From 1d747210d5248f4f4a7e38bb27b212b476d3b3b8 Mon Sep 17 00:00:00 2001
From: Jens Osterkamp <jens@linux.vnet.ibm.com>
Date: Thu, 3 Feb 2011 23:00:09 +0000
Subject: [PATCH 08/51] robustness: proper handling of LINK_DOWN/UP
This patch aims at more robustness in the VDP code, especially with regard
of error situations: link down/up, switch reboot, resends, ...
It contains the following changes:
- limit number of timers in VDP sm
Multiple link up netlink messages can cause more than one VDP
timer for one interface to be started. This causes multiple sendout of ECP
frames. This patch fixes it.
- stop keepalive timers while processing
Keepalive timers have to be stopped while the profile is in any processing.
- proper startup of VDP timer/ECP tx sm
After a LINK_DOWN/LINK_UP again, make sure that the VDP timer is properly
started again, even if multiple LINK_UP events are received.
Ensure that ECP tx state machine is in a proper state after LINK_UP.
- improved sequence nr handling
Given we send out multiple VDP TLVs in on ECP frame, we have to note the
sequence nr used to send it out. This enables us on reception of an ack
frame to identify all acked VSI types and mark them as ackReceived for
further processing.
- improved handling of timeouts and resends
implement a better control of timeouts and resends of ECP frames, if
necessary.
- stop ECP TX sm until ack was received
This should also save us some cycles.
Signed-off-by: Jens Osterkamp <jens@linux.vnet.ibm.com>
Signed-off-by: Petr Sabata <psabata@redhat.com>
---
ecp/ecp.c | 8 +++-
ecp/ecp_rx.c | 1 +
ecp/ecp_tx.c | 4 ++-
include/lldp_vdp.h | 2 +
lldp_vdp.c | 102 ++++++++++++++++++++++++++++++++++++---------------
5 files changed, 84 insertions(+), 33 deletions(-)
diff --git a/ecp/ecp.c b/ecp/ecp.c
index 3798ae6..1f46664 100644
--- a/ecp/ecp.c
+++ b/ecp/ecp.c
@@ -60,14 +60,18 @@ int ecp_init(char *ifname)
goto fail;
}
- vd->ecp.l2 = l2_packet_init(vd->ifname, NULL, ETH_P_ECP,
- ecp_rx_ReceiveFrame, vd, 1);
+ if (!vd->ecp.l2) {
+ vd->ecp.l2 = l2_packet_init(vd->ifname, NULL, ETH_P_ECP,
+ ecp_rx_ReceiveFrame, vd, 1);
+ }
+
if (!vd->ecp.l2) {
LLDPAD_ERR("ERROR: Failed to open register layer 2 access to "
"ETH_P_ECP\n");
goto fail;
}
+ vd->ecp.ackTimerExpired = true;
ecp_tx_run_sm(vd);
ecp_rx_run_sm(vd);
diff --git a/ecp/ecp_rx.c b/ecp/ecp_rx.c
index bfd51fa..6bd6237 100644
--- a/ecp/ecp_rx.c
+++ b/ecp/ecp_rx.c
@@ -268,6 +268,7 @@ void ecp_rx_ReceiveFrame(void *ctx, unsigned int ifindex, const u8 *buf, size_t
vd->ecp.ackReceived = true;
LLDPAD_DBG("%s(%i)-%s: received ack frame \n", __func__, __LINE__, vd->ifname);
ecp_print_framein(vd);
+ vdp_ack_profiles(vd, vd->ecp.seqECPDU);
ecp_tx_run_sm(vd);
vd->ecp.ackReceived = false;
break;
diff --git a/ecp/ecp_tx.c b/ecp/ecp_tx.c
index 54a278f..cc3cc51 100644
--- a/ecp/ecp_tx.c
+++ b/ecp/ecp_tx.c
@@ -172,6 +172,8 @@ bool ecp_build_ECPDU(struct vdp_data *vd)
fb_offset += ptlv->size;
}
+ p->seqnr = vd->ecp.lastSequence;
+
ptlv = free_pkd_tlv(ptlv);
}
@@ -406,7 +408,7 @@ static bool ecp_set_tx_state(struct vdp_data *vd)
return true;
}
ecp_tx_change_state(vd, ECP_TX_WAIT_FOR_ACK);
- return true;
+ return false;
case ECP_TX_WAIT_FOR_ACK:
if (vd->ecp.ackTimerExpired) {
vd->ecp.retries++;
diff --git a/include/lldp_vdp.h b/include/lldp_vdp.h
index 3d1411d..5fc1a22 100644
--- a/include/lldp_vdp.h
+++ b/include/lldp_vdp.h
@@ -126,6 +126,7 @@ struct vsi_profile {
int ackReceived;
int keepaliveTimer;
int state;
+ int seqnr;
bool localChange;
LIST_ENTRY(vsi_profile) profile;
};
@@ -137,6 +138,7 @@ struct vdp_data {
int role;
int keepaliveTimer;
int ackTimer;
+ int nroftimers;
LIST_HEAD(profile_head, vsi_profile) profile_head;
LIST_ENTRY(vdp_data) entry;
};
diff --git a/lldp_vdp.c b/lldp_vdp.c
index 6ebf171..a3781b0 100644
--- a/lldp_vdp.c
+++ b/lldp_vdp.c
@@ -129,6 +129,26 @@ void vdp_print_profile(struct vsi_profile *profile)
LLDPAD_DBG("vlan: %i\n", profile->vlan);
}
+/* vdp_ack_profiles - set ackReceived for all profiles with seqnr
+ * @vd: vd for the interface
+ * @seqnr: seqnr the ack has been received with
+ *
+ * no return value
+ *
+ * set the ackReceived for all profiles which have been sent out with
+ * the seqnr that we now have received the ack for.
+ */
+void vdp_ack_profiles(struct vdp_data *vd, int seqnr)
+{
+ struct vsi_profile *p;
+
+ LIST_FOREACH(p, &vd->profile_head, profile) {
+ if (p->seqnr == seqnr)
+ p->ackReceived = true;
+ }
+
+}
+
/* vdp_somethingChangedLocal - set flag if profile has changed
* @profile: profile to set the flag for
* @flag: set the flag to true or false
@@ -185,6 +205,8 @@ void vdp_timeout_handler(void *eloop_data, void *user_ctx)
vd = (struct vdp_data *) user_ctx;
+ vd->nroftimers--;
+
LIST_FOREACH(p, &vd->profile_head, profile) {
if (p->ackTimer > 0)
p->ackTimer--;
@@ -194,9 +216,15 @@ void vdp_timeout_handler(void *eloop_data, void *user_ctx)
if (vdp_ackTimer_expired(p) ||
vdp_keepaliveTimer_expired(p) ||
- p->ackReceived ||
- p->localChange)
+ p->ackReceived) {
+ LLDPAD_DBG("%s(%i): vdp_ackTimer_expired %i\n", __func__, __LINE__,
+ vdp_ackTimer_expired(p));
+ LLDPAD_DBG("%s(%i): p->ackReceived %i\n", __func__, __LINE__,
+ p->ackReceived);
+ LLDPAD_DBG("%s(%i): vdp_keepaliveTimer_expired %i\n", __func__, __LINE__,
+ vdp_keepaliveTimer_expired(p));
vdp_vsi_sm_station(p);
+ }
}
vdp_start_timer(vd);
@@ -214,6 +242,8 @@ static int vdp_stop_timer(struct vdp_data *vd)
LLDPAD_DBG("%s(%i)-%s: stopping vdp timer\n", __func__, __LINE__,
vd->ifname);
+ vd->nroftimers--;
+
return eloop_cancel_timeout(vdp_timeout_handler, NULL, (void *) vd);
}
@@ -231,6 +261,15 @@ static int vdp_start_timer(struct vdp_data *vd)
secs = 0;
usecs = VDP_TIMER_GRANULARITY;
+ vd->nroftimers++;
+
+ if (vd->nroftimers > 1) {
+ LLDPAD_WARN("%s(%i)-%s: nroftimers %i !\n", __func__, __LINE__,
+ vd->ifname, vd->nroftimers);
+ vd->nroftimers--;
+ return 0;
+ }
+
return eloop_register_timeout(secs, usecs, vdp_timeout_handler, NULL, (void *) vd);
}
@@ -301,6 +340,7 @@ void vdp_vsi_change_station_state(struct vsi_profile *profile, u8 newstate)
break;
case VSI_ASSOC_PROCESSING:
assert((profile->state == VSI_PREASSOCIATED) ||
+ (profile->state == VSI_ASSOCIATED) ||
(profile->state == VSI_UNASSOCIATED));
break;
case VSI_ASSOCIATED:
@@ -308,7 +348,8 @@ void vdp_vsi_change_station_state(struct vsi_profile *profile, u8 newstate)
(profile->state == VSI_ASSOCIATED));
break;
case VSI_PREASSOC_PROCESSING:
- assert(profile->state == VSI_UNASSOCIATED);
+ assert((profile->state == VSI_PREASSOCIATED) ||
+ profile->state == VSI_UNASSOCIATED);
break;
case VSI_PREASSOCIATED:
assert((profile->state == VSI_PREASSOC_PROCESSING) ||
@@ -366,8 +407,6 @@ static bool vdp_vsi_set_station_state(struct vsi_profile *profile)
}
return false;
case VSI_ASSOC_PROCESSING:
- LLDPAD_DBG("profile->ackReceived %i, vdp_ackTimer %i\n",
- profile->ackReceived, profile->ackTimer);
if (profile->ackReceived) {
vdp_vsi_change_station_state(profile, VSI_ASSOCIATED);
return true;
@@ -383,24 +422,36 @@ static bool vdp_vsi_set_station_state(struct vsi_profile *profile)
} else if (profile->mode == VDP_MODE_DEASSOCIATE) {
vdp_vsi_change_station_state(profile, VSI_DEASSOC_PROCESSING);
return true;
+ } else if (vdp_keepaliveTimer_expired(profile)) {
+ vdp_stop_keepaliveTimer(profile);
+ vdp_somethingChangedLocal(profile, true);
+ vdp_vsi_change_station_state(profile, VSI_ASSOC_PROCESSING);
+ return true;
}
return false;
case VSI_PREASSOC_PROCESSING:
+ LLDPAD_DBG("%s(%i): profile->ackReceived %i, vdp_ackTimer %i\n", __func__,
+ __LINE__, profile->ackReceived, profile->ackTimer);
if (profile->ackReceived) {
vdp_vsi_change_station_state(profile, VSI_PREASSOCIATED);
return true;
- } else if (vdp_ackTimer_expired(profile)) {
+ } else if (!profile->ackReceived && vdp_ackTimer_expired(profile)) {
vdp_vsi_change_station_state(profile, VSI_EXIT);
return true;
}
+ return false;
case VSI_PREASSOCIATED:
- if (profile->mode == VDP_MODE_DEASSOCIATE) {
- vdp_vsi_change_station_state(profile, VSI_DEASSOC_PROCESSING);
- return true;
- }
if (profile->mode == VDP_MODE_ASSOCIATE) {
vdp_vsi_change_station_state(profile, VSI_ASSOC_PROCESSING);
return true;
+ } else if (profile->mode == VDP_MODE_DEASSOCIATE) {
+ vdp_vsi_change_station_state(profile, VSI_DEASSOC_PROCESSING);
+ return true;
+ } else if (vdp_keepaliveTimer_expired(profile)) {
+ vdp_stop_keepaliveTimer(profile);
+ vdp_somethingChangedLocal(profile, true);
+ vdp_vsi_change_station_state(profile, VSI_PREASSOC_PROCESSING);
+ return true;
}
return false;
case VSI_DEASSOC_PROCESSING:
@@ -438,55 +489,44 @@ void vdp_vsi_sm_station(struct vsi_profile *profile)
case VSI_UNASSOCIATED:
break;
case VSI_ASSOC_PROCESSING:
+ vdp_stop_keepaliveTimer(profile);
profile->response = VDP_RESPONSE_NO_RESPONSE;
if (profile->localChange) {
ecp_somethingChangedLocal(vd);
+ vdp_start_ackTimer(profile);
ecp_tx_run_sm(vd);
}
- vdp_somethingChangedLocal(profile, false);
- vdp_start_ackTimer(profile);
break;
case VSI_ASSOCIATED:
profile->ackReceived = false;
vdp_somethingChangedLocal(profile, false);
vdp_stop_ackTimer(profile);
- if (vdp_keepaliveTimer_expired(profile)) {
- vdp_somethingChangedLocal(profile, true);
- ecp_somethingChangedLocal(vd);
- ecp_tx_run_sm(vd);
- vdp_start_keepaliveTimer(profile);
- }
+ vdp_start_keepaliveTimer(profile);
break;
case VSI_PREASSOC_PROCESSING:
- /* send out profile */
+ vdp_stop_keepaliveTimer(profile);
profile->response = VDP_RESPONSE_NO_RESPONSE;
if (profile->localChange) {
ecp_somethingChangedLocal(vd);
+ vdp_start_ackTimer(profile);
ecp_tx_run_sm(vd);
}
- vdp_somethingChangedLocal(profile, false);
- vdp_start_ackTimer(profile);
break;
case VSI_PREASSOCIATED:
profile->ackReceived = false;
vdp_somethingChangedLocal(profile, false);
vdp_stop_ackTimer(profile);
- if (vdp_keepaliveTimer_expired(profile)) {
- vdp_somethingChangedLocal(profile, true);
- ecp_somethingChangedLocal(vd);
- ecp_tx_run_sm(vd);
- }
vdp_start_keepaliveTimer(profile);
break;
case VSI_DEASSOC_PROCESSING:
+ vdp_stop_keepaliveTimer(profile);
profile->response = VDP_RESPONSE_NO_RESPONSE;
vdp_stop_keepaliveTimer(profile);
if (profile->localChange) {
ecp_somethingChangedLocal(vd);
+ vdp_start_ackTimer(profile);
ecp_tx_run_sm(vd);
}
- vdp_somethingChangedLocal(profile, false);
- vdp_start_ackTimer(profile);
break;
case VSI_EXIT:
/* TODO: send DEASSOC here ? */
@@ -1175,7 +1215,7 @@ void vdp_ifup(char *ifname)
vd = vdp_data(ifname);
if (vd) {
LLDPAD_WARN("%s:%s vdp data already exists !\n", __func__, ifname);
- goto out_start_timer;
+ goto out_start_again;
}
/* not found, alloc/init per-port module data */
@@ -1204,6 +1244,7 @@ void vdp_ifup(char *ifname)
ud = find_module_user_data_by_if(ifname, &lldp_head, LLDP_MOD_VDP);
LIST_INSERT_HEAD(&ud->head, vd, entry);
+out_start_again:
if (ecp_init(ifname)) {
LLDPAD_ERR("%s:%s unable to init ecp !\n", __func__, ifname);
vdp_ifdown(ifname);
@@ -1213,7 +1254,8 @@ void vdp_ifup(char *ifname)
vd->keepaliveTimer = VDP_KEEPALIVE_TIMER_DEFAULT;
vd->ackTimer = VDP_ACK_TIMER_DEFAULT;
-out_start_timer:
+ LLDPAD_DBG("%s(%i)-%s: starting vdp timer (%i)\n", __func__, __LINE__,
+ vd->ifname, vd->nroftimers);
vdp_start_timer(vd);
LLDPAD_DBG("%s:%s vdp added\n", __func__, ifname);
--
1.7.4.4