  

  or

  $ export INSTRUMENT=global
  $ ./tcpdump ...
  

  The library libbfd is used, therefore the binutils-dev package is required.

• Put the normative reference if any as comments (RFC, etc.).

• Put the format of packets/headers/options as comments if there is no published normative reference.

• The printer may receive incomplete packet in the buffer, truncated at any random position, for example by capturing with -s size option. This means that an attempt to fetch packet data based on the expected format of the packet may run the risk of overrunning the buffer.

  Furthermore, if the packet is complete, but is not correctly formed, that can also cause a printer to overrun the buffer, as it will be fetching packet data based on the expected format of the packet.

  Therefore, integral, IPv4 address, and octet sequence values should be fetched using the GET_*() macros, which are defined in extract.h.

  If your code reads and decodes every byte of the protocol packet, then to ensure proper and complete bounds checks it would be sufficient to read all packet data using the GET_*() macros.

  If your code uses the macros above only on some packet data, then the gaps would have to be bounds-checked using the ND_TCHECK_*() macros:

  ND_TCHECK_n(p), n in { 1, 2, 3, 4, 5, 6, 7, 8, 16 }
  ND_TCHECK_SIZE(p)
  ND_TCHECK_LEN(p, l)
  

  where p points to the data not being decoded. For ND_CHECK_n(), n is the length of the gap, in bytes. For ND_CHECK_SIZE(), the length of the gap, in bytes, is the size of an item of the data type to which p points. For ND_CHECK_LEN(), l is the length of the gap, in bytes.

  For the GET_*() and ND_TCHECK_* macros (if not already done):

  • Assign: ndo->ndo_protocol = "protocol";
  • Define: ND_LONGJMP_FROM_TCHECK before including netdissect.h
  • Make sure that the intersection of GET_*() and ND_TCHECK_*() is minimal, but at the same time their union covers all packet data in all cases.

  You can test the code via:

  sudo ./tcpdump -s snaplen [-v][v][...] -i lo # in a terminal
  sudo tcpreplay -i lo sample.pcap             # in another terminal
  

  You should try several values for snaplen to do various truncation.

• The GET_*() macros that fetch integral values are:

  GET_U_1(p)
  GET_S_1(p)
  GET_BE_U_n(p), n in { 2, 3, 4, 5, 6, 7, 8 }
  GET_BE_S_n(p), n in { 2, 3, 4, 5, 6, 7, 8 }
  GET_LE_U_n(p), n in { 2, 3, 4, 5, 6, 7, 8 }
  GET_LE_S_n(p), n in { 2, 3, 4, 5, 6, 7, 8 }
  

  where p points to the integral value in the packet buffer. The macro returns the integral value at that location.

  U indicates that an unsigned value is fetched; S indicates that a signed value is fetched. For multi-byte values, BE indicates that a big-endian value ("network byte order") is fetched, and LE indicates that a little-endian value is fetched. n is the length, in bytes, of the multi-byte integral value to be fetched.

  In addition to the bounds checking the GET_*() macros perform, using those macros has other advantages:

  • tcpdump runs on both big-endian and little-endian systems, so fetches of multi-byte integral values must be done in a fashion that works regardless of the byte order of the machine running tcpdump. The GET_BE_*() macros will fetch a big-endian value and return a host-byte-order value on both big-endian and little-endian machines, and the GET_LE_*() macros will fetch a little-endian value and return a host-byte-order value on both big-endian and little-endian machines.

  • tcpdump runs on machines that do not support unaligned access to multi-byte values, and packet values are not guaranteed to be aligned on the proper boundary. The GET_BE_*() and GET_LE_*() macros will fetch values even if they are not aligned on the proper boundary.

• The GET_*() macros that fetch IPv4 address values are:

  GET_IPV4_TO_HOST_ORDER(p)
  GET_IPV4_TO_NETWORK_ORDER(p)
  

  where p points to the address in the packet buffer. GET_IPV4_TO_HOST_ORDER() returns the address in the byte order of the host that is running tcpdump; GET_IPV4_TO_NETWORK_ORDER() returns it in network byte order.

  Like the integral GET_*() macros, these macros work correctly on both big-endian and little-endian machines and will fetch values even if they are not aligned on the proper boundary.

• The GET_*() macro that fetches an arbitrary sequences of bytes is:

  GET_CPY_BYTES(dst, p, len)
  

  where dst is the destination to which the sequence of bytes should be copied, p points to the first byte of the sequence of bytes, and len is the number of bytes to be copied. The bytes are copied in the order in which they appear in the packet.

• To fetch a network address and convert it to a printable string, use the following GET_*() macros, defined in addrtoname.h, to perform bounds checks to make sure the entire address is within the buffer and to translate the address to a string to print:

  GET_IPADDR_STRING(p)
  GET_IP6ADDR_STRING(p)
  GET_MAC48_STRING(p)
  GET_EUI64_STRING(p)
  GET_EUI64LE_STRING(p)
  GET_LINKADDR_STRING(p, type, len)
  GET_ISONSAP_STRING(nsap, nsap_length)
  

  GET_IPADDR_STRING() fetches an IPv4 address pointed to by p and returns a string that is either a host name, if the -n flag wasn't specified and a host name could be found for the address, or the standard XXX.XXX.XXX.XXX-style representation of the address.

  GET_IP6ADDR_STRING() fetches an IPv6 address pointed to by p and returns a string that is either a host name, if the -n flag wasn't specified and a host name could be found for the address, or the standard XXXX::XXXX-style representation of the address.

  GET_MAC48_STRING() fetches a 48-bit MAC address (Ethernet, 802.11, etc.) pointed to by p and returns a string that is either a host name, if the -n flag wasn't specified and a host name could be found in the ethers file for the address, or the standard XX:XX:XX:XX:XX:XX-style representation of the address.

  GET_EUI64_STRING() fetches a 64-bit EUI pointed to by p and returns a string that is the standard XX:XX:XX:XX:XX:XX:XX:XX-style representation of the address.

  GET_EUI64LE_STRING() fetches a 64-bit EUI, in reverse byte order, pointed to by p and returns a string that is the standard XX:XX:XX:XX:XX:XX:XX:XX-style representation of the address.

  GET_LINKADDR_STRING() fetches an octet string, of length length and type type, pointed to by p and returns a string whose format depends on the value of type:

  • LINKADDR_MAC48 - if the length is 6, the string has the same value as GET_MAC48_STRING() would return for that address, otherwise, the string is a sequence of XX:XX:... values for the bytes of the address;

  • LINKADDR_FRELAY - the string is "DLCI XXX", where XXX is the DLCI, if the address is a valid Q.922 header, and an error indication otherwise;

  • LINKADDR_EUI64, LINKADDR_ATM, LINKADDR_OTHER - the string is a sequence of XX:XX:... values for the bytes of the address.

• When defining a structure corresponding to a packet or part of a packet, so that a pointer to packet data can be cast to a pointer to that structure and that structure pointer used to refer to fields in the packet, use the nd_* types for the structure members.

  Those types all are aligned only on a 1-byte boundary, so a compiler will not assume that the structure is aligned on a boundary stricter than one byte; there is no guarantee that fields in packets are aligned on any particular boundary.

  This means that all padding in the structure must be explicitly declared as fields in the structure.

  The nd_* types for integral values are:

  • nd_uintN_t, for unsigned integral values, where N is the number of bytes in the value.
  • nd_intN_t, for signed integral values, where N is the number of bytes in the value.

  The nd_* types for IP addresses are:

  • nd_ipv4, for IPv4 addresses;
  • nd_ipv6, for IPv6 addresses.

  The nd_* types for link-layer addresses are:

  • nd_mac48, for MAC-48 (Ethernet, 802.11, etc.) addresses;
  • nd_eui64, for EUI-64 values.

  The nd_* type for a byte in a sequence of bytes is nd_byte; an N-byte sequence should be declared as nd_byte[N].

• Do invalid packet checks in code: Think that your code can receive in input not only a valid packet but any arbitrary random sequence of octets (packet

  • built malformed originally by the sender or by a fuzz tester,
  • became corrupted in transit or for some other reason).

  Print with: nd_print_invalid(ndo); /* to print " (invalid)" */

• Use struct tok for indexed strings and print them with tok2str() or bittok2str() (for flags). All struct tok must end with { 0, NULL }.

• Avoid empty lines in output of printers.

• A commit message must have:

  First line: Capitalized short summary in the imperative (50 chars or less)
  
  If the commit concerns a protocol, the summary line must start with
  "protocol: ".
  
  Body: Detailed explanatory text, if necessary. Fold it to approximately
  72 characters. There must be an empty line separating the summary from
  the body.
  

• Avoid non-ASCII characters in code and commit messages.

• Use the style of the modified sources.

• Don't mix declarations and code.

• Don't use // for comments. Not all C compilers accept C++/C99 comments by default.

• Avoid trailing tabs/spaces