redonkable
/
alistair23-linux
1
0
Fork 0
Code Releases Activity

RxRPC fixes 

-----BEGIN PGP SIGNATURE-----
 
 iQIzBAABCAAdFiEEqG5UsNXhtOCrfGQP+7dXa6fLC2sFAl7FQH4ACgkQ+7dXa6fL
 C2u05BAAlilXCYGZU23/yQmMxxkmcG+jW+oDV9ySNDl6iJT+OtKxHEGReDD4D2f0
 rhaBivgGcOnZy89AGjNrSVROGlwSBXl7ArJcjsfsx8AuNzxHUHQKWlW/k8n87qEt
 NCTze7f65IT6NowgYAFgJn5kIpY/9iKuNiCf6NGL3Z35wqxPvwNs6AQSGM495uvB
 el/ddkr8QzzjI9Ejsgzj94x4DAOjk4T4WzfWMAgyr1OEqz6vKNKkCwSKPySOsQAK
 72JRaGhWA9rfAOkA7nAZpnjHdfFYnkFBOVQzmswOJYRYe3D/QY5D9PUlGIQ5OSjL
 yV5YOi/+AUrSif79NfEYXga0r/NFJMFqBg2zo/eiSrhfZZFZMDcagnGhzpGjbYF1
 IaeIu4q/MQOQybi8m1GJhvFfPOhdKRn731jlsUvEoxK0TonSu/u64eus+qelQxOd
 uiIcu/kLxfPZSznUd8cXZ+Pffce0uBIRWq0nRQZ703TyHY+/gYo7ZGHr/FZNKaK4
 lRNP4Nu3goLQCI40R7y7USnpX+kWfd4mYC9zl+VBSXG1JymYbOezXYrNATBCqpo6
 9VoYtqDdo8ESksFUBqM7fGRDZ20nah6KdRGmnrPU+rpODHZEZmNN7D/rayU31wua
 VIbVw1WluvSVnQ8+b1BwJwvhJQ4CazyGcnbxDqx1zd07EjUiiJI=
 =jJvy
 -----END PGP SIGNATURE-----

Merge tag 'rxrpc-fixes-20200520' of git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs

David Howells says:

====================
rxrpc: Fix retransmission timeout and ACK discard

Here are a couple of fixes and an extra tracepoint for AF_RXRPC:

 (1) Calculate the RTO pretty much as TCP does, rather than making
     something up, including an initial 4s timeout (which causes return
     probes from the fileserver to fail if a packet goes missing), and add
     backoff.

 (2) Fix the discarding of out-of-order received ACKs.  We mustn't let the
     hard-ACK point regress, nor do we want to do unnecessary
     retransmission because the soft-ACK list regresses.  This is not
     trivial, however, due to some loose wording in various old protocol
     specs, the ACK field that should be used for this sometimes has the
     wrong information in it.

 (3) Add a tracepoint to log a discarded ACK.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
alistair/sunxi64-5.7-dsi
David S. Miller 2020-05-22 15:53:30 -07:00
parent 79dde73cf9 441fdee1ea
commit 4629ed2e48
17 changed files with 335 additions and 159 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
fs/afs/fs_probe.c
View File

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

18
fs/afs/vl_probe.c
View File

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

2
include/net/af_rxrpc.h
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_get_peer(struct socket *, struct rxrpc_call *,
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,
rxrpc_user_attach_call_t, unsigned long, gfp_t,
unsigned int);

52
include/trace/events/rxrpc.h
View File

@ -1112,18 +1112,17 @@ TRACE_EVENT(rxrpc_rtt_tx,
TRACE_EVENT(rxrpc_rtt_rx,
TP_PROTO(struct rxrpc_call *call, enum rxrpc_rtt_rx_trace why,
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(
__field(unsigned int, call )
__field(enum rxrpc_rtt_rx_trace, why )
__field(u8, nr )
__field(rxrpc_serial_t, send_serial )
__field(rxrpc_serial_t, resp_serial )
__field(s64, rtt )
__field(u64, avg )
__field(u32, rtt )
__field(u32, rto )
),
TP_fast_assign(
@ -1132,18 +1131,16 @@ TRACE_EVENT(rxrpc_rtt_rx,
__entry->send_serial = send_serial;
__entry->resp_serial = resp_serial;
__entry->rtt = rtt;
__entry->nr = nr;
__entry->avg = avg;
__entry->rto = rto;
),
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,
__print_symbolic(__entry->why, rxrpc_rtt_rx_traces),
__entry->send_serial,
__entry->resp_serial,
__entry->rtt,
__entry->nr,
__entry->avg)
__entry->rto)
);
TRACE_EVENT(rxrpc_timer,
@ -1544,6 +1541,41 @@ TRACE_EVENT(rxrpc_notify_socket,
__entry->serial)
);
TRACE_EVENT(rxrpc_rx_discard_ack,
TP_PROTO(unsigned int debug_id, rxrpc_serial_t serial,
rxrpc_seq_t first_soft_ack, rxrpc_seq_t call_ackr_first,
rxrpc_seq_t prev_pkt, rxrpc_seq_t call_ackr_prev),
TP_ARGS(debug_id, serial, first_soft_ack, call_ackr_first,
prev_pkt, call_ackr_prev),
TP_STRUCT__entry(
__field(unsigned int, debug_id )
__field(rxrpc_serial_t, serial )
__field(rxrpc_seq_t, first_soft_ack)
__field(rxrpc_seq_t, call_ackr_first)
__field(rxrpc_seq_t, prev_pkt)
__field(rxrpc_seq_t, call_ackr_prev)
),
TP_fast_assign(
__entry->debug_id = debug_id;
__entry->serial = serial;
__entry->first_soft_ack = first_soft_ack;
__entry->call_ackr_first = call_ackr_first;
__entry->prev_pkt = prev_pkt;
__entry->call_ackr_prev = call_ackr_prev;
),
TP_printk("c=%08x r=%08x %08x<%08x %08x<%08x",
__entry->debug_id,
__entry->serial,
__entry->first_soft_ack,
__entry->call_ackr_first,
__entry->prev_pkt,
__entry->call_ackr_prev)
);
#endif /* _TRACE_RXRPC_H */
/* This part must be outside protection */

1
net/rxrpc/Makefile
View File

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

25
net/rxrpc/ar-internal.h
View File

@ -7,6 +7,7 @@
#include <linux/atomic.h>
#include <linux/seqlock.h>
#include <linux/win_minmax.h>
#include <net/net_namespace.h>
#include <net/netns/generic.h>
#include <net/sock.h>
@ -311,11 +312,14 @@ struct rxrpc_peer {
#define RXRPC_RTT_CACHE_SIZE 32
spinlock_t rtt_input_lock; /* RTT lock for input routine */
ktime_t rtt_last_req; /* Time of last RTT request */
u64 rtt; /* Current RTT estimate (in nS) */
u64 rtt_sum; /* Sum of cache contents */
u64 rtt_cache[RXRPC_RTT_CACHE_SIZE]; /* Determined RTT cache */
u8 rtt_cursor; /* next entry at which to insert */
u8 rtt_usage; /* amount of cache actually used */
unsigned int rtt_count; /* Number of samples we've got */
u32 srtt_us; /* smoothed round trip time << 3 in usecs */
u32 mdev_us; /* medium deviation */
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 */
};
@ -1041,7 +1045,6 @@ extern unsigned long rxrpc_idle_ack_delay;
extern unsigned int rxrpc_rx_window_size;
extern unsigned int rxrpc_rx_mtu;
extern unsigned int rxrpc_rx_jumbo_max;
extern unsigned long rxrpc_resend_timeout;
extern const s8 rxrpc_ack_priority[];
@ -1069,8 +1072,6 @@ void rxrpc_send_keepalive(struct rxrpc_peer *);
* peer_event.c
*/
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 *);
/*
@ -1102,6 +1103,14 @@ extern const struct seq_operations rxrpc_peer_seq_ops;
void rxrpc_notify_socket(struct rxrpc_call *);
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
*/

2
net/rxrpc/call_accept.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);
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))
rxrpc_propose_ACK(call, RXRPC_ACK_PING, sp->hdr.serial,
true, true,

22
net/rxrpc/call_event.c
View File

@ -111,8 +111,8 @@ static void __rxrpc_propose_ACK(struct rxrpc_call *call, u8 ack_reason,
} else {
unsigned long now = jiffies, ack_at;
if (call->peer->rtt_usage > 0)
ack_at = nsecs_to_jiffies(call->peer->rtt);
if (call->peer->srtt_us != 0)
ack_at = usecs_to_jiffies(call->peer->srtt_us >> 3);
else
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)
{
struct sk_buff *skb;
unsigned long resend_at;
unsigned long resend_at, rto_j;
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;
u8 annotation, anno_type, retrans = 0, unacked = 0;
_enter("{%d,%d}", call->tx_hard_ack, call->tx_top);
if (call->peer->rtt_usage > 1)
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;
rto_j = call->peer->rto_j;
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);
@ -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 += jiffies + rxrpc_resend_timeout;
resend_at += jiffies + rto_j;
WRITE_ONCE(call->resend_at, resend_at);
if (unacked)
@ -234,7 +228,7 @@ static void rxrpc_resend(struct rxrpc_call *call, unsigned long now_j)
rxrpc_timer_set_for_resend);
spin_unlock_bh(&call->lock);
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;
rxrpc_propose_ACK(call, RXRPC_ACK_PING, 0, true, false,
rxrpc_propose_ack_ping_for_lost_ack);

44
net/rxrpc/input.c
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
* 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;
if (ktime_before(skb->tstamp,
ktime_add_ns(call->cong_tstamp,
call->peer->rtt)))
ktime_add_us(call->cong_tstamp,
call->peer->srtt_us >> 3)))
goto out_no_clear_ca;
change = rxrpc_cong_rtt_window_end;
call->cong_tstamp = skb->tstamp;
@ -802,6 +802,30 @@ static void rxrpc_input_soft_acks(struct rxrpc_call *call, u8 *acks,
}
}
/*
* Return true if the ACK is valid - ie. it doesn't appear to have regressed
* with respect to the ack state conveyed by preceding ACKs.
*/
static bool rxrpc_is_ack_valid(struct rxrpc_call *call,
rxrpc_seq_t first_pkt, rxrpc_seq_t prev_pkt)
{
rxrpc_seq_t base = READ_ONCE(call->ackr_first_seq);
if (after(first_pkt, base))
return true; /* The window advanced */
if (before(first_pkt, base))
return false; /* firstPacket regressed */
if (after_eq(prev_pkt, call->ackr_prev_seq))
return true; /* previousPacket hasn't regressed. */
/* Some rx implementations put a serial number in previousPacket. */
if (after_eq(prev_pkt, base + call->tx_winsize))
return false;
return true;
}
/*
* Process an ACK packet.
*
@ -865,9 +889,12 @@ static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb)
}
/* Discard any out-of-order or duplicate ACKs (outside lock). */
if (before(first_soft_ack, call->ackr_first_seq) ||
before(prev_pkt, call->ackr_prev_seq))
if (!rxrpc_is_ack_valid(call, first_soft_ack, prev_pkt)) {
trace_rxrpc_rx_discard_ack(call->debug_id, sp->hdr.serial,
first_soft_ack, call->ackr_first_seq,
prev_pkt, call->ackr_prev_seq);
return;
}
buf.info.rxMTU = 0;
ioffset = offset + nr_acks + 3;
@ -878,9 +905,12 @@ static void rxrpc_input_ack(struct rxrpc_call *call, struct sk_buff *skb)
spin_lock(&call->input_lock);
/* Discard any out-of-order or duplicate ACKs (inside lock). */
if (before(first_soft_ack, call->ackr_first_seq) ||
before(prev_pkt, call->ackr_prev_seq))
if (!rxrpc_is_ack_valid(call, first_soft_ack, prev_pkt)) {
trace_rxrpc_rx_discard_ack(call->debug_id, sp->hdr.serial,
first_soft_ack, call->ackr_first_seq,
prev_pkt, call->ackr_prev_seq);
goto out;
}
call->acks_latest_ts = skb->tstamp;
call->ackr_first_seq = first_soft_ack;

5
net/rxrpc/misc.c
View File

@ -63,11 +63,6 @@ unsigned int rxrpc_rx_mtu = 5692;
*/
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[] = {
[0] = 0,
[RXRPC_ACK_DELAY] = 1,

9
net/rxrpc/output.c
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) ||
retrans ||
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_get_real())))
whdr.flags |= RXRPC_REQUEST_ACK;
@ -423,13 +423,10 @@ done:
if (whdr.flags & RXRPC_REQUEST_ACK) {
call->peer->rtt_last_req = skb->tstamp;
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;
ack_lost_at = nsecs_to_jiffies(2 * call->peer->rtt);
if (ack_lost_at < 1)
ack_lost_at = 1;
ack_lost_at = rxrpc_get_rto_backoff(call->peer, retrans);
ack_lost_at += nowj;
WRITE_ONCE(call->ack_lost_at, ack_lost_at);
rxrpc_reduce_call_timer(call, ack_lost_at, nowj,

46
net/rxrpc/peer_event.c
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.
*/

12
net/rxrpc/peer_object.c
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);
peer->debug_id = atomic_inc_return(&rxrpc_debug_id);
rxrpc_peer_init_rtt(peer);
if (RXRPC_TX_SMSS > 2190)
peer->cong_cwnd = 2;
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);
/**
* 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.
* @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);

8
net/rxrpc/proc.c
View File

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

195
net/rxrpc/rtt.c 100644
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);
}

26
net/rxrpc/sendmsg.c
View File

@ -66,15 +66,14 @@ static int rxrpc_wait_for_tx_window_waitall(struct rxrpc_sock *rx,
struct rxrpc_call *call)
{
rxrpc_seq_t tx_start, tx_win;
signed long rtt2, timeout;
u64 rtt;
signed long rtt, timeout;
rtt = READ_ONCE(call->peer->rtt);
rtt2 = nsecs_to_jiffies64(rtt) * 2;
if (rtt2 < 2)
rtt2 = 2;
rtt = READ_ONCE(call->peer->srtt_us) >> 3;
rtt = usecs_to_jiffies(rtt) * 2;
if (rtt < 2)
rtt = 2;
timeout = rtt2;
timeout = rtt;
tx_start = READ_ONCE(call->tx_hard_ack);
for (;;) {
@ -92,7 +91,7 @@ static int rxrpc_wait_for_tx_window_waitall(struct rxrpc_sock *rx,
return -EINTR;
if (tx_win != tx_start) {
timeout = rtt2;
timeout = rtt;
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);
rxrpc_instant_resend(call, ix);
} 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);
rxrpc_reduce_call_timer(call, resend_at, now,
rxrpc_timer_set_for_send);

9
net/rxrpc/sysctl.c
View File

@ -71,15 +71,6 @@ static struct ctl_table rxrpc_sysctl_table[] = {
.extra1 = (void *)&one_jiffy,
.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 */
{