ci: build & test util-linux on Fedora Rawhide via Packit #1924
Conversation
|
So far it seems to be doing the right thing, even though the collection of
|
|
I will take a look. Thank you for reporting. (I will use this comment to note my analysis.)
|
|
That's cool. @mrc0mmand thanks! |
|
@mrc0mmand is there a way to know the version of kernel used for running the test cases? |
Yup, in this case its |
The test suite prints the version at the beginning, See rpm-build:fedora-rawhide-aarch64 in the list of PR checks, then "Details" -> "Dashboard URL" -> "Build Logs" where is complete output. |
|
@karelzak this should be almost ready, and thanks to #1936 I was able to drop one of the There's one remaining fail on It can be reproduced locally using This seems to work fine in an x86_64 container: |
|
On fedora-rawhide-i386, it ends on NS_GET_USERNS: Linux returns this errno when not able to identify the ioctl. IMHO the problem is the mismatch between the 64bit kernel and 32bit userspace. I guess the best would be to detect this situation and mark the test as KNOWN_FAIL. |
It seems 32bit userspace on 64bit kernel return ENOTTY for NS_GET_USERNS ioctl (for example when execute tests in mock environment). Addresses: #1924 Signed-off-by: Karel Zak <kzak@redhat.com>
|
@mrc0mmand so, are we ready to merge it? ;-) |
Indeed, thanks! I've rebased the PR branch, so let's see how it goes :) |
|
@karelzak unfortunately, the i386 build still fails: I'll sprinkle the test with some debug logs. Also, it looks like 825feef broke build in the CodeQL Action: Full log: https://github.com/util-linux/util-linux/actions/runs/3674306157/jobs/6212366513#step:5:937 |
|
Ah, I see, the workaround from 857038d doesn't work, since |
|
There is no elegant way to detect kernel wordsize (32/64 bit) in an environment like mock. There is possible to use /proc/sys/kernel/osrelease (or uname -a), but this is a distribution-specific string. Things like @t-8ch, what about adding to the /proc something like |
|
@karelzak I'll look into it in a bit. |
|
While it would be possible to add a new sysfs file to skip the test, would it not be better to fix the ioctls 32bit compat? For the time being we would still need a workaround in any case. |
|
I have a patch that makes the ioctl work. |
As all parameters and return values of the ioctls have the same representation on both 32bit and 64bit we can reuse the normal ioctl handler for the compat handler via compat_ptr_ioctl(). All nsfs ioctls return a plain "int" filedescriptor which is a signed 4-byte integer type on both 32bit and 64bit. The only parameter taken is by NS_GET_OWNER_UID and is a pointer to a "uid_t" which is a 4-byte unsigned integer type on both 32bit and 64bit. Fixes: 6786741 ("nsfs: add ioctl to get an owning user namespace for ns file descriptor") Reported-by: Karel Zak <kzak@redhat.com> Link: util-linux/util-linux#1924 (comment) Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
As all parameters and return values of the ioctls have the same representation on both 32bit and 64bit we can reuse the normal ioctl handler for the compat handler via compat_ptr_ioctl(). All nsfs ioctls return a plain "int" filedescriptor which is a signed 4-byte integer type on both 32bit and 64bit. The only parameter taken is by NS_GET_OWNER_UID and is a pointer to a "uid_t" which is a 4-byte unsigned integer type on both 32bit and 64bit. Fixes: 6786741 ("nsfs: add ioctl to get an owning user namespace for ns file descriptor") Reported-by: Karel Zak <kzak@redhat.com> Link: util-linux/util-linux#1924 (comment) Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
|
Yes, I've thought about the ioctl too, and it's definitely a way how to fix the core of the problem, but I still think that we need a way how to detect 32 vs. 64 bits. (I mean not for this issue, but as a generic thing.) |
|
@mrc0mmand can you rebase this PR? I guess with recent @t-8ch's changes; it's ready now. |
Let's run at least the unit-tests on Fedora Rawhide via Packit, as it supports a couple of alternative architectures (aarch64, ppc64le, s390x) as well as i*86, which were source of issues in the past.
That indeed seems to have done the trick, thanks! |
|
@karelzak Do we have a concrete usecase to argue for the new sysfs file? unsigned long wrdsz = CHAR_BIT * sizeof(void *);Anything else? PS: all tests on this PR are green. |
|
The packit logs currently all talk about "util-linux-2.38". It would be clearer to have some indication that is actually a development version. |
As all parameters and return values of the ioctls have the same representation on both 32bit and 64bit we can reuse the normal ioctl handler for the compat handler via compat_ptr_ioctl(). All nsfs ioctls return a plain "int" filedescriptor which is a signed 4-byte integer type on both 32bit and 64bit. The only parameter taken is by NS_GET_OWNER_UID and is a pointer to a "uid_t" which is a 4-byte unsigned integer type on both 32bit and 64bit. Fixes: 6786741 ("nsfs: add ioctl to get an owning user namespace for ns file descriptor") Reported-by: Karel Zak <kzak@redhat.com> Link: util-linux/util-linux#1924 (comment) Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
As all parameters and return values of the ioctls have the same representation on both 32bit and 64bit we can reuse the normal ioctl handler for the compat handler via compat_ptr_ioctl(). All nsfs ioctls return a plain "int" filedescriptor which is a signed 4-byte integer type on both 32bit and 64bit. The only parameter taken is by NS_GET_OWNER_UID and is a pointer to a "uid_t" which is a 4-byte unsigned integer type on both 32bit and 64bit. Fixes: 6786741 ("nsfs: add ioctl to get an owning user namespace for ns file descriptor") Reported-by: Karel Zak <kzak@redhat.com> Link: util-linux/util-linux#1924 (comment) Signed-off-by: Thomas Weißschuh <linux@weissschuh.net> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Let's run at least the unit-tests on Fedora Rawhide via Packit, as it supports a couple of alternative architectures (aarch64, ppc64le, s390x) as well as i*86, which were source of issues in the past.