2005-08-09 20:14:34 -07:00
|
|
|
/*
|
|
|
|
* net/dccp/ccids/ccid3.c
|
|
|
|
*
|
2007-12-12 09:06:14 -07:00
|
|
|
* Copyright (c) 2007 The University of Aberdeen, Scotland, UK
|
2007-05-28 08:23:29 -07:00
|
|
|
* Copyright (c) 2005-7 The University of Waikato, Hamilton, New Zealand.
|
|
|
|
* Copyright (c) 2005-7 Ian McDonald <ian.mcdonald@jandi.co.nz>
|
2005-08-09 20:14:34 -07:00
|
|
|
*
|
|
|
|
* An implementation of the DCCP protocol
|
|
|
|
*
|
|
|
|
* This code has been developed by the University of Waikato WAND
|
|
|
|
* research group. For further information please see http://www.wand.net.nz/
|
|
|
|
*
|
|
|
|
* This code also uses code from Lulea University, rereleased as GPL by its
|
|
|
|
* authors:
|
|
|
|
* Copyright (c) 2003 Nils-Erik Mattsson, Joacim Haggmark, Magnus Erixzon
|
|
|
|
*
|
|
|
|
* Changes to meet Linux coding standards, to make it meet latest ccid3 draft
|
|
|
|
* and to make it work as a loadable module in the DCCP stack written by
|
|
|
|
* Arnaldo Carvalho de Melo <acme@conectiva.com.br>.
|
|
|
|
*
|
|
|
|
* Copyright (c) 2005 Arnaldo Carvalho de Melo <acme@conectiva.com.br>
|
|
|
|
*
|
|
|
|
* This program is free software; you can redistribute it and/or modify
|
|
|
|
* it under the terms of the GNU General Public License as published by
|
|
|
|
* the Free Software Foundation; either version 2 of the License, or
|
|
|
|
* (at your option) any later version.
|
|
|
|
*
|
|
|
|
* This program is distributed in the hope that it will be useful,
|
|
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
|
* GNU General Public License for more details.
|
|
|
|
*
|
|
|
|
* You should have received a copy of the GNU General Public License
|
|
|
|
* along with this program; if not, write to the Free Software
|
|
|
|
* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
|
|
|
|
*/
|
|
|
|
#include "../dccp.h"
|
|
|
|
#include "ccid3.h"
|
|
|
|
|
2007-10-24 05:46:58 -07:00
|
|
|
#include <asm/unaligned.h>
|
|
|
|
|
2006-11-20 13:28:09 -07:00
|
|
|
#ifdef CONFIG_IP_DCCP_CCID3_DEBUG
|
|
|
|
static int ccid3_debug;
|
|
|
|
#define ccid3_pr_debug(format, a...) DCCP_PR_DEBUG(ccid3_debug, format, ##a)
|
2005-08-09 20:14:34 -07:00
|
|
|
#else
|
|
|
|
#define ccid3_pr_debug(format, a...)
|
|
|
|
#endif
|
|
|
|
|
2007-03-20 09:11:24 -07:00
|
|
|
/*
|
|
|
|
* Transmitter Half-Connection Routines
|
|
|
|
*/
|
2006-11-20 13:28:09 -07:00
|
|
|
#ifdef CONFIG_IP_DCCP_CCID3_DEBUG
|
2005-08-09 20:14:34 -07:00
|
|
|
static const char *ccid3_tx_state_name(enum ccid3_hc_tx_states state)
|
|
|
|
{
|
|
|
|
static char *ccid3_state_names[] = {
|
|
|
|
[TFRC_SSTATE_NO_SENT] = "NO_SENT",
|
|
|
|
[TFRC_SSTATE_NO_FBACK] = "NO_FBACK",
|
|
|
|
[TFRC_SSTATE_FBACK] = "FBACK",
|
|
|
|
[TFRC_SSTATE_TERM] = "TERM",
|
|
|
|
};
|
|
|
|
|
|
|
|
return ccid3_state_names[state];
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2006-03-20 22:58:56 -07:00
|
|
|
static void ccid3_hc_tx_set_state(struct sock *sk,
|
|
|
|
enum ccid3_hc_tx_states state)
|
2005-08-09 20:14:34 -07:00
|
|
|
{
|
2005-09-08 22:40:58 -07:00
|
|
|
struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
|
2008-09-03 22:30:19 -07:00
|
|
|
enum ccid3_hc_tx_states oldstate = hctx->state;
|
2005-08-09 20:14:34 -07:00
|
|
|
|
|
|
|
ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
|
2005-08-26 23:51:58 -07:00
|
|
|
dccp_role(sk), sk, ccid3_tx_state_name(oldstate),
|
|
|
|
ccid3_tx_state_name(state));
|
2005-08-09 20:14:34 -07:00
|
|
|
WARN_ON(state == oldstate);
|
2008-09-03 22:30:19 -07:00
|
|
|
hctx->state = state;
|
2005-08-09 20:14:34 -07:00
|
|
|
}
|
|
|
|
|
2007-03-20 11:12:10 -07:00
|
|
|
/*
|
2007-11-20 12:33:17 -07:00
|
|
|
* Compute the initial sending rate X_init in the manner of RFC 3390:
|
|
|
|
*
|
|
|
|
* X_init = min(4 * s, max(2 * s, 4380 bytes)) / RTT
|
|
|
|
*
|
|
|
|
* Note that RFC 3390 uses MSS, RFC 4342 refers to RFC 3390, and rfc3448bis
|
|
|
|
* (rev-02) clarifies the use of RFC 3390 with regard to the above formula.
|
2007-03-20 11:12:10 -07:00
|
|
|
* For consistency with other parts of the code, X_init is scaled by 2^6.
|
|
|
|
*/
|
|
|
|
static inline u64 rfc3390_initial_rate(struct sock *sk)
|
|
|
|
{
|
2007-11-20 12:33:17 -07:00
|
|
|
const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
|
2008-09-03 22:30:19 -07:00
|
|
|
const __u32 w_init = clamp_t(__u32, 4380U, 2 * hctx->s, 4 * hctx->s);
|
2007-03-20 11:12:10 -07:00
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
return scaled_div(w_init << 6, hctx->rtt);
|
2007-03-20 11:12:10 -07:00
|
|
|
}
|
|
|
|
|
dccp ccid-3: Bug fix for the inter-packet scheduling algorithm
This fixes a subtle bug in the calculation of the inter-packet gap and shows
that t_delta, as it is currently used, is not needed. And hence replaced.
The algorithm from RFC 3448, 4.6 below continually computes a send time t_nom,
which is initialised with the current time t_now; t_gran = 1E6 / HZ specifies
the scheduling granularity, s the packet size, and X the sending rate:
t_distance = t_nom - t_now; // in microseconds
t_delta = min(t_ipi, t_gran) / 2; // `delta' parameter in microseconds
if (t_distance >= t_delta) {
reschedule after (t_distance / 1000) milliseconds;
} else {
t_ipi = s / X; // inter-packet interval in usec
t_nom += t_ipi; // compute the next send time
send packet now;
}
1) Description of the bug
-------------------------
Rescheduling requires a conversion into milliseconds, due to this call chain:
* ccid3_hc_tx_send_packet() returns a timeout in milliseconds,
* this value is converted by msecs_to_jiffies() in dccp_write_xmit(),
* and finally used as jiffy-expires-value for sk_reset_timer().
The highest jiffy resolution with HZ=1000 is 1 millisecond, so using a higher
granularity does not make much sense here.
As a consequence, values of t_distance < 1000 are truncated to 0. This issue
has so far been resolved by using instead
if (t_distance >= t_delta + 1000)
reschedule after (t_distance / 1000) milliseconds;
The bug is in artificially inflating t_delta to t_delta' = t_delta + 1000. This
is unnecessarily large, a more adequate value is t_delta' = max(t_delta, 1000).
2) Consequences of using the corrected t_delta'
-----------------------------------------------
Since t_delta <= t_gran/2 = 10^6/(2*HZ), we have t_delta <= 1000 as long as
HZ >= 500. This means that t_delta' = max(1000, t_delta) is constant at 1000.
On the other hand, when using a coarse HZ value of HZ < 500, we have three
sub-cases that can all be reduced to using another constant of t_gran/2.
(a) The first case arises when t_ipi > t_gran. Here t_delta' is the constant
t_delta' = max(1000, t_gran/2) = t_gran/2.
(b) If t_ipi <= 2000 < t_gran = 10^6/HZ usec, then t_delta = t_ipi/2 <= 1000,
so that t_delta' = max(1000, t_delta) = 1000 < t_gran/2.
(c) If 2000 < t_ipi <= t_gran, we have t_delta' = max(t_delta, 1000) = t_ipi/2.
In the second and third cases we have delay values less than t_gran/2, which is
in the order of less than or equal to half a jiffy.
How these are treated depends on how fractions of a jiffy are handled: they
are either always rounded down to 0, or always rounded up to 1 jiffy (assuming
non-zero values). In both cases the error is on average in the order of 50%.
Thus we are not increasing the error when in the second/third case we replace
a value less than t_gran/2 with 0, by setting t_delta' to the constant t_gran/2.
3) Summary
----------
Fixing (1) and considering (2), the patch replaces t_delta with a constant,
whose value depends on CONFIG_HZ, changing the above algorithm to:
if (t_distance >= t_delta')
reschedule after (t_distance / 1000) milliseconds;
where t_delta' = 10^6/(2*HZ) if HZ < 500, and t_delta' = 1000 otherwise.
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
2008-09-03 22:30:19 -07:00
|
|
|
/**
|
|
|
|
* ccid3_update_send_interval - Calculate new t_ipi = s / X_inst
|
|
|
|
* This respects the granularity of X_inst (64 * bytes/second).
|
2006-11-27 15:31:33 -07:00
|
|
|
*/
|
2008-01-06 00:13:58 -07:00
|
|
|
static void ccid3_update_send_interval(struct ccid3_hc_tx_sock *hctx)
|
2005-08-09 20:14:34 -07:00
|
|
|
{
|
2008-09-03 22:30:19 -07:00
|
|
|
hctx->t_ipi = scaled_div32(((u64)hctx->s) << 6, hctx->x);
|
2005-08-09 20:14:34 -07:00
|
|
|
|
dccp ccid-3: Bug fix for the inter-packet scheduling algorithm
This fixes a subtle bug in the calculation of the inter-packet gap and shows
that t_delta, as it is currently used, is not needed. And hence replaced.
The algorithm from RFC 3448, 4.6 below continually computes a send time t_nom,
which is initialised with the current time t_now; t_gran = 1E6 / HZ specifies
the scheduling granularity, s the packet size, and X the sending rate:
t_distance = t_nom - t_now; // in microseconds
t_delta = min(t_ipi, t_gran) / 2; // `delta' parameter in microseconds
if (t_distance >= t_delta) {
reschedule after (t_distance / 1000) milliseconds;
} else {
t_ipi = s / X; // inter-packet interval in usec
t_nom += t_ipi; // compute the next send time
send packet now;
}
1) Description of the bug
-------------------------
Rescheduling requires a conversion into milliseconds, due to this call chain:
* ccid3_hc_tx_send_packet() returns a timeout in milliseconds,
* this value is converted by msecs_to_jiffies() in dccp_write_xmit(),
* and finally used as jiffy-expires-value for sk_reset_timer().
The highest jiffy resolution with HZ=1000 is 1 millisecond, so using a higher
granularity does not make much sense here.
As a consequence, values of t_distance < 1000 are truncated to 0. This issue
has so far been resolved by using instead
if (t_distance >= t_delta + 1000)
reschedule after (t_distance / 1000) milliseconds;
The bug is in artificially inflating t_delta to t_delta' = t_delta + 1000. This
is unnecessarily large, a more adequate value is t_delta' = max(t_delta, 1000).
2) Consequences of using the corrected t_delta'
-----------------------------------------------
Since t_delta <= t_gran/2 = 10^6/(2*HZ), we have t_delta <= 1000 as long as
HZ >= 500. This means that t_delta' = max(1000, t_delta) is constant at 1000.
On the other hand, when using a coarse HZ value of HZ < 500, we have three
sub-cases that can all be reduced to using another constant of t_gran/2.
(a) The first case arises when t_ipi > t_gran. Here t_delta' is the constant
t_delta' = max(1000, t_gran/2) = t_gran/2.
(b) If t_ipi <= 2000 < t_gran = 10^6/HZ usec, then t_delta = t_ipi/2 <= 1000,
so that t_delta' = max(1000, t_delta) = 1000 < t_gran/2.
(c) If 2000 < t_ipi <= t_gran, we have t_delta' = max(t_delta, 1000) = t_ipi/2.
In the second and third cases we have delay values less than t_gran/2, which is
in the order of less than or equal to half a jiffy.
How these are treated depends on how fractions of a jiffy are handled: they
are either always rounded down to 0, or always rounded up to 1 jiffy (assuming
non-zero values). In both cases the error is on average in the order of 50%.
Thus we are not increasing the error when in the second/third case we replace
a value less than t_gran/2 with 0, by setting t_delta' to the constant t_gran/2.
3) Summary
----------
Fixing (1) and considering (2), the patch replaces t_delta with a constant,
whose value depends on CONFIG_HZ, changing the above algorithm to:
if (t_distance >= t_delta')
reschedule after (t_distance / 1000) milliseconds;
where t_delta' = 10^6/(2*HZ) if HZ < 500, and t_delta' = 1000 otherwise.
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
2008-09-03 22:30:19 -07:00
|
|
|
ccid3_pr_debug("t_ipi=%u, s=%u, X=%u\n", hctx->t_ipi,
|
|
|
|
hctx->s, (unsigned)(hctx->x >> 6));
|
2005-08-09 20:14:34 -07:00
|
|
|
}
|
2007-09-25 22:39:16 -07:00
|
|
|
|
2007-11-20 13:01:59 -07:00
|
|
|
static u32 ccid3_hc_tx_idle_rtt(struct ccid3_hc_tx_sock *hctx, ktime_t now)
|
|
|
|
{
|
2008-09-03 22:30:19 -07:00
|
|
|
u32 delta = ktime_us_delta(now, hctx->t_last_win_count);
|
2007-11-20 13:01:59 -07:00
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
return delta / hctx->rtt;
|
2007-11-20 13:01:59 -07:00
|
|
|
}
|
|
|
|
|
2007-09-25 22:39:16 -07:00
|
|
|
/**
|
|
|
|
* ccid3_hc_tx_update_x - Update allowed sending rate X
|
|
|
|
* @stamp: most recent time if available - can be left NULL.
|
|
|
|
* This function tracks draft rfc3448bis, check there for latest details.
|
2006-12-03 09:50:56 -07:00
|
|
|
*
|
2006-12-09 19:02:12 -07:00
|
|
|
* Note: X and X_recv are both stored in units of 64 * bytes/second, to support
|
|
|
|
* fine-grained resolution of sending rates. This requires scaling by 2^6
|
|
|
|
* throughout the code. Only X_calc is unscaled (in bytes/second).
|
|
|
|
*
|
|
|
|
*/
|
2007-09-25 22:39:16 -07:00
|
|
|
static void ccid3_hc_tx_update_x(struct sock *sk, ktime_t *stamp)
|
2005-08-09 20:14:34 -07:00
|
|
|
{
|
2005-09-08 22:40:58 -07:00
|
|
|
struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
|
2008-09-03 22:30:19 -07:00
|
|
|
u64 min_rate = 2 * hctx->x_recv;
|
|
|
|
const u64 old_x = hctx->x;
|
2007-12-17 07:57:43 -07:00
|
|
|
ktime_t now = stamp ? *stamp : ktime_get_real();
|
2005-08-09 20:14:34 -07:00
|
|
|
|
2007-03-20 11:19:07 -07:00
|
|
|
/*
|
|
|
|
* Handle IDLE periods: do not reduce below RFC3390 initial sending rate
|
2007-11-20 13:01:59 -07:00
|
|
|
* when idling [RFC 4342, 5.1]. Definition of idling is from rfc3448bis:
|
|
|
|
* a sender is idle if it has not sent anything over a 2-RTT-period.
|
2007-03-20 11:19:07 -07:00
|
|
|
* For consistency with X and X_recv, min_rate is also scaled by 2^6.
|
|
|
|
*/
|
2007-11-20 13:01:59 -07:00
|
|
|
if (ccid3_hc_tx_idle_rtt(hctx, now) >= 2) {
|
2007-03-20 11:19:07 -07:00
|
|
|
min_rate = rfc3390_initial_rate(sk);
|
2008-09-03 22:30:19 -07:00
|
|
|
min_rate = max(min_rate, 2 * hctx->x_recv);
|
2007-03-20 11:19:07 -07:00
|
|
|
}
|
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
if (hctx->p > 0) {
|
2006-12-09 19:02:12 -07:00
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
hctx->x = min(((u64)hctx->x_calc) << 6, min_rate);
|
|
|
|
hctx->x = max(hctx->x, (((u64)hctx->s) << 6) / TFRC_T_MBI);
|
2006-11-28 14:51:42 -07:00
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
} else if (ktime_us_delta(now, hctx->t_ld) - (s64)hctx->rtt >= 0) {
|
2007-08-19 17:14:27 -07:00
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
hctx->x = min(2 * hctx->x, min_rate);
|
|
|
|
hctx->x = max(hctx->x,
|
|
|
|
scaled_div(((u64)hctx->s) << 6, hctx->rtt));
|
|
|
|
hctx->t_ld = now;
|
2006-12-09 19:00:14 -07:00
|
|
|
}
|
2005-08-27 14:18:18 -07:00
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
if (hctx->x != old_x) {
|
2007-03-20 11:04:30 -07:00
|
|
|
ccid3_pr_debug("X_prev=%u, X_now=%u, X_calc=%u, "
|
|
|
|
"X_recv=%u\n", (unsigned)(old_x >> 6),
|
2008-09-03 22:30:19 -07:00
|
|
|
(unsigned)(hctx->x >> 6), hctx->x_calc,
|
|
|
|
(unsigned)(hctx->x_recv >> 6));
|
2007-03-20 10:49:20 -07:00
|
|
|
|
2007-03-20 10:56:11 -07:00
|
|
|
ccid3_update_send_interval(hctx);
|
2007-03-20 10:49:20 -07:00
|
|
|
}
|
2005-08-09 20:14:34 -07:00
|
|
|
}
|
|
|
|
|
2006-11-28 14:22:33 -07:00
|
|
|
/*
|
2006-12-10 11:01:18 -07:00
|
|
|
* Track the mean packet size `s' (cf. RFC 4342, 5.3 and RFC 3448, 4.1)
|
|
|
|
* @len: DCCP packet payload size in bytes
|
2006-11-28 14:22:33 -07:00
|
|
|
*/
|
|
|
|
static inline void ccid3_hc_tx_update_s(struct ccid3_hc_tx_sock *hctx, int len)
|
|
|
|
{
|
2008-09-03 22:30:19 -07:00
|
|
|
const u16 old_s = hctx->s;
|
2007-03-20 10:56:11 -07:00
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
hctx->s = tfrc_ewma(hctx->s, len, 9);
|
2007-03-20 10:56:11 -07:00
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
if (hctx->s != old_s)
|
2007-03-20 10:56:11 -07:00
|
|
|
ccid3_update_send_interval(hctx);
|
2006-11-28 14:22:33 -07:00
|
|
|
}
|
|
|
|
|
2006-12-09 19:07:37 -07:00
|
|
|
/*
|
2006-12-10 11:01:18 -07:00
|
|
|
* Update Window Counter using the algorithm from [RFC 4342, 8.1].
|
2008-05-27 06:33:54 -07:00
|
|
|
* As elsewhere, RTT > 0 is assumed by using dccp_sample_rtt().
|
2006-12-09 19:07:37 -07:00
|
|
|
*/
|
|
|
|
static inline void ccid3_hc_tx_update_win_count(struct ccid3_hc_tx_sock *hctx,
|
2007-06-16 09:34:02 -07:00
|
|
|
ktime_t now)
|
2006-12-09 19:07:37 -07:00
|
|
|
{
|
2008-09-03 22:30:19 -07:00
|
|
|
u32 delta = ktime_us_delta(now, hctx->t_last_win_count),
|
|
|
|
quarter_rtts = (4 * delta) / hctx->rtt;
|
2006-12-09 19:07:37 -07:00
|
|
|
|
|
|
|
if (quarter_rtts > 0) {
|
2008-09-03 22:30:19 -07:00
|
|
|
hctx->t_last_win_count = now;
|
|
|
|
hctx->last_win_count += min(quarter_rtts, 5U);
|
|
|
|
hctx->last_win_count &= 0xF; /* mod 16 */
|
2006-12-09 19:07:37 -07:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-08-09 20:14:34 -07:00
|
|
|
static void ccid3_hc_tx_no_feedback_timer(unsigned long data)
|
|
|
|
{
|
|
|
|
struct sock *sk = (struct sock *)data;
|
2005-09-08 22:40:58 -07:00
|
|
|
struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
|
2006-11-28 13:34:34 -07:00
|
|
|
unsigned long t_nfb = USEC_PER_SEC / 5;
|
2005-08-09 20:14:34 -07:00
|
|
|
|
|
|
|
bh_lock_sock(sk);
|
|
|
|
if (sock_owned_by_user(sk)) {
|
|
|
|
/* Try again later. */
|
|
|
|
/* XXX: set some sensible MIB */
|
2006-11-27 15:29:27 -07:00
|
|
|
goto restart_timer;
|
2005-08-09 20:14:34 -07:00
|
|
|
}
|
|
|
|
|
2006-12-09 19:14:12 -07:00
|
|
|
ccid3_pr_debug("%s(%p, state=%s) - entry \n", dccp_role(sk), sk,
|
2008-09-03 22:30:19 -07:00
|
|
|
ccid3_tx_state_name(hctx->state));
|
2006-12-09 19:14:12 -07:00
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
if (hctx->state == TFRC_SSTATE_FBACK)
|
2007-12-17 07:57:43 -07:00
|
|
|
ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
|
2008-09-03 22:30:19 -07:00
|
|
|
else if (hctx->state != TFRC_SSTATE_NO_FBACK)
|
2007-12-17 07:57:43 -07:00
|
|
|
goto out;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Determine new allowed sending rate X as per draft rfc3448bis-00, 4.4
|
2008-09-03 22:30:19 -07:00
|
|
|
* RTO is 0 if and only if no feedback has been received yet.
|
2007-12-17 07:57:43 -07:00
|
|
|
*/
|
2008-09-03 22:30:19 -07:00
|
|
|
if (hctx->t_rto == 0 || hctx->p == 0) {
|
2007-12-17 07:57:43 -07:00
|
|
|
|
|
|
|
/* halve send rate directly */
|
2008-09-03 22:30:19 -07:00
|
|
|
hctx->x = max(hctx->x / 2, (((u64)hctx->s) << 6) / TFRC_T_MBI);
|
2007-03-20 10:56:11 -07:00
|
|
|
ccid3_update_send_interval(hctx);
|
2007-12-17 07:57:43 -07:00
|
|
|
} else {
|
2005-08-26 23:51:58 -07:00
|
|
|
/*
|
2007-12-17 07:57:43 -07:00
|
|
|
* Modify the cached value of X_recv
|
2007-03-20 11:19:07 -07:00
|
|
|
*
|
2007-12-17 07:57:43 -07:00
|
|
|
* If (X_calc > 2 * X_recv)
|
2007-03-20 11:19:07 -07:00
|
|
|
* X_recv = max(X_recv / 2, s / (2 * t_mbi));
|
|
|
|
* Else
|
|
|
|
* X_recv = X_calc / 4;
|
|
|
|
*
|
|
|
|
* Note that X_recv is scaled by 2^6 while X_calc is not
|
2005-08-26 23:51:58 -07:00
|
|
|
*/
|
2008-09-03 22:30:19 -07:00
|
|
|
BUG_ON(hctx->p && !hctx->x_calc);
|
2007-03-20 11:19:07 -07:00
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
if (hctx->x_calc > (hctx->x_recv >> 5))
|
|
|
|
hctx->x_recv =
|
|
|
|
max(hctx->x_recv / 2,
|
|
|
|
(((__u64)hctx->s) << 6) / (2 * TFRC_T_MBI));
|
2007-12-17 07:57:43 -07:00
|
|
|
else {
|
2008-09-03 22:30:19 -07:00
|
|
|
hctx->x_recv = hctx->x_calc;
|
|
|
|
hctx->x_recv <<= 4;
|
2005-08-09 20:14:34 -07:00
|
|
|
}
|
2007-09-25 22:39:16 -07:00
|
|
|
ccid3_hc_tx_update_x(sk, NULL);
|
2005-08-09 20:14:34 -07:00
|
|
|
}
|
2007-12-17 07:57:43 -07:00
|
|
|
ccid3_pr_debug("Reduced X to %llu/64 bytes/sec\n",
|
2008-09-03 22:30:19 -07:00
|
|
|
(unsigned long long)hctx->x);
|
2007-12-17 07:57:43 -07:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Set new timeout for the nofeedback timer.
|
|
|
|
* See comments in packet_recv() regarding the value of t_RTO.
|
|
|
|
*/
|
2008-09-03 22:30:19 -07:00
|
|
|
if (unlikely(hctx->t_rto == 0)) /* no feedback received yet */
|
2007-12-17 07:57:43 -07:00
|
|
|
t_nfb = TFRC_INITIAL_TIMEOUT;
|
|
|
|
else
|
2008-09-03 22:30:19 -07:00
|
|
|
t_nfb = max(hctx->t_rto, 2 * hctx->t_ipi);
|
2005-08-09 20:14:34 -07:00
|
|
|
|
2006-11-27 15:29:27 -07:00
|
|
|
restart_timer:
|
2008-09-03 22:30:19 -07:00
|
|
|
sk_reset_timer(sk, &hctx->no_feedback_timer,
|
2007-02-09 07:24:38 -07:00
|
|
|
jiffies + usecs_to_jiffies(t_nfb));
|
2005-08-09 20:14:34 -07:00
|
|
|
out:
|
|
|
|
bh_unlock_sock(sk);
|
|
|
|
sock_put(sk);
|
|
|
|
}
|
|
|
|
|
2006-11-27 07:26:03 -07:00
|
|
|
/*
|
|
|
|
* returns
|
|
|
|
* > 0: delay (in msecs) that should pass before actually sending
|
|
|
|
* = 0: can send immediately
|
|
|
|
* < 0: error condition; do not send packet
|
|
|
|
*/
|
2006-11-28 14:55:06 -07:00
|
|
|
static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
|
2005-08-09 20:14:34 -07:00
|
|
|
{
|
|
|
|
struct dccp_sock *dp = dccp_sk(sk);
|
2005-09-08 22:40:58 -07:00
|
|
|
struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
|
2007-06-16 09:34:02 -07:00
|
|
|
ktime_t now = ktime_get_real();
|
|
|
|
s64 delay;
|
2005-08-09 20:14:34 -07:00
|
|
|
|
|
|
|
/*
|
2006-11-27 07:26:57 -07:00
|
|
|
* This function is called only for Data and DataAck packets. Sending
|
|
|
|
* zero-sized Data(Ack)s is theoretically possible, but for congestion
|
|
|
|
* control this case is pathological - ignore it.
|
2005-08-09 20:14:34 -07:00
|
|
|
*/
|
2006-11-28 14:55:06 -07:00
|
|
|
if (unlikely(skb->len == 0))
|
2006-11-27 07:26:57 -07:00
|
|
|
return -EBADMSG;
|
2005-08-09 20:14:34 -07:00
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
switch (hctx->state) {
|
2005-08-09 20:14:34 -07:00
|
|
|
case TFRC_SSTATE_NO_SENT:
|
2008-09-03 22:30:19 -07:00
|
|
|
sk_reset_timer(sk, &hctx->no_feedback_timer, (jiffies +
|
2007-02-09 07:24:38 -07:00
|
|
|
usecs_to_jiffies(TFRC_INITIAL_TIMEOUT)));
|
2008-09-03 22:30:19 -07:00
|
|
|
hctx->last_win_count = 0;
|
|
|
|
hctx->t_last_win_count = now;
|
2006-11-27 07:13:38 -07:00
|
|
|
|
|
|
|
/* Set t_0 for initial packet */
|
2008-09-03 22:30:19 -07:00
|
|
|
hctx->t_nom = now;
|
2007-03-20 11:31:56 -07:00
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
hctx->s = skb->len;
|
2007-03-20 11:31:56 -07:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Use initial RTT sample when available: recommended by erratum
|
|
|
|
* to RFC 4342. This implements the initialisation procedure of
|
|
|
|
* draft rfc3448bis, section 4.2. Remember, X is scaled by 2^6.
|
|
|
|
*/
|
|
|
|
if (dp->dccps_syn_rtt) {
|
|
|
|
ccid3_pr_debug("SYN RTT = %uus\n", dp->dccps_syn_rtt);
|
2008-09-03 22:30:19 -07:00
|
|
|
hctx->rtt = dp->dccps_syn_rtt;
|
|
|
|
hctx->x = rfc3390_initial_rate(sk);
|
|
|
|
hctx->t_ld = now;
|
2007-03-20 11:31:56 -07:00
|
|
|
} else {
|
2008-06-11 03:19:09 -07:00
|
|
|
/*
|
|
|
|
* Sender does not have RTT sample:
|
|
|
|
* - set fallback RTT (RFC 4340, 3.4) since a RTT value
|
|
|
|
* is needed in several parts (e.g. window counter);
|
|
|
|
* - set sending rate X_pps = 1pps as per RFC 3448, 4.2.
|
|
|
|
*/
|
2008-09-03 22:30:19 -07:00
|
|
|
hctx->rtt = DCCP_FALLBACK_RTT;
|
|
|
|
hctx->x = hctx->s;
|
|
|
|
hctx->x <<= 6;
|
2007-03-20 11:31:56 -07:00
|
|
|
}
|
|
|
|
ccid3_update_send_interval(hctx);
|
|
|
|
|
|
|
|
ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
|
2005-08-09 20:14:34 -07:00
|
|
|
break;
|
|
|
|
case TFRC_SSTATE_NO_FBACK:
|
|
|
|
case TFRC_SSTATE_FBACK:
|
2008-09-03 22:30:19 -07:00
|
|
|
delay = ktime_us_delta(hctx->t_nom, now);
|
2007-03-20 10:49:20 -07:00
|
|
|
ccid3_pr_debug("delay=%ld\n", (long)delay);
|
2006-11-27 07:25:10 -07:00
|
|
|
/*
|
2006-12-10 11:01:18 -07:00
|
|
|
* Scheduling of packet transmissions [RFC 3448, 4.6]
|
2006-11-27 07:25:10 -07:00
|
|
|
*
|
|
|
|
* if (t_now > t_nom - delta)
|
|
|
|
* // send the packet now
|
|
|
|
* else
|
|
|
|
* // send the packet in (t_nom - t_now) milliseconds.
|
|
|
|
*/
|
dccp ccid-3: Bug fix for the inter-packet scheduling algorithm
This fixes a subtle bug in the calculation of the inter-packet gap and shows
that t_delta, as it is currently used, is not needed. And hence replaced.
The algorithm from RFC 3448, 4.6 below continually computes a send time t_nom,
which is initialised with the current time t_now; t_gran = 1E6 / HZ specifies
the scheduling granularity, s the packet size, and X the sending rate:
t_distance = t_nom - t_now; // in microseconds
t_delta = min(t_ipi, t_gran) / 2; // `delta' parameter in microseconds
if (t_distance >= t_delta) {
reschedule after (t_distance / 1000) milliseconds;
} else {
t_ipi = s / X; // inter-packet interval in usec
t_nom += t_ipi; // compute the next send time
send packet now;
}
1) Description of the bug
-------------------------
Rescheduling requires a conversion into milliseconds, due to this call chain:
* ccid3_hc_tx_send_packet() returns a timeout in milliseconds,
* this value is converted by msecs_to_jiffies() in dccp_write_xmit(),
* and finally used as jiffy-expires-value for sk_reset_timer().
The highest jiffy resolution with HZ=1000 is 1 millisecond, so using a higher
granularity does not make much sense here.
As a consequence, values of t_distance < 1000 are truncated to 0. This issue
has so far been resolved by using instead
if (t_distance >= t_delta + 1000)
reschedule after (t_distance / 1000) milliseconds;
The bug is in artificially inflating t_delta to t_delta' = t_delta + 1000. This
is unnecessarily large, a more adequate value is t_delta' = max(t_delta, 1000).
2) Consequences of using the corrected t_delta'
-----------------------------------------------
Since t_delta <= t_gran/2 = 10^6/(2*HZ), we have t_delta <= 1000 as long as
HZ >= 500. This means that t_delta' = max(1000, t_delta) is constant at 1000.
On the other hand, when using a coarse HZ value of HZ < 500, we have three
sub-cases that can all be reduced to using another constant of t_gran/2.
(a) The first case arises when t_ipi > t_gran. Here t_delta' is the constant
t_delta' = max(1000, t_gran/2) = t_gran/2.
(b) If t_ipi <= 2000 < t_gran = 10^6/HZ usec, then t_delta = t_ipi/2 <= 1000,
so that t_delta' = max(1000, t_delta) = 1000 < t_gran/2.
(c) If 2000 < t_ipi <= t_gran, we have t_delta' = max(t_delta, 1000) = t_ipi/2.
In the second and third cases we have delay values less than t_gran/2, which is
in the order of less than or equal to half a jiffy.
How these are treated depends on how fractions of a jiffy are handled: they
are either always rounded down to 0, or always rounded up to 1 jiffy (assuming
non-zero values). In both cases the error is on average in the order of 50%.
Thus we are not increasing the error when in the second/third case we replace
a value less than t_gran/2 with 0, by setting t_delta' to the constant t_gran/2.
3) Summary
----------
Fixing (1) and considering (2), the patch replaces t_delta with a constant,
whose value depends on CONFIG_HZ, changing the above algorithm to:
if (t_distance >= t_delta')
reschedule after (t_distance / 1000) milliseconds;
where t_delta' = 10^6/(2*HZ) if HZ < 500, and t_delta' = 1000 otherwise.
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
2008-09-03 22:30:19 -07:00
|
|
|
if (delay >= TFRC_T_DELTA)
|
|
|
|
return (u32)delay / USEC_PER_MSEC;
|
2006-12-09 19:07:37 -07:00
|
|
|
|
2007-06-16 09:34:02 -07:00
|
|
|
ccid3_hc_tx_update_win_count(hctx, now);
|
2005-08-09 20:14:34 -07:00
|
|
|
break;
|
2006-11-20 13:39:23 -07:00
|
|
|
case TFRC_SSTATE_TERM:
|
2006-12-09 19:14:12 -07:00
|
|
|
DCCP_BUG("%s(%p) - Illegal state TERM", dccp_role(sk), sk);
|
2006-11-27 07:26:03 -07:00
|
|
|
return -EINVAL;
|
2005-08-09 20:14:34 -07:00
|
|
|
}
|
|
|
|
|
2006-11-27 07:26:03 -07:00
|
|
|
/* prepare to send now (add options etc.) */
|
|
|
|
dp->dccps_hc_tx_insert_options = 1;
|
2008-09-03 22:30:19 -07:00
|
|
|
DCCP_SKB_CB(skb)->dccpd_ccval = hctx->last_win_count;
|
2006-12-09 19:08:09 -07:00
|
|
|
|
|
|
|
/* set the nominal send time for the next following packet */
|
2008-09-03 22:30:19 -07:00
|
|
|
hctx->t_nom = ktime_add_us(hctx->t_nom, hctx->t_ipi);
|
2006-11-27 07:26:03 -07:00
|
|
|
return 0;
|
2005-08-09 20:14:34 -07:00
|
|
|
}
|
|
|
|
|
2006-12-10 11:01:18 -07:00
|
|
|
static void ccid3_hc_tx_packet_sent(struct sock *sk, int more,
|
|
|
|
unsigned int len)
|
2005-08-09 20:14:34 -07:00
|
|
|
{
|
2005-09-08 22:40:58 -07:00
|
|
|
struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
|
2005-08-09 20:14:34 -07:00
|
|
|
|
2006-11-28 14:55:06 -07:00
|
|
|
ccid3_hc_tx_update_s(hctx, len);
|
2005-08-09 20:14:34 -07:00
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
if (tfrc_tx_hist_add(&hctx->hist, dccp_sk(sk)->dccps_gss))
|
[DCCP] ccid3: Perform history operations only after packet has been sent
This migrates all packet history operations into the routine
ccid3_hc_tx_packet_sent, thereby removing synchronization problems
that occur when, as before, the operations are spread over multiple
routines.
The following minor simplifications are also applied:
* several simplifications now follow from this change - several tests
are now no longer required
* removal of one unnecessary variable (dp)
Justification:
Currently packet history operations span two different routines,
one of which is likely to pass through several iterations of sleeping
and awakening.
The first routine, ccid3_hc_tx_send_packet, allocates an entry and
sets a few fields. The remaining fields are filled in when the second
routine (which is not within a sleeping context), ccid3_hc_tx_packet_sent,
is called. This has several strong drawbacks:
* it is not necessary to split history operations - all fields can be
filled in by the second routine
* the first routine is called multiple times, until a packet can be sent,
and sleeps meanwhile - this causes a lot of difficulties with regard to
keeping the list consistent
* since both routines do not have a producer-consumer like synchronization,
it is very difficult to maintain data across calls to these routines
* the fact that the routines are called in different contexts (sleeping, not
sleeping) adds further problems
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Acked-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>
2006-12-09 19:09:21 -07:00
|
|
|
DCCP_CRIT("packet history - out of memory!");
|
2005-08-09 20:14:34 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)
|
|
|
|
{
|
2005-09-08 22:40:58 -07:00
|
|
|
struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
|
2008-09-03 22:30:19 -07:00
|
|
|
struct ccid3_options_received *opt_recv = &hctx->options_received;
|
2008-09-03 22:30:19 -07:00
|
|
|
struct tfrc_tx_hist_entry *acked;
|
2007-08-19 17:18:13 -07:00
|
|
|
ktime_t now;
|
2006-11-28 13:34:34 -07:00
|
|
|
unsigned long t_nfb;
|
2007-03-20 11:24:37 -07:00
|
|
|
u32 pinv, r_sample;
|
2005-08-26 23:51:58 -07:00
|
|
|
|
2005-08-09 20:14:34 -07:00
|
|
|
/* we are only interested in ACKs */
|
|
|
|
if (!(DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_ACK ||
|
|
|
|
DCCP_SKB_CB(skb)->dccpd_type == DCCP_PKT_DATAACK))
|
|
|
|
return;
|
2007-12-17 05:25:06 -07:00
|
|
|
/* ... and only in the established state */
|
2008-09-03 22:30:19 -07:00
|
|
|
if (hctx->state != TFRC_SSTATE_FBACK &&
|
|
|
|
hctx->state != TFRC_SSTATE_NO_FBACK)
|
2007-12-17 05:25:06 -07:00
|
|
|
return;
|
2005-08-09 20:14:34 -07:00
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
/*
|
|
|
|
* Locate the acknowledged packet in the TX history.
|
|
|
|
*
|
|
|
|
* Returning "entry not found" here can for instance happen when
|
|
|
|
* - the host has not sent out anything (e.g. a passive server),
|
|
|
|
* - the Ack is outdated (packet with higher Ack number was received),
|
|
|
|
* - it is a bogus Ack (for a packet not sent on this connection).
|
|
|
|
*/
|
|
|
|
acked = tfrc_tx_hist_find_entry(hctx->hist, dccp_hdr_ack_seq(skb));
|
|
|
|
if (acked == NULL)
|
2007-12-17 05:25:06 -07:00
|
|
|
return;
|
2008-09-03 22:30:19 -07:00
|
|
|
/* For the sake of RTT sampling, ignore/remove all older entries */
|
|
|
|
tfrc_tx_hist_purge(&acked->next);
|
|
|
|
|
|
|
|
/* Update the moving average for the RTT estimate (RFC 3448, 4.3) */
|
|
|
|
now = ktime_get_real();
|
|
|
|
r_sample = dccp_sample_rtt(sk, ktime_us_delta(now, acked->stamp));
|
|
|
|
hctx->rtt = tfrc_ewma(hctx->rtt, r_sample, 9);
|
2005-08-09 20:14:34 -07:00
|
|
|
|
2007-12-17 05:25:06 -07:00
|
|
|
/* Update receive rate in units of 64 * bytes/second */
|
2008-09-03 22:30:19 -07:00
|
|
|
hctx->x_recv = opt_recv->ccid3or_receive_rate;
|
|
|
|
hctx->x_recv <<= 6;
|
2005-08-09 20:14:34 -07:00
|
|
|
|
2007-12-17 05:25:06 -07:00
|
|
|
/* Update loss event rate (which is scaled by 1e6) */
|
|
|
|
pinv = opt_recv->ccid3or_loss_event_rate;
|
|
|
|
if (pinv == ~0U || pinv == 0) /* see RFC 4342, 8.5 */
|
2008-09-03 22:30:19 -07:00
|
|
|
hctx->p = 0;
|
2007-12-17 05:25:06 -07:00
|
|
|
else /* can not exceed 100% */
|
2008-09-03 22:30:19 -07:00
|
|
|
hctx->p = scaled_div(1, pinv);
|
2008-09-03 22:30:19 -07:00
|
|
|
|
2007-12-17 07:48:47 -07:00
|
|
|
/*
|
|
|
|
* Update allowed sending rate X as per draft rfc3448bis-00, 4.2/3
|
|
|
|
*/
|
2008-09-03 22:30:19 -07:00
|
|
|
if (hctx->state == TFRC_SSTATE_NO_FBACK) {
|
2007-12-17 07:48:47 -07:00
|
|
|
ccid3_hc_tx_set_state(sk, TFRC_SSTATE_FBACK);
|
2006-12-03 09:50:56 -07:00
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
if (hctx->t_rto == 0) {
|
2007-12-17 07:48:47 -07:00
|
|
|
/*
|
|
|
|
* Initial feedback packet: Larger Initial Windows (4.2)
|
|
|
|
*/
|
2008-09-03 22:30:19 -07:00
|
|
|
hctx->x = rfc3390_initial_rate(sk);
|
|
|
|
hctx->t_ld = now;
|
2006-11-28 14:51:42 -07:00
|
|
|
|
2007-12-17 07:48:47 -07:00
|
|
|
ccid3_update_send_interval(hctx);
|
2005-08-09 20:14:34 -07:00
|
|
|
|
2007-12-17 07:48:47 -07:00
|
|
|
goto done_computing_x;
|
2008-09-03 22:30:19 -07:00
|
|
|
} else if (hctx->p == 0) {
|
2007-12-17 07:48:47 -07:00
|
|
|
/*
|
|
|
|
* First feedback after nofeedback timer expiry (4.3)
|
|
|
|
*/
|
|
|
|
goto done_computing_x;
|
|
|
|
}
|
|
|
|
}
|
2005-08-09 20:14:34 -07:00
|
|
|
|
2007-12-17 07:48:47 -07:00
|
|
|
/* Update sending rate (step 4 of [RFC 3448, 4.3]) */
|
2008-09-03 22:30:19 -07:00
|
|
|
if (hctx->p > 0)
|
|
|
|
hctx->x_calc = tfrc_calc_x(hctx->s, hctx->rtt, hctx->p);
|
2007-12-17 07:48:47 -07:00
|
|
|
ccid3_hc_tx_update_x(sk, &now);
|
2005-08-09 20:14:34 -07:00
|
|
|
|
2007-12-17 07:48:47 -07:00
|
|
|
done_computing_x:
|
|
|
|
ccid3_pr_debug("%s(%p), RTT=%uus (sample=%uus), s=%u, "
|
2007-12-17 05:25:06 -07:00
|
|
|
"p=%u, X_calc=%u, X_recv=%u, X=%u\n",
|
2008-09-03 22:30:19 -07:00
|
|
|
dccp_role(sk), sk, hctx->rtt, r_sample,
|
|
|
|
hctx->s, hctx->p, hctx->x_calc,
|
|
|
|
(unsigned)(hctx->x_recv >> 6),
|
|
|
|
(unsigned)(hctx->x >> 6));
|
2005-08-09 20:14:34 -07:00
|
|
|
|
2007-12-17 05:25:06 -07:00
|
|
|
/* unschedule no feedback timer */
|
2008-09-03 22:30:19 -07:00
|
|
|
sk_stop_timer(sk, &hctx->no_feedback_timer);
|
2005-08-09 20:14:34 -07:00
|
|
|
|
2007-12-17 05:25:06 -07:00
|
|
|
/*
|
|
|
|
* As we have calculated new ipi, delta, t_nom it is possible
|
|
|
|
* that we now can send a packet, so wake up dccp_wait_for_ccid
|
|
|
|
*/
|
|
|
|
sk->sk_write_space(sk);
|
2005-08-10 08:59:38 -07:00
|
|
|
|
2007-12-17 05:25:06 -07:00
|
|
|
/*
|
|
|
|
* Update timeout interval for the nofeedback timer.
|
|
|
|
* We use a configuration option to increase the lower bound.
|
|
|
|
* This can help avoid triggering the nofeedback timer too
|
|
|
|
* often ('spinning') on LANs with small RTTs.
|
|
|
|
*/
|
2008-09-03 22:30:19 -07:00
|
|
|
hctx->t_rto = max_t(u32, 4 * hctx->rtt, (CONFIG_IP_DCCP_CCID3_RTO *
|
|
|
|
(USEC_PER_SEC / 1000)));
|
2007-12-17 05:25:06 -07:00
|
|
|
/*
|
|
|
|
* Schedule no feedback timer to expire in
|
|
|
|
* max(t_RTO, 2 * s/X) = max(t_RTO, 2 * t_ipi)
|
|
|
|
*/
|
2008-09-03 22:30:19 -07:00
|
|
|
t_nfb = max(hctx->t_rto, 2 * hctx->t_ipi);
|
2005-08-09 20:14:34 -07:00
|
|
|
|
2007-12-17 05:25:06 -07:00
|
|
|
ccid3_pr_debug("%s(%p), Scheduled no feedback timer to "
|
|
|
|
"expire in %lu jiffies (%luus)\n",
|
2008-09-03 22:30:19 -07:00
|
|
|
dccp_role(sk), sk, usecs_to_jiffies(t_nfb), t_nfb);
|
2006-12-09 19:14:12 -07:00
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
sk_reset_timer(sk, &hctx->no_feedback_timer,
|
2007-12-17 05:25:06 -07:00
|
|
|
jiffies + usecs_to_jiffies(t_nfb));
|
2005-08-09 20:14:34 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
static int ccid3_hc_tx_parse_options(struct sock *sk, unsigned char option,
|
2005-08-26 23:51:58 -07:00
|
|
|
unsigned char len, u16 idx,
|
|
|
|
unsigned char *value)
|
2005-08-09 20:14:34 -07:00
|
|
|
{
|
|
|
|
int rc = 0;
|
2005-09-08 22:40:58 -07:00
|
|
|
const struct dccp_sock *dp = dccp_sk(sk);
|
|
|
|
struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
|
2008-09-03 22:30:19 -07:00
|
|
|
struct ccid3_options_received *opt_recv = &hctx->options_received;
|
2007-10-24 05:46:58 -07:00
|
|
|
__be32 opt_val;
|
2005-08-09 20:14:34 -07:00
|
|
|
|
|
|
|
if (opt_recv->ccid3or_seqno != dp->dccps_gsr) {
|
|
|
|
opt_recv->ccid3or_seqno = dp->dccps_gsr;
|
|
|
|
opt_recv->ccid3or_loss_event_rate = ~0;
|
|
|
|
opt_recv->ccid3or_loss_intervals_idx = 0;
|
|
|
|
opt_recv->ccid3or_loss_intervals_len = 0;
|
|
|
|
opt_recv->ccid3or_receive_rate = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
switch (option) {
|
|
|
|
case TFRC_OPT_LOSS_EVENT_RATE:
|
2005-09-09 16:01:25 -07:00
|
|
|
if (unlikely(len != 4)) {
|
2006-12-09 19:14:12 -07:00
|
|
|
DCCP_WARN("%s(%p), invalid len %d "
|
2006-11-20 13:39:23 -07:00
|
|
|
"for TFRC_OPT_LOSS_EVENT_RATE\n",
|
|
|
|
dccp_role(sk), sk, len);
|
2005-08-09 20:14:34 -07:00
|
|
|
rc = -EINVAL;
|
|
|
|
} else {
|
2007-10-24 05:46:58 -07:00
|
|
|
opt_val = get_unaligned((__be32 *)value);
|
|
|
|
opt_recv->ccid3or_loss_event_rate = ntohl(opt_val);
|
2006-12-09 19:14:12 -07:00
|
|
|
ccid3_pr_debug("%s(%p), LOSS_EVENT_RATE=%u\n",
|
2005-08-09 20:14:34 -07:00
|
|
|
dccp_role(sk), sk,
|
|
|
|
opt_recv->ccid3or_loss_event_rate);
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case TFRC_OPT_LOSS_INTERVALS:
|
|
|
|
opt_recv->ccid3or_loss_intervals_idx = idx;
|
|
|
|
opt_recv->ccid3or_loss_intervals_len = len;
|
2006-12-09 19:14:12 -07:00
|
|
|
ccid3_pr_debug("%s(%p), LOSS_INTERVALS=(%u, %u)\n",
|
2005-08-09 20:14:34 -07:00
|
|
|
dccp_role(sk), sk,
|
|
|
|
opt_recv->ccid3or_loss_intervals_idx,
|
|
|
|
opt_recv->ccid3or_loss_intervals_len);
|
|
|
|
break;
|
|
|
|
case TFRC_OPT_RECEIVE_RATE:
|
2005-09-09 16:01:25 -07:00
|
|
|
if (unlikely(len != 4)) {
|
2006-12-09 19:14:12 -07:00
|
|
|
DCCP_WARN("%s(%p), invalid len %d "
|
2006-11-20 13:39:23 -07:00
|
|
|
"for TFRC_OPT_RECEIVE_RATE\n",
|
|
|
|
dccp_role(sk), sk, len);
|
2005-08-09 20:14:34 -07:00
|
|
|
rc = -EINVAL;
|
|
|
|
} else {
|
2007-10-24 05:46:58 -07:00
|
|
|
opt_val = get_unaligned((__be32 *)value);
|
|
|
|
opt_recv->ccid3or_receive_rate = ntohl(opt_val);
|
2006-12-09 19:14:12 -07:00
|
|
|
ccid3_pr_debug("%s(%p), RECEIVE_RATE=%u\n",
|
2005-08-09 20:14:34 -07:00
|
|
|
dccp_role(sk), sk,
|
|
|
|
opt_recv->ccid3or_receive_rate);
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
return rc;
|
|
|
|
}
|
|
|
|
|
2006-03-20 20:21:44 -07:00
|
|
|
static int ccid3_hc_tx_init(struct ccid *ccid, struct sock *sk)
|
2005-08-09 20:14:34 -07:00
|
|
|
{
|
2006-03-20 20:21:44 -07:00
|
|
|
struct ccid3_hc_tx_sock *hctx = ccid_priv(ccid);
|
2005-08-09 20:14:34 -07:00
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
hctx->state = TFRC_SSTATE_NO_SENT;
|
|
|
|
hctx->hist = NULL;
|
|
|
|
setup_timer(&hctx->no_feedback_timer,
|
|
|
|
ccid3_hc_tx_no_feedback_timer, (unsigned long)sk);
|
2005-08-09 20:14:34 -07:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void ccid3_hc_tx_exit(struct sock *sk)
|
|
|
|
{
|
2005-09-08 22:40:58 -07:00
|
|
|
struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
|
2005-08-09 20:14:34 -07:00
|
|
|
|
|
|
|
ccid3_hc_tx_set_state(sk, TFRC_SSTATE_TERM);
|
2008-09-03 22:30:19 -07:00
|
|
|
sk_stop_timer(sk, &hctx->no_feedback_timer);
|
2005-08-09 20:14:34 -07:00
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
tfrc_tx_hist_purge(&hctx->hist);
|
2005-08-09 20:14:34 -07:00
|
|
|
}
|
|
|
|
|
2007-03-20 09:11:24 -07:00
|
|
|
static void ccid3_hc_tx_get_info(struct sock *sk, struct tcp_info *info)
|
|
|
|
{
|
dccp ccid-3: No more CCID control blocks in LISTEN state
The CCIDs are activated as last of the features, at the end of the handshake,
were the LISTEN state of the master socket is inherited into the server
state of the child socket. Thus, the only states visible to CCIDs now are
OPEN/PARTOPEN, and the closing states.
This allows to remove tests which were previously necessary to protect
against referencing a socket in the listening state (in CCID3), but which
now have become redundant.
As a further byproduct of enabling the CCIDs only after the connection has been
fully established, several typecast-initialisations of ccid3_hc_{rx,tx}_sock
can now be eliminated:
* the CCID is loaded, so it is not necessary to test if it is NULL,
* if it is possible to load a CCID and leave the private area NULL, then this
is a bug, which should crash loudly - and earlier,
* the test for state==OPEN || state==PARTOPEN now reduces only to the closing
phase (e.g. when the node has received an unexpected Reset).
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Acked-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
2008-09-03 22:30:19 -07:00
|
|
|
info->tcpi_rto = ccid3_hc_tx_sk(sk)->t_rto;
|
|
|
|
info->tcpi_rtt = ccid3_hc_tx_sk(sk)->rtt;
|
2007-03-20 09:11:24 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
static int ccid3_hc_tx_getsockopt(struct sock *sk, const int optname, int len,
|
|
|
|
u32 __user *optval, int __user *optlen)
|
|
|
|
{
|
dccp ccid-3: No more CCID control blocks in LISTEN state
The CCIDs are activated as last of the features, at the end of the handshake,
were the LISTEN state of the master socket is inherited into the server
state of the child socket. Thus, the only states visible to CCIDs now are
OPEN/PARTOPEN, and the closing states.
This allows to remove tests which were previously necessary to protect
against referencing a socket in the listening state (in CCID3), but which
now have become redundant.
As a further byproduct of enabling the CCIDs only after the connection has been
fully established, several typecast-initialisations of ccid3_hc_{rx,tx}_sock
can now be eliminated:
* the CCID is loaded, so it is not necessary to test if it is NULL,
* if it is possible to load a CCID and leave the private area NULL, then this
is a bug, which should crash loudly - and earlier,
* the test for state==OPEN || state==PARTOPEN now reduces only to the closing
phase (e.g. when the node has received an unexpected Reset).
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Acked-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
2008-09-03 22:30:19 -07:00
|
|
|
const struct ccid3_hc_tx_sock *hctx = ccid3_hc_tx_sk(sk);
|
2008-09-03 22:30:19 -07:00
|
|
|
struct tfrc_tx_info tfrc;
|
2007-03-20 09:11:24 -07:00
|
|
|
const void *val;
|
|
|
|
|
|
|
|
switch (optname) {
|
|
|
|
case DCCP_SOCKOPT_CCID_TX_INFO:
|
2008-09-03 22:30:19 -07:00
|
|
|
if (len < sizeof(tfrc))
|
2007-03-20 09:11:24 -07:00
|
|
|
return -EINVAL;
|
2008-09-03 22:30:19 -07:00
|
|
|
tfrc.tfrctx_x = hctx->x;
|
|
|
|
tfrc.tfrctx_x_recv = hctx->x_recv;
|
|
|
|
tfrc.tfrctx_x_calc = hctx->x_calc;
|
|
|
|
tfrc.tfrctx_rtt = hctx->rtt;
|
|
|
|
tfrc.tfrctx_p = hctx->p;
|
|
|
|
tfrc.tfrctx_rto = hctx->t_rto;
|
|
|
|
tfrc.tfrctx_ipi = hctx->t_ipi;
|
|
|
|
len = sizeof(tfrc);
|
|
|
|
val = &tfrc;
|
2007-03-20 09:11:24 -07:00
|
|
|
break;
|
|
|
|
default:
|
|
|
|
return -ENOPROTOOPT;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (put_user(len, optlen) || copy_to_user(optval, val, len))
|
|
|
|
return -EFAULT;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2005-08-09 20:14:34 -07:00
|
|
|
/*
|
2007-03-20 09:11:24 -07:00
|
|
|
* Receiver Half-Connection Routines
|
2005-08-09 20:14:34 -07:00
|
|
|
*/
|
2007-12-06 08:18:11 -07:00
|
|
|
|
|
|
|
/* CCID3 feedback types */
|
|
|
|
enum ccid3_fback_type {
|
|
|
|
CCID3_FBACK_NONE = 0,
|
|
|
|
CCID3_FBACK_INITIAL,
|
|
|
|
CCID3_FBACK_PERIODIC,
|
|
|
|
CCID3_FBACK_PARAM_CHANGE
|
|
|
|
};
|
|
|
|
|
2006-11-20 13:28:09 -07:00
|
|
|
#ifdef CONFIG_IP_DCCP_CCID3_DEBUG
|
2005-08-09 20:14:34 -07:00
|
|
|
static const char *ccid3_rx_state_name(enum ccid3_hc_rx_states state)
|
|
|
|
{
|
|
|
|
static char *ccid3_rx_state_names[] = {
|
|
|
|
[TFRC_RSTATE_NO_DATA] = "NO_DATA",
|
|
|
|
[TFRC_RSTATE_DATA] = "DATA",
|
|
|
|
[TFRC_RSTATE_TERM] = "TERM",
|
|
|
|
};
|
|
|
|
|
|
|
|
return ccid3_rx_state_names[state];
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2006-03-20 22:58:56 -07:00
|
|
|
static void ccid3_hc_rx_set_state(struct sock *sk,
|
|
|
|
enum ccid3_hc_rx_states state)
|
2005-08-09 20:14:34 -07:00
|
|
|
{
|
2005-09-08 22:40:58 -07:00
|
|
|
struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
|
2008-09-03 22:30:19 -07:00
|
|
|
enum ccid3_hc_rx_states oldstate = hcrx->state;
|
2005-08-09 20:14:34 -07:00
|
|
|
|
|
|
|
ccid3_pr_debug("%s(%p) %-8.8s -> %s\n",
|
2005-08-26 23:51:58 -07:00
|
|
|
dccp_role(sk), sk, ccid3_rx_state_name(oldstate),
|
|
|
|
ccid3_rx_state_name(state));
|
2005-08-09 20:14:34 -07:00
|
|
|
WARN_ON(state == oldstate);
|
2008-09-03 22:30:19 -07:00
|
|
|
hcrx->state = state;
|
2005-08-09 20:14:34 -07:00
|
|
|
}
|
|
|
|
|
2007-12-06 08:18:11 -07:00
|
|
|
static void ccid3_hc_rx_send_feedback(struct sock *sk,
|
|
|
|
const struct sk_buff *skb,
|
|
|
|
enum ccid3_fback_type fbtype)
|
2005-08-09 20:14:34 -07:00
|
|
|
{
|
2005-09-08 22:40:58 -07:00
|
|
|
struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
|
2005-08-09 20:14:34 -07:00
|
|
|
struct dccp_sock *dp = dccp_sk(sk);
|
2007-08-19 17:17:51 -07:00
|
|
|
ktime_t now;
|
2007-12-06 08:18:11 -07:00
|
|
|
s64 delta = 0;
|
2005-08-09 20:14:34 -07:00
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
if (unlikely(hcrx->state == TFRC_RSTATE_TERM))
|
2007-12-06 08:18:11 -07:00
|
|
|
return;
|
|
|
|
|
2007-08-19 17:15:37 -07:00
|
|
|
now = ktime_get_real();
|
2005-08-27 14:18:18 -07:00
|
|
|
|
2007-12-06 08:18:11 -07:00
|
|
|
switch (fbtype) {
|
|
|
|
case CCID3_FBACK_INITIAL:
|
2008-09-03 22:30:19 -07:00
|
|
|
hcrx->x_recv = 0;
|
|
|
|
hcrx->p_inverse = ~0U; /* see RFC 4342, 8.5 */
|
2005-08-09 20:14:34 -07:00
|
|
|
break;
|
2007-12-06 08:18:11 -07:00
|
|
|
case CCID3_FBACK_PARAM_CHANGE:
|
|
|
|
/*
|
|
|
|
* When parameters change (new loss or p > p_prev), we do not
|
|
|
|
* have a reliable estimate for R_m of [RFC 3448, 6.2] and so
|
|
|
|
* need to reuse the previous value of X_recv. However, when
|
|
|
|
* X_recv was 0 (due to early loss), this would kill X down to
|
|
|
|
* s/t_mbi (i.e. one packet in 64 seconds).
|
|
|
|
* To avoid such drastic reduction, we approximate X_recv as
|
|
|
|
* the number of bytes since last feedback.
|
|
|
|
* This is a safe fallback, since X is bounded above by X_calc.
|
|
|
|
*/
|
2008-09-03 22:30:19 -07:00
|
|
|
if (hcrx->x_recv > 0)
|
2007-12-06 08:18:11 -07:00
|
|
|
break;
|
|
|
|
/* fall through */
|
|
|
|
case CCID3_FBACK_PERIODIC:
|
2008-09-03 22:30:19 -07:00
|
|
|
delta = ktime_us_delta(now, hcrx->tstamp_last_feedback);
|
2007-12-06 08:18:11 -07:00
|
|
|
if (delta <= 0)
|
|
|
|
DCCP_BUG("delta (%ld) <= 0", (long)delta);
|
|
|
|
else
|
2008-09-03 22:30:19 -07:00
|
|
|
hcrx->x_recv = scaled_div32(hcrx->bytes_recv, delta);
|
2005-08-09 20:14:34 -07:00
|
|
|
break;
|
2007-12-06 08:18:11 -07:00
|
|
|
default:
|
2005-08-09 20:14:34 -07:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
ccid3_pr_debug("Interval %ldusec, X_recv=%u, 1/p=%u\n",
|
|
|
|
(long)delta, hcrx->x_recv, hcrx->p_inverse);
|
2005-08-09 20:14:34 -07:00
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
hcrx->tstamp_last_feedback = now;
|
|
|
|
hcrx->last_counter = dccp_hdr(skb)->dccph_ccval;
|
|
|
|
hcrx->bytes_recv = 0;
|
2005-08-09 20:14:34 -07:00
|
|
|
|
2005-09-08 22:30:07 -07:00
|
|
|
dp->dccps_hc_rx_insert_options = 1;
|
2005-08-09 20:14:34 -07:00
|
|
|
dccp_send_ack(sk);
|
|
|
|
}
|
|
|
|
|
2006-03-20 23:32:06 -07:00
|
|
|
static int ccid3_hc_rx_insert_options(struct sock *sk, struct sk_buff *skb)
|
2005-08-09 20:14:34 -07:00
|
|
|
{
|
dccp ccid-3: No more CCID control blocks in LISTEN state
The CCIDs are activated as last of the features, at the end of the handshake,
were the LISTEN state of the master socket is inherited into the server
state of the child socket. Thus, the only states visible to CCIDs now are
OPEN/PARTOPEN, and the closing states.
This allows to remove tests which were previously necessary to protect
against referencing a socket in the listening state (in CCID3), but which
now have become redundant.
As a further byproduct of enabling the CCIDs only after the connection has been
fully established, several typecast-initialisations of ccid3_hc_{rx,tx}_sock
can now be eliminated:
* the CCID is loaded, so it is not necessary to test if it is NULL,
* if it is possible to load a CCID and leave the private area NULL, then this
is a bug, which should crash loudly - and earlier,
* the test for state==OPEN || state==PARTOPEN now reduces only to the closing
phase (e.g. when the node has received an unexpected Reset).
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Acked-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
2008-09-03 22:30:19 -07:00
|
|
|
const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
|
2006-03-20 20:23:32 -07:00
|
|
|
__be32 x_recv, pinv;
|
2005-08-09 20:14:34 -07:00
|
|
|
|
2005-09-09 16:01:25 -07:00
|
|
|
if (!(sk->sk_state == DCCP_OPEN || sk->sk_state == DCCP_PARTOPEN))
|
2006-03-20 23:32:06 -07:00
|
|
|
return 0;
|
2005-08-09 20:14:34 -07:00
|
|
|
|
2005-08-23 21:51:59 -07:00
|
|
|
if (dccp_packet_without_ack(skb))
|
2006-03-20 23:32:06 -07:00
|
|
|
return 0;
|
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
x_recv = htonl(hcrx->x_recv);
|
|
|
|
pinv = htonl(hcrx->p_inverse);
|
2006-03-20 23:32:06 -07:00
|
|
|
|
2007-12-08 11:26:59 -07:00
|
|
|
if (dccp_insert_option(sk, skb, TFRC_OPT_LOSS_EVENT_RATE,
|
2006-12-10 11:01:18 -07:00
|
|
|
&pinv, sizeof(pinv)) ||
|
2006-03-20 23:32:06 -07:00
|
|
|
dccp_insert_option(sk, skb, TFRC_OPT_RECEIVE_RATE,
|
2006-12-10 11:01:18 -07:00
|
|
|
&x_recv, sizeof(x_recv)))
|
2006-03-20 23:32:06 -07:00
|
|
|
return -1;
|
|
|
|
|
|
|
|
return 0;
|
2005-08-09 20:14:34 -07:00
|
|
|
}
|
|
|
|
|
2007-12-12 09:06:14 -07:00
|
|
|
/** ccid3_first_li - Implements [RFC 3448, 6.3.1]
|
|
|
|
*
|
|
|
|
* Determine the length of the first loss interval via inverse lookup.
|
|
|
|
* Assume that X_recv can be computed by the throughput equation
|
|
|
|
* s
|
|
|
|
* X_recv = --------
|
|
|
|
* R * fval
|
|
|
|
* Find some p such that f(p) = fval; return 1/p (scaled).
|
|
|
|
*/
|
|
|
|
static u32 ccid3_first_li(struct sock *sk)
|
|
|
|
{
|
|
|
|
struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
|
|
|
|
u32 x_recv, p, delta;
|
|
|
|
u64 fval;
|
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
if (hcrx->rtt == 0) {
|
2007-12-12 09:06:14 -07:00
|
|
|
DCCP_WARN("No RTT estimate available, using fallback RTT\n");
|
2008-09-03 22:30:19 -07:00
|
|
|
hcrx->rtt = DCCP_FALLBACK_RTT;
|
2007-12-12 09:06:14 -07:00
|
|
|
}
|
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
delta = ktime_to_us(net_timedelta(hcrx->tstamp_last_feedback));
|
|
|
|
x_recv = scaled_div32(hcrx->bytes_recv, delta);
|
2007-12-12 09:06:14 -07:00
|
|
|
if (x_recv == 0) { /* would also trigger divide-by-zero */
|
|
|
|
DCCP_WARN("X_recv==0\n");
|
2008-09-03 22:30:19 -07:00
|
|
|
if (hcrx->x_recv == 0) {
|
2007-12-12 09:06:14 -07:00
|
|
|
DCCP_BUG("stored value of X_recv is zero");
|
|
|
|
return ~0U;
|
|
|
|
}
|
2008-09-03 22:30:19 -07:00
|
|
|
x_recv = hcrx->x_recv;
|
2007-12-12 09:06:14 -07:00
|
|
|
}
|
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
fval = scaled_div(hcrx->s, hcrx->rtt);
|
2007-12-12 09:06:14 -07:00
|
|
|
fval = scaled_div32(fval, x_recv);
|
|
|
|
p = tfrc_calc_x_reverse_lookup(fval);
|
|
|
|
|
|
|
|
ccid3_pr_debug("%s(%p), receive rate=%u bytes/s, implied "
|
|
|
|
"loss rate=%u\n", dccp_role(sk), sk, x_recv, p);
|
|
|
|
|
|
|
|
return p == 0 ? ~0U : scaled_div(1, p);
|
|
|
|
}
|
|
|
|
|
2007-12-06 08:18:11 -07:00
|
|
|
static void ccid3_hc_rx_packet_recv(struct sock *sk, struct sk_buff *skb)
|
2005-08-09 20:14:34 -07:00
|
|
|
{
|
2005-09-08 22:40:58 -07:00
|
|
|
struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
|
2007-12-06 08:18:11 -07:00
|
|
|
enum ccid3_fback_type do_feedback = CCID3_FBACK_NONE;
|
2008-07-13 03:51:40 -07:00
|
|
|
const u64 ndp = dccp_sk(sk)->dccps_options_received.dccpor_ndp;
|
2007-12-06 08:18:11 -07:00
|
|
|
const bool is_data_packet = dccp_data_packet(skb);
|
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
if (unlikely(hcrx->state == TFRC_RSTATE_NO_DATA)) {
|
2007-12-06 08:18:11 -07:00
|
|
|
if (is_data_packet) {
|
|
|
|
const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
|
|
|
|
do_feedback = CCID3_FBACK_INITIAL;
|
|
|
|
ccid3_hc_rx_set_state(sk, TFRC_RSTATE_DATA);
|
2008-09-03 22:30:19 -07:00
|
|
|
hcrx->s = payload;
|
2007-12-06 08:18:11 -07:00
|
|
|
/*
|
2008-09-03 22:30:19 -07:00
|
|
|
* Not necessary to update bytes_recv here,
|
2007-12-06 08:18:11 -07:00
|
|
|
* since X_recv = 0 for the first feedback packet (cf.
|
|
|
|
* RFC 3448, 6.3) -- gerrit
|
|
|
|
*/
|
2006-08-26 23:40:50 -07:00
|
|
|
}
|
2007-12-06 08:18:11 -07:00
|
|
|
goto update_records;
|
2006-08-26 23:40:50 -07:00
|
|
|
}
|
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
if (tfrc_rx_hist_duplicate(&hcrx->hist, skb))
|
2007-12-06 08:18:11 -07:00
|
|
|
return; /* done receiving */
|
2005-08-09 20:14:34 -07:00
|
|
|
|
2007-12-06 08:18:11 -07:00
|
|
|
if (is_data_packet) {
|
|
|
|
const u32 payload = skb->len - dccp_hdr(skb)->dccph_doff * 4;
|
|
|
|
/*
|
|
|
|
* Update moving-average of s and the sum of received payload bytes
|
|
|
|
*/
|
2008-09-03 22:30:19 -07:00
|
|
|
hcrx->s = tfrc_ewma(hcrx->s, payload, 9);
|
|
|
|
hcrx->bytes_recv += payload;
|
2005-08-09 20:14:34 -07:00
|
|
|
}
|
|
|
|
|
2007-12-06 08:18:11 -07:00
|
|
|
/*
|
2008-07-13 03:51:40 -07:00
|
|
|
* Perform loss detection and handle pending losses
|
2007-12-06 08:18:11 -07:00
|
|
|
*/
|
2008-09-03 22:30:19 -07:00
|
|
|
if (tfrc_rx_handle_loss(&hcrx->hist, &hcrx->li_hist,
|
2008-07-13 03:51:40 -07:00
|
|
|
skb, ndp, ccid3_first_li, sk)) {
|
2007-12-12 09:06:14 -07:00
|
|
|
do_feedback = CCID3_FBACK_PARAM_CHANGE;
|
|
|
|
goto done_receiving;
|
|
|
|
}
|
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
if (tfrc_rx_hist_loss_pending(&hcrx->hist))
|
2008-07-13 03:51:40 -07:00
|
|
|
return; /* done receiving */
|
2006-11-28 14:22:33 -07:00
|
|
|
|
2007-12-06 08:18:11 -07:00
|
|
|
/*
|
|
|
|
* Handle data packets: RTT sampling and monitoring p
|
|
|
|
*/
|
|
|
|
if (unlikely(!is_data_packet))
|
|
|
|
goto update_records;
|
2005-08-27 14:18:18 -07:00
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
if (!tfrc_lh_is_initialised(&hcrx->li_hist)) {
|
|
|
|
const u32 sample = tfrc_rx_hist_sample_rtt(&hcrx->hist, skb);
|
2007-12-06 08:18:11 -07:00
|
|
|
/*
|
|
|
|
* Empty loss history: no loss so far, hence p stays 0.
|
|
|
|
* Sample RTT values, since an RTT estimate is required for the
|
|
|
|
* computation of p when the first loss occurs; RFC 3448, 6.3.1.
|
|
|
|
*/
|
|
|
|
if (sample != 0)
|
2008-09-03 22:30:19 -07:00
|
|
|
hcrx->rtt = tfrc_ewma(hcrx->rtt, sample, 9);
|
2007-12-12 09:06:14 -07:00
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
} else if (tfrc_lh_update_i_mean(&hcrx->li_hist, skb)) {
|
2007-12-12 09:06:14 -07:00
|
|
|
/*
|
|
|
|
* Step (3) of [RFC 3448, 6.1]: Recompute I_mean and, if I_mean
|
|
|
|
* has decreased (resp. p has increased), send feedback now.
|
|
|
|
*/
|
|
|
|
do_feedback = CCID3_FBACK_PARAM_CHANGE;
|
2005-08-09 20:14:34 -07:00
|
|
|
}
|
|
|
|
|
2007-12-06 08:18:11 -07:00
|
|
|
/*
|
|
|
|
* Check if the periodic once-per-RTT feedback is due; RFC 4342, 10.3
|
|
|
|
*/
|
2008-09-03 22:30:19 -07:00
|
|
|
if (SUB16(dccp_hdr(skb)->dccph_ccval, hcrx->last_counter) > 3)
|
2007-12-06 08:18:11 -07:00
|
|
|
do_feedback = CCID3_FBACK_PERIODIC;
|
2006-03-03 18:54:46 -07:00
|
|
|
|
2007-12-06 08:18:11 -07:00
|
|
|
update_records:
|
2008-09-03 22:30:19 -07:00
|
|
|
tfrc_rx_hist_add_packet(&hcrx->hist, skb, ndp);
|
2005-08-09 20:14:34 -07:00
|
|
|
|
2007-12-12 09:06:14 -07:00
|
|
|
done_receiving:
|
2007-12-06 08:18:11 -07:00
|
|
|
if (do_feedback)
|
|
|
|
ccid3_hc_rx_send_feedback(sk, skb, do_feedback);
|
2005-08-09 20:14:34 -07:00
|
|
|
}
|
|
|
|
|
2006-03-20 20:21:44 -07:00
|
|
|
static int ccid3_hc_rx_init(struct ccid *ccid, struct sock *sk)
|
2005-08-09 20:14:34 -07:00
|
|
|
{
|
2006-03-20 20:21:44 -07:00
|
|
|
struct ccid3_hc_rx_sock *hcrx = ccid_priv(ccid);
|
2005-08-09 20:14:34 -07:00
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
hcrx->state = TFRC_RSTATE_NO_DATA;
|
|
|
|
tfrc_lh_init(&hcrx->li_hist);
|
|
|
|
return tfrc_rx_hist_alloc(&hcrx->hist);
|
2005-08-09 20:14:34 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
static void ccid3_hc_rx_exit(struct sock *sk)
|
|
|
|
{
|
2005-09-08 22:40:58 -07:00
|
|
|
struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
|
2005-08-09 20:14:34 -07:00
|
|
|
|
|
|
|
ccid3_hc_rx_set_state(sk, TFRC_RSTATE_TERM);
|
|
|
|
|
2008-09-03 22:30:19 -07:00
|
|
|
tfrc_rx_hist_purge(&hcrx->hist);
|
|
|
|
tfrc_lh_cleanup(&hcrx->li_hist);
|
2005-08-09 20:14:34 -07:00
|
|
|
}
|
|
|
|
|
[DCCP]: Introduce dccp_get_info
And also hc_tx and hc_rx get_info functions for the CCIDs to fill in
information that is specific to them.
For now reusing struct tcp_info, later I'll try to figure out a better
solution, for now its really nice to get this kind of info:
[root@qemu ~]# ./ss -danemi
State Recv-Q Send-Q Local Addr:Port Peer Addr:Port
LISTEN 0 0 *:5001 *:* ino:628 sk:c1340040
mem:(r0,w0,f0,t0) cwnd:0 ssthresh:0
ESTAB 0 0 172.20.0.2:5001 172.20.0.1:32785 ino:629 sk:c13409a0
mem:(r0,w0,f0,t0) ts rto:1000 rtt:0.004/0 cwnd:0 ssthresh:0 rcv_rtt:61.377
This, for instance, shows that we're not congestion controlling ACKs,
as the above output is in the ttcp receiving host, and ttcp is a one
way app, i.e. the received never calls sendmsg, so
ccid_hc_tx_send_packet is never called, so the TX half connection
stays in TFRC_SSTATE_NO_SENT state and hctx_rtt is never calculated,
stays with the value set in ccid3_hc_tx_init, 4us, as show above in
milliseconds (0.004ms), upcoming patches will fix this.
rcv_rtt seems sane tho, matching ping results :-)
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2005-08-23 21:52:35 -07:00
|
|
|
static void ccid3_hc_rx_get_info(struct sock *sk, struct tcp_info *info)
|
|
|
|
{
|
dccp ccid-3: No more CCID control blocks in LISTEN state
The CCIDs are activated as last of the features, at the end of the handshake,
were the LISTEN state of the master socket is inherited into the server
state of the child socket. Thus, the only states visible to CCIDs now are
OPEN/PARTOPEN, and the closing states.
This allows to remove tests which were previously necessary to protect
against referencing a socket in the listening state (in CCID3), but which
now have become redundant.
As a further byproduct of enabling the CCIDs only after the connection has been
fully established, several typecast-initialisations of ccid3_hc_{rx,tx}_sock
can now be eliminated:
* the CCID is loaded, so it is not necessary to test if it is NULL,
* if it is possible to load a CCID and leave the private area NULL, then this
is a bug, which should crash loudly - and earlier,
* the test for state==OPEN || state==PARTOPEN now reduces only to the closing
phase (e.g. when the node has received an unexpected Reset).
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Acked-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
2008-09-03 22:30:19 -07:00
|
|
|
info->tcpi_ca_state = ccid3_hc_rx_sk(sk)->state;
|
2006-12-10 11:01:18 -07:00
|
|
|
info->tcpi_options |= TCPI_OPT_TIMESTAMPS;
|
dccp ccid-3: No more CCID control blocks in LISTEN state
The CCIDs are activated as last of the features, at the end of the handshake,
were the LISTEN state of the master socket is inherited into the server
state of the child socket. Thus, the only states visible to CCIDs now are
OPEN/PARTOPEN, and the closing states.
This allows to remove tests which were previously necessary to protect
against referencing a socket in the listening state (in CCID3), but which
now have become redundant.
As a further byproduct of enabling the CCIDs only after the connection has been
fully established, several typecast-initialisations of ccid3_hc_{rx,tx}_sock
can now be eliminated:
* the CCID is loaded, so it is not necessary to test if it is NULL,
* if it is possible to load a CCID and leave the private area NULL, then this
is a bug, which should crash loudly - and earlier,
* the test for state==OPEN || state==PARTOPEN now reduces only to the closing
phase (e.g. when the node has received an unexpected Reset).
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Acked-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
2008-09-03 22:30:19 -07:00
|
|
|
info->tcpi_rcv_rtt = ccid3_hc_rx_sk(sk)->rtt;
|
[DCCP]: Introduce dccp_get_info
And also hc_tx and hc_rx get_info functions for the CCIDs to fill in
information that is specific to them.
For now reusing struct tcp_info, later I'll try to figure out a better
solution, for now its really nice to get this kind of info:
[root@qemu ~]# ./ss -danemi
State Recv-Q Send-Q Local Addr:Port Peer Addr:Port
LISTEN 0 0 *:5001 *:* ino:628 sk:c1340040
mem:(r0,w0,f0,t0) cwnd:0 ssthresh:0
ESTAB 0 0 172.20.0.2:5001 172.20.0.1:32785 ino:629 sk:c13409a0
mem:(r0,w0,f0,t0) ts rto:1000 rtt:0.004/0 cwnd:0 ssthresh:0 rcv_rtt:61.377
This, for instance, shows that we're not congestion controlling ACKs,
as the above output is in the ttcp receiving host, and ttcp is a one
way app, i.e. the received never calls sendmsg, so
ccid_hc_tx_send_packet is never called, so the TX half connection
stays in TFRC_SSTATE_NO_SENT state and hctx_rtt is never calculated,
stays with the value set in ccid3_hc_tx_init, 4us, as show above in
milliseconds (0.004ms), upcoming patches will fix this.
rcv_rtt seems sane tho, matching ping results :-)
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2005-08-23 21:52:35 -07:00
|
|
|
}
|
|
|
|
|
2005-09-18 00:19:32 -07:00
|
|
|
static int ccid3_hc_rx_getsockopt(struct sock *sk, const int optname, int len,
|
|
|
|
u32 __user *optval, int __user *optlen)
|
|
|
|
{
|
dccp ccid-3: No more CCID control blocks in LISTEN state
The CCIDs are activated as last of the features, at the end of the handshake,
were the LISTEN state of the master socket is inherited into the server
state of the child socket. Thus, the only states visible to CCIDs now are
OPEN/PARTOPEN, and the closing states.
This allows to remove tests which were previously necessary to protect
against referencing a socket in the listening state (in CCID3), but which
now have become redundant.
As a further byproduct of enabling the CCIDs only after the connection has been
fully established, several typecast-initialisations of ccid3_hc_{rx,tx}_sock
can now be eliminated:
* the CCID is loaded, so it is not necessary to test if it is NULL,
* if it is possible to load a CCID and leave the private area NULL, then this
is a bug, which should crash loudly - and earlier,
* the test for state==OPEN || state==PARTOPEN now reduces only to the closing
phase (e.g. when the node has received an unexpected Reset).
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Acked-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
2008-09-03 22:30:19 -07:00
|
|
|
const struct ccid3_hc_rx_sock *hcrx = ccid3_hc_rx_sk(sk);
|
2007-12-17 05:07:44 -07:00
|
|
|
struct tfrc_rx_info rx_info;
|
2005-09-18 00:19:32 -07:00
|
|
|
const void *val;
|
2007-02-09 07:24:38 -07:00
|
|
|
|
2005-09-18 00:19:32 -07:00
|
|
|
switch (optname) {
|
|
|
|
case DCCP_SOCKOPT_CCID_RX_INFO:
|
2007-12-17 05:07:44 -07:00
|
|
|
if (len < sizeof(rx_info))
|
2005-09-18 00:19:32 -07:00
|
|
|
return -EINVAL;
|
2008-09-03 22:30:19 -07:00
|
|
|
rx_info.tfrcrx_x_recv = hcrx->x_recv;
|
|
|
|
rx_info.tfrcrx_rtt = hcrx->rtt;
|
|
|
|
rx_info.tfrcrx_p = hcrx->p_inverse == 0 ? ~0U :
|
|
|
|
scaled_div(1, hcrx->p_inverse);
|
2007-12-17 05:07:44 -07:00
|
|
|
len = sizeof(rx_info);
|
|
|
|
val = &rx_info;
|
2005-09-18 00:19:32 -07:00
|
|
|
break;
|
|
|
|
default:
|
|
|
|
return -ENOPROTOOPT;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (put_user(len, optlen) || copy_to_user(optval, val, len))
|
|
|
|
return -EFAULT;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2006-03-20 20:21:44 -07:00
|
|
|
static struct ccid_operations ccid3 = {
|
2006-09-21 19:26:44 -07:00
|
|
|
.ccid_id = DCCPC_CCID3,
|
2007-12-13 18:33:25 -07:00
|
|
|
.ccid_name = "TCP-Friendly Rate Control",
|
2005-08-09 20:14:34 -07:00
|
|
|
.ccid_owner = THIS_MODULE,
|
2006-03-20 20:21:44 -07:00
|
|
|
.ccid_hc_tx_obj_size = sizeof(struct ccid3_hc_tx_sock),
|
2005-08-09 20:14:34 -07:00
|
|
|
.ccid_hc_tx_init = ccid3_hc_tx_init,
|
|
|
|
.ccid_hc_tx_exit = ccid3_hc_tx_exit,
|
|
|
|
.ccid_hc_tx_send_packet = ccid3_hc_tx_send_packet,
|
|
|
|
.ccid_hc_tx_packet_sent = ccid3_hc_tx_packet_sent,
|
|
|
|
.ccid_hc_tx_packet_recv = ccid3_hc_tx_packet_recv,
|
|
|
|
.ccid_hc_tx_parse_options = ccid3_hc_tx_parse_options,
|
2006-03-20 20:21:44 -07:00
|
|
|
.ccid_hc_rx_obj_size = sizeof(struct ccid3_hc_rx_sock),
|
2005-08-09 20:14:34 -07:00
|
|
|
.ccid_hc_rx_init = ccid3_hc_rx_init,
|
|
|
|
.ccid_hc_rx_exit = ccid3_hc_rx_exit,
|
|
|
|
.ccid_hc_rx_insert_options = ccid3_hc_rx_insert_options,
|
|
|
|
.ccid_hc_rx_packet_recv = ccid3_hc_rx_packet_recv,
|
[DCCP]: Introduce dccp_get_info
And also hc_tx and hc_rx get_info functions for the CCIDs to fill in
information that is specific to them.
For now reusing struct tcp_info, later I'll try to figure out a better
solution, for now its really nice to get this kind of info:
[root@qemu ~]# ./ss -danemi
State Recv-Q Send-Q Local Addr:Port Peer Addr:Port
LISTEN 0 0 *:5001 *:* ino:628 sk:c1340040
mem:(r0,w0,f0,t0) cwnd:0 ssthresh:0
ESTAB 0 0 172.20.0.2:5001 172.20.0.1:32785 ino:629 sk:c13409a0
mem:(r0,w0,f0,t0) ts rto:1000 rtt:0.004/0 cwnd:0 ssthresh:0 rcv_rtt:61.377
This, for instance, shows that we're not congestion controlling ACKs,
as the above output is in the ttcp receiving host, and ttcp is a one
way app, i.e. the received never calls sendmsg, so
ccid_hc_tx_send_packet is never called, so the TX half connection
stays in TFRC_SSTATE_NO_SENT state and hctx_rtt is never calculated,
stays with the value set in ccid3_hc_tx_init, 4us, as show above in
milliseconds (0.004ms), upcoming patches will fix this.
rcv_rtt seems sane tho, matching ping results :-)
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2005-08-23 21:52:35 -07:00
|
|
|
.ccid_hc_rx_get_info = ccid3_hc_rx_get_info,
|
|
|
|
.ccid_hc_tx_get_info = ccid3_hc_tx_get_info,
|
2005-09-18 00:19:32 -07:00
|
|
|
.ccid_hc_rx_getsockopt = ccid3_hc_rx_getsockopt,
|
|
|
|
.ccid_hc_tx_getsockopt = ccid3_hc_tx_getsockopt,
|
2005-08-09 20:14:34 -07:00
|
|
|
};
|
2006-12-10 11:01:18 -07:00
|
|
|
|
2006-11-20 13:28:09 -07:00
|
|
|
#ifdef CONFIG_IP_DCCP_CCID3_DEBUG
|
2008-08-23 04:28:27 -07:00
|
|
|
module_param(ccid3_debug, bool, 0644);
|
2005-08-09 20:14:34 -07:00
|
|
|
MODULE_PARM_DESC(ccid3_debug, "Enable debug messages");
|
2006-11-20 13:28:09 -07:00
|
|
|
#endif
|
2005-08-09 20:14:34 -07:00
|
|
|
|
|
|
|
static __init int ccid3_module_init(void)
|
|
|
|
{
|
2007-12-06 07:28:13 -07:00
|
|
|
return ccid_register(&ccid3);
|
2005-08-09 20:14:34 -07:00
|
|
|
}
|
|
|
|
module_init(ccid3_module_init);
|
|
|
|
|
|
|
|
static __exit void ccid3_module_exit(void)
|
|
|
|
{
|
|
|
|
ccid_unregister(&ccid3);
|
|
|
|
}
|
|
|
|
module_exit(ccid3_module_exit);
|
|
|
|
|
2006-08-26 19:01:30 -07:00
|
|
|
MODULE_AUTHOR("Ian McDonald <ian.mcdonald@jandi.co.nz>, "
|
2005-08-26 23:51:58 -07:00
|
|
|
"Arnaldo Carvalho de Melo <acme@ghostprotocols.net>");
|
2005-08-09 20:14:34 -07:00
|
|
|
MODULE_DESCRIPTION("DCCP TFRC CCID3 CCID");
|
|
|
|
MODULE_LICENSE("GPL");
|
|
|
|
MODULE_ALIAS("net-dccp-ccid-3");
|