1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
3
4 #include <linux/crc32c.h>
5 #include <linux/ctype.h>
6 #include <linux/highmem.h>
7 #include <linux/inet.h>
8 #include <linux/kthread.h>
9 #include <linux/net.h>
10 #include <linux/nsproxy.h>
11 #include <linux/sched/mm.h>
12 #include <linux/slab.h>
13 #include <linux/socket.h>
14 #include <linux/string.h>
15 #ifdef CONFIG_BLOCK
16 #include <linux/bio.h>
17 #endif /* CONFIG_BLOCK */
18 #include <linux/dns_resolver.h>
19 #include <net/tcp.h>
20
21 #include <linux/ceph/ceph_features.h>
22 #include <linux/ceph/libceph.h>
23 #include <linux/ceph/messenger.h>
24 #include <linux/ceph/decode.h>
25 #include <linux/ceph/pagelist.h>
26 #include <linux/export.h>
27
28 /*
29 * Ceph uses the messenger to exchange ceph_msg messages with other
30 * hosts in the system. The messenger provides ordered and reliable
31 * delivery. We tolerate TCP disconnects by reconnecting (with
32 * exponential backoff) in the case of a fault (disconnection, bad
33 * crc, protocol error). Acks allow sent messages to be discarded by
34 * the sender.
35 */
36
37 /*
38 * We track the state of the socket on a given connection using
39 * values defined below. The transition to a new socket state is
40 * handled by a function which verifies we aren't coming from an
41 * unexpected state.
42 *
43 * --------
44 * | NEW* | transient initial state
45 * --------
46 * | con_sock_state_init()
47 * v
48 * ----------
49 * | CLOSED | initialized, but no socket (and no
50 * ---------- TCP connection)
51 * ^ \
52 * | \ con_sock_state_connecting()
53 * | ----------------------
54 * | \
55 * + con_sock_state_closed() \
56 * |+--------------------------- \
57 * | \ \ \
58 * | ----------- \ \
59 * | | CLOSING | socket event; \ \
60 * | ----------- await close \ \
61 * | ^ \ |
62 * | | \ |
63 * | + con_sock_state_closing() \ |
64 * | / \ | |
65 * | / --------------- | |
66 * | / \ v v
67 * | / --------------
68 * | / -----------------| CONNECTING | socket created, TCP
69 * | | / -------------- connect initiated
70 * | | | con_sock_state_connected()
71 * | | v
72 * -------------
73 * | CONNECTED | TCP connection established
74 * -------------
75 *
76 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
77 */
78
79 #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
80 #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
81 #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
82 #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
83 #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
84
85 /*
86 * connection states
87 */
88 #define CON_STATE_CLOSED 1 /* -> PREOPEN */
89 #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
90 #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
91 #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
92 #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
93 #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
94
95 /*
96 * ceph_connection flag bits
97 */
98 #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
99 * messages on errors */
100 #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
101 #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
102 #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
103 #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
104
con_flag_valid(unsigned long con_flag)105 static bool con_flag_valid(unsigned long con_flag)
106 {
107 switch (con_flag) {
108 case CON_FLAG_LOSSYTX:
109 case CON_FLAG_KEEPALIVE_PENDING:
110 case CON_FLAG_WRITE_PENDING:
111 case CON_FLAG_SOCK_CLOSED:
112 case CON_FLAG_BACKOFF:
113 return true;
114 default:
115 return false;
116 }
117 }
118
con_flag_clear(struct ceph_connection * con,unsigned long con_flag)119 static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
120 {
121 BUG_ON(!con_flag_valid(con_flag));
122
123 clear_bit(con_flag, &con->flags);
124 }
125
con_flag_set(struct ceph_connection * con,unsigned long con_flag)126 static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
127 {
128 BUG_ON(!con_flag_valid(con_flag));
129
130 set_bit(con_flag, &con->flags);
131 }
132
con_flag_test(struct ceph_connection * con,unsigned long con_flag)133 static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
134 {
135 BUG_ON(!con_flag_valid(con_flag));
136
137 return test_bit(con_flag, &con->flags);
138 }
139
con_flag_test_and_clear(struct ceph_connection * con,unsigned long con_flag)140 static bool con_flag_test_and_clear(struct ceph_connection *con,
141 unsigned long con_flag)
142 {
143 BUG_ON(!con_flag_valid(con_flag));
144
145 return test_and_clear_bit(con_flag, &con->flags);
146 }
147
con_flag_test_and_set(struct ceph_connection * con,unsigned long con_flag)148 static bool con_flag_test_and_set(struct ceph_connection *con,
149 unsigned long con_flag)
150 {
151 BUG_ON(!con_flag_valid(con_flag));
152
153 return test_and_set_bit(con_flag, &con->flags);
154 }
155
156 /* Slab caches for frequently-allocated structures */
157
158 static struct kmem_cache *ceph_msg_cache;
159
160 /* static tag bytes (protocol control messages) */
161 static char tag_msg = CEPH_MSGR_TAG_MSG;
162 static char tag_ack = CEPH_MSGR_TAG_ACK;
163 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
164 static char tag_keepalive2 = CEPH_MSGR_TAG_KEEPALIVE2;
165
166 #ifdef CONFIG_LOCKDEP
167 static struct lock_class_key socket_class;
168 #endif
169
170 static void queue_con(struct ceph_connection *con);
171 static void cancel_con(struct ceph_connection *con);
172 static void ceph_con_workfn(struct work_struct *);
173 static void con_fault(struct ceph_connection *con);
174
175 /*
176 * Nicely render a sockaddr as a string. An array of formatted
177 * strings is used, to approximate reentrancy.
178 */
179 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
180 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
181 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
182 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
183
184 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
185 static atomic_t addr_str_seq = ATOMIC_INIT(0);
186
187 static struct page *zero_page; /* used in certain error cases */
188
ceph_pr_addr(const struct ceph_entity_addr * addr)189 const char *ceph_pr_addr(const struct ceph_entity_addr *addr)
190 {
191 int i;
192 char *s;
193 struct sockaddr_storage ss = addr->in_addr; /* align */
194 struct sockaddr_in *in4 = (struct sockaddr_in *)&ss;
195 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *)&ss;
196
197 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
198 s = addr_str[i];
199
200 switch (ss.ss_family) {
201 case AF_INET:
202 snprintf(s, MAX_ADDR_STR_LEN, "(%d)%pI4:%hu",
203 le32_to_cpu(addr->type), &in4->sin_addr,
204 ntohs(in4->sin_port));
205 break;
206
207 case AF_INET6:
208 snprintf(s, MAX_ADDR_STR_LEN, "(%d)[%pI6c]:%hu",
209 le32_to_cpu(addr->type), &in6->sin6_addr,
210 ntohs(in6->sin6_port));
211 break;
212
213 default:
214 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
215 ss.ss_family);
216 }
217
218 return s;
219 }
220 EXPORT_SYMBOL(ceph_pr_addr);
221
encode_my_addr(struct ceph_messenger * msgr)222 static void encode_my_addr(struct ceph_messenger *msgr)
223 {
224 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
225 ceph_encode_banner_addr(&msgr->my_enc_addr);
226 }
227
228 /*
229 * work queue for all reading and writing to/from the socket.
230 */
231 static struct workqueue_struct *ceph_msgr_wq;
232
ceph_msgr_slab_init(void)233 static int ceph_msgr_slab_init(void)
234 {
235 BUG_ON(ceph_msg_cache);
236 ceph_msg_cache = KMEM_CACHE(ceph_msg, 0);
237 if (!ceph_msg_cache)
238 return -ENOMEM;
239
240 return 0;
241 }
242
ceph_msgr_slab_exit(void)243 static void ceph_msgr_slab_exit(void)
244 {
245 BUG_ON(!ceph_msg_cache);
246 kmem_cache_destroy(ceph_msg_cache);
247 ceph_msg_cache = NULL;
248 }
249
_ceph_msgr_exit(void)250 static void _ceph_msgr_exit(void)
251 {
252 if (ceph_msgr_wq) {
253 destroy_workqueue(ceph_msgr_wq);
254 ceph_msgr_wq = NULL;
255 }
256
257 BUG_ON(zero_page == NULL);
258 put_page(zero_page);
259 zero_page = NULL;
260
261 ceph_msgr_slab_exit();
262 }
263
ceph_msgr_init(void)264 int __init ceph_msgr_init(void)
265 {
266 if (ceph_msgr_slab_init())
267 return -ENOMEM;
268
269 BUG_ON(zero_page != NULL);
270 zero_page = ZERO_PAGE(0);
271 get_page(zero_page);
272
273 /*
274 * The number of active work items is limited by the number of
275 * connections, so leave @max_active at default.
276 */
277 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
278 if (ceph_msgr_wq)
279 return 0;
280
281 pr_err("msgr_init failed to create workqueue\n");
282 _ceph_msgr_exit();
283
284 return -ENOMEM;
285 }
286
ceph_msgr_exit(void)287 void ceph_msgr_exit(void)
288 {
289 BUG_ON(ceph_msgr_wq == NULL);
290
291 _ceph_msgr_exit();
292 }
293
ceph_msgr_flush(void)294 void ceph_msgr_flush(void)
295 {
296 flush_workqueue(ceph_msgr_wq);
297 }
298 EXPORT_SYMBOL(ceph_msgr_flush);
299
300 /* Connection socket state transition functions */
301
con_sock_state_init(struct ceph_connection * con)302 static void con_sock_state_init(struct ceph_connection *con)
303 {
304 int old_state;
305
306 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
307 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
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_CLOSED);
311 }
312
con_sock_state_connecting(struct ceph_connection * con)313 static void con_sock_state_connecting(struct ceph_connection *con)
314 {
315 int old_state;
316
317 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
318 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
319 printk("%s: unexpected old state %d\n", __func__, old_state);
320 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
321 CON_SOCK_STATE_CONNECTING);
322 }
323
con_sock_state_connected(struct ceph_connection * con)324 static void con_sock_state_connected(struct ceph_connection *con)
325 {
326 int old_state;
327
328 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
329 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
330 printk("%s: unexpected old state %d\n", __func__, old_state);
331 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
332 CON_SOCK_STATE_CONNECTED);
333 }
334
con_sock_state_closing(struct ceph_connection * con)335 static void con_sock_state_closing(struct ceph_connection *con)
336 {
337 int old_state;
338
339 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
340 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
341 old_state != CON_SOCK_STATE_CONNECTED &&
342 old_state != CON_SOCK_STATE_CLOSING))
343 printk("%s: unexpected old state %d\n", __func__, old_state);
344 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
345 CON_SOCK_STATE_CLOSING);
346 }
347
con_sock_state_closed(struct ceph_connection * con)348 static void con_sock_state_closed(struct ceph_connection *con)
349 {
350 int old_state;
351
352 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
353 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
354 old_state != CON_SOCK_STATE_CLOSING &&
355 old_state != CON_SOCK_STATE_CONNECTING &&
356 old_state != CON_SOCK_STATE_CLOSED))
357 printk("%s: unexpected old state %d\n", __func__, old_state);
358 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
359 CON_SOCK_STATE_CLOSED);
360 }
361
362 /*
363 * socket callback functions
364 */
365
366 /* data available on socket, or listen socket received a connect */
ceph_sock_data_ready(struct sock * sk)367 static void ceph_sock_data_ready(struct sock *sk)
368 {
369 struct ceph_connection *con = sk->sk_user_data;
370 if (atomic_read(&con->msgr->stopping)) {
371 return;
372 }
373
374 if (sk->sk_state != TCP_CLOSE_WAIT) {
375 dout("%s on %p state = %lu, queueing work\n", __func__,
376 con, con->state);
377 queue_con(con);
378 }
379 }
380
381 /* socket has buffer space for writing */
ceph_sock_write_space(struct sock * sk)382 static void ceph_sock_write_space(struct sock *sk)
383 {
384 struct ceph_connection *con = sk->sk_user_data;
385
386 /* only queue to workqueue if there is data we want to write,
387 * and there is sufficient space in the socket buffer to accept
388 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
389 * doesn't get called again until try_write() fills the socket
390 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
391 * and net/core/stream.c:sk_stream_write_space().
392 */
393 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
394 if (sk_stream_is_writeable(sk)) {
395 dout("%s %p queueing write work\n", __func__, con);
396 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
397 queue_con(con);
398 }
399 } else {
400 dout("%s %p nothing to write\n", __func__, con);
401 }
402 }
403
404 /* socket's state has changed */
ceph_sock_state_change(struct sock * sk)405 static void ceph_sock_state_change(struct sock *sk)
406 {
407 struct ceph_connection *con = sk->sk_user_data;
408
409 dout("%s %p state = %lu sk_state = %u\n", __func__,
410 con, con->state, sk->sk_state);
411
412 switch (sk->sk_state) {
413 case TCP_CLOSE:
414 dout("%s TCP_CLOSE\n", __func__);
415 /* fall through */
416 case TCP_CLOSE_WAIT:
417 dout("%s TCP_CLOSE_WAIT\n", __func__);
418 con_sock_state_closing(con);
419 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
420 queue_con(con);
421 break;
422 case TCP_ESTABLISHED:
423 dout("%s TCP_ESTABLISHED\n", __func__);
424 con_sock_state_connected(con);
425 queue_con(con);
426 break;
427 default: /* Everything else is uninteresting */
428 break;
429 }
430 }
431
432 /*
433 * set up socket callbacks
434 */
set_sock_callbacks(struct socket * sock,struct ceph_connection * con)435 static void set_sock_callbacks(struct socket *sock,
436 struct ceph_connection *con)
437 {
438 struct sock *sk = sock->sk;
439 sk->sk_user_data = con;
440 sk->sk_data_ready = ceph_sock_data_ready;
441 sk->sk_write_space = ceph_sock_write_space;
442 sk->sk_state_change = ceph_sock_state_change;
443 }
444
445
446 /*
447 * socket helpers
448 */
449
450 /*
451 * initiate connection to a remote socket.
452 */
ceph_tcp_connect(struct ceph_connection * con)453 static int ceph_tcp_connect(struct ceph_connection *con)
454 {
455 struct sockaddr_storage ss = con->peer_addr.in_addr; /* align */
456 struct socket *sock;
457 unsigned int noio_flag;
458 int ret;
459
460 BUG_ON(con->sock);
461
462 /* sock_create_kern() allocates with GFP_KERNEL */
463 noio_flag = memalloc_noio_save();
464 ret = sock_create_kern(read_pnet(&con->msgr->net), ss.ss_family,
465 SOCK_STREAM, IPPROTO_TCP, &sock);
466 memalloc_noio_restore(noio_flag);
467 if (ret)
468 return ret;
469 sock->sk->sk_allocation = GFP_NOFS;
470
471 #ifdef CONFIG_LOCKDEP
472 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
473 #endif
474
475 set_sock_callbacks(sock, con);
476
477 dout("connect %s\n", ceph_pr_addr(&con->peer_addr));
478
479 con_sock_state_connecting(con);
480 ret = sock->ops->connect(sock, (struct sockaddr *)&ss, sizeof(ss),
481 O_NONBLOCK);
482 if (ret == -EINPROGRESS) {
483 dout("connect %s EINPROGRESS sk_state = %u\n",
484 ceph_pr_addr(&con->peer_addr),
485 sock->sk->sk_state);
486 } else if (ret < 0) {
487 pr_err("connect %s error %d\n",
488 ceph_pr_addr(&con->peer_addr), ret);
489 sock_release(sock);
490 return ret;
491 }
492
493 if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY)) {
494 int optval = 1;
495
496 ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
497 (char *)&optval, sizeof(optval));
498 if (ret)
499 pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
500 ret);
501 }
502
503 con->sock = sock;
504 return 0;
505 }
506
507 /*
508 * If @buf is NULL, discard up to @len bytes.
509 */
ceph_tcp_recvmsg(struct socket * sock,void * buf,size_t len)510 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
511 {
512 struct kvec iov = {buf, len};
513 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
514 int r;
515
516 if (!buf)
517 msg.msg_flags |= MSG_TRUNC;
518
519 iov_iter_kvec(&msg.msg_iter, READ, &iov, 1, len);
520 r = sock_recvmsg(sock, &msg, msg.msg_flags);
521 if (r == -EAGAIN)
522 r = 0;
523 return r;
524 }
525
ceph_tcp_recvpage(struct socket * sock,struct page * page,int page_offset,size_t length)526 static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
527 int page_offset, size_t length)
528 {
529 struct bio_vec bvec = {
530 .bv_page = page,
531 .bv_offset = page_offset,
532 .bv_len = length
533 };
534 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
535 int r;
536
537 BUG_ON(page_offset + length > PAGE_SIZE);
538 iov_iter_bvec(&msg.msg_iter, READ, &bvec, 1, length);
539 r = sock_recvmsg(sock, &msg, msg.msg_flags);
540 if (r == -EAGAIN)
541 r = 0;
542 return r;
543 }
544
545 /*
546 * write something. @more is true if caller will be sending more data
547 * shortly.
548 */
ceph_tcp_sendmsg(struct socket * sock,struct kvec * iov,size_t kvlen,size_t len,bool more)549 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
550 size_t kvlen, size_t len, bool 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
566 /*
567 * @more: either or both of MSG_MORE and MSG_SENDPAGE_NOTLAST
568 */
ceph_tcp_sendpage(struct socket * sock,struct page * page,int offset,size_t size,int more)569 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
570 int offset, size_t size, int more)
571 {
572 ssize_t (*sendpage)(struct socket *sock, struct page *page,
573 int offset, size_t size, int flags);
574 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | more;
575 int ret;
576
577 /*
578 * sendpage cannot properly handle pages with page_count == 0,
579 * we need to fall back to sendmsg if that's the case.
580 *
581 * Same goes for slab pages: skb_can_coalesce() allows
582 * coalescing neighboring slab objects into a single frag which
583 * triggers one of hardened usercopy checks.
584 */
585 if (page_count(page) >= 1 && !PageSlab(page))
586 sendpage = sock->ops->sendpage;
587 else
588 sendpage = sock_no_sendpage;
589
590 ret = sendpage(sock, page, offset, size, flags);
591 if (ret == -EAGAIN)
592 ret = 0;
593
594 return ret;
595 }
596
597 /*
598 * Shutdown/close the socket for the given connection.
599 */
con_close_socket(struct ceph_connection * con)600 static int con_close_socket(struct ceph_connection *con)
601 {
602 int rc = 0;
603
604 dout("con_close_socket on %p sock %p\n", con, con->sock);
605 if (con->sock) {
606 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
607 sock_release(con->sock);
608 con->sock = NULL;
609 }
610
611 /*
612 * Forcibly clear the SOCK_CLOSED flag. It gets set
613 * independent of the connection mutex, and we could have
614 * received a socket close event before we had the chance to
615 * shut the socket down.
616 */
617 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
618
619 con_sock_state_closed(con);
620 return rc;
621 }
622
623 /*
624 * Reset a connection. Discard all incoming and outgoing messages
625 * and clear *_seq state.
626 */
ceph_msg_remove(struct ceph_msg * msg)627 static void ceph_msg_remove(struct ceph_msg *msg)
628 {
629 list_del_init(&msg->list_head);
630
631 ceph_msg_put(msg);
632 }
ceph_msg_remove_list(struct list_head * head)633 static void ceph_msg_remove_list(struct list_head *head)
634 {
635 while (!list_empty(head)) {
636 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
637 list_head);
638 ceph_msg_remove(msg);
639 }
640 }
641
reset_connection(struct ceph_connection * con)642 static void reset_connection(struct ceph_connection *con)
643 {
644 /* reset connection, out_queue, msg_ and connect_seq */
645 /* discard existing out_queue and msg_seq */
646 dout("reset_connection %p\n", con);
647 ceph_msg_remove_list(&con->out_queue);
648 ceph_msg_remove_list(&con->out_sent);
649
650 if (con->in_msg) {
651 BUG_ON(con->in_msg->con != con);
652 ceph_msg_put(con->in_msg);
653 con->in_msg = NULL;
654 }
655
656 con->connect_seq = 0;
657 con->out_seq = 0;
658 if (con->out_msg) {
659 BUG_ON(con->out_msg->con != con);
660 ceph_msg_put(con->out_msg);
661 con->out_msg = NULL;
662 }
663 con->in_seq = 0;
664 con->in_seq_acked = 0;
665
666 con->out_skip = 0;
667 }
668
669 /*
670 * mark a peer down. drop any open connections.
671 */
ceph_con_close(struct ceph_connection * con)672 void ceph_con_close(struct ceph_connection *con)
673 {
674 mutex_lock(&con->mutex);
675 dout("con_close %p peer %s\n", con, ceph_pr_addr(&con->peer_addr));
676 con->state = CON_STATE_CLOSED;
677
678 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
679 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
680 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
681 con_flag_clear(con, CON_FLAG_BACKOFF);
682
683 reset_connection(con);
684 con->peer_global_seq = 0;
685 cancel_con(con);
686 con_close_socket(con);
687 mutex_unlock(&con->mutex);
688 }
689 EXPORT_SYMBOL(ceph_con_close);
690
691 /*
692 * Reopen a closed connection, with a new peer address.
693 */
ceph_con_open(struct ceph_connection * con,__u8 entity_type,__u64 entity_num,struct ceph_entity_addr * addr)694 void ceph_con_open(struct ceph_connection *con,
695 __u8 entity_type, __u64 entity_num,
696 struct ceph_entity_addr *addr)
697 {
698 mutex_lock(&con->mutex);
699 dout("con_open %p %s\n", con, ceph_pr_addr(addr));
700
701 WARN_ON(con->state != CON_STATE_CLOSED);
702 con->state = CON_STATE_PREOPEN;
703
704 con->peer_name.type = (__u8) entity_type;
705 con->peer_name.num = cpu_to_le64(entity_num);
706
707 memcpy(&con->peer_addr, addr, sizeof(*addr));
708 con->delay = 0; /* reset backoff memory */
709 mutex_unlock(&con->mutex);
710 queue_con(con);
711 }
712 EXPORT_SYMBOL(ceph_con_open);
713
714 /*
715 * return true if this connection ever successfully opened
716 */
ceph_con_opened(struct ceph_connection * con)717 bool ceph_con_opened(struct ceph_connection *con)
718 {
719 return con->connect_seq > 0;
720 }
721
722 /*
723 * initialize a new connection.
724 */
ceph_con_init(struct ceph_connection * con,void * private,const struct ceph_connection_operations * ops,struct ceph_messenger * msgr)725 void ceph_con_init(struct ceph_connection *con, void *private,
726 const struct ceph_connection_operations *ops,
727 struct ceph_messenger *msgr)
728 {
729 dout("con_init %p\n", con);
730 memset(con, 0, sizeof(*con));
731 con->private = private;
732 con->ops = ops;
733 con->msgr = msgr;
734
735 con_sock_state_init(con);
736
737 mutex_init(&con->mutex);
738 INIT_LIST_HEAD(&con->out_queue);
739 INIT_LIST_HEAD(&con->out_sent);
740 INIT_DELAYED_WORK(&con->work, ceph_con_workfn);
741
742 con->state = CON_STATE_CLOSED;
743 }
744 EXPORT_SYMBOL(ceph_con_init);
745
746
747 /*
748 * We maintain a global counter to order connection attempts. Get
749 * a unique seq greater than @gt.
750 */
get_global_seq(struct ceph_messenger * msgr,u32 gt)751 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
752 {
753 u32 ret;
754
755 spin_lock(&msgr->global_seq_lock);
756 if (msgr->global_seq < gt)
757 msgr->global_seq = gt;
758 ret = ++msgr->global_seq;
759 spin_unlock(&msgr->global_seq_lock);
760 return ret;
761 }
762
con_out_kvec_reset(struct ceph_connection * con)763 static void con_out_kvec_reset(struct ceph_connection *con)
764 {
765 BUG_ON(con->out_skip);
766
767 con->out_kvec_left = 0;
768 con->out_kvec_bytes = 0;
769 con->out_kvec_cur = &con->out_kvec[0];
770 }
771
con_out_kvec_add(struct ceph_connection * con,size_t size,void * data)772 static void con_out_kvec_add(struct ceph_connection *con,
773 size_t size, void *data)
774 {
775 int index = con->out_kvec_left;
776
777 BUG_ON(con->out_skip);
778 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
779
780 con->out_kvec[index].iov_len = size;
781 con->out_kvec[index].iov_base = data;
782 con->out_kvec_left++;
783 con->out_kvec_bytes += size;
784 }
785
786 /*
787 * Chop off a kvec from the end. Return residual number of bytes for
788 * that kvec, i.e. how many bytes would have been written if the kvec
789 * hadn't been nuked.
790 */
con_out_kvec_skip(struct ceph_connection * con)791 static int con_out_kvec_skip(struct ceph_connection *con)
792 {
793 int off = con->out_kvec_cur - con->out_kvec;
794 int skip = 0;
795
796 if (con->out_kvec_bytes > 0) {
797 skip = con->out_kvec[off + con->out_kvec_left - 1].iov_len;
798 BUG_ON(con->out_kvec_bytes < skip);
799 BUG_ON(!con->out_kvec_left);
800 con->out_kvec_bytes -= skip;
801 con->out_kvec_left--;
802 }
803
804 return skip;
805 }
806
807 #ifdef CONFIG_BLOCK
808
809 /*
810 * For a bio data item, a piece is whatever remains of the next
811 * entry in the current bio iovec, or the first entry in the next
812 * bio in the list.
813 */
ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor * cursor,size_t length)814 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
815 size_t length)
816 {
817 struct ceph_msg_data *data = cursor->data;
818 struct ceph_bio_iter *it = &cursor->bio_iter;
819
820 cursor->resid = min_t(size_t, length, data->bio_length);
821 *it = data->bio_pos;
822 if (cursor->resid < it->iter.bi_size)
823 it->iter.bi_size = cursor->resid;
824
825 BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
826 cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter);
827 }
828
ceph_msg_data_bio_next(struct ceph_msg_data_cursor * cursor,size_t * page_offset,size_t * length)829 static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
830 size_t *page_offset,
831 size_t *length)
832 {
833 struct bio_vec bv = bio_iter_iovec(cursor->bio_iter.bio,
834 cursor->bio_iter.iter);
835
836 *page_offset = bv.bv_offset;
837 *length = bv.bv_len;
838 return bv.bv_page;
839 }
840
ceph_msg_data_bio_advance(struct ceph_msg_data_cursor * cursor,size_t bytes)841 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
842 size_t bytes)
843 {
844 struct ceph_bio_iter *it = &cursor->bio_iter;
845 struct page *page = bio_iter_page(it->bio, it->iter);
846
847 BUG_ON(bytes > cursor->resid);
848 BUG_ON(bytes > bio_iter_len(it->bio, it->iter));
849 cursor->resid -= bytes;
850 bio_advance_iter(it->bio, &it->iter, bytes);
851
852 if (!cursor->resid) {
853 BUG_ON(!cursor->last_piece);
854 return false; /* no more data */
855 }
856
857 if (!bytes || (it->iter.bi_size && it->iter.bi_bvec_done &&
858 page == bio_iter_page(it->bio, it->iter)))
859 return false; /* more bytes to process in this segment */
860
861 if (!it->iter.bi_size) {
862 it->bio = it->bio->bi_next;
863 it->iter = it->bio->bi_iter;
864 if (cursor->resid < it->iter.bi_size)
865 it->iter.bi_size = cursor->resid;
866 }
867
868 BUG_ON(cursor->last_piece);
869 BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
870 cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter);
871 return true;
872 }
873 #endif /* CONFIG_BLOCK */
874
ceph_msg_data_bvecs_cursor_init(struct ceph_msg_data_cursor * cursor,size_t length)875 static void ceph_msg_data_bvecs_cursor_init(struct ceph_msg_data_cursor *cursor,
876 size_t length)
877 {
878 struct ceph_msg_data *data = cursor->data;
879 struct bio_vec *bvecs = data->bvec_pos.bvecs;
880
881 cursor->resid = min_t(size_t, length, data->bvec_pos.iter.bi_size);
882 cursor->bvec_iter = data->bvec_pos.iter;
883 cursor->bvec_iter.bi_size = cursor->resid;
884
885 BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
886 cursor->last_piece =
887 cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter);
888 }
889
ceph_msg_data_bvecs_next(struct ceph_msg_data_cursor * cursor,size_t * page_offset,size_t * length)890 static struct page *ceph_msg_data_bvecs_next(struct ceph_msg_data_cursor *cursor,
891 size_t *page_offset,
892 size_t *length)
893 {
894 struct bio_vec bv = bvec_iter_bvec(cursor->data->bvec_pos.bvecs,
895 cursor->bvec_iter);
896
897 *page_offset = bv.bv_offset;
898 *length = bv.bv_len;
899 return bv.bv_page;
900 }
901
ceph_msg_data_bvecs_advance(struct ceph_msg_data_cursor * cursor,size_t bytes)902 static bool ceph_msg_data_bvecs_advance(struct ceph_msg_data_cursor *cursor,
903 size_t bytes)
904 {
905 struct bio_vec *bvecs = cursor->data->bvec_pos.bvecs;
906 struct page *page = bvec_iter_page(bvecs, cursor->bvec_iter);
907
908 BUG_ON(bytes > cursor->resid);
909 BUG_ON(bytes > bvec_iter_len(bvecs, cursor->bvec_iter));
910 cursor->resid -= bytes;
911 bvec_iter_advance(bvecs, &cursor->bvec_iter, bytes);
912
913 if (!cursor->resid) {
914 BUG_ON(!cursor->last_piece);
915 return false; /* no more data */
916 }
917
918 if (!bytes || (cursor->bvec_iter.bi_bvec_done &&
919 page == bvec_iter_page(bvecs, cursor->bvec_iter)))
920 return false; /* more bytes to process in this segment */
921
922 BUG_ON(cursor->last_piece);
923 BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
924 cursor->last_piece =
925 cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter);
926 return true;
927 }
928
929 /*
930 * For a page array, a piece comes from the first page in the array
931 * that has not already been fully consumed.
932 */
ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor * cursor,size_t length)933 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
934 size_t length)
935 {
936 struct ceph_msg_data *data = cursor->data;
937 int page_count;
938
939 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
940
941 BUG_ON(!data->pages);
942 BUG_ON(!data->length);
943
944 cursor->resid = min(length, data->length);
945 page_count = calc_pages_for(data->alignment, (u64)data->length);
946 cursor->page_offset = data->alignment & ~PAGE_MASK;
947 cursor->page_index = 0;
948 BUG_ON(page_count > (int)USHRT_MAX);
949 cursor->page_count = (unsigned short)page_count;
950 BUG_ON(length > SIZE_MAX - cursor->page_offset);
951 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
952 }
953
954 static struct page *
ceph_msg_data_pages_next(struct ceph_msg_data_cursor * cursor,size_t * page_offset,size_t * length)955 ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
956 size_t *page_offset, size_t *length)
957 {
958 struct ceph_msg_data *data = cursor->data;
959
960 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
961
962 BUG_ON(cursor->page_index >= cursor->page_count);
963 BUG_ON(cursor->page_offset >= PAGE_SIZE);
964
965 *page_offset = cursor->page_offset;
966 if (cursor->last_piece)
967 *length = cursor->resid;
968 else
969 *length = PAGE_SIZE - *page_offset;
970
971 return data->pages[cursor->page_index];
972 }
973
ceph_msg_data_pages_advance(struct ceph_msg_data_cursor * cursor,size_t bytes)974 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
975 size_t bytes)
976 {
977 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
978
979 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
980
981 /* Advance the cursor page offset */
982
983 cursor->resid -= bytes;
984 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
985 if (!bytes || cursor->page_offset)
986 return false; /* more bytes to process in the current page */
987
988 if (!cursor->resid)
989 return false; /* no more data */
990
991 /* Move on to the next page; offset is already at 0 */
992
993 BUG_ON(cursor->page_index >= cursor->page_count);
994 cursor->page_index++;
995 cursor->last_piece = cursor->resid <= PAGE_SIZE;
996
997 return true;
998 }
999
1000 /*
1001 * For a pagelist, a piece is whatever remains to be consumed in the
1002 * first page in the list, or the front of the next page.
1003 */
1004 static void
ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor * cursor,size_t length)1005 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
1006 size_t length)
1007 {
1008 struct ceph_msg_data *data = cursor->data;
1009 struct ceph_pagelist *pagelist;
1010 struct page *page;
1011
1012 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1013
1014 pagelist = data->pagelist;
1015 BUG_ON(!pagelist);
1016
1017 if (!length)
1018 return; /* pagelist can be assigned but empty */
1019
1020 BUG_ON(list_empty(&pagelist->head));
1021 page = list_first_entry(&pagelist->head, struct page, lru);
1022
1023 cursor->resid = min(length, pagelist->length);
1024 cursor->page = page;
1025 cursor->offset = 0;
1026 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1027 }
1028
1029 static struct page *
ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor * cursor,size_t * page_offset,size_t * length)1030 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
1031 size_t *page_offset, size_t *length)
1032 {
1033 struct ceph_msg_data *data = cursor->data;
1034 struct ceph_pagelist *pagelist;
1035
1036 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1037
1038 pagelist = data->pagelist;
1039 BUG_ON(!pagelist);
1040
1041 BUG_ON(!cursor->page);
1042 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1043
1044 /* offset of first page in pagelist is always 0 */
1045 *page_offset = cursor->offset & ~PAGE_MASK;
1046 if (cursor->last_piece)
1047 *length = cursor->resid;
1048 else
1049 *length = PAGE_SIZE - *page_offset;
1050
1051 return cursor->page;
1052 }
1053
ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor * cursor,size_t bytes)1054 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
1055 size_t bytes)
1056 {
1057 struct ceph_msg_data *data = cursor->data;
1058 struct ceph_pagelist *pagelist;
1059
1060 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1061
1062 pagelist = data->pagelist;
1063 BUG_ON(!pagelist);
1064
1065 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1066 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1067
1068 /* Advance the cursor offset */
1069
1070 cursor->resid -= bytes;
1071 cursor->offset += bytes;
1072 /* offset of first page in pagelist is always 0 */
1073 if (!bytes || cursor->offset & ~PAGE_MASK)
1074 return false; /* more bytes to process in the current page */
1075
1076 if (!cursor->resid)
1077 return false; /* no more data */
1078
1079 /* Move on to the next page */
1080
1081 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1082 cursor->page = list_next_entry(cursor->page, lru);
1083 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1084
1085 return true;
1086 }
1087
1088 /*
1089 * Message data is handled (sent or received) in pieces, where each
1090 * piece resides on a single page. The network layer might not
1091 * consume an entire piece at once. A data item's cursor keeps
1092 * track of which piece is next to process and how much remains to
1093 * be processed in that piece. It also tracks whether the current
1094 * piece is the last one in the data item.
1095 */
__ceph_msg_data_cursor_init(struct ceph_msg_data_cursor * cursor)1096 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1097 {
1098 size_t length = cursor->total_resid;
1099
1100 switch (cursor->data->type) {
1101 case CEPH_MSG_DATA_PAGELIST:
1102 ceph_msg_data_pagelist_cursor_init(cursor, length);
1103 break;
1104 case CEPH_MSG_DATA_PAGES:
1105 ceph_msg_data_pages_cursor_init(cursor, length);
1106 break;
1107 #ifdef CONFIG_BLOCK
1108 case CEPH_MSG_DATA_BIO:
1109 ceph_msg_data_bio_cursor_init(cursor, length);
1110 break;
1111 #endif /* CONFIG_BLOCK */
1112 case CEPH_MSG_DATA_BVECS:
1113 ceph_msg_data_bvecs_cursor_init(cursor, length);
1114 break;
1115 case CEPH_MSG_DATA_NONE:
1116 default:
1117 /* BUG(); */
1118 break;
1119 }
1120 cursor->need_crc = true;
1121 }
1122
ceph_msg_data_cursor_init(struct ceph_msg * msg,size_t length)1123 static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1124 {
1125 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1126
1127 BUG_ON(!length);
1128 BUG_ON(length > msg->data_length);
1129 BUG_ON(!msg->num_data_items);
1130
1131 cursor->total_resid = length;
1132 cursor->data = msg->data;
1133
1134 __ceph_msg_data_cursor_init(cursor);
1135 }
1136
1137 /*
1138 * Return the page containing the next piece to process for a given
1139 * data item, and supply the page offset and length of that piece.
1140 * Indicate whether this is the last piece in this data item.
1141 */
ceph_msg_data_next(struct ceph_msg_data_cursor * cursor,size_t * page_offset,size_t * length,bool * last_piece)1142 static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1143 size_t *page_offset, size_t *length,
1144 bool *last_piece)
1145 {
1146 struct page *page;
1147
1148 switch (cursor->data->type) {
1149 case CEPH_MSG_DATA_PAGELIST:
1150 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1151 break;
1152 case CEPH_MSG_DATA_PAGES:
1153 page = ceph_msg_data_pages_next(cursor, page_offset, length);
1154 break;
1155 #ifdef CONFIG_BLOCK
1156 case CEPH_MSG_DATA_BIO:
1157 page = ceph_msg_data_bio_next(cursor, page_offset, length);
1158 break;
1159 #endif /* CONFIG_BLOCK */
1160 case CEPH_MSG_DATA_BVECS:
1161 page = ceph_msg_data_bvecs_next(cursor, page_offset, length);
1162 break;
1163 case CEPH_MSG_DATA_NONE:
1164 default:
1165 page = NULL;
1166 break;
1167 }
1168
1169 BUG_ON(!page);
1170 BUG_ON(*page_offset + *length > PAGE_SIZE);
1171 BUG_ON(!*length);
1172 BUG_ON(*length > cursor->resid);
1173 if (last_piece)
1174 *last_piece = cursor->last_piece;
1175
1176 return page;
1177 }
1178
1179 /*
1180 * Returns true if the result moves the cursor on to the next piece
1181 * of the data item.
1182 */
ceph_msg_data_advance(struct ceph_msg_data_cursor * cursor,size_t bytes)1183 static void ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1184 size_t bytes)
1185 {
1186 bool new_piece;
1187
1188 BUG_ON(bytes > cursor->resid);
1189 switch (cursor->data->type) {
1190 case CEPH_MSG_DATA_PAGELIST:
1191 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1192 break;
1193 case CEPH_MSG_DATA_PAGES:
1194 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1195 break;
1196 #ifdef CONFIG_BLOCK
1197 case CEPH_MSG_DATA_BIO:
1198 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1199 break;
1200 #endif /* CONFIG_BLOCK */
1201 case CEPH_MSG_DATA_BVECS:
1202 new_piece = ceph_msg_data_bvecs_advance(cursor, bytes);
1203 break;
1204 case CEPH_MSG_DATA_NONE:
1205 default:
1206 BUG();
1207 break;
1208 }
1209 cursor->total_resid -= bytes;
1210
1211 if (!cursor->resid && cursor->total_resid) {
1212 WARN_ON(!cursor->last_piece);
1213 cursor->data++;
1214 __ceph_msg_data_cursor_init(cursor);
1215 new_piece = true;
1216 }
1217 cursor->need_crc = new_piece;
1218 }
1219
sizeof_footer(struct ceph_connection * con)1220 static size_t sizeof_footer(struct ceph_connection *con)
1221 {
1222 return (con->peer_features & CEPH_FEATURE_MSG_AUTH) ?
1223 sizeof(struct ceph_msg_footer) :
1224 sizeof(struct ceph_msg_footer_old);
1225 }
1226
prepare_message_data(struct ceph_msg * msg,u32 data_len)1227 static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
1228 {
1229 /* Initialize data cursor */
1230
1231 ceph_msg_data_cursor_init(msg, (size_t)data_len);
1232 }
1233
1234 /*
1235 * Prepare footer for currently outgoing message, and finish things
1236 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1237 */
prepare_write_message_footer(struct ceph_connection * con)1238 static void prepare_write_message_footer(struct ceph_connection *con)
1239 {
1240 struct ceph_msg *m = con->out_msg;
1241
1242 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1243
1244 dout("prepare_write_message_footer %p\n", con);
1245 con_out_kvec_add(con, sizeof_footer(con), &m->footer);
1246 if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
1247 if (con->ops->sign_message)
1248 con->ops->sign_message(m);
1249 else
1250 m->footer.sig = 0;
1251 } else {
1252 m->old_footer.flags = m->footer.flags;
1253 }
1254 con->out_more = m->more_to_follow;
1255 con->out_msg_done = true;
1256 }
1257
1258 /*
1259 * Prepare headers for the next outgoing message.
1260 */
prepare_write_message(struct ceph_connection * con)1261 static void prepare_write_message(struct ceph_connection *con)
1262 {
1263 struct ceph_msg *m;
1264 u32 crc;
1265
1266 con_out_kvec_reset(con);
1267 con->out_msg_done = false;
1268
1269 /* Sneak an ack in there first? If we can get it into the same
1270 * TCP packet that's a good thing. */
1271 if (con->in_seq > con->in_seq_acked) {
1272 con->in_seq_acked = con->in_seq;
1273 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1274 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1275 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1276 &con->out_temp_ack);
1277 }
1278
1279 BUG_ON(list_empty(&con->out_queue));
1280 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1281 con->out_msg = m;
1282 BUG_ON(m->con != con);
1283
1284 /* put message on sent list */
1285 ceph_msg_get(m);
1286 list_move_tail(&m->list_head, &con->out_sent);
1287
1288 /*
1289 * only assign outgoing seq # if we haven't sent this message
1290 * yet. if it is requeued, resend with it's original seq.
1291 */
1292 if (m->needs_out_seq) {
1293 m->hdr.seq = cpu_to_le64(++con->out_seq);
1294 m->needs_out_seq = false;
1295
1296 if (con->ops->reencode_message)
1297 con->ops->reencode_message(m);
1298 }
1299
1300 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1301 m, con->out_seq, le16_to_cpu(m->hdr.type),
1302 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1303 m->data_length);
1304 WARN_ON(m->front.iov_len != le32_to_cpu(m->hdr.front_len));
1305 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
1306
1307 /* tag + hdr + front + middle */
1308 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1309 con_out_kvec_add(con, sizeof(con->out_hdr), &con->out_hdr);
1310 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1311
1312 if (m->middle)
1313 con_out_kvec_add(con, m->middle->vec.iov_len,
1314 m->middle->vec.iov_base);
1315
1316 /* fill in hdr crc and finalize hdr */
1317 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1318 con->out_msg->hdr.crc = cpu_to_le32(crc);
1319 memcpy(&con->out_hdr, &con->out_msg->hdr, sizeof(con->out_hdr));
1320
1321 /* fill in front and middle crc, footer */
1322 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1323 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1324 if (m->middle) {
1325 crc = crc32c(0, m->middle->vec.iov_base,
1326 m->middle->vec.iov_len);
1327 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1328 } else
1329 con->out_msg->footer.middle_crc = 0;
1330 dout("%s front_crc %u middle_crc %u\n", __func__,
1331 le32_to_cpu(con->out_msg->footer.front_crc),
1332 le32_to_cpu(con->out_msg->footer.middle_crc));
1333 con->out_msg->footer.flags = 0;
1334
1335 /* is there a data payload? */
1336 con->out_msg->footer.data_crc = 0;
1337 if (m->data_length) {
1338 prepare_message_data(con->out_msg, m->data_length);
1339 con->out_more = 1; /* data + footer will follow */
1340 } else {
1341 /* no, queue up footer too and be done */
1342 prepare_write_message_footer(con);
1343 }
1344
1345 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1346 }
1347
1348 /*
1349 * Prepare an ack.
1350 */
prepare_write_ack(struct ceph_connection * con)1351 static void prepare_write_ack(struct ceph_connection *con)
1352 {
1353 dout("prepare_write_ack %p %llu -> %llu\n", con,
1354 con->in_seq_acked, con->in_seq);
1355 con->in_seq_acked = con->in_seq;
1356
1357 con_out_kvec_reset(con);
1358
1359 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1360
1361 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1362 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1363 &con->out_temp_ack);
1364
1365 con->out_more = 1; /* more will follow.. eventually.. */
1366 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1367 }
1368
1369 /*
1370 * Prepare to share the seq during handshake
1371 */
prepare_write_seq(struct ceph_connection * con)1372 static void prepare_write_seq(struct ceph_connection *con)
1373 {
1374 dout("prepare_write_seq %p %llu -> %llu\n", con,
1375 con->in_seq_acked, con->in_seq);
1376 con->in_seq_acked = con->in_seq;
1377
1378 con_out_kvec_reset(con);
1379
1380 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1381 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1382 &con->out_temp_ack);
1383
1384 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1385 }
1386
1387 /*
1388 * Prepare to write keepalive byte.
1389 */
prepare_write_keepalive(struct ceph_connection * con)1390 static void prepare_write_keepalive(struct ceph_connection *con)
1391 {
1392 dout("prepare_write_keepalive %p\n", con);
1393 con_out_kvec_reset(con);
1394 if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
1395 struct timespec64 now;
1396
1397 ktime_get_real_ts64(&now);
1398 con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
1399 ceph_encode_timespec64(&con->out_temp_keepalive2, &now);
1400 con_out_kvec_add(con, sizeof(con->out_temp_keepalive2),
1401 &con->out_temp_keepalive2);
1402 } else {
1403 con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive);
1404 }
1405 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1406 }
1407
1408 /*
1409 * Connection negotiation.
1410 */
1411
get_connect_authorizer(struct ceph_connection * con)1412 static int get_connect_authorizer(struct ceph_connection *con)
1413 {
1414 struct ceph_auth_handshake *auth;
1415 int auth_proto;
1416
1417 if (!con->ops->get_authorizer) {
1418 con->auth = NULL;
1419 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1420 con->out_connect.authorizer_len = 0;
1421 return 0;
1422 }
1423
1424 auth = con->ops->get_authorizer(con, &auth_proto, con->auth_retry);
1425 if (IS_ERR(auth))
1426 return PTR_ERR(auth);
1427
1428 con->auth = auth;
1429 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1430 con->out_connect.authorizer_len = cpu_to_le32(auth->authorizer_buf_len);
1431 return 0;
1432 }
1433
1434 /*
1435 * We connected to a peer and are saying hello.
1436 */
prepare_write_banner(struct ceph_connection * con)1437 static void prepare_write_banner(struct ceph_connection *con)
1438 {
1439 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1440 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1441 &con->msgr->my_enc_addr);
1442
1443 con->out_more = 0;
1444 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1445 }
1446
__prepare_write_connect(struct ceph_connection * con)1447 static void __prepare_write_connect(struct ceph_connection *con)
1448 {
1449 con_out_kvec_add(con, sizeof(con->out_connect), &con->out_connect);
1450 if (con->auth)
1451 con_out_kvec_add(con, con->auth->authorizer_buf_len,
1452 con->auth->authorizer_buf);
1453
1454 con->out_more = 0;
1455 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1456 }
1457
prepare_write_connect(struct ceph_connection * con)1458 static int prepare_write_connect(struct ceph_connection *con)
1459 {
1460 unsigned int global_seq = get_global_seq(con->msgr, 0);
1461 int proto;
1462 int ret;
1463
1464 switch (con->peer_name.type) {
1465 case CEPH_ENTITY_TYPE_MON:
1466 proto = CEPH_MONC_PROTOCOL;
1467 break;
1468 case CEPH_ENTITY_TYPE_OSD:
1469 proto = CEPH_OSDC_PROTOCOL;
1470 break;
1471 case CEPH_ENTITY_TYPE_MDS:
1472 proto = CEPH_MDSC_PROTOCOL;
1473 break;
1474 default:
1475 BUG();
1476 }
1477
1478 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1479 con->connect_seq, global_seq, proto);
1480
1481 con->out_connect.features =
1482 cpu_to_le64(from_msgr(con->msgr)->supported_features);
1483 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1484 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1485 con->out_connect.global_seq = cpu_to_le32(global_seq);
1486 con->out_connect.protocol_version = cpu_to_le32(proto);
1487 con->out_connect.flags = 0;
1488
1489 ret = get_connect_authorizer(con);
1490 if (ret)
1491 return ret;
1492
1493 __prepare_write_connect(con);
1494 return 0;
1495 }
1496
1497 /*
1498 * write as much of pending kvecs to the socket as we can.
1499 * 1 -> done
1500 * 0 -> socket full, but more to do
1501 * <0 -> error
1502 */
write_partial_kvec(struct ceph_connection * con)1503 static int write_partial_kvec(struct ceph_connection *con)
1504 {
1505 int ret;
1506
1507 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1508 while (con->out_kvec_bytes > 0) {
1509 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1510 con->out_kvec_left, con->out_kvec_bytes,
1511 con->out_more);
1512 if (ret <= 0)
1513 goto out;
1514 con->out_kvec_bytes -= ret;
1515 if (con->out_kvec_bytes == 0)
1516 break; /* done */
1517
1518 /* account for full iov entries consumed */
1519 while (ret >= con->out_kvec_cur->iov_len) {
1520 BUG_ON(!con->out_kvec_left);
1521 ret -= con->out_kvec_cur->iov_len;
1522 con->out_kvec_cur++;
1523 con->out_kvec_left--;
1524 }
1525 /* and for a partially-consumed entry */
1526 if (ret) {
1527 con->out_kvec_cur->iov_len -= ret;
1528 con->out_kvec_cur->iov_base += ret;
1529 }
1530 }
1531 con->out_kvec_left = 0;
1532 ret = 1;
1533 out:
1534 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1535 con->out_kvec_bytes, con->out_kvec_left, ret);
1536 return ret; /* done! */
1537 }
1538
ceph_crc32c_page(u32 crc,struct page * page,unsigned int page_offset,unsigned int length)1539 static u32 ceph_crc32c_page(u32 crc, struct page *page,
1540 unsigned int page_offset,
1541 unsigned int length)
1542 {
1543 char *kaddr;
1544
1545 kaddr = kmap(page);
1546 BUG_ON(kaddr == NULL);
1547 crc = crc32c(crc, kaddr + page_offset, length);
1548 kunmap(page);
1549
1550 return crc;
1551 }
1552 /*
1553 * Write as much message data payload as we can. If we finish, queue
1554 * up the footer.
1555 * 1 -> done, footer is now queued in out_kvec[].
1556 * 0 -> socket full, but more to do
1557 * <0 -> error
1558 */
write_partial_message_data(struct ceph_connection * con)1559 static int write_partial_message_data(struct ceph_connection *con)
1560 {
1561 struct ceph_msg *msg = con->out_msg;
1562 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1563 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
1564 int more = MSG_MORE | MSG_SENDPAGE_NOTLAST;
1565 u32 crc;
1566
1567 dout("%s %p msg %p\n", __func__, con, msg);
1568
1569 if (!msg->num_data_items)
1570 return -EINVAL;
1571
1572 /*
1573 * Iterate through each page that contains data to be
1574 * written, and send as much as possible for each.
1575 *
1576 * If we are calculating the data crc (the default), we will
1577 * need to map the page. If we have no pages, they have
1578 * been revoked, so use the zero page.
1579 */
1580 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1581 while (cursor->total_resid) {
1582 struct page *page;
1583 size_t page_offset;
1584 size_t length;
1585 int ret;
1586
1587 if (!cursor->resid) {
1588 ceph_msg_data_advance(cursor, 0);
1589 continue;
1590 }
1591
1592 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
1593 if (length == cursor->total_resid)
1594 more = MSG_MORE;
1595 ret = ceph_tcp_sendpage(con->sock, page, page_offset, length,
1596 more);
1597 if (ret <= 0) {
1598 if (do_datacrc)
1599 msg->footer.data_crc = cpu_to_le32(crc);
1600
1601 return ret;
1602 }
1603 if (do_datacrc && cursor->need_crc)
1604 crc = ceph_crc32c_page(crc, page, page_offset, length);
1605 ceph_msg_data_advance(cursor, (size_t)ret);
1606 }
1607
1608 dout("%s %p msg %p done\n", __func__, con, msg);
1609
1610 /* prepare and queue up footer, too */
1611 if (do_datacrc)
1612 msg->footer.data_crc = cpu_to_le32(crc);
1613 else
1614 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1615 con_out_kvec_reset(con);
1616 prepare_write_message_footer(con);
1617
1618 return 1; /* must return > 0 to indicate success */
1619 }
1620
1621 /*
1622 * write some zeros
1623 */
write_partial_skip(struct ceph_connection * con)1624 static int write_partial_skip(struct ceph_connection *con)
1625 {
1626 int more = MSG_MORE | MSG_SENDPAGE_NOTLAST;
1627 int ret;
1628
1629 dout("%s %p %d left\n", __func__, con, con->out_skip);
1630 while (con->out_skip > 0) {
1631 size_t size = min(con->out_skip, (int) PAGE_SIZE);
1632
1633 if (size == con->out_skip)
1634 more = MSG_MORE;
1635 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, more);
1636 if (ret <= 0)
1637 goto out;
1638 con->out_skip -= ret;
1639 }
1640 ret = 1;
1641 out:
1642 return ret;
1643 }
1644
1645 /*
1646 * Prepare to read connection handshake, or an ack.
1647 */
prepare_read_banner(struct ceph_connection * con)1648 static void prepare_read_banner(struct ceph_connection *con)
1649 {
1650 dout("prepare_read_banner %p\n", con);
1651 con->in_base_pos = 0;
1652 }
1653
prepare_read_connect(struct ceph_connection * con)1654 static void prepare_read_connect(struct ceph_connection *con)
1655 {
1656 dout("prepare_read_connect %p\n", con);
1657 con->in_base_pos = 0;
1658 }
1659
prepare_read_ack(struct ceph_connection * con)1660 static void prepare_read_ack(struct ceph_connection *con)
1661 {
1662 dout("prepare_read_ack %p\n", con);
1663 con->in_base_pos = 0;
1664 }
1665
prepare_read_seq(struct ceph_connection * con)1666 static void prepare_read_seq(struct ceph_connection *con)
1667 {
1668 dout("prepare_read_seq %p\n", con);
1669 con->in_base_pos = 0;
1670 con->in_tag = CEPH_MSGR_TAG_SEQ;
1671 }
1672
prepare_read_tag(struct ceph_connection * con)1673 static void prepare_read_tag(struct ceph_connection *con)
1674 {
1675 dout("prepare_read_tag %p\n", con);
1676 con->in_base_pos = 0;
1677 con->in_tag = CEPH_MSGR_TAG_READY;
1678 }
1679
prepare_read_keepalive_ack(struct ceph_connection * con)1680 static void prepare_read_keepalive_ack(struct ceph_connection *con)
1681 {
1682 dout("prepare_read_keepalive_ack %p\n", con);
1683 con->in_base_pos = 0;
1684 }
1685
1686 /*
1687 * Prepare to read a message.
1688 */
prepare_read_message(struct ceph_connection * con)1689 static int prepare_read_message(struct ceph_connection *con)
1690 {
1691 dout("prepare_read_message %p\n", con);
1692 BUG_ON(con->in_msg != NULL);
1693 con->in_base_pos = 0;
1694 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1695 return 0;
1696 }
1697
1698
read_partial(struct ceph_connection * con,int end,int size,void * object)1699 static int read_partial(struct ceph_connection *con,
1700 int end, int size, void *object)
1701 {
1702 while (con->in_base_pos < end) {
1703 int left = end - con->in_base_pos;
1704 int have = size - left;
1705 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1706 if (ret <= 0)
1707 return ret;
1708 con->in_base_pos += ret;
1709 }
1710 return 1;
1711 }
1712
1713
1714 /*
1715 * Read all or part of the connect-side handshake on a new connection
1716 */
read_partial_banner(struct ceph_connection * con)1717 static int read_partial_banner(struct ceph_connection *con)
1718 {
1719 int size;
1720 int end;
1721 int ret;
1722
1723 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1724
1725 /* peer's banner */
1726 size = strlen(CEPH_BANNER);
1727 end = size;
1728 ret = read_partial(con, end, size, con->in_banner);
1729 if (ret <= 0)
1730 goto out;
1731
1732 size = sizeof (con->actual_peer_addr);
1733 end += size;
1734 ret = read_partial(con, end, size, &con->actual_peer_addr);
1735 if (ret <= 0)
1736 goto out;
1737 ceph_decode_banner_addr(&con->actual_peer_addr);
1738
1739 size = sizeof (con->peer_addr_for_me);
1740 end += size;
1741 ret = read_partial(con, end, size, &con->peer_addr_for_me);
1742 if (ret <= 0)
1743 goto out;
1744 ceph_decode_banner_addr(&con->peer_addr_for_me);
1745
1746 out:
1747 return ret;
1748 }
1749
read_partial_connect(struct ceph_connection * con)1750 static int read_partial_connect(struct ceph_connection *con)
1751 {
1752 int size;
1753 int end;
1754 int ret;
1755
1756 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1757
1758 size = sizeof (con->in_reply);
1759 end = size;
1760 ret = read_partial(con, end, size, &con->in_reply);
1761 if (ret <= 0)
1762 goto out;
1763
1764 if (con->auth) {
1765 size = le32_to_cpu(con->in_reply.authorizer_len);
1766 if (size > con->auth->authorizer_reply_buf_len) {
1767 pr_err("authorizer reply too big: %d > %zu\n", size,
1768 con->auth->authorizer_reply_buf_len);
1769 ret = -EINVAL;
1770 goto out;
1771 }
1772
1773 end += size;
1774 ret = read_partial(con, end, size,
1775 con->auth->authorizer_reply_buf);
1776 if (ret <= 0)
1777 goto out;
1778 }
1779
1780 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1781 con, (int)con->in_reply.tag,
1782 le32_to_cpu(con->in_reply.connect_seq),
1783 le32_to_cpu(con->in_reply.global_seq));
1784 out:
1785 return ret;
1786 }
1787
1788 /*
1789 * Verify the hello banner looks okay.
1790 */
verify_hello(struct ceph_connection * con)1791 static int verify_hello(struct ceph_connection *con)
1792 {
1793 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1794 pr_err("connect to %s got bad banner\n",
1795 ceph_pr_addr(&con->peer_addr));
1796 con->error_msg = "protocol error, bad banner";
1797 return -1;
1798 }
1799 return 0;
1800 }
1801
addr_is_blank(struct ceph_entity_addr * addr)1802 static bool addr_is_blank(struct ceph_entity_addr *addr)
1803 {
1804 struct sockaddr_storage ss = addr->in_addr; /* align */
1805 struct in_addr *addr4 = &((struct sockaddr_in *)&ss)->sin_addr;
1806 struct in6_addr *addr6 = &((struct sockaddr_in6 *)&ss)->sin6_addr;
1807
1808 switch (ss.ss_family) {
1809 case AF_INET:
1810 return addr4->s_addr == htonl(INADDR_ANY);
1811 case AF_INET6:
1812 return ipv6_addr_any(addr6);
1813 default:
1814 return true;
1815 }
1816 }
1817
addr_port(struct ceph_entity_addr * addr)1818 static int addr_port(struct ceph_entity_addr *addr)
1819 {
1820 switch (get_unaligned(&addr->in_addr.ss_family)) {
1821 case AF_INET:
1822 return ntohs(get_unaligned(&((struct sockaddr_in *)&addr->in_addr)->sin_port));
1823 case AF_INET6:
1824 return ntohs(get_unaligned(&((struct sockaddr_in6 *)&addr->in_addr)->sin6_port));
1825 }
1826 return 0;
1827 }
1828
addr_set_port(struct ceph_entity_addr * addr,int p)1829 static void addr_set_port(struct ceph_entity_addr *addr, int p)
1830 {
1831 switch (get_unaligned(&addr->in_addr.ss_family)) {
1832 case AF_INET:
1833 put_unaligned(htons(p), &((struct sockaddr_in *)&addr->in_addr)->sin_port);
1834 break;
1835 case AF_INET6:
1836 put_unaligned(htons(p), &((struct sockaddr_in6 *)&addr->in_addr)->sin6_port);
1837 break;
1838 }
1839 }
1840
1841 /*
1842 * Unlike other *_pton function semantics, zero indicates success.
1843 */
ceph_pton(const char * str,size_t len,struct ceph_entity_addr * addr,char delim,const char ** ipend)1844 static int ceph_pton(const char *str, size_t len, struct ceph_entity_addr *addr,
1845 char delim, const char **ipend)
1846 {
1847 memset(&addr->in_addr, 0, sizeof(addr->in_addr));
1848
1849 if (in4_pton(str, len, (u8 *)&((struct sockaddr_in *)&addr->in_addr)->sin_addr.s_addr, delim, ipend)) {
1850 put_unaligned(AF_INET, &addr->in_addr.ss_family);
1851 return 0;
1852 }
1853
1854 if (in6_pton(str, len, (u8 *)&((struct sockaddr_in6 *)&addr->in_addr)->sin6_addr.s6_addr, delim, ipend)) {
1855 put_unaligned(AF_INET6, &addr->in_addr.ss_family);
1856 return 0;
1857 }
1858
1859 return -EINVAL;
1860 }
1861
1862 /*
1863 * Extract hostname string and resolve using kernel DNS facility.
1864 */
1865 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
ceph_dns_resolve_name(const char * name,size_t namelen,struct ceph_entity_addr * addr,char delim,const char ** ipend)1866 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1867 struct ceph_entity_addr *addr, char delim, const char **ipend)
1868 {
1869 const char *end, *delim_p;
1870 char *colon_p, *ip_addr = NULL;
1871 int ip_len, ret;
1872
1873 /*
1874 * The end of the hostname occurs immediately preceding the delimiter or
1875 * the port marker (':') where the delimiter takes precedence.
1876 */
1877 delim_p = memchr(name, delim, namelen);
1878 colon_p = memchr(name, ':', namelen);
1879
1880 if (delim_p && colon_p)
1881 end = delim_p < colon_p ? delim_p : colon_p;
1882 else if (!delim_p && colon_p)
1883 end = colon_p;
1884 else {
1885 end = delim_p;
1886 if (!end) /* case: hostname:/ */
1887 end = name + namelen;
1888 }
1889
1890 if (end <= name)
1891 return -EINVAL;
1892
1893 /* do dns_resolve upcall */
1894 ip_len = dns_query(current->nsproxy->net_ns,
1895 NULL, name, end - name, NULL, &ip_addr, NULL, false);
1896 if (ip_len > 0)
1897 ret = ceph_pton(ip_addr, ip_len, addr, -1, NULL);
1898 else
1899 ret = -ESRCH;
1900
1901 kfree(ip_addr);
1902
1903 *ipend = end;
1904
1905 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1906 ret, ret ? "failed" : ceph_pr_addr(addr));
1907
1908 return ret;
1909 }
1910 #else
ceph_dns_resolve_name(const char * name,size_t namelen,struct ceph_entity_addr * addr,char delim,const char ** ipend)1911 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1912 struct ceph_entity_addr *addr, char delim, const char **ipend)
1913 {
1914 return -EINVAL;
1915 }
1916 #endif
1917
1918 /*
1919 * Parse a server name (IP or hostname). If a valid IP address is not found
1920 * then try to extract a hostname to resolve using userspace DNS upcall.
1921 */
ceph_parse_server_name(const char * name,size_t namelen,struct ceph_entity_addr * addr,char delim,const char ** ipend)1922 static int ceph_parse_server_name(const char *name, size_t namelen,
1923 struct ceph_entity_addr *addr, char delim, const char **ipend)
1924 {
1925 int ret;
1926
1927 ret = ceph_pton(name, namelen, addr, delim, ipend);
1928 if (ret)
1929 ret = ceph_dns_resolve_name(name, namelen, addr, delim, ipend);
1930
1931 return ret;
1932 }
1933
1934 /*
1935 * Parse an ip[:port] list into an addr array. Use the default
1936 * monitor port if a port isn't specified.
1937 */
ceph_parse_ips(const char * c,const char * end,struct ceph_entity_addr * addr,int max_count,int * count)1938 int ceph_parse_ips(const char *c, const char *end,
1939 struct ceph_entity_addr *addr,
1940 int max_count, int *count)
1941 {
1942 int i, ret = -EINVAL;
1943 const char *p = c;
1944
1945 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1946 for (i = 0; i < max_count; i++) {
1947 const char *ipend;
1948 int port;
1949 char delim = ',';
1950
1951 if (*p == '[') {
1952 delim = ']';
1953 p++;
1954 }
1955
1956 ret = ceph_parse_server_name(p, end - p, &addr[i], delim, &ipend);
1957 if (ret)
1958 goto bad;
1959 ret = -EINVAL;
1960
1961 p = ipend;
1962
1963 if (delim == ']') {
1964 if (*p != ']') {
1965 dout("missing matching ']'\n");
1966 goto bad;
1967 }
1968 p++;
1969 }
1970
1971 /* port? */
1972 if (p < end && *p == ':') {
1973 port = 0;
1974 p++;
1975 while (p < end && *p >= '0' && *p <= '9') {
1976 port = (port * 10) + (*p - '0');
1977 p++;
1978 }
1979 if (port == 0)
1980 port = CEPH_MON_PORT;
1981 else if (port > 65535)
1982 goto bad;
1983 } else {
1984 port = CEPH_MON_PORT;
1985 }
1986
1987 addr_set_port(&addr[i], port);
1988 addr[i].type = CEPH_ENTITY_ADDR_TYPE_LEGACY;
1989
1990 dout("parse_ips got %s\n", ceph_pr_addr(&addr[i]));
1991
1992 if (p == end)
1993 break;
1994 if (*p != ',')
1995 goto bad;
1996 p++;
1997 }
1998
1999 if (p != end)
2000 goto bad;
2001
2002 if (count)
2003 *count = i + 1;
2004 return 0;
2005
2006 bad:
2007 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
2008 return ret;
2009 }
2010 EXPORT_SYMBOL(ceph_parse_ips);
2011
process_banner(struct ceph_connection * con)2012 static int process_banner(struct ceph_connection *con)
2013 {
2014 dout("process_banner on %p\n", con);
2015
2016 if (verify_hello(con) < 0)
2017 return -1;
2018
2019 /*
2020 * Make sure the other end is who we wanted. note that the other
2021 * end may not yet know their ip address, so if it's 0.0.0.0, give
2022 * them the benefit of the doubt.
2023 */
2024 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
2025 sizeof(con->peer_addr)) != 0 &&
2026 !(addr_is_blank(&con->actual_peer_addr) &&
2027 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
2028 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
2029 ceph_pr_addr(&con->peer_addr),
2030 (int)le32_to_cpu(con->peer_addr.nonce),
2031 ceph_pr_addr(&con->actual_peer_addr),
2032 (int)le32_to_cpu(con->actual_peer_addr.nonce));
2033 con->error_msg = "wrong peer at address";
2034 return -1;
2035 }
2036
2037 /*
2038 * did we learn our address?
2039 */
2040 if (addr_is_blank(&con->msgr->inst.addr)) {
2041 int port = addr_port(&con->msgr->inst.addr);
2042
2043 memcpy(&con->msgr->inst.addr.in_addr,
2044 &con->peer_addr_for_me.in_addr,
2045 sizeof(con->peer_addr_for_me.in_addr));
2046 addr_set_port(&con->msgr->inst.addr, port);
2047 encode_my_addr(con->msgr);
2048 dout("process_banner learned my addr is %s\n",
2049 ceph_pr_addr(&con->msgr->inst.addr));
2050 }
2051
2052 return 0;
2053 }
2054
process_connect(struct ceph_connection * con)2055 static int process_connect(struct ceph_connection *con)
2056 {
2057 u64 sup_feat = from_msgr(con->msgr)->supported_features;
2058 u64 req_feat = from_msgr(con->msgr)->required_features;
2059 u64 server_feat = le64_to_cpu(con->in_reply.features);
2060 int ret;
2061
2062 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
2063
2064 if (con->auth) {
2065 int len = le32_to_cpu(con->in_reply.authorizer_len);
2066
2067 /*
2068 * Any connection that defines ->get_authorizer()
2069 * should also define ->add_authorizer_challenge() and
2070 * ->verify_authorizer_reply().
2071 *
2072 * See get_connect_authorizer().
2073 */
2074 if (con->in_reply.tag == CEPH_MSGR_TAG_CHALLENGE_AUTHORIZER) {
2075 ret = con->ops->add_authorizer_challenge(
2076 con, con->auth->authorizer_reply_buf, len);
2077 if (ret < 0)
2078 return ret;
2079
2080 con_out_kvec_reset(con);
2081 __prepare_write_connect(con);
2082 prepare_read_connect(con);
2083 return 0;
2084 }
2085
2086 if (len) {
2087 ret = con->ops->verify_authorizer_reply(con);
2088 if (ret < 0) {
2089 con->error_msg = "bad authorize reply";
2090 return ret;
2091 }
2092 }
2093 }
2094
2095 switch (con->in_reply.tag) {
2096 case CEPH_MSGR_TAG_FEATURES:
2097 pr_err("%s%lld %s feature set mismatch,"
2098 " my %llx < server's %llx, missing %llx\n",
2099 ENTITY_NAME(con->peer_name),
2100 ceph_pr_addr(&con->peer_addr),
2101 sup_feat, server_feat, server_feat & ~sup_feat);
2102 con->error_msg = "missing required protocol features";
2103 reset_connection(con);
2104 return -1;
2105
2106 case CEPH_MSGR_TAG_BADPROTOVER:
2107 pr_err("%s%lld %s protocol version mismatch,"
2108 " my %d != server's %d\n",
2109 ENTITY_NAME(con->peer_name),
2110 ceph_pr_addr(&con->peer_addr),
2111 le32_to_cpu(con->out_connect.protocol_version),
2112 le32_to_cpu(con->in_reply.protocol_version));
2113 con->error_msg = "protocol version mismatch";
2114 reset_connection(con);
2115 return -1;
2116
2117 case CEPH_MSGR_TAG_BADAUTHORIZER:
2118 con->auth_retry++;
2119 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2120 con->auth_retry);
2121 if (con->auth_retry == 2) {
2122 con->error_msg = "connect authorization failure";
2123 return -1;
2124 }
2125 con_out_kvec_reset(con);
2126 ret = prepare_write_connect(con);
2127 if (ret < 0)
2128 return ret;
2129 prepare_read_connect(con);
2130 break;
2131
2132 case CEPH_MSGR_TAG_RESETSESSION:
2133 /*
2134 * If we connected with a large connect_seq but the peer
2135 * has no record of a session with us (no connection, or
2136 * connect_seq == 0), they will send RESETSESION to indicate
2137 * that they must have reset their session, and may have
2138 * dropped messages.
2139 */
2140 dout("process_connect got RESET peer seq %u\n",
2141 le32_to_cpu(con->in_reply.connect_seq));
2142 pr_err("%s%lld %s connection reset\n",
2143 ENTITY_NAME(con->peer_name),
2144 ceph_pr_addr(&con->peer_addr));
2145 reset_connection(con);
2146 con_out_kvec_reset(con);
2147 ret = prepare_write_connect(con);
2148 if (ret < 0)
2149 return ret;
2150 prepare_read_connect(con);
2151
2152 /* Tell ceph about it. */
2153 mutex_unlock(&con->mutex);
2154 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2155 if (con->ops->peer_reset)
2156 con->ops->peer_reset(con);
2157 mutex_lock(&con->mutex);
2158 if (con->state != CON_STATE_NEGOTIATING)
2159 return -EAGAIN;
2160 break;
2161
2162 case CEPH_MSGR_TAG_RETRY_SESSION:
2163 /*
2164 * If we sent a smaller connect_seq than the peer has, try
2165 * again with a larger value.
2166 */
2167 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2168 le32_to_cpu(con->out_connect.connect_seq),
2169 le32_to_cpu(con->in_reply.connect_seq));
2170 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2171 con_out_kvec_reset(con);
2172 ret = prepare_write_connect(con);
2173 if (ret < 0)
2174 return ret;
2175 prepare_read_connect(con);
2176 break;
2177
2178 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2179 /*
2180 * If we sent a smaller global_seq than the peer has, try
2181 * again with a larger value.
2182 */
2183 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2184 con->peer_global_seq,
2185 le32_to_cpu(con->in_reply.global_seq));
2186 get_global_seq(con->msgr,
2187 le32_to_cpu(con->in_reply.global_seq));
2188 con_out_kvec_reset(con);
2189 ret = prepare_write_connect(con);
2190 if (ret < 0)
2191 return ret;
2192 prepare_read_connect(con);
2193 break;
2194
2195 case CEPH_MSGR_TAG_SEQ:
2196 case CEPH_MSGR_TAG_READY:
2197 if (req_feat & ~server_feat) {
2198 pr_err("%s%lld %s protocol feature mismatch,"
2199 " my required %llx > server's %llx, need %llx\n",
2200 ENTITY_NAME(con->peer_name),
2201 ceph_pr_addr(&con->peer_addr),
2202 req_feat, server_feat, req_feat & ~server_feat);
2203 con->error_msg = "missing required protocol features";
2204 reset_connection(con);
2205 return -1;
2206 }
2207
2208 WARN_ON(con->state != CON_STATE_NEGOTIATING);
2209 con->state = CON_STATE_OPEN;
2210 con->auth_retry = 0; /* we authenticated; clear flag */
2211 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2212 con->connect_seq++;
2213 con->peer_features = server_feat;
2214 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2215 con->peer_global_seq,
2216 le32_to_cpu(con->in_reply.connect_seq),
2217 con->connect_seq);
2218 WARN_ON(con->connect_seq !=
2219 le32_to_cpu(con->in_reply.connect_seq));
2220
2221 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2222 con_flag_set(con, CON_FLAG_LOSSYTX);
2223
2224 con->delay = 0; /* reset backoff memory */
2225
2226 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2227 prepare_write_seq(con);
2228 prepare_read_seq(con);
2229 } else {
2230 prepare_read_tag(con);
2231 }
2232 break;
2233
2234 case CEPH_MSGR_TAG_WAIT:
2235 /*
2236 * If there is a connection race (we are opening
2237 * connections to each other), one of us may just have
2238 * to WAIT. This shouldn't happen if we are the
2239 * client.
2240 */
2241 con->error_msg = "protocol error, got WAIT as client";
2242 return -1;
2243
2244 default:
2245 con->error_msg = "protocol error, garbage tag during connect";
2246 return -1;
2247 }
2248 return 0;
2249 }
2250
2251
2252 /*
2253 * read (part of) an ack
2254 */
read_partial_ack(struct ceph_connection * con)2255 static int read_partial_ack(struct ceph_connection *con)
2256 {
2257 int size = sizeof (con->in_temp_ack);
2258 int end = size;
2259
2260 return read_partial(con, end, size, &con->in_temp_ack);
2261 }
2262
2263 /*
2264 * We can finally discard anything that's been acked.
2265 */
process_ack(struct ceph_connection * con)2266 static void process_ack(struct ceph_connection *con)
2267 {
2268 struct ceph_msg *m;
2269 u64 ack = le64_to_cpu(con->in_temp_ack);
2270 u64 seq;
2271 bool reconnect = (con->in_tag == CEPH_MSGR_TAG_SEQ);
2272 struct list_head *list = reconnect ? &con->out_queue : &con->out_sent;
2273
2274 /*
2275 * In the reconnect case, con_fault() has requeued messages
2276 * in out_sent. We should cleanup old messages according to
2277 * the reconnect seq.
2278 */
2279 while (!list_empty(list)) {
2280 m = list_first_entry(list, struct ceph_msg, list_head);
2281 if (reconnect && m->needs_out_seq)
2282 break;
2283 seq = le64_to_cpu(m->hdr.seq);
2284 if (seq > ack)
2285 break;
2286 dout("got ack for seq %llu type %d at %p\n", seq,
2287 le16_to_cpu(m->hdr.type), m);
2288 m->ack_stamp = jiffies;
2289 ceph_msg_remove(m);
2290 }
2291
2292 prepare_read_tag(con);
2293 }
2294
2295
read_partial_message_section(struct ceph_connection * con,struct kvec * section,unsigned int sec_len,u32 * crc)2296 static int read_partial_message_section(struct ceph_connection *con,
2297 struct kvec *section,
2298 unsigned int sec_len, u32 *crc)
2299 {
2300 int ret, left;
2301
2302 BUG_ON(!section);
2303
2304 while (section->iov_len < sec_len) {
2305 BUG_ON(section->iov_base == NULL);
2306 left = sec_len - section->iov_len;
2307 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2308 section->iov_len, left);
2309 if (ret <= 0)
2310 return ret;
2311 section->iov_len += ret;
2312 }
2313 if (section->iov_len == sec_len)
2314 *crc = crc32c(0, section->iov_base, section->iov_len);
2315
2316 return 1;
2317 }
2318
read_partial_msg_data(struct ceph_connection * con)2319 static int read_partial_msg_data(struct ceph_connection *con)
2320 {
2321 struct ceph_msg *msg = con->in_msg;
2322 struct ceph_msg_data_cursor *cursor = &msg->cursor;
2323 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2324 struct page *page;
2325 size_t page_offset;
2326 size_t length;
2327 u32 crc = 0;
2328 int ret;
2329
2330 if (!msg->num_data_items)
2331 return -EIO;
2332
2333 if (do_datacrc)
2334 crc = con->in_data_crc;
2335 while (cursor->total_resid) {
2336 if (!cursor->resid) {
2337 ceph_msg_data_advance(cursor, 0);
2338 continue;
2339 }
2340
2341 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
2342 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2343 if (ret <= 0) {
2344 if (do_datacrc)
2345 con->in_data_crc = crc;
2346
2347 return ret;
2348 }
2349
2350 if (do_datacrc)
2351 crc = ceph_crc32c_page(crc, page, page_offset, ret);
2352 ceph_msg_data_advance(cursor, (size_t)ret);
2353 }
2354 if (do_datacrc)
2355 con->in_data_crc = crc;
2356
2357 return 1; /* must return > 0 to indicate success */
2358 }
2359
2360 /*
2361 * read (part of) a message.
2362 */
2363 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2364
read_partial_message(struct ceph_connection * con)2365 static int read_partial_message(struct ceph_connection *con)
2366 {
2367 struct ceph_msg *m = con->in_msg;
2368 int size;
2369 int end;
2370 int ret;
2371 unsigned int front_len, middle_len, data_len;
2372 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2373 bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
2374 u64 seq;
2375 u32 crc;
2376
2377 dout("read_partial_message con %p msg %p\n", con, m);
2378
2379 /* header */
2380 size = sizeof (con->in_hdr);
2381 end = size;
2382 ret = read_partial(con, end, size, &con->in_hdr);
2383 if (ret <= 0)
2384 return ret;
2385
2386 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2387 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2388 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
2389 crc, con->in_hdr.crc);
2390 return -EBADMSG;
2391 }
2392
2393 front_len = le32_to_cpu(con->in_hdr.front_len);
2394 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2395 return -EIO;
2396 middle_len = le32_to_cpu(con->in_hdr.middle_len);
2397 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2398 return -EIO;
2399 data_len = le32_to_cpu(con->in_hdr.data_len);
2400 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2401 return -EIO;
2402
2403 /* verify seq# */
2404 seq = le64_to_cpu(con->in_hdr.seq);
2405 if ((s64)seq - (s64)con->in_seq < 1) {
2406 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2407 ENTITY_NAME(con->peer_name),
2408 ceph_pr_addr(&con->peer_addr),
2409 seq, con->in_seq + 1);
2410 con->in_base_pos = -front_len - middle_len - data_len -
2411 sizeof_footer(con);
2412 con->in_tag = CEPH_MSGR_TAG_READY;
2413 return 1;
2414 } else if ((s64)seq - (s64)con->in_seq > 1) {
2415 pr_err("read_partial_message bad seq %lld expected %lld\n",
2416 seq, con->in_seq + 1);
2417 con->error_msg = "bad message sequence # for incoming message";
2418 return -EBADE;
2419 }
2420
2421 /* allocate message? */
2422 if (!con->in_msg) {
2423 int skip = 0;
2424
2425 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2426 front_len, data_len);
2427 ret = ceph_con_in_msg_alloc(con, &skip);
2428 if (ret < 0)
2429 return ret;
2430
2431 BUG_ON(!con->in_msg ^ skip);
2432 if (skip) {
2433 /* skip this message */
2434 dout("alloc_msg said skip message\n");
2435 con->in_base_pos = -front_len - middle_len - data_len -
2436 sizeof_footer(con);
2437 con->in_tag = CEPH_MSGR_TAG_READY;
2438 con->in_seq++;
2439 return 1;
2440 }
2441
2442 BUG_ON(!con->in_msg);
2443 BUG_ON(con->in_msg->con != con);
2444 m = con->in_msg;
2445 m->front.iov_len = 0; /* haven't read it yet */
2446 if (m->middle)
2447 m->middle->vec.iov_len = 0;
2448
2449 /* prepare for data payload, if any */
2450
2451 if (data_len)
2452 prepare_message_data(con->in_msg, data_len);
2453 }
2454
2455 /* front */
2456 ret = read_partial_message_section(con, &m->front, front_len,
2457 &con->in_front_crc);
2458 if (ret <= 0)
2459 return ret;
2460
2461 /* middle */
2462 if (m->middle) {
2463 ret = read_partial_message_section(con, &m->middle->vec,
2464 middle_len,
2465 &con->in_middle_crc);
2466 if (ret <= 0)
2467 return ret;
2468 }
2469
2470 /* (page) data */
2471 if (data_len) {
2472 ret = read_partial_msg_data(con);
2473 if (ret <= 0)
2474 return ret;
2475 }
2476
2477 /* footer */
2478 size = sizeof_footer(con);
2479 end += size;
2480 ret = read_partial(con, end, size, &m->footer);
2481 if (ret <= 0)
2482 return ret;
2483
2484 if (!need_sign) {
2485 m->footer.flags = m->old_footer.flags;
2486 m->footer.sig = 0;
2487 }
2488
2489 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2490 m, front_len, m->footer.front_crc, middle_len,
2491 m->footer.middle_crc, data_len, m->footer.data_crc);
2492
2493 /* crc ok? */
2494 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2495 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2496 m, con->in_front_crc, m->footer.front_crc);
2497 return -EBADMSG;
2498 }
2499 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2500 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2501 m, con->in_middle_crc, m->footer.middle_crc);
2502 return -EBADMSG;
2503 }
2504 if (do_datacrc &&
2505 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2506 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2507 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2508 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2509 return -EBADMSG;
2510 }
2511
2512 if (need_sign && con->ops->check_message_signature &&
2513 con->ops->check_message_signature(m)) {
2514 pr_err("read_partial_message %p signature check failed\n", m);
2515 return -EBADMSG;
2516 }
2517
2518 return 1; /* done! */
2519 }
2520
2521 /*
2522 * Process message. This happens in the worker thread. The callback should
2523 * be careful not to do anything that waits on other incoming messages or it
2524 * may deadlock.
2525 */
process_message(struct ceph_connection * con)2526 static void process_message(struct ceph_connection *con)
2527 {
2528 struct ceph_msg *msg = con->in_msg;
2529
2530 BUG_ON(con->in_msg->con != con);
2531 con->in_msg = NULL;
2532
2533 /* if first message, set peer_name */
2534 if (con->peer_name.type == 0)
2535 con->peer_name = msg->hdr.src;
2536
2537 con->in_seq++;
2538 mutex_unlock(&con->mutex);
2539
2540 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2541 msg, le64_to_cpu(msg->hdr.seq),
2542 ENTITY_NAME(msg->hdr.src),
2543 le16_to_cpu(msg->hdr.type),
2544 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2545 le32_to_cpu(msg->hdr.front_len),
2546 le32_to_cpu(msg->hdr.data_len),
2547 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2548 con->ops->dispatch(con, msg);
2549
2550 mutex_lock(&con->mutex);
2551 }
2552
read_keepalive_ack(struct ceph_connection * con)2553 static int read_keepalive_ack(struct ceph_connection *con)
2554 {
2555 struct ceph_timespec ceph_ts;
2556 size_t size = sizeof(ceph_ts);
2557 int ret = read_partial(con, size, size, &ceph_ts);
2558 if (ret <= 0)
2559 return ret;
2560 ceph_decode_timespec64(&con->last_keepalive_ack, &ceph_ts);
2561 prepare_read_tag(con);
2562 return 1;
2563 }
2564
2565 /*
2566 * Write something to the socket. Called in a worker thread when the
2567 * socket appears to be writeable and we have something ready to send.
2568 */
try_write(struct ceph_connection * con)2569 static int try_write(struct ceph_connection *con)
2570 {
2571 int ret = 1;
2572
2573 dout("try_write start %p state %lu\n", con, con->state);
2574 if (con->state != CON_STATE_PREOPEN &&
2575 con->state != CON_STATE_CONNECTING &&
2576 con->state != CON_STATE_NEGOTIATING &&
2577 con->state != CON_STATE_OPEN)
2578 return 0;
2579
2580 /* open the socket first? */
2581 if (con->state == CON_STATE_PREOPEN) {
2582 BUG_ON(con->sock);
2583 con->state = CON_STATE_CONNECTING;
2584
2585 con_out_kvec_reset(con);
2586 prepare_write_banner(con);
2587 prepare_read_banner(con);
2588
2589 BUG_ON(con->in_msg);
2590 con->in_tag = CEPH_MSGR_TAG_READY;
2591 dout("try_write initiating connect on %p new state %lu\n",
2592 con, con->state);
2593 ret = ceph_tcp_connect(con);
2594 if (ret < 0) {
2595 con->error_msg = "connect error";
2596 goto out;
2597 }
2598 }
2599
2600 more:
2601 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2602 BUG_ON(!con->sock);
2603
2604 /* kvec data queued? */
2605 if (con->out_kvec_left) {
2606 ret = write_partial_kvec(con);
2607 if (ret <= 0)
2608 goto out;
2609 }
2610 if (con->out_skip) {
2611 ret = write_partial_skip(con);
2612 if (ret <= 0)
2613 goto out;
2614 }
2615
2616 /* msg pages? */
2617 if (con->out_msg) {
2618 if (con->out_msg_done) {
2619 ceph_msg_put(con->out_msg);
2620 con->out_msg = NULL; /* we're done with this one */
2621 goto do_next;
2622 }
2623
2624 ret = write_partial_message_data(con);
2625 if (ret == 1)
2626 goto more; /* we need to send the footer, too! */
2627 if (ret == 0)
2628 goto out;
2629 if (ret < 0) {
2630 dout("try_write write_partial_message_data err %d\n",
2631 ret);
2632 goto out;
2633 }
2634 }
2635
2636 do_next:
2637 if (con->state == CON_STATE_OPEN) {
2638 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2639 prepare_write_keepalive(con);
2640 goto more;
2641 }
2642 /* is anything else pending? */
2643 if (!list_empty(&con->out_queue)) {
2644 prepare_write_message(con);
2645 goto more;
2646 }
2647 if (con->in_seq > con->in_seq_acked) {
2648 prepare_write_ack(con);
2649 goto more;
2650 }
2651 }
2652
2653 /* Nothing to do! */
2654 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2655 dout("try_write nothing else to write.\n");
2656 ret = 0;
2657 out:
2658 dout("try_write done on %p ret %d\n", con, ret);
2659 return ret;
2660 }
2661
2662 /*
2663 * Read what we can from the socket.
2664 */
try_read(struct ceph_connection * con)2665 static int try_read(struct ceph_connection *con)
2666 {
2667 int ret = -1;
2668
2669 more:
2670 dout("try_read start on %p state %lu\n", con, con->state);
2671 if (con->state != CON_STATE_CONNECTING &&
2672 con->state != CON_STATE_NEGOTIATING &&
2673 con->state != CON_STATE_OPEN)
2674 return 0;
2675
2676 BUG_ON(!con->sock);
2677
2678 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2679 con->in_base_pos);
2680
2681 if (con->state == CON_STATE_CONNECTING) {
2682 dout("try_read connecting\n");
2683 ret = read_partial_banner(con);
2684 if (ret <= 0)
2685 goto out;
2686 ret = process_banner(con);
2687 if (ret < 0)
2688 goto out;
2689
2690 con->state = CON_STATE_NEGOTIATING;
2691
2692 /*
2693 * Received banner is good, exchange connection info.
2694 * Do not reset out_kvec, as sending our banner raced
2695 * with receiving peer banner after connect completed.
2696 */
2697 ret = prepare_write_connect(con);
2698 if (ret < 0)
2699 goto out;
2700 prepare_read_connect(con);
2701
2702 /* Send connection info before awaiting response */
2703 goto out;
2704 }
2705
2706 if (con->state == CON_STATE_NEGOTIATING) {
2707 dout("try_read negotiating\n");
2708 ret = read_partial_connect(con);
2709 if (ret <= 0)
2710 goto out;
2711 ret = process_connect(con);
2712 if (ret < 0)
2713 goto out;
2714 goto more;
2715 }
2716
2717 WARN_ON(con->state != CON_STATE_OPEN);
2718
2719 if (con->in_base_pos < 0) {
2720 /*
2721 * skipping + discarding content.
2722 */
2723 ret = ceph_tcp_recvmsg(con->sock, NULL, -con->in_base_pos);
2724 if (ret <= 0)
2725 goto out;
2726 dout("skipped %d / %d bytes\n", ret, -con->in_base_pos);
2727 con->in_base_pos += ret;
2728 if (con->in_base_pos)
2729 goto more;
2730 }
2731 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2732 /*
2733 * what's next?
2734 */
2735 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2736 if (ret <= 0)
2737 goto out;
2738 dout("try_read got tag %d\n", (int)con->in_tag);
2739 switch (con->in_tag) {
2740 case CEPH_MSGR_TAG_MSG:
2741 prepare_read_message(con);
2742 break;
2743 case CEPH_MSGR_TAG_ACK:
2744 prepare_read_ack(con);
2745 break;
2746 case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
2747 prepare_read_keepalive_ack(con);
2748 break;
2749 case CEPH_MSGR_TAG_CLOSE:
2750 con_close_socket(con);
2751 con->state = CON_STATE_CLOSED;
2752 goto out;
2753 default:
2754 goto bad_tag;
2755 }
2756 }
2757 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2758 ret = read_partial_message(con);
2759 if (ret <= 0) {
2760 switch (ret) {
2761 case -EBADMSG:
2762 con->error_msg = "bad crc/signature";
2763 /* fall through */
2764 case -EBADE:
2765 ret = -EIO;
2766 break;
2767 case -EIO:
2768 con->error_msg = "io error";
2769 break;
2770 }
2771 goto out;
2772 }
2773 if (con->in_tag == CEPH_MSGR_TAG_READY)
2774 goto more;
2775 process_message(con);
2776 if (con->state == CON_STATE_OPEN)
2777 prepare_read_tag(con);
2778 goto more;
2779 }
2780 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2781 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2782 /*
2783 * the final handshake seq exchange is semantically
2784 * equivalent to an ACK
2785 */
2786 ret = read_partial_ack(con);
2787 if (ret <= 0)
2788 goto out;
2789 process_ack(con);
2790 goto more;
2791 }
2792 if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
2793 ret = read_keepalive_ack(con);
2794 if (ret <= 0)
2795 goto out;
2796 goto more;
2797 }
2798
2799 out:
2800 dout("try_read done on %p ret %d\n", con, ret);
2801 return ret;
2802
2803 bad_tag:
2804 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2805 con->error_msg = "protocol error, garbage tag";
2806 ret = -1;
2807 goto out;
2808 }
2809
2810
2811 /*
2812 * Atomically queue work on a connection after the specified delay.
2813 * Bump @con reference to avoid races with connection teardown.
2814 * Returns 0 if work was queued, or an error code otherwise.
2815 */
queue_con_delay(struct ceph_connection * con,unsigned long delay)2816 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2817 {
2818 if (!con->ops->get(con)) {
2819 dout("%s %p ref count 0\n", __func__, con);
2820 return -ENOENT;
2821 }
2822
2823 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2824 dout("%s %p - already queued\n", __func__, con);
2825 con->ops->put(con);
2826 return -EBUSY;
2827 }
2828
2829 dout("%s %p %lu\n", __func__, con, delay);
2830 return 0;
2831 }
2832
queue_con(struct ceph_connection * con)2833 static void queue_con(struct ceph_connection *con)
2834 {
2835 (void) queue_con_delay(con, 0);
2836 }
2837
cancel_con(struct ceph_connection * con)2838 static void cancel_con(struct ceph_connection *con)
2839 {
2840 if (cancel_delayed_work(&con->work)) {
2841 dout("%s %p\n", __func__, con);
2842 con->ops->put(con);
2843 }
2844 }
2845
con_sock_closed(struct ceph_connection * con)2846 static bool con_sock_closed(struct ceph_connection *con)
2847 {
2848 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2849 return false;
2850
2851 #define CASE(x) \
2852 case CON_STATE_ ## x: \
2853 con->error_msg = "socket closed (con state " #x ")"; \
2854 break;
2855
2856 switch (con->state) {
2857 CASE(CLOSED);
2858 CASE(PREOPEN);
2859 CASE(CONNECTING);
2860 CASE(NEGOTIATING);
2861 CASE(OPEN);
2862 CASE(STANDBY);
2863 default:
2864 pr_warn("%s con %p unrecognized state %lu\n",
2865 __func__, con, con->state);
2866 con->error_msg = "unrecognized con state";
2867 BUG();
2868 break;
2869 }
2870 #undef CASE
2871
2872 return true;
2873 }
2874
con_backoff(struct ceph_connection * con)2875 static bool con_backoff(struct ceph_connection *con)
2876 {
2877 int ret;
2878
2879 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2880 return false;
2881
2882 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2883 if (ret) {
2884 dout("%s: con %p FAILED to back off %lu\n", __func__,
2885 con, con->delay);
2886 BUG_ON(ret == -ENOENT);
2887 con_flag_set(con, CON_FLAG_BACKOFF);
2888 }
2889
2890 return true;
2891 }
2892
2893 /* Finish fault handling; con->mutex must *not* be held here */
2894
con_fault_finish(struct ceph_connection * con)2895 static void con_fault_finish(struct ceph_connection *con)
2896 {
2897 dout("%s %p\n", __func__, con);
2898
2899 /*
2900 * in case we faulted due to authentication, invalidate our
2901 * current tickets so that we can get new ones.
2902 */
2903 if (con->auth_retry) {
2904 dout("auth_retry %d, invalidating\n", con->auth_retry);
2905 if (con->ops->invalidate_authorizer)
2906 con->ops->invalidate_authorizer(con);
2907 con->auth_retry = 0;
2908 }
2909
2910 if (con->ops->fault)
2911 con->ops->fault(con);
2912 }
2913
2914 /*
2915 * Do some work on a connection. Drop a connection ref when we're done.
2916 */
ceph_con_workfn(struct work_struct * work)2917 static void ceph_con_workfn(struct work_struct *work)
2918 {
2919 struct ceph_connection *con = container_of(work, struct ceph_connection,
2920 work.work);
2921 bool fault;
2922
2923 mutex_lock(&con->mutex);
2924 while (true) {
2925 int ret;
2926
2927 if ((fault = con_sock_closed(con))) {
2928 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2929 break;
2930 }
2931 if (con_backoff(con)) {
2932 dout("%s: con %p BACKOFF\n", __func__, con);
2933 break;
2934 }
2935 if (con->state == CON_STATE_STANDBY) {
2936 dout("%s: con %p STANDBY\n", __func__, con);
2937 break;
2938 }
2939 if (con->state == CON_STATE_CLOSED) {
2940 dout("%s: con %p CLOSED\n", __func__, con);
2941 BUG_ON(con->sock);
2942 break;
2943 }
2944 if (con->state == CON_STATE_PREOPEN) {
2945 dout("%s: con %p PREOPEN\n", __func__, con);
2946 BUG_ON(con->sock);
2947 }
2948
2949 ret = try_read(con);
2950 if (ret < 0) {
2951 if (ret == -EAGAIN)
2952 continue;
2953 if (!con->error_msg)
2954 con->error_msg = "socket error on read";
2955 fault = true;
2956 break;
2957 }
2958
2959 ret = try_write(con);
2960 if (ret < 0) {
2961 if (ret == -EAGAIN)
2962 continue;
2963 if (!con->error_msg)
2964 con->error_msg = "socket error on write";
2965 fault = true;
2966 }
2967
2968 break; /* If we make it to here, we're done */
2969 }
2970 if (fault)
2971 con_fault(con);
2972 mutex_unlock(&con->mutex);
2973
2974 if (fault)
2975 con_fault_finish(con);
2976
2977 con->ops->put(con);
2978 }
2979
2980 /*
2981 * Generic error/fault handler. A retry mechanism is used with
2982 * exponential backoff
2983 */
con_fault(struct ceph_connection * con)2984 static void con_fault(struct ceph_connection *con)
2985 {
2986 dout("fault %p state %lu to peer %s\n",
2987 con, con->state, ceph_pr_addr(&con->peer_addr));
2988
2989 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2990 ceph_pr_addr(&con->peer_addr), con->error_msg);
2991 con->error_msg = NULL;
2992
2993 WARN_ON(con->state != CON_STATE_CONNECTING &&
2994 con->state != CON_STATE_NEGOTIATING &&
2995 con->state != CON_STATE_OPEN);
2996
2997 con_close_socket(con);
2998
2999 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
3000 dout("fault on LOSSYTX channel, marking CLOSED\n");
3001 con->state = CON_STATE_CLOSED;
3002 return;
3003 }
3004
3005 if (con->in_msg) {
3006 BUG_ON(con->in_msg->con != con);
3007 ceph_msg_put(con->in_msg);
3008 con->in_msg = NULL;
3009 }
3010
3011 /* Requeue anything that hasn't been acked */
3012 list_splice_init(&con->out_sent, &con->out_queue);
3013
3014 /* If there are no messages queued or keepalive pending, place
3015 * the connection in a STANDBY state */
3016 if (list_empty(&con->out_queue) &&
3017 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
3018 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
3019 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
3020 con->state = CON_STATE_STANDBY;
3021 } else {
3022 /* retry after a delay. */
3023 con->state = CON_STATE_PREOPEN;
3024 if (con->delay == 0)
3025 con->delay = BASE_DELAY_INTERVAL;
3026 else if (con->delay < MAX_DELAY_INTERVAL)
3027 con->delay *= 2;
3028 con_flag_set(con, CON_FLAG_BACKOFF);
3029 queue_con(con);
3030 }
3031 }
3032
3033
ceph_messenger_reset_nonce(struct ceph_messenger * msgr)3034 void ceph_messenger_reset_nonce(struct ceph_messenger *msgr)
3035 {
3036 u32 nonce = le32_to_cpu(msgr->inst.addr.nonce) + 1000000;
3037 msgr->inst.addr.nonce = cpu_to_le32(nonce);
3038 encode_my_addr(msgr);
3039 }
3040
3041 /*
3042 * initialize a new messenger instance
3043 */
ceph_messenger_init(struct ceph_messenger * msgr,struct ceph_entity_addr * myaddr)3044 void ceph_messenger_init(struct ceph_messenger *msgr,
3045 struct ceph_entity_addr *myaddr)
3046 {
3047 spin_lock_init(&msgr->global_seq_lock);
3048
3049 if (myaddr)
3050 msgr->inst.addr = *myaddr;
3051
3052 /* select a random nonce */
3053 msgr->inst.addr.type = 0;
3054 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
3055 encode_my_addr(msgr);
3056
3057 atomic_set(&msgr->stopping, 0);
3058 write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
3059
3060 dout("%s %p\n", __func__, msgr);
3061 }
3062 EXPORT_SYMBOL(ceph_messenger_init);
3063
ceph_messenger_fini(struct ceph_messenger * msgr)3064 void ceph_messenger_fini(struct ceph_messenger *msgr)
3065 {
3066 put_net(read_pnet(&msgr->net));
3067 }
3068 EXPORT_SYMBOL(ceph_messenger_fini);
3069
msg_con_set(struct ceph_msg * msg,struct ceph_connection * con)3070 static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con)
3071 {
3072 if (msg->con)
3073 msg->con->ops->put(msg->con);
3074
3075 msg->con = con ? con->ops->get(con) : NULL;
3076 BUG_ON(msg->con != con);
3077 }
3078
clear_standby(struct ceph_connection * con)3079 static void clear_standby(struct ceph_connection *con)
3080 {
3081 /* come back from STANDBY? */
3082 if (con->state == CON_STATE_STANDBY) {
3083 dout("clear_standby %p and ++connect_seq\n", con);
3084 con->state = CON_STATE_PREOPEN;
3085 con->connect_seq++;
3086 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
3087 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
3088 }
3089 }
3090
3091 /*
3092 * Queue up an outgoing message on the given connection.
3093 */
ceph_con_send(struct ceph_connection * con,struct ceph_msg * msg)3094 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
3095 {
3096 /* set src+dst */
3097 msg->hdr.src = con->msgr->inst.name;
3098 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
3099 msg->needs_out_seq = true;
3100
3101 mutex_lock(&con->mutex);
3102
3103 if (con->state == CON_STATE_CLOSED) {
3104 dout("con_send %p closed, dropping %p\n", con, msg);
3105 ceph_msg_put(msg);
3106 mutex_unlock(&con->mutex);
3107 return;
3108 }
3109
3110 msg_con_set(msg, con);
3111
3112 BUG_ON(!list_empty(&msg->list_head));
3113 list_add_tail(&msg->list_head, &con->out_queue);
3114 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3115 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3116 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3117 le32_to_cpu(msg->hdr.front_len),
3118 le32_to_cpu(msg->hdr.middle_len),
3119 le32_to_cpu(msg->hdr.data_len));
3120
3121 clear_standby(con);
3122 mutex_unlock(&con->mutex);
3123
3124 /* if there wasn't anything waiting to send before, queue
3125 * new work */
3126 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3127 queue_con(con);
3128 }
3129 EXPORT_SYMBOL(ceph_con_send);
3130
3131 /*
3132 * Revoke a message that was previously queued for send
3133 */
ceph_msg_revoke(struct ceph_msg * msg)3134 void ceph_msg_revoke(struct ceph_msg *msg)
3135 {
3136 struct ceph_connection *con = msg->con;
3137
3138 if (!con) {
3139 dout("%s msg %p null con\n", __func__, msg);
3140 return; /* Message not in our possession */
3141 }
3142
3143 mutex_lock(&con->mutex);
3144 if (!list_empty(&msg->list_head)) {
3145 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
3146 list_del_init(&msg->list_head);
3147 msg->hdr.seq = 0;
3148
3149 ceph_msg_put(msg);
3150 }
3151 if (con->out_msg == msg) {
3152 BUG_ON(con->out_skip);
3153 /* footer */
3154 if (con->out_msg_done) {
3155 con->out_skip += con_out_kvec_skip(con);
3156 } else {
3157 BUG_ON(!msg->data_length);
3158 con->out_skip += sizeof_footer(con);
3159 }
3160 /* data, middle, front */
3161 if (msg->data_length)
3162 con->out_skip += msg->cursor.total_resid;
3163 if (msg->middle)
3164 con->out_skip += con_out_kvec_skip(con);
3165 con->out_skip += con_out_kvec_skip(con);
3166
3167 dout("%s %p msg %p - was sending, will write %d skip %d\n",
3168 __func__, con, msg, con->out_kvec_bytes, con->out_skip);
3169 msg->hdr.seq = 0;
3170 con->out_msg = NULL;
3171 ceph_msg_put(msg);
3172 }
3173
3174 mutex_unlock(&con->mutex);
3175 }
3176
3177 /*
3178 * Revoke a message that we may be reading data into
3179 */
ceph_msg_revoke_incoming(struct ceph_msg * msg)3180 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
3181 {
3182 struct ceph_connection *con = msg->con;
3183
3184 if (!con) {
3185 dout("%s msg %p null con\n", __func__, msg);
3186 return; /* Message not in our possession */
3187 }
3188
3189 mutex_lock(&con->mutex);
3190 if (con->in_msg == msg) {
3191 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3192 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3193 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
3194
3195 /* skip rest of message */
3196 dout("%s %p msg %p revoked\n", __func__, con, msg);
3197 con->in_base_pos = con->in_base_pos -
3198 sizeof(struct ceph_msg_header) -
3199 front_len -
3200 middle_len -
3201 data_len -
3202 sizeof(struct ceph_msg_footer);
3203 ceph_msg_put(con->in_msg);
3204 con->in_msg = NULL;
3205 con->in_tag = CEPH_MSGR_TAG_READY;
3206 con->in_seq++;
3207 } else {
3208 dout("%s %p in_msg %p msg %p no-op\n",
3209 __func__, con, con->in_msg, msg);
3210 }
3211 mutex_unlock(&con->mutex);
3212 }
3213
3214 /*
3215 * Queue a keepalive byte to ensure the tcp connection is alive.
3216 */
ceph_con_keepalive(struct ceph_connection * con)3217 void ceph_con_keepalive(struct ceph_connection *con)
3218 {
3219 dout("con_keepalive %p\n", con);
3220 mutex_lock(&con->mutex);
3221 clear_standby(con);
3222 con_flag_set(con, CON_FLAG_KEEPALIVE_PENDING);
3223 mutex_unlock(&con->mutex);
3224
3225 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3226 queue_con(con);
3227 }
3228 EXPORT_SYMBOL(ceph_con_keepalive);
3229
ceph_con_keepalive_expired(struct ceph_connection * con,unsigned long interval)3230 bool ceph_con_keepalive_expired(struct ceph_connection *con,
3231 unsigned long interval)
3232 {
3233 if (interval > 0 &&
3234 (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
3235 struct timespec64 now;
3236 struct timespec64 ts;
3237 ktime_get_real_ts64(&now);
3238 jiffies_to_timespec64(interval, &ts);
3239 ts = timespec64_add(con->last_keepalive_ack, ts);
3240 return timespec64_compare(&now, &ts) >= 0;
3241 }
3242 return false;
3243 }
3244
ceph_msg_data_add(struct ceph_msg * msg)3245 static struct ceph_msg_data *ceph_msg_data_add(struct ceph_msg *msg)
3246 {
3247 BUG_ON(msg->num_data_items >= msg->max_data_items);
3248 return &msg->data[msg->num_data_items++];
3249 }
3250
ceph_msg_data_destroy(struct ceph_msg_data * data)3251 static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3252 {
3253 if (data->type == CEPH_MSG_DATA_PAGELIST)
3254 ceph_pagelist_release(data->pagelist);
3255 }
3256
ceph_msg_data_add_pages(struct ceph_msg * msg,struct page ** pages,size_t length,size_t alignment)3257 void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
3258 size_t length, size_t alignment)
3259 {
3260 struct ceph_msg_data *data;
3261
3262 BUG_ON(!pages);
3263 BUG_ON(!length);
3264
3265 data = ceph_msg_data_add(msg);
3266 data->type = CEPH_MSG_DATA_PAGES;
3267 data->pages = pages;
3268 data->length = length;
3269 data->alignment = alignment & ~PAGE_MASK;
3270
3271 msg->data_length += length;
3272 }
3273 EXPORT_SYMBOL(ceph_msg_data_add_pages);
3274
ceph_msg_data_add_pagelist(struct ceph_msg * msg,struct ceph_pagelist * pagelist)3275 void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
3276 struct ceph_pagelist *pagelist)
3277 {
3278 struct ceph_msg_data *data;
3279
3280 BUG_ON(!pagelist);
3281 BUG_ON(!pagelist->length);
3282
3283 data = ceph_msg_data_add(msg);
3284 data->type = CEPH_MSG_DATA_PAGELIST;
3285 refcount_inc(&pagelist->refcnt);
3286 data->pagelist = pagelist;
3287
3288 msg->data_length += pagelist->length;
3289 }
3290 EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
3291
3292 #ifdef CONFIG_BLOCK
ceph_msg_data_add_bio(struct ceph_msg * msg,struct ceph_bio_iter * bio_pos,u32 length)3293 void ceph_msg_data_add_bio(struct ceph_msg *msg, struct ceph_bio_iter *bio_pos,
3294 u32 length)
3295 {
3296 struct ceph_msg_data *data;
3297
3298 data = ceph_msg_data_add(msg);
3299 data->type = CEPH_MSG_DATA_BIO;
3300 data->bio_pos = *bio_pos;
3301 data->bio_length = length;
3302
3303 msg->data_length += length;
3304 }
3305 EXPORT_SYMBOL(ceph_msg_data_add_bio);
3306 #endif /* CONFIG_BLOCK */
3307
ceph_msg_data_add_bvecs(struct ceph_msg * msg,struct ceph_bvec_iter * bvec_pos)3308 void ceph_msg_data_add_bvecs(struct ceph_msg *msg,
3309 struct ceph_bvec_iter *bvec_pos)
3310 {
3311 struct ceph_msg_data *data;
3312
3313 data = ceph_msg_data_add(msg);
3314 data->type = CEPH_MSG_DATA_BVECS;
3315 data->bvec_pos = *bvec_pos;
3316
3317 msg->data_length += bvec_pos->iter.bi_size;
3318 }
3319 EXPORT_SYMBOL(ceph_msg_data_add_bvecs);
3320
3321 /*
3322 * construct a new message with given type, size
3323 * the new msg has a ref count of 1.
3324 */
ceph_msg_new2(int type,int front_len,int max_data_items,gfp_t flags,bool can_fail)3325 struct ceph_msg *ceph_msg_new2(int type, int front_len, int max_data_items,
3326 gfp_t flags, bool can_fail)
3327 {
3328 struct ceph_msg *m;
3329
3330 m = kmem_cache_zalloc(ceph_msg_cache, flags);
3331 if (m == NULL)
3332 goto out;
3333
3334 m->hdr.type = cpu_to_le16(type);
3335 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3336 m->hdr.front_len = cpu_to_le32(front_len);
3337
3338 INIT_LIST_HEAD(&m->list_head);
3339 kref_init(&m->kref);
3340
3341 /* front */
3342 if (front_len) {
3343 m->front.iov_base = ceph_kvmalloc(front_len, flags);
3344 if (m->front.iov_base == NULL) {
3345 dout("ceph_msg_new can't allocate %d bytes\n",
3346 front_len);
3347 goto out2;
3348 }
3349 } else {
3350 m->front.iov_base = NULL;
3351 }
3352 m->front_alloc_len = m->front.iov_len = front_len;
3353
3354 if (max_data_items) {
3355 m->data = kmalloc_array(max_data_items, sizeof(*m->data),
3356 flags);
3357 if (!m->data)
3358 goto out2;
3359
3360 m->max_data_items = max_data_items;
3361 }
3362
3363 dout("ceph_msg_new %p front %d\n", m, front_len);
3364 return m;
3365
3366 out2:
3367 ceph_msg_put(m);
3368 out:
3369 if (!can_fail) {
3370 pr_err("msg_new can't create type %d front %d\n", type,
3371 front_len);
3372 WARN_ON(1);
3373 } else {
3374 dout("msg_new can't create type %d front %d\n", type,
3375 front_len);
3376 }
3377 return NULL;
3378 }
3379 EXPORT_SYMBOL(ceph_msg_new2);
3380
ceph_msg_new(int type,int front_len,gfp_t flags,bool can_fail)3381 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3382 bool can_fail)
3383 {
3384 return ceph_msg_new2(type, front_len, 0, flags, can_fail);
3385 }
3386 EXPORT_SYMBOL(ceph_msg_new);
3387
3388 /*
3389 * Allocate "middle" portion of a message, if it is needed and wasn't
3390 * allocated by alloc_msg. This allows us to read a small fixed-size
3391 * per-type header in the front and then gracefully fail (i.e.,
3392 * propagate the error to the caller based on info in the front) when
3393 * the middle is too large.
3394 */
ceph_alloc_middle(struct ceph_connection * con,struct ceph_msg * msg)3395 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3396 {
3397 int type = le16_to_cpu(msg->hdr.type);
3398 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3399
3400 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3401 ceph_msg_type_name(type), middle_len);
3402 BUG_ON(!middle_len);
3403 BUG_ON(msg->middle);
3404
3405 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3406 if (!msg->middle)
3407 return -ENOMEM;
3408 return 0;
3409 }
3410
3411 /*
3412 * Allocate a message for receiving an incoming message on a
3413 * connection, and save the result in con->in_msg. Uses the
3414 * connection's private alloc_msg op if available.
3415 *
3416 * Returns 0 on success, or a negative error code.
3417 *
3418 * On success, if we set *skip = 1:
3419 * - the next message should be skipped and ignored.
3420 * - con->in_msg == NULL
3421 * or if we set *skip = 0:
3422 * - con->in_msg is non-null.
3423 * On error (ENOMEM, EAGAIN, ...),
3424 * - con->in_msg == NULL
3425 */
ceph_con_in_msg_alloc(struct ceph_connection * con,int * skip)3426 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3427 {
3428 struct ceph_msg_header *hdr = &con->in_hdr;
3429 int middle_len = le32_to_cpu(hdr->middle_len);
3430 struct ceph_msg *msg;
3431 int ret = 0;
3432
3433 BUG_ON(con->in_msg != NULL);
3434 BUG_ON(!con->ops->alloc_msg);
3435
3436 mutex_unlock(&con->mutex);
3437 msg = con->ops->alloc_msg(con, hdr, skip);
3438 mutex_lock(&con->mutex);
3439 if (con->state != CON_STATE_OPEN) {
3440 if (msg)
3441 ceph_msg_put(msg);
3442 return -EAGAIN;
3443 }
3444 if (msg) {
3445 BUG_ON(*skip);
3446 msg_con_set(msg, con);
3447 con->in_msg = msg;
3448 } else {
3449 /*
3450 * Null message pointer means either we should skip
3451 * this message or we couldn't allocate memory. The
3452 * former is not an error.
3453 */
3454 if (*skip)
3455 return 0;
3456
3457 con->error_msg = "error allocating memory for incoming message";
3458 return -ENOMEM;
3459 }
3460 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3461
3462 if (middle_len && !con->in_msg->middle) {
3463 ret = ceph_alloc_middle(con, con->in_msg);
3464 if (ret < 0) {
3465 ceph_msg_put(con->in_msg);
3466 con->in_msg = NULL;
3467 }
3468 }
3469
3470 return ret;
3471 }
3472
3473
3474 /*
3475 * Free a generically kmalloc'd message.
3476 */
ceph_msg_free(struct ceph_msg * m)3477 static void ceph_msg_free(struct ceph_msg *m)
3478 {
3479 dout("%s %p\n", __func__, m);
3480 kvfree(m->front.iov_base);
3481 kfree(m->data);
3482 kmem_cache_free(ceph_msg_cache, m);
3483 }
3484
ceph_msg_release(struct kref * kref)3485 static void ceph_msg_release(struct kref *kref)
3486 {
3487 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3488 int i;
3489
3490 dout("%s %p\n", __func__, m);
3491 WARN_ON(!list_empty(&m->list_head));
3492
3493 msg_con_set(m, NULL);
3494
3495 /* drop middle, data, if any */
3496 if (m->middle) {
3497 ceph_buffer_put(m->middle);
3498 m->middle = NULL;
3499 }
3500
3501 for (i = 0; i < m->num_data_items; i++)
3502 ceph_msg_data_destroy(&m->data[i]);
3503
3504 if (m->pool)
3505 ceph_msgpool_put(m->pool, m);
3506 else
3507 ceph_msg_free(m);
3508 }
3509
ceph_msg_get(struct ceph_msg * msg)3510 struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3511 {
3512 dout("%s %p (was %d)\n", __func__, msg,
3513 kref_read(&msg->kref));
3514 kref_get(&msg->kref);
3515 return msg;
3516 }
3517 EXPORT_SYMBOL(ceph_msg_get);
3518
ceph_msg_put(struct ceph_msg * msg)3519 void ceph_msg_put(struct ceph_msg *msg)
3520 {
3521 dout("%s %p (was %d)\n", __func__, msg,
3522 kref_read(&msg->kref));
3523 kref_put(&msg->kref, ceph_msg_release);
3524 }
3525 EXPORT_SYMBOL(ceph_msg_put);
3526
ceph_msg_dump(struct ceph_msg * msg)3527 void ceph_msg_dump(struct ceph_msg *msg)
3528 {
3529 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3530 msg->front_alloc_len, msg->data_length);
3531 print_hex_dump(KERN_DEBUG, "header: ",
3532 DUMP_PREFIX_OFFSET, 16, 1,
3533 &msg->hdr, sizeof(msg->hdr), true);
3534 print_hex_dump(KERN_DEBUG, " front: ",
3535 DUMP_PREFIX_OFFSET, 16, 1,
3536 msg->front.iov_base, msg->front.iov_len, true);
3537 if (msg->middle)
3538 print_hex_dump(KERN_DEBUG, "middle: ",
3539 DUMP_PREFIX_OFFSET, 16, 1,
3540 msg->middle->vec.iov_base,
3541 msg->middle->vec.iov_len, true);
3542 print_hex_dump(KERN_DEBUG, "footer: ",
3543 DUMP_PREFIX_OFFSET, 16, 1,
3544 &msg->footer, sizeof(msg->footer), true);
3545 }
3546 EXPORT_SYMBOL(ceph_msg_dump);
3547