Re: [PATCH net 1/4] netfilter: flowtable: do not queue offload work after DESTROY
From: Pablo Neira Ayuso
Date: Sat Nov 21 2026 - 18:23:59 EST
On Wed, Nov 18, 2026 at 12:12:47PM -0800, Florian Fainelli wrote:
> nf_flow_offload_refresh() can be reached by a packet path reader which
> looked the flow up before it was torn down. NF_FLOW_PENDING keeps the
> DESTROY work from being queued while another work item is pending, but
> once the DESTROY work has run and cleared the bit, a reader can take it
> again and queue a REPLACE (or STATS) work for a flow which is already
> being removed.
>
> GC removes a hardware flow after NF_FLOW_HW_DEAD is set, and frees it after
> an RCU grace period, which the reader can outlive in a preemptible RCU
> read-side section. The queued work then dereferences a freed flow and
> the flowtable, and can program the hardware again with a rule keyed by
> the freed flow's cookie.
>
> NF_FLOW_HW_DYING is set when DESTROY is queued, and DESTROY needs
> NF_FLOW_PENDING, so a work item which holds the pending bit and finds
> NF_FLOW_HW_DYING clear prevents DESTROY from being queued until it
> completes. Check it after taking the pending bit, and do not queue
> anything other than DESTROY when it is set.
>
> Not reproduced: this is a race between a long RCU read-side section and
> the GC and offload workers. Found by review of a related change.
>
> Fixes: c29f74e0df7a ("netfilter: nf_flow_table: hardware offload support")
> Signed-off-by: Florian Fainelli <florian.fainelli@xxxxxxxxxxxx>
> ---
> net/netfilter/nf_flow_table_offload.c | 13 +++++++++++++
> 1 file changed, 13 insertions(+)
>
> diff --git a/net/netfilter/nf_flow_table_offload.c b/net/netfilter/nf_flow_table_offload.c
> index 4365859220e6..62f460d38efa 100644
> --- a/net/netfilter/nf_flow_table_offload.c
> +++ b/net/netfilter/nf_flow_table_offload.c
> @@ -1088,6 +1088,19 @@ nf_flow_offload_work_alloc(struct nf_flowtable *flowtable,
> if (test_and_set_bit(NF_FLOW_PENDING, &flow->flags))
> return NULL;
>
> + /* GC only removes a hardware flow after its DESTROY work has run, and
> + * NF_FLOW_PENDING keeps another one from being queued, so nothing can
> + * free the flow while this work is pending, unless DESTROY was queued
> + * before we took the pending bit. A reader which found the flow before
> + * it was torn down must not queue more work then, as the flow could be
> + * freed before that work runs.
> + */
> + if (cmd != FLOW_CLS_DESTROY &&
> + test_bit(NF_FLOW_HW_DYING, &flow->flags)) {
> + clear_bit(NF_FLOW_PENDING, &flow->flags);
> + return NULL;
> + }
Maybe address by checking for the _CLOSING + _TEARDOWN flag? If those
are set on, then hardware will be scheduled for removal soon.
I guess there is also another hypothetical case where _CLOSING is set
on, but HW_DYING was not set on yet?
Probably the HW_DYING bit can go away, _CLOSING was added for hw
offload specifically, to schedule the removal in hardware while the
flow still remains with IPS_OFFLOAD.
diff --git a/net/netfilter/nf_flow_table_offload.c b/net/netfilter/nf_flow_table_offload.c
index 11434cfb105b..7104d332d98a 100644
--- a/net/netfilter/nf_flow_table_offload.c
+++ b/net/netfilter/nf_flow_table_offload.c
@@ -1132,7 +1132,8 @@ void nf_flow_offload_refresh(struct nf_flowtable *flowtable,
if (!offload)
return;
- if (test_bit(NF_FLOW_TEARDOWN, &flow->flags)) {
+ if (test_bit(NF_FLOW_CLOSING, &flow->flags) ||
+ test_bit(NF_FLOW_TEARDOWN, &flow->flags)) {
clear_bit(NF_FLOW_PENDING, &flow->flags);
kfree(offload);
return
> +
> offload = kmalloc_obj(struct flow_offload_work, GFP_ATOMIC);
> if (!offload) {
> clear_bit(NF_FLOW_PENDING, &flow->flags);
> --
> 2.53.0
>
>