1 | // SPDX-License-Identifier: GPL-2.0-or-later |
2 | /* RxRPC packet transmission |
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/net.h> |
11 | #include <linux/gfp.h> |
12 | #include <linux/skbuff.h> |
13 | #include <linux/export.h> |
14 | #include <net/sock.h> |
15 | #include <net/af_rxrpc.h> |
16 | #include <net/udp.h> |
17 | #include "ar-internal.h" |
18 | |
19 | extern int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len); |
20 | |
21 | static ssize_t do_udp_sendmsg(struct socket *socket, struct msghdr *msg, size_t len) |
22 | { |
23 | struct sockaddr *sa = msg->msg_name; |
24 | struct sock *sk = socket->sk; |
25 | |
26 | if (IS_ENABLED(CONFIG_AF_RXRPC_IPV6)) { |
27 | if (sa->sa_family == AF_INET6) { |
28 | if (sk->sk_family != AF_INET6) { |
29 | pr_warn("AF_INET6 address on AF_INET socket\n" ); |
30 | return -ENOPROTOOPT; |
31 | } |
32 | return udpv6_sendmsg(sk, msg, len); |
33 | } |
34 | } |
35 | return udp_sendmsg(sk, msg, len); |
36 | } |
37 | |
38 | struct rxrpc_abort_buffer { |
39 | struct rxrpc_wire_header whdr; |
40 | __be32 abort_code; |
41 | }; |
42 | |
43 | static const char rxrpc_keepalive_string[] = "" ; |
44 | |
45 | /* |
46 | * Increase Tx backoff on transmission failure and clear it on success. |
47 | */ |
48 | static void rxrpc_tx_backoff(struct rxrpc_call *call, int ret) |
49 | { |
50 | if (ret < 0) { |
51 | u16 tx_backoff = READ_ONCE(call->tx_backoff); |
52 | |
53 | if (tx_backoff < HZ) |
54 | WRITE_ONCE(call->tx_backoff, tx_backoff + 1); |
55 | } else { |
56 | WRITE_ONCE(call->tx_backoff, 0); |
57 | } |
58 | } |
59 | |
60 | /* |
61 | * Arrange for a keepalive ping a certain time after we last transmitted. This |
62 | * lets the far side know we're still interested in this call and helps keep |
63 | * the route through any intervening firewall open. |
64 | * |
65 | * Receiving a response to the ping will prevent the ->expect_rx_by timer from |
66 | * expiring. |
67 | */ |
68 | static void rxrpc_set_keepalive(struct rxrpc_call *call) |
69 | { |
70 | unsigned long now = jiffies, keepalive_at = call->next_rx_timo / 6; |
71 | |
72 | keepalive_at += now; |
73 | WRITE_ONCE(call->keepalive_at, keepalive_at); |
74 | rxrpc_reduce_call_timer(call, expire_at: keepalive_at, now, |
75 | why: rxrpc_timer_set_for_keepalive); |
76 | } |
77 | |
78 | /* |
79 | * Fill out an ACK packet. |
80 | */ |
81 | static size_t rxrpc_fill_out_ack(struct rxrpc_connection *conn, |
82 | struct rxrpc_call *call, |
83 | struct rxrpc_txbuf *txb, |
84 | u16 *_rwind) |
85 | { |
86 | struct rxrpc_ackinfo ackinfo; |
87 | unsigned int qsize, sack, wrap, to; |
88 | rxrpc_seq_t window, wtop; |
89 | int rsize; |
90 | u32 mtu, jmax; |
91 | u8 *ackp = txb->acks; |
92 | |
93 | call->ackr_nr_unacked = 0; |
94 | atomic_set(v: &call->ackr_nr_consumed, i: 0); |
95 | rxrpc_inc_stat(call->rxnet, stat_tx_ack_fill); |
96 | clear_bit(nr: RXRPC_CALL_RX_IS_IDLE, addr: &call->flags); |
97 | |
98 | window = call->ackr_window; |
99 | wtop = call->ackr_wtop; |
100 | sack = call->ackr_sack_base % RXRPC_SACK_SIZE; |
101 | txb->ack.firstPacket = htonl(window); |
102 | txb->ack.nAcks = wtop - window; |
103 | |
104 | if (after(seq1: wtop, seq2: window)) { |
105 | wrap = RXRPC_SACK_SIZE - sack; |
106 | to = min_t(unsigned int, txb->ack.nAcks, RXRPC_SACK_SIZE); |
107 | |
108 | if (sack + txb->ack.nAcks <= RXRPC_SACK_SIZE) { |
109 | memcpy(txb->acks, call->ackr_sack_table + sack, txb->ack.nAcks); |
110 | } else { |
111 | memcpy(txb->acks, call->ackr_sack_table + sack, wrap); |
112 | memcpy(txb->acks + wrap, call->ackr_sack_table, |
113 | to - wrap); |
114 | } |
115 | |
116 | ackp += to; |
117 | } else if (before(seq1: wtop, seq2: window)) { |
118 | pr_warn("ack window backward %x %x" , window, wtop); |
119 | } else if (txb->ack.reason == RXRPC_ACK_DELAY) { |
120 | txb->ack.reason = RXRPC_ACK_IDLE; |
121 | } |
122 | |
123 | mtu = conn->peer->if_mtu; |
124 | mtu -= conn->peer->hdrsize; |
125 | jmax = rxrpc_rx_jumbo_max; |
126 | qsize = (window - 1) - call->rx_consumed; |
127 | rsize = max_t(int, call->rx_winsize - qsize, 0); |
128 | *_rwind = rsize; |
129 | ackinfo.rxMTU = htonl(rxrpc_rx_mtu); |
130 | ackinfo.maxMTU = htonl(mtu); |
131 | ackinfo.rwind = htonl(rsize); |
132 | ackinfo.jumbo_max = htonl(jmax); |
133 | |
134 | *ackp++ = 0; |
135 | *ackp++ = 0; |
136 | *ackp++ = 0; |
137 | memcpy(ackp, &ackinfo, sizeof(ackinfo)); |
138 | return txb->ack.nAcks + 3 + sizeof(ackinfo); |
139 | } |
140 | |
141 | /* |
142 | * Record the beginning of an RTT probe. |
143 | */ |
144 | static int rxrpc_begin_rtt_probe(struct rxrpc_call *call, rxrpc_serial_t serial, |
145 | enum rxrpc_rtt_tx_trace why) |
146 | { |
147 | unsigned long avail = call->rtt_avail; |
148 | int rtt_slot = 9; |
149 | |
150 | if (!(avail & RXRPC_CALL_RTT_AVAIL_MASK)) |
151 | goto no_slot; |
152 | |
153 | rtt_slot = __ffs(avail & RXRPC_CALL_RTT_AVAIL_MASK); |
154 | if (!test_and_clear_bit(nr: rtt_slot, addr: &call->rtt_avail)) |
155 | goto no_slot; |
156 | |
157 | call->rtt_serial[rtt_slot] = serial; |
158 | call->rtt_sent_at[rtt_slot] = ktime_get_real(); |
159 | smp_wmb(); /* Write data before avail bit */ |
160 | set_bit(nr: rtt_slot + RXRPC_CALL_RTT_PEND_SHIFT, addr: &call->rtt_avail); |
161 | |
162 | trace_rxrpc_rtt_tx(call, why, slot: rtt_slot, send_serial: serial); |
163 | return rtt_slot; |
164 | |
165 | no_slot: |
166 | trace_rxrpc_rtt_tx(call, why: rxrpc_rtt_tx_no_slot, slot: rtt_slot, send_serial: serial); |
167 | return -1; |
168 | } |
169 | |
170 | /* |
171 | * Cancel an RTT probe. |
172 | */ |
173 | static void rxrpc_cancel_rtt_probe(struct rxrpc_call *call, |
174 | rxrpc_serial_t serial, int rtt_slot) |
175 | { |
176 | if (rtt_slot != -1) { |
177 | clear_bit(nr: rtt_slot + RXRPC_CALL_RTT_PEND_SHIFT, addr: &call->rtt_avail); |
178 | smp_wmb(); /* Clear pending bit before setting slot */ |
179 | set_bit(nr: rtt_slot, addr: &call->rtt_avail); |
180 | trace_rxrpc_rtt_tx(call, why: rxrpc_rtt_tx_cancel, slot: rtt_slot, send_serial: serial); |
181 | } |
182 | } |
183 | |
184 | /* |
185 | * Transmit an ACK packet. |
186 | */ |
187 | int rxrpc_send_ack_packet(struct rxrpc_call *call, struct rxrpc_txbuf *txb) |
188 | { |
189 | struct rxrpc_connection *conn; |
190 | struct msghdr msg; |
191 | struct kvec iov[1]; |
192 | rxrpc_serial_t serial; |
193 | size_t len, n; |
194 | int ret, rtt_slot = -1; |
195 | u16 rwind; |
196 | |
197 | if (test_bit(RXRPC_CALL_DISCONNECTED, &call->flags)) |
198 | return -ECONNRESET; |
199 | |
200 | conn = call->conn; |
201 | |
202 | msg.msg_name = &call->peer->srx.transport; |
203 | msg.msg_namelen = call->peer->srx.transport_len; |
204 | msg.msg_control = NULL; |
205 | msg.msg_controllen = 0; |
206 | msg.msg_flags = 0; |
207 | |
208 | if (txb->ack.reason == RXRPC_ACK_PING) |
209 | txb->wire.flags |= RXRPC_REQUEST_ACK; |
210 | |
211 | n = rxrpc_fill_out_ack(conn, call, txb, rwind: &rwind); |
212 | if (n == 0) |
213 | return 0; |
214 | |
215 | iov[0].iov_base = &txb->wire; |
216 | iov[0].iov_len = sizeof(txb->wire) + sizeof(txb->ack) + n; |
217 | len = iov[0].iov_len; |
218 | |
219 | serial = atomic_inc_return(v: &conn->serial); |
220 | txb->wire.serial = htonl(serial); |
221 | trace_rxrpc_tx_ack(call: call->debug_id, serial, |
222 | ntohl(txb->ack.firstPacket), |
223 | ntohl(txb->ack.serial), reason: txb->ack.reason, n_acks: txb->ack.nAcks, |
224 | rwind); |
225 | |
226 | if (txb->ack.reason == RXRPC_ACK_PING) |
227 | rtt_slot = rxrpc_begin_rtt_probe(call, serial, why: rxrpc_rtt_tx_ping); |
228 | |
229 | rxrpc_inc_stat(call->rxnet, stat_tx_ack_send); |
230 | |
231 | /* Grab the highest received seq as late as possible */ |
232 | txb->ack.previousPacket = htonl(call->rx_highest_seq); |
233 | |
234 | iov_iter_kvec(i: &msg.msg_iter, WRITE, kvec: iov, nr_segs: 1, count: len); |
235 | ret = do_udp_sendmsg(socket: conn->local->socket, msg: &msg, len); |
236 | call->peer->last_tx_at = ktime_get_seconds(); |
237 | if (ret < 0) { |
238 | trace_rxrpc_tx_fail(debug_id: call->debug_id, serial, ret, |
239 | where: rxrpc_tx_point_call_ack); |
240 | } else { |
241 | trace_rxrpc_tx_packet(call_id: call->debug_id, whdr: &txb->wire, |
242 | where: rxrpc_tx_point_call_ack); |
243 | if (txb->wire.flags & RXRPC_REQUEST_ACK) |
244 | call->peer->rtt_last_req = ktime_get_real(); |
245 | } |
246 | rxrpc_tx_backoff(call, ret); |
247 | |
248 | if (!__rxrpc_call_is_complete(call)) { |
249 | if (ret < 0) |
250 | rxrpc_cancel_rtt_probe(call, serial, rtt_slot); |
251 | rxrpc_set_keepalive(call); |
252 | } |
253 | |
254 | return ret; |
255 | } |
256 | |
257 | /* |
258 | * Send an ABORT call packet. |
259 | */ |
260 | int rxrpc_send_abort_packet(struct rxrpc_call *call) |
261 | { |
262 | struct rxrpc_connection *conn; |
263 | struct rxrpc_abort_buffer pkt; |
264 | struct msghdr msg; |
265 | struct kvec iov[1]; |
266 | rxrpc_serial_t serial; |
267 | int ret; |
268 | |
269 | /* Don't bother sending aborts for a client call once the server has |
270 | * hard-ACK'd all of its request data. After that point, we're not |
271 | * going to stop the operation proceeding, and whilst we might limit |
272 | * the reply, it's not worth it if we can send a new call on the same |
273 | * channel instead, thereby closing off this call. |
274 | */ |
275 | if (rxrpc_is_client_call(call) && |
276 | test_bit(RXRPC_CALL_TX_ALL_ACKED, &call->flags)) |
277 | return 0; |
278 | |
279 | if (test_bit(RXRPC_CALL_DISCONNECTED, &call->flags)) |
280 | return -ECONNRESET; |
281 | |
282 | conn = call->conn; |
283 | |
284 | msg.msg_name = &call->peer->srx.transport; |
285 | msg.msg_namelen = call->peer->srx.transport_len; |
286 | msg.msg_control = NULL; |
287 | msg.msg_controllen = 0; |
288 | msg.msg_flags = 0; |
289 | |
290 | pkt.whdr.epoch = htonl(conn->proto.epoch); |
291 | pkt.whdr.cid = htonl(call->cid); |
292 | pkt.whdr.callNumber = htonl(call->call_id); |
293 | pkt.whdr.seq = 0; |
294 | pkt.whdr.type = RXRPC_PACKET_TYPE_ABORT; |
295 | pkt.whdr.flags = conn->out_clientflag; |
296 | pkt.whdr.userStatus = 0; |
297 | pkt.whdr.securityIndex = call->security_ix; |
298 | pkt.whdr._rsvd = 0; |
299 | pkt.whdr.serviceId = htons(call->dest_srx.srx_service); |
300 | pkt.abort_code = htonl(call->abort_code); |
301 | |
302 | iov[0].iov_base = &pkt; |
303 | iov[0].iov_len = sizeof(pkt); |
304 | |
305 | serial = atomic_inc_return(v: &conn->serial); |
306 | pkt.whdr.serial = htonl(serial); |
307 | |
308 | iov_iter_kvec(i: &msg.msg_iter, WRITE, kvec: iov, nr_segs: 1, count: sizeof(pkt)); |
309 | ret = do_udp_sendmsg(socket: conn->local->socket, msg: &msg, len: sizeof(pkt)); |
310 | conn->peer->last_tx_at = ktime_get_seconds(); |
311 | if (ret < 0) |
312 | trace_rxrpc_tx_fail(debug_id: call->debug_id, serial, ret, |
313 | where: rxrpc_tx_point_call_abort); |
314 | else |
315 | trace_rxrpc_tx_packet(call_id: call->debug_id, whdr: &pkt.whdr, |
316 | where: rxrpc_tx_point_call_abort); |
317 | rxrpc_tx_backoff(call, ret); |
318 | return ret; |
319 | } |
320 | |
321 | /* |
322 | * send a packet through the transport endpoint |
323 | */ |
324 | int rxrpc_send_data_packet(struct rxrpc_call *call, struct rxrpc_txbuf *txb) |
325 | { |
326 | enum rxrpc_req_ack_trace why; |
327 | struct rxrpc_connection *conn = call->conn; |
328 | struct msghdr msg; |
329 | struct kvec iov[1]; |
330 | rxrpc_serial_t serial; |
331 | size_t len; |
332 | int ret, rtt_slot = -1; |
333 | |
334 | _enter("%x,{%d}" , txb->seq, txb->len); |
335 | |
336 | /* Each transmission of a Tx packet needs a new serial number */ |
337 | serial = atomic_inc_return(v: &conn->serial); |
338 | txb->wire.serial = htonl(serial); |
339 | |
340 | if (test_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags) && |
341 | txb->seq == 1) |
342 | txb->wire.userStatus = RXRPC_USERSTATUS_SERVICE_UPGRADE; |
343 | |
344 | iov[0].iov_base = &txb->wire; |
345 | iov[0].iov_len = sizeof(txb->wire) + txb->len; |
346 | len = iov[0].iov_len; |
347 | iov_iter_kvec(i: &msg.msg_iter, WRITE, kvec: iov, nr_segs: 1, count: len); |
348 | |
349 | msg.msg_name = &call->peer->srx.transport; |
350 | msg.msg_namelen = call->peer->srx.transport_len; |
351 | msg.msg_control = NULL; |
352 | msg.msg_controllen = 0; |
353 | msg.msg_flags = 0; |
354 | |
355 | /* If our RTT cache needs working on, request an ACK. Also request |
356 | * ACKs if a DATA packet appears to have been lost. |
357 | * |
358 | * However, we mustn't request an ACK on the last reply packet of a |
359 | * service call, lest OpenAFS incorrectly send us an ACK with some |
360 | * soft-ACKs in it and then never follow up with a proper hard ACK. |
361 | */ |
362 | if (txb->wire.flags & RXRPC_REQUEST_ACK) |
363 | why = rxrpc_reqack_already_on; |
364 | else if (test_bit(RXRPC_TXBUF_LAST, &txb->flags) && rxrpc_sending_to_client(txb)) |
365 | why = rxrpc_reqack_no_srv_last; |
366 | else if (test_and_clear_bit(nr: RXRPC_CALL_EV_ACK_LOST, addr: &call->events)) |
367 | why = rxrpc_reqack_ack_lost; |
368 | else if (test_bit(RXRPC_TXBUF_RESENT, &txb->flags)) |
369 | why = rxrpc_reqack_retrans; |
370 | else if (call->cong_mode == RXRPC_CALL_SLOW_START && call->cong_cwnd <= 2) |
371 | why = rxrpc_reqack_slow_start; |
372 | else if (call->tx_winsize <= 2) |
373 | why = rxrpc_reqack_small_txwin; |
374 | else if (call->peer->rtt_count < 3 && txb->seq & 1) |
375 | why = rxrpc_reqack_more_rtt; |
376 | else if (ktime_before(cmp1: ktime_add_ms(kt: call->peer->rtt_last_req, msec: 1000), cmp2: ktime_get_real())) |
377 | why = rxrpc_reqack_old_rtt; |
378 | else |
379 | goto dont_set_request_ack; |
380 | |
381 | rxrpc_inc_stat(call->rxnet, stat_why_req_ack[why]); |
382 | trace_rxrpc_req_ack(call_debug_id: call->debug_id, seq: txb->seq, why); |
383 | if (why != rxrpc_reqack_no_srv_last) |
384 | txb->wire.flags |= RXRPC_REQUEST_ACK; |
385 | dont_set_request_ack: |
386 | |
387 | if (IS_ENABLED(CONFIG_AF_RXRPC_INJECT_LOSS)) { |
388 | static int lose; |
389 | if ((lose++ & 7) == 7) { |
390 | ret = 0; |
391 | trace_rxrpc_tx_data(call, seq: txb->seq, serial, |
392 | flags: txb->wire.flags, |
393 | test_bit(RXRPC_TXBUF_RESENT, &txb->flags), |
394 | lose: true); |
395 | goto done; |
396 | } |
397 | } |
398 | |
399 | trace_rxrpc_tx_data(call, seq: txb->seq, serial, flags: txb->wire.flags, |
400 | test_bit(RXRPC_TXBUF_RESENT, &txb->flags), lose: false); |
401 | |
402 | /* Track what we've attempted to transmit at least once so that the |
403 | * retransmission algorithm doesn't try to resend what we haven't sent |
404 | * yet. However, this can race as we can receive an ACK before we get |
405 | * to this point. But, OTOH, if we won't get an ACK mentioning this |
406 | * packet unless the far side received it (though it could have |
407 | * discarded it anyway and NAK'd it). |
408 | */ |
409 | cmpxchg(&call->tx_transmitted, txb->seq - 1, txb->seq); |
410 | |
411 | /* send the packet with the don't fragment bit set if we currently |
412 | * think it's small enough */ |
413 | if (txb->len >= call->peer->maxdata) |
414 | goto send_fragmentable; |
415 | |
416 | txb->last_sent = ktime_get_real(); |
417 | if (txb->wire.flags & RXRPC_REQUEST_ACK) |
418 | rtt_slot = rxrpc_begin_rtt_probe(call, serial, why: rxrpc_rtt_tx_data); |
419 | |
420 | /* send the packet by UDP |
421 | * - returns -EMSGSIZE if UDP would have to fragment the packet |
422 | * to go out of the interface |
423 | * - in which case, we'll have processed the ICMP error |
424 | * message and update the peer record |
425 | */ |
426 | rxrpc_inc_stat(call->rxnet, stat_tx_data_send); |
427 | ret = do_udp_sendmsg(socket: conn->local->socket, msg: &msg, len); |
428 | conn->peer->last_tx_at = ktime_get_seconds(); |
429 | |
430 | if (ret < 0) { |
431 | rxrpc_inc_stat(call->rxnet, stat_tx_data_send_fail); |
432 | rxrpc_cancel_rtt_probe(call, serial, rtt_slot); |
433 | trace_rxrpc_tx_fail(debug_id: call->debug_id, serial, ret, |
434 | where: rxrpc_tx_point_call_data_nofrag); |
435 | } else { |
436 | trace_rxrpc_tx_packet(call_id: call->debug_id, whdr: &txb->wire, |
437 | where: rxrpc_tx_point_call_data_nofrag); |
438 | } |
439 | |
440 | rxrpc_tx_backoff(call, ret); |
441 | if (ret == -EMSGSIZE) |
442 | goto send_fragmentable; |
443 | |
444 | done: |
445 | if (ret >= 0) { |
446 | call->tx_last_sent = txb->last_sent; |
447 | if (txb->wire.flags & RXRPC_REQUEST_ACK) { |
448 | call->peer->rtt_last_req = txb->last_sent; |
449 | if (call->peer->rtt_count > 1) { |
450 | unsigned long nowj = jiffies, ack_lost_at; |
451 | |
452 | ack_lost_at = rxrpc_get_rto_backoff(call->peer, false); |
453 | ack_lost_at += nowj; |
454 | WRITE_ONCE(call->ack_lost_at, ack_lost_at); |
455 | rxrpc_reduce_call_timer(call, expire_at: ack_lost_at, now: nowj, |
456 | why: rxrpc_timer_set_for_lost_ack); |
457 | } |
458 | } |
459 | |
460 | if (txb->seq == 1 && |
461 | !test_and_set_bit(nr: RXRPC_CALL_BEGAN_RX_TIMER, |
462 | addr: &call->flags)) { |
463 | unsigned long nowj = jiffies, expect_rx_by; |
464 | |
465 | expect_rx_by = nowj + call->next_rx_timo; |
466 | WRITE_ONCE(call->expect_rx_by, expect_rx_by); |
467 | rxrpc_reduce_call_timer(call, expire_at: expect_rx_by, now: nowj, |
468 | why: rxrpc_timer_set_for_normal); |
469 | } |
470 | |
471 | rxrpc_set_keepalive(call); |
472 | } else { |
473 | /* Cancel the call if the initial transmission fails, |
474 | * particularly if that's due to network routing issues that |
475 | * aren't going away anytime soon. The layer above can arrange |
476 | * the retransmission. |
477 | */ |
478 | if (!test_and_set_bit(nr: RXRPC_CALL_BEGAN_RX_TIMER, addr: &call->flags)) |
479 | rxrpc_set_call_completion(call, compl: RXRPC_CALL_LOCAL_ERROR, |
480 | RX_USER_ABORT, error: ret); |
481 | } |
482 | |
483 | _leave(" = %d [%u]" , ret, call->peer->maxdata); |
484 | return ret; |
485 | |
486 | send_fragmentable: |
487 | /* attempt to send this message with fragmentation enabled */ |
488 | _debug("send fragment" ); |
489 | |
490 | txb->last_sent = ktime_get_real(); |
491 | if (txb->wire.flags & RXRPC_REQUEST_ACK) |
492 | rtt_slot = rxrpc_begin_rtt_probe(call, serial, why: rxrpc_rtt_tx_data); |
493 | |
494 | switch (conn->local->srx.transport.family) { |
495 | case AF_INET6: |
496 | case AF_INET: |
497 | ip_sock_set_mtu_discover(sk: conn->local->socket->sk, |
498 | IP_PMTUDISC_DONT); |
499 | rxrpc_inc_stat(call->rxnet, stat_tx_data_send_frag); |
500 | ret = do_udp_sendmsg(socket: conn->local->socket, msg: &msg, len); |
501 | conn->peer->last_tx_at = ktime_get_seconds(); |
502 | |
503 | ip_sock_set_mtu_discover(sk: conn->local->socket->sk, |
504 | IP_PMTUDISC_DO); |
505 | break; |
506 | |
507 | default: |
508 | BUG(); |
509 | } |
510 | |
511 | if (ret < 0) { |
512 | rxrpc_inc_stat(call->rxnet, stat_tx_data_send_fail); |
513 | rxrpc_cancel_rtt_probe(call, serial, rtt_slot); |
514 | trace_rxrpc_tx_fail(debug_id: call->debug_id, serial, ret, |
515 | where: rxrpc_tx_point_call_data_frag); |
516 | } else { |
517 | trace_rxrpc_tx_packet(call_id: call->debug_id, whdr: &txb->wire, |
518 | where: rxrpc_tx_point_call_data_frag); |
519 | } |
520 | rxrpc_tx_backoff(call, ret); |
521 | goto done; |
522 | } |
523 | |
524 | /* |
525 | * Transmit a connection-level abort. |
526 | */ |
527 | void rxrpc_send_conn_abort(struct rxrpc_connection *conn) |
528 | { |
529 | struct rxrpc_wire_header whdr; |
530 | struct msghdr msg; |
531 | struct kvec iov[2]; |
532 | __be32 word; |
533 | size_t len; |
534 | u32 serial; |
535 | int ret; |
536 | |
537 | msg.msg_name = &conn->peer->srx.transport; |
538 | msg.msg_namelen = conn->peer->srx.transport_len; |
539 | msg.msg_control = NULL; |
540 | msg.msg_controllen = 0; |
541 | msg.msg_flags = 0; |
542 | |
543 | whdr.epoch = htonl(conn->proto.epoch); |
544 | whdr.cid = htonl(conn->proto.cid); |
545 | whdr.callNumber = 0; |
546 | whdr.seq = 0; |
547 | whdr.type = RXRPC_PACKET_TYPE_ABORT; |
548 | whdr.flags = conn->out_clientflag; |
549 | whdr.userStatus = 0; |
550 | whdr.securityIndex = conn->security_ix; |
551 | whdr._rsvd = 0; |
552 | whdr.serviceId = htons(conn->service_id); |
553 | |
554 | word = htonl(conn->abort_code); |
555 | |
556 | iov[0].iov_base = &whdr; |
557 | iov[0].iov_len = sizeof(whdr); |
558 | iov[1].iov_base = &word; |
559 | iov[1].iov_len = sizeof(word); |
560 | |
561 | len = iov[0].iov_len + iov[1].iov_len; |
562 | |
563 | serial = atomic_inc_return(v: &conn->serial); |
564 | whdr.serial = htonl(serial); |
565 | |
566 | iov_iter_kvec(i: &msg.msg_iter, WRITE, kvec: iov, nr_segs: 2, count: len); |
567 | ret = do_udp_sendmsg(socket: conn->local->socket, msg: &msg, len); |
568 | if (ret < 0) { |
569 | trace_rxrpc_tx_fail(debug_id: conn->debug_id, serial, ret, |
570 | where: rxrpc_tx_point_conn_abort); |
571 | _debug("sendmsg failed: %d" , ret); |
572 | return; |
573 | } |
574 | |
575 | trace_rxrpc_tx_packet(call_id: conn->debug_id, whdr: &whdr, where: rxrpc_tx_point_conn_abort); |
576 | |
577 | conn->peer->last_tx_at = ktime_get_seconds(); |
578 | } |
579 | |
580 | /* |
581 | * Reject a packet through the local endpoint. |
582 | */ |
583 | void rxrpc_reject_packet(struct rxrpc_local *local, struct sk_buff *skb) |
584 | { |
585 | struct rxrpc_wire_header whdr; |
586 | struct sockaddr_rxrpc srx; |
587 | struct rxrpc_skb_priv *sp = rxrpc_skb(skb); |
588 | struct msghdr msg; |
589 | struct kvec iov[2]; |
590 | size_t size; |
591 | __be32 code; |
592 | int ret, ioc; |
593 | |
594 | rxrpc_see_skb(skb, rxrpc_skb_see_reject); |
595 | |
596 | iov[0].iov_base = &whdr; |
597 | iov[0].iov_len = sizeof(whdr); |
598 | iov[1].iov_base = &code; |
599 | iov[1].iov_len = sizeof(code); |
600 | |
601 | msg.msg_name = &srx.transport; |
602 | msg.msg_control = NULL; |
603 | msg.msg_controllen = 0; |
604 | msg.msg_flags = 0; |
605 | |
606 | memset(&whdr, 0, sizeof(whdr)); |
607 | |
608 | switch (skb->mark) { |
609 | case RXRPC_SKB_MARK_REJECT_BUSY: |
610 | whdr.type = RXRPC_PACKET_TYPE_BUSY; |
611 | size = sizeof(whdr); |
612 | ioc = 1; |
613 | break; |
614 | case RXRPC_SKB_MARK_REJECT_ABORT: |
615 | whdr.type = RXRPC_PACKET_TYPE_ABORT; |
616 | code = htonl(skb->priority); |
617 | size = sizeof(whdr) + sizeof(code); |
618 | ioc = 2; |
619 | break; |
620 | default: |
621 | return; |
622 | } |
623 | |
624 | if (rxrpc_extract_addr_from_skb(&srx, skb) == 0) { |
625 | msg.msg_namelen = srx.transport_len; |
626 | |
627 | whdr.epoch = htonl(sp->hdr.epoch); |
628 | whdr.cid = htonl(sp->hdr.cid); |
629 | whdr.callNumber = htonl(sp->hdr.callNumber); |
630 | whdr.serviceId = htons(sp->hdr.serviceId); |
631 | whdr.flags = sp->hdr.flags; |
632 | whdr.flags ^= RXRPC_CLIENT_INITIATED; |
633 | whdr.flags &= RXRPC_CLIENT_INITIATED; |
634 | |
635 | iov_iter_kvec(i: &msg.msg_iter, WRITE, kvec: iov, nr_segs: ioc, count: size); |
636 | ret = do_udp_sendmsg(socket: local->socket, msg: &msg, len: size); |
637 | if (ret < 0) |
638 | trace_rxrpc_tx_fail(debug_id: local->debug_id, serial: 0, ret, |
639 | where: rxrpc_tx_point_reject); |
640 | else |
641 | trace_rxrpc_tx_packet(call_id: local->debug_id, whdr: &whdr, |
642 | where: rxrpc_tx_point_reject); |
643 | } |
644 | } |
645 | |
646 | /* |
647 | * Send a VERSION reply to a peer as a keepalive. |
648 | */ |
649 | void rxrpc_send_keepalive(struct rxrpc_peer *peer) |
650 | { |
651 | struct rxrpc_wire_header whdr; |
652 | struct msghdr msg; |
653 | struct kvec iov[2]; |
654 | size_t len; |
655 | int ret; |
656 | |
657 | _enter("" ); |
658 | |
659 | msg.msg_name = &peer->srx.transport; |
660 | msg.msg_namelen = peer->srx.transport_len; |
661 | msg.msg_control = NULL; |
662 | msg.msg_controllen = 0; |
663 | msg.msg_flags = 0; |
664 | |
665 | whdr.epoch = htonl(peer->local->rxnet->epoch); |
666 | whdr.cid = 0; |
667 | whdr.callNumber = 0; |
668 | whdr.seq = 0; |
669 | whdr.serial = 0; |
670 | whdr.type = RXRPC_PACKET_TYPE_VERSION; /* Not client-initiated */ |
671 | whdr.flags = RXRPC_LAST_PACKET; |
672 | whdr.userStatus = 0; |
673 | whdr.securityIndex = 0; |
674 | whdr._rsvd = 0; |
675 | whdr.serviceId = 0; |
676 | |
677 | iov[0].iov_base = &whdr; |
678 | iov[0].iov_len = sizeof(whdr); |
679 | iov[1].iov_base = (char *)rxrpc_keepalive_string; |
680 | iov[1].iov_len = sizeof(rxrpc_keepalive_string); |
681 | |
682 | len = iov[0].iov_len + iov[1].iov_len; |
683 | |
684 | iov_iter_kvec(i: &msg.msg_iter, WRITE, kvec: iov, nr_segs: 2, count: len); |
685 | ret = do_udp_sendmsg(socket: peer->local->socket, msg: &msg, len); |
686 | if (ret < 0) |
687 | trace_rxrpc_tx_fail(debug_id: peer->debug_id, serial: 0, ret, |
688 | where: rxrpc_tx_point_version_keepalive); |
689 | else |
690 | trace_rxrpc_tx_packet(call_id: peer->debug_id, whdr: &whdr, |
691 | where: rxrpc_tx_point_version_keepalive); |
692 | |
693 | peer->last_tx_at = ktime_get_seconds(); |
694 | _leave("" ); |
695 | } |
696 | |
697 | /* |
698 | * Schedule an instant Tx resend. |
699 | */ |
700 | static inline void rxrpc_instant_resend(struct rxrpc_call *call, |
701 | struct rxrpc_txbuf *txb) |
702 | { |
703 | if (!__rxrpc_call_is_complete(call)) |
704 | kdebug("resend" ); |
705 | } |
706 | |
707 | /* |
708 | * Transmit one packet. |
709 | */ |
710 | void rxrpc_transmit_one(struct rxrpc_call *call, struct rxrpc_txbuf *txb) |
711 | { |
712 | int ret; |
713 | |
714 | ret = rxrpc_send_data_packet(call, txb); |
715 | if (ret < 0) { |
716 | switch (ret) { |
717 | case -ENETUNREACH: |
718 | case -EHOSTUNREACH: |
719 | case -ECONNREFUSED: |
720 | rxrpc_set_call_completion(call, compl: RXRPC_CALL_LOCAL_ERROR, |
721 | abort_code: 0, error: ret); |
722 | break; |
723 | default: |
724 | _debug("need instant resend %d" , ret); |
725 | rxrpc_instant_resend(call, txb); |
726 | } |
727 | } else { |
728 | unsigned long now = jiffies; |
729 | unsigned long resend_at = now + call->peer->rto_j; |
730 | |
731 | WRITE_ONCE(call->resend_at, resend_at); |
732 | rxrpc_reduce_call_timer(call, expire_at: resend_at, now, |
733 | why: rxrpc_timer_set_for_send); |
734 | } |
735 | } |
736 | |