> The phyint_t itself is part of a stack which is associated with a > particular zone, and that zone will have the link kstats of interest.
We store that context in the case of a shared-stack zone today? > You do agree that having all link kstats have system-wide (i.e. > ALL_ZONES) visibility is a bug, right (or at least at odds with the > intended per-zone link namespace)? Yes, but it's tricky to remove because there are consumers in the kernel that may be operating in contexts that do not readily have access to the right state. For instance, as part of Clearview IPMP I found (and removed) a number of places in the kernel that opened DLPI devices by name -- there may still be others lingering (likely using the /devices path, which already seems broken in the case of zones). -- meem
