Description of problem:

Test fails sometimes with:

/sbin/swapoff -L swapit
psock_read_event: 0x26d2010 g->state = 3, fd = 3, events = 0x1
roc 175 (swapoff_label) took 0.07 seconds
sock_read_event: 0x26d2010 g->state = 3, fd = 3, events = 0x1
sock_write_event: 0x26d2010 g->state = 3, fd = 3, events = 0x2
sock_write_event: writing 40 bytes ...
sock_write_event: wrote 40 bytes
sock_write_event: done writing, calling send_cb
sock_read_event: 0x26d2010 g->state = 3, fd = 3, events = 0x1
proc 85 (zero) took 0.03 seconds
sock_read_event: 0x26d2010 g->state = 3, fd = 3, events = 0x1
sock_write_event: 0x26d2010 g->state = 3, fd = 3, events = 0x2
sock_write_event: writing 40 bytes ...
sock_write_event: wrote 40 bytes
sock_write_event: done writing, calling send_cb
/sbin/blockdev --rereadpt /dev/sdb
guestfsd: error: /sbin/blockdev: BLKRRPART: Device or resource busy
proc 65 (blockdev_rereadpt) took 0.09 seconds
sock_read_event: 0x26d2010 g->state = 3, fd = 3, events = 0x1
sock_read_event: 0x26d2010 g->state = 3, fd = 3, events = 0x1
/sbin/blockdev: BLKRRPART: Device or resource busy
test_swapon_label_0 FAILED

The proximate cause is this commit:

http://git.et.redhat.com/?p=libguestfs.git;a=commitdiff;h=8eb7f1a58f424ad0f210fe638574b5067d2ac97b

However the underlying reason is that the swapoff operation
doesn't finish before the BLKRRPART ioctl is performed.

In the past we have sometimes seen good effects by adding
calls to udev_settle() at places in the code, although we
don't know if that works just by luck/timing.