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/libceph.h>
19#include <linux/ceph/messenger.h>
20#include <linux/ceph/decode.h>
21#include <linux/ceph/pagelist.h>
22#include <linux/export.h>
23
24#define list_entry_next(pos, member)					\
25	list_entry(pos->member.next, typeof(*pos), member)
26
27/*
28 * Ceph uses the messenger to exchange ceph_msg messages with other
29 * hosts in the system.  The messenger provides ordered and reliable
30 * delivery.  We tolerate TCP disconnects by reconnecting (with
31 * exponential backoff) in the case of a fault (disconnection, bad
32 * crc, protocol error).  Acks allow sent messages to be discarded by
33 * the sender.
34 */
35
36/*
37 * We track the state of the socket on a given connection using
38 * values defined below.  The transition to a new socket state is
39 * handled by a function which verifies we aren't coming from an
40 * unexpected state.
41 *
42 *      --------
43 *      | NEW* |  transient initial state
44 *      --------
45 *          | con_sock_state_init()
46 *          v
47 *      ----------
48 *      | CLOSED |  initialized, but no socket (and no
49 *      ----------  TCP connection)
50 *       ^      \
51 *       |       \ con_sock_state_connecting()
52 *       |        ----------------------
53 *       |                              \
54 *       + con_sock_state_closed()       \
55 *       |+---------------------------    \
56 *       | \                          \    \
57 *       |  -----------                \    \
58 *       |  | CLOSING |  socket event;  \    \
59 *       |  -----------  await close     \    \
60 *       |       ^                        \   |
61 *       |       |                         \  |
62 *       |       + con_sock_state_closing() \ |
63 *       |      / \                         | |
64 *       |     /   ---------------          | |
65 *       |    /                   \         v v
66 *       |   /                    --------------
67 *       |  /    -----------------| CONNECTING |  socket created, TCP
68 *       |  |   /                 --------------  connect initiated
69 *       |  |   | con_sock_state_connected()
70 *       |  |   v
71 *      -------------
72 *      | CONNECTED |  TCP connection established
73 *      -------------
74 *
75 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
76 */
77
78#define CON_SOCK_STATE_NEW		0	/* -> CLOSED */
79#define CON_SOCK_STATE_CLOSED		1	/* -> CONNECTING */
80#define CON_SOCK_STATE_CONNECTING	2	/* -> CONNECTED or -> CLOSING */
81#define CON_SOCK_STATE_CONNECTED	3	/* -> CLOSING or -> CLOSED */
82#define CON_SOCK_STATE_CLOSING		4	/* -> CLOSED */
83
84/*
85 * connection states
86 */
87#define CON_STATE_CLOSED        1  /* -> PREOPEN */
88#define CON_STATE_PREOPEN       2  /* -> CONNECTING, CLOSED */
89#define CON_STATE_CONNECTING    3  /* -> NEGOTIATING, CLOSED */
90#define CON_STATE_NEGOTIATING   4  /* -> OPEN, CLOSED */
91#define CON_STATE_OPEN          5  /* -> STANDBY, CLOSED */
92#define CON_STATE_STANDBY       6  /* -> PREOPEN, CLOSED */
93
94/*
95 * ceph_connection flag bits
96 */
97#define CON_FLAG_LOSSYTX           0  /* we can close channel or drop
98				       * messages on errors */
99#define CON_FLAG_KEEPALIVE_PENDING 1  /* we need to send a keepalive */
100#define CON_FLAG_WRITE_PENDING	   2  /* we have data ready to send */
101#define CON_FLAG_SOCK_CLOSED	   3  /* socket state changed to closed */
102#define CON_FLAG_BACKOFF           4  /* need to retry queuing delayed work */
103
104static bool con_flag_valid(unsigned long con_flag)
105{
106	switch (con_flag) {
107	case CON_FLAG_LOSSYTX:
108	case CON_FLAG_KEEPALIVE_PENDING:
109	case CON_FLAG_WRITE_PENDING:
110	case CON_FLAG_SOCK_CLOSED:
111	case CON_FLAG_BACKOFF:
112		return true;
113	default:
114		return false;
115	}
116}
117
118static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
119{
120	BUG_ON(!con_flag_valid(con_flag));
121
122	clear_bit(con_flag, &con->flags);
123}
124
125static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
126{
127	BUG_ON(!con_flag_valid(con_flag));
128
129	set_bit(con_flag, &con->flags);
130}
131
132static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
133{
134	BUG_ON(!con_flag_valid(con_flag));
135
136	return test_bit(con_flag, &con->flags);
137}
138
139static bool con_flag_test_and_clear(struct ceph_connection *con,
140					unsigned long con_flag)
141{
142	BUG_ON(!con_flag_valid(con_flag));
143
144	return test_and_clear_bit(con_flag, &con->flags);
145}
146
147static bool con_flag_test_and_set(struct ceph_connection *con,
148					unsigned long con_flag)
149{
150	BUG_ON(!con_flag_valid(con_flag));
151
152	return test_and_set_bit(con_flag, &con->flags);
153}
154
155/* Slab caches for frequently-allocated structures */
156
157static struct kmem_cache	*ceph_msg_cache;
158static struct kmem_cache	*ceph_msg_data_cache;
159
160/* static tag bytes (protocol control messages) */
161static char tag_msg = CEPH_MSGR_TAG_MSG;
162static char tag_ack = CEPH_MSGR_TAG_ACK;
163static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
164
165#ifdef CONFIG_LOCKDEP
166static struct lock_class_key socket_class;
167#endif
168
169/*
170 * When skipping (ignoring) a block of input we read it into a "skip
171 * buffer," which is this many bytes in size.
172 */
173#define SKIP_BUF_SIZE	1024
174
175static void queue_con(struct ceph_connection *con);
176static void con_work(struct work_struct *);
177static void con_fault(struct ceph_connection *con);
178
179/*
180 * Nicely render a sockaddr as a string.  An array of formatted
181 * strings is used, to approximate reentrancy.
182 */
183#define ADDR_STR_COUNT_LOG	5	/* log2(# address strings in array) */
184#define ADDR_STR_COUNT		(1 << ADDR_STR_COUNT_LOG)
185#define ADDR_STR_COUNT_MASK	(ADDR_STR_COUNT - 1)
186#define MAX_ADDR_STR_LEN	64	/* 54 is enough */
187
188static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
189static atomic_t addr_str_seq = ATOMIC_INIT(0);
190
191static struct page *zero_page;		/* used in certain error cases */
192
193const char *ceph_pr_addr(const struct sockaddr_storage *ss)
194{
195	int i;
196	char *s;
197	struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
198	struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
199
200	i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
201	s = addr_str[i];
202
203	switch (ss->ss_family) {
204	case AF_INET:
205		snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
206			 ntohs(in4->sin_port));
207		break;
208
209	case AF_INET6:
210		snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
211			 ntohs(in6->sin6_port));
212		break;
213
214	default:
215		snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
216			 ss->ss_family);
217	}
218
219	return s;
220}
221EXPORT_SYMBOL(ceph_pr_addr);
222
223static void encode_my_addr(struct ceph_messenger *msgr)
224{
225	memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
226	ceph_encode_addr(&msgr->my_enc_addr);
227}
228
229/*
230 * work queue for all reading and writing to/from the socket.
231 */
232static struct workqueue_struct *ceph_msgr_wq;
233
234static int ceph_msgr_slab_init(void)
235{
236	BUG_ON(ceph_msg_cache);
237	ceph_msg_cache = kmem_cache_create("ceph_msg",
238					sizeof (struct ceph_msg),
239					__alignof__(struct ceph_msg), 0, NULL);
240
241	if (!ceph_msg_cache)
242		return -ENOMEM;
243
244	BUG_ON(ceph_msg_data_cache);
245	ceph_msg_data_cache = kmem_cache_create("ceph_msg_data",
246					sizeof (struct ceph_msg_data),
247					__alignof__(struct ceph_msg_data),
248					0, NULL);
249	if (ceph_msg_data_cache)
250		return 0;
251
252	kmem_cache_destroy(ceph_msg_cache);
253	ceph_msg_cache = NULL;
254
255	return -ENOMEM;
256}
257
258static void ceph_msgr_slab_exit(void)
259{
260	BUG_ON(!ceph_msg_data_cache);
261	kmem_cache_destroy(ceph_msg_data_cache);
262	ceph_msg_data_cache = NULL;
263
264	BUG_ON(!ceph_msg_cache);
265	kmem_cache_destroy(ceph_msg_cache);
266	ceph_msg_cache = NULL;
267}
268
269static void _ceph_msgr_exit(void)
270{
271	if (ceph_msgr_wq) {
272		destroy_workqueue(ceph_msgr_wq);
273		ceph_msgr_wq = NULL;
274	}
275
276	ceph_msgr_slab_exit();
277
278	BUG_ON(zero_page == NULL);
279	kunmap(zero_page);
280	page_cache_release(zero_page);
281	zero_page = NULL;
282}
283
284int ceph_msgr_init(void)
285{
286	BUG_ON(zero_page != NULL);
287	zero_page = ZERO_PAGE(0);
288	page_cache_get(zero_page);
289
290	if (ceph_msgr_slab_init())
291		return -ENOMEM;
292
293	ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0);
294	if (ceph_msgr_wq)
295		return 0;
296
297	pr_err("msgr_init failed to create workqueue\n");
298	_ceph_msgr_exit();
299
300	return -ENOMEM;
301}
302EXPORT_SYMBOL(ceph_msgr_init);
303
304void ceph_msgr_exit(void)
305{
306	BUG_ON(ceph_msgr_wq == NULL);
307
308	_ceph_msgr_exit();
309}
310EXPORT_SYMBOL(ceph_msgr_exit);
311
312void ceph_msgr_flush(void)
313{
314	flush_workqueue(ceph_msgr_wq);
315}
316EXPORT_SYMBOL(ceph_msgr_flush);
317
318/* Connection socket state transition functions */
319
320static void con_sock_state_init(struct ceph_connection *con)
321{
322	int old_state;
323
324	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
325	if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
326		printk("%s: unexpected old state %d\n", __func__, old_state);
327	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
328	     CON_SOCK_STATE_CLOSED);
329}
330
331static void con_sock_state_connecting(struct ceph_connection *con)
332{
333	int old_state;
334
335	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
336	if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
337		printk("%s: unexpected old state %d\n", __func__, old_state);
338	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
339	     CON_SOCK_STATE_CONNECTING);
340}
341
342static void con_sock_state_connected(struct ceph_connection *con)
343{
344	int old_state;
345
346	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
347	if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
348		printk("%s: unexpected old state %d\n", __func__, old_state);
349	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
350	     CON_SOCK_STATE_CONNECTED);
351}
352
353static void con_sock_state_closing(struct ceph_connection *con)
354{
355	int old_state;
356
357	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
358	if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
359			old_state != CON_SOCK_STATE_CONNECTED &&
360			old_state != CON_SOCK_STATE_CLOSING))
361		printk("%s: unexpected old state %d\n", __func__, old_state);
362	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
363	     CON_SOCK_STATE_CLOSING);
364}
365
366static void con_sock_state_closed(struct ceph_connection *con)
367{
368	int old_state;
369
370	old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
371	if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
372		    old_state != CON_SOCK_STATE_CLOSING &&
373		    old_state != CON_SOCK_STATE_CONNECTING &&
374		    old_state != CON_SOCK_STATE_CLOSED))
375		printk("%s: unexpected old state %d\n", __func__, old_state);
376	dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
377	     CON_SOCK_STATE_CLOSED);
378}
379
380/*
381 * socket callback functions
382 */
383
384/* data available on socket, or listen socket received a connect */
385static void ceph_sock_data_ready(struct sock *sk, int count_unused)
386{
387	struct ceph_connection *con = sk->sk_user_data;
388	if (atomic_read(&con->msgr->stopping)) {
389		return;
390	}
391
392	if (sk->sk_state != TCP_CLOSE_WAIT) {
393		dout("%s on %p state = %lu, queueing work\n", __func__,
394		     con, con->state);
395		queue_con(con);
396	}
397}
398
399/* socket has buffer space for writing */
400static void ceph_sock_write_space(struct sock *sk)
401{
402	struct ceph_connection *con = sk->sk_user_data;
403
404	/* only queue to workqueue if there is data we want to write,
405	 * and there is sufficient space in the socket buffer to accept
406	 * more data.  clear SOCK_NOSPACE so that ceph_sock_write_space()
407	 * doesn't get called again until try_write() fills the socket
408	 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
409	 * and net/core/stream.c:sk_stream_write_space().
410	 */
411	if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
412		if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
413			dout("%s %p queueing write work\n", __func__, con);
414			clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
415			queue_con(con);
416		}
417	} else {
418		dout("%s %p nothing to write\n", __func__, con);
419	}
420}
421
422/* socket's state has changed */
423static void ceph_sock_state_change(struct sock *sk)
424{
425	struct ceph_connection *con = sk->sk_user_data;
426
427	dout("%s %p state = %lu sk_state = %u\n", __func__,
428	     con, con->state, sk->sk_state);
429
430	switch (sk->sk_state) {
431	case TCP_CLOSE:
432		dout("%s TCP_CLOSE\n", __func__);
433	case TCP_CLOSE_WAIT:
434		dout("%s TCP_CLOSE_WAIT\n", __func__);
435		con_sock_state_closing(con);
436		con_flag_set(con, CON_FLAG_SOCK_CLOSED);
437		queue_con(con);
438		break;
439	case TCP_ESTABLISHED:
440		dout("%s TCP_ESTABLISHED\n", __func__);
441		con_sock_state_connected(con);
442		queue_con(con);
443		break;
444	default:	/* Everything else is uninteresting */
445		break;
446	}
447}
448
449/*
450 * set up socket callbacks
451 */
452static void set_sock_callbacks(struct socket *sock,
453			       struct ceph_connection *con)
454{
455	struct sock *sk = sock->sk;
456	sk->sk_user_data = con;
457	sk->sk_data_ready = ceph_sock_data_ready;
458	sk->sk_write_space = ceph_sock_write_space;
459	sk->sk_state_change = ceph_sock_state_change;
460}
461
462
463/*
464 * socket helpers
465 */
466
467/*
468 * initiate connection to a remote socket.
469 */
470static int ceph_tcp_connect(struct ceph_connection *con)
471{
472	struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
473	struct socket *sock;
474	int ret;
475
476	BUG_ON(con->sock);
477	ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
478			       IPPROTO_TCP, &sock);
479	if (ret)
480		return ret;
481	sock->sk->sk_allocation = GFP_NOFS;
482
483#ifdef CONFIG_LOCKDEP
484	lockdep_set_class(&sock->sk->sk_lock, &socket_class);
485#endif
486
487	set_sock_callbacks(sock, con);
488
489	dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
490
491	con_sock_state_connecting(con);
492	ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
493				 O_NONBLOCK);
494	if (ret == -EINPROGRESS) {
495		dout("connect %s EINPROGRESS sk_state = %u\n",
496		     ceph_pr_addr(&con->peer_addr.in_addr),
497		     sock->sk->sk_state);
498	} else if (ret < 0) {
499		pr_err("connect %s error %d\n",
500		       ceph_pr_addr(&con->peer_addr.in_addr), ret);
501		sock_release(sock);
502		con->error_msg = "connect error";
503
504		return ret;
505	}
506	con->sock = sock;
507	return 0;
508}
509
510static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
511{
512	struct kvec iov = {buf, len};
513	struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
514	int r;
515
516	r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
517	if (r == -EAGAIN)
518		r = 0;
519	return r;
520}
521
522static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
523		     int page_offset, size_t length)
524{
525	void *kaddr;
526	int ret;
527
528	BUG_ON(page_offset + length > PAGE_SIZE);
529
530	kaddr = kmap(page);
531	BUG_ON(!kaddr);
532	ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
533	kunmap(page);
534
535	return ret;
536}
537
538/*
539 * write something.  @more is true if caller will be sending more data
540 * shortly.
541 */
542static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
543		     size_t kvlen, size_t len, int more)
544{
545	struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
546	int r;
547
548	if (more)
549		msg.msg_flags |= MSG_MORE;
550	else
551		msg.msg_flags |= MSG_EOR;  /* superfluous, but what the hell */
552
553	r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
554	if (r == -EAGAIN)
555		r = 0;
556	return r;
557}
558
559static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
560		     int offset, size_t size, bool more)
561{
562	int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
563	int ret;
564
565	ret = kernel_sendpage(sock, page, offset, size, flags);
566	if (ret == -EAGAIN)
567		ret = 0;
568
569	return ret;
570}
571
572
573/*
574 * Shutdown/close the socket for the given connection.
575 */
576static int con_close_socket(struct ceph_connection *con)
577{
578	int rc = 0;
579
580	dout("con_close_socket on %p sock %p\n", con, con->sock);
581	if (con->sock) {
582		rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
583		sock_release(con->sock);
584		con->sock = NULL;
585	}
586
587	/*
588	 * Forcibly clear the SOCK_CLOSED flag.  It gets set
589	 * independent of the connection mutex, and we could have
590	 * received a socket close event before we had the chance to
591	 * shut the socket down.
592	 */
593	con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
594
595	con_sock_state_closed(con);
596	return rc;
597}
598
599/*
600 * Reset a connection.  Discard all incoming and outgoing messages
601 * and clear *_seq state.
602 */
603static void ceph_msg_remove(struct ceph_msg *msg)
604{
605	list_del_init(&msg->list_head);
606	BUG_ON(msg->con == NULL);
607	msg->con->ops->put(msg->con);
608	msg->con = NULL;
609
610	ceph_msg_put(msg);
611}
612static void ceph_msg_remove_list(struct list_head *head)
613{
614	while (!list_empty(head)) {
615		struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
616							list_head);
617		ceph_msg_remove(msg);
618	}
619}
620
621static void reset_connection(struct ceph_connection *con)
622{
623	/* reset connection, out_queue, msg_ and connect_seq */
624	/* discard existing out_queue and msg_seq */
625	dout("reset_connection %p\n", con);
626	ceph_msg_remove_list(&con->out_queue);
627	ceph_msg_remove_list(&con->out_sent);
628
629	if (con->in_msg) {
630		BUG_ON(con->in_msg->con != con);
631		con->in_msg->con = NULL;
632		ceph_msg_put(con->in_msg);
633		con->in_msg = NULL;
634		con->ops->put(con);
635	}
636
637	con->connect_seq = 0;
638	con->out_seq = 0;
639	if (con->out_msg) {
640		ceph_msg_put(con->out_msg);
641		con->out_msg = NULL;
642	}
643	con->in_seq = 0;
644	con->in_seq_acked = 0;
645}
646
647/*
648 * mark a peer down.  drop any open connections.
649 */
650void ceph_con_close(struct ceph_connection *con)
651{
652	mutex_lock(&con->mutex);
653	dout("con_close %p peer %s\n", con,
654	     ceph_pr_addr(&con->peer_addr.in_addr));
655	con->state = CON_STATE_CLOSED;
656
657	con_flag_clear(con, CON_FLAG_LOSSYTX);	/* so we retry next connect */
658	con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
659	con_flag_clear(con, CON_FLAG_WRITE_PENDING);
660	con_flag_clear(con, CON_FLAG_BACKOFF);
661
662	reset_connection(con);
663	con->peer_global_seq = 0;
664	cancel_delayed_work(&con->work);
665	con_close_socket(con);
666	mutex_unlock(&con->mutex);
667}
668EXPORT_SYMBOL(ceph_con_close);
669
670/*
671 * Reopen a closed connection, with a new peer address.
672 */
673void ceph_con_open(struct ceph_connection *con,
674		   __u8 entity_type, __u64 entity_num,
675		   struct ceph_entity_addr *addr)
676{
677	mutex_lock(&con->mutex);
678	dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
679
680	WARN_ON(con->state != CON_STATE_CLOSED);
681	con->state = CON_STATE_PREOPEN;
682
683	con->peer_name.type = (__u8) entity_type;
684	con->peer_name.num = cpu_to_le64(entity_num);
685
686	memcpy(&con->peer_addr, addr, sizeof(*addr));
687	con->delay = 0;      /* reset backoff memory */
688	mutex_unlock(&con->mutex);
689	queue_con(con);
690}
691EXPORT_SYMBOL(ceph_con_open);
692
693/*
694 * return true if this connection ever successfully opened
695 */
696bool ceph_con_opened(struct ceph_connection *con)
697{
698	return con->connect_seq > 0;
699}
700
701/*
702 * initialize a new connection.
703 */
704void ceph_con_init(struct ceph_connection *con, void *private,
705	const struct ceph_connection_operations *ops,
706	struct ceph_messenger *msgr)
707{
708	dout("con_init %p\n", con);
709	memset(con, 0, sizeof(*con));
710	con->private = private;
711	con->ops = ops;
712	con->msgr = msgr;
713
714	con_sock_state_init(con);
715
716	mutex_init(&con->mutex);
717	INIT_LIST_HEAD(&con->out_queue);
718	INIT_LIST_HEAD(&con->out_sent);
719	INIT_DELAYED_WORK(&con->work, con_work);
720
721	con->state = CON_STATE_CLOSED;
722}
723EXPORT_SYMBOL(ceph_con_init);
724
725
726/*
727 * We maintain a global counter to order connection attempts.  Get
728 * a unique seq greater than @gt.
729 */
730static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
731{
732	u32 ret;
733
734	spin_lock(&msgr->global_seq_lock);
735	if (msgr->global_seq < gt)
736		msgr->global_seq = gt;
737	ret = ++msgr->global_seq;
738	spin_unlock(&msgr->global_seq_lock);
739	return ret;
740}
741
742static void con_out_kvec_reset(struct ceph_connection *con)
743{
744	con->out_kvec_left = 0;
745	con->out_kvec_bytes = 0;
746	con->out_kvec_cur = &con->out_kvec[0];
747}
748
749static void con_out_kvec_add(struct ceph_connection *con,
750				size_t size, void *data)
751{
752	int index;
753
754	index = con->out_kvec_left;
755	BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
756
757	con->out_kvec[index].iov_len = size;
758	con->out_kvec[index].iov_base = data;
759	con->out_kvec_left++;
760	con->out_kvec_bytes += size;
761}
762
763#ifdef CONFIG_BLOCK
764
765/*
766 * For a bio data item, a piece is whatever remains of the next
767 * entry in the current bio iovec, or the first entry in the next
768 * bio in the list.
769 */
770static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
771					size_t length)
772{
773	struct ceph_msg_data *data = cursor->data;
774	struct bio *bio;
775
776	BUG_ON(data->type != CEPH_MSG_DATA_BIO);
777
778	bio = data->bio;
779	BUG_ON(!bio);
780	BUG_ON(!bio->bi_vcnt);
781
782	cursor->resid = min(length, data->bio_length);
783	cursor->bio = bio;
784	cursor->vector_index = 0;
785	cursor->vector_offset = 0;
786	cursor->last_piece = length <= bio->bi_io_vec[0].bv_len;
787}
788
789static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
790						size_t *page_offset,
791						size_t *length)
792{
793	struct ceph_msg_data *data = cursor->data;
794	struct bio *bio;
795	struct bio_vec *bio_vec;
796	unsigned int index;
797
798	BUG_ON(data->type != CEPH_MSG_DATA_BIO);
799
800	bio = cursor->bio;
801	BUG_ON(!bio);
802
803	index = cursor->vector_index;
804	BUG_ON(index >= (unsigned int) bio->bi_vcnt);
805
806	bio_vec = &bio->bi_io_vec[index];
807	BUG_ON(cursor->vector_offset >= bio_vec->bv_len);
808	*page_offset = (size_t) (bio_vec->bv_offset + cursor->vector_offset);
809	BUG_ON(*page_offset >= PAGE_SIZE);
810	if (cursor->last_piece) /* pagelist offset is always 0 */
811		*length = cursor->resid;
812	else
813		*length = (size_t) (bio_vec->bv_len - cursor->vector_offset);
814	BUG_ON(*length > cursor->resid);
815	BUG_ON(*page_offset + *length > PAGE_SIZE);
816
817	return bio_vec->bv_page;
818}
819
820static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
821					size_t bytes)
822{
823	struct bio *bio;
824	struct bio_vec *bio_vec;
825	unsigned int index;
826
827	BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
828
829	bio = cursor->bio;
830	BUG_ON(!bio);
831
832	index = cursor->vector_index;
833	BUG_ON(index >= (unsigned int) bio->bi_vcnt);
834	bio_vec = &bio->bi_io_vec[index];
835
836	/* Advance the cursor offset */
837
838	BUG_ON(cursor->resid < bytes);
839	cursor->resid -= bytes;
840	cursor->vector_offset += bytes;
841	if (cursor->vector_offset < bio_vec->bv_len)
842		return false;	/* more bytes to process in this segment */
843	BUG_ON(cursor->vector_offset != bio_vec->bv_len);
844
845	/* Move on to the next segment, and possibly the next bio */
846
847	if (++index == (unsigned int) bio->bi_vcnt) {
848		bio = bio->bi_next;
849		index = 0;
850	}
851	cursor->bio = bio;
852	cursor->vector_index = index;
853	cursor->vector_offset = 0;
854
855	if (!cursor->last_piece) {
856		BUG_ON(!cursor->resid);
857		BUG_ON(!bio);
858		/* A short read is OK, so use <= rather than == */
859		if (cursor->resid <= bio->bi_io_vec[index].bv_len)
860			cursor->last_piece = true;
861	}
862
863	return true;
864}
865#endif /* CONFIG_BLOCK */
866
867/*
868 * For a page array, a piece comes from the first page in the array
869 * that has not already been fully consumed.
870 */
871static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
872					size_t length)
873{
874	struct ceph_msg_data *data = cursor->data;
875	int page_count;
876
877	BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
878
879	BUG_ON(!data->pages);
880	BUG_ON(!data->length);
881
882	cursor->resid = min(length, data->length);
883	page_count = calc_pages_for(data->alignment, (u64)data->length);
884	cursor->page_offset = data->alignment & ~PAGE_MASK;
885	cursor->page_index = 0;
886	BUG_ON(page_count > (int)USHRT_MAX);
887	cursor->page_count = (unsigned short)page_count;
888	BUG_ON(length > SIZE_MAX - cursor->page_offset);
889	cursor->last_piece = (size_t)cursor->page_offset + length <= PAGE_SIZE;
890}
891
892static struct page *
893ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
894					size_t *page_offset, size_t *length)
895{
896	struct ceph_msg_data *data = cursor->data;
897
898	BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
899
900	BUG_ON(cursor->page_index >= cursor->page_count);
901	BUG_ON(cursor->page_offset >= PAGE_SIZE);
902
903	*page_offset = cursor->page_offset;
904	if (cursor->last_piece)
905		*length = cursor->resid;
906	else
907		*length = PAGE_SIZE - *page_offset;
908
909	return data->pages[cursor->page_index];
910}
911
912static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
913						size_t bytes)
914{
915	BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
916
917	BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
918
919	/* Advance the cursor page offset */
920
921	cursor->resid -= bytes;
922	cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
923	if (!bytes || cursor->page_offset)
924		return false;	/* more bytes to process in the current page */
925
926	/* Move on to the next page; offset is already at 0 */
927
928	BUG_ON(cursor->page_index >= cursor->page_count);
929	cursor->page_index++;
930	cursor->last_piece = cursor->resid <= PAGE_SIZE;
931
932	return true;
933}
934
935/*
936 * For a pagelist, a piece is whatever remains to be consumed in the
937 * first page in the list, or the front of the next page.
938 */
939static void
940ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
941					size_t length)
942{
943	struct ceph_msg_data *data = cursor->data;
944	struct ceph_pagelist *pagelist;
945	struct page *page;
946
947	BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
948
949	pagelist = data->pagelist;
950	BUG_ON(!pagelist);
951
952	if (!length)
953		return;		/* pagelist can be assigned but empty */
954
955	BUG_ON(list_empty(&pagelist->head));
956	page = list_first_entry(&pagelist->head, struct page, lru);
957
958	cursor->resid = min(length, pagelist->length);
959	cursor->page = page;
960	cursor->offset = 0;
961	cursor->last_piece = cursor->resid <= PAGE_SIZE;
962}
963
964static struct page *
965ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
966				size_t *page_offset, size_t *length)
967{
968	struct ceph_msg_data *data = cursor->data;
969	struct ceph_pagelist *pagelist;
970
971	BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
972
973	pagelist = data->pagelist;
974	BUG_ON(!pagelist);
975
976	BUG_ON(!cursor->page);
977	BUG_ON(cursor->offset + cursor->resid != pagelist->length);
978
979	/* offset of first page in pagelist is always 0 */
980	*page_offset = cursor->offset & ~PAGE_MASK;
981	if (cursor->last_piece)
982		*length = cursor->resid;
983	else
984		*length = PAGE_SIZE - *page_offset;
985
986	return cursor->page;
987}
988
989static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
990						size_t bytes)
991{
992	struct ceph_msg_data *data = cursor->data;
993	struct ceph_pagelist *pagelist;
994
995	BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
996
997	pagelist = data->pagelist;
998	BUG_ON(!pagelist);
999
1000	BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1001	BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1002
1003	/* Advance the cursor offset */
1004
1005	cursor->resid -= bytes;
1006	cursor->offset += bytes;
1007	/* offset of first page in pagelist is always 0 */
1008	if (!bytes || cursor->offset & ~PAGE_MASK)
1009		return false;	/* more bytes to process in the current page */
1010
1011	/* Move on to the next page */
1012
1013	BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1014	cursor->page = list_entry_next(cursor->page, lru);
1015	cursor->last_piece = cursor->resid <= PAGE_SIZE;
1016
1017	return true;
1018}
1019
1020/*
1021 * Message data is handled (sent or received) in pieces, where each
1022 * piece resides on a single page.  The network layer might not
1023 * consume an entire piece at once.  A data item's cursor keeps
1024 * track of which piece is next to process and how much remains to
1025 * be processed in that piece.  It also tracks whether the current
1026 * piece is the last one in the data item.
1027 */
1028static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1029{
1030	size_t length = cursor->total_resid;
1031
1032	switch (cursor->data->type) {
1033	case CEPH_MSG_DATA_PAGELIST:
1034		ceph_msg_data_pagelist_cursor_init(cursor, length);
1035		break;
1036	case CEPH_MSG_DATA_PAGES:
1037		ceph_msg_data_pages_cursor_init(cursor, length);
1038		break;
1039#ifdef CONFIG_BLOCK
1040	case CEPH_MSG_DATA_BIO:
1041		ceph_msg_data_bio_cursor_init(cursor, length);
1042		break;
1043#endif /* CONFIG_BLOCK */
1044	case CEPH_MSG_DATA_NONE:
1045	default:
1046		/* BUG(); */
1047		break;
1048	}
1049	cursor->need_crc = true;
1050}
1051
1052static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1053{
1054	struct ceph_msg_data_cursor *cursor = &msg->cursor;
1055	struct ceph_msg_data *data;
1056
1057	BUG_ON(!length);
1058	BUG_ON(length > msg->data_length);
1059	BUG_ON(list_empty(&msg->data));
1060
1061	cursor->data_head = &msg->data;
1062	cursor->total_resid = length;
1063	data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1064	cursor->data = data;
1065
1066	__ceph_msg_data_cursor_init(cursor);
1067}
1068
1069/*
1070 * Return the page containing the next piece to process for a given
1071 * data item, and supply the page offset and length of that piece.
1072 * Indicate whether this is the last piece in this data item.
1073 */
1074static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1075					size_t *page_offset, size_t *length,
1076					bool *last_piece)
1077{
1078	struct page *page;
1079
1080	switch (cursor->data->type) {
1081	case CEPH_MSG_DATA_PAGELIST:
1082		page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1083		break;
1084	case CEPH_MSG_DATA_PAGES:
1085		page = ceph_msg_data_pages_next(cursor, page_offset, length);
1086		break;
1087#ifdef CONFIG_BLOCK
1088	case CEPH_MSG_DATA_BIO:
1089		page = ceph_msg_data_bio_next(cursor, page_offset, length);
1090		break;
1091#endif /* CONFIG_BLOCK */
1092	case CEPH_MSG_DATA_NONE:
1093	default:
1094		page = NULL;
1095		break;
1096	}
1097	BUG_ON(!page);
1098	BUG_ON(*page_offset + *length > PAGE_SIZE);
1099	BUG_ON(!*length);
1100	if (last_piece)
1101		*last_piece = cursor->last_piece;
1102
1103	return page;
1104}
1105
1106/*
1107 * Returns true if the result moves the cursor on to the next piece
1108 * of the data item.
1109 */
1110static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1111				size_t bytes)
1112{
1113	bool new_piece;
1114
1115	BUG_ON(bytes > cursor->resid);
1116	switch (cursor->data->type) {
1117	case CEPH_MSG_DATA_PAGELIST:
1118		new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1119		break;
1120	case CEPH_MSG_DATA_PAGES:
1121		new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1122		break;
1123#ifdef CONFIG_BLOCK
1124	case CEPH_MSG_DATA_BIO:
1125		new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1126		break;
1127#endif /* CONFIG_BLOCK */
1128	case CEPH_MSG_DATA_NONE:
1129	default:
1130		BUG();
1131		break;
1132	}
1133	cursor->total_resid -= bytes;
1134
1135	if (!cursor->resid && cursor->total_resid) {
1136		WARN_ON(!cursor->last_piece);
1137		BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1138		cursor->data = list_entry_next(cursor->data, links);
1139		__ceph_msg_data_cursor_init(cursor);
1140		new_piece = true;
1141	}
1142	cursor->need_crc = new_piece;
1143
1144	return new_piece;
1145}
1146
1147static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
1148{
1149	BUG_ON(!msg);
1150	BUG_ON(!data_len);
1151
1152	/* Initialize data cursor */
1153
1154	ceph_msg_data_cursor_init(msg, (size_t)data_len);
1155}
1156
1157/*
1158 * Prepare footer for currently outgoing message, and finish things
1159 * off.  Assumes out_kvec* are already valid.. we just add on to the end.
1160 */
1161static void prepare_write_message_footer(struct ceph_connection *con)
1162{
1163	struct ceph_msg *m = con->out_msg;
1164	int v = con->out_kvec_left;
1165
1166	m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1167
1168	dout("prepare_write_message_footer %p\n", con);
1169	con->out_kvec_is_msg = true;
1170	con->out_kvec[v].iov_base = &m->footer;
1171	con->out_kvec[v].iov_len = sizeof(m->footer);
1172	con->out_kvec_bytes += sizeof(m->footer);
1173	con->out_kvec_left++;
1174	con->out_more = m->more_to_follow;
1175	con->out_msg_done = true;
1176}
1177
1178/*
1179 * Prepare headers for the next outgoing message.
1180 */
1181static void prepare_write_message(struct ceph_connection *con)
1182{
1183	struct ceph_msg *m;
1184	u32 crc;
1185
1186	con_out_kvec_reset(con);
1187	con->out_kvec_is_msg = true;
1188	con->out_msg_done = false;
1189
1190	/* Sneak an ack in there first?  If we can get it into the same
1191	 * TCP packet that's a good thing. */
1192	if (con->in_seq > con->in_seq_acked) {
1193		con->in_seq_acked = con->in_seq;
1194		con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1195		con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1196		con_out_kvec_add(con, sizeof (con->out_temp_ack),
1197			&con->out_temp_ack);
1198	}
1199
1200	BUG_ON(list_empty(&con->out_queue));
1201	m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1202	con->out_msg = m;
1203	BUG_ON(m->con != con);
1204
1205	/* put message on sent list */
1206	ceph_msg_get(m);
1207	list_move_tail(&m->list_head, &con->out_sent);
1208
1209	/*
1210	 * only assign outgoing seq # if we haven't sent this message
1211	 * yet.  if it is requeued, resend with it's original seq.
1212	 */
1213	if (m->needs_out_seq) {
1214		m->hdr.seq = cpu_to_le64(++con->out_seq);
1215		m->needs_out_seq = false;
1216	}
1217	WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
1218
1219	dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1220	     m, con->out_seq, le16_to_cpu(m->hdr.type),
1221	     le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1222	     m->data_length);
1223	BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1224
1225	/* tag + hdr + front + middle */
1226	con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1227	con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1228	con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1229
1230	if (m->middle)
1231		con_out_kvec_add(con, m->middle->vec.iov_len,
1232			m->middle->vec.iov_base);
1233
1234	/* fill in crc (except data pages), footer */
1235	crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1236	con->out_msg->hdr.crc = cpu_to_le32(crc);
1237	con->out_msg->footer.flags = 0;
1238
1239	crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1240	con->out_msg->footer.front_crc = cpu_to_le32(crc);
1241	if (m->middle) {
1242		crc = crc32c(0, m->middle->vec.iov_base,
1243				m->middle->vec.iov_len);
1244		con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1245	} else
1246		con->out_msg->footer.middle_crc = 0;
1247	dout("%s front_crc %u middle_crc %u\n", __func__,
1248	     le32_to_cpu(con->out_msg->footer.front_crc),
1249	     le32_to_cpu(con->out_msg->footer.middle_crc));
1250
1251	/* is there a data payload? */
1252	con->out_msg->footer.data_crc = 0;
1253	if (m->data_length) {
1254		prepare_message_data(con->out_msg, m->data_length);
1255		con->out_more = 1;  /* data + footer will follow */
1256	} else {
1257		/* no, queue up footer too and be done */
1258		prepare_write_message_footer(con);
1259	}
1260
1261	con_flag_set(con, CON_FLAG_WRITE_PENDING);
1262}
1263
1264/*
1265 * Prepare an ack.
1266 */
1267static void prepare_write_ack(struct ceph_connection *con)
1268{
1269	dout("prepare_write_ack %p %llu -> %llu\n", con,
1270	     con->in_seq_acked, con->in_seq);
1271	con->in_seq_acked = con->in_seq;
1272
1273	con_out_kvec_reset(con);
1274
1275	con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1276
1277	con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1278	con_out_kvec_add(con, sizeof (con->out_temp_ack),
1279				&con->out_temp_ack);
1280
1281	con->out_more = 1;  /* more will follow.. eventually.. */
1282	con_flag_set(con, CON_FLAG_WRITE_PENDING);
1283}
1284
1285/*
1286 * Prepare to share the seq during handshake
1287 */
1288static void prepare_write_seq(struct ceph_connection *con)
1289{
1290	dout("prepare_write_seq %p %llu -> %llu\n", con,
1291	     con->in_seq_acked, con->in_seq);
1292	con->in_seq_acked = con->in_seq;
1293
1294	con_out_kvec_reset(con);
1295
1296	con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1297	con_out_kvec_add(con, sizeof (con->out_temp_ack),
1298			 &con->out_temp_ack);
1299
1300	con_flag_set(con, CON_FLAG_WRITE_PENDING);
1301}
1302
1303/*
1304 * Prepare to write keepalive byte.
1305 */
1306static void prepare_write_keepalive(struct ceph_connection *con)
1307{
1308	dout("prepare_write_keepalive %p\n", con);
1309	con_out_kvec_reset(con);
1310	con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
1311	con_flag_set(con, CON_FLAG_WRITE_PENDING);
1312}
1313
1314/*
1315 * Connection negotiation.
1316 */
1317
1318static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1319						int *auth_proto)
1320{
1321	struct ceph_auth_handshake *auth;
1322
1323	if (!con->ops->get_authorizer) {
1324		con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1325		con->out_connect.authorizer_len = 0;
1326		return NULL;
1327	}
1328
1329	/* Can't hold the mutex while getting authorizer */
1330	mutex_unlock(&con->mutex);
1331	auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
1332	mutex_lock(&con->mutex);
1333
1334	if (IS_ERR(auth))
1335		return auth;
1336	if (con->state != CON_STATE_NEGOTIATING)
1337		return ERR_PTR(-EAGAIN);
1338
1339	con->auth_reply_buf = auth->authorizer_reply_buf;
1340	con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
1341	return auth;
1342}
1343
1344/*
1345 * We connected to a peer and are saying hello.
1346 */
1347static void prepare_write_banner(struct ceph_connection *con)
1348{
1349	con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1350	con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1351					&con->msgr->my_enc_addr);
1352
1353	con->out_more = 0;
1354	con_flag_set(con, CON_FLAG_WRITE_PENDING);
1355}
1356
1357static int prepare_write_connect(struct ceph_connection *con)
1358{
1359	unsigned int global_seq = get_global_seq(con->msgr, 0);
1360	int proto;
1361	int auth_proto;
1362	struct ceph_auth_handshake *auth;
1363
1364	switch (con->peer_name.type) {
1365	case CEPH_ENTITY_TYPE_MON:
1366		proto = CEPH_MONC_PROTOCOL;
1367		break;
1368	case CEPH_ENTITY_TYPE_OSD:
1369		proto = CEPH_OSDC_PROTOCOL;
1370		break;
1371	case CEPH_ENTITY_TYPE_MDS:
1372		proto = CEPH_MDSC_PROTOCOL;
1373		break;
1374	default:
1375		BUG();
1376	}
1377
1378	dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1379	     con->connect_seq, global_seq, proto);
1380
1381	con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
1382	con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1383	con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1384	con->out_connect.global_seq = cpu_to_le32(global_seq);
1385	con->out_connect.protocol_version = cpu_to_le32(proto);
1386	con->out_connect.flags = 0;
1387
1388	auth_proto = CEPH_AUTH_UNKNOWN;
1389	auth = get_connect_authorizer(con, &auth_proto);
1390	if (IS_ERR(auth))
1391		return PTR_ERR(auth);
1392
1393	con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1394	con->out_connect.authorizer_len = auth ?
1395		cpu_to_le32(auth->authorizer_buf_len) : 0;
1396
1397	con_out_kvec_add(con, sizeof (con->out_connect),
1398					&con->out_connect);
1399	if (auth && auth->authorizer_buf_len)
1400		con_out_kvec_add(con, auth->authorizer_buf_len,
1401					auth->authorizer_buf);
1402
1403	con->out_more = 0;
1404	con_flag_set(con, CON_FLAG_WRITE_PENDING);
1405
1406	return 0;
1407}
1408
1409/*
1410 * write as much of pending kvecs to the socket as we can.
1411 *  1 -> done
1412 *  0 -> socket full, but more to do
1413 * <0 -> error
1414 */
1415static int write_partial_kvec(struct ceph_connection *con)
1416{
1417	int ret;
1418
1419	dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1420	while (con->out_kvec_bytes > 0) {
1421		ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1422				       con->out_kvec_left, con->out_kvec_bytes,
1423				       con->out_more);
1424		if (ret <= 0)
1425			goto out;
1426		con->out_kvec_bytes -= ret;
1427		if (con->out_kvec_bytes == 0)
1428			break;            /* done */
1429
1430		/* account for full iov entries consumed */
1431		while (ret >= con->out_kvec_cur->iov_len) {
1432			BUG_ON(!con->out_kvec_left);
1433			ret -= con->out_kvec_cur->iov_len;
1434			con->out_kvec_cur++;
1435			con->out_kvec_left--;
1436		}
1437		/* and for a partially-consumed entry */
1438		if (ret) {
1439			con->out_kvec_cur->iov_len -= ret;
1440			con->out_kvec_cur->iov_base += ret;
1441		}
1442	}
1443	con->out_kvec_left = 0;
1444	con->out_kvec_is_msg = false;
1445	ret = 1;
1446out:
1447	dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1448	     con->out_kvec_bytes, con->out_kvec_left, ret);
1449	return ret;  /* done! */
1450}
1451
1452static u32 ceph_crc32c_page(u32 crc, struct page *page,
1453				unsigned int page_offset,
1454				unsigned int length)
1455{
1456	char *kaddr;
1457
1458	kaddr = kmap(page);
1459	BUG_ON(kaddr == NULL);
1460	crc = crc32c(crc, kaddr + page_offset, length);
1461	kunmap(page);
1462
1463	return crc;
1464}
1465/*
1466 * Write as much message data payload as we can.  If we finish, queue
1467 * up the footer.
1468 *  1 -> done, footer is now queued in out_kvec[].
1469 *  0 -> socket full, but more to do
1470 * <0 -> error
1471 */
1472static int write_partial_message_data(struct ceph_connection *con)
1473{
1474	struct ceph_msg *msg = con->out_msg;
1475	struct ceph_msg_data_cursor *cursor = &msg->cursor;
1476	bool do_datacrc = !con->msgr->nocrc;
1477	u32 crc;
1478
1479	dout("%s %p msg %p\n", __func__, con, msg);
1480
1481	if (list_empty(&msg->data))
1482		return -EINVAL;
1483
1484	/*
1485	 * Iterate through each page that contains data to be
1486	 * written, and send as much as possible for each.
1487	 *
1488	 * If we are calculating the data crc (the default), we will
1489	 * need to map the page.  If we have no pages, they have
1490	 * been revoked, so use the zero page.
1491	 */
1492	crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1493	while (cursor->resid) {
1494		struct page *page;
1495		size_t page_offset;
1496		size_t length;
1497		bool last_piece;
1498		bool need_crc;
1499		int ret;
1500
1501		page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
1502							&last_piece);
1503		ret = ceph_tcp_sendpage(con->sock, page, page_offset,
1504				      length, last_piece);
1505		if (ret <= 0) {
1506			if (do_datacrc)
1507				msg->footer.data_crc = cpu_to_le32(crc);
1508
1509			return ret;
1510		}
1511		if (do_datacrc && cursor->need_crc)
1512			crc = ceph_crc32c_page(crc, page, page_offset, length);
1513		need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret);
1514	}
1515
1516	dout("%s %p msg %p done\n", __func__, con, msg);
1517
1518	/* prepare and queue up footer, too */
1519	if (do_datacrc)
1520		msg->footer.data_crc = cpu_to_le32(crc);
1521	else
1522		msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1523	con_out_kvec_reset(con);
1524	prepare_write_message_footer(con);
1525
1526	return 1;	/* must return > 0 to indicate success */
1527}
1528
1529/*
1530 * write some zeros
1531 */
1532static int write_partial_skip(struct ceph_connection *con)
1533{
1534	int ret;
1535
1536	while (con->out_skip > 0) {
1537		size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
1538
1539		ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
1540		if (ret <= 0)
1541			goto out;
1542		con->out_skip -= ret;
1543	}
1544	ret = 1;
1545out:
1546	return ret;
1547}
1548
1549/*
1550 * Prepare to read connection handshake, or an ack.
1551 */
1552static void prepare_read_banner(struct ceph_connection *con)
1553{
1554	dout("prepare_read_banner %p\n", con);
1555	con->in_base_pos = 0;
1556}
1557
1558static void prepare_read_connect(struct ceph_connection *con)
1559{
1560	dout("prepare_read_connect %p\n", con);
1561	con->in_base_pos = 0;
1562}
1563
1564static void prepare_read_ack(struct ceph_connection *con)
1565{
1566	dout("prepare_read_ack %p\n", con);
1567	con->in_base_pos = 0;
1568}
1569
1570static void prepare_read_seq(struct ceph_connection *con)
1571{
1572	dout("prepare_read_seq %p\n", con);
1573	con->in_base_pos = 0;
1574	con->in_tag = CEPH_MSGR_TAG_SEQ;
1575}
1576
1577static void prepare_read_tag(struct ceph_connection *con)
1578{
1579	dout("prepare_read_tag %p\n", con);
1580	con->in_base_pos = 0;
1581	con->in_tag = CEPH_MSGR_TAG_READY;
1582}
1583
1584/*
1585 * Prepare to read a message.
1586 */
1587static int prepare_read_message(struct ceph_connection *con)
1588{
1589	dout("prepare_read_message %p\n", con);
1590	BUG_ON(con->in_msg != NULL);
1591	con->in_base_pos = 0;
1592	con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1593	return 0;
1594}
1595
1596
1597static int read_partial(struct ceph_connection *con,
1598			int end, int size, void *object)
1599{
1600	while (con->in_base_pos < end) {
1601		int left = end - con->in_base_pos;
1602		int have = size - left;
1603		int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1604		if (ret <= 0)
1605			return ret;
1606		con->in_base_pos += ret;
1607	}
1608	return 1;
1609}
1610
1611
1612/*
1613 * Read all or part of the connect-side handshake on a new connection
1614 */
1615static int read_partial_banner(struct ceph_connection *con)
1616{
1617	int size;
1618	int end;
1619	int ret;
1620
1621	dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1622
1623	/* peer's banner */
1624	size = strlen(CEPH_BANNER);
1625	end = size;
1626	ret = read_partial(con, end, size, con->in_banner);
1627	if (ret <= 0)
1628		goto out;
1629
1630	size = sizeof (con->actual_peer_addr);
1631	end += size;
1632	ret = read_partial(con, end, size, &con->actual_peer_addr);
1633	if (ret <= 0)
1634		goto out;
1635
1636	size = sizeof (con->peer_addr_for_me);
1637	end += size;
1638	ret = read_partial(con, end, size, &con->peer_addr_for_me);
1639	if (ret <= 0)
1640		goto out;
1641
1642out:
1643	return ret;
1644}
1645
1646static int read_partial_connect(struct ceph_connection *con)
1647{
1648	int size;
1649	int end;
1650	int ret;
1651
1652	dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1653
1654	size = sizeof (con->in_reply);
1655	end = size;
1656	ret = read_partial(con, end, size, &con->in_reply);
1657	if (ret <= 0)
1658		goto out;
1659
1660	size = le32_to_cpu(con->in_reply.authorizer_len);
1661	end += size;
1662	ret = read_partial(con, end, size, con->auth_reply_buf);
1663	if (ret <= 0)
1664		goto out;
1665
1666	dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1667	     con, (int)con->in_reply.tag,
1668	     le32_to_cpu(con->in_reply.connect_seq),
1669	     le32_to_cpu(con->in_reply.global_seq));
1670out:
1671	return ret;
1672
1673}
1674
1675/*
1676 * Verify the hello banner looks okay.
1677 */
1678static int verify_hello(struct ceph_connection *con)
1679{
1680	if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1681		pr_err("connect to %s got bad banner\n",
1682		       ceph_pr_addr(&con->peer_addr.in_addr));
1683		con->error_msg = "protocol error, bad banner";
1684		return -1;
1685	}
1686	return 0;
1687}
1688
1689static bool addr_is_blank(struct sockaddr_storage *ss)
1690{
1691	switch (ss->ss_family) {
1692	case AF_INET:
1693		return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1694	case AF_INET6:
1695		return
1696		     ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1697		     ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1698		     ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1699		     ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1700	}
1701	return false;
1702}
1703
1704static int addr_port(struct sockaddr_storage *ss)
1705{
1706	switch (ss->ss_family) {
1707	case AF_INET:
1708		return ntohs(((struct sockaddr_in *)ss)->sin_port);
1709	case AF_INET6:
1710		return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1711	}
1712	return 0;
1713}
1714
1715static void addr_set_port(struct sockaddr_storage *ss, int p)
1716{
1717	switch (ss->ss_family) {
1718	case AF_INET:
1719		((struct sockaddr_in *)ss)->sin_port = htons(p);
1720		break;
1721	case AF_INET6:
1722		((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1723		break;
1724	}
1725}
1726
1727/*
1728 * Unlike other *_pton function semantics, zero indicates success.
1729 */
1730static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1731		char delim, const char **ipend)
1732{
1733	struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1734	struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1735
1736	memset(ss, 0, sizeof(*ss));
1737
1738	if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1739		ss->ss_family = AF_INET;
1740		return 0;
1741	}
1742
1743	if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1744		ss->ss_family = AF_INET6;
1745		return 0;
1746	}
1747
1748	return -EINVAL;
1749}
1750
1751/*
1752 * Extract hostname string and resolve using kernel DNS facility.
1753 */
1754#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1755static int ceph_dns_resolve_name(const char *name, size_t namelen,
1756		struct sockaddr_storage *ss, char delim, const char **ipend)
1757{
1758	const char *end, *delim_p;
1759	char *colon_p, *ip_addr = NULL;
1760	int ip_len, ret;
1761
1762	/*
1763	 * The end of the hostname occurs immediately preceding the delimiter or
1764	 * the port marker (':') where the delimiter takes precedence.
1765	 */
1766	delim_p = memchr(name, delim, namelen);
1767	colon_p = memchr(name, ':', namelen);
1768
1769	if (delim_p && colon_p)
1770		end = delim_p < colon_p ? delim_p : colon_p;
1771	else if (!delim_p && colon_p)
1772		end = colon_p;
1773	else {
1774		end = delim_p;
1775		if (!end) /* case: hostname:/ */
1776			end = name + namelen;
1777	}
1778
1779	if (end <= name)
1780		return -EINVAL;
1781
1782	/* do dns_resolve upcall */
1783	ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1784	if (ip_len > 0)
1785		ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1786	else
1787		ret = -ESRCH;
1788
1789	kfree(ip_addr);
1790
1791	*ipend = end;
1792
1793	pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1794			ret, ret ? "failed" : ceph_pr_addr(ss));
1795
1796	return ret;
1797}
1798#else
1799static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1800		struct sockaddr_storage *ss, char delim, const char **ipend)
1801{
1802	return -EINVAL;
1803}
1804#endif
1805
1806/*
1807 * Parse a server name (IP or hostname). If a valid IP address is not found
1808 * then try to extract a hostname to resolve using userspace DNS upcall.
1809 */
1810static int ceph_parse_server_name(const char *name, size_t namelen,
1811			struct sockaddr_storage *ss, char delim, const char **ipend)
1812{
1813	int ret;
1814
1815	ret = ceph_pton(name, namelen, ss, delim, ipend);
1816	if (ret)
1817		ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1818
1819	return ret;
1820}
1821
1822/*
1823 * Parse an ip[:port] list into an addr array.  Use the default
1824 * monitor port if a port isn't specified.
1825 */
1826int ceph_parse_ips(const char *c, const char *end,
1827		   struct ceph_entity_addr *addr,
1828		   int max_count, int *count)
1829{
1830	int i, ret = -EINVAL;
1831	const char *p = c;
1832
1833	dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1834	for (i = 0; i < max_count; i++) {
1835		const char *ipend;
1836		struct sockaddr_storage *ss = &addr[i].in_addr;
1837		int port;
1838		char delim = ',';
1839
1840		if (*p == '[') {
1841			delim = ']';
1842			p++;
1843		}
1844
1845		ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1846		if (ret)
1847			goto bad;
1848		ret = -EINVAL;
1849
1850		p = ipend;
1851
1852		if (delim == ']') {
1853			if (*p != ']') {
1854				dout("missing matching ']'\n");
1855				goto bad;
1856			}
1857			p++;
1858		}
1859
1860		/* port? */
1861		if (p < end && *p == ':') {
1862			port = 0;
1863			p++;
1864			while (p < end && *p >= '0' && *p <= '9') {
1865				port = (port * 10) + (*p - '0');
1866				p++;
1867			}
1868			if (port > 65535 || port == 0)
1869				goto bad;
1870		} else {
1871			port = CEPH_MON_PORT;
1872		}
1873
1874		addr_set_port(ss, port);
1875
1876		dout("parse_ips got %s\n", ceph_pr_addr(ss));
1877
1878		if (p == end)
1879			break;
1880		if (*p != ',')
1881			goto bad;
1882		p++;
1883	}
1884
1885	if (p != end)
1886		goto bad;
1887
1888	if (count)
1889		*count = i + 1;
1890	return 0;
1891
1892bad:
1893	pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1894	return ret;
1895}
1896EXPORT_SYMBOL(ceph_parse_ips);
1897
1898static int process_banner(struct ceph_connection *con)
1899{
1900	dout("process_banner on %p\n", con);
1901
1902	if (verify_hello(con) < 0)
1903		return -1;
1904
1905	ceph_decode_addr(&con->actual_peer_addr);
1906	ceph_decode_addr(&con->peer_addr_for_me);
1907
1908	/*
1909	 * Make sure the other end is who we wanted.  note that the other
1910	 * end may not yet know their ip address, so if it's 0.0.0.0, give
1911	 * them the benefit of the doubt.
1912	 */
1913	if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1914		   sizeof(con->peer_addr)) != 0 &&
1915	    !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1916	      con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1917		pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1918			   ceph_pr_addr(&con->peer_addr.in_addr),
1919			   (int)le32_to_cpu(con->peer_addr.nonce),
1920			   ceph_pr_addr(&con->actual_peer_addr.in_addr),
1921			   (int)le32_to_cpu(con->actual_peer_addr.nonce));
1922		con->error_msg = "wrong peer at address";
1923		return -1;
1924	}
1925
1926	/*
1927	 * did we learn our address?
1928	 */
1929	if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1930		int port = addr_port(&con->msgr->inst.addr.in_addr);
1931
1932		memcpy(&con->msgr->inst.addr.in_addr,
1933		       &con->peer_addr_for_me.in_addr,
1934		       sizeof(con->peer_addr_for_me.in_addr));
1935		addr_set_port(&con->msgr->inst.addr.in_addr, port);
1936		encode_my_addr(con->msgr);
1937		dout("process_banner learned my addr is %s\n",
1938		     ceph_pr_addr(&con->msgr->inst.addr.in_addr));
1939	}
1940
1941	return 0;
1942}
1943
1944static int process_connect(struct ceph_connection *con)
1945{
1946	u64 sup_feat = con->msgr->supported_features;
1947	u64 req_feat = con->msgr->required_features;
1948	u64 server_feat = le64_to_cpu(con->in_reply.features);
1949	int ret;
1950
1951	dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1952
1953	switch (con->in_reply.tag) {
1954	case CEPH_MSGR_TAG_FEATURES:
1955		pr_err("%s%lld %s feature set mismatch,"
1956		       " my %llx < server's %llx, missing %llx\n",
1957		       ENTITY_NAME(con->peer_name),
1958		       ceph_pr_addr(&con->peer_addr.in_addr),
1959		       sup_feat, server_feat, server_feat & ~sup_feat);
1960		con->error_msg = "missing required protocol features";
1961		reset_connection(con);
1962		return -1;
1963
1964	case CEPH_MSGR_TAG_BADPROTOVER:
1965		pr_err("%s%lld %s protocol version mismatch,"
1966		       " my %d != server's %d\n",
1967		       ENTITY_NAME(con->peer_name),
1968		       ceph_pr_addr(&con->peer_addr.in_addr),
1969		       le32_to_cpu(con->out_connect.protocol_version),
1970		       le32_to_cpu(con->in_reply.protocol_version));
1971		con->error_msg = "protocol version mismatch";
1972		reset_connection(con);
1973		return -1;
1974
1975	case CEPH_MSGR_TAG_BADAUTHORIZER:
1976		con->auth_retry++;
1977		dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1978		     con->auth_retry);
1979		if (con->auth_retry == 2) {
1980			con->error_msg = "connect authorization failure";
1981			return -1;
1982		}
1983		con_out_kvec_reset(con);
1984		ret = prepare_write_connect(con);
1985		if (ret < 0)
1986			return ret;
1987		prepare_read_connect(con);
1988		break;
1989
1990	case CEPH_MSGR_TAG_RESETSESSION:
1991		/*
1992		 * If we connected with a large connect_seq but the peer
1993		 * has no record of a session with us (no connection, or
1994		 * connect_seq == 0), they will send RESETSESION to indicate
1995		 * that they must have reset their session, and may have
1996		 * dropped messages.
1997		 */
1998		dout("process_connect got RESET peer seq %u\n",
1999		     le32_to_cpu(con->in_reply.connect_seq));
2000		pr_err("%s%lld %s connection reset\n",
2001		       ENTITY_NAME(con->peer_name),
2002		       ceph_pr_addr(&con->peer_addr.in_addr));
2003		reset_connection(con);
2004		con_out_kvec_reset(con);
2005		ret = prepare_write_connect(con);
2006		if (ret < 0)
2007			return ret;
2008		prepare_read_connect(con);
2009
2010		/* Tell ceph about it. */
2011		mutex_unlock(&con->mutex);
2012		pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2013		if (con->ops->peer_reset)
2014			con->ops->peer_reset(con);
2015		mutex_lock(&con->mutex);
2016		if (con->state != CON_STATE_NEGOTIATING)
2017			return -EAGAIN;
2018		break;
2019
2020	case CEPH_MSGR_TAG_RETRY_SESSION:
2021		/*
2022		 * If we sent a smaller connect_seq than the peer has, try
2023		 * again with a larger value.
2024		 */
2025		dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2026		     le32_to_cpu(con->out_connect.connect_seq),
2027		     le32_to_cpu(con->in_reply.connect_seq));
2028		con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2029		con_out_kvec_reset(con);
2030		ret = prepare_write_connect(con);
2031		if (ret < 0)
2032			return ret;
2033		prepare_read_connect(con);
2034		break;
2035
2036	case CEPH_MSGR_TAG_RETRY_GLOBAL:
2037		/*
2038		 * If we sent a smaller global_seq than the peer has, try
2039		 * again with a larger value.
2040		 */
2041		dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2042		     con->peer_global_seq,
2043		     le32_to_cpu(con->in_reply.global_seq));
2044		get_global_seq(con->msgr,
2045			       le32_to_cpu(con->in_reply.global_seq));
2046		con_out_kvec_reset(con);
2047		ret = prepare_write_connect(con);
2048		if (ret < 0)
2049			return ret;
2050		prepare_read_connect(con);
2051		break;
2052
2053	case CEPH_MSGR_TAG_SEQ:
2054	case CEPH_MSGR_TAG_READY:
2055		if (req_feat & ~server_feat) {
2056			pr_err("%s%lld %s protocol feature mismatch,"
2057			       " my required %llx > server's %llx, need %llx\n",
2058			       ENTITY_NAME(con->peer_name),
2059			       ceph_pr_addr(&con->peer_addr.in_addr),
2060			       req_feat, server_feat, req_feat & ~server_feat);
2061			con->error_msg = "missing required protocol features";
2062			reset_connection(con);
2063			return -1;
2064		}
2065
2066		WARN_ON(con->state != CON_STATE_NEGOTIATING);
2067		con->state = CON_STATE_OPEN;
2068		con->auth_retry = 0;    /* we authenticated; clear flag */
2069		con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2070		con->connect_seq++;
2071		con->peer_features = server_feat;
2072		dout("process_connect got READY gseq %d cseq %d (%d)\n",
2073		     con->peer_global_seq,
2074		     le32_to_cpu(con->in_reply.connect_seq),
2075		     con->connect_seq);
2076		WARN_ON(con->connect_seq !=
2077			le32_to_cpu(con->in_reply.connect_seq));
2078
2079		if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2080			con_flag_set(con, CON_FLAG_LOSSYTX);
2081
2082		con->delay = 0;      /* reset backoff memory */
2083
2084		if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2085			prepare_write_seq(con);
2086			prepare_read_seq(con);
2087		} else {
2088			prepare_read_tag(con);
2089		}
2090		break;
2091
2092	case CEPH_MSGR_TAG_WAIT:
2093		/*
2094		 * If there is a connection race (we are opening
2095		 * connections to each other), one of us may just have
2096		 * to WAIT.  This shouldn't happen if we are the
2097		 * client.
2098		 */
2099		pr_err("process_connect got WAIT as client\n");
2100		con->error_msg = "protocol error, got WAIT as client";
2101		return -1;
2102
2103	default:
2104		pr_err("connect protocol error, will retry\n");
2105		con->error_msg = "protocol error, garbage tag during connect";
2106		return -1;
2107	}
2108	return 0;
2109}
2110
2111
2112/*
2113 * read (part of) an ack
2114 */
2115static int read_partial_ack(struct ceph_connection *con)
2116{
2117	int size = sizeof (con->in_temp_ack);
2118	int end = size;
2119
2120	return read_partial(con, end, size, &con->in_temp_ack);
2121}
2122
2123/*
2124 * We can finally discard anything that's been acked.
2125 */
2126static void process_ack(struct ceph_connection *con)
2127{
2128	struct ceph_msg *m;
2129	u64 ack = le64_to_cpu(con->in_temp_ack);
2130	u64 seq;
2131
2132	while (!list_empty(&con->out_sent)) {
2133		m = list_first_entry(&con->out_sent, struct ceph_msg,
2134				     list_head);
2135		seq = le64_to_cpu(m->hdr.seq);
2136		if (seq > ack)
2137			break;
2138		dout("got ack for seq %llu type %d at %p\n", seq,
2139		     le16_to_cpu(m->hdr.type), m);
2140		m->ack_stamp = jiffies;
2141		ceph_msg_remove(m);
2142	}
2143	prepare_read_tag(con);
2144}
2145
2146
2147static int read_partial_message_section(struct ceph_connection *con,
2148					struct kvec *section,
2149					unsigned int sec_len, u32 *crc)
2150{
2151	int ret, left;
2152
2153	BUG_ON(!section);
2154
2155	while (section->iov_len < sec_len) {
2156		BUG_ON(section->iov_base == NULL);
2157		left = sec_len - section->iov_len;
2158		ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2159				       section->iov_len, left);
2160		if (ret <= 0)
2161			return ret;
2162		section->iov_len += ret;
2163	}
2164	if (section->iov_len == sec_len)
2165		*crc = crc32c(0, section->iov_base, section->iov_len);
2166
2167	return 1;
2168}
2169
2170static int read_partial_msg_data(struct ceph_connection *con)
2171{
2172	struct ceph_msg *msg = con->in_msg;
2173	struct ceph_msg_data_cursor *cursor = &msg->cursor;
2174	const bool do_datacrc = !con->msgr->nocrc;
2175	struct page *page;
2176	size_t page_offset;
2177	size_t length;
2178	u32 crc = 0;
2179	int ret;
2180
2181	BUG_ON(!msg);
2182	if (list_empty(&msg->data))
2183		return -EIO;
2184
2185	if (do_datacrc)
2186		crc = con->in_data_crc;
2187	while (cursor->resid) {
2188		page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
2189							NULL);
2190		ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2191		if (ret <= 0) {
2192			if (do_datacrc)
2193				con->in_data_crc = crc;
2194
2195			return ret;
2196		}
2197
2198		if (do_datacrc)
2199			crc = ceph_crc32c_page(crc, page, page_offset, ret);
2200		(void) ceph_msg_data_advance(&msg->cursor, (size_t)ret);
2201	}
2202	if (do_datacrc)
2203		con->in_data_crc = crc;
2204
2205	return 1;	/* must return > 0 to indicate success */
2206}
2207
2208/*
2209 * read (part of) a message.
2210 */
2211static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2212
2213static int read_partial_message(struct ceph_connection *con)
2214{
2215	struct ceph_msg *m = con->in_msg;
2216	int size;
2217	int end;
2218	int ret;
2219	unsigned int front_len, middle_len, data_len;
2220	bool do_datacrc = !con->msgr->nocrc;
2221	u64 seq;
2222	u32 crc;
2223
2224	dout("read_partial_message con %p msg %p\n", con, m);
2225
2226	/* header */
2227	size = sizeof (con->in_hdr);
2228	end = size;
2229	ret = read_partial(con, end, size, &con->in_hdr);
2230	if (ret <= 0)
2231		return ret;
2232
2233	crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2234	if (cpu_to_le32(crc) != con->in_hdr.crc) {
2235		pr_err("read_partial_message bad hdr "
2236		       " crc %u != expected %u\n",
2237		       crc, con->in_hdr.crc);
2238		return -EBADMSG;
2239	}
2240
2241	front_len = le32_to_cpu(con->in_hdr.front_len);
2242	if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2243		return -EIO;
2244	middle_len = le32_to_cpu(con->in_hdr.middle_len);
2245	if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2246		return -EIO;
2247	data_len = le32_to_cpu(con->in_hdr.data_len);
2248	if (data_len > CEPH_MSG_MAX_DATA_LEN)
2249		return -EIO;
2250
2251	/* verify seq# */
2252	seq = le64_to_cpu(con->in_hdr.seq);
2253	if ((s64)seq - (s64)con->in_seq < 1) {
2254		pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2255			ENTITY_NAME(con->peer_name),
2256			ceph_pr_addr(&con->peer_addr.in_addr),
2257			seq, con->in_seq + 1);
2258		con->in_base_pos = -front_len - middle_len - data_len -
2259			sizeof(m->footer);
2260		con->in_tag = CEPH_MSGR_TAG_READY;
2261		return 0;
2262	} else if ((s64)seq - (s64)con->in_seq > 1) {
2263		pr_err("read_partial_message bad seq %lld expected %lld\n",
2264		       seq, con->in_seq + 1);
2265		con->error_msg = "bad message sequence # for incoming message";
2266		return -EBADMSG;
2267	}
2268
2269	/* allocate message? */
2270	if (!con->in_msg) {
2271		int skip = 0;
2272
2273		dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2274		     front_len, data_len);
2275		ret = ceph_con_in_msg_alloc(con, &skip);
2276		if (ret < 0)
2277			return ret;
2278
2279		BUG_ON(!con->in_msg ^ skip);
2280		if (con->in_msg && data_len > con->in_msg->data_length) {
2281			pr_warning("%s skipping long message (%u > %zd)\n",
2282				__func__, data_len, con->in_msg->data_length);
2283			ceph_msg_put(con->in_msg);
2284			con->in_msg = NULL;
2285			skip = 1;
2286		}
2287		if (skip) {
2288			/* skip this message */
2289			dout("alloc_msg said skip message\n");
2290			con->in_base_pos = -front_len - middle_len - data_len -
2291				sizeof(m->footer);
2292			con->in_tag = CEPH_MSGR_TAG_READY;
2293			con->in_seq++;
2294			return 0;
2295		}
2296
2297		BUG_ON(!con->in_msg);
2298		BUG_ON(con->in_msg->con != con);
2299		m = con->in_msg;
2300		m->front.iov_len = 0;    /* haven't read it yet */
2301		if (m->middle)
2302			m->middle->vec.iov_len = 0;
2303
2304		/* prepare for data payload, if any */
2305
2306		if (data_len)
2307			prepare_message_data(con->in_msg, data_len);
2308	}
2309
2310	/* front */
2311	ret = read_partial_message_section(con, &m->front, front_len,
2312					   &con->in_front_crc);
2313	if (ret <= 0)
2314		return ret;
2315
2316	/* middle */
2317	if (m->middle) {
2318		ret = read_partial_message_section(con, &m->middle->vec,
2319						   middle_len,
2320						   &con->in_middle_crc);
2321		if (ret <= 0)
2322			return ret;
2323	}
2324
2325	/* (page) data */
2326	if (data_len) {
2327		ret = read_partial_msg_data(con);
2328		if (ret <= 0)
2329			return ret;
2330	}
2331
2332	/* footer */
2333	size = sizeof (m->footer);
2334	end += size;
2335	ret = read_partial(con, end, size, &m->footer);
2336	if (ret <= 0)
2337		return ret;
2338
2339	dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2340	     m, front_len, m->footer.front_crc, middle_len,
2341	     m->footer.middle_crc, data_len, m->footer.data_crc);
2342
2343	/* crc ok? */
2344	if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2345		pr_err("read_partial_message %p front crc %u != exp. %u\n",
2346		       m, con->in_front_crc, m->footer.front_crc);
2347		return -EBADMSG;
2348	}
2349	if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2350		pr_err("read_partial_message %p middle crc %u != exp %u\n",
2351		       m, con->in_middle_crc, m->footer.middle_crc);
2352		return -EBADMSG;
2353	}
2354	if (do_datacrc &&
2355	    (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2356	    con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2357		pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2358		       con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2359		return -EBADMSG;
2360	}
2361
2362	return 1; /* done! */
2363}
2364
2365/*
2366 * Process message.  This happens in the worker thread.  The callback should
2367 * be careful not to do anything that waits on other incoming messages or it
2368 * may deadlock.
2369 */
2370static void process_message(struct ceph_connection *con)
2371{
2372	struct ceph_msg *msg;
2373
2374	BUG_ON(con->in_msg->con != con);
2375	con->in_msg->con = NULL;
2376	msg = con->in_msg;
2377	con->in_msg = NULL;
2378	con->ops->put(con);
2379
2380	/* if first message, set peer_name */
2381	if (con->peer_name.type == 0)
2382		con->peer_name = msg->hdr.src;
2383
2384	con->in_seq++;
2385	mutex_unlock(&con->mutex);
2386
2387	dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2388	     msg, le64_to_cpu(msg->hdr.seq),
2389	     ENTITY_NAME(msg->hdr.src),
2390	     le16_to_cpu(msg->hdr.type),
2391	     ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2392	     le32_to_cpu(msg->hdr.front_len),
2393	     le32_to_cpu(msg->hdr.data_len),
2394	     con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2395	con->ops->dispatch(con, msg);
2396
2397	mutex_lock(&con->mutex);
2398}
2399
2400
2401/*
2402 * Write something to the socket.  Called in a worker thread when the
2403 * socket appears to be writeable and we have something ready to send.
2404 */
2405static int try_write(struct ceph_connection *con)
2406{
2407	int ret = 1;
2408
2409	dout("try_write start %p state %lu\n", con, con->state);
2410
2411more:
2412	dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2413
2414	/* open the socket first? */
2415	if (con->state == CON_STATE_PREOPEN) {
2416		BUG_ON(con->sock);
2417		con->state = CON_STATE_CONNECTING;
2418
2419		con_out_kvec_reset(con);
2420		prepare_write_banner(con);
2421		prepare_read_banner(con);
2422
2423		BUG_ON(con->in_msg);
2424		con->in_tag = CEPH_MSGR_TAG_READY;
2425		dout("try_write initiating connect on %p new state %lu\n",
2426		     con, con->state);
2427		ret = ceph_tcp_connect(con);
2428		if (ret < 0) {
2429			con->error_msg = "connect error";
2430			goto out;
2431		}
2432	}
2433
2434more_kvec:
2435	/* kvec data queued? */
2436	if (con->out_skip) {
2437		ret = write_partial_skip(con);
2438		if (ret <= 0)
2439			goto out;
2440	}
2441	if (con->out_kvec_left) {
2442		ret = write_partial_kvec(con);
2443		if (ret <= 0)
2444			goto out;
2445	}
2446
2447	/* msg pages? */
2448	if (con->out_msg) {
2449		if (con->out_msg_done) {
2450			ceph_msg_put(con->out_msg);
2451			con->out_msg = NULL;   /* we're done with this one */
2452			goto do_next;
2453		}
2454
2455		ret = write_partial_message_data(con);
2456		if (ret == 1)
2457			goto more_kvec;  /* we need to send the footer, too! */
2458		if (ret == 0)
2459			goto out;
2460		if (ret < 0) {
2461			dout("try_write write_partial_message_data err %d\n",
2462			     ret);
2463			goto out;
2464		}
2465	}
2466
2467do_next:
2468	if (con->state == CON_STATE_OPEN) {
2469		/* is anything else pending? */
2470		if (!list_empty(&con->out_queue)) {
2471			prepare_write_message(con);
2472			goto more;
2473		}
2474		if (con->in_seq > con->in_seq_acked) {
2475			prepare_write_ack(con);
2476			goto more;
2477		}
2478		if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2479			prepare_write_keepalive(con);
2480			goto more;
2481		}
2482	}
2483
2484	/* Nothing to do! */
2485	con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2486	dout("try_write nothing else to write.\n");
2487	ret = 0;
2488out:
2489	dout("try_write done on %p ret %d\n", con, ret);
2490	return ret;
2491}
2492
2493
2494
2495/*
2496 * Read what we can from the socket.
2497 */
2498static int try_read(struct ceph_connection *con)
2499{
2500	int ret = -1;
2501
2502more:
2503	dout("try_read start on %p state %lu\n", con, con->state);
2504	if (con->state != CON_STATE_CONNECTING &&
2505	    con->state != CON_STATE_NEGOTIATING &&
2506	    con->state != CON_STATE_OPEN)
2507		return 0;
2508
2509	BUG_ON(!con->sock);
2510
2511	dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2512	     con->in_base_pos);
2513
2514	if (con->state == CON_STATE_CONNECTING) {
2515		dout("try_read connecting\n");
2516		ret = read_partial_banner(con);
2517		if (ret <= 0)
2518			goto out;
2519		ret = process_banner(con);
2520		if (ret < 0)
2521			goto out;
2522
2523		con->state = CON_STATE_NEGOTIATING;
2524
2525		/*
2526		 * Received banner is good, exchange connection info.
2527		 * Do not reset out_kvec, as sending our banner raced
2528		 * with receiving peer banner after connect completed.
2529		 */
2530		ret = prepare_write_connect(con);
2531		if (ret < 0)
2532			goto out;
2533		prepare_read_connect(con);
2534
2535		/* Send connection info before awaiting response */
2536		goto out;
2537	}
2538
2539	if (con->state == CON_STATE_NEGOTIATING) {
2540		dout("try_read negotiating\n");
2541		ret = read_partial_connect(con);
2542		if (ret <= 0)
2543			goto out;
2544		ret = process_connect(con);
2545		if (ret < 0)
2546			goto out;
2547		goto more;
2548	}
2549
2550	WARN_ON(con->state != CON_STATE_OPEN);
2551
2552	if (con->in_base_pos < 0) {
2553		/*
2554		 * skipping + discarding content.
2555		 *
2556		 * FIXME: there must be a better way to do this!
2557		 */
2558		static char buf[SKIP_BUF_SIZE];
2559		int skip = min((int) sizeof (buf), -con->in_base_pos);
2560
2561		dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2562		ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2563		if (ret <= 0)
2564			goto out;
2565		con->in_base_pos += ret;
2566		if (con->in_base_pos)
2567			goto more;
2568	}
2569	if (con->in_tag == CEPH_MSGR_TAG_READY) {
2570		/*
2571		 * what's next?
2572		 */
2573		ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2574		if (ret <= 0)
2575			goto out;
2576		dout("try_read got tag %d\n", (int)con->in_tag);
2577		switch (con->in_tag) {
2578		case CEPH_MSGR_TAG_MSG:
2579			prepare_read_message(con);
2580			break;
2581		case CEPH_MSGR_TAG_ACK:
2582			prepare_read_ack(con);
2583			break;
2584		case CEPH_MSGR_TAG_CLOSE:
2585			con_close_socket(con);
2586			con->state = CON_STATE_CLOSED;
2587			goto out;
2588		default:
2589			goto bad_tag;
2590		}
2591	}
2592	if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2593		ret = read_partial_message(con);
2594		if (ret <= 0) {
2595			switch (ret) {
2596			case -EBADMSG:
2597				con->error_msg = "bad crc";
2598				ret = -EIO;
2599				break;
2600			case -EIO:
2601				con->error_msg = "io error";
2602				break;
2603			}
2604			goto out;
2605		}
2606		if (con->in_tag == CEPH_MSGR_TAG_READY)
2607			goto more;
2608		process_message(con);
2609		if (con->state == CON_STATE_OPEN)
2610			prepare_read_tag(con);
2611		goto more;
2612	}
2613	if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2614	    con->in_tag == CEPH_MSGR_TAG_SEQ) {
2615		/*
2616		 * the final handshake seq exchange is semantically
2617		 * equivalent to an ACK
2618		 */
2619		ret = read_partial_ack(con);
2620		if (ret <= 0)
2621			goto out;
2622		process_ack(con);
2623		goto more;
2624	}
2625
2626out:
2627	dout("try_read done on %p ret %d\n", con, ret);
2628	return ret;
2629
2630bad_tag:
2631	pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2632	con->error_msg = "protocol error, garbage tag";
2633	ret = -1;
2634	goto out;
2635}
2636
2637
2638/*
2639 * Atomically queue work on a connection after the specified delay.
2640 * Bump @con reference to avoid races with connection teardown.
2641 * Returns 0 if work was queued, or an error code otherwise.
2642 */
2643static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2644{
2645	if (!con->ops->get(con)) {
2646		dout("%s %p ref count 0\n", __func__, con);
2647
2648		return -ENOENT;
2649	}
2650
2651	if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2652		dout("%s %p - already queued\n", __func__, con);
2653		con->ops->put(con);
2654
2655		return -EBUSY;
2656	}
2657
2658	dout("%s %p %lu\n", __func__, con, delay);
2659
2660	return 0;
2661}
2662
2663static void queue_con(struct ceph_connection *con)
2664{
2665	(void) queue_con_delay(con, 0);
2666}
2667
2668static bool con_sock_closed(struct ceph_connection *con)
2669{
2670	if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2671		return false;
2672
2673#define CASE(x)								\
2674	case CON_STATE_ ## x:						\
2675		con->error_msg = "socket closed (con state " #x ")";	\
2676		break;
2677
2678	switch (con->state) {
2679	CASE(CLOSED);
2680	CASE(PREOPEN);
2681	CASE(CONNECTING);
2682	CASE(NEGOTIATING);
2683	CASE(OPEN);
2684	CASE(STANDBY);
2685	default:
2686		pr_warning("%s con %p unrecognized state %lu\n",
2687			__func__, con, con->state);
2688		con->error_msg = "unrecognized con state";
2689		BUG();
2690		break;
2691	}
2692#undef CASE
2693
2694	return true;
2695}
2696
2697static bool con_backoff(struct ceph_connection *con)
2698{
2699	int ret;
2700
2701	if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2702		return false;
2703
2704	ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2705	if (ret) {
2706		dout("%s: con %p FAILED to back off %lu\n", __func__,
2707			con, con->delay);
2708		BUG_ON(ret == -ENOENT);
2709		con_flag_set(con, CON_FLAG_BACKOFF);
2710	}
2711
2712	return true;
2713}
2714
2715/* Finish fault handling; con->mutex must *not* be held here */
2716
2717static void con_fault_finish(struct ceph_connection *con)
2718{
2719	/*
2720	 * in case we faulted due to authentication, invalidate our
2721	 * current tickets so that we can get new ones.
2722	 */
2723	if (con->auth_retry && con->ops->invalidate_authorizer) {
2724		dout("calling invalidate_authorizer()\n");
2725		con->ops->invalidate_authorizer(con);
2726	}
2727
2728	if (con->ops->fault)
2729		con->ops->fault(con);
2730}
2731
2732/*
2733 * Do some work on a connection.  Drop a connection ref when we're done.
2734 */
2735static void con_work(struct work_struct *work)
2736{
2737	struct ceph_connection *con = container_of(work, struct ceph_connection,
2738						   work.work);
2739	bool fault;
2740
2741	mutex_lock(&con->mutex);
2742	while (true) {
2743		int ret;
2744
2745		if ((fault = con_sock_closed(con))) {
2746			dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2747			break;
2748		}
2749		if (con_backoff(con)) {
2750			dout("%s: con %p BACKOFF\n", __func__, con);
2751			break;
2752		}
2753		if (con->state == CON_STATE_STANDBY) {
2754			dout("%s: con %p STANDBY\n", __func__, con);
2755			break;
2756		}
2757		if (con->state == CON_STATE_CLOSED) {
2758			dout("%s: con %p CLOSED\n", __func__, con);
2759			BUG_ON(con->sock);
2760			break;
2761		}
2762		if (con->state == CON_STATE_PREOPEN) {
2763			dout("%s: con %p PREOPEN\n", __func__, con);
2764			BUG_ON(con->sock);
2765		}
2766
2767		ret = try_read(con);
2768		if (ret < 0) {
2769			if (ret == -EAGAIN)
2770				continue;
2771			con->error_msg = "socket error on read";
2772			fault = true;
2773			break;
2774		}
2775
2776		ret = try_write(con);
2777		if (ret < 0) {
2778			if (ret == -EAGAIN)
2779				continue;
2780			con->error_msg = "socket error on write";
2781			fault = true;
2782		}
2783
2784		break;	/* If we make it to here, we're done */
2785	}
2786	if (fault)
2787		con_fault(con);
2788	mutex_unlock(&con->mutex);
2789
2790	if (fault)
2791		con_fault_finish(con);
2792
2793	con->ops->put(con);
2794}
2795
2796/*
2797 * Generic error/fault handler.  A retry mechanism is used with
2798 * exponential backoff
2799 */
2800static void con_fault(struct ceph_connection *con)
2801{
2802	pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2803	       ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2804	dout("fault %p state %lu to peer %s\n",
2805	     con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2806
2807	WARN_ON(con->state != CON_STATE_CONNECTING &&
2808	       con->state != CON_STATE_NEGOTIATING &&
2809	       con->state != CON_STATE_OPEN);
2810
2811	con_close_socket(con);
2812
2813	if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2814		dout("fault on LOSSYTX channel, marking CLOSED\n");
2815		con->state = CON_STATE_CLOSED;
2816		return;
2817	}
2818
2819	if (con->in_msg) {
2820		BUG_ON(con->in_msg->con != con);
2821		con->in_msg->con = NULL;
2822		ceph_msg_put(con->in_msg);
2823		con->in_msg = NULL;
2824		con->ops->put(con);
2825	}
2826
2827	/* Requeue anything that hasn't been acked */
2828	list_splice_init(&con->out_sent, &con->out_queue);
2829
2830	/* If there are no messages queued or keepalive pending, place
2831	 * the connection in a STANDBY state */
2832	if (list_empty(&con->out_queue) &&
2833	    !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
2834		dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2835		con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2836		con->state = CON_STATE_STANDBY;
2837	} else {
2838		/* retry after a delay. */
2839		con->state = CON_STATE_PREOPEN;
2840		if (con->delay == 0)
2841			con->delay = BASE_DELAY_INTERVAL;
2842		else if (con->delay < MAX_DELAY_INTERVAL)
2843			con->delay *= 2;
2844		con_flag_set(con, CON_FLAG_BACKOFF);
2845		queue_con(con);
2846	}
2847}
2848
2849
2850
2851/*
2852 * initialize a new messenger instance
2853 */
2854void ceph_messenger_init(struct ceph_messenger *msgr,
2855			struct ceph_entity_addr *myaddr,
2856			u32 supported_features,
2857			u32 required_features,
2858			bool nocrc)
2859{
2860	msgr->supported_features = supported_features;
2861	msgr->required_features = required_features;
2862
2863	spin_lock_init(&msgr->global_seq_lock);
2864
2865	if (myaddr)
2866		msgr->inst.addr = *myaddr;
2867
2868	/* select a random nonce */
2869	msgr->inst.addr.type = 0;
2870	get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2871	encode_my_addr(msgr);
2872	msgr->nocrc = nocrc;
2873
2874	atomic_set(&msgr->stopping, 0);
2875
2876	dout("%s %p\n", __func__, msgr);
2877}
2878EXPORT_SYMBOL(ceph_messenger_init);
2879
2880static void clear_standby(struct ceph_connection *con)
2881{
2882	/* come back from STANDBY? */
2883	if (con->state == CON_STATE_STANDBY) {
2884		dout("clear_standby %p and ++connect_seq\n", con);
2885		con->state = CON_STATE_PREOPEN;
2886		con->connect_seq++;
2887		WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2888		WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
2889	}
2890}
2891
2892/*
2893 * Queue up an outgoing message on the given connection.
2894 */
2895void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2896{
2897	/* set src+dst */
2898	msg->hdr.src = con->msgr->inst.name;
2899	BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2900	msg->needs_out_seq = true;
2901
2902	mutex_lock(&con->mutex);
2903
2904	if (con->state == CON_STATE_CLOSED) {
2905		dout("con_send %p closed, dropping %p\n", con, msg);
2906		ceph_msg_put(msg);
2907		mutex_unlock(&con->mutex);
2908		return;
2909	}
2910
2911	BUG_ON(msg->con != NULL);
2912	msg->con = con->ops->get(con);
2913	BUG_ON(msg->con == NULL);
2914
2915	BUG_ON(!list_empty(&msg->list_head));
2916	list_add_tail(&msg->list_head, &con->out_queue);
2917	dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2918	     ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2919	     ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2920	     le32_to_cpu(msg->hdr.front_len),
2921	     le32_to_cpu(msg->hdr.middle_len),
2922	     le32_to_cpu(msg->hdr.data_len));
2923
2924	clear_standby(con);
2925	mutex_unlock(&con->mutex);
2926
2927	/* if there wasn't anything waiting to send before, queue
2928	 * new work */
2929	if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
2930		queue_con(con);
2931}
2932EXPORT_SYMBOL(ceph_con_send);
2933
2934/*
2935 * Revoke a message that was previously queued for send
2936 */
2937void ceph_msg_revoke(struct ceph_msg *msg)
2938{
2939	struct ceph_connection *con = msg->con;
2940
2941	if (!con)
2942		return;		/* Message not in our possession */
2943
2944	mutex_lock(&con->mutex);
2945	if (!list_empty(&msg->list_head)) {
2946		dout("%s %p msg %p - was on queue\n", __func__, con, msg);
2947		list_del_init(&msg->list_head);
2948		BUG_ON(msg->con == NULL);
2949		msg->con->ops->put(msg->con);
2950		msg->con = NULL;
2951		msg->hdr.seq = 0;
2952
2953		ceph_msg_put(msg);
2954	}
2955	if (con->out_msg == msg) {
2956		dout("%s %p msg %p - was sending\n", __func__, con, msg);
2957		con->out_msg = NULL;
2958		if (con->out_kvec_is_msg) {
2959			con->out_skip = con->out_kvec_bytes;
2960			con->out_kvec_is_msg = false;
2961		}
2962		msg->hdr.seq = 0;
2963
2964		ceph_msg_put(msg);
2965	}
2966	mutex_unlock(&con->mutex);
2967}
2968
2969/*
2970 * Revoke a message that we may be reading data into
2971 */
2972void ceph_msg_revoke_incoming(struct ceph_msg *msg)
2973{
2974	struct ceph_connection *con;
2975
2976	BUG_ON(msg == NULL);
2977	if (!msg->con) {
2978		dout("%s msg %p null con\n", __func__, msg);
2979
2980		return;		/* Message not in our possession */
2981	}
2982
2983	con = msg->con;
2984	mutex_lock(&con->mutex);
2985	if (con->in_msg == msg) {
2986		unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
2987		unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
2988		unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
2989
2990		/* skip rest of message */
2991		dout("%s %p msg %p revoked\n", __func__, con, msg);
2992		con->in_base_pos = con->in_base_pos -
2993				sizeof(struct ceph_msg_header) -
2994				front_len -
2995				middle_len -
2996				data_len -
2997				sizeof(struct ceph_msg_footer);
2998		ceph_msg_put(con->in_msg);
2999		con->in_msg = NULL;
3000		con->in_tag = CEPH_MSGR_TAG_READY;
3001		con->in_seq++;
3002	} else {
3003		dout("%s %p in_msg %p msg %p no-op\n",
3004		     __func__, con, con->in_msg, msg);
3005	}
3006	mutex_unlock(&con->mutex);
3007}
3008
3009/*
3010 * Queue a keepalive byte to ensure the tcp connection is alive.
3011 */
3012void ceph_con_keepalive(struct ceph_connection *con)
3013{
3014	dout("con_keepalive %p\n", con);
3015	mutex_lock(&con->mutex);
3016	clear_standby(con);
3017	mutex_unlock(&con->mutex);
3018	if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3019	    con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3020		queue_con(con);
3021}
3022EXPORT_SYMBOL(ceph_con_keepalive);
3023
3024static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
3025{
3026	struct ceph_msg_data *data;
3027
3028	if (WARN_ON(!ceph_msg_data_type_valid(type)))
3029		return NULL;
3030
3031	data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
3032	if (data)
3033		data->type = type;
3034	INIT_LIST_HEAD(&data->links);
3035
3036	return data;
3037}
3038
3039static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3040{
3041	if (!data)
3042		return;
3043
3044	WARN_ON(!list_empty(&data->links));
3045	if (data->type == CEPH_MSG_DATA_PAGELIST) {
3046		ceph_pagelist_release(data->pagelist);
3047		kfree(data->pagelist);
3048	}
3049	kmem_cache_free(ceph_msg_data_cache, data);
3050}
3051
3052void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
3053		size_t length, size_t alignment)
3054{
3055	struct ceph_msg_data *data;
3056
3057	BUG_ON(!pages);
3058	BUG_ON(!length);
3059
3060	data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3061	BUG_ON(!data);
3062	data->pages = pages;
3063	data->length = length;
3064	data->alignment = alignment & ~PAGE_MASK;
3065
3066	list_add_tail(&data->links, &msg->data);
3067	msg->data_length += length;
3068}
3069EXPORT_SYMBOL(ceph_msg_data_add_pages);
3070
3071void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
3072				struct ceph_pagelist *pagelist)
3073{
3074	struct ceph_msg_data *data;
3075
3076	BUG_ON(!pagelist);
3077	BUG_ON(!pagelist->length);
3078
3079	data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3080	BUG_ON(!data);
3081	data->pagelist = pagelist;
3082
3083	list_add_tail(&data->links, &msg->data);
3084	msg->data_length += pagelist->length;
3085}
3086EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
3087
3088#ifdef	CONFIG_BLOCK
3089void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
3090		size_t length)
3091{
3092	struct ceph_msg_data *data;
3093
3094	BUG_ON(!bio);
3095
3096	data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3097	BUG_ON(!data);
3098	data->bio = bio;
3099	data->bio_length = length;
3100
3101	list_add_tail(&data->links, &msg->data);
3102	msg->data_length += length;
3103}
3104EXPORT_SYMBOL(ceph_msg_data_add_bio);
3105#endif	/* CONFIG_BLOCK */
3106
3107/*
3108 * construct a new message with given type, size
3109 * the new msg has a ref count of 1.
3110 */
3111struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3112			      bool can_fail)
3113{
3114	struct ceph_msg *m;
3115
3116	m = kmem_cache_zalloc(ceph_msg_cache, flags);
3117	if (m == NULL)
3118		goto out;
3119
3120	m->hdr.type = cpu_to_le16(type);
3121	m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3122	m->hdr.front_len = cpu_to_le32(front_len);
3123
3124	INIT_LIST_HEAD(&m->list_head);
3125	kref_init(&m->kref);
3126	INIT_LIST_HEAD(&m->data);
3127
3128	/* front */
3129	m->front_max = front_len;
3130	if (front_len) {
3131		if (front_len > PAGE_CACHE_SIZE) {
3132			m->front.iov_base = __vmalloc(front_len, flags,
3133						      PAGE_KERNEL);
3134			m->front_is_vmalloc = true;
3135		} else {
3136			m->front.iov_base = kmalloc(front_len, flags);
3137		}
3138		if (m->front.iov_base == NULL) {
3139			dout("ceph_msg_new can't allocate %d bytes\n",
3140			     front_len);
3141			goto out2;
3142		}
3143	} else {
3144		m->front.iov_base = NULL;
3145	}
3146	m->front.iov_len = front_len;
3147
3148	dout("ceph_msg_new %p front %d\n", m, front_len);
3149	return m;
3150
3151out2:
3152	ceph_msg_put(m);
3153out:
3154	if (!can_fail) {
3155		pr_err("msg_new can't create type %d front %d\n", type,
3156		       front_len);
3157		WARN_ON(1);
3158	} else {
3159		dout("msg_new can't create type %d front %d\n", type,
3160		     front_len);
3161	}
3162	return NULL;
3163}
3164EXPORT_SYMBOL(ceph_msg_new);
3165
3166/*
3167 * Allocate "middle" portion of a message, if it is needed and wasn't
3168 * allocated by alloc_msg.  This allows us to read a small fixed-size
3169 * per-type header in the front and then gracefully fail (i.e.,
3170 * propagate the error to the caller based on info in the front) when
3171 * the middle is too large.
3172 */
3173static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3174{
3175	int type = le16_to_cpu(msg->hdr.type);
3176	int middle_len = le32_to_cpu(msg->hdr.middle_len);
3177
3178	dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3179	     ceph_msg_type_name(type), middle_len);
3180	BUG_ON(!middle_len);
3181	BUG_ON(msg->middle);
3182
3183	msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3184	if (!msg->middle)
3185		return -ENOMEM;
3186	return 0;
3187}
3188
3189/*
3190 * Allocate a message for receiving an incoming message on a
3191 * connection, and save the result in con->in_msg.  Uses the
3192 * connection's private alloc_msg op if available.
3193 *
3194 * Returns 0 on success, or a negative error code.
3195 *
3196 * On success, if we set *skip = 1:
3197 *  - the next message should be skipped and ignored.
3198 *  - con->in_msg == NULL
3199 * or if we set *skip = 0:
3200 *  - con->in_msg is non-null.
3201 * On error (ENOMEM, EAGAIN, ...),
3202 *  - con->in_msg == NULL
3203 */
3204static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3205{
3206	struct ceph_msg_header *hdr = &con->in_hdr;
3207	int middle_len = le32_to_cpu(hdr->middle_len);
3208	struct ceph_msg *msg;
3209	int ret = 0;
3210
3211	BUG_ON(con->in_msg != NULL);
3212	BUG_ON(!con->ops->alloc_msg);
3213
3214	mutex_unlock(&con->mutex);
3215	msg = con->ops->alloc_msg(con, hdr, skip);
3216	mutex_lock(&con->mutex);
3217	if (con->state != CON_STATE_OPEN) {
3218		if (msg)
3219			ceph_msg_put(msg);
3220		return -EAGAIN;
3221	}
3222	if (msg) {
3223		BUG_ON(*skip);
3224		con->in_msg = msg;
3225		con->in_msg->con = con->ops->get(con);
3226		BUG_ON(con->in_msg->con == NULL);
3227	} else {
3228		/*
3229		 * Null message pointer means either we should skip
3230		 * this message or we couldn't allocate memory.  The
3231		 * former is not an error.
3232		 */
3233		if (*skip)
3234			return 0;
3235		con->error_msg = "error allocating memory for incoming message";
3236
3237		return -ENOMEM;
3238	}
3239	memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3240
3241	if (middle_len && !con->in_msg->middle) {
3242		ret = ceph_alloc_middle(con, con->in_msg);
3243		if (ret < 0) {
3244			ceph_msg_put(con->in_msg);
3245			con->in_msg = NULL;
3246		}
3247	}
3248
3249	return ret;
3250}
3251
3252
3253/*
3254 * Free a generically kmalloc'd message.
3255 */
3256void ceph_msg_kfree(struct ceph_msg *m)
3257{
3258	dout("msg_kfree %p\n", m);
3259	if (m->front_is_vmalloc)
3260		vfree(m->front.iov_base);
3261	else
3262		kfree(m->front.iov_base);
3263	kmem_cache_free(ceph_msg_cache, m);
3264}
3265
3266/*
3267 * Drop a msg ref.  Destroy as needed.
3268 */
3269void ceph_msg_last_put(struct kref *kref)
3270{
3271	struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3272	LIST_HEAD(data);
3273	struct list_head *links;
3274	struct list_head *next;
3275
3276	dout("ceph_msg_put last one on %p\n", m);
3277	WARN_ON(!list_empty(&m->list_head));
3278
3279	/* drop middle, data, if any */
3280	if (m->middle) {
3281		ceph_buffer_put(m->middle);
3282		m->middle = NULL;
3283	}
3284
3285	list_splice_init(&m->data, &data);
3286	list_for_each_safe(links, next, &data) {
3287		struct ceph_msg_data *data;
3288
3289		data = list_entry(links, struct ceph_msg_data, links);
3290		list_del_init(links);
3291		ceph_msg_data_destroy(data);
3292	}
3293	m->data_length = 0;
3294
3295	if (m->pool)
3296		ceph_msgpool_put(m->pool, m);
3297	else
3298		ceph_msg_kfree(m);
3299}
3300EXPORT_SYMBOL(ceph_msg_last_put);
3301
3302void ceph_msg_dump(struct ceph_msg *msg)
3303{
3304	pr_debug("msg_dump %p (front_max %d length %zd)\n", msg,
3305		 msg->front_max, msg->data_length);
3306	print_hex_dump(KERN_DEBUG, "header: ",
3307		       DUMP_PREFIX_OFFSET, 16, 1,
3308		       &msg->hdr, sizeof(msg->hdr), true);
3309	print_hex_dump(KERN_DEBUG, " front: ",
3310		       DUMP_PREFIX_OFFSET, 16, 1,
3311		       msg->front.iov_base, msg->front.iov_len, true);
3312	if (msg->middle)
3313		print_hex_dump(KERN_DEBUG, "middle: ",
3314			       DUMP_PREFIX_OFFSET, 16, 1,
3315			       msg->middle->vec.iov_base,
3316			       msg->middle->vec.iov_len, true);
3317	print_hex_dump(KERN_DEBUG, "footer: ",
3318		       DUMP_PREFIX_OFFSET, 16, 1,
3319		       &msg->footer, sizeof(msg->footer), true);
3320}
3321EXPORT_SYMBOL(ceph_msg_dump);
3322