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/ceph_features.h>
19 #include <linux/ceph/libceph.h>
20 #include <linux/ceph/messenger.h>
21 #include <linux/ceph/decode.h>
22 #include <linux/ceph/pagelist.h>
23 #include <linux/export.h>
25 #define list_entry_next(pos, member) \
26 list_entry(pos->member.next, typeof(*pos), member)
29 * Ceph uses the messenger to exchange ceph_msg messages with other
30 * hosts in the system. The messenger provides ordered and reliable
31 * delivery. We tolerate TCP disconnects by reconnecting (with
32 * exponential backoff) in the case of a fault (disconnection, bad
33 * crc, protocol error). Acks allow sent messages to be discarded by
38 * We track the state of the socket on a given connection using
39 * values defined below. The transition to a new socket state is
40 * handled by a function which verifies we aren't coming from an
44 * | NEW* | transient initial state
46 * | con_sock_state_init()
49 * | CLOSED | initialized, but no socket (and no
50 * ---------- TCP connection)
52 * | \ con_sock_state_connecting()
53 * | ----------------------
55 * + con_sock_state_closed() \
56 * |+--------------------------- \
59 * | | CLOSING | socket event; \ \
60 * | ----------- await close \ \
63 * | + con_sock_state_closing() \ |
65 * | / --------------- | |
68 * | / -----------------| CONNECTING | socket created, TCP
69 * | | / -------------- connect initiated
70 * | | | con_sock_state_connected()
73 * | CONNECTED | TCP connection established
76 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
79 #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
80 #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
81 #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
82 #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
83 #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
88 #define CON_STATE_CLOSED 1 /* -> PREOPEN */
89 #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
90 #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
91 #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
92 #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
93 #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
96 * ceph_connection flag bits
98 #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
99 * messages on errors */
100 #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
101 #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
102 #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
103 #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
105 static bool con_flag_valid(unsigned long con_flag)
108 case CON_FLAG_LOSSYTX:
109 case CON_FLAG_KEEPALIVE_PENDING:
110 case CON_FLAG_WRITE_PENDING:
111 case CON_FLAG_SOCK_CLOSED:
112 case CON_FLAG_BACKOFF:
119 static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
121 BUG_ON(!con_flag_valid(con_flag));
123 clear_bit(con_flag, &con->flags);
126 static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
128 BUG_ON(!con_flag_valid(con_flag));
130 set_bit(con_flag, &con->flags);
133 static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
135 BUG_ON(!con_flag_valid(con_flag));
137 return test_bit(con_flag, &con->flags);
140 static bool con_flag_test_and_clear(struct ceph_connection *con,
141 unsigned long con_flag)
143 BUG_ON(!con_flag_valid(con_flag));
145 return test_and_clear_bit(con_flag, &con->flags);
148 static bool con_flag_test_and_set(struct ceph_connection *con,
149 unsigned long con_flag)
151 BUG_ON(!con_flag_valid(con_flag));
153 return test_and_set_bit(con_flag, &con->flags);
156 /* Slab caches for frequently-allocated structures */
158 static struct kmem_cache *ceph_msg_cache;
159 static struct kmem_cache *ceph_msg_data_cache;
161 /* static tag bytes (protocol control messages) */
162 static char tag_msg = CEPH_MSGR_TAG_MSG;
163 static char tag_ack = CEPH_MSGR_TAG_ACK;
164 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
166 #ifdef CONFIG_LOCKDEP
167 static struct lock_class_key socket_class;
171 * When skipping (ignoring) a block of input we read it into a "skip
172 * buffer," which is this many bytes in size.
174 #define SKIP_BUF_SIZE 1024
176 static void queue_con(struct ceph_connection *con);
177 static void cancel_con(struct ceph_connection *con);
178 static void con_work(struct work_struct *);
179 static void con_fault(struct ceph_connection *con);
182 * Nicely render a sockaddr as a string. An array of formatted
183 * strings is used, to approximate reentrancy.
185 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
186 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
187 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
188 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
190 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
191 static atomic_t addr_str_seq = ATOMIC_INIT(0);
193 static struct page *zero_page; /* used in certain error cases */
195 const char *ceph_pr_addr(const struct sockaddr_storage *ss)
199 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
200 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
202 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
205 switch (ss->ss_family) {
207 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
208 ntohs(in4->sin_port));
212 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
213 ntohs(in6->sin6_port));
217 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
223 EXPORT_SYMBOL(ceph_pr_addr);
225 static void encode_my_addr(struct ceph_messenger *msgr)
227 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
228 ceph_encode_addr(&msgr->my_enc_addr);
232 * work queue for all reading and writing to/from the socket.
234 static struct workqueue_struct *ceph_msgr_wq;
236 static int ceph_msgr_slab_init(void)
238 BUG_ON(ceph_msg_cache);
239 ceph_msg_cache = kmem_cache_create("ceph_msg",
240 sizeof (struct ceph_msg),
241 __alignof__(struct ceph_msg), 0, NULL);
246 BUG_ON(ceph_msg_data_cache);
247 ceph_msg_data_cache = kmem_cache_create("ceph_msg_data",
248 sizeof (struct ceph_msg_data),
249 __alignof__(struct ceph_msg_data),
251 if (ceph_msg_data_cache)
254 kmem_cache_destroy(ceph_msg_cache);
255 ceph_msg_cache = NULL;
260 static void ceph_msgr_slab_exit(void)
262 BUG_ON(!ceph_msg_data_cache);
263 kmem_cache_destroy(ceph_msg_data_cache);
264 ceph_msg_data_cache = NULL;
266 BUG_ON(!ceph_msg_cache);
267 kmem_cache_destroy(ceph_msg_cache);
268 ceph_msg_cache = NULL;
271 static void _ceph_msgr_exit(void)
274 destroy_workqueue(ceph_msgr_wq);
278 ceph_msgr_slab_exit();
280 BUG_ON(zero_page == NULL);
281 page_cache_release(zero_page);
285 int ceph_msgr_init(void)
287 BUG_ON(zero_page != NULL);
288 zero_page = ZERO_PAGE(0);
289 page_cache_get(zero_page);
291 if (ceph_msgr_slab_init())
295 * The number of active work items is limited by the number of
296 * connections, so leave @max_active at default.
298 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
302 pr_err("msgr_init failed to create workqueue\n");
307 EXPORT_SYMBOL(ceph_msgr_init);
309 void ceph_msgr_exit(void)
311 BUG_ON(ceph_msgr_wq == NULL);
315 EXPORT_SYMBOL(ceph_msgr_exit);
317 void ceph_msgr_flush(void)
319 flush_workqueue(ceph_msgr_wq);
321 EXPORT_SYMBOL(ceph_msgr_flush);
323 /* Connection socket state transition functions */
325 static void con_sock_state_init(struct ceph_connection *con)
329 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
330 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
331 printk("%s: unexpected old state %d\n", __func__, old_state);
332 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
333 CON_SOCK_STATE_CLOSED);
336 static void con_sock_state_connecting(struct ceph_connection *con)
340 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
341 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
342 printk("%s: unexpected old state %d\n", __func__, old_state);
343 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
344 CON_SOCK_STATE_CONNECTING);
347 static void con_sock_state_connected(struct ceph_connection *con)
351 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
352 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
353 printk("%s: unexpected old state %d\n", __func__, old_state);
354 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
355 CON_SOCK_STATE_CONNECTED);
358 static void con_sock_state_closing(struct ceph_connection *con)
362 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
363 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
364 old_state != CON_SOCK_STATE_CONNECTED &&
365 old_state != CON_SOCK_STATE_CLOSING))
366 printk("%s: unexpected old state %d\n", __func__, old_state);
367 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
368 CON_SOCK_STATE_CLOSING);
371 static void con_sock_state_closed(struct ceph_connection *con)
375 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
376 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
377 old_state != CON_SOCK_STATE_CLOSING &&
378 old_state != CON_SOCK_STATE_CONNECTING &&
379 old_state != CON_SOCK_STATE_CLOSED))
380 printk("%s: unexpected old state %d\n", __func__, old_state);
381 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
382 CON_SOCK_STATE_CLOSED);
386 * socket callback functions
389 /* data available on socket, or listen socket received a connect */
390 static void ceph_sock_data_ready(struct sock *sk)
392 struct ceph_connection *con = sk->sk_user_data;
393 if (atomic_read(&con->msgr->stopping)) {
397 if (sk->sk_state != TCP_CLOSE_WAIT) {
398 dout("%s on %p state = %lu, queueing work\n", __func__,
404 /* socket has buffer space for writing */
405 static void ceph_sock_write_space(struct sock *sk)
407 struct ceph_connection *con = sk->sk_user_data;
409 /* only queue to workqueue if there is data we want to write,
410 * and there is sufficient space in the socket buffer to accept
411 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
412 * doesn't get called again until try_write() fills the socket
413 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
414 * and net/core/stream.c:sk_stream_write_space().
416 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
417 if (sk_stream_is_writeable(sk)) {
418 dout("%s %p queueing write work\n", __func__, con);
419 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
423 dout("%s %p nothing to write\n", __func__, con);
427 /* socket's state has changed */
428 static void ceph_sock_state_change(struct sock *sk)
430 struct ceph_connection *con = sk->sk_user_data;
432 dout("%s %p state = %lu sk_state = %u\n", __func__,
433 con, con->state, sk->sk_state);
435 switch (sk->sk_state) {
437 dout("%s TCP_CLOSE\n", __func__);
439 dout("%s TCP_CLOSE_WAIT\n", __func__);
440 con_sock_state_closing(con);
441 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
444 case TCP_ESTABLISHED:
445 dout("%s TCP_ESTABLISHED\n", __func__);
446 con_sock_state_connected(con);
449 default: /* Everything else is uninteresting */
455 * set up socket callbacks
457 static void set_sock_callbacks(struct socket *sock,
458 struct ceph_connection *con)
460 struct sock *sk = sock->sk;
461 sk->sk_user_data = con;
462 sk->sk_data_ready = ceph_sock_data_ready;
463 sk->sk_write_space = ceph_sock_write_space;
464 sk->sk_state_change = ceph_sock_state_change;
473 * initiate connection to a remote socket.
475 static int ceph_tcp_connect(struct ceph_connection *con)
477 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
482 ret = sock_create_kern(&init_net, con->peer_addr.in_addr.ss_family,
483 SOCK_STREAM, IPPROTO_TCP, &sock);
486 sock->sk->sk_allocation = GFP_NOFS;
488 #ifdef CONFIG_LOCKDEP
489 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
492 set_sock_callbacks(sock, con);
494 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
496 con_sock_state_connecting(con);
497 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
499 if (ret == -EINPROGRESS) {
500 dout("connect %s EINPROGRESS sk_state = %u\n",
501 ceph_pr_addr(&con->peer_addr.in_addr),
503 } else if (ret < 0) {
504 pr_err("connect %s error %d\n",
505 ceph_pr_addr(&con->peer_addr.in_addr), ret);
510 if (con->msgr->tcp_nodelay) {
513 ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
514 (char *)&optval, sizeof(optval));
516 pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
524 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
526 struct kvec iov = {buf, len};
527 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
530 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
536 static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
537 int page_offset, size_t length)
542 BUG_ON(page_offset + length > PAGE_SIZE);
546 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
553 * write something. @more is true if caller will be sending more data
556 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
557 size_t kvlen, size_t len, int more)
559 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
563 msg.msg_flags |= MSG_MORE;
565 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
567 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
573 static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
574 int offset, size_t size, bool more)
576 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
579 ret = kernel_sendpage(sock, page, offset, size, flags);
586 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
587 int offset, size_t size, bool more)
592 /* sendpage cannot properly handle pages with page_count == 0,
593 * we need to fallback to sendmsg if that's the case */
594 if (page_count(page) >= 1)
595 return __ceph_tcp_sendpage(sock, page, offset, size, more);
597 iov.iov_base = kmap(page) + offset;
599 ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more);
606 * Shutdown/close the socket for the given connection.
608 static int con_close_socket(struct ceph_connection *con)
612 dout("con_close_socket on %p sock %p\n", con, con->sock);
614 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
615 sock_release(con->sock);
620 * Forcibly clear the SOCK_CLOSED flag. It gets set
621 * independent of the connection mutex, and we could have
622 * received a socket close event before we had the chance to
623 * shut the socket down.
625 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
627 con_sock_state_closed(con);
632 * Reset a connection. Discard all incoming and outgoing messages
633 * and clear *_seq state.
635 static void ceph_msg_remove(struct ceph_msg *msg)
637 list_del_init(&msg->list_head);
638 BUG_ON(msg->con == NULL);
639 msg->con->ops->put(msg->con);
644 static void ceph_msg_remove_list(struct list_head *head)
646 while (!list_empty(head)) {
647 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
649 ceph_msg_remove(msg);
653 static void reset_connection(struct ceph_connection *con)
655 /* reset connection, out_queue, msg_ and connect_seq */
656 /* discard existing out_queue and msg_seq */
657 dout("reset_connection %p\n", con);
658 ceph_msg_remove_list(&con->out_queue);
659 ceph_msg_remove_list(&con->out_sent);
662 BUG_ON(con->in_msg->con != con);
663 con->in_msg->con = NULL;
664 ceph_msg_put(con->in_msg);
669 con->connect_seq = 0;
672 ceph_msg_put(con->out_msg);
676 con->in_seq_acked = 0;
680 * mark a peer down. drop any open connections.
682 void ceph_con_close(struct ceph_connection *con)
684 mutex_lock(&con->mutex);
685 dout("con_close %p peer %s\n", con,
686 ceph_pr_addr(&con->peer_addr.in_addr));
687 con->state = CON_STATE_CLOSED;
689 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
690 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
691 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
692 con_flag_clear(con, CON_FLAG_BACKOFF);
694 reset_connection(con);
695 con->peer_global_seq = 0;
697 con_close_socket(con);
698 mutex_unlock(&con->mutex);
700 EXPORT_SYMBOL(ceph_con_close);
703 * Reopen a closed connection, with a new peer address.
705 void ceph_con_open(struct ceph_connection *con,
706 __u8 entity_type, __u64 entity_num,
707 struct ceph_entity_addr *addr)
709 mutex_lock(&con->mutex);
710 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
712 WARN_ON(con->state != CON_STATE_CLOSED);
713 con->state = CON_STATE_PREOPEN;
715 con->peer_name.type = (__u8) entity_type;
716 con->peer_name.num = cpu_to_le64(entity_num);
718 memcpy(&con->peer_addr, addr, sizeof(*addr));
719 con->delay = 0; /* reset backoff memory */
720 mutex_unlock(&con->mutex);
723 EXPORT_SYMBOL(ceph_con_open);
726 * return true if this connection ever successfully opened
728 bool ceph_con_opened(struct ceph_connection *con)
730 return con->connect_seq > 0;
734 * initialize a new connection.
736 void ceph_con_init(struct ceph_connection *con, void *private,
737 const struct ceph_connection_operations *ops,
738 struct ceph_messenger *msgr)
740 dout("con_init %p\n", con);
741 memset(con, 0, sizeof(*con));
742 con->private = private;
746 con_sock_state_init(con);
748 mutex_init(&con->mutex);
749 INIT_LIST_HEAD(&con->out_queue);
750 INIT_LIST_HEAD(&con->out_sent);
751 INIT_DELAYED_WORK(&con->work, con_work);
753 con->state = CON_STATE_CLOSED;
755 EXPORT_SYMBOL(ceph_con_init);
759 * We maintain a global counter to order connection attempts. Get
760 * a unique seq greater than @gt.
762 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
766 spin_lock(&msgr->global_seq_lock);
767 if (msgr->global_seq < gt)
768 msgr->global_seq = gt;
769 ret = ++msgr->global_seq;
770 spin_unlock(&msgr->global_seq_lock);
774 static void con_out_kvec_reset(struct ceph_connection *con)
776 con->out_kvec_left = 0;
777 con->out_kvec_bytes = 0;
778 con->out_kvec_cur = &con->out_kvec[0];
781 static void con_out_kvec_add(struct ceph_connection *con,
782 size_t size, void *data)
786 index = con->out_kvec_left;
787 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
789 con->out_kvec[index].iov_len = size;
790 con->out_kvec[index].iov_base = data;
791 con->out_kvec_left++;
792 con->out_kvec_bytes += size;
798 * For a bio data item, a piece is whatever remains of the next
799 * entry in the current bio iovec, or the first entry in the next
802 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
805 struct ceph_msg_data *data = cursor->data;
808 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
813 cursor->resid = min(length, data->bio_length);
815 cursor->bvec_iter = bio->bi_iter;
817 cursor->resid <= bio_iter_len(bio, cursor->bvec_iter);
820 static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
824 struct ceph_msg_data *data = cursor->data;
826 struct bio_vec bio_vec;
828 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
833 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
835 *page_offset = (size_t) bio_vec.bv_offset;
836 BUG_ON(*page_offset >= PAGE_SIZE);
837 if (cursor->last_piece) /* pagelist offset is always 0 */
838 *length = cursor->resid;
840 *length = (size_t) bio_vec.bv_len;
841 BUG_ON(*length > cursor->resid);
842 BUG_ON(*page_offset + *length > PAGE_SIZE);
844 return bio_vec.bv_page;
847 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
851 struct bio_vec bio_vec;
853 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
858 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
860 /* Advance the cursor offset */
862 BUG_ON(cursor->resid < bytes);
863 cursor->resid -= bytes;
865 bio_advance_iter(bio, &cursor->bvec_iter, bytes);
867 if (bytes < bio_vec.bv_len)
868 return false; /* more bytes to process in this segment */
870 /* Move on to the next segment, and possibly the next bio */
872 if (!cursor->bvec_iter.bi_size) {
876 cursor->bvec_iter = bio->bi_iter;
878 memset(&cursor->bvec_iter, 0,
879 sizeof(cursor->bvec_iter));
882 if (!cursor->last_piece) {
883 BUG_ON(!cursor->resid);
885 /* A short read is OK, so use <= rather than == */
886 if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter))
887 cursor->last_piece = true;
892 #endif /* CONFIG_BLOCK */
895 * For a page array, a piece comes from the first page in the array
896 * that has not already been fully consumed.
898 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
901 struct ceph_msg_data *data = cursor->data;
904 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
906 BUG_ON(!data->pages);
907 BUG_ON(!data->length);
909 cursor->resid = min(length, data->length);
910 page_count = calc_pages_for(data->alignment, (u64)data->length);
911 cursor->page_offset = data->alignment & ~PAGE_MASK;
912 cursor->page_index = 0;
913 BUG_ON(page_count > (int)USHRT_MAX);
914 cursor->page_count = (unsigned short)page_count;
915 BUG_ON(length > SIZE_MAX - cursor->page_offset);
916 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
920 ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
921 size_t *page_offset, size_t *length)
923 struct ceph_msg_data *data = cursor->data;
925 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
927 BUG_ON(cursor->page_index >= cursor->page_count);
928 BUG_ON(cursor->page_offset >= PAGE_SIZE);
930 *page_offset = cursor->page_offset;
931 if (cursor->last_piece)
932 *length = cursor->resid;
934 *length = PAGE_SIZE - *page_offset;
936 return data->pages[cursor->page_index];
939 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
942 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
944 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
946 /* Advance the cursor page offset */
948 cursor->resid -= bytes;
949 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
950 if (!bytes || cursor->page_offset)
951 return false; /* more bytes to process in the current page */
954 return false; /* no more data */
956 /* Move on to the next page; offset is already at 0 */
958 BUG_ON(cursor->page_index >= cursor->page_count);
959 cursor->page_index++;
960 cursor->last_piece = cursor->resid <= PAGE_SIZE;
966 * For a pagelist, a piece is whatever remains to be consumed in the
967 * first page in the list, or the front of the next page.
970 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
973 struct ceph_msg_data *data = cursor->data;
974 struct ceph_pagelist *pagelist;
977 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
979 pagelist = data->pagelist;
983 return; /* pagelist can be assigned but empty */
985 BUG_ON(list_empty(&pagelist->head));
986 page = list_first_entry(&pagelist->head, struct page, lru);
988 cursor->resid = min(length, pagelist->length);
991 cursor->last_piece = cursor->resid <= PAGE_SIZE;
995 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
996 size_t *page_offset, size_t *length)
998 struct ceph_msg_data *data = cursor->data;
999 struct ceph_pagelist *pagelist;
1001 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1003 pagelist = data->pagelist;
1006 BUG_ON(!cursor->page);
1007 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1009 /* offset of first page in pagelist is always 0 */
1010 *page_offset = cursor->offset & ~PAGE_MASK;
1011 if (cursor->last_piece)
1012 *length = cursor->resid;
1014 *length = PAGE_SIZE - *page_offset;
1016 return cursor->page;
1019 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
1022 struct ceph_msg_data *data = cursor->data;
1023 struct ceph_pagelist *pagelist;
1025 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1027 pagelist = data->pagelist;
1030 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1031 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1033 /* Advance the cursor offset */
1035 cursor->resid -= bytes;
1036 cursor->offset += bytes;
1037 /* offset of first page in pagelist is always 0 */
1038 if (!bytes || cursor->offset & ~PAGE_MASK)
1039 return false; /* more bytes to process in the current page */
1042 return false; /* no more data */
1044 /* Move on to the next page */
1046 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1047 cursor->page = list_entry_next(cursor->page, lru);
1048 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1054 * Message data is handled (sent or received) in pieces, where each
1055 * piece resides on a single page. The network layer might not
1056 * consume an entire piece at once. A data item's cursor keeps
1057 * track of which piece is next to process and how much remains to
1058 * be processed in that piece. It also tracks whether the current
1059 * piece is the last one in the data item.
1061 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1063 size_t length = cursor->total_resid;
1065 switch (cursor->data->type) {
1066 case CEPH_MSG_DATA_PAGELIST:
1067 ceph_msg_data_pagelist_cursor_init(cursor, length);
1069 case CEPH_MSG_DATA_PAGES:
1070 ceph_msg_data_pages_cursor_init(cursor, length);
1073 case CEPH_MSG_DATA_BIO:
1074 ceph_msg_data_bio_cursor_init(cursor, length);
1076 #endif /* CONFIG_BLOCK */
1077 case CEPH_MSG_DATA_NONE:
1082 cursor->need_crc = true;
1085 static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1087 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1088 struct ceph_msg_data *data;
1091 BUG_ON(length > msg->data_length);
1092 BUG_ON(list_empty(&msg->data));
1094 cursor->data_head = &msg->data;
1095 cursor->total_resid = length;
1096 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1097 cursor->data = data;
1099 __ceph_msg_data_cursor_init(cursor);
1103 * Return the page containing the next piece to process for a given
1104 * data item, and supply the page offset and length of that piece.
1105 * Indicate whether this is the last piece in this data item.
1107 static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1108 size_t *page_offset, size_t *length,
1113 switch (cursor->data->type) {
1114 case CEPH_MSG_DATA_PAGELIST:
1115 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1117 case CEPH_MSG_DATA_PAGES:
1118 page = ceph_msg_data_pages_next(cursor, page_offset, length);
1121 case CEPH_MSG_DATA_BIO:
1122 page = ceph_msg_data_bio_next(cursor, page_offset, length);
1124 #endif /* CONFIG_BLOCK */
1125 case CEPH_MSG_DATA_NONE:
1131 BUG_ON(*page_offset + *length > PAGE_SIZE);
1134 *last_piece = cursor->last_piece;
1140 * Returns true if the result moves the cursor on to the next piece
1143 static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1148 BUG_ON(bytes > cursor->resid);
1149 switch (cursor->data->type) {
1150 case CEPH_MSG_DATA_PAGELIST:
1151 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1153 case CEPH_MSG_DATA_PAGES:
1154 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1157 case CEPH_MSG_DATA_BIO:
1158 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1160 #endif /* CONFIG_BLOCK */
1161 case CEPH_MSG_DATA_NONE:
1166 cursor->total_resid -= bytes;
1168 if (!cursor->resid && cursor->total_resid) {
1169 WARN_ON(!cursor->last_piece);
1170 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1171 cursor->data = list_entry_next(cursor->data, links);
1172 __ceph_msg_data_cursor_init(cursor);
1175 cursor->need_crc = new_piece;
1180 static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
1185 /* Initialize data cursor */
1187 ceph_msg_data_cursor_init(msg, (size_t)data_len);
1191 * Prepare footer for currently outgoing message, and finish things
1192 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1194 static void prepare_write_message_footer(struct ceph_connection *con)
1196 struct ceph_msg *m = con->out_msg;
1197 int v = con->out_kvec_left;
1199 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1201 dout("prepare_write_message_footer %p\n", con);
1202 con->out_kvec_is_msg = true;
1203 con->out_kvec[v].iov_base = &m->footer;
1204 if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
1205 if (con->ops->sign_message)
1206 con->ops->sign_message(con, m);
1209 con->out_kvec[v].iov_len = sizeof(m->footer);
1210 con->out_kvec_bytes += sizeof(m->footer);
1212 m->old_footer.flags = m->footer.flags;
1213 con->out_kvec[v].iov_len = sizeof(m->old_footer);
1214 con->out_kvec_bytes += sizeof(m->old_footer);
1216 con->out_kvec_left++;
1217 con->out_more = m->more_to_follow;
1218 con->out_msg_done = true;
1222 * Prepare headers for the next outgoing message.
1224 static void prepare_write_message(struct ceph_connection *con)
1229 con_out_kvec_reset(con);
1230 con->out_kvec_is_msg = true;
1231 con->out_msg_done = false;
1233 /* Sneak an ack in there first? If we can get it into the same
1234 * TCP packet that's a good thing. */
1235 if (con->in_seq > con->in_seq_acked) {
1236 con->in_seq_acked = con->in_seq;
1237 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1238 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1239 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1240 &con->out_temp_ack);
1243 BUG_ON(list_empty(&con->out_queue));
1244 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1246 BUG_ON(m->con != con);
1248 /* put message on sent list */
1250 list_move_tail(&m->list_head, &con->out_sent);
1253 * only assign outgoing seq # if we haven't sent this message
1254 * yet. if it is requeued, resend with it's original seq.
1256 if (m->needs_out_seq) {
1257 m->hdr.seq = cpu_to_le64(++con->out_seq);
1258 m->needs_out_seq = false;
1260 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
1262 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1263 m, con->out_seq, le16_to_cpu(m->hdr.type),
1264 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1266 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1268 /* tag + hdr + front + middle */
1269 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1270 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1271 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1274 con_out_kvec_add(con, m->middle->vec.iov_len,
1275 m->middle->vec.iov_base);
1277 /* fill in crc (except data pages), footer */
1278 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1279 con->out_msg->hdr.crc = cpu_to_le32(crc);
1280 con->out_msg->footer.flags = 0;
1282 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1283 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1285 crc = crc32c(0, m->middle->vec.iov_base,
1286 m->middle->vec.iov_len);
1287 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1289 con->out_msg->footer.middle_crc = 0;
1290 dout("%s front_crc %u middle_crc %u\n", __func__,
1291 le32_to_cpu(con->out_msg->footer.front_crc),
1292 le32_to_cpu(con->out_msg->footer.middle_crc));
1294 /* is there a data payload? */
1295 con->out_msg->footer.data_crc = 0;
1296 if (m->data_length) {
1297 prepare_message_data(con->out_msg, m->data_length);
1298 con->out_more = 1; /* data + footer will follow */
1300 /* no, queue up footer too and be done */
1301 prepare_write_message_footer(con);
1304 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1310 static void prepare_write_ack(struct ceph_connection *con)
1312 dout("prepare_write_ack %p %llu -> %llu\n", con,
1313 con->in_seq_acked, con->in_seq);
1314 con->in_seq_acked = con->in_seq;
1316 con_out_kvec_reset(con);
1318 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1320 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1321 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1322 &con->out_temp_ack);
1324 con->out_more = 1; /* more will follow.. eventually.. */
1325 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1329 * Prepare to share the seq during handshake
1331 static void prepare_write_seq(struct ceph_connection *con)
1333 dout("prepare_write_seq %p %llu -> %llu\n", con,
1334 con->in_seq_acked, con->in_seq);
1335 con->in_seq_acked = con->in_seq;
1337 con_out_kvec_reset(con);
1339 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1340 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1341 &con->out_temp_ack);
1343 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1347 * Prepare to write keepalive byte.
1349 static void prepare_write_keepalive(struct ceph_connection *con)
1351 dout("prepare_write_keepalive %p\n", con);
1352 con_out_kvec_reset(con);
1353 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
1354 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1358 * Connection negotiation.
1361 static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1364 struct ceph_auth_handshake *auth;
1366 if (!con->ops->get_authorizer) {
1367 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1368 con->out_connect.authorizer_len = 0;
1372 /* Can't hold the mutex while getting authorizer */
1373 mutex_unlock(&con->mutex);
1374 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
1375 mutex_lock(&con->mutex);
1379 if (con->state != CON_STATE_NEGOTIATING)
1380 return ERR_PTR(-EAGAIN);
1382 con->auth_reply_buf = auth->authorizer_reply_buf;
1383 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
1388 * We connected to a peer and are saying hello.
1390 static void prepare_write_banner(struct ceph_connection *con)
1392 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1393 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1394 &con->msgr->my_enc_addr);
1397 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1400 static int prepare_write_connect(struct ceph_connection *con)
1402 unsigned int global_seq = get_global_seq(con->msgr, 0);
1405 struct ceph_auth_handshake *auth;
1407 switch (con->peer_name.type) {
1408 case CEPH_ENTITY_TYPE_MON:
1409 proto = CEPH_MONC_PROTOCOL;
1411 case CEPH_ENTITY_TYPE_OSD:
1412 proto = CEPH_OSDC_PROTOCOL;
1414 case CEPH_ENTITY_TYPE_MDS:
1415 proto = CEPH_MDSC_PROTOCOL;
1421 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1422 con->connect_seq, global_seq, proto);
1424 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
1425 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1426 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1427 con->out_connect.global_seq = cpu_to_le32(global_seq);
1428 con->out_connect.protocol_version = cpu_to_le32(proto);
1429 con->out_connect.flags = 0;
1431 auth_proto = CEPH_AUTH_UNKNOWN;
1432 auth = get_connect_authorizer(con, &auth_proto);
1434 return PTR_ERR(auth);
1436 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1437 con->out_connect.authorizer_len = auth ?
1438 cpu_to_le32(auth->authorizer_buf_len) : 0;
1440 con_out_kvec_add(con, sizeof (con->out_connect),
1442 if (auth && auth->authorizer_buf_len)
1443 con_out_kvec_add(con, auth->authorizer_buf_len,
1444 auth->authorizer_buf);
1447 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1453 * write as much of pending kvecs to the socket as we can.
1455 * 0 -> socket full, but more to do
1458 static int write_partial_kvec(struct ceph_connection *con)
1462 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1463 while (con->out_kvec_bytes > 0) {
1464 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1465 con->out_kvec_left, con->out_kvec_bytes,
1469 con->out_kvec_bytes -= ret;
1470 if (con->out_kvec_bytes == 0)
1473 /* account for full iov entries consumed */
1474 while (ret >= con->out_kvec_cur->iov_len) {
1475 BUG_ON(!con->out_kvec_left);
1476 ret -= con->out_kvec_cur->iov_len;
1477 con->out_kvec_cur++;
1478 con->out_kvec_left--;
1480 /* and for a partially-consumed entry */
1482 con->out_kvec_cur->iov_len -= ret;
1483 con->out_kvec_cur->iov_base += ret;
1486 con->out_kvec_left = 0;
1487 con->out_kvec_is_msg = false;
1490 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1491 con->out_kvec_bytes, con->out_kvec_left, ret);
1492 return ret; /* done! */
1495 static u32 ceph_crc32c_page(u32 crc, struct page *page,
1496 unsigned int page_offset,
1497 unsigned int length)
1502 BUG_ON(kaddr == NULL);
1503 crc = crc32c(crc, kaddr + page_offset, length);
1509 * Write as much message data payload as we can. If we finish, queue
1511 * 1 -> done, footer is now queued in out_kvec[].
1512 * 0 -> socket full, but more to do
1515 static int write_partial_message_data(struct ceph_connection *con)
1517 struct ceph_msg *msg = con->out_msg;
1518 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1519 bool do_datacrc = !con->msgr->nocrc;
1522 dout("%s %p msg %p\n", __func__, con, msg);
1524 if (list_empty(&msg->data))
1528 * Iterate through each page that contains data to be
1529 * written, and send as much as possible for each.
1531 * If we are calculating the data crc (the default), we will
1532 * need to map the page. If we have no pages, they have
1533 * been revoked, so use the zero page.
1535 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1536 while (cursor->resid) {
1544 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
1546 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
1547 length, !last_piece);
1550 msg->footer.data_crc = cpu_to_le32(crc);
1554 if (do_datacrc && cursor->need_crc)
1555 crc = ceph_crc32c_page(crc, page, page_offset, length);
1556 need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret);
1559 dout("%s %p msg %p done\n", __func__, con, msg);
1561 /* prepare and queue up footer, too */
1563 msg->footer.data_crc = cpu_to_le32(crc);
1565 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1566 con_out_kvec_reset(con);
1567 prepare_write_message_footer(con);
1569 return 1; /* must return > 0 to indicate success */
1575 static int write_partial_skip(struct ceph_connection *con)
1579 while (con->out_skip > 0) {
1580 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
1582 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
1585 con->out_skip -= ret;
1593 * Prepare to read connection handshake, or an ack.
1595 static void prepare_read_banner(struct ceph_connection *con)
1597 dout("prepare_read_banner %p\n", con);
1598 con->in_base_pos = 0;
1601 static void prepare_read_connect(struct ceph_connection *con)
1603 dout("prepare_read_connect %p\n", con);
1604 con->in_base_pos = 0;
1607 static void prepare_read_ack(struct ceph_connection *con)
1609 dout("prepare_read_ack %p\n", con);
1610 con->in_base_pos = 0;
1613 static void prepare_read_seq(struct ceph_connection *con)
1615 dout("prepare_read_seq %p\n", con);
1616 con->in_base_pos = 0;
1617 con->in_tag = CEPH_MSGR_TAG_SEQ;
1620 static void prepare_read_tag(struct ceph_connection *con)
1622 dout("prepare_read_tag %p\n", con);
1623 con->in_base_pos = 0;
1624 con->in_tag = CEPH_MSGR_TAG_READY;
1628 * Prepare to read a message.
1630 static int prepare_read_message(struct ceph_connection *con)
1632 dout("prepare_read_message %p\n", con);
1633 BUG_ON(con->in_msg != NULL);
1634 con->in_base_pos = 0;
1635 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1640 static int read_partial(struct ceph_connection *con,
1641 int end, int size, void *object)
1643 while (con->in_base_pos < end) {
1644 int left = end - con->in_base_pos;
1645 int have = size - left;
1646 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1649 con->in_base_pos += ret;
1656 * Read all or part of the connect-side handshake on a new connection
1658 static int read_partial_banner(struct ceph_connection *con)
1664 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1667 size = strlen(CEPH_BANNER);
1669 ret = read_partial(con, end, size, con->in_banner);
1673 size = sizeof (con->actual_peer_addr);
1675 ret = read_partial(con, end, size, &con->actual_peer_addr);
1679 size = sizeof (con->peer_addr_for_me);
1681 ret = read_partial(con, end, size, &con->peer_addr_for_me);
1689 static int read_partial_connect(struct ceph_connection *con)
1695 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1697 size = sizeof (con->in_reply);
1699 ret = read_partial(con, end, size, &con->in_reply);
1703 size = le32_to_cpu(con->in_reply.authorizer_len);
1705 ret = read_partial(con, end, size, con->auth_reply_buf);
1709 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1710 con, (int)con->in_reply.tag,
1711 le32_to_cpu(con->in_reply.connect_seq),
1712 le32_to_cpu(con->in_reply.global_seq));
1719 * Verify the hello banner looks okay.
1721 static int verify_hello(struct ceph_connection *con)
1723 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1724 pr_err("connect to %s got bad banner\n",
1725 ceph_pr_addr(&con->peer_addr.in_addr));
1726 con->error_msg = "protocol error, bad banner";
1732 static bool addr_is_blank(struct sockaddr_storage *ss)
1734 switch (ss->ss_family) {
1736 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1739 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1740 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1741 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1742 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1747 static int addr_port(struct sockaddr_storage *ss)
1749 switch (ss->ss_family) {
1751 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1753 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1758 static void addr_set_port(struct sockaddr_storage *ss, int p)
1760 switch (ss->ss_family) {
1762 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1765 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1771 * Unlike other *_pton function semantics, zero indicates success.
1773 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1774 char delim, const char **ipend)
1776 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1777 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1779 memset(ss, 0, sizeof(*ss));
1781 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1782 ss->ss_family = AF_INET;
1786 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1787 ss->ss_family = AF_INET6;
1795 * Extract hostname string and resolve using kernel DNS facility.
1797 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1798 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1799 struct sockaddr_storage *ss, char delim, const char **ipend)
1801 const char *end, *delim_p;
1802 char *colon_p, *ip_addr = NULL;
1806 * The end of the hostname occurs immediately preceding the delimiter or
1807 * the port marker (':') where the delimiter takes precedence.
1809 delim_p = memchr(name, delim, namelen);
1810 colon_p = memchr(name, ':', namelen);
1812 if (delim_p && colon_p)
1813 end = delim_p < colon_p ? delim_p : colon_p;
1814 else if (!delim_p && colon_p)
1818 if (!end) /* case: hostname:/ */
1819 end = name + namelen;
1825 /* do dns_resolve upcall */
1826 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1828 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1836 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1837 ret, ret ? "failed" : ceph_pr_addr(ss));
1842 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1843 struct sockaddr_storage *ss, char delim, const char **ipend)
1850 * Parse a server name (IP or hostname). If a valid IP address is not found
1851 * then try to extract a hostname to resolve using userspace DNS upcall.
1853 static int ceph_parse_server_name(const char *name, size_t namelen,
1854 struct sockaddr_storage *ss, char delim, const char **ipend)
1858 ret = ceph_pton(name, namelen, ss, delim, ipend);
1860 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1866 * Parse an ip[:port] list into an addr array. Use the default
1867 * monitor port if a port isn't specified.
1869 int ceph_parse_ips(const char *c, const char *end,
1870 struct ceph_entity_addr *addr,
1871 int max_count, int *count)
1873 int i, ret = -EINVAL;
1876 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1877 for (i = 0; i < max_count; i++) {
1879 struct sockaddr_storage *ss = &addr[i].in_addr;
1888 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1897 dout("missing matching ']'\n");
1904 if (p < end && *p == ':') {
1907 while (p < end && *p >= '0' && *p <= '9') {
1908 port = (port * 10) + (*p - '0');
1912 port = CEPH_MON_PORT;
1913 else if (port > 65535)
1916 port = CEPH_MON_PORT;
1919 addr_set_port(ss, port);
1921 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1938 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1941 EXPORT_SYMBOL(ceph_parse_ips);
1943 static int process_banner(struct ceph_connection *con)
1945 dout("process_banner on %p\n", con);
1947 if (verify_hello(con) < 0)
1950 ceph_decode_addr(&con->actual_peer_addr);
1951 ceph_decode_addr(&con->peer_addr_for_me);
1954 * Make sure the other end is who we wanted. note that the other
1955 * end may not yet know their ip address, so if it's 0.0.0.0, give
1956 * them the benefit of the doubt.
1958 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1959 sizeof(con->peer_addr)) != 0 &&
1960 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1961 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1962 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
1963 ceph_pr_addr(&con->peer_addr.in_addr),
1964 (int)le32_to_cpu(con->peer_addr.nonce),
1965 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1966 (int)le32_to_cpu(con->actual_peer_addr.nonce));
1967 con->error_msg = "wrong peer at address";
1972 * did we learn our address?
1974 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1975 int port = addr_port(&con->msgr->inst.addr.in_addr);
1977 memcpy(&con->msgr->inst.addr.in_addr,
1978 &con->peer_addr_for_me.in_addr,
1979 sizeof(con->peer_addr_for_me.in_addr));
1980 addr_set_port(&con->msgr->inst.addr.in_addr, port);
1981 encode_my_addr(con->msgr);
1982 dout("process_banner learned my addr is %s\n",
1983 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
1989 static int process_connect(struct ceph_connection *con)
1991 u64 sup_feat = con->msgr->supported_features;
1992 u64 req_feat = con->msgr->required_features;
1993 u64 server_feat = ceph_sanitize_features(
1994 le64_to_cpu(con->in_reply.features));
1997 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1999 switch (con->in_reply.tag) {
2000 case CEPH_MSGR_TAG_FEATURES:
2001 pr_err("%s%lld %s feature set mismatch,"
2002 " my %llx < server's %llx, missing %llx\n",
2003 ENTITY_NAME(con->peer_name),
2004 ceph_pr_addr(&con->peer_addr.in_addr),
2005 sup_feat, server_feat, server_feat & ~sup_feat);
2006 con->error_msg = "missing required protocol features";
2007 reset_connection(con);
2010 case CEPH_MSGR_TAG_BADPROTOVER:
2011 pr_err("%s%lld %s protocol version mismatch,"
2012 " my %d != server's %d\n",
2013 ENTITY_NAME(con->peer_name),
2014 ceph_pr_addr(&con->peer_addr.in_addr),
2015 le32_to_cpu(con->out_connect.protocol_version),
2016 le32_to_cpu(con->in_reply.protocol_version));
2017 con->error_msg = "protocol version mismatch";
2018 reset_connection(con);
2021 case CEPH_MSGR_TAG_BADAUTHORIZER:
2023 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2025 if (con->auth_retry == 2) {
2026 con->error_msg = "connect authorization failure";
2029 con_out_kvec_reset(con);
2030 ret = prepare_write_connect(con);
2033 prepare_read_connect(con);
2036 case CEPH_MSGR_TAG_RESETSESSION:
2038 * If we connected with a large connect_seq but the peer
2039 * has no record of a session with us (no connection, or
2040 * connect_seq == 0), they will send RESETSESION to indicate
2041 * that they must have reset their session, and may have
2044 dout("process_connect got RESET peer seq %u\n",
2045 le32_to_cpu(con->in_reply.connect_seq));
2046 pr_err("%s%lld %s connection reset\n",
2047 ENTITY_NAME(con->peer_name),
2048 ceph_pr_addr(&con->peer_addr.in_addr));
2049 reset_connection(con);
2050 con_out_kvec_reset(con);
2051 ret = prepare_write_connect(con);
2054 prepare_read_connect(con);
2056 /* Tell ceph about it. */
2057 mutex_unlock(&con->mutex);
2058 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2059 if (con->ops->peer_reset)
2060 con->ops->peer_reset(con);
2061 mutex_lock(&con->mutex);
2062 if (con->state != CON_STATE_NEGOTIATING)
2066 case CEPH_MSGR_TAG_RETRY_SESSION:
2068 * If we sent a smaller connect_seq than the peer has, try
2069 * again with a larger value.
2071 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2072 le32_to_cpu(con->out_connect.connect_seq),
2073 le32_to_cpu(con->in_reply.connect_seq));
2074 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2075 con_out_kvec_reset(con);
2076 ret = prepare_write_connect(con);
2079 prepare_read_connect(con);
2082 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2084 * If we sent a smaller global_seq than the peer has, try
2085 * again with a larger value.
2087 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2088 con->peer_global_seq,
2089 le32_to_cpu(con->in_reply.global_seq));
2090 get_global_seq(con->msgr,
2091 le32_to_cpu(con->in_reply.global_seq));
2092 con_out_kvec_reset(con);
2093 ret = prepare_write_connect(con);
2096 prepare_read_connect(con);
2099 case CEPH_MSGR_TAG_SEQ:
2100 case CEPH_MSGR_TAG_READY:
2101 if (req_feat & ~server_feat) {
2102 pr_err("%s%lld %s protocol feature mismatch,"
2103 " my required %llx > server's %llx, need %llx\n",
2104 ENTITY_NAME(con->peer_name),
2105 ceph_pr_addr(&con->peer_addr.in_addr),
2106 req_feat, server_feat, req_feat & ~server_feat);
2107 con->error_msg = "missing required protocol features";
2108 reset_connection(con);
2112 WARN_ON(con->state != CON_STATE_NEGOTIATING);
2113 con->state = CON_STATE_OPEN;
2114 con->auth_retry = 0; /* we authenticated; clear flag */
2115 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2117 con->peer_features = server_feat;
2118 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2119 con->peer_global_seq,
2120 le32_to_cpu(con->in_reply.connect_seq),
2122 WARN_ON(con->connect_seq !=
2123 le32_to_cpu(con->in_reply.connect_seq));
2125 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2126 con_flag_set(con, CON_FLAG_LOSSYTX);
2128 con->delay = 0; /* reset backoff memory */
2130 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2131 prepare_write_seq(con);
2132 prepare_read_seq(con);
2134 prepare_read_tag(con);
2138 case CEPH_MSGR_TAG_WAIT:
2140 * If there is a connection race (we are opening
2141 * connections to each other), one of us may just have
2142 * to WAIT. This shouldn't happen if we are the
2145 con->error_msg = "protocol error, got WAIT as client";
2149 con->error_msg = "protocol error, garbage tag during connect";
2157 * read (part of) an ack
2159 static int read_partial_ack(struct ceph_connection *con)
2161 int size = sizeof (con->in_temp_ack);
2164 return read_partial(con, end, size, &con->in_temp_ack);
2168 * We can finally discard anything that's been acked.
2170 static void process_ack(struct ceph_connection *con)
2173 u64 ack = le64_to_cpu(con->in_temp_ack);
2176 while (!list_empty(&con->out_sent)) {
2177 m = list_first_entry(&con->out_sent, struct ceph_msg,
2179 seq = le64_to_cpu(m->hdr.seq);
2182 dout("got ack for seq %llu type %d at %p\n", seq,
2183 le16_to_cpu(m->hdr.type), m);
2184 m->ack_stamp = jiffies;
2187 prepare_read_tag(con);
2191 static int read_partial_message_section(struct ceph_connection *con,
2192 struct kvec *section,
2193 unsigned int sec_len, u32 *crc)
2199 while (section->iov_len < sec_len) {
2200 BUG_ON(section->iov_base == NULL);
2201 left = sec_len - section->iov_len;
2202 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2203 section->iov_len, left);
2206 section->iov_len += ret;
2208 if (section->iov_len == sec_len)
2209 *crc = crc32c(0, section->iov_base, section->iov_len);
2214 static int read_partial_msg_data(struct ceph_connection *con)
2216 struct ceph_msg *msg = con->in_msg;
2217 struct ceph_msg_data_cursor *cursor = &msg->cursor;
2218 const bool do_datacrc = !con->msgr->nocrc;
2226 if (list_empty(&msg->data))
2230 crc = con->in_data_crc;
2231 while (cursor->resid) {
2232 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
2234 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2237 con->in_data_crc = crc;
2243 crc = ceph_crc32c_page(crc, page, page_offset, ret);
2244 (void) ceph_msg_data_advance(&msg->cursor, (size_t)ret);
2247 con->in_data_crc = crc;
2249 return 1; /* must return > 0 to indicate success */
2253 * read (part of) a message.
2255 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2257 static int read_partial_message(struct ceph_connection *con)
2259 struct ceph_msg *m = con->in_msg;
2263 unsigned int front_len, middle_len, data_len;
2264 bool do_datacrc = !con->msgr->nocrc;
2265 bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
2269 dout("read_partial_message con %p msg %p\n", con, m);
2272 size = sizeof (con->in_hdr);
2274 ret = read_partial(con, end, size, &con->in_hdr);
2278 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2279 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2280 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
2281 crc, con->in_hdr.crc);
2285 front_len = le32_to_cpu(con->in_hdr.front_len);
2286 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2288 middle_len = le32_to_cpu(con->in_hdr.middle_len);
2289 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2291 data_len = le32_to_cpu(con->in_hdr.data_len);
2292 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2296 seq = le64_to_cpu(con->in_hdr.seq);
2297 if ((s64)seq - (s64)con->in_seq < 1) {
2298 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2299 ENTITY_NAME(con->peer_name),
2300 ceph_pr_addr(&con->peer_addr.in_addr),
2301 seq, con->in_seq + 1);
2302 con->in_base_pos = -front_len - middle_len - data_len -
2304 con->in_tag = CEPH_MSGR_TAG_READY;
2306 } else if ((s64)seq - (s64)con->in_seq > 1) {
2307 pr_err("read_partial_message bad seq %lld expected %lld\n",
2308 seq, con->in_seq + 1);
2309 con->error_msg = "bad message sequence # for incoming message";
2313 /* allocate message? */
2317 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2318 front_len, data_len);
2319 ret = ceph_con_in_msg_alloc(con, &skip);
2323 BUG_ON(!con->in_msg ^ skip);
2324 if (con->in_msg && data_len > con->in_msg->data_length) {
2325 pr_warn("%s skipping long message (%u > %zd)\n",
2326 __func__, data_len, con->in_msg->data_length);
2327 ceph_msg_put(con->in_msg);
2332 /* skip this message */
2333 dout("alloc_msg said skip message\n");
2334 con->in_base_pos = -front_len - middle_len - data_len -
2336 con->in_tag = CEPH_MSGR_TAG_READY;
2341 BUG_ON(!con->in_msg);
2342 BUG_ON(con->in_msg->con != con);
2344 m->front.iov_len = 0; /* haven't read it yet */
2346 m->middle->vec.iov_len = 0;
2348 /* prepare for data payload, if any */
2351 prepare_message_data(con->in_msg, data_len);
2355 ret = read_partial_message_section(con, &m->front, front_len,
2356 &con->in_front_crc);
2362 ret = read_partial_message_section(con, &m->middle->vec,
2364 &con->in_middle_crc);
2371 ret = read_partial_msg_data(con);
2378 size = sizeof(m->footer);
2380 size = sizeof(m->old_footer);
2383 ret = read_partial(con, end, size, &m->footer);
2388 m->footer.flags = m->old_footer.flags;
2392 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2393 m, front_len, m->footer.front_crc, middle_len,
2394 m->footer.middle_crc, data_len, m->footer.data_crc);
2397 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2398 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2399 m, con->in_front_crc, m->footer.front_crc);
2402 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2403 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2404 m, con->in_middle_crc, m->footer.middle_crc);
2408 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2409 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2410 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2411 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2415 if (need_sign && con->ops->check_message_signature &&
2416 con->ops->check_message_signature(con, m)) {
2417 pr_err("read_partial_message %p signature check failed\n", m);
2421 return 1; /* done! */
2425 * Process message. This happens in the worker thread. The callback should
2426 * be careful not to do anything that waits on other incoming messages or it
2429 static void process_message(struct ceph_connection *con)
2431 struct ceph_msg *msg;
2433 BUG_ON(con->in_msg->con != con);
2434 con->in_msg->con = NULL;
2439 /* if first message, set peer_name */
2440 if (con->peer_name.type == 0)
2441 con->peer_name = msg->hdr.src;
2444 mutex_unlock(&con->mutex);
2446 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2447 msg, le64_to_cpu(msg->hdr.seq),
2448 ENTITY_NAME(msg->hdr.src),
2449 le16_to_cpu(msg->hdr.type),
2450 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2451 le32_to_cpu(msg->hdr.front_len),
2452 le32_to_cpu(msg->hdr.data_len),
2453 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2454 con->ops->dispatch(con, msg);
2456 mutex_lock(&con->mutex);
2461 * Write something to the socket. Called in a worker thread when the
2462 * socket appears to be writeable and we have something ready to send.
2464 static int try_write(struct ceph_connection *con)
2468 dout("try_write start %p state %lu\n", con, con->state);
2471 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2473 /* open the socket first? */
2474 if (con->state == CON_STATE_PREOPEN) {
2476 con->state = CON_STATE_CONNECTING;
2478 con_out_kvec_reset(con);
2479 prepare_write_banner(con);
2480 prepare_read_banner(con);
2482 BUG_ON(con->in_msg);
2483 con->in_tag = CEPH_MSGR_TAG_READY;
2484 dout("try_write initiating connect on %p new state %lu\n",
2486 ret = ceph_tcp_connect(con);
2488 con->error_msg = "connect error";
2494 /* kvec data queued? */
2495 if (con->out_skip) {
2496 ret = write_partial_skip(con);
2500 if (con->out_kvec_left) {
2501 ret = write_partial_kvec(con);
2508 if (con->out_msg_done) {
2509 ceph_msg_put(con->out_msg);
2510 con->out_msg = NULL; /* we're done with this one */
2514 ret = write_partial_message_data(con);
2516 goto more_kvec; /* we need to send the footer, too! */
2520 dout("try_write write_partial_message_data err %d\n",
2527 if (con->state == CON_STATE_OPEN) {
2528 /* is anything else pending? */
2529 if (!list_empty(&con->out_queue)) {
2530 prepare_write_message(con);
2533 if (con->in_seq > con->in_seq_acked) {
2534 prepare_write_ack(con);
2537 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2538 prepare_write_keepalive(con);
2543 /* Nothing to do! */
2544 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2545 dout("try_write nothing else to write.\n");
2548 dout("try_write done on %p ret %d\n", con, ret);
2555 * Read what we can from the socket.
2557 static int try_read(struct ceph_connection *con)
2562 dout("try_read start on %p state %lu\n", con, con->state);
2563 if (con->state != CON_STATE_CONNECTING &&
2564 con->state != CON_STATE_NEGOTIATING &&
2565 con->state != CON_STATE_OPEN)
2570 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2573 if (con->state == CON_STATE_CONNECTING) {
2574 dout("try_read connecting\n");
2575 ret = read_partial_banner(con);
2578 ret = process_banner(con);
2582 con->state = CON_STATE_NEGOTIATING;
2585 * Received banner is good, exchange connection info.
2586 * Do not reset out_kvec, as sending our banner raced
2587 * with receiving peer banner after connect completed.
2589 ret = prepare_write_connect(con);
2592 prepare_read_connect(con);
2594 /* Send connection info before awaiting response */
2598 if (con->state == CON_STATE_NEGOTIATING) {
2599 dout("try_read negotiating\n");
2600 ret = read_partial_connect(con);
2603 ret = process_connect(con);
2609 WARN_ON(con->state != CON_STATE_OPEN);
2611 if (con->in_base_pos < 0) {
2613 * skipping + discarding content.
2615 * FIXME: there must be a better way to do this!
2617 static char buf[SKIP_BUF_SIZE];
2618 int skip = min((int) sizeof (buf), -con->in_base_pos);
2620 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2621 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2624 con->in_base_pos += ret;
2625 if (con->in_base_pos)
2628 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2632 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2635 dout("try_read got tag %d\n", (int)con->in_tag);
2636 switch (con->in_tag) {
2637 case CEPH_MSGR_TAG_MSG:
2638 prepare_read_message(con);
2640 case CEPH_MSGR_TAG_ACK:
2641 prepare_read_ack(con);
2643 case CEPH_MSGR_TAG_CLOSE:
2644 con_close_socket(con);
2645 con->state = CON_STATE_CLOSED;
2651 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2652 ret = read_partial_message(con);
2656 con->error_msg = "bad crc";
2662 con->error_msg = "io error";
2667 if (con->in_tag == CEPH_MSGR_TAG_READY)
2669 process_message(con);
2670 if (con->state == CON_STATE_OPEN)
2671 prepare_read_tag(con);
2674 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2675 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2677 * the final handshake seq exchange is semantically
2678 * equivalent to an ACK
2680 ret = read_partial_ack(con);
2688 dout("try_read done on %p ret %d\n", con, ret);
2692 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2693 con->error_msg = "protocol error, garbage tag";
2700 * Atomically queue work on a connection after the specified delay.
2701 * Bump @con reference to avoid races with connection teardown.
2702 * Returns 0 if work was queued, or an error code otherwise.
2704 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2706 if (!con->ops->get(con)) {
2707 dout("%s %p ref count 0\n", __func__, con);
2711 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2712 dout("%s %p - already queued\n", __func__, con);
2717 dout("%s %p %lu\n", __func__, con, delay);
2721 static void queue_con(struct ceph_connection *con)
2723 (void) queue_con_delay(con, 0);
2726 static void cancel_con(struct ceph_connection *con)
2728 if (cancel_delayed_work(&con->work)) {
2729 dout("%s %p\n", __func__, con);
2734 static bool con_sock_closed(struct ceph_connection *con)
2736 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2740 case CON_STATE_ ## x: \
2741 con->error_msg = "socket closed (con state " #x ")"; \
2744 switch (con->state) {
2752 pr_warn("%s con %p unrecognized state %lu\n",
2753 __func__, con, con->state);
2754 con->error_msg = "unrecognized con state";
2763 static bool con_backoff(struct ceph_connection *con)
2767 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2770 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2772 dout("%s: con %p FAILED to back off %lu\n", __func__,
2774 BUG_ON(ret == -ENOENT);
2775 con_flag_set(con, CON_FLAG_BACKOFF);
2781 /* Finish fault handling; con->mutex must *not* be held here */
2783 static void con_fault_finish(struct ceph_connection *con)
2786 * in case we faulted due to authentication, invalidate our
2787 * current tickets so that we can get new ones.
2789 if (con->auth_retry && con->ops->invalidate_authorizer) {
2790 dout("calling invalidate_authorizer()\n");
2791 con->ops->invalidate_authorizer(con);
2794 if (con->ops->fault)
2795 con->ops->fault(con);
2799 * Do some work on a connection. Drop a connection ref when we're done.
2801 static void con_work(struct work_struct *work)
2803 struct ceph_connection *con = container_of(work, struct ceph_connection,
2807 mutex_lock(&con->mutex);
2811 if ((fault = con_sock_closed(con))) {
2812 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2815 if (con_backoff(con)) {
2816 dout("%s: con %p BACKOFF\n", __func__, con);
2819 if (con->state == CON_STATE_STANDBY) {
2820 dout("%s: con %p STANDBY\n", __func__, con);
2823 if (con->state == CON_STATE_CLOSED) {
2824 dout("%s: con %p CLOSED\n", __func__, con);
2828 if (con->state == CON_STATE_PREOPEN) {
2829 dout("%s: con %p PREOPEN\n", __func__, con);
2833 ret = try_read(con);
2837 if (!con->error_msg)
2838 con->error_msg = "socket error on read";
2843 ret = try_write(con);
2847 if (!con->error_msg)
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 dout("fault %p state %lu to peer %s\n",
2871 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2873 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2874 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2875 con->error_msg = NULL;
2877 WARN_ON(con->state != CON_STATE_CONNECTING &&
2878 con->state != CON_STATE_NEGOTIATING &&
2879 con->state != CON_STATE_OPEN);
2881 con_close_socket(con);
2883 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2884 dout("fault on LOSSYTX channel, marking CLOSED\n");
2885 con->state = CON_STATE_CLOSED;
2890 BUG_ON(con->in_msg->con != con);
2891 con->in_msg->con = NULL;
2892 ceph_msg_put(con->in_msg);
2897 /* Requeue anything that hasn't been acked */
2898 list_splice_init(&con->out_sent, &con->out_queue);
2900 /* If there are no messages queued or keepalive pending, place
2901 * the connection in a STANDBY state */
2902 if (list_empty(&con->out_queue) &&
2903 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
2904 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2905 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2906 con->state = CON_STATE_STANDBY;
2908 /* retry after a delay. */
2909 con->state = CON_STATE_PREOPEN;
2910 if (con->delay == 0)
2911 con->delay = BASE_DELAY_INTERVAL;
2912 else if (con->delay < MAX_DELAY_INTERVAL)
2914 con_flag_set(con, CON_FLAG_BACKOFF);
2922 * initialize a new messenger instance
2924 void ceph_messenger_init(struct ceph_messenger *msgr,
2925 struct ceph_entity_addr *myaddr,
2926 u64 supported_features,
2927 u64 required_features,
2931 msgr->supported_features = supported_features;
2932 msgr->required_features = required_features;
2934 spin_lock_init(&msgr->global_seq_lock);
2937 msgr->inst.addr = *myaddr;
2939 /* select a random nonce */
2940 msgr->inst.addr.type = 0;
2941 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2942 encode_my_addr(msgr);
2943 msgr->nocrc = nocrc;
2944 msgr->tcp_nodelay = tcp_nodelay;
2946 atomic_set(&msgr->stopping, 0);
2948 dout("%s %p\n", __func__, msgr);
2950 EXPORT_SYMBOL(ceph_messenger_init);
2952 static void clear_standby(struct ceph_connection *con)
2954 /* come back from STANDBY? */
2955 if (con->state == CON_STATE_STANDBY) {
2956 dout("clear_standby %p and ++connect_seq\n", con);
2957 con->state = CON_STATE_PREOPEN;
2959 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2960 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
2965 * Queue up an outgoing message on the given connection.
2967 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2970 msg->hdr.src = con->msgr->inst.name;
2971 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2972 msg->needs_out_seq = true;
2974 mutex_lock(&con->mutex);
2976 if (con->state == CON_STATE_CLOSED) {
2977 dout("con_send %p closed, dropping %p\n", con, msg);
2979 mutex_unlock(&con->mutex);
2983 BUG_ON(msg->con != NULL);
2984 msg->con = con->ops->get(con);
2985 BUG_ON(msg->con == NULL);
2987 BUG_ON(!list_empty(&msg->list_head));
2988 list_add_tail(&msg->list_head, &con->out_queue);
2989 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2990 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2991 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2992 le32_to_cpu(msg->hdr.front_len),
2993 le32_to_cpu(msg->hdr.middle_len),
2994 le32_to_cpu(msg->hdr.data_len));
2997 mutex_unlock(&con->mutex);
2999 /* if there wasn't anything waiting to send before, queue
3001 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3004 EXPORT_SYMBOL(ceph_con_send);
3007 * Revoke a message that was previously queued for send
3009 void ceph_msg_revoke(struct ceph_msg *msg)
3011 struct ceph_connection *con = msg->con;
3014 return; /* Message not in our possession */
3016 mutex_lock(&con->mutex);
3017 if (!list_empty(&msg->list_head)) {
3018 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
3019 list_del_init(&msg->list_head);
3020 BUG_ON(msg->con == NULL);
3021 msg->con->ops->put(msg->con);
3027 if (con->out_msg == msg) {
3028 dout("%s %p msg %p - was sending\n", __func__, con, msg);
3029 con->out_msg = NULL;
3030 if (con->out_kvec_is_msg) {
3031 con->out_skip = con->out_kvec_bytes;
3032 con->out_kvec_is_msg = false;
3038 mutex_unlock(&con->mutex);
3042 * Revoke a message that we may be reading data into
3044 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
3046 struct ceph_connection *con;
3048 BUG_ON(msg == NULL);
3050 dout("%s msg %p null con\n", __func__, msg);
3052 return; /* Message not in our possession */
3056 mutex_lock(&con->mutex);
3057 if (con->in_msg == msg) {
3058 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3059 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3060 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
3062 /* skip rest of message */
3063 dout("%s %p msg %p revoked\n", __func__, con, msg);
3064 con->in_base_pos = con->in_base_pos -
3065 sizeof(struct ceph_msg_header) -
3069 sizeof(struct ceph_msg_footer);
3070 ceph_msg_put(con->in_msg);
3072 con->in_tag = CEPH_MSGR_TAG_READY;
3075 dout("%s %p in_msg %p msg %p no-op\n",
3076 __func__, con, con->in_msg, msg);
3078 mutex_unlock(&con->mutex);
3082 * Queue a keepalive byte to ensure the tcp connection is alive.
3084 void ceph_con_keepalive(struct ceph_connection *con)
3086 dout("con_keepalive %p\n", con);
3087 mutex_lock(&con->mutex);
3089 mutex_unlock(&con->mutex);
3090 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3091 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3094 EXPORT_SYMBOL(ceph_con_keepalive);
3096 static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
3098 struct ceph_msg_data *data;
3100 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3103 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
3106 INIT_LIST_HEAD(&data->links);
3111 static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3116 WARN_ON(!list_empty(&data->links));
3117 if (data->type == CEPH_MSG_DATA_PAGELIST)
3118 ceph_pagelist_release(data->pagelist);
3119 kmem_cache_free(ceph_msg_data_cache, data);
3122 void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
3123 size_t length, size_t alignment)
3125 struct ceph_msg_data *data;
3130 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3132 data->pages = pages;
3133 data->length = length;
3134 data->alignment = alignment & ~PAGE_MASK;
3136 list_add_tail(&data->links, &msg->data);
3137 msg->data_length += length;
3139 EXPORT_SYMBOL(ceph_msg_data_add_pages);
3141 void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
3142 struct ceph_pagelist *pagelist)
3144 struct ceph_msg_data *data;
3147 BUG_ON(!pagelist->length);
3149 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3151 data->pagelist = pagelist;
3153 list_add_tail(&data->links, &msg->data);
3154 msg->data_length += pagelist->length;
3156 EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
3159 void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
3162 struct ceph_msg_data *data;
3166 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3169 data->bio_length = length;
3171 list_add_tail(&data->links, &msg->data);
3172 msg->data_length += length;
3174 EXPORT_SYMBOL(ceph_msg_data_add_bio);
3175 #endif /* CONFIG_BLOCK */
3178 * construct a new message with given type, size
3179 * the new msg has a ref count of 1.
3181 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3186 m = kmem_cache_zalloc(ceph_msg_cache, flags);
3190 m->hdr.type = cpu_to_le16(type);
3191 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3192 m->hdr.front_len = cpu_to_le32(front_len);
3194 INIT_LIST_HEAD(&m->list_head);
3195 kref_init(&m->kref);
3196 INIT_LIST_HEAD(&m->data);
3200 m->front.iov_base = ceph_kvmalloc(front_len, flags);
3201 if (m->front.iov_base == NULL) {
3202 dout("ceph_msg_new can't allocate %d bytes\n",
3207 m->front.iov_base = NULL;
3209 m->front_alloc_len = m->front.iov_len = front_len;
3211 dout("ceph_msg_new %p front %d\n", m, front_len);
3218 pr_err("msg_new can't create type %d front %d\n", type,
3222 dout("msg_new can't create type %d front %d\n", type,
3227 EXPORT_SYMBOL(ceph_msg_new);
3230 * Allocate "middle" portion of a message, if it is needed and wasn't
3231 * allocated by alloc_msg. This allows us to read a small fixed-size
3232 * per-type header in the front and then gracefully fail (i.e.,
3233 * propagate the error to the caller based on info in the front) when
3234 * the middle is too large.
3236 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3238 int type = le16_to_cpu(msg->hdr.type);
3239 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3241 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3242 ceph_msg_type_name(type), middle_len);
3243 BUG_ON(!middle_len);
3244 BUG_ON(msg->middle);
3246 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3253 * Allocate a message for receiving an incoming message on a
3254 * connection, and save the result in con->in_msg. Uses the
3255 * connection's private alloc_msg op if available.
3257 * Returns 0 on success, or a negative error code.
3259 * On success, if we set *skip = 1:
3260 * - the next message should be skipped and ignored.
3261 * - con->in_msg == NULL
3262 * or if we set *skip = 0:
3263 * - con->in_msg is non-null.
3264 * On error (ENOMEM, EAGAIN, ...),
3265 * - con->in_msg == NULL
3267 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3269 struct ceph_msg_header *hdr = &con->in_hdr;
3270 int middle_len = le32_to_cpu(hdr->middle_len);
3271 struct ceph_msg *msg;
3274 BUG_ON(con->in_msg != NULL);
3275 BUG_ON(!con->ops->alloc_msg);
3277 mutex_unlock(&con->mutex);
3278 msg = con->ops->alloc_msg(con, hdr, skip);
3279 mutex_lock(&con->mutex);
3280 if (con->state != CON_STATE_OPEN) {
3288 con->in_msg->con = con->ops->get(con);
3289 BUG_ON(con->in_msg->con == NULL);
3292 * Null message pointer means either we should skip
3293 * this message or we couldn't allocate memory. The
3294 * former is not an error.
3299 con->error_msg = "error allocating memory for incoming message";
3302 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3304 if (middle_len && !con->in_msg->middle) {
3305 ret = ceph_alloc_middle(con, con->in_msg);
3307 ceph_msg_put(con->in_msg);
3317 * Free a generically kmalloc'd message.
3319 static void ceph_msg_free(struct ceph_msg *m)
3321 dout("%s %p\n", __func__, m);
3322 kvfree(m->front.iov_base);
3323 kmem_cache_free(ceph_msg_cache, m);
3326 static void ceph_msg_release(struct kref *kref)
3328 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3330 struct list_head *links;
3331 struct list_head *next;
3333 dout("%s %p\n", __func__, m);
3334 WARN_ON(!list_empty(&m->list_head));
3336 /* drop middle, data, if any */
3338 ceph_buffer_put(m->middle);
3342 list_splice_init(&m->data, &data);
3343 list_for_each_safe(links, next, &data) {
3344 struct ceph_msg_data *data;
3346 data = list_entry(links, struct ceph_msg_data, links);
3347 list_del_init(links);
3348 ceph_msg_data_destroy(data);
3353 ceph_msgpool_put(m->pool, m);
3358 struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3360 dout("%s %p (was %d)\n", __func__, msg,
3361 atomic_read(&msg->kref.refcount));
3362 kref_get(&msg->kref);
3365 EXPORT_SYMBOL(ceph_msg_get);
3367 void ceph_msg_put(struct ceph_msg *msg)
3369 dout("%s %p (was %d)\n", __func__, msg,
3370 atomic_read(&msg->kref.refcount));
3371 kref_put(&msg->kref, ceph_msg_release);
3373 EXPORT_SYMBOL(ceph_msg_put);
3375 void ceph_msg_dump(struct ceph_msg *msg)
3377 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3378 msg->front_alloc_len, msg->data_length);
3379 print_hex_dump(KERN_DEBUG, "header: ",
3380 DUMP_PREFIX_OFFSET, 16, 1,
3381 &msg->hdr, sizeof(msg->hdr), true);
3382 print_hex_dump(KERN_DEBUG, " front: ",
3383 DUMP_PREFIX_OFFSET, 16, 1,
3384 msg->front.iov_base, msg->front.iov_len, true);
3386 print_hex_dump(KERN_DEBUG, "middle: ",
3387 DUMP_PREFIX_OFFSET, 16, 1,
3388 msg->middle->vec.iov_base,
3389 msg->middle->vec.iov_len, true);
3390 print_hex_dump(KERN_DEBUG, "footer: ",
3391 DUMP_PREFIX_OFFSET, 16, 1,
3392 &msg->footer, sizeof(msg->footer), true);
3394 EXPORT_SYMBOL(ceph_msg_dump);