From: Eric Dumazet <eric.dumazet@gmail.com>
To: "David S . Miller" <davem@davemloft.net>,
Jakub Kicinski <kuba@kernel.org>
Cc: David Ahern <dsahern@kernel.org>, netdev <netdev@vger.kernel.org>,
Eric Dumazet <edumazet@google.com>,
Eric Dumazet <eric.dumazet@gmail.com>
Subject: [PATCH v2 net 2/2] ipv4: add net_hash_mix() dispersion to fib_info_laddrhash keys
Date: Wed, 19 Jan 2022 02:04:13 -0800 [thread overview]
Message-ID: <20220119100413.4077866-3-eric.dumazet@gmail.com> (raw)
In-Reply-To: <20220119100413.4077866-1-eric.dumazet@gmail.com>
From: Eric Dumazet <edumazet@google.com>
net/ipv4/fib_semantics.c uses a hash table (fib_info_laddrhash)
in which fib_sync_down_addr() can locate fib_info
based on IPv4 local address.
This hash table is resized based on total number of
hashed fib_info, but the hash function is only
using the local address.
For hosts having many active network namespaces,
all fib_info for loopback devices (IPv4 address 127.0.0.1)
are hashed into a single bucket, making netns dismantles
very slow.
Signed-off-by: Eric Dumazet <edumazet@google.com>
---
net/ipv4/fib_semantics.c | 29 +++++++++++++++--------------
1 file changed, 15 insertions(+), 14 deletions(-)
diff --git a/net/ipv4/fib_semantics.c b/net/ipv4/fib_semantics.c
index 9813949da10493de36b9db797b6a5d94fd9bd3b1..b4589861b84c6bc4daa7149e078ad63749c7622f 100644
--- a/net/ipv4/fib_semantics.c
+++ b/net/ipv4/fib_semantics.c
@@ -52,6 +52,7 @@ static DEFINE_SPINLOCK(fib_info_lock);
static struct hlist_head *fib_info_hash;
static struct hlist_head *fib_info_laddrhash;
static unsigned int fib_info_hash_size;
+static unsigned int fib_info_hash_bits;
static unsigned int fib_info_cnt;
#define DEVINDEX_HASHBITS 8
@@ -1247,13 +1248,13 @@ int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
return err;
}
-static inline unsigned int fib_laddr_hashfn(__be32 val)
+static struct hlist_head *
+fib_info_laddrhash_bucket(const struct net *net, __be32 val)
{
- unsigned int mask = (fib_info_hash_size - 1);
+ u32 slot = hash_32(net_hash_mix(net) ^ (__force u32)val,
+ fib_info_hash_bits);
- return ((__force u32)val ^
- ((__force u32)val >> 7) ^
- ((__force u32)val >> 14)) & mask;
+ return &fib_info_laddrhash[slot];
}
static struct hlist_head *fib_info_hash_alloc(int bytes)
@@ -1289,6 +1290,7 @@ static void fib_info_hash_move(struct hlist_head *new_info_hash,
old_info_hash = fib_info_hash;
old_laddrhash = fib_info_laddrhash;
fib_info_hash_size = new_size;
+ fib_info_hash_bits = ilog2(new_size);
for (i = 0; i < old_size; i++) {
struct hlist_head *head = &fib_info_hash[i];
@@ -1306,21 +1308,20 @@ static void fib_info_hash_move(struct hlist_head *new_info_hash,
}
fib_info_hash = new_info_hash;
+ fib_info_laddrhash = new_laddrhash;
for (i = 0; i < old_size; i++) {
- struct hlist_head *lhead = &fib_info_laddrhash[i];
+ struct hlist_head *lhead = &old_laddrhash[i];
struct hlist_node *n;
struct fib_info *fi;
hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
struct hlist_head *ldest;
- unsigned int new_hash;
- new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
- ldest = &new_laddrhash[new_hash];
+ ldest = fib_info_laddrhash_bucket(fi->fib_net,
+ fi->fib_prefsrc);
hlist_add_head(&fi->fib_lhash, ldest);
}
}
- fib_info_laddrhash = new_laddrhash;
spin_unlock_bh(&fib_info_lock);
@@ -1605,7 +1606,7 @@ struct fib_info *fib_create_info(struct fib_config *cfg,
if (fi->fib_prefsrc) {
struct hlist_head *head;
- head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
+ head = fib_info_laddrhash_bucket(net, fi->fib_prefsrc);
hlist_add_head(&fi->fib_lhash, head);
}
if (fi->nh) {
@@ -1877,16 +1878,16 @@ int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
*/
int fib_sync_down_addr(struct net_device *dev, __be32 local)
{
- int ret = 0;
- unsigned int hash = fib_laddr_hashfn(local);
- struct hlist_head *head = &fib_info_laddrhash[hash];
int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
struct net *net = dev_net(dev);
+ struct hlist_head *head;
struct fib_info *fi;
+ int ret = 0;
if (!fib_info_laddrhash || local == 0)
return 0;
+ head = fib_info_laddrhash_bucket(net, local);
hlist_for_each_entry(fi, head, fib_lhash) {
if (!net_eq(fi->fib_net, net) ||
fi->fib_tb_id != tb_id)
--
2.34.1.703.g22d0c6ccf7-goog
next prev parent reply other threads:[~2022-01-19 10:04 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-01-19 10:04 [PATCH v2 net 0/2] ipv4: avoid pathological hash tables Eric Dumazet
2022-01-19 10:04 ` [PATCH v2 net 1/2] ipv4: avoid quadratic behavior in netns dismantle Eric Dumazet
2022-01-19 10:04 ` Eric Dumazet [this message]
2022-01-19 15:40 ` [PATCH v2 net 2/2] ipv4: add net_hash_mix() dispersion to fib_info_laddrhash keys David Ahern
2022-01-19 18:24 ` [PATCH v2 net 0/2] ipv4: avoid pathological hash tables Jakub Kicinski
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20220119100413.4077866-3-eric.dumazet@gmail.com \
--to=eric.dumazet@gmail.com \
--cc=davem@davemloft.net \
--cc=dsahern@kernel.org \
--cc=edumazet@google.com \
--cc=kuba@kernel.org \
--cc=netdev@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.