[PATCH 0/3] Various fixes for flowtable hardware offload

[PATCH 0/3] Various fixes for flowtable hardware offload

[Paul Blakey]

First two related to flushing the pending hardware offload work,
and the third is just a line that was accidently removed.

Paul Blakey (3):
  netfilter: flowtable: Fix hardware flush order on
    nf_flow_table_cleanup
  netfilter: flowtable: Fix missing flush hardware on table free
  netfilter: flowtable: Fix setting forgotten NF_FLOW_HW_DEAD flag

 net/netfilter/nf_flow_table_core.c    | 3 ++-
 net/netfilter/nf_flow_table_offload.c | 1 +
 2 files changed, 3 insertions(+), 1 deletion(-)

-- 
1.8.3.1

[PATCH 1/3] netfilter: flowtable: Fix hardware flush order on nf_flow_table_cleanup

[Paul Blakey]

On netdev down event, nf_flow_table_cleanup() is called for the relevant
device and it cleans all the tables that are on that device.
If one of those tables has hardware offload flag,
nf_flow_table_iterate_cleanup flushes hardware and then runs the gc.
But the gc can queue more hardware work, which will take time to execute.

Instead first add the work, then flush it, to execute it now.

Fixes: c29f74e0df7a ("netfilter: nf_flow_table: hardware offload support")
Signed-off-by: Paul Blakey <paulb@mellanox.com>
---
 net/netfilter/nf_flow_table_core.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/netfilter/nf_flow_table_core.c b/net/netfilter/nf_flow_table_core.c
index 7e91989..14a069c 100644
--- a/net/netfilter/nf_flow_table_core.c
+++ b/net/netfilter/nf_flow_table_core.c
@@ -529,9 +529,9 @@ static void nf_flow_table_do_cleanup(struct flow_offload *flow, void *data)
 static void nf_flow_table_iterate_cleanup(struct nf_flowtable *flowtable,
 					  struct net_device *dev)
 {
-	nf_flow_table_offload_flush(flowtable);
 	nf_flow_table_iterate(flowtable, nf_flow_table_do_cleanup, dev);
 	flush_delayed_work(&flowtable->gc_work);
+	nf_flow_table_offload_flush(flowtable);
 }
 
 void nf_flow_table_cleanup(struct net_device *dev)
-- 
1.8.3.1

[PATCH 2/3] netfilter: flowtable: Fix missing flush hardware on table free

[Paul Blakey]

If entries exist when freeing a hardware offload enabled table,
we queue work for hardware while running the gc iteration.

Execute it (flush) after queueing.

Fixes: c29f74e0df7a ("netfilter: nf_flow_table: hardware offload support")
Signed-off-by: Paul Blakey <paulb@mellanox.com>
---
 net/netfilter/nf_flow_table_core.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/net/netfilter/nf_flow_table_core.c b/net/netfilter/nf_flow_table_core.c
index 14a069c..8af28e1 100644
--- a/net/netfilter/nf_flow_table_core.c
+++ b/net/netfilter/nf_flow_table_core.c
@@ -553,6 +553,7 @@ void nf_flow_table_free(struct nf_flowtable *flow_table)
 	cancel_delayed_work_sync(&flow_table->gc_work);
 	nf_flow_table_iterate(flow_table, nf_flow_table_do_cleanup, NULL);
 	nf_flow_table_iterate(flow_table, nf_flow_offload_gc_step, flow_table);
+	nf_flow_table_offload_flush(flow_table);
 	rhashtable_destroy(&flow_table->rhashtable);
 }
 EXPORT_SYMBOL_GPL(nf_flow_table_free);
-- 
1.8.3.1

[PATCH 3/3] netfilter: flowtable: Fix setting forgotten NF_FLOW_HW_DEAD flag

[Paul Blakey]

During the refactor this was accidently removed.

Fixes: ae29045018c8 ("netfilter: flowtable: add nf_flow_offload_tuple() helper")
Signed-off-by: Paul Blakey <paulb@mellanox.com>
---
 net/netfilter/nf_flow_table_offload.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/net/netfilter/nf_flow_table_offload.c b/net/netfilter/nf_flow_table_offload.c
index c8b70ff..83e1db3 100644
--- a/net/netfilter/nf_flow_table_offload.c
+++ b/net/netfilter/nf_flow_table_offload.c
@@ -675,6 +675,7 @@ static void flow_offload_work_del(struct flow_offload_work *offload)
 {
 	flow_offload_tuple_del(offload, FLOW_OFFLOAD_DIR_ORIGINAL);
 	flow_offload_tuple_del(offload, FLOW_OFFLOAD_DIR_REPLY);
+	set_bit(NF_FLOW_HW_DEAD, &offload->flow->flags);
 }
 
 static void flow_offload_tuple_stats(struct flow_offload_work *offload,
-- 
1.8.3.1

Re: [PATCH 0/3] Various fixes for flowtable hardware offload

[Pablo Neira Ayuso]

On Thu, Jan 30, 2020 at 06:04:34PM +0200, Paul Blakey wrote:
> First two related to flushing the pending hardware offload work,
> and the third is just a line that was accidently removed.

Applied, thanks Paul.