xref: /linux/net/smc/smc_rx.c (revision 3f0a50f345f78183f6e9b39c2f45ca5dcaa511ca)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Shared Memory Communications over RDMA (SMC-R) and RoCE
4  *
5  * Manage RMBE
6  * copy new RMBE data into user space
7  *
8  * Copyright IBM Corp. 2016
9  *
10  * Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
11  */
12 
13 #include <linux/net.h>
14 #include <linux/rcupdate.h>
15 #include <linux/sched/signal.h>
16 
17 #include <net/sock.h>
18 
19 #include "smc.h"
20 #include "smc_core.h"
21 #include "smc_cdc.h"
22 #include "smc_tx.h" /* smc_tx_consumer_update() */
23 #include "smc_rx.h"
24 #include "smc_stats.h"
25 #include "smc_tracepoint.h"
26 
27 /* callback implementation to wakeup consumers blocked with smc_rx_wait().
28  * indirectly called by smc_cdc_msg_recv_action().
29  */
30 static void smc_rx_wake_up(struct sock *sk)
31 {
32 	struct socket_wq *wq;
33 
34 	/* derived from sock_def_readable() */
35 	/* called already in smc_listen_work() */
36 	rcu_read_lock();
37 	wq = rcu_dereference(sk->sk_wq);
38 	if (skwq_has_sleeper(wq))
39 		wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
40 						EPOLLRDNORM | EPOLLRDBAND);
41 	sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
42 	if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
43 	    (sk->sk_state == SMC_CLOSED))
44 		sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_HUP);
45 	rcu_read_unlock();
46 }
47 
48 /* Update consumer cursor
49  *   @conn   connection to update
50  *   @cons   consumer cursor
51  *   @len    number of Bytes consumed
52  *   Returns:
53  *   1 if we should end our receive, 0 otherwise
54  */
55 static int smc_rx_update_consumer(struct smc_sock *smc,
56 				  union smc_host_cursor cons, size_t len)
57 {
58 	struct smc_connection *conn = &smc->conn;
59 	struct sock *sk = &smc->sk;
60 	bool force = false;
61 	int diff, rc = 0;
62 
63 	smc_curs_add(conn->rmb_desc->len, &cons, len);
64 
65 	/* did we process urgent data? */
66 	if (conn->urg_state == SMC_URG_VALID || conn->urg_rx_skip_pend) {
67 		diff = smc_curs_comp(conn->rmb_desc->len, &cons,
68 				     &conn->urg_curs);
69 		if (sock_flag(sk, SOCK_URGINLINE)) {
70 			if (diff == 0) {
71 				force = true;
72 				rc = 1;
73 				conn->urg_state = SMC_URG_READ;
74 			}
75 		} else {
76 			if (diff == 1) {
77 				/* skip urgent byte */
78 				force = true;
79 				smc_curs_add(conn->rmb_desc->len, &cons, 1);
80 				conn->urg_rx_skip_pend = false;
81 			} else if (diff < -1)
82 				/* we read past urgent byte */
83 				conn->urg_state = SMC_URG_READ;
84 		}
85 	}
86 
87 	smc_curs_copy(&conn->local_tx_ctrl.cons, &cons, conn);
88 
89 	/* send consumer cursor update if required */
90 	/* similar to advertising new TCP rcv_wnd if required */
91 	smc_tx_consumer_update(conn, force);
92 
93 	return rc;
94 }
95 
96 static void smc_rx_update_cons(struct smc_sock *smc, size_t len)
97 {
98 	struct smc_connection *conn = &smc->conn;
99 	union smc_host_cursor cons;
100 
101 	smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn);
102 	smc_rx_update_consumer(smc, cons, len);
103 }
104 
105 struct smc_spd_priv {
106 	struct smc_sock *smc;
107 	size_t		 len;
108 };
109 
110 static void smc_rx_pipe_buf_release(struct pipe_inode_info *pipe,
111 				    struct pipe_buffer *buf)
112 {
113 	struct smc_spd_priv *priv = (struct smc_spd_priv *)buf->private;
114 	struct smc_sock *smc = priv->smc;
115 	struct smc_connection *conn;
116 	struct sock *sk = &smc->sk;
117 
118 	if (sk->sk_state == SMC_CLOSED ||
119 	    sk->sk_state == SMC_PEERFINCLOSEWAIT ||
120 	    sk->sk_state == SMC_APPFINCLOSEWAIT)
121 		goto out;
122 	conn = &smc->conn;
123 	lock_sock(sk);
124 	smc_rx_update_cons(smc, priv->len);
125 	release_sock(sk);
126 	if (atomic_sub_and_test(priv->len, &conn->splice_pending))
127 		smc_rx_wake_up(sk);
128 out:
129 	kfree(priv);
130 	put_page(buf->page);
131 	sock_put(sk);
132 }
133 
134 static const struct pipe_buf_operations smc_pipe_ops = {
135 	.release = smc_rx_pipe_buf_release,
136 	.get = generic_pipe_buf_get
137 };
138 
139 static void smc_rx_spd_release(struct splice_pipe_desc *spd,
140 			       unsigned int i)
141 {
142 	put_page(spd->pages[i]);
143 }
144 
145 static int smc_rx_splice(struct pipe_inode_info *pipe, char *src, size_t len,
146 			 struct smc_sock *smc)
147 {
148 	struct splice_pipe_desc spd;
149 	struct partial_page partial;
150 	struct smc_spd_priv *priv;
151 	int bytes;
152 
153 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
154 	if (!priv)
155 		return -ENOMEM;
156 	priv->len = len;
157 	priv->smc = smc;
158 	partial.offset = src - (char *)smc->conn.rmb_desc->cpu_addr;
159 	partial.len = len;
160 	partial.private = (unsigned long)priv;
161 
162 	spd.nr_pages_max = 1;
163 	spd.nr_pages = 1;
164 	spd.pages = &smc->conn.rmb_desc->pages;
165 	spd.partial = &partial;
166 	spd.ops = &smc_pipe_ops;
167 	spd.spd_release = smc_rx_spd_release;
168 
169 	bytes = splice_to_pipe(pipe, &spd);
170 	if (bytes > 0) {
171 		sock_hold(&smc->sk);
172 		get_page(smc->conn.rmb_desc->pages);
173 		atomic_add(bytes, &smc->conn.splice_pending);
174 	}
175 
176 	return bytes;
177 }
178 
179 static int smc_rx_data_available_and_no_splice_pend(struct smc_connection *conn)
180 {
181 	return atomic_read(&conn->bytes_to_rcv) &&
182 	       !atomic_read(&conn->splice_pending);
183 }
184 
185 /* blocks rcvbuf consumer until >=len bytes available or timeout or interrupted
186  *   @smc    smc socket
187  *   @timeo  pointer to max seconds to wait, pointer to value 0 for no timeout
188  *   @fcrit  add'l criterion to evaluate as function pointer
189  * Returns:
190  * 1 if at least 1 byte available in rcvbuf or if socket error/shutdown.
191  * 0 otherwise (nothing in rcvbuf nor timeout, e.g. interrupted).
192  */
193 int smc_rx_wait(struct smc_sock *smc, long *timeo,
194 		int (*fcrit)(struct smc_connection *conn))
195 {
196 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
197 	struct smc_connection *conn = &smc->conn;
198 	struct smc_cdc_conn_state_flags *cflags =
199 					&conn->local_tx_ctrl.conn_state_flags;
200 	struct sock *sk = &smc->sk;
201 	int rc;
202 
203 	if (fcrit(conn))
204 		return 1;
205 	sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
206 	add_wait_queue(sk_sleep(sk), &wait);
207 	rc = sk_wait_event(sk, timeo,
208 			   sk->sk_err ||
209 			   cflags->peer_conn_abort ||
210 			   sk->sk_shutdown & RCV_SHUTDOWN ||
211 			   conn->killed ||
212 			   fcrit(conn),
213 			   &wait);
214 	remove_wait_queue(sk_sleep(sk), &wait);
215 	sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
216 	return rc;
217 }
218 
219 static int smc_rx_recv_urg(struct smc_sock *smc, struct msghdr *msg, int len,
220 			   int flags)
221 {
222 	struct smc_connection *conn = &smc->conn;
223 	union smc_host_cursor cons;
224 	struct sock *sk = &smc->sk;
225 	int rc = 0;
226 
227 	if (sock_flag(sk, SOCK_URGINLINE) ||
228 	    !(conn->urg_state == SMC_URG_VALID) ||
229 	    conn->urg_state == SMC_URG_READ)
230 		return -EINVAL;
231 
232 	SMC_STAT_INC(smc, urg_data_cnt);
233 	if (conn->urg_state == SMC_URG_VALID) {
234 		if (!(flags & MSG_PEEK))
235 			smc->conn.urg_state = SMC_URG_READ;
236 		msg->msg_flags |= MSG_OOB;
237 		if (len > 0) {
238 			if (!(flags & MSG_TRUNC))
239 				rc = memcpy_to_msg(msg, &conn->urg_rx_byte, 1);
240 			len = 1;
241 			smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn);
242 			if (smc_curs_diff(conn->rmb_desc->len, &cons,
243 					  &conn->urg_curs) > 1)
244 				conn->urg_rx_skip_pend = true;
245 			/* Urgent Byte was already accounted for, but trigger
246 			 * skipping the urgent byte in non-inline case
247 			 */
248 			if (!(flags & MSG_PEEK))
249 				smc_rx_update_consumer(smc, cons, 0);
250 		} else {
251 			msg->msg_flags |= MSG_TRUNC;
252 		}
253 
254 		return rc ? -EFAULT : len;
255 	}
256 
257 	if (sk->sk_state == SMC_CLOSED || sk->sk_shutdown & RCV_SHUTDOWN)
258 		return 0;
259 
260 	return -EAGAIN;
261 }
262 
263 static bool smc_rx_recvmsg_data_available(struct smc_sock *smc)
264 {
265 	struct smc_connection *conn = &smc->conn;
266 
267 	if (smc_rx_data_available(conn))
268 		return true;
269 	else if (conn->urg_state == SMC_URG_VALID)
270 		/* we received a single urgent Byte - skip */
271 		smc_rx_update_cons(smc, 0);
272 	return false;
273 }
274 
275 /* smc_rx_recvmsg - receive data from RMBE
276  * @msg:	copy data to receive buffer
277  * @pipe:	copy data to pipe if set - indicates splice() call
278  *
279  * rcvbuf consumer: main API called by socket layer.
280  * Called under sk lock.
281  */
282 int smc_rx_recvmsg(struct smc_sock *smc, struct msghdr *msg,
283 		   struct pipe_inode_info *pipe, size_t len, int flags)
284 {
285 	size_t copylen, read_done = 0, read_remaining = len;
286 	size_t chunk_len, chunk_off, chunk_len_sum;
287 	struct smc_connection *conn = &smc->conn;
288 	int (*func)(struct smc_connection *conn);
289 	union smc_host_cursor cons;
290 	int readable, chunk;
291 	char *rcvbuf_base;
292 	struct sock *sk;
293 	int splbytes;
294 	long timeo;
295 	int target;		/* Read at least these many bytes */
296 	int rc;
297 
298 	if (unlikely(flags & MSG_ERRQUEUE))
299 		return -EINVAL; /* future work for sk.sk_family == AF_SMC */
300 
301 	sk = &smc->sk;
302 	if (sk->sk_state == SMC_LISTEN)
303 		return -ENOTCONN;
304 	if (flags & MSG_OOB)
305 		return smc_rx_recv_urg(smc, msg, len, flags);
306 	timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
307 	target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
308 
309 	readable = atomic_read(&conn->bytes_to_rcv);
310 	if (readable >= conn->rmb_desc->len)
311 		SMC_STAT_RMB_RX_FULL(smc, !conn->lnk);
312 
313 	if (len < readable)
314 		SMC_STAT_RMB_RX_SIZE_SMALL(smc, !conn->lnk);
315 	/* we currently use 1 RMBE per RMB, so RMBE == RMB base addr */
316 	rcvbuf_base = conn->rx_off + conn->rmb_desc->cpu_addr;
317 
318 	do { /* while (read_remaining) */
319 		if (read_done >= target || (pipe && read_done))
320 			break;
321 
322 		if (conn->killed)
323 			break;
324 
325 		if (smc_rx_recvmsg_data_available(smc))
326 			goto copy;
327 
328 		if (sk->sk_shutdown & RCV_SHUTDOWN) {
329 			/* smc_cdc_msg_recv_action() could have run after
330 			 * above smc_rx_recvmsg_data_available()
331 			 */
332 			if (smc_rx_recvmsg_data_available(smc))
333 				goto copy;
334 			break;
335 		}
336 
337 		if (read_done) {
338 			if (sk->sk_err ||
339 			    sk->sk_state == SMC_CLOSED ||
340 			    !timeo ||
341 			    signal_pending(current))
342 				break;
343 		} else {
344 			if (sk->sk_err) {
345 				read_done = sock_error(sk);
346 				break;
347 			}
348 			if (sk->sk_state == SMC_CLOSED) {
349 				if (!sock_flag(sk, SOCK_DONE)) {
350 					/* This occurs when user tries to read
351 					 * from never connected socket.
352 					 */
353 					read_done = -ENOTCONN;
354 					break;
355 				}
356 				break;
357 			}
358 			if (signal_pending(current)) {
359 				read_done = sock_intr_errno(timeo);
360 				break;
361 			}
362 			if (!timeo)
363 				return -EAGAIN;
364 		}
365 
366 		if (!smc_rx_data_available(conn)) {
367 			smc_rx_wait(smc, &timeo, smc_rx_data_available);
368 			continue;
369 		}
370 
371 copy:
372 		/* initialize variables for 1st iteration of subsequent loop */
373 		/* could be just 1 byte, even after waiting on data above */
374 		readable = atomic_read(&conn->bytes_to_rcv);
375 		splbytes = atomic_read(&conn->splice_pending);
376 		if (!readable || (msg && splbytes)) {
377 			if (splbytes)
378 				func = smc_rx_data_available_and_no_splice_pend;
379 			else
380 				func = smc_rx_data_available;
381 			smc_rx_wait(smc, &timeo, func);
382 			continue;
383 		}
384 
385 		smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn);
386 		/* subsequent splice() calls pick up where previous left */
387 		if (splbytes)
388 			smc_curs_add(conn->rmb_desc->len, &cons, splbytes);
389 		if (conn->urg_state == SMC_URG_VALID &&
390 		    sock_flag(&smc->sk, SOCK_URGINLINE) &&
391 		    readable > 1)
392 			readable--;	/* always stop at urgent Byte */
393 		/* not more than what user space asked for */
394 		copylen = min_t(size_t, read_remaining, readable);
395 		/* determine chunks where to read from rcvbuf */
396 		/* either unwrapped case, or 1st chunk of wrapped case */
397 		chunk_len = min_t(size_t, copylen, conn->rmb_desc->len -
398 				  cons.count);
399 		chunk_len_sum = chunk_len;
400 		chunk_off = cons.count;
401 		smc_rmb_sync_sg_for_cpu(conn);
402 		for (chunk = 0; chunk < 2; chunk++) {
403 			if (!(flags & MSG_TRUNC)) {
404 				if (msg) {
405 					rc = memcpy_to_msg(msg, rcvbuf_base +
406 							   chunk_off,
407 							   chunk_len);
408 				} else {
409 					rc = smc_rx_splice(pipe, rcvbuf_base +
410 							chunk_off, chunk_len,
411 							smc);
412 				}
413 				if (rc < 0) {
414 					if (!read_done)
415 						read_done = -EFAULT;
416 					smc_rmb_sync_sg_for_device(conn);
417 					goto out;
418 				}
419 			}
420 			read_remaining -= chunk_len;
421 			read_done += chunk_len;
422 
423 			if (chunk_len_sum == copylen)
424 				break; /* either on 1st or 2nd iteration */
425 			/* prepare next (== 2nd) iteration */
426 			chunk_len = copylen - chunk_len; /* remainder */
427 			chunk_len_sum += chunk_len;
428 			chunk_off = 0; /* modulo offset in recv ring buffer */
429 		}
430 		smc_rmb_sync_sg_for_device(conn);
431 
432 		/* update cursors */
433 		if (!(flags & MSG_PEEK)) {
434 			/* increased in recv tasklet smc_cdc_msg_rcv() */
435 			smp_mb__before_atomic();
436 			atomic_sub(copylen, &conn->bytes_to_rcv);
437 			/* guarantee 0 <= bytes_to_rcv <= rmb_desc->len */
438 			smp_mb__after_atomic();
439 			if (msg && smc_rx_update_consumer(smc, cons, copylen))
440 				goto out;
441 		}
442 
443 		trace_smc_rx_recvmsg(smc, copylen);
444 	} while (read_remaining);
445 out:
446 	return read_done;
447 }
448 
449 /* Initialize receive properties on connection establishment. NB: not __init! */
450 void smc_rx_init(struct smc_sock *smc)
451 {
452 	smc->sk.sk_data_ready = smc_rx_wake_up;
453 	atomic_set(&smc->conn.splice_pending, 0);
454 	smc->conn.urg_state = SMC_URG_READ;
455 }
456