[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CANEQ_++gbwU2Rvcqg5KtngZC1UX1XcjOKfPKRr3Pvxi+VyQX+Q@mail.gmail.com>
Date: Wed, 16 Jun 2021 12:16:52 +0300
From: Amit Klein <aksecurity@...il.com>
To: Greg Kroah-Hartman <gregkh@...uxfoundation.org>
Cc: linux-kernel@...r.kernel.org, stable@...r.kernel.org,
Eric Dumazet <edumazet@...gle.com>, Willy Tarreau <w@....eu>,
"David S. Miller" <davem@...emloft.net>,
Sasha Levin <sashal@...nel.org>
Subject: Re: [PATCH 5.4 175/244] inet: use bigger hash table for IP ID generation
Here is the patch (minus headers, description, etc. - I believe these
can be copied as is from the 5.x patch, but not sure about the
rules...). It can be applied to 4.14.236. If this is OK, I can move on
to 4.9 and 4.4.
net/ipv4/route.c | 41 ++++++++++++++++++++++++++++-------------
1 file changed, 28 insertions(+), 13 deletions(-)
diff --git a/net/ipv4/route.c b/net/ipv4/route.c
index 78d6bc61a1d8..022a2b748da3 100644
--- a/net/ipv4/route.c
+++ b/net/ipv4/route.c
@@ -70,6 +70,7 @@
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/mm.h>
+#include <linux/memblock.h>
#include <linux/string.h>
#include <linux/socket.h>
#include <linux/sockios.h>
@@ -485,8 +486,10 @@ static void ipv4_confirm_neigh(const struct
dst_entry *dst, const void *daddr)
__ipv4_confirm_neigh(dev, *(__force u32 *)pkey);
}
-#define IP_IDENTS_SZ 2048u
-
+/* Hash tables of size 2048..262144 depending on RAM size.
+ * Each bucket uses 8 bytes.
+ */
+static u32 ip_idents_mask __read_mostly;
static atomic_t *ip_idents __read_mostly;
static u32 *ip_tstamps __read_mostly;
@@ -496,12 +499,16 @@ static u32 *ip_tstamps __read_mostly;
*/
u32 ip_idents_reserve(u32 hash, int segs)
{
- u32 *p_tstamp = ip_tstamps + hash % IP_IDENTS_SZ;
- atomic_t *p_id = ip_idents + hash % IP_IDENTS_SZ;
- u32 old = ACCESS_ONCE(*p_tstamp);
- u32 now = (u32)jiffies;
+ u32 bucket, old, now = (u32)jiffies;
+ atomic_t *p_id;
+ u32 *p_tstamp;
u32 delta = 0;
+ bucket = hash & ip_idents_mask;
+ p_tstamp = ip_tstamps + bucket;
+ p_id = ip_idents + bucket;
+ old = READ_ONCE(*p_tstamp);
+
if (old != now && cmpxchg(p_tstamp, old, now) == old)
delta = prandom_u32_max(now - old);
@@ -3099,17 +3106,25 @@ struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;
int __init ip_rt_init(void)
{
int rc = 0;
+ void *idents_hash;
int cpu;
- ip_idents = kmalloc(IP_IDENTS_SZ * sizeof(*ip_idents), GFP_KERNEL);
- if (!ip_idents)
- panic("IP: failed to allocate ip_idents\n");
+ /* For modern hosts, this will use 2 MB of memory */
+ idents_hash = alloc_large_system_hash("IP idents",
+ sizeof(*ip_idents) + sizeof(*ip_tstamps),
+ 0,
+ 16, /* one bucket per 64 KB */
+ HASH_ZERO,
+ NULL,
+ &ip_idents_mask,
+ 2048,
+ 256*1024);
+
+ ip_idents = idents_hash;
- prandom_bytes(ip_idents, IP_IDENTS_SZ * sizeof(*ip_idents));
+ prandom_bytes(ip_idents, (ip_idents_mask + 1) * sizeof(*ip_idents));
- ip_tstamps = kcalloc(IP_IDENTS_SZ, sizeof(*ip_tstamps), GFP_KERNEL);
- if (!ip_tstamps)
- panic("IP: failed to allocate ip_tstamps\n");
+ ip_tstamps = idents_hash + (ip_idents_mask + 1) * sizeof(*ip_idents);
for_each_possible_cpu(cpu) {
struct uncached_list *ul = &per_cpu(rt_uncached_list, cpu);
On Wed, Jun 16, 2021 at 10:16 AM Greg Kroah-Hartman
<gregkh@...uxfoundation.org> wrote:
>
> On Wed, Jun 16, 2021 at 10:02:44AM +0300, Amit Klein wrote:
> > Hi Greg et al.
> >
> > I see that you backported this patch (increasing the IP ID hash table size)
> > to the newer LTS branches more than a month ago. But I don't see that it
> > was backported to older LTS branches (4.19, 4.14, 4.9, 4.4). Is this
> > intentional?
>
> It applies cleanly to 4.19, but not the older ones. If you think it is
> needed there for those kernels, please provide a working backport that
> we can apply.
>
> thanks,
>
> greg k-h
Powered by blists - more mailing lists