BMP: simplified update queuing and better memory performance

This commit is quite a substantial rework of the underlying layers in
BMP TX:

- several unnecessary layers of indirection dropped, including most of
  the original BMP's buffer machinery
- all messages are now written directly into one protocol's buffer
  allocated for the whole time big enough to fit every possible message
- output blocks are allocated by pages and immediately returned when
  used, improving the overall memory footprint
- no intermediary allocation is done from the heap altogether
- there is a documented and configurable limit on the TX queue size

doc/bird.sgml

@@ -3808,6 +3808,15 @@ by default and have to be enabled during installation by the configure option
 routes (in <ref id="bgp-import-table" name="BGP import tables">) and post-policy
 routes (in regular routing tables). All BGP protocols are monitored automatically.
 
+<sect1>Configuration (incomplete)
+<label id="bmp-config">
+
+<p><descrip>
+	<tag><label id="bmp-tx-buffer-limit">tx buffer limit <m/number/</tag>
+	How much data we are going to queue before we call the session stuck
+	and restart it, in megabytes. Default value: 1024 (effectively 1 gigabyte).
+</descrip>
+
 <sect1>Example
 <label id="bmp-exam">
 
@@ -3821,6 +3830,9 @@ protocol bmp {
 
 	# Monitor accepted routes (passed import filters)
 	monitoring rib in post_policy;
+
+	# Allow only 64M of pending data
+	tx buffer limit 64;
 }
 </code>
 

proto/bgp/bgp.h

@@ -650,7 +650,7 @@ int bgp_get_attr(const struct eattr *e, byte *buf, int buflen);
 void bgp_get_route_info(struct rte *, byte *buf);
 int bgp_total_aigp_metric_(rte *e, u64 *metric, const struct adata **ad);
 
-byte * bgp_bmp_encode_rte(struct bgp_channel *c, byte *buf, const net_addr *n, const struct rte *new, const struct rte_src *src);
+byte * bgp_bmp_encode_rte(struct bgp_channel *c, byte *buf, byte *end, const net_addr *n, const struct rte *new, const struct rte_src *src);
 
 #define BGP_AIGP_METRIC		1
 #define BGP_AIGP_MAX		U64(0xffffffffffffffff)

proto/bgp/packets.c

@@ -2482,12 +2482,10 @@ bgp_create_mp_unreach(struct bgp_write_state *s, struct bgp_bucket *buck, byte *
 #ifdef CONFIG_BMP
 
 static byte *
-bgp_create_update_bmp(struct bgp_channel *c, byte *buf, struct bgp_bucket *buck, bool update)
+bgp_create_update_bmp(struct bgp_channel *c, byte *buf, byte *end, struct bgp_bucket *buck, bool update)
 {
   struct bgp_proto *p = (void *) c->c.proto;
-  byte *end = buf + (BGP_MAX_EXT_MSG_LENGTH - BGP_HEADER_LENGTH);
   byte *res = NULL;
-  /* FIXME: must be a bit shorter */
 
   struct lp_state tmpp;
   lp_save(tmp_linpool, &tmpp);
@@ -2535,7 +2533,7 @@ bgp_bmp_prepare_bgp_hdr(byte *buf, const u16 msg_size, const u8 msg_type)
 }
 
 byte *
-bgp_bmp_encode_rte(struct bgp_channel *c, byte *buf, const net_addr *n,
+bgp_bmp_encode_rte(struct bgp_channel *c, byte *buf, byte *end, const net_addr *n,
 		   const struct rte *new, const struct rte_src *src)
 {
 //  struct bgp_proto *p = (void *) c->c.proto;
@@ -2561,7 +2559,7 @@ bgp_bmp_encode_rte(struct bgp_channel *c, byte *buf, const net_addr *n,
   net_copy(px->net, n);
   add_tail(&b->prefixes, &px->buck_node);
 
-  byte *end = bgp_create_update_bmp(c, pkt, b, !!new);
+  end = bgp_create_update_bmp(c, pkt, end, b, !!new);
 
   if (end)
     bgp_bmp_prepare_bgp_hdr(buf, end - buf, PKT_UPDATE);

proto/bmp/Makefile

@@ -1,6 +1,6 @@
-src := bmp.c buffer.c map.c
+src := bmp.c map.c
 obj := $(src-o-files)
 $(all-daemon)
 $(cf-local)
 
-tests_objs := $(tests_objs) $(src-o-files)
+tests_objs := $(tests_objs) $(src-o-files)