I'm about to reattempt the hacks to make tcp_sacktag_write_queue()
use the RB tree lookups.
In order to make the changes easier to review I'm trying to clean
up the function a little bit. This one pulls out the DSACK
detection logic.
I'm starting to pepper get_unaligned() calls around the sack
block accesses as I've been getting a few of these in my logs
on sparc64 lately:
[68089.285478] Kernel unaligned access at TPC[60e3c4]
tcp_sacktag_write_queue+0x40/0x86c
it's pretty easy to make it happen with NOP TCP options and
stuff like that, and we have get_unaligned() calls for other
TCP options already.
Pushed to net-2.6.22
commit d9367183d9d8fd1853e3bc4d0b1af077553e0e8a
Author: David S. Miller <[EMAIL PROTECTED]>
Date: Wed Mar 28 16:27:47 2007 -0700
[TCP]: Extract DSACK detection code from tcp_sacktag_write_queue().
Signed-off-by: David S. Miller <[EMAIL PROTECTED]>
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index c855791..a5a8987 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -936,6 +936,39 @@ static void tcp_update_reordering(struct sock *sk, const
int metric,
* Both of these heuristics are not used in Loss state, when we cannot
* account for retransmits accurately.
*/
+static int tcp_check_dsack(struct tcp_sock *tp, struct sk_buff *ack_skb,
+ struct tcp_sack_block_wire *sp, int num_sacks,
+ u32 prior_snd_una)
+{
+ u32 start_seq_0 = ntohl(get_unaligned(&sp[0].start_seq));
+ u32 end_seq_0 = ntohl(get_unaligned(&sp[0].end_seq));
+ int dup_sack = 0;
+
+ if (before(start_seq_0, TCP_SKB_CB(ack_skb)->ack_seq)) {
+ dup_sack = 1;
+ tp->rx_opt.sack_ok |= 4;
+ NET_INC_STATS_BH(LINUX_MIB_TCPDSACKRECV);
+ } else if (num_sacks > 1) {
+ u32 end_seq_1 = ntohl(get_unaligned(&sp[1].end_seq));
+ u32 start_seq_1 = ntohl(get_unaligned(&sp[1].start_seq));
+
+ if (!after(end_seq_0, end_seq_1) &&
+ !before(start_seq_0, start_seq_1)) {
+ dup_sack = 1;
+ tp->rx_opt.sack_ok |= 4;
+ NET_INC_STATS_BH(LINUX_MIB_TCPDSACKOFORECV);
+ }
+ }
+
+ /* D-SACK for already forgotten data... Do dumb counting. */
+ if (dup_sack &&
+ !after(end_seq_0, prior_snd_una) &&
+ after(end_seq_0, tp->undo_marker))
+ tp->undo_retrans--;
+
+ return dup_sack;
+}
+
static int
tcp_sacktag_write_queue(struct sock *sk, struct sk_buff *ack_skb,
u32 prior_snd_una, u32 *mark_lost_entry_seq)
@@ -963,25 +996,7 @@ tcp_sacktag_write_queue(struct sock *sk, struct sk_buff
*ack_skb,
*mark_lost_entry_seq = tp->highest_sack;
prior_fackets = tp->fackets_out;
- /* Check for D-SACK. */
- if (before(ntohl(sp[0].start_seq), TCP_SKB_CB(ack_skb)->ack_seq)) {
- dup_sack = 1;
- tp->rx_opt.sack_ok |= 4;
- NET_INC_STATS_BH(LINUX_MIB_TCPDSACKRECV);
- } else if (num_sacks > 1 &&
- !after(ntohl(sp[0].end_seq), ntohl(sp[1].end_seq)) &&
- !before(ntohl(sp[0].start_seq),
ntohl(sp[1].start_seq))) {
- dup_sack = 1;
- tp->rx_opt.sack_ok |= 4;
- NET_INC_STATS_BH(LINUX_MIB_TCPDSACKOFORECV);
- }
-
- /* D-SACK for already forgotten data...
- * Do dumb counting. */
- if (dup_sack &&
- !after(ntohl(sp[0].end_seq), prior_snd_una) &&
- after(ntohl(sp[0].end_seq), tp->undo_marker))
- tp->undo_retrans--;
+ dup_sack = tcp_check_dsack(tp, ack_skb, sp, num_sacks, prior_snd_una);
/* Eliminate too old ACKs, but take into
* account more or less fresh ones, they can
-
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
