1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* Management of Tx window, Tx resend, ACKs and out-of-sequence reception 3 * 4 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved. 5 * Written by David Howells (dhowells@redhat.com) 6 */ 7 8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 9 10 #include <linux/module.h> 11 #include <linux/circ_buf.h> 12 #include <linux/net.h> 13 #include <linux/skbuff.h> 14 #include <linux/slab.h> 15 #include <linux/udp.h> 16 #include <net/sock.h> 17 #include <net/af_rxrpc.h> 18 #include "ar-internal.h" 19 20 /* 21 * Propose a PING ACK be sent. 22 */ 23 void rxrpc_propose_ping(struct rxrpc_call *call, u32 serial, 24 enum rxrpc_propose_ack_trace why) 25 { 26 ktime_t delay = ms_to_ktime(READ_ONCE(rxrpc_idle_ack_delay)); 27 ktime_t now = ktime_get_real(); 28 ktime_t ping_at = ktime_add(now, delay); 29 30 trace_rxrpc_propose_ack(call, why, RXRPC_ACK_PING, serial); 31 if (ktime_before(ping_at, call->ping_at)) { 32 call->ping_at = ping_at; 33 trace_rxrpc_timer_set(call, delay, rxrpc_timer_trace_ping); 34 } 35 } 36 37 /* 38 * Propose a DELAY ACK be sent in the future. 39 */ 40 void rxrpc_propose_delay_ACK(struct rxrpc_call *call, rxrpc_serial_t serial, 41 enum rxrpc_propose_ack_trace why) 42 { 43 ktime_t now = ktime_get_real(), delay; 44 45 trace_rxrpc_propose_ack(call, why, RXRPC_ACK_DELAY, serial); 46 47 if (call->srtt_us) 48 delay = (call->srtt_us >> 3) * NSEC_PER_USEC; 49 else 50 delay = ms_to_ktime(READ_ONCE(rxrpc_soft_ack_delay)); 51 ktime_add_ms(delay, call->tx_backoff); 52 53 call->delay_ack_at = ktime_add(now, delay); 54 trace_rxrpc_timer_set(call, delay, rxrpc_timer_trace_delayed_ack); 55 } 56 57 /* 58 * Retransmit one or more packets. 59 */ 60 static bool rxrpc_retransmit_data(struct rxrpc_call *call, 61 struct rxrpc_send_data_req *req) 62 { 63 struct rxrpc_txqueue *tq = req->tq; 64 unsigned int ix = req->seq & RXRPC_TXQ_MASK; 65 struct rxrpc_txbuf *txb = tq->bufs[ix]; 66 67 _enter("%x,%x,%x,%x", tq->qbase, req->seq, ix, txb->debug_id); 68 69 req->retrans = true; 70 trace_rxrpc_retransmit(call, req, txb); 71 72 txb->flags |= RXRPC_TXBUF_RESENT; 73 rxrpc_send_data_packet(call, req); 74 rxrpc_inc_stat(call->rxnet, stat_tx_data_retrans); 75 76 req->tq = NULL; 77 req->n = 0; 78 req->did_send = true; 79 req->now = ktime_get_real(); 80 return true; 81 } 82 83 /* 84 * Perform retransmission of NAK'd and unack'd packets. 85 */ 86 static void rxrpc_resend(struct rxrpc_call *call) 87 { 88 struct rxrpc_send_data_req req = { 89 .now = ktime_get_real(), 90 .trace = rxrpc_txdata_retransmit, 91 }; 92 struct rxrpc_txqueue *tq; 93 94 _enter("{%d,%d}", call->tx_bottom, call->tx_top); 95 96 trace_rxrpc_resend(call, call->acks_highest_serial); 97 98 /* Scan the transmission queue, looking for lost packets. */ 99 for (tq = call->tx_queue; tq; tq = tq->next) { 100 unsigned long lost = tq->segment_lost; 101 102 if (after(tq->qbase, call->tx_transmitted)) 103 break; 104 105 _debug("retr %16lx %u c=%08x [%x]", 106 tq->segment_acked, tq->nr_reported_acks, call->debug_id, tq->qbase); 107 _debug("lost %16lx", lost); 108 109 trace_rxrpc_resend_lost(call, tq, lost); 110 while (lost) { 111 unsigned int ix = __ffs(lost); 112 struct rxrpc_txbuf *txb = tq->bufs[ix]; 113 114 __clear_bit(ix, &lost); 115 rxrpc_see_txbuf(txb, rxrpc_txbuf_see_lost); 116 117 req.tq = tq; 118 req.seq = tq->qbase + ix; 119 req.n = 1; 120 rxrpc_retransmit_data(call, &req); 121 } 122 } 123 124 rxrpc_get_rto_backoff(call, req.did_send); 125 _leave(""); 126 } 127 128 /* 129 * Resend the highest-seq DATA packet so far transmitted for RACK-TLP [RFC8985 7.3]. 130 */ 131 void rxrpc_resend_tlp(struct rxrpc_call *call) 132 { 133 struct rxrpc_send_data_req req = { 134 .now = ktime_get_real(), 135 .seq = call->tx_transmitted, 136 .n = 1, 137 .tlp_probe = true, 138 .trace = rxrpc_txdata_tlp_retransmit, 139 }; 140 141 /* There's a chance it'll be on the tail segment of the queue. */ 142 req.tq = READ_ONCE(call->tx_qtail); 143 if (req.tq && 144 before(call->tx_transmitted, req.tq->qbase + RXRPC_NR_TXQUEUE)) { 145 rxrpc_retransmit_data(call, &req); 146 return; 147 } 148 149 for (req.tq = call->tx_queue; req.tq; req.tq = req.tq->next) { 150 if (after_eq(call->tx_transmitted, req.tq->qbase) && 151 before(call->tx_transmitted, req.tq->qbase + RXRPC_NR_TXQUEUE)) { 152 rxrpc_retransmit_data(call, &req); 153 return; 154 } 155 } 156 } 157 158 /* 159 * Start transmitting the reply to a service. This cancels the need to ACK the 160 * request if we haven't yet done so. 161 */ 162 static void rxrpc_begin_service_reply(struct rxrpc_call *call) 163 { 164 rxrpc_set_call_state(call, RXRPC_CALL_SERVER_SEND_REPLY); 165 if (call->ackr_reason == RXRPC_ACK_DELAY) 166 call->ackr_reason = 0; 167 call->delay_ack_at = KTIME_MAX; 168 trace_rxrpc_timer_can(call, rxrpc_timer_trace_delayed_ack); 169 } 170 171 /* 172 * Close the transmission phase. After this point there is no more data to be 173 * transmitted in the call. 174 */ 175 static void rxrpc_close_tx_phase(struct rxrpc_call *call) 176 { 177 _debug("________awaiting reply/ACK__________"); 178 179 switch (__rxrpc_call_state(call)) { 180 case RXRPC_CALL_CLIENT_SEND_REQUEST: 181 rxrpc_set_call_state(call, RXRPC_CALL_CLIENT_AWAIT_REPLY); 182 break; 183 case RXRPC_CALL_SERVER_SEND_REPLY: 184 rxrpc_set_call_state(call, RXRPC_CALL_SERVER_AWAIT_ACK); 185 break; 186 default: 187 break; 188 } 189 } 190 191 /* 192 * Transmit some as-yet untransmitted data, to a maximum of the supplied limit. 193 */ 194 static void rxrpc_transmit_fresh_data(struct rxrpc_call *call, unsigned int limit, 195 enum rxrpc_txdata_trace trace) 196 { 197 int space = rxrpc_tx_window_space(call); 198 199 if (!test_bit(RXRPC_CALL_EXPOSED, &call->flags)) { 200 if (call->send_top == call->tx_top) 201 return; 202 rxrpc_expose_client_call(call); 203 } 204 205 while (space > 0) { 206 struct rxrpc_send_data_req req = { 207 .now = ktime_get_real(), 208 .seq = call->tx_transmitted + 1, 209 .n = 0, 210 .trace = trace, 211 }; 212 struct rxrpc_txqueue *tq; 213 struct rxrpc_txbuf *txb; 214 rxrpc_seq_t send_top, seq; 215 int limit = min(space, max(call->peer->pmtud_jumbo, 1)); 216 217 /* Order send_top before the contents of the new txbufs and 218 * txqueue pointers 219 */ 220 send_top = smp_load_acquire(&call->send_top); 221 if (call->tx_top == send_top) 222 break; 223 224 trace_rxrpc_transmit(call, send_top, space); 225 226 tq = call->tx_qtail; 227 seq = call->tx_top; 228 trace_rxrpc_tq(call, tq, seq, rxrpc_tq_decant); 229 230 do { 231 int ix; 232 233 seq++; 234 ix = seq & RXRPC_TXQ_MASK; 235 if (!ix) { 236 tq = tq->next; 237 trace_rxrpc_tq(call, tq, seq, rxrpc_tq_decant_advance); 238 } 239 if (!req.tq) 240 req.tq = tq; 241 txb = tq->bufs[ix]; 242 req.n++; 243 if (!txb->jumboable) 244 break; 245 } while (req.n < limit && before(seq, send_top)); 246 247 if (txb->flags & RXRPC_LAST_PACKET) { 248 rxrpc_close_tx_phase(call); 249 tq = NULL; 250 } 251 call->tx_qtail = tq; 252 call->tx_top = seq; 253 254 space -= req.n; 255 rxrpc_send_data_packet(call, &req); 256 } 257 } 258 259 void rxrpc_transmit_some_data(struct rxrpc_call *call, unsigned int limit, 260 enum rxrpc_txdata_trace trace) 261 { 262 switch (__rxrpc_call_state(call)) { 263 case RXRPC_CALL_SERVER_ACK_REQUEST: 264 if (call->tx_bottom == READ_ONCE(call->send_top)) 265 return; 266 rxrpc_begin_service_reply(call); 267 fallthrough; 268 269 case RXRPC_CALL_SERVER_SEND_REPLY: 270 case RXRPC_CALL_CLIENT_SEND_REQUEST: 271 if (!rxrpc_tx_window_space(call)) 272 return; 273 if (call->tx_bottom == READ_ONCE(call->send_top)) { 274 rxrpc_inc_stat(call->rxnet, stat_tx_data_underflow); 275 return; 276 } 277 rxrpc_transmit_fresh_data(call, limit, trace); 278 break; 279 default: 280 return; 281 } 282 } 283 284 /* 285 * Ping the other end to fill our RTT cache and to retrieve the rwind 286 * and MTU parameters. 287 */ 288 static void rxrpc_send_initial_ping(struct rxrpc_call *call) 289 { 290 if (call->rtt_count < 3 || 291 ktime_before(ktime_add_ms(call->rtt_last_req, 1000), 292 ktime_get_real())) 293 rxrpc_send_ACK(call, RXRPC_ACK_PING, 0, 294 rxrpc_propose_ack_ping_for_params); 295 } 296 297 /* 298 * Handle retransmission and deferred ACK/abort generation. 299 */ 300 bool rxrpc_input_call_event(struct rxrpc_call *call) 301 { 302 struct sk_buff *skb; 303 ktime_t now, t; 304 bool did_receive = false, saw_ack = false; 305 s32 abort_code; 306 307 rxrpc_see_call(call, rxrpc_call_see_input); 308 309 //printk("\n--------------------\n"); 310 _enter("{%d,%s,%lx}", 311 call->debug_id, rxrpc_call_states[__rxrpc_call_state(call)], 312 call->events); 313 314 /* Handle abort request locklessly, vs rxrpc_propose_abort(). */ 315 abort_code = smp_load_acquire(&call->send_abort); 316 if (abort_code) { 317 rxrpc_abort_call(call, 0, call->send_abort, call->send_abort_err, 318 call->send_abort_why); 319 goto out; 320 } 321 322 do { 323 skb = __skb_dequeue(&call->rx_queue); 324 if (skb) { 325 struct rxrpc_skb_priv *sp = rxrpc_skb(skb); 326 327 if (__rxrpc_call_is_complete(call) || 328 skb->mark == RXRPC_SKB_MARK_ERROR) { 329 rxrpc_free_skb(skb, rxrpc_skb_put_call_rx); 330 goto out; 331 } 332 333 saw_ack |= sp->hdr.type == RXRPC_PACKET_TYPE_ACK; 334 335 rxrpc_input_call_packet(call, skb); 336 rxrpc_free_skb(skb, rxrpc_skb_put_call_rx); 337 did_receive = true; 338 } 339 340 t = ktime_sub(call->rack_timo_at, ktime_get_real()); 341 if (t <= 0) { 342 trace_rxrpc_timer_exp(call, t, 343 rxrpc_timer_trace_rack_off + call->rack_timer_mode); 344 call->rack_timo_at = KTIME_MAX; 345 rxrpc_rack_timer_expired(call, t); 346 } 347 348 } while (!skb_queue_empty(&call->rx_queue)); 349 350 /* If we see our async-event poke, check for timeout trippage. */ 351 now = ktime_get_real(); 352 t = ktime_sub(call->expect_rx_by, now); 353 if (t <= 0) { 354 trace_rxrpc_timer_exp(call, t, rxrpc_timer_trace_expect_rx); 355 goto expired; 356 } 357 358 t = ktime_sub(call->expect_req_by, now); 359 if (t <= 0) { 360 call->expect_req_by = KTIME_MAX; 361 if (__rxrpc_call_state(call) == RXRPC_CALL_SERVER_RECV_REQUEST) { 362 trace_rxrpc_timer_exp(call, t, rxrpc_timer_trace_idle); 363 goto expired; 364 } 365 } 366 367 t = ktime_sub(READ_ONCE(call->expect_term_by), now); 368 if (t <= 0) { 369 trace_rxrpc_timer_exp(call, t, rxrpc_timer_trace_hard); 370 goto expired; 371 } 372 373 t = ktime_sub(call->delay_ack_at, now); 374 if (t <= 0) { 375 trace_rxrpc_timer_exp(call, t, rxrpc_timer_trace_delayed_ack); 376 call->delay_ack_at = KTIME_MAX; 377 rxrpc_send_ACK(call, RXRPC_ACK_DELAY, 0, 378 rxrpc_propose_ack_delayed_ack); 379 } 380 381 t = ktime_sub(call->ping_at, now); 382 if (t <= 0) { 383 trace_rxrpc_timer_exp(call, t, rxrpc_timer_trace_ping); 384 call->ping_at = KTIME_MAX; 385 rxrpc_send_ACK(call, RXRPC_ACK_PING, 0, 386 rxrpc_propose_ack_ping_for_keepalive); 387 } 388 389 now = ktime_get_real(); 390 t = ktime_sub(call->keepalive_at, now); 391 if (t <= 0) { 392 trace_rxrpc_timer_exp(call, t, rxrpc_timer_trace_keepalive); 393 call->keepalive_at = KTIME_MAX; 394 rxrpc_send_ACK(call, RXRPC_ACK_PING, 0, 395 rxrpc_propose_ack_ping_for_keepalive); 396 } 397 398 if (test_and_clear_bit(RXRPC_CALL_EV_INITIAL_PING, &call->events)) 399 rxrpc_send_initial_ping(call); 400 401 rxrpc_transmit_some_data(call, UINT_MAX, rxrpc_txdata_new_data); 402 403 if (saw_ack) 404 rxrpc_congestion_degrade(call); 405 406 if (did_receive && 407 (__rxrpc_call_state(call) == RXRPC_CALL_CLIENT_SEND_REQUEST || 408 __rxrpc_call_state(call) == RXRPC_CALL_SERVER_SEND_REPLY)) { 409 t = ktime_sub(call->rack_timo_at, ktime_get_real()); 410 trace_rxrpc_rack(call, t); 411 } 412 413 /* Process events */ 414 if (test_and_clear_bit(RXRPC_CALL_EV_ACK_LOST, &call->events)) 415 rxrpc_send_ACK(call, RXRPC_ACK_PING, 0, 416 rxrpc_propose_ack_ping_for_lost_ack); 417 418 if (call->tx_nr_lost > 0 && 419 __rxrpc_call_state(call) != RXRPC_CALL_CLIENT_RECV_REPLY && 420 !test_bit(RXRPC_CALL_TX_ALL_ACKED, &call->flags)) 421 rxrpc_resend(call); 422 423 if (test_and_clear_bit(RXRPC_CALL_RX_IS_IDLE, &call->flags)) 424 rxrpc_send_ACK(call, RXRPC_ACK_IDLE, 0, 425 rxrpc_propose_ack_rx_idle); 426 427 if (call->ackr_nr_unacked > 2) { 428 if (call->rtt_count < 3) 429 rxrpc_send_ACK(call, RXRPC_ACK_PING, 0, 430 rxrpc_propose_ack_ping_for_rtt); 431 else if (ktime_before(ktime_add_ms(call->rtt_last_req, 1000), 432 ktime_get_real())) 433 rxrpc_send_ACK(call, RXRPC_ACK_PING, 0, 434 rxrpc_propose_ack_ping_for_old_rtt); 435 else 436 rxrpc_send_ACK(call, RXRPC_ACK_IDLE, 0, 437 rxrpc_propose_ack_input_data); 438 } 439 440 /* Make sure the timer is restarted */ 441 if (!__rxrpc_call_is_complete(call)) { 442 ktime_t next = READ_ONCE(call->expect_term_by), delay; 443 444 #define set(T) { ktime_t _t = (T); if (ktime_before(_t, next)) next = _t; } 445 446 set(call->expect_req_by); 447 set(call->expect_rx_by); 448 set(call->delay_ack_at); 449 set(call->rack_timo_at); 450 set(call->keepalive_at); 451 set(call->ping_at); 452 453 now = ktime_get_real(); 454 delay = ktime_sub(next, now); 455 if (delay <= 0) { 456 rxrpc_poke_call(call, rxrpc_call_poke_timer_now); 457 } else { 458 unsigned long nowj = jiffies, delayj, nextj; 459 460 delayj = umax(nsecs_to_jiffies(delay), 1); 461 nextj = nowj + delayj; 462 if (time_before(nextj, call->timer.expires) || 463 !timer_pending(&call->timer)) { 464 trace_rxrpc_timer_restart(call, delay, delayj); 465 timer_reduce(&call->timer, nextj); 466 } 467 } 468 } 469 470 out: 471 if (__rxrpc_call_is_complete(call)) { 472 del_timer_sync(&call->timer); 473 if (!test_bit(RXRPC_CALL_DISCONNECTED, &call->flags)) 474 rxrpc_disconnect_call(call); 475 if (call->security) 476 call->security->free_call_crypto(call); 477 } else { 478 if (did_receive && 479 call->peer->ackr_adv_pmtud && 480 call->peer->pmtud_pending) 481 rxrpc_send_probe_for_pmtud(call); 482 } 483 _leave(""); 484 return true; 485 486 expired: 487 if (test_bit(RXRPC_CALL_RX_HEARD, &call->flags) && 488 (int)call->conn->hi_serial - (int)call->rx_serial > 0) { 489 trace_rxrpc_call_reset(call); 490 rxrpc_abort_call(call, 0, RX_CALL_DEAD, -ECONNRESET, 491 rxrpc_abort_call_reset); 492 } else { 493 rxrpc_abort_call(call, 0, RX_CALL_TIMEOUT, -ETIME, 494 rxrpc_abort_call_timeout); 495 } 496 goto out; 497 } 498