Loading...
1#include <linux/ceph/ceph_debug.h>
2
3#include <linux/crc32c.h>
4#include <linux/ctype.h>
5#include <linux/highmem.h>
6#include <linux/inet.h>
7#include <linux/kthread.h>
8#include <linux/net.h>
9#include <linux/nsproxy.h>
10#include <linux/slab.h>
11#include <linux/socket.h>
12#include <linux/string.h>
13#ifdef CONFIG_BLOCK
14#include <linux/bio.h>
15#endif /* CONFIG_BLOCK */
16#include <linux/dns_resolver.h>
17#include <net/tcp.h>
18
19#include <linux/ceph/ceph_features.h>
20#include <linux/ceph/libceph.h>
21#include <linux/ceph/messenger.h>
22#include <linux/ceph/decode.h>
23#include <linux/ceph/pagelist.h>
24#include <linux/export.h>
25
26/*
27 * Ceph uses the messenger to exchange ceph_msg messages with other
28 * hosts in the system. The messenger provides ordered and reliable
29 * delivery. We tolerate TCP disconnects by reconnecting (with
30 * exponential backoff) in the case of a fault (disconnection, bad
31 * crc, protocol error). Acks allow sent messages to be discarded by
32 * the sender.
33 */
34
35/*
36 * We track the state of the socket on a given connection using
37 * values defined below. The transition to a new socket state is
38 * handled by a function which verifies we aren't coming from an
39 * unexpected state.
40 *
41 * --------
42 * | NEW* | transient initial state
43 * --------
44 * | con_sock_state_init()
45 * v
46 * ----------
47 * | CLOSED | initialized, but no socket (and no
48 * ---------- TCP connection)
49 * ^ \
50 * | \ con_sock_state_connecting()
51 * | ----------------------
52 * | \
53 * + con_sock_state_closed() \
54 * |+--------------------------- \
55 * | \ \ \
56 * | ----------- \ \
57 * | | CLOSING | socket event; \ \
58 * | ----------- await close \ \
59 * | ^ \ |
60 * | | \ |
61 * | + con_sock_state_closing() \ |
62 * | / \ | |
63 * | / --------------- | |
64 * | / \ v v
65 * | / --------------
66 * | / -----------------| CONNECTING | socket created, TCP
67 * | | / -------------- connect initiated
68 * | | | con_sock_state_connected()
69 * | | v
70 * -------------
71 * | CONNECTED | TCP connection established
72 * -------------
73 *
74 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
75 */
76
77#define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
78#define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
79#define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
80#define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
81#define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
82
83/*
84 * connection states
85 */
86#define CON_STATE_CLOSED 1 /* -> PREOPEN */
87#define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
88#define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
89#define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
90#define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
91#define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
92
93/*
94 * ceph_connection flag bits
95 */
96#define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
97 * messages on errors */
98#define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
99#define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
100#define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
101#define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
102
103static bool con_flag_valid(unsigned long con_flag)
104{
105 switch (con_flag) {
106 case CON_FLAG_LOSSYTX:
107 case CON_FLAG_KEEPALIVE_PENDING:
108 case CON_FLAG_WRITE_PENDING:
109 case CON_FLAG_SOCK_CLOSED:
110 case CON_FLAG_BACKOFF:
111 return true;
112 default:
113 return false;
114 }
115}
116
117static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
118{
119 BUG_ON(!con_flag_valid(con_flag));
120
121 clear_bit(con_flag, &con->flags);
122}
123
124static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
125{
126 BUG_ON(!con_flag_valid(con_flag));
127
128 set_bit(con_flag, &con->flags);
129}
130
131static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
132{
133 BUG_ON(!con_flag_valid(con_flag));
134
135 return test_bit(con_flag, &con->flags);
136}
137
138static bool con_flag_test_and_clear(struct ceph_connection *con,
139 unsigned long con_flag)
140{
141 BUG_ON(!con_flag_valid(con_flag));
142
143 return test_and_clear_bit(con_flag, &con->flags);
144}
145
146static bool con_flag_test_and_set(struct ceph_connection *con,
147 unsigned long con_flag)
148{
149 BUG_ON(!con_flag_valid(con_flag));
150
151 return test_and_set_bit(con_flag, &con->flags);
152}
153
154/* Slab caches for frequently-allocated structures */
155
156static struct kmem_cache *ceph_msg_cache;
157static struct kmem_cache *ceph_msg_data_cache;
158
159/* static tag bytes (protocol control messages) */
160static char tag_msg = CEPH_MSGR_TAG_MSG;
161static char tag_ack = CEPH_MSGR_TAG_ACK;
162static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
163static char tag_keepalive2 = CEPH_MSGR_TAG_KEEPALIVE2;
164
165#ifdef CONFIG_LOCKDEP
166static struct lock_class_key socket_class;
167#endif
168
169/*
170 * When skipping (ignoring) a block of input we read it into a "skip
171 * buffer," which is this many bytes in size.
172 */
173#define SKIP_BUF_SIZE 1024
174
175static void queue_con(struct ceph_connection *con);
176static void cancel_con(struct ceph_connection *con);
177static void ceph_con_workfn(struct work_struct *);
178static void con_fault(struct ceph_connection *con);
179
180/*
181 * Nicely render a sockaddr as a string. An array of formatted
182 * strings is used, to approximate reentrancy.
183 */
184#define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
185#define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
186#define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
187#define MAX_ADDR_STR_LEN 64 /* 54 is enough */
188
189static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
190static atomic_t addr_str_seq = ATOMIC_INIT(0);
191
192static struct page *zero_page; /* used in certain error cases */
193
194const char *ceph_pr_addr(const struct sockaddr_storage *ss)
195{
196 int i;
197 char *s;
198 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
199 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
200
201 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
202 s = addr_str[i];
203
204 switch (ss->ss_family) {
205 case AF_INET:
206 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
207 ntohs(in4->sin_port));
208 break;
209
210 case AF_INET6:
211 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
212 ntohs(in6->sin6_port));
213 break;
214
215 default:
216 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
217 ss->ss_family);
218 }
219
220 return s;
221}
222EXPORT_SYMBOL(ceph_pr_addr);
223
224static void encode_my_addr(struct ceph_messenger *msgr)
225{
226 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
227 ceph_encode_addr(&msgr->my_enc_addr);
228}
229
230/*
231 * work queue for all reading and writing to/from the socket.
232 */
233static struct workqueue_struct *ceph_msgr_wq;
234
235static int ceph_msgr_slab_init(void)
236{
237 BUG_ON(ceph_msg_cache);
238 ceph_msg_cache = KMEM_CACHE(ceph_msg, 0);
239 if (!ceph_msg_cache)
240 return -ENOMEM;
241
242 BUG_ON(ceph_msg_data_cache);
243 ceph_msg_data_cache = KMEM_CACHE(ceph_msg_data, 0);
244 if (ceph_msg_data_cache)
245 return 0;
246
247 kmem_cache_destroy(ceph_msg_cache);
248 ceph_msg_cache = NULL;
249
250 return -ENOMEM;
251}
252
253static void ceph_msgr_slab_exit(void)
254{
255 BUG_ON(!ceph_msg_data_cache);
256 kmem_cache_destroy(ceph_msg_data_cache);
257 ceph_msg_data_cache = NULL;
258
259 BUG_ON(!ceph_msg_cache);
260 kmem_cache_destroy(ceph_msg_cache);
261 ceph_msg_cache = NULL;
262}
263
264static void _ceph_msgr_exit(void)
265{
266 if (ceph_msgr_wq) {
267 destroy_workqueue(ceph_msgr_wq);
268 ceph_msgr_wq = NULL;
269 }
270
271 BUG_ON(zero_page == NULL);
272 put_page(zero_page);
273 zero_page = NULL;
274
275 ceph_msgr_slab_exit();
276}
277
278int ceph_msgr_init(void)
279{
280 if (ceph_msgr_slab_init())
281 return -ENOMEM;
282
283 BUG_ON(zero_page != NULL);
284 zero_page = ZERO_PAGE(0);
285 get_page(zero_page);
286
287 /*
288 * The number of active work items is limited by the number of
289 * connections, so leave @max_active at default.
290 */
291 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
292 if (ceph_msgr_wq)
293 return 0;
294
295 pr_err("msgr_init failed to create workqueue\n");
296 _ceph_msgr_exit();
297
298 return -ENOMEM;
299}
300EXPORT_SYMBOL(ceph_msgr_init);
301
302void ceph_msgr_exit(void)
303{
304 BUG_ON(ceph_msgr_wq == NULL);
305
306 _ceph_msgr_exit();
307}
308EXPORT_SYMBOL(ceph_msgr_exit);
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 case TCP_CLOSE_WAIT:
432 dout("%s TCP_CLOSE_WAIT\n", __func__);
433 con_sock_state_closing(con);
434 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
435 queue_con(con);
436 break;
437 case TCP_ESTABLISHED:
438 dout("%s TCP_ESTABLISHED\n", __func__);
439 con_sock_state_connected(con);
440 queue_con(con);
441 break;
442 default: /* Everything else is uninteresting */
443 break;
444 }
445}
446
447/*
448 * set up socket callbacks
449 */
450static void set_sock_callbacks(struct socket *sock,
451 struct ceph_connection *con)
452{
453 struct sock *sk = sock->sk;
454 sk->sk_user_data = con;
455 sk->sk_data_ready = ceph_sock_data_ready;
456 sk->sk_write_space = ceph_sock_write_space;
457 sk->sk_state_change = ceph_sock_state_change;
458}
459
460
461/*
462 * socket helpers
463 */
464
465/*
466 * initiate connection to a remote socket.
467 */
468static int ceph_tcp_connect(struct ceph_connection *con)
469{
470 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
471 struct socket *sock;
472 int ret;
473
474 BUG_ON(con->sock);
475 ret = sock_create_kern(read_pnet(&con->msgr->net), paddr->ss_family,
476 SOCK_STREAM, IPPROTO_TCP, &sock);
477 if (ret)
478 return ret;
479 sock->sk->sk_allocation = GFP_NOFS;
480
481#ifdef CONFIG_LOCKDEP
482 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
483#endif
484
485 set_sock_callbacks(sock, con);
486
487 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
488
489 con_sock_state_connecting(con);
490 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
491 O_NONBLOCK);
492 if (ret == -EINPROGRESS) {
493 dout("connect %s EINPROGRESS sk_state = %u\n",
494 ceph_pr_addr(&con->peer_addr.in_addr),
495 sock->sk->sk_state);
496 } else if (ret < 0) {
497 pr_err("connect %s error %d\n",
498 ceph_pr_addr(&con->peer_addr.in_addr), ret);
499 sock_release(sock);
500 return ret;
501 }
502
503 if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY)) {
504 int optval = 1;
505
506 ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
507 (char *)&optval, sizeof(optval));
508 if (ret)
509 pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
510 ret);
511 }
512
513 con->sock = sock;
514 return 0;
515}
516
517static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
518{
519 struct kvec iov = {buf, len};
520 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
521 int r;
522
523 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
524 if (r == -EAGAIN)
525 r = 0;
526 return r;
527}
528
529static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
530 int page_offset, size_t length)
531{
532 void *kaddr;
533 int ret;
534
535 BUG_ON(page_offset + length > PAGE_SIZE);
536
537 kaddr = kmap(page);
538 BUG_ON(!kaddr);
539 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
540 kunmap(page);
541
542 return ret;
543}
544
545/*
546 * write something. @more is true if caller will be sending more data
547 * shortly.
548 */
549static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
550 size_t kvlen, size_t len, int more)
551{
552 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
553 int r;
554
555 if (more)
556 msg.msg_flags |= MSG_MORE;
557 else
558 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
559
560 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
561 if (r == -EAGAIN)
562 r = 0;
563 return r;
564}
565
566static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
567 int offset, size_t size, bool more)
568{
569 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
570 int ret;
571
572 ret = kernel_sendpage(sock, page, offset, size, flags);
573 if (ret == -EAGAIN)
574 ret = 0;
575
576 return ret;
577}
578
579static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
580 int offset, size_t size, bool more)
581{
582 int ret;
583 struct kvec iov;
584
585 /* sendpage cannot properly handle pages with page_count == 0,
586 * we need to fallback to sendmsg if that's the case */
587 if (page_count(page) >= 1)
588 return __ceph_tcp_sendpage(sock, page, offset, size, more);
589
590 iov.iov_base = kmap(page) + offset;
591 iov.iov_len = size;
592 ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more);
593 kunmap(page);
594
595 return ret;
596}
597
598/*
599 * Shutdown/close the socket for the given connection.
600 */
601static int con_close_socket(struct ceph_connection *con)
602{
603 int rc = 0;
604
605 dout("con_close_socket on %p sock %p\n", con, con->sock);
606 if (con->sock) {
607 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
608 sock_release(con->sock);
609 con->sock = NULL;
610 }
611
612 /*
613 * Forcibly clear the SOCK_CLOSED flag. It gets set
614 * independent of the connection mutex, and we could have
615 * received a socket close event before we had the chance to
616 * shut the socket down.
617 */
618 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
619
620 con_sock_state_closed(con);
621 return rc;
622}
623
624/*
625 * Reset a connection. Discard all incoming and outgoing messages
626 * and clear *_seq state.
627 */
628static void ceph_msg_remove(struct ceph_msg *msg)
629{
630 list_del_init(&msg->list_head);
631
632 ceph_msg_put(msg);
633}
634static void ceph_msg_remove_list(struct list_head *head)
635{
636 while (!list_empty(head)) {
637 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
638 list_head);
639 ceph_msg_remove(msg);
640 }
641}
642
643static void reset_connection(struct ceph_connection *con)
644{
645 /* reset connection, out_queue, msg_ and connect_seq */
646 /* discard existing out_queue and msg_seq */
647 dout("reset_connection %p\n", con);
648 ceph_msg_remove_list(&con->out_queue);
649 ceph_msg_remove_list(&con->out_sent);
650
651 if (con->in_msg) {
652 BUG_ON(con->in_msg->con != con);
653 ceph_msg_put(con->in_msg);
654 con->in_msg = NULL;
655 }
656
657 con->connect_seq = 0;
658 con->out_seq = 0;
659 if (con->out_msg) {
660 BUG_ON(con->out_msg->con != con);
661 ceph_msg_put(con->out_msg);
662 con->out_msg = NULL;
663 }
664 con->in_seq = 0;
665 con->in_seq_acked = 0;
666
667 con->out_skip = 0;
668}
669
670/*
671 * mark a peer down. drop any open connections.
672 */
673void ceph_con_close(struct ceph_connection *con)
674{
675 mutex_lock(&con->mutex);
676 dout("con_close %p peer %s\n", con,
677 ceph_pr_addr(&con->peer_addr.in_addr));
678 con->state = CON_STATE_CLOSED;
679
680 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
681 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
682 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
683 con_flag_clear(con, CON_FLAG_BACKOFF);
684
685 reset_connection(con);
686 con->peer_global_seq = 0;
687 cancel_con(con);
688 con_close_socket(con);
689 mutex_unlock(&con->mutex);
690}
691EXPORT_SYMBOL(ceph_con_close);
692
693/*
694 * Reopen a closed connection, with a new peer address.
695 */
696void ceph_con_open(struct ceph_connection *con,
697 __u8 entity_type, __u64 entity_num,
698 struct ceph_entity_addr *addr)
699{
700 mutex_lock(&con->mutex);
701 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
702
703 WARN_ON(con->state != CON_STATE_CLOSED);
704 con->state = CON_STATE_PREOPEN;
705
706 con->peer_name.type = (__u8) entity_type;
707 con->peer_name.num = cpu_to_le64(entity_num);
708
709 memcpy(&con->peer_addr, addr, sizeof(*addr));
710 con->delay = 0; /* reset backoff memory */
711 mutex_unlock(&con->mutex);
712 queue_con(con);
713}
714EXPORT_SYMBOL(ceph_con_open);
715
716/*
717 * return true if this connection ever successfully opened
718 */
719bool ceph_con_opened(struct ceph_connection *con)
720{
721 return con->connect_seq > 0;
722}
723
724/*
725 * initialize a new connection.
726 */
727void ceph_con_init(struct ceph_connection *con, void *private,
728 const struct ceph_connection_operations *ops,
729 struct ceph_messenger *msgr)
730{
731 dout("con_init %p\n", con);
732 memset(con, 0, sizeof(*con));
733 con->private = private;
734 con->ops = ops;
735 con->msgr = msgr;
736
737 con_sock_state_init(con);
738
739 mutex_init(&con->mutex);
740 INIT_LIST_HEAD(&con->out_queue);
741 INIT_LIST_HEAD(&con->out_sent);
742 INIT_DELAYED_WORK(&con->work, ceph_con_workfn);
743
744 con->state = CON_STATE_CLOSED;
745}
746EXPORT_SYMBOL(ceph_con_init);
747
748
749/*
750 * We maintain a global counter to order connection attempts. Get
751 * a unique seq greater than @gt.
752 */
753static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
754{
755 u32 ret;
756
757 spin_lock(&msgr->global_seq_lock);
758 if (msgr->global_seq < gt)
759 msgr->global_seq = gt;
760 ret = ++msgr->global_seq;
761 spin_unlock(&msgr->global_seq_lock);
762 return ret;
763}
764
765static void con_out_kvec_reset(struct ceph_connection *con)
766{
767 BUG_ON(con->out_skip);
768
769 con->out_kvec_left = 0;
770 con->out_kvec_bytes = 0;
771 con->out_kvec_cur = &con->out_kvec[0];
772}
773
774static void con_out_kvec_add(struct ceph_connection *con,
775 size_t size, void *data)
776{
777 int index = con->out_kvec_left;
778
779 BUG_ON(con->out_skip);
780 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
781
782 con->out_kvec[index].iov_len = size;
783 con->out_kvec[index].iov_base = data;
784 con->out_kvec_left++;
785 con->out_kvec_bytes += size;
786}
787
788/*
789 * Chop off a kvec from the end. Return residual number of bytes for
790 * that kvec, i.e. how many bytes would have been written if the kvec
791 * hadn't been nuked.
792 */
793static int con_out_kvec_skip(struct ceph_connection *con)
794{
795 int off = con->out_kvec_cur - con->out_kvec;
796 int skip = 0;
797
798 if (con->out_kvec_bytes > 0) {
799 skip = con->out_kvec[off + con->out_kvec_left - 1].iov_len;
800 BUG_ON(con->out_kvec_bytes < skip);
801 BUG_ON(!con->out_kvec_left);
802 con->out_kvec_bytes -= skip;
803 con->out_kvec_left--;
804 }
805
806 return skip;
807}
808
809#ifdef CONFIG_BLOCK
810
811/*
812 * For a bio data item, a piece is whatever remains of the next
813 * entry in the current bio iovec, or the first entry in the next
814 * bio in the list.
815 */
816static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
817 size_t length)
818{
819 struct ceph_msg_data *data = cursor->data;
820 struct bio *bio;
821
822 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
823
824 bio = data->bio;
825 BUG_ON(!bio);
826
827 cursor->resid = min(length, data->bio_length);
828 cursor->bio = bio;
829 cursor->bvec_iter = bio->bi_iter;
830 cursor->last_piece =
831 cursor->resid <= bio_iter_len(bio, cursor->bvec_iter);
832}
833
834static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
835 size_t *page_offset,
836 size_t *length)
837{
838 struct ceph_msg_data *data = cursor->data;
839 struct bio *bio;
840 struct bio_vec bio_vec;
841
842 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
843
844 bio = cursor->bio;
845 BUG_ON(!bio);
846
847 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
848
849 *page_offset = (size_t) bio_vec.bv_offset;
850 BUG_ON(*page_offset >= PAGE_SIZE);
851 if (cursor->last_piece) /* pagelist offset is always 0 */
852 *length = cursor->resid;
853 else
854 *length = (size_t) bio_vec.bv_len;
855 BUG_ON(*length > cursor->resid);
856 BUG_ON(*page_offset + *length > PAGE_SIZE);
857
858 return bio_vec.bv_page;
859}
860
861static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
862 size_t bytes)
863{
864 struct bio *bio;
865 struct bio_vec bio_vec;
866
867 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
868
869 bio = cursor->bio;
870 BUG_ON(!bio);
871
872 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
873
874 /* Advance the cursor offset */
875
876 BUG_ON(cursor->resid < bytes);
877 cursor->resid -= bytes;
878
879 bio_advance_iter(bio, &cursor->bvec_iter, bytes);
880
881 if (bytes < bio_vec.bv_len)
882 return false; /* more bytes to process in this segment */
883
884 /* Move on to the next segment, and possibly the next bio */
885
886 if (!cursor->bvec_iter.bi_size) {
887 bio = bio->bi_next;
888 cursor->bio = bio;
889 if (bio)
890 cursor->bvec_iter = bio->bi_iter;
891 else
892 memset(&cursor->bvec_iter, 0,
893 sizeof(cursor->bvec_iter));
894 }
895
896 if (!cursor->last_piece) {
897 BUG_ON(!cursor->resid);
898 BUG_ON(!bio);
899 /* A short read is OK, so use <= rather than == */
900 if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter))
901 cursor->last_piece = true;
902 }
903
904 return true;
905}
906#endif /* CONFIG_BLOCK */
907
908/*
909 * For a page array, a piece comes from the first page in the array
910 * that has not already been fully consumed.
911 */
912static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
913 size_t length)
914{
915 struct ceph_msg_data *data = cursor->data;
916 int page_count;
917
918 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
919
920 BUG_ON(!data->pages);
921 BUG_ON(!data->length);
922
923 cursor->resid = min(length, data->length);
924 page_count = calc_pages_for(data->alignment, (u64)data->length);
925 cursor->page_offset = data->alignment & ~PAGE_MASK;
926 cursor->page_index = 0;
927 BUG_ON(page_count > (int)USHRT_MAX);
928 cursor->page_count = (unsigned short)page_count;
929 BUG_ON(length > SIZE_MAX - cursor->page_offset);
930 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
931}
932
933static struct page *
934ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
935 size_t *page_offset, size_t *length)
936{
937 struct ceph_msg_data *data = cursor->data;
938
939 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
940
941 BUG_ON(cursor->page_index >= cursor->page_count);
942 BUG_ON(cursor->page_offset >= PAGE_SIZE);
943
944 *page_offset = cursor->page_offset;
945 if (cursor->last_piece)
946 *length = cursor->resid;
947 else
948 *length = PAGE_SIZE - *page_offset;
949
950 return data->pages[cursor->page_index];
951}
952
953static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
954 size_t bytes)
955{
956 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
957
958 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
959
960 /* Advance the cursor page offset */
961
962 cursor->resid -= bytes;
963 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
964 if (!bytes || cursor->page_offset)
965 return false; /* more bytes to process in the current page */
966
967 if (!cursor->resid)
968 return false; /* no more data */
969
970 /* Move on to the next page; offset is already at 0 */
971
972 BUG_ON(cursor->page_index >= cursor->page_count);
973 cursor->page_index++;
974 cursor->last_piece = cursor->resid <= PAGE_SIZE;
975
976 return true;
977}
978
979/*
980 * For a pagelist, a piece is whatever remains to be consumed in the
981 * first page in the list, or the front of the next page.
982 */
983static void
984ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
985 size_t length)
986{
987 struct ceph_msg_data *data = cursor->data;
988 struct ceph_pagelist *pagelist;
989 struct page *page;
990
991 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
992
993 pagelist = data->pagelist;
994 BUG_ON(!pagelist);
995
996 if (!length)
997 return; /* pagelist can be assigned but empty */
998
999 BUG_ON(list_empty(&pagelist->head));
1000 page = list_first_entry(&pagelist->head, struct page, lru);
1001
1002 cursor->resid = min(length, pagelist->length);
1003 cursor->page = page;
1004 cursor->offset = 0;
1005 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1006}
1007
1008static struct page *
1009ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
1010 size_t *page_offset, size_t *length)
1011{
1012 struct ceph_msg_data *data = cursor->data;
1013 struct ceph_pagelist *pagelist;
1014
1015 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1016
1017 pagelist = data->pagelist;
1018 BUG_ON(!pagelist);
1019
1020 BUG_ON(!cursor->page);
1021 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1022
1023 /* offset of first page in pagelist is always 0 */
1024 *page_offset = cursor->offset & ~PAGE_MASK;
1025 if (cursor->last_piece)
1026 *length = cursor->resid;
1027 else
1028 *length = PAGE_SIZE - *page_offset;
1029
1030 return cursor->page;
1031}
1032
1033static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
1034 size_t bytes)
1035{
1036 struct ceph_msg_data *data = cursor->data;
1037 struct ceph_pagelist *pagelist;
1038
1039 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1040
1041 pagelist = data->pagelist;
1042 BUG_ON(!pagelist);
1043
1044 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1045 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1046
1047 /* Advance the cursor offset */
1048
1049 cursor->resid -= bytes;
1050 cursor->offset += bytes;
1051 /* offset of first page in pagelist is always 0 */
1052 if (!bytes || cursor->offset & ~PAGE_MASK)
1053 return false; /* more bytes to process in the current page */
1054
1055 if (!cursor->resid)
1056 return false; /* no more data */
1057
1058 /* Move on to the next page */
1059
1060 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1061 cursor->page = list_next_entry(cursor->page, lru);
1062 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1063
1064 return true;
1065}
1066
1067/*
1068 * Message data is handled (sent or received) in pieces, where each
1069 * piece resides on a single page. The network layer might not
1070 * consume an entire piece at once. A data item's cursor keeps
1071 * track of which piece is next to process and how much remains to
1072 * be processed in that piece. It also tracks whether the current
1073 * piece is the last one in the data item.
1074 */
1075static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1076{
1077 size_t length = cursor->total_resid;
1078
1079 switch (cursor->data->type) {
1080 case CEPH_MSG_DATA_PAGELIST:
1081 ceph_msg_data_pagelist_cursor_init(cursor, length);
1082 break;
1083 case CEPH_MSG_DATA_PAGES:
1084 ceph_msg_data_pages_cursor_init(cursor, length);
1085 break;
1086#ifdef CONFIG_BLOCK
1087 case CEPH_MSG_DATA_BIO:
1088 ceph_msg_data_bio_cursor_init(cursor, length);
1089 break;
1090#endif /* CONFIG_BLOCK */
1091 case CEPH_MSG_DATA_NONE:
1092 default:
1093 /* BUG(); */
1094 break;
1095 }
1096 cursor->need_crc = true;
1097}
1098
1099static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1100{
1101 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1102 struct ceph_msg_data *data;
1103
1104 BUG_ON(!length);
1105 BUG_ON(length > msg->data_length);
1106 BUG_ON(list_empty(&msg->data));
1107
1108 cursor->data_head = &msg->data;
1109 cursor->total_resid = length;
1110 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1111 cursor->data = data;
1112
1113 __ceph_msg_data_cursor_init(cursor);
1114}
1115
1116/*
1117 * Return the page containing the next piece to process for a given
1118 * data item, and supply the page offset and length of that piece.
1119 * Indicate whether this is the last piece in this data item.
1120 */
1121static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1122 size_t *page_offset, size_t *length,
1123 bool *last_piece)
1124{
1125 struct page *page;
1126
1127 switch (cursor->data->type) {
1128 case CEPH_MSG_DATA_PAGELIST:
1129 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1130 break;
1131 case CEPH_MSG_DATA_PAGES:
1132 page = ceph_msg_data_pages_next(cursor, page_offset, length);
1133 break;
1134#ifdef CONFIG_BLOCK
1135 case CEPH_MSG_DATA_BIO:
1136 page = ceph_msg_data_bio_next(cursor, page_offset, length);
1137 break;
1138#endif /* CONFIG_BLOCK */
1139 case CEPH_MSG_DATA_NONE:
1140 default:
1141 page = NULL;
1142 break;
1143 }
1144 BUG_ON(!page);
1145 BUG_ON(*page_offset + *length > PAGE_SIZE);
1146 BUG_ON(!*length);
1147 if (last_piece)
1148 *last_piece = cursor->last_piece;
1149
1150 return page;
1151}
1152
1153/*
1154 * Returns true if the result moves the cursor on to the next piece
1155 * of the data item.
1156 */
1157static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1158 size_t bytes)
1159{
1160 bool new_piece;
1161
1162 BUG_ON(bytes > cursor->resid);
1163 switch (cursor->data->type) {
1164 case CEPH_MSG_DATA_PAGELIST:
1165 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1166 break;
1167 case CEPH_MSG_DATA_PAGES:
1168 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1169 break;
1170#ifdef CONFIG_BLOCK
1171 case CEPH_MSG_DATA_BIO:
1172 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1173 break;
1174#endif /* CONFIG_BLOCK */
1175 case CEPH_MSG_DATA_NONE:
1176 default:
1177 BUG();
1178 break;
1179 }
1180 cursor->total_resid -= bytes;
1181
1182 if (!cursor->resid && cursor->total_resid) {
1183 WARN_ON(!cursor->last_piece);
1184 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1185 cursor->data = list_next_entry(cursor->data, links);
1186 __ceph_msg_data_cursor_init(cursor);
1187 new_piece = true;
1188 }
1189 cursor->need_crc = new_piece;
1190
1191 return new_piece;
1192}
1193
1194static size_t sizeof_footer(struct ceph_connection *con)
1195{
1196 return (con->peer_features & CEPH_FEATURE_MSG_AUTH) ?
1197 sizeof(struct ceph_msg_footer) :
1198 sizeof(struct ceph_msg_footer_old);
1199}
1200
1201static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
1202{
1203 BUG_ON(!msg);
1204 BUG_ON(!data_len);
1205
1206 /* Initialize data cursor */
1207
1208 ceph_msg_data_cursor_init(msg, (size_t)data_len);
1209}
1210
1211/*
1212 * Prepare footer for currently outgoing message, and finish things
1213 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1214 */
1215static void prepare_write_message_footer(struct ceph_connection *con)
1216{
1217 struct ceph_msg *m = con->out_msg;
1218
1219 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1220
1221 dout("prepare_write_message_footer %p\n", con);
1222 con_out_kvec_add(con, sizeof_footer(con), &m->footer);
1223 if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
1224 if (con->ops->sign_message)
1225 con->ops->sign_message(m);
1226 else
1227 m->footer.sig = 0;
1228 } else {
1229 m->old_footer.flags = m->footer.flags;
1230 }
1231 con->out_more = m->more_to_follow;
1232 con->out_msg_done = true;
1233}
1234
1235/*
1236 * Prepare headers for the next outgoing message.
1237 */
1238static void prepare_write_message(struct ceph_connection *con)
1239{
1240 struct ceph_msg *m;
1241 u32 crc;
1242
1243 con_out_kvec_reset(con);
1244 con->out_msg_done = false;
1245
1246 /* Sneak an ack in there first? If we can get it into the same
1247 * TCP packet that's a good thing. */
1248 if (con->in_seq > con->in_seq_acked) {
1249 con->in_seq_acked = con->in_seq;
1250 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1251 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1252 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1253 &con->out_temp_ack);
1254 }
1255
1256 BUG_ON(list_empty(&con->out_queue));
1257 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1258 con->out_msg = m;
1259 BUG_ON(m->con != con);
1260
1261 /* put message on sent list */
1262 ceph_msg_get(m);
1263 list_move_tail(&m->list_head, &con->out_sent);
1264
1265 /*
1266 * only assign outgoing seq # if we haven't sent this message
1267 * yet. if it is requeued, resend with it's original seq.
1268 */
1269 if (m->needs_out_seq) {
1270 m->hdr.seq = cpu_to_le64(++con->out_seq);
1271 m->needs_out_seq = false;
1272 }
1273 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
1274
1275 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1276 m, con->out_seq, le16_to_cpu(m->hdr.type),
1277 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1278 m->data_length);
1279 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1280
1281 /* tag + hdr + front + middle */
1282 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1283 con_out_kvec_add(con, sizeof(con->out_hdr), &con->out_hdr);
1284 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1285
1286 if (m->middle)
1287 con_out_kvec_add(con, m->middle->vec.iov_len,
1288 m->middle->vec.iov_base);
1289
1290 /* fill in hdr crc and finalize hdr */
1291 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1292 con->out_msg->hdr.crc = cpu_to_le32(crc);
1293 memcpy(&con->out_hdr, &con->out_msg->hdr, sizeof(con->out_hdr));
1294
1295 /* fill in front and middle crc, footer */
1296 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1297 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1298 if (m->middle) {
1299 crc = crc32c(0, m->middle->vec.iov_base,
1300 m->middle->vec.iov_len);
1301 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1302 } else
1303 con->out_msg->footer.middle_crc = 0;
1304 dout("%s front_crc %u middle_crc %u\n", __func__,
1305 le32_to_cpu(con->out_msg->footer.front_crc),
1306 le32_to_cpu(con->out_msg->footer.middle_crc));
1307 con->out_msg->footer.flags = 0;
1308
1309 /* is there a data payload? */
1310 con->out_msg->footer.data_crc = 0;
1311 if (m->data_length) {
1312 prepare_message_data(con->out_msg, m->data_length);
1313 con->out_more = 1; /* data + footer will follow */
1314 } else {
1315 /* no, queue up footer too and be done */
1316 prepare_write_message_footer(con);
1317 }
1318
1319 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1320}
1321
1322/*
1323 * Prepare an ack.
1324 */
1325static void prepare_write_ack(struct ceph_connection *con)
1326{
1327 dout("prepare_write_ack %p %llu -> %llu\n", con,
1328 con->in_seq_acked, con->in_seq);
1329 con->in_seq_acked = con->in_seq;
1330
1331 con_out_kvec_reset(con);
1332
1333 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1334
1335 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1336 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1337 &con->out_temp_ack);
1338
1339 con->out_more = 1; /* more will follow.. eventually.. */
1340 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1341}
1342
1343/*
1344 * Prepare to share the seq during handshake
1345 */
1346static void prepare_write_seq(struct ceph_connection *con)
1347{
1348 dout("prepare_write_seq %p %llu -> %llu\n", con,
1349 con->in_seq_acked, con->in_seq);
1350 con->in_seq_acked = con->in_seq;
1351
1352 con_out_kvec_reset(con);
1353
1354 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1355 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1356 &con->out_temp_ack);
1357
1358 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1359}
1360
1361/*
1362 * Prepare to write keepalive byte.
1363 */
1364static void prepare_write_keepalive(struct ceph_connection *con)
1365{
1366 dout("prepare_write_keepalive %p\n", con);
1367 con_out_kvec_reset(con);
1368 if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
1369 struct timespec now = CURRENT_TIME;
1370
1371 con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
1372 ceph_encode_timespec(&con->out_temp_keepalive2, &now);
1373 con_out_kvec_add(con, sizeof(con->out_temp_keepalive2),
1374 &con->out_temp_keepalive2);
1375 } else {
1376 con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive);
1377 }
1378 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1379}
1380
1381/*
1382 * Connection negotiation.
1383 */
1384
1385static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1386 int *auth_proto)
1387{
1388 struct ceph_auth_handshake *auth;
1389
1390 if (!con->ops->get_authorizer) {
1391 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1392 con->out_connect.authorizer_len = 0;
1393 return NULL;
1394 }
1395
1396 /* Can't hold the mutex while getting authorizer */
1397 mutex_unlock(&con->mutex);
1398 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
1399 mutex_lock(&con->mutex);
1400
1401 if (IS_ERR(auth))
1402 return auth;
1403 if (con->state != CON_STATE_NEGOTIATING)
1404 return ERR_PTR(-EAGAIN);
1405
1406 con->auth_reply_buf = auth->authorizer_reply_buf;
1407 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
1408 return auth;
1409}
1410
1411/*
1412 * We connected to a peer and are saying hello.
1413 */
1414static void prepare_write_banner(struct ceph_connection *con)
1415{
1416 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1417 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1418 &con->msgr->my_enc_addr);
1419
1420 con->out_more = 0;
1421 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1422}
1423
1424static int prepare_write_connect(struct ceph_connection *con)
1425{
1426 unsigned int global_seq = get_global_seq(con->msgr, 0);
1427 int proto;
1428 int auth_proto;
1429 struct ceph_auth_handshake *auth;
1430
1431 switch (con->peer_name.type) {
1432 case CEPH_ENTITY_TYPE_MON:
1433 proto = CEPH_MONC_PROTOCOL;
1434 break;
1435 case CEPH_ENTITY_TYPE_OSD:
1436 proto = CEPH_OSDC_PROTOCOL;
1437 break;
1438 case CEPH_ENTITY_TYPE_MDS:
1439 proto = CEPH_MDSC_PROTOCOL;
1440 break;
1441 default:
1442 BUG();
1443 }
1444
1445 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1446 con->connect_seq, global_seq, proto);
1447
1448 con->out_connect.features =
1449 cpu_to_le64(from_msgr(con->msgr)->supported_features);
1450 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1451 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1452 con->out_connect.global_seq = cpu_to_le32(global_seq);
1453 con->out_connect.protocol_version = cpu_to_le32(proto);
1454 con->out_connect.flags = 0;
1455
1456 auth_proto = CEPH_AUTH_UNKNOWN;
1457 auth = get_connect_authorizer(con, &auth_proto);
1458 if (IS_ERR(auth))
1459 return PTR_ERR(auth);
1460
1461 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1462 con->out_connect.authorizer_len = auth ?
1463 cpu_to_le32(auth->authorizer_buf_len) : 0;
1464
1465 con_out_kvec_add(con, sizeof (con->out_connect),
1466 &con->out_connect);
1467 if (auth && auth->authorizer_buf_len)
1468 con_out_kvec_add(con, auth->authorizer_buf_len,
1469 auth->authorizer_buf);
1470
1471 con->out_more = 0;
1472 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1473
1474 return 0;
1475}
1476
1477/*
1478 * write as much of pending kvecs to the socket as we can.
1479 * 1 -> done
1480 * 0 -> socket full, but more to do
1481 * <0 -> error
1482 */
1483static int write_partial_kvec(struct ceph_connection *con)
1484{
1485 int ret;
1486
1487 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1488 while (con->out_kvec_bytes > 0) {
1489 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1490 con->out_kvec_left, con->out_kvec_bytes,
1491 con->out_more);
1492 if (ret <= 0)
1493 goto out;
1494 con->out_kvec_bytes -= ret;
1495 if (con->out_kvec_bytes == 0)
1496 break; /* done */
1497
1498 /* account for full iov entries consumed */
1499 while (ret >= con->out_kvec_cur->iov_len) {
1500 BUG_ON(!con->out_kvec_left);
1501 ret -= con->out_kvec_cur->iov_len;
1502 con->out_kvec_cur++;
1503 con->out_kvec_left--;
1504 }
1505 /* and for a partially-consumed entry */
1506 if (ret) {
1507 con->out_kvec_cur->iov_len -= ret;
1508 con->out_kvec_cur->iov_base += ret;
1509 }
1510 }
1511 con->out_kvec_left = 0;
1512 ret = 1;
1513out:
1514 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1515 con->out_kvec_bytes, con->out_kvec_left, ret);
1516 return ret; /* done! */
1517}
1518
1519static u32 ceph_crc32c_page(u32 crc, struct page *page,
1520 unsigned int page_offset,
1521 unsigned int length)
1522{
1523 char *kaddr;
1524
1525 kaddr = kmap(page);
1526 BUG_ON(kaddr == NULL);
1527 crc = crc32c(crc, kaddr + page_offset, length);
1528 kunmap(page);
1529
1530 return crc;
1531}
1532/*
1533 * Write as much message data payload as we can. If we finish, queue
1534 * up the footer.
1535 * 1 -> done, footer is now queued in out_kvec[].
1536 * 0 -> socket full, but more to do
1537 * <0 -> error
1538 */
1539static int write_partial_message_data(struct ceph_connection *con)
1540{
1541 struct ceph_msg *msg = con->out_msg;
1542 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1543 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
1544 u32 crc;
1545
1546 dout("%s %p msg %p\n", __func__, con, msg);
1547
1548 if (list_empty(&msg->data))
1549 return -EINVAL;
1550
1551 /*
1552 * Iterate through each page that contains data to be
1553 * written, and send as much as possible for each.
1554 *
1555 * If we are calculating the data crc (the default), we will
1556 * need to map the page. If we have no pages, they have
1557 * been revoked, so use the zero page.
1558 */
1559 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1560 while (cursor->resid) {
1561 struct page *page;
1562 size_t page_offset;
1563 size_t length;
1564 bool last_piece;
1565 bool need_crc;
1566 int ret;
1567
1568 page = ceph_msg_data_next(cursor, &page_offset, &length,
1569 &last_piece);
1570 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
1571 length, !last_piece);
1572 if (ret <= 0) {
1573 if (do_datacrc)
1574 msg->footer.data_crc = cpu_to_le32(crc);
1575
1576 return ret;
1577 }
1578 if (do_datacrc && cursor->need_crc)
1579 crc = ceph_crc32c_page(crc, page, page_offset, length);
1580 need_crc = ceph_msg_data_advance(cursor, (size_t)ret);
1581 }
1582
1583 dout("%s %p msg %p done\n", __func__, con, msg);
1584
1585 /* prepare and queue up footer, too */
1586 if (do_datacrc)
1587 msg->footer.data_crc = cpu_to_le32(crc);
1588 else
1589 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1590 con_out_kvec_reset(con);
1591 prepare_write_message_footer(con);
1592
1593 return 1; /* must return > 0 to indicate success */
1594}
1595
1596/*
1597 * write some zeros
1598 */
1599static int write_partial_skip(struct ceph_connection *con)
1600{
1601 int ret;
1602
1603 dout("%s %p %d left\n", __func__, con, con->out_skip);
1604 while (con->out_skip > 0) {
1605 size_t size = min(con->out_skip, (int) PAGE_SIZE);
1606
1607 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
1608 if (ret <= 0)
1609 goto out;
1610 con->out_skip -= ret;
1611 }
1612 ret = 1;
1613out:
1614 return ret;
1615}
1616
1617/*
1618 * Prepare to read connection handshake, or an ack.
1619 */
1620static void prepare_read_banner(struct ceph_connection *con)
1621{
1622 dout("prepare_read_banner %p\n", con);
1623 con->in_base_pos = 0;
1624}
1625
1626static void prepare_read_connect(struct ceph_connection *con)
1627{
1628 dout("prepare_read_connect %p\n", con);
1629 con->in_base_pos = 0;
1630}
1631
1632static void prepare_read_ack(struct ceph_connection *con)
1633{
1634 dout("prepare_read_ack %p\n", con);
1635 con->in_base_pos = 0;
1636}
1637
1638static void prepare_read_seq(struct ceph_connection *con)
1639{
1640 dout("prepare_read_seq %p\n", con);
1641 con->in_base_pos = 0;
1642 con->in_tag = CEPH_MSGR_TAG_SEQ;
1643}
1644
1645static void prepare_read_tag(struct ceph_connection *con)
1646{
1647 dout("prepare_read_tag %p\n", con);
1648 con->in_base_pos = 0;
1649 con->in_tag = CEPH_MSGR_TAG_READY;
1650}
1651
1652static void prepare_read_keepalive_ack(struct ceph_connection *con)
1653{
1654 dout("prepare_read_keepalive_ack %p\n", con);
1655 con->in_base_pos = 0;
1656}
1657
1658/*
1659 * Prepare to read a message.
1660 */
1661static int prepare_read_message(struct ceph_connection *con)
1662{
1663 dout("prepare_read_message %p\n", con);
1664 BUG_ON(con->in_msg != NULL);
1665 con->in_base_pos = 0;
1666 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1667 return 0;
1668}
1669
1670
1671static int read_partial(struct ceph_connection *con,
1672 int end, int size, void *object)
1673{
1674 while (con->in_base_pos < end) {
1675 int left = end - con->in_base_pos;
1676 int have = size - left;
1677 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1678 if (ret <= 0)
1679 return ret;
1680 con->in_base_pos += ret;
1681 }
1682 return 1;
1683}
1684
1685
1686/*
1687 * Read all or part of the connect-side handshake on a new connection
1688 */
1689static int read_partial_banner(struct ceph_connection *con)
1690{
1691 int size;
1692 int end;
1693 int ret;
1694
1695 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1696
1697 /* peer's banner */
1698 size = strlen(CEPH_BANNER);
1699 end = size;
1700 ret = read_partial(con, end, size, con->in_banner);
1701 if (ret <= 0)
1702 goto out;
1703
1704 size = sizeof (con->actual_peer_addr);
1705 end += size;
1706 ret = read_partial(con, end, size, &con->actual_peer_addr);
1707 if (ret <= 0)
1708 goto out;
1709
1710 size = sizeof (con->peer_addr_for_me);
1711 end += size;
1712 ret = read_partial(con, end, size, &con->peer_addr_for_me);
1713 if (ret <= 0)
1714 goto out;
1715
1716out:
1717 return ret;
1718}
1719
1720static int read_partial_connect(struct ceph_connection *con)
1721{
1722 int size;
1723 int end;
1724 int ret;
1725
1726 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1727
1728 size = sizeof (con->in_reply);
1729 end = size;
1730 ret = read_partial(con, end, size, &con->in_reply);
1731 if (ret <= 0)
1732 goto out;
1733
1734 size = le32_to_cpu(con->in_reply.authorizer_len);
1735 end += size;
1736 ret = read_partial(con, end, size, con->auth_reply_buf);
1737 if (ret <= 0)
1738 goto out;
1739
1740 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1741 con, (int)con->in_reply.tag,
1742 le32_to_cpu(con->in_reply.connect_seq),
1743 le32_to_cpu(con->in_reply.global_seq));
1744out:
1745 return ret;
1746
1747}
1748
1749/*
1750 * Verify the hello banner looks okay.
1751 */
1752static int verify_hello(struct ceph_connection *con)
1753{
1754 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1755 pr_err("connect to %s got bad banner\n",
1756 ceph_pr_addr(&con->peer_addr.in_addr));
1757 con->error_msg = "protocol error, bad banner";
1758 return -1;
1759 }
1760 return 0;
1761}
1762
1763static bool addr_is_blank(struct sockaddr_storage *ss)
1764{
1765 struct in_addr *addr = &((struct sockaddr_in *)ss)->sin_addr;
1766 struct in6_addr *addr6 = &((struct sockaddr_in6 *)ss)->sin6_addr;
1767
1768 switch (ss->ss_family) {
1769 case AF_INET:
1770 return addr->s_addr == htonl(INADDR_ANY);
1771 case AF_INET6:
1772 return ipv6_addr_any(addr6);
1773 default:
1774 return true;
1775 }
1776}
1777
1778static int addr_port(struct sockaddr_storage *ss)
1779{
1780 switch (ss->ss_family) {
1781 case AF_INET:
1782 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1783 case AF_INET6:
1784 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1785 }
1786 return 0;
1787}
1788
1789static void addr_set_port(struct sockaddr_storage *ss, int p)
1790{
1791 switch (ss->ss_family) {
1792 case AF_INET:
1793 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1794 break;
1795 case AF_INET6:
1796 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1797 break;
1798 }
1799}
1800
1801/*
1802 * Unlike other *_pton function semantics, zero indicates success.
1803 */
1804static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1805 char delim, const char **ipend)
1806{
1807 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1808 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1809
1810 memset(ss, 0, sizeof(*ss));
1811
1812 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1813 ss->ss_family = AF_INET;
1814 return 0;
1815 }
1816
1817 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1818 ss->ss_family = AF_INET6;
1819 return 0;
1820 }
1821
1822 return -EINVAL;
1823}
1824
1825/*
1826 * Extract hostname string and resolve using kernel DNS facility.
1827 */
1828#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1829static int ceph_dns_resolve_name(const char *name, size_t namelen,
1830 struct sockaddr_storage *ss, char delim, const char **ipend)
1831{
1832 const char *end, *delim_p;
1833 char *colon_p, *ip_addr = NULL;
1834 int ip_len, ret;
1835
1836 /*
1837 * The end of the hostname occurs immediately preceding the delimiter or
1838 * the port marker (':') where the delimiter takes precedence.
1839 */
1840 delim_p = memchr(name, delim, namelen);
1841 colon_p = memchr(name, ':', namelen);
1842
1843 if (delim_p && colon_p)
1844 end = delim_p < colon_p ? delim_p : colon_p;
1845 else if (!delim_p && colon_p)
1846 end = colon_p;
1847 else {
1848 end = delim_p;
1849 if (!end) /* case: hostname:/ */
1850 end = name + namelen;
1851 }
1852
1853 if (end <= name)
1854 return -EINVAL;
1855
1856 /* do dns_resolve upcall */
1857 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1858 if (ip_len > 0)
1859 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1860 else
1861 ret = -ESRCH;
1862
1863 kfree(ip_addr);
1864
1865 *ipend = end;
1866
1867 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1868 ret, ret ? "failed" : ceph_pr_addr(ss));
1869
1870 return ret;
1871}
1872#else
1873static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1874 struct sockaddr_storage *ss, char delim, const char **ipend)
1875{
1876 return -EINVAL;
1877}
1878#endif
1879
1880/*
1881 * Parse a server name (IP or hostname). If a valid IP address is not found
1882 * then try to extract a hostname to resolve using userspace DNS upcall.
1883 */
1884static int ceph_parse_server_name(const char *name, size_t namelen,
1885 struct sockaddr_storage *ss, char delim, const char **ipend)
1886{
1887 int ret;
1888
1889 ret = ceph_pton(name, namelen, ss, delim, ipend);
1890 if (ret)
1891 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1892
1893 return ret;
1894}
1895
1896/*
1897 * Parse an ip[:port] list into an addr array. Use the default
1898 * monitor port if a port isn't specified.
1899 */
1900int ceph_parse_ips(const char *c, const char *end,
1901 struct ceph_entity_addr *addr,
1902 int max_count, int *count)
1903{
1904 int i, ret = -EINVAL;
1905 const char *p = c;
1906
1907 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1908 for (i = 0; i < max_count; i++) {
1909 const char *ipend;
1910 struct sockaddr_storage *ss = &addr[i].in_addr;
1911 int port;
1912 char delim = ',';
1913
1914 if (*p == '[') {
1915 delim = ']';
1916 p++;
1917 }
1918
1919 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1920 if (ret)
1921 goto bad;
1922 ret = -EINVAL;
1923
1924 p = ipend;
1925
1926 if (delim == ']') {
1927 if (*p != ']') {
1928 dout("missing matching ']'\n");
1929 goto bad;
1930 }
1931 p++;
1932 }
1933
1934 /* port? */
1935 if (p < end && *p == ':') {
1936 port = 0;
1937 p++;
1938 while (p < end && *p >= '0' && *p <= '9') {
1939 port = (port * 10) + (*p - '0');
1940 p++;
1941 }
1942 if (port == 0)
1943 port = CEPH_MON_PORT;
1944 else if (port > 65535)
1945 goto bad;
1946 } else {
1947 port = CEPH_MON_PORT;
1948 }
1949
1950 addr_set_port(ss, port);
1951
1952 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1953
1954 if (p == end)
1955 break;
1956 if (*p != ',')
1957 goto bad;
1958 p++;
1959 }
1960
1961 if (p != end)
1962 goto bad;
1963
1964 if (count)
1965 *count = i + 1;
1966 return 0;
1967
1968bad:
1969 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1970 return ret;
1971}
1972EXPORT_SYMBOL(ceph_parse_ips);
1973
1974static int process_banner(struct ceph_connection *con)
1975{
1976 dout("process_banner on %p\n", con);
1977
1978 if (verify_hello(con) < 0)
1979 return -1;
1980
1981 ceph_decode_addr(&con->actual_peer_addr);
1982 ceph_decode_addr(&con->peer_addr_for_me);
1983
1984 /*
1985 * Make sure the other end is who we wanted. note that the other
1986 * end may not yet know their ip address, so if it's 0.0.0.0, give
1987 * them the benefit of the doubt.
1988 */
1989 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1990 sizeof(con->peer_addr)) != 0 &&
1991 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1992 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1993 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
1994 ceph_pr_addr(&con->peer_addr.in_addr),
1995 (int)le32_to_cpu(con->peer_addr.nonce),
1996 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1997 (int)le32_to_cpu(con->actual_peer_addr.nonce));
1998 con->error_msg = "wrong peer at address";
1999 return -1;
2000 }
2001
2002 /*
2003 * did we learn our address?
2004 */
2005 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
2006 int port = addr_port(&con->msgr->inst.addr.in_addr);
2007
2008 memcpy(&con->msgr->inst.addr.in_addr,
2009 &con->peer_addr_for_me.in_addr,
2010 sizeof(con->peer_addr_for_me.in_addr));
2011 addr_set_port(&con->msgr->inst.addr.in_addr, port);
2012 encode_my_addr(con->msgr);
2013 dout("process_banner learned my addr is %s\n",
2014 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
2015 }
2016
2017 return 0;
2018}
2019
2020static int process_connect(struct ceph_connection *con)
2021{
2022 u64 sup_feat = from_msgr(con->msgr)->supported_features;
2023 u64 req_feat = from_msgr(con->msgr)->required_features;
2024 u64 server_feat = ceph_sanitize_features(
2025 le64_to_cpu(con->in_reply.features));
2026 int ret;
2027
2028 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
2029
2030 switch (con->in_reply.tag) {
2031 case CEPH_MSGR_TAG_FEATURES:
2032 pr_err("%s%lld %s feature set mismatch,"
2033 " my %llx < server's %llx, missing %llx\n",
2034 ENTITY_NAME(con->peer_name),
2035 ceph_pr_addr(&con->peer_addr.in_addr),
2036 sup_feat, server_feat, server_feat & ~sup_feat);
2037 con->error_msg = "missing required protocol features";
2038 reset_connection(con);
2039 return -1;
2040
2041 case CEPH_MSGR_TAG_BADPROTOVER:
2042 pr_err("%s%lld %s protocol version mismatch,"
2043 " my %d != server's %d\n",
2044 ENTITY_NAME(con->peer_name),
2045 ceph_pr_addr(&con->peer_addr.in_addr),
2046 le32_to_cpu(con->out_connect.protocol_version),
2047 le32_to_cpu(con->in_reply.protocol_version));
2048 con->error_msg = "protocol version mismatch";
2049 reset_connection(con);
2050 return -1;
2051
2052 case CEPH_MSGR_TAG_BADAUTHORIZER:
2053 con->auth_retry++;
2054 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2055 con->auth_retry);
2056 if (con->auth_retry == 2) {
2057 con->error_msg = "connect authorization failure";
2058 return -1;
2059 }
2060 con_out_kvec_reset(con);
2061 ret = prepare_write_connect(con);
2062 if (ret < 0)
2063 return ret;
2064 prepare_read_connect(con);
2065 break;
2066
2067 case CEPH_MSGR_TAG_RESETSESSION:
2068 /*
2069 * If we connected with a large connect_seq but the peer
2070 * has no record of a session with us (no connection, or
2071 * connect_seq == 0), they will send RESETSESION to indicate
2072 * that they must have reset their session, and may have
2073 * dropped messages.
2074 */
2075 dout("process_connect got RESET peer seq %u\n",
2076 le32_to_cpu(con->in_reply.connect_seq));
2077 pr_err("%s%lld %s connection reset\n",
2078 ENTITY_NAME(con->peer_name),
2079 ceph_pr_addr(&con->peer_addr.in_addr));
2080 reset_connection(con);
2081 con_out_kvec_reset(con);
2082 ret = prepare_write_connect(con);
2083 if (ret < 0)
2084 return ret;
2085 prepare_read_connect(con);
2086
2087 /* Tell ceph about it. */
2088 mutex_unlock(&con->mutex);
2089 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2090 if (con->ops->peer_reset)
2091 con->ops->peer_reset(con);
2092 mutex_lock(&con->mutex);
2093 if (con->state != CON_STATE_NEGOTIATING)
2094 return -EAGAIN;
2095 break;
2096
2097 case CEPH_MSGR_TAG_RETRY_SESSION:
2098 /*
2099 * If we sent a smaller connect_seq than the peer has, try
2100 * again with a larger value.
2101 */
2102 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2103 le32_to_cpu(con->out_connect.connect_seq),
2104 le32_to_cpu(con->in_reply.connect_seq));
2105 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2106 con_out_kvec_reset(con);
2107 ret = prepare_write_connect(con);
2108 if (ret < 0)
2109 return ret;
2110 prepare_read_connect(con);
2111 break;
2112
2113 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2114 /*
2115 * If we sent a smaller global_seq than the peer has, try
2116 * again with a larger value.
2117 */
2118 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2119 con->peer_global_seq,
2120 le32_to_cpu(con->in_reply.global_seq));
2121 get_global_seq(con->msgr,
2122 le32_to_cpu(con->in_reply.global_seq));
2123 con_out_kvec_reset(con);
2124 ret = prepare_write_connect(con);
2125 if (ret < 0)
2126 return ret;
2127 prepare_read_connect(con);
2128 break;
2129
2130 case CEPH_MSGR_TAG_SEQ:
2131 case CEPH_MSGR_TAG_READY:
2132 if (req_feat & ~server_feat) {
2133 pr_err("%s%lld %s protocol feature mismatch,"
2134 " my required %llx > server's %llx, need %llx\n",
2135 ENTITY_NAME(con->peer_name),
2136 ceph_pr_addr(&con->peer_addr.in_addr),
2137 req_feat, server_feat, req_feat & ~server_feat);
2138 con->error_msg = "missing required protocol features";
2139 reset_connection(con);
2140 return -1;
2141 }
2142
2143 WARN_ON(con->state != CON_STATE_NEGOTIATING);
2144 con->state = CON_STATE_OPEN;
2145 con->auth_retry = 0; /* we authenticated; clear flag */
2146 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2147 con->connect_seq++;
2148 con->peer_features = server_feat;
2149 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2150 con->peer_global_seq,
2151 le32_to_cpu(con->in_reply.connect_seq),
2152 con->connect_seq);
2153 WARN_ON(con->connect_seq !=
2154 le32_to_cpu(con->in_reply.connect_seq));
2155
2156 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2157 con_flag_set(con, CON_FLAG_LOSSYTX);
2158
2159 con->delay = 0; /* reset backoff memory */
2160
2161 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2162 prepare_write_seq(con);
2163 prepare_read_seq(con);
2164 } else {
2165 prepare_read_tag(con);
2166 }
2167 break;
2168
2169 case CEPH_MSGR_TAG_WAIT:
2170 /*
2171 * If there is a connection race (we are opening
2172 * connections to each other), one of us may just have
2173 * to WAIT. This shouldn't happen if we are the
2174 * client.
2175 */
2176 con->error_msg = "protocol error, got WAIT as client";
2177 return -1;
2178
2179 default:
2180 con->error_msg = "protocol error, garbage tag during connect";
2181 return -1;
2182 }
2183 return 0;
2184}
2185
2186
2187/*
2188 * read (part of) an ack
2189 */
2190static int read_partial_ack(struct ceph_connection *con)
2191{
2192 int size = sizeof (con->in_temp_ack);
2193 int end = size;
2194
2195 return read_partial(con, end, size, &con->in_temp_ack);
2196}
2197
2198/*
2199 * We can finally discard anything that's been acked.
2200 */
2201static void process_ack(struct ceph_connection *con)
2202{
2203 struct ceph_msg *m;
2204 u64 ack = le64_to_cpu(con->in_temp_ack);
2205 u64 seq;
2206
2207 while (!list_empty(&con->out_sent)) {
2208 m = list_first_entry(&con->out_sent, struct ceph_msg,
2209 list_head);
2210 seq = le64_to_cpu(m->hdr.seq);
2211 if (seq > ack)
2212 break;
2213 dout("got ack for seq %llu type %d at %p\n", seq,
2214 le16_to_cpu(m->hdr.type), m);
2215 m->ack_stamp = jiffies;
2216 ceph_msg_remove(m);
2217 }
2218 prepare_read_tag(con);
2219}
2220
2221
2222static int read_partial_message_section(struct ceph_connection *con,
2223 struct kvec *section,
2224 unsigned int sec_len, u32 *crc)
2225{
2226 int ret, left;
2227
2228 BUG_ON(!section);
2229
2230 while (section->iov_len < sec_len) {
2231 BUG_ON(section->iov_base == NULL);
2232 left = sec_len - section->iov_len;
2233 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2234 section->iov_len, left);
2235 if (ret <= 0)
2236 return ret;
2237 section->iov_len += ret;
2238 }
2239 if (section->iov_len == sec_len)
2240 *crc = crc32c(0, section->iov_base, section->iov_len);
2241
2242 return 1;
2243}
2244
2245static int read_partial_msg_data(struct ceph_connection *con)
2246{
2247 struct ceph_msg *msg = con->in_msg;
2248 struct ceph_msg_data_cursor *cursor = &msg->cursor;
2249 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2250 struct page *page;
2251 size_t page_offset;
2252 size_t length;
2253 u32 crc = 0;
2254 int ret;
2255
2256 BUG_ON(!msg);
2257 if (list_empty(&msg->data))
2258 return -EIO;
2259
2260 if (do_datacrc)
2261 crc = con->in_data_crc;
2262 while (cursor->resid) {
2263 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
2264 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2265 if (ret <= 0) {
2266 if (do_datacrc)
2267 con->in_data_crc = crc;
2268
2269 return ret;
2270 }
2271
2272 if (do_datacrc)
2273 crc = ceph_crc32c_page(crc, page, page_offset, ret);
2274 (void) ceph_msg_data_advance(cursor, (size_t)ret);
2275 }
2276 if (do_datacrc)
2277 con->in_data_crc = crc;
2278
2279 return 1; /* must return > 0 to indicate success */
2280}
2281
2282/*
2283 * read (part of) a message.
2284 */
2285static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2286
2287static int read_partial_message(struct ceph_connection *con)
2288{
2289 struct ceph_msg *m = con->in_msg;
2290 int size;
2291 int end;
2292 int ret;
2293 unsigned int front_len, middle_len, data_len;
2294 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2295 bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
2296 u64 seq;
2297 u32 crc;
2298
2299 dout("read_partial_message con %p msg %p\n", con, m);
2300
2301 /* header */
2302 size = sizeof (con->in_hdr);
2303 end = size;
2304 ret = read_partial(con, end, size, &con->in_hdr);
2305 if (ret <= 0)
2306 return ret;
2307
2308 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2309 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2310 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
2311 crc, con->in_hdr.crc);
2312 return -EBADMSG;
2313 }
2314
2315 front_len = le32_to_cpu(con->in_hdr.front_len);
2316 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2317 return -EIO;
2318 middle_len = le32_to_cpu(con->in_hdr.middle_len);
2319 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2320 return -EIO;
2321 data_len = le32_to_cpu(con->in_hdr.data_len);
2322 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2323 return -EIO;
2324
2325 /* verify seq# */
2326 seq = le64_to_cpu(con->in_hdr.seq);
2327 if ((s64)seq - (s64)con->in_seq < 1) {
2328 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2329 ENTITY_NAME(con->peer_name),
2330 ceph_pr_addr(&con->peer_addr.in_addr),
2331 seq, con->in_seq + 1);
2332 con->in_base_pos = -front_len - middle_len - data_len -
2333 sizeof_footer(con);
2334 con->in_tag = CEPH_MSGR_TAG_READY;
2335 return 1;
2336 } else if ((s64)seq - (s64)con->in_seq > 1) {
2337 pr_err("read_partial_message bad seq %lld expected %lld\n",
2338 seq, con->in_seq + 1);
2339 con->error_msg = "bad message sequence # for incoming message";
2340 return -EBADE;
2341 }
2342
2343 /* allocate message? */
2344 if (!con->in_msg) {
2345 int skip = 0;
2346
2347 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2348 front_len, data_len);
2349 ret = ceph_con_in_msg_alloc(con, &skip);
2350 if (ret < 0)
2351 return ret;
2352
2353 BUG_ON(!con->in_msg ^ skip);
2354 if (skip) {
2355 /* skip this message */
2356 dout("alloc_msg said skip message\n");
2357 con->in_base_pos = -front_len - middle_len - data_len -
2358 sizeof_footer(con);
2359 con->in_tag = CEPH_MSGR_TAG_READY;
2360 con->in_seq++;
2361 return 1;
2362 }
2363
2364 BUG_ON(!con->in_msg);
2365 BUG_ON(con->in_msg->con != con);
2366 m = con->in_msg;
2367 m->front.iov_len = 0; /* haven't read it yet */
2368 if (m->middle)
2369 m->middle->vec.iov_len = 0;
2370
2371 /* prepare for data payload, if any */
2372
2373 if (data_len)
2374 prepare_message_data(con->in_msg, data_len);
2375 }
2376
2377 /* front */
2378 ret = read_partial_message_section(con, &m->front, front_len,
2379 &con->in_front_crc);
2380 if (ret <= 0)
2381 return ret;
2382
2383 /* middle */
2384 if (m->middle) {
2385 ret = read_partial_message_section(con, &m->middle->vec,
2386 middle_len,
2387 &con->in_middle_crc);
2388 if (ret <= 0)
2389 return ret;
2390 }
2391
2392 /* (page) data */
2393 if (data_len) {
2394 ret = read_partial_msg_data(con);
2395 if (ret <= 0)
2396 return ret;
2397 }
2398
2399 /* footer */
2400 size = sizeof_footer(con);
2401 end += size;
2402 ret = read_partial(con, end, size, &m->footer);
2403 if (ret <= 0)
2404 return ret;
2405
2406 if (!need_sign) {
2407 m->footer.flags = m->old_footer.flags;
2408 m->footer.sig = 0;
2409 }
2410
2411 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2412 m, front_len, m->footer.front_crc, middle_len,
2413 m->footer.middle_crc, data_len, m->footer.data_crc);
2414
2415 /* crc ok? */
2416 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2417 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2418 m, con->in_front_crc, m->footer.front_crc);
2419 return -EBADMSG;
2420 }
2421 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2422 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2423 m, con->in_middle_crc, m->footer.middle_crc);
2424 return -EBADMSG;
2425 }
2426 if (do_datacrc &&
2427 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2428 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2429 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2430 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2431 return -EBADMSG;
2432 }
2433
2434 if (need_sign && con->ops->check_message_signature &&
2435 con->ops->check_message_signature(m)) {
2436 pr_err("read_partial_message %p signature check failed\n", m);
2437 return -EBADMSG;
2438 }
2439
2440 return 1; /* done! */
2441}
2442
2443/*
2444 * Process message. This happens in the worker thread. The callback should
2445 * be careful not to do anything that waits on other incoming messages or it
2446 * may deadlock.
2447 */
2448static void process_message(struct ceph_connection *con)
2449{
2450 struct ceph_msg *msg = con->in_msg;
2451
2452 BUG_ON(con->in_msg->con != con);
2453 con->in_msg = NULL;
2454
2455 /* if first message, set peer_name */
2456 if (con->peer_name.type == 0)
2457 con->peer_name = msg->hdr.src;
2458
2459 con->in_seq++;
2460 mutex_unlock(&con->mutex);
2461
2462 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2463 msg, le64_to_cpu(msg->hdr.seq),
2464 ENTITY_NAME(msg->hdr.src),
2465 le16_to_cpu(msg->hdr.type),
2466 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2467 le32_to_cpu(msg->hdr.front_len),
2468 le32_to_cpu(msg->hdr.data_len),
2469 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2470 con->ops->dispatch(con, msg);
2471
2472 mutex_lock(&con->mutex);
2473}
2474
2475static int read_keepalive_ack(struct ceph_connection *con)
2476{
2477 struct ceph_timespec ceph_ts;
2478 size_t size = sizeof(ceph_ts);
2479 int ret = read_partial(con, size, size, &ceph_ts);
2480 if (ret <= 0)
2481 return ret;
2482 ceph_decode_timespec(&con->last_keepalive_ack, &ceph_ts);
2483 prepare_read_tag(con);
2484 return 1;
2485}
2486
2487/*
2488 * Write something to the socket. Called in a worker thread when the
2489 * socket appears to be writeable and we have something ready to send.
2490 */
2491static int try_write(struct ceph_connection *con)
2492{
2493 int ret = 1;
2494
2495 dout("try_write start %p state %lu\n", con, con->state);
2496
2497more:
2498 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2499
2500 /* open the socket first? */
2501 if (con->state == CON_STATE_PREOPEN) {
2502 BUG_ON(con->sock);
2503 con->state = CON_STATE_CONNECTING;
2504
2505 con_out_kvec_reset(con);
2506 prepare_write_banner(con);
2507 prepare_read_banner(con);
2508
2509 BUG_ON(con->in_msg);
2510 con->in_tag = CEPH_MSGR_TAG_READY;
2511 dout("try_write initiating connect on %p new state %lu\n",
2512 con, con->state);
2513 ret = ceph_tcp_connect(con);
2514 if (ret < 0) {
2515 con->error_msg = "connect error";
2516 goto out;
2517 }
2518 }
2519
2520more_kvec:
2521 /* kvec data queued? */
2522 if (con->out_kvec_left) {
2523 ret = write_partial_kvec(con);
2524 if (ret <= 0)
2525 goto out;
2526 }
2527 if (con->out_skip) {
2528 ret = write_partial_skip(con);
2529 if (ret <= 0)
2530 goto out;
2531 }
2532
2533 /* msg pages? */
2534 if (con->out_msg) {
2535 if (con->out_msg_done) {
2536 ceph_msg_put(con->out_msg);
2537 con->out_msg = NULL; /* we're done with this one */
2538 goto do_next;
2539 }
2540
2541 ret = write_partial_message_data(con);
2542 if (ret == 1)
2543 goto more_kvec; /* we need to send the footer, too! */
2544 if (ret == 0)
2545 goto out;
2546 if (ret < 0) {
2547 dout("try_write write_partial_message_data err %d\n",
2548 ret);
2549 goto out;
2550 }
2551 }
2552
2553do_next:
2554 if (con->state == CON_STATE_OPEN) {
2555 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2556 prepare_write_keepalive(con);
2557 goto more;
2558 }
2559 /* is anything else pending? */
2560 if (!list_empty(&con->out_queue)) {
2561 prepare_write_message(con);
2562 goto more;
2563 }
2564 if (con->in_seq > con->in_seq_acked) {
2565 prepare_write_ack(con);
2566 goto more;
2567 }
2568 }
2569
2570 /* Nothing to do! */
2571 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2572 dout("try_write nothing else to write.\n");
2573 ret = 0;
2574out:
2575 dout("try_write done on %p ret %d\n", con, ret);
2576 return ret;
2577}
2578
2579
2580
2581/*
2582 * Read what we can from the socket.
2583 */
2584static int try_read(struct ceph_connection *con)
2585{
2586 int ret = -1;
2587
2588more:
2589 dout("try_read start on %p state %lu\n", con, con->state);
2590 if (con->state != CON_STATE_CONNECTING &&
2591 con->state != CON_STATE_NEGOTIATING &&
2592 con->state != CON_STATE_OPEN)
2593 return 0;
2594
2595 BUG_ON(!con->sock);
2596
2597 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2598 con->in_base_pos);
2599
2600 if (con->state == CON_STATE_CONNECTING) {
2601 dout("try_read connecting\n");
2602 ret = read_partial_banner(con);
2603 if (ret <= 0)
2604 goto out;
2605 ret = process_banner(con);
2606 if (ret < 0)
2607 goto out;
2608
2609 con->state = CON_STATE_NEGOTIATING;
2610
2611 /*
2612 * Received banner is good, exchange connection info.
2613 * Do not reset out_kvec, as sending our banner raced
2614 * with receiving peer banner after connect completed.
2615 */
2616 ret = prepare_write_connect(con);
2617 if (ret < 0)
2618 goto out;
2619 prepare_read_connect(con);
2620
2621 /* Send connection info before awaiting response */
2622 goto out;
2623 }
2624
2625 if (con->state == CON_STATE_NEGOTIATING) {
2626 dout("try_read negotiating\n");
2627 ret = read_partial_connect(con);
2628 if (ret <= 0)
2629 goto out;
2630 ret = process_connect(con);
2631 if (ret < 0)
2632 goto out;
2633 goto more;
2634 }
2635
2636 WARN_ON(con->state != CON_STATE_OPEN);
2637
2638 if (con->in_base_pos < 0) {
2639 /*
2640 * skipping + discarding content.
2641 *
2642 * FIXME: there must be a better way to do this!
2643 */
2644 static char buf[SKIP_BUF_SIZE];
2645 int skip = min((int) sizeof (buf), -con->in_base_pos);
2646
2647 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2648 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2649 if (ret <= 0)
2650 goto out;
2651 con->in_base_pos += ret;
2652 if (con->in_base_pos)
2653 goto more;
2654 }
2655 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2656 /*
2657 * what's next?
2658 */
2659 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2660 if (ret <= 0)
2661 goto out;
2662 dout("try_read got tag %d\n", (int)con->in_tag);
2663 switch (con->in_tag) {
2664 case CEPH_MSGR_TAG_MSG:
2665 prepare_read_message(con);
2666 break;
2667 case CEPH_MSGR_TAG_ACK:
2668 prepare_read_ack(con);
2669 break;
2670 case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
2671 prepare_read_keepalive_ack(con);
2672 break;
2673 case CEPH_MSGR_TAG_CLOSE:
2674 con_close_socket(con);
2675 con->state = CON_STATE_CLOSED;
2676 goto out;
2677 default:
2678 goto bad_tag;
2679 }
2680 }
2681 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2682 ret = read_partial_message(con);
2683 if (ret <= 0) {
2684 switch (ret) {
2685 case -EBADMSG:
2686 con->error_msg = "bad crc/signature";
2687 /* fall through */
2688 case -EBADE:
2689 ret = -EIO;
2690 break;
2691 case -EIO:
2692 con->error_msg = "io error";
2693 break;
2694 }
2695 goto out;
2696 }
2697 if (con->in_tag == CEPH_MSGR_TAG_READY)
2698 goto more;
2699 process_message(con);
2700 if (con->state == CON_STATE_OPEN)
2701 prepare_read_tag(con);
2702 goto more;
2703 }
2704 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2705 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2706 /*
2707 * the final handshake seq exchange is semantically
2708 * equivalent to an ACK
2709 */
2710 ret = read_partial_ack(con);
2711 if (ret <= 0)
2712 goto out;
2713 process_ack(con);
2714 goto more;
2715 }
2716 if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
2717 ret = read_keepalive_ack(con);
2718 if (ret <= 0)
2719 goto out;
2720 goto more;
2721 }
2722
2723out:
2724 dout("try_read done on %p ret %d\n", con, ret);
2725 return ret;
2726
2727bad_tag:
2728 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2729 con->error_msg = "protocol error, garbage tag";
2730 ret = -1;
2731 goto out;
2732}
2733
2734
2735/*
2736 * Atomically queue work on a connection after the specified delay.
2737 * Bump @con reference to avoid races with connection teardown.
2738 * Returns 0 if work was queued, or an error code otherwise.
2739 */
2740static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2741{
2742 if (!con->ops->get(con)) {
2743 dout("%s %p ref count 0\n", __func__, con);
2744 return -ENOENT;
2745 }
2746
2747 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2748 dout("%s %p - already queued\n", __func__, con);
2749 con->ops->put(con);
2750 return -EBUSY;
2751 }
2752
2753 dout("%s %p %lu\n", __func__, con, delay);
2754 return 0;
2755}
2756
2757static void queue_con(struct ceph_connection *con)
2758{
2759 (void) queue_con_delay(con, 0);
2760}
2761
2762static void cancel_con(struct ceph_connection *con)
2763{
2764 if (cancel_delayed_work(&con->work)) {
2765 dout("%s %p\n", __func__, con);
2766 con->ops->put(con);
2767 }
2768}
2769
2770static bool con_sock_closed(struct ceph_connection *con)
2771{
2772 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2773 return false;
2774
2775#define CASE(x) \
2776 case CON_STATE_ ## x: \
2777 con->error_msg = "socket closed (con state " #x ")"; \
2778 break;
2779
2780 switch (con->state) {
2781 CASE(CLOSED);
2782 CASE(PREOPEN);
2783 CASE(CONNECTING);
2784 CASE(NEGOTIATING);
2785 CASE(OPEN);
2786 CASE(STANDBY);
2787 default:
2788 pr_warn("%s con %p unrecognized state %lu\n",
2789 __func__, con, con->state);
2790 con->error_msg = "unrecognized con state";
2791 BUG();
2792 break;
2793 }
2794#undef CASE
2795
2796 return true;
2797}
2798
2799static bool con_backoff(struct ceph_connection *con)
2800{
2801 int ret;
2802
2803 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2804 return false;
2805
2806 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2807 if (ret) {
2808 dout("%s: con %p FAILED to back off %lu\n", __func__,
2809 con, con->delay);
2810 BUG_ON(ret == -ENOENT);
2811 con_flag_set(con, CON_FLAG_BACKOFF);
2812 }
2813
2814 return true;
2815}
2816
2817/* Finish fault handling; con->mutex must *not* be held here */
2818
2819static void con_fault_finish(struct ceph_connection *con)
2820{
2821 dout("%s %p\n", __func__, con);
2822
2823 /*
2824 * in case we faulted due to authentication, invalidate our
2825 * current tickets so that we can get new ones.
2826 */
2827 if (con->auth_retry) {
2828 dout("auth_retry %d, invalidating\n", con->auth_retry);
2829 if (con->ops->invalidate_authorizer)
2830 con->ops->invalidate_authorizer(con);
2831 con->auth_retry = 0;
2832 }
2833
2834 if (con->ops->fault)
2835 con->ops->fault(con);
2836}
2837
2838/*
2839 * Do some work on a connection. Drop a connection ref when we're done.
2840 */
2841static void ceph_con_workfn(struct work_struct *work)
2842{
2843 struct ceph_connection *con = container_of(work, struct ceph_connection,
2844 work.work);
2845 bool fault;
2846
2847 mutex_lock(&con->mutex);
2848 while (true) {
2849 int ret;
2850
2851 if ((fault = con_sock_closed(con))) {
2852 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2853 break;
2854 }
2855 if (con_backoff(con)) {
2856 dout("%s: con %p BACKOFF\n", __func__, con);
2857 break;
2858 }
2859 if (con->state == CON_STATE_STANDBY) {
2860 dout("%s: con %p STANDBY\n", __func__, con);
2861 break;
2862 }
2863 if (con->state == CON_STATE_CLOSED) {
2864 dout("%s: con %p CLOSED\n", __func__, con);
2865 BUG_ON(con->sock);
2866 break;
2867 }
2868 if (con->state == CON_STATE_PREOPEN) {
2869 dout("%s: con %p PREOPEN\n", __func__, con);
2870 BUG_ON(con->sock);
2871 }
2872
2873 ret = try_read(con);
2874 if (ret < 0) {
2875 if (ret == -EAGAIN)
2876 continue;
2877 if (!con->error_msg)
2878 con->error_msg = "socket error on read";
2879 fault = true;
2880 break;
2881 }
2882
2883 ret = try_write(con);
2884 if (ret < 0) {
2885 if (ret == -EAGAIN)
2886 continue;
2887 if (!con->error_msg)
2888 con->error_msg = "socket error on write";
2889 fault = true;
2890 }
2891
2892 break; /* If we make it to here, we're done */
2893 }
2894 if (fault)
2895 con_fault(con);
2896 mutex_unlock(&con->mutex);
2897
2898 if (fault)
2899 con_fault_finish(con);
2900
2901 con->ops->put(con);
2902}
2903
2904/*
2905 * Generic error/fault handler. A retry mechanism is used with
2906 * exponential backoff
2907 */
2908static void con_fault(struct ceph_connection *con)
2909{
2910 dout("fault %p state %lu to peer %s\n",
2911 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2912
2913 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2914 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2915 con->error_msg = NULL;
2916
2917 WARN_ON(con->state != CON_STATE_CONNECTING &&
2918 con->state != CON_STATE_NEGOTIATING &&
2919 con->state != CON_STATE_OPEN);
2920
2921 con_close_socket(con);
2922
2923 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2924 dout("fault on LOSSYTX channel, marking CLOSED\n");
2925 con->state = CON_STATE_CLOSED;
2926 return;
2927 }
2928
2929 if (con->in_msg) {
2930 BUG_ON(con->in_msg->con != con);
2931 ceph_msg_put(con->in_msg);
2932 con->in_msg = NULL;
2933 }
2934
2935 /* Requeue anything that hasn't been acked */
2936 list_splice_init(&con->out_sent, &con->out_queue);
2937
2938 /* If there are no messages queued or keepalive pending, place
2939 * the connection in a STANDBY state */
2940 if (list_empty(&con->out_queue) &&
2941 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
2942 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2943 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2944 con->state = CON_STATE_STANDBY;
2945 } else {
2946 /* retry after a delay. */
2947 con->state = CON_STATE_PREOPEN;
2948 if (con->delay == 0)
2949 con->delay = BASE_DELAY_INTERVAL;
2950 else if (con->delay < MAX_DELAY_INTERVAL)
2951 con->delay *= 2;
2952 con_flag_set(con, CON_FLAG_BACKOFF);
2953 queue_con(con);
2954 }
2955}
2956
2957
2958
2959/*
2960 * initialize a new messenger instance
2961 */
2962void ceph_messenger_init(struct ceph_messenger *msgr,
2963 struct ceph_entity_addr *myaddr)
2964{
2965 spin_lock_init(&msgr->global_seq_lock);
2966
2967 if (myaddr)
2968 msgr->inst.addr = *myaddr;
2969
2970 /* select a random nonce */
2971 msgr->inst.addr.type = 0;
2972 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2973 encode_my_addr(msgr);
2974
2975 atomic_set(&msgr->stopping, 0);
2976 write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
2977
2978 dout("%s %p\n", __func__, msgr);
2979}
2980EXPORT_SYMBOL(ceph_messenger_init);
2981
2982void ceph_messenger_fini(struct ceph_messenger *msgr)
2983{
2984 put_net(read_pnet(&msgr->net));
2985}
2986EXPORT_SYMBOL(ceph_messenger_fini);
2987
2988static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con)
2989{
2990 if (msg->con)
2991 msg->con->ops->put(msg->con);
2992
2993 msg->con = con ? con->ops->get(con) : NULL;
2994 BUG_ON(msg->con != con);
2995}
2996
2997static void clear_standby(struct ceph_connection *con)
2998{
2999 /* come back from STANDBY? */
3000 if (con->state == CON_STATE_STANDBY) {
3001 dout("clear_standby %p and ++connect_seq\n", con);
3002 con->state = CON_STATE_PREOPEN;
3003 con->connect_seq++;
3004 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
3005 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
3006 }
3007}
3008
3009/*
3010 * Queue up an outgoing message on the given connection.
3011 */
3012void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
3013{
3014 /* set src+dst */
3015 msg->hdr.src = con->msgr->inst.name;
3016 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
3017 msg->needs_out_seq = true;
3018
3019 mutex_lock(&con->mutex);
3020
3021 if (con->state == CON_STATE_CLOSED) {
3022 dout("con_send %p closed, dropping %p\n", con, msg);
3023 ceph_msg_put(msg);
3024 mutex_unlock(&con->mutex);
3025 return;
3026 }
3027
3028 msg_con_set(msg, con);
3029
3030 BUG_ON(!list_empty(&msg->list_head));
3031 list_add_tail(&msg->list_head, &con->out_queue);
3032 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3033 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3034 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3035 le32_to_cpu(msg->hdr.front_len),
3036 le32_to_cpu(msg->hdr.middle_len),
3037 le32_to_cpu(msg->hdr.data_len));
3038
3039 clear_standby(con);
3040 mutex_unlock(&con->mutex);
3041
3042 /* if there wasn't anything waiting to send before, queue
3043 * new work */
3044 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3045 queue_con(con);
3046}
3047EXPORT_SYMBOL(ceph_con_send);
3048
3049/*
3050 * Revoke a message that was previously queued for send
3051 */
3052void ceph_msg_revoke(struct ceph_msg *msg)
3053{
3054 struct ceph_connection *con = msg->con;
3055
3056 if (!con) {
3057 dout("%s msg %p null con\n", __func__, msg);
3058 return; /* Message not in our possession */
3059 }
3060
3061 mutex_lock(&con->mutex);
3062 if (!list_empty(&msg->list_head)) {
3063 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
3064 list_del_init(&msg->list_head);
3065 msg->hdr.seq = 0;
3066
3067 ceph_msg_put(msg);
3068 }
3069 if (con->out_msg == msg) {
3070 BUG_ON(con->out_skip);
3071 /* footer */
3072 if (con->out_msg_done) {
3073 con->out_skip += con_out_kvec_skip(con);
3074 } else {
3075 BUG_ON(!msg->data_length);
3076 con->out_skip += sizeof_footer(con);
3077 }
3078 /* data, middle, front */
3079 if (msg->data_length)
3080 con->out_skip += msg->cursor.total_resid;
3081 if (msg->middle)
3082 con->out_skip += con_out_kvec_skip(con);
3083 con->out_skip += con_out_kvec_skip(con);
3084
3085 dout("%s %p msg %p - was sending, will write %d skip %d\n",
3086 __func__, con, msg, con->out_kvec_bytes, con->out_skip);
3087 msg->hdr.seq = 0;
3088 con->out_msg = NULL;
3089 ceph_msg_put(msg);
3090 }
3091
3092 mutex_unlock(&con->mutex);
3093}
3094
3095/*
3096 * Revoke a message that we may be reading data into
3097 */
3098void ceph_msg_revoke_incoming(struct ceph_msg *msg)
3099{
3100 struct ceph_connection *con = msg->con;
3101
3102 if (!con) {
3103 dout("%s msg %p null con\n", __func__, msg);
3104 return; /* Message not in our possession */
3105 }
3106
3107 mutex_lock(&con->mutex);
3108 if (con->in_msg == msg) {
3109 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3110 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3111 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
3112
3113 /* skip rest of message */
3114 dout("%s %p msg %p revoked\n", __func__, con, msg);
3115 con->in_base_pos = con->in_base_pos -
3116 sizeof(struct ceph_msg_header) -
3117 front_len -
3118 middle_len -
3119 data_len -
3120 sizeof(struct ceph_msg_footer);
3121 ceph_msg_put(con->in_msg);
3122 con->in_msg = NULL;
3123 con->in_tag = CEPH_MSGR_TAG_READY;
3124 con->in_seq++;
3125 } else {
3126 dout("%s %p in_msg %p msg %p no-op\n",
3127 __func__, con, con->in_msg, msg);
3128 }
3129 mutex_unlock(&con->mutex);
3130}
3131
3132/*
3133 * Queue a keepalive byte to ensure the tcp connection is alive.
3134 */
3135void ceph_con_keepalive(struct ceph_connection *con)
3136{
3137 dout("con_keepalive %p\n", con);
3138 mutex_lock(&con->mutex);
3139 clear_standby(con);
3140 mutex_unlock(&con->mutex);
3141 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3142 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3143 queue_con(con);
3144}
3145EXPORT_SYMBOL(ceph_con_keepalive);
3146
3147bool ceph_con_keepalive_expired(struct ceph_connection *con,
3148 unsigned long interval)
3149{
3150 if (interval > 0 &&
3151 (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
3152 struct timespec now = CURRENT_TIME;
3153 struct timespec ts;
3154 jiffies_to_timespec(interval, &ts);
3155 ts = timespec_add(con->last_keepalive_ack, ts);
3156 return timespec_compare(&now, &ts) >= 0;
3157 }
3158 return false;
3159}
3160
3161static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
3162{
3163 struct ceph_msg_data *data;
3164
3165 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3166 return NULL;
3167
3168 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
3169 if (data)
3170 data->type = type;
3171 INIT_LIST_HEAD(&data->links);
3172
3173 return data;
3174}
3175
3176static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3177{
3178 if (!data)
3179 return;
3180
3181 WARN_ON(!list_empty(&data->links));
3182 if (data->type == CEPH_MSG_DATA_PAGELIST)
3183 ceph_pagelist_release(data->pagelist);
3184 kmem_cache_free(ceph_msg_data_cache, data);
3185}
3186
3187void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
3188 size_t length, size_t alignment)
3189{
3190 struct ceph_msg_data *data;
3191
3192 BUG_ON(!pages);
3193 BUG_ON(!length);
3194
3195 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3196 BUG_ON(!data);
3197 data->pages = pages;
3198 data->length = length;
3199 data->alignment = alignment & ~PAGE_MASK;
3200
3201 list_add_tail(&data->links, &msg->data);
3202 msg->data_length += length;
3203}
3204EXPORT_SYMBOL(ceph_msg_data_add_pages);
3205
3206void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
3207 struct ceph_pagelist *pagelist)
3208{
3209 struct ceph_msg_data *data;
3210
3211 BUG_ON(!pagelist);
3212 BUG_ON(!pagelist->length);
3213
3214 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3215 BUG_ON(!data);
3216 data->pagelist = pagelist;
3217
3218 list_add_tail(&data->links, &msg->data);
3219 msg->data_length += pagelist->length;
3220}
3221EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
3222
3223#ifdef CONFIG_BLOCK
3224void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
3225 size_t length)
3226{
3227 struct ceph_msg_data *data;
3228
3229 BUG_ON(!bio);
3230
3231 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3232 BUG_ON(!data);
3233 data->bio = bio;
3234 data->bio_length = length;
3235
3236 list_add_tail(&data->links, &msg->data);
3237 msg->data_length += length;
3238}
3239EXPORT_SYMBOL(ceph_msg_data_add_bio);
3240#endif /* CONFIG_BLOCK */
3241
3242/*
3243 * construct a new message with given type, size
3244 * the new msg has a ref count of 1.
3245 */
3246struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3247 bool can_fail)
3248{
3249 struct ceph_msg *m;
3250
3251 m = kmem_cache_zalloc(ceph_msg_cache, flags);
3252 if (m == NULL)
3253 goto out;
3254
3255 m->hdr.type = cpu_to_le16(type);
3256 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3257 m->hdr.front_len = cpu_to_le32(front_len);
3258
3259 INIT_LIST_HEAD(&m->list_head);
3260 kref_init(&m->kref);
3261 INIT_LIST_HEAD(&m->data);
3262
3263 /* front */
3264 if (front_len) {
3265 m->front.iov_base = ceph_kvmalloc(front_len, flags);
3266 if (m->front.iov_base == NULL) {
3267 dout("ceph_msg_new can't allocate %d bytes\n",
3268 front_len);
3269 goto out2;
3270 }
3271 } else {
3272 m->front.iov_base = NULL;
3273 }
3274 m->front_alloc_len = m->front.iov_len = front_len;
3275
3276 dout("ceph_msg_new %p front %d\n", m, front_len);
3277 return m;
3278
3279out2:
3280 ceph_msg_put(m);
3281out:
3282 if (!can_fail) {
3283 pr_err("msg_new can't create type %d front %d\n", type,
3284 front_len);
3285 WARN_ON(1);
3286 } else {
3287 dout("msg_new can't create type %d front %d\n", type,
3288 front_len);
3289 }
3290 return NULL;
3291}
3292EXPORT_SYMBOL(ceph_msg_new);
3293
3294/*
3295 * Allocate "middle" portion of a message, if it is needed and wasn't
3296 * allocated by alloc_msg. This allows us to read a small fixed-size
3297 * per-type header in the front and then gracefully fail (i.e.,
3298 * propagate the error to the caller based on info in the front) when
3299 * the middle is too large.
3300 */
3301static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3302{
3303 int type = le16_to_cpu(msg->hdr.type);
3304 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3305
3306 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3307 ceph_msg_type_name(type), middle_len);
3308 BUG_ON(!middle_len);
3309 BUG_ON(msg->middle);
3310
3311 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3312 if (!msg->middle)
3313 return -ENOMEM;
3314 return 0;
3315}
3316
3317/*
3318 * Allocate a message for receiving an incoming message on a
3319 * connection, and save the result in con->in_msg. Uses the
3320 * connection's private alloc_msg op if available.
3321 *
3322 * Returns 0 on success, or a negative error code.
3323 *
3324 * On success, if we set *skip = 1:
3325 * - the next message should be skipped and ignored.
3326 * - con->in_msg == NULL
3327 * or if we set *skip = 0:
3328 * - con->in_msg is non-null.
3329 * On error (ENOMEM, EAGAIN, ...),
3330 * - con->in_msg == NULL
3331 */
3332static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3333{
3334 struct ceph_msg_header *hdr = &con->in_hdr;
3335 int middle_len = le32_to_cpu(hdr->middle_len);
3336 struct ceph_msg *msg;
3337 int ret = 0;
3338
3339 BUG_ON(con->in_msg != NULL);
3340 BUG_ON(!con->ops->alloc_msg);
3341
3342 mutex_unlock(&con->mutex);
3343 msg = con->ops->alloc_msg(con, hdr, skip);
3344 mutex_lock(&con->mutex);
3345 if (con->state != CON_STATE_OPEN) {
3346 if (msg)
3347 ceph_msg_put(msg);
3348 return -EAGAIN;
3349 }
3350 if (msg) {
3351 BUG_ON(*skip);
3352 msg_con_set(msg, con);
3353 con->in_msg = msg;
3354 } else {
3355 /*
3356 * Null message pointer means either we should skip
3357 * this message or we couldn't allocate memory. The
3358 * former is not an error.
3359 */
3360 if (*skip)
3361 return 0;
3362
3363 con->error_msg = "error allocating memory for incoming message";
3364 return -ENOMEM;
3365 }
3366 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3367
3368 if (middle_len && !con->in_msg->middle) {
3369 ret = ceph_alloc_middle(con, con->in_msg);
3370 if (ret < 0) {
3371 ceph_msg_put(con->in_msg);
3372 con->in_msg = NULL;
3373 }
3374 }
3375
3376 return ret;
3377}
3378
3379
3380/*
3381 * Free a generically kmalloc'd message.
3382 */
3383static void ceph_msg_free(struct ceph_msg *m)
3384{
3385 dout("%s %p\n", __func__, m);
3386 kvfree(m->front.iov_base);
3387 kmem_cache_free(ceph_msg_cache, m);
3388}
3389
3390static void ceph_msg_release(struct kref *kref)
3391{
3392 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3393 struct ceph_msg_data *data, *next;
3394
3395 dout("%s %p\n", __func__, m);
3396 WARN_ON(!list_empty(&m->list_head));
3397
3398 msg_con_set(m, NULL);
3399
3400 /* drop middle, data, if any */
3401 if (m->middle) {
3402 ceph_buffer_put(m->middle);
3403 m->middle = NULL;
3404 }
3405
3406 list_for_each_entry_safe(data, next, &m->data, links) {
3407 list_del_init(&data->links);
3408 ceph_msg_data_destroy(data);
3409 }
3410 m->data_length = 0;
3411
3412 if (m->pool)
3413 ceph_msgpool_put(m->pool, m);
3414 else
3415 ceph_msg_free(m);
3416}
3417
3418struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3419{
3420 dout("%s %p (was %d)\n", __func__, msg,
3421 atomic_read(&msg->kref.refcount));
3422 kref_get(&msg->kref);
3423 return msg;
3424}
3425EXPORT_SYMBOL(ceph_msg_get);
3426
3427void ceph_msg_put(struct ceph_msg *msg)
3428{
3429 dout("%s %p (was %d)\n", __func__, msg,
3430 atomic_read(&msg->kref.refcount));
3431 kref_put(&msg->kref, ceph_msg_release);
3432}
3433EXPORT_SYMBOL(ceph_msg_put);
3434
3435void ceph_msg_dump(struct ceph_msg *msg)
3436{
3437 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3438 msg->front_alloc_len, msg->data_length);
3439 print_hex_dump(KERN_DEBUG, "header: ",
3440 DUMP_PREFIX_OFFSET, 16, 1,
3441 &msg->hdr, sizeof(msg->hdr), true);
3442 print_hex_dump(KERN_DEBUG, " front: ",
3443 DUMP_PREFIX_OFFSET, 16, 1,
3444 msg->front.iov_base, msg->front.iov_len, true);
3445 if (msg->middle)
3446 print_hex_dump(KERN_DEBUG, "middle: ",
3447 DUMP_PREFIX_OFFSET, 16, 1,
3448 msg->middle->vec.iov_base,
3449 msg->middle->vec.iov_len, true);
3450 print_hex_dump(KERN_DEBUG, "footer: ",
3451 DUMP_PREFIX_OFFSET, 16, 1,
3452 &msg->footer, sizeof(msg->footer), true);
3453}
3454EXPORT_SYMBOL(ceph_msg_dump);
1// SPDX-License-Identifier: GPL-2.0
2#include <linux/ceph/ceph_debug.h>
3
4#include <linux/crc32c.h>
5#include <linux/ctype.h>
6#include <linux/highmem.h>
7#include <linux/inet.h>
8#include <linux/kthread.h>
9#include <linux/net.h>
10#include <linux/nsproxy.h>
11#include <linux/sched/mm.h>
12#include <linux/slab.h>
13#include <linux/socket.h>
14#include <linux/string.h>
15#ifdef CONFIG_BLOCK
16#include <linux/bio.h>
17#endif /* CONFIG_BLOCK */
18#include <linux/dns_resolver.h>
19#include <net/tcp.h>
20#include <trace/events/sock.h>
21
22#include <linux/ceph/ceph_features.h>
23#include <linux/ceph/libceph.h>
24#include <linux/ceph/messenger.h>
25#include <linux/ceph/decode.h>
26#include <linux/ceph/pagelist.h>
27#include <linux/export.h>
28
29/*
30 * Ceph uses the messenger to exchange ceph_msg messages with other
31 * hosts in the system. The messenger provides ordered and reliable
32 * delivery. We tolerate TCP disconnects by reconnecting (with
33 * exponential backoff) in the case of a fault (disconnection, bad
34 * crc, protocol error). Acks allow sent messages to be discarded by
35 * the sender.
36 */
37
38/*
39 * We track the state of the socket on a given connection using
40 * values defined below. The transition to a new socket state is
41 * handled by a function which verifies we aren't coming from an
42 * unexpected state.
43 *
44 * --------
45 * | NEW* | transient initial state
46 * --------
47 * | con_sock_state_init()
48 * v
49 * ----------
50 * | CLOSED | initialized, but no socket (and no
51 * ---------- TCP connection)
52 * ^ \
53 * | \ con_sock_state_connecting()
54 * | ----------------------
55 * | \
56 * + con_sock_state_closed() \
57 * |+--------------------------- \
58 * | \ \ \
59 * | ----------- \ \
60 * | | CLOSING | socket event; \ \
61 * | ----------- await close \ \
62 * | ^ \ |
63 * | | \ |
64 * | + con_sock_state_closing() \ |
65 * | / \ | |
66 * | / --------------- | |
67 * | / \ v v
68 * | / --------------
69 * | / -----------------| CONNECTING | socket created, TCP
70 * | | / -------------- connect initiated
71 * | | | con_sock_state_connected()
72 * | | v
73 * -------------
74 * | CONNECTED | TCP connection established
75 * -------------
76 *
77 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
78 */
79
80#define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
81#define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
82#define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
83#define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
84#define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
85
86static bool con_flag_valid(unsigned long con_flag)
87{
88 switch (con_flag) {
89 case CEPH_CON_F_LOSSYTX:
90 case CEPH_CON_F_KEEPALIVE_PENDING:
91 case CEPH_CON_F_WRITE_PENDING:
92 case CEPH_CON_F_SOCK_CLOSED:
93 case CEPH_CON_F_BACKOFF:
94 return true;
95 default:
96 return false;
97 }
98}
99
100void ceph_con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
101{
102 BUG_ON(!con_flag_valid(con_flag));
103
104 clear_bit(con_flag, &con->flags);
105}
106
107void ceph_con_flag_set(struct ceph_connection *con, unsigned long con_flag)
108{
109 BUG_ON(!con_flag_valid(con_flag));
110
111 set_bit(con_flag, &con->flags);
112}
113
114bool ceph_con_flag_test(struct ceph_connection *con, unsigned long con_flag)
115{
116 BUG_ON(!con_flag_valid(con_flag));
117
118 return test_bit(con_flag, &con->flags);
119}
120
121bool ceph_con_flag_test_and_clear(struct ceph_connection *con,
122 unsigned long con_flag)
123{
124 BUG_ON(!con_flag_valid(con_flag));
125
126 return test_and_clear_bit(con_flag, &con->flags);
127}
128
129bool ceph_con_flag_test_and_set(struct ceph_connection *con,
130 unsigned long con_flag)
131{
132 BUG_ON(!con_flag_valid(con_flag));
133
134 return test_and_set_bit(con_flag, &con->flags);
135}
136
137/* Slab caches for frequently-allocated structures */
138
139static struct kmem_cache *ceph_msg_cache;
140
141#ifdef CONFIG_LOCKDEP
142static struct lock_class_key socket_class;
143#endif
144
145static void queue_con(struct ceph_connection *con);
146static void cancel_con(struct ceph_connection *con);
147static void ceph_con_workfn(struct work_struct *);
148static void con_fault(struct ceph_connection *con);
149
150/*
151 * Nicely render a sockaddr as a string. An array of formatted
152 * strings is used, to approximate reentrancy.
153 */
154#define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
155#define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
156#define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
157#define MAX_ADDR_STR_LEN 64 /* 54 is enough */
158
159static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
160static atomic_t addr_str_seq = ATOMIC_INIT(0);
161
162struct page *ceph_zero_page; /* used in certain error cases */
163
164const char *ceph_pr_addr(const struct ceph_entity_addr *addr)
165{
166 int i;
167 char *s;
168 struct sockaddr_storage ss = addr->in_addr; /* align */
169 struct sockaddr_in *in4 = (struct sockaddr_in *)&ss;
170 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *)&ss;
171
172 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
173 s = addr_str[i];
174
175 switch (ss.ss_family) {
176 case AF_INET:
177 snprintf(s, MAX_ADDR_STR_LEN, "(%d)%pI4:%hu",
178 le32_to_cpu(addr->type), &in4->sin_addr,
179 ntohs(in4->sin_port));
180 break;
181
182 case AF_INET6:
183 snprintf(s, MAX_ADDR_STR_LEN, "(%d)[%pI6c]:%hu",
184 le32_to_cpu(addr->type), &in6->sin6_addr,
185 ntohs(in6->sin6_port));
186 break;
187
188 default:
189 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
190 ss.ss_family);
191 }
192
193 return s;
194}
195EXPORT_SYMBOL(ceph_pr_addr);
196
197void ceph_encode_my_addr(struct ceph_messenger *msgr)
198{
199 if (!ceph_msgr2(from_msgr(msgr))) {
200 memcpy(&msgr->my_enc_addr, &msgr->inst.addr,
201 sizeof(msgr->my_enc_addr));
202 ceph_encode_banner_addr(&msgr->my_enc_addr);
203 }
204}
205
206/*
207 * work queue for all reading and writing to/from the socket.
208 */
209static struct workqueue_struct *ceph_msgr_wq;
210
211static int ceph_msgr_slab_init(void)
212{
213 BUG_ON(ceph_msg_cache);
214 ceph_msg_cache = KMEM_CACHE(ceph_msg, 0);
215 if (!ceph_msg_cache)
216 return -ENOMEM;
217
218 return 0;
219}
220
221static void ceph_msgr_slab_exit(void)
222{
223 BUG_ON(!ceph_msg_cache);
224 kmem_cache_destroy(ceph_msg_cache);
225 ceph_msg_cache = NULL;
226}
227
228static void _ceph_msgr_exit(void)
229{
230 if (ceph_msgr_wq) {
231 destroy_workqueue(ceph_msgr_wq);
232 ceph_msgr_wq = NULL;
233 }
234
235 BUG_ON(!ceph_zero_page);
236 put_page(ceph_zero_page);
237 ceph_zero_page = NULL;
238
239 ceph_msgr_slab_exit();
240}
241
242int __init ceph_msgr_init(void)
243{
244 if (ceph_msgr_slab_init())
245 return -ENOMEM;
246
247 BUG_ON(ceph_zero_page);
248 ceph_zero_page = ZERO_PAGE(0);
249 get_page(ceph_zero_page);
250
251 /*
252 * The number of active work items is limited by the number of
253 * connections, so leave @max_active at default.
254 */
255 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
256 if (ceph_msgr_wq)
257 return 0;
258
259 pr_err("msgr_init failed to create workqueue\n");
260 _ceph_msgr_exit();
261
262 return -ENOMEM;
263}
264
265void ceph_msgr_exit(void)
266{
267 BUG_ON(ceph_msgr_wq == NULL);
268
269 _ceph_msgr_exit();
270}
271
272void ceph_msgr_flush(void)
273{
274 flush_workqueue(ceph_msgr_wq);
275}
276EXPORT_SYMBOL(ceph_msgr_flush);
277
278/* Connection socket state transition functions */
279
280static void con_sock_state_init(struct ceph_connection *con)
281{
282 int old_state;
283
284 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
285 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
286 printk("%s: unexpected old state %d\n", __func__, old_state);
287 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
288 CON_SOCK_STATE_CLOSED);
289}
290
291static void con_sock_state_connecting(struct ceph_connection *con)
292{
293 int old_state;
294
295 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
296 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
297 printk("%s: unexpected old state %d\n", __func__, old_state);
298 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
299 CON_SOCK_STATE_CONNECTING);
300}
301
302static void con_sock_state_connected(struct ceph_connection *con)
303{
304 int old_state;
305
306 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
307 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
308 printk("%s: unexpected old state %d\n", __func__, old_state);
309 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
310 CON_SOCK_STATE_CONNECTED);
311}
312
313static void con_sock_state_closing(struct ceph_connection *con)
314{
315 int old_state;
316
317 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
318 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
319 old_state != CON_SOCK_STATE_CONNECTED &&
320 old_state != CON_SOCK_STATE_CLOSING))
321 printk("%s: unexpected old state %d\n", __func__, old_state);
322 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
323 CON_SOCK_STATE_CLOSING);
324}
325
326static void con_sock_state_closed(struct ceph_connection *con)
327{
328 int old_state;
329
330 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
331 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
332 old_state != CON_SOCK_STATE_CLOSING &&
333 old_state != CON_SOCK_STATE_CONNECTING &&
334 old_state != CON_SOCK_STATE_CLOSED))
335 printk("%s: unexpected old state %d\n", __func__, old_state);
336 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
337 CON_SOCK_STATE_CLOSED);
338}
339
340/*
341 * socket callback functions
342 */
343
344/* data available on socket, or listen socket received a connect */
345static void ceph_sock_data_ready(struct sock *sk)
346{
347 struct ceph_connection *con = sk->sk_user_data;
348
349 trace_sk_data_ready(sk);
350
351 if (atomic_read(&con->msgr->stopping)) {
352 return;
353 }
354
355 if (sk->sk_state != TCP_CLOSE_WAIT) {
356 dout("%s %p state = %d, queueing work\n", __func__,
357 con, con->state);
358 queue_con(con);
359 }
360}
361
362/* socket has buffer space for writing */
363static void ceph_sock_write_space(struct sock *sk)
364{
365 struct ceph_connection *con = sk->sk_user_data;
366
367 /* only queue to workqueue if there is data we want to write,
368 * and there is sufficient space in the socket buffer to accept
369 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
370 * doesn't get called again until try_write() fills the socket
371 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
372 * and net/core/stream.c:sk_stream_write_space().
373 */
374 if (ceph_con_flag_test(con, CEPH_CON_F_WRITE_PENDING)) {
375 if (sk_stream_is_writeable(sk)) {
376 dout("%s %p queueing write work\n", __func__, con);
377 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
378 queue_con(con);
379 }
380 } else {
381 dout("%s %p nothing to write\n", __func__, con);
382 }
383}
384
385/* socket's state has changed */
386static void ceph_sock_state_change(struct sock *sk)
387{
388 struct ceph_connection *con = sk->sk_user_data;
389
390 dout("%s %p state = %d sk_state = %u\n", __func__,
391 con, con->state, sk->sk_state);
392
393 switch (sk->sk_state) {
394 case TCP_CLOSE:
395 dout("%s TCP_CLOSE\n", __func__);
396 fallthrough;
397 case TCP_CLOSE_WAIT:
398 dout("%s TCP_CLOSE_WAIT\n", __func__);
399 con_sock_state_closing(con);
400 ceph_con_flag_set(con, CEPH_CON_F_SOCK_CLOSED);
401 queue_con(con);
402 break;
403 case TCP_ESTABLISHED:
404 dout("%s TCP_ESTABLISHED\n", __func__);
405 con_sock_state_connected(con);
406 queue_con(con);
407 break;
408 default: /* Everything else is uninteresting */
409 break;
410 }
411}
412
413/*
414 * set up socket callbacks
415 */
416static void set_sock_callbacks(struct socket *sock,
417 struct ceph_connection *con)
418{
419 struct sock *sk = sock->sk;
420 sk->sk_user_data = con;
421 sk->sk_data_ready = ceph_sock_data_ready;
422 sk->sk_write_space = ceph_sock_write_space;
423 sk->sk_state_change = ceph_sock_state_change;
424}
425
426
427/*
428 * socket helpers
429 */
430
431/*
432 * initiate connection to a remote socket.
433 */
434int ceph_tcp_connect(struct ceph_connection *con)
435{
436 struct sockaddr_storage ss = con->peer_addr.in_addr; /* align */
437 struct socket *sock;
438 unsigned int noio_flag;
439 int ret;
440
441 dout("%s con %p peer_addr %s\n", __func__, con,
442 ceph_pr_addr(&con->peer_addr));
443 BUG_ON(con->sock);
444
445 /* sock_create_kern() allocates with GFP_KERNEL */
446 noio_flag = memalloc_noio_save();
447 ret = sock_create_kern(read_pnet(&con->msgr->net), ss.ss_family,
448 SOCK_STREAM, IPPROTO_TCP, &sock);
449 memalloc_noio_restore(noio_flag);
450 if (ret)
451 return ret;
452 sock->sk->sk_allocation = GFP_NOFS;
453 sock->sk->sk_use_task_frag = false;
454
455#ifdef CONFIG_LOCKDEP
456 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
457#endif
458
459 set_sock_callbacks(sock, con);
460
461 con_sock_state_connecting(con);
462 ret = kernel_connect(sock, (struct sockaddr *)&ss, sizeof(ss),
463 O_NONBLOCK);
464 if (ret == -EINPROGRESS) {
465 dout("connect %s EINPROGRESS sk_state = %u\n",
466 ceph_pr_addr(&con->peer_addr),
467 sock->sk->sk_state);
468 } else if (ret < 0) {
469 pr_err("connect %s error %d\n",
470 ceph_pr_addr(&con->peer_addr), ret);
471 sock_release(sock);
472 return ret;
473 }
474
475 if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY))
476 tcp_sock_set_nodelay(sock->sk);
477
478 con->sock = sock;
479 return 0;
480}
481
482/*
483 * Shutdown/close the socket for the given connection.
484 */
485int ceph_con_close_socket(struct ceph_connection *con)
486{
487 int rc = 0;
488
489 dout("%s con %p sock %p\n", __func__, con, con->sock);
490 if (con->sock) {
491 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
492 sock_release(con->sock);
493 con->sock = NULL;
494 }
495
496 /*
497 * Forcibly clear the SOCK_CLOSED flag. It gets set
498 * independent of the connection mutex, and we could have
499 * received a socket close event before we had the chance to
500 * shut the socket down.
501 */
502 ceph_con_flag_clear(con, CEPH_CON_F_SOCK_CLOSED);
503
504 con_sock_state_closed(con);
505 return rc;
506}
507
508static void ceph_con_reset_protocol(struct ceph_connection *con)
509{
510 dout("%s con %p\n", __func__, con);
511
512 ceph_con_close_socket(con);
513 if (con->in_msg) {
514 WARN_ON(con->in_msg->con != con);
515 ceph_msg_put(con->in_msg);
516 con->in_msg = NULL;
517 }
518 if (con->out_msg) {
519 WARN_ON(con->out_msg->con != con);
520 ceph_msg_put(con->out_msg);
521 con->out_msg = NULL;
522 }
523 if (con->bounce_page) {
524 __free_page(con->bounce_page);
525 con->bounce_page = NULL;
526 }
527
528 if (ceph_msgr2(from_msgr(con->msgr)))
529 ceph_con_v2_reset_protocol(con);
530 else
531 ceph_con_v1_reset_protocol(con);
532}
533
534/*
535 * Reset a connection. Discard all incoming and outgoing messages
536 * and clear *_seq state.
537 */
538static void ceph_msg_remove(struct ceph_msg *msg)
539{
540 list_del_init(&msg->list_head);
541
542 ceph_msg_put(msg);
543}
544
545static void ceph_msg_remove_list(struct list_head *head)
546{
547 while (!list_empty(head)) {
548 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
549 list_head);
550 ceph_msg_remove(msg);
551 }
552}
553
554void ceph_con_reset_session(struct ceph_connection *con)
555{
556 dout("%s con %p\n", __func__, con);
557
558 WARN_ON(con->in_msg);
559 WARN_ON(con->out_msg);
560 ceph_msg_remove_list(&con->out_queue);
561 ceph_msg_remove_list(&con->out_sent);
562 con->out_seq = 0;
563 con->in_seq = 0;
564 con->in_seq_acked = 0;
565
566 if (ceph_msgr2(from_msgr(con->msgr)))
567 ceph_con_v2_reset_session(con);
568 else
569 ceph_con_v1_reset_session(con);
570}
571
572/*
573 * mark a peer down. drop any open connections.
574 */
575void ceph_con_close(struct ceph_connection *con)
576{
577 mutex_lock(&con->mutex);
578 dout("con_close %p peer %s\n", con, ceph_pr_addr(&con->peer_addr));
579 con->state = CEPH_CON_S_CLOSED;
580
581 ceph_con_flag_clear(con, CEPH_CON_F_LOSSYTX); /* so we retry next
582 connect */
583 ceph_con_flag_clear(con, CEPH_CON_F_KEEPALIVE_PENDING);
584 ceph_con_flag_clear(con, CEPH_CON_F_WRITE_PENDING);
585 ceph_con_flag_clear(con, CEPH_CON_F_BACKOFF);
586
587 ceph_con_reset_protocol(con);
588 ceph_con_reset_session(con);
589 cancel_con(con);
590 mutex_unlock(&con->mutex);
591}
592EXPORT_SYMBOL(ceph_con_close);
593
594/*
595 * Reopen a closed connection, with a new peer address.
596 */
597void ceph_con_open(struct ceph_connection *con,
598 __u8 entity_type, __u64 entity_num,
599 struct ceph_entity_addr *addr)
600{
601 mutex_lock(&con->mutex);
602 dout("con_open %p %s\n", con, ceph_pr_addr(addr));
603
604 WARN_ON(con->state != CEPH_CON_S_CLOSED);
605 con->state = CEPH_CON_S_PREOPEN;
606
607 con->peer_name.type = (__u8) entity_type;
608 con->peer_name.num = cpu_to_le64(entity_num);
609
610 memcpy(&con->peer_addr, addr, sizeof(*addr));
611 con->delay = 0; /* reset backoff memory */
612 mutex_unlock(&con->mutex);
613 queue_con(con);
614}
615EXPORT_SYMBOL(ceph_con_open);
616
617/*
618 * return true if this connection ever successfully opened
619 */
620bool ceph_con_opened(struct ceph_connection *con)
621{
622 if (ceph_msgr2(from_msgr(con->msgr)))
623 return ceph_con_v2_opened(con);
624
625 return ceph_con_v1_opened(con);
626}
627
628/*
629 * initialize a new connection.
630 */
631void ceph_con_init(struct ceph_connection *con, void *private,
632 const struct ceph_connection_operations *ops,
633 struct ceph_messenger *msgr)
634{
635 dout("con_init %p\n", con);
636 memset(con, 0, sizeof(*con));
637 con->private = private;
638 con->ops = ops;
639 con->msgr = msgr;
640
641 con_sock_state_init(con);
642
643 mutex_init(&con->mutex);
644 INIT_LIST_HEAD(&con->out_queue);
645 INIT_LIST_HEAD(&con->out_sent);
646 INIT_DELAYED_WORK(&con->work, ceph_con_workfn);
647
648 con->state = CEPH_CON_S_CLOSED;
649}
650EXPORT_SYMBOL(ceph_con_init);
651
652/*
653 * We maintain a global counter to order connection attempts. Get
654 * a unique seq greater than @gt.
655 */
656u32 ceph_get_global_seq(struct ceph_messenger *msgr, u32 gt)
657{
658 u32 ret;
659
660 spin_lock(&msgr->global_seq_lock);
661 if (msgr->global_seq < gt)
662 msgr->global_seq = gt;
663 ret = ++msgr->global_seq;
664 spin_unlock(&msgr->global_seq_lock);
665 return ret;
666}
667
668/*
669 * Discard messages that have been acked by the server.
670 */
671void ceph_con_discard_sent(struct ceph_connection *con, u64 ack_seq)
672{
673 struct ceph_msg *msg;
674 u64 seq;
675
676 dout("%s con %p ack_seq %llu\n", __func__, con, ack_seq);
677 while (!list_empty(&con->out_sent)) {
678 msg = list_first_entry(&con->out_sent, struct ceph_msg,
679 list_head);
680 WARN_ON(msg->needs_out_seq);
681 seq = le64_to_cpu(msg->hdr.seq);
682 if (seq > ack_seq)
683 break;
684
685 dout("%s con %p discarding msg %p seq %llu\n", __func__, con,
686 msg, seq);
687 ceph_msg_remove(msg);
688 }
689}
690
691/*
692 * Discard messages that have been requeued in con_fault(), up to
693 * reconnect_seq. This avoids gratuitously resending messages that
694 * the server had received and handled prior to reconnect.
695 */
696void ceph_con_discard_requeued(struct ceph_connection *con, u64 reconnect_seq)
697{
698 struct ceph_msg *msg;
699 u64 seq;
700
701 dout("%s con %p reconnect_seq %llu\n", __func__, con, reconnect_seq);
702 while (!list_empty(&con->out_queue)) {
703 msg = list_first_entry(&con->out_queue, struct ceph_msg,
704 list_head);
705 if (msg->needs_out_seq)
706 break;
707 seq = le64_to_cpu(msg->hdr.seq);
708 if (seq > reconnect_seq)
709 break;
710
711 dout("%s con %p discarding msg %p seq %llu\n", __func__, con,
712 msg, seq);
713 ceph_msg_remove(msg);
714 }
715}
716
717#ifdef CONFIG_BLOCK
718
719/*
720 * For a bio data item, a piece is whatever remains of the next
721 * entry in the current bio iovec, or the first entry in the next
722 * bio in the list.
723 */
724static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
725 size_t length)
726{
727 struct ceph_msg_data *data = cursor->data;
728 struct ceph_bio_iter *it = &cursor->bio_iter;
729
730 cursor->resid = min_t(size_t, length, data->bio_length);
731 *it = data->bio_pos;
732 if (cursor->resid < it->iter.bi_size)
733 it->iter.bi_size = cursor->resid;
734
735 BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
736}
737
738static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
739 size_t *page_offset,
740 size_t *length)
741{
742 struct bio_vec bv = bio_iter_iovec(cursor->bio_iter.bio,
743 cursor->bio_iter.iter);
744
745 *page_offset = bv.bv_offset;
746 *length = bv.bv_len;
747 return bv.bv_page;
748}
749
750static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
751 size_t bytes)
752{
753 struct ceph_bio_iter *it = &cursor->bio_iter;
754 struct page *page = bio_iter_page(it->bio, it->iter);
755
756 BUG_ON(bytes > cursor->resid);
757 BUG_ON(bytes > bio_iter_len(it->bio, it->iter));
758 cursor->resid -= bytes;
759 bio_advance_iter(it->bio, &it->iter, bytes);
760
761 if (!cursor->resid)
762 return false; /* no more data */
763
764 if (!bytes || (it->iter.bi_size && it->iter.bi_bvec_done &&
765 page == bio_iter_page(it->bio, it->iter)))
766 return false; /* more bytes to process in this segment */
767
768 if (!it->iter.bi_size) {
769 it->bio = it->bio->bi_next;
770 it->iter = it->bio->bi_iter;
771 if (cursor->resid < it->iter.bi_size)
772 it->iter.bi_size = cursor->resid;
773 }
774
775 BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
776 return true;
777}
778#endif /* CONFIG_BLOCK */
779
780static void ceph_msg_data_bvecs_cursor_init(struct ceph_msg_data_cursor *cursor,
781 size_t length)
782{
783 struct ceph_msg_data *data = cursor->data;
784 struct bio_vec *bvecs = data->bvec_pos.bvecs;
785
786 cursor->resid = min_t(size_t, length, data->bvec_pos.iter.bi_size);
787 cursor->bvec_iter = data->bvec_pos.iter;
788 cursor->bvec_iter.bi_size = cursor->resid;
789
790 BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
791}
792
793static struct page *ceph_msg_data_bvecs_next(struct ceph_msg_data_cursor *cursor,
794 size_t *page_offset,
795 size_t *length)
796{
797 struct bio_vec bv = bvec_iter_bvec(cursor->data->bvec_pos.bvecs,
798 cursor->bvec_iter);
799
800 *page_offset = bv.bv_offset;
801 *length = bv.bv_len;
802 return bv.bv_page;
803}
804
805static bool ceph_msg_data_bvecs_advance(struct ceph_msg_data_cursor *cursor,
806 size_t bytes)
807{
808 struct bio_vec *bvecs = cursor->data->bvec_pos.bvecs;
809 struct page *page = bvec_iter_page(bvecs, cursor->bvec_iter);
810
811 BUG_ON(bytes > cursor->resid);
812 BUG_ON(bytes > bvec_iter_len(bvecs, cursor->bvec_iter));
813 cursor->resid -= bytes;
814 bvec_iter_advance(bvecs, &cursor->bvec_iter, bytes);
815
816 if (!cursor->resid)
817 return false; /* no more data */
818
819 if (!bytes || (cursor->bvec_iter.bi_bvec_done &&
820 page == bvec_iter_page(bvecs, cursor->bvec_iter)))
821 return false; /* more bytes to process in this segment */
822
823 BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
824 return true;
825}
826
827/*
828 * For a page array, a piece comes from the first page in the array
829 * that has not already been fully consumed.
830 */
831static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
832 size_t length)
833{
834 struct ceph_msg_data *data = cursor->data;
835 int page_count;
836
837 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
838
839 BUG_ON(!data->pages);
840 BUG_ON(!data->length);
841
842 cursor->resid = min(length, data->length);
843 page_count = calc_pages_for(data->alignment, (u64)data->length);
844 cursor->page_offset = data->alignment & ~PAGE_MASK;
845 cursor->page_index = 0;
846 BUG_ON(page_count > (int)USHRT_MAX);
847 cursor->page_count = (unsigned short)page_count;
848 BUG_ON(length > SIZE_MAX - cursor->page_offset);
849}
850
851static struct page *
852ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
853 size_t *page_offset, size_t *length)
854{
855 struct ceph_msg_data *data = cursor->data;
856
857 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
858
859 BUG_ON(cursor->page_index >= cursor->page_count);
860 BUG_ON(cursor->page_offset >= PAGE_SIZE);
861
862 *page_offset = cursor->page_offset;
863 *length = min_t(size_t, cursor->resid, PAGE_SIZE - *page_offset);
864 return data->pages[cursor->page_index];
865}
866
867static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
868 size_t bytes)
869{
870 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
871
872 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
873
874 /* Advance the cursor page offset */
875
876 cursor->resid -= bytes;
877 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
878 if (!bytes || cursor->page_offset)
879 return false; /* more bytes to process in the current page */
880
881 if (!cursor->resid)
882 return false; /* no more data */
883
884 /* Move on to the next page; offset is already at 0 */
885
886 BUG_ON(cursor->page_index >= cursor->page_count);
887 cursor->page_index++;
888 return true;
889}
890
891/*
892 * For a pagelist, a piece is whatever remains to be consumed in the
893 * first page in the list, or the front of the next page.
894 */
895static void
896ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
897 size_t length)
898{
899 struct ceph_msg_data *data = cursor->data;
900 struct ceph_pagelist *pagelist;
901 struct page *page;
902
903 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
904
905 pagelist = data->pagelist;
906 BUG_ON(!pagelist);
907
908 if (!length)
909 return; /* pagelist can be assigned but empty */
910
911 BUG_ON(list_empty(&pagelist->head));
912 page = list_first_entry(&pagelist->head, struct page, lru);
913
914 cursor->resid = min(length, pagelist->length);
915 cursor->page = page;
916 cursor->offset = 0;
917}
918
919static struct page *
920ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
921 size_t *page_offset, size_t *length)
922{
923 struct ceph_msg_data *data = cursor->data;
924 struct ceph_pagelist *pagelist;
925
926 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
927
928 pagelist = data->pagelist;
929 BUG_ON(!pagelist);
930
931 BUG_ON(!cursor->page);
932 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
933
934 /* offset of first page in pagelist is always 0 */
935 *page_offset = cursor->offset & ~PAGE_MASK;
936 *length = min_t(size_t, cursor->resid, PAGE_SIZE - *page_offset);
937 return cursor->page;
938}
939
940static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
941 size_t bytes)
942{
943 struct ceph_msg_data *data = cursor->data;
944 struct ceph_pagelist *pagelist;
945
946 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
947
948 pagelist = data->pagelist;
949 BUG_ON(!pagelist);
950
951 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
952 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
953
954 /* Advance the cursor offset */
955
956 cursor->resid -= bytes;
957 cursor->offset += bytes;
958 /* offset of first page in pagelist is always 0 */
959 if (!bytes || cursor->offset & ~PAGE_MASK)
960 return false; /* more bytes to process in the current page */
961
962 if (!cursor->resid)
963 return false; /* no more data */
964
965 /* Move on to the next page */
966
967 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
968 cursor->page = list_next_entry(cursor->page, lru);
969 return true;
970}
971
972static void ceph_msg_data_iter_cursor_init(struct ceph_msg_data_cursor *cursor,
973 size_t length)
974{
975 struct ceph_msg_data *data = cursor->data;
976
977 cursor->iov_iter = data->iter;
978 cursor->lastlen = 0;
979 iov_iter_truncate(&cursor->iov_iter, length);
980 cursor->resid = iov_iter_count(&cursor->iov_iter);
981}
982
983static struct page *ceph_msg_data_iter_next(struct ceph_msg_data_cursor *cursor,
984 size_t *page_offset, size_t *length)
985{
986 struct page *page;
987 ssize_t len;
988
989 if (cursor->lastlen)
990 iov_iter_revert(&cursor->iov_iter, cursor->lastlen);
991
992 len = iov_iter_get_pages2(&cursor->iov_iter, &page, PAGE_SIZE,
993 1, page_offset);
994 BUG_ON(len < 0);
995
996 cursor->lastlen = len;
997
998 /*
999 * FIXME: The assumption is that the pages represented by the iov_iter
1000 * are pinned, with the references held by the upper-level
1001 * callers, or by virtue of being under writeback. Eventually,
1002 * we'll get an iov_iter_get_pages2 variant that doesn't take
1003 * page refs. Until then, just put the page ref.
1004 */
1005 VM_BUG_ON_PAGE(!PageWriteback(page) && page_count(page) < 2, page);
1006 put_page(page);
1007
1008 *length = min_t(size_t, len, cursor->resid);
1009 return page;
1010}
1011
1012static bool ceph_msg_data_iter_advance(struct ceph_msg_data_cursor *cursor,
1013 size_t bytes)
1014{
1015 BUG_ON(bytes > cursor->resid);
1016 cursor->resid -= bytes;
1017
1018 if (bytes < cursor->lastlen) {
1019 cursor->lastlen -= bytes;
1020 } else {
1021 iov_iter_advance(&cursor->iov_iter, bytes - cursor->lastlen);
1022 cursor->lastlen = 0;
1023 }
1024
1025 return cursor->resid;
1026}
1027
1028/*
1029 * Message data is handled (sent or received) in pieces, where each
1030 * piece resides on a single page. The network layer might not
1031 * consume an entire piece at once. A data item's cursor keeps
1032 * track of which piece is next to process and how much remains to
1033 * be processed in that piece. It also tracks whether the current
1034 * piece is the last one in the data item.
1035 */
1036static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1037{
1038 size_t length = cursor->total_resid;
1039
1040 switch (cursor->data->type) {
1041 case CEPH_MSG_DATA_PAGELIST:
1042 ceph_msg_data_pagelist_cursor_init(cursor, length);
1043 break;
1044 case CEPH_MSG_DATA_PAGES:
1045 ceph_msg_data_pages_cursor_init(cursor, length);
1046 break;
1047#ifdef CONFIG_BLOCK
1048 case CEPH_MSG_DATA_BIO:
1049 ceph_msg_data_bio_cursor_init(cursor, length);
1050 break;
1051#endif /* CONFIG_BLOCK */
1052 case CEPH_MSG_DATA_BVECS:
1053 ceph_msg_data_bvecs_cursor_init(cursor, length);
1054 break;
1055 case CEPH_MSG_DATA_ITER:
1056 ceph_msg_data_iter_cursor_init(cursor, length);
1057 break;
1058 case CEPH_MSG_DATA_NONE:
1059 default:
1060 /* BUG(); */
1061 break;
1062 }
1063 cursor->need_crc = true;
1064}
1065
1066void ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor,
1067 struct ceph_msg *msg, size_t length)
1068{
1069 BUG_ON(!length);
1070 BUG_ON(length > msg->data_length);
1071 BUG_ON(!msg->num_data_items);
1072
1073 cursor->total_resid = length;
1074 cursor->data = msg->data;
1075 cursor->sr_resid = 0;
1076
1077 __ceph_msg_data_cursor_init(cursor);
1078}
1079
1080/*
1081 * Return the page containing the next piece to process for a given
1082 * data item, and supply the page offset and length of that piece.
1083 * Indicate whether this is the last piece in this data item.
1084 */
1085struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1086 size_t *page_offset, size_t *length)
1087{
1088 struct page *page;
1089
1090 switch (cursor->data->type) {
1091 case CEPH_MSG_DATA_PAGELIST:
1092 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1093 break;
1094 case CEPH_MSG_DATA_PAGES:
1095 page = ceph_msg_data_pages_next(cursor, page_offset, length);
1096 break;
1097#ifdef CONFIG_BLOCK
1098 case CEPH_MSG_DATA_BIO:
1099 page = ceph_msg_data_bio_next(cursor, page_offset, length);
1100 break;
1101#endif /* CONFIG_BLOCK */
1102 case CEPH_MSG_DATA_BVECS:
1103 page = ceph_msg_data_bvecs_next(cursor, page_offset, length);
1104 break;
1105 case CEPH_MSG_DATA_ITER:
1106 page = ceph_msg_data_iter_next(cursor, page_offset, length);
1107 break;
1108 case CEPH_MSG_DATA_NONE:
1109 default:
1110 page = NULL;
1111 break;
1112 }
1113
1114 BUG_ON(!page);
1115 BUG_ON(*page_offset + *length > PAGE_SIZE);
1116 BUG_ON(!*length);
1117 BUG_ON(*length > cursor->resid);
1118
1119 return page;
1120}
1121
1122/*
1123 * Returns true if the result moves the cursor on to the next piece
1124 * of the data item.
1125 */
1126void ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor, size_t bytes)
1127{
1128 bool new_piece;
1129
1130 BUG_ON(bytes > cursor->resid);
1131 switch (cursor->data->type) {
1132 case CEPH_MSG_DATA_PAGELIST:
1133 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1134 break;
1135 case CEPH_MSG_DATA_PAGES:
1136 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1137 break;
1138#ifdef CONFIG_BLOCK
1139 case CEPH_MSG_DATA_BIO:
1140 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1141 break;
1142#endif /* CONFIG_BLOCK */
1143 case CEPH_MSG_DATA_BVECS:
1144 new_piece = ceph_msg_data_bvecs_advance(cursor, bytes);
1145 break;
1146 case CEPH_MSG_DATA_ITER:
1147 new_piece = ceph_msg_data_iter_advance(cursor, bytes);
1148 break;
1149 case CEPH_MSG_DATA_NONE:
1150 default:
1151 BUG();
1152 break;
1153 }
1154 cursor->total_resid -= bytes;
1155
1156 if (!cursor->resid && cursor->total_resid) {
1157 cursor->data++;
1158 __ceph_msg_data_cursor_init(cursor);
1159 new_piece = true;
1160 }
1161 cursor->need_crc = new_piece;
1162}
1163
1164u32 ceph_crc32c_page(u32 crc, struct page *page, unsigned int page_offset,
1165 unsigned int length)
1166{
1167 char *kaddr;
1168
1169 kaddr = kmap(page);
1170 BUG_ON(kaddr == NULL);
1171 crc = crc32c(crc, kaddr + page_offset, length);
1172 kunmap(page);
1173
1174 return crc;
1175}
1176
1177bool ceph_addr_is_blank(const struct ceph_entity_addr *addr)
1178{
1179 struct sockaddr_storage ss = addr->in_addr; /* align */
1180 struct in_addr *addr4 = &((struct sockaddr_in *)&ss)->sin_addr;
1181 struct in6_addr *addr6 = &((struct sockaddr_in6 *)&ss)->sin6_addr;
1182
1183 switch (ss.ss_family) {
1184 case AF_INET:
1185 return addr4->s_addr == htonl(INADDR_ANY);
1186 case AF_INET6:
1187 return ipv6_addr_any(addr6);
1188 default:
1189 return true;
1190 }
1191}
1192EXPORT_SYMBOL(ceph_addr_is_blank);
1193
1194int ceph_addr_port(const struct ceph_entity_addr *addr)
1195{
1196 switch (get_unaligned(&addr->in_addr.ss_family)) {
1197 case AF_INET:
1198 return ntohs(get_unaligned(&((struct sockaddr_in *)&addr->in_addr)->sin_port));
1199 case AF_INET6:
1200 return ntohs(get_unaligned(&((struct sockaddr_in6 *)&addr->in_addr)->sin6_port));
1201 }
1202 return 0;
1203}
1204
1205void ceph_addr_set_port(struct ceph_entity_addr *addr, int p)
1206{
1207 switch (get_unaligned(&addr->in_addr.ss_family)) {
1208 case AF_INET:
1209 put_unaligned(htons(p), &((struct sockaddr_in *)&addr->in_addr)->sin_port);
1210 break;
1211 case AF_INET6:
1212 put_unaligned(htons(p), &((struct sockaddr_in6 *)&addr->in_addr)->sin6_port);
1213 break;
1214 }
1215}
1216
1217/*
1218 * Unlike other *_pton function semantics, zero indicates success.
1219 */
1220static int ceph_pton(const char *str, size_t len, struct ceph_entity_addr *addr,
1221 char delim, const char **ipend)
1222{
1223 memset(&addr->in_addr, 0, sizeof(addr->in_addr));
1224
1225 if (in4_pton(str, len, (u8 *)&((struct sockaddr_in *)&addr->in_addr)->sin_addr.s_addr, delim, ipend)) {
1226 put_unaligned(AF_INET, &addr->in_addr.ss_family);
1227 return 0;
1228 }
1229
1230 if (in6_pton(str, len, (u8 *)&((struct sockaddr_in6 *)&addr->in_addr)->sin6_addr.s6_addr, delim, ipend)) {
1231 put_unaligned(AF_INET6, &addr->in_addr.ss_family);
1232 return 0;
1233 }
1234
1235 return -EINVAL;
1236}
1237
1238/*
1239 * Extract hostname string and resolve using kernel DNS facility.
1240 */
1241#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1242static int ceph_dns_resolve_name(const char *name, size_t namelen,
1243 struct ceph_entity_addr *addr, char delim, const char **ipend)
1244{
1245 const char *end, *delim_p;
1246 char *colon_p, *ip_addr = NULL;
1247 int ip_len, ret;
1248
1249 /*
1250 * The end of the hostname occurs immediately preceding the delimiter or
1251 * the port marker (':') where the delimiter takes precedence.
1252 */
1253 delim_p = memchr(name, delim, namelen);
1254 colon_p = memchr(name, ':', namelen);
1255
1256 if (delim_p && colon_p)
1257 end = delim_p < colon_p ? delim_p : colon_p;
1258 else if (!delim_p && colon_p)
1259 end = colon_p;
1260 else {
1261 end = delim_p;
1262 if (!end) /* case: hostname:/ */
1263 end = name + namelen;
1264 }
1265
1266 if (end <= name)
1267 return -EINVAL;
1268
1269 /* do dns_resolve upcall */
1270 ip_len = dns_query(current->nsproxy->net_ns,
1271 NULL, name, end - name, NULL, &ip_addr, NULL, false);
1272 if (ip_len > 0)
1273 ret = ceph_pton(ip_addr, ip_len, addr, -1, NULL);
1274 else
1275 ret = -ESRCH;
1276
1277 kfree(ip_addr);
1278
1279 *ipend = end;
1280
1281 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1282 ret, ret ? "failed" : ceph_pr_addr(addr));
1283
1284 return ret;
1285}
1286#else
1287static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1288 struct ceph_entity_addr *addr, char delim, const char **ipend)
1289{
1290 return -EINVAL;
1291}
1292#endif
1293
1294/*
1295 * Parse a server name (IP or hostname). If a valid IP address is not found
1296 * then try to extract a hostname to resolve using userspace DNS upcall.
1297 */
1298static int ceph_parse_server_name(const char *name, size_t namelen,
1299 struct ceph_entity_addr *addr, char delim, const char **ipend)
1300{
1301 int ret;
1302
1303 ret = ceph_pton(name, namelen, addr, delim, ipend);
1304 if (ret)
1305 ret = ceph_dns_resolve_name(name, namelen, addr, delim, ipend);
1306
1307 return ret;
1308}
1309
1310/*
1311 * Parse an ip[:port] list into an addr array. Use the default
1312 * monitor port if a port isn't specified.
1313 */
1314int ceph_parse_ips(const char *c, const char *end,
1315 struct ceph_entity_addr *addr,
1316 int max_count, int *count, char delim)
1317{
1318 int i, ret = -EINVAL;
1319 const char *p = c;
1320
1321 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1322 for (i = 0; i < max_count; i++) {
1323 char cur_delim = delim;
1324 const char *ipend;
1325 int port;
1326
1327 if (*p == '[') {
1328 cur_delim = ']';
1329 p++;
1330 }
1331
1332 ret = ceph_parse_server_name(p, end - p, &addr[i], cur_delim,
1333 &ipend);
1334 if (ret)
1335 goto bad;
1336 ret = -EINVAL;
1337
1338 p = ipend;
1339
1340 if (cur_delim == ']') {
1341 if (*p != ']') {
1342 dout("missing matching ']'\n");
1343 goto bad;
1344 }
1345 p++;
1346 }
1347
1348 /* port? */
1349 if (p < end && *p == ':') {
1350 port = 0;
1351 p++;
1352 while (p < end && *p >= '0' && *p <= '9') {
1353 port = (port * 10) + (*p - '0');
1354 p++;
1355 }
1356 if (port == 0)
1357 port = CEPH_MON_PORT;
1358 else if (port > 65535)
1359 goto bad;
1360 } else {
1361 port = CEPH_MON_PORT;
1362 }
1363
1364 ceph_addr_set_port(&addr[i], port);
1365 /*
1366 * We want the type to be set according to ms_mode
1367 * option, but options are normally parsed after mon
1368 * addresses. Rather than complicating parsing, set
1369 * to LEGACY and override in build_initial_monmap()
1370 * for mon addresses and ceph_messenger_init() for
1371 * ip option.
1372 */
1373 addr[i].type = CEPH_ENTITY_ADDR_TYPE_LEGACY;
1374 addr[i].nonce = 0;
1375
1376 dout("%s got %s\n", __func__, ceph_pr_addr(&addr[i]));
1377
1378 if (p == end)
1379 break;
1380 if (*p != delim)
1381 goto bad;
1382 p++;
1383 }
1384
1385 if (p != end)
1386 goto bad;
1387
1388 if (count)
1389 *count = i + 1;
1390 return 0;
1391
1392bad:
1393 return ret;
1394}
1395
1396/*
1397 * Process message. This happens in the worker thread. The callback should
1398 * be careful not to do anything that waits on other incoming messages or it
1399 * may deadlock.
1400 */
1401void ceph_con_process_message(struct ceph_connection *con)
1402{
1403 struct ceph_msg *msg = con->in_msg;
1404
1405 BUG_ON(con->in_msg->con != con);
1406 con->in_msg = NULL;
1407
1408 /* if first message, set peer_name */
1409 if (con->peer_name.type == 0)
1410 con->peer_name = msg->hdr.src;
1411
1412 con->in_seq++;
1413 mutex_unlock(&con->mutex);
1414
1415 dout("===== %p %llu from %s%lld %d=%s len %d+%d+%d (%u %u %u) =====\n",
1416 msg, le64_to_cpu(msg->hdr.seq),
1417 ENTITY_NAME(msg->hdr.src),
1418 le16_to_cpu(msg->hdr.type),
1419 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
1420 le32_to_cpu(msg->hdr.front_len),
1421 le32_to_cpu(msg->hdr.middle_len),
1422 le32_to_cpu(msg->hdr.data_len),
1423 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
1424 con->ops->dispatch(con, msg);
1425
1426 mutex_lock(&con->mutex);
1427}
1428
1429/*
1430 * Atomically queue work on a connection after the specified delay.
1431 * Bump @con reference to avoid races with connection teardown.
1432 * Returns 0 if work was queued, or an error code otherwise.
1433 */
1434static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
1435{
1436 if (!con->ops->get(con)) {
1437 dout("%s %p ref count 0\n", __func__, con);
1438 return -ENOENT;
1439 }
1440
1441 if (delay >= HZ)
1442 delay = round_jiffies_relative(delay);
1443
1444 dout("%s %p %lu\n", __func__, con, delay);
1445 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
1446 dout("%s %p - already queued\n", __func__, con);
1447 con->ops->put(con);
1448 return -EBUSY;
1449 }
1450
1451 return 0;
1452}
1453
1454static void queue_con(struct ceph_connection *con)
1455{
1456 (void) queue_con_delay(con, 0);
1457}
1458
1459static void cancel_con(struct ceph_connection *con)
1460{
1461 if (cancel_delayed_work(&con->work)) {
1462 dout("%s %p\n", __func__, con);
1463 con->ops->put(con);
1464 }
1465}
1466
1467static bool con_sock_closed(struct ceph_connection *con)
1468{
1469 if (!ceph_con_flag_test_and_clear(con, CEPH_CON_F_SOCK_CLOSED))
1470 return false;
1471
1472#define CASE(x) \
1473 case CEPH_CON_S_ ## x: \
1474 con->error_msg = "socket closed (con state " #x ")"; \
1475 break;
1476
1477 switch (con->state) {
1478 CASE(CLOSED);
1479 CASE(PREOPEN);
1480 CASE(V1_BANNER);
1481 CASE(V1_CONNECT_MSG);
1482 CASE(V2_BANNER_PREFIX);
1483 CASE(V2_BANNER_PAYLOAD);
1484 CASE(V2_HELLO);
1485 CASE(V2_AUTH);
1486 CASE(V2_AUTH_SIGNATURE);
1487 CASE(V2_SESSION_CONNECT);
1488 CASE(V2_SESSION_RECONNECT);
1489 CASE(OPEN);
1490 CASE(STANDBY);
1491 default:
1492 BUG();
1493 }
1494#undef CASE
1495
1496 return true;
1497}
1498
1499static bool con_backoff(struct ceph_connection *con)
1500{
1501 int ret;
1502
1503 if (!ceph_con_flag_test_and_clear(con, CEPH_CON_F_BACKOFF))
1504 return false;
1505
1506 ret = queue_con_delay(con, con->delay);
1507 if (ret) {
1508 dout("%s: con %p FAILED to back off %lu\n", __func__,
1509 con, con->delay);
1510 BUG_ON(ret == -ENOENT);
1511 ceph_con_flag_set(con, CEPH_CON_F_BACKOFF);
1512 }
1513
1514 return true;
1515}
1516
1517/* Finish fault handling; con->mutex must *not* be held here */
1518
1519static void con_fault_finish(struct ceph_connection *con)
1520{
1521 dout("%s %p\n", __func__, con);
1522
1523 /*
1524 * in case we faulted due to authentication, invalidate our
1525 * current tickets so that we can get new ones.
1526 */
1527 if (con->v1.auth_retry) {
1528 dout("auth_retry %d, invalidating\n", con->v1.auth_retry);
1529 if (con->ops->invalidate_authorizer)
1530 con->ops->invalidate_authorizer(con);
1531 con->v1.auth_retry = 0;
1532 }
1533
1534 if (con->ops->fault)
1535 con->ops->fault(con);
1536}
1537
1538/*
1539 * Do some work on a connection. Drop a connection ref when we're done.
1540 */
1541static void ceph_con_workfn(struct work_struct *work)
1542{
1543 struct ceph_connection *con = container_of(work, struct ceph_connection,
1544 work.work);
1545 bool fault;
1546
1547 mutex_lock(&con->mutex);
1548 while (true) {
1549 int ret;
1550
1551 if ((fault = con_sock_closed(con))) {
1552 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
1553 break;
1554 }
1555 if (con_backoff(con)) {
1556 dout("%s: con %p BACKOFF\n", __func__, con);
1557 break;
1558 }
1559 if (con->state == CEPH_CON_S_STANDBY) {
1560 dout("%s: con %p STANDBY\n", __func__, con);
1561 break;
1562 }
1563 if (con->state == CEPH_CON_S_CLOSED) {
1564 dout("%s: con %p CLOSED\n", __func__, con);
1565 BUG_ON(con->sock);
1566 break;
1567 }
1568 if (con->state == CEPH_CON_S_PREOPEN) {
1569 dout("%s: con %p PREOPEN\n", __func__, con);
1570 BUG_ON(con->sock);
1571 }
1572
1573 if (ceph_msgr2(from_msgr(con->msgr)))
1574 ret = ceph_con_v2_try_read(con);
1575 else
1576 ret = ceph_con_v1_try_read(con);
1577 if (ret < 0) {
1578 if (ret == -EAGAIN)
1579 continue;
1580 if (!con->error_msg)
1581 con->error_msg = "socket error on read";
1582 fault = true;
1583 break;
1584 }
1585
1586 if (ceph_msgr2(from_msgr(con->msgr)))
1587 ret = ceph_con_v2_try_write(con);
1588 else
1589 ret = ceph_con_v1_try_write(con);
1590 if (ret < 0) {
1591 if (ret == -EAGAIN)
1592 continue;
1593 if (!con->error_msg)
1594 con->error_msg = "socket error on write";
1595 fault = true;
1596 }
1597
1598 break; /* If we make it to here, we're done */
1599 }
1600 if (fault)
1601 con_fault(con);
1602 mutex_unlock(&con->mutex);
1603
1604 if (fault)
1605 con_fault_finish(con);
1606
1607 con->ops->put(con);
1608}
1609
1610/*
1611 * Generic error/fault handler. A retry mechanism is used with
1612 * exponential backoff
1613 */
1614static void con_fault(struct ceph_connection *con)
1615{
1616 dout("fault %p state %d to peer %s\n",
1617 con, con->state, ceph_pr_addr(&con->peer_addr));
1618
1619 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
1620 ceph_pr_addr(&con->peer_addr), con->error_msg);
1621 con->error_msg = NULL;
1622
1623 WARN_ON(con->state == CEPH_CON_S_STANDBY ||
1624 con->state == CEPH_CON_S_CLOSED);
1625
1626 ceph_con_reset_protocol(con);
1627
1628 if (ceph_con_flag_test(con, CEPH_CON_F_LOSSYTX)) {
1629 dout("fault on LOSSYTX channel, marking CLOSED\n");
1630 con->state = CEPH_CON_S_CLOSED;
1631 return;
1632 }
1633
1634 /* Requeue anything that hasn't been acked */
1635 list_splice_init(&con->out_sent, &con->out_queue);
1636
1637 /* If there are no messages queued or keepalive pending, place
1638 * the connection in a STANDBY state */
1639 if (list_empty(&con->out_queue) &&
1640 !ceph_con_flag_test(con, CEPH_CON_F_KEEPALIVE_PENDING)) {
1641 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
1642 ceph_con_flag_clear(con, CEPH_CON_F_WRITE_PENDING);
1643 con->state = CEPH_CON_S_STANDBY;
1644 } else {
1645 /* retry after a delay. */
1646 con->state = CEPH_CON_S_PREOPEN;
1647 if (!con->delay) {
1648 con->delay = BASE_DELAY_INTERVAL;
1649 } else if (con->delay < MAX_DELAY_INTERVAL) {
1650 con->delay *= 2;
1651 if (con->delay > MAX_DELAY_INTERVAL)
1652 con->delay = MAX_DELAY_INTERVAL;
1653 }
1654 ceph_con_flag_set(con, CEPH_CON_F_BACKOFF);
1655 queue_con(con);
1656 }
1657}
1658
1659void ceph_messenger_reset_nonce(struct ceph_messenger *msgr)
1660{
1661 u32 nonce = le32_to_cpu(msgr->inst.addr.nonce) + 1000000;
1662 msgr->inst.addr.nonce = cpu_to_le32(nonce);
1663 ceph_encode_my_addr(msgr);
1664}
1665
1666/*
1667 * initialize a new messenger instance
1668 */
1669void ceph_messenger_init(struct ceph_messenger *msgr,
1670 struct ceph_entity_addr *myaddr)
1671{
1672 spin_lock_init(&msgr->global_seq_lock);
1673
1674 if (myaddr) {
1675 memcpy(&msgr->inst.addr.in_addr, &myaddr->in_addr,
1676 sizeof(msgr->inst.addr.in_addr));
1677 ceph_addr_set_port(&msgr->inst.addr, 0);
1678 }
1679
1680 /*
1681 * Since nautilus, clients are identified using type ANY.
1682 * For msgr1, ceph_encode_banner_addr() munges it to NONE.
1683 */
1684 msgr->inst.addr.type = CEPH_ENTITY_ADDR_TYPE_ANY;
1685
1686 /* generate a random non-zero nonce */
1687 do {
1688 get_random_bytes(&msgr->inst.addr.nonce,
1689 sizeof(msgr->inst.addr.nonce));
1690 } while (!msgr->inst.addr.nonce);
1691 ceph_encode_my_addr(msgr);
1692
1693 atomic_set(&msgr->stopping, 0);
1694 write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
1695
1696 dout("%s %p\n", __func__, msgr);
1697}
1698
1699void ceph_messenger_fini(struct ceph_messenger *msgr)
1700{
1701 put_net(read_pnet(&msgr->net));
1702}
1703
1704static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con)
1705{
1706 if (msg->con)
1707 msg->con->ops->put(msg->con);
1708
1709 msg->con = con ? con->ops->get(con) : NULL;
1710 BUG_ON(msg->con != con);
1711}
1712
1713static void clear_standby(struct ceph_connection *con)
1714{
1715 /* come back from STANDBY? */
1716 if (con->state == CEPH_CON_S_STANDBY) {
1717 dout("clear_standby %p and ++connect_seq\n", con);
1718 con->state = CEPH_CON_S_PREOPEN;
1719 con->v1.connect_seq++;
1720 WARN_ON(ceph_con_flag_test(con, CEPH_CON_F_WRITE_PENDING));
1721 WARN_ON(ceph_con_flag_test(con, CEPH_CON_F_KEEPALIVE_PENDING));
1722 }
1723}
1724
1725/*
1726 * Queue up an outgoing message on the given connection.
1727 *
1728 * Consumes a ref on @msg.
1729 */
1730void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
1731{
1732 /* set src+dst */
1733 msg->hdr.src = con->msgr->inst.name;
1734 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
1735 msg->needs_out_seq = true;
1736
1737 mutex_lock(&con->mutex);
1738
1739 if (con->state == CEPH_CON_S_CLOSED) {
1740 dout("con_send %p closed, dropping %p\n", con, msg);
1741 ceph_msg_put(msg);
1742 mutex_unlock(&con->mutex);
1743 return;
1744 }
1745
1746 msg_con_set(msg, con);
1747
1748 BUG_ON(!list_empty(&msg->list_head));
1749 list_add_tail(&msg->list_head, &con->out_queue);
1750 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
1751 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
1752 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
1753 le32_to_cpu(msg->hdr.front_len),
1754 le32_to_cpu(msg->hdr.middle_len),
1755 le32_to_cpu(msg->hdr.data_len));
1756
1757 clear_standby(con);
1758 mutex_unlock(&con->mutex);
1759
1760 /* if there wasn't anything waiting to send before, queue
1761 * new work */
1762 if (!ceph_con_flag_test_and_set(con, CEPH_CON_F_WRITE_PENDING))
1763 queue_con(con);
1764}
1765EXPORT_SYMBOL(ceph_con_send);
1766
1767/*
1768 * Revoke a message that was previously queued for send
1769 */
1770void ceph_msg_revoke(struct ceph_msg *msg)
1771{
1772 struct ceph_connection *con = msg->con;
1773
1774 if (!con) {
1775 dout("%s msg %p null con\n", __func__, msg);
1776 return; /* Message not in our possession */
1777 }
1778
1779 mutex_lock(&con->mutex);
1780 if (list_empty(&msg->list_head)) {
1781 WARN_ON(con->out_msg == msg);
1782 dout("%s con %p msg %p not linked\n", __func__, con, msg);
1783 mutex_unlock(&con->mutex);
1784 return;
1785 }
1786
1787 dout("%s con %p msg %p was linked\n", __func__, con, msg);
1788 msg->hdr.seq = 0;
1789 ceph_msg_remove(msg);
1790
1791 if (con->out_msg == msg) {
1792 WARN_ON(con->state != CEPH_CON_S_OPEN);
1793 dout("%s con %p msg %p was sending\n", __func__, con, msg);
1794 if (ceph_msgr2(from_msgr(con->msgr)))
1795 ceph_con_v2_revoke(con);
1796 else
1797 ceph_con_v1_revoke(con);
1798 ceph_msg_put(con->out_msg);
1799 con->out_msg = NULL;
1800 } else {
1801 dout("%s con %p msg %p not current, out_msg %p\n", __func__,
1802 con, msg, con->out_msg);
1803 }
1804 mutex_unlock(&con->mutex);
1805}
1806
1807/*
1808 * Revoke a message that we may be reading data into
1809 */
1810void ceph_msg_revoke_incoming(struct ceph_msg *msg)
1811{
1812 struct ceph_connection *con = msg->con;
1813
1814 if (!con) {
1815 dout("%s msg %p null con\n", __func__, msg);
1816 return; /* Message not in our possession */
1817 }
1818
1819 mutex_lock(&con->mutex);
1820 if (con->in_msg == msg) {
1821 WARN_ON(con->state != CEPH_CON_S_OPEN);
1822 dout("%s con %p msg %p was recving\n", __func__, con, msg);
1823 if (ceph_msgr2(from_msgr(con->msgr)))
1824 ceph_con_v2_revoke_incoming(con);
1825 else
1826 ceph_con_v1_revoke_incoming(con);
1827 ceph_msg_put(con->in_msg);
1828 con->in_msg = NULL;
1829 } else {
1830 dout("%s con %p msg %p not current, in_msg %p\n", __func__,
1831 con, msg, con->in_msg);
1832 }
1833 mutex_unlock(&con->mutex);
1834}
1835
1836/*
1837 * Queue a keepalive byte to ensure the tcp connection is alive.
1838 */
1839void ceph_con_keepalive(struct ceph_connection *con)
1840{
1841 dout("con_keepalive %p\n", con);
1842 mutex_lock(&con->mutex);
1843 clear_standby(con);
1844 ceph_con_flag_set(con, CEPH_CON_F_KEEPALIVE_PENDING);
1845 mutex_unlock(&con->mutex);
1846
1847 if (!ceph_con_flag_test_and_set(con, CEPH_CON_F_WRITE_PENDING))
1848 queue_con(con);
1849}
1850EXPORT_SYMBOL(ceph_con_keepalive);
1851
1852bool ceph_con_keepalive_expired(struct ceph_connection *con,
1853 unsigned long interval)
1854{
1855 if (interval > 0 &&
1856 (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
1857 struct timespec64 now;
1858 struct timespec64 ts;
1859 ktime_get_real_ts64(&now);
1860 jiffies_to_timespec64(interval, &ts);
1861 ts = timespec64_add(con->last_keepalive_ack, ts);
1862 return timespec64_compare(&now, &ts) >= 0;
1863 }
1864 return false;
1865}
1866
1867static struct ceph_msg_data *ceph_msg_data_add(struct ceph_msg *msg)
1868{
1869 BUG_ON(msg->num_data_items >= msg->max_data_items);
1870 return &msg->data[msg->num_data_items++];
1871}
1872
1873static void ceph_msg_data_destroy(struct ceph_msg_data *data)
1874{
1875 if (data->type == CEPH_MSG_DATA_PAGES && data->own_pages) {
1876 int num_pages = calc_pages_for(data->alignment, data->length);
1877 ceph_release_page_vector(data->pages, num_pages);
1878 } else if (data->type == CEPH_MSG_DATA_PAGELIST) {
1879 ceph_pagelist_release(data->pagelist);
1880 }
1881}
1882
1883void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
1884 size_t length, size_t alignment, bool own_pages)
1885{
1886 struct ceph_msg_data *data;
1887
1888 BUG_ON(!pages);
1889 BUG_ON(!length);
1890
1891 data = ceph_msg_data_add(msg);
1892 data->type = CEPH_MSG_DATA_PAGES;
1893 data->pages = pages;
1894 data->length = length;
1895 data->alignment = alignment & ~PAGE_MASK;
1896 data->own_pages = own_pages;
1897
1898 msg->data_length += length;
1899}
1900EXPORT_SYMBOL(ceph_msg_data_add_pages);
1901
1902void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
1903 struct ceph_pagelist *pagelist)
1904{
1905 struct ceph_msg_data *data;
1906
1907 BUG_ON(!pagelist);
1908 BUG_ON(!pagelist->length);
1909
1910 data = ceph_msg_data_add(msg);
1911 data->type = CEPH_MSG_DATA_PAGELIST;
1912 refcount_inc(&pagelist->refcnt);
1913 data->pagelist = pagelist;
1914
1915 msg->data_length += pagelist->length;
1916}
1917EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
1918
1919#ifdef CONFIG_BLOCK
1920void ceph_msg_data_add_bio(struct ceph_msg *msg, struct ceph_bio_iter *bio_pos,
1921 u32 length)
1922{
1923 struct ceph_msg_data *data;
1924
1925 data = ceph_msg_data_add(msg);
1926 data->type = CEPH_MSG_DATA_BIO;
1927 data->bio_pos = *bio_pos;
1928 data->bio_length = length;
1929
1930 msg->data_length += length;
1931}
1932EXPORT_SYMBOL(ceph_msg_data_add_bio);
1933#endif /* CONFIG_BLOCK */
1934
1935void ceph_msg_data_add_bvecs(struct ceph_msg *msg,
1936 struct ceph_bvec_iter *bvec_pos)
1937{
1938 struct ceph_msg_data *data;
1939
1940 data = ceph_msg_data_add(msg);
1941 data->type = CEPH_MSG_DATA_BVECS;
1942 data->bvec_pos = *bvec_pos;
1943
1944 msg->data_length += bvec_pos->iter.bi_size;
1945}
1946EXPORT_SYMBOL(ceph_msg_data_add_bvecs);
1947
1948void ceph_msg_data_add_iter(struct ceph_msg *msg,
1949 struct iov_iter *iter)
1950{
1951 struct ceph_msg_data *data;
1952
1953 data = ceph_msg_data_add(msg);
1954 data->type = CEPH_MSG_DATA_ITER;
1955 data->iter = *iter;
1956
1957 msg->data_length += iov_iter_count(&data->iter);
1958}
1959
1960/*
1961 * construct a new message with given type, size
1962 * the new msg has a ref count of 1.
1963 */
1964struct ceph_msg *ceph_msg_new2(int type, int front_len, int max_data_items,
1965 gfp_t flags, bool can_fail)
1966{
1967 struct ceph_msg *m;
1968
1969 m = kmem_cache_zalloc(ceph_msg_cache, flags);
1970 if (m == NULL)
1971 goto out;
1972
1973 m->hdr.type = cpu_to_le16(type);
1974 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
1975 m->hdr.front_len = cpu_to_le32(front_len);
1976
1977 INIT_LIST_HEAD(&m->list_head);
1978 kref_init(&m->kref);
1979
1980 /* front */
1981 if (front_len) {
1982 m->front.iov_base = kvmalloc(front_len, flags);
1983 if (m->front.iov_base == NULL) {
1984 dout("ceph_msg_new can't allocate %d bytes\n",
1985 front_len);
1986 goto out2;
1987 }
1988 } else {
1989 m->front.iov_base = NULL;
1990 }
1991 m->front_alloc_len = m->front.iov_len = front_len;
1992
1993 if (max_data_items) {
1994 m->data = kmalloc_array(max_data_items, sizeof(*m->data),
1995 flags);
1996 if (!m->data)
1997 goto out2;
1998
1999 m->max_data_items = max_data_items;
2000 }
2001
2002 dout("ceph_msg_new %p front %d\n", m, front_len);
2003 return m;
2004
2005out2:
2006 ceph_msg_put(m);
2007out:
2008 if (!can_fail) {
2009 pr_err("msg_new can't create type %d front %d\n", type,
2010 front_len);
2011 WARN_ON(1);
2012 } else {
2013 dout("msg_new can't create type %d front %d\n", type,
2014 front_len);
2015 }
2016 return NULL;
2017}
2018EXPORT_SYMBOL(ceph_msg_new2);
2019
2020struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
2021 bool can_fail)
2022{
2023 return ceph_msg_new2(type, front_len, 0, flags, can_fail);
2024}
2025EXPORT_SYMBOL(ceph_msg_new);
2026
2027/*
2028 * Allocate "middle" portion of a message, if it is needed and wasn't
2029 * allocated by alloc_msg. This allows us to read a small fixed-size
2030 * per-type header in the front and then gracefully fail (i.e.,
2031 * propagate the error to the caller based on info in the front) when
2032 * the middle is too large.
2033 */
2034static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
2035{
2036 int type = le16_to_cpu(msg->hdr.type);
2037 int middle_len = le32_to_cpu(msg->hdr.middle_len);
2038
2039 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
2040 ceph_msg_type_name(type), middle_len);
2041 BUG_ON(!middle_len);
2042 BUG_ON(msg->middle);
2043
2044 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
2045 if (!msg->middle)
2046 return -ENOMEM;
2047 return 0;
2048}
2049
2050/*
2051 * Allocate a message for receiving an incoming message on a
2052 * connection, and save the result in con->in_msg. Uses the
2053 * connection's private alloc_msg op if available.
2054 *
2055 * Returns 0 on success, or a negative error code.
2056 *
2057 * On success, if we set *skip = 1:
2058 * - the next message should be skipped and ignored.
2059 * - con->in_msg == NULL
2060 * or if we set *skip = 0:
2061 * - con->in_msg is non-null.
2062 * On error (ENOMEM, EAGAIN, ...),
2063 * - con->in_msg == NULL
2064 */
2065int ceph_con_in_msg_alloc(struct ceph_connection *con,
2066 struct ceph_msg_header *hdr, int *skip)
2067{
2068 int middle_len = le32_to_cpu(hdr->middle_len);
2069 struct ceph_msg *msg;
2070 int ret = 0;
2071
2072 BUG_ON(con->in_msg != NULL);
2073 BUG_ON(!con->ops->alloc_msg);
2074
2075 mutex_unlock(&con->mutex);
2076 msg = con->ops->alloc_msg(con, hdr, skip);
2077 mutex_lock(&con->mutex);
2078 if (con->state != CEPH_CON_S_OPEN) {
2079 if (msg)
2080 ceph_msg_put(msg);
2081 return -EAGAIN;
2082 }
2083 if (msg) {
2084 BUG_ON(*skip);
2085 msg_con_set(msg, con);
2086 con->in_msg = msg;
2087 } else {
2088 /*
2089 * Null message pointer means either we should skip
2090 * this message or we couldn't allocate memory. The
2091 * former is not an error.
2092 */
2093 if (*skip)
2094 return 0;
2095
2096 con->error_msg = "error allocating memory for incoming message";
2097 return -ENOMEM;
2098 }
2099 memcpy(&con->in_msg->hdr, hdr, sizeof(*hdr));
2100
2101 if (middle_len && !con->in_msg->middle) {
2102 ret = ceph_alloc_middle(con, con->in_msg);
2103 if (ret < 0) {
2104 ceph_msg_put(con->in_msg);
2105 con->in_msg = NULL;
2106 }
2107 }
2108
2109 return ret;
2110}
2111
2112void ceph_con_get_out_msg(struct ceph_connection *con)
2113{
2114 struct ceph_msg *msg;
2115
2116 BUG_ON(list_empty(&con->out_queue));
2117 msg = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
2118 WARN_ON(msg->con != con);
2119
2120 /*
2121 * Put the message on "sent" list using a ref from ceph_con_send().
2122 * It is put when the message is acked or revoked.
2123 */
2124 list_move_tail(&msg->list_head, &con->out_sent);
2125
2126 /*
2127 * Only assign outgoing seq # if we haven't sent this message
2128 * yet. If it is requeued, resend with it's original seq.
2129 */
2130 if (msg->needs_out_seq) {
2131 msg->hdr.seq = cpu_to_le64(++con->out_seq);
2132 msg->needs_out_seq = false;
2133
2134 if (con->ops->reencode_message)
2135 con->ops->reencode_message(msg);
2136 }
2137
2138 /*
2139 * Get a ref for out_msg. It is put when we are done sending the
2140 * message or in case of a fault.
2141 */
2142 WARN_ON(con->out_msg);
2143 con->out_msg = ceph_msg_get(msg);
2144}
2145
2146/*
2147 * Free a generically kmalloc'd message.
2148 */
2149static void ceph_msg_free(struct ceph_msg *m)
2150{
2151 dout("%s %p\n", __func__, m);
2152 kvfree(m->front.iov_base);
2153 kfree(m->data);
2154 kmem_cache_free(ceph_msg_cache, m);
2155}
2156
2157static void ceph_msg_release(struct kref *kref)
2158{
2159 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
2160 int i;
2161
2162 dout("%s %p\n", __func__, m);
2163 WARN_ON(!list_empty(&m->list_head));
2164
2165 msg_con_set(m, NULL);
2166
2167 /* drop middle, data, if any */
2168 if (m->middle) {
2169 ceph_buffer_put(m->middle);
2170 m->middle = NULL;
2171 }
2172
2173 for (i = 0; i < m->num_data_items; i++)
2174 ceph_msg_data_destroy(&m->data[i]);
2175
2176 if (m->pool)
2177 ceph_msgpool_put(m->pool, m);
2178 else
2179 ceph_msg_free(m);
2180}
2181
2182struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
2183{
2184 dout("%s %p (was %d)\n", __func__, msg,
2185 kref_read(&msg->kref));
2186 kref_get(&msg->kref);
2187 return msg;
2188}
2189EXPORT_SYMBOL(ceph_msg_get);
2190
2191void ceph_msg_put(struct ceph_msg *msg)
2192{
2193 dout("%s %p (was %d)\n", __func__, msg,
2194 kref_read(&msg->kref));
2195 kref_put(&msg->kref, ceph_msg_release);
2196}
2197EXPORT_SYMBOL(ceph_msg_put);
2198
2199void ceph_msg_dump(struct ceph_msg *msg)
2200{
2201 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
2202 msg->front_alloc_len, msg->data_length);
2203 print_hex_dump(KERN_DEBUG, "header: ",
2204 DUMP_PREFIX_OFFSET, 16, 1,
2205 &msg->hdr, sizeof(msg->hdr), true);
2206 print_hex_dump(KERN_DEBUG, " front: ",
2207 DUMP_PREFIX_OFFSET, 16, 1,
2208 msg->front.iov_base, msg->front.iov_len, true);
2209 if (msg->middle)
2210 print_hex_dump(KERN_DEBUG, "middle: ",
2211 DUMP_PREFIX_OFFSET, 16, 1,
2212 msg->middle->vec.iov_base,
2213 msg->middle->vec.iov_len, true);
2214 print_hex_dump(KERN_DEBUG, "footer: ",
2215 DUMP_PREFIX_OFFSET, 16, 1,
2216 &msg->footer, sizeof(msg->footer), true);
2217}
2218EXPORT_SYMBOL(ceph_msg_dump);