Loading...
1// SPDX-License-Identifier: GPL-2.0
2#include <linux/ceph/ceph_debug.h>
3
4#include <linux/crc32c.h>
5#include <linux/ctype.h>
6#include <linux/highmem.h>
7#include <linux/inet.h>
8#include <linux/kthread.h>
9#include <linux/net.h>
10#include <linux/nsproxy.h>
11#include <linux/sched/mm.h>
12#include <linux/slab.h>
13#include <linux/socket.h>
14#include <linux/string.h>
15#ifdef CONFIG_BLOCK
16#include <linux/bio.h>
17#endif /* CONFIG_BLOCK */
18#include <linux/dns_resolver.h>
19#include <net/tcp.h>
20
21#include <linux/ceph/ceph_features.h>
22#include <linux/ceph/libceph.h>
23#include <linux/ceph/messenger.h>
24#include <linux/ceph/decode.h>
25#include <linux/ceph/pagelist.h>
26#include <linux/export.h>
27
28/*
29 * Ceph uses the messenger to exchange ceph_msg messages with other
30 * hosts in the system. The messenger provides ordered and reliable
31 * delivery. We tolerate TCP disconnects by reconnecting (with
32 * exponential backoff) in the case of a fault (disconnection, bad
33 * crc, protocol error). Acks allow sent messages to be discarded by
34 * the sender.
35 */
36
37/*
38 * We track the state of the socket on a given connection using
39 * values defined below. The transition to a new socket state is
40 * handled by a function which verifies we aren't coming from an
41 * unexpected state.
42 *
43 * --------
44 * | NEW* | transient initial state
45 * --------
46 * | con_sock_state_init()
47 * v
48 * ----------
49 * | CLOSED | initialized, but no socket (and no
50 * ---------- TCP connection)
51 * ^ \
52 * | \ con_sock_state_connecting()
53 * | ----------------------
54 * | \
55 * + con_sock_state_closed() \
56 * |+--------------------------- \
57 * | \ \ \
58 * | ----------- \ \
59 * | | CLOSING | socket event; \ \
60 * | ----------- await close \ \
61 * | ^ \ |
62 * | | \ |
63 * | + con_sock_state_closing() \ |
64 * | / \ | |
65 * | / --------------- | |
66 * | / \ v v
67 * | / --------------
68 * | / -----------------| CONNECTING | socket created, TCP
69 * | | / -------------- connect initiated
70 * | | | con_sock_state_connected()
71 * | | v
72 * -------------
73 * | CONNECTED | TCP connection established
74 * -------------
75 *
76 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
77 */
78
79#define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
80#define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
81#define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
82#define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
83#define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
84
85/*
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;
165static char tag_keepalive2 = CEPH_MSGR_TAG_KEEPALIVE2;
166
167#ifdef CONFIG_LOCKDEP
168static struct lock_class_key socket_class;
169#endif
170
171/*
172 * When skipping (ignoring) a block of input we read it into a "skip
173 * buffer," which is this many bytes in size.
174 */
175#define SKIP_BUF_SIZE 1024
176
177static void queue_con(struct ceph_connection *con);
178static void cancel_con(struct ceph_connection *con);
179static void ceph_con_workfn(struct work_struct *);
180static void con_fault(struct ceph_connection *con);
181
182/*
183 * Nicely render a sockaddr as a string. An array of formatted
184 * strings is used, to approximate reentrancy.
185 */
186#define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
187#define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
188#define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
189#define MAX_ADDR_STR_LEN 64 /* 54 is enough */
190
191static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
192static atomic_t addr_str_seq = ATOMIC_INIT(0);
193
194static struct page *zero_page; /* used in certain error cases */
195
196const char *ceph_pr_addr(const struct sockaddr_storage *ss)
197{
198 int i;
199 char *s;
200 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
201 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
202
203 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
204 s = addr_str[i];
205
206 switch (ss->ss_family) {
207 case AF_INET:
208 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
209 ntohs(in4->sin_port));
210 break;
211
212 case AF_INET6:
213 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
214 ntohs(in6->sin6_port));
215 break;
216
217 default:
218 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
219 ss->ss_family);
220 }
221
222 return s;
223}
224EXPORT_SYMBOL(ceph_pr_addr);
225
226static void encode_my_addr(struct ceph_messenger *msgr)
227{
228 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
229 ceph_encode_addr(&msgr->my_enc_addr);
230}
231
232/*
233 * work queue for all reading and writing to/from the socket.
234 */
235static struct workqueue_struct *ceph_msgr_wq;
236
237static int ceph_msgr_slab_init(void)
238{
239 BUG_ON(ceph_msg_cache);
240 ceph_msg_cache = KMEM_CACHE(ceph_msg, 0);
241 if (!ceph_msg_cache)
242 return -ENOMEM;
243
244 BUG_ON(ceph_msg_data_cache);
245 ceph_msg_data_cache = KMEM_CACHE(ceph_msg_data, 0);
246 if (ceph_msg_data_cache)
247 return 0;
248
249 kmem_cache_destroy(ceph_msg_cache);
250 ceph_msg_cache = NULL;
251
252 return -ENOMEM;
253}
254
255static void ceph_msgr_slab_exit(void)
256{
257 BUG_ON(!ceph_msg_data_cache);
258 kmem_cache_destroy(ceph_msg_data_cache);
259 ceph_msg_data_cache = NULL;
260
261 BUG_ON(!ceph_msg_cache);
262 kmem_cache_destroy(ceph_msg_cache);
263 ceph_msg_cache = NULL;
264}
265
266static void _ceph_msgr_exit(void)
267{
268 if (ceph_msgr_wq) {
269 destroy_workqueue(ceph_msgr_wq);
270 ceph_msgr_wq = NULL;
271 }
272
273 BUG_ON(zero_page == NULL);
274 put_page(zero_page);
275 zero_page = NULL;
276
277 ceph_msgr_slab_exit();
278}
279
280int __init ceph_msgr_init(void)
281{
282 if (ceph_msgr_slab_init())
283 return -ENOMEM;
284
285 BUG_ON(zero_page != NULL);
286 zero_page = ZERO_PAGE(0);
287 get_page(zero_page);
288
289 /*
290 * The number of active work items is limited by the number of
291 * connections, so leave @max_active at default.
292 */
293 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
294 if (ceph_msgr_wq)
295 return 0;
296
297 pr_err("msgr_init failed to create workqueue\n");
298 _ceph_msgr_exit();
299
300 return -ENOMEM;
301}
302
303void ceph_msgr_exit(void)
304{
305 BUG_ON(ceph_msgr_wq == NULL);
306
307 _ceph_msgr_exit();
308}
309
310void ceph_msgr_flush(void)
311{
312 flush_workqueue(ceph_msgr_wq);
313}
314EXPORT_SYMBOL(ceph_msgr_flush);
315
316/* Connection socket state transition functions */
317
318static void con_sock_state_init(struct ceph_connection *con)
319{
320 int old_state;
321
322 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
323 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
324 printk("%s: unexpected old state %d\n", __func__, old_state);
325 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
326 CON_SOCK_STATE_CLOSED);
327}
328
329static void con_sock_state_connecting(struct ceph_connection *con)
330{
331 int old_state;
332
333 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
334 if (WARN_ON(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_CONNECTING);
338}
339
340static void con_sock_state_connected(struct ceph_connection *con)
341{
342 int old_state;
343
344 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
345 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
346 printk("%s: unexpected old state %d\n", __func__, old_state);
347 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
348 CON_SOCK_STATE_CONNECTED);
349}
350
351static void con_sock_state_closing(struct ceph_connection *con)
352{
353 int old_state;
354
355 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
356 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
357 old_state != CON_SOCK_STATE_CONNECTED &&
358 old_state != CON_SOCK_STATE_CLOSING))
359 printk("%s: unexpected old state %d\n", __func__, old_state);
360 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
361 CON_SOCK_STATE_CLOSING);
362}
363
364static void con_sock_state_closed(struct ceph_connection *con)
365{
366 int old_state;
367
368 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
369 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
370 old_state != CON_SOCK_STATE_CLOSING &&
371 old_state != CON_SOCK_STATE_CONNECTING &&
372 old_state != CON_SOCK_STATE_CLOSED))
373 printk("%s: unexpected old state %d\n", __func__, old_state);
374 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
375 CON_SOCK_STATE_CLOSED);
376}
377
378/*
379 * socket callback functions
380 */
381
382/* data available on socket, or listen socket received a connect */
383static void ceph_sock_data_ready(struct sock *sk)
384{
385 struct ceph_connection *con = sk->sk_user_data;
386 if (atomic_read(&con->msgr->stopping)) {
387 return;
388 }
389
390 if (sk->sk_state != TCP_CLOSE_WAIT) {
391 dout("%s on %p state = %lu, queueing work\n", __func__,
392 con, con->state);
393 queue_con(con);
394 }
395}
396
397/* socket has buffer space for writing */
398static void ceph_sock_write_space(struct sock *sk)
399{
400 struct ceph_connection *con = sk->sk_user_data;
401
402 /* only queue to workqueue if there is data we want to write,
403 * and there is sufficient space in the socket buffer to accept
404 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
405 * doesn't get called again until try_write() fills the socket
406 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
407 * and net/core/stream.c:sk_stream_write_space().
408 */
409 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
410 if (sk_stream_is_writeable(sk)) {
411 dout("%s %p queueing write work\n", __func__, con);
412 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
413 queue_con(con);
414 }
415 } else {
416 dout("%s %p nothing to write\n", __func__, con);
417 }
418}
419
420/* socket's state has changed */
421static void ceph_sock_state_change(struct sock *sk)
422{
423 struct ceph_connection *con = sk->sk_user_data;
424
425 dout("%s %p state = %lu sk_state = %u\n", __func__,
426 con, con->state, sk->sk_state);
427
428 switch (sk->sk_state) {
429 case TCP_CLOSE:
430 dout("%s TCP_CLOSE\n", __func__);
431 /* fall through */
432 case TCP_CLOSE_WAIT:
433 dout("%s TCP_CLOSE_WAIT\n", __func__);
434 con_sock_state_closing(con);
435 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
436 queue_con(con);
437 break;
438 case TCP_ESTABLISHED:
439 dout("%s TCP_ESTABLISHED\n", __func__);
440 con_sock_state_connected(con);
441 queue_con(con);
442 break;
443 default: /* Everything else is uninteresting */
444 break;
445 }
446}
447
448/*
449 * set up socket callbacks
450 */
451static void set_sock_callbacks(struct socket *sock,
452 struct ceph_connection *con)
453{
454 struct sock *sk = sock->sk;
455 sk->sk_user_data = con;
456 sk->sk_data_ready = ceph_sock_data_ready;
457 sk->sk_write_space = ceph_sock_write_space;
458 sk->sk_state_change = ceph_sock_state_change;
459}
460
461
462/*
463 * socket helpers
464 */
465
466/*
467 * initiate connection to a remote socket.
468 */
469static int ceph_tcp_connect(struct ceph_connection *con)
470{
471 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
472 struct socket *sock;
473 unsigned int noio_flag;
474 int ret;
475
476 BUG_ON(con->sock);
477
478 /* sock_create_kern() allocates with GFP_KERNEL */
479 noio_flag = memalloc_noio_save();
480 ret = sock_create_kern(read_pnet(&con->msgr->net), paddr->ss_family,
481 SOCK_STREAM, IPPROTO_TCP, &sock);
482 memalloc_noio_restore(noio_flag);
483 if (ret)
484 return ret;
485 sock->sk->sk_allocation = GFP_NOFS;
486
487#ifdef CONFIG_LOCKDEP
488 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
489#endif
490
491 set_sock_callbacks(sock, con);
492
493 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
494
495 con_sock_state_connecting(con);
496 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
497 O_NONBLOCK);
498 if (ret == -EINPROGRESS) {
499 dout("connect %s EINPROGRESS sk_state = %u\n",
500 ceph_pr_addr(&con->peer_addr.in_addr),
501 sock->sk->sk_state);
502 } else if (ret < 0) {
503 pr_err("connect %s error %d\n",
504 ceph_pr_addr(&con->peer_addr.in_addr), ret);
505 sock_release(sock);
506 return ret;
507 }
508
509 if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY)) {
510 int optval = 1;
511
512 ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
513 (char *)&optval, sizeof(optval));
514 if (ret)
515 pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
516 ret);
517 }
518
519 con->sock = sock;
520 return 0;
521}
522
523static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
524{
525 struct kvec iov = {buf, len};
526 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
527 int r;
528
529 iov_iter_kvec(&msg.msg_iter, READ | ITER_KVEC, &iov, 1, len);
530 r = sock_recvmsg(sock, &msg, msg.msg_flags);
531 if (r == -EAGAIN)
532 r = 0;
533 return r;
534}
535
536static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
537 int page_offset, size_t length)
538{
539 struct bio_vec bvec = {
540 .bv_page = page,
541 .bv_offset = page_offset,
542 .bv_len = length
543 };
544 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
545 int r;
546
547 BUG_ON(page_offset + length > PAGE_SIZE);
548 iov_iter_bvec(&msg.msg_iter, READ | ITER_BVEC, &bvec, 1, length);
549 r = sock_recvmsg(sock, &msg, msg.msg_flags);
550 if (r == -EAGAIN)
551 r = 0;
552 return r;
553}
554
555/*
556 * write something. @more is true if caller will be sending more data
557 * shortly.
558 */
559static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
560 size_t kvlen, size_t len, int more)
561{
562 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
563 int r;
564
565 if (more)
566 msg.msg_flags |= MSG_MORE;
567 else
568 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
569
570 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
571 if (r == -EAGAIN)
572 r = 0;
573 return r;
574}
575
576static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
577 int offset, size_t size, bool more)
578{
579 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
580 int ret;
581
582 ret = kernel_sendpage(sock, page, offset, size, flags);
583 if (ret == -EAGAIN)
584 ret = 0;
585
586 return ret;
587}
588
589static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
590 int offset, size_t size, bool more)
591{
592 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
593 struct bio_vec bvec;
594 int ret;
595
596 /* sendpage cannot properly handle pages with page_count == 0,
597 * we need to fallback to sendmsg if that's the case */
598 if (page_count(page) >= 1)
599 return __ceph_tcp_sendpage(sock, page, offset, size, more);
600
601 bvec.bv_page = page;
602 bvec.bv_offset = offset;
603 bvec.bv_len = size;
604
605 if (more)
606 msg.msg_flags |= MSG_MORE;
607 else
608 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
609
610 iov_iter_bvec(&msg.msg_iter, WRITE | ITER_BVEC, &bvec, 1, size);
611 ret = sock_sendmsg(sock, &msg);
612 if (ret == -EAGAIN)
613 ret = 0;
614
615 return ret;
616}
617
618/*
619 * Shutdown/close the socket for the given connection.
620 */
621static int con_close_socket(struct ceph_connection *con)
622{
623 int rc = 0;
624
625 dout("con_close_socket on %p sock %p\n", con, con->sock);
626 if (con->sock) {
627 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
628 sock_release(con->sock);
629 con->sock = NULL;
630 }
631
632 /*
633 * Forcibly clear the SOCK_CLOSED flag. It gets set
634 * independent of the connection mutex, and we could have
635 * received a socket close event before we had the chance to
636 * shut the socket down.
637 */
638 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
639
640 con_sock_state_closed(con);
641 return rc;
642}
643
644/*
645 * Reset a connection. Discard all incoming and outgoing messages
646 * and clear *_seq state.
647 */
648static void ceph_msg_remove(struct ceph_msg *msg)
649{
650 list_del_init(&msg->list_head);
651
652 ceph_msg_put(msg);
653}
654static void ceph_msg_remove_list(struct list_head *head)
655{
656 while (!list_empty(head)) {
657 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
658 list_head);
659 ceph_msg_remove(msg);
660 }
661}
662
663static void reset_connection(struct ceph_connection *con)
664{
665 /* reset connection, out_queue, msg_ and connect_seq */
666 /* discard existing out_queue and msg_seq */
667 dout("reset_connection %p\n", con);
668 ceph_msg_remove_list(&con->out_queue);
669 ceph_msg_remove_list(&con->out_sent);
670
671 if (con->in_msg) {
672 BUG_ON(con->in_msg->con != con);
673 ceph_msg_put(con->in_msg);
674 con->in_msg = NULL;
675 }
676
677 con->connect_seq = 0;
678 con->out_seq = 0;
679 if (con->out_msg) {
680 BUG_ON(con->out_msg->con != con);
681 ceph_msg_put(con->out_msg);
682 con->out_msg = NULL;
683 }
684 con->in_seq = 0;
685 con->in_seq_acked = 0;
686
687 con->out_skip = 0;
688}
689
690/*
691 * mark a peer down. drop any open connections.
692 */
693void ceph_con_close(struct ceph_connection *con)
694{
695 mutex_lock(&con->mutex);
696 dout("con_close %p peer %s\n", con,
697 ceph_pr_addr(&con->peer_addr.in_addr));
698 con->state = CON_STATE_CLOSED;
699
700 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
701 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
702 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
703 con_flag_clear(con, CON_FLAG_BACKOFF);
704
705 reset_connection(con);
706 con->peer_global_seq = 0;
707 cancel_con(con);
708 con_close_socket(con);
709 mutex_unlock(&con->mutex);
710}
711EXPORT_SYMBOL(ceph_con_close);
712
713/*
714 * Reopen a closed connection, with a new peer address.
715 */
716void ceph_con_open(struct ceph_connection *con,
717 __u8 entity_type, __u64 entity_num,
718 struct ceph_entity_addr *addr)
719{
720 mutex_lock(&con->mutex);
721 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
722
723 WARN_ON(con->state != CON_STATE_CLOSED);
724 con->state = CON_STATE_PREOPEN;
725
726 con->peer_name.type = (__u8) entity_type;
727 con->peer_name.num = cpu_to_le64(entity_num);
728
729 memcpy(&con->peer_addr, addr, sizeof(*addr));
730 con->delay = 0; /* reset backoff memory */
731 mutex_unlock(&con->mutex);
732 queue_con(con);
733}
734EXPORT_SYMBOL(ceph_con_open);
735
736/*
737 * return true if this connection ever successfully opened
738 */
739bool ceph_con_opened(struct ceph_connection *con)
740{
741 return con->connect_seq > 0;
742}
743
744/*
745 * initialize a new connection.
746 */
747void ceph_con_init(struct ceph_connection *con, void *private,
748 const struct ceph_connection_operations *ops,
749 struct ceph_messenger *msgr)
750{
751 dout("con_init %p\n", con);
752 memset(con, 0, sizeof(*con));
753 con->private = private;
754 con->ops = ops;
755 con->msgr = msgr;
756
757 con_sock_state_init(con);
758
759 mutex_init(&con->mutex);
760 INIT_LIST_HEAD(&con->out_queue);
761 INIT_LIST_HEAD(&con->out_sent);
762 INIT_DELAYED_WORK(&con->work, ceph_con_workfn);
763
764 con->state = CON_STATE_CLOSED;
765}
766EXPORT_SYMBOL(ceph_con_init);
767
768
769/*
770 * We maintain a global counter to order connection attempts. Get
771 * a unique seq greater than @gt.
772 */
773static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
774{
775 u32 ret;
776
777 spin_lock(&msgr->global_seq_lock);
778 if (msgr->global_seq < gt)
779 msgr->global_seq = gt;
780 ret = ++msgr->global_seq;
781 spin_unlock(&msgr->global_seq_lock);
782 return ret;
783}
784
785static void con_out_kvec_reset(struct ceph_connection *con)
786{
787 BUG_ON(con->out_skip);
788
789 con->out_kvec_left = 0;
790 con->out_kvec_bytes = 0;
791 con->out_kvec_cur = &con->out_kvec[0];
792}
793
794static void con_out_kvec_add(struct ceph_connection *con,
795 size_t size, void *data)
796{
797 int index = con->out_kvec_left;
798
799 BUG_ON(con->out_skip);
800 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
801
802 con->out_kvec[index].iov_len = size;
803 con->out_kvec[index].iov_base = data;
804 con->out_kvec_left++;
805 con->out_kvec_bytes += size;
806}
807
808/*
809 * Chop off a kvec from the end. Return residual number of bytes for
810 * that kvec, i.e. how many bytes would have been written if the kvec
811 * hadn't been nuked.
812 */
813static int con_out_kvec_skip(struct ceph_connection *con)
814{
815 int off = con->out_kvec_cur - con->out_kvec;
816 int skip = 0;
817
818 if (con->out_kvec_bytes > 0) {
819 skip = con->out_kvec[off + con->out_kvec_left - 1].iov_len;
820 BUG_ON(con->out_kvec_bytes < skip);
821 BUG_ON(!con->out_kvec_left);
822 con->out_kvec_bytes -= skip;
823 con->out_kvec_left--;
824 }
825
826 return skip;
827}
828
829#ifdef CONFIG_BLOCK
830
831/*
832 * For a bio data item, a piece is whatever remains of the next
833 * entry in the current bio iovec, or the first entry in the next
834 * bio in the list.
835 */
836static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
837 size_t length)
838{
839 struct ceph_msg_data *data = cursor->data;
840 struct ceph_bio_iter *it = &cursor->bio_iter;
841
842 cursor->resid = min_t(size_t, length, data->bio_length);
843 *it = data->bio_pos;
844 if (cursor->resid < it->iter.bi_size)
845 it->iter.bi_size = cursor->resid;
846
847 BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
848 cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter);
849}
850
851static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
852 size_t *page_offset,
853 size_t *length)
854{
855 struct bio_vec bv = bio_iter_iovec(cursor->bio_iter.bio,
856 cursor->bio_iter.iter);
857
858 *page_offset = bv.bv_offset;
859 *length = bv.bv_len;
860 return bv.bv_page;
861}
862
863static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
864 size_t bytes)
865{
866 struct ceph_bio_iter *it = &cursor->bio_iter;
867
868 BUG_ON(bytes > cursor->resid);
869 BUG_ON(bytes > bio_iter_len(it->bio, it->iter));
870 cursor->resid -= bytes;
871 bio_advance_iter(it->bio, &it->iter, bytes);
872
873 if (!cursor->resid) {
874 BUG_ON(!cursor->last_piece);
875 return false; /* no more data */
876 }
877
878 if (!bytes || (it->iter.bi_size && it->iter.bi_bvec_done))
879 return false; /* more bytes to process in this segment */
880
881 if (!it->iter.bi_size) {
882 it->bio = it->bio->bi_next;
883 it->iter = it->bio->bi_iter;
884 if (cursor->resid < it->iter.bi_size)
885 it->iter.bi_size = cursor->resid;
886 }
887
888 BUG_ON(cursor->last_piece);
889 BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
890 cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter);
891 return true;
892}
893#endif /* CONFIG_BLOCK */
894
895static void ceph_msg_data_bvecs_cursor_init(struct ceph_msg_data_cursor *cursor,
896 size_t length)
897{
898 struct ceph_msg_data *data = cursor->data;
899 struct bio_vec *bvecs = data->bvec_pos.bvecs;
900
901 cursor->resid = min_t(size_t, length, data->bvec_pos.iter.bi_size);
902 cursor->bvec_iter = data->bvec_pos.iter;
903 cursor->bvec_iter.bi_size = cursor->resid;
904
905 BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
906 cursor->last_piece =
907 cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter);
908}
909
910static struct page *ceph_msg_data_bvecs_next(struct ceph_msg_data_cursor *cursor,
911 size_t *page_offset,
912 size_t *length)
913{
914 struct bio_vec bv = bvec_iter_bvec(cursor->data->bvec_pos.bvecs,
915 cursor->bvec_iter);
916
917 *page_offset = bv.bv_offset;
918 *length = bv.bv_len;
919 return bv.bv_page;
920}
921
922static bool ceph_msg_data_bvecs_advance(struct ceph_msg_data_cursor *cursor,
923 size_t bytes)
924{
925 struct bio_vec *bvecs = cursor->data->bvec_pos.bvecs;
926
927 BUG_ON(bytes > cursor->resid);
928 BUG_ON(bytes > bvec_iter_len(bvecs, cursor->bvec_iter));
929 cursor->resid -= bytes;
930 bvec_iter_advance(bvecs, &cursor->bvec_iter, bytes);
931
932 if (!cursor->resid) {
933 BUG_ON(!cursor->last_piece);
934 return false; /* no more data */
935 }
936
937 if (!bytes || cursor->bvec_iter.bi_bvec_done)
938 return false; /* more bytes to process in this segment */
939
940 BUG_ON(cursor->last_piece);
941 BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
942 cursor->last_piece =
943 cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter);
944 return true;
945}
946
947/*
948 * For a page array, a piece comes from the first page in the array
949 * that has not already been fully consumed.
950 */
951static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
952 size_t length)
953{
954 struct ceph_msg_data *data = cursor->data;
955 int page_count;
956
957 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
958
959 BUG_ON(!data->pages);
960 BUG_ON(!data->length);
961
962 cursor->resid = min(length, data->length);
963 page_count = calc_pages_for(data->alignment, (u64)data->length);
964 cursor->page_offset = data->alignment & ~PAGE_MASK;
965 cursor->page_index = 0;
966 BUG_ON(page_count > (int)USHRT_MAX);
967 cursor->page_count = (unsigned short)page_count;
968 BUG_ON(length > SIZE_MAX - cursor->page_offset);
969 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
970}
971
972static struct page *
973ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
974 size_t *page_offset, size_t *length)
975{
976 struct ceph_msg_data *data = cursor->data;
977
978 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
979
980 BUG_ON(cursor->page_index >= cursor->page_count);
981 BUG_ON(cursor->page_offset >= PAGE_SIZE);
982
983 *page_offset = cursor->page_offset;
984 if (cursor->last_piece)
985 *length = cursor->resid;
986 else
987 *length = PAGE_SIZE - *page_offset;
988
989 return data->pages[cursor->page_index];
990}
991
992static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
993 size_t bytes)
994{
995 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
996
997 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
998
999 /* Advance the cursor page offset */
1000
1001 cursor->resid -= bytes;
1002 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
1003 if (!bytes || cursor->page_offset)
1004 return false; /* more bytes to process in the current page */
1005
1006 if (!cursor->resid)
1007 return false; /* no more data */
1008
1009 /* Move on to the next page; offset is already at 0 */
1010
1011 BUG_ON(cursor->page_index >= cursor->page_count);
1012 cursor->page_index++;
1013 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1014
1015 return true;
1016}
1017
1018/*
1019 * For a pagelist, a piece is whatever remains to be consumed in the
1020 * first page in the list, or the front of the next page.
1021 */
1022static void
1023ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
1024 size_t length)
1025{
1026 struct ceph_msg_data *data = cursor->data;
1027 struct ceph_pagelist *pagelist;
1028 struct page *page;
1029
1030 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1031
1032 pagelist = data->pagelist;
1033 BUG_ON(!pagelist);
1034
1035 if (!length)
1036 return; /* pagelist can be assigned but empty */
1037
1038 BUG_ON(list_empty(&pagelist->head));
1039 page = list_first_entry(&pagelist->head, struct page, lru);
1040
1041 cursor->resid = min(length, pagelist->length);
1042 cursor->page = page;
1043 cursor->offset = 0;
1044 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1045}
1046
1047static struct page *
1048ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
1049 size_t *page_offset, size_t *length)
1050{
1051 struct ceph_msg_data *data = cursor->data;
1052 struct ceph_pagelist *pagelist;
1053
1054 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1055
1056 pagelist = data->pagelist;
1057 BUG_ON(!pagelist);
1058
1059 BUG_ON(!cursor->page);
1060 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1061
1062 /* offset of first page in pagelist is always 0 */
1063 *page_offset = cursor->offset & ~PAGE_MASK;
1064 if (cursor->last_piece)
1065 *length = cursor->resid;
1066 else
1067 *length = PAGE_SIZE - *page_offset;
1068
1069 return cursor->page;
1070}
1071
1072static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
1073 size_t bytes)
1074{
1075 struct ceph_msg_data *data = cursor->data;
1076 struct ceph_pagelist *pagelist;
1077
1078 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1079
1080 pagelist = data->pagelist;
1081 BUG_ON(!pagelist);
1082
1083 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1084 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1085
1086 /* Advance the cursor offset */
1087
1088 cursor->resid -= bytes;
1089 cursor->offset += bytes;
1090 /* offset of first page in pagelist is always 0 */
1091 if (!bytes || cursor->offset & ~PAGE_MASK)
1092 return false; /* more bytes to process in the current page */
1093
1094 if (!cursor->resid)
1095 return false; /* no more data */
1096
1097 /* Move on to the next page */
1098
1099 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1100 cursor->page = list_next_entry(cursor->page, lru);
1101 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1102
1103 return true;
1104}
1105
1106/*
1107 * Message data is handled (sent or received) in pieces, where each
1108 * piece resides on a single page. The network layer might not
1109 * consume an entire piece at once. A data item's cursor keeps
1110 * track of which piece is next to process and how much remains to
1111 * be processed in that piece. It also tracks whether the current
1112 * piece is the last one in the data item.
1113 */
1114static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1115{
1116 size_t length = cursor->total_resid;
1117
1118 switch (cursor->data->type) {
1119 case CEPH_MSG_DATA_PAGELIST:
1120 ceph_msg_data_pagelist_cursor_init(cursor, length);
1121 break;
1122 case CEPH_MSG_DATA_PAGES:
1123 ceph_msg_data_pages_cursor_init(cursor, length);
1124 break;
1125#ifdef CONFIG_BLOCK
1126 case CEPH_MSG_DATA_BIO:
1127 ceph_msg_data_bio_cursor_init(cursor, length);
1128 break;
1129#endif /* CONFIG_BLOCK */
1130 case CEPH_MSG_DATA_BVECS:
1131 ceph_msg_data_bvecs_cursor_init(cursor, length);
1132 break;
1133 case CEPH_MSG_DATA_NONE:
1134 default:
1135 /* BUG(); */
1136 break;
1137 }
1138 cursor->need_crc = true;
1139}
1140
1141static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1142{
1143 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1144 struct ceph_msg_data *data;
1145
1146 BUG_ON(!length);
1147 BUG_ON(length > msg->data_length);
1148 BUG_ON(list_empty(&msg->data));
1149
1150 cursor->data_head = &msg->data;
1151 cursor->total_resid = length;
1152 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1153 cursor->data = data;
1154
1155 __ceph_msg_data_cursor_init(cursor);
1156}
1157
1158/*
1159 * Return the page containing the next piece to process for a given
1160 * data item, and supply the page offset and length of that piece.
1161 * Indicate whether this is the last piece in this data item.
1162 */
1163static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1164 size_t *page_offset, size_t *length,
1165 bool *last_piece)
1166{
1167 struct page *page;
1168
1169 switch (cursor->data->type) {
1170 case CEPH_MSG_DATA_PAGELIST:
1171 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1172 break;
1173 case CEPH_MSG_DATA_PAGES:
1174 page = ceph_msg_data_pages_next(cursor, page_offset, length);
1175 break;
1176#ifdef CONFIG_BLOCK
1177 case CEPH_MSG_DATA_BIO:
1178 page = ceph_msg_data_bio_next(cursor, page_offset, length);
1179 break;
1180#endif /* CONFIG_BLOCK */
1181 case CEPH_MSG_DATA_BVECS:
1182 page = ceph_msg_data_bvecs_next(cursor, page_offset, length);
1183 break;
1184 case CEPH_MSG_DATA_NONE:
1185 default:
1186 page = NULL;
1187 break;
1188 }
1189
1190 BUG_ON(!page);
1191 BUG_ON(*page_offset + *length > PAGE_SIZE);
1192 BUG_ON(!*length);
1193 BUG_ON(*length > cursor->resid);
1194 if (last_piece)
1195 *last_piece = cursor->last_piece;
1196
1197 return page;
1198}
1199
1200/*
1201 * Returns true if the result moves the cursor on to the next piece
1202 * of the data item.
1203 */
1204static void ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1205 size_t bytes)
1206{
1207 bool new_piece;
1208
1209 BUG_ON(bytes > cursor->resid);
1210 switch (cursor->data->type) {
1211 case CEPH_MSG_DATA_PAGELIST:
1212 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1213 break;
1214 case CEPH_MSG_DATA_PAGES:
1215 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1216 break;
1217#ifdef CONFIG_BLOCK
1218 case CEPH_MSG_DATA_BIO:
1219 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1220 break;
1221#endif /* CONFIG_BLOCK */
1222 case CEPH_MSG_DATA_BVECS:
1223 new_piece = ceph_msg_data_bvecs_advance(cursor, bytes);
1224 break;
1225 case CEPH_MSG_DATA_NONE:
1226 default:
1227 BUG();
1228 break;
1229 }
1230 cursor->total_resid -= bytes;
1231
1232 if (!cursor->resid && cursor->total_resid) {
1233 WARN_ON(!cursor->last_piece);
1234 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1235 cursor->data = list_next_entry(cursor->data, links);
1236 __ceph_msg_data_cursor_init(cursor);
1237 new_piece = true;
1238 }
1239 cursor->need_crc = new_piece;
1240}
1241
1242static size_t sizeof_footer(struct ceph_connection *con)
1243{
1244 return (con->peer_features & CEPH_FEATURE_MSG_AUTH) ?
1245 sizeof(struct ceph_msg_footer) :
1246 sizeof(struct ceph_msg_footer_old);
1247}
1248
1249static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
1250{
1251 BUG_ON(!msg);
1252 BUG_ON(!data_len);
1253
1254 /* Initialize data cursor */
1255
1256 ceph_msg_data_cursor_init(msg, (size_t)data_len);
1257}
1258
1259/*
1260 * Prepare footer for currently outgoing message, and finish things
1261 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1262 */
1263static void prepare_write_message_footer(struct ceph_connection *con)
1264{
1265 struct ceph_msg *m = con->out_msg;
1266
1267 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1268
1269 dout("prepare_write_message_footer %p\n", con);
1270 con_out_kvec_add(con, sizeof_footer(con), &m->footer);
1271 if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
1272 if (con->ops->sign_message)
1273 con->ops->sign_message(m);
1274 else
1275 m->footer.sig = 0;
1276 } else {
1277 m->old_footer.flags = m->footer.flags;
1278 }
1279 con->out_more = m->more_to_follow;
1280 con->out_msg_done = true;
1281}
1282
1283/*
1284 * Prepare headers for the next outgoing message.
1285 */
1286static void prepare_write_message(struct ceph_connection *con)
1287{
1288 struct ceph_msg *m;
1289 u32 crc;
1290
1291 con_out_kvec_reset(con);
1292 con->out_msg_done = false;
1293
1294 /* Sneak an ack in there first? If we can get it into the same
1295 * TCP packet that's a good thing. */
1296 if (con->in_seq > con->in_seq_acked) {
1297 con->in_seq_acked = con->in_seq;
1298 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1299 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1300 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1301 &con->out_temp_ack);
1302 }
1303
1304 BUG_ON(list_empty(&con->out_queue));
1305 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1306 con->out_msg = m;
1307 BUG_ON(m->con != con);
1308
1309 /* put message on sent list */
1310 ceph_msg_get(m);
1311 list_move_tail(&m->list_head, &con->out_sent);
1312
1313 /*
1314 * only assign outgoing seq # if we haven't sent this message
1315 * yet. if it is requeued, resend with it's original seq.
1316 */
1317 if (m->needs_out_seq) {
1318 m->hdr.seq = cpu_to_le64(++con->out_seq);
1319 m->needs_out_seq = false;
1320
1321 if (con->ops->reencode_message)
1322 con->ops->reencode_message(m);
1323 }
1324
1325 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1326 m, con->out_seq, le16_to_cpu(m->hdr.type),
1327 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1328 m->data_length);
1329 WARN_ON(m->front.iov_len != le32_to_cpu(m->hdr.front_len));
1330 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
1331
1332 /* tag + hdr + front + middle */
1333 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1334 con_out_kvec_add(con, sizeof(con->out_hdr), &con->out_hdr);
1335 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1336
1337 if (m->middle)
1338 con_out_kvec_add(con, m->middle->vec.iov_len,
1339 m->middle->vec.iov_base);
1340
1341 /* fill in hdr crc and finalize hdr */
1342 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1343 con->out_msg->hdr.crc = cpu_to_le32(crc);
1344 memcpy(&con->out_hdr, &con->out_msg->hdr, sizeof(con->out_hdr));
1345
1346 /* fill in front and middle crc, footer */
1347 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1348 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1349 if (m->middle) {
1350 crc = crc32c(0, m->middle->vec.iov_base,
1351 m->middle->vec.iov_len);
1352 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1353 } else
1354 con->out_msg->footer.middle_crc = 0;
1355 dout("%s front_crc %u middle_crc %u\n", __func__,
1356 le32_to_cpu(con->out_msg->footer.front_crc),
1357 le32_to_cpu(con->out_msg->footer.middle_crc));
1358 con->out_msg->footer.flags = 0;
1359
1360 /* is there a data payload? */
1361 con->out_msg->footer.data_crc = 0;
1362 if (m->data_length) {
1363 prepare_message_data(con->out_msg, m->data_length);
1364 con->out_more = 1; /* data + footer will follow */
1365 } else {
1366 /* no, queue up footer too and be done */
1367 prepare_write_message_footer(con);
1368 }
1369
1370 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1371}
1372
1373/*
1374 * Prepare an ack.
1375 */
1376static void prepare_write_ack(struct ceph_connection *con)
1377{
1378 dout("prepare_write_ack %p %llu -> %llu\n", con,
1379 con->in_seq_acked, con->in_seq);
1380 con->in_seq_acked = con->in_seq;
1381
1382 con_out_kvec_reset(con);
1383
1384 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1385
1386 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1387 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1388 &con->out_temp_ack);
1389
1390 con->out_more = 1; /* more will follow.. eventually.. */
1391 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1392}
1393
1394/*
1395 * Prepare to share the seq during handshake
1396 */
1397static void prepare_write_seq(struct ceph_connection *con)
1398{
1399 dout("prepare_write_seq %p %llu -> %llu\n", con,
1400 con->in_seq_acked, con->in_seq);
1401 con->in_seq_acked = con->in_seq;
1402
1403 con_out_kvec_reset(con);
1404
1405 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1406 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1407 &con->out_temp_ack);
1408
1409 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1410}
1411
1412/*
1413 * Prepare to write keepalive byte.
1414 */
1415static void prepare_write_keepalive(struct ceph_connection *con)
1416{
1417 dout("prepare_write_keepalive %p\n", con);
1418 con_out_kvec_reset(con);
1419 if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
1420 struct timespec now;
1421
1422 ktime_get_real_ts(&now);
1423 con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
1424 ceph_encode_timespec(&con->out_temp_keepalive2, &now);
1425 con_out_kvec_add(con, sizeof(con->out_temp_keepalive2),
1426 &con->out_temp_keepalive2);
1427 } else {
1428 con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive);
1429 }
1430 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1431}
1432
1433/*
1434 * Connection negotiation.
1435 */
1436
1437static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1438 int *auth_proto)
1439{
1440 struct ceph_auth_handshake *auth;
1441
1442 if (!con->ops->get_authorizer) {
1443 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1444 con->out_connect.authorizer_len = 0;
1445 return NULL;
1446 }
1447
1448 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
1449 if (IS_ERR(auth))
1450 return auth;
1451
1452 con->auth_reply_buf = auth->authorizer_reply_buf;
1453 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
1454 return auth;
1455}
1456
1457/*
1458 * We connected to a peer and are saying hello.
1459 */
1460static void prepare_write_banner(struct ceph_connection *con)
1461{
1462 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1463 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1464 &con->msgr->my_enc_addr);
1465
1466 con->out_more = 0;
1467 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1468}
1469
1470static int prepare_write_connect(struct ceph_connection *con)
1471{
1472 unsigned int global_seq = get_global_seq(con->msgr, 0);
1473 int proto;
1474 int auth_proto;
1475 struct ceph_auth_handshake *auth;
1476
1477 switch (con->peer_name.type) {
1478 case CEPH_ENTITY_TYPE_MON:
1479 proto = CEPH_MONC_PROTOCOL;
1480 break;
1481 case CEPH_ENTITY_TYPE_OSD:
1482 proto = CEPH_OSDC_PROTOCOL;
1483 break;
1484 case CEPH_ENTITY_TYPE_MDS:
1485 proto = CEPH_MDSC_PROTOCOL;
1486 break;
1487 default:
1488 BUG();
1489 }
1490
1491 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1492 con->connect_seq, global_seq, proto);
1493
1494 con->out_connect.features =
1495 cpu_to_le64(from_msgr(con->msgr)->supported_features);
1496 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1497 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1498 con->out_connect.global_seq = cpu_to_le32(global_seq);
1499 con->out_connect.protocol_version = cpu_to_le32(proto);
1500 con->out_connect.flags = 0;
1501
1502 auth_proto = CEPH_AUTH_UNKNOWN;
1503 auth = get_connect_authorizer(con, &auth_proto);
1504 if (IS_ERR(auth))
1505 return PTR_ERR(auth);
1506
1507 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1508 con->out_connect.authorizer_len = auth ?
1509 cpu_to_le32(auth->authorizer_buf_len) : 0;
1510
1511 con_out_kvec_add(con, sizeof (con->out_connect),
1512 &con->out_connect);
1513 if (auth && auth->authorizer_buf_len)
1514 con_out_kvec_add(con, auth->authorizer_buf_len,
1515 auth->authorizer_buf);
1516
1517 con->out_more = 0;
1518 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1519
1520 return 0;
1521}
1522
1523/*
1524 * write as much of pending kvecs to the socket as we can.
1525 * 1 -> done
1526 * 0 -> socket full, but more to do
1527 * <0 -> error
1528 */
1529static int write_partial_kvec(struct ceph_connection *con)
1530{
1531 int ret;
1532
1533 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1534 while (con->out_kvec_bytes > 0) {
1535 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1536 con->out_kvec_left, con->out_kvec_bytes,
1537 con->out_more);
1538 if (ret <= 0)
1539 goto out;
1540 con->out_kvec_bytes -= ret;
1541 if (con->out_kvec_bytes == 0)
1542 break; /* done */
1543
1544 /* account for full iov entries consumed */
1545 while (ret >= con->out_kvec_cur->iov_len) {
1546 BUG_ON(!con->out_kvec_left);
1547 ret -= con->out_kvec_cur->iov_len;
1548 con->out_kvec_cur++;
1549 con->out_kvec_left--;
1550 }
1551 /* and for a partially-consumed entry */
1552 if (ret) {
1553 con->out_kvec_cur->iov_len -= ret;
1554 con->out_kvec_cur->iov_base += ret;
1555 }
1556 }
1557 con->out_kvec_left = 0;
1558 ret = 1;
1559out:
1560 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1561 con->out_kvec_bytes, con->out_kvec_left, ret);
1562 return ret; /* done! */
1563}
1564
1565static u32 ceph_crc32c_page(u32 crc, struct page *page,
1566 unsigned int page_offset,
1567 unsigned int length)
1568{
1569 char *kaddr;
1570
1571 kaddr = kmap(page);
1572 BUG_ON(kaddr == NULL);
1573 crc = crc32c(crc, kaddr + page_offset, length);
1574 kunmap(page);
1575
1576 return crc;
1577}
1578/*
1579 * Write as much message data payload as we can. If we finish, queue
1580 * up the footer.
1581 * 1 -> done, footer is now queued in out_kvec[].
1582 * 0 -> socket full, but more to do
1583 * <0 -> error
1584 */
1585static int write_partial_message_data(struct ceph_connection *con)
1586{
1587 struct ceph_msg *msg = con->out_msg;
1588 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1589 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
1590 u32 crc;
1591
1592 dout("%s %p msg %p\n", __func__, con, msg);
1593
1594 if (list_empty(&msg->data))
1595 return -EINVAL;
1596
1597 /*
1598 * Iterate through each page that contains data to be
1599 * written, and send as much as possible for each.
1600 *
1601 * If we are calculating the data crc (the default), we will
1602 * need to map the page. If we have no pages, they have
1603 * been revoked, so use the zero page.
1604 */
1605 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1606 while (cursor->total_resid) {
1607 struct page *page;
1608 size_t page_offset;
1609 size_t length;
1610 bool last_piece;
1611 int ret;
1612
1613 if (!cursor->resid) {
1614 ceph_msg_data_advance(cursor, 0);
1615 continue;
1616 }
1617
1618 page = ceph_msg_data_next(cursor, &page_offset, &length,
1619 &last_piece);
1620 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
1621 length, !last_piece);
1622 if (ret <= 0) {
1623 if (do_datacrc)
1624 msg->footer.data_crc = cpu_to_le32(crc);
1625
1626 return ret;
1627 }
1628 if (do_datacrc && cursor->need_crc)
1629 crc = ceph_crc32c_page(crc, page, page_offset, length);
1630 ceph_msg_data_advance(cursor, (size_t)ret);
1631 }
1632
1633 dout("%s %p msg %p done\n", __func__, con, msg);
1634
1635 /* prepare and queue up footer, too */
1636 if (do_datacrc)
1637 msg->footer.data_crc = cpu_to_le32(crc);
1638 else
1639 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1640 con_out_kvec_reset(con);
1641 prepare_write_message_footer(con);
1642
1643 return 1; /* must return > 0 to indicate success */
1644}
1645
1646/*
1647 * write some zeros
1648 */
1649static int write_partial_skip(struct ceph_connection *con)
1650{
1651 int ret;
1652
1653 dout("%s %p %d left\n", __func__, con, con->out_skip);
1654 while (con->out_skip > 0) {
1655 size_t size = min(con->out_skip, (int) PAGE_SIZE);
1656
1657 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
1658 if (ret <= 0)
1659 goto out;
1660 con->out_skip -= ret;
1661 }
1662 ret = 1;
1663out:
1664 return ret;
1665}
1666
1667/*
1668 * Prepare to read connection handshake, or an ack.
1669 */
1670static void prepare_read_banner(struct ceph_connection *con)
1671{
1672 dout("prepare_read_banner %p\n", con);
1673 con->in_base_pos = 0;
1674}
1675
1676static void prepare_read_connect(struct ceph_connection *con)
1677{
1678 dout("prepare_read_connect %p\n", con);
1679 con->in_base_pos = 0;
1680}
1681
1682static void prepare_read_ack(struct ceph_connection *con)
1683{
1684 dout("prepare_read_ack %p\n", con);
1685 con->in_base_pos = 0;
1686}
1687
1688static void prepare_read_seq(struct ceph_connection *con)
1689{
1690 dout("prepare_read_seq %p\n", con);
1691 con->in_base_pos = 0;
1692 con->in_tag = CEPH_MSGR_TAG_SEQ;
1693}
1694
1695static void prepare_read_tag(struct ceph_connection *con)
1696{
1697 dout("prepare_read_tag %p\n", con);
1698 con->in_base_pos = 0;
1699 con->in_tag = CEPH_MSGR_TAG_READY;
1700}
1701
1702static void prepare_read_keepalive_ack(struct ceph_connection *con)
1703{
1704 dout("prepare_read_keepalive_ack %p\n", con);
1705 con->in_base_pos = 0;
1706}
1707
1708/*
1709 * Prepare to read a message.
1710 */
1711static int prepare_read_message(struct ceph_connection *con)
1712{
1713 dout("prepare_read_message %p\n", con);
1714 BUG_ON(con->in_msg != NULL);
1715 con->in_base_pos = 0;
1716 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1717 return 0;
1718}
1719
1720
1721static int read_partial(struct ceph_connection *con,
1722 int end, int size, void *object)
1723{
1724 while (con->in_base_pos < end) {
1725 int left = end - con->in_base_pos;
1726 int have = size - left;
1727 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1728 if (ret <= 0)
1729 return ret;
1730 con->in_base_pos += ret;
1731 }
1732 return 1;
1733}
1734
1735
1736/*
1737 * Read all or part of the connect-side handshake on a new connection
1738 */
1739static int read_partial_banner(struct ceph_connection *con)
1740{
1741 int size;
1742 int end;
1743 int ret;
1744
1745 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1746
1747 /* peer's banner */
1748 size = strlen(CEPH_BANNER);
1749 end = size;
1750 ret = read_partial(con, end, size, con->in_banner);
1751 if (ret <= 0)
1752 goto out;
1753
1754 size = sizeof (con->actual_peer_addr);
1755 end += size;
1756 ret = read_partial(con, end, size, &con->actual_peer_addr);
1757 if (ret <= 0)
1758 goto out;
1759
1760 size = sizeof (con->peer_addr_for_me);
1761 end += size;
1762 ret = read_partial(con, end, size, &con->peer_addr_for_me);
1763 if (ret <= 0)
1764 goto out;
1765
1766out:
1767 return ret;
1768}
1769
1770static int read_partial_connect(struct ceph_connection *con)
1771{
1772 int size;
1773 int end;
1774 int ret;
1775
1776 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1777
1778 size = sizeof (con->in_reply);
1779 end = size;
1780 ret = read_partial(con, end, size, &con->in_reply);
1781 if (ret <= 0)
1782 goto out;
1783
1784 size = le32_to_cpu(con->in_reply.authorizer_len);
1785 end += size;
1786 ret = read_partial(con, end, size, con->auth_reply_buf);
1787 if (ret <= 0)
1788 goto out;
1789
1790 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1791 con, (int)con->in_reply.tag,
1792 le32_to_cpu(con->in_reply.connect_seq),
1793 le32_to_cpu(con->in_reply.global_seq));
1794out:
1795 return ret;
1796
1797}
1798
1799/*
1800 * Verify the hello banner looks okay.
1801 */
1802static int verify_hello(struct ceph_connection *con)
1803{
1804 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1805 pr_err("connect to %s got bad banner\n",
1806 ceph_pr_addr(&con->peer_addr.in_addr));
1807 con->error_msg = "protocol error, bad banner";
1808 return -1;
1809 }
1810 return 0;
1811}
1812
1813static bool addr_is_blank(struct sockaddr_storage *ss)
1814{
1815 struct in_addr *addr = &((struct sockaddr_in *)ss)->sin_addr;
1816 struct in6_addr *addr6 = &((struct sockaddr_in6 *)ss)->sin6_addr;
1817
1818 switch (ss->ss_family) {
1819 case AF_INET:
1820 return addr->s_addr == htonl(INADDR_ANY);
1821 case AF_INET6:
1822 return ipv6_addr_any(addr6);
1823 default:
1824 return true;
1825 }
1826}
1827
1828static int addr_port(struct sockaddr_storage *ss)
1829{
1830 switch (ss->ss_family) {
1831 case AF_INET:
1832 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1833 case AF_INET6:
1834 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1835 }
1836 return 0;
1837}
1838
1839static void addr_set_port(struct sockaddr_storage *ss, int p)
1840{
1841 switch (ss->ss_family) {
1842 case AF_INET:
1843 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1844 break;
1845 case AF_INET6:
1846 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1847 break;
1848 }
1849}
1850
1851/*
1852 * Unlike other *_pton function semantics, zero indicates success.
1853 */
1854static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1855 char delim, const char **ipend)
1856{
1857 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1858 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1859
1860 memset(ss, 0, sizeof(*ss));
1861
1862 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1863 ss->ss_family = AF_INET;
1864 return 0;
1865 }
1866
1867 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1868 ss->ss_family = AF_INET6;
1869 return 0;
1870 }
1871
1872 return -EINVAL;
1873}
1874
1875/*
1876 * Extract hostname string and resolve using kernel DNS facility.
1877 */
1878#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1879static int ceph_dns_resolve_name(const char *name, size_t namelen,
1880 struct sockaddr_storage *ss, char delim, const char **ipend)
1881{
1882 const char *end, *delim_p;
1883 char *colon_p, *ip_addr = NULL;
1884 int ip_len, ret;
1885
1886 /*
1887 * The end of the hostname occurs immediately preceding the delimiter or
1888 * the port marker (':') where the delimiter takes precedence.
1889 */
1890 delim_p = memchr(name, delim, namelen);
1891 colon_p = memchr(name, ':', namelen);
1892
1893 if (delim_p && colon_p)
1894 end = delim_p < colon_p ? delim_p : colon_p;
1895 else if (!delim_p && colon_p)
1896 end = colon_p;
1897 else {
1898 end = delim_p;
1899 if (!end) /* case: hostname:/ */
1900 end = name + namelen;
1901 }
1902
1903 if (end <= name)
1904 return -EINVAL;
1905
1906 /* do dns_resolve upcall */
1907 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1908 if (ip_len > 0)
1909 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1910 else
1911 ret = -ESRCH;
1912
1913 kfree(ip_addr);
1914
1915 *ipend = end;
1916
1917 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1918 ret, ret ? "failed" : ceph_pr_addr(ss));
1919
1920 return ret;
1921}
1922#else
1923static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1924 struct sockaddr_storage *ss, char delim, const char **ipend)
1925{
1926 return -EINVAL;
1927}
1928#endif
1929
1930/*
1931 * Parse a server name (IP or hostname). If a valid IP address is not found
1932 * then try to extract a hostname to resolve using userspace DNS upcall.
1933 */
1934static int ceph_parse_server_name(const char *name, size_t namelen,
1935 struct sockaddr_storage *ss, char delim, const char **ipend)
1936{
1937 int ret;
1938
1939 ret = ceph_pton(name, namelen, ss, delim, ipend);
1940 if (ret)
1941 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1942
1943 return ret;
1944}
1945
1946/*
1947 * Parse an ip[:port] list into an addr array. Use the default
1948 * monitor port if a port isn't specified.
1949 */
1950int ceph_parse_ips(const char *c, const char *end,
1951 struct ceph_entity_addr *addr,
1952 int max_count, int *count)
1953{
1954 int i, ret = -EINVAL;
1955 const char *p = c;
1956
1957 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1958 for (i = 0; i < max_count; i++) {
1959 const char *ipend;
1960 struct sockaddr_storage *ss = &addr[i].in_addr;
1961 int port;
1962 char delim = ',';
1963
1964 if (*p == '[') {
1965 delim = ']';
1966 p++;
1967 }
1968
1969 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1970 if (ret)
1971 goto bad;
1972 ret = -EINVAL;
1973
1974 p = ipend;
1975
1976 if (delim == ']') {
1977 if (*p != ']') {
1978 dout("missing matching ']'\n");
1979 goto bad;
1980 }
1981 p++;
1982 }
1983
1984 /* port? */
1985 if (p < end && *p == ':') {
1986 port = 0;
1987 p++;
1988 while (p < end && *p >= '0' && *p <= '9') {
1989 port = (port * 10) + (*p - '0');
1990 p++;
1991 }
1992 if (port == 0)
1993 port = CEPH_MON_PORT;
1994 else if (port > 65535)
1995 goto bad;
1996 } else {
1997 port = CEPH_MON_PORT;
1998 }
1999
2000 addr_set_port(ss, port);
2001
2002 dout("parse_ips got %s\n", ceph_pr_addr(ss));
2003
2004 if (p == end)
2005 break;
2006 if (*p != ',')
2007 goto bad;
2008 p++;
2009 }
2010
2011 if (p != end)
2012 goto bad;
2013
2014 if (count)
2015 *count = i + 1;
2016 return 0;
2017
2018bad:
2019 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
2020 return ret;
2021}
2022EXPORT_SYMBOL(ceph_parse_ips);
2023
2024static int process_banner(struct ceph_connection *con)
2025{
2026 dout("process_banner on %p\n", con);
2027
2028 if (verify_hello(con) < 0)
2029 return -1;
2030
2031 ceph_decode_addr(&con->actual_peer_addr);
2032 ceph_decode_addr(&con->peer_addr_for_me);
2033
2034 /*
2035 * Make sure the other end is who we wanted. note that the other
2036 * end may not yet know their ip address, so if it's 0.0.0.0, give
2037 * them the benefit of the doubt.
2038 */
2039 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
2040 sizeof(con->peer_addr)) != 0 &&
2041 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
2042 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
2043 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
2044 ceph_pr_addr(&con->peer_addr.in_addr),
2045 (int)le32_to_cpu(con->peer_addr.nonce),
2046 ceph_pr_addr(&con->actual_peer_addr.in_addr),
2047 (int)le32_to_cpu(con->actual_peer_addr.nonce));
2048 con->error_msg = "wrong peer at address";
2049 return -1;
2050 }
2051
2052 /*
2053 * did we learn our address?
2054 */
2055 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
2056 int port = addr_port(&con->msgr->inst.addr.in_addr);
2057
2058 memcpy(&con->msgr->inst.addr.in_addr,
2059 &con->peer_addr_for_me.in_addr,
2060 sizeof(con->peer_addr_for_me.in_addr));
2061 addr_set_port(&con->msgr->inst.addr.in_addr, port);
2062 encode_my_addr(con->msgr);
2063 dout("process_banner learned my addr is %s\n",
2064 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
2065 }
2066
2067 return 0;
2068}
2069
2070static int process_connect(struct ceph_connection *con)
2071{
2072 u64 sup_feat = from_msgr(con->msgr)->supported_features;
2073 u64 req_feat = from_msgr(con->msgr)->required_features;
2074 u64 server_feat = le64_to_cpu(con->in_reply.features);
2075 int ret;
2076
2077 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
2078
2079 if (con->auth_reply_buf) {
2080 /*
2081 * Any connection that defines ->get_authorizer()
2082 * should also define ->verify_authorizer_reply().
2083 * See get_connect_authorizer().
2084 */
2085 ret = con->ops->verify_authorizer_reply(con);
2086 if (ret < 0) {
2087 con->error_msg = "bad authorize reply";
2088 return ret;
2089 }
2090 }
2091
2092 switch (con->in_reply.tag) {
2093 case CEPH_MSGR_TAG_FEATURES:
2094 pr_err("%s%lld %s feature set mismatch,"
2095 " my %llx < server's %llx, missing %llx\n",
2096 ENTITY_NAME(con->peer_name),
2097 ceph_pr_addr(&con->peer_addr.in_addr),
2098 sup_feat, server_feat, server_feat & ~sup_feat);
2099 con->error_msg = "missing required protocol features";
2100 reset_connection(con);
2101 return -1;
2102
2103 case CEPH_MSGR_TAG_BADPROTOVER:
2104 pr_err("%s%lld %s protocol version mismatch,"
2105 " my %d != server's %d\n",
2106 ENTITY_NAME(con->peer_name),
2107 ceph_pr_addr(&con->peer_addr.in_addr),
2108 le32_to_cpu(con->out_connect.protocol_version),
2109 le32_to_cpu(con->in_reply.protocol_version));
2110 con->error_msg = "protocol version mismatch";
2111 reset_connection(con);
2112 return -1;
2113
2114 case CEPH_MSGR_TAG_BADAUTHORIZER:
2115 con->auth_retry++;
2116 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2117 con->auth_retry);
2118 if (con->auth_retry == 2) {
2119 con->error_msg = "connect authorization failure";
2120 return -1;
2121 }
2122 con_out_kvec_reset(con);
2123 ret = prepare_write_connect(con);
2124 if (ret < 0)
2125 return ret;
2126 prepare_read_connect(con);
2127 break;
2128
2129 case CEPH_MSGR_TAG_RESETSESSION:
2130 /*
2131 * If we connected with a large connect_seq but the peer
2132 * has no record of a session with us (no connection, or
2133 * connect_seq == 0), they will send RESETSESION to indicate
2134 * that they must have reset their session, and may have
2135 * dropped messages.
2136 */
2137 dout("process_connect got RESET peer seq %u\n",
2138 le32_to_cpu(con->in_reply.connect_seq));
2139 pr_err("%s%lld %s connection reset\n",
2140 ENTITY_NAME(con->peer_name),
2141 ceph_pr_addr(&con->peer_addr.in_addr));
2142 reset_connection(con);
2143 con_out_kvec_reset(con);
2144 ret = prepare_write_connect(con);
2145 if (ret < 0)
2146 return ret;
2147 prepare_read_connect(con);
2148
2149 /* Tell ceph about it. */
2150 mutex_unlock(&con->mutex);
2151 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2152 if (con->ops->peer_reset)
2153 con->ops->peer_reset(con);
2154 mutex_lock(&con->mutex);
2155 if (con->state != CON_STATE_NEGOTIATING)
2156 return -EAGAIN;
2157 break;
2158
2159 case CEPH_MSGR_TAG_RETRY_SESSION:
2160 /*
2161 * If we sent a smaller connect_seq than the peer has, try
2162 * again with a larger value.
2163 */
2164 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2165 le32_to_cpu(con->out_connect.connect_seq),
2166 le32_to_cpu(con->in_reply.connect_seq));
2167 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2168 con_out_kvec_reset(con);
2169 ret = prepare_write_connect(con);
2170 if (ret < 0)
2171 return ret;
2172 prepare_read_connect(con);
2173 break;
2174
2175 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2176 /*
2177 * If we sent a smaller global_seq than the peer has, try
2178 * again with a larger value.
2179 */
2180 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2181 con->peer_global_seq,
2182 le32_to_cpu(con->in_reply.global_seq));
2183 get_global_seq(con->msgr,
2184 le32_to_cpu(con->in_reply.global_seq));
2185 con_out_kvec_reset(con);
2186 ret = prepare_write_connect(con);
2187 if (ret < 0)
2188 return ret;
2189 prepare_read_connect(con);
2190 break;
2191
2192 case CEPH_MSGR_TAG_SEQ:
2193 case CEPH_MSGR_TAG_READY:
2194 if (req_feat & ~server_feat) {
2195 pr_err("%s%lld %s protocol feature mismatch,"
2196 " my required %llx > server's %llx, need %llx\n",
2197 ENTITY_NAME(con->peer_name),
2198 ceph_pr_addr(&con->peer_addr.in_addr),
2199 req_feat, server_feat, req_feat & ~server_feat);
2200 con->error_msg = "missing required protocol features";
2201 reset_connection(con);
2202 return -1;
2203 }
2204
2205 WARN_ON(con->state != CON_STATE_NEGOTIATING);
2206 con->state = CON_STATE_OPEN;
2207 con->auth_retry = 0; /* we authenticated; clear flag */
2208 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2209 con->connect_seq++;
2210 con->peer_features = server_feat;
2211 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2212 con->peer_global_seq,
2213 le32_to_cpu(con->in_reply.connect_seq),
2214 con->connect_seq);
2215 WARN_ON(con->connect_seq !=
2216 le32_to_cpu(con->in_reply.connect_seq));
2217
2218 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2219 con_flag_set(con, CON_FLAG_LOSSYTX);
2220
2221 con->delay = 0; /* reset backoff memory */
2222
2223 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2224 prepare_write_seq(con);
2225 prepare_read_seq(con);
2226 } else {
2227 prepare_read_tag(con);
2228 }
2229 break;
2230
2231 case CEPH_MSGR_TAG_WAIT:
2232 /*
2233 * If there is a connection race (we are opening
2234 * connections to each other), one of us may just have
2235 * to WAIT. This shouldn't happen if we are the
2236 * client.
2237 */
2238 con->error_msg = "protocol error, got WAIT as client";
2239 return -1;
2240
2241 default:
2242 con->error_msg = "protocol error, garbage tag during connect";
2243 return -1;
2244 }
2245 return 0;
2246}
2247
2248
2249/*
2250 * read (part of) an ack
2251 */
2252static int read_partial_ack(struct ceph_connection *con)
2253{
2254 int size = sizeof (con->in_temp_ack);
2255 int end = size;
2256
2257 return read_partial(con, end, size, &con->in_temp_ack);
2258}
2259
2260/*
2261 * We can finally discard anything that's been acked.
2262 */
2263static void process_ack(struct ceph_connection *con)
2264{
2265 struct ceph_msg *m;
2266 u64 ack = le64_to_cpu(con->in_temp_ack);
2267 u64 seq;
2268 bool reconnect = (con->in_tag == CEPH_MSGR_TAG_SEQ);
2269 struct list_head *list = reconnect ? &con->out_queue : &con->out_sent;
2270
2271 /*
2272 * In the reconnect case, con_fault() has requeued messages
2273 * in out_sent. We should cleanup old messages according to
2274 * the reconnect seq.
2275 */
2276 while (!list_empty(list)) {
2277 m = list_first_entry(list, struct ceph_msg, list_head);
2278 if (reconnect && m->needs_out_seq)
2279 break;
2280 seq = le64_to_cpu(m->hdr.seq);
2281 if (seq > ack)
2282 break;
2283 dout("got ack for seq %llu type %d at %p\n", seq,
2284 le16_to_cpu(m->hdr.type), m);
2285 m->ack_stamp = jiffies;
2286 ceph_msg_remove(m);
2287 }
2288
2289 prepare_read_tag(con);
2290}
2291
2292
2293static int read_partial_message_section(struct ceph_connection *con,
2294 struct kvec *section,
2295 unsigned int sec_len, u32 *crc)
2296{
2297 int ret, left;
2298
2299 BUG_ON(!section);
2300
2301 while (section->iov_len < sec_len) {
2302 BUG_ON(section->iov_base == NULL);
2303 left = sec_len - section->iov_len;
2304 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2305 section->iov_len, left);
2306 if (ret <= 0)
2307 return ret;
2308 section->iov_len += ret;
2309 }
2310 if (section->iov_len == sec_len)
2311 *crc = crc32c(0, section->iov_base, section->iov_len);
2312
2313 return 1;
2314}
2315
2316static int read_partial_msg_data(struct ceph_connection *con)
2317{
2318 struct ceph_msg *msg = con->in_msg;
2319 struct ceph_msg_data_cursor *cursor = &msg->cursor;
2320 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2321 struct page *page;
2322 size_t page_offset;
2323 size_t length;
2324 u32 crc = 0;
2325 int ret;
2326
2327 BUG_ON(!msg);
2328 if (list_empty(&msg->data))
2329 return -EIO;
2330
2331 if (do_datacrc)
2332 crc = con->in_data_crc;
2333 while (cursor->total_resid) {
2334 if (!cursor->resid) {
2335 ceph_msg_data_advance(cursor, 0);
2336 continue;
2337 }
2338
2339 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
2340 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2341 if (ret <= 0) {
2342 if (do_datacrc)
2343 con->in_data_crc = crc;
2344
2345 return ret;
2346 }
2347
2348 if (do_datacrc)
2349 crc = ceph_crc32c_page(crc, page, page_offset, ret);
2350 ceph_msg_data_advance(cursor, (size_t)ret);
2351 }
2352 if (do_datacrc)
2353 con->in_data_crc = crc;
2354
2355 return 1; /* must return > 0 to indicate success */
2356}
2357
2358/*
2359 * read (part of) a message.
2360 */
2361static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2362
2363static int read_partial_message(struct ceph_connection *con)
2364{
2365 struct ceph_msg *m = con->in_msg;
2366 int size;
2367 int end;
2368 int ret;
2369 unsigned int front_len, middle_len, data_len;
2370 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2371 bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
2372 u64 seq;
2373 u32 crc;
2374
2375 dout("read_partial_message con %p msg %p\n", con, m);
2376
2377 /* header */
2378 size = sizeof (con->in_hdr);
2379 end = size;
2380 ret = read_partial(con, end, size, &con->in_hdr);
2381 if (ret <= 0)
2382 return ret;
2383
2384 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2385 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2386 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
2387 crc, con->in_hdr.crc);
2388 return -EBADMSG;
2389 }
2390
2391 front_len = le32_to_cpu(con->in_hdr.front_len);
2392 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2393 return -EIO;
2394 middle_len = le32_to_cpu(con->in_hdr.middle_len);
2395 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2396 return -EIO;
2397 data_len = le32_to_cpu(con->in_hdr.data_len);
2398 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2399 return -EIO;
2400
2401 /* verify seq# */
2402 seq = le64_to_cpu(con->in_hdr.seq);
2403 if ((s64)seq - (s64)con->in_seq < 1) {
2404 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2405 ENTITY_NAME(con->peer_name),
2406 ceph_pr_addr(&con->peer_addr.in_addr),
2407 seq, con->in_seq + 1);
2408 con->in_base_pos = -front_len - middle_len - data_len -
2409 sizeof_footer(con);
2410 con->in_tag = CEPH_MSGR_TAG_READY;
2411 return 1;
2412 } else if ((s64)seq - (s64)con->in_seq > 1) {
2413 pr_err("read_partial_message bad seq %lld expected %lld\n",
2414 seq, con->in_seq + 1);
2415 con->error_msg = "bad message sequence # for incoming message";
2416 return -EBADE;
2417 }
2418
2419 /* allocate message? */
2420 if (!con->in_msg) {
2421 int skip = 0;
2422
2423 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2424 front_len, data_len);
2425 ret = ceph_con_in_msg_alloc(con, &skip);
2426 if (ret < 0)
2427 return ret;
2428
2429 BUG_ON(!con->in_msg ^ skip);
2430 if (skip) {
2431 /* skip this message */
2432 dout("alloc_msg said skip message\n");
2433 con->in_base_pos = -front_len - middle_len - data_len -
2434 sizeof_footer(con);
2435 con->in_tag = CEPH_MSGR_TAG_READY;
2436 con->in_seq++;
2437 return 1;
2438 }
2439
2440 BUG_ON(!con->in_msg);
2441 BUG_ON(con->in_msg->con != con);
2442 m = con->in_msg;
2443 m->front.iov_len = 0; /* haven't read it yet */
2444 if (m->middle)
2445 m->middle->vec.iov_len = 0;
2446
2447 /* prepare for data payload, if any */
2448
2449 if (data_len)
2450 prepare_message_data(con->in_msg, data_len);
2451 }
2452
2453 /* front */
2454 ret = read_partial_message_section(con, &m->front, front_len,
2455 &con->in_front_crc);
2456 if (ret <= 0)
2457 return ret;
2458
2459 /* middle */
2460 if (m->middle) {
2461 ret = read_partial_message_section(con, &m->middle->vec,
2462 middle_len,
2463 &con->in_middle_crc);
2464 if (ret <= 0)
2465 return ret;
2466 }
2467
2468 /* (page) data */
2469 if (data_len) {
2470 ret = read_partial_msg_data(con);
2471 if (ret <= 0)
2472 return ret;
2473 }
2474
2475 /* footer */
2476 size = sizeof_footer(con);
2477 end += size;
2478 ret = read_partial(con, end, size, &m->footer);
2479 if (ret <= 0)
2480 return ret;
2481
2482 if (!need_sign) {
2483 m->footer.flags = m->old_footer.flags;
2484 m->footer.sig = 0;
2485 }
2486
2487 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2488 m, front_len, m->footer.front_crc, middle_len,
2489 m->footer.middle_crc, data_len, m->footer.data_crc);
2490
2491 /* crc ok? */
2492 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2493 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2494 m, con->in_front_crc, m->footer.front_crc);
2495 return -EBADMSG;
2496 }
2497 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2498 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2499 m, con->in_middle_crc, m->footer.middle_crc);
2500 return -EBADMSG;
2501 }
2502 if (do_datacrc &&
2503 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2504 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2505 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2506 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2507 return -EBADMSG;
2508 }
2509
2510 if (need_sign && con->ops->check_message_signature &&
2511 con->ops->check_message_signature(m)) {
2512 pr_err("read_partial_message %p signature check failed\n", m);
2513 return -EBADMSG;
2514 }
2515
2516 return 1; /* done! */
2517}
2518
2519/*
2520 * Process message. This happens in the worker thread. The callback should
2521 * be careful not to do anything that waits on other incoming messages or it
2522 * may deadlock.
2523 */
2524static void process_message(struct ceph_connection *con)
2525{
2526 struct ceph_msg *msg = con->in_msg;
2527
2528 BUG_ON(con->in_msg->con != con);
2529 con->in_msg = NULL;
2530
2531 /* if first message, set peer_name */
2532 if (con->peer_name.type == 0)
2533 con->peer_name = msg->hdr.src;
2534
2535 con->in_seq++;
2536 mutex_unlock(&con->mutex);
2537
2538 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2539 msg, le64_to_cpu(msg->hdr.seq),
2540 ENTITY_NAME(msg->hdr.src),
2541 le16_to_cpu(msg->hdr.type),
2542 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2543 le32_to_cpu(msg->hdr.front_len),
2544 le32_to_cpu(msg->hdr.data_len),
2545 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2546 con->ops->dispatch(con, msg);
2547
2548 mutex_lock(&con->mutex);
2549}
2550
2551static int read_keepalive_ack(struct ceph_connection *con)
2552{
2553 struct ceph_timespec ceph_ts;
2554 size_t size = sizeof(ceph_ts);
2555 int ret = read_partial(con, size, size, &ceph_ts);
2556 if (ret <= 0)
2557 return ret;
2558 ceph_decode_timespec(&con->last_keepalive_ack, &ceph_ts);
2559 prepare_read_tag(con);
2560 return 1;
2561}
2562
2563/*
2564 * Write something to the socket. Called in a worker thread when the
2565 * socket appears to be writeable and we have something ready to send.
2566 */
2567static int try_write(struct ceph_connection *con)
2568{
2569 int ret = 1;
2570
2571 dout("try_write start %p state %lu\n", con, con->state);
2572 if (con->state != CON_STATE_PREOPEN &&
2573 con->state != CON_STATE_CONNECTING &&
2574 con->state != CON_STATE_NEGOTIATING &&
2575 con->state != CON_STATE_OPEN)
2576 return 0;
2577
2578more:
2579 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2580
2581 /* open the socket first? */
2582 if (con->state == CON_STATE_PREOPEN) {
2583 BUG_ON(con->sock);
2584 con->state = CON_STATE_CONNECTING;
2585
2586 con_out_kvec_reset(con);
2587 prepare_write_banner(con);
2588 prepare_read_banner(con);
2589
2590 BUG_ON(con->in_msg);
2591 con->in_tag = CEPH_MSGR_TAG_READY;
2592 dout("try_write initiating connect on %p new state %lu\n",
2593 con, con->state);
2594 ret = ceph_tcp_connect(con);
2595 if (ret < 0) {
2596 con->error_msg = "connect error";
2597 goto out;
2598 }
2599 }
2600
2601more_kvec:
2602 BUG_ON(!con->sock);
2603
2604 /* kvec data queued? */
2605 if (con->out_kvec_left) {
2606 ret = write_partial_kvec(con);
2607 if (ret <= 0)
2608 goto out;
2609 }
2610 if (con->out_skip) {
2611 ret = write_partial_skip(con);
2612 if (ret <= 0)
2613 goto out;
2614 }
2615
2616 /* msg pages? */
2617 if (con->out_msg) {
2618 if (con->out_msg_done) {
2619 ceph_msg_put(con->out_msg);
2620 con->out_msg = NULL; /* we're done with this one */
2621 goto do_next;
2622 }
2623
2624 ret = write_partial_message_data(con);
2625 if (ret == 1)
2626 goto more_kvec; /* we need to send the footer, too! */
2627 if (ret == 0)
2628 goto out;
2629 if (ret < 0) {
2630 dout("try_write write_partial_message_data err %d\n",
2631 ret);
2632 goto out;
2633 }
2634 }
2635
2636do_next:
2637 if (con->state == CON_STATE_OPEN) {
2638 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2639 prepare_write_keepalive(con);
2640 goto more;
2641 }
2642 /* is anything else pending? */
2643 if (!list_empty(&con->out_queue)) {
2644 prepare_write_message(con);
2645 goto more;
2646 }
2647 if (con->in_seq > con->in_seq_acked) {
2648 prepare_write_ack(con);
2649 goto more;
2650 }
2651 }
2652
2653 /* Nothing to do! */
2654 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2655 dout("try_write nothing else to write.\n");
2656 ret = 0;
2657out:
2658 dout("try_write done on %p ret %d\n", con, ret);
2659 return ret;
2660}
2661
2662
2663
2664/*
2665 * Read what we can from the socket.
2666 */
2667static int try_read(struct ceph_connection *con)
2668{
2669 int ret = -1;
2670
2671more:
2672 dout("try_read start on %p state %lu\n", con, con->state);
2673 if (con->state != CON_STATE_CONNECTING &&
2674 con->state != CON_STATE_NEGOTIATING &&
2675 con->state != CON_STATE_OPEN)
2676 return 0;
2677
2678 BUG_ON(!con->sock);
2679
2680 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2681 con->in_base_pos);
2682
2683 if (con->state == CON_STATE_CONNECTING) {
2684 dout("try_read connecting\n");
2685 ret = read_partial_banner(con);
2686 if (ret <= 0)
2687 goto out;
2688 ret = process_banner(con);
2689 if (ret < 0)
2690 goto out;
2691
2692 con->state = CON_STATE_NEGOTIATING;
2693
2694 /*
2695 * Received banner is good, exchange connection info.
2696 * Do not reset out_kvec, as sending our banner raced
2697 * with receiving peer banner after connect completed.
2698 */
2699 ret = prepare_write_connect(con);
2700 if (ret < 0)
2701 goto out;
2702 prepare_read_connect(con);
2703
2704 /* Send connection info before awaiting response */
2705 goto out;
2706 }
2707
2708 if (con->state == CON_STATE_NEGOTIATING) {
2709 dout("try_read negotiating\n");
2710 ret = read_partial_connect(con);
2711 if (ret <= 0)
2712 goto out;
2713 ret = process_connect(con);
2714 if (ret < 0)
2715 goto out;
2716 goto more;
2717 }
2718
2719 WARN_ON(con->state != CON_STATE_OPEN);
2720
2721 if (con->in_base_pos < 0) {
2722 /*
2723 * skipping + discarding content.
2724 *
2725 * FIXME: there must be a better way to do this!
2726 */
2727 static char buf[SKIP_BUF_SIZE];
2728 int skip = min((int) sizeof (buf), -con->in_base_pos);
2729
2730 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2731 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2732 if (ret <= 0)
2733 goto out;
2734 con->in_base_pos += ret;
2735 if (con->in_base_pos)
2736 goto more;
2737 }
2738 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2739 /*
2740 * what's next?
2741 */
2742 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2743 if (ret <= 0)
2744 goto out;
2745 dout("try_read got tag %d\n", (int)con->in_tag);
2746 switch (con->in_tag) {
2747 case CEPH_MSGR_TAG_MSG:
2748 prepare_read_message(con);
2749 break;
2750 case CEPH_MSGR_TAG_ACK:
2751 prepare_read_ack(con);
2752 break;
2753 case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
2754 prepare_read_keepalive_ack(con);
2755 break;
2756 case CEPH_MSGR_TAG_CLOSE:
2757 con_close_socket(con);
2758 con->state = CON_STATE_CLOSED;
2759 goto out;
2760 default:
2761 goto bad_tag;
2762 }
2763 }
2764 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2765 ret = read_partial_message(con);
2766 if (ret <= 0) {
2767 switch (ret) {
2768 case -EBADMSG:
2769 con->error_msg = "bad crc/signature";
2770 /* fall through */
2771 case -EBADE:
2772 ret = -EIO;
2773 break;
2774 case -EIO:
2775 con->error_msg = "io error";
2776 break;
2777 }
2778 goto out;
2779 }
2780 if (con->in_tag == CEPH_MSGR_TAG_READY)
2781 goto more;
2782 process_message(con);
2783 if (con->state == CON_STATE_OPEN)
2784 prepare_read_tag(con);
2785 goto more;
2786 }
2787 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2788 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2789 /*
2790 * the final handshake seq exchange is semantically
2791 * equivalent to an ACK
2792 */
2793 ret = read_partial_ack(con);
2794 if (ret <= 0)
2795 goto out;
2796 process_ack(con);
2797 goto more;
2798 }
2799 if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
2800 ret = read_keepalive_ack(con);
2801 if (ret <= 0)
2802 goto out;
2803 goto more;
2804 }
2805
2806out:
2807 dout("try_read done on %p ret %d\n", con, ret);
2808 return ret;
2809
2810bad_tag:
2811 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2812 con->error_msg = "protocol error, garbage tag";
2813 ret = -1;
2814 goto out;
2815}
2816
2817
2818/*
2819 * Atomically queue work on a connection after the specified delay.
2820 * Bump @con reference to avoid races with connection teardown.
2821 * Returns 0 if work was queued, or an error code otherwise.
2822 */
2823static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2824{
2825 if (!con->ops->get(con)) {
2826 dout("%s %p ref count 0\n", __func__, con);
2827 return -ENOENT;
2828 }
2829
2830 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2831 dout("%s %p - already queued\n", __func__, con);
2832 con->ops->put(con);
2833 return -EBUSY;
2834 }
2835
2836 dout("%s %p %lu\n", __func__, con, delay);
2837 return 0;
2838}
2839
2840static void queue_con(struct ceph_connection *con)
2841{
2842 (void) queue_con_delay(con, 0);
2843}
2844
2845static void cancel_con(struct ceph_connection *con)
2846{
2847 if (cancel_delayed_work(&con->work)) {
2848 dout("%s %p\n", __func__, con);
2849 con->ops->put(con);
2850 }
2851}
2852
2853static bool con_sock_closed(struct ceph_connection *con)
2854{
2855 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2856 return false;
2857
2858#define CASE(x) \
2859 case CON_STATE_ ## x: \
2860 con->error_msg = "socket closed (con state " #x ")"; \
2861 break;
2862
2863 switch (con->state) {
2864 CASE(CLOSED);
2865 CASE(PREOPEN);
2866 CASE(CONNECTING);
2867 CASE(NEGOTIATING);
2868 CASE(OPEN);
2869 CASE(STANDBY);
2870 default:
2871 pr_warn("%s con %p unrecognized state %lu\n",
2872 __func__, con, con->state);
2873 con->error_msg = "unrecognized con state";
2874 BUG();
2875 break;
2876 }
2877#undef CASE
2878
2879 return true;
2880}
2881
2882static bool con_backoff(struct ceph_connection *con)
2883{
2884 int ret;
2885
2886 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2887 return false;
2888
2889 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2890 if (ret) {
2891 dout("%s: con %p FAILED to back off %lu\n", __func__,
2892 con, con->delay);
2893 BUG_ON(ret == -ENOENT);
2894 con_flag_set(con, CON_FLAG_BACKOFF);
2895 }
2896
2897 return true;
2898}
2899
2900/* Finish fault handling; con->mutex must *not* be held here */
2901
2902static void con_fault_finish(struct ceph_connection *con)
2903{
2904 dout("%s %p\n", __func__, con);
2905
2906 /*
2907 * in case we faulted due to authentication, invalidate our
2908 * current tickets so that we can get new ones.
2909 */
2910 if (con->auth_retry) {
2911 dout("auth_retry %d, invalidating\n", con->auth_retry);
2912 if (con->ops->invalidate_authorizer)
2913 con->ops->invalidate_authorizer(con);
2914 con->auth_retry = 0;
2915 }
2916
2917 if (con->ops->fault)
2918 con->ops->fault(con);
2919}
2920
2921/*
2922 * Do some work on a connection. Drop a connection ref when we're done.
2923 */
2924static void ceph_con_workfn(struct work_struct *work)
2925{
2926 struct ceph_connection *con = container_of(work, struct ceph_connection,
2927 work.work);
2928 bool fault;
2929
2930 mutex_lock(&con->mutex);
2931 while (true) {
2932 int ret;
2933
2934 if ((fault = con_sock_closed(con))) {
2935 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2936 break;
2937 }
2938 if (con_backoff(con)) {
2939 dout("%s: con %p BACKOFF\n", __func__, con);
2940 break;
2941 }
2942 if (con->state == CON_STATE_STANDBY) {
2943 dout("%s: con %p STANDBY\n", __func__, con);
2944 break;
2945 }
2946 if (con->state == CON_STATE_CLOSED) {
2947 dout("%s: con %p CLOSED\n", __func__, con);
2948 BUG_ON(con->sock);
2949 break;
2950 }
2951 if (con->state == CON_STATE_PREOPEN) {
2952 dout("%s: con %p PREOPEN\n", __func__, con);
2953 BUG_ON(con->sock);
2954 }
2955
2956 ret = try_read(con);
2957 if (ret < 0) {
2958 if (ret == -EAGAIN)
2959 continue;
2960 if (!con->error_msg)
2961 con->error_msg = "socket error on read";
2962 fault = true;
2963 break;
2964 }
2965
2966 ret = try_write(con);
2967 if (ret < 0) {
2968 if (ret == -EAGAIN)
2969 continue;
2970 if (!con->error_msg)
2971 con->error_msg = "socket error on write";
2972 fault = true;
2973 }
2974
2975 break; /* If we make it to here, we're done */
2976 }
2977 if (fault)
2978 con_fault(con);
2979 mutex_unlock(&con->mutex);
2980
2981 if (fault)
2982 con_fault_finish(con);
2983
2984 con->ops->put(con);
2985}
2986
2987/*
2988 * Generic error/fault handler. A retry mechanism is used with
2989 * exponential backoff
2990 */
2991static void con_fault(struct ceph_connection *con)
2992{
2993 dout("fault %p state %lu to peer %s\n",
2994 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2995
2996 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2997 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2998 con->error_msg = NULL;
2999
3000 WARN_ON(con->state != CON_STATE_CONNECTING &&
3001 con->state != CON_STATE_NEGOTIATING &&
3002 con->state != CON_STATE_OPEN);
3003
3004 con_close_socket(con);
3005
3006 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
3007 dout("fault on LOSSYTX channel, marking CLOSED\n");
3008 con->state = CON_STATE_CLOSED;
3009 return;
3010 }
3011
3012 if (con->in_msg) {
3013 BUG_ON(con->in_msg->con != con);
3014 ceph_msg_put(con->in_msg);
3015 con->in_msg = NULL;
3016 }
3017
3018 /* Requeue anything that hasn't been acked */
3019 list_splice_init(&con->out_sent, &con->out_queue);
3020
3021 /* If there are no messages queued or keepalive pending, place
3022 * the connection in a STANDBY state */
3023 if (list_empty(&con->out_queue) &&
3024 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
3025 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
3026 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
3027 con->state = CON_STATE_STANDBY;
3028 } else {
3029 /* retry after a delay. */
3030 con->state = CON_STATE_PREOPEN;
3031 if (con->delay == 0)
3032 con->delay = BASE_DELAY_INTERVAL;
3033 else if (con->delay < MAX_DELAY_INTERVAL)
3034 con->delay *= 2;
3035 con_flag_set(con, CON_FLAG_BACKOFF);
3036 queue_con(con);
3037 }
3038}
3039
3040
3041
3042/*
3043 * initialize a new messenger instance
3044 */
3045void ceph_messenger_init(struct ceph_messenger *msgr,
3046 struct ceph_entity_addr *myaddr)
3047{
3048 spin_lock_init(&msgr->global_seq_lock);
3049
3050 if (myaddr)
3051 msgr->inst.addr = *myaddr;
3052
3053 /* select a random nonce */
3054 msgr->inst.addr.type = 0;
3055 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
3056 encode_my_addr(msgr);
3057
3058 atomic_set(&msgr->stopping, 0);
3059 write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
3060
3061 dout("%s %p\n", __func__, msgr);
3062}
3063EXPORT_SYMBOL(ceph_messenger_init);
3064
3065void ceph_messenger_fini(struct ceph_messenger *msgr)
3066{
3067 put_net(read_pnet(&msgr->net));
3068}
3069EXPORT_SYMBOL(ceph_messenger_fini);
3070
3071static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con)
3072{
3073 if (msg->con)
3074 msg->con->ops->put(msg->con);
3075
3076 msg->con = con ? con->ops->get(con) : NULL;
3077 BUG_ON(msg->con != con);
3078}
3079
3080static void clear_standby(struct ceph_connection *con)
3081{
3082 /* come back from STANDBY? */
3083 if (con->state == CON_STATE_STANDBY) {
3084 dout("clear_standby %p and ++connect_seq\n", con);
3085 con->state = CON_STATE_PREOPEN;
3086 con->connect_seq++;
3087 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
3088 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
3089 }
3090}
3091
3092/*
3093 * Queue up an outgoing message on the given connection.
3094 */
3095void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
3096{
3097 /* set src+dst */
3098 msg->hdr.src = con->msgr->inst.name;
3099 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
3100 msg->needs_out_seq = true;
3101
3102 mutex_lock(&con->mutex);
3103
3104 if (con->state == CON_STATE_CLOSED) {
3105 dout("con_send %p closed, dropping %p\n", con, msg);
3106 ceph_msg_put(msg);
3107 mutex_unlock(&con->mutex);
3108 return;
3109 }
3110
3111 msg_con_set(msg, con);
3112
3113 BUG_ON(!list_empty(&msg->list_head));
3114 list_add_tail(&msg->list_head, &con->out_queue);
3115 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3116 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3117 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3118 le32_to_cpu(msg->hdr.front_len),
3119 le32_to_cpu(msg->hdr.middle_len),
3120 le32_to_cpu(msg->hdr.data_len));
3121
3122 clear_standby(con);
3123 mutex_unlock(&con->mutex);
3124
3125 /* if there wasn't anything waiting to send before, queue
3126 * new work */
3127 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3128 queue_con(con);
3129}
3130EXPORT_SYMBOL(ceph_con_send);
3131
3132/*
3133 * Revoke a message that was previously queued for send
3134 */
3135void ceph_msg_revoke(struct ceph_msg *msg)
3136{
3137 struct ceph_connection *con = msg->con;
3138
3139 if (!con) {
3140 dout("%s msg %p null con\n", __func__, msg);
3141 return; /* Message not in our possession */
3142 }
3143
3144 mutex_lock(&con->mutex);
3145 if (!list_empty(&msg->list_head)) {
3146 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
3147 list_del_init(&msg->list_head);
3148 msg->hdr.seq = 0;
3149
3150 ceph_msg_put(msg);
3151 }
3152 if (con->out_msg == msg) {
3153 BUG_ON(con->out_skip);
3154 /* footer */
3155 if (con->out_msg_done) {
3156 con->out_skip += con_out_kvec_skip(con);
3157 } else {
3158 BUG_ON(!msg->data_length);
3159 con->out_skip += sizeof_footer(con);
3160 }
3161 /* data, middle, front */
3162 if (msg->data_length)
3163 con->out_skip += msg->cursor.total_resid;
3164 if (msg->middle)
3165 con->out_skip += con_out_kvec_skip(con);
3166 con->out_skip += con_out_kvec_skip(con);
3167
3168 dout("%s %p msg %p - was sending, will write %d skip %d\n",
3169 __func__, con, msg, con->out_kvec_bytes, con->out_skip);
3170 msg->hdr.seq = 0;
3171 con->out_msg = NULL;
3172 ceph_msg_put(msg);
3173 }
3174
3175 mutex_unlock(&con->mutex);
3176}
3177
3178/*
3179 * Revoke a message that we may be reading data into
3180 */
3181void ceph_msg_revoke_incoming(struct ceph_msg *msg)
3182{
3183 struct ceph_connection *con = msg->con;
3184
3185 if (!con) {
3186 dout("%s msg %p null con\n", __func__, msg);
3187 return; /* Message not in our possession */
3188 }
3189
3190 mutex_lock(&con->mutex);
3191 if (con->in_msg == msg) {
3192 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3193 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3194 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
3195
3196 /* skip rest of message */
3197 dout("%s %p msg %p revoked\n", __func__, con, msg);
3198 con->in_base_pos = con->in_base_pos -
3199 sizeof(struct ceph_msg_header) -
3200 front_len -
3201 middle_len -
3202 data_len -
3203 sizeof(struct ceph_msg_footer);
3204 ceph_msg_put(con->in_msg);
3205 con->in_msg = NULL;
3206 con->in_tag = CEPH_MSGR_TAG_READY;
3207 con->in_seq++;
3208 } else {
3209 dout("%s %p in_msg %p msg %p no-op\n",
3210 __func__, con, con->in_msg, msg);
3211 }
3212 mutex_unlock(&con->mutex);
3213}
3214
3215/*
3216 * Queue a keepalive byte to ensure the tcp connection is alive.
3217 */
3218void ceph_con_keepalive(struct ceph_connection *con)
3219{
3220 dout("con_keepalive %p\n", con);
3221 mutex_lock(&con->mutex);
3222 clear_standby(con);
3223 mutex_unlock(&con->mutex);
3224 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3225 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3226 queue_con(con);
3227}
3228EXPORT_SYMBOL(ceph_con_keepalive);
3229
3230bool ceph_con_keepalive_expired(struct ceph_connection *con,
3231 unsigned long interval)
3232{
3233 if (interval > 0 &&
3234 (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
3235 struct timespec now;
3236 struct timespec ts;
3237 ktime_get_real_ts(&now);
3238 jiffies_to_timespec(interval, &ts);
3239 ts = timespec_add(con->last_keepalive_ack, ts);
3240 return timespec_compare(&now, &ts) >= 0;
3241 }
3242 return false;
3243}
3244
3245static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
3246{
3247 struct ceph_msg_data *data;
3248
3249 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3250 return NULL;
3251
3252 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
3253 if (!data)
3254 return NULL;
3255
3256 data->type = type;
3257 INIT_LIST_HEAD(&data->links);
3258
3259 return data;
3260}
3261
3262static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3263{
3264 if (!data)
3265 return;
3266
3267 WARN_ON(!list_empty(&data->links));
3268 if (data->type == CEPH_MSG_DATA_PAGELIST)
3269 ceph_pagelist_release(data->pagelist);
3270 kmem_cache_free(ceph_msg_data_cache, data);
3271}
3272
3273void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
3274 size_t length, size_t alignment)
3275{
3276 struct ceph_msg_data *data;
3277
3278 BUG_ON(!pages);
3279 BUG_ON(!length);
3280
3281 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3282 BUG_ON(!data);
3283 data->pages = pages;
3284 data->length = length;
3285 data->alignment = alignment & ~PAGE_MASK;
3286
3287 list_add_tail(&data->links, &msg->data);
3288 msg->data_length += length;
3289}
3290EXPORT_SYMBOL(ceph_msg_data_add_pages);
3291
3292void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
3293 struct ceph_pagelist *pagelist)
3294{
3295 struct ceph_msg_data *data;
3296
3297 BUG_ON(!pagelist);
3298 BUG_ON(!pagelist->length);
3299
3300 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3301 BUG_ON(!data);
3302 data->pagelist = pagelist;
3303
3304 list_add_tail(&data->links, &msg->data);
3305 msg->data_length += pagelist->length;
3306}
3307EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
3308
3309#ifdef CONFIG_BLOCK
3310void ceph_msg_data_add_bio(struct ceph_msg *msg, struct ceph_bio_iter *bio_pos,
3311 u32 length)
3312{
3313 struct ceph_msg_data *data;
3314
3315 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3316 BUG_ON(!data);
3317 data->bio_pos = *bio_pos;
3318 data->bio_length = length;
3319
3320 list_add_tail(&data->links, &msg->data);
3321 msg->data_length += length;
3322}
3323EXPORT_SYMBOL(ceph_msg_data_add_bio);
3324#endif /* CONFIG_BLOCK */
3325
3326void ceph_msg_data_add_bvecs(struct ceph_msg *msg,
3327 struct ceph_bvec_iter *bvec_pos)
3328{
3329 struct ceph_msg_data *data;
3330
3331 data = ceph_msg_data_create(CEPH_MSG_DATA_BVECS);
3332 BUG_ON(!data);
3333 data->bvec_pos = *bvec_pos;
3334
3335 list_add_tail(&data->links, &msg->data);
3336 msg->data_length += bvec_pos->iter.bi_size;
3337}
3338EXPORT_SYMBOL(ceph_msg_data_add_bvecs);
3339
3340/*
3341 * construct a new message with given type, size
3342 * the new msg has a ref count of 1.
3343 */
3344struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3345 bool can_fail)
3346{
3347 struct ceph_msg *m;
3348
3349 m = kmem_cache_zalloc(ceph_msg_cache, flags);
3350 if (m == NULL)
3351 goto out;
3352
3353 m->hdr.type = cpu_to_le16(type);
3354 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3355 m->hdr.front_len = cpu_to_le32(front_len);
3356
3357 INIT_LIST_HEAD(&m->list_head);
3358 kref_init(&m->kref);
3359 INIT_LIST_HEAD(&m->data);
3360
3361 /* front */
3362 if (front_len) {
3363 m->front.iov_base = ceph_kvmalloc(front_len, flags);
3364 if (m->front.iov_base == NULL) {
3365 dout("ceph_msg_new can't allocate %d bytes\n",
3366 front_len);
3367 goto out2;
3368 }
3369 } else {
3370 m->front.iov_base = NULL;
3371 }
3372 m->front_alloc_len = m->front.iov_len = front_len;
3373
3374 dout("ceph_msg_new %p front %d\n", m, front_len);
3375 return m;
3376
3377out2:
3378 ceph_msg_put(m);
3379out:
3380 if (!can_fail) {
3381 pr_err("msg_new can't create type %d front %d\n", type,
3382 front_len);
3383 WARN_ON(1);
3384 } else {
3385 dout("msg_new can't create type %d front %d\n", type,
3386 front_len);
3387 }
3388 return NULL;
3389}
3390EXPORT_SYMBOL(ceph_msg_new);
3391
3392/*
3393 * Allocate "middle" portion of a message, if it is needed and wasn't
3394 * allocated by alloc_msg. This allows us to read a small fixed-size
3395 * per-type header in the front and then gracefully fail (i.e.,
3396 * propagate the error to the caller based on info in the front) when
3397 * the middle is too large.
3398 */
3399static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3400{
3401 int type = le16_to_cpu(msg->hdr.type);
3402 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3403
3404 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3405 ceph_msg_type_name(type), middle_len);
3406 BUG_ON(!middle_len);
3407 BUG_ON(msg->middle);
3408
3409 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3410 if (!msg->middle)
3411 return -ENOMEM;
3412 return 0;
3413}
3414
3415/*
3416 * Allocate a message for receiving an incoming message on a
3417 * connection, and save the result in con->in_msg. Uses the
3418 * connection's private alloc_msg op if available.
3419 *
3420 * Returns 0 on success, or a negative error code.
3421 *
3422 * On success, if we set *skip = 1:
3423 * - the next message should be skipped and ignored.
3424 * - con->in_msg == NULL
3425 * or if we set *skip = 0:
3426 * - con->in_msg is non-null.
3427 * On error (ENOMEM, EAGAIN, ...),
3428 * - con->in_msg == NULL
3429 */
3430static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3431{
3432 struct ceph_msg_header *hdr = &con->in_hdr;
3433 int middle_len = le32_to_cpu(hdr->middle_len);
3434 struct ceph_msg *msg;
3435 int ret = 0;
3436
3437 BUG_ON(con->in_msg != NULL);
3438 BUG_ON(!con->ops->alloc_msg);
3439
3440 mutex_unlock(&con->mutex);
3441 msg = con->ops->alloc_msg(con, hdr, skip);
3442 mutex_lock(&con->mutex);
3443 if (con->state != CON_STATE_OPEN) {
3444 if (msg)
3445 ceph_msg_put(msg);
3446 return -EAGAIN;
3447 }
3448 if (msg) {
3449 BUG_ON(*skip);
3450 msg_con_set(msg, con);
3451 con->in_msg = msg;
3452 } else {
3453 /*
3454 * Null message pointer means either we should skip
3455 * this message or we couldn't allocate memory. The
3456 * former is not an error.
3457 */
3458 if (*skip)
3459 return 0;
3460
3461 con->error_msg = "error allocating memory for incoming message";
3462 return -ENOMEM;
3463 }
3464 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3465
3466 if (middle_len && !con->in_msg->middle) {
3467 ret = ceph_alloc_middle(con, con->in_msg);
3468 if (ret < 0) {
3469 ceph_msg_put(con->in_msg);
3470 con->in_msg = NULL;
3471 }
3472 }
3473
3474 return ret;
3475}
3476
3477
3478/*
3479 * Free a generically kmalloc'd message.
3480 */
3481static void ceph_msg_free(struct ceph_msg *m)
3482{
3483 dout("%s %p\n", __func__, m);
3484 kvfree(m->front.iov_base);
3485 kmem_cache_free(ceph_msg_cache, m);
3486}
3487
3488static void ceph_msg_release(struct kref *kref)
3489{
3490 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3491 struct ceph_msg_data *data, *next;
3492
3493 dout("%s %p\n", __func__, m);
3494 WARN_ON(!list_empty(&m->list_head));
3495
3496 msg_con_set(m, NULL);
3497
3498 /* drop middle, data, if any */
3499 if (m->middle) {
3500 ceph_buffer_put(m->middle);
3501 m->middle = NULL;
3502 }
3503
3504 list_for_each_entry_safe(data, next, &m->data, links) {
3505 list_del_init(&data->links);
3506 ceph_msg_data_destroy(data);
3507 }
3508 m->data_length = 0;
3509
3510 if (m->pool)
3511 ceph_msgpool_put(m->pool, m);
3512 else
3513 ceph_msg_free(m);
3514}
3515
3516struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3517{
3518 dout("%s %p (was %d)\n", __func__, msg,
3519 kref_read(&msg->kref));
3520 kref_get(&msg->kref);
3521 return msg;
3522}
3523EXPORT_SYMBOL(ceph_msg_get);
3524
3525void ceph_msg_put(struct ceph_msg *msg)
3526{
3527 dout("%s %p (was %d)\n", __func__, msg,
3528 kref_read(&msg->kref));
3529 kref_put(&msg->kref, ceph_msg_release);
3530}
3531EXPORT_SYMBOL(ceph_msg_put);
3532
3533void ceph_msg_dump(struct ceph_msg *msg)
3534{
3535 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3536 msg->front_alloc_len, msg->data_length);
3537 print_hex_dump(KERN_DEBUG, "header: ",
3538 DUMP_PREFIX_OFFSET, 16, 1,
3539 &msg->hdr, sizeof(msg->hdr), true);
3540 print_hex_dump(KERN_DEBUG, " front: ",
3541 DUMP_PREFIX_OFFSET, 16, 1,
3542 msg->front.iov_base, msg->front.iov_len, true);
3543 if (msg->middle)
3544 print_hex_dump(KERN_DEBUG, "middle: ",
3545 DUMP_PREFIX_OFFSET, 16, 1,
3546 msg->middle->vec.iov_base,
3547 msg->middle->vec.iov_len, true);
3548 print_hex_dump(KERN_DEBUG, "footer: ",
3549 DUMP_PREFIX_OFFSET, 16, 1,
3550 &msg->footer, sizeof(msg->footer), true);
3551}
3552EXPORT_SYMBOL(ceph_msg_dump);
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#include <linux/bio.h>
13#include <linux/blkdev.h>
14#include <linux/dns_resolver.h>
15#include <net/tcp.h>
16
17#include <linux/ceph/libceph.h>
18#include <linux/ceph/messenger.h>
19#include <linux/ceph/decode.h>
20#include <linux/ceph/pagelist.h>
21#include <linux/export.h>
22
23/*
24 * Ceph uses the messenger to exchange ceph_msg messages with other
25 * hosts in the system. The messenger provides ordered and reliable
26 * delivery. We tolerate TCP disconnects by reconnecting (with
27 * exponential backoff) in the case of a fault (disconnection, bad
28 * crc, protocol error). Acks allow sent messages to be discarded by
29 * the sender.
30 */
31
32/* static tag bytes (protocol control messages) */
33static char tag_msg = CEPH_MSGR_TAG_MSG;
34static char tag_ack = CEPH_MSGR_TAG_ACK;
35static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
36
37#ifdef CONFIG_LOCKDEP
38static struct lock_class_key socket_class;
39#endif
40
41/*
42 * When skipping (ignoring) a block of input we read it into a "skip
43 * buffer," which is this many bytes in size.
44 */
45#define SKIP_BUF_SIZE 1024
46
47static void queue_con(struct ceph_connection *con);
48static void con_work(struct work_struct *);
49static void ceph_fault(struct ceph_connection *con);
50
51/*
52 * Nicely render a sockaddr as a string. An array of formatted
53 * strings is used, to approximate reentrancy.
54 */
55#define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
56#define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
57#define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
58#define MAX_ADDR_STR_LEN 64 /* 54 is enough */
59
60static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
61static atomic_t addr_str_seq = ATOMIC_INIT(0);
62
63static struct page *zero_page; /* used in certain error cases */
64
65const char *ceph_pr_addr(const struct sockaddr_storage *ss)
66{
67 int i;
68 char *s;
69 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
70 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
71
72 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
73 s = addr_str[i];
74
75 switch (ss->ss_family) {
76 case AF_INET:
77 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
78 ntohs(in4->sin_port));
79 break;
80
81 case AF_INET6:
82 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
83 ntohs(in6->sin6_port));
84 break;
85
86 default:
87 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
88 ss->ss_family);
89 }
90
91 return s;
92}
93EXPORT_SYMBOL(ceph_pr_addr);
94
95static void encode_my_addr(struct ceph_messenger *msgr)
96{
97 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
98 ceph_encode_addr(&msgr->my_enc_addr);
99}
100
101/*
102 * work queue for all reading and writing to/from the socket.
103 */
104static struct workqueue_struct *ceph_msgr_wq;
105
106void _ceph_msgr_exit(void)
107{
108 if (ceph_msgr_wq) {
109 destroy_workqueue(ceph_msgr_wq);
110 ceph_msgr_wq = NULL;
111 }
112
113 BUG_ON(zero_page == NULL);
114 kunmap(zero_page);
115 page_cache_release(zero_page);
116 zero_page = NULL;
117}
118
119int ceph_msgr_init(void)
120{
121 BUG_ON(zero_page != NULL);
122 zero_page = ZERO_PAGE(0);
123 page_cache_get(zero_page);
124
125 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0);
126 if (ceph_msgr_wq)
127 return 0;
128
129 pr_err("msgr_init failed to create workqueue\n");
130 _ceph_msgr_exit();
131
132 return -ENOMEM;
133}
134EXPORT_SYMBOL(ceph_msgr_init);
135
136void ceph_msgr_exit(void)
137{
138 BUG_ON(ceph_msgr_wq == NULL);
139
140 _ceph_msgr_exit();
141}
142EXPORT_SYMBOL(ceph_msgr_exit);
143
144void ceph_msgr_flush(void)
145{
146 flush_workqueue(ceph_msgr_wq);
147}
148EXPORT_SYMBOL(ceph_msgr_flush);
149
150
151/*
152 * socket callback functions
153 */
154
155/* data available on socket, or listen socket received a connect */
156static void ceph_data_ready(struct sock *sk, int count_unused)
157{
158 struct ceph_connection *con = sk->sk_user_data;
159
160 if (sk->sk_state != TCP_CLOSE_WAIT) {
161 dout("ceph_data_ready on %p state = %lu, queueing work\n",
162 con, con->state);
163 queue_con(con);
164 }
165}
166
167/* socket has buffer space for writing */
168static void ceph_write_space(struct sock *sk)
169{
170 struct ceph_connection *con = sk->sk_user_data;
171
172 /* only queue to workqueue if there is data we want to write,
173 * and there is sufficient space in the socket buffer to accept
174 * more data. clear SOCK_NOSPACE so that ceph_write_space()
175 * doesn't get called again until try_write() fills the socket
176 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
177 * and net/core/stream.c:sk_stream_write_space().
178 */
179 if (test_bit(WRITE_PENDING, &con->state)) {
180 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
181 dout("ceph_write_space %p queueing write work\n", con);
182 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
183 queue_con(con);
184 }
185 } else {
186 dout("ceph_write_space %p nothing to write\n", con);
187 }
188}
189
190/* socket's state has changed */
191static void ceph_state_change(struct sock *sk)
192{
193 struct ceph_connection *con = sk->sk_user_data;
194
195 dout("ceph_state_change %p state = %lu sk_state = %u\n",
196 con, con->state, sk->sk_state);
197
198 if (test_bit(CLOSED, &con->state))
199 return;
200
201 switch (sk->sk_state) {
202 case TCP_CLOSE:
203 dout("ceph_state_change TCP_CLOSE\n");
204 case TCP_CLOSE_WAIT:
205 dout("ceph_state_change TCP_CLOSE_WAIT\n");
206 if (test_and_set_bit(SOCK_CLOSED, &con->state) == 0) {
207 if (test_bit(CONNECTING, &con->state))
208 con->error_msg = "connection failed";
209 else
210 con->error_msg = "socket closed";
211 queue_con(con);
212 }
213 break;
214 case TCP_ESTABLISHED:
215 dout("ceph_state_change TCP_ESTABLISHED\n");
216 queue_con(con);
217 break;
218 default: /* Everything else is uninteresting */
219 break;
220 }
221}
222
223/*
224 * set up socket callbacks
225 */
226static void set_sock_callbacks(struct socket *sock,
227 struct ceph_connection *con)
228{
229 struct sock *sk = sock->sk;
230 sk->sk_user_data = con;
231 sk->sk_data_ready = ceph_data_ready;
232 sk->sk_write_space = ceph_write_space;
233 sk->sk_state_change = ceph_state_change;
234}
235
236
237/*
238 * socket helpers
239 */
240
241/*
242 * initiate connection to a remote socket.
243 */
244static int ceph_tcp_connect(struct ceph_connection *con)
245{
246 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
247 struct socket *sock;
248 int ret;
249
250 BUG_ON(con->sock);
251 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
252 IPPROTO_TCP, &sock);
253 if (ret)
254 return ret;
255 sock->sk->sk_allocation = GFP_NOFS;
256
257#ifdef CONFIG_LOCKDEP
258 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
259#endif
260
261 set_sock_callbacks(sock, con);
262
263 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
264
265 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
266 O_NONBLOCK);
267 if (ret == -EINPROGRESS) {
268 dout("connect %s EINPROGRESS sk_state = %u\n",
269 ceph_pr_addr(&con->peer_addr.in_addr),
270 sock->sk->sk_state);
271 } else if (ret < 0) {
272 pr_err("connect %s error %d\n",
273 ceph_pr_addr(&con->peer_addr.in_addr), ret);
274 sock_release(sock);
275 con->error_msg = "connect error";
276
277 return ret;
278 }
279 con->sock = sock;
280
281 return 0;
282}
283
284static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
285{
286 struct kvec iov = {buf, len};
287 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
288 int r;
289
290 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
291 if (r == -EAGAIN)
292 r = 0;
293 return r;
294}
295
296/*
297 * write something. @more is true if caller will be sending more data
298 * shortly.
299 */
300static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
301 size_t kvlen, size_t len, int more)
302{
303 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
304 int r;
305
306 if (more)
307 msg.msg_flags |= MSG_MORE;
308 else
309 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
310
311 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
312 if (r == -EAGAIN)
313 r = 0;
314 return r;
315}
316
317static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
318 int offset, size_t size, int more)
319{
320 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
321 int ret;
322
323 ret = kernel_sendpage(sock, page, offset, size, flags);
324 if (ret == -EAGAIN)
325 ret = 0;
326
327 return ret;
328}
329
330
331/*
332 * Shutdown/close the socket for the given connection.
333 */
334static int con_close_socket(struct ceph_connection *con)
335{
336 int rc;
337
338 dout("con_close_socket on %p sock %p\n", con, con->sock);
339 if (!con->sock)
340 return 0;
341 set_bit(SOCK_CLOSED, &con->state);
342 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
343 sock_release(con->sock);
344 con->sock = NULL;
345 clear_bit(SOCK_CLOSED, &con->state);
346 return rc;
347}
348
349/*
350 * Reset a connection. Discard all incoming and outgoing messages
351 * and clear *_seq state.
352 */
353static void ceph_msg_remove(struct ceph_msg *msg)
354{
355 list_del_init(&msg->list_head);
356 ceph_msg_put(msg);
357}
358static void ceph_msg_remove_list(struct list_head *head)
359{
360 while (!list_empty(head)) {
361 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
362 list_head);
363 ceph_msg_remove(msg);
364 }
365}
366
367static void reset_connection(struct ceph_connection *con)
368{
369 /* reset connection, out_queue, msg_ and connect_seq */
370 /* discard existing out_queue and msg_seq */
371 ceph_msg_remove_list(&con->out_queue);
372 ceph_msg_remove_list(&con->out_sent);
373
374 if (con->in_msg) {
375 ceph_msg_put(con->in_msg);
376 con->in_msg = NULL;
377 }
378
379 con->connect_seq = 0;
380 con->out_seq = 0;
381 if (con->out_msg) {
382 ceph_msg_put(con->out_msg);
383 con->out_msg = NULL;
384 }
385 con->in_seq = 0;
386 con->in_seq_acked = 0;
387}
388
389/*
390 * mark a peer down. drop any open connections.
391 */
392void ceph_con_close(struct ceph_connection *con)
393{
394 dout("con_close %p peer %s\n", con,
395 ceph_pr_addr(&con->peer_addr.in_addr));
396 set_bit(CLOSED, &con->state); /* in case there's queued work */
397 clear_bit(STANDBY, &con->state); /* avoid connect_seq bump */
398 clear_bit(LOSSYTX, &con->state); /* so we retry next connect */
399 clear_bit(KEEPALIVE_PENDING, &con->state);
400 clear_bit(WRITE_PENDING, &con->state);
401 mutex_lock(&con->mutex);
402 reset_connection(con);
403 con->peer_global_seq = 0;
404 cancel_delayed_work(&con->work);
405 mutex_unlock(&con->mutex);
406 queue_con(con);
407}
408EXPORT_SYMBOL(ceph_con_close);
409
410/*
411 * Reopen a closed connection, with a new peer address.
412 */
413void ceph_con_open(struct ceph_connection *con, struct ceph_entity_addr *addr)
414{
415 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
416 set_bit(OPENING, &con->state);
417 clear_bit(CLOSED, &con->state);
418 memcpy(&con->peer_addr, addr, sizeof(*addr));
419 con->delay = 0; /* reset backoff memory */
420 queue_con(con);
421}
422EXPORT_SYMBOL(ceph_con_open);
423
424/*
425 * return true if this connection ever successfully opened
426 */
427bool ceph_con_opened(struct ceph_connection *con)
428{
429 return con->connect_seq > 0;
430}
431
432/*
433 * generic get/put
434 */
435struct ceph_connection *ceph_con_get(struct ceph_connection *con)
436{
437 int nref = __atomic_add_unless(&con->nref, 1, 0);
438
439 dout("con_get %p nref = %d -> %d\n", con, nref, nref + 1);
440
441 return nref ? con : NULL;
442}
443
444void ceph_con_put(struct ceph_connection *con)
445{
446 int nref = atomic_dec_return(&con->nref);
447
448 BUG_ON(nref < 0);
449 if (nref == 0) {
450 BUG_ON(con->sock);
451 kfree(con);
452 }
453 dout("con_put %p nref = %d -> %d\n", con, nref + 1, nref);
454}
455
456/*
457 * initialize a new connection.
458 */
459void ceph_con_init(struct ceph_messenger *msgr, struct ceph_connection *con)
460{
461 dout("con_init %p\n", con);
462 memset(con, 0, sizeof(*con));
463 atomic_set(&con->nref, 1);
464 con->msgr = msgr;
465 mutex_init(&con->mutex);
466 INIT_LIST_HEAD(&con->out_queue);
467 INIT_LIST_HEAD(&con->out_sent);
468 INIT_DELAYED_WORK(&con->work, con_work);
469}
470EXPORT_SYMBOL(ceph_con_init);
471
472
473/*
474 * We maintain a global counter to order connection attempts. Get
475 * a unique seq greater than @gt.
476 */
477static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
478{
479 u32 ret;
480
481 spin_lock(&msgr->global_seq_lock);
482 if (msgr->global_seq < gt)
483 msgr->global_seq = gt;
484 ret = ++msgr->global_seq;
485 spin_unlock(&msgr->global_seq_lock);
486 return ret;
487}
488
489static void ceph_con_out_kvec_reset(struct ceph_connection *con)
490{
491 con->out_kvec_left = 0;
492 con->out_kvec_bytes = 0;
493 con->out_kvec_cur = &con->out_kvec[0];
494}
495
496static void ceph_con_out_kvec_add(struct ceph_connection *con,
497 size_t size, void *data)
498{
499 int index;
500
501 index = con->out_kvec_left;
502 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
503
504 con->out_kvec[index].iov_len = size;
505 con->out_kvec[index].iov_base = data;
506 con->out_kvec_left++;
507 con->out_kvec_bytes += size;
508}
509
510/*
511 * Prepare footer for currently outgoing message, and finish things
512 * off. Assumes out_kvec* are already valid.. we just add on to the end.
513 */
514static void prepare_write_message_footer(struct ceph_connection *con)
515{
516 struct ceph_msg *m = con->out_msg;
517 int v = con->out_kvec_left;
518
519 dout("prepare_write_message_footer %p\n", con);
520 con->out_kvec_is_msg = true;
521 con->out_kvec[v].iov_base = &m->footer;
522 con->out_kvec[v].iov_len = sizeof(m->footer);
523 con->out_kvec_bytes += sizeof(m->footer);
524 con->out_kvec_left++;
525 con->out_more = m->more_to_follow;
526 con->out_msg_done = true;
527}
528
529/*
530 * Prepare headers for the next outgoing message.
531 */
532static void prepare_write_message(struct ceph_connection *con)
533{
534 struct ceph_msg *m;
535 u32 crc;
536
537 ceph_con_out_kvec_reset(con);
538 con->out_kvec_is_msg = true;
539 con->out_msg_done = false;
540
541 /* Sneak an ack in there first? If we can get it into the same
542 * TCP packet that's a good thing. */
543 if (con->in_seq > con->in_seq_acked) {
544 con->in_seq_acked = con->in_seq;
545 ceph_con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
546 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
547 ceph_con_out_kvec_add(con, sizeof (con->out_temp_ack),
548 &con->out_temp_ack);
549 }
550
551 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
552 con->out_msg = m;
553
554 /* put message on sent list */
555 ceph_msg_get(m);
556 list_move_tail(&m->list_head, &con->out_sent);
557
558 /*
559 * only assign outgoing seq # if we haven't sent this message
560 * yet. if it is requeued, resend with it's original seq.
561 */
562 if (m->needs_out_seq) {
563 m->hdr.seq = cpu_to_le64(++con->out_seq);
564 m->needs_out_seq = false;
565 }
566#ifdef CONFIG_BLOCK
567 else
568 m->bio_iter = NULL;
569#endif
570
571 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n",
572 m, con->out_seq, le16_to_cpu(m->hdr.type),
573 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
574 le32_to_cpu(m->hdr.data_len),
575 m->nr_pages);
576 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
577
578 /* tag + hdr + front + middle */
579 ceph_con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
580 ceph_con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
581 ceph_con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
582
583 if (m->middle)
584 ceph_con_out_kvec_add(con, m->middle->vec.iov_len,
585 m->middle->vec.iov_base);
586
587 /* fill in crc (except data pages), footer */
588 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
589 con->out_msg->hdr.crc = cpu_to_le32(crc);
590 con->out_msg->footer.flags = CEPH_MSG_FOOTER_COMPLETE;
591
592 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
593 con->out_msg->footer.front_crc = cpu_to_le32(crc);
594 if (m->middle) {
595 crc = crc32c(0, m->middle->vec.iov_base,
596 m->middle->vec.iov_len);
597 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
598 } else
599 con->out_msg->footer.middle_crc = 0;
600 con->out_msg->footer.data_crc = 0;
601 dout("prepare_write_message front_crc %u data_crc %u\n",
602 le32_to_cpu(con->out_msg->footer.front_crc),
603 le32_to_cpu(con->out_msg->footer.middle_crc));
604
605 /* is there a data payload? */
606 if (le32_to_cpu(m->hdr.data_len) > 0) {
607 /* initialize page iterator */
608 con->out_msg_pos.page = 0;
609 if (m->pages)
610 con->out_msg_pos.page_pos = m->page_alignment;
611 else
612 con->out_msg_pos.page_pos = 0;
613 con->out_msg_pos.data_pos = 0;
614 con->out_msg_pos.did_page_crc = false;
615 con->out_more = 1; /* data + footer will follow */
616 } else {
617 /* no, queue up footer too and be done */
618 prepare_write_message_footer(con);
619 }
620
621 set_bit(WRITE_PENDING, &con->state);
622}
623
624/*
625 * Prepare an ack.
626 */
627static void prepare_write_ack(struct ceph_connection *con)
628{
629 dout("prepare_write_ack %p %llu -> %llu\n", con,
630 con->in_seq_acked, con->in_seq);
631 con->in_seq_acked = con->in_seq;
632
633 ceph_con_out_kvec_reset(con);
634
635 ceph_con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
636
637 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
638 ceph_con_out_kvec_add(con, sizeof (con->out_temp_ack),
639 &con->out_temp_ack);
640
641 con->out_more = 1; /* more will follow.. eventually.. */
642 set_bit(WRITE_PENDING, &con->state);
643}
644
645/*
646 * Prepare to write keepalive byte.
647 */
648static void prepare_write_keepalive(struct ceph_connection *con)
649{
650 dout("prepare_write_keepalive %p\n", con);
651 ceph_con_out_kvec_reset(con);
652 ceph_con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
653 set_bit(WRITE_PENDING, &con->state);
654}
655
656/*
657 * Connection negotiation.
658 */
659
660static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
661 int *auth_proto)
662{
663 struct ceph_auth_handshake *auth;
664
665 if (!con->ops->get_authorizer) {
666 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
667 con->out_connect.authorizer_len = 0;
668
669 return NULL;
670 }
671
672 /* Can't hold the mutex while getting authorizer */
673
674 mutex_unlock(&con->mutex);
675
676 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
677
678 mutex_lock(&con->mutex);
679
680 if (IS_ERR(auth))
681 return auth;
682 if (test_bit(CLOSED, &con->state) || test_bit(OPENING, &con->state))
683 return ERR_PTR(-EAGAIN);
684
685 con->auth_reply_buf = auth->authorizer_reply_buf;
686 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
687
688
689 return auth;
690}
691
692/*
693 * We connected to a peer and are saying hello.
694 */
695static void prepare_write_banner(struct ceph_connection *con)
696{
697 ceph_con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
698 ceph_con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
699 &con->msgr->my_enc_addr);
700
701 con->out_more = 0;
702 set_bit(WRITE_PENDING, &con->state);
703}
704
705static int prepare_write_connect(struct ceph_connection *con)
706{
707 unsigned int global_seq = get_global_seq(con->msgr, 0);
708 int proto;
709 int auth_proto;
710 struct ceph_auth_handshake *auth;
711
712 switch (con->peer_name.type) {
713 case CEPH_ENTITY_TYPE_MON:
714 proto = CEPH_MONC_PROTOCOL;
715 break;
716 case CEPH_ENTITY_TYPE_OSD:
717 proto = CEPH_OSDC_PROTOCOL;
718 break;
719 case CEPH_ENTITY_TYPE_MDS:
720 proto = CEPH_MDSC_PROTOCOL;
721 break;
722 default:
723 BUG();
724 }
725
726 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
727 con->connect_seq, global_seq, proto);
728
729 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
730 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
731 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
732 con->out_connect.global_seq = cpu_to_le32(global_seq);
733 con->out_connect.protocol_version = cpu_to_le32(proto);
734 con->out_connect.flags = 0;
735
736 auth_proto = CEPH_AUTH_UNKNOWN;
737 auth = get_connect_authorizer(con, &auth_proto);
738 if (IS_ERR(auth))
739 return PTR_ERR(auth);
740
741 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
742 con->out_connect.authorizer_len = auth ?
743 cpu_to_le32(auth->authorizer_buf_len) : 0;
744
745 ceph_con_out_kvec_add(con, sizeof (con->out_connect),
746 &con->out_connect);
747 if (auth && auth->authorizer_buf_len)
748 ceph_con_out_kvec_add(con, auth->authorizer_buf_len,
749 auth->authorizer_buf);
750
751 con->out_more = 0;
752 set_bit(WRITE_PENDING, &con->state);
753
754 return 0;
755}
756
757/*
758 * write as much of pending kvecs to the socket as we can.
759 * 1 -> done
760 * 0 -> socket full, but more to do
761 * <0 -> error
762 */
763static int write_partial_kvec(struct ceph_connection *con)
764{
765 int ret;
766
767 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
768 while (con->out_kvec_bytes > 0) {
769 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
770 con->out_kvec_left, con->out_kvec_bytes,
771 con->out_more);
772 if (ret <= 0)
773 goto out;
774 con->out_kvec_bytes -= ret;
775 if (con->out_kvec_bytes == 0)
776 break; /* done */
777
778 /* account for full iov entries consumed */
779 while (ret >= con->out_kvec_cur->iov_len) {
780 BUG_ON(!con->out_kvec_left);
781 ret -= con->out_kvec_cur->iov_len;
782 con->out_kvec_cur++;
783 con->out_kvec_left--;
784 }
785 /* and for a partially-consumed entry */
786 if (ret) {
787 con->out_kvec_cur->iov_len -= ret;
788 con->out_kvec_cur->iov_base += ret;
789 }
790 }
791 con->out_kvec_left = 0;
792 con->out_kvec_is_msg = false;
793 ret = 1;
794out:
795 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
796 con->out_kvec_bytes, con->out_kvec_left, ret);
797 return ret; /* done! */
798}
799
800#ifdef CONFIG_BLOCK
801static void init_bio_iter(struct bio *bio, struct bio **iter, int *seg)
802{
803 if (!bio) {
804 *iter = NULL;
805 *seg = 0;
806 return;
807 }
808 *iter = bio;
809 *seg = bio->bi_idx;
810}
811
812static void iter_bio_next(struct bio **bio_iter, int *seg)
813{
814 if (*bio_iter == NULL)
815 return;
816
817 BUG_ON(*seg >= (*bio_iter)->bi_vcnt);
818
819 (*seg)++;
820 if (*seg == (*bio_iter)->bi_vcnt)
821 init_bio_iter((*bio_iter)->bi_next, bio_iter, seg);
822}
823#endif
824
825/*
826 * Write as much message data payload as we can. If we finish, queue
827 * up the footer.
828 * 1 -> done, footer is now queued in out_kvec[].
829 * 0 -> socket full, but more to do
830 * <0 -> error
831 */
832static int write_partial_msg_pages(struct ceph_connection *con)
833{
834 struct ceph_msg *msg = con->out_msg;
835 unsigned int data_len = le32_to_cpu(msg->hdr.data_len);
836 size_t len;
837 bool do_datacrc = !con->msgr->nocrc;
838 int ret;
839 int total_max_write;
840 int in_trail = 0;
841 size_t trail_len = (msg->trail ? msg->trail->length : 0);
842
843 dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
844 con, con->out_msg, con->out_msg_pos.page, con->out_msg->nr_pages,
845 con->out_msg_pos.page_pos);
846
847#ifdef CONFIG_BLOCK
848 if (msg->bio && !msg->bio_iter)
849 init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg);
850#endif
851
852 while (data_len > con->out_msg_pos.data_pos) {
853 struct page *page = NULL;
854 int max_write = PAGE_SIZE;
855 int bio_offset = 0;
856
857 total_max_write = data_len - trail_len -
858 con->out_msg_pos.data_pos;
859
860 /*
861 * if we are calculating the data crc (the default), we need
862 * to map the page. if our pages[] has been revoked, use the
863 * zero page.
864 */
865
866 /* have we reached the trail part of the data? */
867 if (con->out_msg_pos.data_pos >= data_len - trail_len) {
868 in_trail = 1;
869
870 total_max_write = data_len - con->out_msg_pos.data_pos;
871
872 page = list_first_entry(&msg->trail->head,
873 struct page, lru);
874 max_write = PAGE_SIZE;
875 } else if (msg->pages) {
876 page = msg->pages[con->out_msg_pos.page];
877 } else if (msg->pagelist) {
878 page = list_first_entry(&msg->pagelist->head,
879 struct page, lru);
880#ifdef CONFIG_BLOCK
881 } else if (msg->bio) {
882 struct bio_vec *bv;
883
884 bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg);
885 page = bv->bv_page;
886 bio_offset = bv->bv_offset;
887 max_write = bv->bv_len;
888#endif
889 } else {
890 page = zero_page;
891 }
892 len = min_t(int, max_write - con->out_msg_pos.page_pos,
893 total_max_write);
894
895 if (do_datacrc && !con->out_msg_pos.did_page_crc) {
896 void *base;
897 u32 crc;
898 u32 tmpcrc = le32_to_cpu(con->out_msg->footer.data_crc);
899 char *kaddr;
900
901 kaddr = kmap(page);
902 BUG_ON(kaddr == NULL);
903 base = kaddr + con->out_msg_pos.page_pos + bio_offset;
904 crc = crc32c(tmpcrc, base, len);
905 con->out_msg->footer.data_crc = cpu_to_le32(crc);
906 con->out_msg_pos.did_page_crc = true;
907 }
908 ret = ceph_tcp_sendpage(con->sock, page,
909 con->out_msg_pos.page_pos + bio_offset,
910 len, 1);
911
912 if (do_datacrc)
913 kunmap(page);
914
915 if (ret <= 0)
916 goto out;
917
918 con->out_msg_pos.data_pos += ret;
919 con->out_msg_pos.page_pos += ret;
920 if (ret == len) {
921 con->out_msg_pos.page_pos = 0;
922 con->out_msg_pos.page++;
923 con->out_msg_pos.did_page_crc = false;
924 if (in_trail)
925 list_move_tail(&page->lru,
926 &msg->trail->head);
927 else if (msg->pagelist)
928 list_move_tail(&page->lru,
929 &msg->pagelist->head);
930#ifdef CONFIG_BLOCK
931 else if (msg->bio)
932 iter_bio_next(&msg->bio_iter, &msg->bio_seg);
933#endif
934 }
935 }
936
937 dout("write_partial_msg_pages %p msg %p done\n", con, msg);
938
939 /* prepare and queue up footer, too */
940 if (!do_datacrc)
941 con->out_msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
942 ceph_con_out_kvec_reset(con);
943 prepare_write_message_footer(con);
944 ret = 1;
945out:
946 return ret;
947}
948
949/*
950 * write some zeros
951 */
952static int write_partial_skip(struct ceph_connection *con)
953{
954 int ret;
955
956 while (con->out_skip > 0) {
957 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
958
959 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, 1);
960 if (ret <= 0)
961 goto out;
962 con->out_skip -= ret;
963 }
964 ret = 1;
965out:
966 return ret;
967}
968
969/*
970 * Prepare to read connection handshake, or an ack.
971 */
972static void prepare_read_banner(struct ceph_connection *con)
973{
974 dout("prepare_read_banner %p\n", con);
975 con->in_base_pos = 0;
976}
977
978static void prepare_read_connect(struct ceph_connection *con)
979{
980 dout("prepare_read_connect %p\n", con);
981 con->in_base_pos = 0;
982}
983
984static void prepare_read_ack(struct ceph_connection *con)
985{
986 dout("prepare_read_ack %p\n", con);
987 con->in_base_pos = 0;
988}
989
990static void prepare_read_tag(struct ceph_connection *con)
991{
992 dout("prepare_read_tag %p\n", con);
993 con->in_base_pos = 0;
994 con->in_tag = CEPH_MSGR_TAG_READY;
995}
996
997/*
998 * Prepare to read a message.
999 */
1000static int prepare_read_message(struct ceph_connection *con)
1001{
1002 dout("prepare_read_message %p\n", con);
1003 BUG_ON(con->in_msg != NULL);
1004 con->in_base_pos = 0;
1005 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1006 return 0;
1007}
1008
1009
1010static int read_partial(struct ceph_connection *con,
1011 int end, int size, void *object)
1012{
1013 while (con->in_base_pos < end) {
1014 int left = end - con->in_base_pos;
1015 int have = size - left;
1016 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1017 if (ret <= 0)
1018 return ret;
1019 con->in_base_pos += ret;
1020 }
1021 return 1;
1022}
1023
1024
1025/*
1026 * Read all or part of the connect-side handshake on a new connection
1027 */
1028static int read_partial_banner(struct ceph_connection *con)
1029{
1030 int size;
1031 int end;
1032 int ret;
1033
1034 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1035
1036 /* peer's banner */
1037 size = strlen(CEPH_BANNER);
1038 end = size;
1039 ret = read_partial(con, end, size, con->in_banner);
1040 if (ret <= 0)
1041 goto out;
1042
1043 size = sizeof (con->actual_peer_addr);
1044 end += size;
1045 ret = read_partial(con, end, size, &con->actual_peer_addr);
1046 if (ret <= 0)
1047 goto out;
1048
1049 size = sizeof (con->peer_addr_for_me);
1050 end += size;
1051 ret = read_partial(con, end, size, &con->peer_addr_for_me);
1052 if (ret <= 0)
1053 goto out;
1054
1055out:
1056 return ret;
1057}
1058
1059static int read_partial_connect(struct ceph_connection *con)
1060{
1061 int size;
1062 int end;
1063 int ret;
1064
1065 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1066
1067 size = sizeof (con->in_reply);
1068 end = size;
1069 ret = read_partial(con, end, size, &con->in_reply);
1070 if (ret <= 0)
1071 goto out;
1072
1073 size = le32_to_cpu(con->in_reply.authorizer_len);
1074 end += size;
1075 ret = read_partial(con, end, size, con->auth_reply_buf);
1076 if (ret <= 0)
1077 goto out;
1078
1079 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1080 con, (int)con->in_reply.tag,
1081 le32_to_cpu(con->in_reply.connect_seq),
1082 le32_to_cpu(con->in_reply.global_seq));
1083out:
1084 return ret;
1085
1086}
1087
1088/*
1089 * Verify the hello banner looks okay.
1090 */
1091static int verify_hello(struct ceph_connection *con)
1092{
1093 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1094 pr_err("connect to %s got bad banner\n",
1095 ceph_pr_addr(&con->peer_addr.in_addr));
1096 con->error_msg = "protocol error, bad banner";
1097 return -1;
1098 }
1099 return 0;
1100}
1101
1102static bool addr_is_blank(struct sockaddr_storage *ss)
1103{
1104 switch (ss->ss_family) {
1105 case AF_INET:
1106 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1107 case AF_INET6:
1108 return
1109 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1110 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1111 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1112 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1113 }
1114 return false;
1115}
1116
1117static int addr_port(struct sockaddr_storage *ss)
1118{
1119 switch (ss->ss_family) {
1120 case AF_INET:
1121 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1122 case AF_INET6:
1123 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1124 }
1125 return 0;
1126}
1127
1128static void addr_set_port(struct sockaddr_storage *ss, int p)
1129{
1130 switch (ss->ss_family) {
1131 case AF_INET:
1132 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1133 break;
1134 case AF_INET6:
1135 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1136 break;
1137 }
1138}
1139
1140/*
1141 * Unlike other *_pton function semantics, zero indicates success.
1142 */
1143static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1144 char delim, const char **ipend)
1145{
1146 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1147 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1148
1149 memset(ss, 0, sizeof(*ss));
1150
1151 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1152 ss->ss_family = AF_INET;
1153 return 0;
1154 }
1155
1156 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1157 ss->ss_family = AF_INET6;
1158 return 0;
1159 }
1160
1161 return -EINVAL;
1162}
1163
1164/*
1165 * Extract hostname string and resolve using kernel DNS facility.
1166 */
1167#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1168static int ceph_dns_resolve_name(const char *name, size_t namelen,
1169 struct sockaddr_storage *ss, char delim, const char **ipend)
1170{
1171 const char *end, *delim_p;
1172 char *colon_p, *ip_addr = NULL;
1173 int ip_len, ret;
1174
1175 /*
1176 * The end of the hostname occurs immediately preceding the delimiter or
1177 * the port marker (':') where the delimiter takes precedence.
1178 */
1179 delim_p = memchr(name, delim, namelen);
1180 colon_p = memchr(name, ':', namelen);
1181
1182 if (delim_p && colon_p)
1183 end = delim_p < colon_p ? delim_p : colon_p;
1184 else if (!delim_p && colon_p)
1185 end = colon_p;
1186 else {
1187 end = delim_p;
1188 if (!end) /* case: hostname:/ */
1189 end = name + namelen;
1190 }
1191
1192 if (end <= name)
1193 return -EINVAL;
1194
1195 /* do dns_resolve upcall */
1196 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1197 if (ip_len > 0)
1198 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1199 else
1200 ret = -ESRCH;
1201
1202 kfree(ip_addr);
1203
1204 *ipend = end;
1205
1206 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1207 ret, ret ? "failed" : ceph_pr_addr(ss));
1208
1209 return ret;
1210}
1211#else
1212static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1213 struct sockaddr_storage *ss, char delim, const char **ipend)
1214{
1215 return -EINVAL;
1216}
1217#endif
1218
1219/*
1220 * Parse a server name (IP or hostname). If a valid IP address is not found
1221 * then try to extract a hostname to resolve using userspace DNS upcall.
1222 */
1223static int ceph_parse_server_name(const char *name, size_t namelen,
1224 struct sockaddr_storage *ss, char delim, const char **ipend)
1225{
1226 int ret;
1227
1228 ret = ceph_pton(name, namelen, ss, delim, ipend);
1229 if (ret)
1230 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1231
1232 return ret;
1233}
1234
1235/*
1236 * Parse an ip[:port] list into an addr array. Use the default
1237 * monitor port if a port isn't specified.
1238 */
1239int ceph_parse_ips(const char *c, const char *end,
1240 struct ceph_entity_addr *addr,
1241 int max_count, int *count)
1242{
1243 int i, ret = -EINVAL;
1244 const char *p = c;
1245
1246 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1247 for (i = 0; i < max_count; i++) {
1248 const char *ipend;
1249 struct sockaddr_storage *ss = &addr[i].in_addr;
1250 int port;
1251 char delim = ',';
1252
1253 if (*p == '[') {
1254 delim = ']';
1255 p++;
1256 }
1257
1258 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1259 if (ret)
1260 goto bad;
1261 ret = -EINVAL;
1262
1263 p = ipend;
1264
1265 if (delim == ']') {
1266 if (*p != ']') {
1267 dout("missing matching ']'\n");
1268 goto bad;
1269 }
1270 p++;
1271 }
1272
1273 /* port? */
1274 if (p < end && *p == ':') {
1275 port = 0;
1276 p++;
1277 while (p < end && *p >= '0' && *p <= '9') {
1278 port = (port * 10) + (*p - '0');
1279 p++;
1280 }
1281 if (port > 65535 || port == 0)
1282 goto bad;
1283 } else {
1284 port = CEPH_MON_PORT;
1285 }
1286
1287 addr_set_port(ss, port);
1288
1289 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1290
1291 if (p == end)
1292 break;
1293 if (*p != ',')
1294 goto bad;
1295 p++;
1296 }
1297
1298 if (p != end)
1299 goto bad;
1300
1301 if (count)
1302 *count = i + 1;
1303 return 0;
1304
1305bad:
1306 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1307 return ret;
1308}
1309EXPORT_SYMBOL(ceph_parse_ips);
1310
1311static int process_banner(struct ceph_connection *con)
1312{
1313 dout("process_banner on %p\n", con);
1314
1315 if (verify_hello(con) < 0)
1316 return -1;
1317
1318 ceph_decode_addr(&con->actual_peer_addr);
1319 ceph_decode_addr(&con->peer_addr_for_me);
1320
1321 /*
1322 * Make sure the other end is who we wanted. note that the other
1323 * end may not yet know their ip address, so if it's 0.0.0.0, give
1324 * them the benefit of the doubt.
1325 */
1326 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1327 sizeof(con->peer_addr)) != 0 &&
1328 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1329 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1330 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1331 ceph_pr_addr(&con->peer_addr.in_addr),
1332 (int)le32_to_cpu(con->peer_addr.nonce),
1333 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1334 (int)le32_to_cpu(con->actual_peer_addr.nonce));
1335 con->error_msg = "wrong peer at address";
1336 return -1;
1337 }
1338
1339 /*
1340 * did we learn our address?
1341 */
1342 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1343 int port = addr_port(&con->msgr->inst.addr.in_addr);
1344
1345 memcpy(&con->msgr->inst.addr.in_addr,
1346 &con->peer_addr_for_me.in_addr,
1347 sizeof(con->peer_addr_for_me.in_addr));
1348 addr_set_port(&con->msgr->inst.addr.in_addr, port);
1349 encode_my_addr(con->msgr);
1350 dout("process_banner learned my addr is %s\n",
1351 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
1352 }
1353
1354 set_bit(NEGOTIATING, &con->state);
1355 prepare_read_connect(con);
1356 return 0;
1357}
1358
1359static void fail_protocol(struct ceph_connection *con)
1360{
1361 reset_connection(con);
1362 set_bit(CLOSED, &con->state); /* in case there's queued work */
1363
1364 mutex_unlock(&con->mutex);
1365 if (con->ops->bad_proto)
1366 con->ops->bad_proto(con);
1367 mutex_lock(&con->mutex);
1368}
1369
1370static int process_connect(struct ceph_connection *con)
1371{
1372 u64 sup_feat = con->msgr->supported_features;
1373 u64 req_feat = con->msgr->required_features;
1374 u64 server_feat = le64_to_cpu(con->in_reply.features);
1375 int ret;
1376
1377 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1378
1379 switch (con->in_reply.tag) {
1380 case CEPH_MSGR_TAG_FEATURES:
1381 pr_err("%s%lld %s feature set mismatch,"
1382 " my %llx < server's %llx, missing %llx\n",
1383 ENTITY_NAME(con->peer_name),
1384 ceph_pr_addr(&con->peer_addr.in_addr),
1385 sup_feat, server_feat, server_feat & ~sup_feat);
1386 con->error_msg = "missing required protocol features";
1387 fail_protocol(con);
1388 return -1;
1389
1390 case CEPH_MSGR_TAG_BADPROTOVER:
1391 pr_err("%s%lld %s protocol version mismatch,"
1392 " my %d != server's %d\n",
1393 ENTITY_NAME(con->peer_name),
1394 ceph_pr_addr(&con->peer_addr.in_addr),
1395 le32_to_cpu(con->out_connect.protocol_version),
1396 le32_to_cpu(con->in_reply.protocol_version));
1397 con->error_msg = "protocol version mismatch";
1398 fail_protocol(con);
1399 return -1;
1400
1401 case CEPH_MSGR_TAG_BADAUTHORIZER:
1402 con->auth_retry++;
1403 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1404 con->auth_retry);
1405 if (con->auth_retry == 2) {
1406 con->error_msg = "connect authorization failure";
1407 return -1;
1408 }
1409 con->auth_retry = 1;
1410 ceph_con_out_kvec_reset(con);
1411 ret = prepare_write_connect(con);
1412 if (ret < 0)
1413 return ret;
1414 prepare_read_connect(con);
1415 break;
1416
1417 case CEPH_MSGR_TAG_RESETSESSION:
1418 /*
1419 * If we connected with a large connect_seq but the peer
1420 * has no record of a session with us (no connection, or
1421 * connect_seq == 0), they will send RESETSESION to indicate
1422 * that they must have reset their session, and may have
1423 * dropped messages.
1424 */
1425 dout("process_connect got RESET peer seq %u\n",
1426 le32_to_cpu(con->in_reply.connect_seq));
1427 pr_err("%s%lld %s connection reset\n",
1428 ENTITY_NAME(con->peer_name),
1429 ceph_pr_addr(&con->peer_addr.in_addr));
1430 reset_connection(con);
1431 ceph_con_out_kvec_reset(con);
1432 ret = prepare_write_connect(con);
1433 if (ret < 0)
1434 return ret;
1435 prepare_read_connect(con);
1436
1437 /* Tell ceph about it. */
1438 mutex_unlock(&con->mutex);
1439 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
1440 if (con->ops->peer_reset)
1441 con->ops->peer_reset(con);
1442 mutex_lock(&con->mutex);
1443 if (test_bit(CLOSED, &con->state) ||
1444 test_bit(OPENING, &con->state))
1445 return -EAGAIN;
1446 break;
1447
1448 case CEPH_MSGR_TAG_RETRY_SESSION:
1449 /*
1450 * If we sent a smaller connect_seq than the peer has, try
1451 * again with a larger value.
1452 */
1453 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
1454 le32_to_cpu(con->out_connect.connect_seq),
1455 le32_to_cpu(con->in_reply.connect_seq));
1456 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
1457 ceph_con_out_kvec_reset(con);
1458 ret = prepare_write_connect(con);
1459 if (ret < 0)
1460 return ret;
1461 prepare_read_connect(con);
1462 break;
1463
1464 case CEPH_MSGR_TAG_RETRY_GLOBAL:
1465 /*
1466 * If we sent a smaller global_seq than the peer has, try
1467 * again with a larger value.
1468 */
1469 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1470 con->peer_global_seq,
1471 le32_to_cpu(con->in_reply.global_seq));
1472 get_global_seq(con->msgr,
1473 le32_to_cpu(con->in_reply.global_seq));
1474 ceph_con_out_kvec_reset(con);
1475 ret = prepare_write_connect(con);
1476 if (ret < 0)
1477 return ret;
1478 prepare_read_connect(con);
1479 break;
1480
1481 case CEPH_MSGR_TAG_READY:
1482 if (req_feat & ~server_feat) {
1483 pr_err("%s%lld %s protocol feature mismatch,"
1484 " my required %llx > server's %llx, need %llx\n",
1485 ENTITY_NAME(con->peer_name),
1486 ceph_pr_addr(&con->peer_addr.in_addr),
1487 req_feat, server_feat, req_feat & ~server_feat);
1488 con->error_msg = "missing required protocol features";
1489 fail_protocol(con);
1490 return -1;
1491 }
1492 clear_bit(CONNECTING, &con->state);
1493 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
1494 con->connect_seq++;
1495 con->peer_features = server_feat;
1496 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1497 con->peer_global_seq,
1498 le32_to_cpu(con->in_reply.connect_seq),
1499 con->connect_seq);
1500 WARN_ON(con->connect_seq !=
1501 le32_to_cpu(con->in_reply.connect_seq));
1502
1503 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
1504 set_bit(LOSSYTX, &con->state);
1505
1506 prepare_read_tag(con);
1507 break;
1508
1509 case CEPH_MSGR_TAG_WAIT:
1510 /*
1511 * If there is a connection race (we are opening
1512 * connections to each other), one of us may just have
1513 * to WAIT. This shouldn't happen if we are the
1514 * client.
1515 */
1516 pr_err("process_connect got WAIT as client\n");
1517 con->error_msg = "protocol error, got WAIT as client";
1518 return -1;
1519
1520 default:
1521 pr_err("connect protocol error, will retry\n");
1522 con->error_msg = "protocol error, garbage tag during connect";
1523 return -1;
1524 }
1525 return 0;
1526}
1527
1528
1529/*
1530 * read (part of) an ack
1531 */
1532static int read_partial_ack(struct ceph_connection *con)
1533{
1534 int size = sizeof (con->in_temp_ack);
1535 int end = size;
1536
1537 return read_partial(con, end, size, &con->in_temp_ack);
1538}
1539
1540
1541/*
1542 * We can finally discard anything that's been acked.
1543 */
1544static void process_ack(struct ceph_connection *con)
1545{
1546 struct ceph_msg *m;
1547 u64 ack = le64_to_cpu(con->in_temp_ack);
1548 u64 seq;
1549
1550 while (!list_empty(&con->out_sent)) {
1551 m = list_first_entry(&con->out_sent, struct ceph_msg,
1552 list_head);
1553 seq = le64_to_cpu(m->hdr.seq);
1554 if (seq > ack)
1555 break;
1556 dout("got ack for seq %llu type %d at %p\n", seq,
1557 le16_to_cpu(m->hdr.type), m);
1558 m->ack_stamp = jiffies;
1559 ceph_msg_remove(m);
1560 }
1561 prepare_read_tag(con);
1562}
1563
1564
1565
1566
1567static int read_partial_message_section(struct ceph_connection *con,
1568 struct kvec *section,
1569 unsigned int sec_len, u32 *crc)
1570{
1571 int ret, left;
1572
1573 BUG_ON(!section);
1574
1575 while (section->iov_len < sec_len) {
1576 BUG_ON(section->iov_base == NULL);
1577 left = sec_len - section->iov_len;
1578 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
1579 section->iov_len, left);
1580 if (ret <= 0)
1581 return ret;
1582 section->iov_len += ret;
1583 }
1584 if (section->iov_len == sec_len)
1585 *crc = crc32c(0, section->iov_base, section->iov_len);
1586
1587 return 1;
1588}
1589
1590static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con,
1591 struct ceph_msg_header *hdr,
1592 int *skip);
1593
1594
1595static int read_partial_message_pages(struct ceph_connection *con,
1596 struct page **pages,
1597 unsigned int data_len, bool do_datacrc)
1598{
1599 void *p;
1600 int ret;
1601 int left;
1602
1603 left = min((int)(data_len - con->in_msg_pos.data_pos),
1604 (int)(PAGE_SIZE - con->in_msg_pos.page_pos));
1605 /* (page) data */
1606 BUG_ON(pages == NULL);
1607 p = kmap(pages[con->in_msg_pos.page]);
1608 ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos,
1609 left);
1610 if (ret > 0 && do_datacrc)
1611 con->in_data_crc =
1612 crc32c(con->in_data_crc,
1613 p + con->in_msg_pos.page_pos, ret);
1614 kunmap(pages[con->in_msg_pos.page]);
1615 if (ret <= 0)
1616 return ret;
1617 con->in_msg_pos.data_pos += ret;
1618 con->in_msg_pos.page_pos += ret;
1619 if (con->in_msg_pos.page_pos == PAGE_SIZE) {
1620 con->in_msg_pos.page_pos = 0;
1621 con->in_msg_pos.page++;
1622 }
1623
1624 return ret;
1625}
1626
1627#ifdef CONFIG_BLOCK
1628static int read_partial_message_bio(struct ceph_connection *con,
1629 struct bio **bio_iter, int *bio_seg,
1630 unsigned int data_len, bool do_datacrc)
1631{
1632 struct bio_vec *bv = bio_iovec_idx(*bio_iter, *bio_seg);
1633 void *p;
1634 int ret, left;
1635
1636 if (IS_ERR(bv))
1637 return PTR_ERR(bv);
1638
1639 left = min((int)(data_len - con->in_msg_pos.data_pos),
1640 (int)(bv->bv_len - con->in_msg_pos.page_pos));
1641
1642 p = kmap(bv->bv_page) + bv->bv_offset;
1643
1644 ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos,
1645 left);
1646 if (ret > 0 && do_datacrc)
1647 con->in_data_crc =
1648 crc32c(con->in_data_crc,
1649 p + con->in_msg_pos.page_pos, ret);
1650 kunmap(bv->bv_page);
1651 if (ret <= 0)
1652 return ret;
1653 con->in_msg_pos.data_pos += ret;
1654 con->in_msg_pos.page_pos += ret;
1655 if (con->in_msg_pos.page_pos == bv->bv_len) {
1656 con->in_msg_pos.page_pos = 0;
1657 iter_bio_next(bio_iter, bio_seg);
1658 }
1659
1660 return ret;
1661}
1662#endif
1663
1664/*
1665 * read (part of) a message.
1666 */
1667static int read_partial_message(struct ceph_connection *con)
1668{
1669 struct ceph_msg *m = con->in_msg;
1670 int size;
1671 int end;
1672 int ret;
1673 unsigned int front_len, middle_len, data_len;
1674 bool do_datacrc = !con->msgr->nocrc;
1675 int skip;
1676 u64 seq;
1677 u32 crc;
1678
1679 dout("read_partial_message con %p msg %p\n", con, m);
1680
1681 /* header */
1682 size = sizeof (con->in_hdr);
1683 end = size;
1684 ret = read_partial(con, end, size, &con->in_hdr);
1685 if (ret <= 0)
1686 return ret;
1687
1688 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
1689 if (cpu_to_le32(crc) != con->in_hdr.crc) {
1690 pr_err("read_partial_message bad hdr "
1691 " crc %u != expected %u\n",
1692 crc, con->in_hdr.crc);
1693 return -EBADMSG;
1694 }
1695
1696 front_len = le32_to_cpu(con->in_hdr.front_len);
1697 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
1698 return -EIO;
1699 middle_len = le32_to_cpu(con->in_hdr.middle_len);
1700 if (middle_len > CEPH_MSG_MAX_DATA_LEN)
1701 return -EIO;
1702 data_len = le32_to_cpu(con->in_hdr.data_len);
1703 if (data_len > CEPH_MSG_MAX_DATA_LEN)
1704 return -EIO;
1705
1706 /* verify seq# */
1707 seq = le64_to_cpu(con->in_hdr.seq);
1708 if ((s64)seq - (s64)con->in_seq < 1) {
1709 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1710 ENTITY_NAME(con->peer_name),
1711 ceph_pr_addr(&con->peer_addr.in_addr),
1712 seq, con->in_seq + 1);
1713 con->in_base_pos = -front_len - middle_len - data_len -
1714 sizeof(m->footer);
1715 con->in_tag = CEPH_MSGR_TAG_READY;
1716 return 0;
1717 } else if ((s64)seq - (s64)con->in_seq > 1) {
1718 pr_err("read_partial_message bad seq %lld expected %lld\n",
1719 seq, con->in_seq + 1);
1720 con->error_msg = "bad message sequence # for incoming message";
1721 return -EBADMSG;
1722 }
1723
1724 /* allocate message? */
1725 if (!con->in_msg) {
1726 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
1727 con->in_hdr.front_len, con->in_hdr.data_len);
1728 skip = 0;
1729 con->in_msg = ceph_alloc_msg(con, &con->in_hdr, &skip);
1730 if (skip) {
1731 /* skip this message */
1732 dout("alloc_msg said skip message\n");
1733 BUG_ON(con->in_msg);
1734 con->in_base_pos = -front_len - middle_len - data_len -
1735 sizeof(m->footer);
1736 con->in_tag = CEPH_MSGR_TAG_READY;
1737 con->in_seq++;
1738 return 0;
1739 }
1740 if (!con->in_msg) {
1741 con->error_msg =
1742 "error allocating memory for incoming message";
1743 return -ENOMEM;
1744 }
1745 m = con->in_msg;
1746 m->front.iov_len = 0; /* haven't read it yet */
1747 if (m->middle)
1748 m->middle->vec.iov_len = 0;
1749
1750 con->in_msg_pos.page = 0;
1751 if (m->pages)
1752 con->in_msg_pos.page_pos = m->page_alignment;
1753 else
1754 con->in_msg_pos.page_pos = 0;
1755 con->in_msg_pos.data_pos = 0;
1756 }
1757
1758 /* front */
1759 ret = read_partial_message_section(con, &m->front, front_len,
1760 &con->in_front_crc);
1761 if (ret <= 0)
1762 return ret;
1763
1764 /* middle */
1765 if (m->middle) {
1766 ret = read_partial_message_section(con, &m->middle->vec,
1767 middle_len,
1768 &con->in_middle_crc);
1769 if (ret <= 0)
1770 return ret;
1771 }
1772#ifdef CONFIG_BLOCK
1773 if (m->bio && !m->bio_iter)
1774 init_bio_iter(m->bio, &m->bio_iter, &m->bio_seg);
1775#endif
1776
1777 /* (page) data */
1778 while (con->in_msg_pos.data_pos < data_len) {
1779 if (m->pages) {
1780 ret = read_partial_message_pages(con, m->pages,
1781 data_len, do_datacrc);
1782 if (ret <= 0)
1783 return ret;
1784#ifdef CONFIG_BLOCK
1785 } else if (m->bio) {
1786
1787 ret = read_partial_message_bio(con,
1788 &m->bio_iter, &m->bio_seg,
1789 data_len, do_datacrc);
1790 if (ret <= 0)
1791 return ret;
1792#endif
1793 } else {
1794 BUG_ON(1);
1795 }
1796 }
1797
1798 /* footer */
1799 size = sizeof (m->footer);
1800 end += size;
1801 ret = read_partial(con, end, size, &m->footer);
1802 if (ret <= 0)
1803 return ret;
1804
1805 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1806 m, front_len, m->footer.front_crc, middle_len,
1807 m->footer.middle_crc, data_len, m->footer.data_crc);
1808
1809 /* crc ok? */
1810 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
1811 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1812 m, con->in_front_crc, m->footer.front_crc);
1813 return -EBADMSG;
1814 }
1815 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
1816 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1817 m, con->in_middle_crc, m->footer.middle_crc);
1818 return -EBADMSG;
1819 }
1820 if (do_datacrc &&
1821 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
1822 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
1823 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
1824 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
1825 return -EBADMSG;
1826 }
1827
1828 return 1; /* done! */
1829}
1830
1831/*
1832 * Process message. This happens in the worker thread. The callback should
1833 * be careful not to do anything that waits on other incoming messages or it
1834 * may deadlock.
1835 */
1836static void process_message(struct ceph_connection *con)
1837{
1838 struct ceph_msg *msg;
1839
1840 msg = con->in_msg;
1841 con->in_msg = NULL;
1842
1843 /* if first message, set peer_name */
1844 if (con->peer_name.type == 0)
1845 con->peer_name = msg->hdr.src;
1846
1847 con->in_seq++;
1848 mutex_unlock(&con->mutex);
1849
1850 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
1851 msg, le64_to_cpu(msg->hdr.seq),
1852 ENTITY_NAME(msg->hdr.src),
1853 le16_to_cpu(msg->hdr.type),
1854 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
1855 le32_to_cpu(msg->hdr.front_len),
1856 le32_to_cpu(msg->hdr.data_len),
1857 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
1858 con->ops->dispatch(con, msg);
1859
1860 mutex_lock(&con->mutex);
1861 prepare_read_tag(con);
1862}
1863
1864
1865/*
1866 * Write something to the socket. Called in a worker thread when the
1867 * socket appears to be writeable and we have something ready to send.
1868 */
1869static int try_write(struct ceph_connection *con)
1870{
1871 int ret = 1;
1872
1873 dout("try_write start %p state %lu nref %d\n", con, con->state,
1874 atomic_read(&con->nref));
1875
1876more:
1877 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
1878
1879 /* open the socket first? */
1880 if (con->sock == NULL) {
1881 ceph_con_out_kvec_reset(con);
1882 prepare_write_banner(con);
1883 ret = prepare_write_connect(con);
1884 if (ret < 0)
1885 goto out;
1886 prepare_read_banner(con);
1887 set_bit(CONNECTING, &con->state);
1888 clear_bit(NEGOTIATING, &con->state);
1889
1890 BUG_ON(con->in_msg);
1891 con->in_tag = CEPH_MSGR_TAG_READY;
1892 dout("try_write initiating connect on %p new state %lu\n",
1893 con, con->state);
1894 ret = ceph_tcp_connect(con);
1895 if (ret < 0) {
1896 con->error_msg = "connect error";
1897 goto out;
1898 }
1899 }
1900
1901more_kvec:
1902 /* kvec data queued? */
1903 if (con->out_skip) {
1904 ret = write_partial_skip(con);
1905 if (ret <= 0)
1906 goto out;
1907 }
1908 if (con->out_kvec_left) {
1909 ret = write_partial_kvec(con);
1910 if (ret <= 0)
1911 goto out;
1912 }
1913
1914 /* msg pages? */
1915 if (con->out_msg) {
1916 if (con->out_msg_done) {
1917 ceph_msg_put(con->out_msg);
1918 con->out_msg = NULL; /* we're done with this one */
1919 goto do_next;
1920 }
1921
1922 ret = write_partial_msg_pages(con);
1923 if (ret == 1)
1924 goto more_kvec; /* we need to send the footer, too! */
1925 if (ret == 0)
1926 goto out;
1927 if (ret < 0) {
1928 dout("try_write write_partial_msg_pages err %d\n",
1929 ret);
1930 goto out;
1931 }
1932 }
1933
1934do_next:
1935 if (!test_bit(CONNECTING, &con->state)) {
1936 /* is anything else pending? */
1937 if (!list_empty(&con->out_queue)) {
1938 prepare_write_message(con);
1939 goto more;
1940 }
1941 if (con->in_seq > con->in_seq_acked) {
1942 prepare_write_ack(con);
1943 goto more;
1944 }
1945 if (test_and_clear_bit(KEEPALIVE_PENDING, &con->state)) {
1946 prepare_write_keepalive(con);
1947 goto more;
1948 }
1949 }
1950
1951 /* Nothing to do! */
1952 clear_bit(WRITE_PENDING, &con->state);
1953 dout("try_write nothing else to write.\n");
1954 ret = 0;
1955out:
1956 dout("try_write done on %p ret %d\n", con, ret);
1957 return ret;
1958}
1959
1960
1961
1962/*
1963 * Read what we can from the socket.
1964 */
1965static int try_read(struct ceph_connection *con)
1966{
1967 int ret = -1;
1968
1969 if (!con->sock)
1970 return 0;
1971
1972 if (test_bit(STANDBY, &con->state))
1973 return 0;
1974
1975 dout("try_read start on %p\n", con);
1976
1977more:
1978 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
1979 con->in_base_pos);
1980
1981 /*
1982 * process_connect and process_message drop and re-take
1983 * con->mutex. make sure we handle a racing close or reopen.
1984 */
1985 if (test_bit(CLOSED, &con->state) ||
1986 test_bit(OPENING, &con->state)) {
1987 ret = -EAGAIN;
1988 goto out;
1989 }
1990
1991 if (test_bit(CONNECTING, &con->state)) {
1992 if (!test_bit(NEGOTIATING, &con->state)) {
1993 dout("try_read connecting\n");
1994 ret = read_partial_banner(con);
1995 if (ret <= 0)
1996 goto out;
1997 ret = process_banner(con);
1998 if (ret < 0)
1999 goto out;
2000 }
2001 ret = read_partial_connect(con);
2002 if (ret <= 0)
2003 goto out;
2004 ret = process_connect(con);
2005 if (ret < 0)
2006 goto out;
2007 goto more;
2008 }
2009
2010 if (con->in_base_pos < 0) {
2011 /*
2012 * skipping + discarding content.
2013 *
2014 * FIXME: there must be a better way to do this!
2015 */
2016 static char buf[SKIP_BUF_SIZE];
2017 int skip = min((int) sizeof (buf), -con->in_base_pos);
2018
2019 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2020 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2021 if (ret <= 0)
2022 goto out;
2023 con->in_base_pos += ret;
2024 if (con->in_base_pos)
2025 goto more;
2026 }
2027 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2028 /*
2029 * what's next?
2030 */
2031 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2032 if (ret <= 0)
2033 goto out;
2034 dout("try_read got tag %d\n", (int)con->in_tag);
2035 switch (con->in_tag) {
2036 case CEPH_MSGR_TAG_MSG:
2037 prepare_read_message(con);
2038 break;
2039 case CEPH_MSGR_TAG_ACK:
2040 prepare_read_ack(con);
2041 break;
2042 case CEPH_MSGR_TAG_CLOSE:
2043 set_bit(CLOSED, &con->state); /* fixme */
2044 goto out;
2045 default:
2046 goto bad_tag;
2047 }
2048 }
2049 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2050 ret = read_partial_message(con);
2051 if (ret <= 0) {
2052 switch (ret) {
2053 case -EBADMSG:
2054 con->error_msg = "bad crc";
2055 ret = -EIO;
2056 break;
2057 case -EIO:
2058 con->error_msg = "io error";
2059 break;
2060 }
2061 goto out;
2062 }
2063 if (con->in_tag == CEPH_MSGR_TAG_READY)
2064 goto more;
2065 process_message(con);
2066 goto more;
2067 }
2068 if (con->in_tag == CEPH_MSGR_TAG_ACK) {
2069 ret = read_partial_ack(con);
2070 if (ret <= 0)
2071 goto out;
2072 process_ack(con);
2073 goto more;
2074 }
2075
2076out:
2077 dout("try_read done on %p ret %d\n", con, ret);
2078 return ret;
2079
2080bad_tag:
2081 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2082 con->error_msg = "protocol error, garbage tag";
2083 ret = -1;
2084 goto out;
2085}
2086
2087
2088/*
2089 * Atomically queue work on a connection. Bump @con reference to
2090 * avoid races with connection teardown.
2091 */
2092static void queue_con(struct ceph_connection *con)
2093{
2094 if (test_bit(DEAD, &con->state)) {
2095 dout("queue_con %p ignoring: DEAD\n",
2096 con);
2097 return;
2098 }
2099
2100 if (!con->ops->get(con)) {
2101 dout("queue_con %p ref count 0\n", con);
2102 return;
2103 }
2104
2105 if (!queue_delayed_work(ceph_msgr_wq, &con->work, 0)) {
2106 dout("queue_con %p - already queued\n", con);
2107 con->ops->put(con);
2108 } else {
2109 dout("queue_con %p\n", con);
2110 }
2111}
2112
2113/*
2114 * Do some work on a connection. Drop a connection ref when we're done.
2115 */
2116static void con_work(struct work_struct *work)
2117{
2118 struct ceph_connection *con = container_of(work, struct ceph_connection,
2119 work.work);
2120 int ret;
2121
2122 mutex_lock(&con->mutex);
2123restart:
2124 if (test_and_clear_bit(BACKOFF, &con->state)) {
2125 dout("con_work %p backing off\n", con);
2126 if (queue_delayed_work(ceph_msgr_wq, &con->work,
2127 round_jiffies_relative(con->delay))) {
2128 dout("con_work %p backoff %lu\n", con, con->delay);
2129 mutex_unlock(&con->mutex);
2130 return;
2131 } else {
2132 con->ops->put(con);
2133 dout("con_work %p FAILED to back off %lu\n", con,
2134 con->delay);
2135 }
2136 }
2137
2138 if (test_bit(STANDBY, &con->state)) {
2139 dout("con_work %p STANDBY\n", con);
2140 goto done;
2141 }
2142 if (test_bit(CLOSED, &con->state)) { /* e.g. if we are replaced */
2143 dout("con_work CLOSED\n");
2144 con_close_socket(con);
2145 goto done;
2146 }
2147 if (test_and_clear_bit(OPENING, &con->state)) {
2148 /* reopen w/ new peer */
2149 dout("con_work OPENING\n");
2150 con_close_socket(con);
2151 }
2152
2153 if (test_and_clear_bit(SOCK_CLOSED, &con->state))
2154 goto fault;
2155
2156 ret = try_read(con);
2157 if (ret == -EAGAIN)
2158 goto restart;
2159 if (ret < 0)
2160 goto fault;
2161
2162 ret = try_write(con);
2163 if (ret == -EAGAIN)
2164 goto restart;
2165 if (ret < 0)
2166 goto fault;
2167
2168done:
2169 mutex_unlock(&con->mutex);
2170done_unlocked:
2171 con->ops->put(con);
2172 return;
2173
2174fault:
2175 mutex_unlock(&con->mutex);
2176 ceph_fault(con); /* error/fault path */
2177 goto done_unlocked;
2178}
2179
2180
2181/*
2182 * Generic error/fault handler. A retry mechanism is used with
2183 * exponential backoff
2184 */
2185static void ceph_fault(struct ceph_connection *con)
2186{
2187 pr_err("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2188 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2189 dout("fault %p state %lu to peer %s\n",
2190 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2191
2192 if (test_bit(LOSSYTX, &con->state)) {
2193 dout("fault on LOSSYTX channel\n");
2194 goto out;
2195 }
2196
2197 mutex_lock(&con->mutex);
2198 if (test_bit(CLOSED, &con->state))
2199 goto out_unlock;
2200
2201 con_close_socket(con);
2202
2203 if (con->in_msg) {
2204 ceph_msg_put(con->in_msg);
2205 con->in_msg = NULL;
2206 }
2207
2208 /* Requeue anything that hasn't been acked */
2209 list_splice_init(&con->out_sent, &con->out_queue);
2210
2211 /* If there are no messages queued or keepalive pending, place
2212 * the connection in a STANDBY state */
2213 if (list_empty(&con->out_queue) &&
2214 !test_bit(KEEPALIVE_PENDING, &con->state)) {
2215 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2216 clear_bit(WRITE_PENDING, &con->state);
2217 set_bit(STANDBY, &con->state);
2218 } else {
2219 /* retry after a delay. */
2220 if (con->delay == 0)
2221 con->delay = BASE_DELAY_INTERVAL;
2222 else if (con->delay < MAX_DELAY_INTERVAL)
2223 con->delay *= 2;
2224 con->ops->get(con);
2225 if (queue_delayed_work(ceph_msgr_wq, &con->work,
2226 round_jiffies_relative(con->delay))) {
2227 dout("fault queued %p delay %lu\n", con, con->delay);
2228 } else {
2229 con->ops->put(con);
2230 dout("fault failed to queue %p delay %lu, backoff\n",
2231 con, con->delay);
2232 /*
2233 * In many cases we see a socket state change
2234 * while con_work is running and end up
2235 * queuing (non-delayed) work, such that we
2236 * can't backoff with a delay. Set a flag so
2237 * that when con_work restarts we schedule the
2238 * delay then.
2239 */
2240 set_bit(BACKOFF, &con->state);
2241 }
2242 }
2243
2244out_unlock:
2245 mutex_unlock(&con->mutex);
2246out:
2247 /*
2248 * in case we faulted due to authentication, invalidate our
2249 * current tickets so that we can get new ones.
2250 */
2251 if (con->auth_retry && con->ops->invalidate_authorizer) {
2252 dout("calling invalidate_authorizer()\n");
2253 con->ops->invalidate_authorizer(con);
2254 }
2255
2256 if (con->ops->fault)
2257 con->ops->fault(con);
2258}
2259
2260
2261
2262/*
2263 * create a new messenger instance
2264 */
2265struct ceph_messenger *ceph_messenger_create(struct ceph_entity_addr *myaddr,
2266 u32 supported_features,
2267 u32 required_features)
2268{
2269 struct ceph_messenger *msgr;
2270
2271 msgr = kzalloc(sizeof(*msgr), GFP_KERNEL);
2272 if (msgr == NULL)
2273 return ERR_PTR(-ENOMEM);
2274
2275 msgr->supported_features = supported_features;
2276 msgr->required_features = required_features;
2277
2278 spin_lock_init(&msgr->global_seq_lock);
2279
2280 if (myaddr)
2281 msgr->inst.addr = *myaddr;
2282
2283 /* select a random nonce */
2284 msgr->inst.addr.type = 0;
2285 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2286 encode_my_addr(msgr);
2287
2288 dout("messenger_create %p\n", msgr);
2289 return msgr;
2290}
2291EXPORT_SYMBOL(ceph_messenger_create);
2292
2293void ceph_messenger_destroy(struct ceph_messenger *msgr)
2294{
2295 dout("destroy %p\n", msgr);
2296 kfree(msgr);
2297 dout("destroyed messenger %p\n", msgr);
2298}
2299EXPORT_SYMBOL(ceph_messenger_destroy);
2300
2301static void clear_standby(struct ceph_connection *con)
2302{
2303 /* come back from STANDBY? */
2304 if (test_and_clear_bit(STANDBY, &con->state)) {
2305 mutex_lock(&con->mutex);
2306 dout("clear_standby %p and ++connect_seq\n", con);
2307 con->connect_seq++;
2308 WARN_ON(test_bit(WRITE_PENDING, &con->state));
2309 WARN_ON(test_bit(KEEPALIVE_PENDING, &con->state));
2310 mutex_unlock(&con->mutex);
2311 }
2312}
2313
2314/*
2315 * Queue up an outgoing message on the given connection.
2316 */
2317void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2318{
2319 if (test_bit(CLOSED, &con->state)) {
2320 dout("con_send %p closed, dropping %p\n", con, msg);
2321 ceph_msg_put(msg);
2322 return;
2323 }
2324
2325 /* set src+dst */
2326 msg->hdr.src = con->msgr->inst.name;
2327
2328 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2329
2330 msg->needs_out_seq = true;
2331
2332 /* queue */
2333 mutex_lock(&con->mutex);
2334 BUG_ON(!list_empty(&msg->list_head));
2335 list_add_tail(&msg->list_head, &con->out_queue);
2336 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2337 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2338 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2339 le32_to_cpu(msg->hdr.front_len),
2340 le32_to_cpu(msg->hdr.middle_len),
2341 le32_to_cpu(msg->hdr.data_len));
2342 mutex_unlock(&con->mutex);
2343
2344 /* if there wasn't anything waiting to send before, queue
2345 * new work */
2346 clear_standby(con);
2347 if (test_and_set_bit(WRITE_PENDING, &con->state) == 0)
2348 queue_con(con);
2349}
2350EXPORT_SYMBOL(ceph_con_send);
2351
2352/*
2353 * Revoke a message that was previously queued for send
2354 */
2355void ceph_con_revoke(struct ceph_connection *con, struct ceph_msg *msg)
2356{
2357 mutex_lock(&con->mutex);
2358 if (!list_empty(&msg->list_head)) {
2359 dout("con_revoke %p msg %p - was on queue\n", con, msg);
2360 list_del_init(&msg->list_head);
2361 ceph_msg_put(msg);
2362 msg->hdr.seq = 0;
2363 }
2364 if (con->out_msg == msg) {
2365 dout("con_revoke %p msg %p - was sending\n", con, msg);
2366 con->out_msg = NULL;
2367 if (con->out_kvec_is_msg) {
2368 con->out_skip = con->out_kvec_bytes;
2369 con->out_kvec_is_msg = false;
2370 }
2371 ceph_msg_put(msg);
2372 msg->hdr.seq = 0;
2373 }
2374 mutex_unlock(&con->mutex);
2375}
2376
2377/*
2378 * Revoke a message that we may be reading data into
2379 */
2380void ceph_con_revoke_message(struct ceph_connection *con, struct ceph_msg *msg)
2381{
2382 mutex_lock(&con->mutex);
2383 if (con->in_msg && con->in_msg == msg) {
2384 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
2385 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
2386 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
2387
2388 /* skip rest of message */
2389 dout("con_revoke_pages %p msg %p revoked\n", con, msg);
2390 con->in_base_pos = con->in_base_pos -
2391 sizeof(struct ceph_msg_header) -
2392 front_len -
2393 middle_len -
2394 data_len -
2395 sizeof(struct ceph_msg_footer);
2396 ceph_msg_put(con->in_msg);
2397 con->in_msg = NULL;
2398 con->in_tag = CEPH_MSGR_TAG_READY;
2399 con->in_seq++;
2400 } else {
2401 dout("con_revoke_pages %p msg %p pages %p no-op\n",
2402 con, con->in_msg, msg);
2403 }
2404 mutex_unlock(&con->mutex);
2405}
2406
2407/*
2408 * Queue a keepalive byte to ensure the tcp connection is alive.
2409 */
2410void ceph_con_keepalive(struct ceph_connection *con)
2411{
2412 dout("con_keepalive %p\n", con);
2413 clear_standby(con);
2414 if (test_and_set_bit(KEEPALIVE_PENDING, &con->state) == 0 &&
2415 test_and_set_bit(WRITE_PENDING, &con->state) == 0)
2416 queue_con(con);
2417}
2418EXPORT_SYMBOL(ceph_con_keepalive);
2419
2420
2421/*
2422 * construct a new message with given type, size
2423 * the new msg has a ref count of 1.
2424 */
2425struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
2426 bool can_fail)
2427{
2428 struct ceph_msg *m;
2429
2430 m = kmalloc(sizeof(*m), flags);
2431 if (m == NULL)
2432 goto out;
2433 kref_init(&m->kref);
2434 INIT_LIST_HEAD(&m->list_head);
2435
2436 m->hdr.tid = 0;
2437 m->hdr.type = cpu_to_le16(type);
2438 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
2439 m->hdr.version = 0;
2440 m->hdr.front_len = cpu_to_le32(front_len);
2441 m->hdr.middle_len = 0;
2442 m->hdr.data_len = 0;
2443 m->hdr.data_off = 0;
2444 m->hdr.reserved = 0;
2445 m->footer.front_crc = 0;
2446 m->footer.middle_crc = 0;
2447 m->footer.data_crc = 0;
2448 m->footer.flags = 0;
2449 m->front_max = front_len;
2450 m->front_is_vmalloc = false;
2451 m->more_to_follow = false;
2452 m->ack_stamp = 0;
2453 m->pool = NULL;
2454
2455 /* middle */
2456 m->middle = NULL;
2457
2458 /* data */
2459 m->nr_pages = 0;
2460 m->page_alignment = 0;
2461 m->pages = NULL;
2462 m->pagelist = NULL;
2463 m->bio = NULL;
2464 m->bio_iter = NULL;
2465 m->bio_seg = 0;
2466 m->trail = NULL;
2467
2468 /* front */
2469 if (front_len) {
2470 if (front_len > PAGE_CACHE_SIZE) {
2471 m->front.iov_base = __vmalloc(front_len, flags,
2472 PAGE_KERNEL);
2473 m->front_is_vmalloc = true;
2474 } else {
2475 m->front.iov_base = kmalloc(front_len, flags);
2476 }
2477 if (m->front.iov_base == NULL) {
2478 dout("ceph_msg_new can't allocate %d bytes\n",
2479 front_len);
2480 goto out2;
2481 }
2482 } else {
2483 m->front.iov_base = NULL;
2484 }
2485 m->front.iov_len = front_len;
2486
2487 dout("ceph_msg_new %p front %d\n", m, front_len);
2488 return m;
2489
2490out2:
2491 ceph_msg_put(m);
2492out:
2493 if (!can_fail) {
2494 pr_err("msg_new can't create type %d front %d\n", type,
2495 front_len);
2496 WARN_ON(1);
2497 } else {
2498 dout("msg_new can't create type %d front %d\n", type,
2499 front_len);
2500 }
2501 return NULL;
2502}
2503EXPORT_SYMBOL(ceph_msg_new);
2504
2505/*
2506 * Allocate "middle" portion of a message, if it is needed and wasn't
2507 * allocated by alloc_msg. This allows us to read a small fixed-size
2508 * per-type header in the front and then gracefully fail (i.e.,
2509 * propagate the error to the caller based on info in the front) when
2510 * the middle is too large.
2511 */
2512static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
2513{
2514 int type = le16_to_cpu(msg->hdr.type);
2515 int middle_len = le32_to_cpu(msg->hdr.middle_len);
2516
2517 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
2518 ceph_msg_type_name(type), middle_len);
2519 BUG_ON(!middle_len);
2520 BUG_ON(msg->middle);
2521
2522 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
2523 if (!msg->middle)
2524 return -ENOMEM;
2525 return 0;
2526}
2527
2528/*
2529 * Generic message allocator, for incoming messages.
2530 */
2531static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con,
2532 struct ceph_msg_header *hdr,
2533 int *skip)
2534{
2535 int type = le16_to_cpu(hdr->type);
2536 int front_len = le32_to_cpu(hdr->front_len);
2537 int middle_len = le32_to_cpu(hdr->middle_len);
2538 struct ceph_msg *msg = NULL;
2539 int ret;
2540
2541 if (con->ops->alloc_msg) {
2542 mutex_unlock(&con->mutex);
2543 msg = con->ops->alloc_msg(con, hdr, skip);
2544 mutex_lock(&con->mutex);
2545 if (!msg || *skip)
2546 return NULL;
2547 }
2548 if (!msg) {
2549 *skip = 0;
2550 msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
2551 if (!msg) {
2552 pr_err("unable to allocate msg type %d len %d\n",
2553 type, front_len);
2554 return NULL;
2555 }
2556 msg->page_alignment = le16_to_cpu(hdr->data_off);
2557 }
2558 memcpy(&msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2559
2560 if (middle_len && !msg->middle) {
2561 ret = ceph_alloc_middle(con, msg);
2562 if (ret < 0) {
2563 ceph_msg_put(msg);
2564 return NULL;
2565 }
2566 }
2567
2568 return msg;
2569}
2570
2571
2572/*
2573 * Free a generically kmalloc'd message.
2574 */
2575void ceph_msg_kfree(struct ceph_msg *m)
2576{
2577 dout("msg_kfree %p\n", m);
2578 if (m->front_is_vmalloc)
2579 vfree(m->front.iov_base);
2580 else
2581 kfree(m->front.iov_base);
2582 kfree(m);
2583}
2584
2585/*
2586 * Drop a msg ref. Destroy as needed.
2587 */
2588void ceph_msg_last_put(struct kref *kref)
2589{
2590 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
2591
2592 dout("ceph_msg_put last one on %p\n", m);
2593 WARN_ON(!list_empty(&m->list_head));
2594
2595 /* drop middle, data, if any */
2596 if (m->middle) {
2597 ceph_buffer_put(m->middle);
2598 m->middle = NULL;
2599 }
2600 m->nr_pages = 0;
2601 m->pages = NULL;
2602
2603 if (m->pagelist) {
2604 ceph_pagelist_release(m->pagelist);
2605 kfree(m->pagelist);
2606 m->pagelist = NULL;
2607 }
2608
2609 m->trail = NULL;
2610
2611 if (m->pool)
2612 ceph_msgpool_put(m->pool, m);
2613 else
2614 ceph_msg_kfree(m);
2615}
2616EXPORT_SYMBOL(ceph_msg_last_put);
2617
2618void ceph_msg_dump(struct ceph_msg *msg)
2619{
2620 pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg,
2621 msg->front_max, msg->nr_pages);
2622 print_hex_dump(KERN_DEBUG, "header: ",
2623 DUMP_PREFIX_OFFSET, 16, 1,
2624 &msg->hdr, sizeof(msg->hdr), true);
2625 print_hex_dump(KERN_DEBUG, " front: ",
2626 DUMP_PREFIX_OFFSET, 16, 1,
2627 msg->front.iov_base, msg->front.iov_len, true);
2628 if (msg->middle)
2629 print_hex_dump(KERN_DEBUG, "middle: ",
2630 DUMP_PREFIX_OFFSET, 16, 1,
2631 msg->middle->vec.iov_base,
2632 msg->middle->vec.iov_len, true);
2633 print_hex_dump(KERN_DEBUG, "footer: ",
2634 DUMP_PREFIX_OFFSET, 16, 1,
2635 &msg->footer, sizeof(msg->footer), true);
2636}
2637EXPORT_SYMBOL(ceph_msg_dump);