• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0-only
2 /******************************************************************************
3 *******************************************************************************
4 **
5 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
6 **  Copyright (C) 2004-2009 Red Hat, Inc.  All rights reserved.
7 **
8 **
9 *******************************************************************************
10 ******************************************************************************/
11 
12 /*
13  * lowcomms.c
14  *
15  * This is the "low-level" comms layer.
16  *
17  * It is responsible for sending/receiving messages
18  * from other nodes in the cluster.
19  *
20  * Cluster nodes are referred to by their nodeids. nodeids are
21  * simply 32 bit numbers to the locking module - if they need to
22  * be expanded for the cluster infrastructure then that is its
23  * responsibility. It is this layer's
24  * responsibility to resolve these into IP address or
25  * whatever it needs for inter-node communication.
26  *
27  * The comms level is two kernel threads that deal mainly with
28  * the receiving of messages from other nodes and passing them
29  * up to the mid-level comms layer (which understands the
30  * message format) for execution by the locking core, and
31  * a send thread which does all the setting up of connections
32  * to remote nodes and the sending of data. Threads are not allowed
33  * to send their own data because it may cause them to wait in times
34  * of high load. Also, this way, the sending thread can collect together
35  * messages bound for one node and send them in one block.
36  *
37  * lowcomms will choose to use either TCP or SCTP as its transport layer
38  * depending on the configuration variable 'protocol'. This should be set
39  * to 0 (default) for TCP or 1 for SCTP. It should be configured using a
40  * cluster-wide mechanism as it must be the same on all nodes of the cluster
41  * for the DLM to function.
42  *
43  */
44 
45 #include <asm/ioctls.h>
46 #include <net/sock.h>
47 #include <net/tcp.h>
48 #include <linux/pagemap.h>
49 #include <linux/file.h>
50 #include <linux/mutex.h>
51 #include <linux/sctp.h>
52 #include <linux/slab.h>
53 #include <net/sctp/sctp.h>
54 #include <net/ipv6.h>
55 
56 #include "dlm_internal.h"
57 #include "lowcomms.h"
58 #include "midcomms.h"
59 #include "config.h"
60 
61 #define NEEDED_RMEM (4*1024*1024)
62 #define CONN_HASH_SIZE 32
63 
64 /* Number of messages to send before rescheduling */
65 #define MAX_SEND_MSG_COUNT 25
66 #define DLM_SHUTDOWN_WAIT_TIMEOUT msecs_to_jiffies(10000)
67 
68 struct connection {
69 	struct socket *sock;	/* NULL if not connected */
70 	uint32_t nodeid;	/* So we know who we are in the list */
71 	struct mutex sock_mutex;
72 	unsigned long flags;
73 #define CF_READ_PENDING 1
74 #define CF_WRITE_PENDING 2
75 #define CF_INIT_PENDING 4
76 #define CF_IS_OTHERCON 5
77 #define CF_CLOSE 6
78 #define CF_APP_LIMITED 7
79 #define CF_CLOSING 8
80 #define CF_SHUTDOWN 9
81 	struct list_head writequeue;  /* List of outgoing writequeue_entries */
82 	spinlock_t writequeue_lock;
83 	int (*rx_action) (struct connection *);	/* What to do when active */
84 	void (*connect_action) (struct connection *);	/* What to do to connect */
85 	void (*shutdown_action)(struct connection *con); /* What to do to shutdown */
86 	int retries;
87 #define MAX_CONNECT_RETRIES 3
88 	struct hlist_node list;
89 	struct connection *othercon;
90 	struct work_struct rwork; /* Receive workqueue */
91 	struct work_struct swork; /* Send workqueue */
92 	wait_queue_head_t shutdown_wait; /* wait for graceful shutdown */
93 	unsigned char *rx_buf;
94 	int rx_buflen;
95 	int rx_leftover;
96 	struct rcu_head rcu;
97 };
98 #define sock2con(x) ((struct connection *)(x)->sk_user_data)
99 
100 /* An entry waiting to be sent */
101 struct writequeue_entry {
102 	struct list_head list;
103 	struct page *page;
104 	int offset;
105 	int len;
106 	int end;
107 	int users;
108 	struct connection *con;
109 };
110 
111 struct dlm_node_addr {
112 	struct list_head list;
113 	int nodeid;
114 	int addr_count;
115 	int curr_addr_index;
116 	struct sockaddr_storage *addr[DLM_MAX_ADDR_COUNT];
117 };
118 
119 static struct listen_sock_callbacks {
120 	void (*sk_error_report)(struct sock *);
121 	void (*sk_data_ready)(struct sock *);
122 	void (*sk_state_change)(struct sock *);
123 	void (*sk_write_space)(struct sock *);
124 } listen_sock;
125 
126 static LIST_HEAD(dlm_node_addrs);
127 static DEFINE_SPINLOCK(dlm_node_addrs_spin);
128 
129 static struct sockaddr_storage *dlm_local_addr[DLM_MAX_ADDR_COUNT];
130 static int dlm_local_count;
131 static int dlm_allow_conn;
132 
133 /* Work queues */
134 static struct workqueue_struct *recv_workqueue;
135 static struct workqueue_struct *send_workqueue;
136 
137 static struct hlist_head connection_hash[CONN_HASH_SIZE];
138 static DEFINE_SPINLOCK(connections_lock);
139 DEFINE_STATIC_SRCU(connections_srcu);
140 
141 static void process_recv_sockets(struct work_struct *work);
142 static void process_send_sockets(struct work_struct *work);
143 
144 
145 /* This is deliberately very simple because most clusters have simple
146    sequential nodeids, so we should be able to go straight to a connection
147    struct in the array */
nodeid_hash(int nodeid)148 static inline int nodeid_hash(int nodeid)
149 {
150 	return nodeid & (CONN_HASH_SIZE-1);
151 }
152 
__find_con(int nodeid)153 static struct connection *__find_con(int nodeid)
154 {
155 	int r, idx;
156 	struct connection *con;
157 
158 	r = nodeid_hash(nodeid);
159 
160 	idx = srcu_read_lock(&connections_srcu);
161 	hlist_for_each_entry_rcu(con, &connection_hash[r], list) {
162 		if (con->nodeid == nodeid) {
163 			srcu_read_unlock(&connections_srcu, idx);
164 			return con;
165 		}
166 	}
167 	srcu_read_unlock(&connections_srcu, idx);
168 
169 	return NULL;
170 }
171 
172 /*
173  * If 'allocation' is zero then we don't attempt to create a new
174  * connection structure for this node.
175  */
nodeid2con(int nodeid,gfp_t alloc)176 static struct connection *nodeid2con(int nodeid, gfp_t alloc)
177 {
178 	struct connection *con, *tmp;
179 	int r;
180 
181 	con = __find_con(nodeid);
182 	if (con || !alloc)
183 		return con;
184 
185 	con = kzalloc(sizeof(*con), alloc);
186 	if (!con)
187 		return NULL;
188 
189 	con->rx_buflen = dlm_config.ci_buffer_size;
190 	con->rx_buf = kmalloc(con->rx_buflen, GFP_NOFS);
191 	if (!con->rx_buf) {
192 		kfree(con);
193 		return NULL;
194 	}
195 
196 	con->nodeid = nodeid;
197 	mutex_init(&con->sock_mutex);
198 	INIT_LIST_HEAD(&con->writequeue);
199 	spin_lock_init(&con->writequeue_lock);
200 	INIT_WORK(&con->swork, process_send_sockets);
201 	INIT_WORK(&con->rwork, process_recv_sockets);
202 	init_waitqueue_head(&con->shutdown_wait);
203 
204 	/* Setup action pointers for child sockets */
205 	if (con->nodeid) {
206 		struct connection *zerocon = __find_con(0);
207 
208 		con->connect_action = zerocon->connect_action;
209 		if (!con->rx_action)
210 			con->rx_action = zerocon->rx_action;
211 	}
212 
213 	r = nodeid_hash(nodeid);
214 
215 	spin_lock(&connections_lock);
216 	/* Because multiple workqueues/threads calls this function it can
217 	 * race on multiple cpu's. Instead of locking hot path __find_con()
218 	 * we just check in rare cases of recently added nodes again
219 	 * under protection of connections_lock. If this is the case we
220 	 * abort our connection creation and return the existing connection.
221 	 */
222 	tmp = __find_con(nodeid);
223 	if (tmp) {
224 		spin_unlock(&connections_lock);
225 		kfree(con->rx_buf);
226 		kfree(con);
227 		return tmp;
228 	}
229 
230 	hlist_add_head_rcu(&con->list, &connection_hash[r]);
231 	spin_unlock(&connections_lock);
232 
233 	return con;
234 }
235 
236 /* Loop round all connections */
foreach_conn(void (* conn_func)(struct connection * c))237 static void foreach_conn(void (*conn_func)(struct connection *c))
238 {
239 	int i, idx;
240 	struct connection *con;
241 
242 	idx = srcu_read_lock(&connections_srcu);
243 	for (i = 0; i < CONN_HASH_SIZE; i++) {
244 		hlist_for_each_entry_rcu(con, &connection_hash[i], list)
245 			conn_func(con);
246 	}
247 	srcu_read_unlock(&connections_srcu, idx);
248 }
249 
find_node_addr(int nodeid)250 static struct dlm_node_addr *find_node_addr(int nodeid)
251 {
252 	struct dlm_node_addr *na;
253 
254 	list_for_each_entry(na, &dlm_node_addrs, list) {
255 		if (na->nodeid == nodeid)
256 			return na;
257 	}
258 	return NULL;
259 }
260 
addr_compare(struct sockaddr_storage * x,struct sockaddr_storage * y)261 static int addr_compare(struct sockaddr_storage *x, struct sockaddr_storage *y)
262 {
263 	switch (x->ss_family) {
264 	case AF_INET: {
265 		struct sockaddr_in *sinx = (struct sockaddr_in *)x;
266 		struct sockaddr_in *siny = (struct sockaddr_in *)y;
267 		if (sinx->sin_addr.s_addr != siny->sin_addr.s_addr)
268 			return 0;
269 		if (sinx->sin_port != siny->sin_port)
270 			return 0;
271 		break;
272 	}
273 	case AF_INET6: {
274 		struct sockaddr_in6 *sinx = (struct sockaddr_in6 *)x;
275 		struct sockaddr_in6 *siny = (struct sockaddr_in6 *)y;
276 		if (!ipv6_addr_equal(&sinx->sin6_addr, &siny->sin6_addr))
277 			return 0;
278 		if (sinx->sin6_port != siny->sin6_port)
279 			return 0;
280 		break;
281 	}
282 	default:
283 		return 0;
284 	}
285 	return 1;
286 }
287 
nodeid_to_addr(int nodeid,struct sockaddr_storage * sas_out,struct sockaddr * sa_out,bool try_new_addr)288 static int nodeid_to_addr(int nodeid, struct sockaddr_storage *sas_out,
289 			  struct sockaddr *sa_out, bool try_new_addr)
290 {
291 	struct sockaddr_storage sas;
292 	struct dlm_node_addr *na;
293 
294 	if (!dlm_local_count)
295 		return -1;
296 
297 	spin_lock(&dlm_node_addrs_spin);
298 	na = find_node_addr(nodeid);
299 	if (na && na->addr_count) {
300 		memcpy(&sas, na->addr[na->curr_addr_index],
301 		       sizeof(struct sockaddr_storage));
302 
303 		if (try_new_addr) {
304 			na->curr_addr_index++;
305 			if (na->curr_addr_index == na->addr_count)
306 				na->curr_addr_index = 0;
307 		}
308 	}
309 	spin_unlock(&dlm_node_addrs_spin);
310 
311 	if (!na)
312 		return -EEXIST;
313 
314 	if (!na->addr_count)
315 		return -ENOENT;
316 
317 	if (sas_out)
318 		memcpy(sas_out, &sas, sizeof(struct sockaddr_storage));
319 
320 	if (!sa_out)
321 		return 0;
322 
323 	if (dlm_local_addr[0]->ss_family == AF_INET) {
324 		struct sockaddr_in *in4  = (struct sockaddr_in *) &sas;
325 		struct sockaddr_in *ret4 = (struct sockaddr_in *) sa_out;
326 		ret4->sin_addr.s_addr = in4->sin_addr.s_addr;
327 	} else {
328 		struct sockaddr_in6 *in6  = (struct sockaddr_in6 *) &sas;
329 		struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) sa_out;
330 		ret6->sin6_addr = in6->sin6_addr;
331 	}
332 
333 	return 0;
334 }
335 
addr_to_nodeid(struct sockaddr_storage * addr,int * nodeid)336 static int addr_to_nodeid(struct sockaddr_storage *addr, int *nodeid)
337 {
338 	struct dlm_node_addr *na;
339 	int rv = -EEXIST;
340 	int addr_i;
341 
342 	spin_lock(&dlm_node_addrs_spin);
343 	list_for_each_entry(na, &dlm_node_addrs, list) {
344 		if (!na->addr_count)
345 			continue;
346 
347 		for (addr_i = 0; addr_i < na->addr_count; addr_i++) {
348 			if (addr_compare(na->addr[addr_i], addr)) {
349 				*nodeid = na->nodeid;
350 				rv = 0;
351 				goto unlock;
352 			}
353 		}
354 	}
355 unlock:
356 	spin_unlock(&dlm_node_addrs_spin);
357 	return rv;
358 }
359 
dlm_lowcomms_addr(int nodeid,struct sockaddr_storage * addr,int len)360 int dlm_lowcomms_addr(int nodeid, struct sockaddr_storage *addr, int len)
361 {
362 	struct sockaddr_storage *new_addr;
363 	struct dlm_node_addr *new_node, *na;
364 
365 	new_node = kzalloc(sizeof(struct dlm_node_addr), GFP_NOFS);
366 	if (!new_node)
367 		return -ENOMEM;
368 
369 	new_addr = kzalloc(sizeof(struct sockaddr_storage), GFP_NOFS);
370 	if (!new_addr) {
371 		kfree(new_node);
372 		return -ENOMEM;
373 	}
374 
375 	memcpy(new_addr, addr, len);
376 
377 	spin_lock(&dlm_node_addrs_spin);
378 	na = find_node_addr(nodeid);
379 	if (!na) {
380 		new_node->nodeid = nodeid;
381 		new_node->addr[0] = new_addr;
382 		new_node->addr_count = 1;
383 		list_add(&new_node->list, &dlm_node_addrs);
384 		spin_unlock(&dlm_node_addrs_spin);
385 		return 0;
386 	}
387 
388 	if (na->addr_count >= DLM_MAX_ADDR_COUNT) {
389 		spin_unlock(&dlm_node_addrs_spin);
390 		kfree(new_addr);
391 		kfree(new_node);
392 		return -ENOSPC;
393 	}
394 
395 	na->addr[na->addr_count++] = new_addr;
396 	spin_unlock(&dlm_node_addrs_spin);
397 	kfree(new_node);
398 	return 0;
399 }
400 
401 /* Data available on socket or listen socket received a connect */
lowcomms_data_ready(struct sock * sk)402 static void lowcomms_data_ready(struct sock *sk)
403 {
404 	struct connection *con;
405 
406 	read_lock_bh(&sk->sk_callback_lock);
407 	con = sock2con(sk);
408 	if (con && !test_and_set_bit(CF_READ_PENDING, &con->flags))
409 		queue_work(recv_workqueue, &con->rwork);
410 	read_unlock_bh(&sk->sk_callback_lock);
411 }
412 
lowcomms_write_space(struct sock * sk)413 static void lowcomms_write_space(struct sock *sk)
414 {
415 	struct connection *con;
416 
417 	read_lock_bh(&sk->sk_callback_lock);
418 	con = sock2con(sk);
419 	if (!con)
420 		goto out;
421 
422 	clear_bit(SOCK_NOSPACE, &con->sock->flags);
423 
424 	if (test_and_clear_bit(CF_APP_LIMITED, &con->flags)) {
425 		con->sock->sk->sk_write_pending--;
426 		clear_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags);
427 	}
428 
429 	queue_work(send_workqueue, &con->swork);
430 out:
431 	read_unlock_bh(&sk->sk_callback_lock);
432 }
433 
lowcomms_connect_sock(struct connection * con)434 static inline void lowcomms_connect_sock(struct connection *con)
435 {
436 	if (test_bit(CF_CLOSE, &con->flags))
437 		return;
438 	queue_work(send_workqueue, &con->swork);
439 	cond_resched();
440 }
441 
lowcomms_state_change(struct sock * sk)442 static void lowcomms_state_change(struct sock *sk)
443 {
444 	/* SCTP layer is not calling sk_data_ready when the connection
445 	 * is done, so we catch the signal through here. Also, it
446 	 * doesn't switch socket state when entering shutdown, so we
447 	 * skip the write in that case.
448 	 */
449 	if (sk->sk_shutdown) {
450 		if (sk->sk_shutdown == RCV_SHUTDOWN)
451 			lowcomms_data_ready(sk);
452 	} else if (sk->sk_state == TCP_ESTABLISHED) {
453 		lowcomms_write_space(sk);
454 	}
455 }
456 
dlm_lowcomms_connect_node(int nodeid)457 int dlm_lowcomms_connect_node(int nodeid)
458 {
459 	struct connection *con;
460 
461 	if (nodeid == dlm_our_nodeid())
462 		return 0;
463 
464 	con = nodeid2con(nodeid, GFP_NOFS);
465 	if (!con)
466 		return -ENOMEM;
467 	lowcomms_connect_sock(con);
468 	return 0;
469 }
470 
lowcomms_error_report(struct sock * sk)471 static void lowcomms_error_report(struct sock *sk)
472 {
473 	struct connection *con;
474 	void (*orig_report)(struct sock *) = NULL;
475 	struct inet_sock *inet;
476 
477 	read_lock_bh(&sk->sk_callback_lock);
478 	con = sock2con(sk);
479 	if (con == NULL)
480 		goto out;
481 
482 	orig_report = listen_sock.sk_error_report;
483 
484 	inet = inet_sk(sk);
485 	switch (sk->sk_family) {
486 	case AF_INET:
487 		printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
488 				   "sending to node %d at %pI4, dport %d, "
489 				   "sk_err=%d/%d\n", dlm_our_nodeid(),
490 				   con->nodeid, &inet->inet_daddr,
491 				   ntohs(inet->inet_dport), sk->sk_err,
492 				   sk->sk_err_soft);
493 		break;
494 #if IS_ENABLED(CONFIG_IPV6)
495 	case AF_INET6:
496 		printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
497 				   "sending to node %d at %pI6c, "
498 				   "dport %d, sk_err=%d/%d\n", dlm_our_nodeid(),
499 				   con->nodeid, &sk->sk_v6_daddr,
500 				   ntohs(inet->inet_dport), sk->sk_err,
501 				   sk->sk_err_soft);
502 		break;
503 #endif
504 	default:
505 		printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
506 				   "invalid socket family %d set, "
507 				   "sk_err=%d/%d\n", dlm_our_nodeid(),
508 				   sk->sk_family, sk->sk_err, sk->sk_err_soft);
509 		goto out;
510 	}
511 out:
512 	read_unlock_bh(&sk->sk_callback_lock);
513 	if (orig_report)
514 		orig_report(sk);
515 }
516 
517 /* Note: sk_callback_lock must be locked before calling this function. */
save_listen_callbacks(struct socket * sock)518 static void save_listen_callbacks(struct socket *sock)
519 {
520 	struct sock *sk = sock->sk;
521 
522 	listen_sock.sk_data_ready = sk->sk_data_ready;
523 	listen_sock.sk_state_change = sk->sk_state_change;
524 	listen_sock.sk_write_space = sk->sk_write_space;
525 	listen_sock.sk_error_report = sk->sk_error_report;
526 }
527 
restore_callbacks(struct socket * sock)528 static void restore_callbacks(struct socket *sock)
529 {
530 	struct sock *sk = sock->sk;
531 
532 	write_lock_bh(&sk->sk_callback_lock);
533 	sk->sk_user_data = NULL;
534 	sk->sk_data_ready = listen_sock.sk_data_ready;
535 	sk->sk_state_change = listen_sock.sk_state_change;
536 	sk->sk_write_space = listen_sock.sk_write_space;
537 	sk->sk_error_report = listen_sock.sk_error_report;
538 	write_unlock_bh(&sk->sk_callback_lock);
539 }
540 
541 /* Make a socket active */
add_sock(struct socket * sock,struct connection * con)542 static void add_sock(struct socket *sock, struct connection *con)
543 {
544 	struct sock *sk = sock->sk;
545 
546 	write_lock_bh(&sk->sk_callback_lock);
547 	con->sock = sock;
548 
549 	sk->sk_user_data = con;
550 	/* Install a data_ready callback */
551 	sk->sk_data_ready = lowcomms_data_ready;
552 	sk->sk_write_space = lowcomms_write_space;
553 	sk->sk_state_change = lowcomms_state_change;
554 	sk->sk_allocation = GFP_NOFS;
555 	sk->sk_error_report = lowcomms_error_report;
556 	write_unlock_bh(&sk->sk_callback_lock);
557 }
558 
559 /* Add the port number to an IPv6 or 4 sockaddr and return the address
560    length */
make_sockaddr(struct sockaddr_storage * saddr,uint16_t port,int * addr_len)561 static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port,
562 			  int *addr_len)
563 {
564 	saddr->ss_family =  dlm_local_addr[0]->ss_family;
565 	if (saddr->ss_family == AF_INET) {
566 		struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr;
567 		in4_addr->sin_port = cpu_to_be16(port);
568 		*addr_len = sizeof(struct sockaddr_in);
569 		memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero));
570 	} else {
571 		struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr;
572 		in6_addr->sin6_port = cpu_to_be16(port);
573 		*addr_len = sizeof(struct sockaddr_in6);
574 	}
575 	memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len);
576 }
577 
578 /* Close a remote connection and tidy up */
close_connection(struct connection * con,bool and_other,bool tx,bool rx)579 static void close_connection(struct connection *con, bool and_other,
580 			     bool tx, bool rx)
581 {
582 	bool closing = test_and_set_bit(CF_CLOSING, &con->flags);
583 
584 	if (tx && !closing && cancel_work_sync(&con->swork)) {
585 		log_print("canceled swork for node %d", con->nodeid);
586 		clear_bit(CF_WRITE_PENDING, &con->flags);
587 	}
588 	if (rx && !closing && cancel_work_sync(&con->rwork)) {
589 		log_print("canceled rwork for node %d", con->nodeid);
590 		clear_bit(CF_READ_PENDING, &con->flags);
591 	}
592 
593 	mutex_lock(&con->sock_mutex);
594 	if (con->sock) {
595 		restore_callbacks(con->sock);
596 		sock_release(con->sock);
597 		con->sock = NULL;
598 	}
599 	if (con->othercon && and_other) {
600 		/* Will only re-enter once. */
601 		close_connection(con->othercon, false, tx, rx);
602 	}
603 
604 	con->rx_leftover = 0;
605 	con->retries = 0;
606 	mutex_unlock(&con->sock_mutex);
607 	clear_bit(CF_CLOSING, &con->flags);
608 }
609 
shutdown_connection(struct connection * con)610 static void shutdown_connection(struct connection *con)
611 {
612 	int ret;
613 
614 	flush_work(&con->swork);
615 
616 	mutex_lock(&con->sock_mutex);
617 	/* nothing to shutdown */
618 	if (!con->sock) {
619 		mutex_unlock(&con->sock_mutex);
620 		return;
621 	}
622 
623 	set_bit(CF_SHUTDOWN, &con->flags);
624 	ret = kernel_sock_shutdown(con->sock, SHUT_WR);
625 	mutex_unlock(&con->sock_mutex);
626 	if (ret) {
627 		log_print("Connection %p failed to shutdown: %d will force close",
628 			  con, ret);
629 		goto force_close;
630 	} else {
631 		ret = wait_event_timeout(con->shutdown_wait,
632 					 !test_bit(CF_SHUTDOWN, &con->flags),
633 					 DLM_SHUTDOWN_WAIT_TIMEOUT);
634 		if (ret == 0) {
635 			log_print("Connection %p shutdown timed out, will force close",
636 				  con);
637 			goto force_close;
638 		}
639 	}
640 
641 	return;
642 
643 force_close:
644 	clear_bit(CF_SHUTDOWN, &con->flags);
645 	close_connection(con, false, true, true);
646 }
647 
dlm_tcp_shutdown(struct connection * con)648 static void dlm_tcp_shutdown(struct connection *con)
649 {
650 	if (con->othercon)
651 		shutdown_connection(con->othercon);
652 	shutdown_connection(con);
653 }
654 
con_realloc_receive_buf(struct connection * con,int newlen)655 static int con_realloc_receive_buf(struct connection *con, int newlen)
656 {
657 	unsigned char *newbuf;
658 
659 	newbuf = kmalloc(newlen, GFP_NOFS);
660 	if (!newbuf)
661 		return -ENOMEM;
662 
663 	/* copy any leftover from last receive */
664 	if (con->rx_leftover)
665 		memmove(newbuf, con->rx_buf, con->rx_leftover);
666 
667 	/* swap to new buffer space */
668 	kfree(con->rx_buf);
669 	con->rx_buflen = newlen;
670 	con->rx_buf = newbuf;
671 
672 	return 0;
673 }
674 
675 /* Data received from remote end */
receive_from_sock(struct connection * con)676 static int receive_from_sock(struct connection *con)
677 {
678 	int call_again_soon = 0;
679 	struct msghdr msg;
680 	struct kvec iov;
681 	int ret, buflen;
682 
683 	mutex_lock(&con->sock_mutex);
684 
685 	if (con->sock == NULL) {
686 		ret = -EAGAIN;
687 		goto out_close;
688 	}
689 
690 	if (con->nodeid == 0) {
691 		ret = -EINVAL;
692 		goto out_close;
693 	}
694 
695 	/* realloc if we get new buffer size to read out */
696 	buflen = dlm_config.ci_buffer_size;
697 	if (con->rx_buflen != buflen && con->rx_leftover <= buflen) {
698 		ret = con_realloc_receive_buf(con, buflen);
699 		if (ret < 0)
700 			goto out_resched;
701 	}
702 
703 	/* calculate new buffer parameter regarding last receive and
704 	 * possible leftover bytes
705 	 */
706 	iov.iov_base = con->rx_buf + con->rx_leftover;
707 	iov.iov_len = con->rx_buflen - con->rx_leftover;
708 
709 	memset(&msg, 0, sizeof(msg));
710 	msg.msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
711 	ret = kernel_recvmsg(con->sock, &msg, &iov, 1, iov.iov_len,
712 			     msg.msg_flags);
713 	if (ret <= 0)
714 		goto out_close;
715 	else if (ret == iov.iov_len)
716 		call_again_soon = 1;
717 
718 	/* new buflen according readed bytes and leftover from last receive */
719 	buflen = ret + con->rx_leftover;
720 	ret = dlm_process_incoming_buffer(con->nodeid, con->rx_buf, buflen);
721 	if (ret < 0)
722 		goto out_close;
723 
724 	/* calculate leftover bytes from process and put it into begin of
725 	 * the receive buffer, so next receive we have the full message
726 	 * at the start address of the receive buffer.
727 	 */
728 	con->rx_leftover = buflen - ret;
729 	if (con->rx_leftover) {
730 		memmove(con->rx_buf, con->rx_buf + ret,
731 			con->rx_leftover);
732 		call_again_soon = true;
733 	}
734 
735 	if (call_again_soon)
736 		goto out_resched;
737 
738 	mutex_unlock(&con->sock_mutex);
739 	return 0;
740 
741 out_resched:
742 	if (!test_and_set_bit(CF_READ_PENDING, &con->flags))
743 		queue_work(recv_workqueue, &con->rwork);
744 	mutex_unlock(&con->sock_mutex);
745 	return -EAGAIN;
746 
747 out_close:
748 	mutex_unlock(&con->sock_mutex);
749 	if (ret != -EAGAIN) {
750 		/* Reconnect when there is something to send */
751 		close_connection(con, false, true, false);
752 		if (ret == 0) {
753 			log_print("connection %p got EOF from %d",
754 				  con, con->nodeid);
755 			/* handling for tcp shutdown */
756 			clear_bit(CF_SHUTDOWN, &con->flags);
757 			wake_up(&con->shutdown_wait);
758 			/* signal to breaking receive worker */
759 			ret = -1;
760 		}
761 	}
762 	return ret;
763 }
764 
765 /* Listening socket is busy, accept a connection */
accept_from_sock(struct connection * con)766 static int accept_from_sock(struct connection *con)
767 {
768 	int result;
769 	struct sockaddr_storage peeraddr;
770 	struct socket *newsock;
771 	int len;
772 	int nodeid;
773 	struct connection *newcon;
774 	struct connection *addcon;
775 	unsigned int mark;
776 
777 	if (!dlm_allow_conn) {
778 		return -1;
779 	}
780 
781 	mutex_lock_nested(&con->sock_mutex, 0);
782 
783 	if (!con->sock) {
784 		mutex_unlock(&con->sock_mutex);
785 		return -ENOTCONN;
786 	}
787 
788 	result = kernel_accept(con->sock, &newsock, O_NONBLOCK);
789 	if (result < 0)
790 		goto accept_err;
791 
792 	/* Get the connected socket's peer */
793 	memset(&peeraddr, 0, sizeof(peeraddr));
794 	len = newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr, 2);
795 	if (len < 0) {
796 		result = -ECONNABORTED;
797 		goto accept_err;
798 	}
799 
800 	/* Get the new node's NODEID */
801 	make_sockaddr(&peeraddr, 0, &len);
802 	if (addr_to_nodeid(&peeraddr, &nodeid)) {
803 		unsigned char *b=(unsigned char *)&peeraddr;
804 		log_print("connect from non cluster node");
805 		print_hex_dump_bytes("ss: ", DUMP_PREFIX_NONE,
806 				     b, sizeof(struct sockaddr_storage));
807 		sock_release(newsock);
808 		mutex_unlock(&con->sock_mutex);
809 		return -1;
810 	}
811 
812 	dlm_comm_mark(nodeid, &mark);
813 	sock_set_mark(newsock->sk, mark);
814 
815 	log_print("got connection from %d", nodeid);
816 
817 	/*  Check to see if we already have a connection to this node. This
818 	 *  could happen if the two nodes initiate a connection at roughly
819 	 *  the same time and the connections cross on the wire.
820 	 *  In this case we store the incoming one in "othercon"
821 	 */
822 	newcon = nodeid2con(nodeid, GFP_NOFS);
823 	if (!newcon) {
824 		result = -ENOMEM;
825 		goto accept_err;
826 	}
827 	mutex_lock_nested(&newcon->sock_mutex, 1);
828 	if (newcon->sock) {
829 		struct connection *othercon = newcon->othercon;
830 
831 		if (!othercon) {
832 			othercon = kzalloc(sizeof(*othercon), GFP_NOFS);
833 			if (!othercon) {
834 				log_print("failed to allocate incoming socket");
835 				mutex_unlock(&newcon->sock_mutex);
836 				result = -ENOMEM;
837 				goto accept_err;
838 			}
839 
840 			othercon->rx_buflen = dlm_config.ci_buffer_size;
841 			othercon->rx_buf = kmalloc(othercon->rx_buflen, GFP_NOFS);
842 			if (!othercon->rx_buf) {
843 				mutex_unlock(&newcon->sock_mutex);
844 				kfree(othercon);
845 				log_print("failed to allocate incoming socket receive buffer");
846 				result = -ENOMEM;
847 				goto accept_err;
848 			}
849 
850 			othercon->nodeid = nodeid;
851 			othercon->rx_action = receive_from_sock;
852 			mutex_init(&othercon->sock_mutex);
853 			INIT_LIST_HEAD(&othercon->writequeue);
854 			spin_lock_init(&othercon->writequeue_lock);
855 			INIT_WORK(&othercon->swork, process_send_sockets);
856 			INIT_WORK(&othercon->rwork, process_recv_sockets);
857 			init_waitqueue_head(&othercon->shutdown_wait);
858 			set_bit(CF_IS_OTHERCON, &othercon->flags);
859 		} else {
860 			/* close other sock con if we have something new */
861 			close_connection(othercon, false, true, false);
862 		}
863 
864 		mutex_lock_nested(&othercon->sock_mutex, 2);
865 		newcon->othercon = othercon;
866 		add_sock(newsock, othercon);
867 		addcon = othercon;
868 		mutex_unlock(&othercon->sock_mutex);
869 	}
870 	else {
871 		newcon->rx_action = receive_from_sock;
872 		/* accept copies the sk after we've saved the callbacks, so we
873 		   don't want to save them a second time or comm errors will
874 		   result in calling sk_error_report recursively. */
875 		add_sock(newsock, newcon);
876 		addcon = newcon;
877 	}
878 
879 	mutex_unlock(&newcon->sock_mutex);
880 
881 	/*
882 	 * Add it to the active queue in case we got data
883 	 * between processing the accept adding the socket
884 	 * to the read_sockets list
885 	 */
886 	if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags))
887 		queue_work(recv_workqueue, &addcon->rwork);
888 	mutex_unlock(&con->sock_mutex);
889 
890 	return 0;
891 
892 accept_err:
893 	mutex_unlock(&con->sock_mutex);
894 	if (newsock)
895 		sock_release(newsock);
896 
897 	if (result != -EAGAIN)
898 		log_print("error accepting connection from node: %d", result);
899 	return result;
900 }
901 
free_entry(struct writequeue_entry * e)902 static void free_entry(struct writequeue_entry *e)
903 {
904 	__free_page(e->page);
905 	kfree(e);
906 }
907 
908 /*
909  * writequeue_entry_complete - try to delete and free write queue entry
910  * @e: write queue entry to try to delete
911  * @completed: bytes completed
912  *
913  * writequeue_lock must be held.
914  */
writequeue_entry_complete(struct writequeue_entry * e,int completed)915 static void writequeue_entry_complete(struct writequeue_entry *e, int completed)
916 {
917 	e->offset += completed;
918 	e->len -= completed;
919 
920 	if (e->len == 0 && e->users == 0) {
921 		list_del(&e->list);
922 		free_entry(e);
923 	}
924 }
925 
926 /*
927  * sctp_bind_addrs - bind a SCTP socket to all our addresses
928  */
sctp_bind_addrs(struct connection * con,uint16_t port)929 static int sctp_bind_addrs(struct connection *con, uint16_t port)
930 {
931 	struct sockaddr_storage localaddr;
932 	struct sockaddr *addr = (struct sockaddr *)&localaddr;
933 	int i, addr_len, result = 0;
934 
935 	for (i = 0; i < dlm_local_count; i++) {
936 		memcpy(&localaddr, dlm_local_addr[i], sizeof(localaddr));
937 		make_sockaddr(&localaddr, port, &addr_len);
938 
939 		if (!i)
940 			result = kernel_bind(con->sock, addr, addr_len);
941 		else
942 			result = sock_bind_add(con->sock->sk, addr, addr_len);
943 
944 		if (result < 0) {
945 			log_print("Can't bind to %d addr number %d, %d.\n",
946 				  port, i + 1, result);
947 			break;
948 		}
949 	}
950 	return result;
951 }
952 
953 /* Initiate an SCTP association.
954    This is a special case of send_to_sock() in that we don't yet have a
955    peeled-off socket for this association, so we use the listening socket
956    and add the primary IP address of the remote node.
957  */
sctp_connect_to_sock(struct connection * con)958 static void sctp_connect_to_sock(struct connection *con)
959 {
960 	struct sockaddr_storage daddr;
961 	int result;
962 	int addr_len;
963 	struct socket *sock;
964 	unsigned int mark;
965 
966 	if (con->nodeid == 0) {
967 		log_print("attempt to connect sock 0 foiled");
968 		return;
969 	}
970 
971 	dlm_comm_mark(con->nodeid, &mark);
972 
973 	mutex_lock(&con->sock_mutex);
974 
975 	/* Some odd races can cause double-connects, ignore them */
976 	if (con->retries++ > MAX_CONNECT_RETRIES)
977 		goto out;
978 
979 	if (con->sock) {
980 		log_print("node %d already connected.", con->nodeid);
981 		goto out;
982 	}
983 
984 	memset(&daddr, 0, sizeof(daddr));
985 	result = nodeid_to_addr(con->nodeid, &daddr, NULL, true);
986 	if (result < 0) {
987 		log_print("no address for nodeid %d", con->nodeid);
988 		goto out;
989 	}
990 
991 	/* Create a socket to communicate with */
992 	result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
993 				  SOCK_STREAM, IPPROTO_SCTP, &sock);
994 	if (result < 0)
995 		goto socket_err;
996 
997 	sock_set_mark(sock->sk, mark);
998 
999 	con->rx_action = receive_from_sock;
1000 	con->connect_action = sctp_connect_to_sock;
1001 	add_sock(sock, con);
1002 
1003 	/* Bind to all addresses. */
1004 	if (sctp_bind_addrs(con, 0))
1005 		goto bind_err;
1006 
1007 	make_sockaddr(&daddr, dlm_config.ci_tcp_port, &addr_len);
1008 
1009 	log_print("connecting to %d", con->nodeid);
1010 
1011 	/* Turn off Nagle's algorithm */
1012 	sctp_sock_set_nodelay(sock->sk);
1013 
1014 	/*
1015 	 * Make sock->ops->connect() function return in specified time,
1016 	 * since O_NONBLOCK argument in connect() function does not work here,
1017 	 * then, we should restore the default value of this attribute.
1018 	 */
1019 	sock_set_sndtimeo(sock->sk, 5);
1020 	result = sock->ops->connect(sock, (struct sockaddr *)&daddr, addr_len,
1021 				   0);
1022 	sock_set_sndtimeo(sock->sk, 0);
1023 
1024 	if (result == -EINPROGRESS)
1025 		result = 0;
1026 	if (result == 0)
1027 		goto out;
1028 
1029 bind_err:
1030 	con->sock = NULL;
1031 	sock_release(sock);
1032 
1033 socket_err:
1034 	/*
1035 	 * Some errors are fatal and this list might need adjusting. For other
1036 	 * errors we try again until the max number of retries is reached.
1037 	 */
1038 	if (result != -EHOSTUNREACH &&
1039 	    result != -ENETUNREACH &&
1040 	    result != -ENETDOWN &&
1041 	    result != -EINVAL &&
1042 	    result != -EPROTONOSUPPORT) {
1043 		log_print("connect %d try %d error %d", con->nodeid,
1044 			  con->retries, result);
1045 		mutex_unlock(&con->sock_mutex);
1046 		msleep(1000);
1047 		lowcomms_connect_sock(con);
1048 		return;
1049 	}
1050 
1051 out:
1052 	mutex_unlock(&con->sock_mutex);
1053 }
1054 
1055 /* Connect a new socket to its peer */
tcp_connect_to_sock(struct connection * con)1056 static void tcp_connect_to_sock(struct connection *con)
1057 {
1058 	struct sockaddr_storage saddr, src_addr;
1059 	int addr_len;
1060 	struct socket *sock = NULL;
1061 	unsigned int mark;
1062 	int result;
1063 
1064 	if (con->nodeid == 0) {
1065 		log_print("attempt to connect sock 0 foiled");
1066 		return;
1067 	}
1068 
1069 	dlm_comm_mark(con->nodeid, &mark);
1070 
1071 	mutex_lock(&con->sock_mutex);
1072 	if (con->retries++ > MAX_CONNECT_RETRIES)
1073 		goto out;
1074 
1075 	/* Some odd races can cause double-connects, ignore them */
1076 	if (con->sock)
1077 		goto out;
1078 
1079 	/* Create a socket to communicate with */
1080 	result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1081 				  SOCK_STREAM, IPPROTO_TCP, &sock);
1082 	if (result < 0)
1083 		goto out_err;
1084 
1085 	sock_set_mark(sock->sk, mark);
1086 
1087 	memset(&saddr, 0, sizeof(saddr));
1088 	result = nodeid_to_addr(con->nodeid, &saddr, NULL, false);
1089 	if (result < 0) {
1090 		log_print("no address for nodeid %d", con->nodeid);
1091 		goto out_err;
1092 	}
1093 
1094 	con->rx_action = receive_from_sock;
1095 	con->connect_action = tcp_connect_to_sock;
1096 	con->shutdown_action = dlm_tcp_shutdown;
1097 	add_sock(sock, con);
1098 
1099 	/* Bind to our cluster-known address connecting to avoid
1100 	   routing problems */
1101 	memcpy(&src_addr, dlm_local_addr[0], sizeof(src_addr));
1102 	make_sockaddr(&src_addr, 0, &addr_len);
1103 	result = sock->ops->bind(sock, (struct sockaddr *) &src_addr,
1104 				 addr_len);
1105 	if (result < 0) {
1106 		log_print("could not bind for connect: %d", result);
1107 		/* This *may* not indicate a critical error */
1108 	}
1109 
1110 	make_sockaddr(&saddr, dlm_config.ci_tcp_port, &addr_len);
1111 
1112 	log_print("connecting to %d", con->nodeid);
1113 
1114 	/* Turn off Nagle's algorithm */
1115 	tcp_sock_set_nodelay(sock->sk);
1116 
1117 	result = sock->ops->connect(sock, (struct sockaddr *)&saddr, addr_len,
1118 				   O_NONBLOCK);
1119 	if (result == -EINPROGRESS)
1120 		result = 0;
1121 	if (result == 0)
1122 		goto out;
1123 
1124 out_err:
1125 	if (con->sock) {
1126 		sock_release(con->sock);
1127 		con->sock = NULL;
1128 	} else if (sock) {
1129 		sock_release(sock);
1130 	}
1131 	/*
1132 	 * Some errors are fatal and this list might need adjusting. For other
1133 	 * errors we try again until the max number of retries is reached.
1134 	 */
1135 	if (result != -EHOSTUNREACH &&
1136 	    result != -ENETUNREACH &&
1137 	    result != -ENETDOWN &&
1138 	    result != -EINVAL &&
1139 	    result != -EPROTONOSUPPORT) {
1140 		log_print("connect %d try %d error %d", con->nodeid,
1141 			  con->retries, result);
1142 		mutex_unlock(&con->sock_mutex);
1143 		msleep(1000);
1144 		lowcomms_connect_sock(con);
1145 		return;
1146 	}
1147 out:
1148 	mutex_unlock(&con->sock_mutex);
1149 	return;
1150 }
1151 
tcp_create_listen_sock(struct connection * con,struct sockaddr_storage * saddr)1152 static struct socket *tcp_create_listen_sock(struct connection *con,
1153 					     struct sockaddr_storage *saddr)
1154 {
1155 	struct socket *sock = NULL;
1156 	int result = 0;
1157 	int addr_len;
1158 
1159 	if (dlm_local_addr[0]->ss_family == AF_INET)
1160 		addr_len = sizeof(struct sockaddr_in);
1161 	else
1162 		addr_len = sizeof(struct sockaddr_in6);
1163 
1164 	/* Create a socket to communicate with */
1165 	result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1166 				  SOCK_STREAM, IPPROTO_TCP, &sock);
1167 	if (result < 0) {
1168 		log_print("Can't create listening comms socket");
1169 		goto create_out;
1170 	}
1171 
1172 	sock_set_mark(sock->sk, dlm_config.ci_mark);
1173 
1174 	/* Turn off Nagle's algorithm */
1175 	tcp_sock_set_nodelay(sock->sk);
1176 
1177 	sock_set_reuseaddr(sock->sk);
1178 
1179 	write_lock_bh(&sock->sk->sk_callback_lock);
1180 	sock->sk->sk_user_data = con;
1181 	save_listen_callbacks(sock);
1182 	con->rx_action = accept_from_sock;
1183 	con->connect_action = tcp_connect_to_sock;
1184 	write_unlock_bh(&sock->sk->sk_callback_lock);
1185 
1186 	/* Bind to our port */
1187 	make_sockaddr(saddr, dlm_config.ci_tcp_port, &addr_len);
1188 	result = sock->ops->bind(sock, (struct sockaddr *) saddr, addr_len);
1189 	if (result < 0) {
1190 		log_print("Can't bind to port %d", dlm_config.ci_tcp_port);
1191 		sock_release(sock);
1192 		sock = NULL;
1193 		con->sock = NULL;
1194 		goto create_out;
1195 	}
1196 	sock_set_keepalive(sock->sk);
1197 
1198 	result = sock->ops->listen(sock, 5);
1199 	if (result < 0) {
1200 		log_print("Can't listen on port %d", dlm_config.ci_tcp_port);
1201 		sock_release(sock);
1202 		sock = NULL;
1203 		goto create_out;
1204 	}
1205 
1206 create_out:
1207 	return sock;
1208 }
1209 
1210 /* Get local addresses */
init_local(void)1211 static void init_local(void)
1212 {
1213 	struct sockaddr_storage sas, *addr;
1214 	int i;
1215 
1216 	dlm_local_count = 0;
1217 	for (i = 0; i < DLM_MAX_ADDR_COUNT; i++) {
1218 		if (dlm_our_addr(&sas, i))
1219 			break;
1220 
1221 		addr = kmemdup(&sas, sizeof(*addr), GFP_NOFS);
1222 		if (!addr)
1223 			break;
1224 		dlm_local_addr[dlm_local_count++] = addr;
1225 	}
1226 }
1227 
deinit_local(void)1228 static void deinit_local(void)
1229 {
1230 	int i;
1231 
1232 	for (i = 0; i < dlm_local_count; i++)
1233 		kfree(dlm_local_addr[i]);
1234 }
1235 
1236 /* Initialise SCTP socket and bind to all interfaces */
sctp_listen_for_all(void)1237 static int sctp_listen_for_all(void)
1238 {
1239 	struct socket *sock = NULL;
1240 	int result = -EINVAL;
1241 	struct connection *con = nodeid2con(0, GFP_NOFS);
1242 
1243 	if (!con)
1244 		return -ENOMEM;
1245 
1246 	log_print("Using SCTP for communications");
1247 
1248 	result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1249 				  SOCK_STREAM, IPPROTO_SCTP, &sock);
1250 	if (result < 0) {
1251 		log_print("Can't create comms socket, check SCTP is loaded");
1252 		goto out;
1253 	}
1254 
1255 	sock_set_rcvbuf(sock->sk, NEEDED_RMEM);
1256 	sock_set_mark(sock->sk, dlm_config.ci_mark);
1257 	sctp_sock_set_nodelay(sock->sk);
1258 
1259 	write_lock_bh(&sock->sk->sk_callback_lock);
1260 	/* Init con struct */
1261 	sock->sk->sk_user_data = con;
1262 	save_listen_callbacks(sock);
1263 	con->sock = sock;
1264 	con->sock->sk->sk_data_ready = lowcomms_data_ready;
1265 	con->rx_action = accept_from_sock;
1266 	con->connect_action = sctp_connect_to_sock;
1267 
1268 	write_unlock_bh(&sock->sk->sk_callback_lock);
1269 
1270 	/* Bind to all addresses. */
1271 	if (sctp_bind_addrs(con, dlm_config.ci_tcp_port))
1272 		goto create_delsock;
1273 
1274 	result = sock->ops->listen(sock, 5);
1275 	if (result < 0) {
1276 		log_print("Can't set socket listening");
1277 		goto create_delsock;
1278 	}
1279 
1280 	return 0;
1281 
1282 create_delsock:
1283 	sock_release(sock);
1284 	con->sock = NULL;
1285 out:
1286 	return result;
1287 }
1288 
tcp_listen_for_all(void)1289 static int tcp_listen_for_all(void)
1290 {
1291 	struct socket *sock = NULL;
1292 	struct connection *con = nodeid2con(0, GFP_NOFS);
1293 	int result = -EINVAL;
1294 
1295 	if (!con)
1296 		return -ENOMEM;
1297 
1298 	/* We don't support multi-homed hosts */
1299 	if (dlm_local_addr[1] != NULL) {
1300 		log_print("TCP protocol can't handle multi-homed hosts, "
1301 			  "try SCTP");
1302 		return -EINVAL;
1303 	}
1304 
1305 	log_print("Using TCP for communications");
1306 
1307 	sock = tcp_create_listen_sock(con, dlm_local_addr[0]);
1308 	if (sock) {
1309 		add_sock(sock, con);
1310 		result = 0;
1311 	}
1312 	else {
1313 		result = -EADDRINUSE;
1314 	}
1315 
1316 	return result;
1317 }
1318 
1319 
1320 
new_writequeue_entry(struct connection * con,gfp_t allocation)1321 static struct writequeue_entry *new_writequeue_entry(struct connection *con,
1322 						     gfp_t allocation)
1323 {
1324 	struct writequeue_entry *entry;
1325 
1326 	entry = kmalloc(sizeof(struct writequeue_entry), allocation);
1327 	if (!entry)
1328 		return NULL;
1329 
1330 	entry->page = alloc_page(allocation);
1331 	if (!entry->page) {
1332 		kfree(entry);
1333 		return NULL;
1334 	}
1335 
1336 	entry->offset = 0;
1337 	entry->len = 0;
1338 	entry->end = 0;
1339 	entry->users = 0;
1340 	entry->con = con;
1341 
1342 	return entry;
1343 }
1344 
dlm_lowcomms_get_buffer(int nodeid,int len,gfp_t allocation,char ** ppc)1345 void *dlm_lowcomms_get_buffer(int nodeid, int len, gfp_t allocation, char **ppc)
1346 {
1347 	struct connection *con;
1348 	struct writequeue_entry *e;
1349 	int offset = 0;
1350 
1351 	con = nodeid2con(nodeid, allocation);
1352 	if (!con)
1353 		return NULL;
1354 
1355 	spin_lock(&con->writequeue_lock);
1356 	e = list_entry(con->writequeue.prev, struct writequeue_entry, list);
1357 	if ((&e->list == &con->writequeue) ||
1358 	    (PAGE_SIZE - e->end < len)) {
1359 		e = NULL;
1360 	} else {
1361 		offset = e->end;
1362 		e->end += len;
1363 		e->users++;
1364 	}
1365 	spin_unlock(&con->writequeue_lock);
1366 
1367 	if (e) {
1368 	got_one:
1369 		*ppc = page_address(e->page) + offset;
1370 		return e;
1371 	}
1372 
1373 	e = new_writequeue_entry(con, allocation);
1374 	if (e) {
1375 		spin_lock(&con->writequeue_lock);
1376 		offset = e->end;
1377 		e->end += len;
1378 		e->users++;
1379 		list_add_tail(&e->list, &con->writequeue);
1380 		spin_unlock(&con->writequeue_lock);
1381 		goto got_one;
1382 	}
1383 	return NULL;
1384 }
1385 
dlm_lowcomms_commit_buffer(void * mh)1386 void dlm_lowcomms_commit_buffer(void *mh)
1387 {
1388 	struct writequeue_entry *e = (struct writequeue_entry *)mh;
1389 	struct connection *con = e->con;
1390 	int users;
1391 
1392 	spin_lock(&con->writequeue_lock);
1393 	users = --e->users;
1394 	if (users)
1395 		goto out;
1396 	e->len = e->end - e->offset;
1397 	spin_unlock(&con->writequeue_lock);
1398 
1399 	queue_work(send_workqueue, &con->swork);
1400 	return;
1401 
1402 out:
1403 	spin_unlock(&con->writequeue_lock);
1404 	return;
1405 }
1406 
1407 /* Send a message */
send_to_sock(struct connection * con)1408 static void send_to_sock(struct connection *con)
1409 {
1410 	int ret = 0;
1411 	const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
1412 	struct writequeue_entry *e;
1413 	int len, offset;
1414 	int count = 0;
1415 
1416 	mutex_lock(&con->sock_mutex);
1417 	if (con->sock == NULL)
1418 		goto out_connect;
1419 
1420 	spin_lock(&con->writequeue_lock);
1421 	for (;;) {
1422 		e = list_entry(con->writequeue.next, struct writequeue_entry,
1423 			       list);
1424 		if ((struct list_head *) e == &con->writequeue)
1425 			break;
1426 
1427 		len = e->len;
1428 		offset = e->offset;
1429 		BUG_ON(len == 0 && e->users == 0);
1430 		spin_unlock(&con->writequeue_lock);
1431 
1432 		ret = 0;
1433 		if (len) {
1434 			ret = kernel_sendpage(con->sock, e->page, offset, len,
1435 					      msg_flags);
1436 			if (ret == -EAGAIN || ret == 0) {
1437 				if (ret == -EAGAIN &&
1438 				    test_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags) &&
1439 				    !test_and_set_bit(CF_APP_LIMITED, &con->flags)) {
1440 					/* Notify TCP that we're limited by the
1441 					 * application window size.
1442 					 */
1443 					set_bit(SOCK_NOSPACE, &con->sock->flags);
1444 					con->sock->sk->sk_write_pending++;
1445 				}
1446 				cond_resched();
1447 				goto out;
1448 			} else if (ret < 0)
1449 				goto send_error;
1450 		}
1451 
1452 		/* Don't starve people filling buffers */
1453 		if (++count >= MAX_SEND_MSG_COUNT) {
1454 			cond_resched();
1455 			count = 0;
1456 		}
1457 
1458 		spin_lock(&con->writequeue_lock);
1459 		writequeue_entry_complete(e, ret);
1460 	}
1461 	spin_unlock(&con->writequeue_lock);
1462 out:
1463 	mutex_unlock(&con->sock_mutex);
1464 	return;
1465 
1466 send_error:
1467 	mutex_unlock(&con->sock_mutex);
1468 	close_connection(con, false, false, true);
1469 	/* Requeue the send work. When the work daemon runs again, it will try
1470 	   a new connection, then call this function again. */
1471 	queue_work(send_workqueue, &con->swork);
1472 	return;
1473 
1474 out_connect:
1475 	mutex_unlock(&con->sock_mutex);
1476 	queue_work(send_workqueue, &con->swork);
1477 	cond_resched();
1478 }
1479 
clean_one_writequeue(struct connection * con)1480 static void clean_one_writequeue(struct connection *con)
1481 {
1482 	struct writequeue_entry *e, *safe;
1483 
1484 	spin_lock(&con->writequeue_lock);
1485 	list_for_each_entry_safe(e, safe, &con->writequeue, list) {
1486 		list_del(&e->list);
1487 		free_entry(e);
1488 	}
1489 	spin_unlock(&con->writequeue_lock);
1490 }
1491 
1492 /* Called from recovery when it knows that a node has
1493    left the cluster */
dlm_lowcomms_close(int nodeid)1494 int dlm_lowcomms_close(int nodeid)
1495 {
1496 	struct connection *con;
1497 	struct dlm_node_addr *na;
1498 
1499 	log_print("closing connection to node %d", nodeid);
1500 	con = nodeid2con(nodeid, 0);
1501 	if (con) {
1502 		set_bit(CF_CLOSE, &con->flags);
1503 		close_connection(con, true, true, true);
1504 		clean_one_writequeue(con);
1505 	}
1506 
1507 	spin_lock(&dlm_node_addrs_spin);
1508 	na = find_node_addr(nodeid);
1509 	if (na) {
1510 		list_del(&na->list);
1511 		while (na->addr_count--)
1512 			kfree(na->addr[na->addr_count]);
1513 		kfree(na);
1514 	}
1515 	spin_unlock(&dlm_node_addrs_spin);
1516 
1517 	return 0;
1518 }
1519 
1520 /* Receive workqueue function */
process_recv_sockets(struct work_struct * work)1521 static void process_recv_sockets(struct work_struct *work)
1522 {
1523 	struct connection *con = container_of(work, struct connection, rwork);
1524 	int err;
1525 
1526 	clear_bit(CF_READ_PENDING, &con->flags);
1527 	do {
1528 		err = con->rx_action(con);
1529 	} while (!err);
1530 }
1531 
1532 /* Send workqueue function */
process_send_sockets(struct work_struct * work)1533 static void process_send_sockets(struct work_struct *work)
1534 {
1535 	struct connection *con = container_of(work, struct connection, swork);
1536 
1537 	clear_bit(CF_WRITE_PENDING, &con->flags);
1538 	if (con->sock == NULL) /* not mutex protected so check it inside too */
1539 		con->connect_action(con);
1540 	if (!list_empty(&con->writequeue))
1541 		send_to_sock(con);
1542 }
1543 
work_stop(void)1544 static void work_stop(void)
1545 {
1546 	if (recv_workqueue)
1547 		destroy_workqueue(recv_workqueue);
1548 	if (send_workqueue)
1549 		destroy_workqueue(send_workqueue);
1550 }
1551 
work_start(void)1552 static int work_start(void)
1553 {
1554 	recv_workqueue = alloc_workqueue("dlm_recv",
1555 					 WQ_UNBOUND | WQ_MEM_RECLAIM, 1);
1556 	if (!recv_workqueue) {
1557 		log_print("can't start dlm_recv");
1558 		return -ENOMEM;
1559 	}
1560 
1561 	send_workqueue = alloc_workqueue("dlm_send",
1562 					 WQ_UNBOUND | WQ_MEM_RECLAIM, 1);
1563 	if (!send_workqueue) {
1564 		log_print("can't start dlm_send");
1565 		destroy_workqueue(recv_workqueue);
1566 		return -ENOMEM;
1567 	}
1568 
1569 	return 0;
1570 }
1571 
_stop_conn(struct connection * con,bool and_other)1572 static void _stop_conn(struct connection *con, bool and_other)
1573 {
1574 	mutex_lock(&con->sock_mutex);
1575 	set_bit(CF_CLOSE, &con->flags);
1576 	set_bit(CF_READ_PENDING, &con->flags);
1577 	set_bit(CF_WRITE_PENDING, &con->flags);
1578 	if (con->sock && con->sock->sk) {
1579 		write_lock_bh(&con->sock->sk->sk_callback_lock);
1580 		con->sock->sk->sk_user_data = NULL;
1581 		write_unlock_bh(&con->sock->sk->sk_callback_lock);
1582 	}
1583 	if (con->othercon && and_other)
1584 		_stop_conn(con->othercon, false);
1585 	mutex_unlock(&con->sock_mutex);
1586 }
1587 
stop_conn(struct connection * con)1588 static void stop_conn(struct connection *con)
1589 {
1590 	_stop_conn(con, true);
1591 }
1592 
shutdown_conn(struct connection * con)1593 static void shutdown_conn(struct connection *con)
1594 {
1595 	if (con->shutdown_action)
1596 		con->shutdown_action(con);
1597 }
1598 
connection_release(struct rcu_head * rcu)1599 static void connection_release(struct rcu_head *rcu)
1600 {
1601 	struct connection *con = container_of(rcu, struct connection, rcu);
1602 
1603 	kfree(con->rx_buf);
1604 	kfree(con);
1605 }
1606 
free_conn(struct connection * con)1607 static void free_conn(struct connection *con)
1608 {
1609 	close_connection(con, true, true, true);
1610 	spin_lock(&connections_lock);
1611 	hlist_del_rcu(&con->list);
1612 	spin_unlock(&connections_lock);
1613 	if (con->othercon) {
1614 		clean_one_writequeue(con->othercon);
1615 		call_rcu(&con->othercon->rcu, connection_release);
1616 	}
1617 	clean_one_writequeue(con);
1618 	call_rcu(&con->rcu, connection_release);
1619 }
1620 
work_flush(void)1621 static void work_flush(void)
1622 {
1623 	int ok, idx;
1624 	int i;
1625 	struct connection *con;
1626 
1627 	do {
1628 		ok = 1;
1629 		foreach_conn(stop_conn);
1630 		if (recv_workqueue)
1631 			flush_workqueue(recv_workqueue);
1632 		if (send_workqueue)
1633 			flush_workqueue(send_workqueue);
1634 		idx = srcu_read_lock(&connections_srcu);
1635 		for (i = 0; i < CONN_HASH_SIZE && ok; i++) {
1636 			hlist_for_each_entry_rcu(con, &connection_hash[i],
1637 						 list) {
1638 				ok &= test_bit(CF_READ_PENDING, &con->flags);
1639 				ok &= test_bit(CF_WRITE_PENDING, &con->flags);
1640 				if (con->othercon) {
1641 					ok &= test_bit(CF_READ_PENDING,
1642 						       &con->othercon->flags);
1643 					ok &= test_bit(CF_WRITE_PENDING,
1644 						       &con->othercon->flags);
1645 				}
1646 			}
1647 		}
1648 		srcu_read_unlock(&connections_srcu, idx);
1649 	} while (!ok);
1650 }
1651 
dlm_lowcomms_stop(void)1652 void dlm_lowcomms_stop(void)
1653 {
1654 	/* Set all the flags to prevent any
1655 	   socket activity.
1656 	*/
1657 	dlm_allow_conn = 0;
1658 
1659 	if (recv_workqueue)
1660 		flush_workqueue(recv_workqueue);
1661 	if (send_workqueue)
1662 		flush_workqueue(send_workqueue);
1663 
1664 	foreach_conn(shutdown_conn);
1665 	work_flush();
1666 	foreach_conn(free_conn);
1667 	work_stop();
1668 	deinit_local();
1669 }
1670 
dlm_lowcomms_start(void)1671 int dlm_lowcomms_start(void)
1672 {
1673 	int error = -EINVAL;
1674 	struct connection *con;
1675 	int i;
1676 
1677 	for (i = 0; i < CONN_HASH_SIZE; i++)
1678 		INIT_HLIST_HEAD(&connection_hash[i]);
1679 
1680 	init_local();
1681 	if (!dlm_local_count) {
1682 		error = -ENOTCONN;
1683 		log_print("no local IP address has been set");
1684 		goto fail;
1685 	}
1686 
1687 	error = work_start();
1688 	if (error)
1689 		goto fail;
1690 
1691 	dlm_allow_conn = 1;
1692 
1693 	/* Start listening */
1694 	if (dlm_config.ci_protocol == 0)
1695 		error = tcp_listen_for_all();
1696 	else
1697 		error = sctp_listen_for_all();
1698 	if (error)
1699 		goto fail_unlisten;
1700 
1701 	return 0;
1702 
1703 fail_unlisten:
1704 	dlm_allow_conn = 0;
1705 	con = nodeid2con(0,0);
1706 	if (con)
1707 		free_conn(con);
1708 fail:
1709 	return error;
1710 }
1711 
dlm_lowcomms_exit(void)1712 void dlm_lowcomms_exit(void)
1713 {
1714 	struct dlm_node_addr *na, *safe;
1715 
1716 	spin_lock(&dlm_node_addrs_spin);
1717 	list_for_each_entry_safe(na, safe, &dlm_node_addrs, list) {
1718 		list_del(&na->list);
1719 		while (na->addr_count--)
1720 			kfree(na->addr[na->addr_count]);
1721 		kfree(na);
1722 	}
1723 	spin_unlock(&dlm_node_addrs_spin);
1724 }
1725