> On 07/10/20 11:29, Misono Tomohiro wrote:
> > Currently it is unknown whether virtiofsd will be built at
> > configuration time. It will be automatically built when dependency is
> > met. Also, required libraries are not clear.
> >
> > To make this clear, add configure option --{enable,disable}-virtiofsd.
> > The default is the same as current (enabled if available) like many
> > other options. When --enable-virtiofsd is given and dependency is not
> > met, we get:
> >
> > ERROR: virtiofsd requires libcap-ng devel, seccomp devel, vhost user
> > and tools support
> >
> > In addition, configuration summary now includes virtiofsd entry:
> >
> > build virtiofs daemon: YES/NO
> >
> > Sidenote: this patch defines CONFIG_VIRTIOFSD for config-host.mak to
> > avoid duplicate dependency check in tools/meson.build.
> >
> > Signed-off-by: Misono Tomohiro <misono.tomoh...@jp.fujitsu.com>
>
> Hi Misono,
>
> can you please handle the option via meson_options.txt? That is, just pass
> the value (auto/enabled/disabled) through
> from configure to meson, and handle the default in tools/meson.build. The
> logic will be something like this:
>
> have_virtiofsd = (targetos == 'linux' and
> 'CONFIG_SECCOMP' in config_host and
> 'CONFIG_LIBCAP_NG' in config_host)
>
> if get_option('virtiofsd').enabled()
> if not have_virtiofsd
> if targetos != 'linux'
> error('virtiofsd requires Linux')
> else
> error('virtiofsd requires libcap-ng-devel and seccomp-devel')
> endif
> endif
> elif get_option('virtiofsd').disabled() or not have_tools or \
> not 'CONFIG_VHOST_USER' in config_host
> have_virtiofsd = false
> endif
>
> if have_virtiofsd
> subdir('virtiofsd')
> endif
>
> This is because, when adding the option, there are some conditions that
> should disable virtiofsd by default but can be
> overridden with --enable-virtiofsd.
>
> More information on how to create a new Meson option can be found in
> docs/devel/build-system.rst.
Hi Paolo
Thanks a lot for the clear explanation. I will update the patch to follow the
meson style.
I realized virtiofsd actually needs tools (i.e. "--disable-tools
--enable-virtiofsd"
does not work with above meson.build) since virtiofsd requires libvhost_user
which will
be built ony when tools are built. So, I will keep the current dependency check
(except 'have_system').
BTW, while testing the updated patch, I noticed current master branch (as of
10/08) fails to execute virtiofsd.
backtrace from coredump shows:
#0 get_relocated_path (dir=0x560f4d2f2ef0 "/usr/local/var/run/virtiofsd") at
../util/cutils.c:924
#1 0x0000560f4baab6da in qemu_get_local_state_pathname
(relative_pathname=relative_pathname@entry=0x560f4bac6cf1 "run/virtiofsd")
at ../util/oslib-posix.c:345
#2 0x0000560f4baa1b09 in fv_socket_lock (se=0x560f4d2f2f20) at
../tools/virtiofsd/fuse_virtio.c:865
#3 fv_create_listen_socket (se=0x560f4d2f2f20) at
../tools/virtiofsd/fuse_virtio.c:906
#4 virtio_session_mount (se=0x560f4d2f2f20) at
../tools/virtiofsd/fuse_virtio.c:968
#5 0x0000560f4ba99725 in main (argc=<optimized out>, argv=<optimized out>) at
../tools/virtiofsd/passthrough_ll.c:2947
So, this is related to:
https://github.com/qemu/qemu/commit/f4f5ed2cbde65acaa1bd88d00cc23fa8bf6b5ed9#diff-ae9b732998587b609c0854bae40b2fe6R924
Adding "qemu_init_exec_dir(argv[0]);" in virtiofs's main() seems solve the
problem,
but is it correct fix?
Regards,
Misono
>
> Thanks,
>
> Paolo
>
> > -have_virtiofsd = (have_system and
> > - have_tools and
> > - 'CONFIG_LINUX' in config_host and
> > - 'CONFIG_SECCOMP' in config_host and
> > - 'CONFIG_LIBCAP_NG' in config_host and
> > - 'CONFIG_VHOST_USER' in config_host)
> > -
> > -if have_virtiofsd
> > +if 'CONFIG_VIRTIOFSD' in config_host
> > subdir('virtiofsd')
> > endif
> >
