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:
1008 * Return the page containing the next piece to process for a given
1009 * data item, and supply the page offset and length of that piece.
1010 * Indicate whether this is the last piece in this data item.
1012 static struct page *ceph_msg_data_next(struct ceph_msg_data *data,
1013 size_t *page_offset,
1019 switch (data->type) {
1020 case CEPH_MSG_DATA_PAGELIST:
1021 page = ceph_msg_data_pagelist_next(data, page_offset, length);
1023 case CEPH_MSG_DATA_PAGES:
1024 page = ceph_msg_data_pages_next(data, page_offset, length);
1027 case CEPH_MSG_DATA_BIO:
1028 page = ceph_msg_data_bio_next(data, page_offset, length);
1030 #endif /* CONFIG_BLOCK */
1031 case CEPH_MSG_DATA_NONE:
1037 BUG_ON(*page_offset + *length > PAGE_SIZE);
1040 *last_piece = data->cursor.last_piece;
1046 * Returns true if the result moves the cursor on to the next piece
1049 static bool ceph_msg_data_advance(struct ceph_msg_data *data, size_t bytes)
1051 struct ceph_msg_data_cursor *cursor = &data->cursor;
1054 BUG_ON(bytes > cursor->resid);
1055 switch (data->type) {
1056 case CEPH_MSG_DATA_PAGELIST:
1057 new_piece = ceph_msg_data_pagelist_advance(data, bytes);
1059 case CEPH_MSG_DATA_PAGES:
1060 new_piece = ceph_msg_data_pages_advance(data, bytes);
1063 case CEPH_MSG_DATA_BIO:
1064 new_piece = ceph_msg_data_bio_advance(data, bytes);
1066 #endif /* CONFIG_BLOCK */
1067 case CEPH_MSG_DATA_NONE:
1076 static void prepare_message_data(struct ceph_msg *msg,
1077 struct ceph_msg_pos *msg_pos)
1083 data_len = le32_to_cpu(msg->hdr.data_len);
1086 /* initialize page iterator */
1088 if (ceph_msg_has_pages(msg))
1089 msg_pos->page_pos = msg->p.alignment;
1091 msg_pos->page_pos = 0;
1092 msg_pos->data_pos = 0;
1094 /* Initialize data cursors */
1097 if (ceph_msg_has_bio(msg))
1098 ceph_msg_data_cursor_init(&msg->b, data_len);
1099 #endif /* CONFIG_BLOCK */
1100 if (ceph_msg_has_pages(msg))
1101 ceph_msg_data_cursor_init(&msg->p, data_len);
1102 if (ceph_msg_has_pagelist(msg))
1103 ceph_msg_data_cursor_init(&msg->l, data_len);
1105 msg_pos->did_page_crc = false;
1109 * Prepare footer for currently outgoing message, and finish things
1110 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1112 static void prepare_write_message_footer(struct ceph_connection *con)
1114 struct ceph_msg *m = con->out_msg;
1115 int v = con->out_kvec_left;
1117 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1119 dout("prepare_write_message_footer %p\n", con);
1120 con->out_kvec_is_msg = true;
1121 con->out_kvec[v].iov_base = &m->footer;
1122 con->out_kvec[v].iov_len = sizeof(m->footer);
1123 con->out_kvec_bytes += sizeof(m->footer);
1124 con->out_kvec_left++;
1125 con->out_more = m->more_to_follow;
1126 con->out_msg_done = true;
1130 * Prepare headers for the next outgoing message.
1132 static void prepare_write_message(struct ceph_connection *con)
1137 con_out_kvec_reset(con);
1138 con->out_kvec_is_msg = true;
1139 con->out_msg_done = false;
1141 /* Sneak an ack in there first? If we can get it into the same
1142 * TCP packet that's a good thing. */
1143 if (con->in_seq > con->in_seq_acked) {
1144 con->in_seq_acked = con->in_seq;
1145 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1146 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1147 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1148 &con->out_temp_ack);
1151 BUG_ON(list_empty(&con->out_queue));
1152 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1154 BUG_ON(m->con != con);
1156 /* put message on sent list */
1158 list_move_tail(&m->list_head, &con->out_sent);
1161 * only assign outgoing seq # if we haven't sent this message
1162 * yet. if it is requeued, resend with it's original seq.
1164 if (m->needs_out_seq) {
1165 m->hdr.seq = cpu_to_le64(++con->out_seq);
1166 m->needs_out_seq = false;
1169 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d (%zd)\n",
1170 m, con->out_seq, le16_to_cpu(m->hdr.type),
1171 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1172 le32_to_cpu(m->hdr.data_len), m->p.length);
1173 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1175 /* tag + hdr + front + middle */
1176 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1177 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1178 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1181 con_out_kvec_add(con, m->middle->vec.iov_len,
1182 m->middle->vec.iov_base);
1184 /* fill in crc (except data pages), footer */
1185 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1186 con->out_msg->hdr.crc = cpu_to_le32(crc);
1187 con->out_msg->footer.flags = 0;
1189 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1190 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1192 crc = crc32c(0, m->middle->vec.iov_base,
1193 m->middle->vec.iov_len);
1194 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1196 con->out_msg->footer.middle_crc = 0;
1197 dout("%s front_crc %u middle_crc %u\n", __func__,
1198 le32_to_cpu(con->out_msg->footer.front_crc),
1199 le32_to_cpu(con->out_msg->footer.middle_crc));
1201 /* is there a data payload? */
1202 con->out_msg->footer.data_crc = 0;
1203 if (m->hdr.data_len) {
1204 prepare_message_data(con->out_msg, &con->out_msg_pos);
1205 con->out_more = 1; /* data + footer will follow */
1207 /* no, queue up footer too and be done */
1208 prepare_write_message_footer(con);
1211 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1217 static void prepare_write_ack(struct ceph_connection *con)
1219 dout("prepare_write_ack %p %llu -> %llu\n", con,
1220 con->in_seq_acked, con->in_seq);
1221 con->in_seq_acked = con->in_seq;
1223 con_out_kvec_reset(con);
1225 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1227 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1228 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1229 &con->out_temp_ack);
1231 con->out_more = 1; /* more will follow.. eventually.. */
1232 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1236 * Prepare to share the seq during handshake
1238 static void prepare_write_seq(struct ceph_connection *con)
1240 dout("prepare_write_seq %p %llu -> %llu\n", con,
1241 con->in_seq_acked, con->in_seq);
1242 con->in_seq_acked = con->in_seq;
1244 con_out_kvec_reset(con);
1246 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1247 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1248 &con->out_temp_ack);
1250 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1254 * Prepare to write keepalive byte.
1256 static void prepare_write_keepalive(struct ceph_connection *con)
1258 dout("prepare_write_keepalive %p\n", con);
1259 con_out_kvec_reset(con);
1260 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
1261 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1265 * Connection negotiation.
1268 static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1271 struct ceph_auth_handshake *auth;
1273 if (!con->ops->get_authorizer) {
1274 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1275 con->out_connect.authorizer_len = 0;
1279 /* Can't hold the mutex while getting authorizer */
1280 mutex_unlock(&con->mutex);
1281 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
1282 mutex_lock(&con->mutex);
1286 if (con->state != CON_STATE_NEGOTIATING)
1287 return ERR_PTR(-EAGAIN);
1289 con->auth_reply_buf = auth->authorizer_reply_buf;
1290 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
1295 * We connected to a peer and are saying hello.
1297 static void prepare_write_banner(struct ceph_connection *con)
1299 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1300 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1301 &con->msgr->my_enc_addr);
1304 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1307 static int prepare_write_connect(struct ceph_connection *con)
1309 unsigned int global_seq = get_global_seq(con->msgr, 0);
1312 struct ceph_auth_handshake *auth;
1314 switch (con->peer_name.type) {
1315 case CEPH_ENTITY_TYPE_MON:
1316 proto = CEPH_MONC_PROTOCOL;
1318 case CEPH_ENTITY_TYPE_OSD:
1319 proto = CEPH_OSDC_PROTOCOL;
1321 case CEPH_ENTITY_TYPE_MDS:
1322 proto = CEPH_MDSC_PROTOCOL;
1328 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1329 con->connect_seq, global_seq, proto);
1331 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
1332 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1333 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1334 con->out_connect.global_seq = cpu_to_le32(global_seq);
1335 con->out_connect.protocol_version = cpu_to_le32(proto);
1336 con->out_connect.flags = 0;
1338 auth_proto = CEPH_AUTH_UNKNOWN;
1339 auth = get_connect_authorizer(con, &auth_proto);
1341 return PTR_ERR(auth);
1343 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1344 con->out_connect.authorizer_len = auth ?
1345 cpu_to_le32(auth->authorizer_buf_len) : 0;
1347 con_out_kvec_add(con, sizeof (con->out_connect),
1349 if (auth && auth->authorizer_buf_len)
1350 con_out_kvec_add(con, auth->authorizer_buf_len,
1351 auth->authorizer_buf);
1354 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1360 * write as much of pending kvecs to the socket as we can.
1362 * 0 -> socket full, but more to do
1365 static int write_partial_kvec(struct ceph_connection *con)
1369 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1370 while (con->out_kvec_bytes > 0) {
1371 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1372 con->out_kvec_left, con->out_kvec_bytes,
1376 con->out_kvec_bytes -= ret;
1377 if (con->out_kvec_bytes == 0)
1380 /* account for full iov entries consumed */
1381 while (ret >= con->out_kvec_cur->iov_len) {
1382 BUG_ON(!con->out_kvec_left);
1383 ret -= con->out_kvec_cur->iov_len;
1384 con->out_kvec_cur++;
1385 con->out_kvec_left--;
1387 /* and for a partially-consumed entry */
1389 con->out_kvec_cur->iov_len -= ret;
1390 con->out_kvec_cur->iov_base += ret;
1393 con->out_kvec_left = 0;
1394 con->out_kvec_is_msg = false;
1397 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1398 con->out_kvec_bytes, con->out_kvec_left, ret);
1399 return ret; /* done! */
1402 static void out_msg_pos_next(struct ceph_connection *con, struct page *page,
1403 size_t len, size_t sent)
1405 struct ceph_msg *msg = con->out_msg;
1406 struct ceph_msg_pos *msg_pos = &con->out_msg_pos;
1407 bool need_crc = false;
1412 msg_pos->data_pos += sent;
1413 msg_pos->page_pos += sent;
1414 if (ceph_msg_has_pages(msg))
1415 need_crc = ceph_msg_data_advance(&msg->p, sent);
1416 else if (ceph_msg_has_pagelist(msg))
1417 need_crc = ceph_msg_data_advance(&msg->l, sent);
1419 else if (ceph_msg_has_bio(msg))
1420 need_crc = ceph_msg_data_advance(&msg->b, sent);
1421 #endif /* CONFIG_BLOCK */
1422 BUG_ON(need_crc && sent != len);
1427 BUG_ON(sent != len);
1428 msg_pos->page_pos = 0;
1430 msg_pos->did_page_crc = false;
1433 static void in_msg_pos_next(struct ceph_connection *con, size_t len,
1436 struct ceph_msg *msg = con->in_msg;
1437 struct ceph_msg_pos *msg_pos = &con->in_msg_pos;
1442 msg_pos->data_pos += received;
1443 msg_pos->page_pos += received;
1445 if (ceph_msg_has_bio(msg))
1446 (void) ceph_msg_data_advance(&msg->b, received);
1447 #endif /* CONFIG_BLOCK */
1451 BUG_ON(received != len);
1452 msg_pos->page_pos = 0;
1456 static u32 ceph_crc32c_page(u32 crc, struct page *page,
1457 unsigned int page_offset,
1458 unsigned int length)
1463 BUG_ON(kaddr == NULL);
1464 crc = crc32c(crc, kaddr + page_offset, length);
1470 * Write as much message data payload as we can. If we finish, queue
1472 * 1 -> done, footer is now queued in out_kvec[].
1473 * 0 -> socket full, but more to do
1476 static int write_partial_message_data(struct ceph_connection *con)
1478 struct ceph_msg *msg = con->out_msg;
1479 struct ceph_msg_pos *msg_pos = &con->out_msg_pos;
1480 unsigned int data_len = le32_to_cpu(msg->hdr.data_len);
1481 bool do_datacrc = !con->msgr->nocrc;
1484 dout("%s %p msg %p page %d offset %d\n", __func__,
1485 con, msg, msg_pos->page, msg_pos->page_pos);
1488 * Iterate through each page that contains data to be
1489 * written, and send as much as possible for each.
1491 * If we are calculating the data crc (the default), we will
1492 * need to map the page. If we have no pages, they have
1493 * been revoked, so use the zero page.
1495 while (data_len > msg_pos->data_pos) {
1501 if (ceph_msg_has_pages(msg)) {
1502 page = ceph_msg_data_next(&msg->p, &page_offset,
1503 &length, &last_piece);
1504 } else if (ceph_msg_has_pagelist(msg)) {
1505 page = ceph_msg_data_next(&msg->l, &page_offset,
1506 &length, &last_piece);
1508 } else if (ceph_msg_has_bio(msg)) {
1509 page = ceph_msg_data_next(&msg->b, &page_offset,
1510 &length, &last_piece);
1513 size_t resid = data_len - msg_pos->data_pos;
1516 page_offset = msg_pos->page_pos;
1517 length = PAGE_SIZE - page_offset;
1518 length = min(resid, length);
1519 last_piece = length == resid;
1521 if (do_datacrc && !msg_pos->did_page_crc) {
1522 u32 crc = le32_to_cpu(msg->footer.data_crc);
1524 crc = ceph_crc32c_page(crc, page, page_offset, length);
1525 msg->footer.data_crc = cpu_to_le32(crc);
1526 msg_pos->did_page_crc = true;
1528 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
1529 length, last_piece);
1533 out_msg_pos_next(con, page, length, (size_t) ret);
1536 dout("%s %p msg %p done\n", __func__, con, msg);
1538 /* prepare and queue up footer, too */
1540 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1541 con_out_kvec_reset(con);
1542 prepare_write_message_footer(con);
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 ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2191 static int read_partial_message_pages(struct ceph_connection *con,
2192 unsigned int data_len, bool do_datacrc)
2194 struct ceph_msg *msg = con->in_msg;
2195 struct ceph_msg_pos *msg_pos = &con->in_msg_pos;
2196 struct page **pages;
2204 pages = msg->p.pages;
2205 BUG_ON(pages == NULL);
2206 page = pages[msg_pos->page];
2207 page_offset = msg_pos->page_pos;
2208 BUG_ON(msg_pos->data_pos >= data_len);
2209 left = data_len - msg_pos->data_pos;
2210 BUG_ON(page_offset >= PAGE_SIZE);
2211 length = min_t(unsigned int, PAGE_SIZE - page_offset, left);
2213 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2218 con->in_data_crc = ceph_crc32c_page(con->in_data_crc, page,
2221 in_msg_pos_next(con, length, ret);
2227 static int read_partial_message_bio(struct ceph_connection *con,
2228 unsigned int data_len, bool do_datacrc)
2230 struct ceph_msg *msg = con->in_msg;
2238 page = ceph_msg_data_next(&msg->b, &page_offset, &length, NULL);
2240 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2245 con->in_data_crc = ceph_crc32c_page(con->in_data_crc, page,
2248 in_msg_pos_next(con, length, ret);
2254 static int read_partial_msg_data(struct ceph_connection *con)
2256 struct ceph_msg *msg = con->in_msg;
2257 struct ceph_msg_pos *msg_pos = &con->in_msg_pos;
2258 const bool do_datacrc = !con->msgr->nocrc;
2259 unsigned int data_len;
2264 data_len = le32_to_cpu(con->in_hdr.data_len);
2265 while (msg_pos->data_pos < data_len) {
2266 if (ceph_msg_has_pages(msg)) {
2267 ret = read_partial_message_pages(con, data_len,
2272 } else if (ceph_msg_has_bio(msg)) {
2273 ret = read_partial_message_bio(con,
2274 data_len, do_datacrc);
2283 return 1; /* must return > 0 to indicate success */
2287 * read (part of) a message.
2289 static int read_partial_message(struct ceph_connection *con)
2291 struct ceph_msg *m = con->in_msg;
2295 unsigned int front_len, middle_len, data_len;
2296 bool do_datacrc = !con->msgr->nocrc;
2300 dout("read_partial_message con %p msg %p\n", con, m);
2303 size = sizeof (con->in_hdr);
2305 ret = read_partial(con, end, size, &con->in_hdr);
2309 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2310 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2311 pr_err("read_partial_message bad hdr "
2312 " crc %u != expected %u\n",
2313 crc, con->in_hdr.crc);
2317 front_len = le32_to_cpu(con->in_hdr.front_len);
2318 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2320 middle_len = le32_to_cpu(con->in_hdr.middle_len);
2321 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2323 data_len = le32_to_cpu(con->in_hdr.data_len);
2324 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2328 seq = le64_to_cpu(con->in_hdr.seq);
2329 if ((s64)seq - (s64)con->in_seq < 1) {
2330 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2331 ENTITY_NAME(con->peer_name),
2332 ceph_pr_addr(&con->peer_addr.in_addr),
2333 seq, con->in_seq + 1);
2334 con->in_base_pos = -front_len - middle_len - data_len -
2336 con->in_tag = CEPH_MSGR_TAG_READY;
2338 } else if ((s64)seq - (s64)con->in_seq > 1) {
2339 pr_err("read_partial_message bad seq %lld expected %lld\n",
2340 seq, con->in_seq + 1);
2341 con->error_msg = "bad message sequence # for incoming message";
2345 /* allocate message? */
2349 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2350 front_len, data_len);
2351 ret = ceph_con_in_msg_alloc(con, &skip);
2355 /* skip this message */
2356 dout("alloc_msg said skip message\n");
2357 BUG_ON(con->in_msg);
2358 con->in_base_pos = -front_len - middle_len - data_len -
2360 con->in_tag = CEPH_MSGR_TAG_READY;
2365 BUG_ON(!con->in_msg);
2366 BUG_ON(con->in_msg->con != con);
2368 m->front.iov_len = 0; /* haven't read it yet */
2370 m->middle->vec.iov_len = 0;
2372 /* prepare for data payload, if any */
2375 prepare_message_data(con->in_msg, &con->in_msg_pos);
2379 ret = read_partial_message_section(con, &m->front, front_len,
2380 &con->in_front_crc);
2386 ret = read_partial_message_section(con, &m->middle->vec,
2388 &con->in_middle_crc);
2395 ret = read_partial_msg_data(con);
2401 size = sizeof (m->footer);
2403 ret = read_partial(con, end, size, &m->footer);
2407 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2408 m, front_len, m->footer.front_crc, middle_len,
2409 m->footer.middle_crc, data_len, m->footer.data_crc);
2412 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2413 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2414 m, con->in_front_crc, m->footer.front_crc);
2417 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2418 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2419 m, con->in_middle_crc, m->footer.middle_crc);
2423 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2424 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2425 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2426 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2430 return 1; /* done! */
2434 * Process message. This happens in the worker thread. The callback should
2435 * be careful not to do anything that waits on other incoming messages or it
2438 static void process_message(struct ceph_connection *con)
2440 struct ceph_msg *msg;
2442 BUG_ON(con->in_msg->con != con);
2443 con->in_msg->con = NULL;
2448 /* if first message, set peer_name */
2449 if (con->peer_name.type == 0)
2450 con->peer_name = msg->hdr.src;
2453 mutex_unlock(&con->mutex);
2455 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2456 msg, le64_to_cpu(msg->hdr.seq),
2457 ENTITY_NAME(msg->hdr.src),
2458 le16_to_cpu(msg->hdr.type),
2459 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2460 le32_to_cpu(msg->hdr.front_len),
2461 le32_to_cpu(msg->hdr.data_len),
2462 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2463 con->ops->dispatch(con, msg);
2465 mutex_lock(&con->mutex);
2470 * Write something to the socket. Called in a worker thread when the
2471 * socket appears to be writeable and we have something ready to send.
2473 static int try_write(struct ceph_connection *con)
2477 dout("try_write start %p state %lu\n", con, con->state);
2480 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2482 /* open the socket first? */
2483 if (con->state == CON_STATE_PREOPEN) {
2485 con->state = CON_STATE_CONNECTING;
2487 con_out_kvec_reset(con);
2488 prepare_write_banner(con);
2489 prepare_read_banner(con);
2491 BUG_ON(con->in_msg);
2492 con->in_tag = CEPH_MSGR_TAG_READY;
2493 dout("try_write initiating connect on %p new state %lu\n",
2495 ret = ceph_tcp_connect(con);
2497 con->error_msg = "connect error";
2503 /* kvec data queued? */
2504 if (con->out_skip) {
2505 ret = write_partial_skip(con);
2509 if (con->out_kvec_left) {
2510 ret = write_partial_kvec(con);
2517 if (con->out_msg_done) {
2518 ceph_msg_put(con->out_msg);
2519 con->out_msg = NULL; /* we're done with this one */
2523 ret = write_partial_message_data(con);
2525 goto more_kvec; /* we need to send the footer, too! */
2529 dout("try_write write_partial_message_data err %d\n",
2536 if (con->state == CON_STATE_OPEN) {
2537 /* is anything else pending? */
2538 if (!list_empty(&con->out_queue)) {
2539 prepare_write_message(con);
2542 if (con->in_seq > con->in_seq_acked) {
2543 prepare_write_ack(con);
2546 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2547 prepare_write_keepalive(con);
2552 /* Nothing to do! */
2553 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2554 dout("try_write nothing else to write.\n");
2557 dout("try_write done on %p ret %d\n", con, ret);
2564 * Read what we can from the socket.
2566 static int try_read(struct ceph_connection *con)
2571 dout("try_read start on %p state %lu\n", con, con->state);
2572 if (con->state != CON_STATE_CONNECTING &&
2573 con->state != CON_STATE_NEGOTIATING &&
2574 con->state != CON_STATE_OPEN)
2579 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2582 if (con->state == CON_STATE_CONNECTING) {
2583 dout("try_read connecting\n");
2584 ret = read_partial_banner(con);
2587 ret = process_banner(con);
2591 con->state = CON_STATE_NEGOTIATING;
2594 * Received banner is good, exchange connection info.
2595 * Do not reset out_kvec, as sending our banner raced
2596 * with receiving peer banner after connect completed.
2598 ret = prepare_write_connect(con);
2601 prepare_read_connect(con);
2603 /* Send connection info before awaiting response */
2607 if (con->state == CON_STATE_NEGOTIATING) {
2608 dout("try_read negotiating\n");
2609 ret = read_partial_connect(con);
2612 ret = process_connect(con);
2618 WARN_ON(con->state != CON_STATE_OPEN);
2620 if (con->in_base_pos < 0) {
2622 * skipping + discarding content.
2624 * FIXME: there must be a better way to do this!
2626 static char buf[SKIP_BUF_SIZE];
2627 int skip = min((int) sizeof (buf), -con->in_base_pos);
2629 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2630 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2633 con->in_base_pos += ret;
2634 if (con->in_base_pos)
2637 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2641 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2644 dout("try_read got tag %d\n", (int)con->in_tag);
2645 switch (con->in_tag) {
2646 case CEPH_MSGR_TAG_MSG:
2647 prepare_read_message(con);
2649 case CEPH_MSGR_TAG_ACK:
2650 prepare_read_ack(con);
2652 case CEPH_MSGR_TAG_CLOSE:
2653 con_close_socket(con);
2654 con->state = CON_STATE_CLOSED;
2660 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2661 ret = read_partial_message(con);
2665 con->error_msg = "bad crc";
2669 con->error_msg = "io error";
2674 if (con->in_tag == CEPH_MSGR_TAG_READY)
2676 process_message(con);
2677 if (con->state == CON_STATE_OPEN)
2678 prepare_read_tag(con);
2681 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2682 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2684 * the final handshake seq exchange is semantically
2685 * equivalent to an ACK
2687 ret = read_partial_ack(con);
2695 dout("try_read done on %p ret %d\n", con, ret);
2699 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2700 con->error_msg = "protocol error, garbage tag";
2707 * Atomically queue work on a connection after the specified delay.
2708 * Bump @con reference to avoid races with connection teardown.
2709 * Returns 0 if work was queued, or an error code otherwise.
2711 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2713 if (!con->ops->get(con)) {
2714 dout("%s %p ref count 0\n", __func__, con);
2719 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2720 dout("%s %p - already queued\n", __func__, con);
2726 dout("%s %p %lu\n", __func__, con, delay);
2731 static void queue_con(struct ceph_connection *con)
2733 (void) queue_con_delay(con, 0);
2736 static bool con_sock_closed(struct ceph_connection *con)
2738 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2742 case CON_STATE_ ## x: \
2743 con->error_msg = "socket closed (con state " #x ")"; \
2746 switch (con->state) {
2754 pr_warning("%s con %p unrecognized state %lu\n",
2755 __func__, con, con->state);
2756 con->error_msg = "unrecognized con state";
2765 static bool con_backoff(struct ceph_connection *con)
2769 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2772 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2774 dout("%s: con %p FAILED to back off %lu\n", __func__,
2776 BUG_ON(ret == -ENOENT);
2777 con_flag_set(con, CON_FLAG_BACKOFF);
2783 /* Finish fault handling; con->mutex must *not* be held here */
2785 static void con_fault_finish(struct ceph_connection *con)
2788 * in case we faulted due to authentication, invalidate our
2789 * current tickets so that we can get new ones.
2791 if (con->auth_retry && con->ops->invalidate_authorizer) {
2792 dout("calling invalidate_authorizer()\n");
2793 con->ops->invalidate_authorizer(con);
2796 if (con->ops->fault)
2797 con->ops->fault(con);
2801 * Do some work on a connection. Drop a connection ref when we're done.
2803 static void con_work(struct work_struct *work)
2805 struct ceph_connection *con = container_of(work, struct ceph_connection,
2809 mutex_lock(&con->mutex);
2813 if ((fault = con_sock_closed(con))) {
2814 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2817 if (con_backoff(con)) {
2818 dout("%s: con %p BACKOFF\n", __func__, con);
2821 if (con->state == CON_STATE_STANDBY) {
2822 dout("%s: con %p STANDBY\n", __func__, con);
2825 if (con->state == CON_STATE_CLOSED) {
2826 dout("%s: con %p CLOSED\n", __func__, con);
2830 if (con->state == CON_STATE_PREOPEN) {
2831 dout("%s: con %p PREOPEN\n", __func__, con);
2835 ret = try_read(con);
2839 con->error_msg = "socket error on read";
2844 ret = try_write(con);
2848 con->error_msg = "socket error on write";
2852 break; /* If we make it to here, we're done */
2856 mutex_unlock(&con->mutex);
2859 con_fault_finish(con);
2865 * Generic error/fault handler. A retry mechanism is used with
2866 * exponential backoff
2868 static void con_fault(struct ceph_connection *con)
2870 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2871 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2872 dout("fault %p state %lu to peer %s\n",
2873 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2875 WARN_ON(con->state != CON_STATE_CONNECTING &&
2876 con->state != CON_STATE_NEGOTIATING &&
2877 con->state != CON_STATE_OPEN);
2879 con_close_socket(con);
2881 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2882 dout("fault on LOSSYTX channel, marking CLOSED\n");
2883 con->state = CON_STATE_CLOSED;
2888 BUG_ON(con->in_msg->con != con);
2889 con->in_msg->con = NULL;
2890 ceph_msg_put(con->in_msg);
2895 /* Requeue anything that hasn't been acked */
2896 list_splice_init(&con->out_sent, &con->out_queue);
2898 /* If there are no messages queued or keepalive pending, place
2899 * the connection in a STANDBY state */
2900 if (list_empty(&con->out_queue) &&
2901 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
2902 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2903 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2904 con->state = CON_STATE_STANDBY;
2906 /* retry after a delay. */
2907 con->state = CON_STATE_PREOPEN;
2908 if (con->delay == 0)
2909 con->delay = BASE_DELAY_INTERVAL;
2910 else if (con->delay < MAX_DELAY_INTERVAL)
2912 con_flag_set(con, CON_FLAG_BACKOFF);
2920 * initialize a new messenger instance
2922 void ceph_messenger_init(struct ceph_messenger *msgr,
2923 struct ceph_entity_addr *myaddr,
2924 u32 supported_features,
2925 u32 required_features,
2928 msgr->supported_features = supported_features;
2929 msgr->required_features = required_features;
2931 spin_lock_init(&msgr->global_seq_lock);
2934 msgr->inst.addr = *myaddr;
2936 /* select a random nonce */
2937 msgr->inst.addr.type = 0;
2938 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2939 encode_my_addr(msgr);
2940 msgr->nocrc = nocrc;
2942 atomic_set(&msgr->stopping, 0);
2944 dout("%s %p\n", __func__, msgr);
2946 EXPORT_SYMBOL(ceph_messenger_init);
2948 static void clear_standby(struct ceph_connection *con)
2950 /* come back from STANDBY? */
2951 if (con->state == CON_STATE_STANDBY) {
2952 dout("clear_standby %p and ++connect_seq\n", con);
2953 con->state = CON_STATE_PREOPEN;
2955 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2956 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
2961 * Queue up an outgoing message on the given connection.
2963 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2966 msg->hdr.src = con->msgr->inst.name;
2967 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2968 msg->needs_out_seq = true;
2970 mutex_lock(&con->mutex);
2972 if (con->state == CON_STATE_CLOSED) {
2973 dout("con_send %p closed, dropping %p\n", con, msg);
2975 mutex_unlock(&con->mutex);
2979 BUG_ON(msg->con != NULL);
2980 msg->con = con->ops->get(con);
2981 BUG_ON(msg->con == NULL);
2983 BUG_ON(!list_empty(&msg->list_head));
2984 list_add_tail(&msg->list_head, &con->out_queue);
2985 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2986 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2987 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2988 le32_to_cpu(msg->hdr.front_len),
2989 le32_to_cpu(msg->hdr.middle_len),
2990 le32_to_cpu(msg->hdr.data_len));
2993 mutex_unlock(&con->mutex);
2995 /* if there wasn't anything waiting to send before, queue
2997 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3000 EXPORT_SYMBOL(ceph_con_send);
3003 * Revoke a message that was previously queued for send
3005 void ceph_msg_revoke(struct ceph_msg *msg)
3007 struct ceph_connection *con = msg->con;
3010 return; /* Message not in our possession */
3012 mutex_lock(&con->mutex);
3013 if (!list_empty(&msg->list_head)) {
3014 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
3015 list_del_init(&msg->list_head);
3016 BUG_ON(msg->con == NULL);
3017 msg->con->ops->put(msg->con);
3023 if (con->out_msg == msg) {
3024 dout("%s %p msg %p - was sending\n", __func__, con, msg);
3025 con->out_msg = NULL;
3026 if (con->out_kvec_is_msg) {
3027 con->out_skip = con->out_kvec_bytes;
3028 con->out_kvec_is_msg = false;
3034 mutex_unlock(&con->mutex);
3038 * Revoke a message that we may be reading data into
3040 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
3042 struct ceph_connection *con;
3044 BUG_ON(msg == NULL);
3046 dout("%s msg %p null con\n", __func__, msg);
3048 return; /* Message not in our possession */
3052 mutex_lock(&con->mutex);
3053 if (con->in_msg == msg) {
3054 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3055 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3056 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
3058 /* skip rest of message */
3059 dout("%s %p msg %p revoked\n", __func__, con, msg);
3060 con->in_base_pos = con->in_base_pos -
3061 sizeof(struct ceph_msg_header) -
3065 sizeof(struct ceph_msg_footer);
3066 ceph_msg_put(con->in_msg);
3068 con->in_tag = CEPH_MSGR_TAG_READY;
3071 dout("%s %p in_msg %p msg %p no-op\n",
3072 __func__, con, con->in_msg, msg);
3074 mutex_unlock(&con->mutex);
3078 * Queue a keepalive byte to ensure the tcp connection is alive.
3080 void ceph_con_keepalive(struct ceph_connection *con)
3082 dout("con_keepalive %p\n", con);
3083 mutex_lock(&con->mutex);
3085 mutex_unlock(&con->mutex);
3086 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3087 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3090 EXPORT_SYMBOL(ceph_con_keepalive);
3092 static void ceph_msg_data_init(struct ceph_msg_data *data)
3094 data->type = CEPH_MSG_DATA_NONE;
3097 void ceph_msg_data_set_pages(struct ceph_msg *msg, struct page **pages,
3098 size_t length, size_t alignment)
3102 BUG_ON(msg->p.type != CEPH_MSG_DATA_NONE);
3104 msg->p.type = CEPH_MSG_DATA_PAGES;
3105 msg->p.pages = pages;
3106 msg->p.length = length;
3107 msg->p.alignment = alignment & ~PAGE_MASK;
3109 EXPORT_SYMBOL(ceph_msg_data_set_pages);
3111 void ceph_msg_data_set_pagelist(struct ceph_msg *msg,
3112 struct ceph_pagelist *pagelist)
3115 BUG_ON(!pagelist->length);
3116 BUG_ON(msg->l.type != CEPH_MSG_DATA_NONE);
3118 msg->l.type = CEPH_MSG_DATA_PAGELIST;
3119 msg->l.pagelist = pagelist;
3121 EXPORT_SYMBOL(ceph_msg_data_set_pagelist);
3123 void ceph_msg_data_set_bio(struct ceph_msg *msg, struct bio *bio)
3126 BUG_ON(msg->b.type != CEPH_MSG_DATA_NONE);
3128 msg->b.type = CEPH_MSG_DATA_BIO;
3131 EXPORT_SYMBOL(ceph_msg_data_set_bio);
3134 * construct a new message with given type, size
3135 * the new msg has a ref count of 1.
3137 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3142 m = kzalloc(sizeof(*m), flags);
3146 m->hdr.type = cpu_to_le16(type);
3147 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3148 m->hdr.front_len = cpu_to_le32(front_len);
3150 INIT_LIST_HEAD(&m->list_head);
3151 kref_init(&m->kref);
3153 ceph_msg_data_init(&m->p);
3154 ceph_msg_data_init(&m->l);
3155 ceph_msg_data_init(&m->b);
3158 m->front_max = front_len;
3160 if (front_len > PAGE_CACHE_SIZE) {
3161 m->front.iov_base = __vmalloc(front_len, flags,
3163 m->front_is_vmalloc = true;
3165 m->front.iov_base = kmalloc(front_len, flags);
3167 if (m->front.iov_base == NULL) {
3168 dout("ceph_msg_new can't allocate %d bytes\n",
3173 m->front.iov_base = NULL;
3175 m->front.iov_len = front_len;
3177 dout("ceph_msg_new %p front %d\n", m, front_len);
3184 pr_err("msg_new can't create type %d front %d\n", type,
3188 dout("msg_new can't create type %d front %d\n", type,
3193 EXPORT_SYMBOL(ceph_msg_new);
3196 * Allocate "middle" portion of a message, if it is needed and wasn't
3197 * allocated by alloc_msg. This allows us to read a small fixed-size
3198 * per-type header in the front and then gracefully fail (i.e.,
3199 * propagate the error to the caller based on info in the front) when
3200 * the middle is too large.
3202 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3204 int type = le16_to_cpu(msg->hdr.type);
3205 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3207 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3208 ceph_msg_type_name(type), middle_len);
3209 BUG_ON(!middle_len);
3210 BUG_ON(msg->middle);
3212 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3219 * Allocate a message for receiving an incoming message on a
3220 * connection, and save the result in con->in_msg. Uses the
3221 * connection's private alloc_msg op if available.
3223 * Returns 0 on success, or a negative error code.
3225 * On success, if we set *skip = 1:
3226 * - the next message should be skipped and ignored.
3227 * - con->in_msg == NULL
3228 * or if we set *skip = 0:
3229 * - con->in_msg is non-null.
3230 * On error (ENOMEM, EAGAIN, ...),
3231 * - con->in_msg == NULL
3233 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3235 struct ceph_msg_header *hdr = &con->in_hdr;
3236 int middle_len = le32_to_cpu(hdr->middle_len);
3237 struct ceph_msg *msg;
3240 BUG_ON(con->in_msg != NULL);
3241 BUG_ON(!con->ops->alloc_msg);
3243 mutex_unlock(&con->mutex);
3244 msg = con->ops->alloc_msg(con, hdr, skip);
3245 mutex_lock(&con->mutex);
3246 if (con->state != CON_STATE_OPEN) {
3254 con->in_msg->con = con->ops->get(con);
3255 BUG_ON(con->in_msg->con == NULL);
3258 * Null message pointer means either we should skip
3259 * this message or we couldn't allocate memory. The
3260 * former is not an error.
3264 con->error_msg = "error allocating memory for incoming message";
3268 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3270 if (middle_len && !con->in_msg->middle) {
3271 ret = ceph_alloc_middle(con, con->in_msg);
3273 ceph_msg_put(con->in_msg);
3283 * Free a generically kmalloc'd message.
3285 void ceph_msg_kfree(struct ceph_msg *m)
3287 dout("msg_kfree %p\n", m);
3288 if (m->front_is_vmalloc)
3289 vfree(m->front.iov_base);
3291 kfree(m->front.iov_base);
3296 * Drop a msg ref. Destroy as needed.
3298 void ceph_msg_last_put(struct kref *kref)
3300 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3302 dout("ceph_msg_put last one on %p\n", m);
3303 WARN_ON(!list_empty(&m->list_head));
3305 /* drop middle, data, if any */
3307 ceph_buffer_put(m->middle);
3310 if (ceph_msg_has_pages(m)) {
3313 m->p.type = CEPH_OSD_DATA_TYPE_NONE;
3315 if (ceph_msg_has_pagelist(m)) {
3316 ceph_pagelist_release(m->l.pagelist);
3317 kfree(m->l.pagelist);
3318 m->l.pagelist = NULL;
3319 m->l.type = CEPH_OSD_DATA_TYPE_NONE;
3321 if (ceph_msg_has_bio(m)) {
3323 m->b.type = CEPH_OSD_DATA_TYPE_NONE;
3327 ceph_msgpool_put(m->pool, m);
3331 EXPORT_SYMBOL(ceph_msg_last_put);
3333 void ceph_msg_dump(struct ceph_msg *msg)
3335 pr_debug("msg_dump %p (front_max %d length %zd)\n", msg,
3336 msg->front_max, msg->p.length);
3337 print_hex_dump(KERN_DEBUG, "header: ",
3338 DUMP_PREFIX_OFFSET, 16, 1,
3339 &msg->hdr, sizeof(msg->hdr), true);
3340 print_hex_dump(KERN_DEBUG, " front: ",
3341 DUMP_PREFIX_OFFSET, 16, 1,
3342 msg->front.iov_base, msg->front.iov_len, true);
3344 print_hex_dump(KERN_DEBUG, "middle: ",
3345 DUMP_PREFIX_OFFSET, 16, 1,
3346 msg->middle->vec.iov_base,
3347 msg->middle->vec.iov_len, true);
3348 print_hex_dump(KERN_DEBUG, "footer: ",
3349 DUMP_PREFIX_OFFSET, 16, 1,
3350 &msg->footer, sizeof(msg->footer), true);
3352 EXPORT_SYMBOL(ceph_msg_dump);