lkml.org 
[lkml]   [2021]   [Jun]   [16]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    Patch in this message
    /
    From
    Date
    SubjectRe: [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@linuxfoundation.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

    \
     
     \ /
      Last update: 2021-06-16 11:18    [W:4.200 / U:0.008 seconds]
    ©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site