1 Some Information for Contributors
2 ---------------------------------
3 Thank you for considering to make a contribution to tcpdump! Please use the
4 guidelines below to achieve the best results and experience for everyone.
7 How to report bugs and other problems
8 -------------------------------------
9 To report a security issue (segfault, buffer overflow, infinite loop, arbitrary
10 code execution etc) please send an e-mail to security@tcpdump.org, do not use
13 To report a non-security problem (failure to compile, incorrect output in the
14 protocol printout, missing support for a particular protocol etc) please check
15 first that it reproduces with the latest stable release of tcpdump and the latest
16 stable release of libpcap. If it does, please check that the problem reproduces
17 with the current git master branch of tcpdump and the current git master branch of
18 libpcap. If it does (and it is not a security-related problem, otherwise see
19 above), please navigate to https://github.com/the-tcpdump-group/tcpdump/issues
20 and check if the problem has already been reported. If it has not, please open
21 a new issue and provide the following details:
23 * tcpdump and libpcap version (tcpdump --version)
24 * operating system name and version and any other details that may be relevant
25 (uname -a, compiler name and version, CPU type etc.)
26 * custom configure/CMake flags, if any
27 * statement of the problem
30 Please note that if you know exactly how to solve the problem and the solution
31 would not be too intrusive, it would be best to contribute some development time
32 and to open a pull request instead as discussed below.
34 Still not sure how to do? Feel free to [subscribe](https://www.tcpdump.org/#mailing-lists)
35 to the mailing list tcpdump-workers@lists.tcpdump.org and ask!
38 How to add new code and to update existing code
39 -----------------------------------------------
41 0) Check that there isn't a pull request already opened for the changes you
44 1) Fork the Tcpdump repository on GitHub from
45 https://github.com/the-tcpdump-group/tcpdump
46 (See https://help.github.com/articles/fork-a-repo/)
48 2) Setup optional continuous integration (CI) builds
49 You can setup Travis CI, Cirrus CI and AppVeyor builds for your fork to
50 test your changes on Linux, FreeBSD, macOS and Windows before opening a
52 (See https://docs.travis-ci.com/user/tutorial/ for information
53 on setting up Travis CI; go to https://ci.appveyor.com/login and log
54 in with your GitHub account and select "NEW PROJECT" to set up an
57 3) Setup your git working copy
58 git clone https://github.com/<username>/tcpdump.git
60 git remote add upstream https://github.com/the-tcpdump-group/tcpdump
63 4) Do a 'touch .devel' in your working directory.
64 Currently, the effect is
65 a) add (via configure, in Makefile) some warnings options ( -Wall
66 -Wmissing-prototypes -Wstrict-prototypes, ...) to the compiler if it
67 supports these options,
68 b) have the Makefile support "make depend" and the configure script run it.
70 5) Configure and build
71 ./configure && make -s && make check
74 The tests directory contains regression tests of the dissection of captured
75 packets. Those captured packets were saved running tcpdump with option "-w
76 sample.pcap". Additional options, such as "-n", are used to create relevant
77 and reproducible output; "-#" is used to indicate which particular packets
78 have output that differs. The tests are run with the TZ environment
79 variable set to GMT0, so that UTC, rather than the local time where the
80 tests are being run, is used when "local time" values are printed. The
81 actual test compares the current text output with the expected result
82 (sample.out) saved from a previous version.
84 Any new/updated fields in a dissector must be present in a sample.pcap file
85 and the corresponding output file.
87 Configuration is set in tests/TESTLIST.
88 Each line in this file has the following format:
89 test-name sample.pcap sample.out tcpdump-options
91 The sample.out file can be produced as follows:
92 (cd tests && TZ=GMT0 ../tcpdump -# -n -r sample.pcap tcpdump-options > sample.out)
94 Or, for convenience, use "./update-test.sh test-name"
96 It is often useful to have test outputs with different verbosity levels
97 (none, -v, -vv, -vvv, etc.) depending on the code.
99 7) Test with 'make check'
100 Don't send a pull request if 'make check' gives failed tests.
102 8) Try to rebase your commits to keep the history simple.
103 git rebase upstream/master
104 (If the rebase fails and you cannot resolve, issue "git rebase --abort"
105 and ask for help in the pull request comment.)
107 9) Once 100% happy, put your work into your forked repository.
109 This will trigger your fork CI tests, if set up in step 2.
111 10) Initiate and send a pull request
112 (See https://help.github.com/articles/using-pull-requests/)
113 This will trigger the upstream repository CI tests.
116 Code style and generic remarks
117 ------------------------------
118 a) A thorough reading of some other printers code is useful.
120 b) Put the normative reference if any as comments (RFC, etc.).
122 c) Put the format of packets/headers/options as comments if there is no
123 published normative reference.
125 d) The printer may receive incomplete packet in the buffer, truncated at any
126 random position, for example by capturing with '-s size' option.
127 If your code reads and decodes every byte of the protocol packet, then to
128 ensure proper and complete bounds checks it would be sufficient to read all
129 packet data using the GET_*() macros, typically:
132 GET_BE_U_n(p), n in { 2, 3, 4, 5, 6, 7, 8 }
133 GET_BE_S_n(p), n in { 2, 3, 4, 5, 6, 7, 8 }
134 If your code uses the macros above only on some packet data, then the gaps
135 would have to be bounds-checked using the ND_TCHECK_*() macros:
136 ND_TCHECK_n(p), n in { 1, 2, 3, 4, 5, 6, 7, 8, 16 }
139 For the ND_TCHECK_* macros (if not already done):
140 Assign: ndo->ndo_protocol = "protocol";
141 Define: ND_LONGJMP_FROM_TCHECK before including netdissect.h
142 Make sure that the intersection of GET_*() and ND_TCHECK_*() is minimal,
143 but at the same time their union covers all packet data in all cases.
144 You can test the code via:
145 sudo ./tcpdump -s snaplen [-v][v][...] -i lo # in a terminal
146 sudo tcpreplay -i lo sample.pcap # in another terminal
147 You should try several values for snaplen to do various truncation.
149 e) Do invalid packet checks in code: Think that your code can receive in input
150 not only a valid packet but any arbitrary random sequence of octets (packet
151 - built malformed originally by the sender or by a fuzz tester,
152 - became corrupted in transit or for some other reason).
153 Print with: nd_print_invalid(ndo); /* to print " (invalid)" */
155 f) Use 'struct tok' for indexed strings and print them with
156 tok2str() or bittok2str() (for flags).
158 g) Avoid empty lines in output of printers.
160 h) A commit message must have:
161 First line: Capitalized short summary in the imperative (70 chars or less)
163 Body: Detailed explanatory text, if necessary. Fold it to approximately
164 72 characters. There must be an empty line separating the summary from
167 i) Avoid non-ASCII characters in code and commit messages.
169 j) Use the style of the modified sources.
171 k) Don't mix declarations and code
173 l) Don't use // for comments
174 Not all C compilers accept C++/C99 comments by default.
176 m) Avoid trailing tabs/spaces