xref: /openbmc/linux/net/ceph/messenger.c (revision d2999e1b)
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 con_work(struct work_struct *);
178 static void con_fault(struct ceph_connection *con);
179 
180 /*
181  * Nicely render a sockaddr as a string.  An array of formatted
182  * strings is used, to approximate reentrancy.
183  */
184 #define ADDR_STR_COUNT_LOG	5	/* log2(# address strings in array) */
185 #define ADDR_STR_COUNT		(1 << ADDR_STR_COUNT_LOG)
186 #define ADDR_STR_COUNT_MASK	(ADDR_STR_COUNT - 1)
187 #define MAX_ADDR_STR_LEN	64	/* 54 is enough */
188 
189 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
190 static atomic_t addr_str_seq = ATOMIC_INIT(0);
191 
192 static struct page *zero_page;		/* used in certain error cases */
193 
194 const char *ceph_pr_addr(const struct sockaddr_storage *ss)
195 {
196 	int i;
197 	char *s;
198 	struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
199 	struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
200 
201 	i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
202 	s = addr_str[i];
203 
204 	switch (ss->ss_family) {
205 	case AF_INET:
206 		snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
207 			 ntohs(in4->sin_port));
208 		break;
209 
210 	case AF_INET6:
211 		snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
212 			 ntohs(in6->sin6_port));
213 		break;
214 
215 	default:
216 		snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
217 			 ss->ss_family);
218 	}
219 
220 	return s;
221 }
222 EXPORT_SYMBOL(ceph_pr_addr);
223 
224 static void encode_my_addr(struct ceph_messenger *msgr)
225 {
226 	memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
227 	ceph_encode_addr(&msgr->my_enc_addr);
228 }
229 
230 /*
231  * work queue for all reading and writing to/from the socket.
232  */
233 static struct workqueue_struct *ceph_msgr_wq;
234 
235 static int ceph_msgr_slab_init(void)
236 {
237 	BUG_ON(ceph_msg_cache);
238 	ceph_msg_cache = kmem_cache_create("ceph_msg",
239 					sizeof (struct ceph_msg),
240 					__alignof__(struct ceph_msg), 0, NULL);
241 
242 	if (!ceph_msg_cache)
243 		return -ENOMEM;
244 
245 	BUG_ON(ceph_msg_data_cache);
246 	ceph_msg_data_cache = kmem_cache_create("ceph_msg_data",
247 					sizeof (struct ceph_msg_data),
248 					__alignof__(struct ceph_msg_data),
249 					0, NULL);
250 	if (ceph_msg_data_cache)
251 		return 0;
252 
253 	kmem_cache_destroy(ceph_msg_cache);
254 	ceph_msg_cache = NULL;
255 
256 	return -ENOMEM;
257 }
258 
259 static void ceph_msgr_slab_exit(void)
260 {
261 	BUG_ON(!ceph_msg_data_cache);
262 	kmem_cache_destroy(ceph_msg_data_cache);
263 	ceph_msg_data_cache = NULL;
264 
265 	BUG_ON(!ceph_msg_cache);
266 	kmem_cache_destroy(ceph_msg_cache);
267 	ceph_msg_cache = NULL;
268 }
269 
270 static void _ceph_msgr_exit(void)
271 {
272 	if (ceph_msgr_wq) {
273 		destroy_workqueue(ceph_msgr_wq);
274 		ceph_msgr_wq = NULL;
275 	}
276 
277 	ceph_msgr_slab_exit();
278 
279 	BUG_ON(zero_page == NULL);
280 	kunmap(zero_page);
281 	page_cache_release(zero_page);
282 	zero_page = NULL;
283 }
284 
285 int ceph_msgr_init(void)
286 {
287 	BUG_ON(zero_page != NULL);
288 	zero_page = ZERO_PAGE(0);
289 	page_cache_get(zero_page);
290 
291 	if (ceph_msgr_slab_init())
292 		return -ENOMEM;
293 
294 	ceph_msgr_wq = alloc_workqueue("ceph-msgr", 0, 0);
295 	if (ceph_msgr_wq)
296 		return 0;
297 
298 	pr_err("msgr_init failed to create workqueue\n");
299 	_ceph_msgr_exit();
300 
301 	return -ENOMEM;
302 }
303 EXPORT_SYMBOL(ceph_msgr_init);
304 
305 void ceph_msgr_exit(void)
306 {
307 	BUG_ON(ceph_msgr_wq == NULL);
308 
309 	_ceph_msgr_exit();
310 }
311 EXPORT_SYMBOL(ceph_msgr_exit);
312 
313 void ceph_msgr_flush(void)
314 {
315 	flush_workqueue(ceph_msgr_wq);
316 }
317 EXPORT_SYMBOL(ceph_msgr_flush);
318 
319 /* Connection socket state transition functions */
320 
321 static void con_sock_state_init(struct ceph_connection *con)
322 {
323 	int old_state;
324 
325 	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
326 	if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
327 		printk("%s: unexpected old state %d\n", __func__, old_state);
328 	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
329 	     CON_SOCK_STATE_CLOSED);
330 }
331 
332 static void con_sock_state_connecting(struct ceph_connection *con)
333 {
334 	int old_state;
335 
336 	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
337 	if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
338 		printk("%s: unexpected old state %d\n", __func__, old_state);
339 	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
340 	     CON_SOCK_STATE_CONNECTING);
341 }
342 
343 static void con_sock_state_connected(struct ceph_connection *con)
344 {
345 	int old_state;
346 
347 	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
348 	if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
349 		printk("%s: unexpected old state %d\n", __func__, old_state);
350 	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
351 	     CON_SOCK_STATE_CONNECTED);
352 }
353 
354 static void con_sock_state_closing(struct ceph_connection *con)
355 {
356 	int old_state;
357 
358 	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
359 	if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
360 			old_state != CON_SOCK_STATE_CONNECTED &&
361 			old_state != CON_SOCK_STATE_CLOSING))
362 		printk("%s: unexpected old state %d\n", __func__, old_state);
363 	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
364 	     CON_SOCK_STATE_CLOSING);
365 }
366 
367 static void con_sock_state_closed(struct ceph_connection *con)
368 {
369 	int old_state;
370 
371 	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
372 	if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
373 		    old_state != CON_SOCK_STATE_CLOSING &&
374 		    old_state != CON_SOCK_STATE_CONNECTING &&
375 		    old_state != CON_SOCK_STATE_CLOSED))
376 		printk("%s: unexpected old state %d\n", __func__, old_state);
377 	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
378 	     CON_SOCK_STATE_CLOSED);
379 }
380 
381 /*
382  * socket callback functions
383  */
384 
385 /* data available on socket, or listen socket received a connect */
386 static void ceph_sock_data_ready(struct sock *sk)
387 {
388 	struct ceph_connection *con = sk->sk_user_data;
389 	if (atomic_read(&con->msgr->stopping)) {
390 		return;
391 	}
392 
393 	if (sk->sk_state != TCP_CLOSE_WAIT) {
394 		dout("%s on %p state = %lu, queueing work\n", __func__,
395 		     con, con->state);
396 		queue_con(con);
397 	}
398 }
399 
400 /* socket has buffer space for writing */
401 static void ceph_sock_write_space(struct sock *sk)
402 {
403 	struct ceph_connection *con = sk->sk_user_data;
404 
405 	/* only queue to workqueue if there is data we want to write,
406 	 * and there is sufficient space in the socket buffer to accept
407 	 * more data.  clear SOCK_NOSPACE so that ceph_sock_write_space()
408 	 * doesn't get called again until try_write() fills the socket
409 	 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
410 	 * and net/core/stream.c:sk_stream_write_space().
411 	 */
412 	if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
413 		if (sk_stream_is_writeable(sk)) {
414 			dout("%s %p queueing write work\n", __func__, con);
415 			clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
416 			queue_con(con);
417 		}
418 	} else {
419 		dout("%s %p nothing to write\n", __func__, con);
420 	}
421 }
422 
423 /* socket's state has changed */
424 static void ceph_sock_state_change(struct sock *sk)
425 {
426 	struct ceph_connection *con = sk->sk_user_data;
427 
428 	dout("%s %p state = %lu sk_state = %u\n", __func__,
429 	     con, con->state, sk->sk_state);
430 
431 	switch (sk->sk_state) {
432 	case TCP_CLOSE:
433 		dout("%s TCP_CLOSE\n", __func__);
434 	case TCP_CLOSE_WAIT:
435 		dout("%s TCP_CLOSE_WAIT\n", __func__);
436 		con_sock_state_closing(con);
437 		con_flag_set(con, CON_FLAG_SOCK_CLOSED);
438 		queue_con(con);
439 		break;
440 	case TCP_ESTABLISHED:
441 		dout("%s TCP_ESTABLISHED\n", __func__);
442 		con_sock_state_connected(con);
443 		queue_con(con);
444 		break;
445 	default:	/* Everything else is uninteresting */
446 		break;
447 	}
448 }
449 
450 /*
451  * set up socket callbacks
452  */
453 static void set_sock_callbacks(struct socket *sock,
454 			       struct ceph_connection *con)
455 {
456 	struct sock *sk = sock->sk;
457 	sk->sk_user_data = con;
458 	sk->sk_data_ready = ceph_sock_data_ready;
459 	sk->sk_write_space = ceph_sock_write_space;
460 	sk->sk_state_change = ceph_sock_state_change;
461 }
462 
463 
464 /*
465  * socket helpers
466  */
467 
468 /*
469  * initiate connection to a remote socket.
470  */
471 static int ceph_tcp_connect(struct ceph_connection *con)
472 {
473 	struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
474 	struct socket *sock;
475 	int ret;
476 
477 	BUG_ON(con->sock);
478 	ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
479 			       IPPROTO_TCP, &sock);
480 	if (ret)
481 		return ret;
482 	sock->sk->sk_allocation = GFP_NOFS;
483 
484 #ifdef CONFIG_LOCKDEP
485 	lockdep_set_class(&sock->sk->sk_lock, &socket_class);
486 #endif
487 
488 	set_sock_callbacks(sock, con);
489 
490 	dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
491 
492 	con_sock_state_connecting(con);
493 	ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
494 				 O_NONBLOCK);
495 	if (ret == -EINPROGRESS) {
496 		dout("connect %s EINPROGRESS sk_state = %u\n",
497 		     ceph_pr_addr(&con->peer_addr.in_addr),
498 		     sock->sk->sk_state);
499 	} else if (ret < 0) {
500 		pr_err("connect %s error %d\n",
501 		       ceph_pr_addr(&con->peer_addr.in_addr), ret);
502 		sock_release(sock);
503 		con->error_msg = "connect error";
504 
505 		return ret;
506 	}
507 	con->sock = sock;
508 	return 0;
509 }
510 
511 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
512 {
513 	struct kvec iov = {buf, len};
514 	struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
515 	int r;
516 
517 	r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
518 	if (r == -EAGAIN)
519 		r = 0;
520 	return r;
521 }
522 
523 static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
524 		     int page_offset, size_t length)
525 {
526 	void *kaddr;
527 	int ret;
528 
529 	BUG_ON(page_offset + length > PAGE_SIZE);
530 
531 	kaddr = kmap(page);
532 	BUG_ON(!kaddr);
533 	ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
534 	kunmap(page);
535 
536 	return ret;
537 }
538 
539 /*
540  * write something.  @more is true if caller will be sending more data
541  * shortly.
542  */
543 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
544 		     size_t kvlen, size_t len, int more)
545 {
546 	struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
547 	int r;
548 
549 	if (more)
550 		msg.msg_flags |= MSG_MORE;
551 	else
552 		msg.msg_flags |= MSG_EOR;  /* superfluous, but what the hell */
553 
554 	r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
555 	if (r == -EAGAIN)
556 		r = 0;
557 	return r;
558 }
559 
560 static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
561 		     int offset, size_t size, bool more)
562 {
563 	int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
564 	int ret;
565 
566 	ret = kernel_sendpage(sock, page, offset, size, flags);
567 	if (ret == -EAGAIN)
568 		ret = 0;
569 
570 	return ret;
571 }
572 
573 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
574 		     int offset, size_t size, bool more)
575 {
576 	int ret;
577 	struct kvec iov;
578 
579 	/* sendpage cannot properly handle pages with page_count == 0,
580 	 * we need to fallback to sendmsg if that's the case */
581 	if (page_count(page) >= 1)
582 		return __ceph_tcp_sendpage(sock, page, offset, size, more);
583 
584 	iov.iov_base = kmap(page) + offset;
585 	iov.iov_len = size;
586 	ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more);
587 	kunmap(page);
588 
589 	return ret;
590 }
591 
592 /*
593  * Shutdown/close the socket for the given connection.
594  */
595 static int con_close_socket(struct ceph_connection *con)
596 {
597 	int rc = 0;
598 
599 	dout("con_close_socket on %p sock %p\n", con, con->sock);
600 	if (con->sock) {
601 		rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
602 		sock_release(con->sock);
603 		con->sock = NULL;
604 	}
605 
606 	/*
607 	 * Forcibly clear the SOCK_CLOSED flag.  It gets set
608 	 * independent of the connection mutex, and we could have
609 	 * received a socket close event before we had the chance to
610 	 * shut the socket down.
611 	 */
612 	con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
613 
614 	con_sock_state_closed(con);
615 	return rc;
616 }
617 
618 /*
619  * Reset a connection.  Discard all incoming and outgoing messages
620  * and clear *_seq state.
621  */
622 static void ceph_msg_remove(struct ceph_msg *msg)
623 {
624 	list_del_init(&msg->list_head);
625 	BUG_ON(msg->con == NULL);
626 	msg->con->ops->put(msg->con);
627 	msg->con = NULL;
628 
629 	ceph_msg_put(msg);
630 }
631 static void ceph_msg_remove_list(struct list_head *head)
632 {
633 	while (!list_empty(head)) {
634 		struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
635 							list_head);
636 		ceph_msg_remove(msg);
637 	}
638 }
639 
640 static void reset_connection(struct ceph_connection *con)
641 {
642 	/* reset connection, out_queue, msg_ and connect_seq */
643 	/* discard existing out_queue and msg_seq */
644 	dout("reset_connection %p\n", con);
645 	ceph_msg_remove_list(&con->out_queue);
646 	ceph_msg_remove_list(&con->out_sent);
647 
648 	if (con->in_msg) {
649 		BUG_ON(con->in_msg->con != con);
650 		con->in_msg->con = NULL;
651 		ceph_msg_put(con->in_msg);
652 		con->in_msg = NULL;
653 		con->ops->put(con);
654 	}
655 
656 	con->connect_seq = 0;
657 	con->out_seq = 0;
658 	if (con->out_msg) {
659 		ceph_msg_put(con->out_msg);
660 		con->out_msg = NULL;
661 	}
662 	con->in_seq = 0;
663 	con->in_seq_acked = 0;
664 }
665 
666 /*
667  * mark a peer down.  drop any open connections.
668  */
669 void ceph_con_close(struct ceph_connection *con)
670 {
671 	mutex_lock(&con->mutex);
672 	dout("con_close %p peer %s\n", con,
673 	     ceph_pr_addr(&con->peer_addr.in_addr));
674 	con->state = CON_STATE_CLOSED;
675 
676 	con_flag_clear(con, CON_FLAG_LOSSYTX);	/* so we retry next connect */
677 	con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
678 	con_flag_clear(con, CON_FLAG_WRITE_PENDING);
679 	con_flag_clear(con, CON_FLAG_BACKOFF);
680 
681 	reset_connection(con);
682 	con->peer_global_seq = 0;
683 	cancel_delayed_work(&con->work);
684 	con_close_socket(con);
685 	mutex_unlock(&con->mutex);
686 }
687 EXPORT_SYMBOL(ceph_con_close);
688 
689 /*
690  * Reopen a closed connection, with a new peer address.
691  */
692 void ceph_con_open(struct ceph_connection *con,
693 		   __u8 entity_type, __u64 entity_num,
694 		   struct ceph_entity_addr *addr)
695 {
696 	mutex_lock(&con->mutex);
697 	dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
698 
699 	WARN_ON(con->state != CON_STATE_CLOSED);
700 	con->state = CON_STATE_PREOPEN;
701 
702 	con->peer_name.type = (__u8) entity_type;
703 	con->peer_name.num = cpu_to_le64(entity_num);
704 
705 	memcpy(&con->peer_addr, addr, sizeof(*addr));
706 	con->delay = 0;      /* reset backoff memory */
707 	mutex_unlock(&con->mutex);
708 	queue_con(con);
709 }
710 EXPORT_SYMBOL(ceph_con_open);
711 
712 /*
713  * return true if this connection ever successfully opened
714  */
715 bool ceph_con_opened(struct ceph_connection *con)
716 {
717 	return con->connect_seq > 0;
718 }
719 
720 /*
721  * initialize a new connection.
722  */
723 void ceph_con_init(struct ceph_connection *con, void *private,
724 	const struct ceph_connection_operations *ops,
725 	struct ceph_messenger *msgr)
726 {
727 	dout("con_init %p\n", con);
728 	memset(con, 0, sizeof(*con));
729 	con->private = private;
730 	con->ops = ops;
731 	con->msgr = msgr;
732 
733 	con_sock_state_init(con);
734 
735 	mutex_init(&con->mutex);
736 	INIT_LIST_HEAD(&con->out_queue);
737 	INIT_LIST_HEAD(&con->out_sent);
738 	INIT_DELAYED_WORK(&con->work, con_work);
739 
740 	con->state = CON_STATE_CLOSED;
741 }
742 EXPORT_SYMBOL(ceph_con_init);
743 
744 
745 /*
746  * We maintain a global counter to order connection attempts.  Get
747  * a unique seq greater than @gt.
748  */
749 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
750 {
751 	u32 ret;
752 
753 	spin_lock(&msgr->global_seq_lock);
754 	if (msgr->global_seq < gt)
755 		msgr->global_seq = gt;
756 	ret = ++msgr->global_seq;
757 	spin_unlock(&msgr->global_seq_lock);
758 	return ret;
759 }
760 
761 static void con_out_kvec_reset(struct ceph_connection *con)
762 {
763 	con->out_kvec_left = 0;
764 	con->out_kvec_bytes = 0;
765 	con->out_kvec_cur = &con->out_kvec[0];
766 }
767 
768 static void con_out_kvec_add(struct ceph_connection *con,
769 				size_t size, void *data)
770 {
771 	int index;
772 
773 	index = con->out_kvec_left;
774 	BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
775 
776 	con->out_kvec[index].iov_len = size;
777 	con->out_kvec[index].iov_base = data;
778 	con->out_kvec_left++;
779 	con->out_kvec_bytes += size;
780 }
781 
782 #ifdef CONFIG_BLOCK
783 
784 /*
785  * For a bio data item, a piece is whatever remains of the next
786  * entry in the current bio iovec, or the first entry in the next
787  * bio in the list.
788  */
789 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
790 					size_t length)
791 {
792 	struct ceph_msg_data *data = cursor->data;
793 	struct bio *bio;
794 
795 	BUG_ON(data->type != CEPH_MSG_DATA_BIO);
796 
797 	bio = data->bio;
798 	BUG_ON(!bio);
799 
800 	cursor->resid = min(length, data->bio_length);
801 	cursor->bio = bio;
802 	cursor->bvec_iter = bio->bi_iter;
803 	cursor->last_piece =
804 		cursor->resid <= bio_iter_len(bio, cursor->bvec_iter);
805 }
806 
807 static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
808 						size_t *page_offset,
809 						size_t *length)
810 {
811 	struct ceph_msg_data *data = cursor->data;
812 	struct bio *bio;
813 	struct bio_vec bio_vec;
814 
815 	BUG_ON(data->type != CEPH_MSG_DATA_BIO);
816 
817 	bio = cursor->bio;
818 	BUG_ON(!bio);
819 
820 	bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
821 
822 	*page_offset = (size_t) bio_vec.bv_offset;
823 	BUG_ON(*page_offset >= PAGE_SIZE);
824 	if (cursor->last_piece) /* pagelist offset is always 0 */
825 		*length = cursor->resid;
826 	else
827 		*length = (size_t) bio_vec.bv_len;
828 	BUG_ON(*length > cursor->resid);
829 	BUG_ON(*page_offset + *length > PAGE_SIZE);
830 
831 	return bio_vec.bv_page;
832 }
833 
834 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
835 					size_t bytes)
836 {
837 	struct bio *bio;
838 	struct bio_vec bio_vec;
839 
840 	BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
841 
842 	bio = cursor->bio;
843 	BUG_ON(!bio);
844 
845 	bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
846 
847 	/* Advance the cursor offset */
848 
849 	BUG_ON(cursor->resid < bytes);
850 	cursor->resid -= bytes;
851 
852 	bio_advance_iter(bio, &cursor->bvec_iter, bytes);
853 
854 	if (bytes < bio_vec.bv_len)
855 		return false;	/* more bytes to process in this segment */
856 
857 	/* Move on to the next segment, and possibly the next bio */
858 
859 	if (!cursor->bvec_iter.bi_size) {
860 		bio = bio->bi_next;
861 		cursor->bio = bio;
862 		if (bio)
863 			cursor->bvec_iter = bio->bi_iter;
864 		else
865 			memset(&cursor->bvec_iter, 0,
866 			       sizeof(cursor->bvec_iter));
867 	}
868 
869 	if (!cursor->last_piece) {
870 		BUG_ON(!cursor->resid);
871 		BUG_ON(!bio);
872 		/* A short read is OK, so use <= rather than == */
873 		if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter))
874 			cursor->last_piece = true;
875 	}
876 
877 	return true;
878 }
879 #endif /* CONFIG_BLOCK */
880 
881 /*
882  * For a page array, a piece comes from the first page in the array
883  * that has not already been fully consumed.
884  */
885 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
886 					size_t length)
887 {
888 	struct ceph_msg_data *data = cursor->data;
889 	int page_count;
890 
891 	BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
892 
893 	BUG_ON(!data->pages);
894 	BUG_ON(!data->length);
895 
896 	cursor->resid = min(length, data->length);
897 	page_count = calc_pages_for(data->alignment, (u64)data->length);
898 	cursor->page_offset = data->alignment & ~PAGE_MASK;
899 	cursor->page_index = 0;
900 	BUG_ON(page_count > (int)USHRT_MAX);
901 	cursor->page_count = (unsigned short)page_count;
902 	BUG_ON(length > SIZE_MAX - cursor->page_offset);
903 	cursor->last_piece = (size_t)cursor->page_offset + length <= PAGE_SIZE;
904 }
905 
906 static struct page *
907 ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
908 					size_t *page_offset, size_t *length)
909 {
910 	struct ceph_msg_data *data = cursor->data;
911 
912 	BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
913 
914 	BUG_ON(cursor->page_index >= cursor->page_count);
915 	BUG_ON(cursor->page_offset >= PAGE_SIZE);
916 
917 	*page_offset = cursor->page_offset;
918 	if (cursor->last_piece)
919 		*length = cursor->resid;
920 	else
921 		*length = PAGE_SIZE - *page_offset;
922 
923 	return data->pages[cursor->page_index];
924 }
925 
926 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
927 						size_t bytes)
928 {
929 	BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
930 
931 	BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
932 
933 	/* Advance the cursor page offset */
934 
935 	cursor->resid -= bytes;
936 	cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
937 	if (!bytes || cursor->page_offset)
938 		return false;	/* more bytes to process in the current page */
939 
940 	if (!cursor->resid)
941 		return false;   /* no more data */
942 
943 	/* Move on to the next page; offset is already at 0 */
944 
945 	BUG_ON(cursor->page_index >= cursor->page_count);
946 	cursor->page_index++;
947 	cursor->last_piece = cursor->resid <= PAGE_SIZE;
948 
949 	return true;
950 }
951 
952 /*
953  * For a pagelist, a piece is whatever remains to be consumed in the
954  * first page in the list, or the front of the next page.
955  */
956 static void
957 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
958 					size_t length)
959 {
960 	struct ceph_msg_data *data = cursor->data;
961 	struct ceph_pagelist *pagelist;
962 	struct page *page;
963 
964 	BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
965 
966 	pagelist = data->pagelist;
967 	BUG_ON(!pagelist);
968 
969 	if (!length)
970 		return;		/* pagelist can be assigned but empty */
971 
972 	BUG_ON(list_empty(&pagelist->head));
973 	page = list_first_entry(&pagelist->head, struct page, lru);
974 
975 	cursor->resid = min(length, pagelist->length);
976 	cursor->page = page;
977 	cursor->offset = 0;
978 	cursor->last_piece = cursor->resid <= PAGE_SIZE;
979 }
980 
981 static struct page *
982 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
983 				size_t *page_offset, size_t *length)
984 {
985 	struct ceph_msg_data *data = cursor->data;
986 	struct ceph_pagelist *pagelist;
987 
988 	BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
989 
990 	pagelist = data->pagelist;
991 	BUG_ON(!pagelist);
992 
993 	BUG_ON(!cursor->page);
994 	BUG_ON(cursor->offset + cursor->resid != pagelist->length);
995 
996 	/* offset of first page in pagelist is always 0 */
997 	*page_offset = cursor->offset & ~PAGE_MASK;
998 	if (cursor->last_piece)
999 		*length = cursor->resid;
1000 	else
1001 		*length = PAGE_SIZE - *page_offset;
1002 
1003 	return cursor->page;
1004 }
1005 
1006 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
1007 						size_t bytes)
1008 {
1009 	struct ceph_msg_data *data = cursor->data;
1010 	struct ceph_pagelist *pagelist;
1011 
1012 	BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1013 
1014 	pagelist = data->pagelist;
1015 	BUG_ON(!pagelist);
1016 
1017 	BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1018 	BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1019 
1020 	/* Advance the cursor offset */
1021 
1022 	cursor->resid -= bytes;
1023 	cursor->offset += bytes;
1024 	/* offset of first page in pagelist is always 0 */
1025 	if (!bytes || cursor->offset & ~PAGE_MASK)
1026 		return false;	/* more bytes to process in the current page */
1027 
1028 	if (!cursor->resid)
1029 		return false;   /* no more data */
1030 
1031 	/* Move on to the next page */
1032 
1033 	BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1034 	cursor->page = list_entry_next(cursor->page, lru);
1035 	cursor->last_piece = cursor->resid <= PAGE_SIZE;
1036 
1037 	return true;
1038 }
1039 
1040 /*
1041  * Message data is handled (sent or received) in pieces, where each
1042  * piece resides on a single page.  The network layer might not
1043  * consume an entire piece at once.  A data item's cursor keeps
1044  * track of which piece is next to process and how much remains to
1045  * be processed in that piece.  It also tracks whether the current
1046  * piece is the last one in the data item.
1047  */
1048 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1049 {
1050 	size_t length = cursor->total_resid;
1051 
1052 	switch (cursor->data->type) {
1053 	case CEPH_MSG_DATA_PAGELIST:
1054 		ceph_msg_data_pagelist_cursor_init(cursor, length);
1055 		break;
1056 	case CEPH_MSG_DATA_PAGES:
1057 		ceph_msg_data_pages_cursor_init(cursor, length);
1058 		break;
1059 #ifdef CONFIG_BLOCK
1060 	case CEPH_MSG_DATA_BIO:
1061 		ceph_msg_data_bio_cursor_init(cursor, length);
1062 		break;
1063 #endif /* CONFIG_BLOCK */
1064 	case CEPH_MSG_DATA_NONE:
1065 	default:
1066 		/* BUG(); */
1067 		break;
1068 	}
1069 	cursor->need_crc = true;
1070 }
1071 
1072 static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1073 {
1074 	struct ceph_msg_data_cursor *cursor = &msg->cursor;
1075 	struct ceph_msg_data *data;
1076 
1077 	BUG_ON(!length);
1078 	BUG_ON(length > msg->data_length);
1079 	BUG_ON(list_empty(&msg->data));
1080 
1081 	cursor->data_head = &msg->data;
1082 	cursor->total_resid = length;
1083 	data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1084 	cursor->data = data;
1085 
1086 	__ceph_msg_data_cursor_init(cursor);
1087 }
1088 
1089 /*
1090  * Return the page containing the next piece to process for a given
1091  * data item, and supply the page offset and length of that piece.
1092  * Indicate whether this is the last piece in this data item.
1093  */
1094 static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1095 					size_t *page_offset, size_t *length,
1096 					bool *last_piece)
1097 {
1098 	struct page *page;
1099 
1100 	switch (cursor->data->type) {
1101 	case CEPH_MSG_DATA_PAGELIST:
1102 		page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1103 		break;
1104 	case CEPH_MSG_DATA_PAGES:
1105 		page = ceph_msg_data_pages_next(cursor, page_offset, length);
1106 		break;
1107 #ifdef CONFIG_BLOCK
1108 	case CEPH_MSG_DATA_BIO:
1109 		page = ceph_msg_data_bio_next(cursor, page_offset, length);
1110 		break;
1111 #endif /* CONFIG_BLOCK */
1112 	case CEPH_MSG_DATA_NONE:
1113 	default:
1114 		page = NULL;
1115 		break;
1116 	}
1117 	BUG_ON(!page);
1118 	BUG_ON(*page_offset + *length > PAGE_SIZE);
1119 	BUG_ON(!*length);
1120 	if (last_piece)
1121 		*last_piece = cursor->last_piece;
1122 
1123 	return page;
1124 }
1125 
1126 /*
1127  * Returns true if the result moves the cursor on to the next piece
1128  * of the data item.
1129  */
1130 static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1131 				size_t bytes)
1132 {
1133 	bool new_piece;
1134 
1135 	BUG_ON(bytes > cursor->resid);
1136 	switch (cursor->data->type) {
1137 	case CEPH_MSG_DATA_PAGELIST:
1138 		new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1139 		break;
1140 	case CEPH_MSG_DATA_PAGES:
1141 		new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1142 		break;
1143 #ifdef CONFIG_BLOCK
1144 	case CEPH_MSG_DATA_BIO:
1145 		new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1146 		break;
1147 #endif /* CONFIG_BLOCK */
1148 	case CEPH_MSG_DATA_NONE:
1149 	default:
1150 		BUG();
1151 		break;
1152 	}
1153 	cursor->total_resid -= bytes;
1154 
1155 	if (!cursor->resid && cursor->total_resid) {
1156 		WARN_ON(!cursor->last_piece);
1157 		BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1158 		cursor->data = list_entry_next(cursor->data, links);
1159 		__ceph_msg_data_cursor_init(cursor);
1160 		new_piece = true;
1161 	}
1162 	cursor->need_crc = new_piece;
1163 
1164 	return new_piece;
1165 }
1166 
1167 static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
1168 {
1169 	BUG_ON(!msg);
1170 	BUG_ON(!data_len);
1171 
1172 	/* Initialize data cursor */
1173 
1174 	ceph_msg_data_cursor_init(msg, (size_t)data_len);
1175 }
1176 
1177 /*
1178  * Prepare footer for currently outgoing message, and finish things
1179  * off.  Assumes out_kvec* are already valid.. we just add on to the end.
1180  */
1181 static void prepare_write_message_footer(struct ceph_connection *con)
1182 {
1183 	struct ceph_msg *m = con->out_msg;
1184 	int v = con->out_kvec_left;
1185 
1186 	m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1187 
1188 	dout("prepare_write_message_footer %p\n", con);
1189 	con->out_kvec_is_msg = true;
1190 	con->out_kvec[v].iov_base = &m->footer;
1191 	con->out_kvec[v].iov_len = sizeof(m->footer);
1192 	con->out_kvec_bytes += sizeof(m->footer);
1193 	con->out_kvec_left++;
1194 	con->out_more = m->more_to_follow;
1195 	con->out_msg_done = true;
1196 }
1197 
1198 /*
1199  * Prepare headers for the next outgoing message.
1200  */
1201 static void prepare_write_message(struct ceph_connection *con)
1202 {
1203 	struct ceph_msg *m;
1204 	u32 crc;
1205 
1206 	con_out_kvec_reset(con);
1207 	con->out_kvec_is_msg = true;
1208 	con->out_msg_done = false;
1209 
1210 	/* Sneak an ack in there first?  If we can get it into the same
1211 	 * TCP packet that's a good thing. */
1212 	if (con->in_seq > con->in_seq_acked) {
1213 		con->in_seq_acked = con->in_seq;
1214 		con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1215 		con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1216 		con_out_kvec_add(con, sizeof (con->out_temp_ack),
1217 			&con->out_temp_ack);
1218 	}
1219 
1220 	BUG_ON(list_empty(&con->out_queue));
1221 	m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1222 	con->out_msg = m;
1223 	BUG_ON(m->con != con);
1224 
1225 	/* put message on sent list */
1226 	ceph_msg_get(m);
1227 	list_move_tail(&m->list_head, &con->out_sent);
1228 
1229 	/*
1230 	 * only assign outgoing seq # if we haven't sent this message
1231 	 * yet.  if it is requeued, resend with it's original seq.
1232 	 */
1233 	if (m->needs_out_seq) {
1234 		m->hdr.seq = cpu_to_le64(++con->out_seq);
1235 		m->needs_out_seq = false;
1236 	}
1237 	WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
1238 
1239 	dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1240 	     m, con->out_seq, le16_to_cpu(m->hdr.type),
1241 	     le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1242 	     m->data_length);
1243 	BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1244 
1245 	/* tag + hdr + front + middle */
1246 	con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1247 	con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1248 	con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1249 
1250 	if (m->middle)
1251 		con_out_kvec_add(con, m->middle->vec.iov_len,
1252 			m->middle->vec.iov_base);
1253 
1254 	/* fill in crc (except data pages), footer */
1255 	crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1256 	con->out_msg->hdr.crc = cpu_to_le32(crc);
1257 	con->out_msg->footer.flags = 0;
1258 
1259 	crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1260 	con->out_msg->footer.front_crc = cpu_to_le32(crc);
1261 	if (m->middle) {
1262 		crc = crc32c(0, m->middle->vec.iov_base,
1263 				m->middle->vec.iov_len);
1264 		con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1265 	} else
1266 		con->out_msg->footer.middle_crc = 0;
1267 	dout("%s front_crc %u middle_crc %u\n", __func__,
1268 	     le32_to_cpu(con->out_msg->footer.front_crc),
1269 	     le32_to_cpu(con->out_msg->footer.middle_crc));
1270 
1271 	/* is there a data payload? */
1272 	con->out_msg->footer.data_crc = 0;
1273 	if (m->data_length) {
1274 		prepare_message_data(con->out_msg, m->data_length);
1275 		con->out_more = 1;  /* data + footer will follow */
1276 	} else {
1277 		/* no, queue up footer too and be done */
1278 		prepare_write_message_footer(con);
1279 	}
1280 
1281 	con_flag_set(con, CON_FLAG_WRITE_PENDING);
1282 }
1283 
1284 /*
1285  * Prepare an ack.
1286  */
1287 static void prepare_write_ack(struct ceph_connection *con)
1288 {
1289 	dout("prepare_write_ack %p %llu -> %llu\n", con,
1290 	     con->in_seq_acked, con->in_seq);
1291 	con->in_seq_acked = con->in_seq;
1292 
1293 	con_out_kvec_reset(con);
1294 
1295 	con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1296 
1297 	con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1298 	con_out_kvec_add(con, sizeof (con->out_temp_ack),
1299 				&con->out_temp_ack);
1300 
1301 	con->out_more = 1;  /* more will follow.. eventually.. */
1302 	con_flag_set(con, CON_FLAG_WRITE_PENDING);
1303 }
1304 
1305 /*
1306  * Prepare to share the seq during handshake
1307  */
1308 static void prepare_write_seq(struct ceph_connection *con)
1309 {
1310 	dout("prepare_write_seq %p %llu -> %llu\n", con,
1311 	     con->in_seq_acked, con->in_seq);
1312 	con->in_seq_acked = con->in_seq;
1313 
1314 	con_out_kvec_reset(con);
1315 
1316 	con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1317 	con_out_kvec_add(con, sizeof (con->out_temp_ack),
1318 			 &con->out_temp_ack);
1319 
1320 	con_flag_set(con, CON_FLAG_WRITE_PENDING);
1321 }
1322 
1323 /*
1324  * Prepare to write keepalive byte.
1325  */
1326 static void prepare_write_keepalive(struct ceph_connection *con)
1327 {
1328 	dout("prepare_write_keepalive %p\n", con);
1329 	con_out_kvec_reset(con);
1330 	con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
1331 	con_flag_set(con, CON_FLAG_WRITE_PENDING);
1332 }
1333 
1334 /*
1335  * Connection negotiation.
1336  */
1337 
1338 static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1339 						int *auth_proto)
1340 {
1341 	struct ceph_auth_handshake *auth;
1342 
1343 	if (!con->ops->get_authorizer) {
1344 		con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1345 		con->out_connect.authorizer_len = 0;
1346 		return NULL;
1347 	}
1348 
1349 	/* Can't hold the mutex while getting authorizer */
1350 	mutex_unlock(&con->mutex);
1351 	auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
1352 	mutex_lock(&con->mutex);
1353 
1354 	if (IS_ERR(auth))
1355 		return auth;
1356 	if (con->state != CON_STATE_NEGOTIATING)
1357 		return ERR_PTR(-EAGAIN);
1358 
1359 	con->auth_reply_buf = auth->authorizer_reply_buf;
1360 	con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
1361 	return auth;
1362 }
1363 
1364 /*
1365  * We connected to a peer and are saying hello.
1366  */
1367 static void prepare_write_banner(struct ceph_connection *con)
1368 {
1369 	con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1370 	con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1371 					&con->msgr->my_enc_addr);
1372 
1373 	con->out_more = 0;
1374 	con_flag_set(con, CON_FLAG_WRITE_PENDING);
1375 }
1376 
1377 static int prepare_write_connect(struct ceph_connection *con)
1378 {
1379 	unsigned int global_seq = get_global_seq(con->msgr, 0);
1380 	int proto;
1381 	int auth_proto;
1382 	struct ceph_auth_handshake *auth;
1383 
1384 	switch (con->peer_name.type) {
1385 	case CEPH_ENTITY_TYPE_MON:
1386 		proto = CEPH_MONC_PROTOCOL;
1387 		break;
1388 	case CEPH_ENTITY_TYPE_OSD:
1389 		proto = CEPH_OSDC_PROTOCOL;
1390 		break;
1391 	case CEPH_ENTITY_TYPE_MDS:
1392 		proto = CEPH_MDSC_PROTOCOL;
1393 		break;
1394 	default:
1395 		BUG();
1396 	}
1397 
1398 	dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1399 	     con->connect_seq, global_seq, proto);
1400 
1401 	con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
1402 	con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1403 	con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1404 	con->out_connect.global_seq = cpu_to_le32(global_seq);
1405 	con->out_connect.protocol_version = cpu_to_le32(proto);
1406 	con->out_connect.flags = 0;
1407 
1408 	auth_proto = CEPH_AUTH_UNKNOWN;
1409 	auth = get_connect_authorizer(con, &auth_proto);
1410 	if (IS_ERR(auth))
1411 		return PTR_ERR(auth);
1412 
1413 	con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1414 	con->out_connect.authorizer_len = auth ?
1415 		cpu_to_le32(auth->authorizer_buf_len) : 0;
1416 
1417 	con_out_kvec_add(con, sizeof (con->out_connect),
1418 					&con->out_connect);
1419 	if (auth && auth->authorizer_buf_len)
1420 		con_out_kvec_add(con, auth->authorizer_buf_len,
1421 					auth->authorizer_buf);
1422 
1423 	con->out_more = 0;
1424 	con_flag_set(con, CON_FLAG_WRITE_PENDING);
1425 
1426 	return 0;
1427 }
1428 
1429 /*
1430  * write as much of pending kvecs to the socket as we can.
1431  *  1 -> done
1432  *  0 -> socket full, but more to do
1433  * <0 -> error
1434  */
1435 static int write_partial_kvec(struct ceph_connection *con)
1436 {
1437 	int ret;
1438 
1439 	dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1440 	while (con->out_kvec_bytes > 0) {
1441 		ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1442 				       con->out_kvec_left, con->out_kvec_bytes,
1443 				       con->out_more);
1444 		if (ret <= 0)
1445 			goto out;
1446 		con->out_kvec_bytes -= ret;
1447 		if (con->out_kvec_bytes == 0)
1448 			break;            /* done */
1449 
1450 		/* account for full iov entries consumed */
1451 		while (ret >= con->out_kvec_cur->iov_len) {
1452 			BUG_ON(!con->out_kvec_left);
1453 			ret -= con->out_kvec_cur->iov_len;
1454 			con->out_kvec_cur++;
1455 			con->out_kvec_left--;
1456 		}
1457 		/* and for a partially-consumed entry */
1458 		if (ret) {
1459 			con->out_kvec_cur->iov_len -= ret;
1460 			con->out_kvec_cur->iov_base += ret;
1461 		}
1462 	}
1463 	con->out_kvec_left = 0;
1464 	con->out_kvec_is_msg = false;
1465 	ret = 1;
1466 out:
1467 	dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1468 	     con->out_kvec_bytes, con->out_kvec_left, ret);
1469 	return ret;  /* done! */
1470 }
1471 
1472 static u32 ceph_crc32c_page(u32 crc, struct page *page,
1473 				unsigned int page_offset,
1474 				unsigned int length)
1475 {
1476 	char *kaddr;
1477 
1478 	kaddr = kmap(page);
1479 	BUG_ON(kaddr == NULL);
1480 	crc = crc32c(crc, kaddr + page_offset, length);
1481 	kunmap(page);
1482 
1483 	return crc;
1484 }
1485 /*
1486  * Write as much message data payload as we can.  If we finish, queue
1487  * up the footer.
1488  *  1 -> done, footer is now queued in out_kvec[].
1489  *  0 -> socket full, but more to do
1490  * <0 -> error
1491  */
1492 static int write_partial_message_data(struct ceph_connection *con)
1493 {
1494 	struct ceph_msg *msg = con->out_msg;
1495 	struct ceph_msg_data_cursor *cursor = &msg->cursor;
1496 	bool do_datacrc = !con->msgr->nocrc;
1497 	u32 crc;
1498 
1499 	dout("%s %p msg %p\n", __func__, con, msg);
1500 
1501 	if (list_empty(&msg->data))
1502 		return -EINVAL;
1503 
1504 	/*
1505 	 * Iterate through each page that contains data to be
1506 	 * written, and send as much as possible for each.
1507 	 *
1508 	 * If we are calculating the data crc (the default), we will
1509 	 * need to map the page.  If we have no pages, they have
1510 	 * been revoked, so use the zero page.
1511 	 */
1512 	crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1513 	while (cursor->resid) {
1514 		struct page *page;
1515 		size_t page_offset;
1516 		size_t length;
1517 		bool last_piece;
1518 		bool need_crc;
1519 		int ret;
1520 
1521 		page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
1522 							&last_piece);
1523 		ret = ceph_tcp_sendpage(con->sock, page, page_offset,
1524 				      length, last_piece);
1525 		if (ret <= 0) {
1526 			if (do_datacrc)
1527 				msg->footer.data_crc = cpu_to_le32(crc);
1528 
1529 			return ret;
1530 		}
1531 		if (do_datacrc && cursor->need_crc)
1532 			crc = ceph_crc32c_page(crc, page, page_offset, length);
1533 		need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret);
1534 	}
1535 
1536 	dout("%s %p msg %p done\n", __func__, con, msg);
1537 
1538 	/* prepare and queue up footer, too */
1539 	if (do_datacrc)
1540 		msg->footer.data_crc = cpu_to_le32(crc);
1541 	else
1542 		msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1543 	con_out_kvec_reset(con);
1544 	prepare_write_message_footer(con);
1545 
1546 	return 1;	/* must return > 0 to indicate success */
1547 }
1548 
1549 /*
1550  * write some zeros
1551  */
1552 static int write_partial_skip(struct ceph_connection *con)
1553 {
1554 	int ret;
1555 
1556 	while (con->out_skip > 0) {
1557 		size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
1558 
1559 		ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
1560 		if (ret <= 0)
1561 			goto out;
1562 		con->out_skip -= ret;
1563 	}
1564 	ret = 1;
1565 out:
1566 	return ret;
1567 }
1568 
1569 /*
1570  * Prepare to read connection handshake, or an ack.
1571  */
1572 static void prepare_read_banner(struct ceph_connection *con)
1573 {
1574 	dout("prepare_read_banner %p\n", con);
1575 	con->in_base_pos = 0;
1576 }
1577 
1578 static void prepare_read_connect(struct ceph_connection *con)
1579 {
1580 	dout("prepare_read_connect %p\n", con);
1581 	con->in_base_pos = 0;
1582 }
1583 
1584 static void prepare_read_ack(struct ceph_connection *con)
1585 {
1586 	dout("prepare_read_ack %p\n", con);
1587 	con->in_base_pos = 0;
1588 }
1589 
1590 static void prepare_read_seq(struct ceph_connection *con)
1591 {
1592 	dout("prepare_read_seq %p\n", con);
1593 	con->in_base_pos = 0;
1594 	con->in_tag = CEPH_MSGR_TAG_SEQ;
1595 }
1596 
1597 static void prepare_read_tag(struct ceph_connection *con)
1598 {
1599 	dout("prepare_read_tag %p\n", con);
1600 	con->in_base_pos = 0;
1601 	con->in_tag = CEPH_MSGR_TAG_READY;
1602 }
1603 
1604 /*
1605  * Prepare to read a message.
1606  */
1607 static int prepare_read_message(struct ceph_connection *con)
1608 {
1609 	dout("prepare_read_message %p\n", con);
1610 	BUG_ON(con->in_msg != NULL);
1611 	con->in_base_pos = 0;
1612 	con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1613 	return 0;
1614 }
1615 
1616 
1617 static int read_partial(struct ceph_connection *con,
1618 			int end, int size, void *object)
1619 {
1620 	while (con->in_base_pos < end) {
1621 		int left = end - con->in_base_pos;
1622 		int have = size - left;
1623 		int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1624 		if (ret <= 0)
1625 			return ret;
1626 		con->in_base_pos += ret;
1627 	}
1628 	return 1;
1629 }
1630 
1631 
1632 /*
1633  * Read all or part of the connect-side handshake on a new connection
1634  */
1635 static int read_partial_banner(struct ceph_connection *con)
1636 {
1637 	int size;
1638 	int end;
1639 	int ret;
1640 
1641 	dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1642 
1643 	/* peer's banner */
1644 	size = strlen(CEPH_BANNER);
1645 	end = size;
1646 	ret = read_partial(con, end, size, con->in_banner);
1647 	if (ret <= 0)
1648 		goto out;
1649 
1650 	size = sizeof (con->actual_peer_addr);
1651 	end += size;
1652 	ret = read_partial(con, end, size, &con->actual_peer_addr);
1653 	if (ret <= 0)
1654 		goto out;
1655 
1656 	size = sizeof (con->peer_addr_for_me);
1657 	end += size;
1658 	ret = read_partial(con, end, size, &con->peer_addr_for_me);
1659 	if (ret <= 0)
1660 		goto out;
1661 
1662 out:
1663 	return ret;
1664 }
1665 
1666 static int read_partial_connect(struct ceph_connection *con)
1667 {
1668 	int size;
1669 	int end;
1670 	int ret;
1671 
1672 	dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1673 
1674 	size = sizeof (con->in_reply);
1675 	end = size;
1676 	ret = read_partial(con, end, size, &con->in_reply);
1677 	if (ret <= 0)
1678 		goto out;
1679 
1680 	size = le32_to_cpu(con->in_reply.authorizer_len);
1681 	end += size;
1682 	ret = read_partial(con, end, size, con->auth_reply_buf);
1683 	if (ret <= 0)
1684 		goto out;
1685 
1686 	dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1687 	     con, (int)con->in_reply.tag,
1688 	     le32_to_cpu(con->in_reply.connect_seq),
1689 	     le32_to_cpu(con->in_reply.global_seq));
1690 out:
1691 	return ret;
1692 
1693 }
1694 
1695 /*
1696  * Verify the hello banner looks okay.
1697  */
1698 static int verify_hello(struct ceph_connection *con)
1699 {
1700 	if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1701 		pr_err("connect to %s got bad banner\n",
1702 		       ceph_pr_addr(&con->peer_addr.in_addr));
1703 		con->error_msg = "protocol error, bad banner";
1704 		return -1;
1705 	}
1706 	return 0;
1707 }
1708 
1709 static bool addr_is_blank(struct sockaddr_storage *ss)
1710 {
1711 	switch (ss->ss_family) {
1712 	case AF_INET:
1713 		return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1714 	case AF_INET6:
1715 		return
1716 		     ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1717 		     ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1718 		     ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1719 		     ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1720 	}
1721 	return false;
1722 }
1723 
1724 static int addr_port(struct sockaddr_storage *ss)
1725 {
1726 	switch (ss->ss_family) {
1727 	case AF_INET:
1728 		return ntohs(((struct sockaddr_in *)ss)->sin_port);
1729 	case AF_INET6:
1730 		return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1731 	}
1732 	return 0;
1733 }
1734 
1735 static void addr_set_port(struct sockaddr_storage *ss, int p)
1736 {
1737 	switch (ss->ss_family) {
1738 	case AF_INET:
1739 		((struct sockaddr_in *)ss)->sin_port = htons(p);
1740 		break;
1741 	case AF_INET6:
1742 		((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1743 		break;
1744 	}
1745 }
1746 
1747 /*
1748  * Unlike other *_pton function semantics, zero indicates success.
1749  */
1750 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1751 		char delim, const char **ipend)
1752 {
1753 	struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1754 	struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1755 
1756 	memset(ss, 0, sizeof(*ss));
1757 
1758 	if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1759 		ss->ss_family = AF_INET;
1760 		return 0;
1761 	}
1762 
1763 	if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1764 		ss->ss_family = AF_INET6;
1765 		return 0;
1766 	}
1767 
1768 	return -EINVAL;
1769 }
1770 
1771 /*
1772  * Extract hostname string and resolve using kernel DNS facility.
1773  */
1774 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1775 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1776 		struct sockaddr_storage *ss, char delim, const char **ipend)
1777 {
1778 	const char *end, *delim_p;
1779 	char *colon_p, *ip_addr = NULL;
1780 	int ip_len, ret;
1781 
1782 	/*
1783 	 * The end of the hostname occurs immediately preceding the delimiter or
1784 	 * the port marker (':') where the delimiter takes precedence.
1785 	 */
1786 	delim_p = memchr(name, delim, namelen);
1787 	colon_p = memchr(name, ':', namelen);
1788 
1789 	if (delim_p && colon_p)
1790 		end = delim_p < colon_p ? delim_p : colon_p;
1791 	else if (!delim_p && colon_p)
1792 		end = colon_p;
1793 	else {
1794 		end = delim_p;
1795 		if (!end) /* case: hostname:/ */
1796 			end = name + namelen;
1797 	}
1798 
1799 	if (end <= name)
1800 		return -EINVAL;
1801 
1802 	/* do dns_resolve upcall */
1803 	ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1804 	if (ip_len > 0)
1805 		ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1806 	else
1807 		ret = -ESRCH;
1808 
1809 	kfree(ip_addr);
1810 
1811 	*ipend = end;
1812 
1813 	pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1814 			ret, ret ? "failed" : ceph_pr_addr(ss));
1815 
1816 	return ret;
1817 }
1818 #else
1819 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1820 		struct sockaddr_storage *ss, char delim, const char **ipend)
1821 {
1822 	return -EINVAL;
1823 }
1824 #endif
1825 
1826 /*
1827  * Parse a server name (IP or hostname). If a valid IP address is not found
1828  * then try to extract a hostname to resolve using userspace DNS upcall.
1829  */
1830 static int ceph_parse_server_name(const char *name, size_t namelen,
1831 			struct sockaddr_storage *ss, char delim, const char **ipend)
1832 {
1833 	int ret;
1834 
1835 	ret = ceph_pton(name, namelen, ss, delim, ipend);
1836 	if (ret)
1837 		ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1838 
1839 	return ret;
1840 }
1841 
1842 /*
1843  * Parse an ip[:port] list into an addr array.  Use the default
1844  * monitor port if a port isn't specified.
1845  */
1846 int ceph_parse_ips(const char *c, const char *end,
1847 		   struct ceph_entity_addr *addr,
1848 		   int max_count, int *count)
1849 {
1850 	int i, ret = -EINVAL;
1851 	const char *p = c;
1852 
1853 	dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1854 	for (i = 0; i < max_count; i++) {
1855 		const char *ipend;
1856 		struct sockaddr_storage *ss = &addr[i].in_addr;
1857 		int port;
1858 		char delim = ',';
1859 
1860 		if (*p == '[') {
1861 			delim = ']';
1862 			p++;
1863 		}
1864 
1865 		ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1866 		if (ret)
1867 			goto bad;
1868 		ret = -EINVAL;
1869 
1870 		p = ipend;
1871 
1872 		if (delim == ']') {
1873 			if (*p != ']') {
1874 				dout("missing matching ']'\n");
1875 				goto bad;
1876 			}
1877 			p++;
1878 		}
1879 
1880 		/* port? */
1881 		if (p < end && *p == ':') {
1882 			port = 0;
1883 			p++;
1884 			while (p < end && *p >= '0' && *p <= '9') {
1885 				port = (port * 10) + (*p - '0');
1886 				p++;
1887 			}
1888 			if (port == 0)
1889 				port = CEPH_MON_PORT;
1890 			else if (port > 65535)
1891 				goto bad;
1892 		} else {
1893 			port = CEPH_MON_PORT;
1894 		}
1895 
1896 		addr_set_port(ss, port);
1897 
1898 		dout("parse_ips got %s\n", ceph_pr_addr(ss));
1899 
1900 		if (p == end)
1901 			break;
1902 		if (*p != ',')
1903 			goto bad;
1904 		p++;
1905 	}
1906 
1907 	if (p != end)
1908 		goto bad;
1909 
1910 	if (count)
1911 		*count = i + 1;
1912 	return 0;
1913 
1914 bad:
1915 	pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1916 	return ret;
1917 }
1918 EXPORT_SYMBOL(ceph_parse_ips);
1919 
1920 static int process_banner(struct ceph_connection *con)
1921 {
1922 	dout("process_banner on %p\n", con);
1923 
1924 	if (verify_hello(con) < 0)
1925 		return -1;
1926 
1927 	ceph_decode_addr(&con->actual_peer_addr);
1928 	ceph_decode_addr(&con->peer_addr_for_me);
1929 
1930 	/*
1931 	 * Make sure the other end is who we wanted.  note that the other
1932 	 * end may not yet know their ip address, so if it's 0.0.0.0, give
1933 	 * them the benefit of the doubt.
1934 	 */
1935 	if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1936 		   sizeof(con->peer_addr)) != 0 &&
1937 	    !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1938 	      con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1939 		pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1940 			   ceph_pr_addr(&con->peer_addr.in_addr),
1941 			   (int)le32_to_cpu(con->peer_addr.nonce),
1942 			   ceph_pr_addr(&con->actual_peer_addr.in_addr),
1943 			   (int)le32_to_cpu(con->actual_peer_addr.nonce));
1944 		con->error_msg = "wrong peer at address";
1945 		return -1;
1946 	}
1947 
1948 	/*
1949 	 * did we learn our address?
1950 	 */
1951 	if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1952 		int port = addr_port(&con->msgr->inst.addr.in_addr);
1953 
1954 		memcpy(&con->msgr->inst.addr.in_addr,
1955 		       &con->peer_addr_for_me.in_addr,
1956 		       sizeof(con->peer_addr_for_me.in_addr));
1957 		addr_set_port(&con->msgr->inst.addr.in_addr, port);
1958 		encode_my_addr(con->msgr);
1959 		dout("process_banner learned my addr is %s\n",
1960 		     ceph_pr_addr(&con->msgr->inst.addr.in_addr));
1961 	}
1962 
1963 	return 0;
1964 }
1965 
1966 static int process_connect(struct ceph_connection *con)
1967 {
1968 	u64 sup_feat = con->msgr->supported_features;
1969 	u64 req_feat = con->msgr->required_features;
1970 	u64 server_feat = ceph_sanitize_features(
1971 				le64_to_cpu(con->in_reply.features));
1972 	int ret;
1973 
1974 	dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1975 
1976 	switch (con->in_reply.tag) {
1977 	case CEPH_MSGR_TAG_FEATURES:
1978 		pr_err("%s%lld %s feature set mismatch,"
1979 		       " my %llx < server's %llx, missing %llx\n",
1980 		       ENTITY_NAME(con->peer_name),
1981 		       ceph_pr_addr(&con->peer_addr.in_addr),
1982 		       sup_feat, server_feat, server_feat & ~sup_feat);
1983 		con->error_msg = "missing required protocol features";
1984 		reset_connection(con);
1985 		return -1;
1986 
1987 	case CEPH_MSGR_TAG_BADPROTOVER:
1988 		pr_err("%s%lld %s protocol version mismatch,"
1989 		       " my %d != server's %d\n",
1990 		       ENTITY_NAME(con->peer_name),
1991 		       ceph_pr_addr(&con->peer_addr.in_addr),
1992 		       le32_to_cpu(con->out_connect.protocol_version),
1993 		       le32_to_cpu(con->in_reply.protocol_version));
1994 		con->error_msg = "protocol version mismatch";
1995 		reset_connection(con);
1996 		return -1;
1997 
1998 	case CEPH_MSGR_TAG_BADAUTHORIZER:
1999 		con->auth_retry++;
2000 		dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2001 		     con->auth_retry);
2002 		if (con->auth_retry == 2) {
2003 			con->error_msg = "connect authorization failure";
2004 			return -1;
2005 		}
2006 		con_out_kvec_reset(con);
2007 		ret = prepare_write_connect(con);
2008 		if (ret < 0)
2009 			return ret;
2010 		prepare_read_connect(con);
2011 		break;
2012 
2013 	case CEPH_MSGR_TAG_RESETSESSION:
2014 		/*
2015 		 * If we connected with a large connect_seq but the peer
2016 		 * has no record of a session with us (no connection, or
2017 		 * connect_seq == 0), they will send RESETSESION to indicate
2018 		 * that they must have reset their session, and may have
2019 		 * dropped messages.
2020 		 */
2021 		dout("process_connect got RESET peer seq %u\n",
2022 		     le32_to_cpu(con->in_reply.connect_seq));
2023 		pr_err("%s%lld %s connection reset\n",
2024 		       ENTITY_NAME(con->peer_name),
2025 		       ceph_pr_addr(&con->peer_addr.in_addr));
2026 		reset_connection(con);
2027 		con_out_kvec_reset(con);
2028 		ret = prepare_write_connect(con);
2029 		if (ret < 0)
2030 			return ret;
2031 		prepare_read_connect(con);
2032 
2033 		/* Tell ceph about it. */
2034 		mutex_unlock(&con->mutex);
2035 		pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2036 		if (con->ops->peer_reset)
2037 			con->ops->peer_reset(con);
2038 		mutex_lock(&con->mutex);
2039 		if (con->state != CON_STATE_NEGOTIATING)
2040 			return -EAGAIN;
2041 		break;
2042 
2043 	case CEPH_MSGR_TAG_RETRY_SESSION:
2044 		/*
2045 		 * If we sent a smaller connect_seq than the peer has, try
2046 		 * again with a larger value.
2047 		 */
2048 		dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2049 		     le32_to_cpu(con->out_connect.connect_seq),
2050 		     le32_to_cpu(con->in_reply.connect_seq));
2051 		con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2052 		con_out_kvec_reset(con);
2053 		ret = prepare_write_connect(con);
2054 		if (ret < 0)
2055 			return ret;
2056 		prepare_read_connect(con);
2057 		break;
2058 
2059 	case CEPH_MSGR_TAG_RETRY_GLOBAL:
2060 		/*
2061 		 * If we sent a smaller global_seq than the peer has, try
2062 		 * again with a larger value.
2063 		 */
2064 		dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2065 		     con->peer_global_seq,
2066 		     le32_to_cpu(con->in_reply.global_seq));
2067 		get_global_seq(con->msgr,
2068 			       le32_to_cpu(con->in_reply.global_seq));
2069 		con_out_kvec_reset(con);
2070 		ret = prepare_write_connect(con);
2071 		if (ret < 0)
2072 			return ret;
2073 		prepare_read_connect(con);
2074 		break;
2075 
2076 	case CEPH_MSGR_TAG_SEQ:
2077 	case CEPH_MSGR_TAG_READY:
2078 		if (req_feat & ~server_feat) {
2079 			pr_err("%s%lld %s protocol feature mismatch,"
2080 			       " my required %llx > server's %llx, need %llx\n",
2081 			       ENTITY_NAME(con->peer_name),
2082 			       ceph_pr_addr(&con->peer_addr.in_addr),
2083 			       req_feat, server_feat, req_feat & ~server_feat);
2084 			con->error_msg = "missing required protocol features";
2085 			reset_connection(con);
2086 			return -1;
2087 		}
2088 
2089 		WARN_ON(con->state != CON_STATE_NEGOTIATING);
2090 		con->state = CON_STATE_OPEN;
2091 		con->auth_retry = 0;    /* we authenticated; clear flag */
2092 		con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2093 		con->connect_seq++;
2094 		con->peer_features = server_feat;
2095 		dout("process_connect got READY gseq %d cseq %d (%d)\n",
2096 		     con->peer_global_seq,
2097 		     le32_to_cpu(con->in_reply.connect_seq),
2098 		     con->connect_seq);
2099 		WARN_ON(con->connect_seq !=
2100 			le32_to_cpu(con->in_reply.connect_seq));
2101 
2102 		if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2103 			con_flag_set(con, CON_FLAG_LOSSYTX);
2104 
2105 		con->delay = 0;      /* reset backoff memory */
2106 
2107 		if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2108 			prepare_write_seq(con);
2109 			prepare_read_seq(con);
2110 		} else {
2111 			prepare_read_tag(con);
2112 		}
2113 		break;
2114 
2115 	case CEPH_MSGR_TAG_WAIT:
2116 		/*
2117 		 * If there is a connection race (we are opening
2118 		 * connections to each other), one of us may just have
2119 		 * to WAIT.  This shouldn't happen if we are the
2120 		 * client.
2121 		 */
2122 		pr_err("process_connect got WAIT as client\n");
2123 		con->error_msg = "protocol error, got WAIT as client";
2124 		return -1;
2125 
2126 	default:
2127 		pr_err("connect protocol error, will retry\n");
2128 		con->error_msg = "protocol error, garbage tag during connect";
2129 		return -1;
2130 	}
2131 	return 0;
2132 }
2133 
2134 
2135 /*
2136  * read (part of) an ack
2137  */
2138 static int read_partial_ack(struct ceph_connection *con)
2139 {
2140 	int size = sizeof (con->in_temp_ack);
2141 	int end = size;
2142 
2143 	return read_partial(con, end, size, &con->in_temp_ack);
2144 }
2145 
2146 /*
2147  * We can finally discard anything that's been acked.
2148  */
2149 static void process_ack(struct ceph_connection *con)
2150 {
2151 	struct ceph_msg *m;
2152 	u64 ack = le64_to_cpu(con->in_temp_ack);
2153 	u64 seq;
2154 
2155 	while (!list_empty(&con->out_sent)) {
2156 		m = list_first_entry(&con->out_sent, struct ceph_msg,
2157 				     list_head);
2158 		seq = le64_to_cpu(m->hdr.seq);
2159 		if (seq > ack)
2160 			break;
2161 		dout("got ack for seq %llu type %d at %p\n", seq,
2162 		     le16_to_cpu(m->hdr.type), m);
2163 		m->ack_stamp = jiffies;
2164 		ceph_msg_remove(m);
2165 	}
2166 	prepare_read_tag(con);
2167 }
2168 
2169 
2170 static int read_partial_message_section(struct ceph_connection *con,
2171 					struct kvec *section,
2172 					unsigned int sec_len, u32 *crc)
2173 {
2174 	int ret, left;
2175 
2176 	BUG_ON(!section);
2177 
2178 	while (section->iov_len < sec_len) {
2179 		BUG_ON(section->iov_base == NULL);
2180 		left = sec_len - section->iov_len;
2181 		ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2182 				       section->iov_len, left);
2183 		if (ret <= 0)
2184 			return ret;
2185 		section->iov_len += ret;
2186 	}
2187 	if (section->iov_len == sec_len)
2188 		*crc = crc32c(0, section->iov_base, section->iov_len);
2189 
2190 	return 1;
2191 }
2192 
2193 static int read_partial_msg_data(struct ceph_connection *con)
2194 {
2195 	struct ceph_msg *msg = con->in_msg;
2196 	struct ceph_msg_data_cursor *cursor = &msg->cursor;
2197 	const bool do_datacrc = !con->msgr->nocrc;
2198 	struct page *page;
2199 	size_t page_offset;
2200 	size_t length;
2201 	u32 crc = 0;
2202 	int ret;
2203 
2204 	BUG_ON(!msg);
2205 	if (list_empty(&msg->data))
2206 		return -EIO;
2207 
2208 	if (do_datacrc)
2209 		crc = con->in_data_crc;
2210 	while (cursor->resid) {
2211 		page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
2212 							NULL);
2213 		ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2214 		if (ret <= 0) {
2215 			if (do_datacrc)
2216 				con->in_data_crc = crc;
2217 
2218 			return ret;
2219 		}
2220 
2221 		if (do_datacrc)
2222 			crc = ceph_crc32c_page(crc, page, page_offset, ret);
2223 		(void) ceph_msg_data_advance(&msg->cursor, (size_t)ret);
2224 	}
2225 	if (do_datacrc)
2226 		con->in_data_crc = crc;
2227 
2228 	return 1;	/* must return > 0 to indicate success */
2229 }
2230 
2231 /*
2232  * read (part of) a message.
2233  */
2234 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2235 
2236 static int read_partial_message(struct ceph_connection *con)
2237 {
2238 	struct ceph_msg *m = con->in_msg;
2239 	int size;
2240 	int end;
2241 	int ret;
2242 	unsigned int front_len, middle_len, data_len;
2243 	bool do_datacrc = !con->msgr->nocrc;
2244 	u64 seq;
2245 	u32 crc;
2246 
2247 	dout("read_partial_message con %p msg %p\n", con, m);
2248 
2249 	/* header */
2250 	size = sizeof (con->in_hdr);
2251 	end = size;
2252 	ret = read_partial(con, end, size, &con->in_hdr);
2253 	if (ret <= 0)
2254 		return ret;
2255 
2256 	crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2257 	if (cpu_to_le32(crc) != con->in_hdr.crc) {
2258 		pr_err("read_partial_message bad hdr "
2259 		       " crc %u != expected %u\n",
2260 		       crc, con->in_hdr.crc);
2261 		return -EBADMSG;
2262 	}
2263 
2264 	front_len = le32_to_cpu(con->in_hdr.front_len);
2265 	if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2266 		return -EIO;
2267 	middle_len = le32_to_cpu(con->in_hdr.middle_len);
2268 	if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2269 		return -EIO;
2270 	data_len = le32_to_cpu(con->in_hdr.data_len);
2271 	if (data_len > CEPH_MSG_MAX_DATA_LEN)
2272 		return -EIO;
2273 
2274 	/* verify seq# */
2275 	seq = le64_to_cpu(con->in_hdr.seq);
2276 	if ((s64)seq - (s64)con->in_seq < 1) {
2277 		pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2278 			ENTITY_NAME(con->peer_name),
2279 			ceph_pr_addr(&con->peer_addr.in_addr),
2280 			seq, con->in_seq + 1);
2281 		con->in_base_pos = -front_len - middle_len - data_len -
2282 			sizeof(m->footer);
2283 		con->in_tag = CEPH_MSGR_TAG_READY;
2284 		return 0;
2285 	} else if ((s64)seq - (s64)con->in_seq > 1) {
2286 		pr_err("read_partial_message bad seq %lld expected %lld\n",
2287 		       seq, con->in_seq + 1);
2288 		con->error_msg = "bad message sequence # for incoming message";
2289 		return -EBADMSG;
2290 	}
2291 
2292 	/* allocate message? */
2293 	if (!con->in_msg) {
2294 		int skip = 0;
2295 
2296 		dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2297 		     front_len, data_len);
2298 		ret = ceph_con_in_msg_alloc(con, &skip);
2299 		if (ret < 0)
2300 			return ret;
2301 
2302 		BUG_ON(!con->in_msg ^ skip);
2303 		if (con->in_msg && data_len > con->in_msg->data_length) {
2304 			pr_warning("%s skipping long message (%u > %zd)\n",
2305 				__func__, data_len, con->in_msg->data_length);
2306 			ceph_msg_put(con->in_msg);
2307 			con->in_msg = NULL;
2308 			skip = 1;
2309 		}
2310 		if (skip) {
2311 			/* skip this message */
2312 			dout("alloc_msg said skip message\n");
2313 			con->in_base_pos = -front_len - middle_len - data_len -
2314 				sizeof(m->footer);
2315 			con->in_tag = CEPH_MSGR_TAG_READY;
2316 			con->in_seq++;
2317 			return 0;
2318 		}
2319 
2320 		BUG_ON(!con->in_msg);
2321 		BUG_ON(con->in_msg->con != con);
2322 		m = con->in_msg;
2323 		m->front.iov_len = 0;    /* haven't read it yet */
2324 		if (m->middle)
2325 			m->middle->vec.iov_len = 0;
2326 
2327 		/* prepare for data payload, if any */
2328 
2329 		if (data_len)
2330 			prepare_message_data(con->in_msg, data_len);
2331 	}
2332 
2333 	/* front */
2334 	ret = read_partial_message_section(con, &m->front, front_len,
2335 					   &con->in_front_crc);
2336 	if (ret <= 0)
2337 		return ret;
2338 
2339 	/* middle */
2340 	if (m->middle) {
2341 		ret = read_partial_message_section(con, &m->middle->vec,
2342 						   middle_len,
2343 						   &con->in_middle_crc);
2344 		if (ret <= 0)
2345 			return ret;
2346 	}
2347 
2348 	/* (page) data */
2349 	if (data_len) {
2350 		ret = read_partial_msg_data(con);
2351 		if (ret <= 0)
2352 			return ret;
2353 	}
2354 
2355 	/* footer */
2356 	size = sizeof (m->footer);
2357 	end += size;
2358 	ret = read_partial(con, end, size, &m->footer);
2359 	if (ret <= 0)
2360 		return ret;
2361 
2362 	dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2363 	     m, front_len, m->footer.front_crc, middle_len,
2364 	     m->footer.middle_crc, data_len, m->footer.data_crc);
2365 
2366 	/* crc ok? */
2367 	if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2368 		pr_err("read_partial_message %p front crc %u != exp. %u\n",
2369 		       m, con->in_front_crc, m->footer.front_crc);
2370 		return -EBADMSG;
2371 	}
2372 	if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2373 		pr_err("read_partial_message %p middle crc %u != exp %u\n",
2374 		       m, con->in_middle_crc, m->footer.middle_crc);
2375 		return -EBADMSG;
2376 	}
2377 	if (do_datacrc &&
2378 	    (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2379 	    con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2380 		pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2381 		       con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2382 		return -EBADMSG;
2383 	}
2384 
2385 	return 1; /* done! */
2386 }
2387 
2388 /*
2389  * Process message.  This happens in the worker thread.  The callback should
2390  * be careful not to do anything that waits on other incoming messages or it
2391  * may deadlock.
2392  */
2393 static void process_message(struct ceph_connection *con)
2394 {
2395 	struct ceph_msg *msg;
2396 
2397 	BUG_ON(con->in_msg->con != con);
2398 	con->in_msg->con = NULL;
2399 	msg = con->in_msg;
2400 	con->in_msg = NULL;
2401 	con->ops->put(con);
2402 
2403 	/* if first message, set peer_name */
2404 	if (con->peer_name.type == 0)
2405 		con->peer_name = msg->hdr.src;
2406 
2407 	con->in_seq++;
2408 	mutex_unlock(&con->mutex);
2409 
2410 	dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2411 	     msg, le64_to_cpu(msg->hdr.seq),
2412 	     ENTITY_NAME(msg->hdr.src),
2413 	     le16_to_cpu(msg->hdr.type),
2414 	     ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2415 	     le32_to_cpu(msg->hdr.front_len),
2416 	     le32_to_cpu(msg->hdr.data_len),
2417 	     con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2418 	con->ops->dispatch(con, msg);
2419 
2420 	mutex_lock(&con->mutex);
2421 }
2422 
2423 
2424 /*
2425  * Write something to the socket.  Called in a worker thread when the
2426  * socket appears to be writeable and we have something ready to send.
2427  */
2428 static int try_write(struct ceph_connection *con)
2429 {
2430 	int ret = 1;
2431 
2432 	dout("try_write start %p state %lu\n", con, con->state);
2433 
2434 more:
2435 	dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2436 
2437 	/* open the socket first? */
2438 	if (con->state == CON_STATE_PREOPEN) {
2439 		BUG_ON(con->sock);
2440 		con->state = CON_STATE_CONNECTING;
2441 
2442 		con_out_kvec_reset(con);
2443 		prepare_write_banner(con);
2444 		prepare_read_banner(con);
2445 
2446 		BUG_ON(con->in_msg);
2447 		con->in_tag = CEPH_MSGR_TAG_READY;
2448 		dout("try_write initiating connect on %p new state %lu\n",
2449 		     con, con->state);
2450 		ret = ceph_tcp_connect(con);
2451 		if (ret < 0) {
2452 			con->error_msg = "connect error";
2453 			goto out;
2454 		}
2455 	}
2456 
2457 more_kvec:
2458 	/* kvec data queued? */
2459 	if (con->out_skip) {
2460 		ret = write_partial_skip(con);
2461 		if (ret <= 0)
2462 			goto out;
2463 	}
2464 	if (con->out_kvec_left) {
2465 		ret = write_partial_kvec(con);
2466 		if (ret <= 0)
2467 			goto out;
2468 	}
2469 
2470 	/* msg pages? */
2471 	if (con->out_msg) {
2472 		if (con->out_msg_done) {
2473 			ceph_msg_put(con->out_msg);
2474 			con->out_msg = NULL;   /* we're done with this one */
2475 			goto do_next;
2476 		}
2477 
2478 		ret = write_partial_message_data(con);
2479 		if (ret == 1)
2480 			goto more_kvec;  /* we need to send the footer, too! */
2481 		if (ret == 0)
2482 			goto out;
2483 		if (ret < 0) {
2484 			dout("try_write write_partial_message_data err %d\n",
2485 			     ret);
2486 			goto out;
2487 		}
2488 	}
2489 
2490 do_next:
2491 	if (con->state == CON_STATE_OPEN) {
2492 		/* is anything else pending? */
2493 		if (!list_empty(&con->out_queue)) {
2494 			prepare_write_message(con);
2495 			goto more;
2496 		}
2497 		if (con->in_seq > con->in_seq_acked) {
2498 			prepare_write_ack(con);
2499 			goto more;
2500 		}
2501 		if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2502 			prepare_write_keepalive(con);
2503 			goto more;
2504 		}
2505 	}
2506 
2507 	/* Nothing to do! */
2508 	con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2509 	dout("try_write nothing else to write.\n");
2510 	ret = 0;
2511 out:
2512 	dout("try_write done on %p ret %d\n", con, ret);
2513 	return ret;
2514 }
2515 
2516 
2517 
2518 /*
2519  * Read what we can from the socket.
2520  */
2521 static int try_read(struct ceph_connection *con)
2522 {
2523 	int ret = -1;
2524 
2525 more:
2526 	dout("try_read start on %p state %lu\n", con, con->state);
2527 	if (con->state != CON_STATE_CONNECTING &&
2528 	    con->state != CON_STATE_NEGOTIATING &&
2529 	    con->state != CON_STATE_OPEN)
2530 		return 0;
2531 
2532 	BUG_ON(!con->sock);
2533 
2534 	dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2535 	     con->in_base_pos);
2536 
2537 	if (con->state == CON_STATE_CONNECTING) {
2538 		dout("try_read connecting\n");
2539 		ret = read_partial_banner(con);
2540 		if (ret <= 0)
2541 			goto out;
2542 		ret = process_banner(con);
2543 		if (ret < 0)
2544 			goto out;
2545 
2546 		con->state = CON_STATE_NEGOTIATING;
2547 
2548 		/*
2549 		 * Received banner is good, exchange connection info.
2550 		 * Do not reset out_kvec, as sending our banner raced
2551 		 * with receiving peer banner after connect completed.
2552 		 */
2553 		ret = prepare_write_connect(con);
2554 		if (ret < 0)
2555 			goto out;
2556 		prepare_read_connect(con);
2557 
2558 		/* Send connection info before awaiting response */
2559 		goto out;
2560 	}
2561 
2562 	if (con->state == CON_STATE_NEGOTIATING) {
2563 		dout("try_read negotiating\n");
2564 		ret = read_partial_connect(con);
2565 		if (ret <= 0)
2566 			goto out;
2567 		ret = process_connect(con);
2568 		if (ret < 0)
2569 			goto out;
2570 		goto more;
2571 	}
2572 
2573 	WARN_ON(con->state != CON_STATE_OPEN);
2574 
2575 	if (con->in_base_pos < 0) {
2576 		/*
2577 		 * skipping + discarding content.
2578 		 *
2579 		 * FIXME: there must be a better way to do this!
2580 		 */
2581 		static char buf[SKIP_BUF_SIZE];
2582 		int skip = min((int) sizeof (buf), -con->in_base_pos);
2583 
2584 		dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2585 		ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2586 		if (ret <= 0)
2587 			goto out;
2588 		con->in_base_pos += ret;
2589 		if (con->in_base_pos)
2590 			goto more;
2591 	}
2592 	if (con->in_tag == CEPH_MSGR_TAG_READY) {
2593 		/*
2594 		 * what's next?
2595 		 */
2596 		ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2597 		if (ret <= 0)
2598 			goto out;
2599 		dout("try_read got tag %d\n", (int)con->in_tag);
2600 		switch (con->in_tag) {
2601 		case CEPH_MSGR_TAG_MSG:
2602 			prepare_read_message(con);
2603 			break;
2604 		case CEPH_MSGR_TAG_ACK:
2605 			prepare_read_ack(con);
2606 			break;
2607 		case CEPH_MSGR_TAG_CLOSE:
2608 			con_close_socket(con);
2609 			con->state = CON_STATE_CLOSED;
2610 			goto out;
2611 		default:
2612 			goto bad_tag;
2613 		}
2614 	}
2615 	if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2616 		ret = read_partial_message(con);
2617 		if (ret <= 0) {
2618 			switch (ret) {
2619 			case -EBADMSG:
2620 				con->error_msg = "bad crc";
2621 				ret = -EIO;
2622 				break;
2623 			case -EIO:
2624 				con->error_msg = "io error";
2625 				break;
2626 			}
2627 			goto out;
2628 		}
2629 		if (con->in_tag == CEPH_MSGR_TAG_READY)
2630 			goto more;
2631 		process_message(con);
2632 		if (con->state == CON_STATE_OPEN)
2633 			prepare_read_tag(con);
2634 		goto more;
2635 	}
2636 	if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2637 	    con->in_tag == CEPH_MSGR_TAG_SEQ) {
2638 		/*
2639 		 * the final handshake seq exchange is semantically
2640 		 * equivalent to an ACK
2641 		 */
2642 		ret = read_partial_ack(con);
2643 		if (ret <= 0)
2644 			goto out;
2645 		process_ack(con);
2646 		goto more;
2647 	}
2648 
2649 out:
2650 	dout("try_read done on %p ret %d\n", con, ret);
2651 	return ret;
2652 
2653 bad_tag:
2654 	pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2655 	con->error_msg = "protocol error, garbage tag";
2656 	ret = -1;
2657 	goto out;
2658 }
2659 
2660 
2661 /*
2662  * Atomically queue work on a connection after the specified delay.
2663  * Bump @con reference to avoid races with connection teardown.
2664  * Returns 0 if work was queued, or an error code otherwise.
2665  */
2666 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2667 {
2668 	if (!con->ops->get(con)) {
2669 		dout("%s %p ref count 0\n", __func__, con);
2670 
2671 		return -ENOENT;
2672 	}
2673 
2674 	if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2675 		dout("%s %p - already queued\n", __func__, con);
2676 		con->ops->put(con);
2677 
2678 		return -EBUSY;
2679 	}
2680 
2681 	dout("%s %p %lu\n", __func__, con, delay);
2682 
2683 	return 0;
2684 }
2685 
2686 static void queue_con(struct ceph_connection *con)
2687 {
2688 	(void) queue_con_delay(con, 0);
2689 }
2690 
2691 static bool con_sock_closed(struct ceph_connection *con)
2692 {
2693 	if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2694 		return false;
2695 
2696 #define CASE(x)								\
2697 	case CON_STATE_ ## x:						\
2698 		con->error_msg = "socket closed (con state " #x ")";	\
2699 		break;
2700 
2701 	switch (con->state) {
2702 	CASE(CLOSED);
2703 	CASE(PREOPEN);
2704 	CASE(CONNECTING);
2705 	CASE(NEGOTIATING);
2706 	CASE(OPEN);
2707 	CASE(STANDBY);
2708 	default:
2709 		pr_warning("%s con %p unrecognized state %lu\n",
2710 			__func__, con, con->state);
2711 		con->error_msg = "unrecognized con state";
2712 		BUG();
2713 		break;
2714 	}
2715 #undef CASE
2716 
2717 	return true;
2718 }
2719 
2720 static bool con_backoff(struct ceph_connection *con)
2721 {
2722 	int ret;
2723 
2724 	if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2725 		return false;
2726 
2727 	ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2728 	if (ret) {
2729 		dout("%s: con %p FAILED to back off %lu\n", __func__,
2730 			con, con->delay);
2731 		BUG_ON(ret == -ENOENT);
2732 		con_flag_set(con, CON_FLAG_BACKOFF);
2733 	}
2734 
2735 	return true;
2736 }
2737 
2738 /* Finish fault handling; con->mutex must *not* be held here */
2739 
2740 static void con_fault_finish(struct ceph_connection *con)
2741 {
2742 	/*
2743 	 * in case we faulted due to authentication, invalidate our
2744 	 * current tickets so that we can get new ones.
2745 	 */
2746 	if (con->auth_retry && con->ops->invalidate_authorizer) {
2747 		dout("calling invalidate_authorizer()\n");
2748 		con->ops->invalidate_authorizer(con);
2749 	}
2750 
2751 	if (con->ops->fault)
2752 		con->ops->fault(con);
2753 }
2754 
2755 /*
2756  * Do some work on a connection.  Drop a connection ref when we're done.
2757  */
2758 static void con_work(struct work_struct *work)
2759 {
2760 	struct ceph_connection *con = container_of(work, struct ceph_connection,
2761 						   work.work);
2762 	bool fault;
2763 
2764 	mutex_lock(&con->mutex);
2765 	while (true) {
2766 		int ret;
2767 
2768 		if ((fault = con_sock_closed(con))) {
2769 			dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2770 			break;
2771 		}
2772 		if (con_backoff(con)) {
2773 			dout("%s: con %p BACKOFF\n", __func__, con);
2774 			break;
2775 		}
2776 		if (con->state == CON_STATE_STANDBY) {
2777 			dout("%s: con %p STANDBY\n", __func__, con);
2778 			break;
2779 		}
2780 		if (con->state == CON_STATE_CLOSED) {
2781 			dout("%s: con %p CLOSED\n", __func__, con);
2782 			BUG_ON(con->sock);
2783 			break;
2784 		}
2785 		if (con->state == CON_STATE_PREOPEN) {
2786 			dout("%s: con %p PREOPEN\n", __func__, con);
2787 			BUG_ON(con->sock);
2788 		}
2789 
2790 		ret = try_read(con);
2791 		if (ret < 0) {
2792 			if (ret == -EAGAIN)
2793 				continue;
2794 			con->error_msg = "socket error on read";
2795 			fault = true;
2796 			break;
2797 		}
2798 
2799 		ret = try_write(con);
2800 		if (ret < 0) {
2801 			if (ret == -EAGAIN)
2802 				continue;
2803 			con->error_msg = "socket error on write";
2804 			fault = true;
2805 		}
2806 
2807 		break;	/* If we make it to here, we're done */
2808 	}
2809 	if (fault)
2810 		con_fault(con);
2811 	mutex_unlock(&con->mutex);
2812 
2813 	if (fault)
2814 		con_fault_finish(con);
2815 
2816 	con->ops->put(con);
2817 }
2818 
2819 /*
2820  * Generic error/fault handler.  A retry mechanism is used with
2821  * exponential backoff
2822  */
2823 static void con_fault(struct ceph_connection *con)
2824 {
2825 	pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2826 	       ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2827 	dout("fault %p state %lu to peer %s\n",
2828 	     con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2829 
2830 	WARN_ON(con->state != CON_STATE_CONNECTING &&
2831 	       con->state != CON_STATE_NEGOTIATING &&
2832 	       con->state != CON_STATE_OPEN);
2833 
2834 	con_close_socket(con);
2835 
2836 	if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2837 		dout("fault on LOSSYTX channel, marking CLOSED\n");
2838 		con->state = CON_STATE_CLOSED;
2839 		return;
2840 	}
2841 
2842 	if (con->in_msg) {
2843 		BUG_ON(con->in_msg->con != con);
2844 		con->in_msg->con = NULL;
2845 		ceph_msg_put(con->in_msg);
2846 		con->in_msg = NULL;
2847 		con->ops->put(con);
2848 	}
2849 
2850 	/* Requeue anything that hasn't been acked */
2851 	list_splice_init(&con->out_sent, &con->out_queue);
2852 
2853 	/* If there are no messages queued or keepalive pending, place
2854 	 * the connection in a STANDBY state */
2855 	if (list_empty(&con->out_queue) &&
2856 	    !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
2857 		dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2858 		con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2859 		con->state = CON_STATE_STANDBY;
2860 	} else {
2861 		/* retry after a delay. */
2862 		con->state = CON_STATE_PREOPEN;
2863 		if (con->delay == 0)
2864 			con->delay = BASE_DELAY_INTERVAL;
2865 		else if (con->delay < MAX_DELAY_INTERVAL)
2866 			con->delay *= 2;
2867 		con_flag_set(con, CON_FLAG_BACKOFF);
2868 		queue_con(con);
2869 	}
2870 }
2871 
2872 
2873 
2874 /*
2875  * initialize a new messenger instance
2876  */
2877 void ceph_messenger_init(struct ceph_messenger *msgr,
2878 			struct ceph_entity_addr *myaddr,
2879 			u64 supported_features,
2880 			u64 required_features,
2881 			bool nocrc)
2882 {
2883 	msgr->supported_features = supported_features;
2884 	msgr->required_features = required_features;
2885 
2886 	spin_lock_init(&msgr->global_seq_lock);
2887 
2888 	if (myaddr)
2889 		msgr->inst.addr = *myaddr;
2890 
2891 	/* select a random nonce */
2892 	msgr->inst.addr.type = 0;
2893 	get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2894 	encode_my_addr(msgr);
2895 	msgr->nocrc = nocrc;
2896 
2897 	atomic_set(&msgr->stopping, 0);
2898 
2899 	dout("%s %p\n", __func__, msgr);
2900 }
2901 EXPORT_SYMBOL(ceph_messenger_init);
2902 
2903 static void clear_standby(struct ceph_connection *con)
2904 {
2905 	/* come back from STANDBY? */
2906 	if (con->state == CON_STATE_STANDBY) {
2907 		dout("clear_standby %p and ++connect_seq\n", con);
2908 		con->state = CON_STATE_PREOPEN;
2909 		con->connect_seq++;
2910 		WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2911 		WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
2912 	}
2913 }
2914 
2915 /*
2916  * Queue up an outgoing message on the given connection.
2917  */
2918 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2919 {
2920 	/* set src+dst */
2921 	msg->hdr.src = con->msgr->inst.name;
2922 	BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2923 	msg->needs_out_seq = true;
2924 
2925 	mutex_lock(&con->mutex);
2926 
2927 	if (con->state == CON_STATE_CLOSED) {
2928 		dout("con_send %p closed, dropping %p\n", con, msg);
2929 		ceph_msg_put(msg);
2930 		mutex_unlock(&con->mutex);
2931 		return;
2932 	}
2933 
2934 	BUG_ON(msg->con != NULL);
2935 	msg->con = con->ops->get(con);
2936 	BUG_ON(msg->con == NULL);
2937 
2938 	BUG_ON(!list_empty(&msg->list_head));
2939 	list_add_tail(&msg->list_head, &con->out_queue);
2940 	dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2941 	     ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2942 	     ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2943 	     le32_to_cpu(msg->hdr.front_len),
2944 	     le32_to_cpu(msg->hdr.middle_len),
2945 	     le32_to_cpu(msg->hdr.data_len));
2946 
2947 	clear_standby(con);
2948 	mutex_unlock(&con->mutex);
2949 
2950 	/* if there wasn't anything waiting to send before, queue
2951 	 * new work */
2952 	if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
2953 		queue_con(con);
2954 }
2955 EXPORT_SYMBOL(ceph_con_send);
2956 
2957 /*
2958  * Revoke a message that was previously queued for send
2959  */
2960 void ceph_msg_revoke(struct ceph_msg *msg)
2961 {
2962 	struct ceph_connection *con = msg->con;
2963 
2964 	if (!con)
2965 		return;		/* Message not in our possession */
2966 
2967 	mutex_lock(&con->mutex);
2968 	if (!list_empty(&msg->list_head)) {
2969 		dout("%s %p msg %p - was on queue\n", __func__, con, msg);
2970 		list_del_init(&msg->list_head);
2971 		BUG_ON(msg->con == NULL);
2972 		msg->con->ops->put(msg->con);
2973 		msg->con = NULL;
2974 		msg->hdr.seq = 0;
2975 
2976 		ceph_msg_put(msg);
2977 	}
2978 	if (con->out_msg == msg) {
2979 		dout("%s %p msg %p - was sending\n", __func__, con, msg);
2980 		con->out_msg = NULL;
2981 		if (con->out_kvec_is_msg) {
2982 			con->out_skip = con->out_kvec_bytes;
2983 			con->out_kvec_is_msg = false;
2984 		}
2985 		msg->hdr.seq = 0;
2986 
2987 		ceph_msg_put(msg);
2988 	}
2989 	mutex_unlock(&con->mutex);
2990 }
2991 
2992 /*
2993  * Revoke a message that we may be reading data into
2994  */
2995 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
2996 {
2997 	struct ceph_connection *con;
2998 
2999 	BUG_ON(msg == NULL);
3000 	if (!msg->con) {
3001 		dout("%s msg %p null con\n", __func__, msg);
3002 
3003 		return;		/* Message not in our possession */
3004 	}
3005 
3006 	con = msg->con;
3007 	mutex_lock(&con->mutex);
3008 	if (con->in_msg == msg) {
3009 		unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3010 		unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3011 		unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
3012 
3013 		/* skip rest of message */
3014 		dout("%s %p msg %p revoked\n", __func__, con, msg);
3015 		con->in_base_pos = con->in_base_pos -
3016 				sizeof(struct ceph_msg_header) -
3017 				front_len -
3018 				middle_len -
3019 				data_len -
3020 				sizeof(struct ceph_msg_footer);
3021 		ceph_msg_put(con->in_msg);
3022 		con->in_msg = NULL;
3023 		con->in_tag = CEPH_MSGR_TAG_READY;
3024 		con->in_seq++;
3025 	} else {
3026 		dout("%s %p in_msg %p msg %p no-op\n",
3027 		     __func__, con, con->in_msg, msg);
3028 	}
3029 	mutex_unlock(&con->mutex);
3030 }
3031 
3032 /*
3033  * Queue a keepalive byte to ensure the tcp connection is alive.
3034  */
3035 void ceph_con_keepalive(struct ceph_connection *con)
3036 {
3037 	dout("con_keepalive %p\n", con);
3038 	mutex_lock(&con->mutex);
3039 	clear_standby(con);
3040 	mutex_unlock(&con->mutex);
3041 	if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3042 	    con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3043 		queue_con(con);
3044 }
3045 EXPORT_SYMBOL(ceph_con_keepalive);
3046 
3047 static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
3048 {
3049 	struct ceph_msg_data *data;
3050 
3051 	if (WARN_ON(!ceph_msg_data_type_valid(type)))
3052 		return NULL;
3053 
3054 	data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
3055 	if (data)
3056 		data->type = type;
3057 	INIT_LIST_HEAD(&data->links);
3058 
3059 	return data;
3060 }
3061 
3062 static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3063 {
3064 	if (!data)
3065 		return;
3066 
3067 	WARN_ON(!list_empty(&data->links));
3068 	if (data->type == CEPH_MSG_DATA_PAGELIST) {
3069 		ceph_pagelist_release(data->pagelist);
3070 		kfree(data->pagelist);
3071 	}
3072 	kmem_cache_free(ceph_msg_data_cache, data);
3073 }
3074 
3075 void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
3076 		size_t length, size_t alignment)
3077 {
3078 	struct ceph_msg_data *data;
3079 
3080 	BUG_ON(!pages);
3081 	BUG_ON(!length);
3082 
3083 	data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3084 	BUG_ON(!data);
3085 	data->pages = pages;
3086 	data->length = length;
3087 	data->alignment = alignment & ~PAGE_MASK;
3088 
3089 	list_add_tail(&data->links, &msg->data);
3090 	msg->data_length += length;
3091 }
3092 EXPORT_SYMBOL(ceph_msg_data_add_pages);
3093 
3094 void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
3095 				struct ceph_pagelist *pagelist)
3096 {
3097 	struct ceph_msg_data *data;
3098 
3099 	BUG_ON(!pagelist);
3100 	BUG_ON(!pagelist->length);
3101 
3102 	data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3103 	BUG_ON(!data);
3104 	data->pagelist = pagelist;
3105 
3106 	list_add_tail(&data->links, &msg->data);
3107 	msg->data_length += pagelist->length;
3108 }
3109 EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
3110 
3111 #ifdef	CONFIG_BLOCK
3112 void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
3113 		size_t length)
3114 {
3115 	struct ceph_msg_data *data;
3116 
3117 	BUG_ON(!bio);
3118 
3119 	data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3120 	BUG_ON(!data);
3121 	data->bio = bio;
3122 	data->bio_length = length;
3123 
3124 	list_add_tail(&data->links, &msg->data);
3125 	msg->data_length += length;
3126 }
3127 EXPORT_SYMBOL(ceph_msg_data_add_bio);
3128 #endif	/* CONFIG_BLOCK */
3129 
3130 /*
3131  * construct a new message with given type, size
3132  * the new msg has a ref count of 1.
3133  */
3134 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3135 			      bool can_fail)
3136 {
3137 	struct ceph_msg *m;
3138 
3139 	m = kmem_cache_zalloc(ceph_msg_cache, flags);
3140 	if (m == NULL)
3141 		goto out;
3142 
3143 	m->hdr.type = cpu_to_le16(type);
3144 	m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3145 	m->hdr.front_len = cpu_to_le32(front_len);
3146 
3147 	INIT_LIST_HEAD(&m->list_head);
3148 	kref_init(&m->kref);
3149 	INIT_LIST_HEAD(&m->data);
3150 
3151 	/* front */
3152 	if (front_len) {
3153 		m->front.iov_base = ceph_kvmalloc(front_len, flags);
3154 		if (m->front.iov_base == NULL) {
3155 			dout("ceph_msg_new can't allocate %d bytes\n",
3156 			     front_len);
3157 			goto out2;
3158 		}
3159 	} else {
3160 		m->front.iov_base = NULL;
3161 	}
3162 	m->front_alloc_len = m->front.iov_len = front_len;
3163 
3164 	dout("ceph_msg_new %p front %d\n", m, front_len);
3165 	return m;
3166 
3167 out2:
3168 	ceph_msg_put(m);
3169 out:
3170 	if (!can_fail) {
3171 		pr_err("msg_new can't create type %d front %d\n", type,
3172 		       front_len);
3173 		WARN_ON(1);
3174 	} else {
3175 		dout("msg_new can't create type %d front %d\n", type,
3176 		     front_len);
3177 	}
3178 	return NULL;
3179 }
3180 EXPORT_SYMBOL(ceph_msg_new);
3181 
3182 /*
3183  * Allocate "middle" portion of a message, if it is needed and wasn't
3184  * allocated by alloc_msg.  This allows us to read a small fixed-size
3185  * per-type header in the front and then gracefully fail (i.e.,
3186  * propagate the error to the caller based on info in the front) when
3187  * the middle is too large.
3188  */
3189 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3190 {
3191 	int type = le16_to_cpu(msg->hdr.type);
3192 	int middle_len = le32_to_cpu(msg->hdr.middle_len);
3193 
3194 	dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3195 	     ceph_msg_type_name(type), middle_len);
3196 	BUG_ON(!middle_len);
3197 	BUG_ON(msg->middle);
3198 
3199 	msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3200 	if (!msg->middle)
3201 		return -ENOMEM;
3202 	return 0;
3203 }
3204 
3205 /*
3206  * Allocate a message for receiving an incoming message on a
3207  * connection, and save the result in con->in_msg.  Uses the
3208  * connection's private alloc_msg op if available.
3209  *
3210  * Returns 0 on success, or a negative error code.
3211  *
3212  * On success, if we set *skip = 1:
3213  *  - the next message should be skipped and ignored.
3214  *  - con->in_msg == NULL
3215  * or if we set *skip = 0:
3216  *  - con->in_msg is non-null.
3217  * On error (ENOMEM, EAGAIN, ...),
3218  *  - con->in_msg == NULL
3219  */
3220 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3221 {
3222 	struct ceph_msg_header *hdr = &con->in_hdr;
3223 	int middle_len = le32_to_cpu(hdr->middle_len);
3224 	struct ceph_msg *msg;
3225 	int ret = 0;
3226 
3227 	BUG_ON(con->in_msg != NULL);
3228 	BUG_ON(!con->ops->alloc_msg);
3229 
3230 	mutex_unlock(&con->mutex);
3231 	msg = con->ops->alloc_msg(con, hdr, skip);
3232 	mutex_lock(&con->mutex);
3233 	if (con->state != CON_STATE_OPEN) {
3234 		if (msg)
3235 			ceph_msg_put(msg);
3236 		return -EAGAIN;
3237 	}
3238 	if (msg) {
3239 		BUG_ON(*skip);
3240 		con->in_msg = msg;
3241 		con->in_msg->con = con->ops->get(con);
3242 		BUG_ON(con->in_msg->con == NULL);
3243 	} else {
3244 		/*
3245 		 * Null message pointer means either we should skip
3246 		 * this message or we couldn't allocate memory.  The
3247 		 * former is not an error.
3248 		 */
3249 		if (*skip)
3250 			return 0;
3251 		con->error_msg = "error allocating memory for incoming message";
3252 
3253 		return -ENOMEM;
3254 	}
3255 	memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3256 
3257 	if (middle_len && !con->in_msg->middle) {
3258 		ret = ceph_alloc_middle(con, con->in_msg);
3259 		if (ret < 0) {
3260 			ceph_msg_put(con->in_msg);
3261 			con->in_msg = NULL;
3262 		}
3263 	}
3264 
3265 	return ret;
3266 }
3267 
3268 
3269 /*
3270  * Free a generically kmalloc'd message.
3271  */
3272 void ceph_msg_kfree(struct ceph_msg *m)
3273 {
3274 	dout("msg_kfree %p\n", m);
3275 	ceph_kvfree(m->front.iov_base);
3276 	kmem_cache_free(ceph_msg_cache, m);
3277 }
3278 
3279 /*
3280  * Drop a msg ref.  Destroy as needed.
3281  */
3282 void ceph_msg_last_put(struct kref *kref)
3283 {
3284 	struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3285 	LIST_HEAD(data);
3286 	struct list_head *links;
3287 	struct list_head *next;
3288 
3289 	dout("ceph_msg_put last one on %p\n", m);
3290 	WARN_ON(!list_empty(&m->list_head));
3291 
3292 	/* drop middle, data, if any */
3293 	if (m->middle) {
3294 		ceph_buffer_put(m->middle);
3295 		m->middle = NULL;
3296 	}
3297 
3298 	list_splice_init(&m->data, &data);
3299 	list_for_each_safe(links, next, &data) {
3300 		struct ceph_msg_data *data;
3301 
3302 		data = list_entry(links, struct ceph_msg_data, links);
3303 		list_del_init(links);
3304 		ceph_msg_data_destroy(data);
3305 	}
3306 	m->data_length = 0;
3307 
3308 	if (m->pool)
3309 		ceph_msgpool_put(m->pool, m);
3310 	else
3311 		ceph_msg_kfree(m);
3312 }
3313 EXPORT_SYMBOL(ceph_msg_last_put);
3314 
3315 void ceph_msg_dump(struct ceph_msg *msg)
3316 {
3317 	pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3318 		 msg->front_alloc_len, msg->data_length);
3319 	print_hex_dump(KERN_DEBUG, "header: ",
3320 		       DUMP_PREFIX_OFFSET, 16, 1,
3321 		       &msg->hdr, sizeof(msg->hdr), true);
3322 	print_hex_dump(KERN_DEBUG, " front: ",
3323 		       DUMP_PREFIX_OFFSET, 16, 1,
3324 		       msg->front.iov_base, msg->front.iov_len, true);
3325 	if (msg->middle)
3326 		print_hex_dump(KERN_DEBUG, "middle: ",
3327 			       DUMP_PREFIX_OFFSET, 16, 1,
3328 			       msg->middle->vec.iov_base,
3329 			       msg->middle->vec.iov_len, true);
3330 	print_hex_dump(KERN_DEBUG, "footer: ",
3331 		       DUMP_PREFIX_OFFSET, 16, 1,
3332 		       &msg->footer, sizeof(msg->footer), true);
3333 }
3334 EXPORT_SYMBOL(ceph_msg_dump);
3335