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_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("ceph_data_ready on %p state = %lu, queueing work\n",
167 /* socket has buffer space for writing */
168 static void ceph_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_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("ceph_write_space %p queueing write work\n", con);
182 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
186 dout("ceph_write_space %p nothing to write\n", con);
190 /* socket's state has changed */
191 static void ceph_state_change(struct sock *sk)
193 struct ceph_connection *con = sk->sk_user_data;
195 dout("ceph_state_change %p state = %lu sk_state = %u\n",
196 con, con->state, sk->sk_state);
198 if (test_bit(CLOSED, &con->state))
201 switch (sk->sk_state) {
203 dout("ceph_state_change TCP_CLOSE\n");
205 dout("ceph_state_change TCP_CLOSE_WAIT\n");
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("ceph_state_change TCP_ESTABLISHED\n");
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_data_ready;
232 sk->sk_write_space = ceph_write_space;
233 sk->sk_state_change = ceph_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 ceph_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 ceph_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 ceph_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 ceph_con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
546 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
547 ceph_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 ceph_con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
576 ceph_con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
577 ceph_con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
580 ceph_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 ceph_con_out_kvec_reset(con);
631 ceph_con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
633 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
634 ceph_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 ceph_con_out_kvec_reset(con);
648 ceph_con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
649 set_bit(WRITE_PENDING, &con->state);
653 * Connection negotiation.
656 static int prepare_connect_authorizer(struct ceph_connection *con)
660 int auth_protocol = 0;
662 mutex_unlock(&con->mutex);
663 if (con->ops->get_authorizer)
664 con->ops->get_authorizer(con, &auth_buf, &auth_len,
665 &auth_protocol, &con->auth_reply_buf,
666 &con->auth_reply_buf_len,
668 mutex_lock(&con->mutex);
670 if (test_bit(CLOSED, &con->state) ||
671 test_bit(OPENING, &con->state))
674 con->out_connect.authorizer_protocol = cpu_to_le32(auth_protocol);
675 con->out_connect.authorizer_len = cpu_to_le32(auth_len);
678 ceph_con_out_kvec_add(con, auth_len, auth_buf);
684 * We connected to a peer and are saying hello.
686 static void prepare_write_banner(struct ceph_messenger *msgr,
687 struct ceph_connection *con)
689 ceph_con_out_kvec_reset(con);
690 ceph_con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
691 ceph_con_out_kvec_add(con, sizeof (msgr->my_enc_addr),
695 set_bit(WRITE_PENDING, &con->state);
698 static int prepare_write_connect(struct ceph_messenger *msgr,
699 struct ceph_connection *con,
702 unsigned int global_seq = get_global_seq(con->msgr, 0);
705 switch (con->peer_name.type) {
706 case CEPH_ENTITY_TYPE_MON:
707 proto = CEPH_MONC_PROTOCOL;
709 case CEPH_ENTITY_TYPE_OSD:
710 proto = CEPH_OSDC_PROTOCOL;
712 case CEPH_ENTITY_TYPE_MDS:
713 proto = CEPH_MDSC_PROTOCOL;
719 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
720 con->connect_seq, global_seq, proto);
722 con->out_connect.features = cpu_to_le64(msgr->supported_features);
723 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
724 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
725 con->out_connect.global_seq = cpu_to_le32(global_seq);
726 con->out_connect.protocol_version = cpu_to_le32(proto);
727 con->out_connect.flags = 0;
730 prepare_write_banner(msgr, con);
732 ceph_con_out_kvec_reset(con);
733 ceph_con_out_kvec_add(con, sizeof (con->out_connect), &con->out_connect);
736 set_bit(WRITE_PENDING, &con->state);
738 return prepare_connect_authorizer(con);
742 * write as much of pending kvecs to the socket as we can.
744 * 0 -> socket full, but more to do
747 static int write_partial_kvec(struct ceph_connection *con)
751 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
752 while (con->out_kvec_bytes > 0) {
753 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
754 con->out_kvec_left, con->out_kvec_bytes,
758 con->out_kvec_bytes -= ret;
759 if (con->out_kvec_bytes == 0)
762 /* account for full iov entries consumed */
763 while (ret >= con->out_kvec_cur->iov_len) {
764 BUG_ON(!con->out_kvec_left);
765 ret -= con->out_kvec_cur->iov_len;
767 con->out_kvec_left--;
769 /* and for a partially-consumed entry */
771 con->out_kvec_cur->iov_len -= ret;
772 con->out_kvec_cur->iov_base += ret;
775 con->out_kvec_left = 0;
776 con->out_kvec_is_msg = false;
779 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
780 con->out_kvec_bytes, con->out_kvec_left, ret);
781 return ret; /* done! */
785 static void init_bio_iter(struct bio *bio, struct bio **iter, int *seg)
796 static void iter_bio_next(struct bio **bio_iter, int *seg)
798 if (*bio_iter == NULL)
801 BUG_ON(*seg >= (*bio_iter)->bi_vcnt);
804 if (*seg == (*bio_iter)->bi_vcnt)
805 init_bio_iter((*bio_iter)->bi_next, bio_iter, seg);
810 * Write as much message data payload as we can. If we finish, queue
812 * 1 -> done, footer is now queued in out_kvec[].
813 * 0 -> socket full, but more to do
816 static int write_partial_msg_pages(struct ceph_connection *con)
818 struct ceph_msg *msg = con->out_msg;
819 unsigned int data_len = le32_to_cpu(msg->hdr.data_len);
821 bool do_datacrc = !con->msgr->nocrc;
825 size_t trail_len = (msg->trail ? msg->trail->length : 0);
827 dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
828 con, con->out_msg, con->out_msg_pos.page, con->out_msg->nr_pages,
829 con->out_msg_pos.page_pos);
832 if (msg->bio && !msg->bio_iter)
833 init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg);
836 while (data_len > con->out_msg_pos.data_pos) {
837 struct page *page = NULL;
838 int max_write = PAGE_SIZE;
841 total_max_write = data_len - trail_len -
842 con->out_msg_pos.data_pos;
845 * if we are calculating the data crc (the default), we need
846 * to map the page. if our pages[] has been revoked, use the
850 /* have we reached the trail part of the data? */
851 if (con->out_msg_pos.data_pos >= data_len - trail_len) {
854 total_max_write = data_len - con->out_msg_pos.data_pos;
856 page = list_first_entry(&msg->trail->head,
858 max_write = PAGE_SIZE;
859 } else if (msg->pages) {
860 page = msg->pages[con->out_msg_pos.page];
861 } else if (msg->pagelist) {
862 page = list_first_entry(&msg->pagelist->head,
865 } else if (msg->bio) {
868 bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg);
870 bio_offset = bv->bv_offset;
871 max_write = bv->bv_len;
876 len = min_t(int, max_write - con->out_msg_pos.page_pos,
879 if (do_datacrc && !con->out_msg_pos.did_page_crc) {
882 u32 tmpcrc = le32_to_cpu(con->out_msg->footer.data_crc);
886 BUG_ON(kaddr == NULL);
887 base = kaddr + con->out_msg_pos.page_pos + bio_offset;
888 crc = crc32c(tmpcrc, base, len);
889 con->out_msg->footer.data_crc = cpu_to_le32(crc);
890 con->out_msg_pos.did_page_crc = true;
892 ret = ceph_tcp_sendpage(con->sock, page,
893 con->out_msg_pos.page_pos + bio_offset,
902 con->out_msg_pos.data_pos += ret;
903 con->out_msg_pos.page_pos += ret;
905 con->out_msg_pos.page_pos = 0;
906 con->out_msg_pos.page++;
907 con->out_msg_pos.did_page_crc = false;
909 list_move_tail(&page->lru,
911 else if (msg->pagelist)
912 list_move_tail(&page->lru,
913 &msg->pagelist->head);
916 iter_bio_next(&msg->bio_iter, &msg->bio_seg);
921 dout("write_partial_msg_pages %p msg %p done\n", con, msg);
923 /* prepare and queue up footer, too */
925 con->out_msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
926 ceph_con_out_kvec_reset(con);
927 prepare_write_message_footer(con);
936 static int write_partial_skip(struct ceph_connection *con)
940 while (con->out_skip > 0) {
941 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
943 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, 1);
946 con->out_skip -= ret;
954 * Prepare to read connection handshake, or an ack.
956 static void prepare_read_banner(struct ceph_connection *con)
958 dout("prepare_read_banner %p\n", con);
959 con->in_base_pos = 0;
962 static void prepare_read_connect(struct ceph_connection *con)
964 dout("prepare_read_connect %p\n", con);
965 con->in_base_pos = 0;
968 static void prepare_read_ack(struct ceph_connection *con)
970 dout("prepare_read_ack %p\n", con);
971 con->in_base_pos = 0;
974 static void prepare_read_tag(struct ceph_connection *con)
976 dout("prepare_read_tag %p\n", con);
977 con->in_base_pos = 0;
978 con->in_tag = CEPH_MSGR_TAG_READY;
982 * Prepare to read a message.
984 static int prepare_read_message(struct ceph_connection *con)
986 dout("prepare_read_message %p\n", con);
987 BUG_ON(con->in_msg != NULL);
988 con->in_base_pos = 0;
989 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
994 static int read_partial(struct ceph_connection *con,
995 int *to, int size, void *object)
998 while (con->in_base_pos < *to) {
999 int left = *to - con->in_base_pos;
1000 int have = size - left;
1001 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1004 con->in_base_pos += ret;
1011 * Read all or part of the connect-side handshake on a new connection
1013 static int read_partial_banner(struct ceph_connection *con)
1017 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1020 ret = read_partial(con, &to, strlen(CEPH_BANNER), con->in_banner);
1023 ret = read_partial(con, &to, sizeof(con->actual_peer_addr),
1024 &con->actual_peer_addr);
1027 ret = read_partial(con, &to, sizeof(con->peer_addr_for_me),
1028 &con->peer_addr_for_me);
1035 static int read_partial_connect(struct ceph_connection *con)
1039 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1041 ret = read_partial(con, &to, sizeof(con->in_reply), &con->in_reply);
1044 ret = read_partial(con, &to, le32_to_cpu(con->in_reply.authorizer_len),
1045 con->auth_reply_buf);
1049 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1050 con, (int)con->in_reply.tag,
1051 le32_to_cpu(con->in_reply.connect_seq),
1052 le32_to_cpu(con->in_reply.global_seq));
1059 * Verify the hello banner looks okay.
1061 static int verify_hello(struct ceph_connection *con)
1063 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1064 pr_err("connect to %s got bad banner\n",
1065 ceph_pr_addr(&con->peer_addr.in_addr));
1066 con->error_msg = "protocol error, bad banner";
1072 static bool addr_is_blank(struct sockaddr_storage *ss)
1074 switch (ss->ss_family) {
1076 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1079 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1080 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1081 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1082 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1087 static int addr_port(struct sockaddr_storage *ss)
1089 switch (ss->ss_family) {
1091 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1093 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1098 static void addr_set_port(struct sockaddr_storage *ss, int p)
1100 switch (ss->ss_family) {
1102 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1105 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1111 * Unlike other *_pton function semantics, zero indicates success.
1113 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1114 char delim, const char **ipend)
1116 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1117 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1119 memset(ss, 0, sizeof(*ss));
1121 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1122 ss->ss_family = AF_INET;
1126 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1127 ss->ss_family = AF_INET6;
1135 * Extract hostname string and resolve using kernel DNS facility.
1137 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1138 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1139 struct sockaddr_storage *ss, char delim, const char **ipend)
1141 const char *end, *delim_p;
1142 char *colon_p, *ip_addr = NULL;
1146 * The end of the hostname occurs immediately preceding the delimiter or
1147 * the port marker (':') where the delimiter takes precedence.
1149 delim_p = memchr(name, delim, namelen);
1150 colon_p = memchr(name, ':', namelen);
1152 if (delim_p && colon_p)
1153 end = delim_p < colon_p ? delim_p : colon_p;
1154 else if (!delim_p && colon_p)
1158 if (!end) /* case: hostname:/ */
1159 end = name + namelen;
1165 /* do dns_resolve upcall */
1166 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1168 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1176 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1177 ret, ret ? "failed" : ceph_pr_addr(ss));
1182 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1183 struct sockaddr_storage *ss, char delim, const char **ipend)
1190 * Parse a server name (IP or hostname). If a valid IP address is not found
1191 * then try to extract a hostname to resolve using userspace DNS upcall.
1193 static int ceph_parse_server_name(const char *name, size_t namelen,
1194 struct sockaddr_storage *ss, char delim, const char **ipend)
1198 ret = ceph_pton(name, namelen, ss, delim, ipend);
1200 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1206 * Parse an ip[:port] list into an addr array. Use the default
1207 * monitor port if a port isn't specified.
1209 int ceph_parse_ips(const char *c, const char *end,
1210 struct ceph_entity_addr *addr,
1211 int max_count, int *count)
1213 int i, ret = -EINVAL;
1216 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1217 for (i = 0; i < max_count; i++) {
1219 struct sockaddr_storage *ss = &addr[i].in_addr;
1228 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1237 dout("missing matching ']'\n");
1244 if (p < end && *p == ':') {
1247 while (p < end && *p >= '0' && *p <= '9') {
1248 port = (port * 10) + (*p - '0');
1251 if (port > 65535 || port == 0)
1254 port = CEPH_MON_PORT;
1257 addr_set_port(ss, port);
1259 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1276 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1279 EXPORT_SYMBOL(ceph_parse_ips);
1281 static int process_banner(struct ceph_connection *con)
1283 dout("process_banner on %p\n", con);
1285 if (verify_hello(con) < 0)
1288 ceph_decode_addr(&con->actual_peer_addr);
1289 ceph_decode_addr(&con->peer_addr_for_me);
1292 * Make sure the other end is who we wanted. note that the other
1293 * end may not yet know their ip address, so if it's 0.0.0.0, give
1294 * them the benefit of the doubt.
1296 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1297 sizeof(con->peer_addr)) != 0 &&
1298 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1299 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1300 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1301 ceph_pr_addr(&con->peer_addr.in_addr),
1302 (int)le32_to_cpu(con->peer_addr.nonce),
1303 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1304 (int)le32_to_cpu(con->actual_peer_addr.nonce));
1305 con->error_msg = "wrong peer at address";
1310 * did we learn our address?
1312 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1313 int port = addr_port(&con->msgr->inst.addr.in_addr);
1315 memcpy(&con->msgr->inst.addr.in_addr,
1316 &con->peer_addr_for_me.in_addr,
1317 sizeof(con->peer_addr_for_me.in_addr));
1318 addr_set_port(&con->msgr->inst.addr.in_addr, port);
1319 encode_my_addr(con->msgr);
1320 dout("process_banner learned my addr is %s\n",
1321 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
1324 set_bit(NEGOTIATING, &con->state);
1325 prepare_read_connect(con);
1329 static void fail_protocol(struct ceph_connection *con)
1331 reset_connection(con);
1332 set_bit(CLOSED, &con->state); /* in case there's queued work */
1334 mutex_unlock(&con->mutex);
1335 if (con->ops->bad_proto)
1336 con->ops->bad_proto(con);
1337 mutex_lock(&con->mutex);
1340 static int process_connect(struct ceph_connection *con)
1342 u64 sup_feat = con->msgr->supported_features;
1343 u64 req_feat = con->msgr->required_features;
1344 u64 server_feat = le64_to_cpu(con->in_reply.features);
1347 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1349 switch (con->in_reply.tag) {
1350 case CEPH_MSGR_TAG_FEATURES:
1351 pr_err("%s%lld %s feature set mismatch,"
1352 " my %llx < server's %llx, missing %llx\n",
1353 ENTITY_NAME(con->peer_name),
1354 ceph_pr_addr(&con->peer_addr.in_addr),
1355 sup_feat, server_feat, server_feat & ~sup_feat);
1356 con->error_msg = "missing required protocol features";
1360 case CEPH_MSGR_TAG_BADPROTOVER:
1361 pr_err("%s%lld %s protocol version mismatch,"
1362 " my %d != server's %d\n",
1363 ENTITY_NAME(con->peer_name),
1364 ceph_pr_addr(&con->peer_addr.in_addr),
1365 le32_to_cpu(con->out_connect.protocol_version),
1366 le32_to_cpu(con->in_reply.protocol_version));
1367 con->error_msg = "protocol version mismatch";
1371 case CEPH_MSGR_TAG_BADAUTHORIZER:
1373 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1375 if (con->auth_retry == 2) {
1376 con->error_msg = "connect authorization failure";
1379 con->auth_retry = 1;
1380 ret = prepare_write_connect(con->msgr, con, 0);
1383 prepare_read_connect(con);
1386 case CEPH_MSGR_TAG_RESETSESSION:
1388 * If we connected with a large connect_seq but the peer
1389 * has no record of a session with us (no connection, or
1390 * connect_seq == 0), they will send RESETSESION to indicate
1391 * that they must have reset their session, and may have
1394 dout("process_connect got RESET peer seq %u\n",
1395 le32_to_cpu(con->in_connect.connect_seq));
1396 pr_err("%s%lld %s connection reset\n",
1397 ENTITY_NAME(con->peer_name),
1398 ceph_pr_addr(&con->peer_addr.in_addr));
1399 reset_connection(con);
1400 prepare_write_connect(con->msgr, con, 0);
1401 prepare_read_connect(con);
1403 /* Tell ceph about it. */
1404 mutex_unlock(&con->mutex);
1405 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
1406 if (con->ops->peer_reset)
1407 con->ops->peer_reset(con);
1408 mutex_lock(&con->mutex);
1409 if (test_bit(CLOSED, &con->state) ||
1410 test_bit(OPENING, &con->state))
1414 case CEPH_MSGR_TAG_RETRY_SESSION:
1416 * If we sent a smaller connect_seq than the peer has, try
1417 * again with a larger value.
1419 dout("process_connect got RETRY my seq = %u, peer_seq = %u\n",
1420 le32_to_cpu(con->out_connect.connect_seq),
1421 le32_to_cpu(con->in_connect.connect_seq));
1422 con->connect_seq = le32_to_cpu(con->in_connect.connect_seq);
1423 prepare_write_connect(con->msgr, con, 0);
1424 prepare_read_connect(con);
1427 case CEPH_MSGR_TAG_RETRY_GLOBAL:
1429 * If we sent a smaller global_seq than the peer has, try
1430 * again with a larger value.
1432 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1433 con->peer_global_seq,
1434 le32_to_cpu(con->in_connect.global_seq));
1435 get_global_seq(con->msgr,
1436 le32_to_cpu(con->in_connect.global_seq));
1437 prepare_write_connect(con->msgr, con, 0);
1438 prepare_read_connect(con);
1441 case CEPH_MSGR_TAG_READY:
1442 if (req_feat & ~server_feat) {
1443 pr_err("%s%lld %s protocol feature mismatch,"
1444 " my required %llx > server's %llx, need %llx\n",
1445 ENTITY_NAME(con->peer_name),
1446 ceph_pr_addr(&con->peer_addr.in_addr),
1447 req_feat, server_feat, req_feat & ~server_feat);
1448 con->error_msg = "missing required protocol features";
1452 clear_bit(CONNECTING, &con->state);
1453 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
1455 con->peer_features = server_feat;
1456 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1457 con->peer_global_seq,
1458 le32_to_cpu(con->in_reply.connect_seq),
1460 WARN_ON(con->connect_seq !=
1461 le32_to_cpu(con->in_reply.connect_seq));
1463 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
1464 set_bit(LOSSYTX, &con->state);
1466 prepare_read_tag(con);
1469 case CEPH_MSGR_TAG_WAIT:
1471 * If there is a connection race (we are opening
1472 * connections to each other), one of us may just have
1473 * to WAIT. This shouldn't happen if we are the
1476 pr_err("process_connect got WAIT as client\n");
1477 con->error_msg = "protocol error, got WAIT as client";
1481 pr_err("connect protocol error, will retry\n");
1482 con->error_msg = "protocol error, garbage tag during connect";
1490 * read (part of) an ack
1492 static int read_partial_ack(struct ceph_connection *con)
1496 return read_partial(con, &to, sizeof(con->in_temp_ack),
1502 * We can finally discard anything that's been acked.
1504 static void process_ack(struct ceph_connection *con)
1507 u64 ack = le64_to_cpu(con->in_temp_ack);
1510 while (!list_empty(&con->out_sent)) {
1511 m = list_first_entry(&con->out_sent, struct ceph_msg,
1513 seq = le64_to_cpu(m->hdr.seq);
1516 dout("got ack for seq %llu type %d at %p\n", seq,
1517 le16_to_cpu(m->hdr.type), m);
1518 m->ack_stamp = jiffies;
1521 prepare_read_tag(con);
1527 static int read_partial_message_section(struct ceph_connection *con,
1528 struct kvec *section,
1529 unsigned int sec_len, u32 *crc)
1535 while (section->iov_len < sec_len) {
1536 BUG_ON(section->iov_base == NULL);
1537 left = sec_len - section->iov_len;
1538 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
1539 section->iov_len, left);
1542 section->iov_len += ret;
1544 if (section->iov_len == sec_len)
1545 *crc = crc32c(0, section->iov_base, section->iov_len);
1550 static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con,
1551 struct ceph_msg_header *hdr,
1555 static int read_partial_message_pages(struct ceph_connection *con,
1556 struct page **pages,
1557 unsigned int data_len, bool do_datacrc)
1563 left = min((int)(data_len - con->in_msg_pos.data_pos),
1564 (int)(PAGE_SIZE - con->in_msg_pos.page_pos));
1566 BUG_ON(pages == NULL);
1567 p = kmap(pages[con->in_msg_pos.page]);
1568 ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos,
1570 if (ret > 0 && do_datacrc)
1572 crc32c(con->in_data_crc,
1573 p + con->in_msg_pos.page_pos, ret);
1574 kunmap(pages[con->in_msg_pos.page]);
1577 con->in_msg_pos.data_pos += ret;
1578 con->in_msg_pos.page_pos += ret;
1579 if (con->in_msg_pos.page_pos == PAGE_SIZE) {
1580 con->in_msg_pos.page_pos = 0;
1581 con->in_msg_pos.page++;
1588 static int read_partial_message_bio(struct ceph_connection *con,
1589 struct bio **bio_iter, int *bio_seg,
1590 unsigned int data_len, bool do_datacrc)
1592 struct bio_vec *bv = bio_iovec_idx(*bio_iter, *bio_seg);
1599 left = min((int)(data_len - con->in_msg_pos.data_pos),
1600 (int)(bv->bv_len - con->in_msg_pos.page_pos));
1602 p = kmap(bv->bv_page) + bv->bv_offset;
1604 ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos,
1606 if (ret > 0 && do_datacrc)
1608 crc32c(con->in_data_crc,
1609 p + con->in_msg_pos.page_pos, ret);
1610 kunmap(bv->bv_page);
1613 con->in_msg_pos.data_pos += ret;
1614 con->in_msg_pos.page_pos += ret;
1615 if (con->in_msg_pos.page_pos == bv->bv_len) {
1616 con->in_msg_pos.page_pos = 0;
1617 iter_bio_next(bio_iter, bio_seg);
1625 * read (part of) a message.
1627 static int read_partial_message(struct ceph_connection *con)
1629 struct ceph_msg *m = con->in_msg;
1632 unsigned int front_len, middle_len, data_len;
1633 bool do_datacrc = !con->msgr->nocrc;
1638 dout("read_partial_message con %p msg %p\n", con, m);
1641 while (con->in_base_pos < sizeof(con->in_hdr)) {
1642 left = sizeof(con->in_hdr) - con->in_base_pos;
1643 ret = ceph_tcp_recvmsg(con->sock,
1644 (char *)&con->in_hdr + con->in_base_pos,
1648 con->in_base_pos += ret;
1651 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
1652 if (cpu_to_le32(crc) != con->in_hdr.crc) {
1653 pr_err("read_partial_message bad hdr "
1654 " crc %u != expected %u\n",
1655 crc, con->in_hdr.crc);
1659 front_len = le32_to_cpu(con->in_hdr.front_len);
1660 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
1662 middle_len = le32_to_cpu(con->in_hdr.middle_len);
1663 if (middle_len > CEPH_MSG_MAX_DATA_LEN)
1665 data_len = le32_to_cpu(con->in_hdr.data_len);
1666 if (data_len > CEPH_MSG_MAX_DATA_LEN)
1670 seq = le64_to_cpu(con->in_hdr.seq);
1671 if ((s64)seq - (s64)con->in_seq < 1) {
1672 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1673 ENTITY_NAME(con->peer_name),
1674 ceph_pr_addr(&con->peer_addr.in_addr),
1675 seq, con->in_seq + 1);
1676 con->in_base_pos = -front_len - middle_len - data_len -
1678 con->in_tag = CEPH_MSGR_TAG_READY;
1680 } else if ((s64)seq - (s64)con->in_seq > 1) {
1681 pr_err("read_partial_message bad seq %lld expected %lld\n",
1682 seq, con->in_seq + 1);
1683 con->error_msg = "bad message sequence # for incoming message";
1687 /* allocate message? */
1689 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
1690 con->in_hdr.front_len, con->in_hdr.data_len);
1692 con->in_msg = ceph_alloc_msg(con, &con->in_hdr, &skip);
1694 /* skip this message */
1695 dout("alloc_msg said skip message\n");
1696 BUG_ON(con->in_msg);
1697 con->in_base_pos = -front_len - middle_len - data_len -
1699 con->in_tag = CEPH_MSGR_TAG_READY;
1705 "error allocating memory for incoming message";
1709 m->front.iov_len = 0; /* haven't read it yet */
1711 m->middle->vec.iov_len = 0;
1713 con->in_msg_pos.page = 0;
1715 con->in_msg_pos.page_pos = m->page_alignment;
1717 con->in_msg_pos.page_pos = 0;
1718 con->in_msg_pos.data_pos = 0;
1722 ret = read_partial_message_section(con, &m->front, front_len,
1723 &con->in_front_crc);
1729 ret = read_partial_message_section(con, &m->middle->vec,
1731 &con->in_middle_crc);
1736 if (m->bio && !m->bio_iter)
1737 init_bio_iter(m->bio, &m->bio_iter, &m->bio_seg);
1741 while (con->in_msg_pos.data_pos < data_len) {
1743 ret = read_partial_message_pages(con, m->pages,
1744 data_len, do_datacrc);
1748 } else if (m->bio) {
1750 ret = read_partial_message_bio(con,
1751 &m->bio_iter, &m->bio_seg,
1752 data_len, do_datacrc);
1762 to = sizeof(m->hdr) + sizeof(m->footer);
1763 while (con->in_base_pos < to) {
1764 left = to - con->in_base_pos;
1765 ret = ceph_tcp_recvmsg(con->sock, (char *)&m->footer +
1766 (con->in_base_pos - sizeof(m->hdr)),
1770 con->in_base_pos += ret;
1772 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1773 m, front_len, m->footer.front_crc, middle_len,
1774 m->footer.middle_crc, data_len, m->footer.data_crc);
1777 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
1778 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1779 m, con->in_front_crc, m->footer.front_crc);
1782 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
1783 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1784 m, con->in_middle_crc, m->footer.middle_crc);
1788 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
1789 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
1790 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
1791 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
1795 return 1; /* done! */
1799 * Process message. This happens in the worker thread. The callback should
1800 * be careful not to do anything that waits on other incoming messages or it
1803 static void process_message(struct ceph_connection *con)
1805 struct ceph_msg *msg;
1810 /* if first message, set peer_name */
1811 if (con->peer_name.type == 0)
1812 con->peer_name = msg->hdr.src;
1815 mutex_unlock(&con->mutex);
1817 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
1818 msg, le64_to_cpu(msg->hdr.seq),
1819 ENTITY_NAME(msg->hdr.src),
1820 le16_to_cpu(msg->hdr.type),
1821 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
1822 le32_to_cpu(msg->hdr.front_len),
1823 le32_to_cpu(msg->hdr.data_len),
1824 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
1825 con->ops->dispatch(con, msg);
1827 mutex_lock(&con->mutex);
1828 prepare_read_tag(con);
1833 * Write something to the socket. Called in a worker thread when the
1834 * socket appears to be writeable and we have something ready to send.
1836 static int try_write(struct ceph_connection *con)
1838 struct ceph_messenger *msgr = con->msgr;
1841 dout("try_write start %p state %lu nref %d\n", con, con->state,
1842 atomic_read(&con->nref));
1845 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
1847 /* open the socket first? */
1848 if (con->sock == NULL) {
1849 prepare_write_connect(msgr, con, 1);
1850 prepare_read_banner(con);
1851 set_bit(CONNECTING, &con->state);
1852 clear_bit(NEGOTIATING, &con->state);
1854 BUG_ON(con->in_msg);
1855 con->in_tag = CEPH_MSGR_TAG_READY;
1856 dout("try_write initiating connect on %p new state %lu\n",
1858 ret = ceph_tcp_connect(con);
1860 con->error_msg = "connect error";
1866 /* kvec data queued? */
1867 if (con->out_skip) {
1868 ret = write_partial_skip(con);
1872 if (con->out_kvec_left) {
1873 ret = write_partial_kvec(con);
1880 if (con->out_msg_done) {
1881 ceph_msg_put(con->out_msg);
1882 con->out_msg = NULL; /* we're done with this one */
1886 ret = write_partial_msg_pages(con);
1888 goto more_kvec; /* we need to send the footer, too! */
1892 dout("try_write write_partial_msg_pages err %d\n",
1899 if (!test_bit(CONNECTING, &con->state)) {
1900 /* is anything else pending? */
1901 if (!list_empty(&con->out_queue)) {
1902 prepare_write_message(con);
1905 if (con->in_seq > con->in_seq_acked) {
1906 prepare_write_ack(con);
1909 if (test_and_clear_bit(KEEPALIVE_PENDING, &con->state)) {
1910 prepare_write_keepalive(con);
1915 /* Nothing to do! */
1916 clear_bit(WRITE_PENDING, &con->state);
1917 dout("try_write nothing else to write.\n");
1920 dout("try_write done on %p ret %d\n", con, ret);
1927 * Read what we can from the socket.
1929 static int try_read(struct ceph_connection *con)
1936 if (test_bit(STANDBY, &con->state))
1939 dout("try_read start on %p\n", con);
1942 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
1946 * process_connect and process_message drop and re-take
1947 * con->mutex. make sure we handle a racing close or reopen.
1949 if (test_bit(CLOSED, &con->state) ||
1950 test_bit(OPENING, &con->state)) {
1955 if (test_bit(CONNECTING, &con->state)) {
1956 if (!test_bit(NEGOTIATING, &con->state)) {
1957 dout("try_read connecting\n");
1958 ret = read_partial_banner(con);
1961 ret = process_banner(con);
1965 ret = read_partial_connect(con);
1968 ret = process_connect(con);
1974 if (con->in_base_pos < 0) {
1976 * skipping + discarding content.
1978 * FIXME: there must be a better way to do this!
1980 static char buf[SKIP_BUF_SIZE];
1981 int skip = min((int) sizeof (buf), -con->in_base_pos);
1983 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
1984 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
1987 con->in_base_pos += ret;
1988 if (con->in_base_pos)
1991 if (con->in_tag == CEPH_MSGR_TAG_READY) {
1995 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
1998 dout("try_read got tag %d\n", (int)con->in_tag);
1999 switch (con->in_tag) {
2000 case CEPH_MSGR_TAG_MSG:
2001 prepare_read_message(con);
2003 case CEPH_MSGR_TAG_ACK:
2004 prepare_read_ack(con);
2006 case CEPH_MSGR_TAG_CLOSE:
2007 set_bit(CLOSED, &con->state); /* fixme */
2013 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2014 ret = read_partial_message(con);
2018 con->error_msg = "bad crc";
2022 con->error_msg = "io error";
2027 if (con->in_tag == CEPH_MSGR_TAG_READY)
2029 process_message(con);
2032 if (con->in_tag == CEPH_MSGR_TAG_ACK) {
2033 ret = read_partial_ack(con);
2041 dout("try_read done on %p ret %d\n", con, ret);
2045 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2046 con->error_msg = "protocol error, garbage tag";
2053 * Atomically queue work on a connection. Bump @con reference to
2054 * avoid races with connection teardown.
2056 static void queue_con(struct ceph_connection *con)
2058 if (test_bit(DEAD, &con->state)) {
2059 dout("queue_con %p ignoring: DEAD\n",
2064 if (!con->ops->get(con)) {
2065 dout("queue_con %p ref count 0\n", con);
2069 if (!queue_delayed_work(ceph_msgr_wq, &con->work, 0)) {
2070 dout("queue_con %p - already queued\n", con);
2073 dout("queue_con %p\n", con);
2078 * Do some work on a connection. Drop a connection ref when we're done.
2080 static void con_work(struct work_struct *work)
2082 struct ceph_connection *con = container_of(work, struct ceph_connection,
2086 mutex_lock(&con->mutex);
2088 if (test_and_clear_bit(BACKOFF, &con->state)) {
2089 dout("con_work %p backing off\n", con);
2090 if (queue_delayed_work(ceph_msgr_wq, &con->work,
2091 round_jiffies_relative(con->delay))) {
2092 dout("con_work %p backoff %lu\n", con, con->delay);
2093 mutex_unlock(&con->mutex);
2097 dout("con_work %p FAILED to back off %lu\n", con,
2102 if (test_bit(STANDBY, &con->state)) {
2103 dout("con_work %p STANDBY\n", con);
2106 if (test_bit(CLOSED, &con->state)) { /* e.g. if we are replaced */
2107 dout("con_work CLOSED\n");
2108 con_close_socket(con);
2111 if (test_and_clear_bit(OPENING, &con->state)) {
2112 /* reopen w/ new peer */
2113 dout("con_work OPENING\n");
2114 con_close_socket(con);
2117 if (test_and_clear_bit(SOCK_CLOSED, &con->state))
2120 ret = try_read(con);
2126 ret = try_write(con);
2133 mutex_unlock(&con->mutex);
2139 mutex_unlock(&con->mutex);
2140 ceph_fault(con); /* error/fault path */
2146 * Generic error/fault handler. A retry mechanism is used with
2147 * exponential backoff
2149 static void ceph_fault(struct ceph_connection *con)
2151 pr_err("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2152 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2153 dout("fault %p state %lu to peer %s\n",
2154 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2156 if (test_bit(LOSSYTX, &con->state)) {
2157 dout("fault on LOSSYTX channel\n");
2161 mutex_lock(&con->mutex);
2162 if (test_bit(CLOSED, &con->state))
2165 con_close_socket(con);
2168 ceph_msg_put(con->in_msg);
2172 /* Requeue anything that hasn't been acked */
2173 list_splice_init(&con->out_sent, &con->out_queue);
2175 /* If there are no messages queued or keepalive pending, place
2176 * the connection in a STANDBY state */
2177 if (list_empty(&con->out_queue) &&
2178 !test_bit(KEEPALIVE_PENDING, &con->state)) {
2179 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2180 clear_bit(WRITE_PENDING, &con->state);
2181 set_bit(STANDBY, &con->state);
2183 /* retry after a delay. */
2184 if (con->delay == 0)
2185 con->delay = BASE_DELAY_INTERVAL;
2186 else if (con->delay < MAX_DELAY_INTERVAL)
2189 if (queue_delayed_work(ceph_msgr_wq, &con->work,
2190 round_jiffies_relative(con->delay))) {
2191 dout("fault queued %p delay %lu\n", con, con->delay);
2194 dout("fault failed to queue %p delay %lu, backoff\n",
2197 * In many cases we see a socket state change
2198 * while con_work is running and end up
2199 * queuing (non-delayed) work, such that we
2200 * can't backoff with a delay. Set a flag so
2201 * that when con_work restarts we schedule the
2204 set_bit(BACKOFF, &con->state);
2209 mutex_unlock(&con->mutex);
2212 * in case we faulted due to authentication, invalidate our
2213 * current tickets so that we can get new ones.
2215 if (con->auth_retry && con->ops->invalidate_authorizer) {
2216 dout("calling invalidate_authorizer()\n");
2217 con->ops->invalidate_authorizer(con);
2220 if (con->ops->fault)
2221 con->ops->fault(con);
2227 * create a new messenger instance
2229 struct ceph_messenger *ceph_messenger_create(struct ceph_entity_addr *myaddr,
2230 u32 supported_features,
2231 u32 required_features)
2233 struct ceph_messenger *msgr;
2235 msgr = kzalloc(sizeof(*msgr), GFP_KERNEL);
2237 return ERR_PTR(-ENOMEM);
2239 msgr->supported_features = supported_features;
2240 msgr->required_features = required_features;
2242 spin_lock_init(&msgr->global_seq_lock);
2245 msgr->inst.addr = *myaddr;
2247 /* select a random nonce */
2248 msgr->inst.addr.type = 0;
2249 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2250 encode_my_addr(msgr);
2252 dout("messenger_create %p\n", msgr);
2255 EXPORT_SYMBOL(ceph_messenger_create);
2257 void ceph_messenger_destroy(struct ceph_messenger *msgr)
2259 dout("destroy %p\n", msgr);
2261 dout("destroyed messenger %p\n", msgr);
2263 EXPORT_SYMBOL(ceph_messenger_destroy);
2265 static void clear_standby(struct ceph_connection *con)
2267 /* come back from STANDBY? */
2268 if (test_and_clear_bit(STANDBY, &con->state)) {
2269 mutex_lock(&con->mutex);
2270 dout("clear_standby %p and ++connect_seq\n", con);
2272 WARN_ON(test_bit(WRITE_PENDING, &con->state));
2273 WARN_ON(test_bit(KEEPALIVE_PENDING, &con->state));
2274 mutex_unlock(&con->mutex);
2279 * Queue up an outgoing message on the given connection.
2281 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2283 if (test_bit(CLOSED, &con->state)) {
2284 dout("con_send %p closed, dropping %p\n", con, msg);
2290 msg->hdr.src = con->msgr->inst.name;
2292 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2294 msg->needs_out_seq = true;
2297 mutex_lock(&con->mutex);
2298 BUG_ON(!list_empty(&msg->list_head));
2299 list_add_tail(&msg->list_head, &con->out_queue);
2300 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2301 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2302 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2303 le32_to_cpu(msg->hdr.front_len),
2304 le32_to_cpu(msg->hdr.middle_len),
2305 le32_to_cpu(msg->hdr.data_len));
2306 mutex_unlock(&con->mutex);
2308 /* if there wasn't anything waiting to send before, queue
2311 if (test_and_set_bit(WRITE_PENDING, &con->state) == 0)
2314 EXPORT_SYMBOL(ceph_con_send);
2317 * Revoke a message that was previously queued for send
2319 void ceph_con_revoke(struct ceph_connection *con, struct ceph_msg *msg)
2321 mutex_lock(&con->mutex);
2322 if (!list_empty(&msg->list_head)) {
2323 dout("con_revoke %p msg %p - was on queue\n", con, msg);
2324 list_del_init(&msg->list_head);
2328 if (con->out_msg == msg) {
2329 dout("con_revoke %p msg %p - was sending\n", con, msg);
2330 con->out_msg = NULL;
2331 if (con->out_kvec_is_msg) {
2332 con->out_skip = con->out_kvec_bytes;
2333 con->out_kvec_is_msg = false;
2338 mutex_unlock(&con->mutex);
2342 * Revoke a message that we may be reading data into
2344 void ceph_con_revoke_message(struct ceph_connection *con, struct ceph_msg *msg)
2346 mutex_lock(&con->mutex);
2347 if (con->in_msg && con->in_msg == msg) {
2348 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
2349 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
2350 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
2352 /* skip rest of message */
2353 dout("con_revoke_pages %p msg %p revoked\n", con, msg);
2354 con->in_base_pos = con->in_base_pos -
2355 sizeof(struct ceph_msg_header) -
2359 sizeof(struct ceph_msg_footer);
2360 ceph_msg_put(con->in_msg);
2362 con->in_tag = CEPH_MSGR_TAG_READY;
2365 dout("con_revoke_pages %p msg %p pages %p no-op\n",
2366 con, con->in_msg, msg);
2368 mutex_unlock(&con->mutex);
2372 * Queue a keepalive byte to ensure the tcp connection is alive.
2374 void ceph_con_keepalive(struct ceph_connection *con)
2376 dout("con_keepalive %p\n", con);
2378 if (test_and_set_bit(KEEPALIVE_PENDING, &con->state) == 0 &&
2379 test_and_set_bit(WRITE_PENDING, &con->state) == 0)
2382 EXPORT_SYMBOL(ceph_con_keepalive);
2386 * construct a new message with given type, size
2387 * the new msg has a ref count of 1.
2389 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
2394 m = kmalloc(sizeof(*m), flags);
2397 kref_init(&m->kref);
2398 INIT_LIST_HEAD(&m->list_head);
2401 m->hdr.type = cpu_to_le16(type);
2402 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
2404 m->hdr.front_len = cpu_to_le32(front_len);
2405 m->hdr.middle_len = 0;
2406 m->hdr.data_len = 0;
2407 m->hdr.data_off = 0;
2408 m->hdr.reserved = 0;
2409 m->footer.front_crc = 0;
2410 m->footer.middle_crc = 0;
2411 m->footer.data_crc = 0;
2412 m->footer.flags = 0;
2413 m->front_max = front_len;
2414 m->front_is_vmalloc = false;
2415 m->more_to_follow = false;
2424 m->page_alignment = 0;
2434 if (front_len > PAGE_CACHE_SIZE) {
2435 m->front.iov_base = __vmalloc(front_len, flags,
2437 m->front_is_vmalloc = true;
2439 m->front.iov_base = kmalloc(front_len, flags);
2441 if (m->front.iov_base == NULL) {
2442 dout("ceph_msg_new can't allocate %d bytes\n",
2447 m->front.iov_base = NULL;
2449 m->front.iov_len = front_len;
2451 dout("ceph_msg_new %p front %d\n", m, front_len);
2458 pr_err("msg_new can't create type %d front %d\n", type,
2462 dout("msg_new can't create type %d front %d\n", type,
2467 EXPORT_SYMBOL(ceph_msg_new);
2470 * Allocate "middle" portion of a message, if it is needed and wasn't
2471 * allocated by alloc_msg. This allows us to read a small fixed-size
2472 * per-type header in the front and then gracefully fail (i.e.,
2473 * propagate the error to the caller based on info in the front) when
2474 * the middle is too large.
2476 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
2478 int type = le16_to_cpu(msg->hdr.type);
2479 int middle_len = le32_to_cpu(msg->hdr.middle_len);
2481 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
2482 ceph_msg_type_name(type), middle_len);
2483 BUG_ON(!middle_len);
2484 BUG_ON(msg->middle);
2486 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
2493 * Generic message allocator, for incoming messages.
2495 static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con,
2496 struct ceph_msg_header *hdr,
2499 int type = le16_to_cpu(hdr->type);
2500 int front_len = le32_to_cpu(hdr->front_len);
2501 int middle_len = le32_to_cpu(hdr->middle_len);
2502 struct ceph_msg *msg = NULL;
2505 if (con->ops->alloc_msg) {
2506 mutex_unlock(&con->mutex);
2507 msg = con->ops->alloc_msg(con, hdr, skip);
2508 mutex_lock(&con->mutex);
2514 msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
2516 pr_err("unable to allocate msg type %d len %d\n",
2520 msg->page_alignment = le16_to_cpu(hdr->data_off);
2522 memcpy(&msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2524 if (middle_len && !msg->middle) {
2525 ret = ceph_alloc_middle(con, msg);
2537 * Free a generically kmalloc'd message.
2539 void ceph_msg_kfree(struct ceph_msg *m)
2541 dout("msg_kfree %p\n", m);
2542 if (m->front_is_vmalloc)
2543 vfree(m->front.iov_base);
2545 kfree(m->front.iov_base);
2550 * Drop a msg ref. Destroy as needed.
2552 void ceph_msg_last_put(struct kref *kref)
2554 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
2556 dout("ceph_msg_put last one on %p\n", m);
2557 WARN_ON(!list_empty(&m->list_head));
2559 /* drop middle, data, if any */
2561 ceph_buffer_put(m->middle);
2568 ceph_pagelist_release(m->pagelist);
2576 ceph_msgpool_put(m->pool, m);
2580 EXPORT_SYMBOL(ceph_msg_last_put);
2582 void ceph_msg_dump(struct ceph_msg *msg)
2584 pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg,
2585 msg->front_max, msg->nr_pages);
2586 print_hex_dump(KERN_DEBUG, "header: ",
2587 DUMP_PREFIX_OFFSET, 16, 1,
2588 &msg->hdr, sizeof(msg->hdr), true);
2589 print_hex_dump(KERN_DEBUG, " front: ",
2590 DUMP_PREFIX_OFFSET, 16, 1,
2591 msg->front.iov_base, msg->front.iov_len, true);
2593 print_hex_dump(KERN_DEBUG, "middle: ",
2594 DUMP_PREFIX_OFFSET, 16, 1,
2595 msg->middle->vec.iov_base,
2596 msg->middle->vec.iov_len, true);
2597 print_hex_dump(KERN_DEBUG, "footer: ",
2598 DUMP_PREFIX_OFFSET, 16, 1,
2599 &msg->footer, sizeof(msg->footer), true);
2601 EXPORT_SYMBOL(ceph_msg_dump);