1 #include "ceph_debug.h"
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>
9 #include <linux/socket.h>
10 #include <linux/string.h>
14 #include "messenger.h"
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
27 /* static tag bytes (protocol control messages) */
28 static char tag_msg = CEPH_MSGR_TAG_MSG;
29 static char tag_ack = CEPH_MSGR_TAG_ACK;
30 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
33 static void queue_con(struct ceph_connection *con);
34 static void con_work(struct work_struct *);
35 static void ceph_fault(struct ceph_connection *con);
37 const char *ceph_name_type_str(int 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 "???";
50 * nicely render a sockaddr as a string.
52 #define MAX_ADDR_STR 20
53 static char addr_str[MAX_ADDR_STR][40];
54 static DEFINE_SPINLOCK(addr_str_lock);
55 static int last_addr_str;
57 const char *pr_addr(const struct sockaddr_storage *ss)
61 struct sockaddr_in *in4 = (void *)ss;
62 unsigned char *quad = (void *)&in4->sin_addr.s_addr;
63 struct sockaddr_in6 *in6 = (void *)ss;
65 spin_lock(&addr_str_lock);
67 if (last_addr_str == MAX_ADDR_STR)
69 spin_unlock(&addr_str_lock);
72 switch (ss->ss_family) {
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));
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));
96 sprintf(s, "(unknown sockaddr family %d)", (int)ss->ss_family);
102 static void encode_my_addr(struct ceph_messenger *msgr)
104 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
105 ceph_encode_addr(&msgr->my_enc_addr);
109 * work queue for all reading and writing to/from the socket.
111 struct workqueue_struct *ceph_msgr_wq;
113 int __init ceph_msgr_init(void)
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);
125 void ceph_msgr_exit(void)
127 destroy_workqueue(ceph_msgr_wq);
131 * socket callback functions
134 /* data available on socket, or listen socket received a connect */
135 static void ceph_data_ready(struct sock *sk, int count_unused)
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",
146 /* socket has buffer space for writing */
147 static void ceph_write_space(struct sock *sk)
149 struct ceph_connection *con =
150 (struct ceph_connection *)sk->sk_user_data;
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);
157 dout("ceph_write_space %p nothing to write\n", con);
160 /* since we have our own write_space, clear the SOCK_NOSPACE flag */
161 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
164 /* socket's state has changed */
165 static void ceph_state_change(struct sock *sk)
167 struct ceph_connection *con =
168 (struct ceph_connection *)sk->sk_user_data;
170 dout("ceph_state_change %p state = %lu sk_state = %u\n",
171 con, con->state, sk->sk_state);
173 if (test_bit(CLOSED, &con->state))
176 switch (sk->sk_state) {
178 dout("ceph_state_change TCP_CLOSE\n");
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";
185 con->error_msg = "socket closed";
189 case TCP_ESTABLISHED:
190 dout("ceph_state_change TCP_ESTABLISHED\n");
197 * set up socket callbacks
199 static void set_sock_callbacks(struct socket *sock,
200 struct ceph_connection *con)
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;
215 * initiate connection to a remote socket.
217 static struct socket *ceph_tcp_connect(struct ceph_connection *con)
219 struct sockaddr *paddr = (struct sockaddr *)&con->peer_addr.in_addr;
224 ret = sock_create_kern(AF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
228 sock->sk->sk_allocation = GFP_NOFS;
230 set_sock_callbacks(sock, con);
232 dout("connect %s\n", pr_addr(&con->peer_addr.in_addr));
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),
242 pr_err("connect %s error %d\n",
243 pr_addr(&con->peer_addr.in_addr), ret);
246 con->error_msg = "connect error";
254 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
256 struct kvec iov = {buf, len};
257 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
259 return kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
263 * write something. @more is true if caller will be sending more data
266 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
267 size_t kvlen, size_t len, int more)
269 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
272 msg.msg_flags |= MSG_MORE;
274 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
276 return kernel_sendmsg(sock, &msg, iov, kvlen, len);
281 * Shutdown/close the socket for the given connection.
283 static int con_close_socket(struct ceph_connection *con)
287 dout("con_close_socket on %p sock %p\n", con, con->sock);
290 set_bit(SOCK_CLOSED, &con->state);
291 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
292 sock_release(con->sock);
294 clear_bit(SOCK_CLOSED, &con->state);
299 * Reset a connection. Discard all incoming and outgoing messages
300 * and clear *_seq state.
302 static void ceph_msg_remove(struct ceph_msg *msg)
304 list_del_init(&msg->list_head);
307 static void ceph_msg_remove_list(struct list_head *head)
309 while (!list_empty(head)) {
310 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
312 ceph_msg_remove(msg);
316 static void reset_connection(struct ceph_connection *con)
318 /* reset connection, out_queue, msg_ and connect_seq */
319 /* discard existing out_queue and msg_seq */
320 ceph_msg_remove_list(&con->out_queue);
321 ceph_msg_remove_list(&con->out_sent);
324 ceph_msg_put(con->in_msg);
328 con->connect_seq = 0;
331 ceph_msg_put(con->out_msg);
338 * mark a peer down. drop any open connections.
340 void ceph_con_close(struct ceph_connection *con)
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 */
345 mutex_lock(&con->mutex);
346 reset_connection(con);
347 cancel_delayed_work(&con->work);
348 mutex_unlock(&con->mutex);
353 * Reopen a closed connection, with a new peer address.
355 void ceph_con_open(struct ceph_connection *con, struct ceph_entity_addr *addr)
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));
361 con->delay = 0; /* reset backoff memory */
368 struct ceph_connection *ceph_con_get(struct ceph_connection *con)
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))
377 void ceph_con_put(struct ceph_connection *con)
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)) {
389 * initialize a new connection.
391 void ceph_con_init(struct ceph_messenger *msgr, struct ceph_connection *con)
393 dout("con_init %p\n", con);
394 memset(con, 0, sizeof(*con));
395 atomic_set(&con->nref, 1);
397 mutex_init(&con->mutex);
398 INIT_LIST_HEAD(&con->out_queue);
399 INIT_LIST_HEAD(&con->out_sent);
400 INIT_DELAYED_WORK(&con->work, con_work);
405 * We maintain a global counter to order connection attempts. Get
406 * a unique seq greater than @gt.
408 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
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);
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.
425 static void prepare_write_message_footer(struct ceph_connection *con, int v)
427 struct ceph_msg *m = con->out_msg;
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;
436 con->out_msg_done = true;
440 * Prepare headers for the next outgoing message.
442 static void prepare_write_message(struct ceph_connection *con)
447 con->out_kvec_bytes = 0;
448 con->out_kvec_is_msg = true;
449 con->out_msg_done = false;
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);
463 m = list_first_entry(&con->out_queue,
464 struct ceph_msg, list_head);
466 if (test_bit(LOSSYTX, &con->state)) {
467 list_del_init(&m->list_head);
469 /* put message on sent list */
471 list_move_tail(&m->list_head, &con->out_sent);
474 m->hdr.seq = cpu_to_le64(++con->out_seq);
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),
481 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
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;
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;
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));
504 con->out_msg->footer.middle_crc =
505 cpu_to_le32(crc32c(0, m->middle->vec.iov_base,
506 m->middle->vec.iov_len));
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));
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 */
524 /* no, queue up footer too and be done */
525 prepare_write_message_footer(con, v);
528 set_bit(WRITE_PENDING, &con->state);
534 static void prepare_write_ack(struct ceph_connection *con)
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;
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);
553 * Prepare to write keepalive byte.
555 static void prepare_write_keepalive(struct ceph_connection *con)
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);
567 * Connection negotiation.
570 static void prepare_connect_authorizer(struct ceph_connection *con)
574 int auth_protocol = 0;
576 mutex_unlock(&con->mutex);
577 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,
582 mutex_lock(&con->mutex);
584 con->out_connect.authorizer_protocol = cpu_to_le32(auth_protocol);
585 con->out_connect.authorizer_len = cpu_to_le32(auth_len);
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;
594 * We connected to a peer and are saying hello.
596 static void prepare_write_banner(struct ceph_messenger *msgr,
597 struct ceph_connection *con)
599 int len = strlen(CEPH_BANNER);
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;
609 set_bit(WRITE_PENDING, &con->state);
612 static void prepare_write_connect(struct ceph_messenger *msgr,
613 struct ceph_connection *con,
616 unsigned global_seq = get_global_seq(con->msgr, 0);
619 switch (con->peer_name.type) {
620 case CEPH_ENTITY_TYPE_MON:
621 proto = CEPH_MONC_PROTOCOL;
623 case CEPH_ENTITY_TYPE_OSD:
624 proto = CEPH_OSDC_PROTOCOL;
626 case CEPH_ENTITY_TYPE_MDS:
627 proto = CEPH_MDSC_PROTOCOL;
633 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
634 con->connect_seq, global_seq, proto);
636 con->out_connect.features = CEPH_FEATURE_SUPPORTED;
637 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;
644 con->out_kvec_left = 0;
645 con->out_kvec_bytes = 0;
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);
651 con->out_kvec_cur = con->out_kvec;
653 set_bit(WRITE_PENDING, &con->state);
655 prepare_connect_authorizer(con);
660 * write as much of pending kvecs to the socket as we can.
662 * 0 -> socket full, but more to do
665 static int write_partial_kvec(struct ceph_connection *con)
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,
676 con->out_kvec_bytes -= ret;
677 if (con->out_kvec_bytes == 0)
680 if (ret >= con->out_kvec_cur->iov_len) {
681 ret -= con->out_kvec_cur->iov_len;
683 con->out_kvec_left--;
685 con->out_kvec_cur->iov_len -= ret;
686 con->out_kvec_cur->iov_base += ret;
692 con->out_kvec_left = 0;
693 con->out_kvec_is_msg = false;
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! */
702 * Write as much message data payload as we can. If we finish, queue
704 * 1 -> done, footer is now queued in out_kvec[].
705 * 0 -> socket full, but more to do
708 static int write_partial_msg_pages(struct ceph_connection *con)
710 struct ceph_msg *msg = con->out_msg;
711 unsigned data_len = le32_to_cpu(msg->hdr.data_len);
713 int crc = con->msgr->nocrc;
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);
720 while (con->out_msg_pos.page < con->out_msg->nr_pages) {
721 struct page *page = NULL;
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
730 page = msg->pages[con->out_msg_pos.page];
733 } else if (msg->pagelist) {
734 page = list_first_entry(&msg->pagelist->head,
739 page = con->msgr->zero_page;
741 kaddr = page_address(con->msgr->zero_page);
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);
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;
755 ret = kernel_sendpage(con->sock, page,
756 con->out_msg_pos.page_pos, len,
757 MSG_DONTWAIT | MSG_NOSIGNAL |
760 if (crc && (msg->pages || msg->pagelist))
766 con->out_msg_pos.data_pos += ret;
767 con->out_msg_pos.page_pos += ret;
769 con->out_msg_pos.page_pos = 0;
770 con->out_msg_pos.page++;
771 con->out_msg_pos.did_page_crc = 0;
773 list_move_tail(&page->lru,
774 &msg->pagelist->head);
778 dout("write_partial_msg_pages %p msg %p done\n", con, msg);
780 /* prepare and queue up footer, too */
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);
795 static int write_partial_skip(struct ceph_connection *con)
799 while (con->out_skip > 0) {
801 .iov_base = page_address(con->msgr->zero_page),
802 .iov_len = min(con->out_skip, (int)PAGE_CACHE_SIZE)
805 ret = ceph_tcp_sendmsg(con->sock, &iov, 1, iov.iov_len, 1);
808 con->out_skip -= ret;
816 * Prepare to read connection handshake, or an ack.
818 static void prepare_read_banner(struct ceph_connection *con)
820 dout("prepare_read_banner %p\n", con);
821 con->in_base_pos = 0;
824 static void prepare_read_connect(struct ceph_connection *con)
826 dout("prepare_read_connect %p\n", con);
827 con->in_base_pos = 0;
830 static void prepare_read_connect_retry(struct ceph_connection *con)
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);
837 static void prepare_read_ack(struct ceph_connection *con)
839 dout("prepare_read_ack %p\n", con);
840 con->in_base_pos = 0;
843 static void prepare_read_tag(struct ceph_connection *con)
845 dout("prepare_read_tag %p\n", con);
846 con->in_base_pos = 0;
847 con->in_tag = CEPH_MSGR_TAG_READY;
851 * Prepare to read a message.
853 static int prepare_read_message(struct ceph_connection *con)
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;
863 static int read_partial(struct ceph_connection *con,
864 int *to, int size, void *object)
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);
873 con->in_base_pos += ret;
880 * Read all or part of the connect-side handshake on a new connection
882 static int read_partial_banner(struct ceph_connection *con)
886 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
889 ret = read_partial(con, &to, strlen(CEPH_BANNER), con->in_banner);
892 ret = read_partial(con, &to, sizeof(con->actual_peer_addr),
893 &con->actual_peer_addr);
896 ret = read_partial(con, &to, sizeof(con->peer_addr_for_me),
897 &con->peer_addr_for_me);
904 static int read_partial_connect(struct ceph_connection *con)
908 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
910 ret = read_partial(con, &to, sizeof(con->in_reply), &con->in_reply);
913 ret = read_partial(con, &to, le32_to_cpu(con->in_reply.authorizer_len),
914 con->auth_reply_buf);
918 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),
921 le32_to_cpu(con->in_reply.global_seq));
928 * Verify the hello banner looks okay.
930 static int verify_hello(struct ceph_connection *con)
932 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
933 pr_err("connect to %s got bad banner\n",
934 pr_addr(&con->peer_addr.in_addr));
935 con->error_msg = "protocol error, bad banner";
941 static bool addr_is_blank(struct sockaddr_storage *ss)
943 switch (ss->ss_family) {
945 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
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;
956 static int addr_port(struct sockaddr_storage *ss)
958 switch (ss->ss_family) {
960 return ntohs(((struct sockaddr_in *)ss)->sin_port);
962 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
967 static void addr_set_port(struct sockaddr_storage *ss, int p)
969 switch (ss->ss_family) {
971 ((struct sockaddr_in *)ss)->sin_port = htons(p);
973 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
978 * Parse an ip[:port] list into an addr array. Use the default
979 * monitor port if a port isn't specified.
981 int ceph_parse_ips(const char *c, const char *end,
982 struct ceph_entity_addr *addr,
983 int max_count, int *count)
988 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
989 for (i = 0; i < max_count; i++) {
991 struct sockaddr_storage *ss = &addr[i].in_addr;
992 struct sockaddr_in *in4 = (void *)ss;
993 struct sockaddr_in6 *in6 = (void *)ss;
996 memset(ss, 0, sizeof(*ss));
997 if (in4_pton(p, end - p, (u8 *)&in4->sin_addr.s_addr,
999 ss->ss_family = AF_INET;
1000 } else if (in6_pton(p, end - p, (u8 *)&in6->sin6_addr.s6_addr,
1002 ss->ss_family = AF_INET6;
1009 if (p < end && *p == ':') {
1012 while (p < end && *p >= '0' && *p <= '9') {
1013 port = (port * 10) + (*p - '0');
1016 if (port > 65535 || port == 0)
1019 port = CEPH_MON_PORT;
1022 addr_set_port(ss, port);
1024 dout("parse_ips got %s\n", pr_addr(ss));
1041 pr_err("parse_ips bad ip '%s'\n", c);
1045 static int process_banner(struct ceph_connection *con)
1047 dout("process_banner on %p\n", con);
1049 if (verify_hello(con) < 0)
1052 ceph_decode_addr(&con->actual_peer_addr);
1053 ceph_decode_addr(&con->peer_addr_for_me);
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.
1060 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1061 sizeof(con->peer_addr)) != 0 &&
1062 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1063 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1064 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));
1069 con->error_msg = "wrong peer at address";
1074 * did we learn our address?
1076 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1077 int port = addr_port(&con->msgr->inst.addr.in_addr);
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);
1083 encode_my_addr(con->msgr);
1084 dout("process_banner learned my addr is %s\n",
1085 pr_addr(&con->msgr->inst.addr.in_addr));
1088 set_bit(NEGOTIATING, &con->state);
1089 prepare_read_connect(con);
1093 static void fail_protocol(struct ceph_connection *con)
1095 reset_connection(con);
1096 set_bit(CLOSED, &con->state); /* in case there's queued work */
1098 mutex_unlock(&con->mutex);
1099 if (con->ops->bad_proto)
1100 con->ops->bad_proto(con);
1101 mutex_lock(&con->mutex);
1104 static int process_connect(struct ceph_connection *con)
1106 u64 sup_feat = CEPH_FEATURE_SUPPORTED;
1107 u64 req_feat = CEPH_FEATURE_REQUIRED;
1108 u64 server_feat = le64_to_cpu(con->in_reply.features);
1110 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1112 switch (con->in_reply.tag) {
1113 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";
1123 case CEPH_MSGR_TAG_BADPROTOVER:
1124 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";
1134 case CEPH_MSGR_TAG_BADAUTHORIZER:
1136 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1138 if (con->auth_retry == 2) {
1139 con->error_msg = "connect authorization failure";
1140 reset_connection(con);
1141 set_bit(CLOSED, &con->state);
1144 con->auth_retry = 1;
1145 prepare_write_connect(con->msgr, con, 0);
1146 prepare_read_connect_retry(con);
1149 case CEPH_MSGR_TAG_RESETSESSION:
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
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);
1163 prepare_write_connect(con->msgr, con, 0);
1164 prepare_read_connect(con);
1166 /* Tell ceph about it. */
1167 mutex_unlock(&con->mutex);
1168 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
1169 if (con->ops->peer_reset)
1170 con->ops->peer_reset(con);
1171 mutex_lock(&con->mutex);
1174 case CEPH_MSGR_TAG_RETRY_SESSION:
1176 * If we sent a smaller connect_seq than the peer has, try
1177 * again with a larger value.
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);
1183 prepare_write_connect(con->msgr, con, 0);
1184 prepare_read_connect(con);
1187 case CEPH_MSGR_TAG_RETRY_GLOBAL:
1189 * If we sent a smaller global_seq than the peer has, try
1190 * again with a larger value.
1192 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1193 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));
1197 prepare_write_connect(con->msgr, con, 0);
1198 prepare_read_connect(con);
1201 case CEPH_MSGR_TAG_READY:
1202 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";
1212 clear_bit(CONNECTING, &con->state);
1213 con->peer_global_seq = le32_to_cpu(con->in_reply.global_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),
1219 WARN_ON(con->connect_seq !=
1220 le32_to_cpu(con->in_reply.connect_seq));
1222 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
1223 set_bit(LOSSYTX, &con->state);
1225 prepare_read_tag(con);
1228 case CEPH_MSGR_TAG_WAIT:
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
1235 pr_err("process_connect peer connecting WAIT\n");
1238 pr_err("connect protocol error, will retry\n");
1239 con->error_msg = "protocol error, garbage tag during connect";
1247 * read (part of) an ack
1249 static int read_partial_ack(struct ceph_connection *con)
1253 return read_partial(con, &to, sizeof(con->in_temp_ack),
1259 * We can finally discard anything that's been acked.
1261 static void process_ack(struct ceph_connection *con)
1264 u64 ack = le64_to_cpu(con->in_temp_ack);
1267 while (!list_empty(&con->out_sent)) {
1268 m = list_first_entry(&con->out_sent, struct ceph_msg,
1270 seq = le64_to_cpu(m->hdr.seq);
1273 dout("got ack for seq %llu type %d at %p\n", seq,
1274 le16_to_cpu(m->hdr.type), m);
1277 prepare_read_tag(con);
1283 static int read_partial_message_section(struct ceph_connection *con,
1284 struct kvec *section, unsigned int sec_len,
1292 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);
1299 section->iov_len += ret;
1300 if (section->iov_len == sec_len)
1301 *crc = crc32c(0, section->iov_base,
1308 static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con,
1309 struct ceph_msg_header *hdr,
1312 * read (part of) a message.
1314 static int read_partial_message(struct ceph_connection *con)
1316 struct ceph_msg *m = con->in_msg;
1320 unsigned front_len, middle_len, data_len, data_off;
1321 int datacrc = con->msgr->nocrc;
1324 dout("read_partial_message con %p msg %p\n", con, m);
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,
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);
1346 front_len = le32_to_cpu(con->in_hdr.front_len);
1347 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
1349 middle_len = le32_to_cpu(con->in_hdr.middle_len);
1350 if (middle_len > CEPH_MSG_MAX_DATA_LEN)
1352 data_len = le32_to_cpu(con->in_hdr.data_len);
1353 if (data_len > CEPH_MSG_MAX_DATA_LEN)
1355 data_off = le16_to_cpu(con->in_hdr.data_off);
1357 /* allocate message? */
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);
1361 con->in_msg = ceph_alloc_msg(con, &con->in_hdr, &skip);
1363 /* skip this message */
1364 dout("alloc_msg returned NULL, skipping message\n");
1365 con->in_base_pos = -front_len - middle_len - data_len -
1367 con->in_tag = CEPH_MSGR_TAG_READY;
1370 if (IS_ERR(con->in_msg)) {
1371 ret = PTR_ERR(con->in_msg);
1374 "error allocating memory for incoming message";
1378 m->front.iov_len = 0; /* haven't read it yet */
1380 m->middle->vec.iov_len = 0;
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;
1388 ret = read_partial_message_section(con, &m->front, front_len,
1389 &con->in_front_crc);
1395 ret = read_partial_message_section(con, &m->middle->vec, middle_len,
1396 &con->in_middle_crc);
1402 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,
1409 if (ret > 0 && datacrc)
1411 crc32c(con->in_data_crc,
1412 p + con->in_msg_pos.page_pos, ret);
1413 kunmap(m->pages[con->in_msg_pos.page]);
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++;
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)),
1433 con->in_base_pos += ret;
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);
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);
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);
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));
1458 return 1; /* done! */
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
1466 static void process_message(struct ceph_connection *con)
1468 struct ceph_msg *msg;
1473 /* if first message, set peer_name */
1474 if (con->peer_name.type == 0)
1475 con->peer_name = msg->hdr.src.name;
1478 mutex_unlock(&con->mutex);
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);
1490 mutex_lock(&con->mutex);
1491 prepare_read_tag(con);
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.
1499 static int try_write(struct ceph_connection *con)
1501 struct ceph_messenger *msgr = con->msgr;
1504 dout("try_write start %p state %lu nref %d\n", con, con->state,
1505 atomic_read(&con->nref));
1507 mutex_lock(&con->mutex);
1509 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
1511 /* open the socket first? */
1512 if (con->sock == NULL) {
1514 * if we were STANDBY and are reconnecting _this_
1515 * connection, bump connect_seq now. Always bump
1518 if (test_and_clear_bit(STANDBY, &con->state))
1521 prepare_write_banner(msgr, con);
1522 prepare_write_connect(msgr, con, 1);
1523 prepare_read_banner(con);
1524 set_bit(CONNECTING, &con->state);
1525 clear_bit(NEGOTIATING, &con->state);
1527 BUG_ON(con->in_msg);
1528 con->in_tag = CEPH_MSGR_TAG_READY;
1529 dout("try_write initiating connect on %p new state %lu\n",
1531 con->sock = ceph_tcp_connect(con);
1532 if (IS_ERR(con->sock)) {
1534 con->error_msg = "connect error";
1541 /* kvec data queued? */
1542 if (con->out_skip) {
1543 ret = write_partial_skip(con);
1547 dout("try_write write_partial_skip err %d\n", ret);
1551 if (con->out_kvec_left) {
1552 ret = write_partial_kvec(con);
1559 if (con->out_msg_done) {
1560 ceph_msg_put(con->out_msg);
1561 con->out_msg = NULL; /* we're done with this one */
1565 ret = write_partial_msg_pages(con);
1567 goto more_kvec; /* we need to send the footer, too! */
1571 dout("try_write write_partial_msg_pages err %d\n",
1578 if (!test_bit(CONNECTING, &con->state)) {
1579 /* is anything else pending? */
1580 if (!list_empty(&con->out_queue)) {
1581 prepare_write_message(con);
1584 if (con->in_seq > con->in_seq_acked) {
1585 prepare_write_ack(con);
1588 if (test_and_clear_bit(KEEPALIVE_PENDING, &con->state)) {
1589 prepare_write_keepalive(con);
1594 /* Nothing to do! */
1595 clear_bit(WRITE_PENDING, &con->state);
1596 dout("try_write nothing else to write.\n");
1600 mutex_unlock(&con->mutex);
1601 dout("try_write done on %p\n", con);
1608 * Read what we can from the socket.
1610 static int try_read(struct ceph_connection *con)
1612 struct ceph_messenger *msgr;
1618 if (test_bit(STANDBY, &con->state))
1621 dout("try_read start on %p\n", con);
1624 mutex_lock(&con->mutex);
1627 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
1629 if (test_bit(CONNECTING, &con->state)) {
1630 if (!test_bit(NEGOTIATING, &con->state)) {
1631 dout("try_read connecting\n");
1632 ret = read_partial_banner(con);
1635 if (process_banner(con) < 0) {
1640 ret = read_partial_connect(con);
1643 if (process_connect(con) < 0) {
1650 if (con->in_base_pos < 0) {
1652 * skipping + discarding content.
1654 * FIXME: there must be a better way to do this!
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);
1662 con->in_base_pos += ret;
1663 if (con->in_base_pos)
1666 if (con->in_tag == CEPH_MSGR_TAG_READY) {
1670 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
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);
1678 case CEPH_MSGR_TAG_ACK:
1679 prepare_read_ack(con);
1681 case CEPH_MSGR_TAG_CLOSE:
1682 set_bit(CLOSED, &con->state); /* fixme */
1688 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
1689 ret = read_partial_message(con);
1693 con->error_msg = "bad crc";
1697 con->error_msg = "io error";
1703 if (con->in_tag == CEPH_MSGR_TAG_READY)
1705 process_message(con);
1708 if (con->in_tag == CEPH_MSGR_TAG_ACK) {
1709 ret = read_partial_ack(con);
1719 mutex_unlock(&con->mutex);
1720 dout("try_read done on %p\n", con);
1724 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
1725 con->error_msg = "protocol error, garbage tag";
1732 * Atomically queue work on a connection. Bump @con reference to
1733 * avoid races with connection teardown.
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.
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).
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.
1749 static void queue_con(struct ceph_connection *con)
1751 if (test_bit(DEAD, &con->state)) {
1752 dout("queue_con %p ignoring: DEAD\n",
1757 if (!con->ops->get(con)) {
1758 dout("queue_con %p ref count 0\n", con);
1762 set_bit(QUEUED, &con->state);
1763 if (test_bit(BUSY, &con->state)) {
1764 dout("queue_con %p - already BUSY\n", con);
1766 } else if (!queue_work(ceph_msgr_wq, &con->work.work)) {
1767 dout("queue_con %p - already queued\n", con);
1770 dout("queue_con %p\n", con);
1775 * Do some work on a connection. Drop a connection ref when we're done.
1777 static void con_work(struct work_struct *work)
1779 struct ceph_connection *con = container_of(work, struct ceph_connection,
1784 if (test_and_set_bit(BUSY, &con->state) != 0) {
1785 dout("con_work %p BUSY already set\n", con);
1788 dout("con_work %p start, clearing QUEUED\n", con);
1789 clear_bit(QUEUED, &con->state);
1791 if (test_bit(CLOSED, &con->state)) { /* e.g. if we are replaced */
1792 dout("con_work CLOSED\n");
1793 con_close_socket(con);
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);
1802 if (test_and_clear_bit(SOCK_CLOSED, &con->state) ||
1803 try_read(con) < 0 ||
1804 try_write(con) < 0) {
1806 ceph_fault(con); /* error/fault path */
1810 clear_bit(BUSY, &con->state);
1811 dout("con->state=%lu\n", con->state);
1812 if (test_bit(QUEUED, &con->state)) {
1813 if (!backoff || test_bit(OPENING, &con->state)) {
1814 dout("con_work %p QUEUED reset, looping\n", con);
1817 dout("con_work %p QUEUED reset, but just faulted\n", con);
1818 clear_bit(QUEUED, &con->state);
1820 dout("con_work %p done\n", con);
1828 * Generic error/fault handler. A retry mechanism is used with
1829 * exponential backoff
1831 static void ceph_fault(struct ceph_connection *con)
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));
1838 if (test_bit(LOSSYTX, &con->state)) {
1839 dout("fault on LOSSYTX channel\n");
1843 clear_bit(BUSY, &con->state); /* to avoid an improbable race */
1845 mutex_lock(&con->mutex);
1846 if (test_bit(CLOSED, &con->state))
1849 con_close_socket(con);
1852 ceph_msg_put(con->in_msg);
1857 * in case we faulted due to authentication, invalidate our
1858 * current tickets so that we can get new ones.
1860 if (con->auth_retry && con->ops->invalidate_authorizer) {
1861 dout("calling invalidate_authorizer()\n");
1862 con->ops->invalidate_authorizer(con);
1865 /* If there are no messages in the queue, place the connection
1866 * in a STANDBY state (i.e., don't try to reconnect just yet). */
1867 if (list_empty(&con->out_queue) && !con->out_keepalive_pending) {
1868 dout("fault setting STANDBY\n");
1869 set_bit(STANDBY, &con->state);
1870 mutex_unlock(&con->mutex);
1874 /* Requeue anything that hasn't been acked, and retry after a
1876 list_splice_init(&con->out_sent, &con->out_queue);
1878 if (con->delay == 0)
1879 con->delay = BASE_DELAY_INTERVAL;
1880 else if (con->delay < MAX_DELAY_INTERVAL)
1883 /* explicitly schedule work to try to reconnect again later. */
1884 dout("fault queueing %p delay %lu\n", con, con->delay);
1886 if (queue_delayed_work(ceph_msgr_wq, &con->work,
1887 round_jiffies_relative(con->delay)) == 0)
1891 mutex_unlock(&con->mutex);
1893 if (con->ops->fault)
1894 con->ops->fault(con);
1900 * create a new messenger instance
1902 struct ceph_messenger *ceph_messenger_create(struct ceph_entity_addr *myaddr)
1904 struct ceph_messenger *msgr;
1906 msgr = kzalloc(sizeof(*msgr), GFP_KERNEL);
1908 return ERR_PTR(-ENOMEM);
1910 spin_lock_init(&msgr->global_seq_lock);
1912 /* the zero page is needed if a request is "canceled" while the message
1913 * is being written over the socket */
1914 msgr->zero_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1915 if (!msgr->zero_page) {
1917 return ERR_PTR(-ENOMEM);
1919 kmap(msgr->zero_page);
1922 msgr->inst.addr = *myaddr;
1924 /* select a random nonce */
1925 msgr->inst.addr.type = 0;
1926 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
1927 encode_my_addr(msgr);
1929 dout("messenger_create %p\n", msgr);
1933 void ceph_messenger_destroy(struct ceph_messenger *msgr)
1935 dout("destroy %p\n", msgr);
1936 kunmap(msgr->zero_page);
1937 __free_page(msgr->zero_page);
1939 dout("destroyed messenger %p\n", msgr);
1943 * Queue up an outgoing message on the given connection.
1945 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
1947 if (test_bit(CLOSED, &con->state)) {
1948 dout("con_send %p closed, dropping %p\n", con, msg);
1954 msg->hdr.src.name = con->msgr->inst.name;
1955 msg->hdr.src.addr = con->msgr->my_enc_addr;
1956 msg->hdr.orig_src = msg->hdr.src;
1959 mutex_lock(&con->mutex);
1960 BUG_ON(!list_empty(&msg->list_head));
1961 list_add_tail(&msg->list_head, &con->out_queue);
1962 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
1963 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
1964 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
1965 le32_to_cpu(msg->hdr.front_len),
1966 le32_to_cpu(msg->hdr.middle_len),
1967 le32_to_cpu(msg->hdr.data_len));
1968 mutex_unlock(&con->mutex);
1970 /* if there wasn't anything waiting to send before, queue
1972 if (test_and_set_bit(WRITE_PENDING, &con->state) == 0)
1977 * Revoke a message that was previously queued for send
1979 void ceph_con_revoke(struct ceph_connection *con, struct ceph_msg *msg)
1981 mutex_lock(&con->mutex);
1982 if (!list_empty(&msg->list_head)) {
1983 dout("con_revoke %p msg %p\n", con, msg);
1984 list_del_init(&msg->list_head);
1987 if (con->out_msg == msg) {
1988 ceph_msg_put(con->out_msg);
1989 con->out_msg = NULL;
1991 if (con->out_kvec_is_msg) {
1992 con->out_skip = con->out_kvec_bytes;
1993 con->out_kvec_is_msg = false;
1996 dout("con_revoke %p msg %p - not queued (sent?)\n", con, msg);
1998 mutex_unlock(&con->mutex);
2002 * Revoke a message that we may be reading data into
2004 void ceph_con_revoke_message(struct ceph_connection *con, struct ceph_msg *msg)
2006 mutex_lock(&con->mutex);
2007 if (con->in_msg && con->in_msg == msg) {
2008 unsigned front_len = le32_to_cpu(con->in_hdr.front_len);
2009 unsigned middle_len = le32_to_cpu(con->in_hdr.middle_len);
2010 unsigned data_len = le32_to_cpu(con->in_hdr.data_len);
2012 /* skip rest of message */
2013 dout("con_revoke_pages %p msg %p revoked\n", con, msg);
2014 con->in_base_pos = con->in_base_pos -
2015 sizeof(struct ceph_msg_header) -
2019 sizeof(struct ceph_msg_footer);
2020 ceph_msg_put(con->in_msg);
2022 con->in_tag = CEPH_MSGR_TAG_READY;
2024 dout("con_revoke_pages %p msg %p pages %p no-op\n",
2025 con, con->in_msg, msg);
2027 mutex_unlock(&con->mutex);
2031 * Queue a keepalive byte to ensure the tcp connection is alive.
2033 void ceph_con_keepalive(struct ceph_connection *con)
2035 if (test_and_set_bit(KEEPALIVE_PENDING, &con->state) == 0 &&
2036 test_and_set_bit(WRITE_PENDING, &con->state) == 0)
2042 * construct a new message with given type, size
2043 * the new msg has a ref count of 1.
2045 struct ceph_msg *ceph_msg_new(int type, int front_len,
2046 int page_len, int page_off, struct page **pages)
2050 m = kmalloc(sizeof(*m), GFP_NOFS);
2053 kref_init(&m->kref);
2054 INIT_LIST_HEAD(&m->list_head);
2056 m->hdr.type = cpu_to_le16(type);
2057 m->hdr.front_len = cpu_to_le32(front_len);
2058 m->hdr.middle_len = 0;
2059 m->hdr.data_len = cpu_to_le32(page_len);
2060 m->hdr.data_off = cpu_to_le16(page_off);
2061 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
2062 m->footer.front_crc = 0;
2063 m->footer.middle_crc = 0;
2064 m->footer.data_crc = 0;
2065 m->front_max = front_len;
2066 m->front_is_vmalloc = false;
2067 m->more_to_follow = false;
2072 if (front_len > PAGE_CACHE_SIZE) {
2073 m->front.iov_base = __vmalloc(front_len, GFP_NOFS,
2075 m->front_is_vmalloc = true;
2077 m->front.iov_base = kmalloc(front_len, GFP_NOFS);
2079 if (m->front.iov_base == NULL) {
2080 pr_err("msg_new can't allocate %d bytes\n",
2085 m->front.iov_base = NULL;
2087 m->front.iov_len = front_len;
2093 m->nr_pages = calc_pages_for(page_off, page_len);
2097 dout("ceph_msg_new %p page %d~%d -> %d\n", m, page_off, page_len,
2104 pr_err("msg_new can't create type %d len %d\n", type, front_len);
2105 return ERR_PTR(-ENOMEM);
2109 * Allocate "middle" portion of a message, if it is needed and wasn't
2110 * allocated by alloc_msg. This allows us to read a small fixed-size
2111 * per-type header in the front and then gracefully fail (i.e.,
2112 * propagate the error to the caller based on info in the front) when
2113 * the middle is too large.
2115 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
2117 int type = le16_to_cpu(msg->hdr.type);
2118 int middle_len = le32_to_cpu(msg->hdr.middle_len);
2120 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
2121 ceph_msg_type_name(type), middle_len);
2122 BUG_ON(!middle_len);
2123 BUG_ON(msg->middle);
2125 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
2132 * Generic message allocator, for incoming messages.
2134 static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con,
2135 struct ceph_msg_header *hdr,
2138 int type = le16_to_cpu(hdr->type);
2139 int front_len = le32_to_cpu(hdr->front_len);
2140 int middle_len = le32_to_cpu(hdr->middle_len);
2141 struct ceph_msg *msg = NULL;
2144 if (con->ops->alloc_msg) {
2145 mutex_unlock(&con->mutex);
2146 msg = con->ops->alloc_msg(con, hdr, skip);
2147 mutex_lock(&con->mutex);
2156 msg = ceph_msg_new(type, front_len, 0, 0, NULL);
2158 pr_err("unable to allocate msg type %d len %d\n",
2160 return ERR_PTR(-ENOMEM);
2163 memcpy(&msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2166 ret = ceph_alloc_middle(con, msg);
2179 * Free a generically kmalloc'd message.
2181 void ceph_msg_kfree(struct ceph_msg *m)
2183 dout("msg_kfree %p\n", m);
2184 if (m->front_is_vmalloc)
2185 vfree(m->front.iov_base);
2187 kfree(m->front.iov_base);
2192 * Drop a msg ref. Destroy as needed.
2194 void ceph_msg_last_put(struct kref *kref)
2196 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
2198 dout("ceph_msg_put last one on %p\n", m);
2199 WARN_ON(!list_empty(&m->list_head));
2201 /* drop middle, data, if any */
2203 ceph_buffer_put(m->middle);
2210 ceph_pagelist_release(m->pagelist);
2216 ceph_msgpool_put(m->pool, m);
2221 void ceph_msg_dump(struct ceph_msg *msg)
2223 pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg,
2224 msg->front_max, msg->nr_pages);
2225 print_hex_dump(KERN_DEBUG, "header: ",
2226 DUMP_PREFIX_OFFSET, 16, 1,
2227 &msg->hdr, sizeof(msg->hdr), true);
2228 print_hex_dump(KERN_DEBUG, " front: ",
2229 DUMP_PREFIX_OFFSET, 16, 1,
2230 msg->front.iov_base, msg->front.iov_len, true);
2232 print_hex_dump(KERN_DEBUG, "middle: ",
2233 DUMP_PREFIX_OFFSET, 16, 1,
2234 msg->middle->vec.iov_base,
2235 msg->middle->vec.iov_len, true);
2236 print_hex_dump(KERN_DEBUG, "footer: ",
2237 DUMP_PREFIX_OFFSET, 16, 1,
2238 &msg->footer, sizeof(msg->footer), true);