Patches are very much welcome!

If you want to get a change included in cjdns, the best thing to do is start by asking in IRC if the change fits the spirit of the project, then developing your change in your own git tree and then asking for it to be merged in with the others.

Small patches are easy to include and large ones are hard to validate. Consider breaking your evil plans up into a bunch of nice little changes which are easy to understand and prove safe.

Rust interop

Cjdns codebase contains both C and Rust code. If you are planning on implementing new functionality, it is better to do it in Rust because it simplifies code review and maintainability.

Dependencies - ask first

If you're planning to add any new dependencies to cjdns, ask before you add them. Binary size is important, cjdns runs on some small devices and pull requests adding large and/or needless dependencies will be rejected. If you want to include a new dependency, have a very good reason and ask before you do the work.

FFI

There are two kinds of foreign function interface in cjdns: cffi and rffi. Cffi is Rust code which is generated from C code and Rffi is C code generated from Rust.

rust/cjdns_sys/src/cffi.rs is generated from rust/cjdns_sys/cffi.h
rust/cjdns_sys/Rffi.h is generated from rust/cjdns_sys/src/rffi.rs

Normally these files are checked in to git and do not change, but if you need to add a C function to be called from Rust, or a Rust function to be called from C, you need to edit the relevant file and then rebuild cjdns_sys with generate-cffi and/or generate-rffi feature.

cd rust/cjdns_sys
cargo build --features generate-rffi
cargo build --features generate-cffi

Running these will update the generated files.

Minutiae:

Cjdns has no issue tracker, this is on purpose, it's for the same reasons that git has no issue tracker.
Everything gets merged to the crashey branch, merges to master only happen during releases, so making your pull requests against crashey is nicest.
Discussion happens on IRC, you can either start by joining and suggesting the change you want to make (or asking if anyone wants to make it for you) or you can do some work, make a pull request and begin the conversation on github.

Codestyle:

Indentation: 4 spaces, tabs are not in the codebase.
Trailing whitespace is not in the codebase, Windows users make sure you have git configured to remove carriage return characters as lines in the codebase are \n delineated.
File names and structures are CamelCase with first letter capital.
All globally visible functions shall begin with the name of the file in which they are defined followed by an underscore and then the name of the function. AKA ThingThatDoesStuff_doStuff()
Functions and local variables shall use camelCase names with first letter lowercase.
Structures declared in header files must begin with, or be, the name of the header in which they are declared.
Global variables as well as static local variables are forbidden. Constants are acceptable.
All preprocessor definitions in header files must contain the name of the header file followed by an underscore and the definition name in all capitals AKA #define SillyMath_VALUE_OF_PI 3 or #define SillyMath_DIVIDE(a,b) (a / b) it is sometimes acceptable for macros to use camel case as is done in Endian.h, use judgement.

If there is a better way, come to irc and announce it, this code style has evolved to where it is now.

Assert.h

Lining your code with assertions is great! You'll find a few macros in Assert.h to help you. If an assertion is cheap or in a cold codepath or you otherwise feel it's important that it's never skipped, use Assert_true(), if you want the assertion to be skipped on small hardware where -DPARANOIA is disabled, use Assert_ifParanoid(), if your assertion might be triggered by "bad nodes" in a realistic network and is for simulation only, use Assert_ifTesting() and it will only be included if -DTESTING is passed. If your assertion has side-effects, you must use Assert_true() because these are macros which are completely removed if they are disabled.

Testing

Any file in a /test/ subdirectory which ends with _test.c will be compiled as a test and added to the testing regime through some nodejs hackery. you might Tests can fail by returning non-zero or using Assert_true() statements, whatever makes sense.

All patches which add tests will be addressed before any patches which don't.

Debugging

    sudo gdb ./cjdroute -ex 'set follow-fork-mode child' -ex 'r < /etc/cjdroute.conf'

If it crashes, type backtrace to get some useful information. The backtrace will show where in the program it crashed and where called that and where called that, etc. Down the left side of the screen are numbers next to the function names, you can select one of these for further inspection using the select-frame command. Once you've selected a frame you can print arguments to the function or local variables (maybe not if compiler optimization was enabled!). The print command will help you extract the value of a local variable, an argument to the function or really anything. In gdb, tab complete works for both commands and variables/arguments so if in doubt, hit tab :)

To stop the program in the debugger, use ctrl+c, this will put you in the debugger shell.

Once in the debugger shell, to quit the debugger use ctrl+d, if the program is running it will prompt you, another ctrl+d will be taken as a "yes, please quit".

Valgrind

Generally doesn't work because Poly1305 (on most processors) does crypto on floating point registers and valgrind interprets the assembly and does not properly interpret floating point instructions so under valgrind tests will fail and the engine will not run correctly. See: http://www.valgrind.org/docs/manual/manual-core.html#manual-core.limits

Profiling

The best way to profile cjdns is using Brendan Gregg's FlameGraph generator. http://www.brendangregg.com/FlameGraphs/cpuflamegraphs.html You can do this on Linux using the perf utility.

sudo perf record -a -g -F 997 -p `pidof cjdroute`
# let this run for a while, put cjdroute through some exercises
<ctrl+c>
sudo perf script | ../FlameGraph/stackcollapse-perf.pl > ./cjdns-stackcollapse.out
../FlameGraph/flamegraph.pl < ./cjdns-stackcollapse.out > ./cjdns-stackcollapse.svg
chromium ./cjdns-stackcollapse.svg

Simulating

Cjdns comes with it's own simulator, it will create n nodes and link them together however you wish. It's like having many cjdns processes all running together but they're all in the same process so it is much more efficient. You can set admin credentials on one node and then use the admin tools to access it as you would an ordinary router. You will however need private keys whose public keys hash to ip addresses beginning with fc. To make these keys, use the makekeys utility.

./makekeys | head -n 32 > keys.txt

To convert the list of keys into a simulator configuration, use makesim.js, note there are interesting constants inside of makesim.js which you might want to alter.

node ./tools/lib/makesim.js keys.txt > ~/my-cjdns-simulation.json

Once you have a simulation setup, you may want to add your admin credentials to one of the nodes so you can inspect it, dump the table, etc...

Example simulation config entry with added admin block:

    "fc5c:0537:606a:3d7e:c9f0:2103:4dcd:6bc8": {
      "privateKey": "0dc3d33bbffc2d16c175df463110c6d164714a40d23db2f83539664b7365a5b6",
      "peers": [
        "fc1c:84bf:9557:1ce4:4862:fd82:5105:9984",
        "fc1c:90e4:2953:938b:47cc:65c3:6540:dff6",
        "fc1b:7ea7:911f:f2d5:7685:ac22:6c4f:8b15"
      ],
      "admin":
      {
          "bind": "127.0.0.1:11234",
          "password": "the_password_you_will_use_to_connect"
      }
    },

And to start it up (in the debugger):

gdb ./sybilsim -ex 'r < ~/my-cjdns-simulation.json'

BUG: Sometimes the simulator doesn't really start up correctly! If you could figure out what is going wrong, your help would be most appreciated, if not, you can just quit and then restart it again and it should start up ok.

HACKING.md 8.2 KB Permalink History Raw