The motivation of the implementation is to allow any foreign guest OSes
with systemd v259+ to boot using systemd-nspawn(1), otherwise any
systemd service requiring importing credentials will fail, including
journald, D-Bus service and getty. We used to work around this issue by
patching systemd to retry the clone() without CLONE_NEWNS, which is not
and can not be implemented in QEMU user emulation.

mount(8) from util-linux worked prior to this patch because it will
fallback to the old mount(2) syscall if fsmount(8) series of syscalls
is not available.

Tested good on an x86-64 host, with:

- an AOSC OS mips64r6el guest, which boots successfully with full systemd
  support using systemd-nspawn(1).
- a Gentoo ppc (big endian) guest, which boots successfully with full
  systemd support, using systemd-nspawn(1).

Changes from v2:

* Apply the review from Pierrick Bouvier[1], making sure that the buffer
  is unlocked before returning error in the fsconfig handler.
* Clearify some comments in syscall.c.

Changes from v1:

* Add missing guard for FSCONFIG_CMD_CREATE_EXCL in strace.c, since it is
  only available since 6.6.
* Fix some wording issues in the cover letter.

[1]: https://lists.nongnu.org/archive/html/qemu-devel/2026-05/msg07794.html

Xinhui Yang (2):
  linux-user: implement fsmount(2) series of syscalls
  linux-user/strace: add fsmount series of syscalls

 linux-user/strace.c    | 105 +++++++++++++++++++++++++++++++++++++++++
 linux-user/strace.list |  15 ++++++
 linux-user/syscall.c   |  91 +++++++++++++++++++++++++++++++++++
 3 files changed, 211 insertions(+)

-- 
2.52.0


Reply via email to