blob: 9ea7b763c8dce3346ec530b1f726c1c5d1b63e74 [file] [log] [blame]
Sage Weil31b80062009-10-06 11:31:13 -07001#include "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/socket.h>
10#include <linux/string.h>
11#include <net/tcp.h>
12
13#include "super.h"
14#include "messenger.h"
Sage Weil63f2d212009-11-03 15:17:56 -080015#include "decode.h"
Sage Weil58bb3b32009-12-23 12:12:31 -080016#include "pagelist.h"
Sage Weil31b80062009-10-06 11:31:13 -070017
18/*
19 * Ceph uses the messenger to exchange ceph_msg messages with other
20 * hosts in the system. The messenger provides ordered and reliable
21 * delivery. We tolerate TCP disconnects by reconnecting (with
22 * exponential backoff) in the case of a fault (disconnection, bad
23 * crc, protocol error). Acks allow sent messages to be discarded by
24 * the sender.
25 */
26
27/* static tag bytes (protocol control messages) */
28static char tag_msg = CEPH_MSGR_TAG_MSG;
29static char tag_ack = CEPH_MSGR_TAG_ACK;
30static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
31
32
33static void queue_con(struct ceph_connection *con);
34static void con_work(struct work_struct *);
35static void ceph_fault(struct ceph_connection *con);
36
37const char *ceph_name_type_str(int t)
38{
39 switch (t) {
40 case CEPH_ENTITY_TYPE_MON: return "mon";
41 case CEPH_ENTITY_TYPE_MDS: return "mds";
42 case CEPH_ENTITY_TYPE_OSD: return "osd";
43 case CEPH_ENTITY_TYPE_CLIENT: return "client";
44 case CEPH_ENTITY_TYPE_ADMIN: return "admin";
45 default: return "???";
46 }
47}
48
49/*
50 * nicely render a sockaddr as a string.
51 */
52#define MAX_ADDR_STR 20
53static char addr_str[MAX_ADDR_STR][40];
54static DEFINE_SPINLOCK(addr_str_lock);
55static int last_addr_str;
56
57const char *pr_addr(const struct sockaddr_storage *ss)
58{
59 int i;
60 char *s;
61 struct sockaddr_in *in4 = (void *)ss;
62 unsigned char *quad = (void *)&in4->sin_addr.s_addr;
63 struct sockaddr_in6 *in6 = (void *)ss;
64
65 spin_lock(&addr_str_lock);
66 i = last_addr_str++;
67 if (last_addr_str == MAX_ADDR_STR)
68 last_addr_str = 0;
69 spin_unlock(&addr_str_lock);
70 s = addr_str[i];
71
72 switch (ss->ss_family) {
73 case AF_INET:
74 sprintf(s, "%u.%u.%u.%u:%u",
75 (unsigned int)quad[0],
76 (unsigned int)quad[1],
77 (unsigned int)quad[2],
78 (unsigned int)quad[3],
79 (unsigned int)ntohs(in4->sin_port));
80 break;
81
82 case AF_INET6:
83 sprintf(s, "%04x:%04x:%04x:%04x:%04x:%04x:%04x:%04x:%u",
84 in6->sin6_addr.s6_addr16[0],
85 in6->sin6_addr.s6_addr16[1],
86 in6->sin6_addr.s6_addr16[2],
87 in6->sin6_addr.s6_addr16[3],
88 in6->sin6_addr.s6_addr16[4],
89 in6->sin6_addr.s6_addr16[5],
90 in6->sin6_addr.s6_addr16[6],
91 in6->sin6_addr.s6_addr16[7],
92 (unsigned int)ntohs(in6->sin6_port));
93 break;
94
95 default:
96 sprintf(s, "(unknown sockaddr family %d)", (int)ss->ss_family);
97 }
98
99 return s;
100}
101
Sage Weil63f2d212009-11-03 15:17:56 -0800102static void encode_my_addr(struct ceph_messenger *msgr)
103{
104 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
105 ceph_encode_addr(&msgr->my_enc_addr);
106}
107
Sage Weil31b80062009-10-06 11:31:13 -0700108/*
109 * work queue for all reading and writing to/from the socket.
110 */
111struct workqueue_struct *ceph_msgr_wq;
112
113int __init ceph_msgr_init(void)
114{
115 ceph_msgr_wq = create_workqueue("ceph-msgr");
116 if (IS_ERR(ceph_msgr_wq)) {
117 int ret = PTR_ERR(ceph_msgr_wq);
118 pr_err("msgr_init failed to create workqueue: %d\n", ret);
119 ceph_msgr_wq = NULL;
120 return ret;
121 }
122 return 0;
123}
124
125void ceph_msgr_exit(void)
126{
127 destroy_workqueue(ceph_msgr_wq);
128}
129
130/*
131 * socket callback functions
132 */
133
134/* data available on socket, or listen socket received a connect */
135static void ceph_data_ready(struct sock *sk, int count_unused)
136{
137 struct ceph_connection *con =
138 (struct ceph_connection *)sk->sk_user_data;
139 if (sk->sk_state != TCP_CLOSE_WAIT) {
140 dout("ceph_data_ready on %p state = %lu, queueing work\n",
141 con, con->state);
142 queue_con(con);
143 }
144}
145
146/* socket has buffer space for writing */
147static void ceph_write_space(struct sock *sk)
148{
149 struct ceph_connection *con =
150 (struct ceph_connection *)sk->sk_user_data;
151
152 /* only queue to workqueue if there is data we want to write. */
153 if (test_bit(WRITE_PENDING, &con->state)) {
154 dout("ceph_write_space %p queueing write work\n", con);
155 queue_con(con);
156 } else {
157 dout("ceph_write_space %p nothing to write\n", con);
158 }
159
160 /* since we have our own write_space, clear the SOCK_NOSPACE flag */
161 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
162}
163
164/* socket's state has changed */
165static void ceph_state_change(struct sock *sk)
166{
167 struct ceph_connection *con =
168 (struct ceph_connection *)sk->sk_user_data;
169
170 dout("ceph_state_change %p state = %lu sk_state = %u\n",
171 con, con->state, sk->sk_state);
172
173 if (test_bit(CLOSED, &con->state))
174 return;
175
176 switch (sk->sk_state) {
177 case TCP_CLOSE:
178 dout("ceph_state_change TCP_CLOSE\n");
179 case TCP_CLOSE_WAIT:
180 dout("ceph_state_change TCP_CLOSE_WAIT\n");
181 if (test_and_set_bit(SOCK_CLOSED, &con->state) == 0) {
182 if (test_bit(CONNECTING, &con->state))
183 con->error_msg = "connection failed";
184 else
185 con->error_msg = "socket closed";
186 queue_con(con);
187 }
188 break;
189 case TCP_ESTABLISHED:
190 dout("ceph_state_change TCP_ESTABLISHED\n");
191 queue_con(con);
192 break;
193 }
194}
195
196/*
197 * set up socket callbacks
198 */
199static void set_sock_callbacks(struct socket *sock,
200 struct ceph_connection *con)
201{
202 struct sock *sk = sock->sk;
203 sk->sk_user_data = (void *)con;
204 sk->sk_data_ready = ceph_data_ready;
205 sk->sk_write_space = ceph_write_space;
206 sk->sk_state_change = ceph_state_change;
207}
208
209
210/*
211 * socket helpers
212 */
213
214/*
215 * initiate connection to a remote socket.
216 */
217static struct socket *ceph_tcp_connect(struct ceph_connection *con)
218{
219 struct sockaddr *paddr = (struct sockaddr *)&con->peer_addr.in_addr;
220 struct socket *sock;
221 int ret;
222
223 BUG_ON(con->sock);
224 ret = sock_create_kern(AF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
225 if (ret)
226 return ERR_PTR(ret);
227 con->sock = sock;
228 sock->sk->sk_allocation = GFP_NOFS;
229
230 set_sock_callbacks(sock, con);
231
232 dout("connect %s\n", pr_addr(&con->peer_addr.in_addr));
233
234 ret = sock->ops->connect(sock, paddr, sizeof(*paddr), O_NONBLOCK);
235 if (ret == -EINPROGRESS) {
236 dout("connect %s EINPROGRESS sk_state = %u\n",
237 pr_addr(&con->peer_addr.in_addr),
238 sock->sk->sk_state);
239 ret = 0;
240 }
241 if (ret < 0) {
242 pr_err("connect %s error %d\n",
243 pr_addr(&con->peer_addr.in_addr), ret);
244 sock_release(sock);
245 con->sock = NULL;
246 con->error_msg = "connect error";
247 }
248
249 if (ret < 0)
250 return ERR_PTR(ret);
251 return sock;
252}
253
254static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
255{
256 struct kvec iov = {buf, len};
257 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
258
259 return kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
260}
261
262/*
263 * write something. @more is true if caller will be sending more data
264 * shortly.
265 */
266static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
267 size_t kvlen, size_t len, int more)
268{
269 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
270
271 if (more)
272 msg.msg_flags |= MSG_MORE;
273 else
274 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
275
276 return kernel_sendmsg(sock, &msg, iov, kvlen, len);
277}
278
279
280/*
281 * Shutdown/close the socket for the given connection.
282 */
283static int con_close_socket(struct ceph_connection *con)
284{
285 int rc;
286
287 dout("con_close_socket on %p sock %p\n", con, con->sock);
288 if (!con->sock)
289 return 0;
290 set_bit(SOCK_CLOSED, &con->state);
291 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
292 sock_release(con->sock);
293 con->sock = NULL;
294 clear_bit(SOCK_CLOSED, &con->state);
295 return rc;
296}
297
298/*
299 * Reset a connection. Discard all incoming and outgoing messages
300 * and clear *_seq state.
301 */
302static void ceph_msg_remove(struct ceph_msg *msg)
303{
304 list_del_init(&msg->list_head);
305 ceph_msg_put(msg);
306}
307static void ceph_msg_remove_list(struct list_head *head)
308{
309 while (!list_empty(head)) {
310 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
311 list_head);
312 ceph_msg_remove(msg);
313 }
314}
315
316static void reset_connection(struct ceph_connection *con)
317{
318 /* reset connection, out_queue, msg_ and connect_seq */
319 /* discard existing out_queue and msg_seq */
Sage Weil31b80062009-10-06 11:31:13 -0700320 ceph_msg_remove_list(&con->out_queue);
321 ceph_msg_remove_list(&con->out_sent);
322
Sage Weilcf3e5c42009-12-11 09:48:05 -0800323 if (con->in_msg) {
324 ceph_msg_put(con->in_msg);
325 con->in_msg = NULL;
326 }
327
Sage Weil31b80062009-10-06 11:31:13 -0700328 con->connect_seq = 0;
329 con->out_seq = 0;
Sage Weilc86a2932009-12-14 14:04:30 -0800330 if (con->out_msg) {
331 ceph_msg_put(con->out_msg);
332 con->out_msg = NULL;
333 }
Sage Weil31b80062009-10-06 11:31:13 -0700334 con->in_seq = 0;
Sage Weil31b80062009-10-06 11:31:13 -0700335}
336
337/*
338 * mark a peer down. drop any open connections.
339 */
340void ceph_con_close(struct ceph_connection *con)
341{
342 dout("con_close %p peer %s\n", con, pr_addr(&con->peer_addr.in_addr));
343 set_bit(CLOSED, &con->state); /* in case there's queued work */
344 clear_bit(STANDBY, &con->state); /* avoid connect_seq bump */
Sage Weilec302642009-12-22 10:43:42 -0800345 mutex_lock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -0700346 reset_connection(con);
Sage Weil91e45ce32010-02-15 12:05:09 -0800347 cancel_delayed_work(&con->work);
Sage Weilec302642009-12-22 10:43:42 -0800348 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -0700349 queue_con(con);
350}
351
352/*
Sage Weil31b80062009-10-06 11:31:13 -0700353 * Reopen a closed connection, with a new peer address.
354 */
355void ceph_con_open(struct ceph_connection *con, struct ceph_entity_addr *addr)
356{
357 dout("con_open %p %s\n", con, pr_addr(&addr->in_addr));
358 set_bit(OPENING, &con->state);
359 clear_bit(CLOSED, &con->state);
360 memcpy(&con->peer_addr, addr, sizeof(*addr));
Sage Weil03c677e2009-11-20 15:14:15 -0800361 con->delay = 0; /* reset backoff memory */
Sage Weil31b80062009-10-06 11:31:13 -0700362 queue_con(con);
363}
364
365/*
366 * generic get/put
367 */
368struct ceph_connection *ceph_con_get(struct ceph_connection *con)
369{
370 dout("con_get %p nref = %d -> %d\n", con,
371 atomic_read(&con->nref), atomic_read(&con->nref) + 1);
372 if (atomic_inc_not_zero(&con->nref))
373 return con;
374 return NULL;
375}
376
377void ceph_con_put(struct ceph_connection *con)
378{
379 dout("con_put %p nref = %d -> %d\n", con,
380 atomic_read(&con->nref), atomic_read(&con->nref) - 1);
381 BUG_ON(atomic_read(&con->nref) == 0);
382 if (atomic_dec_and_test(&con->nref)) {
Sage Weil71ececd2009-11-18 11:27:06 -0800383 BUG_ON(con->sock);
Sage Weil31b80062009-10-06 11:31:13 -0700384 kfree(con);
385 }
386}
387
388/*
389 * initialize a new connection.
390 */
391void ceph_con_init(struct ceph_messenger *msgr, struct ceph_connection *con)
392{
393 dout("con_init %p\n", con);
394 memset(con, 0, sizeof(*con));
395 atomic_set(&con->nref, 1);
396 con->msgr = msgr;
Sage Weilec302642009-12-22 10:43:42 -0800397 mutex_init(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -0700398 INIT_LIST_HEAD(&con->out_queue);
399 INIT_LIST_HEAD(&con->out_sent);
400 INIT_DELAYED_WORK(&con->work, con_work);
401}
402
403
404/*
405 * We maintain a global counter to order connection attempts. Get
406 * a unique seq greater than @gt.
407 */
408static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
409{
410 u32 ret;
411
412 spin_lock(&msgr->global_seq_lock);
413 if (msgr->global_seq < gt)
414 msgr->global_seq = gt;
415 ret = ++msgr->global_seq;
416 spin_unlock(&msgr->global_seq_lock);
417 return ret;
418}
419
420
421/*
422 * Prepare footer for currently outgoing message, and finish things
423 * off. Assumes out_kvec* are already valid.. we just add on to the end.
424 */
425static void prepare_write_message_footer(struct ceph_connection *con, int v)
426{
427 struct ceph_msg *m = con->out_msg;
428
429 dout("prepare_write_message_footer %p\n", con);
430 con->out_kvec_is_msg = true;
431 con->out_kvec[v].iov_base = &m->footer;
432 con->out_kvec[v].iov_len = sizeof(m->footer);
433 con->out_kvec_bytes += sizeof(m->footer);
434 con->out_kvec_left++;
435 con->out_more = m->more_to_follow;
Sage Weilc86a2932009-12-14 14:04:30 -0800436 con->out_msg_done = true;
Sage Weil31b80062009-10-06 11:31:13 -0700437}
438
439/*
440 * Prepare headers for the next outgoing message.
441 */
442static void prepare_write_message(struct ceph_connection *con)
443{
444 struct ceph_msg *m;
445 int v = 0;
446
447 con->out_kvec_bytes = 0;
448 con->out_kvec_is_msg = true;
Sage Weilc86a2932009-12-14 14:04:30 -0800449 con->out_msg_done = false;
Sage Weil31b80062009-10-06 11:31:13 -0700450
451 /* Sneak an ack in there first? If we can get it into the same
452 * TCP packet that's a good thing. */
453 if (con->in_seq > con->in_seq_acked) {
454 con->in_seq_acked = con->in_seq;
455 con->out_kvec[v].iov_base = &tag_ack;
456 con->out_kvec[v++].iov_len = 1;
457 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
458 con->out_kvec[v].iov_base = &con->out_temp_ack;
459 con->out_kvec[v++].iov_len = sizeof(con->out_temp_ack);
460 con->out_kvec_bytes = 1 + sizeof(con->out_temp_ack);
461 }
462
Sage Weil31b80062009-10-06 11:31:13 -0700463 m = list_first_entry(&con->out_queue,
464 struct ceph_msg, list_head);
Sage Weilc86a2932009-12-14 14:04:30 -0800465 con->out_msg = m;
Sage Weilb3d1dbb2009-12-14 14:58:11 -0800466 if (test_bit(LOSSYTX, &con->state)) {
Sage Weil6c5d1a42010-02-13 20:29:31 -0800467 list_del_init(&m->list_head);
468 } else {
Sage Weilb3d1dbb2009-12-14 14:58:11 -0800469 /* put message on sent list */
470 ceph_msg_get(m);
471 list_move_tail(&m->list_head, &con->out_sent);
Sage Weilb3d1dbb2009-12-14 14:58:11 -0800472 }
Sage Weil31b80062009-10-06 11:31:13 -0700473
474 m->hdr.seq = cpu_to_le64(++con->out_seq);
475
476 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n",
477 m, con->out_seq, le16_to_cpu(m->hdr.type),
478 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
479 le32_to_cpu(m->hdr.data_len),
480 m->nr_pages);
481 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
482
483 /* tag + hdr + front + middle */
484 con->out_kvec[v].iov_base = &tag_msg;
485 con->out_kvec[v++].iov_len = 1;
486 con->out_kvec[v].iov_base = &m->hdr;
487 con->out_kvec[v++].iov_len = sizeof(m->hdr);
488 con->out_kvec[v++] = m->front;
489 if (m->middle)
490 con->out_kvec[v++] = m->middle->vec;
491 con->out_kvec_left = v;
492 con->out_kvec_bytes += 1 + sizeof(m->hdr) + m->front.iov_len +
493 (m->middle ? m->middle->vec.iov_len : 0);
494 con->out_kvec_cur = con->out_kvec;
495
496 /* fill in crc (except data pages), footer */
497 con->out_msg->hdr.crc =
498 cpu_to_le32(crc32c(0, (void *)&m->hdr,
499 sizeof(m->hdr) - sizeof(m->hdr.crc)));
500 con->out_msg->footer.flags = CEPH_MSG_FOOTER_COMPLETE;
501 con->out_msg->footer.front_crc =
502 cpu_to_le32(crc32c(0, m->front.iov_base, m->front.iov_len));
503 if (m->middle)
504 con->out_msg->footer.middle_crc =
505 cpu_to_le32(crc32c(0, m->middle->vec.iov_base,
506 m->middle->vec.iov_len));
507 else
508 con->out_msg->footer.middle_crc = 0;
509 con->out_msg->footer.data_crc = 0;
510 dout("prepare_write_message front_crc %u data_crc %u\n",
511 le32_to_cpu(con->out_msg->footer.front_crc),
512 le32_to_cpu(con->out_msg->footer.middle_crc));
513
514 /* is there a data payload? */
515 if (le32_to_cpu(m->hdr.data_len) > 0) {
516 /* initialize page iterator */
517 con->out_msg_pos.page = 0;
518 con->out_msg_pos.page_pos =
519 le16_to_cpu(m->hdr.data_off) & ~PAGE_MASK;
520 con->out_msg_pos.data_pos = 0;
521 con->out_msg_pos.did_page_crc = 0;
522 con->out_more = 1; /* data + footer will follow */
523 } else {
524 /* no, queue up footer too and be done */
525 prepare_write_message_footer(con, v);
526 }
527
528 set_bit(WRITE_PENDING, &con->state);
529}
530
531/*
532 * Prepare an ack.
533 */
534static void prepare_write_ack(struct ceph_connection *con)
535{
536 dout("prepare_write_ack %p %llu -> %llu\n", con,
537 con->in_seq_acked, con->in_seq);
538 con->in_seq_acked = con->in_seq;
539
540 con->out_kvec[0].iov_base = &tag_ack;
541 con->out_kvec[0].iov_len = 1;
542 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
543 con->out_kvec[1].iov_base = &con->out_temp_ack;
544 con->out_kvec[1].iov_len = sizeof(con->out_temp_ack);
545 con->out_kvec_left = 2;
546 con->out_kvec_bytes = 1 + sizeof(con->out_temp_ack);
547 con->out_kvec_cur = con->out_kvec;
548 con->out_more = 1; /* more will follow.. eventually.. */
549 set_bit(WRITE_PENDING, &con->state);
550}
551
552/*
553 * Prepare to write keepalive byte.
554 */
555static void prepare_write_keepalive(struct ceph_connection *con)
556{
557 dout("prepare_write_keepalive %p\n", con);
558 con->out_kvec[0].iov_base = &tag_keepalive;
559 con->out_kvec[0].iov_len = 1;
560 con->out_kvec_left = 1;
561 con->out_kvec_bytes = 1;
562 con->out_kvec_cur = con->out_kvec;
563 set_bit(WRITE_PENDING, &con->state);
564}
565
566/*
567 * Connection negotiation.
568 */
569
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800570static void prepare_connect_authorizer(struct ceph_connection *con)
571{
572 void *auth_buf;
573 int auth_len = 0;
574 int auth_protocol = 0;
575
Sage Weilec302642009-12-22 10:43:42 -0800576 mutex_unlock(&con->mutex);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800577 if (con->ops->get_authorizer)
578 con->ops->get_authorizer(con, &auth_buf, &auth_len,
579 &auth_protocol, &con->auth_reply_buf,
580 &con->auth_reply_buf_len,
581 con->auth_retry);
Sage Weilec302642009-12-22 10:43:42 -0800582 mutex_lock(&con->mutex);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800583
584 con->out_connect.authorizer_protocol = cpu_to_le32(auth_protocol);
585 con->out_connect.authorizer_len = cpu_to_le32(auth_len);
586
587 con->out_kvec[con->out_kvec_left].iov_base = auth_buf;
588 con->out_kvec[con->out_kvec_left].iov_len = auth_len;
589 con->out_kvec_left++;
590 con->out_kvec_bytes += auth_len;
591}
592
Sage Weil31b80062009-10-06 11:31:13 -0700593/*
594 * We connected to a peer and are saying hello.
595 */
Sage Weileed0ef22009-11-10 14:34:36 -0800596static void prepare_write_banner(struct ceph_messenger *msgr,
597 struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -0700598{
599 int len = strlen(CEPH_BANNER);
Sage Weileed0ef22009-11-10 14:34:36 -0800600
601 con->out_kvec[0].iov_base = CEPH_BANNER;
602 con->out_kvec[0].iov_len = len;
603 con->out_kvec[1].iov_base = &msgr->my_enc_addr;
604 con->out_kvec[1].iov_len = sizeof(msgr->my_enc_addr);
605 con->out_kvec_left = 2;
606 con->out_kvec_bytes = len + sizeof(msgr->my_enc_addr);
607 con->out_kvec_cur = con->out_kvec;
608 con->out_more = 0;
609 set_bit(WRITE_PENDING, &con->state);
610}
611
612static void prepare_write_connect(struct ceph_messenger *msgr,
613 struct ceph_connection *con,
614 int after_banner)
615{
Sage Weil31b80062009-10-06 11:31:13 -0700616 unsigned global_seq = get_global_seq(con->msgr, 0);
617 int proto;
618
619 switch (con->peer_name.type) {
620 case CEPH_ENTITY_TYPE_MON:
621 proto = CEPH_MONC_PROTOCOL;
622 break;
623 case CEPH_ENTITY_TYPE_OSD:
624 proto = CEPH_OSDC_PROTOCOL;
625 break;
626 case CEPH_ENTITY_TYPE_MDS:
627 proto = CEPH_MDSC_PROTOCOL;
628 break;
629 default:
630 BUG();
631 }
632
633 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
634 con->connect_seq, global_seq, proto);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800635
Sage Weil04a419f2009-12-23 09:30:21 -0800636 con->out_connect.features = CEPH_FEATURE_SUPPORTED;
Sage Weil31b80062009-10-06 11:31:13 -0700637 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
638 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
639 con->out_connect.global_seq = cpu_to_le32(global_seq);
640 con->out_connect.protocol_version = cpu_to_le32(proto);
641 con->out_connect.flags = 0;
Sage Weil31b80062009-10-06 11:31:13 -0700642
Sage Weileed0ef22009-11-10 14:34:36 -0800643 if (!after_banner) {
644 con->out_kvec_left = 0;
645 con->out_kvec_bytes = 0;
646 }
647 con->out_kvec[con->out_kvec_left].iov_base = &con->out_connect;
648 con->out_kvec[con->out_kvec_left].iov_len = sizeof(con->out_connect);
649 con->out_kvec_left++;
650 con->out_kvec_bytes += sizeof(con->out_connect);
Sage Weil31b80062009-10-06 11:31:13 -0700651 con->out_kvec_cur = con->out_kvec;
652 con->out_more = 0;
653 set_bit(WRITE_PENDING, &con->state);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800654
655 prepare_connect_authorizer(con);
Sage Weil31b80062009-10-06 11:31:13 -0700656}
657
658
659/*
660 * write as much of pending kvecs to the socket as we can.
661 * 1 -> done
662 * 0 -> socket full, but more to do
663 * <0 -> error
664 */
665static int write_partial_kvec(struct ceph_connection *con)
666{
667 int ret;
668
669 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
670 while (con->out_kvec_bytes > 0) {
671 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
672 con->out_kvec_left, con->out_kvec_bytes,
673 con->out_more);
674 if (ret <= 0)
675 goto out;
676 con->out_kvec_bytes -= ret;
677 if (con->out_kvec_bytes == 0)
678 break; /* done */
679 while (ret > 0) {
680 if (ret >= con->out_kvec_cur->iov_len) {
681 ret -= con->out_kvec_cur->iov_len;
682 con->out_kvec_cur++;
683 con->out_kvec_left--;
684 } else {
685 con->out_kvec_cur->iov_len -= ret;
686 con->out_kvec_cur->iov_base += ret;
687 ret = 0;
688 break;
689 }
690 }
691 }
692 con->out_kvec_left = 0;
693 con->out_kvec_is_msg = false;
694 ret = 1;
695out:
696 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
697 con->out_kvec_bytes, con->out_kvec_left, ret);
698 return ret; /* done! */
699}
700
701/*
702 * Write as much message data payload as we can. If we finish, queue
703 * up the footer.
704 * 1 -> done, footer is now queued in out_kvec[].
705 * 0 -> socket full, but more to do
706 * <0 -> error
707 */
708static int write_partial_msg_pages(struct ceph_connection *con)
709{
710 struct ceph_msg *msg = con->out_msg;
711 unsigned data_len = le32_to_cpu(msg->hdr.data_len);
712 size_t len;
713 int crc = con->msgr->nocrc;
714 int ret;
715
716 dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
717 con, con->out_msg, con->out_msg_pos.page, con->out_msg->nr_pages,
718 con->out_msg_pos.page_pos);
719
720 while (con->out_msg_pos.page < con->out_msg->nr_pages) {
721 struct page *page = NULL;
722 void *kaddr = NULL;
723
724 /*
725 * if we are calculating the data crc (the default), we need
726 * to map the page. if our pages[] has been revoked, use the
727 * zero page.
728 */
729 if (msg->pages) {
730 page = msg->pages[con->out_msg_pos.page];
731 if (crc)
732 kaddr = kmap(page);
Sage Weil58bb3b32009-12-23 12:12:31 -0800733 } else if (msg->pagelist) {
734 page = list_first_entry(&msg->pagelist->head,
735 struct page, lru);
736 if (crc)
737 kaddr = kmap(page);
Sage Weil31b80062009-10-06 11:31:13 -0700738 } else {
739 page = con->msgr->zero_page;
740 if (crc)
741 kaddr = page_address(con->msgr->zero_page);
742 }
743 len = min((int)(PAGE_SIZE - con->out_msg_pos.page_pos),
744 (int)(data_len - con->out_msg_pos.data_pos));
745 if (crc && !con->out_msg_pos.did_page_crc) {
746 void *base = kaddr + con->out_msg_pos.page_pos;
747 u32 tmpcrc = le32_to_cpu(con->out_msg->footer.data_crc);
748
749 BUG_ON(kaddr == NULL);
750 con->out_msg->footer.data_crc =
751 cpu_to_le32(crc32c(tmpcrc, base, len));
752 con->out_msg_pos.did_page_crc = 1;
753 }
754
755 ret = kernel_sendpage(con->sock, page,
756 con->out_msg_pos.page_pos, len,
757 MSG_DONTWAIT | MSG_NOSIGNAL |
758 MSG_MORE);
759
Sage Weil58bb3b32009-12-23 12:12:31 -0800760 if (crc && (msg->pages || msg->pagelist))
Sage Weil31b80062009-10-06 11:31:13 -0700761 kunmap(page);
762
763 if (ret <= 0)
764 goto out;
765
766 con->out_msg_pos.data_pos += ret;
767 con->out_msg_pos.page_pos += ret;
768 if (ret == len) {
769 con->out_msg_pos.page_pos = 0;
770 con->out_msg_pos.page++;
771 con->out_msg_pos.did_page_crc = 0;
Sage Weil58bb3b32009-12-23 12:12:31 -0800772 if (msg->pagelist)
773 list_move_tail(&page->lru,
774 &msg->pagelist->head);
Sage Weil31b80062009-10-06 11:31:13 -0700775 }
776 }
777
778 dout("write_partial_msg_pages %p msg %p done\n", con, msg);
779
780 /* prepare and queue up footer, too */
781 if (!crc)
782 con->out_msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
783 con->out_kvec_bytes = 0;
784 con->out_kvec_left = 0;
785 con->out_kvec_cur = con->out_kvec;
786 prepare_write_message_footer(con, 0);
787 ret = 1;
788out:
789 return ret;
790}
791
792/*
793 * write some zeros
794 */
795static int write_partial_skip(struct ceph_connection *con)
796{
797 int ret;
798
799 while (con->out_skip > 0) {
800 struct kvec iov = {
801 .iov_base = page_address(con->msgr->zero_page),
802 .iov_len = min(con->out_skip, (int)PAGE_CACHE_SIZE)
803 };
804
805 ret = ceph_tcp_sendmsg(con->sock, &iov, 1, iov.iov_len, 1);
806 if (ret <= 0)
807 goto out;
808 con->out_skip -= ret;
809 }
810 ret = 1;
811out:
812 return ret;
813}
814
815/*
816 * Prepare to read connection handshake, or an ack.
817 */
Sage Weileed0ef22009-11-10 14:34:36 -0800818static void prepare_read_banner(struct ceph_connection *con)
819{
820 dout("prepare_read_banner %p\n", con);
821 con->in_base_pos = 0;
822}
823
Sage Weil31b80062009-10-06 11:31:13 -0700824static void prepare_read_connect(struct ceph_connection *con)
825{
826 dout("prepare_read_connect %p\n", con);
827 con->in_base_pos = 0;
828}
829
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800830static void prepare_read_connect_retry(struct ceph_connection *con)
831{
832 dout("prepare_read_connect_retry %p\n", con);
833 con->in_base_pos = strlen(CEPH_BANNER) + sizeof(con->actual_peer_addr)
834 + sizeof(con->peer_addr_for_me);
835}
836
Sage Weil31b80062009-10-06 11:31:13 -0700837static void prepare_read_ack(struct ceph_connection *con)
838{
839 dout("prepare_read_ack %p\n", con);
840 con->in_base_pos = 0;
841}
842
843static void prepare_read_tag(struct ceph_connection *con)
844{
845 dout("prepare_read_tag %p\n", con);
846 con->in_base_pos = 0;
847 con->in_tag = CEPH_MSGR_TAG_READY;
848}
849
850/*
851 * Prepare to read a message.
852 */
853static int prepare_read_message(struct ceph_connection *con)
854{
855 dout("prepare_read_message %p\n", con);
856 BUG_ON(con->in_msg != NULL);
857 con->in_base_pos = 0;
858 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
859 return 0;
860}
861
862
863static int read_partial(struct ceph_connection *con,
864 int *to, int size, void *object)
865{
866 *to += size;
867 while (con->in_base_pos < *to) {
868 int left = *to - con->in_base_pos;
869 int have = size - left;
870 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
871 if (ret <= 0)
872 return ret;
873 con->in_base_pos += ret;
874 }
875 return 1;
876}
877
878
879/*
880 * Read all or part of the connect-side handshake on a new connection
881 */
Sage Weileed0ef22009-11-10 14:34:36 -0800882static int read_partial_banner(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -0700883{
884 int ret, to = 0;
885
Sage Weileed0ef22009-11-10 14:34:36 -0800886 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
Sage Weil31b80062009-10-06 11:31:13 -0700887
888 /* peer's banner */
889 ret = read_partial(con, &to, strlen(CEPH_BANNER), con->in_banner);
890 if (ret <= 0)
891 goto out;
892 ret = read_partial(con, &to, sizeof(con->actual_peer_addr),
893 &con->actual_peer_addr);
894 if (ret <= 0)
895 goto out;
896 ret = read_partial(con, &to, sizeof(con->peer_addr_for_me),
897 &con->peer_addr_for_me);
898 if (ret <= 0)
899 goto out;
Sage Weileed0ef22009-11-10 14:34:36 -0800900out:
901 return ret;
902}
903
904static int read_partial_connect(struct ceph_connection *con)
905{
906 int ret, to = 0;
907
908 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
909
Sage Weil31b80062009-10-06 11:31:13 -0700910 ret = read_partial(con, &to, sizeof(con->in_reply), &con->in_reply);
911 if (ret <= 0)
912 goto out;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800913 ret = read_partial(con, &to, le32_to_cpu(con->in_reply.authorizer_len),
914 con->auth_reply_buf);
915 if (ret <= 0)
916 goto out;
Sage Weil31b80062009-10-06 11:31:13 -0700917
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800918 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
919 con, (int)con->in_reply.tag,
920 le32_to_cpu(con->in_reply.connect_seq),
Sage Weil31b80062009-10-06 11:31:13 -0700921 le32_to_cpu(con->in_reply.global_seq));
922out:
923 return ret;
Sage Weileed0ef22009-11-10 14:34:36 -0800924
Sage Weil31b80062009-10-06 11:31:13 -0700925}
926
927/*
928 * Verify the hello banner looks okay.
929 */
930static int verify_hello(struct ceph_connection *con)
931{
932 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
Sage Weil13e38c82009-10-09 16:36:34 -0700933 pr_err("connect to %s got bad banner\n",
Sage Weil31b80062009-10-06 11:31:13 -0700934 pr_addr(&con->peer_addr.in_addr));
935 con->error_msg = "protocol error, bad banner";
936 return -1;
937 }
938 return 0;
939}
940
941static bool addr_is_blank(struct sockaddr_storage *ss)
942{
943 switch (ss->ss_family) {
944 case AF_INET:
945 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
946 case AF_INET6:
947 return
948 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
949 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
950 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
951 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
952 }
953 return false;
954}
955
956static int addr_port(struct sockaddr_storage *ss)
957{
958 switch (ss->ss_family) {
959 case AF_INET:
Sage Weilf28bcfb2009-11-04 11:46:35 -0800960 return ntohs(((struct sockaddr_in *)ss)->sin_port);
Sage Weil31b80062009-10-06 11:31:13 -0700961 case AF_INET6:
Sage Weilf28bcfb2009-11-04 11:46:35 -0800962 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
Sage Weil31b80062009-10-06 11:31:13 -0700963 }
964 return 0;
965}
966
967static void addr_set_port(struct sockaddr_storage *ss, int p)
968{
969 switch (ss->ss_family) {
970 case AF_INET:
971 ((struct sockaddr_in *)ss)->sin_port = htons(p);
972 case AF_INET6:
973 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
974 }
975}
976
977/*
978 * Parse an ip[:port] list into an addr array. Use the default
979 * monitor port if a port isn't specified.
980 */
981int ceph_parse_ips(const char *c, const char *end,
982 struct ceph_entity_addr *addr,
983 int max_count, int *count)
984{
985 int i;
986 const char *p = c;
987
988 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
989 for (i = 0; i < max_count; i++) {
990 const char *ipend;
991 struct sockaddr_storage *ss = &addr[i].in_addr;
992 struct sockaddr_in *in4 = (void *)ss;
993 struct sockaddr_in6 *in6 = (void *)ss;
994 int port;
995
996 memset(ss, 0, sizeof(*ss));
997 if (in4_pton(p, end - p, (u8 *)&in4->sin_addr.s_addr,
998 ',', &ipend)) {
999 ss->ss_family = AF_INET;
1000 } else if (in6_pton(p, end - p, (u8 *)&in6->sin6_addr.s6_addr,
1001 ',', &ipend)) {
1002 ss->ss_family = AF_INET6;
1003 } else {
1004 goto bad;
1005 }
1006 p = ipend;
1007
1008 /* port? */
1009 if (p < end && *p == ':') {
1010 port = 0;
1011 p++;
1012 while (p < end && *p >= '0' && *p <= '9') {
1013 port = (port * 10) + (*p - '0');
1014 p++;
1015 }
1016 if (port > 65535 || port == 0)
1017 goto bad;
1018 } else {
1019 port = CEPH_MON_PORT;
1020 }
1021
1022 addr_set_port(ss, port);
1023
1024 dout("parse_ips got %s\n", pr_addr(ss));
1025
1026 if (p == end)
1027 break;
1028 if (*p != ',')
1029 goto bad;
1030 p++;
1031 }
1032
1033 if (p != end)
1034 goto bad;
1035
1036 if (count)
1037 *count = i + 1;
1038 return 0;
1039
1040bad:
1041 pr_err("parse_ips bad ip '%s'\n", c);
1042 return -EINVAL;
1043}
1044
Sage Weileed0ef22009-11-10 14:34:36 -08001045static int process_banner(struct ceph_connection *con)
Sage Weil31b80062009-10-06 11:31:13 -07001046{
Sage Weileed0ef22009-11-10 14:34:36 -08001047 dout("process_banner on %p\n", con);
Sage Weil31b80062009-10-06 11:31:13 -07001048
1049 if (verify_hello(con) < 0)
1050 return -1;
1051
Sage Weil63f2d212009-11-03 15:17:56 -08001052 ceph_decode_addr(&con->actual_peer_addr);
1053 ceph_decode_addr(&con->peer_addr_for_me);
1054
Sage Weil31b80062009-10-06 11:31:13 -07001055 /*
1056 * Make sure the other end is who we wanted. note that the other
1057 * end may not yet know their ip address, so if it's 0.0.0.0, give
1058 * them the benefit of the doubt.
1059 */
Sage Weil103e2d32010-01-07 16:12:36 -08001060 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1061 sizeof(con->peer_addr)) != 0 &&
Sage Weil31b80062009-10-06 11:31:13 -07001062 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1063 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
Sage Weil103e2d32010-01-07 16:12:36 -08001064 pr_warning("wrong peer, want %s/%lld, got %s/%lld\n",
1065 pr_addr(&con->peer_addr.in_addr),
1066 le64_to_cpu(con->peer_addr.nonce),
1067 pr_addr(&con->actual_peer_addr.in_addr),
1068 le64_to_cpu(con->actual_peer_addr.nonce));
Sage Weil58bb3b32009-12-23 12:12:31 -08001069 con->error_msg = "wrong peer at address";
Sage Weil31b80062009-10-06 11:31:13 -07001070 return -1;
1071 }
1072
1073 /*
1074 * did we learn our address?
1075 */
1076 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1077 int port = addr_port(&con->msgr->inst.addr.in_addr);
1078
1079 memcpy(&con->msgr->inst.addr.in_addr,
1080 &con->peer_addr_for_me.in_addr,
1081 sizeof(con->peer_addr_for_me.in_addr));
1082 addr_set_port(&con->msgr->inst.addr.in_addr, port);
Sage Weil63f2d212009-11-03 15:17:56 -08001083 encode_my_addr(con->msgr);
Sage Weileed0ef22009-11-10 14:34:36 -08001084 dout("process_banner learned my addr is %s\n",
Sage Weil31b80062009-10-06 11:31:13 -07001085 pr_addr(&con->msgr->inst.addr.in_addr));
1086 }
1087
Sage Weileed0ef22009-11-10 14:34:36 -08001088 set_bit(NEGOTIATING, &con->state);
1089 prepare_read_connect(con);
1090 return 0;
1091}
1092
Sage Weil04a419f2009-12-23 09:30:21 -08001093static void fail_protocol(struct ceph_connection *con)
1094{
1095 reset_connection(con);
1096 set_bit(CLOSED, &con->state); /* in case there's queued work */
1097
1098 mutex_unlock(&con->mutex);
1099 if (con->ops->bad_proto)
1100 con->ops->bad_proto(con);
1101 mutex_lock(&con->mutex);
1102}
1103
Sage Weileed0ef22009-11-10 14:34:36 -08001104static int process_connect(struct ceph_connection *con)
1105{
Sage Weil04a419f2009-12-23 09:30:21 -08001106 u64 sup_feat = CEPH_FEATURE_SUPPORTED;
1107 u64 req_feat = CEPH_FEATURE_REQUIRED;
1108 u64 server_feat = le64_to_cpu(con->in_reply.features);
1109
Sage Weileed0ef22009-11-10 14:34:36 -08001110 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1111
Sage Weil31b80062009-10-06 11:31:13 -07001112 switch (con->in_reply.tag) {
Sage Weil04a419f2009-12-23 09:30:21 -08001113 case CEPH_MSGR_TAG_FEATURES:
1114 pr_err("%s%lld %s feature set mismatch,"
1115 " my %llx < server's %llx, missing %llx\n",
1116 ENTITY_NAME(con->peer_name),
1117 pr_addr(&con->peer_addr.in_addr),
1118 sup_feat, server_feat, server_feat & ~sup_feat);
1119 con->error_msg = "missing required protocol features";
1120 fail_protocol(con);
1121 return -1;
1122
Sage Weil31b80062009-10-06 11:31:13 -07001123 case CEPH_MSGR_TAG_BADPROTOVER:
Sage Weil31b80062009-10-06 11:31:13 -07001124 pr_err("%s%lld %s protocol version mismatch,"
1125 " my %d != server's %d\n",
1126 ENTITY_NAME(con->peer_name),
1127 pr_addr(&con->peer_addr.in_addr),
1128 le32_to_cpu(con->out_connect.protocol_version),
1129 le32_to_cpu(con->in_reply.protocol_version));
1130 con->error_msg = "protocol version mismatch";
Sage Weil04a419f2009-12-23 09:30:21 -08001131 fail_protocol(con);
Sage Weil31b80062009-10-06 11:31:13 -07001132 return -1;
1133
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001134 case CEPH_MSGR_TAG_BADAUTHORIZER:
1135 con->auth_retry++;
1136 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1137 con->auth_retry);
1138 if (con->auth_retry == 2) {
1139 con->error_msg = "connect authorization failure";
1140 reset_connection(con);
1141 set_bit(CLOSED, &con->state);
1142 return -1;
1143 }
1144 con->auth_retry = 1;
1145 prepare_write_connect(con->msgr, con, 0);
1146 prepare_read_connect_retry(con);
1147 break;
Sage Weil31b80062009-10-06 11:31:13 -07001148
1149 case CEPH_MSGR_TAG_RESETSESSION:
1150 /*
1151 * If we connected with a large connect_seq but the peer
1152 * has no record of a session with us (no connection, or
1153 * connect_seq == 0), they will send RESETSESION to indicate
1154 * that they must have reset their session, and may have
1155 * dropped messages.
1156 */
1157 dout("process_connect got RESET peer seq %u\n",
1158 le32_to_cpu(con->in_connect.connect_seq));
1159 pr_err("%s%lld %s connection reset\n",
1160 ENTITY_NAME(con->peer_name),
1161 pr_addr(&con->peer_addr.in_addr));
1162 reset_connection(con);
Sage Weileed0ef22009-11-10 14:34:36 -08001163 prepare_write_connect(con->msgr, con, 0);
Sage Weil31b80062009-10-06 11:31:13 -07001164 prepare_read_connect(con);
1165
1166 /* Tell ceph about it. */
Sage Weilec302642009-12-22 10:43:42 -08001167 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07001168 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
1169 if (con->ops->peer_reset)
1170 con->ops->peer_reset(con);
Sage Weilec302642009-12-22 10:43:42 -08001171 mutex_lock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07001172 break;
1173
1174 case CEPH_MSGR_TAG_RETRY_SESSION:
1175 /*
1176 * If we sent a smaller connect_seq than the peer has, try
1177 * again with a larger value.
1178 */
1179 dout("process_connect got RETRY my seq = %u, peer_seq = %u\n",
1180 le32_to_cpu(con->out_connect.connect_seq),
1181 le32_to_cpu(con->in_connect.connect_seq));
1182 con->connect_seq = le32_to_cpu(con->in_connect.connect_seq);
Sage Weileed0ef22009-11-10 14:34:36 -08001183 prepare_write_connect(con->msgr, con, 0);
Sage Weil31b80062009-10-06 11:31:13 -07001184 prepare_read_connect(con);
1185 break;
1186
1187 case CEPH_MSGR_TAG_RETRY_GLOBAL:
1188 /*
1189 * If we sent a smaller global_seq than the peer has, try
1190 * again with a larger value.
1191 */
Sage Weileed0ef22009-11-10 14:34:36 -08001192 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
Sage Weil31b80062009-10-06 11:31:13 -07001193 con->peer_global_seq,
1194 le32_to_cpu(con->in_connect.global_seq));
1195 get_global_seq(con->msgr,
1196 le32_to_cpu(con->in_connect.global_seq));
Sage Weileed0ef22009-11-10 14:34:36 -08001197 prepare_write_connect(con->msgr, con, 0);
Sage Weil31b80062009-10-06 11:31:13 -07001198 prepare_read_connect(con);
1199 break;
1200
1201 case CEPH_MSGR_TAG_READY:
Sage Weil04a419f2009-12-23 09:30:21 -08001202 if (req_feat & ~server_feat) {
1203 pr_err("%s%lld %s protocol feature mismatch,"
1204 " my required %llx > server's %llx, need %llx\n",
1205 ENTITY_NAME(con->peer_name),
1206 pr_addr(&con->peer_addr.in_addr),
1207 req_feat, server_feat, req_feat & ~server_feat);
1208 con->error_msg = "missing required protocol features";
1209 fail_protocol(con);
1210 return -1;
1211 }
Sage Weil31b80062009-10-06 11:31:13 -07001212 clear_bit(CONNECTING, &con->state);
Sage Weil31b80062009-10-06 11:31:13 -07001213 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
1214 con->connect_seq++;
1215 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1216 con->peer_global_seq,
1217 le32_to_cpu(con->in_reply.connect_seq),
1218 con->connect_seq);
1219 WARN_ON(con->connect_seq !=
1220 le32_to_cpu(con->in_reply.connect_seq));
Sage Weil92ac41d2009-12-14 14:56:56 -08001221
1222 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
1223 set_bit(LOSSYTX, &con->state);
1224
Sage Weil31b80062009-10-06 11:31:13 -07001225 prepare_read_tag(con);
1226 break;
1227
1228 case CEPH_MSGR_TAG_WAIT:
1229 /*
1230 * If there is a connection race (we are opening
1231 * connections to each other), one of us may just have
1232 * to WAIT. This shouldn't happen if we are the
1233 * client.
1234 */
1235 pr_err("process_connect peer connecting WAIT\n");
1236
1237 default:
1238 pr_err("connect protocol error, will retry\n");
1239 con->error_msg = "protocol error, garbage tag during connect";
1240 return -1;
1241 }
1242 return 0;
1243}
1244
1245
1246/*
1247 * read (part of) an ack
1248 */
1249static int read_partial_ack(struct ceph_connection *con)
1250{
1251 int to = 0;
1252
1253 return read_partial(con, &to, sizeof(con->in_temp_ack),
1254 &con->in_temp_ack);
1255}
1256
1257
1258/*
1259 * We can finally discard anything that's been acked.
1260 */
1261static void process_ack(struct ceph_connection *con)
1262{
1263 struct ceph_msg *m;
1264 u64 ack = le64_to_cpu(con->in_temp_ack);
1265 u64 seq;
1266
Sage Weil31b80062009-10-06 11:31:13 -07001267 while (!list_empty(&con->out_sent)) {
1268 m = list_first_entry(&con->out_sent, struct ceph_msg,
1269 list_head);
1270 seq = le64_to_cpu(m->hdr.seq);
1271 if (seq > ack)
1272 break;
1273 dout("got ack for seq %llu type %d at %p\n", seq,
1274 le16_to_cpu(m->hdr.type), m);
1275 ceph_msg_remove(m);
1276 }
Sage Weil31b80062009-10-06 11:31:13 -07001277 prepare_read_tag(con);
1278}
1279
1280
1281
1282
Yehuda Sadeh24504182010-01-08 13:58:34 -08001283static int read_partial_message_section(struct ceph_connection *con,
1284 struct kvec *section, unsigned int sec_len,
1285 u32 *crc)
1286{
1287 int left;
1288 int ret;
Sage Weil31b80062009-10-06 11:31:13 -07001289
Yehuda Sadeh24504182010-01-08 13:58:34 -08001290 BUG_ON(!section);
Sage Weil31b80062009-10-06 11:31:13 -07001291
Yehuda Sadeh24504182010-01-08 13:58:34 -08001292 while (section->iov_len < sec_len) {
1293 BUG_ON(section->iov_base == NULL);
1294 left = sec_len - section->iov_len;
1295 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
1296 section->iov_len, left);
1297 if (ret <= 0)
1298 return ret;
1299 section->iov_len += ret;
1300 if (section->iov_len == sec_len)
1301 *crc = crc32c(0, section->iov_base,
1302 section->iov_len);
1303 }
1304
1305 return 1;
1306}
1307
1308static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con,
1309 struct ceph_msg_header *hdr,
1310 int *skip);
Sage Weil31b80062009-10-06 11:31:13 -07001311/*
1312 * read (part of) a message.
1313 */
1314static int read_partial_message(struct ceph_connection *con)
1315{
1316 struct ceph_msg *m = con->in_msg;
1317 void *p;
1318 int ret;
Yehuda Sadeh9d7f0f12010-01-11 10:32:02 -08001319 int to, left;
Sage Weil31b80062009-10-06 11:31:13 -07001320 unsigned front_len, middle_len, data_len, data_off;
1321 int datacrc = con->msgr->nocrc;
Yehuda Sadeh24504182010-01-08 13:58:34 -08001322 int skip;
Sage Weil31b80062009-10-06 11:31:13 -07001323
1324 dout("read_partial_message con %p msg %p\n", con, m);
1325
1326 /* header */
1327 while (con->in_base_pos < sizeof(con->in_hdr)) {
1328 left = sizeof(con->in_hdr) - con->in_base_pos;
1329 ret = ceph_tcp_recvmsg(con->sock,
1330 (char *)&con->in_hdr + con->in_base_pos,
1331 left);
1332 if (ret <= 0)
1333 return ret;
1334 con->in_base_pos += ret;
1335 if (con->in_base_pos == sizeof(con->in_hdr)) {
1336 u32 crc = crc32c(0, (void *)&con->in_hdr,
1337 sizeof(con->in_hdr) - sizeof(con->in_hdr.crc));
1338 if (crc != le32_to_cpu(con->in_hdr.crc)) {
1339 pr_err("read_partial_message bad hdr "
1340 " crc %u != expected %u\n",
1341 crc, con->in_hdr.crc);
1342 return -EBADMSG;
1343 }
1344 }
1345 }
Sage Weil31b80062009-10-06 11:31:13 -07001346 front_len = le32_to_cpu(con->in_hdr.front_len);
1347 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
1348 return -EIO;
1349 middle_len = le32_to_cpu(con->in_hdr.middle_len);
1350 if (middle_len > CEPH_MSG_MAX_DATA_LEN)
1351 return -EIO;
1352 data_len = le32_to_cpu(con->in_hdr.data_len);
1353 if (data_len > CEPH_MSG_MAX_DATA_LEN)
1354 return -EIO;
Yehuda Sadeh9d7f0f12010-01-11 10:32:02 -08001355 data_off = le16_to_cpu(con->in_hdr.data_off);
Sage Weil31b80062009-10-06 11:31:13 -07001356
1357 /* allocate message? */
1358 if (!con->in_msg) {
1359 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
1360 con->in_hdr.front_len, con->in_hdr.data_len);
Yehuda Sadeh24504182010-01-08 13:58:34 -08001361 con->in_msg = ceph_alloc_msg(con, &con->in_hdr, &skip);
1362 if (skip) {
Sage Weil31b80062009-10-06 11:31:13 -07001363 /* skip this message */
Sage Weil5b3a4db2010-02-19 21:43:23 -08001364 dout("alloc_msg returned NULL, skipping message\n");
Sage Weil31b80062009-10-06 11:31:13 -07001365 con->in_base_pos = -front_len - middle_len - data_len -
1366 sizeof(m->footer);
1367 con->in_tag = CEPH_MSGR_TAG_READY;
1368 return 0;
1369 }
1370 if (IS_ERR(con->in_msg)) {
1371 ret = PTR_ERR(con->in_msg);
1372 con->in_msg = NULL;
Sage Weil5b3a4db2010-02-19 21:43:23 -08001373 con->error_msg =
1374 "error allocating memory for incoming message";
Sage Weil31b80062009-10-06 11:31:13 -07001375 return ret;
1376 }
1377 m = con->in_msg;
1378 m->front.iov_len = 0; /* haven't read it yet */
Yehuda Sadeh24504182010-01-08 13:58:34 -08001379 if (m->middle)
1380 m->middle->vec.iov_len = 0;
Yehuda Sadeh9d7f0f12010-01-11 10:32:02 -08001381
1382 con->in_msg_pos.page = 0;
1383 con->in_msg_pos.page_pos = data_off & ~PAGE_MASK;
1384 con->in_msg_pos.data_pos = 0;
Sage Weil31b80062009-10-06 11:31:13 -07001385 }
1386
1387 /* front */
Yehuda Sadeh24504182010-01-08 13:58:34 -08001388 ret = read_partial_message_section(con, &m->front, front_len,
1389 &con->in_front_crc);
1390 if (ret <= 0)
1391 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07001392
1393 /* middle */
Yehuda Sadeh24504182010-01-08 13:58:34 -08001394 if (m->middle) {
1395 ret = read_partial_message_section(con, &m->middle->vec, middle_len,
1396 &con->in_middle_crc);
Sage Weil31b80062009-10-06 11:31:13 -07001397 if (ret <= 0)
1398 return ret;
Sage Weil31b80062009-10-06 11:31:13 -07001399 }
1400
1401 /* (page) data */
Sage Weil31b80062009-10-06 11:31:13 -07001402 while (con->in_msg_pos.data_pos < data_len) {
1403 left = min((int)(data_len - con->in_msg_pos.data_pos),
1404 (int)(PAGE_SIZE - con->in_msg_pos.page_pos));
1405 BUG_ON(m->pages == NULL);
1406 p = kmap(m->pages[con->in_msg_pos.page]);
1407 ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos,
1408 left);
1409 if (ret > 0 && datacrc)
1410 con->in_data_crc =
1411 crc32c(con->in_data_crc,
1412 p + con->in_msg_pos.page_pos, ret);
1413 kunmap(m->pages[con->in_msg_pos.page]);
1414 if (ret <= 0)
1415 return ret;
1416 con->in_msg_pos.data_pos += ret;
1417 con->in_msg_pos.page_pos += ret;
1418 if (con->in_msg_pos.page_pos == PAGE_SIZE) {
1419 con->in_msg_pos.page_pos = 0;
1420 con->in_msg_pos.page++;
1421 }
1422 }
1423
Sage Weil31b80062009-10-06 11:31:13 -07001424 /* footer */
1425 to = sizeof(m->hdr) + sizeof(m->footer);
1426 while (con->in_base_pos < to) {
1427 left = to - con->in_base_pos;
1428 ret = ceph_tcp_recvmsg(con->sock, (char *)&m->footer +
1429 (con->in_base_pos - sizeof(m->hdr)),
1430 left);
1431 if (ret <= 0)
1432 return ret;
1433 con->in_base_pos += ret;
1434 }
1435 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1436 m, front_len, m->footer.front_crc, middle_len,
1437 m->footer.middle_crc, data_len, m->footer.data_crc);
1438
1439 /* crc ok? */
1440 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
1441 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1442 m, con->in_front_crc, m->footer.front_crc);
1443 return -EBADMSG;
1444 }
1445 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
1446 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1447 m, con->in_middle_crc, m->footer.middle_crc);
1448 return -EBADMSG;
1449 }
1450 if (datacrc &&
1451 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
1452 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
1453 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
1454 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
1455 return -EBADMSG;
1456 }
1457
1458 return 1; /* done! */
1459}
1460
1461/*
1462 * Process message. This happens in the worker thread. The callback should
1463 * be careful not to do anything that waits on other incoming messages or it
1464 * may deadlock.
1465 */
1466static void process_message(struct ceph_connection *con)
1467{
Sage Weil5e095e82009-12-14 14:30:34 -08001468 struct ceph_msg *msg;
Sage Weil31b80062009-10-06 11:31:13 -07001469
Sage Weil5e095e82009-12-14 14:30:34 -08001470 msg = con->in_msg;
Sage Weil31b80062009-10-06 11:31:13 -07001471 con->in_msg = NULL;
1472
1473 /* if first message, set peer_name */
1474 if (con->peer_name.type == 0)
1475 con->peer_name = msg->hdr.src.name;
1476
Sage Weil31b80062009-10-06 11:31:13 -07001477 con->in_seq++;
Sage Weilec302642009-12-22 10:43:42 -08001478 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07001479
1480 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
1481 msg, le64_to_cpu(msg->hdr.seq),
1482 ENTITY_NAME(msg->hdr.src.name),
1483 le16_to_cpu(msg->hdr.type),
1484 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
1485 le32_to_cpu(msg->hdr.front_len),
1486 le32_to_cpu(msg->hdr.data_len),
1487 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
1488 con->ops->dispatch(con, msg);
Sage Weilec302642009-12-22 10:43:42 -08001489
1490 mutex_lock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07001491 prepare_read_tag(con);
1492}
1493
1494
1495/*
1496 * Write something to the socket. Called in a worker thread when the
1497 * socket appears to be writeable and we have something ready to send.
1498 */
1499static int try_write(struct ceph_connection *con)
1500{
1501 struct ceph_messenger *msgr = con->msgr;
1502 int ret = 1;
1503
1504 dout("try_write start %p state %lu nref %d\n", con, con->state,
1505 atomic_read(&con->nref));
1506
Sage Weilec302642009-12-22 10:43:42 -08001507 mutex_lock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07001508more:
1509 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
1510
1511 /* open the socket first? */
1512 if (con->sock == NULL) {
1513 /*
1514 * if we were STANDBY and are reconnecting _this_
1515 * connection, bump connect_seq now. Always bump
1516 * global_seq.
1517 */
1518 if (test_and_clear_bit(STANDBY, &con->state))
1519 con->connect_seq++;
1520
Sage Weileed0ef22009-11-10 14:34:36 -08001521 prepare_write_banner(msgr, con);
1522 prepare_write_connect(msgr, con, 1);
1523 prepare_read_banner(con);
Sage Weil31b80062009-10-06 11:31:13 -07001524 set_bit(CONNECTING, &con->state);
Sage Weileed0ef22009-11-10 14:34:36 -08001525 clear_bit(NEGOTIATING, &con->state);
Sage Weil31b80062009-10-06 11:31:13 -07001526
Sage Weilcf3e5c42009-12-11 09:48:05 -08001527 BUG_ON(con->in_msg);
Sage Weil31b80062009-10-06 11:31:13 -07001528 con->in_tag = CEPH_MSGR_TAG_READY;
1529 dout("try_write initiating connect on %p new state %lu\n",
1530 con, con->state);
1531 con->sock = ceph_tcp_connect(con);
1532 if (IS_ERR(con->sock)) {
1533 con->sock = NULL;
1534 con->error_msg = "connect error";
1535 ret = -1;
1536 goto out;
1537 }
1538 }
1539
1540more_kvec:
1541 /* kvec data queued? */
1542 if (con->out_skip) {
1543 ret = write_partial_skip(con);
1544 if (ret <= 0)
1545 goto done;
1546 if (ret < 0) {
1547 dout("try_write write_partial_skip err %d\n", ret);
1548 goto done;
1549 }
1550 }
1551 if (con->out_kvec_left) {
1552 ret = write_partial_kvec(con);
1553 if (ret <= 0)
1554 goto done;
Sage Weil31b80062009-10-06 11:31:13 -07001555 }
1556
1557 /* msg pages? */
1558 if (con->out_msg) {
Sage Weilc86a2932009-12-14 14:04:30 -08001559 if (con->out_msg_done) {
1560 ceph_msg_put(con->out_msg);
1561 con->out_msg = NULL; /* we're done with this one */
1562 goto do_next;
1563 }
1564
Sage Weil31b80062009-10-06 11:31:13 -07001565 ret = write_partial_msg_pages(con);
1566 if (ret == 1)
1567 goto more_kvec; /* we need to send the footer, too! */
1568 if (ret == 0)
1569 goto done;
1570 if (ret < 0) {
1571 dout("try_write write_partial_msg_pages err %d\n",
1572 ret);
1573 goto done;
1574 }
1575 }
1576
Sage Weilc86a2932009-12-14 14:04:30 -08001577do_next:
Sage Weil31b80062009-10-06 11:31:13 -07001578 if (!test_bit(CONNECTING, &con->state)) {
1579 /* is anything else pending? */
1580 if (!list_empty(&con->out_queue)) {
1581 prepare_write_message(con);
1582 goto more;
1583 }
1584 if (con->in_seq > con->in_seq_acked) {
1585 prepare_write_ack(con);
1586 goto more;
1587 }
1588 if (test_and_clear_bit(KEEPALIVE_PENDING, &con->state)) {
1589 prepare_write_keepalive(con);
1590 goto more;
1591 }
1592 }
1593
1594 /* Nothing to do! */
1595 clear_bit(WRITE_PENDING, &con->state);
1596 dout("try_write nothing else to write.\n");
1597done:
1598 ret = 0;
1599out:
Sage Weilec302642009-12-22 10:43:42 -08001600 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07001601 dout("try_write done on %p\n", con);
1602 return ret;
1603}
1604
1605
1606
1607/*
1608 * Read what we can from the socket.
1609 */
1610static int try_read(struct ceph_connection *con)
1611{
1612 struct ceph_messenger *msgr;
1613 int ret = -1;
1614
1615 if (!con->sock)
1616 return 0;
1617
1618 if (test_bit(STANDBY, &con->state))
1619 return 0;
1620
1621 dout("try_read start on %p\n", con);
1622 msgr = con->msgr;
1623
Sage Weilec302642009-12-22 10:43:42 -08001624 mutex_lock(&con->mutex);
1625
Sage Weil31b80062009-10-06 11:31:13 -07001626more:
1627 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
1628 con->in_base_pos);
1629 if (test_bit(CONNECTING, &con->state)) {
Sage Weileed0ef22009-11-10 14:34:36 -08001630 if (!test_bit(NEGOTIATING, &con->state)) {
1631 dout("try_read connecting\n");
1632 ret = read_partial_banner(con);
1633 if (ret <= 0)
1634 goto done;
1635 if (process_banner(con) < 0) {
1636 ret = -1;
1637 goto out;
1638 }
1639 }
Sage Weil31b80062009-10-06 11:31:13 -07001640 ret = read_partial_connect(con);
1641 if (ret <= 0)
1642 goto done;
1643 if (process_connect(con) < 0) {
1644 ret = -1;
1645 goto out;
1646 }
1647 goto more;
1648 }
1649
1650 if (con->in_base_pos < 0) {
1651 /*
1652 * skipping + discarding content.
1653 *
1654 * FIXME: there must be a better way to do this!
1655 */
1656 static char buf[1024];
1657 int skip = min(1024, -con->in_base_pos);
1658 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
1659 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
1660 if (ret <= 0)
1661 goto done;
1662 con->in_base_pos += ret;
1663 if (con->in_base_pos)
1664 goto more;
1665 }
1666 if (con->in_tag == CEPH_MSGR_TAG_READY) {
1667 /*
1668 * what's next?
1669 */
1670 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
1671 if (ret <= 0)
1672 goto done;
1673 dout("try_read got tag %d\n", (int)con->in_tag);
1674 switch (con->in_tag) {
1675 case CEPH_MSGR_TAG_MSG:
1676 prepare_read_message(con);
1677 break;
1678 case CEPH_MSGR_TAG_ACK:
1679 prepare_read_ack(con);
1680 break;
1681 case CEPH_MSGR_TAG_CLOSE:
1682 set_bit(CLOSED, &con->state); /* fixme */
1683 goto done;
1684 default:
1685 goto bad_tag;
1686 }
1687 }
1688 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
1689 ret = read_partial_message(con);
1690 if (ret <= 0) {
1691 switch (ret) {
1692 case -EBADMSG:
1693 con->error_msg = "bad crc";
1694 ret = -EIO;
1695 goto out;
1696 case -EIO:
1697 con->error_msg = "io error";
1698 goto out;
1699 default:
1700 goto done;
1701 }
1702 }
1703 if (con->in_tag == CEPH_MSGR_TAG_READY)
1704 goto more;
1705 process_message(con);
1706 goto more;
1707 }
1708 if (con->in_tag == CEPH_MSGR_TAG_ACK) {
1709 ret = read_partial_ack(con);
1710 if (ret <= 0)
1711 goto done;
1712 process_ack(con);
1713 goto more;
1714 }
1715
1716done:
1717 ret = 0;
1718out:
Sage Weilec302642009-12-22 10:43:42 -08001719 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07001720 dout("try_read done on %p\n", con);
1721 return ret;
1722
1723bad_tag:
1724 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
1725 con->error_msg = "protocol error, garbage tag";
1726 ret = -1;
1727 goto out;
1728}
1729
1730
1731/*
1732 * Atomically queue work on a connection. Bump @con reference to
1733 * avoid races with connection teardown.
1734 *
1735 * There is some trickery going on with QUEUED and BUSY because we
1736 * only want a _single_ thread operating on each connection at any
1737 * point in time, but we want to use all available CPUs.
1738 *
1739 * The worker thread only proceeds if it can atomically set BUSY. It
1740 * clears QUEUED and does it's thing. When it thinks it's done, it
1741 * clears BUSY, then rechecks QUEUED.. if it's set again, it loops
1742 * (tries again to set BUSY).
1743 *
1744 * To queue work, we first set QUEUED, _then_ if BUSY isn't set, we
1745 * try to queue work. If that fails (work is already queued, or BUSY)
1746 * we give up (work also already being done or is queued) but leave QUEUED
1747 * set so that the worker thread will loop if necessary.
1748 */
1749static void queue_con(struct ceph_connection *con)
1750{
1751 if (test_bit(DEAD, &con->state)) {
1752 dout("queue_con %p ignoring: DEAD\n",
1753 con);
1754 return;
1755 }
1756
1757 if (!con->ops->get(con)) {
1758 dout("queue_con %p ref count 0\n", con);
1759 return;
1760 }
1761
1762 set_bit(QUEUED, &con->state);
1763 if (test_bit(BUSY, &con->state)) {
1764 dout("queue_con %p - already BUSY\n", con);
1765 con->ops->put(con);
1766 } else if (!queue_work(ceph_msgr_wq, &con->work.work)) {
1767 dout("queue_con %p - already queued\n", con);
1768 con->ops->put(con);
1769 } else {
1770 dout("queue_con %p\n", con);
1771 }
1772}
1773
1774/*
1775 * Do some work on a connection. Drop a connection ref when we're done.
1776 */
1777static void con_work(struct work_struct *work)
1778{
1779 struct ceph_connection *con = container_of(work, struct ceph_connection,
1780 work.work);
1781 int backoff = 0;
1782
1783more:
1784 if (test_and_set_bit(BUSY, &con->state) != 0) {
1785 dout("con_work %p BUSY already set\n", con);
1786 goto out;
1787 }
1788 dout("con_work %p start, clearing QUEUED\n", con);
1789 clear_bit(QUEUED, &con->state);
1790
1791 if (test_bit(CLOSED, &con->state)) { /* e.g. if we are replaced */
1792 dout("con_work CLOSED\n");
1793 con_close_socket(con);
1794 goto done;
1795 }
1796 if (test_and_clear_bit(OPENING, &con->state)) {
1797 /* reopen w/ new peer */
1798 dout("con_work OPENING\n");
1799 con_close_socket(con);
1800 }
1801
1802 if (test_and_clear_bit(SOCK_CLOSED, &con->state) ||
1803 try_read(con) < 0 ||
1804 try_write(con) < 0) {
1805 backoff = 1;
1806 ceph_fault(con); /* error/fault path */
1807 }
1808
1809done:
1810 clear_bit(BUSY, &con->state);
1811 dout("con->state=%lu\n", con->state);
1812 if (test_bit(QUEUED, &con->state)) {
Sage Weile2663ab2010-02-16 22:01:03 -08001813 if (!backoff || test_bit(OPENING, &con->state)) {
Sage Weil31b80062009-10-06 11:31:13 -07001814 dout("con_work %p QUEUED reset, looping\n", con);
1815 goto more;
1816 }
1817 dout("con_work %p QUEUED reset, but just faulted\n", con);
1818 clear_bit(QUEUED, &con->state);
1819 }
1820 dout("con_work %p done\n", con);
1821
1822out:
1823 con->ops->put(con);
1824}
1825
1826
1827/*
1828 * Generic error/fault handler. A retry mechanism is used with
1829 * exponential backoff
1830 */
1831static void ceph_fault(struct ceph_connection *con)
1832{
1833 pr_err("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
1834 pr_addr(&con->peer_addr.in_addr), con->error_msg);
1835 dout("fault %p state %lu to peer %s\n",
1836 con, con->state, pr_addr(&con->peer_addr.in_addr));
1837
1838 if (test_bit(LOSSYTX, &con->state)) {
1839 dout("fault on LOSSYTX channel\n");
1840 goto out;
1841 }
1842
1843 clear_bit(BUSY, &con->state); /* to avoid an improbable race */
1844
Sage Weilec302642009-12-22 10:43:42 -08001845 mutex_lock(&con->mutex);
Sage Weil91e45ce32010-02-15 12:05:09 -08001846 if (test_bit(CLOSED, &con->state))
1847 goto out_unlock;
Sage Weilec302642009-12-22 10:43:42 -08001848
Sage Weil31b80062009-10-06 11:31:13 -07001849 con_close_socket(con);
Sage Weil5e095e82009-12-14 14:30:34 -08001850
1851 if (con->in_msg) {
1852 ceph_msg_put(con->in_msg);
1853 con->in_msg = NULL;
1854 }
Sage Weil31b80062009-10-06 11:31:13 -07001855
Sage Weil9bd2e6f2010-02-02 16:21:06 -08001856
Sage Weil31b80062009-10-06 11:31:13 -07001857 /* If there are no messages in the queue, place the connection
1858 * in a STANDBY state (i.e., don't try to reconnect just yet). */
Sage Weil31b80062009-10-06 11:31:13 -07001859 if (list_empty(&con->out_queue) && !con->out_keepalive_pending) {
1860 dout("fault setting STANDBY\n");
1861 set_bit(STANDBY, &con->state);
Sage Weilec302642009-12-22 10:43:42 -08001862 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07001863 goto out;
1864 }
1865
1866 /* Requeue anything that hasn't been acked, and retry after a
1867 * delay. */
1868 list_splice_init(&con->out_sent, &con->out_queue);
Sage Weil31b80062009-10-06 11:31:13 -07001869
1870 if (con->delay == 0)
1871 con->delay = BASE_DELAY_INTERVAL;
1872 else if (con->delay < MAX_DELAY_INTERVAL)
1873 con->delay *= 2;
1874
1875 /* explicitly schedule work to try to reconnect again later. */
1876 dout("fault queueing %p delay %lu\n", con, con->delay);
1877 con->ops->get(con);
1878 if (queue_delayed_work(ceph_msgr_wq, &con->work,
1879 round_jiffies_relative(con->delay)) == 0)
1880 con->ops->put(con);
1881
Sage Weil91e45ce32010-02-15 12:05:09 -08001882out_unlock:
1883 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07001884out:
Sage Weil161fd652010-02-25 12:38:57 -08001885 /*
1886 * in case we faulted due to authentication, invalidate our
1887 * current tickets so that we can get new ones.
1888 */
1889 if (con->auth_retry && con->ops->invalidate_authorizer) {
1890 dout("calling invalidate_authorizer()\n");
1891 con->ops->invalidate_authorizer(con);
1892 }
1893
Sage Weil31b80062009-10-06 11:31:13 -07001894 if (con->ops->fault)
1895 con->ops->fault(con);
1896}
1897
1898
1899
1900/*
1901 * create a new messenger instance
1902 */
1903struct ceph_messenger *ceph_messenger_create(struct ceph_entity_addr *myaddr)
1904{
1905 struct ceph_messenger *msgr;
1906
1907 msgr = kzalloc(sizeof(*msgr), GFP_KERNEL);
1908 if (msgr == NULL)
1909 return ERR_PTR(-ENOMEM);
1910
1911 spin_lock_init(&msgr->global_seq_lock);
1912
1913 /* the zero page is needed if a request is "canceled" while the message
1914 * is being written over the socket */
1915 msgr->zero_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1916 if (!msgr->zero_page) {
1917 kfree(msgr);
1918 return ERR_PTR(-ENOMEM);
1919 }
1920 kmap(msgr->zero_page);
1921
1922 if (myaddr)
1923 msgr->inst.addr = *myaddr;
1924
1925 /* select a random nonce */
Sage Weilac8839d2010-01-27 14:28:10 -08001926 msgr->inst.addr.type = 0;
Sage Weil103e2d32010-01-07 16:12:36 -08001927 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
Sage Weil63f2d212009-11-03 15:17:56 -08001928 encode_my_addr(msgr);
Sage Weil31b80062009-10-06 11:31:13 -07001929
1930 dout("messenger_create %p\n", msgr);
1931 return msgr;
1932}
1933
1934void ceph_messenger_destroy(struct ceph_messenger *msgr)
1935{
1936 dout("destroy %p\n", msgr);
1937 kunmap(msgr->zero_page);
1938 __free_page(msgr->zero_page);
1939 kfree(msgr);
1940 dout("destroyed messenger %p\n", msgr);
1941}
1942
1943/*
1944 * Queue up an outgoing message on the given connection.
1945 */
1946void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
1947{
1948 if (test_bit(CLOSED, &con->state)) {
1949 dout("con_send %p closed, dropping %p\n", con, msg);
1950 ceph_msg_put(msg);
1951 return;
1952 }
1953
1954 /* set src+dst */
Sage Weil63f2d212009-11-03 15:17:56 -08001955 msg->hdr.src.name = con->msgr->inst.name;
1956 msg->hdr.src.addr = con->msgr->my_enc_addr;
1957 msg->hdr.orig_src = msg->hdr.src;
Sage Weil31b80062009-10-06 11:31:13 -07001958
1959 /* queue */
Sage Weilec302642009-12-22 10:43:42 -08001960 mutex_lock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07001961 BUG_ON(!list_empty(&msg->list_head));
1962 list_add_tail(&msg->list_head, &con->out_queue);
1963 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
1964 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
1965 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
1966 le32_to_cpu(msg->hdr.front_len),
1967 le32_to_cpu(msg->hdr.middle_len),
1968 le32_to_cpu(msg->hdr.data_len));
Sage Weilec302642009-12-22 10:43:42 -08001969 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07001970
1971 /* if there wasn't anything waiting to send before, queue
1972 * new work */
1973 if (test_and_set_bit(WRITE_PENDING, &con->state) == 0)
1974 queue_con(con);
1975}
1976
1977/*
1978 * Revoke a message that was previously queued for send
1979 */
1980void ceph_con_revoke(struct ceph_connection *con, struct ceph_msg *msg)
1981{
Sage Weilec302642009-12-22 10:43:42 -08001982 mutex_lock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07001983 if (!list_empty(&msg->list_head)) {
1984 dout("con_revoke %p msg %p\n", con, msg);
1985 list_del_init(&msg->list_head);
1986 ceph_msg_put(msg);
1987 msg->hdr.seq = 0;
Sage Weilc86a2932009-12-14 14:04:30 -08001988 if (con->out_msg == msg) {
1989 ceph_msg_put(con->out_msg);
Sage Weil31b80062009-10-06 11:31:13 -07001990 con->out_msg = NULL;
Sage Weilc86a2932009-12-14 14:04:30 -08001991 }
Sage Weil31b80062009-10-06 11:31:13 -07001992 if (con->out_kvec_is_msg) {
1993 con->out_skip = con->out_kvec_bytes;
1994 con->out_kvec_is_msg = false;
1995 }
1996 } else {
1997 dout("con_revoke %p msg %p - not queued (sent?)\n", con, msg);
1998 }
Sage Weilec302642009-12-22 10:43:42 -08001999 mutex_unlock(&con->mutex);
Sage Weil31b80062009-10-06 11:31:13 -07002000}
2001
2002/*
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -08002003 * Revoke a message that we may be reading data into
Sage Weil350b1c32009-12-22 10:45:45 -08002004 */
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -08002005void ceph_con_revoke_message(struct ceph_connection *con, struct ceph_msg *msg)
Sage Weil350b1c32009-12-22 10:45:45 -08002006{
2007 mutex_lock(&con->mutex);
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -08002008 if (con->in_msg && con->in_msg == msg) {
2009 unsigned front_len = le32_to_cpu(con->in_hdr.front_len);
2010 unsigned middle_len = le32_to_cpu(con->in_hdr.middle_len);
Sage Weil350b1c32009-12-22 10:45:45 -08002011 unsigned data_len = le32_to_cpu(con->in_hdr.data_len);
2012
2013 /* skip rest of message */
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -08002014 dout("con_revoke_pages %p msg %p revoked\n", con, msg);
Sage Weil350b1c32009-12-22 10:45:45 -08002015 con->in_base_pos = con->in_base_pos -
2016 sizeof(struct ceph_msg_header) -
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -08002017 front_len -
2018 middle_len -
2019 data_len -
Sage Weil350b1c32009-12-22 10:45:45 -08002020 sizeof(struct ceph_msg_footer);
Sage Weil350b1c32009-12-22 10:45:45 -08002021 ceph_msg_put(con->in_msg);
2022 con->in_msg = NULL;
2023 con->in_tag = CEPH_MSGR_TAG_READY;
2024 } else {
2025 dout("con_revoke_pages %p msg %p pages %p no-op\n",
Yehuda Sadeh0d59ab82010-01-13 17:03:23 -08002026 con, con->in_msg, msg);
Sage Weil350b1c32009-12-22 10:45:45 -08002027 }
2028 mutex_unlock(&con->mutex);
2029}
2030
2031/*
Sage Weil31b80062009-10-06 11:31:13 -07002032 * Queue a keepalive byte to ensure the tcp connection is alive.
2033 */
2034void ceph_con_keepalive(struct ceph_connection *con)
2035{
2036 if (test_and_set_bit(KEEPALIVE_PENDING, &con->state) == 0 &&
2037 test_and_set_bit(WRITE_PENDING, &con->state) == 0)
2038 queue_con(con);
2039}
2040
2041
2042/*
2043 * construct a new message with given type, size
2044 * the new msg has a ref count of 1.
2045 */
2046struct ceph_msg *ceph_msg_new(int type, int front_len,
2047 int page_len, int page_off, struct page **pages)
2048{
2049 struct ceph_msg *m;
2050
2051 m = kmalloc(sizeof(*m), GFP_NOFS);
2052 if (m == NULL)
2053 goto out;
Sage Weilc2e552e2009-12-07 15:55:05 -08002054 kref_init(&m->kref);
Sage Weil31b80062009-10-06 11:31:13 -07002055 INIT_LIST_HEAD(&m->list_head);
2056
2057 m->hdr.type = cpu_to_le16(type);
2058 m->hdr.front_len = cpu_to_le32(front_len);
2059 m->hdr.middle_len = 0;
2060 m->hdr.data_len = cpu_to_le32(page_len);
2061 m->hdr.data_off = cpu_to_le16(page_off);
2062 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
2063 m->footer.front_crc = 0;
2064 m->footer.middle_crc = 0;
2065 m->footer.data_crc = 0;
2066 m->front_max = front_len;
2067 m->front_is_vmalloc = false;
2068 m->more_to_follow = false;
2069 m->pool = NULL;
2070
2071 /* front */
2072 if (front_len) {
2073 if (front_len > PAGE_CACHE_SIZE) {
2074 m->front.iov_base = __vmalloc(front_len, GFP_NOFS,
2075 PAGE_KERNEL);
2076 m->front_is_vmalloc = true;
2077 } else {
2078 m->front.iov_base = kmalloc(front_len, GFP_NOFS);
2079 }
2080 if (m->front.iov_base == NULL) {
2081 pr_err("msg_new can't allocate %d bytes\n",
2082 front_len);
2083 goto out2;
2084 }
2085 } else {
2086 m->front.iov_base = NULL;
2087 }
2088 m->front.iov_len = front_len;
2089
2090 /* middle */
2091 m->middle = NULL;
2092
2093 /* data */
2094 m->nr_pages = calc_pages_for(page_off, page_len);
2095 m->pages = pages;
Sage Weil58bb3b32009-12-23 12:12:31 -08002096 m->pagelist = NULL;
Sage Weil31b80062009-10-06 11:31:13 -07002097
2098 dout("ceph_msg_new %p page %d~%d -> %d\n", m, page_off, page_len,
2099 m->nr_pages);
2100 return m;
2101
2102out2:
2103 ceph_msg_put(m);
2104out:
2105 pr_err("msg_new can't create type %d len %d\n", type, front_len);
2106 return ERR_PTR(-ENOMEM);
2107}
2108
2109/*
Sage Weil31b80062009-10-06 11:31:13 -07002110 * Allocate "middle" portion of a message, if it is needed and wasn't
2111 * allocated by alloc_msg. This allows us to read a small fixed-size
2112 * per-type header in the front and then gracefully fail (i.e.,
2113 * propagate the error to the caller based on info in the front) when
2114 * the middle is too large.
2115 */
Yehuda Sadeh24504182010-01-08 13:58:34 -08002116static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
Sage Weil31b80062009-10-06 11:31:13 -07002117{
2118 int type = le16_to_cpu(msg->hdr.type);
2119 int middle_len = le32_to_cpu(msg->hdr.middle_len);
2120
2121 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
2122 ceph_msg_type_name(type), middle_len);
2123 BUG_ON(!middle_len);
2124 BUG_ON(msg->middle);
2125
Sage Weilb6c1d5b2009-12-07 12:17:17 -08002126 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
Sage Weil31b80062009-10-06 11:31:13 -07002127 if (!msg->middle)
2128 return -ENOMEM;
2129 return 0;
2130}
2131
Yehuda Sadeh24504182010-01-08 13:58:34 -08002132/*
2133 * Generic message allocator, for incoming messages.
2134 */
2135static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con,
2136 struct ceph_msg_header *hdr,
2137 int *skip)
2138{
2139 int type = le16_to_cpu(hdr->type);
2140 int front_len = le32_to_cpu(hdr->front_len);
2141 int middle_len = le32_to_cpu(hdr->middle_len);
2142 struct ceph_msg *msg = NULL;
2143 int ret;
2144
2145 if (con->ops->alloc_msg) {
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002146 mutex_unlock(&con->mutex);
Yehuda Sadeh24504182010-01-08 13:58:34 -08002147 msg = con->ops->alloc_msg(con, hdr, skip);
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08002148 mutex_lock(&con->mutex);
Yehuda Sadeh24504182010-01-08 13:58:34 -08002149 if (IS_ERR(msg))
2150 return msg;
2151
2152 if (*skip)
2153 return NULL;
2154 }
2155 if (!msg) {
2156 *skip = 0;
2157 msg = ceph_msg_new(type, front_len, 0, 0, NULL);
2158 if (!msg) {
2159 pr_err("unable to allocate msg type %d len %d\n",
2160 type, front_len);
2161 return ERR_PTR(-ENOMEM);
2162 }
2163 }
Yehuda Sadeh9d7f0f12010-01-11 10:32:02 -08002164 memcpy(&msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
Yehuda Sadeh24504182010-01-08 13:58:34 -08002165
2166 if (middle_len) {
2167 ret = ceph_alloc_middle(con, msg);
2168
2169 if (ret < 0) {
2170 ceph_msg_put(msg);
2171 return msg;
2172 }
2173 }
Yehuda Sadeh9d7f0f12010-01-11 10:32:02 -08002174
Yehuda Sadeh24504182010-01-08 13:58:34 -08002175 return msg;
2176}
2177
Sage Weil31b80062009-10-06 11:31:13 -07002178
2179/*
2180 * Free a generically kmalloc'd message.
2181 */
2182void ceph_msg_kfree(struct ceph_msg *m)
2183{
2184 dout("msg_kfree %p\n", m);
2185 if (m->front_is_vmalloc)
2186 vfree(m->front.iov_base);
2187 else
2188 kfree(m->front.iov_base);
2189 kfree(m);
2190}
2191
2192/*
2193 * Drop a msg ref. Destroy as needed.
2194 */
Sage Weilc2e552e2009-12-07 15:55:05 -08002195void ceph_msg_last_put(struct kref *kref)
Sage Weil31b80062009-10-06 11:31:13 -07002196{
Sage Weilc2e552e2009-12-07 15:55:05 -08002197 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
Sage Weil31b80062009-10-06 11:31:13 -07002198
Sage Weilc2e552e2009-12-07 15:55:05 -08002199 dout("ceph_msg_put last one on %p\n", m);
2200 WARN_ON(!list_empty(&m->list_head));
Sage Weil31b80062009-10-06 11:31:13 -07002201
Sage Weilc2e552e2009-12-07 15:55:05 -08002202 /* drop middle, data, if any */
2203 if (m->middle) {
2204 ceph_buffer_put(m->middle);
2205 m->middle = NULL;
Sage Weil31b80062009-10-06 11:31:13 -07002206 }
Sage Weilc2e552e2009-12-07 15:55:05 -08002207 m->nr_pages = 0;
2208 m->pages = NULL;
2209
Sage Weil58bb3b32009-12-23 12:12:31 -08002210 if (m->pagelist) {
2211 ceph_pagelist_release(m->pagelist);
2212 kfree(m->pagelist);
2213 m->pagelist = NULL;
2214 }
2215
Sage Weilc2e552e2009-12-07 15:55:05 -08002216 if (m->pool)
2217 ceph_msgpool_put(m->pool, m);
2218 else
2219 ceph_msg_kfree(m);
Sage Weil31b80062009-10-06 11:31:13 -07002220}
Sage Weil9ec7cab2009-12-14 15:13:47 -08002221
2222void ceph_msg_dump(struct ceph_msg *msg)
2223{
2224 pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg,
2225 msg->front_max, msg->nr_pages);
2226 print_hex_dump(KERN_DEBUG, "header: ",
2227 DUMP_PREFIX_OFFSET, 16, 1,
2228 &msg->hdr, sizeof(msg->hdr), true);
2229 print_hex_dump(KERN_DEBUG, " front: ",
2230 DUMP_PREFIX_OFFSET, 16, 1,
2231 msg->front.iov_base, msg->front.iov_len, true);
2232 if (msg->middle)
2233 print_hex_dump(KERN_DEBUG, "middle: ",
2234 DUMP_PREFIX_OFFSET, 16, 1,
2235 msg->middle->vec.iov_base,
2236 msg->middle->vec.iov_len, true);
2237 print_hex_dump(KERN_DEBUG, "footer: ",
2238 DUMP_PREFIX_OFFSET, 16, 1,
2239 &msg->footer, sizeof(msg->footer), true);
2240}