batman-adv: tp_meter: consolidate congestion control variables

Congestion control variables in batadv_tp_sender were scattered across the
struct without clear grouping, making it difficult to reason about which
fields require cwnd_lock (now "cc_lock") protection. These should be
combined in a structure to make it more easily visible which variable
should be read/modified with the cc_lock held.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
This commit is contained in:
Sven Eckelmann 2026-05-12 19:37:05 +02:00
parent a7eff94024
commit 4f117f5626
No known key found for this signature in database
GPG Key ID: 4D0F772BD314F5CB
2 changed files with 128 additions and 122 deletions

View File

@ -143,12 +143,12 @@ static u32 batadv_tp_cwnd(u32 base, u32 increment, u32 min)
* increased by MSS * MSS / CWND for every unique received ACK
*/
static void batadv_tp_update_cwnd(struct batadv_tp_sender *tp_vars, u32 mss)
__must_hold(&tp_vars->cwnd_lock)
__must_hold(&tp_vars->cc_lock)
{
/* slow start... */
if (tp_vars->cwnd <= tp_vars->ss_threshold) {
tp_vars->dec_cwnd = 0;
tp_vars->cwnd = batadv_tp_cwnd(tp_vars->cwnd, mss, mss);
if (tp_vars->cc.cwnd <= tp_vars->cc.ss_threshold) {
tp_vars->cc.dec_cwnd = 0;
tp_vars->cc.cwnd = batadv_tp_cwnd(tp_vars->cc.cwnd, mss, mss);
return;
}
@ -156,13 +156,13 @@ static void batadv_tp_update_cwnd(struct batadv_tp_sender *tp_vars, u32 mss)
mss = min_t(u32, mss, (1U << 14) - 1);
/* increment CWND at least of 1 (section 3.1 of RFC5681) */
tp_vars->dec_cwnd += max_t(u32, 1U << 3,
((mss * mss) << 3) / tp_vars->cwnd);
if (tp_vars->dec_cwnd < (mss << 3))
tp_vars->cc.dec_cwnd += max_t(u32, 1U << 3,
((mss * mss) << 3) / tp_vars->cc.cwnd);
if (tp_vars->cc.dec_cwnd < (mss << 3))
return;
tp_vars->cwnd = batadv_tp_cwnd(tp_vars->cwnd, mss, mss);
tp_vars->dec_cwnd = 0;
tp_vars->cc.cwnd = batadv_tp_cwnd(tp_vars->cc.cwnd, mss, mss);
tp_vars->cc.dec_cwnd = 0;
}
/**
@ -172,7 +172,7 @@ static void batadv_tp_update_cwnd(struct batadv_tp_sender *tp_vars, u32 mss)
*/
static void batadv_tp_update_rto(struct batadv_tp_sender *tp_vars,
u32 new_rtt)
__must_hold(&tp_vars->cwnd_lock)
__must_hold(&tp_vars->cc_lock)
{
long m = new_rtt;
@ -182,24 +182,24 @@ static void batadv_tp_update_rto(struct batadv_tp_sender *tp_vars,
* It's tricky to understand. Don't lose hair please.
* Inspired by tcp_rtt_estimator() tcp_input.c
*/
if (tp_vars->srtt != 0) {
m -= (tp_vars->srtt >> 3); /* m is now error in rtt est */
tp_vars->srtt += m; /* rtt = 7/8 srtt + 1/8 new */
if (tp_vars->cc.srtt != 0) {
m -= (tp_vars->cc.srtt >> 3); /* m is now error in rtt est */
tp_vars->cc.srtt += m; /* rtt = 7/8 srtt + 1/8 new */
if (m < 0)
m = -m;
m -= (tp_vars->rttvar >> 2);
tp_vars->rttvar += m; /* mdev ~= 3/4 rttvar + 1/4 new */
m -= (tp_vars->cc.rttvar >> 2);
tp_vars->cc.rttvar += m; /* mdev ~= 3/4 rttvar + 1/4 new */
} else {
/* first measure getting in */
tp_vars->srtt = m << 3; /* take the measured time to be srtt */
tp_vars->rttvar = m << 1; /* new_rtt / 2 */
tp_vars->cc.srtt = m << 3; /* take the measured time to be srtt */
tp_vars->cc.rttvar = m << 1; /* new_rtt / 2 */
}
/* rto = srtt + 4 * rttvar.
* rttvar is scaled by 4, therefore doesn't need to be multiplied
*/
WRITE_ONCE(tp_vars->rto, (tp_vars->srtt >> 3) + tp_vars->rttvar);
WRITE_ONCE(tp_vars->cc.rto, (tp_vars->cc.srtt >> 3) + tp_vars->cc.rttvar);
}
/**
@ -450,11 +450,11 @@ static void batadv_tp_sender_end(struct batadv_priv *bat_priv,
batadv_dbg(BATADV_DBG_TP_METER, bat_priv,
"Last timing stats: SRTT=%ums RTTVAR=%ums RTO=%ums\n",
tp_vars->srtt >> 3, tp_vars->rttvar >> 2, tp_vars->rto);
tp_vars->cc.srtt >> 3, tp_vars->cc.rttvar >> 2, tp_vars->cc.rto);
batadv_dbg(BATADV_DBG_TP_METER, bat_priv,
"Final values: cwnd=%u ss_threshold=%u\n",
tp_vars->cwnd, tp_vars->ss_threshold);
tp_vars->cc.cwnd, tp_vars->cc.ss_threshold);
session_cookie = batadv_tp_session_cookie(tp_vars->common.session,
tp_vars->icmp_uid);
@ -521,7 +521,7 @@ static void batadv_tp_reset_sender_timer(struct batadv_tp_sender *tp_vars)
return;
mod_timer(&tp_vars->common.timer,
jiffies + msecs_to_jiffies(READ_ONCE(tp_vars->rto)));
jiffies + msecs_to_jiffies(READ_ONCE(tp_vars->cc.rto)));
}
/**
@ -540,11 +540,11 @@ static void batadv_tp_sender_timeout(struct timer_list *t)
if (batadv_tp_sender_stopped(tp_vars))
return;
spin_lock_bh(&tp_vars->cwnd_lock);
spin_lock_bh(&tp_vars->cc_lock);
/* if the user waited long enough...shutdown the test */
if (unlikely(tp_vars->rto >= BATADV_TP_MAX_RTO)) {
spin_unlock_bh(&tp_vars->cwnd_lock);
if (unlikely(tp_vars->cc.rto >= BATADV_TP_MAX_RTO)) {
spin_unlock_bh(&tp_vars->cc_lock);
batadv_tp_sender_shutdown(tp_vars,
BATADV_TP_REASON_DST_UNREACHABLE);
return;
@ -553,22 +553,22 @@ static void batadv_tp_sender_timeout(struct timer_list *t)
/* RTO exponential backoff
* Details in Section 5.5 of RFC6298
*/
WRITE_ONCE(tp_vars->rto, tp_vars->rto * 2);
WRITE_ONCE(tp_vars->cc.rto, tp_vars->cc.rto * 2);
tp_vars->ss_threshold = tp_vars->cwnd >> 1;
if (tp_vars->ss_threshold < BATADV_TP_PLEN * 2)
tp_vars->ss_threshold = BATADV_TP_PLEN * 2;
tp_vars->cc.ss_threshold = tp_vars->cc.cwnd >> 1;
if (tp_vars->cc.ss_threshold < BATADV_TP_PLEN * 2)
tp_vars->cc.ss_threshold = BATADV_TP_PLEN * 2;
batadv_dbg(BATADV_DBG_TP_METER, bat_priv,
"Meter: RTO fired during test towards %pM! cwnd=%u new ss_thr=%u, resetting last_sent to %u\n",
tp_vars->common.other_end, tp_vars->cwnd, tp_vars->ss_threshold,
tp_vars->last_acked);
tp_vars->common.other_end, tp_vars->cc.cwnd, tp_vars->cc.ss_threshold,
tp_vars->cc.last_acked);
tp_vars->cwnd = BATADV_TP_PLEN * 3;
tp_vars->cc.cwnd = BATADV_TP_PLEN * 3;
WRITE_ONCE(tp_vars->last_sent, tp_vars->last_acked);
WRITE_ONCE(tp_vars->cc.last_sent, tp_vars->cc.last_acked);
spin_unlock_bh(&tp_vars->cwnd_lock);
spin_unlock_bh(&tp_vars->cc_lock);
/* resend the non-ACKed packets.. */
wake_up(&tp_vars->more_bytes);
@ -698,81 +698,81 @@ static enum batadv_tp_ack_reaction
batadv_tp_handle_ack(struct batadv_priv *bat_priv,
struct batadv_tp_sender *tp_vars,
u32 recv_ack, size_t mss)
__must_hold(&tp_vars->cwnd_lock)
__must_hold(&tp_vars->cc_lock)
{
enum batadv_tp_ack_reaction reaction;
if (batadv_seq_before(recv_ack, tp_vars->last_acked))
if (batadv_seq_before(recv_ack, tp_vars->cc.last_acked))
return BATADV_TP_ACK_REACTION_OLD_ACK;
/* check if this ACK is a duplicate */
if (tp_vars->last_acked == recv_ack) {
if (tp_vars->cc.last_acked == recv_ack) {
/* if this is the third duplicate ACK do Fast Retransmit */
if (tp_vars->dup_acks > 3)
if (tp_vars->cc.dup_acks > 3)
return BATADV_TP_ACK_REACTION_IGNORE;
tp_vars->dup_acks++;
if (tp_vars->dup_acks != 3)
tp_vars->cc.dup_acks++;
if (tp_vars->cc.dup_acks != 3)
return BATADV_TP_ACK_REACTION_IGNORE;
if (!batadv_seq_before(tp_vars->recover, recv_ack))
if (!batadv_seq_before(tp_vars->cc.recover, recv_ack))
return BATADV_TP_ACK_REACTION_IGNORE;
/* Fast Recovery */
tp_vars->fast_recovery = true;
tp_vars->cc.fast_recovery = true;
/* Set recover to the last outstanding seqno when Fast Recovery
* is entered. RFC6582, Section 3.2, step 1
*/
tp_vars->recover = tp_vars->last_sent;
tp_vars->ss_threshold = tp_vars->cwnd >> 1;
tp_vars->cc.recover = tp_vars->cc.last_sent;
tp_vars->cc.ss_threshold = tp_vars->cc.cwnd >> 1;
batadv_dbg(BATADV_DBG_TP_METER, bat_priv,
"Meter: Fast Recovery, (cur cwnd=%u) ss_thr=%u last_sent=%u recv_ack=%u\n",
tp_vars->cwnd, tp_vars->ss_threshold,
tp_vars->last_sent, recv_ack);
tp_vars->cwnd = batadv_tp_cwnd(tp_vars->ss_threshold, 3 * mss,
mss);
tp_vars->dec_cwnd = 0;
WRITE_ONCE(tp_vars->last_sent, recv_ack);
tp_vars->cc.cwnd, tp_vars->cc.ss_threshold,
tp_vars->cc.last_sent, recv_ack);
tp_vars->cc.cwnd = batadv_tp_cwnd(tp_vars->cc.ss_threshold, 3 * mss,
mss);
tp_vars->cc.dec_cwnd = 0;
WRITE_ONCE(tp_vars->cc.last_sent, recv_ack);
return BATADV_TP_ACK_REACTION_RESEND_WAKEUP;
}
/* count the acked data */
atomic64_add(recv_ack - tp_vars->last_acked, &tp_vars->tot_sent);
atomic64_add(recv_ack - tp_vars->cc.last_acked, &tp_vars->tot_sent);
/* reset the duplicate ACKs counter */
tp_vars->dup_acks = 0;
tp_vars->cc.dup_acks = 0;
if (tp_vars->fast_recovery) {
if (tp_vars->cc.fast_recovery) {
/* partial ACK */
if (batadv_seq_before(recv_ack, tp_vars->recover)) {
if (batadv_seq_before(recv_ack, tp_vars->cc.recover)) {
/* this is another hole in the window. React
* immediately as specified by NewReno (see
* Section 3.2 of RFC6582 for details)
*/
reaction = BATADV_TP_ACK_REACTION_RESEND_WAKEUP;
tp_vars->cwnd = batadv_tp_cwnd(tp_vars->cwnd,
mss, mss);
tp_vars->cc.cwnd = batadv_tp_cwnd(tp_vars->cc.cwnd,
mss, mss);
} else {
tp_vars->fast_recovery = false;
tp_vars->cc.fast_recovery = false;
/* set cwnd to the value of ss_threshold at the
* moment that Fast Recovery was entered.
* RFC6582, Section 3.2, step 3
*/
tp_vars->cwnd = batadv_tp_cwnd(tp_vars->ss_threshold,
0, mss);
tp_vars->cc.cwnd = batadv_tp_cwnd(tp_vars->cc.ss_threshold,
0, mss);
reaction = BATADV_TP_ACK_REACTION_WAKEUP;
}
} else {
if (recv_ack - tp_vars->last_acked >= mss)
if (recv_ack - tp_vars->cc.last_acked >= mss)
batadv_tp_update_cwnd(tp_vars, mss);
reaction = BATADV_TP_ACK_REACTION_WAKEUP;
}
/* move the Transmit Window */
WRITE_ONCE(tp_vars->last_acked, recv_ack);
WRITE_ONCE(tp_vars->cc.last_acked, recv_ack);
return reaction;
}
@ -814,7 +814,7 @@ static void batadv_tp_recv_ack(struct batadv_priv *bat_priv,
goto out;
/* old ACK? silently drop it.. */
if (batadv_seq_before(recv_ack, READ_ONCE(tp_vars->last_acked)))
if (batadv_seq_before(recv_ack, READ_ONCE(tp_vars->cc.last_acked)))
goto out;
primary_if = batadv_primary_if_get_selected(bat_priv);
@ -825,14 +825,14 @@ static void batadv_tp_recv_ack(struct batadv_priv *bat_priv,
if (unlikely(!orig_node))
goto out;
spin_lock_bh(&tp_vars->cwnd_lock);
spin_lock_bh(&tp_vars->cc_lock);
/* update RTO with the new sampled RTT, if any */
rtt = jiffies_to_msecs(jiffies) - ntohl(icmp->timestamp);
if (icmp->timestamp && rtt)
batadv_tp_update_rto(tp_vars, rtt);
reaction = batadv_tp_handle_ack(bat_priv, tp_vars, recv_ack, mss);
spin_unlock_bh(&tp_vars->cwnd_lock);
spin_unlock_bh(&tp_vars->cc_lock);
if (reaction == BATADV_TP_ACK_REACTION_OLD_ACK)
goto out;
@ -873,16 +873,16 @@ static bool batadv_tp_avail(struct batadv_tp_sender *tp_vars,
{
u32 win_left, win_limit;
spin_lock_bh(&tp_vars->cwnd_lock);
spin_lock_bh(&tp_vars->cc_lock);
win_limit = tp_vars->last_acked + tp_vars->cwnd;
win_limit = tp_vars->cc.last_acked + tp_vars->cc.cwnd;
if (batadv_seq_before(tp_vars->last_sent, win_limit))
win_left = win_limit - tp_vars->last_sent;
if (batadv_seq_before(tp_vars->cc.last_sent, win_limit))
win_left = win_limit - tp_vars->cc.last_sent;
else
win_left = 0;
spin_unlock_bh(&tp_vars->cwnd_lock);
spin_unlock_bh(&tp_vars->cc_lock);
return win_left >= payload_len;
}
@ -964,7 +964,7 @@ static int batadv_tp_send(void *arg)
* the size of the unicast header
*/
packet_len = payload_len + sizeof(struct batadv_unicast_packet);
last_sent = READ_ONCE(tp_vars->last_sent);
last_sent = READ_ONCE(tp_vars->cc.last_sent);
err = batadv_tp_send_msg(tp_vars, primary_if->net_dev->dev_addr,
orig_node, last_sent, packet_len,
@ -983,10 +983,10 @@ static int batadv_tp_send(void *arg)
/* right-shift the TWND */
if (!err) {
spin_lock_bh(&tp_vars->cwnd_lock);
if (tp_vars->last_sent == last_sent)
WRITE_ONCE(tp_vars->last_sent, last_sent + payload_len);
spin_unlock_bh(&tp_vars->cwnd_lock);
spin_lock_bh(&tp_vars->cc_lock);
if (tp_vars->cc.last_sent == last_sent)
WRITE_ONCE(tp_vars->cc.last_sent, last_sent + payload_len);
spin_unlock_bh(&tp_vars->cc_lock);
}
cond_resched();
@ -1104,30 +1104,30 @@ void batadv_tp_start(struct batadv_priv *bat_priv, const u8 *dst,
memcpy(tp_vars->common.session, session_id, sizeof(session_id));
tp_vars->icmp_uid = icmp_uid;
WRITE_ONCE(tp_vars->last_sent, BATADV_TP_FIRST_SEQ);
WRITE_ONCE(tp_vars->dup_acks, 0);
WRITE_ONCE(tp_vars->last_acked, BATADV_TP_FIRST_SEQ);
tp_vars->fast_recovery = false;
tp_vars->recover = BATADV_TP_FIRST_SEQ;
WRITE_ONCE(tp_vars->cc.last_sent, BATADV_TP_FIRST_SEQ);
WRITE_ONCE(tp_vars->cc.dup_acks, 0);
WRITE_ONCE(tp_vars->cc.last_acked, BATADV_TP_FIRST_SEQ);
tp_vars->cc.fast_recovery = false;
tp_vars->cc.recover = BATADV_TP_FIRST_SEQ;
/* initialise the CWND to 3*MSS (Section 3.1 in RFC5681).
* For batman-adv the MSS is the size of the payload received by the
* mesh_interface, hence its MTU
*/
tp_vars->cwnd = BATADV_TP_PLEN * 3;
tp_vars->dec_cwnd = 0;
tp_vars->cc.cwnd = BATADV_TP_PLEN * 3;
tp_vars->cc.dec_cwnd = 0;
/* at the beginning initialise the SS threshold to the biggest possible
* window size, hence the AWND size
*/
tp_vars->ss_threshold = BATADV_TP_AWND;
tp_vars->cc.ss_threshold = BATADV_TP_AWND;
/* RTO initial value is 3 seconds.
* Details in Section 2.1 of RFC6298
*/
WRITE_ONCE(tp_vars->rto, 1000);
tp_vars->srtt = 0;
tp_vars->rttvar = 0;
WRITE_ONCE(tp_vars->cc.rto, 1000);
tp_vars->cc.srtt = 0;
tp_vars->cc.rttvar = 0;
atomic64_set(&tp_vars->tot_sent, 0);
@ -1143,7 +1143,7 @@ void batadv_tp_start(struct batadv_priv *bat_priv, const u8 *dst,
spin_lock_init(&tp_vars->common.unacked_lock);
INIT_LIST_HEAD(&tp_vars->common.unacked_list);
spin_lock_init(&tp_vars->cwnd_lock);
spin_lock_init(&tp_vars->cc_lock);
tp_vars->prerandom_offset = 0;
spin_lock_init(&tp_vars->prerandom_lock);

View File

@ -1367,6 +1367,45 @@ struct batadv_tp_vars_common {
struct rcu_head rcu;
};
/** struct batadv_tp_sender_cc - congestion control variables */
struct batadv_tp_sender_cc {
/** @fast_recovery: true if in Fast Recovery mode */
bool fast_recovery:1;
/** @dup_acks: duplicate ACKs counter */
u8 dup_acks;
/** @dec_cwnd: decimal part of the cwnd used during linear growth */
u16 dec_cwnd;
/** @cwnd: current size of the congestion window */
u32 cwnd;
/**
* @ss_threshold: Slow Start threshold. Once cwnd exceeds this value the
* connection switches to the Congestion Avoidance state
*/
u32 ss_threshold;
/** @last_acked: last acked byte */
u32 last_acked;
/** @last_sent: last sent byte, not yet acked */
u32 last_sent;
/** @recover: last sent seqno when entering Fast Recovery */
u32 recover;
/** @rto: sender timeout */
u32 rto;
/** @srtt: smoothed RTT scaled by 2^3 */
u32 srtt;
/** @rttvar: RTT variation scaled by 2^2 */
u32 rttvar;
};
/**
* struct batadv_tp_sender - sender tp meter private variables per session
*/
@ -1383,15 +1422,6 @@ struct batadv_tp_sender {
*/
atomic_t send_result;
/** @last_sent: last sent byte, not yet acked */
u32 last_sent;
/** @fast_recovery: true if in Fast Recovery mode */
bool fast_recovery:1;
/** @recover: last sent seqno when entering Fast Recovery */
u32 recover;
/** @finish_work: work item for the finishing procedure */
struct delayed_work finish_work;
@ -1404,39 +1434,15 @@ struct batadv_tp_sender {
/** @icmp_uid: local ICMP "socket" index */
u8 icmp_uid;
/** @dec_cwnd: decimal part of the cwnd used during linear growth */
u16 dec_cwnd;
/** @cc: congestion control variables */
struct batadv_tp_sender_cc cc;
/** @cwnd: current size of the congestion window */
u32 cwnd;
/** @cwnd_lock: lock do protect congestion control variables */
spinlock_t cwnd_lock;
/**
* @ss_threshold: Slow Start threshold. Once cwnd exceeds this value the
* connection switches to the Congestion Avoidance state
*/
u32 ss_threshold;
/** @last_acked: last acked byte */
u32 last_acked;
/** @cc_lock: lock do protect @cc */
spinlock_t cc_lock;
/** @tot_sent: amount of data sent/ACKed so far */
atomic64_t tot_sent;
/** @dup_acks: duplicate ACKs counter */
u8 dup_acks;
/** @rto: sender timeout */
u32 rto;
/** @srtt: smoothed RTT scaled by 2^3 */
u32 srtt;
/** @rttvar: RTT variation scaled by 2^2 */
u32 rttvar;
/**
* @more_bytes: waiting queue anchor when waiting for more ack/retry
* timeout