X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Date: Wed, 13 Apr 2022 00:34:57 +0200 From: "Alexandre Belloni" Subject: Re: [OE-core] [PATCH v3 1/1] apt: add apt selftest to test signed package feeds Message-ID: References: <20220411205036.8298-1-fntoth@gmail.com> <20220411205036.8298-2-fntoth@gmail.com> <6cb8d5aa-dd1f-40ab-8f10-50449f1e8438@gmail.com> <2521f751cc6ff0682c41df568d5ac497a41dc60f.camel@linuxfoundation.org> <3993add1-31c5-0c2d-88ae-e4f11dfb3859@gmail.com> MIME-Version: 1.0 In-Reply-To: <3993add1-31c5-0c2d-88ae-e4f11dfb3859@gmail.com> Content-Type: text/plain; charset="utf-8" Content-Disposition: inline Content-Transfer-Encoding: 8bit List-id: To: Ferry Toth Cc: Richard Purdie , openembedded-core@lists.openembedded.org, Xavier Berger , Alexander Kanavin On 13/04/2022 00:20:40+0200, Ferry Toth wrote: > Hi, > > Op 12-04-2022 om 23:51 schreef Richard Purdie: > > On Tue, 2022-04-12 at 23:48 +0200, Alexandre Belloni wrote: > > > On 12/04/2022 23:32:49+0200, Ferry Toth wrote: > > > > Hi > > > > > > > > Op 12-04-2022 om 16:16 schreef Alexandre Belloni: > > > > > Hello, > > > > > > > > > > On 11/04/2022 22:50:36+0200, Ferry Toth wrote: > > > > > > From: Ferry Toth > > > > > > > > > > > > Since Gatesgarth apt (1.8.2) has become more strict and doesn’t allow unsigned repositories by default. > > > > > > Currently when building images this requirement is worked around by using [allow-insecure=yes] and > > > > > > equivalently when performing selftest. > > > > > > > > > > > > Patches "gpg-sign: Add parameters to gpg signature function" and "package_manager: sign DEB package feeds" > > > > > > enable signed DEB package feeds. This patch adds a runtime test for apt derived from the test_testimage_dnf > > > > > > test. It creates a signed deb package feed, runs a qemu image to install the key and performs some package > > > > > > management. To be able to install the key the gnupg package is added to the testimage. > > > > > > > > > > > > > > > > This went through the autobuilders and it seems this still fails: > > > > > > > > That is disappointing. > > > > > > > > > https://autobuilder.yoctoproject.org/typhoon/#/builders/87/builds/3437/steps/15/logs/stdio > > > > > > > > > > ERROR: package-index-1.0-r0 do_package_index: Could not get gpg version: Command '['/home/pokybuild/yocto-worker/oe-selftest-ubuntu/build/build-st-34525/tmp/hosttools/gpg', '--agent-program=/home/pokybuild/yocto-worker/oe-selftest-ubuntu/build/build-st-34525/tmp/hosttools/gpg-agent|--auto-expand-secmem', '--version', '--no-permission-warning']' returned non-zero exit status 2. > > > > > ERROR: Logfile of failure stored in: /home/pokybuild/yocto-worker/oe-selftest-ubuntu/build/build-st-34525/tmp/work/core2-64-poky-linux/package-index/1.0-r0/temp/log.do_package_index.53841 > > > > > NOTE: recipe package-index-1.0-r0: task do_package_index: Failed > > > > > > > > In fact package_index is failing, which is outside this patch code. > > > > > > > > > ERROR: Task (/home/pokybuild/yocto-worker/oe-selftest-ubuntu/build/meta/recipes-core/meta/package-index.bb:do_package_index) failed with exit code '1' > > > > > > > > > > This was ubuntu 16.04 so maybe gpg on the distro is too old (1.4.20) but > > > > > I'm not sure as I think you are using gnupg-native. > > > > > > > > I would have expected gnupg-native, but the log line above shows hosttools > > > > is being used. But the same would happen for signed rpm and ipk feeds right? > > > > > > > > Did we get the correct one tested? I see 55173d in next and then reverted by > > > > Richard. But that was v2. > > > > > > > > > > This was https://git.yoctoproject.org/poky-contrib/commit/?id=5abda438ce762fc7b8e065e3e9063820c758918e > > This is the correct one. > > > > Just to be sure, I've started on ubuntu1604 both master and this branch, > > > we'll see if this reproduces. > > > > Firstly, this is occurring in the newly added test so this is being triggered by > > the new code. I suspect what is happening is that gnupg-native isn't being built > > before the test and this means that it is falling back to the system gpg. The > > system gpg is too old on that worker so it fails. > > Certainly > > > You can probably reproduce locally by not having a gpg on your build system > > (move it out the way temporarily?). > > Thanks for the tip. Not sure if I can remove the package, but IIUC it's the > executable that needs to be present so I can just move it out of the way. > > > If I'm right (and I'm just guessing), the fix is to add the missing dependency > > to ensure gpg is one we've built. > > I know how to add dependency in a recipe, but where to add here? > > I already have 'bitbake('gnupg-native -c addto_recipe_sysroot')' > Should I run 'bitbake('gnupg-native')' before that? > > I copied these lines from test_testimage_dnf, shouldn't that have similar > problems? > sign_rpm.bbclass has PACKAGE_WRITE_DEPS += "gnupg-native", doesn't that solve this issue? -- Alexandre Belloni, co-owner and COO, Bootlin Embedded Linux and Kernel engineering https://bootlin.com