]> git.proxmox.com Git - mirror_frr.git/commitdiff
bgpd: move bgp i/o to a separate source file
authorQuentin Young <qlyoung@cumulusnetworks.com>
Tue, 18 Apr 2017 18:11:43 +0000 (18:11 +0000)
committerQuentin Young <qlyoung@cumulusnetworks.com>
Thu, 30 Nov 2017 21:17:59 +0000 (16:17 -0500)
After implement threading, bgp_packet.c was serving the double purpose
of consolidating packet parsing functionality and handling actual I/O
operations. This is somewhat messy and difficult to understand. I've
thus moved all code and data structures for handling threaded packet
writes to bgp_io.[ch].

Although bgp_io.[ch] only handles writes at the moment to keep the noise
on this commit series down, for organization purposes, it's probably
best to move bgp_read() and its trappings into here as well and
restructure that code so that read()'s happen in the pthread and packet
processing happens on the main thread.

Signed-off-by: Quentin Young <qlyoung@cumulusnetworks.com>
bgpd/Makefile.am
bgpd/bgp_fsm.c
bgpd/bgp_io.c [new file with mode: 0644]
bgpd/bgp_io.h [new file with mode: 0644]
bgpd/bgp_packet.c
bgpd/bgp_packet.h
bgpd/bgpd.c
bgpd/bgpd.h

index c8791687ea66948a60d52f17b38664a1b0ba803d..b0d34dc43b8115ba16d888ef2063c4a32b849887 100644 (file)
@@ -86,7 +86,7 @@ libbgp_a_SOURCES = \
        bgp_nht.c bgp_updgrp.c bgp_updgrp_packet.c bgp_updgrp_adv.c bgp_bfd.c \
        bgp_encap_tlv.c $(BGP_VNC_RFAPI_SRC) bgp_attr_evpn.c \
        bgp_evpn.c bgp_evpn_vty.c bgp_vpn.c bgp_label.c bgp_rd.c \
-       bgp_keepalives.c
+       bgp_keepalives.c bgp_io.c
 
 noinst_HEADERS = \
        bgp_memory.h \
@@ -99,6 +99,7 @@ noinst_HEADERS = \
        bgp_updgrp.h bgp_bfd.h bgp_encap_tlv.h bgp_encap_types.h \
        $(BGP_VNC_RFAPI_HD) bgp_attr_evpn.h bgp_evpn.h bgp_evpn_vty.h \
         bgp_vpn.h bgp_label.h bgp_rd.h bgp_evpn_private.h bgp_keepalives.h \
+       bgp_io.h
 
 bgpd_SOURCES = bgp_main.c
 bgpd_LDADD = libbgp.a  $(BGP_VNC_RFP_LIB) ../lib/libfrr.la @LIBCAP@ @LIBM@
index d965b49c4900569ad9cc7c23308fafb9bbef89aa..95e2f157cce778052ccc32e4c2776cb65c3af4bb 100644 (file)
@@ -50,6 +50,7 @@
 #include "bgpd/bgp_bfd.h"
 #include "bgpd/bgp_memory.h"
 #include "bgpd/bgp_keepalives.h"
+#include "bgpd/bgp_io.h"
 
 DEFINE_HOOK(peer_backward_transition, (struct peer * peer), (peer))
 DEFINE_HOOK(peer_established, (struct peer * peer), (peer))
@@ -1037,6 +1038,7 @@ int bgp_stop(struct peer *peer)
        BGP_TIMER_OFF(peer->t_holdtime);
        peer_keepalives_off(peer);
        BGP_TIMER_OFF(peer->t_routeadv);
+       BGP_TIMER_OFF(peer->t_generate_updgrp_packets);
 
        /* Stream reset. */
        peer->packet_size = 0;
diff --git a/bgpd/bgp_io.c b/bgpd/bgp_io.c
new file mode 100644 (file)
index 0000000..5d14737
--- /dev/null
@@ -0,0 +1,307 @@
+/*
+  BGP I/O.
+  Implements a consumer thread to flush packets destined for remote peers.
+
+  Copyright (C) 2017  Cumulus Networks
+
+  This program is free software; you can redistribute it and/or modify
+  it under the terms of the GNU General Public License as published by
+  the Free Software Foundation; either version 2 of the License, or
+  (at your option) any later version.
+
+  This program is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU General Public License
+  along with this program; see the file COPYING; if not, write to the
+  Free Software Foundation, Inc., 51 Franklin St, Fifth Floor, Boston,
+  MA 02110-1301 USA
+ */
+
+#include <zebra.h>
+#include <sys/time.h>
+#include <pthread.h>
+
+#include "thread.h"
+#include "hash.h"
+#include "stream.h"
+#include "memory.h"
+#include "log.h"
+#include "monotime.h"
+#include "network.h"
+#include "frr_pthread.h"
+
+#include "bgpd/bgpd.h"
+#include "bgpd/bgp_io.h"
+#include "bgpd/bgp_debug.h"
+#include "bgpd/bgp_packet.h"
+#include "bgpd/bgp_fsm.h"
+
+static int bgp_write(struct peer *);
+static void peer_process_writes(struct hash_backet *, void *);
+
+bool bgp_packet_writes_thread_run = false;
+
+/* Hash table of peers to operate on, associated synchronization primitives and
+ * hash table callbacks.
+ * ------------------------------------------------------------------------ */
+static struct hash *peerhash;
+/* Mutex to protect hash table */
+static pthread_mutex_t *peerhash_mtx;
+/* Condition variable used to notify the write thread that there is work to do
+ */
+static pthread_cond_t *write_cond;
+
+static unsigned int peer_hash_key_make(void *p)
+{
+       struct peer *peer = p;
+       return sockunion_hash(&peer->su);
+}
+
+static int peer_hash_cmp(const void *p1, const void *p2)
+{
+       const struct peer *peer1 = p1;
+       const struct peer *peer2 = p2;
+       return (sockunion_same(&peer1->su, &peer2->su)
+               && CHECK_FLAG(peer1->flags, PEER_FLAG_CONFIG_NODE)
+                          == CHECK_FLAG(peer2->flags, PEER_FLAG_CONFIG_NODE));
+}
+/* ------------------------------------------------------------------------ */
+
+void peer_writes_init(void)
+{
+       peerhash_mtx = XCALLOC(MTYPE_PTHREAD, sizeof(pthread_mutex_t));
+       write_cond = XCALLOC(MTYPE_PTHREAD, sizeof(pthread_cond_t));
+
+       // initialize mutex
+       pthread_mutex_init(peerhash_mtx, NULL);
+
+       // use monotonic clock with condition variable
+       pthread_condattr_t attrs;
+       pthread_condattr_init(&attrs);
+       pthread_condattr_setclock(&attrs, CLOCK_MONOTONIC);
+       pthread_cond_init(write_cond, &attrs);
+       pthread_condattr_destroy(&attrs);
+
+       // initialize peerhash
+       peerhash = hash_create_size(2048, peer_hash_key_make, peer_hash_cmp);
+}
+
+static void peer_writes_finish(void *arg)
+{
+       bgp_packet_writes_thread_run = false;
+
+       if (peerhash)
+               hash_free(peerhash);
+
+       peerhash = NULL;
+
+       pthread_mutex_unlock(peerhash_mtx);
+       pthread_mutex_destroy(peerhash_mtx);
+       pthread_cond_destroy(write_cond);
+
+       XFREE(MTYPE_PTHREAD, peerhash_mtx);
+       XFREE(MTYPE_PTHREAD, write_cond);
+}
+
+void *peer_writes_start(void *arg)
+{
+       struct timeval currtime = {0, 0};
+       struct timeval sleeptime = {0, 500};
+       struct timespec next_update = {0, 0};
+
+       pthread_mutex_lock(peerhash_mtx);
+
+       // register cleanup handler
+       pthread_cleanup_push(&peer_writes_finish, NULL);
+
+       bgp_packet_writes_thread_run = true;
+
+       while (bgp_packet_writes_thread_run) {
+               // wait around until next update time
+               if (peerhash->count > 0)
+                       pthread_cond_timedwait(write_cond, peerhash_mtx,
+                                              &next_update);
+               else // wait around until we have some peers
+                       while (peerhash->count == 0
+                              && bgp_packet_writes_thread_run)
+                               pthread_cond_wait(write_cond, peerhash_mtx);
+
+               hash_iterate(peerhash, peer_process_writes, NULL);
+
+               monotime(&currtime);
+               timeradd(&currtime, &sleeptime, &currtime);
+               TIMEVAL_TO_TIMESPEC(&currtime, &next_update);
+       }
+
+       // clean up
+       pthread_cleanup_pop(1);
+
+       return NULL;
+}
+
+int peer_writes_stop(void **result)
+{
+       struct frr_pthread *fpt = frr_pthread_get(PTHREAD_WRITE);
+       bgp_packet_writes_thread_run = false;
+       peer_writes_wake();
+       pthread_join(fpt->thread, result);
+       return 0;
+}
+
+void peer_writes_on(struct peer *peer)
+{
+       if (peer->status == Deleted)
+               return;
+
+       pthread_mutex_lock(peerhash_mtx);
+       {
+               if (!hash_lookup(peerhash, peer)) {
+                       hash_get(peerhash, peer, hash_alloc_intern);
+                       peer_lock(peer);
+               }
+
+               SET_FLAG(peer->thread_flags, PEER_THREAD_WRITES_ON);
+       }
+       pthread_mutex_unlock(peerhash_mtx);
+       peer_writes_wake();
+}
+
+void peer_writes_off(struct peer *peer)
+{
+       pthread_mutex_lock(peerhash_mtx);
+       {
+               if (hash_release(peerhash, peer)) {
+                       peer_unlock(peer);
+                       fprintf(stderr, "Releasing %p\n", peer);
+               }
+
+               UNSET_FLAG(peer->thread_flags, PEER_THREAD_WRITES_ON);
+       }
+       pthread_mutex_unlock(peerhash_mtx);
+}
+
+void peer_writes_wake()
+{
+       pthread_cond_signal(write_cond);
+}
+
+/**
+ * Callback for hash_iterate. Takes a hash bucket, unwraps it into a peer and
+ * synchronously calls bgp_write() on the peer.
+ */
+static void peer_process_writes(struct hash_backet *hb, void *arg)
+{
+       static struct peer *peer;
+       peer = hb->data;
+       pthread_mutex_lock(&peer->obuf_mtx);
+       {
+               bgp_write(peer);
+       }
+       pthread_mutex_unlock(&peer->obuf_mtx);
+
+       // dispatch job on main thread
+       BGP_TIMER_ON(peer->t_generate_updgrp_packets,
+                    bgp_generate_updgrp_packets, 100);
+}
+
+/**
+ * Flush peer output buffer.
+ *
+ * This function pops packets off of peer->obuf and writes them to peer->fd.
+ * The amount of packets written is equal to the minimum of peer->wpkt_quanta
+ * and the number of packets on the output buffer.
+ *
+ * If write() returns an error, the appropriate FSM event is generated.
+ *
+ * The return value is equal to the number of packets written
+ * (which may be zero).
+ */
+static int bgp_write(struct peer *peer)
+{
+       u_char type;
+       struct stream *s;
+       int num;
+       int update_last_write = 0;
+       unsigned int count = 0;
+       unsigned int oc = 0;
+
+       /* Write packets. The number of packets written is the value of
+        * bgp->wpkt_quanta or the size of the output buffer, whichever is
+        * smaller.*/
+       while (count < peer->bgp->wpkt_quanta
+              && (s = stream_fifo_head(peer->obuf))) {
+               int writenum;
+               do {
+                       writenum = stream_get_endp(s) - stream_get_getp(s);
+                       num = write(peer->fd, STREAM_PNT(s), writenum);
+
+                       if (num < 0) {
+                               if (!ERRNO_IO_RETRY(errno))
+                                       BGP_EVENT_ADD(peer, TCP_fatal_error);
+
+                               goto done;
+                       } else if (num != writenum) // incomplete write
+                               stream_forward_getp(s, num);
+
+               } while (num != writenum);
+
+               /* Retrieve BGP packet type. */
+               stream_set_getp(s, BGP_MARKER_SIZE + 2);
+               type = stream_getc(s);
+
+               switch (type) {
+               case BGP_MSG_OPEN:
+                       peer->open_out++;
+                       break;
+               case BGP_MSG_UPDATE:
+                       peer->update_out++;
+                       break;
+               case BGP_MSG_NOTIFY:
+                       peer->notify_out++;
+                       /* Double start timer. */
+                       peer->v_start *= 2;
+
+                       /* Overflow check. */
+                       if (peer->v_start >= (60 * 2))
+                               peer->v_start = (60 * 2);
+
+                       /* Handle Graceful Restart case where the state changes
+                          to
+                          Connect instead of Idle */
+                       /* Flush any existing events */
+                       BGP_EVENT_ADD(peer, BGP_Stop);
+                       goto done;
+
+               case BGP_MSG_KEEPALIVE:
+                       peer->keepalive_out++;
+                       break;
+               case BGP_MSG_ROUTE_REFRESH_NEW:
+               case BGP_MSG_ROUTE_REFRESH_OLD:
+                       peer->refresh_out++;
+                       break;
+               case BGP_MSG_CAPABILITY:
+                       peer->dynamic_cap_out++;
+                       break;
+               }
+
+               count++;
+               /* OK we send packet so delete it. */
+               stream_free(stream_fifo_pop(peer->obuf));
+               update_last_write = 1;
+       }
+
+done : {
+       /* Update last_update if UPDATEs were written. */
+       if (peer->update_out > oc)
+               peer->last_update = bgp_clock();
+
+       /* If we TXed any flavor of packet update last_write */
+       if (update_last_write)
+               peer->last_write = bgp_clock();
+}
+
+       return count;
+}
diff --git a/bgpd/bgp_io.h b/bgpd/bgp_io.h
new file mode 100644 (file)
index 0000000..7b81b8e
--- /dev/null
@@ -0,0 +1,104 @@
+/*
+  BGP I/O.
+  Implements a consumer thread to flush packets destined for remote peers.
+
+  Copyright (C) 2017  Cumulus Networks
+
+  This program is free software; you can redistribute it and/or modify
+  it under the terms of the GNU General Public License as published by
+  the Free Software Foundation; either version 2 of the License, or
+  (at your option) any later version.
+
+  This program is distributed in the hope that it will be useful, but
+  WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+  General Public License for more details.
+
+  You should have received a copy of the GNU General Public License
+  along with this program; see the file COPYING; if not, write to the
+  Free Software Foundation, Inc., 51 Franklin St, Fifth Floor, Boston,
+  MA 02110-1301 USA
+ */
+
+#ifndef _FRR_BGP_IO_H
+#define _FRR_BGP_IO_H
+
+#include "bgpd/bgpd.h"
+
+/**
+ * Control variable for write thread.
+ *
+ * Setting this variable to false and calling peer_writes_wake() will
+ * eventually result in thread termination.
+ */
+extern bool bgp_packet_writes_thread_run;
+
+/**
+ * Initializes data structures and flags for the write thread.
+ *
+ * This function should be called from the main thread before
+ * peer_writes_start() is invoked.
+ */
+extern void peer_writes_init(void);
+
+/**
+ * Start function for write thread.
+ *
+ * This function should be passed to pthread_create() during BGP startup.
+ */
+extern void *peer_writes_start(void *arg);
+
+/**
+ * Start function for write thread.
+ *
+ * Uninitializes all resources and stops the thread.
+ *
+ * @param result -- where to store data result, unused
+ */
+extern int peer_writes_stop(void **result);
+
+/**
+ * Registers a peer with the write thread.
+ *
+ * This function adds the peer to an internal data structure, which must be
+ * locked for write access. This call will block until the structure can be
+ * locked.
+ *
+ * After this function is called, any packets placed on peer->obuf will be
+ * written to peer->fd at regular intervals.
+ *
+ * This function increments the peer reference counter with peer_lock().
+ *
+ * If the peer is already registered, nothing happens.
+ *
+ * @param peer - peer to register
+ */
+extern void peer_writes_on(struct peer *peer);
+
+/**
+ * Deregisters a peer with the write thread.
+ *
+ * This function removes the peer from an internal data structure, which must
+ * be locked for write access. This call will block until the structure can be
+ * locked.
+ *
+ * After this function is called, any packets placed on peer->obuf will not be
+ * written to peer->fd.
+ *
+ * This function decrements the peer reference counter with peer_unlock().
+ *
+ * If the peer is not registered, nothing happens.
+ *
+ * @param peer - peer to deregister
+ */
+extern void peer_writes_off(struct peer *peer);
+
+/**
+ * Notifies the write thread that there is work to be done.
+ *
+ * This function has the effect of waking the write thread if it is sleeping.
+ * If the thread is not sleeping, this signal will be ignored.
+ */
+extern void peer_writes_wake(void);
+
+#endif /* _FRR_BGP_IO_H */
index 5994df23ece07754002eeb11bff07fcfa3ca3d7d..a89d72cc649094e12b512950efe502cac5470bb3 100644 (file)
@@ -1,4 +1,6 @@
 /* BGP packet management routine.
+ * Contains utility functions for constructing and consuming BGP messages.
+ * Copyright (C) 2017 Cumulus Networks
  * Copyright (C) 1999 Kunihiro Ishiguro
  *
  * This file is part of GNU Zebra.
@@ -34,7 +36,6 @@
 #include "plist.h"
 #include "queue.h"
 #include "filter.h"
-#include "lib/frr_pthread.h"
 
 #include "bgpd/bgpd.h"
 #include "bgpd/bgp_table.h"
@@ -56,6 +57,7 @@
 #include "bgpd/bgp_vty.h"
 #include "bgpd/bgp_updgrp.h"
 #include "bgpd/bgp_label.h"
+#include "bgpd/bgp_io.h"
 
 /* Linked list of active peers */
 static pthread_mutex_t *plist_mtx;
@@ -99,17 +101,6 @@ int bgp_packet_set_size(struct stream *s)
        return cp;
 }
 
-/**
- * Push a packet onto the beginning of the peer's output queue.
- * Must be externally synchronized around 'peer'.
- */
-static void bgp_packet_add_unsafe(struct peer *peer, struct stream *s)
-{
-       /* Add packet to the end of list. */
-       stream_fifo_push(peer->obuf, s);
-       peer_writes_wake();
-}
-
 /*
  * Push a packet onto the beginning of the peer's output queue.
  * This function acquires the peer's write mutex before proceeding.
@@ -117,17 +108,10 @@ static void bgp_packet_add_unsafe(struct peer *peer, struct stream *s)
 static void bgp_packet_add(struct peer *peer, struct stream *s)
 {
        pthread_mutex_lock(&peer->obuf_mtx);
-       bgp_packet_add_unsafe(peer, s);
+       stream_fifo_push(peer->obuf, s);
        pthread_mutex_unlock(&peer->obuf_mtx);
-}
 
-/**
- * Pop a packet off the end of the peer's output queue.
- * Must be externally synchronized around 'peer'.
- */
-static void bgp_packet_delete_unsafe(struct peer *peer)
-{
-       stream_free(stream_fifo_pop(peer->obuf));
+       peer_writes_wake();
 }
 
 static struct stream *bgp_update_packet_eor(struct peer *peer, afi_t afi,
@@ -172,10 +156,18 @@ static struct stream *bgp_update_packet_eor(struct peer *peer, afi_t afi,
        return s;
 }
 
-/* Get next packet to be written.  */
-static struct stream *bgp_write_packet(struct peer *peer)
+/*
+ * Enqueue onto the peer's output buffer any packets which are pending for the
+ * update group it is a member of.
+ *
+ * XXX: Severely needs performance work.
+ */
+int bgp_generate_updgrp_packets(struct thread *thread)
 {
-       struct stream *s = NULL;
+       struct peer *peer = THREAD_ARG(thread);
+       peer->t_generate_updgrp_packets = NULL;
+
+       struct stream *s;
        struct peer_af *paf;
        struct bpacket *next_pkt;
        afi_t afi;
@@ -187,104 +179,86 @@ static struct stream *bgp_write_packet(struct peer *peer)
         * update-delay post processing).
         */
        if (peer->status != Established)
-               return NULL;
+               return 0;
 
        if (peer->bgp && peer->bgp->main_peers_update_hold)
-               return NULL;
-
-       FOREACH_AFI_SAFI (afi, safi) {
-               paf = peer_af_find(peer, afi, safi);
-               if (!paf || !PAF_SUBGRP(paf))
-                       continue;
-               next_pkt = paf->next_pkt_to_send;
-
-               /* Try to generate a packet for the peer if we are at
-                * the end of
-                * the list. Always try to push out WITHDRAWs first. */
-               if (!next_pkt || !next_pkt->buffer) {
-                       next_pkt = subgroup_withdraw_packet(PAF_SUBGRP(paf));
-                       if (!next_pkt || !next_pkt->buffer)
-                               subgroup_update_packet(PAF_SUBGRP(paf));
-                       next_pkt = paf->next_pkt_to_send;
-               }
+               return 0;
 
-                       /* Try to generate a packet for the peer if we are at
-                        * the end of
-                        * the list. Always try to push out WITHDRAWs first. */
-                       if (!next_pkt || !next_pkt->buffer) {
-                               next_pkt = subgroup_withdraw_packet(
-                                       PAF_SUBGRP(paf));
-                               if (!next_pkt || !next_pkt->buffer)
-                                       subgroup_update_packet(PAF_SUBGRP(paf));
+       do {
+               s = NULL;
+               for (afi = AFI_IP; afi < AFI_MAX; afi++)
+                       for (safi = SAFI_UNICAST; safi < SAFI_MAX; safi++) {
+                               paf = peer_af_find(peer, afi, safi);
+                               if (!paf || !PAF_SUBGRP(paf))
+                                       continue;
                                next_pkt = paf->next_pkt_to_send;
-                       }
-
-                       /* If we still don't have a packet to send to the peer,
-                        * then
-                        * try to find out out if we have to send eor or if not,
-                        * skip to
-                        * the next AFI, SAFI.
-                        * Don't send the EOR prematurely... if the subgroup's
-                        * coalesce
-                        * timer is running, the adjacency-out structure is not
-                        * created
-                        * yet.
-                        */
-                       if (!next_pkt || !next_pkt->buffer) {
-                               if (CHECK_FLAG(peer->cap,
-                                              PEER_CAP_RESTART_RCV)) {
-                                       if (!(PAF_SUBGRP(paf))->t_coalesce
-                                           && peer->afc_nego[afi][safi]
-                                           && peer->synctime
-                                           && !CHECK_FLAG(
-                                                      peer->af_sflags[afi]
-                                                                     [safi],
-                                                      PEER_STATUS_EOR_SEND)) {
-                                               SET_FLAG(peer->af_sflags[afi]
-                                                                       [safi],
-                                                        PEER_STATUS_EOR_SEND);
 
-                                               if ((s = bgp_update_packet_eor(
-                                                            peer, afi, safi)))
-                                                       bgp_packet_add(peer, s);
+                               /* Try to generate a packet for the peer if we
+                                * are at the end of
+                                * the list. Always try to push out WITHDRAWs
+                                * first. */
+                               if (!next_pkt || !next_pkt->buffer) {
+                                       next_pkt = subgroup_withdraw_packet(
+                                               PAF_SUBGRP(paf));
+                                       if (!next_pkt || !next_pkt->buffer)
+                                               subgroup_update_packet(
+                                                       PAF_SUBGRP(paf));
+                                       next_pkt = paf->next_pkt_to_send;
+                               }
 
-                                               return s;
+                               /* If we still don't have a packet to send to
+                                * the peer, then
+                                * try to find out out if we have to send eor or
+                                * if not, skip to
+                                * the next AFI, SAFI.
+                                * Don't send the EOR prematurely... if the
+                                * subgroup's coalesce
+                                * timer is running, the adjacency-out structure
+                                * is not created
+                                * yet.
+                                */
+                               if (!next_pkt || !next_pkt->buffer) {
+                                       if (CHECK_FLAG(peer->cap,
+                                                      PEER_CAP_RESTART_RCV)) {
+                                               if (!(PAF_SUBGRP(paf))
+                                                            ->t_coalesce
+                                                   && peer->afc_nego[afi][safi]
+                                                   && peer->synctime
+                                                   && !CHECK_FLAG(
+                                                              peer->af_sflags
+                                                                      [afi]
+                                                                      [safi],
+                                                              PEER_STATUS_EOR_SEND)) {
+                                                       SET_FLAG(
+                                                               peer->af_sflags
+                                                                       [afi]
+                                                                       [safi],
+                                                               PEER_STATUS_EOR_SEND);
+
+                                                       if ((s = bgp_update_packet_eor(
+                                                                    peer, afi,
+                                                                    safi)))
+                                                               bgp_packet_add(
+                                                                       peer,
+                                                                       s);
+                                               }
                                        }
+                                       continue;
                                }
-                       }
-                       continue;
-               }
-
 
-                       /* Found a packet template to send, overwrite packet
-                        * with appropriate
-                        * attributes from peer and advance peer */
-                       s = bpacket_reformat_for_peer(next_pkt, paf);
-                       bgp_packet_add(peer, s);
-                       bpacket_queue_advance_peer(paf);
-                       return s;
-               }
-
-       return NULL;
-}
 
-static int bgp_generate_updgrp_packets(struct thread *thread)
-{
-       struct listnode *ln;
-       struct peer *peer;
-       pthread_mutex_lock(plist_mtx);
-       {
-               for (ALL_LIST_ELEMENTS_RO(plist, ln, peer))
-                       while (bgp_write_packet(peer))
-                               ;
+                               /* Found a packet template to send, overwrite
+                                * packet with appropriate
+                                * attributes from peer and advance peer */
+                               s = bpacket_reformat_for_peer(next_pkt, paf);
+                               bgp_packet_add(peer, s);
+                               bpacket_queue_advance_peer(paf);
+                       }
+       } while (s);
 
-               t_generate_updgrp_packets = NULL;
-       }
-       pthread_mutex_unlock(plist_mtx);
        return 0;
 }
 
-
 /*
  * Creates a BGP Keepalive packet and appends it to the peer's output queue.
  */
@@ -2168,266 +2142,3 @@ done:
 
        return 0;
 }
-
-/* ------------- write thread ------------------ */
-
-/**
- * Flush peer output buffer.
- *
- * This function pops packets off of peer->obuf and writes them to peer->fd.
- * The amount of packets written is equal to the minimum of peer->wpkt_quanta
- * and the number of packets on the output buffer.
- *
- * If write() returns an error, the appropriate FSM event is generated.
- *
- * The return value is equal to the number of packets written
- * (which may be zero).
- */
-static int bgp_write(struct peer *peer)
-{
-       u_char type;
-       struct stream *s;
-       int num;
-       int update_last_write = 0;
-       unsigned int count = 0;
-       unsigned int oc = 0;
-
-       /* Write packets. The number of packets written is the value of
-        * bgp->wpkt_quanta or the size of the output buffer, whichever is
-        * smaller.*/
-       while (count < peer->bgp->wpkt_quanta
-              && (s = stream_fifo_head(peer->obuf))) {
-               int writenum;
-               do { // write a full packet, or return on error
-                       writenum = stream_get_endp(s) - stream_get_getp(s);
-                       num = write(peer->fd, STREAM_PNT(s), writenum);
-
-                       if (num < 0) {
-                               if (!ERRNO_IO_RETRY(errno))
-                                       BGP_EVENT_ADD(peer, TCP_fatal_error);
-
-                               goto done;
-                       } else if (num != writenum) // incomplete write
-                               stream_forward_getp(s, num);
-
-               } while (num != writenum);
-
-               /* Retrieve BGP packet type. */
-               stream_set_getp(s, BGP_MARKER_SIZE + 2);
-               type = stream_getc(s);
-
-               switch (type) {
-               case BGP_MSG_OPEN:
-                       peer->open_out++;
-                       break;
-               case BGP_MSG_UPDATE:
-                       peer->update_out++;
-                       break;
-               case BGP_MSG_NOTIFY:
-                       peer->notify_out++;
-                       /* Double start timer. */
-                       peer->v_start *= 2;
-
-                       /* Overflow check. */
-                       if (peer->v_start >= (60 * 2))
-                               peer->v_start = (60 * 2);
-
-                       /* Handle Graceful Restart case where the state changes
-                          to
-                          Connect instead of Idle */
-                       /* Flush any existing events */
-                       BGP_EVENT_ADD(peer, BGP_Stop);
-                       goto done;
-
-               case BGP_MSG_KEEPALIVE:
-                       peer->keepalive_out++;
-                       break;
-               case BGP_MSG_ROUTE_REFRESH_NEW:
-               case BGP_MSG_ROUTE_REFRESH_OLD:
-                       peer->refresh_out++;
-                       break;
-               case BGP_MSG_CAPABILITY:
-                       peer->dynamic_cap_out++;
-                       break;
-               }
-
-               count++;
-               /* OK we send packet so delete it. */
-               bgp_packet_delete_unsafe(peer);
-               update_last_write = 1;
-       }
-
-done : {
-       /* Update last_update if UPDATEs were written. */
-       if (peer->update_out > oc)
-               peer->last_update = bgp_clock();
-
-       /* If we TXed any flavor of packet update last_write */
-       if (update_last_write)
-               peer->last_write = bgp_clock();
-}
-
-       return count;
-}
-
-void peer_writes_init(void)
-{
-       plist_mtx = XCALLOC(MTYPE_PTHREAD, sizeof(pthread_mutex_t));
-       write_cond = XCALLOC(MTYPE_PTHREAD, sizeof(pthread_cond_t));
-
-       // initialize mutex
-       pthread_mutex_init(plist_mtx, NULL);
-
-       // use monotonic clock with condition variable
-       pthread_condattr_t attrs;
-       pthread_condattr_init(&attrs);
-       pthread_condattr_setclock(&attrs, CLOCK_MONOTONIC);
-       pthread_cond_init(write_cond, &attrs);
-       pthread_condattr_destroy(&attrs);
-
-       // initialize peerlist
-       plist = list_new();
-}
-
-static void peer_writes_finish(void *arg)
-{
-       bgp_packet_writes_thread_run = false;
-
-       if (plist)
-               list_delete(plist);
-
-       plist = NULL;
-
-       pthread_mutex_unlock(plist_mtx);
-       pthread_mutex_destroy(plist_mtx);
-       pthread_cond_destroy(write_cond);
-
-       XFREE(MTYPE_PTHREAD, plist_mtx);
-       XFREE(MTYPE_PTHREAD, write_cond);
-}
-
-/**
- * Entry function for peer packet flushing pthread.
- *
- * peer_writes_init() must be called prior to this.
- */
-void *peer_writes_start(void *arg)
-{
-       struct timeval currtime = {0, 0};
-       struct timeval sleeptime = {0, 500};
-       struct timespec next_update = {0, 0};
-
-       struct listnode *ln;
-       struct peer *peer;
-
-       pthread_mutex_lock(plist_mtx);
-
-       // register cleanup handler
-       pthread_cleanup_push(&peer_writes_finish, NULL);
-
-       bgp_packet_writes_thread_run = true;
-
-       while (bgp_packet_writes_thread_run) { // wait around until next update
-                                              // time
-               if (plist->count > 0)
-                       pthread_cond_timedwait(write_cond, plist_mtx,
-                                              &next_update);
-               else // wait around until we have some peers
-                       while (plist->count == 0
-                              && bgp_packet_writes_thread_run)
-                               pthread_cond_wait(write_cond, plist_mtx);
-
-               for (ALL_LIST_ELEMENTS_RO(plist, ln, peer)) {
-                       pthread_mutex_lock(&peer->obuf_mtx);
-                       {
-                               bgp_write(peer);
-                       }
-                       pthread_mutex_unlock(&peer->obuf_mtx);
-
-                       if (!bgp_packet_writes_thread_run)
-                               break;
-               }
-
-               // schedule update packet generation on main thread
-               if (!t_generate_updgrp_packets)
-                       t_generate_updgrp_packets = thread_add_event(
-                               bm->master, bgp_generate_updgrp_packets, NULL,
-                               0);
-
-               monotime(&currtime);
-               timeradd(&currtime, &sleeptime, &currtime);
-               TIMEVAL_TO_TIMESPEC(&currtime, &next_update);
-       }
-
-       // clean up
-       pthread_cleanup_pop(1);
-
-       return NULL;
-}
-
-/**
- * Turns on packet writing for a peer.
- */
-void peer_writes_on(struct peer *peer)
-{
-       if (peer->status == Deleted)
-               return;
-
-       pthread_mutex_lock(plist_mtx);
-       {
-               struct listnode *ln, *nn;
-               struct peer *p;
-
-               // make sure this peer isn't already in the list
-               for (ALL_LIST_ELEMENTS(plist, ln, nn, p))
-                       if (p == peer) {
-                               pthread_mutex_unlock(plist_mtx);
-                               return;
-                       }
-
-               peer_lock(peer);
-               listnode_add(plist, peer);
-
-               SET_FLAG(peer->thread_flags, PEER_THREAD_WRITES_ON);
-       }
-       pthread_mutex_unlock(plist_mtx);
-       peer_writes_wake();
-}
-
-/**
- * Turns off packet writing for a peer.
- */
-void peer_writes_off(struct peer *peer)
-{
-       struct listnode *ln, *nn;
-       struct peer *p;
-       pthread_mutex_lock(plist_mtx);
-       {
-               for (ALL_LIST_ELEMENTS(plist, ln, nn, p))
-                       if (p == peer) {
-                               list_delete_node(plist, ln);
-                               peer_unlock(peer);
-                               break;
-                       }
-
-               UNSET_FLAG(peer->thread_flags, PEER_THREAD_WRITES_ON);
-       }
-       pthread_mutex_unlock(plist_mtx);
-}
-
-/**
- * Wakes up the write thread to do work.
- */
-void peer_writes_wake()
-{
-       pthread_cond_signal(write_cond);
-}
-
-int peer_writes_stop(void **result)
-{
-       struct frr_pthread *fpt = frr_pthread_get(PTHREAD_WRITE);
-       bgp_packet_writes_thread_run = false;
-       peer_writes_wake();
-       pthread_join(fpt->thread, result);
-       return 0;
-}
index 2c252012f9e6b6538b0677c903a2323e03108bdf..d7080d7fb63861d74450f40940b0f17918c63d66 100644 (file)
@@ -67,11 +67,6 @@ extern int bgp_packet_set_size(struct stream *s);
 /* Control variable for write thread. */
 extern bool bgp_packet_writes_thread_run;
 
-extern void peer_writes_init(void);
-extern void *peer_writes_start(void *arg);
-extern void peer_writes_on(struct peer *peer);
-extern void peer_writes_off(struct peer *peer);
-extern void peer_writes_wake(void);
-extern int peer_writes_stop(void **result);
+extern int bgp_generate_updgrp_packets(struct thread *);
 
 #endif /* _QUAGGA_BGP_PACKET_H */
index 306a31e409baec1d9f45d190caa34f30651ac8e3..cfe5d5c670b0f6d5eb7430098d05102c49954694 100644 (file)
@@ -77,6 +77,7 @@
 #include "bgpd/bgp_memory.h"
 #include "bgpd/bgp_evpn_vty.h"
 #include "bgpd/bgp_keepalives.h"
+#include "bgpd/bgp_io.h"
 
 
 DEFINE_MTYPE_STATIC(BGPD, PEER_TX_SHUTDOWN_MSG, "Peer shutdown message (TX)");
index 0e80226ebeeaddb6505327e8e7648f345603fd4c..4fb784e2455b8cc139e7f519341769bb2627006b 100644 (file)
@@ -806,6 +806,7 @@ struct peer {
        struct thread *t_pmax_restart;
        struct thread *t_gr_restart;
        struct thread *t_gr_stale;
+       struct thread *t_generate_updgrp_packets;
 
        /* Thread flags. */
        u_int16_t thread_flags;