]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
b38d7324 UB |
2 | /* |
3 | * Shared Memory Communications over RDMA (SMC-R) and RoCE | |
4 | * | |
5 | * Socket Closing - normal and abnormal | |
6 | * | |
7 | * Copyright IBM Corp. 2016 | |
8 | * | |
9 | * Author(s): Ursula Braun <ubraun@linux.vnet.ibm.com> | |
10 | */ | |
11 | ||
12 | #include <linux/workqueue.h> | |
c3edc401 IM |
13 | #include <linux/sched/signal.h> |
14 | ||
b38d7324 UB |
15 | #include <net/sock.h> |
16 | ||
17 | #include "smc.h" | |
18 | #include "smc_tx.h" | |
19 | #include "smc_cdc.h" | |
20 | #include "smc_close.h" | |
21 | ||
127f4970 UB |
22 | #define SMC_CLOSE_WAIT_LISTEN_CLCSOCK_TIME (5 * HZ) |
23 | ||
fd57770d KG |
24 | /* release the clcsock that is assigned to the smc_sock */ |
25 | void smc_clcsock_release(struct smc_sock *smc) | |
26 | { | |
27 | struct socket *tcp; | |
28 | ||
29 | if (smc->listen_smc && current_work() != &smc->smc_listen_work) | |
30 | cancel_work_sync(&smc->smc_listen_work); | |
31 | mutex_lock(&smc->clcsock_release_lock); | |
32 | if (smc->clcsock) { | |
33 | tcp = smc->clcsock; | |
34 | smc->clcsock = NULL; | |
35 | sock_release(tcp); | |
36 | } | |
37 | mutex_unlock(&smc->clcsock_release_lock); | |
38 | } | |
39 | ||
b38d7324 UB |
40 | static void smc_close_cleanup_listen(struct sock *parent) |
41 | { | |
42 | struct sock *sk; | |
43 | ||
44 | /* Close non-accepted connections */ | |
45 | while ((sk = smc_accept_dequeue(parent, NULL))) | |
46 | smc_close_non_accepted(sk); | |
47 | } | |
48 | ||
b38d7324 UB |
49 | /* wait for sndbuf data being transmitted */ |
50 | static void smc_close_stream_wait(struct smc_sock *smc, long timeout) | |
51 | { | |
52 | DEFINE_WAIT_FUNC(wait, woken_wake_function); | |
53 | struct sock *sk = &smc->sk; | |
54 | ||
55 | if (!timeout) | |
56 | return; | |
57 | ||
58 | if (!smc_tx_prepared_sends(&smc->conn)) | |
59 | return; | |
60 | ||
61 | smc->wait_close_tx_prepared = 1; | |
62 | add_wait_queue(sk_sleep(sk), &wait); | |
63 | while (!signal_pending(current) && timeout) { | |
64 | int rc; | |
65 | ||
66 | rc = sk_wait_event(sk, &timeout, | |
67 | !smc_tx_prepared_sends(&smc->conn) || | |
d18963cf | 68 | sk->sk_err == ECONNABORTED || |
b2900980 UB |
69 | sk->sk_err == ECONNRESET || |
70 | smc->conn.killed, | |
b38d7324 UB |
71 | &wait); |
72 | if (rc) | |
73 | break; | |
74 | } | |
75 | remove_wait_queue(sk_sleep(sk), &wait); | |
76 | smc->wait_close_tx_prepared = 0; | |
77 | } | |
78 | ||
79 | void smc_close_wake_tx_prepared(struct smc_sock *smc) | |
80 | { | |
81 | if (smc->wait_close_tx_prepared) | |
82 | /* wake up socket closing */ | |
83 | smc->sk.sk_state_change(&smc->sk); | |
84 | } | |
85 | ||
86 | static int smc_close_wr(struct smc_connection *conn) | |
87 | { | |
88 | conn->local_tx_ctrl.conn_state_flags.peer_done_writing = 1; | |
89 | ||
90 | return smc_cdc_get_slot_and_msg_send(conn); | |
91 | } | |
92 | ||
93 | static int smc_close_final(struct smc_connection *conn) | |
94 | { | |
95 | if (atomic_read(&conn->bytes_to_rcv)) | |
96 | conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1; | |
97 | else | |
98 | conn->local_tx_ctrl.conn_state_flags.peer_conn_closed = 1; | |
b2900980 UB |
99 | if (conn->killed) |
100 | return -EPIPE; | |
b38d7324 UB |
101 | |
102 | return smc_cdc_get_slot_and_msg_send(conn); | |
103 | } | |
104 | ||
105 | static int smc_close_abort(struct smc_connection *conn) | |
106 | { | |
107 | conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1; | |
108 | ||
109 | return smc_cdc_get_slot_and_msg_send(conn); | |
110 | } | |
111 | ||
112 | /* terminate smc socket abnormally - active abort | |
732720fa | 113 | * link group is terminated, i.e. RDMA communication no longer possible |
b38d7324 | 114 | */ |
a8ae890b | 115 | static void smc_close_active_abort(struct smc_sock *smc) |
b38d7324 | 116 | { |
3163c507 UB |
117 | struct sock *sk = &smc->sk; |
118 | ||
dd65d87a UB |
119 | if (sk->sk_state != SMC_INIT && smc->clcsock && smc->clcsock->sk) { |
120 | sk->sk_err = ECONNABORTED; | |
121 | if (smc->clcsock && smc->clcsock->sk) { | |
122 | smc->clcsock->sk->sk_err = ECONNABORTED; | |
123 | smc->clcsock->sk->sk_state_change(smc->clcsock->sk); | |
124 | } | |
b38d7324 | 125 | } |
3163c507 | 126 | switch (sk->sk_state) { |
46c28dbd | 127 | case SMC_ACTIVE: |
3163c507 | 128 | sk->sk_state = SMC_PEERABORTWAIT; |
611b63a1 UB |
129 | release_sock(sk); |
130 | cancel_delayed_work_sync(&smc->conn.tx_work); | |
131 | lock_sock(sk); | |
d18963cf | 132 | sk->sk_state = SMC_CLOSED; |
51f1de79 | 133 | sock_put(sk); /* passive closing */ |
b38d7324 UB |
134 | break; |
135 | case SMC_APPCLOSEWAIT1: | |
136 | case SMC_APPCLOSEWAIT2: | |
611b63a1 UB |
137 | release_sock(sk); |
138 | cancel_delayed_work_sync(&smc->conn.tx_work); | |
139 | lock_sock(sk); | |
d18963cf | 140 | sk->sk_state = SMC_CLOSED; |
b38d7324 UB |
141 | break; |
142 | case SMC_PEERCLOSEWAIT1: | |
143 | case SMC_PEERCLOSEWAIT2: | |
d18963cf UB |
144 | case SMC_PEERFINCLOSEWAIT: |
145 | sk->sk_state = SMC_CLOSED; | |
51f1de79 | 146 | sock_put(sk); /* passive closing */ |
b38d7324 UB |
147 | break; |
148 | case SMC_PROCESSABORT: | |
149 | case SMC_APPFINCLOSEWAIT: | |
3163c507 | 150 | sk->sk_state = SMC_CLOSED; |
b38d7324 | 151 | break; |
dd65d87a | 152 | case SMC_INIT: |
b38d7324 UB |
153 | case SMC_PEERABORTWAIT: |
154 | case SMC_CLOSED: | |
155 | break; | |
156 | } | |
157 | ||
3163c507 UB |
158 | sock_set_flag(sk, SOCK_DEAD); |
159 | sk->sk_state_change(sk); | |
b38d7324 UB |
160 | } |
161 | ||
a98bf8c0 UB |
162 | static inline bool smc_close_sent_any_close(struct smc_connection *conn) |
163 | { | |
164 | return conn->local_tx_ctrl.conn_state_flags.peer_conn_abort || | |
165 | conn->local_tx_ctrl.conn_state_flags.peer_conn_closed; | |
166 | } | |
167 | ||
b38d7324 UB |
168 | int smc_close_active(struct smc_sock *smc) |
169 | { | |
170 | struct smc_cdc_conn_state_flags *txflags = | |
171 | &smc->conn.local_tx_ctrl.conn_state_flags; | |
b38d7324 UB |
172 | struct smc_connection *conn = &smc->conn; |
173 | struct sock *sk = &smc->sk; | |
174 | int old_state; | |
8c96feee | 175 | long timeout; |
b38d7324 UB |
176 | int rc = 0; |
177 | ||
8c96feee UB |
178 | timeout = current->flags & PF_EXITING ? |
179 | 0 : sock_flag(sk, SOCK_LINGER) ? | |
180 | sk->sk_lingertime : SMC_MAX_STREAM_WAIT_TIMEOUT; | |
b38d7324 | 181 | |
b38d7324 | 182 | old_state = sk->sk_state; |
bbb96bf2 UB |
183 | again: |
184 | switch (sk->sk_state) { | |
b38d7324 UB |
185 | case SMC_INIT: |
186 | sk->sk_state = SMC_CLOSED; | |
b38d7324 UB |
187 | break; |
188 | case SMC_LISTEN: | |
189 | sk->sk_state = SMC_CLOSED; | |
190 | sk->sk_state_change(sk); /* wake up accept */ | |
191 | if (smc->clcsock && smc->clcsock->sk) { | |
192 | rc = kernel_sock_shutdown(smc->clcsock, SHUT_RDWR); | |
193 | /* wake up kernel_accept of smc_tcp_listen_worker */ | |
194 | smc->clcsock->sk->sk_data_ready(smc->clcsock->sk); | |
195 | } | |
b38d7324 | 196 | smc_close_cleanup_listen(sk); |
3d502067 UB |
197 | release_sock(sk); |
198 | flush_work(&smc->tcp_listen_work); | |
199 | lock_sock(sk); | |
b38d7324 UB |
200 | break; |
201 | case SMC_ACTIVE: | |
202 | smc_close_stream_wait(smc, timeout); | |
203 | release_sock(sk); | |
18e537cd | 204 | cancel_delayed_work_sync(&conn->tx_work); |
b38d7324 UB |
205 | lock_sock(sk); |
206 | if (sk->sk_state == SMC_ACTIVE) { | |
207 | /* send close request */ | |
208 | rc = smc_close_final(conn); | |
209 | sk->sk_state = SMC_PEERCLOSEWAIT1; | |
210 | } else { | |
211 | /* peer event has changed the state */ | |
212 | goto again; | |
213 | } | |
214 | break; | |
215 | case SMC_APPFINCLOSEWAIT: | |
216 | /* socket already shutdown wr or both (active close) */ | |
217 | if (txflags->peer_done_writing && | |
a98bf8c0 | 218 | !smc_close_sent_any_close(conn)) { |
b38d7324 UB |
219 | /* just shutdown wr done, send close request */ |
220 | rc = smc_close_final(conn); | |
221 | } | |
222 | sk->sk_state = SMC_CLOSED; | |
b38d7324 UB |
223 | break; |
224 | case SMC_APPCLOSEWAIT1: | |
225 | case SMC_APPCLOSEWAIT2: | |
226 | if (!smc_cdc_rxed_any_close(conn)) | |
227 | smc_close_stream_wait(smc, timeout); | |
228 | release_sock(sk); | |
18e537cd | 229 | cancel_delayed_work_sync(&conn->tx_work); |
b38d7324 | 230 | lock_sock(sk); |
bbb96bf2 UB |
231 | if (sk->sk_state != SMC_APPCLOSEWAIT1 && |
232 | sk->sk_state != SMC_APPCLOSEWAIT2) | |
233 | goto again; | |
234 | /* confirm close from peer */ | |
235 | rc = smc_close_final(conn); | |
51f1de79 | 236 | if (smc_cdc_rxed_any_close(conn)) { |
b38d7324 UB |
237 | /* peer has closed the socket already */ |
238 | sk->sk_state = SMC_CLOSED; | |
51f1de79 UB |
239 | sock_put(sk); /* postponed passive closing */ |
240 | } else { | |
b38d7324 UB |
241 | /* peer has just issued a shutdown write */ |
242 | sk->sk_state = SMC_PEERFINCLOSEWAIT; | |
51f1de79 | 243 | } |
b38d7324 UB |
244 | break; |
245 | case SMC_PEERCLOSEWAIT1: | |
246 | case SMC_PEERCLOSEWAIT2: | |
a98bf8c0 UB |
247 | if (txflags->peer_done_writing && |
248 | !smc_close_sent_any_close(conn)) { | |
249 | /* just shutdown wr done, send close request */ | |
250 | rc = smc_close_final(conn); | |
251 | } | |
252 | /* peer sending PeerConnectionClosed will cause transition */ | |
253 | break; | |
b38d7324 UB |
254 | case SMC_PEERFINCLOSEWAIT: |
255 | /* peer sending PeerConnectionClosed will cause transition */ | |
256 | break; | |
257 | case SMC_PROCESSABORT: | |
d18963cf | 258 | rc = smc_close_abort(conn); |
b38d7324 | 259 | sk->sk_state = SMC_CLOSED; |
b38d7324 UB |
260 | break; |
261 | case SMC_PEERABORTWAIT: | |
d18963cf UB |
262 | sk->sk_state = SMC_CLOSED; |
263 | break; | |
b38d7324 UB |
264 | case SMC_CLOSED: |
265 | /* nothing to do, add tracing in future patch */ | |
266 | break; | |
267 | } | |
268 | ||
269 | if (old_state != sk->sk_state) | |
3163c507 | 270 | sk->sk_state_change(sk); |
b38d7324 UB |
271 | return rc; |
272 | } | |
273 | ||
274 | static void smc_close_passive_abort_received(struct smc_sock *smc) | |
275 | { | |
276 | struct smc_cdc_conn_state_flags *txflags = | |
277 | &smc->conn.local_tx_ctrl.conn_state_flags; | |
278 | struct sock *sk = &smc->sk; | |
279 | ||
280 | switch (sk->sk_state) { | |
51f1de79 | 281 | case SMC_INIT: |
b38d7324 | 282 | case SMC_ACTIVE: |
b38d7324 | 283 | case SMC_APPCLOSEWAIT1: |
51f1de79 UB |
284 | sk->sk_state = SMC_PROCESSABORT; |
285 | sock_put(sk); /* passive closing */ | |
286 | break; | |
287 | case SMC_APPFINCLOSEWAIT: | |
b38d7324 UB |
288 | sk->sk_state = SMC_PROCESSABORT; |
289 | break; | |
290 | case SMC_PEERCLOSEWAIT1: | |
291 | case SMC_PEERCLOSEWAIT2: | |
292 | if (txflags->peer_done_writing && | |
51f1de79 | 293 | !smc_close_sent_any_close(&smc->conn)) |
b38d7324 | 294 | /* just shutdown, but not yet closed locally */ |
b38d7324 | 295 | sk->sk_state = SMC_PROCESSABORT; |
51f1de79 | 296 | else |
b38d7324 | 297 | sk->sk_state = SMC_CLOSED; |
51f1de79 | 298 | sock_put(sk); /* passive closing */ |
b38d7324 | 299 | break; |
51f1de79 | 300 | case SMC_APPCLOSEWAIT2: |
b38d7324 | 301 | case SMC_PEERFINCLOSEWAIT: |
51f1de79 UB |
302 | sk->sk_state = SMC_CLOSED; |
303 | sock_put(sk); /* passive closing */ | |
304 | break; | |
b38d7324 UB |
305 | case SMC_PEERABORTWAIT: |
306 | sk->sk_state = SMC_CLOSED; | |
307 | break; | |
b38d7324 UB |
308 | case SMC_PROCESSABORT: |
309 | /* nothing to do, add tracing in future patch */ | |
310 | break; | |
311 | } | |
312 | } | |
313 | ||
732720fa UB |
314 | /* Either some kind of closing has been received: peer_conn_closed, |
315 | * peer_conn_abort, or peer_done_writing | |
316 | * or the link group of the connection terminates abnormally. | |
b38d7324 | 317 | */ |
46c28dbd | 318 | static void smc_close_passive_work(struct work_struct *work) |
b38d7324 | 319 | { |
46c28dbd UB |
320 | struct smc_connection *conn = container_of(work, |
321 | struct smc_connection, | |
322 | close_work); | |
323 | struct smc_sock *smc = container_of(conn, struct smc_sock, conn); | |
324 | struct smc_cdc_conn_state_flags *rxflags; | |
fd57770d | 325 | bool release_clcsock = false; |
b38d7324 UB |
326 | struct sock *sk = &smc->sk; |
327 | int old_state; | |
328 | ||
3163c507 | 329 | lock_sock(sk); |
b38d7324 UB |
330 | old_state = sk->sk_state; |
331 | ||
b2900980 | 332 | if (conn->killed) { |
46c28dbd UB |
333 | /* abnormal termination */ |
334 | smc_close_active_abort(smc); | |
335 | goto wakeup; | |
336 | } | |
337 | ||
3163c507 | 338 | rxflags = &conn->local_rx_ctrl.conn_state_flags; |
b38d7324 | 339 | if (rxflags->peer_conn_abort) { |
732720fa | 340 | /* peer has not received all data */ |
b38d7324 | 341 | smc_close_passive_abort_received(smc); |
611b63a1 UB |
342 | release_sock(&smc->sk); |
343 | cancel_delayed_work_sync(&conn->tx_work); | |
344 | lock_sock(&smc->sk); | |
b38d7324 UB |
345 | goto wakeup; |
346 | } | |
347 | ||
348 | switch (sk->sk_state) { | |
349 | case SMC_INIT: | |
84b799a2 | 350 | sk->sk_state = SMC_APPCLOSEWAIT1; |
b38d7324 UB |
351 | break; |
352 | case SMC_ACTIVE: | |
353 | sk->sk_state = SMC_APPCLOSEWAIT1; | |
51f1de79 UB |
354 | /* postpone sock_put() for passive closing to cover |
355 | * received SEND_SHUTDOWN as well | |
356 | */ | |
b38d7324 UB |
357 | break; |
358 | case SMC_PEERCLOSEWAIT1: | |
359 | if (rxflags->peer_done_writing) | |
360 | sk->sk_state = SMC_PEERCLOSEWAIT2; | |
7f6b437e GS |
361 | /* fall through */ |
362 | /* to check for closing */ | |
b38d7324 | 363 | case SMC_PEERCLOSEWAIT2: |
3163c507 | 364 | if (!smc_cdc_rxed_any_close(conn)) |
b38d7324 UB |
365 | break; |
366 | if (sock_flag(sk, SOCK_DEAD) && | |
a98bf8c0 | 367 | smc_close_sent_any_close(conn)) { |
b38d7324 UB |
368 | /* smc_release has already been called locally */ |
369 | sk->sk_state = SMC_CLOSED; | |
370 | } else { | |
371 | /* just shutdown, but not yet closed locally */ | |
372 | sk->sk_state = SMC_APPFINCLOSEWAIT; | |
373 | } | |
51f1de79 | 374 | sock_put(sk); /* passive closing */ |
b38d7324 | 375 | break; |
5ac92a00 | 376 | case SMC_PEERFINCLOSEWAIT: |
51f1de79 | 377 | if (smc_cdc_rxed_any_close(conn)) { |
5ac92a00 | 378 | sk->sk_state = SMC_CLOSED; |
51f1de79 UB |
379 | sock_put(sk); /* passive closing */ |
380 | } | |
5ac92a00 | 381 | break; |
b38d7324 UB |
382 | case SMC_APPCLOSEWAIT1: |
383 | case SMC_APPCLOSEWAIT2: | |
51f1de79 UB |
384 | /* postpone sock_put() for passive closing to cover |
385 | * received SEND_SHUTDOWN as well | |
386 | */ | |
387 | break; | |
b38d7324 UB |
388 | case SMC_APPFINCLOSEWAIT: |
389 | case SMC_PEERABORTWAIT: | |
390 | case SMC_PROCESSABORT: | |
391 | case SMC_CLOSED: | |
392 | /* nothing to do, add tracing in future patch */ | |
393 | break; | |
394 | } | |
395 | ||
396 | wakeup: | |
b38d7324 UB |
397 | sk->sk_data_ready(sk); /* wakeup blocked rcvbuf consumers */ |
398 | sk->sk_write_space(sk); /* wakeup blocked sndbuf producers */ | |
399 | ||
a98bf8c0 UB |
400 | if (old_state != sk->sk_state) { |
401 | sk->sk_state_change(sk); | |
402 | if ((sk->sk_state == SMC_CLOSED) && | |
b03faa1f | 403 | (sock_flag(sk, SOCK_DEAD) || !sk->sk_socket)) { |
3163c507 | 404 | smc_conn_free(conn); |
fd57770d KG |
405 | if (smc->clcsock) |
406 | release_clcsock = true; | |
b03faa1f | 407 | } |
b38d7324 | 408 | } |
3163c507 | 409 | release_sock(sk); |
fd57770d KG |
410 | if (release_clcsock) |
411 | smc_clcsock_release(smc); | |
51f1de79 | 412 | sock_put(sk); /* sock_hold done by schedulers of close_work */ |
b38d7324 UB |
413 | } |
414 | ||
415 | int smc_close_shutdown_write(struct smc_sock *smc) | |
416 | { | |
417 | struct smc_connection *conn = &smc->conn; | |
b38d7324 UB |
418 | struct sock *sk = &smc->sk; |
419 | int old_state; | |
8c96feee | 420 | long timeout; |
b38d7324 UB |
421 | int rc = 0; |
422 | ||
8c96feee UB |
423 | timeout = current->flags & PF_EXITING ? |
424 | 0 : sock_flag(sk, SOCK_LINGER) ? | |
425 | sk->sk_lingertime : SMC_MAX_STREAM_WAIT_TIMEOUT; | |
b38d7324 | 426 | |
b38d7324 | 427 | old_state = sk->sk_state; |
bbb96bf2 UB |
428 | again: |
429 | switch (sk->sk_state) { | |
b38d7324 UB |
430 | case SMC_ACTIVE: |
431 | smc_close_stream_wait(smc, timeout); | |
432 | release_sock(sk); | |
18e537cd | 433 | cancel_delayed_work_sync(&conn->tx_work); |
b38d7324 | 434 | lock_sock(sk); |
bbb96bf2 UB |
435 | if (sk->sk_state != SMC_ACTIVE) |
436 | goto again; | |
b38d7324 UB |
437 | /* send close wr request */ |
438 | rc = smc_close_wr(conn); | |
bbb96bf2 | 439 | sk->sk_state = SMC_PEERCLOSEWAIT1; |
b38d7324 UB |
440 | break; |
441 | case SMC_APPCLOSEWAIT1: | |
442 | /* passive close */ | |
443 | if (!smc_cdc_rxed_any_close(conn)) | |
444 | smc_close_stream_wait(smc, timeout); | |
445 | release_sock(sk); | |
18e537cd | 446 | cancel_delayed_work_sync(&conn->tx_work); |
b38d7324 | 447 | lock_sock(sk); |
bbb96bf2 UB |
448 | if (sk->sk_state != SMC_APPCLOSEWAIT1) |
449 | goto again; | |
b38d7324 UB |
450 | /* confirm close from peer */ |
451 | rc = smc_close_wr(conn); | |
452 | sk->sk_state = SMC_APPCLOSEWAIT2; | |
453 | break; | |
454 | case SMC_APPCLOSEWAIT2: | |
455 | case SMC_PEERFINCLOSEWAIT: | |
456 | case SMC_PEERCLOSEWAIT1: | |
457 | case SMC_PEERCLOSEWAIT2: | |
458 | case SMC_APPFINCLOSEWAIT: | |
459 | case SMC_PROCESSABORT: | |
460 | case SMC_PEERABORTWAIT: | |
461 | /* nothing to do, add tracing in future patch */ | |
462 | break; | |
463 | } | |
464 | ||
465 | if (old_state != sk->sk_state) | |
3163c507 | 466 | sk->sk_state_change(sk); |
b38d7324 UB |
467 | return rc; |
468 | } | |
46c28dbd UB |
469 | |
470 | /* Initialize close properties on connection establishment. */ | |
471 | void smc_close_init(struct smc_sock *smc) | |
472 | { | |
473 | INIT_WORK(&smc->conn.close_work, smc_close_passive_work); | |
474 | } |