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 /* Slab caches for frequently-allocated structures */
157 static struct kmem_cache *ceph_msg_cache;
158 static struct kmem_cache *ceph_msg_data_cache;
160 /* static tag bytes (protocol control messages) */
161 static char tag_msg = CEPH_MSGR_TAG_MSG;
162 static char tag_ack = CEPH_MSGR_TAG_ACK;
163 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
165 #ifdef CONFIG_LOCKDEP
166 static struct lock_class_key socket_class;
170 * When skipping (ignoring) a block of input we read it into a "skip
171 * buffer," which is this many bytes in size.
173 #define SKIP_BUF_SIZE 1024
175 static void queue_con(struct ceph_connection *con);
176 static void con_work(struct work_struct *);
177 static void con_fault(struct ceph_connection *con);
180 * Nicely render a sockaddr as a string. An array of formatted
181 * strings is used, to approximate reentrancy.
183 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
184 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
185 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
186 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
188 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
189 static atomic_t addr_str_seq = ATOMIC_INIT(0);
191 static struct page *zero_page; /* used in certain error cases */
193 const char *ceph_pr_addr(const struct sockaddr_storage *ss)
197 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
198 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
200 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
203 switch (ss->ss_family) {
205 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
206 ntohs(in4->sin_port));
210 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
211 ntohs(in6->sin6_port));
215 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
221 EXPORT_SYMBOL(ceph_pr_addr);
223 static void encode_my_addr(struct ceph_messenger *msgr)
225 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
226 ceph_encode_addr(&msgr->my_enc_addr);
230 * work queue for all reading and writing to/from the socket.
232 static struct workqueue_struct *ceph_msgr_wq;
234 static int ceph_msgr_slab_init(void)
236 BUG_ON(ceph_msg_cache);
237 ceph_msg_cache = kmem_cache_create("ceph_msg",
238 sizeof (struct ceph_msg),
239 __alignof__(struct ceph_msg), 0, NULL);
244 BUG_ON(ceph_msg_data_cache);
245 ceph_msg_data_cache = kmem_cache_create("ceph_msg_data",
246 sizeof (struct ceph_msg_data),
247 __alignof__(struct ceph_msg_data),
249 if (ceph_msg_data_cache)
252 kmem_cache_destroy(ceph_msg_cache);
253 ceph_msg_cache = NULL;
258 static void ceph_msgr_slab_exit(void)
260 BUG_ON(!ceph_msg_data_cache);
261 kmem_cache_destroy(ceph_msg_data_cache);
262 ceph_msg_data_cache = NULL;
264 BUG_ON(!ceph_msg_cache);
265 kmem_cache_destroy(ceph_msg_cache);
266 ceph_msg_cache = NULL;
269 static void _ceph_msgr_exit(void)
272 destroy_workqueue(ceph_msgr_wq);
276 ceph_msgr_slab_exit();
278 BUG_ON(zero_page == NULL);
280 page_cache_release(zero_page);
284 int ceph_msgr_init(void)
286 BUG_ON(zero_page != NULL);
287 zero_page = ZERO_PAGE(0);
288 page_cache_get(zero_page);
290 if (ceph_msgr_slab_init())
293 ceph_msgr_wq = alloc_workqueue("ceph-msgr",
294 WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
298 pr_err("msgr_init failed to create workqueue\n");
303 EXPORT_SYMBOL(ceph_msgr_init);
305 void ceph_msgr_exit(void)
307 BUG_ON(ceph_msgr_wq == NULL);
311 EXPORT_SYMBOL(ceph_msgr_exit);
313 void ceph_msgr_flush(void)
315 flush_workqueue(ceph_msgr_wq);
317 EXPORT_SYMBOL(ceph_msgr_flush);
319 /* Connection socket state transition functions */
321 static void con_sock_state_init(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_NEW))
327 printk("%s: unexpected old state %d\n", __func__, old_state);
328 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
329 CON_SOCK_STATE_CLOSED);
332 static void con_sock_state_connecting(struct ceph_connection *con)
336 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
337 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
338 printk("%s: unexpected old state %d\n", __func__, old_state);
339 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
340 CON_SOCK_STATE_CONNECTING);
343 static void con_sock_state_connected(struct ceph_connection *con)
347 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
348 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
349 printk("%s: unexpected old state %d\n", __func__, old_state);
350 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
351 CON_SOCK_STATE_CONNECTED);
354 static void con_sock_state_closing(struct ceph_connection *con)
358 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
359 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
360 old_state != CON_SOCK_STATE_CONNECTED &&
361 old_state != CON_SOCK_STATE_CLOSING))
362 printk("%s: unexpected old state %d\n", __func__, old_state);
363 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
364 CON_SOCK_STATE_CLOSING);
367 static void con_sock_state_closed(struct ceph_connection *con)
371 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
372 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
373 old_state != CON_SOCK_STATE_CLOSING &&
374 old_state != CON_SOCK_STATE_CONNECTING &&
375 old_state != CON_SOCK_STATE_CLOSED))
376 printk("%s: unexpected old state %d\n", __func__, old_state);
377 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
378 CON_SOCK_STATE_CLOSED);
382 * socket callback functions
385 /* data available on socket, or listen socket received a connect */
386 static void ceph_sock_data_ready(struct sock *sk, int count_unused)
388 struct ceph_connection *con = sk->sk_user_data;
389 if (atomic_read(&con->msgr->stopping)) {
393 if (sk->sk_state != TCP_CLOSE_WAIT) {
394 dout("%s on %p state = %lu, queueing work\n", __func__,
400 /* socket has buffer space for writing */
401 static void ceph_sock_write_space(struct sock *sk)
403 struct ceph_connection *con = sk->sk_user_data;
405 /* only queue to workqueue if there is data we want to write,
406 * and there is sufficient space in the socket buffer to accept
407 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
408 * doesn't get called again until try_write() fills the socket
409 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
410 * and net/core/stream.c:sk_stream_write_space().
412 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
413 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
414 dout("%s %p queueing write work\n", __func__, con);
415 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
419 dout("%s %p nothing to write\n", __func__, con);
423 /* socket's state has changed */
424 static void ceph_sock_state_change(struct sock *sk)
426 struct ceph_connection *con = sk->sk_user_data;
428 dout("%s %p state = %lu sk_state = %u\n", __func__,
429 con, con->state, sk->sk_state);
431 switch (sk->sk_state) {
433 dout("%s TCP_CLOSE\n", __func__);
435 dout("%s TCP_CLOSE_WAIT\n", __func__);
436 con_sock_state_closing(con);
437 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
440 case TCP_ESTABLISHED:
441 dout("%s TCP_ESTABLISHED\n", __func__);
442 con_sock_state_connected(con);
445 default: /* Everything else is uninteresting */
451 * set up socket callbacks
453 static void set_sock_callbacks(struct socket *sock,
454 struct ceph_connection *con)
456 struct sock *sk = sock->sk;
457 sk->sk_user_data = con;
458 sk->sk_data_ready = ceph_sock_data_ready;
459 sk->sk_write_space = ceph_sock_write_space;
460 sk->sk_state_change = ceph_sock_state_change;
469 * initiate connection to a remote socket.
471 static int ceph_tcp_connect(struct ceph_connection *con)
473 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
478 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
482 sock->sk->sk_allocation = GFP_NOFS;
484 #ifdef CONFIG_LOCKDEP
485 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
488 set_sock_callbacks(sock, con);
490 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
492 con_sock_state_connecting(con);
493 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
495 if (ret == -EINPROGRESS) {
496 dout("connect %s EINPROGRESS sk_state = %u\n",
497 ceph_pr_addr(&con->peer_addr.in_addr),
499 } else if (ret < 0) {
500 pr_err("connect %s error %d\n",
501 ceph_pr_addr(&con->peer_addr.in_addr), ret);
503 con->error_msg = "connect error";
511 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
513 struct kvec iov = {buf, len};
514 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
517 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
523 static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
524 int page_offset, size_t length)
529 BUG_ON(page_offset + length > PAGE_SIZE);
533 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
540 * write something. @more is true if caller will be sending more data
543 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
544 size_t kvlen, size_t len, int more)
546 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
550 msg.msg_flags |= MSG_MORE;
552 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
554 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
560 static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
561 int offset, size_t size, bool more)
563 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
566 ret = kernel_sendpage(sock, page, offset, size, flags);
573 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
574 int offset, size_t size, bool more)
579 /* sendpage cannot properly handle pages with page_count == 0,
580 * we need to fallback to sendmsg if that's the case */
581 if (page_count(page) >= 1)
582 return __ceph_tcp_sendpage(sock, page, offset, size, more);
584 iov.iov_base = kmap(page) + offset;
586 ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more);
593 * Shutdown/close the socket for the given connection.
595 static int con_close_socket(struct ceph_connection *con)
599 dout("con_close_socket on %p sock %p\n", con, con->sock);
601 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
602 sock_release(con->sock);
607 * Forcibly clear the SOCK_CLOSED flag. It gets set
608 * independent of the connection mutex, and we could have
609 * received a socket close event before we had the chance to
610 * shut the socket down.
612 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
614 con_sock_state_closed(con);
619 * Reset a connection. Discard all incoming and outgoing messages
620 * and clear *_seq state.
622 static void ceph_msg_remove(struct ceph_msg *msg)
624 list_del_init(&msg->list_head);
625 BUG_ON(msg->con == NULL);
626 msg->con->ops->put(msg->con);
631 static void ceph_msg_remove_list(struct list_head *head)
633 while (!list_empty(head)) {
634 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
636 ceph_msg_remove(msg);
640 static void reset_connection(struct ceph_connection *con)
642 /* reset connection, out_queue, msg_ and connect_seq */
643 /* discard existing out_queue and msg_seq */
644 dout("reset_connection %p\n", con);
645 ceph_msg_remove_list(&con->out_queue);
646 ceph_msg_remove_list(&con->out_sent);
649 BUG_ON(con->in_msg->con != con);
650 con->in_msg->con = NULL;
651 ceph_msg_put(con->in_msg);
656 con->connect_seq = 0;
659 ceph_msg_put(con->out_msg);
663 con->in_seq_acked = 0;
667 * mark a peer down. drop any open connections.
669 void ceph_con_close(struct ceph_connection *con)
671 mutex_lock(&con->mutex);
672 dout("con_close %p peer %s\n", con,
673 ceph_pr_addr(&con->peer_addr.in_addr));
674 con->state = CON_STATE_CLOSED;
676 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
677 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
678 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
679 con_flag_clear(con, CON_FLAG_BACKOFF);
681 reset_connection(con);
682 con->peer_global_seq = 0;
683 cancel_delayed_work(&con->work);
684 con_close_socket(con);
685 mutex_unlock(&con->mutex);
687 EXPORT_SYMBOL(ceph_con_close);
690 * Reopen a closed connection, with a new peer address.
692 void ceph_con_open(struct ceph_connection *con,
693 __u8 entity_type, __u64 entity_num,
694 struct ceph_entity_addr *addr)
696 mutex_lock(&con->mutex);
697 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
699 WARN_ON(con->state != CON_STATE_CLOSED);
700 con->state = CON_STATE_PREOPEN;
702 con->peer_name.type = (__u8) entity_type;
703 con->peer_name.num = cpu_to_le64(entity_num);
705 memcpy(&con->peer_addr, addr, sizeof(*addr));
706 con->delay = 0; /* reset backoff memory */
707 mutex_unlock(&con->mutex);
710 EXPORT_SYMBOL(ceph_con_open);
713 * return true if this connection ever successfully opened
715 bool ceph_con_opened(struct ceph_connection *con)
717 return con->connect_seq > 0;
721 * initialize a new connection.
723 void ceph_con_init(struct ceph_connection *con, void *private,
724 const struct ceph_connection_operations *ops,
725 struct ceph_messenger *msgr)
727 dout("con_init %p\n", con);
728 memset(con, 0, sizeof(*con));
729 con->private = private;
733 con_sock_state_init(con);
735 mutex_init(&con->mutex);
736 INIT_LIST_HEAD(&con->out_queue);
737 INIT_LIST_HEAD(&con->out_sent);
738 INIT_DELAYED_WORK(&con->work, con_work);
740 con->state = CON_STATE_CLOSED;
742 EXPORT_SYMBOL(ceph_con_init);
746 * We maintain a global counter to order connection attempts. Get
747 * a unique seq greater than @gt.
749 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
753 spin_lock(&msgr->global_seq_lock);
754 if (msgr->global_seq < gt)
755 msgr->global_seq = gt;
756 ret = ++msgr->global_seq;
757 spin_unlock(&msgr->global_seq_lock);
761 static void con_out_kvec_reset(struct ceph_connection *con)
763 con->out_kvec_left = 0;
764 con->out_kvec_bytes = 0;
765 con->out_kvec_cur = &con->out_kvec[0];
768 static void con_out_kvec_add(struct ceph_connection *con,
769 size_t size, void *data)
773 index = con->out_kvec_left;
774 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
776 con->out_kvec[index].iov_len = size;
777 con->out_kvec[index].iov_base = data;
778 con->out_kvec_left++;
779 con->out_kvec_bytes += size;
785 * For a bio data item, a piece is whatever remains of the next
786 * entry in the current bio iovec, or the first entry in the next
789 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
792 struct ceph_msg_data *data = cursor->data;
795 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
799 BUG_ON(!bio->bi_vcnt);
801 cursor->resid = min(length, data->bio_length);
803 cursor->vector_index = 0;
804 cursor->vector_offset = 0;
805 cursor->last_piece = length <= bio->bi_io_vec[0].bv_len;
808 static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
812 struct ceph_msg_data *data = cursor->data;
814 struct bio_vec *bio_vec;
817 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
822 index = cursor->vector_index;
823 BUG_ON(index >= (unsigned int) bio->bi_vcnt);
825 bio_vec = &bio->bi_io_vec[index];
826 BUG_ON(cursor->vector_offset >= bio_vec->bv_len);
827 *page_offset = (size_t) (bio_vec->bv_offset + cursor->vector_offset);
828 BUG_ON(*page_offset >= PAGE_SIZE);
829 if (cursor->last_piece) /* pagelist offset is always 0 */
830 *length = cursor->resid;
832 *length = (size_t) (bio_vec->bv_len - cursor->vector_offset);
833 BUG_ON(*length > cursor->resid);
834 BUG_ON(*page_offset + *length > PAGE_SIZE);
836 return bio_vec->bv_page;
839 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
843 struct bio_vec *bio_vec;
846 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
851 index = cursor->vector_index;
852 BUG_ON(index >= (unsigned int) bio->bi_vcnt);
853 bio_vec = &bio->bi_io_vec[index];
855 /* Advance the cursor offset */
857 BUG_ON(cursor->resid < bytes);
858 cursor->resid -= bytes;
859 cursor->vector_offset += bytes;
860 if (cursor->vector_offset < bio_vec->bv_len)
861 return false; /* more bytes to process in this segment */
862 BUG_ON(cursor->vector_offset != bio_vec->bv_len);
864 /* Move on to the next segment, and possibly the next bio */
866 if (++index == (unsigned int) bio->bi_vcnt) {
871 cursor->vector_index = index;
872 cursor->vector_offset = 0;
874 if (!cursor->last_piece) {
875 BUG_ON(!cursor->resid);
877 /* A short read is OK, so use <= rather than == */
878 if (cursor->resid <= bio->bi_io_vec[index].bv_len)
879 cursor->last_piece = true;
884 #endif /* CONFIG_BLOCK */
887 * For a page array, a piece comes from the first page in the array
888 * that has not already been fully consumed.
890 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
893 struct ceph_msg_data *data = cursor->data;
896 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
898 BUG_ON(!data->pages);
899 BUG_ON(!data->length);
901 cursor->resid = min(length, data->length);
902 page_count = calc_pages_for(data->alignment, (u64)data->length);
903 cursor->page_offset = data->alignment & ~PAGE_MASK;
904 cursor->page_index = 0;
905 BUG_ON(page_count > (int)USHRT_MAX);
906 cursor->page_count = (unsigned short)page_count;
907 BUG_ON(length > SIZE_MAX - cursor->page_offset);
908 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
912 ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
913 size_t *page_offset, size_t *length)
915 struct ceph_msg_data *data = cursor->data;
917 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
919 BUG_ON(cursor->page_index >= cursor->page_count);
920 BUG_ON(cursor->page_offset >= PAGE_SIZE);
922 *page_offset = cursor->page_offset;
923 if (cursor->last_piece)
924 *length = cursor->resid;
926 *length = PAGE_SIZE - *page_offset;
928 return data->pages[cursor->page_index];
931 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
934 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
936 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
938 /* Advance the cursor page offset */
940 cursor->resid -= bytes;
941 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
942 if (!bytes || cursor->page_offset)
943 return false; /* more bytes to process in the current page */
945 /* Move on to the next page; offset is already at 0 */
947 BUG_ON(cursor->page_index >= cursor->page_count);
948 cursor->page_index++;
949 cursor->last_piece = cursor->resid <= PAGE_SIZE;
955 * For a pagelist, a piece is whatever remains to be consumed in the
956 * first page in the list, or the front of the next page.
959 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
962 struct ceph_msg_data *data = cursor->data;
963 struct ceph_pagelist *pagelist;
966 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
968 pagelist = data->pagelist;
972 return; /* pagelist can be assigned but empty */
974 BUG_ON(list_empty(&pagelist->head));
975 page = list_first_entry(&pagelist->head, struct page, lru);
977 cursor->resid = min(length, pagelist->length);
980 cursor->last_piece = cursor->resid <= PAGE_SIZE;
984 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
985 size_t *page_offset, size_t *length)
987 struct ceph_msg_data *data = cursor->data;
988 struct ceph_pagelist *pagelist;
990 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
992 pagelist = data->pagelist;
995 BUG_ON(!cursor->page);
996 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
998 /* offset of first page in pagelist is always 0 */
999 *page_offset = cursor->offset & ~PAGE_MASK;
1000 if (cursor->last_piece)
1001 *length = cursor->resid;
1003 *length = PAGE_SIZE - *page_offset;
1005 return cursor->page;
1008 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
1011 struct ceph_msg_data *data = cursor->data;
1012 struct ceph_pagelist *pagelist;
1014 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1016 pagelist = data->pagelist;
1019 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1020 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1022 /* Advance the cursor offset */
1024 cursor->resid -= bytes;
1025 cursor->offset += bytes;
1026 /* offset of first page in pagelist is always 0 */
1027 if (!bytes || cursor->offset & ~PAGE_MASK)
1028 return false; /* more bytes to process in the current page */
1030 /* Move on to the next page */
1032 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1033 cursor->page = list_entry_next(cursor->page, lru);
1034 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1040 * Message data is handled (sent or received) in pieces, where each
1041 * piece resides on a single page. The network layer might not
1042 * consume an entire piece at once. A data item's cursor keeps
1043 * track of which piece is next to process and how much remains to
1044 * be processed in that piece. It also tracks whether the current
1045 * piece is the last one in the data item.
1047 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1049 size_t length = cursor->total_resid;
1051 switch (cursor->data->type) {
1052 case CEPH_MSG_DATA_PAGELIST:
1053 ceph_msg_data_pagelist_cursor_init(cursor, length);
1055 case CEPH_MSG_DATA_PAGES:
1056 ceph_msg_data_pages_cursor_init(cursor, length);
1059 case CEPH_MSG_DATA_BIO:
1060 ceph_msg_data_bio_cursor_init(cursor, length);
1062 #endif /* CONFIG_BLOCK */
1063 case CEPH_MSG_DATA_NONE:
1068 cursor->need_crc = true;
1071 static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1073 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1074 struct ceph_msg_data *data;
1077 BUG_ON(length > msg->data_length);
1078 BUG_ON(list_empty(&msg->data));
1080 cursor->data_head = &msg->data;
1081 cursor->total_resid = length;
1082 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1083 cursor->data = data;
1085 __ceph_msg_data_cursor_init(cursor);
1089 * Return the page containing the next piece to process for a given
1090 * data item, and supply the page offset and length of that piece.
1091 * Indicate whether this is the last piece in this data item.
1093 static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1094 size_t *page_offset, size_t *length,
1099 switch (cursor->data->type) {
1100 case CEPH_MSG_DATA_PAGELIST:
1101 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1103 case CEPH_MSG_DATA_PAGES:
1104 page = ceph_msg_data_pages_next(cursor, page_offset, length);
1107 case CEPH_MSG_DATA_BIO:
1108 page = ceph_msg_data_bio_next(cursor, page_offset, length);
1110 #endif /* CONFIG_BLOCK */
1111 case CEPH_MSG_DATA_NONE:
1117 BUG_ON(*page_offset + *length > PAGE_SIZE);
1120 *last_piece = cursor->last_piece;
1126 * Returns true if the result moves the cursor on to the next piece
1129 static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1134 BUG_ON(bytes > cursor->resid);
1135 switch (cursor->data->type) {
1136 case CEPH_MSG_DATA_PAGELIST:
1137 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1139 case CEPH_MSG_DATA_PAGES:
1140 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1143 case CEPH_MSG_DATA_BIO:
1144 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1146 #endif /* CONFIG_BLOCK */
1147 case CEPH_MSG_DATA_NONE:
1152 cursor->total_resid -= bytes;
1154 if (!cursor->resid && cursor->total_resid) {
1155 WARN_ON(!cursor->last_piece);
1156 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1157 cursor->data = list_entry_next(cursor->data, links);
1158 __ceph_msg_data_cursor_init(cursor);
1161 cursor->need_crc = new_piece;
1166 static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
1171 /* Initialize data cursor */
1173 ceph_msg_data_cursor_init(msg, (size_t)data_len);
1177 * Prepare footer for currently outgoing message, and finish things
1178 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1180 static void prepare_write_message_footer(struct ceph_connection *con)
1182 struct ceph_msg *m = con->out_msg;
1183 int v = con->out_kvec_left;
1185 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1187 dout("prepare_write_message_footer %p\n", con);
1188 con->out_kvec_is_msg = true;
1189 con->out_kvec[v].iov_base = &m->footer;
1190 con->out_kvec[v].iov_len = sizeof(m->footer);
1191 con->out_kvec_bytes += sizeof(m->footer);
1192 con->out_kvec_left++;
1193 con->out_more = m->more_to_follow;
1194 con->out_msg_done = true;
1198 * Prepare headers for the next outgoing message.
1200 static void prepare_write_message(struct ceph_connection *con)
1205 con_out_kvec_reset(con);
1206 con->out_kvec_is_msg = true;
1207 con->out_msg_done = false;
1209 /* Sneak an ack in there first? If we can get it into the same
1210 * TCP packet that's a good thing. */
1211 if (con->in_seq > con->in_seq_acked) {
1212 con->in_seq_acked = con->in_seq;
1213 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1214 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1215 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1216 &con->out_temp_ack);
1219 BUG_ON(list_empty(&con->out_queue));
1220 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1222 BUG_ON(m->con != con);
1224 /* put message on sent list */
1226 list_move_tail(&m->list_head, &con->out_sent);
1229 * only assign outgoing seq # if we haven't sent this message
1230 * yet. if it is requeued, resend with it's original seq.
1232 if (m->needs_out_seq) {
1233 m->hdr.seq = cpu_to_le64(++con->out_seq);
1234 m->needs_out_seq = false;
1236 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
1238 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1239 m, con->out_seq, le16_to_cpu(m->hdr.type),
1240 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1242 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1244 /* tag + hdr + front + middle */
1245 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1246 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1247 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1250 con_out_kvec_add(con, m->middle->vec.iov_len,
1251 m->middle->vec.iov_base);
1253 /* fill in crc (except data pages), footer */
1254 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1255 con->out_msg->hdr.crc = cpu_to_le32(crc);
1256 con->out_msg->footer.flags = 0;
1258 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1259 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1261 crc = crc32c(0, m->middle->vec.iov_base,
1262 m->middle->vec.iov_len);
1263 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1265 con->out_msg->footer.middle_crc = 0;
1266 dout("%s front_crc %u middle_crc %u\n", __func__,
1267 le32_to_cpu(con->out_msg->footer.front_crc),
1268 le32_to_cpu(con->out_msg->footer.middle_crc));
1270 /* is there a data payload? */
1271 con->out_msg->footer.data_crc = 0;
1272 if (m->data_length) {
1273 prepare_message_data(con->out_msg, m->data_length);
1274 con->out_more = 1; /* data + footer will follow */
1276 /* no, queue up footer too and be done */
1277 prepare_write_message_footer(con);
1280 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1286 static void prepare_write_ack(struct ceph_connection *con)
1288 dout("prepare_write_ack %p %llu -> %llu\n", con,
1289 con->in_seq_acked, con->in_seq);
1290 con->in_seq_acked = con->in_seq;
1292 con_out_kvec_reset(con);
1294 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1296 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1297 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1298 &con->out_temp_ack);
1300 con->out_more = 1; /* more will follow.. eventually.. */
1301 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1305 * Prepare to share the seq during handshake
1307 static void prepare_write_seq(struct ceph_connection *con)
1309 dout("prepare_write_seq %p %llu -> %llu\n", con,
1310 con->in_seq_acked, con->in_seq);
1311 con->in_seq_acked = con->in_seq;
1313 con_out_kvec_reset(con);
1315 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1316 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1317 &con->out_temp_ack);
1319 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1323 * Prepare to write keepalive byte.
1325 static void prepare_write_keepalive(struct ceph_connection *con)
1327 dout("prepare_write_keepalive %p\n", con);
1328 con_out_kvec_reset(con);
1329 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
1330 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1334 * Connection negotiation.
1337 static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1340 struct ceph_auth_handshake *auth;
1342 if (!con->ops->get_authorizer) {
1343 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1344 con->out_connect.authorizer_len = 0;
1348 /* Can't hold the mutex while getting authorizer */
1349 mutex_unlock(&con->mutex);
1350 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
1351 mutex_lock(&con->mutex);
1355 if (con->state != CON_STATE_NEGOTIATING)
1356 return ERR_PTR(-EAGAIN);
1358 con->auth_reply_buf = auth->authorizer_reply_buf;
1359 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
1364 * We connected to a peer and are saying hello.
1366 static void prepare_write_banner(struct ceph_connection *con)
1368 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1369 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1370 &con->msgr->my_enc_addr);
1373 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1376 static int prepare_write_connect(struct ceph_connection *con)
1378 unsigned int global_seq = get_global_seq(con->msgr, 0);
1381 struct ceph_auth_handshake *auth;
1383 switch (con->peer_name.type) {
1384 case CEPH_ENTITY_TYPE_MON:
1385 proto = CEPH_MONC_PROTOCOL;
1387 case CEPH_ENTITY_TYPE_OSD:
1388 proto = CEPH_OSDC_PROTOCOL;
1390 case CEPH_ENTITY_TYPE_MDS:
1391 proto = CEPH_MDSC_PROTOCOL;
1397 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1398 con->connect_seq, global_seq, proto);
1400 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
1401 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1402 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1403 con->out_connect.global_seq = cpu_to_le32(global_seq);
1404 con->out_connect.protocol_version = cpu_to_le32(proto);
1405 con->out_connect.flags = 0;
1407 auth_proto = CEPH_AUTH_UNKNOWN;
1408 auth = get_connect_authorizer(con, &auth_proto);
1410 return PTR_ERR(auth);
1412 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1413 con->out_connect.authorizer_len = auth ?
1414 cpu_to_le32(auth->authorizer_buf_len) : 0;
1416 con_out_kvec_add(con, sizeof (con->out_connect),
1418 if (auth && auth->authorizer_buf_len)
1419 con_out_kvec_add(con, auth->authorizer_buf_len,
1420 auth->authorizer_buf);
1423 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1429 * write as much of pending kvecs to the socket as we can.
1431 * 0 -> socket full, but more to do
1434 static int write_partial_kvec(struct ceph_connection *con)
1438 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1439 while (con->out_kvec_bytes > 0) {
1440 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1441 con->out_kvec_left, con->out_kvec_bytes,
1445 con->out_kvec_bytes -= ret;
1446 if (con->out_kvec_bytes == 0)
1449 /* account for full iov entries consumed */
1450 while (ret >= con->out_kvec_cur->iov_len) {
1451 BUG_ON(!con->out_kvec_left);
1452 ret -= con->out_kvec_cur->iov_len;
1453 con->out_kvec_cur++;
1454 con->out_kvec_left--;
1456 /* and for a partially-consumed entry */
1458 con->out_kvec_cur->iov_len -= ret;
1459 con->out_kvec_cur->iov_base += ret;
1462 con->out_kvec_left = 0;
1463 con->out_kvec_is_msg = false;
1466 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1467 con->out_kvec_bytes, con->out_kvec_left, ret);
1468 return ret; /* done! */
1471 static u32 ceph_crc32c_page(u32 crc, struct page *page,
1472 unsigned int page_offset,
1473 unsigned int length)
1478 BUG_ON(kaddr == NULL);
1479 crc = crc32c(crc, kaddr + page_offset, length);
1485 * Write as much message data payload as we can. If we finish, queue
1487 * 1 -> done, footer is now queued in out_kvec[].
1488 * 0 -> socket full, but more to do
1491 static int write_partial_message_data(struct ceph_connection *con)
1493 struct ceph_msg *msg = con->out_msg;
1494 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1495 bool do_datacrc = !con->msgr->nocrc;
1498 dout("%s %p msg %p\n", __func__, con, msg);
1500 if (list_empty(&msg->data))
1504 * Iterate through each page that contains data to be
1505 * written, and send as much as possible for each.
1507 * If we are calculating the data crc (the default), we will
1508 * need to map the page. If we have no pages, they have
1509 * been revoked, so use the zero page.
1511 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1512 while (cursor->resid) {
1520 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
1522 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
1523 length, last_piece);
1526 msg->footer.data_crc = cpu_to_le32(crc);
1530 if (do_datacrc && cursor->need_crc)
1531 crc = ceph_crc32c_page(crc, page, page_offset, length);
1532 need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret);
1535 dout("%s %p msg %p done\n", __func__, con, msg);
1537 /* prepare and queue up footer, too */
1539 msg->footer.data_crc = cpu_to_le32(crc);
1541 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1542 con_out_kvec_reset(con);
1543 prepare_write_message_footer(con);
1545 return 1; /* must return > 0 to indicate success */
1551 static int write_partial_skip(struct ceph_connection *con)
1555 while (con->out_skip > 0) {
1556 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
1558 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
1561 con->out_skip -= ret;
1569 * Prepare to read connection handshake, or an ack.
1571 static void prepare_read_banner(struct ceph_connection *con)
1573 dout("prepare_read_banner %p\n", con);
1574 con->in_base_pos = 0;
1577 static void prepare_read_connect(struct ceph_connection *con)
1579 dout("prepare_read_connect %p\n", con);
1580 con->in_base_pos = 0;
1583 static void prepare_read_ack(struct ceph_connection *con)
1585 dout("prepare_read_ack %p\n", con);
1586 con->in_base_pos = 0;
1589 static void prepare_read_seq(struct ceph_connection *con)
1591 dout("prepare_read_seq %p\n", con);
1592 con->in_base_pos = 0;
1593 con->in_tag = CEPH_MSGR_TAG_SEQ;
1596 static void prepare_read_tag(struct ceph_connection *con)
1598 dout("prepare_read_tag %p\n", con);
1599 con->in_base_pos = 0;
1600 con->in_tag = CEPH_MSGR_TAG_READY;
1604 * Prepare to read a message.
1606 static int prepare_read_message(struct ceph_connection *con)
1608 dout("prepare_read_message %p\n", con);
1609 BUG_ON(con->in_msg != NULL);
1610 con->in_base_pos = 0;
1611 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1616 static int read_partial(struct ceph_connection *con,
1617 int end, int size, void *object)
1619 while (con->in_base_pos < end) {
1620 int left = end - con->in_base_pos;
1621 int have = size - left;
1622 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1625 con->in_base_pos += ret;
1632 * Read all or part of the connect-side handshake on a new connection
1634 static int read_partial_banner(struct ceph_connection *con)
1640 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1643 size = strlen(CEPH_BANNER);
1645 ret = read_partial(con, end, size, con->in_banner);
1649 size = sizeof (con->actual_peer_addr);
1651 ret = read_partial(con, end, size, &con->actual_peer_addr);
1655 size = sizeof (con->peer_addr_for_me);
1657 ret = read_partial(con, end, size, &con->peer_addr_for_me);
1665 static int read_partial_connect(struct ceph_connection *con)
1671 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1673 size = sizeof (con->in_reply);
1675 ret = read_partial(con, end, size, &con->in_reply);
1679 size = le32_to_cpu(con->in_reply.authorizer_len);
1681 ret = read_partial(con, end, size, con->auth_reply_buf);
1685 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1686 con, (int)con->in_reply.tag,
1687 le32_to_cpu(con->in_reply.connect_seq),
1688 le32_to_cpu(con->in_reply.global_seq));
1695 * Verify the hello banner looks okay.
1697 static int verify_hello(struct ceph_connection *con)
1699 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1700 pr_err("connect to %s got bad banner\n",
1701 ceph_pr_addr(&con->peer_addr.in_addr));
1702 con->error_msg = "protocol error, bad banner";
1708 static bool addr_is_blank(struct sockaddr_storage *ss)
1710 switch (ss->ss_family) {
1712 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1715 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1716 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1717 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1718 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1723 static int addr_port(struct sockaddr_storage *ss)
1725 switch (ss->ss_family) {
1727 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1729 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1734 static void addr_set_port(struct sockaddr_storage *ss, int p)
1736 switch (ss->ss_family) {
1738 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1741 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1747 * Unlike other *_pton function semantics, zero indicates success.
1749 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1750 char delim, const char **ipend)
1752 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1753 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1755 memset(ss, 0, sizeof(*ss));
1757 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1758 ss->ss_family = AF_INET;
1762 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1763 ss->ss_family = AF_INET6;
1771 * Extract hostname string and resolve using kernel DNS facility.
1773 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1774 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1775 struct sockaddr_storage *ss, char delim, const char **ipend)
1777 const char *end, *delim_p;
1778 char *colon_p, *ip_addr = NULL;
1782 * The end of the hostname occurs immediately preceding the delimiter or
1783 * the port marker (':') where the delimiter takes precedence.
1785 delim_p = memchr(name, delim, namelen);
1786 colon_p = memchr(name, ':', namelen);
1788 if (delim_p && colon_p)
1789 end = delim_p < colon_p ? delim_p : colon_p;
1790 else if (!delim_p && colon_p)
1794 if (!end) /* case: hostname:/ */
1795 end = name + namelen;
1801 /* do dns_resolve upcall */
1802 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1804 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1812 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1813 ret, ret ? "failed" : ceph_pr_addr(ss));
1818 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1819 struct sockaddr_storage *ss, char delim, const char **ipend)
1826 * Parse a server name (IP or hostname). If a valid IP address is not found
1827 * then try to extract a hostname to resolve using userspace DNS upcall.
1829 static int ceph_parse_server_name(const char *name, size_t namelen,
1830 struct sockaddr_storage *ss, char delim, const char **ipend)
1834 ret = ceph_pton(name, namelen, ss, delim, ipend);
1836 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1842 * Parse an ip[:port] list into an addr array. Use the default
1843 * monitor port if a port isn't specified.
1845 int ceph_parse_ips(const char *c, const char *end,
1846 struct ceph_entity_addr *addr,
1847 int max_count, int *count)
1849 int i, ret = -EINVAL;
1852 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1853 for (i = 0; i < max_count; i++) {
1855 struct sockaddr_storage *ss = &addr[i].in_addr;
1864 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1873 dout("missing matching ']'\n");
1880 if (p < end && *p == ':') {
1883 while (p < end && *p >= '0' && *p <= '9') {
1884 port = (port * 10) + (*p - '0');
1887 if (port > 65535 || port == 0)
1890 port = CEPH_MON_PORT;
1893 addr_set_port(ss, port);
1895 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1912 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1915 EXPORT_SYMBOL(ceph_parse_ips);
1917 static int process_banner(struct ceph_connection *con)
1919 dout("process_banner on %p\n", con);
1921 if (verify_hello(con) < 0)
1924 ceph_decode_addr(&con->actual_peer_addr);
1925 ceph_decode_addr(&con->peer_addr_for_me);
1928 * Make sure the other end is who we wanted. note that the other
1929 * end may not yet know their ip address, so if it's 0.0.0.0, give
1930 * them the benefit of the doubt.
1932 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1933 sizeof(con->peer_addr)) != 0 &&
1934 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1935 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1936 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1937 ceph_pr_addr(&con->peer_addr.in_addr),
1938 (int)le32_to_cpu(con->peer_addr.nonce),
1939 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1940 (int)le32_to_cpu(con->actual_peer_addr.nonce));
1941 con->error_msg = "wrong peer at address";
1946 * did we learn our address?
1948 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1949 int port = addr_port(&con->msgr->inst.addr.in_addr);
1951 memcpy(&con->msgr->inst.addr.in_addr,
1952 &con->peer_addr_for_me.in_addr,
1953 sizeof(con->peer_addr_for_me.in_addr));
1954 addr_set_port(&con->msgr->inst.addr.in_addr, port);
1955 encode_my_addr(con->msgr);
1956 dout("process_banner learned my addr is %s\n",
1957 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
1963 static int process_connect(struct ceph_connection *con)
1965 u64 sup_feat = con->msgr->supported_features;
1966 u64 req_feat = con->msgr->required_features;
1967 u64 server_feat = le64_to_cpu(con->in_reply.features);
1970 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1972 switch (con->in_reply.tag) {
1973 case CEPH_MSGR_TAG_FEATURES:
1974 pr_err("%s%lld %s feature set mismatch,"
1975 " my %llx < server's %llx, missing %llx\n",
1976 ENTITY_NAME(con->peer_name),
1977 ceph_pr_addr(&con->peer_addr.in_addr),
1978 sup_feat, server_feat, server_feat & ~sup_feat);
1979 con->error_msg = "missing required protocol features";
1980 reset_connection(con);
1983 case CEPH_MSGR_TAG_BADPROTOVER:
1984 pr_err("%s%lld %s protocol version mismatch,"
1985 " my %d != server's %d\n",
1986 ENTITY_NAME(con->peer_name),
1987 ceph_pr_addr(&con->peer_addr.in_addr),
1988 le32_to_cpu(con->out_connect.protocol_version),
1989 le32_to_cpu(con->in_reply.protocol_version));
1990 con->error_msg = "protocol version mismatch";
1991 reset_connection(con);
1994 case CEPH_MSGR_TAG_BADAUTHORIZER:
1996 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1998 if (con->auth_retry == 2) {
1999 con->error_msg = "connect authorization failure";
2002 con_out_kvec_reset(con);
2003 ret = prepare_write_connect(con);
2006 prepare_read_connect(con);
2009 case CEPH_MSGR_TAG_RESETSESSION:
2011 * If we connected with a large connect_seq but the peer
2012 * has no record of a session with us (no connection, or
2013 * connect_seq == 0), they will send RESETSESION to indicate
2014 * that they must have reset their session, and may have
2017 dout("process_connect got RESET peer seq %u\n",
2018 le32_to_cpu(con->in_reply.connect_seq));
2019 pr_err("%s%lld %s connection reset\n",
2020 ENTITY_NAME(con->peer_name),
2021 ceph_pr_addr(&con->peer_addr.in_addr));
2022 reset_connection(con);
2023 con_out_kvec_reset(con);
2024 ret = prepare_write_connect(con);
2027 prepare_read_connect(con);
2029 /* Tell ceph about it. */
2030 mutex_unlock(&con->mutex);
2031 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2032 if (con->ops->peer_reset)
2033 con->ops->peer_reset(con);
2034 mutex_lock(&con->mutex);
2035 if (con->state != CON_STATE_NEGOTIATING)
2039 case CEPH_MSGR_TAG_RETRY_SESSION:
2041 * If we sent a smaller connect_seq than the peer has, try
2042 * again with a larger value.
2044 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2045 le32_to_cpu(con->out_connect.connect_seq),
2046 le32_to_cpu(con->in_reply.connect_seq));
2047 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2048 con_out_kvec_reset(con);
2049 ret = prepare_write_connect(con);
2052 prepare_read_connect(con);
2055 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2057 * If we sent a smaller global_seq than the peer has, try
2058 * again with a larger value.
2060 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2061 con->peer_global_seq,
2062 le32_to_cpu(con->in_reply.global_seq));
2063 get_global_seq(con->msgr,
2064 le32_to_cpu(con->in_reply.global_seq));
2065 con_out_kvec_reset(con);
2066 ret = prepare_write_connect(con);
2069 prepare_read_connect(con);
2072 case CEPH_MSGR_TAG_SEQ:
2073 case CEPH_MSGR_TAG_READY:
2074 if (req_feat & ~server_feat) {
2075 pr_err("%s%lld %s protocol feature mismatch,"
2076 " my required %llx > server's %llx, need %llx\n",
2077 ENTITY_NAME(con->peer_name),
2078 ceph_pr_addr(&con->peer_addr.in_addr),
2079 req_feat, server_feat, req_feat & ~server_feat);
2080 con->error_msg = "missing required protocol features";
2081 reset_connection(con);
2085 WARN_ON(con->state != CON_STATE_NEGOTIATING);
2086 con->state = CON_STATE_OPEN;
2087 con->auth_retry = 0; /* we authenticated; clear flag */
2088 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2090 con->peer_features = server_feat;
2091 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2092 con->peer_global_seq,
2093 le32_to_cpu(con->in_reply.connect_seq),
2095 WARN_ON(con->connect_seq !=
2096 le32_to_cpu(con->in_reply.connect_seq));
2098 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2099 con_flag_set(con, CON_FLAG_LOSSYTX);
2101 con->delay = 0; /* reset backoff memory */
2103 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2104 prepare_write_seq(con);
2105 prepare_read_seq(con);
2107 prepare_read_tag(con);
2111 case CEPH_MSGR_TAG_WAIT:
2113 * If there is a connection race (we are opening
2114 * connections to each other), one of us may just have
2115 * to WAIT. This shouldn't happen if we are the
2118 pr_err("process_connect got WAIT as client\n");
2119 con->error_msg = "protocol error, got WAIT as client";
2123 pr_err("connect protocol error, will retry\n");
2124 con->error_msg = "protocol error, garbage tag during connect";
2132 * read (part of) an ack
2134 static int read_partial_ack(struct ceph_connection *con)
2136 int size = sizeof (con->in_temp_ack);
2139 return read_partial(con, end, size, &con->in_temp_ack);
2143 * We can finally discard anything that's been acked.
2145 static void process_ack(struct ceph_connection *con)
2148 u64 ack = le64_to_cpu(con->in_temp_ack);
2151 while (!list_empty(&con->out_sent)) {
2152 m = list_first_entry(&con->out_sent, struct ceph_msg,
2154 seq = le64_to_cpu(m->hdr.seq);
2157 dout("got ack for seq %llu type %d at %p\n", seq,
2158 le16_to_cpu(m->hdr.type), m);
2159 m->ack_stamp = jiffies;
2162 prepare_read_tag(con);
2166 static int read_partial_message_section(struct ceph_connection *con,
2167 struct kvec *section,
2168 unsigned int sec_len, u32 *crc)
2174 while (section->iov_len < sec_len) {
2175 BUG_ON(section->iov_base == NULL);
2176 left = sec_len - section->iov_len;
2177 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2178 section->iov_len, left);
2181 section->iov_len += ret;
2183 if (section->iov_len == sec_len)
2184 *crc = crc32c(0, section->iov_base, section->iov_len);
2189 static int read_partial_msg_data(struct ceph_connection *con)
2191 struct ceph_msg *msg = con->in_msg;
2192 struct ceph_msg_data_cursor *cursor = &msg->cursor;
2193 const bool do_datacrc = !con->msgr->nocrc;
2201 if (list_empty(&msg->data))
2205 crc = con->in_data_crc;
2206 while (cursor->resid) {
2207 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
2209 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2212 con->in_data_crc = crc;
2218 crc = ceph_crc32c_page(crc, page, page_offset, ret);
2219 (void) ceph_msg_data_advance(&msg->cursor, (size_t)ret);
2222 con->in_data_crc = crc;
2224 return 1; /* must return > 0 to indicate success */
2228 * read (part of) a message.
2230 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2232 static int read_partial_message(struct ceph_connection *con)
2234 struct ceph_msg *m = con->in_msg;
2238 unsigned int front_len, middle_len, data_len;
2239 bool do_datacrc = !con->msgr->nocrc;
2243 dout("read_partial_message con %p msg %p\n", con, m);
2246 size = sizeof (con->in_hdr);
2248 ret = read_partial(con, end, size, &con->in_hdr);
2252 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2253 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2254 pr_err("read_partial_message bad hdr "
2255 " crc %u != expected %u\n",
2256 crc, con->in_hdr.crc);
2260 front_len = le32_to_cpu(con->in_hdr.front_len);
2261 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2263 middle_len = le32_to_cpu(con->in_hdr.middle_len);
2264 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2266 data_len = le32_to_cpu(con->in_hdr.data_len);
2267 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2271 seq = le64_to_cpu(con->in_hdr.seq);
2272 if ((s64)seq - (s64)con->in_seq < 1) {
2273 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2274 ENTITY_NAME(con->peer_name),
2275 ceph_pr_addr(&con->peer_addr.in_addr),
2276 seq, con->in_seq + 1);
2277 con->in_base_pos = -front_len - middle_len - data_len -
2279 con->in_tag = CEPH_MSGR_TAG_READY;
2281 } else if ((s64)seq - (s64)con->in_seq > 1) {
2282 pr_err("read_partial_message bad seq %lld expected %lld\n",
2283 seq, con->in_seq + 1);
2284 con->error_msg = "bad message sequence # for incoming message";
2288 /* allocate message? */
2292 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2293 front_len, data_len);
2294 ret = ceph_con_in_msg_alloc(con, &skip);
2298 BUG_ON(!con->in_msg ^ skip);
2299 if (con->in_msg && data_len > con->in_msg->data_length) {
2300 pr_warning("%s skipping long message (%u > %zd)\n",
2301 __func__, data_len, con->in_msg->data_length);
2302 ceph_msg_put(con->in_msg);
2307 /* skip this message */
2308 dout("alloc_msg said skip message\n");
2309 con->in_base_pos = -front_len - middle_len - data_len -
2311 con->in_tag = CEPH_MSGR_TAG_READY;
2316 BUG_ON(!con->in_msg);
2317 BUG_ON(con->in_msg->con != con);
2319 m->front.iov_len = 0; /* haven't read it yet */
2321 m->middle->vec.iov_len = 0;
2323 /* prepare for data payload, if any */
2326 prepare_message_data(con->in_msg, data_len);
2330 ret = read_partial_message_section(con, &m->front, front_len,
2331 &con->in_front_crc);
2337 ret = read_partial_message_section(con, &m->middle->vec,
2339 &con->in_middle_crc);
2346 ret = read_partial_msg_data(con);
2352 size = sizeof (m->footer);
2354 ret = read_partial(con, end, size, &m->footer);
2358 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2359 m, front_len, m->footer.front_crc, middle_len,
2360 m->footer.middle_crc, data_len, m->footer.data_crc);
2363 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2364 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2365 m, con->in_front_crc, m->footer.front_crc);
2368 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2369 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2370 m, con->in_middle_crc, m->footer.middle_crc);
2374 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2375 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2376 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2377 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2381 return 1; /* done! */
2385 * Process message. This happens in the worker thread. The callback should
2386 * be careful not to do anything that waits on other incoming messages or it
2389 static void process_message(struct ceph_connection *con)
2391 struct ceph_msg *msg;
2393 BUG_ON(con->in_msg->con != con);
2394 con->in_msg->con = NULL;
2399 /* if first message, set peer_name */
2400 if (con->peer_name.type == 0)
2401 con->peer_name = msg->hdr.src;
2404 mutex_unlock(&con->mutex);
2406 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2407 msg, le64_to_cpu(msg->hdr.seq),
2408 ENTITY_NAME(msg->hdr.src),
2409 le16_to_cpu(msg->hdr.type),
2410 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2411 le32_to_cpu(msg->hdr.front_len),
2412 le32_to_cpu(msg->hdr.data_len),
2413 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2414 con->ops->dispatch(con, msg);
2416 mutex_lock(&con->mutex);
2421 * Write something to the socket. Called in a worker thread when the
2422 * socket appears to be writeable and we have something ready to send.
2424 static int try_write(struct ceph_connection *con)
2428 dout("try_write start %p state %lu\n", con, con->state);
2431 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2433 /* open the socket first? */
2434 if (con->state == CON_STATE_PREOPEN) {
2436 con->state = CON_STATE_CONNECTING;
2438 con_out_kvec_reset(con);
2439 prepare_write_banner(con);
2440 prepare_read_banner(con);
2442 BUG_ON(con->in_msg);
2443 con->in_tag = CEPH_MSGR_TAG_READY;
2444 dout("try_write initiating connect on %p new state %lu\n",
2446 ret = ceph_tcp_connect(con);
2448 con->error_msg = "connect error";
2454 /* kvec data queued? */
2455 if (con->out_skip) {
2456 ret = write_partial_skip(con);
2460 if (con->out_kvec_left) {
2461 ret = write_partial_kvec(con);
2468 if (con->out_msg_done) {
2469 ceph_msg_put(con->out_msg);
2470 con->out_msg = NULL; /* we're done with this one */
2474 ret = write_partial_message_data(con);
2476 goto more_kvec; /* we need to send the footer, too! */
2480 dout("try_write write_partial_message_data err %d\n",
2487 if (con->state == CON_STATE_OPEN) {
2488 /* is anything else pending? */
2489 if (!list_empty(&con->out_queue)) {
2490 prepare_write_message(con);
2493 if (con->in_seq > con->in_seq_acked) {
2494 prepare_write_ack(con);
2497 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2498 prepare_write_keepalive(con);
2503 /* Nothing to do! */
2504 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2505 dout("try_write nothing else to write.\n");
2508 dout("try_write done on %p ret %d\n", con, ret);
2515 * Read what we can from the socket.
2517 static int try_read(struct ceph_connection *con)
2522 dout("try_read start on %p state %lu\n", con, con->state);
2523 if (con->state != CON_STATE_CONNECTING &&
2524 con->state != CON_STATE_NEGOTIATING &&
2525 con->state != CON_STATE_OPEN)
2530 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2533 if (con->state == CON_STATE_CONNECTING) {
2534 dout("try_read connecting\n");
2535 ret = read_partial_banner(con);
2538 ret = process_banner(con);
2542 con->state = CON_STATE_NEGOTIATING;
2545 * Received banner is good, exchange connection info.
2546 * Do not reset out_kvec, as sending our banner raced
2547 * with receiving peer banner after connect completed.
2549 ret = prepare_write_connect(con);
2552 prepare_read_connect(con);
2554 /* Send connection info before awaiting response */
2558 if (con->state == CON_STATE_NEGOTIATING) {
2559 dout("try_read negotiating\n");
2560 ret = read_partial_connect(con);
2563 ret = process_connect(con);
2569 WARN_ON(con->state != CON_STATE_OPEN);
2571 if (con->in_base_pos < 0) {
2573 * skipping + discarding content.
2575 * FIXME: there must be a better way to do this!
2577 static char buf[SKIP_BUF_SIZE];
2578 int skip = min((int) sizeof (buf), -con->in_base_pos);
2580 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2581 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2584 con->in_base_pos += ret;
2585 if (con->in_base_pos)
2588 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2592 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2595 dout("try_read got tag %d\n", (int)con->in_tag);
2596 switch (con->in_tag) {
2597 case CEPH_MSGR_TAG_MSG:
2598 prepare_read_message(con);
2600 case CEPH_MSGR_TAG_ACK:
2601 prepare_read_ack(con);
2603 case CEPH_MSGR_TAG_CLOSE:
2604 con_close_socket(con);
2605 con->state = CON_STATE_CLOSED;
2611 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2612 ret = read_partial_message(con);
2616 con->error_msg = "bad crc";
2620 con->error_msg = "io error";
2625 if (con->in_tag == CEPH_MSGR_TAG_READY)
2627 process_message(con);
2628 if (con->state == CON_STATE_OPEN)
2629 prepare_read_tag(con);
2632 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2633 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2635 * the final handshake seq exchange is semantically
2636 * equivalent to an ACK
2638 ret = read_partial_ack(con);
2646 dout("try_read done on %p ret %d\n", con, ret);
2650 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2651 con->error_msg = "protocol error, garbage tag";
2658 * Atomically queue work on a connection after the specified delay.
2659 * Bump @con reference to avoid races with connection teardown.
2660 * Returns 0 if work was queued, or an error code otherwise.
2662 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2664 if (!con->ops->get(con)) {
2665 dout("%s %p ref count 0\n", __func__, con);
2670 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2671 dout("%s %p - already queued\n", __func__, con);
2677 dout("%s %p %lu\n", __func__, con, delay);
2682 static void queue_con(struct ceph_connection *con)
2684 (void) queue_con_delay(con, 0);
2687 static bool con_sock_closed(struct ceph_connection *con)
2689 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2693 case CON_STATE_ ## x: \
2694 con->error_msg = "socket closed (con state " #x ")"; \
2697 switch (con->state) {
2705 pr_warning("%s con %p unrecognized state %lu\n",
2706 __func__, con, con->state);
2707 con->error_msg = "unrecognized con state";
2716 static bool con_backoff(struct ceph_connection *con)
2720 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2723 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2725 dout("%s: con %p FAILED to back off %lu\n", __func__,
2727 BUG_ON(ret == -ENOENT);
2728 con_flag_set(con, CON_FLAG_BACKOFF);
2734 /* Finish fault handling; con->mutex must *not* be held here */
2736 static void con_fault_finish(struct ceph_connection *con)
2739 * in case we faulted due to authentication, invalidate our
2740 * current tickets so that we can get new ones.
2742 if (con->auth_retry && con->ops->invalidate_authorizer) {
2743 dout("calling invalidate_authorizer()\n");
2744 con->ops->invalidate_authorizer(con);
2747 if (con->ops->fault)
2748 con->ops->fault(con);
2752 * Do some work on a connection. Drop a connection ref when we're done.
2754 static void con_work(struct work_struct *work)
2756 struct ceph_connection *con = container_of(work, struct ceph_connection,
2760 mutex_lock(&con->mutex);
2764 if ((fault = con_sock_closed(con))) {
2765 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2768 if (con_backoff(con)) {
2769 dout("%s: con %p BACKOFF\n", __func__, con);
2772 if (con->state == CON_STATE_STANDBY) {
2773 dout("%s: con %p STANDBY\n", __func__, con);
2776 if (con->state == CON_STATE_CLOSED) {
2777 dout("%s: con %p CLOSED\n", __func__, con);
2781 if (con->state == CON_STATE_PREOPEN) {
2782 dout("%s: con %p PREOPEN\n", __func__, con);
2786 ret = try_read(con);
2790 con->error_msg = "socket error on read";
2795 ret = try_write(con);
2799 con->error_msg = "socket error on write";
2803 break; /* If we make it to here, we're done */
2807 mutex_unlock(&con->mutex);
2810 con_fault_finish(con);
2816 * Generic error/fault handler. A retry mechanism is used with
2817 * exponential backoff
2819 static void con_fault(struct ceph_connection *con)
2821 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2822 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2823 dout("fault %p state %lu to peer %s\n",
2824 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2826 WARN_ON(con->state != CON_STATE_CONNECTING &&
2827 con->state != CON_STATE_NEGOTIATING &&
2828 con->state != CON_STATE_OPEN);
2830 con_close_socket(con);
2832 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2833 dout("fault on LOSSYTX channel, marking CLOSED\n");
2834 con->state = CON_STATE_CLOSED;
2839 BUG_ON(con->in_msg->con != con);
2840 con->in_msg->con = NULL;
2841 ceph_msg_put(con->in_msg);
2846 /* Requeue anything that hasn't been acked */
2847 list_splice_init(&con->out_sent, &con->out_queue);
2849 /* If there are no messages queued or keepalive pending, place
2850 * the connection in a STANDBY state */
2851 if (list_empty(&con->out_queue) &&
2852 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
2853 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2854 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2855 con->state = CON_STATE_STANDBY;
2857 /* retry after a delay. */
2858 con->state = CON_STATE_PREOPEN;
2859 if (con->delay == 0)
2860 con->delay = BASE_DELAY_INTERVAL;
2861 else if (con->delay < MAX_DELAY_INTERVAL)
2863 con_flag_set(con, CON_FLAG_BACKOFF);
2871 * initialize a new messenger instance
2873 void ceph_messenger_init(struct ceph_messenger *msgr,
2874 struct ceph_entity_addr *myaddr,
2875 u32 supported_features,
2876 u32 required_features,
2879 msgr->supported_features = supported_features;
2880 msgr->required_features = required_features;
2882 spin_lock_init(&msgr->global_seq_lock);
2885 msgr->inst.addr = *myaddr;
2887 /* select a random nonce */
2888 msgr->inst.addr.type = 0;
2889 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2890 encode_my_addr(msgr);
2891 msgr->nocrc = nocrc;
2893 atomic_set(&msgr->stopping, 0);
2895 dout("%s %p\n", __func__, msgr);
2897 EXPORT_SYMBOL(ceph_messenger_init);
2899 static void clear_standby(struct ceph_connection *con)
2901 /* come back from STANDBY? */
2902 if (con->state == CON_STATE_STANDBY) {
2903 dout("clear_standby %p and ++connect_seq\n", con);
2904 con->state = CON_STATE_PREOPEN;
2906 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2907 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
2912 * Queue up an outgoing message on the given connection.
2914 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2917 msg->hdr.src = con->msgr->inst.name;
2918 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2919 msg->needs_out_seq = true;
2921 mutex_lock(&con->mutex);
2923 if (con->state == CON_STATE_CLOSED) {
2924 dout("con_send %p closed, dropping %p\n", con, msg);
2926 mutex_unlock(&con->mutex);
2930 BUG_ON(msg->con != NULL);
2931 msg->con = con->ops->get(con);
2932 BUG_ON(msg->con == NULL);
2934 BUG_ON(!list_empty(&msg->list_head));
2935 list_add_tail(&msg->list_head, &con->out_queue);
2936 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2937 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2938 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2939 le32_to_cpu(msg->hdr.front_len),
2940 le32_to_cpu(msg->hdr.middle_len),
2941 le32_to_cpu(msg->hdr.data_len));
2944 mutex_unlock(&con->mutex);
2946 /* if there wasn't anything waiting to send before, queue
2948 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
2951 EXPORT_SYMBOL(ceph_con_send);
2954 * Revoke a message that was previously queued for send
2956 void ceph_msg_revoke(struct ceph_msg *msg)
2958 struct ceph_connection *con = msg->con;
2961 return; /* Message not in our possession */
2963 mutex_lock(&con->mutex);
2964 if (!list_empty(&msg->list_head)) {
2965 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
2966 list_del_init(&msg->list_head);
2967 BUG_ON(msg->con == NULL);
2968 msg->con->ops->put(msg->con);
2974 if (con->out_msg == msg) {
2975 dout("%s %p msg %p - was sending\n", __func__, con, msg);
2976 con->out_msg = NULL;
2977 if (con->out_kvec_is_msg) {
2978 con->out_skip = con->out_kvec_bytes;
2979 con->out_kvec_is_msg = false;
2985 mutex_unlock(&con->mutex);
2989 * Revoke a message that we may be reading data into
2991 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
2993 struct ceph_connection *con;
2995 BUG_ON(msg == NULL);
2997 dout("%s msg %p null con\n", __func__, msg);
2999 return; /* Message not in our possession */
3003 mutex_lock(&con->mutex);
3004 if (con->in_msg == msg) {
3005 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3006 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3007 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
3009 /* skip rest of message */
3010 dout("%s %p msg %p revoked\n", __func__, con, msg);
3011 con->in_base_pos = con->in_base_pos -
3012 sizeof(struct ceph_msg_header) -
3016 sizeof(struct ceph_msg_footer);
3017 ceph_msg_put(con->in_msg);
3019 con->in_tag = CEPH_MSGR_TAG_READY;
3022 dout("%s %p in_msg %p msg %p no-op\n",
3023 __func__, con, con->in_msg, msg);
3025 mutex_unlock(&con->mutex);
3029 * Queue a keepalive byte to ensure the tcp connection is alive.
3031 void ceph_con_keepalive(struct ceph_connection *con)
3033 dout("con_keepalive %p\n", con);
3034 mutex_lock(&con->mutex);
3036 mutex_unlock(&con->mutex);
3037 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3038 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3041 EXPORT_SYMBOL(ceph_con_keepalive);
3043 static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
3045 struct ceph_msg_data *data;
3047 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3050 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
3053 INIT_LIST_HEAD(&data->links);
3058 static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3063 WARN_ON(!list_empty(&data->links));
3064 if (data->type == CEPH_MSG_DATA_PAGELIST) {
3065 ceph_pagelist_release(data->pagelist);
3066 kfree(data->pagelist);
3068 kmem_cache_free(ceph_msg_data_cache, data);
3071 void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
3072 size_t length, size_t alignment)
3074 struct ceph_msg_data *data;
3079 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3081 data->pages = pages;
3082 data->length = length;
3083 data->alignment = alignment & ~PAGE_MASK;
3085 list_add_tail(&data->links, &msg->data);
3086 msg->data_length += length;
3088 EXPORT_SYMBOL(ceph_msg_data_add_pages);
3090 void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
3091 struct ceph_pagelist *pagelist)
3093 struct ceph_msg_data *data;
3096 BUG_ON(!pagelist->length);
3098 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3100 data->pagelist = pagelist;
3102 list_add_tail(&data->links, &msg->data);
3103 msg->data_length += pagelist->length;
3105 EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
3108 void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
3111 struct ceph_msg_data *data;
3115 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3118 data->bio_length = length;
3120 list_add_tail(&data->links, &msg->data);
3121 msg->data_length += length;
3123 EXPORT_SYMBOL(ceph_msg_data_add_bio);
3124 #endif /* CONFIG_BLOCK */
3127 * construct a new message with given type, size
3128 * the new msg has a ref count of 1.
3130 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3135 m = kmem_cache_zalloc(ceph_msg_cache, flags);
3139 m->hdr.type = cpu_to_le16(type);
3140 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3141 m->hdr.front_len = cpu_to_le32(front_len);
3143 INIT_LIST_HEAD(&m->list_head);
3144 kref_init(&m->kref);
3145 INIT_LIST_HEAD(&m->data);
3148 m->front_alloc_len = front_len;
3150 if (front_len > PAGE_CACHE_SIZE) {
3151 m->front.iov_base = __vmalloc(front_len, flags,
3153 m->front_is_vmalloc = true;
3155 m->front.iov_base = kmalloc(front_len, flags);
3157 if (m->front.iov_base == NULL) {
3158 dout("ceph_msg_new can't allocate %d bytes\n",
3163 m->front.iov_base = NULL;
3165 m->front.iov_len = front_len;
3167 dout("ceph_msg_new %p front %d\n", m, front_len);
3174 pr_err("msg_new can't create type %d front %d\n", type,
3178 dout("msg_new can't create type %d front %d\n", type,
3183 EXPORT_SYMBOL(ceph_msg_new);
3186 * Allocate "middle" portion of a message, if it is needed and wasn't
3187 * allocated by alloc_msg. This allows us to read a small fixed-size
3188 * per-type header in the front and then gracefully fail (i.e.,
3189 * propagate the error to the caller based on info in the front) when
3190 * the middle is too large.
3192 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3194 int type = le16_to_cpu(msg->hdr.type);
3195 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3197 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3198 ceph_msg_type_name(type), middle_len);
3199 BUG_ON(!middle_len);
3200 BUG_ON(msg->middle);
3202 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3209 * Allocate a message for receiving an incoming message on a
3210 * connection, and save the result in con->in_msg. Uses the
3211 * connection's private alloc_msg op if available.
3213 * Returns 0 on success, or a negative error code.
3215 * On success, if we set *skip = 1:
3216 * - the next message should be skipped and ignored.
3217 * - con->in_msg == NULL
3218 * or if we set *skip = 0:
3219 * - con->in_msg is non-null.
3220 * On error (ENOMEM, EAGAIN, ...),
3221 * - con->in_msg == NULL
3223 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3225 struct ceph_msg_header *hdr = &con->in_hdr;
3226 int middle_len = le32_to_cpu(hdr->middle_len);
3227 struct ceph_msg *msg;
3230 BUG_ON(con->in_msg != NULL);
3231 BUG_ON(!con->ops->alloc_msg);
3233 mutex_unlock(&con->mutex);
3234 msg = con->ops->alloc_msg(con, hdr, skip);
3235 mutex_lock(&con->mutex);
3236 if (con->state != CON_STATE_OPEN) {
3244 con->in_msg->con = con->ops->get(con);
3245 BUG_ON(con->in_msg->con == NULL);
3248 * Null message pointer means either we should skip
3249 * this message or we couldn't allocate memory. The
3250 * former is not an error.
3254 con->error_msg = "error allocating memory for incoming message";
3258 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3260 if (middle_len && !con->in_msg->middle) {
3261 ret = ceph_alloc_middle(con, con->in_msg);
3263 ceph_msg_put(con->in_msg);
3273 * Free a generically kmalloc'd message.
3275 void ceph_msg_kfree(struct ceph_msg *m)
3277 dout("msg_kfree %p\n", m);
3278 if (m->front_is_vmalloc)
3279 vfree(m->front.iov_base);
3281 kfree(m->front.iov_base);
3282 kmem_cache_free(ceph_msg_cache, m);
3286 * Drop a msg ref. Destroy as needed.
3288 void ceph_msg_last_put(struct kref *kref)
3290 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3292 struct list_head *links;
3293 struct list_head *next;
3295 dout("ceph_msg_put last one on %p\n", m);
3296 WARN_ON(!list_empty(&m->list_head));
3298 /* drop middle, data, if any */
3300 ceph_buffer_put(m->middle);
3304 list_splice_init(&m->data, &data);
3305 list_for_each_safe(links, next, &data) {
3306 struct ceph_msg_data *data;
3308 data = list_entry(links, struct ceph_msg_data, links);
3309 list_del_init(links);
3310 ceph_msg_data_destroy(data);
3315 ceph_msgpool_put(m->pool, m);
3319 EXPORT_SYMBOL(ceph_msg_last_put);
3321 void ceph_msg_dump(struct ceph_msg *msg)
3323 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3324 msg->front_alloc_len, msg->data_length);
3325 print_hex_dump(KERN_DEBUG, "header: ",
3326 DUMP_PREFIX_OFFSET, 16, 1,
3327 &msg->hdr, sizeof(msg->hdr), true);
3328 print_hex_dump(KERN_DEBUG, " front: ",
3329 DUMP_PREFIX_OFFSET, 16, 1,
3330 msg->front.iov_base, msg->front.iov_len, true);
3332 print_hex_dump(KERN_DEBUG, "middle: ",
3333 DUMP_PREFIX_OFFSET, 16, 1,
3334 msg->middle->vec.iov_base,
3335 msg->middle->vec.iov_len, true);
3336 print_hex_dump(KERN_DEBUG, "footer: ",
3337 DUMP_PREFIX_OFFSET, 16, 1,
3338 &msg->footer, sizeof(msg->footer), true);
3340 EXPORT_SYMBOL(ceph_msg_dump);