xref: /openbmc/linux/net/ceph/messenger.c (revision 3c6a73cc)
1 #include <linux/ceph/ceph_debug.h>
2 
3 #include <linux/crc32c.h>
4 #include <linux/ctype.h>
5 #include <linux/highmem.h>
6 #include <linux/inet.h>
7 #include <linux/kthread.h>
8 #include <linux/net.h>
9 #include <linux/slab.h>
10 #include <linux/socket.h>
11 #include <linux/string.h>
12 #ifdef	CONFIG_BLOCK
13 #include <linux/bio.h>
14 #endif	/* CONFIG_BLOCK */
15 #include <linux/dns_resolver.h>
16 #include <net/tcp.h>
17 
18 #include <linux/ceph/ceph_features.h>
19 #include <linux/ceph/libceph.h>
20 #include <linux/ceph/messenger.h>
21 #include <linux/ceph/decode.h>
22 #include <linux/ceph/pagelist.h>
23 #include <linux/export.h>
24 
25 #define list_entry_next(pos, member)					\
26 	list_entry(pos->member.next, typeof(*pos), member)
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 static struct kmem_cache	*ceph_msg_data_cache;
160 
161 /* static tag bytes (protocol control messages) */
162 static char tag_msg = CEPH_MSGR_TAG_MSG;
163 static char tag_ack = CEPH_MSGR_TAG_ACK;
164 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
165 
166 #ifdef CONFIG_LOCKDEP
167 static struct lock_class_key socket_class;
168 #endif
169 
170 /*
171  * When skipping (ignoring) a block of input we read it into a "skip
172  * buffer," which is this many bytes in size.
173  */
174 #define SKIP_BUF_SIZE	1024
175 
176 static void queue_con(struct ceph_connection *con);
177 static void cancel_con(struct ceph_connection *con);
178 static void con_work(struct work_struct *);
179 static void con_fault(struct ceph_connection *con);
180 
181 /*
182  * Nicely render a sockaddr as a string.  An array of formatted
183  * strings is used, to approximate reentrancy.
184  */
185 #define ADDR_STR_COUNT_LOG	5	/* log2(# address strings in array) */
186 #define ADDR_STR_COUNT		(1 << ADDR_STR_COUNT_LOG)
187 #define ADDR_STR_COUNT_MASK	(ADDR_STR_COUNT - 1)
188 #define MAX_ADDR_STR_LEN	64	/* 54 is enough */
189 
190 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
191 static atomic_t addr_str_seq = ATOMIC_INIT(0);
192 
193 static struct page *zero_page;		/* used in certain error cases */
194 
195 const char *ceph_pr_addr(const struct sockaddr_storage *ss)
196 {
197 	int i;
198 	char *s;
199 	struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
200 	struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
201 
202 	i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
203 	s = addr_str[i];
204 
205 	switch (ss->ss_family) {
206 	case AF_INET:
207 		snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
208 			 ntohs(in4->sin_port));
209 		break;
210 
211 	case AF_INET6:
212 		snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
213 			 ntohs(in6->sin6_port));
214 		break;
215 
216 	default:
217 		snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
218 			 ss->ss_family);
219 	}
220 
221 	return s;
222 }
223 EXPORT_SYMBOL(ceph_pr_addr);
224 
225 static void encode_my_addr(struct ceph_messenger *msgr)
226 {
227 	memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
228 	ceph_encode_addr(&msgr->my_enc_addr);
229 }
230 
231 /*
232  * work queue for all reading and writing to/from the socket.
233  */
234 static struct workqueue_struct *ceph_msgr_wq;
235 
236 static int ceph_msgr_slab_init(void)
237 {
238 	BUG_ON(ceph_msg_cache);
239 	ceph_msg_cache = kmem_cache_create("ceph_msg",
240 					sizeof (struct ceph_msg),
241 					__alignof__(struct ceph_msg), 0, NULL);
242 
243 	if (!ceph_msg_cache)
244 		return -ENOMEM;
245 
246 	BUG_ON(ceph_msg_data_cache);
247 	ceph_msg_data_cache = kmem_cache_create("ceph_msg_data",
248 					sizeof (struct ceph_msg_data),
249 					__alignof__(struct ceph_msg_data),
250 					0, NULL);
251 	if (ceph_msg_data_cache)
252 		return 0;
253 
254 	kmem_cache_destroy(ceph_msg_cache);
255 	ceph_msg_cache = NULL;
256 
257 	return -ENOMEM;
258 }
259 
260 static void ceph_msgr_slab_exit(void)
261 {
262 	BUG_ON(!ceph_msg_data_cache);
263 	kmem_cache_destroy(ceph_msg_data_cache);
264 	ceph_msg_data_cache = NULL;
265 
266 	BUG_ON(!ceph_msg_cache);
267 	kmem_cache_destroy(ceph_msg_cache);
268 	ceph_msg_cache = NULL;
269 }
270 
271 static void _ceph_msgr_exit(void)
272 {
273 	if (ceph_msgr_wq) {
274 		destroy_workqueue(ceph_msgr_wq);
275 		ceph_msgr_wq = NULL;
276 	}
277 
278 	ceph_msgr_slab_exit();
279 
280 	BUG_ON(zero_page == NULL);
281 	kunmap(zero_page);
282 	page_cache_release(zero_page);
283 	zero_page = NULL;
284 }
285 
286 int ceph_msgr_init(void)
287 {
288 	BUG_ON(zero_page != NULL);
289 	zero_page = ZERO_PAGE(0);
290 	page_cache_get(zero_page);
291 
292 	if (ceph_msgr_slab_init())
293 		return -ENOMEM;
294 
295 	/*
296 	 * The number of active work items is limited by the number of
297 	 * connections, so leave @max_active at default.
298 	 */
299 	ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
300 	if (ceph_msgr_wq)
301 		return 0;
302 
303 	pr_err("msgr_init failed to create workqueue\n");
304 	_ceph_msgr_exit();
305 
306 	return -ENOMEM;
307 }
308 EXPORT_SYMBOL(ceph_msgr_init);
309 
310 void ceph_msgr_exit(void)
311 {
312 	BUG_ON(ceph_msgr_wq == NULL);
313 
314 	_ceph_msgr_exit();
315 }
316 EXPORT_SYMBOL(ceph_msgr_exit);
317 
318 void ceph_msgr_flush(void)
319 {
320 	flush_workqueue(ceph_msgr_wq);
321 }
322 EXPORT_SYMBOL(ceph_msgr_flush);
323 
324 /* Connection socket state transition functions */
325 
326 static void con_sock_state_init(struct ceph_connection *con)
327 {
328 	int old_state;
329 
330 	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
331 	if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
332 		printk("%s: unexpected old state %d\n", __func__, old_state);
333 	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
334 	     CON_SOCK_STATE_CLOSED);
335 }
336 
337 static void con_sock_state_connecting(struct ceph_connection *con)
338 {
339 	int old_state;
340 
341 	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
342 	if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
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_CONNECTING);
346 }
347 
348 static void con_sock_state_connected(struct ceph_connection *con)
349 {
350 	int old_state;
351 
352 	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
353 	if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
354 		printk("%s: unexpected old state %d\n", __func__, old_state);
355 	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
356 	     CON_SOCK_STATE_CONNECTED);
357 }
358 
359 static void con_sock_state_closing(struct ceph_connection *con)
360 {
361 	int old_state;
362 
363 	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
364 	if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
365 			old_state != CON_SOCK_STATE_CONNECTED &&
366 			old_state != CON_SOCK_STATE_CLOSING))
367 		printk("%s: unexpected old state %d\n", __func__, old_state);
368 	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
369 	     CON_SOCK_STATE_CLOSING);
370 }
371 
372 static void con_sock_state_closed(struct ceph_connection *con)
373 {
374 	int old_state;
375 
376 	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
377 	if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
378 		    old_state != CON_SOCK_STATE_CLOSING &&
379 		    old_state != CON_SOCK_STATE_CONNECTING &&
380 		    old_state != CON_SOCK_STATE_CLOSED))
381 		printk("%s: unexpected old state %d\n", __func__, old_state);
382 	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
383 	     CON_SOCK_STATE_CLOSED);
384 }
385 
386 /*
387  * socket callback functions
388  */
389 
390 /* data available on socket, or listen socket received a connect */
391 static void ceph_sock_data_ready(struct sock *sk)
392 {
393 	struct ceph_connection *con = sk->sk_user_data;
394 	if (atomic_read(&con->msgr->stopping)) {
395 		return;
396 	}
397 
398 	if (sk->sk_state != TCP_CLOSE_WAIT) {
399 		dout("%s on %p state = %lu, queueing work\n", __func__,
400 		     con, con->state);
401 		queue_con(con);
402 	}
403 }
404 
405 /* socket has buffer space for writing */
406 static void ceph_sock_write_space(struct sock *sk)
407 {
408 	struct ceph_connection *con = sk->sk_user_data;
409 
410 	/* only queue to workqueue if there is data we want to write,
411 	 * and there is sufficient space in the socket buffer to accept
412 	 * more data.  clear SOCK_NOSPACE so that ceph_sock_write_space()
413 	 * doesn't get called again until try_write() fills the socket
414 	 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
415 	 * and net/core/stream.c:sk_stream_write_space().
416 	 */
417 	if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
418 		if (sk_stream_is_writeable(sk)) {
419 			dout("%s %p queueing write work\n", __func__, con);
420 			clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
421 			queue_con(con);
422 		}
423 	} else {
424 		dout("%s %p nothing to write\n", __func__, con);
425 	}
426 }
427 
428 /* socket's state has changed */
429 static void ceph_sock_state_change(struct sock *sk)
430 {
431 	struct ceph_connection *con = sk->sk_user_data;
432 
433 	dout("%s %p state = %lu sk_state = %u\n", __func__,
434 	     con, con->state, sk->sk_state);
435 
436 	switch (sk->sk_state) {
437 	case TCP_CLOSE:
438 		dout("%s TCP_CLOSE\n", __func__);
439 	case TCP_CLOSE_WAIT:
440 		dout("%s TCP_CLOSE_WAIT\n", __func__);
441 		con_sock_state_closing(con);
442 		con_flag_set(con, CON_FLAG_SOCK_CLOSED);
443 		queue_con(con);
444 		break;
445 	case TCP_ESTABLISHED:
446 		dout("%s TCP_ESTABLISHED\n", __func__);
447 		con_sock_state_connected(con);
448 		queue_con(con);
449 		break;
450 	default:	/* Everything else is uninteresting */
451 		break;
452 	}
453 }
454 
455 /*
456  * set up socket callbacks
457  */
458 static void set_sock_callbacks(struct socket *sock,
459 			       struct ceph_connection *con)
460 {
461 	struct sock *sk = sock->sk;
462 	sk->sk_user_data = con;
463 	sk->sk_data_ready = ceph_sock_data_ready;
464 	sk->sk_write_space = ceph_sock_write_space;
465 	sk->sk_state_change = ceph_sock_state_change;
466 }
467 
468 
469 /*
470  * socket helpers
471  */
472 
473 /*
474  * initiate connection to a remote socket.
475  */
476 static int ceph_tcp_connect(struct ceph_connection *con)
477 {
478 	struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
479 	struct socket *sock;
480 	int ret;
481 
482 	BUG_ON(con->sock);
483 	ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
484 			       IPPROTO_TCP, &sock);
485 	if (ret)
486 		return ret;
487 	sock->sk->sk_allocation = GFP_NOFS;
488 
489 #ifdef CONFIG_LOCKDEP
490 	lockdep_set_class(&sock->sk->sk_lock, &socket_class);
491 #endif
492 
493 	set_sock_callbacks(sock, con);
494 
495 	dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
496 
497 	con_sock_state_connecting(con);
498 	ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
499 				 O_NONBLOCK);
500 	if (ret == -EINPROGRESS) {
501 		dout("connect %s EINPROGRESS sk_state = %u\n",
502 		     ceph_pr_addr(&con->peer_addr.in_addr),
503 		     sock->sk->sk_state);
504 	} else if (ret < 0) {
505 		pr_err("connect %s error %d\n",
506 		       ceph_pr_addr(&con->peer_addr.in_addr), ret);
507 		sock_release(sock);
508 		con->error_msg = "connect error";
509 
510 		return ret;
511 	}
512 	con->sock = sock;
513 	return 0;
514 }
515 
516 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
517 {
518 	struct kvec iov = {buf, len};
519 	struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
520 	int r;
521 
522 	r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
523 	if (r == -EAGAIN)
524 		r = 0;
525 	return r;
526 }
527 
528 static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
529 		     int page_offset, size_t length)
530 {
531 	void *kaddr;
532 	int ret;
533 
534 	BUG_ON(page_offset + length > PAGE_SIZE);
535 
536 	kaddr = kmap(page);
537 	BUG_ON(!kaddr);
538 	ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
539 	kunmap(page);
540 
541 	return ret;
542 }
543 
544 /*
545  * write something.  @more is true if caller will be sending more data
546  * shortly.
547  */
548 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
549 		     size_t kvlen, size_t len, int more)
550 {
551 	struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
552 	int r;
553 
554 	if (more)
555 		msg.msg_flags |= MSG_MORE;
556 	else
557 		msg.msg_flags |= MSG_EOR;  /* superfluous, but what the hell */
558 
559 	r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
560 	if (r == -EAGAIN)
561 		r = 0;
562 	return r;
563 }
564 
565 static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
566 		     int offset, size_t size, bool more)
567 {
568 	int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
569 	int ret;
570 
571 	ret = kernel_sendpage(sock, page, offset, size, flags);
572 	if (ret == -EAGAIN)
573 		ret = 0;
574 
575 	return ret;
576 }
577 
578 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
579 		     int offset, size_t size, bool more)
580 {
581 	int ret;
582 	struct kvec iov;
583 
584 	/* sendpage cannot properly handle pages with page_count == 0,
585 	 * we need to fallback to sendmsg if that's the case */
586 	if (page_count(page) >= 1)
587 		return __ceph_tcp_sendpage(sock, page, offset, size, more);
588 
589 	iov.iov_base = kmap(page) + offset;
590 	iov.iov_len = size;
591 	ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more);
592 	kunmap(page);
593 
594 	return ret;
595 }
596 
597 /*
598  * Shutdown/close the socket for the given connection.
599  */
600 static int con_close_socket(struct ceph_connection *con)
601 {
602 	int rc = 0;
603 
604 	dout("con_close_socket on %p sock %p\n", con, con->sock);
605 	if (con->sock) {
606 		rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
607 		sock_release(con->sock);
608 		con->sock = NULL;
609 	}
610 
611 	/*
612 	 * Forcibly clear the SOCK_CLOSED flag.  It gets set
613 	 * independent of the connection mutex, and we could have
614 	 * received a socket close event before we had the chance to
615 	 * shut the socket down.
616 	 */
617 	con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
618 
619 	con_sock_state_closed(con);
620 	return rc;
621 }
622 
623 /*
624  * Reset a connection.  Discard all incoming and outgoing messages
625  * and clear *_seq state.
626  */
627 static void ceph_msg_remove(struct ceph_msg *msg)
628 {
629 	list_del_init(&msg->list_head);
630 	BUG_ON(msg->con == NULL);
631 	msg->con->ops->put(msg->con);
632 	msg->con = NULL;
633 
634 	ceph_msg_put(msg);
635 }
636 static void ceph_msg_remove_list(struct list_head *head)
637 {
638 	while (!list_empty(head)) {
639 		struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
640 							list_head);
641 		ceph_msg_remove(msg);
642 	}
643 }
644 
645 static void reset_connection(struct ceph_connection *con)
646 {
647 	/* reset connection, out_queue, msg_ and connect_seq */
648 	/* discard existing out_queue and msg_seq */
649 	dout("reset_connection %p\n", con);
650 	ceph_msg_remove_list(&con->out_queue);
651 	ceph_msg_remove_list(&con->out_sent);
652 
653 	if (con->in_msg) {
654 		BUG_ON(con->in_msg->con != con);
655 		con->in_msg->con = NULL;
656 		ceph_msg_put(con->in_msg);
657 		con->in_msg = NULL;
658 		con->ops->put(con);
659 	}
660 
661 	con->connect_seq = 0;
662 	con->out_seq = 0;
663 	if (con->out_msg) {
664 		ceph_msg_put(con->out_msg);
665 		con->out_msg = NULL;
666 	}
667 	con->in_seq = 0;
668 	con->in_seq_acked = 0;
669 }
670 
671 /*
672  * mark a peer down.  drop any open connections.
673  */
674 void ceph_con_close(struct ceph_connection *con)
675 {
676 	mutex_lock(&con->mutex);
677 	dout("con_close %p peer %s\n", con,
678 	     ceph_pr_addr(&con->peer_addr.in_addr));
679 	con->state = CON_STATE_CLOSED;
680 
681 	con_flag_clear(con, CON_FLAG_LOSSYTX);	/* so we retry next connect */
682 	con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
683 	con_flag_clear(con, CON_FLAG_WRITE_PENDING);
684 	con_flag_clear(con, CON_FLAG_BACKOFF);
685 
686 	reset_connection(con);
687 	con->peer_global_seq = 0;
688 	cancel_con(con);
689 	con_close_socket(con);
690 	mutex_unlock(&con->mutex);
691 }
692 EXPORT_SYMBOL(ceph_con_close);
693 
694 /*
695  * Reopen a closed connection, with a new peer address.
696  */
697 void ceph_con_open(struct ceph_connection *con,
698 		   __u8 entity_type, __u64 entity_num,
699 		   struct ceph_entity_addr *addr)
700 {
701 	mutex_lock(&con->mutex);
702 	dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
703 
704 	WARN_ON(con->state != CON_STATE_CLOSED);
705 	con->state = CON_STATE_PREOPEN;
706 
707 	con->peer_name.type = (__u8) entity_type;
708 	con->peer_name.num = cpu_to_le64(entity_num);
709 
710 	memcpy(&con->peer_addr, addr, sizeof(*addr));
711 	con->delay = 0;      /* reset backoff memory */
712 	mutex_unlock(&con->mutex);
713 	queue_con(con);
714 }
715 EXPORT_SYMBOL(ceph_con_open);
716 
717 /*
718  * return true if this connection ever successfully opened
719  */
720 bool ceph_con_opened(struct ceph_connection *con)
721 {
722 	return con->connect_seq > 0;
723 }
724 
725 /*
726  * initialize a new connection.
727  */
728 void ceph_con_init(struct ceph_connection *con, void *private,
729 	const struct ceph_connection_operations *ops,
730 	struct ceph_messenger *msgr)
731 {
732 	dout("con_init %p\n", con);
733 	memset(con, 0, sizeof(*con));
734 	con->private = private;
735 	con->ops = ops;
736 	con->msgr = msgr;
737 
738 	con_sock_state_init(con);
739 
740 	mutex_init(&con->mutex);
741 	INIT_LIST_HEAD(&con->out_queue);
742 	INIT_LIST_HEAD(&con->out_sent);
743 	INIT_DELAYED_WORK(&con->work, con_work);
744 
745 	con->state = CON_STATE_CLOSED;
746 }
747 EXPORT_SYMBOL(ceph_con_init);
748 
749 
750 /*
751  * We maintain a global counter to order connection attempts.  Get
752  * a unique seq greater than @gt.
753  */
754 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
755 {
756 	u32 ret;
757 
758 	spin_lock(&msgr->global_seq_lock);
759 	if (msgr->global_seq < gt)
760 		msgr->global_seq = gt;
761 	ret = ++msgr->global_seq;
762 	spin_unlock(&msgr->global_seq_lock);
763 	return ret;
764 }
765 
766 static void con_out_kvec_reset(struct ceph_connection *con)
767 {
768 	con->out_kvec_left = 0;
769 	con->out_kvec_bytes = 0;
770 	con->out_kvec_cur = &con->out_kvec[0];
771 }
772 
773 static void con_out_kvec_add(struct ceph_connection *con,
774 				size_t size, void *data)
775 {
776 	int index;
777 
778 	index = con->out_kvec_left;
779 	BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
780 
781 	con->out_kvec[index].iov_len = size;
782 	con->out_kvec[index].iov_base = data;
783 	con->out_kvec_left++;
784 	con->out_kvec_bytes += size;
785 }
786 
787 #ifdef CONFIG_BLOCK
788 
789 /*
790  * For a bio data item, a piece is whatever remains of the next
791  * entry in the current bio iovec, or the first entry in the next
792  * bio in the list.
793  */
794 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
795 					size_t length)
796 {
797 	struct ceph_msg_data *data = cursor->data;
798 	struct bio *bio;
799 
800 	BUG_ON(data->type != CEPH_MSG_DATA_BIO);
801 
802 	bio = data->bio;
803 	BUG_ON(!bio);
804 
805 	cursor->resid = min(length, data->bio_length);
806 	cursor->bio = bio;
807 	cursor->bvec_iter = bio->bi_iter;
808 	cursor->last_piece =
809 		cursor->resid <= bio_iter_len(bio, cursor->bvec_iter);
810 }
811 
812 static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
813 						size_t *page_offset,
814 						size_t *length)
815 {
816 	struct ceph_msg_data *data = cursor->data;
817 	struct bio *bio;
818 	struct bio_vec bio_vec;
819 
820 	BUG_ON(data->type != CEPH_MSG_DATA_BIO);
821 
822 	bio = cursor->bio;
823 	BUG_ON(!bio);
824 
825 	bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
826 
827 	*page_offset = (size_t) bio_vec.bv_offset;
828 	BUG_ON(*page_offset >= PAGE_SIZE);
829 	if (cursor->last_piece) /* pagelist offset is always 0 */
830 		*length = cursor->resid;
831 	else
832 		*length = (size_t) bio_vec.bv_len;
833 	BUG_ON(*length > cursor->resid);
834 	BUG_ON(*page_offset + *length > PAGE_SIZE);
835 
836 	return bio_vec.bv_page;
837 }
838 
839 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
840 					size_t bytes)
841 {
842 	struct bio *bio;
843 	struct bio_vec bio_vec;
844 
845 	BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
846 
847 	bio = cursor->bio;
848 	BUG_ON(!bio);
849 
850 	bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
851 
852 	/* Advance the cursor offset */
853 
854 	BUG_ON(cursor->resid < bytes);
855 	cursor->resid -= bytes;
856 
857 	bio_advance_iter(bio, &cursor->bvec_iter, bytes);
858 
859 	if (bytes < bio_vec.bv_len)
860 		return false;	/* more bytes to process in this segment */
861 
862 	/* Move on to the next segment, and possibly the next bio */
863 
864 	if (!cursor->bvec_iter.bi_size) {
865 		bio = bio->bi_next;
866 		cursor->bio = bio;
867 		if (bio)
868 			cursor->bvec_iter = bio->bi_iter;
869 		else
870 			memset(&cursor->bvec_iter, 0,
871 			       sizeof(cursor->bvec_iter));
872 	}
873 
874 	if (!cursor->last_piece) {
875 		BUG_ON(!cursor->resid);
876 		BUG_ON(!bio);
877 		/* A short read is OK, so use <= rather than == */
878 		if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter))
879 			cursor->last_piece = true;
880 	}
881 
882 	return true;
883 }
884 #endif /* CONFIG_BLOCK */
885 
886 /*
887  * For a page array, a piece comes from the first page in the array
888  * that has not already been fully consumed.
889  */
890 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
891 					size_t length)
892 {
893 	struct ceph_msg_data *data = cursor->data;
894 	int page_count;
895 
896 	BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
897 
898 	BUG_ON(!data->pages);
899 	BUG_ON(!data->length);
900 
901 	cursor->resid = min(length, data->length);
902 	page_count = calc_pages_for(data->alignment, (u64)data->length);
903 	cursor->page_offset = data->alignment & ~PAGE_MASK;
904 	cursor->page_index = 0;
905 	BUG_ON(page_count > (int)USHRT_MAX);
906 	cursor->page_count = (unsigned short)page_count;
907 	BUG_ON(length > SIZE_MAX - cursor->page_offset);
908 	cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
909 }
910 
911 static struct page *
912 ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
913 					size_t *page_offset, size_t *length)
914 {
915 	struct ceph_msg_data *data = cursor->data;
916 
917 	BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
918 
919 	BUG_ON(cursor->page_index >= cursor->page_count);
920 	BUG_ON(cursor->page_offset >= PAGE_SIZE);
921 
922 	*page_offset = cursor->page_offset;
923 	if (cursor->last_piece)
924 		*length = cursor->resid;
925 	else
926 		*length = PAGE_SIZE - *page_offset;
927 
928 	return data->pages[cursor->page_index];
929 }
930 
931 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
932 						size_t bytes)
933 {
934 	BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
935 
936 	BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
937 
938 	/* Advance the cursor page offset */
939 
940 	cursor->resid -= bytes;
941 	cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
942 	if (!bytes || cursor->page_offset)
943 		return false;	/* more bytes to process in the current page */
944 
945 	if (!cursor->resid)
946 		return false;   /* no more data */
947 
948 	/* Move on to the next page; offset is already at 0 */
949 
950 	BUG_ON(cursor->page_index >= cursor->page_count);
951 	cursor->page_index++;
952 	cursor->last_piece = cursor->resid <= PAGE_SIZE;
953 
954 	return true;
955 }
956 
957 /*
958  * For a pagelist, a piece is whatever remains to be consumed in the
959  * first page in the list, or the front of the next page.
960  */
961 static void
962 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
963 					size_t length)
964 {
965 	struct ceph_msg_data *data = cursor->data;
966 	struct ceph_pagelist *pagelist;
967 	struct page *page;
968 
969 	BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
970 
971 	pagelist = data->pagelist;
972 	BUG_ON(!pagelist);
973 
974 	if (!length)
975 		return;		/* pagelist can be assigned but empty */
976 
977 	BUG_ON(list_empty(&pagelist->head));
978 	page = list_first_entry(&pagelist->head, struct page, lru);
979 
980 	cursor->resid = min(length, pagelist->length);
981 	cursor->page = page;
982 	cursor->offset = 0;
983 	cursor->last_piece = cursor->resid <= PAGE_SIZE;
984 }
985 
986 static struct page *
987 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
988 				size_t *page_offset, size_t *length)
989 {
990 	struct ceph_msg_data *data = cursor->data;
991 	struct ceph_pagelist *pagelist;
992 
993 	BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
994 
995 	pagelist = data->pagelist;
996 	BUG_ON(!pagelist);
997 
998 	BUG_ON(!cursor->page);
999 	BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1000 
1001 	/* offset of first page in pagelist is always 0 */
1002 	*page_offset = cursor->offset & ~PAGE_MASK;
1003 	if (cursor->last_piece)
1004 		*length = cursor->resid;
1005 	else
1006 		*length = PAGE_SIZE - *page_offset;
1007 
1008 	return cursor->page;
1009 }
1010 
1011 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
1012 						size_t bytes)
1013 {
1014 	struct ceph_msg_data *data = cursor->data;
1015 	struct ceph_pagelist *pagelist;
1016 
1017 	BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1018 
1019 	pagelist = data->pagelist;
1020 	BUG_ON(!pagelist);
1021 
1022 	BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1023 	BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1024 
1025 	/* Advance the cursor offset */
1026 
1027 	cursor->resid -= bytes;
1028 	cursor->offset += bytes;
1029 	/* offset of first page in pagelist is always 0 */
1030 	if (!bytes || cursor->offset & ~PAGE_MASK)
1031 		return false;	/* more bytes to process in the current page */
1032 
1033 	if (!cursor->resid)
1034 		return false;   /* no more data */
1035 
1036 	/* Move on to the next page */
1037 
1038 	BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1039 	cursor->page = list_entry_next(cursor->page, lru);
1040 	cursor->last_piece = cursor->resid <= PAGE_SIZE;
1041 
1042 	return true;
1043 }
1044 
1045 /*
1046  * Message data is handled (sent or received) in pieces, where each
1047  * piece resides on a single page.  The network layer might not
1048  * consume an entire piece at once.  A data item's cursor keeps
1049  * track of which piece is next to process and how much remains to
1050  * be processed in that piece.  It also tracks whether the current
1051  * piece is the last one in the data item.
1052  */
1053 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1054 {
1055 	size_t length = cursor->total_resid;
1056 
1057 	switch (cursor->data->type) {
1058 	case CEPH_MSG_DATA_PAGELIST:
1059 		ceph_msg_data_pagelist_cursor_init(cursor, length);
1060 		break;
1061 	case CEPH_MSG_DATA_PAGES:
1062 		ceph_msg_data_pages_cursor_init(cursor, length);
1063 		break;
1064 #ifdef CONFIG_BLOCK
1065 	case CEPH_MSG_DATA_BIO:
1066 		ceph_msg_data_bio_cursor_init(cursor, length);
1067 		break;
1068 #endif /* CONFIG_BLOCK */
1069 	case CEPH_MSG_DATA_NONE:
1070 	default:
1071 		/* BUG(); */
1072 		break;
1073 	}
1074 	cursor->need_crc = true;
1075 }
1076 
1077 static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1078 {
1079 	struct ceph_msg_data_cursor *cursor = &msg->cursor;
1080 	struct ceph_msg_data *data;
1081 
1082 	BUG_ON(!length);
1083 	BUG_ON(length > msg->data_length);
1084 	BUG_ON(list_empty(&msg->data));
1085 
1086 	cursor->data_head = &msg->data;
1087 	cursor->total_resid = length;
1088 	data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1089 	cursor->data = data;
1090 
1091 	__ceph_msg_data_cursor_init(cursor);
1092 }
1093 
1094 /*
1095  * Return the page containing the next piece to process for a given
1096  * data item, and supply the page offset and length of that piece.
1097  * Indicate whether this is the last piece in this data item.
1098  */
1099 static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1100 					size_t *page_offset, size_t *length,
1101 					bool *last_piece)
1102 {
1103 	struct page *page;
1104 
1105 	switch (cursor->data->type) {
1106 	case CEPH_MSG_DATA_PAGELIST:
1107 		page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1108 		break;
1109 	case CEPH_MSG_DATA_PAGES:
1110 		page = ceph_msg_data_pages_next(cursor, page_offset, length);
1111 		break;
1112 #ifdef CONFIG_BLOCK
1113 	case CEPH_MSG_DATA_BIO:
1114 		page = ceph_msg_data_bio_next(cursor, page_offset, length);
1115 		break;
1116 #endif /* CONFIG_BLOCK */
1117 	case CEPH_MSG_DATA_NONE:
1118 	default:
1119 		page = NULL;
1120 		break;
1121 	}
1122 	BUG_ON(!page);
1123 	BUG_ON(*page_offset + *length > PAGE_SIZE);
1124 	BUG_ON(!*length);
1125 	if (last_piece)
1126 		*last_piece = cursor->last_piece;
1127 
1128 	return page;
1129 }
1130 
1131 /*
1132  * Returns true if the result moves the cursor on to the next piece
1133  * of the data item.
1134  */
1135 static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1136 				size_t bytes)
1137 {
1138 	bool new_piece;
1139 
1140 	BUG_ON(bytes > cursor->resid);
1141 	switch (cursor->data->type) {
1142 	case CEPH_MSG_DATA_PAGELIST:
1143 		new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1144 		break;
1145 	case CEPH_MSG_DATA_PAGES:
1146 		new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1147 		break;
1148 #ifdef CONFIG_BLOCK
1149 	case CEPH_MSG_DATA_BIO:
1150 		new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1151 		break;
1152 #endif /* CONFIG_BLOCK */
1153 	case CEPH_MSG_DATA_NONE:
1154 	default:
1155 		BUG();
1156 		break;
1157 	}
1158 	cursor->total_resid -= bytes;
1159 
1160 	if (!cursor->resid && cursor->total_resid) {
1161 		WARN_ON(!cursor->last_piece);
1162 		BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1163 		cursor->data = list_entry_next(cursor->data, links);
1164 		__ceph_msg_data_cursor_init(cursor);
1165 		new_piece = true;
1166 	}
1167 	cursor->need_crc = new_piece;
1168 
1169 	return new_piece;
1170 }
1171 
1172 static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
1173 {
1174 	BUG_ON(!msg);
1175 	BUG_ON(!data_len);
1176 
1177 	/* Initialize data cursor */
1178 
1179 	ceph_msg_data_cursor_init(msg, (size_t)data_len);
1180 }
1181 
1182 /*
1183  * Prepare footer for currently outgoing message, and finish things
1184  * off.  Assumes out_kvec* are already valid.. we just add on to the end.
1185  */
1186 static void prepare_write_message_footer(struct ceph_connection *con)
1187 {
1188 	struct ceph_msg *m = con->out_msg;
1189 	int v = con->out_kvec_left;
1190 
1191 	m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1192 
1193 	dout("prepare_write_message_footer %p\n", con);
1194 	con->out_kvec_is_msg = true;
1195 	con->out_kvec[v].iov_base = &m->footer;
1196 	con->out_kvec[v].iov_len = sizeof(m->footer);
1197 	con->out_kvec_bytes += sizeof(m->footer);
1198 	con->out_kvec_left++;
1199 	con->out_more = m->more_to_follow;
1200 	con->out_msg_done = true;
1201 }
1202 
1203 /*
1204  * Prepare headers for the next outgoing message.
1205  */
1206 static void prepare_write_message(struct ceph_connection *con)
1207 {
1208 	struct ceph_msg *m;
1209 	u32 crc;
1210 
1211 	con_out_kvec_reset(con);
1212 	con->out_kvec_is_msg = true;
1213 	con->out_msg_done = false;
1214 
1215 	/* Sneak an ack in there first?  If we can get it into the same
1216 	 * TCP packet that's a good thing. */
1217 	if (con->in_seq > con->in_seq_acked) {
1218 		con->in_seq_acked = con->in_seq;
1219 		con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1220 		con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1221 		con_out_kvec_add(con, sizeof (con->out_temp_ack),
1222 			&con->out_temp_ack);
1223 	}
1224 
1225 	BUG_ON(list_empty(&con->out_queue));
1226 	m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1227 	con->out_msg = m;
1228 	BUG_ON(m->con != con);
1229 
1230 	/* put message on sent list */
1231 	ceph_msg_get(m);
1232 	list_move_tail(&m->list_head, &con->out_sent);
1233 
1234 	/*
1235 	 * only assign outgoing seq # if we haven't sent this message
1236 	 * yet.  if it is requeued, resend with it's original seq.
1237 	 */
1238 	if (m->needs_out_seq) {
1239 		m->hdr.seq = cpu_to_le64(++con->out_seq);
1240 		m->needs_out_seq = false;
1241 	}
1242 	WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
1243 
1244 	dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1245 	     m, con->out_seq, le16_to_cpu(m->hdr.type),
1246 	     le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1247 	     m->data_length);
1248 	BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1249 
1250 	/* tag + hdr + front + middle */
1251 	con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1252 	con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1253 	con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1254 
1255 	if (m->middle)
1256 		con_out_kvec_add(con, m->middle->vec.iov_len,
1257 			m->middle->vec.iov_base);
1258 
1259 	/* fill in crc (except data pages), footer */
1260 	crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1261 	con->out_msg->hdr.crc = cpu_to_le32(crc);
1262 	con->out_msg->footer.flags = 0;
1263 
1264 	crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1265 	con->out_msg->footer.front_crc = cpu_to_le32(crc);
1266 	if (m->middle) {
1267 		crc = crc32c(0, m->middle->vec.iov_base,
1268 				m->middle->vec.iov_len);
1269 		con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1270 	} else
1271 		con->out_msg->footer.middle_crc = 0;
1272 	dout("%s front_crc %u middle_crc %u\n", __func__,
1273 	     le32_to_cpu(con->out_msg->footer.front_crc),
1274 	     le32_to_cpu(con->out_msg->footer.middle_crc));
1275 
1276 	/* is there a data payload? */
1277 	con->out_msg->footer.data_crc = 0;
1278 	if (m->data_length) {
1279 		prepare_message_data(con->out_msg, m->data_length);
1280 		con->out_more = 1;  /* data + footer will follow */
1281 	} else {
1282 		/* no, queue up footer too and be done */
1283 		prepare_write_message_footer(con);
1284 	}
1285 
1286 	con_flag_set(con, CON_FLAG_WRITE_PENDING);
1287 }
1288 
1289 /*
1290  * Prepare an ack.
1291  */
1292 static void prepare_write_ack(struct ceph_connection *con)
1293 {
1294 	dout("prepare_write_ack %p %llu -> %llu\n", con,
1295 	     con->in_seq_acked, con->in_seq);
1296 	con->in_seq_acked = con->in_seq;
1297 
1298 	con_out_kvec_reset(con);
1299 
1300 	con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1301 
1302 	con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1303 	con_out_kvec_add(con, sizeof (con->out_temp_ack),
1304 				&con->out_temp_ack);
1305 
1306 	con->out_more = 1;  /* more will follow.. eventually.. */
1307 	con_flag_set(con, CON_FLAG_WRITE_PENDING);
1308 }
1309 
1310 /*
1311  * Prepare to share the seq during handshake
1312  */
1313 static void prepare_write_seq(struct ceph_connection *con)
1314 {
1315 	dout("prepare_write_seq %p %llu -> %llu\n", con,
1316 	     con->in_seq_acked, con->in_seq);
1317 	con->in_seq_acked = con->in_seq;
1318 
1319 	con_out_kvec_reset(con);
1320 
1321 	con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1322 	con_out_kvec_add(con, sizeof (con->out_temp_ack),
1323 			 &con->out_temp_ack);
1324 
1325 	con_flag_set(con, CON_FLAG_WRITE_PENDING);
1326 }
1327 
1328 /*
1329  * Prepare to write keepalive byte.
1330  */
1331 static void prepare_write_keepalive(struct ceph_connection *con)
1332 {
1333 	dout("prepare_write_keepalive %p\n", con);
1334 	con_out_kvec_reset(con);
1335 	con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
1336 	con_flag_set(con, CON_FLAG_WRITE_PENDING);
1337 }
1338 
1339 /*
1340  * Connection negotiation.
1341  */
1342 
1343 static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1344 						int *auth_proto)
1345 {
1346 	struct ceph_auth_handshake *auth;
1347 
1348 	if (!con->ops->get_authorizer) {
1349 		con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1350 		con->out_connect.authorizer_len = 0;
1351 		return NULL;
1352 	}
1353 
1354 	/* Can't hold the mutex while getting authorizer */
1355 	mutex_unlock(&con->mutex);
1356 	auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
1357 	mutex_lock(&con->mutex);
1358 
1359 	if (IS_ERR(auth))
1360 		return auth;
1361 	if (con->state != CON_STATE_NEGOTIATING)
1362 		return ERR_PTR(-EAGAIN);
1363 
1364 	con->auth_reply_buf = auth->authorizer_reply_buf;
1365 	con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
1366 	return auth;
1367 }
1368 
1369 /*
1370  * We connected to a peer and are saying hello.
1371  */
1372 static void prepare_write_banner(struct ceph_connection *con)
1373 {
1374 	con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1375 	con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1376 					&con->msgr->my_enc_addr);
1377 
1378 	con->out_more = 0;
1379 	con_flag_set(con, CON_FLAG_WRITE_PENDING);
1380 }
1381 
1382 static int prepare_write_connect(struct ceph_connection *con)
1383 {
1384 	unsigned int global_seq = get_global_seq(con->msgr, 0);
1385 	int proto;
1386 	int auth_proto;
1387 	struct ceph_auth_handshake *auth;
1388 
1389 	switch (con->peer_name.type) {
1390 	case CEPH_ENTITY_TYPE_MON:
1391 		proto = CEPH_MONC_PROTOCOL;
1392 		break;
1393 	case CEPH_ENTITY_TYPE_OSD:
1394 		proto = CEPH_OSDC_PROTOCOL;
1395 		break;
1396 	case CEPH_ENTITY_TYPE_MDS:
1397 		proto = CEPH_MDSC_PROTOCOL;
1398 		break;
1399 	default:
1400 		BUG();
1401 	}
1402 
1403 	dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1404 	     con->connect_seq, global_seq, proto);
1405 
1406 	con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
1407 	con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1408 	con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1409 	con->out_connect.global_seq = cpu_to_le32(global_seq);
1410 	con->out_connect.protocol_version = cpu_to_le32(proto);
1411 	con->out_connect.flags = 0;
1412 
1413 	auth_proto = CEPH_AUTH_UNKNOWN;
1414 	auth = get_connect_authorizer(con, &auth_proto);
1415 	if (IS_ERR(auth))
1416 		return PTR_ERR(auth);
1417 
1418 	con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1419 	con->out_connect.authorizer_len = auth ?
1420 		cpu_to_le32(auth->authorizer_buf_len) : 0;
1421 
1422 	con_out_kvec_add(con, sizeof (con->out_connect),
1423 					&con->out_connect);
1424 	if (auth && auth->authorizer_buf_len)
1425 		con_out_kvec_add(con, auth->authorizer_buf_len,
1426 					auth->authorizer_buf);
1427 
1428 	con->out_more = 0;
1429 	con_flag_set(con, CON_FLAG_WRITE_PENDING);
1430 
1431 	return 0;
1432 }
1433 
1434 /*
1435  * write as much of pending kvecs to the socket as we can.
1436  *  1 -> done
1437  *  0 -> socket full, but more to do
1438  * <0 -> error
1439  */
1440 static int write_partial_kvec(struct ceph_connection *con)
1441 {
1442 	int ret;
1443 
1444 	dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1445 	while (con->out_kvec_bytes > 0) {
1446 		ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1447 				       con->out_kvec_left, con->out_kvec_bytes,
1448 				       con->out_more);
1449 		if (ret <= 0)
1450 			goto out;
1451 		con->out_kvec_bytes -= ret;
1452 		if (con->out_kvec_bytes == 0)
1453 			break;            /* done */
1454 
1455 		/* account for full iov entries consumed */
1456 		while (ret >= con->out_kvec_cur->iov_len) {
1457 			BUG_ON(!con->out_kvec_left);
1458 			ret -= con->out_kvec_cur->iov_len;
1459 			con->out_kvec_cur++;
1460 			con->out_kvec_left--;
1461 		}
1462 		/* and for a partially-consumed entry */
1463 		if (ret) {
1464 			con->out_kvec_cur->iov_len -= ret;
1465 			con->out_kvec_cur->iov_base += ret;
1466 		}
1467 	}
1468 	con->out_kvec_left = 0;
1469 	con->out_kvec_is_msg = false;
1470 	ret = 1;
1471 out:
1472 	dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1473 	     con->out_kvec_bytes, con->out_kvec_left, ret);
1474 	return ret;  /* done! */
1475 }
1476 
1477 static u32 ceph_crc32c_page(u32 crc, struct page *page,
1478 				unsigned int page_offset,
1479 				unsigned int length)
1480 {
1481 	char *kaddr;
1482 
1483 	kaddr = kmap(page);
1484 	BUG_ON(kaddr == NULL);
1485 	crc = crc32c(crc, kaddr + page_offset, length);
1486 	kunmap(page);
1487 
1488 	return crc;
1489 }
1490 /*
1491  * Write as much message data payload as we can.  If we finish, queue
1492  * up the footer.
1493  *  1 -> done, footer is now queued in out_kvec[].
1494  *  0 -> socket full, but more to do
1495  * <0 -> error
1496  */
1497 static int write_partial_message_data(struct ceph_connection *con)
1498 {
1499 	struct ceph_msg *msg = con->out_msg;
1500 	struct ceph_msg_data_cursor *cursor = &msg->cursor;
1501 	bool do_datacrc = !con->msgr->nocrc;
1502 	u32 crc;
1503 
1504 	dout("%s %p msg %p\n", __func__, con, msg);
1505 
1506 	if (list_empty(&msg->data))
1507 		return -EINVAL;
1508 
1509 	/*
1510 	 * Iterate through each page that contains data to be
1511 	 * written, and send as much as possible for each.
1512 	 *
1513 	 * If we are calculating the data crc (the default), we will
1514 	 * need to map the page.  If we have no pages, they have
1515 	 * been revoked, so use the zero page.
1516 	 */
1517 	crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1518 	while (cursor->resid) {
1519 		struct page *page;
1520 		size_t page_offset;
1521 		size_t length;
1522 		bool last_piece;
1523 		bool need_crc;
1524 		int ret;
1525 
1526 		page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
1527 							&last_piece);
1528 		ret = ceph_tcp_sendpage(con->sock, page, page_offset,
1529 				      length, last_piece);
1530 		if (ret <= 0) {
1531 			if (do_datacrc)
1532 				msg->footer.data_crc = cpu_to_le32(crc);
1533 
1534 			return ret;
1535 		}
1536 		if (do_datacrc && cursor->need_crc)
1537 			crc = ceph_crc32c_page(crc, page, page_offset, length);
1538 		need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret);
1539 	}
1540 
1541 	dout("%s %p msg %p done\n", __func__, con, msg);
1542 
1543 	/* prepare and queue up footer, too */
1544 	if (do_datacrc)
1545 		msg->footer.data_crc = cpu_to_le32(crc);
1546 	else
1547 		msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1548 	con_out_kvec_reset(con);
1549 	prepare_write_message_footer(con);
1550 
1551 	return 1;	/* must return > 0 to indicate success */
1552 }
1553 
1554 /*
1555  * write some zeros
1556  */
1557 static int write_partial_skip(struct ceph_connection *con)
1558 {
1559 	int ret;
1560 
1561 	while (con->out_skip > 0) {
1562 		size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
1563 
1564 		ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
1565 		if (ret <= 0)
1566 			goto out;
1567 		con->out_skip -= ret;
1568 	}
1569 	ret = 1;
1570 out:
1571 	return ret;
1572 }
1573 
1574 /*
1575  * Prepare to read connection handshake, or an ack.
1576  */
1577 static void prepare_read_banner(struct ceph_connection *con)
1578 {
1579 	dout("prepare_read_banner %p\n", con);
1580 	con->in_base_pos = 0;
1581 }
1582 
1583 static void prepare_read_connect(struct ceph_connection *con)
1584 {
1585 	dout("prepare_read_connect %p\n", con);
1586 	con->in_base_pos = 0;
1587 }
1588 
1589 static void prepare_read_ack(struct ceph_connection *con)
1590 {
1591 	dout("prepare_read_ack %p\n", con);
1592 	con->in_base_pos = 0;
1593 }
1594 
1595 static void prepare_read_seq(struct ceph_connection *con)
1596 {
1597 	dout("prepare_read_seq %p\n", con);
1598 	con->in_base_pos = 0;
1599 	con->in_tag = CEPH_MSGR_TAG_SEQ;
1600 }
1601 
1602 static void prepare_read_tag(struct ceph_connection *con)
1603 {
1604 	dout("prepare_read_tag %p\n", con);
1605 	con->in_base_pos = 0;
1606 	con->in_tag = CEPH_MSGR_TAG_READY;
1607 }
1608 
1609 /*
1610  * Prepare to read a message.
1611  */
1612 static int prepare_read_message(struct ceph_connection *con)
1613 {
1614 	dout("prepare_read_message %p\n", con);
1615 	BUG_ON(con->in_msg != NULL);
1616 	con->in_base_pos = 0;
1617 	con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1618 	return 0;
1619 }
1620 
1621 
1622 static int read_partial(struct ceph_connection *con,
1623 			int end, int size, void *object)
1624 {
1625 	while (con->in_base_pos < end) {
1626 		int left = end - con->in_base_pos;
1627 		int have = size - left;
1628 		int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1629 		if (ret <= 0)
1630 			return ret;
1631 		con->in_base_pos += ret;
1632 	}
1633 	return 1;
1634 }
1635 
1636 
1637 /*
1638  * Read all or part of the connect-side handshake on a new connection
1639  */
1640 static int read_partial_banner(struct ceph_connection *con)
1641 {
1642 	int size;
1643 	int end;
1644 	int ret;
1645 
1646 	dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1647 
1648 	/* peer's banner */
1649 	size = strlen(CEPH_BANNER);
1650 	end = size;
1651 	ret = read_partial(con, end, size, con->in_banner);
1652 	if (ret <= 0)
1653 		goto out;
1654 
1655 	size = sizeof (con->actual_peer_addr);
1656 	end += size;
1657 	ret = read_partial(con, end, size, &con->actual_peer_addr);
1658 	if (ret <= 0)
1659 		goto out;
1660 
1661 	size = sizeof (con->peer_addr_for_me);
1662 	end += size;
1663 	ret = read_partial(con, end, size, &con->peer_addr_for_me);
1664 	if (ret <= 0)
1665 		goto out;
1666 
1667 out:
1668 	return ret;
1669 }
1670 
1671 static int read_partial_connect(struct ceph_connection *con)
1672 {
1673 	int size;
1674 	int end;
1675 	int ret;
1676 
1677 	dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1678 
1679 	size = sizeof (con->in_reply);
1680 	end = size;
1681 	ret = read_partial(con, end, size, &con->in_reply);
1682 	if (ret <= 0)
1683 		goto out;
1684 
1685 	size = le32_to_cpu(con->in_reply.authorizer_len);
1686 	end += size;
1687 	ret = read_partial(con, end, size, con->auth_reply_buf);
1688 	if (ret <= 0)
1689 		goto out;
1690 
1691 	dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1692 	     con, (int)con->in_reply.tag,
1693 	     le32_to_cpu(con->in_reply.connect_seq),
1694 	     le32_to_cpu(con->in_reply.global_seq));
1695 out:
1696 	return ret;
1697 
1698 }
1699 
1700 /*
1701  * Verify the hello banner looks okay.
1702  */
1703 static int verify_hello(struct ceph_connection *con)
1704 {
1705 	if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1706 		pr_err("connect to %s got bad banner\n",
1707 		       ceph_pr_addr(&con->peer_addr.in_addr));
1708 		con->error_msg = "protocol error, bad banner";
1709 		return -1;
1710 	}
1711 	return 0;
1712 }
1713 
1714 static bool addr_is_blank(struct sockaddr_storage *ss)
1715 {
1716 	switch (ss->ss_family) {
1717 	case AF_INET:
1718 		return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1719 	case AF_INET6:
1720 		return
1721 		     ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1722 		     ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1723 		     ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1724 		     ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1725 	}
1726 	return false;
1727 }
1728 
1729 static int addr_port(struct sockaddr_storage *ss)
1730 {
1731 	switch (ss->ss_family) {
1732 	case AF_INET:
1733 		return ntohs(((struct sockaddr_in *)ss)->sin_port);
1734 	case AF_INET6:
1735 		return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1736 	}
1737 	return 0;
1738 }
1739 
1740 static void addr_set_port(struct sockaddr_storage *ss, int p)
1741 {
1742 	switch (ss->ss_family) {
1743 	case AF_INET:
1744 		((struct sockaddr_in *)ss)->sin_port = htons(p);
1745 		break;
1746 	case AF_INET6:
1747 		((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1748 		break;
1749 	}
1750 }
1751 
1752 /*
1753  * Unlike other *_pton function semantics, zero indicates success.
1754  */
1755 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1756 		char delim, const char **ipend)
1757 {
1758 	struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1759 	struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1760 
1761 	memset(ss, 0, sizeof(*ss));
1762 
1763 	if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1764 		ss->ss_family = AF_INET;
1765 		return 0;
1766 	}
1767 
1768 	if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1769 		ss->ss_family = AF_INET6;
1770 		return 0;
1771 	}
1772 
1773 	return -EINVAL;
1774 }
1775 
1776 /*
1777  * Extract hostname string and resolve using kernel DNS facility.
1778  */
1779 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1780 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1781 		struct sockaddr_storage *ss, char delim, const char **ipend)
1782 {
1783 	const char *end, *delim_p;
1784 	char *colon_p, *ip_addr = NULL;
1785 	int ip_len, ret;
1786 
1787 	/*
1788 	 * The end of the hostname occurs immediately preceding the delimiter or
1789 	 * the port marker (':') where the delimiter takes precedence.
1790 	 */
1791 	delim_p = memchr(name, delim, namelen);
1792 	colon_p = memchr(name, ':', namelen);
1793 
1794 	if (delim_p && colon_p)
1795 		end = delim_p < colon_p ? delim_p : colon_p;
1796 	else if (!delim_p && colon_p)
1797 		end = colon_p;
1798 	else {
1799 		end = delim_p;
1800 		if (!end) /* case: hostname:/ */
1801 			end = name + namelen;
1802 	}
1803 
1804 	if (end <= name)
1805 		return -EINVAL;
1806 
1807 	/* do dns_resolve upcall */
1808 	ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1809 	if (ip_len > 0)
1810 		ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1811 	else
1812 		ret = -ESRCH;
1813 
1814 	kfree(ip_addr);
1815 
1816 	*ipend = end;
1817 
1818 	pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1819 			ret, ret ? "failed" : ceph_pr_addr(ss));
1820 
1821 	return ret;
1822 }
1823 #else
1824 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1825 		struct sockaddr_storage *ss, char delim, const char **ipend)
1826 {
1827 	return -EINVAL;
1828 }
1829 #endif
1830 
1831 /*
1832  * Parse a server name (IP or hostname). If a valid IP address is not found
1833  * then try to extract a hostname to resolve using userspace DNS upcall.
1834  */
1835 static int ceph_parse_server_name(const char *name, size_t namelen,
1836 			struct sockaddr_storage *ss, char delim, const char **ipend)
1837 {
1838 	int ret;
1839 
1840 	ret = ceph_pton(name, namelen, ss, delim, ipend);
1841 	if (ret)
1842 		ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1843 
1844 	return ret;
1845 }
1846 
1847 /*
1848  * Parse an ip[:port] list into an addr array.  Use the default
1849  * monitor port if a port isn't specified.
1850  */
1851 int ceph_parse_ips(const char *c, const char *end,
1852 		   struct ceph_entity_addr *addr,
1853 		   int max_count, int *count)
1854 {
1855 	int i, ret = -EINVAL;
1856 	const char *p = c;
1857 
1858 	dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1859 	for (i = 0; i < max_count; i++) {
1860 		const char *ipend;
1861 		struct sockaddr_storage *ss = &addr[i].in_addr;
1862 		int port;
1863 		char delim = ',';
1864 
1865 		if (*p == '[') {
1866 			delim = ']';
1867 			p++;
1868 		}
1869 
1870 		ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1871 		if (ret)
1872 			goto bad;
1873 		ret = -EINVAL;
1874 
1875 		p = ipend;
1876 
1877 		if (delim == ']') {
1878 			if (*p != ']') {
1879 				dout("missing matching ']'\n");
1880 				goto bad;
1881 			}
1882 			p++;
1883 		}
1884 
1885 		/* port? */
1886 		if (p < end && *p == ':') {
1887 			port = 0;
1888 			p++;
1889 			while (p < end && *p >= '0' && *p <= '9') {
1890 				port = (port * 10) + (*p - '0');
1891 				p++;
1892 			}
1893 			if (port == 0)
1894 				port = CEPH_MON_PORT;
1895 			else if (port > 65535)
1896 				goto bad;
1897 		} else {
1898 			port = CEPH_MON_PORT;
1899 		}
1900 
1901 		addr_set_port(ss, port);
1902 
1903 		dout("parse_ips got %s\n", ceph_pr_addr(ss));
1904 
1905 		if (p == end)
1906 			break;
1907 		if (*p != ',')
1908 			goto bad;
1909 		p++;
1910 	}
1911 
1912 	if (p != end)
1913 		goto bad;
1914 
1915 	if (count)
1916 		*count = i + 1;
1917 	return 0;
1918 
1919 bad:
1920 	pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1921 	return ret;
1922 }
1923 EXPORT_SYMBOL(ceph_parse_ips);
1924 
1925 static int process_banner(struct ceph_connection *con)
1926 {
1927 	dout("process_banner on %p\n", con);
1928 
1929 	if (verify_hello(con) < 0)
1930 		return -1;
1931 
1932 	ceph_decode_addr(&con->actual_peer_addr);
1933 	ceph_decode_addr(&con->peer_addr_for_me);
1934 
1935 	/*
1936 	 * Make sure the other end is who we wanted.  note that the other
1937 	 * end may not yet know their ip address, so if it's 0.0.0.0, give
1938 	 * them the benefit of the doubt.
1939 	 */
1940 	if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1941 		   sizeof(con->peer_addr)) != 0 &&
1942 	    !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1943 	      con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1944 		pr_warn("wrong peer, want %s/%d, got %s/%d\n",
1945 			ceph_pr_addr(&con->peer_addr.in_addr),
1946 			(int)le32_to_cpu(con->peer_addr.nonce),
1947 			ceph_pr_addr(&con->actual_peer_addr.in_addr),
1948 			(int)le32_to_cpu(con->actual_peer_addr.nonce));
1949 		con->error_msg = "wrong peer at address";
1950 		return -1;
1951 	}
1952 
1953 	/*
1954 	 * did we learn our address?
1955 	 */
1956 	if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1957 		int port = addr_port(&con->msgr->inst.addr.in_addr);
1958 
1959 		memcpy(&con->msgr->inst.addr.in_addr,
1960 		       &con->peer_addr_for_me.in_addr,
1961 		       sizeof(con->peer_addr_for_me.in_addr));
1962 		addr_set_port(&con->msgr->inst.addr.in_addr, port);
1963 		encode_my_addr(con->msgr);
1964 		dout("process_banner learned my addr is %s\n",
1965 		     ceph_pr_addr(&con->msgr->inst.addr.in_addr));
1966 	}
1967 
1968 	return 0;
1969 }
1970 
1971 static int process_connect(struct ceph_connection *con)
1972 {
1973 	u64 sup_feat = con->msgr->supported_features;
1974 	u64 req_feat = con->msgr->required_features;
1975 	u64 server_feat = ceph_sanitize_features(
1976 				le64_to_cpu(con->in_reply.features));
1977 	int ret;
1978 
1979 	dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1980 
1981 	switch (con->in_reply.tag) {
1982 	case CEPH_MSGR_TAG_FEATURES:
1983 		pr_err("%s%lld %s feature set mismatch,"
1984 		       " my %llx < server's %llx, missing %llx\n",
1985 		       ENTITY_NAME(con->peer_name),
1986 		       ceph_pr_addr(&con->peer_addr.in_addr),
1987 		       sup_feat, server_feat, server_feat & ~sup_feat);
1988 		con->error_msg = "missing required protocol features";
1989 		reset_connection(con);
1990 		return -1;
1991 
1992 	case CEPH_MSGR_TAG_BADPROTOVER:
1993 		pr_err("%s%lld %s protocol version mismatch,"
1994 		       " my %d != server's %d\n",
1995 		       ENTITY_NAME(con->peer_name),
1996 		       ceph_pr_addr(&con->peer_addr.in_addr),
1997 		       le32_to_cpu(con->out_connect.protocol_version),
1998 		       le32_to_cpu(con->in_reply.protocol_version));
1999 		con->error_msg = "protocol version mismatch";
2000 		reset_connection(con);
2001 		return -1;
2002 
2003 	case CEPH_MSGR_TAG_BADAUTHORIZER:
2004 		con->auth_retry++;
2005 		dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2006 		     con->auth_retry);
2007 		if (con->auth_retry == 2) {
2008 			con->error_msg = "connect authorization failure";
2009 			return -1;
2010 		}
2011 		con_out_kvec_reset(con);
2012 		ret = prepare_write_connect(con);
2013 		if (ret < 0)
2014 			return ret;
2015 		prepare_read_connect(con);
2016 		break;
2017 
2018 	case CEPH_MSGR_TAG_RESETSESSION:
2019 		/*
2020 		 * If we connected with a large connect_seq but the peer
2021 		 * has no record of a session with us (no connection, or
2022 		 * connect_seq == 0), they will send RESETSESION to indicate
2023 		 * that they must have reset their session, and may have
2024 		 * dropped messages.
2025 		 */
2026 		dout("process_connect got RESET peer seq %u\n",
2027 		     le32_to_cpu(con->in_reply.connect_seq));
2028 		pr_err("%s%lld %s connection reset\n",
2029 		       ENTITY_NAME(con->peer_name),
2030 		       ceph_pr_addr(&con->peer_addr.in_addr));
2031 		reset_connection(con);
2032 		con_out_kvec_reset(con);
2033 		ret = prepare_write_connect(con);
2034 		if (ret < 0)
2035 			return ret;
2036 		prepare_read_connect(con);
2037 
2038 		/* Tell ceph about it. */
2039 		mutex_unlock(&con->mutex);
2040 		pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2041 		if (con->ops->peer_reset)
2042 			con->ops->peer_reset(con);
2043 		mutex_lock(&con->mutex);
2044 		if (con->state != CON_STATE_NEGOTIATING)
2045 			return -EAGAIN;
2046 		break;
2047 
2048 	case CEPH_MSGR_TAG_RETRY_SESSION:
2049 		/*
2050 		 * If we sent a smaller connect_seq than the peer has, try
2051 		 * again with a larger value.
2052 		 */
2053 		dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2054 		     le32_to_cpu(con->out_connect.connect_seq),
2055 		     le32_to_cpu(con->in_reply.connect_seq));
2056 		con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2057 		con_out_kvec_reset(con);
2058 		ret = prepare_write_connect(con);
2059 		if (ret < 0)
2060 			return ret;
2061 		prepare_read_connect(con);
2062 		break;
2063 
2064 	case CEPH_MSGR_TAG_RETRY_GLOBAL:
2065 		/*
2066 		 * If we sent a smaller global_seq than the peer has, try
2067 		 * again with a larger value.
2068 		 */
2069 		dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2070 		     con->peer_global_seq,
2071 		     le32_to_cpu(con->in_reply.global_seq));
2072 		get_global_seq(con->msgr,
2073 			       le32_to_cpu(con->in_reply.global_seq));
2074 		con_out_kvec_reset(con);
2075 		ret = prepare_write_connect(con);
2076 		if (ret < 0)
2077 			return ret;
2078 		prepare_read_connect(con);
2079 		break;
2080 
2081 	case CEPH_MSGR_TAG_SEQ:
2082 	case CEPH_MSGR_TAG_READY:
2083 		if (req_feat & ~server_feat) {
2084 			pr_err("%s%lld %s protocol feature mismatch,"
2085 			       " my required %llx > server's %llx, need %llx\n",
2086 			       ENTITY_NAME(con->peer_name),
2087 			       ceph_pr_addr(&con->peer_addr.in_addr),
2088 			       req_feat, server_feat, req_feat & ~server_feat);
2089 			con->error_msg = "missing required protocol features";
2090 			reset_connection(con);
2091 			return -1;
2092 		}
2093 
2094 		WARN_ON(con->state != CON_STATE_NEGOTIATING);
2095 		con->state = CON_STATE_OPEN;
2096 		con->auth_retry = 0;    /* we authenticated; clear flag */
2097 		con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2098 		con->connect_seq++;
2099 		con->peer_features = server_feat;
2100 		dout("process_connect got READY gseq %d cseq %d (%d)\n",
2101 		     con->peer_global_seq,
2102 		     le32_to_cpu(con->in_reply.connect_seq),
2103 		     con->connect_seq);
2104 		WARN_ON(con->connect_seq !=
2105 			le32_to_cpu(con->in_reply.connect_seq));
2106 
2107 		if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2108 			con_flag_set(con, CON_FLAG_LOSSYTX);
2109 
2110 		con->delay = 0;      /* reset backoff memory */
2111 
2112 		if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2113 			prepare_write_seq(con);
2114 			prepare_read_seq(con);
2115 		} else {
2116 			prepare_read_tag(con);
2117 		}
2118 		break;
2119 
2120 	case CEPH_MSGR_TAG_WAIT:
2121 		/*
2122 		 * If there is a connection race (we are opening
2123 		 * connections to each other), one of us may just have
2124 		 * to WAIT.  This shouldn't happen if we are the
2125 		 * client.
2126 		 */
2127 		pr_err("process_connect got WAIT as client\n");
2128 		con->error_msg = "protocol error, got WAIT as client";
2129 		return -1;
2130 
2131 	default:
2132 		pr_err("connect protocol error, will retry\n");
2133 		con->error_msg = "protocol error, garbage tag during connect";
2134 		return -1;
2135 	}
2136 	return 0;
2137 }
2138 
2139 
2140 /*
2141  * read (part of) an ack
2142  */
2143 static int read_partial_ack(struct ceph_connection *con)
2144 {
2145 	int size = sizeof (con->in_temp_ack);
2146 	int end = size;
2147 
2148 	return read_partial(con, end, size, &con->in_temp_ack);
2149 }
2150 
2151 /*
2152  * We can finally discard anything that's been acked.
2153  */
2154 static void process_ack(struct ceph_connection *con)
2155 {
2156 	struct ceph_msg *m;
2157 	u64 ack = le64_to_cpu(con->in_temp_ack);
2158 	u64 seq;
2159 
2160 	while (!list_empty(&con->out_sent)) {
2161 		m = list_first_entry(&con->out_sent, struct ceph_msg,
2162 				     list_head);
2163 		seq = le64_to_cpu(m->hdr.seq);
2164 		if (seq > ack)
2165 			break;
2166 		dout("got ack for seq %llu type %d at %p\n", seq,
2167 		     le16_to_cpu(m->hdr.type), m);
2168 		m->ack_stamp = jiffies;
2169 		ceph_msg_remove(m);
2170 	}
2171 	prepare_read_tag(con);
2172 }
2173 
2174 
2175 static int read_partial_message_section(struct ceph_connection *con,
2176 					struct kvec *section,
2177 					unsigned int sec_len, u32 *crc)
2178 {
2179 	int ret, left;
2180 
2181 	BUG_ON(!section);
2182 
2183 	while (section->iov_len < sec_len) {
2184 		BUG_ON(section->iov_base == NULL);
2185 		left = sec_len - section->iov_len;
2186 		ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2187 				       section->iov_len, left);
2188 		if (ret <= 0)
2189 			return ret;
2190 		section->iov_len += ret;
2191 	}
2192 	if (section->iov_len == sec_len)
2193 		*crc = crc32c(0, section->iov_base, section->iov_len);
2194 
2195 	return 1;
2196 }
2197 
2198 static int read_partial_msg_data(struct ceph_connection *con)
2199 {
2200 	struct ceph_msg *msg = con->in_msg;
2201 	struct ceph_msg_data_cursor *cursor = &msg->cursor;
2202 	const bool do_datacrc = !con->msgr->nocrc;
2203 	struct page *page;
2204 	size_t page_offset;
2205 	size_t length;
2206 	u32 crc = 0;
2207 	int ret;
2208 
2209 	BUG_ON(!msg);
2210 	if (list_empty(&msg->data))
2211 		return -EIO;
2212 
2213 	if (do_datacrc)
2214 		crc = con->in_data_crc;
2215 	while (cursor->resid) {
2216 		page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
2217 							NULL);
2218 		ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2219 		if (ret <= 0) {
2220 			if (do_datacrc)
2221 				con->in_data_crc = crc;
2222 
2223 			return ret;
2224 		}
2225 
2226 		if (do_datacrc)
2227 			crc = ceph_crc32c_page(crc, page, page_offset, ret);
2228 		(void) ceph_msg_data_advance(&msg->cursor, (size_t)ret);
2229 	}
2230 	if (do_datacrc)
2231 		con->in_data_crc = crc;
2232 
2233 	return 1;	/* must return > 0 to indicate success */
2234 }
2235 
2236 /*
2237  * read (part of) a message.
2238  */
2239 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2240 
2241 static int read_partial_message(struct ceph_connection *con)
2242 {
2243 	struct ceph_msg *m = con->in_msg;
2244 	int size;
2245 	int end;
2246 	int ret;
2247 	unsigned int front_len, middle_len, data_len;
2248 	bool do_datacrc = !con->msgr->nocrc;
2249 	u64 seq;
2250 	u32 crc;
2251 
2252 	dout("read_partial_message con %p msg %p\n", con, m);
2253 
2254 	/* header */
2255 	size = sizeof (con->in_hdr);
2256 	end = size;
2257 	ret = read_partial(con, end, size, &con->in_hdr);
2258 	if (ret <= 0)
2259 		return ret;
2260 
2261 	crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2262 	if (cpu_to_le32(crc) != con->in_hdr.crc) {
2263 		pr_err("read_partial_message bad hdr "
2264 		       " crc %u != expected %u\n",
2265 		       crc, con->in_hdr.crc);
2266 		return -EBADMSG;
2267 	}
2268 
2269 	front_len = le32_to_cpu(con->in_hdr.front_len);
2270 	if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2271 		return -EIO;
2272 	middle_len = le32_to_cpu(con->in_hdr.middle_len);
2273 	if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2274 		return -EIO;
2275 	data_len = le32_to_cpu(con->in_hdr.data_len);
2276 	if (data_len > CEPH_MSG_MAX_DATA_LEN)
2277 		return -EIO;
2278 
2279 	/* verify seq# */
2280 	seq = le64_to_cpu(con->in_hdr.seq);
2281 	if ((s64)seq - (s64)con->in_seq < 1) {
2282 		pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2283 			ENTITY_NAME(con->peer_name),
2284 			ceph_pr_addr(&con->peer_addr.in_addr),
2285 			seq, con->in_seq + 1);
2286 		con->in_base_pos = -front_len - middle_len - data_len -
2287 			sizeof(m->footer);
2288 		con->in_tag = CEPH_MSGR_TAG_READY;
2289 		return 0;
2290 	} else if ((s64)seq - (s64)con->in_seq > 1) {
2291 		pr_err("read_partial_message bad seq %lld expected %lld\n",
2292 		       seq, con->in_seq + 1);
2293 		con->error_msg = "bad message sequence # for incoming message";
2294 		return -EBADMSG;
2295 	}
2296 
2297 	/* allocate message? */
2298 	if (!con->in_msg) {
2299 		int skip = 0;
2300 
2301 		dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2302 		     front_len, data_len);
2303 		ret = ceph_con_in_msg_alloc(con, &skip);
2304 		if (ret < 0)
2305 			return ret;
2306 
2307 		BUG_ON(!con->in_msg ^ skip);
2308 		if (con->in_msg && data_len > con->in_msg->data_length) {
2309 			pr_warn("%s skipping long message (%u > %zd)\n",
2310 				__func__, data_len, con->in_msg->data_length);
2311 			ceph_msg_put(con->in_msg);
2312 			con->in_msg = NULL;
2313 			skip = 1;
2314 		}
2315 		if (skip) {
2316 			/* skip this message */
2317 			dout("alloc_msg said skip message\n");
2318 			con->in_base_pos = -front_len - middle_len - data_len -
2319 				sizeof(m->footer);
2320 			con->in_tag = CEPH_MSGR_TAG_READY;
2321 			con->in_seq++;
2322 			return 0;
2323 		}
2324 
2325 		BUG_ON(!con->in_msg);
2326 		BUG_ON(con->in_msg->con != con);
2327 		m = con->in_msg;
2328 		m->front.iov_len = 0;    /* haven't read it yet */
2329 		if (m->middle)
2330 			m->middle->vec.iov_len = 0;
2331 
2332 		/* prepare for data payload, if any */
2333 
2334 		if (data_len)
2335 			prepare_message_data(con->in_msg, data_len);
2336 	}
2337 
2338 	/* front */
2339 	ret = read_partial_message_section(con, &m->front, front_len,
2340 					   &con->in_front_crc);
2341 	if (ret <= 0)
2342 		return ret;
2343 
2344 	/* middle */
2345 	if (m->middle) {
2346 		ret = read_partial_message_section(con, &m->middle->vec,
2347 						   middle_len,
2348 						   &con->in_middle_crc);
2349 		if (ret <= 0)
2350 			return ret;
2351 	}
2352 
2353 	/* (page) data */
2354 	if (data_len) {
2355 		ret = read_partial_msg_data(con);
2356 		if (ret <= 0)
2357 			return ret;
2358 	}
2359 
2360 	/* footer */
2361 	size = sizeof (m->footer);
2362 	end += size;
2363 	ret = read_partial(con, end, size, &m->footer);
2364 	if (ret <= 0)
2365 		return ret;
2366 
2367 	dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2368 	     m, front_len, m->footer.front_crc, middle_len,
2369 	     m->footer.middle_crc, data_len, m->footer.data_crc);
2370 
2371 	/* crc ok? */
2372 	if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2373 		pr_err("read_partial_message %p front crc %u != exp. %u\n",
2374 		       m, con->in_front_crc, m->footer.front_crc);
2375 		return -EBADMSG;
2376 	}
2377 	if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2378 		pr_err("read_partial_message %p middle crc %u != exp %u\n",
2379 		       m, con->in_middle_crc, m->footer.middle_crc);
2380 		return -EBADMSG;
2381 	}
2382 	if (do_datacrc &&
2383 	    (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2384 	    con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2385 		pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2386 		       con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2387 		return -EBADMSG;
2388 	}
2389 
2390 	return 1; /* done! */
2391 }
2392 
2393 /*
2394  * Process message.  This happens in the worker thread.  The callback should
2395  * be careful not to do anything that waits on other incoming messages or it
2396  * may deadlock.
2397  */
2398 static void process_message(struct ceph_connection *con)
2399 {
2400 	struct ceph_msg *msg;
2401 
2402 	BUG_ON(con->in_msg->con != con);
2403 	con->in_msg->con = NULL;
2404 	msg = con->in_msg;
2405 	con->in_msg = NULL;
2406 	con->ops->put(con);
2407 
2408 	/* if first message, set peer_name */
2409 	if (con->peer_name.type == 0)
2410 		con->peer_name = msg->hdr.src;
2411 
2412 	con->in_seq++;
2413 	mutex_unlock(&con->mutex);
2414 
2415 	dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2416 	     msg, le64_to_cpu(msg->hdr.seq),
2417 	     ENTITY_NAME(msg->hdr.src),
2418 	     le16_to_cpu(msg->hdr.type),
2419 	     ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2420 	     le32_to_cpu(msg->hdr.front_len),
2421 	     le32_to_cpu(msg->hdr.data_len),
2422 	     con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2423 	con->ops->dispatch(con, msg);
2424 
2425 	mutex_lock(&con->mutex);
2426 }
2427 
2428 
2429 /*
2430  * Write something to the socket.  Called in a worker thread when the
2431  * socket appears to be writeable and we have something ready to send.
2432  */
2433 static int try_write(struct ceph_connection *con)
2434 {
2435 	int ret = 1;
2436 
2437 	dout("try_write start %p state %lu\n", con, con->state);
2438 
2439 more:
2440 	dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2441 
2442 	/* open the socket first? */
2443 	if (con->state == CON_STATE_PREOPEN) {
2444 		BUG_ON(con->sock);
2445 		con->state = CON_STATE_CONNECTING;
2446 
2447 		con_out_kvec_reset(con);
2448 		prepare_write_banner(con);
2449 		prepare_read_banner(con);
2450 
2451 		BUG_ON(con->in_msg);
2452 		con->in_tag = CEPH_MSGR_TAG_READY;
2453 		dout("try_write initiating connect on %p new state %lu\n",
2454 		     con, con->state);
2455 		ret = ceph_tcp_connect(con);
2456 		if (ret < 0) {
2457 			con->error_msg = "connect error";
2458 			goto out;
2459 		}
2460 	}
2461 
2462 more_kvec:
2463 	/* kvec data queued? */
2464 	if (con->out_skip) {
2465 		ret = write_partial_skip(con);
2466 		if (ret <= 0)
2467 			goto out;
2468 	}
2469 	if (con->out_kvec_left) {
2470 		ret = write_partial_kvec(con);
2471 		if (ret <= 0)
2472 			goto out;
2473 	}
2474 
2475 	/* msg pages? */
2476 	if (con->out_msg) {
2477 		if (con->out_msg_done) {
2478 			ceph_msg_put(con->out_msg);
2479 			con->out_msg = NULL;   /* we're done with this one */
2480 			goto do_next;
2481 		}
2482 
2483 		ret = write_partial_message_data(con);
2484 		if (ret == 1)
2485 			goto more_kvec;  /* we need to send the footer, too! */
2486 		if (ret == 0)
2487 			goto out;
2488 		if (ret < 0) {
2489 			dout("try_write write_partial_message_data err %d\n",
2490 			     ret);
2491 			goto out;
2492 		}
2493 	}
2494 
2495 do_next:
2496 	if (con->state == CON_STATE_OPEN) {
2497 		/* is anything else pending? */
2498 		if (!list_empty(&con->out_queue)) {
2499 			prepare_write_message(con);
2500 			goto more;
2501 		}
2502 		if (con->in_seq > con->in_seq_acked) {
2503 			prepare_write_ack(con);
2504 			goto more;
2505 		}
2506 		if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2507 			prepare_write_keepalive(con);
2508 			goto more;
2509 		}
2510 	}
2511 
2512 	/* Nothing to do! */
2513 	con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2514 	dout("try_write nothing else to write.\n");
2515 	ret = 0;
2516 out:
2517 	dout("try_write done on %p ret %d\n", con, ret);
2518 	return ret;
2519 }
2520 
2521 
2522 
2523 /*
2524  * Read what we can from the socket.
2525  */
2526 static int try_read(struct ceph_connection *con)
2527 {
2528 	int ret = -1;
2529 
2530 more:
2531 	dout("try_read start on %p state %lu\n", con, con->state);
2532 	if (con->state != CON_STATE_CONNECTING &&
2533 	    con->state != CON_STATE_NEGOTIATING &&
2534 	    con->state != CON_STATE_OPEN)
2535 		return 0;
2536 
2537 	BUG_ON(!con->sock);
2538 
2539 	dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2540 	     con->in_base_pos);
2541 
2542 	if (con->state == CON_STATE_CONNECTING) {
2543 		dout("try_read connecting\n");
2544 		ret = read_partial_banner(con);
2545 		if (ret <= 0)
2546 			goto out;
2547 		ret = process_banner(con);
2548 		if (ret < 0)
2549 			goto out;
2550 
2551 		con->state = CON_STATE_NEGOTIATING;
2552 
2553 		/*
2554 		 * Received banner is good, exchange connection info.
2555 		 * Do not reset out_kvec, as sending our banner raced
2556 		 * with receiving peer banner after connect completed.
2557 		 */
2558 		ret = prepare_write_connect(con);
2559 		if (ret < 0)
2560 			goto out;
2561 		prepare_read_connect(con);
2562 
2563 		/* Send connection info before awaiting response */
2564 		goto out;
2565 	}
2566 
2567 	if (con->state == CON_STATE_NEGOTIATING) {
2568 		dout("try_read negotiating\n");
2569 		ret = read_partial_connect(con);
2570 		if (ret <= 0)
2571 			goto out;
2572 		ret = process_connect(con);
2573 		if (ret < 0)
2574 			goto out;
2575 		goto more;
2576 	}
2577 
2578 	WARN_ON(con->state != CON_STATE_OPEN);
2579 
2580 	if (con->in_base_pos < 0) {
2581 		/*
2582 		 * skipping + discarding content.
2583 		 *
2584 		 * FIXME: there must be a better way to do this!
2585 		 */
2586 		static char buf[SKIP_BUF_SIZE];
2587 		int skip = min((int) sizeof (buf), -con->in_base_pos);
2588 
2589 		dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2590 		ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2591 		if (ret <= 0)
2592 			goto out;
2593 		con->in_base_pos += ret;
2594 		if (con->in_base_pos)
2595 			goto more;
2596 	}
2597 	if (con->in_tag == CEPH_MSGR_TAG_READY) {
2598 		/*
2599 		 * what's next?
2600 		 */
2601 		ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2602 		if (ret <= 0)
2603 			goto out;
2604 		dout("try_read got tag %d\n", (int)con->in_tag);
2605 		switch (con->in_tag) {
2606 		case CEPH_MSGR_TAG_MSG:
2607 			prepare_read_message(con);
2608 			break;
2609 		case CEPH_MSGR_TAG_ACK:
2610 			prepare_read_ack(con);
2611 			break;
2612 		case CEPH_MSGR_TAG_CLOSE:
2613 			con_close_socket(con);
2614 			con->state = CON_STATE_CLOSED;
2615 			goto out;
2616 		default:
2617 			goto bad_tag;
2618 		}
2619 	}
2620 	if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2621 		ret = read_partial_message(con);
2622 		if (ret <= 0) {
2623 			switch (ret) {
2624 			case -EBADMSG:
2625 				con->error_msg = "bad crc";
2626 				ret = -EIO;
2627 				break;
2628 			case -EIO:
2629 				con->error_msg = "io error";
2630 				break;
2631 			}
2632 			goto out;
2633 		}
2634 		if (con->in_tag == CEPH_MSGR_TAG_READY)
2635 			goto more;
2636 		process_message(con);
2637 		if (con->state == CON_STATE_OPEN)
2638 			prepare_read_tag(con);
2639 		goto more;
2640 	}
2641 	if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2642 	    con->in_tag == CEPH_MSGR_TAG_SEQ) {
2643 		/*
2644 		 * the final handshake seq exchange is semantically
2645 		 * equivalent to an ACK
2646 		 */
2647 		ret = read_partial_ack(con);
2648 		if (ret <= 0)
2649 			goto out;
2650 		process_ack(con);
2651 		goto more;
2652 	}
2653 
2654 out:
2655 	dout("try_read done on %p ret %d\n", con, ret);
2656 	return ret;
2657 
2658 bad_tag:
2659 	pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2660 	con->error_msg = "protocol error, garbage tag";
2661 	ret = -1;
2662 	goto out;
2663 }
2664 
2665 
2666 /*
2667  * Atomically queue work on a connection after the specified delay.
2668  * Bump @con reference to avoid races with connection teardown.
2669  * Returns 0 if work was queued, or an error code otherwise.
2670  */
2671 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2672 {
2673 	if (!con->ops->get(con)) {
2674 		dout("%s %p ref count 0\n", __func__, con);
2675 		return -ENOENT;
2676 	}
2677 
2678 	if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2679 		dout("%s %p - already queued\n", __func__, con);
2680 		con->ops->put(con);
2681 		return -EBUSY;
2682 	}
2683 
2684 	dout("%s %p %lu\n", __func__, con, delay);
2685 	return 0;
2686 }
2687 
2688 static void queue_con(struct ceph_connection *con)
2689 {
2690 	(void) queue_con_delay(con, 0);
2691 }
2692 
2693 static void cancel_con(struct ceph_connection *con)
2694 {
2695 	if (cancel_delayed_work(&con->work)) {
2696 		dout("%s %p\n", __func__, con);
2697 		con->ops->put(con);
2698 	}
2699 }
2700 
2701 static bool con_sock_closed(struct ceph_connection *con)
2702 {
2703 	if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2704 		return false;
2705 
2706 #define CASE(x)								\
2707 	case CON_STATE_ ## x:						\
2708 		con->error_msg = "socket closed (con state " #x ")";	\
2709 		break;
2710 
2711 	switch (con->state) {
2712 	CASE(CLOSED);
2713 	CASE(PREOPEN);
2714 	CASE(CONNECTING);
2715 	CASE(NEGOTIATING);
2716 	CASE(OPEN);
2717 	CASE(STANDBY);
2718 	default:
2719 		pr_warn("%s con %p unrecognized state %lu\n",
2720 			__func__, con, con->state);
2721 		con->error_msg = "unrecognized con state";
2722 		BUG();
2723 		break;
2724 	}
2725 #undef CASE
2726 
2727 	return true;
2728 }
2729 
2730 static bool con_backoff(struct ceph_connection *con)
2731 {
2732 	int ret;
2733 
2734 	if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2735 		return false;
2736 
2737 	ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2738 	if (ret) {
2739 		dout("%s: con %p FAILED to back off %lu\n", __func__,
2740 			con, con->delay);
2741 		BUG_ON(ret == -ENOENT);
2742 		con_flag_set(con, CON_FLAG_BACKOFF);
2743 	}
2744 
2745 	return true;
2746 }
2747 
2748 /* Finish fault handling; con->mutex must *not* be held here */
2749 
2750 static void con_fault_finish(struct ceph_connection *con)
2751 {
2752 	/*
2753 	 * in case we faulted due to authentication, invalidate our
2754 	 * current tickets so that we can get new ones.
2755 	 */
2756 	if (con->auth_retry && con->ops->invalidate_authorizer) {
2757 		dout("calling invalidate_authorizer()\n");
2758 		con->ops->invalidate_authorizer(con);
2759 	}
2760 
2761 	if (con->ops->fault)
2762 		con->ops->fault(con);
2763 }
2764 
2765 /*
2766  * Do some work on a connection.  Drop a connection ref when we're done.
2767  */
2768 static void con_work(struct work_struct *work)
2769 {
2770 	struct ceph_connection *con = container_of(work, struct ceph_connection,
2771 						   work.work);
2772 	bool fault;
2773 
2774 	mutex_lock(&con->mutex);
2775 	while (true) {
2776 		int ret;
2777 
2778 		if ((fault = con_sock_closed(con))) {
2779 			dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2780 			break;
2781 		}
2782 		if (con_backoff(con)) {
2783 			dout("%s: con %p BACKOFF\n", __func__, con);
2784 			break;
2785 		}
2786 		if (con->state == CON_STATE_STANDBY) {
2787 			dout("%s: con %p STANDBY\n", __func__, con);
2788 			break;
2789 		}
2790 		if (con->state == CON_STATE_CLOSED) {
2791 			dout("%s: con %p CLOSED\n", __func__, con);
2792 			BUG_ON(con->sock);
2793 			break;
2794 		}
2795 		if (con->state == CON_STATE_PREOPEN) {
2796 			dout("%s: con %p PREOPEN\n", __func__, con);
2797 			BUG_ON(con->sock);
2798 		}
2799 
2800 		ret = try_read(con);
2801 		if (ret < 0) {
2802 			if (ret == -EAGAIN)
2803 				continue;
2804 			con->error_msg = "socket error on read";
2805 			fault = true;
2806 			break;
2807 		}
2808 
2809 		ret = try_write(con);
2810 		if (ret < 0) {
2811 			if (ret == -EAGAIN)
2812 				continue;
2813 			con->error_msg = "socket error on write";
2814 			fault = true;
2815 		}
2816 
2817 		break;	/* If we make it to here, we're done */
2818 	}
2819 	if (fault)
2820 		con_fault(con);
2821 	mutex_unlock(&con->mutex);
2822 
2823 	if (fault)
2824 		con_fault_finish(con);
2825 
2826 	con->ops->put(con);
2827 }
2828 
2829 /*
2830  * Generic error/fault handler.  A retry mechanism is used with
2831  * exponential backoff
2832  */
2833 static void con_fault(struct ceph_connection *con)
2834 {
2835 	pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2836 		ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2837 	dout("fault %p state %lu to peer %s\n",
2838 	     con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2839 
2840 	WARN_ON(con->state != CON_STATE_CONNECTING &&
2841 	       con->state != CON_STATE_NEGOTIATING &&
2842 	       con->state != CON_STATE_OPEN);
2843 
2844 	con_close_socket(con);
2845 
2846 	if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2847 		dout("fault on LOSSYTX channel, marking CLOSED\n");
2848 		con->state = CON_STATE_CLOSED;
2849 		return;
2850 	}
2851 
2852 	if (con->in_msg) {
2853 		BUG_ON(con->in_msg->con != con);
2854 		con->in_msg->con = NULL;
2855 		ceph_msg_put(con->in_msg);
2856 		con->in_msg = NULL;
2857 		con->ops->put(con);
2858 	}
2859 
2860 	/* Requeue anything that hasn't been acked */
2861 	list_splice_init(&con->out_sent, &con->out_queue);
2862 
2863 	/* If there are no messages queued or keepalive pending, place
2864 	 * the connection in a STANDBY state */
2865 	if (list_empty(&con->out_queue) &&
2866 	    !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
2867 		dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2868 		con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2869 		con->state = CON_STATE_STANDBY;
2870 	} else {
2871 		/* retry after a delay. */
2872 		con->state = CON_STATE_PREOPEN;
2873 		if (con->delay == 0)
2874 			con->delay = BASE_DELAY_INTERVAL;
2875 		else if (con->delay < MAX_DELAY_INTERVAL)
2876 			con->delay *= 2;
2877 		con_flag_set(con, CON_FLAG_BACKOFF);
2878 		queue_con(con);
2879 	}
2880 }
2881 
2882 
2883 
2884 /*
2885  * initialize a new messenger instance
2886  */
2887 void ceph_messenger_init(struct ceph_messenger *msgr,
2888 			struct ceph_entity_addr *myaddr,
2889 			u64 supported_features,
2890 			u64 required_features,
2891 			bool nocrc)
2892 {
2893 	msgr->supported_features = supported_features;
2894 	msgr->required_features = required_features;
2895 
2896 	spin_lock_init(&msgr->global_seq_lock);
2897 
2898 	if (myaddr)
2899 		msgr->inst.addr = *myaddr;
2900 
2901 	/* select a random nonce */
2902 	msgr->inst.addr.type = 0;
2903 	get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2904 	encode_my_addr(msgr);
2905 	msgr->nocrc = nocrc;
2906 
2907 	atomic_set(&msgr->stopping, 0);
2908 
2909 	dout("%s %p\n", __func__, msgr);
2910 }
2911 EXPORT_SYMBOL(ceph_messenger_init);
2912 
2913 static void clear_standby(struct ceph_connection *con)
2914 {
2915 	/* come back from STANDBY? */
2916 	if (con->state == CON_STATE_STANDBY) {
2917 		dout("clear_standby %p and ++connect_seq\n", con);
2918 		con->state = CON_STATE_PREOPEN;
2919 		con->connect_seq++;
2920 		WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2921 		WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
2922 	}
2923 }
2924 
2925 /*
2926  * Queue up an outgoing message on the given connection.
2927  */
2928 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2929 {
2930 	/* set src+dst */
2931 	msg->hdr.src = con->msgr->inst.name;
2932 	BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2933 	msg->needs_out_seq = true;
2934 
2935 	mutex_lock(&con->mutex);
2936 
2937 	if (con->state == CON_STATE_CLOSED) {
2938 		dout("con_send %p closed, dropping %p\n", con, msg);
2939 		ceph_msg_put(msg);
2940 		mutex_unlock(&con->mutex);
2941 		return;
2942 	}
2943 
2944 	BUG_ON(msg->con != NULL);
2945 	msg->con = con->ops->get(con);
2946 	BUG_ON(msg->con == NULL);
2947 
2948 	BUG_ON(!list_empty(&msg->list_head));
2949 	list_add_tail(&msg->list_head, &con->out_queue);
2950 	dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2951 	     ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2952 	     ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2953 	     le32_to_cpu(msg->hdr.front_len),
2954 	     le32_to_cpu(msg->hdr.middle_len),
2955 	     le32_to_cpu(msg->hdr.data_len));
2956 
2957 	clear_standby(con);
2958 	mutex_unlock(&con->mutex);
2959 
2960 	/* if there wasn't anything waiting to send before, queue
2961 	 * new work */
2962 	if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
2963 		queue_con(con);
2964 }
2965 EXPORT_SYMBOL(ceph_con_send);
2966 
2967 /*
2968  * Revoke a message that was previously queued for send
2969  */
2970 void ceph_msg_revoke(struct ceph_msg *msg)
2971 {
2972 	struct ceph_connection *con = msg->con;
2973 
2974 	if (!con)
2975 		return;		/* Message not in our possession */
2976 
2977 	mutex_lock(&con->mutex);
2978 	if (!list_empty(&msg->list_head)) {
2979 		dout("%s %p msg %p - was on queue\n", __func__, con, msg);
2980 		list_del_init(&msg->list_head);
2981 		BUG_ON(msg->con == NULL);
2982 		msg->con->ops->put(msg->con);
2983 		msg->con = NULL;
2984 		msg->hdr.seq = 0;
2985 
2986 		ceph_msg_put(msg);
2987 	}
2988 	if (con->out_msg == msg) {
2989 		dout("%s %p msg %p - was sending\n", __func__, con, msg);
2990 		con->out_msg = NULL;
2991 		if (con->out_kvec_is_msg) {
2992 			con->out_skip = con->out_kvec_bytes;
2993 			con->out_kvec_is_msg = false;
2994 		}
2995 		msg->hdr.seq = 0;
2996 
2997 		ceph_msg_put(msg);
2998 	}
2999 	mutex_unlock(&con->mutex);
3000 }
3001 
3002 /*
3003  * Revoke a message that we may be reading data into
3004  */
3005 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
3006 {
3007 	struct ceph_connection *con;
3008 
3009 	BUG_ON(msg == NULL);
3010 	if (!msg->con) {
3011 		dout("%s msg %p null con\n", __func__, msg);
3012 
3013 		return;		/* Message not in our possession */
3014 	}
3015 
3016 	con = msg->con;
3017 	mutex_lock(&con->mutex);
3018 	if (con->in_msg == msg) {
3019 		unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3020 		unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3021 		unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
3022 
3023 		/* skip rest of message */
3024 		dout("%s %p msg %p revoked\n", __func__, con, msg);
3025 		con->in_base_pos = con->in_base_pos -
3026 				sizeof(struct ceph_msg_header) -
3027 				front_len -
3028 				middle_len -
3029 				data_len -
3030 				sizeof(struct ceph_msg_footer);
3031 		ceph_msg_put(con->in_msg);
3032 		con->in_msg = NULL;
3033 		con->in_tag = CEPH_MSGR_TAG_READY;
3034 		con->in_seq++;
3035 	} else {
3036 		dout("%s %p in_msg %p msg %p no-op\n",
3037 		     __func__, con, con->in_msg, msg);
3038 	}
3039 	mutex_unlock(&con->mutex);
3040 }
3041 
3042 /*
3043  * Queue a keepalive byte to ensure the tcp connection is alive.
3044  */
3045 void ceph_con_keepalive(struct ceph_connection *con)
3046 {
3047 	dout("con_keepalive %p\n", con);
3048 	mutex_lock(&con->mutex);
3049 	clear_standby(con);
3050 	mutex_unlock(&con->mutex);
3051 	if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3052 	    con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3053 		queue_con(con);
3054 }
3055 EXPORT_SYMBOL(ceph_con_keepalive);
3056 
3057 static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
3058 {
3059 	struct ceph_msg_data *data;
3060 
3061 	if (WARN_ON(!ceph_msg_data_type_valid(type)))
3062 		return NULL;
3063 
3064 	data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
3065 	if (data)
3066 		data->type = type;
3067 	INIT_LIST_HEAD(&data->links);
3068 
3069 	return data;
3070 }
3071 
3072 static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3073 {
3074 	if (!data)
3075 		return;
3076 
3077 	WARN_ON(!list_empty(&data->links));
3078 	if (data->type == CEPH_MSG_DATA_PAGELIST)
3079 		ceph_pagelist_release(data->pagelist);
3080 	kmem_cache_free(ceph_msg_data_cache, data);
3081 }
3082 
3083 void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
3084 		size_t length, size_t alignment)
3085 {
3086 	struct ceph_msg_data *data;
3087 
3088 	BUG_ON(!pages);
3089 	BUG_ON(!length);
3090 
3091 	data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3092 	BUG_ON(!data);
3093 	data->pages = pages;
3094 	data->length = length;
3095 	data->alignment = alignment & ~PAGE_MASK;
3096 
3097 	list_add_tail(&data->links, &msg->data);
3098 	msg->data_length += length;
3099 }
3100 EXPORT_SYMBOL(ceph_msg_data_add_pages);
3101 
3102 void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
3103 				struct ceph_pagelist *pagelist)
3104 {
3105 	struct ceph_msg_data *data;
3106 
3107 	BUG_ON(!pagelist);
3108 	BUG_ON(!pagelist->length);
3109 
3110 	data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3111 	BUG_ON(!data);
3112 	data->pagelist = pagelist;
3113 
3114 	list_add_tail(&data->links, &msg->data);
3115 	msg->data_length += pagelist->length;
3116 }
3117 EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
3118 
3119 #ifdef	CONFIG_BLOCK
3120 void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
3121 		size_t length)
3122 {
3123 	struct ceph_msg_data *data;
3124 
3125 	BUG_ON(!bio);
3126 
3127 	data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3128 	BUG_ON(!data);
3129 	data->bio = bio;
3130 	data->bio_length = length;
3131 
3132 	list_add_tail(&data->links, &msg->data);
3133 	msg->data_length += length;
3134 }
3135 EXPORT_SYMBOL(ceph_msg_data_add_bio);
3136 #endif	/* CONFIG_BLOCK */
3137 
3138 /*
3139  * construct a new message with given type, size
3140  * the new msg has a ref count of 1.
3141  */
3142 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3143 			      bool can_fail)
3144 {
3145 	struct ceph_msg *m;
3146 
3147 	m = kmem_cache_zalloc(ceph_msg_cache, flags);
3148 	if (m == NULL)
3149 		goto out;
3150 
3151 	m->hdr.type = cpu_to_le16(type);
3152 	m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3153 	m->hdr.front_len = cpu_to_le32(front_len);
3154 
3155 	INIT_LIST_HEAD(&m->list_head);
3156 	kref_init(&m->kref);
3157 	INIT_LIST_HEAD(&m->data);
3158 
3159 	/* front */
3160 	if (front_len) {
3161 		m->front.iov_base = ceph_kvmalloc(front_len, flags);
3162 		if (m->front.iov_base == NULL) {
3163 			dout("ceph_msg_new can't allocate %d bytes\n",
3164 			     front_len);
3165 			goto out2;
3166 		}
3167 	} else {
3168 		m->front.iov_base = NULL;
3169 	}
3170 	m->front_alloc_len = m->front.iov_len = front_len;
3171 
3172 	dout("ceph_msg_new %p front %d\n", m, front_len);
3173 	return m;
3174 
3175 out2:
3176 	ceph_msg_put(m);
3177 out:
3178 	if (!can_fail) {
3179 		pr_err("msg_new can't create type %d front %d\n", type,
3180 		       front_len);
3181 		WARN_ON(1);
3182 	} else {
3183 		dout("msg_new can't create type %d front %d\n", type,
3184 		     front_len);
3185 	}
3186 	return NULL;
3187 }
3188 EXPORT_SYMBOL(ceph_msg_new);
3189 
3190 /*
3191  * Allocate "middle" portion of a message, if it is needed and wasn't
3192  * allocated by alloc_msg.  This allows us to read a small fixed-size
3193  * per-type header in the front and then gracefully fail (i.e.,
3194  * propagate the error to the caller based on info in the front) when
3195  * the middle is too large.
3196  */
3197 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3198 {
3199 	int type = le16_to_cpu(msg->hdr.type);
3200 	int middle_len = le32_to_cpu(msg->hdr.middle_len);
3201 
3202 	dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3203 	     ceph_msg_type_name(type), middle_len);
3204 	BUG_ON(!middle_len);
3205 	BUG_ON(msg->middle);
3206 
3207 	msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3208 	if (!msg->middle)
3209 		return -ENOMEM;
3210 	return 0;
3211 }
3212 
3213 /*
3214  * Allocate a message for receiving an incoming message on a
3215  * connection, and save the result in con->in_msg.  Uses the
3216  * connection's private alloc_msg op if available.
3217  *
3218  * Returns 0 on success, or a negative error code.
3219  *
3220  * On success, if we set *skip = 1:
3221  *  - the next message should be skipped and ignored.
3222  *  - con->in_msg == NULL
3223  * or if we set *skip = 0:
3224  *  - con->in_msg is non-null.
3225  * On error (ENOMEM, EAGAIN, ...),
3226  *  - con->in_msg == NULL
3227  */
3228 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3229 {
3230 	struct ceph_msg_header *hdr = &con->in_hdr;
3231 	int middle_len = le32_to_cpu(hdr->middle_len);
3232 	struct ceph_msg *msg;
3233 	int ret = 0;
3234 
3235 	BUG_ON(con->in_msg != NULL);
3236 	BUG_ON(!con->ops->alloc_msg);
3237 
3238 	mutex_unlock(&con->mutex);
3239 	msg = con->ops->alloc_msg(con, hdr, skip);
3240 	mutex_lock(&con->mutex);
3241 	if (con->state != CON_STATE_OPEN) {
3242 		if (msg)
3243 			ceph_msg_put(msg);
3244 		return -EAGAIN;
3245 	}
3246 	if (msg) {
3247 		BUG_ON(*skip);
3248 		con->in_msg = msg;
3249 		con->in_msg->con = con->ops->get(con);
3250 		BUG_ON(con->in_msg->con == NULL);
3251 	} else {
3252 		/*
3253 		 * Null message pointer means either we should skip
3254 		 * this message or we couldn't allocate memory.  The
3255 		 * former is not an error.
3256 		 */
3257 		if (*skip)
3258 			return 0;
3259 		con->error_msg = "error allocating memory for incoming message";
3260 
3261 		return -ENOMEM;
3262 	}
3263 	memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3264 
3265 	if (middle_len && !con->in_msg->middle) {
3266 		ret = ceph_alloc_middle(con, con->in_msg);
3267 		if (ret < 0) {
3268 			ceph_msg_put(con->in_msg);
3269 			con->in_msg = NULL;
3270 		}
3271 	}
3272 
3273 	return ret;
3274 }
3275 
3276 
3277 /*
3278  * Free a generically kmalloc'd message.
3279  */
3280 static void ceph_msg_free(struct ceph_msg *m)
3281 {
3282 	dout("%s %p\n", __func__, m);
3283 	ceph_kvfree(m->front.iov_base);
3284 	kmem_cache_free(ceph_msg_cache, m);
3285 }
3286 
3287 static void ceph_msg_release(struct kref *kref)
3288 {
3289 	struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3290 	LIST_HEAD(data);
3291 	struct list_head *links;
3292 	struct list_head *next;
3293 
3294 	dout("%s %p\n", __func__, m);
3295 	WARN_ON(!list_empty(&m->list_head));
3296 
3297 	/* drop middle, data, if any */
3298 	if (m->middle) {
3299 		ceph_buffer_put(m->middle);
3300 		m->middle = NULL;
3301 	}
3302 
3303 	list_splice_init(&m->data, &data);
3304 	list_for_each_safe(links, next, &data) {
3305 		struct ceph_msg_data *data;
3306 
3307 		data = list_entry(links, struct ceph_msg_data, links);
3308 		list_del_init(links);
3309 		ceph_msg_data_destroy(data);
3310 	}
3311 	m->data_length = 0;
3312 
3313 	if (m->pool)
3314 		ceph_msgpool_put(m->pool, m);
3315 	else
3316 		ceph_msg_free(m);
3317 }
3318 
3319 struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3320 {
3321 	dout("%s %p (was %d)\n", __func__, msg,
3322 	     atomic_read(&msg->kref.refcount));
3323 	kref_get(&msg->kref);
3324 	return msg;
3325 }
3326 EXPORT_SYMBOL(ceph_msg_get);
3327 
3328 void ceph_msg_put(struct ceph_msg *msg)
3329 {
3330 	dout("%s %p (was %d)\n", __func__, msg,
3331 	     atomic_read(&msg->kref.refcount));
3332 	kref_put(&msg->kref, ceph_msg_release);
3333 }
3334 EXPORT_SYMBOL(ceph_msg_put);
3335 
3336 void ceph_msg_dump(struct ceph_msg *msg)
3337 {
3338 	pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3339 		 msg->front_alloc_len, msg->data_length);
3340 	print_hex_dump(KERN_DEBUG, "header: ",
3341 		       DUMP_PREFIX_OFFSET, 16, 1,
3342 		       &msg->hdr, sizeof(msg->hdr), true);
3343 	print_hex_dump(KERN_DEBUG, " front: ",
3344 		       DUMP_PREFIX_OFFSET, 16, 1,
3345 		       msg->front.iov_base, msg->front.iov_len, true);
3346 	if (msg->middle)
3347 		print_hex_dump(KERN_DEBUG, "middle: ",
3348 			       DUMP_PREFIX_OFFSET, 16, 1,
3349 			       msg->middle->vec.iov_base,
3350 			       msg->middle->vec.iov_len, true);
3351 	print_hex_dump(KERN_DEBUG, "footer: ",
3352 		       DUMP_PREFIX_OFFSET, 16, 1,
3353 		       &msg->footer, sizeof(msg->footer), true);
3354 }
3355 EXPORT_SYMBOL(ceph_msg_dump);
3356