connection.c 20.2 KB
Newer Older
A
Andy Grover 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34
/*
 * Copyright (c) 2006 Oracle.  All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     Redistribution and use in source and binary forms, with or
 *     without modification, are permitted provided that the following
 *     conditions are met:
 *
 *      - Redistributions of source code must retain the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer.
 *
 *      - Redistributions in binary form must reproduce the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer in the documentation and/or other materials
 *        provided with the distribution.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 *
 */
#include <linux/kernel.h>
#include <linux/list.h>
35
#include <linux/slab.h>
36
#include <linux/export.h>
A
Andy Grover 已提交
37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53
#include <net/inet_hashtables.h>

#include "rds.h"
#include "loop.h"

#define RDS_CONNECTION_HASH_BITS 12
#define RDS_CONNECTION_HASH_ENTRIES (1 << RDS_CONNECTION_HASH_BITS)
#define RDS_CONNECTION_HASH_MASK (RDS_CONNECTION_HASH_ENTRIES - 1)

/* converting this to RCU is a chore for another day.. */
static DEFINE_SPINLOCK(rds_conn_lock);
static unsigned long rds_conn_count;
static struct hlist_head rds_conn_hash[RDS_CONNECTION_HASH_ENTRIES];
static struct kmem_cache *rds_conn_slab;

static struct hlist_head *rds_conn_bucket(__be32 laddr, __be32 faddr)
{
54 55 56 57 58 59
	static u32 rds_hash_secret __read_mostly;

	unsigned long hash;

	net_get_random_once(&rds_hash_secret, sizeof(rds_hash_secret));

A
Andy Grover 已提交
60
	/* Pass NULL, don't need struct net for hash */
61 62 63
	hash = __inet_ehashfn(be32_to_cpu(laddr), 0,
			      be32_to_cpu(faddr), 0,
			      rds_hash_secret);
A
Andy Grover 已提交
64 65 66 67 68 69 70 71
	return &rds_conn_hash[hash & RDS_CONNECTION_HASH_MASK];
}

#define rds_conn_info_set(var, test, suffix) do {		\
	if (test)						\
		var |= RDS_INFO_CONNECTION_FLAG_##suffix;	\
} while (0)

72
/* rcu read lock must be held or the connection spinlock */
73 74
static struct rds_connection *rds_conn_lookup(struct net *net,
					      struct hlist_head *head,
A
Andy Grover 已提交
75 76 77 78 79
					      __be32 laddr, __be32 faddr,
					      struct rds_transport *trans)
{
	struct rds_connection *conn, *ret = NULL;

80
	hlist_for_each_entry_rcu(conn, head, c_hash_node) {
A
Andy Grover 已提交
81
		if (conn->c_faddr == faddr && conn->c_laddr == laddr &&
82
		    conn->c_trans == trans && net == rds_conn_net(conn)) {
A
Andy Grover 已提交
83 84 85 86 87 88 89 90 91 92 93 94 95 96 97
			ret = conn;
			break;
		}
	}
	rdsdebug("returning conn %p for %pI4 -> %pI4\n", ret,
		 &laddr, &faddr);
	return ret;
}

/*
 * This is called by transports as they're bringing down a connection.
 * It clears partial message state so that the transport can start sending
 * and receiving over this connection again in the future.  It is up to
 * the transport to have serialized this call with its send and recv.
 */
98
static void rds_conn_path_reset(struct rds_conn_path *cp)
A
Andy Grover 已提交
99
{
100 101
	struct rds_connection *conn = cp->cp_conn;

A
Andy Grover 已提交
102 103 104 105
	rdsdebug("connection %pI4 to %pI4 reset\n",
	  &conn->c_laddr, &conn->c_faddr);

	rds_stats_inc(s_conn_reset);
106 107
	rds_send_path_reset(cp);
	cp->cp_flags = 0;
A
Andy Grover 已提交
108 109 110 111 112 113 114

	/* Do not clear next_rx_seq here, else we cannot distinguish
	 * retransmitted packets from new packets, and will hand all
	 * of them to the application. That is not consistent with the
	 * reliability guarantees of RDS. */
}

115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135
static void __rds_conn_path_init(struct rds_connection *conn,
				 struct rds_conn_path *cp, bool is_outgoing)
{
	spin_lock_init(&cp->cp_lock);
	cp->cp_next_tx_seq = 1;
	init_waitqueue_head(&cp->cp_waitq);
	INIT_LIST_HEAD(&cp->cp_send_queue);
	INIT_LIST_HEAD(&cp->cp_retrans);

	cp->cp_conn = conn;
	atomic_set(&cp->cp_state, RDS_CONN_DOWN);
	cp->cp_send_gen = 0;
	cp->cp_reconnect_jiffies = 0;
	INIT_DELAYED_WORK(&cp->cp_send_w, rds_send_worker);
	INIT_DELAYED_WORK(&cp->cp_recv_w, rds_recv_worker);
	INIT_DELAYED_WORK(&cp->cp_conn_w, rds_connect_worker);
	INIT_WORK(&cp->cp_down_w, rds_shutdown_worker);
	mutex_init(&cp->cp_cm_lock);
	cp->cp_flags = 0;
}

A
Andy Grover 已提交
136 137 138 139 140 141 142 143
/*
 * There is only every one 'conn' for a given pair of addresses in the
 * system at a time.  They contain messages to be retransmitted and so
 * span the lifetime of the actual underlying transport connections.
 *
 * For now they are not garbage collected once they're created.  They
 * are torn down as the module is removed, if ever.
 */
144 145
static struct rds_connection *__rds_conn_create(struct net *net,
						__be32 laddr, __be32 faddr,
A
Andy Grover 已提交
146 147 148
				       struct rds_transport *trans, gfp_t gfp,
				       int is_outgoing)
{
149
	struct rds_connection *conn, *parent = NULL;
A
Andy Grover 已提交
150
	struct hlist_head *head = rds_conn_bucket(laddr, faddr);
151
	struct rds_transport *loop_trans;
A
Andy Grover 已提交
152
	unsigned long flags;
153
	int ret, i;
154
	int npaths = (trans->t_mp_capable ? RDS_MPATH_WORKERS : 1);
A
Andy Grover 已提交
155

156
	rcu_read_lock();
157
	conn = rds_conn_lookup(net, head, laddr, faddr, trans);
158
	if (conn && conn->c_loopback && conn->c_trans != &rds_loop_transport &&
159
	    laddr == faddr && !is_outgoing) {
A
Andy Grover 已提交
160 161 162 163 164 165 166
		/* This is a looped back IB connection, and we're
		 * called by the code handling the incoming connect.
		 * We need a second connection object into which we
		 * can stick the other QP. */
		parent = conn;
		conn = parent->c_passive;
	}
167
	rcu_read_unlock();
A
Andy Grover 已提交
168 169 170
	if (conn)
		goto out;

171
	conn = kmem_cache_zalloc(rds_conn_slab, gfp);
172
	if (!conn) {
A
Andy Grover 已提交
173 174 175
		conn = ERR_PTR(-ENOMEM);
		goto out;
	}
176 177 178 179 180 181
	conn->c_path = kcalloc(npaths, sizeof(struct rds_conn_path), gfp);
	if (!conn->c_path) {
		kmem_cache_free(rds_conn_slab, conn);
		conn = ERR_PTR(-ENOMEM);
		goto out;
	}
A
Andy Grover 已提交
182 183 184 185 186

	INIT_HLIST_NODE(&conn->c_hash_node);
	conn->c_laddr = laddr;
	conn->c_faddr = faddr;

187
	rds_conn_net_set(conn, net);
A
Andy Grover 已提交
188 189 190

	ret = rds_cong_get_maps(conn);
	if (ret) {
191
		kfree(conn->c_path);
A
Andy Grover 已提交
192 193 194 195 196 197 198 199 200 201
		kmem_cache_free(rds_conn_slab, conn);
		conn = ERR_PTR(ret);
		goto out;
	}

	/*
	 * This is where a connection becomes loopback.  If *any* RDS sockets
	 * can bind to the destination address then we'd rather the messages
	 * flow through loopback rather than either transport.
	 */
202
	loop_trans = rds_trans_get_preferred(net, faddr);
203 204
	if (loop_trans) {
		rds_trans_put(loop_trans);
A
Andy Grover 已提交
205 206 207 208 209 210 211 212 213 214 215 216
		conn->c_loopback = 1;
		if (is_outgoing && trans->t_prefer_loopback) {
			/* "outgoing" connection - and the transport
			 * says it wants the connection handled by the
			 * loopback transport. This is what TCP does.
			 */
			trans = &rds_loop_transport;
		}
	}

	conn->c_trans = trans;

217
	init_waitqueue_head(&conn->c_hs_waitq);
218
	for (i = 0; i < npaths; i++) {
219 220 221 222
		__rds_conn_path_init(conn, &conn->c_path[i],
				     is_outgoing);
		conn->c_path[i].cp_index = i;
	}
A
Andy Grover 已提交
223 224
	ret = trans->conn_alloc(conn, gfp);
	if (ret) {
225
		kfree(conn->c_path);
A
Andy Grover 已提交
226 227 228 229 230 231 232 233 234 235
		kmem_cache_free(rds_conn_slab, conn);
		conn = ERR_PTR(ret);
		goto out;
	}

	rdsdebug("allocated conn %p for %pI4 -> %pI4 over %s %s\n",
	  conn, &laddr, &faddr,
	  trans->t_name ? trans->t_name : "[unknown]",
	  is_outgoing ? "(outgoing)" : "");

236 237 238 239 240 241 242
	/*
	 * Since we ran without holding the conn lock, someone could
	 * have created the same conn (either normal or passive) in the
	 * interim. We check while holding the lock. If we won, we complete
	 * init and return our conn. If we lost, we rollback and return the
	 * other one.
	 */
A
Andy Grover 已提交
243
	spin_lock_irqsave(&rds_conn_lock, flags);
244 245 246
	if (parent) {
		/* Creating passive conn */
		if (parent->c_passive) {
247
			trans->conn_free(conn->c_path[0].cp_transport_data);
248
			kfree(conn->c_path);
249 250 251
			kmem_cache_free(rds_conn_slab, conn);
			conn = parent->c_passive;
		} else {
A
Andy Grover 已提交
252
			parent->c_passive = conn;
253 254 255
			rds_cong_add_conn(conn);
			rds_conn_count++;
		}
A
Andy Grover 已提交
256
	} else {
257 258 259
		/* Creating normal conn */
		struct rds_connection *found;

260
		found = rds_conn_lookup(net, head, laddr, faddr, trans);
261
		if (found) {
262 263 264
			struct rds_conn_path *cp;
			int i;

265
			for (i = 0; i < npaths; i++) {
266
				cp = &conn->c_path[i];
267 268 269 270 271 272
				/* The ->conn_alloc invocation may have
				 * allocated resource for all paths, so all
				 * of them may have to be freed here.
				 */
				if (cp->cp_transport_data)
					trans->conn_free(cp->cp_transport_data);
273
			}
274
			kfree(conn->c_path);
275 276 277
			kmem_cache_free(rds_conn_slab, conn);
			conn = found;
		} else {
278 279
			conn->c_my_gen_num = rds_gen_num;
			conn->c_peer_gen_num = 0;
280
			hlist_add_head_rcu(&conn->c_hash_node, head);
281 282 283
			rds_cong_add_conn(conn);
			rds_conn_count++;
		}
A
Andy Grover 已提交
284 285 286 287 288 289 290
	}
	spin_unlock_irqrestore(&rds_conn_lock, flags);

out:
	return conn;
}

291 292
struct rds_connection *rds_conn_create(struct net *net,
				       __be32 laddr, __be32 faddr,
A
Andy Grover 已提交
293 294
				       struct rds_transport *trans, gfp_t gfp)
{
295
	return __rds_conn_create(net, laddr, faddr, trans, gfp, 0);
A
Andy Grover 已提交
296
}
A
Andy Grover 已提交
297
EXPORT_SYMBOL_GPL(rds_conn_create);
A
Andy Grover 已提交
298

299 300
struct rds_connection *rds_conn_create_outgoing(struct net *net,
						__be32 laddr, __be32 faddr,
A
Andy Grover 已提交
301 302
				       struct rds_transport *trans, gfp_t gfp)
{
303
	return __rds_conn_create(net, laddr, faddr, trans, gfp, 1);
A
Andy Grover 已提交
304
}
A
Andy Grover 已提交
305
EXPORT_SYMBOL_GPL(rds_conn_create_outgoing);
A
Andy Grover 已提交
306

307
void rds_conn_shutdown(struct rds_conn_path *cp)
308
{
309 310
	struct rds_connection *conn = cp->cp_conn;

311
	/* shut it down unless it's down already */
312
	if (!rds_conn_path_transition(cp, RDS_CONN_DOWN, RDS_CONN_DOWN)) {
313 314 315 316 317 318 319
		/*
		 * Quiesce the connection mgmt handlers before we start tearing
		 * things down. We don't hold the mutex for the entire
		 * duration of the shutdown operation, else we may be
		 * deadlocking with the CM handler. Instead, the CM event
		 * handler is supposed to check for state DISCONNECTING
		 */
320 321 322 323 324 325 326 327 328
		mutex_lock(&cp->cp_cm_lock);
		if (!rds_conn_path_transition(cp, RDS_CONN_UP,
					      RDS_CONN_DISCONNECTING) &&
		    !rds_conn_path_transition(cp, RDS_CONN_ERROR,
					      RDS_CONN_DISCONNECTING)) {
			rds_conn_path_error(cp,
					    "shutdown called in state %d\n",
					    atomic_read(&cp->cp_state));
			mutex_unlock(&cp->cp_cm_lock);
329 330
			return;
		}
331
		mutex_unlock(&cp->cp_cm_lock);
332

333 334 335 336
		wait_event(cp->cp_waitq,
			   !test_bit(RDS_IN_XMIT, &cp->cp_flags));
		wait_event(cp->cp_waitq,
			   !test_bit(RDS_RECV_REFILL, &cp->cp_flags));
337

338
		conn->c_trans->conn_path_shutdown(cp);
339
		rds_conn_path_reset(cp);
340

341
		if (!rds_conn_path_transition(cp, RDS_CONN_DISCONNECTING,
342 343
					      RDS_CONN_DOWN) &&
		    !rds_conn_path_transition(cp, RDS_CONN_ERROR,
344
					      RDS_CONN_DOWN)) {
345 346
			/* This can happen - eg when we're in the middle of tearing
			 * down the connection, and someone unloads the rds module.
347
			 * Quite reproducible with loopback connections.
348
			 * Mostly harmless.
349 350 351 352 353 354
			 *
			 * Note that this also happens with rds-tcp because
			 * we could have triggered rds_conn_path_drop in irq
			 * mode from rds_tcp_state change on the receipt of
			 * a FIN, thus we need to recheck for RDS_CONN_ERROR
			 * here.
355
			 */
356 357 358 359
			rds_conn_path_error(cp, "%s: failed to transition "
					    "to state DOWN, current state "
					    "is %d\n", __func__,
					    atomic_read(&cp->cp_state));
360 361 362 363 364 365 366 367
			return;
		}
	}

	/* Then reconnect if it's still live.
	 * The passive side of an IB loopback connection is never added
	 * to the conn hash, so we never trigger a reconnect on this
	 * conn - the reconnect is always triggered by the active peer. */
368
	cancel_delayed_work_sync(&cp->cp_conn_w);
369 370 371
	rcu_read_lock();
	if (!hlist_unhashed(&conn->c_hash_node)) {
		rcu_read_unlock();
372
		rds_queue_reconnect(cp);
373 374 375
	} else {
		rcu_read_unlock();
	}
376 377
}

378 379 380 381 382 383 384
/* destroy a single rds_conn_path. rds_conn_destroy() iterates over
 * all paths using rds_conn_path_destroy()
 */
static void rds_conn_path_destroy(struct rds_conn_path *cp)
{
	struct rds_message *rm, *rtmp;

385 386 387
	if (!cp->cp_transport_data)
		return;

388 389 390 391
	/* make sure lingering queued work won't try to ref the conn */
	cancel_delayed_work_sync(&cp->cp_send_w);
	cancel_delayed_work_sync(&cp->cp_recv_w);

392 393 394
	rds_conn_path_drop(cp, true);
	flush_work(&cp->cp_down_w);

395 396 397 398 399 400 401 402 403 404 405 406 407 408
	/* tear down queued messages */
	list_for_each_entry_safe(rm, rtmp,
				 &cp->cp_send_queue,
				 m_conn_item) {
		list_del_init(&rm->m_conn_item);
		BUG_ON(!list_empty(&rm->m_sock_item));
		rds_message_put(rm);
	}
	if (cp->cp_xmit_rm)
		rds_message_put(cp->cp_xmit_rm);

	cp->cp_conn->c_trans->conn_free(cp->cp_transport_data);
}

409 410
/*
 * Stop and free a connection.
411 412 413 414
 *
 * This can only be used in very limited circumstances.  It assumes that once
 * the conn has been shutdown that no one else is referencing the connection.
 * We can only ensure this in the rmmod path in the current code.
415
 */
A
Andy Grover 已提交
416 417
void rds_conn_destroy(struct rds_connection *conn)
{
418
	unsigned long flags;
419 420
	int i;
	struct rds_conn_path *cp;
421
	int npaths = (conn->c_trans->t_mp_capable ? RDS_MPATH_WORKERS : 1);
A
Andy Grover 已提交
422 423 424 425 426

	rdsdebug("freeing conn %p for %pI4 -> "
		 "%pI4\n", conn, &conn->c_laddr,
		 &conn->c_faddr);

427
	conn->c_destroy_in_prog = 1;
428 429
	/* Ensure conn will not be scheduled for reconnect */
	spin_lock_irq(&rds_conn_lock);
430
	hlist_del_init_rcu(&conn->c_hash_node);
431
	spin_unlock_irq(&rds_conn_lock);
432 433
	synchronize_rcu();

434
	/* shut the connection down */
435
	for (i = 0; i < npaths; i++) {
436 437 438
		cp = &conn->c_path[i];
		rds_conn_path_destroy(cp);
		BUG_ON(!list_empty(&cp->cp_retrans));
A
Andy Grover 已提交
439 440 441 442 443 444 445 446 447
	}

	/*
	 * The congestion maps aren't freed up here.  They're
	 * freed by rds_cong_exit() after all the connections
	 * have been freed.
	 */
	rds_cong_remove_conn(conn);

448
	put_net(conn->c_net);
449
	kfree(conn->c_path);
A
Andy Grover 已提交
450 451
	kmem_cache_free(rds_conn_slab, conn);

452
	spin_lock_irqsave(&rds_conn_lock, flags);
A
Andy Grover 已提交
453
	rds_conn_count--;
454
	spin_unlock_irqrestore(&rds_conn_lock, flags);
A
Andy Grover 已提交
455
}
A
Andy Grover 已提交
456
EXPORT_SYMBOL_GPL(rds_conn_destroy);
A
Andy Grover 已提交
457 458 459 460 461 462 463 464 465 466 467

static void rds_conn_message_info(struct socket *sock, unsigned int len,
				  struct rds_info_iterator *iter,
				  struct rds_info_lengths *lens,
				  int want_send)
{
	struct hlist_head *head;
	struct list_head *list;
	struct rds_connection *conn;
	struct rds_message *rm;
	unsigned int total = 0;
468
	unsigned long flags;
A
Andy Grover 已提交
469
	size_t i;
470
	int j;
A
Andy Grover 已提交
471 472 473

	len /= sizeof(struct rds_info_message);

474
	rcu_read_lock();
A
Andy Grover 已提交
475 476 477

	for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash);
	     i++, head++) {
478
		hlist_for_each_entry_rcu(conn, head, c_hash_node) {
479
			struct rds_conn_path *cp;
480 481 482 483
			int npaths;

			npaths = (conn->c_trans->t_mp_capable ?
				 RDS_MPATH_WORKERS : 1);
484

485
			for (j = 0; j < npaths; j++) {
486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505
				cp = &conn->c_path[j];
				if (want_send)
					list = &cp->cp_send_queue;
				else
					list = &cp->cp_retrans;

				spin_lock_irqsave(&cp->cp_lock, flags);

				/* XXX too lazy to maintain counts.. */
				list_for_each_entry(rm, list, m_conn_item) {
					total++;
					if (total <= len)
						rds_inc_info_copy(&rm->m_inc,
								  iter,
								  conn->c_laddr,
								  conn->c_faddr,
								  0);
				}

				spin_unlock_irqrestore(&cp->cp_lock, flags);
A
Andy Grover 已提交
506 507 508
			}
		}
	}
509
	rcu_read_unlock();
A
Andy Grover 已提交
510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540

	lens->nr = total;
	lens->each = sizeof(struct rds_info_message);
}

static void rds_conn_message_info_send(struct socket *sock, unsigned int len,
				       struct rds_info_iterator *iter,
				       struct rds_info_lengths *lens)
{
	rds_conn_message_info(sock, len, iter, lens, 1);
}

static void rds_conn_message_info_retrans(struct socket *sock,
					  unsigned int len,
					  struct rds_info_iterator *iter,
					  struct rds_info_lengths *lens)
{
	rds_conn_message_info(sock, len, iter, lens, 0);
}

void rds_for_each_conn_info(struct socket *sock, unsigned int len,
			  struct rds_info_iterator *iter,
			  struct rds_info_lengths *lens,
			  int (*visitor)(struct rds_connection *, void *),
			  size_t item_len)
{
	uint64_t buffer[(item_len + 7) / 8];
	struct hlist_head *head;
	struct rds_connection *conn;
	size_t i;

541
	rcu_read_lock();
A
Andy Grover 已提交
542 543 544 545 546 547

	lens->nr = 0;
	lens->each = item_len;

	for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash);
	     i++, head++) {
548
		hlist_for_each_entry_rcu(conn, head, c_hash_node) {
A
Andy Grover 已提交
549 550 551 552 553 554 555 556 557 558 559 560 561 562 563

			/* XXX no c_lock usage.. */
			if (!visitor(conn, buffer))
				continue;

			/* We copy as much as we can fit in the buffer,
			 * but we count all items so that the caller
			 * can resize the buffer. */
			if (len >= item_len) {
				rds_info_copy(iter, buffer, item_len);
				len -= item_len;
			}
			lens->nr++;
		}
	}
564
	rcu_read_unlock();
A
Andy Grover 已提交
565
}
A
Andy Grover 已提交
566
EXPORT_SYMBOL_GPL(rds_for_each_conn_info);
A
Andy Grover 已提交
567

568 569 570 571 572
static void rds_walk_conn_path_info(struct socket *sock, unsigned int len,
				    struct rds_info_iterator *iter,
				    struct rds_info_lengths *lens,
				    int (*visitor)(struct rds_conn_path *, void *),
				    size_t item_len)
573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588
{
	u64  buffer[(item_len + 7) / 8];
	struct hlist_head *head;
	struct rds_connection *conn;
	size_t i;
	int j;

	rcu_read_lock();

	lens->nr = 0;
	lens->each = item_len;

	for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash);
	     i++, head++) {
		hlist_for_each_entry_rcu(conn, head, c_hash_node) {
			struct rds_conn_path *cp;
589
			int npaths;
590

591 592 593
			npaths = (conn->c_trans->t_mp_capable ?
				 RDS_MPATH_WORKERS : 1);
			for (j = 0; j < npaths; j++) {
594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615
				cp = &conn->c_path[j];

				/* XXX no cp_lock usage.. */
				if (!visitor(cp, buffer))
					continue;
			}

			/* We copy as much as we can fit in the buffer,
			 * but we count all items so that the caller
			 * can resize the buffer.
			 */
			if (len >= item_len) {
				rds_info_copy(iter, buffer, item_len);
				len -= item_len;
			}
			lens->nr++;
		}
	}
	rcu_read_unlock();
}

static int rds_conn_info_visitor(struct rds_conn_path *cp, void *buffer)
A
Andy Grover 已提交
616 617 618
{
	struct rds_info_connection *cinfo = buffer;

619 620 621 622 623
	cinfo->next_tx_seq = cp->cp_next_tx_seq;
	cinfo->next_rx_seq = cp->cp_next_rx_seq;
	cinfo->laddr = cp->cp_conn->c_laddr;
	cinfo->faddr = cp->cp_conn->c_faddr;
	strncpy(cinfo->transport, cp->cp_conn->c_trans->t_name,
A
Andy Grover 已提交
624 625 626
		sizeof(cinfo->transport));
	cinfo->flags = 0;

627
	rds_conn_info_set(cinfo->flags, test_bit(RDS_IN_XMIT, &cp->cp_flags),
628
			  SENDING);
A
Andy Grover 已提交
629 630
	/* XXX Future: return the state rather than these funky bits */
	rds_conn_info_set(cinfo->flags,
631
			  atomic_read(&cp->cp_state) == RDS_CONN_CONNECTING,
A
Andy Grover 已提交
632 633
			  CONNECTING);
	rds_conn_info_set(cinfo->flags,
634
			  atomic_read(&cp->cp_state) == RDS_CONN_UP,
A
Andy Grover 已提交
635 636 637 638 639 640 641 642
			  CONNECTED);
	return 1;
}

static void rds_conn_info(struct socket *sock, unsigned int len,
			  struct rds_info_iterator *iter,
			  struct rds_info_lengths *lens)
{
643
	rds_walk_conn_path_info(sock, len, iter, lens,
A
Andy Grover 已提交
644 645 646 647
				rds_conn_info_visitor,
				sizeof(struct rds_info_connection));
}

648
int rds_conn_init(void)
A
Andy Grover 已提交
649 650 651 652
{
	rds_conn_slab = kmem_cache_create("rds_connection",
					  sizeof(struct rds_connection),
					  0, 0, NULL);
653
	if (!rds_conn_slab)
A
Andy Grover 已提交
654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682
		return -ENOMEM;

	rds_info_register_func(RDS_INFO_CONNECTIONS, rds_conn_info);
	rds_info_register_func(RDS_INFO_SEND_MESSAGES,
			       rds_conn_message_info_send);
	rds_info_register_func(RDS_INFO_RETRANS_MESSAGES,
			       rds_conn_message_info_retrans);

	return 0;
}

void rds_conn_exit(void)
{
	rds_loop_exit();

	WARN_ON(!hlist_empty(rds_conn_hash));

	kmem_cache_destroy(rds_conn_slab);

	rds_info_deregister_func(RDS_INFO_CONNECTIONS, rds_conn_info);
	rds_info_deregister_func(RDS_INFO_SEND_MESSAGES,
				 rds_conn_message_info_send);
	rds_info_deregister_func(RDS_INFO_RETRANS_MESSAGES,
				 rds_conn_message_info_retrans);
}

/*
 * Force a disconnect
 */
683
void rds_conn_path_drop(struct rds_conn_path *cp, bool destroy)
684 685
{
	atomic_set(&cp->cp_state, RDS_CONN_ERROR);
686 687 688 689

	if (!destroy && cp->cp_conn->c_destroy_in_prog)
		return;

690 691 692 693
	queue_work(rds_wq, &cp->cp_down_w);
}
EXPORT_SYMBOL_GPL(rds_conn_path_drop);

A
Andy Grover 已提交
694 695
void rds_conn_drop(struct rds_connection *conn)
{
696
	WARN_ON(conn->c_trans->t_mp_capable);
697
	rds_conn_path_drop(&conn->c_path[0], false);
A
Andy Grover 已提交
698
}
A
Andy Grover 已提交
699
EXPORT_SYMBOL_GPL(rds_conn_drop);
A
Andy Grover 已提交
700

701 702 703 704
/*
 * If the connection is down, trigger a connect. We may have scheduled a
 * delayed reconnect however - in this case we should not interfere.
 */
705 706 707 708 709 710
void rds_conn_path_connect_if_down(struct rds_conn_path *cp)
{
	if (rds_conn_path_state(cp) == RDS_CONN_DOWN &&
	    !test_and_set_bit(RDS_RECONNECT_PENDING, &cp->cp_flags))
		queue_delayed_work(rds_wq, &cp->cp_conn_w, 0);
}
711
EXPORT_SYMBOL_GPL(rds_conn_path_connect_if_down);
712

713 714
void rds_conn_connect_if_down(struct rds_connection *conn)
{
715 716
	WARN_ON(conn->c_trans->t_mp_capable);
	rds_conn_path_connect_if_down(&conn->c_path[0]);
717 718 719
}
EXPORT_SYMBOL_GPL(rds_conn_connect_if_down);

720 721 722 723 724 725 726 727 728
void
__rds_conn_path_error(struct rds_conn_path *cp, const char *fmt, ...)
{
	va_list ap;

	va_start(ap, fmt);
	vprintk(fmt, ap);
	va_end(ap);

729
	rds_conn_path_drop(cp, false);
730
}