On Fri, Sep 14, 2018 at 1:59 PM Willem de Bruijn
<willemdebruijn.ker...@gmail.com> wrote:
>
> From: Willem de Bruijn <will...@google.com>
>
> Add net_offload flag NET_OFF_FLAG_GRO_OFF. If set, a net_offload will
> not be used for gro receive processing.
>
> Also add sysctl helper proc_do_net_offload that toggles this flag and
> register sysctls net.{core,ipv4,ipv6}.gro
>
> Signed-off-by: Willem de Bruijn <will...@google.com>
> ---
> diff --git a/net/core/dev.c b/net/core/dev.c
> index 20d9552afd38..0fd5273bc931 100644
> --- a/net/core/dev.c
> +++ b/net/core/dev.c
> @@ -154,6 +154,7 @@
> #define GRO_MAX_HEAD (MAX_HEADER + 128)
>
> static DEFINE_SPINLOCK(ptype_lock);
> +DEFINE_SPINLOCK(offload_lock);
> struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
> struct list_head ptype_all __read_mostly; /* Taps */
> static struct list_head offload_base __read_mostly;
> diff --git a/net/core/sysctl_net_core.c b/net/core/sysctl_net_core.c
> index b1a2c5e38530..d2d72afdd9eb 100644
> --- a/net/core/sysctl_net_core.c
> +++ b/net/core/sysctl_net_core.c
> @@ -15,6 +15,7 @@
> #include <linux/vmalloc.h>
> #include <linux/init.h>
> #include <linux/slab.h>
> +#include <linux/bitmap.h>
>
> #include <net/ip.h>
> #include <net/sock.h>
> @@ -34,6 +35,58 @@ static int net_msg_warn; /* Unused, but still a sysctl
> */
> int sysctl_fb_tunnels_only_for_init_net __read_mostly = 0;
> EXPORT_SYMBOL(sysctl_fb_tunnels_only_for_init_net);
>
> +extern spinlock_t offload_lock;
> +
> +#define NET_OFF_TBL_LEN 256
> +
> +int proc_do_net_offload(struct ctl_table *ctl, int write, void __user
> *buffer,
> + size_t *lenp, loff_t *ppos)
> +{
> + unsigned long bitmap[NET_OFF_TBL_LEN / (sizeof(unsigned long) << 3)];
> + struct ctl_table tbl = { .maxlen = NET_OFF_TBL_LEN, .data = bitmap };
> + unsigned long flag = (unsigned long) ctl->extra2;
> + struct net_offload __rcu **offs = ctl->extra1;
> + struct net_offload *off;
> + int i, ret;
> +
> + memset(bitmap, 0, sizeof(bitmap));
> +
> + spin_lock(&offload_lock);
> +
> + for (i = 0; i < tbl.maxlen; i++) {
> + off = rcu_dereference_protected(offs[i],
> lockdep_is_held(&offload_lock));
> + if (off && off->flags & flag) {
This does not actually work as is. No protocol will have this flag set
out of the box.
I was in the middle of rewriting some of this when it became topical,
so I sent it out for discussion. It's bound not to be the only bug of
the patchset as is. I'll work through them to get it back in shape.
