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 /* State values for ceph_connection->sock_state; NEW is assumed to be 0 */
34 #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
35 #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
36 #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
37 #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
38 #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
40 /* static tag bytes (protocol control messages) */
41 static char tag_msg = CEPH_MSGR_TAG_MSG;
42 static char tag_ack = CEPH_MSGR_TAG_ACK;
43 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
46 static struct lock_class_key socket_class;
50 * When skipping (ignoring) a block of input we read it into a "skip
51 * buffer," which is this many bytes in size.
53 #define SKIP_BUF_SIZE 1024
55 static void queue_con(struct ceph_connection *con);
56 static void con_work(struct work_struct *);
57 static void ceph_fault(struct ceph_connection *con);
60 * Nicely render a sockaddr as a string. An array of formatted
61 * strings is used, to approximate reentrancy.
63 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
64 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
65 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
66 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
68 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
69 static atomic_t addr_str_seq = ATOMIC_INIT(0);
71 static struct page *zero_page; /* used in certain error cases */
73 const char *ceph_pr_addr(const struct sockaddr_storage *ss)
77 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
78 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
80 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
83 switch (ss->ss_family) {
85 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
86 ntohs(in4->sin_port));
90 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
91 ntohs(in6->sin6_port));
95 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
101 EXPORT_SYMBOL(ceph_pr_addr);
103 static void encode_my_addr(struct ceph_messenger *msgr)
105 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
106 ceph_encode_addr(&msgr->my_enc_addr);
110 * work queue for all reading and writing to/from the socket.
112 static struct workqueue_struct *ceph_msgr_wq;
114 void _ceph_msgr_exit(void)
117 destroy_workqueue(ceph_msgr_wq);
121 BUG_ON(zero_page == NULL);
123 page_cache_release(zero_page);
127 int ceph_msgr_init(void)
129 BUG_ON(zero_page != NULL);
130 zero_page = ZERO_PAGE(0);
131 page_cache_get(zero_page);
133 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0);
137 pr_err("msgr_init failed to create workqueue\n");
142 EXPORT_SYMBOL(ceph_msgr_init);
144 void ceph_msgr_exit(void)
146 BUG_ON(ceph_msgr_wq == NULL);
150 EXPORT_SYMBOL(ceph_msgr_exit);
152 void ceph_msgr_flush(void)
154 flush_workqueue(ceph_msgr_wq);
156 EXPORT_SYMBOL(ceph_msgr_flush);
158 /* Connection socket state transition functions */
160 static void con_sock_state_init(struct ceph_connection *con)
164 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
165 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
166 printk("%s: unexpected old state %d\n", __func__, old_state);
169 static void con_sock_state_connecting(struct ceph_connection *con)
173 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
174 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
175 printk("%s: unexpected old state %d\n", __func__, old_state);
178 static void con_sock_state_connected(struct ceph_connection *con)
182 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
183 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
184 printk("%s: unexpected old state %d\n", __func__, old_state);
187 static void con_sock_state_closing(struct ceph_connection *con)
191 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
192 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
193 old_state != CON_SOCK_STATE_CONNECTED &&
194 old_state != CON_SOCK_STATE_CLOSING))
195 printk("%s: unexpected old state %d\n", __func__, old_state);
198 static void con_sock_state_closed(struct ceph_connection *con)
202 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
203 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
204 old_state != CON_SOCK_STATE_CLOSING))
205 printk("%s: unexpected old state %d\n", __func__, old_state);
209 * socket callback functions
212 /* data available on socket, or listen socket received a connect */
213 static void ceph_sock_data_ready(struct sock *sk, int count_unused)
215 struct ceph_connection *con = sk->sk_user_data;
217 if (sk->sk_state != TCP_CLOSE_WAIT) {
218 dout("%s on %p state = %lu, queueing work\n", __func__,
224 /* socket has buffer space for writing */
225 static void ceph_sock_write_space(struct sock *sk)
227 struct ceph_connection *con = sk->sk_user_data;
229 /* only queue to workqueue if there is data we want to write,
230 * and there is sufficient space in the socket buffer to accept
231 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
232 * doesn't get called again until try_write() fills the socket
233 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
234 * and net/core/stream.c:sk_stream_write_space().
236 if (test_bit(WRITE_PENDING, &con->flags)) {
237 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
238 dout("%s %p queueing write work\n", __func__, con);
239 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
243 dout("%s %p nothing to write\n", __func__, con);
247 /* socket's state has changed */
248 static void ceph_sock_state_change(struct sock *sk)
250 struct ceph_connection *con = sk->sk_user_data;
252 dout("%s %p state = %lu sk_state = %u\n", __func__,
253 con, con->state, sk->sk_state);
255 if (test_bit(CLOSED, &con->state))
258 switch (sk->sk_state) {
260 dout("%s TCP_CLOSE\n", __func__);
262 dout("%s TCP_CLOSE_WAIT\n", __func__);
263 con_sock_state_closing(con);
264 if (test_and_set_bit(SOCK_CLOSED, &con->flags) == 0) {
265 if (test_bit(CONNECTING, &con->state))
266 con->error_msg = "connection failed";
268 con->error_msg = "socket closed";
272 case TCP_ESTABLISHED:
273 dout("%s TCP_ESTABLISHED\n", __func__);
274 con_sock_state_connected(con);
277 default: /* Everything else is uninteresting */
283 * set up socket callbacks
285 static void set_sock_callbacks(struct socket *sock,
286 struct ceph_connection *con)
288 struct sock *sk = sock->sk;
289 sk->sk_user_data = con;
290 sk->sk_data_ready = ceph_sock_data_ready;
291 sk->sk_write_space = ceph_sock_write_space;
292 sk->sk_state_change = ceph_sock_state_change;
301 * initiate connection to a remote socket.
303 static int ceph_tcp_connect(struct ceph_connection *con)
305 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
310 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
314 sock->sk->sk_allocation = GFP_NOFS;
316 #ifdef CONFIG_LOCKDEP
317 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
320 set_sock_callbacks(sock, con);
322 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
324 con_sock_state_connecting(con);
325 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
327 if (ret == -EINPROGRESS) {
328 dout("connect %s EINPROGRESS sk_state = %u\n",
329 ceph_pr_addr(&con->peer_addr.in_addr),
331 } else if (ret < 0) {
332 pr_err("connect %s error %d\n",
333 ceph_pr_addr(&con->peer_addr.in_addr), ret);
335 con->error_msg = "connect error";
343 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
345 struct kvec iov = {buf, len};
346 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
349 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
356 * write something. @more is true if caller will be sending more data
359 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
360 size_t kvlen, size_t len, int more)
362 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
366 msg.msg_flags |= MSG_MORE;
368 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
370 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
376 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
377 int offset, size_t size, int more)
379 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
382 ret = kernel_sendpage(sock, page, offset, size, flags);
391 * Shutdown/close the socket for the given connection.
393 static int con_close_socket(struct ceph_connection *con)
397 dout("con_close_socket on %p sock %p\n", con, con->sock);
400 set_bit(SOCK_CLOSED, &con->state);
401 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
402 sock_release(con->sock);
404 clear_bit(SOCK_CLOSED, &con->state);
405 con_sock_state_closed(con);
410 * Reset a connection. Discard all incoming and outgoing messages
411 * and clear *_seq state.
413 static void ceph_msg_remove(struct ceph_msg *msg)
415 list_del_init(&msg->list_head);
416 BUG_ON(msg->con == NULL);
417 msg->con->ops->put(msg->con);
422 static void ceph_msg_remove_list(struct list_head *head)
424 while (!list_empty(head)) {
425 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
427 ceph_msg_remove(msg);
431 static void reset_connection(struct ceph_connection *con)
433 /* reset connection, out_queue, msg_ and connect_seq */
434 /* discard existing out_queue and msg_seq */
435 ceph_msg_remove_list(&con->out_queue);
436 ceph_msg_remove_list(&con->out_sent);
439 BUG_ON(con->in_msg->con != con);
440 con->in_msg->con = NULL;
441 ceph_msg_put(con->in_msg);
446 con->connect_seq = 0;
449 ceph_msg_put(con->out_msg);
453 con->in_seq_acked = 0;
457 * mark a peer down. drop any open connections.
459 void ceph_con_close(struct ceph_connection *con)
461 dout("con_close %p peer %s\n", con,
462 ceph_pr_addr(&con->peer_addr.in_addr));
463 clear_bit(NEGOTIATING, &con->state);
464 clear_bit(STANDBY, &con->state); /* avoid connect_seq bump */
465 set_bit(CLOSED, &con->state);
467 clear_bit(LOSSYTX, &con->flags); /* so we retry next connect */
468 clear_bit(KEEPALIVE_PENDING, &con->flags);
469 clear_bit(WRITE_PENDING, &con->flags);
471 mutex_lock(&con->mutex);
472 reset_connection(con);
473 con->peer_global_seq = 0;
474 cancel_delayed_work(&con->work);
475 mutex_unlock(&con->mutex);
478 EXPORT_SYMBOL(ceph_con_close);
481 * Reopen a closed connection, with a new peer address.
483 void ceph_con_open(struct ceph_connection *con, struct ceph_entity_addr *addr)
485 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
486 set_bit(OPENING, &con->state);
487 WARN_ON(!test_and_clear_bit(CLOSED, &con->state));
489 memcpy(&con->peer_addr, addr, sizeof(*addr));
490 con->delay = 0; /* reset backoff memory */
493 EXPORT_SYMBOL(ceph_con_open);
496 * return true if this connection ever successfully opened
498 bool ceph_con_opened(struct ceph_connection *con)
500 return con->connect_seq > 0;
506 struct ceph_connection *ceph_con_get(struct ceph_connection *con)
508 int nref = __atomic_add_unless(&con->nref, 1, 0);
510 dout("con_get %p nref = %d -> %d\n", con, nref, nref + 1);
512 return nref ? con : NULL;
515 void ceph_con_put(struct ceph_connection *con)
517 int nref = atomic_dec_return(&con->nref);
524 dout("con_put %p nref = %d -> %d\n", con, nref + 1, nref);
528 * initialize a new connection.
530 void ceph_con_init(struct ceph_connection *con, void *private,
531 const struct ceph_connection_operations *ops,
532 struct ceph_messenger *msgr, __u8 entity_type, __u64 entity_num)
534 dout("con_init %p\n", con);
535 memset(con, 0, sizeof(*con));
536 con->private = private;
538 atomic_set(&con->nref, 1);
541 con_sock_state_init(con);
543 con->peer_name.type = (__u8) entity_type;
544 con->peer_name.num = cpu_to_le64(entity_num);
546 mutex_init(&con->mutex);
547 INIT_LIST_HEAD(&con->out_queue);
548 INIT_LIST_HEAD(&con->out_sent);
549 INIT_DELAYED_WORK(&con->work, con_work);
551 set_bit(CLOSED, &con->state);
553 EXPORT_SYMBOL(ceph_con_init);
557 * We maintain a global counter to order connection attempts. Get
558 * a unique seq greater than @gt.
560 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
564 spin_lock(&msgr->global_seq_lock);
565 if (msgr->global_seq < gt)
566 msgr->global_seq = gt;
567 ret = ++msgr->global_seq;
568 spin_unlock(&msgr->global_seq_lock);
572 static void con_out_kvec_reset(struct ceph_connection *con)
574 con->out_kvec_left = 0;
575 con->out_kvec_bytes = 0;
576 con->out_kvec_cur = &con->out_kvec[0];
579 static void con_out_kvec_add(struct ceph_connection *con,
580 size_t size, void *data)
584 index = con->out_kvec_left;
585 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
587 con->out_kvec[index].iov_len = size;
588 con->out_kvec[index].iov_base = data;
589 con->out_kvec_left++;
590 con->out_kvec_bytes += size;
594 * Prepare footer for currently outgoing message, and finish things
595 * off. Assumes out_kvec* are already valid.. we just add on to the end.
597 static void prepare_write_message_footer(struct ceph_connection *con)
599 struct ceph_msg *m = con->out_msg;
600 int v = con->out_kvec_left;
602 dout("prepare_write_message_footer %p\n", con);
603 con->out_kvec_is_msg = true;
604 con->out_kvec[v].iov_base = &m->footer;
605 con->out_kvec[v].iov_len = sizeof(m->footer);
606 con->out_kvec_bytes += sizeof(m->footer);
607 con->out_kvec_left++;
608 con->out_more = m->more_to_follow;
609 con->out_msg_done = true;
613 * Prepare headers for the next outgoing message.
615 static void prepare_write_message(struct ceph_connection *con)
620 con_out_kvec_reset(con);
621 con->out_kvec_is_msg = true;
622 con->out_msg_done = false;
624 /* Sneak an ack in there first? If we can get it into the same
625 * TCP packet that's a good thing. */
626 if (con->in_seq > con->in_seq_acked) {
627 con->in_seq_acked = con->in_seq;
628 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
629 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
630 con_out_kvec_add(con, sizeof (con->out_temp_ack),
634 BUG_ON(list_empty(&con->out_queue));
635 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
637 BUG_ON(m->con != con);
639 /* put message on sent list */
641 list_move_tail(&m->list_head, &con->out_sent);
644 * only assign outgoing seq # if we haven't sent this message
645 * yet. if it is requeued, resend with it's original seq.
647 if (m->needs_out_seq) {
648 m->hdr.seq = cpu_to_le64(++con->out_seq);
649 m->needs_out_seq = false;
656 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n",
657 m, con->out_seq, le16_to_cpu(m->hdr.type),
658 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
659 le32_to_cpu(m->hdr.data_len),
661 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
663 /* tag + hdr + front + middle */
664 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
665 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
666 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
669 con_out_kvec_add(con, m->middle->vec.iov_len,
670 m->middle->vec.iov_base);
672 /* fill in crc (except data pages), footer */
673 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
674 con->out_msg->hdr.crc = cpu_to_le32(crc);
675 con->out_msg->footer.flags = CEPH_MSG_FOOTER_COMPLETE;
677 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
678 con->out_msg->footer.front_crc = cpu_to_le32(crc);
680 crc = crc32c(0, m->middle->vec.iov_base,
681 m->middle->vec.iov_len);
682 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
684 con->out_msg->footer.middle_crc = 0;
685 con->out_msg->footer.data_crc = 0;
686 dout("prepare_write_message front_crc %u data_crc %u\n",
687 le32_to_cpu(con->out_msg->footer.front_crc),
688 le32_to_cpu(con->out_msg->footer.middle_crc));
690 /* is there a data payload? */
691 if (le32_to_cpu(m->hdr.data_len) > 0) {
692 /* initialize page iterator */
693 con->out_msg_pos.page = 0;
695 con->out_msg_pos.page_pos = m->page_alignment;
697 con->out_msg_pos.page_pos = 0;
698 con->out_msg_pos.data_pos = 0;
699 con->out_msg_pos.did_page_crc = false;
700 con->out_more = 1; /* data + footer will follow */
702 /* no, queue up footer too and be done */
703 prepare_write_message_footer(con);
706 set_bit(WRITE_PENDING, &con->flags);
712 static void prepare_write_ack(struct ceph_connection *con)
714 dout("prepare_write_ack %p %llu -> %llu\n", con,
715 con->in_seq_acked, con->in_seq);
716 con->in_seq_acked = con->in_seq;
718 con_out_kvec_reset(con);
720 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
722 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
723 con_out_kvec_add(con, sizeof (con->out_temp_ack),
726 con->out_more = 1; /* more will follow.. eventually.. */
727 set_bit(WRITE_PENDING, &con->flags);
731 * Prepare to write keepalive byte.
733 static void prepare_write_keepalive(struct ceph_connection *con)
735 dout("prepare_write_keepalive %p\n", con);
736 con_out_kvec_reset(con);
737 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
738 set_bit(WRITE_PENDING, &con->flags);
742 * Connection negotiation.
745 static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
748 struct ceph_auth_handshake *auth;
750 if (!con->ops->get_authorizer) {
751 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
752 con->out_connect.authorizer_len = 0;
757 /* Can't hold the mutex while getting authorizer */
759 mutex_unlock(&con->mutex);
761 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
763 mutex_lock(&con->mutex);
767 if (test_bit(CLOSED, &con->state) || test_bit(OPENING, &con->flags))
768 return ERR_PTR(-EAGAIN);
770 con->auth_reply_buf = auth->authorizer_reply_buf;
771 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
778 * We connected to a peer and are saying hello.
780 static void prepare_write_banner(struct ceph_connection *con)
782 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
783 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
784 &con->msgr->my_enc_addr);
787 set_bit(WRITE_PENDING, &con->flags);
790 static int prepare_write_connect(struct ceph_connection *con)
792 unsigned int global_seq = get_global_seq(con->msgr, 0);
795 struct ceph_auth_handshake *auth;
797 switch (con->peer_name.type) {
798 case CEPH_ENTITY_TYPE_MON:
799 proto = CEPH_MONC_PROTOCOL;
801 case CEPH_ENTITY_TYPE_OSD:
802 proto = CEPH_OSDC_PROTOCOL;
804 case CEPH_ENTITY_TYPE_MDS:
805 proto = CEPH_MDSC_PROTOCOL;
811 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
812 con->connect_seq, global_seq, proto);
814 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
815 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
816 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
817 con->out_connect.global_seq = cpu_to_le32(global_seq);
818 con->out_connect.protocol_version = cpu_to_le32(proto);
819 con->out_connect.flags = 0;
821 auth_proto = CEPH_AUTH_UNKNOWN;
822 auth = get_connect_authorizer(con, &auth_proto);
824 return PTR_ERR(auth);
826 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
827 con->out_connect.authorizer_len = auth ?
828 cpu_to_le32(auth->authorizer_buf_len) : 0;
830 con_out_kvec_add(con, sizeof (con->out_connect),
832 if (auth && auth->authorizer_buf_len)
833 con_out_kvec_add(con, auth->authorizer_buf_len,
834 auth->authorizer_buf);
837 set_bit(WRITE_PENDING, &con->flags);
843 * write as much of pending kvecs to the socket as we can.
845 * 0 -> socket full, but more to do
848 static int write_partial_kvec(struct ceph_connection *con)
852 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
853 while (con->out_kvec_bytes > 0) {
854 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
855 con->out_kvec_left, con->out_kvec_bytes,
859 con->out_kvec_bytes -= ret;
860 if (con->out_kvec_bytes == 0)
863 /* account for full iov entries consumed */
864 while (ret >= con->out_kvec_cur->iov_len) {
865 BUG_ON(!con->out_kvec_left);
866 ret -= con->out_kvec_cur->iov_len;
868 con->out_kvec_left--;
870 /* and for a partially-consumed entry */
872 con->out_kvec_cur->iov_len -= ret;
873 con->out_kvec_cur->iov_base += ret;
876 con->out_kvec_left = 0;
877 con->out_kvec_is_msg = false;
880 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
881 con->out_kvec_bytes, con->out_kvec_left, ret);
882 return ret; /* done! */
886 static void init_bio_iter(struct bio *bio, struct bio **iter, int *seg)
897 static void iter_bio_next(struct bio **bio_iter, int *seg)
899 if (*bio_iter == NULL)
902 BUG_ON(*seg >= (*bio_iter)->bi_vcnt);
905 if (*seg == (*bio_iter)->bi_vcnt)
906 init_bio_iter((*bio_iter)->bi_next, bio_iter, seg);
911 * Write as much message data payload as we can. If we finish, queue
913 * 1 -> done, footer is now queued in out_kvec[].
914 * 0 -> socket full, but more to do
917 static int write_partial_msg_pages(struct ceph_connection *con)
919 struct ceph_msg *msg = con->out_msg;
920 unsigned int data_len = le32_to_cpu(msg->hdr.data_len);
922 bool do_datacrc = !con->msgr->nocrc;
926 size_t trail_len = (msg->trail ? msg->trail->length : 0);
928 dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
929 con, con->out_msg, con->out_msg_pos.page, con->out_msg->nr_pages,
930 con->out_msg_pos.page_pos);
933 if (msg->bio && !msg->bio_iter)
934 init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg);
937 while (data_len > con->out_msg_pos.data_pos) {
938 struct page *page = NULL;
939 int max_write = PAGE_SIZE;
942 total_max_write = data_len - trail_len -
943 con->out_msg_pos.data_pos;
946 * if we are calculating the data crc (the default), we need
947 * to map the page. if our pages[] has been revoked, use the
951 /* have we reached the trail part of the data? */
952 if (con->out_msg_pos.data_pos >= data_len - trail_len) {
955 total_max_write = data_len - con->out_msg_pos.data_pos;
957 page = list_first_entry(&msg->trail->head,
959 max_write = PAGE_SIZE;
960 } else if (msg->pages) {
961 page = msg->pages[con->out_msg_pos.page];
962 } else if (msg->pagelist) {
963 page = list_first_entry(&msg->pagelist->head,
966 } else if (msg->bio) {
969 bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg);
971 bio_offset = bv->bv_offset;
972 max_write = bv->bv_len;
977 len = min_t(int, max_write - con->out_msg_pos.page_pos,
980 if (do_datacrc && !con->out_msg_pos.did_page_crc) {
983 u32 tmpcrc = le32_to_cpu(con->out_msg->footer.data_crc);
987 BUG_ON(kaddr == NULL);
988 base = kaddr + con->out_msg_pos.page_pos + bio_offset;
989 crc = crc32c(tmpcrc, base, len);
990 con->out_msg->footer.data_crc = cpu_to_le32(crc);
991 con->out_msg_pos.did_page_crc = true;
993 ret = ceph_tcp_sendpage(con->sock, page,
994 con->out_msg_pos.page_pos + bio_offset,
1003 con->out_msg_pos.data_pos += ret;
1004 con->out_msg_pos.page_pos += ret;
1006 con->out_msg_pos.page_pos = 0;
1007 con->out_msg_pos.page++;
1008 con->out_msg_pos.did_page_crc = false;
1010 list_move_tail(&page->lru,
1012 else if (msg->pagelist)
1013 list_move_tail(&page->lru,
1014 &msg->pagelist->head);
1017 iter_bio_next(&msg->bio_iter, &msg->bio_seg);
1022 dout("write_partial_msg_pages %p msg %p done\n", con, msg);
1024 /* prepare and queue up footer, too */
1026 con->out_msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1027 con_out_kvec_reset(con);
1028 prepare_write_message_footer(con);
1037 static int write_partial_skip(struct ceph_connection *con)
1041 while (con->out_skip > 0) {
1042 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
1044 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, 1);
1047 con->out_skip -= ret;
1055 * Prepare to read connection handshake, or an ack.
1057 static void prepare_read_banner(struct ceph_connection *con)
1059 dout("prepare_read_banner %p\n", con);
1060 con->in_base_pos = 0;
1063 static void prepare_read_connect(struct ceph_connection *con)
1065 dout("prepare_read_connect %p\n", con);
1066 con->in_base_pos = 0;
1069 static void prepare_read_ack(struct ceph_connection *con)
1071 dout("prepare_read_ack %p\n", con);
1072 con->in_base_pos = 0;
1075 static void prepare_read_tag(struct ceph_connection *con)
1077 dout("prepare_read_tag %p\n", con);
1078 con->in_base_pos = 0;
1079 con->in_tag = CEPH_MSGR_TAG_READY;
1083 * Prepare to read a message.
1085 static int prepare_read_message(struct ceph_connection *con)
1087 dout("prepare_read_message %p\n", con);
1088 BUG_ON(con->in_msg != NULL);
1089 con->in_base_pos = 0;
1090 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1095 static int read_partial(struct ceph_connection *con,
1096 int end, int size, void *object)
1098 while (con->in_base_pos < end) {
1099 int left = end - con->in_base_pos;
1100 int have = size - left;
1101 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1104 con->in_base_pos += ret;
1111 * Read all or part of the connect-side handshake on a new connection
1113 static int read_partial_banner(struct ceph_connection *con)
1119 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1122 size = strlen(CEPH_BANNER);
1124 ret = read_partial(con, end, size, con->in_banner);
1128 size = sizeof (con->actual_peer_addr);
1130 ret = read_partial(con, end, size, &con->actual_peer_addr);
1134 size = sizeof (con->peer_addr_for_me);
1136 ret = read_partial(con, end, size, &con->peer_addr_for_me);
1144 static int read_partial_connect(struct ceph_connection *con)
1150 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1152 size = sizeof (con->in_reply);
1154 ret = read_partial(con, end, size, &con->in_reply);
1158 size = le32_to_cpu(con->in_reply.authorizer_len);
1160 ret = read_partial(con, end, size, con->auth_reply_buf);
1164 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1165 con, (int)con->in_reply.tag,
1166 le32_to_cpu(con->in_reply.connect_seq),
1167 le32_to_cpu(con->in_reply.global_seq));
1174 * Verify the hello banner looks okay.
1176 static int verify_hello(struct ceph_connection *con)
1178 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1179 pr_err("connect to %s got bad banner\n",
1180 ceph_pr_addr(&con->peer_addr.in_addr));
1181 con->error_msg = "protocol error, bad banner";
1187 static bool addr_is_blank(struct sockaddr_storage *ss)
1189 switch (ss->ss_family) {
1191 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1194 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1195 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1196 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1197 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1202 static int addr_port(struct sockaddr_storage *ss)
1204 switch (ss->ss_family) {
1206 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1208 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1213 static void addr_set_port(struct sockaddr_storage *ss, int p)
1215 switch (ss->ss_family) {
1217 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1220 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1226 * Unlike other *_pton function semantics, zero indicates success.
1228 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1229 char delim, const char **ipend)
1231 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1232 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1234 memset(ss, 0, sizeof(*ss));
1236 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1237 ss->ss_family = AF_INET;
1241 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1242 ss->ss_family = AF_INET6;
1250 * Extract hostname string and resolve using kernel DNS facility.
1252 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1253 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1254 struct sockaddr_storage *ss, char delim, const char **ipend)
1256 const char *end, *delim_p;
1257 char *colon_p, *ip_addr = NULL;
1261 * The end of the hostname occurs immediately preceding the delimiter or
1262 * the port marker (':') where the delimiter takes precedence.
1264 delim_p = memchr(name, delim, namelen);
1265 colon_p = memchr(name, ':', namelen);
1267 if (delim_p && colon_p)
1268 end = delim_p < colon_p ? delim_p : colon_p;
1269 else if (!delim_p && colon_p)
1273 if (!end) /* case: hostname:/ */
1274 end = name + namelen;
1280 /* do dns_resolve upcall */
1281 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1283 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1291 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1292 ret, ret ? "failed" : ceph_pr_addr(ss));
1297 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1298 struct sockaddr_storage *ss, char delim, const char **ipend)
1305 * Parse a server name (IP or hostname). If a valid IP address is not found
1306 * then try to extract a hostname to resolve using userspace DNS upcall.
1308 static int ceph_parse_server_name(const char *name, size_t namelen,
1309 struct sockaddr_storage *ss, char delim, const char **ipend)
1313 ret = ceph_pton(name, namelen, ss, delim, ipend);
1315 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1321 * Parse an ip[:port] list into an addr array. Use the default
1322 * monitor port if a port isn't specified.
1324 int ceph_parse_ips(const char *c, const char *end,
1325 struct ceph_entity_addr *addr,
1326 int max_count, int *count)
1328 int i, ret = -EINVAL;
1331 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1332 for (i = 0; i < max_count; i++) {
1334 struct sockaddr_storage *ss = &addr[i].in_addr;
1343 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1352 dout("missing matching ']'\n");
1359 if (p < end && *p == ':') {
1362 while (p < end && *p >= '0' && *p <= '9') {
1363 port = (port * 10) + (*p - '0');
1366 if (port > 65535 || port == 0)
1369 port = CEPH_MON_PORT;
1372 addr_set_port(ss, port);
1374 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1391 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1394 EXPORT_SYMBOL(ceph_parse_ips);
1396 static int process_banner(struct ceph_connection *con)
1398 dout("process_banner on %p\n", con);
1400 if (verify_hello(con) < 0)
1403 ceph_decode_addr(&con->actual_peer_addr);
1404 ceph_decode_addr(&con->peer_addr_for_me);
1407 * Make sure the other end is who we wanted. note that the other
1408 * end may not yet know their ip address, so if it's 0.0.0.0, give
1409 * them the benefit of the doubt.
1411 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1412 sizeof(con->peer_addr)) != 0 &&
1413 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1414 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1415 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1416 ceph_pr_addr(&con->peer_addr.in_addr),
1417 (int)le32_to_cpu(con->peer_addr.nonce),
1418 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1419 (int)le32_to_cpu(con->actual_peer_addr.nonce));
1420 con->error_msg = "wrong peer at address";
1425 * did we learn our address?
1427 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1428 int port = addr_port(&con->msgr->inst.addr.in_addr);
1430 memcpy(&con->msgr->inst.addr.in_addr,
1431 &con->peer_addr_for_me.in_addr,
1432 sizeof(con->peer_addr_for_me.in_addr));
1433 addr_set_port(&con->msgr->inst.addr.in_addr, port);
1434 encode_my_addr(con->msgr);
1435 dout("process_banner learned my addr is %s\n",
1436 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
1439 set_bit(NEGOTIATING, &con->state);
1440 prepare_read_connect(con);
1444 static void fail_protocol(struct ceph_connection *con)
1446 reset_connection(con);
1447 set_bit(CLOSED, &con->state); /* in case there's queued work */
1450 static int process_connect(struct ceph_connection *con)
1452 u64 sup_feat = con->msgr->supported_features;
1453 u64 req_feat = con->msgr->required_features;
1454 u64 server_feat = le64_to_cpu(con->in_reply.features);
1457 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1459 switch (con->in_reply.tag) {
1460 case CEPH_MSGR_TAG_FEATURES:
1461 pr_err("%s%lld %s feature set mismatch,"
1462 " my %llx < server's %llx, missing %llx\n",
1463 ENTITY_NAME(con->peer_name),
1464 ceph_pr_addr(&con->peer_addr.in_addr),
1465 sup_feat, server_feat, server_feat & ~sup_feat);
1466 con->error_msg = "missing required protocol features";
1470 case CEPH_MSGR_TAG_BADPROTOVER:
1471 pr_err("%s%lld %s protocol version mismatch,"
1472 " my %d != server's %d\n",
1473 ENTITY_NAME(con->peer_name),
1474 ceph_pr_addr(&con->peer_addr.in_addr),
1475 le32_to_cpu(con->out_connect.protocol_version),
1476 le32_to_cpu(con->in_reply.protocol_version));
1477 con->error_msg = "protocol version mismatch";
1481 case CEPH_MSGR_TAG_BADAUTHORIZER:
1483 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1485 if (con->auth_retry == 2) {
1486 con->error_msg = "connect authorization failure";
1489 con->auth_retry = 1;
1490 con_out_kvec_reset(con);
1491 ret = prepare_write_connect(con);
1494 prepare_read_connect(con);
1497 case CEPH_MSGR_TAG_RESETSESSION:
1499 * If we connected with a large connect_seq but the peer
1500 * has no record of a session with us (no connection, or
1501 * connect_seq == 0), they will send RESETSESION to indicate
1502 * that they must have reset their session, and may have
1505 dout("process_connect got RESET peer seq %u\n",
1506 le32_to_cpu(con->in_connect.connect_seq));
1507 pr_err("%s%lld %s connection reset\n",
1508 ENTITY_NAME(con->peer_name),
1509 ceph_pr_addr(&con->peer_addr.in_addr));
1510 reset_connection(con);
1511 con_out_kvec_reset(con);
1512 ret = prepare_write_connect(con);
1515 prepare_read_connect(con);
1517 /* Tell ceph about it. */
1518 mutex_unlock(&con->mutex);
1519 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
1520 if (con->ops->peer_reset)
1521 con->ops->peer_reset(con);
1522 mutex_lock(&con->mutex);
1523 if (test_bit(CLOSED, &con->state) ||
1524 test_bit(OPENING, &con->state))
1528 case CEPH_MSGR_TAG_RETRY_SESSION:
1530 * If we sent a smaller connect_seq than the peer has, try
1531 * again with a larger value.
1533 dout("process_connect got RETRY my seq = %u, peer_seq = %u\n",
1534 le32_to_cpu(con->out_connect.connect_seq),
1535 le32_to_cpu(con->in_connect.connect_seq));
1536 con->connect_seq = le32_to_cpu(con->in_connect.connect_seq);
1537 con_out_kvec_reset(con);
1538 ret = prepare_write_connect(con);
1541 prepare_read_connect(con);
1544 case CEPH_MSGR_TAG_RETRY_GLOBAL:
1546 * If we sent a smaller global_seq than the peer has, try
1547 * again with a larger value.
1549 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1550 con->peer_global_seq,
1551 le32_to_cpu(con->in_connect.global_seq));
1552 get_global_seq(con->msgr,
1553 le32_to_cpu(con->in_connect.global_seq));
1554 con_out_kvec_reset(con);
1555 ret = prepare_write_connect(con);
1558 prepare_read_connect(con);
1561 case CEPH_MSGR_TAG_READY:
1562 if (req_feat & ~server_feat) {
1563 pr_err("%s%lld %s protocol feature mismatch,"
1564 " my required %llx > server's %llx, need %llx\n",
1565 ENTITY_NAME(con->peer_name),
1566 ceph_pr_addr(&con->peer_addr.in_addr),
1567 req_feat, server_feat, req_feat & ~server_feat);
1568 con->error_msg = "missing required protocol features";
1572 clear_bit(CONNECTING, &con->state);
1573 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
1575 con->peer_features = server_feat;
1576 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1577 con->peer_global_seq,
1578 le32_to_cpu(con->in_reply.connect_seq),
1580 WARN_ON(con->connect_seq !=
1581 le32_to_cpu(con->in_reply.connect_seq));
1583 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
1584 set_bit(LOSSYTX, &con->flags);
1586 prepare_read_tag(con);
1589 case CEPH_MSGR_TAG_WAIT:
1591 * If there is a connection race (we are opening
1592 * connections to each other), one of us may just have
1593 * to WAIT. This shouldn't happen if we are the
1596 pr_err("process_connect got WAIT as client\n");
1597 con->error_msg = "protocol error, got WAIT as client";
1601 pr_err("connect protocol error, will retry\n");
1602 con->error_msg = "protocol error, garbage tag during connect";
1610 * read (part of) an ack
1612 static int read_partial_ack(struct ceph_connection *con)
1614 int size = sizeof (con->in_temp_ack);
1617 return read_partial(con, end, size, &con->in_temp_ack);
1622 * We can finally discard anything that's been acked.
1624 static void process_ack(struct ceph_connection *con)
1627 u64 ack = le64_to_cpu(con->in_temp_ack);
1630 while (!list_empty(&con->out_sent)) {
1631 m = list_first_entry(&con->out_sent, struct ceph_msg,
1633 seq = le64_to_cpu(m->hdr.seq);
1636 dout("got ack for seq %llu type %d at %p\n", seq,
1637 le16_to_cpu(m->hdr.type), m);
1638 m->ack_stamp = jiffies;
1641 prepare_read_tag(con);
1647 static int read_partial_message_section(struct ceph_connection *con,
1648 struct kvec *section,
1649 unsigned int sec_len, u32 *crc)
1655 while (section->iov_len < sec_len) {
1656 BUG_ON(section->iov_base == NULL);
1657 left = sec_len - section->iov_len;
1658 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
1659 section->iov_len, left);
1662 section->iov_len += ret;
1664 if (section->iov_len == sec_len)
1665 *crc = crc32c(0, section->iov_base, section->iov_len);
1670 static bool ceph_con_in_msg_alloc(struct ceph_connection *con,
1671 struct ceph_msg_header *hdr);
1674 static int read_partial_message_pages(struct ceph_connection *con,
1675 struct page **pages,
1676 unsigned int data_len, bool do_datacrc)
1682 left = min((int)(data_len - con->in_msg_pos.data_pos),
1683 (int)(PAGE_SIZE - con->in_msg_pos.page_pos));
1685 BUG_ON(pages == NULL);
1686 p = kmap(pages[con->in_msg_pos.page]);
1687 ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos,
1689 if (ret > 0 && do_datacrc)
1691 crc32c(con->in_data_crc,
1692 p + con->in_msg_pos.page_pos, ret);
1693 kunmap(pages[con->in_msg_pos.page]);
1696 con->in_msg_pos.data_pos += ret;
1697 con->in_msg_pos.page_pos += ret;
1698 if (con->in_msg_pos.page_pos == PAGE_SIZE) {
1699 con->in_msg_pos.page_pos = 0;
1700 con->in_msg_pos.page++;
1707 static int read_partial_message_bio(struct ceph_connection *con,
1708 struct bio **bio_iter, int *bio_seg,
1709 unsigned int data_len, bool do_datacrc)
1711 struct bio_vec *bv = bio_iovec_idx(*bio_iter, *bio_seg);
1718 left = min((int)(data_len - con->in_msg_pos.data_pos),
1719 (int)(bv->bv_len - con->in_msg_pos.page_pos));
1721 p = kmap(bv->bv_page) + bv->bv_offset;
1723 ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos,
1725 if (ret > 0 && do_datacrc)
1727 crc32c(con->in_data_crc,
1728 p + con->in_msg_pos.page_pos, ret);
1729 kunmap(bv->bv_page);
1732 con->in_msg_pos.data_pos += ret;
1733 con->in_msg_pos.page_pos += ret;
1734 if (con->in_msg_pos.page_pos == bv->bv_len) {
1735 con->in_msg_pos.page_pos = 0;
1736 iter_bio_next(bio_iter, bio_seg);
1744 * read (part of) a message.
1746 static int read_partial_message(struct ceph_connection *con)
1748 struct ceph_msg *m = con->in_msg;
1752 unsigned int front_len, middle_len, data_len;
1753 bool do_datacrc = !con->msgr->nocrc;
1757 dout("read_partial_message con %p msg %p\n", con, m);
1760 size = sizeof (con->in_hdr);
1762 ret = read_partial(con, end, size, &con->in_hdr);
1766 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
1767 if (cpu_to_le32(crc) != con->in_hdr.crc) {
1768 pr_err("read_partial_message bad hdr "
1769 " crc %u != expected %u\n",
1770 crc, con->in_hdr.crc);
1774 front_len = le32_to_cpu(con->in_hdr.front_len);
1775 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
1777 middle_len = le32_to_cpu(con->in_hdr.middle_len);
1778 if (middle_len > CEPH_MSG_MAX_DATA_LEN)
1780 data_len = le32_to_cpu(con->in_hdr.data_len);
1781 if (data_len > CEPH_MSG_MAX_DATA_LEN)
1785 seq = le64_to_cpu(con->in_hdr.seq);
1786 if ((s64)seq - (s64)con->in_seq < 1) {
1787 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1788 ENTITY_NAME(con->peer_name),
1789 ceph_pr_addr(&con->peer_addr.in_addr),
1790 seq, con->in_seq + 1);
1791 con->in_base_pos = -front_len - middle_len - data_len -
1793 con->in_tag = CEPH_MSGR_TAG_READY;
1795 } else if ((s64)seq - (s64)con->in_seq > 1) {
1796 pr_err("read_partial_message bad seq %lld expected %lld\n",
1797 seq, con->in_seq + 1);
1798 con->error_msg = "bad message sequence # for incoming message";
1802 /* allocate message? */
1804 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
1805 con->in_hdr.front_len, con->in_hdr.data_len);
1806 if (ceph_con_in_msg_alloc(con, &con->in_hdr)) {
1807 /* skip this message */
1808 dout("alloc_msg said skip message\n");
1809 BUG_ON(con->in_msg);
1810 con->in_base_pos = -front_len - middle_len - data_len -
1812 con->in_tag = CEPH_MSGR_TAG_READY;
1818 "error allocating memory for incoming message";
1822 BUG_ON(con->in_msg->con != con);
1824 m->front.iov_len = 0; /* haven't read it yet */
1826 m->middle->vec.iov_len = 0;
1828 con->in_msg_pos.page = 0;
1830 con->in_msg_pos.page_pos = m->page_alignment;
1832 con->in_msg_pos.page_pos = 0;
1833 con->in_msg_pos.data_pos = 0;
1837 ret = read_partial_message_section(con, &m->front, front_len,
1838 &con->in_front_crc);
1844 ret = read_partial_message_section(con, &m->middle->vec,
1846 &con->in_middle_crc);
1851 if (m->bio && !m->bio_iter)
1852 init_bio_iter(m->bio, &m->bio_iter, &m->bio_seg);
1856 while (con->in_msg_pos.data_pos < data_len) {
1858 ret = read_partial_message_pages(con, m->pages,
1859 data_len, do_datacrc);
1863 } else if (m->bio) {
1865 ret = read_partial_message_bio(con,
1866 &m->bio_iter, &m->bio_seg,
1867 data_len, do_datacrc);
1877 size = sizeof (m->footer);
1879 ret = read_partial(con, end, size, &m->footer);
1883 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1884 m, front_len, m->footer.front_crc, middle_len,
1885 m->footer.middle_crc, data_len, m->footer.data_crc);
1888 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
1889 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1890 m, con->in_front_crc, m->footer.front_crc);
1893 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
1894 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1895 m, con->in_middle_crc, m->footer.middle_crc);
1899 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
1900 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
1901 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
1902 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
1906 return 1; /* done! */
1910 * Process message. This happens in the worker thread. The callback should
1911 * be careful not to do anything that waits on other incoming messages or it
1914 static void process_message(struct ceph_connection *con)
1916 struct ceph_msg *msg;
1918 BUG_ON(con->in_msg->con != con);
1919 con->in_msg->con = NULL;
1924 /* if first message, set peer_name */
1925 if (con->peer_name.type == 0)
1926 con->peer_name = msg->hdr.src;
1929 mutex_unlock(&con->mutex);
1931 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
1932 msg, le64_to_cpu(msg->hdr.seq),
1933 ENTITY_NAME(msg->hdr.src),
1934 le16_to_cpu(msg->hdr.type),
1935 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
1936 le32_to_cpu(msg->hdr.front_len),
1937 le32_to_cpu(msg->hdr.data_len),
1938 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
1939 con->ops->dispatch(con, msg);
1941 mutex_lock(&con->mutex);
1942 prepare_read_tag(con);
1947 * Write something to the socket. Called in a worker thread when the
1948 * socket appears to be writeable and we have something ready to send.
1950 static int try_write(struct ceph_connection *con)
1954 dout("try_write start %p state %lu nref %d\n", con, con->state,
1955 atomic_read(&con->nref));
1958 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
1960 /* open the socket first? */
1961 if (con->sock == NULL) {
1962 clear_bit(NEGOTIATING, &con->state);
1963 set_bit(CONNECTING, &con->state);
1965 con_out_kvec_reset(con);
1966 prepare_write_banner(con);
1967 ret = prepare_write_connect(con);
1970 prepare_read_banner(con);
1972 BUG_ON(con->in_msg);
1973 con->in_tag = CEPH_MSGR_TAG_READY;
1974 dout("try_write initiating connect on %p new state %lu\n",
1976 ret = ceph_tcp_connect(con);
1978 con->error_msg = "connect error";
1984 /* kvec data queued? */
1985 if (con->out_skip) {
1986 ret = write_partial_skip(con);
1990 if (con->out_kvec_left) {
1991 ret = write_partial_kvec(con);
1998 if (con->out_msg_done) {
1999 ceph_msg_put(con->out_msg);
2000 con->out_msg = NULL; /* we're done with this one */
2004 ret = write_partial_msg_pages(con);
2006 goto more_kvec; /* we need to send the footer, too! */
2010 dout("try_write write_partial_msg_pages err %d\n",
2017 if (!test_bit(CONNECTING, &con->state)) {
2018 /* is anything else pending? */
2019 if (!list_empty(&con->out_queue)) {
2020 prepare_write_message(con);
2023 if (con->in_seq > con->in_seq_acked) {
2024 prepare_write_ack(con);
2027 if (test_and_clear_bit(KEEPALIVE_PENDING, &con->flags)) {
2028 prepare_write_keepalive(con);
2033 /* Nothing to do! */
2034 clear_bit(WRITE_PENDING, &con->flags);
2035 dout("try_write nothing else to write.\n");
2038 dout("try_write done on %p ret %d\n", con, ret);
2045 * Read what we can from the socket.
2047 static int try_read(struct ceph_connection *con)
2054 if (test_bit(STANDBY, &con->state))
2057 dout("try_read start on %p\n", con);
2060 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2064 * process_connect and process_message drop and re-take
2065 * con->mutex. make sure we handle a racing close or reopen.
2067 if (test_bit(CLOSED, &con->state) ||
2068 test_bit(OPENING, &con->state)) {
2073 if (test_bit(CONNECTING, &con->state)) {
2074 if (!test_bit(NEGOTIATING, &con->state)) {
2075 dout("try_read connecting\n");
2076 ret = read_partial_banner(con);
2079 ret = process_banner(con);
2083 ret = read_partial_connect(con);
2086 ret = process_connect(con);
2092 if (con->in_base_pos < 0) {
2094 * skipping + discarding content.
2096 * FIXME: there must be a better way to do this!
2098 static char buf[SKIP_BUF_SIZE];
2099 int skip = min((int) sizeof (buf), -con->in_base_pos);
2101 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2102 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2105 con->in_base_pos += ret;
2106 if (con->in_base_pos)
2109 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2113 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2116 dout("try_read got tag %d\n", (int)con->in_tag);
2117 switch (con->in_tag) {
2118 case CEPH_MSGR_TAG_MSG:
2119 prepare_read_message(con);
2121 case CEPH_MSGR_TAG_ACK:
2122 prepare_read_ack(con);
2124 case CEPH_MSGR_TAG_CLOSE:
2125 set_bit(CLOSED, &con->state); /* fixme */
2131 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2132 ret = read_partial_message(con);
2136 con->error_msg = "bad crc";
2140 con->error_msg = "io error";
2145 if (con->in_tag == CEPH_MSGR_TAG_READY)
2147 process_message(con);
2150 if (con->in_tag == CEPH_MSGR_TAG_ACK) {
2151 ret = read_partial_ack(con);
2159 dout("try_read done on %p ret %d\n", con, ret);
2163 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2164 con->error_msg = "protocol error, garbage tag";
2171 * Atomically queue work on a connection. Bump @con reference to
2172 * avoid races with connection teardown.
2174 static void queue_con(struct ceph_connection *con)
2176 if (!con->ops->get(con)) {
2177 dout("queue_con %p ref count 0\n", con);
2181 if (!queue_delayed_work(ceph_msgr_wq, &con->work, 0)) {
2182 dout("queue_con %p - already queued\n", con);
2185 dout("queue_con %p\n", con);
2190 * Do some work on a connection. Drop a connection ref when we're done.
2192 static void con_work(struct work_struct *work)
2194 struct ceph_connection *con = container_of(work, struct ceph_connection,
2198 mutex_lock(&con->mutex);
2200 if (test_and_clear_bit(BACKOFF, &con->flags)) {
2201 dout("con_work %p backing off\n", con);
2202 if (queue_delayed_work(ceph_msgr_wq, &con->work,
2203 round_jiffies_relative(con->delay))) {
2204 dout("con_work %p backoff %lu\n", con, con->delay);
2205 mutex_unlock(&con->mutex);
2209 dout("con_work %p FAILED to back off %lu\n", con,
2214 if (test_bit(STANDBY, &con->state)) {
2215 dout("con_work %p STANDBY\n", con);
2218 if (test_bit(CLOSED, &con->state)) { /* e.g. if we are replaced */
2219 dout("con_work CLOSED\n");
2220 con_close_socket(con);
2223 if (test_and_clear_bit(OPENING, &con->state)) {
2224 /* reopen w/ new peer */
2225 dout("con_work OPENING\n");
2226 con_close_socket(con);
2229 if (test_and_clear_bit(SOCK_CLOSED, &con->flags))
2232 ret = try_read(con);
2238 ret = try_write(con);
2245 mutex_unlock(&con->mutex);
2251 mutex_unlock(&con->mutex);
2252 ceph_fault(con); /* error/fault path */
2258 * Generic error/fault handler. A retry mechanism is used with
2259 * exponential backoff
2261 static void ceph_fault(struct ceph_connection *con)
2263 pr_err("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2264 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2265 dout("fault %p state %lu to peer %s\n",
2266 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2268 if (test_bit(LOSSYTX, &con->flags)) {
2269 dout("fault on LOSSYTX channel\n");
2273 mutex_lock(&con->mutex);
2274 if (test_bit(CLOSED, &con->state))
2277 con_close_socket(con);
2280 BUG_ON(con->in_msg->con != con);
2281 con->in_msg->con = NULL;
2282 ceph_msg_put(con->in_msg);
2287 /* Requeue anything that hasn't been acked */
2288 list_splice_init(&con->out_sent, &con->out_queue);
2290 /* If there are no messages queued or keepalive pending, place
2291 * the connection in a STANDBY state */
2292 if (list_empty(&con->out_queue) &&
2293 !test_bit(KEEPALIVE_PENDING, &con->flags)) {
2294 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2295 clear_bit(WRITE_PENDING, &con->flags);
2296 set_bit(STANDBY, &con->state);
2298 /* retry after a delay. */
2299 if (con->delay == 0)
2300 con->delay = BASE_DELAY_INTERVAL;
2301 else if (con->delay < MAX_DELAY_INTERVAL)
2304 if (queue_delayed_work(ceph_msgr_wq, &con->work,
2305 round_jiffies_relative(con->delay))) {
2306 dout("fault queued %p delay %lu\n", con, con->delay);
2309 dout("fault failed to queue %p delay %lu, backoff\n",
2312 * In many cases we see a socket state change
2313 * while con_work is running and end up
2314 * queuing (non-delayed) work, such that we
2315 * can't backoff with a delay. Set a flag so
2316 * that when con_work restarts we schedule the
2319 set_bit(BACKOFF, &con->flags);
2324 mutex_unlock(&con->mutex);
2327 * in case we faulted due to authentication, invalidate our
2328 * current tickets so that we can get new ones.
2330 if (con->auth_retry && con->ops->invalidate_authorizer) {
2331 dout("calling invalidate_authorizer()\n");
2332 con->ops->invalidate_authorizer(con);
2335 if (con->ops->fault)
2336 con->ops->fault(con);
2342 * initialize a new messenger instance
2344 void ceph_messenger_init(struct ceph_messenger *msgr,
2345 struct ceph_entity_addr *myaddr,
2346 u32 supported_features,
2347 u32 required_features,
2350 msgr->supported_features = supported_features;
2351 msgr->required_features = required_features;
2353 spin_lock_init(&msgr->global_seq_lock);
2356 msgr->inst.addr = *myaddr;
2358 /* select a random nonce */
2359 msgr->inst.addr.type = 0;
2360 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2361 encode_my_addr(msgr);
2362 msgr->nocrc = nocrc;
2364 dout("%s %p\n", __func__, msgr);
2366 EXPORT_SYMBOL(ceph_messenger_init);
2368 static void clear_standby(struct ceph_connection *con)
2370 /* come back from STANDBY? */
2371 if (test_and_clear_bit(STANDBY, &con->state)) {
2372 mutex_lock(&con->mutex);
2373 dout("clear_standby %p and ++connect_seq\n", con);
2375 WARN_ON(test_bit(WRITE_PENDING, &con->flags));
2376 WARN_ON(test_bit(KEEPALIVE_PENDING, &con->flags));
2377 mutex_unlock(&con->mutex);
2382 * Queue up an outgoing message on the given connection.
2384 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2386 if (test_bit(CLOSED, &con->state)) {
2387 dout("con_send %p closed, dropping %p\n", con, msg);
2393 msg->hdr.src = con->msgr->inst.name;
2395 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2397 msg->needs_out_seq = true;
2400 mutex_lock(&con->mutex);
2402 BUG_ON(msg->con != NULL);
2403 msg->con = con->ops->get(con);
2404 BUG_ON(msg->con == NULL);
2406 BUG_ON(!list_empty(&msg->list_head));
2407 list_add_tail(&msg->list_head, &con->out_queue);
2408 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2409 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2410 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2411 le32_to_cpu(msg->hdr.front_len),
2412 le32_to_cpu(msg->hdr.middle_len),
2413 le32_to_cpu(msg->hdr.data_len));
2414 mutex_unlock(&con->mutex);
2416 /* if there wasn't anything waiting to send before, queue
2419 if (test_and_set_bit(WRITE_PENDING, &con->flags) == 0)
2422 EXPORT_SYMBOL(ceph_con_send);
2425 * Revoke a message that was previously queued for send
2427 void ceph_msg_revoke(struct ceph_msg *msg)
2429 struct ceph_connection *con = msg->con;
2432 return; /* Message not in our possession */
2434 mutex_lock(&con->mutex);
2435 if (!list_empty(&msg->list_head)) {
2436 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
2437 list_del_init(&msg->list_head);
2438 BUG_ON(msg->con == NULL);
2439 msg->con->ops->put(msg->con);
2445 if (con->out_msg == msg) {
2446 dout("%s %p msg %p - was sending\n", __func__, con, msg);
2447 con->out_msg = NULL;
2448 if (con->out_kvec_is_msg) {
2449 con->out_skip = con->out_kvec_bytes;
2450 con->out_kvec_is_msg = false;
2456 mutex_unlock(&con->mutex);
2460 * Revoke a message that we may be reading data into
2462 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
2464 struct ceph_connection *con;
2466 BUG_ON(msg == NULL);
2468 dout("%s msg %p null con\n", __func__, msg);
2470 return; /* Message not in our possession */
2474 mutex_lock(&con->mutex);
2475 if (con->in_msg == msg) {
2476 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
2477 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
2478 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
2480 /* skip rest of message */
2481 dout("%s %p msg %p revoked\n", __func__, con, msg);
2482 con->in_base_pos = con->in_base_pos -
2483 sizeof(struct ceph_msg_header) -
2487 sizeof(struct ceph_msg_footer);
2488 ceph_msg_put(con->in_msg);
2490 con->in_tag = CEPH_MSGR_TAG_READY;
2493 dout("%s %p in_msg %p msg %p no-op\n",
2494 __func__, con, con->in_msg, msg);
2496 mutex_unlock(&con->mutex);
2500 * Queue a keepalive byte to ensure the tcp connection is alive.
2502 void ceph_con_keepalive(struct ceph_connection *con)
2504 dout("con_keepalive %p\n", con);
2506 if (test_and_set_bit(KEEPALIVE_PENDING, &con->flags) == 0 &&
2507 test_and_set_bit(WRITE_PENDING, &con->flags) == 0)
2510 EXPORT_SYMBOL(ceph_con_keepalive);
2514 * construct a new message with given type, size
2515 * the new msg has a ref count of 1.
2517 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
2522 m = kmalloc(sizeof(*m), flags);
2525 kref_init(&m->kref);
2528 INIT_LIST_HEAD(&m->list_head);
2531 m->hdr.type = cpu_to_le16(type);
2532 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
2534 m->hdr.front_len = cpu_to_le32(front_len);
2535 m->hdr.middle_len = 0;
2536 m->hdr.data_len = 0;
2537 m->hdr.data_off = 0;
2538 m->hdr.reserved = 0;
2539 m->footer.front_crc = 0;
2540 m->footer.middle_crc = 0;
2541 m->footer.data_crc = 0;
2542 m->footer.flags = 0;
2543 m->front_max = front_len;
2544 m->front_is_vmalloc = false;
2545 m->more_to_follow = false;
2554 m->page_alignment = 0;
2564 if (front_len > PAGE_CACHE_SIZE) {
2565 m->front.iov_base = __vmalloc(front_len, flags,
2567 m->front_is_vmalloc = true;
2569 m->front.iov_base = kmalloc(front_len, flags);
2571 if (m->front.iov_base == NULL) {
2572 dout("ceph_msg_new can't allocate %d bytes\n",
2577 m->front.iov_base = NULL;
2579 m->front.iov_len = front_len;
2581 dout("ceph_msg_new %p front %d\n", m, front_len);
2588 pr_err("msg_new can't create type %d front %d\n", type,
2592 dout("msg_new can't create type %d front %d\n", type,
2597 EXPORT_SYMBOL(ceph_msg_new);
2600 * Allocate "middle" portion of a message, if it is needed and wasn't
2601 * allocated by alloc_msg. This allows us to read a small fixed-size
2602 * per-type header in the front and then gracefully fail (i.e.,
2603 * propagate the error to the caller based on info in the front) when
2604 * the middle is too large.
2606 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
2608 int type = le16_to_cpu(msg->hdr.type);
2609 int middle_len = le32_to_cpu(msg->hdr.middle_len);
2611 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
2612 ceph_msg_type_name(type), middle_len);
2613 BUG_ON(!middle_len);
2614 BUG_ON(msg->middle);
2616 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
2623 * Allocate a message for receiving an incoming message on a
2624 * connection, and save the result in con->in_msg. Uses the
2625 * connection's private alloc_msg op if available.
2627 * Returns true if the message should be skipped, false otherwise.
2628 * If true is returned (skip message), con->in_msg will be NULL.
2629 * If false is returned, con->in_msg will contain a pointer to the
2630 * newly-allocated message, or NULL in case of memory exhaustion.
2632 static bool ceph_con_in_msg_alloc(struct ceph_connection *con,
2633 struct ceph_msg_header *hdr)
2635 int type = le16_to_cpu(hdr->type);
2636 int front_len = le32_to_cpu(hdr->front_len);
2637 int middle_len = le32_to_cpu(hdr->middle_len);
2640 BUG_ON(con->in_msg != NULL);
2642 if (con->ops->alloc_msg) {
2645 mutex_unlock(&con->mutex);
2646 con->in_msg = con->ops->alloc_msg(con, hdr, &skip);
2647 mutex_lock(&con->mutex);
2649 con->in_msg->con = con->ops->get(con);
2650 BUG_ON(con->in_msg->con == NULL);
2659 con->in_msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
2661 pr_err("unable to allocate msg type %d len %d\n",
2665 con->in_msg->con = con->ops->get(con);
2666 BUG_ON(con->in_msg->con == NULL);
2667 con->in_msg->page_alignment = le16_to_cpu(hdr->data_off);
2669 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2671 if (middle_len && !con->in_msg->middle) {
2672 ret = ceph_alloc_middle(con, con->in_msg);
2674 ceph_msg_put(con->in_msg);
2684 * Free a generically kmalloc'd message.
2686 void ceph_msg_kfree(struct ceph_msg *m)
2688 dout("msg_kfree %p\n", m);
2689 if (m->front_is_vmalloc)
2690 vfree(m->front.iov_base);
2692 kfree(m->front.iov_base);
2697 * Drop a msg ref. Destroy as needed.
2699 void ceph_msg_last_put(struct kref *kref)
2701 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
2703 dout("ceph_msg_put last one on %p\n", m);
2704 WARN_ON(!list_empty(&m->list_head));
2706 /* drop middle, data, if any */
2708 ceph_buffer_put(m->middle);
2715 ceph_pagelist_release(m->pagelist);
2723 ceph_msgpool_put(m->pool, m);
2727 EXPORT_SYMBOL(ceph_msg_last_put);
2729 void ceph_msg_dump(struct ceph_msg *msg)
2731 pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg,
2732 msg->front_max, msg->nr_pages);
2733 print_hex_dump(KERN_DEBUG, "header: ",
2734 DUMP_PREFIX_OFFSET, 16, 1,
2735 &msg->hdr, sizeof(msg->hdr), true);
2736 print_hex_dump(KERN_DEBUG, " front: ",
2737 DUMP_PREFIX_OFFSET, 16, 1,
2738 msg->front.iov_base, msg->front.iov_len, true);
2740 print_hex_dump(KERN_DEBUG, "middle: ",
2741 DUMP_PREFIX_OFFSET, 16, 1,
2742 msg->middle->vec.iov_base,
2743 msg->middle->vec.iov_len, true);
2744 print_hex_dump(KERN_DEBUG, "footer: ",
2745 DUMP_PREFIX_OFFSET, 16, 1,
2746 &msg->footer, sizeof(msg->footer), true);
2748 EXPORT_SYMBOL(ceph_msg_dump);