idmapd: daemonize earlier

daemon_init() calls closeall() to close any "unneeded" file descriptors.
This causes the following issue with idmapd on systems that are
configured to have SSSD handle local users and groups:

1. During startup, rpc.idmapd calls getpwnam("nobody") and
   getgrnam("nobody")
2. sss_nss_mc_get{pw,gr}nam from libnss_sss open the cache files for
   users and groups and store the fd's in a struct sss_cli_mc_ctx.  The
   passwd cache is fd 3 and the group cache is fd 4.
3. idmapd calls daemon_init() which sets fd's 0, 1, and 2 to /dev/null
   and fd 3 to the write end of the pipe that the child uses to report
   it's startup status.  It then closes all fd's >= 4.
4. idmapd then calls event_init() which leads to epoll_create1() which
   returns 4 as the epoll fd.
6. An NFSv4 mount request comes in from a client, triggering an nfsdcb
   callback.
7. idmapd calls getgrgid_r() which leads to a call to
   sss_nss_check_header() which determines that the cache needs to be
   reinitialized.  sss_nss_mc_destroy_ctx() is called, which closes the
   ctx->fd which now corresponds to the epoll file rather than the group
   cache file.
8. event_dispatch() calls epoll_wait() with epfd=4, and -EBADF is
   returned.  idmapd logs the following error and exits with  a nonzero
   status:

rpc.idmapd[650]: main: event_dispatch returns errno 9 (Bad file descriptor)

Moving the deamon_init() call so that it happens before get{pw,gr}nam()
fixes this.

Signed-off-by: Scott Mayhew <smayhew@redhat.com>
Signed-off-by: Steve Dickson <steved@redhat.com>

0 files changed, 0 insertions, 0 deletions