Created attachment 113722 [details] backtrace I tried to build kauth branch on a Fedora Rawhide (F29) test virtual machine. When I launch partitionmanager I get no device listed. Looking in the settings, I tried to change the backend from sfdisk to Dummy, as soon as I click on apply, partitionmanager crashes with the attached backtrace.
Hmm, seems to work here. I could change to Dummy without a crash. I should probably try to build on F29 too to see if I can reproduce... Strangely, crash happens in the code that reads colours from settings file. Was this a new system? Or did you run older versions of KPM there (although, it shouldn't interfere in my oppinion).
(In reply to Andrius Štikonas from comment #1) > Strangely, crash happens in the code that reads colours from settings file. > Was this a new system? Or did you run older versions of KPM there (although, > it shouldn't interfere in my oppinion). This was a clean new system. I've also tried to uninstall the kauth version, install and run the stable version and the retry with the kauth version, but it still crashes. If you want to check build logs for something strange or odd build flags, here it is: https://copr.fedorainfracloud.org/coprs/mattia/Testing/build/772749/ It's also strange that if I install that build on my real machine which is running F28, when I launch partitionmanager I get the prompt which asks me for authentication; on F29 I get no prompt. In both cases the program can't see any disk.
I've tried to manually compile kauth branches on Fedora 28 (VirtualBox with no devices). Didn't crash even when I switched to dummybackend. But so far I don't see anything wrong with those packages either. P.S. I only compiled with -DCMAKE_INSTALL_PREFIX=/usr I'll try with your RPMs and some devices later...
So today I tried to add a virtual hard drive to my Fedora 28 virtual machine. KPM seemed to work. It correctly asked for password using polkit dialog. I could switch to dummy backend too. Although, later I observed crash on exit which I don't see on my main system... Maybe Qt 5.10?
Hi Mattia, I can see the crash with your packages on F28 too. Yet when I tried self compiled build it worked. I guess we need to try to find out what is different...
(In reply to Andrius Štikonas from comment #5) > Hi Mattia, > > I can see the crash with your packages on F28 too. Yet when I tried self > compiled build it worked. > > I guess we need to try to find out what is different... Maybe one of the security flags injected by Fedora build system leads to this difference? If you can give me any advice I can try to to some build excluding that flag.
(In reply to Mattia from comment #6) > (In reply to Andrius Štikonas from comment #5) > > Hi Mattia, > > > > I can see the crash with your packages on F28 too. Yet when I tried self > > compiled build it worked. > > > > I guess we need to try to find out what is different... > > Maybe one of the security flags injected by Fedora build system leads to > this difference? > If you can give me any advice I can try to to some build excluding that flag. Could be, since it crashes in libc. Well, one fix that might workaround it is to switch away from std::vector to std:array for those colours. But I would like to test it before comitting. I can probably test it by patching your source in srpm with my changes. But things are quite slow since I am working in Virtual Machine and furthermore mouse is not working in VirtualBox (F28) :(.
Oh, actually I now remembered why I changed it from std::array to std::vector. It was to help keep ABI compatibility when adding support to new file systems. So I would like to avoid switching back to std::array. QVector might work though if we can't figure it out. By the way, are you on #fedora-kde?
Ok, I think I can see the bug in the code. I can probably fix it even without being able to reproduce the crash.
Git commit cc43e8b706c400e3a4c1c84671608ca6771766af by Andrius Štikonas. Committed on 07/07/2018 at 13:12. Pushed by stikonas into branch 'kauth'. Fix out of bounds access. M +1 -1 src/util/guihelpers.cpp https://commits.kde.org/partitionmanager/cc43e8b706c400e3a4c1c84671608ca6771766af
Can you test now with this fix? So I guess harderned fedora compile settings indeed caused crash instead of allowing out of bounds access.
By the way, on Fedora I also some some errors where somehow helper was complaining that cryptographic signature is invalid. Have you seen those? I should probably test a bit more later if you compile new RPMs...
I'm waiting for the new build to be completed, unfortunately COPR seems to be down at the moment. https://copr.fedorainfracloud.org/coprs/mattia/Testing/build/774961/ I will test the fix when the build will be completed. I didn't see any error related to cryptographic signature, but in my previous tests I can't get the polkit dialog neither. I'm on #fedora-kde with nick mattiaV, although I'm rarely logged in. Thanks
(In reply to Mattia from comment #13) > I'm waiting for the new build to be completed, unfortunately COPR seems to > be down at the moment. > https://copr.fedorainfracloud.org/coprs/mattia/Testing/build/774961/ > > I will test the fix when the build will be completed. I didn't see any error > related to cryptographic signature, but in my previous tests I can't get the > polkit dialog neither. > > I'm on #fedora-kde with nick mattiaV, although I'm rarely logged in. > > Thanks Well, if polkit dialog does not work, reopen this.
I tested it on Fedora 28 and modulo those "Invalid cryptographic signature" errors which I sometimes (not always) see on Fedora when changing the backend. I haven't tested it on F29.
I confirm all the issues are now fixed. Switching to dummy backend does not make PM crash anymore. About the polkit dialog which wasn't showed, launching PM from console showed a lot of "qca does not support rsa" messages. I've found that by installing package qca-qt5-ossl I can successfully authenticate and see the devices. From console I can also see the "Inavlid cryptographic signature errors" every time I switch the backend. I will open a new ticket for that, because I've noticed some other strange behaviour here.
(In reply to Mattia from comment #16) > I confirm all the issues are now fixed. > Switching to dummy backend does not make PM crash anymore. > > About the polkit dialog which wasn't showed, launching PM from console > showed a lot of "qca does not support rsa" messages. I've found that by > installing package qca-qt5-ossl I can successfully authenticate and see the > devices. > > From console I can also see the "Invalid cryptographic signature errors" > every time I switch the backend. I will open a new ticket for that, because > I've noticed some other strange behaviour here. Hmm, I guess qca came with no backends. Here I have qca with botan backend and it also works, so not just openssl. (I am not and expert on RPM packaging, but if it allows something like OR for dependencies, you might allow different qca backends).