queue-6.1/tcp-fix-bind-regression-for-v6-only-wildcard-and-v4-.patch

   1 From 961f27c6b980bd03433c02a3b405cc917c093f0f Mon Sep 17 00:00:00 2001
   2 From: Sasha Levin <sashal@kernel.org>
   3 Date: Tue, 26 Mar 2024 13:42:45 -0700
   4 Subject: tcp: Fix bind() regression for v6-only wildcard and v4(-mapped-v6)
   5  non-wildcard addresses.
   6
   7 From: Kuniyuki Iwashima <kuniyu@amazon.com>
   8
   9 [ Upstream commit d91ef1e1b55f730bee8ce286b02b7bdccbc42973 ]
  10
  11 Jianguo Wu reported another bind() regression introduced by bhash2.
  12
  13 Calling bind() for the following 3 addresses on the same port, the
  14 3rd one should fail but now succeeds.
  15
  16   1. 0.0.0.0 or ::ffff:0.0.0.0
  17   2. [::] w/ IPV6_V6ONLY
  18   3. IPv4 non-wildcard address or v4-mapped-v6 non-wildcard address
  19
  20 The first two bind() create tb2 like this:
  21
  22   bhash2 -> tb2(:: w/ IPV6_V6ONLY) -> tb2(0.0.0.0)
  23
  24 The 3rd bind() will match with the IPv6 only wildcard address bucket
  25 in inet_bind2_bucket_match_addr_any(), however, no conflicting socket
  26 exists in the bucket.  So, inet_bhash2_conflict() will returns false,
  27 and thus, inet_bhash2_addr_any_conflict() returns false consequently.
  28
  29 As a result, the 3rd bind() bypasses conflict check, which should be
  30 done against the IPv4 wildcard address bucket.
  31
  32 So, in inet_bhash2_addr_any_conflict(), we must iterate over all buckets.
  33
  34 Note that we cannot add ipv6_only flag for inet_bind2_bucket as it
  35 would confuse the following patetrn.
  36
  37   1. [::] w/ SO_REUSE{ADDR,PORT} and IPV6_V6ONLY
  38   2. [::] w/ SO_REUSE{ADDR,PORT}
  39   3. IPv4 non-wildcard address or v4-mapped-v6 non-wildcard address
  40
  41 The first bind() would create a bucket with ipv6_only flag true,
  42 the second bind() would add the [::] socket into the same bucket,
  43 and the third bind() could succeed based on the wrong assumption
  44 that ipv6_only bucket would not conflict with v4(-mapped-v6) address.
  45
  46 Fixes: 28044fc1d495 ("net: Add a bhash2 table hashed by port and address")
  47 Diagnosed-by: Jianguo Wu <wujianguo106@163.com>
  48 Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
  49 Link: https://lore.kernel.org/r/20240326204251.51301-3-kuniyu@amazon.com
  50 Signed-off-by: Jakub Kicinski <kuba@kernel.org>
  51 Signed-off-by: Sasha Levin <sashal@kernel.org>
  52 ---
  53  net/ipv4/inet_connection_sock.c | 19 +++++++++++--------
  54  1 file changed, 11 insertions(+), 8 deletions(-)
  55
  56 diff --git a/net/ipv4/inet_connection_sock.c b/net/ipv4/inet_connection_sock.c
  57 index f7832d4253820..8407098a59391 100644
  58 --- a/net/ipv4/inet_connection_sock.c
  59 +++ b/net/ipv4/inet_connection_sock.c
  60 @@ -289,6 +289,7 @@ static bool inet_bhash2_addr_any_conflict(const struct sock *sk, int port, int l
  61         struct sock_reuseport *reuseport_cb;
  62         struct inet_bind_hashbucket *head2;
  63         struct inet_bind2_bucket *tb2;
  64 +       bool conflict = false;
  65         bool reuseport_cb_ok;
  66
  67         rcu_read_lock();
  68 @@ -301,18 +302,20 @@ static bool inet_bhash2_addr_any_conflict(const struct sock *sk, int port, int l
  69
  70         spin_lock(&head2->lock);
  71
  72 -       inet_bind_bucket_for_each(tb2, &head2->chain)
  73 -               if (inet_bind2_bucket_match_addr_any(tb2, net, port, l3mdev, sk))
  74 -                       break;
  75 +       inet_bind_bucket_for_each(tb2, &head2->chain) {
  76 +               if (!inet_bind2_bucket_match_addr_any(tb2, net, port, l3mdev, sk))
  77 +                       continue;
  78
  79 -       if (tb2 && inet_bhash2_conflict(sk, tb2, uid, relax, reuseport_cb_ok,
  80 -                                       reuseport_ok)) {
  81 -               spin_unlock(&head2->lock);
  82 -               return true;
  83 +               if (!inet_bhash2_conflict(sk, tb2, uid, relax, reuseport_cb_ok, reuseport_ok))
  84 +                       continue;
  85 +
  86 +               conflict = true;
  87 +               break;
  88         }
  89
  90         spin_unlock(&head2->lock);
  91 -       return false;
  92 +
  93 +       return conflict;
  94  }
  95
  96  /*
  97 --
  98 2.43.0
  99