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>
13 #include <linux/bio.h>
14 #endif /* CONFIG_BLOCK */
15 #include <linux/dns_resolver.h>
18 #include <linux/ceph/libceph.h>
19 #include <linux/ceph/messenger.h>
20 #include <linux/ceph/decode.h>
21 #include <linux/ceph/pagelist.h>
22 #include <linux/export.h>
24 #define list_entry_next(pos, member) \
25 list_entry(pos->member.next, typeof(*pos), member)
28 * Ceph uses the messenger to exchange ceph_msg messages with other
29 * hosts in the system. The messenger provides ordered and reliable
30 * delivery. We tolerate TCP disconnects by reconnecting (with
31 * exponential backoff) in the case of a fault (disconnection, bad
32 * crc, protocol error). Acks allow sent messages to be discarded by
37 * We track the state of the socket on a given connection using
38 * values defined below. The transition to a new socket state is
39 * handled by a function which verifies we aren't coming from an
43 * | NEW* | transient initial state
45 * | con_sock_state_init()
48 * | CLOSED | initialized, but no socket (and no
49 * ---------- TCP connection)
51 * | \ con_sock_state_connecting()
52 * | ----------------------
54 * + con_sock_state_closed() \
55 * |+--------------------------- \
58 * | | CLOSING | socket event; \ \
59 * | ----------- await close \ \
62 * | + con_sock_state_closing() \ |
64 * | / --------------- | |
67 * | / -----------------| CONNECTING | socket created, TCP
68 * | | / -------------- connect initiated
69 * | | | con_sock_state_connected()
72 * | CONNECTED | TCP connection established
75 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
78 #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
79 #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
80 #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
81 #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
82 #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
87 #define CON_STATE_CLOSED 1 /* -> PREOPEN */
88 #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
89 #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
90 #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
91 #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
92 #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
95 * ceph_connection flag bits
97 #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
98 * messages on errors */
99 #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
100 #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
101 #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
102 #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
104 static bool con_flag_valid(unsigned long con_flag)
107 case CON_FLAG_LOSSYTX:
108 case CON_FLAG_KEEPALIVE_PENDING:
109 case CON_FLAG_WRITE_PENDING:
110 case CON_FLAG_SOCK_CLOSED:
111 case CON_FLAG_BACKOFF:
118 static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
120 BUG_ON(!con_flag_valid(con_flag));
122 clear_bit(con_flag, &con->flags);
125 static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
127 BUG_ON(!con_flag_valid(con_flag));
129 set_bit(con_flag, &con->flags);
132 static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
134 BUG_ON(!con_flag_valid(con_flag));
136 return test_bit(con_flag, &con->flags);
139 static bool con_flag_test_and_clear(struct ceph_connection *con,
140 unsigned long con_flag)
142 BUG_ON(!con_flag_valid(con_flag));
144 return test_and_clear_bit(con_flag, &con->flags);
147 static bool con_flag_test_and_set(struct ceph_connection *con,
148 unsigned long con_flag)
150 BUG_ON(!con_flag_valid(con_flag));
152 return test_and_set_bit(con_flag, &con->flags);
155 /* static tag bytes (protocol control messages) */
156 static char tag_msg = CEPH_MSGR_TAG_MSG;
157 static char tag_ack = CEPH_MSGR_TAG_ACK;
158 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
160 #ifdef CONFIG_LOCKDEP
161 static struct lock_class_key socket_class;
165 * When skipping (ignoring) a block of input we read it into a "skip
166 * buffer," which is this many bytes in size.
168 #define SKIP_BUF_SIZE 1024
170 static void queue_con(struct ceph_connection *con);
171 static void con_work(struct work_struct *);
172 static void con_fault(struct ceph_connection *con);
175 * Nicely render a sockaddr as a string. An array of formatted
176 * strings is used, to approximate reentrancy.
178 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
179 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
180 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
181 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
183 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
184 static atomic_t addr_str_seq = ATOMIC_INIT(0);
186 static struct page *zero_page; /* used in certain error cases */
188 const char *ceph_pr_addr(const struct sockaddr_storage *ss)
192 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
193 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
195 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
198 switch (ss->ss_family) {
200 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
201 ntohs(in4->sin_port));
205 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
206 ntohs(in6->sin6_port));
210 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
216 EXPORT_SYMBOL(ceph_pr_addr);
218 static void encode_my_addr(struct ceph_messenger *msgr)
220 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
221 ceph_encode_addr(&msgr->my_enc_addr);
225 * work queue for all reading and writing to/from the socket.
227 static struct workqueue_struct *ceph_msgr_wq;
229 static void _ceph_msgr_exit(void)
232 destroy_workqueue(ceph_msgr_wq);
236 BUG_ON(zero_page == NULL);
238 page_cache_release(zero_page);
242 int ceph_msgr_init(void)
244 BUG_ON(zero_page != NULL);
245 zero_page = ZERO_PAGE(0);
246 page_cache_get(zero_page);
248 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0);
252 pr_err("msgr_init failed to create workqueue\n");
257 EXPORT_SYMBOL(ceph_msgr_init);
259 void ceph_msgr_exit(void)
261 BUG_ON(ceph_msgr_wq == NULL);
265 EXPORT_SYMBOL(ceph_msgr_exit);
267 void ceph_msgr_flush(void)
269 flush_workqueue(ceph_msgr_wq);
271 EXPORT_SYMBOL(ceph_msgr_flush);
273 /* Connection socket state transition functions */
275 static void con_sock_state_init(struct ceph_connection *con)
279 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
280 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
281 printk("%s: unexpected old state %d\n", __func__, old_state);
282 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
283 CON_SOCK_STATE_CLOSED);
286 static void con_sock_state_connecting(struct ceph_connection *con)
290 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
291 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
292 printk("%s: unexpected old state %d\n", __func__, old_state);
293 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
294 CON_SOCK_STATE_CONNECTING);
297 static void con_sock_state_connected(struct ceph_connection *con)
301 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
302 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
303 printk("%s: unexpected old state %d\n", __func__, old_state);
304 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
305 CON_SOCK_STATE_CONNECTED);
308 static void con_sock_state_closing(struct ceph_connection *con)
312 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
313 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
314 old_state != CON_SOCK_STATE_CONNECTED &&
315 old_state != CON_SOCK_STATE_CLOSING))
316 printk("%s: unexpected old state %d\n", __func__, old_state);
317 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
318 CON_SOCK_STATE_CLOSING);
321 static void con_sock_state_closed(struct ceph_connection *con)
325 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
326 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
327 old_state != CON_SOCK_STATE_CLOSING &&
328 old_state != CON_SOCK_STATE_CONNECTING &&
329 old_state != CON_SOCK_STATE_CLOSED))
330 printk("%s: unexpected old state %d\n", __func__, old_state);
331 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
332 CON_SOCK_STATE_CLOSED);
336 * socket callback functions
339 /* data available on socket, or listen socket received a connect */
340 static void ceph_sock_data_ready(struct sock *sk, int count_unused)
342 struct ceph_connection *con = sk->sk_user_data;
343 if (atomic_read(&con->msgr->stopping)) {
347 if (sk->sk_state != TCP_CLOSE_WAIT) {
348 dout("%s on %p state = %lu, queueing work\n", __func__,
354 /* socket has buffer space for writing */
355 static void ceph_sock_write_space(struct sock *sk)
357 struct ceph_connection *con = sk->sk_user_data;
359 /* only queue to workqueue if there is data we want to write,
360 * and there is sufficient space in the socket buffer to accept
361 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
362 * doesn't get called again until try_write() fills the socket
363 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
364 * and net/core/stream.c:sk_stream_write_space().
366 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
367 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
368 dout("%s %p queueing write work\n", __func__, con);
369 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
373 dout("%s %p nothing to write\n", __func__, con);
377 /* socket's state has changed */
378 static void ceph_sock_state_change(struct sock *sk)
380 struct ceph_connection *con = sk->sk_user_data;
382 dout("%s %p state = %lu sk_state = %u\n", __func__,
383 con, con->state, sk->sk_state);
385 switch (sk->sk_state) {
387 dout("%s TCP_CLOSE\n", __func__);
389 dout("%s TCP_CLOSE_WAIT\n", __func__);
390 con_sock_state_closing(con);
391 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
394 case TCP_ESTABLISHED:
395 dout("%s TCP_ESTABLISHED\n", __func__);
396 con_sock_state_connected(con);
399 default: /* Everything else is uninteresting */
405 * set up socket callbacks
407 static void set_sock_callbacks(struct socket *sock,
408 struct ceph_connection *con)
410 struct sock *sk = sock->sk;
411 sk->sk_user_data = con;
412 sk->sk_data_ready = ceph_sock_data_ready;
413 sk->sk_write_space = ceph_sock_write_space;
414 sk->sk_state_change = ceph_sock_state_change;
423 * initiate connection to a remote socket.
425 static int ceph_tcp_connect(struct ceph_connection *con)
427 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
432 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
436 sock->sk->sk_allocation = GFP_NOFS;
438 #ifdef CONFIG_LOCKDEP
439 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
442 set_sock_callbacks(sock, con);
444 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
446 con_sock_state_connecting(con);
447 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
449 if (ret == -EINPROGRESS) {
450 dout("connect %s EINPROGRESS sk_state = %u\n",
451 ceph_pr_addr(&con->peer_addr.in_addr),
453 } else if (ret < 0) {
454 pr_err("connect %s error %d\n",
455 ceph_pr_addr(&con->peer_addr.in_addr), ret);
457 con->error_msg = "connect error";
465 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
467 struct kvec iov = {buf, len};
468 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
471 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
477 static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
478 int page_offset, size_t length)
483 BUG_ON(page_offset + length > PAGE_SIZE);
487 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
494 * write something. @more is true if caller will be sending more data
497 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
498 size_t kvlen, size_t len, int more)
500 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
504 msg.msg_flags |= MSG_MORE;
506 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
508 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
514 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
515 int offset, size_t size, bool more)
517 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
520 ret = kernel_sendpage(sock, page, offset, size, flags);
529 * Shutdown/close the socket for the given connection.
531 static int con_close_socket(struct ceph_connection *con)
535 dout("con_close_socket on %p sock %p\n", con, con->sock);
537 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
538 sock_release(con->sock);
543 * Forcibly clear the SOCK_CLOSED flag. It gets set
544 * independent of the connection mutex, and we could have
545 * received a socket close event before we had the chance to
546 * shut the socket down.
548 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
550 con_sock_state_closed(con);
555 * Reset a connection. Discard all incoming and outgoing messages
556 * and clear *_seq state.
558 static void ceph_msg_remove(struct ceph_msg *msg)
560 list_del_init(&msg->list_head);
561 BUG_ON(msg->con == NULL);
562 msg->con->ops->put(msg->con);
567 static void ceph_msg_remove_list(struct list_head *head)
569 while (!list_empty(head)) {
570 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
572 ceph_msg_remove(msg);
576 static void reset_connection(struct ceph_connection *con)
578 /* reset connection, out_queue, msg_ and connect_seq */
579 /* discard existing out_queue and msg_seq */
580 dout("reset_connection %p\n", con);
581 ceph_msg_remove_list(&con->out_queue);
582 ceph_msg_remove_list(&con->out_sent);
585 BUG_ON(con->in_msg->con != con);
586 con->in_msg->con = NULL;
587 ceph_msg_put(con->in_msg);
592 con->connect_seq = 0;
595 ceph_msg_put(con->out_msg);
599 con->in_seq_acked = 0;
603 * mark a peer down. drop any open connections.
605 void ceph_con_close(struct ceph_connection *con)
607 mutex_lock(&con->mutex);
608 dout("con_close %p peer %s\n", con,
609 ceph_pr_addr(&con->peer_addr.in_addr));
610 con->state = CON_STATE_CLOSED;
612 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
613 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
614 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
615 con_flag_clear(con, CON_FLAG_BACKOFF);
617 reset_connection(con);
618 con->peer_global_seq = 0;
619 cancel_delayed_work(&con->work);
620 con_close_socket(con);
621 mutex_unlock(&con->mutex);
623 EXPORT_SYMBOL(ceph_con_close);
626 * Reopen a closed connection, with a new peer address.
628 void ceph_con_open(struct ceph_connection *con,
629 __u8 entity_type, __u64 entity_num,
630 struct ceph_entity_addr *addr)
632 mutex_lock(&con->mutex);
633 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
635 WARN_ON(con->state != CON_STATE_CLOSED);
636 con->state = CON_STATE_PREOPEN;
638 con->peer_name.type = (__u8) entity_type;
639 con->peer_name.num = cpu_to_le64(entity_num);
641 memcpy(&con->peer_addr, addr, sizeof(*addr));
642 con->delay = 0; /* reset backoff memory */
643 mutex_unlock(&con->mutex);
646 EXPORT_SYMBOL(ceph_con_open);
649 * return true if this connection ever successfully opened
651 bool ceph_con_opened(struct ceph_connection *con)
653 return con->connect_seq > 0;
657 * initialize a new connection.
659 void ceph_con_init(struct ceph_connection *con, void *private,
660 const struct ceph_connection_operations *ops,
661 struct ceph_messenger *msgr)
663 dout("con_init %p\n", con);
664 memset(con, 0, sizeof(*con));
665 con->private = private;
669 con_sock_state_init(con);
671 mutex_init(&con->mutex);
672 INIT_LIST_HEAD(&con->out_queue);
673 INIT_LIST_HEAD(&con->out_sent);
674 INIT_DELAYED_WORK(&con->work, con_work);
676 con->state = CON_STATE_CLOSED;
678 EXPORT_SYMBOL(ceph_con_init);
682 * We maintain a global counter to order connection attempts. Get
683 * a unique seq greater than @gt.
685 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
689 spin_lock(&msgr->global_seq_lock);
690 if (msgr->global_seq < gt)
691 msgr->global_seq = gt;
692 ret = ++msgr->global_seq;
693 spin_unlock(&msgr->global_seq_lock);
697 static void con_out_kvec_reset(struct ceph_connection *con)
699 con->out_kvec_left = 0;
700 con->out_kvec_bytes = 0;
701 con->out_kvec_cur = &con->out_kvec[0];
704 static void con_out_kvec_add(struct ceph_connection *con,
705 size_t size, void *data)
709 index = con->out_kvec_left;
710 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
712 con->out_kvec[index].iov_len = size;
713 con->out_kvec[index].iov_base = data;
714 con->out_kvec_left++;
715 con->out_kvec_bytes += size;
719 static void init_bio_iter(struct bio *bio, struct bio **bio_iter,
720 unsigned int *bio_seg)
728 *bio_seg = (unsigned int) bio->bi_idx;
731 static void iter_bio_next(struct bio **bio_iter, unsigned int *seg)
733 if (*bio_iter == NULL)
736 BUG_ON(*seg >= (*bio_iter)->bi_vcnt);
739 if (*seg == (*bio_iter)->bi_vcnt)
740 init_bio_iter((*bio_iter)->bi_next, bio_iter, seg);
744 * For a bio data item, a piece is whatever remains of the next
745 * entry in the current bio iovec, or the first entry in the next
748 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data *data)
750 struct ceph_msg_data_cursor *cursor = &data->cursor;
753 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
757 BUG_ON(!bio->bi_vcnt);
758 /* resid = bio->bi_size */
761 cursor->vector_index = 0;
762 cursor->vector_offset = 0;
763 cursor->last_piece = !bio->bi_next && bio->bi_vcnt == 1;
766 static struct page *ceph_msg_data_bio_next(struct ceph_msg_data *data,
770 struct ceph_msg_data_cursor *cursor = &data->cursor;
772 struct bio_vec *bio_vec;
775 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
780 index = cursor->vector_index;
781 BUG_ON(index >= (unsigned int) bio->bi_vcnt);
783 bio_vec = &bio->bi_io_vec[index];
784 BUG_ON(cursor->vector_offset >= bio_vec->bv_len);
785 *page_offset = (size_t) (bio_vec->bv_offset + cursor->vector_offset);
786 BUG_ON(*page_offset >= PAGE_SIZE);
787 *length = (size_t) (bio_vec->bv_len - cursor->vector_offset);
788 BUG_ON(*length > PAGE_SIZE);
790 return bio_vec->bv_page;
793 static bool ceph_msg_data_bio_advance(struct ceph_msg_data *data, size_t bytes)
795 struct ceph_msg_data_cursor *cursor = &data->cursor;
797 struct bio_vec *bio_vec;
800 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
805 index = cursor->vector_index;
806 BUG_ON(index >= (unsigned int) bio->bi_vcnt);
807 bio_vec = &bio->bi_io_vec[index];
808 BUG_ON(cursor->vector_offset + bytes > bio_vec->bv_len);
810 /* Advance the cursor offset */
812 cursor->vector_offset += bytes;
813 if (cursor->vector_offset < bio_vec->bv_len)
814 return false; /* more bytes to process in this segment */
816 /* Move on to the next segment, and possibly the next bio */
818 if (++cursor->vector_index == (unsigned int) bio->bi_vcnt) {
821 cursor->vector_index = 0;
823 cursor->vector_offset = 0;
825 if (!cursor->last_piece && bio && !bio->bi_next)
826 if (cursor->vector_index == (unsigned int) bio->bi_vcnt - 1)
827 cursor->last_piece = true;
834 * For a page array, a piece comes from the first page in the array
835 * that has not already been fully consumed.
837 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data *data)
839 struct ceph_msg_data_cursor *cursor = &data->cursor;
842 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
844 BUG_ON(!data->pages);
845 BUG_ON(!data->length);
847 page_count = calc_pages_for(data->alignment, (u64)data->length);
848 BUG_ON(page_count > (int) USHRT_MAX);
849 cursor->resid = data->length;
850 cursor->page_offset = data->alignment & ~PAGE_MASK;
851 cursor->page_index = 0;
852 cursor->page_count = (unsigned short) page_count;
853 cursor->last_piece = cursor->page_count == 1;
856 static struct page *ceph_msg_data_pages_next(struct ceph_msg_data *data,
860 struct ceph_msg_data_cursor *cursor = &data->cursor;
862 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
864 BUG_ON(cursor->page_index >= cursor->page_count);
865 BUG_ON(cursor->page_offset >= PAGE_SIZE);
866 BUG_ON(!cursor->resid);
868 *page_offset = cursor->page_offset;
869 if (cursor->last_piece) {
870 BUG_ON(*page_offset + cursor->resid > PAGE_SIZE);
871 *length = cursor->resid;
873 *length = PAGE_SIZE - *page_offset;
876 return data->pages[cursor->page_index];
879 static bool ceph_msg_data_pages_advance(struct ceph_msg_data *data,
882 struct ceph_msg_data_cursor *cursor = &data->cursor;
884 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
886 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
887 BUG_ON(bytes > cursor->resid);
889 /* Advance the cursor page offset */
891 cursor->resid -= bytes;
892 cursor->page_offset += bytes;
893 if (!bytes || cursor->page_offset & ~PAGE_MASK)
894 return false; /* more bytes to process in the current page */
896 /* Move on to the next page */
898 BUG_ON(cursor->page_index >= cursor->page_count);
899 cursor->page_offset = 0;
900 cursor->page_index++;
901 cursor->last_piece = cursor->page_index == cursor->page_count - 1;
907 * For a pagelist, a piece is whatever remains to be consumed in the
908 * first page in the list, or the front of the next page.
910 static void ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data *data)
912 struct ceph_msg_data_cursor *cursor = &data->cursor;
913 struct ceph_pagelist *pagelist;
916 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
918 pagelist = data->pagelist;
920 if (!pagelist->length)
921 return; /* pagelist can be assigned but empty */
923 BUG_ON(list_empty(&pagelist->head));
924 page = list_first_entry(&pagelist->head, struct page, lru);
928 cursor->last_piece = pagelist->length <= PAGE_SIZE;
931 static struct page *ceph_msg_data_pagelist_next(struct ceph_msg_data *data,
935 struct ceph_msg_data_cursor *cursor = &data->cursor;
936 struct ceph_pagelist *pagelist;
939 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
941 pagelist = data->pagelist;
944 BUG_ON(!cursor->page);
945 BUG_ON(cursor->offset >= pagelist->length);
947 if (cursor->last_piece) {
948 /* pagelist offset is always 0 */
949 piece_end = pagelist->length & ~PAGE_MASK;
951 piece_end = PAGE_SIZE;
953 piece_end = PAGE_SIZE;
955 *page_offset = cursor->offset & ~PAGE_MASK;
956 *length = piece_end - *page_offset;
958 return data->cursor.page;
961 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data *data,
964 struct ceph_msg_data_cursor *cursor = &data->cursor;
965 struct ceph_pagelist *pagelist;
967 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
969 pagelist = data->pagelist;
971 BUG_ON(!cursor->page);
972 BUG_ON(cursor->offset + bytes > pagelist->length);
973 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
975 /* Advance the cursor offset */
977 cursor->offset += bytes;
978 /* pagelist offset is always 0 */
979 if (!bytes || cursor->offset & ~PAGE_MASK)
980 return false; /* more bytes to process in the current page */
982 /* Move on to the next page */
984 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
985 cursor->page = list_entry_next(cursor->page, lru);
987 /* cursor offset is at page boundary; pagelist offset is always 0 */
988 if (pagelist->length - cursor->offset <= PAGE_SIZE)
989 cursor->last_piece = true;
995 * Message data is handled (sent or received) in pieces, where each
996 * piece resides on a single page. The network layer might not
997 * consume an entire piece at once. A data item's cursor keeps
998 * track of which piece is next to process and how much remains to
999 * be processed in that piece. It also tracks whether the current
1000 * piece is the last one in the data item.
1002 static void ceph_msg_data_cursor_init(struct ceph_msg_data *data)
1004 switch (data->type) {
1005 case CEPH_MSG_DATA_PAGELIST:
1006 ceph_msg_data_pagelist_cursor_init(data);
1008 case CEPH_MSG_DATA_PAGES:
1009 ceph_msg_data_pages_cursor_init(data);
1012 case CEPH_MSG_DATA_BIO:
1013 ceph_msg_data_bio_cursor_init(data);
1015 #endif /* CONFIG_BLOCK */
1016 case CEPH_MSG_DATA_NONE:
1024 * Return the page containing the next piece to process for a given
1025 * data item, and supply the page offset and length of that piece.
1026 * Indicate whether this is the last piece in this data item.
1028 static struct page *ceph_msg_data_next(struct ceph_msg_data *data,
1029 size_t *page_offset,
1035 switch (data->type) {
1036 case CEPH_MSG_DATA_PAGELIST:
1037 page = ceph_msg_data_pagelist_next(data, page_offset, length);
1039 case CEPH_MSG_DATA_PAGES:
1040 page = ceph_msg_data_pages_next(data, page_offset, length);
1043 case CEPH_MSG_DATA_BIO:
1044 page = ceph_msg_data_bio_next(data, page_offset, length);
1046 #endif /* CONFIG_BLOCK */
1047 case CEPH_MSG_DATA_NONE:
1053 BUG_ON(*page_offset + *length > PAGE_SIZE);
1056 *last_piece = data->cursor.last_piece;
1062 * Returns true if the result moves the cursor on to the next piece
1065 static bool ceph_msg_data_advance(struct ceph_msg_data *data, size_t bytes)
1069 switch (data->type) {
1070 case CEPH_MSG_DATA_PAGELIST:
1071 new_piece = ceph_msg_data_pagelist_advance(data, bytes);
1073 case CEPH_MSG_DATA_PAGES:
1074 new_piece = ceph_msg_data_pages_advance(data, bytes);
1077 case CEPH_MSG_DATA_BIO:
1078 new_piece = ceph_msg_data_bio_advance(data, bytes);
1080 #endif /* CONFIG_BLOCK */
1081 case CEPH_MSG_DATA_NONE:
1090 static void prepare_message_data(struct ceph_msg *msg,
1091 struct ceph_msg_pos *msg_pos)
1094 BUG_ON(!msg->hdr.data_len);
1096 /* initialize page iterator */
1098 if (ceph_msg_has_pages(msg))
1099 msg_pos->page_pos = msg->p.alignment;
1101 msg_pos->page_pos = 0;
1103 if (ceph_msg_has_bio(msg))
1104 init_bio_iter(msg->b.bio, &msg->b.bio_iter, &msg->b.bio_seg);
1106 msg_pos->data_pos = 0;
1108 /* Initialize data cursors */
1111 if (ceph_msg_has_bio(msg))
1112 ceph_msg_data_cursor_init(&msg->b);
1113 #endif /* CONFIG_BLOCK */
1114 if (ceph_msg_has_pages(msg))
1115 ceph_msg_data_cursor_init(&msg->p);
1116 if (ceph_msg_has_pagelist(msg))
1117 ceph_msg_data_cursor_init(&msg->l);
1119 msg_pos->did_page_crc = false;
1123 * Prepare footer for currently outgoing message, and finish things
1124 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1126 static void prepare_write_message_footer(struct ceph_connection *con)
1128 struct ceph_msg *m = con->out_msg;
1129 int v = con->out_kvec_left;
1131 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1133 dout("prepare_write_message_footer %p\n", con);
1134 con->out_kvec_is_msg = true;
1135 con->out_kvec[v].iov_base = &m->footer;
1136 con->out_kvec[v].iov_len = sizeof(m->footer);
1137 con->out_kvec_bytes += sizeof(m->footer);
1138 con->out_kvec_left++;
1139 con->out_more = m->more_to_follow;
1140 con->out_msg_done = true;
1144 * Prepare headers for the next outgoing message.
1146 static void prepare_write_message(struct ceph_connection *con)
1151 con_out_kvec_reset(con);
1152 con->out_kvec_is_msg = true;
1153 con->out_msg_done = false;
1155 /* Sneak an ack in there first? If we can get it into the same
1156 * TCP packet that's a good thing. */
1157 if (con->in_seq > con->in_seq_acked) {
1158 con->in_seq_acked = con->in_seq;
1159 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1160 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1161 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1162 &con->out_temp_ack);
1165 BUG_ON(list_empty(&con->out_queue));
1166 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1168 BUG_ON(m->con != con);
1170 /* put message on sent list */
1172 list_move_tail(&m->list_head, &con->out_sent);
1175 * only assign outgoing seq # if we haven't sent this message
1176 * yet. if it is requeued, resend with it's original seq.
1178 if (m->needs_out_seq) {
1179 m->hdr.seq = cpu_to_le64(++con->out_seq);
1180 m->needs_out_seq = false;
1183 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d (%zd)\n",
1184 m, con->out_seq, le16_to_cpu(m->hdr.type),
1185 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1186 le32_to_cpu(m->hdr.data_len), m->p.length);
1187 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1189 /* tag + hdr + front + middle */
1190 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1191 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1192 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1195 con_out_kvec_add(con, m->middle->vec.iov_len,
1196 m->middle->vec.iov_base);
1198 /* fill in crc (except data pages), footer */
1199 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1200 con->out_msg->hdr.crc = cpu_to_le32(crc);
1201 con->out_msg->footer.flags = 0;
1203 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1204 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1206 crc = crc32c(0, m->middle->vec.iov_base,
1207 m->middle->vec.iov_len);
1208 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1210 con->out_msg->footer.middle_crc = 0;
1211 dout("%s front_crc %u middle_crc %u\n", __func__,
1212 le32_to_cpu(con->out_msg->footer.front_crc),
1213 le32_to_cpu(con->out_msg->footer.middle_crc));
1215 /* is there a data payload? */
1216 con->out_msg->footer.data_crc = 0;
1217 if (m->hdr.data_len) {
1218 prepare_message_data(con->out_msg, &con->out_msg_pos);
1219 con->out_more = 1; /* data + footer will follow */
1221 /* no, queue up footer too and be done */
1222 prepare_write_message_footer(con);
1225 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1231 static void prepare_write_ack(struct ceph_connection *con)
1233 dout("prepare_write_ack %p %llu -> %llu\n", con,
1234 con->in_seq_acked, con->in_seq);
1235 con->in_seq_acked = con->in_seq;
1237 con_out_kvec_reset(con);
1239 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1241 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1242 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1243 &con->out_temp_ack);
1245 con->out_more = 1; /* more will follow.. eventually.. */
1246 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1250 * Prepare to share the seq during handshake
1252 static void prepare_write_seq(struct ceph_connection *con)
1254 dout("prepare_write_seq %p %llu -> %llu\n", con,
1255 con->in_seq_acked, con->in_seq);
1256 con->in_seq_acked = con->in_seq;
1258 con_out_kvec_reset(con);
1260 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1261 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1262 &con->out_temp_ack);
1264 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1268 * Prepare to write keepalive byte.
1270 static void prepare_write_keepalive(struct ceph_connection *con)
1272 dout("prepare_write_keepalive %p\n", con);
1273 con_out_kvec_reset(con);
1274 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
1275 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1279 * Connection negotiation.
1282 static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1285 struct ceph_auth_handshake *auth;
1287 if (!con->ops->get_authorizer) {
1288 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1289 con->out_connect.authorizer_len = 0;
1293 /* Can't hold the mutex while getting authorizer */
1294 mutex_unlock(&con->mutex);
1295 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
1296 mutex_lock(&con->mutex);
1300 if (con->state != CON_STATE_NEGOTIATING)
1301 return ERR_PTR(-EAGAIN);
1303 con->auth_reply_buf = auth->authorizer_reply_buf;
1304 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
1309 * We connected to a peer and are saying hello.
1311 static void prepare_write_banner(struct ceph_connection *con)
1313 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1314 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1315 &con->msgr->my_enc_addr);
1318 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1321 static int prepare_write_connect(struct ceph_connection *con)
1323 unsigned int global_seq = get_global_seq(con->msgr, 0);
1326 struct ceph_auth_handshake *auth;
1328 switch (con->peer_name.type) {
1329 case CEPH_ENTITY_TYPE_MON:
1330 proto = CEPH_MONC_PROTOCOL;
1332 case CEPH_ENTITY_TYPE_OSD:
1333 proto = CEPH_OSDC_PROTOCOL;
1335 case CEPH_ENTITY_TYPE_MDS:
1336 proto = CEPH_MDSC_PROTOCOL;
1342 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1343 con->connect_seq, global_seq, proto);
1345 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
1346 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1347 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1348 con->out_connect.global_seq = cpu_to_le32(global_seq);
1349 con->out_connect.protocol_version = cpu_to_le32(proto);
1350 con->out_connect.flags = 0;
1352 auth_proto = CEPH_AUTH_UNKNOWN;
1353 auth = get_connect_authorizer(con, &auth_proto);
1355 return PTR_ERR(auth);
1357 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1358 con->out_connect.authorizer_len = auth ?
1359 cpu_to_le32(auth->authorizer_buf_len) : 0;
1361 con_out_kvec_add(con, sizeof (con->out_connect),
1363 if (auth && auth->authorizer_buf_len)
1364 con_out_kvec_add(con, auth->authorizer_buf_len,
1365 auth->authorizer_buf);
1368 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1374 * write as much of pending kvecs to the socket as we can.
1376 * 0 -> socket full, but more to do
1379 static int write_partial_kvec(struct ceph_connection *con)
1383 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1384 while (con->out_kvec_bytes > 0) {
1385 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1386 con->out_kvec_left, con->out_kvec_bytes,
1390 con->out_kvec_bytes -= ret;
1391 if (con->out_kvec_bytes == 0)
1394 /* account for full iov entries consumed */
1395 while (ret >= con->out_kvec_cur->iov_len) {
1396 BUG_ON(!con->out_kvec_left);
1397 ret -= con->out_kvec_cur->iov_len;
1398 con->out_kvec_cur++;
1399 con->out_kvec_left--;
1401 /* and for a partially-consumed entry */
1403 con->out_kvec_cur->iov_len -= ret;
1404 con->out_kvec_cur->iov_base += ret;
1407 con->out_kvec_left = 0;
1408 con->out_kvec_is_msg = false;
1411 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1412 con->out_kvec_bytes, con->out_kvec_left, ret);
1413 return ret; /* done! */
1416 static void out_msg_pos_next(struct ceph_connection *con, struct page *page,
1417 size_t len, size_t sent)
1419 struct ceph_msg *msg = con->out_msg;
1420 struct ceph_msg_pos *msg_pos = &con->out_msg_pos;
1421 bool need_crc = false;
1426 msg_pos->data_pos += sent;
1427 msg_pos->page_pos += sent;
1428 if (ceph_msg_has_pages(msg))
1429 need_crc = ceph_msg_data_advance(&msg->p, sent);
1430 else if (ceph_msg_has_pagelist(msg))
1431 need_crc = ceph_msg_data_advance(&msg->l, sent);
1433 else if (ceph_msg_has_bio(msg))
1434 need_crc = ceph_msg_data_advance(&msg->b, sent);
1435 #endif /* CONFIG_BLOCK */
1436 BUG_ON(need_crc && sent != len);
1441 BUG_ON(sent != len);
1442 msg_pos->page_pos = 0;
1444 msg_pos->did_page_crc = false;
1447 static void in_msg_pos_next(struct ceph_connection *con, size_t len,
1450 struct ceph_msg *msg = con->in_msg;
1451 struct ceph_msg_pos *msg_pos = &con->in_msg_pos;
1456 msg_pos->data_pos += received;
1457 msg_pos->page_pos += received;
1461 BUG_ON(received != len);
1462 msg_pos->page_pos = 0;
1466 iter_bio_next(&msg->b.bio_iter, &msg->b.bio_seg);
1467 #endif /* CONFIG_BLOCK */
1470 static u32 ceph_crc32c_page(u32 crc, struct page *page,
1471 unsigned int page_offset,
1472 unsigned int length)
1477 BUG_ON(kaddr == NULL);
1478 crc = crc32c(crc, kaddr + page_offset, length);
1484 * Write as much message data payload as we can. If we finish, queue
1486 * 1 -> done, footer is now queued in out_kvec[].
1487 * 0 -> socket full, but more to do
1490 static int write_partial_message_data(struct ceph_connection *con)
1492 struct ceph_msg *msg = con->out_msg;
1493 struct ceph_msg_pos *msg_pos = &con->out_msg_pos;
1494 unsigned int data_len = le32_to_cpu(msg->hdr.data_len);
1495 bool do_datacrc = !con->msgr->nocrc;
1498 dout("%s %p msg %p page %d offset %d\n", __func__,
1499 con, msg, msg_pos->page, msg_pos->page_pos);
1502 * Iterate through each page that contains data to be
1503 * written, and send as much as possible for each.
1505 * If we are calculating the data crc (the default), we will
1506 * need to map the page. If we have no pages, they have
1507 * been revoked, so use the zero page.
1509 while (data_len > msg_pos->data_pos) {
1515 if (ceph_msg_has_pages(msg)) {
1516 page = ceph_msg_data_next(&msg->p, &page_offset,
1517 &length, &last_piece);
1518 } else if (ceph_msg_has_pagelist(msg)) {
1519 page = ceph_msg_data_next(&msg->l, &page_offset,
1520 &length, &last_piece);
1522 } else if (ceph_msg_has_bio(msg)) {
1523 page = ceph_msg_data_next(&msg->b, &page_offset,
1524 &length, &last_piece);
1527 size_t resid = data_len - msg_pos->data_pos;
1530 page_offset = msg_pos->page_pos;
1531 length = PAGE_SIZE - page_offset;
1532 length = min(resid, length);
1533 last_piece = length == resid;
1535 if (do_datacrc && !msg_pos->did_page_crc) {
1536 u32 crc = le32_to_cpu(msg->footer.data_crc);
1538 crc = ceph_crc32c_page(crc, page, page_offset, length);
1539 msg->footer.data_crc = cpu_to_le32(crc);
1540 msg_pos->did_page_crc = true;
1542 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
1543 length, last_piece);
1547 out_msg_pos_next(con, page, length, (size_t) ret);
1550 dout("%s %p msg %p done\n", __func__, con, msg);
1552 /* prepare and queue up footer, too */
1554 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1555 con_out_kvec_reset(con);
1556 prepare_write_message_footer(con);
1565 static int write_partial_skip(struct ceph_connection *con)
1569 while (con->out_skip > 0) {
1570 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
1572 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
1575 con->out_skip -= ret;
1583 * Prepare to read connection handshake, or an ack.
1585 static void prepare_read_banner(struct ceph_connection *con)
1587 dout("prepare_read_banner %p\n", con);
1588 con->in_base_pos = 0;
1591 static void prepare_read_connect(struct ceph_connection *con)
1593 dout("prepare_read_connect %p\n", con);
1594 con->in_base_pos = 0;
1597 static void prepare_read_ack(struct ceph_connection *con)
1599 dout("prepare_read_ack %p\n", con);
1600 con->in_base_pos = 0;
1603 static void prepare_read_seq(struct ceph_connection *con)
1605 dout("prepare_read_seq %p\n", con);
1606 con->in_base_pos = 0;
1607 con->in_tag = CEPH_MSGR_TAG_SEQ;
1610 static void prepare_read_tag(struct ceph_connection *con)
1612 dout("prepare_read_tag %p\n", con);
1613 con->in_base_pos = 0;
1614 con->in_tag = CEPH_MSGR_TAG_READY;
1618 * Prepare to read a message.
1620 static int prepare_read_message(struct ceph_connection *con)
1622 dout("prepare_read_message %p\n", con);
1623 BUG_ON(con->in_msg != NULL);
1624 con->in_base_pos = 0;
1625 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1630 static int read_partial(struct ceph_connection *con,
1631 int end, int size, void *object)
1633 while (con->in_base_pos < end) {
1634 int left = end - con->in_base_pos;
1635 int have = size - left;
1636 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1639 con->in_base_pos += ret;
1646 * Read all or part of the connect-side handshake on a new connection
1648 static int read_partial_banner(struct ceph_connection *con)
1654 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1657 size = strlen(CEPH_BANNER);
1659 ret = read_partial(con, end, size, con->in_banner);
1663 size = sizeof (con->actual_peer_addr);
1665 ret = read_partial(con, end, size, &con->actual_peer_addr);
1669 size = sizeof (con->peer_addr_for_me);
1671 ret = read_partial(con, end, size, &con->peer_addr_for_me);
1679 static int read_partial_connect(struct ceph_connection *con)
1685 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1687 size = sizeof (con->in_reply);
1689 ret = read_partial(con, end, size, &con->in_reply);
1693 size = le32_to_cpu(con->in_reply.authorizer_len);
1695 ret = read_partial(con, end, size, con->auth_reply_buf);
1699 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1700 con, (int)con->in_reply.tag,
1701 le32_to_cpu(con->in_reply.connect_seq),
1702 le32_to_cpu(con->in_reply.global_seq));
1709 * Verify the hello banner looks okay.
1711 static int verify_hello(struct ceph_connection *con)
1713 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1714 pr_err("connect to %s got bad banner\n",
1715 ceph_pr_addr(&con->peer_addr.in_addr));
1716 con->error_msg = "protocol error, bad banner";
1722 static bool addr_is_blank(struct sockaddr_storage *ss)
1724 switch (ss->ss_family) {
1726 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1729 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1730 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1731 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1732 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1737 static int addr_port(struct sockaddr_storage *ss)
1739 switch (ss->ss_family) {
1741 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1743 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1748 static void addr_set_port(struct sockaddr_storage *ss, int p)
1750 switch (ss->ss_family) {
1752 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1755 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1761 * Unlike other *_pton function semantics, zero indicates success.
1763 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1764 char delim, const char **ipend)
1766 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1767 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1769 memset(ss, 0, sizeof(*ss));
1771 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1772 ss->ss_family = AF_INET;
1776 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1777 ss->ss_family = AF_INET6;
1785 * Extract hostname string and resolve using kernel DNS facility.
1787 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1788 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1789 struct sockaddr_storage *ss, char delim, const char **ipend)
1791 const char *end, *delim_p;
1792 char *colon_p, *ip_addr = NULL;
1796 * The end of the hostname occurs immediately preceding the delimiter or
1797 * the port marker (':') where the delimiter takes precedence.
1799 delim_p = memchr(name, delim, namelen);
1800 colon_p = memchr(name, ':', namelen);
1802 if (delim_p && colon_p)
1803 end = delim_p < colon_p ? delim_p : colon_p;
1804 else if (!delim_p && colon_p)
1808 if (!end) /* case: hostname:/ */
1809 end = name + namelen;
1815 /* do dns_resolve upcall */
1816 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1818 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1826 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1827 ret, ret ? "failed" : ceph_pr_addr(ss));
1832 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1833 struct sockaddr_storage *ss, char delim, const char **ipend)
1840 * Parse a server name (IP or hostname). If a valid IP address is not found
1841 * then try to extract a hostname to resolve using userspace DNS upcall.
1843 static int ceph_parse_server_name(const char *name, size_t namelen,
1844 struct sockaddr_storage *ss, char delim, const char **ipend)
1848 ret = ceph_pton(name, namelen, ss, delim, ipend);
1850 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1856 * Parse an ip[:port] list into an addr array. Use the default
1857 * monitor port if a port isn't specified.
1859 int ceph_parse_ips(const char *c, const char *end,
1860 struct ceph_entity_addr *addr,
1861 int max_count, int *count)
1863 int i, ret = -EINVAL;
1866 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1867 for (i = 0; i < max_count; i++) {
1869 struct sockaddr_storage *ss = &addr[i].in_addr;
1878 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1887 dout("missing matching ']'\n");
1894 if (p < end && *p == ':') {
1897 while (p < end && *p >= '0' && *p <= '9') {
1898 port = (port * 10) + (*p - '0');
1901 if (port > 65535 || port == 0)
1904 port = CEPH_MON_PORT;
1907 addr_set_port(ss, port);
1909 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1926 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1929 EXPORT_SYMBOL(ceph_parse_ips);
1931 static int process_banner(struct ceph_connection *con)
1933 dout("process_banner on %p\n", con);
1935 if (verify_hello(con) < 0)
1938 ceph_decode_addr(&con->actual_peer_addr);
1939 ceph_decode_addr(&con->peer_addr_for_me);
1942 * Make sure the other end is who we wanted. note that the other
1943 * end may not yet know their ip address, so if it's 0.0.0.0, give
1944 * them the benefit of the doubt.
1946 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1947 sizeof(con->peer_addr)) != 0 &&
1948 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1949 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1950 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1951 ceph_pr_addr(&con->peer_addr.in_addr),
1952 (int)le32_to_cpu(con->peer_addr.nonce),
1953 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1954 (int)le32_to_cpu(con->actual_peer_addr.nonce));
1955 con->error_msg = "wrong peer at address";
1960 * did we learn our address?
1962 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1963 int port = addr_port(&con->msgr->inst.addr.in_addr);
1965 memcpy(&con->msgr->inst.addr.in_addr,
1966 &con->peer_addr_for_me.in_addr,
1967 sizeof(con->peer_addr_for_me.in_addr));
1968 addr_set_port(&con->msgr->inst.addr.in_addr, port);
1969 encode_my_addr(con->msgr);
1970 dout("process_banner learned my addr is %s\n",
1971 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
1977 static int process_connect(struct ceph_connection *con)
1979 u64 sup_feat = con->msgr->supported_features;
1980 u64 req_feat = con->msgr->required_features;
1981 u64 server_feat = le64_to_cpu(con->in_reply.features);
1984 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1986 switch (con->in_reply.tag) {
1987 case CEPH_MSGR_TAG_FEATURES:
1988 pr_err("%s%lld %s feature set mismatch,"
1989 " my %llx < server's %llx, missing %llx\n",
1990 ENTITY_NAME(con->peer_name),
1991 ceph_pr_addr(&con->peer_addr.in_addr),
1992 sup_feat, server_feat, server_feat & ~sup_feat);
1993 con->error_msg = "missing required protocol features";
1994 reset_connection(con);
1997 case CEPH_MSGR_TAG_BADPROTOVER:
1998 pr_err("%s%lld %s protocol version mismatch,"
1999 " my %d != server's %d\n",
2000 ENTITY_NAME(con->peer_name),
2001 ceph_pr_addr(&con->peer_addr.in_addr),
2002 le32_to_cpu(con->out_connect.protocol_version),
2003 le32_to_cpu(con->in_reply.protocol_version));
2004 con->error_msg = "protocol version mismatch";
2005 reset_connection(con);
2008 case CEPH_MSGR_TAG_BADAUTHORIZER:
2010 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2012 if (con->auth_retry == 2) {
2013 con->error_msg = "connect authorization failure";
2016 con->auth_retry = 1;
2017 con_out_kvec_reset(con);
2018 ret = prepare_write_connect(con);
2021 prepare_read_connect(con);
2024 case CEPH_MSGR_TAG_RESETSESSION:
2026 * If we connected with a large connect_seq but the peer
2027 * has no record of a session with us (no connection, or
2028 * connect_seq == 0), they will send RESETSESION to indicate
2029 * that they must have reset their session, and may have
2032 dout("process_connect got RESET peer seq %u\n",
2033 le32_to_cpu(con->in_reply.connect_seq));
2034 pr_err("%s%lld %s connection reset\n",
2035 ENTITY_NAME(con->peer_name),
2036 ceph_pr_addr(&con->peer_addr.in_addr));
2037 reset_connection(con);
2038 con_out_kvec_reset(con);
2039 ret = prepare_write_connect(con);
2042 prepare_read_connect(con);
2044 /* Tell ceph about it. */
2045 mutex_unlock(&con->mutex);
2046 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2047 if (con->ops->peer_reset)
2048 con->ops->peer_reset(con);
2049 mutex_lock(&con->mutex);
2050 if (con->state != CON_STATE_NEGOTIATING)
2054 case CEPH_MSGR_TAG_RETRY_SESSION:
2056 * If we sent a smaller connect_seq than the peer has, try
2057 * again with a larger value.
2059 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2060 le32_to_cpu(con->out_connect.connect_seq),
2061 le32_to_cpu(con->in_reply.connect_seq));
2062 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2063 con_out_kvec_reset(con);
2064 ret = prepare_write_connect(con);
2067 prepare_read_connect(con);
2070 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2072 * If we sent a smaller global_seq than the peer has, try
2073 * again with a larger value.
2075 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2076 con->peer_global_seq,
2077 le32_to_cpu(con->in_reply.global_seq));
2078 get_global_seq(con->msgr,
2079 le32_to_cpu(con->in_reply.global_seq));
2080 con_out_kvec_reset(con);
2081 ret = prepare_write_connect(con);
2084 prepare_read_connect(con);
2087 case CEPH_MSGR_TAG_SEQ:
2088 case CEPH_MSGR_TAG_READY:
2089 if (req_feat & ~server_feat) {
2090 pr_err("%s%lld %s protocol feature mismatch,"
2091 " my required %llx > server's %llx, need %llx\n",
2092 ENTITY_NAME(con->peer_name),
2093 ceph_pr_addr(&con->peer_addr.in_addr),
2094 req_feat, server_feat, req_feat & ~server_feat);
2095 con->error_msg = "missing required protocol features";
2096 reset_connection(con);
2100 WARN_ON(con->state != CON_STATE_NEGOTIATING);
2101 con->state = CON_STATE_OPEN;
2103 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2105 con->peer_features = server_feat;
2106 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2107 con->peer_global_seq,
2108 le32_to_cpu(con->in_reply.connect_seq),
2110 WARN_ON(con->connect_seq !=
2111 le32_to_cpu(con->in_reply.connect_seq));
2113 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2114 con_flag_set(con, CON_FLAG_LOSSYTX);
2116 con->delay = 0; /* reset backoff memory */
2118 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2119 prepare_write_seq(con);
2120 prepare_read_seq(con);
2122 prepare_read_tag(con);
2126 case CEPH_MSGR_TAG_WAIT:
2128 * If there is a connection race (we are opening
2129 * connections to each other), one of us may just have
2130 * to WAIT. This shouldn't happen if we are the
2133 pr_err("process_connect got WAIT as client\n");
2134 con->error_msg = "protocol error, got WAIT as client";
2138 pr_err("connect protocol error, will retry\n");
2139 con->error_msg = "protocol error, garbage tag during connect";
2147 * read (part of) an ack
2149 static int read_partial_ack(struct ceph_connection *con)
2151 int size = sizeof (con->in_temp_ack);
2154 return read_partial(con, end, size, &con->in_temp_ack);
2158 * We can finally discard anything that's been acked.
2160 static void process_ack(struct ceph_connection *con)
2163 u64 ack = le64_to_cpu(con->in_temp_ack);
2166 while (!list_empty(&con->out_sent)) {
2167 m = list_first_entry(&con->out_sent, struct ceph_msg,
2169 seq = le64_to_cpu(m->hdr.seq);
2172 dout("got ack for seq %llu type %d at %p\n", seq,
2173 le16_to_cpu(m->hdr.type), m);
2174 m->ack_stamp = jiffies;
2177 prepare_read_tag(con);
2181 static int read_partial_message_section(struct ceph_connection *con,
2182 struct kvec *section,
2183 unsigned int sec_len, u32 *crc)
2189 while (section->iov_len < sec_len) {
2190 BUG_ON(section->iov_base == NULL);
2191 left = sec_len - section->iov_len;
2192 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2193 section->iov_len, left);
2196 section->iov_len += ret;
2198 if (section->iov_len == sec_len)
2199 *crc = crc32c(0, section->iov_base, section->iov_len);
2204 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2206 static int read_partial_message_pages(struct ceph_connection *con,
2207 struct page **pages,
2208 unsigned int data_len, bool do_datacrc)
2210 struct ceph_msg_pos *msg_pos = &con->in_msg_pos;
2218 BUG_ON(pages == NULL);
2219 page = pages[msg_pos->page];
2220 page_offset = msg_pos->page_pos;
2221 BUG_ON(msg_pos->data_pos >= data_len);
2222 left = data_len - msg_pos->data_pos;
2223 BUG_ON(page_offset >= PAGE_SIZE);
2224 length = min_t(unsigned int, PAGE_SIZE - page_offset, left);
2226 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2231 con->in_data_crc = ceph_crc32c_page(con->in_data_crc, page,
2234 in_msg_pos_next(con, length, ret);
2240 static int read_partial_message_bio(struct ceph_connection *con,
2241 unsigned int data_len, bool do_datacrc)
2243 struct ceph_msg *msg = con->in_msg;
2244 struct ceph_msg_pos *msg_pos = &con->in_msg_pos;
2253 BUG_ON(!msg->b.bio_iter);
2254 bv = bio_iovec_idx(msg->b.bio_iter, msg->b.bio_seg);
2256 page_offset = bv->bv_offset + msg_pos->page_pos;
2257 BUG_ON(msg_pos->data_pos >= data_len);
2258 left = data_len - msg_pos->data_pos;
2259 BUG_ON(msg_pos->page_pos >= bv->bv_len);
2260 length = min_t(unsigned int, bv->bv_len - msg_pos->page_pos, left);
2262 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2267 con->in_data_crc = ceph_crc32c_page(con->in_data_crc, page,
2270 in_msg_pos_next(con, length, ret);
2276 static int read_partial_msg_data(struct ceph_connection *con)
2278 struct ceph_msg *msg = con->in_msg;
2279 struct ceph_msg_pos *msg_pos = &con->in_msg_pos;
2280 const bool do_datacrc = !con->msgr->nocrc;
2281 unsigned int data_len;
2286 data_len = le32_to_cpu(con->in_hdr.data_len);
2287 while (msg_pos->data_pos < data_len) {
2288 if (ceph_msg_has_pages(msg)) {
2289 ret = read_partial_message_pages(con, msg->p.pages,
2290 data_len, do_datacrc);
2294 } else if (ceph_msg_has_bio(msg)) {
2295 ret = read_partial_message_bio(con,
2296 data_len, do_datacrc);
2305 return 1; /* must return > 0 to indicate success */
2309 * read (part of) a message.
2311 static int read_partial_message(struct ceph_connection *con)
2313 struct ceph_msg *m = con->in_msg;
2317 unsigned int front_len, middle_len, data_len;
2318 bool do_datacrc = !con->msgr->nocrc;
2322 dout("read_partial_message con %p msg %p\n", con, m);
2325 size = sizeof (con->in_hdr);
2327 ret = read_partial(con, end, size, &con->in_hdr);
2331 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2332 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2333 pr_err("read_partial_message bad hdr "
2334 " crc %u != expected %u\n",
2335 crc, con->in_hdr.crc);
2339 front_len = le32_to_cpu(con->in_hdr.front_len);
2340 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2342 middle_len = le32_to_cpu(con->in_hdr.middle_len);
2343 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2345 data_len = le32_to_cpu(con->in_hdr.data_len);
2346 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2350 seq = le64_to_cpu(con->in_hdr.seq);
2351 if ((s64)seq - (s64)con->in_seq < 1) {
2352 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2353 ENTITY_NAME(con->peer_name),
2354 ceph_pr_addr(&con->peer_addr.in_addr),
2355 seq, con->in_seq + 1);
2356 con->in_base_pos = -front_len - middle_len - data_len -
2358 con->in_tag = CEPH_MSGR_TAG_READY;
2360 } else if ((s64)seq - (s64)con->in_seq > 1) {
2361 pr_err("read_partial_message bad seq %lld expected %lld\n",
2362 seq, con->in_seq + 1);
2363 con->error_msg = "bad message sequence # for incoming message";
2367 /* allocate message? */
2371 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2372 front_len, data_len);
2373 ret = ceph_con_in_msg_alloc(con, &skip);
2377 /* skip this message */
2378 dout("alloc_msg said skip message\n");
2379 BUG_ON(con->in_msg);
2380 con->in_base_pos = -front_len - middle_len - data_len -
2382 con->in_tag = CEPH_MSGR_TAG_READY;
2387 BUG_ON(!con->in_msg);
2388 BUG_ON(con->in_msg->con != con);
2390 m->front.iov_len = 0; /* haven't read it yet */
2392 m->middle->vec.iov_len = 0;
2394 /* prepare for data payload, if any */
2397 prepare_message_data(con->in_msg, &con->in_msg_pos);
2401 ret = read_partial_message_section(con, &m->front, front_len,
2402 &con->in_front_crc);
2408 ret = read_partial_message_section(con, &m->middle->vec,
2410 &con->in_middle_crc);
2417 ret = read_partial_msg_data(con);
2423 size = sizeof (m->footer);
2425 ret = read_partial(con, end, size, &m->footer);
2429 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2430 m, front_len, m->footer.front_crc, middle_len,
2431 m->footer.middle_crc, data_len, m->footer.data_crc);
2434 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2435 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2436 m, con->in_front_crc, m->footer.front_crc);
2439 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2440 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2441 m, con->in_middle_crc, m->footer.middle_crc);
2445 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2446 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2447 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2448 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2452 return 1; /* done! */
2456 * Process message. This happens in the worker thread. The callback should
2457 * be careful not to do anything that waits on other incoming messages or it
2460 static void process_message(struct ceph_connection *con)
2462 struct ceph_msg *msg;
2464 BUG_ON(con->in_msg->con != con);
2465 con->in_msg->con = NULL;
2470 /* if first message, set peer_name */
2471 if (con->peer_name.type == 0)
2472 con->peer_name = msg->hdr.src;
2475 mutex_unlock(&con->mutex);
2477 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2478 msg, le64_to_cpu(msg->hdr.seq),
2479 ENTITY_NAME(msg->hdr.src),
2480 le16_to_cpu(msg->hdr.type),
2481 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2482 le32_to_cpu(msg->hdr.front_len),
2483 le32_to_cpu(msg->hdr.data_len),
2484 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2485 con->ops->dispatch(con, msg);
2487 mutex_lock(&con->mutex);
2492 * Write something to the socket. Called in a worker thread when the
2493 * socket appears to be writeable and we have something ready to send.
2495 static int try_write(struct ceph_connection *con)
2499 dout("try_write start %p state %lu\n", con, con->state);
2502 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2504 /* open the socket first? */
2505 if (con->state == CON_STATE_PREOPEN) {
2507 con->state = CON_STATE_CONNECTING;
2509 con_out_kvec_reset(con);
2510 prepare_write_banner(con);
2511 prepare_read_banner(con);
2513 BUG_ON(con->in_msg);
2514 con->in_tag = CEPH_MSGR_TAG_READY;
2515 dout("try_write initiating connect on %p new state %lu\n",
2517 ret = ceph_tcp_connect(con);
2519 con->error_msg = "connect error";
2525 /* kvec data queued? */
2526 if (con->out_skip) {
2527 ret = write_partial_skip(con);
2531 if (con->out_kvec_left) {
2532 ret = write_partial_kvec(con);
2539 if (con->out_msg_done) {
2540 ceph_msg_put(con->out_msg);
2541 con->out_msg = NULL; /* we're done with this one */
2545 ret = write_partial_message_data(con);
2547 goto more_kvec; /* we need to send the footer, too! */
2551 dout("try_write write_partial_message_data err %d\n",
2558 if (con->state == CON_STATE_OPEN) {
2559 /* is anything else pending? */
2560 if (!list_empty(&con->out_queue)) {
2561 prepare_write_message(con);
2564 if (con->in_seq > con->in_seq_acked) {
2565 prepare_write_ack(con);
2568 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2569 prepare_write_keepalive(con);
2574 /* Nothing to do! */
2575 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2576 dout("try_write nothing else to write.\n");
2579 dout("try_write done on %p ret %d\n", con, ret);
2586 * Read what we can from the socket.
2588 static int try_read(struct ceph_connection *con)
2593 dout("try_read start on %p state %lu\n", con, con->state);
2594 if (con->state != CON_STATE_CONNECTING &&
2595 con->state != CON_STATE_NEGOTIATING &&
2596 con->state != CON_STATE_OPEN)
2601 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2604 if (con->state == CON_STATE_CONNECTING) {
2605 dout("try_read connecting\n");
2606 ret = read_partial_banner(con);
2609 ret = process_banner(con);
2613 con->state = CON_STATE_NEGOTIATING;
2616 * Received banner is good, exchange connection info.
2617 * Do not reset out_kvec, as sending our banner raced
2618 * with receiving peer banner after connect completed.
2620 ret = prepare_write_connect(con);
2623 prepare_read_connect(con);
2625 /* Send connection info before awaiting response */
2629 if (con->state == CON_STATE_NEGOTIATING) {
2630 dout("try_read negotiating\n");
2631 ret = read_partial_connect(con);
2634 ret = process_connect(con);
2640 WARN_ON(con->state != CON_STATE_OPEN);
2642 if (con->in_base_pos < 0) {
2644 * skipping + discarding content.
2646 * FIXME: there must be a better way to do this!
2648 static char buf[SKIP_BUF_SIZE];
2649 int skip = min((int) sizeof (buf), -con->in_base_pos);
2651 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2652 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2655 con->in_base_pos += ret;
2656 if (con->in_base_pos)
2659 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2663 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2666 dout("try_read got tag %d\n", (int)con->in_tag);
2667 switch (con->in_tag) {
2668 case CEPH_MSGR_TAG_MSG:
2669 prepare_read_message(con);
2671 case CEPH_MSGR_TAG_ACK:
2672 prepare_read_ack(con);
2674 case CEPH_MSGR_TAG_CLOSE:
2675 con_close_socket(con);
2676 con->state = CON_STATE_CLOSED;
2682 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2683 ret = read_partial_message(con);
2687 con->error_msg = "bad crc";
2691 con->error_msg = "io error";
2696 if (con->in_tag == CEPH_MSGR_TAG_READY)
2698 process_message(con);
2699 if (con->state == CON_STATE_OPEN)
2700 prepare_read_tag(con);
2703 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2704 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2706 * the final handshake seq exchange is semantically
2707 * equivalent to an ACK
2709 ret = read_partial_ack(con);
2717 dout("try_read done on %p ret %d\n", con, ret);
2721 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2722 con->error_msg = "protocol error, garbage tag";
2729 * Atomically queue work on a connection after the specified delay.
2730 * Bump @con reference to avoid races with connection teardown.
2731 * Returns 0 if work was queued, or an error code otherwise.
2733 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2735 if (!con->ops->get(con)) {
2736 dout("%s %p ref count 0\n", __func__, con);
2741 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2742 dout("%s %p - already queued\n", __func__, con);
2748 dout("%s %p %lu\n", __func__, con, delay);
2753 static void queue_con(struct ceph_connection *con)
2755 (void) queue_con_delay(con, 0);
2758 static bool con_sock_closed(struct ceph_connection *con)
2760 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2764 case CON_STATE_ ## x: \
2765 con->error_msg = "socket closed (con state " #x ")"; \
2768 switch (con->state) {
2776 pr_warning("%s con %p unrecognized state %lu\n",
2777 __func__, con, con->state);
2778 con->error_msg = "unrecognized con state";
2787 static bool con_backoff(struct ceph_connection *con)
2791 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2794 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2796 dout("%s: con %p FAILED to back off %lu\n", __func__,
2798 BUG_ON(ret == -ENOENT);
2799 con_flag_set(con, CON_FLAG_BACKOFF);
2805 /* Finish fault handling; con->mutex must *not* be held here */
2807 static void con_fault_finish(struct ceph_connection *con)
2810 * in case we faulted due to authentication, invalidate our
2811 * current tickets so that we can get new ones.
2813 if (con->auth_retry && con->ops->invalidate_authorizer) {
2814 dout("calling invalidate_authorizer()\n");
2815 con->ops->invalidate_authorizer(con);
2818 if (con->ops->fault)
2819 con->ops->fault(con);
2823 * Do some work on a connection. Drop a connection ref when we're done.
2825 static void con_work(struct work_struct *work)
2827 struct ceph_connection *con = container_of(work, struct ceph_connection,
2831 mutex_lock(&con->mutex);
2835 if ((fault = con_sock_closed(con))) {
2836 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2839 if (con_backoff(con)) {
2840 dout("%s: con %p BACKOFF\n", __func__, con);
2843 if (con->state == CON_STATE_STANDBY) {
2844 dout("%s: con %p STANDBY\n", __func__, con);
2847 if (con->state == CON_STATE_CLOSED) {
2848 dout("%s: con %p CLOSED\n", __func__, con);
2852 if (con->state == CON_STATE_PREOPEN) {
2853 dout("%s: con %p PREOPEN\n", __func__, con);
2857 ret = try_read(con);
2861 con->error_msg = "socket error on read";
2866 ret = try_write(con);
2870 con->error_msg = "socket error on write";
2874 break; /* If we make it to here, we're done */
2878 mutex_unlock(&con->mutex);
2881 con_fault_finish(con);
2887 * Generic error/fault handler. A retry mechanism is used with
2888 * exponential backoff
2890 static void con_fault(struct ceph_connection *con)
2892 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2893 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2894 dout("fault %p state %lu to peer %s\n",
2895 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2897 WARN_ON(con->state != CON_STATE_CONNECTING &&
2898 con->state != CON_STATE_NEGOTIATING &&
2899 con->state != CON_STATE_OPEN);
2901 con_close_socket(con);
2903 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2904 dout("fault on LOSSYTX channel, marking CLOSED\n");
2905 con->state = CON_STATE_CLOSED;
2910 BUG_ON(con->in_msg->con != con);
2911 con->in_msg->con = NULL;
2912 ceph_msg_put(con->in_msg);
2917 /* Requeue anything that hasn't been acked */
2918 list_splice_init(&con->out_sent, &con->out_queue);
2920 /* If there are no messages queued or keepalive pending, place
2921 * the connection in a STANDBY state */
2922 if (list_empty(&con->out_queue) &&
2923 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
2924 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2925 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2926 con->state = CON_STATE_STANDBY;
2928 /* retry after a delay. */
2929 con->state = CON_STATE_PREOPEN;
2930 if (con->delay == 0)
2931 con->delay = BASE_DELAY_INTERVAL;
2932 else if (con->delay < MAX_DELAY_INTERVAL)
2934 con_flag_set(con, CON_FLAG_BACKOFF);
2942 * initialize a new messenger instance
2944 void ceph_messenger_init(struct ceph_messenger *msgr,
2945 struct ceph_entity_addr *myaddr,
2946 u32 supported_features,
2947 u32 required_features,
2950 msgr->supported_features = supported_features;
2951 msgr->required_features = required_features;
2953 spin_lock_init(&msgr->global_seq_lock);
2956 msgr->inst.addr = *myaddr;
2958 /* select a random nonce */
2959 msgr->inst.addr.type = 0;
2960 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2961 encode_my_addr(msgr);
2962 msgr->nocrc = nocrc;
2964 atomic_set(&msgr->stopping, 0);
2966 dout("%s %p\n", __func__, msgr);
2968 EXPORT_SYMBOL(ceph_messenger_init);
2970 static void clear_standby(struct ceph_connection *con)
2972 /* come back from STANDBY? */
2973 if (con->state == CON_STATE_STANDBY) {
2974 dout("clear_standby %p and ++connect_seq\n", con);
2975 con->state = CON_STATE_PREOPEN;
2977 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2978 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
2983 * Queue up an outgoing message on the given connection.
2985 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2988 msg->hdr.src = con->msgr->inst.name;
2989 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2990 msg->needs_out_seq = true;
2992 mutex_lock(&con->mutex);
2994 if (con->state == CON_STATE_CLOSED) {
2995 dout("con_send %p closed, dropping %p\n", con, msg);
2997 mutex_unlock(&con->mutex);
3001 BUG_ON(msg->con != NULL);
3002 msg->con = con->ops->get(con);
3003 BUG_ON(msg->con == NULL);
3005 BUG_ON(!list_empty(&msg->list_head));
3006 list_add_tail(&msg->list_head, &con->out_queue);
3007 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3008 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3009 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3010 le32_to_cpu(msg->hdr.front_len),
3011 le32_to_cpu(msg->hdr.middle_len),
3012 le32_to_cpu(msg->hdr.data_len));
3015 mutex_unlock(&con->mutex);
3017 /* if there wasn't anything waiting to send before, queue
3019 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3022 EXPORT_SYMBOL(ceph_con_send);
3025 * Revoke a message that was previously queued for send
3027 void ceph_msg_revoke(struct ceph_msg *msg)
3029 struct ceph_connection *con = msg->con;
3032 return; /* Message not in our possession */
3034 mutex_lock(&con->mutex);
3035 if (!list_empty(&msg->list_head)) {
3036 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
3037 list_del_init(&msg->list_head);
3038 BUG_ON(msg->con == NULL);
3039 msg->con->ops->put(msg->con);
3045 if (con->out_msg == msg) {
3046 dout("%s %p msg %p - was sending\n", __func__, con, msg);
3047 con->out_msg = NULL;
3048 if (con->out_kvec_is_msg) {
3049 con->out_skip = con->out_kvec_bytes;
3050 con->out_kvec_is_msg = false;
3056 mutex_unlock(&con->mutex);
3060 * Revoke a message that we may be reading data into
3062 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
3064 struct ceph_connection *con;
3066 BUG_ON(msg == NULL);
3068 dout("%s msg %p null con\n", __func__, msg);
3070 return; /* Message not in our possession */
3074 mutex_lock(&con->mutex);
3075 if (con->in_msg == msg) {
3076 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3077 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3078 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
3080 /* skip rest of message */
3081 dout("%s %p msg %p revoked\n", __func__, con, msg);
3082 con->in_base_pos = con->in_base_pos -
3083 sizeof(struct ceph_msg_header) -
3087 sizeof(struct ceph_msg_footer);
3088 ceph_msg_put(con->in_msg);
3090 con->in_tag = CEPH_MSGR_TAG_READY;
3093 dout("%s %p in_msg %p msg %p no-op\n",
3094 __func__, con, con->in_msg, msg);
3096 mutex_unlock(&con->mutex);
3100 * Queue a keepalive byte to ensure the tcp connection is alive.
3102 void ceph_con_keepalive(struct ceph_connection *con)
3104 dout("con_keepalive %p\n", con);
3105 mutex_lock(&con->mutex);
3107 mutex_unlock(&con->mutex);
3108 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3109 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3112 EXPORT_SYMBOL(ceph_con_keepalive);
3114 static void ceph_msg_data_init(struct ceph_msg_data *data)
3116 data->type = CEPH_MSG_DATA_NONE;
3119 void ceph_msg_data_set_pages(struct ceph_msg *msg, struct page **pages,
3120 size_t length, size_t alignment)
3124 BUG_ON(msg->p.type != CEPH_MSG_DATA_NONE);
3126 msg->p.type = CEPH_MSG_DATA_PAGES;
3127 msg->p.pages = pages;
3128 msg->p.length = length;
3129 msg->p.alignment = alignment & ~PAGE_MASK;
3131 EXPORT_SYMBOL(ceph_msg_data_set_pages);
3133 void ceph_msg_data_set_pagelist(struct ceph_msg *msg,
3134 struct ceph_pagelist *pagelist)
3137 BUG_ON(!pagelist->length);
3138 BUG_ON(msg->l.type != CEPH_MSG_DATA_NONE);
3140 msg->l.type = CEPH_MSG_DATA_PAGELIST;
3141 msg->l.pagelist = pagelist;
3143 EXPORT_SYMBOL(ceph_msg_data_set_pagelist);
3145 void ceph_msg_data_set_bio(struct ceph_msg *msg, struct bio *bio)
3148 BUG_ON(msg->b.type != CEPH_MSG_DATA_NONE);
3150 msg->b.type = CEPH_MSG_DATA_BIO;
3153 EXPORT_SYMBOL(ceph_msg_data_set_bio);
3156 * construct a new message with given type, size
3157 * the new msg has a ref count of 1.
3159 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3164 m = kzalloc(sizeof(*m), flags);
3168 m->hdr.type = cpu_to_le16(type);
3169 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3170 m->hdr.front_len = cpu_to_le32(front_len);
3172 INIT_LIST_HEAD(&m->list_head);
3173 kref_init(&m->kref);
3175 ceph_msg_data_init(&m->p);
3176 ceph_msg_data_init(&m->l);
3177 ceph_msg_data_init(&m->b);
3180 m->front_max = front_len;
3182 if (front_len > PAGE_CACHE_SIZE) {
3183 m->front.iov_base = __vmalloc(front_len, flags,
3185 m->front_is_vmalloc = true;
3187 m->front.iov_base = kmalloc(front_len, flags);
3189 if (m->front.iov_base == NULL) {
3190 dout("ceph_msg_new can't allocate %d bytes\n",
3195 m->front.iov_base = NULL;
3197 m->front.iov_len = front_len;
3199 dout("ceph_msg_new %p front %d\n", m, front_len);
3206 pr_err("msg_new can't create type %d front %d\n", type,
3210 dout("msg_new can't create type %d front %d\n", type,
3215 EXPORT_SYMBOL(ceph_msg_new);
3218 * Allocate "middle" portion of a message, if it is needed and wasn't
3219 * allocated by alloc_msg. This allows us to read a small fixed-size
3220 * per-type header in the front and then gracefully fail (i.e.,
3221 * propagate the error to the caller based on info in the front) when
3222 * the middle is too large.
3224 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3226 int type = le16_to_cpu(msg->hdr.type);
3227 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3229 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3230 ceph_msg_type_name(type), middle_len);
3231 BUG_ON(!middle_len);
3232 BUG_ON(msg->middle);
3234 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3241 * Allocate a message for receiving an incoming message on a
3242 * connection, and save the result in con->in_msg. Uses the
3243 * connection's private alloc_msg op if available.
3245 * Returns 0 on success, or a negative error code.
3247 * On success, if we set *skip = 1:
3248 * - the next message should be skipped and ignored.
3249 * - con->in_msg == NULL
3250 * or if we set *skip = 0:
3251 * - con->in_msg is non-null.
3252 * On error (ENOMEM, EAGAIN, ...),
3253 * - con->in_msg == NULL
3255 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3257 struct ceph_msg_header *hdr = &con->in_hdr;
3258 int middle_len = le32_to_cpu(hdr->middle_len);
3259 struct ceph_msg *msg;
3262 BUG_ON(con->in_msg != NULL);
3263 BUG_ON(!con->ops->alloc_msg);
3265 mutex_unlock(&con->mutex);
3266 msg = con->ops->alloc_msg(con, hdr, skip);
3267 mutex_lock(&con->mutex);
3268 if (con->state != CON_STATE_OPEN) {
3276 con->in_msg->con = con->ops->get(con);
3277 BUG_ON(con->in_msg->con == NULL);
3280 * Null message pointer means either we should skip
3281 * this message or we couldn't allocate memory. The
3282 * former is not an error.
3286 con->error_msg = "error allocating memory for incoming message";
3290 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3292 if (middle_len && !con->in_msg->middle) {
3293 ret = ceph_alloc_middle(con, con->in_msg);
3295 ceph_msg_put(con->in_msg);
3305 * Free a generically kmalloc'd message.
3307 void ceph_msg_kfree(struct ceph_msg *m)
3309 dout("msg_kfree %p\n", m);
3310 if (m->front_is_vmalloc)
3311 vfree(m->front.iov_base);
3313 kfree(m->front.iov_base);
3318 * Drop a msg ref. Destroy as needed.
3320 void ceph_msg_last_put(struct kref *kref)
3322 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3324 dout("ceph_msg_put last one on %p\n", m);
3325 WARN_ON(!list_empty(&m->list_head));
3327 /* drop middle, data, if any */
3329 ceph_buffer_put(m->middle);
3332 if (ceph_msg_has_pages(m)) {
3337 if (ceph_msg_has_pagelist(m)) {
3338 ceph_pagelist_release(m->l.pagelist);
3339 kfree(m->l.pagelist);
3340 m->l.pagelist = NULL;
3344 ceph_msgpool_put(m->pool, m);
3348 EXPORT_SYMBOL(ceph_msg_last_put);
3350 void ceph_msg_dump(struct ceph_msg *msg)
3352 pr_debug("msg_dump %p (front_max %d length %zd)\n", msg,
3353 msg->front_max, msg->p.length);
3354 print_hex_dump(KERN_DEBUG, "header: ",
3355 DUMP_PREFIX_OFFSET, 16, 1,
3356 &msg->hdr, sizeof(msg->hdr), true);
3357 print_hex_dump(KERN_DEBUG, " front: ",
3358 DUMP_PREFIX_OFFSET, 16, 1,
3359 msg->front.iov_base, msg->front.iov_len, true);
3361 print_hex_dump(KERN_DEBUG, "middle: ",
3362 DUMP_PREFIX_OFFSET, 16, 1,
3363 msg->middle->vec.iov_base,
3364 msg->middle->vec.iov_len, true);
3365 print_hex_dump(KERN_DEBUG, "footer: ",
3366 DUMP_PREFIX_OFFSET, 16, 1,
3367 &msg->footer, sizeof(msg->footer), true);
3369 EXPORT_SYMBOL(ceph_msg_dump);