rxrpc: Fix the excessive initial retransmission timeout

rxrpc currently uses a fixed 4s retransmission timeout until the RTT is
sufficiently sampled.  This can cause problems with some fileservers with
calls to the cache manager in the afs filesystem being dropped from the
fileserver because a packet goes missing and the retransmission timeout is
greater than the call expiry timeout.

Fix this by:

 (1) Copying the RTT/RTO calculation code from Linux's TCP implementation
     and altering it to fit rxrpc.

 (2) Altering the various users of the RTT to make use of the new SRTT
     value.

 (3) Replacing the use of rxrpc_resend_timeout to use the calculated RTO
     value instead (which is needed in jiffies), along with a backoff.

Notes:

 (1) rxrpc provides RTT samples by matching the serial numbers on outgoing
     DATA packets that have the RXRPC_REQUEST_ACK set and PING ACK packets
     against the reference serial number in incoming REQUESTED ACK and
     PING-RESPONSE ACK packets.

 (2) Each packet that is transmitted on an rxrpc connection gets a new
     per-connection serial number, even for retransmissions, so an ACK can
     be cross-referenced to a specific trigger packet.  This allows RTT
     information to be drawn from retransmitted DATA packets also.

 (3) rxrpc maintains the RTT/RTO state on the rxrpc_peer record rather than
     on an rxrpc_call because many RPC calls won't live long enough to
     generate more than one sample.

 (4) The calculated SRTT value is in units of 8ths of a microsecond rather
     than nanoseconds.

The (S)RTT and RTO values are displayed in /proc/net/rxrpc/peers.

Fixes: 17926a7932 ([AF_RXRPC]: Provide secure RxRPC sockets for use by userspace and kernel both"")
Signed-off-by: David Howells <dhowells@redhat.com>
This commit is contained in:
David Howells 2020-05-11 14:54:34 +01:00
parent 42c556fef9
commit c410bf0193
17 changed files with 266 additions and 155 deletions

View File

@ -32,9 +32,8 @@ void afs_fileserver_probe_result(struct afs_call *call)
struct afs_server *server = call->server; struct afs_server *server = call->server;
unsigned int server_index = call->server_index; unsigned int server_index = call->server_index;
unsigned int index = call->addr_ix; unsigned int index = call->addr_ix;
unsigned int rtt = UINT_MAX; unsigned int rtt_us;
bool have_result = false; bool have_result = false;
u64 _rtt;
int ret = call->error; int ret = call->error;
_enter("%pU,%u", &server->uuid, index); _enter("%pU,%u", &server->uuid, index);
@ -93,15 +92,9 @@ responded:
} }
} }
/* Get the RTT and scale it to fit into a 32-bit value that represents rtt_us = rxrpc_kernel_get_srtt(call->net->socket, call->rxcall);
* over a minute of time so that we can access it with one instruction if (rtt_us < server->probe.rtt) {
* on a 32-bit system. server->probe.rtt = rtt_us;
*/
_rtt = rxrpc_kernel_get_rtt(call->net->socket, call->rxcall);
_rtt /= 64;
rtt = (_rtt > UINT_MAX) ? UINT_MAX : _rtt;
if (rtt < server->probe.rtt) {
server->probe.rtt = rtt;
alist->preferred = index; alist->preferred = index;
have_result = true; have_result = true;
} }
@ -113,8 +106,7 @@ out:
spin_unlock(&server->probe_lock); spin_unlock(&server->probe_lock);
_debug("probe [%u][%u] %pISpc rtt=%u ret=%d", _debug("probe [%u][%u] %pISpc rtt=%u ret=%d",
server_index, index, &alist->addrs[index].transport, server_index, index, &alist->addrs[index].transport, rtt_us, ret);
(unsigned int)rtt, ret);
have_result |= afs_fs_probe_done(server); have_result |= afs_fs_probe_done(server);
if (have_result) if (have_result)

View File

@ -31,10 +31,9 @@ void afs_vlserver_probe_result(struct afs_call *call)
struct afs_addr_list *alist = call->alist; struct afs_addr_list *alist = call->alist;
struct afs_vlserver *server = call->vlserver; struct afs_vlserver *server = call->vlserver;
unsigned int server_index = call->server_index; unsigned int server_index = call->server_index;
unsigned int rtt_us = 0;
unsigned int index = call->addr_ix; unsigned int index = call->addr_ix;
unsigned int rtt = UINT_MAX;
bool have_result = false; bool have_result = false;
u64 _rtt;
int ret = call->error; int ret = call->error;
_enter("%s,%u,%u,%d,%d", server->name, server_index, index, ret, call->abort_code); _enter("%s,%u,%u,%d,%d", server->name, server_index, index, ret, call->abort_code);
@ -93,15 +92,9 @@ responded:
} }
} }
/* Get the RTT and scale it to fit into a 32-bit value that represents rtt_us = rxrpc_kernel_get_srtt(call->net->socket, call->rxcall);
* over a minute of time so that we can access it with one instruction if (rtt_us < server->probe.rtt) {
* on a 32-bit system. server->probe.rtt = rtt_us;
*/
_rtt = rxrpc_kernel_get_rtt(call->net->socket, call->rxcall);
_rtt /= 64;
rtt = (_rtt > UINT_MAX) ? UINT_MAX : _rtt;
if (rtt < server->probe.rtt) {
server->probe.rtt = rtt;
alist->preferred = index; alist->preferred = index;
have_result = true; have_result = true;
} }
@ -113,8 +106,7 @@ out:
spin_unlock(&server->probe_lock); spin_unlock(&server->probe_lock);
_debug("probe [%u][%u] %pISpc rtt=%u ret=%d", _debug("probe [%u][%u] %pISpc rtt=%u ret=%d",
server_index, index, &alist->addrs[index].transport, server_index, index, &alist->addrs[index].transport, rtt_us, ret);
(unsigned int)rtt, ret);
have_result |= afs_vl_probe_done(server); have_result |= afs_vl_probe_done(server);
if (have_result) { if (have_result) {

View File

@ -59,7 +59,7 @@ bool rxrpc_kernel_abort_call(struct socket *, struct rxrpc_call *,
void rxrpc_kernel_end_call(struct socket *, struct rxrpc_call *); void rxrpc_kernel_end_call(struct socket *, struct rxrpc_call *);
void rxrpc_kernel_get_peer(struct socket *, struct rxrpc_call *, void rxrpc_kernel_get_peer(struct socket *, struct rxrpc_call *,
struct sockaddr_rxrpc *); struct sockaddr_rxrpc *);
u64 rxrpc_kernel_get_rtt(struct socket *, struct rxrpc_call *); u32 rxrpc_kernel_get_srtt(struct socket *, struct rxrpc_call *);
int rxrpc_kernel_charge_accept(struct socket *, rxrpc_notify_rx_t, int rxrpc_kernel_charge_accept(struct socket *, rxrpc_notify_rx_t,
rxrpc_user_attach_call_t, unsigned long, gfp_t, rxrpc_user_attach_call_t, unsigned long, gfp_t,
unsigned int); unsigned int);

View File

@ -1112,18 +1112,17 @@ TRACE_EVENT(rxrpc_rtt_tx,
TRACE_EVENT(rxrpc_rtt_rx, TRACE_EVENT(rxrpc_rtt_rx,
TP_PROTO(struct rxrpc_call *call, enum rxrpc_rtt_rx_trace why, TP_PROTO(struct rxrpc_call *call, enum rxrpc_rtt_rx_trace why,
rxrpc_serial_t send_serial, rxrpc_serial_t resp_serial, rxrpc_serial_t send_serial, rxrpc_serial_t resp_serial,
s64 rtt, u8 nr, s64 avg), u32 rtt, u32 rto),
TP_ARGS(call, why, send_serial, resp_serial, rtt, nr, avg), TP_ARGS(call, why, send_serial, resp_serial, rtt, rto),
TP_STRUCT__entry( TP_STRUCT__entry(
__field(unsigned int, call ) __field(unsigned int, call )
__field(enum rxrpc_rtt_rx_trace, why ) __field(enum rxrpc_rtt_rx_trace, why )
__field(u8, nr )
__field(rxrpc_serial_t, send_serial ) __field(rxrpc_serial_t, send_serial )
__field(rxrpc_serial_t, resp_serial ) __field(rxrpc_serial_t, resp_serial )
__field(s64, rtt ) __field(u32, rtt )
__field(u64, avg ) __field(u32, rto )
), ),
TP_fast_assign( TP_fast_assign(
@ -1132,18 +1131,16 @@ TRACE_EVENT(rxrpc_rtt_rx,
__entry->send_serial = send_serial; __entry->send_serial = send_serial;
__entry->resp_serial = resp_serial; __entry->resp_serial = resp_serial;
__entry->rtt = rtt; __entry->rtt = rtt;
__entry->nr = nr; __entry->rto = rto;
__entry->avg = avg;
), ),
TP_printk("c=%08x %s sr=%08x rr=%08x rtt=%lld nr=%u avg=%lld", TP_printk("c=%08x %s sr=%08x rr=%08x rtt=%u rto=%u",
__entry->call, __entry->call,
__print_symbolic(__entry->why, rxrpc_rtt_rx_traces), __print_symbolic(__entry->why, rxrpc_rtt_rx_traces),
__entry->send_serial, __entry->send_serial,
__entry->resp_serial, __entry->resp_serial,
__entry->rtt, __entry->rtt,
__entry->nr, __entry->rto)
__entry->avg)
); );
TRACE_EVENT(rxrpc_timer, TRACE_EVENT(rxrpc_timer,

View File

@ -25,6 +25,7 @@ rxrpc-y := \
peer_event.o \ peer_event.o \
peer_object.o \ peer_object.o \
recvmsg.o \ recvmsg.o \
rtt.o \
security.o \ security.o \
sendmsg.o \ sendmsg.o \
skbuff.o \ skbuff.o \

View File

@ -7,6 +7,7 @@
#include <linux/atomic.h> #include <linux/atomic.h>
#include <linux/seqlock.h> #include <linux/seqlock.h>
#include <linux/win_minmax.h>
#include <net/net_namespace.h> #include <net/net_namespace.h>
#include <net/netns/generic.h> #include <net/netns/generic.h>
#include <net/sock.h> #include <net/sock.h>
@ -311,11 +312,14 @@ struct rxrpc_peer {
#define RXRPC_RTT_CACHE_SIZE 32 #define RXRPC_RTT_CACHE_SIZE 32
spinlock_t rtt_input_lock; /* RTT lock for input routine */ spinlock_t rtt_input_lock; /* RTT lock for input routine */
ktime_t rtt_last_req; /* Time of last RTT request */ ktime_t rtt_last_req; /* Time of last RTT request */
u64 rtt; /* Current RTT estimate (in nS) */ unsigned int rtt_count; /* Number of samples we've got */
u64 rtt_sum; /* Sum of cache contents */
u64 rtt_cache[RXRPC_RTT_CACHE_SIZE]; /* Determined RTT cache */ u32 srtt_us; /* smoothed round trip time << 3 in usecs */
u8 rtt_cursor; /* next entry at which to insert */ u32 mdev_us; /* medium deviation */
u8 rtt_usage; /* amount of cache actually used */ u32 mdev_max_us; /* maximal mdev for the last rtt period */
u32 rttvar_us; /* smoothed mdev_max */
u32 rto_j; /* Retransmission timeout in jiffies */
u8 backoff; /* Backoff timeout */
u8 cong_cwnd; /* Congestion window size */ u8 cong_cwnd; /* Congestion window size */
}; };
@ -1041,7 +1045,6 @@ extern unsigned long rxrpc_idle_ack_delay;
extern unsigned int rxrpc_rx_window_size; extern unsigned int rxrpc_rx_window_size;
extern unsigned int rxrpc_rx_mtu; extern unsigned int rxrpc_rx_mtu;
extern unsigned int rxrpc_rx_jumbo_max; extern unsigned int rxrpc_rx_jumbo_max;
extern unsigned long rxrpc_resend_timeout;
extern const s8 rxrpc_ack_priority[]; extern const s8 rxrpc_ack_priority[];
@ -1069,8 +1072,6 @@ void rxrpc_send_keepalive(struct rxrpc_peer *);
* peer_event.c * peer_event.c
*/ */
void rxrpc_error_report(struct sock *); void rxrpc_error_report(struct sock *);
void rxrpc_peer_add_rtt(struct rxrpc_call *, enum rxrpc_rtt_rx_trace,
rxrpc_serial_t, rxrpc_serial_t, ktime_t, ktime_t);
void rxrpc_peer_keepalive_worker(struct work_struct *); void rxrpc_peer_keepalive_worker(struct work_struct *);
/* /*
@ -1102,6 +1103,14 @@ extern const struct seq_operations rxrpc_peer_seq_ops;
void rxrpc_notify_socket(struct rxrpc_call *); void rxrpc_notify_socket(struct rxrpc_call *);
int rxrpc_recvmsg(struct socket *, struct msghdr *, size_t, int); int rxrpc_recvmsg(struct socket *, struct msghdr *, size_t, int);
/*
* rtt.c
*/
void rxrpc_peer_add_rtt(struct rxrpc_call *, enum rxrpc_rtt_rx_trace,
rxrpc_serial_t, rxrpc_serial_t, ktime_t, ktime_t);
unsigned long rxrpc_get_rto_backoff(struct rxrpc_peer *, bool);
void rxrpc_peer_init_rtt(struct rxrpc_peer *);
/* /*
* rxkad.c * rxkad.c
*/ */

View File

@ -248,7 +248,7 @@ static void rxrpc_send_ping(struct rxrpc_call *call, struct sk_buff *skb)
struct rxrpc_skb_priv *sp = rxrpc_skb(skb); struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
ktime_t now = skb->tstamp; ktime_t now = skb->tstamp;
if (call->peer->rtt_usage < 3 || if (call->peer->rtt_count < 3 ||
ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000), now)) ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000), now))
rxrpc_propose_ACK(call, RXRPC_ACK_PING, sp->hdr.serial, rxrpc_propose_ACK(call, RXRPC_ACK_PING, sp->hdr.serial,
true, true, true, true,

View File

@ -111,8 +111,8 @@ static void __rxrpc_propose_ACK(struct rxrpc_call *call, u8 ack_reason,
} else { } else {
unsigned long now = jiffies, ack_at; unsigned long now = jiffies, ack_at;
if (call->peer->rtt_usage > 0) if (call->peer->srtt_us != 0)
ack_at = nsecs_to_jiffies(call->peer->rtt); ack_at = usecs_to_jiffies(call->peer->srtt_us >> 3);
else else
ack_at = expiry; ack_at = expiry;
@ -157,24 +157,18 @@ static void rxrpc_congestion_timeout(struct rxrpc_call *call)
static void rxrpc_resend(struct rxrpc_call *call, unsigned long now_j) static void rxrpc_resend(struct rxrpc_call *call, unsigned long now_j)
{ {
struct sk_buff *skb; struct sk_buff *skb;
unsigned long resend_at; unsigned long resend_at, rto_j;
rxrpc_seq_t cursor, seq, top; rxrpc_seq_t cursor, seq, top;
ktime_t now, max_age, oldest, ack_ts, timeout, min_timeo; ktime_t now, max_age, oldest, ack_ts;
int ix; int ix;
u8 annotation, anno_type, retrans = 0, unacked = 0; u8 annotation, anno_type, retrans = 0, unacked = 0;
_enter("{%d,%d}", call->tx_hard_ack, call->tx_top); _enter("{%d,%d}", call->tx_hard_ack, call->tx_top);
if (call->peer->rtt_usage > 1) rto_j = call->peer->rto_j;
timeout = ns_to_ktime(call->peer->rtt * 3 / 2);
else
timeout = ms_to_ktime(rxrpc_resend_timeout);
min_timeo = ns_to_ktime((1000000000 / HZ) * 4);
if (ktime_before(timeout, min_timeo))
timeout = min_timeo;
now = ktime_get_real(); now = ktime_get_real();
max_age = ktime_sub(now, timeout); max_age = ktime_sub(now, jiffies_to_usecs(rto_j));
spin_lock_bh(&call->lock); spin_lock_bh(&call->lock);
@ -219,7 +213,7 @@ static void rxrpc_resend(struct rxrpc_call *call, unsigned long now_j)
} }
resend_at = nsecs_to_jiffies(ktime_to_ns(ktime_sub(now, oldest))); resend_at = nsecs_to_jiffies(ktime_to_ns(ktime_sub(now, oldest)));
resend_at += jiffies + rxrpc_resend_timeout; resend_at += jiffies + rto_j;
WRITE_ONCE(call->resend_at, resend_at); WRITE_ONCE(call->resend_at, resend_at);
if (unacked) if (unacked)
@ -234,7 +228,7 @@ static void rxrpc_resend(struct rxrpc_call *call, unsigned long now_j)
rxrpc_timer_set_for_resend); rxrpc_timer_set_for_resend);
spin_unlock_bh(&call->lock); spin_unlock_bh(&call->lock);
ack_ts = ktime_sub(now, call->acks_latest_ts); ack_ts = ktime_sub(now, call->acks_latest_ts);
if (ktime_to_ns(ack_ts) < call->peer->rtt) if (ktime_to_us(ack_ts) < (call->peer->srtt_us >> 3))
goto out; goto out;
rxrpc_propose_ACK(call, RXRPC_ACK_PING, 0, true, false, rxrpc_propose_ACK(call, RXRPC_ACK_PING, 0, true, false,
rxrpc_propose_ack_ping_for_lost_ack); rxrpc_propose_ack_ping_for_lost_ack);

View File

@ -91,11 +91,11 @@ static void rxrpc_congestion_management(struct rxrpc_call *call,
/* We analyse the number of packets that get ACK'd per RTT /* We analyse the number of packets that get ACK'd per RTT
* period and increase the window if we managed to fill it. * period and increase the window if we managed to fill it.
*/ */
if (call->peer->rtt_usage == 0) if (call->peer->rtt_count == 0)
goto out; goto out;
if (ktime_before(skb->tstamp, if (ktime_before(skb->tstamp,
ktime_add_ns(call->cong_tstamp, ktime_add_us(call->cong_tstamp,
call->peer->rtt))) call->peer->srtt_us >> 3)))
goto out_no_clear_ca; goto out_no_clear_ca;
change = rxrpc_cong_rtt_window_end; change = rxrpc_cong_rtt_window_end;
call->cong_tstamp = skb->tstamp; call->cong_tstamp = skb->tstamp;

View File

@ -63,11 +63,6 @@ unsigned int rxrpc_rx_mtu = 5692;
*/ */
unsigned int rxrpc_rx_jumbo_max = 4; unsigned int rxrpc_rx_jumbo_max = 4;
/*
* Time till packet resend (in milliseconds).
*/
unsigned long rxrpc_resend_timeout = 4 * HZ;
const s8 rxrpc_ack_priority[] = { const s8 rxrpc_ack_priority[] = {
[0] = 0, [0] = 0,
[RXRPC_ACK_DELAY] = 1, [RXRPC_ACK_DELAY] = 1,

View File

@ -369,7 +369,7 @@ int rxrpc_send_data_packet(struct rxrpc_call *call, struct sk_buff *skb,
(test_and_clear_bit(RXRPC_CALL_EV_ACK_LOST, &call->events) || (test_and_clear_bit(RXRPC_CALL_EV_ACK_LOST, &call->events) ||
retrans || retrans ||
call->cong_mode == RXRPC_CALL_SLOW_START || call->cong_mode == RXRPC_CALL_SLOW_START ||
(call->peer->rtt_usage < 3 && sp->hdr.seq & 1) || (call->peer->rtt_count < 3 && sp->hdr.seq & 1) ||
ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000), ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000),
ktime_get_real()))) ktime_get_real())))
whdr.flags |= RXRPC_REQUEST_ACK; whdr.flags |= RXRPC_REQUEST_ACK;
@ -423,13 +423,10 @@ done:
if (whdr.flags & RXRPC_REQUEST_ACK) { if (whdr.flags & RXRPC_REQUEST_ACK) {
call->peer->rtt_last_req = skb->tstamp; call->peer->rtt_last_req = skb->tstamp;
trace_rxrpc_rtt_tx(call, rxrpc_rtt_tx_data, serial); trace_rxrpc_rtt_tx(call, rxrpc_rtt_tx_data, serial);
if (call->peer->rtt_usage > 1) { if (call->peer->rtt_count > 1) {
unsigned long nowj = jiffies, ack_lost_at; unsigned long nowj = jiffies, ack_lost_at;
ack_lost_at = nsecs_to_jiffies(2 * call->peer->rtt); ack_lost_at = rxrpc_get_rto_backoff(call->peer, retrans);
if (ack_lost_at < 1)
ack_lost_at = 1;
ack_lost_at += nowj; ack_lost_at += nowj;
WRITE_ONCE(call->ack_lost_at, ack_lost_at); WRITE_ONCE(call->ack_lost_at, ack_lost_at);
rxrpc_reduce_call_timer(call, ack_lost_at, nowj, rxrpc_reduce_call_timer(call, ack_lost_at, nowj,

View File

@ -295,52 +295,6 @@ static void rxrpc_distribute_error(struct rxrpc_peer *peer, int error,
} }
} }
/*
* Add RTT information to cache. This is called in softirq mode and has
* exclusive access to the peer RTT data.
*/
void rxrpc_peer_add_rtt(struct rxrpc_call *call, enum rxrpc_rtt_rx_trace why,
rxrpc_serial_t send_serial, rxrpc_serial_t resp_serial,
ktime_t send_time, ktime_t resp_time)
{
struct rxrpc_peer *peer = call->peer;
s64 rtt;
u64 sum = peer->rtt_sum, avg;
u8 cursor = peer->rtt_cursor, usage = peer->rtt_usage;
rtt = ktime_to_ns(ktime_sub(resp_time, send_time));
if (rtt < 0)
return;
spin_lock(&peer->rtt_input_lock);
/* Replace the oldest datum in the RTT buffer */
sum -= peer->rtt_cache[cursor];
sum += rtt;
peer->rtt_cache[cursor] = rtt;
peer->rtt_cursor = (cursor + 1) & (RXRPC_RTT_CACHE_SIZE - 1);
peer->rtt_sum = sum;
if (usage < RXRPC_RTT_CACHE_SIZE) {
usage++;
peer->rtt_usage = usage;
}
spin_unlock(&peer->rtt_input_lock);
/* Now recalculate the average */
if (usage == RXRPC_RTT_CACHE_SIZE) {
avg = sum / RXRPC_RTT_CACHE_SIZE;
} else {
avg = sum;
do_div(avg, usage);
}
/* Don't need to update this under lock */
peer->rtt = avg;
trace_rxrpc_rtt_rx(call, why, send_serial, resp_serial, rtt,
usage, avg);
}
/* /*
* Perform keep-alive pings. * Perform keep-alive pings.
*/ */

View File

@ -225,6 +225,8 @@ struct rxrpc_peer *rxrpc_alloc_peer(struct rxrpc_local *local, gfp_t gfp)
spin_lock_init(&peer->rtt_input_lock); spin_lock_init(&peer->rtt_input_lock);
peer->debug_id = atomic_inc_return(&rxrpc_debug_id); peer->debug_id = atomic_inc_return(&rxrpc_debug_id);
rxrpc_peer_init_rtt(peer);
if (RXRPC_TX_SMSS > 2190) if (RXRPC_TX_SMSS > 2190)
peer->cong_cwnd = 2; peer->cong_cwnd = 2;
else if (RXRPC_TX_SMSS > 1095) else if (RXRPC_TX_SMSS > 1095)
@ -497,14 +499,14 @@ void rxrpc_kernel_get_peer(struct socket *sock, struct rxrpc_call *call,
EXPORT_SYMBOL(rxrpc_kernel_get_peer); EXPORT_SYMBOL(rxrpc_kernel_get_peer);
/** /**
* rxrpc_kernel_get_rtt - Get a call's peer RTT * rxrpc_kernel_get_srtt - Get a call's peer smoothed RTT
* @sock: The socket on which the call is in progress. * @sock: The socket on which the call is in progress.
* @call: The call to query * @call: The call to query
* *
* Get the call's peer RTT. * Get the call's peer smoothed RTT.
*/ */
u64 rxrpc_kernel_get_rtt(struct socket *sock, struct rxrpc_call *call) u32 rxrpc_kernel_get_srtt(struct socket *sock, struct rxrpc_call *call)
{ {
return call->peer->rtt; return call->peer->srtt_us >> 3;
} }
EXPORT_SYMBOL(rxrpc_kernel_get_rtt); EXPORT_SYMBOL(rxrpc_kernel_get_srtt);

View File

@ -222,7 +222,7 @@ static int rxrpc_peer_seq_show(struct seq_file *seq, void *v)
seq_puts(seq, seq_puts(seq,
"Proto Local " "Proto Local "
" Remote " " Remote "
" Use CW MTU LastUse RTT Rc\n" " Use CW MTU LastUse RTT RTO\n"
); );
return 0; return 0;
} }
@ -236,15 +236,15 @@ static int rxrpc_peer_seq_show(struct seq_file *seq, void *v)
now = ktime_get_seconds(); now = ktime_get_seconds();
seq_printf(seq, seq_printf(seq,
"UDP %-47.47s %-47.47s %3u" "UDP %-47.47s %-47.47s %3u"
" %3u %5u %6llus %12llu %2u\n", " %3u %5u %6llus %8u %8u\n",
lbuff, lbuff,
rbuff, rbuff,
atomic_read(&peer->usage), atomic_read(&peer->usage),
peer->cong_cwnd, peer->cong_cwnd,
peer->mtu, peer->mtu,
now - peer->last_tx_at, now - peer->last_tx_at,
peer->rtt, peer->srtt_us >> 3,
peer->rtt_cursor); jiffies_to_usecs(peer->rto_j));
return 0; return 0;
} }

195
net/rxrpc/rtt.c Normal file
View File

@ -0,0 +1,195 @@
// SPDX-License-Identifier: GPL-2.0
/* RTT/RTO calculation.
*
* Adapted from TCP for AF_RXRPC by David Howells (dhowells@redhat.com)
*
* https://tools.ietf.org/html/rfc6298
* https://tools.ietf.org/html/rfc1122#section-4.2.3.1
* http://ccr.sigcomm.org/archive/1995/jan95/ccr-9501-partridge87.pdf
*/
#include <linux/net.h>
#include "ar-internal.h"
#define RXRPC_RTO_MAX ((unsigned)(120 * HZ))
#define RXRPC_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
#define rxrpc_jiffies32 ((u32)jiffies) /* As rxrpc_jiffies32 */
#define rxrpc_min_rtt_wlen 300 /* As sysctl_tcp_min_rtt_wlen */
static u32 rxrpc_rto_min_us(struct rxrpc_peer *peer)
{
return 200;
}
static u32 __rxrpc_set_rto(const struct rxrpc_peer *peer)
{
return _usecs_to_jiffies((peer->srtt_us >> 3) + peer->rttvar_us);
}
static u32 rxrpc_bound_rto(u32 rto)
{
return min(rto, RXRPC_RTO_MAX);
}
/*
* Called to compute a smoothed rtt estimate. The data fed to this
* routine either comes from timestamps, or from segments that were
* known _not_ to have been retransmitted [see Karn/Partridge
* Proceedings SIGCOMM 87]. The algorithm is from the SIGCOMM 88
* piece by Van Jacobson.
* NOTE: the next three routines used to be one big routine.
* To save cycles in the RFC 1323 implementation it was better to break
* it up into three procedures. -- erics
*/
static void rxrpc_rtt_estimator(struct rxrpc_peer *peer, long sample_rtt_us)
{
long m = sample_rtt_us; /* RTT */
u32 srtt = peer->srtt_us;
/* The following amusing code comes from Jacobson's
* article in SIGCOMM '88. Note that rtt and mdev
* are scaled versions of rtt and mean deviation.
* This is designed to be as fast as possible
* m stands for "measurement".
*
* 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 increased quickly, decrease too quickly
* 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 (srtt != 0) {
m -= (srtt >> 3); /* m is now error in rtt est */
srtt += m; /* rtt = 7/8 rtt + 1/8 new */
if (m < 0) {
m = -m; /* m is now abs(error) */
m -= (peer->mdev_us >> 2); /* similar update on mdev */
/* This is similar to one of Eifel findings.
* Eifel blocks mdev updates when rtt decreases.
* This solution is a bit different: we use finer gain
* for mdev in this case (alpha*beta).
* Like Eifel it also prevents growth of rto,
* but also it limits too fast rto decreases,
* happening in pure Eifel.
*/
if (m > 0)
m >>= 3;
} else {
m -= (peer->mdev_us >> 2); /* similar update on mdev */
}
peer->mdev_us += m; /* mdev = 3/4 mdev + 1/4 new */
if (peer->mdev_us > peer->mdev_max_us) {
peer->mdev_max_us = peer->mdev_us;
if (peer->mdev_max_us > peer->rttvar_us)
peer->rttvar_us = peer->mdev_max_us;
}
} else {
/* no previous measure. */
srtt = m << 3; /* take the measured time to be rtt */
peer->mdev_us = m << 1; /* make sure rto = 3*rtt */
peer->rttvar_us = max(peer->mdev_us, rxrpc_rto_min_us(peer));
peer->mdev_max_us = peer->rttvar_us;
}
peer->srtt_us = max(1U, srtt);
}
/*
* Calculate rto without backoff. This is the second half of Van Jacobson's
* routine referred to above.
*/
static void rxrpc_set_rto(struct rxrpc_peer *peer)
{
u32 rto;
/* 1. If rtt variance happened to be less 50msec, it is hallucination.
* It cannot be less due to utterly erratic ACK generation made
* at least by solaris and freebsd. "Erratic ACKs" has _nothing_
* to do with delayed acks, because at cwnd>2 true delack timeout
* is invisible. Actually, Linux-2.4 also generates erratic
* ACKs in some circumstances.
*/
rto = __rxrpc_set_rto(peer);
/* 2. Fixups made earlier cannot be right.
* If we do not estimate RTO correctly without them,
* all the algo is pure shit and should be replaced
* with correct one. It is exactly, which we pretend to do.
*/
/* NOTE: clamping at RXRPC_RTO_MIN is not required, current algo
* guarantees that rto is higher.
*/
peer->rto_j = rxrpc_bound_rto(rto);
}
static void rxrpc_ack_update_rtt(struct rxrpc_peer *peer, long rtt_us)
{
if (rtt_us < 0)
return;
//rxrpc_update_rtt_min(peer, rtt_us);
rxrpc_rtt_estimator(peer, rtt_us);
rxrpc_set_rto(peer);
/* RFC6298: only reset backoff on valid RTT measurement. */
peer->backoff = 0;
}
/*
* Add RTT information to cache. This is called in softirq mode and has
* exclusive access to the peer RTT data.
*/
void rxrpc_peer_add_rtt(struct rxrpc_call *call, enum rxrpc_rtt_rx_trace why,
rxrpc_serial_t send_serial, rxrpc_serial_t resp_serial,
ktime_t send_time, ktime_t resp_time)
{
struct rxrpc_peer *peer = call->peer;
s64 rtt_us;
rtt_us = ktime_to_us(ktime_sub(resp_time, send_time));
if (rtt_us < 0)
return;
spin_lock(&peer->rtt_input_lock);
rxrpc_ack_update_rtt(peer, rtt_us);
if (peer->rtt_count < 3)
peer->rtt_count++;
spin_unlock(&peer->rtt_input_lock);
trace_rxrpc_rtt_rx(call, why, send_serial, resp_serial,
peer->srtt_us >> 3, peer->rto_j);
}
/*
* Get the retransmission timeout to set in jiffies, backing it off each time
* we retransmit.
*/
unsigned long rxrpc_get_rto_backoff(struct rxrpc_peer *peer, bool retrans)
{
u64 timo_j;
u8 backoff = READ_ONCE(peer->backoff);
timo_j = peer->rto_j;
timo_j <<= backoff;
if (retrans && timo_j * 2 <= RXRPC_RTO_MAX)
WRITE_ONCE(peer->backoff, backoff + 1);
if (timo_j < 1)
timo_j = 1;
return timo_j;
}
void rxrpc_peer_init_rtt(struct rxrpc_peer *peer)
{
peer->rto_j = RXRPC_TIMEOUT_INIT;
peer->mdev_us = jiffies_to_usecs(RXRPC_TIMEOUT_INIT);
peer->backoff = 0;
//minmax_reset(&peer->rtt_min, rxrpc_jiffies32, ~0U);
}

View File

@ -66,15 +66,14 @@ static int rxrpc_wait_for_tx_window_waitall(struct rxrpc_sock *rx,
struct rxrpc_call *call) struct rxrpc_call *call)
{ {
rxrpc_seq_t tx_start, tx_win; rxrpc_seq_t tx_start, tx_win;
signed long rtt2, timeout; signed long rtt, timeout;
u64 rtt;
rtt = READ_ONCE(call->peer->rtt); rtt = READ_ONCE(call->peer->srtt_us) >> 3;
rtt2 = nsecs_to_jiffies64(rtt) * 2; rtt = usecs_to_jiffies(rtt) * 2;
if (rtt2 < 2) if (rtt < 2)
rtt2 = 2; rtt = 2;
timeout = rtt2; timeout = rtt;
tx_start = READ_ONCE(call->tx_hard_ack); tx_start = READ_ONCE(call->tx_hard_ack);
for (;;) { for (;;) {
@ -92,7 +91,7 @@ static int rxrpc_wait_for_tx_window_waitall(struct rxrpc_sock *rx,
return -EINTR; return -EINTR;
if (tx_win != tx_start) { if (tx_win != tx_start) {
timeout = rtt2; timeout = rtt;
tx_start = tx_win; tx_start = tx_win;
} }
@ -271,16 +270,9 @@ static int rxrpc_queue_packet(struct rxrpc_sock *rx, struct rxrpc_call *call,
_debug("need instant resend %d", ret); _debug("need instant resend %d", ret);
rxrpc_instant_resend(call, ix); rxrpc_instant_resend(call, ix);
} else { } else {
unsigned long now = jiffies, resend_at; unsigned long now = jiffies;
unsigned long resend_at = now + call->peer->rto_j;
if (call->peer->rtt_usage > 1)
resend_at = nsecs_to_jiffies(call->peer->rtt * 3 / 2);
else
resend_at = rxrpc_resend_timeout;
if (resend_at < 1)
resend_at = 1;
resend_at += now;
WRITE_ONCE(call->resend_at, resend_at); WRITE_ONCE(call->resend_at, resend_at);
rxrpc_reduce_call_timer(call, resend_at, now, rxrpc_reduce_call_timer(call, resend_at, now,
rxrpc_timer_set_for_send); rxrpc_timer_set_for_send);

View File

@ -71,15 +71,6 @@ static struct ctl_table rxrpc_sysctl_table[] = {
.extra1 = (void *)&one_jiffy, .extra1 = (void *)&one_jiffy,
.extra2 = (void *)&max_jiffies, .extra2 = (void *)&max_jiffies,
}, },
{
.procname = "resend_timeout",
.data = &rxrpc_resend_timeout,
.maxlen = sizeof(unsigned long),
.mode = 0644,
.proc_handler = proc_doulongvec_ms_jiffies_minmax,
.extra1 = (void *)&one_jiffy,
.extra2 = (void *)&max_jiffies,
},
/* Non-time values */ /* Non-time values */
{ {