Subject: inet: use bigger hash table for IP ID generation (backported to 4.19) From: Amit Klein aksecurity@gmail.com

This is a backport to 4.19 of the following patch, originally developed by Eric Dumazet.

In commit 73f156a6e8c1 ("inetpeer: get rid of ip_id_count") I used a very small hash table that could be abused by patient attackers to reveal sensitive information.

Switch to a dynamic sizing, depending on RAM size.

Typical big hosts will now use 128x more storage (2 MB) to get a similar increase in security and reduction of hash collisions.

As a bonus, use of alloc_large_system_hash() spreads allocated memory among all NUMA nodes.

Fixes: 73f156a6e8c1 ("inetpeer: get rid of ip_id_count") Reported-by: Amit Klein aksecurity@gmail.com Cc: stable@vger.kernel.org Cc: Eric Dumazet edumazet@google.com Cc: Willy Tarreau w@1wt.eu --- net/ipv4/route.c | 42 ++++++++++++++++++++++++++++-------------- 1 file changed, 28 insertions(+), 14 deletions(-)

(limited to 'net/ipv4/route.c')

diff --git a/net/ipv4/route.c b/net/ipv4/route.c index 0470442ff61d6..ea916df1bbf5e 100644 --- a/net/ipv4/route.c +++ b/net/ipv4/route.c @@ -66,6 +66,7 @@ #include <linux/types.h> #include <linux/kernel.h> #include <linux/mm.h> +#include <linux/bootmem.h> #include <linux/string.h> #include <linux/socket.h> #include <linux/sockios.h> @@ -452,8 +453,10 @@ 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;

@@ -463,12 +466,16 @@ 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 = READ_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);

@@ -3557,18 +3564,25 @@ struct ip_rt_acct __percpu *ip_rt_acct __read_mostly;

int __init ip_rt_init(void) { + void *idents_hash; int cpu;

- ip_idents = kmalloc_array(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);