aboutsummaryrefslogtreecommitdiffstats
path: root/fs/ceph/messenger.c
diff options
context:
space:
mode:
Diffstat (limited to 'fs/ceph/messenger.c')
-rw-r--r--fs/ceph/messenger.c2019
1 files changed, 2019 insertions, 0 deletions
diff --git a/fs/ceph/messenger.c b/fs/ceph/messenger.c
new file mode 100644
index 000000000000..63f7f1359385
--- /dev/null
+++ b/fs/ceph/messenger.c
@@ -0,0 +1,2019 @@
1#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"
15
16/*
17 * Ceph uses the messenger to exchange ceph_msg messages with other
18 * hosts in the system. The messenger provides ordered and reliable
19 * delivery. We tolerate TCP disconnects by reconnecting (with
20 * exponential backoff) in the case of a fault (disconnection, bad
21 * crc, protocol error). Acks allow sent messages to be discarded by
22 * the sender.
23 */
24
25/* static tag bytes (protocol control messages) */
26static char tag_msg = CEPH_MSGR_TAG_MSG;
27static char tag_ack = CEPH_MSGR_TAG_ACK;
28static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
29
30
31static void queue_con(struct ceph_connection *con);
32static void con_work(struct work_struct *);
33static void ceph_fault(struct ceph_connection *con);
34
35const char *ceph_name_type_str(int t)
36{
37 switch (t) {
38 case CEPH_ENTITY_TYPE_MON: return "mon";
39 case CEPH_ENTITY_TYPE_MDS: return "mds";
40 case CEPH_ENTITY_TYPE_OSD: return "osd";
41 case CEPH_ENTITY_TYPE_CLIENT: return "client";
42 case CEPH_ENTITY_TYPE_ADMIN: return "admin";
43 default: return "???";
44 }
45}
46
47/*
48 * nicely render a sockaddr as a string.
49 */
50#define MAX_ADDR_STR 20
51static char addr_str[MAX_ADDR_STR][40];
52static DEFINE_SPINLOCK(addr_str_lock);
53static int last_addr_str;
54
55const char *pr_addr(const struct sockaddr_storage *ss)
56{
57 int i;
58 char *s;
59 struct sockaddr_in *in4 = (void *)ss;
60 unsigned char *quad = (void *)&in4->sin_addr.s_addr;
61 struct sockaddr_in6 *in6 = (void *)ss;
62
63 spin_lock(&addr_str_lock);
64 i = last_addr_str++;
65 if (last_addr_str == MAX_ADDR_STR)
66 last_addr_str = 0;
67 spin_unlock(&addr_str_lock);
68 s = addr_str[i];
69
70 switch (ss->ss_family) {
71 case AF_INET:
72 sprintf(s, "%u.%u.%u.%u:%u",
73 (unsigned int)quad[0],
74 (unsigned int)quad[1],
75 (unsigned int)quad[2],
76 (unsigned int)quad[3],
77 (unsigned int)ntohs(in4->sin_port));
78 break;
79
80 case AF_INET6:
81 sprintf(s, "%04x:%04x:%04x:%04x:%04x:%04x:%04x:%04x:%u",
82 in6->sin6_addr.s6_addr16[0],
83 in6->sin6_addr.s6_addr16[1],
84 in6->sin6_addr.s6_addr16[2],
85 in6->sin6_addr.s6_addr16[3],
86 in6->sin6_addr.s6_addr16[4],
87 in6->sin6_addr.s6_addr16[5],
88 in6->sin6_addr.s6_addr16[6],
89 in6->sin6_addr.s6_addr16[7],
90 (unsigned int)ntohs(in6->sin6_port));
91 break;
92
93 default:
94 sprintf(s, "(unknown sockaddr family %d)", (int)ss->ss_family);
95 }
96
97 return s;
98}
99
100/*
101 * work queue for all reading and writing to/from the socket.
102 */
103struct workqueue_struct *ceph_msgr_wq;
104
105int __init ceph_msgr_init(void)
106{
107 ceph_msgr_wq = create_workqueue("ceph-msgr");
108 if (IS_ERR(ceph_msgr_wq)) {
109 int ret = PTR_ERR(ceph_msgr_wq);
110 pr_err("msgr_init failed to create workqueue: %d\n", ret);
111 ceph_msgr_wq = NULL;
112 return ret;
113 }
114 return 0;
115}
116
117void ceph_msgr_exit(void)
118{
119 destroy_workqueue(ceph_msgr_wq);
120}
121
122/*
123 * socket callback functions
124 */
125
126/* data available on socket, or listen socket received a connect */
127static void ceph_data_ready(struct sock *sk, int count_unused)
128{
129 struct ceph_connection *con =
130 (struct ceph_connection *)sk->sk_user_data;
131 if (sk->sk_state != TCP_CLOSE_WAIT) {
132 dout("ceph_data_ready on %p state = %lu, queueing work\n",
133 con, con->state);
134 queue_con(con);
135 }
136}
137
138/* socket has buffer space for writing */
139static void ceph_write_space(struct sock *sk)
140{
141 struct ceph_connection *con =
142 (struct ceph_connection *)sk->sk_user_data;
143
144 /* only queue to workqueue if there is data we want to write. */
145 if (test_bit(WRITE_PENDING, &con->state)) {
146 dout("ceph_write_space %p queueing write work\n", con);
147 queue_con(con);
148 } else {
149 dout("ceph_write_space %p nothing to write\n", con);
150 }
151
152 /* since we have our own write_space, clear the SOCK_NOSPACE flag */
153 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
154}
155
156/* socket's state has changed */
157static void ceph_state_change(struct sock *sk)
158{
159 struct ceph_connection *con =
160 (struct ceph_connection *)sk->sk_user_data;
161
162 dout("ceph_state_change %p state = %lu sk_state = %u\n",
163 con, con->state, sk->sk_state);
164
165 if (test_bit(CLOSED, &con->state))
166 return;
167
168 switch (sk->sk_state) {
169 case TCP_CLOSE:
170 dout("ceph_state_change TCP_CLOSE\n");
171 case TCP_CLOSE_WAIT:
172 dout("ceph_state_change TCP_CLOSE_WAIT\n");
173 if (test_and_set_bit(SOCK_CLOSED, &con->state) == 0) {
174 if (test_bit(CONNECTING, &con->state))
175 con->error_msg = "connection failed";
176 else
177 con->error_msg = "socket closed";
178 queue_con(con);
179 }
180 break;
181 case TCP_ESTABLISHED:
182 dout("ceph_state_change TCP_ESTABLISHED\n");
183 queue_con(con);
184 break;
185 }
186}
187
188/*
189 * set up socket callbacks
190 */
191static void set_sock_callbacks(struct socket *sock,
192 struct ceph_connection *con)
193{
194 struct sock *sk = sock->sk;
195 sk->sk_user_data = (void *)con;
196 sk->sk_data_ready = ceph_data_ready;
197 sk->sk_write_space = ceph_write_space;
198 sk->sk_state_change = ceph_state_change;
199}
200
201
202/*
203 * socket helpers
204 */
205
206/*
207 * initiate connection to a remote socket.
208 */
209static struct socket *ceph_tcp_connect(struct ceph_connection *con)
210{
211 struct sockaddr *paddr = (struct sockaddr *)&con->peer_addr.in_addr;
212 struct socket *sock;
213 int ret;
214
215 BUG_ON(con->sock);
216 ret = sock_create_kern(AF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
217 if (ret)
218 return ERR_PTR(ret);
219 con->sock = sock;
220 sock->sk->sk_allocation = GFP_NOFS;
221
222 set_sock_callbacks(sock, con);
223
224 dout("connect %s\n", pr_addr(&con->peer_addr.in_addr));
225
226 ret = sock->ops->connect(sock, paddr, sizeof(*paddr), O_NONBLOCK);
227 if (ret == -EINPROGRESS) {
228 dout("connect %s EINPROGRESS sk_state = %u\n",
229 pr_addr(&con->peer_addr.in_addr),
230 sock->sk->sk_state);
231 ret = 0;
232 }
233 if (ret < 0) {
234 pr_err("connect %s error %d\n",
235 pr_addr(&con->peer_addr.in_addr), ret);
236 sock_release(sock);
237 con->sock = NULL;
238 con->error_msg = "connect error";
239 }
240
241 if (ret < 0)
242 return ERR_PTR(ret);
243 return sock;
244}
245
246static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
247{
248 struct kvec iov = {buf, len};
249 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
250
251 return kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
252}
253
254/*
255 * write something. @more is true if caller will be sending more data
256 * shortly.
257 */
258static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
259 size_t kvlen, size_t len, int more)
260{
261 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
262
263 if (more)
264 msg.msg_flags |= MSG_MORE;
265 else
266 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
267
268 return kernel_sendmsg(sock, &msg, iov, kvlen, len);
269}
270
271
272/*
273 * Shutdown/close the socket for the given connection.
274 */
275static int con_close_socket(struct ceph_connection *con)
276{
277 int rc;
278
279 dout("con_close_socket on %p sock %p\n", con, con->sock);
280 if (!con->sock)
281 return 0;
282 set_bit(SOCK_CLOSED, &con->state);
283 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
284 sock_release(con->sock);
285 con->sock = NULL;
286 clear_bit(SOCK_CLOSED, &con->state);
287 return rc;
288}
289
290/*
291 * Reset a connection. Discard all incoming and outgoing messages
292 * and clear *_seq state.
293 */
294static void ceph_msg_remove(struct ceph_msg *msg)
295{
296 list_del_init(&msg->list_head);
297 ceph_msg_put(msg);
298}
299static void ceph_msg_remove_list(struct list_head *head)
300{
301 while (!list_empty(head)) {
302 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
303 list_head);
304 ceph_msg_remove(msg);
305 }
306}
307
308static void reset_connection(struct ceph_connection *con)
309{
310 /* reset connection, out_queue, msg_ and connect_seq */
311 /* discard existing out_queue and msg_seq */
312 mutex_lock(&con->out_mutex);
313 ceph_msg_remove_list(&con->out_queue);
314 ceph_msg_remove_list(&con->out_sent);
315
316 con->connect_seq = 0;
317 con->out_seq = 0;
318 con->out_msg = NULL;
319 con->in_seq = 0;
320 mutex_unlock(&con->out_mutex);
321}
322
323/*
324 * mark a peer down. drop any open connections.
325 */
326void ceph_con_close(struct ceph_connection *con)
327{
328 dout("con_close %p peer %s\n", con, pr_addr(&con->peer_addr.in_addr));
329 set_bit(CLOSED, &con->state); /* in case there's queued work */
330 clear_bit(STANDBY, &con->state); /* avoid connect_seq bump */
331 reset_connection(con);
332 queue_con(con);
333}
334
335/*
336 * clean up connection state
337 */
338void ceph_con_shutdown(struct ceph_connection *con)
339{
340 dout("con_shutdown %p\n", con);
341 reset_connection(con);
342 set_bit(DEAD, &con->state);
343 con_close_socket(con); /* silently ignore errors */
344}
345
346/*
347 * Reopen a closed connection, with a new peer address.
348 */
349void ceph_con_open(struct ceph_connection *con, struct ceph_entity_addr *addr)
350{
351 dout("con_open %p %s\n", con, pr_addr(&addr->in_addr));
352 set_bit(OPENING, &con->state);
353 clear_bit(CLOSED, &con->state);
354 memcpy(&con->peer_addr, addr, sizeof(*addr));
355 queue_con(con);
356}
357
358/*
359 * generic get/put
360 */
361struct ceph_connection *ceph_con_get(struct ceph_connection *con)
362{
363 dout("con_get %p nref = %d -> %d\n", con,
364 atomic_read(&con->nref), atomic_read(&con->nref) + 1);
365 if (atomic_inc_not_zero(&con->nref))
366 return con;
367 return NULL;
368}
369
370void ceph_con_put(struct ceph_connection *con)
371{
372 dout("con_put %p nref = %d -> %d\n", con,
373 atomic_read(&con->nref), atomic_read(&con->nref) - 1);
374 BUG_ON(atomic_read(&con->nref) == 0);
375 if (atomic_dec_and_test(&con->nref)) {
376 ceph_con_shutdown(con);
377 kfree(con);
378 }
379}
380
381/*
382 * initialize a new connection.
383 */
384void ceph_con_init(struct ceph_messenger *msgr, struct ceph_connection *con)
385{
386 dout("con_init %p\n", con);
387 memset(con, 0, sizeof(*con));
388 atomic_set(&con->nref, 1);
389 con->msgr = msgr;
390 mutex_init(&con->out_mutex);
391 INIT_LIST_HEAD(&con->out_queue);
392 INIT_LIST_HEAD(&con->out_sent);
393 INIT_DELAYED_WORK(&con->work, con_work);
394}
395
396
397/*
398 * We maintain a global counter to order connection attempts. Get
399 * a unique seq greater than @gt.
400 */
401static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
402{
403 u32 ret;
404
405 spin_lock(&msgr->global_seq_lock);
406 if (msgr->global_seq < gt)
407 msgr->global_seq = gt;
408 ret = ++msgr->global_seq;
409 spin_unlock(&msgr->global_seq_lock);
410 return ret;
411}
412
413
414/*
415 * Prepare footer for currently outgoing message, and finish things
416 * off. Assumes out_kvec* are already valid.. we just add on to the end.
417 */
418static void prepare_write_message_footer(struct ceph_connection *con, int v)
419{
420 struct ceph_msg *m = con->out_msg;
421
422 dout("prepare_write_message_footer %p\n", con);
423 con->out_kvec_is_msg = true;
424 con->out_kvec[v].iov_base = &m->footer;
425 con->out_kvec[v].iov_len = sizeof(m->footer);
426 con->out_kvec_bytes += sizeof(m->footer);
427 con->out_kvec_left++;
428 con->out_more = m->more_to_follow;
429 con->out_msg = NULL; /* we're done with this one */
430}
431
432/*
433 * Prepare headers for the next outgoing message.
434 */
435static void prepare_write_message(struct ceph_connection *con)
436{
437 struct ceph_msg *m;
438 int v = 0;
439
440 con->out_kvec_bytes = 0;
441 con->out_kvec_is_msg = true;
442
443 /* Sneak an ack in there first? If we can get it into the same
444 * TCP packet that's a good thing. */
445 if (con->in_seq > con->in_seq_acked) {
446 con->in_seq_acked = con->in_seq;
447 con->out_kvec[v].iov_base = &tag_ack;
448 con->out_kvec[v++].iov_len = 1;
449 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
450 con->out_kvec[v].iov_base = &con->out_temp_ack;
451 con->out_kvec[v++].iov_len = sizeof(con->out_temp_ack);
452 con->out_kvec_bytes = 1 + sizeof(con->out_temp_ack);
453 }
454
455 /* move message to sending/sent list */
456 m = list_first_entry(&con->out_queue,
457 struct ceph_msg, list_head);
458 list_move_tail(&m->list_head, &con->out_sent);
459 con->out_msg = m; /* we don't bother taking a reference here. */
460
461 m->hdr.seq = cpu_to_le64(++con->out_seq);
462
463 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n",
464 m, con->out_seq, le16_to_cpu(m->hdr.type),
465 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
466 le32_to_cpu(m->hdr.data_len),
467 m->nr_pages);
468 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
469
470 /* tag + hdr + front + middle */
471 con->out_kvec[v].iov_base = &tag_msg;
472 con->out_kvec[v++].iov_len = 1;
473 con->out_kvec[v].iov_base = &m->hdr;
474 con->out_kvec[v++].iov_len = sizeof(m->hdr);
475 con->out_kvec[v++] = m->front;
476 if (m->middle)
477 con->out_kvec[v++] = m->middle->vec;
478 con->out_kvec_left = v;
479 con->out_kvec_bytes += 1 + sizeof(m->hdr) + m->front.iov_len +
480 (m->middle ? m->middle->vec.iov_len : 0);
481 con->out_kvec_cur = con->out_kvec;
482
483 /* fill in crc (except data pages), footer */
484 con->out_msg->hdr.crc =
485 cpu_to_le32(crc32c(0, (void *)&m->hdr,
486 sizeof(m->hdr) - sizeof(m->hdr.crc)));
487 con->out_msg->footer.flags = CEPH_MSG_FOOTER_COMPLETE;
488 con->out_msg->footer.front_crc =
489 cpu_to_le32(crc32c(0, m->front.iov_base, m->front.iov_len));
490 if (m->middle)
491 con->out_msg->footer.middle_crc =
492 cpu_to_le32(crc32c(0, m->middle->vec.iov_base,
493 m->middle->vec.iov_len));
494 else
495 con->out_msg->footer.middle_crc = 0;
496 con->out_msg->footer.data_crc = 0;
497 dout("prepare_write_message front_crc %u data_crc %u\n",
498 le32_to_cpu(con->out_msg->footer.front_crc),
499 le32_to_cpu(con->out_msg->footer.middle_crc));
500
501 /* is there a data payload? */
502 if (le32_to_cpu(m->hdr.data_len) > 0) {
503 /* initialize page iterator */
504 con->out_msg_pos.page = 0;
505 con->out_msg_pos.page_pos =
506 le16_to_cpu(m->hdr.data_off) & ~PAGE_MASK;
507 con->out_msg_pos.data_pos = 0;
508 con->out_msg_pos.did_page_crc = 0;
509 con->out_more = 1; /* data + footer will follow */
510 } else {
511 /* no, queue up footer too and be done */
512 prepare_write_message_footer(con, v);
513 }
514
515 set_bit(WRITE_PENDING, &con->state);
516}
517
518/*
519 * Prepare an ack.
520 */
521static void prepare_write_ack(struct ceph_connection *con)
522{
523 dout("prepare_write_ack %p %llu -> %llu\n", con,
524 con->in_seq_acked, con->in_seq);
525 con->in_seq_acked = con->in_seq;
526
527 con->out_kvec[0].iov_base = &tag_ack;
528 con->out_kvec[0].iov_len = 1;
529 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
530 con->out_kvec[1].iov_base = &con->out_temp_ack;
531 con->out_kvec[1].iov_len = sizeof(con->out_temp_ack);
532 con->out_kvec_left = 2;
533 con->out_kvec_bytes = 1 + sizeof(con->out_temp_ack);
534 con->out_kvec_cur = con->out_kvec;
535 con->out_more = 1; /* more will follow.. eventually.. */
536 set_bit(WRITE_PENDING, &con->state);
537}
538
539/*
540 * Prepare to write keepalive byte.
541 */
542static void prepare_write_keepalive(struct ceph_connection *con)
543{
544 dout("prepare_write_keepalive %p\n", con);
545 con->out_kvec[0].iov_base = &tag_keepalive;
546 con->out_kvec[0].iov_len = 1;
547 con->out_kvec_left = 1;
548 con->out_kvec_bytes = 1;
549 con->out_kvec_cur = con->out_kvec;
550 set_bit(WRITE_PENDING, &con->state);
551}
552
553/*
554 * Connection negotiation.
555 */
556
557/*
558 * We connected to a peer and are saying hello.
559 */
560static void prepare_write_connect(struct ceph_messenger *msgr,
561 struct ceph_connection *con)
562{
563 int len = strlen(CEPH_BANNER);
564 unsigned global_seq = get_global_seq(con->msgr, 0);
565 int proto;
566
567 switch (con->peer_name.type) {
568 case CEPH_ENTITY_TYPE_MON:
569 proto = CEPH_MONC_PROTOCOL;
570 break;
571 case CEPH_ENTITY_TYPE_OSD:
572 proto = CEPH_OSDC_PROTOCOL;
573 break;
574 case CEPH_ENTITY_TYPE_MDS:
575 proto = CEPH_MDSC_PROTOCOL;
576 break;
577 default:
578 BUG();
579 }
580
581 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
582 con->connect_seq, global_seq, proto);
583 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
584 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
585 con->out_connect.global_seq = cpu_to_le32(global_seq);
586 con->out_connect.protocol_version = cpu_to_le32(proto);
587 con->out_connect.flags = 0;
588 if (test_bit(LOSSYTX, &con->state))
589 con->out_connect.flags = CEPH_MSG_CONNECT_LOSSY;
590
591 con->out_kvec[0].iov_base = CEPH_BANNER;
592 con->out_kvec[0].iov_len = len;
593 con->out_kvec[1].iov_base = &msgr->inst.addr;
594 con->out_kvec[1].iov_len = sizeof(msgr->inst.addr);
595 con->out_kvec[2].iov_base = &con->out_connect;
596 con->out_kvec[2].iov_len = sizeof(con->out_connect);
597 con->out_kvec_left = 3;
598 con->out_kvec_bytes = len + sizeof(msgr->inst.addr) +
599 sizeof(con->out_connect);
600 con->out_kvec_cur = con->out_kvec;
601 con->out_more = 0;
602 set_bit(WRITE_PENDING, &con->state);
603}
604
605static void prepare_write_connect_retry(struct ceph_messenger *msgr,
606 struct ceph_connection *con)
607{
608 dout("prepare_write_connect_retry %p\n", con);
609 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
610 con->out_connect.global_seq =
611 cpu_to_le32(get_global_seq(con->msgr, 0));
612
613 con->out_kvec[0].iov_base = &con->out_connect;
614 con->out_kvec[0].iov_len = sizeof(con->out_connect);
615 con->out_kvec_left = 1;
616 con->out_kvec_bytes = sizeof(con->out_connect);
617 con->out_kvec_cur = con->out_kvec;
618 con->out_more = 0;
619 set_bit(WRITE_PENDING, &con->state);
620}
621
622
623/*
624 * write as much of pending kvecs to the socket as we can.
625 * 1 -> done
626 * 0 -> socket full, but more to do
627 * <0 -> error
628 */
629static int write_partial_kvec(struct ceph_connection *con)
630{
631 int ret;
632
633 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
634 while (con->out_kvec_bytes > 0) {
635 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
636 con->out_kvec_left, con->out_kvec_bytes,
637 con->out_more);
638 if (ret <= 0)
639 goto out;
640 con->out_kvec_bytes -= ret;
641 if (con->out_kvec_bytes == 0)
642 break; /* done */
643 while (ret > 0) {
644 if (ret >= con->out_kvec_cur->iov_len) {
645 ret -= con->out_kvec_cur->iov_len;
646 con->out_kvec_cur++;
647 con->out_kvec_left--;
648 } else {
649 con->out_kvec_cur->iov_len -= ret;
650 con->out_kvec_cur->iov_base += ret;
651 ret = 0;
652 break;
653 }
654 }
655 }
656 con->out_kvec_left = 0;
657 con->out_kvec_is_msg = false;
658 ret = 1;
659out:
660 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
661 con->out_kvec_bytes, con->out_kvec_left, ret);
662 return ret; /* done! */
663}
664
665/*
666 * Write as much message data payload as we can. If we finish, queue
667 * up the footer.
668 * 1 -> done, footer is now queued in out_kvec[].
669 * 0 -> socket full, but more to do
670 * <0 -> error
671 */
672static int write_partial_msg_pages(struct ceph_connection *con)
673{
674 struct ceph_msg *msg = con->out_msg;
675 unsigned data_len = le32_to_cpu(msg->hdr.data_len);
676 size_t len;
677 int crc = con->msgr->nocrc;
678 int ret;
679
680 dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
681 con, con->out_msg, con->out_msg_pos.page, con->out_msg->nr_pages,
682 con->out_msg_pos.page_pos);
683
684 while (con->out_msg_pos.page < con->out_msg->nr_pages) {
685 struct page *page = NULL;
686 void *kaddr = NULL;
687
688 /*
689 * if we are calculating the data crc (the default), we need
690 * to map the page. if our pages[] has been revoked, use the
691 * zero page.
692 */
693 if (msg->pages) {
694 page = msg->pages[con->out_msg_pos.page];
695 if (crc)
696 kaddr = kmap(page);
697 } else {
698 page = con->msgr->zero_page;
699 if (crc)
700 kaddr = page_address(con->msgr->zero_page);
701 }
702 len = min((int)(PAGE_SIZE - con->out_msg_pos.page_pos),
703 (int)(data_len - con->out_msg_pos.data_pos));
704 if (crc && !con->out_msg_pos.did_page_crc) {
705 void *base = kaddr + con->out_msg_pos.page_pos;
706 u32 tmpcrc = le32_to_cpu(con->out_msg->footer.data_crc);
707
708 BUG_ON(kaddr == NULL);
709 con->out_msg->footer.data_crc =
710 cpu_to_le32(crc32c(tmpcrc, base, len));
711 con->out_msg_pos.did_page_crc = 1;
712 }
713
714 ret = kernel_sendpage(con->sock, page,
715 con->out_msg_pos.page_pos, len,
716 MSG_DONTWAIT | MSG_NOSIGNAL |
717 MSG_MORE);
718
719 if (crc && msg->pages)
720 kunmap(page);
721
722 if (ret <= 0)
723 goto out;
724
725 con->out_msg_pos.data_pos += ret;
726 con->out_msg_pos.page_pos += ret;
727 if (ret == len) {
728 con->out_msg_pos.page_pos = 0;
729 con->out_msg_pos.page++;
730 con->out_msg_pos.did_page_crc = 0;
731 }
732 }
733
734 dout("write_partial_msg_pages %p msg %p done\n", con, msg);
735
736 /* prepare and queue up footer, too */
737 if (!crc)
738 con->out_msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
739 con->out_kvec_bytes = 0;
740 con->out_kvec_left = 0;
741 con->out_kvec_cur = con->out_kvec;
742 prepare_write_message_footer(con, 0);
743 ret = 1;
744out:
745 return ret;
746}
747
748/*
749 * write some zeros
750 */
751static int write_partial_skip(struct ceph_connection *con)
752{
753 int ret;
754
755 while (con->out_skip > 0) {
756 struct kvec iov = {
757 .iov_base = page_address(con->msgr->zero_page),
758 .iov_len = min(con->out_skip, (int)PAGE_CACHE_SIZE)
759 };
760
761 ret = ceph_tcp_sendmsg(con->sock, &iov, 1, iov.iov_len, 1);
762 if (ret <= 0)
763 goto out;
764 con->out_skip -= ret;
765 }
766 ret = 1;
767out:
768 return ret;
769}
770
771/*
772 * Prepare to read connection handshake, or an ack.
773 */
774static void prepare_read_connect(struct ceph_connection *con)
775{
776 dout("prepare_read_connect %p\n", con);
777 con->in_base_pos = 0;
778}
779
780static void prepare_read_ack(struct ceph_connection *con)
781{
782 dout("prepare_read_ack %p\n", con);
783 con->in_base_pos = 0;
784}
785
786static void prepare_read_tag(struct ceph_connection *con)
787{
788 dout("prepare_read_tag %p\n", con);
789 con->in_base_pos = 0;
790 con->in_tag = CEPH_MSGR_TAG_READY;
791}
792
793/*
794 * Prepare to read a message.
795 */
796static int prepare_read_message(struct ceph_connection *con)
797{
798 dout("prepare_read_message %p\n", con);
799 BUG_ON(con->in_msg != NULL);
800 con->in_base_pos = 0;
801 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
802 return 0;
803}
804
805
806static int read_partial(struct ceph_connection *con,
807 int *to, int size, void *object)
808{
809 *to += size;
810 while (con->in_base_pos < *to) {
811 int left = *to - con->in_base_pos;
812 int have = size - left;
813 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
814 if (ret <= 0)
815 return ret;
816 con->in_base_pos += ret;
817 }
818 return 1;
819}
820
821
822/*
823 * Read all or part of the connect-side handshake on a new connection
824 */
825static int read_partial_connect(struct ceph_connection *con)
826{
827 int ret, to = 0;
828
829 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
830
831 /* peer's banner */
832 ret = read_partial(con, &to, strlen(CEPH_BANNER), con->in_banner);
833 if (ret <= 0)
834 goto out;
835 ret = read_partial(con, &to, sizeof(con->actual_peer_addr),
836 &con->actual_peer_addr);
837 if (ret <= 0)
838 goto out;
839 ret = read_partial(con, &to, sizeof(con->peer_addr_for_me),
840 &con->peer_addr_for_me);
841 if (ret <= 0)
842 goto out;
843 ret = read_partial(con, &to, sizeof(con->in_reply), &con->in_reply);
844 if (ret <= 0)
845 goto out;
846
847 dout("read_partial_connect %p connect_seq = %u, global_seq = %u\n",
848 con, le32_to_cpu(con->in_reply.connect_seq),
849 le32_to_cpu(con->in_reply.global_seq));
850out:
851 return ret;
852}
853
854/*
855 * Verify the hello banner looks okay.
856 */
857static int verify_hello(struct ceph_connection *con)
858{
859 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
860 pr_err("connect to/from %s has bad banner\n",
861 pr_addr(&con->peer_addr.in_addr));
862 con->error_msg = "protocol error, bad banner";
863 return -1;
864 }
865 return 0;
866}
867
868static bool addr_is_blank(struct sockaddr_storage *ss)
869{
870 switch (ss->ss_family) {
871 case AF_INET:
872 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
873 case AF_INET6:
874 return
875 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
876 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
877 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
878 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
879 }
880 return false;
881}
882
883static int addr_port(struct sockaddr_storage *ss)
884{
885 switch (ss->ss_family) {
886 case AF_INET:
887 return ((struct sockaddr_in *)ss)->sin_port;
888 case AF_INET6:
889 return ((struct sockaddr_in6 *)ss)->sin6_port;
890 }
891 return 0;
892}
893
894static void addr_set_port(struct sockaddr_storage *ss, int p)
895{
896 switch (ss->ss_family) {
897 case AF_INET:
898 ((struct sockaddr_in *)ss)->sin_port = htons(p);
899 case AF_INET6:
900 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
901 }
902}
903
904/*
905 * Parse an ip[:port] list into an addr array. Use the default
906 * monitor port if a port isn't specified.
907 */
908int ceph_parse_ips(const char *c, const char *end,
909 struct ceph_entity_addr *addr,
910 int max_count, int *count)
911{
912 int i;
913 const char *p = c;
914
915 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
916 for (i = 0; i < max_count; i++) {
917 const char *ipend;
918 struct sockaddr_storage *ss = &addr[i].in_addr;
919 struct sockaddr_in *in4 = (void *)ss;
920 struct sockaddr_in6 *in6 = (void *)ss;
921 int port;
922
923 memset(ss, 0, sizeof(*ss));
924 if (in4_pton(p, end - p, (u8 *)&in4->sin_addr.s_addr,
925 ',', &ipend)) {
926 ss->ss_family = AF_INET;
927 } else if (in6_pton(p, end - p, (u8 *)&in6->sin6_addr.s6_addr,
928 ',', &ipend)) {
929 ss->ss_family = AF_INET6;
930 } else {
931 goto bad;
932 }
933 p = ipend;
934
935 /* port? */
936 if (p < end && *p == ':') {
937 port = 0;
938 p++;
939 while (p < end && *p >= '0' && *p <= '9') {
940 port = (port * 10) + (*p - '0');
941 p++;
942 }
943 if (port > 65535 || port == 0)
944 goto bad;
945 } else {
946 port = CEPH_MON_PORT;
947 }
948
949 addr_set_port(ss, port);
950
951 dout("parse_ips got %s\n", pr_addr(ss));
952
953 if (p == end)
954 break;
955 if (*p != ',')
956 goto bad;
957 p++;
958 }
959
960 if (p != end)
961 goto bad;
962
963 if (count)
964 *count = i + 1;
965 return 0;
966
967bad:
968 pr_err("parse_ips bad ip '%s'\n", c);
969 return -EINVAL;
970}
971
972static int process_connect(struct ceph_connection *con)
973{
974 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
975
976 if (verify_hello(con) < 0)
977 return -1;
978
979 /*
980 * Make sure the other end is who we wanted. note that the other
981 * end may not yet know their ip address, so if it's 0.0.0.0, give
982 * them the benefit of the doubt.
983 */
984 if (!ceph_entity_addr_is_local(&con->peer_addr,
985 &con->actual_peer_addr) &&
986 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
987 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
988 pr_err("wrong peer, want %s/%d, "
989 "got %s/%d, wtf\n",
990 pr_addr(&con->peer_addr.in_addr),
991 con->peer_addr.nonce,
992 pr_addr(&con->actual_peer_addr.in_addr),
993 con->actual_peer_addr.nonce);
994 con->error_msg = "protocol error, wrong peer";
995 return -1;
996 }
997
998 /*
999 * did we learn our address?
1000 */
1001 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1002 int port = addr_port(&con->msgr->inst.addr.in_addr);
1003
1004 memcpy(&con->msgr->inst.addr.in_addr,
1005 &con->peer_addr_for_me.in_addr,
1006 sizeof(con->peer_addr_for_me.in_addr));
1007 addr_set_port(&con->msgr->inst.addr.in_addr, port);
1008 dout("process_connect learned my addr is %s\n",
1009 pr_addr(&con->msgr->inst.addr.in_addr));
1010 }
1011
1012 switch (con->in_reply.tag) {
1013 case CEPH_MSGR_TAG_BADPROTOVER:
1014 dout("process_connect got BADPROTOVER my %d != their %d\n",
1015 le32_to_cpu(con->out_connect.protocol_version),
1016 le32_to_cpu(con->in_reply.protocol_version));
1017 pr_err("%s%lld %s protocol version mismatch,"
1018 " my %d != server's %d\n",
1019 ENTITY_NAME(con->peer_name),
1020 pr_addr(&con->peer_addr.in_addr),
1021 le32_to_cpu(con->out_connect.protocol_version),
1022 le32_to_cpu(con->in_reply.protocol_version));
1023 con->error_msg = "protocol version mismatch";
1024 if (con->ops->bad_proto)
1025 con->ops->bad_proto(con);
1026 reset_connection(con);
1027 set_bit(CLOSED, &con->state); /* in case there's queued work */
1028 return -1;
1029
1030
1031 case CEPH_MSGR_TAG_RESETSESSION:
1032 /*
1033 * If we connected with a large connect_seq but the peer
1034 * has no record of a session with us (no connection, or
1035 * connect_seq == 0), they will send RESETSESION to indicate
1036 * that they must have reset their session, and may have
1037 * dropped messages.
1038 */
1039 dout("process_connect got RESET peer seq %u\n",
1040 le32_to_cpu(con->in_connect.connect_seq));
1041 pr_err("%s%lld %s connection reset\n",
1042 ENTITY_NAME(con->peer_name),
1043 pr_addr(&con->peer_addr.in_addr));
1044 reset_connection(con);
1045 prepare_write_connect_retry(con->msgr, con);
1046 prepare_read_connect(con);
1047
1048 /* Tell ceph about it. */
1049 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
1050 if (con->ops->peer_reset)
1051 con->ops->peer_reset(con);
1052 break;
1053
1054 case CEPH_MSGR_TAG_RETRY_SESSION:
1055 /*
1056 * If we sent a smaller connect_seq than the peer has, try
1057 * again with a larger value.
1058 */
1059 dout("process_connect got RETRY my seq = %u, peer_seq = %u\n",
1060 le32_to_cpu(con->out_connect.connect_seq),
1061 le32_to_cpu(con->in_connect.connect_seq));
1062 con->connect_seq = le32_to_cpu(con->in_connect.connect_seq);
1063 prepare_write_connect_retry(con->msgr, con);
1064 prepare_read_connect(con);
1065 break;
1066
1067 case CEPH_MSGR_TAG_RETRY_GLOBAL:
1068 /*
1069 * If we sent a smaller global_seq than the peer has, try
1070 * again with a larger value.
1071 */
1072 dout("process_connect got RETRY_GLOBAL my %u, peer_gseq = %u\n",
1073 con->peer_global_seq,
1074 le32_to_cpu(con->in_connect.global_seq));
1075 get_global_seq(con->msgr,
1076 le32_to_cpu(con->in_connect.global_seq));
1077 prepare_write_connect_retry(con->msgr, con);
1078 prepare_read_connect(con);
1079 break;
1080
1081 case CEPH_MSGR_TAG_READY:
1082 clear_bit(CONNECTING, &con->state);
1083 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
1084 set_bit(LOSSYRX, &con->state);
1085 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
1086 con->connect_seq++;
1087 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1088 con->peer_global_seq,
1089 le32_to_cpu(con->in_reply.connect_seq),
1090 con->connect_seq);
1091 WARN_ON(con->connect_seq !=
1092 le32_to_cpu(con->in_reply.connect_seq));
1093
1094 con->delay = 0; /* reset backoff memory */
1095 prepare_read_tag(con);
1096 break;
1097
1098 case CEPH_MSGR_TAG_WAIT:
1099 /*
1100 * If there is a connection race (we are opening
1101 * connections to each other), one of us may just have
1102 * to WAIT. This shouldn't happen if we are the
1103 * client.
1104 */
1105 pr_err("process_connect peer connecting WAIT\n");
1106
1107 default:
1108 pr_err("connect protocol error, will retry\n");
1109 con->error_msg = "protocol error, garbage tag during connect";
1110 return -1;
1111 }
1112 return 0;
1113}
1114
1115
1116/*
1117 * read (part of) an ack
1118 */
1119static int read_partial_ack(struct ceph_connection *con)
1120{
1121 int to = 0;
1122
1123 return read_partial(con, &to, sizeof(con->in_temp_ack),
1124 &con->in_temp_ack);
1125}
1126
1127
1128/*
1129 * We can finally discard anything that's been acked.
1130 */
1131static void process_ack(struct ceph_connection *con)
1132{
1133 struct ceph_msg *m;
1134 u64 ack = le64_to_cpu(con->in_temp_ack);
1135 u64 seq;
1136
1137 mutex_lock(&con->out_mutex);
1138 while (!list_empty(&con->out_sent)) {
1139 m = list_first_entry(&con->out_sent, struct ceph_msg,
1140 list_head);
1141 seq = le64_to_cpu(m->hdr.seq);
1142 if (seq > ack)
1143 break;
1144 dout("got ack for seq %llu type %d at %p\n", seq,
1145 le16_to_cpu(m->hdr.type), m);
1146 ceph_msg_remove(m);
1147 }
1148 mutex_unlock(&con->out_mutex);
1149 prepare_read_tag(con);
1150}
1151
1152
1153
1154
1155
1156
1157/*
1158 * read (part of) a message.
1159 */
1160static int read_partial_message(struct ceph_connection *con)
1161{
1162 struct ceph_msg *m = con->in_msg;
1163 void *p;
1164 int ret;
1165 int to, want, left;
1166 unsigned front_len, middle_len, data_len, data_off;
1167 int datacrc = con->msgr->nocrc;
1168
1169 dout("read_partial_message con %p msg %p\n", con, m);
1170
1171 /* header */
1172 while (con->in_base_pos < sizeof(con->in_hdr)) {
1173 left = sizeof(con->in_hdr) - con->in_base_pos;
1174 ret = ceph_tcp_recvmsg(con->sock,
1175 (char *)&con->in_hdr + con->in_base_pos,
1176 left);
1177 if (ret <= 0)
1178 return ret;
1179 con->in_base_pos += ret;
1180 if (con->in_base_pos == sizeof(con->in_hdr)) {
1181 u32 crc = crc32c(0, (void *)&con->in_hdr,
1182 sizeof(con->in_hdr) - sizeof(con->in_hdr.crc));
1183 if (crc != le32_to_cpu(con->in_hdr.crc)) {
1184 pr_err("read_partial_message bad hdr "
1185 " crc %u != expected %u\n",
1186 crc, con->in_hdr.crc);
1187 return -EBADMSG;
1188 }
1189 }
1190 }
1191
1192 front_len = le32_to_cpu(con->in_hdr.front_len);
1193 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
1194 return -EIO;
1195 middle_len = le32_to_cpu(con->in_hdr.middle_len);
1196 if (middle_len > CEPH_MSG_MAX_DATA_LEN)
1197 return -EIO;
1198 data_len = le32_to_cpu(con->in_hdr.data_len);
1199 if (data_len > CEPH_MSG_MAX_DATA_LEN)
1200 return -EIO;
1201
1202 /* allocate message? */
1203 if (!con->in_msg) {
1204 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
1205 con->in_hdr.front_len, con->in_hdr.data_len);
1206 con->in_msg = con->ops->alloc_msg(con, &con->in_hdr);
1207 if (!con->in_msg) {
1208 /* skip this message */
1209 dout("alloc_msg returned NULL, skipping message\n");
1210 con->in_base_pos = -front_len - middle_len - data_len -
1211 sizeof(m->footer);
1212 con->in_tag = CEPH_MSGR_TAG_READY;
1213 return 0;
1214 }
1215 if (IS_ERR(con->in_msg)) {
1216 ret = PTR_ERR(con->in_msg);
1217 con->in_msg = NULL;
1218 con->error_msg = "out of memory for incoming message";
1219 return ret;
1220 }
1221 m = con->in_msg;
1222 m->front.iov_len = 0; /* haven't read it yet */
1223 memcpy(&m->hdr, &con->in_hdr, sizeof(con->in_hdr));
1224 }
1225
1226 /* front */
1227 while (m->front.iov_len < front_len) {
1228 BUG_ON(m->front.iov_base == NULL);
1229 left = front_len - m->front.iov_len;
1230 ret = ceph_tcp_recvmsg(con->sock, (char *)m->front.iov_base +
1231 m->front.iov_len, left);
1232 if (ret <= 0)
1233 return ret;
1234 m->front.iov_len += ret;
1235 if (m->front.iov_len == front_len)
1236 con->in_front_crc = crc32c(0, m->front.iov_base,
1237 m->front.iov_len);
1238 }
1239
1240 /* middle */
1241 while (middle_len > 0 && (!m->middle ||
1242 m->middle->vec.iov_len < middle_len)) {
1243 if (m->middle == NULL) {
1244 ret = -EOPNOTSUPP;
1245 if (con->ops->alloc_middle)
1246 ret = con->ops->alloc_middle(con, m);
1247 if (ret < 0) {
1248 dout("alloc_middle failed, skipping payload\n");
1249 con->in_base_pos = -middle_len - data_len
1250 - sizeof(m->footer);
1251 ceph_msg_put(con->in_msg);
1252 con->in_msg = NULL;
1253 con->in_tag = CEPH_MSGR_TAG_READY;
1254 return 0;
1255 }
1256 m->middle->vec.iov_len = 0;
1257 }
1258 left = middle_len - m->middle->vec.iov_len;
1259 ret = ceph_tcp_recvmsg(con->sock,
1260 (char *)m->middle->vec.iov_base +
1261 m->middle->vec.iov_len, left);
1262 if (ret <= 0)
1263 return ret;
1264 m->middle->vec.iov_len += ret;
1265 if (m->middle->vec.iov_len == middle_len)
1266 con->in_middle_crc = crc32c(0, m->middle->vec.iov_base,
1267 m->middle->vec.iov_len);
1268 }
1269
1270 /* (page) data */
1271 data_off = le16_to_cpu(m->hdr.data_off);
1272 if (data_len == 0)
1273 goto no_data;
1274
1275 if (m->nr_pages == 0) {
1276 con->in_msg_pos.page = 0;
1277 con->in_msg_pos.page_pos = data_off & ~PAGE_MASK;
1278 con->in_msg_pos.data_pos = 0;
1279 /* find pages for data payload */
1280 want = calc_pages_for(data_off & ~PAGE_MASK, data_len);
1281 ret = -1;
1282 if (con->ops->prepare_pages)
1283 ret = con->ops->prepare_pages(con, m, want);
1284 if (ret < 0) {
1285 dout("%p prepare_pages failed, skipping payload\n", m);
1286 con->in_base_pos = -data_len - sizeof(m->footer);
1287 ceph_msg_put(con->in_msg);
1288 con->in_msg = NULL;
1289 con->in_tag = CEPH_MSGR_TAG_READY;
1290 return 0;
1291 }
1292 BUG_ON(m->nr_pages < want);
1293 }
1294 while (con->in_msg_pos.data_pos < data_len) {
1295 left = min((int)(data_len - con->in_msg_pos.data_pos),
1296 (int)(PAGE_SIZE - con->in_msg_pos.page_pos));
1297 BUG_ON(m->pages == NULL);
1298 p = kmap(m->pages[con->in_msg_pos.page]);
1299 ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos,
1300 left);
1301 if (ret > 0 && datacrc)
1302 con->in_data_crc =
1303 crc32c(con->in_data_crc,
1304 p + con->in_msg_pos.page_pos, ret);
1305 kunmap(m->pages[con->in_msg_pos.page]);
1306 if (ret <= 0)
1307 return ret;
1308 con->in_msg_pos.data_pos += ret;
1309 con->in_msg_pos.page_pos += ret;
1310 if (con->in_msg_pos.page_pos == PAGE_SIZE) {
1311 con->in_msg_pos.page_pos = 0;
1312 con->in_msg_pos.page++;
1313 }
1314 }
1315
1316no_data:
1317 /* footer */
1318 to = sizeof(m->hdr) + sizeof(m->footer);
1319 while (con->in_base_pos < to) {
1320 left = to - con->in_base_pos;
1321 ret = ceph_tcp_recvmsg(con->sock, (char *)&m->footer +
1322 (con->in_base_pos - sizeof(m->hdr)),
1323 left);
1324 if (ret <= 0)
1325 return ret;
1326 con->in_base_pos += ret;
1327 }
1328 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1329 m, front_len, m->footer.front_crc, middle_len,
1330 m->footer.middle_crc, data_len, m->footer.data_crc);
1331
1332 /* crc ok? */
1333 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
1334 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1335 m, con->in_front_crc, m->footer.front_crc);
1336 return -EBADMSG;
1337 }
1338 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
1339 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1340 m, con->in_middle_crc, m->footer.middle_crc);
1341 return -EBADMSG;
1342 }
1343 if (datacrc &&
1344 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
1345 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
1346 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
1347 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
1348 return -EBADMSG;
1349 }
1350
1351 return 1; /* done! */
1352}
1353
1354/*
1355 * Process message. This happens in the worker thread. The callback should
1356 * be careful not to do anything that waits on other incoming messages or it
1357 * may deadlock.
1358 */
1359static void process_message(struct ceph_connection *con)
1360{
1361 struct ceph_msg *msg = con->in_msg;
1362
1363 con->in_msg = NULL;
1364
1365 /* if first message, set peer_name */
1366 if (con->peer_name.type == 0)
1367 con->peer_name = msg->hdr.src.name;
1368
1369 mutex_lock(&con->out_mutex);
1370 con->in_seq++;
1371 mutex_unlock(&con->out_mutex);
1372
1373 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
1374 msg, le64_to_cpu(msg->hdr.seq),
1375 ENTITY_NAME(msg->hdr.src.name),
1376 le16_to_cpu(msg->hdr.type),
1377 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
1378 le32_to_cpu(msg->hdr.front_len),
1379 le32_to_cpu(msg->hdr.data_len),
1380 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
1381 con->ops->dispatch(con, msg);
1382 prepare_read_tag(con);
1383}
1384
1385
1386/*
1387 * Write something to the socket. Called in a worker thread when the
1388 * socket appears to be writeable and we have something ready to send.
1389 */
1390static int try_write(struct ceph_connection *con)
1391{
1392 struct ceph_messenger *msgr = con->msgr;
1393 int ret = 1;
1394
1395 dout("try_write start %p state %lu nref %d\n", con, con->state,
1396 atomic_read(&con->nref));
1397
1398 mutex_lock(&con->out_mutex);
1399more:
1400 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
1401
1402 /* open the socket first? */
1403 if (con->sock == NULL) {
1404 /*
1405 * if we were STANDBY and are reconnecting _this_
1406 * connection, bump connect_seq now. Always bump
1407 * global_seq.
1408 */
1409 if (test_and_clear_bit(STANDBY, &con->state))
1410 con->connect_seq++;
1411
1412 prepare_write_connect(msgr, con);
1413 prepare_read_connect(con);
1414 set_bit(CONNECTING, &con->state);
1415
1416 con->in_tag = CEPH_MSGR_TAG_READY;
1417 dout("try_write initiating connect on %p new state %lu\n",
1418 con, con->state);
1419 con->sock = ceph_tcp_connect(con);
1420 if (IS_ERR(con->sock)) {
1421 con->sock = NULL;
1422 con->error_msg = "connect error";
1423 ret = -1;
1424 goto out;
1425 }
1426 }
1427
1428more_kvec:
1429 /* kvec data queued? */
1430 if (con->out_skip) {
1431 ret = write_partial_skip(con);
1432 if (ret <= 0)
1433 goto done;
1434 if (ret < 0) {
1435 dout("try_write write_partial_skip err %d\n", ret);
1436 goto done;
1437 }
1438 }
1439 if (con->out_kvec_left) {
1440 ret = write_partial_kvec(con);
1441 if (ret <= 0)
1442 goto done;
1443 if (ret < 0) {
1444 dout("try_write write_partial_kvec err %d\n", ret);
1445 goto done;
1446 }
1447 }
1448
1449 /* msg pages? */
1450 if (con->out_msg) {
1451 ret = write_partial_msg_pages(con);
1452 if (ret == 1)
1453 goto more_kvec; /* we need to send the footer, too! */
1454 if (ret == 0)
1455 goto done;
1456 if (ret < 0) {
1457 dout("try_write write_partial_msg_pages err %d\n",
1458 ret);
1459 goto done;
1460 }
1461 }
1462
1463 if (!test_bit(CONNECTING, &con->state)) {
1464 /* is anything else pending? */
1465 if (!list_empty(&con->out_queue)) {
1466 prepare_write_message(con);
1467 goto more;
1468 }
1469 if (con->in_seq > con->in_seq_acked) {
1470 prepare_write_ack(con);
1471 goto more;
1472 }
1473 if (test_and_clear_bit(KEEPALIVE_PENDING, &con->state)) {
1474 prepare_write_keepalive(con);
1475 goto more;
1476 }
1477 }
1478
1479 /* Nothing to do! */
1480 clear_bit(WRITE_PENDING, &con->state);
1481 dout("try_write nothing else to write.\n");
1482done:
1483 ret = 0;
1484out:
1485 mutex_unlock(&con->out_mutex);
1486 dout("try_write done on %p\n", con);
1487 return ret;
1488}
1489
1490
1491
1492/*
1493 * Read what we can from the socket.
1494 */
1495static int try_read(struct ceph_connection *con)
1496{
1497 struct ceph_messenger *msgr;
1498 int ret = -1;
1499
1500 if (!con->sock)
1501 return 0;
1502
1503 if (test_bit(STANDBY, &con->state))
1504 return 0;
1505
1506 dout("try_read start on %p\n", con);
1507 msgr = con->msgr;
1508
1509more:
1510 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
1511 con->in_base_pos);
1512 if (test_bit(CONNECTING, &con->state)) {
1513 dout("try_read connecting\n");
1514 ret = read_partial_connect(con);
1515 if (ret <= 0)
1516 goto done;
1517 if (process_connect(con) < 0) {
1518 ret = -1;
1519 goto out;
1520 }
1521 goto more;
1522 }
1523
1524 if (con->in_base_pos < 0) {
1525 /*
1526 * skipping + discarding content.
1527 *
1528 * FIXME: there must be a better way to do this!
1529 */
1530 static char buf[1024];
1531 int skip = min(1024, -con->in_base_pos);
1532 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
1533 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
1534 if (ret <= 0)
1535 goto done;
1536 con->in_base_pos += ret;
1537 if (con->in_base_pos)
1538 goto more;
1539 }
1540 if (con->in_tag == CEPH_MSGR_TAG_READY) {
1541 /*
1542 * what's next?
1543 */
1544 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
1545 if (ret <= 0)
1546 goto done;
1547 dout("try_read got tag %d\n", (int)con->in_tag);
1548 switch (con->in_tag) {
1549 case CEPH_MSGR_TAG_MSG:
1550 prepare_read_message(con);
1551 break;
1552 case CEPH_MSGR_TAG_ACK:
1553 prepare_read_ack(con);
1554 break;
1555 case CEPH_MSGR_TAG_CLOSE:
1556 set_bit(CLOSED, &con->state); /* fixme */
1557 goto done;
1558 default:
1559 goto bad_tag;
1560 }
1561 }
1562 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
1563 ret = read_partial_message(con);
1564 if (ret <= 0) {
1565 switch (ret) {
1566 case -EBADMSG:
1567 con->error_msg = "bad crc";
1568 ret = -EIO;
1569 goto out;
1570 case -EIO:
1571 con->error_msg = "io error";
1572 goto out;
1573 default:
1574 goto done;
1575 }
1576 }
1577 if (con->in_tag == CEPH_MSGR_TAG_READY)
1578 goto more;
1579 process_message(con);
1580 goto more;
1581 }
1582 if (con->in_tag == CEPH_MSGR_TAG_ACK) {
1583 ret = read_partial_ack(con);
1584 if (ret <= 0)
1585 goto done;
1586 process_ack(con);
1587 goto more;
1588 }
1589
1590done:
1591 ret = 0;
1592out:
1593 dout("try_read done on %p\n", con);
1594 return ret;
1595
1596bad_tag:
1597 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
1598 con->error_msg = "protocol error, garbage tag";
1599 ret = -1;
1600 goto out;
1601}
1602
1603
1604/*
1605 * Atomically queue work on a connection. Bump @con reference to
1606 * avoid races with connection teardown.
1607 *
1608 * There is some trickery going on with QUEUED and BUSY because we
1609 * only want a _single_ thread operating on each connection at any
1610 * point in time, but we want to use all available CPUs.
1611 *
1612 * The worker thread only proceeds if it can atomically set BUSY. It
1613 * clears QUEUED and does it's thing. When it thinks it's done, it
1614 * clears BUSY, then rechecks QUEUED.. if it's set again, it loops
1615 * (tries again to set BUSY).
1616 *
1617 * To queue work, we first set QUEUED, _then_ if BUSY isn't set, we
1618 * try to queue work. If that fails (work is already queued, or BUSY)
1619 * we give up (work also already being done or is queued) but leave QUEUED
1620 * set so that the worker thread will loop if necessary.
1621 */
1622static void queue_con(struct ceph_connection *con)
1623{
1624 if (test_bit(DEAD, &con->state)) {
1625 dout("queue_con %p ignoring: DEAD\n",
1626 con);
1627 return;
1628 }
1629
1630 if (!con->ops->get(con)) {
1631 dout("queue_con %p ref count 0\n", con);
1632 return;
1633 }
1634
1635 set_bit(QUEUED, &con->state);
1636 if (test_bit(BUSY, &con->state)) {
1637 dout("queue_con %p - already BUSY\n", con);
1638 con->ops->put(con);
1639 } else if (!queue_work(ceph_msgr_wq, &con->work.work)) {
1640 dout("queue_con %p - already queued\n", con);
1641 con->ops->put(con);
1642 } else {
1643 dout("queue_con %p\n", con);
1644 }
1645}
1646
1647/*
1648 * Do some work on a connection. Drop a connection ref when we're done.
1649 */
1650static void con_work(struct work_struct *work)
1651{
1652 struct ceph_connection *con = container_of(work, struct ceph_connection,
1653 work.work);
1654 int backoff = 0;
1655
1656more:
1657 if (test_and_set_bit(BUSY, &con->state) != 0) {
1658 dout("con_work %p BUSY already set\n", con);
1659 goto out;
1660 }
1661 dout("con_work %p start, clearing QUEUED\n", con);
1662 clear_bit(QUEUED, &con->state);
1663
1664 if (test_bit(CLOSED, &con->state)) { /* e.g. if we are replaced */
1665 dout("con_work CLOSED\n");
1666 con_close_socket(con);
1667 goto done;
1668 }
1669 if (test_and_clear_bit(OPENING, &con->state)) {
1670 /* reopen w/ new peer */
1671 dout("con_work OPENING\n");
1672 con_close_socket(con);
1673 }
1674
1675 if (test_and_clear_bit(SOCK_CLOSED, &con->state) ||
1676 try_read(con) < 0 ||
1677 try_write(con) < 0) {
1678 backoff = 1;
1679 ceph_fault(con); /* error/fault path */
1680 }
1681
1682done:
1683 clear_bit(BUSY, &con->state);
1684 dout("con->state=%lu\n", con->state);
1685 if (test_bit(QUEUED, &con->state)) {
1686 if (!backoff) {
1687 dout("con_work %p QUEUED reset, looping\n", con);
1688 goto more;
1689 }
1690 dout("con_work %p QUEUED reset, but just faulted\n", con);
1691 clear_bit(QUEUED, &con->state);
1692 }
1693 dout("con_work %p done\n", con);
1694
1695out:
1696 con->ops->put(con);
1697}
1698
1699
1700/*
1701 * Generic error/fault handler. A retry mechanism is used with
1702 * exponential backoff
1703 */
1704static void ceph_fault(struct ceph_connection *con)
1705{
1706 pr_err("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
1707 pr_addr(&con->peer_addr.in_addr), con->error_msg);
1708 dout("fault %p state %lu to peer %s\n",
1709 con, con->state, pr_addr(&con->peer_addr.in_addr));
1710
1711 if (test_bit(LOSSYTX, &con->state)) {
1712 dout("fault on LOSSYTX channel\n");
1713 goto out;
1714 }
1715
1716 clear_bit(BUSY, &con->state); /* to avoid an improbable race */
1717
1718 con_close_socket(con);
1719 con->in_msg = NULL;
1720
1721 /* If there are no messages in the queue, place the connection
1722 * in a STANDBY state (i.e., don't try to reconnect just yet). */
1723 mutex_lock(&con->out_mutex);
1724 if (list_empty(&con->out_queue) && !con->out_keepalive_pending) {
1725 dout("fault setting STANDBY\n");
1726 set_bit(STANDBY, &con->state);
1727 mutex_unlock(&con->out_mutex);
1728 goto out;
1729 }
1730
1731 /* Requeue anything that hasn't been acked, and retry after a
1732 * delay. */
1733 list_splice_init(&con->out_sent, &con->out_queue);
1734 mutex_unlock(&con->out_mutex);
1735
1736 if (con->delay == 0)
1737 con->delay = BASE_DELAY_INTERVAL;
1738 else if (con->delay < MAX_DELAY_INTERVAL)
1739 con->delay *= 2;
1740
1741 /* explicitly schedule work to try to reconnect again later. */
1742 dout("fault queueing %p delay %lu\n", con, con->delay);
1743 con->ops->get(con);
1744 if (queue_delayed_work(ceph_msgr_wq, &con->work,
1745 round_jiffies_relative(con->delay)) == 0)
1746 con->ops->put(con);
1747
1748out:
1749 if (con->ops->fault)
1750 con->ops->fault(con);
1751}
1752
1753
1754
1755/*
1756 * create a new messenger instance
1757 */
1758struct ceph_messenger *ceph_messenger_create(struct ceph_entity_addr *myaddr)
1759{
1760 struct ceph_messenger *msgr;
1761
1762 msgr = kzalloc(sizeof(*msgr), GFP_KERNEL);
1763 if (msgr == NULL)
1764 return ERR_PTR(-ENOMEM);
1765
1766 spin_lock_init(&msgr->global_seq_lock);
1767
1768 /* the zero page is needed if a request is "canceled" while the message
1769 * is being written over the socket */
1770 msgr->zero_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1771 if (!msgr->zero_page) {
1772 kfree(msgr);
1773 return ERR_PTR(-ENOMEM);
1774 }
1775 kmap(msgr->zero_page);
1776
1777 if (myaddr)
1778 msgr->inst.addr = *myaddr;
1779
1780 /* select a random nonce */
1781 get_random_bytes(&msgr->inst.addr.nonce,
1782 sizeof(msgr->inst.addr.nonce));
1783
1784 dout("messenger_create %p\n", msgr);
1785 return msgr;
1786}
1787
1788void ceph_messenger_destroy(struct ceph_messenger *msgr)
1789{
1790 dout("destroy %p\n", msgr);
1791 kunmap(msgr->zero_page);
1792 __free_page(msgr->zero_page);
1793 kfree(msgr);
1794 dout("destroyed messenger %p\n", msgr);
1795}
1796
1797/*
1798 * Queue up an outgoing message on the given connection.
1799 */
1800void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
1801{
1802 if (test_bit(CLOSED, &con->state)) {
1803 dout("con_send %p closed, dropping %p\n", con, msg);
1804 ceph_msg_put(msg);
1805 return;
1806 }
1807
1808 /* set src+dst */
1809 msg->hdr.src = con->msgr->inst;
1810 msg->hdr.orig_src = con->msgr->inst;
1811 msg->hdr.dst_erank = con->peer_addr.erank;
1812
1813 /* queue */
1814 mutex_lock(&con->out_mutex);
1815 BUG_ON(!list_empty(&msg->list_head));
1816 list_add_tail(&msg->list_head, &con->out_queue);
1817 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
1818 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
1819 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
1820 le32_to_cpu(msg->hdr.front_len),
1821 le32_to_cpu(msg->hdr.middle_len),
1822 le32_to_cpu(msg->hdr.data_len));
1823 mutex_unlock(&con->out_mutex);
1824
1825 /* if there wasn't anything waiting to send before, queue
1826 * new work */
1827 if (test_and_set_bit(WRITE_PENDING, &con->state) == 0)
1828 queue_con(con);
1829}
1830
1831/*
1832 * Revoke a message that was previously queued for send
1833 */
1834void ceph_con_revoke(struct ceph_connection *con, struct ceph_msg *msg)
1835{
1836 mutex_lock(&con->out_mutex);
1837 if (!list_empty(&msg->list_head)) {
1838 dout("con_revoke %p msg %p\n", con, msg);
1839 list_del_init(&msg->list_head);
1840 ceph_msg_put(msg);
1841 msg->hdr.seq = 0;
1842 if (con->out_msg == msg)
1843 con->out_msg = NULL;
1844 if (con->out_kvec_is_msg) {
1845 con->out_skip = con->out_kvec_bytes;
1846 con->out_kvec_is_msg = false;
1847 }
1848 } else {
1849 dout("con_revoke %p msg %p - not queued (sent?)\n", con, msg);
1850 }
1851 mutex_unlock(&con->out_mutex);
1852}
1853
1854/*
1855 * Queue a keepalive byte to ensure the tcp connection is alive.
1856 */
1857void ceph_con_keepalive(struct ceph_connection *con)
1858{
1859 if (test_and_set_bit(KEEPALIVE_PENDING, &con->state) == 0 &&
1860 test_and_set_bit(WRITE_PENDING, &con->state) == 0)
1861 queue_con(con);
1862}
1863
1864
1865/*
1866 * construct a new message with given type, size
1867 * the new msg has a ref count of 1.
1868 */
1869struct ceph_msg *ceph_msg_new(int type, int front_len,
1870 int page_len, int page_off, struct page **pages)
1871{
1872 struct ceph_msg *m;
1873
1874 m = kmalloc(sizeof(*m), GFP_NOFS);
1875 if (m == NULL)
1876 goto out;
1877 atomic_set(&m->nref, 1);
1878 INIT_LIST_HEAD(&m->list_head);
1879
1880 m->hdr.type = cpu_to_le16(type);
1881 m->hdr.front_len = cpu_to_le32(front_len);
1882 m->hdr.middle_len = 0;
1883 m->hdr.data_len = cpu_to_le32(page_len);
1884 m->hdr.data_off = cpu_to_le16(page_off);
1885 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
1886 m->footer.front_crc = 0;
1887 m->footer.middle_crc = 0;
1888 m->footer.data_crc = 0;
1889 m->front_max = front_len;
1890 m->front_is_vmalloc = false;
1891 m->more_to_follow = false;
1892 m->pool = NULL;
1893
1894 /* front */
1895 if (front_len) {
1896 if (front_len > PAGE_CACHE_SIZE) {
1897 m->front.iov_base = __vmalloc(front_len, GFP_NOFS,
1898 PAGE_KERNEL);
1899 m->front_is_vmalloc = true;
1900 } else {
1901 m->front.iov_base = kmalloc(front_len, GFP_NOFS);
1902 }
1903 if (m->front.iov_base == NULL) {
1904 pr_err("msg_new can't allocate %d bytes\n",
1905 front_len);
1906 goto out2;
1907 }
1908 } else {
1909 m->front.iov_base = NULL;
1910 }
1911 m->front.iov_len = front_len;
1912
1913 /* middle */
1914 m->middle = NULL;
1915
1916 /* data */
1917 m->nr_pages = calc_pages_for(page_off, page_len);
1918 m->pages = pages;
1919
1920 dout("ceph_msg_new %p page %d~%d -> %d\n", m, page_off, page_len,
1921 m->nr_pages);
1922 return m;
1923
1924out2:
1925 ceph_msg_put(m);
1926out:
1927 pr_err("msg_new can't create type %d len %d\n", type, front_len);
1928 return ERR_PTR(-ENOMEM);
1929}
1930
1931/*
1932 * Generic message allocator, for incoming messages.
1933 */
1934struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con,
1935 struct ceph_msg_header *hdr)
1936{
1937 int type = le16_to_cpu(hdr->type);
1938 int front_len = le32_to_cpu(hdr->front_len);
1939 struct ceph_msg *msg = ceph_msg_new(type, front_len, 0, 0, NULL);
1940
1941 if (!msg) {
1942 pr_err("unable to allocate msg type %d len %d\n",
1943 type, front_len);
1944 return ERR_PTR(-ENOMEM);
1945 }
1946 return msg;
1947}
1948
1949/*
1950 * Allocate "middle" portion of a message, if it is needed and wasn't
1951 * allocated by alloc_msg. This allows us to read a small fixed-size
1952 * per-type header in the front and then gracefully fail (i.e.,
1953 * propagate the error to the caller based on info in the front) when
1954 * the middle is too large.
1955 */
1956int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
1957{
1958 int type = le16_to_cpu(msg->hdr.type);
1959 int middle_len = le32_to_cpu(msg->hdr.middle_len);
1960
1961 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
1962 ceph_msg_type_name(type), middle_len);
1963 BUG_ON(!middle_len);
1964 BUG_ON(msg->middle);
1965
1966 msg->middle = ceph_buffer_new_alloc(middle_len, GFP_NOFS);
1967 if (!msg->middle)
1968 return -ENOMEM;
1969 return 0;
1970}
1971
1972
1973/*
1974 * Free a generically kmalloc'd message.
1975 */
1976void ceph_msg_kfree(struct ceph_msg *m)
1977{
1978 dout("msg_kfree %p\n", m);
1979 if (m->front_is_vmalloc)
1980 vfree(m->front.iov_base);
1981 else
1982 kfree(m->front.iov_base);
1983 kfree(m);
1984}
1985
1986/*
1987 * Drop a msg ref. Destroy as needed.
1988 */
1989void ceph_msg_put(struct ceph_msg *m)
1990{
1991 dout("ceph_msg_put %p %d -> %d\n", m, atomic_read(&m->nref),
1992 atomic_read(&m->nref)-1);
1993 if (atomic_read(&m->nref) <= 0) {
1994 pr_err("bad ceph_msg_put on %p %llu %d=%s %d+%d\n",
1995 m, le64_to_cpu(m->hdr.seq),
1996 le16_to_cpu(m->hdr.type),
1997 ceph_msg_type_name(le16_to_cpu(m->hdr.type)),
1998 le32_to_cpu(m->hdr.front_len),
1999 le32_to_cpu(m->hdr.data_len));
2000 WARN_ON(1);
2001 }
2002 if (atomic_dec_and_test(&m->nref)) {
2003 dout("ceph_msg_put last one on %p\n", m);
2004 WARN_ON(!list_empty(&m->list_head));
2005
2006 /* drop middle, data, if any */
2007 if (m->middle) {
2008 ceph_buffer_put(m->middle);
2009 m->middle = NULL;
2010 }
2011 m->nr_pages = 0;
2012 m->pages = NULL;
2013
2014 if (m->pool)
2015 ceph_msgpool_put(m->pool, m);
2016 else
2017 ceph_msg_kfree(m);
2018 }
2019}
generated by cgit v1.2.2 (git 2.25.0) at 2024-12-24 08:11:28 -0500