xref: /openbmc/linux/net/ceph/messenger.c (revision 2a598d0b)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
3 
4 #include <linux/crc32c.h>
5 #include <linux/ctype.h>
6 #include <linux/highmem.h>
7 #include <linux/inet.h>
8 #include <linux/kthread.h>
9 #include <linux/net.h>
10 #include <linux/nsproxy.h>
11 #include <linux/sched/mm.h>
12 #include <linux/slab.h>
13 #include <linux/socket.h>
14 #include <linux/string.h>
15 #ifdef	CONFIG_BLOCK
16 #include <linux/bio.h>
17 #endif	/* CONFIG_BLOCK */
18 #include <linux/dns_resolver.h>
19 #include <net/tcp.h>
20 #include <trace/events/sock.h>
21 
22 #include <linux/ceph/ceph_features.h>
23 #include <linux/ceph/libceph.h>
24 #include <linux/ceph/messenger.h>
25 #include <linux/ceph/decode.h>
26 #include <linux/ceph/pagelist.h>
27 #include <linux/export.h>
28 
29 /*
30  * Ceph uses the messenger to exchange ceph_msg messages with other
31  * hosts in the system.  The messenger provides ordered and reliable
32  * delivery.  We tolerate TCP disconnects by reconnecting (with
33  * exponential backoff) in the case of a fault (disconnection, bad
34  * crc, protocol error).  Acks allow sent messages to be discarded by
35  * the sender.
36  */
37 
38 /*
39  * We track the state of the socket on a given connection using
40  * values defined below.  The transition to a new socket state is
41  * handled by a function which verifies we aren't coming from an
42  * unexpected state.
43  *
44  *      --------
45  *      | NEW* |  transient initial state
46  *      --------
47  *          | con_sock_state_init()
48  *          v
49  *      ----------
50  *      | CLOSED |  initialized, but no socket (and no
51  *      ----------  TCP connection)
52  *       ^      \
53  *       |       \ con_sock_state_connecting()
54  *       |        ----------------------
55  *       |                              \
56  *       + con_sock_state_closed()       \
57  *       |+---------------------------    \
58  *       | \                          \    \
59  *       |  -----------                \    \
60  *       |  | CLOSING |  socket event;  \    \
61  *       |  -----------  await close     \    \
62  *       |       ^                        \   |
63  *       |       |                         \  |
64  *       |       + con_sock_state_closing() \ |
65  *       |      / \                         | |
66  *       |     /   ---------------          | |
67  *       |    /                   \         v v
68  *       |   /                    --------------
69  *       |  /    -----------------| CONNECTING |  socket created, TCP
70  *       |  |   /                 --------------  connect initiated
71  *       |  |   | con_sock_state_connected()
72  *       |  |   v
73  *      -------------
74  *      | CONNECTED |  TCP connection established
75  *      -------------
76  *
77  * State values for ceph_connection->sock_state; NEW is assumed to be 0.
78  */
79 
80 #define CON_SOCK_STATE_NEW		0	/* -> CLOSED */
81 #define CON_SOCK_STATE_CLOSED		1	/* -> CONNECTING */
82 #define CON_SOCK_STATE_CONNECTING	2	/* -> CONNECTED or -> CLOSING */
83 #define CON_SOCK_STATE_CONNECTED	3	/* -> CLOSING or -> CLOSED */
84 #define CON_SOCK_STATE_CLOSING		4	/* -> CLOSED */
85 
86 static bool con_flag_valid(unsigned long con_flag)
87 {
88 	switch (con_flag) {
89 	case CEPH_CON_F_LOSSYTX:
90 	case CEPH_CON_F_KEEPALIVE_PENDING:
91 	case CEPH_CON_F_WRITE_PENDING:
92 	case CEPH_CON_F_SOCK_CLOSED:
93 	case CEPH_CON_F_BACKOFF:
94 		return true;
95 	default:
96 		return false;
97 	}
98 }
99 
100 void ceph_con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
101 {
102 	BUG_ON(!con_flag_valid(con_flag));
103 
104 	clear_bit(con_flag, &con->flags);
105 }
106 
107 void ceph_con_flag_set(struct ceph_connection *con, unsigned long con_flag)
108 {
109 	BUG_ON(!con_flag_valid(con_flag));
110 
111 	set_bit(con_flag, &con->flags);
112 }
113 
114 bool ceph_con_flag_test(struct ceph_connection *con, unsigned long con_flag)
115 {
116 	BUG_ON(!con_flag_valid(con_flag));
117 
118 	return test_bit(con_flag, &con->flags);
119 }
120 
121 bool ceph_con_flag_test_and_clear(struct ceph_connection *con,
122 				  unsigned long con_flag)
123 {
124 	BUG_ON(!con_flag_valid(con_flag));
125 
126 	return test_and_clear_bit(con_flag, &con->flags);
127 }
128 
129 bool ceph_con_flag_test_and_set(struct ceph_connection *con,
130 				unsigned long con_flag)
131 {
132 	BUG_ON(!con_flag_valid(con_flag));
133 
134 	return test_and_set_bit(con_flag, &con->flags);
135 }
136 
137 /* Slab caches for frequently-allocated structures */
138 
139 static struct kmem_cache	*ceph_msg_cache;
140 
141 #ifdef CONFIG_LOCKDEP
142 static struct lock_class_key socket_class;
143 #endif
144 
145 static void queue_con(struct ceph_connection *con);
146 static void cancel_con(struct ceph_connection *con);
147 static void ceph_con_workfn(struct work_struct *);
148 static void con_fault(struct ceph_connection *con);
149 
150 /*
151  * Nicely render a sockaddr as a string.  An array of formatted
152  * strings is used, to approximate reentrancy.
153  */
154 #define ADDR_STR_COUNT_LOG	5	/* log2(# address strings in array) */
155 #define ADDR_STR_COUNT		(1 << ADDR_STR_COUNT_LOG)
156 #define ADDR_STR_COUNT_MASK	(ADDR_STR_COUNT - 1)
157 #define MAX_ADDR_STR_LEN	64	/* 54 is enough */
158 
159 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
160 static atomic_t addr_str_seq = ATOMIC_INIT(0);
161 
162 struct page *ceph_zero_page;		/* used in certain error cases */
163 
164 const char *ceph_pr_addr(const struct ceph_entity_addr *addr)
165 {
166 	int i;
167 	char *s;
168 	struct sockaddr_storage ss = addr->in_addr; /* align */
169 	struct sockaddr_in *in4 = (struct sockaddr_in *)&ss;
170 	struct sockaddr_in6 *in6 = (struct sockaddr_in6 *)&ss;
171 
172 	i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
173 	s = addr_str[i];
174 
175 	switch (ss.ss_family) {
176 	case AF_INET:
177 		snprintf(s, MAX_ADDR_STR_LEN, "(%d)%pI4:%hu",
178 			 le32_to_cpu(addr->type), &in4->sin_addr,
179 			 ntohs(in4->sin_port));
180 		break;
181 
182 	case AF_INET6:
183 		snprintf(s, MAX_ADDR_STR_LEN, "(%d)[%pI6c]:%hu",
184 			 le32_to_cpu(addr->type), &in6->sin6_addr,
185 			 ntohs(in6->sin6_port));
186 		break;
187 
188 	default:
189 		snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
190 			 ss.ss_family);
191 	}
192 
193 	return s;
194 }
195 EXPORT_SYMBOL(ceph_pr_addr);
196 
197 void ceph_encode_my_addr(struct ceph_messenger *msgr)
198 {
199 	if (!ceph_msgr2(from_msgr(msgr))) {
200 		memcpy(&msgr->my_enc_addr, &msgr->inst.addr,
201 		       sizeof(msgr->my_enc_addr));
202 		ceph_encode_banner_addr(&msgr->my_enc_addr);
203 	}
204 }
205 
206 /*
207  * work queue for all reading and writing to/from the socket.
208  */
209 static struct workqueue_struct *ceph_msgr_wq;
210 
211 static int ceph_msgr_slab_init(void)
212 {
213 	BUG_ON(ceph_msg_cache);
214 	ceph_msg_cache = KMEM_CACHE(ceph_msg, 0);
215 	if (!ceph_msg_cache)
216 		return -ENOMEM;
217 
218 	return 0;
219 }
220 
221 static void ceph_msgr_slab_exit(void)
222 {
223 	BUG_ON(!ceph_msg_cache);
224 	kmem_cache_destroy(ceph_msg_cache);
225 	ceph_msg_cache = NULL;
226 }
227 
228 static void _ceph_msgr_exit(void)
229 {
230 	if (ceph_msgr_wq) {
231 		destroy_workqueue(ceph_msgr_wq);
232 		ceph_msgr_wq = NULL;
233 	}
234 
235 	BUG_ON(!ceph_zero_page);
236 	put_page(ceph_zero_page);
237 	ceph_zero_page = NULL;
238 
239 	ceph_msgr_slab_exit();
240 }
241 
242 int __init ceph_msgr_init(void)
243 {
244 	if (ceph_msgr_slab_init())
245 		return -ENOMEM;
246 
247 	BUG_ON(ceph_zero_page);
248 	ceph_zero_page = ZERO_PAGE(0);
249 	get_page(ceph_zero_page);
250 
251 	/*
252 	 * The number of active work items is limited by the number of
253 	 * connections, so leave @max_active at default.
254 	 */
255 	ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
256 	if (ceph_msgr_wq)
257 		return 0;
258 
259 	pr_err("msgr_init failed to create workqueue\n");
260 	_ceph_msgr_exit();
261 
262 	return -ENOMEM;
263 }
264 
265 void ceph_msgr_exit(void)
266 {
267 	BUG_ON(ceph_msgr_wq == NULL);
268 
269 	_ceph_msgr_exit();
270 }
271 
272 void ceph_msgr_flush(void)
273 {
274 	flush_workqueue(ceph_msgr_wq);
275 }
276 EXPORT_SYMBOL(ceph_msgr_flush);
277 
278 /* Connection socket state transition functions */
279 
280 static void con_sock_state_init(struct ceph_connection *con)
281 {
282 	int old_state;
283 
284 	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
285 	if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
286 		printk("%s: unexpected old state %d\n", __func__, old_state);
287 	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
288 	     CON_SOCK_STATE_CLOSED);
289 }
290 
291 static void con_sock_state_connecting(struct ceph_connection *con)
292 {
293 	int old_state;
294 
295 	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
296 	if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
297 		printk("%s: unexpected old state %d\n", __func__, old_state);
298 	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
299 	     CON_SOCK_STATE_CONNECTING);
300 }
301 
302 static void con_sock_state_connected(struct ceph_connection *con)
303 {
304 	int old_state;
305 
306 	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
307 	if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
308 		printk("%s: unexpected old state %d\n", __func__, old_state);
309 	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
310 	     CON_SOCK_STATE_CONNECTED);
311 }
312 
313 static void con_sock_state_closing(struct ceph_connection *con)
314 {
315 	int old_state;
316 
317 	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
318 	if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
319 			old_state != CON_SOCK_STATE_CONNECTED &&
320 			old_state != CON_SOCK_STATE_CLOSING))
321 		printk("%s: unexpected old state %d\n", __func__, old_state);
322 	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
323 	     CON_SOCK_STATE_CLOSING);
324 }
325 
326 static void con_sock_state_closed(struct ceph_connection *con)
327 {
328 	int old_state;
329 
330 	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
331 	if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
332 		    old_state != CON_SOCK_STATE_CLOSING &&
333 		    old_state != CON_SOCK_STATE_CONNECTING &&
334 		    old_state != CON_SOCK_STATE_CLOSED))
335 		printk("%s: unexpected old state %d\n", __func__, old_state);
336 	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
337 	     CON_SOCK_STATE_CLOSED);
338 }
339 
340 /*
341  * socket callback functions
342  */
343 
344 /* data available on socket, or listen socket received a connect */
345 static void ceph_sock_data_ready(struct sock *sk)
346 {
347 	struct ceph_connection *con = sk->sk_user_data;
348 
349 	trace_sk_data_ready(sk);
350 
351 	if (atomic_read(&con->msgr->stopping)) {
352 		return;
353 	}
354 
355 	if (sk->sk_state != TCP_CLOSE_WAIT) {
356 		dout("%s %p state = %d, queueing work\n", __func__,
357 		     con, con->state);
358 		queue_con(con);
359 	}
360 }
361 
362 /* socket has buffer space for writing */
363 static void ceph_sock_write_space(struct sock *sk)
364 {
365 	struct ceph_connection *con = sk->sk_user_data;
366 
367 	/* only queue to workqueue if there is data we want to write,
368 	 * and there is sufficient space in the socket buffer to accept
369 	 * more data.  clear SOCK_NOSPACE so that ceph_sock_write_space()
370 	 * doesn't get called again until try_write() fills the socket
371 	 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
372 	 * and net/core/stream.c:sk_stream_write_space().
373 	 */
374 	if (ceph_con_flag_test(con, CEPH_CON_F_WRITE_PENDING)) {
375 		if (sk_stream_is_writeable(sk)) {
376 			dout("%s %p queueing write work\n", __func__, con);
377 			clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
378 			queue_con(con);
379 		}
380 	} else {
381 		dout("%s %p nothing to write\n", __func__, con);
382 	}
383 }
384 
385 /* socket's state has changed */
386 static void ceph_sock_state_change(struct sock *sk)
387 {
388 	struct ceph_connection *con = sk->sk_user_data;
389 
390 	dout("%s %p state = %d sk_state = %u\n", __func__,
391 	     con, con->state, sk->sk_state);
392 
393 	switch (sk->sk_state) {
394 	case TCP_CLOSE:
395 		dout("%s TCP_CLOSE\n", __func__);
396 		fallthrough;
397 	case TCP_CLOSE_WAIT:
398 		dout("%s TCP_CLOSE_WAIT\n", __func__);
399 		con_sock_state_closing(con);
400 		ceph_con_flag_set(con, CEPH_CON_F_SOCK_CLOSED);
401 		queue_con(con);
402 		break;
403 	case TCP_ESTABLISHED:
404 		dout("%s TCP_ESTABLISHED\n", __func__);
405 		con_sock_state_connected(con);
406 		queue_con(con);
407 		break;
408 	default:	/* Everything else is uninteresting */
409 		break;
410 	}
411 }
412 
413 /*
414  * set up socket callbacks
415  */
416 static void set_sock_callbacks(struct socket *sock,
417 			       struct ceph_connection *con)
418 {
419 	struct sock *sk = sock->sk;
420 	sk->sk_user_data = con;
421 	sk->sk_data_ready = ceph_sock_data_ready;
422 	sk->sk_write_space = ceph_sock_write_space;
423 	sk->sk_state_change = ceph_sock_state_change;
424 }
425 
426 
427 /*
428  * socket helpers
429  */
430 
431 /*
432  * initiate connection to a remote socket.
433  */
434 int ceph_tcp_connect(struct ceph_connection *con)
435 {
436 	struct sockaddr_storage ss = con->peer_addr.in_addr; /* align */
437 	struct socket *sock;
438 	unsigned int noio_flag;
439 	int ret;
440 
441 	dout("%s con %p peer_addr %s\n", __func__, con,
442 	     ceph_pr_addr(&con->peer_addr));
443 	BUG_ON(con->sock);
444 
445 	/* sock_create_kern() allocates with GFP_KERNEL */
446 	noio_flag = memalloc_noio_save();
447 	ret = sock_create_kern(read_pnet(&con->msgr->net), ss.ss_family,
448 			       SOCK_STREAM, IPPROTO_TCP, &sock);
449 	memalloc_noio_restore(noio_flag);
450 	if (ret)
451 		return ret;
452 	sock->sk->sk_allocation = GFP_NOFS;
453 	sock->sk->sk_use_task_frag = false;
454 
455 #ifdef CONFIG_LOCKDEP
456 	lockdep_set_class(&sock->sk->sk_lock, &socket_class);
457 #endif
458 
459 	set_sock_callbacks(sock, con);
460 
461 	con_sock_state_connecting(con);
462 	ret = sock->ops->connect(sock, (struct sockaddr *)&ss, sizeof(ss),
463 				 O_NONBLOCK);
464 	if (ret == -EINPROGRESS) {
465 		dout("connect %s EINPROGRESS sk_state = %u\n",
466 		     ceph_pr_addr(&con->peer_addr),
467 		     sock->sk->sk_state);
468 	} else if (ret < 0) {
469 		pr_err("connect %s error %d\n",
470 		       ceph_pr_addr(&con->peer_addr), ret);
471 		sock_release(sock);
472 		return ret;
473 	}
474 
475 	if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY))
476 		tcp_sock_set_nodelay(sock->sk);
477 
478 	con->sock = sock;
479 	return 0;
480 }
481 
482 /*
483  * Shutdown/close the socket for the given connection.
484  */
485 int ceph_con_close_socket(struct ceph_connection *con)
486 {
487 	int rc = 0;
488 
489 	dout("%s con %p sock %p\n", __func__, con, con->sock);
490 	if (con->sock) {
491 		rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
492 		sock_release(con->sock);
493 		con->sock = NULL;
494 	}
495 
496 	/*
497 	 * Forcibly clear the SOCK_CLOSED flag.  It gets set
498 	 * independent of the connection mutex, and we could have
499 	 * received a socket close event before we had the chance to
500 	 * shut the socket down.
501 	 */
502 	ceph_con_flag_clear(con, CEPH_CON_F_SOCK_CLOSED);
503 
504 	con_sock_state_closed(con);
505 	return rc;
506 }
507 
508 static void ceph_con_reset_protocol(struct ceph_connection *con)
509 {
510 	dout("%s con %p\n", __func__, con);
511 
512 	ceph_con_close_socket(con);
513 	if (con->in_msg) {
514 		WARN_ON(con->in_msg->con != con);
515 		ceph_msg_put(con->in_msg);
516 		con->in_msg = NULL;
517 	}
518 	if (con->out_msg) {
519 		WARN_ON(con->out_msg->con != con);
520 		ceph_msg_put(con->out_msg);
521 		con->out_msg = NULL;
522 	}
523 	if (con->bounce_page) {
524 		__free_page(con->bounce_page);
525 		con->bounce_page = NULL;
526 	}
527 
528 	if (ceph_msgr2(from_msgr(con->msgr)))
529 		ceph_con_v2_reset_protocol(con);
530 	else
531 		ceph_con_v1_reset_protocol(con);
532 }
533 
534 /*
535  * Reset a connection.  Discard all incoming and outgoing messages
536  * and clear *_seq state.
537  */
538 static void ceph_msg_remove(struct ceph_msg *msg)
539 {
540 	list_del_init(&msg->list_head);
541 
542 	ceph_msg_put(msg);
543 }
544 
545 static void ceph_msg_remove_list(struct list_head *head)
546 {
547 	while (!list_empty(head)) {
548 		struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
549 							list_head);
550 		ceph_msg_remove(msg);
551 	}
552 }
553 
554 void ceph_con_reset_session(struct ceph_connection *con)
555 {
556 	dout("%s con %p\n", __func__, con);
557 
558 	WARN_ON(con->in_msg);
559 	WARN_ON(con->out_msg);
560 	ceph_msg_remove_list(&con->out_queue);
561 	ceph_msg_remove_list(&con->out_sent);
562 	con->out_seq = 0;
563 	con->in_seq = 0;
564 	con->in_seq_acked = 0;
565 
566 	if (ceph_msgr2(from_msgr(con->msgr)))
567 		ceph_con_v2_reset_session(con);
568 	else
569 		ceph_con_v1_reset_session(con);
570 }
571 
572 /*
573  * mark a peer down.  drop any open connections.
574  */
575 void ceph_con_close(struct ceph_connection *con)
576 {
577 	mutex_lock(&con->mutex);
578 	dout("con_close %p peer %s\n", con, ceph_pr_addr(&con->peer_addr));
579 	con->state = CEPH_CON_S_CLOSED;
580 
581 	ceph_con_flag_clear(con, CEPH_CON_F_LOSSYTX);  /* so we retry next
582 							  connect */
583 	ceph_con_flag_clear(con, CEPH_CON_F_KEEPALIVE_PENDING);
584 	ceph_con_flag_clear(con, CEPH_CON_F_WRITE_PENDING);
585 	ceph_con_flag_clear(con, CEPH_CON_F_BACKOFF);
586 
587 	ceph_con_reset_protocol(con);
588 	ceph_con_reset_session(con);
589 	cancel_con(con);
590 	mutex_unlock(&con->mutex);
591 }
592 EXPORT_SYMBOL(ceph_con_close);
593 
594 /*
595  * Reopen a closed connection, with a new peer address.
596  */
597 void ceph_con_open(struct ceph_connection *con,
598 		   __u8 entity_type, __u64 entity_num,
599 		   struct ceph_entity_addr *addr)
600 {
601 	mutex_lock(&con->mutex);
602 	dout("con_open %p %s\n", con, ceph_pr_addr(addr));
603 
604 	WARN_ON(con->state != CEPH_CON_S_CLOSED);
605 	con->state = CEPH_CON_S_PREOPEN;
606 
607 	con->peer_name.type = (__u8) entity_type;
608 	con->peer_name.num = cpu_to_le64(entity_num);
609 
610 	memcpy(&con->peer_addr, addr, sizeof(*addr));
611 	con->delay = 0;      /* reset backoff memory */
612 	mutex_unlock(&con->mutex);
613 	queue_con(con);
614 }
615 EXPORT_SYMBOL(ceph_con_open);
616 
617 /*
618  * return true if this connection ever successfully opened
619  */
620 bool ceph_con_opened(struct ceph_connection *con)
621 {
622 	if (ceph_msgr2(from_msgr(con->msgr)))
623 		return ceph_con_v2_opened(con);
624 
625 	return ceph_con_v1_opened(con);
626 }
627 
628 /*
629  * initialize a new connection.
630  */
631 void ceph_con_init(struct ceph_connection *con, void *private,
632 	const struct ceph_connection_operations *ops,
633 	struct ceph_messenger *msgr)
634 {
635 	dout("con_init %p\n", con);
636 	memset(con, 0, sizeof(*con));
637 	con->private = private;
638 	con->ops = ops;
639 	con->msgr = msgr;
640 
641 	con_sock_state_init(con);
642 
643 	mutex_init(&con->mutex);
644 	INIT_LIST_HEAD(&con->out_queue);
645 	INIT_LIST_HEAD(&con->out_sent);
646 	INIT_DELAYED_WORK(&con->work, ceph_con_workfn);
647 
648 	con->state = CEPH_CON_S_CLOSED;
649 }
650 EXPORT_SYMBOL(ceph_con_init);
651 
652 /*
653  * We maintain a global counter to order connection attempts.  Get
654  * a unique seq greater than @gt.
655  */
656 u32 ceph_get_global_seq(struct ceph_messenger *msgr, u32 gt)
657 {
658 	u32 ret;
659 
660 	spin_lock(&msgr->global_seq_lock);
661 	if (msgr->global_seq < gt)
662 		msgr->global_seq = gt;
663 	ret = ++msgr->global_seq;
664 	spin_unlock(&msgr->global_seq_lock);
665 	return ret;
666 }
667 
668 /*
669  * Discard messages that have been acked by the server.
670  */
671 void ceph_con_discard_sent(struct ceph_connection *con, u64 ack_seq)
672 {
673 	struct ceph_msg *msg;
674 	u64 seq;
675 
676 	dout("%s con %p ack_seq %llu\n", __func__, con, ack_seq);
677 	while (!list_empty(&con->out_sent)) {
678 		msg = list_first_entry(&con->out_sent, struct ceph_msg,
679 				       list_head);
680 		WARN_ON(msg->needs_out_seq);
681 		seq = le64_to_cpu(msg->hdr.seq);
682 		if (seq > ack_seq)
683 			break;
684 
685 		dout("%s con %p discarding msg %p seq %llu\n", __func__, con,
686 		     msg, seq);
687 		ceph_msg_remove(msg);
688 	}
689 }
690 
691 /*
692  * Discard messages that have been requeued in con_fault(), up to
693  * reconnect_seq.  This avoids gratuitously resending messages that
694  * the server had received and handled prior to reconnect.
695  */
696 void ceph_con_discard_requeued(struct ceph_connection *con, u64 reconnect_seq)
697 {
698 	struct ceph_msg *msg;
699 	u64 seq;
700 
701 	dout("%s con %p reconnect_seq %llu\n", __func__, con, reconnect_seq);
702 	while (!list_empty(&con->out_queue)) {
703 		msg = list_first_entry(&con->out_queue, struct ceph_msg,
704 				       list_head);
705 		if (msg->needs_out_seq)
706 			break;
707 		seq = le64_to_cpu(msg->hdr.seq);
708 		if (seq > reconnect_seq)
709 			break;
710 
711 		dout("%s con %p discarding msg %p seq %llu\n", __func__, con,
712 		     msg, seq);
713 		ceph_msg_remove(msg);
714 	}
715 }
716 
717 #ifdef CONFIG_BLOCK
718 
719 /*
720  * For a bio data item, a piece is whatever remains of the next
721  * entry in the current bio iovec, or the first entry in the next
722  * bio in the list.
723  */
724 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
725 					size_t length)
726 {
727 	struct ceph_msg_data *data = cursor->data;
728 	struct ceph_bio_iter *it = &cursor->bio_iter;
729 
730 	cursor->resid = min_t(size_t, length, data->bio_length);
731 	*it = data->bio_pos;
732 	if (cursor->resid < it->iter.bi_size)
733 		it->iter.bi_size = cursor->resid;
734 
735 	BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
736 }
737 
738 static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
739 						size_t *page_offset,
740 						size_t *length)
741 {
742 	struct bio_vec bv = bio_iter_iovec(cursor->bio_iter.bio,
743 					   cursor->bio_iter.iter);
744 
745 	*page_offset = bv.bv_offset;
746 	*length = bv.bv_len;
747 	return bv.bv_page;
748 }
749 
750 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
751 					size_t bytes)
752 {
753 	struct ceph_bio_iter *it = &cursor->bio_iter;
754 	struct page *page = bio_iter_page(it->bio, it->iter);
755 
756 	BUG_ON(bytes > cursor->resid);
757 	BUG_ON(bytes > bio_iter_len(it->bio, it->iter));
758 	cursor->resid -= bytes;
759 	bio_advance_iter(it->bio, &it->iter, bytes);
760 
761 	if (!cursor->resid)
762 		return false;   /* no more data */
763 
764 	if (!bytes || (it->iter.bi_size && it->iter.bi_bvec_done &&
765 		       page == bio_iter_page(it->bio, it->iter)))
766 		return false;	/* more bytes to process in this segment */
767 
768 	if (!it->iter.bi_size) {
769 		it->bio = it->bio->bi_next;
770 		it->iter = it->bio->bi_iter;
771 		if (cursor->resid < it->iter.bi_size)
772 			it->iter.bi_size = cursor->resid;
773 	}
774 
775 	BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
776 	return true;
777 }
778 #endif /* CONFIG_BLOCK */
779 
780 static void ceph_msg_data_bvecs_cursor_init(struct ceph_msg_data_cursor *cursor,
781 					size_t length)
782 {
783 	struct ceph_msg_data *data = cursor->data;
784 	struct bio_vec *bvecs = data->bvec_pos.bvecs;
785 
786 	cursor->resid = min_t(size_t, length, data->bvec_pos.iter.bi_size);
787 	cursor->bvec_iter = data->bvec_pos.iter;
788 	cursor->bvec_iter.bi_size = cursor->resid;
789 
790 	BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
791 }
792 
793 static struct page *ceph_msg_data_bvecs_next(struct ceph_msg_data_cursor *cursor,
794 						size_t *page_offset,
795 						size_t *length)
796 {
797 	struct bio_vec bv = bvec_iter_bvec(cursor->data->bvec_pos.bvecs,
798 					   cursor->bvec_iter);
799 
800 	*page_offset = bv.bv_offset;
801 	*length = bv.bv_len;
802 	return bv.bv_page;
803 }
804 
805 static bool ceph_msg_data_bvecs_advance(struct ceph_msg_data_cursor *cursor,
806 					size_t bytes)
807 {
808 	struct bio_vec *bvecs = cursor->data->bvec_pos.bvecs;
809 	struct page *page = bvec_iter_page(bvecs, cursor->bvec_iter);
810 
811 	BUG_ON(bytes > cursor->resid);
812 	BUG_ON(bytes > bvec_iter_len(bvecs, cursor->bvec_iter));
813 	cursor->resid -= bytes;
814 	bvec_iter_advance(bvecs, &cursor->bvec_iter, bytes);
815 
816 	if (!cursor->resid)
817 		return false;   /* no more data */
818 
819 	if (!bytes || (cursor->bvec_iter.bi_bvec_done &&
820 		       page == bvec_iter_page(bvecs, cursor->bvec_iter)))
821 		return false;	/* more bytes to process in this segment */
822 
823 	BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
824 	return true;
825 }
826 
827 /*
828  * For a page array, a piece comes from the first page in the array
829  * that has not already been fully consumed.
830  */
831 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
832 					size_t length)
833 {
834 	struct ceph_msg_data *data = cursor->data;
835 	int page_count;
836 
837 	BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
838 
839 	BUG_ON(!data->pages);
840 	BUG_ON(!data->length);
841 
842 	cursor->resid = min(length, data->length);
843 	page_count = calc_pages_for(data->alignment, (u64)data->length);
844 	cursor->page_offset = data->alignment & ~PAGE_MASK;
845 	cursor->page_index = 0;
846 	BUG_ON(page_count > (int)USHRT_MAX);
847 	cursor->page_count = (unsigned short)page_count;
848 	BUG_ON(length > SIZE_MAX - cursor->page_offset);
849 }
850 
851 static struct page *
852 ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
853 					size_t *page_offset, size_t *length)
854 {
855 	struct ceph_msg_data *data = cursor->data;
856 
857 	BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
858 
859 	BUG_ON(cursor->page_index >= cursor->page_count);
860 	BUG_ON(cursor->page_offset >= PAGE_SIZE);
861 
862 	*page_offset = cursor->page_offset;
863 	*length = min_t(size_t, cursor->resid, PAGE_SIZE - *page_offset);
864 	return data->pages[cursor->page_index];
865 }
866 
867 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
868 						size_t bytes)
869 {
870 	BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
871 
872 	BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
873 
874 	/* Advance the cursor page offset */
875 
876 	cursor->resid -= bytes;
877 	cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
878 	if (!bytes || cursor->page_offset)
879 		return false;	/* more bytes to process in the current page */
880 
881 	if (!cursor->resid)
882 		return false;   /* no more data */
883 
884 	/* Move on to the next page; offset is already at 0 */
885 
886 	BUG_ON(cursor->page_index >= cursor->page_count);
887 	cursor->page_index++;
888 	return true;
889 }
890 
891 /*
892  * For a pagelist, a piece is whatever remains to be consumed in the
893  * first page in the list, or the front of the next page.
894  */
895 static void
896 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
897 					size_t length)
898 {
899 	struct ceph_msg_data *data = cursor->data;
900 	struct ceph_pagelist *pagelist;
901 	struct page *page;
902 
903 	BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
904 
905 	pagelist = data->pagelist;
906 	BUG_ON(!pagelist);
907 
908 	if (!length)
909 		return;		/* pagelist can be assigned but empty */
910 
911 	BUG_ON(list_empty(&pagelist->head));
912 	page = list_first_entry(&pagelist->head, struct page, lru);
913 
914 	cursor->resid = min(length, pagelist->length);
915 	cursor->page = page;
916 	cursor->offset = 0;
917 }
918 
919 static struct page *
920 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
921 				size_t *page_offset, size_t *length)
922 {
923 	struct ceph_msg_data *data = cursor->data;
924 	struct ceph_pagelist *pagelist;
925 
926 	BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
927 
928 	pagelist = data->pagelist;
929 	BUG_ON(!pagelist);
930 
931 	BUG_ON(!cursor->page);
932 	BUG_ON(cursor->offset + cursor->resid != pagelist->length);
933 
934 	/* offset of first page in pagelist is always 0 */
935 	*page_offset = cursor->offset & ~PAGE_MASK;
936 	*length = min_t(size_t, cursor->resid, PAGE_SIZE - *page_offset);
937 	return cursor->page;
938 }
939 
940 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
941 						size_t bytes)
942 {
943 	struct ceph_msg_data *data = cursor->data;
944 	struct ceph_pagelist *pagelist;
945 
946 	BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
947 
948 	pagelist = data->pagelist;
949 	BUG_ON(!pagelist);
950 
951 	BUG_ON(cursor->offset + cursor->resid != pagelist->length);
952 	BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
953 
954 	/* Advance the cursor offset */
955 
956 	cursor->resid -= bytes;
957 	cursor->offset += bytes;
958 	/* offset of first page in pagelist is always 0 */
959 	if (!bytes || cursor->offset & ~PAGE_MASK)
960 		return false;	/* more bytes to process in the current page */
961 
962 	if (!cursor->resid)
963 		return false;   /* no more data */
964 
965 	/* Move on to the next page */
966 
967 	BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
968 	cursor->page = list_next_entry(cursor->page, lru);
969 	return true;
970 }
971 
972 /*
973  * Message data is handled (sent or received) in pieces, where each
974  * piece resides on a single page.  The network layer might not
975  * consume an entire piece at once.  A data item's cursor keeps
976  * track of which piece is next to process and how much remains to
977  * be processed in that piece.  It also tracks whether the current
978  * piece is the last one in the data item.
979  */
980 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
981 {
982 	size_t length = cursor->total_resid;
983 
984 	switch (cursor->data->type) {
985 	case CEPH_MSG_DATA_PAGELIST:
986 		ceph_msg_data_pagelist_cursor_init(cursor, length);
987 		break;
988 	case CEPH_MSG_DATA_PAGES:
989 		ceph_msg_data_pages_cursor_init(cursor, length);
990 		break;
991 #ifdef CONFIG_BLOCK
992 	case CEPH_MSG_DATA_BIO:
993 		ceph_msg_data_bio_cursor_init(cursor, length);
994 		break;
995 #endif /* CONFIG_BLOCK */
996 	case CEPH_MSG_DATA_BVECS:
997 		ceph_msg_data_bvecs_cursor_init(cursor, length);
998 		break;
999 	case CEPH_MSG_DATA_NONE:
1000 	default:
1001 		/* BUG(); */
1002 		break;
1003 	}
1004 	cursor->need_crc = true;
1005 }
1006 
1007 void ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor,
1008 			       struct ceph_msg *msg, size_t length)
1009 {
1010 	BUG_ON(!length);
1011 	BUG_ON(length > msg->data_length);
1012 	BUG_ON(!msg->num_data_items);
1013 
1014 	cursor->total_resid = length;
1015 	cursor->data = msg->data;
1016 
1017 	__ceph_msg_data_cursor_init(cursor);
1018 }
1019 
1020 /*
1021  * Return the page containing the next piece to process for a given
1022  * data item, and supply the page offset and length of that piece.
1023  * Indicate whether this is the last piece in this data item.
1024  */
1025 struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1026 				size_t *page_offset, size_t *length)
1027 {
1028 	struct page *page;
1029 
1030 	switch (cursor->data->type) {
1031 	case CEPH_MSG_DATA_PAGELIST:
1032 		page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1033 		break;
1034 	case CEPH_MSG_DATA_PAGES:
1035 		page = ceph_msg_data_pages_next(cursor, page_offset, length);
1036 		break;
1037 #ifdef CONFIG_BLOCK
1038 	case CEPH_MSG_DATA_BIO:
1039 		page = ceph_msg_data_bio_next(cursor, page_offset, length);
1040 		break;
1041 #endif /* CONFIG_BLOCK */
1042 	case CEPH_MSG_DATA_BVECS:
1043 		page = ceph_msg_data_bvecs_next(cursor, page_offset, length);
1044 		break;
1045 	case CEPH_MSG_DATA_NONE:
1046 	default:
1047 		page = NULL;
1048 		break;
1049 	}
1050 
1051 	BUG_ON(!page);
1052 	BUG_ON(*page_offset + *length > PAGE_SIZE);
1053 	BUG_ON(!*length);
1054 	BUG_ON(*length > cursor->resid);
1055 
1056 	return page;
1057 }
1058 
1059 /*
1060  * Returns true if the result moves the cursor on to the next piece
1061  * of the data item.
1062  */
1063 void ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor, size_t bytes)
1064 {
1065 	bool new_piece;
1066 
1067 	BUG_ON(bytes > cursor->resid);
1068 	switch (cursor->data->type) {
1069 	case CEPH_MSG_DATA_PAGELIST:
1070 		new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1071 		break;
1072 	case CEPH_MSG_DATA_PAGES:
1073 		new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1074 		break;
1075 #ifdef CONFIG_BLOCK
1076 	case CEPH_MSG_DATA_BIO:
1077 		new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1078 		break;
1079 #endif /* CONFIG_BLOCK */
1080 	case CEPH_MSG_DATA_BVECS:
1081 		new_piece = ceph_msg_data_bvecs_advance(cursor, bytes);
1082 		break;
1083 	case CEPH_MSG_DATA_NONE:
1084 	default:
1085 		BUG();
1086 		break;
1087 	}
1088 	cursor->total_resid -= bytes;
1089 
1090 	if (!cursor->resid && cursor->total_resid) {
1091 		cursor->data++;
1092 		__ceph_msg_data_cursor_init(cursor);
1093 		new_piece = true;
1094 	}
1095 	cursor->need_crc = new_piece;
1096 }
1097 
1098 u32 ceph_crc32c_page(u32 crc, struct page *page, unsigned int page_offset,
1099 		     unsigned int length)
1100 {
1101 	char *kaddr;
1102 
1103 	kaddr = kmap(page);
1104 	BUG_ON(kaddr == NULL);
1105 	crc = crc32c(crc, kaddr + page_offset, length);
1106 	kunmap(page);
1107 
1108 	return crc;
1109 }
1110 
1111 bool ceph_addr_is_blank(const struct ceph_entity_addr *addr)
1112 {
1113 	struct sockaddr_storage ss = addr->in_addr; /* align */
1114 	struct in_addr *addr4 = &((struct sockaddr_in *)&ss)->sin_addr;
1115 	struct in6_addr *addr6 = &((struct sockaddr_in6 *)&ss)->sin6_addr;
1116 
1117 	switch (ss.ss_family) {
1118 	case AF_INET:
1119 		return addr4->s_addr == htonl(INADDR_ANY);
1120 	case AF_INET6:
1121 		return ipv6_addr_any(addr6);
1122 	default:
1123 		return true;
1124 	}
1125 }
1126 
1127 int ceph_addr_port(const struct ceph_entity_addr *addr)
1128 {
1129 	switch (get_unaligned(&addr->in_addr.ss_family)) {
1130 	case AF_INET:
1131 		return ntohs(get_unaligned(&((struct sockaddr_in *)&addr->in_addr)->sin_port));
1132 	case AF_INET6:
1133 		return ntohs(get_unaligned(&((struct sockaddr_in6 *)&addr->in_addr)->sin6_port));
1134 	}
1135 	return 0;
1136 }
1137 
1138 void ceph_addr_set_port(struct ceph_entity_addr *addr, int p)
1139 {
1140 	switch (get_unaligned(&addr->in_addr.ss_family)) {
1141 	case AF_INET:
1142 		put_unaligned(htons(p), &((struct sockaddr_in *)&addr->in_addr)->sin_port);
1143 		break;
1144 	case AF_INET6:
1145 		put_unaligned(htons(p), &((struct sockaddr_in6 *)&addr->in_addr)->sin6_port);
1146 		break;
1147 	}
1148 }
1149 
1150 /*
1151  * Unlike other *_pton function semantics, zero indicates success.
1152  */
1153 static int ceph_pton(const char *str, size_t len, struct ceph_entity_addr *addr,
1154 		char delim, const char **ipend)
1155 {
1156 	memset(&addr->in_addr, 0, sizeof(addr->in_addr));
1157 
1158 	if (in4_pton(str, len, (u8 *)&((struct sockaddr_in *)&addr->in_addr)->sin_addr.s_addr, delim, ipend)) {
1159 		put_unaligned(AF_INET, &addr->in_addr.ss_family);
1160 		return 0;
1161 	}
1162 
1163 	if (in6_pton(str, len, (u8 *)&((struct sockaddr_in6 *)&addr->in_addr)->sin6_addr.s6_addr, delim, ipend)) {
1164 		put_unaligned(AF_INET6, &addr->in_addr.ss_family);
1165 		return 0;
1166 	}
1167 
1168 	return -EINVAL;
1169 }
1170 
1171 /*
1172  * Extract hostname string and resolve using kernel DNS facility.
1173  */
1174 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1175 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1176 		struct ceph_entity_addr *addr, char delim, const char **ipend)
1177 {
1178 	const char *end, *delim_p;
1179 	char *colon_p, *ip_addr = NULL;
1180 	int ip_len, ret;
1181 
1182 	/*
1183 	 * The end of the hostname occurs immediately preceding the delimiter or
1184 	 * the port marker (':') where the delimiter takes precedence.
1185 	 */
1186 	delim_p = memchr(name, delim, namelen);
1187 	colon_p = memchr(name, ':', namelen);
1188 
1189 	if (delim_p && colon_p)
1190 		end = delim_p < colon_p ? delim_p : colon_p;
1191 	else if (!delim_p && colon_p)
1192 		end = colon_p;
1193 	else {
1194 		end = delim_p;
1195 		if (!end) /* case: hostname:/ */
1196 			end = name + namelen;
1197 	}
1198 
1199 	if (end <= name)
1200 		return -EINVAL;
1201 
1202 	/* do dns_resolve upcall */
1203 	ip_len = dns_query(current->nsproxy->net_ns,
1204 			   NULL, name, end - name, NULL, &ip_addr, NULL, false);
1205 	if (ip_len > 0)
1206 		ret = ceph_pton(ip_addr, ip_len, addr, -1, NULL);
1207 	else
1208 		ret = -ESRCH;
1209 
1210 	kfree(ip_addr);
1211 
1212 	*ipend = end;
1213 
1214 	pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1215 			ret, ret ? "failed" : ceph_pr_addr(addr));
1216 
1217 	return ret;
1218 }
1219 #else
1220 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1221 		struct ceph_entity_addr *addr, char delim, const char **ipend)
1222 {
1223 	return -EINVAL;
1224 }
1225 #endif
1226 
1227 /*
1228  * Parse a server name (IP or hostname). If a valid IP address is not found
1229  * then try to extract a hostname to resolve using userspace DNS upcall.
1230  */
1231 static int ceph_parse_server_name(const char *name, size_t namelen,
1232 		struct ceph_entity_addr *addr, char delim, const char **ipend)
1233 {
1234 	int ret;
1235 
1236 	ret = ceph_pton(name, namelen, addr, delim, ipend);
1237 	if (ret)
1238 		ret = ceph_dns_resolve_name(name, namelen, addr, delim, ipend);
1239 
1240 	return ret;
1241 }
1242 
1243 /*
1244  * Parse an ip[:port] list into an addr array.  Use the default
1245  * monitor port if a port isn't specified.
1246  */
1247 int ceph_parse_ips(const char *c, const char *end,
1248 		   struct ceph_entity_addr *addr,
1249 		   int max_count, int *count, char delim)
1250 {
1251 	int i, ret = -EINVAL;
1252 	const char *p = c;
1253 
1254 	dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1255 	for (i = 0; i < max_count; i++) {
1256 		char cur_delim = delim;
1257 		const char *ipend;
1258 		int port;
1259 
1260 		if (*p == '[') {
1261 			cur_delim = ']';
1262 			p++;
1263 		}
1264 
1265 		ret = ceph_parse_server_name(p, end - p, &addr[i], cur_delim,
1266 					     &ipend);
1267 		if (ret)
1268 			goto bad;
1269 		ret = -EINVAL;
1270 
1271 		p = ipend;
1272 
1273 		if (cur_delim == ']') {
1274 			if (*p != ']') {
1275 				dout("missing matching ']'\n");
1276 				goto bad;
1277 			}
1278 			p++;
1279 		}
1280 
1281 		/* port? */
1282 		if (p < end && *p == ':') {
1283 			port = 0;
1284 			p++;
1285 			while (p < end && *p >= '0' && *p <= '9') {
1286 				port = (port * 10) + (*p - '0');
1287 				p++;
1288 			}
1289 			if (port == 0)
1290 				port = CEPH_MON_PORT;
1291 			else if (port > 65535)
1292 				goto bad;
1293 		} else {
1294 			port = CEPH_MON_PORT;
1295 		}
1296 
1297 		ceph_addr_set_port(&addr[i], port);
1298 		/*
1299 		 * We want the type to be set according to ms_mode
1300 		 * option, but options are normally parsed after mon
1301 		 * addresses.  Rather than complicating parsing, set
1302 		 * to LEGACY and override in build_initial_monmap()
1303 		 * for mon addresses and ceph_messenger_init() for
1304 		 * ip option.
1305 		 */
1306 		addr[i].type = CEPH_ENTITY_ADDR_TYPE_LEGACY;
1307 		addr[i].nonce = 0;
1308 
1309 		dout("%s got %s\n", __func__, ceph_pr_addr(&addr[i]));
1310 
1311 		if (p == end)
1312 			break;
1313 		if (*p != delim)
1314 			goto bad;
1315 		p++;
1316 	}
1317 
1318 	if (p != end)
1319 		goto bad;
1320 
1321 	if (count)
1322 		*count = i + 1;
1323 	return 0;
1324 
1325 bad:
1326 	return ret;
1327 }
1328 
1329 /*
1330  * Process message.  This happens in the worker thread.  The callback should
1331  * be careful not to do anything that waits on other incoming messages or it
1332  * may deadlock.
1333  */
1334 void ceph_con_process_message(struct ceph_connection *con)
1335 {
1336 	struct ceph_msg *msg = con->in_msg;
1337 
1338 	BUG_ON(con->in_msg->con != con);
1339 	con->in_msg = NULL;
1340 
1341 	/* if first message, set peer_name */
1342 	if (con->peer_name.type == 0)
1343 		con->peer_name = msg->hdr.src;
1344 
1345 	con->in_seq++;
1346 	mutex_unlock(&con->mutex);
1347 
1348 	dout("===== %p %llu from %s%lld %d=%s len %d+%d+%d (%u %u %u) =====\n",
1349 	     msg, le64_to_cpu(msg->hdr.seq),
1350 	     ENTITY_NAME(msg->hdr.src),
1351 	     le16_to_cpu(msg->hdr.type),
1352 	     ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
1353 	     le32_to_cpu(msg->hdr.front_len),
1354 	     le32_to_cpu(msg->hdr.middle_len),
1355 	     le32_to_cpu(msg->hdr.data_len),
1356 	     con->in_front_crc, con->in_middle_crc, con->in_data_crc);
1357 	con->ops->dispatch(con, msg);
1358 
1359 	mutex_lock(&con->mutex);
1360 }
1361 
1362 /*
1363  * Atomically queue work on a connection after the specified delay.
1364  * Bump @con reference to avoid races with connection teardown.
1365  * Returns 0 if work was queued, or an error code otherwise.
1366  */
1367 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
1368 {
1369 	if (!con->ops->get(con)) {
1370 		dout("%s %p ref count 0\n", __func__, con);
1371 		return -ENOENT;
1372 	}
1373 
1374 	if (delay >= HZ)
1375 		delay = round_jiffies_relative(delay);
1376 
1377 	dout("%s %p %lu\n", __func__, con, delay);
1378 	if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
1379 		dout("%s %p - already queued\n", __func__, con);
1380 		con->ops->put(con);
1381 		return -EBUSY;
1382 	}
1383 
1384 	return 0;
1385 }
1386 
1387 static void queue_con(struct ceph_connection *con)
1388 {
1389 	(void) queue_con_delay(con, 0);
1390 }
1391 
1392 static void cancel_con(struct ceph_connection *con)
1393 {
1394 	if (cancel_delayed_work(&con->work)) {
1395 		dout("%s %p\n", __func__, con);
1396 		con->ops->put(con);
1397 	}
1398 }
1399 
1400 static bool con_sock_closed(struct ceph_connection *con)
1401 {
1402 	if (!ceph_con_flag_test_and_clear(con, CEPH_CON_F_SOCK_CLOSED))
1403 		return false;
1404 
1405 #define CASE(x)								\
1406 	case CEPH_CON_S_ ## x:						\
1407 		con->error_msg = "socket closed (con state " #x ")";	\
1408 		break;
1409 
1410 	switch (con->state) {
1411 	CASE(CLOSED);
1412 	CASE(PREOPEN);
1413 	CASE(V1_BANNER);
1414 	CASE(V1_CONNECT_MSG);
1415 	CASE(V2_BANNER_PREFIX);
1416 	CASE(V2_BANNER_PAYLOAD);
1417 	CASE(V2_HELLO);
1418 	CASE(V2_AUTH);
1419 	CASE(V2_AUTH_SIGNATURE);
1420 	CASE(V2_SESSION_CONNECT);
1421 	CASE(V2_SESSION_RECONNECT);
1422 	CASE(OPEN);
1423 	CASE(STANDBY);
1424 	default:
1425 		BUG();
1426 	}
1427 #undef CASE
1428 
1429 	return true;
1430 }
1431 
1432 static bool con_backoff(struct ceph_connection *con)
1433 {
1434 	int ret;
1435 
1436 	if (!ceph_con_flag_test_and_clear(con, CEPH_CON_F_BACKOFF))
1437 		return false;
1438 
1439 	ret = queue_con_delay(con, con->delay);
1440 	if (ret) {
1441 		dout("%s: con %p FAILED to back off %lu\n", __func__,
1442 			con, con->delay);
1443 		BUG_ON(ret == -ENOENT);
1444 		ceph_con_flag_set(con, CEPH_CON_F_BACKOFF);
1445 	}
1446 
1447 	return true;
1448 }
1449 
1450 /* Finish fault handling; con->mutex must *not* be held here */
1451 
1452 static void con_fault_finish(struct ceph_connection *con)
1453 {
1454 	dout("%s %p\n", __func__, con);
1455 
1456 	/*
1457 	 * in case we faulted due to authentication, invalidate our
1458 	 * current tickets so that we can get new ones.
1459 	 */
1460 	if (con->v1.auth_retry) {
1461 		dout("auth_retry %d, invalidating\n", con->v1.auth_retry);
1462 		if (con->ops->invalidate_authorizer)
1463 			con->ops->invalidate_authorizer(con);
1464 		con->v1.auth_retry = 0;
1465 	}
1466 
1467 	if (con->ops->fault)
1468 		con->ops->fault(con);
1469 }
1470 
1471 /*
1472  * Do some work on a connection.  Drop a connection ref when we're done.
1473  */
1474 static void ceph_con_workfn(struct work_struct *work)
1475 {
1476 	struct ceph_connection *con = container_of(work, struct ceph_connection,
1477 						   work.work);
1478 	bool fault;
1479 
1480 	mutex_lock(&con->mutex);
1481 	while (true) {
1482 		int ret;
1483 
1484 		if ((fault = con_sock_closed(con))) {
1485 			dout("%s: con %p SOCK_CLOSED\n", __func__, con);
1486 			break;
1487 		}
1488 		if (con_backoff(con)) {
1489 			dout("%s: con %p BACKOFF\n", __func__, con);
1490 			break;
1491 		}
1492 		if (con->state == CEPH_CON_S_STANDBY) {
1493 			dout("%s: con %p STANDBY\n", __func__, con);
1494 			break;
1495 		}
1496 		if (con->state == CEPH_CON_S_CLOSED) {
1497 			dout("%s: con %p CLOSED\n", __func__, con);
1498 			BUG_ON(con->sock);
1499 			break;
1500 		}
1501 		if (con->state == CEPH_CON_S_PREOPEN) {
1502 			dout("%s: con %p PREOPEN\n", __func__, con);
1503 			BUG_ON(con->sock);
1504 		}
1505 
1506 		if (ceph_msgr2(from_msgr(con->msgr)))
1507 			ret = ceph_con_v2_try_read(con);
1508 		else
1509 			ret = ceph_con_v1_try_read(con);
1510 		if (ret < 0) {
1511 			if (ret == -EAGAIN)
1512 				continue;
1513 			if (!con->error_msg)
1514 				con->error_msg = "socket error on read";
1515 			fault = true;
1516 			break;
1517 		}
1518 
1519 		if (ceph_msgr2(from_msgr(con->msgr)))
1520 			ret = ceph_con_v2_try_write(con);
1521 		else
1522 			ret = ceph_con_v1_try_write(con);
1523 		if (ret < 0) {
1524 			if (ret == -EAGAIN)
1525 				continue;
1526 			if (!con->error_msg)
1527 				con->error_msg = "socket error on write";
1528 			fault = true;
1529 		}
1530 
1531 		break;	/* If we make it to here, we're done */
1532 	}
1533 	if (fault)
1534 		con_fault(con);
1535 	mutex_unlock(&con->mutex);
1536 
1537 	if (fault)
1538 		con_fault_finish(con);
1539 
1540 	con->ops->put(con);
1541 }
1542 
1543 /*
1544  * Generic error/fault handler.  A retry mechanism is used with
1545  * exponential backoff
1546  */
1547 static void con_fault(struct ceph_connection *con)
1548 {
1549 	dout("fault %p state %d to peer %s\n",
1550 	     con, con->state, ceph_pr_addr(&con->peer_addr));
1551 
1552 	pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
1553 		ceph_pr_addr(&con->peer_addr), con->error_msg);
1554 	con->error_msg = NULL;
1555 
1556 	WARN_ON(con->state == CEPH_CON_S_STANDBY ||
1557 		con->state == CEPH_CON_S_CLOSED);
1558 
1559 	ceph_con_reset_protocol(con);
1560 
1561 	if (ceph_con_flag_test(con, CEPH_CON_F_LOSSYTX)) {
1562 		dout("fault on LOSSYTX channel, marking CLOSED\n");
1563 		con->state = CEPH_CON_S_CLOSED;
1564 		return;
1565 	}
1566 
1567 	/* Requeue anything that hasn't been acked */
1568 	list_splice_init(&con->out_sent, &con->out_queue);
1569 
1570 	/* If there are no messages queued or keepalive pending, place
1571 	 * the connection in a STANDBY state */
1572 	if (list_empty(&con->out_queue) &&
1573 	    !ceph_con_flag_test(con, CEPH_CON_F_KEEPALIVE_PENDING)) {
1574 		dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
1575 		ceph_con_flag_clear(con, CEPH_CON_F_WRITE_PENDING);
1576 		con->state = CEPH_CON_S_STANDBY;
1577 	} else {
1578 		/* retry after a delay. */
1579 		con->state = CEPH_CON_S_PREOPEN;
1580 		if (!con->delay) {
1581 			con->delay = BASE_DELAY_INTERVAL;
1582 		} else if (con->delay < MAX_DELAY_INTERVAL) {
1583 			con->delay *= 2;
1584 			if (con->delay > MAX_DELAY_INTERVAL)
1585 				con->delay = MAX_DELAY_INTERVAL;
1586 		}
1587 		ceph_con_flag_set(con, CEPH_CON_F_BACKOFF);
1588 		queue_con(con);
1589 	}
1590 }
1591 
1592 void ceph_messenger_reset_nonce(struct ceph_messenger *msgr)
1593 {
1594 	u32 nonce = le32_to_cpu(msgr->inst.addr.nonce) + 1000000;
1595 	msgr->inst.addr.nonce = cpu_to_le32(nonce);
1596 	ceph_encode_my_addr(msgr);
1597 }
1598 
1599 /*
1600  * initialize a new messenger instance
1601  */
1602 void ceph_messenger_init(struct ceph_messenger *msgr,
1603 			 struct ceph_entity_addr *myaddr)
1604 {
1605 	spin_lock_init(&msgr->global_seq_lock);
1606 
1607 	if (myaddr) {
1608 		memcpy(&msgr->inst.addr.in_addr, &myaddr->in_addr,
1609 		       sizeof(msgr->inst.addr.in_addr));
1610 		ceph_addr_set_port(&msgr->inst.addr, 0);
1611 	}
1612 
1613 	/*
1614 	 * Since nautilus, clients are identified using type ANY.
1615 	 * For msgr1, ceph_encode_banner_addr() munges it to NONE.
1616 	 */
1617 	msgr->inst.addr.type = CEPH_ENTITY_ADDR_TYPE_ANY;
1618 
1619 	/* generate a random non-zero nonce */
1620 	do {
1621 		get_random_bytes(&msgr->inst.addr.nonce,
1622 				 sizeof(msgr->inst.addr.nonce));
1623 	} while (!msgr->inst.addr.nonce);
1624 	ceph_encode_my_addr(msgr);
1625 
1626 	atomic_set(&msgr->stopping, 0);
1627 	write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
1628 
1629 	dout("%s %p\n", __func__, msgr);
1630 }
1631 
1632 void ceph_messenger_fini(struct ceph_messenger *msgr)
1633 {
1634 	put_net(read_pnet(&msgr->net));
1635 }
1636 
1637 static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con)
1638 {
1639 	if (msg->con)
1640 		msg->con->ops->put(msg->con);
1641 
1642 	msg->con = con ? con->ops->get(con) : NULL;
1643 	BUG_ON(msg->con != con);
1644 }
1645 
1646 static void clear_standby(struct ceph_connection *con)
1647 {
1648 	/* come back from STANDBY? */
1649 	if (con->state == CEPH_CON_S_STANDBY) {
1650 		dout("clear_standby %p and ++connect_seq\n", con);
1651 		con->state = CEPH_CON_S_PREOPEN;
1652 		con->v1.connect_seq++;
1653 		WARN_ON(ceph_con_flag_test(con, CEPH_CON_F_WRITE_PENDING));
1654 		WARN_ON(ceph_con_flag_test(con, CEPH_CON_F_KEEPALIVE_PENDING));
1655 	}
1656 }
1657 
1658 /*
1659  * Queue up an outgoing message on the given connection.
1660  *
1661  * Consumes a ref on @msg.
1662  */
1663 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
1664 {
1665 	/* set src+dst */
1666 	msg->hdr.src = con->msgr->inst.name;
1667 	BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
1668 	msg->needs_out_seq = true;
1669 
1670 	mutex_lock(&con->mutex);
1671 
1672 	if (con->state == CEPH_CON_S_CLOSED) {
1673 		dout("con_send %p closed, dropping %p\n", con, msg);
1674 		ceph_msg_put(msg);
1675 		mutex_unlock(&con->mutex);
1676 		return;
1677 	}
1678 
1679 	msg_con_set(msg, con);
1680 
1681 	BUG_ON(!list_empty(&msg->list_head));
1682 	list_add_tail(&msg->list_head, &con->out_queue);
1683 	dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
1684 	     ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
1685 	     ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
1686 	     le32_to_cpu(msg->hdr.front_len),
1687 	     le32_to_cpu(msg->hdr.middle_len),
1688 	     le32_to_cpu(msg->hdr.data_len));
1689 
1690 	clear_standby(con);
1691 	mutex_unlock(&con->mutex);
1692 
1693 	/* if there wasn't anything waiting to send before, queue
1694 	 * new work */
1695 	if (!ceph_con_flag_test_and_set(con, CEPH_CON_F_WRITE_PENDING))
1696 		queue_con(con);
1697 }
1698 EXPORT_SYMBOL(ceph_con_send);
1699 
1700 /*
1701  * Revoke a message that was previously queued for send
1702  */
1703 void ceph_msg_revoke(struct ceph_msg *msg)
1704 {
1705 	struct ceph_connection *con = msg->con;
1706 
1707 	if (!con) {
1708 		dout("%s msg %p null con\n", __func__, msg);
1709 		return;		/* Message not in our possession */
1710 	}
1711 
1712 	mutex_lock(&con->mutex);
1713 	if (list_empty(&msg->list_head)) {
1714 		WARN_ON(con->out_msg == msg);
1715 		dout("%s con %p msg %p not linked\n", __func__, con, msg);
1716 		mutex_unlock(&con->mutex);
1717 		return;
1718 	}
1719 
1720 	dout("%s con %p msg %p was linked\n", __func__, con, msg);
1721 	msg->hdr.seq = 0;
1722 	ceph_msg_remove(msg);
1723 
1724 	if (con->out_msg == msg) {
1725 		WARN_ON(con->state != CEPH_CON_S_OPEN);
1726 		dout("%s con %p msg %p was sending\n", __func__, con, msg);
1727 		if (ceph_msgr2(from_msgr(con->msgr)))
1728 			ceph_con_v2_revoke(con);
1729 		else
1730 			ceph_con_v1_revoke(con);
1731 		ceph_msg_put(con->out_msg);
1732 		con->out_msg = NULL;
1733 	} else {
1734 		dout("%s con %p msg %p not current, out_msg %p\n", __func__,
1735 		     con, msg, con->out_msg);
1736 	}
1737 	mutex_unlock(&con->mutex);
1738 }
1739 
1740 /*
1741  * Revoke a message that we may be reading data into
1742  */
1743 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
1744 {
1745 	struct ceph_connection *con = msg->con;
1746 
1747 	if (!con) {
1748 		dout("%s msg %p null con\n", __func__, msg);
1749 		return;		/* Message not in our possession */
1750 	}
1751 
1752 	mutex_lock(&con->mutex);
1753 	if (con->in_msg == msg) {
1754 		WARN_ON(con->state != CEPH_CON_S_OPEN);
1755 		dout("%s con %p msg %p was recving\n", __func__, con, msg);
1756 		if (ceph_msgr2(from_msgr(con->msgr)))
1757 			ceph_con_v2_revoke_incoming(con);
1758 		else
1759 			ceph_con_v1_revoke_incoming(con);
1760 		ceph_msg_put(con->in_msg);
1761 		con->in_msg = NULL;
1762 	} else {
1763 		dout("%s con %p msg %p not current, in_msg %p\n", __func__,
1764 		     con, msg, con->in_msg);
1765 	}
1766 	mutex_unlock(&con->mutex);
1767 }
1768 
1769 /*
1770  * Queue a keepalive byte to ensure the tcp connection is alive.
1771  */
1772 void ceph_con_keepalive(struct ceph_connection *con)
1773 {
1774 	dout("con_keepalive %p\n", con);
1775 	mutex_lock(&con->mutex);
1776 	clear_standby(con);
1777 	ceph_con_flag_set(con, CEPH_CON_F_KEEPALIVE_PENDING);
1778 	mutex_unlock(&con->mutex);
1779 
1780 	if (!ceph_con_flag_test_and_set(con, CEPH_CON_F_WRITE_PENDING))
1781 		queue_con(con);
1782 }
1783 EXPORT_SYMBOL(ceph_con_keepalive);
1784 
1785 bool ceph_con_keepalive_expired(struct ceph_connection *con,
1786 			       unsigned long interval)
1787 {
1788 	if (interval > 0 &&
1789 	    (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
1790 		struct timespec64 now;
1791 		struct timespec64 ts;
1792 		ktime_get_real_ts64(&now);
1793 		jiffies_to_timespec64(interval, &ts);
1794 		ts = timespec64_add(con->last_keepalive_ack, ts);
1795 		return timespec64_compare(&now, &ts) >= 0;
1796 	}
1797 	return false;
1798 }
1799 
1800 static struct ceph_msg_data *ceph_msg_data_add(struct ceph_msg *msg)
1801 {
1802 	BUG_ON(msg->num_data_items >= msg->max_data_items);
1803 	return &msg->data[msg->num_data_items++];
1804 }
1805 
1806 static void ceph_msg_data_destroy(struct ceph_msg_data *data)
1807 {
1808 	if (data->type == CEPH_MSG_DATA_PAGES && data->own_pages) {
1809 		int num_pages = calc_pages_for(data->alignment, data->length);
1810 		ceph_release_page_vector(data->pages, num_pages);
1811 	} else if (data->type == CEPH_MSG_DATA_PAGELIST) {
1812 		ceph_pagelist_release(data->pagelist);
1813 	}
1814 }
1815 
1816 void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
1817 			     size_t length, size_t alignment, bool own_pages)
1818 {
1819 	struct ceph_msg_data *data;
1820 
1821 	BUG_ON(!pages);
1822 	BUG_ON(!length);
1823 
1824 	data = ceph_msg_data_add(msg);
1825 	data->type = CEPH_MSG_DATA_PAGES;
1826 	data->pages = pages;
1827 	data->length = length;
1828 	data->alignment = alignment & ~PAGE_MASK;
1829 	data->own_pages = own_pages;
1830 
1831 	msg->data_length += length;
1832 }
1833 EXPORT_SYMBOL(ceph_msg_data_add_pages);
1834 
1835 void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
1836 				struct ceph_pagelist *pagelist)
1837 {
1838 	struct ceph_msg_data *data;
1839 
1840 	BUG_ON(!pagelist);
1841 	BUG_ON(!pagelist->length);
1842 
1843 	data = ceph_msg_data_add(msg);
1844 	data->type = CEPH_MSG_DATA_PAGELIST;
1845 	refcount_inc(&pagelist->refcnt);
1846 	data->pagelist = pagelist;
1847 
1848 	msg->data_length += pagelist->length;
1849 }
1850 EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
1851 
1852 #ifdef	CONFIG_BLOCK
1853 void ceph_msg_data_add_bio(struct ceph_msg *msg, struct ceph_bio_iter *bio_pos,
1854 			   u32 length)
1855 {
1856 	struct ceph_msg_data *data;
1857 
1858 	data = ceph_msg_data_add(msg);
1859 	data->type = CEPH_MSG_DATA_BIO;
1860 	data->bio_pos = *bio_pos;
1861 	data->bio_length = length;
1862 
1863 	msg->data_length += length;
1864 }
1865 EXPORT_SYMBOL(ceph_msg_data_add_bio);
1866 #endif	/* CONFIG_BLOCK */
1867 
1868 void ceph_msg_data_add_bvecs(struct ceph_msg *msg,
1869 			     struct ceph_bvec_iter *bvec_pos)
1870 {
1871 	struct ceph_msg_data *data;
1872 
1873 	data = ceph_msg_data_add(msg);
1874 	data->type = CEPH_MSG_DATA_BVECS;
1875 	data->bvec_pos = *bvec_pos;
1876 
1877 	msg->data_length += bvec_pos->iter.bi_size;
1878 }
1879 EXPORT_SYMBOL(ceph_msg_data_add_bvecs);
1880 
1881 /*
1882  * construct a new message with given type, size
1883  * the new msg has a ref count of 1.
1884  */
1885 struct ceph_msg *ceph_msg_new2(int type, int front_len, int max_data_items,
1886 			       gfp_t flags, bool can_fail)
1887 {
1888 	struct ceph_msg *m;
1889 
1890 	m = kmem_cache_zalloc(ceph_msg_cache, flags);
1891 	if (m == NULL)
1892 		goto out;
1893 
1894 	m->hdr.type = cpu_to_le16(type);
1895 	m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
1896 	m->hdr.front_len = cpu_to_le32(front_len);
1897 
1898 	INIT_LIST_HEAD(&m->list_head);
1899 	kref_init(&m->kref);
1900 
1901 	/* front */
1902 	if (front_len) {
1903 		m->front.iov_base = kvmalloc(front_len, flags);
1904 		if (m->front.iov_base == NULL) {
1905 			dout("ceph_msg_new can't allocate %d bytes\n",
1906 			     front_len);
1907 			goto out2;
1908 		}
1909 	} else {
1910 		m->front.iov_base = NULL;
1911 	}
1912 	m->front_alloc_len = m->front.iov_len = front_len;
1913 
1914 	if (max_data_items) {
1915 		m->data = kmalloc_array(max_data_items, sizeof(*m->data),
1916 					flags);
1917 		if (!m->data)
1918 			goto out2;
1919 
1920 		m->max_data_items = max_data_items;
1921 	}
1922 
1923 	dout("ceph_msg_new %p front %d\n", m, front_len);
1924 	return m;
1925 
1926 out2:
1927 	ceph_msg_put(m);
1928 out:
1929 	if (!can_fail) {
1930 		pr_err("msg_new can't create type %d front %d\n", type,
1931 		       front_len);
1932 		WARN_ON(1);
1933 	} else {
1934 		dout("msg_new can't create type %d front %d\n", type,
1935 		     front_len);
1936 	}
1937 	return NULL;
1938 }
1939 EXPORT_SYMBOL(ceph_msg_new2);
1940 
1941 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
1942 			      bool can_fail)
1943 {
1944 	return ceph_msg_new2(type, front_len, 0, flags, can_fail);
1945 }
1946 EXPORT_SYMBOL(ceph_msg_new);
1947 
1948 /*
1949  * Allocate "middle" portion of a message, if it is needed and wasn't
1950  * allocated by alloc_msg.  This allows us to read a small fixed-size
1951  * per-type header in the front and then gracefully fail (i.e.,
1952  * propagate the error to the caller based on info in the front) when
1953  * the middle is too large.
1954  */
1955 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
1956 {
1957 	int type = le16_to_cpu(msg->hdr.type);
1958 	int middle_len = le32_to_cpu(msg->hdr.middle_len);
1959 
1960 	dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
1961 	     ceph_msg_type_name(type), middle_len);
1962 	BUG_ON(!middle_len);
1963 	BUG_ON(msg->middle);
1964 
1965 	msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
1966 	if (!msg->middle)
1967 		return -ENOMEM;
1968 	return 0;
1969 }
1970 
1971 /*
1972  * Allocate a message for receiving an incoming message on a
1973  * connection, and save the result in con->in_msg.  Uses the
1974  * connection's private alloc_msg op if available.
1975  *
1976  * Returns 0 on success, or a negative error code.
1977  *
1978  * On success, if we set *skip = 1:
1979  *  - the next message should be skipped and ignored.
1980  *  - con->in_msg == NULL
1981  * or if we set *skip = 0:
1982  *  - con->in_msg is non-null.
1983  * On error (ENOMEM, EAGAIN, ...),
1984  *  - con->in_msg == NULL
1985  */
1986 int ceph_con_in_msg_alloc(struct ceph_connection *con,
1987 			  struct ceph_msg_header *hdr, int *skip)
1988 {
1989 	int middle_len = le32_to_cpu(hdr->middle_len);
1990 	struct ceph_msg *msg;
1991 	int ret = 0;
1992 
1993 	BUG_ON(con->in_msg != NULL);
1994 	BUG_ON(!con->ops->alloc_msg);
1995 
1996 	mutex_unlock(&con->mutex);
1997 	msg = con->ops->alloc_msg(con, hdr, skip);
1998 	mutex_lock(&con->mutex);
1999 	if (con->state != CEPH_CON_S_OPEN) {
2000 		if (msg)
2001 			ceph_msg_put(msg);
2002 		return -EAGAIN;
2003 	}
2004 	if (msg) {
2005 		BUG_ON(*skip);
2006 		msg_con_set(msg, con);
2007 		con->in_msg = msg;
2008 	} else {
2009 		/*
2010 		 * Null message pointer means either we should skip
2011 		 * this message or we couldn't allocate memory.  The
2012 		 * former is not an error.
2013 		 */
2014 		if (*skip)
2015 			return 0;
2016 
2017 		con->error_msg = "error allocating memory for incoming message";
2018 		return -ENOMEM;
2019 	}
2020 	memcpy(&con->in_msg->hdr, hdr, sizeof(*hdr));
2021 
2022 	if (middle_len && !con->in_msg->middle) {
2023 		ret = ceph_alloc_middle(con, con->in_msg);
2024 		if (ret < 0) {
2025 			ceph_msg_put(con->in_msg);
2026 			con->in_msg = NULL;
2027 		}
2028 	}
2029 
2030 	return ret;
2031 }
2032 
2033 void ceph_con_get_out_msg(struct ceph_connection *con)
2034 {
2035 	struct ceph_msg *msg;
2036 
2037 	BUG_ON(list_empty(&con->out_queue));
2038 	msg = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
2039 	WARN_ON(msg->con != con);
2040 
2041 	/*
2042 	 * Put the message on "sent" list using a ref from ceph_con_send().
2043 	 * It is put when the message is acked or revoked.
2044 	 */
2045 	list_move_tail(&msg->list_head, &con->out_sent);
2046 
2047 	/*
2048 	 * Only assign outgoing seq # if we haven't sent this message
2049 	 * yet.  If it is requeued, resend with it's original seq.
2050 	 */
2051 	if (msg->needs_out_seq) {
2052 		msg->hdr.seq = cpu_to_le64(++con->out_seq);
2053 		msg->needs_out_seq = false;
2054 
2055 		if (con->ops->reencode_message)
2056 			con->ops->reencode_message(msg);
2057 	}
2058 
2059 	/*
2060 	 * Get a ref for out_msg.  It is put when we are done sending the
2061 	 * message or in case of a fault.
2062 	 */
2063 	WARN_ON(con->out_msg);
2064 	con->out_msg = ceph_msg_get(msg);
2065 }
2066 
2067 /*
2068  * Free a generically kmalloc'd message.
2069  */
2070 static void ceph_msg_free(struct ceph_msg *m)
2071 {
2072 	dout("%s %p\n", __func__, m);
2073 	kvfree(m->front.iov_base);
2074 	kfree(m->data);
2075 	kmem_cache_free(ceph_msg_cache, m);
2076 }
2077 
2078 static void ceph_msg_release(struct kref *kref)
2079 {
2080 	struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
2081 	int i;
2082 
2083 	dout("%s %p\n", __func__, m);
2084 	WARN_ON(!list_empty(&m->list_head));
2085 
2086 	msg_con_set(m, NULL);
2087 
2088 	/* drop middle, data, if any */
2089 	if (m->middle) {
2090 		ceph_buffer_put(m->middle);
2091 		m->middle = NULL;
2092 	}
2093 
2094 	for (i = 0; i < m->num_data_items; i++)
2095 		ceph_msg_data_destroy(&m->data[i]);
2096 
2097 	if (m->pool)
2098 		ceph_msgpool_put(m->pool, m);
2099 	else
2100 		ceph_msg_free(m);
2101 }
2102 
2103 struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
2104 {
2105 	dout("%s %p (was %d)\n", __func__, msg,
2106 	     kref_read(&msg->kref));
2107 	kref_get(&msg->kref);
2108 	return msg;
2109 }
2110 EXPORT_SYMBOL(ceph_msg_get);
2111 
2112 void ceph_msg_put(struct ceph_msg *msg)
2113 {
2114 	dout("%s %p (was %d)\n", __func__, msg,
2115 	     kref_read(&msg->kref));
2116 	kref_put(&msg->kref, ceph_msg_release);
2117 }
2118 EXPORT_SYMBOL(ceph_msg_put);
2119 
2120 void ceph_msg_dump(struct ceph_msg *msg)
2121 {
2122 	pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
2123 		 msg->front_alloc_len, msg->data_length);
2124 	print_hex_dump(KERN_DEBUG, "header: ",
2125 		       DUMP_PREFIX_OFFSET, 16, 1,
2126 		       &msg->hdr, sizeof(msg->hdr), true);
2127 	print_hex_dump(KERN_DEBUG, " front: ",
2128 		       DUMP_PREFIX_OFFSET, 16, 1,
2129 		       msg->front.iov_base, msg->front.iov_len, true);
2130 	if (msg->middle)
2131 		print_hex_dump(KERN_DEBUG, "middle: ",
2132 			       DUMP_PREFIX_OFFSET, 16, 1,
2133 			       msg->middle->vec.iov_base,
2134 			       msg->middle->vec.iov_len, true);
2135 	print_hex_dump(KERN_DEBUG, "footer: ",
2136 		       DUMP_PREFIX_OFFSET, 16, 1,
2137 		       &msg->footer, sizeof(msg->footer), true);
2138 }
2139 EXPORT_SYMBOL(ceph_msg_dump);
2140