[no-relnote] fix e2e failures for nvidia-container-cli test cases

[cdesiniotis]

Newer versions of Ubuntu container images (starting with 26.04) introduced changes that broke nvidia-container-cli tests. In particular, the libcap shared library, the pivot_root binary, and /sbin/ldconfig.real are not included by default. This differs from prior releases. To account for this and fix the test failures, the libcap2 and utils-linux-extra (provided pivot_root) packages have been added as prerequisites that get installed in the nested container runner used in our e2e tests. In addition, /sbin/ldconfig.real is made a symlink to /sbin/ldconfig if it doesn't exist.

The nvidia-container-cli tests create a nested container runner here with the ubuntu image. I suspect the e2e tests started failing on main once ubuntu:latest started corresponding to ubuntu:26.04. Note the differences below between the ubuntu:24.04 and ubuntu:26.04 images.

$ docker run --rm -it ubuntu:24.04
root@587a8a3b954e:/# dpkg -l | grep libcap
ii  libcap-ng0:amd64        0.8.4-2build2                amd64        alternate POSIX capabilities library
ii  libcap2:amd64           1:2.66-5ubuntu2.2            amd64        POSIX 1003.1e capabilities (library)
root@587a8a3b954e:/# ls -ltr /sbin/ldconfig*
-rwxr-xr-x 1 root root 1051280 Jan 30 08:27 /sbin/ldconfig.real
-rwxr-xr-x 1 root root     387 Jan 30 08:27 /sbin/ldconfig
root@587a8a3b954e:/# which pivot_root
/usr/sbin/pivot_root

$ docker run --rm -it ubuntu:26.04
root@d113cb089dcc:/# dpkg -l | grep libcap
ii  libcap-ng0:amd64               0.8.5-4build5                    amd64        alternate POSIX capabilities library
root@d113cb089dcc:/# ls -ltr /sbin/ldconfig*
-rwxr-xr-x 1 root root 1083728 Mar 31 20:35 /sbin/ldconfig
root@d113cb089dcc:/# which pivot_root
root@d113cb089dcc:/#

Note, I have provisioned an Ubuntu 26.04 system and see both libcap2 and utils-linux-extra come by default with a standard ISO. I do notice that only /sbin/ldconfig exists (and not /sbin/ldconfig.real), but this does not cause any issues. I installed the NVIDIA Container Toolkit using our documentation, and [nvidia-container-cli.ldconfig] correctly gets set to @/sbin/ldconfig in /etc/nvidia-container-runtime/config.toml. I am able to run GPU containers without any issue (even when I force that the legacy nvidia-container-runtime-hook gets used).

[cdesiniotis]

/cherry-pick release-1.19

[cdesiniotis]

Latest test failure on main: https://github.com/NVIDIA/nvidia-container-toolkit/actions/runs/25462288168

[coveralls]

Coverage Report for CI Build 25464698998

Coverage remained the same at 43.323%

Details

• Coverage remained the same as the base build.
• Patch coverage: No coverable lines changed in this PR.
• No coverage regressions found.

Uncovered Changes

No uncovered changes found.

Coverage Regressions

No coverage regressions found.

---

Coverage Stats

Relevant Lines:	14856
Covered Lines:	6436
Line Coverage:	43.32%
Coverage Strength:	0.48 hits per line

---

💛 - Coveralls

[github-actions]

🤖 Backport PR created for release-1.19: #1806 ✅