xref: /linux/net/ceph/messenger.c (revision 3839a7460721b87501134697b7b90c45dcc7825d)
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 
21 #include <linux/ceph/ceph_features.h>
22 #include <linux/ceph/libceph.h>
23 #include <linux/ceph/messenger.h>
24 #include <linux/ceph/decode.h>
25 #include <linux/ceph/pagelist.h>
26 #include <linux/export.h>
27 
28 /*
29  * Ceph uses the messenger to exchange ceph_msg messages with other
30  * hosts in the system.  The messenger provides ordered and reliable
31  * delivery.  We tolerate TCP disconnects by reconnecting (with
32  * exponential backoff) in the case of a fault (disconnection, bad
33  * crc, protocol error).  Acks allow sent messages to be discarded by
34  * the sender.
35  */
36 
37 /*
38  * We track the state of the socket on a given connection using
39  * values defined below.  The transition to a new socket state is
40  * handled by a function which verifies we aren't coming from an
41  * unexpected state.
42  *
43  *      --------
44  *      | NEW* |  transient initial state
45  *      --------
46  *          | con_sock_state_init()
47  *          v
48  *      ----------
49  *      | CLOSED |  initialized, but no socket (and no
50  *      ----------  TCP connection)
51  *       ^      \
52  *       |       \ con_sock_state_connecting()
53  *       |        ----------------------
54  *       |                              \
55  *       + con_sock_state_closed()       \
56  *       |+---------------------------    \
57  *       | \                          \    \
58  *       |  -----------                \    \
59  *       |  | CLOSING |  socket event;  \    \
60  *       |  -----------  await close     \    \
61  *       |       ^                        \   |
62  *       |       |                         \  |
63  *       |       + con_sock_state_closing() \ |
64  *       |      / \                         | |
65  *       |     /   ---------------          | |
66  *       |    /                   \         v v
67  *       |   /                    --------------
68  *       |  /    -----------------| CONNECTING |  socket created, TCP
69  *       |  |   /                 --------------  connect initiated
70  *       |  |   | con_sock_state_connected()
71  *       |  |   v
72  *      -------------
73  *      | CONNECTED |  TCP connection established
74  *      -------------
75  *
76  * State values for ceph_connection->sock_state; NEW is assumed to be 0.
77  */
78 
79 #define CON_SOCK_STATE_NEW		0	/* -> CLOSED */
80 #define CON_SOCK_STATE_CLOSED		1	/* -> CONNECTING */
81 #define CON_SOCK_STATE_CONNECTING	2	/* -> CONNECTED or -> CLOSING */
82 #define CON_SOCK_STATE_CONNECTED	3	/* -> CLOSING or -> CLOSED */
83 #define CON_SOCK_STATE_CLOSING		4	/* -> CLOSED */
84 
85 /*
86  * connection states
87  */
88 #define CON_STATE_CLOSED        1  /* -> PREOPEN */
89 #define CON_STATE_PREOPEN       2  /* -> CONNECTING, CLOSED */
90 #define CON_STATE_CONNECTING    3  /* -> NEGOTIATING, CLOSED */
91 #define CON_STATE_NEGOTIATING   4  /* -> OPEN, CLOSED */
92 #define CON_STATE_OPEN          5  /* -> STANDBY, CLOSED */
93 #define CON_STATE_STANDBY       6  /* -> PREOPEN, CLOSED */
94 
95 /*
96  * ceph_connection flag bits
97  */
98 #define CON_FLAG_LOSSYTX           0  /* we can close channel or drop
99 				       * messages on errors */
100 #define CON_FLAG_KEEPALIVE_PENDING 1  /* we need to send a keepalive */
101 #define CON_FLAG_WRITE_PENDING	   2  /* we have data ready to send */
102 #define CON_FLAG_SOCK_CLOSED	   3  /* socket state changed to closed */
103 #define CON_FLAG_BACKOFF           4  /* need to retry queuing delayed work */
104 
105 static bool con_flag_valid(unsigned long con_flag)
106 {
107 	switch (con_flag) {
108 	case CON_FLAG_LOSSYTX:
109 	case CON_FLAG_KEEPALIVE_PENDING:
110 	case CON_FLAG_WRITE_PENDING:
111 	case CON_FLAG_SOCK_CLOSED:
112 	case CON_FLAG_BACKOFF:
113 		return true;
114 	default:
115 		return false;
116 	}
117 }
118 
119 static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
120 {
121 	BUG_ON(!con_flag_valid(con_flag));
122 
123 	clear_bit(con_flag, &con->flags);
124 }
125 
126 static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
127 {
128 	BUG_ON(!con_flag_valid(con_flag));
129 
130 	set_bit(con_flag, &con->flags);
131 }
132 
133 static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
134 {
135 	BUG_ON(!con_flag_valid(con_flag));
136 
137 	return test_bit(con_flag, &con->flags);
138 }
139 
140 static bool con_flag_test_and_clear(struct ceph_connection *con,
141 					unsigned long con_flag)
142 {
143 	BUG_ON(!con_flag_valid(con_flag));
144 
145 	return test_and_clear_bit(con_flag, &con->flags);
146 }
147 
148 static bool con_flag_test_and_set(struct ceph_connection *con,
149 					unsigned long con_flag)
150 {
151 	BUG_ON(!con_flag_valid(con_flag));
152 
153 	return test_and_set_bit(con_flag, &con->flags);
154 }
155 
156 /* Slab caches for frequently-allocated structures */
157 
158 static struct kmem_cache	*ceph_msg_cache;
159 
160 /* static tag bytes (protocol control messages) */
161 static char tag_msg = CEPH_MSGR_TAG_MSG;
162 static char tag_ack = CEPH_MSGR_TAG_ACK;
163 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
164 static char tag_keepalive2 = CEPH_MSGR_TAG_KEEPALIVE2;
165 
166 #ifdef CONFIG_LOCKDEP
167 static struct lock_class_key socket_class;
168 #endif
169 
170 static void queue_con(struct ceph_connection *con);
171 static void cancel_con(struct ceph_connection *con);
172 static void ceph_con_workfn(struct work_struct *);
173 static void con_fault(struct ceph_connection *con);
174 
175 /*
176  * Nicely render a sockaddr as a string.  An array of formatted
177  * strings is used, to approximate reentrancy.
178  */
179 #define ADDR_STR_COUNT_LOG	5	/* log2(# address strings in array) */
180 #define ADDR_STR_COUNT		(1 << ADDR_STR_COUNT_LOG)
181 #define ADDR_STR_COUNT_MASK	(ADDR_STR_COUNT - 1)
182 #define MAX_ADDR_STR_LEN	64	/* 54 is enough */
183 
184 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
185 static atomic_t addr_str_seq = ATOMIC_INIT(0);
186 
187 static struct page *zero_page;		/* used in certain error cases */
188 
189 const char *ceph_pr_addr(const struct ceph_entity_addr *addr)
190 {
191 	int i;
192 	char *s;
193 	struct sockaddr_storage ss = addr->in_addr; /* align */
194 	struct sockaddr_in *in4 = (struct sockaddr_in *)&ss;
195 	struct sockaddr_in6 *in6 = (struct sockaddr_in6 *)&ss;
196 
197 	i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
198 	s = addr_str[i];
199 
200 	switch (ss.ss_family) {
201 	case AF_INET:
202 		snprintf(s, MAX_ADDR_STR_LEN, "(%d)%pI4:%hu",
203 			 le32_to_cpu(addr->type), &in4->sin_addr,
204 			 ntohs(in4->sin_port));
205 		break;
206 
207 	case AF_INET6:
208 		snprintf(s, MAX_ADDR_STR_LEN, "(%d)[%pI6c]:%hu",
209 			 le32_to_cpu(addr->type), &in6->sin6_addr,
210 			 ntohs(in6->sin6_port));
211 		break;
212 
213 	default:
214 		snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
215 			 ss.ss_family);
216 	}
217 
218 	return s;
219 }
220 EXPORT_SYMBOL(ceph_pr_addr);
221 
222 static void encode_my_addr(struct ceph_messenger *msgr)
223 {
224 	memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
225 	ceph_encode_banner_addr(&msgr->my_enc_addr);
226 }
227 
228 /*
229  * work queue for all reading and writing to/from the socket.
230  */
231 static struct workqueue_struct *ceph_msgr_wq;
232 
233 static int ceph_msgr_slab_init(void)
234 {
235 	BUG_ON(ceph_msg_cache);
236 	ceph_msg_cache = KMEM_CACHE(ceph_msg, 0);
237 	if (!ceph_msg_cache)
238 		return -ENOMEM;
239 
240 	return 0;
241 }
242 
243 static void ceph_msgr_slab_exit(void)
244 {
245 	BUG_ON(!ceph_msg_cache);
246 	kmem_cache_destroy(ceph_msg_cache);
247 	ceph_msg_cache = NULL;
248 }
249 
250 static void _ceph_msgr_exit(void)
251 {
252 	if (ceph_msgr_wq) {
253 		destroy_workqueue(ceph_msgr_wq);
254 		ceph_msgr_wq = NULL;
255 	}
256 
257 	BUG_ON(zero_page == NULL);
258 	put_page(zero_page);
259 	zero_page = NULL;
260 
261 	ceph_msgr_slab_exit();
262 }
263 
264 int __init ceph_msgr_init(void)
265 {
266 	if (ceph_msgr_slab_init())
267 		return -ENOMEM;
268 
269 	BUG_ON(zero_page != NULL);
270 	zero_page = ZERO_PAGE(0);
271 	get_page(zero_page);
272 
273 	/*
274 	 * The number of active work items is limited by the number of
275 	 * connections, so leave @max_active at default.
276 	 */
277 	ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
278 	if (ceph_msgr_wq)
279 		return 0;
280 
281 	pr_err("msgr_init failed to create workqueue\n");
282 	_ceph_msgr_exit();
283 
284 	return -ENOMEM;
285 }
286 
287 void ceph_msgr_exit(void)
288 {
289 	BUG_ON(ceph_msgr_wq == NULL);
290 
291 	_ceph_msgr_exit();
292 }
293 
294 void ceph_msgr_flush(void)
295 {
296 	flush_workqueue(ceph_msgr_wq);
297 }
298 EXPORT_SYMBOL(ceph_msgr_flush);
299 
300 /* Connection socket state transition functions */
301 
302 static void con_sock_state_init(struct ceph_connection *con)
303 {
304 	int old_state;
305 
306 	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
307 	if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
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_CLOSED);
311 }
312 
313 static void con_sock_state_connecting(struct ceph_connection *con)
314 {
315 	int old_state;
316 
317 	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
318 	if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
319 		printk("%s: unexpected old state %d\n", __func__, old_state);
320 	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
321 	     CON_SOCK_STATE_CONNECTING);
322 }
323 
324 static void con_sock_state_connected(struct ceph_connection *con)
325 {
326 	int old_state;
327 
328 	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
329 	if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
330 		printk("%s: unexpected old state %d\n", __func__, old_state);
331 	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
332 	     CON_SOCK_STATE_CONNECTED);
333 }
334 
335 static void con_sock_state_closing(struct ceph_connection *con)
336 {
337 	int old_state;
338 
339 	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
340 	if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
341 			old_state != CON_SOCK_STATE_CONNECTED &&
342 			old_state != CON_SOCK_STATE_CLOSING))
343 		printk("%s: unexpected old state %d\n", __func__, old_state);
344 	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
345 	     CON_SOCK_STATE_CLOSING);
346 }
347 
348 static void con_sock_state_closed(struct ceph_connection *con)
349 {
350 	int old_state;
351 
352 	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
353 	if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
354 		    old_state != CON_SOCK_STATE_CLOSING &&
355 		    old_state != CON_SOCK_STATE_CONNECTING &&
356 		    old_state != CON_SOCK_STATE_CLOSED))
357 		printk("%s: unexpected old state %d\n", __func__, old_state);
358 	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
359 	     CON_SOCK_STATE_CLOSED);
360 }
361 
362 /*
363  * socket callback functions
364  */
365 
366 /* data available on socket, or listen socket received a connect */
367 static void ceph_sock_data_ready(struct sock *sk)
368 {
369 	struct ceph_connection *con = sk->sk_user_data;
370 	if (atomic_read(&con->msgr->stopping)) {
371 		return;
372 	}
373 
374 	if (sk->sk_state != TCP_CLOSE_WAIT) {
375 		dout("%s on %p state = %lu, queueing work\n", __func__,
376 		     con, con->state);
377 		queue_con(con);
378 	}
379 }
380 
381 /* socket has buffer space for writing */
382 static void ceph_sock_write_space(struct sock *sk)
383 {
384 	struct ceph_connection *con = sk->sk_user_data;
385 
386 	/* only queue to workqueue if there is data we want to write,
387 	 * and there is sufficient space in the socket buffer to accept
388 	 * more data.  clear SOCK_NOSPACE so that ceph_sock_write_space()
389 	 * doesn't get called again until try_write() fills the socket
390 	 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
391 	 * and net/core/stream.c:sk_stream_write_space().
392 	 */
393 	if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
394 		if (sk_stream_is_writeable(sk)) {
395 			dout("%s %p queueing write work\n", __func__, con);
396 			clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
397 			queue_con(con);
398 		}
399 	} else {
400 		dout("%s %p nothing to write\n", __func__, con);
401 	}
402 }
403 
404 /* socket's state has changed */
405 static void ceph_sock_state_change(struct sock *sk)
406 {
407 	struct ceph_connection *con = sk->sk_user_data;
408 
409 	dout("%s %p state = %lu sk_state = %u\n", __func__,
410 	     con, con->state, sk->sk_state);
411 
412 	switch (sk->sk_state) {
413 	case TCP_CLOSE:
414 		dout("%s TCP_CLOSE\n", __func__);
415 		/* fall through */
416 	case TCP_CLOSE_WAIT:
417 		dout("%s TCP_CLOSE_WAIT\n", __func__);
418 		con_sock_state_closing(con);
419 		con_flag_set(con, CON_FLAG_SOCK_CLOSED);
420 		queue_con(con);
421 		break;
422 	case TCP_ESTABLISHED:
423 		dout("%s TCP_ESTABLISHED\n", __func__);
424 		con_sock_state_connected(con);
425 		queue_con(con);
426 		break;
427 	default:	/* Everything else is uninteresting */
428 		break;
429 	}
430 }
431 
432 /*
433  * set up socket callbacks
434  */
435 static void set_sock_callbacks(struct socket *sock,
436 			       struct ceph_connection *con)
437 {
438 	struct sock *sk = sock->sk;
439 	sk->sk_user_data = con;
440 	sk->sk_data_ready = ceph_sock_data_ready;
441 	sk->sk_write_space = ceph_sock_write_space;
442 	sk->sk_state_change = ceph_sock_state_change;
443 }
444 
445 
446 /*
447  * socket helpers
448  */
449 
450 /*
451  * initiate connection to a remote socket.
452  */
453 static int ceph_tcp_connect(struct ceph_connection *con)
454 {
455 	struct sockaddr_storage ss = con->peer_addr.in_addr; /* align */
456 	struct socket *sock;
457 	unsigned int noio_flag;
458 	int ret;
459 
460 	BUG_ON(con->sock);
461 
462 	/* sock_create_kern() allocates with GFP_KERNEL */
463 	noio_flag = memalloc_noio_save();
464 	ret = sock_create_kern(read_pnet(&con->msgr->net), ss.ss_family,
465 			       SOCK_STREAM, IPPROTO_TCP, &sock);
466 	memalloc_noio_restore(noio_flag);
467 	if (ret)
468 		return ret;
469 	sock->sk->sk_allocation = GFP_NOFS;
470 
471 #ifdef CONFIG_LOCKDEP
472 	lockdep_set_class(&sock->sk->sk_lock, &socket_class);
473 #endif
474 
475 	set_sock_callbacks(sock, con);
476 
477 	dout("connect %s\n", ceph_pr_addr(&con->peer_addr));
478 
479 	con_sock_state_connecting(con);
480 	ret = sock->ops->connect(sock, (struct sockaddr *)&ss, sizeof(ss),
481 				 O_NONBLOCK);
482 	if (ret == -EINPROGRESS) {
483 		dout("connect %s EINPROGRESS sk_state = %u\n",
484 		     ceph_pr_addr(&con->peer_addr),
485 		     sock->sk->sk_state);
486 	} else if (ret < 0) {
487 		pr_err("connect %s error %d\n",
488 		       ceph_pr_addr(&con->peer_addr), ret);
489 		sock_release(sock);
490 		return ret;
491 	}
492 
493 	if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY))
494 		tcp_sock_set_nodelay(sock->sk);
495 
496 	con->sock = sock;
497 	return 0;
498 }
499 
500 /*
501  * If @buf is NULL, discard up to @len bytes.
502  */
503 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
504 {
505 	struct kvec iov = {buf, len};
506 	struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
507 	int r;
508 
509 	if (!buf)
510 		msg.msg_flags |= MSG_TRUNC;
511 
512 	iov_iter_kvec(&msg.msg_iter, READ, &iov, 1, len);
513 	r = sock_recvmsg(sock, &msg, msg.msg_flags);
514 	if (r == -EAGAIN)
515 		r = 0;
516 	return r;
517 }
518 
519 static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
520 		     int page_offset, size_t length)
521 {
522 	struct bio_vec bvec = {
523 		.bv_page = page,
524 		.bv_offset = page_offset,
525 		.bv_len = length
526 	};
527 	struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
528 	int r;
529 
530 	BUG_ON(page_offset + length > PAGE_SIZE);
531 	iov_iter_bvec(&msg.msg_iter, READ, &bvec, 1, length);
532 	r = sock_recvmsg(sock, &msg, msg.msg_flags);
533 	if (r == -EAGAIN)
534 		r = 0;
535 	return r;
536 }
537 
538 /*
539  * write something.  @more is true if caller will be sending more data
540  * shortly.
541  */
542 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
543 			    size_t kvlen, size_t len, bool more)
544 {
545 	struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
546 	int r;
547 
548 	if (more)
549 		msg.msg_flags |= MSG_MORE;
550 	else
551 		msg.msg_flags |= MSG_EOR;  /* superfluous, but what the hell */
552 
553 	r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
554 	if (r == -EAGAIN)
555 		r = 0;
556 	return r;
557 }
558 
559 /*
560  * @more: either or both of MSG_MORE and MSG_SENDPAGE_NOTLAST
561  */
562 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
563 			     int offset, size_t size, int more)
564 {
565 	ssize_t (*sendpage)(struct socket *sock, struct page *page,
566 			    int offset, size_t size, int flags);
567 	int flags = MSG_DONTWAIT | MSG_NOSIGNAL | more;
568 	int ret;
569 
570 	/*
571 	 * sendpage cannot properly handle pages with page_count == 0,
572 	 * we need to fall back to sendmsg if that's the case.
573 	 *
574 	 * Same goes for slab pages: skb_can_coalesce() allows
575 	 * coalescing neighboring slab objects into a single frag which
576 	 * triggers one of hardened usercopy checks.
577 	 */
578 	if (page_count(page) >= 1 && !PageSlab(page))
579 		sendpage = sock->ops->sendpage;
580 	else
581 		sendpage = sock_no_sendpage;
582 
583 	ret = sendpage(sock, page, offset, size, flags);
584 	if (ret == -EAGAIN)
585 		ret = 0;
586 
587 	return ret;
588 }
589 
590 /*
591  * Shutdown/close the socket for the given connection.
592  */
593 static int con_close_socket(struct ceph_connection *con)
594 {
595 	int rc = 0;
596 
597 	dout("con_close_socket on %p sock %p\n", con, con->sock);
598 	if (con->sock) {
599 		rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
600 		sock_release(con->sock);
601 		con->sock = NULL;
602 	}
603 
604 	/*
605 	 * Forcibly clear the SOCK_CLOSED flag.  It gets set
606 	 * independent of the connection mutex, and we could have
607 	 * received a socket close event before we had the chance to
608 	 * shut the socket down.
609 	 */
610 	con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
611 
612 	con_sock_state_closed(con);
613 	return rc;
614 }
615 
616 /*
617  * Reset a connection.  Discard all incoming and outgoing messages
618  * and clear *_seq state.
619  */
620 static void ceph_msg_remove(struct ceph_msg *msg)
621 {
622 	list_del_init(&msg->list_head);
623 
624 	ceph_msg_put(msg);
625 }
626 static void ceph_msg_remove_list(struct list_head *head)
627 {
628 	while (!list_empty(head)) {
629 		struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
630 							list_head);
631 		ceph_msg_remove(msg);
632 	}
633 }
634 
635 static void reset_connection(struct ceph_connection *con)
636 {
637 	/* reset connection, out_queue, msg_ and connect_seq */
638 	/* discard existing out_queue and msg_seq */
639 	dout("reset_connection %p\n", con);
640 	ceph_msg_remove_list(&con->out_queue);
641 	ceph_msg_remove_list(&con->out_sent);
642 
643 	if (con->in_msg) {
644 		BUG_ON(con->in_msg->con != con);
645 		ceph_msg_put(con->in_msg);
646 		con->in_msg = NULL;
647 	}
648 
649 	con->connect_seq = 0;
650 	con->out_seq = 0;
651 	if (con->out_msg) {
652 		BUG_ON(con->out_msg->con != con);
653 		ceph_msg_put(con->out_msg);
654 		con->out_msg = NULL;
655 	}
656 	con->in_seq = 0;
657 	con->in_seq_acked = 0;
658 
659 	con->out_skip = 0;
660 }
661 
662 /*
663  * mark a peer down.  drop any open connections.
664  */
665 void ceph_con_close(struct ceph_connection *con)
666 {
667 	mutex_lock(&con->mutex);
668 	dout("con_close %p peer %s\n", con, ceph_pr_addr(&con->peer_addr));
669 	con->state = CON_STATE_CLOSED;
670 
671 	con_flag_clear(con, CON_FLAG_LOSSYTX);	/* so we retry next connect */
672 	con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
673 	con_flag_clear(con, CON_FLAG_WRITE_PENDING);
674 	con_flag_clear(con, CON_FLAG_BACKOFF);
675 
676 	reset_connection(con);
677 	con->peer_global_seq = 0;
678 	cancel_con(con);
679 	con_close_socket(con);
680 	mutex_unlock(&con->mutex);
681 }
682 EXPORT_SYMBOL(ceph_con_close);
683 
684 /*
685  * Reopen a closed connection, with a new peer address.
686  */
687 void ceph_con_open(struct ceph_connection *con,
688 		   __u8 entity_type, __u64 entity_num,
689 		   struct ceph_entity_addr *addr)
690 {
691 	mutex_lock(&con->mutex);
692 	dout("con_open %p %s\n", con, ceph_pr_addr(addr));
693 
694 	WARN_ON(con->state != CON_STATE_CLOSED);
695 	con->state = CON_STATE_PREOPEN;
696 
697 	con->peer_name.type = (__u8) entity_type;
698 	con->peer_name.num = cpu_to_le64(entity_num);
699 
700 	memcpy(&con->peer_addr, addr, sizeof(*addr));
701 	con->delay = 0;      /* reset backoff memory */
702 	mutex_unlock(&con->mutex);
703 	queue_con(con);
704 }
705 EXPORT_SYMBOL(ceph_con_open);
706 
707 /*
708  * return true if this connection ever successfully opened
709  */
710 bool ceph_con_opened(struct ceph_connection *con)
711 {
712 	return con->connect_seq > 0;
713 }
714 
715 /*
716  * initialize a new connection.
717  */
718 void ceph_con_init(struct ceph_connection *con, void *private,
719 	const struct ceph_connection_operations *ops,
720 	struct ceph_messenger *msgr)
721 {
722 	dout("con_init %p\n", con);
723 	memset(con, 0, sizeof(*con));
724 	con->private = private;
725 	con->ops = ops;
726 	con->msgr = msgr;
727 
728 	con_sock_state_init(con);
729 
730 	mutex_init(&con->mutex);
731 	INIT_LIST_HEAD(&con->out_queue);
732 	INIT_LIST_HEAD(&con->out_sent);
733 	INIT_DELAYED_WORK(&con->work, ceph_con_workfn);
734 
735 	con->state = CON_STATE_CLOSED;
736 }
737 EXPORT_SYMBOL(ceph_con_init);
738 
739 
740 /*
741  * We maintain a global counter to order connection attempts.  Get
742  * a unique seq greater than @gt.
743  */
744 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
745 {
746 	u32 ret;
747 
748 	spin_lock(&msgr->global_seq_lock);
749 	if (msgr->global_seq < gt)
750 		msgr->global_seq = gt;
751 	ret = ++msgr->global_seq;
752 	spin_unlock(&msgr->global_seq_lock);
753 	return ret;
754 }
755 
756 static void con_out_kvec_reset(struct ceph_connection *con)
757 {
758 	BUG_ON(con->out_skip);
759 
760 	con->out_kvec_left = 0;
761 	con->out_kvec_bytes = 0;
762 	con->out_kvec_cur = &con->out_kvec[0];
763 }
764 
765 static void con_out_kvec_add(struct ceph_connection *con,
766 				size_t size, void *data)
767 {
768 	int index = con->out_kvec_left;
769 
770 	BUG_ON(con->out_skip);
771 	BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
772 
773 	con->out_kvec[index].iov_len = size;
774 	con->out_kvec[index].iov_base = data;
775 	con->out_kvec_left++;
776 	con->out_kvec_bytes += size;
777 }
778 
779 /*
780  * Chop off a kvec from the end.  Return residual number of bytes for
781  * that kvec, i.e. how many bytes would have been written if the kvec
782  * hadn't been nuked.
783  */
784 static int con_out_kvec_skip(struct ceph_connection *con)
785 {
786 	int off = con->out_kvec_cur - con->out_kvec;
787 	int skip = 0;
788 
789 	if (con->out_kvec_bytes > 0) {
790 		skip = con->out_kvec[off + con->out_kvec_left - 1].iov_len;
791 		BUG_ON(con->out_kvec_bytes < skip);
792 		BUG_ON(!con->out_kvec_left);
793 		con->out_kvec_bytes -= skip;
794 		con->out_kvec_left--;
795 	}
796 
797 	return skip;
798 }
799 
800 #ifdef CONFIG_BLOCK
801 
802 /*
803  * For a bio data item, a piece is whatever remains of the next
804  * entry in the current bio iovec, or the first entry in the next
805  * bio in the list.
806  */
807 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
808 					size_t length)
809 {
810 	struct ceph_msg_data *data = cursor->data;
811 	struct ceph_bio_iter *it = &cursor->bio_iter;
812 
813 	cursor->resid = min_t(size_t, length, data->bio_length);
814 	*it = data->bio_pos;
815 	if (cursor->resid < it->iter.bi_size)
816 		it->iter.bi_size = cursor->resid;
817 
818 	BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
819 	cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter);
820 }
821 
822 static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
823 						size_t *page_offset,
824 						size_t *length)
825 {
826 	struct bio_vec bv = bio_iter_iovec(cursor->bio_iter.bio,
827 					   cursor->bio_iter.iter);
828 
829 	*page_offset = bv.bv_offset;
830 	*length = bv.bv_len;
831 	return bv.bv_page;
832 }
833 
834 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
835 					size_t bytes)
836 {
837 	struct ceph_bio_iter *it = &cursor->bio_iter;
838 	struct page *page = bio_iter_page(it->bio, it->iter);
839 
840 	BUG_ON(bytes > cursor->resid);
841 	BUG_ON(bytes > bio_iter_len(it->bio, it->iter));
842 	cursor->resid -= bytes;
843 	bio_advance_iter(it->bio, &it->iter, bytes);
844 
845 	if (!cursor->resid) {
846 		BUG_ON(!cursor->last_piece);
847 		return false;   /* no more data */
848 	}
849 
850 	if (!bytes || (it->iter.bi_size && it->iter.bi_bvec_done &&
851 		       page == bio_iter_page(it->bio, it->iter)))
852 		return false;	/* more bytes to process in this segment */
853 
854 	if (!it->iter.bi_size) {
855 		it->bio = it->bio->bi_next;
856 		it->iter = it->bio->bi_iter;
857 		if (cursor->resid < it->iter.bi_size)
858 			it->iter.bi_size = cursor->resid;
859 	}
860 
861 	BUG_ON(cursor->last_piece);
862 	BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
863 	cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter);
864 	return true;
865 }
866 #endif /* CONFIG_BLOCK */
867 
868 static void ceph_msg_data_bvecs_cursor_init(struct ceph_msg_data_cursor *cursor,
869 					size_t length)
870 {
871 	struct ceph_msg_data *data = cursor->data;
872 	struct bio_vec *bvecs = data->bvec_pos.bvecs;
873 
874 	cursor->resid = min_t(size_t, length, data->bvec_pos.iter.bi_size);
875 	cursor->bvec_iter = data->bvec_pos.iter;
876 	cursor->bvec_iter.bi_size = cursor->resid;
877 
878 	BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
879 	cursor->last_piece =
880 	    cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter);
881 }
882 
883 static struct page *ceph_msg_data_bvecs_next(struct ceph_msg_data_cursor *cursor,
884 						size_t *page_offset,
885 						size_t *length)
886 {
887 	struct bio_vec bv = bvec_iter_bvec(cursor->data->bvec_pos.bvecs,
888 					   cursor->bvec_iter);
889 
890 	*page_offset = bv.bv_offset;
891 	*length = bv.bv_len;
892 	return bv.bv_page;
893 }
894 
895 static bool ceph_msg_data_bvecs_advance(struct ceph_msg_data_cursor *cursor,
896 					size_t bytes)
897 {
898 	struct bio_vec *bvecs = cursor->data->bvec_pos.bvecs;
899 	struct page *page = bvec_iter_page(bvecs, cursor->bvec_iter);
900 
901 	BUG_ON(bytes > cursor->resid);
902 	BUG_ON(bytes > bvec_iter_len(bvecs, cursor->bvec_iter));
903 	cursor->resid -= bytes;
904 	bvec_iter_advance(bvecs, &cursor->bvec_iter, bytes);
905 
906 	if (!cursor->resid) {
907 		BUG_ON(!cursor->last_piece);
908 		return false;   /* no more data */
909 	}
910 
911 	if (!bytes || (cursor->bvec_iter.bi_bvec_done &&
912 		       page == bvec_iter_page(bvecs, cursor->bvec_iter)))
913 		return false;	/* more bytes to process in this segment */
914 
915 	BUG_ON(cursor->last_piece);
916 	BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
917 	cursor->last_piece =
918 	    cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter);
919 	return true;
920 }
921 
922 /*
923  * For a page array, a piece comes from the first page in the array
924  * that has not already been fully consumed.
925  */
926 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
927 					size_t length)
928 {
929 	struct ceph_msg_data *data = cursor->data;
930 	int page_count;
931 
932 	BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
933 
934 	BUG_ON(!data->pages);
935 	BUG_ON(!data->length);
936 
937 	cursor->resid = min(length, data->length);
938 	page_count = calc_pages_for(data->alignment, (u64)data->length);
939 	cursor->page_offset = data->alignment & ~PAGE_MASK;
940 	cursor->page_index = 0;
941 	BUG_ON(page_count > (int)USHRT_MAX);
942 	cursor->page_count = (unsigned short)page_count;
943 	BUG_ON(length > SIZE_MAX - cursor->page_offset);
944 	cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
945 }
946 
947 static struct page *
948 ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
949 					size_t *page_offset, size_t *length)
950 {
951 	struct ceph_msg_data *data = cursor->data;
952 
953 	BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
954 
955 	BUG_ON(cursor->page_index >= cursor->page_count);
956 	BUG_ON(cursor->page_offset >= PAGE_SIZE);
957 
958 	*page_offset = cursor->page_offset;
959 	if (cursor->last_piece)
960 		*length = cursor->resid;
961 	else
962 		*length = PAGE_SIZE - *page_offset;
963 
964 	return data->pages[cursor->page_index];
965 }
966 
967 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
968 						size_t bytes)
969 {
970 	BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
971 
972 	BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
973 
974 	/* Advance the cursor page offset */
975 
976 	cursor->resid -= bytes;
977 	cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
978 	if (!bytes || cursor->page_offset)
979 		return false;	/* more bytes to process in the current page */
980 
981 	if (!cursor->resid)
982 		return false;   /* no more data */
983 
984 	/* Move on to the next page; offset is already at 0 */
985 
986 	BUG_ON(cursor->page_index >= cursor->page_count);
987 	cursor->page_index++;
988 	cursor->last_piece = cursor->resid <= PAGE_SIZE;
989 
990 	return true;
991 }
992 
993 /*
994  * For a pagelist, a piece is whatever remains to be consumed in the
995  * first page in the list, or the front of the next page.
996  */
997 static void
998 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
999 					size_t length)
1000 {
1001 	struct ceph_msg_data *data = cursor->data;
1002 	struct ceph_pagelist *pagelist;
1003 	struct page *page;
1004 
1005 	BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1006 
1007 	pagelist = data->pagelist;
1008 	BUG_ON(!pagelist);
1009 
1010 	if (!length)
1011 		return;		/* pagelist can be assigned but empty */
1012 
1013 	BUG_ON(list_empty(&pagelist->head));
1014 	page = list_first_entry(&pagelist->head, struct page, lru);
1015 
1016 	cursor->resid = min(length, pagelist->length);
1017 	cursor->page = page;
1018 	cursor->offset = 0;
1019 	cursor->last_piece = cursor->resid <= PAGE_SIZE;
1020 }
1021 
1022 static struct page *
1023 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
1024 				size_t *page_offset, size_t *length)
1025 {
1026 	struct ceph_msg_data *data = cursor->data;
1027 	struct ceph_pagelist *pagelist;
1028 
1029 	BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1030 
1031 	pagelist = data->pagelist;
1032 	BUG_ON(!pagelist);
1033 
1034 	BUG_ON(!cursor->page);
1035 	BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1036 
1037 	/* offset of first page in pagelist is always 0 */
1038 	*page_offset = cursor->offset & ~PAGE_MASK;
1039 	if (cursor->last_piece)
1040 		*length = cursor->resid;
1041 	else
1042 		*length = PAGE_SIZE - *page_offset;
1043 
1044 	return cursor->page;
1045 }
1046 
1047 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
1048 						size_t bytes)
1049 {
1050 	struct ceph_msg_data *data = cursor->data;
1051 	struct ceph_pagelist *pagelist;
1052 
1053 	BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1054 
1055 	pagelist = data->pagelist;
1056 	BUG_ON(!pagelist);
1057 
1058 	BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1059 	BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1060 
1061 	/* Advance the cursor offset */
1062 
1063 	cursor->resid -= bytes;
1064 	cursor->offset += bytes;
1065 	/* offset of first page in pagelist is always 0 */
1066 	if (!bytes || cursor->offset & ~PAGE_MASK)
1067 		return false;	/* more bytes to process in the current page */
1068 
1069 	if (!cursor->resid)
1070 		return false;   /* no more data */
1071 
1072 	/* Move on to the next page */
1073 
1074 	BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1075 	cursor->page = list_next_entry(cursor->page, lru);
1076 	cursor->last_piece = cursor->resid <= PAGE_SIZE;
1077 
1078 	return true;
1079 }
1080 
1081 /*
1082  * Message data is handled (sent or received) in pieces, where each
1083  * piece resides on a single page.  The network layer might not
1084  * consume an entire piece at once.  A data item's cursor keeps
1085  * track of which piece is next to process and how much remains to
1086  * be processed in that piece.  It also tracks whether the current
1087  * piece is the last one in the data item.
1088  */
1089 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1090 {
1091 	size_t length = cursor->total_resid;
1092 
1093 	switch (cursor->data->type) {
1094 	case CEPH_MSG_DATA_PAGELIST:
1095 		ceph_msg_data_pagelist_cursor_init(cursor, length);
1096 		break;
1097 	case CEPH_MSG_DATA_PAGES:
1098 		ceph_msg_data_pages_cursor_init(cursor, length);
1099 		break;
1100 #ifdef CONFIG_BLOCK
1101 	case CEPH_MSG_DATA_BIO:
1102 		ceph_msg_data_bio_cursor_init(cursor, length);
1103 		break;
1104 #endif /* CONFIG_BLOCK */
1105 	case CEPH_MSG_DATA_BVECS:
1106 		ceph_msg_data_bvecs_cursor_init(cursor, length);
1107 		break;
1108 	case CEPH_MSG_DATA_NONE:
1109 	default:
1110 		/* BUG(); */
1111 		break;
1112 	}
1113 	cursor->need_crc = true;
1114 }
1115 
1116 static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1117 {
1118 	struct ceph_msg_data_cursor *cursor = &msg->cursor;
1119 
1120 	BUG_ON(!length);
1121 	BUG_ON(length > msg->data_length);
1122 	BUG_ON(!msg->num_data_items);
1123 
1124 	cursor->total_resid = length;
1125 	cursor->data = msg->data;
1126 
1127 	__ceph_msg_data_cursor_init(cursor);
1128 }
1129 
1130 /*
1131  * Return the page containing the next piece to process for a given
1132  * data item, and supply the page offset and length of that piece.
1133  * Indicate whether this is the last piece in this data item.
1134  */
1135 static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1136 					size_t *page_offset, size_t *length,
1137 					bool *last_piece)
1138 {
1139 	struct page *page;
1140 
1141 	switch (cursor->data->type) {
1142 	case CEPH_MSG_DATA_PAGELIST:
1143 		page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1144 		break;
1145 	case CEPH_MSG_DATA_PAGES:
1146 		page = ceph_msg_data_pages_next(cursor, page_offset, length);
1147 		break;
1148 #ifdef CONFIG_BLOCK
1149 	case CEPH_MSG_DATA_BIO:
1150 		page = ceph_msg_data_bio_next(cursor, page_offset, length);
1151 		break;
1152 #endif /* CONFIG_BLOCK */
1153 	case CEPH_MSG_DATA_BVECS:
1154 		page = ceph_msg_data_bvecs_next(cursor, page_offset, length);
1155 		break;
1156 	case CEPH_MSG_DATA_NONE:
1157 	default:
1158 		page = NULL;
1159 		break;
1160 	}
1161 
1162 	BUG_ON(!page);
1163 	BUG_ON(*page_offset + *length > PAGE_SIZE);
1164 	BUG_ON(!*length);
1165 	BUG_ON(*length > cursor->resid);
1166 	if (last_piece)
1167 		*last_piece = cursor->last_piece;
1168 
1169 	return page;
1170 }
1171 
1172 /*
1173  * Returns true if the result moves the cursor on to the next piece
1174  * of the data item.
1175  */
1176 static void ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1177 				  size_t bytes)
1178 {
1179 	bool new_piece;
1180 
1181 	BUG_ON(bytes > cursor->resid);
1182 	switch (cursor->data->type) {
1183 	case CEPH_MSG_DATA_PAGELIST:
1184 		new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1185 		break;
1186 	case CEPH_MSG_DATA_PAGES:
1187 		new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1188 		break;
1189 #ifdef CONFIG_BLOCK
1190 	case CEPH_MSG_DATA_BIO:
1191 		new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1192 		break;
1193 #endif /* CONFIG_BLOCK */
1194 	case CEPH_MSG_DATA_BVECS:
1195 		new_piece = ceph_msg_data_bvecs_advance(cursor, bytes);
1196 		break;
1197 	case CEPH_MSG_DATA_NONE:
1198 	default:
1199 		BUG();
1200 		break;
1201 	}
1202 	cursor->total_resid -= bytes;
1203 
1204 	if (!cursor->resid && cursor->total_resid) {
1205 		WARN_ON(!cursor->last_piece);
1206 		cursor->data++;
1207 		__ceph_msg_data_cursor_init(cursor);
1208 		new_piece = true;
1209 	}
1210 	cursor->need_crc = new_piece;
1211 }
1212 
1213 static size_t sizeof_footer(struct ceph_connection *con)
1214 {
1215 	return (con->peer_features & CEPH_FEATURE_MSG_AUTH) ?
1216 	    sizeof(struct ceph_msg_footer) :
1217 	    sizeof(struct ceph_msg_footer_old);
1218 }
1219 
1220 static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
1221 {
1222 	/* Initialize data cursor */
1223 
1224 	ceph_msg_data_cursor_init(msg, (size_t)data_len);
1225 }
1226 
1227 /*
1228  * Prepare footer for currently outgoing message, and finish things
1229  * off.  Assumes out_kvec* are already valid.. we just add on to the end.
1230  */
1231 static void prepare_write_message_footer(struct ceph_connection *con)
1232 {
1233 	struct ceph_msg *m = con->out_msg;
1234 
1235 	m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1236 
1237 	dout("prepare_write_message_footer %p\n", con);
1238 	con_out_kvec_add(con, sizeof_footer(con), &m->footer);
1239 	if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
1240 		if (con->ops->sign_message)
1241 			con->ops->sign_message(m);
1242 		else
1243 			m->footer.sig = 0;
1244 	} else {
1245 		m->old_footer.flags = m->footer.flags;
1246 	}
1247 	con->out_more = m->more_to_follow;
1248 	con->out_msg_done = true;
1249 }
1250 
1251 /*
1252  * Prepare headers for the next outgoing message.
1253  */
1254 static void prepare_write_message(struct ceph_connection *con)
1255 {
1256 	struct ceph_msg *m;
1257 	u32 crc;
1258 
1259 	con_out_kvec_reset(con);
1260 	con->out_msg_done = false;
1261 
1262 	/* Sneak an ack in there first?  If we can get it into the same
1263 	 * TCP packet that's a good thing. */
1264 	if (con->in_seq > con->in_seq_acked) {
1265 		con->in_seq_acked = con->in_seq;
1266 		con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1267 		con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1268 		con_out_kvec_add(con, sizeof (con->out_temp_ack),
1269 			&con->out_temp_ack);
1270 	}
1271 
1272 	BUG_ON(list_empty(&con->out_queue));
1273 	m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1274 	con->out_msg = m;
1275 	BUG_ON(m->con != con);
1276 
1277 	/* put message on sent list */
1278 	ceph_msg_get(m);
1279 	list_move_tail(&m->list_head, &con->out_sent);
1280 
1281 	/*
1282 	 * only assign outgoing seq # if we haven't sent this message
1283 	 * yet.  if it is requeued, resend with it's original seq.
1284 	 */
1285 	if (m->needs_out_seq) {
1286 		m->hdr.seq = cpu_to_le64(++con->out_seq);
1287 		m->needs_out_seq = false;
1288 
1289 		if (con->ops->reencode_message)
1290 			con->ops->reencode_message(m);
1291 	}
1292 
1293 	dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1294 	     m, con->out_seq, le16_to_cpu(m->hdr.type),
1295 	     le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1296 	     m->data_length);
1297 	WARN_ON(m->front.iov_len != le32_to_cpu(m->hdr.front_len));
1298 	WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
1299 
1300 	/* tag + hdr + front + middle */
1301 	con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1302 	con_out_kvec_add(con, sizeof(con->out_hdr), &con->out_hdr);
1303 	con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1304 
1305 	if (m->middle)
1306 		con_out_kvec_add(con, m->middle->vec.iov_len,
1307 			m->middle->vec.iov_base);
1308 
1309 	/* fill in hdr crc and finalize hdr */
1310 	crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1311 	con->out_msg->hdr.crc = cpu_to_le32(crc);
1312 	memcpy(&con->out_hdr, &con->out_msg->hdr, sizeof(con->out_hdr));
1313 
1314 	/* fill in front and middle crc, footer */
1315 	crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1316 	con->out_msg->footer.front_crc = cpu_to_le32(crc);
1317 	if (m->middle) {
1318 		crc = crc32c(0, m->middle->vec.iov_base,
1319 				m->middle->vec.iov_len);
1320 		con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1321 	} else
1322 		con->out_msg->footer.middle_crc = 0;
1323 	dout("%s front_crc %u middle_crc %u\n", __func__,
1324 	     le32_to_cpu(con->out_msg->footer.front_crc),
1325 	     le32_to_cpu(con->out_msg->footer.middle_crc));
1326 	con->out_msg->footer.flags = 0;
1327 
1328 	/* is there a data payload? */
1329 	con->out_msg->footer.data_crc = 0;
1330 	if (m->data_length) {
1331 		prepare_message_data(con->out_msg, m->data_length);
1332 		con->out_more = 1;  /* data + footer will follow */
1333 	} else {
1334 		/* no, queue up footer too and be done */
1335 		prepare_write_message_footer(con);
1336 	}
1337 
1338 	con_flag_set(con, CON_FLAG_WRITE_PENDING);
1339 }
1340 
1341 /*
1342  * Prepare an ack.
1343  */
1344 static void prepare_write_ack(struct ceph_connection *con)
1345 {
1346 	dout("prepare_write_ack %p %llu -> %llu\n", con,
1347 	     con->in_seq_acked, con->in_seq);
1348 	con->in_seq_acked = con->in_seq;
1349 
1350 	con_out_kvec_reset(con);
1351 
1352 	con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1353 
1354 	con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1355 	con_out_kvec_add(con, sizeof (con->out_temp_ack),
1356 				&con->out_temp_ack);
1357 
1358 	con->out_more = 1;  /* more will follow.. eventually.. */
1359 	con_flag_set(con, CON_FLAG_WRITE_PENDING);
1360 }
1361 
1362 /*
1363  * Prepare to share the seq during handshake
1364  */
1365 static void prepare_write_seq(struct ceph_connection *con)
1366 {
1367 	dout("prepare_write_seq %p %llu -> %llu\n", con,
1368 	     con->in_seq_acked, con->in_seq);
1369 	con->in_seq_acked = con->in_seq;
1370 
1371 	con_out_kvec_reset(con);
1372 
1373 	con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1374 	con_out_kvec_add(con, sizeof (con->out_temp_ack),
1375 			 &con->out_temp_ack);
1376 
1377 	con_flag_set(con, CON_FLAG_WRITE_PENDING);
1378 }
1379 
1380 /*
1381  * Prepare to write keepalive byte.
1382  */
1383 static void prepare_write_keepalive(struct ceph_connection *con)
1384 {
1385 	dout("prepare_write_keepalive %p\n", con);
1386 	con_out_kvec_reset(con);
1387 	if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
1388 		struct timespec64 now;
1389 
1390 		ktime_get_real_ts64(&now);
1391 		con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
1392 		ceph_encode_timespec64(&con->out_temp_keepalive2, &now);
1393 		con_out_kvec_add(con, sizeof(con->out_temp_keepalive2),
1394 				 &con->out_temp_keepalive2);
1395 	} else {
1396 		con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive);
1397 	}
1398 	con_flag_set(con, CON_FLAG_WRITE_PENDING);
1399 }
1400 
1401 /*
1402  * Connection negotiation.
1403  */
1404 
1405 static int get_connect_authorizer(struct ceph_connection *con)
1406 {
1407 	struct ceph_auth_handshake *auth;
1408 	int auth_proto;
1409 
1410 	if (!con->ops->get_authorizer) {
1411 		con->auth = NULL;
1412 		con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1413 		con->out_connect.authorizer_len = 0;
1414 		return 0;
1415 	}
1416 
1417 	auth = con->ops->get_authorizer(con, &auth_proto, con->auth_retry);
1418 	if (IS_ERR(auth))
1419 		return PTR_ERR(auth);
1420 
1421 	con->auth = auth;
1422 	con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1423 	con->out_connect.authorizer_len = cpu_to_le32(auth->authorizer_buf_len);
1424 	return 0;
1425 }
1426 
1427 /*
1428  * We connected to a peer and are saying hello.
1429  */
1430 static void prepare_write_banner(struct ceph_connection *con)
1431 {
1432 	con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1433 	con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1434 					&con->msgr->my_enc_addr);
1435 
1436 	con->out_more = 0;
1437 	con_flag_set(con, CON_FLAG_WRITE_PENDING);
1438 }
1439 
1440 static void __prepare_write_connect(struct ceph_connection *con)
1441 {
1442 	con_out_kvec_add(con, sizeof(con->out_connect), &con->out_connect);
1443 	if (con->auth)
1444 		con_out_kvec_add(con, con->auth->authorizer_buf_len,
1445 				 con->auth->authorizer_buf);
1446 
1447 	con->out_more = 0;
1448 	con_flag_set(con, CON_FLAG_WRITE_PENDING);
1449 }
1450 
1451 static int prepare_write_connect(struct ceph_connection *con)
1452 {
1453 	unsigned int global_seq = get_global_seq(con->msgr, 0);
1454 	int proto;
1455 	int ret;
1456 
1457 	switch (con->peer_name.type) {
1458 	case CEPH_ENTITY_TYPE_MON:
1459 		proto = CEPH_MONC_PROTOCOL;
1460 		break;
1461 	case CEPH_ENTITY_TYPE_OSD:
1462 		proto = CEPH_OSDC_PROTOCOL;
1463 		break;
1464 	case CEPH_ENTITY_TYPE_MDS:
1465 		proto = CEPH_MDSC_PROTOCOL;
1466 		break;
1467 	default:
1468 		BUG();
1469 	}
1470 
1471 	dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1472 	     con->connect_seq, global_seq, proto);
1473 
1474 	con->out_connect.features =
1475 	    cpu_to_le64(from_msgr(con->msgr)->supported_features);
1476 	con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1477 	con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1478 	con->out_connect.global_seq = cpu_to_le32(global_seq);
1479 	con->out_connect.protocol_version = cpu_to_le32(proto);
1480 	con->out_connect.flags = 0;
1481 
1482 	ret = get_connect_authorizer(con);
1483 	if (ret)
1484 		return ret;
1485 
1486 	__prepare_write_connect(con);
1487 	return 0;
1488 }
1489 
1490 /*
1491  * write as much of pending kvecs to the socket as we can.
1492  *  1 -> done
1493  *  0 -> socket full, but more to do
1494  * <0 -> error
1495  */
1496 static int write_partial_kvec(struct ceph_connection *con)
1497 {
1498 	int ret;
1499 
1500 	dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1501 	while (con->out_kvec_bytes > 0) {
1502 		ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1503 				       con->out_kvec_left, con->out_kvec_bytes,
1504 				       con->out_more);
1505 		if (ret <= 0)
1506 			goto out;
1507 		con->out_kvec_bytes -= ret;
1508 		if (con->out_kvec_bytes == 0)
1509 			break;            /* done */
1510 
1511 		/* account for full iov entries consumed */
1512 		while (ret >= con->out_kvec_cur->iov_len) {
1513 			BUG_ON(!con->out_kvec_left);
1514 			ret -= con->out_kvec_cur->iov_len;
1515 			con->out_kvec_cur++;
1516 			con->out_kvec_left--;
1517 		}
1518 		/* and for a partially-consumed entry */
1519 		if (ret) {
1520 			con->out_kvec_cur->iov_len -= ret;
1521 			con->out_kvec_cur->iov_base += ret;
1522 		}
1523 	}
1524 	con->out_kvec_left = 0;
1525 	ret = 1;
1526 out:
1527 	dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1528 	     con->out_kvec_bytes, con->out_kvec_left, ret);
1529 	return ret;  /* done! */
1530 }
1531 
1532 static u32 ceph_crc32c_page(u32 crc, struct page *page,
1533 				unsigned int page_offset,
1534 				unsigned int length)
1535 {
1536 	char *kaddr;
1537 
1538 	kaddr = kmap(page);
1539 	BUG_ON(kaddr == NULL);
1540 	crc = crc32c(crc, kaddr + page_offset, length);
1541 	kunmap(page);
1542 
1543 	return crc;
1544 }
1545 /*
1546  * Write as much message data payload as we can.  If we finish, queue
1547  * up the footer.
1548  *  1 -> done, footer is now queued in out_kvec[].
1549  *  0 -> socket full, but more to do
1550  * <0 -> error
1551  */
1552 static int write_partial_message_data(struct ceph_connection *con)
1553 {
1554 	struct ceph_msg *msg = con->out_msg;
1555 	struct ceph_msg_data_cursor *cursor = &msg->cursor;
1556 	bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
1557 	int more = MSG_MORE | MSG_SENDPAGE_NOTLAST;
1558 	u32 crc;
1559 
1560 	dout("%s %p msg %p\n", __func__, con, msg);
1561 
1562 	if (!msg->num_data_items)
1563 		return -EINVAL;
1564 
1565 	/*
1566 	 * Iterate through each page that contains data to be
1567 	 * written, and send as much as possible for each.
1568 	 *
1569 	 * If we are calculating the data crc (the default), we will
1570 	 * need to map the page.  If we have no pages, they have
1571 	 * been revoked, so use the zero page.
1572 	 */
1573 	crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1574 	while (cursor->total_resid) {
1575 		struct page *page;
1576 		size_t page_offset;
1577 		size_t length;
1578 		int ret;
1579 
1580 		if (!cursor->resid) {
1581 			ceph_msg_data_advance(cursor, 0);
1582 			continue;
1583 		}
1584 
1585 		page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
1586 		if (length == cursor->total_resid)
1587 			more = MSG_MORE;
1588 		ret = ceph_tcp_sendpage(con->sock, page, page_offset, length,
1589 					more);
1590 		if (ret <= 0) {
1591 			if (do_datacrc)
1592 				msg->footer.data_crc = cpu_to_le32(crc);
1593 
1594 			return ret;
1595 		}
1596 		if (do_datacrc && cursor->need_crc)
1597 			crc = ceph_crc32c_page(crc, page, page_offset, length);
1598 		ceph_msg_data_advance(cursor, (size_t)ret);
1599 	}
1600 
1601 	dout("%s %p msg %p done\n", __func__, con, msg);
1602 
1603 	/* prepare and queue up footer, too */
1604 	if (do_datacrc)
1605 		msg->footer.data_crc = cpu_to_le32(crc);
1606 	else
1607 		msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1608 	con_out_kvec_reset(con);
1609 	prepare_write_message_footer(con);
1610 
1611 	return 1;	/* must return > 0 to indicate success */
1612 }
1613 
1614 /*
1615  * write some zeros
1616  */
1617 static int write_partial_skip(struct ceph_connection *con)
1618 {
1619 	int more = MSG_MORE | MSG_SENDPAGE_NOTLAST;
1620 	int ret;
1621 
1622 	dout("%s %p %d left\n", __func__, con, con->out_skip);
1623 	while (con->out_skip > 0) {
1624 		size_t size = min(con->out_skip, (int) PAGE_SIZE);
1625 
1626 		if (size == con->out_skip)
1627 			more = MSG_MORE;
1628 		ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, more);
1629 		if (ret <= 0)
1630 			goto out;
1631 		con->out_skip -= ret;
1632 	}
1633 	ret = 1;
1634 out:
1635 	return ret;
1636 }
1637 
1638 /*
1639  * Prepare to read connection handshake, or an ack.
1640  */
1641 static void prepare_read_banner(struct ceph_connection *con)
1642 {
1643 	dout("prepare_read_banner %p\n", con);
1644 	con->in_base_pos = 0;
1645 }
1646 
1647 static void prepare_read_connect(struct ceph_connection *con)
1648 {
1649 	dout("prepare_read_connect %p\n", con);
1650 	con->in_base_pos = 0;
1651 }
1652 
1653 static void prepare_read_ack(struct ceph_connection *con)
1654 {
1655 	dout("prepare_read_ack %p\n", con);
1656 	con->in_base_pos = 0;
1657 }
1658 
1659 static void prepare_read_seq(struct ceph_connection *con)
1660 {
1661 	dout("prepare_read_seq %p\n", con);
1662 	con->in_base_pos = 0;
1663 	con->in_tag = CEPH_MSGR_TAG_SEQ;
1664 }
1665 
1666 static void prepare_read_tag(struct ceph_connection *con)
1667 {
1668 	dout("prepare_read_tag %p\n", con);
1669 	con->in_base_pos = 0;
1670 	con->in_tag = CEPH_MSGR_TAG_READY;
1671 }
1672 
1673 static void prepare_read_keepalive_ack(struct ceph_connection *con)
1674 {
1675 	dout("prepare_read_keepalive_ack %p\n", con);
1676 	con->in_base_pos = 0;
1677 }
1678 
1679 /*
1680  * Prepare to read a message.
1681  */
1682 static int prepare_read_message(struct ceph_connection *con)
1683 {
1684 	dout("prepare_read_message %p\n", con);
1685 	BUG_ON(con->in_msg != NULL);
1686 	con->in_base_pos = 0;
1687 	con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1688 	return 0;
1689 }
1690 
1691 
1692 static int read_partial(struct ceph_connection *con,
1693 			int end, int size, void *object)
1694 {
1695 	while (con->in_base_pos < end) {
1696 		int left = end - con->in_base_pos;
1697 		int have = size - left;
1698 		int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1699 		if (ret <= 0)
1700 			return ret;
1701 		con->in_base_pos += ret;
1702 	}
1703 	return 1;
1704 }
1705 
1706 
1707 /*
1708  * Read all or part of the connect-side handshake on a new connection
1709  */
1710 static int read_partial_banner(struct ceph_connection *con)
1711 {
1712 	int size;
1713 	int end;
1714 	int ret;
1715 
1716 	dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1717 
1718 	/* peer's banner */
1719 	size = strlen(CEPH_BANNER);
1720 	end = size;
1721 	ret = read_partial(con, end, size, con->in_banner);
1722 	if (ret <= 0)
1723 		goto out;
1724 
1725 	size = sizeof (con->actual_peer_addr);
1726 	end += size;
1727 	ret = read_partial(con, end, size, &con->actual_peer_addr);
1728 	if (ret <= 0)
1729 		goto out;
1730 	ceph_decode_banner_addr(&con->actual_peer_addr);
1731 
1732 	size = sizeof (con->peer_addr_for_me);
1733 	end += size;
1734 	ret = read_partial(con, end, size, &con->peer_addr_for_me);
1735 	if (ret <= 0)
1736 		goto out;
1737 	ceph_decode_banner_addr(&con->peer_addr_for_me);
1738 
1739 out:
1740 	return ret;
1741 }
1742 
1743 static int read_partial_connect(struct ceph_connection *con)
1744 {
1745 	int size;
1746 	int end;
1747 	int ret;
1748 
1749 	dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1750 
1751 	size = sizeof (con->in_reply);
1752 	end = size;
1753 	ret = read_partial(con, end, size, &con->in_reply);
1754 	if (ret <= 0)
1755 		goto out;
1756 
1757 	if (con->auth) {
1758 		size = le32_to_cpu(con->in_reply.authorizer_len);
1759 		if (size > con->auth->authorizer_reply_buf_len) {
1760 			pr_err("authorizer reply too big: %d > %zu\n", size,
1761 			       con->auth->authorizer_reply_buf_len);
1762 			ret = -EINVAL;
1763 			goto out;
1764 		}
1765 
1766 		end += size;
1767 		ret = read_partial(con, end, size,
1768 				   con->auth->authorizer_reply_buf);
1769 		if (ret <= 0)
1770 			goto out;
1771 	}
1772 
1773 	dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1774 	     con, (int)con->in_reply.tag,
1775 	     le32_to_cpu(con->in_reply.connect_seq),
1776 	     le32_to_cpu(con->in_reply.global_seq));
1777 out:
1778 	return ret;
1779 }
1780 
1781 /*
1782  * Verify the hello banner looks okay.
1783  */
1784 static int verify_hello(struct ceph_connection *con)
1785 {
1786 	if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1787 		pr_err("connect to %s got bad banner\n",
1788 		       ceph_pr_addr(&con->peer_addr));
1789 		con->error_msg = "protocol error, bad banner";
1790 		return -1;
1791 	}
1792 	return 0;
1793 }
1794 
1795 static bool addr_is_blank(struct ceph_entity_addr *addr)
1796 {
1797 	struct sockaddr_storage ss = addr->in_addr; /* align */
1798 	struct in_addr *addr4 = &((struct sockaddr_in *)&ss)->sin_addr;
1799 	struct in6_addr *addr6 = &((struct sockaddr_in6 *)&ss)->sin6_addr;
1800 
1801 	switch (ss.ss_family) {
1802 	case AF_INET:
1803 		return addr4->s_addr == htonl(INADDR_ANY);
1804 	case AF_INET6:
1805 		return ipv6_addr_any(addr6);
1806 	default:
1807 		return true;
1808 	}
1809 }
1810 
1811 static int addr_port(struct ceph_entity_addr *addr)
1812 {
1813 	switch (get_unaligned(&addr->in_addr.ss_family)) {
1814 	case AF_INET:
1815 		return ntohs(get_unaligned(&((struct sockaddr_in *)&addr->in_addr)->sin_port));
1816 	case AF_INET6:
1817 		return ntohs(get_unaligned(&((struct sockaddr_in6 *)&addr->in_addr)->sin6_port));
1818 	}
1819 	return 0;
1820 }
1821 
1822 static void addr_set_port(struct ceph_entity_addr *addr, int p)
1823 {
1824 	switch (get_unaligned(&addr->in_addr.ss_family)) {
1825 	case AF_INET:
1826 		put_unaligned(htons(p), &((struct sockaddr_in *)&addr->in_addr)->sin_port);
1827 		break;
1828 	case AF_INET6:
1829 		put_unaligned(htons(p), &((struct sockaddr_in6 *)&addr->in_addr)->sin6_port);
1830 		break;
1831 	}
1832 }
1833 
1834 /*
1835  * Unlike other *_pton function semantics, zero indicates success.
1836  */
1837 static int ceph_pton(const char *str, size_t len, struct ceph_entity_addr *addr,
1838 		char delim, const char **ipend)
1839 {
1840 	memset(&addr->in_addr, 0, sizeof(addr->in_addr));
1841 
1842 	if (in4_pton(str, len, (u8 *)&((struct sockaddr_in *)&addr->in_addr)->sin_addr.s_addr, delim, ipend)) {
1843 		put_unaligned(AF_INET, &addr->in_addr.ss_family);
1844 		return 0;
1845 	}
1846 
1847 	if (in6_pton(str, len, (u8 *)&((struct sockaddr_in6 *)&addr->in_addr)->sin6_addr.s6_addr, delim, ipend)) {
1848 		put_unaligned(AF_INET6, &addr->in_addr.ss_family);
1849 		return 0;
1850 	}
1851 
1852 	return -EINVAL;
1853 }
1854 
1855 /*
1856  * Extract hostname string and resolve using kernel DNS facility.
1857  */
1858 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1859 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1860 		struct ceph_entity_addr *addr, char delim, const char **ipend)
1861 {
1862 	const char *end, *delim_p;
1863 	char *colon_p, *ip_addr = NULL;
1864 	int ip_len, ret;
1865 
1866 	/*
1867 	 * The end of the hostname occurs immediately preceding the delimiter or
1868 	 * the port marker (':') where the delimiter takes precedence.
1869 	 */
1870 	delim_p = memchr(name, delim, namelen);
1871 	colon_p = memchr(name, ':', namelen);
1872 
1873 	if (delim_p && colon_p)
1874 		end = delim_p < colon_p ? delim_p : colon_p;
1875 	else if (!delim_p && colon_p)
1876 		end = colon_p;
1877 	else {
1878 		end = delim_p;
1879 		if (!end) /* case: hostname:/ */
1880 			end = name + namelen;
1881 	}
1882 
1883 	if (end <= name)
1884 		return -EINVAL;
1885 
1886 	/* do dns_resolve upcall */
1887 	ip_len = dns_query(current->nsproxy->net_ns,
1888 			   NULL, name, end - name, NULL, &ip_addr, NULL, false);
1889 	if (ip_len > 0)
1890 		ret = ceph_pton(ip_addr, ip_len, addr, -1, NULL);
1891 	else
1892 		ret = -ESRCH;
1893 
1894 	kfree(ip_addr);
1895 
1896 	*ipend = end;
1897 
1898 	pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1899 			ret, ret ? "failed" : ceph_pr_addr(addr));
1900 
1901 	return ret;
1902 }
1903 #else
1904 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1905 		struct ceph_entity_addr *addr, char delim, const char **ipend)
1906 {
1907 	return -EINVAL;
1908 }
1909 #endif
1910 
1911 /*
1912  * Parse a server name (IP or hostname). If a valid IP address is not found
1913  * then try to extract a hostname to resolve using userspace DNS upcall.
1914  */
1915 static int ceph_parse_server_name(const char *name, size_t namelen,
1916 		struct ceph_entity_addr *addr, char delim, const char **ipend)
1917 {
1918 	int ret;
1919 
1920 	ret = ceph_pton(name, namelen, addr, delim, ipend);
1921 	if (ret)
1922 		ret = ceph_dns_resolve_name(name, namelen, addr, delim, ipend);
1923 
1924 	return ret;
1925 }
1926 
1927 /*
1928  * Parse an ip[:port] list into an addr array.  Use the default
1929  * monitor port if a port isn't specified.
1930  */
1931 int ceph_parse_ips(const char *c, const char *end,
1932 		   struct ceph_entity_addr *addr,
1933 		   int max_count, int *count)
1934 {
1935 	int i, ret = -EINVAL;
1936 	const char *p = c;
1937 
1938 	dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1939 	for (i = 0; i < max_count; i++) {
1940 		const char *ipend;
1941 		int port;
1942 		char delim = ',';
1943 
1944 		if (*p == '[') {
1945 			delim = ']';
1946 			p++;
1947 		}
1948 
1949 		ret = ceph_parse_server_name(p, end - p, &addr[i], delim, &ipend);
1950 		if (ret)
1951 			goto bad;
1952 		ret = -EINVAL;
1953 
1954 		p = ipend;
1955 
1956 		if (delim == ']') {
1957 			if (*p != ']') {
1958 				dout("missing matching ']'\n");
1959 				goto bad;
1960 			}
1961 			p++;
1962 		}
1963 
1964 		/* port? */
1965 		if (p < end && *p == ':') {
1966 			port = 0;
1967 			p++;
1968 			while (p < end && *p >= '0' && *p <= '9') {
1969 				port = (port * 10) + (*p - '0');
1970 				p++;
1971 			}
1972 			if (port == 0)
1973 				port = CEPH_MON_PORT;
1974 			else if (port > 65535)
1975 				goto bad;
1976 		} else {
1977 			port = CEPH_MON_PORT;
1978 		}
1979 
1980 		addr_set_port(&addr[i], port);
1981 		addr[i].type = CEPH_ENTITY_ADDR_TYPE_LEGACY;
1982 
1983 		dout("parse_ips got %s\n", ceph_pr_addr(&addr[i]));
1984 
1985 		if (p == end)
1986 			break;
1987 		if (*p != ',')
1988 			goto bad;
1989 		p++;
1990 	}
1991 
1992 	if (p != end)
1993 		goto bad;
1994 
1995 	if (count)
1996 		*count = i + 1;
1997 	return 0;
1998 
1999 bad:
2000 	return ret;
2001 }
2002 
2003 static int process_banner(struct ceph_connection *con)
2004 {
2005 	dout("process_banner on %p\n", con);
2006 
2007 	if (verify_hello(con) < 0)
2008 		return -1;
2009 
2010 	/*
2011 	 * Make sure the other end is who we wanted.  note that the other
2012 	 * end may not yet know their ip address, so if it's 0.0.0.0, give
2013 	 * them the benefit of the doubt.
2014 	 */
2015 	if (memcmp(&con->peer_addr, &con->actual_peer_addr,
2016 		   sizeof(con->peer_addr)) != 0 &&
2017 	    !(addr_is_blank(&con->actual_peer_addr) &&
2018 	      con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
2019 		pr_warn("wrong peer, want %s/%d, got %s/%d\n",
2020 			ceph_pr_addr(&con->peer_addr),
2021 			(int)le32_to_cpu(con->peer_addr.nonce),
2022 			ceph_pr_addr(&con->actual_peer_addr),
2023 			(int)le32_to_cpu(con->actual_peer_addr.nonce));
2024 		con->error_msg = "wrong peer at address";
2025 		return -1;
2026 	}
2027 
2028 	/*
2029 	 * did we learn our address?
2030 	 */
2031 	if (addr_is_blank(&con->msgr->inst.addr)) {
2032 		int port = addr_port(&con->msgr->inst.addr);
2033 
2034 		memcpy(&con->msgr->inst.addr.in_addr,
2035 		       &con->peer_addr_for_me.in_addr,
2036 		       sizeof(con->peer_addr_for_me.in_addr));
2037 		addr_set_port(&con->msgr->inst.addr, port);
2038 		encode_my_addr(con->msgr);
2039 		dout("process_banner learned my addr is %s\n",
2040 		     ceph_pr_addr(&con->msgr->inst.addr));
2041 	}
2042 
2043 	return 0;
2044 }
2045 
2046 static int process_connect(struct ceph_connection *con)
2047 {
2048 	u64 sup_feat = from_msgr(con->msgr)->supported_features;
2049 	u64 req_feat = from_msgr(con->msgr)->required_features;
2050 	u64 server_feat = le64_to_cpu(con->in_reply.features);
2051 	int ret;
2052 
2053 	dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
2054 
2055 	if (con->auth) {
2056 		int len = le32_to_cpu(con->in_reply.authorizer_len);
2057 
2058 		/*
2059 		 * Any connection that defines ->get_authorizer()
2060 		 * should also define ->add_authorizer_challenge() and
2061 		 * ->verify_authorizer_reply().
2062 		 *
2063 		 * See get_connect_authorizer().
2064 		 */
2065 		if (con->in_reply.tag == CEPH_MSGR_TAG_CHALLENGE_AUTHORIZER) {
2066 			ret = con->ops->add_authorizer_challenge(
2067 				    con, con->auth->authorizer_reply_buf, len);
2068 			if (ret < 0)
2069 				return ret;
2070 
2071 			con_out_kvec_reset(con);
2072 			__prepare_write_connect(con);
2073 			prepare_read_connect(con);
2074 			return 0;
2075 		}
2076 
2077 		if (len) {
2078 			ret = con->ops->verify_authorizer_reply(con);
2079 			if (ret < 0) {
2080 				con->error_msg = "bad authorize reply";
2081 				return ret;
2082 			}
2083 		}
2084 	}
2085 
2086 	switch (con->in_reply.tag) {
2087 	case CEPH_MSGR_TAG_FEATURES:
2088 		pr_err("%s%lld %s feature set mismatch,"
2089 		       " my %llx < server's %llx, missing %llx\n",
2090 		       ENTITY_NAME(con->peer_name),
2091 		       ceph_pr_addr(&con->peer_addr),
2092 		       sup_feat, server_feat, server_feat & ~sup_feat);
2093 		con->error_msg = "missing required protocol features";
2094 		reset_connection(con);
2095 		return -1;
2096 
2097 	case CEPH_MSGR_TAG_BADPROTOVER:
2098 		pr_err("%s%lld %s protocol version mismatch,"
2099 		       " my %d != server's %d\n",
2100 		       ENTITY_NAME(con->peer_name),
2101 		       ceph_pr_addr(&con->peer_addr),
2102 		       le32_to_cpu(con->out_connect.protocol_version),
2103 		       le32_to_cpu(con->in_reply.protocol_version));
2104 		con->error_msg = "protocol version mismatch";
2105 		reset_connection(con);
2106 		return -1;
2107 
2108 	case CEPH_MSGR_TAG_BADAUTHORIZER:
2109 		con->auth_retry++;
2110 		dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2111 		     con->auth_retry);
2112 		if (con->auth_retry == 2) {
2113 			con->error_msg = "connect authorization failure";
2114 			return -1;
2115 		}
2116 		con_out_kvec_reset(con);
2117 		ret = prepare_write_connect(con);
2118 		if (ret < 0)
2119 			return ret;
2120 		prepare_read_connect(con);
2121 		break;
2122 
2123 	case CEPH_MSGR_TAG_RESETSESSION:
2124 		/*
2125 		 * If we connected with a large connect_seq but the peer
2126 		 * has no record of a session with us (no connection, or
2127 		 * connect_seq == 0), they will send RESETSESION to indicate
2128 		 * that they must have reset their session, and may have
2129 		 * dropped messages.
2130 		 */
2131 		dout("process_connect got RESET peer seq %u\n",
2132 		     le32_to_cpu(con->in_reply.connect_seq));
2133 		pr_err("%s%lld %s connection reset\n",
2134 		       ENTITY_NAME(con->peer_name),
2135 		       ceph_pr_addr(&con->peer_addr));
2136 		reset_connection(con);
2137 		con_out_kvec_reset(con);
2138 		ret = prepare_write_connect(con);
2139 		if (ret < 0)
2140 			return ret;
2141 		prepare_read_connect(con);
2142 
2143 		/* Tell ceph about it. */
2144 		mutex_unlock(&con->mutex);
2145 		pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2146 		if (con->ops->peer_reset)
2147 			con->ops->peer_reset(con);
2148 		mutex_lock(&con->mutex);
2149 		if (con->state != CON_STATE_NEGOTIATING)
2150 			return -EAGAIN;
2151 		break;
2152 
2153 	case CEPH_MSGR_TAG_RETRY_SESSION:
2154 		/*
2155 		 * If we sent a smaller connect_seq than the peer has, try
2156 		 * again with a larger value.
2157 		 */
2158 		dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2159 		     le32_to_cpu(con->out_connect.connect_seq),
2160 		     le32_to_cpu(con->in_reply.connect_seq));
2161 		con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2162 		con_out_kvec_reset(con);
2163 		ret = prepare_write_connect(con);
2164 		if (ret < 0)
2165 			return ret;
2166 		prepare_read_connect(con);
2167 		break;
2168 
2169 	case CEPH_MSGR_TAG_RETRY_GLOBAL:
2170 		/*
2171 		 * If we sent a smaller global_seq than the peer has, try
2172 		 * again with a larger value.
2173 		 */
2174 		dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2175 		     con->peer_global_seq,
2176 		     le32_to_cpu(con->in_reply.global_seq));
2177 		get_global_seq(con->msgr,
2178 			       le32_to_cpu(con->in_reply.global_seq));
2179 		con_out_kvec_reset(con);
2180 		ret = prepare_write_connect(con);
2181 		if (ret < 0)
2182 			return ret;
2183 		prepare_read_connect(con);
2184 		break;
2185 
2186 	case CEPH_MSGR_TAG_SEQ:
2187 	case CEPH_MSGR_TAG_READY:
2188 		if (req_feat & ~server_feat) {
2189 			pr_err("%s%lld %s protocol feature mismatch,"
2190 			       " my required %llx > server's %llx, need %llx\n",
2191 			       ENTITY_NAME(con->peer_name),
2192 			       ceph_pr_addr(&con->peer_addr),
2193 			       req_feat, server_feat, req_feat & ~server_feat);
2194 			con->error_msg = "missing required protocol features";
2195 			reset_connection(con);
2196 			return -1;
2197 		}
2198 
2199 		WARN_ON(con->state != CON_STATE_NEGOTIATING);
2200 		con->state = CON_STATE_OPEN;
2201 		con->auth_retry = 0;    /* we authenticated; clear flag */
2202 		con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2203 		con->connect_seq++;
2204 		con->peer_features = server_feat;
2205 		dout("process_connect got READY gseq %d cseq %d (%d)\n",
2206 		     con->peer_global_seq,
2207 		     le32_to_cpu(con->in_reply.connect_seq),
2208 		     con->connect_seq);
2209 		WARN_ON(con->connect_seq !=
2210 			le32_to_cpu(con->in_reply.connect_seq));
2211 
2212 		if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2213 			con_flag_set(con, CON_FLAG_LOSSYTX);
2214 
2215 		con->delay = 0;      /* reset backoff memory */
2216 
2217 		if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2218 			prepare_write_seq(con);
2219 			prepare_read_seq(con);
2220 		} else {
2221 			prepare_read_tag(con);
2222 		}
2223 		break;
2224 
2225 	case CEPH_MSGR_TAG_WAIT:
2226 		/*
2227 		 * If there is a connection race (we are opening
2228 		 * connections to each other), one of us may just have
2229 		 * to WAIT.  This shouldn't happen if we are the
2230 		 * client.
2231 		 */
2232 		con->error_msg = "protocol error, got WAIT as client";
2233 		return -1;
2234 
2235 	default:
2236 		con->error_msg = "protocol error, garbage tag during connect";
2237 		return -1;
2238 	}
2239 	return 0;
2240 }
2241 
2242 
2243 /*
2244  * read (part of) an ack
2245  */
2246 static int read_partial_ack(struct ceph_connection *con)
2247 {
2248 	int size = sizeof (con->in_temp_ack);
2249 	int end = size;
2250 
2251 	return read_partial(con, end, size, &con->in_temp_ack);
2252 }
2253 
2254 /*
2255  * We can finally discard anything that's been acked.
2256  */
2257 static void process_ack(struct ceph_connection *con)
2258 {
2259 	struct ceph_msg *m;
2260 	u64 ack = le64_to_cpu(con->in_temp_ack);
2261 	u64 seq;
2262 	bool reconnect = (con->in_tag == CEPH_MSGR_TAG_SEQ);
2263 	struct list_head *list = reconnect ? &con->out_queue : &con->out_sent;
2264 
2265 	/*
2266 	 * In the reconnect case, con_fault() has requeued messages
2267 	 * in out_sent. We should cleanup old messages according to
2268 	 * the reconnect seq.
2269 	 */
2270 	while (!list_empty(list)) {
2271 		m = list_first_entry(list, struct ceph_msg, list_head);
2272 		if (reconnect && m->needs_out_seq)
2273 			break;
2274 		seq = le64_to_cpu(m->hdr.seq);
2275 		if (seq > ack)
2276 			break;
2277 		dout("got ack for seq %llu type %d at %p\n", seq,
2278 		     le16_to_cpu(m->hdr.type), m);
2279 		m->ack_stamp = jiffies;
2280 		ceph_msg_remove(m);
2281 	}
2282 
2283 	prepare_read_tag(con);
2284 }
2285 
2286 
2287 static int read_partial_message_section(struct ceph_connection *con,
2288 					struct kvec *section,
2289 					unsigned int sec_len, u32 *crc)
2290 {
2291 	int ret, left;
2292 
2293 	BUG_ON(!section);
2294 
2295 	while (section->iov_len < sec_len) {
2296 		BUG_ON(section->iov_base == NULL);
2297 		left = sec_len - section->iov_len;
2298 		ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2299 				       section->iov_len, left);
2300 		if (ret <= 0)
2301 			return ret;
2302 		section->iov_len += ret;
2303 	}
2304 	if (section->iov_len == sec_len)
2305 		*crc = crc32c(0, section->iov_base, section->iov_len);
2306 
2307 	return 1;
2308 }
2309 
2310 static int read_partial_msg_data(struct ceph_connection *con)
2311 {
2312 	struct ceph_msg *msg = con->in_msg;
2313 	struct ceph_msg_data_cursor *cursor = &msg->cursor;
2314 	bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2315 	struct page *page;
2316 	size_t page_offset;
2317 	size_t length;
2318 	u32 crc = 0;
2319 	int ret;
2320 
2321 	if (!msg->num_data_items)
2322 		return -EIO;
2323 
2324 	if (do_datacrc)
2325 		crc = con->in_data_crc;
2326 	while (cursor->total_resid) {
2327 		if (!cursor->resid) {
2328 			ceph_msg_data_advance(cursor, 0);
2329 			continue;
2330 		}
2331 
2332 		page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
2333 		ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2334 		if (ret <= 0) {
2335 			if (do_datacrc)
2336 				con->in_data_crc = crc;
2337 
2338 			return ret;
2339 		}
2340 
2341 		if (do_datacrc)
2342 			crc = ceph_crc32c_page(crc, page, page_offset, ret);
2343 		ceph_msg_data_advance(cursor, (size_t)ret);
2344 	}
2345 	if (do_datacrc)
2346 		con->in_data_crc = crc;
2347 
2348 	return 1;	/* must return > 0 to indicate success */
2349 }
2350 
2351 /*
2352  * read (part of) a message.
2353  */
2354 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2355 
2356 static int read_partial_message(struct ceph_connection *con)
2357 {
2358 	struct ceph_msg *m = con->in_msg;
2359 	int size;
2360 	int end;
2361 	int ret;
2362 	unsigned int front_len, middle_len, data_len;
2363 	bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2364 	bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
2365 	u64 seq;
2366 	u32 crc;
2367 
2368 	dout("read_partial_message con %p msg %p\n", con, m);
2369 
2370 	/* header */
2371 	size = sizeof (con->in_hdr);
2372 	end = size;
2373 	ret = read_partial(con, end, size, &con->in_hdr);
2374 	if (ret <= 0)
2375 		return ret;
2376 
2377 	crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2378 	if (cpu_to_le32(crc) != con->in_hdr.crc) {
2379 		pr_err("read_partial_message bad hdr crc %u != expected %u\n",
2380 		       crc, con->in_hdr.crc);
2381 		return -EBADMSG;
2382 	}
2383 
2384 	front_len = le32_to_cpu(con->in_hdr.front_len);
2385 	if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2386 		return -EIO;
2387 	middle_len = le32_to_cpu(con->in_hdr.middle_len);
2388 	if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2389 		return -EIO;
2390 	data_len = le32_to_cpu(con->in_hdr.data_len);
2391 	if (data_len > CEPH_MSG_MAX_DATA_LEN)
2392 		return -EIO;
2393 
2394 	/* verify seq# */
2395 	seq = le64_to_cpu(con->in_hdr.seq);
2396 	if ((s64)seq - (s64)con->in_seq < 1) {
2397 		pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2398 			ENTITY_NAME(con->peer_name),
2399 			ceph_pr_addr(&con->peer_addr),
2400 			seq, con->in_seq + 1);
2401 		con->in_base_pos = -front_len - middle_len - data_len -
2402 			sizeof_footer(con);
2403 		con->in_tag = CEPH_MSGR_TAG_READY;
2404 		return 1;
2405 	} else if ((s64)seq - (s64)con->in_seq > 1) {
2406 		pr_err("read_partial_message bad seq %lld expected %lld\n",
2407 		       seq, con->in_seq + 1);
2408 		con->error_msg = "bad message sequence # for incoming message";
2409 		return -EBADE;
2410 	}
2411 
2412 	/* allocate message? */
2413 	if (!con->in_msg) {
2414 		int skip = 0;
2415 
2416 		dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2417 		     front_len, data_len);
2418 		ret = ceph_con_in_msg_alloc(con, &skip);
2419 		if (ret < 0)
2420 			return ret;
2421 
2422 		BUG_ON(!con->in_msg ^ skip);
2423 		if (skip) {
2424 			/* skip this message */
2425 			dout("alloc_msg said skip message\n");
2426 			con->in_base_pos = -front_len - middle_len - data_len -
2427 				sizeof_footer(con);
2428 			con->in_tag = CEPH_MSGR_TAG_READY;
2429 			con->in_seq++;
2430 			return 1;
2431 		}
2432 
2433 		BUG_ON(!con->in_msg);
2434 		BUG_ON(con->in_msg->con != con);
2435 		m = con->in_msg;
2436 		m->front.iov_len = 0;    /* haven't read it yet */
2437 		if (m->middle)
2438 			m->middle->vec.iov_len = 0;
2439 
2440 		/* prepare for data payload, if any */
2441 
2442 		if (data_len)
2443 			prepare_message_data(con->in_msg, data_len);
2444 	}
2445 
2446 	/* front */
2447 	ret = read_partial_message_section(con, &m->front, front_len,
2448 					   &con->in_front_crc);
2449 	if (ret <= 0)
2450 		return ret;
2451 
2452 	/* middle */
2453 	if (m->middle) {
2454 		ret = read_partial_message_section(con, &m->middle->vec,
2455 						   middle_len,
2456 						   &con->in_middle_crc);
2457 		if (ret <= 0)
2458 			return ret;
2459 	}
2460 
2461 	/* (page) data */
2462 	if (data_len) {
2463 		ret = read_partial_msg_data(con);
2464 		if (ret <= 0)
2465 			return ret;
2466 	}
2467 
2468 	/* footer */
2469 	size = sizeof_footer(con);
2470 	end += size;
2471 	ret = read_partial(con, end, size, &m->footer);
2472 	if (ret <= 0)
2473 		return ret;
2474 
2475 	if (!need_sign) {
2476 		m->footer.flags = m->old_footer.flags;
2477 		m->footer.sig = 0;
2478 	}
2479 
2480 	dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2481 	     m, front_len, m->footer.front_crc, middle_len,
2482 	     m->footer.middle_crc, data_len, m->footer.data_crc);
2483 
2484 	/* crc ok? */
2485 	if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2486 		pr_err("read_partial_message %p front crc %u != exp. %u\n",
2487 		       m, con->in_front_crc, m->footer.front_crc);
2488 		return -EBADMSG;
2489 	}
2490 	if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2491 		pr_err("read_partial_message %p middle crc %u != exp %u\n",
2492 		       m, con->in_middle_crc, m->footer.middle_crc);
2493 		return -EBADMSG;
2494 	}
2495 	if (do_datacrc &&
2496 	    (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2497 	    con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2498 		pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2499 		       con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2500 		return -EBADMSG;
2501 	}
2502 
2503 	if (need_sign && con->ops->check_message_signature &&
2504 	    con->ops->check_message_signature(m)) {
2505 		pr_err("read_partial_message %p signature check failed\n", m);
2506 		return -EBADMSG;
2507 	}
2508 
2509 	return 1; /* done! */
2510 }
2511 
2512 /*
2513  * Process message.  This happens in the worker thread.  The callback should
2514  * be careful not to do anything that waits on other incoming messages or it
2515  * may deadlock.
2516  */
2517 static void process_message(struct ceph_connection *con)
2518 {
2519 	struct ceph_msg *msg = con->in_msg;
2520 
2521 	BUG_ON(con->in_msg->con != con);
2522 	con->in_msg = NULL;
2523 
2524 	/* if first message, set peer_name */
2525 	if (con->peer_name.type == 0)
2526 		con->peer_name = msg->hdr.src;
2527 
2528 	con->in_seq++;
2529 	mutex_unlock(&con->mutex);
2530 
2531 	dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2532 	     msg, le64_to_cpu(msg->hdr.seq),
2533 	     ENTITY_NAME(msg->hdr.src),
2534 	     le16_to_cpu(msg->hdr.type),
2535 	     ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2536 	     le32_to_cpu(msg->hdr.front_len),
2537 	     le32_to_cpu(msg->hdr.data_len),
2538 	     con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2539 	con->ops->dispatch(con, msg);
2540 
2541 	mutex_lock(&con->mutex);
2542 }
2543 
2544 static int read_keepalive_ack(struct ceph_connection *con)
2545 {
2546 	struct ceph_timespec ceph_ts;
2547 	size_t size = sizeof(ceph_ts);
2548 	int ret = read_partial(con, size, size, &ceph_ts);
2549 	if (ret <= 0)
2550 		return ret;
2551 	ceph_decode_timespec64(&con->last_keepalive_ack, &ceph_ts);
2552 	prepare_read_tag(con);
2553 	return 1;
2554 }
2555 
2556 /*
2557  * Write something to the socket.  Called in a worker thread when the
2558  * socket appears to be writeable and we have something ready to send.
2559  */
2560 static int try_write(struct ceph_connection *con)
2561 {
2562 	int ret = 1;
2563 
2564 	dout("try_write start %p state %lu\n", con, con->state);
2565 	if (con->state != CON_STATE_PREOPEN &&
2566 	    con->state != CON_STATE_CONNECTING &&
2567 	    con->state != CON_STATE_NEGOTIATING &&
2568 	    con->state != CON_STATE_OPEN)
2569 		return 0;
2570 
2571 	/* open the socket first? */
2572 	if (con->state == CON_STATE_PREOPEN) {
2573 		BUG_ON(con->sock);
2574 		con->state = CON_STATE_CONNECTING;
2575 
2576 		con_out_kvec_reset(con);
2577 		prepare_write_banner(con);
2578 		prepare_read_banner(con);
2579 
2580 		BUG_ON(con->in_msg);
2581 		con->in_tag = CEPH_MSGR_TAG_READY;
2582 		dout("try_write initiating connect on %p new state %lu\n",
2583 		     con, con->state);
2584 		ret = ceph_tcp_connect(con);
2585 		if (ret < 0) {
2586 			con->error_msg = "connect error";
2587 			goto out;
2588 		}
2589 	}
2590 
2591 more:
2592 	dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2593 	BUG_ON(!con->sock);
2594 
2595 	/* kvec data queued? */
2596 	if (con->out_kvec_left) {
2597 		ret = write_partial_kvec(con);
2598 		if (ret <= 0)
2599 			goto out;
2600 	}
2601 	if (con->out_skip) {
2602 		ret = write_partial_skip(con);
2603 		if (ret <= 0)
2604 			goto out;
2605 	}
2606 
2607 	/* msg pages? */
2608 	if (con->out_msg) {
2609 		if (con->out_msg_done) {
2610 			ceph_msg_put(con->out_msg);
2611 			con->out_msg = NULL;   /* we're done with this one */
2612 			goto do_next;
2613 		}
2614 
2615 		ret = write_partial_message_data(con);
2616 		if (ret == 1)
2617 			goto more;  /* we need to send the footer, too! */
2618 		if (ret == 0)
2619 			goto out;
2620 		if (ret < 0) {
2621 			dout("try_write write_partial_message_data err %d\n",
2622 			     ret);
2623 			goto out;
2624 		}
2625 	}
2626 
2627 do_next:
2628 	if (con->state == CON_STATE_OPEN) {
2629 		if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2630 			prepare_write_keepalive(con);
2631 			goto more;
2632 		}
2633 		/* is anything else pending? */
2634 		if (!list_empty(&con->out_queue)) {
2635 			prepare_write_message(con);
2636 			goto more;
2637 		}
2638 		if (con->in_seq > con->in_seq_acked) {
2639 			prepare_write_ack(con);
2640 			goto more;
2641 		}
2642 	}
2643 
2644 	/* Nothing to do! */
2645 	con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2646 	dout("try_write nothing else to write.\n");
2647 	ret = 0;
2648 out:
2649 	dout("try_write done on %p ret %d\n", con, ret);
2650 	return ret;
2651 }
2652 
2653 /*
2654  * Read what we can from the socket.
2655  */
2656 static int try_read(struct ceph_connection *con)
2657 {
2658 	int ret = -1;
2659 
2660 more:
2661 	dout("try_read start on %p state %lu\n", con, con->state);
2662 	if (con->state != CON_STATE_CONNECTING &&
2663 	    con->state != CON_STATE_NEGOTIATING &&
2664 	    con->state != CON_STATE_OPEN)
2665 		return 0;
2666 
2667 	BUG_ON(!con->sock);
2668 
2669 	dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2670 	     con->in_base_pos);
2671 
2672 	if (con->state == CON_STATE_CONNECTING) {
2673 		dout("try_read connecting\n");
2674 		ret = read_partial_banner(con);
2675 		if (ret <= 0)
2676 			goto out;
2677 		ret = process_banner(con);
2678 		if (ret < 0)
2679 			goto out;
2680 
2681 		con->state = CON_STATE_NEGOTIATING;
2682 
2683 		/*
2684 		 * Received banner is good, exchange connection info.
2685 		 * Do not reset out_kvec, as sending our banner raced
2686 		 * with receiving peer banner after connect completed.
2687 		 */
2688 		ret = prepare_write_connect(con);
2689 		if (ret < 0)
2690 			goto out;
2691 		prepare_read_connect(con);
2692 
2693 		/* Send connection info before awaiting response */
2694 		goto out;
2695 	}
2696 
2697 	if (con->state == CON_STATE_NEGOTIATING) {
2698 		dout("try_read negotiating\n");
2699 		ret = read_partial_connect(con);
2700 		if (ret <= 0)
2701 			goto out;
2702 		ret = process_connect(con);
2703 		if (ret < 0)
2704 			goto out;
2705 		goto more;
2706 	}
2707 
2708 	WARN_ON(con->state != CON_STATE_OPEN);
2709 
2710 	if (con->in_base_pos < 0) {
2711 		/*
2712 		 * skipping + discarding content.
2713 		 */
2714 		ret = ceph_tcp_recvmsg(con->sock, NULL, -con->in_base_pos);
2715 		if (ret <= 0)
2716 			goto out;
2717 		dout("skipped %d / %d bytes\n", ret, -con->in_base_pos);
2718 		con->in_base_pos += ret;
2719 		if (con->in_base_pos)
2720 			goto more;
2721 	}
2722 	if (con->in_tag == CEPH_MSGR_TAG_READY) {
2723 		/*
2724 		 * what's next?
2725 		 */
2726 		ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2727 		if (ret <= 0)
2728 			goto out;
2729 		dout("try_read got tag %d\n", (int)con->in_tag);
2730 		switch (con->in_tag) {
2731 		case CEPH_MSGR_TAG_MSG:
2732 			prepare_read_message(con);
2733 			break;
2734 		case CEPH_MSGR_TAG_ACK:
2735 			prepare_read_ack(con);
2736 			break;
2737 		case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
2738 			prepare_read_keepalive_ack(con);
2739 			break;
2740 		case CEPH_MSGR_TAG_CLOSE:
2741 			con_close_socket(con);
2742 			con->state = CON_STATE_CLOSED;
2743 			goto out;
2744 		default:
2745 			goto bad_tag;
2746 		}
2747 	}
2748 	if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2749 		ret = read_partial_message(con);
2750 		if (ret <= 0) {
2751 			switch (ret) {
2752 			case -EBADMSG:
2753 				con->error_msg = "bad crc/signature";
2754 				/* fall through */
2755 			case -EBADE:
2756 				ret = -EIO;
2757 				break;
2758 			case -EIO:
2759 				con->error_msg = "io error";
2760 				break;
2761 			}
2762 			goto out;
2763 		}
2764 		if (con->in_tag == CEPH_MSGR_TAG_READY)
2765 			goto more;
2766 		process_message(con);
2767 		if (con->state == CON_STATE_OPEN)
2768 			prepare_read_tag(con);
2769 		goto more;
2770 	}
2771 	if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2772 	    con->in_tag == CEPH_MSGR_TAG_SEQ) {
2773 		/*
2774 		 * the final handshake seq exchange is semantically
2775 		 * equivalent to an ACK
2776 		 */
2777 		ret = read_partial_ack(con);
2778 		if (ret <= 0)
2779 			goto out;
2780 		process_ack(con);
2781 		goto more;
2782 	}
2783 	if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
2784 		ret = read_keepalive_ack(con);
2785 		if (ret <= 0)
2786 			goto out;
2787 		goto more;
2788 	}
2789 
2790 out:
2791 	dout("try_read done on %p ret %d\n", con, ret);
2792 	return ret;
2793 
2794 bad_tag:
2795 	pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2796 	con->error_msg = "protocol error, garbage tag";
2797 	ret = -1;
2798 	goto out;
2799 }
2800 
2801 
2802 /*
2803  * Atomically queue work on a connection after the specified delay.
2804  * Bump @con reference to avoid races with connection teardown.
2805  * Returns 0 if work was queued, or an error code otherwise.
2806  */
2807 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2808 {
2809 	if (!con->ops->get(con)) {
2810 		dout("%s %p ref count 0\n", __func__, con);
2811 		return -ENOENT;
2812 	}
2813 
2814 	if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2815 		dout("%s %p - already queued\n", __func__, con);
2816 		con->ops->put(con);
2817 		return -EBUSY;
2818 	}
2819 
2820 	dout("%s %p %lu\n", __func__, con, delay);
2821 	return 0;
2822 }
2823 
2824 static void queue_con(struct ceph_connection *con)
2825 {
2826 	(void) queue_con_delay(con, 0);
2827 }
2828 
2829 static void cancel_con(struct ceph_connection *con)
2830 {
2831 	if (cancel_delayed_work(&con->work)) {
2832 		dout("%s %p\n", __func__, con);
2833 		con->ops->put(con);
2834 	}
2835 }
2836 
2837 static bool con_sock_closed(struct ceph_connection *con)
2838 {
2839 	if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2840 		return false;
2841 
2842 #define CASE(x)								\
2843 	case CON_STATE_ ## x:						\
2844 		con->error_msg = "socket closed (con state " #x ")";	\
2845 		break;
2846 
2847 	switch (con->state) {
2848 	CASE(CLOSED);
2849 	CASE(PREOPEN);
2850 	CASE(CONNECTING);
2851 	CASE(NEGOTIATING);
2852 	CASE(OPEN);
2853 	CASE(STANDBY);
2854 	default:
2855 		pr_warn("%s con %p unrecognized state %lu\n",
2856 			__func__, con, con->state);
2857 		con->error_msg = "unrecognized con state";
2858 		BUG();
2859 		break;
2860 	}
2861 #undef CASE
2862 
2863 	return true;
2864 }
2865 
2866 static bool con_backoff(struct ceph_connection *con)
2867 {
2868 	int ret;
2869 
2870 	if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2871 		return false;
2872 
2873 	ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2874 	if (ret) {
2875 		dout("%s: con %p FAILED to back off %lu\n", __func__,
2876 			con, con->delay);
2877 		BUG_ON(ret == -ENOENT);
2878 		con_flag_set(con, CON_FLAG_BACKOFF);
2879 	}
2880 
2881 	return true;
2882 }
2883 
2884 /* Finish fault handling; con->mutex must *not* be held here */
2885 
2886 static void con_fault_finish(struct ceph_connection *con)
2887 {
2888 	dout("%s %p\n", __func__, con);
2889 
2890 	/*
2891 	 * in case we faulted due to authentication, invalidate our
2892 	 * current tickets so that we can get new ones.
2893 	 */
2894 	if (con->auth_retry) {
2895 		dout("auth_retry %d, invalidating\n", con->auth_retry);
2896 		if (con->ops->invalidate_authorizer)
2897 			con->ops->invalidate_authorizer(con);
2898 		con->auth_retry = 0;
2899 	}
2900 
2901 	if (con->ops->fault)
2902 		con->ops->fault(con);
2903 }
2904 
2905 /*
2906  * Do some work on a connection.  Drop a connection ref when we're done.
2907  */
2908 static void ceph_con_workfn(struct work_struct *work)
2909 {
2910 	struct ceph_connection *con = container_of(work, struct ceph_connection,
2911 						   work.work);
2912 	bool fault;
2913 
2914 	mutex_lock(&con->mutex);
2915 	while (true) {
2916 		int ret;
2917 
2918 		if ((fault = con_sock_closed(con))) {
2919 			dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2920 			break;
2921 		}
2922 		if (con_backoff(con)) {
2923 			dout("%s: con %p BACKOFF\n", __func__, con);
2924 			break;
2925 		}
2926 		if (con->state == CON_STATE_STANDBY) {
2927 			dout("%s: con %p STANDBY\n", __func__, con);
2928 			break;
2929 		}
2930 		if (con->state == CON_STATE_CLOSED) {
2931 			dout("%s: con %p CLOSED\n", __func__, con);
2932 			BUG_ON(con->sock);
2933 			break;
2934 		}
2935 		if (con->state == CON_STATE_PREOPEN) {
2936 			dout("%s: con %p PREOPEN\n", __func__, con);
2937 			BUG_ON(con->sock);
2938 		}
2939 
2940 		ret = try_read(con);
2941 		if (ret < 0) {
2942 			if (ret == -EAGAIN)
2943 				continue;
2944 			if (!con->error_msg)
2945 				con->error_msg = "socket error on read";
2946 			fault = true;
2947 			break;
2948 		}
2949 
2950 		ret = try_write(con);
2951 		if (ret < 0) {
2952 			if (ret == -EAGAIN)
2953 				continue;
2954 			if (!con->error_msg)
2955 				con->error_msg = "socket error on write";
2956 			fault = true;
2957 		}
2958 
2959 		break;	/* If we make it to here, we're done */
2960 	}
2961 	if (fault)
2962 		con_fault(con);
2963 	mutex_unlock(&con->mutex);
2964 
2965 	if (fault)
2966 		con_fault_finish(con);
2967 
2968 	con->ops->put(con);
2969 }
2970 
2971 /*
2972  * Generic error/fault handler.  A retry mechanism is used with
2973  * exponential backoff
2974  */
2975 static void con_fault(struct ceph_connection *con)
2976 {
2977 	dout("fault %p state %lu to peer %s\n",
2978 	     con, con->state, ceph_pr_addr(&con->peer_addr));
2979 
2980 	pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2981 		ceph_pr_addr(&con->peer_addr), con->error_msg);
2982 	con->error_msg = NULL;
2983 
2984 	WARN_ON(con->state != CON_STATE_CONNECTING &&
2985 	       con->state != CON_STATE_NEGOTIATING &&
2986 	       con->state != CON_STATE_OPEN);
2987 
2988 	con_close_socket(con);
2989 
2990 	if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2991 		dout("fault on LOSSYTX channel, marking CLOSED\n");
2992 		con->state = CON_STATE_CLOSED;
2993 		return;
2994 	}
2995 
2996 	if (con->in_msg) {
2997 		BUG_ON(con->in_msg->con != con);
2998 		ceph_msg_put(con->in_msg);
2999 		con->in_msg = NULL;
3000 	}
3001 
3002 	/* Requeue anything that hasn't been acked */
3003 	list_splice_init(&con->out_sent, &con->out_queue);
3004 
3005 	/* If there are no messages queued or keepalive pending, place
3006 	 * the connection in a STANDBY state */
3007 	if (list_empty(&con->out_queue) &&
3008 	    !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
3009 		dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
3010 		con_flag_clear(con, CON_FLAG_WRITE_PENDING);
3011 		con->state = CON_STATE_STANDBY;
3012 	} else {
3013 		/* retry after a delay. */
3014 		con->state = CON_STATE_PREOPEN;
3015 		if (con->delay == 0)
3016 			con->delay = BASE_DELAY_INTERVAL;
3017 		else if (con->delay < MAX_DELAY_INTERVAL)
3018 			con->delay *= 2;
3019 		con_flag_set(con, CON_FLAG_BACKOFF);
3020 		queue_con(con);
3021 	}
3022 }
3023 
3024 
3025 void ceph_messenger_reset_nonce(struct ceph_messenger *msgr)
3026 {
3027 	u32 nonce = le32_to_cpu(msgr->inst.addr.nonce) + 1000000;
3028 	msgr->inst.addr.nonce = cpu_to_le32(nonce);
3029 	encode_my_addr(msgr);
3030 }
3031 
3032 /*
3033  * initialize a new messenger instance
3034  */
3035 void ceph_messenger_init(struct ceph_messenger *msgr,
3036 			 struct ceph_entity_addr *myaddr)
3037 {
3038 	spin_lock_init(&msgr->global_seq_lock);
3039 
3040 	if (myaddr)
3041 		msgr->inst.addr = *myaddr;
3042 
3043 	/* select a random nonce */
3044 	msgr->inst.addr.type = 0;
3045 	get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
3046 	encode_my_addr(msgr);
3047 
3048 	atomic_set(&msgr->stopping, 0);
3049 	write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
3050 
3051 	dout("%s %p\n", __func__, msgr);
3052 }
3053 EXPORT_SYMBOL(ceph_messenger_init);
3054 
3055 void ceph_messenger_fini(struct ceph_messenger *msgr)
3056 {
3057 	put_net(read_pnet(&msgr->net));
3058 }
3059 EXPORT_SYMBOL(ceph_messenger_fini);
3060 
3061 static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con)
3062 {
3063 	if (msg->con)
3064 		msg->con->ops->put(msg->con);
3065 
3066 	msg->con = con ? con->ops->get(con) : NULL;
3067 	BUG_ON(msg->con != con);
3068 }
3069 
3070 static void clear_standby(struct ceph_connection *con)
3071 {
3072 	/* come back from STANDBY? */
3073 	if (con->state == CON_STATE_STANDBY) {
3074 		dout("clear_standby %p and ++connect_seq\n", con);
3075 		con->state = CON_STATE_PREOPEN;
3076 		con->connect_seq++;
3077 		WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
3078 		WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
3079 	}
3080 }
3081 
3082 /*
3083  * Queue up an outgoing message on the given connection.
3084  */
3085 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
3086 {
3087 	/* set src+dst */
3088 	msg->hdr.src = con->msgr->inst.name;
3089 	BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
3090 	msg->needs_out_seq = true;
3091 
3092 	mutex_lock(&con->mutex);
3093 
3094 	if (con->state == CON_STATE_CLOSED) {
3095 		dout("con_send %p closed, dropping %p\n", con, msg);
3096 		ceph_msg_put(msg);
3097 		mutex_unlock(&con->mutex);
3098 		return;
3099 	}
3100 
3101 	msg_con_set(msg, con);
3102 
3103 	BUG_ON(!list_empty(&msg->list_head));
3104 	list_add_tail(&msg->list_head, &con->out_queue);
3105 	dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3106 	     ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3107 	     ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3108 	     le32_to_cpu(msg->hdr.front_len),
3109 	     le32_to_cpu(msg->hdr.middle_len),
3110 	     le32_to_cpu(msg->hdr.data_len));
3111 
3112 	clear_standby(con);
3113 	mutex_unlock(&con->mutex);
3114 
3115 	/* if there wasn't anything waiting to send before, queue
3116 	 * new work */
3117 	if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3118 		queue_con(con);
3119 }
3120 EXPORT_SYMBOL(ceph_con_send);
3121 
3122 /*
3123  * Revoke a message that was previously queued for send
3124  */
3125 void ceph_msg_revoke(struct ceph_msg *msg)
3126 {
3127 	struct ceph_connection *con = msg->con;
3128 
3129 	if (!con) {
3130 		dout("%s msg %p null con\n", __func__, msg);
3131 		return;		/* Message not in our possession */
3132 	}
3133 
3134 	mutex_lock(&con->mutex);
3135 	if (!list_empty(&msg->list_head)) {
3136 		dout("%s %p msg %p - was on queue\n", __func__, con, msg);
3137 		list_del_init(&msg->list_head);
3138 		msg->hdr.seq = 0;
3139 
3140 		ceph_msg_put(msg);
3141 	}
3142 	if (con->out_msg == msg) {
3143 		BUG_ON(con->out_skip);
3144 		/* footer */
3145 		if (con->out_msg_done) {
3146 			con->out_skip += con_out_kvec_skip(con);
3147 		} else {
3148 			BUG_ON(!msg->data_length);
3149 			con->out_skip += sizeof_footer(con);
3150 		}
3151 		/* data, middle, front */
3152 		if (msg->data_length)
3153 			con->out_skip += msg->cursor.total_resid;
3154 		if (msg->middle)
3155 			con->out_skip += con_out_kvec_skip(con);
3156 		con->out_skip += con_out_kvec_skip(con);
3157 
3158 		dout("%s %p msg %p - was sending, will write %d skip %d\n",
3159 		     __func__, con, msg, con->out_kvec_bytes, con->out_skip);
3160 		msg->hdr.seq = 0;
3161 		con->out_msg = NULL;
3162 		ceph_msg_put(msg);
3163 	}
3164 
3165 	mutex_unlock(&con->mutex);
3166 }
3167 
3168 /*
3169  * Revoke a message that we may be reading data into
3170  */
3171 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
3172 {
3173 	struct ceph_connection *con = msg->con;
3174 
3175 	if (!con) {
3176 		dout("%s msg %p null con\n", __func__, msg);
3177 		return;		/* Message not in our possession */
3178 	}
3179 
3180 	mutex_lock(&con->mutex);
3181 	if (con->in_msg == msg) {
3182 		unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3183 		unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3184 		unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
3185 
3186 		/* skip rest of message */
3187 		dout("%s %p msg %p revoked\n", __func__, con, msg);
3188 		con->in_base_pos = con->in_base_pos -
3189 				sizeof(struct ceph_msg_header) -
3190 				front_len -
3191 				middle_len -
3192 				data_len -
3193 				sizeof(struct ceph_msg_footer);
3194 		ceph_msg_put(con->in_msg);
3195 		con->in_msg = NULL;
3196 		con->in_tag = CEPH_MSGR_TAG_READY;
3197 		con->in_seq++;
3198 	} else {
3199 		dout("%s %p in_msg %p msg %p no-op\n",
3200 		     __func__, con, con->in_msg, msg);
3201 	}
3202 	mutex_unlock(&con->mutex);
3203 }
3204 
3205 /*
3206  * Queue a keepalive byte to ensure the tcp connection is alive.
3207  */
3208 void ceph_con_keepalive(struct ceph_connection *con)
3209 {
3210 	dout("con_keepalive %p\n", con);
3211 	mutex_lock(&con->mutex);
3212 	clear_standby(con);
3213 	con_flag_set(con, CON_FLAG_KEEPALIVE_PENDING);
3214 	mutex_unlock(&con->mutex);
3215 
3216 	if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3217 		queue_con(con);
3218 }
3219 EXPORT_SYMBOL(ceph_con_keepalive);
3220 
3221 bool ceph_con_keepalive_expired(struct ceph_connection *con,
3222 			       unsigned long interval)
3223 {
3224 	if (interval > 0 &&
3225 	    (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
3226 		struct timespec64 now;
3227 		struct timespec64 ts;
3228 		ktime_get_real_ts64(&now);
3229 		jiffies_to_timespec64(interval, &ts);
3230 		ts = timespec64_add(con->last_keepalive_ack, ts);
3231 		return timespec64_compare(&now, &ts) >= 0;
3232 	}
3233 	return false;
3234 }
3235 
3236 static struct ceph_msg_data *ceph_msg_data_add(struct ceph_msg *msg)
3237 {
3238 	BUG_ON(msg->num_data_items >= msg->max_data_items);
3239 	return &msg->data[msg->num_data_items++];
3240 }
3241 
3242 static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3243 {
3244 	if (data->type == CEPH_MSG_DATA_PAGES && data->own_pages) {
3245 		int num_pages = calc_pages_for(data->alignment, data->length);
3246 		ceph_release_page_vector(data->pages, num_pages);
3247 	} else if (data->type == CEPH_MSG_DATA_PAGELIST) {
3248 		ceph_pagelist_release(data->pagelist);
3249 	}
3250 }
3251 
3252 void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
3253 			     size_t length, size_t alignment, bool own_pages)
3254 {
3255 	struct ceph_msg_data *data;
3256 
3257 	BUG_ON(!pages);
3258 	BUG_ON(!length);
3259 
3260 	data = ceph_msg_data_add(msg);
3261 	data->type = CEPH_MSG_DATA_PAGES;
3262 	data->pages = pages;
3263 	data->length = length;
3264 	data->alignment = alignment & ~PAGE_MASK;
3265 	data->own_pages = own_pages;
3266 
3267 	msg->data_length += length;
3268 }
3269 EXPORT_SYMBOL(ceph_msg_data_add_pages);
3270 
3271 void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
3272 				struct ceph_pagelist *pagelist)
3273 {
3274 	struct ceph_msg_data *data;
3275 
3276 	BUG_ON(!pagelist);
3277 	BUG_ON(!pagelist->length);
3278 
3279 	data = ceph_msg_data_add(msg);
3280 	data->type = CEPH_MSG_DATA_PAGELIST;
3281 	refcount_inc(&pagelist->refcnt);
3282 	data->pagelist = pagelist;
3283 
3284 	msg->data_length += pagelist->length;
3285 }
3286 EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
3287 
3288 #ifdef	CONFIG_BLOCK
3289 void ceph_msg_data_add_bio(struct ceph_msg *msg, struct ceph_bio_iter *bio_pos,
3290 			   u32 length)
3291 {
3292 	struct ceph_msg_data *data;
3293 
3294 	data = ceph_msg_data_add(msg);
3295 	data->type = CEPH_MSG_DATA_BIO;
3296 	data->bio_pos = *bio_pos;
3297 	data->bio_length = length;
3298 
3299 	msg->data_length += length;
3300 }
3301 EXPORT_SYMBOL(ceph_msg_data_add_bio);
3302 #endif	/* CONFIG_BLOCK */
3303 
3304 void ceph_msg_data_add_bvecs(struct ceph_msg *msg,
3305 			     struct ceph_bvec_iter *bvec_pos)
3306 {
3307 	struct ceph_msg_data *data;
3308 
3309 	data = ceph_msg_data_add(msg);
3310 	data->type = CEPH_MSG_DATA_BVECS;
3311 	data->bvec_pos = *bvec_pos;
3312 
3313 	msg->data_length += bvec_pos->iter.bi_size;
3314 }
3315 EXPORT_SYMBOL(ceph_msg_data_add_bvecs);
3316 
3317 /*
3318  * construct a new message with given type, size
3319  * the new msg has a ref count of 1.
3320  */
3321 struct ceph_msg *ceph_msg_new2(int type, int front_len, int max_data_items,
3322 			       gfp_t flags, bool can_fail)
3323 {
3324 	struct ceph_msg *m;
3325 
3326 	m = kmem_cache_zalloc(ceph_msg_cache, flags);
3327 	if (m == NULL)
3328 		goto out;
3329 
3330 	m->hdr.type = cpu_to_le16(type);
3331 	m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3332 	m->hdr.front_len = cpu_to_le32(front_len);
3333 
3334 	INIT_LIST_HEAD(&m->list_head);
3335 	kref_init(&m->kref);
3336 
3337 	/* front */
3338 	if (front_len) {
3339 		m->front.iov_base = ceph_kvmalloc(front_len, flags);
3340 		if (m->front.iov_base == NULL) {
3341 			dout("ceph_msg_new can't allocate %d bytes\n",
3342 			     front_len);
3343 			goto out2;
3344 		}
3345 	} else {
3346 		m->front.iov_base = NULL;
3347 	}
3348 	m->front_alloc_len = m->front.iov_len = front_len;
3349 
3350 	if (max_data_items) {
3351 		m->data = kmalloc_array(max_data_items, sizeof(*m->data),
3352 					flags);
3353 		if (!m->data)
3354 			goto out2;
3355 
3356 		m->max_data_items = max_data_items;
3357 	}
3358 
3359 	dout("ceph_msg_new %p front %d\n", m, front_len);
3360 	return m;
3361 
3362 out2:
3363 	ceph_msg_put(m);
3364 out:
3365 	if (!can_fail) {
3366 		pr_err("msg_new can't create type %d front %d\n", type,
3367 		       front_len);
3368 		WARN_ON(1);
3369 	} else {
3370 		dout("msg_new can't create type %d front %d\n", type,
3371 		     front_len);
3372 	}
3373 	return NULL;
3374 }
3375 EXPORT_SYMBOL(ceph_msg_new2);
3376 
3377 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3378 			      bool can_fail)
3379 {
3380 	return ceph_msg_new2(type, front_len, 0, flags, can_fail);
3381 }
3382 EXPORT_SYMBOL(ceph_msg_new);
3383 
3384 /*
3385  * Allocate "middle" portion of a message, if it is needed and wasn't
3386  * allocated by alloc_msg.  This allows us to read a small fixed-size
3387  * per-type header in the front and then gracefully fail (i.e.,
3388  * propagate the error to the caller based on info in the front) when
3389  * the middle is too large.
3390  */
3391 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3392 {
3393 	int type = le16_to_cpu(msg->hdr.type);
3394 	int middle_len = le32_to_cpu(msg->hdr.middle_len);
3395 
3396 	dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3397 	     ceph_msg_type_name(type), middle_len);
3398 	BUG_ON(!middle_len);
3399 	BUG_ON(msg->middle);
3400 
3401 	msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3402 	if (!msg->middle)
3403 		return -ENOMEM;
3404 	return 0;
3405 }
3406 
3407 /*
3408  * Allocate a message for receiving an incoming message on a
3409  * connection, and save the result in con->in_msg.  Uses the
3410  * connection's private alloc_msg op if available.
3411  *
3412  * Returns 0 on success, or a negative error code.
3413  *
3414  * On success, if we set *skip = 1:
3415  *  - the next message should be skipped and ignored.
3416  *  - con->in_msg == NULL
3417  * or if we set *skip = 0:
3418  *  - con->in_msg is non-null.
3419  * On error (ENOMEM, EAGAIN, ...),
3420  *  - con->in_msg == NULL
3421  */
3422 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3423 {
3424 	struct ceph_msg_header *hdr = &con->in_hdr;
3425 	int middle_len = le32_to_cpu(hdr->middle_len);
3426 	struct ceph_msg *msg;
3427 	int ret = 0;
3428 
3429 	BUG_ON(con->in_msg != NULL);
3430 	BUG_ON(!con->ops->alloc_msg);
3431 
3432 	mutex_unlock(&con->mutex);
3433 	msg = con->ops->alloc_msg(con, hdr, skip);
3434 	mutex_lock(&con->mutex);
3435 	if (con->state != CON_STATE_OPEN) {
3436 		if (msg)
3437 			ceph_msg_put(msg);
3438 		return -EAGAIN;
3439 	}
3440 	if (msg) {
3441 		BUG_ON(*skip);
3442 		msg_con_set(msg, con);
3443 		con->in_msg = msg;
3444 	} else {
3445 		/*
3446 		 * Null message pointer means either we should skip
3447 		 * this message or we couldn't allocate memory.  The
3448 		 * former is not an error.
3449 		 */
3450 		if (*skip)
3451 			return 0;
3452 
3453 		con->error_msg = "error allocating memory for incoming message";
3454 		return -ENOMEM;
3455 	}
3456 	memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3457 
3458 	if (middle_len && !con->in_msg->middle) {
3459 		ret = ceph_alloc_middle(con, con->in_msg);
3460 		if (ret < 0) {
3461 			ceph_msg_put(con->in_msg);
3462 			con->in_msg = NULL;
3463 		}
3464 	}
3465 
3466 	return ret;
3467 }
3468 
3469 
3470 /*
3471  * Free a generically kmalloc'd message.
3472  */
3473 static void ceph_msg_free(struct ceph_msg *m)
3474 {
3475 	dout("%s %p\n", __func__, m);
3476 	kvfree(m->front.iov_base);
3477 	kfree(m->data);
3478 	kmem_cache_free(ceph_msg_cache, m);
3479 }
3480 
3481 static void ceph_msg_release(struct kref *kref)
3482 {
3483 	struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3484 	int i;
3485 
3486 	dout("%s %p\n", __func__, m);
3487 	WARN_ON(!list_empty(&m->list_head));
3488 
3489 	msg_con_set(m, NULL);
3490 
3491 	/* drop middle, data, if any */
3492 	if (m->middle) {
3493 		ceph_buffer_put(m->middle);
3494 		m->middle = NULL;
3495 	}
3496 
3497 	for (i = 0; i < m->num_data_items; i++)
3498 		ceph_msg_data_destroy(&m->data[i]);
3499 
3500 	if (m->pool)
3501 		ceph_msgpool_put(m->pool, m);
3502 	else
3503 		ceph_msg_free(m);
3504 }
3505 
3506 struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3507 {
3508 	dout("%s %p (was %d)\n", __func__, msg,
3509 	     kref_read(&msg->kref));
3510 	kref_get(&msg->kref);
3511 	return msg;
3512 }
3513 EXPORT_SYMBOL(ceph_msg_get);
3514 
3515 void ceph_msg_put(struct ceph_msg *msg)
3516 {
3517 	dout("%s %p (was %d)\n", __func__, msg,
3518 	     kref_read(&msg->kref));
3519 	kref_put(&msg->kref, ceph_msg_release);
3520 }
3521 EXPORT_SYMBOL(ceph_msg_put);
3522 
3523 void ceph_msg_dump(struct ceph_msg *msg)
3524 {
3525 	pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3526 		 msg->front_alloc_len, msg->data_length);
3527 	print_hex_dump(KERN_DEBUG, "header: ",
3528 		       DUMP_PREFIX_OFFSET, 16, 1,
3529 		       &msg->hdr, sizeof(msg->hdr), true);
3530 	print_hex_dump(KERN_DEBUG, " front: ",
3531 		       DUMP_PREFIX_OFFSET, 16, 1,
3532 		       msg->front.iov_base, msg->front.iov_len, true);
3533 	if (msg->middle)
3534 		print_hex_dump(KERN_DEBUG, "middle: ",
3535 			       DUMP_PREFIX_OFFSET, 16, 1,
3536 			       msg->middle->vec.iov_base,
3537 			       msg->middle->vec.iov_len, true);
3538 	print_hex_dump(KERN_DEBUG, "footer: ",
3539 		       DUMP_PREFIX_OFFSET, 16, 1,
3540 		       &msg->footer, sizeof(msg->footer), true);
3541 }
3542 EXPORT_SYMBOL(ceph_msg_dump);
3543