% Examples of CCS Reports % Bradley M. Kuhn & Karen M. Sandler % Tuesday 9 May 2017 # CCS Complete, Corresponding Source # How GPLv3 says CCS.
> The "Corresponding Source" for a work in object code form means all the source code needed to generate, install, and (for an executable work) run the object code and to modify the work, including scripts to control those activities.

— GPLv3§1

# How GPLv2 says CCS.
> You may copy and distribute the Program (or a work based on it, under > § 2) in object code or executable form under the terms of § 1 > & 2 above provided that you … [a]ccompany it with the complete > corresponding machine-readable source code … The source code for a > work means the preferred form of the work for making modifications to it. > For an executable work, complete source code means all the source code for > all modules it contains, plus any associated interface definition files, > plus the scripts used to control compilation and installation of the > executable.

— GPLv2§3

# The 11 Words That Consumed My Life + GPLv2 enforcement, for embedded products, is all about the these eleven words. + I could give an entire talk on any one of these 11 words. + Yes, I can even give 20-30 minute treatises on each use of “the”. + Yet, when enforcement processes are at their best, they're about the spirit behind these words, not the words themselves.
> the scripts used to control compilation and installation of the executable.

— GPLv2§3

# The 11 Words That Consumed My Life + Basic reference rule: + Can a developer reasonably skilled in the art of embedded software build your sources, take the (copylefted) executables and install them? + Enforcement spends its most attention on testing CSS “candidates” to verify that.
> the scripts used to control compilation and installation of the executable.

— GPLv2§3

# CCS "Round" Reports + Evaluate each CCS candidate. + Someone "skilled in the art" attempts to build. + Detailed notes are sent, asking for new CCS candidate "round". + Some anonyomized real world examples. # No Build Instructions
The primary issues we found were a dearth of build instructions as well as a complete lack of installation instructions. There was no information that mentioned how one might build each package so we had to guess about which Makefile and/or build script to run for each package. And in many cases it was not possible or straight-forward to build - this must be resolved in the next source candidate. # Making General Recommendations
We generally recommend that the source release be a single file (ie. one tarball containing all packages required for the build) that includes a README or similar in the main directory explaining exactly how to build and install all of the packages. See section 21.2 of http://compliance.guide/pristine-example for an excellent example. # Suspicious, But Not Captious.
Also, we noticed that some packages mentioned in the "OPEN SOURCE SOFTWARE NOTICE" included with the device (and also downloaded as part of the source release; see Open_Source_Software_Notice_and_Privacy_Policy.pdf ) could not be found in the source release. For example, we found "Software: Samba 3.0.XX" in the notice, but could not find any trace of Samba in the source release. Please ensure that all the software included in the notice is included in the source release as well. # Nesting Doll Packages
Once extracted, the 3 .rar files above produce the following output files: * busybox-1.21.1.rar * AB_A0101.123.tar.gz * source.rar * a small text file that gives two-word descriptions of the above files # Actual(ly Trying to) Build
This file included no instructions for how one might build it so we tried to run "make" but received the following error: $ make .../busybox-1.21.1/scripts/gcc-version.sh: line 11: arm-none-linux-gnueabi-gcc: command not found # Toolchain? + The toolchain is rarely considered mandatory as part of “the scripts”. + Admittedly, it doesn't *control* compilation, it *is* compilation. + The script here is explaining precisely what type of toolchain is needed. + Something like: “GCC vX built with the following ./configure line” is usually adequate. + But including the toolchain is a nice step to make it easy for your users.
> the scripts used to **control compilation** and installation of the executable.

— GPLv2§3

# We Guess at Compiler Anyway
So we searched for an arm-none-linux-gnueabi- cross-compiler in the other files but could not find one. We then tried to use our own (be editing the PATH appropriately), which did get us past this error. Note that this is not acceptable in a source release - the cross-compiler that a user must use needs to be clearly indicated (name, version, etc.) and/or included with the source release. # Feedback on Small Problems
Once we had the custom cross-compiler configured, we then ran into these errors: $ make .../busybox-1.21.1/scripts/gen_build_files.sh: Permission denied make: *** [gen_build_files] Error 127 $ make .../busybox-1.21.1/scripts/mkconfigs: Permission denied make: *** [include/config/MARKER] Error 126 $ make /bin/sh: applets/usage_compressed: Permission denied make[1]: *** [include/usage_compressed.h] Error 126 make: *** [applets_dir] Error 2 $ make .../busybox-1.21.1/scripts/trylink: Permission denied make: *** [busybox_unstripped] Error 126 In each case, we found the mentioned file and then added executable permissions to it (ie. "chmod u+x scripts/gen_build_files.sh"). This must be fixed in the next source release - please set the executable bits on the above files appropriately in the archive file you distribute. # Install Instructions missing
After fixing the above, a "busybox" binary was generated. However, there were no instructions to indicate how one might install this binary on the device. Such instructions are required by GPLv2, under which BusyBox is licensed. Please include the instructions in your next source release. # Build "Only Seems" To build
For the AAB_A0101.123.tar.gz package, we ran "./build.sh", the build took about 140 seconds, which is less than one would expect for building all of the programs listed in the "OPEN SOURCE SOFTWARE NOTICE". The only files we could immediately find that were clearly the result of this "./build.sh" invocation were some kernel image binaries, found in path/path/path/path/path/KERNEL_OBJ . This path was not mentioned at all and we had to guess at where they might be. # Maybe Proprietary Kernel Modules?
Furthermore, there were no .ko files generated, which is abnormal for a build of the kernel, Linux. Please ensure that all .ko files which are used on the system are generated with "./build.sh" or a similar script. # Weird versioning
* The following libraries have different versions in the firmware than is built from the candidate CCS. Specifically, your candidate CCS contains version "1800", and the firmware has version "2400". Since most of these libraries are licensed under the LGPL, you are required to have the complete, corresponding source present for the correct version as distributed in the firmware. You also must include the "scripts to control compilation and installation of the executable". * lib/libgio-2.0.so.0.2400.2 * lib/libglib-2.0.so.0.2400.2 * lib/libgmodule-2.0.so.0.2400.2 * lib/libgobject-2.0.so.0.2400.2 * lib/libgthread-2.0.so.0.2400.2 * lib/libz.so.1.2.5 (version 1.2.2 is provided in the sources) # Weird Build Issues Over Many Candidates
You mentioned in your Round 6 commentary that you have corrected the thatlib issues. However, we are unable to see what you mean. There are now two copies of thatlib, one in 2624.7_524/uclinux-rootfs/lib/thatlib/, as well as the one in yourlibs. We aren't sure which one you intend to be built to generate the binaries on the firmware. When we try to build the yourlibs one from scratch, by cleaning the whole area, we get the following build issues. Here's what we did: # Getting Really build-technical
We ran: make -C libsrc/thatlib install which did not work because of a missing Makefile error. We read the build source and discovered that the Makefile, etc, for that directory is generated by running: cd libsrc/thatlib/thatlib-0.9.22_mipsel-uclibc; sh configure_thatlib_mipsel-uclibc Once we did that make -C libsrc/thatlib install worked correctly. The only remaining binaries were in build source and discovered that the Makefile, etc, for that directory is generated by running: cd libsrc/thatlib/thatlib-0.9.22_mipsel-uclibc; sh configure_thatlib_mipsel-uclibc # Getting Really build-technical
Once we did that make -C libsrc/thatlib install worked correctly. The only remaining binaries were in ./libsrc/thatlib/\{YOURLIB_ROOT_DIR\}/ which looks like a build with a misconfigured environment somehow, so we simply removed that directory. Then, after running make clean, thatlib failed with the following errors. Random .o/.so files laying around in the thatlib source directory, and then it failing to build correctly after they are removed. If there some set of .so files you claim are not required as part of the C&CS since thatlib is LGPL'd, we understand that, but the rest of the sources must build and install those other .so's. Here's the build error we get in the bdvdlibs version: # Getting Really build-technical
mkdir .libs/libthatlibwm_default.a.tmp (cd .libs/libthatlibwm_default.a.tmp && ar x ../../.libs/libthatlibwm_default.a) mkdir .libs/libthatlibwm_default.a.tmp (cd .libs/libthatlibwm_default.a.tmp && ar x ../../.libs/libthatlibwm_default.a) /opt/toolchains/crosstools_sf-linux-2.6.18.0_gcc-4.2-9ts_uclibc-nptl-0.9.29-20070423_20080702/bin//mipsel-uclibc- ld -o libthatlibwm_default.o -r .libs/libthatlibwm_default.a.tmp/*.o /opt/toolchains/crosstools_sf-linux-2.6.18.0_gcc-4.2-9ts_uclibc-nptl-0.9.29-20070423_20080702/bin//mipsel-uclibc- ld: .libs/libthatlibwm_default.a.tmp/default.o: Relocations in generic ELF (EM: 3) /opt/toolchains/crosstools_sf-linux-2.6.18.0_gcc-4.2-9ts_uclibc-nptl-0.9.29-20070423_20080702/bin//mipsel-uclibc- ld: .libs/libthatlibwm_default.a.tmp/default.o: Relocations in generic ELF (EM: 3) /opt/toolchains/crosstools_sf-linux-2.6.18.0_gcc-4.2-9ts_uclibc-nptl-0.9.29-20070423_20080702/bin//mipsel-uclibc- ld: .libs/libthatlibwm_default.a.tmp/default.o: Relocations in generic ELF (EM: 3) /opt/toolchains/crosstools_sf-linux-2.6.18.0_gcc-4.2-9ts_uclibc-nptl-0.9.29-20070423_20080702/bin//mipsel-uclibc- ld: .libs/libthatlibwm_default.a.tmp/default.o: Relocations in generic ELF (EM: 3) .libs/libthatlibwm_default.a.tmp/default.o: could not read symbols: File in wrong format make[4]: *** [libthatlibwm_default.o] Error 1 # Proprietary Linux Modules Are Everywhere
We did find one .ko file that was already included in the package, but wasn't built when we ran "./build.sh". This is path/path/android_X.X/device/device-type/mydevice.ko , which notes that its license is "GPL v2" in the modinfo, but for which we could find no source code in the source release. Please ensure that the source code for mydevice.ko is included in the next source candidate. # Proprietary Linux Modules Are Everywhere
* The following files are derivative of the kernel named Linux and therefore covered by the GPL. However, no source code, scripts to control compilation nor installation are included in your CCS candidate: lib/modules/myfilesystem.ko lib/modules/mydevicecontroller.ko lib/modules/myblockdevice.ko lib/modules/mypcicard.ko # Non-Technical GPL Compliance Issues
Regarding over the air updates: we'd like to see a screenshot or other details documenting what has now been implemented by BestBuy to make sure the offer for source appears to users appropriately after upgrade. There was a consensus reached on the last conference call how this would be done, so we only need follow up and implementation on that. # Binary Comparison.
Note that we did not receive a firmware image to compare this with (though we do have the device). Company's website did not appear to have any firmware images available for download. It would be helpful to have such an image for the next CCS check. The above source candidate was downloaded from http:///sourcez.company.com/en/search/index.htm?keywords=X1234Y, which was alluded to in Company's 2017-01-18 email to us that said: "You can check this website http://sourcez.company.com/en/search/index.htm " The email did not mention how to use that website, but we found that by entering "X1234Y" into the top right search box that we could find the source file list. Note that the offer for source included in the web UI of the device said to email NAME@COMPANY.com , which is how the above instructions for downloading the source were received. # A Pristine Example + Enforcement must often use a “know it when I see it” standard. + i.e., can we take your CCS build it, and install it? + We've reached compliant CCS with hundreds of companies: + but that didn't mean the CCS was pretty. + Thanks to ThinkPenguin, we finally have an example of beautiful embedded product compliance. # Lessons Learned from Pristine Example + The full paper for this talk is available online: + [compliance.guide/pristine-example](http://compliance.guide/pristine-example) + It's part of the larger tutorial called [*Copyleft and the GNU General Public License: A Comprehensive Tutorial and Guide*](https://copyleft.org/guide/) at copyleft.org. # Give a roadmap in a README + Scripts doesn't only mean shell scripts and Makefiles. + Think of the script of a play or movie. + If your build process includes human intervention … + … then the script are a written explanation of what the human must do.
> **the scripts** used to control compilation and installation of the executable.

— GPLv2§3

# ThinkPengiun's README A file called “README” at the top-level directory said: In order to build firmware images for your router, the following needs to be installed: gcc, binutils, bzip2, flex, python, perl, make, find, grep, diff, unzip, gawk, getopt, libz-dev and libc headers. Please use “make menuconfig” to configure your appreciated configuration for the toolchain and firmware. Please note that the default configuration is what was used to build the firmware image for your router. It is advised that you use this configuration. Simply running “make” will build your firmware. The build system will download all sources, build the cross-compile toolchain, the kernel and all chosen applications. To build your own firmware you need to have access to a GNU/Linux system (case-sensitive filesystem required). # Make Sure It Builds + Can your CCS pass this test? + Give you source release to another developer from another department. + Ask them to follow the instructions you wrote. + They should get the equivalent binaries you get in building. + Very few organizations bother to do this. + It's probably the most useful step to verify compliance, yet *no* compliance process recommendations I've ever seen include this.
> the scripts used to **control compilation** and installation of the executable.

— GPLv2§3

# It's not “make install” + Server system software can offer a “make install” that reasonable works to meet installation requirements. + Embedded products are admittedly difficult to install. + To comply here, you'll usually just have write out the instructions. + It is required; don't skip this part.
> the scripts used to **control** compilation and **installation** of the executable.

— GPLv2§3

# Missing hardware components + Inclusion of specialized installation hardware is not a “script”. + In our ThinkPenguin example, we had to go buy a USB serial adapter to install the modified firmware. + Just tell the user what they have to go buy for the install to work. # More Info / Talk License + Specific Sections of Copyleft Guide relating to these topics: - [The Pristine Example](https://copyleft.org/guide/comprehensive-gpl-guidech22.html#x29-15900021) - [Details of a Compliant Distribution](https://copyleft.org/guide/comprehensive-gpl-guidech16.html#x21-12700015)

Presentation and slides are: Copyright © Bradley M. Kuhn (2008–2011, 2015, 2017), Karen M. Sandler (2017), and are licensed under the Creative Commons Attribution-Share Alike 4.0 International License.