Loading...
1#include <linux/ceph/ceph_debug.h>
2
3#include <linux/crc32c.h>
4#include <linux/ctype.h>
5#include <linux/highmem.h>
6#include <linux/inet.h>
7#include <linux/kthread.h>
8#include <linux/net.h>
9#include <linux/slab.h>
10#include <linux/socket.h>
11#include <linux/string.h>
12#ifdef CONFIG_BLOCK
13#include <linux/bio.h>
14#endif /* CONFIG_BLOCK */
15#include <linux/dns_resolver.h>
16#include <net/tcp.h>
17
18#include <linux/ceph/ceph_features.h>
19#include <linux/ceph/libceph.h>
20#include <linux/ceph/messenger.h>
21#include <linux/ceph/decode.h>
22#include <linux/ceph/pagelist.h>
23#include <linux/export.h>
24
25#define list_entry_next(pos, member) \
26 list_entry(pos->member.next, typeof(*pos), member)
27
28/*
29 * Ceph uses the messenger to exchange ceph_msg messages with other
30 * hosts in the system. The messenger provides ordered and reliable
31 * delivery. We tolerate TCP disconnects by reconnecting (with
32 * exponential backoff) in the case of a fault (disconnection, bad
33 * crc, protocol error). Acks allow sent messages to be discarded by
34 * the sender.
35 */
36
37/*
38 * We track the state of the socket on a given connection using
39 * values defined below. The transition to a new socket state is
40 * handled by a function which verifies we aren't coming from an
41 * unexpected state.
42 *
43 * --------
44 * | NEW* | transient initial state
45 * --------
46 * | con_sock_state_init()
47 * v
48 * ----------
49 * | CLOSED | initialized, but no socket (and no
50 * ---------- TCP connection)
51 * ^ \
52 * | \ con_sock_state_connecting()
53 * | ----------------------
54 * | \
55 * + con_sock_state_closed() \
56 * |+--------------------------- \
57 * | \ \ \
58 * | ----------- \ \
59 * | | CLOSING | socket event; \ \
60 * | ----------- await close \ \
61 * | ^ \ |
62 * | | \ |
63 * | + con_sock_state_closing() \ |
64 * | / \ | |
65 * | / --------------- | |
66 * | / \ v v
67 * | / --------------
68 * | / -----------------| CONNECTING | socket created, TCP
69 * | | / -------------- connect initiated
70 * | | | con_sock_state_connected()
71 * | | v
72 * -------------
73 * | CONNECTED | TCP connection established
74 * -------------
75 *
76 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
77 */
78
79#define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
80#define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
81#define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
82#define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
83#define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
84
85/*
86 * connection states
87 */
88#define CON_STATE_CLOSED 1 /* -> PREOPEN */
89#define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
90#define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
91#define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
92#define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
93#define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
94
95/*
96 * ceph_connection flag bits
97 */
98#define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
99 * messages on errors */
100#define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
101#define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
102#define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
103#define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
104
105static bool con_flag_valid(unsigned long con_flag)
106{
107 switch (con_flag) {
108 case CON_FLAG_LOSSYTX:
109 case CON_FLAG_KEEPALIVE_PENDING:
110 case CON_FLAG_WRITE_PENDING:
111 case CON_FLAG_SOCK_CLOSED:
112 case CON_FLAG_BACKOFF:
113 return true;
114 default:
115 return false;
116 }
117}
118
119static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
120{
121 BUG_ON(!con_flag_valid(con_flag));
122
123 clear_bit(con_flag, &con->flags);
124}
125
126static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
127{
128 BUG_ON(!con_flag_valid(con_flag));
129
130 set_bit(con_flag, &con->flags);
131}
132
133static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
134{
135 BUG_ON(!con_flag_valid(con_flag));
136
137 return test_bit(con_flag, &con->flags);
138}
139
140static bool con_flag_test_and_clear(struct ceph_connection *con,
141 unsigned long con_flag)
142{
143 BUG_ON(!con_flag_valid(con_flag));
144
145 return test_and_clear_bit(con_flag, &con->flags);
146}
147
148static bool con_flag_test_and_set(struct ceph_connection *con,
149 unsigned long con_flag)
150{
151 BUG_ON(!con_flag_valid(con_flag));
152
153 return test_and_set_bit(con_flag, &con->flags);
154}
155
156/* Slab caches for frequently-allocated structures */
157
158static struct kmem_cache *ceph_msg_cache;
159static struct kmem_cache *ceph_msg_data_cache;
160
161/* static tag bytes (protocol control messages) */
162static char tag_msg = CEPH_MSGR_TAG_MSG;
163static char tag_ack = CEPH_MSGR_TAG_ACK;
164static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
165
166#ifdef CONFIG_LOCKDEP
167static struct lock_class_key socket_class;
168#endif
169
170/*
171 * When skipping (ignoring) a block of input we read it into a "skip
172 * buffer," which is this many bytes in size.
173 */
174#define SKIP_BUF_SIZE 1024
175
176static void queue_con(struct ceph_connection *con);
177static void con_work(struct work_struct *);
178static void con_fault(struct ceph_connection *con);
179
180/*
181 * Nicely render a sockaddr as a string. An array of formatted
182 * strings is used, to approximate reentrancy.
183 */
184#define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
185#define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
186#define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
187#define MAX_ADDR_STR_LEN 64 /* 54 is enough */
188
189static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
190static atomic_t addr_str_seq = ATOMIC_INIT(0);
191
192static struct page *zero_page; /* used in certain error cases */
193
194const char *ceph_pr_addr(const struct sockaddr_storage *ss)
195{
196 int i;
197 char *s;
198 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
199 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
200
201 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
202 s = addr_str[i];
203
204 switch (ss->ss_family) {
205 case AF_INET:
206 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
207 ntohs(in4->sin_port));
208 break;
209
210 case AF_INET6:
211 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
212 ntohs(in6->sin6_port));
213 break;
214
215 default:
216 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
217 ss->ss_family);
218 }
219
220 return s;
221}
222EXPORT_SYMBOL(ceph_pr_addr);
223
224static void encode_my_addr(struct ceph_messenger *msgr)
225{
226 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
227 ceph_encode_addr(&msgr->my_enc_addr);
228}
229
230/*
231 * work queue for all reading and writing to/from the socket.
232 */
233static struct workqueue_struct *ceph_msgr_wq;
234
235static int ceph_msgr_slab_init(void)
236{
237 BUG_ON(ceph_msg_cache);
238 ceph_msg_cache = kmem_cache_create("ceph_msg",
239 sizeof (struct ceph_msg),
240 __alignof__(struct ceph_msg), 0, NULL);
241
242 if (!ceph_msg_cache)
243 return -ENOMEM;
244
245 BUG_ON(ceph_msg_data_cache);
246 ceph_msg_data_cache = kmem_cache_create("ceph_msg_data",
247 sizeof (struct ceph_msg_data),
248 __alignof__(struct ceph_msg_data),
249 0, NULL);
250 if (ceph_msg_data_cache)
251 return 0;
252
253 kmem_cache_destroy(ceph_msg_cache);
254 ceph_msg_cache = NULL;
255
256 return -ENOMEM;
257}
258
259static void ceph_msgr_slab_exit(void)
260{
261 BUG_ON(!ceph_msg_data_cache);
262 kmem_cache_destroy(ceph_msg_data_cache);
263 ceph_msg_data_cache = NULL;
264
265 BUG_ON(!ceph_msg_cache);
266 kmem_cache_destroy(ceph_msg_cache);
267 ceph_msg_cache = NULL;
268}
269
270static void _ceph_msgr_exit(void)
271{
272 if (ceph_msgr_wq) {
273 destroy_workqueue(ceph_msgr_wq);
274 ceph_msgr_wq = NULL;
275 }
276
277 ceph_msgr_slab_exit();
278
279 BUG_ON(zero_page == NULL);
280 kunmap(zero_page);
281 page_cache_release(zero_page);
282 zero_page = NULL;
283}
284
285int ceph_msgr_init(void)
286{
287 BUG_ON(zero_page != NULL);
288 zero_page = ZERO_PAGE(0);
289 page_cache_get(zero_page);
290
291 if (ceph_msgr_slab_init())
292 return -ENOMEM;
293
294 ceph_msgr_wq = alloc_workqueue("ceph-msgr", 0, 0);
295 if (ceph_msgr_wq)
296 return 0;
297
298 pr_err("msgr_init failed to create workqueue\n");
299 _ceph_msgr_exit();
300
301 return -ENOMEM;
302}
303EXPORT_SYMBOL(ceph_msgr_init);
304
305void ceph_msgr_exit(void)
306{
307 BUG_ON(ceph_msgr_wq == NULL);
308
309 _ceph_msgr_exit();
310}
311EXPORT_SYMBOL(ceph_msgr_exit);
312
313void ceph_msgr_flush(void)
314{
315 flush_workqueue(ceph_msgr_wq);
316}
317EXPORT_SYMBOL(ceph_msgr_flush);
318
319/* Connection socket state transition functions */
320
321static void con_sock_state_init(struct ceph_connection *con)
322{
323 int old_state;
324
325 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
326 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
327 printk("%s: unexpected old state %d\n", __func__, old_state);
328 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
329 CON_SOCK_STATE_CLOSED);
330}
331
332static void con_sock_state_connecting(struct ceph_connection *con)
333{
334 int old_state;
335
336 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
337 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
338 printk("%s: unexpected old state %d\n", __func__, old_state);
339 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
340 CON_SOCK_STATE_CONNECTING);
341}
342
343static void con_sock_state_connected(struct ceph_connection *con)
344{
345 int old_state;
346
347 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
348 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
349 printk("%s: unexpected old state %d\n", __func__, old_state);
350 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
351 CON_SOCK_STATE_CONNECTED);
352}
353
354static void con_sock_state_closing(struct ceph_connection *con)
355{
356 int old_state;
357
358 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
359 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
360 old_state != CON_SOCK_STATE_CONNECTED &&
361 old_state != CON_SOCK_STATE_CLOSING))
362 printk("%s: unexpected old state %d\n", __func__, old_state);
363 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
364 CON_SOCK_STATE_CLOSING);
365}
366
367static void con_sock_state_closed(struct ceph_connection *con)
368{
369 int old_state;
370
371 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
372 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
373 old_state != CON_SOCK_STATE_CLOSING &&
374 old_state != CON_SOCK_STATE_CONNECTING &&
375 old_state != CON_SOCK_STATE_CLOSED))
376 printk("%s: unexpected old state %d\n", __func__, old_state);
377 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
378 CON_SOCK_STATE_CLOSED);
379}
380
381/*
382 * socket callback functions
383 */
384
385/* data available on socket, or listen socket received a connect */
386static void ceph_sock_data_ready(struct sock *sk)
387{
388 struct ceph_connection *con = sk->sk_user_data;
389 if (atomic_read(&con->msgr->stopping)) {
390 return;
391 }
392
393 if (sk->sk_state != TCP_CLOSE_WAIT) {
394 dout("%s on %p state = %lu, queueing work\n", __func__,
395 con, con->state);
396 queue_con(con);
397 }
398}
399
400/* socket has buffer space for writing */
401static void ceph_sock_write_space(struct sock *sk)
402{
403 struct ceph_connection *con = sk->sk_user_data;
404
405 /* only queue to workqueue if there is data we want to write,
406 * and there is sufficient space in the socket buffer to accept
407 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
408 * doesn't get called again until try_write() fills the socket
409 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
410 * and net/core/stream.c:sk_stream_write_space().
411 */
412 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
413 if (sk_stream_is_writeable(sk)) {
414 dout("%s %p queueing write work\n", __func__, con);
415 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
416 queue_con(con);
417 }
418 } else {
419 dout("%s %p nothing to write\n", __func__, con);
420 }
421}
422
423/* socket's state has changed */
424static void ceph_sock_state_change(struct sock *sk)
425{
426 struct ceph_connection *con = sk->sk_user_data;
427
428 dout("%s %p state = %lu sk_state = %u\n", __func__,
429 con, con->state, sk->sk_state);
430
431 switch (sk->sk_state) {
432 case TCP_CLOSE:
433 dout("%s TCP_CLOSE\n", __func__);
434 case TCP_CLOSE_WAIT:
435 dout("%s TCP_CLOSE_WAIT\n", __func__);
436 con_sock_state_closing(con);
437 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
438 queue_con(con);
439 break;
440 case TCP_ESTABLISHED:
441 dout("%s TCP_ESTABLISHED\n", __func__);
442 con_sock_state_connected(con);
443 queue_con(con);
444 break;
445 default: /* Everything else is uninteresting */
446 break;
447 }
448}
449
450/*
451 * set up socket callbacks
452 */
453static void set_sock_callbacks(struct socket *sock,
454 struct ceph_connection *con)
455{
456 struct sock *sk = sock->sk;
457 sk->sk_user_data = con;
458 sk->sk_data_ready = ceph_sock_data_ready;
459 sk->sk_write_space = ceph_sock_write_space;
460 sk->sk_state_change = ceph_sock_state_change;
461}
462
463
464/*
465 * socket helpers
466 */
467
468/*
469 * initiate connection to a remote socket.
470 */
471static int ceph_tcp_connect(struct ceph_connection *con)
472{
473 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
474 struct socket *sock;
475 int ret;
476
477 BUG_ON(con->sock);
478 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
479 IPPROTO_TCP, &sock);
480 if (ret)
481 return ret;
482 sock->sk->sk_allocation = GFP_NOFS;
483
484#ifdef CONFIG_LOCKDEP
485 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
486#endif
487
488 set_sock_callbacks(sock, con);
489
490 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
491
492 con_sock_state_connecting(con);
493 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
494 O_NONBLOCK);
495 if (ret == -EINPROGRESS) {
496 dout("connect %s EINPROGRESS sk_state = %u\n",
497 ceph_pr_addr(&con->peer_addr.in_addr),
498 sock->sk->sk_state);
499 } else if (ret < 0) {
500 pr_err("connect %s error %d\n",
501 ceph_pr_addr(&con->peer_addr.in_addr), ret);
502 sock_release(sock);
503 con->error_msg = "connect error";
504
505 return ret;
506 }
507 con->sock = sock;
508 return 0;
509}
510
511static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
512{
513 struct kvec iov = {buf, len};
514 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
515 int r;
516
517 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
518 if (r == -EAGAIN)
519 r = 0;
520 return r;
521}
522
523static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
524 int page_offset, size_t length)
525{
526 void *kaddr;
527 int ret;
528
529 BUG_ON(page_offset + length > PAGE_SIZE);
530
531 kaddr = kmap(page);
532 BUG_ON(!kaddr);
533 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
534 kunmap(page);
535
536 return ret;
537}
538
539/*
540 * write something. @more is true if caller will be sending more data
541 * shortly.
542 */
543static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
544 size_t kvlen, size_t len, int more)
545{
546 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
547 int r;
548
549 if (more)
550 msg.msg_flags |= MSG_MORE;
551 else
552 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
553
554 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
555 if (r == -EAGAIN)
556 r = 0;
557 return r;
558}
559
560static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
561 int offset, size_t size, bool more)
562{
563 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
564 int ret;
565
566 ret = kernel_sendpage(sock, page, offset, size, flags);
567 if (ret == -EAGAIN)
568 ret = 0;
569
570 return ret;
571}
572
573static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
574 int offset, size_t size, bool more)
575{
576 int ret;
577 struct kvec iov;
578
579 /* sendpage cannot properly handle pages with page_count == 0,
580 * we need to fallback to sendmsg if that's the case */
581 if (page_count(page) >= 1)
582 return __ceph_tcp_sendpage(sock, page, offset, size, more);
583
584 iov.iov_base = kmap(page) + offset;
585 iov.iov_len = size;
586 ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more);
587 kunmap(page);
588
589 return ret;
590}
591
592/*
593 * Shutdown/close the socket for the given connection.
594 */
595static int con_close_socket(struct ceph_connection *con)
596{
597 int rc = 0;
598
599 dout("con_close_socket on %p sock %p\n", con, con->sock);
600 if (con->sock) {
601 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
602 sock_release(con->sock);
603 con->sock = NULL;
604 }
605
606 /*
607 * Forcibly clear the SOCK_CLOSED flag. It gets set
608 * independent of the connection mutex, and we could have
609 * received a socket close event before we had the chance to
610 * shut the socket down.
611 */
612 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
613
614 con_sock_state_closed(con);
615 return rc;
616}
617
618/*
619 * Reset a connection. Discard all incoming and outgoing messages
620 * and clear *_seq state.
621 */
622static void ceph_msg_remove(struct ceph_msg *msg)
623{
624 list_del_init(&msg->list_head);
625 BUG_ON(msg->con == NULL);
626 msg->con->ops->put(msg->con);
627 msg->con = NULL;
628
629 ceph_msg_put(msg);
630}
631static void ceph_msg_remove_list(struct list_head *head)
632{
633 while (!list_empty(head)) {
634 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
635 list_head);
636 ceph_msg_remove(msg);
637 }
638}
639
640static void reset_connection(struct ceph_connection *con)
641{
642 /* reset connection, out_queue, msg_ and connect_seq */
643 /* discard existing out_queue and msg_seq */
644 dout("reset_connection %p\n", con);
645 ceph_msg_remove_list(&con->out_queue);
646 ceph_msg_remove_list(&con->out_sent);
647
648 if (con->in_msg) {
649 BUG_ON(con->in_msg->con != con);
650 con->in_msg->con = NULL;
651 ceph_msg_put(con->in_msg);
652 con->in_msg = NULL;
653 con->ops->put(con);
654 }
655
656 con->connect_seq = 0;
657 con->out_seq = 0;
658 if (con->out_msg) {
659 ceph_msg_put(con->out_msg);
660 con->out_msg = NULL;
661 }
662 con->in_seq = 0;
663 con->in_seq_acked = 0;
664}
665
666/*
667 * mark a peer down. drop any open connections.
668 */
669void ceph_con_close(struct ceph_connection *con)
670{
671 mutex_lock(&con->mutex);
672 dout("con_close %p peer %s\n", con,
673 ceph_pr_addr(&con->peer_addr.in_addr));
674 con->state = CON_STATE_CLOSED;
675
676 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
677 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
678 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
679 con_flag_clear(con, CON_FLAG_BACKOFF);
680
681 reset_connection(con);
682 con->peer_global_seq = 0;
683 cancel_delayed_work(&con->work);
684 con_close_socket(con);
685 mutex_unlock(&con->mutex);
686}
687EXPORT_SYMBOL(ceph_con_close);
688
689/*
690 * Reopen a closed connection, with a new peer address.
691 */
692void ceph_con_open(struct ceph_connection *con,
693 __u8 entity_type, __u64 entity_num,
694 struct ceph_entity_addr *addr)
695{
696 mutex_lock(&con->mutex);
697 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
698
699 WARN_ON(con->state != CON_STATE_CLOSED);
700 con->state = CON_STATE_PREOPEN;
701
702 con->peer_name.type = (__u8) entity_type;
703 con->peer_name.num = cpu_to_le64(entity_num);
704
705 memcpy(&con->peer_addr, addr, sizeof(*addr));
706 con->delay = 0; /* reset backoff memory */
707 mutex_unlock(&con->mutex);
708 queue_con(con);
709}
710EXPORT_SYMBOL(ceph_con_open);
711
712/*
713 * return true if this connection ever successfully opened
714 */
715bool ceph_con_opened(struct ceph_connection *con)
716{
717 return con->connect_seq > 0;
718}
719
720/*
721 * initialize a new connection.
722 */
723void ceph_con_init(struct ceph_connection *con, void *private,
724 const struct ceph_connection_operations *ops,
725 struct ceph_messenger *msgr)
726{
727 dout("con_init %p\n", con);
728 memset(con, 0, sizeof(*con));
729 con->private = private;
730 con->ops = ops;
731 con->msgr = msgr;
732
733 con_sock_state_init(con);
734
735 mutex_init(&con->mutex);
736 INIT_LIST_HEAD(&con->out_queue);
737 INIT_LIST_HEAD(&con->out_sent);
738 INIT_DELAYED_WORK(&con->work, con_work);
739
740 con->state = CON_STATE_CLOSED;
741}
742EXPORT_SYMBOL(ceph_con_init);
743
744
745/*
746 * We maintain a global counter to order connection attempts. Get
747 * a unique seq greater than @gt.
748 */
749static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
750{
751 u32 ret;
752
753 spin_lock(&msgr->global_seq_lock);
754 if (msgr->global_seq < gt)
755 msgr->global_seq = gt;
756 ret = ++msgr->global_seq;
757 spin_unlock(&msgr->global_seq_lock);
758 return ret;
759}
760
761static void con_out_kvec_reset(struct ceph_connection *con)
762{
763 con->out_kvec_left = 0;
764 con->out_kvec_bytes = 0;
765 con->out_kvec_cur = &con->out_kvec[0];
766}
767
768static void con_out_kvec_add(struct ceph_connection *con,
769 size_t size, void *data)
770{
771 int index;
772
773 index = con->out_kvec_left;
774 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
775
776 con->out_kvec[index].iov_len = size;
777 con->out_kvec[index].iov_base = data;
778 con->out_kvec_left++;
779 con->out_kvec_bytes += size;
780}
781
782#ifdef CONFIG_BLOCK
783
784/*
785 * For a bio data item, a piece is whatever remains of the next
786 * entry in the current bio iovec, or the first entry in the next
787 * bio in the list.
788 */
789static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
790 size_t length)
791{
792 struct ceph_msg_data *data = cursor->data;
793 struct bio *bio;
794
795 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
796
797 bio = data->bio;
798 BUG_ON(!bio);
799
800 cursor->resid = min(length, data->bio_length);
801 cursor->bio = bio;
802 cursor->bvec_iter = bio->bi_iter;
803 cursor->last_piece =
804 cursor->resid <= bio_iter_len(bio, cursor->bvec_iter);
805}
806
807static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
808 size_t *page_offset,
809 size_t *length)
810{
811 struct ceph_msg_data *data = cursor->data;
812 struct bio *bio;
813 struct bio_vec bio_vec;
814
815 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
816
817 bio = cursor->bio;
818 BUG_ON(!bio);
819
820 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
821
822 *page_offset = (size_t) bio_vec.bv_offset;
823 BUG_ON(*page_offset >= PAGE_SIZE);
824 if (cursor->last_piece) /* pagelist offset is always 0 */
825 *length = cursor->resid;
826 else
827 *length = (size_t) bio_vec.bv_len;
828 BUG_ON(*length > cursor->resid);
829 BUG_ON(*page_offset + *length > PAGE_SIZE);
830
831 return bio_vec.bv_page;
832}
833
834static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
835 size_t bytes)
836{
837 struct bio *bio;
838 struct bio_vec bio_vec;
839
840 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
841
842 bio = cursor->bio;
843 BUG_ON(!bio);
844
845 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
846
847 /* Advance the cursor offset */
848
849 BUG_ON(cursor->resid < bytes);
850 cursor->resid -= bytes;
851
852 bio_advance_iter(bio, &cursor->bvec_iter, bytes);
853
854 if (bytes < bio_vec.bv_len)
855 return false; /* more bytes to process in this segment */
856
857 /* Move on to the next segment, and possibly the next bio */
858
859 if (!cursor->bvec_iter.bi_size) {
860 bio = bio->bi_next;
861 cursor->bio = bio;
862 if (bio)
863 cursor->bvec_iter = bio->bi_iter;
864 else
865 memset(&cursor->bvec_iter, 0,
866 sizeof(cursor->bvec_iter));
867 }
868
869 if (!cursor->last_piece) {
870 BUG_ON(!cursor->resid);
871 BUG_ON(!bio);
872 /* A short read is OK, so use <= rather than == */
873 if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter))
874 cursor->last_piece = true;
875 }
876
877 return true;
878}
879#endif /* CONFIG_BLOCK */
880
881/*
882 * For a page array, a piece comes from the first page in the array
883 * that has not already been fully consumed.
884 */
885static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
886 size_t length)
887{
888 struct ceph_msg_data *data = cursor->data;
889 int page_count;
890
891 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
892
893 BUG_ON(!data->pages);
894 BUG_ON(!data->length);
895
896 cursor->resid = min(length, data->length);
897 page_count = calc_pages_for(data->alignment, (u64)data->length);
898 cursor->page_offset = data->alignment & ~PAGE_MASK;
899 cursor->page_index = 0;
900 BUG_ON(page_count > (int)USHRT_MAX);
901 cursor->page_count = (unsigned short)page_count;
902 BUG_ON(length > SIZE_MAX - cursor->page_offset);
903 cursor->last_piece = (size_t)cursor->page_offset + length <= PAGE_SIZE;
904}
905
906static struct page *
907ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
908 size_t *page_offset, size_t *length)
909{
910 struct ceph_msg_data *data = cursor->data;
911
912 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
913
914 BUG_ON(cursor->page_index >= cursor->page_count);
915 BUG_ON(cursor->page_offset >= PAGE_SIZE);
916
917 *page_offset = cursor->page_offset;
918 if (cursor->last_piece)
919 *length = cursor->resid;
920 else
921 *length = PAGE_SIZE - *page_offset;
922
923 return data->pages[cursor->page_index];
924}
925
926static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
927 size_t bytes)
928{
929 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
930
931 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
932
933 /* Advance the cursor page offset */
934
935 cursor->resid -= bytes;
936 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
937 if (!bytes || cursor->page_offset)
938 return false; /* more bytes to process in the current page */
939
940 if (!cursor->resid)
941 return false; /* no more data */
942
943 /* Move on to the next page; offset is already at 0 */
944
945 BUG_ON(cursor->page_index >= cursor->page_count);
946 cursor->page_index++;
947 cursor->last_piece = cursor->resid <= PAGE_SIZE;
948
949 return true;
950}
951
952/*
953 * For a pagelist, a piece is whatever remains to be consumed in the
954 * first page in the list, or the front of the next page.
955 */
956static void
957ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
958 size_t length)
959{
960 struct ceph_msg_data *data = cursor->data;
961 struct ceph_pagelist *pagelist;
962 struct page *page;
963
964 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
965
966 pagelist = data->pagelist;
967 BUG_ON(!pagelist);
968
969 if (!length)
970 return; /* pagelist can be assigned but empty */
971
972 BUG_ON(list_empty(&pagelist->head));
973 page = list_first_entry(&pagelist->head, struct page, lru);
974
975 cursor->resid = min(length, pagelist->length);
976 cursor->page = page;
977 cursor->offset = 0;
978 cursor->last_piece = cursor->resid <= PAGE_SIZE;
979}
980
981static struct page *
982ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
983 size_t *page_offset, size_t *length)
984{
985 struct ceph_msg_data *data = cursor->data;
986 struct ceph_pagelist *pagelist;
987
988 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
989
990 pagelist = data->pagelist;
991 BUG_ON(!pagelist);
992
993 BUG_ON(!cursor->page);
994 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
995
996 /* offset of first page in pagelist is always 0 */
997 *page_offset = cursor->offset & ~PAGE_MASK;
998 if (cursor->last_piece)
999 *length = cursor->resid;
1000 else
1001 *length = PAGE_SIZE - *page_offset;
1002
1003 return cursor->page;
1004}
1005
1006static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
1007 size_t bytes)
1008{
1009 struct ceph_msg_data *data = cursor->data;
1010 struct ceph_pagelist *pagelist;
1011
1012 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1013
1014 pagelist = data->pagelist;
1015 BUG_ON(!pagelist);
1016
1017 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1018 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1019
1020 /* Advance the cursor offset */
1021
1022 cursor->resid -= bytes;
1023 cursor->offset += bytes;
1024 /* offset of first page in pagelist is always 0 */
1025 if (!bytes || cursor->offset & ~PAGE_MASK)
1026 return false; /* more bytes to process in the current page */
1027
1028 if (!cursor->resid)
1029 return false; /* no more data */
1030
1031 /* Move on to the next page */
1032
1033 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1034 cursor->page = list_entry_next(cursor->page, lru);
1035 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1036
1037 return true;
1038}
1039
1040/*
1041 * Message data is handled (sent or received) in pieces, where each
1042 * piece resides on a single page. The network layer might not
1043 * consume an entire piece at once. A data item's cursor keeps
1044 * track of which piece is next to process and how much remains to
1045 * be processed in that piece. It also tracks whether the current
1046 * piece is the last one in the data item.
1047 */
1048static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1049{
1050 size_t length = cursor->total_resid;
1051
1052 switch (cursor->data->type) {
1053 case CEPH_MSG_DATA_PAGELIST:
1054 ceph_msg_data_pagelist_cursor_init(cursor, length);
1055 break;
1056 case CEPH_MSG_DATA_PAGES:
1057 ceph_msg_data_pages_cursor_init(cursor, length);
1058 break;
1059#ifdef CONFIG_BLOCK
1060 case CEPH_MSG_DATA_BIO:
1061 ceph_msg_data_bio_cursor_init(cursor, length);
1062 break;
1063#endif /* CONFIG_BLOCK */
1064 case CEPH_MSG_DATA_NONE:
1065 default:
1066 /* BUG(); */
1067 break;
1068 }
1069 cursor->need_crc = true;
1070}
1071
1072static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1073{
1074 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1075 struct ceph_msg_data *data;
1076
1077 BUG_ON(!length);
1078 BUG_ON(length > msg->data_length);
1079 BUG_ON(list_empty(&msg->data));
1080
1081 cursor->data_head = &msg->data;
1082 cursor->total_resid = length;
1083 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1084 cursor->data = data;
1085
1086 __ceph_msg_data_cursor_init(cursor);
1087}
1088
1089/*
1090 * Return the page containing the next piece to process for a given
1091 * data item, and supply the page offset and length of that piece.
1092 * Indicate whether this is the last piece in this data item.
1093 */
1094static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1095 size_t *page_offset, size_t *length,
1096 bool *last_piece)
1097{
1098 struct page *page;
1099
1100 switch (cursor->data->type) {
1101 case CEPH_MSG_DATA_PAGELIST:
1102 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1103 break;
1104 case CEPH_MSG_DATA_PAGES:
1105 page = ceph_msg_data_pages_next(cursor, page_offset, length);
1106 break;
1107#ifdef CONFIG_BLOCK
1108 case CEPH_MSG_DATA_BIO:
1109 page = ceph_msg_data_bio_next(cursor, page_offset, length);
1110 break;
1111#endif /* CONFIG_BLOCK */
1112 case CEPH_MSG_DATA_NONE:
1113 default:
1114 page = NULL;
1115 break;
1116 }
1117 BUG_ON(!page);
1118 BUG_ON(*page_offset + *length > PAGE_SIZE);
1119 BUG_ON(!*length);
1120 if (last_piece)
1121 *last_piece = cursor->last_piece;
1122
1123 return page;
1124}
1125
1126/*
1127 * Returns true if the result moves the cursor on to the next piece
1128 * of the data item.
1129 */
1130static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1131 size_t bytes)
1132{
1133 bool new_piece;
1134
1135 BUG_ON(bytes > cursor->resid);
1136 switch (cursor->data->type) {
1137 case CEPH_MSG_DATA_PAGELIST:
1138 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1139 break;
1140 case CEPH_MSG_DATA_PAGES:
1141 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1142 break;
1143#ifdef CONFIG_BLOCK
1144 case CEPH_MSG_DATA_BIO:
1145 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1146 break;
1147#endif /* CONFIG_BLOCK */
1148 case CEPH_MSG_DATA_NONE:
1149 default:
1150 BUG();
1151 break;
1152 }
1153 cursor->total_resid -= bytes;
1154
1155 if (!cursor->resid && cursor->total_resid) {
1156 WARN_ON(!cursor->last_piece);
1157 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1158 cursor->data = list_entry_next(cursor->data, links);
1159 __ceph_msg_data_cursor_init(cursor);
1160 new_piece = true;
1161 }
1162 cursor->need_crc = new_piece;
1163
1164 return new_piece;
1165}
1166
1167static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
1168{
1169 BUG_ON(!msg);
1170 BUG_ON(!data_len);
1171
1172 /* Initialize data cursor */
1173
1174 ceph_msg_data_cursor_init(msg, (size_t)data_len);
1175}
1176
1177/*
1178 * Prepare footer for currently outgoing message, and finish things
1179 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1180 */
1181static void prepare_write_message_footer(struct ceph_connection *con)
1182{
1183 struct ceph_msg *m = con->out_msg;
1184 int v = con->out_kvec_left;
1185
1186 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1187
1188 dout("prepare_write_message_footer %p\n", con);
1189 con->out_kvec_is_msg = true;
1190 con->out_kvec[v].iov_base = &m->footer;
1191 con->out_kvec[v].iov_len = sizeof(m->footer);
1192 con->out_kvec_bytes += sizeof(m->footer);
1193 con->out_kvec_left++;
1194 con->out_more = m->more_to_follow;
1195 con->out_msg_done = true;
1196}
1197
1198/*
1199 * Prepare headers for the next outgoing message.
1200 */
1201static void prepare_write_message(struct ceph_connection *con)
1202{
1203 struct ceph_msg *m;
1204 u32 crc;
1205
1206 con_out_kvec_reset(con);
1207 con->out_kvec_is_msg = true;
1208 con->out_msg_done = false;
1209
1210 /* Sneak an ack in there first? If we can get it into the same
1211 * TCP packet that's a good thing. */
1212 if (con->in_seq > con->in_seq_acked) {
1213 con->in_seq_acked = con->in_seq;
1214 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1215 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1216 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1217 &con->out_temp_ack);
1218 }
1219
1220 BUG_ON(list_empty(&con->out_queue));
1221 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1222 con->out_msg = m;
1223 BUG_ON(m->con != con);
1224
1225 /* put message on sent list */
1226 ceph_msg_get(m);
1227 list_move_tail(&m->list_head, &con->out_sent);
1228
1229 /*
1230 * only assign outgoing seq # if we haven't sent this message
1231 * yet. if it is requeued, resend with it's original seq.
1232 */
1233 if (m->needs_out_seq) {
1234 m->hdr.seq = cpu_to_le64(++con->out_seq);
1235 m->needs_out_seq = false;
1236 }
1237 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
1238
1239 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1240 m, con->out_seq, le16_to_cpu(m->hdr.type),
1241 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1242 m->data_length);
1243 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1244
1245 /* tag + hdr + front + middle */
1246 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1247 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1248 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1249
1250 if (m->middle)
1251 con_out_kvec_add(con, m->middle->vec.iov_len,
1252 m->middle->vec.iov_base);
1253
1254 /* fill in crc (except data pages), footer */
1255 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1256 con->out_msg->hdr.crc = cpu_to_le32(crc);
1257 con->out_msg->footer.flags = 0;
1258
1259 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1260 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1261 if (m->middle) {
1262 crc = crc32c(0, m->middle->vec.iov_base,
1263 m->middle->vec.iov_len);
1264 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1265 } else
1266 con->out_msg->footer.middle_crc = 0;
1267 dout("%s front_crc %u middle_crc %u\n", __func__,
1268 le32_to_cpu(con->out_msg->footer.front_crc),
1269 le32_to_cpu(con->out_msg->footer.middle_crc));
1270
1271 /* is there a data payload? */
1272 con->out_msg->footer.data_crc = 0;
1273 if (m->data_length) {
1274 prepare_message_data(con->out_msg, m->data_length);
1275 con->out_more = 1; /* data + footer will follow */
1276 } else {
1277 /* no, queue up footer too and be done */
1278 prepare_write_message_footer(con);
1279 }
1280
1281 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1282}
1283
1284/*
1285 * Prepare an ack.
1286 */
1287static void prepare_write_ack(struct ceph_connection *con)
1288{
1289 dout("prepare_write_ack %p %llu -> %llu\n", con,
1290 con->in_seq_acked, con->in_seq);
1291 con->in_seq_acked = con->in_seq;
1292
1293 con_out_kvec_reset(con);
1294
1295 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1296
1297 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1298 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1299 &con->out_temp_ack);
1300
1301 con->out_more = 1; /* more will follow.. eventually.. */
1302 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1303}
1304
1305/*
1306 * Prepare to share the seq during handshake
1307 */
1308static void prepare_write_seq(struct ceph_connection *con)
1309{
1310 dout("prepare_write_seq %p %llu -> %llu\n", con,
1311 con->in_seq_acked, con->in_seq);
1312 con->in_seq_acked = con->in_seq;
1313
1314 con_out_kvec_reset(con);
1315
1316 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1317 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1318 &con->out_temp_ack);
1319
1320 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1321}
1322
1323/*
1324 * Prepare to write keepalive byte.
1325 */
1326static void prepare_write_keepalive(struct ceph_connection *con)
1327{
1328 dout("prepare_write_keepalive %p\n", con);
1329 con_out_kvec_reset(con);
1330 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
1331 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1332}
1333
1334/*
1335 * Connection negotiation.
1336 */
1337
1338static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1339 int *auth_proto)
1340{
1341 struct ceph_auth_handshake *auth;
1342
1343 if (!con->ops->get_authorizer) {
1344 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1345 con->out_connect.authorizer_len = 0;
1346 return NULL;
1347 }
1348
1349 /* Can't hold the mutex while getting authorizer */
1350 mutex_unlock(&con->mutex);
1351 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
1352 mutex_lock(&con->mutex);
1353
1354 if (IS_ERR(auth))
1355 return auth;
1356 if (con->state != CON_STATE_NEGOTIATING)
1357 return ERR_PTR(-EAGAIN);
1358
1359 con->auth_reply_buf = auth->authorizer_reply_buf;
1360 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
1361 return auth;
1362}
1363
1364/*
1365 * We connected to a peer and are saying hello.
1366 */
1367static void prepare_write_banner(struct ceph_connection *con)
1368{
1369 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1370 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1371 &con->msgr->my_enc_addr);
1372
1373 con->out_more = 0;
1374 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1375}
1376
1377static int prepare_write_connect(struct ceph_connection *con)
1378{
1379 unsigned int global_seq = get_global_seq(con->msgr, 0);
1380 int proto;
1381 int auth_proto;
1382 struct ceph_auth_handshake *auth;
1383
1384 switch (con->peer_name.type) {
1385 case CEPH_ENTITY_TYPE_MON:
1386 proto = CEPH_MONC_PROTOCOL;
1387 break;
1388 case CEPH_ENTITY_TYPE_OSD:
1389 proto = CEPH_OSDC_PROTOCOL;
1390 break;
1391 case CEPH_ENTITY_TYPE_MDS:
1392 proto = CEPH_MDSC_PROTOCOL;
1393 break;
1394 default:
1395 BUG();
1396 }
1397
1398 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1399 con->connect_seq, global_seq, proto);
1400
1401 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
1402 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1403 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1404 con->out_connect.global_seq = cpu_to_le32(global_seq);
1405 con->out_connect.protocol_version = cpu_to_le32(proto);
1406 con->out_connect.flags = 0;
1407
1408 auth_proto = CEPH_AUTH_UNKNOWN;
1409 auth = get_connect_authorizer(con, &auth_proto);
1410 if (IS_ERR(auth))
1411 return PTR_ERR(auth);
1412
1413 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1414 con->out_connect.authorizer_len = auth ?
1415 cpu_to_le32(auth->authorizer_buf_len) : 0;
1416
1417 con_out_kvec_add(con, sizeof (con->out_connect),
1418 &con->out_connect);
1419 if (auth && auth->authorizer_buf_len)
1420 con_out_kvec_add(con, auth->authorizer_buf_len,
1421 auth->authorizer_buf);
1422
1423 con->out_more = 0;
1424 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1425
1426 return 0;
1427}
1428
1429/*
1430 * write as much of pending kvecs to the socket as we can.
1431 * 1 -> done
1432 * 0 -> socket full, but more to do
1433 * <0 -> error
1434 */
1435static int write_partial_kvec(struct ceph_connection *con)
1436{
1437 int ret;
1438
1439 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1440 while (con->out_kvec_bytes > 0) {
1441 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1442 con->out_kvec_left, con->out_kvec_bytes,
1443 con->out_more);
1444 if (ret <= 0)
1445 goto out;
1446 con->out_kvec_bytes -= ret;
1447 if (con->out_kvec_bytes == 0)
1448 break; /* done */
1449
1450 /* account for full iov entries consumed */
1451 while (ret >= con->out_kvec_cur->iov_len) {
1452 BUG_ON(!con->out_kvec_left);
1453 ret -= con->out_kvec_cur->iov_len;
1454 con->out_kvec_cur++;
1455 con->out_kvec_left--;
1456 }
1457 /* and for a partially-consumed entry */
1458 if (ret) {
1459 con->out_kvec_cur->iov_len -= ret;
1460 con->out_kvec_cur->iov_base += ret;
1461 }
1462 }
1463 con->out_kvec_left = 0;
1464 con->out_kvec_is_msg = false;
1465 ret = 1;
1466out:
1467 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1468 con->out_kvec_bytes, con->out_kvec_left, ret);
1469 return ret; /* done! */
1470}
1471
1472static u32 ceph_crc32c_page(u32 crc, struct page *page,
1473 unsigned int page_offset,
1474 unsigned int length)
1475{
1476 char *kaddr;
1477
1478 kaddr = kmap(page);
1479 BUG_ON(kaddr == NULL);
1480 crc = crc32c(crc, kaddr + page_offset, length);
1481 kunmap(page);
1482
1483 return crc;
1484}
1485/*
1486 * Write as much message data payload as we can. If we finish, queue
1487 * up the footer.
1488 * 1 -> done, footer is now queued in out_kvec[].
1489 * 0 -> socket full, but more to do
1490 * <0 -> error
1491 */
1492static int write_partial_message_data(struct ceph_connection *con)
1493{
1494 struct ceph_msg *msg = con->out_msg;
1495 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1496 bool do_datacrc = !con->msgr->nocrc;
1497 u32 crc;
1498
1499 dout("%s %p msg %p\n", __func__, con, msg);
1500
1501 if (list_empty(&msg->data))
1502 return -EINVAL;
1503
1504 /*
1505 * Iterate through each page that contains data to be
1506 * written, and send as much as possible for each.
1507 *
1508 * If we are calculating the data crc (the default), we will
1509 * need to map the page. If we have no pages, they have
1510 * been revoked, so use the zero page.
1511 */
1512 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1513 while (cursor->resid) {
1514 struct page *page;
1515 size_t page_offset;
1516 size_t length;
1517 bool last_piece;
1518 bool need_crc;
1519 int ret;
1520
1521 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
1522 &last_piece);
1523 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
1524 length, last_piece);
1525 if (ret <= 0) {
1526 if (do_datacrc)
1527 msg->footer.data_crc = cpu_to_le32(crc);
1528
1529 return ret;
1530 }
1531 if (do_datacrc && cursor->need_crc)
1532 crc = ceph_crc32c_page(crc, page, page_offset, length);
1533 need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret);
1534 }
1535
1536 dout("%s %p msg %p done\n", __func__, con, msg);
1537
1538 /* prepare and queue up footer, too */
1539 if (do_datacrc)
1540 msg->footer.data_crc = cpu_to_le32(crc);
1541 else
1542 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1543 con_out_kvec_reset(con);
1544 prepare_write_message_footer(con);
1545
1546 return 1; /* must return > 0 to indicate success */
1547}
1548
1549/*
1550 * write some zeros
1551 */
1552static int write_partial_skip(struct ceph_connection *con)
1553{
1554 int ret;
1555
1556 while (con->out_skip > 0) {
1557 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
1558
1559 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
1560 if (ret <= 0)
1561 goto out;
1562 con->out_skip -= ret;
1563 }
1564 ret = 1;
1565out:
1566 return ret;
1567}
1568
1569/*
1570 * Prepare to read connection handshake, or an ack.
1571 */
1572static void prepare_read_banner(struct ceph_connection *con)
1573{
1574 dout("prepare_read_banner %p\n", con);
1575 con->in_base_pos = 0;
1576}
1577
1578static void prepare_read_connect(struct ceph_connection *con)
1579{
1580 dout("prepare_read_connect %p\n", con);
1581 con->in_base_pos = 0;
1582}
1583
1584static void prepare_read_ack(struct ceph_connection *con)
1585{
1586 dout("prepare_read_ack %p\n", con);
1587 con->in_base_pos = 0;
1588}
1589
1590static void prepare_read_seq(struct ceph_connection *con)
1591{
1592 dout("prepare_read_seq %p\n", con);
1593 con->in_base_pos = 0;
1594 con->in_tag = CEPH_MSGR_TAG_SEQ;
1595}
1596
1597static void prepare_read_tag(struct ceph_connection *con)
1598{
1599 dout("prepare_read_tag %p\n", con);
1600 con->in_base_pos = 0;
1601 con->in_tag = CEPH_MSGR_TAG_READY;
1602}
1603
1604/*
1605 * Prepare to read a message.
1606 */
1607static int prepare_read_message(struct ceph_connection *con)
1608{
1609 dout("prepare_read_message %p\n", con);
1610 BUG_ON(con->in_msg != NULL);
1611 con->in_base_pos = 0;
1612 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1613 return 0;
1614}
1615
1616
1617static int read_partial(struct ceph_connection *con,
1618 int end, int size, void *object)
1619{
1620 while (con->in_base_pos < end) {
1621 int left = end - con->in_base_pos;
1622 int have = size - left;
1623 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1624 if (ret <= 0)
1625 return ret;
1626 con->in_base_pos += ret;
1627 }
1628 return 1;
1629}
1630
1631
1632/*
1633 * Read all or part of the connect-side handshake on a new connection
1634 */
1635static int read_partial_banner(struct ceph_connection *con)
1636{
1637 int size;
1638 int end;
1639 int ret;
1640
1641 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1642
1643 /* peer's banner */
1644 size = strlen(CEPH_BANNER);
1645 end = size;
1646 ret = read_partial(con, end, size, con->in_banner);
1647 if (ret <= 0)
1648 goto out;
1649
1650 size = sizeof (con->actual_peer_addr);
1651 end += size;
1652 ret = read_partial(con, end, size, &con->actual_peer_addr);
1653 if (ret <= 0)
1654 goto out;
1655
1656 size = sizeof (con->peer_addr_for_me);
1657 end += size;
1658 ret = read_partial(con, end, size, &con->peer_addr_for_me);
1659 if (ret <= 0)
1660 goto out;
1661
1662out:
1663 return ret;
1664}
1665
1666static int read_partial_connect(struct ceph_connection *con)
1667{
1668 int size;
1669 int end;
1670 int ret;
1671
1672 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1673
1674 size = sizeof (con->in_reply);
1675 end = size;
1676 ret = read_partial(con, end, size, &con->in_reply);
1677 if (ret <= 0)
1678 goto out;
1679
1680 size = le32_to_cpu(con->in_reply.authorizer_len);
1681 end += size;
1682 ret = read_partial(con, end, size, con->auth_reply_buf);
1683 if (ret <= 0)
1684 goto out;
1685
1686 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1687 con, (int)con->in_reply.tag,
1688 le32_to_cpu(con->in_reply.connect_seq),
1689 le32_to_cpu(con->in_reply.global_seq));
1690out:
1691 return ret;
1692
1693}
1694
1695/*
1696 * Verify the hello banner looks okay.
1697 */
1698static int verify_hello(struct ceph_connection *con)
1699{
1700 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1701 pr_err("connect to %s got bad banner\n",
1702 ceph_pr_addr(&con->peer_addr.in_addr));
1703 con->error_msg = "protocol error, bad banner";
1704 return -1;
1705 }
1706 return 0;
1707}
1708
1709static bool addr_is_blank(struct sockaddr_storage *ss)
1710{
1711 switch (ss->ss_family) {
1712 case AF_INET:
1713 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1714 case AF_INET6:
1715 return
1716 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1717 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1718 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1719 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1720 }
1721 return false;
1722}
1723
1724static int addr_port(struct sockaddr_storage *ss)
1725{
1726 switch (ss->ss_family) {
1727 case AF_INET:
1728 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1729 case AF_INET6:
1730 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1731 }
1732 return 0;
1733}
1734
1735static void addr_set_port(struct sockaddr_storage *ss, int p)
1736{
1737 switch (ss->ss_family) {
1738 case AF_INET:
1739 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1740 break;
1741 case AF_INET6:
1742 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1743 break;
1744 }
1745}
1746
1747/*
1748 * Unlike other *_pton function semantics, zero indicates success.
1749 */
1750static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1751 char delim, const char **ipend)
1752{
1753 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1754 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1755
1756 memset(ss, 0, sizeof(*ss));
1757
1758 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1759 ss->ss_family = AF_INET;
1760 return 0;
1761 }
1762
1763 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1764 ss->ss_family = AF_INET6;
1765 return 0;
1766 }
1767
1768 return -EINVAL;
1769}
1770
1771/*
1772 * Extract hostname string and resolve using kernel DNS facility.
1773 */
1774#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1775static int ceph_dns_resolve_name(const char *name, size_t namelen,
1776 struct sockaddr_storage *ss, char delim, const char **ipend)
1777{
1778 const char *end, *delim_p;
1779 char *colon_p, *ip_addr = NULL;
1780 int ip_len, ret;
1781
1782 /*
1783 * The end of the hostname occurs immediately preceding the delimiter or
1784 * the port marker (':') where the delimiter takes precedence.
1785 */
1786 delim_p = memchr(name, delim, namelen);
1787 colon_p = memchr(name, ':', namelen);
1788
1789 if (delim_p && colon_p)
1790 end = delim_p < colon_p ? delim_p : colon_p;
1791 else if (!delim_p && colon_p)
1792 end = colon_p;
1793 else {
1794 end = delim_p;
1795 if (!end) /* case: hostname:/ */
1796 end = name + namelen;
1797 }
1798
1799 if (end <= name)
1800 return -EINVAL;
1801
1802 /* do dns_resolve upcall */
1803 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1804 if (ip_len > 0)
1805 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1806 else
1807 ret = -ESRCH;
1808
1809 kfree(ip_addr);
1810
1811 *ipend = end;
1812
1813 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1814 ret, ret ? "failed" : ceph_pr_addr(ss));
1815
1816 return ret;
1817}
1818#else
1819static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1820 struct sockaddr_storage *ss, char delim, const char **ipend)
1821{
1822 return -EINVAL;
1823}
1824#endif
1825
1826/*
1827 * Parse a server name (IP or hostname). If a valid IP address is not found
1828 * then try to extract a hostname to resolve using userspace DNS upcall.
1829 */
1830static int ceph_parse_server_name(const char *name, size_t namelen,
1831 struct sockaddr_storage *ss, char delim, const char **ipend)
1832{
1833 int ret;
1834
1835 ret = ceph_pton(name, namelen, ss, delim, ipend);
1836 if (ret)
1837 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1838
1839 return ret;
1840}
1841
1842/*
1843 * Parse an ip[:port] list into an addr array. Use the default
1844 * monitor port if a port isn't specified.
1845 */
1846int ceph_parse_ips(const char *c, const char *end,
1847 struct ceph_entity_addr *addr,
1848 int max_count, int *count)
1849{
1850 int i, ret = -EINVAL;
1851 const char *p = c;
1852
1853 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1854 for (i = 0; i < max_count; i++) {
1855 const char *ipend;
1856 struct sockaddr_storage *ss = &addr[i].in_addr;
1857 int port;
1858 char delim = ',';
1859
1860 if (*p == '[') {
1861 delim = ']';
1862 p++;
1863 }
1864
1865 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1866 if (ret)
1867 goto bad;
1868 ret = -EINVAL;
1869
1870 p = ipend;
1871
1872 if (delim == ']') {
1873 if (*p != ']') {
1874 dout("missing matching ']'\n");
1875 goto bad;
1876 }
1877 p++;
1878 }
1879
1880 /* port? */
1881 if (p < end && *p == ':') {
1882 port = 0;
1883 p++;
1884 while (p < end && *p >= '0' && *p <= '9') {
1885 port = (port * 10) + (*p - '0');
1886 p++;
1887 }
1888 if (port == 0)
1889 port = CEPH_MON_PORT;
1890 else if (port > 65535)
1891 goto bad;
1892 } else {
1893 port = CEPH_MON_PORT;
1894 }
1895
1896 addr_set_port(ss, port);
1897
1898 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1899
1900 if (p == end)
1901 break;
1902 if (*p != ',')
1903 goto bad;
1904 p++;
1905 }
1906
1907 if (p != end)
1908 goto bad;
1909
1910 if (count)
1911 *count = i + 1;
1912 return 0;
1913
1914bad:
1915 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1916 return ret;
1917}
1918EXPORT_SYMBOL(ceph_parse_ips);
1919
1920static int process_banner(struct ceph_connection *con)
1921{
1922 dout("process_banner on %p\n", con);
1923
1924 if (verify_hello(con) < 0)
1925 return -1;
1926
1927 ceph_decode_addr(&con->actual_peer_addr);
1928 ceph_decode_addr(&con->peer_addr_for_me);
1929
1930 /*
1931 * Make sure the other end is who we wanted. note that the other
1932 * end may not yet know their ip address, so if it's 0.0.0.0, give
1933 * them the benefit of the doubt.
1934 */
1935 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1936 sizeof(con->peer_addr)) != 0 &&
1937 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1938 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1939 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1940 ceph_pr_addr(&con->peer_addr.in_addr),
1941 (int)le32_to_cpu(con->peer_addr.nonce),
1942 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1943 (int)le32_to_cpu(con->actual_peer_addr.nonce));
1944 con->error_msg = "wrong peer at address";
1945 return -1;
1946 }
1947
1948 /*
1949 * did we learn our address?
1950 */
1951 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1952 int port = addr_port(&con->msgr->inst.addr.in_addr);
1953
1954 memcpy(&con->msgr->inst.addr.in_addr,
1955 &con->peer_addr_for_me.in_addr,
1956 sizeof(con->peer_addr_for_me.in_addr));
1957 addr_set_port(&con->msgr->inst.addr.in_addr, port);
1958 encode_my_addr(con->msgr);
1959 dout("process_banner learned my addr is %s\n",
1960 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
1961 }
1962
1963 return 0;
1964}
1965
1966static int process_connect(struct ceph_connection *con)
1967{
1968 u64 sup_feat = con->msgr->supported_features;
1969 u64 req_feat = con->msgr->required_features;
1970 u64 server_feat = ceph_sanitize_features(
1971 le64_to_cpu(con->in_reply.features));
1972 int ret;
1973
1974 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1975
1976 switch (con->in_reply.tag) {
1977 case CEPH_MSGR_TAG_FEATURES:
1978 pr_err("%s%lld %s feature set mismatch,"
1979 " my %llx < server's %llx, missing %llx\n",
1980 ENTITY_NAME(con->peer_name),
1981 ceph_pr_addr(&con->peer_addr.in_addr),
1982 sup_feat, server_feat, server_feat & ~sup_feat);
1983 con->error_msg = "missing required protocol features";
1984 reset_connection(con);
1985 return -1;
1986
1987 case CEPH_MSGR_TAG_BADPROTOVER:
1988 pr_err("%s%lld %s protocol version mismatch,"
1989 " my %d != server's %d\n",
1990 ENTITY_NAME(con->peer_name),
1991 ceph_pr_addr(&con->peer_addr.in_addr),
1992 le32_to_cpu(con->out_connect.protocol_version),
1993 le32_to_cpu(con->in_reply.protocol_version));
1994 con->error_msg = "protocol version mismatch";
1995 reset_connection(con);
1996 return -1;
1997
1998 case CEPH_MSGR_TAG_BADAUTHORIZER:
1999 con->auth_retry++;
2000 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2001 con->auth_retry);
2002 if (con->auth_retry == 2) {
2003 con->error_msg = "connect authorization failure";
2004 return -1;
2005 }
2006 con_out_kvec_reset(con);
2007 ret = prepare_write_connect(con);
2008 if (ret < 0)
2009 return ret;
2010 prepare_read_connect(con);
2011 break;
2012
2013 case CEPH_MSGR_TAG_RESETSESSION:
2014 /*
2015 * If we connected with a large connect_seq but the peer
2016 * has no record of a session with us (no connection, or
2017 * connect_seq == 0), they will send RESETSESION to indicate
2018 * that they must have reset their session, and may have
2019 * dropped messages.
2020 */
2021 dout("process_connect got RESET peer seq %u\n",
2022 le32_to_cpu(con->in_reply.connect_seq));
2023 pr_err("%s%lld %s connection reset\n",
2024 ENTITY_NAME(con->peer_name),
2025 ceph_pr_addr(&con->peer_addr.in_addr));
2026 reset_connection(con);
2027 con_out_kvec_reset(con);
2028 ret = prepare_write_connect(con);
2029 if (ret < 0)
2030 return ret;
2031 prepare_read_connect(con);
2032
2033 /* Tell ceph about it. */
2034 mutex_unlock(&con->mutex);
2035 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2036 if (con->ops->peer_reset)
2037 con->ops->peer_reset(con);
2038 mutex_lock(&con->mutex);
2039 if (con->state != CON_STATE_NEGOTIATING)
2040 return -EAGAIN;
2041 break;
2042
2043 case CEPH_MSGR_TAG_RETRY_SESSION:
2044 /*
2045 * If we sent a smaller connect_seq than the peer has, try
2046 * again with a larger value.
2047 */
2048 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2049 le32_to_cpu(con->out_connect.connect_seq),
2050 le32_to_cpu(con->in_reply.connect_seq));
2051 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2052 con_out_kvec_reset(con);
2053 ret = prepare_write_connect(con);
2054 if (ret < 0)
2055 return ret;
2056 prepare_read_connect(con);
2057 break;
2058
2059 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2060 /*
2061 * If we sent a smaller global_seq than the peer has, try
2062 * again with a larger value.
2063 */
2064 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2065 con->peer_global_seq,
2066 le32_to_cpu(con->in_reply.global_seq));
2067 get_global_seq(con->msgr,
2068 le32_to_cpu(con->in_reply.global_seq));
2069 con_out_kvec_reset(con);
2070 ret = prepare_write_connect(con);
2071 if (ret < 0)
2072 return ret;
2073 prepare_read_connect(con);
2074 break;
2075
2076 case CEPH_MSGR_TAG_SEQ:
2077 case CEPH_MSGR_TAG_READY:
2078 if (req_feat & ~server_feat) {
2079 pr_err("%s%lld %s protocol feature mismatch,"
2080 " my required %llx > server's %llx, need %llx\n",
2081 ENTITY_NAME(con->peer_name),
2082 ceph_pr_addr(&con->peer_addr.in_addr),
2083 req_feat, server_feat, req_feat & ~server_feat);
2084 con->error_msg = "missing required protocol features";
2085 reset_connection(con);
2086 return -1;
2087 }
2088
2089 WARN_ON(con->state != CON_STATE_NEGOTIATING);
2090 con->state = CON_STATE_OPEN;
2091 con->auth_retry = 0; /* we authenticated; clear flag */
2092 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2093 con->connect_seq++;
2094 con->peer_features = server_feat;
2095 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2096 con->peer_global_seq,
2097 le32_to_cpu(con->in_reply.connect_seq),
2098 con->connect_seq);
2099 WARN_ON(con->connect_seq !=
2100 le32_to_cpu(con->in_reply.connect_seq));
2101
2102 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2103 con_flag_set(con, CON_FLAG_LOSSYTX);
2104
2105 con->delay = 0; /* reset backoff memory */
2106
2107 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2108 prepare_write_seq(con);
2109 prepare_read_seq(con);
2110 } else {
2111 prepare_read_tag(con);
2112 }
2113 break;
2114
2115 case CEPH_MSGR_TAG_WAIT:
2116 /*
2117 * If there is a connection race (we are opening
2118 * connections to each other), one of us may just have
2119 * to WAIT. This shouldn't happen if we are the
2120 * client.
2121 */
2122 pr_err("process_connect got WAIT as client\n");
2123 con->error_msg = "protocol error, got WAIT as client";
2124 return -1;
2125
2126 default:
2127 pr_err("connect protocol error, will retry\n");
2128 con->error_msg = "protocol error, garbage tag during connect";
2129 return -1;
2130 }
2131 return 0;
2132}
2133
2134
2135/*
2136 * read (part of) an ack
2137 */
2138static int read_partial_ack(struct ceph_connection *con)
2139{
2140 int size = sizeof (con->in_temp_ack);
2141 int end = size;
2142
2143 return read_partial(con, end, size, &con->in_temp_ack);
2144}
2145
2146/*
2147 * We can finally discard anything that's been acked.
2148 */
2149static void process_ack(struct ceph_connection *con)
2150{
2151 struct ceph_msg *m;
2152 u64 ack = le64_to_cpu(con->in_temp_ack);
2153 u64 seq;
2154
2155 while (!list_empty(&con->out_sent)) {
2156 m = list_first_entry(&con->out_sent, struct ceph_msg,
2157 list_head);
2158 seq = le64_to_cpu(m->hdr.seq);
2159 if (seq > ack)
2160 break;
2161 dout("got ack for seq %llu type %d at %p\n", seq,
2162 le16_to_cpu(m->hdr.type), m);
2163 m->ack_stamp = jiffies;
2164 ceph_msg_remove(m);
2165 }
2166 prepare_read_tag(con);
2167}
2168
2169
2170static int read_partial_message_section(struct ceph_connection *con,
2171 struct kvec *section,
2172 unsigned int sec_len, u32 *crc)
2173{
2174 int ret, left;
2175
2176 BUG_ON(!section);
2177
2178 while (section->iov_len < sec_len) {
2179 BUG_ON(section->iov_base == NULL);
2180 left = sec_len - section->iov_len;
2181 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2182 section->iov_len, left);
2183 if (ret <= 0)
2184 return ret;
2185 section->iov_len += ret;
2186 }
2187 if (section->iov_len == sec_len)
2188 *crc = crc32c(0, section->iov_base, section->iov_len);
2189
2190 return 1;
2191}
2192
2193static int read_partial_msg_data(struct ceph_connection *con)
2194{
2195 struct ceph_msg *msg = con->in_msg;
2196 struct ceph_msg_data_cursor *cursor = &msg->cursor;
2197 const bool do_datacrc = !con->msgr->nocrc;
2198 struct page *page;
2199 size_t page_offset;
2200 size_t length;
2201 u32 crc = 0;
2202 int ret;
2203
2204 BUG_ON(!msg);
2205 if (list_empty(&msg->data))
2206 return -EIO;
2207
2208 if (do_datacrc)
2209 crc = con->in_data_crc;
2210 while (cursor->resid) {
2211 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
2212 NULL);
2213 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2214 if (ret <= 0) {
2215 if (do_datacrc)
2216 con->in_data_crc = crc;
2217
2218 return ret;
2219 }
2220
2221 if (do_datacrc)
2222 crc = ceph_crc32c_page(crc, page, page_offset, ret);
2223 (void) ceph_msg_data_advance(&msg->cursor, (size_t)ret);
2224 }
2225 if (do_datacrc)
2226 con->in_data_crc = crc;
2227
2228 return 1; /* must return > 0 to indicate success */
2229}
2230
2231/*
2232 * read (part of) a message.
2233 */
2234static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2235
2236static int read_partial_message(struct ceph_connection *con)
2237{
2238 struct ceph_msg *m = con->in_msg;
2239 int size;
2240 int end;
2241 int ret;
2242 unsigned int front_len, middle_len, data_len;
2243 bool do_datacrc = !con->msgr->nocrc;
2244 u64 seq;
2245 u32 crc;
2246
2247 dout("read_partial_message con %p msg %p\n", con, m);
2248
2249 /* header */
2250 size = sizeof (con->in_hdr);
2251 end = size;
2252 ret = read_partial(con, end, size, &con->in_hdr);
2253 if (ret <= 0)
2254 return ret;
2255
2256 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2257 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2258 pr_err("read_partial_message bad hdr "
2259 " crc %u != expected %u\n",
2260 crc, con->in_hdr.crc);
2261 return -EBADMSG;
2262 }
2263
2264 front_len = le32_to_cpu(con->in_hdr.front_len);
2265 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2266 return -EIO;
2267 middle_len = le32_to_cpu(con->in_hdr.middle_len);
2268 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2269 return -EIO;
2270 data_len = le32_to_cpu(con->in_hdr.data_len);
2271 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2272 return -EIO;
2273
2274 /* verify seq# */
2275 seq = le64_to_cpu(con->in_hdr.seq);
2276 if ((s64)seq - (s64)con->in_seq < 1) {
2277 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2278 ENTITY_NAME(con->peer_name),
2279 ceph_pr_addr(&con->peer_addr.in_addr),
2280 seq, con->in_seq + 1);
2281 con->in_base_pos = -front_len - middle_len - data_len -
2282 sizeof(m->footer);
2283 con->in_tag = CEPH_MSGR_TAG_READY;
2284 return 0;
2285 } else if ((s64)seq - (s64)con->in_seq > 1) {
2286 pr_err("read_partial_message bad seq %lld expected %lld\n",
2287 seq, con->in_seq + 1);
2288 con->error_msg = "bad message sequence # for incoming message";
2289 return -EBADMSG;
2290 }
2291
2292 /* allocate message? */
2293 if (!con->in_msg) {
2294 int skip = 0;
2295
2296 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2297 front_len, data_len);
2298 ret = ceph_con_in_msg_alloc(con, &skip);
2299 if (ret < 0)
2300 return ret;
2301
2302 BUG_ON(!con->in_msg ^ skip);
2303 if (con->in_msg && data_len > con->in_msg->data_length) {
2304 pr_warning("%s skipping long message (%u > %zd)\n",
2305 __func__, data_len, con->in_msg->data_length);
2306 ceph_msg_put(con->in_msg);
2307 con->in_msg = NULL;
2308 skip = 1;
2309 }
2310 if (skip) {
2311 /* skip this message */
2312 dout("alloc_msg said skip message\n");
2313 con->in_base_pos = -front_len - middle_len - data_len -
2314 sizeof(m->footer);
2315 con->in_tag = CEPH_MSGR_TAG_READY;
2316 con->in_seq++;
2317 return 0;
2318 }
2319
2320 BUG_ON(!con->in_msg);
2321 BUG_ON(con->in_msg->con != con);
2322 m = con->in_msg;
2323 m->front.iov_len = 0; /* haven't read it yet */
2324 if (m->middle)
2325 m->middle->vec.iov_len = 0;
2326
2327 /* prepare for data payload, if any */
2328
2329 if (data_len)
2330 prepare_message_data(con->in_msg, data_len);
2331 }
2332
2333 /* front */
2334 ret = read_partial_message_section(con, &m->front, front_len,
2335 &con->in_front_crc);
2336 if (ret <= 0)
2337 return ret;
2338
2339 /* middle */
2340 if (m->middle) {
2341 ret = read_partial_message_section(con, &m->middle->vec,
2342 middle_len,
2343 &con->in_middle_crc);
2344 if (ret <= 0)
2345 return ret;
2346 }
2347
2348 /* (page) data */
2349 if (data_len) {
2350 ret = read_partial_msg_data(con);
2351 if (ret <= 0)
2352 return ret;
2353 }
2354
2355 /* footer */
2356 size = sizeof (m->footer);
2357 end += size;
2358 ret = read_partial(con, end, size, &m->footer);
2359 if (ret <= 0)
2360 return ret;
2361
2362 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2363 m, front_len, m->footer.front_crc, middle_len,
2364 m->footer.middle_crc, data_len, m->footer.data_crc);
2365
2366 /* crc ok? */
2367 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2368 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2369 m, con->in_front_crc, m->footer.front_crc);
2370 return -EBADMSG;
2371 }
2372 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2373 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2374 m, con->in_middle_crc, m->footer.middle_crc);
2375 return -EBADMSG;
2376 }
2377 if (do_datacrc &&
2378 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2379 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2380 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2381 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2382 return -EBADMSG;
2383 }
2384
2385 return 1; /* done! */
2386}
2387
2388/*
2389 * Process message. This happens in the worker thread. The callback should
2390 * be careful not to do anything that waits on other incoming messages or it
2391 * may deadlock.
2392 */
2393static void process_message(struct ceph_connection *con)
2394{
2395 struct ceph_msg *msg;
2396
2397 BUG_ON(con->in_msg->con != con);
2398 con->in_msg->con = NULL;
2399 msg = con->in_msg;
2400 con->in_msg = NULL;
2401 con->ops->put(con);
2402
2403 /* if first message, set peer_name */
2404 if (con->peer_name.type == 0)
2405 con->peer_name = msg->hdr.src;
2406
2407 con->in_seq++;
2408 mutex_unlock(&con->mutex);
2409
2410 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2411 msg, le64_to_cpu(msg->hdr.seq),
2412 ENTITY_NAME(msg->hdr.src),
2413 le16_to_cpu(msg->hdr.type),
2414 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2415 le32_to_cpu(msg->hdr.front_len),
2416 le32_to_cpu(msg->hdr.data_len),
2417 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2418 con->ops->dispatch(con, msg);
2419
2420 mutex_lock(&con->mutex);
2421}
2422
2423
2424/*
2425 * Write something to the socket. Called in a worker thread when the
2426 * socket appears to be writeable and we have something ready to send.
2427 */
2428static int try_write(struct ceph_connection *con)
2429{
2430 int ret = 1;
2431
2432 dout("try_write start %p state %lu\n", con, con->state);
2433
2434more:
2435 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2436
2437 /* open the socket first? */
2438 if (con->state == CON_STATE_PREOPEN) {
2439 BUG_ON(con->sock);
2440 con->state = CON_STATE_CONNECTING;
2441
2442 con_out_kvec_reset(con);
2443 prepare_write_banner(con);
2444 prepare_read_banner(con);
2445
2446 BUG_ON(con->in_msg);
2447 con->in_tag = CEPH_MSGR_TAG_READY;
2448 dout("try_write initiating connect on %p new state %lu\n",
2449 con, con->state);
2450 ret = ceph_tcp_connect(con);
2451 if (ret < 0) {
2452 con->error_msg = "connect error";
2453 goto out;
2454 }
2455 }
2456
2457more_kvec:
2458 /* kvec data queued? */
2459 if (con->out_skip) {
2460 ret = write_partial_skip(con);
2461 if (ret <= 0)
2462 goto out;
2463 }
2464 if (con->out_kvec_left) {
2465 ret = write_partial_kvec(con);
2466 if (ret <= 0)
2467 goto out;
2468 }
2469
2470 /* msg pages? */
2471 if (con->out_msg) {
2472 if (con->out_msg_done) {
2473 ceph_msg_put(con->out_msg);
2474 con->out_msg = NULL; /* we're done with this one */
2475 goto do_next;
2476 }
2477
2478 ret = write_partial_message_data(con);
2479 if (ret == 1)
2480 goto more_kvec; /* we need to send the footer, too! */
2481 if (ret == 0)
2482 goto out;
2483 if (ret < 0) {
2484 dout("try_write write_partial_message_data err %d\n",
2485 ret);
2486 goto out;
2487 }
2488 }
2489
2490do_next:
2491 if (con->state == CON_STATE_OPEN) {
2492 /* is anything else pending? */
2493 if (!list_empty(&con->out_queue)) {
2494 prepare_write_message(con);
2495 goto more;
2496 }
2497 if (con->in_seq > con->in_seq_acked) {
2498 prepare_write_ack(con);
2499 goto more;
2500 }
2501 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2502 prepare_write_keepalive(con);
2503 goto more;
2504 }
2505 }
2506
2507 /* Nothing to do! */
2508 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2509 dout("try_write nothing else to write.\n");
2510 ret = 0;
2511out:
2512 dout("try_write done on %p ret %d\n", con, ret);
2513 return ret;
2514}
2515
2516
2517
2518/*
2519 * Read what we can from the socket.
2520 */
2521static int try_read(struct ceph_connection *con)
2522{
2523 int ret = -1;
2524
2525more:
2526 dout("try_read start on %p state %lu\n", con, con->state);
2527 if (con->state != CON_STATE_CONNECTING &&
2528 con->state != CON_STATE_NEGOTIATING &&
2529 con->state != CON_STATE_OPEN)
2530 return 0;
2531
2532 BUG_ON(!con->sock);
2533
2534 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2535 con->in_base_pos);
2536
2537 if (con->state == CON_STATE_CONNECTING) {
2538 dout("try_read connecting\n");
2539 ret = read_partial_banner(con);
2540 if (ret <= 0)
2541 goto out;
2542 ret = process_banner(con);
2543 if (ret < 0)
2544 goto out;
2545
2546 con->state = CON_STATE_NEGOTIATING;
2547
2548 /*
2549 * Received banner is good, exchange connection info.
2550 * Do not reset out_kvec, as sending our banner raced
2551 * with receiving peer banner after connect completed.
2552 */
2553 ret = prepare_write_connect(con);
2554 if (ret < 0)
2555 goto out;
2556 prepare_read_connect(con);
2557
2558 /* Send connection info before awaiting response */
2559 goto out;
2560 }
2561
2562 if (con->state == CON_STATE_NEGOTIATING) {
2563 dout("try_read negotiating\n");
2564 ret = read_partial_connect(con);
2565 if (ret <= 0)
2566 goto out;
2567 ret = process_connect(con);
2568 if (ret < 0)
2569 goto out;
2570 goto more;
2571 }
2572
2573 WARN_ON(con->state != CON_STATE_OPEN);
2574
2575 if (con->in_base_pos < 0) {
2576 /*
2577 * skipping + discarding content.
2578 *
2579 * FIXME: there must be a better way to do this!
2580 */
2581 static char buf[SKIP_BUF_SIZE];
2582 int skip = min((int) sizeof (buf), -con->in_base_pos);
2583
2584 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2585 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2586 if (ret <= 0)
2587 goto out;
2588 con->in_base_pos += ret;
2589 if (con->in_base_pos)
2590 goto more;
2591 }
2592 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2593 /*
2594 * what's next?
2595 */
2596 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2597 if (ret <= 0)
2598 goto out;
2599 dout("try_read got tag %d\n", (int)con->in_tag);
2600 switch (con->in_tag) {
2601 case CEPH_MSGR_TAG_MSG:
2602 prepare_read_message(con);
2603 break;
2604 case CEPH_MSGR_TAG_ACK:
2605 prepare_read_ack(con);
2606 break;
2607 case CEPH_MSGR_TAG_CLOSE:
2608 con_close_socket(con);
2609 con->state = CON_STATE_CLOSED;
2610 goto out;
2611 default:
2612 goto bad_tag;
2613 }
2614 }
2615 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2616 ret = read_partial_message(con);
2617 if (ret <= 0) {
2618 switch (ret) {
2619 case -EBADMSG:
2620 con->error_msg = "bad crc";
2621 ret = -EIO;
2622 break;
2623 case -EIO:
2624 con->error_msg = "io error";
2625 break;
2626 }
2627 goto out;
2628 }
2629 if (con->in_tag == CEPH_MSGR_TAG_READY)
2630 goto more;
2631 process_message(con);
2632 if (con->state == CON_STATE_OPEN)
2633 prepare_read_tag(con);
2634 goto more;
2635 }
2636 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2637 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2638 /*
2639 * the final handshake seq exchange is semantically
2640 * equivalent to an ACK
2641 */
2642 ret = read_partial_ack(con);
2643 if (ret <= 0)
2644 goto out;
2645 process_ack(con);
2646 goto more;
2647 }
2648
2649out:
2650 dout("try_read done on %p ret %d\n", con, ret);
2651 return ret;
2652
2653bad_tag:
2654 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2655 con->error_msg = "protocol error, garbage tag";
2656 ret = -1;
2657 goto out;
2658}
2659
2660
2661/*
2662 * Atomically queue work on a connection after the specified delay.
2663 * Bump @con reference to avoid races with connection teardown.
2664 * Returns 0 if work was queued, or an error code otherwise.
2665 */
2666static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2667{
2668 if (!con->ops->get(con)) {
2669 dout("%s %p ref count 0\n", __func__, con);
2670
2671 return -ENOENT;
2672 }
2673
2674 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2675 dout("%s %p - already queued\n", __func__, con);
2676 con->ops->put(con);
2677
2678 return -EBUSY;
2679 }
2680
2681 dout("%s %p %lu\n", __func__, con, delay);
2682
2683 return 0;
2684}
2685
2686static void queue_con(struct ceph_connection *con)
2687{
2688 (void) queue_con_delay(con, 0);
2689}
2690
2691static bool con_sock_closed(struct ceph_connection *con)
2692{
2693 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2694 return false;
2695
2696#define CASE(x) \
2697 case CON_STATE_ ## x: \
2698 con->error_msg = "socket closed (con state " #x ")"; \
2699 break;
2700
2701 switch (con->state) {
2702 CASE(CLOSED);
2703 CASE(PREOPEN);
2704 CASE(CONNECTING);
2705 CASE(NEGOTIATING);
2706 CASE(OPEN);
2707 CASE(STANDBY);
2708 default:
2709 pr_warning("%s con %p unrecognized state %lu\n",
2710 __func__, con, con->state);
2711 con->error_msg = "unrecognized con state";
2712 BUG();
2713 break;
2714 }
2715#undef CASE
2716
2717 return true;
2718}
2719
2720static bool con_backoff(struct ceph_connection *con)
2721{
2722 int ret;
2723
2724 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2725 return false;
2726
2727 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2728 if (ret) {
2729 dout("%s: con %p FAILED to back off %lu\n", __func__,
2730 con, con->delay);
2731 BUG_ON(ret == -ENOENT);
2732 con_flag_set(con, CON_FLAG_BACKOFF);
2733 }
2734
2735 return true;
2736}
2737
2738/* Finish fault handling; con->mutex must *not* be held here */
2739
2740static void con_fault_finish(struct ceph_connection *con)
2741{
2742 /*
2743 * in case we faulted due to authentication, invalidate our
2744 * current tickets so that we can get new ones.
2745 */
2746 if (con->auth_retry && con->ops->invalidate_authorizer) {
2747 dout("calling invalidate_authorizer()\n");
2748 con->ops->invalidate_authorizer(con);
2749 }
2750
2751 if (con->ops->fault)
2752 con->ops->fault(con);
2753}
2754
2755/*
2756 * Do some work on a connection. Drop a connection ref when we're done.
2757 */
2758static void con_work(struct work_struct *work)
2759{
2760 struct ceph_connection *con = container_of(work, struct ceph_connection,
2761 work.work);
2762 bool fault;
2763
2764 mutex_lock(&con->mutex);
2765 while (true) {
2766 int ret;
2767
2768 if ((fault = con_sock_closed(con))) {
2769 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2770 break;
2771 }
2772 if (con_backoff(con)) {
2773 dout("%s: con %p BACKOFF\n", __func__, con);
2774 break;
2775 }
2776 if (con->state == CON_STATE_STANDBY) {
2777 dout("%s: con %p STANDBY\n", __func__, con);
2778 break;
2779 }
2780 if (con->state == CON_STATE_CLOSED) {
2781 dout("%s: con %p CLOSED\n", __func__, con);
2782 BUG_ON(con->sock);
2783 break;
2784 }
2785 if (con->state == CON_STATE_PREOPEN) {
2786 dout("%s: con %p PREOPEN\n", __func__, con);
2787 BUG_ON(con->sock);
2788 }
2789
2790 ret = try_read(con);
2791 if (ret < 0) {
2792 if (ret == -EAGAIN)
2793 continue;
2794 con->error_msg = "socket error on read";
2795 fault = true;
2796 break;
2797 }
2798
2799 ret = try_write(con);
2800 if (ret < 0) {
2801 if (ret == -EAGAIN)
2802 continue;
2803 con->error_msg = "socket error on write";
2804 fault = true;
2805 }
2806
2807 break; /* If we make it to here, we're done */
2808 }
2809 if (fault)
2810 con_fault(con);
2811 mutex_unlock(&con->mutex);
2812
2813 if (fault)
2814 con_fault_finish(con);
2815
2816 con->ops->put(con);
2817}
2818
2819/*
2820 * Generic error/fault handler. A retry mechanism is used with
2821 * exponential backoff
2822 */
2823static void con_fault(struct ceph_connection *con)
2824{
2825 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2826 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2827 dout("fault %p state %lu to peer %s\n",
2828 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2829
2830 WARN_ON(con->state != CON_STATE_CONNECTING &&
2831 con->state != CON_STATE_NEGOTIATING &&
2832 con->state != CON_STATE_OPEN);
2833
2834 con_close_socket(con);
2835
2836 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2837 dout("fault on LOSSYTX channel, marking CLOSED\n");
2838 con->state = CON_STATE_CLOSED;
2839 return;
2840 }
2841
2842 if (con->in_msg) {
2843 BUG_ON(con->in_msg->con != con);
2844 con->in_msg->con = NULL;
2845 ceph_msg_put(con->in_msg);
2846 con->in_msg = NULL;
2847 con->ops->put(con);
2848 }
2849
2850 /* Requeue anything that hasn't been acked */
2851 list_splice_init(&con->out_sent, &con->out_queue);
2852
2853 /* If there are no messages queued or keepalive pending, place
2854 * the connection in a STANDBY state */
2855 if (list_empty(&con->out_queue) &&
2856 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
2857 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2858 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2859 con->state = CON_STATE_STANDBY;
2860 } else {
2861 /* retry after a delay. */
2862 con->state = CON_STATE_PREOPEN;
2863 if (con->delay == 0)
2864 con->delay = BASE_DELAY_INTERVAL;
2865 else if (con->delay < MAX_DELAY_INTERVAL)
2866 con->delay *= 2;
2867 con_flag_set(con, CON_FLAG_BACKOFF);
2868 queue_con(con);
2869 }
2870}
2871
2872
2873
2874/*
2875 * initialize a new messenger instance
2876 */
2877void ceph_messenger_init(struct ceph_messenger *msgr,
2878 struct ceph_entity_addr *myaddr,
2879 u64 supported_features,
2880 u64 required_features,
2881 bool nocrc)
2882{
2883 msgr->supported_features = supported_features;
2884 msgr->required_features = required_features;
2885
2886 spin_lock_init(&msgr->global_seq_lock);
2887
2888 if (myaddr)
2889 msgr->inst.addr = *myaddr;
2890
2891 /* select a random nonce */
2892 msgr->inst.addr.type = 0;
2893 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2894 encode_my_addr(msgr);
2895 msgr->nocrc = nocrc;
2896
2897 atomic_set(&msgr->stopping, 0);
2898
2899 dout("%s %p\n", __func__, msgr);
2900}
2901EXPORT_SYMBOL(ceph_messenger_init);
2902
2903static void clear_standby(struct ceph_connection *con)
2904{
2905 /* come back from STANDBY? */
2906 if (con->state == CON_STATE_STANDBY) {
2907 dout("clear_standby %p and ++connect_seq\n", con);
2908 con->state = CON_STATE_PREOPEN;
2909 con->connect_seq++;
2910 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2911 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
2912 }
2913}
2914
2915/*
2916 * Queue up an outgoing message on the given connection.
2917 */
2918void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2919{
2920 /* set src+dst */
2921 msg->hdr.src = con->msgr->inst.name;
2922 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2923 msg->needs_out_seq = true;
2924
2925 mutex_lock(&con->mutex);
2926
2927 if (con->state == CON_STATE_CLOSED) {
2928 dout("con_send %p closed, dropping %p\n", con, msg);
2929 ceph_msg_put(msg);
2930 mutex_unlock(&con->mutex);
2931 return;
2932 }
2933
2934 BUG_ON(msg->con != NULL);
2935 msg->con = con->ops->get(con);
2936 BUG_ON(msg->con == NULL);
2937
2938 BUG_ON(!list_empty(&msg->list_head));
2939 list_add_tail(&msg->list_head, &con->out_queue);
2940 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2941 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2942 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2943 le32_to_cpu(msg->hdr.front_len),
2944 le32_to_cpu(msg->hdr.middle_len),
2945 le32_to_cpu(msg->hdr.data_len));
2946
2947 clear_standby(con);
2948 mutex_unlock(&con->mutex);
2949
2950 /* if there wasn't anything waiting to send before, queue
2951 * new work */
2952 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
2953 queue_con(con);
2954}
2955EXPORT_SYMBOL(ceph_con_send);
2956
2957/*
2958 * Revoke a message that was previously queued for send
2959 */
2960void ceph_msg_revoke(struct ceph_msg *msg)
2961{
2962 struct ceph_connection *con = msg->con;
2963
2964 if (!con)
2965 return; /* Message not in our possession */
2966
2967 mutex_lock(&con->mutex);
2968 if (!list_empty(&msg->list_head)) {
2969 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
2970 list_del_init(&msg->list_head);
2971 BUG_ON(msg->con == NULL);
2972 msg->con->ops->put(msg->con);
2973 msg->con = NULL;
2974 msg->hdr.seq = 0;
2975
2976 ceph_msg_put(msg);
2977 }
2978 if (con->out_msg == msg) {
2979 dout("%s %p msg %p - was sending\n", __func__, con, msg);
2980 con->out_msg = NULL;
2981 if (con->out_kvec_is_msg) {
2982 con->out_skip = con->out_kvec_bytes;
2983 con->out_kvec_is_msg = false;
2984 }
2985 msg->hdr.seq = 0;
2986
2987 ceph_msg_put(msg);
2988 }
2989 mutex_unlock(&con->mutex);
2990}
2991
2992/*
2993 * Revoke a message that we may be reading data into
2994 */
2995void ceph_msg_revoke_incoming(struct ceph_msg *msg)
2996{
2997 struct ceph_connection *con;
2998
2999 BUG_ON(msg == NULL);
3000 if (!msg->con) {
3001 dout("%s msg %p null con\n", __func__, msg);
3002
3003 return; /* Message not in our possession */
3004 }
3005
3006 con = msg->con;
3007 mutex_lock(&con->mutex);
3008 if (con->in_msg == msg) {
3009 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3010 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3011 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
3012
3013 /* skip rest of message */
3014 dout("%s %p msg %p revoked\n", __func__, con, msg);
3015 con->in_base_pos = con->in_base_pos -
3016 sizeof(struct ceph_msg_header) -
3017 front_len -
3018 middle_len -
3019 data_len -
3020 sizeof(struct ceph_msg_footer);
3021 ceph_msg_put(con->in_msg);
3022 con->in_msg = NULL;
3023 con->in_tag = CEPH_MSGR_TAG_READY;
3024 con->in_seq++;
3025 } else {
3026 dout("%s %p in_msg %p msg %p no-op\n",
3027 __func__, con, con->in_msg, msg);
3028 }
3029 mutex_unlock(&con->mutex);
3030}
3031
3032/*
3033 * Queue a keepalive byte to ensure the tcp connection is alive.
3034 */
3035void ceph_con_keepalive(struct ceph_connection *con)
3036{
3037 dout("con_keepalive %p\n", con);
3038 mutex_lock(&con->mutex);
3039 clear_standby(con);
3040 mutex_unlock(&con->mutex);
3041 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3042 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3043 queue_con(con);
3044}
3045EXPORT_SYMBOL(ceph_con_keepalive);
3046
3047static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
3048{
3049 struct ceph_msg_data *data;
3050
3051 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3052 return NULL;
3053
3054 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
3055 if (data)
3056 data->type = type;
3057 INIT_LIST_HEAD(&data->links);
3058
3059 return data;
3060}
3061
3062static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3063{
3064 if (!data)
3065 return;
3066
3067 WARN_ON(!list_empty(&data->links));
3068 if (data->type == CEPH_MSG_DATA_PAGELIST) {
3069 ceph_pagelist_release(data->pagelist);
3070 kfree(data->pagelist);
3071 }
3072 kmem_cache_free(ceph_msg_data_cache, data);
3073}
3074
3075void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
3076 size_t length, size_t alignment)
3077{
3078 struct ceph_msg_data *data;
3079
3080 BUG_ON(!pages);
3081 BUG_ON(!length);
3082
3083 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3084 BUG_ON(!data);
3085 data->pages = pages;
3086 data->length = length;
3087 data->alignment = alignment & ~PAGE_MASK;
3088
3089 list_add_tail(&data->links, &msg->data);
3090 msg->data_length += length;
3091}
3092EXPORT_SYMBOL(ceph_msg_data_add_pages);
3093
3094void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
3095 struct ceph_pagelist *pagelist)
3096{
3097 struct ceph_msg_data *data;
3098
3099 BUG_ON(!pagelist);
3100 BUG_ON(!pagelist->length);
3101
3102 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3103 BUG_ON(!data);
3104 data->pagelist = pagelist;
3105
3106 list_add_tail(&data->links, &msg->data);
3107 msg->data_length += pagelist->length;
3108}
3109EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
3110
3111#ifdef CONFIG_BLOCK
3112void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
3113 size_t length)
3114{
3115 struct ceph_msg_data *data;
3116
3117 BUG_ON(!bio);
3118
3119 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3120 BUG_ON(!data);
3121 data->bio = bio;
3122 data->bio_length = length;
3123
3124 list_add_tail(&data->links, &msg->data);
3125 msg->data_length += length;
3126}
3127EXPORT_SYMBOL(ceph_msg_data_add_bio);
3128#endif /* CONFIG_BLOCK */
3129
3130/*
3131 * construct a new message with given type, size
3132 * the new msg has a ref count of 1.
3133 */
3134struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3135 bool can_fail)
3136{
3137 struct ceph_msg *m;
3138
3139 m = kmem_cache_zalloc(ceph_msg_cache, flags);
3140 if (m == NULL)
3141 goto out;
3142
3143 m->hdr.type = cpu_to_le16(type);
3144 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3145 m->hdr.front_len = cpu_to_le32(front_len);
3146
3147 INIT_LIST_HEAD(&m->list_head);
3148 kref_init(&m->kref);
3149 INIT_LIST_HEAD(&m->data);
3150
3151 /* front */
3152 if (front_len) {
3153 m->front.iov_base = ceph_kvmalloc(front_len, flags);
3154 if (m->front.iov_base == NULL) {
3155 dout("ceph_msg_new can't allocate %d bytes\n",
3156 front_len);
3157 goto out2;
3158 }
3159 } else {
3160 m->front.iov_base = NULL;
3161 }
3162 m->front_alloc_len = m->front.iov_len = front_len;
3163
3164 dout("ceph_msg_new %p front %d\n", m, front_len);
3165 return m;
3166
3167out2:
3168 ceph_msg_put(m);
3169out:
3170 if (!can_fail) {
3171 pr_err("msg_new can't create type %d front %d\n", type,
3172 front_len);
3173 WARN_ON(1);
3174 } else {
3175 dout("msg_new can't create type %d front %d\n", type,
3176 front_len);
3177 }
3178 return NULL;
3179}
3180EXPORT_SYMBOL(ceph_msg_new);
3181
3182/*
3183 * Allocate "middle" portion of a message, if it is needed and wasn't
3184 * allocated by alloc_msg. This allows us to read a small fixed-size
3185 * per-type header in the front and then gracefully fail (i.e.,
3186 * propagate the error to the caller based on info in the front) when
3187 * the middle is too large.
3188 */
3189static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3190{
3191 int type = le16_to_cpu(msg->hdr.type);
3192 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3193
3194 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3195 ceph_msg_type_name(type), middle_len);
3196 BUG_ON(!middle_len);
3197 BUG_ON(msg->middle);
3198
3199 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3200 if (!msg->middle)
3201 return -ENOMEM;
3202 return 0;
3203}
3204
3205/*
3206 * Allocate a message for receiving an incoming message on a
3207 * connection, and save the result in con->in_msg. Uses the
3208 * connection's private alloc_msg op if available.
3209 *
3210 * Returns 0 on success, or a negative error code.
3211 *
3212 * On success, if we set *skip = 1:
3213 * - the next message should be skipped and ignored.
3214 * - con->in_msg == NULL
3215 * or if we set *skip = 0:
3216 * - con->in_msg is non-null.
3217 * On error (ENOMEM, EAGAIN, ...),
3218 * - con->in_msg == NULL
3219 */
3220static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3221{
3222 struct ceph_msg_header *hdr = &con->in_hdr;
3223 int middle_len = le32_to_cpu(hdr->middle_len);
3224 struct ceph_msg *msg;
3225 int ret = 0;
3226
3227 BUG_ON(con->in_msg != NULL);
3228 BUG_ON(!con->ops->alloc_msg);
3229
3230 mutex_unlock(&con->mutex);
3231 msg = con->ops->alloc_msg(con, hdr, skip);
3232 mutex_lock(&con->mutex);
3233 if (con->state != CON_STATE_OPEN) {
3234 if (msg)
3235 ceph_msg_put(msg);
3236 return -EAGAIN;
3237 }
3238 if (msg) {
3239 BUG_ON(*skip);
3240 con->in_msg = msg;
3241 con->in_msg->con = con->ops->get(con);
3242 BUG_ON(con->in_msg->con == NULL);
3243 } else {
3244 /*
3245 * Null message pointer means either we should skip
3246 * this message or we couldn't allocate memory. The
3247 * former is not an error.
3248 */
3249 if (*skip)
3250 return 0;
3251 con->error_msg = "error allocating memory for incoming message";
3252
3253 return -ENOMEM;
3254 }
3255 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3256
3257 if (middle_len && !con->in_msg->middle) {
3258 ret = ceph_alloc_middle(con, con->in_msg);
3259 if (ret < 0) {
3260 ceph_msg_put(con->in_msg);
3261 con->in_msg = NULL;
3262 }
3263 }
3264
3265 return ret;
3266}
3267
3268
3269/*
3270 * Free a generically kmalloc'd message.
3271 */
3272void ceph_msg_kfree(struct ceph_msg *m)
3273{
3274 dout("msg_kfree %p\n", m);
3275 ceph_kvfree(m->front.iov_base);
3276 kmem_cache_free(ceph_msg_cache, m);
3277}
3278
3279/*
3280 * Drop a msg ref. Destroy as needed.
3281 */
3282void ceph_msg_last_put(struct kref *kref)
3283{
3284 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3285 LIST_HEAD(data);
3286 struct list_head *links;
3287 struct list_head *next;
3288
3289 dout("ceph_msg_put last one on %p\n", m);
3290 WARN_ON(!list_empty(&m->list_head));
3291
3292 /* drop middle, data, if any */
3293 if (m->middle) {
3294 ceph_buffer_put(m->middle);
3295 m->middle = NULL;
3296 }
3297
3298 list_splice_init(&m->data, &data);
3299 list_for_each_safe(links, next, &data) {
3300 struct ceph_msg_data *data;
3301
3302 data = list_entry(links, struct ceph_msg_data, links);
3303 list_del_init(links);
3304 ceph_msg_data_destroy(data);
3305 }
3306 m->data_length = 0;
3307
3308 if (m->pool)
3309 ceph_msgpool_put(m->pool, m);
3310 else
3311 ceph_msg_kfree(m);
3312}
3313EXPORT_SYMBOL(ceph_msg_last_put);
3314
3315void ceph_msg_dump(struct ceph_msg *msg)
3316{
3317 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3318 msg->front_alloc_len, msg->data_length);
3319 print_hex_dump(KERN_DEBUG, "header: ",
3320 DUMP_PREFIX_OFFSET, 16, 1,
3321 &msg->hdr, sizeof(msg->hdr), true);
3322 print_hex_dump(KERN_DEBUG, " front: ",
3323 DUMP_PREFIX_OFFSET, 16, 1,
3324 msg->front.iov_base, msg->front.iov_len, true);
3325 if (msg->middle)
3326 print_hex_dump(KERN_DEBUG, "middle: ",
3327 DUMP_PREFIX_OFFSET, 16, 1,
3328 msg->middle->vec.iov_base,
3329 msg->middle->vec.iov_len, true);
3330 print_hex_dump(KERN_DEBUG, "footer: ",
3331 DUMP_PREFIX_OFFSET, 16, 1,
3332 &msg->footer, sizeof(msg->footer), true);
3333}
3334EXPORT_SYMBOL(ceph_msg_dump);
1// SPDX-License-Identifier: GPL-2.0
2#include <linux/ceph/ceph_debug.h>
3
4#include <linux/crc32c.h>
5#include <linux/ctype.h>
6#include <linux/highmem.h>
7#include <linux/inet.h>
8#include <linux/kthread.h>
9#include <linux/net.h>
10#include <linux/nsproxy.h>
11#include <linux/sched/mm.h>
12#include <linux/slab.h>
13#include <linux/socket.h>
14#include <linux/string.h>
15#ifdef CONFIG_BLOCK
16#include <linux/bio.h>
17#endif /* CONFIG_BLOCK */
18#include <linux/dns_resolver.h>
19#include <net/tcp.h>
20#include <trace/events/sock.h>
21
22#include <linux/ceph/ceph_features.h>
23#include <linux/ceph/libceph.h>
24#include <linux/ceph/messenger.h>
25#include <linux/ceph/decode.h>
26#include <linux/ceph/pagelist.h>
27#include <linux/export.h>
28
29/*
30 * Ceph uses the messenger to exchange ceph_msg messages with other
31 * hosts in the system. The messenger provides ordered and reliable
32 * delivery. We tolerate TCP disconnects by reconnecting (with
33 * exponential backoff) in the case of a fault (disconnection, bad
34 * crc, protocol error). Acks allow sent messages to be discarded by
35 * the sender.
36 */
37
38/*
39 * We track the state of the socket on a given connection using
40 * values defined below. The transition to a new socket state is
41 * handled by a function which verifies we aren't coming from an
42 * unexpected state.
43 *
44 * --------
45 * | NEW* | transient initial state
46 * --------
47 * | con_sock_state_init()
48 * v
49 * ----------
50 * | CLOSED | initialized, but no socket (and no
51 * ---------- TCP connection)
52 * ^ \
53 * | \ con_sock_state_connecting()
54 * | ----------------------
55 * | \
56 * + con_sock_state_closed() \
57 * |+--------------------------- \
58 * | \ \ \
59 * | ----------- \ \
60 * | | CLOSING | socket event; \ \
61 * | ----------- await close \ \
62 * | ^ \ |
63 * | | \ |
64 * | + con_sock_state_closing() \ |
65 * | / \ | |
66 * | / --------------- | |
67 * | / \ v v
68 * | / --------------
69 * | / -----------------| CONNECTING | socket created, TCP
70 * | | / -------------- connect initiated
71 * | | | con_sock_state_connected()
72 * | | v
73 * -------------
74 * | CONNECTED | TCP connection established
75 * -------------
76 *
77 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
78 */
79
80#define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
81#define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
82#define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
83#define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
84#define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
85
86static bool con_flag_valid(unsigned long con_flag)
87{
88 switch (con_flag) {
89 case CEPH_CON_F_LOSSYTX:
90 case CEPH_CON_F_KEEPALIVE_PENDING:
91 case CEPH_CON_F_WRITE_PENDING:
92 case CEPH_CON_F_SOCK_CLOSED:
93 case CEPH_CON_F_BACKOFF:
94 return true;
95 default:
96 return false;
97 }
98}
99
100void ceph_con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
101{
102 BUG_ON(!con_flag_valid(con_flag));
103
104 clear_bit(con_flag, &con->flags);
105}
106
107void ceph_con_flag_set(struct ceph_connection *con, unsigned long con_flag)
108{
109 BUG_ON(!con_flag_valid(con_flag));
110
111 set_bit(con_flag, &con->flags);
112}
113
114bool ceph_con_flag_test(struct ceph_connection *con, unsigned long con_flag)
115{
116 BUG_ON(!con_flag_valid(con_flag));
117
118 return test_bit(con_flag, &con->flags);
119}
120
121bool ceph_con_flag_test_and_clear(struct ceph_connection *con,
122 unsigned long con_flag)
123{
124 BUG_ON(!con_flag_valid(con_flag));
125
126 return test_and_clear_bit(con_flag, &con->flags);
127}
128
129bool ceph_con_flag_test_and_set(struct ceph_connection *con,
130 unsigned long con_flag)
131{
132 BUG_ON(!con_flag_valid(con_flag));
133
134 return test_and_set_bit(con_flag, &con->flags);
135}
136
137/* Slab caches for frequently-allocated structures */
138
139static struct kmem_cache *ceph_msg_cache;
140
141#ifdef CONFIG_LOCKDEP
142static struct lock_class_key socket_class;
143#endif
144
145static void queue_con(struct ceph_connection *con);
146static void cancel_con(struct ceph_connection *con);
147static void ceph_con_workfn(struct work_struct *);
148static void con_fault(struct ceph_connection *con);
149
150/*
151 * Nicely render a sockaddr as a string. An array of formatted
152 * strings is used, to approximate reentrancy.
153 */
154#define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
155#define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
156#define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
157#define MAX_ADDR_STR_LEN 64 /* 54 is enough */
158
159static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
160static atomic_t addr_str_seq = ATOMIC_INIT(0);
161
162struct page *ceph_zero_page; /* used in certain error cases */
163
164const char *ceph_pr_addr(const struct ceph_entity_addr *addr)
165{
166 int i;
167 char *s;
168 struct sockaddr_storage ss = addr->in_addr; /* align */
169 struct sockaddr_in *in4 = (struct sockaddr_in *)&ss;
170 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *)&ss;
171
172 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
173 s = addr_str[i];
174
175 switch (ss.ss_family) {
176 case AF_INET:
177 snprintf(s, MAX_ADDR_STR_LEN, "(%d)%pI4:%hu",
178 le32_to_cpu(addr->type), &in4->sin_addr,
179 ntohs(in4->sin_port));
180 break;
181
182 case AF_INET6:
183 snprintf(s, MAX_ADDR_STR_LEN, "(%d)[%pI6c]:%hu",
184 le32_to_cpu(addr->type), &in6->sin6_addr,
185 ntohs(in6->sin6_port));
186 break;
187
188 default:
189 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
190 ss.ss_family);
191 }
192
193 return s;
194}
195EXPORT_SYMBOL(ceph_pr_addr);
196
197void ceph_encode_my_addr(struct ceph_messenger *msgr)
198{
199 if (!ceph_msgr2(from_msgr(msgr))) {
200 memcpy(&msgr->my_enc_addr, &msgr->inst.addr,
201 sizeof(msgr->my_enc_addr));
202 ceph_encode_banner_addr(&msgr->my_enc_addr);
203 }
204}
205
206/*
207 * work queue for all reading and writing to/from the socket.
208 */
209static struct workqueue_struct *ceph_msgr_wq;
210
211static int ceph_msgr_slab_init(void)
212{
213 BUG_ON(ceph_msg_cache);
214 ceph_msg_cache = KMEM_CACHE(ceph_msg, 0);
215 if (!ceph_msg_cache)
216 return -ENOMEM;
217
218 return 0;
219}
220
221static void ceph_msgr_slab_exit(void)
222{
223 BUG_ON(!ceph_msg_cache);
224 kmem_cache_destroy(ceph_msg_cache);
225 ceph_msg_cache = NULL;
226}
227
228static void _ceph_msgr_exit(void)
229{
230 if (ceph_msgr_wq) {
231 destroy_workqueue(ceph_msgr_wq);
232 ceph_msgr_wq = NULL;
233 }
234
235 BUG_ON(!ceph_zero_page);
236 put_page(ceph_zero_page);
237 ceph_zero_page = NULL;
238
239 ceph_msgr_slab_exit();
240}
241
242int __init ceph_msgr_init(void)
243{
244 if (ceph_msgr_slab_init())
245 return -ENOMEM;
246
247 BUG_ON(ceph_zero_page);
248 ceph_zero_page = ZERO_PAGE(0);
249 get_page(ceph_zero_page);
250
251 /*
252 * The number of active work items is limited by the number of
253 * connections, so leave @max_active at default.
254 */
255 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
256 if (ceph_msgr_wq)
257 return 0;
258
259 pr_err("msgr_init failed to create workqueue\n");
260 _ceph_msgr_exit();
261
262 return -ENOMEM;
263}
264
265void ceph_msgr_exit(void)
266{
267 BUG_ON(ceph_msgr_wq == NULL);
268
269 _ceph_msgr_exit();
270}
271
272void ceph_msgr_flush(void)
273{
274 flush_workqueue(ceph_msgr_wq);
275}
276EXPORT_SYMBOL(ceph_msgr_flush);
277
278/* Connection socket state transition functions */
279
280static void con_sock_state_init(struct ceph_connection *con)
281{
282 int old_state;
283
284 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
285 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
286 printk("%s: unexpected old state %d\n", __func__, old_state);
287 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
288 CON_SOCK_STATE_CLOSED);
289}
290
291static void con_sock_state_connecting(struct ceph_connection *con)
292{
293 int old_state;
294
295 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
296 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
297 printk("%s: unexpected old state %d\n", __func__, old_state);
298 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
299 CON_SOCK_STATE_CONNECTING);
300}
301
302static void con_sock_state_connected(struct ceph_connection *con)
303{
304 int old_state;
305
306 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
307 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
308 printk("%s: unexpected old state %d\n", __func__, old_state);
309 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
310 CON_SOCK_STATE_CONNECTED);
311}
312
313static void con_sock_state_closing(struct ceph_connection *con)
314{
315 int old_state;
316
317 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
318 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
319 old_state != CON_SOCK_STATE_CONNECTED &&
320 old_state != CON_SOCK_STATE_CLOSING))
321 printk("%s: unexpected old state %d\n", __func__, old_state);
322 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
323 CON_SOCK_STATE_CLOSING);
324}
325
326static void con_sock_state_closed(struct ceph_connection *con)
327{
328 int old_state;
329
330 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
331 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
332 old_state != CON_SOCK_STATE_CLOSING &&
333 old_state != CON_SOCK_STATE_CONNECTING &&
334 old_state != CON_SOCK_STATE_CLOSED))
335 printk("%s: unexpected old state %d\n", __func__, old_state);
336 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
337 CON_SOCK_STATE_CLOSED);
338}
339
340/*
341 * socket callback functions
342 */
343
344/* data available on socket, or listen socket received a connect */
345static void ceph_sock_data_ready(struct sock *sk)
346{
347 struct ceph_connection *con = sk->sk_user_data;
348
349 trace_sk_data_ready(sk);
350
351 if (atomic_read(&con->msgr->stopping)) {
352 return;
353 }
354
355 if (sk->sk_state != TCP_CLOSE_WAIT) {
356 dout("%s %p state = %d, queueing work\n", __func__,
357 con, con->state);
358 queue_con(con);
359 }
360}
361
362/* socket has buffer space for writing */
363static void ceph_sock_write_space(struct sock *sk)
364{
365 struct ceph_connection *con = sk->sk_user_data;
366
367 /* only queue to workqueue if there is data we want to write,
368 * and there is sufficient space in the socket buffer to accept
369 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
370 * doesn't get called again until try_write() fills the socket
371 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
372 * and net/core/stream.c:sk_stream_write_space().
373 */
374 if (ceph_con_flag_test(con, CEPH_CON_F_WRITE_PENDING)) {
375 if (sk_stream_is_writeable(sk)) {
376 dout("%s %p queueing write work\n", __func__, con);
377 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
378 queue_con(con);
379 }
380 } else {
381 dout("%s %p nothing to write\n", __func__, con);
382 }
383}
384
385/* socket's state has changed */
386static void ceph_sock_state_change(struct sock *sk)
387{
388 struct ceph_connection *con = sk->sk_user_data;
389
390 dout("%s %p state = %d sk_state = %u\n", __func__,
391 con, con->state, sk->sk_state);
392
393 switch (sk->sk_state) {
394 case TCP_CLOSE:
395 dout("%s TCP_CLOSE\n", __func__);
396 fallthrough;
397 case TCP_CLOSE_WAIT:
398 dout("%s TCP_CLOSE_WAIT\n", __func__);
399 con_sock_state_closing(con);
400 ceph_con_flag_set(con, CEPH_CON_F_SOCK_CLOSED);
401 queue_con(con);
402 break;
403 case TCP_ESTABLISHED:
404 dout("%s TCP_ESTABLISHED\n", __func__);
405 con_sock_state_connected(con);
406 queue_con(con);
407 break;
408 default: /* Everything else is uninteresting */
409 break;
410 }
411}
412
413/*
414 * set up socket callbacks
415 */
416static void set_sock_callbacks(struct socket *sock,
417 struct ceph_connection *con)
418{
419 struct sock *sk = sock->sk;
420 sk->sk_user_data = con;
421 sk->sk_data_ready = ceph_sock_data_ready;
422 sk->sk_write_space = ceph_sock_write_space;
423 sk->sk_state_change = ceph_sock_state_change;
424}
425
426
427/*
428 * socket helpers
429 */
430
431/*
432 * initiate connection to a remote socket.
433 */
434int ceph_tcp_connect(struct ceph_connection *con)
435{
436 struct sockaddr_storage ss = con->peer_addr.in_addr; /* align */
437 struct socket *sock;
438 unsigned int noio_flag;
439 int ret;
440
441 dout("%s con %p peer_addr %s\n", __func__, con,
442 ceph_pr_addr(&con->peer_addr));
443 BUG_ON(con->sock);
444
445 /* sock_create_kern() allocates with GFP_KERNEL */
446 noio_flag = memalloc_noio_save();
447 ret = sock_create_kern(read_pnet(&con->msgr->net), ss.ss_family,
448 SOCK_STREAM, IPPROTO_TCP, &sock);
449 memalloc_noio_restore(noio_flag);
450 if (ret)
451 return ret;
452 sock->sk->sk_allocation = GFP_NOFS;
453 sock->sk->sk_use_task_frag = false;
454
455#ifdef CONFIG_LOCKDEP
456 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
457#endif
458
459 set_sock_callbacks(sock, con);
460
461 con_sock_state_connecting(con);
462 ret = kernel_connect(sock, (struct sockaddr *)&ss, sizeof(ss),
463 O_NONBLOCK);
464 if (ret == -EINPROGRESS) {
465 dout("connect %s EINPROGRESS sk_state = %u\n",
466 ceph_pr_addr(&con->peer_addr),
467 sock->sk->sk_state);
468 } else if (ret < 0) {
469 pr_err("connect %s error %d\n",
470 ceph_pr_addr(&con->peer_addr), ret);
471 sock_release(sock);
472 return ret;
473 }
474
475 if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY))
476 tcp_sock_set_nodelay(sock->sk);
477
478 con->sock = sock;
479 return 0;
480}
481
482/*
483 * Shutdown/close the socket for the given connection.
484 */
485int ceph_con_close_socket(struct ceph_connection *con)
486{
487 int rc = 0;
488
489 dout("%s con %p sock %p\n", __func__, con, con->sock);
490 if (con->sock) {
491 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
492 sock_release(con->sock);
493 con->sock = NULL;
494 }
495
496 /*
497 * Forcibly clear the SOCK_CLOSED flag. It gets set
498 * independent of the connection mutex, and we could have
499 * received a socket close event before we had the chance to
500 * shut the socket down.
501 */
502 ceph_con_flag_clear(con, CEPH_CON_F_SOCK_CLOSED);
503
504 con_sock_state_closed(con);
505 return rc;
506}
507
508static void ceph_con_reset_protocol(struct ceph_connection *con)
509{
510 dout("%s con %p\n", __func__, con);
511
512 ceph_con_close_socket(con);
513 if (con->in_msg) {
514 WARN_ON(con->in_msg->con != con);
515 ceph_msg_put(con->in_msg);
516 con->in_msg = NULL;
517 }
518 if (con->out_msg) {
519 WARN_ON(con->out_msg->con != con);
520 ceph_msg_put(con->out_msg);
521 con->out_msg = NULL;
522 }
523 if (con->bounce_page) {
524 __free_page(con->bounce_page);
525 con->bounce_page = NULL;
526 }
527
528 if (ceph_msgr2(from_msgr(con->msgr)))
529 ceph_con_v2_reset_protocol(con);
530 else
531 ceph_con_v1_reset_protocol(con);
532}
533
534/*
535 * Reset a connection. Discard all incoming and outgoing messages
536 * and clear *_seq state.
537 */
538static void ceph_msg_remove(struct ceph_msg *msg)
539{
540 list_del_init(&msg->list_head);
541
542 ceph_msg_put(msg);
543}
544
545static void ceph_msg_remove_list(struct list_head *head)
546{
547 while (!list_empty(head)) {
548 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
549 list_head);
550 ceph_msg_remove(msg);
551 }
552}
553
554void ceph_con_reset_session(struct ceph_connection *con)
555{
556 dout("%s con %p\n", __func__, con);
557
558 WARN_ON(con->in_msg);
559 WARN_ON(con->out_msg);
560 ceph_msg_remove_list(&con->out_queue);
561 ceph_msg_remove_list(&con->out_sent);
562 con->out_seq = 0;
563 con->in_seq = 0;
564 con->in_seq_acked = 0;
565
566 if (ceph_msgr2(from_msgr(con->msgr)))
567 ceph_con_v2_reset_session(con);
568 else
569 ceph_con_v1_reset_session(con);
570}
571
572/*
573 * mark a peer down. drop any open connections.
574 */
575void ceph_con_close(struct ceph_connection *con)
576{
577 mutex_lock(&con->mutex);
578 dout("con_close %p peer %s\n", con, ceph_pr_addr(&con->peer_addr));
579 con->state = CEPH_CON_S_CLOSED;
580
581 ceph_con_flag_clear(con, CEPH_CON_F_LOSSYTX); /* so we retry next
582 connect */
583 ceph_con_flag_clear(con, CEPH_CON_F_KEEPALIVE_PENDING);
584 ceph_con_flag_clear(con, CEPH_CON_F_WRITE_PENDING);
585 ceph_con_flag_clear(con, CEPH_CON_F_BACKOFF);
586
587 ceph_con_reset_protocol(con);
588 ceph_con_reset_session(con);
589 cancel_con(con);
590 mutex_unlock(&con->mutex);
591}
592EXPORT_SYMBOL(ceph_con_close);
593
594/*
595 * Reopen a closed connection, with a new peer address.
596 */
597void ceph_con_open(struct ceph_connection *con,
598 __u8 entity_type, __u64 entity_num,
599 struct ceph_entity_addr *addr)
600{
601 mutex_lock(&con->mutex);
602 dout("con_open %p %s\n", con, ceph_pr_addr(addr));
603
604 WARN_ON(con->state != CEPH_CON_S_CLOSED);
605 con->state = CEPH_CON_S_PREOPEN;
606
607 con->peer_name.type = (__u8) entity_type;
608 con->peer_name.num = cpu_to_le64(entity_num);
609
610 memcpy(&con->peer_addr, addr, sizeof(*addr));
611 con->delay = 0; /* reset backoff memory */
612 mutex_unlock(&con->mutex);
613 queue_con(con);
614}
615EXPORT_SYMBOL(ceph_con_open);
616
617/*
618 * return true if this connection ever successfully opened
619 */
620bool ceph_con_opened(struct ceph_connection *con)
621{
622 if (ceph_msgr2(from_msgr(con->msgr)))
623 return ceph_con_v2_opened(con);
624
625 return ceph_con_v1_opened(con);
626}
627
628/*
629 * initialize a new connection.
630 */
631void ceph_con_init(struct ceph_connection *con, void *private,
632 const struct ceph_connection_operations *ops,
633 struct ceph_messenger *msgr)
634{
635 dout("con_init %p\n", con);
636 memset(con, 0, sizeof(*con));
637 con->private = private;
638 con->ops = ops;
639 con->msgr = msgr;
640
641 con_sock_state_init(con);
642
643 mutex_init(&con->mutex);
644 INIT_LIST_HEAD(&con->out_queue);
645 INIT_LIST_HEAD(&con->out_sent);
646 INIT_DELAYED_WORK(&con->work, ceph_con_workfn);
647
648 con->state = CEPH_CON_S_CLOSED;
649}
650EXPORT_SYMBOL(ceph_con_init);
651
652/*
653 * We maintain a global counter to order connection attempts. Get
654 * a unique seq greater than @gt.
655 */
656u32 ceph_get_global_seq(struct ceph_messenger *msgr, u32 gt)
657{
658 u32 ret;
659
660 spin_lock(&msgr->global_seq_lock);
661 if (msgr->global_seq < gt)
662 msgr->global_seq = gt;
663 ret = ++msgr->global_seq;
664 spin_unlock(&msgr->global_seq_lock);
665 return ret;
666}
667
668/*
669 * Discard messages that have been acked by the server.
670 */
671void ceph_con_discard_sent(struct ceph_connection *con, u64 ack_seq)
672{
673 struct ceph_msg *msg;
674 u64 seq;
675
676 dout("%s con %p ack_seq %llu\n", __func__, con, ack_seq);
677 while (!list_empty(&con->out_sent)) {
678 msg = list_first_entry(&con->out_sent, struct ceph_msg,
679 list_head);
680 WARN_ON(msg->needs_out_seq);
681 seq = le64_to_cpu(msg->hdr.seq);
682 if (seq > ack_seq)
683 break;
684
685 dout("%s con %p discarding msg %p seq %llu\n", __func__, con,
686 msg, seq);
687 ceph_msg_remove(msg);
688 }
689}
690
691/*
692 * Discard messages that have been requeued in con_fault(), up to
693 * reconnect_seq. This avoids gratuitously resending messages that
694 * the server had received and handled prior to reconnect.
695 */
696void ceph_con_discard_requeued(struct ceph_connection *con, u64 reconnect_seq)
697{
698 struct ceph_msg *msg;
699 u64 seq;
700
701 dout("%s con %p reconnect_seq %llu\n", __func__, con, reconnect_seq);
702 while (!list_empty(&con->out_queue)) {
703 msg = list_first_entry(&con->out_queue, struct ceph_msg,
704 list_head);
705 if (msg->needs_out_seq)
706 break;
707 seq = le64_to_cpu(msg->hdr.seq);
708 if (seq > reconnect_seq)
709 break;
710
711 dout("%s con %p discarding msg %p seq %llu\n", __func__, con,
712 msg, seq);
713 ceph_msg_remove(msg);
714 }
715}
716
717#ifdef CONFIG_BLOCK
718
719/*
720 * For a bio data item, a piece is whatever remains of the next
721 * entry in the current bio iovec, or the first entry in the next
722 * bio in the list.
723 */
724static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
725 size_t length)
726{
727 struct ceph_msg_data *data = cursor->data;
728 struct ceph_bio_iter *it = &cursor->bio_iter;
729
730 cursor->resid = min_t(size_t, length, data->bio_length);
731 *it = data->bio_pos;
732 if (cursor->resid < it->iter.bi_size)
733 it->iter.bi_size = cursor->resid;
734
735 BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
736}
737
738static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
739 size_t *page_offset,
740 size_t *length)
741{
742 struct bio_vec bv = bio_iter_iovec(cursor->bio_iter.bio,
743 cursor->bio_iter.iter);
744
745 *page_offset = bv.bv_offset;
746 *length = bv.bv_len;
747 return bv.bv_page;
748}
749
750static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
751 size_t bytes)
752{
753 struct ceph_bio_iter *it = &cursor->bio_iter;
754 struct page *page = bio_iter_page(it->bio, it->iter);
755
756 BUG_ON(bytes > cursor->resid);
757 BUG_ON(bytes > bio_iter_len(it->bio, it->iter));
758 cursor->resid -= bytes;
759 bio_advance_iter(it->bio, &it->iter, bytes);
760
761 if (!cursor->resid)
762 return false; /* no more data */
763
764 if (!bytes || (it->iter.bi_size && it->iter.bi_bvec_done &&
765 page == bio_iter_page(it->bio, it->iter)))
766 return false; /* more bytes to process in this segment */
767
768 if (!it->iter.bi_size) {
769 it->bio = it->bio->bi_next;
770 it->iter = it->bio->bi_iter;
771 if (cursor->resid < it->iter.bi_size)
772 it->iter.bi_size = cursor->resid;
773 }
774
775 BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
776 return true;
777}
778#endif /* CONFIG_BLOCK */
779
780static void ceph_msg_data_bvecs_cursor_init(struct ceph_msg_data_cursor *cursor,
781 size_t length)
782{
783 struct ceph_msg_data *data = cursor->data;
784 struct bio_vec *bvecs = data->bvec_pos.bvecs;
785
786 cursor->resid = min_t(size_t, length, data->bvec_pos.iter.bi_size);
787 cursor->bvec_iter = data->bvec_pos.iter;
788 cursor->bvec_iter.bi_size = cursor->resid;
789
790 BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
791}
792
793static struct page *ceph_msg_data_bvecs_next(struct ceph_msg_data_cursor *cursor,
794 size_t *page_offset,
795 size_t *length)
796{
797 struct bio_vec bv = bvec_iter_bvec(cursor->data->bvec_pos.bvecs,
798 cursor->bvec_iter);
799
800 *page_offset = bv.bv_offset;
801 *length = bv.bv_len;
802 return bv.bv_page;
803}
804
805static bool ceph_msg_data_bvecs_advance(struct ceph_msg_data_cursor *cursor,
806 size_t bytes)
807{
808 struct bio_vec *bvecs = cursor->data->bvec_pos.bvecs;
809 struct page *page = bvec_iter_page(bvecs, cursor->bvec_iter);
810
811 BUG_ON(bytes > cursor->resid);
812 BUG_ON(bytes > bvec_iter_len(bvecs, cursor->bvec_iter));
813 cursor->resid -= bytes;
814 bvec_iter_advance(bvecs, &cursor->bvec_iter, bytes);
815
816 if (!cursor->resid)
817 return false; /* no more data */
818
819 if (!bytes || (cursor->bvec_iter.bi_bvec_done &&
820 page == bvec_iter_page(bvecs, cursor->bvec_iter)))
821 return false; /* more bytes to process in this segment */
822
823 BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
824 return true;
825}
826
827/*
828 * For a page array, a piece comes from the first page in the array
829 * that has not already been fully consumed.
830 */
831static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
832 size_t length)
833{
834 struct ceph_msg_data *data = cursor->data;
835 int page_count;
836
837 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
838
839 BUG_ON(!data->pages);
840 BUG_ON(!data->length);
841
842 cursor->resid = min(length, data->length);
843 page_count = calc_pages_for(data->alignment, (u64)data->length);
844 cursor->page_offset = data->alignment & ~PAGE_MASK;
845 cursor->page_index = 0;
846 BUG_ON(page_count > (int)USHRT_MAX);
847 cursor->page_count = (unsigned short)page_count;
848 BUG_ON(length > SIZE_MAX - cursor->page_offset);
849}
850
851static struct page *
852ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
853 size_t *page_offset, size_t *length)
854{
855 struct ceph_msg_data *data = cursor->data;
856
857 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
858
859 BUG_ON(cursor->page_index >= cursor->page_count);
860 BUG_ON(cursor->page_offset >= PAGE_SIZE);
861
862 *page_offset = cursor->page_offset;
863 *length = min_t(size_t, cursor->resid, PAGE_SIZE - *page_offset);
864 return data->pages[cursor->page_index];
865}
866
867static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
868 size_t bytes)
869{
870 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
871
872 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
873
874 /* Advance the cursor page offset */
875
876 cursor->resid -= bytes;
877 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
878 if (!bytes || cursor->page_offset)
879 return false; /* more bytes to process in the current page */
880
881 if (!cursor->resid)
882 return false; /* no more data */
883
884 /* Move on to the next page; offset is already at 0 */
885
886 BUG_ON(cursor->page_index >= cursor->page_count);
887 cursor->page_index++;
888 return true;
889}
890
891/*
892 * For a pagelist, a piece is whatever remains to be consumed in the
893 * first page in the list, or the front of the next page.
894 */
895static void
896ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
897 size_t length)
898{
899 struct ceph_msg_data *data = cursor->data;
900 struct ceph_pagelist *pagelist;
901 struct page *page;
902
903 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
904
905 pagelist = data->pagelist;
906 BUG_ON(!pagelist);
907
908 if (!length)
909 return; /* pagelist can be assigned but empty */
910
911 BUG_ON(list_empty(&pagelist->head));
912 page = list_first_entry(&pagelist->head, struct page, lru);
913
914 cursor->resid = min(length, pagelist->length);
915 cursor->page = page;
916 cursor->offset = 0;
917}
918
919static struct page *
920ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
921 size_t *page_offset, size_t *length)
922{
923 struct ceph_msg_data *data = cursor->data;
924 struct ceph_pagelist *pagelist;
925
926 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
927
928 pagelist = data->pagelist;
929 BUG_ON(!pagelist);
930
931 BUG_ON(!cursor->page);
932 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
933
934 /* offset of first page in pagelist is always 0 */
935 *page_offset = cursor->offset & ~PAGE_MASK;
936 *length = min_t(size_t, cursor->resid, PAGE_SIZE - *page_offset);
937 return cursor->page;
938}
939
940static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
941 size_t bytes)
942{
943 struct ceph_msg_data *data = cursor->data;
944 struct ceph_pagelist *pagelist;
945
946 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
947
948 pagelist = data->pagelist;
949 BUG_ON(!pagelist);
950
951 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
952 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
953
954 /* Advance the cursor offset */
955
956 cursor->resid -= bytes;
957 cursor->offset += bytes;
958 /* offset of first page in pagelist is always 0 */
959 if (!bytes || cursor->offset & ~PAGE_MASK)
960 return false; /* more bytes to process in the current page */
961
962 if (!cursor->resid)
963 return false; /* no more data */
964
965 /* Move on to the next page */
966
967 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
968 cursor->page = list_next_entry(cursor->page, lru);
969 return true;
970}
971
972static void ceph_msg_data_iter_cursor_init(struct ceph_msg_data_cursor *cursor,
973 size_t length)
974{
975 struct ceph_msg_data *data = cursor->data;
976
977 cursor->iov_iter = data->iter;
978 cursor->lastlen = 0;
979 iov_iter_truncate(&cursor->iov_iter, length);
980 cursor->resid = iov_iter_count(&cursor->iov_iter);
981}
982
983static struct page *ceph_msg_data_iter_next(struct ceph_msg_data_cursor *cursor,
984 size_t *page_offset, size_t *length)
985{
986 struct page *page;
987 ssize_t len;
988
989 if (cursor->lastlen)
990 iov_iter_revert(&cursor->iov_iter, cursor->lastlen);
991
992 len = iov_iter_get_pages2(&cursor->iov_iter, &page, PAGE_SIZE,
993 1, page_offset);
994 BUG_ON(len < 0);
995
996 cursor->lastlen = len;
997
998 /*
999 * FIXME: The assumption is that the pages represented by the iov_iter
1000 * are pinned, with the references held by the upper-level
1001 * callers, or by virtue of being under writeback. Eventually,
1002 * we'll get an iov_iter_get_pages2 variant that doesn't take
1003 * page refs. Until then, just put the page ref.
1004 */
1005 VM_BUG_ON_PAGE(!PageWriteback(page) && page_count(page) < 2, page);
1006 put_page(page);
1007
1008 *length = min_t(size_t, len, cursor->resid);
1009 return page;
1010}
1011
1012static bool ceph_msg_data_iter_advance(struct ceph_msg_data_cursor *cursor,
1013 size_t bytes)
1014{
1015 BUG_ON(bytes > cursor->resid);
1016 cursor->resid -= bytes;
1017
1018 if (bytes < cursor->lastlen) {
1019 cursor->lastlen -= bytes;
1020 } else {
1021 iov_iter_advance(&cursor->iov_iter, bytes - cursor->lastlen);
1022 cursor->lastlen = 0;
1023 }
1024
1025 return cursor->resid;
1026}
1027
1028/*
1029 * Message data is handled (sent or received) in pieces, where each
1030 * piece resides on a single page. The network layer might not
1031 * consume an entire piece at once. A data item's cursor keeps
1032 * track of which piece is next to process and how much remains to
1033 * be processed in that piece. It also tracks whether the current
1034 * piece is the last one in the data item.
1035 */
1036static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1037{
1038 size_t length = cursor->total_resid;
1039
1040 switch (cursor->data->type) {
1041 case CEPH_MSG_DATA_PAGELIST:
1042 ceph_msg_data_pagelist_cursor_init(cursor, length);
1043 break;
1044 case CEPH_MSG_DATA_PAGES:
1045 ceph_msg_data_pages_cursor_init(cursor, length);
1046 break;
1047#ifdef CONFIG_BLOCK
1048 case CEPH_MSG_DATA_BIO:
1049 ceph_msg_data_bio_cursor_init(cursor, length);
1050 break;
1051#endif /* CONFIG_BLOCK */
1052 case CEPH_MSG_DATA_BVECS:
1053 ceph_msg_data_bvecs_cursor_init(cursor, length);
1054 break;
1055 case CEPH_MSG_DATA_ITER:
1056 ceph_msg_data_iter_cursor_init(cursor, length);
1057 break;
1058 case CEPH_MSG_DATA_NONE:
1059 default:
1060 /* BUG(); */
1061 break;
1062 }
1063 cursor->need_crc = true;
1064}
1065
1066void ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor,
1067 struct ceph_msg *msg, size_t length)
1068{
1069 BUG_ON(!length);
1070 BUG_ON(length > msg->data_length);
1071 BUG_ON(!msg->num_data_items);
1072
1073 cursor->total_resid = length;
1074 cursor->data = msg->data;
1075 cursor->sr_resid = 0;
1076
1077 __ceph_msg_data_cursor_init(cursor);
1078}
1079
1080/*
1081 * Return the page containing the next piece to process for a given
1082 * data item, and supply the page offset and length of that piece.
1083 * Indicate whether this is the last piece in this data item.
1084 */
1085struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1086 size_t *page_offset, size_t *length)
1087{
1088 struct page *page;
1089
1090 switch (cursor->data->type) {
1091 case CEPH_MSG_DATA_PAGELIST:
1092 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1093 break;
1094 case CEPH_MSG_DATA_PAGES:
1095 page = ceph_msg_data_pages_next(cursor, page_offset, length);
1096 break;
1097#ifdef CONFIG_BLOCK
1098 case CEPH_MSG_DATA_BIO:
1099 page = ceph_msg_data_bio_next(cursor, page_offset, length);
1100 break;
1101#endif /* CONFIG_BLOCK */
1102 case CEPH_MSG_DATA_BVECS:
1103 page = ceph_msg_data_bvecs_next(cursor, page_offset, length);
1104 break;
1105 case CEPH_MSG_DATA_ITER:
1106 page = ceph_msg_data_iter_next(cursor, page_offset, length);
1107 break;
1108 case CEPH_MSG_DATA_NONE:
1109 default:
1110 page = NULL;
1111 break;
1112 }
1113
1114 BUG_ON(!page);
1115 BUG_ON(*page_offset + *length > PAGE_SIZE);
1116 BUG_ON(!*length);
1117 BUG_ON(*length > cursor->resid);
1118
1119 return page;
1120}
1121
1122/*
1123 * Returns true if the result moves the cursor on to the next piece
1124 * of the data item.
1125 */
1126void ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor, size_t bytes)
1127{
1128 bool new_piece;
1129
1130 BUG_ON(bytes > cursor->resid);
1131 switch (cursor->data->type) {
1132 case CEPH_MSG_DATA_PAGELIST:
1133 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1134 break;
1135 case CEPH_MSG_DATA_PAGES:
1136 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1137 break;
1138#ifdef CONFIG_BLOCK
1139 case CEPH_MSG_DATA_BIO:
1140 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1141 break;
1142#endif /* CONFIG_BLOCK */
1143 case CEPH_MSG_DATA_BVECS:
1144 new_piece = ceph_msg_data_bvecs_advance(cursor, bytes);
1145 break;
1146 case CEPH_MSG_DATA_ITER:
1147 new_piece = ceph_msg_data_iter_advance(cursor, bytes);
1148 break;
1149 case CEPH_MSG_DATA_NONE:
1150 default:
1151 BUG();
1152 break;
1153 }
1154 cursor->total_resid -= bytes;
1155
1156 if (!cursor->resid && cursor->total_resid) {
1157 cursor->data++;
1158 __ceph_msg_data_cursor_init(cursor);
1159 new_piece = true;
1160 }
1161 cursor->need_crc = new_piece;
1162}
1163
1164u32 ceph_crc32c_page(u32 crc, struct page *page, unsigned int page_offset,
1165 unsigned int length)
1166{
1167 char *kaddr;
1168
1169 kaddr = kmap(page);
1170 BUG_ON(kaddr == NULL);
1171 crc = crc32c(crc, kaddr + page_offset, length);
1172 kunmap(page);
1173
1174 return crc;
1175}
1176
1177bool ceph_addr_is_blank(const struct ceph_entity_addr *addr)
1178{
1179 struct sockaddr_storage ss = addr->in_addr; /* align */
1180 struct in_addr *addr4 = &((struct sockaddr_in *)&ss)->sin_addr;
1181 struct in6_addr *addr6 = &((struct sockaddr_in6 *)&ss)->sin6_addr;
1182
1183 switch (ss.ss_family) {
1184 case AF_INET:
1185 return addr4->s_addr == htonl(INADDR_ANY);
1186 case AF_INET6:
1187 return ipv6_addr_any(addr6);
1188 default:
1189 return true;
1190 }
1191}
1192EXPORT_SYMBOL(ceph_addr_is_blank);
1193
1194int ceph_addr_port(const struct ceph_entity_addr *addr)
1195{
1196 switch (get_unaligned(&addr->in_addr.ss_family)) {
1197 case AF_INET:
1198 return ntohs(get_unaligned(&((struct sockaddr_in *)&addr->in_addr)->sin_port));
1199 case AF_INET6:
1200 return ntohs(get_unaligned(&((struct sockaddr_in6 *)&addr->in_addr)->sin6_port));
1201 }
1202 return 0;
1203}
1204
1205void ceph_addr_set_port(struct ceph_entity_addr *addr, int p)
1206{
1207 switch (get_unaligned(&addr->in_addr.ss_family)) {
1208 case AF_INET:
1209 put_unaligned(htons(p), &((struct sockaddr_in *)&addr->in_addr)->sin_port);
1210 break;
1211 case AF_INET6:
1212 put_unaligned(htons(p), &((struct sockaddr_in6 *)&addr->in_addr)->sin6_port);
1213 break;
1214 }
1215}
1216
1217/*
1218 * Unlike other *_pton function semantics, zero indicates success.
1219 */
1220static int ceph_pton(const char *str, size_t len, struct ceph_entity_addr *addr,
1221 char delim, const char **ipend)
1222{
1223 memset(&addr->in_addr, 0, sizeof(addr->in_addr));
1224
1225 if (in4_pton(str, len, (u8 *)&((struct sockaddr_in *)&addr->in_addr)->sin_addr.s_addr, delim, ipend)) {
1226 put_unaligned(AF_INET, &addr->in_addr.ss_family);
1227 return 0;
1228 }
1229
1230 if (in6_pton(str, len, (u8 *)&((struct sockaddr_in6 *)&addr->in_addr)->sin6_addr.s6_addr, delim, ipend)) {
1231 put_unaligned(AF_INET6, &addr->in_addr.ss_family);
1232 return 0;
1233 }
1234
1235 return -EINVAL;
1236}
1237
1238/*
1239 * Extract hostname string and resolve using kernel DNS facility.
1240 */
1241#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1242static int ceph_dns_resolve_name(const char *name, size_t namelen,
1243 struct ceph_entity_addr *addr, char delim, const char **ipend)
1244{
1245 const char *end, *delim_p;
1246 char *colon_p, *ip_addr = NULL;
1247 int ip_len, ret;
1248
1249 /*
1250 * The end of the hostname occurs immediately preceding the delimiter or
1251 * the port marker (':') where the delimiter takes precedence.
1252 */
1253 delim_p = memchr(name, delim, namelen);
1254 colon_p = memchr(name, ':', namelen);
1255
1256 if (delim_p && colon_p)
1257 end = delim_p < colon_p ? delim_p : colon_p;
1258 else if (!delim_p && colon_p)
1259 end = colon_p;
1260 else {
1261 end = delim_p;
1262 if (!end) /* case: hostname:/ */
1263 end = name + namelen;
1264 }
1265
1266 if (end <= name)
1267 return -EINVAL;
1268
1269 /* do dns_resolve upcall */
1270 ip_len = dns_query(current->nsproxy->net_ns,
1271 NULL, name, end - name, NULL, &ip_addr, NULL, false);
1272 if (ip_len > 0)
1273 ret = ceph_pton(ip_addr, ip_len, addr, -1, NULL);
1274 else
1275 ret = -ESRCH;
1276
1277 kfree(ip_addr);
1278
1279 *ipend = end;
1280
1281 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1282 ret, ret ? "failed" : ceph_pr_addr(addr));
1283
1284 return ret;
1285}
1286#else
1287static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1288 struct ceph_entity_addr *addr, char delim, const char **ipend)
1289{
1290 return -EINVAL;
1291}
1292#endif
1293
1294/*
1295 * Parse a server name (IP or hostname). If a valid IP address is not found
1296 * then try to extract a hostname to resolve using userspace DNS upcall.
1297 */
1298static int ceph_parse_server_name(const char *name, size_t namelen,
1299 struct ceph_entity_addr *addr, char delim, const char **ipend)
1300{
1301 int ret;
1302
1303 ret = ceph_pton(name, namelen, addr, delim, ipend);
1304 if (ret)
1305 ret = ceph_dns_resolve_name(name, namelen, addr, delim, ipend);
1306
1307 return ret;
1308}
1309
1310/*
1311 * Parse an ip[:port] list into an addr array. Use the default
1312 * monitor port if a port isn't specified.
1313 */
1314int ceph_parse_ips(const char *c, const char *end,
1315 struct ceph_entity_addr *addr,
1316 int max_count, int *count, char delim)
1317{
1318 int i, ret = -EINVAL;
1319 const char *p = c;
1320
1321 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1322 for (i = 0; i < max_count; i++) {
1323 char cur_delim = delim;
1324 const char *ipend;
1325 int port;
1326
1327 if (*p == '[') {
1328 cur_delim = ']';
1329 p++;
1330 }
1331
1332 ret = ceph_parse_server_name(p, end - p, &addr[i], cur_delim,
1333 &ipend);
1334 if (ret)
1335 goto bad;
1336 ret = -EINVAL;
1337
1338 p = ipend;
1339
1340 if (cur_delim == ']') {
1341 if (*p != ']') {
1342 dout("missing matching ']'\n");
1343 goto bad;
1344 }
1345 p++;
1346 }
1347
1348 /* port? */
1349 if (p < end && *p == ':') {
1350 port = 0;
1351 p++;
1352 while (p < end && *p >= '0' && *p <= '9') {
1353 port = (port * 10) + (*p - '0');
1354 p++;
1355 }
1356 if (port == 0)
1357 port = CEPH_MON_PORT;
1358 else if (port > 65535)
1359 goto bad;
1360 } else {
1361 port = CEPH_MON_PORT;
1362 }
1363
1364 ceph_addr_set_port(&addr[i], port);
1365 /*
1366 * We want the type to be set according to ms_mode
1367 * option, but options are normally parsed after mon
1368 * addresses. Rather than complicating parsing, set
1369 * to LEGACY and override in build_initial_monmap()
1370 * for mon addresses and ceph_messenger_init() for
1371 * ip option.
1372 */
1373 addr[i].type = CEPH_ENTITY_ADDR_TYPE_LEGACY;
1374 addr[i].nonce = 0;
1375
1376 dout("%s got %s\n", __func__, ceph_pr_addr(&addr[i]));
1377
1378 if (p == end)
1379 break;
1380 if (*p != delim)
1381 goto bad;
1382 p++;
1383 }
1384
1385 if (p != end)
1386 goto bad;
1387
1388 if (count)
1389 *count = i + 1;
1390 return 0;
1391
1392bad:
1393 return ret;
1394}
1395
1396/*
1397 * Process message. This happens in the worker thread. The callback should
1398 * be careful not to do anything that waits on other incoming messages or it
1399 * may deadlock.
1400 */
1401void ceph_con_process_message(struct ceph_connection *con)
1402{
1403 struct ceph_msg *msg = con->in_msg;
1404
1405 BUG_ON(con->in_msg->con != con);
1406 con->in_msg = NULL;
1407
1408 /* if first message, set peer_name */
1409 if (con->peer_name.type == 0)
1410 con->peer_name = msg->hdr.src;
1411
1412 con->in_seq++;
1413 mutex_unlock(&con->mutex);
1414
1415 dout("===== %p %llu from %s%lld %d=%s len %d+%d+%d (%u %u %u) =====\n",
1416 msg, le64_to_cpu(msg->hdr.seq),
1417 ENTITY_NAME(msg->hdr.src),
1418 le16_to_cpu(msg->hdr.type),
1419 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
1420 le32_to_cpu(msg->hdr.front_len),
1421 le32_to_cpu(msg->hdr.middle_len),
1422 le32_to_cpu(msg->hdr.data_len),
1423 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
1424 con->ops->dispatch(con, msg);
1425
1426 mutex_lock(&con->mutex);
1427}
1428
1429/*
1430 * Atomically queue work on a connection after the specified delay.
1431 * Bump @con reference to avoid races with connection teardown.
1432 * Returns 0 if work was queued, or an error code otherwise.
1433 */
1434static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
1435{
1436 if (!con->ops->get(con)) {
1437 dout("%s %p ref count 0\n", __func__, con);
1438 return -ENOENT;
1439 }
1440
1441 if (delay >= HZ)
1442 delay = round_jiffies_relative(delay);
1443
1444 dout("%s %p %lu\n", __func__, con, delay);
1445 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
1446 dout("%s %p - already queued\n", __func__, con);
1447 con->ops->put(con);
1448 return -EBUSY;
1449 }
1450
1451 return 0;
1452}
1453
1454static void queue_con(struct ceph_connection *con)
1455{
1456 (void) queue_con_delay(con, 0);
1457}
1458
1459static void cancel_con(struct ceph_connection *con)
1460{
1461 if (cancel_delayed_work(&con->work)) {
1462 dout("%s %p\n", __func__, con);
1463 con->ops->put(con);
1464 }
1465}
1466
1467static bool con_sock_closed(struct ceph_connection *con)
1468{
1469 if (!ceph_con_flag_test_and_clear(con, CEPH_CON_F_SOCK_CLOSED))
1470 return false;
1471
1472#define CASE(x) \
1473 case CEPH_CON_S_ ## x: \
1474 con->error_msg = "socket closed (con state " #x ")"; \
1475 break;
1476
1477 switch (con->state) {
1478 CASE(CLOSED);
1479 CASE(PREOPEN);
1480 CASE(V1_BANNER);
1481 CASE(V1_CONNECT_MSG);
1482 CASE(V2_BANNER_PREFIX);
1483 CASE(V2_BANNER_PAYLOAD);
1484 CASE(V2_HELLO);
1485 CASE(V2_AUTH);
1486 CASE(V2_AUTH_SIGNATURE);
1487 CASE(V2_SESSION_CONNECT);
1488 CASE(V2_SESSION_RECONNECT);
1489 CASE(OPEN);
1490 CASE(STANDBY);
1491 default:
1492 BUG();
1493 }
1494#undef CASE
1495
1496 return true;
1497}
1498
1499static bool con_backoff(struct ceph_connection *con)
1500{
1501 int ret;
1502
1503 if (!ceph_con_flag_test_and_clear(con, CEPH_CON_F_BACKOFF))
1504 return false;
1505
1506 ret = queue_con_delay(con, con->delay);
1507 if (ret) {
1508 dout("%s: con %p FAILED to back off %lu\n", __func__,
1509 con, con->delay);
1510 BUG_ON(ret == -ENOENT);
1511 ceph_con_flag_set(con, CEPH_CON_F_BACKOFF);
1512 }
1513
1514 return true;
1515}
1516
1517/* Finish fault handling; con->mutex must *not* be held here */
1518
1519static void con_fault_finish(struct ceph_connection *con)
1520{
1521 dout("%s %p\n", __func__, con);
1522
1523 /*
1524 * in case we faulted due to authentication, invalidate our
1525 * current tickets so that we can get new ones.
1526 */
1527 if (con->v1.auth_retry) {
1528 dout("auth_retry %d, invalidating\n", con->v1.auth_retry);
1529 if (con->ops->invalidate_authorizer)
1530 con->ops->invalidate_authorizer(con);
1531 con->v1.auth_retry = 0;
1532 }
1533
1534 if (con->ops->fault)
1535 con->ops->fault(con);
1536}
1537
1538/*
1539 * Do some work on a connection. Drop a connection ref when we're done.
1540 */
1541static void ceph_con_workfn(struct work_struct *work)
1542{
1543 struct ceph_connection *con = container_of(work, struct ceph_connection,
1544 work.work);
1545 bool fault;
1546
1547 mutex_lock(&con->mutex);
1548 while (true) {
1549 int ret;
1550
1551 if ((fault = con_sock_closed(con))) {
1552 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
1553 break;
1554 }
1555 if (con_backoff(con)) {
1556 dout("%s: con %p BACKOFF\n", __func__, con);
1557 break;
1558 }
1559 if (con->state == CEPH_CON_S_STANDBY) {
1560 dout("%s: con %p STANDBY\n", __func__, con);
1561 break;
1562 }
1563 if (con->state == CEPH_CON_S_CLOSED) {
1564 dout("%s: con %p CLOSED\n", __func__, con);
1565 BUG_ON(con->sock);
1566 break;
1567 }
1568 if (con->state == CEPH_CON_S_PREOPEN) {
1569 dout("%s: con %p PREOPEN\n", __func__, con);
1570 BUG_ON(con->sock);
1571 }
1572
1573 if (ceph_msgr2(from_msgr(con->msgr)))
1574 ret = ceph_con_v2_try_read(con);
1575 else
1576 ret = ceph_con_v1_try_read(con);
1577 if (ret < 0) {
1578 if (ret == -EAGAIN)
1579 continue;
1580 if (!con->error_msg)
1581 con->error_msg = "socket error on read";
1582 fault = true;
1583 break;
1584 }
1585
1586 if (ceph_msgr2(from_msgr(con->msgr)))
1587 ret = ceph_con_v2_try_write(con);
1588 else
1589 ret = ceph_con_v1_try_write(con);
1590 if (ret < 0) {
1591 if (ret == -EAGAIN)
1592 continue;
1593 if (!con->error_msg)
1594 con->error_msg = "socket error on write";
1595 fault = true;
1596 }
1597
1598 break; /* If we make it to here, we're done */
1599 }
1600 if (fault)
1601 con_fault(con);
1602 mutex_unlock(&con->mutex);
1603
1604 if (fault)
1605 con_fault_finish(con);
1606
1607 con->ops->put(con);
1608}
1609
1610/*
1611 * Generic error/fault handler. A retry mechanism is used with
1612 * exponential backoff
1613 */
1614static void con_fault(struct ceph_connection *con)
1615{
1616 dout("fault %p state %d to peer %s\n",
1617 con, con->state, ceph_pr_addr(&con->peer_addr));
1618
1619 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
1620 ceph_pr_addr(&con->peer_addr), con->error_msg);
1621 con->error_msg = NULL;
1622
1623 WARN_ON(con->state == CEPH_CON_S_STANDBY ||
1624 con->state == CEPH_CON_S_CLOSED);
1625
1626 ceph_con_reset_protocol(con);
1627
1628 if (ceph_con_flag_test(con, CEPH_CON_F_LOSSYTX)) {
1629 dout("fault on LOSSYTX channel, marking CLOSED\n");
1630 con->state = CEPH_CON_S_CLOSED;
1631 return;
1632 }
1633
1634 /* Requeue anything that hasn't been acked */
1635 list_splice_init(&con->out_sent, &con->out_queue);
1636
1637 /* If there are no messages queued or keepalive pending, place
1638 * the connection in a STANDBY state */
1639 if (list_empty(&con->out_queue) &&
1640 !ceph_con_flag_test(con, CEPH_CON_F_KEEPALIVE_PENDING)) {
1641 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
1642 ceph_con_flag_clear(con, CEPH_CON_F_WRITE_PENDING);
1643 con->state = CEPH_CON_S_STANDBY;
1644 } else {
1645 /* retry after a delay. */
1646 con->state = CEPH_CON_S_PREOPEN;
1647 if (!con->delay) {
1648 con->delay = BASE_DELAY_INTERVAL;
1649 } else if (con->delay < MAX_DELAY_INTERVAL) {
1650 con->delay *= 2;
1651 if (con->delay > MAX_DELAY_INTERVAL)
1652 con->delay = MAX_DELAY_INTERVAL;
1653 }
1654 ceph_con_flag_set(con, CEPH_CON_F_BACKOFF);
1655 queue_con(con);
1656 }
1657}
1658
1659void ceph_messenger_reset_nonce(struct ceph_messenger *msgr)
1660{
1661 u32 nonce = le32_to_cpu(msgr->inst.addr.nonce) + 1000000;
1662 msgr->inst.addr.nonce = cpu_to_le32(nonce);
1663 ceph_encode_my_addr(msgr);
1664}
1665
1666/*
1667 * initialize a new messenger instance
1668 */
1669void ceph_messenger_init(struct ceph_messenger *msgr,
1670 struct ceph_entity_addr *myaddr)
1671{
1672 spin_lock_init(&msgr->global_seq_lock);
1673
1674 if (myaddr) {
1675 memcpy(&msgr->inst.addr.in_addr, &myaddr->in_addr,
1676 sizeof(msgr->inst.addr.in_addr));
1677 ceph_addr_set_port(&msgr->inst.addr, 0);
1678 }
1679
1680 /*
1681 * Since nautilus, clients are identified using type ANY.
1682 * For msgr1, ceph_encode_banner_addr() munges it to NONE.
1683 */
1684 msgr->inst.addr.type = CEPH_ENTITY_ADDR_TYPE_ANY;
1685
1686 /* generate a random non-zero nonce */
1687 do {
1688 get_random_bytes(&msgr->inst.addr.nonce,
1689 sizeof(msgr->inst.addr.nonce));
1690 } while (!msgr->inst.addr.nonce);
1691 ceph_encode_my_addr(msgr);
1692
1693 atomic_set(&msgr->stopping, 0);
1694 write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
1695
1696 dout("%s %p\n", __func__, msgr);
1697}
1698
1699void ceph_messenger_fini(struct ceph_messenger *msgr)
1700{
1701 put_net(read_pnet(&msgr->net));
1702}
1703
1704static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con)
1705{
1706 if (msg->con)
1707 msg->con->ops->put(msg->con);
1708
1709 msg->con = con ? con->ops->get(con) : NULL;
1710 BUG_ON(msg->con != con);
1711}
1712
1713static void clear_standby(struct ceph_connection *con)
1714{
1715 /* come back from STANDBY? */
1716 if (con->state == CEPH_CON_S_STANDBY) {
1717 dout("clear_standby %p and ++connect_seq\n", con);
1718 con->state = CEPH_CON_S_PREOPEN;
1719 con->v1.connect_seq++;
1720 WARN_ON(ceph_con_flag_test(con, CEPH_CON_F_WRITE_PENDING));
1721 WARN_ON(ceph_con_flag_test(con, CEPH_CON_F_KEEPALIVE_PENDING));
1722 }
1723}
1724
1725/*
1726 * Queue up an outgoing message on the given connection.
1727 *
1728 * Consumes a ref on @msg.
1729 */
1730void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
1731{
1732 /* set src+dst */
1733 msg->hdr.src = con->msgr->inst.name;
1734 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
1735 msg->needs_out_seq = true;
1736
1737 mutex_lock(&con->mutex);
1738
1739 if (con->state == CEPH_CON_S_CLOSED) {
1740 dout("con_send %p closed, dropping %p\n", con, msg);
1741 ceph_msg_put(msg);
1742 mutex_unlock(&con->mutex);
1743 return;
1744 }
1745
1746 msg_con_set(msg, con);
1747
1748 BUG_ON(!list_empty(&msg->list_head));
1749 list_add_tail(&msg->list_head, &con->out_queue);
1750 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
1751 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
1752 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
1753 le32_to_cpu(msg->hdr.front_len),
1754 le32_to_cpu(msg->hdr.middle_len),
1755 le32_to_cpu(msg->hdr.data_len));
1756
1757 clear_standby(con);
1758 mutex_unlock(&con->mutex);
1759
1760 /* if there wasn't anything waiting to send before, queue
1761 * new work */
1762 if (!ceph_con_flag_test_and_set(con, CEPH_CON_F_WRITE_PENDING))
1763 queue_con(con);
1764}
1765EXPORT_SYMBOL(ceph_con_send);
1766
1767/*
1768 * Revoke a message that was previously queued for send
1769 */
1770void ceph_msg_revoke(struct ceph_msg *msg)
1771{
1772 struct ceph_connection *con = msg->con;
1773
1774 if (!con) {
1775 dout("%s msg %p null con\n", __func__, msg);
1776 return; /* Message not in our possession */
1777 }
1778
1779 mutex_lock(&con->mutex);
1780 if (list_empty(&msg->list_head)) {
1781 WARN_ON(con->out_msg == msg);
1782 dout("%s con %p msg %p not linked\n", __func__, con, msg);
1783 mutex_unlock(&con->mutex);
1784 return;
1785 }
1786
1787 dout("%s con %p msg %p was linked\n", __func__, con, msg);
1788 msg->hdr.seq = 0;
1789 ceph_msg_remove(msg);
1790
1791 if (con->out_msg == msg) {
1792 WARN_ON(con->state != CEPH_CON_S_OPEN);
1793 dout("%s con %p msg %p was sending\n", __func__, con, msg);
1794 if (ceph_msgr2(from_msgr(con->msgr)))
1795 ceph_con_v2_revoke(con);
1796 else
1797 ceph_con_v1_revoke(con);
1798 ceph_msg_put(con->out_msg);
1799 con->out_msg = NULL;
1800 } else {
1801 dout("%s con %p msg %p not current, out_msg %p\n", __func__,
1802 con, msg, con->out_msg);
1803 }
1804 mutex_unlock(&con->mutex);
1805}
1806
1807/*
1808 * Revoke a message that we may be reading data into
1809 */
1810void ceph_msg_revoke_incoming(struct ceph_msg *msg)
1811{
1812 struct ceph_connection *con = msg->con;
1813
1814 if (!con) {
1815 dout("%s msg %p null con\n", __func__, msg);
1816 return; /* Message not in our possession */
1817 }
1818
1819 mutex_lock(&con->mutex);
1820 if (con->in_msg == msg) {
1821 WARN_ON(con->state != CEPH_CON_S_OPEN);
1822 dout("%s con %p msg %p was recving\n", __func__, con, msg);
1823 if (ceph_msgr2(from_msgr(con->msgr)))
1824 ceph_con_v2_revoke_incoming(con);
1825 else
1826 ceph_con_v1_revoke_incoming(con);
1827 ceph_msg_put(con->in_msg);
1828 con->in_msg = NULL;
1829 } else {
1830 dout("%s con %p msg %p not current, in_msg %p\n", __func__,
1831 con, msg, con->in_msg);
1832 }
1833 mutex_unlock(&con->mutex);
1834}
1835
1836/*
1837 * Queue a keepalive byte to ensure the tcp connection is alive.
1838 */
1839void ceph_con_keepalive(struct ceph_connection *con)
1840{
1841 dout("con_keepalive %p\n", con);
1842 mutex_lock(&con->mutex);
1843 clear_standby(con);
1844 ceph_con_flag_set(con, CEPH_CON_F_KEEPALIVE_PENDING);
1845 mutex_unlock(&con->mutex);
1846
1847 if (!ceph_con_flag_test_and_set(con, CEPH_CON_F_WRITE_PENDING))
1848 queue_con(con);
1849}
1850EXPORT_SYMBOL(ceph_con_keepalive);
1851
1852bool ceph_con_keepalive_expired(struct ceph_connection *con,
1853 unsigned long interval)
1854{
1855 if (interval > 0 &&
1856 (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
1857 struct timespec64 now;
1858 struct timespec64 ts;
1859 ktime_get_real_ts64(&now);
1860 jiffies_to_timespec64(interval, &ts);
1861 ts = timespec64_add(con->last_keepalive_ack, ts);
1862 return timespec64_compare(&now, &ts) >= 0;
1863 }
1864 return false;
1865}
1866
1867static struct ceph_msg_data *ceph_msg_data_add(struct ceph_msg *msg)
1868{
1869 BUG_ON(msg->num_data_items >= msg->max_data_items);
1870 return &msg->data[msg->num_data_items++];
1871}
1872
1873static void ceph_msg_data_destroy(struct ceph_msg_data *data)
1874{
1875 if (data->type == CEPH_MSG_DATA_PAGES && data->own_pages) {
1876 int num_pages = calc_pages_for(data->alignment, data->length);
1877 ceph_release_page_vector(data->pages, num_pages);
1878 } else if (data->type == CEPH_MSG_DATA_PAGELIST) {
1879 ceph_pagelist_release(data->pagelist);
1880 }
1881}
1882
1883void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
1884 size_t length, size_t alignment, bool own_pages)
1885{
1886 struct ceph_msg_data *data;
1887
1888 BUG_ON(!pages);
1889 BUG_ON(!length);
1890
1891 data = ceph_msg_data_add(msg);
1892 data->type = CEPH_MSG_DATA_PAGES;
1893 data->pages = pages;
1894 data->length = length;
1895 data->alignment = alignment & ~PAGE_MASK;
1896 data->own_pages = own_pages;
1897
1898 msg->data_length += length;
1899}
1900EXPORT_SYMBOL(ceph_msg_data_add_pages);
1901
1902void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
1903 struct ceph_pagelist *pagelist)
1904{
1905 struct ceph_msg_data *data;
1906
1907 BUG_ON(!pagelist);
1908 BUG_ON(!pagelist->length);
1909
1910 data = ceph_msg_data_add(msg);
1911 data->type = CEPH_MSG_DATA_PAGELIST;
1912 refcount_inc(&pagelist->refcnt);
1913 data->pagelist = pagelist;
1914
1915 msg->data_length += pagelist->length;
1916}
1917EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
1918
1919#ifdef CONFIG_BLOCK
1920void ceph_msg_data_add_bio(struct ceph_msg *msg, struct ceph_bio_iter *bio_pos,
1921 u32 length)
1922{
1923 struct ceph_msg_data *data;
1924
1925 data = ceph_msg_data_add(msg);
1926 data->type = CEPH_MSG_DATA_BIO;
1927 data->bio_pos = *bio_pos;
1928 data->bio_length = length;
1929
1930 msg->data_length += length;
1931}
1932EXPORT_SYMBOL(ceph_msg_data_add_bio);
1933#endif /* CONFIG_BLOCK */
1934
1935void ceph_msg_data_add_bvecs(struct ceph_msg *msg,
1936 struct ceph_bvec_iter *bvec_pos)
1937{
1938 struct ceph_msg_data *data;
1939
1940 data = ceph_msg_data_add(msg);
1941 data->type = CEPH_MSG_DATA_BVECS;
1942 data->bvec_pos = *bvec_pos;
1943
1944 msg->data_length += bvec_pos->iter.bi_size;
1945}
1946EXPORT_SYMBOL(ceph_msg_data_add_bvecs);
1947
1948void ceph_msg_data_add_iter(struct ceph_msg *msg,
1949 struct iov_iter *iter)
1950{
1951 struct ceph_msg_data *data;
1952
1953 data = ceph_msg_data_add(msg);
1954 data->type = CEPH_MSG_DATA_ITER;
1955 data->iter = *iter;
1956
1957 msg->data_length += iov_iter_count(&data->iter);
1958}
1959
1960/*
1961 * construct a new message with given type, size
1962 * the new msg has a ref count of 1.
1963 */
1964struct ceph_msg *ceph_msg_new2(int type, int front_len, int max_data_items,
1965 gfp_t flags, bool can_fail)
1966{
1967 struct ceph_msg *m;
1968
1969 m = kmem_cache_zalloc(ceph_msg_cache, flags);
1970 if (m == NULL)
1971 goto out;
1972
1973 m->hdr.type = cpu_to_le16(type);
1974 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
1975 m->hdr.front_len = cpu_to_le32(front_len);
1976
1977 INIT_LIST_HEAD(&m->list_head);
1978 kref_init(&m->kref);
1979
1980 /* front */
1981 if (front_len) {
1982 m->front.iov_base = kvmalloc(front_len, flags);
1983 if (m->front.iov_base == NULL) {
1984 dout("ceph_msg_new can't allocate %d bytes\n",
1985 front_len);
1986 goto out2;
1987 }
1988 } else {
1989 m->front.iov_base = NULL;
1990 }
1991 m->front_alloc_len = m->front.iov_len = front_len;
1992
1993 if (max_data_items) {
1994 m->data = kmalloc_array(max_data_items, sizeof(*m->data),
1995 flags);
1996 if (!m->data)
1997 goto out2;
1998
1999 m->max_data_items = max_data_items;
2000 }
2001
2002 dout("ceph_msg_new %p front %d\n", m, front_len);
2003 return m;
2004
2005out2:
2006 ceph_msg_put(m);
2007out:
2008 if (!can_fail) {
2009 pr_err("msg_new can't create type %d front %d\n", type,
2010 front_len);
2011 WARN_ON(1);
2012 } else {
2013 dout("msg_new can't create type %d front %d\n", type,
2014 front_len);
2015 }
2016 return NULL;
2017}
2018EXPORT_SYMBOL(ceph_msg_new2);
2019
2020struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
2021 bool can_fail)
2022{
2023 return ceph_msg_new2(type, front_len, 0, flags, can_fail);
2024}
2025EXPORT_SYMBOL(ceph_msg_new);
2026
2027/*
2028 * Allocate "middle" portion of a message, if it is needed and wasn't
2029 * allocated by alloc_msg. This allows us to read a small fixed-size
2030 * per-type header in the front and then gracefully fail (i.e.,
2031 * propagate the error to the caller based on info in the front) when
2032 * the middle is too large.
2033 */
2034static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
2035{
2036 int type = le16_to_cpu(msg->hdr.type);
2037 int middle_len = le32_to_cpu(msg->hdr.middle_len);
2038
2039 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
2040 ceph_msg_type_name(type), middle_len);
2041 BUG_ON(!middle_len);
2042 BUG_ON(msg->middle);
2043
2044 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
2045 if (!msg->middle)
2046 return -ENOMEM;
2047 return 0;
2048}
2049
2050/*
2051 * Allocate a message for receiving an incoming message on a
2052 * connection, and save the result in con->in_msg. Uses the
2053 * connection's private alloc_msg op if available.
2054 *
2055 * Returns 0 on success, or a negative error code.
2056 *
2057 * On success, if we set *skip = 1:
2058 * - the next message should be skipped and ignored.
2059 * - con->in_msg == NULL
2060 * or if we set *skip = 0:
2061 * - con->in_msg is non-null.
2062 * On error (ENOMEM, EAGAIN, ...),
2063 * - con->in_msg == NULL
2064 */
2065int ceph_con_in_msg_alloc(struct ceph_connection *con,
2066 struct ceph_msg_header *hdr, int *skip)
2067{
2068 int middle_len = le32_to_cpu(hdr->middle_len);
2069 struct ceph_msg *msg;
2070 int ret = 0;
2071
2072 BUG_ON(con->in_msg != NULL);
2073 BUG_ON(!con->ops->alloc_msg);
2074
2075 mutex_unlock(&con->mutex);
2076 msg = con->ops->alloc_msg(con, hdr, skip);
2077 mutex_lock(&con->mutex);
2078 if (con->state != CEPH_CON_S_OPEN) {
2079 if (msg)
2080 ceph_msg_put(msg);
2081 return -EAGAIN;
2082 }
2083 if (msg) {
2084 BUG_ON(*skip);
2085 msg_con_set(msg, con);
2086 con->in_msg = msg;
2087 } else {
2088 /*
2089 * Null message pointer means either we should skip
2090 * this message or we couldn't allocate memory. The
2091 * former is not an error.
2092 */
2093 if (*skip)
2094 return 0;
2095
2096 con->error_msg = "error allocating memory for incoming message";
2097 return -ENOMEM;
2098 }
2099 memcpy(&con->in_msg->hdr, hdr, sizeof(*hdr));
2100
2101 if (middle_len && !con->in_msg->middle) {
2102 ret = ceph_alloc_middle(con, con->in_msg);
2103 if (ret < 0) {
2104 ceph_msg_put(con->in_msg);
2105 con->in_msg = NULL;
2106 }
2107 }
2108
2109 return ret;
2110}
2111
2112void ceph_con_get_out_msg(struct ceph_connection *con)
2113{
2114 struct ceph_msg *msg;
2115
2116 BUG_ON(list_empty(&con->out_queue));
2117 msg = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
2118 WARN_ON(msg->con != con);
2119
2120 /*
2121 * Put the message on "sent" list using a ref from ceph_con_send().
2122 * It is put when the message is acked or revoked.
2123 */
2124 list_move_tail(&msg->list_head, &con->out_sent);
2125
2126 /*
2127 * Only assign outgoing seq # if we haven't sent this message
2128 * yet. If it is requeued, resend with it's original seq.
2129 */
2130 if (msg->needs_out_seq) {
2131 msg->hdr.seq = cpu_to_le64(++con->out_seq);
2132 msg->needs_out_seq = false;
2133
2134 if (con->ops->reencode_message)
2135 con->ops->reencode_message(msg);
2136 }
2137
2138 /*
2139 * Get a ref for out_msg. It is put when we are done sending the
2140 * message or in case of a fault.
2141 */
2142 WARN_ON(con->out_msg);
2143 con->out_msg = ceph_msg_get(msg);
2144}
2145
2146/*
2147 * Free a generically kmalloc'd message.
2148 */
2149static void ceph_msg_free(struct ceph_msg *m)
2150{
2151 dout("%s %p\n", __func__, m);
2152 kvfree(m->front.iov_base);
2153 kfree(m->data);
2154 kmem_cache_free(ceph_msg_cache, m);
2155}
2156
2157static void ceph_msg_release(struct kref *kref)
2158{
2159 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
2160 int i;
2161
2162 dout("%s %p\n", __func__, m);
2163 WARN_ON(!list_empty(&m->list_head));
2164
2165 msg_con_set(m, NULL);
2166
2167 /* drop middle, data, if any */
2168 if (m->middle) {
2169 ceph_buffer_put(m->middle);
2170 m->middle = NULL;
2171 }
2172
2173 for (i = 0; i < m->num_data_items; i++)
2174 ceph_msg_data_destroy(&m->data[i]);
2175
2176 if (m->pool)
2177 ceph_msgpool_put(m->pool, m);
2178 else
2179 ceph_msg_free(m);
2180}
2181
2182struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
2183{
2184 dout("%s %p (was %d)\n", __func__, msg,
2185 kref_read(&msg->kref));
2186 kref_get(&msg->kref);
2187 return msg;
2188}
2189EXPORT_SYMBOL(ceph_msg_get);
2190
2191void ceph_msg_put(struct ceph_msg *msg)
2192{
2193 dout("%s %p (was %d)\n", __func__, msg,
2194 kref_read(&msg->kref));
2195 kref_put(&msg->kref, ceph_msg_release);
2196}
2197EXPORT_SYMBOL(ceph_msg_put);
2198
2199void ceph_msg_dump(struct ceph_msg *msg)
2200{
2201 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
2202 msg->front_alloc_len, msg->data_length);
2203 print_hex_dump(KERN_DEBUG, "header: ",
2204 DUMP_PREFIX_OFFSET, 16, 1,
2205 &msg->hdr, sizeof(msg->hdr), true);
2206 print_hex_dump(KERN_DEBUG, " front: ",
2207 DUMP_PREFIX_OFFSET, 16, 1,
2208 msg->front.iov_base, msg->front.iov_len, true);
2209 if (msg->middle)
2210 print_hex_dump(KERN_DEBUG, "middle: ",
2211 DUMP_PREFIX_OFFSET, 16, 1,
2212 msg->middle->vec.iov_base,
2213 msg->middle->vec.iov_len, true);
2214 print_hex_dump(KERN_DEBUG, "footer: ",
2215 DUMP_PREFIX_OFFSET, 16, 1,
2216 &msg->footer, sizeof(msg->footer), true);
2217}
2218EXPORT_SYMBOL(ceph_msg_dump);