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;
721 * For a bio data item, a piece is whatever remains of the next
722 * entry in the current bio iovec, or the first entry in the next
725 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data *data,
728 struct ceph_msg_data_cursor *cursor = &data->cursor;
731 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
735 BUG_ON(!bio->bi_vcnt);
737 cursor->resid = length;
739 cursor->vector_index = 0;
740 cursor->vector_offset = 0;
741 cursor->last_piece = length <= bio->bi_io_vec[0].bv_len;
744 static struct page *ceph_msg_data_bio_next(struct ceph_msg_data *data,
748 struct ceph_msg_data_cursor *cursor = &data->cursor;
750 struct bio_vec *bio_vec;
753 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
758 index = cursor->vector_index;
759 BUG_ON(index >= (unsigned int) bio->bi_vcnt);
761 bio_vec = &bio->bi_io_vec[index];
762 BUG_ON(cursor->vector_offset >= bio_vec->bv_len);
763 *page_offset = (size_t) (bio_vec->bv_offset + cursor->vector_offset);
764 BUG_ON(*page_offset >= PAGE_SIZE);
765 if (cursor->last_piece) /* pagelist offset is always 0 */
766 *length = cursor->resid;
768 *length = (size_t) (bio_vec->bv_len - cursor->vector_offset);
769 BUG_ON(*length > PAGE_SIZE);
770 BUG_ON(*length > cursor->resid);
772 return bio_vec->bv_page;
775 static bool ceph_msg_data_bio_advance(struct ceph_msg_data *data, size_t bytes)
777 struct ceph_msg_data_cursor *cursor = &data->cursor;
779 struct bio_vec *bio_vec;
782 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
787 index = cursor->vector_index;
788 BUG_ON(index >= (unsigned int) bio->bi_vcnt);
789 bio_vec = &bio->bi_io_vec[index];
791 /* Advance the cursor offset */
793 BUG_ON(cursor->resid < bytes);
794 cursor->resid -= bytes;
795 cursor->vector_offset += bytes;
796 if (cursor->vector_offset < bio_vec->bv_len)
797 return false; /* more bytes to process in this segment */
798 BUG_ON(cursor->vector_offset != bio_vec->bv_len);
800 /* Move on to the next segment, and possibly the next bio */
802 if (++index == (unsigned int) bio->bi_vcnt) {
807 cursor->vector_index = index;
808 cursor->vector_offset = 0;
810 if (!cursor->last_piece) {
811 BUG_ON(!cursor->resid);
813 /* A short read is OK, so use <= rather than == */
814 if (cursor->resid <= bio->bi_io_vec[index].bv_len)
815 cursor->last_piece = true;
823 * For a page array, a piece comes from the first page in the array
824 * that has not already been fully consumed.
826 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data *data,
829 struct ceph_msg_data_cursor *cursor = &data->cursor;
832 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
834 BUG_ON(!data->pages);
835 BUG_ON(!data->length);
836 BUG_ON(length != data->length);
838 cursor->resid = length;
839 page_count = calc_pages_for(data->alignment, (u64)data->length);
840 cursor->page_offset = data->alignment & ~PAGE_MASK;
841 cursor->page_index = 0;
842 BUG_ON(page_count > (int) USHRT_MAX);
843 cursor->page_count = (unsigned short) page_count;
844 cursor->last_piece = length <= PAGE_SIZE;
847 static struct page *ceph_msg_data_pages_next(struct ceph_msg_data *data,
851 struct ceph_msg_data_cursor *cursor = &data->cursor;
853 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
855 BUG_ON(cursor->page_index >= cursor->page_count);
856 BUG_ON(cursor->page_offset >= PAGE_SIZE);
858 *page_offset = cursor->page_offset;
859 if (cursor->last_piece)
860 *length = cursor->resid;
862 *length = PAGE_SIZE - *page_offset;
864 return data->pages[cursor->page_index];
867 static bool ceph_msg_data_pages_advance(struct ceph_msg_data *data,
870 struct ceph_msg_data_cursor *cursor = &data->cursor;
872 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
874 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
876 /* Advance the cursor page offset */
878 cursor->resid -= bytes;
879 cursor->page_offset += bytes;
880 if (!bytes || cursor->page_offset & ~PAGE_MASK)
881 return false; /* more bytes to process in the current page */
883 /* Move on to the next page */
885 BUG_ON(cursor->page_index >= cursor->page_count);
886 cursor->page_offset = 0;
887 cursor->page_index++;
888 cursor->last_piece = cursor->resid <= PAGE_SIZE;
894 * For a pagelist, a piece is whatever remains to be consumed in the
895 * first page in the list, or the front of the next page.
897 static void ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data *data,
900 struct ceph_msg_data_cursor *cursor = &data->cursor;
901 struct ceph_pagelist *pagelist;
904 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
906 pagelist = data->pagelist;
908 BUG_ON(length != pagelist->length);
911 return; /* pagelist can be assigned but empty */
913 BUG_ON(list_empty(&pagelist->head));
914 page = list_first_entry(&pagelist->head, struct page, lru);
916 cursor->resid = length;
919 cursor->last_piece = length <= PAGE_SIZE;
922 static struct page *ceph_msg_data_pagelist_next(struct ceph_msg_data *data,
926 struct ceph_msg_data_cursor *cursor = &data->cursor;
927 struct ceph_pagelist *pagelist;
929 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
931 pagelist = data->pagelist;
934 BUG_ON(!cursor->page);
935 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
937 *page_offset = cursor->offset & ~PAGE_MASK;
938 if (cursor->last_piece) /* pagelist offset is always 0 */
939 *length = cursor->resid;
941 *length = PAGE_SIZE - *page_offset;
943 return data->cursor.page;
946 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data *data,
949 struct ceph_msg_data_cursor *cursor = &data->cursor;
950 struct ceph_pagelist *pagelist;
952 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
954 pagelist = data->pagelist;
957 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
958 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
960 /* Advance the cursor offset */
962 cursor->resid -= bytes;
963 cursor->offset += bytes;
964 /* pagelist offset is always 0 */
965 if (!bytes || cursor->offset & ~PAGE_MASK)
966 return false; /* more bytes to process in the current page */
968 /* Move on to the next page */
970 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
971 cursor->page = list_entry_next(cursor->page, lru);
972 cursor->last_piece = cursor->resid <= PAGE_SIZE;
978 * Message data is handled (sent or received) in pieces, where each
979 * piece resides on a single page. The network layer might not
980 * consume an entire piece at once. A data item's cursor keeps
981 * track of which piece is next to process and how much remains to
982 * be processed in that piece. It also tracks whether the current
983 * piece is the last one in the data item.
985 static void ceph_msg_data_cursor_init(struct ceph_msg_data *data,
988 switch (data->type) {
989 case CEPH_MSG_DATA_PAGELIST:
990 ceph_msg_data_pagelist_cursor_init(data, length);
992 case CEPH_MSG_DATA_PAGES:
993 ceph_msg_data_pages_cursor_init(data, length);
996 case CEPH_MSG_DATA_BIO:
997 ceph_msg_data_bio_cursor_init(data, length);
999 #endif /* CONFIG_BLOCK */
1000 case CEPH_MSG_DATA_NONE:
1005 data->cursor.need_crc = true;
1009 * Return the page containing the next piece to process for a given
1010 * data item, and supply the page offset and length of that piece.
1011 * Indicate whether this is the last piece in this data item.
1013 static struct page *ceph_msg_data_next(struct ceph_msg_data *data,
1014 size_t *page_offset,
1020 switch (data->type) {
1021 case CEPH_MSG_DATA_PAGELIST:
1022 page = ceph_msg_data_pagelist_next(data, page_offset, length);
1024 case CEPH_MSG_DATA_PAGES:
1025 page = ceph_msg_data_pages_next(data, page_offset, length);
1028 case CEPH_MSG_DATA_BIO:
1029 page = ceph_msg_data_bio_next(data, page_offset, length);
1031 #endif /* CONFIG_BLOCK */
1032 case CEPH_MSG_DATA_NONE:
1038 BUG_ON(*page_offset + *length > PAGE_SIZE);
1041 *last_piece = data->cursor.last_piece;
1047 * Returns true if the result moves the cursor on to the next piece
1050 static bool ceph_msg_data_advance(struct ceph_msg_data *data, size_t bytes)
1052 struct ceph_msg_data_cursor *cursor = &data->cursor;
1055 BUG_ON(bytes > cursor->resid);
1056 switch (data->type) {
1057 case CEPH_MSG_DATA_PAGELIST:
1058 new_piece = ceph_msg_data_pagelist_advance(data, bytes);
1060 case CEPH_MSG_DATA_PAGES:
1061 new_piece = ceph_msg_data_pages_advance(data, bytes);
1064 case CEPH_MSG_DATA_BIO:
1065 new_piece = ceph_msg_data_bio_advance(data, bytes);
1067 #endif /* CONFIG_BLOCK */
1068 case CEPH_MSG_DATA_NONE:
1073 data->cursor.need_crc = new_piece;
1078 static void prepare_message_data(struct ceph_msg *msg)
1084 data_len = le32_to_cpu(msg->hdr.data_len);
1087 /* Initialize data cursor */
1089 ceph_msg_data_cursor_init(&msg->data, data_len);
1093 * Prepare footer for currently outgoing message, and finish things
1094 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1096 static void prepare_write_message_footer(struct ceph_connection *con)
1098 struct ceph_msg *m = con->out_msg;
1099 int v = con->out_kvec_left;
1101 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1103 dout("prepare_write_message_footer %p\n", con);
1104 con->out_kvec_is_msg = true;
1105 con->out_kvec[v].iov_base = &m->footer;
1106 con->out_kvec[v].iov_len = sizeof(m->footer);
1107 con->out_kvec_bytes += sizeof(m->footer);
1108 con->out_kvec_left++;
1109 con->out_more = m->more_to_follow;
1110 con->out_msg_done = true;
1114 * Prepare headers for the next outgoing message.
1116 static void prepare_write_message(struct ceph_connection *con)
1121 con_out_kvec_reset(con);
1122 con->out_kvec_is_msg = true;
1123 con->out_msg_done = false;
1125 /* Sneak an ack in there first? If we can get it into the same
1126 * TCP packet that's a good thing. */
1127 if (con->in_seq > con->in_seq_acked) {
1128 con->in_seq_acked = con->in_seq;
1129 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1130 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1131 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1132 &con->out_temp_ack);
1135 BUG_ON(list_empty(&con->out_queue));
1136 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1138 BUG_ON(m->con != con);
1140 /* put message on sent list */
1142 list_move_tail(&m->list_head, &con->out_sent);
1145 * only assign outgoing seq # if we haven't sent this message
1146 * yet. if it is requeued, resend with it's original seq.
1148 if (m->needs_out_seq) {
1149 m->hdr.seq = cpu_to_le64(++con->out_seq);
1150 m->needs_out_seq = false;
1153 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d\n",
1154 m, con->out_seq, le16_to_cpu(m->hdr.type),
1155 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1156 le32_to_cpu(m->hdr.data_len));
1157 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1159 /* tag + hdr + front + middle */
1160 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1161 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1162 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1165 con_out_kvec_add(con, m->middle->vec.iov_len,
1166 m->middle->vec.iov_base);
1168 /* fill in crc (except data pages), footer */
1169 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1170 con->out_msg->hdr.crc = cpu_to_le32(crc);
1171 con->out_msg->footer.flags = 0;
1173 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1174 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1176 crc = crc32c(0, m->middle->vec.iov_base,
1177 m->middle->vec.iov_len);
1178 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1180 con->out_msg->footer.middle_crc = 0;
1181 dout("%s front_crc %u middle_crc %u\n", __func__,
1182 le32_to_cpu(con->out_msg->footer.front_crc),
1183 le32_to_cpu(con->out_msg->footer.middle_crc));
1185 /* is there a data payload? */
1186 con->out_msg->footer.data_crc = 0;
1187 if (m->hdr.data_len) {
1188 prepare_message_data(con->out_msg);
1189 con->out_more = 1; /* data + footer will follow */
1191 /* no, queue up footer too and be done */
1192 prepare_write_message_footer(con);
1195 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1201 static void prepare_write_ack(struct ceph_connection *con)
1203 dout("prepare_write_ack %p %llu -> %llu\n", con,
1204 con->in_seq_acked, con->in_seq);
1205 con->in_seq_acked = con->in_seq;
1207 con_out_kvec_reset(con);
1209 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1211 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1212 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1213 &con->out_temp_ack);
1215 con->out_more = 1; /* more will follow.. eventually.. */
1216 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1220 * Prepare to share the seq during handshake
1222 static void prepare_write_seq(struct ceph_connection *con)
1224 dout("prepare_write_seq %p %llu -> %llu\n", con,
1225 con->in_seq_acked, con->in_seq);
1226 con->in_seq_acked = con->in_seq;
1228 con_out_kvec_reset(con);
1230 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1231 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1232 &con->out_temp_ack);
1234 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1238 * Prepare to write keepalive byte.
1240 static void prepare_write_keepalive(struct ceph_connection *con)
1242 dout("prepare_write_keepalive %p\n", con);
1243 con_out_kvec_reset(con);
1244 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
1245 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1249 * Connection negotiation.
1252 static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1255 struct ceph_auth_handshake *auth;
1257 if (!con->ops->get_authorizer) {
1258 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1259 con->out_connect.authorizer_len = 0;
1263 /* Can't hold the mutex while getting authorizer */
1264 mutex_unlock(&con->mutex);
1265 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
1266 mutex_lock(&con->mutex);
1270 if (con->state != CON_STATE_NEGOTIATING)
1271 return ERR_PTR(-EAGAIN);
1273 con->auth_reply_buf = auth->authorizer_reply_buf;
1274 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
1279 * We connected to a peer and are saying hello.
1281 static void prepare_write_banner(struct ceph_connection *con)
1283 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1284 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1285 &con->msgr->my_enc_addr);
1288 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1291 static int prepare_write_connect(struct ceph_connection *con)
1293 unsigned int global_seq = get_global_seq(con->msgr, 0);
1296 struct ceph_auth_handshake *auth;
1298 switch (con->peer_name.type) {
1299 case CEPH_ENTITY_TYPE_MON:
1300 proto = CEPH_MONC_PROTOCOL;
1302 case CEPH_ENTITY_TYPE_OSD:
1303 proto = CEPH_OSDC_PROTOCOL;
1305 case CEPH_ENTITY_TYPE_MDS:
1306 proto = CEPH_MDSC_PROTOCOL;
1312 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1313 con->connect_seq, global_seq, proto);
1315 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
1316 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1317 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1318 con->out_connect.global_seq = cpu_to_le32(global_seq);
1319 con->out_connect.protocol_version = cpu_to_le32(proto);
1320 con->out_connect.flags = 0;
1322 auth_proto = CEPH_AUTH_UNKNOWN;
1323 auth = get_connect_authorizer(con, &auth_proto);
1325 return PTR_ERR(auth);
1327 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1328 con->out_connect.authorizer_len = auth ?
1329 cpu_to_le32(auth->authorizer_buf_len) : 0;
1331 con_out_kvec_add(con, sizeof (con->out_connect),
1333 if (auth && auth->authorizer_buf_len)
1334 con_out_kvec_add(con, auth->authorizer_buf_len,
1335 auth->authorizer_buf);
1338 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1344 * write as much of pending kvecs to the socket as we can.
1346 * 0 -> socket full, but more to do
1349 static int write_partial_kvec(struct ceph_connection *con)
1353 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1354 while (con->out_kvec_bytes > 0) {
1355 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1356 con->out_kvec_left, con->out_kvec_bytes,
1360 con->out_kvec_bytes -= ret;
1361 if (con->out_kvec_bytes == 0)
1364 /* account for full iov entries consumed */
1365 while (ret >= con->out_kvec_cur->iov_len) {
1366 BUG_ON(!con->out_kvec_left);
1367 ret -= con->out_kvec_cur->iov_len;
1368 con->out_kvec_cur++;
1369 con->out_kvec_left--;
1371 /* and for a partially-consumed entry */
1373 con->out_kvec_cur->iov_len -= ret;
1374 con->out_kvec_cur->iov_base += ret;
1377 con->out_kvec_left = 0;
1378 con->out_kvec_is_msg = false;
1381 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1382 con->out_kvec_bytes, con->out_kvec_left, ret);
1383 return ret; /* done! */
1386 static void out_msg_pos_next(struct ceph_connection *con, struct page *page,
1387 size_t len, size_t sent)
1389 struct ceph_msg *msg = con->out_msg;
1395 need_crc = ceph_msg_data_advance(&msg->data, sent);
1396 BUG_ON(need_crc && sent != len);
1401 BUG_ON(sent != len);
1404 static void in_msg_pos_next(struct ceph_connection *con, size_t len,
1407 struct ceph_msg *msg = con->in_msg;
1412 (void) ceph_msg_data_advance(&msg->data, received);
1417 BUG_ON(received != len);
1420 static u32 ceph_crc32c_page(u32 crc, struct page *page,
1421 unsigned int page_offset,
1422 unsigned int length)
1427 BUG_ON(kaddr == NULL);
1428 crc = crc32c(crc, kaddr + page_offset, length);
1434 * Write as much message data payload as we can. If we finish, queue
1436 * 1 -> done, footer is now queued in out_kvec[].
1437 * 0 -> socket full, but more to do
1440 static int write_partial_message_data(struct ceph_connection *con)
1442 struct ceph_msg *msg = con->out_msg;
1443 struct ceph_msg_data_cursor *cursor = &msg->data.cursor;
1444 bool do_datacrc = !con->msgr->nocrc;
1447 dout("%s %p msg %p\n", __func__, con, msg);
1449 if (WARN_ON(!ceph_msg_has_data(msg)))
1453 * Iterate through each page that contains data to be
1454 * written, and send as much as possible for each.
1456 * If we are calculating the data crc (the default), we will
1457 * need to map the page. If we have no pages, they have
1458 * been revoked, so use the zero page.
1460 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1461 while (cursor->resid) {
1468 page = ceph_msg_data_next(&msg->data, &page_offset, &length,
1470 if (do_datacrc && cursor->need_crc)
1471 crc = ceph_crc32c_page(crc, page, page_offset, length);
1472 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
1473 length, last_piece);
1476 msg->footer.data_crc = cpu_to_le32(crc);
1480 out_msg_pos_next(con, page, length, (size_t) ret);
1483 dout("%s %p msg %p done\n", __func__, con, msg);
1485 /* prepare and queue up footer, too */
1487 msg->footer.data_crc = cpu_to_le32(crc);
1489 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1490 con_out_kvec_reset(con);
1491 prepare_write_message_footer(con);
1493 return 1; /* must return > 0 to indicate success */
1499 static int write_partial_skip(struct ceph_connection *con)
1503 while (con->out_skip > 0) {
1504 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
1506 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
1509 con->out_skip -= ret;
1517 * Prepare to read connection handshake, or an ack.
1519 static void prepare_read_banner(struct ceph_connection *con)
1521 dout("prepare_read_banner %p\n", con);
1522 con->in_base_pos = 0;
1525 static void prepare_read_connect(struct ceph_connection *con)
1527 dout("prepare_read_connect %p\n", con);
1528 con->in_base_pos = 0;
1531 static void prepare_read_ack(struct ceph_connection *con)
1533 dout("prepare_read_ack %p\n", con);
1534 con->in_base_pos = 0;
1537 static void prepare_read_seq(struct ceph_connection *con)
1539 dout("prepare_read_seq %p\n", con);
1540 con->in_base_pos = 0;
1541 con->in_tag = CEPH_MSGR_TAG_SEQ;
1544 static void prepare_read_tag(struct ceph_connection *con)
1546 dout("prepare_read_tag %p\n", con);
1547 con->in_base_pos = 0;
1548 con->in_tag = CEPH_MSGR_TAG_READY;
1552 * Prepare to read a message.
1554 static int prepare_read_message(struct ceph_connection *con)
1556 dout("prepare_read_message %p\n", con);
1557 BUG_ON(con->in_msg != NULL);
1558 con->in_base_pos = 0;
1559 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1564 static int read_partial(struct ceph_connection *con,
1565 int end, int size, void *object)
1567 while (con->in_base_pos < end) {
1568 int left = end - con->in_base_pos;
1569 int have = size - left;
1570 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1573 con->in_base_pos += ret;
1580 * Read all or part of the connect-side handshake on a new connection
1582 static int read_partial_banner(struct ceph_connection *con)
1588 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1591 size = strlen(CEPH_BANNER);
1593 ret = read_partial(con, end, size, con->in_banner);
1597 size = sizeof (con->actual_peer_addr);
1599 ret = read_partial(con, end, size, &con->actual_peer_addr);
1603 size = sizeof (con->peer_addr_for_me);
1605 ret = read_partial(con, end, size, &con->peer_addr_for_me);
1613 static int read_partial_connect(struct ceph_connection *con)
1619 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1621 size = sizeof (con->in_reply);
1623 ret = read_partial(con, end, size, &con->in_reply);
1627 size = le32_to_cpu(con->in_reply.authorizer_len);
1629 ret = read_partial(con, end, size, con->auth_reply_buf);
1633 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1634 con, (int)con->in_reply.tag,
1635 le32_to_cpu(con->in_reply.connect_seq),
1636 le32_to_cpu(con->in_reply.global_seq));
1643 * Verify the hello banner looks okay.
1645 static int verify_hello(struct ceph_connection *con)
1647 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1648 pr_err("connect to %s got bad banner\n",
1649 ceph_pr_addr(&con->peer_addr.in_addr));
1650 con->error_msg = "protocol error, bad banner";
1656 static bool addr_is_blank(struct sockaddr_storage *ss)
1658 switch (ss->ss_family) {
1660 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1663 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1664 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1665 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1666 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1671 static int addr_port(struct sockaddr_storage *ss)
1673 switch (ss->ss_family) {
1675 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1677 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1682 static void addr_set_port(struct sockaddr_storage *ss, int p)
1684 switch (ss->ss_family) {
1686 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1689 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1695 * Unlike other *_pton function semantics, zero indicates success.
1697 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1698 char delim, const char **ipend)
1700 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1701 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1703 memset(ss, 0, sizeof(*ss));
1705 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1706 ss->ss_family = AF_INET;
1710 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1711 ss->ss_family = AF_INET6;
1719 * Extract hostname string and resolve using kernel DNS facility.
1721 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1722 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1723 struct sockaddr_storage *ss, char delim, const char **ipend)
1725 const char *end, *delim_p;
1726 char *colon_p, *ip_addr = NULL;
1730 * The end of the hostname occurs immediately preceding the delimiter or
1731 * the port marker (':') where the delimiter takes precedence.
1733 delim_p = memchr(name, delim, namelen);
1734 colon_p = memchr(name, ':', namelen);
1736 if (delim_p && colon_p)
1737 end = delim_p < colon_p ? delim_p : colon_p;
1738 else if (!delim_p && colon_p)
1742 if (!end) /* case: hostname:/ */
1743 end = name + namelen;
1749 /* do dns_resolve upcall */
1750 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1752 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1760 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1761 ret, ret ? "failed" : ceph_pr_addr(ss));
1766 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1767 struct sockaddr_storage *ss, char delim, const char **ipend)
1774 * Parse a server name (IP or hostname). If a valid IP address is not found
1775 * then try to extract a hostname to resolve using userspace DNS upcall.
1777 static int ceph_parse_server_name(const char *name, size_t namelen,
1778 struct sockaddr_storage *ss, char delim, const char **ipend)
1782 ret = ceph_pton(name, namelen, ss, delim, ipend);
1784 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1790 * Parse an ip[:port] list into an addr array. Use the default
1791 * monitor port if a port isn't specified.
1793 int ceph_parse_ips(const char *c, const char *end,
1794 struct ceph_entity_addr *addr,
1795 int max_count, int *count)
1797 int i, ret = -EINVAL;
1800 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1801 for (i = 0; i < max_count; i++) {
1803 struct sockaddr_storage *ss = &addr[i].in_addr;
1812 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1821 dout("missing matching ']'\n");
1828 if (p < end && *p == ':') {
1831 while (p < end && *p >= '0' && *p <= '9') {
1832 port = (port * 10) + (*p - '0');
1835 if (port > 65535 || port == 0)
1838 port = CEPH_MON_PORT;
1841 addr_set_port(ss, port);
1843 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1860 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1863 EXPORT_SYMBOL(ceph_parse_ips);
1865 static int process_banner(struct ceph_connection *con)
1867 dout("process_banner on %p\n", con);
1869 if (verify_hello(con) < 0)
1872 ceph_decode_addr(&con->actual_peer_addr);
1873 ceph_decode_addr(&con->peer_addr_for_me);
1876 * Make sure the other end is who we wanted. note that the other
1877 * end may not yet know their ip address, so if it's 0.0.0.0, give
1878 * them the benefit of the doubt.
1880 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1881 sizeof(con->peer_addr)) != 0 &&
1882 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1883 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1884 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1885 ceph_pr_addr(&con->peer_addr.in_addr),
1886 (int)le32_to_cpu(con->peer_addr.nonce),
1887 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1888 (int)le32_to_cpu(con->actual_peer_addr.nonce));
1889 con->error_msg = "wrong peer at address";
1894 * did we learn our address?
1896 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1897 int port = addr_port(&con->msgr->inst.addr.in_addr);
1899 memcpy(&con->msgr->inst.addr.in_addr,
1900 &con->peer_addr_for_me.in_addr,
1901 sizeof(con->peer_addr_for_me.in_addr));
1902 addr_set_port(&con->msgr->inst.addr.in_addr, port);
1903 encode_my_addr(con->msgr);
1904 dout("process_banner learned my addr is %s\n",
1905 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
1911 static int process_connect(struct ceph_connection *con)
1913 u64 sup_feat = con->msgr->supported_features;
1914 u64 req_feat = con->msgr->required_features;
1915 u64 server_feat = le64_to_cpu(con->in_reply.features);
1918 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1920 switch (con->in_reply.tag) {
1921 case CEPH_MSGR_TAG_FEATURES:
1922 pr_err("%s%lld %s feature set mismatch,"
1923 " my %llx < server's %llx, missing %llx\n",
1924 ENTITY_NAME(con->peer_name),
1925 ceph_pr_addr(&con->peer_addr.in_addr),
1926 sup_feat, server_feat, server_feat & ~sup_feat);
1927 con->error_msg = "missing required protocol features";
1928 reset_connection(con);
1931 case CEPH_MSGR_TAG_BADPROTOVER:
1932 pr_err("%s%lld %s protocol version mismatch,"
1933 " my %d != server's %d\n",
1934 ENTITY_NAME(con->peer_name),
1935 ceph_pr_addr(&con->peer_addr.in_addr),
1936 le32_to_cpu(con->out_connect.protocol_version),
1937 le32_to_cpu(con->in_reply.protocol_version));
1938 con->error_msg = "protocol version mismatch";
1939 reset_connection(con);
1942 case CEPH_MSGR_TAG_BADAUTHORIZER:
1944 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1946 if (con->auth_retry == 2) {
1947 con->error_msg = "connect authorization failure";
1950 con_out_kvec_reset(con);
1951 ret = prepare_write_connect(con);
1954 prepare_read_connect(con);
1957 case CEPH_MSGR_TAG_RESETSESSION:
1959 * If we connected with a large connect_seq but the peer
1960 * has no record of a session with us (no connection, or
1961 * connect_seq == 0), they will send RESETSESION to indicate
1962 * that they must have reset their session, and may have
1965 dout("process_connect got RESET peer seq %u\n",
1966 le32_to_cpu(con->in_reply.connect_seq));
1967 pr_err("%s%lld %s connection reset\n",
1968 ENTITY_NAME(con->peer_name),
1969 ceph_pr_addr(&con->peer_addr.in_addr));
1970 reset_connection(con);
1971 con_out_kvec_reset(con);
1972 ret = prepare_write_connect(con);
1975 prepare_read_connect(con);
1977 /* Tell ceph about it. */
1978 mutex_unlock(&con->mutex);
1979 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
1980 if (con->ops->peer_reset)
1981 con->ops->peer_reset(con);
1982 mutex_lock(&con->mutex);
1983 if (con->state != CON_STATE_NEGOTIATING)
1987 case CEPH_MSGR_TAG_RETRY_SESSION:
1989 * If we sent a smaller connect_seq than the peer has, try
1990 * again with a larger value.
1992 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
1993 le32_to_cpu(con->out_connect.connect_seq),
1994 le32_to_cpu(con->in_reply.connect_seq));
1995 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
1996 con_out_kvec_reset(con);
1997 ret = prepare_write_connect(con);
2000 prepare_read_connect(con);
2003 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2005 * If we sent a smaller global_seq than the peer has, try
2006 * again with a larger value.
2008 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2009 con->peer_global_seq,
2010 le32_to_cpu(con->in_reply.global_seq));
2011 get_global_seq(con->msgr,
2012 le32_to_cpu(con->in_reply.global_seq));
2013 con_out_kvec_reset(con);
2014 ret = prepare_write_connect(con);
2017 prepare_read_connect(con);
2020 case CEPH_MSGR_TAG_SEQ:
2021 case CEPH_MSGR_TAG_READY:
2022 if (req_feat & ~server_feat) {
2023 pr_err("%s%lld %s protocol feature mismatch,"
2024 " my required %llx > server's %llx, need %llx\n",
2025 ENTITY_NAME(con->peer_name),
2026 ceph_pr_addr(&con->peer_addr.in_addr),
2027 req_feat, server_feat, req_feat & ~server_feat);
2028 con->error_msg = "missing required protocol features";
2029 reset_connection(con);
2033 WARN_ON(con->state != CON_STATE_NEGOTIATING);
2034 con->state = CON_STATE_OPEN;
2035 con->auth_retry = 0; /* we authenticated; clear flag */
2036 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2038 con->peer_features = server_feat;
2039 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2040 con->peer_global_seq,
2041 le32_to_cpu(con->in_reply.connect_seq),
2043 WARN_ON(con->connect_seq !=
2044 le32_to_cpu(con->in_reply.connect_seq));
2046 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2047 con_flag_set(con, CON_FLAG_LOSSYTX);
2049 con->delay = 0; /* reset backoff memory */
2051 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2052 prepare_write_seq(con);
2053 prepare_read_seq(con);
2055 prepare_read_tag(con);
2059 case CEPH_MSGR_TAG_WAIT:
2061 * If there is a connection race (we are opening
2062 * connections to each other), one of us may just have
2063 * to WAIT. This shouldn't happen if we are the
2066 pr_err("process_connect got WAIT as client\n");
2067 con->error_msg = "protocol error, got WAIT as client";
2071 pr_err("connect protocol error, will retry\n");
2072 con->error_msg = "protocol error, garbage tag during connect";
2080 * read (part of) an ack
2082 static int read_partial_ack(struct ceph_connection *con)
2084 int size = sizeof (con->in_temp_ack);
2087 return read_partial(con, end, size, &con->in_temp_ack);
2091 * We can finally discard anything that's been acked.
2093 static void process_ack(struct ceph_connection *con)
2096 u64 ack = le64_to_cpu(con->in_temp_ack);
2099 while (!list_empty(&con->out_sent)) {
2100 m = list_first_entry(&con->out_sent, struct ceph_msg,
2102 seq = le64_to_cpu(m->hdr.seq);
2105 dout("got ack for seq %llu type %d at %p\n", seq,
2106 le16_to_cpu(m->hdr.type), m);
2107 m->ack_stamp = jiffies;
2110 prepare_read_tag(con);
2114 static int read_partial_message_section(struct ceph_connection *con,
2115 struct kvec *section,
2116 unsigned int sec_len, u32 *crc)
2122 while (section->iov_len < sec_len) {
2123 BUG_ON(section->iov_base == NULL);
2124 left = sec_len - section->iov_len;
2125 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2126 section->iov_len, left);
2129 section->iov_len += ret;
2131 if (section->iov_len == sec_len)
2132 *crc = crc32c(0, section->iov_base, section->iov_len);
2137 static int read_partial_msg_data(struct ceph_connection *con)
2139 struct ceph_msg *msg = con->in_msg;
2140 struct ceph_msg_data_cursor *cursor = &msg->data.cursor;
2141 const bool do_datacrc = !con->msgr->nocrc;
2149 if (WARN_ON(!ceph_msg_has_data(msg)))
2153 crc = con->in_data_crc;
2154 while (cursor->resid) {
2155 page = ceph_msg_data_next(&msg->data, &page_offset, &length,
2157 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2160 con->in_data_crc = crc;
2166 crc = ceph_crc32c_page(crc, page, page_offset, ret);
2167 in_msg_pos_next(con, length, ret);
2170 con->in_data_crc = crc;
2172 return 1; /* must return > 0 to indicate success */
2176 * read (part of) a message.
2178 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2180 static int read_partial_message(struct ceph_connection *con)
2182 struct ceph_msg *m = con->in_msg;
2186 unsigned int front_len, middle_len, data_len;
2187 bool do_datacrc = !con->msgr->nocrc;
2191 dout("read_partial_message con %p msg %p\n", con, m);
2194 size = sizeof (con->in_hdr);
2196 ret = read_partial(con, end, size, &con->in_hdr);
2200 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2201 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2202 pr_err("read_partial_message bad hdr "
2203 " crc %u != expected %u\n",
2204 crc, con->in_hdr.crc);
2208 front_len = le32_to_cpu(con->in_hdr.front_len);
2209 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2211 middle_len = le32_to_cpu(con->in_hdr.middle_len);
2212 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2214 data_len = le32_to_cpu(con->in_hdr.data_len);
2215 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2219 seq = le64_to_cpu(con->in_hdr.seq);
2220 if ((s64)seq - (s64)con->in_seq < 1) {
2221 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2222 ENTITY_NAME(con->peer_name),
2223 ceph_pr_addr(&con->peer_addr.in_addr),
2224 seq, con->in_seq + 1);
2225 con->in_base_pos = -front_len - middle_len - data_len -
2227 con->in_tag = CEPH_MSGR_TAG_READY;
2229 } else if ((s64)seq - (s64)con->in_seq > 1) {
2230 pr_err("read_partial_message bad seq %lld expected %lld\n",
2231 seq, con->in_seq + 1);
2232 con->error_msg = "bad message sequence # for incoming message";
2236 /* allocate message? */
2240 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2241 front_len, data_len);
2242 ret = ceph_con_in_msg_alloc(con, &skip);
2246 /* skip this message */
2247 dout("alloc_msg said skip message\n");
2248 BUG_ON(con->in_msg);
2249 con->in_base_pos = -front_len - middle_len - data_len -
2251 con->in_tag = CEPH_MSGR_TAG_READY;
2256 BUG_ON(!con->in_msg);
2257 BUG_ON(con->in_msg->con != con);
2259 m->front.iov_len = 0; /* haven't read it yet */
2261 m->middle->vec.iov_len = 0;
2263 /* prepare for data payload, if any */
2266 prepare_message_data(con->in_msg);
2270 ret = read_partial_message_section(con, &m->front, front_len,
2271 &con->in_front_crc);
2277 ret = read_partial_message_section(con, &m->middle->vec,
2279 &con->in_middle_crc);
2286 ret = read_partial_msg_data(con);
2292 size = sizeof (m->footer);
2294 ret = read_partial(con, end, size, &m->footer);
2298 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2299 m, front_len, m->footer.front_crc, middle_len,
2300 m->footer.middle_crc, data_len, m->footer.data_crc);
2303 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2304 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2305 m, con->in_front_crc, m->footer.front_crc);
2308 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2309 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2310 m, con->in_middle_crc, m->footer.middle_crc);
2314 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2315 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2316 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2317 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2321 return 1; /* done! */
2325 * Process message. This happens in the worker thread. The callback should
2326 * be careful not to do anything that waits on other incoming messages or it
2329 static void process_message(struct ceph_connection *con)
2331 struct ceph_msg *msg;
2333 BUG_ON(con->in_msg->con != con);
2334 con->in_msg->con = NULL;
2339 /* if first message, set peer_name */
2340 if (con->peer_name.type == 0)
2341 con->peer_name = msg->hdr.src;
2344 mutex_unlock(&con->mutex);
2346 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2347 msg, le64_to_cpu(msg->hdr.seq),
2348 ENTITY_NAME(msg->hdr.src),
2349 le16_to_cpu(msg->hdr.type),
2350 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2351 le32_to_cpu(msg->hdr.front_len),
2352 le32_to_cpu(msg->hdr.data_len),
2353 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2354 con->ops->dispatch(con, msg);
2356 mutex_lock(&con->mutex);
2361 * Write something to the socket. Called in a worker thread when the
2362 * socket appears to be writeable and we have something ready to send.
2364 static int try_write(struct ceph_connection *con)
2368 dout("try_write start %p state %lu\n", con, con->state);
2371 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2373 /* open the socket first? */
2374 if (con->state == CON_STATE_PREOPEN) {
2376 con->state = CON_STATE_CONNECTING;
2378 con_out_kvec_reset(con);
2379 prepare_write_banner(con);
2380 prepare_read_banner(con);
2382 BUG_ON(con->in_msg);
2383 con->in_tag = CEPH_MSGR_TAG_READY;
2384 dout("try_write initiating connect on %p new state %lu\n",
2386 ret = ceph_tcp_connect(con);
2388 con->error_msg = "connect error";
2394 /* kvec data queued? */
2395 if (con->out_skip) {
2396 ret = write_partial_skip(con);
2400 if (con->out_kvec_left) {
2401 ret = write_partial_kvec(con);
2408 if (con->out_msg_done) {
2409 ceph_msg_put(con->out_msg);
2410 con->out_msg = NULL; /* we're done with this one */
2414 ret = write_partial_message_data(con);
2416 goto more_kvec; /* we need to send the footer, too! */
2420 dout("try_write write_partial_message_data err %d\n",
2427 if (con->state == CON_STATE_OPEN) {
2428 /* is anything else pending? */
2429 if (!list_empty(&con->out_queue)) {
2430 prepare_write_message(con);
2433 if (con->in_seq > con->in_seq_acked) {
2434 prepare_write_ack(con);
2437 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2438 prepare_write_keepalive(con);
2443 /* Nothing to do! */
2444 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2445 dout("try_write nothing else to write.\n");
2448 dout("try_write done on %p ret %d\n", con, ret);
2455 * Read what we can from the socket.
2457 static int try_read(struct ceph_connection *con)
2462 dout("try_read start on %p state %lu\n", con, con->state);
2463 if (con->state != CON_STATE_CONNECTING &&
2464 con->state != CON_STATE_NEGOTIATING &&
2465 con->state != CON_STATE_OPEN)
2470 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2473 if (con->state == CON_STATE_CONNECTING) {
2474 dout("try_read connecting\n");
2475 ret = read_partial_banner(con);
2478 ret = process_banner(con);
2482 con->state = CON_STATE_NEGOTIATING;
2485 * Received banner is good, exchange connection info.
2486 * Do not reset out_kvec, as sending our banner raced
2487 * with receiving peer banner after connect completed.
2489 ret = prepare_write_connect(con);
2492 prepare_read_connect(con);
2494 /* Send connection info before awaiting response */
2498 if (con->state == CON_STATE_NEGOTIATING) {
2499 dout("try_read negotiating\n");
2500 ret = read_partial_connect(con);
2503 ret = process_connect(con);
2509 WARN_ON(con->state != CON_STATE_OPEN);
2511 if (con->in_base_pos < 0) {
2513 * skipping + discarding content.
2515 * FIXME: there must be a better way to do this!
2517 static char buf[SKIP_BUF_SIZE];
2518 int skip = min((int) sizeof (buf), -con->in_base_pos);
2520 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2521 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2524 con->in_base_pos += ret;
2525 if (con->in_base_pos)
2528 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2532 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2535 dout("try_read got tag %d\n", (int)con->in_tag);
2536 switch (con->in_tag) {
2537 case CEPH_MSGR_TAG_MSG:
2538 prepare_read_message(con);
2540 case CEPH_MSGR_TAG_ACK:
2541 prepare_read_ack(con);
2543 case CEPH_MSGR_TAG_CLOSE:
2544 con_close_socket(con);
2545 con->state = CON_STATE_CLOSED;
2551 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2552 ret = read_partial_message(con);
2556 con->error_msg = "bad crc";
2560 con->error_msg = "io error";
2565 if (con->in_tag == CEPH_MSGR_TAG_READY)
2567 process_message(con);
2568 if (con->state == CON_STATE_OPEN)
2569 prepare_read_tag(con);
2572 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2573 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2575 * the final handshake seq exchange is semantically
2576 * equivalent to an ACK
2578 ret = read_partial_ack(con);
2586 dout("try_read done on %p ret %d\n", con, ret);
2590 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2591 con->error_msg = "protocol error, garbage tag";
2598 * Atomically queue work on a connection after the specified delay.
2599 * Bump @con reference to avoid races with connection teardown.
2600 * Returns 0 if work was queued, or an error code otherwise.
2602 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2604 if (!con->ops->get(con)) {
2605 dout("%s %p ref count 0\n", __func__, con);
2610 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2611 dout("%s %p - already queued\n", __func__, con);
2617 dout("%s %p %lu\n", __func__, con, delay);
2622 static void queue_con(struct ceph_connection *con)
2624 (void) queue_con_delay(con, 0);
2627 static bool con_sock_closed(struct ceph_connection *con)
2629 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2633 case CON_STATE_ ## x: \
2634 con->error_msg = "socket closed (con state " #x ")"; \
2637 switch (con->state) {
2645 pr_warning("%s con %p unrecognized state %lu\n",
2646 __func__, con, con->state);
2647 con->error_msg = "unrecognized con state";
2656 static bool con_backoff(struct ceph_connection *con)
2660 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2663 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2665 dout("%s: con %p FAILED to back off %lu\n", __func__,
2667 BUG_ON(ret == -ENOENT);
2668 con_flag_set(con, CON_FLAG_BACKOFF);
2674 /* Finish fault handling; con->mutex must *not* be held here */
2676 static void con_fault_finish(struct ceph_connection *con)
2679 * in case we faulted due to authentication, invalidate our
2680 * current tickets so that we can get new ones.
2682 if (con->auth_retry && con->ops->invalidate_authorizer) {
2683 dout("calling invalidate_authorizer()\n");
2684 con->ops->invalidate_authorizer(con);
2687 if (con->ops->fault)
2688 con->ops->fault(con);
2692 * Do some work on a connection. Drop a connection ref when we're done.
2694 static void con_work(struct work_struct *work)
2696 struct ceph_connection *con = container_of(work, struct ceph_connection,
2700 mutex_lock(&con->mutex);
2704 if ((fault = con_sock_closed(con))) {
2705 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2708 if (con_backoff(con)) {
2709 dout("%s: con %p BACKOFF\n", __func__, con);
2712 if (con->state == CON_STATE_STANDBY) {
2713 dout("%s: con %p STANDBY\n", __func__, con);
2716 if (con->state == CON_STATE_CLOSED) {
2717 dout("%s: con %p CLOSED\n", __func__, con);
2721 if (con->state == CON_STATE_PREOPEN) {
2722 dout("%s: con %p PREOPEN\n", __func__, con);
2726 ret = try_read(con);
2730 con->error_msg = "socket error on read";
2735 ret = try_write(con);
2739 con->error_msg = "socket error on write";
2743 break; /* If we make it to here, we're done */
2747 mutex_unlock(&con->mutex);
2750 con_fault_finish(con);
2756 * Generic error/fault handler. A retry mechanism is used with
2757 * exponential backoff
2759 static void con_fault(struct ceph_connection *con)
2761 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2762 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2763 dout("fault %p state %lu to peer %s\n",
2764 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2766 WARN_ON(con->state != CON_STATE_CONNECTING &&
2767 con->state != CON_STATE_NEGOTIATING &&
2768 con->state != CON_STATE_OPEN);
2770 con_close_socket(con);
2772 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2773 dout("fault on LOSSYTX channel, marking CLOSED\n");
2774 con->state = CON_STATE_CLOSED;
2779 BUG_ON(con->in_msg->con != con);
2780 con->in_msg->con = NULL;
2781 ceph_msg_put(con->in_msg);
2786 /* Requeue anything that hasn't been acked */
2787 list_splice_init(&con->out_sent, &con->out_queue);
2789 /* If there are no messages queued or keepalive pending, place
2790 * the connection in a STANDBY state */
2791 if (list_empty(&con->out_queue) &&
2792 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
2793 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2794 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2795 con->state = CON_STATE_STANDBY;
2797 /* retry after a delay. */
2798 con->state = CON_STATE_PREOPEN;
2799 if (con->delay == 0)
2800 con->delay = BASE_DELAY_INTERVAL;
2801 else if (con->delay < MAX_DELAY_INTERVAL)
2803 con_flag_set(con, CON_FLAG_BACKOFF);
2811 * initialize a new messenger instance
2813 void ceph_messenger_init(struct ceph_messenger *msgr,
2814 struct ceph_entity_addr *myaddr,
2815 u32 supported_features,
2816 u32 required_features,
2819 msgr->supported_features = supported_features;
2820 msgr->required_features = required_features;
2822 spin_lock_init(&msgr->global_seq_lock);
2825 msgr->inst.addr = *myaddr;
2827 /* select a random nonce */
2828 msgr->inst.addr.type = 0;
2829 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2830 encode_my_addr(msgr);
2831 msgr->nocrc = nocrc;
2833 atomic_set(&msgr->stopping, 0);
2835 dout("%s %p\n", __func__, msgr);
2837 EXPORT_SYMBOL(ceph_messenger_init);
2839 static void clear_standby(struct ceph_connection *con)
2841 /* come back from STANDBY? */
2842 if (con->state == CON_STATE_STANDBY) {
2843 dout("clear_standby %p and ++connect_seq\n", con);
2844 con->state = CON_STATE_PREOPEN;
2846 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2847 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
2852 * Queue up an outgoing message on the given connection.
2854 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2857 msg->hdr.src = con->msgr->inst.name;
2858 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2859 msg->needs_out_seq = true;
2861 mutex_lock(&con->mutex);
2863 if (con->state == CON_STATE_CLOSED) {
2864 dout("con_send %p closed, dropping %p\n", con, msg);
2866 mutex_unlock(&con->mutex);
2870 BUG_ON(msg->con != NULL);
2871 msg->con = con->ops->get(con);
2872 BUG_ON(msg->con == NULL);
2874 BUG_ON(!list_empty(&msg->list_head));
2875 list_add_tail(&msg->list_head, &con->out_queue);
2876 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2877 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2878 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2879 le32_to_cpu(msg->hdr.front_len),
2880 le32_to_cpu(msg->hdr.middle_len),
2881 le32_to_cpu(msg->hdr.data_len));
2884 mutex_unlock(&con->mutex);
2886 /* if there wasn't anything waiting to send before, queue
2888 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
2891 EXPORT_SYMBOL(ceph_con_send);
2894 * Revoke a message that was previously queued for send
2896 void ceph_msg_revoke(struct ceph_msg *msg)
2898 struct ceph_connection *con = msg->con;
2901 return; /* Message not in our possession */
2903 mutex_lock(&con->mutex);
2904 if (!list_empty(&msg->list_head)) {
2905 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
2906 list_del_init(&msg->list_head);
2907 BUG_ON(msg->con == NULL);
2908 msg->con->ops->put(msg->con);
2914 if (con->out_msg == msg) {
2915 dout("%s %p msg %p - was sending\n", __func__, con, msg);
2916 con->out_msg = NULL;
2917 if (con->out_kvec_is_msg) {
2918 con->out_skip = con->out_kvec_bytes;
2919 con->out_kvec_is_msg = false;
2925 mutex_unlock(&con->mutex);
2929 * Revoke a message that we may be reading data into
2931 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
2933 struct ceph_connection *con;
2935 BUG_ON(msg == NULL);
2937 dout("%s msg %p null con\n", __func__, msg);
2939 return; /* Message not in our possession */
2943 mutex_lock(&con->mutex);
2944 if (con->in_msg == msg) {
2945 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
2946 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
2947 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
2949 /* skip rest of message */
2950 dout("%s %p msg %p revoked\n", __func__, con, msg);
2951 con->in_base_pos = con->in_base_pos -
2952 sizeof(struct ceph_msg_header) -
2956 sizeof(struct ceph_msg_footer);
2957 ceph_msg_put(con->in_msg);
2959 con->in_tag = CEPH_MSGR_TAG_READY;
2962 dout("%s %p in_msg %p msg %p no-op\n",
2963 __func__, con, con->in_msg, msg);
2965 mutex_unlock(&con->mutex);
2969 * Queue a keepalive byte to ensure the tcp connection is alive.
2971 void ceph_con_keepalive(struct ceph_connection *con)
2973 dout("con_keepalive %p\n", con);
2974 mutex_lock(&con->mutex);
2976 mutex_unlock(&con->mutex);
2977 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
2978 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
2981 EXPORT_SYMBOL(ceph_con_keepalive);
2983 static void ceph_msg_data_init(struct ceph_msg_data *data)
2985 data->type = CEPH_MSG_DATA_NONE;
2988 void ceph_msg_data_set_pages(struct ceph_msg *msg, struct page **pages,
2989 size_t length, size_t alignment)
2993 BUG_ON(msg->data.type != CEPH_MSG_DATA_NONE);
2995 msg->data.type = CEPH_MSG_DATA_PAGES;
2996 msg->data.pages = pages;
2997 msg->data.length = length;
2998 msg->data.alignment = alignment & ~PAGE_MASK;
3000 EXPORT_SYMBOL(ceph_msg_data_set_pages);
3002 void ceph_msg_data_set_pagelist(struct ceph_msg *msg,
3003 struct ceph_pagelist *pagelist)
3006 BUG_ON(!pagelist->length);
3007 BUG_ON(msg->data.type != CEPH_MSG_DATA_NONE);
3009 msg->data.type = CEPH_MSG_DATA_PAGELIST;
3010 msg->data.pagelist = pagelist;
3012 EXPORT_SYMBOL(ceph_msg_data_set_pagelist);
3014 void ceph_msg_data_set_bio(struct ceph_msg *msg, struct bio *bio)
3017 BUG_ON(msg->data.type != CEPH_MSG_DATA_NONE);
3019 msg->data.type = CEPH_MSG_DATA_BIO;
3020 msg->data.bio = bio;
3022 EXPORT_SYMBOL(ceph_msg_data_set_bio);
3025 * construct a new message with given type, size
3026 * the new msg has a ref count of 1.
3028 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3033 m = kzalloc(sizeof(*m), flags);
3037 m->hdr.type = cpu_to_le16(type);
3038 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3039 m->hdr.front_len = cpu_to_le32(front_len);
3041 INIT_LIST_HEAD(&m->list_head);
3042 kref_init(&m->kref);
3044 ceph_msg_data_init(&m->data);
3047 m->front_max = front_len;
3049 if (front_len > PAGE_CACHE_SIZE) {
3050 m->front.iov_base = __vmalloc(front_len, flags,
3052 m->front_is_vmalloc = true;
3054 m->front.iov_base = kmalloc(front_len, flags);
3056 if (m->front.iov_base == NULL) {
3057 dout("ceph_msg_new can't allocate %d bytes\n",
3062 m->front.iov_base = NULL;
3064 m->front.iov_len = front_len;
3066 dout("ceph_msg_new %p front %d\n", m, front_len);
3073 pr_err("msg_new can't create type %d front %d\n", type,
3077 dout("msg_new can't create type %d front %d\n", type,
3082 EXPORT_SYMBOL(ceph_msg_new);
3085 * Allocate "middle" portion of a message, if it is needed and wasn't
3086 * allocated by alloc_msg. This allows us to read a small fixed-size
3087 * per-type header in the front and then gracefully fail (i.e.,
3088 * propagate the error to the caller based on info in the front) when
3089 * the middle is too large.
3091 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3093 int type = le16_to_cpu(msg->hdr.type);
3094 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3096 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3097 ceph_msg_type_name(type), middle_len);
3098 BUG_ON(!middle_len);
3099 BUG_ON(msg->middle);
3101 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3108 * Allocate a message for receiving an incoming message on a
3109 * connection, and save the result in con->in_msg. Uses the
3110 * connection's private alloc_msg op if available.
3112 * Returns 0 on success, or a negative error code.
3114 * On success, if we set *skip = 1:
3115 * - the next message should be skipped and ignored.
3116 * - con->in_msg == NULL
3117 * or if we set *skip = 0:
3118 * - con->in_msg is non-null.
3119 * On error (ENOMEM, EAGAIN, ...),
3120 * - con->in_msg == NULL
3122 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3124 struct ceph_msg_header *hdr = &con->in_hdr;
3125 int middle_len = le32_to_cpu(hdr->middle_len);
3126 struct ceph_msg *msg;
3129 BUG_ON(con->in_msg != NULL);
3130 BUG_ON(!con->ops->alloc_msg);
3132 mutex_unlock(&con->mutex);
3133 msg = con->ops->alloc_msg(con, hdr, skip);
3134 mutex_lock(&con->mutex);
3135 if (con->state != CON_STATE_OPEN) {
3143 con->in_msg->con = con->ops->get(con);
3144 BUG_ON(con->in_msg->con == NULL);
3147 * Null message pointer means either we should skip
3148 * this message or we couldn't allocate memory. The
3149 * former is not an error.
3153 con->error_msg = "error allocating memory for incoming message";
3157 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3159 if (middle_len && !con->in_msg->middle) {
3160 ret = ceph_alloc_middle(con, con->in_msg);
3162 ceph_msg_put(con->in_msg);
3172 * Free a generically kmalloc'd message.
3174 void ceph_msg_kfree(struct ceph_msg *m)
3176 dout("msg_kfree %p\n", m);
3177 if (m->front_is_vmalloc)
3178 vfree(m->front.iov_base);
3180 kfree(m->front.iov_base);
3185 * Drop a msg ref. Destroy as needed.
3187 void ceph_msg_last_put(struct kref *kref)
3189 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3191 dout("ceph_msg_put last one on %p\n", m);
3192 WARN_ON(!list_empty(&m->list_head));
3194 /* drop middle, data, if any */
3196 ceph_buffer_put(m->middle);
3199 if (ceph_msg_has_data(m)) {
3200 if (m->data.type == CEPH_MSG_DATA_PAGELIST) {
3201 ceph_pagelist_release(m->data.pagelist);
3202 kfree(m->data.pagelist);
3204 memset(&m->data, 0, sizeof m->data);
3205 ceph_msg_data_init(&m->data);
3209 ceph_msgpool_put(m->pool, m);
3213 EXPORT_SYMBOL(ceph_msg_last_put);
3215 void ceph_msg_dump(struct ceph_msg *msg)
3217 pr_debug("msg_dump %p (front_max %d length %zd)\n", msg,
3218 msg->front_max, msg->data.length);
3219 print_hex_dump(KERN_DEBUG, "header: ",
3220 DUMP_PREFIX_OFFSET, 16, 1,
3221 &msg->hdr, sizeof(msg->hdr), true);
3222 print_hex_dump(KERN_DEBUG, " front: ",
3223 DUMP_PREFIX_OFFSET, 16, 1,
3224 msg->front.iov_base, msg->front.iov_len, true);
3226 print_hex_dump(KERN_DEBUG, "middle: ",
3227 DUMP_PREFIX_OFFSET, 16, 1,
3228 msg->middle->vec.iov_base,
3229 msg->middle->vec.iov_len, true);
3230 print_hex_dump(KERN_DEBUG, "footer: ",
3231 DUMP_PREFIX_OFFSET, 16, 1,
3232 &msg->footer, sizeof(msg->footer), true);
3234 EXPORT_SYMBOL(ceph_msg_dump);