[5/5] opensm: Resend LFTs/VLArb/SL2VL MADs in case of error

Commit Message

Jens Domke Feb. 4, 2014, 5:54 a.m. UTC

Dear Alex,

the memset call in sl2vl_update_table causes segmentation faults if 
force_update=1, since p_tbl won't get anything assigned and remains NULL.
Please, find a possible fix attached.

Regards,
Jens

On 03.02.14 20:05, Alex Netes wrote:
> There are several MADs that we only SET during the sweep (and never
> GET).
> Zero the stored block, so in case the MAD will end up with error,
> we will resend it during the next sweep.
>
> Signed-off-by: Alex Netes <[e-mail]>
> ---
>   opensm/osm_qos.c       |   13 +++++++++++++
>   opensm/osm_ucast_mgr.c |    7 +++++++
>   2 files changed, 20 insertions(+), 0 deletions(-)
>
> diff --git a/opensm/osm_qos.c b/opensm/osm_qos.c
> index a301803..473e3c8 100644
> --- a/opensm/osm_qos.c
> +++ b/opensm/osm_qos.c
> @@ -183,6 +183,13 @@ static ib_api_status_t vlarb_update_table_block(osm_sm_t * sm,
>   	if (!p_mad)
>   		return IB_INSUFFICIENT_MEMORY;
>
> +	/*
> +	 * Zero the stored VL Arbitration block, so in case the MAD will
> +	 * end up with error, we will resend it in the next sweep.
> +	 */
> +	memset(&p->vl_arb[block_num], 0,
> +	       block_length * sizeof(block.vl_entry[0]));
> +
>   	cl_qlist_insert_tail(mad_list, &p_mad->list_item);
>
>   	return IB_SUCCESS;
> @@ -272,6 +279,12 @@ static ib_api_status_t sl2vl_update_table(osm_sm_t * sm, osm_physp_t * p,
>   	if (!p_mad)
>   		return IB_INSUFFICIENT_MEMORY;
>
> +	/*
> +	 * Zero the stored SL2VL block, so in case the MAD will
> +	 * end up with error, we will resend it in the next sweep.
> +	 */
> +	memset(p_tbl, 0, sizeof(tbl));
> +
>   	cl_qlist_insert_tail(mad_list, &p_mad->list_item);
>   	return IB_SUCCESS;
>   }
> diff --git a/opensm/osm_ucast_mgr.c b/opensm/osm_ucast_mgr.c
> index 8194307..c8a7360 100644
> --- a/opensm/osm_ucast_mgr.c
> +++ b/opensm/osm_ucast_mgr.c
> @@ -1002,6 +1002,13 @@ static int set_lft_block(IN osm_switch_t *p_sw, IN osm_ucast_mgr_t *p_mgr,
>   		    IB_SMP_DATA_SIZE))
>   		return 0;
>
> +	/*
> +	 * Zero the stored LFT block, so in case the MAD will end up
> +	 * with error, we will resend it in the next sweep.
> +	 */
> +	memset(p_sw->lft + block_id_ho * IB_SMP_DATA_SIZE, OSM_NO_PATH,
> +	       IB_SMP_DATA_SIZE);
> +
>   	OSM_LOG(p_mgr->p_log, OSM_LOG_DEBUG,
>   		"Writing FT block %u to switch 0x%" PRIx64 "\n", block_id_ho,
>   		cl_ntoh64(context.lft_context.node_guid));
>
From 3cbe8f10c4ab7d83c5898b67e42d9e99be355c05 Mon Sep 17 00:00:00 2001
From: Jens Domke <[e-mail]>
Date: Tue, 4 Feb 2014 14:47:44 +0900
Subject: [PATCH 1/1] osm_qos.c: fix potential segmentation fault

if force_update=1, then p_tbl remains NULL and therefore memset
crashes

Signed-off-by: Jens Domke <[e-mail]>
---
 opensm/osm_qos.c |    5 +++--
 1 files changed, 3 insertions(+), 2 deletions(-)

Patch

diff --git a/opensm/osm_qos.c b/opensm/osm_qos.c
index 473e3c8..76f0ff6 100644
--- a/opensm/osm_qos.c
+++ b/opensm/osm_qos.c
@@ -252,7 +252,7 @@  static ib_api_status_t sl2vl_update_table(osm_sm_t * sm, osm_physp_t * p,
 					  const ib_slvl_table_t * sl2vl_table,
 					  cl_qlist_t *mad_list)
 {
-	ib_slvl_table_t tbl, *p_tbl;
+	ib_slvl_table_t tbl, *p_tbl = NULL;
 	unsigned vl_mask;
 	uint8_t vl1, vl2;
 	int i;
@@ -283,7 +283,8 @@  static ib_api_status_t sl2vl_update_table(osm_sm_t * sm, osm_physp_t * p,
 	 * Zero the stored SL2VL block, so in case the MAD will
 	 * end up with error, we will resend it in the next sweep.
 	 */
-	memset(p_tbl, 0, sizeof(tbl));
+	if (p_tbl)
+		memset(p_tbl, 0, sizeof(tbl));
 
 	cl_qlist_insert_tail(mad_list, &p_mad->list_item);
 	return IB_SUCCESS;