On Wed, 2022-09-21 at 09:13 +0200, Peter Bergin wrote:

Hi Richard,

On 2022-09-21 00:11, Richard Purdie wrote:

On Tue, 2022-09-20 at 23:05 +0200, Peter Bergin wrote:

Hi,

I'm working on a recipe for libtraceevent
(https://git.kernel.org/pub/scm/libs/libtrace/libtraceevent.git) that is
needed for newer trace-cmd. I'm confused about an error I get during
do_package that I want some help with.

File: '<snip>/sources/poky/meta/lib/oe/sstatesig.py', lineno: 587,
function: process
     0583:                    else:
     0584:                        add_perm(stat.S_IXOTH, 'x')
     0585:
     0586:                    try:
 *** 0587:                        update_hash(" %10s" %
pwd.getpwuid(s.st_uid).pw_name)
     0588:                        update_hash(" %10s" %
grp.getgrgid(s.st_gid).gr_name)
     0589:                    except KeyError as e:
     0590:                        bb.warn("KeyError in %s" % path)
     0591:                        msg = ("KeyError: %s\nPath %s is
owned by uid %d, gid %d, which doesn't match "
Exception: Exception: KeyError: 'getpwuid(): uid not found: 1000'
Path ./package/usr/lib/libtraceevent.a is owned by uid 1000, gid 1004,
which doesn't match any user/group on target. This may be due to host
contamination.

ERROR: Logfile of failure stored in:
<snip>/build/<machine>/tmp/work/cortexa53-crypto-poky-linux/libtraceevent/1.6.2-r0/temp/log.do_package.3574262
ERROR: Task
(<snip>/sources/meta-openembedded/meta-oe/recipes-kernel/trace-cmd/libtraceevent_1.6.2.bb:do_package)
failed with exit code '1'

What happens is that during do_package the directory ${WORKDIR}/package
is moved to ${WORKDIR}/sstate-build-package and the sstate is
calculated. In that sequence in sstatesig.py the owner of the files are
checked and used in the hash calculations. By adding debug prints I have
seen that in normal case the s.st_uid in the code above evaluates to 0
for the files and directories. When looking in the file system my user
(1000/1004) owns them but I guess PSEUDO is involved here to fake root?
For some reason when checking the libraries build by libtraceevent in
./usr/lib/ the files reports to be owned by uid 1000 and the exection
above is raised. It is valid for ./usr/lib/libtraceevent.a and also
other files in the ./usr/lib directory. Before the files in ./usr/lib is
checked files from /usr/include has been checked for libtraceevent and
they have s.st_uid=0.

I've spent some time trying to understand this. Give up now and try this
forum if I can get some ideas. I share the recipe below.

Have you tried putting a:

chown root.root ${D}<files>

into do_install and checking if you force the ownership, whether that
helps?

It helped to add 'chown -R root.root ${D}${libdir}'. Thanks!

But curios to understand why. Anyway when looking into ${D} after I
added chown the files is owned by the build user. I guess this is some
pseudo function to have additional permission somewhere? Is is easy to
explain? And normally I have a feeling that I don't need this, why is
this package special?

If you place a file into ${D} with something like cp, it doesn't have
specific ownership information set. Recipes which use cp to install
files usually have chown/chgrp calls to change the build user to root
or other appropriate ownership.

For this reason, most recipes use "install" which handles this behind
the scenes for many cases.

I'd guess the Makefile in this case uses cp rather than install.

Cheers,

Richard