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_cursor *cursor,
728 struct ceph_msg_data *data = cursor->data;
731 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
735 BUG_ON(!bio->bi_vcnt);
737 cursor->resid = min(length, data->bio_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_cursor *cursor,
748 struct ceph_msg_data *data = cursor->data;
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 > cursor->resid);
770 BUG_ON(*page_offset + *length > PAGE_SIZE);
772 return bio_vec->bv_page;
775 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
779 struct bio_vec *bio_vec;
782 BUG_ON(cursor->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;
820 #endif /* CONFIG_BLOCK */
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_cursor *cursor,
829 struct ceph_msg_data *data = cursor->data;
832 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
834 BUG_ON(!data->pages);
835 BUG_ON(!data->length);
837 cursor->resid = min(length, data->length);
838 page_count = calc_pages_for(data->alignment, (u64)data->length);
839 cursor->page_offset = data->alignment & ~PAGE_MASK;
840 cursor->page_index = 0;
841 BUG_ON(page_count > (int)USHRT_MAX);
842 cursor->page_count = (unsigned short)page_count;
843 BUG_ON(length > SIZE_MAX - cursor->page_offset);
844 cursor->last_piece = (size_t)cursor->page_offset + length <= PAGE_SIZE;
848 ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
849 size_t *page_offset, size_t *length)
851 struct ceph_msg_data *data = cursor->data;
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_cursor *cursor,
870 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
872 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
874 /* Advance the cursor page offset */
876 cursor->resid -= bytes;
877 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
878 if (!bytes || cursor->page_offset)
879 return false; /* more bytes to process in the current page */
881 /* Move on to the next page; offset is already at 0 */
883 BUG_ON(cursor->page_index >= cursor->page_count);
884 cursor->page_index++;
885 cursor->last_piece = cursor->resid <= PAGE_SIZE;
891 * For a pagelist, a piece is whatever remains to be consumed in the
892 * first page in the list, or the front of the next page.
895 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
898 struct ceph_msg_data *data = cursor->data;
899 struct ceph_pagelist *pagelist;
902 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
904 pagelist = data->pagelist;
908 return; /* pagelist can be assigned but empty */
910 BUG_ON(list_empty(&pagelist->head));
911 page = list_first_entry(&pagelist->head, struct page, lru);
913 cursor->resid = min(length, pagelist->length);
916 cursor->last_piece = cursor->resid <= PAGE_SIZE;
920 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
921 size_t *page_offset, size_t *length)
923 struct ceph_msg_data *data = cursor->data;
924 struct ceph_pagelist *pagelist;
926 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
928 pagelist = data->pagelist;
931 BUG_ON(!cursor->page);
932 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
934 /* offset of first page in pagelist is always 0 */
935 *page_offset = cursor->offset & ~PAGE_MASK;
936 if (cursor->last_piece)
937 *length = cursor->resid;
939 *length = PAGE_SIZE - *page_offset;
944 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
947 struct ceph_msg_data *data = cursor->data;
948 struct ceph_pagelist *pagelist;
950 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
952 pagelist = data->pagelist;
955 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
956 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
958 /* Advance the cursor offset */
960 cursor->resid -= bytes;
961 cursor->offset += bytes;
962 /* offset of first page in pagelist is always 0 */
963 if (!bytes || cursor->offset & ~PAGE_MASK)
964 return false; /* more bytes to process in the current page */
966 /* Move on to the next page */
968 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
969 cursor->page = list_entry_next(cursor->page, lru);
970 cursor->last_piece = cursor->resid <= PAGE_SIZE;
976 * Message data is handled (sent or received) in pieces, where each
977 * piece resides on a single page. The network layer might not
978 * consume an entire piece at once. A data item's cursor keeps
979 * track of which piece is next to process and how much remains to
980 * be processed in that piece. It also tracks whether the current
981 * piece is the last one in the data item.
983 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
985 size_t length = cursor->total_resid;
987 switch (cursor->data->type) {
988 case CEPH_MSG_DATA_PAGELIST:
989 ceph_msg_data_pagelist_cursor_init(cursor, length);
991 case CEPH_MSG_DATA_PAGES:
992 ceph_msg_data_pages_cursor_init(cursor, length);
995 case CEPH_MSG_DATA_BIO:
996 ceph_msg_data_bio_cursor_init(cursor, length);
998 #endif /* CONFIG_BLOCK */
999 case CEPH_MSG_DATA_NONE:
1004 cursor->need_crc = true;
1007 static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1009 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1010 struct ceph_msg_data *data;
1013 BUG_ON(length > msg->data_length);
1014 BUG_ON(list_empty(&msg->data));
1016 cursor->data_head = &msg->data;
1017 cursor->total_resid = length;
1018 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1019 cursor->data = data;
1021 __ceph_msg_data_cursor_init(cursor);
1025 * Return the page containing the next piece to process for a given
1026 * data item, and supply the page offset and length of that piece.
1027 * Indicate whether this is the last piece in this data item.
1029 static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1030 size_t *page_offset, size_t *length,
1035 switch (cursor->data->type) {
1036 case CEPH_MSG_DATA_PAGELIST:
1037 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1039 case CEPH_MSG_DATA_PAGES:
1040 page = ceph_msg_data_pages_next(cursor, page_offset, length);
1043 case CEPH_MSG_DATA_BIO:
1044 page = ceph_msg_data_bio_next(cursor, page_offset, length);
1046 #endif /* CONFIG_BLOCK */
1047 case CEPH_MSG_DATA_NONE:
1053 BUG_ON(*page_offset + *length > PAGE_SIZE);
1056 *last_piece = cursor->last_piece;
1062 * Returns true if the result moves the cursor on to the next piece
1065 static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1070 BUG_ON(bytes > cursor->resid);
1071 switch (cursor->data->type) {
1072 case CEPH_MSG_DATA_PAGELIST:
1073 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1075 case CEPH_MSG_DATA_PAGES:
1076 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1079 case CEPH_MSG_DATA_BIO:
1080 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1082 #endif /* CONFIG_BLOCK */
1083 case CEPH_MSG_DATA_NONE:
1088 cursor->total_resid -= bytes;
1090 if (!cursor->resid && cursor->total_resid) {
1091 WARN_ON(!cursor->last_piece);
1092 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1093 cursor->data = list_entry_next(cursor->data, links);
1094 __ceph_msg_data_cursor_init(cursor);
1097 cursor->need_crc = new_piece;
1102 static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
1107 /* Initialize data cursor */
1109 ceph_msg_data_cursor_init(msg, (size_t)data_len);
1113 * Prepare footer for currently outgoing message, and finish things
1114 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1116 static void prepare_write_message_footer(struct ceph_connection *con)
1118 struct ceph_msg *m = con->out_msg;
1119 int v = con->out_kvec_left;
1121 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1123 dout("prepare_write_message_footer %p\n", con);
1124 con->out_kvec_is_msg = true;
1125 con->out_kvec[v].iov_base = &m->footer;
1126 con->out_kvec[v].iov_len = sizeof(m->footer);
1127 con->out_kvec_bytes += sizeof(m->footer);
1128 con->out_kvec_left++;
1129 con->out_more = m->more_to_follow;
1130 con->out_msg_done = true;
1134 * Prepare headers for the next outgoing message.
1136 static void prepare_write_message(struct ceph_connection *con)
1141 con_out_kvec_reset(con);
1142 con->out_kvec_is_msg = true;
1143 con->out_msg_done = false;
1145 /* Sneak an ack in there first? If we can get it into the same
1146 * TCP packet that's a good thing. */
1147 if (con->in_seq > con->in_seq_acked) {
1148 con->in_seq_acked = con->in_seq;
1149 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1150 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1151 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1152 &con->out_temp_ack);
1155 BUG_ON(list_empty(&con->out_queue));
1156 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1158 BUG_ON(m->con != con);
1160 /* put message on sent list */
1162 list_move_tail(&m->list_head, &con->out_sent);
1165 * only assign outgoing seq # if we haven't sent this message
1166 * yet. if it is requeued, resend with it's original seq.
1168 if (m->needs_out_seq) {
1169 m->hdr.seq = cpu_to_le64(++con->out_seq);
1170 m->needs_out_seq = false;
1172 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
1174 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1175 m, con->out_seq, le16_to_cpu(m->hdr.type),
1176 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1178 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1180 /* tag + hdr + front + middle */
1181 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1182 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1183 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1186 con_out_kvec_add(con, m->middle->vec.iov_len,
1187 m->middle->vec.iov_base);
1189 /* fill in crc (except data pages), footer */
1190 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1191 con->out_msg->hdr.crc = cpu_to_le32(crc);
1192 con->out_msg->footer.flags = 0;
1194 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1195 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1197 crc = crc32c(0, m->middle->vec.iov_base,
1198 m->middle->vec.iov_len);
1199 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1201 con->out_msg->footer.middle_crc = 0;
1202 dout("%s front_crc %u middle_crc %u\n", __func__,
1203 le32_to_cpu(con->out_msg->footer.front_crc),
1204 le32_to_cpu(con->out_msg->footer.middle_crc));
1206 /* is there a data payload? */
1207 con->out_msg->footer.data_crc = 0;
1208 if (m->data_length) {
1209 prepare_message_data(con->out_msg, m->data_length);
1210 con->out_more = 1; /* data + footer will follow */
1212 /* no, queue up footer too and be done */
1213 prepare_write_message_footer(con);
1216 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1222 static void prepare_write_ack(struct ceph_connection *con)
1224 dout("prepare_write_ack %p %llu -> %llu\n", con,
1225 con->in_seq_acked, con->in_seq);
1226 con->in_seq_acked = con->in_seq;
1228 con_out_kvec_reset(con);
1230 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1232 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1233 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1234 &con->out_temp_ack);
1236 con->out_more = 1; /* more will follow.. eventually.. */
1237 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1241 * Prepare to share the seq during handshake
1243 static void prepare_write_seq(struct ceph_connection *con)
1245 dout("prepare_write_seq %p %llu -> %llu\n", con,
1246 con->in_seq_acked, con->in_seq);
1247 con->in_seq_acked = con->in_seq;
1249 con_out_kvec_reset(con);
1251 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1252 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1253 &con->out_temp_ack);
1255 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1259 * Prepare to write keepalive byte.
1261 static void prepare_write_keepalive(struct ceph_connection *con)
1263 dout("prepare_write_keepalive %p\n", con);
1264 con_out_kvec_reset(con);
1265 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
1266 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1270 * Connection negotiation.
1273 static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1276 struct ceph_auth_handshake *auth;
1278 if (!con->ops->get_authorizer) {
1279 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1280 con->out_connect.authorizer_len = 0;
1284 /* Can't hold the mutex while getting authorizer */
1285 mutex_unlock(&con->mutex);
1286 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
1287 mutex_lock(&con->mutex);
1291 if (con->state != CON_STATE_NEGOTIATING)
1292 return ERR_PTR(-EAGAIN);
1294 con->auth_reply_buf = auth->authorizer_reply_buf;
1295 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
1300 * We connected to a peer and are saying hello.
1302 static void prepare_write_banner(struct ceph_connection *con)
1304 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1305 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1306 &con->msgr->my_enc_addr);
1309 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1312 static int prepare_write_connect(struct ceph_connection *con)
1314 unsigned int global_seq = get_global_seq(con->msgr, 0);
1317 struct ceph_auth_handshake *auth;
1319 switch (con->peer_name.type) {
1320 case CEPH_ENTITY_TYPE_MON:
1321 proto = CEPH_MONC_PROTOCOL;
1323 case CEPH_ENTITY_TYPE_OSD:
1324 proto = CEPH_OSDC_PROTOCOL;
1326 case CEPH_ENTITY_TYPE_MDS:
1327 proto = CEPH_MDSC_PROTOCOL;
1333 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1334 con->connect_seq, global_seq, proto);
1336 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
1337 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1338 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1339 con->out_connect.global_seq = cpu_to_le32(global_seq);
1340 con->out_connect.protocol_version = cpu_to_le32(proto);
1341 con->out_connect.flags = 0;
1343 auth_proto = CEPH_AUTH_UNKNOWN;
1344 auth = get_connect_authorizer(con, &auth_proto);
1346 return PTR_ERR(auth);
1348 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1349 con->out_connect.authorizer_len = auth ?
1350 cpu_to_le32(auth->authorizer_buf_len) : 0;
1352 con_out_kvec_add(con, sizeof (con->out_connect),
1354 if (auth && auth->authorizer_buf_len)
1355 con_out_kvec_add(con, auth->authorizer_buf_len,
1356 auth->authorizer_buf);
1359 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1365 * write as much of pending kvecs to the socket as we can.
1367 * 0 -> socket full, but more to do
1370 static int write_partial_kvec(struct ceph_connection *con)
1374 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1375 while (con->out_kvec_bytes > 0) {
1376 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1377 con->out_kvec_left, con->out_kvec_bytes,
1381 con->out_kvec_bytes -= ret;
1382 if (con->out_kvec_bytes == 0)
1385 /* account for full iov entries consumed */
1386 while (ret >= con->out_kvec_cur->iov_len) {
1387 BUG_ON(!con->out_kvec_left);
1388 ret -= con->out_kvec_cur->iov_len;
1389 con->out_kvec_cur++;
1390 con->out_kvec_left--;
1392 /* and for a partially-consumed entry */
1394 con->out_kvec_cur->iov_len -= ret;
1395 con->out_kvec_cur->iov_base += ret;
1398 con->out_kvec_left = 0;
1399 con->out_kvec_is_msg = false;
1402 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1403 con->out_kvec_bytes, con->out_kvec_left, ret);
1404 return ret; /* done! */
1407 static u32 ceph_crc32c_page(u32 crc, struct page *page,
1408 unsigned int page_offset,
1409 unsigned int length)
1414 BUG_ON(kaddr == NULL);
1415 crc = crc32c(crc, kaddr + page_offset, length);
1421 * Write as much message data payload as we can. If we finish, queue
1423 * 1 -> done, footer is now queued in out_kvec[].
1424 * 0 -> socket full, but more to do
1427 static int write_partial_message_data(struct ceph_connection *con)
1429 struct ceph_msg *msg = con->out_msg;
1430 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1431 bool do_datacrc = !con->msgr->nocrc;
1434 dout("%s %p msg %p\n", __func__, con, msg);
1436 if (list_empty(&msg->data))
1440 * Iterate through each page that contains data to be
1441 * written, and send as much as possible for each.
1443 * If we are calculating the data crc (the default), we will
1444 * need to map the page. If we have no pages, they have
1445 * been revoked, so use the zero page.
1447 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1448 while (cursor->resid) {
1456 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
1458 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
1459 length, last_piece);
1462 msg->footer.data_crc = cpu_to_le32(crc);
1466 if (do_datacrc && cursor->need_crc)
1467 crc = ceph_crc32c_page(crc, page, page_offset, length);
1468 need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret);
1471 dout("%s %p msg %p done\n", __func__, con, msg);
1473 /* prepare and queue up footer, too */
1475 msg->footer.data_crc = cpu_to_le32(crc);
1477 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1478 con_out_kvec_reset(con);
1479 prepare_write_message_footer(con);
1481 return 1; /* must return > 0 to indicate success */
1487 static int write_partial_skip(struct ceph_connection *con)
1491 while (con->out_skip > 0) {
1492 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
1494 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
1497 con->out_skip -= ret;
1505 * Prepare to read connection handshake, or an ack.
1507 static void prepare_read_banner(struct ceph_connection *con)
1509 dout("prepare_read_banner %p\n", con);
1510 con->in_base_pos = 0;
1513 static void prepare_read_connect(struct ceph_connection *con)
1515 dout("prepare_read_connect %p\n", con);
1516 con->in_base_pos = 0;
1519 static void prepare_read_ack(struct ceph_connection *con)
1521 dout("prepare_read_ack %p\n", con);
1522 con->in_base_pos = 0;
1525 static void prepare_read_seq(struct ceph_connection *con)
1527 dout("prepare_read_seq %p\n", con);
1528 con->in_base_pos = 0;
1529 con->in_tag = CEPH_MSGR_TAG_SEQ;
1532 static void prepare_read_tag(struct ceph_connection *con)
1534 dout("prepare_read_tag %p\n", con);
1535 con->in_base_pos = 0;
1536 con->in_tag = CEPH_MSGR_TAG_READY;
1540 * Prepare to read a message.
1542 static int prepare_read_message(struct ceph_connection *con)
1544 dout("prepare_read_message %p\n", con);
1545 BUG_ON(con->in_msg != NULL);
1546 con->in_base_pos = 0;
1547 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1552 static int read_partial(struct ceph_connection *con,
1553 int end, int size, void *object)
1555 while (con->in_base_pos < end) {
1556 int left = end - con->in_base_pos;
1557 int have = size - left;
1558 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1561 con->in_base_pos += ret;
1568 * Read all or part of the connect-side handshake on a new connection
1570 static int read_partial_banner(struct ceph_connection *con)
1576 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1579 size = strlen(CEPH_BANNER);
1581 ret = read_partial(con, end, size, con->in_banner);
1585 size = sizeof (con->actual_peer_addr);
1587 ret = read_partial(con, end, size, &con->actual_peer_addr);
1591 size = sizeof (con->peer_addr_for_me);
1593 ret = read_partial(con, end, size, &con->peer_addr_for_me);
1601 static int read_partial_connect(struct ceph_connection *con)
1607 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1609 size = sizeof (con->in_reply);
1611 ret = read_partial(con, end, size, &con->in_reply);
1615 size = le32_to_cpu(con->in_reply.authorizer_len);
1617 ret = read_partial(con, end, size, con->auth_reply_buf);
1621 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1622 con, (int)con->in_reply.tag,
1623 le32_to_cpu(con->in_reply.connect_seq),
1624 le32_to_cpu(con->in_reply.global_seq));
1631 * Verify the hello banner looks okay.
1633 static int verify_hello(struct ceph_connection *con)
1635 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1636 pr_err("connect to %s got bad banner\n",
1637 ceph_pr_addr(&con->peer_addr.in_addr));
1638 con->error_msg = "protocol error, bad banner";
1644 static bool addr_is_blank(struct sockaddr_storage *ss)
1646 switch (ss->ss_family) {
1648 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1651 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1652 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1653 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1654 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1659 static int addr_port(struct sockaddr_storage *ss)
1661 switch (ss->ss_family) {
1663 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1665 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1670 static void addr_set_port(struct sockaddr_storage *ss, int p)
1672 switch (ss->ss_family) {
1674 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1677 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1683 * Unlike other *_pton function semantics, zero indicates success.
1685 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1686 char delim, const char **ipend)
1688 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1689 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1691 memset(ss, 0, sizeof(*ss));
1693 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1694 ss->ss_family = AF_INET;
1698 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1699 ss->ss_family = AF_INET6;
1707 * Extract hostname string and resolve using kernel DNS facility.
1709 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1710 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1711 struct sockaddr_storage *ss, char delim, const char **ipend)
1713 const char *end, *delim_p;
1714 char *colon_p, *ip_addr = NULL;
1718 * The end of the hostname occurs immediately preceding the delimiter or
1719 * the port marker (':') where the delimiter takes precedence.
1721 delim_p = memchr(name, delim, namelen);
1722 colon_p = memchr(name, ':', namelen);
1724 if (delim_p && colon_p)
1725 end = delim_p < colon_p ? delim_p : colon_p;
1726 else if (!delim_p && colon_p)
1730 if (!end) /* case: hostname:/ */
1731 end = name + namelen;
1737 /* do dns_resolve upcall */
1738 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1740 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1748 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1749 ret, ret ? "failed" : ceph_pr_addr(ss));
1754 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1755 struct sockaddr_storage *ss, char delim, const char **ipend)
1762 * Parse a server name (IP or hostname). If a valid IP address is not found
1763 * then try to extract a hostname to resolve using userspace DNS upcall.
1765 static int ceph_parse_server_name(const char *name, size_t namelen,
1766 struct sockaddr_storage *ss, char delim, const char **ipend)
1770 ret = ceph_pton(name, namelen, ss, delim, ipend);
1772 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1778 * Parse an ip[:port] list into an addr array. Use the default
1779 * monitor port if a port isn't specified.
1781 int ceph_parse_ips(const char *c, const char *end,
1782 struct ceph_entity_addr *addr,
1783 int max_count, int *count)
1785 int i, ret = -EINVAL;
1788 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1789 for (i = 0; i < max_count; i++) {
1791 struct sockaddr_storage *ss = &addr[i].in_addr;
1800 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1809 dout("missing matching ']'\n");
1816 if (p < end && *p == ':') {
1819 while (p < end && *p >= '0' && *p <= '9') {
1820 port = (port * 10) + (*p - '0');
1823 if (port > 65535 || port == 0)
1826 port = CEPH_MON_PORT;
1829 addr_set_port(ss, port);
1831 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1848 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1851 EXPORT_SYMBOL(ceph_parse_ips);
1853 static int process_banner(struct ceph_connection *con)
1855 dout("process_banner on %p\n", con);
1857 if (verify_hello(con) < 0)
1860 ceph_decode_addr(&con->actual_peer_addr);
1861 ceph_decode_addr(&con->peer_addr_for_me);
1864 * Make sure the other end is who we wanted. note that the other
1865 * end may not yet know their ip address, so if it's 0.0.0.0, give
1866 * them the benefit of the doubt.
1868 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1869 sizeof(con->peer_addr)) != 0 &&
1870 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1871 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1872 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1873 ceph_pr_addr(&con->peer_addr.in_addr),
1874 (int)le32_to_cpu(con->peer_addr.nonce),
1875 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1876 (int)le32_to_cpu(con->actual_peer_addr.nonce));
1877 con->error_msg = "wrong peer at address";
1882 * did we learn our address?
1884 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1885 int port = addr_port(&con->msgr->inst.addr.in_addr);
1887 memcpy(&con->msgr->inst.addr.in_addr,
1888 &con->peer_addr_for_me.in_addr,
1889 sizeof(con->peer_addr_for_me.in_addr));
1890 addr_set_port(&con->msgr->inst.addr.in_addr, port);
1891 encode_my_addr(con->msgr);
1892 dout("process_banner learned my addr is %s\n",
1893 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
1899 static int process_connect(struct ceph_connection *con)
1901 u64 sup_feat = con->msgr->supported_features;
1902 u64 req_feat = con->msgr->required_features;
1903 u64 server_feat = le64_to_cpu(con->in_reply.features);
1906 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1908 switch (con->in_reply.tag) {
1909 case CEPH_MSGR_TAG_FEATURES:
1910 pr_err("%s%lld %s feature set mismatch,"
1911 " my %llx < server's %llx, missing %llx\n",
1912 ENTITY_NAME(con->peer_name),
1913 ceph_pr_addr(&con->peer_addr.in_addr),
1914 sup_feat, server_feat, server_feat & ~sup_feat);
1915 con->error_msg = "missing required protocol features";
1916 reset_connection(con);
1919 case CEPH_MSGR_TAG_BADPROTOVER:
1920 pr_err("%s%lld %s protocol version mismatch,"
1921 " my %d != server's %d\n",
1922 ENTITY_NAME(con->peer_name),
1923 ceph_pr_addr(&con->peer_addr.in_addr),
1924 le32_to_cpu(con->out_connect.protocol_version),
1925 le32_to_cpu(con->in_reply.protocol_version));
1926 con->error_msg = "protocol version mismatch";
1927 reset_connection(con);
1930 case CEPH_MSGR_TAG_BADAUTHORIZER:
1932 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1934 if (con->auth_retry == 2) {
1935 con->error_msg = "connect authorization failure";
1938 con_out_kvec_reset(con);
1939 ret = prepare_write_connect(con);
1942 prepare_read_connect(con);
1945 case CEPH_MSGR_TAG_RESETSESSION:
1947 * If we connected with a large connect_seq but the peer
1948 * has no record of a session with us (no connection, or
1949 * connect_seq == 0), they will send RESETSESION to indicate
1950 * that they must have reset their session, and may have
1953 dout("process_connect got RESET peer seq %u\n",
1954 le32_to_cpu(con->in_reply.connect_seq));
1955 pr_err("%s%lld %s connection reset\n",
1956 ENTITY_NAME(con->peer_name),
1957 ceph_pr_addr(&con->peer_addr.in_addr));
1958 reset_connection(con);
1959 con_out_kvec_reset(con);
1960 ret = prepare_write_connect(con);
1963 prepare_read_connect(con);
1965 /* Tell ceph about it. */
1966 mutex_unlock(&con->mutex);
1967 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
1968 if (con->ops->peer_reset)
1969 con->ops->peer_reset(con);
1970 mutex_lock(&con->mutex);
1971 if (con->state != CON_STATE_NEGOTIATING)
1975 case CEPH_MSGR_TAG_RETRY_SESSION:
1977 * If we sent a smaller connect_seq than the peer has, try
1978 * again with a larger value.
1980 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
1981 le32_to_cpu(con->out_connect.connect_seq),
1982 le32_to_cpu(con->in_reply.connect_seq));
1983 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
1984 con_out_kvec_reset(con);
1985 ret = prepare_write_connect(con);
1988 prepare_read_connect(con);
1991 case CEPH_MSGR_TAG_RETRY_GLOBAL:
1993 * If we sent a smaller global_seq than the peer has, try
1994 * again with a larger value.
1996 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1997 con->peer_global_seq,
1998 le32_to_cpu(con->in_reply.global_seq));
1999 get_global_seq(con->msgr,
2000 le32_to_cpu(con->in_reply.global_seq));
2001 con_out_kvec_reset(con);
2002 ret = prepare_write_connect(con);
2005 prepare_read_connect(con);
2008 case CEPH_MSGR_TAG_SEQ:
2009 case CEPH_MSGR_TAG_READY:
2010 if (req_feat & ~server_feat) {
2011 pr_err("%s%lld %s protocol feature mismatch,"
2012 " my required %llx > server's %llx, need %llx\n",
2013 ENTITY_NAME(con->peer_name),
2014 ceph_pr_addr(&con->peer_addr.in_addr),
2015 req_feat, server_feat, req_feat & ~server_feat);
2016 con->error_msg = "missing required protocol features";
2017 reset_connection(con);
2021 WARN_ON(con->state != CON_STATE_NEGOTIATING);
2022 con->state = CON_STATE_OPEN;
2023 con->auth_retry = 0; /* we authenticated; clear flag */
2024 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2026 con->peer_features = server_feat;
2027 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2028 con->peer_global_seq,
2029 le32_to_cpu(con->in_reply.connect_seq),
2031 WARN_ON(con->connect_seq !=
2032 le32_to_cpu(con->in_reply.connect_seq));
2034 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2035 con_flag_set(con, CON_FLAG_LOSSYTX);
2037 con->delay = 0; /* reset backoff memory */
2039 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2040 prepare_write_seq(con);
2041 prepare_read_seq(con);
2043 prepare_read_tag(con);
2047 case CEPH_MSGR_TAG_WAIT:
2049 * If there is a connection race (we are opening
2050 * connections to each other), one of us may just have
2051 * to WAIT. This shouldn't happen if we are the
2054 pr_err("process_connect got WAIT as client\n");
2055 con->error_msg = "protocol error, got WAIT as client";
2059 pr_err("connect protocol error, will retry\n");
2060 con->error_msg = "protocol error, garbage tag during connect";
2068 * read (part of) an ack
2070 static int read_partial_ack(struct ceph_connection *con)
2072 int size = sizeof (con->in_temp_ack);
2075 return read_partial(con, end, size, &con->in_temp_ack);
2079 * We can finally discard anything that's been acked.
2081 static void process_ack(struct ceph_connection *con)
2084 u64 ack = le64_to_cpu(con->in_temp_ack);
2087 while (!list_empty(&con->out_sent)) {
2088 m = list_first_entry(&con->out_sent, struct ceph_msg,
2090 seq = le64_to_cpu(m->hdr.seq);
2093 dout("got ack for seq %llu type %d at %p\n", seq,
2094 le16_to_cpu(m->hdr.type), m);
2095 m->ack_stamp = jiffies;
2098 prepare_read_tag(con);
2102 static int read_partial_message_section(struct ceph_connection *con,
2103 struct kvec *section,
2104 unsigned int sec_len, u32 *crc)
2110 while (section->iov_len < sec_len) {
2111 BUG_ON(section->iov_base == NULL);
2112 left = sec_len - section->iov_len;
2113 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2114 section->iov_len, left);
2117 section->iov_len += ret;
2119 if (section->iov_len == sec_len)
2120 *crc = crc32c(0, section->iov_base, section->iov_len);
2125 static int read_partial_msg_data(struct ceph_connection *con)
2127 struct ceph_msg *msg = con->in_msg;
2128 struct ceph_msg_data_cursor *cursor = &msg->cursor;
2129 const bool do_datacrc = !con->msgr->nocrc;
2137 if (list_empty(&msg->data))
2141 crc = con->in_data_crc;
2142 while (cursor->resid) {
2143 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
2145 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2148 con->in_data_crc = crc;
2154 crc = ceph_crc32c_page(crc, page, page_offset, ret);
2155 (void) ceph_msg_data_advance(&msg->cursor, (size_t)ret);
2158 con->in_data_crc = crc;
2160 return 1; /* must return > 0 to indicate success */
2164 * read (part of) a message.
2166 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2168 static int read_partial_message(struct ceph_connection *con)
2170 struct ceph_msg *m = con->in_msg;
2174 unsigned int front_len, middle_len, data_len;
2175 bool do_datacrc = !con->msgr->nocrc;
2179 dout("read_partial_message con %p msg %p\n", con, m);
2182 size = sizeof (con->in_hdr);
2184 ret = read_partial(con, end, size, &con->in_hdr);
2188 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2189 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2190 pr_err("read_partial_message bad hdr "
2191 " crc %u != expected %u\n",
2192 crc, con->in_hdr.crc);
2196 front_len = le32_to_cpu(con->in_hdr.front_len);
2197 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2199 middle_len = le32_to_cpu(con->in_hdr.middle_len);
2200 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2202 data_len = le32_to_cpu(con->in_hdr.data_len);
2203 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2207 seq = le64_to_cpu(con->in_hdr.seq);
2208 if ((s64)seq - (s64)con->in_seq < 1) {
2209 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2210 ENTITY_NAME(con->peer_name),
2211 ceph_pr_addr(&con->peer_addr.in_addr),
2212 seq, con->in_seq + 1);
2213 con->in_base_pos = -front_len - middle_len - data_len -
2215 con->in_tag = CEPH_MSGR_TAG_READY;
2217 } else if ((s64)seq - (s64)con->in_seq > 1) {
2218 pr_err("read_partial_message bad seq %lld expected %lld\n",
2219 seq, con->in_seq + 1);
2220 con->error_msg = "bad message sequence # for incoming message";
2224 /* allocate message? */
2228 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2229 front_len, data_len);
2230 ret = ceph_con_in_msg_alloc(con, &skip);
2234 BUG_ON(!con->in_msg ^ skip);
2235 if (con->in_msg && data_len > con->in_msg->data_length) {
2236 pr_warning("%s skipping long message (%u > %zd)\n",
2237 __func__, data_len, con->in_msg->data_length);
2238 ceph_msg_put(con->in_msg);
2243 /* skip this message */
2244 dout("alloc_msg said skip message\n");
2245 con->in_base_pos = -front_len - middle_len - data_len -
2247 con->in_tag = CEPH_MSGR_TAG_READY;
2252 BUG_ON(!con->in_msg);
2253 BUG_ON(con->in_msg->con != con);
2255 m->front.iov_len = 0; /* haven't read it yet */
2257 m->middle->vec.iov_len = 0;
2259 /* prepare for data payload, if any */
2262 prepare_message_data(con->in_msg, data_len);
2266 ret = read_partial_message_section(con, &m->front, front_len,
2267 &con->in_front_crc);
2273 ret = read_partial_message_section(con, &m->middle->vec,
2275 &con->in_middle_crc);
2282 ret = read_partial_msg_data(con);
2288 size = sizeof (m->footer);
2290 ret = read_partial(con, end, size, &m->footer);
2294 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2295 m, front_len, m->footer.front_crc, middle_len,
2296 m->footer.middle_crc, data_len, m->footer.data_crc);
2299 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2300 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2301 m, con->in_front_crc, m->footer.front_crc);
2304 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2305 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2306 m, con->in_middle_crc, m->footer.middle_crc);
2310 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2311 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2312 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2313 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2317 return 1; /* done! */
2321 * Process message. This happens in the worker thread. The callback should
2322 * be careful not to do anything that waits on other incoming messages or it
2325 static void process_message(struct ceph_connection *con)
2327 struct ceph_msg *msg;
2329 BUG_ON(con->in_msg->con != con);
2330 con->in_msg->con = NULL;
2335 /* if first message, set peer_name */
2336 if (con->peer_name.type == 0)
2337 con->peer_name = msg->hdr.src;
2340 mutex_unlock(&con->mutex);
2342 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2343 msg, le64_to_cpu(msg->hdr.seq),
2344 ENTITY_NAME(msg->hdr.src),
2345 le16_to_cpu(msg->hdr.type),
2346 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2347 le32_to_cpu(msg->hdr.front_len),
2348 le32_to_cpu(msg->hdr.data_len),
2349 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2350 con->ops->dispatch(con, msg);
2352 mutex_lock(&con->mutex);
2357 * Write something to the socket. Called in a worker thread when the
2358 * socket appears to be writeable and we have something ready to send.
2360 static int try_write(struct ceph_connection *con)
2364 dout("try_write start %p state %lu\n", con, con->state);
2367 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2369 /* open the socket first? */
2370 if (con->state == CON_STATE_PREOPEN) {
2372 con->state = CON_STATE_CONNECTING;
2374 con_out_kvec_reset(con);
2375 prepare_write_banner(con);
2376 prepare_read_banner(con);
2378 BUG_ON(con->in_msg);
2379 con->in_tag = CEPH_MSGR_TAG_READY;
2380 dout("try_write initiating connect on %p new state %lu\n",
2382 ret = ceph_tcp_connect(con);
2384 con->error_msg = "connect error";
2390 /* kvec data queued? */
2391 if (con->out_skip) {
2392 ret = write_partial_skip(con);
2396 if (con->out_kvec_left) {
2397 ret = write_partial_kvec(con);
2404 if (con->out_msg_done) {
2405 ceph_msg_put(con->out_msg);
2406 con->out_msg = NULL; /* we're done with this one */
2410 ret = write_partial_message_data(con);
2412 goto more_kvec; /* we need to send the footer, too! */
2416 dout("try_write write_partial_message_data err %d\n",
2423 if (con->state == CON_STATE_OPEN) {
2424 /* is anything else pending? */
2425 if (!list_empty(&con->out_queue)) {
2426 prepare_write_message(con);
2429 if (con->in_seq > con->in_seq_acked) {
2430 prepare_write_ack(con);
2433 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2434 prepare_write_keepalive(con);
2439 /* Nothing to do! */
2440 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2441 dout("try_write nothing else to write.\n");
2444 dout("try_write done on %p ret %d\n", con, ret);
2451 * Read what we can from the socket.
2453 static int try_read(struct ceph_connection *con)
2458 dout("try_read start on %p state %lu\n", con, con->state);
2459 if (con->state != CON_STATE_CONNECTING &&
2460 con->state != CON_STATE_NEGOTIATING &&
2461 con->state != CON_STATE_OPEN)
2466 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2469 if (con->state == CON_STATE_CONNECTING) {
2470 dout("try_read connecting\n");
2471 ret = read_partial_banner(con);
2474 ret = process_banner(con);
2478 con->state = CON_STATE_NEGOTIATING;
2481 * Received banner is good, exchange connection info.
2482 * Do not reset out_kvec, as sending our banner raced
2483 * with receiving peer banner after connect completed.
2485 ret = prepare_write_connect(con);
2488 prepare_read_connect(con);
2490 /* Send connection info before awaiting response */
2494 if (con->state == CON_STATE_NEGOTIATING) {
2495 dout("try_read negotiating\n");
2496 ret = read_partial_connect(con);
2499 ret = process_connect(con);
2505 WARN_ON(con->state != CON_STATE_OPEN);
2507 if (con->in_base_pos < 0) {
2509 * skipping + discarding content.
2511 * FIXME: there must be a better way to do this!
2513 static char buf[SKIP_BUF_SIZE];
2514 int skip = min((int) sizeof (buf), -con->in_base_pos);
2516 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2517 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2520 con->in_base_pos += ret;
2521 if (con->in_base_pos)
2524 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2528 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2531 dout("try_read got tag %d\n", (int)con->in_tag);
2532 switch (con->in_tag) {
2533 case CEPH_MSGR_TAG_MSG:
2534 prepare_read_message(con);
2536 case CEPH_MSGR_TAG_ACK:
2537 prepare_read_ack(con);
2539 case CEPH_MSGR_TAG_CLOSE:
2540 con_close_socket(con);
2541 con->state = CON_STATE_CLOSED;
2547 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2548 ret = read_partial_message(con);
2552 con->error_msg = "bad crc";
2556 con->error_msg = "io error";
2561 if (con->in_tag == CEPH_MSGR_TAG_READY)
2563 process_message(con);
2564 if (con->state == CON_STATE_OPEN)
2565 prepare_read_tag(con);
2568 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2569 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2571 * the final handshake seq exchange is semantically
2572 * equivalent to an ACK
2574 ret = read_partial_ack(con);
2582 dout("try_read done on %p ret %d\n", con, ret);
2586 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2587 con->error_msg = "protocol error, garbage tag";
2594 * Atomically queue work on a connection after the specified delay.
2595 * Bump @con reference to avoid races with connection teardown.
2596 * Returns 0 if work was queued, or an error code otherwise.
2598 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2600 if (!con->ops->get(con)) {
2601 dout("%s %p ref count 0\n", __func__, con);
2606 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2607 dout("%s %p - already queued\n", __func__, con);
2613 dout("%s %p %lu\n", __func__, con, delay);
2618 static void queue_con(struct ceph_connection *con)
2620 (void) queue_con_delay(con, 0);
2623 static bool con_sock_closed(struct ceph_connection *con)
2625 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2629 case CON_STATE_ ## x: \
2630 con->error_msg = "socket closed (con state " #x ")"; \
2633 switch (con->state) {
2641 pr_warning("%s con %p unrecognized state %lu\n",
2642 __func__, con, con->state);
2643 con->error_msg = "unrecognized con state";
2652 static bool con_backoff(struct ceph_connection *con)
2656 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2659 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2661 dout("%s: con %p FAILED to back off %lu\n", __func__,
2663 BUG_ON(ret == -ENOENT);
2664 con_flag_set(con, CON_FLAG_BACKOFF);
2670 /* Finish fault handling; con->mutex must *not* be held here */
2672 static void con_fault_finish(struct ceph_connection *con)
2675 * in case we faulted due to authentication, invalidate our
2676 * current tickets so that we can get new ones.
2678 if (con->auth_retry && con->ops->invalidate_authorizer) {
2679 dout("calling invalidate_authorizer()\n");
2680 con->ops->invalidate_authorizer(con);
2683 if (con->ops->fault)
2684 con->ops->fault(con);
2688 * Do some work on a connection. Drop a connection ref when we're done.
2690 static void con_work(struct work_struct *work)
2692 struct ceph_connection *con = container_of(work, struct ceph_connection,
2696 mutex_lock(&con->mutex);
2700 if ((fault = con_sock_closed(con))) {
2701 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2704 if (con_backoff(con)) {
2705 dout("%s: con %p BACKOFF\n", __func__, con);
2708 if (con->state == CON_STATE_STANDBY) {
2709 dout("%s: con %p STANDBY\n", __func__, con);
2712 if (con->state == CON_STATE_CLOSED) {
2713 dout("%s: con %p CLOSED\n", __func__, con);
2717 if (con->state == CON_STATE_PREOPEN) {
2718 dout("%s: con %p PREOPEN\n", __func__, con);
2722 ret = try_read(con);
2726 con->error_msg = "socket error on read";
2731 ret = try_write(con);
2735 con->error_msg = "socket error on write";
2739 break; /* If we make it to here, we're done */
2743 mutex_unlock(&con->mutex);
2746 con_fault_finish(con);
2752 * Generic error/fault handler. A retry mechanism is used with
2753 * exponential backoff
2755 static void con_fault(struct ceph_connection *con)
2757 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2758 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2759 dout("fault %p state %lu to peer %s\n",
2760 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2762 WARN_ON(con->state != CON_STATE_CONNECTING &&
2763 con->state != CON_STATE_NEGOTIATING &&
2764 con->state != CON_STATE_OPEN);
2766 con_close_socket(con);
2768 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2769 dout("fault on LOSSYTX channel, marking CLOSED\n");
2770 con->state = CON_STATE_CLOSED;
2775 BUG_ON(con->in_msg->con != con);
2776 con->in_msg->con = NULL;
2777 ceph_msg_put(con->in_msg);
2782 /* Requeue anything that hasn't been acked */
2783 list_splice_init(&con->out_sent, &con->out_queue);
2785 /* If there are no messages queued or keepalive pending, place
2786 * the connection in a STANDBY state */
2787 if (list_empty(&con->out_queue) &&
2788 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
2789 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2790 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2791 con->state = CON_STATE_STANDBY;
2793 /* retry after a delay. */
2794 con->state = CON_STATE_PREOPEN;
2795 if (con->delay == 0)
2796 con->delay = BASE_DELAY_INTERVAL;
2797 else if (con->delay < MAX_DELAY_INTERVAL)
2799 con_flag_set(con, CON_FLAG_BACKOFF);
2807 * initialize a new messenger instance
2809 void ceph_messenger_init(struct ceph_messenger *msgr,
2810 struct ceph_entity_addr *myaddr,
2811 u32 supported_features,
2812 u32 required_features,
2815 msgr->supported_features = supported_features;
2816 msgr->required_features = required_features;
2818 spin_lock_init(&msgr->global_seq_lock);
2821 msgr->inst.addr = *myaddr;
2823 /* select a random nonce */
2824 msgr->inst.addr.type = 0;
2825 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2826 encode_my_addr(msgr);
2827 msgr->nocrc = nocrc;
2829 atomic_set(&msgr->stopping, 0);
2831 dout("%s %p\n", __func__, msgr);
2833 EXPORT_SYMBOL(ceph_messenger_init);
2835 static void clear_standby(struct ceph_connection *con)
2837 /* come back from STANDBY? */
2838 if (con->state == CON_STATE_STANDBY) {
2839 dout("clear_standby %p and ++connect_seq\n", con);
2840 con->state = CON_STATE_PREOPEN;
2842 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2843 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
2848 * Queue up an outgoing message on the given connection.
2850 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2853 msg->hdr.src = con->msgr->inst.name;
2854 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2855 msg->needs_out_seq = true;
2857 mutex_lock(&con->mutex);
2859 if (con->state == CON_STATE_CLOSED) {
2860 dout("con_send %p closed, dropping %p\n", con, msg);
2862 mutex_unlock(&con->mutex);
2866 BUG_ON(msg->con != NULL);
2867 msg->con = con->ops->get(con);
2868 BUG_ON(msg->con == NULL);
2870 BUG_ON(!list_empty(&msg->list_head));
2871 list_add_tail(&msg->list_head, &con->out_queue);
2872 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2873 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2874 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2875 le32_to_cpu(msg->hdr.front_len),
2876 le32_to_cpu(msg->hdr.middle_len),
2877 le32_to_cpu(msg->hdr.data_len));
2880 mutex_unlock(&con->mutex);
2882 /* if there wasn't anything waiting to send before, queue
2884 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
2887 EXPORT_SYMBOL(ceph_con_send);
2890 * Revoke a message that was previously queued for send
2892 void ceph_msg_revoke(struct ceph_msg *msg)
2894 struct ceph_connection *con = msg->con;
2897 return; /* Message not in our possession */
2899 mutex_lock(&con->mutex);
2900 if (!list_empty(&msg->list_head)) {
2901 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
2902 list_del_init(&msg->list_head);
2903 BUG_ON(msg->con == NULL);
2904 msg->con->ops->put(msg->con);
2910 if (con->out_msg == msg) {
2911 dout("%s %p msg %p - was sending\n", __func__, con, msg);
2912 con->out_msg = NULL;
2913 if (con->out_kvec_is_msg) {
2914 con->out_skip = con->out_kvec_bytes;
2915 con->out_kvec_is_msg = false;
2921 mutex_unlock(&con->mutex);
2925 * Revoke a message that we may be reading data into
2927 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
2929 struct ceph_connection *con;
2931 BUG_ON(msg == NULL);
2933 dout("%s msg %p null con\n", __func__, msg);
2935 return; /* Message not in our possession */
2939 mutex_lock(&con->mutex);
2940 if (con->in_msg == msg) {
2941 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
2942 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
2943 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
2945 /* skip rest of message */
2946 dout("%s %p msg %p revoked\n", __func__, con, msg);
2947 con->in_base_pos = con->in_base_pos -
2948 sizeof(struct ceph_msg_header) -
2952 sizeof(struct ceph_msg_footer);
2953 ceph_msg_put(con->in_msg);
2955 con->in_tag = CEPH_MSGR_TAG_READY;
2958 dout("%s %p in_msg %p msg %p no-op\n",
2959 __func__, con, con->in_msg, msg);
2961 mutex_unlock(&con->mutex);
2965 * Queue a keepalive byte to ensure the tcp connection is alive.
2967 void ceph_con_keepalive(struct ceph_connection *con)
2969 dout("con_keepalive %p\n", con);
2970 mutex_lock(&con->mutex);
2972 mutex_unlock(&con->mutex);
2973 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
2974 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
2977 EXPORT_SYMBOL(ceph_con_keepalive);
2979 static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
2981 struct ceph_msg_data *data;
2983 if (WARN_ON(!ceph_msg_data_type_valid(type)))
2986 data = kzalloc(sizeof (*data), GFP_NOFS);
2989 INIT_LIST_HEAD(&data->links);
2994 static void ceph_msg_data_destroy(struct ceph_msg_data *data)
2999 WARN_ON(!list_empty(&data->links));
3000 if (data->type == CEPH_MSG_DATA_PAGELIST) {
3001 ceph_pagelist_release(data->pagelist);
3002 kfree(data->pagelist);
3007 void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
3008 size_t length, size_t alignment)
3010 struct ceph_msg_data *data;
3015 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3017 data->pages = pages;
3018 data->length = length;
3019 data->alignment = alignment & ~PAGE_MASK;
3021 list_add_tail(&data->links, &msg->data);
3022 msg->data_length += length;
3024 EXPORT_SYMBOL(ceph_msg_data_add_pages);
3026 void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
3027 struct ceph_pagelist *pagelist)
3029 struct ceph_msg_data *data;
3032 BUG_ON(!pagelist->length);
3034 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3036 data->pagelist = pagelist;
3038 list_add_tail(&data->links, &msg->data);
3039 msg->data_length += pagelist->length;
3041 EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
3044 void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
3047 struct ceph_msg_data *data;
3051 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3054 data->bio_length = length;
3056 list_add_tail(&data->links, &msg->data);
3057 msg->data_length += length;
3059 EXPORT_SYMBOL(ceph_msg_data_add_bio);
3060 #endif /* CONFIG_BLOCK */
3063 * construct a new message with given type, size
3064 * the new msg has a ref count of 1.
3066 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3071 m = kzalloc(sizeof(*m), flags);
3075 m->hdr.type = cpu_to_le16(type);
3076 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3077 m->hdr.front_len = cpu_to_le32(front_len);
3079 INIT_LIST_HEAD(&m->list_head);
3080 kref_init(&m->kref);
3081 INIT_LIST_HEAD(&m->data);
3084 m->front_max = front_len;
3086 if (front_len > PAGE_CACHE_SIZE) {
3087 m->front.iov_base = __vmalloc(front_len, flags,
3089 m->front_is_vmalloc = true;
3091 m->front.iov_base = kmalloc(front_len, flags);
3093 if (m->front.iov_base == NULL) {
3094 dout("ceph_msg_new can't allocate %d bytes\n",
3099 m->front.iov_base = NULL;
3101 m->front.iov_len = front_len;
3103 dout("ceph_msg_new %p front %d\n", m, front_len);
3110 pr_err("msg_new can't create type %d front %d\n", type,
3114 dout("msg_new can't create type %d front %d\n", type,
3119 EXPORT_SYMBOL(ceph_msg_new);
3122 * Allocate "middle" portion of a message, if it is needed and wasn't
3123 * allocated by alloc_msg. This allows us to read a small fixed-size
3124 * per-type header in the front and then gracefully fail (i.e.,
3125 * propagate the error to the caller based on info in the front) when
3126 * the middle is too large.
3128 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3130 int type = le16_to_cpu(msg->hdr.type);
3131 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3133 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3134 ceph_msg_type_name(type), middle_len);
3135 BUG_ON(!middle_len);
3136 BUG_ON(msg->middle);
3138 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3145 * Allocate a message for receiving an incoming message on a
3146 * connection, and save the result in con->in_msg. Uses the
3147 * connection's private alloc_msg op if available.
3149 * Returns 0 on success, or a negative error code.
3151 * On success, if we set *skip = 1:
3152 * - the next message should be skipped and ignored.
3153 * - con->in_msg == NULL
3154 * or if we set *skip = 0:
3155 * - con->in_msg is non-null.
3156 * On error (ENOMEM, EAGAIN, ...),
3157 * - con->in_msg == NULL
3159 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3161 struct ceph_msg_header *hdr = &con->in_hdr;
3162 int middle_len = le32_to_cpu(hdr->middle_len);
3163 struct ceph_msg *msg;
3166 BUG_ON(con->in_msg != NULL);
3167 BUG_ON(!con->ops->alloc_msg);
3169 mutex_unlock(&con->mutex);
3170 msg = con->ops->alloc_msg(con, hdr, skip);
3171 mutex_lock(&con->mutex);
3172 if (con->state != CON_STATE_OPEN) {
3180 con->in_msg->con = con->ops->get(con);
3181 BUG_ON(con->in_msg->con == NULL);
3184 * Null message pointer means either we should skip
3185 * this message or we couldn't allocate memory. The
3186 * former is not an error.
3190 con->error_msg = "error allocating memory for incoming message";
3194 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3196 if (middle_len && !con->in_msg->middle) {
3197 ret = ceph_alloc_middle(con, con->in_msg);
3199 ceph_msg_put(con->in_msg);
3209 * Free a generically kmalloc'd message.
3211 void ceph_msg_kfree(struct ceph_msg *m)
3213 dout("msg_kfree %p\n", m);
3214 if (m->front_is_vmalloc)
3215 vfree(m->front.iov_base);
3217 kfree(m->front.iov_base);
3222 * Drop a msg ref. Destroy as needed.
3224 void ceph_msg_last_put(struct kref *kref)
3226 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3228 struct list_head *links;
3229 struct list_head *next;
3231 dout("ceph_msg_put last one on %p\n", m);
3232 WARN_ON(!list_empty(&m->list_head));
3234 /* drop middle, data, if any */
3236 ceph_buffer_put(m->middle);
3240 list_splice_init(&m->data, &data);
3241 list_for_each_safe(links, next, &data) {
3242 struct ceph_msg_data *data;
3244 data = list_entry(links, struct ceph_msg_data, links);
3245 list_del_init(links);
3246 ceph_msg_data_destroy(data);
3251 ceph_msgpool_put(m->pool, m);
3255 EXPORT_SYMBOL(ceph_msg_last_put);
3257 void ceph_msg_dump(struct ceph_msg *msg)
3259 pr_debug("msg_dump %p (front_max %d length %zd)\n", msg,
3260 msg->front_max, msg->data_length);
3261 print_hex_dump(KERN_DEBUG, "header: ",
3262 DUMP_PREFIX_OFFSET, 16, 1,
3263 &msg->hdr, sizeof(msg->hdr), true);
3264 print_hex_dump(KERN_DEBUG, " front: ",
3265 DUMP_PREFIX_OFFSET, 16, 1,
3266 msg->front.iov_base, msg->front.iov_len, true);
3268 print_hex_dump(KERN_DEBUG, "middle: ",
3269 DUMP_PREFIX_OFFSET, 16, 1,
3270 msg->middle->vec.iov_base,
3271 msg->middle->vec.iov_len, true);
3272 print_hex_dump(KERN_DEBUG, "footer: ",
3273 DUMP_PREFIX_OFFSET, 16, 1,
3274 &msg->footer, sizeof(msg->footer), true);
3276 EXPORT_SYMBOL(ceph_msg_dump);