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>
16 #include <linux/ceph/libceph.h>
17 #include <linux/ceph/messenger.h>
18 #include <linux/ceph/decode.h>
19 #include <linux/ceph/pagelist.h>
22 * Ceph uses the messenger to exchange ceph_msg messages with other
23 * hosts in the system. The messenger provides ordered and reliable
24 * delivery. We tolerate TCP disconnects by reconnecting (with
25 * exponential backoff) in the case of a fault (disconnection, bad
26 * crc, protocol error). Acks allow sent messages to be discarded by
30 /* static tag bytes (protocol control messages) */
31 static char tag_msg = CEPH_MSGR_TAG_MSG;
32 static char tag_ack = CEPH_MSGR_TAG_ACK;
33 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
36 static struct lock_class_key socket_class;
40 static void queue_con(struct ceph_connection *con);
41 static void con_work(struct work_struct *);
42 static void ceph_fault(struct ceph_connection *con);
45 * nicely render a sockaddr as a string.
47 #define MAX_ADDR_STR 20
48 #define MAX_ADDR_STR_LEN 60
49 static char addr_str[MAX_ADDR_STR][MAX_ADDR_STR_LEN];
50 static DEFINE_SPINLOCK(addr_str_lock);
51 static int last_addr_str;
53 const char *ceph_pr_addr(const struct sockaddr_storage *ss)
57 struct sockaddr_in *in4 = (void *)ss;
58 struct sockaddr_in6 *in6 = (void *)ss;
60 spin_lock(&addr_str_lock);
62 if (last_addr_str == MAX_ADDR_STR)
64 spin_unlock(&addr_str_lock);
67 switch (ss->ss_family) {
69 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%u", &in4->sin_addr,
70 (unsigned int)ntohs(in4->sin_port));
74 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%u", &in6->sin6_addr,
75 (unsigned int)ntohs(in6->sin6_port));
79 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %d)",
85 EXPORT_SYMBOL(ceph_pr_addr);
87 static void encode_my_addr(struct ceph_messenger *msgr)
89 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
90 ceph_encode_addr(&msgr->my_enc_addr);
94 * work queue for all reading and writing to/from the socket.
96 struct workqueue_struct *ceph_msgr_wq;
98 int ceph_msgr_init(void)
100 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0);
102 pr_err("msgr_init failed to create workqueue\n");
107 EXPORT_SYMBOL(ceph_msgr_init);
109 void ceph_msgr_exit(void)
111 destroy_workqueue(ceph_msgr_wq);
113 EXPORT_SYMBOL(ceph_msgr_exit);
115 void ceph_msgr_flush(void)
117 flush_workqueue(ceph_msgr_wq);
119 EXPORT_SYMBOL(ceph_msgr_flush);
123 * socket callback functions
126 /* data available on socket, or listen socket received a connect */
127 static void ceph_data_ready(struct sock *sk, int count_unused)
129 struct ceph_connection *con =
130 (struct ceph_connection *)sk->sk_user_data;
131 if (sk->sk_state != TCP_CLOSE_WAIT) {
132 dout("ceph_data_ready on %p state = %lu, queueing work\n",
138 /* socket has buffer space for writing */
139 static void ceph_write_space(struct sock *sk)
141 struct ceph_connection *con =
142 (struct ceph_connection *)sk->sk_user_data;
144 /* only queue to workqueue if there is data we want to write. */
145 if (test_bit(WRITE_PENDING, &con->state)) {
146 dout("ceph_write_space %p queueing write work\n", con);
149 dout("ceph_write_space %p nothing to write\n", con);
152 /* since we have our own write_space, clear the SOCK_NOSPACE flag */
153 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
156 /* socket's state has changed */
157 static void ceph_state_change(struct sock *sk)
159 struct ceph_connection *con =
160 (struct ceph_connection *)sk->sk_user_data;
162 dout("ceph_state_change %p state = %lu sk_state = %u\n",
163 con, con->state, sk->sk_state);
165 if (test_bit(CLOSED, &con->state))
168 switch (sk->sk_state) {
170 dout("ceph_state_change TCP_CLOSE\n");
172 dout("ceph_state_change TCP_CLOSE_WAIT\n");
173 if (test_and_set_bit(SOCK_CLOSED, &con->state) == 0) {
174 if (test_bit(CONNECTING, &con->state))
175 con->error_msg = "connection failed";
177 con->error_msg = "socket closed";
181 case TCP_ESTABLISHED:
182 dout("ceph_state_change TCP_ESTABLISHED\n");
189 * set up socket callbacks
191 static void set_sock_callbacks(struct socket *sock,
192 struct ceph_connection *con)
194 struct sock *sk = sock->sk;
195 sk->sk_user_data = (void *)con;
196 sk->sk_data_ready = ceph_data_ready;
197 sk->sk_write_space = ceph_write_space;
198 sk->sk_state_change = ceph_state_change;
207 * initiate connection to a remote socket.
209 static struct socket *ceph_tcp_connect(struct ceph_connection *con)
211 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
216 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
221 sock->sk->sk_allocation = GFP_NOFS;
223 #ifdef CONFIG_LOCKDEP
224 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
227 set_sock_callbacks(sock, con);
229 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
231 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
233 if (ret == -EINPROGRESS) {
234 dout("connect %s EINPROGRESS sk_state = %u\n",
235 ceph_pr_addr(&con->peer_addr.in_addr),
240 pr_err("connect %s error %d\n",
241 ceph_pr_addr(&con->peer_addr.in_addr), ret);
244 con->error_msg = "connect error";
252 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
254 struct kvec iov = {buf, len};
255 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
258 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
265 * write something. @more is true if caller will be sending more data
268 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
269 size_t kvlen, size_t len, int more)
271 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
275 msg.msg_flags |= MSG_MORE;
277 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
279 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
287 * Shutdown/close the socket for the given connection.
289 static int con_close_socket(struct ceph_connection *con)
293 dout("con_close_socket on %p sock %p\n", con, con->sock);
296 set_bit(SOCK_CLOSED, &con->state);
297 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
298 sock_release(con->sock);
300 clear_bit(SOCK_CLOSED, &con->state);
305 * Reset a connection. Discard all incoming and outgoing messages
306 * and clear *_seq state.
308 static void ceph_msg_remove(struct ceph_msg *msg)
310 list_del_init(&msg->list_head);
313 static void ceph_msg_remove_list(struct list_head *head)
315 while (!list_empty(head)) {
316 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
318 ceph_msg_remove(msg);
322 static void reset_connection(struct ceph_connection *con)
324 /* reset connection, out_queue, msg_ and connect_seq */
325 /* discard existing out_queue and msg_seq */
326 ceph_msg_remove_list(&con->out_queue);
327 ceph_msg_remove_list(&con->out_sent);
330 ceph_msg_put(con->in_msg);
334 con->connect_seq = 0;
337 ceph_msg_put(con->out_msg);
341 con->in_seq_acked = 0;
345 * mark a peer down. drop any open connections.
347 void ceph_con_close(struct ceph_connection *con)
349 dout("con_close %p peer %s\n", con,
350 ceph_pr_addr(&con->peer_addr.in_addr));
351 set_bit(CLOSED, &con->state); /* in case there's queued work */
352 clear_bit(STANDBY, &con->state); /* avoid connect_seq bump */
353 clear_bit(LOSSYTX, &con->state); /* so we retry next connect */
354 clear_bit(KEEPALIVE_PENDING, &con->state);
355 clear_bit(WRITE_PENDING, &con->state);
356 mutex_lock(&con->mutex);
357 reset_connection(con);
358 con->peer_global_seq = 0;
359 cancel_delayed_work(&con->work);
360 mutex_unlock(&con->mutex);
363 EXPORT_SYMBOL(ceph_con_close);
366 * Reopen a closed connection, with a new peer address.
368 void ceph_con_open(struct ceph_connection *con, struct ceph_entity_addr *addr)
370 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
371 set_bit(OPENING, &con->state);
372 clear_bit(CLOSED, &con->state);
373 memcpy(&con->peer_addr, addr, sizeof(*addr));
374 con->delay = 0; /* reset backoff memory */
377 EXPORT_SYMBOL(ceph_con_open);
380 * return true if this connection ever successfully opened
382 bool ceph_con_opened(struct ceph_connection *con)
384 return con->connect_seq > 0;
390 struct ceph_connection *ceph_con_get(struct ceph_connection *con)
392 dout("con_get %p nref = %d -> %d\n", con,
393 atomic_read(&con->nref), atomic_read(&con->nref) + 1);
394 if (atomic_inc_not_zero(&con->nref))
399 void ceph_con_put(struct ceph_connection *con)
401 dout("con_put %p nref = %d -> %d\n", con,
402 atomic_read(&con->nref), atomic_read(&con->nref) - 1);
403 BUG_ON(atomic_read(&con->nref) == 0);
404 if (atomic_dec_and_test(&con->nref)) {
411 * initialize a new connection.
413 void ceph_con_init(struct ceph_messenger *msgr, struct ceph_connection *con)
415 dout("con_init %p\n", con);
416 memset(con, 0, sizeof(*con));
417 atomic_set(&con->nref, 1);
419 mutex_init(&con->mutex);
420 INIT_LIST_HEAD(&con->out_queue);
421 INIT_LIST_HEAD(&con->out_sent);
422 INIT_DELAYED_WORK(&con->work, con_work);
424 EXPORT_SYMBOL(ceph_con_init);
428 * We maintain a global counter to order connection attempts. Get
429 * a unique seq greater than @gt.
431 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
435 spin_lock(&msgr->global_seq_lock);
436 if (msgr->global_seq < gt)
437 msgr->global_seq = gt;
438 ret = ++msgr->global_seq;
439 spin_unlock(&msgr->global_seq_lock);
445 * Prepare footer for currently outgoing message, and finish things
446 * off. Assumes out_kvec* are already valid.. we just add on to the end.
448 static void prepare_write_message_footer(struct ceph_connection *con, int v)
450 struct ceph_msg *m = con->out_msg;
452 dout("prepare_write_message_footer %p\n", con);
453 con->out_kvec_is_msg = true;
454 con->out_kvec[v].iov_base = &m->footer;
455 con->out_kvec[v].iov_len = sizeof(m->footer);
456 con->out_kvec_bytes += sizeof(m->footer);
457 con->out_kvec_left++;
458 con->out_more = m->more_to_follow;
459 con->out_msg_done = true;
463 * Prepare headers for the next outgoing message.
465 static void prepare_write_message(struct ceph_connection *con)
470 con->out_kvec_bytes = 0;
471 con->out_kvec_is_msg = true;
472 con->out_msg_done = false;
474 /* Sneak an ack in there first? If we can get it into the same
475 * TCP packet that's a good thing. */
476 if (con->in_seq > con->in_seq_acked) {
477 con->in_seq_acked = con->in_seq;
478 con->out_kvec[v].iov_base = &tag_ack;
479 con->out_kvec[v++].iov_len = 1;
480 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
481 con->out_kvec[v].iov_base = &con->out_temp_ack;
482 con->out_kvec[v++].iov_len = sizeof(con->out_temp_ack);
483 con->out_kvec_bytes = 1 + sizeof(con->out_temp_ack);
486 m = list_first_entry(&con->out_queue,
487 struct ceph_msg, list_head);
490 /* put message on sent list */
492 list_move_tail(&m->list_head, &con->out_sent);
495 * only assign outgoing seq # if we haven't sent this message
496 * yet. if it is requeued, resend with it's original seq.
498 if (m->needs_out_seq) {
499 m->hdr.seq = cpu_to_le64(++con->out_seq);
500 m->needs_out_seq = false;
503 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n",
504 m, con->out_seq, le16_to_cpu(m->hdr.type),
505 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
506 le32_to_cpu(m->hdr.data_len),
508 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
510 /* tag + hdr + front + middle */
511 con->out_kvec[v].iov_base = &tag_msg;
512 con->out_kvec[v++].iov_len = 1;
513 con->out_kvec[v].iov_base = &m->hdr;
514 con->out_kvec[v++].iov_len = sizeof(m->hdr);
515 con->out_kvec[v++] = m->front;
517 con->out_kvec[v++] = m->middle->vec;
518 con->out_kvec_left = v;
519 con->out_kvec_bytes += 1 + sizeof(m->hdr) + m->front.iov_len +
520 (m->middle ? m->middle->vec.iov_len : 0);
521 con->out_kvec_cur = con->out_kvec;
523 /* fill in crc (except data pages), footer */
524 con->out_msg->hdr.crc =
525 cpu_to_le32(crc32c(0, (void *)&m->hdr,
526 sizeof(m->hdr) - sizeof(m->hdr.crc)));
527 con->out_msg->footer.flags = CEPH_MSG_FOOTER_COMPLETE;
528 con->out_msg->footer.front_crc =
529 cpu_to_le32(crc32c(0, m->front.iov_base, m->front.iov_len));
531 con->out_msg->footer.middle_crc =
532 cpu_to_le32(crc32c(0, m->middle->vec.iov_base,
533 m->middle->vec.iov_len));
535 con->out_msg->footer.middle_crc = 0;
536 con->out_msg->footer.data_crc = 0;
537 dout("prepare_write_message front_crc %u data_crc %u\n",
538 le32_to_cpu(con->out_msg->footer.front_crc),
539 le32_to_cpu(con->out_msg->footer.middle_crc));
541 /* is there a data payload? */
542 if (le32_to_cpu(m->hdr.data_len) > 0) {
543 /* initialize page iterator */
544 con->out_msg_pos.page = 0;
546 con->out_msg_pos.page_pos = m->page_alignment;
548 con->out_msg_pos.page_pos = 0;
549 con->out_msg_pos.data_pos = 0;
550 con->out_msg_pos.did_page_crc = 0;
551 con->out_more = 1; /* data + footer will follow */
553 /* no, queue up footer too and be done */
554 prepare_write_message_footer(con, v);
557 set_bit(WRITE_PENDING, &con->state);
563 static void prepare_write_ack(struct ceph_connection *con)
565 dout("prepare_write_ack %p %llu -> %llu\n", con,
566 con->in_seq_acked, con->in_seq);
567 con->in_seq_acked = con->in_seq;
569 con->out_kvec[0].iov_base = &tag_ack;
570 con->out_kvec[0].iov_len = 1;
571 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
572 con->out_kvec[1].iov_base = &con->out_temp_ack;
573 con->out_kvec[1].iov_len = sizeof(con->out_temp_ack);
574 con->out_kvec_left = 2;
575 con->out_kvec_bytes = 1 + sizeof(con->out_temp_ack);
576 con->out_kvec_cur = con->out_kvec;
577 con->out_more = 1; /* more will follow.. eventually.. */
578 set_bit(WRITE_PENDING, &con->state);
582 * Prepare to write keepalive byte.
584 static void prepare_write_keepalive(struct ceph_connection *con)
586 dout("prepare_write_keepalive %p\n", con);
587 con->out_kvec[0].iov_base = &tag_keepalive;
588 con->out_kvec[0].iov_len = 1;
589 con->out_kvec_left = 1;
590 con->out_kvec_bytes = 1;
591 con->out_kvec_cur = con->out_kvec;
592 set_bit(WRITE_PENDING, &con->state);
596 * Connection negotiation.
599 static int prepare_connect_authorizer(struct ceph_connection *con)
603 int auth_protocol = 0;
605 mutex_unlock(&con->mutex);
606 if (con->ops->get_authorizer)
607 con->ops->get_authorizer(con, &auth_buf, &auth_len,
608 &auth_protocol, &con->auth_reply_buf,
609 &con->auth_reply_buf_len,
611 mutex_lock(&con->mutex);
613 if (test_bit(CLOSED, &con->state) ||
614 test_bit(OPENING, &con->state))
617 con->out_connect.authorizer_protocol = cpu_to_le32(auth_protocol);
618 con->out_connect.authorizer_len = cpu_to_le32(auth_len);
621 con->out_kvec[con->out_kvec_left].iov_base = auth_buf;
622 con->out_kvec[con->out_kvec_left].iov_len = auth_len;
623 con->out_kvec_left++;
624 con->out_kvec_bytes += auth_len;
630 * We connected to a peer and are saying hello.
632 static void prepare_write_banner(struct ceph_messenger *msgr,
633 struct ceph_connection *con)
635 int len = strlen(CEPH_BANNER);
637 con->out_kvec[0].iov_base = CEPH_BANNER;
638 con->out_kvec[0].iov_len = len;
639 con->out_kvec[1].iov_base = &msgr->my_enc_addr;
640 con->out_kvec[1].iov_len = sizeof(msgr->my_enc_addr);
641 con->out_kvec_left = 2;
642 con->out_kvec_bytes = len + sizeof(msgr->my_enc_addr);
643 con->out_kvec_cur = con->out_kvec;
645 set_bit(WRITE_PENDING, &con->state);
648 static int prepare_write_connect(struct ceph_messenger *msgr,
649 struct ceph_connection *con,
652 unsigned global_seq = get_global_seq(con->msgr, 0);
655 switch (con->peer_name.type) {
656 case CEPH_ENTITY_TYPE_MON:
657 proto = CEPH_MONC_PROTOCOL;
659 case CEPH_ENTITY_TYPE_OSD:
660 proto = CEPH_OSDC_PROTOCOL;
662 case CEPH_ENTITY_TYPE_MDS:
663 proto = CEPH_MDSC_PROTOCOL;
669 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
670 con->connect_seq, global_seq, proto);
672 con->out_connect.features = cpu_to_le64(msgr->supported_features);
673 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
674 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
675 con->out_connect.global_seq = cpu_to_le32(global_seq);
676 con->out_connect.protocol_version = cpu_to_le32(proto);
677 con->out_connect.flags = 0;
680 con->out_kvec_left = 0;
681 con->out_kvec_bytes = 0;
683 con->out_kvec[con->out_kvec_left].iov_base = &con->out_connect;
684 con->out_kvec[con->out_kvec_left].iov_len = sizeof(con->out_connect);
685 con->out_kvec_left++;
686 con->out_kvec_bytes += sizeof(con->out_connect);
687 con->out_kvec_cur = con->out_kvec;
689 set_bit(WRITE_PENDING, &con->state);
691 return prepare_connect_authorizer(con);
696 * write as much of pending kvecs to the socket as we can.
698 * 0 -> socket full, but more to do
701 static int write_partial_kvec(struct ceph_connection *con)
705 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
706 while (con->out_kvec_bytes > 0) {
707 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
708 con->out_kvec_left, con->out_kvec_bytes,
712 con->out_kvec_bytes -= ret;
713 if (con->out_kvec_bytes == 0)
716 if (ret >= con->out_kvec_cur->iov_len) {
717 ret -= con->out_kvec_cur->iov_len;
719 con->out_kvec_left--;
721 con->out_kvec_cur->iov_len -= ret;
722 con->out_kvec_cur->iov_base += ret;
728 con->out_kvec_left = 0;
729 con->out_kvec_is_msg = false;
732 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
733 con->out_kvec_bytes, con->out_kvec_left, ret);
734 return ret; /* done! */
738 static void init_bio_iter(struct bio *bio, struct bio **iter, int *seg)
749 static void iter_bio_next(struct bio **bio_iter, int *seg)
751 if (*bio_iter == NULL)
754 BUG_ON(*seg >= (*bio_iter)->bi_vcnt);
757 if (*seg == (*bio_iter)->bi_vcnt)
758 init_bio_iter((*bio_iter)->bi_next, bio_iter, seg);
763 * Write as much message data payload as we can. If we finish, queue
765 * 1 -> done, footer is now queued in out_kvec[].
766 * 0 -> socket full, but more to do
769 static int write_partial_msg_pages(struct ceph_connection *con)
771 struct ceph_msg *msg = con->out_msg;
772 unsigned data_len = le32_to_cpu(msg->hdr.data_len);
774 int crc = con->msgr->nocrc;
778 size_t trail_len = (msg->trail ? msg->trail->length : 0);
780 dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
781 con, con->out_msg, con->out_msg_pos.page, con->out_msg->nr_pages,
782 con->out_msg_pos.page_pos);
785 if (msg->bio && !msg->bio_iter)
786 init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg);
789 while (data_len > con->out_msg_pos.data_pos) {
790 struct page *page = NULL;
792 int max_write = PAGE_SIZE;
795 total_max_write = data_len - trail_len -
796 con->out_msg_pos.data_pos;
799 * if we are calculating the data crc (the default), we need
800 * to map the page. if our pages[] has been revoked, use the
804 /* have we reached the trail part of the data? */
805 if (con->out_msg_pos.data_pos >= data_len - trail_len) {
808 total_max_write = data_len - con->out_msg_pos.data_pos;
810 page = list_first_entry(&msg->trail->head,
814 max_write = PAGE_SIZE;
815 } else if (msg->pages) {
816 page = msg->pages[con->out_msg_pos.page];
819 } else if (msg->pagelist) {
820 page = list_first_entry(&msg->pagelist->head,
825 } else if (msg->bio) {
828 bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg);
830 page_shift = bv->bv_offset;
832 kaddr = kmap(page) + page_shift;
833 max_write = bv->bv_len;
836 page = con->msgr->zero_page;
838 kaddr = page_address(con->msgr->zero_page);
840 len = min_t(int, max_write - con->out_msg_pos.page_pos,
843 if (crc && !con->out_msg_pos.did_page_crc) {
844 void *base = kaddr + con->out_msg_pos.page_pos;
845 u32 tmpcrc = le32_to_cpu(con->out_msg->footer.data_crc);
847 BUG_ON(kaddr == NULL);
848 con->out_msg->footer.data_crc =
849 cpu_to_le32(crc32c(tmpcrc, base, len));
850 con->out_msg_pos.did_page_crc = 1;
852 ret = kernel_sendpage(con->sock, page,
853 con->out_msg_pos.page_pos + page_shift,
855 MSG_DONTWAIT | MSG_NOSIGNAL |
859 (msg->pages || msg->pagelist || msg->bio || in_trail))
867 con->out_msg_pos.data_pos += ret;
868 con->out_msg_pos.page_pos += ret;
870 con->out_msg_pos.page_pos = 0;
871 con->out_msg_pos.page++;
872 con->out_msg_pos.did_page_crc = 0;
874 list_move_tail(&page->lru,
876 else if (msg->pagelist)
877 list_move_tail(&page->lru,
878 &msg->pagelist->head);
881 iter_bio_next(&msg->bio_iter, &msg->bio_seg);
886 dout("write_partial_msg_pages %p msg %p done\n", con, msg);
888 /* prepare and queue up footer, too */
890 con->out_msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
891 con->out_kvec_bytes = 0;
892 con->out_kvec_left = 0;
893 con->out_kvec_cur = con->out_kvec;
894 prepare_write_message_footer(con, 0);
903 static int write_partial_skip(struct ceph_connection *con)
907 while (con->out_skip > 0) {
909 .iov_base = page_address(con->msgr->zero_page),
910 .iov_len = min(con->out_skip, (int)PAGE_CACHE_SIZE)
913 ret = ceph_tcp_sendmsg(con->sock, &iov, 1, iov.iov_len, 1);
916 con->out_skip -= ret;
924 * Prepare to read connection handshake, or an ack.
926 static void prepare_read_banner(struct ceph_connection *con)
928 dout("prepare_read_banner %p\n", con);
929 con->in_base_pos = 0;
932 static void prepare_read_connect(struct ceph_connection *con)
934 dout("prepare_read_connect %p\n", con);
935 con->in_base_pos = 0;
938 static void prepare_read_ack(struct ceph_connection *con)
940 dout("prepare_read_ack %p\n", con);
941 con->in_base_pos = 0;
944 static void prepare_read_tag(struct ceph_connection *con)
946 dout("prepare_read_tag %p\n", con);
947 con->in_base_pos = 0;
948 con->in_tag = CEPH_MSGR_TAG_READY;
952 * Prepare to read a message.
954 static int prepare_read_message(struct ceph_connection *con)
956 dout("prepare_read_message %p\n", con);
957 BUG_ON(con->in_msg != NULL);
958 con->in_base_pos = 0;
959 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
964 static int read_partial(struct ceph_connection *con,
965 int *to, int size, void *object)
968 while (con->in_base_pos < *to) {
969 int left = *to - con->in_base_pos;
970 int have = size - left;
971 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
974 con->in_base_pos += ret;
981 * Read all or part of the connect-side handshake on a new connection
983 static int read_partial_banner(struct ceph_connection *con)
987 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
990 ret = read_partial(con, &to, strlen(CEPH_BANNER), con->in_banner);
993 ret = read_partial(con, &to, sizeof(con->actual_peer_addr),
994 &con->actual_peer_addr);
997 ret = read_partial(con, &to, sizeof(con->peer_addr_for_me),
998 &con->peer_addr_for_me);
1005 static int read_partial_connect(struct ceph_connection *con)
1009 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1011 ret = read_partial(con, &to, sizeof(con->in_reply), &con->in_reply);
1014 ret = read_partial(con, &to, le32_to_cpu(con->in_reply.authorizer_len),
1015 con->auth_reply_buf);
1019 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1020 con, (int)con->in_reply.tag,
1021 le32_to_cpu(con->in_reply.connect_seq),
1022 le32_to_cpu(con->in_reply.global_seq));
1029 * Verify the hello banner looks okay.
1031 static int verify_hello(struct ceph_connection *con)
1033 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1034 pr_err("connect to %s got bad banner\n",
1035 ceph_pr_addr(&con->peer_addr.in_addr));
1036 con->error_msg = "protocol error, bad banner";
1042 static bool addr_is_blank(struct sockaddr_storage *ss)
1044 switch (ss->ss_family) {
1046 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1049 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1050 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1051 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1052 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1057 static int addr_port(struct sockaddr_storage *ss)
1059 switch (ss->ss_family) {
1061 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1063 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1068 static void addr_set_port(struct sockaddr_storage *ss, int p)
1070 switch (ss->ss_family) {
1072 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1075 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1081 * Parse an ip[:port] list into an addr array. Use the default
1082 * monitor port if a port isn't specified.
1084 int ceph_parse_ips(const char *c, const char *end,
1085 struct ceph_entity_addr *addr,
1086 int max_count, int *count)
1091 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1092 for (i = 0; i < max_count; i++) {
1094 struct sockaddr_storage *ss = &addr[i].in_addr;
1095 struct sockaddr_in *in4 = (void *)ss;
1096 struct sockaddr_in6 *in6 = (void *)ss;
1105 memset(ss, 0, sizeof(*ss));
1106 if (in4_pton(p, end - p, (u8 *)&in4->sin_addr.s_addr,
1108 ss->ss_family = AF_INET;
1109 else if (in6_pton(p, end - p, (u8 *)&in6->sin6_addr.s6_addr,
1111 ss->ss_family = AF_INET6;
1118 dout("missing matching ']'\n");
1125 if (p < end && *p == ':') {
1128 while (p < end && *p >= '0' && *p <= '9') {
1129 port = (port * 10) + (*p - '0');
1132 if (port > 65535 || port == 0)
1135 port = CEPH_MON_PORT;
1138 addr_set_port(ss, port);
1140 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1157 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1160 EXPORT_SYMBOL(ceph_parse_ips);
1162 static int process_banner(struct ceph_connection *con)
1164 dout("process_banner on %p\n", con);
1166 if (verify_hello(con) < 0)
1169 ceph_decode_addr(&con->actual_peer_addr);
1170 ceph_decode_addr(&con->peer_addr_for_me);
1173 * Make sure the other end is who we wanted. note that the other
1174 * end may not yet know their ip address, so if it's 0.0.0.0, give
1175 * them the benefit of the doubt.
1177 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1178 sizeof(con->peer_addr)) != 0 &&
1179 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1180 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1181 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1182 ceph_pr_addr(&con->peer_addr.in_addr),
1183 (int)le32_to_cpu(con->peer_addr.nonce),
1184 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1185 (int)le32_to_cpu(con->actual_peer_addr.nonce));
1186 con->error_msg = "wrong peer at address";
1191 * did we learn our address?
1193 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1194 int port = addr_port(&con->msgr->inst.addr.in_addr);
1196 memcpy(&con->msgr->inst.addr.in_addr,
1197 &con->peer_addr_for_me.in_addr,
1198 sizeof(con->peer_addr_for_me.in_addr));
1199 addr_set_port(&con->msgr->inst.addr.in_addr, port);
1200 encode_my_addr(con->msgr);
1201 dout("process_banner learned my addr is %s\n",
1202 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
1205 set_bit(NEGOTIATING, &con->state);
1206 prepare_read_connect(con);
1210 static void fail_protocol(struct ceph_connection *con)
1212 reset_connection(con);
1213 set_bit(CLOSED, &con->state); /* in case there's queued work */
1215 mutex_unlock(&con->mutex);
1216 if (con->ops->bad_proto)
1217 con->ops->bad_proto(con);
1218 mutex_lock(&con->mutex);
1221 static int process_connect(struct ceph_connection *con)
1223 u64 sup_feat = con->msgr->supported_features;
1224 u64 req_feat = con->msgr->required_features;
1225 u64 server_feat = le64_to_cpu(con->in_reply.features);
1228 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1230 switch (con->in_reply.tag) {
1231 case CEPH_MSGR_TAG_FEATURES:
1232 pr_err("%s%lld %s feature set mismatch,"
1233 " my %llx < server's %llx, missing %llx\n",
1234 ENTITY_NAME(con->peer_name),
1235 ceph_pr_addr(&con->peer_addr.in_addr),
1236 sup_feat, server_feat, server_feat & ~sup_feat);
1237 con->error_msg = "missing required protocol features";
1241 case CEPH_MSGR_TAG_BADPROTOVER:
1242 pr_err("%s%lld %s protocol version mismatch,"
1243 " my %d != server's %d\n",
1244 ENTITY_NAME(con->peer_name),
1245 ceph_pr_addr(&con->peer_addr.in_addr),
1246 le32_to_cpu(con->out_connect.protocol_version),
1247 le32_to_cpu(con->in_reply.protocol_version));
1248 con->error_msg = "protocol version mismatch";
1252 case CEPH_MSGR_TAG_BADAUTHORIZER:
1254 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1256 if (con->auth_retry == 2) {
1257 con->error_msg = "connect authorization failure";
1260 con->auth_retry = 1;
1261 ret = prepare_write_connect(con->msgr, con, 0);
1264 prepare_read_connect(con);
1267 case CEPH_MSGR_TAG_RESETSESSION:
1269 * If we connected with a large connect_seq but the peer
1270 * has no record of a session with us (no connection, or
1271 * connect_seq == 0), they will send RESETSESION to indicate
1272 * that they must have reset their session, and may have
1275 dout("process_connect got RESET peer seq %u\n",
1276 le32_to_cpu(con->in_connect.connect_seq));
1277 pr_err("%s%lld %s connection reset\n",
1278 ENTITY_NAME(con->peer_name),
1279 ceph_pr_addr(&con->peer_addr.in_addr));
1280 reset_connection(con);
1281 prepare_write_connect(con->msgr, con, 0);
1282 prepare_read_connect(con);
1284 /* Tell ceph about it. */
1285 mutex_unlock(&con->mutex);
1286 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
1287 if (con->ops->peer_reset)
1288 con->ops->peer_reset(con);
1289 mutex_lock(&con->mutex);
1290 if (test_bit(CLOSED, &con->state) ||
1291 test_bit(OPENING, &con->state))
1295 case CEPH_MSGR_TAG_RETRY_SESSION:
1297 * If we sent a smaller connect_seq than the peer has, try
1298 * again with a larger value.
1300 dout("process_connect got RETRY my seq = %u, peer_seq = %u\n",
1301 le32_to_cpu(con->out_connect.connect_seq),
1302 le32_to_cpu(con->in_connect.connect_seq));
1303 con->connect_seq = le32_to_cpu(con->in_connect.connect_seq);
1304 prepare_write_connect(con->msgr, con, 0);
1305 prepare_read_connect(con);
1308 case CEPH_MSGR_TAG_RETRY_GLOBAL:
1310 * If we sent a smaller global_seq than the peer has, try
1311 * again with a larger value.
1313 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1314 con->peer_global_seq,
1315 le32_to_cpu(con->in_connect.global_seq));
1316 get_global_seq(con->msgr,
1317 le32_to_cpu(con->in_connect.global_seq));
1318 prepare_write_connect(con->msgr, con, 0);
1319 prepare_read_connect(con);
1322 case CEPH_MSGR_TAG_READY:
1323 if (req_feat & ~server_feat) {
1324 pr_err("%s%lld %s protocol feature mismatch,"
1325 " my required %llx > server's %llx, need %llx\n",
1326 ENTITY_NAME(con->peer_name),
1327 ceph_pr_addr(&con->peer_addr.in_addr),
1328 req_feat, server_feat, req_feat & ~server_feat);
1329 con->error_msg = "missing required protocol features";
1333 clear_bit(CONNECTING, &con->state);
1334 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
1336 con->peer_features = server_feat;
1337 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1338 con->peer_global_seq,
1339 le32_to_cpu(con->in_reply.connect_seq),
1341 WARN_ON(con->connect_seq !=
1342 le32_to_cpu(con->in_reply.connect_seq));
1344 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
1345 set_bit(LOSSYTX, &con->state);
1347 prepare_read_tag(con);
1350 case CEPH_MSGR_TAG_WAIT:
1352 * If there is a connection race (we are opening
1353 * connections to each other), one of us may just have
1354 * to WAIT. This shouldn't happen if we are the
1357 pr_err("process_connect got WAIT as client\n");
1358 con->error_msg = "protocol error, got WAIT as client";
1362 pr_err("connect protocol error, will retry\n");
1363 con->error_msg = "protocol error, garbage tag during connect";
1371 * read (part of) an ack
1373 static int read_partial_ack(struct ceph_connection *con)
1377 return read_partial(con, &to, sizeof(con->in_temp_ack),
1383 * We can finally discard anything that's been acked.
1385 static void process_ack(struct ceph_connection *con)
1388 u64 ack = le64_to_cpu(con->in_temp_ack);
1391 while (!list_empty(&con->out_sent)) {
1392 m = list_first_entry(&con->out_sent, struct ceph_msg,
1394 seq = le64_to_cpu(m->hdr.seq);
1397 dout("got ack for seq %llu type %d at %p\n", seq,
1398 le16_to_cpu(m->hdr.type), m);
1399 m->ack_stamp = jiffies;
1402 prepare_read_tag(con);
1408 static int read_partial_message_section(struct ceph_connection *con,
1409 struct kvec *section,
1410 unsigned int sec_len, u32 *crc)
1416 while (section->iov_len < sec_len) {
1417 BUG_ON(section->iov_base == NULL);
1418 left = sec_len - section->iov_len;
1419 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
1420 section->iov_len, left);
1423 section->iov_len += ret;
1424 if (section->iov_len == sec_len)
1425 *crc = crc32c(0, section->iov_base,
1432 static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con,
1433 struct ceph_msg_header *hdr,
1437 static int read_partial_message_pages(struct ceph_connection *con,
1438 struct page **pages,
1439 unsigned data_len, int datacrc)
1445 left = min((int)(data_len - con->in_msg_pos.data_pos),
1446 (int)(PAGE_SIZE - con->in_msg_pos.page_pos));
1448 BUG_ON(pages == NULL);
1449 p = kmap(pages[con->in_msg_pos.page]);
1450 ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos,
1452 if (ret > 0 && datacrc)
1454 crc32c(con->in_data_crc,
1455 p + con->in_msg_pos.page_pos, ret);
1456 kunmap(pages[con->in_msg_pos.page]);
1459 con->in_msg_pos.data_pos += ret;
1460 con->in_msg_pos.page_pos += ret;
1461 if (con->in_msg_pos.page_pos == PAGE_SIZE) {
1462 con->in_msg_pos.page_pos = 0;
1463 con->in_msg_pos.page++;
1470 static int read_partial_message_bio(struct ceph_connection *con,
1471 struct bio **bio_iter, int *bio_seg,
1472 unsigned data_len, int datacrc)
1474 struct bio_vec *bv = bio_iovec_idx(*bio_iter, *bio_seg);
1481 left = min((int)(data_len - con->in_msg_pos.data_pos),
1482 (int)(bv->bv_len - con->in_msg_pos.page_pos));
1484 p = kmap(bv->bv_page) + bv->bv_offset;
1486 ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos,
1488 if (ret > 0 && datacrc)
1490 crc32c(con->in_data_crc,
1491 p + con->in_msg_pos.page_pos, ret);
1492 kunmap(bv->bv_page);
1495 con->in_msg_pos.data_pos += ret;
1496 con->in_msg_pos.page_pos += ret;
1497 if (con->in_msg_pos.page_pos == bv->bv_len) {
1498 con->in_msg_pos.page_pos = 0;
1499 iter_bio_next(bio_iter, bio_seg);
1507 * read (part of) a message.
1509 static int read_partial_message(struct ceph_connection *con)
1511 struct ceph_msg *m = con->in_msg;
1514 unsigned front_len, middle_len, data_len;
1515 int datacrc = con->msgr->nocrc;
1519 dout("read_partial_message con %p msg %p\n", con, m);
1522 while (con->in_base_pos < sizeof(con->in_hdr)) {
1523 left = sizeof(con->in_hdr) - con->in_base_pos;
1524 ret = ceph_tcp_recvmsg(con->sock,
1525 (char *)&con->in_hdr + con->in_base_pos,
1529 con->in_base_pos += ret;
1530 if (con->in_base_pos == sizeof(con->in_hdr)) {
1531 u32 crc = crc32c(0, (void *)&con->in_hdr,
1532 sizeof(con->in_hdr) - sizeof(con->in_hdr.crc));
1533 if (crc != le32_to_cpu(con->in_hdr.crc)) {
1534 pr_err("read_partial_message bad hdr "
1535 " crc %u != expected %u\n",
1536 crc, con->in_hdr.crc);
1541 front_len = le32_to_cpu(con->in_hdr.front_len);
1542 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
1544 middle_len = le32_to_cpu(con->in_hdr.middle_len);
1545 if (middle_len > CEPH_MSG_MAX_DATA_LEN)
1547 data_len = le32_to_cpu(con->in_hdr.data_len);
1548 if (data_len > CEPH_MSG_MAX_DATA_LEN)
1552 seq = le64_to_cpu(con->in_hdr.seq);
1553 if ((s64)seq - (s64)con->in_seq < 1) {
1554 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1555 ENTITY_NAME(con->peer_name),
1556 ceph_pr_addr(&con->peer_addr.in_addr),
1557 seq, con->in_seq + 1);
1558 con->in_base_pos = -front_len - middle_len - data_len -
1560 con->in_tag = CEPH_MSGR_TAG_READY;
1562 } else if ((s64)seq - (s64)con->in_seq > 1) {
1563 pr_err("read_partial_message bad seq %lld expected %lld\n",
1564 seq, con->in_seq + 1);
1565 con->error_msg = "bad message sequence # for incoming message";
1569 /* allocate message? */
1571 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
1572 con->in_hdr.front_len, con->in_hdr.data_len);
1574 con->in_msg = ceph_alloc_msg(con, &con->in_hdr, &skip);
1576 /* skip this message */
1577 dout("alloc_msg said skip message\n");
1578 BUG_ON(con->in_msg);
1579 con->in_base_pos = -front_len - middle_len - data_len -
1581 con->in_tag = CEPH_MSGR_TAG_READY;
1587 "error allocating memory for incoming message";
1591 m->front.iov_len = 0; /* haven't read it yet */
1593 m->middle->vec.iov_len = 0;
1595 con->in_msg_pos.page = 0;
1597 con->in_msg_pos.page_pos = m->page_alignment;
1599 con->in_msg_pos.page_pos = 0;
1600 con->in_msg_pos.data_pos = 0;
1604 ret = read_partial_message_section(con, &m->front, front_len,
1605 &con->in_front_crc);
1611 ret = read_partial_message_section(con, &m->middle->vec,
1613 &con->in_middle_crc);
1618 if (m->bio && !m->bio_iter)
1619 init_bio_iter(m->bio, &m->bio_iter, &m->bio_seg);
1623 while (con->in_msg_pos.data_pos < data_len) {
1625 ret = read_partial_message_pages(con, m->pages,
1630 } else if (m->bio) {
1632 ret = read_partial_message_bio(con,
1633 &m->bio_iter, &m->bio_seg,
1644 to = sizeof(m->hdr) + sizeof(m->footer);
1645 while (con->in_base_pos < to) {
1646 left = to - con->in_base_pos;
1647 ret = ceph_tcp_recvmsg(con->sock, (char *)&m->footer +
1648 (con->in_base_pos - sizeof(m->hdr)),
1652 con->in_base_pos += ret;
1654 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1655 m, front_len, m->footer.front_crc, middle_len,
1656 m->footer.middle_crc, data_len, m->footer.data_crc);
1659 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
1660 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1661 m, con->in_front_crc, m->footer.front_crc);
1664 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
1665 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1666 m, con->in_middle_crc, m->footer.middle_crc);
1670 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
1671 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
1672 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
1673 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
1677 return 1; /* done! */
1681 * Process message. This happens in the worker thread. The callback should
1682 * be careful not to do anything that waits on other incoming messages or it
1685 static void process_message(struct ceph_connection *con)
1687 struct ceph_msg *msg;
1692 /* if first message, set peer_name */
1693 if (con->peer_name.type == 0)
1694 con->peer_name = msg->hdr.src;
1697 mutex_unlock(&con->mutex);
1699 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
1700 msg, le64_to_cpu(msg->hdr.seq),
1701 ENTITY_NAME(msg->hdr.src),
1702 le16_to_cpu(msg->hdr.type),
1703 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
1704 le32_to_cpu(msg->hdr.front_len),
1705 le32_to_cpu(msg->hdr.data_len),
1706 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
1707 con->ops->dispatch(con, msg);
1709 mutex_lock(&con->mutex);
1710 prepare_read_tag(con);
1715 * Write something to the socket. Called in a worker thread when the
1716 * socket appears to be writeable and we have something ready to send.
1718 static int try_write(struct ceph_connection *con)
1720 struct ceph_messenger *msgr = con->msgr;
1723 dout("try_write start %p state %lu nref %d\n", con, con->state,
1724 atomic_read(&con->nref));
1727 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
1729 /* open the socket first? */
1730 if (con->sock == NULL) {
1731 prepare_write_banner(msgr, con);
1732 prepare_write_connect(msgr, con, 1);
1733 prepare_read_banner(con);
1734 set_bit(CONNECTING, &con->state);
1735 clear_bit(NEGOTIATING, &con->state);
1737 BUG_ON(con->in_msg);
1738 con->in_tag = CEPH_MSGR_TAG_READY;
1739 dout("try_write initiating connect on %p new state %lu\n",
1741 con->sock = ceph_tcp_connect(con);
1742 if (IS_ERR(con->sock)) {
1744 con->error_msg = "connect error";
1751 /* kvec data queued? */
1752 if (con->out_skip) {
1753 ret = write_partial_skip(con);
1757 if (con->out_kvec_left) {
1758 ret = write_partial_kvec(con);
1765 if (con->out_msg_done) {
1766 ceph_msg_put(con->out_msg);
1767 con->out_msg = NULL; /* we're done with this one */
1771 ret = write_partial_msg_pages(con);
1773 goto more_kvec; /* we need to send the footer, too! */
1777 dout("try_write write_partial_msg_pages err %d\n",
1784 if (!test_bit(CONNECTING, &con->state)) {
1785 /* is anything else pending? */
1786 if (!list_empty(&con->out_queue)) {
1787 prepare_write_message(con);
1790 if (con->in_seq > con->in_seq_acked) {
1791 prepare_write_ack(con);
1794 if (test_and_clear_bit(KEEPALIVE_PENDING, &con->state)) {
1795 prepare_write_keepalive(con);
1800 /* Nothing to do! */
1801 clear_bit(WRITE_PENDING, &con->state);
1802 dout("try_write nothing else to write.\n");
1805 dout("try_write done on %p ret %d\n", con, ret);
1812 * Read what we can from the socket.
1814 static int try_read(struct ceph_connection *con)
1821 if (test_bit(STANDBY, &con->state))
1824 dout("try_read start on %p\n", con);
1827 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
1831 * process_connect and process_message drop and re-take
1832 * con->mutex. make sure we handle a racing close or reopen.
1834 if (test_bit(CLOSED, &con->state) ||
1835 test_bit(OPENING, &con->state)) {
1840 if (test_bit(CONNECTING, &con->state)) {
1841 if (!test_bit(NEGOTIATING, &con->state)) {
1842 dout("try_read connecting\n");
1843 ret = read_partial_banner(con);
1846 ret = process_banner(con);
1850 ret = read_partial_connect(con);
1853 ret = process_connect(con);
1859 if (con->in_base_pos < 0) {
1861 * skipping + discarding content.
1863 * FIXME: there must be a better way to do this!
1865 static char buf[1024];
1866 int skip = min(1024, -con->in_base_pos);
1867 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
1868 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
1871 con->in_base_pos += ret;
1872 if (con->in_base_pos)
1875 if (con->in_tag == CEPH_MSGR_TAG_READY) {
1879 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
1882 dout("try_read got tag %d\n", (int)con->in_tag);
1883 switch (con->in_tag) {
1884 case CEPH_MSGR_TAG_MSG:
1885 prepare_read_message(con);
1887 case CEPH_MSGR_TAG_ACK:
1888 prepare_read_ack(con);
1890 case CEPH_MSGR_TAG_CLOSE:
1891 set_bit(CLOSED, &con->state); /* fixme */
1897 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
1898 ret = read_partial_message(con);
1902 con->error_msg = "bad crc";
1906 con->error_msg = "io error";
1911 if (con->in_tag == CEPH_MSGR_TAG_READY)
1913 process_message(con);
1916 if (con->in_tag == CEPH_MSGR_TAG_ACK) {
1917 ret = read_partial_ack(con);
1925 dout("try_read done on %p ret %d\n", con, ret);
1929 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
1930 con->error_msg = "protocol error, garbage tag";
1937 * Atomically queue work on a connection. Bump @con reference to
1938 * avoid races with connection teardown.
1940 static void queue_con(struct ceph_connection *con)
1942 if (test_bit(DEAD, &con->state)) {
1943 dout("queue_con %p ignoring: DEAD\n",
1948 if (!con->ops->get(con)) {
1949 dout("queue_con %p ref count 0\n", con);
1953 if (!queue_delayed_work(ceph_msgr_wq, &con->work, 0)) {
1954 dout("queue_con %p - already queued\n", con);
1957 dout("queue_con %p\n", con);
1962 * Do some work on a connection. Drop a connection ref when we're done.
1964 static void con_work(struct work_struct *work)
1966 struct ceph_connection *con = container_of(work, struct ceph_connection,
1970 mutex_lock(&con->mutex);
1972 if (test_and_clear_bit(BACKOFF, &con->state)) {
1973 dout("con_work %p backing off\n", con);
1974 if (queue_delayed_work(ceph_msgr_wq, &con->work,
1975 round_jiffies_relative(con->delay))) {
1976 dout("con_work %p backoff %lu\n", con, con->delay);
1977 mutex_unlock(&con->mutex);
1981 dout("con_work %p FAILED to back off %lu\n", con,
1986 if (test_bit(STANDBY, &con->state)) {
1987 dout("con_work %p STANDBY\n", con);
1990 if (test_bit(CLOSED, &con->state)) { /* e.g. if we are replaced */
1991 dout("con_work CLOSED\n");
1992 con_close_socket(con);
1995 if (test_and_clear_bit(OPENING, &con->state)) {
1996 /* reopen w/ new peer */
1997 dout("con_work OPENING\n");
1998 con_close_socket(con);
2001 if (test_and_clear_bit(SOCK_CLOSED, &con->state))
2004 ret = try_read(con);
2010 ret = try_write(con);
2017 mutex_unlock(&con->mutex);
2023 mutex_unlock(&con->mutex);
2024 ceph_fault(con); /* error/fault path */
2030 * Generic error/fault handler. A retry mechanism is used with
2031 * exponential backoff
2033 static void ceph_fault(struct ceph_connection *con)
2035 pr_err("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2036 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2037 dout("fault %p state %lu to peer %s\n",
2038 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2040 if (test_bit(LOSSYTX, &con->state)) {
2041 dout("fault on LOSSYTX channel\n");
2045 mutex_lock(&con->mutex);
2046 if (test_bit(CLOSED, &con->state))
2049 con_close_socket(con);
2052 ceph_msg_put(con->in_msg);
2056 /* Requeue anything that hasn't been acked */
2057 list_splice_init(&con->out_sent, &con->out_queue);
2059 /* If there are no messages queued or keepalive pending, place
2060 * the connection in a STANDBY state */
2061 if (list_empty(&con->out_queue) &&
2062 !test_bit(KEEPALIVE_PENDING, &con->state)) {
2063 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2064 clear_bit(WRITE_PENDING, &con->state);
2065 set_bit(STANDBY, &con->state);
2067 /* retry after a delay. */
2068 if (con->delay == 0)
2069 con->delay = BASE_DELAY_INTERVAL;
2070 else if (con->delay < MAX_DELAY_INTERVAL)
2073 if (queue_delayed_work(ceph_msgr_wq, &con->work,
2074 round_jiffies_relative(con->delay))) {
2075 dout("fault queued %p delay %lu\n", con, con->delay);
2078 dout("fault failed to queue %p delay %lu, backoff\n",
2081 * In many cases we see a socket state change
2082 * while con_work is running and end up
2083 * queuing (non-delayed) work, such that we
2084 * can't backoff with a delay. Set a flag so
2085 * that when con_work restarts we schedule the
2088 set_bit(BACKOFF, &con->state);
2093 mutex_unlock(&con->mutex);
2096 * in case we faulted due to authentication, invalidate our
2097 * current tickets so that we can get new ones.
2099 if (con->auth_retry && con->ops->invalidate_authorizer) {
2100 dout("calling invalidate_authorizer()\n");
2101 con->ops->invalidate_authorizer(con);
2104 if (con->ops->fault)
2105 con->ops->fault(con);
2111 * create a new messenger instance
2113 struct ceph_messenger *ceph_messenger_create(struct ceph_entity_addr *myaddr,
2114 u32 supported_features,
2115 u32 required_features)
2117 struct ceph_messenger *msgr;
2119 msgr = kzalloc(sizeof(*msgr), GFP_KERNEL);
2121 return ERR_PTR(-ENOMEM);
2123 msgr->supported_features = supported_features;
2124 msgr->required_features = required_features;
2126 spin_lock_init(&msgr->global_seq_lock);
2128 /* the zero page is needed if a request is "canceled" while the message
2129 * is being written over the socket */
2130 msgr->zero_page = __page_cache_alloc(GFP_KERNEL | __GFP_ZERO);
2131 if (!msgr->zero_page) {
2133 return ERR_PTR(-ENOMEM);
2135 kmap(msgr->zero_page);
2138 msgr->inst.addr = *myaddr;
2140 /* select a random nonce */
2141 msgr->inst.addr.type = 0;
2142 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2143 encode_my_addr(msgr);
2145 dout("messenger_create %p\n", msgr);
2148 EXPORT_SYMBOL(ceph_messenger_create);
2150 void ceph_messenger_destroy(struct ceph_messenger *msgr)
2152 dout("destroy %p\n", msgr);
2153 kunmap(msgr->zero_page);
2154 __free_page(msgr->zero_page);
2156 dout("destroyed messenger %p\n", msgr);
2158 EXPORT_SYMBOL(ceph_messenger_destroy);
2160 static void clear_standby(struct ceph_connection *con)
2162 /* come back from STANDBY? */
2163 if (test_and_clear_bit(STANDBY, &con->state)) {
2164 mutex_lock(&con->mutex);
2165 dout("clear_standby %p and ++connect_seq\n", con);
2167 WARN_ON(test_bit(WRITE_PENDING, &con->state));
2168 WARN_ON(test_bit(KEEPALIVE_PENDING, &con->state));
2169 mutex_unlock(&con->mutex);
2174 * Queue up an outgoing message on the given connection.
2176 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2178 if (test_bit(CLOSED, &con->state)) {
2179 dout("con_send %p closed, dropping %p\n", con, msg);
2185 msg->hdr.src = con->msgr->inst.name;
2187 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2189 msg->needs_out_seq = true;
2192 mutex_lock(&con->mutex);
2193 BUG_ON(!list_empty(&msg->list_head));
2194 list_add_tail(&msg->list_head, &con->out_queue);
2195 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2196 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2197 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2198 le32_to_cpu(msg->hdr.front_len),
2199 le32_to_cpu(msg->hdr.middle_len),
2200 le32_to_cpu(msg->hdr.data_len));
2201 mutex_unlock(&con->mutex);
2203 /* if there wasn't anything waiting to send before, queue
2206 if (test_and_set_bit(WRITE_PENDING, &con->state) == 0)
2209 EXPORT_SYMBOL(ceph_con_send);
2212 * Revoke a message that was previously queued for send
2214 void ceph_con_revoke(struct ceph_connection *con, struct ceph_msg *msg)
2216 mutex_lock(&con->mutex);
2217 if (!list_empty(&msg->list_head)) {
2218 dout("con_revoke %p msg %p - was on queue\n", con, msg);
2219 list_del_init(&msg->list_head);
2223 if (con->out_msg == msg) {
2224 dout("con_revoke %p msg %p - was sending\n", con, msg);
2225 con->out_msg = NULL;
2226 if (con->out_kvec_is_msg) {
2227 con->out_skip = con->out_kvec_bytes;
2228 con->out_kvec_is_msg = false;
2233 mutex_unlock(&con->mutex);
2237 * Revoke a message that we may be reading data into
2239 void ceph_con_revoke_message(struct ceph_connection *con, struct ceph_msg *msg)
2241 mutex_lock(&con->mutex);
2242 if (con->in_msg && con->in_msg == msg) {
2243 unsigned front_len = le32_to_cpu(con->in_hdr.front_len);
2244 unsigned middle_len = le32_to_cpu(con->in_hdr.middle_len);
2245 unsigned data_len = le32_to_cpu(con->in_hdr.data_len);
2247 /* skip rest of message */
2248 dout("con_revoke_pages %p msg %p revoked\n", con, msg);
2249 con->in_base_pos = con->in_base_pos -
2250 sizeof(struct ceph_msg_header) -
2254 sizeof(struct ceph_msg_footer);
2255 ceph_msg_put(con->in_msg);
2257 con->in_tag = CEPH_MSGR_TAG_READY;
2260 dout("con_revoke_pages %p msg %p pages %p no-op\n",
2261 con, con->in_msg, msg);
2263 mutex_unlock(&con->mutex);
2267 * Queue a keepalive byte to ensure the tcp connection is alive.
2269 void ceph_con_keepalive(struct ceph_connection *con)
2271 dout("con_keepalive %p\n", con);
2273 if (test_and_set_bit(KEEPALIVE_PENDING, &con->state) == 0 &&
2274 test_and_set_bit(WRITE_PENDING, &con->state) == 0)
2277 EXPORT_SYMBOL(ceph_con_keepalive);
2281 * construct a new message with given type, size
2282 * the new msg has a ref count of 1.
2284 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags)
2288 m = kmalloc(sizeof(*m), flags);
2291 kref_init(&m->kref);
2292 INIT_LIST_HEAD(&m->list_head);
2295 m->hdr.type = cpu_to_le16(type);
2296 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
2298 m->hdr.front_len = cpu_to_le32(front_len);
2299 m->hdr.middle_len = 0;
2300 m->hdr.data_len = 0;
2301 m->hdr.data_off = 0;
2302 m->hdr.reserved = 0;
2303 m->footer.front_crc = 0;
2304 m->footer.middle_crc = 0;
2305 m->footer.data_crc = 0;
2306 m->footer.flags = 0;
2307 m->front_max = front_len;
2308 m->front_is_vmalloc = false;
2309 m->more_to_follow = false;
2317 m->page_alignment = 0;
2327 if (front_len > PAGE_CACHE_SIZE) {
2328 m->front.iov_base = __vmalloc(front_len, flags,
2330 m->front_is_vmalloc = true;
2332 m->front.iov_base = kmalloc(front_len, flags);
2334 if (m->front.iov_base == NULL) {
2335 pr_err("msg_new can't allocate %d bytes\n",
2340 m->front.iov_base = NULL;
2342 m->front.iov_len = front_len;
2344 dout("ceph_msg_new %p front %d\n", m, front_len);
2350 pr_err("msg_new can't create type %d front %d\n", type, front_len);
2353 EXPORT_SYMBOL(ceph_msg_new);
2356 * Allocate "middle" portion of a message, if it is needed and wasn't
2357 * allocated by alloc_msg. This allows us to read a small fixed-size
2358 * per-type header in the front and then gracefully fail (i.e.,
2359 * propagate the error to the caller based on info in the front) when
2360 * the middle is too large.
2362 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
2364 int type = le16_to_cpu(msg->hdr.type);
2365 int middle_len = le32_to_cpu(msg->hdr.middle_len);
2367 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
2368 ceph_msg_type_name(type), middle_len);
2369 BUG_ON(!middle_len);
2370 BUG_ON(msg->middle);
2372 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
2379 * Generic message allocator, for incoming messages.
2381 static struct ceph_msg *ceph_alloc_msg(struct ceph_connection *con,
2382 struct ceph_msg_header *hdr,
2385 int type = le16_to_cpu(hdr->type);
2386 int front_len = le32_to_cpu(hdr->front_len);
2387 int middle_len = le32_to_cpu(hdr->middle_len);
2388 struct ceph_msg *msg = NULL;
2391 if (con->ops->alloc_msg) {
2392 mutex_unlock(&con->mutex);
2393 msg = con->ops->alloc_msg(con, hdr, skip);
2394 mutex_lock(&con->mutex);
2400 msg = ceph_msg_new(type, front_len, GFP_NOFS);
2402 pr_err("unable to allocate msg type %d len %d\n",
2406 msg->page_alignment = le16_to_cpu(hdr->data_off);
2408 memcpy(&msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2410 if (middle_len && !msg->middle) {
2411 ret = ceph_alloc_middle(con, msg);
2423 * Free a generically kmalloc'd message.
2425 void ceph_msg_kfree(struct ceph_msg *m)
2427 dout("msg_kfree %p\n", m);
2428 if (m->front_is_vmalloc)
2429 vfree(m->front.iov_base);
2431 kfree(m->front.iov_base);
2436 * Drop a msg ref. Destroy as needed.
2438 void ceph_msg_last_put(struct kref *kref)
2440 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
2442 dout("ceph_msg_put last one on %p\n", m);
2443 WARN_ON(!list_empty(&m->list_head));
2445 /* drop middle, data, if any */
2447 ceph_buffer_put(m->middle);
2454 ceph_pagelist_release(m->pagelist);
2462 ceph_msgpool_put(m->pool, m);
2466 EXPORT_SYMBOL(ceph_msg_last_put);
2468 void ceph_msg_dump(struct ceph_msg *msg)
2470 pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg,
2471 msg->front_max, msg->nr_pages);
2472 print_hex_dump(KERN_DEBUG, "header: ",
2473 DUMP_PREFIX_OFFSET, 16, 1,
2474 &msg->hdr, sizeof(msg->hdr), true);
2475 print_hex_dump(KERN_DEBUG, " front: ",
2476 DUMP_PREFIX_OFFSET, 16, 1,
2477 msg->front.iov_base, msg->front.iov_len, true);
2479 print_hex_dump(KERN_DEBUG, "middle: ",
2480 DUMP_PREFIX_OFFSET, 16, 1,
2481 msg->middle->vec.iov_base,
2482 msg->middle->vec.iov_len, true);
2483 print_hex_dump(KERN_DEBUG, "footer: ",
2484 DUMP_PREFIX_OFFSET, 16, 1,
2485 &msg->footer, sizeof(msg->footer), true);
2487 EXPORT_SYMBOL(ceph_msg_dump);