bird/filter/trie.c
Ondrej Zajicek (work) 13225f1dbf Filter: Faster prefix sets
Use 16-way (4bit) branching in prefix trie instead of basic binary
branching. The change makes IPv4 prefix sets almost 3x faster, but
with more memory consumption and much more complicated algorithm.

Together with a previous filter change, it makes IPv4 prefix sets
about ~4.3x faster and slightly smaller (on my test data).
2021-09-25 16:06:43 +02:00

626 lines
17 KiB
C

/*
* Filters: Trie for prefix sets
*
* Copyright 2009 Ondrej Zajicek <santiago@crfreenet.org>
*
* Can be freely distributed and used under the terms of the GNU GPL.
*/
/**
* DOC: Trie for prefix sets
*
* We use a (compressed) trie to represent prefix sets. Every node
* in the trie represents one prefix (&addr/&plen) and &plen also
* indicates the index of the bit in the address that is used to
* branch at the node. If we need to represent just a set of
* prefixes, it would be simple, but we have to represent a
* set of prefix patterns. Each prefix pattern consists of
* &ppaddr/&pplen and two integers: &low and &high, and a prefix
* &paddr/&plen matches that pattern if the first MIN(&plen, &pplen)
* bits of &paddr and &ppaddr are the same and &low <= &plen <= &high.
*
* We use a bitmask (&accept) to represent accepted prefix lengths
* at a node. As there are 33 prefix lengths (0..32 for IPv4), but
* there is just one prefix of zero length in the whole trie so we
* have &zero flag in &f_trie (indicating whether the trie accepts
* prefix 0.0.0.0/0) as a special case, and &accept bitmask
* represents accepted prefix lengths from 1 to 32.
*
* There are two cases in prefix matching - a match when the length
* of the prefix is smaller that the length of the prefix pattern,
* (&plen < &pplen) and otherwise. The second case is simple - we
* just walk through the trie and look at every visited node
* whether that prefix accepts our prefix length (&plen). The
* first case is tricky - we don't want to examine every descendant
* of a final node, so (when we create the trie) we have to propagate
* that information from nodes to their ascendants.
*
* Suppose that we have two masks (M1 and M2) for a node. Mask M1
* represents accepted prefix lengths by just the node and mask M2
* represents accepted prefix lengths by the node or any of its
* descendants. Therefore M2 is a bitwise or of M1 and children's
* M2 and this is a maintained invariant during trie building.
* Basically, when we want to match a prefix, we walk through the trie,
* check mask M1 for our prefix length and when we came to
* final node, we check mask M2.
*
* There are two differences in the real implementation. First,
* we use a compressed trie so there is a case that we skip our
* final node (if it is not in the trie) and we came to node that
* is either extension of our prefix, or completely out of path
* In the first case, we also have to check M2.
*
* Second, we really need not to maintain two separate bitmasks.
* Checks for mask M1 are always larger than &applen and we need
* just the first &pplen bits of mask M2 (if trie compression
* hadn't been used it would suffice to know just $applen-th bit),
* so we have to store them together in &accept mask - the first
* &pplen bits of mask M2 and then mask M1.
*
* There are four cases when we walk through a trie:
*
* - we are in NULL
* - we are out of path (prefixes are inconsistent)
* - we are in the wanted (final) node (node length == &plen)
* - we are beyond the end of path (node length > &plen)
* - we are still on path and keep walking (node length < &plen)
*
* The walking code in trie_match_prefix() is structured according to
* these cases.
*/
#include "nest/bird.h"
#include "lib/string.h"
#include "conf/conf.h"
#include "filter/filter.h"
#include "filter/data.h"
/*
* In the trie_add_prefix(), we use ip_addr (assuming that it is the same as
* ip6_addr) to handle both IPv4 and IPv6 prefixes. In contrast to rest of the
* BIRD, IPv4 addresses are just zero-padded from right. That is why we have
* ipt_from_ip4() and ipt_to_ip4() macros below.
*/
#define ipa_mkmask(x) ip6_mkmask(x)
#define ipa_masklen(x) ip6_masklen(&x)
#define ipa_pxlen(x,y) ip6_pxlen(x,y)
#define ipa_getbit(a,p) ip6_getbit(a,p)
#define ipa_getbits(a,p,n) ip6_getbits(a,p,n)
#define ipa_setbits(a,p,n) ip6_setbits(a,p,n)
#define trie_local_mask(a,b,c) trie_local_mask6(a,b,c)
#define ipt_from_ip4(x) _MI6(_I(x), 0, 0, 0)
#define ipt_to_ip4(x) _MI4(_I0(x))
/**
* f_new_trie - allocates and returns a new empty trie
* @lp: linear pool to allocate items from
* @data_size: user data attached to node
*/
struct f_trie *
f_new_trie(linpool *lp, uint data_size)
{
struct f_trie * ret;
ret = lp_allocz(lp, sizeof(struct f_trie) + data_size);
ret->lp = lp;
ret->ipv4 = -1;
ret->data_size = data_size;
return ret;
}
static inline struct f_trie_node4 *
new_node4(struct f_trie *t, uint plen, uint local, ip4_addr paddr, ip4_addr pmask, ip4_addr amask)
{
struct f_trie_node4 *n = lp_allocz(t->lp, sizeof(struct f_trie_node4) + t->data_size);
n->plen = plen;
n->local = local;
n->addr = paddr;
n->mask = pmask;
n->accept = amask;
return n;
}
static inline struct f_trie_node6 *
new_node6(struct f_trie *t, uint plen, uint local, ip6_addr paddr, ip6_addr pmask, ip6_addr amask)
{
struct f_trie_node6 *n = lp_allocz(t->lp, sizeof(struct f_trie_node6) + t->data_size);
n->plen = plen;
n->local = local;
n->addr = paddr;
n->mask = pmask;
n->accept = amask;
return n;
}
static inline struct f_trie_node *
new_node(struct f_trie *t, uint plen, uint local, ip_addr paddr, ip_addr pmask, ip_addr amask)
{
if (t->ipv4)
return (struct f_trie_node *) new_node4(t, plen, local, ipt_to_ip4(paddr), ipt_to_ip4(pmask), ipt_to_ip4(amask));
else
return (struct f_trie_node *) new_node6(t, plen, local, ipa_to_ip6(paddr), ipa_to_ip6(pmask), ipa_to_ip6(amask));
}
static inline void
attach_node4(struct f_trie_node4 *parent, struct f_trie_node4 *child)
{
parent->c[ip4_getbits(child->addr, parent->plen, TRIE_STEP)] = child;
}
static inline void
attach_node6(struct f_trie_node6 *parent, struct f_trie_node6 *child)
{
parent->c[ip6_getbits(child->addr, parent->plen, TRIE_STEP)] = child;
}
static inline void
attach_node(struct f_trie_node *parent, struct f_trie_node *child, int v4)
{
if (v4)
attach_node4(&parent->v4, &child->v4);
else
attach_node6(&parent->v6, &child->v6);
}
static inline uint
trie_local_mask4(ip4_addr px, uint plen, uint nlen)
{
uint step = plen - nlen;
uint pos = (1u << step) + ip4_getbits(px, nlen, step);
return 1u << pos;
}
static inline uint
trie_local_mask6(ip6_addr px, uint plen, uint nlen)
{
uint step = plen - nlen;
uint pos = (1u << step) + ip6_getbits(px, nlen, step);
return 1u << pos;
}
static inline uint
trie_amask_to_local(ip_addr px, ip_addr amask, uint nlen)
{
uint local = 0;
for (uint plen = MAX(nlen, 1); plen < (nlen + TRIE_STEP); plen++)
if (ipa_getbit(amask, plen - 1))
local |= trie_local_mask(px, plen, nlen);
return local;
}
#define GET_ADDR(N,F,X) ((X) ? ipt_from_ip4((N)->v4.F) : ipa_from_ip6((N)->v6.F))
#define SET_ADDR(N,F,X,V) ({ if (X) (N)->v4.F =ipt_to_ip4(V); else (N)->v6.F =ipa_to_ip6(V); })
#define ADD_LOCAL(N,X,V) ({ uint v_ = (V); if (X) (N)->v4.local |= v_; else (N)->v6.local |= v_; })
#define GET_CHILD(N,F,X,I) ((X) ? (struct f_trie_node *) (N)->v4.c[I] : (struct f_trie_node *) (N)->v6.c[I])
static void *
trie_add_node(struct f_trie *t, uint plen, ip_addr px, uint local, uint l, uint h)
{
uint l_ = l ? (l - 1) : 0;
ip_addr amask = (l_ < h) ? ipa_xor(ipa_mkmask(l_), ipa_mkmask(h)) : IPA_NONE;
ip_addr pmask = ipa_mkmask(plen);
ip_addr paddr = ipa_and(px, pmask);
struct f_trie_node *o = NULL;
struct f_trie_node *n = &t->root;
int v4 = t->ipv4;
/* Add all bits for each active level (0x0002 0x000c 0x00f0 0xff00) */
for (uint i = 0; i < TRIE_STEP; i++)
if ((l <= (plen + i)) && ((plen + i) <= h))
local |= ((1u << (1u << i)) - 1) << (1u << i);
DBG("Insert node %I/%u (%I %x)\n", paddr, plen, amask, local);
while (n)
{
ip_addr naddr = GET_ADDR(n, addr, v4);
ip_addr nmask = GET_ADDR(n, mask, v4);
ip_addr accept = GET_ADDR(n, accept, v4);
ip_addr cmask = ipa_and(nmask, pmask);
uint nlen = v4 ? n->v4.plen : n->v6.plen;
DBG("Found node %I/%u (%I %x)\n",
naddr, nlen, accept, v4 ? n->v4.local : n->v6.local);
if (ipa_compare(ipa_and(paddr, cmask), ipa_and(naddr, cmask)))
{
/* We are out of path - we have to add branching node 'b'
between node 'o' and node 'n', and attach new node 'a'
as the other child of 'b'. */
int blen = ROUND_DOWN_POW2(ipa_pxlen(paddr, naddr), TRIE_STEP);
ip_addr bmask = ipa_mkmask(blen);
ip_addr baddr = ipa_and(px, bmask);
/* Merge accept masks from children to get accept mask for node 'b' */
ip_addr baccm = ipa_and(ipa_or(amask, accept), bmask);
uint bloc = trie_amask_to_local(naddr, accept, blen) |
trie_amask_to_local(paddr, amask, blen);
struct f_trie_node *a = new_node(t, plen, local, paddr, pmask, amask);
struct f_trie_node *b = new_node(t, blen, bloc, baddr, bmask, baccm);
attach_node(o, b, v4);
attach_node(b, n, v4);
attach_node(b, a, v4);
DBG("Case 1\n");
return a;
}
if (plen < nlen)
{
/* We add new node 'a' between node 'o' and node 'n' */
amask = ipa_or(amask, ipa_and(accept, pmask));
local |= trie_amask_to_local(naddr, accept, plen);
struct f_trie_node *a = new_node(t, plen, local, paddr, pmask, amask);
attach_node(o, a, v4);
attach_node(a, n, v4);
DBG("Case 2\n");
return a;
}
if (plen == nlen)
{
/* We already found added node in trie. Just update accept and local mask */
accept = ipa_or(accept, amask);
SET_ADDR(n, accept, v4, accept);
ADD_LOCAL(n, v4, local);
DBG("Case 3\n");
return n;
}
/* Update accept mask part M2 and go deeper */
accept = ipa_or(accept, ipa_and(amask, nmask));
SET_ADDR(n, accept, v4, accept);
ADD_LOCAL(n, v4, trie_amask_to_local(paddr, amask, nlen));
DBG("Step %u\n", ipa_getbits(paddr, nlen));
/* n->plen < plen and plen <= 32 (128) */
o = n;
n = GET_CHILD(n, c, v4, ipa_getbits(paddr, nlen, TRIE_STEP));
}
/* We add new tail node 'a' after node 'o' */
struct f_trie_node *a = new_node(t, plen, local, paddr, pmask, amask);
attach_node(o, a, v4);
DBG("Case 4\n");
return a;
}
/**
* trie_add_prefix
* @t: trie to add to
* @net: IP network prefix
* @l: prefix lower bound
* @h: prefix upper bound
*
* Adds prefix (prefix pattern) @n to trie @t. @l and @h are lower
* and upper bounds on accepted prefix lengths, both inclusive.
* 0 <= l, h <= 32 (128 for IPv6).
*
* Returns a pointer to the allocated node. The function can return a pointer to
* an existing node if @px and @plen are the same. If px/plen == 0/0 (or ::/0),
* a pointer to the root node is returned. Returns NULL when called with
* mismatched IPv4/IPv6 net type.
*/
void *
trie_add_prefix(struct f_trie *t, const net_addr *net, uint l, uint h)
{
uint plen = net_pxlen(net);
ip_addr px;
int v4;
switch (net->type)
{
case NET_IP4: px = ipt_from_ip4(net4_prefix(net)); v4 = 1; break;
case NET_IP6: px = ipa_from_ip6(net6_prefix(net)); v4 = 0; break;
default: bug("invalid type");
}
if (t->ipv4 != v4)
{
if (t->ipv4 < 0)
t->ipv4 = v4;
else
return NULL;
}
DBG("\nInsert net %N (%u-%u)\n", net, l, h);
if (l == 0)
t->zero = 1;
if (h < plen)
plen = h;
/* Primary node length, plen rounded down */
uint nlen = ROUND_DOWN_POW2(plen, TRIE_STEP);
if (plen == nlen)
return trie_add_node(t, nlen, px, 0, l, h);
/* Secondary node length, plen rouned up */
uint slen = nlen + TRIE_STEP;
void *node = NULL;
/*
* For unaligned prefix lengths it is more complicated. We need to encode
* matching prefixes of lengths from l to h. There are three cases of lengths:
*
* 1) 0..nlen are encoded by the accept mask of the primary node
* 2) nlen..(slen-1) are encoded by the local mask of the primary node
* 3) slen..max are encoded in secondary nodes
*/
if (l < slen)
{
uint local = 0;
/* Compute local bits for accepted nlen..(slen-1) prefixes */
for (uint i = 0; i < TRIE_STEP; i++)
if ((l <= (nlen + i)) && ((nlen + i) <= h))
{
uint pos = (1u << i) + ipa_getbits(px, nlen, i);
uint len = ((nlen + i) <= plen) ? 1 : (1u << (nlen + i - plen));
/* We need to fill 'len' bits starting at 'pos' position */
local |= ((1u << len) - 1) << pos;
}
/* Add the primary node */
node = trie_add_node(t, nlen, px, local, l, nlen);
}
if (slen <= h)
{
uint l2 = MAX(l, slen);
uint max = (1u << (slen - plen));
/* Add secondary nodes */
for (uint i = 0; i < max; i++)
node = trie_add_node(t, slen, ipa_setbits(px, slen - 1, i), 0, l2, h);
}
return node;
}
static int
trie_match_net4(const struct f_trie *t, ip4_addr px, uint plen)
{
ip4_addr pmask = ip4_mkmask(plen);
ip4_addr paddr = ip4_and(px, pmask);
if (plen == 0)
return t->zero;
int plentest = plen - 1;
uint nlen = ROUND_DOWN_POW2(plen, TRIE_STEP);
uint local = trie_local_mask4(px, plen, nlen);
const struct f_trie_node4 *n = &t->root.v4;
while (n)
{
ip4_addr cmask = ip4_and(n->mask, pmask);
/* We are out of path */
if (ip4_compare(ip4_and(paddr, cmask), ip4_and(n->addr, cmask)))
return 0;
/* Check local mask */
if ((n->plen == nlen) && (n->local & local))
return 1;
/* Check accept mask */
if (ip4_getbit(n->accept, plentest))
return 1;
/* We finished trie walk and still no match */
if (plen <= n->plen)
return 0;
/* Choose children */
n = n->c[ip4_getbits(paddr, n->plen, TRIE_STEP)];
}
return 0;
}
static int
trie_match_net6(const struct f_trie *t, ip6_addr px, uint plen)
{
ip6_addr pmask = ip6_mkmask(plen);
ip6_addr paddr = ip6_and(px, pmask);
if (plen == 0)
return t->zero;
int plentest = plen - 1;
uint nlen = ROUND_DOWN_POW2(plen, TRIE_STEP);
uint local = trie_local_mask6(px, plen, nlen);
const struct f_trie_node6 *n = &t->root.v6;
while (n)
{
ip6_addr cmask = ip6_and(n->mask, pmask);
/* We are out of path */
if (ip6_compare(ip6_and(paddr, cmask), ip6_and(n->addr, cmask)))
return 0;
/* Check local mask */
if ((n->plen == nlen) && (n->local & local))
return 1;
/* Check accept mask */
if (ip6_getbit(n->accept, plentest))
return 1;
/* We finished trie walk and still no match */
if (plen <= n->plen)
return 0;
/* Choose children */
n = n->c[ip6_getbits(paddr, n->plen, TRIE_STEP)];
}
return 0;
}
/**
* trie_match_net
* @t: trie
* @n: net address
*
* Tries to find a matching net in the trie such that
* prefix @n matches that prefix pattern. Returns 1 if there
* is such prefix pattern in the trie.
*/
int
trie_match_net(const struct f_trie *t, const net_addr *n)
{
switch (n->type)
{
case NET_IP4:
case NET_VPN4:
case NET_ROA4:
return t->ipv4 ? trie_match_net4(t, net4_prefix(n), net_pxlen(n)) : 0;
case NET_IP6:
case NET_VPN6:
case NET_ROA6:
return !t->ipv4 ? trie_match_net6(t, net6_prefix(n), net_pxlen(n)) : 0;
default:
return 0;
}
}
static int
trie_node_same4(const struct f_trie_node4 *t1, const struct f_trie_node4 *t2)
{
if ((t1 == NULL) && (t2 == NULL))
return 1;
if ((t1 == NULL) || (t2 == NULL))
return 0;
if ((t1->plen != t2->plen) ||
(! ip4_equal(t1->addr, t2->addr)) ||
(! ip4_equal(t1->accept, t2->accept)))
return 0;
for (uint i = 0; i < (1 << TRIE_STEP); i++)
if (! trie_node_same4(t1->c[i], t2->c[i]))
return 0;
return 1;
}
static int
trie_node_same6(const struct f_trie_node6 *t1, const struct f_trie_node6 *t2)
{
if ((t1 == NULL) && (t2 == NULL))
return 1;
if ((t1 == NULL) || (t2 == NULL))
return 0;
if ((t1->plen != t2->plen) ||
(! ip6_equal(t1->addr, t2->addr)) ||
(! ip6_equal(t1->accept, t2->accept)))
return 0;
for (uint i = 0; i < (1 << TRIE_STEP); i++)
if (! trie_node_same6(t1->c[i], t2->c[i]))
return 0;
return 1;
}
/**
* trie_same
* @t1: first trie to be compared
* @t2: second one
*
* Compares two tries and returns 1 if they are same
*/
int
trie_same(const struct f_trie *t1, const struct f_trie *t2)
{
if ((t1->zero != t2->zero) || (t1->ipv4 != t2->ipv4))
return 0;
if (t1->ipv4)
return trie_node_same4(&t1->root.v4, &t2->root.v4);
else
return trie_node_same6(&t1->root.v6, &t2->root.v6);
}
static void
trie_node_format4(const struct f_trie_node4 *t, buffer *buf)
{
if (t == NULL)
return;
if (ip4_nonzero(t->accept))
buffer_print(buf, "%I4/%d{%I4}, ", t->addr, t->plen, t->accept);
for (uint i = 0; i < (1 << TRIE_STEP); i++)
trie_node_format4(t->c[i], buf);
}
static void
trie_node_format6(const struct f_trie_node6 *t, buffer *buf)
{
if (t == NULL)
return;
if (ip6_nonzero(t->accept))
buffer_print(buf, "%I6/%d{%I6}, ", t->addr, t->plen, t->accept);
for (uint i = 0; i < (1 << TRIE_STEP); i++)
trie_node_format6(t->c[i], buf);
}
/**
* trie_format
* @t: trie to be formatted
* @buf: destination buffer
*
* Prints the trie to the supplied buffer.
*/
void
trie_format(const struct f_trie *t, buffer *buf)
{
buffer_puts(buf, "[");
if (t->zero)
buffer_print(buf, "%I/%d, ", t->ipv4 ? IPA_NONE4 : IPA_NONE6, 0);
if (t->ipv4)
trie_node_format4(&t->root.v4, buf);
else
trie_node_format6(&t->root.v6, buf);
if (buf->pos == buf->end)
return;
/* Undo last separator */
if (buf->pos[-1] != '[')
buf->pos -= 2;
buffer_puts(buf, "]");
}