xref: /linux/net/ceph/messenger.c (revision 3f0a50f345f78183f6e9b39c2f45ca5dcaa511ca)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
3 
4 #include <linux/crc32c.h>
5 #include <linux/ctype.h>
6 #include <linux/highmem.h>
7 #include <linux/inet.h>
8 #include <linux/kthread.h>
9 #include <linux/net.h>
10 #include <linux/nsproxy.h>
11 #include <linux/sched/mm.h>
12 #include <linux/slab.h>
13 #include <linux/socket.h>
14 #include <linux/string.h>
15 #ifdef	CONFIG_BLOCK
16 #include <linux/bio.h>
17 #endif	/* CONFIG_BLOCK */
18 #include <linux/dns_resolver.h>
19 #include <net/tcp.h>
20 
21 #include <linux/ceph/ceph_features.h>
22 #include <linux/ceph/libceph.h>
23 #include <linux/ceph/messenger.h>
24 #include <linux/ceph/decode.h>
25 #include <linux/ceph/pagelist.h>
26 #include <linux/export.h>
27 
28 /*
29  * Ceph uses the messenger to exchange ceph_msg messages with other
30  * hosts in the system.  The messenger provides ordered and reliable
31  * delivery.  We tolerate TCP disconnects by reconnecting (with
32  * exponential backoff) in the case of a fault (disconnection, bad
33  * crc, protocol error).  Acks allow sent messages to be discarded by
34  * the sender.
35  */
36 
37 /*
38  * We track the state of the socket on a given connection using
39  * values defined below.  The transition to a new socket state is
40  * handled by a function which verifies we aren't coming from an
41  * unexpected state.
42  *
43  *      --------
44  *      | NEW* |  transient initial state
45  *      --------
46  *          | con_sock_state_init()
47  *          v
48  *      ----------
49  *      | CLOSED |  initialized, but no socket (and no
50  *      ----------  TCP connection)
51  *       ^      \
52  *       |       \ con_sock_state_connecting()
53  *       |        ----------------------
54  *       |                              \
55  *       + con_sock_state_closed()       \
56  *       |+---------------------------    \
57  *       | \                          \    \
58  *       |  -----------                \    \
59  *       |  | CLOSING |  socket event;  \    \
60  *       |  -----------  await close     \    \
61  *       |       ^                        \   |
62  *       |       |                         \  |
63  *       |       + con_sock_state_closing() \ |
64  *       |      / \                         | |
65  *       |     /   ---------------          | |
66  *       |    /                   \         v v
67  *       |   /                    --------------
68  *       |  /    -----------------| CONNECTING |  socket created, TCP
69  *       |  |   /                 --------------  connect initiated
70  *       |  |   | con_sock_state_connected()
71  *       |  |   v
72  *      -------------
73  *      | CONNECTED |  TCP connection established
74  *      -------------
75  *
76  * State values for ceph_connection->sock_state; NEW is assumed to be 0.
77  */
78 
79 #define CON_SOCK_STATE_NEW		0	/* -> CLOSED */
80 #define CON_SOCK_STATE_CLOSED		1	/* -> CONNECTING */
81 #define CON_SOCK_STATE_CONNECTING	2	/* -> CONNECTED or -> CLOSING */
82 #define CON_SOCK_STATE_CONNECTED	3	/* -> CLOSING or -> CLOSED */
83 #define CON_SOCK_STATE_CLOSING		4	/* -> CLOSED */
84 
85 static bool con_flag_valid(unsigned long con_flag)
86 {
87 	switch (con_flag) {
88 	case CEPH_CON_F_LOSSYTX:
89 	case CEPH_CON_F_KEEPALIVE_PENDING:
90 	case CEPH_CON_F_WRITE_PENDING:
91 	case CEPH_CON_F_SOCK_CLOSED:
92 	case CEPH_CON_F_BACKOFF:
93 		return true;
94 	default:
95 		return false;
96 	}
97 }
98 
99 void ceph_con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
100 {
101 	BUG_ON(!con_flag_valid(con_flag));
102 
103 	clear_bit(con_flag, &con->flags);
104 }
105 
106 void ceph_con_flag_set(struct ceph_connection *con, unsigned long con_flag)
107 {
108 	BUG_ON(!con_flag_valid(con_flag));
109 
110 	set_bit(con_flag, &con->flags);
111 }
112 
113 bool ceph_con_flag_test(struct ceph_connection *con, unsigned long con_flag)
114 {
115 	BUG_ON(!con_flag_valid(con_flag));
116 
117 	return test_bit(con_flag, &con->flags);
118 }
119 
120 bool ceph_con_flag_test_and_clear(struct ceph_connection *con,
121 				  unsigned long con_flag)
122 {
123 	BUG_ON(!con_flag_valid(con_flag));
124 
125 	return test_and_clear_bit(con_flag, &con->flags);
126 }
127 
128 bool ceph_con_flag_test_and_set(struct ceph_connection *con,
129 				unsigned long con_flag)
130 {
131 	BUG_ON(!con_flag_valid(con_flag));
132 
133 	return test_and_set_bit(con_flag, &con->flags);
134 }
135 
136 /* Slab caches for frequently-allocated structures */
137 
138 static struct kmem_cache	*ceph_msg_cache;
139 
140 #ifdef CONFIG_LOCKDEP
141 static struct lock_class_key socket_class;
142 #endif
143 
144 static void queue_con(struct ceph_connection *con);
145 static void cancel_con(struct ceph_connection *con);
146 static void ceph_con_workfn(struct work_struct *);
147 static void con_fault(struct ceph_connection *con);
148 
149 /*
150  * Nicely render a sockaddr as a string.  An array of formatted
151  * strings is used, to approximate reentrancy.
152  */
153 #define ADDR_STR_COUNT_LOG	5	/* log2(# address strings in array) */
154 #define ADDR_STR_COUNT		(1 << ADDR_STR_COUNT_LOG)
155 #define ADDR_STR_COUNT_MASK	(ADDR_STR_COUNT - 1)
156 #define MAX_ADDR_STR_LEN	64	/* 54 is enough */
157 
158 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
159 static atomic_t addr_str_seq = ATOMIC_INIT(0);
160 
161 struct page *ceph_zero_page;		/* used in certain error cases */
162 
163 const char *ceph_pr_addr(const struct ceph_entity_addr *addr)
164 {
165 	int i;
166 	char *s;
167 	struct sockaddr_storage ss = addr->in_addr; /* align */
168 	struct sockaddr_in *in4 = (struct sockaddr_in *)&ss;
169 	struct sockaddr_in6 *in6 = (struct sockaddr_in6 *)&ss;
170 
171 	i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
172 	s = addr_str[i];
173 
174 	switch (ss.ss_family) {
175 	case AF_INET:
176 		snprintf(s, MAX_ADDR_STR_LEN, "(%d)%pI4:%hu",
177 			 le32_to_cpu(addr->type), &in4->sin_addr,
178 			 ntohs(in4->sin_port));
179 		break;
180 
181 	case AF_INET6:
182 		snprintf(s, MAX_ADDR_STR_LEN, "(%d)[%pI6c]:%hu",
183 			 le32_to_cpu(addr->type), &in6->sin6_addr,
184 			 ntohs(in6->sin6_port));
185 		break;
186 
187 	default:
188 		snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
189 			 ss.ss_family);
190 	}
191 
192 	return s;
193 }
194 EXPORT_SYMBOL(ceph_pr_addr);
195 
196 void ceph_encode_my_addr(struct ceph_messenger *msgr)
197 {
198 	if (!ceph_msgr2(from_msgr(msgr))) {
199 		memcpy(&msgr->my_enc_addr, &msgr->inst.addr,
200 		       sizeof(msgr->my_enc_addr));
201 		ceph_encode_banner_addr(&msgr->my_enc_addr);
202 	}
203 }
204 
205 /*
206  * work queue for all reading and writing to/from the socket.
207  */
208 static struct workqueue_struct *ceph_msgr_wq;
209 
210 static int ceph_msgr_slab_init(void)
211 {
212 	BUG_ON(ceph_msg_cache);
213 	ceph_msg_cache = KMEM_CACHE(ceph_msg, 0);
214 	if (!ceph_msg_cache)
215 		return -ENOMEM;
216 
217 	return 0;
218 }
219 
220 static void ceph_msgr_slab_exit(void)
221 {
222 	BUG_ON(!ceph_msg_cache);
223 	kmem_cache_destroy(ceph_msg_cache);
224 	ceph_msg_cache = NULL;
225 }
226 
227 static void _ceph_msgr_exit(void)
228 {
229 	if (ceph_msgr_wq) {
230 		destroy_workqueue(ceph_msgr_wq);
231 		ceph_msgr_wq = NULL;
232 	}
233 
234 	BUG_ON(!ceph_zero_page);
235 	put_page(ceph_zero_page);
236 	ceph_zero_page = NULL;
237 
238 	ceph_msgr_slab_exit();
239 }
240 
241 int __init ceph_msgr_init(void)
242 {
243 	if (ceph_msgr_slab_init())
244 		return -ENOMEM;
245 
246 	BUG_ON(ceph_zero_page);
247 	ceph_zero_page = ZERO_PAGE(0);
248 	get_page(ceph_zero_page);
249 
250 	/*
251 	 * The number of active work items is limited by the number of
252 	 * connections, so leave @max_active at default.
253 	 */
254 	ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
255 	if (ceph_msgr_wq)
256 		return 0;
257 
258 	pr_err("msgr_init failed to create workqueue\n");
259 	_ceph_msgr_exit();
260 
261 	return -ENOMEM;
262 }
263 
264 void ceph_msgr_exit(void)
265 {
266 	BUG_ON(ceph_msgr_wq == NULL);
267 
268 	_ceph_msgr_exit();
269 }
270 
271 void ceph_msgr_flush(void)
272 {
273 	flush_workqueue(ceph_msgr_wq);
274 }
275 EXPORT_SYMBOL(ceph_msgr_flush);
276 
277 /* Connection socket state transition functions */
278 
279 static void con_sock_state_init(struct ceph_connection *con)
280 {
281 	int old_state;
282 
283 	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
284 	if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
285 		printk("%s: unexpected old state %d\n", __func__, old_state);
286 	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
287 	     CON_SOCK_STATE_CLOSED);
288 }
289 
290 static void con_sock_state_connecting(struct ceph_connection *con)
291 {
292 	int old_state;
293 
294 	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
295 	if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
296 		printk("%s: unexpected old state %d\n", __func__, old_state);
297 	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
298 	     CON_SOCK_STATE_CONNECTING);
299 }
300 
301 static void con_sock_state_connected(struct ceph_connection *con)
302 {
303 	int old_state;
304 
305 	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
306 	if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
307 		printk("%s: unexpected old state %d\n", __func__, old_state);
308 	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
309 	     CON_SOCK_STATE_CONNECTED);
310 }
311 
312 static void con_sock_state_closing(struct ceph_connection *con)
313 {
314 	int old_state;
315 
316 	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
317 	if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
318 			old_state != CON_SOCK_STATE_CONNECTED &&
319 			old_state != CON_SOCK_STATE_CLOSING))
320 		printk("%s: unexpected old state %d\n", __func__, old_state);
321 	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
322 	     CON_SOCK_STATE_CLOSING);
323 }
324 
325 static void con_sock_state_closed(struct ceph_connection *con)
326 {
327 	int old_state;
328 
329 	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
330 	if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
331 		    old_state != CON_SOCK_STATE_CLOSING &&
332 		    old_state != CON_SOCK_STATE_CONNECTING &&
333 		    old_state != CON_SOCK_STATE_CLOSED))
334 		printk("%s: unexpected old state %d\n", __func__, old_state);
335 	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
336 	     CON_SOCK_STATE_CLOSED);
337 }
338 
339 /*
340  * socket callback functions
341  */
342 
343 /* data available on socket, or listen socket received a connect */
344 static void ceph_sock_data_ready(struct sock *sk)
345 {
346 	struct ceph_connection *con = sk->sk_user_data;
347 	if (atomic_read(&con->msgr->stopping)) {
348 		return;
349 	}
350 
351 	if (sk->sk_state != TCP_CLOSE_WAIT) {
352 		dout("%s %p state = %d, queueing work\n", __func__,
353 		     con, con->state);
354 		queue_con(con);
355 	}
356 }
357 
358 /* socket has buffer space for writing */
359 static void ceph_sock_write_space(struct sock *sk)
360 {
361 	struct ceph_connection *con = sk->sk_user_data;
362 
363 	/* only queue to workqueue if there is data we want to write,
364 	 * and there is sufficient space in the socket buffer to accept
365 	 * more data.  clear SOCK_NOSPACE so that ceph_sock_write_space()
366 	 * doesn't get called again until try_write() fills the socket
367 	 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
368 	 * and net/core/stream.c:sk_stream_write_space().
369 	 */
370 	if (ceph_con_flag_test(con, CEPH_CON_F_WRITE_PENDING)) {
371 		if (sk_stream_is_writeable(sk)) {
372 			dout("%s %p queueing write work\n", __func__, con);
373 			clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
374 			queue_con(con);
375 		}
376 	} else {
377 		dout("%s %p nothing to write\n", __func__, con);
378 	}
379 }
380 
381 /* socket's state has changed */
382 static void ceph_sock_state_change(struct sock *sk)
383 {
384 	struct ceph_connection *con = sk->sk_user_data;
385 
386 	dout("%s %p state = %d sk_state = %u\n", __func__,
387 	     con, con->state, sk->sk_state);
388 
389 	switch (sk->sk_state) {
390 	case TCP_CLOSE:
391 		dout("%s TCP_CLOSE\n", __func__);
392 		fallthrough;
393 	case TCP_CLOSE_WAIT:
394 		dout("%s TCP_CLOSE_WAIT\n", __func__);
395 		con_sock_state_closing(con);
396 		ceph_con_flag_set(con, CEPH_CON_F_SOCK_CLOSED);
397 		queue_con(con);
398 		break;
399 	case TCP_ESTABLISHED:
400 		dout("%s TCP_ESTABLISHED\n", __func__);
401 		con_sock_state_connected(con);
402 		queue_con(con);
403 		break;
404 	default:	/* Everything else is uninteresting */
405 		break;
406 	}
407 }
408 
409 /*
410  * set up socket callbacks
411  */
412 static void set_sock_callbacks(struct socket *sock,
413 			       struct ceph_connection *con)
414 {
415 	struct sock *sk = sock->sk;
416 	sk->sk_user_data = con;
417 	sk->sk_data_ready = ceph_sock_data_ready;
418 	sk->sk_write_space = ceph_sock_write_space;
419 	sk->sk_state_change = ceph_sock_state_change;
420 }
421 
422 
423 /*
424  * socket helpers
425  */
426 
427 /*
428  * initiate connection to a remote socket.
429  */
430 int ceph_tcp_connect(struct ceph_connection *con)
431 {
432 	struct sockaddr_storage ss = con->peer_addr.in_addr; /* align */
433 	struct socket *sock;
434 	unsigned int noio_flag;
435 	int ret;
436 
437 	dout("%s con %p peer_addr %s\n", __func__, con,
438 	     ceph_pr_addr(&con->peer_addr));
439 	BUG_ON(con->sock);
440 
441 	/* sock_create_kern() allocates with GFP_KERNEL */
442 	noio_flag = memalloc_noio_save();
443 	ret = sock_create_kern(read_pnet(&con->msgr->net), ss.ss_family,
444 			       SOCK_STREAM, IPPROTO_TCP, &sock);
445 	memalloc_noio_restore(noio_flag);
446 	if (ret)
447 		return ret;
448 	sock->sk->sk_allocation = GFP_NOFS;
449 
450 #ifdef CONFIG_LOCKDEP
451 	lockdep_set_class(&sock->sk->sk_lock, &socket_class);
452 #endif
453 
454 	set_sock_callbacks(sock, con);
455 
456 	con_sock_state_connecting(con);
457 	ret = sock->ops->connect(sock, (struct sockaddr *)&ss, sizeof(ss),
458 				 O_NONBLOCK);
459 	if (ret == -EINPROGRESS) {
460 		dout("connect %s EINPROGRESS sk_state = %u\n",
461 		     ceph_pr_addr(&con->peer_addr),
462 		     sock->sk->sk_state);
463 	} else if (ret < 0) {
464 		pr_err("connect %s error %d\n",
465 		       ceph_pr_addr(&con->peer_addr), ret);
466 		sock_release(sock);
467 		return ret;
468 	}
469 
470 	if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY))
471 		tcp_sock_set_nodelay(sock->sk);
472 
473 	con->sock = sock;
474 	return 0;
475 }
476 
477 /*
478  * Shutdown/close the socket for the given connection.
479  */
480 int ceph_con_close_socket(struct ceph_connection *con)
481 {
482 	int rc = 0;
483 
484 	dout("%s con %p sock %p\n", __func__, con, con->sock);
485 	if (con->sock) {
486 		rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
487 		sock_release(con->sock);
488 		con->sock = NULL;
489 	}
490 
491 	/*
492 	 * Forcibly clear the SOCK_CLOSED flag.  It gets set
493 	 * independent of the connection mutex, and we could have
494 	 * received a socket close event before we had the chance to
495 	 * shut the socket down.
496 	 */
497 	ceph_con_flag_clear(con, CEPH_CON_F_SOCK_CLOSED);
498 
499 	con_sock_state_closed(con);
500 	return rc;
501 }
502 
503 static void ceph_con_reset_protocol(struct ceph_connection *con)
504 {
505 	dout("%s con %p\n", __func__, con);
506 
507 	ceph_con_close_socket(con);
508 	if (con->in_msg) {
509 		WARN_ON(con->in_msg->con != con);
510 		ceph_msg_put(con->in_msg);
511 		con->in_msg = NULL;
512 	}
513 	if (con->out_msg) {
514 		WARN_ON(con->out_msg->con != con);
515 		ceph_msg_put(con->out_msg);
516 		con->out_msg = NULL;
517 	}
518 	if (con->bounce_page) {
519 		__free_page(con->bounce_page);
520 		con->bounce_page = NULL;
521 	}
522 
523 	if (ceph_msgr2(from_msgr(con->msgr)))
524 		ceph_con_v2_reset_protocol(con);
525 	else
526 		ceph_con_v1_reset_protocol(con);
527 }
528 
529 /*
530  * Reset a connection.  Discard all incoming and outgoing messages
531  * and clear *_seq state.
532  */
533 static void ceph_msg_remove(struct ceph_msg *msg)
534 {
535 	list_del_init(&msg->list_head);
536 
537 	ceph_msg_put(msg);
538 }
539 
540 static void ceph_msg_remove_list(struct list_head *head)
541 {
542 	while (!list_empty(head)) {
543 		struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
544 							list_head);
545 		ceph_msg_remove(msg);
546 	}
547 }
548 
549 void ceph_con_reset_session(struct ceph_connection *con)
550 {
551 	dout("%s con %p\n", __func__, con);
552 
553 	WARN_ON(con->in_msg);
554 	WARN_ON(con->out_msg);
555 	ceph_msg_remove_list(&con->out_queue);
556 	ceph_msg_remove_list(&con->out_sent);
557 	con->out_seq = 0;
558 	con->in_seq = 0;
559 	con->in_seq_acked = 0;
560 
561 	if (ceph_msgr2(from_msgr(con->msgr)))
562 		ceph_con_v2_reset_session(con);
563 	else
564 		ceph_con_v1_reset_session(con);
565 }
566 
567 /*
568  * mark a peer down.  drop any open connections.
569  */
570 void ceph_con_close(struct ceph_connection *con)
571 {
572 	mutex_lock(&con->mutex);
573 	dout("con_close %p peer %s\n", con, ceph_pr_addr(&con->peer_addr));
574 	con->state = CEPH_CON_S_CLOSED;
575 
576 	ceph_con_flag_clear(con, CEPH_CON_F_LOSSYTX);  /* so we retry next
577 							  connect */
578 	ceph_con_flag_clear(con, CEPH_CON_F_KEEPALIVE_PENDING);
579 	ceph_con_flag_clear(con, CEPH_CON_F_WRITE_PENDING);
580 	ceph_con_flag_clear(con, CEPH_CON_F_BACKOFF);
581 
582 	ceph_con_reset_protocol(con);
583 	ceph_con_reset_session(con);
584 	cancel_con(con);
585 	mutex_unlock(&con->mutex);
586 }
587 EXPORT_SYMBOL(ceph_con_close);
588 
589 /*
590  * Reopen a closed connection, with a new peer address.
591  */
592 void ceph_con_open(struct ceph_connection *con,
593 		   __u8 entity_type, __u64 entity_num,
594 		   struct ceph_entity_addr *addr)
595 {
596 	mutex_lock(&con->mutex);
597 	dout("con_open %p %s\n", con, ceph_pr_addr(addr));
598 
599 	WARN_ON(con->state != CEPH_CON_S_CLOSED);
600 	con->state = CEPH_CON_S_PREOPEN;
601 
602 	con->peer_name.type = (__u8) entity_type;
603 	con->peer_name.num = cpu_to_le64(entity_num);
604 
605 	memcpy(&con->peer_addr, addr, sizeof(*addr));
606 	con->delay = 0;      /* reset backoff memory */
607 	mutex_unlock(&con->mutex);
608 	queue_con(con);
609 }
610 EXPORT_SYMBOL(ceph_con_open);
611 
612 /*
613  * return true if this connection ever successfully opened
614  */
615 bool ceph_con_opened(struct ceph_connection *con)
616 {
617 	if (ceph_msgr2(from_msgr(con->msgr)))
618 		return ceph_con_v2_opened(con);
619 
620 	return ceph_con_v1_opened(con);
621 }
622 
623 /*
624  * initialize a new connection.
625  */
626 void ceph_con_init(struct ceph_connection *con, void *private,
627 	const struct ceph_connection_operations *ops,
628 	struct ceph_messenger *msgr)
629 {
630 	dout("con_init %p\n", con);
631 	memset(con, 0, sizeof(*con));
632 	con->private = private;
633 	con->ops = ops;
634 	con->msgr = msgr;
635 
636 	con_sock_state_init(con);
637 
638 	mutex_init(&con->mutex);
639 	INIT_LIST_HEAD(&con->out_queue);
640 	INIT_LIST_HEAD(&con->out_sent);
641 	INIT_DELAYED_WORK(&con->work, ceph_con_workfn);
642 
643 	con->state = CEPH_CON_S_CLOSED;
644 }
645 EXPORT_SYMBOL(ceph_con_init);
646 
647 /*
648  * We maintain a global counter to order connection attempts.  Get
649  * a unique seq greater than @gt.
650  */
651 u32 ceph_get_global_seq(struct ceph_messenger *msgr, u32 gt)
652 {
653 	u32 ret;
654 
655 	spin_lock(&msgr->global_seq_lock);
656 	if (msgr->global_seq < gt)
657 		msgr->global_seq = gt;
658 	ret = ++msgr->global_seq;
659 	spin_unlock(&msgr->global_seq_lock);
660 	return ret;
661 }
662 
663 /*
664  * Discard messages that have been acked by the server.
665  */
666 void ceph_con_discard_sent(struct ceph_connection *con, u64 ack_seq)
667 {
668 	struct ceph_msg *msg;
669 	u64 seq;
670 
671 	dout("%s con %p ack_seq %llu\n", __func__, con, ack_seq);
672 	while (!list_empty(&con->out_sent)) {
673 		msg = list_first_entry(&con->out_sent, struct ceph_msg,
674 				       list_head);
675 		WARN_ON(msg->needs_out_seq);
676 		seq = le64_to_cpu(msg->hdr.seq);
677 		if (seq > ack_seq)
678 			break;
679 
680 		dout("%s con %p discarding msg %p seq %llu\n", __func__, con,
681 		     msg, seq);
682 		ceph_msg_remove(msg);
683 	}
684 }
685 
686 /*
687  * Discard messages that have been requeued in con_fault(), up to
688  * reconnect_seq.  This avoids gratuitously resending messages that
689  * the server had received and handled prior to reconnect.
690  */
691 void ceph_con_discard_requeued(struct ceph_connection *con, u64 reconnect_seq)
692 {
693 	struct ceph_msg *msg;
694 	u64 seq;
695 
696 	dout("%s con %p reconnect_seq %llu\n", __func__, con, reconnect_seq);
697 	while (!list_empty(&con->out_queue)) {
698 		msg = list_first_entry(&con->out_queue, struct ceph_msg,
699 				       list_head);
700 		if (msg->needs_out_seq)
701 			break;
702 		seq = le64_to_cpu(msg->hdr.seq);
703 		if (seq > reconnect_seq)
704 			break;
705 
706 		dout("%s con %p discarding msg %p seq %llu\n", __func__, con,
707 		     msg, seq);
708 		ceph_msg_remove(msg);
709 	}
710 }
711 
712 #ifdef CONFIG_BLOCK
713 
714 /*
715  * For a bio data item, a piece is whatever remains of the next
716  * entry in the current bio iovec, or the first entry in the next
717  * bio in the list.
718  */
719 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
720 					size_t length)
721 {
722 	struct ceph_msg_data *data = cursor->data;
723 	struct ceph_bio_iter *it = &cursor->bio_iter;
724 
725 	cursor->resid = min_t(size_t, length, data->bio_length);
726 	*it = data->bio_pos;
727 	if (cursor->resid < it->iter.bi_size)
728 		it->iter.bi_size = cursor->resid;
729 
730 	BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
731 	cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter);
732 }
733 
734 static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
735 						size_t *page_offset,
736 						size_t *length)
737 {
738 	struct bio_vec bv = bio_iter_iovec(cursor->bio_iter.bio,
739 					   cursor->bio_iter.iter);
740 
741 	*page_offset = bv.bv_offset;
742 	*length = bv.bv_len;
743 	return bv.bv_page;
744 }
745 
746 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
747 					size_t bytes)
748 {
749 	struct ceph_bio_iter *it = &cursor->bio_iter;
750 	struct page *page = bio_iter_page(it->bio, it->iter);
751 
752 	BUG_ON(bytes > cursor->resid);
753 	BUG_ON(bytes > bio_iter_len(it->bio, it->iter));
754 	cursor->resid -= bytes;
755 	bio_advance_iter(it->bio, &it->iter, bytes);
756 
757 	if (!cursor->resid) {
758 		BUG_ON(!cursor->last_piece);
759 		return false;   /* no more data */
760 	}
761 
762 	if (!bytes || (it->iter.bi_size && it->iter.bi_bvec_done &&
763 		       page == bio_iter_page(it->bio, it->iter)))
764 		return false;	/* more bytes to process in this segment */
765 
766 	if (!it->iter.bi_size) {
767 		it->bio = it->bio->bi_next;
768 		it->iter = it->bio->bi_iter;
769 		if (cursor->resid < it->iter.bi_size)
770 			it->iter.bi_size = cursor->resid;
771 	}
772 
773 	BUG_ON(cursor->last_piece);
774 	BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
775 	cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter);
776 	return true;
777 }
778 #endif /* CONFIG_BLOCK */
779 
780 static void ceph_msg_data_bvecs_cursor_init(struct ceph_msg_data_cursor *cursor,
781 					size_t length)
782 {
783 	struct ceph_msg_data *data = cursor->data;
784 	struct bio_vec *bvecs = data->bvec_pos.bvecs;
785 
786 	cursor->resid = min_t(size_t, length, data->bvec_pos.iter.bi_size);
787 	cursor->bvec_iter = data->bvec_pos.iter;
788 	cursor->bvec_iter.bi_size = cursor->resid;
789 
790 	BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
791 	cursor->last_piece =
792 	    cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter);
793 }
794 
795 static struct page *ceph_msg_data_bvecs_next(struct ceph_msg_data_cursor *cursor,
796 						size_t *page_offset,
797 						size_t *length)
798 {
799 	struct bio_vec bv = bvec_iter_bvec(cursor->data->bvec_pos.bvecs,
800 					   cursor->bvec_iter);
801 
802 	*page_offset = bv.bv_offset;
803 	*length = bv.bv_len;
804 	return bv.bv_page;
805 }
806 
807 static bool ceph_msg_data_bvecs_advance(struct ceph_msg_data_cursor *cursor,
808 					size_t bytes)
809 {
810 	struct bio_vec *bvecs = cursor->data->bvec_pos.bvecs;
811 	struct page *page = bvec_iter_page(bvecs, cursor->bvec_iter);
812 
813 	BUG_ON(bytes > cursor->resid);
814 	BUG_ON(bytes > bvec_iter_len(bvecs, cursor->bvec_iter));
815 	cursor->resid -= bytes;
816 	bvec_iter_advance(bvecs, &cursor->bvec_iter, bytes);
817 
818 	if (!cursor->resid) {
819 		BUG_ON(!cursor->last_piece);
820 		return false;   /* no more data */
821 	}
822 
823 	if (!bytes || (cursor->bvec_iter.bi_bvec_done &&
824 		       page == bvec_iter_page(bvecs, cursor->bvec_iter)))
825 		return false;	/* more bytes to process in this segment */
826 
827 	BUG_ON(cursor->last_piece);
828 	BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
829 	cursor->last_piece =
830 	    cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter);
831 	return true;
832 }
833 
834 /*
835  * For a page array, a piece comes from the first page in the array
836  * that has not already been fully consumed.
837  */
838 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
839 					size_t length)
840 {
841 	struct ceph_msg_data *data = cursor->data;
842 	int page_count;
843 
844 	BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
845 
846 	BUG_ON(!data->pages);
847 	BUG_ON(!data->length);
848 
849 	cursor->resid = min(length, data->length);
850 	page_count = calc_pages_for(data->alignment, (u64)data->length);
851 	cursor->page_offset = data->alignment & ~PAGE_MASK;
852 	cursor->page_index = 0;
853 	BUG_ON(page_count > (int)USHRT_MAX);
854 	cursor->page_count = (unsigned short)page_count;
855 	BUG_ON(length > SIZE_MAX - cursor->page_offset);
856 	cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
857 }
858 
859 static struct page *
860 ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
861 					size_t *page_offset, size_t *length)
862 {
863 	struct ceph_msg_data *data = cursor->data;
864 
865 	BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
866 
867 	BUG_ON(cursor->page_index >= cursor->page_count);
868 	BUG_ON(cursor->page_offset >= PAGE_SIZE);
869 
870 	*page_offset = cursor->page_offset;
871 	if (cursor->last_piece)
872 		*length = cursor->resid;
873 	else
874 		*length = PAGE_SIZE - *page_offset;
875 
876 	return data->pages[cursor->page_index];
877 }
878 
879 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
880 						size_t bytes)
881 {
882 	BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
883 
884 	BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
885 
886 	/* Advance the cursor page offset */
887 
888 	cursor->resid -= bytes;
889 	cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
890 	if (!bytes || cursor->page_offset)
891 		return false;	/* more bytes to process in the current page */
892 
893 	if (!cursor->resid)
894 		return false;   /* no more data */
895 
896 	/* Move on to the next page; offset is already at 0 */
897 
898 	BUG_ON(cursor->page_index >= cursor->page_count);
899 	cursor->page_index++;
900 	cursor->last_piece = cursor->resid <= PAGE_SIZE;
901 
902 	return true;
903 }
904 
905 /*
906  * For a pagelist, a piece is whatever remains to be consumed in the
907  * first page in the list, or the front of the next page.
908  */
909 static void
910 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
911 					size_t length)
912 {
913 	struct ceph_msg_data *data = cursor->data;
914 	struct ceph_pagelist *pagelist;
915 	struct page *page;
916 
917 	BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
918 
919 	pagelist = data->pagelist;
920 	BUG_ON(!pagelist);
921 
922 	if (!length)
923 		return;		/* pagelist can be assigned but empty */
924 
925 	BUG_ON(list_empty(&pagelist->head));
926 	page = list_first_entry(&pagelist->head, struct page, lru);
927 
928 	cursor->resid = min(length, pagelist->length);
929 	cursor->page = page;
930 	cursor->offset = 0;
931 	cursor->last_piece = cursor->resid <= PAGE_SIZE;
932 }
933 
934 static struct page *
935 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
936 				size_t *page_offset, size_t *length)
937 {
938 	struct ceph_msg_data *data = cursor->data;
939 	struct ceph_pagelist *pagelist;
940 
941 	BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
942 
943 	pagelist = data->pagelist;
944 	BUG_ON(!pagelist);
945 
946 	BUG_ON(!cursor->page);
947 	BUG_ON(cursor->offset + cursor->resid != pagelist->length);
948 
949 	/* offset of first page in pagelist is always 0 */
950 	*page_offset = cursor->offset & ~PAGE_MASK;
951 	if (cursor->last_piece)
952 		*length = cursor->resid;
953 	else
954 		*length = PAGE_SIZE - *page_offset;
955 
956 	return cursor->page;
957 }
958 
959 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
960 						size_t bytes)
961 {
962 	struct ceph_msg_data *data = cursor->data;
963 	struct ceph_pagelist *pagelist;
964 
965 	BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
966 
967 	pagelist = data->pagelist;
968 	BUG_ON(!pagelist);
969 
970 	BUG_ON(cursor->offset + cursor->resid != pagelist->length);
971 	BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
972 
973 	/* Advance the cursor offset */
974 
975 	cursor->resid -= bytes;
976 	cursor->offset += bytes;
977 	/* offset of first page in pagelist is always 0 */
978 	if (!bytes || cursor->offset & ~PAGE_MASK)
979 		return false;	/* more bytes to process in the current page */
980 
981 	if (!cursor->resid)
982 		return false;   /* no more data */
983 
984 	/* Move on to the next page */
985 
986 	BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
987 	cursor->page = list_next_entry(cursor->page, lru);
988 	cursor->last_piece = cursor->resid <= PAGE_SIZE;
989 
990 	return true;
991 }
992 
993 /*
994  * Message data is handled (sent or received) in pieces, where each
995  * piece resides on a single page.  The network layer might not
996  * consume an entire piece at once.  A data item's cursor keeps
997  * track of which piece is next to process and how much remains to
998  * be processed in that piece.  It also tracks whether the current
999  * piece is the last one in the data item.
1000  */
1001 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1002 {
1003 	size_t length = cursor->total_resid;
1004 
1005 	switch (cursor->data->type) {
1006 	case CEPH_MSG_DATA_PAGELIST:
1007 		ceph_msg_data_pagelist_cursor_init(cursor, length);
1008 		break;
1009 	case CEPH_MSG_DATA_PAGES:
1010 		ceph_msg_data_pages_cursor_init(cursor, length);
1011 		break;
1012 #ifdef CONFIG_BLOCK
1013 	case CEPH_MSG_DATA_BIO:
1014 		ceph_msg_data_bio_cursor_init(cursor, length);
1015 		break;
1016 #endif /* CONFIG_BLOCK */
1017 	case CEPH_MSG_DATA_BVECS:
1018 		ceph_msg_data_bvecs_cursor_init(cursor, length);
1019 		break;
1020 	case CEPH_MSG_DATA_NONE:
1021 	default:
1022 		/* BUG(); */
1023 		break;
1024 	}
1025 	cursor->need_crc = true;
1026 }
1027 
1028 void ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor,
1029 			       struct ceph_msg *msg, size_t length)
1030 {
1031 	BUG_ON(!length);
1032 	BUG_ON(length > msg->data_length);
1033 	BUG_ON(!msg->num_data_items);
1034 
1035 	cursor->total_resid = length;
1036 	cursor->data = msg->data;
1037 
1038 	__ceph_msg_data_cursor_init(cursor);
1039 }
1040 
1041 /*
1042  * Return the page containing the next piece to process for a given
1043  * data item, and supply the page offset and length of that piece.
1044  * Indicate whether this is the last piece in this data item.
1045  */
1046 struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1047 				size_t *page_offset, size_t *length,
1048 				bool *last_piece)
1049 {
1050 	struct page *page;
1051 
1052 	switch (cursor->data->type) {
1053 	case CEPH_MSG_DATA_PAGELIST:
1054 		page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1055 		break;
1056 	case CEPH_MSG_DATA_PAGES:
1057 		page = ceph_msg_data_pages_next(cursor, page_offset, length);
1058 		break;
1059 #ifdef CONFIG_BLOCK
1060 	case CEPH_MSG_DATA_BIO:
1061 		page = ceph_msg_data_bio_next(cursor, page_offset, length);
1062 		break;
1063 #endif /* CONFIG_BLOCK */
1064 	case CEPH_MSG_DATA_BVECS:
1065 		page = ceph_msg_data_bvecs_next(cursor, page_offset, length);
1066 		break;
1067 	case CEPH_MSG_DATA_NONE:
1068 	default:
1069 		page = NULL;
1070 		break;
1071 	}
1072 
1073 	BUG_ON(!page);
1074 	BUG_ON(*page_offset + *length > PAGE_SIZE);
1075 	BUG_ON(!*length);
1076 	BUG_ON(*length > cursor->resid);
1077 	if (last_piece)
1078 		*last_piece = cursor->last_piece;
1079 
1080 	return page;
1081 }
1082 
1083 /*
1084  * Returns true if the result moves the cursor on to the next piece
1085  * of the data item.
1086  */
1087 void ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor, size_t bytes)
1088 {
1089 	bool new_piece;
1090 
1091 	BUG_ON(bytes > cursor->resid);
1092 	switch (cursor->data->type) {
1093 	case CEPH_MSG_DATA_PAGELIST:
1094 		new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1095 		break;
1096 	case CEPH_MSG_DATA_PAGES:
1097 		new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1098 		break;
1099 #ifdef CONFIG_BLOCK
1100 	case CEPH_MSG_DATA_BIO:
1101 		new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1102 		break;
1103 #endif /* CONFIG_BLOCK */
1104 	case CEPH_MSG_DATA_BVECS:
1105 		new_piece = ceph_msg_data_bvecs_advance(cursor, bytes);
1106 		break;
1107 	case CEPH_MSG_DATA_NONE:
1108 	default:
1109 		BUG();
1110 		break;
1111 	}
1112 	cursor->total_resid -= bytes;
1113 
1114 	if (!cursor->resid && cursor->total_resid) {
1115 		WARN_ON(!cursor->last_piece);
1116 		cursor->data++;
1117 		__ceph_msg_data_cursor_init(cursor);
1118 		new_piece = true;
1119 	}
1120 	cursor->need_crc = new_piece;
1121 }
1122 
1123 u32 ceph_crc32c_page(u32 crc, struct page *page, unsigned int page_offset,
1124 		     unsigned int length)
1125 {
1126 	char *kaddr;
1127 
1128 	kaddr = kmap(page);
1129 	BUG_ON(kaddr == NULL);
1130 	crc = crc32c(crc, kaddr + page_offset, length);
1131 	kunmap(page);
1132 
1133 	return crc;
1134 }
1135 
1136 bool ceph_addr_is_blank(const struct ceph_entity_addr *addr)
1137 {
1138 	struct sockaddr_storage ss = addr->in_addr; /* align */
1139 	struct in_addr *addr4 = &((struct sockaddr_in *)&ss)->sin_addr;
1140 	struct in6_addr *addr6 = &((struct sockaddr_in6 *)&ss)->sin6_addr;
1141 
1142 	switch (ss.ss_family) {
1143 	case AF_INET:
1144 		return addr4->s_addr == htonl(INADDR_ANY);
1145 	case AF_INET6:
1146 		return ipv6_addr_any(addr6);
1147 	default:
1148 		return true;
1149 	}
1150 }
1151 
1152 int ceph_addr_port(const struct ceph_entity_addr *addr)
1153 {
1154 	switch (get_unaligned(&addr->in_addr.ss_family)) {
1155 	case AF_INET:
1156 		return ntohs(get_unaligned(&((struct sockaddr_in *)&addr->in_addr)->sin_port));
1157 	case AF_INET6:
1158 		return ntohs(get_unaligned(&((struct sockaddr_in6 *)&addr->in_addr)->sin6_port));
1159 	}
1160 	return 0;
1161 }
1162 
1163 void ceph_addr_set_port(struct ceph_entity_addr *addr, int p)
1164 {
1165 	switch (get_unaligned(&addr->in_addr.ss_family)) {
1166 	case AF_INET:
1167 		put_unaligned(htons(p), &((struct sockaddr_in *)&addr->in_addr)->sin_port);
1168 		break;
1169 	case AF_INET6:
1170 		put_unaligned(htons(p), &((struct sockaddr_in6 *)&addr->in_addr)->sin6_port);
1171 		break;
1172 	}
1173 }
1174 
1175 /*
1176  * Unlike other *_pton function semantics, zero indicates success.
1177  */
1178 static int ceph_pton(const char *str, size_t len, struct ceph_entity_addr *addr,
1179 		char delim, const char **ipend)
1180 {
1181 	memset(&addr->in_addr, 0, sizeof(addr->in_addr));
1182 
1183 	if (in4_pton(str, len, (u8 *)&((struct sockaddr_in *)&addr->in_addr)->sin_addr.s_addr, delim, ipend)) {
1184 		put_unaligned(AF_INET, &addr->in_addr.ss_family);
1185 		return 0;
1186 	}
1187 
1188 	if (in6_pton(str, len, (u8 *)&((struct sockaddr_in6 *)&addr->in_addr)->sin6_addr.s6_addr, delim, ipend)) {
1189 		put_unaligned(AF_INET6, &addr->in_addr.ss_family);
1190 		return 0;
1191 	}
1192 
1193 	return -EINVAL;
1194 }
1195 
1196 /*
1197  * Extract hostname string and resolve using kernel DNS facility.
1198  */
1199 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1200 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1201 		struct ceph_entity_addr *addr, char delim, const char **ipend)
1202 {
1203 	const char *end, *delim_p;
1204 	char *colon_p, *ip_addr = NULL;
1205 	int ip_len, ret;
1206 
1207 	/*
1208 	 * The end of the hostname occurs immediately preceding the delimiter or
1209 	 * the port marker (':') where the delimiter takes precedence.
1210 	 */
1211 	delim_p = memchr(name, delim, namelen);
1212 	colon_p = memchr(name, ':', namelen);
1213 
1214 	if (delim_p && colon_p)
1215 		end = delim_p < colon_p ? delim_p : colon_p;
1216 	else if (!delim_p && colon_p)
1217 		end = colon_p;
1218 	else {
1219 		end = delim_p;
1220 		if (!end) /* case: hostname:/ */
1221 			end = name + namelen;
1222 	}
1223 
1224 	if (end <= name)
1225 		return -EINVAL;
1226 
1227 	/* do dns_resolve upcall */
1228 	ip_len = dns_query(current->nsproxy->net_ns,
1229 			   NULL, name, end - name, NULL, &ip_addr, NULL, false);
1230 	if (ip_len > 0)
1231 		ret = ceph_pton(ip_addr, ip_len, addr, -1, NULL);
1232 	else
1233 		ret = -ESRCH;
1234 
1235 	kfree(ip_addr);
1236 
1237 	*ipend = end;
1238 
1239 	pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1240 			ret, ret ? "failed" : ceph_pr_addr(addr));
1241 
1242 	return ret;
1243 }
1244 #else
1245 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1246 		struct ceph_entity_addr *addr, char delim, const char **ipend)
1247 {
1248 	return -EINVAL;
1249 }
1250 #endif
1251 
1252 /*
1253  * Parse a server name (IP or hostname). If a valid IP address is not found
1254  * then try to extract a hostname to resolve using userspace DNS upcall.
1255  */
1256 static int ceph_parse_server_name(const char *name, size_t namelen,
1257 		struct ceph_entity_addr *addr, char delim, const char **ipend)
1258 {
1259 	int ret;
1260 
1261 	ret = ceph_pton(name, namelen, addr, delim, ipend);
1262 	if (ret)
1263 		ret = ceph_dns_resolve_name(name, namelen, addr, delim, ipend);
1264 
1265 	return ret;
1266 }
1267 
1268 /*
1269  * Parse an ip[:port] list into an addr array.  Use the default
1270  * monitor port if a port isn't specified.
1271  */
1272 int ceph_parse_ips(const char *c, const char *end,
1273 		   struct ceph_entity_addr *addr,
1274 		   int max_count, int *count, char delim)
1275 {
1276 	int i, ret = -EINVAL;
1277 	const char *p = c;
1278 
1279 	dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1280 	for (i = 0; i < max_count; i++) {
1281 		char cur_delim = delim;
1282 		const char *ipend;
1283 		int port;
1284 
1285 		if (*p == '[') {
1286 			cur_delim = ']';
1287 			p++;
1288 		}
1289 
1290 		ret = ceph_parse_server_name(p, end - p, &addr[i], cur_delim,
1291 					     &ipend);
1292 		if (ret)
1293 			goto bad;
1294 		ret = -EINVAL;
1295 
1296 		p = ipend;
1297 
1298 		if (cur_delim == ']') {
1299 			if (*p != ']') {
1300 				dout("missing matching ']'\n");
1301 				goto bad;
1302 			}
1303 			p++;
1304 		}
1305 
1306 		/* port? */
1307 		if (p < end && *p == ':') {
1308 			port = 0;
1309 			p++;
1310 			while (p < end && *p >= '0' && *p <= '9') {
1311 				port = (port * 10) + (*p - '0');
1312 				p++;
1313 			}
1314 			if (port == 0)
1315 				port = CEPH_MON_PORT;
1316 			else if (port > 65535)
1317 				goto bad;
1318 		} else {
1319 			port = CEPH_MON_PORT;
1320 		}
1321 
1322 		ceph_addr_set_port(&addr[i], port);
1323 		/*
1324 		 * We want the type to be set according to ms_mode
1325 		 * option, but options are normally parsed after mon
1326 		 * addresses.  Rather than complicating parsing, set
1327 		 * to LEGACY and override in build_initial_monmap()
1328 		 * for mon addresses and ceph_messenger_init() for
1329 		 * ip option.
1330 		 */
1331 		addr[i].type = CEPH_ENTITY_ADDR_TYPE_LEGACY;
1332 		addr[i].nonce = 0;
1333 
1334 		dout("%s got %s\n", __func__, ceph_pr_addr(&addr[i]));
1335 
1336 		if (p == end)
1337 			break;
1338 		if (*p != delim)
1339 			goto bad;
1340 		p++;
1341 	}
1342 
1343 	if (p != end)
1344 		goto bad;
1345 
1346 	if (count)
1347 		*count = i + 1;
1348 	return 0;
1349 
1350 bad:
1351 	return ret;
1352 }
1353 
1354 /*
1355  * Process message.  This happens in the worker thread.  The callback should
1356  * be careful not to do anything that waits on other incoming messages or it
1357  * may deadlock.
1358  */
1359 void ceph_con_process_message(struct ceph_connection *con)
1360 {
1361 	struct ceph_msg *msg = con->in_msg;
1362 
1363 	BUG_ON(con->in_msg->con != con);
1364 	con->in_msg = NULL;
1365 
1366 	/* if first message, set peer_name */
1367 	if (con->peer_name.type == 0)
1368 		con->peer_name = msg->hdr.src;
1369 
1370 	con->in_seq++;
1371 	mutex_unlock(&con->mutex);
1372 
1373 	dout("===== %p %llu from %s%lld %d=%s len %d+%d+%d (%u %u %u) =====\n",
1374 	     msg, le64_to_cpu(msg->hdr.seq),
1375 	     ENTITY_NAME(msg->hdr.src),
1376 	     le16_to_cpu(msg->hdr.type),
1377 	     ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
1378 	     le32_to_cpu(msg->hdr.front_len),
1379 	     le32_to_cpu(msg->hdr.middle_len),
1380 	     le32_to_cpu(msg->hdr.data_len),
1381 	     con->in_front_crc, con->in_middle_crc, con->in_data_crc);
1382 	con->ops->dispatch(con, msg);
1383 
1384 	mutex_lock(&con->mutex);
1385 }
1386 
1387 /*
1388  * Atomically queue work on a connection after the specified delay.
1389  * Bump @con reference to avoid races with connection teardown.
1390  * Returns 0 if work was queued, or an error code otherwise.
1391  */
1392 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
1393 {
1394 	if (!con->ops->get(con)) {
1395 		dout("%s %p ref count 0\n", __func__, con);
1396 		return -ENOENT;
1397 	}
1398 
1399 	if (delay >= HZ)
1400 		delay = round_jiffies_relative(delay);
1401 
1402 	dout("%s %p %lu\n", __func__, con, delay);
1403 	if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
1404 		dout("%s %p - already queued\n", __func__, con);
1405 		con->ops->put(con);
1406 		return -EBUSY;
1407 	}
1408 
1409 	return 0;
1410 }
1411 
1412 static void queue_con(struct ceph_connection *con)
1413 {
1414 	(void) queue_con_delay(con, 0);
1415 }
1416 
1417 static void cancel_con(struct ceph_connection *con)
1418 {
1419 	if (cancel_delayed_work(&con->work)) {
1420 		dout("%s %p\n", __func__, con);
1421 		con->ops->put(con);
1422 	}
1423 }
1424 
1425 static bool con_sock_closed(struct ceph_connection *con)
1426 {
1427 	if (!ceph_con_flag_test_and_clear(con, CEPH_CON_F_SOCK_CLOSED))
1428 		return false;
1429 
1430 #define CASE(x)								\
1431 	case CEPH_CON_S_ ## x:						\
1432 		con->error_msg = "socket closed (con state " #x ")";	\
1433 		break;
1434 
1435 	switch (con->state) {
1436 	CASE(CLOSED);
1437 	CASE(PREOPEN);
1438 	CASE(V1_BANNER);
1439 	CASE(V1_CONNECT_MSG);
1440 	CASE(V2_BANNER_PREFIX);
1441 	CASE(V2_BANNER_PAYLOAD);
1442 	CASE(V2_HELLO);
1443 	CASE(V2_AUTH);
1444 	CASE(V2_AUTH_SIGNATURE);
1445 	CASE(V2_SESSION_CONNECT);
1446 	CASE(V2_SESSION_RECONNECT);
1447 	CASE(OPEN);
1448 	CASE(STANDBY);
1449 	default:
1450 		BUG();
1451 	}
1452 #undef CASE
1453 
1454 	return true;
1455 }
1456 
1457 static bool con_backoff(struct ceph_connection *con)
1458 {
1459 	int ret;
1460 
1461 	if (!ceph_con_flag_test_and_clear(con, CEPH_CON_F_BACKOFF))
1462 		return false;
1463 
1464 	ret = queue_con_delay(con, con->delay);
1465 	if (ret) {
1466 		dout("%s: con %p FAILED to back off %lu\n", __func__,
1467 			con, con->delay);
1468 		BUG_ON(ret == -ENOENT);
1469 		ceph_con_flag_set(con, CEPH_CON_F_BACKOFF);
1470 	}
1471 
1472 	return true;
1473 }
1474 
1475 /* Finish fault handling; con->mutex must *not* be held here */
1476 
1477 static void con_fault_finish(struct ceph_connection *con)
1478 {
1479 	dout("%s %p\n", __func__, con);
1480 
1481 	/*
1482 	 * in case we faulted due to authentication, invalidate our
1483 	 * current tickets so that we can get new ones.
1484 	 */
1485 	if (con->v1.auth_retry) {
1486 		dout("auth_retry %d, invalidating\n", con->v1.auth_retry);
1487 		if (con->ops->invalidate_authorizer)
1488 			con->ops->invalidate_authorizer(con);
1489 		con->v1.auth_retry = 0;
1490 	}
1491 
1492 	if (con->ops->fault)
1493 		con->ops->fault(con);
1494 }
1495 
1496 /*
1497  * Do some work on a connection.  Drop a connection ref when we're done.
1498  */
1499 static void ceph_con_workfn(struct work_struct *work)
1500 {
1501 	struct ceph_connection *con = container_of(work, struct ceph_connection,
1502 						   work.work);
1503 	bool fault;
1504 
1505 	mutex_lock(&con->mutex);
1506 	while (true) {
1507 		int ret;
1508 
1509 		if ((fault = con_sock_closed(con))) {
1510 			dout("%s: con %p SOCK_CLOSED\n", __func__, con);
1511 			break;
1512 		}
1513 		if (con_backoff(con)) {
1514 			dout("%s: con %p BACKOFF\n", __func__, con);
1515 			break;
1516 		}
1517 		if (con->state == CEPH_CON_S_STANDBY) {
1518 			dout("%s: con %p STANDBY\n", __func__, con);
1519 			break;
1520 		}
1521 		if (con->state == CEPH_CON_S_CLOSED) {
1522 			dout("%s: con %p CLOSED\n", __func__, con);
1523 			BUG_ON(con->sock);
1524 			break;
1525 		}
1526 		if (con->state == CEPH_CON_S_PREOPEN) {
1527 			dout("%s: con %p PREOPEN\n", __func__, con);
1528 			BUG_ON(con->sock);
1529 		}
1530 
1531 		if (ceph_msgr2(from_msgr(con->msgr)))
1532 			ret = ceph_con_v2_try_read(con);
1533 		else
1534 			ret = ceph_con_v1_try_read(con);
1535 		if (ret < 0) {
1536 			if (ret == -EAGAIN)
1537 				continue;
1538 			if (!con->error_msg)
1539 				con->error_msg = "socket error on read";
1540 			fault = true;
1541 			break;
1542 		}
1543 
1544 		if (ceph_msgr2(from_msgr(con->msgr)))
1545 			ret = ceph_con_v2_try_write(con);
1546 		else
1547 			ret = ceph_con_v1_try_write(con);
1548 		if (ret < 0) {
1549 			if (ret == -EAGAIN)
1550 				continue;
1551 			if (!con->error_msg)
1552 				con->error_msg = "socket error on write";
1553 			fault = true;
1554 		}
1555 
1556 		break;	/* If we make it to here, we're done */
1557 	}
1558 	if (fault)
1559 		con_fault(con);
1560 	mutex_unlock(&con->mutex);
1561 
1562 	if (fault)
1563 		con_fault_finish(con);
1564 
1565 	con->ops->put(con);
1566 }
1567 
1568 /*
1569  * Generic error/fault handler.  A retry mechanism is used with
1570  * exponential backoff
1571  */
1572 static void con_fault(struct ceph_connection *con)
1573 {
1574 	dout("fault %p state %d to peer %s\n",
1575 	     con, con->state, ceph_pr_addr(&con->peer_addr));
1576 
1577 	pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
1578 		ceph_pr_addr(&con->peer_addr), con->error_msg);
1579 	con->error_msg = NULL;
1580 
1581 	WARN_ON(con->state == CEPH_CON_S_STANDBY ||
1582 		con->state == CEPH_CON_S_CLOSED);
1583 
1584 	ceph_con_reset_protocol(con);
1585 
1586 	if (ceph_con_flag_test(con, CEPH_CON_F_LOSSYTX)) {
1587 		dout("fault on LOSSYTX channel, marking CLOSED\n");
1588 		con->state = CEPH_CON_S_CLOSED;
1589 		return;
1590 	}
1591 
1592 	/* Requeue anything that hasn't been acked */
1593 	list_splice_init(&con->out_sent, &con->out_queue);
1594 
1595 	/* If there are no messages queued or keepalive pending, place
1596 	 * the connection in a STANDBY state */
1597 	if (list_empty(&con->out_queue) &&
1598 	    !ceph_con_flag_test(con, CEPH_CON_F_KEEPALIVE_PENDING)) {
1599 		dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
1600 		ceph_con_flag_clear(con, CEPH_CON_F_WRITE_PENDING);
1601 		con->state = CEPH_CON_S_STANDBY;
1602 	} else {
1603 		/* retry after a delay. */
1604 		con->state = CEPH_CON_S_PREOPEN;
1605 		if (!con->delay) {
1606 			con->delay = BASE_DELAY_INTERVAL;
1607 		} else if (con->delay < MAX_DELAY_INTERVAL) {
1608 			con->delay *= 2;
1609 			if (con->delay > MAX_DELAY_INTERVAL)
1610 				con->delay = MAX_DELAY_INTERVAL;
1611 		}
1612 		ceph_con_flag_set(con, CEPH_CON_F_BACKOFF);
1613 		queue_con(con);
1614 	}
1615 }
1616 
1617 void ceph_messenger_reset_nonce(struct ceph_messenger *msgr)
1618 {
1619 	u32 nonce = le32_to_cpu(msgr->inst.addr.nonce) + 1000000;
1620 	msgr->inst.addr.nonce = cpu_to_le32(nonce);
1621 	ceph_encode_my_addr(msgr);
1622 }
1623 
1624 /*
1625  * initialize a new messenger instance
1626  */
1627 void ceph_messenger_init(struct ceph_messenger *msgr,
1628 			 struct ceph_entity_addr *myaddr)
1629 {
1630 	spin_lock_init(&msgr->global_seq_lock);
1631 
1632 	if (myaddr) {
1633 		memcpy(&msgr->inst.addr.in_addr, &myaddr->in_addr,
1634 		       sizeof(msgr->inst.addr.in_addr));
1635 		ceph_addr_set_port(&msgr->inst.addr, 0);
1636 	}
1637 
1638 	/*
1639 	 * Since nautilus, clients are identified using type ANY.
1640 	 * For msgr1, ceph_encode_banner_addr() munges it to NONE.
1641 	 */
1642 	msgr->inst.addr.type = CEPH_ENTITY_ADDR_TYPE_ANY;
1643 
1644 	/* generate a random non-zero nonce */
1645 	do {
1646 		get_random_bytes(&msgr->inst.addr.nonce,
1647 				 sizeof(msgr->inst.addr.nonce));
1648 	} while (!msgr->inst.addr.nonce);
1649 	ceph_encode_my_addr(msgr);
1650 
1651 	atomic_set(&msgr->stopping, 0);
1652 	write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
1653 
1654 	dout("%s %p\n", __func__, msgr);
1655 }
1656 
1657 void ceph_messenger_fini(struct ceph_messenger *msgr)
1658 {
1659 	put_net(read_pnet(&msgr->net));
1660 }
1661 
1662 static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con)
1663 {
1664 	if (msg->con)
1665 		msg->con->ops->put(msg->con);
1666 
1667 	msg->con = con ? con->ops->get(con) : NULL;
1668 	BUG_ON(msg->con != con);
1669 }
1670 
1671 static void clear_standby(struct ceph_connection *con)
1672 {
1673 	/* come back from STANDBY? */
1674 	if (con->state == CEPH_CON_S_STANDBY) {
1675 		dout("clear_standby %p and ++connect_seq\n", con);
1676 		con->state = CEPH_CON_S_PREOPEN;
1677 		con->v1.connect_seq++;
1678 		WARN_ON(ceph_con_flag_test(con, CEPH_CON_F_WRITE_PENDING));
1679 		WARN_ON(ceph_con_flag_test(con, CEPH_CON_F_KEEPALIVE_PENDING));
1680 	}
1681 }
1682 
1683 /*
1684  * Queue up an outgoing message on the given connection.
1685  *
1686  * Consumes a ref on @msg.
1687  */
1688 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
1689 {
1690 	/* set src+dst */
1691 	msg->hdr.src = con->msgr->inst.name;
1692 	BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
1693 	msg->needs_out_seq = true;
1694 
1695 	mutex_lock(&con->mutex);
1696 
1697 	if (con->state == CEPH_CON_S_CLOSED) {
1698 		dout("con_send %p closed, dropping %p\n", con, msg);
1699 		ceph_msg_put(msg);
1700 		mutex_unlock(&con->mutex);
1701 		return;
1702 	}
1703 
1704 	msg_con_set(msg, con);
1705 
1706 	BUG_ON(!list_empty(&msg->list_head));
1707 	list_add_tail(&msg->list_head, &con->out_queue);
1708 	dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
1709 	     ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
1710 	     ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
1711 	     le32_to_cpu(msg->hdr.front_len),
1712 	     le32_to_cpu(msg->hdr.middle_len),
1713 	     le32_to_cpu(msg->hdr.data_len));
1714 
1715 	clear_standby(con);
1716 	mutex_unlock(&con->mutex);
1717 
1718 	/* if there wasn't anything waiting to send before, queue
1719 	 * new work */
1720 	if (!ceph_con_flag_test_and_set(con, CEPH_CON_F_WRITE_PENDING))
1721 		queue_con(con);
1722 }
1723 EXPORT_SYMBOL(ceph_con_send);
1724 
1725 /*
1726  * Revoke a message that was previously queued for send
1727  */
1728 void ceph_msg_revoke(struct ceph_msg *msg)
1729 {
1730 	struct ceph_connection *con = msg->con;
1731 
1732 	if (!con) {
1733 		dout("%s msg %p null con\n", __func__, msg);
1734 		return;		/* Message not in our possession */
1735 	}
1736 
1737 	mutex_lock(&con->mutex);
1738 	if (list_empty(&msg->list_head)) {
1739 		WARN_ON(con->out_msg == msg);
1740 		dout("%s con %p msg %p not linked\n", __func__, con, msg);
1741 		mutex_unlock(&con->mutex);
1742 		return;
1743 	}
1744 
1745 	dout("%s con %p msg %p was linked\n", __func__, con, msg);
1746 	msg->hdr.seq = 0;
1747 	ceph_msg_remove(msg);
1748 
1749 	if (con->out_msg == msg) {
1750 		WARN_ON(con->state != CEPH_CON_S_OPEN);
1751 		dout("%s con %p msg %p was sending\n", __func__, con, msg);
1752 		if (ceph_msgr2(from_msgr(con->msgr)))
1753 			ceph_con_v2_revoke(con);
1754 		else
1755 			ceph_con_v1_revoke(con);
1756 		ceph_msg_put(con->out_msg);
1757 		con->out_msg = NULL;
1758 	} else {
1759 		dout("%s con %p msg %p not current, out_msg %p\n", __func__,
1760 		     con, msg, con->out_msg);
1761 	}
1762 	mutex_unlock(&con->mutex);
1763 }
1764 
1765 /*
1766  * Revoke a message that we may be reading data into
1767  */
1768 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
1769 {
1770 	struct ceph_connection *con = msg->con;
1771 
1772 	if (!con) {
1773 		dout("%s msg %p null con\n", __func__, msg);
1774 		return;		/* Message not in our possession */
1775 	}
1776 
1777 	mutex_lock(&con->mutex);
1778 	if (con->in_msg == msg) {
1779 		WARN_ON(con->state != CEPH_CON_S_OPEN);
1780 		dout("%s con %p msg %p was recving\n", __func__, con, msg);
1781 		if (ceph_msgr2(from_msgr(con->msgr)))
1782 			ceph_con_v2_revoke_incoming(con);
1783 		else
1784 			ceph_con_v1_revoke_incoming(con);
1785 		ceph_msg_put(con->in_msg);
1786 		con->in_msg = NULL;
1787 	} else {
1788 		dout("%s con %p msg %p not current, in_msg %p\n", __func__,
1789 		     con, msg, con->in_msg);
1790 	}
1791 	mutex_unlock(&con->mutex);
1792 }
1793 
1794 /*
1795  * Queue a keepalive byte to ensure the tcp connection is alive.
1796  */
1797 void ceph_con_keepalive(struct ceph_connection *con)
1798 {
1799 	dout("con_keepalive %p\n", con);
1800 	mutex_lock(&con->mutex);
1801 	clear_standby(con);
1802 	ceph_con_flag_set(con, CEPH_CON_F_KEEPALIVE_PENDING);
1803 	mutex_unlock(&con->mutex);
1804 
1805 	if (!ceph_con_flag_test_and_set(con, CEPH_CON_F_WRITE_PENDING))
1806 		queue_con(con);
1807 }
1808 EXPORT_SYMBOL(ceph_con_keepalive);
1809 
1810 bool ceph_con_keepalive_expired(struct ceph_connection *con,
1811 			       unsigned long interval)
1812 {
1813 	if (interval > 0 &&
1814 	    (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
1815 		struct timespec64 now;
1816 		struct timespec64 ts;
1817 		ktime_get_real_ts64(&now);
1818 		jiffies_to_timespec64(interval, &ts);
1819 		ts = timespec64_add(con->last_keepalive_ack, ts);
1820 		return timespec64_compare(&now, &ts) >= 0;
1821 	}
1822 	return false;
1823 }
1824 
1825 static struct ceph_msg_data *ceph_msg_data_add(struct ceph_msg *msg)
1826 {
1827 	BUG_ON(msg->num_data_items >= msg->max_data_items);
1828 	return &msg->data[msg->num_data_items++];
1829 }
1830 
1831 static void ceph_msg_data_destroy(struct ceph_msg_data *data)
1832 {
1833 	if (data->type == CEPH_MSG_DATA_PAGES && data->own_pages) {
1834 		int num_pages = calc_pages_for(data->alignment, data->length);
1835 		ceph_release_page_vector(data->pages, num_pages);
1836 	} else if (data->type == CEPH_MSG_DATA_PAGELIST) {
1837 		ceph_pagelist_release(data->pagelist);
1838 	}
1839 }
1840 
1841 void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
1842 			     size_t length, size_t alignment, bool own_pages)
1843 {
1844 	struct ceph_msg_data *data;
1845 
1846 	BUG_ON(!pages);
1847 	BUG_ON(!length);
1848 
1849 	data = ceph_msg_data_add(msg);
1850 	data->type = CEPH_MSG_DATA_PAGES;
1851 	data->pages = pages;
1852 	data->length = length;
1853 	data->alignment = alignment & ~PAGE_MASK;
1854 	data->own_pages = own_pages;
1855 
1856 	msg->data_length += length;
1857 }
1858 EXPORT_SYMBOL(ceph_msg_data_add_pages);
1859 
1860 void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
1861 				struct ceph_pagelist *pagelist)
1862 {
1863 	struct ceph_msg_data *data;
1864 
1865 	BUG_ON(!pagelist);
1866 	BUG_ON(!pagelist->length);
1867 
1868 	data = ceph_msg_data_add(msg);
1869 	data->type = CEPH_MSG_DATA_PAGELIST;
1870 	refcount_inc(&pagelist->refcnt);
1871 	data->pagelist = pagelist;
1872 
1873 	msg->data_length += pagelist->length;
1874 }
1875 EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
1876 
1877 #ifdef	CONFIG_BLOCK
1878 void ceph_msg_data_add_bio(struct ceph_msg *msg, struct ceph_bio_iter *bio_pos,
1879 			   u32 length)
1880 {
1881 	struct ceph_msg_data *data;
1882 
1883 	data = ceph_msg_data_add(msg);
1884 	data->type = CEPH_MSG_DATA_BIO;
1885 	data->bio_pos = *bio_pos;
1886 	data->bio_length = length;
1887 
1888 	msg->data_length += length;
1889 }
1890 EXPORT_SYMBOL(ceph_msg_data_add_bio);
1891 #endif	/* CONFIG_BLOCK */
1892 
1893 void ceph_msg_data_add_bvecs(struct ceph_msg *msg,
1894 			     struct ceph_bvec_iter *bvec_pos)
1895 {
1896 	struct ceph_msg_data *data;
1897 
1898 	data = ceph_msg_data_add(msg);
1899 	data->type = CEPH_MSG_DATA_BVECS;
1900 	data->bvec_pos = *bvec_pos;
1901 
1902 	msg->data_length += bvec_pos->iter.bi_size;
1903 }
1904 EXPORT_SYMBOL(ceph_msg_data_add_bvecs);
1905 
1906 /*
1907  * construct a new message with given type, size
1908  * the new msg has a ref count of 1.
1909  */
1910 struct ceph_msg *ceph_msg_new2(int type, int front_len, int max_data_items,
1911 			       gfp_t flags, bool can_fail)
1912 {
1913 	struct ceph_msg *m;
1914 
1915 	m = kmem_cache_zalloc(ceph_msg_cache, flags);
1916 	if (m == NULL)
1917 		goto out;
1918 
1919 	m->hdr.type = cpu_to_le16(type);
1920 	m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
1921 	m->hdr.front_len = cpu_to_le32(front_len);
1922 
1923 	INIT_LIST_HEAD(&m->list_head);
1924 	kref_init(&m->kref);
1925 
1926 	/* front */
1927 	if (front_len) {
1928 		m->front.iov_base = kvmalloc(front_len, flags);
1929 		if (m->front.iov_base == NULL) {
1930 			dout("ceph_msg_new can't allocate %d bytes\n",
1931 			     front_len);
1932 			goto out2;
1933 		}
1934 	} else {
1935 		m->front.iov_base = NULL;
1936 	}
1937 	m->front_alloc_len = m->front.iov_len = front_len;
1938 
1939 	if (max_data_items) {
1940 		m->data = kmalloc_array(max_data_items, sizeof(*m->data),
1941 					flags);
1942 		if (!m->data)
1943 			goto out2;
1944 
1945 		m->max_data_items = max_data_items;
1946 	}
1947 
1948 	dout("ceph_msg_new %p front %d\n", m, front_len);
1949 	return m;
1950 
1951 out2:
1952 	ceph_msg_put(m);
1953 out:
1954 	if (!can_fail) {
1955 		pr_err("msg_new can't create type %d front %d\n", type,
1956 		       front_len);
1957 		WARN_ON(1);
1958 	} else {
1959 		dout("msg_new can't create type %d front %d\n", type,
1960 		     front_len);
1961 	}
1962 	return NULL;
1963 }
1964 EXPORT_SYMBOL(ceph_msg_new2);
1965 
1966 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
1967 			      bool can_fail)
1968 {
1969 	return ceph_msg_new2(type, front_len, 0, flags, can_fail);
1970 }
1971 EXPORT_SYMBOL(ceph_msg_new);
1972 
1973 /*
1974  * Allocate "middle" portion of a message, if it is needed and wasn't
1975  * allocated by alloc_msg.  This allows us to read a small fixed-size
1976  * per-type header in the front and then gracefully fail (i.e.,
1977  * propagate the error to the caller based on info in the front) when
1978  * the middle is too large.
1979  */
1980 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
1981 {
1982 	int type = le16_to_cpu(msg->hdr.type);
1983 	int middle_len = le32_to_cpu(msg->hdr.middle_len);
1984 
1985 	dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
1986 	     ceph_msg_type_name(type), middle_len);
1987 	BUG_ON(!middle_len);
1988 	BUG_ON(msg->middle);
1989 
1990 	msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
1991 	if (!msg->middle)
1992 		return -ENOMEM;
1993 	return 0;
1994 }
1995 
1996 /*
1997  * Allocate a message for receiving an incoming message on a
1998  * connection, and save the result in con->in_msg.  Uses the
1999  * connection's private alloc_msg op if available.
2000  *
2001  * Returns 0 on success, or a negative error code.
2002  *
2003  * On success, if we set *skip = 1:
2004  *  - the next message should be skipped and ignored.
2005  *  - con->in_msg == NULL
2006  * or if we set *skip = 0:
2007  *  - con->in_msg is non-null.
2008  * On error (ENOMEM, EAGAIN, ...),
2009  *  - con->in_msg == NULL
2010  */
2011 int ceph_con_in_msg_alloc(struct ceph_connection *con,
2012 			  struct ceph_msg_header *hdr, int *skip)
2013 {
2014 	int middle_len = le32_to_cpu(hdr->middle_len);
2015 	struct ceph_msg *msg;
2016 	int ret = 0;
2017 
2018 	BUG_ON(con->in_msg != NULL);
2019 	BUG_ON(!con->ops->alloc_msg);
2020 
2021 	mutex_unlock(&con->mutex);
2022 	msg = con->ops->alloc_msg(con, hdr, skip);
2023 	mutex_lock(&con->mutex);
2024 	if (con->state != CEPH_CON_S_OPEN) {
2025 		if (msg)
2026 			ceph_msg_put(msg);
2027 		return -EAGAIN;
2028 	}
2029 	if (msg) {
2030 		BUG_ON(*skip);
2031 		msg_con_set(msg, con);
2032 		con->in_msg = msg;
2033 	} else {
2034 		/*
2035 		 * Null message pointer means either we should skip
2036 		 * this message or we couldn't allocate memory.  The
2037 		 * former is not an error.
2038 		 */
2039 		if (*skip)
2040 			return 0;
2041 
2042 		con->error_msg = "error allocating memory for incoming message";
2043 		return -ENOMEM;
2044 	}
2045 	memcpy(&con->in_msg->hdr, hdr, sizeof(*hdr));
2046 
2047 	if (middle_len && !con->in_msg->middle) {
2048 		ret = ceph_alloc_middle(con, con->in_msg);
2049 		if (ret < 0) {
2050 			ceph_msg_put(con->in_msg);
2051 			con->in_msg = NULL;
2052 		}
2053 	}
2054 
2055 	return ret;
2056 }
2057 
2058 void ceph_con_get_out_msg(struct ceph_connection *con)
2059 {
2060 	struct ceph_msg *msg;
2061 
2062 	BUG_ON(list_empty(&con->out_queue));
2063 	msg = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
2064 	WARN_ON(msg->con != con);
2065 
2066 	/*
2067 	 * Put the message on "sent" list using a ref from ceph_con_send().
2068 	 * It is put when the message is acked or revoked.
2069 	 */
2070 	list_move_tail(&msg->list_head, &con->out_sent);
2071 
2072 	/*
2073 	 * Only assign outgoing seq # if we haven't sent this message
2074 	 * yet.  If it is requeued, resend with it's original seq.
2075 	 */
2076 	if (msg->needs_out_seq) {
2077 		msg->hdr.seq = cpu_to_le64(++con->out_seq);
2078 		msg->needs_out_seq = false;
2079 
2080 		if (con->ops->reencode_message)
2081 			con->ops->reencode_message(msg);
2082 	}
2083 
2084 	/*
2085 	 * Get a ref for out_msg.  It is put when we are done sending the
2086 	 * message or in case of a fault.
2087 	 */
2088 	WARN_ON(con->out_msg);
2089 	con->out_msg = ceph_msg_get(msg);
2090 }
2091 
2092 /*
2093  * Free a generically kmalloc'd message.
2094  */
2095 static void ceph_msg_free(struct ceph_msg *m)
2096 {
2097 	dout("%s %p\n", __func__, m);
2098 	kvfree(m->front.iov_base);
2099 	kfree(m->data);
2100 	kmem_cache_free(ceph_msg_cache, m);
2101 }
2102 
2103 static void ceph_msg_release(struct kref *kref)
2104 {
2105 	struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
2106 	int i;
2107 
2108 	dout("%s %p\n", __func__, m);
2109 	WARN_ON(!list_empty(&m->list_head));
2110 
2111 	msg_con_set(m, NULL);
2112 
2113 	/* drop middle, data, if any */
2114 	if (m->middle) {
2115 		ceph_buffer_put(m->middle);
2116 		m->middle = NULL;
2117 	}
2118 
2119 	for (i = 0; i < m->num_data_items; i++)
2120 		ceph_msg_data_destroy(&m->data[i]);
2121 
2122 	if (m->pool)
2123 		ceph_msgpool_put(m->pool, m);
2124 	else
2125 		ceph_msg_free(m);
2126 }
2127 
2128 struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
2129 {
2130 	dout("%s %p (was %d)\n", __func__, msg,
2131 	     kref_read(&msg->kref));
2132 	kref_get(&msg->kref);
2133 	return msg;
2134 }
2135 EXPORT_SYMBOL(ceph_msg_get);
2136 
2137 void ceph_msg_put(struct ceph_msg *msg)
2138 {
2139 	dout("%s %p (was %d)\n", __func__, msg,
2140 	     kref_read(&msg->kref));
2141 	kref_put(&msg->kref, ceph_msg_release);
2142 }
2143 EXPORT_SYMBOL(ceph_msg_put);
2144 
2145 void ceph_msg_dump(struct ceph_msg *msg)
2146 {
2147 	pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
2148 		 msg->front_alloc_len, msg->data_length);
2149 	print_hex_dump(KERN_DEBUG, "header: ",
2150 		       DUMP_PREFIX_OFFSET, 16, 1,
2151 		       &msg->hdr, sizeof(msg->hdr), true);
2152 	print_hex_dump(KERN_DEBUG, " front: ",
2153 		       DUMP_PREFIX_OFFSET, 16, 1,
2154 		       msg->front.iov_base, msg->front.iov_len, true);
2155 	if (msg->middle)
2156 		print_hex_dump(KERN_DEBUG, "middle: ",
2157 			       DUMP_PREFIX_OFFSET, 16, 1,
2158 			       msg->middle->vec.iov_base,
2159 			       msg->middle->vec.iov_len, true);
2160 	print_hex_dump(KERN_DEBUG, "footer: ",
2161 		       DUMP_PREFIX_OFFSET, 16, 1,
2162 		       &msg->footer, sizeof(msg->footer), true);
2163 }
2164 EXPORT_SYMBOL(ceph_msg_dump);
2165