[email protected] wrote:
On (04/24/09 09:42), Erik Nordmark wrote:
Vasumathi and I were talking about this.. even today (with ifconfig)
the plumb and the SLIFADDR are separate operations, triggered from userspace,
so there is a window where the 0'th interface has been plumbed with 0.0.0.0
but not yet SLIFADDR'ed, and anything can happen (including two
SLIFADDRs racing). I suggested that we should add a new ioctl that will
atomically (SLIFFADDR the 0'th interface if it is 0.0.0.0, else do the ADDIF).
Are you concerned about two admins doing their independent ifconfigs, or
what is the source of the race? (Adding kernel complexity to handle
dueling admins isn't worth while.)
no I am concerned that when someone call ipadm_create_addr() the library
has to figure out if the interface has to be created, and if it doesnt,
if it should do SLIFADDR or ADDIF. These decisions cannot be done
atomically from the library.
Why would the library have to be concerned about them not being atomic
*unless* there are dualing admins?
Erik
_______________________________________________
networking-discuss mailing list
[email protected]
