projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
libceph: collapse all data items into one
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / net / ceph / messenger.c
... / ...
CommitLineData
1#include <linux/ceph/ceph_debug.h>
2
3#include <linux/crc32c.h>
4#include <linux/ctype.h>
5#include <linux/highmem.h>
6#include <linux/inet.h>
7#include <linux/kthread.h>
8#include <linux/net.h>
9#include <linux/slab.h>
10#include <linux/socket.h>
11#include <linux/string.h>
12#ifdef CONFIG_BLOCK
13#include <linux/bio.h>
14#endif /* CONFIG_BLOCK */
15#include <linux/dns_resolver.h>
16#include <net/tcp.h>
17
18#include <linux/ceph/libceph.h>
19#include <linux/ceph/messenger.h>
20#include <linux/ceph/decode.h>
21#include <linux/ceph/pagelist.h>
22#include <linux/export.h>
23
24#define list_entry_next(pos, member) \
25 list_entry(pos->member.next, typeof(*pos), member)
26
27/*
28 * Ceph uses the messenger to exchange ceph_msg messages with other
29 * hosts in the system. The messenger provides ordered and reliable
30 * delivery. We tolerate TCP disconnects by reconnecting (with
31 * exponential backoff) in the case of a fault (disconnection, bad
32 * crc, protocol error). Acks allow sent messages to be discarded by
33 * the sender.
34 */
35
36/*
37 * We track the state of the socket on a given connection using
38 * values defined below. The transition to a new socket state is
39 * handled by a function which verifies we aren't coming from an
40 * unexpected state.
41 *
42 * --------
43 * | NEW* | transient initial state
44 * --------
45 * | con_sock_state_init()
46 * v
47 * ----------
48 * | CLOSED | initialized, but no socket (and no
49 * ---------- TCP connection)
50 * ^ \
51 * | \ con_sock_state_connecting()
52 * | ----------------------
53 * | \
54 * + con_sock_state_closed() \
55 * |+--------------------------- \
56 * | \ \ \
57 * | ----------- \ \
58 * | | CLOSING | socket event; \ \
59 * | ----------- await close \ \
60 * | ^ \ |
61 * | | \ |
62 * | + con_sock_state_closing() \ |
63 * | / \ | |
64 * | / --------------- | |
65 * | / \ v v
66 * | / --------------
67 * | / -----------------| CONNECTING | socket created, TCP
68 * | | / -------------- connect initiated
69 * | | | con_sock_state_connected()
70 * | | v
71 * -------------
72 * | CONNECTED | TCP connection established
73 * -------------
74 *
75 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
76 */
77
78#define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
79#define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
80#define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
81#define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
82#define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
83
84/*
85 * connection states
86 */
87#define CON_STATE_CLOSED 1 /* -> PREOPEN */
88#define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
89#define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
90#define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
91#define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
92#define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
93
94/*
95 * ceph_connection flag bits
96 */
97#define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
98 * messages on errors */
99#define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
100#define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
101#define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
102#define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
103
104static bool con_flag_valid(unsigned long con_flag)
105{
106 switch (con_flag) {
107 case CON_FLAG_LOSSYTX:
108 case CON_FLAG_KEEPALIVE_PENDING:
109 case CON_FLAG_WRITE_PENDING:
110 case CON_FLAG_SOCK_CLOSED:
111 case CON_FLAG_BACKOFF:
112 return true;
113 default:
114 return false;
115 }
116}
117
118static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
119{
120 BUG_ON(!con_flag_valid(con_flag));
121
122 clear_bit(con_flag, &con->flags);
123}
124
125static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
126{
127 BUG_ON(!con_flag_valid(con_flag));
128
129 set_bit(con_flag, &con->flags);
130}
131
132static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
133{
134 BUG_ON(!con_flag_valid(con_flag));
135
136 return test_bit(con_flag, &con->flags);
137}
138
139static bool con_flag_test_and_clear(struct ceph_connection *con,
140 unsigned long con_flag)
141{
142 BUG_ON(!con_flag_valid(con_flag));
143
144 return test_and_clear_bit(con_flag, &con->flags);
145}
146
147static bool con_flag_test_and_set(struct ceph_connection *con,
148 unsigned long con_flag)
149{
150 BUG_ON(!con_flag_valid(con_flag));
151
152 return test_and_set_bit(con_flag, &con->flags);
153}
154
155/* static tag bytes (protocol control messages) */
156static char tag_msg = CEPH_MSGR_TAG_MSG;
157static char tag_ack = CEPH_MSGR_TAG_ACK;
158static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
159
160#ifdef CONFIG_LOCKDEP
161static struct lock_class_key socket_class;
162#endif
163
164/*
165 * When skipping (ignoring) a block of input we read it into a "skip
166 * buffer," which is this many bytes in size.
167 */
168#define SKIP_BUF_SIZE 1024
169
170static void queue_con(struct ceph_connection *con);
171static void con_work(struct work_struct *);
172static void con_fault(struct ceph_connection *con);
173
174/*
175 * Nicely render a sockaddr as a string. An array of formatted
176 * strings is used, to approximate reentrancy.
177 */
178#define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
179#define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
180#define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
181#define MAX_ADDR_STR_LEN 64 /* 54 is enough */
182
183static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
184static atomic_t addr_str_seq = ATOMIC_INIT(0);
185
186static struct page *zero_page; /* used in certain error cases */
187
188const char *ceph_pr_addr(const struct sockaddr_storage *ss)
189{
190 int i;
191 char *s;
192 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
193 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
194
195 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
196 s = addr_str[i];
197
198 switch (ss->ss_family) {
199 case AF_INET:
200 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
201 ntohs(in4->sin_port));
202 break;
203
204 case AF_INET6:
205 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
206 ntohs(in6->sin6_port));
207 break;
208
209 default:
210 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
211 ss->ss_family);
212 }
213
214 return s;
215}
216EXPORT_SYMBOL(ceph_pr_addr);
217
218static void encode_my_addr(struct ceph_messenger *msgr)
219{
220 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
221 ceph_encode_addr(&msgr->my_enc_addr);
222}
223
224/*
225 * work queue for all reading and writing to/from the socket.
226 */
227static struct workqueue_struct *ceph_msgr_wq;
228
229static void _ceph_msgr_exit(void)
230{
231 if (ceph_msgr_wq) {
232 destroy_workqueue(ceph_msgr_wq);
233 ceph_msgr_wq = NULL;
234 }
235
236 BUG_ON(zero_page == NULL);
237 kunmap(zero_page);
238 page_cache_release(zero_page);
239 zero_page = NULL;
240}
241
242int ceph_msgr_init(void)
243{
244 BUG_ON(zero_page != NULL);
245 zero_page = ZERO_PAGE(0);
246 page_cache_get(zero_page);
247
248 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0);
249 if (ceph_msgr_wq)
250 return 0;
251
252 pr_err("msgr_init failed to create workqueue\n");
253 _ceph_msgr_exit();
254
255 return -ENOMEM;
256}
257EXPORT_SYMBOL(ceph_msgr_init);
258
259void ceph_msgr_exit(void)
260{
261 BUG_ON(ceph_msgr_wq == NULL);
262
263 _ceph_msgr_exit();
264}
265EXPORT_SYMBOL(ceph_msgr_exit);
266
267void ceph_msgr_flush(void)
268{
269 flush_workqueue(ceph_msgr_wq);
270}
271EXPORT_SYMBOL(ceph_msgr_flush);
272
273/* Connection socket state transition functions */
274
275static void con_sock_state_init(struct ceph_connection *con)
276{
277 int old_state;
278
279 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
280 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
281 printk("%s: unexpected old state %d\n", __func__, old_state);
282 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
283 CON_SOCK_STATE_CLOSED);
284}
285
286static void con_sock_state_connecting(struct ceph_connection *con)
287{
288 int old_state;
289
290 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
291 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
292 printk("%s: unexpected old state %d\n", __func__, old_state);
293 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
294 CON_SOCK_STATE_CONNECTING);
295}
296
297static void con_sock_state_connected(struct ceph_connection *con)
298{
299 int old_state;
300
301 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
302 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
303 printk("%s: unexpected old state %d\n", __func__, old_state);
304 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
305 CON_SOCK_STATE_CONNECTED);
306}
307
308static void con_sock_state_closing(struct ceph_connection *con)
309{
310 int old_state;
311
312 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
313 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
314 old_state != CON_SOCK_STATE_CONNECTED &&
315 old_state != CON_SOCK_STATE_CLOSING))
316 printk("%s: unexpected old state %d\n", __func__, old_state);
317 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
318 CON_SOCK_STATE_CLOSING);
319}
320
321static void con_sock_state_closed(struct ceph_connection *con)
322{
323 int old_state;
324
325 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
326 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
327 old_state != CON_SOCK_STATE_CLOSING &&
328 old_state != CON_SOCK_STATE_CONNECTING &&
329 old_state != CON_SOCK_STATE_CLOSED))
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_CLOSED);
333}
334
335/*
336 * socket callback functions
337 */
338
339/* data available on socket, or listen socket received a connect */
340static void ceph_sock_data_ready(struct sock *sk, int count_unused)
341{
342 struct ceph_connection *con = sk->sk_user_data;
343 if (atomic_read(&con->msgr->stopping)) {
344 return;
345 }
346
347 if (sk->sk_state != TCP_CLOSE_WAIT) {
348 dout("%s on %p state = %lu, queueing work\n", __func__,
349 con, con->state);
350 queue_con(con);
351 }
352}
353
354/* socket has buffer space for writing */
355static void ceph_sock_write_space(struct sock *sk)
356{
357 struct ceph_connection *con = sk->sk_user_data;
358
359 /* only queue to workqueue if there is data we want to write,
360 * and there is sufficient space in the socket buffer to accept
361 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
362 * doesn't get called again until try_write() fills the socket
363 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
364 * and net/core/stream.c:sk_stream_write_space().
365 */
366 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
367 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
368 dout("%s %p queueing write work\n", __func__, con);
369 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
370 queue_con(con);
371 }
372 } else {
373 dout("%s %p nothing to write\n", __func__, con);
374 }
375}
376
377/* socket's state has changed */
378static void ceph_sock_state_change(struct sock *sk)
379{
380 struct ceph_connection *con = sk->sk_user_data;
381
382 dout("%s %p state = %lu sk_state = %u\n", __func__,
383 con, con->state, sk->sk_state);
384
385 switch (sk->sk_state) {
386 case TCP_CLOSE:
387 dout("%s TCP_CLOSE\n", __func__);
388 case TCP_CLOSE_WAIT:
389 dout("%s TCP_CLOSE_WAIT\n", __func__);
390 con_sock_state_closing(con);
391 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
392 queue_con(con);
393 break;
394 case TCP_ESTABLISHED:
395 dout("%s TCP_ESTABLISHED\n", __func__);
396 con_sock_state_connected(con);
397 queue_con(con);
398 break;
399 default: /* Everything else is uninteresting */
400 break;
401 }
402}
403
404/*
405 * set up socket callbacks
406 */
407static void set_sock_callbacks(struct socket *sock,
408 struct ceph_connection *con)
409{
410 struct sock *sk = sock->sk;
411 sk->sk_user_data = con;
412 sk->sk_data_ready = ceph_sock_data_ready;
413 sk->sk_write_space = ceph_sock_write_space;
414 sk->sk_state_change = ceph_sock_state_change;
415}
416
417
418/*
419 * socket helpers
420 */
421
422/*
423 * initiate connection to a remote socket.
424 */
425static int ceph_tcp_connect(struct ceph_connection *con)
426{
427 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
428 struct socket *sock;
429 int ret;
430
431 BUG_ON(con->sock);
432 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
433 IPPROTO_TCP, &sock);
434 if (ret)
435 return ret;
436 sock->sk->sk_allocation = GFP_NOFS;
437
438#ifdef CONFIG_LOCKDEP
439 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
440#endif
441
442 set_sock_callbacks(sock, con);
443
444 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
445
446 con_sock_state_connecting(con);
447 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
448 O_NONBLOCK);
449 if (ret == -EINPROGRESS) {
450 dout("connect %s EINPROGRESS sk_state = %u\n",
451 ceph_pr_addr(&con->peer_addr.in_addr),
452 sock->sk->sk_state);
453 } else if (ret < 0) {
454 pr_err("connect %s error %d\n",
455 ceph_pr_addr(&con->peer_addr.in_addr), ret);
456 sock_release(sock);
457 con->error_msg = "connect error";
458
459 return ret;
460 }
461 con->sock = sock;
462 return 0;
463}
464
465static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
466{
467 struct kvec iov = {buf, len};
468 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
469 int r;
470
471 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
472 if (r == -EAGAIN)
473 r = 0;
474 return r;
475}
476
477static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
478 int page_offset, size_t length)
479{
480 void *kaddr;
481 int ret;
482
483 BUG_ON(page_offset + length > PAGE_SIZE);
484
485 kaddr = kmap(page);
486 BUG_ON(!kaddr);
487 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
488 kunmap(page);
489
490 return ret;
491}
492
493/*
494 * write something. @more is true if caller will be sending more data
495 * shortly.
496 */
497static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
498 size_t kvlen, size_t len, int more)
499{
500 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
501 int r;
502
503 if (more)
504 msg.msg_flags |= MSG_MORE;
505 else
506 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
507
508 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
509 if (r == -EAGAIN)
510 r = 0;
511 return r;
512}
513
514static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
515 int offset, size_t size, bool more)
516{
517 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
518 int ret;
519
520 ret = kernel_sendpage(sock, page, offset, size, flags);
521 if (ret == -EAGAIN)
522 ret = 0;
523
524 return ret;
525}
526
527
528/*
529 * Shutdown/close the socket for the given connection.
530 */
531static int con_close_socket(struct ceph_connection *con)
532{
533 int rc = 0;
534
535 dout("con_close_socket on %p sock %p\n", con, con->sock);
536 if (con->sock) {
537 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
538 sock_release(con->sock);
539 con->sock = NULL;
540 }
541
542 /*
543 * Forcibly clear the SOCK_CLOSED flag. It gets set
544 * independent of the connection mutex, and we could have
545 * received a socket close event before we had the chance to
546 * shut the socket down.
547 */
548 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
549
550 con_sock_state_closed(con);
551 return rc;
552}
553
554/*
555 * Reset a connection. Discard all incoming and outgoing messages
556 * and clear *_seq state.
557 */
558static void ceph_msg_remove(struct ceph_msg *msg)
559{
560 list_del_init(&msg->list_head);
561 BUG_ON(msg->con == NULL);
562 msg->con->ops->put(msg->con);
563 msg->con = NULL;
564
565 ceph_msg_put(msg);
566}
567static void ceph_msg_remove_list(struct list_head *head)
568{
569 while (!list_empty(head)) {
570 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
571 list_head);
572 ceph_msg_remove(msg);
573 }
574}
575
576static void reset_connection(struct ceph_connection *con)
577{
578 /* reset connection, out_queue, msg_ and connect_seq */
579 /* discard existing out_queue and msg_seq */
580 dout("reset_connection %p\n", con);
581 ceph_msg_remove_list(&con->out_queue);
582 ceph_msg_remove_list(&con->out_sent);
583
584 if (con->in_msg) {
585 BUG_ON(con->in_msg->con != con);
586 con->in_msg->con = NULL;
587 ceph_msg_put(con->in_msg);
588 con->in_msg = NULL;
589 con->ops->put(con);
590 }
591
592 con->connect_seq = 0;
593 con->out_seq = 0;
594 if (con->out_msg) {
595 ceph_msg_put(con->out_msg);
596 con->out_msg = NULL;
597 }
598 con->in_seq = 0;
599 con->in_seq_acked = 0;
600}
601
602/*
603 * mark a peer down. drop any open connections.
604 */
605void ceph_con_close(struct ceph_connection *con)
606{
607 mutex_lock(&con->mutex);
608 dout("con_close %p peer %s\n", con,
609 ceph_pr_addr(&con->peer_addr.in_addr));
610 con->state = CON_STATE_CLOSED;
611
612 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
613 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
614 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
615 con_flag_clear(con, CON_FLAG_BACKOFF);
616
617 reset_connection(con);
618 con->peer_global_seq = 0;
619 cancel_delayed_work(&con->work);
620 con_close_socket(con);
621 mutex_unlock(&con->mutex);
622}
623EXPORT_SYMBOL(ceph_con_close);
624
625/*
626 * Reopen a closed connection, with a new peer address.
627 */
628void ceph_con_open(struct ceph_connection *con,
629 __u8 entity_type, __u64 entity_num,
630 struct ceph_entity_addr *addr)
631{
632 mutex_lock(&con->mutex);
633 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
634
635 WARN_ON(con->state != CON_STATE_CLOSED);
636 con->state = CON_STATE_PREOPEN;
637
638 con->peer_name.type = (__u8) entity_type;
639 con->peer_name.num = cpu_to_le64(entity_num);
640
641 memcpy(&con->peer_addr, addr, sizeof(*addr));
642 con->delay = 0; /* reset backoff memory */
643 mutex_unlock(&con->mutex);
644 queue_con(con);
645}
646EXPORT_SYMBOL(ceph_con_open);
647
648/*
649 * return true if this connection ever successfully opened
650 */
651bool ceph_con_opened(struct ceph_connection *con)
652{
653 return con->connect_seq > 0;
654}
655
656/*
657 * initialize a new connection.
658 */
659void ceph_con_init(struct ceph_connection *con, void *private,
660 const struct ceph_connection_operations *ops,
661 struct ceph_messenger *msgr)
662{
663 dout("con_init %p\n", con);
664 memset(con, 0, sizeof(*con));
665 con->private = private;
666 con->ops = ops;
667 con->msgr = msgr;
668
669 con_sock_state_init(con);
670
671 mutex_init(&con->mutex);
672 INIT_LIST_HEAD(&con->out_queue);
673 INIT_LIST_HEAD(&con->out_sent);
674 INIT_DELAYED_WORK(&con->work, con_work);
675
676 con->state = CON_STATE_CLOSED;
677}
678EXPORT_SYMBOL(ceph_con_init);
679
680
681/*
682 * We maintain a global counter to order connection attempts. Get
683 * a unique seq greater than @gt.
684 */
685static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
686{
687 u32 ret;
688
689 spin_lock(&msgr->global_seq_lock);
690 if (msgr->global_seq < gt)
691 msgr->global_seq = gt;
692 ret = ++msgr->global_seq;
693 spin_unlock(&msgr->global_seq_lock);
694 return ret;
695}
696
697static void con_out_kvec_reset(struct ceph_connection *con)
698{
699 con->out_kvec_left = 0;
700 con->out_kvec_bytes = 0;
701 con->out_kvec_cur = &con->out_kvec[0];
702}
703
704static void con_out_kvec_add(struct ceph_connection *con,
705 size_t size, void *data)
706{
707 int index;
708
709 index = con->out_kvec_left;
710 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
711
712 con->out_kvec[index].iov_len = size;
713 con->out_kvec[index].iov_base = data;
714 con->out_kvec_left++;
715 con->out_kvec_bytes += size;
716}
717
718#ifdef CONFIG_BLOCK
719
720/*
721 * For a bio data item, a piece is whatever remains of the next
722 * entry in the current bio iovec, or the first entry in the next
723 * bio in the list.
724 */
725static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data *data,
726 size_t length)
727{
728 struct ceph_msg_data_cursor *cursor = &data->cursor;
729 struct bio *bio;
730
731 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
732
733 bio = data->bio;
734 BUG_ON(!bio);
735 BUG_ON(!bio->bi_vcnt);
736
737 cursor->resid = length;
738 cursor->bio = bio;
739 cursor->vector_index = 0;
740 cursor->vector_offset = 0;
741 cursor->last_piece = length <= bio->bi_io_vec[0].bv_len;
742}
743
744static struct page *ceph_msg_data_bio_next(struct ceph_msg_data *data,
745 size_t *page_offset,
746 size_t *length)
747{
748 struct ceph_msg_data_cursor *cursor = &data->cursor;
749 struct bio *bio;
750 struct bio_vec *bio_vec;
751 unsigned int index;
752
753 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
754
755 bio = cursor->bio;
756 BUG_ON(!bio);
757
758 index = cursor->vector_index;
759 BUG_ON(index >= (unsigned int) bio->bi_vcnt);
760
761 bio_vec = &bio->bi_io_vec[index];
762 BUG_ON(cursor->vector_offset >= bio_vec->bv_len);
763 *page_offset = (size_t) (bio_vec->bv_offset + cursor->vector_offset);
764 BUG_ON(*page_offset >= PAGE_SIZE);
765 if (cursor->last_piece) /* pagelist offset is always 0 */
766 *length = cursor->resid;
767 else
768 *length = (size_t) (bio_vec->bv_len - cursor->vector_offset);
769 BUG_ON(*length > PAGE_SIZE);
770 BUG_ON(*length > cursor->resid);
771
772 return bio_vec->bv_page;
773}
774
775static bool ceph_msg_data_bio_advance(struct ceph_msg_data *data, size_t bytes)
776{
777 struct ceph_msg_data_cursor *cursor = &data->cursor;
778 struct bio *bio;
779 struct bio_vec *bio_vec;
780 unsigned int index;
781
782 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
783
784 bio = cursor->bio;
785 BUG_ON(!bio);
786
787 index = cursor->vector_index;
788 BUG_ON(index >= (unsigned int) bio->bi_vcnt);
789 bio_vec = &bio->bi_io_vec[index];
790
791 /* Advance the cursor offset */
792
793 BUG_ON(cursor->resid < bytes);
794 cursor->resid -= bytes;
795 cursor->vector_offset += bytes;
796 if (cursor->vector_offset < bio_vec->bv_len)
797 return false; /* more bytes to process in this segment */
798 BUG_ON(cursor->vector_offset != bio_vec->bv_len);
799
800 /* Move on to the next segment, and possibly the next bio */
801
802 if (++index == (unsigned int) bio->bi_vcnt) {
803 bio = bio->bi_next;
804 index = 0;
805 }
806 cursor->bio = bio;
807 cursor->vector_index = index;
808 cursor->vector_offset = 0;
809
810 if (!cursor->last_piece) {
811 BUG_ON(!cursor->resid);
812 BUG_ON(!bio);
813 /* A short read is OK, so use <= rather than == */
814 if (cursor->resid <= bio->bi_io_vec[index].bv_len)
815 cursor->last_piece = true;
816 }
817
818 return true;
819}
820#endif
821
822/*
823 * For a page array, a piece comes from the first page in the array
824 * that has not already been fully consumed.
825 */
826static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data *data,
827 size_t length)
828{
829 struct ceph_msg_data_cursor *cursor = &data->cursor;
830 int page_count;
831
832 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
833
834 BUG_ON(!data->pages);
835 BUG_ON(!data->length);
836 BUG_ON(length != data->length);
837
838 cursor->resid = length;
839 page_count = calc_pages_for(data->alignment, (u64)data->length);
840 cursor->page_offset = data->alignment & ~PAGE_MASK;
841 cursor->page_index = 0;
842 BUG_ON(page_count > (int) USHRT_MAX);
843 cursor->page_count = (unsigned short) page_count;
844 cursor->last_piece = length <= PAGE_SIZE;
845}
846
847static struct page *ceph_msg_data_pages_next(struct ceph_msg_data *data,
848 size_t *page_offset,
849 size_t *length)
850{
851 struct ceph_msg_data_cursor *cursor = &data->cursor;
852
853 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
854
855 BUG_ON(cursor->page_index >= cursor->page_count);
856 BUG_ON(cursor->page_offset >= PAGE_SIZE);
857
858 *page_offset = cursor->page_offset;
859 if (cursor->last_piece)
860 *length = cursor->resid;
861 else
862 *length = PAGE_SIZE - *page_offset;
863
864 return data->pages[cursor->page_index];
865}
866
867static bool ceph_msg_data_pages_advance(struct ceph_msg_data *data,
868 size_t bytes)
869{
870 struct ceph_msg_data_cursor *cursor = &data->cursor;
871
872 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
873
874 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
875
876 /* Advance the cursor page offset */
877
878 cursor->resid -= bytes;
879 cursor->page_offset += bytes;
880 if (!bytes || cursor->page_offset & ~PAGE_MASK)
881 return false; /* more bytes to process in the current page */
882
883 /* Move on to the next page */
884
885 BUG_ON(cursor->page_index >= cursor->page_count);
886 cursor->page_offset = 0;
887 cursor->page_index++;
888 cursor->last_piece = cursor->resid <= PAGE_SIZE;
889
890 return true;
891}
892
893/*
894 * For a pagelist, a piece is whatever remains to be consumed in the
895 * first page in the list, or the front of the next page.
896 */
897static void ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data *data,
898 size_t length)
899{
900 struct ceph_msg_data_cursor *cursor = &data->cursor;
901 struct ceph_pagelist *pagelist;
902 struct page *page;
903
904 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
905
906 pagelist = data->pagelist;
907 BUG_ON(!pagelist);
908 BUG_ON(length != pagelist->length);
909
910 if (!length)
911 return; /* pagelist can be assigned but empty */
912
913 BUG_ON(list_empty(&pagelist->head));
914 page = list_first_entry(&pagelist->head, struct page, lru);
915
916 cursor->resid = length;
917 cursor->page = page;
918 cursor->offset = 0;
919 cursor->last_piece = length <= PAGE_SIZE;
920}
921
922static struct page *ceph_msg_data_pagelist_next(struct ceph_msg_data *data,
923 size_t *page_offset,
924 size_t *length)
925{
926 struct ceph_msg_data_cursor *cursor = &data->cursor;
927 struct ceph_pagelist *pagelist;
928
929 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
930
931 pagelist = data->pagelist;
932 BUG_ON(!pagelist);
933
934 BUG_ON(!cursor->page);
935 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
936
937 *page_offset = cursor->offset & ~PAGE_MASK;
938 if (cursor->last_piece) /* pagelist offset is always 0 */
939 *length = cursor->resid;
940 else
941 *length = PAGE_SIZE - *page_offset;
942
943 return data->cursor.page;
944}
945
946static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data *data,
947 size_t bytes)
948{
949 struct ceph_msg_data_cursor *cursor = &data->cursor;
950 struct ceph_pagelist *pagelist;
951
952 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
953
954 pagelist = data->pagelist;
955 BUG_ON(!pagelist);
956
957 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
958 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
959
960 /* Advance the cursor offset */
961
962 cursor->resid -= bytes;
963 cursor->offset += bytes;
964 /* pagelist offset is always 0 */
965 if (!bytes || cursor->offset & ~PAGE_MASK)
966 return false; /* more bytes to process in the current page */
967
968 /* Move on to the next page */
969
970 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
971 cursor->page = list_entry_next(cursor->page, lru);
972 cursor->last_piece = cursor->resid <= PAGE_SIZE;
973
974 return true;
975}
976
977/*
978 * Message data is handled (sent or received) in pieces, where each
979 * piece resides on a single page. The network layer might not
980 * consume an entire piece at once. A data item's cursor keeps
981 * track of which piece is next to process and how much remains to
982 * be processed in that piece. It also tracks whether the current
983 * piece is the last one in the data item.
984 */
985static void ceph_msg_data_cursor_init(struct ceph_msg_data *data,
986 size_t length)
987{
988 switch (data->type) {
989 case CEPH_MSG_DATA_PAGELIST:
990 ceph_msg_data_pagelist_cursor_init(data, length);
991 break;
992 case CEPH_MSG_DATA_PAGES:
993 ceph_msg_data_pages_cursor_init(data, length);
994 break;
995#ifdef CONFIG_BLOCK
996 case CEPH_MSG_DATA_BIO:
997 ceph_msg_data_bio_cursor_init(data, length);
998 break;
999#endif /* CONFIG_BLOCK */
1000 case CEPH_MSG_DATA_NONE:
1001 default:
1002 /* BUG(); */
1003 break;
1004 }
1005}
1006
1007/*
1008 * Return the page containing the next piece to process for a given
1009 * data item, and supply the page offset and length of that piece.
1010 * Indicate whether this is the last piece in this data item.
1011 */
1012static struct page *ceph_msg_data_next(struct ceph_msg_data *data,
1013 size_t *page_offset,
1014 size_t *length,
1015 bool *last_piece)
1016{
1017 struct page *page;
1018
1019 switch (data->type) {
1020 case CEPH_MSG_DATA_PAGELIST:
1021 page = ceph_msg_data_pagelist_next(data, page_offset, length);
1022 break;
1023 case CEPH_MSG_DATA_PAGES:
1024 page = ceph_msg_data_pages_next(data, page_offset, length);
1025 break;
1026#ifdef CONFIG_BLOCK
1027 case CEPH_MSG_DATA_BIO:
1028 page = ceph_msg_data_bio_next(data, page_offset, length);
1029 break;
1030#endif /* CONFIG_BLOCK */
1031 case CEPH_MSG_DATA_NONE:
1032 default:
1033 page = NULL;
1034 break;
1035 }
1036 BUG_ON(!page);
1037 BUG_ON(*page_offset + *length > PAGE_SIZE);
1038 BUG_ON(!*length);
1039 if (last_piece)
1040 *last_piece = data->cursor.last_piece;
1041
1042 return page;
1043}
1044
1045/*
1046 * Returns true if the result moves the cursor on to the next piece
1047 * of the data item.
1048 */
1049static bool ceph_msg_data_advance(struct ceph_msg_data *data, size_t bytes)
1050{
1051 struct ceph_msg_data_cursor *cursor = &data->cursor;
1052 bool new_piece;
1053
1054 BUG_ON(bytes > cursor->resid);
1055 switch (data->type) {
1056 case CEPH_MSG_DATA_PAGELIST:
1057 new_piece = ceph_msg_data_pagelist_advance(data, bytes);
1058 break;
1059 case CEPH_MSG_DATA_PAGES:
1060 new_piece = ceph_msg_data_pages_advance(data, bytes);
1061 break;
1062#ifdef CONFIG_BLOCK
1063 case CEPH_MSG_DATA_BIO:
1064 new_piece = ceph_msg_data_bio_advance(data, bytes);
1065 break;
1066#endif /* CONFIG_BLOCK */
1067 case CEPH_MSG_DATA_NONE:
1068 default:
1069 BUG();
1070 break;
1071 }
1072
1073 return new_piece;
1074}
1075
1076static void prepare_message_data(struct ceph_msg *msg,
1077 struct ceph_msg_pos *msg_pos)
1078{
1079 size_t data_len;
1080
1081 BUG_ON(!msg);
1082
1083 data_len = le32_to_cpu(msg->hdr.data_len);
1084 BUG_ON(!data_len);
1085
1086 /* initialize page iterator */
1087 msg_pos->page = 0;
1088 if (ceph_msg_has_data(msg))
1089 msg_pos->page_pos = msg->data.alignment;
1090 else
1091 msg_pos->page_pos = 0;
1092 msg_pos->data_pos = 0;
1093
1094 /* Initialize data cursor */
1095
1096 ceph_msg_data_cursor_init(&msg->data, data_len);
1097
1098 msg_pos->did_page_crc = false;
1099}
1100
1101/*
1102 * Prepare footer for currently outgoing message, and finish things
1103 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1104 */
1105static void prepare_write_message_footer(struct ceph_connection *con)
1106{
1107 struct ceph_msg *m = con->out_msg;
1108 int v = con->out_kvec_left;
1109
1110 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1111
1112 dout("prepare_write_message_footer %p\n", con);
1113 con->out_kvec_is_msg = true;
1114 con->out_kvec[v].iov_base = &m->footer;
1115 con->out_kvec[v].iov_len = sizeof(m->footer);
1116 con->out_kvec_bytes += sizeof(m->footer);
1117 con->out_kvec_left++;
1118 con->out_more = m->more_to_follow;
1119 con->out_msg_done = true;
1120}
1121
1122/*
1123 * Prepare headers for the next outgoing message.
1124 */
1125static void prepare_write_message(struct ceph_connection *con)
1126{
1127 struct ceph_msg *m;
1128 u32 crc;
1129
1130 con_out_kvec_reset(con);
1131 con->out_kvec_is_msg = true;
1132 con->out_msg_done = false;
1133
1134 /* Sneak an ack in there first? If we can get it into the same
1135 * TCP packet that's a good thing. */
1136 if (con->in_seq > con->in_seq_acked) {
1137 con->in_seq_acked = con->in_seq;
1138 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1139 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1140 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1141 &con->out_temp_ack);
1142 }
1143
1144 BUG_ON(list_empty(&con->out_queue));
1145 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1146 con->out_msg = m;
1147 BUG_ON(m->con != con);
1148
1149 /* put message on sent list */
1150 ceph_msg_get(m);
1151 list_move_tail(&m->list_head, &con->out_sent);
1152
1153 /*
1154 * only assign outgoing seq # if we haven't sent this message
1155 * yet. if it is requeued, resend with it's original seq.
1156 */
1157 if (m->needs_out_seq) {
1158 m->hdr.seq = cpu_to_le64(++con->out_seq);
1159 m->needs_out_seq = false;
1160 }
1161
1162 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d\n",
1163 m, con->out_seq, le16_to_cpu(m->hdr.type),
1164 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1165 le32_to_cpu(m->hdr.data_len));
1166 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1167
1168 /* tag + hdr + front + middle */
1169 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1170 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1171 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1172
1173 if (m->middle)
1174 con_out_kvec_add(con, m->middle->vec.iov_len,
1175 m->middle->vec.iov_base);
1176
1177 /* fill in crc (except data pages), footer */
1178 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1179 con->out_msg->hdr.crc = cpu_to_le32(crc);
1180 con->out_msg->footer.flags = 0;
1181
1182 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1183 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1184 if (m->middle) {
1185 crc = crc32c(0, m->middle->vec.iov_base,
1186 m->middle->vec.iov_len);
1187 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1188 } else
1189 con->out_msg->footer.middle_crc = 0;
1190 dout("%s front_crc %u middle_crc %u\n", __func__,
1191 le32_to_cpu(con->out_msg->footer.front_crc),
1192 le32_to_cpu(con->out_msg->footer.middle_crc));
1193
1194 /* is there a data payload? */
1195 con->out_msg->footer.data_crc = 0;
1196 if (m->hdr.data_len) {
1197 prepare_message_data(con->out_msg, &con->out_msg_pos);
1198 con->out_more = 1; /* data + footer will follow */
1199 } else {
1200 /* no, queue up footer too and be done */
1201 prepare_write_message_footer(con);
1202 }
1203
1204 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1205}
1206
1207/*
1208 * Prepare an ack.
1209 */
1210static void prepare_write_ack(struct ceph_connection *con)
1211{
1212 dout("prepare_write_ack %p %llu -> %llu\n", con,
1213 con->in_seq_acked, con->in_seq);
1214 con->in_seq_acked = con->in_seq;
1215
1216 con_out_kvec_reset(con);
1217
1218 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1219
1220 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1221 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1222 &con->out_temp_ack);
1223
1224 con->out_more = 1; /* more will follow.. eventually.. */
1225 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1226}
1227
1228/*
1229 * Prepare to share the seq during handshake
1230 */
1231static void prepare_write_seq(struct ceph_connection *con)
1232{
1233 dout("prepare_write_seq %p %llu -> %llu\n", con,
1234 con->in_seq_acked, con->in_seq);
1235 con->in_seq_acked = con->in_seq;
1236
1237 con_out_kvec_reset(con);
1238
1239 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1240 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1241 &con->out_temp_ack);
1242
1243 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1244}
1245
1246/*
1247 * Prepare to write keepalive byte.
1248 */
1249static void prepare_write_keepalive(struct ceph_connection *con)
1250{
1251 dout("prepare_write_keepalive %p\n", con);
1252 con_out_kvec_reset(con);
1253 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
1254 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1255}
1256
1257/*
1258 * Connection negotiation.
1259 */
1260
1261static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1262 int *auth_proto)
1263{
1264 struct ceph_auth_handshake *auth;
1265
1266 if (!con->ops->get_authorizer) {
1267 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1268 con->out_connect.authorizer_len = 0;
1269 return NULL;
1270 }
1271
1272 /* Can't hold the mutex while getting authorizer */
1273 mutex_unlock(&con->mutex);
1274 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
1275 mutex_lock(&con->mutex);
1276
1277 if (IS_ERR(auth))
1278 return auth;
1279 if (con->state != CON_STATE_NEGOTIATING)
1280 return ERR_PTR(-EAGAIN);
1281
1282 con->auth_reply_buf = auth->authorizer_reply_buf;
1283 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
1284 return auth;
1285}
1286
1287/*
1288 * We connected to a peer and are saying hello.
1289 */
1290static void prepare_write_banner(struct ceph_connection *con)
1291{
1292 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1293 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1294 &con->msgr->my_enc_addr);
1295
1296 con->out_more = 0;
1297 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1298}
1299
1300static int prepare_write_connect(struct ceph_connection *con)
1301{
1302 unsigned int global_seq = get_global_seq(con->msgr, 0);
1303 int proto;
1304 int auth_proto;
1305 struct ceph_auth_handshake *auth;
1306
1307 switch (con->peer_name.type) {
1308 case CEPH_ENTITY_TYPE_MON:
1309 proto = CEPH_MONC_PROTOCOL;
1310 break;
1311 case CEPH_ENTITY_TYPE_OSD:
1312 proto = CEPH_OSDC_PROTOCOL;
1313 break;
1314 case CEPH_ENTITY_TYPE_MDS:
1315 proto = CEPH_MDSC_PROTOCOL;
1316 break;
1317 default:
1318 BUG();
1319 }
1320
1321 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1322 con->connect_seq, global_seq, proto);
1323
1324 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
1325 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1326 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1327 con->out_connect.global_seq = cpu_to_le32(global_seq);
1328 con->out_connect.protocol_version = cpu_to_le32(proto);
1329 con->out_connect.flags = 0;
1330
1331 auth_proto = CEPH_AUTH_UNKNOWN;
1332 auth = get_connect_authorizer(con, &auth_proto);
1333 if (IS_ERR(auth))
1334 return PTR_ERR(auth);
1335
1336 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1337 con->out_connect.authorizer_len = auth ?
1338 cpu_to_le32(auth->authorizer_buf_len) : 0;
1339
1340 con_out_kvec_add(con, sizeof (con->out_connect),
1341 &con->out_connect);
1342 if (auth && auth->authorizer_buf_len)
1343 con_out_kvec_add(con, auth->authorizer_buf_len,
1344 auth->authorizer_buf);
1345
1346 con->out_more = 0;
1347 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1348
1349 return 0;
1350}
1351
1352/*
1353 * write as much of pending kvecs to the socket as we can.
1354 * 1 -> done
1355 * 0 -> socket full, but more to do
1356 * <0 -> error
1357 */
1358static int write_partial_kvec(struct ceph_connection *con)
1359{
1360 int ret;
1361
1362 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1363 while (con->out_kvec_bytes > 0) {
1364 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1365 con->out_kvec_left, con->out_kvec_bytes,
1366 con->out_more);
1367 if (ret <= 0)
1368 goto out;
1369 con->out_kvec_bytes -= ret;
1370 if (con->out_kvec_bytes == 0)
1371 break; /* done */
1372
1373 /* account for full iov entries consumed */
1374 while (ret >= con->out_kvec_cur->iov_len) {
1375 BUG_ON(!con->out_kvec_left);
1376 ret -= con->out_kvec_cur->iov_len;
1377 con->out_kvec_cur++;
1378 con->out_kvec_left--;
1379 }
1380 /* and for a partially-consumed entry */
1381 if (ret) {
1382 con->out_kvec_cur->iov_len -= ret;
1383 con->out_kvec_cur->iov_base += ret;
1384 }
1385 }
1386 con->out_kvec_left = 0;
1387 con->out_kvec_is_msg = false;
1388 ret = 1;
1389out:
1390 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1391 con->out_kvec_bytes, con->out_kvec_left, ret);
1392 return ret; /* done! */
1393}
1394
1395static void out_msg_pos_next(struct ceph_connection *con, struct page *page,
1396 size_t len, size_t sent)
1397{
1398 struct ceph_msg *msg = con->out_msg;
1399 struct ceph_msg_pos *msg_pos = &con->out_msg_pos;
1400 bool need_crc = false;
1401
1402 BUG_ON(!msg);
1403 BUG_ON(!sent);
1404
1405 msg_pos->data_pos += sent;
1406 msg_pos->page_pos += sent;
1407 need_crc = ceph_msg_data_advance(&msg->data, sent);
1408 BUG_ON(need_crc && sent != len);
1409
1410 if (sent < len)
1411 return;
1412
1413 BUG_ON(sent != len);
1414 msg_pos->page_pos = 0;
1415 msg_pos->page++;
1416 msg_pos->did_page_crc = false;
1417}
1418
1419static void in_msg_pos_next(struct ceph_connection *con, size_t len,
1420 size_t received)
1421{
1422 struct ceph_msg *msg = con->in_msg;
1423 struct ceph_msg_pos *msg_pos = &con->in_msg_pos;
1424
1425 BUG_ON(!msg);
1426 BUG_ON(!received);
1427
1428 msg_pos->data_pos += received;
1429 msg_pos->page_pos += received;
1430 (void) ceph_msg_data_advance(&msg->data, received);
1431
1432 if (received < len)
1433 return;
1434
1435 BUG_ON(received != len);
1436 msg_pos->page_pos = 0;
1437 msg_pos->page++;
1438}
1439
1440static u32 ceph_crc32c_page(u32 crc, struct page *page,
1441 unsigned int page_offset,
1442 unsigned int length)
1443{
1444 char *kaddr;
1445
1446 kaddr = kmap(page);
1447 BUG_ON(kaddr == NULL);
1448 crc = crc32c(crc, kaddr + page_offset, length);
1449 kunmap(page);
1450
1451 return crc;
1452}
1453/*
1454 * Write as much message data payload as we can. If we finish, queue
1455 * up the footer.
1456 * 1 -> done, footer is now queued in out_kvec[].
1457 * 0 -> socket full, but more to do
1458 * <0 -> error
1459 */
1460static int write_partial_message_data(struct ceph_connection *con)
1461{
1462 struct ceph_msg *msg = con->out_msg;
1463 struct ceph_msg_pos *msg_pos = &con->out_msg_pos;
1464 unsigned int data_len = le32_to_cpu(msg->hdr.data_len);
1465 bool do_datacrc = !con->msgr->nocrc;
1466 int ret;
1467
1468 dout("%s %p msg %p page %d offset %d\n", __func__,
1469 con, msg, msg_pos->page, msg_pos->page_pos);
1470
1471 if (WARN_ON(!ceph_msg_has_data(msg)))
1472 return -EINVAL;
1473
1474 /*
1475 * Iterate through each page that contains data to be
1476 * written, and send as much as possible for each.
1477 *
1478 * If we are calculating the data crc (the default), we will
1479 * need to map the page. If we have no pages, they have
1480 * been revoked, so use the zero page.
1481 */
1482 while (data_len > msg_pos->data_pos) {
1483 struct page *page;
1484 size_t page_offset;
1485 size_t length;
1486 bool last_piece;
1487
1488 page = ceph_msg_data_next(&msg->data, &page_offset, &length,
1489 &last_piece);
1490 if (do_datacrc && !msg_pos->did_page_crc) {
1491 u32 crc = le32_to_cpu(msg->footer.data_crc);
1492
1493 crc = ceph_crc32c_page(crc, page, page_offset, length);
1494 msg->footer.data_crc = cpu_to_le32(crc);
1495 msg_pos->did_page_crc = true;
1496 }
1497 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
1498 length, last_piece);
1499 if (ret <= 0)
1500 goto out;
1501
1502 out_msg_pos_next(con, page, length, (size_t) ret);
1503 }
1504
1505 dout("%s %p msg %p done\n", __func__, con, msg);
1506
1507 /* prepare and queue up footer, too */
1508 if (!do_datacrc)
1509 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1510 con_out_kvec_reset(con);
1511 prepare_write_message_footer(con);
1512 ret = 1;
1513out:
1514 return ret;
1515}
1516
1517/*
1518 * write some zeros
1519 */
1520static int write_partial_skip(struct ceph_connection *con)
1521{
1522 int ret;
1523
1524 while (con->out_skip > 0) {
1525 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
1526
1527 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
1528 if (ret <= 0)
1529 goto out;
1530 con->out_skip -= ret;
1531 }
1532 ret = 1;
1533out:
1534 return ret;
1535}
1536
1537/*
1538 * Prepare to read connection handshake, or an ack.
1539 */
1540static void prepare_read_banner(struct ceph_connection *con)
1541{
1542 dout("prepare_read_banner %p\n", con);
1543 con->in_base_pos = 0;
1544}
1545
1546static void prepare_read_connect(struct ceph_connection *con)
1547{
1548 dout("prepare_read_connect %p\n", con);
1549 con->in_base_pos = 0;
1550}
1551
1552static void prepare_read_ack(struct ceph_connection *con)
1553{
1554 dout("prepare_read_ack %p\n", con);
1555 con->in_base_pos = 0;
1556}
1557
1558static void prepare_read_seq(struct ceph_connection *con)
1559{
1560 dout("prepare_read_seq %p\n", con);
1561 con->in_base_pos = 0;
1562 con->in_tag = CEPH_MSGR_TAG_SEQ;
1563}
1564
1565static void prepare_read_tag(struct ceph_connection *con)
1566{
1567 dout("prepare_read_tag %p\n", con);
1568 con->in_base_pos = 0;
1569 con->in_tag = CEPH_MSGR_TAG_READY;
1570}
1571
1572/*
1573 * Prepare to read a message.
1574 */
1575static int prepare_read_message(struct ceph_connection *con)
1576{
1577 dout("prepare_read_message %p\n", con);
1578 BUG_ON(con->in_msg != NULL);
1579 con->in_base_pos = 0;
1580 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1581 return 0;
1582}
1583
1584
1585static int read_partial(struct ceph_connection *con,
1586 int end, int size, void *object)
1587{
1588 while (con->in_base_pos < end) {
1589 int left = end - con->in_base_pos;
1590 int have = size - left;
1591 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1592 if (ret <= 0)
1593 return ret;
1594 con->in_base_pos += ret;
1595 }
1596 return 1;
1597}
1598
1599
1600/*
1601 * Read all or part of the connect-side handshake on a new connection
1602 */
1603static int read_partial_banner(struct ceph_connection *con)
1604{
1605 int size;
1606 int end;
1607 int ret;
1608
1609 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1610
1611 /* peer's banner */
1612 size = strlen(CEPH_BANNER);
1613 end = size;
1614 ret = read_partial(con, end, size, con->in_banner);
1615 if (ret <= 0)
1616 goto out;
1617
1618 size = sizeof (con->actual_peer_addr);
1619 end += size;
1620 ret = read_partial(con, end, size, &con->actual_peer_addr);
1621 if (ret <= 0)
1622 goto out;
1623
1624 size = sizeof (con->peer_addr_for_me);
1625 end += size;
1626 ret = read_partial(con, end, size, &con->peer_addr_for_me);
1627 if (ret <= 0)
1628 goto out;
1629
1630out:
1631 return ret;
1632}
1633
1634static int read_partial_connect(struct ceph_connection *con)
1635{
1636 int size;
1637 int end;
1638 int ret;
1639
1640 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1641
1642 size = sizeof (con->in_reply);
1643 end = size;
1644 ret = read_partial(con, end, size, &con->in_reply);
1645 if (ret <= 0)
1646 goto out;
1647
1648 size = le32_to_cpu(con->in_reply.authorizer_len);
1649 end += size;
1650 ret = read_partial(con, end, size, con->auth_reply_buf);
1651 if (ret <= 0)
1652 goto out;
1653
1654 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1655 con, (int)con->in_reply.tag,
1656 le32_to_cpu(con->in_reply.connect_seq),
1657 le32_to_cpu(con->in_reply.global_seq));
1658out:
1659 return ret;
1660
1661}
1662
1663/*
1664 * Verify the hello banner looks okay.
1665 */
1666static int verify_hello(struct ceph_connection *con)
1667{
1668 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1669 pr_err("connect to %s got bad banner\n",
1670 ceph_pr_addr(&con->peer_addr.in_addr));
1671 con->error_msg = "protocol error, bad banner";
1672 return -1;
1673 }
1674 return 0;
1675}
1676
1677static bool addr_is_blank(struct sockaddr_storage *ss)
1678{
1679 switch (ss->ss_family) {
1680 case AF_INET:
1681 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1682 case AF_INET6:
1683 return
1684 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1685 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1686 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1687 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1688 }
1689 return false;
1690}
1691
1692static int addr_port(struct sockaddr_storage *ss)
1693{
1694 switch (ss->ss_family) {
1695 case AF_INET:
1696 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1697 case AF_INET6:
1698 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1699 }
1700 return 0;
1701}
1702
1703static void addr_set_port(struct sockaddr_storage *ss, int p)
1704{
1705 switch (ss->ss_family) {
1706 case AF_INET:
1707 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1708 break;
1709 case AF_INET6:
1710 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1711 break;
1712 }
1713}
1714
1715/*
1716 * Unlike other *_pton function semantics, zero indicates success.
1717 */
1718static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1719 char delim, const char **ipend)
1720{
1721 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1722 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1723
1724 memset(ss, 0, sizeof(*ss));
1725
1726 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1727 ss->ss_family = AF_INET;
1728 return 0;
1729 }
1730
1731 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1732 ss->ss_family = AF_INET6;
1733 return 0;
1734 }
1735
1736 return -EINVAL;
1737}
1738
1739/*
1740 * Extract hostname string and resolve using kernel DNS facility.
1741 */
1742#ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1743static int ceph_dns_resolve_name(const char *name, size_t namelen,
1744 struct sockaddr_storage *ss, char delim, const char **ipend)
1745{
1746 const char *end, *delim_p;
1747 char *colon_p, *ip_addr = NULL;
1748 int ip_len, ret;
1749
1750 /*
1751 * The end of the hostname occurs immediately preceding the delimiter or
1752 * the port marker (':') where the delimiter takes precedence.
1753 */
1754 delim_p = memchr(name, delim, namelen);
1755 colon_p = memchr(name, ':', namelen);
1756
1757 if (delim_p && colon_p)
1758 end = delim_p < colon_p ? delim_p : colon_p;
1759 else if (!delim_p && colon_p)
1760 end = colon_p;
1761 else {
1762 end = delim_p;
1763 if (!end) /* case: hostname:/ */
1764 end = name + namelen;
1765 }
1766
1767 if (end <= name)
1768 return -EINVAL;
1769
1770 /* do dns_resolve upcall */
1771 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1772 if (ip_len > 0)
1773 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1774 else
1775 ret = -ESRCH;
1776
1777 kfree(ip_addr);
1778
1779 *ipend = end;
1780
1781 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1782 ret, ret ? "failed" : ceph_pr_addr(ss));
1783
1784 return ret;
1785}
1786#else
1787static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1788 struct sockaddr_storage *ss, char delim, const char **ipend)
1789{
1790 return -EINVAL;
1791}
1792#endif
1793
1794/*
1795 * Parse a server name (IP or hostname). If a valid IP address is not found
1796 * then try to extract a hostname to resolve using userspace DNS upcall.
1797 */
1798static int ceph_parse_server_name(const char *name, size_t namelen,
1799 struct sockaddr_storage *ss, char delim, const char **ipend)
1800{
1801 int ret;
1802
1803 ret = ceph_pton(name, namelen, ss, delim, ipend);
1804 if (ret)
1805 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1806
1807 return ret;
1808}
1809
1810/*
1811 * Parse an ip[:port] list into an addr array. Use the default
1812 * monitor port if a port isn't specified.
1813 */
1814int ceph_parse_ips(const char *c, const char *end,
1815 struct ceph_entity_addr *addr,
1816 int max_count, int *count)
1817{
1818 int i, ret = -EINVAL;
1819 const char *p = c;
1820
1821 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1822 for (i = 0; i < max_count; i++) {
1823 const char *ipend;
1824 struct sockaddr_storage *ss = &addr[i].in_addr;
1825 int port;
1826 char delim = ',';
1827
1828 if (*p == '[') {
1829 delim = ']';
1830 p++;
1831 }
1832
1833 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1834 if (ret)
1835 goto bad;
1836 ret = -EINVAL;
1837
1838 p = ipend;
1839
1840 if (delim == ']') {
1841 if (*p != ']') {
1842 dout("missing matching ']'\n");
1843 goto bad;
1844 }
1845 p++;
1846 }
1847
1848 /* port? */
1849 if (p < end && *p == ':') {
1850 port = 0;
1851 p++;
1852 while (p < end && *p >= '0' && *p <= '9') {
1853 port = (port * 10) + (*p - '0');
1854 p++;
1855 }
1856 if (port > 65535 || port == 0)
1857 goto bad;
1858 } else {
1859 port = CEPH_MON_PORT;
1860 }
1861
1862 addr_set_port(ss, port);
1863
1864 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1865
1866 if (p == end)
1867 break;
1868 if (*p != ',')
1869 goto bad;
1870 p++;
1871 }
1872
1873 if (p != end)
1874 goto bad;
1875
1876 if (count)
1877 *count = i + 1;
1878 return 0;
1879
1880bad:
1881 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1882 return ret;
1883}
1884EXPORT_SYMBOL(ceph_parse_ips);
1885
1886static int process_banner(struct ceph_connection *con)
1887{
1888 dout("process_banner on %p\n", con);
1889
1890 if (verify_hello(con) < 0)
1891 return -1;
1892
1893 ceph_decode_addr(&con->actual_peer_addr);
1894 ceph_decode_addr(&con->peer_addr_for_me);
1895
1896 /*
1897 * Make sure the other end is who we wanted. note that the other
1898 * end may not yet know their ip address, so if it's 0.0.0.0, give
1899 * them the benefit of the doubt.
1900 */
1901 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1902 sizeof(con->peer_addr)) != 0 &&
1903 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1904 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1905 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1906 ceph_pr_addr(&con->peer_addr.in_addr),
1907 (int)le32_to_cpu(con->peer_addr.nonce),
1908 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1909 (int)le32_to_cpu(con->actual_peer_addr.nonce));
1910 con->error_msg = "wrong peer at address";
1911 return -1;
1912 }
1913
1914 /*
1915 * did we learn our address?
1916 */
1917 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1918 int port = addr_port(&con->msgr->inst.addr.in_addr);
1919
1920 memcpy(&con->msgr->inst.addr.in_addr,
1921 &con->peer_addr_for_me.in_addr,
1922 sizeof(con->peer_addr_for_me.in_addr));
1923 addr_set_port(&con->msgr->inst.addr.in_addr, port);
1924 encode_my_addr(con->msgr);
1925 dout("process_banner learned my addr is %s\n",
1926 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
1927 }
1928
1929 return 0;
1930}
1931
1932static int process_connect(struct ceph_connection *con)
1933{
1934 u64 sup_feat = con->msgr->supported_features;
1935 u64 req_feat = con->msgr->required_features;
1936 u64 server_feat = le64_to_cpu(con->in_reply.features);
1937 int ret;
1938
1939 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1940
1941 switch (con->in_reply.tag) {
1942 case CEPH_MSGR_TAG_FEATURES:
1943 pr_err("%s%lld %s feature set mismatch,"
1944 " my %llx < server's %llx, missing %llx\n",
1945 ENTITY_NAME(con->peer_name),
1946 ceph_pr_addr(&con->peer_addr.in_addr),
1947 sup_feat, server_feat, server_feat & ~sup_feat);
1948 con->error_msg = "missing required protocol features";
1949 reset_connection(con);
1950 return -1;
1951
1952 case CEPH_MSGR_TAG_BADPROTOVER:
1953 pr_err("%s%lld %s protocol version mismatch,"
1954 " my %d != server's %d\n",
1955 ENTITY_NAME(con->peer_name),
1956 ceph_pr_addr(&con->peer_addr.in_addr),
1957 le32_to_cpu(con->out_connect.protocol_version),
1958 le32_to_cpu(con->in_reply.protocol_version));
1959 con->error_msg = "protocol version mismatch";
1960 reset_connection(con);
1961 return -1;
1962
1963 case CEPH_MSGR_TAG_BADAUTHORIZER:
1964 con->auth_retry++;
1965 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1966 con->auth_retry);
1967 if (con->auth_retry == 2) {
1968 con->error_msg = "connect authorization failure";
1969 return -1;
1970 }
1971 con_out_kvec_reset(con);
1972 ret = prepare_write_connect(con);
1973 if (ret < 0)
1974 return ret;
1975 prepare_read_connect(con);
1976 break;
1977
1978 case CEPH_MSGR_TAG_RESETSESSION:
1979 /*
1980 * If we connected with a large connect_seq but the peer
1981 * has no record of a session with us (no connection, or
1982 * connect_seq == 0), they will send RESETSESION to indicate
1983 * that they must have reset their session, and may have
1984 * dropped messages.
1985 */
1986 dout("process_connect got RESET peer seq %u\n",
1987 le32_to_cpu(con->in_reply.connect_seq));
1988 pr_err("%s%lld %s connection reset\n",
1989 ENTITY_NAME(con->peer_name),
1990 ceph_pr_addr(&con->peer_addr.in_addr));
1991 reset_connection(con);
1992 con_out_kvec_reset(con);
1993 ret = prepare_write_connect(con);
1994 if (ret < 0)
1995 return ret;
1996 prepare_read_connect(con);
1997
1998 /* Tell ceph about it. */
1999 mutex_unlock(&con->mutex);
2000 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2001 if (con->ops->peer_reset)
2002 con->ops->peer_reset(con);
2003 mutex_lock(&con->mutex);
2004 if (con->state != CON_STATE_NEGOTIATING)
2005 return -EAGAIN;
2006 break;
2007
2008 case CEPH_MSGR_TAG_RETRY_SESSION:
2009 /*
2010 * If we sent a smaller connect_seq than the peer has, try
2011 * again with a larger value.
2012 */
2013 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2014 le32_to_cpu(con->out_connect.connect_seq),
2015 le32_to_cpu(con->in_reply.connect_seq));
2016 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2017 con_out_kvec_reset(con);
2018 ret = prepare_write_connect(con);
2019 if (ret < 0)
2020 return ret;
2021 prepare_read_connect(con);
2022 break;
2023
2024 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2025 /*
2026 * If we sent a smaller global_seq than the peer has, try
2027 * again with a larger value.
2028 */
2029 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2030 con->peer_global_seq,
2031 le32_to_cpu(con->in_reply.global_seq));
2032 get_global_seq(con->msgr,
2033 le32_to_cpu(con->in_reply.global_seq));
2034 con_out_kvec_reset(con);
2035 ret = prepare_write_connect(con);
2036 if (ret < 0)
2037 return ret;
2038 prepare_read_connect(con);
2039 break;
2040
2041 case CEPH_MSGR_TAG_SEQ:
2042 case CEPH_MSGR_TAG_READY:
2043 if (req_feat & ~server_feat) {
2044 pr_err("%s%lld %s protocol feature mismatch,"
2045 " my required %llx > server's %llx, need %llx\n",
2046 ENTITY_NAME(con->peer_name),
2047 ceph_pr_addr(&con->peer_addr.in_addr),
2048 req_feat, server_feat, req_feat & ~server_feat);
2049 con->error_msg = "missing required protocol features";
2050 reset_connection(con);
2051 return -1;
2052 }
2053
2054 WARN_ON(con->state != CON_STATE_NEGOTIATING);
2055 con->state = CON_STATE_OPEN;
2056 con->auth_retry = 0; /* we authenticated; clear flag */
2057 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2058 con->connect_seq++;
2059 con->peer_features = server_feat;
2060 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2061 con->peer_global_seq,
2062 le32_to_cpu(con->in_reply.connect_seq),
2063 con->connect_seq);
2064 WARN_ON(con->connect_seq !=
2065 le32_to_cpu(con->in_reply.connect_seq));
2066
2067 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2068 con_flag_set(con, CON_FLAG_LOSSYTX);
2069
2070 con->delay = 0; /* reset backoff memory */
2071
2072 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2073 prepare_write_seq(con);
2074 prepare_read_seq(con);
2075 } else {
2076 prepare_read_tag(con);
2077 }
2078 break;
2079
2080 case CEPH_MSGR_TAG_WAIT:
2081 /*
2082 * If there is a connection race (we are opening
2083 * connections to each other), one of us may just have
2084 * to WAIT. This shouldn't happen if we are the
2085 * client.
2086 */
2087 pr_err("process_connect got WAIT as client\n");
2088 con->error_msg = "protocol error, got WAIT as client";
2089 return -1;
2090
2091 default:
2092 pr_err("connect protocol error, will retry\n");
2093 con->error_msg = "protocol error, garbage tag during connect";
2094 return -1;
2095 }
2096 return 0;
2097}
2098
2099
2100/*
2101 * read (part of) an ack
2102 */
2103static int read_partial_ack(struct ceph_connection *con)
2104{
2105 int size = sizeof (con->in_temp_ack);
2106 int end = size;
2107
2108 return read_partial(con, end, size, &con->in_temp_ack);
2109}
2110
2111/*
2112 * We can finally discard anything that's been acked.
2113 */
2114static void process_ack(struct ceph_connection *con)
2115{
2116 struct ceph_msg *m;
2117 u64 ack = le64_to_cpu(con->in_temp_ack);
2118 u64 seq;
2119
2120 while (!list_empty(&con->out_sent)) {
2121 m = list_first_entry(&con->out_sent, struct ceph_msg,
2122 list_head);
2123 seq = le64_to_cpu(m->hdr.seq);
2124 if (seq > ack)
2125 break;
2126 dout("got ack for seq %llu type %d at %p\n", seq,
2127 le16_to_cpu(m->hdr.type), m);
2128 m->ack_stamp = jiffies;
2129 ceph_msg_remove(m);
2130 }
2131 prepare_read_tag(con);
2132}
2133
2134
2135static int read_partial_message_section(struct ceph_connection *con,
2136 struct kvec *section,
2137 unsigned int sec_len, u32 *crc)
2138{
2139 int ret, left;
2140
2141 BUG_ON(!section);
2142
2143 while (section->iov_len < sec_len) {
2144 BUG_ON(section->iov_base == NULL);
2145 left = sec_len - section->iov_len;
2146 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2147 section->iov_len, left);
2148 if (ret <= 0)
2149 return ret;
2150 section->iov_len += ret;
2151 }
2152 if (section->iov_len == sec_len)
2153 *crc = crc32c(0, section->iov_base, section->iov_len);
2154
2155 return 1;
2156}
2157
2158static int read_partial_msg_data(struct ceph_connection *con)
2159{
2160 struct ceph_msg *msg = con->in_msg;
2161 struct ceph_msg_pos *msg_pos = &con->in_msg_pos;
2162 const bool do_datacrc = !con->msgr->nocrc;
2163 unsigned int data_len;
2164 struct page *page;
2165 size_t page_offset;
2166 size_t length;
2167 int ret;
2168
2169 BUG_ON(!msg);
2170 if (WARN_ON(!ceph_msg_has_data(msg)))
2171 return -EIO;
2172
2173 data_len = le32_to_cpu(con->in_hdr.data_len);
2174 while (msg_pos->data_pos < data_len) {
2175 page = ceph_msg_data_next(&msg->data, &page_offset, &length,
2176 NULL);
2177 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2178 if (ret <= 0)
2179 return ret;
2180
2181 if (do_datacrc)
2182 con->in_data_crc = ceph_crc32c_page(con->in_data_crc,
2183 page, page_offset, ret);
2184 in_msg_pos_next(con, length, ret);
2185 }
2186
2187 return 1; /* must return > 0 to indicate success */
2188}
2189
2190/*
2191 * read (part of) a message.
2192 */
2193static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2194
2195static int read_partial_message(struct ceph_connection *con)
2196{
2197 struct ceph_msg *m = con->in_msg;
2198 int size;
2199 int end;
2200 int ret;
2201 unsigned int front_len, middle_len, data_len;
2202 bool do_datacrc = !con->msgr->nocrc;
2203 u64 seq;
2204 u32 crc;
2205
2206 dout("read_partial_message con %p msg %p\n", con, m);
2207
2208 /* header */
2209 size = sizeof (con->in_hdr);
2210 end = size;
2211 ret = read_partial(con, end, size, &con->in_hdr);
2212 if (ret <= 0)
2213 return ret;
2214
2215 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2216 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2217 pr_err("read_partial_message bad hdr "
2218 " crc %u != expected %u\n",
2219 crc, con->in_hdr.crc);
2220 return -EBADMSG;
2221 }
2222
2223 front_len = le32_to_cpu(con->in_hdr.front_len);
2224 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2225 return -EIO;
2226 middle_len = le32_to_cpu(con->in_hdr.middle_len);
2227 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2228 return -EIO;
2229 data_len = le32_to_cpu(con->in_hdr.data_len);
2230 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2231 return -EIO;
2232
2233 /* verify seq# */
2234 seq = le64_to_cpu(con->in_hdr.seq);
2235 if ((s64)seq - (s64)con->in_seq < 1) {
2236 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2237 ENTITY_NAME(con->peer_name),
2238 ceph_pr_addr(&con->peer_addr.in_addr),
2239 seq, con->in_seq + 1);
2240 con->in_base_pos = -front_len - middle_len - data_len -
2241 sizeof(m->footer);
2242 con->in_tag = CEPH_MSGR_TAG_READY;
2243 return 0;
2244 } else if ((s64)seq - (s64)con->in_seq > 1) {
2245 pr_err("read_partial_message bad seq %lld expected %lld\n",
2246 seq, con->in_seq + 1);
2247 con->error_msg = "bad message sequence # for incoming message";
2248 return -EBADMSG;
2249 }
2250
2251 /* allocate message? */
2252 if (!con->in_msg) {
2253 int skip = 0;
2254
2255 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2256 front_len, data_len);
2257 ret = ceph_con_in_msg_alloc(con, &skip);
2258 if (ret < 0)
2259 return ret;
2260 if (skip) {
2261 /* skip this message */
2262 dout("alloc_msg said skip message\n");
2263 BUG_ON(con->in_msg);
2264 con->in_base_pos = -front_len - middle_len - data_len -
2265 sizeof(m->footer);
2266 con->in_tag = CEPH_MSGR_TAG_READY;
2267 con->in_seq++;
2268 return 0;
2269 }
2270
2271 BUG_ON(!con->in_msg);
2272 BUG_ON(con->in_msg->con != con);
2273 m = con->in_msg;
2274 m->front.iov_len = 0; /* haven't read it yet */
2275 if (m->middle)
2276 m->middle->vec.iov_len = 0;
2277
2278 /* prepare for data payload, if any */
2279
2280 if (data_len)
2281 prepare_message_data(con->in_msg, &con->in_msg_pos);
2282 }
2283
2284 /* front */
2285 ret = read_partial_message_section(con, &m->front, front_len,
2286 &con->in_front_crc);
2287 if (ret <= 0)
2288 return ret;
2289
2290 /* middle */
2291 if (m->middle) {
2292 ret = read_partial_message_section(con, &m->middle->vec,
2293 middle_len,
2294 &con->in_middle_crc);
2295 if (ret <= 0)
2296 return ret;
2297 }
2298
2299 /* (page) data */
2300 if (data_len) {
2301 ret = read_partial_msg_data(con);
2302 if (ret <= 0)
2303 return ret;
2304 }
2305
2306 /* footer */
2307 size = sizeof (m->footer);
2308 end += size;
2309 ret = read_partial(con, end, size, &m->footer);
2310 if (ret <= 0)
2311 return ret;
2312
2313 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2314 m, front_len, m->footer.front_crc, middle_len,
2315 m->footer.middle_crc, data_len, m->footer.data_crc);
2316
2317 /* crc ok? */
2318 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2319 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2320 m, con->in_front_crc, m->footer.front_crc);
2321 return -EBADMSG;
2322 }
2323 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2324 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2325 m, con->in_middle_crc, m->footer.middle_crc);
2326 return -EBADMSG;
2327 }
2328 if (do_datacrc &&
2329 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2330 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2331 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2332 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2333 return -EBADMSG;
2334 }
2335
2336 return 1; /* done! */
2337}
2338
2339/*
2340 * Process message. This happens in the worker thread. The callback should
2341 * be careful not to do anything that waits on other incoming messages or it
2342 * may deadlock.
2343 */
2344static void process_message(struct ceph_connection *con)
2345{
2346 struct ceph_msg *msg;
2347
2348 BUG_ON(con->in_msg->con != con);
2349 con->in_msg->con = NULL;
2350 msg = con->in_msg;
2351 con->in_msg = NULL;
2352 con->ops->put(con);
2353
2354 /* if first message, set peer_name */
2355 if (con->peer_name.type == 0)
2356 con->peer_name = msg->hdr.src;
2357
2358 con->in_seq++;
2359 mutex_unlock(&con->mutex);
2360
2361 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2362 msg, le64_to_cpu(msg->hdr.seq),
2363 ENTITY_NAME(msg->hdr.src),
2364 le16_to_cpu(msg->hdr.type),
2365 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2366 le32_to_cpu(msg->hdr.front_len),
2367 le32_to_cpu(msg->hdr.data_len),
2368 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2369 con->ops->dispatch(con, msg);
2370
2371 mutex_lock(&con->mutex);
2372}
2373
2374
2375/*
2376 * Write something to the socket. Called in a worker thread when the
2377 * socket appears to be writeable and we have something ready to send.
2378 */
2379static int try_write(struct ceph_connection *con)
2380{
2381 int ret = 1;
2382
2383 dout("try_write start %p state %lu\n", con, con->state);
2384
2385more:
2386 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2387
2388 /* open the socket first? */
2389 if (con->state == CON_STATE_PREOPEN) {
2390 BUG_ON(con->sock);
2391 con->state = CON_STATE_CONNECTING;
2392
2393 con_out_kvec_reset(con);
2394 prepare_write_banner(con);
2395 prepare_read_banner(con);
2396
2397 BUG_ON(con->in_msg);
2398 con->in_tag = CEPH_MSGR_TAG_READY;
2399 dout("try_write initiating connect on %p new state %lu\n",
2400 con, con->state);
2401 ret = ceph_tcp_connect(con);
2402 if (ret < 0) {
2403 con->error_msg = "connect error";
2404 goto out;
2405 }
2406 }
2407
2408more_kvec:
2409 /* kvec data queued? */
2410 if (con->out_skip) {
2411 ret = write_partial_skip(con);
2412 if (ret <= 0)
2413 goto out;
2414 }
2415 if (con->out_kvec_left) {
2416 ret = write_partial_kvec(con);
2417 if (ret <= 0)
2418 goto out;
2419 }
2420
2421 /* msg pages? */
2422 if (con->out_msg) {
2423 if (con->out_msg_done) {
2424 ceph_msg_put(con->out_msg);
2425 con->out_msg = NULL; /* we're done with this one */
2426 goto do_next;
2427 }
2428
2429 ret = write_partial_message_data(con);
2430 if (ret == 1)
2431 goto more_kvec; /* we need to send the footer, too! */
2432 if (ret == 0)
2433 goto out;
2434 if (ret < 0) {
2435 dout("try_write write_partial_message_data err %d\n",
2436 ret);
2437 goto out;
2438 }
2439 }
2440
2441do_next:
2442 if (con->state == CON_STATE_OPEN) {
2443 /* is anything else pending? */
2444 if (!list_empty(&con->out_queue)) {
2445 prepare_write_message(con);
2446 goto more;
2447 }
2448 if (con->in_seq > con->in_seq_acked) {
2449 prepare_write_ack(con);
2450 goto more;
2451 }
2452 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2453 prepare_write_keepalive(con);
2454 goto more;
2455 }
2456 }
2457
2458 /* Nothing to do! */
2459 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2460 dout("try_write nothing else to write.\n");
2461 ret = 0;
2462out:
2463 dout("try_write done on %p ret %d\n", con, ret);
2464 return ret;
2465}
2466
2467
2468
2469/*
2470 * Read what we can from the socket.
2471 */
2472static int try_read(struct ceph_connection *con)
2473{
2474 int ret = -1;
2475
2476more:
2477 dout("try_read start on %p state %lu\n", con, con->state);
2478 if (con->state != CON_STATE_CONNECTING &&
2479 con->state != CON_STATE_NEGOTIATING &&
2480 con->state != CON_STATE_OPEN)
2481 return 0;
2482
2483 BUG_ON(!con->sock);
2484
2485 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2486 con->in_base_pos);
2487
2488 if (con->state == CON_STATE_CONNECTING) {
2489 dout("try_read connecting\n");
2490 ret = read_partial_banner(con);
2491 if (ret <= 0)
2492 goto out;
2493 ret = process_banner(con);
2494 if (ret < 0)
2495 goto out;
2496
2497 con->state = CON_STATE_NEGOTIATING;
2498
2499 /*
2500 * Received banner is good, exchange connection info.
2501 * Do not reset out_kvec, as sending our banner raced
2502 * with receiving peer banner after connect completed.
2503 */
2504 ret = prepare_write_connect(con);
2505 if (ret < 0)
2506 goto out;
2507 prepare_read_connect(con);
2508
2509 /* Send connection info before awaiting response */
2510 goto out;
2511 }
2512
2513 if (con->state == CON_STATE_NEGOTIATING) {
2514 dout("try_read negotiating\n");
2515 ret = read_partial_connect(con);
2516 if (ret <= 0)
2517 goto out;
2518 ret = process_connect(con);
2519 if (ret < 0)
2520 goto out;
2521 goto more;
2522 }
2523
2524 WARN_ON(con->state != CON_STATE_OPEN);
2525
2526 if (con->in_base_pos < 0) {
2527 /*
2528 * skipping + discarding content.
2529 *
2530 * FIXME: there must be a better way to do this!
2531 */
2532 static char buf[SKIP_BUF_SIZE];
2533 int skip = min((int) sizeof (buf), -con->in_base_pos);
2534
2535 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2536 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2537 if (ret <= 0)
2538 goto out;
2539 con->in_base_pos += ret;
2540 if (con->in_base_pos)
2541 goto more;
2542 }
2543 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2544 /*
2545 * what's next?
2546 */
2547 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2548 if (ret <= 0)
2549 goto out;
2550 dout("try_read got tag %d\n", (int)con->in_tag);
2551 switch (con->in_tag) {
2552 case CEPH_MSGR_TAG_MSG:
2553 prepare_read_message(con);
2554 break;
2555 case CEPH_MSGR_TAG_ACK:
2556 prepare_read_ack(con);
2557 break;
2558 case CEPH_MSGR_TAG_CLOSE:
2559 con_close_socket(con);
2560 con->state = CON_STATE_CLOSED;
2561 goto out;
2562 default:
2563 goto bad_tag;
2564 }
2565 }
2566 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2567 ret = read_partial_message(con);
2568 if (ret <= 0) {
2569 switch (ret) {
2570 case -EBADMSG:
2571 con->error_msg = "bad crc";
2572 ret = -EIO;
2573 break;
2574 case -EIO:
2575 con->error_msg = "io error";
2576 break;
2577 }
2578 goto out;
2579 }
2580 if (con->in_tag == CEPH_MSGR_TAG_READY)
2581 goto more;
2582 process_message(con);
2583 if (con->state == CON_STATE_OPEN)
2584 prepare_read_tag(con);
2585 goto more;
2586 }
2587 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2588 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2589 /*
2590 * the final handshake seq exchange is semantically
2591 * equivalent to an ACK
2592 */
2593 ret = read_partial_ack(con);
2594 if (ret <= 0)
2595 goto out;
2596 process_ack(con);
2597 goto more;
2598 }
2599
2600out:
2601 dout("try_read done on %p ret %d\n", con, ret);
2602 return ret;
2603
2604bad_tag:
2605 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2606 con->error_msg = "protocol error, garbage tag";
2607 ret = -1;
2608 goto out;
2609}
2610
2611
2612/*
2613 * Atomically queue work on a connection after the specified delay.
2614 * Bump @con reference to avoid races with connection teardown.
2615 * Returns 0 if work was queued, or an error code otherwise.
2616 */
2617static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2618{
2619 if (!con->ops->get(con)) {
2620 dout("%s %p ref count 0\n", __func__, con);
2621
2622 return -ENOENT;
2623 }
2624
2625 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2626 dout("%s %p - already queued\n", __func__, con);
2627 con->ops->put(con);
2628
2629 return -EBUSY;
2630 }
2631
2632 dout("%s %p %lu\n", __func__, con, delay);
2633
2634 return 0;
2635}
2636
2637static void queue_con(struct ceph_connection *con)
2638{
2639 (void) queue_con_delay(con, 0);
2640}
2641
2642static bool con_sock_closed(struct ceph_connection *con)
2643{
2644 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2645 return false;
2646
2647#define CASE(x) \
2648 case CON_STATE_ ## x: \
2649 con->error_msg = "socket closed (con state " #x ")"; \
2650 break;
2651
2652 switch (con->state) {
2653 CASE(CLOSED);
2654 CASE(PREOPEN);
2655 CASE(CONNECTING);
2656 CASE(NEGOTIATING);
2657 CASE(OPEN);
2658 CASE(STANDBY);
2659 default:
2660 pr_warning("%s con %p unrecognized state %lu\n",
2661 __func__, con, con->state);
2662 con->error_msg = "unrecognized con state";
2663 BUG();
2664 break;
2665 }
2666#undef CASE
2667
2668 return true;
2669}
2670
2671static bool con_backoff(struct ceph_connection *con)
2672{
2673 int ret;
2674
2675 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2676 return false;
2677
2678 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2679 if (ret) {
2680 dout("%s: con %p FAILED to back off %lu\n", __func__,
2681 con, con->delay);
2682 BUG_ON(ret == -ENOENT);
2683 con_flag_set(con, CON_FLAG_BACKOFF);
2684 }
2685
2686 return true;
2687}
2688
2689/* Finish fault handling; con->mutex must *not* be held here */
2690
2691static void con_fault_finish(struct ceph_connection *con)
2692{
2693 /*
2694 * in case we faulted due to authentication, invalidate our
2695 * current tickets so that we can get new ones.
2696 */
2697 if (con->auth_retry && con->ops->invalidate_authorizer) {
2698 dout("calling invalidate_authorizer()\n");
2699 con->ops->invalidate_authorizer(con);
2700 }
2701
2702 if (con->ops->fault)
2703 con->ops->fault(con);
2704}
2705
2706/*
2707 * Do some work on a connection. Drop a connection ref when we're done.
2708 */
2709static void con_work(struct work_struct *work)
2710{
2711 struct ceph_connection *con = container_of(work, struct ceph_connection,
2712 work.work);
2713 bool fault;
2714
2715 mutex_lock(&con->mutex);
2716 while (true) {
2717 int ret;
2718
2719 if ((fault = con_sock_closed(con))) {
2720 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2721 break;
2722 }
2723 if (con_backoff(con)) {
2724 dout("%s: con %p BACKOFF\n", __func__, con);
2725 break;
2726 }
2727 if (con->state == CON_STATE_STANDBY) {
2728 dout("%s: con %p STANDBY\n", __func__, con);
2729 break;
2730 }
2731 if (con->state == CON_STATE_CLOSED) {
2732 dout("%s: con %p CLOSED\n", __func__, con);
2733 BUG_ON(con->sock);
2734 break;
2735 }
2736 if (con->state == CON_STATE_PREOPEN) {
2737 dout("%s: con %p PREOPEN\n", __func__, con);
2738 BUG_ON(con->sock);
2739 }
2740
2741 ret = try_read(con);
2742 if (ret < 0) {
2743 if (ret == -EAGAIN)
2744 continue;
2745 con->error_msg = "socket error on read";
2746 fault = true;
2747 break;
2748 }
2749
2750 ret = try_write(con);
2751 if (ret < 0) {
2752 if (ret == -EAGAIN)
2753 continue;
2754 con->error_msg = "socket error on write";
2755 fault = true;
2756 }
2757
2758 break; /* If we make it to here, we're done */
2759 }
2760 if (fault)
2761 con_fault(con);
2762 mutex_unlock(&con->mutex);
2763
2764 if (fault)
2765 con_fault_finish(con);
2766
2767 con->ops->put(con);
2768}
2769
2770/*
2771 * Generic error/fault handler. A retry mechanism is used with
2772 * exponential backoff
2773 */
2774static void con_fault(struct ceph_connection *con)
2775{
2776 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2777 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2778 dout("fault %p state %lu to peer %s\n",
2779 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2780
2781 WARN_ON(con->state != CON_STATE_CONNECTING &&
2782 con->state != CON_STATE_NEGOTIATING &&
2783 con->state != CON_STATE_OPEN);
2784
2785 con_close_socket(con);
2786
2787 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2788 dout("fault on LOSSYTX channel, marking CLOSED\n");
2789 con->state = CON_STATE_CLOSED;
2790 return;
2791 }
2792
2793 if (con->in_msg) {
2794 BUG_ON(con->in_msg->con != con);
2795 con->in_msg->con = NULL;
2796 ceph_msg_put(con->in_msg);
2797 con->in_msg = NULL;
2798 con->ops->put(con);
2799 }
2800
2801 /* Requeue anything that hasn't been acked */
2802 list_splice_init(&con->out_sent, &con->out_queue);
2803
2804 /* If there are no messages queued or keepalive pending, place
2805 * the connection in a STANDBY state */
2806 if (list_empty(&con->out_queue) &&
2807 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
2808 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2809 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2810 con->state = CON_STATE_STANDBY;
2811 } else {
2812 /* retry after a delay. */
2813 con->state = CON_STATE_PREOPEN;
2814 if (con->delay == 0)
2815 con->delay = BASE_DELAY_INTERVAL;
2816 else if (con->delay < MAX_DELAY_INTERVAL)
2817 con->delay *= 2;
2818 con_flag_set(con, CON_FLAG_BACKOFF);
2819 queue_con(con);
2820 }
2821}
2822
2823
2824
2825/*
2826 * initialize a new messenger instance
2827 */
2828void ceph_messenger_init(struct ceph_messenger *msgr,
2829 struct ceph_entity_addr *myaddr,
2830 u32 supported_features,
2831 u32 required_features,
2832 bool nocrc)
2833{
2834 msgr->supported_features = supported_features;
2835 msgr->required_features = required_features;
2836
2837 spin_lock_init(&msgr->global_seq_lock);
2838
2839 if (myaddr)
2840 msgr->inst.addr = *myaddr;
2841
2842 /* select a random nonce */
2843 msgr->inst.addr.type = 0;
2844 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2845 encode_my_addr(msgr);
2846 msgr->nocrc = nocrc;
2847
2848 atomic_set(&msgr->stopping, 0);
2849
2850 dout("%s %p\n", __func__, msgr);
2851}
2852EXPORT_SYMBOL(ceph_messenger_init);
2853
2854static void clear_standby(struct ceph_connection *con)
2855{
2856 /* come back from STANDBY? */
2857 if (con->state == CON_STATE_STANDBY) {
2858 dout("clear_standby %p and ++connect_seq\n", con);
2859 con->state = CON_STATE_PREOPEN;
2860 con->connect_seq++;
2861 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2862 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
2863 }
2864}
2865
2866/*
2867 * Queue up an outgoing message on the given connection.
2868 */
2869void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2870{
2871 /* set src+dst */
2872 msg->hdr.src = con->msgr->inst.name;
2873 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2874 msg->needs_out_seq = true;
2875
2876 mutex_lock(&con->mutex);
2877
2878 if (con->state == CON_STATE_CLOSED) {
2879 dout("con_send %p closed, dropping %p\n", con, msg);
2880 ceph_msg_put(msg);
2881 mutex_unlock(&con->mutex);
2882 return;
2883 }
2884
2885 BUG_ON(msg->con != NULL);
2886 msg->con = con->ops->get(con);
2887 BUG_ON(msg->con == NULL);
2888
2889 BUG_ON(!list_empty(&msg->list_head));
2890 list_add_tail(&msg->list_head, &con->out_queue);
2891 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2892 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2893 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2894 le32_to_cpu(msg->hdr.front_len),
2895 le32_to_cpu(msg->hdr.middle_len),
2896 le32_to_cpu(msg->hdr.data_len));
2897
2898 clear_standby(con);
2899 mutex_unlock(&con->mutex);
2900
2901 /* if there wasn't anything waiting to send before, queue
2902 * new work */
2903 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
2904 queue_con(con);
2905}
2906EXPORT_SYMBOL(ceph_con_send);
2907
2908/*
2909 * Revoke a message that was previously queued for send
2910 */
2911void ceph_msg_revoke(struct ceph_msg *msg)
2912{
2913 struct ceph_connection *con = msg->con;
2914
2915 if (!con)
2916 return; /* Message not in our possession */
2917
2918 mutex_lock(&con->mutex);
2919 if (!list_empty(&msg->list_head)) {
2920 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
2921 list_del_init(&msg->list_head);
2922 BUG_ON(msg->con == NULL);
2923 msg->con->ops->put(msg->con);
2924 msg->con = NULL;
2925 msg->hdr.seq = 0;
2926
2927 ceph_msg_put(msg);
2928 }
2929 if (con->out_msg == msg) {
2930 dout("%s %p msg %p - was sending\n", __func__, con, msg);
2931 con->out_msg = NULL;
2932 if (con->out_kvec_is_msg) {
2933 con->out_skip = con->out_kvec_bytes;
2934 con->out_kvec_is_msg = false;
2935 }
2936 msg->hdr.seq = 0;
2937
2938 ceph_msg_put(msg);
2939 }
2940 mutex_unlock(&con->mutex);
2941}
2942
2943/*
2944 * Revoke a message that we may be reading data into
2945 */
2946void ceph_msg_revoke_incoming(struct ceph_msg *msg)
2947{
2948 struct ceph_connection *con;
2949
2950 BUG_ON(msg == NULL);
2951 if (!msg->con) {
2952 dout("%s msg %p null con\n", __func__, msg);
2953
2954 return; /* Message not in our possession */
2955 }
2956
2957 con = msg->con;
2958 mutex_lock(&con->mutex);
2959 if (con->in_msg == msg) {
2960 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
2961 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
2962 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
2963
2964 /* skip rest of message */
2965 dout("%s %p msg %p revoked\n", __func__, con, msg);
2966 con->in_base_pos = con->in_base_pos -
2967 sizeof(struct ceph_msg_header) -
2968 front_len -
2969 middle_len -
2970 data_len -
2971 sizeof(struct ceph_msg_footer);
2972 ceph_msg_put(con->in_msg);
2973 con->in_msg = NULL;
2974 con->in_tag = CEPH_MSGR_TAG_READY;
2975 con->in_seq++;
2976 } else {
2977 dout("%s %p in_msg %p msg %p no-op\n",
2978 __func__, con, con->in_msg, msg);
2979 }
2980 mutex_unlock(&con->mutex);
2981}
2982
2983/*
2984 * Queue a keepalive byte to ensure the tcp connection is alive.
2985 */
2986void ceph_con_keepalive(struct ceph_connection *con)
2987{
2988 dout("con_keepalive %p\n", con);
2989 mutex_lock(&con->mutex);
2990 clear_standby(con);
2991 mutex_unlock(&con->mutex);
2992 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
2993 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
2994 queue_con(con);
2995}
2996EXPORT_SYMBOL(ceph_con_keepalive);
2997
2998static void ceph_msg_data_init(struct ceph_msg_data *data)
2999{
3000 data->type = CEPH_MSG_DATA_NONE;
3001}
3002
3003void ceph_msg_data_set_pages(struct ceph_msg *msg, struct page **pages,
3004 size_t length, size_t alignment)
3005{
3006 BUG_ON(!pages);
3007 BUG_ON(!length);
3008 BUG_ON(msg->data.type != CEPH_MSG_DATA_NONE);
3009
3010 msg->data.type = CEPH_MSG_DATA_PAGES;
3011 msg->data.pages = pages;
3012 msg->data.length = length;
3013 msg->data.alignment = alignment & ~PAGE_MASK;
3014}
3015EXPORT_SYMBOL(ceph_msg_data_set_pages);
3016
3017void ceph_msg_data_set_pagelist(struct ceph_msg *msg,
3018 struct ceph_pagelist *pagelist)
3019{
3020 BUG_ON(!pagelist);
3021 BUG_ON(!pagelist->length);
3022 BUG_ON(msg->data.type != CEPH_MSG_DATA_NONE);
3023
3024 msg->data.type = CEPH_MSG_DATA_PAGELIST;
3025 msg->data.pagelist = pagelist;
3026}
3027EXPORT_SYMBOL(ceph_msg_data_set_pagelist);
3028
3029void ceph_msg_data_set_bio(struct ceph_msg *msg, struct bio *bio)
3030{
3031 BUG_ON(!bio);
3032 BUG_ON(msg->data.type != CEPH_MSG_DATA_NONE);
3033
3034 msg->data.type = CEPH_MSG_DATA_BIO;
3035 msg->data.bio = bio;
3036}
3037EXPORT_SYMBOL(ceph_msg_data_set_bio);
3038
3039/*
3040 * construct a new message with given type, size
3041 * the new msg has a ref count of 1.
3042 */
3043struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3044 bool can_fail)
3045{
3046 struct ceph_msg *m;
3047
3048 m = kzalloc(sizeof(*m), flags);
3049 if (m == NULL)
3050 goto out;
3051
3052 m->hdr.type = cpu_to_le16(type);
3053 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3054 m->hdr.front_len = cpu_to_le32(front_len);
3055
3056 INIT_LIST_HEAD(&m->list_head);
3057 kref_init(&m->kref);
3058
3059 ceph_msg_data_init(&m->data);
3060
3061 /* front */
3062 m->front_max = front_len;
3063 if (front_len) {
3064 if (front_len > PAGE_CACHE_SIZE) {
3065 m->front.iov_base = __vmalloc(front_len, flags,
3066 PAGE_KERNEL);
3067 m->front_is_vmalloc = true;
3068 } else {
3069 m->front.iov_base = kmalloc(front_len, flags);
3070 }
3071 if (m->front.iov_base == NULL) {
3072 dout("ceph_msg_new can't allocate %d bytes\n",
3073 front_len);
3074 goto out2;
3075 }
3076 } else {
3077 m->front.iov_base = NULL;
3078 }
3079 m->front.iov_len = front_len;
3080
3081 dout("ceph_msg_new %p front %d\n", m, front_len);
3082 return m;
3083
3084out2:
3085 ceph_msg_put(m);
3086out:
3087 if (!can_fail) {
3088 pr_err("msg_new can't create type %d front %d\n", type,
3089 front_len);
3090 WARN_ON(1);
3091 } else {
3092 dout("msg_new can't create type %d front %d\n", type,
3093 front_len);
3094 }
3095 return NULL;
3096}
3097EXPORT_SYMBOL(ceph_msg_new);
3098
3099/*
3100 * Allocate "middle" portion of a message, if it is needed and wasn't
3101 * allocated by alloc_msg. This allows us to read a small fixed-size
3102 * per-type header in the front and then gracefully fail (i.e.,
3103 * propagate the error to the caller based on info in the front) when
3104 * the middle is too large.
3105 */
3106static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3107{
3108 int type = le16_to_cpu(msg->hdr.type);
3109 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3110
3111 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3112 ceph_msg_type_name(type), middle_len);
3113 BUG_ON(!middle_len);
3114 BUG_ON(msg->middle);
3115
3116 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3117 if (!msg->middle)
3118 return -ENOMEM;
3119 return 0;
3120}
3121
3122/*
3123 * Allocate a message for receiving an incoming message on a
3124 * connection, and save the result in con->in_msg. Uses the
3125 * connection's private alloc_msg op if available.
3126 *
3127 * Returns 0 on success, or a negative error code.
3128 *
3129 * On success, if we set *skip = 1:
3130 * - the next message should be skipped and ignored.
3131 * - con->in_msg == NULL
3132 * or if we set *skip = 0:
3133 * - con->in_msg is non-null.
3134 * On error (ENOMEM, EAGAIN, ...),
3135 * - con->in_msg == NULL
3136 */
3137static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3138{
3139 struct ceph_msg_header *hdr = &con->in_hdr;
3140 int middle_len = le32_to_cpu(hdr->middle_len);
3141 struct ceph_msg *msg;
3142 int ret = 0;
3143
3144 BUG_ON(con->in_msg != NULL);
3145 BUG_ON(!con->ops->alloc_msg);
3146
3147 mutex_unlock(&con->mutex);
3148 msg = con->ops->alloc_msg(con, hdr, skip);
3149 mutex_lock(&con->mutex);
3150 if (con->state != CON_STATE_OPEN) {
3151 if (msg)
3152 ceph_msg_put(msg);
3153 return -EAGAIN;
3154 }
3155 if (msg) {
3156 BUG_ON(*skip);
3157 con->in_msg = msg;
3158 con->in_msg->con = con->ops->get(con);
3159 BUG_ON(con->in_msg->con == NULL);
3160 } else {
3161 /*
3162 * Null message pointer means either we should skip
3163 * this message or we couldn't allocate memory. The
3164 * former is not an error.
3165 */
3166 if (*skip)
3167 return 0;
3168 con->error_msg = "error allocating memory for incoming message";
3169
3170 return -ENOMEM;
3171 }
3172 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3173
3174 if (middle_len && !con->in_msg->middle) {
3175 ret = ceph_alloc_middle(con, con->in_msg);
3176 if (ret < 0) {
3177 ceph_msg_put(con->in_msg);
3178 con->in_msg = NULL;
3179 }
3180 }
3181
3182 return ret;
3183}
3184
3185
3186/*
3187 * Free a generically kmalloc'd message.
3188 */
3189void ceph_msg_kfree(struct ceph_msg *m)
3190{
3191 dout("msg_kfree %p\n", m);
3192 if (m->front_is_vmalloc)
3193 vfree(m->front.iov_base);
3194 else
3195 kfree(m->front.iov_base);
3196 kfree(m);
3197}
3198
3199/*
3200 * Drop a msg ref. Destroy as needed.
3201 */
3202void ceph_msg_last_put(struct kref *kref)
3203{
3204 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3205
3206 dout("ceph_msg_put last one on %p\n", m);
3207 WARN_ON(!list_empty(&m->list_head));
3208
3209 /* drop middle, data, if any */
3210 if (m->middle) {
3211 ceph_buffer_put(m->middle);
3212 m->middle = NULL;
3213 }
3214 if (ceph_msg_has_data(m)) {
3215 if (m->data.type == CEPH_MSG_DATA_PAGELIST) {
3216 ceph_pagelist_release(m->data.pagelist);
3217 kfree(m->data.pagelist);
3218 }
3219 memset(&m->data, 0, sizeof m->data);
3220 ceph_msg_data_init(&m->data);
3221 }
3222
3223 if (m->pool)
3224 ceph_msgpool_put(m->pool, m);
3225 else
3226 ceph_msg_kfree(m);
3227}
3228EXPORT_SYMBOL(ceph_msg_last_put);
3229
3230void ceph_msg_dump(struct ceph_msg *msg)
3231{
3232 pr_debug("msg_dump %p (front_max %d length %zd)\n", msg,
3233 msg->front_max, msg->data.length);
3234 print_hex_dump(KERN_DEBUG, "header: ",
3235 DUMP_PREFIX_OFFSET, 16, 1,
3236 &msg->hdr, sizeof(msg->hdr), true);
3237 print_hex_dump(KERN_DEBUG, " front: ",
3238 DUMP_PREFIX_OFFSET, 16, 1,
3239 msg->front.iov_base, msg->front.iov_len, true);
3240 if (msg->middle)
3241 print_hex_dump(KERN_DEBUG, "middle: ",
3242 DUMP_PREFIX_OFFSET, 16, 1,
3243 msg->middle->vec.iov_base,
3244 msg->middle->vec.iov_len, true);
3245 print_hex_dump(KERN_DEBUG, "footer: ",
3246 DUMP_PREFIX_OFFSET, 16, 1,
3247 &msg->footer, sizeof(msg->footer), true);
3248}
3249EXPORT_SYMBOL(ceph_msg_dump);
kernel source for telekom puls (alcatel/mediatek)