This is implemented in a way similar to lib/slists.h, but it took some
more effort to make rehashing not disturb the readers. We do it by just
taking _highest_ k bits of ipa_hash as our hash value and sorting each
box by whole ipa_hash().
Consult FIB_ITERATE_* macros in nest/route.h.
Implemented fib_check() debugging function and also rewrote the rehashing
algorithm to use better thresholds and not to waste time by rehashing
forth and back.
we can use very simple function which is monotonic with respect
to re-hashing:
n ^= n >> 16;
n ^= n << 10;
h = (n >> (16 - o)) & ((1 << o) - 1);
where o is table order. Statistical analysis for both backbone routing
table and local OSPF routing tables gives values near theoretical
optimum for uniform distribution (see ips.c for formulae).
The trick is very simple: We always calculate a 16-bit hash value n and
use o most significant bits (this gives us monotonity wrt. rehashing
if we sort the chains by the value of n). The first shift/xor pair
reduces the IP address to a 16-bit one, the second pair makes higher
bits of the 16-bit value uniformly distributed even for tables containing
lots of long prefixes (typical interior routing case with 24-bit or even
longer prefixes).
o Interface syncing is now a part of krt and it can have configurable
parameters. Actually, the only one is scan rate now :)
o Kernel routing table syncing is now synchronized with interface
syncing (we need the most recent version of the interface list
to prevent lots of routes to non-existent destinations from
appearing). Instead of its own timer, we just check if it's
route scan time after each iface list scan.
o Syncing of device routes implemented.
o CONFIG_AUTO_ROUTES should control syncing of automatic device routes.
o Rewrote krt_remove_route() to really remove routes :)
o Better diagnostics.
o Fixed a couple of bugs.
o rte can now contain a pointer to both cached and uncached rta. Protocols
which don't need their own attribute caching can now just fill-in a rta,
link it to rte without any calls to attribute cache and call rte_update()
which will replace rte->attrs by a cached copy.
o In order to support this, one of previously pad bytes in struct rta
now holds new attribute flags (RTAF_CACHED). If you call rte_update()
with uncached rta, you _must_ clear these flags. In other cases rta_lookup()
sets it appropriately.
o Added rte_free() which is useful when you construct a rte and then the
circumstances change and you decide not to use it for an update. (Needed
for temporary rte's in kernel syncer...)
- cfg_strcpy() -> cfg_strdup()
- mempool -> linpool, mp_* -> lp_* [to avoid confusion with memblock, mb_*]
Anyway, it might be better to stop ranting about names and do some *real* work.
intended to serve as an example of interface pattern list use. As a side
effect, you can disable generating of device routes by disabling
this protocol.
o iface_patt_match(list, iface) -- match interface against list
o iface_patts_equal(a, b, c) -- compare whether two pattern lists are
equal or not. c(x,y) is called for comparison of protocol-dependent
data.
o cfg_alloc(size) -- generic memory allocation
o cfg_allocu(size) -- unaligned memory allocation
o cfg_allocz(size) -- zeroed memory allocation
o cfg_strcpy(str) -- allocate a copy of a string
Also fixed a bug in lexing of string literals.
