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 * When skipping (ignoring) a block of input we read it into a "skip
43 * buffer," which is this many bytes in size.
45 #define SKIP_BUF_SIZE 1024
47 static void queue_con(struct ceph_connection *con);
48 static void con_work(struct work_struct *);
49 static void ceph_fault(struct ceph_connection *con);
52 * Nicely render a sockaddr as a string. An array of formatted
53 * strings is used, to approximate reentrancy.
55 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
56 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
57 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
58 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
60 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
61 static atomic_t addr_str_seq = ATOMIC_INIT(0);
63 static struct page *zero_page; /* used in certain error cases */
65 const char *ceph_pr_addr(const struct sockaddr_storage *ss)
69 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
70 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
72 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
75 switch (ss->ss_family) {
77 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
78 ntohs(in4->sin_port));
82 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
83 ntohs(in6->sin6_port));
87 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
93 EXPORT_SYMBOL(ceph_pr_addr);
95 static void encode_my_addr(struct ceph_messenger *msgr)
97 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
98 ceph_encode_addr(&msgr->my_enc_addr);
102 * work queue for all reading and writing to/from the socket.
104 static struct workqueue_struct *ceph_msgr_wq;
106 void _ceph_msgr_exit(void)
109 destroy_workqueue(ceph_msgr_wq);
113 BUG_ON(zero_page == NULL);
115 page_cache_release(zero_page);
119 int ceph_msgr_init(void)
121 BUG_ON(zero_page != NULL);
122 zero_page = ZERO_PAGE(0);
123 page_cache_get(zero_page);
125 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0);
129 pr_err("msgr_init failed to create workqueue\n");
134 EXPORT_SYMBOL(ceph_msgr_init);
136 void ceph_msgr_exit(void)
138 BUG_ON(ceph_msgr_wq == NULL);
142 EXPORT_SYMBOL(ceph_msgr_exit);
144 void ceph_msgr_flush(void)
146 flush_workqueue(ceph_msgr_wq);
148 EXPORT_SYMBOL(ceph_msgr_flush);
152 * socket callback functions
155 /* data available on socket, or listen socket received a connect */
156 static void ceph_sock_data_ready(struct sock *sk, int count_unused)
158 struct ceph_connection *con = sk->sk_user_data;
160 if (sk->sk_state != TCP_CLOSE_WAIT) {
161 dout("%s on %p state = %lu, queueing work\n", __func__,
167 /* socket has buffer space for writing */
168 static void ceph_sock_write_space(struct sock *sk)
170 struct ceph_connection *con = sk->sk_user_data;
172 /* only queue to workqueue if there is data we want to write,
173 * and there is sufficient space in the socket buffer to accept
174 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
175 * doesn't get called again until try_write() fills the socket
176 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
177 * and net/core/stream.c:sk_stream_write_space().
179 if (test_bit(WRITE_PENDING, &con->state)) {
180 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
181 dout("%s %p queueing write work\n", __func__, con);
182 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
186 dout("%s %p nothing to write\n", __func__, con);
190 /* socket's state has changed */
191 static void ceph_sock_state_change(struct sock *sk)
193 struct ceph_connection *con = sk->sk_user_data;
195 dout("%s %p state = %lu sk_state = %u\n", __func__,
196 con, con->state, sk->sk_state);
198 if (test_bit(CLOSED, &con->state))
201 switch (sk->sk_state) {
203 dout("%s TCP_CLOSE\n", __func__);
205 dout("%s TCP_CLOSE_WAIT\n", __func__);
206 if (test_and_set_bit(SOCK_CLOSED, &con->state) == 0) {
207 if (test_bit(CONNECTING, &con->state))
208 con->error_msg = "connection failed";
210 con->error_msg = "socket closed";
214 case TCP_ESTABLISHED:
215 dout("%s TCP_ESTABLISHED\n", __func__);
218 default: /* Everything else is uninteresting */
224 * set up socket callbacks
226 static void set_sock_callbacks(struct socket *sock,
227 struct ceph_connection *con)
229 struct sock *sk = sock->sk;
230 sk->sk_user_data = con;
231 sk->sk_data_ready = ceph_sock_data_ready;
232 sk->sk_write_space = ceph_sock_write_space;
233 sk->sk_state_change = ceph_sock_state_change;
242 * initiate connection to a remote socket.
244 static int ceph_tcp_connect(struct ceph_connection *con)
246 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
251 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
255 sock->sk->sk_allocation = GFP_NOFS;
257 #ifdef CONFIG_LOCKDEP
258 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
261 set_sock_callbacks(sock, con);
263 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
265 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
267 if (ret == -EINPROGRESS) {
268 dout("connect %s EINPROGRESS sk_state = %u\n",
269 ceph_pr_addr(&con->peer_addr.in_addr),
271 } else if (ret < 0) {
272 pr_err("connect %s error %d\n",
273 ceph_pr_addr(&con->peer_addr.in_addr), ret);
275 con->error_msg = "connect error";
284 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
286 struct kvec iov = {buf, len};
287 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
290 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
297 * write something. @more is true if caller will be sending more data
300 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
301 size_t kvlen, size_t len, int more)
303 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
307 msg.msg_flags |= MSG_MORE;
309 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
311 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
317 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
318 int offset, size_t size, int more)
320 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
323 ret = kernel_sendpage(sock, page, offset, size, flags);
332 * Shutdown/close the socket for the given connection.
334 static int con_close_socket(struct ceph_connection *con)
338 dout("con_close_socket on %p sock %p\n", con, con->sock);
341 set_bit(SOCK_CLOSED, &con->state);
342 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
343 sock_release(con->sock);
345 clear_bit(SOCK_CLOSED, &con->state);
350 * Reset a connection. Discard all incoming and outgoing messages
351 * and clear *_seq state.
353 static void ceph_msg_remove(struct ceph_msg *msg)
355 list_del_init(&msg->list_head);
358 static void ceph_msg_remove_list(struct list_head *head)
360 while (!list_empty(head)) {
361 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
363 ceph_msg_remove(msg);
367 static void reset_connection(struct ceph_connection *con)
369 /* reset connection, out_queue, msg_ and connect_seq */
370 /* discard existing out_queue and msg_seq */
371 ceph_msg_remove_list(&con->out_queue);
372 ceph_msg_remove_list(&con->out_sent);
375 ceph_msg_put(con->in_msg);
379 con->connect_seq = 0;
382 ceph_msg_put(con->out_msg);
386 con->in_seq_acked = 0;
390 * mark a peer down. drop any open connections.
392 void ceph_con_close(struct ceph_connection *con)
394 dout("con_close %p peer %s\n", con,
395 ceph_pr_addr(&con->peer_addr.in_addr));
396 set_bit(CLOSED, &con->state); /* in case there's queued work */
397 clear_bit(STANDBY, &con->state); /* avoid connect_seq bump */
398 clear_bit(LOSSYTX, &con->state); /* so we retry next connect */
399 clear_bit(KEEPALIVE_PENDING, &con->state);
400 clear_bit(WRITE_PENDING, &con->state);
401 mutex_lock(&con->mutex);
402 reset_connection(con);
403 con->peer_global_seq = 0;
404 cancel_delayed_work(&con->work);
405 mutex_unlock(&con->mutex);
408 EXPORT_SYMBOL(ceph_con_close);
411 * Reopen a closed connection, with a new peer address.
413 void ceph_con_open(struct ceph_connection *con, struct ceph_entity_addr *addr)
415 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
416 set_bit(OPENING, &con->state);
417 clear_bit(CLOSED, &con->state);
418 memcpy(&con->peer_addr, addr, sizeof(*addr));
419 con->delay = 0; /* reset backoff memory */
422 EXPORT_SYMBOL(ceph_con_open);
425 * return true if this connection ever successfully opened
427 bool ceph_con_opened(struct ceph_connection *con)
429 return con->connect_seq > 0;
435 struct ceph_connection *ceph_con_get(struct ceph_connection *con)
437 int nref = __atomic_add_unless(&con->nref, 1, 0);
439 dout("con_get %p nref = %d -> %d\n", con, nref, nref + 1);
441 return nref ? con : NULL;
444 void ceph_con_put(struct ceph_connection *con)
446 int nref = atomic_dec_return(&con->nref);
453 dout("con_put %p nref = %d -> %d\n", con, nref + 1, nref);
457 * initialize a new connection.
459 void ceph_con_init(struct ceph_messenger *msgr, struct ceph_connection *con)
461 dout("con_init %p\n", con);
462 memset(con, 0, sizeof(*con));
463 atomic_set(&con->nref, 1);
465 mutex_init(&con->mutex);
466 INIT_LIST_HEAD(&con->out_queue);
467 INIT_LIST_HEAD(&con->out_sent);
468 INIT_DELAYED_WORK(&con->work, con_work);
470 EXPORT_SYMBOL(ceph_con_init);
474 * We maintain a global counter to order connection attempts. Get
475 * a unique seq greater than @gt.
477 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
481 spin_lock(&msgr->global_seq_lock);
482 if (msgr->global_seq < gt)
483 msgr->global_seq = gt;
484 ret = ++msgr->global_seq;
485 spin_unlock(&msgr->global_seq_lock);
489 static void con_out_kvec_reset(struct ceph_connection *con)
491 con->out_kvec_left = 0;
492 con->out_kvec_bytes = 0;
493 con->out_kvec_cur = &con->out_kvec[0];
496 static void con_out_kvec_add(struct ceph_connection *con,
497 size_t size, void *data)
501 index = con->out_kvec_left;
502 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
504 con->out_kvec[index].iov_len = size;
505 con->out_kvec[index].iov_base = data;
506 con->out_kvec_left++;
507 con->out_kvec_bytes += size;
511 * Prepare footer for currently outgoing message, and finish things
512 * off. Assumes out_kvec* are already valid.. we just add on to the end.
514 static void prepare_write_message_footer(struct ceph_connection *con)
516 struct ceph_msg *m = con->out_msg;
517 int v = con->out_kvec_left;
519 dout("prepare_write_message_footer %p\n", con);
520 con->out_kvec_is_msg = true;
521 con->out_kvec[v].iov_base = &m->footer;
522 con->out_kvec[v].iov_len = sizeof(m->footer);
523 con->out_kvec_bytes += sizeof(m->footer);
524 con->out_kvec_left++;
525 con->out_more = m->more_to_follow;
526 con->out_msg_done = true;
530 * Prepare headers for the next outgoing message.
532 static void prepare_write_message(struct ceph_connection *con)
537 con_out_kvec_reset(con);
538 con->out_kvec_is_msg = true;
539 con->out_msg_done = false;
541 /* Sneak an ack in there first? If we can get it into the same
542 * TCP packet that's a good thing. */
543 if (con->in_seq > con->in_seq_acked) {
544 con->in_seq_acked = con->in_seq;
545 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
546 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
547 con_out_kvec_add(con, sizeof (con->out_temp_ack),
551 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
554 /* put message on sent list */
556 list_move_tail(&m->list_head, &con->out_sent);
559 * only assign outgoing seq # if we haven't sent this message
560 * yet. if it is requeued, resend with it's original seq.
562 if (m->needs_out_seq) {
563 m->hdr.seq = cpu_to_le64(++con->out_seq);
564 m->needs_out_seq = false;
567 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n",
568 m, con->out_seq, le16_to_cpu(m->hdr.type),
569 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
570 le32_to_cpu(m->hdr.data_len),
572 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
574 /* tag + hdr + front + middle */
575 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
576 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
577 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
580 con_out_kvec_add(con, m->middle->vec.iov_len,
581 m->middle->vec.iov_base);
583 /* fill in crc (except data pages), footer */
584 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
585 con->out_msg->hdr.crc = cpu_to_le32(crc);
586 con->out_msg->footer.flags = CEPH_MSG_FOOTER_COMPLETE;
588 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
589 con->out_msg->footer.front_crc = cpu_to_le32(crc);
591 crc = crc32c(0, m->middle->vec.iov_base,
592 m->middle->vec.iov_len);
593 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
595 con->out_msg->footer.middle_crc = 0;
596 con->out_msg->footer.data_crc = 0;
597 dout("prepare_write_message front_crc %u data_crc %u\n",
598 le32_to_cpu(con->out_msg->footer.front_crc),
599 le32_to_cpu(con->out_msg->footer.middle_crc));
601 /* is there a data payload? */
602 if (le32_to_cpu(m->hdr.data_len) > 0) {
603 /* initialize page iterator */
604 con->out_msg_pos.page = 0;
606 con->out_msg_pos.page_pos = m->page_alignment;
608 con->out_msg_pos.page_pos = 0;
609 con->out_msg_pos.data_pos = 0;
610 con->out_msg_pos.did_page_crc = false;
611 con->out_more = 1; /* data + footer will follow */
613 /* no, queue up footer too and be done */
614 prepare_write_message_footer(con);
617 set_bit(WRITE_PENDING, &con->state);
623 static void prepare_write_ack(struct ceph_connection *con)
625 dout("prepare_write_ack %p %llu -> %llu\n", con,
626 con->in_seq_acked, con->in_seq);
627 con->in_seq_acked = con->in_seq;
629 con_out_kvec_reset(con);
631 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
633 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
634 con_out_kvec_add(con, sizeof (con->out_temp_ack),
637 con->out_more = 1; /* more will follow.. eventually.. */
638 set_bit(WRITE_PENDING, &con->state);
642 * Prepare to write keepalive byte.
644 static void prepare_write_keepalive(struct ceph_connection *con)
646 dout("prepare_write_keepalive %p\n", con);
647 con_out_kvec_reset(con);
648 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
649 set_bit(WRITE_PENDING, &con->state);
653 * Connection negotiation.
656 static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
659 struct ceph_auth_handshake *auth;
661 if (!con->ops->get_authorizer) {
662 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
663 con->out_connect.authorizer_len = 0;
668 /* Can't hold the mutex while getting authorizer */
670 mutex_unlock(&con->mutex);
672 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
674 mutex_lock(&con->mutex);
678 if (test_bit(CLOSED, &con->state) || test_bit(OPENING, &con->state))
679 return ERR_PTR(-EAGAIN);
681 con->auth_reply_buf = auth->authorizer_reply_buf;
682 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
689 * We connected to a peer and are saying hello.
691 static void prepare_write_banner(struct ceph_connection *con)
693 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
694 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
695 &con->msgr->my_enc_addr);
698 set_bit(WRITE_PENDING, &con->state);
701 static int prepare_write_connect(struct ceph_connection *con)
703 unsigned global_seq = get_global_seq(con->msgr, 0);
706 struct ceph_auth_handshake *auth;
708 switch (con->peer_name.type) {
709 case CEPH_ENTITY_TYPE_MON:
710 proto = CEPH_MONC_PROTOCOL;
712 case CEPH_ENTITY_TYPE_OSD:
713 proto = CEPH_OSDC_PROTOCOL;
715 case CEPH_ENTITY_TYPE_MDS:
716 proto = CEPH_MDSC_PROTOCOL;
722 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
723 con->connect_seq, global_seq, proto);
725 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
726 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
727 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
728 con->out_connect.global_seq = cpu_to_le32(global_seq);
729 con->out_connect.protocol_version = cpu_to_le32(proto);
730 con->out_connect.flags = 0;
732 auth_proto = CEPH_AUTH_UNKNOWN;
733 auth = get_connect_authorizer(con, &auth_proto);
735 return PTR_ERR(auth);
737 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
738 con->out_connect.authorizer_len = auth ?
739 cpu_to_le32(auth->authorizer_buf_len) : 0;
741 con_out_kvec_add(con, sizeof (con->out_connect),
743 if (auth && auth->authorizer_buf_len)
744 con_out_kvec_add(con, auth->authorizer_buf_len,
745 auth->authorizer_buf);
748 set_bit(WRITE_PENDING, &con->state);
754 * write as much of pending kvecs to the socket as we can.
756 * 0 -> socket full, but more to do
759 static int write_partial_kvec(struct ceph_connection *con)
763 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
764 while (con->out_kvec_bytes > 0) {
765 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
766 con->out_kvec_left, con->out_kvec_bytes,
770 con->out_kvec_bytes -= ret;
771 if (con->out_kvec_bytes == 0)
774 /* account for full iov entries consumed */
775 while (ret >= con->out_kvec_cur->iov_len) {
776 BUG_ON(!con->out_kvec_left);
777 ret -= con->out_kvec_cur->iov_len;
779 con->out_kvec_left--;
781 /* and for a partially-consumed entry */
783 con->out_kvec_cur->iov_len -= ret;
784 con->out_kvec_cur->iov_base += ret;
787 con->out_kvec_left = 0;
788 con->out_kvec_is_msg = false;
791 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
792 con->out_kvec_bytes, con->out_kvec_left, ret);
793 return ret; /* done! */
797 static void init_bio_iter(struct bio *bio, struct bio **iter, int *seg)
808 static void iter_bio_next(struct bio **bio_iter, int *seg)
810 if (*bio_iter == NULL)
813 BUG_ON(*seg >= (*bio_iter)->bi_vcnt);
816 if (*seg == (*bio_iter)->bi_vcnt)
817 init_bio_iter((*bio_iter)->bi_next, bio_iter, seg);
822 * Write as much message data payload as we can. If we finish, queue
824 * 1 -> done, footer is now queued in out_kvec[].
825 * 0 -> socket full, but more to do
828 static int write_partial_msg_pages(struct ceph_connection *con)
830 struct ceph_msg *msg = con->out_msg;
831 unsigned data_len = le32_to_cpu(msg->hdr.data_len);
833 bool do_datacrc = !con->msgr->nocrc;
837 size_t trail_len = (msg->trail ? msg->trail->length : 0);
839 dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
840 con, con->out_msg, con->out_msg_pos.page, con->out_msg->nr_pages,
841 con->out_msg_pos.page_pos);
844 if (msg->bio && !msg->bio_iter)
845 init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg);
848 while (data_len > con->out_msg_pos.data_pos) {
849 struct page *page = NULL;
850 int max_write = PAGE_SIZE;
853 total_max_write = data_len - trail_len -
854 con->out_msg_pos.data_pos;
857 * if we are calculating the data crc (the default), we need
858 * to map the page. if our pages[] has been revoked, use the
862 /* have we reached the trail part of the data? */
863 if (con->out_msg_pos.data_pos >= data_len - trail_len) {
866 total_max_write = data_len - con->out_msg_pos.data_pos;
868 page = list_first_entry(&msg->trail->head,
870 max_write = PAGE_SIZE;
871 } else if (msg->pages) {
872 page = msg->pages[con->out_msg_pos.page];
873 } else if (msg->pagelist) {
874 page = list_first_entry(&msg->pagelist->head,
877 } else if (msg->bio) {
880 bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg);
882 bio_offset = bv->bv_offset;
883 max_write = bv->bv_len;
888 len = min_t(int, max_write - con->out_msg_pos.page_pos,
891 if (do_datacrc && !con->out_msg_pos.did_page_crc) {
894 u32 tmpcrc = le32_to_cpu(con->out_msg->footer.data_crc);
898 BUG_ON(kaddr == NULL);
899 base = kaddr + con->out_msg_pos.page_pos + bio_offset;
900 crc = crc32c(tmpcrc, base, len);
901 con->out_msg->footer.data_crc = cpu_to_le32(crc);
902 con->out_msg_pos.did_page_crc = true;
904 ret = ceph_tcp_sendpage(con->sock, page,
905 con->out_msg_pos.page_pos + bio_offset,
914 con->out_msg_pos.data_pos += ret;
915 con->out_msg_pos.page_pos += ret;
917 con->out_msg_pos.page_pos = 0;
918 con->out_msg_pos.page++;
919 con->out_msg_pos.did_page_crc = false;
921 list_move_tail(&page->lru,
923 else if (msg->pagelist)
924 list_move_tail(&page->lru,
925 &msg->pagelist->head);
928 iter_bio_next(&msg->bio_iter, &msg->bio_seg);
933 dout("write_partial_msg_pages %p msg %p done\n", con, msg);
935 /* prepare and queue up footer, too */
937 con->out_msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
938 con_out_kvec_reset(con);
939 prepare_write_message_footer(con);
948 static int write_partial_skip(struct ceph_connection *con)
952 while (con->out_skip > 0) {
953 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
955 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, 1);
958 con->out_skip -= ret;
966 * Prepare to read connection handshake, or an ack.
968 static void prepare_read_banner(struct ceph_connection *con)
970 dout("prepare_read_banner %p\n", con);
971 con->in_base_pos = 0;
974 static void prepare_read_connect(struct ceph_connection *con)
976 dout("prepare_read_connect %p\n", con);
977 con->in_base_pos = 0;
980 static void prepare_read_ack(struct ceph_connection *con)
982 dout("prepare_read_ack %p\n", con);
983 con->in_base_pos = 0;
986 static void prepare_read_tag(struct ceph_connection *con)
988 dout("prepare_read_tag %p\n", con);
989 con->in_base_pos = 0;
990 con->in_tag = CEPH_MSGR_TAG_READY;
994 * Prepare to read a message.
996 static int prepare_read_message(struct ceph_connection *con)
998 dout("prepare_read_message %p\n", con);
999 BUG_ON(con->in_msg != NULL);
1000 con->in_base_pos = 0;
1001 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1006 static int read_partial(struct ceph_connection *con,
1007 int end, int size, void *object)
1009 while (con->in_base_pos < end) {
1010 int left = end - con->in_base_pos;
1011 int have = size - left;
1012 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1015 con->in_base_pos += ret;
1022 * Read all or part of the connect-side handshake on a new connection
1024 static int read_partial_banner(struct ceph_connection *con)
1030 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1033 size = strlen(CEPH_BANNER);
1035 ret = read_partial(con, end, size, con->in_banner);
1039 size = sizeof (con->actual_peer_addr);
1041 ret = read_partial(con, end, size, &con->actual_peer_addr);
1045 size = sizeof (con->peer_addr_for_me);
1047 ret = read_partial(con, end, size, &con->peer_addr_for_me);
1055 static int read_partial_connect(struct ceph_connection *con)
1061 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1063 size = sizeof (con->in_reply);
1065 ret = read_partial(con, end, size, &con->in_reply);
1069 size = le32_to_cpu(con->in_reply.authorizer_len);
1071 ret = read_partial(con, end, size, con->auth_reply_buf);
1075 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1076 con, (int)con->in_reply.tag,
1077 le32_to_cpu(con->in_reply.connect_seq),
1078 le32_to_cpu(con->in_reply.global_seq));
1085 * Verify the hello banner looks okay.
1087 static int verify_hello(struct ceph_connection *con)
1089 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1090 pr_err("connect to %s got bad banner\n",
1091 ceph_pr_addr(&con->peer_addr.in_addr));
1092 con->error_msg = "protocol error, bad banner";
1098 static bool addr_is_blank(struct sockaddr_storage *ss)
1100 switch (ss->ss_family) {
1102 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1105 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1106 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1107 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1108 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1113 static int addr_port(struct sockaddr_storage *ss)
1115 switch (ss->ss_family) {
1117 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1119 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1124 static void addr_set_port(struct sockaddr_storage *ss, int p)
1126 switch (ss->ss_family) {
1128 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1131 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1137 * Unlike other *_pton function semantics, zero indicates success.
1139 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1140 char delim, const char **ipend)
1142 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1143 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1145 memset(ss, 0, sizeof(*ss));
1147 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1148 ss->ss_family = AF_INET;
1152 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1153 ss->ss_family = AF_INET6;
1161 * Extract hostname string and resolve using kernel DNS facility.
1163 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1164 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1165 struct sockaddr_storage *ss, char delim, const char **ipend)
1167 const char *end, *delim_p;
1168 char *colon_p, *ip_addr = NULL;
1172 * The end of the hostname occurs immediately preceding the delimiter or
1173 * the port marker (':') where the delimiter takes precedence.
1175 delim_p = memchr(name, delim, namelen);
1176 colon_p = memchr(name, ':', namelen);
1178 if (delim_p && colon_p)
1179 end = delim_p < colon_p ? delim_p : colon_p;
1180 else if (!delim_p && colon_p)
1184 if (!end) /* case: hostname:/ */
1185 end = name + namelen;
1191 /* do dns_resolve upcall */
1192 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1194 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1202 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1203 ret, ret ? "failed" : ceph_pr_addr(ss));
1208 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1209 struct sockaddr_storage *ss, char delim, const char **ipend)
1216 * Parse a server name (IP or hostname). If a valid IP address is not found
1217 * then try to extract a hostname to resolve using userspace DNS upcall.
1219 static int ceph_parse_server_name(const char *name, size_t namelen,
1220 struct sockaddr_storage *ss, char delim, const char **ipend)
1224 ret = ceph_pton(name, namelen, ss, delim, ipend);
1226 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1232 * Parse an ip[:port] list into an addr array. Use the default
1233 * monitor port if a port isn't specified.
1235 int ceph_parse_ips(const char *c, const char *end,
1236 struct ceph_entity_addr *addr,
1237 int max_count, int *count)
1239 int i, ret = -EINVAL;
1242 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1243 for (i = 0; i < max_count; i++) {
1245 struct sockaddr_storage *ss = &addr[i].in_addr;
1254 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1263 dout("missing matching ']'\n");
1270 if (p < end && *p == ':') {
1273 while (p < end && *p >= '0' && *p <= '9') {
1274 port = (port * 10) + (*p - '0');
1277 if (port > 65535 || port == 0)
1280 port = CEPH_MON_PORT;
1283 addr_set_port(ss, port);
1285 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1302 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1305 EXPORT_SYMBOL(ceph_parse_ips);
1307 static int process_banner(struct ceph_connection *con)
1309 dout("process_banner on %p\n", con);
1311 if (verify_hello(con) < 0)
1314 ceph_decode_addr(&con->actual_peer_addr);
1315 ceph_decode_addr(&con->peer_addr_for_me);
1318 * Make sure the other end is who we wanted. note that the other
1319 * end may not yet know their ip address, so if it's 0.0.0.0, give
1320 * them the benefit of the doubt.
1322 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1323 sizeof(con->peer_addr)) != 0 &&
1324 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1325 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1326 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1327 ceph_pr_addr(&con->peer_addr.in_addr),
1328 (int)le32_to_cpu(con->peer_addr.nonce),
1329 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1330 (int)le32_to_cpu(con->actual_peer_addr.nonce));
1331 con->error_msg = "wrong peer at address";
1336 * did we learn our address?
1338 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1339 int port = addr_port(&con->msgr->inst.addr.in_addr);
1341 memcpy(&con->msgr->inst.addr.in_addr,
1342 &con->peer_addr_for_me.in_addr,
1343 sizeof(con->peer_addr_for_me.in_addr));
1344 addr_set_port(&con->msgr->inst.addr.in_addr, port);
1345 encode_my_addr(con->msgr);
1346 dout("process_banner learned my addr is %s\n",
1347 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
1350 set_bit(NEGOTIATING, &con->state);
1351 prepare_read_connect(con);
1355 static void fail_protocol(struct ceph_connection *con)
1357 reset_connection(con);
1358 set_bit(CLOSED, &con->state); /* in case there's queued work */
1361 static int process_connect(struct ceph_connection *con)
1363 u64 sup_feat = con->msgr->supported_features;
1364 u64 req_feat = con->msgr->required_features;
1365 u64 server_feat = le64_to_cpu(con->in_reply.features);
1368 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1370 switch (con->in_reply.tag) {
1371 case CEPH_MSGR_TAG_FEATURES:
1372 pr_err("%s%lld %s feature set mismatch,"
1373 " my %llx < server's %llx, missing %llx\n",
1374 ENTITY_NAME(con->peer_name),
1375 ceph_pr_addr(&con->peer_addr.in_addr),
1376 sup_feat, server_feat, server_feat & ~sup_feat);
1377 con->error_msg = "missing required protocol features";
1381 case CEPH_MSGR_TAG_BADPROTOVER:
1382 pr_err("%s%lld %s protocol version mismatch,"
1383 " my %d != server's %d\n",
1384 ENTITY_NAME(con->peer_name),
1385 ceph_pr_addr(&con->peer_addr.in_addr),
1386 le32_to_cpu(con->out_connect.protocol_version),
1387 le32_to_cpu(con->in_reply.protocol_version));
1388 con->error_msg = "protocol version mismatch";
1392 case CEPH_MSGR_TAG_BADAUTHORIZER:
1394 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1396 if (con->auth_retry == 2) {
1397 con->error_msg = "connect authorization failure";
1400 con->auth_retry = 1;
1401 con_out_kvec_reset(con);
1402 ret = prepare_write_connect(con);
1405 prepare_read_connect(con);
1408 case CEPH_MSGR_TAG_RESETSESSION:
1410 * If we connected with a large connect_seq but the peer
1411 * has no record of a session with us (no connection, or
1412 * connect_seq == 0), they will send RESETSESION to indicate
1413 * that they must have reset their session, and may have
1416 dout("process_connect got RESET peer seq %u\n",
1417 le32_to_cpu(con->in_connect.connect_seq));
1418 pr_err("%s%lld %s connection reset\n",
1419 ENTITY_NAME(con->peer_name),
1420 ceph_pr_addr(&con->peer_addr.in_addr));
1421 reset_connection(con);
1422 con_out_kvec_reset(con);
1423 ret = prepare_write_connect(con);
1426 prepare_read_connect(con);
1428 /* Tell ceph about it. */
1429 mutex_unlock(&con->mutex);
1430 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
1431 if (con->ops->peer_reset)
1432 con->ops->peer_reset(con);
1433 mutex_lock(&con->mutex);
1434 if (test_bit(CLOSED, &con->state) ||
1435 test_bit(OPENING, &con->state))
1439 case CEPH_MSGR_TAG_RETRY_SESSION:
1441 * If we sent a smaller connect_seq than the peer has, try
1442 * again with a larger value.
1444 dout("process_connect got RETRY my seq = %u, peer_seq = %u\n",
1445 le32_to_cpu(con->out_connect.connect_seq),
1446 le32_to_cpu(con->in_connect.connect_seq));
1447 con->connect_seq = le32_to_cpu(con->in_connect.connect_seq);
1448 con_out_kvec_reset(con);
1449 ret = prepare_write_connect(con);
1452 prepare_read_connect(con);
1455 case CEPH_MSGR_TAG_RETRY_GLOBAL:
1457 * If we sent a smaller global_seq than the peer has, try
1458 * again with a larger value.
1460 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1461 con->peer_global_seq,
1462 le32_to_cpu(con->in_connect.global_seq));
1463 get_global_seq(con->msgr,
1464 le32_to_cpu(con->in_connect.global_seq));
1465 con_out_kvec_reset(con);
1466 ret = prepare_write_connect(con);
1469 prepare_read_connect(con);
1472 case CEPH_MSGR_TAG_READY:
1473 if (req_feat & ~server_feat) {
1474 pr_err("%s%lld %s protocol feature mismatch,"
1475 " my required %llx > server's %llx, need %llx\n",
1476 ENTITY_NAME(con->peer_name),
1477 ceph_pr_addr(&con->peer_addr.in_addr),
1478 req_feat, server_feat, req_feat & ~server_feat);
1479 con->error_msg = "missing required protocol features";
1483 clear_bit(CONNECTING, &con->state);
1484 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
1486 con->peer_features = server_feat;
1487 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1488 con->peer_global_seq,
1489 le32_to_cpu(con->in_reply.connect_seq),
1491 WARN_ON(con->connect_seq !=
1492 le32_to_cpu(con->in_reply.connect_seq));
1494 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
1495 set_bit(LOSSYTX, &con->state);
1497 prepare_read_tag(con);
1500 case CEPH_MSGR_TAG_WAIT:
1502 * If there is a connection race (we are opening
1503 * connections to each other), one of us may just have
1504 * to WAIT. This shouldn't happen if we are the
1507 pr_err("process_connect got WAIT as client\n");
1508 con->error_msg = "protocol error, got WAIT as client";
1512 pr_err("connect protocol error, will retry\n");
1513 con->error_msg = "protocol error, garbage tag during connect";
1521 * read (part of) an ack
1523 static int read_partial_ack(struct ceph_connection *con)
1525 int size = sizeof (con->in_temp_ack);
1528 return read_partial(con, end, size, &con->in_temp_ack);
1533 * We can finally discard anything that's been acked.
1535 static void process_ack(struct ceph_connection *con)
1538 u64 ack = le64_to_cpu(con->in_temp_ack);
1541 while (!list_empty(&con->out_sent)) {
1542 m = list_first_entry(&con->out_sent, struct ceph_msg,
1544 seq = le64_to_cpu(m->hdr.seq);
1547 dout("got ack for seq %llu type %d at %p\n", seq,
1548 le16_to_cpu(m->hdr.type), m);
1549 m->ack_stamp = jiffies;
1552 prepare_read_tag(con);
1558 static int read_partial_message_section(struct ceph_connection *con,
1559 struct kvec *section,
1560 unsigned int sec_len, u32 *crc)
1566 while (section->iov_len < sec_len) {
1567 BUG_ON(section->iov_base == NULL);
1568 left = sec_len - section->iov_len;
1569 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
1570 section->iov_len, left);
1573 section->iov_len += ret;
1575 if (section->iov_len == sec_len)
1576 *crc = crc32c(0, section->iov_base, section->iov_len);
1581 static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con,
1582 struct ceph_msg_header *hdr,
1586 static int read_partial_message_pages(struct ceph_connection *con,
1587 struct page **pages,
1588 unsigned data_len, bool do_datacrc)
1594 left = min((int)(data_len - con->in_msg_pos.data_pos),
1595 (int)(PAGE_SIZE - con->in_msg_pos.page_pos));
1597 BUG_ON(pages == NULL);
1598 p = kmap(pages[con->in_msg_pos.page]);
1599 ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos,
1601 if (ret > 0 && do_datacrc)
1603 crc32c(con->in_data_crc,
1604 p + con->in_msg_pos.page_pos, ret);
1605 kunmap(pages[con->in_msg_pos.page]);
1608 con->in_msg_pos.data_pos += ret;
1609 con->in_msg_pos.page_pos += ret;
1610 if (con->in_msg_pos.page_pos == PAGE_SIZE) {
1611 con->in_msg_pos.page_pos = 0;
1612 con->in_msg_pos.page++;
1619 static int read_partial_message_bio(struct ceph_connection *con,
1620 struct bio **bio_iter, int *bio_seg,
1621 unsigned data_len, bool do_datacrc)
1623 struct bio_vec *bv = bio_iovec_idx(*bio_iter, *bio_seg);
1630 left = min((int)(data_len - con->in_msg_pos.data_pos),
1631 (int)(bv->bv_len - con->in_msg_pos.page_pos));
1633 p = kmap(bv->bv_page) + bv->bv_offset;
1635 ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos,
1637 if (ret > 0 && do_datacrc)
1639 crc32c(con->in_data_crc,
1640 p + con->in_msg_pos.page_pos, ret);
1641 kunmap(bv->bv_page);
1644 con->in_msg_pos.data_pos += ret;
1645 con->in_msg_pos.page_pos += ret;
1646 if (con->in_msg_pos.page_pos == bv->bv_len) {
1647 con->in_msg_pos.page_pos = 0;
1648 iter_bio_next(bio_iter, bio_seg);
1656 * read (part of) a message.
1658 static int read_partial_message(struct ceph_connection *con)
1660 struct ceph_msg *m = con->in_msg;
1664 unsigned front_len, middle_len, data_len;
1665 bool do_datacrc = !con->msgr->nocrc;
1670 dout("read_partial_message con %p msg %p\n", con, m);
1673 size = sizeof (con->in_hdr);
1675 ret = read_partial(con, end, size, &con->in_hdr);
1679 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
1680 if (cpu_to_le32(crc) != con->in_hdr.crc) {
1681 pr_err("read_partial_message bad hdr "
1682 " crc %u != expected %u\n",
1683 crc, con->in_hdr.crc);
1687 front_len = le32_to_cpu(con->in_hdr.front_len);
1688 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
1690 middle_len = le32_to_cpu(con->in_hdr.middle_len);
1691 if (middle_len > CEPH_MSG_MAX_DATA_LEN)
1693 data_len = le32_to_cpu(con->in_hdr.data_len);
1694 if (data_len > CEPH_MSG_MAX_DATA_LEN)
1698 seq = le64_to_cpu(con->in_hdr.seq);
1699 if ((s64)seq - (s64)con->in_seq < 1) {
1700 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1701 ENTITY_NAME(con->peer_name),
1702 ceph_pr_addr(&con->peer_addr.in_addr),
1703 seq, con->in_seq + 1);
1704 con->in_base_pos = -front_len - middle_len - data_len -
1706 con->in_tag = CEPH_MSGR_TAG_READY;
1708 } else if ((s64)seq - (s64)con->in_seq > 1) {
1709 pr_err("read_partial_message bad seq %lld expected %lld\n",
1710 seq, con->in_seq + 1);
1711 con->error_msg = "bad message sequence # for incoming message";
1715 /* allocate message? */
1717 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
1718 con->in_hdr.front_len, con->in_hdr.data_len);
1720 con->in_msg = ceph_alloc_msg(con, &con->in_hdr, &skip);
1722 /* skip this message */
1723 dout("alloc_msg said skip message\n");
1724 BUG_ON(con->in_msg);
1725 con->in_base_pos = -front_len - middle_len - data_len -
1727 con->in_tag = CEPH_MSGR_TAG_READY;
1733 "error allocating memory for incoming message";
1737 m->front.iov_len = 0; /* haven't read it yet */
1739 m->middle->vec.iov_len = 0;
1741 con->in_msg_pos.page = 0;
1743 con->in_msg_pos.page_pos = m->page_alignment;
1745 con->in_msg_pos.page_pos = 0;
1746 con->in_msg_pos.data_pos = 0;
1750 ret = read_partial_message_section(con, &m->front, front_len,
1751 &con->in_front_crc);
1757 ret = read_partial_message_section(con, &m->middle->vec,
1759 &con->in_middle_crc);
1764 if (m->bio && !m->bio_iter)
1765 init_bio_iter(m->bio, &m->bio_iter, &m->bio_seg);
1769 while (con->in_msg_pos.data_pos < data_len) {
1771 ret = read_partial_message_pages(con, m->pages,
1772 data_len, do_datacrc);
1776 } else if (m->bio) {
1778 ret = read_partial_message_bio(con,
1779 &m->bio_iter, &m->bio_seg,
1780 data_len, do_datacrc);
1790 size = sizeof (m->footer);
1792 ret = read_partial(con, end, size, &m->footer);
1796 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1797 m, front_len, m->footer.front_crc, middle_len,
1798 m->footer.middle_crc, data_len, m->footer.data_crc);
1801 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
1802 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1803 m, con->in_front_crc, m->footer.front_crc);
1806 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
1807 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1808 m, con->in_middle_crc, m->footer.middle_crc);
1812 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
1813 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
1814 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
1815 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
1819 return 1; /* done! */
1823 * Process message. This happens in the worker thread. The callback should
1824 * be careful not to do anything that waits on other incoming messages or it
1827 static void process_message(struct ceph_connection *con)
1829 struct ceph_msg *msg;
1834 /* if first message, set peer_name */
1835 if (con->peer_name.type == 0)
1836 con->peer_name = msg->hdr.src;
1839 mutex_unlock(&con->mutex);
1841 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
1842 msg, le64_to_cpu(msg->hdr.seq),
1843 ENTITY_NAME(msg->hdr.src),
1844 le16_to_cpu(msg->hdr.type),
1845 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
1846 le32_to_cpu(msg->hdr.front_len),
1847 le32_to_cpu(msg->hdr.data_len),
1848 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
1849 con->ops->dispatch(con, msg);
1851 mutex_lock(&con->mutex);
1852 prepare_read_tag(con);
1857 * Write something to the socket. Called in a worker thread when the
1858 * socket appears to be writeable and we have something ready to send.
1860 static int try_write(struct ceph_connection *con)
1864 dout("try_write start %p state %lu nref %d\n", con, con->state,
1865 atomic_read(&con->nref));
1868 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
1870 /* open the socket first? */
1871 if (con->sock == NULL) {
1872 con_out_kvec_reset(con);
1873 prepare_write_banner(con);
1874 ret = prepare_write_connect(con);
1877 prepare_read_banner(con);
1878 set_bit(CONNECTING, &con->state);
1879 clear_bit(NEGOTIATING, &con->state);
1881 BUG_ON(con->in_msg);
1882 con->in_tag = CEPH_MSGR_TAG_READY;
1883 dout("try_write initiating connect on %p new state %lu\n",
1885 ret = ceph_tcp_connect(con);
1887 con->error_msg = "connect error";
1893 /* kvec data queued? */
1894 if (con->out_skip) {
1895 ret = write_partial_skip(con);
1899 if (con->out_kvec_left) {
1900 ret = write_partial_kvec(con);
1907 if (con->out_msg_done) {
1908 ceph_msg_put(con->out_msg);
1909 con->out_msg = NULL; /* we're done with this one */
1913 ret = write_partial_msg_pages(con);
1915 goto more_kvec; /* we need to send the footer, too! */
1919 dout("try_write write_partial_msg_pages err %d\n",
1926 if (!test_bit(CONNECTING, &con->state)) {
1927 /* is anything else pending? */
1928 if (!list_empty(&con->out_queue)) {
1929 prepare_write_message(con);
1932 if (con->in_seq > con->in_seq_acked) {
1933 prepare_write_ack(con);
1936 if (test_and_clear_bit(KEEPALIVE_PENDING, &con->state)) {
1937 prepare_write_keepalive(con);
1942 /* Nothing to do! */
1943 clear_bit(WRITE_PENDING, &con->state);
1944 dout("try_write nothing else to write.\n");
1947 dout("try_write done on %p ret %d\n", con, ret);
1954 * Read what we can from the socket.
1956 static int try_read(struct ceph_connection *con)
1963 if (test_bit(STANDBY, &con->state))
1966 dout("try_read start on %p\n", con);
1969 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
1973 * process_connect and process_message drop and re-take
1974 * con->mutex. make sure we handle a racing close or reopen.
1976 if (test_bit(CLOSED, &con->state) ||
1977 test_bit(OPENING, &con->state)) {
1982 if (test_bit(CONNECTING, &con->state)) {
1983 if (!test_bit(NEGOTIATING, &con->state)) {
1984 dout("try_read connecting\n");
1985 ret = read_partial_banner(con);
1988 ret = process_banner(con);
1992 ret = read_partial_connect(con);
1995 ret = process_connect(con);
2001 if (con->in_base_pos < 0) {
2003 * skipping + discarding content.
2005 * FIXME: there must be a better way to do this!
2007 static char buf[SKIP_BUF_SIZE];
2008 int skip = min((int) sizeof (buf), -con->in_base_pos);
2010 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2011 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2014 con->in_base_pos += ret;
2015 if (con->in_base_pos)
2018 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2022 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2025 dout("try_read got tag %d\n", (int)con->in_tag);
2026 switch (con->in_tag) {
2027 case CEPH_MSGR_TAG_MSG:
2028 prepare_read_message(con);
2030 case CEPH_MSGR_TAG_ACK:
2031 prepare_read_ack(con);
2033 case CEPH_MSGR_TAG_CLOSE:
2034 set_bit(CLOSED, &con->state); /* fixme */
2040 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2041 ret = read_partial_message(con);
2045 con->error_msg = "bad crc";
2049 con->error_msg = "io error";
2054 if (con->in_tag == CEPH_MSGR_TAG_READY)
2056 process_message(con);
2059 if (con->in_tag == CEPH_MSGR_TAG_ACK) {
2060 ret = read_partial_ack(con);
2068 dout("try_read done on %p ret %d\n", con, ret);
2072 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2073 con->error_msg = "protocol error, garbage tag";
2080 * Atomically queue work on a connection. Bump @con reference to
2081 * avoid races with connection teardown.
2083 static void queue_con(struct ceph_connection *con)
2085 if (!con->ops->get(con)) {
2086 dout("queue_con %p ref count 0\n", con);
2090 if (!queue_delayed_work(ceph_msgr_wq, &con->work, 0)) {
2091 dout("queue_con %p - already queued\n", con);
2094 dout("queue_con %p\n", con);
2099 * Do some work on a connection. Drop a connection ref when we're done.
2101 static void con_work(struct work_struct *work)
2103 struct ceph_connection *con = container_of(work, struct ceph_connection,
2107 mutex_lock(&con->mutex);
2109 if (test_and_clear_bit(BACKOFF, &con->state)) {
2110 dout("con_work %p backing off\n", con);
2111 if (queue_delayed_work(ceph_msgr_wq, &con->work,
2112 round_jiffies_relative(con->delay))) {
2113 dout("con_work %p backoff %lu\n", con, con->delay);
2114 mutex_unlock(&con->mutex);
2118 dout("con_work %p FAILED to back off %lu\n", con,
2123 if (test_bit(STANDBY, &con->state)) {
2124 dout("con_work %p STANDBY\n", con);
2127 if (test_bit(CLOSED, &con->state)) { /* e.g. if we are replaced */
2128 dout("con_work CLOSED\n");
2129 con_close_socket(con);
2132 if (test_and_clear_bit(OPENING, &con->state)) {
2133 /* reopen w/ new peer */
2134 dout("con_work OPENING\n");
2135 con_close_socket(con);
2138 if (test_and_clear_bit(SOCK_CLOSED, &con->state))
2141 ret = try_read(con);
2147 ret = try_write(con);
2154 mutex_unlock(&con->mutex);
2160 mutex_unlock(&con->mutex);
2161 ceph_fault(con); /* error/fault path */
2167 * Generic error/fault handler. A retry mechanism is used with
2168 * exponential backoff
2170 static void ceph_fault(struct ceph_connection *con)
2172 pr_err("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2173 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2174 dout("fault %p state %lu to peer %s\n",
2175 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2177 if (test_bit(LOSSYTX, &con->state)) {
2178 dout("fault on LOSSYTX channel\n");
2182 mutex_lock(&con->mutex);
2183 if (test_bit(CLOSED, &con->state))
2186 con_close_socket(con);
2189 ceph_msg_put(con->in_msg);
2193 /* Requeue anything that hasn't been acked */
2194 list_splice_init(&con->out_sent, &con->out_queue);
2196 /* If there are no messages queued or keepalive pending, place
2197 * the connection in a STANDBY state */
2198 if (list_empty(&con->out_queue) &&
2199 !test_bit(KEEPALIVE_PENDING, &con->state)) {
2200 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2201 clear_bit(WRITE_PENDING, &con->state);
2202 set_bit(STANDBY, &con->state);
2204 /* retry after a delay. */
2205 if (con->delay == 0)
2206 con->delay = BASE_DELAY_INTERVAL;
2207 else if (con->delay < MAX_DELAY_INTERVAL)
2210 if (queue_delayed_work(ceph_msgr_wq, &con->work,
2211 round_jiffies_relative(con->delay))) {
2212 dout("fault queued %p delay %lu\n", con, con->delay);
2215 dout("fault failed to queue %p delay %lu, backoff\n",
2218 * In many cases we see a socket state change
2219 * while con_work is running and end up
2220 * queuing (non-delayed) work, such that we
2221 * can't backoff with a delay. Set a flag so
2222 * that when con_work restarts we schedule the
2225 set_bit(BACKOFF, &con->state);
2230 mutex_unlock(&con->mutex);
2233 * in case we faulted due to authentication, invalidate our
2234 * current tickets so that we can get new ones.
2236 if (con->auth_retry && con->ops->invalidate_authorizer) {
2237 dout("calling invalidate_authorizer()\n");
2238 con->ops->invalidate_authorizer(con);
2241 if (con->ops->fault)
2242 con->ops->fault(con);
2248 * create a new messenger instance
2250 struct ceph_messenger *ceph_messenger_create(struct ceph_entity_addr *myaddr,
2251 u32 supported_features,
2252 u32 required_features)
2254 struct ceph_messenger *msgr;
2256 msgr = kzalloc(sizeof(*msgr), GFP_KERNEL);
2258 return ERR_PTR(-ENOMEM);
2260 msgr->supported_features = supported_features;
2261 msgr->required_features = required_features;
2263 spin_lock_init(&msgr->global_seq_lock);
2266 msgr->inst.addr = *myaddr;
2268 /* select a random nonce */
2269 msgr->inst.addr.type = 0;
2270 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2271 encode_my_addr(msgr);
2273 dout("messenger_create %p\n", msgr);
2276 EXPORT_SYMBOL(ceph_messenger_create);
2278 void ceph_messenger_destroy(struct ceph_messenger *msgr)
2280 dout("destroy %p\n", msgr);
2282 dout("destroyed messenger %p\n", msgr);
2284 EXPORT_SYMBOL(ceph_messenger_destroy);
2286 static void clear_standby(struct ceph_connection *con)
2288 /* come back from STANDBY? */
2289 if (test_and_clear_bit(STANDBY, &con->state)) {
2290 mutex_lock(&con->mutex);
2291 dout("clear_standby %p and ++connect_seq\n", con);
2293 WARN_ON(test_bit(WRITE_PENDING, &con->state));
2294 WARN_ON(test_bit(KEEPALIVE_PENDING, &con->state));
2295 mutex_unlock(&con->mutex);
2300 * Queue up an outgoing message on the given connection.
2302 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2304 if (test_bit(CLOSED, &con->state)) {
2305 dout("con_send %p closed, dropping %p\n", con, msg);
2311 msg->hdr.src = con->msgr->inst.name;
2313 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2315 msg->needs_out_seq = true;
2318 mutex_lock(&con->mutex);
2319 BUG_ON(!list_empty(&msg->list_head));
2320 list_add_tail(&msg->list_head, &con->out_queue);
2321 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2322 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2323 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2324 le32_to_cpu(msg->hdr.front_len),
2325 le32_to_cpu(msg->hdr.middle_len),
2326 le32_to_cpu(msg->hdr.data_len));
2327 mutex_unlock(&con->mutex);
2329 /* if there wasn't anything waiting to send before, queue
2332 if (test_and_set_bit(WRITE_PENDING, &con->state) == 0)
2335 EXPORT_SYMBOL(ceph_con_send);
2338 * Revoke a message that was previously queued for send
2340 void ceph_con_revoke(struct ceph_connection *con, struct ceph_msg *msg)
2342 mutex_lock(&con->mutex);
2343 if (!list_empty(&msg->list_head)) {
2344 dout("con_revoke %p msg %p - was on queue\n", con, msg);
2345 list_del_init(&msg->list_head);
2349 if (con->out_msg == msg) {
2350 dout("con_revoke %p msg %p - was sending\n", con, msg);
2351 con->out_msg = NULL;
2352 if (con->out_kvec_is_msg) {
2353 con->out_skip = con->out_kvec_bytes;
2354 con->out_kvec_is_msg = false;
2359 mutex_unlock(&con->mutex);
2363 * Revoke a message that we may be reading data into
2365 void ceph_con_revoke_message(struct ceph_connection *con, struct ceph_msg *msg)
2367 mutex_lock(&con->mutex);
2368 if (con->in_msg && con->in_msg == msg) {
2369 unsigned front_len = le32_to_cpu(con->in_hdr.front_len);
2370 unsigned middle_len = le32_to_cpu(con->in_hdr.middle_len);
2371 unsigned data_len = le32_to_cpu(con->in_hdr.data_len);
2373 /* skip rest of message */
2374 dout("con_revoke_pages %p msg %p revoked\n", con, msg);
2375 con->in_base_pos = con->in_base_pos -
2376 sizeof(struct ceph_msg_header) -
2380 sizeof(struct ceph_msg_footer);
2381 ceph_msg_put(con->in_msg);
2383 con->in_tag = CEPH_MSGR_TAG_READY;
2386 dout("con_revoke_pages %p msg %p pages %p no-op\n",
2387 con, con->in_msg, msg);
2389 mutex_unlock(&con->mutex);
2393 * Queue a keepalive byte to ensure the tcp connection is alive.
2395 void ceph_con_keepalive(struct ceph_connection *con)
2397 dout("con_keepalive %p\n", con);
2399 if (test_and_set_bit(KEEPALIVE_PENDING, &con->state) == 0 &&
2400 test_and_set_bit(WRITE_PENDING, &con->state) == 0)
2403 EXPORT_SYMBOL(ceph_con_keepalive);
2407 * construct a new message with given type, size
2408 * the new msg has a ref count of 1.
2410 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
2415 m = kmalloc(sizeof(*m), flags);
2418 kref_init(&m->kref);
2419 INIT_LIST_HEAD(&m->list_head);
2422 m->hdr.type = cpu_to_le16(type);
2423 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
2425 m->hdr.front_len = cpu_to_le32(front_len);
2426 m->hdr.middle_len = 0;
2427 m->hdr.data_len = 0;
2428 m->hdr.data_off = 0;
2429 m->hdr.reserved = 0;
2430 m->footer.front_crc = 0;
2431 m->footer.middle_crc = 0;
2432 m->footer.data_crc = 0;
2433 m->footer.flags = 0;
2434 m->front_max = front_len;
2435 m->front_is_vmalloc = false;
2436 m->more_to_follow = false;
2445 m->page_alignment = 0;
2455 if (front_len > PAGE_CACHE_SIZE) {
2456 m->front.iov_base = __vmalloc(front_len, flags,
2458 m->front_is_vmalloc = true;
2460 m->front.iov_base = kmalloc(front_len, flags);
2462 if (m->front.iov_base == NULL) {
2463 dout("ceph_msg_new can't allocate %d bytes\n",
2468 m->front.iov_base = NULL;
2470 m->front.iov_len = front_len;
2472 dout("ceph_msg_new %p front %d\n", m, front_len);
2479 pr_err("msg_new can't create type %d front %d\n", type,
2483 dout("msg_new can't create type %d front %d\n", type,
2488 EXPORT_SYMBOL(ceph_msg_new);
2491 * Allocate "middle" portion of a message, if it is needed and wasn't
2492 * allocated by alloc_msg. This allows us to read a small fixed-size
2493 * per-type header in the front and then gracefully fail (i.e.,
2494 * propagate the error to the caller based on info in the front) when
2495 * the middle is too large.
2497 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
2499 int type = le16_to_cpu(msg->hdr.type);
2500 int middle_len = le32_to_cpu(msg->hdr.middle_len);
2502 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
2503 ceph_msg_type_name(type), middle_len);
2504 BUG_ON(!middle_len);
2505 BUG_ON(msg->middle);
2507 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
2514 * Generic message allocator, for incoming messages.
2516 static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con,
2517 struct ceph_msg_header *hdr,
2520 int type = le16_to_cpu(hdr->type);
2521 int front_len = le32_to_cpu(hdr->front_len);
2522 int middle_len = le32_to_cpu(hdr->middle_len);
2523 struct ceph_msg *msg = NULL;
2526 if (con->ops->alloc_msg) {
2527 mutex_unlock(&con->mutex);
2528 msg = con->ops->alloc_msg(con, hdr, skip);
2529 mutex_lock(&con->mutex);
2535 msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
2537 pr_err("unable to allocate msg type %d len %d\n",
2541 msg->page_alignment = le16_to_cpu(hdr->data_off);
2543 memcpy(&msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2545 if (middle_len && !msg->middle) {
2546 ret = ceph_alloc_middle(con, msg);
2558 * Free a generically kmalloc'd message.
2560 void ceph_msg_kfree(struct ceph_msg *m)
2562 dout("msg_kfree %p\n", m);
2563 if (m->front_is_vmalloc)
2564 vfree(m->front.iov_base);
2566 kfree(m->front.iov_base);
2571 * Drop a msg ref. Destroy as needed.
2573 void ceph_msg_last_put(struct kref *kref)
2575 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
2577 dout("ceph_msg_put last one on %p\n", m);
2578 WARN_ON(!list_empty(&m->list_head));
2580 /* drop middle, data, if any */
2582 ceph_buffer_put(m->middle);
2589 ceph_pagelist_release(m->pagelist);
2597 ceph_msgpool_put(m->pool, m);
2601 EXPORT_SYMBOL(ceph_msg_last_put);
2603 void ceph_msg_dump(struct ceph_msg *msg)
2605 pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg,
2606 msg->front_max, msg->nr_pages);
2607 print_hex_dump(KERN_DEBUG, "header: ",
2608 DUMP_PREFIX_OFFSET, 16, 1,
2609 &msg->hdr, sizeof(msg->hdr), true);
2610 print_hex_dump(KERN_DEBUG, " front: ",
2611 DUMP_PREFIX_OFFSET, 16, 1,
2612 msg->front.iov_base, msg->front.iov_len, true);
2614 print_hex_dump(KERN_DEBUG, "middle: ",
2615 DUMP_PREFIX_OFFSET, 16, 1,
2616 msg->middle->vec.iov_base,
2617 msg->middle->vec.iov_len, true);
2618 print_hex_dump(KERN_DEBUG, "footer: ",
2619 DUMP_PREFIX_OFFSET, 16, 1,
2620 &msg->footer, sizeof(msg->footer), true);
2622 EXPORT_SYMBOL(ceph_msg_dump);