1 #include <linux/ceph/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/slab.h>
10 #include <linux/socket.h>
11 #include <linux/string.h>
12 #include <linux/bio.h>
13 #include <linux/blkdev.h>
14 #include <linux/dns_resolver.h>
17 #include <linux/ceph/libceph.h>
18 #include <linux/ceph/messenger.h>
19 #include <linux/ceph/decode.h>
20 #include <linux/ceph/pagelist.h>
21 #include <linux/export.h>
24 * Ceph uses the messenger to exchange ceph_msg messages with other
25 * hosts in the system. The messenger provides ordered and reliable
26 * delivery. We tolerate TCP disconnects by reconnecting (with
27 * exponential backoff) in the case of a fault (disconnection, bad
28 * crc, protocol error). Acks allow sent messages to be discarded by
32 /* static tag bytes (protocol control messages) */
33 static char tag_msg = CEPH_MSGR_TAG_MSG;
34 static char tag_ack = CEPH_MSGR_TAG_ACK;
35 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
38 static struct lock_class_key socket_class;
42 static void queue_con(struct ceph_connection *con);
43 static void con_work(struct work_struct *);
44 static void ceph_fault(struct ceph_connection *con);
47 * nicely render a sockaddr as a string.
49 #define MAX_ADDR_STR 20
50 #define MAX_ADDR_STR_LEN 60
51 static char addr_str[MAX_ADDR_STR][MAX_ADDR_STR_LEN];
52 static DEFINE_SPINLOCK(addr_str_lock);
53 static int last_addr_str;
55 const char *ceph_pr_addr(const struct sockaddr_storage *ss)
59 struct sockaddr_in *in4 = (void *)ss;
60 struct sockaddr_in6 *in6 = (void *)ss;
62 spin_lock(&addr_str_lock);
64 if (last_addr_str == MAX_ADDR_STR)
66 spin_unlock(&addr_str_lock);
69 switch (ss->ss_family) {
71 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%u", &in4->sin_addr,
72 (unsigned int)ntohs(in4->sin_port));
76 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%u", &in6->sin6_addr,
77 (unsigned int)ntohs(in6->sin6_port));
81 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %d)",
87 EXPORT_SYMBOL(ceph_pr_addr);
89 static void encode_my_addr(struct ceph_messenger *msgr)
91 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
92 ceph_encode_addr(&msgr->my_enc_addr);
96 * work queue for all reading and writing to/from the socket.
98 struct workqueue_struct *ceph_msgr_wq;
100 int ceph_msgr_init(void)
102 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0);
104 pr_err("msgr_init failed to create workqueue\n");
109 EXPORT_SYMBOL(ceph_msgr_init);
111 void ceph_msgr_exit(void)
113 destroy_workqueue(ceph_msgr_wq);
115 EXPORT_SYMBOL(ceph_msgr_exit);
117 void ceph_msgr_flush(void)
119 flush_workqueue(ceph_msgr_wq);
121 EXPORT_SYMBOL(ceph_msgr_flush);
125 * socket callback functions
128 /* data available on socket, or listen socket received a connect */
129 static void ceph_data_ready(struct sock *sk, int count_unused)
131 struct ceph_connection *con =
132 (struct ceph_connection *)sk->sk_user_data;
133 if (sk->sk_state != TCP_CLOSE_WAIT) {
134 dout("ceph_data_ready on %p state = %lu, queueing work\n",
140 /* socket has buffer space for writing */
141 static void ceph_write_space(struct sock *sk)
143 struct ceph_connection *con =
144 (struct ceph_connection *)sk->sk_user_data;
146 /* only queue to workqueue if there is data we want to write. */
147 if (test_bit(WRITE_PENDING, &con->state)) {
148 dout("ceph_write_space %p queueing write work\n", con);
151 dout("ceph_write_space %p nothing to write\n", con);
154 /* since we have our own write_space, clear the SOCK_NOSPACE flag */
155 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
158 /* socket's state has changed */
159 static void ceph_state_change(struct sock *sk)
161 struct ceph_connection *con =
162 (struct ceph_connection *)sk->sk_user_data;
164 dout("ceph_state_change %p state = %lu sk_state = %u\n",
165 con, con->state, sk->sk_state);
167 if (test_bit(CLOSED, &con->state))
170 switch (sk->sk_state) {
172 dout("ceph_state_change TCP_CLOSE\n");
174 dout("ceph_state_change TCP_CLOSE_WAIT\n");
175 if (test_and_set_bit(SOCK_CLOSED, &con->state) == 0) {
176 if (test_bit(CONNECTING, &con->state))
177 con->error_msg = "connection failed";
179 con->error_msg = "socket closed";
183 case TCP_ESTABLISHED:
184 dout("ceph_state_change TCP_ESTABLISHED\n");
191 * set up socket callbacks
193 static void set_sock_callbacks(struct socket *sock,
194 struct ceph_connection *con)
196 struct sock *sk = sock->sk;
197 sk->sk_user_data = (void *)con;
198 sk->sk_data_ready = ceph_data_ready;
199 sk->sk_write_space = ceph_write_space;
200 sk->sk_state_change = ceph_state_change;
209 * initiate connection to a remote socket.
211 static struct socket *ceph_tcp_connect(struct ceph_connection *con)
213 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
218 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
223 sock->sk->sk_allocation = GFP_NOFS;
225 #ifdef CONFIG_LOCKDEP
226 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
229 set_sock_callbacks(sock, con);
231 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
233 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
235 if (ret == -EINPROGRESS) {
236 dout("connect %s EINPROGRESS sk_state = %u\n",
237 ceph_pr_addr(&con->peer_addr.in_addr),
242 pr_err("connect %s error %d\n",
243 ceph_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 };
260 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
267 * write something. @more is true if caller will be sending more data
270 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
271 size_t kvlen, size_t len, int more)
273 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
277 msg.msg_flags |= MSG_MORE;
279 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
281 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
287 static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
288 int offset, size_t size, bool more)
290 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
293 ret = kernel_sendpage(sock, page, offset, size, flags);
300 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
301 int offset, size_t size, bool more)
306 /* sendpage cannot properly handle pages with page_count == 0,
307 * we need to fallback to sendmsg if that's the case */
308 if (page_count(page) >= 1)
309 return __ceph_tcp_sendpage(sock, page, offset, size, more);
311 iov.iov_base = kmap(page) + offset;
313 ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more);
320 * Shutdown/close the socket for the given connection.
322 static int con_close_socket(struct ceph_connection *con)
326 dout("con_close_socket on %p sock %p\n", con, con->sock);
329 set_bit(SOCK_CLOSED, &con->state);
330 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
331 sock_release(con->sock);
333 clear_bit(SOCK_CLOSED, &con->state);
338 * Reset a connection. Discard all incoming and outgoing messages
339 * and clear *_seq state.
341 static void ceph_msg_remove(struct ceph_msg *msg)
343 list_del_init(&msg->list_head);
346 static void ceph_msg_remove_list(struct list_head *head)
348 while (!list_empty(head)) {
349 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
351 ceph_msg_remove(msg);
355 static void reset_connection(struct ceph_connection *con)
357 /* reset connection, out_queue, msg_ and connect_seq */
358 /* discard existing out_queue and msg_seq */
359 ceph_msg_remove_list(&con->out_queue);
360 ceph_msg_remove_list(&con->out_sent);
363 ceph_msg_put(con->in_msg);
367 con->connect_seq = 0;
370 ceph_msg_put(con->out_msg);
374 con->in_seq_acked = 0;
378 * mark a peer down. drop any open connections.
380 void ceph_con_close(struct ceph_connection *con)
382 dout("con_close %p peer %s\n", con,
383 ceph_pr_addr(&con->peer_addr.in_addr));
384 set_bit(CLOSED, &con->state); /* in case there's queued work */
385 clear_bit(STANDBY, &con->state); /* avoid connect_seq bump */
386 clear_bit(LOSSYTX, &con->state); /* so we retry next connect */
387 clear_bit(KEEPALIVE_PENDING, &con->state);
388 clear_bit(WRITE_PENDING, &con->state);
389 mutex_lock(&con->mutex);
390 reset_connection(con);
391 con->peer_global_seq = 0;
392 cancel_delayed_work(&con->work);
393 mutex_unlock(&con->mutex);
396 EXPORT_SYMBOL(ceph_con_close);
399 * Reopen a closed connection, with a new peer address.
401 void ceph_con_open(struct ceph_connection *con, struct ceph_entity_addr *addr)
403 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
404 set_bit(OPENING, &con->state);
405 clear_bit(CLOSED, &con->state);
406 memcpy(&con->peer_addr, addr, sizeof(*addr));
407 con->delay = 0; /* reset backoff memory */
410 EXPORT_SYMBOL(ceph_con_open);
413 * return true if this connection ever successfully opened
415 bool ceph_con_opened(struct ceph_connection *con)
417 return con->connect_seq > 0;
423 struct ceph_connection *ceph_con_get(struct ceph_connection *con)
425 dout("con_get %p nref = %d -> %d\n", con,
426 atomic_read(&con->nref), atomic_read(&con->nref) + 1);
427 if (atomic_inc_not_zero(&con->nref))
432 void ceph_con_put(struct ceph_connection *con)
434 dout("con_put %p nref = %d -> %d\n", con,
435 atomic_read(&con->nref), atomic_read(&con->nref) - 1);
436 BUG_ON(atomic_read(&con->nref) == 0);
437 if (atomic_dec_and_test(&con->nref)) {
444 * initialize a new connection.
446 void ceph_con_init(struct ceph_messenger *msgr, struct ceph_connection *con)
448 dout("con_init %p\n", con);
449 memset(con, 0, sizeof(*con));
450 atomic_set(&con->nref, 1);
452 mutex_init(&con->mutex);
453 INIT_LIST_HEAD(&con->out_queue);
454 INIT_LIST_HEAD(&con->out_sent);
455 INIT_DELAYED_WORK(&con->work, con_work);
457 EXPORT_SYMBOL(ceph_con_init);
461 * We maintain a global counter to order connection attempts. Get
462 * a unique seq greater than @gt.
464 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
468 spin_lock(&msgr->global_seq_lock);
469 if (msgr->global_seq < gt)
470 msgr->global_seq = gt;
471 ret = ++msgr->global_seq;
472 spin_unlock(&msgr->global_seq_lock);
478 * Prepare footer for currently outgoing message, and finish things
479 * off. Assumes out_kvec* are already valid.. we just add on to the end.
481 static void prepare_write_message_footer(struct ceph_connection *con, int v)
483 struct ceph_msg *m = con->out_msg;
485 dout("prepare_write_message_footer %p\n", con);
486 con->out_kvec_is_msg = true;
487 con->out_kvec[v].iov_base = &m->footer;
488 con->out_kvec[v].iov_len = sizeof(m->footer);
489 con->out_kvec_bytes += sizeof(m->footer);
490 con->out_kvec_left++;
491 con->out_more = m->more_to_follow;
492 con->out_msg_done = true;
496 * Prepare headers for the next outgoing message.
498 static void prepare_write_message(struct ceph_connection *con)
503 con->out_kvec_bytes = 0;
504 con->out_kvec_is_msg = true;
505 con->out_msg_done = false;
507 /* Sneak an ack in there first? If we can get it into the same
508 * TCP packet that's a good thing. */
509 if (con->in_seq > con->in_seq_acked) {
510 con->in_seq_acked = con->in_seq;
511 con->out_kvec[v].iov_base = &tag_ack;
512 con->out_kvec[v++].iov_len = 1;
513 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
514 con->out_kvec[v].iov_base = &con->out_temp_ack;
515 con->out_kvec[v++].iov_len = sizeof(con->out_temp_ack);
516 con->out_kvec_bytes = 1 + sizeof(con->out_temp_ack);
519 m = list_first_entry(&con->out_queue,
520 struct ceph_msg, list_head);
523 /* put message on sent list */
525 list_move_tail(&m->list_head, &con->out_sent);
528 * only assign outgoing seq # if we haven't sent this message
529 * yet. if it is requeued, resend with it's original seq.
531 if (m->needs_out_seq) {
532 m->hdr.seq = cpu_to_le64(++con->out_seq);
533 m->needs_out_seq = false;
536 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n",
537 m, con->out_seq, le16_to_cpu(m->hdr.type),
538 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
539 le32_to_cpu(m->hdr.data_len),
541 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
543 /* tag + hdr + front + middle */
544 con->out_kvec[v].iov_base = &tag_msg;
545 con->out_kvec[v++].iov_len = 1;
546 con->out_kvec[v].iov_base = &m->hdr;
547 con->out_kvec[v++].iov_len = sizeof(m->hdr);
548 con->out_kvec[v++] = m->front;
550 con->out_kvec[v++] = m->middle->vec;
551 con->out_kvec_left = v;
552 con->out_kvec_bytes += 1 + sizeof(m->hdr) + m->front.iov_len +
553 (m->middle ? m->middle->vec.iov_len : 0);
554 con->out_kvec_cur = con->out_kvec;
556 /* fill in crc (except data pages), footer */
557 con->out_msg->hdr.crc =
558 cpu_to_le32(crc32c(0, (void *)&m->hdr,
559 sizeof(m->hdr) - sizeof(m->hdr.crc)));
560 con->out_msg->footer.flags = CEPH_MSG_FOOTER_COMPLETE;
561 con->out_msg->footer.front_crc =
562 cpu_to_le32(crc32c(0, m->front.iov_base, m->front.iov_len));
564 con->out_msg->footer.middle_crc =
565 cpu_to_le32(crc32c(0, m->middle->vec.iov_base,
566 m->middle->vec.iov_len));
568 con->out_msg->footer.middle_crc = 0;
569 con->out_msg->footer.data_crc = 0;
570 dout("prepare_write_message front_crc %u data_crc %u\n",
571 le32_to_cpu(con->out_msg->footer.front_crc),
572 le32_to_cpu(con->out_msg->footer.middle_crc));
574 /* is there a data payload? */
575 if (le32_to_cpu(m->hdr.data_len) > 0) {
576 /* initialize page iterator */
577 con->out_msg_pos.page = 0;
579 con->out_msg_pos.page_pos = m->page_alignment;
581 con->out_msg_pos.page_pos = 0;
582 con->out_msg_pos.data_pos = 0;
583 con->out_msg_pos.did_page_crc = 0;
584 con->out_more = 1; /* data + footer will follow */
586 /* no, queue up footer too and be done */
587 prepare_write_message_footer(con, v);
590 set_bit(WRITE_PENDING, &con->state);
596 static void prepare_write_ack(struct ceph_connection *con)
598 dout("prepare_write_ack %p %llu -> %llu\n", con,
599 con->in_seq_acked, con->in_seq);
600 con->in_seq_acked = con->in_seq;
602 con->out_kvec[0].iov_base = &tag_ack;
603 con->out_kvec[0].iov_len = 1;
604 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
605 con->out_kvec[1].iov_base = &con->out_temp_ack;
606 con->out_kvec[1].iov_len = sizeof(con->out_temp_ack);
607 con->out_kvec_left = 2;
608 con->out_kvec_bytes = 1 + sizeof(con->out_temp_ack);
609 con->out_kvec_cur = con->out_kvec;
610 con->out_more = 1; /* more will follow.. eventually.. */
611 set_bit(WRITE_PENDING, &con->state);
615 * Prepare to write keepalive byte.
617 static void prepare_write_keepalive(struct ceph_connection *con)
619 dout("prepare_write_keepalive %p\n", con);
620 con->out_kvec[0].iov_base = &tag_keepalive;
621 con->out_kvec[0].iov_len = 1;
622 con->out_kvec_left = 1;
623 con->out_kvec_bytes = 1;
624 con->out_kvec_cur = con->out_kvec;
625 set_bit(WRITE_PENDING, &con->state);
629 * Connection negotiation.
632 static int prepare_connect_authorizer(struct ceph_connection *con)
636 int auth_protocol = 0;
638 mutex_unlock(&con->mutex);
639 if (con->ops->get_authorizer)
640 con->ops->get_authorizer(con, &auth_buf, &auth_len,
641 &auth_protocol, &con->auth_reply_buf,
642 &con->auth_reply_buf_len,
644 mutex_lock(&con->mutex);
646 if (test_bit(CLOSED, &con->state) ||
647 test_bit(OPENING, &con->state))
650 con->out_connect.authorizer_protocol = cpu_to_le32(auth_protocol);
651 con->out_connect.authorizer_len = cpu_to_le32(auth_len);
654 con->out_kvec[con->out_kvec_left].iov_base = auth_buf;
655 con->out_kvec[con->out_kvec_left].iov_len = auth_len;
656 con->out_kvec_left++;
657 con->out_kvec_bytes += auth_len;
663 * We connected to a peer and are saying hello.
665 static void prepare_write_banner(struct ceph_messenger *msgr,
666 struct ceph_connection *con)
668 int len = strlen(CEPH_BANNER);
670 con->out_kvec[0].iov_base = CEPH_BANNER;
671 con->out_kvec[0].iov_len = len;
672 con->out_kvec[1].iov_base = &msgr->my_enc_addr;
673 con->out_kvec[1].iov_len = sizeof(msgr->my_enc_addr);
674 con->out_kvec_left = 2;
675 con->out_kvec_bytes = len + sizeof(msgr->my_enc_addr);
676 con->out_kvec_cur = con->out_kvec;
678 set_bit(WRITE_PENDING, &con->state);
681 static int prepare_write_connect(struct ceph_messenger *msgr,
682 struct ceph_connection *con,
685 unsigned global_seq = get_global_seq(con->msgr, 0);
688 switch (con->peer_name.type) {
689 case CEPH_ENTITY_TYPE_MON:
690 proto = CEPH_MONC_PROTOCOL;
692 case CEPH_ENTITY_TYPE_OSD:
693 proto = CEPH_OSDC_PROTOCOL;
695 case CEPH_ENTITY_TYPE_MDS:
696 proto = CEPH_MDSC_PROTOCOL;
702 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
703 con->connect_seq, global_seq, proto);
705 con->out_connect.features = cpu_to_le64(msgr->supported_features);
706 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
707 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
708 con->out_connect.global_seq = cpu_to_le32(global_seq);
709 con->out_connect.protocol_version = cpu_to_le32(proto);
710 con->out_connect.flags = 0;
713 con->out_kvec_left = 0;
714 con->out_kvec_bytes = 0;
716 con->out_kvec[con->out_kvec_left].iov_base = &con->out_connect;
717 con->out_kvec[con->out_kvec_left].iov_len = sizeof(con->out_connect);
718 con->out_kvec_left++;
719 con->out_kvec_bytes += sizeof(con->out_connect);
720 con->out_kvec_cur = con->out_kvec;
722 set_bit(WRITE_PENDING, &con->state);
724 return prepare_connect_authorizer(con);
729 * write as much of pending kvecs to the socket as we can.
731 * 0 -> socket full, but more to do
734 static int write_partial_kvec(struct ceph_connection *con)
738 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
739 while (con->out_kvec_bytes > 0) {
740 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
741 con->out_kvec_left, con->out_kvec_bytes,
745 con->out_kvec_bytes -= ret;
746 if (con->out_kvec_bytes == 0)
749 if (ret >= con->out_kvec_cur->iov_len) {
750 ret -= con->out_kvec_cur->iov_len;
752 con->out_kvec_left--;
754 con->out_kvec_cur->iov_len -= ret;
755 con->out_kvec_cur->iov_base += ret;
761 con->out_kvec_left = 0;
762 con->out_kvec_is_msg = false;
765 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
766 con->out_kvec_bytes, con->out_kvec_left, ret);
767 return ret; /* done! */
771 static void init_bio_iter(struct bio *bio, struct bio **iter, int *seg)
782 static void iter_bio_next(struct bio **bio_iter, int *seg)
784 if (*bio_iter == NULL)
787 BUG_ON(*seg >= (*bio_iter)->bi_vcnt);
790 if (*seg == (*bio_iter)->bi_vcnt)
791 init_bio_iter((*bio_iter)->bi_next, bio_iter, seg);
796 * Write as much message data payload as we can. If we finish, queue
798 * 1 -> done, footer is now queued in out_kvec[].
799 * 0 -> socket full, but more to do
802 static int write_partial_msg_pages(struct ceph_connection *con)
804 struct ceph_msg *msg = con->out_msg;
805 unsigned data_len = le32_to_cpu(msg->hdr.data_len);
807 int crc = con->msgr->nocrc;
811 size_t trail_len = (msg->trail ? msg->trail->length : 0);
813 dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
814 con, con->out_msg, con->out_msg_pos.page, con->out_msg->nr_pages,
815 con->out_msg_pos.page_pos);
818 if (msg->bio && !msg->bio_iter)
819 init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg);
822 while (data_len > con->out_msg_pos.data_pos) {
823 struct page *page = NULL;
825 int max_write = PAGE_SIZE;
828 total_max_write = data_len - trail_len -
829 con->out_msg_pos.data_pos;
832 * if we are calculating the data crc (the default), we need
833 * to map the page. if our pages[] has been revoked, use the
837 /* have we reached the trail part of the data? */
838 if (con->out_msg_pos.data_pos >= data_len - trail_len) {
841 total_max_write = data_len - con->out_msg_pos.data_pos;
843 page = list_first_entry(&msg->trail->head,
847 max_write = PAGE_SIZE;
848 } else if (msg->pages) {
849 page = msg->pages[con->out_msg_pos.page];
852 } else if (msg->pagelist) {
853 page = list_first_entry(&msg->pagelist->head,
858 } else if (msg->bio) {
861 bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg);
863 page_shift = bv->bv_offset;
865 kaddr = kmap(page) + page_shift;
866 max_write = bv->bv_len;
869 page = con->msgr->zero_page;
871 kaddr = page_address(con->msgr->zero_page);
873 len = min_t(int, max_write - con->out_msg_pos.page_pos,
876 if (crc && !con->out_msg_pos.did_page_crc) {
877 void *base = kaddr + con->out_msg_pos.page_pos;
878 u32 tmpcrc = le32_to_cpu(con->out_msg->footer.data_crc);
880 BUG_ON(kaddr == NULL);
881 con->out_msg->footer.data_crc =
882 cpu_to_le32(crc32c(tmpcrc, base, len));
883 con->out_msg_pos.did_page_crc = 1;
885 ret = ceph_tcp_sendpage(con->sock, page,
886 con->out_msg_pos.page_pos + page_shift,
890 (msg->pages || msg->pagelist || msg->bio || in_trail))
896 con->out_msg_pos.data_pos += ret;
897 con->out_msg_pos.page_pos += ret;
899 con->out_msg_pos.page_pos = 0;
900 con->out_msg_pos.page++;
901 con->out_msg_pos.did_page_crc = 0;
903 list_move_tail(&page->lru,
905 else if (msg->pagelist)
906 list_move_tail(&page->lru,
907 &msg->pagelist->head);
910 iter_bio_next(&msg->bio_iter, &msg->bio_seg);
915 dout("write_partial_msg_pages %p msg %p done\n", con, msg);
917 /* prepare and queue up footer, too */
919 con->out_msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
920 con->out_kvec_bytes = 0;
921 con->out_kvec_left = 0;
922 con->out_kvec_cur = con->out_kvec;
923 prepare_write_message_footer(con, 0);
932 static int write_partial_skip(struct ceph_connection *con)
936 while (con->out_skip > 0) {
938 .iov_base = page_address(con->msgr->zero_page),
939 .iov_len = min(con->out_skip, (int)PAGE_CACHE_SIZE)
942 ret = ceph_tcp_sendmsg(con->sock, &iov, 1, iov.iov_len, 1);
945 con->out_skip -= ret;
953 * Prepare to read connection handshake, or an ack.
955 static void prepare_read_banner(struct ceph_connection *con)
957 dout("prepare_read_banner %p\n", con);
958 con->in_base_pos = 0;
961 static void prepare_read_connect(struct ceph_connection *con)
963 dout("prepare_read_connect %p\n", con);
964 con->in_base_pos = 0;
967 static void prepare_read_ack(struct ceph_connection *con)
969 dout("prepare_read_ack %p\n", con);
970 con->in_base_pos = 0;
973 static void prepare_read_tag(struct ceph_connection *con)
975 dout("prepare_read_tag %p\n", con);
976 con->in_base_pos = 0;
977 con->in_tag = CEPH_MSGR_TAG_READY;
981 * Prepare to read a message.
983 static int prepare_read_message(struct ceph_connection *con)
985 dout("prepare_read_message %p\n", con);
986 BUG_ON(con->in_msg != NULL);
987 con->in_base_pos = 0;
988 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
993 static int read_partial(struct ceph_connection *con,
994 int *to, int size, void *object)
997 while (con->in_base_pos < *to) {
998 int left = *to - con->in_base_pos;
999 int have = size - left;
1000 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1003 con->in_base_pos += ret;
1010 * Read all or part of the connect-side handshake on a new connection
1012 static int read_partial_banner(struct ceph_connection *con)
1016 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1019 ret = read_partial(con, &to, strlen(CEPH_BANNER), con->in_banner);
1022 ret = read_partial(con, &to, sizeof(con->actual_peer_addr),
1023 &con->actual_peer_addr);
1026 ret = read_partial(con, &to, sizeof(con->peer_addr_for_me),
1027 &con->peer_addr_for_me);
1034 static int read_partial_connect(struct ceph_connection *con)
1038 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1040 ret = read_partial(con, &to, sizeof(con->in_reply), &con->in_reply);
1043 ret = read_partial(con, &to, le32_to_cpu(con->in_reply.authorizer_len),
1044 con->auth_reply_buf);
1048 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1049 con, (int)con->in_reply.tag,
1050 le32_to_cpu(con->in_reply.connect_seq),
1051 le32_to_cpu(con->in_reply.global_seq));
1058 * Verify the hello banner looks okay.
1060 static int verify_hello(struct ceph_connection *con)
1062 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1063 pr_err("connect to %s got bad banner\n",
1064 ceph_pr_addr(&con->peer_addr.in_addr));
1065 con->error_msg = "protocol error, bad banner";
1071 static bool addr_is_blank(struct sockaddr_storage *ss)
1073 switch (ss->ss_family) {
1075 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1078 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1079 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1080 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1081 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1086 static int addr_port(struct sockaddr_storage *ss)
1088 switch (ss->ss_family) {
1090 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1092 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1097 static void addr_set_port(struct sockaddr_storage *ss, int p)
1099 switch (ss->ss_family) {
1101 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1104 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1110 * Unlike other *_pton function semantics, zero indicates success.
1112 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1113 char delim, const char **ipend)
1115 struct sockaddr_in *in4 = (void *)ss;
1116 struct sockaddr_in6 *in6 = (void *)ss;
1118 memset(ss, 0, sizeof(*ss));
1120 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1121 ss->ss_family = AF_INET;
1125 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1126 ss->ss_family = AF_INET6;
1134 * Extract hostname string and resolve using kernel DNS facility.
1136 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1137 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1138 struct sockaddr_storage *ss, char delim, const char **ipend)
1140 const char *end, *delim_p;
1141 char *colon_p, *ip_addr = NULL;
1145 * The end of the hostname occurs immediately preceding the delimiter or
1146 * the port marker (':') where the delimiter takes precedence.
1148 delim_p = memchr(name, delim, namelen);
1149 colon_p = memchr(name, ':', namelen);
1151 if (delim_p && colon_p)
1152 end = delim_p < colon_p ? delim_p : colon_p;
1153 else if (!delim_p && colon_p)
1157 if (!end) /* case: hostname:/ */
1158 end = name + namelen;
1164 /* do dns_resolve upcall */
1165 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1167 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1175 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1176 ret, ret ? "failed" : ceph_pr_addr(ss));
1181 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1182 struct sockaddr_storage *ss, char delim, const char **ipend)
1189 * Parse a server name (IP or hostname). If a valid IP address is not found
1190 * then try to extract a hostname to resolve using userspace DNS upcall.
1192 static int ceph_parse_server_name(const char *name, size_t namelen,
1193 struct sockaddr_storage *ss, char delim, const char **ipend)
1197 ret = ceph_pton(name, namelen, ss, delim, ipend);
1199 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1205 * Parse an ip[:port] list into an addr array. Use the default
1206 * monitor port if a port isn't specified.
1208 int ceph_parse_ips(const char *c, const char *end,
1209 struct ceph_entity_addr *addr,
1210 int max_count, int *count)
1212 int i, ret = -EINVAL;
1215 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1216 for (i = 0; i < max_count; i++) {
1218 struct sockaddr_storage *ss = &addr[i].in_addr;
1227 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1236 dout("missing matching ']'\n");
1243 if (p < end && *p == ':') {
1246 while (p < end && *p >= '0' && *p <= '9') {
1247 port = (port * 10) + (*p - '0');
1250 if (port > 65535 || port == 0)
1253 port = CEPH_MON_PORT;
1256 addr_set_port(ss, port);
1258 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1275 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1278 EXPORT_SYMBOL(ceph_parse_ips);
1280 static int process_banner(struct ceph_connection *con)
1282 dout("process_banner on %p\n", con);
1284 if (verify_hello(con) < 0)
1287 ceph_decode_addr(&con->actual_peer_addr);
1288 ceph_decode_addr(&con->peer_addr_for_me);
1291 * Make sure the other end is who we wanted. note that the other
1292 * end may not yet know their ip address, so if it's 0.0.0.0, give
1293 * them the benefit of the doubt.
1295 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1296 sizeof(con->peer_addr)) != 0 &&
1297 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1298 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1299 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1300 ceph_pr_addr(&con->peer_addr.in_addr),
1301 (int)le32_to_cpu(con->peer_addr.nonce),
1302 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1303 (int)le32_to_cpu(con->actual_peer_addr.nonce));
1304 con->error_msg = "wrong peer at address";
1309 * did we learn our address?
1311 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1312 int port = addr_port(&con->msgr->inst.addr.in_addr);
1314 memcpy(&con->msgr->inst.addr.in_addr,
1315 &con->peer_addr_for_me.in_addr,
1316 sizeof(con->peer_addr_for_me.in_addr));
1317 addr_set_port(&con->msgr->inst.addr.in_addr, port);
1318 encode_my_addr(con->msgr);
1319 dout("process_banner learned my addr is %s\n",
1320 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
1323 set_bit(NEGOTIATING, &con->state);
1324 prepare_read_connect(con);
1328 static void fail_protocol(struct ceph_connection *con)
1330 reset_connection(con);
1331 set_bit(CLOSED, &con->state); /* in case there's queued work */
1333 mutex_unlock(&con->mutex);
1334 if (con->ops->bad_proto)
1335 con->ops->bad_proto(con);
1336 mutex_lock(&con->mutex);
1339 static int process_connect(struct ceph_connection *con)
1341 u64 sup_feat = con->msgr->supported_features;
1342 u64 req_feat = con->msgr->required_features;
1343 u64 server_feat = le64_to_cpu(con->in_reply.features);
1346 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1348 switch (con->in_reply.tag) {
1349 case CEPH_MSGR_TAG_FEATURES:
1350 pr_err("%s%lld %s feature set mismatch,"
1351 " my %llx < server's %llx, missing %llx\n",
1352 ENTITY_NAME(con->peer_name),
1353 ceph_pr_addr(&con->peer_addr.in_addr),
1354 sup_feat, server_feat, server_feat & ~sup_feat);
1355 con->error_msg = "missing required protocol features";
1359 case CEPH_MSGR_TAG_BADPROTOVER:
1360 pr_err("%s%lld %s protocol version mismatch,"
1361 " my %d != server's %d\n",
1362 ENTITY_NAME(con->peer_name),
1363 ceph_pr_addr(&con->peer_addr.in_addr),
1364 le32_to_cpu(con->out_connect.protocol_version),
1365 le32_to_cpu(con->in_reply.protocol_version));
1366 con->error_msg = "protocol version mismatch";
1370 case CEPH_MSGR_TAG_BADAUTHORIZER:
1372 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1374 if (con->auth_retry == 2) {
1375 con->error_msg = "connect authorization failure";
1378 con->auth_retry = 1;
1379 ret = prepare_write_connect(con->msgr, con, 0);
1382 prepare_read_connect(con);
1385 case CEPH_MSGR_TAG_RESETSESSION:
1387 * If we connected with a large connect_seq but the peer
1388 * has no record of a session with us (no connection, or
1389 * connect_seq == 0), they will send RESETSESION to indicate
1390 * that they must have reset their session, and may have
1393 dout("process_connect got RESET peer seq %u\n",
1394 le32_to_cpu(con->in_connect.connect_seq));
1395 pr_err("%s%lld %s connection reset\n",
1396 ENTITY_NAME(con->peer_name),
1397 ceph_pr_addr(&con->peer_addr.in_addr));
1398 reset_connection(con);
1399 prepare_write_connect(con->msgr, con, 0);
1400 prepare_read_connect(con);
1402 /* Tell ceph about it. */
1403 mutex_unlock(&con->mutex);
1404 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
1405 if (con->ops->peer_reset)
1406 con->ops->peer_reset(con);
1407 mutex_lock(&con->mutex);
1408 if (test_bit(CLOSED, &con->state) ||
1409 test_bit(OPENING, &con->state))
1413 case CEPH_MSGR_TAG_RETRY_SESSION:
1415 * If we sent a smaller connect_seq than the peer has, try
1416 * again with a larger value.
1418 dout("process_connect got RETRY my seq = %u, peer_seq = %u\n",
1419 le32_to_cpu(con->out_connect.connect_seq),
1420 le32_to_cpu(con->in_connect.connect_seq));
1421 con->connect_seq = le32_to_cpu(con->in_connect.connect_seq);
1422 prepare_write_connect(con->msgr, con, 0);
1423 prepare_read_connect(con);
1426 case CEPH_MSGR_TAG_RETRY_GLOBAL:
1428 * If we sent a smaller global_seq than the peer has, try
1429 * again with a larger value.
1431 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1432 con->peer_global_seq,
1433 le32_to_cpu(con->in_connect.global_seq));
1434 get_global_seq(con->msgr,
1435 le32_to_cpu(con->in_connect.global_seq));
1436 prepare_write_connect(con->msgr, con, 0);
1437 prepare_read_connect(con);
1440 case CEPH_MSGR_TAG_READY:
1441 if (req_feat & ~server_feat) {
1442 pr_err("%s%lld %s protocol feature mismatch,"
1443 " my required %llx > server's %llx, need %llx\n",
1444 ENTITY_NAME(con->peer_name),
1445 ceph_pr_addr(&con->peer_addr.in_addr),
1446 req_feat, server_feat, req_feat & ~server_feat);
1447 con->error_msg = "missing required protocol features";
1451 clear_bit(CONNECTING, &con->state);
1452 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
1454 con->peer_features = server_feat;
1455 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1456 con->peer_global_seq,
1457 le32_to_cpu(con->in_reply.connect_seq),
1459 WARN_ON(con->connect_seq !=
1460 le32_to_cpu(con->in_reply.connect_seq));
1462 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
1463 set_bit(LOSSYTX, &con->state);
1465 prepare_read_tag(con);
1468 case CEPH_MSGR_TAG_WAIT:
1470 * If there is a connection race (we are opening
1471 * connections to each other), one of us may just have
1472 * to WAIT. This shouldn't happen if we are the
1475 pr_err("process_connect got WAIT as client\n");
1476 con->error_msg = "protocol error, got WAIT as client";
1480 pr_err("connect protocol error, will retry\n");
1481 con->error_msg = "protocol error, garbage tag during connect";
1489 * read (part of) an ack
1491 static int read_partial_ack(struct ceph_connection *con)
1495 return read_partial(con, &to, sizeof(con->in_temp_ack),
1501 * We can finally discard anything that's been acked.
1503 static void process_ack(struct ceph_connection *con)
1506 u64 ack = le64_to_cpu(con->in_temp_ack);
1509 while (!list_empty(&con->out_sent)) {
1510 m = list_first_entry(&con->out_sent, struct ceph_msg,
1512 seq = le64_to_cpu(m->hdr.seq);
1515 dout("got ack for seq %llu type %d at %p\n", seq,
1516 le16_to_cpu(m->hdr.type), m);
1517 m->ack_stamp = jiffies;
1520 prepare_read_tag(con);
1526 static int read_partial_message_section(struct ceph_connection *con,
1527 struct kvec *section,
1528 unsigned int sec_len, u32 *crc)
1534 while (section->iov_len < sec_len) {
1535 BUG_ON(section->iov_base == NULL);
1536 left = sec_len - section->iov_len;
1537 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
1538 section->iov_len, left);
1541 section->iov_len += ret;
1542 if (section->iov_len == sec_len)
1543 *crc = crc32c(0, section->iov_base,
1550 static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con,
1551 struct ceph_msg_header *hdr,
1555 static int read_partial_message_pages(struct ceph_connection *con,
1556 struct page **pages,
1557 unsigned data_len, int datacrc)
1563 left = min((int)(data_len - con->in_msg_pos.data_pos),
1564 (int)(PAGE_SIZE - con->in_msg_pos.page_pos));
1566 BUG_ON(pages == NULL);
1567 p = kmap(pages[con->in_msg_pos.page]);
1568 ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos,
1570 if (ret > 0 && datacrc)
1572 crc32c(con->in_data_crc,
1573 p + con->in_msg_pos.page_pos, ret);
1574 kunmap(pages[con->in_msg_pos.page]);
1577 con->in_msg_pos.data_pos += ret;
1578 con->in_msg_pos.page_pos += ret;
1579 if (con->in_msg_pos.page_pos == PAGE_SIZE) {
1580 con->in_msg_pos.page_pos = 0;
1581 con->in_msg_pos.page++;
1588 static int read_partial_message_bio(struct ceph_connection *con,
1589 struct bio **bio_iter, int *bio_seg,
1590 unsigned data_len, int datacrc)
1592 struct bio_vec *bv = bio_iovec_idx(*bio_iter, *bio_seg);
1599 left = min((int)(data_len - con->in_msg_pos.data_pos),
1600 (int)(bv->bv_len - con->in_msg_pos.page_pos));
1602 p = kmap(bv->bv_page) + bv->bv_offset;
1604 ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos,
1606 if (ret > 0 && datacrc)
1608 crc32c(con->in_data_crc,
1609 p + con->in_msg_pos.page_pos, ret);
1610 kunmap(bv->bv_page);
1613 con->in_msg_pos.data_pos += ret;
1614 con->in_msg_pos.page_pos += ret;
1615 if (con->in_msg_pos.page_pos == bv->bv_len) {
1616 con->in_msg_pos.page_pos = 0;
1617 iter_bio_next(bio_iter, bio_seg);
1625 * read (part of) a message.
1627 static int read_partial_message(struct ceph_connection *con)
1629 struct ceph_msg *m = con->in_msg;
1632 unsigned front_len, middle_len, data_len;
1633 int datacrc = con->msgr->nocrc;
1637 dout("read_partial_message con %p msg %p\n", con, m);
1640 while (con->in_base_pos < sizeof(con->in_hdr)) {
1641 left = sizeof(con->in_hdr) - con->in_base_pos;
1642 ret = ceph_tcp_recvmsg(con->sock,
1643 (char *)&con->in_hdr + con->in_base_pos,
1647 con->in_base_pos += ret;
1648 if (con->in_base_pos == sizeof(con->in_hdr)) {
1649 u32 crc = crc32c(0, (void *)&con->in_hdr,
1650 sizeof(con->in_hdr) - sizeof(con->in_hdr.crc));
1651 if (crc != le32_to_cpu(con->in_hdr.crc)) {
1652 pr_err("read_partial_message bad hdr "
1653 " crc %u != expected %u\n",
1654 crc, con->in_hdr.crc);
1659 front_len = le32_to_cpu(con->in_hdr.front_len);
1660 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
1662 middle_len = le32_to_cpu(con->in_hdr.middle_len);
1663 if (middle_len > CEPH_MSG_MAX_DATA_LEN)
1665 data_len = le32_to_cpu(con->in_hdr.data_len);
1666 if (data_len > CEPH_MSG_MAX_DATA_LEN)
1670 seq = le64_to_cpu(con->in_hdr.seq);
1671 if ((s64)seq - (s64)con->in_seq < 1) {
1672 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1673 ENTITY_NAME(con->peer_name),
1674 ceph_pr_addr(&con->peer_addr.in_addr),
1675 seq, con->in_seq + 1);
1676 con->in_base_pos = -front_len - middle_len - data_len -
1678 con->in_tag = CEPH_MSGR_TAG_READY;
1680 } else if ((s64)seq - (s64)con->in_seq > 1) {
1681 pr_err("read_partial_message bad seq %lld expected %lld\n",
1682 seq, con->in_seq + 1);
1683 con->error_msg = "bad message sequence # for incoming message";
1687 /* allocate message? */
1689 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
1690 con->in_hdr.front_len, con->in_hdr.data_len);
1692 con->in_msg = ceph_alloc_msg(con, &con->in_hdr, &skip);
1694 /* skip this message */
1695 dout("alloc_msg said skip message\n");
1696 BUG_ON(con->in_msg);
1697 con->in_base_pos = -front_len - middle_len - data_len -
1699 con->in_tag = CEPH_MSGR_TAG_READY;
1705 "error allocating memory for incoming message";
1709 m->front.iov_len = 0; /* haven't read it yet */
1711 m->middle->vec.iov_len = 0;
1713 con->in_msg_pos.page = 0;
1715 con->in_msg_pos.page_pos = m->page_alignment;
1717 con->in_msg_pos.page_pos = 0;
1718 con->in_msg_pos.data_pos = 0;
1722 ret = read_partial_message_section(con, &m->front, front_len,
1723 &con->in_front_crc);
1729 ret = read_partial_message_section(con, &m->middle->vec,
1731 &con->in_middle_crc);
1736 if (m->bio && !m->bio_iter)
1737 init_bio_iter(m->bio, &m->bio_iter, &m->bio_seg);
1741 while (con->in_msg_pos.data_pos < data_len) {
1743 ret = read_partial_message_pages(con, m->pages,
1748 } else if (m->bio) {
1750 ret = read_partial_message_bio(con,
1751 &m->bio_iter, &m->bio_seg,
1762 to = sizeof(m->hdr) + sizeof(m->footer);
1763 while (con->in_base_pos < to) {
1764 left = to - con->in_base_pos;
1765 ret = ceph_tcp_recvmsg(con->sock, (char *)&m->footer +
1766 (con->in_base_pos - sizeof(m->hdr)),
1770 con->in_base_pos += ret;
1772 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1773 m, front_len, m->footer.front_crc, middle_len,
1774 m->footer.middle_crc, data_len, m->footer.data_crc);
1777 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
1778 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1779 m, con->in_front_crc, m->footer.front_crc);
1782 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
1783 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1784 m, con->in_middle_crc, m->footer.middle_crc);
1788 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
1789 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
1790 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
1791 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
1795 return 1; /* done! */
1799 * Process message. This happens in the worker thread. The callback should
1800 * be careful not to do anything that waits on other incoming messages or it
1803 static void process_message(struct ceph_connection *con)
1805 struct ceph_msg *msg;
1810 /* if first message, set peer_name */
1811 if (con->peer_name.type == 0)
1812 con->peer_name = msg->hdr.src;
1815 mutex_unlock(&con->mutex);
1817 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
1818 msg, le64_to_cpu(msg->hdr.seq),
1819 ENTITY_NAME(msg->hdr.src),
1820 le16_to_cpu(msg->hdr.type),
1821 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
1822 le32_to_cpu(msg->hdr.front_len),
1823 le32_to_cpu(msg->hdr.data_len),
1824 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
1825 con->ops->dispatch(con, msg);
1827 mutex_lock(&con->mutex);
1828 prepare_read_tag(con);
1833 * Write something to the socket. Called in a worker thread when the
1834 * socket appears to be writeable and we have something ready to send.
1836 static int try_write(struct ceph_connection *con)
1838 struct ceph_messenger *msgr = con->msgr;
1841 dout("try_write start %p state %lu nref %d\n", con, con->state,
1842 atomic_read(&con->nref));
1845 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
1847 /* open the socket first? */
1848 if (con->sock == NULL) {
1849 prepare_write_banner(msgr, con);
1850 prepare_write_connect(msgr, con, 1);
1851 prepare_read_banner(con);
1852 set_bit(CONNECTING, &con->state);
1853 clear_bit(NEGOTIATING, &con->state);
1855 BUG_ON(con->in_msg);
1856 con->in_tag = CEPH_MSGR_TAG_READY;
1857 dout("try_write initiating connect on %p new state %lu\n",
1859 con->sock = ceph_tcp_connect(con);
1860 if (IS_ERR(con->sock)) {
1862 con->error_msg = "connect error";
1869 /* kvec data queued? */
1870 if (con->out_skip) {
1871 ret = write_partial_skip(con);
1875 if (con->out_kvec_left) {
1876 ret = write_partial_kvec(con);
1883 if (con->out_msg_done) {
1884 ceph_msg_put(con->out_msg);
1885 con->out_msg = NULL; /* we're done with this one */
1889 ret = write_partial_msg_pages(con);
1891 goto more_kvec; /* we need to send the footer, too! */
1895 dout("try_write write_partial_msg_pages err %d\n",
1902 if (!test_bit(CONNECTING, &con->state)) {
1903 /* is anything else pending? */
1904 if (!list_empty(&con->out_queue)) {
1905 prepare_write_message(con);
1908 if (con->in_seq > con->in_seq_acked) {
1909 prepare_write_ack(con);
1912 if (test_and_clear_bit(KEEPALIVE_PENDING, &con->state)) {
1913 prepare_write_keepalive(con);
1918 /* Nothing to do! */
1919 clear_bit(WRITE_PENDING, &con->state);
1920 dout("try_write nothing else to write.\n");
1923 dout("try_write done on %p ret %d\n", con, ret);
1930 * Read what we can from the socket.
1932 static int try_read(struct ceph_connection *con)
1939 if (test_bit(STANDBY, &con->state))
1942 dout("try_read start on %p\n", con);
1945 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
1949 * process_connect and process_message drop and re-take
1950 * con->mutex. make sure we handle a racing close or reopen.
1952 if (test_bit(CLOSED, &con->state) ||
1953 test_bit(OPENING, &con->state)) {
1958 if (test_bit(CONNECTING, &con->state)) {
1959 if (!test_bit(NEGOTIATING, &con->state)) {
1960 dout("try_read connecting\n");
1961 ret = read_partial_banner(con);
1964 ret = process_banner(con);
1968 ret = read_partial_connect(con);
1971 ret = process_connect(con);
1977 if (con->in_base_pos < 0) {
1979 * skipping + discarding content.
1981 * FIXME: there must be a better way to do this!
1983 static char buf[1024];
1984 int skip = min(1024, -con->in_base_pos);
1985 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
1986 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
1989 con->in_base_pos += ret;
1990 if (con->in_base_pos)
1993 if (con->in_tag == CEPH_MSGR_TAG_READY) {
1997 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2000 dout("try_read got tag %d\n", (int)con->in_tag);
2001 switch (con->in_tag) {
2002 case CEPH_MSGR_TAG_MSG:
2003 prepare_read_message(con);
2005 case CEPH_MSGR_TAG_ACK:
2006 prepare_read_ack(con);
2008 case CEPH_MSGR_TAG_CLOSE:
2009 set_bit(CLOSED, &con->state); /* fixme */
2015 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2016 ret = read_partial_message(con);
2020 con->error_msg = "bad crc";
2024 con->error_msg = "io error";
2029 if (con->in_tag == CEPH_MSGR_TAG_READY)
2031 process_message(con);
2034 if (con->in_tag == CEPH_MSGR_TAG_ACK) {
2035 ret = read_partial_ack(con);
2043 dout("try_read done on %p ret %d\n", con, ret);
2047 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2048 con->error_msg = "protocol error, garbage tag";
2055 * Atomically queue work on a connection. Bump @con reference to
2056 * avoid races with connection teardown.
2058 static void queue_con(struct ceph_connection *con)
2060 if (test_bit(DEAD, &con->state)) {
2061 dout("queue_con %p ignoring: DEAD\n",
2066 if (!con->ops->get(con)) {
2067 dout("queue_con %p ref count 0\n", con);
2071 if (!queue_delayed_work(ceph_msgr_wq, &con->work, 0)) {
2072 dout("queue_con %p - already queued\n", con);
2075 dout("queue_con %p\n", con);
2080 * Do some work on a connection. Drop a connection ref when we're done.
2082 static void con_work(struct work_struct *work)
2084 struct ceph_connection *con = container_of(work, struct ceph_connection,
2088 mutex_lock(&con->mutex);
2090 if (test_and_clear_bit(BACKOFF, &con->state)) {
2091 dout("con_work %p backing off\n", con);
2092 if (queue_delayed_work(ceph_msgr_wq, &con->work,
2093 round_jiffies_relative(con->delay))) {
2094 dout("con_work %p backoff %lu\n", con, con->delay);
2095 mutex_unlock(&con->mutex);
2099 dout("con_work %p FAILED to back off %lu\n", con,
2104 if (test_bit(STANDBY, &con->state)) {
2105 dout("con_work %p STANDBY\n", con);
2108 if (test_bit(CLOSED, &con->state)) { /* e.g. if we are replaced */
2109 dout("con_work CLOSED\n");
2110 con_close_socket(con);
2113 if (test_and_clear_bit(OPENING, &con->state)) {
2114 /* reopen w/ new peer */
2115 dout("con_work OPENING\n");
2116 con_close_socket(con);
2119 if (test_and_clear_bit(SOCK_CLOSED, &con->state))
2122 ret = try_read(con);
2128 ret = try_write(con);
2135 mutex_unlock(&con->mutex);
2141 mutex_unlock(&con->mutex);
2142 ceph_fault(con); /* error/fault path */
2148 * Generic error/fault handler. A retry mechanism is used with
2149 * exponential backoff
2151 static void ceph_fault(struct ceph_connection *con)
2153 pr_err("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2154 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2155 dout("fault %p state %lu to peer %s\n",
2156 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2158 if (test_bit(LOSSYTX, &con->state)) {
2159 dout("fault on LOSSYTX channel\n");
2163 mutex_lock(&con->mutex);
2164 if (test_bit(CLOSED, &con->state))
2167 con_close_socket(con);
2170 ceph_msg_put(con->in_msg);
2174 /* Requeue anything that hasn't been acked */
2175 list_splice_init(&con->out_sent, &con->out_queue);
2177 /* If there are no messages queued or keepalive pending, place
2178 * the connection in a STANDBY state */
2179 if (list_empty(&con->out_queue) &&
2180 !test_bit(KEEPALIVE_PENDING, &con->state)) {
2181 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2182 clear_bit(WRITE_PENDING, &con->state);
2183 set_bit(STANDBY, &con->state);
2185 /* retry after a delay. */
2186 if (con->delay == 0)
2187 con->delay = BASE_DELAY_INTERVAL;
2188 else if (con->delay < MAX_DELAY_INTERVAL)
2191 if (queue_delayed_work(ceph_msgr_wq, &con->work,
2192 round_jiffies_relative(con->delay))) {
2193 dout("fault queued %p delay %lu\n", con, con->delay);
2196 dout("fault failed to queue %p delay %lu, backoff\n",
2199 * In many cases we see a socket state change
2200 * while con_work is running and end up
2201 * queuing (non-delayed) work, such that we
2202 * can't backoff with a delay. Set a flag so
2203 * that when con_work restarts we schedule the
2206 set_bit(BACKOFF, &con->state);
2211 mutex_unlock(&con->mutex);
2214 * in case we faulted due to authentication, invalidate our
2215 * current tickets so that we can get new ones.
2217 if (con->auth_retry && con->ops->invalidate_authorizer) {
2218 dout("calling invalidate_authorizer()\n");
2219 con->ops->invalidate_authorizer(con);
2222 if (con->ops->fault)
2223 con->ops->fault(con);
2229 * create a new messenger instance
2231 struct ceph_messenger *ceph_messenger_create(struct ceph_entity_addr *myaddr,
2232 u32 supported_features,
2233 u32 required_features)
2235 struct ceph_messenger *msgr;
2237 msgr = kzalloc(sizeof(*msgr), GFP_KERNEL);
2239 return ERR_PTR(-ENOMEM);
2241 msgr->supported_features = supported_features;
2242 msgr->required_features = required_features;
2244 spin_lock_init(&msgr->global_seq_lock);
2246 /* the zero page is needed if a request is "canceled" while the message
2247 * is being written over the socket */
2248 msgr->zero_page = __page_cache_alloc(GFP_KERNEL | __GFP_ZERO);
2249 if (!msgr->zero_page) {
2251 return ERR_PTR(-ENOMEM);
2253 kmap(msgr->zero_page);
2256 msgr->inst.addr = *myaddr;
2258 /* select a random nonce */
2259 msgr->inst.addr.type = 0;
2260 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2261 encode_my_addr(msgr);
2263 dout("messenger_create %p\n", msgr);
2266 EXPORT_SYMBOL(ceph_messenger_create);
2268 void ceph_messenger_destroy(struct ceph_messenger *msgr)
2270 dout("destroy %p\n", msgr);
2271 kunmap(msgr->zero_page);
2272 __free_page(msgr->zero_page);
2274 dout("destroyed messenger %p\n", msgr);
2276 EXPORT_SYMBOL(ceph_messenger_destroy);
2278 static void clear_standby(struct ceph_connection *con)
2280 /* come back from STANDBY? */
2281 if (test_and_clear_bit(STANDBY, &con->state)) {
2282 mutex_lock(&con->mutex);
2283 dout("clear_standby %p and ++connect_seq\n", con);
2285 WARN_ON(test_bit(WRITE_PENDING, &con->state));
2286 WARN_ON(test_bit(KEEPALIVE_PENDING, &con->state));
2287 mutex_unlock(&con->mutex);
2292 * Queue up an outgoing message on the given connection.
2294 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2296 if (test_bit(CLOSED, &con->state)) {
2297 dout("con_send %p closed, dropping %p\n", con, msg);
2303 msg->hdr.src = con->msgr->inst.name;
2305 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2307 msg->needs_out_seq = true;
2310 mutex_lock(&con->mutex);
2311 BUG_ON(!list_empty(&msg->list_head));
2312 list_add_tail(&msg->list_head, &con->out_queue);
2313 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2314 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2315 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2316 le32_to_cpu(msg->hdr.front_len),
2317 le32_to_cpu(msg->hdr.middle_len),
2318 le32_to_cpu(msg->hdr.data_len));
2319 mutex_unlock(&con->mutex);
2321 /* if there wasn't anything waiting to send before, queue
2324 if (test_and_set_bit(WRITE_PENDING, &con->state) == 0)
2327 EXPORT_SYMBOL(ceph_con_send);
2330 * Revoke a message that was previously queued for send
2332 void ceph_con_revoke(struct ceph_connection *con, struct ceph_msg *msg)
2334 mutex_lock(&con->mutex);
2335 if (!list_empty(&msg->list_head)) {
2336 dout("con_revoke %p msg %p - was on queue\n", con, msg);
2337 list_del_init(&msg->list_head);
2341 if (con->out_msg == msg) {
2342 dout("con_revoke %p msg %p - was sending\n", con, msg);
2343 con->out_msg = NULL;
2344 if (con->out_kvec_is_msg) {
2345 con->out_skip = con->out_kvec_bytes;
2346 con->out_kvec_is_msg = false;
2351 mutex_unlock(&con->mutex);
2355 * Revoke a message that we may be reading data into
2357 void ceph_con_revoke_message(struct ceph_connection *con, struct ceph_msg *msg)
2359 mutex_lock(&con->mutex);
2360 if (con->in_msg && con->in_msg == msg) {
2361 unsigned front_len = le32_to_cpu(con->in_hdr.front_len);
2362 unsigned middle_len = le32_to_cpu(con->in_hdr.middle_len);
2363 unsigned data_len = le32_to_cpu(con->in_hdr.data_len);
2365 /* skip rest of message */
2366 dout("con_revoke_pages %p msg %p revoked\n", con, msg);
2367 con->in_base_pos = con->in_base_pos -
2368 sizeof(struct ceph_msg_header) -
2372 sizeof(struct ceph_msg_footer);
2373 ceph_msg_put(con->in_msg);
2375 con->in_tag = CEPH_MSGR_TAG_READY;
2378 dout("con_revoke_pages %p msg %p pages %p no-op\n",
2379 con, con->in_msg, msg);
2381 mutex_unlock(&con->mutex);
2385 * Queue a keepalive byte to ensure the tcp connection is alive.
2387 void ceph_con_keepalive(struct ceph_connection *con)
2389 dout("con_keepalive %p\n", con);
2391 if (test_and_set_bit(KEEPALIVE_PENDING, &con->state) == 0 &&
2392 test_and_set_bit(WRITE_PENDING, &con->state) == 0)
2395 EXPORT_SYMBOL(ceph_con_keepalive);
2399 * construct a new message with given type, size
2400 * the new msg has a ref count of 1.
2402 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
2407 m = kmalloc(sizeof(*m), flags);
2410 kref_init(&m->kref);
2411 INIT_LIST_HEAD(&m->list_head);
2414 m->hdr.type = cpu_to_le16(type);
2415 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
2417 m->hdr.front_len = cpu_to_le32(front_len);
2418 m->hdr.middle_len = 0;
2419 m->hdr.data_len = 0;
2420 m->hdr.data_off = 0;
2421 m->hdr.reserved = 0;
2422 m->footer.front_crc = 0;
2423 m->footer.middle_crc = 0;
2424 m->footer.data_crc = 0;
2425 m->footer.flags = 0;
2426 m->front_max = front_len;
2427 m->front_is_vmalloc = false;
2428 m->more_to_follow = false;
2437 m->page_alignment = 0;
2447 if (front_len > PAGE_CACHE_SIZE) {
2448 m->front.iov_base = __vmalloc(front_len, flags,
2450 m->front_is_vmalloc = true;
2452 m->front.iov_base = kmalloc(front_len, flags);
2454 if (m->front.iov_base == NULL) {
2455 dout("ceph_msg_new can't allocate %d bytes\n",
2460 m->front.iov_base = NULL;
2462 m->front.iov_len = front_len;
2464 dout("ceph_msg_new %p front %d\n", m, front_len);
2471 pr_err("msg_new can't create type %d front %d\n", type,
2475 dout("msg_new can't create type %d front %d\n", type,
2480 EXPORT_SYMBOL(ceph_msg_new);
2483 * Allocate "middle" portion of a message, if it is needed and wasn't
2484 * allocated by alloc_msg. This allows us to read a small fixed-size
2485 * per-type header in the front and then gracefully fail (i.e.,
2486 * propagate the error to the caller based on info in the front) when
2487 * the middle is too large.
2489 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
2491 int type = le16_to_cpu(msg->hdr.type);
2492 int middle_len = le32_to_cpu(msg->hdr.middle_len);
2494 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
2495 ceph_msg_type_name(type), middle_len);
2496 BUG_ON(!middle_len);
2497 BUG_ON(msg->middle);
2499 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
2506 * Generic message allocator, for incoming messages.
2508 static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con,
2509 struct ceph_msg_header *hdr,
2512 int type = le16_to_cpu(hdr->type);
2513 int front_len = le32_to_cpu(hdr->front_len);
2514 int middle_len = le32_to_cpu(hdr->middle_len);
2515 struct ceph_msg *msg = NULL;
2518 if (con->ops->alloc_msg) {
2519 mutex_unlock(&con->mutex);
2520 msg = con->ops->alloc_msg(con, hdr, skip);
2521 mutex_lock(&con->mutex);
2527 msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
2529 pr_err("unable to allocate msg type %d len %d\n",
2533 msg->page_alignment = le16_to_cpu(hdr->data_off);
2535 memcpy(&msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2537 if (middle_len && !msg->middle) {
2538 ret = ceph_alloc_middle(con, msg);
2550 * Free a generically kmalloc'd message.
2552 void ceph_msg_kfree(struct ceph_msg *m)
2554 dout("msg_kfree %p\n", m);
2555 if (m->front_is_vmalloc)
2556 vfree(m->front.iov_base);
2558 kfree(m->front.iov_base);
2563 * Drop a msg ref. Destroy as needed.
2565 void ceph_msg_last_put(struct kref *kref)
2567 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
2569 dout("ceph_msg_put last one on %p\n", m);
2570 WARN_ON(!list_empty(&m->list_head));
2572 /* drop middle, data, if any */
2574 ceph_buffer_put(m->middle);
2581 ceph_pagelist_release(m->pagelist);
2589 ceph_msgpool_put(m->pool, m);
2593 EXPORT_SYMBOL(ceph_msg_last_put);
2595 void ceph_msg_dump(struct ceph_msg *msg)
2597 pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg,
2598 msg->front_max, msg->nr_pages);
2599 print_hex_dump(KERN_DEBUG, "header: ",
2600 DUMP_PREFIX_OFFSET, 16, 1,
2601 &msg->hdr, sizeof(msg->hdr), true);
2602 print_hex_dump(KERN_DEBUG, " front: ",
2603 DUMP_PREFIX_OFFSET, 16, 1,
2604 msg->front.iov_base, msg->front.iov_len, true);
2606 print_hex_dump(KERN_DEBUG, "middle: ",
2607 DUMP_PREFIX_OFFSET, 16, 1,
2608 msg->middle->vec.iov_base,
2609 msg->middle->vec.iov_len, true);
2610 print_hex_dump(KERN_DEBUG, "footer: ",
2611 DUMP_PREFIX_OFFSET, 16, 1,
2612 &msg->footer, sizeof(msg->footer), true);
2614 EXPORT_SYMBOL(ceph_msg_dump);