patch-2.4.0-test7 linux/net/ipv4/tcp_input.c
Next file: linux/net/ipv4/tcp_ipv4.c
Previous file: linux/net/ipv4/tcp.c
Back to the patch index
Back to the overall index
- Lines: 4942
- Date:
Fri Aug 18 10:26:25 2000
- Orig file:
v2.4.0-test6/linux/net/ipv4/tcp_input.c
- Orig date:
Wed Apr 26 16:34:09 2000
diff -u --recursive --new-file v2.4.0-test6/linux/net/ipv4/tcp_input.c linux/net/ipv4/tcp_input.c
@@ -5,7 +5,7 @@
*
* Implementation of the Transmission Control Protocol(TCP).
*
- * Version: $Id: tcp_input.c,v 1.193 2000/04/20 14:41:16 davem Exp $
+ * Version: $Id: tcp_input.c,v 1.198 2000/08/15 20:15:23 davem Exp $
*
* Authors: Ross Biro, <bir7@leland.Stanford.Edu>
* Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
@@ -55,20 +55,15 @@
* work without delayed acks.
* Andi Kleen: Process packets with PSH set in the
* fast path.
+ * J Hadi Salim: ECN support
*/
-#include <linux/config.h>
#include <linux/mm.h>
#include <linux/sysctl.h>
#include <net/tcp.h>
#include <net/inet_common.h>
#include <linux/ipsec.h>
-#ifdef CONFIG_SYSCTL
-#define SYNC_INIT 0 /* let the user enable it */
-#else
-#define SYNC_INIT 1
-#endif
/* These are on by default so the code paths get tested.
* For the final 2.2 this may be undone at our discretion. -DaveM
@@ -76,33 +71,39 @@
int sysctl_tcp_timestamps = 1;
int sysctl_tcp_window_scaling = 1;
int sysctl_tcp_sack = 1;
+int sysctl_tcp_fack = 1;
+int sysctl_tcp_reordering = TCP_FASTRETRANS_THRESH;
+int sysctl_tcp_ecn = 1;
+int sysctl_tcp_dsack = 1;
+int sysctl_tcp_app_win = 31;
+int sysctl_tcp_adv_win_scale = 2;
-int sysctl_tcp_syncookies = SYNC_INIT;
-int sysctl_tcp_stdurg;
-int sysctl_tcp_rfc1337;
-int sysctl_tcp_tw_recycle = 1;
-int sysctl_tcp_abort_on_overflow = 0;
+int sysctl_tcp_stdurg = 0;
+int sysctl_tcp_rfc1337 = 0;
int sysctl_tcp_max_orphans = NR_FILE;
-int sysctl_tcp_max_tw_buckets = NR_FILE*2;
-static int prune_queue(struct sock *sk);
+#define FLAG_DATA 0x01 /* Incoming frame contained data. */
+#define FLAG_WIN_UPDATE 0x02 /* Incoming ACK was a window update. */
+#define FLAG_DATA_ACKED 0x04 /* This ACK acknowledged new data. */
+#define FLAG_RETRANS_DATA_ACKED 0x08 /* "" "" some of which was retransmitted. */
+#define FLAG_SYN_ACKED 0x10 /* This ACK acknowledged SYN. */
+#define FLAG_DATA_SACKED 0x20 /* New SACK. */
+#define FLAG_ECE 0x40 /* ECE in this ACK */
+#define FLAG_DATA_LOST 0x80 /* SACK detected data lossage. */
+#define FLAG_SLOWPATH 0x100 /* Do not skip RFC checks for window update.*/
+
+#define FLAG_ACKED (FLAG_DATA_ACKED|FLAG_SYN_ACKED)
+#define FLAG_NOT_DUP (FLAG_DATA|FLAG_WIN_UPDATE|FLAG_ACKED)
+#define FLAG_CA_ALERT (FLAG_DATA_SACKED|FLAG_ECE)
+#define FLAG_FORWARD_PROGRESS (FLAG_ACKED|FLAG_DATA_SACKED)
+
+#define IsReno(tp) ((tp)->sack_ok == 0)
+#define IsFack(tp) ((tp)->sack_ok & 2)
+
+#define TCP_REMNANT (TCP_FLAG_FIN|TCP_FLAG_URG|TCP_FLAG_SYN|TCP_FLAG_PSH)
-/*
- * Adapt the MSS value used to make delayed ack decision to the
+/* Adapt the MSS value used to make delayed ack decision to the
* real world.
- *
- * The constant 536 hasn't any good meaning. In IPv4 world
- * MTU may be smaller, though it contradicts to RFC1122, which
- * states that MSS must be at least 536.
- * We use the constant to do not ACK each second
- * packet in a stream of tiny size packets.
- * It means that super-low mtu links will be aggressively delacked.
- * Seems, it is even good. If they have so low mtu, they are weirdly
- * slow.
- *
- * AK: BTW it may be useful to add an option to lock the rcv_mss.
- * this way the beowulf people wouldn't need ugly patches to get the
- * ack frequencies they want and it would be an elegant way to tune delack.
*/
static __inline__ void tcp_measure_rcv_mss(struct tcp_opt *tp, struct sk_buff *skb)
{
@@ -117,6 +118,9 @@
len = skb->len;
if (len >= tp->ack.rcv_mss) {
tp->ack.rcv_mss = len;
+ /* Dubious? Rather, it is final cut. 8) */
+ if (tcp_flag_word(skb->h.th)&TCP_REMNANT)
+ tp->ack.pending |= TCP_ACK_PUSHED;
} else {
/* Otherwise, we make more careful check taking into account,
* that SACKs block is variable.
@@ -124,37 +128,47 @@
* "len" is invariant segment length, including TCP header.
*/
len = skb->tail - skb->h.raw;
- if (len >= TCP_MIN_RCVMSS + sizeof(struct tcphdr)) {
+ if (len >= TCP_MIN_RCVMSS + sizeof(struct tcphdr) ||
+ /* If PSH is not set, packet should be
+ * full sized, provided peer TCP is not badly broken.
+ * This observation (if it is correct 8)) allows
+ * to handle super-low mtu links fairly.
+ */
+ (len >= TCP_MIN_MSS + sizeof(struct tcphdr) &&
+ !(tcp_flag_word(skb->h.th)&TCP_REMNANT))) {
/* Subtract also invariant (if peer is RFC compliant),
* tcp header plus fixed timestamp option length.
* Resulting "len" is MSS free of SACK jitter.
*/
len -= tp->tcp_header_len;
- if (len == lss)
- tp->ack.rcv_mss = len;
tp->ack.last_seg_size = len;
+ if (len == lss) {
+ tp->ack.rcv_mss = len;
+ return;
+ }
}
+ tp->ack.pending |= TCP_ACK_PUSHED;
}
}
-
-static __inline__ void tcp_enter_quickack_mode(struct tcp_opt *tp)
+static void tcp_incr_quickack(struct tcp_opt *tp)
{
- unsigned quickacks = tcp_receive_window(tp)/(2*tp->ack.rcv_mss);
+ unsigned quickacks = tp->rcv_wnd/(2*tp->ack.rcv_mss);
- tp->ack.quick = max(min(quickacks, 127), 1);
+ if (quickacks==0)
+ quickacks=2;
+ if (quickacks > tp->ack.quick)
+ tp->ack.quick = min(quickacks, TCP_MAX_QUICKACKS);
+}
- if (!tp->tstamp_ok && tp->ack.quick>2) {
- /* Quick ACKs are _dangerous_, if RTTM is not used.
- * See comment in tcp_init_metrics(). We still help
- * them to overcome the most difficult, initial
- * phase of slow start.
- */
- tp->ack.quick = 2;
- }
+void tcp_enter_quickack_mode(struct tcp_opt *tp)
+{
+ tcp_incr_quickack(tp);
+ tp->ack.pingpong = 0;
+ tp->ack.ato = TCP_ATO_MIN;
}
-/* Send ACKs quickly, if "quick" count is not ehausted
+/* Send ACKs quickly, if "quick" count is not exhausted
* and the session is not interactive.
*/
@@ -163,6 +177,173 @@
return (tp->ack.quick && !tp->ack.pingpong);
}
+/* Buffer size and advertised window tuning.
+ *
+ * 1. Tuning sk->sndbuf, when connection enters established state.
+ */
+
+static void tcp_fixup_sndbuf(struct sock *sk)
+{
+ struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
+ int sndmem = tp->mss_clamp+MAX_TCP_HEADER+16+sizeof(struct sk_buff);
+
+ if (sk->sndbuf < 3*sndmem)
+ sk->sndbuf = min(3*sndmem, sysctl_tcp_wmem[2]);
+}
+
+/* 2. Tuning advertised window (window_clamp, rcv_ssthresh)
+ *
+ * All tcp_full_space() is split to two parts: "network" buffer, allocated
+ * forward and advertised in receiver window (tp->rcv_wnd) and
+ * "application buffer", required to isolate scheduling/application
+ * latencies from network.
+ * window_clamp is maximal advertised window. It can be less than
+ * tcp_full_space(), in this case tcp_full_space() - window_clamp
+ * is reserved for "application" buffer. The less window_clamp is
+ * the smoother our behaviour from viewpoint of network, but the lower
+ * throughput and the higher sensitivity of the connection to losses. 8)
+ *
+ * rcv_ssthresh is more strict window_clamp used at "slow start"
+ * phase to predict further behaviour of this connection.
+ * It is used for two goals:
+ * - to enforce header prediction at sender, even when application
+ * requires some significant "application buffer". It is check #1.
+ * - to prevent pruning of receive queue because of misprediction
+ * of receiver window. Check #2.
+ *
+ * The scheme does not work when sender sends good segments opening
+ * window and then starts to feed us spagetti. But it should work
+ * in common situations. Otherwise, we have to rely on queue collapsing.
+ */
+
+/* Slow part of check#2. */
+static int
+__tcp_grow_window(struct sock *sk, struct tcp_opt *tp, struct sk_buff *skb)
+{
+ /* Optimize this! */
+ int truesize = tcp_win_from_space(skb->truesize)/2;
+ int window = tcp_full_space(sk)/2;
+
+ while (tp->rcv_ssthresh <= window) {
+ if (truesize <= skb->len)
+ return 2*tp->ack.rcv_mss;
+
+ truesize >>= 1;
+ window >>= 1;
+ }
+ return 0;
+}
+
+static __inline__ void
+tcp_grow_window(struct sock *sk, struct tcp_opt *tp, struct sk_buff *skb)
+{
+ /* Check #1 */
+ if (tp->rcv_ssthresh < tp->window_clamp &&
+ (int)tp->rcv_ssthresh < tcp_space(sk) &&
+ !tcp_memory_pressure) {
+ int incr;
+
+ /* Check #2. Increase window, if skb with such overhead
+ * will fit to rcvbuf in future.
+ */
+ if (tcp_win_from_space(skb->truesize) <= skb->len)
+ incr = 2*tp->advmss;
+ else
+ incr = __tcp_grow_window(sk, tp, skb);
+
+ if (incr) {
+ tp->rcv_ssthresh = min(tp->rcv_ssthresh + incr, tp->window_clamp);
+ tp->ack.quick |= 1;
+ }
+ }
+}
+
+/* 3. Tuning rcvbuf, when connection enters established state. */
+
+static void tcp_fixup_rcvbuf(struct sock *sk)
+{
+ struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
+ int rcvmem = tp->advmss+MAX_TCP_HEADER+16+sizeof(struct sk_buff);
+
+ /* Try to select rcvbuf so that 4 mss-sized segments
+ * will fit to window and correspoding skbs will fit to our rcvbuf.
+ * (was 3; 4 is minimum to allow fast retransmit to work.)
+ */
+ while (tcp_win_from_space(rcvmem) < tp->advmss)
+ rcvmem += 128;
+ if (sk->rcvbuf < 4*rcvmem)
+ sk->rcvbuf = min(4*rcvmem, sysctl_tcp_rmem[2]);
+}
+
+/* 4. Try to fixup all. It is made iimediately after connection enters
+ * established state.
+ */
+static void tcp_init_buffer_space(struct sock *sk)
+{
+ struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
+ int maxwin;
+
+ if (!(sk->userlocks&SOCK_RCVBUF_LOCK))
+ tcp_fixup_rcvbuf(sk);
+ if (!(sk->userlocks&SOCK_SNDBUF_LOCK))
+ tcp_fixup_sndbuf(sk);
+
+ maxwin = tcp_full_space(sk);
+
+ if (tp->window_clamp >= maxwin) {
+ tp->window_clamp = maxwin;
+
+ if (sysctl_tcp_app_win && maxwin>4*tp->advmss)
+ tp->window_clamp = max(maxwin-(maxwin>>sysctl_tcp_app_win), 4*tp->advmss);
+ }
+
+ /* Force reservation of one segment. */
+ if (sysctl_tcp_app_win &&
+ tp->window_clamp > 2*tp->advmss &&
+ tp->window_clamp + tp->advmss > maxwin)
+ tp->window_clamp = max(2*tp->advmss, maxwin-tp->advmss);
+
+ tp->rcv_ssthresh = min(tp->rcv_ssthresh, tp->window_clamp);
+ tp->snd_cwnd_stamp = tcp_time_stamp;
+}
+
+/* 5. Recalculate window clamp after socket hit its memory bounds. */
+static void tcp_clamp_window(struct sock *sk, struct tcp_opt *tp)
+{
+ struct sk_buff *skb;
+ int app_win = tp->rcv_nxt - tp->copied_seq;
+ int ofo_win = 0;
+
+ tp->ack.quick = 0;
+
+ skb_queue_walk(&tp->out_of_order_queue, skb) {
+ ofo_win += skb->len;
+ }
+
+ /* If overcommit is due to out of order segments,
+ * do not clamp window. Try to expand rcvbuf instead.
+ */
+ if (ofo_win) {
+ if (sk->rcvbuf < sysctl_tcp_rmem[2] &&
+ !(sk->userlocks&SOCK_RCVBUF_LOCK) &&
+ !tcp_memory_pressure &&
+ atomic_read(&tcp_memory_allocated) < sysctl_tcp_mem[0])
+ sk->rcvbuf = min(atomic_read(&sk->rmem_alloc), sysctl_tcp_rmem[2]);
+ }
+ if (atomic_read(&sk->rmem_alloc) > sk->rcvbuf) {
+ app_win += ofo_win;
+ if (atomic_read(&sk->rmem_alloc) >= 2*sk->rcvbuf)
+ app_win >>= 1;
+ if (app_win > tp->ack.rcv_mss)
+ app_win -= tp->ack.rcv_mss;
+ app_win = max(app_win, 2*tp->advmss);
+
+ if (!ofo_win)
+ tp->window_clamp = min(tp->window_clamp, app_win);
+ tp->rcv_ssthresh = min(tp->window_clamp, 2*tp->advmss);
+ }
+}
+
/* There is something which you must keep in mind when you analyze the
* behavior of the tp->ato delayed ack timeout interval. When a
* connection starts up, we want to ack as quickly as possible. The
@@ -173,14 +354,13 @@
* each ACK we send, he increments snd_cwnd and transmits more of his
* queue. -DaveM
*/
-static void tcp_event_data_recv(struct tcp_opt *tp, struct sk_buff *skb)
+static void tcp_event_data_recv(struct sock *sk, struct tcp_opt *tp, struct sk_buff *skb)
{
u32 now;
- tcp_measure_rcv_mss(tp, skb);
+ tcp_schedule_ack(tp);
- tp->ack.pending = 1;
- tp->ack.rcv_segs++;
+ tcp_measure_rcv_mss(tp, skb);
now = tcp_time_stamp;
@@ -188,37 +368,31 @@
/* The _first_ data packet received, initialize
* delayed ACK engine.
*/
-
- /* Help sender leave slow start quickly. */
tcp_enter_quickack_mode(tp);
-
- /* Pingpong is off, session is not interactive by default */
- tp->ack.pingpong = 0;
-
- /* ATO is minimal */
- tp->ack.ato = TCP_ATO_MIN;
} else {
int m = now - tp->ack.lrcvtime;
- if (m > TCP_ATO_MAX/2) {
- /* Do not touch ATO, if interval is out of bounds.
- * It will be deflated by delack timer, if our peer
- * really sends too rarely.
+ if (m <= TCP_ATO_MIN/2) {
+ /* The fastest case is the first. */
+ tp->ack.ato = (tp->ack.ato>>1) + TCP_ATO_MIN/2;
+ } else if (m < tp->ack.ato) {
+ tp->ack.ato = (tp->ack.ato>>1) + m;
+ if (tp->ack.ato > tp->rto)
+ tp->ack.ato = tp->rto;
+ } else if (m > tp->rto) {
+ /* Too long gap. Apparently sender falled to
+ * restart window, so that we send ACKs quickly.
*/
- if (m > tp->rto) {
- /* Too long gap. Apparently sender falled to
- * restart window, so that we send ACKs quickly.
- */
- tcp_enter_quickack_mode(tp);
- }
- } else {
- if (m <= 0)
- m = TCP_ATO_MIN/2;
- if (m <= tp->ack.ato)
- tp->ack.ato = (tp->ack.ato >> 1) + m;
+ tcp_incr_quickack(tp);
+ tcp_mem_reclaim(sk);
}
}
tp->ack.lrcvtime = now;
+
+ TCP_ECN_check_ce(tp, skb);
+
+ if (skb->len >= 128)
+ tcp_grow_window(sk, tp, skb);
}
/* Called to compute a smoothed rtt estimate. The data fed to this
@@ -230,7 +404,6 @@
* To save cycles in the RFC 1323 implementation it was better to break
* it up into three procedures. -- erics
*/
-
static __inline__ void tcp_rtt_estimator(struct tcp_opt *tp, __u32 mrtt)
{
long m = mrtt; /* RTT */
@@ -243,6 +416,13 @@
*
* On a 1990 paper the rto value is changed to:
* RTO = rtt + 4 * mdev
+ *
+ * Funny. This algorithm seems to be very broken.
+ * These formulae increase RTO, when it should be decreased, increase
+ * too slowly, when it should be incresed fastly, decrease too fastly
+ * etc. I guess in BSD RTO takes ONE value, so that it is absolutely
+ * does not matter how to _calculate_ it. Seems, it was trap
+ * that VJ failed to avoid. 8)
*/
if(m == 0)
m = 1;
@@ -263,16 +443,27 @@
/* Calculate rto without backoff. This is the second half of Van Jacobson's
* routine referred to above.
*/
-
static __inline__ void tcp_set_rto(struct tcp_opt *tp)
{
tp->rto = (tp->srtt >> 3) + tp->mdev;
/* I am not enough educated to understand this magic.
* However, it smells bad. snd_cwnd>31 is common case.
*/
+ /* OK, I found comment in 2.0 source tree, it deserves
+ * to be reproduced:
+ * ====
+ * Note: Jacobson's algorithm is fine on BSD which has a 1/2 second
+ * granularity clock, but with our 1/100 second granularity clock we
+ * become too sensitive to minor changes in the round trip time.
+ * We add in two compensating factors. First we multiply by 5/4.
+ * For large congestion windows this allows us to tolerate burst
+ * traffic delaying up to 1/4 of our packets. We also add in
+ * a rtt / cong_window term. For small congestion windows this allows
+ * a single packet delay, but has negligible effect
+ * on the compensation for large windows.
+ */
tp->rto += (tp->rto >> 2) + (tp->rto >> (tp->snd_cwnd-1));
}
-
/* Keep the rto between HZ/5 and 120*HZ. 120*HZ is the upper bound
* on packet lifetime in the internet. We need the HZ/5 lower
@@ -292,11 +483,12 @@
tp->rto = TCP_RTO_MAX;
}
+
/* Save metrics learned by this TCP session.
This function is called only, when TCP finishes sucessfully
i.e. when it enters TIME-WAIT or goes from LAST-ACK to CLOSE.
*/
-static void tcp_update_metrics(struct sock *sk)
+void tcp_update_metrics(struct sock *sk)
{
struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
struct dst_entry *dst = __sk_dst_get(sk);
@@ -344,19 +536,20 @@
dst->rttvar -= (dst->rttvar - m)>>2;
}
- if (tp->snd_ssthresh == 0x7FFFFFFF) {
+ if (tp->snd_ssthresh >= 0xFFFF) {
/* Slow start still did not finish. */
if (dst->ssthresh &&
!(dst->mxlock&(1<<RTAX_SSTHRESH)) &&
- tp->snd_cwnd > dst->ssthresh)
- dst->ssthresh = tp->snd_cwnd;
+ (tp->snd_cwnd>>1) > dst->ssthresh)
+ dst->ssthresh = (tp->snd_cwnd>>1);
if (!(dst->mxlock&(1<<RTAX_CWND)) &&
tp->snd_cwnd > dst->cwnd)
dst->cwnd = tp->snd_cwnd;
- } else if (tp->snd_cwnd >= tp->snd_ssthresh && !tp->high_seq) {
+ } else if (tp->snd_cwnd > tp->snd_ssthresh &&
+ tp->ca_state == TCP_CA_Open) {
/* Cong. avoidance phase, cwnd is reliable. */
if (!(dst->mxlock&(1<<RTAX_SSTHRESH)))
- dst->ssthresh = tp->snd_cwnd;
+ dst->ssthresh = max(tp->snd_cwnd>>1, tp->snd_ssthresh);
if (!(dst->mxlock&(1<<RTAX_CWND)))
dst->cwnd = (dst->cwnd + tp->snd_cwnd)>>1;
} else {
@@ -370,9 +563,37 @@
tp->snd_ssthresh > dst->ssthresh)
dst->ssthresh = tp->snd_ssthresh;
}
+
+ if (!(dst->mxlock&(1<<RTAX_REORDERING))) {
+ if (dst->reordering < tp->reordering &&
+ tp->reordering != sysctl_tcp_reordering)
+ dst->reordering = tp->reordering;
+ }
}
}
+/* Increase initial CWND conservatively: if estimated
+ * RTT is low enough (<20msec) or if we have some preset ssthresh.
+ *
+ * Numbers are taken from RFC1414.
+ */
+__u32 tcp_init_cwnd(struct tcp_opt *tp)
+{
+ __u32 cwnd;
+
+ if (tp->mss_cache > 1460)
+ return 2;
+
+ cwnd = (tp->mss_cache > 1095) ? 3 : 4;
+
+ if (!tp->srtt || (tp->snd_ssthresh >= 0xFFFF && tp->srtt > ((HZ/50)<<3)))
+ cwnd = 2;
+ else if (cwnd > tp->snd_ssthresh)
+ cwnd = tp->snd_ssthresh;
+
+ return min(cwnd, tp->snd_cwnd_clamp);
+}
+
/* Initialize metrics on socket. */
static void tcp_init_metrics(struct sock *sk)
@@ -392,6 +613,10 @@
if (tp->snd_ssthresh > tp->snd_cwnd_clamp)
tp->snd_ssthresh = tp->snd_cwnd_clamp;
}
+ if (dst->reordering && tp->reordering != dst->reordering) {
+ tp->sack_ok &= ~2;
+ tp->reordering = dst->reordering;
+ }
if (dst->rtt == 0)
goto reset;
@@ -422,9 +647,9 @@
if (tp->rto < TCP_TIMEOUT_INIT && !tp->saw_tstamp)
goto reset;
tp->snd_cwnd = tcp_init_cwnd(tp);
+ tp->snd_cwnd_stamp = tcp_time_stamp;
return;
-
reset:
/* Play conservative. If timestamps are not
* supported, TCP will fail to recalculate correct
@@ -437,402 +662,967 @@
}
}
-/* WARNING: this must not be called if tp->saw_tstamp was false. */
-extern __inline__ void
-tcp_replace_ts_recent(struct sock *sk, struct tcp_opt *tp, u32 seq)
+static void tcp_update_reordering(struct tcp_opt *tp, int metric, int ts)
{
- if (!after(seq, tp->rcv_wup)) {
- /* PAWS bug workaround wrt. ACK frames, the PAWS discard
- * extra check below makes sure this can only happen
- * for pure ACK frames. -DaveM
- *
- * Not only, also it occurs for expired timestamps
- * and RSTs with bad timestamp option. --ANK
- */
+ if (metric > tp->reordering) {
+ tp->reordering = min(TCP_MAX_REORDERING, metric);
- if((s32)(tp->rcv_tsval - tp->ts_recent) >= 0 ||
- xtime.tv_sec >= tp->ts_recent_stamp + TCP_PAWS_24DAYS) {
- tp->ts_recent = tp->rcv_tsval;
- tp->ts_recent_stamp = xtime.tv_sec;
- }
+ /* This exciting event is worth to be remembered. 8) */
+ if (ts)
+ NET_INC_STATS_BH(TCPTSReorder);
+ else if (IsReno(tp))
+ NET_INC_STATS_BH(TCPRenoReorder);
+ else if (IsFack(tp))
+ NET_INC_STATS_BH(TCPFACKReorder);
+ else
+ NET_INC_STATS_BH(TCPSACKReorder);
+#if FASTRETRANS_DEBUG > 1
+ printk(KERN_DEBUG "Disorder%d %d %u f%u s%u rr%d\n",
+ tp->sack_ok, tp->ca_state,
+ tp->reordering, tp->fackets_out, tp->sacked_out,
+ tp->undo_marker ? tp->undo_retrans : 0);
+#endif
+ /* Disable FACK yet. */
+ tp->sack_ok &= ~2;
}
}
-extern __inline__ int tcp_paws_discard(struct tcp_opt *tp, struct sk_buff *skb)
-{
- return ((s32)(tp->rcv_tsval - tp->ts_recent) < 0 &&
- xtime.tv_sec < tp->ts_recent_stamp + TCP_PAWS_24DAYS
+/* This procedure tags the retransmission queue when SACKs arrive.
+ *
+ * We have three tag bits: SACKED(S), RETRANS(R) and LOST(L).
+ * Packets in queue with these bits set are counted in variables
+ * sacked_out, retrans_out and lost_out, correspondingly.
+ *
+ * Valid combinations are:
+ * Tag InFlight Description
+ * 0 1 - orig segment is in flight.
+ * S 0 - nothing flies, orig reached receiver.
+ * L 0 - nothing flies, orig lost by net.
+ * R 2 - both orig and retransmit are in flight.
+ * L|R 1 - orig is lost, retransmit is in flight.
+ * S|R 1 - orig reached receiver, retrans is still in flight.
+ * (L|S|R is logically valid, it could occur when L|R is sacked,
+ * but it is equivalent to plain S and code short-curcuits it to S.
+ * L|S is logically invalid, it would mean -1 packet in flight 8))
+ *
+ * These 6 states form finite state machine, controlled by the following events:
+ * 1. New ACK (+SACK) arrives. (tcp_sacktag_write_queue())
+ * 2. Retransmission. (tcp_retransmit_skb(), tcp_xmit_retransmit_queue())
+ * 3. Loss detection event of one of three flavors:
+ * A. Scoreboard estimator decided the packet is lost.
+ * A'. Reno "three dupacks" marks head of queue lost.
+ * A''. Its FACK modfication, head until snd.fack is lost.
+ * B. SACK arrives sacking data transmitted after never retransmitted
+ * hole was sent out.
+ * C. SACK arrives sacking SND.NXT at the moment, when the
+ * segment was retransmitted.
+ * 4. D-SACK added new rule: D-SACK changes any tag to S.
+ *
+ * It is pleasant to note, that state diagram turns out to be commutative,
+ * so that we are allowed not to be bothered by order of our actions,
+ * when multiple events arrive simultaneously. (see the function below).
+ *
+ * Reordering detection.
+ * --------------------
+ * Reordering metric is maximal distance, which a packet can be displaced
+ * in packet stream. With SACKs we can estimate it:
+ *
+ * 1. SACK fills old hole and the corresponding segment was not
+ * ever retransmitted -> reordering. Alas, we cannot use it
+ * when segment was retransmitted.
+ * 2. The last flaw is solved with D-SACK. D-SACK arrives
+ * for retransmitted and already SACKed segment -> reordering..
+ * Both of these heuristics are not used in Loss state, when we cannot
+ * account for retransmits accurately.
+ */
+static int
+tcp_sacktag_write_queue(struct sock *sk, struct sk_buff *ack_skb, u32 prior_snd_una)
+{
+ struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
+ unsigned char *ptr = ack_skb->h.raw + TCP_SKB_CB(ack_skb)->sacked;
+ struct tcp_sack_block *sp = (struct tcp_sack_block *)(ptr+2);
+ int num_sacks = (ptr[1] - TCPOLEN_SACK_BASE)>>3;
+ int reord = tp->packets_out;
+ int prior_fackets;
+ u32 lost_retrans = 0;
+ int flag = 0;
+ int i;
- /* Sorry, PAWS as specified is broken wrt. pure-ACKs -DaveM
+ if (!tp->sacked_out)
+ tp->fackets_out = 0;
+ prior_fackets = tp->fackets_out;
+
+ for (i=0; i<num_sacks; i++, sp++) {
+ struct sk_buff *skb;
+ __u32 start_seq = ntohl(sp->start_seq);
+ __u32 end_seq = ntohl(sp->end_seq);
+ int fack_count = 0;
+ int dup_sack = 0;
- I cannot see quitely as all the idea behind PAWS
- is destroyed 8)
+ /* Check for D-SACK. */
+ if (i == 0) {
+ u32 ack = TCP_SKB_CB(ack_skb)->ack_seq;
+
+ if (before(start_seq, ack)) {
+ dup_sack = 1;
+ NET_INC_STATS_BH(TCPDSACKRecv);
+ } else if (num_sacks > 1 &&
+ !after(end_seq, ntohl(sp[1].end_seq)) &&
+ !before(start_seq, ntohl(sp[1].start_seq))) {
+ dup_sack = 1;
+ NET_INC_STATS_BH(TCPDSACKOfoRecv);
+ }
+
+ /* D-SACK for already forgotten data...
+ * Do dumb counting. */
+ if (dup_sack &&
+ !after(end_seq, prior_snd_una) &&
+ after(end_seq, tp->undo_marker))
+ tp->undo_retrans--;
+
+ /* Eliminate too old ACKs, but take into
+ * account more or less fresh ones, they can
+ * contain valid SACK info.
+ */
+ if (before(ack, prior_snd_una-tp->max_window))
+ return 0;
+ }
- The problem is only in reordering duplicate ACKs.
- Hence, we can check this rare case more carefully.
-
- 1. Check that it is really duplicate ACK (ack==snd_una)
- 2. Give it some small "replay" window (~RTO)
-
- We do not know units of foreign ts values, but make conservative
- assumption that they are >=1ms. It solves problem
- noted in Dave's mail to tcpimpl and does not harm PAWS. --ANK
- */
- && (TCP_SKB_CB(skb)->seq != TCP_SKB_CB(skb)->end_seq ||
- TCP_SKB_CB(skb)->ack_seq != tp->snd_una ||
- !skb->h.th->ack ||
- (s32)(tp->ts_recent - tp->rcv_tsval) > (tp->rto*1024)/HZ));
-}
+ /* Event "B" in the comment above. */
+ if (after(end_seq, tp->high_seq))
+ flag |= FLAG_DATA_LOST;
+
+ for_retrans_queue(skb, sk, tp) {
+ u8 sacked = TCP_SKB_CB(skb)->sacked;
+ int in_sack;
+ /* The retransmission queue is always in order, so
+ * we can short-circuit the walk early.
+ */
+ if(!before(TCP_SKB_CB(skb)->seq, end_seq))
+ break;
-static int __tcp_sequence(struct tcp_opt *tp, u32 seq, u32 end_seq)
-{
- u32 end_window = tp->rcv_wup + tp->rcv_wnd;
-#ifdef TCP_FORMAL_WINDOW
- u32 rcv_wnd = tcp_receive_window(tp);
-#else
- u32 rcv_wnd = tp->rcv_wnd;
-#endif
+ fack_count++;
- if (rcv_wnd &&
- after(end_seq, tp->rcv_nxt) &&
- before(seq, end_window))
- return 1;
- if (seq != end_window)
- return 0;
- return (seq == end_seq);
-}
+ in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq) &&
+ !before(end_seq, TCP_SKB_CB(skb)->end_seq);
-/* This functions checks to see if the tcp header is actually acceptable. */
-extern __inline__ int tcp_sequence(struct tcp_opt *tp, u32 seq, u32 end_seq)
-{
-#ifdef TCP_FORMAL_WINDOW
- u32 rcv_wnd = tcp_receive_window(tp);
-#else
- u32 rcv_wnd = tp->rcv_wnd;
-#endif
- if (seq == tp->rcv_nxt)
- return (rcv_wnd || (end_seq == seq));
+ /* Account D-SACK for retransmitted packet. */
+ if ((dup_sack && in_sack) &&
+ (sacked & TCPCB_RETRANS) &&
+ after(TCP_SKB_CB(skb)->end_seq, tp->undo_marker))
+ tp->undo_retrans--;
+
+ /* The frame is ACKed. */
+ if (!after(TCP_SKB_CB(skb)->end_seq, tp->snd_una)) {
+ if (sacked&TCPCB_RETRANS) {
+ if ((dup_sack && in_sack) &&
+ (sacked&TCPCB_SACKED_ACKED))
+ reord = min(fack_count, reord);
+ } else {
+ /* If it was in a hole, we detected reordering. */
+ if (fack_count < prior_fackets &&
+ !(sacked&TCPCB_SACKED_ACKED))
+ reord = min(fack_count, reord);
+ }
- return __tcp_sequence(tp, seq, end_seq);
-}
+ /* Nothing to do; acked frame is about to be dropped. */
+ continue;
+ }
-/* When we get a reset we do this. */
-static void tcp_reset(struct sock *sk)
-{
- /* We want the right error as BSD sees it (and indeed as we do). */
- switch (sk->state) {
- case TCP_SYN_SENT:
- sk->err = ECONNREFUSED;
- break;
- case TCP_CLOSE_WAIT:
- sk->err = EPIPE;
- break;
- case TCP_CLOSE:
- return;
- default:
- sk->err = ECONNRESET;
- }
+ if ((sacked&TCPCB_SACKED_RETRANS) &&
+ after(end_seq, TCP_SKB_CB(skb)->ack_seq) &&
+ (!lost_retrans || after(end_seq, lost_retrans)))
+ lost_retrans = end_seq;
- if (!sk->dead)
- sk->error_report(sk);
+ if (!in_sack)
+ continue;
- tcp_done(sk);
-}
+ if (!(sacked&TCPCB_SACKED_ACKED)) {
+ if (sacked & TCPCB_SACKED_RETRANS) {
+ /* If the segment is not tagged as lost,
+ * we do not clear RETRANS, believing
+ * that retransmission is still in flight.
+ */
+ if (sacked & TCPCB_LOST) {
+ TCP_SKB_CB(skb)->sacked &= ~(TCPCB_LOST|TCPCB_SACKED_RETRANS);
+ tp->lost_out--;
+ tp->retrans_out--;
+ }
+ } else {
+ /* New sack for not retransmitted frame,
+ * which was in hole. It is reordering.
+ */
+ if (!(sacked & TCPCB_RETRANS) &&
+ fack_count < prior_fackets)
+ reord = min(fack_count, reord);
+
+ if (sacked & TCPCB_LOST) {
+ TCP_SKB_CB(skb)->sacked &= ~TCPCB_LOST;
+ tp->lost_out--;
+ }
+ }
-/* This tags the retransmission queue when SACKs arrive. */
-static void tcp_sacktag_write_queue(struct sock *sk, struct tcp_sack_block *sp, int nsacks)
-{
- struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
- int i = nsacks;
+ TCP_SKB_CB(skb)->sacked |= TCPCB_SACKED_ACKED;
+ flag |= FLAG_DATA_SACKED;
+ tp->sacked_out++;
- while(i--) {
- struct sk_buff *skb = skb_peek(&sk->write_queue);
- __u32 start_seq = ntohl(sp->start_seq);
- __u32 end_seq = ntohl(sp->end_seq);
- int fack_count = 0;
+ if (fack_count > tp->fackets_out)
+ tp->fackets_out = fack_count;
+ } else {
+ if (dup_sack && (sacked&TCPCB_RETRANS))
+ reord = min(fack_count, reord);
+ }
- while((skb != NULL) &&
- (skb != tp->send_head) &&
- (skb != (struct sk_buff *)&sk->write_queue)) {
- /* The retransmission queue is always in order, so
- * we can short-circuit the walk early.
+ /* D-SACK. We can detect redundant retransmission
+ * in S|R and plain R frames and clear it.
+ * undo_retrans is decreased above, L|R frames
+ * are accounted above as well.
*/
- if(after(TCP_SKB_CB(skb)->seq, end_seq))
- break;
+ if (dup_sack &&
+ (TCP_SKB_CB(skb)->sacked&TCPCB_SACKED_RETRANS)) {
+ TCP_SKB_CB(skb)->sacked &= ~TCPCB_SACKED_RETRANS;
+ tp->retrans_out--;
+ }
+ }
+ }
- /* We play conservative, we don't allow SACKS to partially
- * tag a sequence space.
- */
- fack_count++;
- if(!after(start_seq, TCP_SKB_CB(skb)->seq) &&
- !before(end_seq, TCP_SKB_CB(skb)->end_seq)) {
- /* If this was a retransmitted frame, account for it. */
- if((TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) &&
- tp->retrans_out)
- tp->retrans_out--;
- TCP_SKB_CB(skb)->sacked |= TCPCB_SACKED_ACKED;
+ /* Check for lost retransmit. This superb idea is
+ * borrowed from "ratehalving". Event "C".
+ * Later note: FACK people cheated me again 8),
+ * we have to account for reordering! Ugly,
+ * but should help.
+ */
+ if (lost_retrans && tp->ca_state == TCP_CA_Recovery) {
+ struct sk_buff *skb;
- /* RULE: All new SACKs will either decrease retrans_out
- * or advance fackets_out.
- */
- if(fack_count > tp->fackets_out)
- tp->fackets_out = fack_count;
+ for_retrans_queue(skb, sk, tp) {
+ if (after(TCP_SKB_CB(skb)->seq, lost_retrans))
+ break;
+ if (!after(TCP_SKB_CB(skb)->end_seq, tp->snd_una))
+ continue;
+ if ((TCP_SKB_CB(skb)->sacked&TCPCB_SACKED_RETRANS) &&
+ after(lost_retrans, TCP_SKB_CB(skb)->ack_seq) &&
+ (IsFack(tp) ||
+ !before(lost_retrans, TCP_SKB_CB(skb)->ack_seq+tp->reordering*tp->mss_cache))) {
+ TCP_SKB_CB(skb)->sacked &= ~TCPCB_SACKED_RETRANS;
+ tp->retrans_out--;
+
+ if (!(TCP_SKB_CB(skb)->sacked&(TCPCB_LOST|TCPCB_SACKED_ACKED))) {
+ tp->lost_out++;
+ TCP_SKB_CB(skb)->sacked |= TCPCB_LOST;
+ flag |= FLAG_DATA_SACKED;
+ NET_INC_STATS_BH(TCPLostRetransmit);
+ }
}
- skb = skb->next;
}
- sp++; /* Move on to the next SACK block. */
}
+
+ tp->left_out = tp->sacked_out + tp->lost_out;
+
+ if (reord < tp->fackets_out && tp->ca_state != TCP_CA_Loss)
+ tcp_update_reordering(tp, (tp->fackets_out+1)-reord, 0);
+
+#if FASTRETRANS_DEBUG > 0
+ BUG_TRAP((int)tp->sacked_out >= 0);
+ BUG_TRAP((int)tp->lost_out >= 0);
+ BUG_TRAP((int)tp->retrans_out >= 0);
+ BUG_TRAP((int)tcp_packets_in_flight(tp) >= 0);
+#endif
+ return flag;
}
-/* Look for tcp options. Normally only called on SYN and SYNACK packets.
- * But, this can also be called on packets in the established flow when
- * the fast version below fails.
+void tcp_clear_retrans(struct tcp_opt *tp)
+{
+ tp->left_out = 0;
+ tp->retrans_out = 0;
+
+ tp->fackets_out = 0;
+ tp->sacked_out = 0;
+ tp->lost_out = 0;
+
+ tp->undo_marker = 0;
+ tp->undo_retrans = 0;
+}
+
+/* Enter Loss state. If "how" is not zero, forget all SACK information
+ * and reset tags completely, otherwise preserve SACKs. If receiver
+ * dropped its ofo queue, we will know this due to reneging detection.
*/
-void tcp_parse_options(struct sock *sk, struct tcphdr *th, struct tcp_opt *tp, int no_fancy)
+void tcp_enter_loss(struct sock *sk, int how)
{
- unsigned char *ptr;
- int length=(th->doff*4)-sizeof(struct tcphdr);
+ struct tcp_opt *tp = &sk->tp_pinfo.af_tcp;
+ struct sk_buff *skb;
+ int cnt = 0;
- ptr = (unsigned char *)(th + 1);
- tp->saw_tstamp = 0;
+ /* Reduce ssthresh if it has not yet been made inside this window. */
+ if (tp->ca_state <= TCP_CA_Disorder ||
+ tp->snd_una == tp->high_seq ||
+ (tp->ca_state == TCP_CA_Loss && !tp->retransmits)) {
+ tp->prior_ssthresh = tcp_current_ssthresh(tp);
+ tp->snd_ssthresh = tcp_recalc_ssthresh(tp);
+ }
+ tp->snd_cwnd = 1;
+ tp->snd_cwnd_cnt = 0;
+ tp->snd_cwnd_stamp = tcp_time_stamp;
+
+ tcp_clear_retrans(tp);
+
+ /* Push undo marker, if it was plain RTO and nothing
+ * was retransmitted. */
+ if (!how)
+ tp->undo_marker = tp->snd_una;
+
+ for_retrans_queue(skb, sk, tp) {
+ cnt++;
+ if (TCP_SKB_CB(skb)->sacked&TCPCB_RETRANS)
+ tp->undo_marker = 0;
+ TCP_SKB_CB(skb)->sacked &= (~TCPCB_TAGBITS)|TCPCB_SACKED_ACKED;
+ if (!(TCP_SKB_CB(skb)->sacked&TCPCB_SACKED_ACKED) || how) {
+ TCP_SKB_CB(skb)->sacked &= ~TCPCB_SACKED_ACKED;
+ TCP_SKB_CB(skb)->sacked |= TCPCB_LOST;
+ tp->lost_out++;
+ } else {
+ tp->sacked_out++;
+ tp->fackets_out = cnt;
+ }
+ }
+ tp->left_out = tp->sacked_out + tp->lost_out;
- while(length>0) {
- int opcode=*ptr++;
- int opsize;
+ tp->reordering = min(tp->reordering, sysctl_tcp_reordering);
+ tp->ca_state = TCP_CA_Loss;
+ tp->high_seq = tp->snd_nxt;
+ TCP_ECN_queue_cwr(tp);
+}
- switch (opcode) {
- case TCPOPT_EOL:
- return;
- case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
- length--;
- continue;
- default:
- opsize=*ptr++;
- if (opsize < 2) /* "silly options" */
- return;
- if (opsize > length)
- break; /* don't parse partial options */
- switch(opcode) {
- case TCPOPT_MSS:
- if(opsize==TCPOLEN_MSS && th->syn) {
- u16 in_mss = ntohs(*(__u16 *)ptr);
- if (in_mss) {
- if (tp->user_mss && tp->user_mss < in_mss)
- in_mss = tp->user_mss;
- tp->mss_clamp = in_mss;
- }
- }
- break;
- case TCPOPT_WINDOW:
- if(opsize==TCPOLEN_WINDOW && th->syn)
- if (!no_fancy && sysctl_tcp_window_scaling) {
- tp->wscale_ok = 1;
- tp->snd_wscale = *(__u8 *)ptr;
- if(tp->snd_wscale > 14) {
- if(net_ratelimit())
- printk("tcp_parse_options: Illegal window "
- "scaling value %d >14 received.",
- tp->snd_wscale);
- tp->snd_wscale = 14;
- }
- }
- break;
- case TCPOPT_TIMESTAMP:
- if(opsize==TCPOLEN_TIMESTAMP) {
- if (sysctl_tcp_timestamps && !no_fancy) {
- tp->tstamp_ok = 1;
- tp->saw_tstamp = 1;
- tp->rcv_tsval = ntohl(*(__u32 *)ptr);
- tp->rcv_tsecr = ntohl(*(__u32 *)(ptr+4));
- }
- }
- break;
- case TCPOPT_SACK_PERM:
- if(opsize==TCPOLEN_SACK_PERM && th->syn) {
- if (sysctl_tcp_sack && !no_fancy) {
- tp->sack_ok = 1;
- tp->num_sacks = 0;
- }
- }
- break;
+static int tcp_check_sack_reneging(struct sock *sk, struct tcp_opt *tp)
+{
+ struct sk_buff *skb;
- case TCPOPT_SACK:
- if((opsize >= (TCPOLEN_SACK_BASE + TCPOLEN_SACK_PERBLOCK)) &&
- sysctl_tcp_sack && (sk != NULL) && !th->syn) {
- int sack_bytes = opsize - TCPOLEN_SACK_BASE;
+ /* If ACK arrived pointing to a remembered SACK,
+ * it means that our remembered SACKs do not reflect
+ * real state of receiver i.e.
+ * receiver _host_ is heavily congested (or buggy).
+ * Do processing similar to RTO timeout.
+ */
+ if ((skb = skb_peek(&sk->write_queue)) != NULL &&
+ (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)) {
+ NET_INC_STATS_BH(TCPSACKReneging);
+
+ tcp_enter_loss(sk, 1);
+ tp->retransmits++;
+ tcp_retransmit_skb(sk, skb_peek(&sk->write_queue));
+ tcp_reset_xmit_timer(sk, TCP_TIME_RETRANS, tp->rto);
+ return 1;
+ }
+ return 0;
+}
+
+static inline int tcp_fackets_out(struct tcp_opt *tp)
+{
+ return IsReno(tp) ? tp->sacked_out+1 : tp->fackets_out;
+}
- if(!(sack_bytes % TCPOLEN_SACK_PERBLOCK)) {
- int num_sacks = sack_bytes >> 3;
- struct tcp_sack_block *sackp;
- sackp = (struct tcp_sack_block *)ptr;
- tcp_sacktag_write_queue(sk, sackp, num_sacks);
- }
- }
- };
- ptr+=opsize-2;
- length-=opsize;
- };
+/* Linux NewReno/SACK/FACK/ECN state machine.
+ * --------------------------------------
+ *
+ * "Open" Normal state, no dubious events, fast path.
+ * "Disorder" In all the respects it is "Open",
+ * but requires a bit more attention. It is entered when
+ * we see some SACKs or dupacks. It is split of "Open"
+ * mainly to move some processing from fast path to slow one.
+ * "CWR" CWND was reduced due to some Congestion Notification event.
+ * It can be ECN, ICMP source quench, local device congestion.
+ * "Recovery" CWND was reduced, we are fast-retransmitting.
+ * "Loss" CWND was reduced due to RTO timeout or SACK reneging.
+ *
+ * tcp_fastretrans_alert() is entered:
+ * - each incoming ACK, if state is not "Open"
+ * - when arrived ACK is unusual, namely:
+ * * SACK
+ * * Duplicate ACK.
+ * * ECN ECE.
+ *
+ * Counting packets in flight is pretty simple.
+ *
+ * in_flight = packets_out - left_out + retrans_out
+ *
+ * packets_out is SND.NXT-SND.UNA counted in packets.
+ *
+ * retrans_out is number of retransmitted segments.
+ *
+ * left_out is number of segments left network, but not ACKed yet.
+ *
+ * left_out = sacked_out + lost_out
+ *
+ * sacked_out: Packets, which arrived to receiver out of order
+ * and hence not ACKed. With SACKs this number is simply
+ * amount of SACKed data. Even without SACKs
+ * it is easy to give pretty reliable estimate of this number,
+ * counting duplicate ACKs.
+ *
+ * lost_out: Packets lost by network. TCP has no explicit
+ * "loss notification" feedback from network (for now).
+ * It means that this number can be only _guessed_.
+ * Actually, it is the heuristics to predict lossage that
+ * distinguishes different algorithms.
+ *
+ * F.e. after RTO, when all the queue is considered as lost,
+ * lost_out = packets_out and in_flight = retrans_out.
+ *
+ * Essentially, we have now two algorithms counting
+ * lost packets.
+ *
+ * FACK: It is the simplest heuristics. As soon as we decided
+ * that something is lost, we decide that _all_ not SACKed
+ * packets until the most forward SACK are lost. I.e.
+ * lost_out = fackets_out - sacked_out and left_out = fackets_out.
+ * It is absolutely correct estimate, if network does not reorder
+ * packets. And it loses any connection to reality when reordering
+ * takes place. We use FACK by default until reordering
+ * is suspected on the path to this destination.
+ *
+ * NewReno: when Recovery is entered, we assume that one segment
+ * is lost (classic Reno). While we are in Recovery and
+ * a partial ACK arrives, we assume that one more packet
+ * is lost (NewReno). This heuristics are the same in NewReno
+ * and SACK.
+ *
+ * Imagine, that's all! Forget about all this shamanism about CWND inflation
+ * deflation etc. CWND is real congestion window, never inflated, changes
+ * only according to classic VJ rules.
+ *
+ * Really tricky (and requiring careful tuning) part of algorithm
+ * is hidden in functions tcp_time_to_recover() and tcp_xmit_retransmit_queue().
+ * The first determines the moment _when_ we should reduce CWND and,
+ * hence, slow down forward transmission. In fact, it determines the moment
+ * when we decide that hole is caused by loss, rather than by a reorder.
+ *
+ * tcp_xmit_retransmit_queue() decides, _what_ we should retransmit to fill
+ * holes, caused by lost packets.
+ *
+ * And the most logically complicated part of algorithm is undo
+ * heuristics. We detect false retransmits due to both too early
+ * fast retransmit (reordering) and underestimated RTO, analyzing
+ * timestamps and D-SACKs. When we detect that some segments were
+ * retransmitted by mistake and CWND reduction was wrong, we undo
+ * window reduction and abort recovery phase. This logic is hidden
+ * inside several functions named tcp_try_undo_<something>.
+ */
+
+/* This function decides, when we should leave Disordered state
+ * and enter Recovery phase, reducing congestion window.
+ *
+ * Main question: may we further continue forward transmission
+ * with the same cwnd?
+ */
+static int
+tcp_time_to_recover(struct sock *sk, struct tcp_opt *tp)
+{
+ /* Trick#1: The loss is proven. */
+ if (tp->lost_out)
+ return 1;
+
+ /* Not-A-Trick#2 : Classic rule... */
+ if (tcp_fackets_out(tp) > tp->reordering)
+ return 1;
+
+ /* Trick#3: It is still not OK... But will it be useful to delay
+ * recovery more?
+ */
+ if (tp->packets_out <= tp->reordering &&
+ tp->sacked_out >= max(tp->packets_out/2, sysctl_tcp_reordering) &&
+ !tcp_may_send_now(sk, tp)) {
+ /* We have nothing to send. This connection is limited
+ * either by receiver window or by application.
+ */
+ return 1;
}
+
+ return 0;
}
-/* Fast parse options. This hopes to only see timestamps.
- * If it is wrong it falls back on tcp_parse_options().
+/* If we receive more dupacks than we expected counting segments
+ * in assumption of absent reordering, interpret this as reordering.
+ * The only another reason could be bug in receiver TCP.
*/
-static __inline__ int tcp_fast_parse_options(struct sock *sk, struct tcphdr *th, struct tcp_opt *tp)
+static void tcp_check_reno_reordering(struct tcp_opt *tp, int addend)
{
- /* If we didn't send out any options ignore them all. */
- if (tp->tcp_header_len == sizeof(struct tcphdr))
- return 0;
- if (th->doff == sizeof(struct tcphdr)>>2) {
- tp->saw_tstamp = 0;
- return 0;
- } else if (th->doff == (sizeof(struct tcphdr)>>2)+(TCPOLEN_TSTAMP_ALIGNED>>2)) {
- __u32 *ptr = (__u32 *)(th + 1);
- if (*ptr == __constant_ntohl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16)
- | (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP)) {
- tp->saw_tstamp = 1;
- ++ptr;
- tp->rcv_tsval = ntohl(*ptr);
- ++ptr;
- tp->rcv_tsecr = ntohl(*ptr);
- return 1;
+ if (tp->sacked_out + 1 > tp->packets_out) {
+ tp->sacked_out = tp->packets_out ? tp->packets_out - 1 : 0;
+ tcp_update_reordering(tp, tp->packets_out+addend, 0);
+ }
+}
+
+/* Emulate SACKs for SACKless connection: account for a new dupack. */
+
+static void tcp_add_reno_sack(struct tcp_opt *tp)
+{
+ ++tp->sacked_out;
+ tcp_check_reno_reordering(tp, 0);
+ tp->left_out = tp->sacked_out + tp->lost_out;
+}
+
+/* Account for ACK, ACKing some data in Reno Recovery phase. */
+
+static void tcp_remove_reno_sacks(struct sock *sk, struct tcp_opt *tp, int acked)
+{
+ if (acked > 0) {
+ /* One ACK eated lost packet. Must eat! */
+ BUG_TRAP(tp->lost_out == 0);
+
+ /* The rest eat duplicate ACKs. */
+ if (acked-1 >= tp->sacked_out)
+ tp->sacked_out = 0;
+ else
+ tp->sacked_out -= acked-1;
+ }
+ tcp_check_reno_reordering(tp, acked);
+ tp->left_out = tp->sacked_out + tp->lost_out;
+}
+
+static inline void tcp_reset_reno_sack(struct tcp_opt *tp)
+{
+ tp->sacked_out = 0;
+ tp->left_out = tp->lost_out;
+}
+
+/* Mark head of queue up as lost. */
+static void
+tcp_mark_head_lost(struct sock *sk, struct tcp_opt *tp, int packets, u32 high_seq)
+{
+ struct sk_buff *skb;
+ int cnt = packets;
+
+ BUG_TRAP(cnt <= tp->packets_out);
+
+ for_retrans_queue(skb, sk, tp) {
+ if (--cnt < 0 || after(TCP_SKB_CB(skb)->end_seq, high_seq))
+ break;
+ if (!(TCP_SKB_CB(skb)->sacked&TCPCB_TAGBITS)) {
+ TCP_SKB_CB(skb)->sacked |= TCPCB_LOST;
+ tp->lost_out++;
}
}
- tcp_parse_options(sk, th, tp, 0);
- return 1;
+ tp->left_out = tp->sacked_out + tp->lost_out;
}
-#define FLAG_DATA 0x01 /* Incoming frame contained data. */
-#define FLAG_WIN_UPDATE 0x02 /* Incoming ACK was a window update. */
-#define FLAG_DATA_ACKED 0x04 /* This ACK acknowledged new data. */
-#define FLAG_RETRANS_DATA_ACKED 0x08 /* "" "" some of which was retransmitted. */
-#define FLAG_SYN_ACKED 0x10 /* This ACK acknowledged new data. */
+/* Account newly detected lost packet(s) */
+
+static void tcp_update_scoreboard(struct sock *sk, struct tcp_opt *tp)
+{
+ if (IsFack(tp)) {
+ int lost = tp->fackets_out - tp->reordering;
+ if (lost <= 0)
+ lost = 1;
+ tcp_mark_head_lost(sk, tp, lost, tp->high_seq);
+ } else {
+ tcp_mark_head_lost(sk, tp, 1, tp->high_seq);
+ }
+}
+
+/* CWND moderation, preventing bursts due to too big ACKs
+ * in dubious situations.
+ */
+static __inline__ void tcp_moderate_cwnd(struct tcp_opt *tp)
+{
+ tp->snd_cwnd = min(tp->snd_cwnd,
+ tcp_packets_in_flight(tp)+tcp_max_burst(tp));
+ tp->snd_cwnd_stamp = tcp_time_stamp;
+}
+
+/* Decrease cwnd each second ack. */
+
+static void tcp_cwnd_down(struct tcp_opt *tp)
+{
+ int decr = tp->snd_cwnd_cnt + 1;
+
+ tp->snd_cwnd_cnt = decr&1;
+ decr >>= 1;
+
+ if (decr && tp->snd_cwnd > tp->snd_ssthresh/2)
+ tp->snd_cwnd -= decr;
+
+ tp->snd_cwnd = min(tp->snd_cwnd, tcp_packets_in_flight(tp)+1);
+ tp->snd_cwnd_stamp = tcp_time_stamp;
+}
+
+/* Nothing was retransmitted or returned timestamp is less
+ * than timestamp of the first retransmission.
+ */
+static __inline__ int tcp_packet_delayed(struct tcp_opt *tp)
+{
+ return !tp->retrans_stamp ||
+ (tp->saw_tstamp &&
+ (__s32)(tp->rcv_tsecr - tp->retrans_stamp) < 0);
+}
+
+/* Undo procedures. */
+
+#if FASTRETRANS_DEBUG > 1
+static void DBGUNDO(struct sock *sk, struct tcp_opt *tp, const char *msg)
+{
+ printk(KERN_DEBUG "Undo %s %u.%u.%u.%u/%u c%u l%u ss%u/%u p%u\n",
+ msg,
+ NIPQUAD(sk->daddr), ntohs(sk->dport),
+ tp->snd_cwnd, tp->left_out,
+ tp->snd_ssthresh, tp->prior_ssthresh, tp->packets_out);
+}
+#else
+#define DBGUNDO(x...) do { } while (0)
+#endif
+
+static void tcp_undo_cwr(struct tcp_opt *tp, int undo)
+{
+ if (tp->prior_ssthresh) {
+ tp->snd_cwnd = max(tp->snd_cwnd, tp->snd_ssthresh<<1);
+ if (undo && tp->prior_ssthresh > tp->snd_ssthresh)
+ tp->snd_ssthresh = tp->prior_ssthresh;
+ } else {
+ tp->snd_cwnd = max(tp->snd_cwnd, tp->snd_ssthresh);
+ }
+ tcp_moderate_cwnd(tp);
+ tp->snd_cwnd_stamp = tcp_time_stamp;
+}
+
+static inline int tcp_may_undo(struct tcp_opt *tp)
+{
+ return tp->undo_marker &&
+ (!tp->undo_retrans || tcp_packet_delayed(tp));
+}
+
+/* People celebrate: "We love our President!" */
+static int tcp_try_undo_recovery(struct sock *sk, struct tcp_opt *tp)
+{
+ if (tcp_may_undo(tp)) {
+ /* Happy end! We did not retransmit anything
+ * or our original transmission succeeded.
+ */
+ DBGUNDO(sk, tp, tp->ca_state == TCP_CA_Loss ? "loss" : "retrans");
+ tcp_undo_cwr(tp, 1);
+ if (tp->ca_state == TCP_CA_Loss)
+ NET_INC_STATS_BH(TCPLossUndo);
+ else
+ NET_INC_STATS_BH(TCPFullUndo);
+ tp->undo_marker = 0;
+ }
+ if (tp->snd_una == tp->high_seq && IsReno(tp)) {
+ /* Hold old state until something *above* high_seq
+ * is ACKed. For Reno it is MUST to prevent false
+ * fast retransmits (RFC2582). SACK TCP is safe. */
+ tcp_moderate_cwnd(tp);
+ return 1;
+ }
+ tp->ca_state = TCP_CA_Open;
+ return 0;
+}
+
+/* Try to undo cwnd reduction, because D-SACKs acked all retransmitted data */
+static void tcp_try_undo_dsack(struct sock *sk, struct tcp_opt *tp)
+{
+ if (tp->undo_marker && !tp->undo_retrans) {
+ DBGUNDO(sk, tp, "D-SACK");
+ tcp_undo_cwr(tp, 1);
+ tp->undo_marker = 0;
+ NET_INC_STATS_BH(TCPDSACKUndo);
+ }
+}
+
+/* Undo during fast recovery after partial ACK. */
+
+static int tcp_try_undo_partial(struct sock *sk, struct tcp_opt *tp, int acked)
+{
+ /* Partial ACK arrived. Force Hoe's retransmit. */
+ int failed = IsReno(tp) || tp->fackets_out>tp->reordering;
+
+ if (tcp_may_undo(tp)) {
+ /* Plain luck! Hole if filled with delayed
+ * packet, rather than with a retransmit.
+ */
+ if (tp->retrans_out == 0)
+ tp->retrans_stamp = 0;
+
+ tcp_update_reordering(tp, tcp_fackets_out(tp)+acked, 1);
+
+ DBGUNDO(sk, tp, "Hoe");
+ tcp_undo_cwr(tp, 0);
+ NET_INC_STATS_BH(TCPPartialUndo);
+
+ /* So... Do not make Hoe's retransmit yet.
+ * If the first packet was delayed, the rest
+ * ones are most probably delayed as well.
+ */
+ failed = 0;
+ }
+ return failed;
+}
+
+/* Undo during loss recovery after partial ACK. */
+static int tcp_try_undo_loss(struct sock *sk, struct tcp_opt *tp)
+{
+ if (tcp_may_undo(tp)) {
+ struct sk_buff *skb;
+ for_retrans_queue(skb, sk, tp) {
+ TCP_SKB_CB(skb)->sacked &= ~TCPCB_LOST;
+ }
+ DBGUNDO(sk, tp, "partial loss");
+ tp->lost_out = 0;
+ tp->left_out = tp->sacked_out;
+ tcp_undo_cwr(tp, 1);
+ NET_INC_STATS_BH(TCPLossUndo);
+ tp->retransmits = 0;
+ tp->undo_marker = 0;
+ if (!IsReno(tp)) {
+ tp->ca_state = TCP_CA_Open;
+ tp->backoff = 0;
+ }
+ return 1;
+ }
+ return 0;
+}
+
+static __inline__ void tcp_complete_cwr(struct tcp_opt *tp)
+{
+ tp->snd_cwnd = min(tp->snd_cwnd, tp->snd_ssthresh);
+ tp->snd_cwnd_stamp = tcp_time_stamp;
+}
-static __inline__ void clear_fast_retransmit(struct tcp_opt *tp)
+static void tcp_try_to_open(struct sock *sk, struct tcp_opt *tp, int flag)
{
- if (tp->dup_acks > 3)
- tp->snd_cwnd = (tp->snd_ssthresh);
+ tp->left_out = tp->sacked_out;
- tp->dup_acks = 0;
+ if (tp->retrans_out == 0)
+ tp->retrans_stamp = 0;
+
+ if (flag&FLAG_ECE)
+ tcp_enter_cwr(tp);
+
+ if (tp->ca_state != TCP_CA_CWR) {
+ int state = TCP_CA_Open;
+
+ if (tp->left_out ||
+ tp->retrans_out ||
+ tp->undo_marker)
+ state = TCP_CA_Disorder;
+
+ if (tp->ca_state != state) {
+ tp->ca_state = state;
+ tp->high_seq = tp->snd_nxt;
+ }
+ tcp_moderate_cwnd(tp);
+ } else {
+ tcp_cwnd_down(tp);
+ }
+}
+
+/* Process an event, which can update packets-in-flight not trivially.
+ * Main goal of this function is to calculate new estimate for left_out,
+ * taking into account both packets sitting in receiver's buffer and
+ * packets lost by network.
+ *
+ * Besides that it does CWND reduction, when packet loss is detected
+ * and changes state of machine.
+ *
+ * It does _not_ decide what to send, it is made in function
+ * tcp_xmit_retransmit_queue().
+ */
+static void
+tcp_fastretrans_alert(struct sock *sk, u32 prior_snd_una,
+ int prior_packets, int flag)
+{
+ struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
+ int is_dupack = (tp->snd_una == prior_snd_una && !(flag&FLAG_NOT_DUP));
+
+ /* Some technical things:
+ * 1. Reno does not count dupacks (sacked_out) automatically. */
+ if (!tp->packets_out)
+ tp->sacked_out = 0;
+ /* 2. SACK counts snd_fack in packets inaccurately. */
+ if (tp->sacked_out == 0)
+ tp->fackets_out = 0;
+
+ /* Now state machine starts.
+ * A. ECE, hence prohibit cwnd undoing, the reduction is required. */
+ if (flag&FLAG_ECE)
+ tp->prior_ssthresh = 0;
+
+ /* B. In all the states check for reneging SACKs. */
+ if (tp->sacked_out && tcp_check_sack_reneging(sk, tp))
+ return;
+
+ /* C. Process data loss notification, provided it is valid. */
+ if ((flag&FLAG_DATA_LOST) &&
+ before(tp->snd_una, tp->high_seq) &&
+ tp->ca_state != TCP_CA_Open &&
+ tp->fackets_out > tp->reordering) {
+ tcp_mark_head_lost(sk, tp, tp->fackets_out-tp->reordering, tp->high_seq);
+ NET_INC_STATS_BH(TCPLoss);
+ }
+
+ /* D. Synchronize left_out to current state. */
+ tp->left_out = tp->sacked_out + tp->lost_out;
+
+ /* E. Check state exit conditions. State can be terminated
+ * when high_seq is ACKed. */
+ if (tp->ca_state == TCP_CA_Open) {
+ BUG_TRAP(tp->retrans_out == 0);
+ tp->retrans_stamp = 0;
+ } else if (!before(tp->snd_una, tp->high_seq)) {
+ switch (tp->ca_state) {
+ case TCP_CA_Loss:
+ tp->retransmits = 0;
+ if (tcp_try_undo_recovery(sk, tp))
+ return;
+ tp->backoff = 0;
+ break;
+
+ case TCP_CA_CWR:
+ /* CWR is to be held something *above* high_seq
+ * is ACKed for CWR bit to reach receiver. */
+ if (tp->snd_una != tp->high_seq) {
+ tcp_complete_cwr(tp);
+ tp->ca_state = TCP_CA_Open;
+ }
+ break;
+
+ case TCP_CA_Disorder:
+ tcp_try_undo_dsack(sk, tp);
+ if (IsReno(tp) || !tp->undo_marker) {
+ tp->undo_marker = 0;
+ tp->ca_state = TCP_CA_Open;
+ }
+ break;
+
+ case TCP_CA_Recovery:
+ if (IsReno(tp))
+ tcp_reset_reno_sack(tp);
+ if (tcp_try_undo_recovery(sk, tp))
+ return;
+ tcp_complete_cwr(tp);
+ break;
+ }
+ }
+
+ /* F. Process state. */
+ switch (tp->ca_state) {
+ case TCP_CA_Recovery:
+ if (prior_snd_una == tp->snd_una) {
+ if (IsReno(tp) && is_dupack)
+ tcp_add_reno_sack(tp);
+ } else {
+ int acked = prior_packets - tp->packets_out;
+ if (IsReno(tp))
+ tcp_remove_reno_sacks(sk, tp, acked);
+ is_dupack = tcp_try_undo_partial(sk, tp, acked);
+ }
+ break;
+ case TCP_CA_Loss:
+ if (flag & FLAG_ACKED)
+ tcp_reset_xmit_timer(sk, TCP_TIME_RETRANS, tp->rto);
+ if (!tcp_try_undo_loss(sk, tp)) {
+ tcp_moderate_cwnd(tp);
+ tcp_xmit_retransmit_queue(sk);
+ return;
+ }
+ if (tp->ca_state != TCP_CA_Open)
+ return;
+ /* Loss is undone; fall through to processing in Open state. */
+ default:
+ if (IsReno(tp)) {
+ if (tp->snd_una != prior_snd_una)
+ tcp_reset_reno_sack(tp);
+ if (is_dupack)
+ tcp_add_reno_sack(tp);
+ }
+
+ if (tp->ca_state == TCP_CA_Disorder)
+ tcp_try_undo_dsack(sk, tp);
+
+ if (!tcp_time_to_recover(sk, tp)) {
+ tcp_try_to_open(sk, tp, flag);
+ return;
+ }
+
+ /* Otherwise enter Recovery state */
+
+ if (IsReno(tp))
+ NET_INC_STATS_BH(TCPRenoRecovery);
+ else
+ NET_INC_STATS_BH(TCPSackRecovery);
+
+ tp->high_seq = tp->snd_nxt;
+ tp->prior_ssthresh = 0;
+ tp->undo_marker = tp->snd_una;
+ tp->undo_retrans = tp->retrans_out;
+
+ if (tp->ca_state < TCP_CA_CWR) {
+ if (!(flag&FLAG_ECE))
+ tp->prior_ssthresh = tcp_current_ssthresh(tp);
+ tp->snd_ssthresh = tcp_recalc_ssthresh(tp);
+ TCP_ECN_queue_cwr(tp);
+ }
+
+ tp->snd_cwnd_cnt = 0;
+ tp->ca_state = TCP_CA_Recovery;
+ }
+
+ if (is_dupack)
+ tcp_update_scoreboard(sk, tp);
+ tcp_cwnd_down(tp);
+ tcp_xmit_retransmit_queue(sk);
+}
+
+/* Read draft-ietf-tcplw-high-performance before mucking
+ * with this code. (Superceeds RFC1323)
+ */
+static void tcp_ack_saw_tstamp(struct tcp_opt *tp)
+{
+ __u32 seq_rtt;
+
+ /* RTTM Rule: A TSecr value received in a segment is used to
+ * update the averaged RTT measurement only if the segment
+ * acknowledges some new data, i.e., only if it advances the
+ * left edge of the send window.
+ *
+ * See draft-ietf-tcplw-high-performance-00, section 3.3.
+ * 1998/04/10 Andrey V. Savochkin <saw@msu.ru>
+ */
+ seq_rtt = tcp_time_stamp - tp->rcv_tsecr;
+ tcp_rtt_estimator(tp, seq_rtt);
+ tcp_set_rto(tp);
+ tp->rto <<= tp->backoff;
+ tcp_bound_rto(tp);
}
-/* NOTE: This code assumes that tp->dup_acks gets cleared when a
- * retransmit timer fires.
- */
-static void tcp_fast_retrans(struct sock *sk, u32 ack, int not_dup)
+static void tcp_ack_no_tstamp(struct tcp_opt *tp, u32 seq_rtt, int flag)
{
- struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
-
- /* Note: If not_dup is set this implies we got a
- * data carrying packet or a window update.
- * This carries no new information about possible
- * lost packets, so we have to ignore it for the purposes
- * of counting duplicate acks. Ideally this does not imply we
- * should stop our fast retransmit phase, more acks may come
- * later without data to help us. Unfortunately this would make
- * the code below much more complex. For now if I see such
- * a packet I clear the fast retransmit phase.
+ /* We don't have a timestamp. Can only use
+ * packets that are not retransmitted to determine
+ * rtt estimates. Also, we must not reset the
+ * backoff for rto until we get a non-retransmitted
+ * packet. This allows us to deal with a situation
+ * where the network delay has increased suddenly.
+ * I.e. Karn's algorithm. (SIGCOMM '87, p5.)
*/
- if (ack == tp->snd_una && tp->packets_out && (not_dup == 0)) {
- /* This is the standard reno style fast retransmit branch. */
-
- /* 1. When the third duplicate ack is received, set ssthresh
- * to one half the current congestion window, but no less
- * than two segments. Retransmit the missing segment.
- */
- if (tp->high_seq == 0 || after(ack, tp->high_seq)) {
- tp->dup_acks++;
- if ((tp->fackets_out > 3) || (tp->dup_acks == 3)) {
- __tcp_enter_cong_avoid(tp);
- /* ... and account for 3 ACKs, which are
- * already received to this time.
- */
- tp->snd_cwnd += 3;
-
- if(!tp->fackets_out)
- tcp_retransmit_skb(sk,
- skb_peek(&sk->write_queue));
- else
- tcp_fack_retransmit(sk);
- tcp_reset_xmit_timer(sk, TCP_TIME_RETRANS, tp->rto);
- }
- } else if (++tp->dup_acks > 3) {
- /* 2. Each time another duplicate ACK arrives, increment
- * cwnd by the segment size. [...] Transmit a packet...
- *
- * Packet transmission will be done on normal flow processing
- * since we're not in "retransmit mode". We do not use
- * duplicate ACKs to artificially inflate the congestion
- * window when doing FACK.
- */
- if(!tp->fackets_out) {
- tp->snd_cwnd++;
- } else {
- /* Fill any further holes which may have
- * appeared.
- *
- * We may want to change this to run every
- * further multiple-of-3 dup ack increments,
- * to be more robust against out-of-order
- * packet delivery. -DaveM
- */
- tcp_fack_retransmit(sk);
- }
- }
- } else if (tp->high_seq != 0) {
- /* In this branch we deal with clearing the Floyd style
- * block on duplicate fast retransmits, and if requested
- * we do Hoe style secondary fast retransmits.
- */
- if (!before(ack, tp->high_seq) || (not_dup & FLAG_DATA) != 0) {
- /* Once we have acked all the packets up to high_seq
- * we are done this fast retransmit phase.
- * Alternatively data arrived. In this case we
- * Have to abort the fast retransmit attempt.
- * Note that we do want to accept a window
- * update since this is expected with Hoe's algorithm.
- */
- clear_fast_retransmit(tp);
- /* After we have cleared up to high_seq we can
- * clear the Floyd style block.
- */
- if (!before(ack, tp->high_seq)) {
- tp->high_seq = 0;
- tp->fackets_out = 0;
- }
- } else if (tp->dup_acks >= 3) {
- if (!tp->fackets_out) {
- /* Hoe Style. We didn't ack the whole
- * window. Take this as a cue that
- * another packet was lost and retransmit it.
- * Don't muck with the congestion window here.
- * Note that we have to be careful not to
- * act if this was a window update and it
- * didn't ack new data, since this does
- * not indicate a packet left the system.
- * We can test this by just checking
- * if ack changed from snd_una, since
- * the only way to get here without advancing
- * from snd_una is if this was a window update.
- */
- if (ack != tp->snd_una && before(ack, tp->high_seq)) {
- tcp_retransmit_skb(sk,
- skb_peek(&sk->write_queue));
- tcp_reset_xmit_timer(sk, TCP_TIME_RETRANS, tp->rto);
- }
- } else {
- /* FACK style, fill any remaining holes in
- * receiver's queue.
- */
- tcp_fack_retransmit(sk);
- }
- }
+ if (!tp->retransmits && !(flag & FLAG_RETRANS_DATA_ACKED)) {
+ tp->backoff = 0;
+ tcp_rtt_estimator(tp, seq_rtt);
+ tcp_set_rto(tp);
+ tcp_bound_rto(tp);
}
}
+static __inline__ void
+tcp_ack_update_rtt(struct tcp_opt *tp, int flag, u32 seq_rtt)
+{
+ if (tp->saw_tstamp)
+ tcp_ack_saw_tstamp(tp);
+ else
+ tcp_ack_no_tstamp(tp, seq_rtt, flag);
+}
+
/* This is Jacobson's slow start and congestion avoidance.
* SIGCOMM '88, p. 328.
*/
@@ -855,31 +1645,38 @@
}
}
+static __inline__ void tcp_ack_packets_out(struct sock *sk, struct tcp_opt *tp)
+{
+ if (tp->packets_out==0) {
+ tcp_clear_xmit_timer(sk, TCP_TIME_RETRANS);
+ } else {
+ struct sk_buff *skb = skb_peek(&sk->write_queue);
+ __u32 when = tp->rto - (tcp_time_stamp - TCP_SKB_CB(skb)->when);
+
+ if ((__s32)when <= 0)
+ when = TCP_RTO_MIN;
+ tcp_reset_xmit_timer(sk, TCP_TIME_RETRANS, when);
+ }
+}
+
/* Remove acknowledged frames from the retransmission queue. */
-static int tcp_clean_rtx_queue(struct sock *sk, __u32 ack,
- __u32 *seq, __u32 *seq_rtt)
+static int tcp_clean_rtx_queue(struct sock *sk)
{
struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
struct sk_buff *skb;
__u32 now = tcp_time_stamp;
int acked = 0;
-
- /* If we are retransmitting, and this ACK clears up to
- * the retransmit head, or further, then clear our state.
- */
- if (tp->retrans_head != NULL &&
- !before(ack, TCP_SKB_CB(tp->retrans_head)->end_seq))
- tp->retrans_head = NULL;
+ __u32 seq_rtt = 0; /* F..g gcc... */
while((skb=skb_peek(&sk->write_queue)) && (skb != tp->send_head)) {
struct tcp_skb_cb *scb = TCP_SKB_CB(skb);
__u8 sacked = scb->sacked;
-
+
/* If our packet is before the ack sequence we can
* discard it as it's confirmed to have arrived at
* the other end.
*/
- if (after(scb->end_seq, ack))
+ if (after(scb->end_seq, tp->snd_una))
break;
/* Initial outgoing SYN's get put onto the write_queue
@@ -889,711 +1686,482 @@
* connection startup slow start one packet too
* quickly. This is severely frowned upon behavior.
*/
- if((sacked & TCPCB_SACKED_RETRANS) && tp->retrans_out)
- tp->retrans_out--;
if(!(scb->flags & TCPCB_FLAG_SYN)) {
acked |= FLAG_DATA_ACKED;
- if(sacked & TCPCB_SACKED_RETRANS)
- acked |= FLAG_RETRANS_DATA_ACKED;
- if(tp->fackets_out)
- tp->fackets_out--;
} else {
acked |= FLAG_SYN_ACKED;
- /* This is pure paranoia. */
- tp->retrans_head = NULL;
}
+
+ if (sacked) {
+ if(sacked & TCPCB_RETRANS) {
+ if(sacked & TCPCB_SACKED_RETRANS)
+ tp->retrans_out--;
+ acked |= FLAG_RETRANS_DATA_ACKED;
+ }
+ if(sacked & TCPCB_SACKED_ACKED)
+ tp->sacked_out--;
+ if(sacked & TCPCB_LOST)
+ tp->lost_out--;
+ }
+ if(tp->fackets_out)
+ tp->fackets_out--;
tp->packets_out--;
- *seq = scb->seq;
- *seq_rtt = now - scb->when;
+ seq_rtt = now - scb->when;
__skb_unlink(skb, skb->list);
- kfree_skb(skb);
+ tcp_free_skb(sk, skb);
+ }
+
+ if (acked&FLAG_ACKED) {
+ tcp_ack_update_rtt(tp, acked, seq_rtt);
+ tcp_ack_packets_out(sk, tp);
+ }
+
+#if FASTRETRANS_DEBUG > 0
+ BUG_TRAP((int)tp->sacked_out >= 0);
+ BUG_TRAP((int)tp->lost_out >= 0);
+ BUG_TRAP((int)tp->retrans_out >= 0);
+ if (tp->packets_out==0 && tp->sack_ok) {
+ if (tp->lost_out) {
+ printk(KERN_DEBUG "Leak l=%u %d\n", tp->lost_out, tp->ca_state);
+ tp->lost_out = 0;
+ }
+ if (tp->sacked_out) {
+ printk(KERN_DEBUG "Leak s=%u %d\n", tp->sacked_out, tp->ca_state);
+ tp->sacked_out = 0;
+ }
+ if (tp->retrans_out) {
+ printk(KERN_DEBUG "Leak r=%u %d\n", tp->retrans_out, tp->ca_state);
+ tp->retrans_out = 0;
+ }
}
+#endif
return acked;
}
-static void tcp_ack_probe(struct sock *sk, __u32 ack)
+static void tcp_ack_probe(struct sock *sk)
{
struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
-
+
/* Was it a usable window open? */
- if (tp->send_head != NULL) {
- if (!after(TCP_SKB_CB(tp->send_head)->end_seq, ack + tp->snd_wnd)) {
- tp->backoff = 0;
- tcp_clear_xmit_timer(sk, TCP_TIME_PROBE0);
- /* If packets_out==0, socket must be waked up by
- * subsequent tcp_data_snd_check(). This function is
- * not for random using!
- */
- } else if (!tp->packets_out) {
- tcp_reset_xmit_timer(sk, TCP_TIME_PROBE0,
- min(tp->rto << tp->backoff, TCP_RTO_MAX));
- }
+ if (!after(TCP_SKB_CB(tp->send_head)->end_seq, tp->snd_una + tp->snd_wnd)) {
+ tp->backoff = 0;
+ tcp_clear_xmit_timer(sk, TCP_TIME_PROBE0);
+ /* Socket must be waked up by subsequent tcp_data_snd_check().
+ * This function is not for random using!
+ */
+ } else {
+ tcp_reset_xmit_timer(sk, TCP_TIME_PROBE0,
+ min(tp->rto << tp->backoff, TCP_RTO_MAX));
}
}
-/* Should we open up the congestion window? */
-static __inline__ int should_advance_cwnd(struct tcp_opt *tp, int flag)
+static __inline__ int tcp_ack_is_dubious(struct tcp_opt *tp, int flag)
{
- /* Data must have been acked. */
- if ((flag & FLAG_DATA_ACKED) == 0)
- return 0;
-
- /* Some of the data acked was retransmitted somehow? */
- if ((flag & FLAG_RETRANS_DATA_ACKED) != 0) {
- /* We advance in all cases except during
- * non-FACK fast retransmit/recovery.
- */
- if (tp->fackets_out != 0 ||
- tp->retransmits != 0)
- return 1;
+ return (!(flag & FLAG_NOT_DUP) || (flag & FLAG_CA_ALERT) ||
+ tp->ca_state != TCP_CA_Open);
+}
- /* Non-FACK fast retransmit does it's own
- * congestion window management, don't get
- * in the way.
- */
- return 0;
- }
+static __inline__ int tcp_may_raise_cwnd(struct tcp_opt *tp, int flag)
+{
+ return (!(flag & FLAG_ECE) || tp->snd_cwnd < tp->snd_ssthresh) &&
+ !((1<<tp->ca_state)&(TCPF_CA_Recovery|TCPF_CA_CWR));
+}
- /* New non-retransmitted data acked, always advance. */
- return 1;
+/* Check that window update is acceptable.
+ * The function assumes that snd_una<=ack<=snd_next.
+ */
+static __inline__ int
+tcp_may_update_window(struct tcp_opt *tp, u32 ack, u32 ack_seq, u32 nwin)
+{
+ return (after(ack, tp->snd_una) ||
+ after(ack_seq, tp->snd_wl1) ||
+ (ack_seq == tp->snd_wl1 && nwin > tp->snd_wnd));
}
-/* Read draft-ietf-tcplw-high-performance before mucking
- * with this code. (Superceeds RFC1323)
+/* Update our send window.
+ *
+ * Window update algorithm, described in RFC793/RFC1122 (used in linux-2.2
+ * and in FreeBSD. NetBSD's one is even worse.) is wrong.
*/
-static void tcp_ack_saw_tstamp(struct sock *sk, struct tcp_opt *tp,
- u32 seq, u32 ack, int flag)
+static int tcp_ack_update_window(struct sock *sk, struct tcp_opt *tp,
+ struct sk_buff *skb, u32 ack, u32 ack_seq)
{
- __u32 seq_rtt;
+ int flag = 0;
+ u32 nwin = ntohs(skb->h.th->window) << tp->snd_wscale;
- /* RTTM Rule: A TSecr value received in a segment is used to
- * update the averaged RTT measurement only if the segment
- * acknowledges some new data, i.e., only if it advances the
- * left edge of the send window.
- *
- * See draft-ietf-tcplw-high-performance-00, section 3.3.
- * 1998/04/10 Andrey V. Savochkin <saw@msu.ru>
- */
- if (!(flag & (FLAG_DATA_ACKED|FLAG_SYN_ACKED)))
- return;
+ if (tcp_may_update_window(tp, ack, ack_seq, nwin)) {
+ flag |= FLAG_WIN_UPDATE;
+ tcp_update_wl(tp, ack, ack_seq);
- seq_rtt = tcp_time_stamp - tp->rcv_tsecr;
- tcp_rtt_estimator(tp, seq_rtt);
- if (tp->retransmits) {
- if (tp->packets_out == 0) {
- tp->retransmits = 0;
- tp->fackets_out = 0;
- tp->retrans_out = 0;
- tp->backoff = 0;
- tcp_set_rto(tp);
- } else {
- /* Still retransmitting, use backoff */
- tcp_set_rto(tp);
- tp->rto = tp->rto << tp->backoff;
+ if (tp->snd_wnd != nwin) {
+ tp->snd_wnd = nwin;
+
+ /* Note, it is the only place, where
+ * fast path is recovered for sending TCP.
+ */
+ if (skb_queue_len(&tp->out_of_order_queue) == 0 &&
+#ifdef TCP_FORMAL_WINDOW
+ tcp_receive_window(tp) &&
+#endif
+ !tp->urg_data)
+ tcp_fast_path_on(tp);
+
+ if (nwin > tp->max_window) {
+ tp->max_window = nwin;
+ tcp_sync_mss(sk, tp->pmtu_cookie);
+ }
}
- } else {
- tcp_set_rto(tp);
}
- tcp_bound_rto(tp);
-}
-
-static __inline__ void tcp_ack_packets_out(struct sock *sk, struct tcp_opt *tp)
-{
- struct sk_buff *skb = skb_peek(&sk->write_queue);
+ tp->snd_una = ack;
#ifdef TCP_DEBUG
- /* It occured in 2.3, because of racy timers. Namely,
- * retransmit timer did not check packets_out and retransmitted
- * send_head sometimes and, hence, messed all the write_queue.
- * Now it is impossible, I bet. --ANK
- */
- if (skb == NULL) {
- printk("Sucks! packets_out=%d, sk=%p, %d\n", tp->packets_out, sk, sk->state);
- return;
+ if (before(tp->snd_una + tp->snd_wnd, tp->snd_nxt)) {
+ if (net_ratelimit())
+ printk(KERN_DEBUG "TCP: peer shrinks window. Bad, what else can I say?\n");
}
#endif
- /* Some data was ACK'd, if still retransmitting (due to a
- * timeout), resend more of the retransmit queue. The
- * congestion window is handled properly by that code.
- */
- if (tp->retransmits) {
- tcp_xmit_retransmit_queue(sk);
- tcp_reset_xmit_timer(sk, TCP_TIME_RETRANS, tp->rto);
- } else {
- __u32 when = tp->rto - (tcp_time_stamp - TCP_SKB_CB(skb)->when);
- if ((__s32)when < 0)
- when = 1;
- tcp_reset_xmit_timer(sk, TCP_TIME_RETRANS, when);
- }
+ return flag;
}
/* This routine deals with incoming acks, but not outgoing ones. */
-static int tcp_ack(struct sock *sk, struct tcphdr *th,
- u32 ack_seq, u32 ack, int len)
+static int tcp_ack(struct sock *sk, struct sk_buff *skb, int flag)
{
struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
- int flag = 0;
- u32 seq = 0;
- u32 seq_rtt = 0;
-
- if(sk->state == TCP_CLOSE)
- return 1; /* Dead, can't ack any more so why bother */
+ u32 prior_snd_una = tp->snd_una;
+ u32 ack_seq = TCP_SKB_CB(skb)->seq;
+ u32 ack = TCP_SKB_CB(skb)->ack_seq;
+ u32 prior_in_flight;
+ int prior_packets;
/* If the ack is newer than sent or older than previous acks
* then we can probably ignore it.
*/
- if (after(ack, tp->snd_nxt) || before(ack, tp->snd_una))
+ if (after(ack, tp->snd_nxt))
goto uninteresting_ack;
- /* If there is data set flag 1 */
- if (len != th->doff*4)
- flag |= FLAG_DATA;
-
- /* Update our send window. */
-
- /* This is the window update code as per RFC 793
- * snd_wl{1,2} are used to prevent unordered
- * segments from shrinking the window
- */
- if (before(tp->snd_wl1, ack_seq) ||
- (tp->snd_wl1 == ack_seq && !after(tp->snd_wl2, ack))) {
- u32 nwin = ntohs(th->window) << tp->snd_wscale;
-
- if ((tp->snd_wl2 != ack) || (nwin > tp->snd_wnd)) {
- flag |= FLAG_WIN_UPDATE;
- if (tp->snd_wnd != nwin) {
- tp->snd_wnd = nwin;
+ if (before(ack, prior_snd_una))
+ goto old_ack;
- /* Note, it is the only place, where
- * fast path is recovered for sending TCP.
- */
- if (skb_queue_len(&tp->out_of_order_queue) == 0 &&
-#ifdef TCP_FORMAL_WINDOW
- tcp_receive_window(tp) &&
-#endif
- !tp->urg_data)
- tcp_fast_path_on(tp);
+ if (!(flag&FLAG_SLOWPATH) && after(ack, prior_snd_una)) {
+ /* Window is constant, pure forward advance.
+ * No more checks are required.
+ * Note, we use the fact that SND.UNA>=SND.WL2.
+ */
+ tcp_update_wl(tp, ack, ack_seq);
+ tp->snd_una = ack;
+ flag |= FLAG_WIN_UPDATE;
- if (nwin > tp->max_window) {
- tp->max_window = nwin;
- tcp_sync_mss(sk, tp->pmtu_cookie);
- }
- }
+ NET_INC_STATS_BH(TCPHPAcks);
+ } else {
+ if (ack_seq != TCP_SKB_CB(skb)->end_seq)
+ flag |= FLAG_DATA;
+ else
+ NET_INC_STATS_BH(TCPPureAcks);
- tp->snd_wl1 = ack_seq;
- tp->snd_wl2 = ack;
- }
- }
+ flag |= tcp_ack_update_window(sk, tp, skb, ack, ack_seq);
- /* BEWARE! From this place and until return from this function
- * snd_nxt and snd_wnd are out of sync. All the routines, called
- * from here must get "ack" as argument or they should not depend
- * on right edge of window. It is _UGLY_. It cries to be fixed. --ANK
- */
+ if (TCP_SKB_CB(skb)->sacked)
+ flag |= tcp_sacktag_write_queue(sk, skb, prior_snd_una);
+
+ if (TCP_ECN_rcv_ecn_echo(tp, skb->h.th))
+ flag |= FLAG_ECE;
+ }
/* We passed data and got it acked, remove any soft error
* log. Something worked...
*/
sk->err_soft = 0;
- tp->probes_out = 0;
tp->rcv_tstamp = tcp_time_stamp;
+ if ((prior_packets = tp->packets_out) == 0)
+ goto no_queue;
- /* See if we can take anything off of the retransmit queue. */
- flag |= tcp_clean_rtx_queue(sk, ack, &seq, &seq_rtt);
-
- /* If this ack opens up a zero window, clear backoff. It was
- * being used to time the probes, and is probably far higher than
- * it needs to be for normal retransmission.
- */
- if (tcp_timer_is_set(sk, TCP_TIME_PROBE0))
- tcp_ack_probe(sk, ack);
-
- /* We must do this here, before code below clears out important
- * state contained in tp->fackets_out and tp->retransmits. -DaveM
- */
- if (should_advance_cwnd(tp, flag))
- tcp_cong_avoid(tp);
+ prior_in_flight = tcp_packets_in_flight(tp);
- /* If we have a timestamp, we always do rtt estimates. */
- if (tp->saw_tstamp) {
- tcp_ack_saw_tstamp(sk, tp, seq, ack, flag);
- } else {
- /* If we were retransmiting don't count rtt estimate. */
- if (tp->retransmits) {
- if (tp->packets_out == 0) {
- tp->retransmits = 0;
- tp->fackets_out = 0;
- tp->retrans_out = 0;
- }
- } else {
- /* We don't have a timestamp. Can only use
- * packets that are not retransmitted to determine
- * rtt estimates. Also, we must not reset the
- * backoff for rto until we get a non-retransmitted
- * packet. This allows us to deal with a situation
- * where the network delay has increased suddenly.
- * I.e. Karn's algorithm. (SIGCOMM '87, p5.)
- */
- if (flag & (FLAG_DATA_ACKED|FLAG_SYN_ACKED)) {
- if(!(flag & FLAG_RETRANS_DATA_ACKED)) {
- tp->backoff = 0;
- tcp_rtt_estimator(tp, seq_rtt);
- tcp_set_rto(tp);
- tcp_bound_rto(tp);
- }
- }
- }
- }
+ /* See if we can take anything off of the retransmit queue. */
+ flag |= tcp_clean_rtx_queue(sk);
- if (tp->packets_out) {
- if (flag & FLAG_DATA_ACKED)
- tcp_ack_packets_out(sk, tp);
+ if (tcp_ack_is_dubious(tp, flag)) {
+ /* Advanve CWND, if state allows this. */
+ if ((flag&FLAG_DATA_ACKED) && prior_in_flight >= tp->snd_cwnd &&
+ tcp_may_raise_cwnd(tp, flag))
+ tcp_cong_avoid(tp);
+ tcp_fastretrans_alert(sk, prior_snd_una, prior_packets, flag);
} else {
- tcp_clear_xmit_timer(sk, TCP_TIME_RETRANS);
+ if ((flag&FLAG_DATA_ACKED) && prior_in_flight >= tp->snd_cwnd)
+ tcp_cong_avoid(tp);
}
- flag &= (FLAG_DATA | FLAG_WIN_UPDATE);
- if ((ack == tp->snd_una && tp->packets_out && flag == 0) ||
- (tp->high_seq != 0)) {
- tcp_fast_retrans(sk, ack, flag);
- } else {
- /* Clear any aborted fast retransmit starts. */
- tp->dup_acks = 0;
- }
- /* It is not a brain fart, I thought a bit now. 8)
- *
- * Forward progress is indicated, if:
- * 1. the ack acknowledges new data.
- * 2. or the ack is duplicate, but it is caused by new segment
- * arrival. This case is filtered by:
- * - it contains no data, syn or fin.
- * - it does not update window.
- * 3. or new SACK. It is difficult to check, so that we ignore it.
- *
- * Forward progress is also indicated by arrival new data,
- * which was caused by window open from our side. This case is more
- * difficult and it is made (alas, incorrectly) in tcp_data_queue().
- * --ANK (990513)
- */
- if (ack != tp->snd_una || (flag == 0 && !th->fin))
+ if ((flag & FLAG_FORWARD_PROGRESS) || !(flag&FLAG_NOT_DUP))
dst_confirm(sk->dst_cache);
- if (ack != tp->snd_una)
- tp->sorry = 1;
+ return 1;
+
+no_queue:
+ tp->probes_out = 0;
- /* Remember the highest ack received. */
- tp->snd_una = ack;
+ /* If this ack opens up a zero window, clear backoff. It was
+ * being used to time the probes, and is probably far higher than
+ * it needs to be for normal retransmission.
+ */
+ if (tp->send_head)
+ tcp_ack_probe(sk);
return 1;
+old_ack:
+ if (TCP_SKB_CB(skb)->sacked)
+ tcp_sacktag_write_queue(sk, skb, prior_snd_una);
+
uninteresting_ack:
- SOCK_DEBUG(sk, "Ack ignored %u %u\n", ack, tp->snd_nxt);
+ SOCK_DEBUG(sk, "Ack %u out of %u:%u\n", ack, tp->snd_una, tp->snd_nxt);
return 0;
}
-int tcp_paws_check(struct tcp_opt *tp, int rst)
-{
- if ((s32)(tp->rcv_tsval - tp->ts_recent) >= 0)
- return 0;
- if (xtime.tv_sec >= tp->ts_recent_stamp + TCP_PAWS_24DAYS)
- return 0;
-
- /* RST segments are not recommended to carry timestamp,
- and, if they do, it is recommended to ignore PAWS because
- "their cleanup function should take precedence over timestamps."
- Certainly, it is mistake. It is necessary to understand the reasons
- of this constraint to relax it: if peer reboots, clock may go
- out-of-sync and half-open connections will not be reset.
- Actually, the problem would be not existing if all
- the implementations followed draft about maintaining clock
- via reboots. Linux-2.2 DOES NOT!
- However, we can relax time bounds for RST segments to MSL.
- */
- if (rst && xtime.tv_sec >= tp->ts_recent_stamp + TCP_PAWS_MSL)
- return 0;
- return 1;
-}
-
-static __inline__ int tcp_in_window(u32 seq, u32 end_seq, u32 s_win, u32 e_win)
+/* Look for tcp options. Normally only called on SYN and SYNACK packets.
+ * But, this can also be called on packets in the established flow when
+ * the fast version below fails.
+ */
+void tcp_parse_options(struct sk_buff *skb, struct tcp_opt *tp)
{
- if (seq == s_win)
- return 1;
- if (after(end_seq, s_win) && before(seq, e_win))
- return 1;
- return (seq == e_win && seq == end_seq);
-}
+ unsigned char *ptr;
+ struct tcphdr *th = skb->h.th;
+ int length=(th->doff*4)-sizeof(struct tcphdr);
+
+ ptr = (unsigned char *)(th + 1);
+ tp->saw_tstamp = 0;
-/* New-style handling of TIME_WAIT sockets. */
+ while(length>0) {
+ int opcode=*ptr++;
+ int opsize;
-/* Must be called with locally disabled BHs. */
-void tcp_timewait_kill(struct tcp_tw_bucket *tw)
-{
- struct tcp_ehash_bucket *ehead;
- struct tcp_bind_hashbucket *bhead;
- struct tcp_bind_bucket *tb;
+ switch (opcode) {
+ case TCPOPT_EOL:
+ return;
+ case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
+ length--;
+ continue;
+ default:
+ opsize=*ptr++;
+ if (opsize < 2) /* "silly options" */
+ return;
+ if (opsize > length)
+ break; /* don't parse partial options */
+ switch(opcode) {
+ case TCPOPT_MSS:
+ if(opsize==TCPOLEN_MSS && th->syn) {
+ u16 in_mss = ntohs(*(__u16 *)ptr);
+ if (in_mss) {
+ if (tp->user_mss && tp->user_mss < in_mss)
+ in_mss = tp->user_mss;
+ tp->mss_clamp = in_mss;
+ }
+ }
+ break;
+ case TCPOPT_WINDOW:
+ if(opsize==TCPOLEN_WINDOW && th->syn)
+ if (sysctl_tcp_window_scaling) {
+ tp->wscale_ok = 1;
+ tp->snd_wscale = *(__u8 *)ptr;
+ if(tp->snd_wscale > 14) {
+ if(net_ratelimit())
+ printk("tcp_parse_options: Illegal window "
+ "scaling value %d >14 received.",
+ tp->snd_wscale);
+ tp->snd_wscale = 14;
+ }
+ }
+ break;
+ case TCPOPT_TIMESTAMP:
+ if(opsize==TCPOLEN_TIMESTAMP) {
+ if (sysctl_tcp_timestamps) {
+ tp->tstamp_ok = 1;
+ tp->saw_tstamp = 1;
+ tp->rcv_tsval = ntohl(*(__u32 *)ptr);
+ tp->rcv_tsecr = ntohl(*(__u32 *)(ptr+4));
+ }
+ }
+ break;
+ case TCPOPT_SACK_PERM:
+ if(opsize==TCPOLEN_SACK_PERM && th->syn) {
+ if (sysctl_tcp_sack) {
+ tp->sack_ok = 1;
+ tcp_sack_reset(tp);
+ }
+ }
+ break;
- /* Unlink from established hashes. */
- ehead = &tcp_ehash[tw->hashent];
- write_lock(&ehead->lock);
- if (!tw->pprev) {
- write_unlock(&ehead->lock);
- return;
- }
- if(tw->next)
- tw->next->pprev = tw->pprev;
- *(tw->pprev) = tw->next;
- tw->pprev = NULL;
- write_unlock(&ehead->lock);
-
- /* Disassociate with bind bucket. */
- bhead = &tcp_bhash[tcp_bhashfn(tw->num)];
- spin_lock(&bhead->lock);
- if ((tb = tw->tb) != NULL) {
- if(tw->bind_next)
- tw->bind_next->bind_pprev = tw->bind_pprev;
- *(tw->bind_pprev) = tw->bind_next;
- tw->tb = NULL;
- if (tb->owners == NULL) {
- if (tb->next)
- tb->next->pprev = tb->pprev;
- *(tb->pprev) = tb->next;
- kmem_cache_free(tcp_bucket_cachep, tb);
- }
- }
- spin_unlock(&bhead->lock);
-
-#ifdef INET_REFCNT_DEBUG
- if (atomic_read(&tw->refcnt) != 1) {
- printk(KERN_DEBUG "tw_bucket %p refcnt=%d\n", tw, atomic_read(&tw->refcnt));
+ case TCPOPT_SACK:
+ if((opsize >= (TCPOLEN_SACK_BASE + TCPOLEN_SACK_PERBLOCK)) &&
+ !((opsize - TCPOLEN_SACK_BASE) % TCPOLEN_SACK_PERBLOCK) &&
+ tp->sack_ok) {
+ TCP_SKB_CB(skb)->sacked = (ptr - 2) - (unsigned char *)th;
+ }
+ };
+ ptr+=opsize-2;
+ length-=opsize;
+ };
}
-#endif
- tcp_tw_put(tw);
}
-/*
- * * Main purpose of TIME-WAIT state is to close connection gracefully,
- * when one of ends sits in LAST-ACK or CLOSING retransmitting FIN
- * (and, probably, tail of data) and one or more our ACKs are lost.
- * * What is TIME-WAIT timeout? It is associated with maximal packet
- * lifetime in the internet, which results in wrong conclusion, that
- * it is set to catch "old duplicate segments" wandering out of their path.
- * It is not quite correct. This timeout is calculated so that it exceeds
- * maximal retransmision timeout enough to allow to lose one (or more)
- * segments sent by peer and our ACKs. This time may be calculated from RTO.
- * * When TIME-WAIT socket receives RST, it means that another end
- * finally closed and we are allowed to kill TIME-WAIT too.
- * * Second purpose of TIME-WAIT is catching old duplicate segments.
- * Well, certainly it is pure paranoia, but if we load TIME-WAIT
- * with this semantics, we MUST NOT kill TIME-WAIT state with RSTs.
- * * If we invented some more clever way to catch duplicates
- * (f.e. based on PAWS), we could truncate TIME-WAIT to several RTOs.
- *
- * The algorithm below is based on FORMAL INTERPRETATION of RFCs.
- * When you compare it to RFCs, please, read section SEGMENT ARRIVES
- * from the very beginning.
- *
- * NOTE. With recycling (and later with fin-wait-2) TW bucket
- * is _not_ stateless. It means, that strictly speaking we must
- * spinlock it. I do not want! Well, probability of misbehaviour
- * is ridiculously low and, seems, we could use some mb() tricks
- * to avoid misread sequence numbers, states etc. --ANK
- */
-enum tcp_tw_status
-tcp_timewait_state_process(struct tcp_tw_bucket *tw, struct sk_buff *skb,
- struct tcphdr *th, unsigned len)
-{
- struct tcp_opt tp;
- int paws_reject = 0;
-
- tp.saw_tstamp = 0;
- if (th->doff > (sizeof(struct tcphdr)>>2) && tw->ts_recent_stamp) {
- tcp_parse_options(NULL, th, &tp, 0);
-
- if (tp.saw_tstamp) {
- tp.ts_recent = tw->ts_recent;
- tp.ts_recent_stamp = tw->ts_recent_stamp;
- paws_reject = tcp_paws_check(&tp, th->rst);
- }
- }
-
- if (tw->substate == TCP_FIN_WAIT2) {
- /* Just repeat all the checks of tcp_rcv_state_process() */
-
- /* Out of window, send ACK */
- if (paws_reject ||
- !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
- tw->rcv_nxt, tw->rcv_nxt + tw->rcv_wnd))
- return TCP_TW_ACK;
-
- if (th->rst)
- goto kill;
-
- if (th->syn && TCP_SKB_CB(skb)->seq != tw->syn_seq)
- goto kill_with_rst;
-
- /* Dup ACK? */
- if (!after(TCP_SKB_CB(skb)->end_seq, tw->rcv_nxt) ||
- TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) {
- tcp_tw_put(tw);
- return TCP_TW_SUCCESS;
- }
-
- /* New data or FIN. If new data arrive after half-duplex close,
- * reset.
- */
- if (!th->fin || TCP_SKB_CB(skb)->end_seq != tw->rcv_nxt+1) {
-kill_with_rst:
- tcp_tw_deschedule(tw);
- tcp_timewait_kill(tw);
- tcp_tw_put(tw);
- return TCP_TW_RST;
- }
-
- /* FIN arrived, enter true time-wait state. */
- tw->substate = TCP_TIME_WAIT;
- tw->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
- if (tp.saw_tstamp) {
- tw->ts_recent_stamp = xtime.tv_sec;
- tw->ts_recent = tp.rcv_tsval;
- }
-
- /* I am shamed, but failed to make it more elegant.
- * Yes, it is direct reference to IP, which is impossible
- * to generalize to IPv6. Taking into account that IPv6
- * do not undertsnad recycling in any case, it not
- * a big problem in practice. --ANK */
- if (tw->family == AF_INET &&
- sysctl_tcp_tw_recycle && tw->ts_recent_stamp &&
- tcp_v4_tw_remember_stamp(tw))
- tcp_tw_schedule(tw, tw->timeout);
- else
- tcp_tw_schedule(tw, TCP_TIMEWAIT_LEN);
- return TCP_TW_ACK;
- }
-
- /*
- * Now real TIME-WAIT state.
- *
- * RFC 1122:
- * "When a connection is [...] on TIME-WAIT state [...]
- * [a TCP] MAY accept a new SYN from the remote TCP to
- * reopen the connection directly, if it:
- *
- * (1) assigns its initial sequence number for the new
- * connection to be larger than the largest sequence
- * number it used on the previous connection incarnation,
- * and
- *
- * (2) returns to TIME-WAIT state if the SYN turns out
- * to be an old duplicate".
- */
-
- if (!paws_reject &&
- (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq &&
- TCP_SKB_CB(skb)->seq == tw->rcv_nxt)) {
- /* In window segment, it may be only reset or bare ack. */
-
- if (th->rst) {
- /* This is TIME_WAIT assasination, in two flavors.
- * Oh well... nobody has a sufficient solution to this
- * protocol bug yet.
- */
- if (sysctl_tcp_rfc1337 == 0) {
-kill:
- tcp_tw_deschedule(tw);
- tcp_timewait_kill(tw);
- tcp_tw_put(tw);
- return TCP_TW_SUCCESS;
- }
- }
- tcp_tw_schedule(tw, TCP_TIMEWAIT_LEN);
-
- if (tp.saw_tstamp) {
- tw->ts_recent = tp.rcv_tsval;
- tw->ts_recent_stamp = xtime.tv_sec;
+/* Fast parse options. This hopes to only see timestamps.
+ * If it is wrong it falls back on tcp_parse_options().
+ */
+static __inline__ int tcp_fast_parse_options(struct sk_buff *skb, struct tcphdr *th, struct tcp_opt *tp)
+{
+ if (th->doff == sizeof(struct tcphdr)>>2) {
+ tp->saw_tstamp = 0;
+ return 0;
+ } else if (th->doff == (sizeof(struct tcphdr)>>2)+(TCPOLEN_TSTAMP_ALIGNED>>2)) {
+ __u32 *ptr = (__u32 *)(th + 1);
+ if (*ptr == __constant_ntohl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16)
+ | (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP)) {
+ tp->saw_tstamp = 1;
+ ++ptr;
+ tp->rcv_tsval = ntohl(*ptr);
+ ++ptr;
+ tp->rcv_tsecr = ntohl(*ptr);
+ return 1;
}
-
- tcp_tw_put(tw);
- return TCP_TW_SUCCESS;
- }
-
- /* Out of window segment.
-
- All the segments are ACKed immediately.
-
- The only exception is new SYN. We accept it, if it is
- not old duplicate and we are not in danger to be killed
- by delayed old duplicates. RFC check is that it has
- newer sequence number works at rates <40Mbit/sec.
- However, if paws works, it is reliable AND even more,
- we even may relax silly seq space cutoff.
-
- RED-PEN: we violate main RFC requirement, if this SYN will appear
- old duplicate (i.e. we receive RST in reply to SYN-ACK),
- we must return socket to time-wait state. It is not good,
- but not fatal yet.
- */
-
- if (th->syn && !th->rst && !th->ack && !paws_reject &&
- (after(TCP_SKB_CB(skb)->seq, tw->rcv_nxt) ||
- (tp.saw_tstamp && (s32)(tw->ts_recent - tp.rcv_tsval) < 0))) {
- u32 isn = tw->snd_nxt + 2;
- if (isn == 0)
- isn++;
- TCP_SKB_CB(skb)->when = isn;
- return TCP_TW_SYN;
}
+ tcp_parse_options(skb, tp);
+ return 1;
+}
- if (paws_reject)
- NET_INC_STATS_BH(PAWSEstabRejected);
+extern __inline__ void
+tcp_store_ts_recent(struct tcp_opt *tp)
+{
+ tp->ts_recent = tp->rcv_tsval;
+ tp->ts_recent_stamp = xtime.tv_sec;
+}
- if(!th->rst) {
- /* In this case we must reset the TIMEWAIT timer.
+extern __inline__ void
+tcp_replace_ts_recent(struct tcp_opt *tp, u32 seq)
+{
+ if (tp->saw_tstamp && !after(seq, tp->rcv_wup)) {
+ /* PAWS bug workaround wrt. ACK frames, the PAWS discard
+ * extra check below makes sure this can only happen
+ * for pure ACK frames. -DaveM
*
- * If it is ACKless SYN it may be both old duplicate
- * and new good SYN with random sequence number <rcv_nxt.
- * Do not reschedule in the last case.
+ * Not only, also it occurs for expired timestamps.
*/
- if (paws_reject || th->ack)
- tcp_tw_schedule(tw, TCP_TIMEWAIT_LEN);
- /* Send ACK. Note, we do not put the bucket,
- * it will be released by caller.
- */
- return TCP_TW_ACK;
+ if((s32)(tp->rcv_tsval - tp->ts_recent) >= 0 ||
+ xtime.tv_sec >= tp->ts_recent_stamp + TCP_PAWS_24DAYS)
+ tcp_store_ts_recent(tp);
}
- tcp_tw_put(tw);
- return TCP_TW_SUCCESS;
}
-/* Enter the time wait state. This is called with locally disabled BH.
- * Essentially we whip up a timewait bucket, copy the
- * relevant info into it from the SK, and mess with hash chains
- * and list linkage.
+/* Sorry, PAWS as specified is broken wrt. pure-ACKs -DaveM
+ *
+ * It is not fatal. If this ACK does _not_ change critical state (seqs, window)
+ * it can pass through stack. So, the following predicate verifies that
+ * this segment is not used for anything but congestion avoidance or
+ * fast retransmit. Moreover, we even are able to eliminate most of such
+ * second order effects, if we apply some small "replay" window (~RTO)
+ * to timestamp space.
+ *
+ * All these measures still do not guarantee that we reject wrapped ACKs
+ * on networks with high bandwidth, when sequence space is recycled fastly,
+ * but it guarantees that such events will be very rare and do not affect
+ * connection seriously. This doesn't look nice, but alas, PAWS is really
+ * buggy extension.
+ *
+ * [ Later note. Even worse! It is buggy for segments _with_ data. RFC
+ * states that events when retransmit arrives after original data are rare.
+ * It is a blatant lie. VJ forgot about fast retransmit! 8)8) It is
+ * the biggest problem on large power networks even with minor reordering.
+ * OK, let's give it small replay window. If peer clock is even 1hz, it is safe
+ * up to bandwidth of 18Gigabit/sec. 8) ]
*/
-static void __tcp_tw_hashdance(struct sock *sk, struct tcp_tw_bucket *tw)
+
+static int tcp_disordered_ack(struct tcp_opt *tp, struct sk_buff *skb)
{
- struct tcp_ehash_bucket *ehead = &tcp_ehash[sk->hashent];
- struct tcp_bind_hashbucket *bhead;
- struct sock **head, *sktw;
+ struct tcphdr *th = skb->h.th;
+ u32 seq = TCP_SKB_CB(skb)->seq;
+ u32 ack = TCP_SKB_CB(skb)->ack_seq;
- write_lock(&ehead->lock);
+ return (/* 1. Pure ACK with correct sequence number. */
+ (th->ack && seq == TCP_SKB_CB(skb)->end_seq && seq == tp->rcv_nxt) &&
- /* Step 1: Remove SK from established hash. */
- if (sk->pprev) {
- if(sk->next)
- sk->next->pprev = sk->pprev;
- *sk->pprev = sk->next;
- sk->pprev = NULL;
- sock_prot_dec_use(sk->prot);
- }
+ /* 2. ... and duplicate ACK. */
+ ack == tp->snd_una &&
- /* Step 2: Hash TW into TIMEWAIT half of established hash table. */
- head = &(ehead + tcp_ehash_size)->chain;
- sktw = (struct sock *)tw;
- if((sktw->next = *head) != NULL)
- (*head)->pprev = &sktw->next;
- *head = sktw;
- sktw->pprev = head;
- atomic_inc(&tw->refcnt);
+ /* 3. ... and does not update window. */
+ !tcp_may_update_window(tp, ack, seq, ntohs(th->window)<<tp->snd_wscale) &&
- write_unlock(&ehead->lock);
+ /* 4. ... and sits in replay window. */
+ (s32)(tp->ts_recent - tp->rcv_tsval) <= (tp->rto*1024)/HZ);
+}
- /* Step 3: Put TW into bind hash. Original socket stays there too.
- Note, that any socket with sk->num!=0 MUST be bound in binding
- cache, even if it is closed.
- */
- bhead = &tcp_bhash[tcp_bhashfn(sk->num)];
- spin_lock(&bhead->lock);
- tw->tb = (struct tcp_bind_bucket *)sk->prev;
- BUG_TRAP(sk->prev!=NULL);
- if ((tw->bind_next = tw->tb->owners) != NULL)
- tw->tb->owners->bind_pprev = &tw->bind_next;
- tw->tb->owners = (struct sock*)tw;
- tw->bind_pprev = &tw->tb->owners;
- spin_unlock(&bhead->lock);
+extern __inline__ int tcp_paws_discard(struct tcp_opt *tp, struct sk_buff *skb)
+{
+ return ((s32)(tp->ts_recent - tp->rcv_tsval) > TCP_PAWS_WINDOW &&
+ xtime.tv_sec < tp->ts_recent_stamp + TCP_PAWS_24DAYS &&
+ !tcp_disordered_ack(tp, skb));
}
-/*
- * Move a socket to time-wait or dead fin-wait-2 state.
- */
-void tcp_time_wait(struct sock *sk, int state, int timeo)
+static int __tcp_sequence(struct tcp_opt *tp, u32 seq, u32 end_seq)
{
- struct tcp_tw_bucket *tw = NULL;
- struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
- int recycle_ok = 0;
+ u32 end_window = tp->rcv_wup + tp->rcv_wnd;
+#ifdef TCP_FORMAL_WINDOW
+ u32 rcv_wnd = tcp_receive_window(tp);
+#else
+ u32 rcv_wnd = tp->rcv_wnd;
+#endif
- if (sysctl_tcp_tw_recycle && tp->ts_recent_stamp)
- recycle_ok = tp->af_specific->remember_stamp(sk);
+ if (rcv_wnd &&
+ after(end_seq, tp->rcv_nxt) &&
+ before(seq, end_window))
+ return 1;
+ if (seq != end_window)
+ return 0;
+ return (seq == end_seq);
+}
- if (tcp_tw_count < sysctl_tcp_max_tw_buckets)
- tw = kmem_cache_alloc(tcp_timewait_cachep, SLAB_ATOMIC);
-
- if(tw != NULL) {
- int rto = (tp->rto<<2) - (tp->rto>>1);
-
- /* Give us an identity. */
- tw->daddr = sk->daddr;
- tw->rcv_saddr = sk->rcv_saddr;
- tw->bound_dev_if= sk->bound_dev_if;
- tw->num = sk->num;
- tw->state = TCP_TIME_WAIT;
- tw->substate = state;
- tw->sport = sk->sport;
- tw->dport = sk->dport;
- tw->family = sk->family;
- tw->reuse = sk->reuse;
- tw->rcv_wscale = tp->rcv_wscale;
- atomic_set(&tw->refcnt, 0);
-
- tw->hashent = sk->hashent;
- tw->rcv_nxt = tp->rcv_nxt;
- tw->snd_nxt = tp->snd_nxt;
- tw->rcv_wnd = tcp_receive_window(tp);
- tw->syn_seq = tp->syn_seq;
- tw->ts_recent = tp->ts_recent;
- tw->ts_recent_stamp= tp->ts_recent_stamp;
- tw->pprev_death = NULL;
-
-#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
- if(tw->family == PF_INET6) {
- memcpy(&tw->v6_daddr,
- &sk->net_pinfo.af_inet6.daddr,
- sizeof(struct in6_addr));
- memcpy(&tw->v6_rcv_saddr,
- &sk->net_pinfo.af_inet6.rcv_saddr,
- sizeof(struct in6_addr));
- }
+/* This functions checks to see if the tcp header is actually acceptable.
+ *
+ * Actually, our check is seriously broken, we must accept RST,ACK,URG
+ * even on zero window effectively trimming data. It is RFC, guys.
+ * But our check is so beautiful, that I do not want to repair it
+ * now. However, taking into account those stupid plans to start to
+ * send some texts with RST, we have to handle at least this case. --ANK
+ */
+extern __inline__ int tcp_sequence(struct tcp_opt *tp, u32 seq, u32 end_seq, int rst)
+{
+#ifdef TCP_FORMAL_WINDOW
+ u32 rcv_wnd = tcp_receive_window(tp);
+#else
+ u32 rcv_wnd = tp->rcv_wnd;
#endif
- /* Linkage updates. */
- __tcp_tw_hashdance(sk, tw);
-
- /* Get the TIME_WAIT timeout firing. */
- if (timeo < rto)
- timeo = rto;
+ if (seq == tp->rcv_nxt)
+ return (rcv_wnd || (end_seq == seq) || rst);
- if (recycle_ok) {
- tw->timeout = rto;
- } else {
- tw->timeout = TCP_TIMEWAIT_LEN;
- if (state == TCP_TIME_WAIT)
- timeo = TCP_TIMEWAIT_LEN;
- }
+ return __tcp_sequence(tp, seq, end_seq);
+}
- tcp_tw_schedule(tw, timeo);
- } else {
- /* Sorry, if we're out of memory, just CLOSE this
- * socket up. We've got bigger problems than
- * non-graceful socket closings.
- */
- if (net_ratelimit())
- printk(KERN_INFO "TCP: time wait bucket table overflow\n");
+/* When we get a reset we do this. */
+static void tcp_reset(struct sock *sk)
+{
+ /* We want the right error as BSD sees it (and indeed as we do). */
+ switch (sk->state) {
+ case TCP_SYN_SENT:
+ sk->err = ECONNREFUSED;
+ break;
+ case TCP_CLOSE_WAIT:
+ sk->err = EPIPE;
+ break;
+ case TCP_CLOSE:
+ return;
+ default:
+ sk->err = ECONNRESET;
}
- tcp_update_metrics(sk);
+ if (!sk->dead)
+ sk->error_report(sk);
+
tcp_done(sk);
}
@@ -1611,22 +2179,22 @@
*
* If we are in FINWAIT-2, a received FIN moves us to TIME-WAIT.
*/
-
static void tcp_fin(struct sk_buff *skb, struct sock *sk, struct tcphdr *th)
{
struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
tp->fin_seq = TCP_SKB_CB(skb)->end_seq;
- tp->ack.pending = 1;
- tp->ack.quick = 0;
+ tcp_schedule_ack(tp);
sk->shutdown |= RCV_SHUTDOWN;
+ sk->done = 1;
switch(sk->state) {
case TCP_SYN_RECV:
case TCP_ESTABLISHED:
/* Move to CLOSE_WAIT */
tcp_set_state(sk, TCP_CLOSE_WAIT);
+ tp->ack.pingpong = 1;
break;
case TCP_CLOSE_WAIT:
@@ -1644,6 +2212,7 @@
* happens, we must ack the received FIN and
* enter the CLOSING state.
*/
+ tcp_send_ack(sk);
tcp_set_state(sk, TCP_CLOSING);
break;
case TCP_FIN_WAIT2:
@@ -1664,7 +2233,8 @@
*/
__skb_queue_purge(&tp->out_of_order_queue);
if (tp->sack_ok)
- tp->num_sacks = 0;
+ tcp_sack_reset(tp);
+ tcp_mem_reclaim(sk);
if (!sk->dead) {
sk->state_change(sk);
@@ -1677,51 +2247,90 @@
}
}
+static __inline__ int
+tcp_sack_extend(struct tcp_sack_block *sp, u32 seq, u32 end_seq)
+{
+ if (!after(seq, sp->end_seq) && !after(sp->start_seq, end_seq)) {
+ if (before(seq, sp->start_seq))
+ sp->start_seq = seq;
+ if (after(end_seq, sp->end_seq))
+ sp->end_seq = end_seq;
+ return 1;
+ }
+ return 0;
+}
+
+static __inline__ void tcp_dsack_set(struct tcp_opt *tp, u32 seq, u32 end_seq)
+{
+ if (tp->sack_ok && sysctl_tcp_dsack) {
+ if (before(seq, tp->rcv_nxt))
+ NET_INC_STATS_BH(TCPDSACKOldSent);
+ else
+ NET_INC_STATS_BH(TCPDSACKOfoSent);
+
+ tp->dsack = 1;
+ tp->duplicate_sack[0].start_seq = seq;
+ tp->duplicate_sack[0].end_seq = end_seq;
+ tp->eff_sacks = min(tp->num_sacks+1, 4-tp->tstamp_ok);
+ }
+}
+
+static __inline__ void tcp_dsack_extend(struct tcp_opt *tp, u32 seq, u32 end_seq)
+{
+ if (!tp->dsack)
+ tcp_dsack_set(tp, seq, end_seq);
+ else
+ tcp_sack_extend(tp->duplicate_sack, seq, end_seq);
+}
+
+static void tcp_send_dupack(struct sock *sk, struct sk_buff *skb)
+{
+ struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
+
+ if (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq &&
+ before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) {
+ NET_INC_STATS_BH(DelayedACKLost);
+ tcp_enter_quickack_mode(tp);
+
+ if (tp->sack_ok && sysctl_tcp_dsack) {
+ u32 end_seq = TCP_SKB_CB(skb)->end_seq;
+
+ if (after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt))
+ end_seq = tp->rcv_nxt;
+ tcp_dsack_set(tp, TCP_SKB_CB(skb)->seq, end_seq);
+ }
+ }
+
+ tcp_send_ack(sk);
+}
+
/* These routines update the SACK block as out-of-order packets arrive or
* in-order packets close up the sequence space.
*/
-static void tcp_sack_maybe_coalesce(struct tcp_opt *tp, struct tcp_sack_block *sp)
+static void tcp_sack_maybe_coalesce(struct tcp_opt *tp)
{
- int this_sack, num_sacks = tp->num_sacks;
- struct tcp_sack_block *swalk = &tp->selective_acks[0];
+ int this_sack;
+ struct tcp_sack_block *sp = &tp->selective_acks[0];
+ struct tcp_sack_block *swalk = sp+1;
- /* If more than one SACK block, see if the recent change to SP eats into
+ /* See if the recent change to the first SACK eats into
* or hits the sequence space of other SACK blocks, if so coalesce.
*/
- if(num_sacks != 1) {
- for(this_sack = 0; this_sack < num_sacks; this_sack++, swalk++) {
- if(swalk == sp)
- continue;
+ for (this_sack = 1; this_sack < tp->num_sacks; ) {
+ if (tcp_sack_extend(sp, swalk->start_seq, swalk->end_seq)) {
+ int i;
- /* First case, bottom of SP moves into top of the
- * sequence space of SWALK.
- */
- if(between(sp->start_seq, swalk->start_seq, swalk->end_seq)) {
- sp->start_seq = swalk->start_seq;
- goto coalesce;
- }
- /* Second case, top of SP moves into bottom of the
- * sequence space of SWALK.
+ /* Zap SWALK, by moving every further SACK up by one slot.
+ * Decrease num_sacks.
*/
- if(between(sp->end_seq, swalk->start_seq, swalk->end_seq)) {
- sp->end_seq = swalk->end_seq;
- goto coalesce;
- }
+ tp->num_sacks--;
+ tp->eff_sacks = min(tp->num_sacks+tp->dsack, 4-tp->tstamp_ok);
+ for(i=this_sack; i < tp->num_sacks; i++)
+ sp[i] = sp[i+1];
+ continue;
}
+ this_sack++, swalk++;
}
- /* SP is the only SACK, or no coalescing cases found. */
- return;
-
-coalesce:
- /* Zap SWALK, by moving every further SACK up by one slot.
- * Decrease num_sacks.
- */
- for(; this_sack < num_sacks-1; this_sack++, swalk++) {
- struct tcp_sack_block *next = (swalk + 1);
- swalk->start_seq = next->start_seq;
- swalk->end_seq = next->end_seq;
- }
- tp->num_sacks--;
}
static __inline__ void tcp_sack_swap(struct tcp_sack_block *sack1, struct tcp_sack_block *sack2)
@@ -1737,151 +2346,117 @@
sack2->end_seq = tmp;
}
-static void tcp_sack_new_ofo_skb(struct sock *sk, struct sk_buff *skb)
+static void tcp_sack_new_ofo_skb(struct sock *sk, u32 seq, u32 end_seq)
{
struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
struct tcp_sack_block *sp = &tp->selective_acks[0];
int cur_sacks = tp->num_sacks;
+ int this_sack;
if (!cur_sacks)
goto new_sack;
- /* Optimize for the common case, new ofo frames arrive
- * "in order". ;-) This also satisfies the requirements
- * of RFC2018 about ordering of SACKs.
- */
- if(sp->end_seq == TCP_SKB_CB(skb)->seq) {
- sp->end_seq = TCP_SKB_CB(skb)->end_seq;
- tcp_sack_maybe_coalesce(tp, sp);
- } else if(sp->start_seq == TCP_SKB_CB(skb)->end_seq) {
- /* Re-ordered arrival, in this case, can be optimized
- * as well.
- */
- sp->start_seq = TCP_SKB_CB(skb)->seq;
- tcp_sack_maybe_coalesce(tp, sp);
- } else {
- struct tcp_sack_block *swap = sp + 1;
- int this_sack, max_sacks = (tp->tstamp_ok ? 3 : 4);
-
- /* Oh well, we have to move things around.
- * Try to find a SACK we can tack this onto.
- */
-
- for(this_sack = 1; this_sack < cur_sacks; this_sack++, swap++) {
- if((swap->end_seq == TCP_SKB_CB(skb)->seq) ||
- (swap->start_seq == TCP_SKB_CB(skb)->end_seq)) {
- if(swap->end_seq == TCP_SKB_CB(skb)->seq)
- swap->end_seq = TCP_SKB_CB(skb)->end_seq;
- else
- swap->start_seq = TCP_SKB_CB(skb)->seq;
- tcp_sack_swap(sp, swap);
- tcp_sack_maybe_coalesce(tp, sp);
- return;
- }
- }
-
- /* Could not find an adjacent existing SACK, build a new one,
- * put it at the front, and shift everyone else down. We
- * always know there is at least one SACK present already here.
- *
- * If the sack array is full, forget about the last one.
- */
- if (cur_sacks >= max_sacks) {
- cur_sacks--;
- tp->num_sacks--;
- }
- while(cur_sacks >= 1) {
- struct tcp_sack_block *this = &tp->selective_acks[cur_sacks];
- struct tcp_sack_block *prev = (this - 1);
- this->start_seq = prev->start_seq;
- this->end_seq = prev->end_seq;
- cur_sacks--;
+ for (this_sack=0; this_sack<cur_sacks; this_sack++, sp++) {
+ if (tcp_sack_extend(sp, seq, end_seq)) {
+ /* Rotate this_sack to the first one. */
+ for (; this_sack>0; this_sack--, sp--)
+ tcp_sack_swap(sp, sp-1);
+ if (cur_sacks > 1)
+ tcp_sack_maybe_coalesce(tp);
+ return;
}
+ }
- new_sack:
- /* Build the new head SACK, and we're done. */
- sp->start_seq = TCP_SKB_CB(skb)->seq;
- sp->end_seq = TCP_SKB_CB(skb)->end_seq;
- tp->num_sacks++;
+ /* Could not find an adjacent existing SACK, build a new one,
+ * put it at the front, and shift everyone else down. We
+ * always know there is at least one SACK present already here.
+ *
+ * If the sack array is full, forget about the last one.
+ */
+ if (this_sack >= 4) {
+ this_sack--;
+ tp->num_sacks--;
+ sp--;
}
+ for(; this_sack > 0; this_sack--, sp--)
+ *sp = *(sp-1);
+
+new_sack:
+ /* Build the new head SACK, and we're done. */
+ sp->start_seq = seq;
+ sp->end_seq = end_seq;
+ tp->num_sacks++;
+ tp->eff_sacks = min(tp->num_sacks+tp->dsack, 4-tp->tstamp_ok);
}
-static void tcp_sack_remove_skb(struct tcp_opt *tp, struct sk_buff *skb)
+/* RCV.NXT advances, some SACKs should be eaten. */
+
+static void tcp_sack_remove(struct tcp_opt *tp)
{
struct tcp_sack_block *sp = &tp->selective_acks[0];
int num_sacks = tp->num_sacks;
int this_sack;
- /* This is an in order data segment _or_ an out-of-order SKB being
- * moved to the receive queue, so we know this removed SKB will eat
- * from the front of a SACK.
- */
- for(this_sack = 0; this_sack < num_sacks; this_sack++, sp++) {
- /* Check if the start of the sack is covered by skb. */
- if(!before(sp->start_seq, TCP_SKB_CB(skb)->seq) &&
- before(sp->start_seq, TCP_SKB_CB(skb)->end_seq))
- break;
- }
-
- /* This should only happen if so many SACKs get built that some get
- * pushed out before we get here, or we eat some in sequence packets
- * which are before the first SACK block.
- */
- if(this_sack >= num_sacks)
+ /* Empty ofo queue, hence, all the SACKs are eaten. Clear. */
+ if (skb_queue_len(&tp->out_of_order_queue) == 0) {
+ tp->num_sacks = 0;
+ tp->eff_sacks = tp->dsack;
return;
+ }
- sp->start_seq = TCP_SKB_CB(skb)->end_seq;
- if(!before(sp->start_seq, sp->end_seq)) {
- /* Zap this SACK, by moving forward any other SACKS. */
- for(this_sack += 1; this_sack < num_sacks; this_sack++, sp++) {
- struct tcp_sack_block *next = (sp + 1);
- sp->start_seq = next->start_seq;
- sp->end_seq = next->end_seq;
+ for(this_sack = 0; this_sack < num_sacks; ) {
+ /* Check if the start of the sack is covered by RCV.NXT. */
+ if (!before(tp->rcv_nxt, sp->start_seq)) {
+ int i;
+
+ /* RCV.NXT must cover all the block! */
+ BUG_TRAP(!before(tp->rcv_nxt, sp->end_seq));
+
+ /* Zap this SACK, by moving forward any other SACKS. */
+ for (i=this_sack+1; i < num_sacks; i++)
+ sp[i-1] = sp[i];
+ num_sacks--;
+ continue;
}
- tp->num_sacks--;
+ this_sack++;
+ sp++;
}
-}
-
-static void tcp_sack_extend(struct tcp_opt *tp, struct sk_buff *old_skb, struct sk_buff *new_skb)
-{
- struct tcp_sack_block *sp = &tp->selective_acks[0];
- int num_sacks = tp->num_sacks;
- int this_sack;
-
- for(this_sack = 0; this_sack < num_sacks; this_sack++, sp++) {
- if(sp->end_seq == TCP_SKB_CB(old_skb)->end_seq)
- break;
+ if (num_sacks != tp->num_sacks) {
+ tp->num_sacks = num_sacks;
+ tp->eff_sacks = min(tp->num_sacks+tp->dsack, 4-tp->tstamp_ok);
}
- if(this_sack >= num_sacks)
- return;
- sp->end_seq = TCP_SKB_CB(new_skb)->end_seq;
}
-
/* This one checks to see if we can put data from the
* out_of_order queue into the receive_queue.
*/
static void tcp_ofo_queue(struct sock *sk)
{
- struct sk_buff *skb;
struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
+ __u32 dsack_high = tp->rcv_nxt;
+ struct sk_buff *skb;
- while ((skb = skb_peek(&tp->out_of_order_queue))) {
+ while ((skb = skb_peek(&tp->out_of_order_queue)) != NULL) {
if (after(TCP_SKB_CB(skb)->seq, tp->rcv_nxt))
break;
+ if (before(TCP_SKB_CB(skb)->seq, dsack_high)) {
+ __u32 dsack = dsack_high;
+ if (before(TCP_SKB_CB(skb)->end_seq, dsack_high))
+ dsack_high = TCP_SKB_CB(skb)->end_seq;
+ tcp_dsack_extend(tp, TCP_SKB_CB(skb)->seq, dsack);
+ }
+
if (!after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt)) {
SOCK_DEBUG(sk, "ofo packet was already received \n");
__skb_unlink(skb, skb->list);
- kfree_skb(skb);
+ __kfree_skb(skb);
continue;
}
SOCK_DEBUG(sk, "ofo requeuing : rcv_next %X seq %X - %X\n",
tp->rcv_nxt, TCP_SKB_CB(skb)->seq,
TCP_SKB_CB(skb)->end_seq);
- if(tp->sack_ok)
- tcp_sack_remove_skb(tp, skb);
__skb_unlink(skb, skb->list);
__skb_queue_tail(&sk->receive_queue, skb);
tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
@@ -1892,10 +2467,14 @@
static void tcp_data_queue(struct sock *sk, struct sk_buff *skb)
{
- struct sk_buff *skb1;
struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
int eaten = 0;
+ if (tp->dsack) {
+ tp->dsack = 0;
+ tp->eff_sacks = min(tp->num_sacks, 4-tp->tstamp_ok);
+ }
+
/* Queue data for delivery to the user.
* Packets in sequence go to the receive queue.
* Out of sequence packets to the out_of_order_queue.
@@ -1924,20 +2503,27 @@
if (!eaten) {
queue_and_out:
- skb_set_owner_r(skb, sk);
+ tcp_set_owner_r(skb, sk);
__skb_queue_tail(&sk->receive_queue, skb);
}
- dst_confirm(sk->dst_cache);
tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
if(skb->len)
- tcp_event_data_recv(tp, skb);
+ tcp_event_data_recv(sk, tp, skb);
if(skb->h.th->fin)
tcp_fin(skb, sk, skb->h.th);
- /* This may have eaten into a SACK block. */
- if(tp->sack_ok && tp->num_sacks)
- tcp_sack_remove_skb(tp, skb);
- tcp_ofo_queue(sk);
+ if (skb_queue_len(&tp->out_of_order_queue)) {
+ tcp_ofo_queue(sk);
+
+ /* RFC2581. 4.2. SHOULD send immediate ACK, when
+ * gap in queue is filled.
+ */
+ if (skb_queue_len(&tp->out_of_order_queue) == 0)
+ tp->ack.pingpong = 0;
+ }
+
+ if(tp->num_sacks)
+ tcp_sack_remove(tp);
/* Turn on fast path. */
if (skb_queue_len(&tp->out_of_order_queue) == 0 &&
@@ -1948,24 +2534,28 @@
tcp_fast_path_on(tp);
if (eaten) {
- kfree_skb(skb);
+ __kfree_skb(skb);
} else if (!sk->dead)
sk->data_ready(sk, 0);
return;
}
+#ifdef TCP_DEBUG
/* An old packet, either a retransmit or some packet got lost. */
if (!after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt)) {
/* A retransmit, 2nd most common case. Force an imediate ack.
*
* It is impossible, seq is checked by top level.
*/
- NETDEBUG(printk("retransmit in tcp_data_queue: seq %X\n", TCP_SKB_CB(skb)->seq));
+ printk("BUG: retransmit in tcp_data_queue: seq %X\n", TCP_SKB_CB(skb)->seq);
tcp_enter_quickack_mode(tp);
- tp->ack.pending = 1;
- kfree_skb(skb);
+ tcp_schedule_ack(tp);
+ __kfree_skb(skb);
return;
}
+#endif
+
+ tcp_enter_quickack_mode(tp);
if (before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) {
/* Partial packet, seq < rcv_next < end_seq */
@@ -1973,67 +2563,198 @@
tp->rcv_nxt, TCP_SKB_CB(skb)->seq,
TCP_SKB_CB(skb)->end_seq);
+ tcp_dsack_set(tp, TCP_SKB_CB(skb)->seq, tp->rcv_nxt);
goto queue_and_out;
}
- /* Ok. This is an out_of_order segment, force an ack. */
- tp->ack.pending = 1;
+ TCP_ECN_check_ce(tp, skb);
/* Disable header prediction. */
tp->pred_flags = 0;
-
+ tcp_schedule_ack(tp);
SOCK_DEBUG(sk, "out of order segment: rcv_next %X seq %X - %X\n",
tp->rcv_nxt, TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq);
- skb_set_owner_r(skb, sk);
+ tcp_set_owner_r(skb, sk);
if (skb_peek(&tp->out_of_order_queue) == NULL) {
/* Initial out of order segment, build 1 SACK. */
if(tp->sack_ok) {
tp->num_sacks = 1;
+ tp->dsack = 0;
+ tp->eff_sacks = 1;
tp->selective_acks[0].start_seq = TCP_SKB_CB(skb)->seq;
tp->selective_acks[0].end_seq = TCP_SKB_CB(skb)->end_seq;
}
__skb_queue_head(&tp->out_of_order_queue,skb);
} else {
- for(skb1=tp->out_of_order_queue.prev; ; skb1 = skb1->prev) {
- /* Already there. */
- if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb1)->seq) {
- if (skb->len >= skb1->len) {
- if(tp->sack_ok)
- tcp_sack_extend(tp, skb1, skb);
- __skb_append(skb1, skb);
- __skb_unlink(skb1, skb1->list);
- kfree_skb(skb1);
- } else {
- /* A duplicate, smaller than what is in the
- * out-of-order queue right now, toss it.
- */
- kfree_skb(skb);
- }
+ struct sk_buff *skb1=tp->out_of_order_queue.prev;
+ u32 seq = TCP_SKB_CB(skb)->seq;
+ u32 end_seq = TCP_SKB_CB(skb)->end_seq;
+
+ if (seq == TCP_SKB_CB(skb1)->end_seq) {
+ __skb_append(skb1, skb);
+
+ if (tp->num_sacks == 0 ||
+ tp->selective_acks[0].end_seq != seq)
+ goto add_sack;
+
+ /* Common case: data arrive in order after hole. */
+ tp->selective_acks[0].end_seq = end_seq;
+ return;
+ }
+
+ /* Find place to insert this segment. */
+ do {
+ if (!after(TCP_SKB_CB(skb1)->seq, seq))
break;
+ } while ((skb1=skb1->prev) != (struct sk_buff*)&tp->out_of_order_queue);
+
+ /* Do skb overlap to previous one? */
+ if (skb1 != (struct sk_buff*)&tp->out_of_order_queue &&
+ before(seq, TCP_SKB_CB(skb1)->end_seq)) {
+ if (!after(end_seq, TCP_SKB_CB(skb1)->end_seq)) {
+ /* All the bits are present. Drop. */
+ __kfree_skb(skb);
+ tcp_dsack_set(tp, seq, end_seq);
+ goto add_sack;
+ }
+ if (after(seq, TCP_SKB_CB(skb1)->seq)) {
+ /* Partial overlap. */
+ tcp_dsack_set(tp, seq, TCP_SKB_CB(skb1)->end_seq);
+ } else {
+ skb1 = skb1->prev;
}
-
- if (after(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb1)->seq)) {
- __skb_append(skb1, skb);
- if(tp->sack_ok)
- tcp_sack_new_ofo_skb(sk, skb);
- break;
+ }
+ __skb_insert(skb, skb1, skb1->next, &tp->out_of_order_queue);
+
+ /* And clean segments covered by new one as whole. */
+ while ((skb1 = skb->next) != (struct sk_buff*)&tp->out_of_order_queue &&
+ after(end_seq, TCP_SKB_CB(skb1)->seq)) {
+ if (before(end_seq, TCP_SKB_CB(skb1)->end_seq)) {
+ tcp_dsack_extend(tp, TCP_SKB_CB(skb1)->seq, end_seq);
+ break;
+ }
+ __skb_unlink(skb1, skb1->list);
+ tcp_dsack_extend(tp, TCP_SKB_CB(skb1)->seq, TCP_SKB_CB(skb1)->end_seq);
+ __kfree_skb(skb1);
+ }
+
+add_sack:
+ if (tp->sack_ok)
+ tcp_sack_new_ofo_skb(sk, seq, end_seq);
+ }
+}
+
+
+static void tcp_collapse_queue(struct sock *sk, struct sk_buff_head *q)
+{
+ struct sk_buff *skb = skb_peek(q);
+ struct sk_buff *skb_next;
+
+ while (skb &&
+ skb != (struct sk_buff *)q &&
+ (skb_next = skb->next) != (struct sk_buff *)q) {
+ struct tcp_skb_cb *scb = TCP_SKB_CB(skb);
+ struct tcp_skb_cb *scb_next = TCP_SKB_CB(skb_next);
+
+ if (scb->end_seq == scb_next->seq &&
+ skb_tailroom(skb) >= skb_next->len &&
+#define TCP_DONT_COLLAPSE (TCP_FLAG_FIN|TCP_FLAG_URG|TCP_FLAG_SYN)
+ !(tcp_flag_word(skb->h.th)&TCP_DONT_COLLAPSE) &&
+ !(tcp_flag_word(skb_next->h.th)&TCP_DONT_COLLAPSE)) {
+ /* OK to collapse two skbs to one */
+ memcpy(skb_put(skb, skb_next->len), skb_next->data, skb_next->len);
+ __skb_unlink(skb_next, skb_next->list);
+ scb->end_seq = scb_next->end_seq;
+ __kfree_skb(skb_next);
+ NET_INC_STATS_BH(TCPRcvCollapsed);
+ } else {
+ /* Lots of spare tailroom, reallocate this skb to trim it. */
+ if (tcp_win_from_space(skb->truesize) > skb->len &&
+ skb_tailroom(skb) > sizeof(struct sk_buff) + 16) {
+ struct sk_buff *nskb;
+
+ nskb = skb_copy_expand(skb, skb_headroom(skb), 0, GFP_ATOMIC);
+ if (nskb) {
+ tcp_set_owner_r(nskb, sk);
+ memcpy(nskb->data-skb_headroom(skb),
+ skb->data-skb_headroom(skb),
+ skb_headroom(skb));
+ __skb_append(skb, nskb);
+ __skb_unlink(skb, skb->list);
+ __kfree_skb(skb);
+ }
}
+ skb = skb_next;
+ }
+ }
+}
+
+/* Clean the out_of_order queue if we can, trying to get
+ * the socket within its memory limits again.
+ *
+ * Return less than zero if we should start dropping frames
+ * until the socket owning process reads some of the data
+ * to stabilize the situation.
+ */
+static int tcp_prune_queue(struct sock *sk)
+{
+ struct tcp_opt *tp = &sk->tp_pinfo.af_tcp;
+
+ SOCK_DEBUG(sk, "prune_queue: c=%x\n", tp->copied_seq);
+
+ NET_INC_STATS_BH(PruneCalled);
+
+ if (atomic_read(&sk->rmem_alloc) >= sk->rcvbuf)
+ tcp_clamp_window(sk, tp);
+ else if (tcp_memory_pressure)
+ tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4*tp->advmss);
+
+ tcp_collapse_queue(sk, &sk->receive_queue);
+ tcp_collapse_queue(sk, &tp->out_of_order_queue);
+ tcp_mem_reclaim(sk);
+
+ if (atomic_read(&sk->rmem_alloc) <= sk->rcvbuf)
+ return 0;
+
+ /* Collapsing did not help, destructive actions follow.
+ * This must not ever occur. */
- /* See if we've hit the start. If so insert. */
- if (skb1 == skb_peek(&tp->out_of_order_queue)) {
- __skb_queue_head(&tp->out_of_order_queue,skb);
- if(tp->sack_ok)
- tcp_sack_new_ofo_skb(sk, skb);
- break;
- }
- }
+ /* First, purge the out_of_order queue. */
+ if (skb_queue_len(&tp->out_of_order_queue)) {
+ net_statistics[smp_processor_id()*2].OfoPruned += skb_queue_len(&tp->out_of_order_queue);
+ __skb_queue_purge(&tp->out_of_order_queue);
+
+ /* Reset SACK state. A conforming SACK implementation will
+ * do the same at a timeout based retransmit. When a connection
+ * is in a sad state like this, we care only about integrity
+ * of the connection not performance.
+ */
+ if(tp->sack_ok)
+ tcp_sack_reset(tp);
+ tcp_mem_reclaim(sk);
}
- return;
+
+ if(atomic_read(&sk->rmem_alloc) <= sk->rcvbuf)
+ return 0;
+
+ /* If we are really being abused, tell the caller to silently
+ * drop receive data on the floor. It will get retransmitted
+ * and hopefully then we'll have sufficient space.
+ */
+ NET_INC_STATS_BH(RcvPruned);
+
+ /* Massive buffer overcommit. */
+ return -1;
}
+static inline int tcp_rmem_schedule(struct sock *sk, struct sk_buff *skb)
+{
+ return (int)skb->truesize <= sk->forward_alloc ||
+ tcp_mem_schedule(sk, skb->truesize, 1);
+}
/*
* This routine handles the data. If there is room in the buffer,
@@ -2053,53 +2774,103 @@
if (skb->len == 0 && !th->fin)
goto drop;
+ TCP_ECN_accept_cwr(tp, skb);
+
/*
* If our receive queue has grown past its limits shrink it.
* Make sure to do this before moving rcv_nxt, otherwise
* data might be acked for that we don't have enough room.
*/
- if (atomic_read(&sk->rmem_alloc) > sk->rcvbuf) {
- if (prune_queue(sk) < 0) {
- /* Still not enough room. That can happen when
- * skb->true_size differs significantly from skb->len.
- */
+ if (atomic_read(&sk->rmem_alloc) > sk->rcvbuf ||
+ !tcp_rmem_schedule(sk, skb)) {
+ if (tcp_prune_queue(sk) < 0 || !tcp_rmem_schedule(sk, skb))
goto drop;
- }
}
tcp_data_queue(sk, skb);
+#ifdef TCP_DEBUG
if (before(tp->rcv_nxt, tp->copied_seq)) {
printk(KERN_DEBUG "*** tcp.c:tcp_data bug acked < copied\n");
tp->rcv_nxt = tp->copied_seq;
}
+#endif
return;
drop:
- kfree_skb(skb);
+ __kfree_skb(skb);
+}
+
+/* RFC2861, slow part. Adjust cwnd, after it was not full during one rto.
+ * As additional protections, we do not touch cwnd in retransmission phases,
+ * and if application hit its sndbuf limit recently.
+ */
+void tcp_cwnd_application_limited(struct sock *sk)
+{
+ struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
+
+ if (tp->ca_state == TCP_CA_Open &&
+ sk->socket && !test_bit(SOCK_NOSPACE, &sk->socket->flags)) {
+ /* Limited by application or receiver window. */
+ u32 win_used = max(tp->snd_cwnd_used, 2);
+ if (win_used < tp->snd_cwnd) {
+ tp->snd_ssthresh = tcp_current_ssthresh(tp);
+ tp->snd_cwnd = (tp->snd_cwnd+win_used)>>1;
+ }
+ tp->snd_cwnd_used = 0;
+ }
+ tp->snd_cwnd_stamp = tcp_time_stamp;
}
+
/* When incoming ACK allowed to free some skb from write_queue,
- * we remember this in flag tp->sorry and wake up socket on the exit
- * from tcp input handler. Probably, handler has already eat this space
- * sending ACK and cloned frames from tcp_write_xmit().
+ * we remember this event in flag tp->queue_shrunk and wake up socket
+ * on the exit from tcp input handler.
*/
-static __inline__ void tcp_new_space(struct sock *sk)
+static void tcp_new_space(struct sock *sk)
{
struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
- struct socket *sock;
- tp->sorry = 0;
+ if (tp->packets_out < tp->snd_cwnd &&
+ !(sk->userlocks&SOCK_SNDBUF_LOCK) &&
+ !tcp_memory_pressure &&
+ atomic_read(&tcp_memory_allocated) < sysctl_tcp_mem[0]) {
+ int sndmem, demanded;
+
+ sndmem = tp->mss_clamp+MAX_TCP_HEADER+16+sizeof(struct sk_buff);
+ demanded = max(tp->snd_cwnd, tp->reordering+1);
+ sndmem *= 2*demanded;
+ if (sndmem > sk->sndbuf)
+ sk->sndbuf = min(sndmem, sysctl_tcp_wmem[2]);
+ tp->snd_cwnd_stamp = tcp_time_stamp;
+ }
+
+ /* Wakeup users. */
+ if (tcp_wspace(sk) >= tcp_min_write_space(sk)) {
+ struct socket *sock = sk->socket;
- if (sock_wspace(sk) >= tcp_min_write_space(sk) &&
- (sock = sk->socket) != NULL) {
clear_bit(SOCK_NOSPACE, &sock->flags);
if (sk->sleep && waitqueue_active(sk->sleep))
wake_up_interruptible(sk->sleep);
- if (sock->fasync_list)
+ if (sock->fasync_list && !(sk->shutdown&SEND_SHUTDOWN))
sock_wake_async(sock, 2, POLL_OUT);
+
+ /* Satisfy those who hook write_space() callback. */
+ if (sk->write_space != tcp_write_space)
+ sk->write_space(sk);
+ }
+}
+
+static inline void tcp_check_space(struct sock *sk)
+{
+ struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
+
+ if (tp->queue_shrunk) {
+ tp->queue_shrunk = 0;
+ if (sk->socket && test_bit(SOCK_NOSPACE, &sk->socket->flags))
+ tcp_new_space(sk);
}
}
@@ -2118,7 +2889,8 @@
struct sk_buff *skb = sk->tp_pinfo.af_tcp.send_head;
if (skb != NULL)
- __tcp_data_snd_check(sk, skb);
+ __tcp_data_snd_check(sk, skb);
+ tcp_check_space(sk);
}
/*
@@ -2128,32 +2900,15 @@
{
struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
- /* This also takes care of updating the window.
- * This if statement needs to be simplified.
- *
- * Rules for delaying an ack:
- * - delay time <= 0.5 HZ
- * - we don't have a window update to send
- * - must send at least every 2 full sized packets
- * - must send an ACK if we have any out of order data
- *
- * With an extra heuristic to handle loss of packet
- * situations and also helping the sender leave slow
- * start in an expediant manner.
- */
-
- /* More than one full frame received or... */
+ /* More than one full frame received... */
if (((tp->rcv_nxt - tp->rcv_wup) > tp->ack.rcv_mss
-#ifdef TCP_MORE_COARSE_ACKS
- /* Avoid to send immediate ACK from input path, if it
- * does not advance window far enough. tcp_recvmsg() will do this.
+ /* ... and right edge of window advances far enough.
+ * (tcp_recvmsg() will send ACK otherwise). Or...
*/
- && (!sysctl_tcp_retrans_collapse || __tcp_select_window(sk) >= tp->rcv_wnd)
-#endif
- ) ||
+ && __tcp_select_window(sk) >= tp->rcv_wnd) ||
/* We ACK each frame or... */
tcp_in_quickack_mode(tp) ||
- /* We have out of order data or */
+ /* We have out of order data. */
(ofo_possible &&
skb_peek(&tp->out_of_order_queue) != NULL)) {
/* Then ack it now */
@@ -2167,14 +2922,13 @@
static __inline__ void tcp_ack_snd_check(struct sock *sk)
{
struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
- if (tp->ack.pending == 0) {
+ if (!tcp_ack_scheduled(tp)) {
/* We sent a data segment already. */
return;
}
__tcp_ack_snd_check(sk, 1);
}
-
/*
* This routine is only called when we have urgent data
* signalled. Its the 'slow' part of tcp_urg. It could be
@@ -2248,92 +3002,6 @@
}
}
-/* Clean the out_of_order queue if we can, trying to get
- * the socket within its memory limits again.
- *
- * Return less than zero if we should start dropping frames
- * until the socket owning process reads some of the data
- * to stabilize the situation.
- */
-static int prune_queue(struct sock *sk)
-{
- struct tcp_opt *tp = &sk->tp_pinfo.af_tcp;
- struct sk_buff *skb;
- int pruned = 0;
-
- SOCK_DEBUG(sk, "prune_queue: c=%x\n", tp->copied_seq);
-
- NET_INC_STATS_BH(PruneCalled);
-
- /* First, purge the out_of_order queue. */
- skb = __skb_dequeue_tail(&tp->out_of_order_queue);
- if(skb != NULL) {
- /* Free it all. */
- do {
- pruned += skb->len;
- net_statistics[smp_processor_id()*2].OfoPruned += skb->len;
- kfree_skb(skb);
- skb = __skb_dequeue_tail(&tp->out_of_order_queue);
- } while(skb != NULL);
-
- /* Reset SACK state. A conforming SACK implementation will
- * do the same at a timeout based retransmit. When a connection
- * is in a sad state like this, we care only about integrity
- * of the connection not performance.
- */
- if(tp->sack_ok)
- tp->num_sacks = 0;
- }
-
- /* If we are really being abused, tell the caller to silently
- * drop receive data on the floor. It will get retransmitted
- * and hopefully then we'll have sufficient space.
- *
- * We used to try to purge the in-order packets too, but that
- * turns out to be deadly and fraught with races. Consider:
- *
- * 1) If we acked the data, we absolutely cannot drop the
- * packet. This data would then never be retransmitted.
- * 2) It is possible, with a proper sequence of events involving
- * delayed acks and backlog queue handling, to have the user
- * read the data before it gets acked. The previous code
- * here got this wrong, and it lead to data corruption.
- * 3) Too much state changes happen when the FIN arrives, so once
- * we've seen that we can't remove any in-order data safely.
- *
- * The net result is that removing in-order receive data is too
- * complex for anyones sanity. So we don't do it anymore. But
- * if we are really having our buffer space abused we stop accepting
- * new receive data.
- *
- * 8) The arguments are interesting, but I even cannot imagine
- * what kind of arguments could force us to drop NICE, ALREADY
- * RECEIVED DATA only to get one more packet? --ANK
- *
- * FIXME: it should recompute SACK state and only remove enough
- * buffers to get into bounds again. The current scheme loses
- * badly sometimes on links with large RTT, especially when
- * the driver has high overhead per skb.
- * (increasing the rcvbuf is not enough because it inflates the
- * the window too, disabling flow control effectively) -AK
- *
- * Mmm... Why not to scale it seprately then? Just replace
- * / WINDOW_ADVERTISE_DIVISOR with >> sk->window_advertise_scale
- * and adjust it dynamically, when TCP window flow control
- * fails? -ANK
- */
-
- tp->ack.quick = 0;
-
- if(atomic_read(&sk->rmem_alloc) < (sk->rcvbuf << 1))
- return 0;
-
- NET_INC_STATS_BH(RcvPruned);
-
- /* Massive buffer overcommit. */
- return -1;
-}
-
static int tcp_copy_to_iovec(struct sock *sk, struct sk_buff *skb, int hlen)
{
struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
@@ -2454,9 +3122,6 @@
* We do checksum and copy also but from device to kernel.
*/
- /* RED-PEN. Using static variables to pass function arguments
- * cannot be good idea...
- */
tp->saw_tstamp = 0;
/* pred_flags is 0xS?10 << 16 + snd_wnd
@@ -2468,7 +3133,7 @@
* PSH flag is ignored.
*/
- if ((tcp_flag_word(th) & ~(TCP_RESERVED_BITS|TCP_FLAG_PSH)) == tp->pred_flags &&
+ if ((tcp_flag_word(th) & TCP_HP_BITS) == tp->pred_flags &&
TCP_SKB_CB(skb)->seq == tp->rcv_nxt) {
int tcp_header_len = tp->tcp_header_len;
@@ -2500,10 +3165,8 @@
* seq == rcv_nxt and rcv_wup <= rcv_nxt.
* Hence, check seq<=rcv_wup reduces to:
*/
- if (tp->rcv_nxt == tp->rcv_wup) {
- tp->ts_recent = tp->rcv_tsval;
- tp->ts_recent_stamp = xtime.tv_sec;
- }
+ if (tp->rcv_nxt == tp->rcv_wup)
+ tcp_store_ts_recent(tp);
}
if (len <= tcp_header_len) {
@@ -2512,18 +3175,15 @@
/* We know that such packets are checksummed
* on entry.
*/
- tcp_ack(sk, th, TCP_SKB_CB(skb)->seq,
- TCP_SKB_CB(skb)->ack_seq, len);
- kfree_skb(skb);
+ tcp_ack(sk, skb, 0);
+ __kfree_skb(skb);
tcp_data_snd_check(sk);
- if (tp->sorry)
- tcp_new_space(sk);
return 0;
} else { /* Header too small */
TCP_INC_STATS_BH(TcpInErrs);
goto discard;
}
- } else if (TCP_SKB_CB(skb)->ack_seq == tp->snd_una) {
+ } else {
int eaten = 0;
if (tp->ucopy.task == current &&
@@ -2546,67 +3206,59 @@
if (tcp_checksum_complete_user(sk, skb))
goto csum_error;
- if (atomic_read(&sk->rmem_alloc) > sk->rcvbuf)
+ if ((int)skb->truesize > sk->forward_alloc)
goto step5;
NET_INC_STATS_BH(TCPHPHits);
/* Bulk data transfer: receiver */
__skb_pull(skb,tcp_header_len);
-
- /* DO NOT notify forward progress here.
- * It saves dozen of CPU instructions in fast path. --ANK
- * And where is it signaled then ? -AK
- * Nowhere. 8) --ANK
- */
__skb_queue_tail(&sk->receive_queue, skb);
- skb_set_owner_r(skb, sk);
-
+ tcp_set_owner_r(skb, sk);
tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
-
- /* FIN bit check is not done since if FIN is set in
- * this frame, the pred_flags won't match up. -DaveM
- */
- sk->data_ready(sk, 0);
}
- tcp_event_data_recv(tp, skb);
+ tcp_event_data_recv(sk, tp, skb);
+
+ if (TCP_SKB_CB(skb)->ack_seq != tp->snd_una) {
+ /* Well, only one small jumplet in fast path... */
+ tcp_ack(sk, skb, FLAG_DATA);
+ tcp_data_snd_check(sk);
+ if (!tcp_ack_scheduled(tp))
+ goto no_ack;
+ }
-#ifdef TCP_MORE_COARSE_ACKS
if (eaten) {
if (tcp_in_quickack_mode(tp)) {
tcp_send_ack(sk);
} else {
tcp_send_delayed_ack(sk);
}
- } else
-#endif
- __tcp_ack_snd_check(sk, 0);
+ } else {
+ __tcp_ack_snd_check(sk, 0);
+ }
+no_ack:
if (eaten)
- kfree_skb(skb);
+ __kfree_skb(skb);
+ else
+ sk->data_ready(sk, 0);
return 0;
}
- /* Packet is in sequence, flags are trivial;
- * only ACK is strange. Jump to step 5.
- */
- if (tcp_checksum_complete_user(sk, skb))
- goto csum_error;
- goto step5;
}
slow_path:
- if (tcp_checksum_complete_user(sk, skb))
+ if (len < (th->doff<<2) || tcp_checksum_complete_user(sk, skb))
goto csum_error;
/*
* RFC1323: H1. Apply PAWS check first.
*/
- if (tcp_fast_parse_options(sk, th, tp) && tp->saw_tstamp &&
+ if (tcp_fast_parse_options(skb, th, tp) && tp->saw_tstamp &&
tcp_paws_discard(tp, skb)) {
if (!th->rst) {
NET_INC_STATS_BH(PAWSEstabRejected);
- tcp_send_ack(sk);
+ tcp_send_dupack(sk, skb);
goto discard;
}
/* Resets are accepted even if PAWS failed.
@@ -2620,23 +3272,15 @@
* Standard slow path.
*/
- if (!tcp_sequence(tp, TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq)) {
+ if (!tcp_sequence(tp, TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq, th->rst)) {
/* RFC793, page 37: "In all states except SYN-SENT, all reset
* (RST) segments are validated by checking their SEQ-fields."
* And page 69: "If an incoming segment is not acceptable,
* an acknowledgment should be sent in reply (unless the RST bit
* is set, if so drop the segment and return)".
*/
- if (th->rst)
- goto discard;
- if (after(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) {
- SOCK_DEBUG(sk, "seq:%d end:%d wup:%d wnd:%d\n",
- TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
- tp->rcv_wup, tp->rcv_wnd);
- }
- tcp_enter_quickack_mode(tp);
- tcp_send_ack(sk);
- NET_INC_STATS_BH(DelayedACKLost);
+ if (!th->rst)
+ tcp_send_dupack(sk, skb);
goto discard;
}
@@ -2645,378 +3289,43 @@
goto discard;
}
- if (tp->saw_tstamp) {
- tcp_replace_ts_recent(sk, tp,
- TCP_SKB_CB(skb)->seq);
- }
+ tcp_replace_ts_recent(tp, TCP_SKB_CB(skb)->seq);
if(th->syn && TCP_SKB_CB(skb)->seq != tp->syn_seq) {
- SOCK_DEBUG(sk, "syn in established state\n");
TCP_INC_STATS_BH(TcpInErrs);
+ NET_INC_STATS_BH(TCPAbortOnSyn);
tcp_reset(sk);
return 1;
}
step5:
if(th->ack)
- tcp_ack(sk, th, TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->ack_seq, len);
-
+ tcp_ack(sk, skb, FLAG_SLOWPATH);
+
/* Process urgent data. */
tcp_urg(sk, th, len);
/* step 7: process the segment text */
tcp_data(skb, sk, len);
- /* Be careful, tcp_data() may have put this into TIME_WAIT. */
- if(sk->state != TCP_CLOSE) {
- tcp_data_snd_check(sk);
- tcp_ack_snd_check(sk);
- if (tp->sorry)
- tcp_new_space(sk);
- }
-
+ tcp_data_snd_check(sk);
+ tcp_ack_snd_check(sk);
return 0;
csum_error:
TCP_INC_STATS_BH(TcpInErrs);
discard:
- kfree_skb(skb);
+ __kfree_skb(skb);
return 0;
}
-
-/* This is not only more efficient than what we used to do, it eliminates
- * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM
- *
- * Actually, we could lots of memory writes here. tp of listening
- * socket contains all necessary default parameters.
- */
-struct sock *tcp_create_openreq_child(struct sock *sk, struct open_request *req, struct sk_buff *skb)
-{
- struct sock *newsk = sk_alloc(PF_INET, GFP_ATOMIC, 0);
-
- if(newsk != NULL) {
- struct tcp_opt *newtp;
-#ifdef CONFIG_FILTER
- struct sk_filter *filter;
-#endif
-
- memcpy(newsk, sk, sizeof(*newsk));
- newsk->state = TCP_SYN_RECV;
-
- /* SANITY */
- newsk->pprev = NULL;
- newsk->prev = NULL;
-
- /* Clone the TCP header template */
- newsk->dport = req->rmt_port;
-
- sock_lock_init(newsk);
- bh_lock_sock(newsk);
-
- atomic_set(&newsk->rmem_alloc, 0);
- skb_queue_head_init(&newsk->receive_queue);
- atomic_set(&newsk->wmem_alloc, 0);
- skb_queue_head_init(&newsk->write_queue);
- atomic_set(&newsk->omem_alloc, 0);
-
- newsk->done = 0;
- newsk->proc = 0;
- newsk->backlog.head = newsk->backlog.tail = NULL;
- skb_queue_head_init(&newsk->error_queue);
- newsk->write_space = tcp_write_space;
-#ifdef CONFIG_FILTER
- if ((filter = newsk->filter) != NULL)
- sk_filter_charge(newsk, filter);
-#endif
-
- /* Now setup tcp_opt */
- newtp = &(newsk->tp_pinfo.af_tcp);
- newtp->pred_flags = 0;
- newtp->rcv_nxt = req->rcv_isn + 1;
- newtp->snd_nxt = req->snt_isn + 1;
- newtp->snd_una = req->snt_isn + 1;
- newtp->snd_sml = req->snt_isn + 1;
-
- tcp_delack_init(newtp);
- if (skb->len >= 536)
- newtp->ack.last_seg_size = skb->len;
-
- tcp_prequeue_init(newtp);
-
- newtp->snd_wl1 = req->rcv_isn;
- newtp->snd_wl2 = req->snt_isn;
-
- newtp->retransmits = 0;
- newtp->backoff = 0;
- newtp->srtt = 0;
- newtp->mdev = TCP_TIMEOUT_INIT;
- newtp->rto = TCP_TIMEOUT_INIT;
-
- newtp->packets_out = 0;
- newtp->fackets_out = 0;
- newtp->retrans_out = 0;
- newtp->snd_ssthresh = 0x7fffffff;
-
- /* So many TCP implementations out there (incorrectly) count the
- * initial SYN frame in their delayed-ACK and congestion control
- * algorithms that we must have the following bandaid to talk
- * efficiently to them. -DaveM
- */
- newtp->snd_cwnd = 2;
- newtp->snd_cwnd_cnt = 0;
- newtp->high_seq = 0;
-
- newtp->dup_acks = 0;
- tcp_init_xmit_timers(newsk);
- skb_queue_head_init(&newtp->out_of_order_queue);
- newtp->send_head = newtp->retrans_head = NULL;
- newtp->rcv_wup = req->rcv_isn + 1;
- newtp->write_seq = req->snt_isn + 1;
- newtp->copied_seq = req->rcv_isn + 1;
-
- newtp->saw_tstamp = 0;
-
- newtp->probes_out = 0;
- newtp->num_sacks = 0;
- newtp->syn_seq = req->rcv_isn;
- newtp->fin_seq = req->rcv_isn;
- newtp->urg_data = 0;
- newtp->listen_opt = NULL;
- newtp->accept_queue = newtp->accept_queue_tail = NULL;
- /* Deinitialize syn_wait_lock to trap illegal accesses. */
- memset(&newtp->syn_wait_lock, 0, sizeof(newtp->syn_wait_lock));
-
- /* Back to base struct sock members. */
- newsk->err = 0;
- newsk->priority = 0;
- atomic_set(&newsk->refcnt, 1);
-#ifdef INET_REFCNT_DEBUG
- atomic_inc(&inet_sock_nr);
-#endif
-
- if (newsk->keepopen)
- tcp_reset_keepalive_timer(newsk, keepalive_time_when(newtp));
- newsk->socket = NULL;
- newsk->sleep = NULL;
-
- newtp->tstamp_ok = req->tstamp_ok;
- if((newtp->sack_ok = req->sack_ok) != 0)
- newtp->num_sacks = 0;
- newtp->window_clamp = req->window_clamp;
- newtp->rcv_wnd = req->rcv_wnd;
- newtp->wscale_ok = req->wscale_ok;
- if (newtp->wscale_ok) {
- newtp->snd_wscale = req->snd_wscale;
- newtp->rcv_wscale = req->rcv_wscale;
- } else {
- newtp->snd_wscale = newtp->rcv_wscale = 0;
- newtp->window_clamp = min(newtp->window_clamp,65535);
- }
- newtp->snd_wnd = ntohs(skb->h.th->window) << newtp->snd_wscale;
- newtp->max_window = newtp->snd_wnd;
-
- if (newtp->tstamp_ok) {
- newtp->ts_recent = req->ts_recent;
- newtp->ts_recent_stamp = xtime.tv_sec;
- newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED;
- } else {
- newtp->ts_recent_stamp = 0;
- newtp->tcp_header_len = sizeof(struct tcphdr);
- }
- newtp->mss_clamp = req->mss;
- }
- return newsk;
-}
-
-/*
- * Process an incoming packet for SYN_RECV sockets represented
- * as an open_request.
- */
-
-struct sock *tcp_check_req(struct sock *sk,struct sk_buff *skb,
- struct open_request *req,
- struct open_request **prev)
-{
- struct tcphdr *th = skb->h.th;
- struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
- u32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK);
- int paws_reject = 0;
- struct tcp_opt ttp;
- struct sock *child;
-
- ttp.saw_tstamp = 0;
- if (th->doff > (sizeof(struct tcphdr)>>2)) {
- tcp_parse_options(NULL, th, &ttp, 0);
-
- if (ttp.saw_tstamp) {
- ttp.ts_recent = req->ts_recent;
- /* We do not store true stamp, but it is not required,
- * it can be estimated (approximately)
- * from another data.
- */
- ttp.ts_recent_stamp = xtime.tv_sec - ((TCP_TIMEOUT_INIT/HZ)<<req->retrans);
- paws_reject = tcp_paws_check(&ttp, th->rst);
- }
- }
-
- /* Check for pure retransmited SYN. */
- if (TCP_SKB_CB(skb)->seq == req->rcv_isn &&
- flg == TCP_FLAG_SYN &&
- !paws_reject) {
- /*
- * RFC793 draws (Incorrectly! It was fixed in RFC1122)
- * this case on figure 6 and figure 8, but formal
- * protocol description says NOTHING.
- * To be more exact, it says that we should send ACK,
- * because this segment (at least, if it has no data)
- * is out of window.
- *
- * CONCLUSION: RFC793 (even with RFC1122) DOES NOT
- * describe SYN-RECV state. All the description
- * is wrong, we cannot believe to it and should
- * rely only on common sense and implementation
- * experience.
- *
- * Enforce "SYN-ACK" according to figure 8, figure 6
- * of RFC793, fixed by RFC1122.
- */
- req->class->rtx_syn_ack(sk, req, NULL);
- return NULL;
- }
-
- /* Further reproduces section "SEGMENT ARRIVES"
- for state SYN-RECEIVED of RFC793.
- It is broken, however, it does not work only
- when SYNs are crossed, which is impossible in our
- case.
-
- But generally, we should (RFC lies!) to accept ACK
- from SYNACK both here and in tcp_rcv_state_process().
- tcp_rcv_state_process() does not, hence, we do not too.
-
- Note that the case is absolutely generic:
- we cannot optimize anything here without
- violating protocol. All the checks must be made
- before attempt to create socket.
- */
-
- /* RFC793: "first check sequence number". */
-
- if (paws_reject || !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
- req->rcv_isn+1, req->rcv_isn+1+req->rcv_wnd)) {
- /* Out of window: send ACK and drop. */
- if (!(flg & TCP_FLAG_RST))
- req->class->send_ack(skb, req);
- if (paws_reject)
- NET_INC_STATS_BH(PAWSEstabRejected);
- return NULL;
- }
-
- /* In sequence, PAWS is OK. */
-
- if (ttp.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, req->rcv_isn+1))
- req->ts_recent = ttp.rcv_tsval;
-
- if (TCP_SKB_CB(skb)->seq == req->rcv_isn) {
- /* Truncate SYN, it is out of window starting
- at req->rcv_isn+1. */
- flg &= ~TCP_FLAG_SYN;
- }
-
- /* RFC793: "second check the RST bit" and
- * "fourth, check the SYN bit"
- */
- if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN))
- goto embryonic_reset;
-
- /* RFC793: "fifth check the ACK field" */
-
- if (!(flg & TCP_FLAG_ACK))
- return NULL;
-
- /* Invalid ACK: reset will be sent by listening socket */
- if (TCP_SKB_CB(skb)->ack_seq != req->snt_isn+1)
- return sk;
- /* Also, it would be not so bad idea to check rcv_tsecr, which
- * is essentially ACK extension and too early or too late values
- * should cause reset in unsynchronized states.
- */
-
- /* If TCP_DEFER_ACCEPT is set, drop bare ACK. */
- if (tp->defer_accept && TCP_SKB_CB(skb)->end_seq == req->rcv_isn+1) {
- req->acked = 1;
- return NULL;
- }
-
- /* OK, ACK is valid, create big socket and
- * feed this segment to it. It will repeat all
- * the tests. THIS SEGMENT MUST MOVE SOCKET TO
- * ESTABLISHED STATE. If it will be dropped after
- * socket is created, wait for troubles.
- */
- child = tp->af_specific->syn_recv_sock(sk, skb, req, NULL);
- if (child == NULL)
- goto listen_overflow;
-
- tcp_synq_unlink(tp, req, prev);
- tcp_synq_removed(sk, req);
-
- tcp_acceptq_queue(sk, req, child);
- return child;
-
-listen_overflow:
- if (!sysctl_tcp_abort_on_overflow) {
- req->acked = 1;
- return NULL;
- }
-
-embryonic_reset:
- NET_INC_STATS_BH(EmbryonicRsts);
- if (!(flg & TCP_FLAG_RST))
- req->class->send_reset(skb);
-
- tcp_synq_drop(sk, req, prev);
- return NULL;
-}
-
-/*
- * Queue segment on the new socket if the new socket is active,
- * otherwise we just shortcircuit this and continue with
- * the new socket.
- */
-
-int tcp_child_process(struct sock *parent, struct sock *child,
- struct sk_buff *skb)
-{
- int ret = 0;
- int state = child->state;
-
- if (child->lock.users == 0) {
- ret = tcp_rcv_state_process(child, skb, skb->h.th, skb->len);
-
- /* Wakeup parent, send SIGIO */
- if (state == TCP_SYN_RECV && child->state != state)
- parent->data_ready(parent, 0);
- } else {
- /* Alas, it is possible again, because we do lookup
- * in main socket hash table and lock on listening
- * socket does not protect us more.
- */
- sk_add_backlog(child, skb);
- }
-
- bh_unlock_sock(child);
- return ret;
-}
-
static int tcp_rcv_synsent_state_process(struct sock *sk, struct sk_buff *skb,
struct tcphdr *th, unsigned len)
{
struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
- tcp_parse_options(sk, th, tp, 0);
+ tcp_parse_options(skb, tp);
if (th->ack) {
/* rfc793:
@@ -3027,24 +3336,12 @@
* a reset (unless the RST bit is set, if so drop
* the segment and return)"
*
- * I cite this place to emphasize one essential
- * detail, this check is different of one
- * in established state: SND.UNA <= SEG.ACK <= SND.NXT.
- * SEG_ACK == SND.UNA == ISS is invalid in SYN-SENT,
- * because we have no previous data sent before SYN.
- * --ANK(990513)
- *
* We do not send data with SYN, so that RFC-correct
* test reduces to:
*/
if (TCP_SKB_CB(skb)->ack_seq != tp->snd_nxt)
return 1;
- /* Check not from any RFC, but it is evident consequence
- * of combining PAWS and usual SYN-SENT logic: ACK _is_
- * checked in SYN-SENT unlike another states, hence
- * echoed tstamp must be checked too.
- */
if (tp->saw_tstamp) {
if (tp->rcv_tsecr == 0) {
/* Workaround for bug in linux-2.1 and early
@@ -3055,13 +3352,9 @@
tp->saw_tstamp = 0;
/* But do not forget to store peer's timestamp! */
- if (th->syn) {
- tp->ts_recent = tp->rcv_tsval;
- tp->ts_recent_stamp = xtime.tv_sec;
- }
- } else if ((__s32)(tp->rcv_tsecr - tcp_time_stamp) > 0 ||
- (__s32)(tp->rcv_tsecr - tp->syn_stamp) < 0) {
- NETDEBUG(if (net_ratelimit()) printk(KERN_DEBUG "TCP: synsent reject.\n"));
+ if (th->syn)
+ tcp_store_ts_recent(tp);
+ } else if (!between(tp->rcv_tsecr, tp->retrans_stamp, tcp_time_stamp)) {
NET_INC_STATS_BH(PAWSActiveRejected);
return 1;
}
@@ -3095,30 +3388,12 @@
* are acceptable then ...
* (our SYN has been ACKed), change the connection
* state to ESTABLISHED..."
- *
- * Do you see? SYN-less ACKs in SYN-SENT state are
- * completely ignored.
- *
- * The bug causing stalled SYN-SENT sockets
- * was here: tcp_ack advanced snd_una and canceled
- * retransmit timer, so that bare ACK received
- * in SYN-SENT state (even with invalid ack==ISS,
- * because tcp_ack check is too weak for SYN-SENT)
- * causes moving socket to invalid semi-SYN-SENT,
- * semi-ESTABLISHED state and connection hangs.
- * --ANK (990514)
- *
- * Bare ACK is valid, however.
- * Actually, RFC793 requires to send such ACK
- * in reply to any out of window packet.
- * It is wrong, but Linux also send such
- * useless ACKs sometimes.
- * --ANK (990724)
*/
+ TCP_ECN_rcv_synack(tp, th);
+
tp->snd_wl1 = TCP_SKB_CB(skb)->seq;
- tcp_ack(sk,th, TCP_SKB_CB(skb)->seq,
- TCP_SKB_CB(skb)->ack_seq, len);
+ tcp_ack(sk, skb, FLAG_SLOWPATH);
/* Ok.. it's good. Set up sequence numbers and
* move to established.
@@ -3130,12 +3405,10 @@
* never scaled.
*/
tp->snd_wnd = ntohs(th->window);
- tp->snd_wl1 = TCP_SKB_CB(skb)->seq;
- tp->snd_wl2 = TCP_SKB_CB(skb)->ack_seq;
+ tcp_init_wl(tp, TCP_SKB_CB(skb)->ack_seq, TCP_SKB_CB(skb)->seq);
+ tp->syn_seq = TCP_SKB_CB(skb)->seq;
tp->fin_seq = TCP_SKB_CB(skb)->seq;
- tcp_set_state(sk, TCP_ESTABLISHED);
-
if (tp->wscale_ok == 0) {
tp->snd_wscale = tp->rcv_wscale = 0;
tp->window_clamp = min(tp->window_clamp,65535);
@@ -3144,12 +3417,14 @@
if (tp->tstamp_ok) {
tp->tcp_header_len =
sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED;
+ tp->advmss -= TCPOLEN_TSTAMP_ALIGNED;
} else
tp->tcp_header_len = sizeof(struct tcphdr);
- if (tp->saw_tstamp) {
- tp->ts_recent = tp->rcv_tsval;
- tp->ts_recent_stamp = xtime.tv_sec;
- }
+ if (tp->saw_tstamp)
+ tcp_store_ts_recent(tp);
+ if (tp->sack_ok && sysctl_tcp_fack)
+ tp->sack_ok |= 2;
+
tcp_sync_mss(sk, tp->pmtu_cookie);
tcp_initialize_rcv_mss(sk);
tcp_init_metrics(sk);
@@ -3158,15 +3433,24 @@
if (sk->keepopen)
tcp_reset_keepalive_timer(sk, keepalive_time_when(tp));
+ if (tp->snd_wscale == 0)
+ __tcp_fast_path_on(tp, tp->snd_wnd);
+ else
+ tp->pred_flags = 0;
+
+ /* Remember, tcp_poll() does not lock socket!
+ * Change state from SYN-SENT only after copied_seq
+ * is initilized. */
tp->copied_seq = tp->rcv_nxt;
- __tcp_fast_path_on(tp, tp->snd_wnd);
+ mb();
+ tcp_set_state(sk, TCP_ESTABLISHED);
if(!sk->dead) {
sk->state_change(sk);
sk_wake_async(sk, 0, POLL_OUT);
}
- if (tp->write_pending) {
+ if (tp->write_pending || tp->defer_accept) {
/* Save one ACK. Data will be ready after
* several ticks, if write_pending is set.
*
@@ -3174,11 +3458,10 @@
* look so _wonderfully_ clever, that I was not able
* to stand against the temptation 8) --ANK
*/
- tp->ack.pending = 1;
+ tcp_schedule_ack(tp);
tp->ack.lrcvtime = tcp_time_stamp;
tcp_enter_quickack_mode(tp);
- tp->ack.ato = TCP_ATO_MIN;
- tcp_reset_xmit_timer(sk, TCP_TIME_DACK, TCP_DELACK_MIN);
+ tcp_reset_xmit_timer(sk, TCP_TIME_DACK, TCP_DELACK_MAX);
goto discard;
} else {
tcp_send_ack(sk);
@@ -3204,20 +3487,12 @@
if (th->syn) {
/* We see SYN without ACK. It is attempt of
- * simultaneous connect with crossed SYNs.
- *
- * The previous version of the code
- * checked for "connecting to self"
- * here. that check is done now in
- * tcp_connect.
- *
- * RED-PEN: BTW, it does not. 8)
+ * simultaneous connect with crossed SYNs.
+ * Particularly, it can be connect to self.
*/
tcp_set_state(sk, TCP_SYN_RECV);
- if (tp->saw_tstamp) {
- tp->ts_recent = tp->rcv_tsval;
- tp->ts_recent_stamp = xtime.tv_sec;
- }
+ if (tp->saw_tstamp)
+ tcp_store_ts_recent(tp);
tp->rcv_nxt = TCP_SKB_CB(skb)->seq + 1;
tp->rcv_wup = TCP_SKB_CB(skb)->seq + 1;
@@ -3232,6 +3507,8 @@
tcp_sync_mss(sk, tp->pmtu_cookie);
tcp_initialize_rcv_mss(sk);
+ TCP_ECN_rcv_syn(tp, th);
+
tcp_send_synack(sk);
#if 0
/* Note, we could accept data and URG from this segment.
@@ -3251,7 +3528,7 @@
*/
discard:
- kfree_skb(skb);
+ __kfree_skb(skb);
return 0;
}
@@ -3273,35 +3550,6 @@
switch (sk->state) {
case TCP_CLOSE:
- /* When state == CLOSED, hash lookup always fails.
- *
- * But, there is a back door, the backlog queue.
- * If we have a sequence of packets in the backlog
- * during __release_sock() which have a sequence such
- * that:
- * packet X causes entry to TCP_CLOSE state
- * ...
- * packet X + N has FIN bit set
- *
- * We report a (luckily) harmless error in this case.
- * The issue is that backlog queue processing bypasses
- * any hash lookups (we know which socket packets are for).
- * The correct behavior here is what 2.0.x did, since
- * a TCP_CLOSE socket does not exist. Drop the frame
- * and send a RST back to the other end.
- */
-
- /* 1. The socket may be moved to TIME-WAIT state.
- 2. While this socket was locked, another socket
- with the same identity could be created.
- 3. To continue?
-
- CONCLUSION: discard and only discard!
-
- Alternative would be relookup and recurse into tcp_v?_rcv
- (not *_do_rcv) to work with timewait and listen states
- correctly.
- */
goto discard;
case TCP_LISTEN:
@@ -3340,56 +3588,20 @@
goto step6;
}
- /* Parse the tcp_options present on this header.
- * By this point we really only expect timestamps.
- * Note that this really has to be here and not later for PAWS
- * (RFC1323) to work.
- */
- if (tcp_fast_parse_options(sk, th, tp) && tp->saw_tstamp &&
+ if (tcp_fast_parse_options(skb, th, tp) && tp->saw_tstamp &&
tcp_paws_discard(tp, skb)) {
if (!th->rst) {
- tcp_send_ack(sk);
+ NET_INC_STATS_BH(PAWSEstabRejected);
+ tcp_send_dupack(sk, skb);
goto discard;
}
/* Reset is accepted even if it did not pass PAWS. */
}
- /* The silly FIN test here is necessary to see an advancing ACK in
- * retransmitted FIN frames properly. Consider the following sequence:
- *
- * host1 --> host2 FIN XSEQ:XSEQ(0) ack YSEQ
- * host2 --> host1 FIN YSEQ:YSEQ(0) ack XSEQ
- * host1 --> host2 XSEQ:XSEQ(0) ack YSEQ+1
- * host2 --> host1 FIN YSEQ:YSEQ(0) ack XSEQ+1 (fails tcp_sequence test)
- *
- * At this point the connection will deadlock with host1 believing
- * that his FIN is never ACK'd, and thus it will retransmit it's FIN
- * forever. The following fix is from Taral (taral@taral.net).
- *
- * RED-PEN. Seems, the above is not true.
- * If at least one end is RFC compliant, it will send ACK to
- * out of window FIN and, hence, move peer to TIME-WAIT.
- * I comment out this line. --ANK
- *
- * RED-PEN. DANGER! tcp_sequence check rejects also SYN-ACKs
- * received in SYN-RECV. The problem is that description of
- * segment processing in SYN-RECV state in RFC792 is WRONG.
- * Correct check would accept ACK from this SYN-ACK, see
- * figures 6 and 8 (fixed by RFC1122). Compare this
- * to problem with FIN, they smell similarly. --ANK
- */
-
/* step 1: check sequence number */
- if (!tcp_sequence(tp, TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq)
-#if 0
- && !(th->fin && TCP_SKB_CB(skb)->end_seq == tp->rcv_nxt)
-#endif
- ) {
- if (!th->rst) {
- NET_INC_STATS_BH(DelayedACKLost);
- tcp_enter_quickack_mode(tp);
- tcp_send_ack(sk);
- }
+ if (!tcp_sequence(tp, TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq, th->rst)) {
+ if (!th->rst)
+ tcp_send_dupack(sk, skb);
goto discard;
}
@@ -3399,10 +3611,7 @@
goto discard;
}
- if (tp->saw_tstamp) {
- tcp_replace_ts_recent(sk, tp,
- TCP_SKB_CB(skb)->seq);
- }
+ tcp_replace_ts_recent(tp, TCP_SKB_CB(skb)->seq);
/* step 3: check security and precedence [ignored] */
@@ -3423,47 +3632,51 @@
*/
if (th->syn && TCP_SKB_CB(skb)->seq != tp->syn_seq) {
+ NET_INC_STATS_BH(TCPAbortOnSyn);
tcp_reset(sk);
return 1;
}
/* step 5: check the ACK field */
if (th->ack) {
- int acceptable = tcp_ack(sk, th, TCP_SKB_CB(skb)->seq,
- TCP_SKB_CB(skb)->ack_seq, len);
+ int acceptable = tcp_ack(sk, skb, FLAG_SLOWPATH);
switch(sk->state) {
case TCP_SYN_RECV:
if (acceptable) {
- tcp_set_state(sk, TCP_ESTABLISHED);
tp->copied_seq = tp->rcv_nxt;
+ mb();
+ tcp_set_state(sk, TCP_ESTABLISHED);
/* Note, that this wakeup is only for marginal
* crossed SYN case. Passively open sockets
* are not waked up, because sk->sleep == NULL
* and sk->socket == NULL.
*/
- if (!sk->dead) {
+ if (sk->socket) {
sk->state_change(sk);
sk_wake_async(sk,0,POLL_OUT);
}
tp->snd_una = TCP_SKB_CB(skb)->ack_seq;
tp->snd_wnd = ntohs(th->window) << tp->snd_wscale;
- tp->snd_wl1 = TCP_SKB_CB(skb)->seq;
- tp->snd_wl2 = TCP_SKB_CB(skb)->ack_seq;
+ tcp_init_wl(tp, TCP_SKB_CB(skb)->ack_seq, TCP_SKB_CB(skb)->seq);
/* tcp_ack considers this ACK as duplicate
* and does not calculate rtt.
* Fix it at least with timestamps.
*/
if (tp->saw_tstamp && !tp->srtt)
- tcp_ack_saw_tstamp(sk, tp, 0, 0, FLAG_SYN_ACKED);
+ tcp_ack_saw_tstamp(tp);
+
+ if (tp->tstamp_ok)
+ tp->advmss -= TCPOLEN_TSTAMP_ALIGNED;
tcp_init_metrics(sk);
+ tcp_initialize_rcv_mss(sk);
+ tcp_init_buffer_space(sk);
tcp_fast_path_on(tp);
} else {
- SOCK_DEBUG(sk, "bad ack\n");
return 1;
}
break;
@@ -3484,6 +3697,7 @@
(TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq &&
after(TCP_SKB_CB(skb)->end_seq - th->fin, tp->rcv_nxt))) {
tcp_done(sk);
+ NET_INC_STATS_BH(TCPAbortOnData);
return 1;
}
@@ -3543,6 +3757,7 @@
if (sk->shutdown & RCV_SHUTDOWN) {
if (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq &&
after(TCP_SKB_CB(skb)->end_seq - th->fin, tp->rcv_nxt)) {
+ NET_INC_STATS_BH(TCPAbortOnData);
tcp_reset(sk);
return 1;
}
@@ -3558,13 +3773,11 @@
if (sk->state != TCP_CLOSE) {
tcp_data_snd_check(sk);
tcp_ack_snd_check(sk);
- if (tp->sorry)
- tcp_new_space(sk);
}
if (!queued) {
discard:
- kfree_skb(skb);
+ __kfree_skb(skb);
}
return 0;
}
FUNET's LINUX-ADM group, linux-adm@nic.funet.fi
TCL-scripts by Sam Shen (who was at: slshen@lbl.gov)