于 2015年03月01日 07:20, John Heffner 写道:

quoted

diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index a2a796c..c418829 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c

@@ -1837,11 +1837,13 @@ static int tcp_mtu_probe(struct sock *sk)
         struct tcp_sock *tp = tcp_sk(sk);
         struct inet_connection_sock *icsk = inet_csk(sk);
         struct sk_buff *skb, *nskb, *next;
+       struct net *net = sock_net(sk);
         int len;
         int probe_size;
         int size_needed;
         int copy;
         int mss_now;
+       int interval;

         /* Not currently probing/verifying,
          * not in recovery,

@@ -1854,11 +1856,17 @@ static int tcp_mtu_probe(struct sock *sk)
             tp->rx_opt.num_sacks || tp->rx_opt.dsack)
                 return -1;

-       /* Very simple search strategy: just double the MSS. */
+       /* Use binary search for probe_size bewteen tcp_mss_base,
+        * and current mss_clamp. if (search_high - search_low)
+        * smaller than a threshold, backoff from probing.
+        */
         mss_now = tcp_current_mss(sk);
-       probe_size = 2 * tp->mss_cache;
+       probe_size = (icsk->icsk_mtup.search_high +
+                     icsk->icsk_mtup.search_low) >> 1;
         size_needed = probe_size + (tp->reordering + 1) * tp->mss_cache;
-       if (probe_size > tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_high)) {
+       interval = icsk->icsk_mtup.search_high - icsk->icsk_mtup.search_low;
+       if (probe_size > tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_high) ||
+           interval < net->ipv4.sysctl_tcp_probe_threshold) {
                 /* TODO: set timer for probe_converge_event */
                 return -1;
         }

A couple things: the local variable probe_size here is TCP segment
size, while search_low and search_high are IP datagram sizes.  Use
tcp_mtu_to_mss to subtract headers.  Also, I think if you set
sysctl_tcp_probe_threshold <= 0, this will keep probing indefinitely
at search_low (not useful).  You probably want to test interval <
max(1, sysctl_tcp_probe_threshold).

Thanks for the comments, will update in next version.
btw, I'm little confused about two points here:

1. Checking if there is enough user data available in write queue with size_needed,
    Is there any special consideration here involving (tp->reordering + 1) * tp->mss_cache ?
    Since the assembly always copies "probe_size" bytes data from the write queue.

    size_needed = probe_size + (tp->reordering + 1) * tp->mss_cache;

2. Traverse write queue to build probing packet with "probe_size" bytes segment,
    when will the final nskb len exceeding probe_size? which trigger the break in line:1971

1971                 if (len >= probe_size)
1972                         break;
1973         }
1974         tcp_init_tso_segs(sk, nskb, nskb->len);

   -John