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/nsproxy.h>
10 #include <linux/slab.h>
11 #include <linux/socket.h>
12 #include <linux/string.h>
14 #include <linux/bio.h>
15 #endif /* CONFIG_BLOCK */
16 #include <linux/dns_resolver.h>
19 #include <linux/ceph/ceph_features.h>
20 #include <linux/ceph/libceph.h>
21 #include <linux/ceph/messenger.h>
22 #include <linux/ceph/decode.h>
23 #include <linux/ceph/pagelist.h>
24 #include <linux/export.h>
26 #define list_entry_next(pos, member) \
27 list_entry(pos->member.next, typeof(*pos), member)
30 * Ceph uses the messenger to exchange ceph_msg messages with other
31 * hosts in the system. The messenger provides ordered and reliable
32 * delivery. We tolerate TCP disconnects by reconnecting (with
33 * exponential backoff) in the case of a fault (disconnection, bad
34 * crc, protocol error). Acks allow sent messages to be discarded by
39 * We track the state of the socket on a given connection using
40 * values defined below. The transition to a new socket state is
41 * handled by a function which verifies we aren't coming from an
45 * | NEW* | transient initial state
47 * | con_sock_state_init()
50 * | CLOSED | initialized, but no socket (and no
51 * ---------- TCP connection)
53 * | \ con_sock_state_connecting()
54 * | ----------------------
56 * + con_sock_state_closed() \
57 * |+--------------------------- \
60 * | | CLOSING | socket event; \ \
61 * | ----------- await close \ \
64 * | + con_sock_state_closing() \ |
66 * | / --------------- | |
69 * | / -----------------| CONNECTING | socket created, TCP
70 * | | / -------------- connect initiated
71 * | | | con_sock_state_connected()
74 * | CONNECTED | TCP connection established
77 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
80 #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
81 #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
82 #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
83 #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
84 #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
89 #define CON_STATE_CLOSED 1 /* -> PREOPEN */
90 #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
91 #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
92 #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
93 #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
94 #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
97 * ceph_connection flag bits
99 #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
100 * messages on errors */
101 #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
102 #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
103 #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
104 #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
106 static bool con_flag_valid(unsigned long con_flag)
109 case CON_FLAG_LOSSYTX:
110 case CON_FLAG_KEEPALIVE_PENDING:
111 case CON_FLAG_WRITE_PENDING:
112 case CON_FLAG_SOCK_CLOSED:
113 case CON_FLAG_BACKOFF:
120 static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
122 BUG_ON(!con_flag_valid(con_flag));
124 clear_bit(con_flag, &con->flags);
127 static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
129 BUG_ON(!con_flag_valid(con_flag));
131 set_bit(con_flag, &con->flags);
134 static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
136 BUG_ON(!con_flag_valid(con_flag));
138 return test_bit(con_flag, &con->flags);
141 static bool con_flag_test_and_clear(struct ceph_connection *con,
142 unsigned long con_flag)
144 BUG_ON(!con_flag_valid(con_flag));
146 return test_and_clear_bit(con_flag, &con->flags);
149 static bool con_flag_test_and_set(struct ceph_connection *con,
150 unsigned long con_flag)
152 BUG_ON(!con_flag_valid(con_flag));
154 return test_and_set_bit(con_flag, &con->flags);
157 /* Slab caches for frequently-allocated structures */
159 static struct kmem_cache *ceph_msg_cache;
160 static struct kmem_cache *ceph_msg_data_cache;
162 /* static tag bytes (protocol control messages) */
163 static char tag_msg = CEPH_MSGR_TAG_MSG;
164 static char tag_ack = CEPH_MSGR_TAG_ACK;
165 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
166 static char tag_keepalive2 = CEPH_MSGR_TAG_KEEPALIVE2;
168 #ifdef CONFIG_LOCKDEP
169 static struct lock_class_key socket_class;
173 * When skipping (ignoring) a block of input we read it into a "skip
174 * buffer," which is this many bytes in size.
176 #define SKIP_BUF_SIZE 1024
178 static void queue_con(struct ceph_connection *con);
179 static void cancel_con(struct ceph_connection *con);
180 static void ceph_con_workfn(struct work_struct *);
181 static void con_fault(struct ceph_connection *con);
184 * Nicely render a sockaddr as a string. An array of formatted
185 * strings is used, to approximate reentrancy.
187 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
188 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
189 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
190 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
192 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
193 static atomic_t addr_str_seq = ATOMIC_INIT(0);
195 static struct page *zero_page; /* used in certain error cases */
197 const char *ceph_pr_addr(const struct sockaddr_storage *ss)
201 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
202 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
204 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
207 switch (ss->ss_family) {
209 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
210 ntohs(in4->sin_port));
214 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
215 ntohs(in6->sin6_port));
219 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
225 EXPORT_SYMBOL(ceph_pr_addr);
227 static void encode_my_addr(struct ceph_messenger *msgr)
229 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
230 ceph_encode_addr(&msgr->my_enc_addr);
234 * work queue for all reading and writing to/from the socket.
236 static struct workqueue_struct *ceph_msgr_wq;
238 static int ceph_msgr_slab_init(void)
240 BUG_ON(ceph_msg_cache);
241 ceph_msg_cache = kmem_cache_create("ceph_msg",
242 sizeof (struct ceph_msg),
243 __alignof__(struct ceph_msg), 0, NULL);
248 BUG_ON(ceph_msg_data_cache);
249 ceph_msg_data_cache = kmem_cache_create("ceph_msg_data",
250 sizeof (struct ceph_msg_data),
251 __alignof__(struct ceph_msg_data),
253 if (ceph_msg_data_cache)
256 kmem_cache_destroy(ceph_msg_cache);
257 ceph_msg_cache = NULL;
262 static void ceph_msgr_slab_exit(void)
264 BUG_ON(!ceph_msg_data_cache);
265 kmem_cache_destroy(ceph_msg_data_cache);
266 ceph_msg_data_cache = NULL;
268 BUG_ON(!ceph_msg_cache);
269 kmem_cache_destroy(ceph_msg_cache);
270 ceph_msg_cache = NULL;
273 static void _ceph_msgr_exit(void)
276 destroy_workqueue(ceph_msgr_wq);
280 BUG_ON(zero_page == NULL);
281 page_cache_release(zero_page);
284 ceph_msgr_slab_exit();
287 int ceph_msgr_init(void)
289 if (ceph_msgr_slab_init())
292 BUG_ON(zero_page != NULL);
293 zero_page = ZERO_PAGE(0);
294 page_cache_get(zero_page);
297 * The number of active work items is limited by the number of
298 * connections, so leave @max_active at default.
300 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
304 pr_err("msgr_init failed to create workqueue\n");
309 EXPORT_SYMBOL(ceph_msgr_init);
311 void ceph_msgr_exit(void)
313 BUG_ON(ceph_msgr_wq == NULL);
317 EXPORT_SYMBOL(ceph_msgr_exit);
319 void ceph_msgr_flush(void)
321 flush_workqueue(ceph_msgr_wq);
323 EXPORT_SYMBOL(ceph_msgr_flush);
325 /* Connection socket state transition functions */
327 static void con_sock_state_init(struct ceph_connection *con)
331 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
332 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
333 printk("%s: unexpected old state %d\n", __func__, old_state);
334 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
335 CON_SOCK_STATE_CLOSED);
338 static void con_sock_state_connecting(struct ceph_connection *con)
342 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
343 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
344 printk("%s: unexpected old state %d\n", __func__, old_state);
345 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
346 CON_SOCK_STATE_CONNECTING);
349 static void con_sock_state_connected(struct ceph_connection *con)
353 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
354 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
355 printk("%s: unexpected old state %d\n", __func__, old_state);
356 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
357 CON_SOCK_STATE_CONNECTED);
360 static void con_sock_state_closing(struct ceph_connection *con)
364 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
365 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
366 old_state != CON_SOCK_STATE_CONNECTED &&
367 old_state != CON_SOCK_STATE_CLOSING))
368 printk("%s: unexpected old state %d\n", __func__, old_state);
369 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
370 CON_SOCK_STATE_CLOSING);
373 static void con_sock_state_closed(struct ceph_connection *con)
377 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
378 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
379 old_state != CON_SOCK_STATE_CLOSING &&
380 old_state != CON_SOCK_STATE_CONNECTING &&
381 old_state != CON_SOCK_STATE_CLOSED))
382 printk("%s: unexpected old state %d\n", __func__, old_state);
383 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
384 CON_SOCK_STATE_CLOSED);
388 * socket callback functions
391 /* data available on socket, or listen socket received a connect */
392 static void ceph_sock_data_ready(struct sock *sk)
394 struct ceph_connection *con = sk->sk_user_data;
395 if (atomic_read(&con->msgr->stopping)) {
399 if (sk->sk_state != TCP_CLOSE_WAIT) {
400 dout("%s on %p state = %lu, queueing work\n", __func__,
406 /* socket has buffer space for writing */
407 static void ceph_sock_write_space(struct sock *sk)
409 struct ceph_connection *con = sk->sk_user_data;
411 /* only queue to workqueue if there is data we want to write,
412 * and there is sufficient space in the socket buffer to accept
413 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
414 * doesn't get called again until try_write() fills the socket
415 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
416 * and net/core/stream.c:sk_stream_write_space().
418 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
419 if (sk_stream_is_writeable(sk)) {
420 dout("%s %p queueing write work\n", __func__, con);
421 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
425 dout("%s %p nothing to write\n", __func__, con);
429 /* socket's state has changed */
430 static void ceph_sock_state_change(struct sock *sk)
432 struct ceph_connection *con = sk->sk_user_data;
434 dout("%s %p state = %lu sk_state = %u\n", __func__,
435 con, con->state, sk->sk_state);
437 switch (sk->sk_state) {
439 dout("%s TCP_CLOSE\n", __func__);
441 dout("%s TCP_CLOSE_WAIT\n", __func__);
442 con_sock_state_closing(con);
443 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
446 case TCP_ESTABLISHED:
447 dout("%s TCP_ESTABLISHED\n", __func__);
448 con_sock_state_connected(con);
451 default: /* Everything else is uninteresting */
457 * set up socket callbacks
459 static void set_sock_callbacks(struct socket *sock,
460 struct ceph_connection *con)
462 struct sock *sk = sock->sk;
463 sk->sk_user_data = con;
464 sk->sk_data_ready = ceph_sock_data_ready;
465 sk->sk_write_space = ceph_sock_write_space;
466 sk->sk_state_change = ceph_sock_state_change;
475 * initiate connection to a remote socket.
477 static int ceph_tcp_connect(struct ceph_connection *con)
479 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
484 ret = sock_create_kern(read_pnet(&con->msgr->net), paddr->ss_family,
485 SOCK_STREAM, IPPROTO_TCP, &sock);
488 sock->sk->sk_allocation = GFP_NOFS;
490 #ifdef CONFIG_LOCKDEP
491 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
494 set_sock_callbacks(sock, con);
496 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
498 con_sock_state_connecting(con);
499 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
501 if (ret == -EINPROGRESS) {
502 dout("connect %s EINPROGRESS sk_state = %u\n",
503 ceph_pr_addr(&con->peer_addr.in_addr),
505 } else if (ret < 0) {
506 pr_err("connect %s error %d\n",
507 ceph_pr_addr(&con->peer_addr.in_addr), ret);
512 if (con->msgr->tcp_nodelay) {
515 ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
516 (char *)&optval, sizeof(optval));
518 pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
526 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
528 struct kvec iov = {buf, len};
529 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
532 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
538 static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
539 int page_offset, size_t length)
544 BUG_ON(page_offset + length > PAGE_SIZE);
548 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
555 * write something. @more is true if caller will be sending more data
558 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
559 size_t kvlen, size_t len, int more)
561 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
565 msg.msg_flags |= MSG_MORE;
567 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
569 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
575 static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
576 int offset, size_t size, bool more)
578 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
581 ret = kernel_sendpage(sock, page, offset, size, flags);
588 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
589 int offset, size_t size, bool more)
594 /* sendpage cannot properly handle pages with page_count == 0,
595 * we need to fallback to sendmsg if that's the case */
596 if (page_count(page) >= 1)
597 return __ceph_tcp_sendpage(sock, page, offset, size, more);
599 iov.iov_base = kmap(page) + offset;
601 ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more);
608 * Shutdown/close the socket for the given connection.
610 static int con_close_socket(struct ceph_connection *con)
614 dout("con_close_socket on %p sock %p\n", con, con->sock);
616 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
617 sock_release(con->sock);
622 * Forcibly clear the SOCK_CLOSED flag. It gets set
623 * independent of the connection mutex, and we could have
624 * received a socket close event before we had the chance to
625 * shut the socket down.
627 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
629 con_sock_state_closed(con);
634 * Reset a connection. Discard all incoming and outgoing messages
635 * and clear *_seq state.
637 static void ceph_msg_remove(struct ceph_msg *msg)
639 list_del_init(&msg->list_head);
640 BUG_ON(msg->con == NULL);
641 msg->con->ops->put(msg->con);
646 static void ceph_msg_remove_list(struct list_head *head)
648 while (!list_empty(head)) {
649 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
651 ceph_msg_remove(msg);
655 static void reset_connection(struct ceph_connection *con)
657 /* reset connection, out_queue, msg_ and connect_seq */
658 /* discard existing out_queue and msg_seq */
659 dout("reset_connection %p\n", con);
660 ceph_msg_remove_list(&con->out_queue);
661 ceph_msg_remove_list(&con->out_sent);
664 BUG_ON(con->in_msg->con != con);
665 con->in_msg->con = NULL;
666 ceph_msg_put(con->in_msg);
671 con->connect_seq = 0;
674 ceph_msg_put(con->out_msg);
678 con->in_seq_acked = 0;
682 * mark a peer down. drop any open connections.
684 void ceph_con_close(struct ceph_connection *con)
686 mutex_lock(&con->mutex);
687 dout("con_close %p peer %s\n", con,
688 ceph_pr_addr(&con->peer_addr.in_addr));
689 con->state = CON_STATE_CLOSED;
691 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
692 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
693 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
694 con_flag_clear(con, CON_FLAG_BACKOFF);
696 reset_connection(con);
697 con->peer_global_seq = 0;
699 con_close_socket(con);
700 mutex_unlock(&con->mutex);
702 EXPORT_SYMBOL(ceph_con_close);
705 * Reopen a closed connection, with a new peer address.
707 void ceph_con_open(struct ceph_connection *con,
708 __u8 entity_type, __u64 entity_num,
709 struct ceph_entity_addr *addr)
711 mutex_lock(&con->mutex);
712 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
714 WARN_ON(con->state != CON_STATE_CLOSED);
715 con->state = CON_STATE_PREOPEN;
717 con->peer_name.type = (__u8) entity_type;
718 con->peer_name.num = cpu_to_le64(entity_num);
720 memcpy(&con->peer_addr, addr, sizeof(*addr));
721 con->delay = 0; /* reset backoff memory */
722 mutex_unlock(&con->mutex);
725 EXPORT_SYMBOL(ceph_con_open);
728 * return true if this connection ever successfully opened
730 bool ceph_con_opened(struct ceph_connection *con)
732 return con->connect_seq > 0;
736 * initialize a new connection.
738 void ceph_con_init(struct ceph_connection *con, void *private,
739 const struct ceph_connection_operations *ops,
740 struct ceph_messenger *msgr)
742 dout("con_init %p\n", con);
743 memset(con, 0, sizeof(*con));
744 con->private = private;
748 con_sock_state_init(con);
750 mutex_init(&con->mutex);
751 INIT_LIST_HEAD(&con->out_queue);
752 INIT_LIST_HEAD(&con->out_sent);
753 INIT_DELAYED_WORK(&con->work, ceph_con_workfn);
755 con->state = CON_STATE_CLOSED;
757 EXPORT_SYMBOL(ceph_con_init);
761 * We maintain a global counter to order connection attempts. Get
762 * a unique seq greater than @gt.
764 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
768 spin_lock(&msgr->global_seq_lock);
769 if (msgr->global_seq < gt)
770 msgr->global_seq = gt;
771 ret = ++msgr->global_seq;
772 spin_unlock(&msgr->global_seq_lock);
776 static void con_out_kvec_reset(struct ceph_connection *con)
778 con->out_kvec_left = 0;
779 con->out_kvec_bytes = 0;
780 con->out_kvec_cur = &con->out_kvec[0];
783 static void con_out_kvec_add(struct ceph_connection *con,
784 size_t size, void *data)
788 index = con->out_kvec_left;
789 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
791 con->out_kvec[index].iov_len = size;
792 con->out_kvec[index].iov_base = data;
793 con->out_kvec_left++;
794 con->out_kvec_bytes += size;
800 * For a bio data item, a piece is whatever remains of the next
801 * entry in the current bio iovec, or the first entry in the next
804 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
807 struct ceph_msg_data *data = cursor->data;
810 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
815 cursor->resid = min(length, data->bio_length);
817 cursor->bvec_iter = bio->bi_iter;
819 cursor->resid <= bio_iter_len(bio, cursor->bvec_iter);
822 static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
826 struct ceph_msg_data *data = cursor->data;
828 struct bio_vec bio_vec;
830 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
835 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
837 *page_offset = (size_t) bio_vec.bv_offset;
838 BUG_ON(*page_offset >= PAGE_SIZE);
839 if (cursor->last_piece) /* pagelist offset is always 0 */
840 *length = cursor->resid;
842 *length = (size_t) bio_vec.bv_len;
843 BUG_ON(*length > cursor->resid);
844 BUG_ON(*page_offset + *length > PAGE_SIZE);
846 return bio_vec.bv_page;
849 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
853 struct bio_vec bio_vec;
855 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
860 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
862 /* Advance the cursor offset */
864 BUG_ON(cursor->resid < bytes);
865 cursor->resid -= bytes;
867 bio_advance_iter(bio, &cursor->bvec_iter, bytes);
869 if (bytes < bio_vec.bv_len)
870 return false; /* more bytes to process in this segment */
872 /* Move on to the next segment, and possibly the next bio */
874 if (!cursor->bvec_iter.bi_size) {
878 cursor->bvec_iter = bio->bi_iter;
880 memset(&cursor->bvec_iter, 0,
881 sizeof(cursor->bvec_iter));
884 if (!cursor->last_piece) {
885 BUG_ON(!cursor->resid);
887 /* A short read is OK, so use <= rather than == */
888 if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter))
889 cursor->last_piece = true;
894 #endif /* CONFIG_BLOCK */
897 * For a page array, a piece comes from the first page in the array
898 * that has not already been fully consumed.
900 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
903 struct ceph_msg_data *data = cursor->data;
906 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
908 BUG_ON(!data->pages);
909 BUG_ON(!data->length);
911 cursor->resid = min(length, data->length);
912 page_count = calc_pages_for(data->alignment, (u64)data->length);
913 cursor->page_offset = data->alignment & ~PAGE_MASK;
914 cursor->page_index = 0;
915 BUG_ON(page_count > (int)USHRT_MAX);
916 cursor->page_count = (unsigned short)page_count;
917 BUG_ON(length > SIZE_MAX - cursor->page_offset);
918 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
922 ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
923 size_t *page_offset, size_t *length)
925 struct ceph_msg_data *data = cursor->data;
927 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
929 BUG_ON(cursor->page_index >= cursor->page_count);
930 BUG_ON(cursor->page_offset >= PAGE_SIZE);
932 *page_offset = cursor->page_offset;
933 if (cursor->last_piece)
934 *length = cursor->resid;
936 *length = PAGE_SIZE - *page_offset;
938 return data->pages[cursor->page_index];
941 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
944 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
946 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
948 /* Advance the cursor page offset */
950 cursor->resid -= bytes;
951 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
952 if (!bytes || cursor->page_offset)
953 return false; /* more bytes to process in the current page */
956 return false; /* no more data */
958 /* Move on to the next page; offset is already at 0 */
960 BUG_ON(cursor->page_index >= cursor->page_count);
961 cursor->page_index++;
962 cursor->last_piece = cursor->resid <= PAGE_SIZE;
968 * For a pagelist, a piece is whatever remains to be consumed in the
969 * first page in the list, or the front of the next page.
972 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
975 struct ceph_msg_data *data = cursor->data;
976 struct ceph_pagelist *pagelist;
979 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
981 pagelist = data->pagelist;
985 return; /* pagelist can be assigned but empty */
987 BUG_ON(list_empty(&pagelist->head));
988 page = list_first_entry(&pagelist->head, struct page, lru);
990 cursor->resid = min(length, pagelist->length);
993 cursor->last_piece = cursor->resid <= PAGE_SIZE;
997 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
998 size_t *page_offset, size_t *length)
1000 struct ceph_msg_data *data = cursor->data;
1001 struct ceph_pagelist *pagelist;
1003 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1005 pagelist = data->pagelist;
1008 BUG_ON(!cursor->page);
1009 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1011 /* offset of first page in pagelist is always 0 */
1012 *page_offset = cursor->offset & ~PAGE_MASK;
1013 if (cursor->last_piece)
1014 *length = cursor->resid;
1016 *length = PAGE_SIZE - *page_offset;
1018 return cursor->page;
1021 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
1024 struct ceph_msg_data *data = cursor->data;
1025 struct ceph_pagelist *pagelist;
1027 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1029 pagelist = data->pagelist;
1032 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1033 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1035 /* Advance the cursor offset */
1037 cursor->resid -= bytes;
1038 cursor->offset += bytes;
1039 /* offset of first page in pagelist is always 0 */
1040 if (!bytes || cursor->offset & ~PAGE_MASK)
1041 return false; /* more bytes to process in the current page */
1044 return false; /* no more data */
1046 /* Move on to the next page */
1048 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1049 cursor->page = list_entry_next(cursor->page, lru);
1050 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1056 * Message data is handled (sent or received) in pieces, where each
1057 * piece resides on a single page. The network layer might not
1058 * consume an entire piece at once. A data item's cursor keeps
1059 * track of which piece is next to process and how much remains to
1060 * be processed in that piece. It also tracks whether the current
1061 * piece is the last one in the data item.
1063 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1065 size_t length = cursor->total_resid;
1067 switch (cursor->data->type) {
1068 case CEPH_MSG_DATA_PAGELIST:
1069 ceph_msg_data_pagelist_cursor_init(cursor, length);
1071 case CEPH_MSG_DATA_PAGES:
1072 ceph_msg_data_pages_cursor_init(cursor, length);
1075 case CEPH_MSG_DATA_BIO:
1076 ceph_msg_data_bio_cursor_init(cursor, length);
1078 #endif /* CONFIG_BLOCK */
1079 case CEPH_MSG_DATA_NONE:
1084 cursor->need_crc = true;
1087 static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1089 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1090 struct ceph_msg_data *data;
1093 BUG_ON(length > msg->data_length);
1094 BUG_ON(list_empty(&msg->data));
1096 cursor->data_head = &msg->data;
1097 cursor->total_resid = length;
1098 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1099 cursor->data = data;
1101 __ceph_msg_data_cursor_init(cursor);
1105 * Return the page containing the next piece to process for a given
1106 * data item, and supply the page offset and length of that piece.
1107 * Indicate whether this is the last piece in this data item.
1109 static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1110 size_t *page_offset, size_t *length,
1115 switch (cursor->data->type) {
1116 case CEPH_MSG_DATA_PAGELIST:
1117 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1119 case CEPH_MSG_DATA_PAGES:
1120 page = ceph_msg_data_pages_next(cursor, page_offset, length);
1123 case CEPH_MSG_DATA_BIO:
1124 page = ceph_msg_data_bio_next(cursor, page_offset, length);
1126 #endif /* CONFIG_BLOCK */
1127 case CEPH_MSG_DATA_NONE:
1133 BUG_ON(*page_offset + *length > PAGE_SIZE);
1136 *last_piece = cursor->last_piece;
1142 * Returns true if the result moves the cursor on to the next piece
1145 static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1150 BUG_ON(bytes > cursor->resid);
1151 switch (cursor->data->type) {
1152 case CEPH_MSG_DATA_PAGELIST:
1153 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1155 case CEPH_MSG_DATA_PAGES:
1156 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1159 case CEPH_MSG_DATA_BIO:
1160 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1162 #endif /* CONFIG_BLOCK */
1163 case CEPH_MSG_DATA_NONE:
1168 cursor->total_resid -= bytes;
1170 if (!cursor->resid && cursor->total_resid) {
1171 WARN_ON(!cursor->last_piece);
1172 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1173 cursor->data = list_entry_next(cursor->data, links);
1174 __ceph_msg_data_cursor_init(cursor);
1177 cursor->need_crc = new_piece;
1182 static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
1187 /* Initialize data cursor */
1189 ceph_msg_data_cursor_init(msg, (size_t)data_len);
1193 * Prepare footer for currently outgoing message, and finish things
1194 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1196 static void prepare_write_message_footer(struct ceph_connection *con)
1198 struct ceph_msg *m = con->out_msg;
1199 int v = con->out_kvec_left;
1201 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1203 dout("prepare_write_message_footer %p\n", con);
1204 con->out_kvec_is_msg = true;
1205 con->out_kvec[v].iov_base = &m->footer;
1206 if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
1207 if (con->ops->sign_message)
1208 con->ops->sign_message(con, m);
1211 con->out_kvec[v].iov_len = sizeof(m->footer);
1212 con->out_kvec_bytes += sizeof(m->footer);
1214 m->old_footer.flags = m->footer.flags;
1215 con->out_kvec[v].iov_len = sizeof(m->old_footer);
1216 con->out_kvec_bytes += sizeof(m->old_footer);
1218 con->out_kvec_left++;
1219 con->out_more = m->more_to_follow;
1220 con->out_msg_done = true;
1224 * Prepare headers for the next outgoing message.
1226 static void prepare_write_message(struct ceph_connection *con)
1231 con_out_kvec_reset(con);
1232 con->out_kvec_is_msg = true;
1233 con->out_msg_done = false;
1235 /* Sneak an ack in there first? If we can get it into the same
1236 * TCP packet that's a good thing. */
1237 if (con->in_seq > con->in_seq_acked) {
1238 con->in_seq_acked = con->in_seq;
1239 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1240 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1241 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1242 &con->out_temp_ack);
1245 BUG_ON(list_empty(&con->out_queue));
1246 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1248 BUG_ON(m->con != con);
1250 /* put message on sent list */
1252 list_move_tail(&m->list_head, &con->out_sent);
1255 * only assign outgoing seq # if we haven't sent this message
1256 * yet. if it is requeued, resend with it's original seq.
1258 if (m->needs_out_seq) {
1259 m->hdr.seq = cpu_to_le64(++con->out_seq);
1260 m->needs_out_seq = false;
1262 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
1264 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1265 m, con->out_seq, le16_to_cpu(m->hdr.type),
1266 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1268 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1270 /* tag + hdr + front + middle */
1271 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1272 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1273 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1276 con_out_kvec_add(con, m->middle->vec.iov_len,
1277 m->middle->vec.iov_base);
1279 /* fill in crc (except data pages), footer */
1280 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1281 con->out_msg->hdr.crc = cpu_to_le32(crc);
1282 con->out_msg->footer.flags = 0;
1284 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1285 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1287 crc = crc32c(0, m->middle->vec.iov_base,
1288 m->middle->vec.iov_len);
1289 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1291 con->out_msg->footer.middle_crc = 0;
1292 dout("%s front_crc %u middle_crc %u\n", __func__,
1293 le32_to_cpu(con->out_msg->footer.front_crc),
1294 le32_to_cpu(con->out_msg->footer.middle_crc));
1296 /* is there a data payload? */
1297 con->out_msg->footer.data_crc = 0;
1298 if (m->data_length) {
1299 prepare_message_data(con->out_msg, m->data_length);
1300 con->out_more = 1; /* data + footer will follow */
1302 /* no, queue up footer too and be done */
1303 prepare_write_message_footer(con);
1306 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1312 static void prepare_write_ack(struct ceph_connection *con)
1314 dout("prepare_write_ack %p %llu -> %llu\n", con,
1315 con->in_seq_acked, con->in_seq);
1316 con->in_seq_acked = con->in_seq;
1318 con_out_kvec_reset(con);
1320 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1322 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1323 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1324 &con->out_temp_ack);
1326 con->out_more = 1; /* more will follow.. eventually.. */
1327 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1331 * Prepare to share the seq during handshake
1333 static void prepare_write_seq(struct ceph_connection *con)
1335 dout("prepare_write_seq %p %llu -> %llu\n", con,
1336 con->in_seq_acked, con->in_seq);
1337 con->in_seq_acked = con->in_seq;
1339 con_out_kvec_reset(con);
1341 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1342 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1343 &con->out_temp_ack);
1345 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1349 * Prepare to write keepalive byte.
1351 static void prepare_write_keepalive(struct ceph_connection *con)
1353 dout("prepare_write_keepalive %p\n", con);
1354 con_out_kvec_reset(con);
1355 if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
1356 struct timespec ts = CURRENT_TIME;
1357 struct ceph_timespec ceph_ts;
1358 ceph_encode_timespec(&ceph_ts, &ts);
1359 con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
1360 con_out_kvec_add(con, sizeof(ceph_ts), &ceph_ts);
1362 con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive);
1364 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1368 * Connection negotiation.
1371 static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1374 struct ceph_auth_handshake *auth;
1376 if (!con->ops->get_authorizer) {
1377 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1378 con->out_connect.authorizer_len = 0;
1382 /* Can't hold the mutex while getting authorizer */
1383 mutex_unlock(&con->mutex);
1384 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
1385 mutex_lock(&con->mutex);
1389 if (con->state != CON_STATE_NEGOTIATING)
1390 return ERR_PTR(-EAGAIN);
1392 con->auth_reply_buf = auth->authorizer_reply_buf;
1393 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
1398 * We connected to a peer and are saying hello.
1400 static void prepare_write_banner(struct ceph_connection *con)
1402 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1403 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1404 &con->msgr->my_enc_addr);
1407 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1410 static int prepare_write_connect(struct ceph_connection *con)
1412 unsigned int global_seq = get_global_seq(con->msgr, 0);
1415 struct ceph_auth_handshake *auth;
1417 switch (con->peer_name.type) {
1418 case CEPH_ENTITY_TYPE_MON:
1419 proto = CEPH_MONC_PROTOCOL;
1421 case CEPH_ENTITY_TYPE_OSD:
1422 proto = CEPH_OSDC_PROTOCOL;
1424 case CEPH_ENTITY_TYPE_MDS:
1425 proto = CEPH_MDSC_PROTOCOL;
1431 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1432 con->connect_seq, global_seq, proto);
1434 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
1435 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1436 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1437 con->out_connect.global_seq = cpu_to_le32(global_seq);
1438 con->out_connect.protocol_version = cpu_to_le32(proto);
1439 con->out_connect.flags = 0;
1441 auth_proto = CEPH_AUTH_UNKNOWN;
1442 auth = get_connect_authorizer(con, &auth_proto);
1444 return PTR_ERR(auth);
1446 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1447 con->out_connect.authorizer_len = auth ?
1448 cpu_to_le32(auth->authorizer_buf_len) : 0;
1450 con_out_kvec_add(con, sizeof (con->out_connect),
1452 if (auth && auth->authorizer_buf_len)
1453 con_out_kvec_add(con, auth->authorizer_buf_len,
1454 auth->authorizer_buf);
1457 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1463 * write as much of pending kvecs to the socket as we can.
1465 * 0 -> socket full, but more to do
1468 static int write_partial_kvec(struct ceph_connection *con)
1472 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1473 while (con->out_kvec_bytes > 0) {
1474 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1475 con->out_kvec_left, con->out_kvec_bytes,
1479 con->out_kvec_bytes -= ret;
1480 if (con->out_kvec_bytes == 0)
1483 /* account for full iov entries consumed */
1484 while (ret >= con->out_kvec_cur->iov_len) {
1485 BUG_ON(!con->out_kvec_left);
1486 ret -= con->out_kvec_cur->iov_len;
1487 con->out_kvec_cur++;
1488 con->out_kvec_left--;
1490 /* and for a partially-consumed entry */
1492 con->out_kvec_cur->iov_len -= ret;
1493 con->out_kvec_cur->iov_base += ret;
1496 con->out_kvec_left = 0;
1497 con->out_kvec_is_msg = false;
1500 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1501 con->out_kvec_bytes, con->out_kvec_left, ret);
1502 return ret; /* done! */
1505 static u32 ceph_crc32c_page(u32 crc, struct page *page,
1506 unsigned int page_offset,
1507 unsigned int length)
1512 BUG_ON(kaddr == NULL);
1513 crc = crc32c(crc, kaddr + page_offset, length);
1519 * Write as much message data payload as we can. If we finish, queue
1521 * 1 -> done, footer is now queued in out_kvec[].
1522 * 0 -> socket full, but more to do
1525 static int write_partial_message_data(struct ceph_connection *con)
1527 struct ceph_msg *msg = con->out_msg;
1528 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1529 bool do_datacrc = !con->msgr->nocrc;
1532 dout("%s %p msg %p\n", __func__, con, msg);
1534 if (list_empty(&msg->data))
1538 * Iterate through each page that contains data to be
1539 * written, and send as much as possible for each.
1541 * If we are calculating the data crc (the default), we will
1542 * need to map the page. If we have no pages, they have
1543 * been revoked, so use the zero page.
1545 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1546 while (cursor->resid) {
1554 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
1556 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
1557 length, !last_piece);
1560 msg->footer.data_crc = cpu_to_le32(crc);
1564 if (do_datacrc && cursor->need_crc)
1565 crc = ceph_crc32c_page(crc, page, page_offset, length);
1566 need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret);
1569 dout("%s %p msg %p done\n", __func__, con, msg);
1571 /* prepare and queue up footer, too */
1573 msg->footer.data_crc = cpu_to_le32(crc);
1575 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1576 con_out_kvec_reset(con);
1577 prepare_write_message_footer(con);
1579 return 1; /* must return > 0 to indicate success */
1585 static int write_partial_skip(struct ceph_connection *con)
1589 while (con->out_skip > 0) {
1590 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
1592 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
1595 con->out_skip -= ret;
1603 * Prepare to read connection handshake, or an ack.
1605 static void prepare_read_banner(struct ceph_connection *con)
1607 dout("prepare_read_banner %p\n", con);
1608 con->in_base_pos = 0;
1611 static void prepare_read_connect(struct ceph_connection *con)
1613 dout("prepare_read_connect %p\n", con);
1614 con->in_base_pos = 0;
1617 static void prepare_read_ack(struct ceph_connection *con)
1619 dout("prepare_read_ack %p\n", con);
1620 con->in_base_pos = 0;
1623 static void prepare_read_seq(struct ceph_connection *con)
1625 dout("prepare_read_seq %p\n", con);
1626 con->in_base_pos = 0;
1627 con->in_tag = CEPH_MSGR_TAG_SEQ;
1630 static void prepare_read_tag(struct ceph_connection *con)
1632 dout("prepare_read_tag %p\n", con);
1633 con->in_base_pos = 0;
1634 con->in_tag = CEPH_MSGR_TAG_READY;
1637 static void prepare_read_keepalive_ack(struct ceph_connection *con)
1639 dout("prepare_read_keepalive_ack %p\n", con);
1640 con->in_base_pos = 0;
1644 * Prepare to read a message.
1646 static int prepare_read_message(struct ceph_connection *con)
1648 dout("prepare_read_message %p\n", con);
1649 BUG_ON(con->in_msg != NULL);
1650 con->in_base_pos = 0;
1651 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1656 static int read_partial(struct ceph_connection *con,
1657 int end, int size, void *object)
1659 while (con->in_base_pos < end) {
1660 int left = end - con->in_base_pos;
1661 int have = size - left;
1662 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1665 con->in_base_pos += ret;
1672 * Read all or part of the connect-side handshake on a new connection
1674 static int read_partial_banner(struct ceph_connection *con)
1680 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1683 size = strlen(CEPH_BANNER);
1685 ret = read_partial(con, end, size, con->in_banner);
1689 size = sizeof (con->actual_peer_addr);
1691 ret = read_partial(con, end, size, &con->actual_peer_addr);
1695 size = sizeof (con->peer_addr_for_me);
1697 ret = read_partial(con, end, size, &con->peer_addr_for_me);
1705 static int read_partial_connect(struct ceph_connection *con)
1711 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1713 size = sizeof (con->in_reply);
1715 ret = read_partial(con, end, size, &con->in_reply);
1719 size = le32_to_cpu(con->in_reply.authorizer_len);
1721 ret = read_partial(con, end, size, con->auth_reply_buf);
1725 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1726 con, (int)con->in_reply.tag,
1727 le32_to_cpu(con->in_reply.connect_seq),
1728 le32_to_cpu(con->in_reply.global_seq));
1735 * Verify the hello banner looks okay.
1737 static int verify_hello(struct ceph_connection *con)
1739 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1740 pr_err("connect to %s got bad banner\n",
1741 ceph_pr_addr(&con->peer_addr.in_addr));
1742 con->error_msg = "protocol error, bad banner";
1748 static bool addr_is_blank(struct sockaddr_storage *ss)
1750 struct in_addr *addr = &((struct sockaddr_in *)ss)->sin_addr;
1751 struct in6_addr *addr6 = &((struct sockaddr_in6 *)ss)->sin6_addr;
1753 switch (ss->ss_family) {
1755 return addr->s_addr == htonl(INADDR_ANY);
1757 return ipv6_addr_any(addr6);
1763 static int addr_port(struct sockaddr_storage *ss)
1765 switch (ss->ss_family) {
1767 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1769 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1774 static void addr_set_port(struct sockaddr_storage *ss, int p)
1776 switch (ss->ss_family) {
1778 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1781 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1787 * Unlike other *_pton function semantics, zero indicates success.
1789 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1790 char delim, const char **ipend)
1792 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1793 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1795 memset(ss, 0, sizeof(*ss));
1797 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1798 ss->ss_family = AF_INET;
1802 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1803 ss->ss_family = AF_INET6;
1811 * Extract hostname string and resolve using kernel DNS facility.
1813 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1814 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1815 struct sockaddr_storage *ss, char delim, const char **ipend)
1817 const char *end, *delim_p;
1818 char *colon_p, *ip_addr = NULL;
1822 * The end of the hostname occurs immediately preceding the delimiter or
1823 * the port marker (':') where the delimiter takes precedence.
1825 delim_p = memchr(name, delim, namelen);
1826 colon_p = memchr(name, ':', namelen);
1828 if (delim_p && colon_p)
1829 end = delim_p < colon_p ? delim_p : colon_p;
1830 else if (!delim_p && colon_p)
1834 if (!end) /* case: hostname:/ */
1835 end = name + namelen;
1841 /* do dns_resolve upcall */
1842 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1844 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1852 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1853 ret, ret ? "failed" : ceph_pr_addr(ss));
1858 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1859 struct sockaddr_storage *ss, char delim, const char **ipend)
1866 * Parse a server name (IP or hostname). If a valid IP address is not found
1867 * then try to extract a hostname to resolve using userspace DNS upcall.
1869 static int ceph_parse_server_name(const char *name, size_t namelen,
1870 struct sockaddr_storage *ss, char delim, const char **ipend)
1874 ret = ceph_pton(name, namelen, ss, delim, ipend);
1876 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1882 * Parse an ip[:port] list into an addr array. Use the default
1883 * monitor port if a port isn't specified.
1885 int ceph_parse_ips(const char *c, const char *end,
1886 struct ceph_entity_addr *addr,
1887 int max_count, int *count)
1889 int i, ret = -EINVAL;
1892 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1893 for (i = 0; i < max_count; i++) {
1895 struct sockaddr_storage *ss = &addr[i].in_addr;
1904 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1913 dout("missing matching ']'\n");
1920 if (p < end && *p == ':') {
1923 while (p < end && *p >= '0' && *p <= '9') {
1924 port = (port * 10) + (*p - '0');
1928 port = CEPH_MON_PORT;
1929 else if (port > 65535)
1932 port = CEPH_MON_PORT;
1935 addr_set_port(ss, port);
1937 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1954 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1957 EXPORT_SYMBOL(ceph_parse_ips);
1959 static int process_banner(struct ceph_connection *con)
1961 dout("process_banner on %p\n", con);
1963 if (verify_hello(con) < 0)
1966 ceph_decode_addr(&con->actual_peer_addr);
1967 ceph_decode_addr(&con->peer_addr_for_me);
1970 * Make sure the other end is who we wanted. note that the other
1971 * end may not yet know their ip address, so if it's 0.0.0.0, give
1972 * them the benefit of the doubt.
1974 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1975 sizeof(con->peer_addr)) != 0 &&
1976 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1977 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1978 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
1979 ceph_pr_addr(&con->peer_addr.in_addr),
1980 (int)le32_to_cpu(con->peer_addr.nonce),
1981 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1982 (int)le32_to_cpu(con->actual_peer_addr.nonce));
1983 con->error_msg = "wrong peer at address";
1988 * did we learn our address?
1990 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1991 int port = addr_port(&con->msgr->inst.addr.in_addr);
1993 memcpy(&con->msgr->inst.addr.in_addr,
1994 &con->peer_addr_for_me.in_addr,
1995 sizeof(con->peer_addr_for_me.in_addr));
1996 addr_set_port(&con->msgr->inst.addr.in_addr, port);
1997 encode_my_addr(con->msgr);
1998 dout("process_banner learned my addr is %s\n",
1999 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
2005 static int process_connect(struct ceph_connection *con)
2007 u64 sup_feat = con->msgr->supported_features;
2008 u64 req_feat = con->msgr->required_features;
2009 u64 server_feat = ceph_sanitize_features(
2010 le64_to_cpu(con->in_reply.features));
2013 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
2015 switch (con->in_reply.tag) {
2016 case CEPH_MSGR_TAG_FEATURES:
2017 pr_err("%s%lld %s feature set mismatch,"
2018 " my %llx < server's %llx, missing %llx\n",
2019 ENTITY_NAME(con->peer_name),
2020 ceph_pr_addr(&con->peer_addr.in_addr),
2021 sup_feat, server_feat, server_feat & ~sup_feat);
2022 con->error_msg = "missing required protocol features";
2023 reset_connection(con);
2026 case CEPH_MSGR_TAG_BADPROTOVER:
2027 pr_err("%s%lld %s protocol version mismatch,"
2028 " my %d != server's %d\n",
2029 ENTITY_NAME(con->peer_name),
2030 ceph_pr_addr(&con->peer_addr.in_addr),
2031 le32_to_cpu(con->out_connect.protocol_version),
2032 le32_to_cpu(con->in_reply.protocol_version));
2033 con->error_msg = "protocol version mismatch";
2034 reset_connection(con);
2037 case CEPH_MSGR_TAG_BADAUTHORIZER:
2039 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2041 if (con->auth_retry == 2) {
2042 con->error_msg = "connect authorization failure";
2045 con_out_kvec_reset(con);
2046 ret = prepare_write_connect(con);
2049 prepare_read_connect(con);
2052 case CEPH_MSGR_TAG_RESETSESSION:
2054 * If we connected with a large connect_seq but the peer
2055 * has no record of a session with us (no connection, or
2056 * connect_seq == 0), they will send RESETSESION to indicate
2057 * that they must have reset their session, and may have
2060 dout("process_connect got RESET peer seq %u\n",
2061 le32_to_cpu(con->in_reply.connect_seq));
2062 pr_err("%s%lld %s connection reset\n",
2063 ENTITY_NAME(con->peer_name),
2064 ceph_pr_addr(&con->peer_addr.in_addr));
2065 reset_connection(con);
2066 con_out_kvec_reset(con);
2067 ret = prepare_write_connect(con);
2070 prepare_read_connect(con);
2072 /* Tell ceph about it. */
2073 mutex_unlock(&con->mutex);
2074 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2075 if (con->ops->peer_reset)
2076 con->ops->peer_reset(con);
2077 mutex_lock(&con->mutex);
2078 if (con->state != CON_STATE_NEGOTIATING)
2082 case CEPH_MSGR_TAG_RETRY_SESSION:
2084 * If we sent a smaller connect_seq than the peer has, try
2085 * again with a larger value.
2087 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2088 le32_to_cpu(con->out_connect.connect_seq),
2089 le32_to_cpu(con->in_reply.connect_seq));
2090 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2091 con_out_kvec_reset(con);
2092 ret = prepare_write_connect(con);
2095 prepare_read_connect(con);
2098 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2100 * If we sent a smaller global_seq than the peer has, try
2101 * again with a larger value.
2103 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2104 con->peer_global_seq,
2105 le32_to_cpu(con->in_reply.global_seq));
2106 get_global_seq(con->msgr,
2107 le32_to_cpu(con->in_reply.global_seq));
2108 con_out_kvec_reset(con);
2109 ret = prepare_write_connect(con);
2112 prepare_read_connect(con);
2115 case CEPH_MSGR_TAG_SEQ:
2116 case CEPH_MSGR_TAG_READY:
2117 if (req_feat & ~server_feat) {
2118 pr_err("%s%lld %s protocol feature mismatch,"
2119 " my required %llx > server's %llx, need %llx\n",
2120 ENTITY_NAME(con->peer_name),
2121 ceph_pr_addr(&con->peer_addr.in_addr),
2122 req_feat, server_feat, req_feat & ~server_feat);
2123 con->error_msg = "missing required protocol features";
2124 reset_connection(con);
2128 WARN_ON(con->state != CON_STATE_NEGOTIATING);
2129 con->state = CON_STATE_OPEN;
2130 con->auth_retry = 0; /* we authenticated; clear flag */
2131 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2133 con->peer_features = server_feat;
2134 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2135 con->peer_global_seq,
2136 le32_to_cpu(con->in_reply.connect_seq),
2138 WARN_ON(con->connect_seq !=
2139 le32_to_cpu(con->in_reply.connect_seq));
2141 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2142 con_flag_set(con, CON_FLAG_LOSSYTX);
2144 con->delay = 0; /* reset backoff memory */
2146 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2147 prepare_write_seq(con);
2148 prepare_read_seq(con);
2150 prepare_read_tag(con);
2154 case CEPH_MSGR_TAG_WAIT:
2156 * If there is a connection race (we are opening
2157 * connections to each other), one of us may just have
2158 * to WAIT. This shouldn't happen if we are the
2161 con->error_msg = "protocol error, got WAIT as client";
2165 con->error_msg = "protocol error, garbage tag during connect";
2173 * read (part of) an ack
2175 static int read_partial_ack(struct ceph_connection *con)
2177 int size = sizeof (con->in_temp_ack);
2180 return read_partial(con, end, size, &con->in_temp_ack);
2184 * We can finally discard anything that's been acked.
2186 static void process_ack(struct ceph_connection *con)
2189 u64 ack = le64_to_cpu(con->in_temp_ack);
2192 while (!list_empty(&con->out_sent)) {
2193 m = list_first_entry(&con->out_sent, struct ceph_msg,
2195 seq = le64_to_cpu(m->hdr.seq);
2198 dout("got ack for seq %llu type %d at %p\n", seq,
2199 le16_to_cpu(m->hdr.type), m);
2200 m->ack_stamp = jiffies;
2203 prepare_read_tag(con);
2207 static int read_partial_message_section(struct ceph_connection *con,
2208 struct kvec *section,
2209 unsigned int sec_len, u32 *crc)
2215 while (section->iov_len < sec_len) {
2216 BUG_ON(section->iov_base == NULL);
2217 left = sec_len - section->iov_len;
2218 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2219 section->iov_len, left);
2222 section->iov_len += ret;
2224 if (section->iov_len == sec_len)
2225 *crc = crc32c(0, section->iov_base, section->iov_len);
2230 static int read_partial_msg_data(struct ceph_connection *con)
2232 struct ceph_msg *msg = con->in_msg;
2233 struct ceph_msg_data_cursor *cursor = &msg->cursor;
2234 const bool do_datacrc = !con->msgr->nocrc;
2242 if (list_empty(&msg->data))
2246 crc = con->in_data_crc;
2247 while (cursor->resid) {
2248 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
2250 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2253 con->in_data_crc = crc;
2259 crc = ceph_crc32c_page(crc, page, page_offset, ret);
2260 (void) ceph_msg_data_advance(&msg->cursor, (size_t)ret);
2263 con->in_data_crc = crc;
2265 return 1; /* must return > 0 to indicate success */
2269 * read (part of) a message.
2271 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2273 static int read_partial_message(struct ceph_connection *con)
2275 struct ceph_msg *m = con->in_msg;
2279 unsigned int front_len, middle_len, data_len;
2280 bool do_datacrc = !con->msgr->nocrc;
2281 bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
2285 dout("read_partial_message con %p msg %p\n", con, m);
2288 size = sizeof (con->in_hdr);
2290 ret = read_partial(con, end, size, &con->in_hdr);
2294 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2295 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2296 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
2297 crc, con->in_hdr.crc);
2301 front_len = le32_to_cpu(con->in_hdr.front_len);
2302 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2304 middle_len = le32_to_cpu(con->in_hdr.middle_len);
2305 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2307 data_len = le32_to_cpu(con->in_hdr.data_len);
2308 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2312 seq = le64_to_cpu(con->in_hdr.seq);
2313 if ((s64)seq - (s64)con->in_seq < 1) {
2314 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2315 ENTITY_NAME(con->peer_name),
2316 ceph_pr_addr(&con->peer_addr.in_addr),
2317 seq, con->in_seq + 1);
2318 con->in_base_pos = -front_len - middle_len - data_len -
2320 con->in_tag = CEPH_MSGR_TAG_READY;
2322 } else if ((s64)seq - (s64)con->in_seq > 1) {
2323 pr_err("read_partial_message bad seq %lld expected %lld\n",
2324 seq, con->in_seq + 1);
2325 con->error_msg = "bad message sequence # for incoming message";
2329 /* allocate message? */
2333 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2334 front_len, data_len);
2335 ret = ceph_con_in_msg_alloc(con, &skip);
2339 BUG_ON(!con->in_msg ^ skip);
2341 /* skip this message */
2342 dout("alloc_msg said skip message\n");
2343 con->in_base_pos = -front_len - middle_len - data_len -
2345 con->in_tag = CEPH_MSGR_TAG_READY;
2350 BUG_ON(!con->in_msg);
2351 BUG_ON(con->in_msg->con != con);
2353 m->front.iov_len = 0; /* haven't read it yet */
2355 m->middle->vec.iov_len = 0;
2357 /* prepare for data payload, if any */
2360 prepare_message_data(con->in_msg, data_len);
2364 ret = read_partial_message_section(con, &m->front, front_len,
2365 &con->in_front_crc);
2371 ret = read_partial_message_section(con, &m->middle->vec,
2373 &con->in_middle_crc);
2380 ret = read_partial_msg_data(con);
2387 size = sizeof(m->footer);
2389 size = sizeof(m->old_footer);
2392 ret = read_partial(con, end, size, &m->footer);
2397 m->footer.flags = m->old_footer.flags;
2401 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2402 m, front_len, m->footer.front_crc, middle_len,
2403 m->footer.middle_crc, data_len, m->footer.data_crc);
2406 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2407 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2408 m, con->in_front_crc, m->footer.front_crc);
2411 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2412 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2413 m, con->in_middle_crc, m->footer.middle_crc);
2417 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2418 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2419 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2420 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2424 if (need_sign && con->ops->check_message_signature &&
2425 con->ops->check_message_signature(con, m)) {
2426 pr_err("read_partial_message %p signature check failed\n", m);
2430 return 1; /* done! */
2434 * Process message. This happens in the worker thread. The callback should
2435 * be careful not to do anything that waits on other incoming messages or it
2438 static void process_message(struct ceph_connection *con)
2440 struct ceph_msg *msg;
2442 BUG_ON(con->in_msg->con != con);
2443 con->in_msg->con = NULL;
2448 /* if first message, set peer_name */
2449 if (con->peer_name.type == 0)
2450 con->peer_name = msg->hdr.src;
2453 mutex_unlock(&con->mutex);
2455 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2456 msg, le64_to_cpu(msg->hdr.seq),
2457 ENTITY_NAME(msg->hdr.src),
2458 le16_to_cpu(msg->hdr.type),
2459 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2460 le32_to_cpu(msg->hdr.front_len),
2461 le32_to_cpu(msg->hdr.data_len),
2462 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2463 con->ops->dispatch(con, msg);
2465 mutex_lock(&con->mutex);
2468 static int read_keepalive_ack(struct ceph_connection *con)
2470 struct ceph_timespec ceph_ts;
2471 size_t size = sizeof(ceph_ts);
2472 int ret = read_partial(con, size, size, &ceph_ts);
2475 ceph_decode_timespec(&con->last_keepalive_ack, &ceph_ts);
2476 prepare_read_tag(con);
2481 * Write something to the socket. Called in a worker thread when the
2482 * socket appears to be writeable and we have something ready to send.
2484 static int try_write(struct ceph_connection *con)
2488 dout("try_write start %p state %lu\n", con, con->state);
2491 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2493 /* open the socket first? */
2494 if (con->state == CON_STATE_PREOPEN) {
2496 con->state = CON_STATE_CONNECTING;
2498 con_out_kvec_reset(con);
2499 prepare_write_banner(con);
2500 prepare_read_banner(con);
2502 BUG_ON(con->in_msg);
2503 con->in_tag = CEPH_MSGR_TAG_READY;
2504 dout("try_write initiating connect on %p new state %lu\n",
2506 ret = ceph_tcp_connect(con);
2508 con->error_msg = "connect error";
2514 /* kvec data queued? */
2515 if (con->out_skip) {
2516 ret = write_partial_skip(con);
2520 if (con->out_kvec_left) {
2521 ret = write_partial_kvec(con);
2528 if (con->out_msg_done) {
2529 ceph_msg_put(con->out_msg);
2530 con->out_msg = NULL; /* we're done with this one */
2534 ret = write_partial_message_data(con);
2536 goto more_kvec; /* we need to send the footer, too! */
2540 dout("try_write write_partial_message_data err %d\n",
2547 if (con->state == CON_STATE_OPEN) {
2548 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2549 prepare_write_keepalive(con);
2552 /* is anything else pending? */
2553 if (!list_empty(&con->out_queue)) {
2554 prepare_write_message(con);
2557 if (con->in_seq > con->in_seq_acked) {
2558 prepare_write_ack(con);
2563 /* Nothing to do! */
2564 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2565 dout("try_write nothing else to write.\n");
2568 dout("try_write done on %p ret %d\n", con, ret);
2575 * Read what we can from the socket.
2577 static int try_read(struct ceph_connection *con)
2582 dout("try_read start on %p state %lu\n", con, con->state);
2583 if (con->state != CON_STATE_CONNECTING &&
2584 con->state != CON_STATE_NEGOTIATING &&
2585 con->state != CON_STATE_OPEN)
2590 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2593 if (con->state == CON_STATE_CONNECTING) {
2594 dout("try_read connecting\n");
2595 ret = read_partial_banner(con);
2598 ret = process_banner(con);
2602 con->state = CON_STATE_NEGOTIATING;
2605 * Received banner is good, exchange connection info.
2606 * Do not reset out_kvec, as sending our banner raced
2607 * with receiving peer banner after connect completed.
2609 ret = prepare_write_connect(con);
2612 prepare_read_connect(con);
2614 /* Send connection info before awaiting response */
2618 if (con->state == CON_STATE_NEGOTIATING) {
2619 dout("try_read negotiating\n");
2620 ret = read_partial_connect(con);
2623 ret = process_connect(con);
2629 WARN_ON(con->state != CON_STATE_OPEN);
2631 if (con->in_base_pos < 0) {
2633 * skipping + discarding content.
2635 * FIXME: there must be a better way to do this!
2637 static char buf[SKIP_BUF_SIZE];
2638 int skip = min((int) sizeof (buf), -con->in_base_pos);
2640 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2641 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2644 con->in_base_pos += ret;
2645 if (con->in_base_pos)
2648 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2652 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2655 dout("try_read got tag %d\n", (int)con->in_tag);
2656 switch (con->in_tag) {
2657 case CEPH_MSGR_TAG_MSG:
2658 prepare_read_message(con);
2660 case CEPH_MSGR_TAG_ACK:
2661 prepare_read_ack(con);
2663 case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
2664 prepare_read_keepalive_ack(con);
2666 case CEPH_MSGR_TAG_CLOSE:
2667 con_close_socket(con);
2668 con->state = CON_STATE_CLOSED;
2674 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2675 ret = read_partial_message(con);
2679 con->error_msg = "bad crc";
2685 con->error_msg = "io error";
2690 if (con->in_tag == CEPH_MSGR_TAG_READY)
2692 process_message(con);
2693 if (con->state == CON_STATE_OPEN)
2694 prepare_read_tag(con);
2697 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2698 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2700 * the final handshake seq exchange is semantically
2701 * equivalent to an ACK
2703 ret = read_partial_ack(con);
2709 if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
2710 ret = read_keepalive_ack(con);
2717 dout("try_read done on %p ret %d\n", con, ret);
2721 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2722 con->error_msg = "protocol error, garbage tag";
2729 * Atomically queue work on a connection after the specified delay.
2730 * Bump @con reference to avoid races with connection teardown.
2731 * Returns 0 if work was queued, or an error code otherwise.
2733 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2735 if (!con->ops->get(con)) {
2736 dout("%s %p ref count 0\n", __func__, con);
2740 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2741 dout("%s %p - already queued\n", __func__, con);
2746 dout("%s %p %lu\n", __func__, con, delay);
2750 static void queue_con(struct ceph_connection *con)
2752 (void) queue_con_delay(con, 0);
2755 static void cancel_con(struct ceph_connection *con)
2757 if (cancel_delayed_work(&con->work)) {
2758 dout("%s %p\n", __func__, con);
2763 static bool con_sock_closed(struct ceph_connection *con)
2765 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2769 case CON_STATE_ ## x: \
2770 con->error_msg = "socket closed (con state " #x ")"; \
2773 switch (con->state) {
2781 pr_warn("%s con %p unrecognized state %lu\n",
2782 __func__, con, con->state);
2783 con->error_msg = "unrecognized con state";
2792 static bool con_backoff(struct ceph_connection *con)
2796 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2799 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2801 dout("%s: con %p FAILED to back off %lu\n", __func__,
2803 BUG_ON(ret == -ENOENT);
2804 con_flag_set(con, CON_FLAG_BACKOFF);
2810 /* Finish fault handling; con->mutex must *not* be held here */
2812 static void con_fault_finish(struct ceph_connection *con)
2815 * in case we faulted due to authentication, invalidate our
2816 * current tickets so that we can get new ones.
2818 if (con->auth_retry && con->ops->invalidate_authorizer) {
2819 dout("calling invalidate_authorizer()\n");
2820 con->ops->invalidate_authorizer(con);
2823 if (con->ops->fault)
2824 con->ops->fault(con);
2828 * Do some work on a connection. Drop a connection ref when we're done.
2830 static void ceph_con_workfn(struct work_struct *work)
2832 struct ceph_connection *con = container_of(work, struct ceph_connection,
2836 mutex_lock(&con->mutex);
2840 if ((fault = con_sock_closed(con))) {
2841 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2844 if (con_backoff(con)) {
2845 dout("%s: con %p BACKOFF\n", __func__, con);
2848 if (con->state == CON_STATE_STANDBY) {
2849 dout("%s: con %p STANDBY\n", __func__, con);
2852 if (con->state == CON_STATE_CLOSED) {
2853 dout("%s: con %p CLOSED\n", __func__, con);
2857 if (con->state == CON_STATE_PREOPEN) {
2858 dout("%s: con %p PREOPEN\n", __func__, con);
2862 ret = try_read(con);
2866 if (!con->error_msg)
2867 con->error_msg = "socket error on read";
2872 ret = try_write(con);
2876 if (!con->error_msg)
2877 con->error_msg = "socket error on write";
2881 break; /* If we make it to here, we're done */
2885 mutex_unlock(&con->mutex);
2888 con_fault_finish(con);
2894 * Generic error/fault handler. A retry mechanism is used with
2895 * exponential backoff
2897 static void con_fault(struct ceph_connection *con)
2899 dout("fault %p state %lu to peer %s\n",
2900 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2902 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2903 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2904 con->error_msg = NULL;
2906 WARN_ON(con->state != CON_STATE_CONNECTING &&
2907 con->state != CON_STATE_NEGOTIATING &&
2908 con->state != CON_STATE_OPEN);
2910 con_close_socket(con);
2912 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2913 dout("fault on LOSSYTX channel, marking CLOSED\n");
2914 con->state = CON_STATE_CLOSED;
2919 BUG_ON(con->in_msg->con != con);
2920 con->in_msg->con = NULL;
2921 ceph_msg_put(con->in_msg);
2926 /* Requeue anything that hasn't been acked */
2927 list_splice_init(&con->out_sent, &con->out_queue);
2929 /* If there are no messages queued or keepalive pending, place
2930 * the connection in a STANDBY state */
2931 if (list_empty(&con->out_queue) &&
2932 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
2933 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2934 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2935 con->state = CON_STATE_STANDBY;
2937 /* retry after a delay. */
2938 con->state = CON_STATE_PREOPEN;
2939 if (con->delay == 0)
2940 con->delay = BASE_DELAY_INTERVAL;
2941 else if (con->delay < MAX_DELAY_INTERVAL)
2943 con_flag_set(con, CON_FLAG_BACKOFF);
2951 * initialize a new messenger instance
2953 void ceph_messenger_init(struct ceph_messenger *msgr,
2954 struct ceph_entity_addr *myaddr,
2955 u64 supported_features,
2956 u64 required_features,
2960 msgr->supported_features = supported_features;
2961 msgr->required_features = required_features;
2963 spin_lock_init(&msgr->global_seq_lock);
2966 msgr->inst.addr = *myaddr;
2968 /* select a random nonce */
2969 msgr->inst.addr.type = 0;
2970 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2971 encode_my_addr(msgr);
2972 msgr->nocrc = nocrc;
2973 msgr->tcp_nodelay = tcp_nodelay;
2975 atomic_set(&msgr->stopping, 0);
2976 write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
2978 dout("%s %p\n", __func__, msgr);
2980 EXPORT_SYMBOL(ceph_messenger_init);
2982 void ceph_messenger_fini(struct ceph_messenger *msgr)
2984 put_net(read_pnet(&msgr->net));
2986 EXPORT_SYMBOL(ceph_messenger_fini);
2988 static void clear_standby(struct ceph_connection *con)
2990 /* come back from STANDBY? */
2991 if (con->state == CON_STATE_STANDBY) {
2992 dout("clear_standby %p and ++connect_seq\n", con);
2993 con->state = CON_STATE_PREOPEN;
2995 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2996 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
3001 * Queue up an outgoing message on the given connection.
3003 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
3006 msg->hdr.src = con->msgr->inst.name;
3007 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
3008 msg->needs_out_seq = true;
3010 mutex_lock(&con->mutex);
3012 if (con->state == CON_STATE_CLOSED) {
3013 dout("con_send %p closed, dropping %p\n", con, msg);
3015 mutex_unlock(&con->mutex);
3019 BUG_ON(msg->con != NULL);
3020 msg->con = con->ops->get(con);
3021 BUG_ON(msg->con == NULL);
3023 BUG_ON(!list_empty(&msg->list_head));
3024 list_add_tail(&msg->list_head, &con->out_queue);
3025 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3026 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3027 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3028 le32_to_cpu(msg->hdr.front_len),
3029 le32_to_cpu(msg->hdr.middle_len),
3030 le32_to_cpu(msg->hdr.data_len));
3033 mutex_unlock(&con->mutex);
3035 /* if there wasn't anything waiting to send before, queue
3037 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3040 EXPORT_SYMBOL(ceph_con_send);
3043 * Revoke a message that was previously queued for send
3045 void ceph_msg_revoke(struct ceph_msg *msg)
3047 struct ceph_connection *con = msg->con;
3050 return; /* Message not in our possession */
3052 mutex_lock(&con->mutex);
3053 if (!list_empty(&msg->list_head)) {
3054 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
3055 list_del_init(&msg->list_head);
3056 BUG_ON(msg->con == NULL);
3057 msg->con->ops->put(msg->con);
3063 if (con->out_msg == msg) {
3064 dout("%s %p msg %p - was sending\n", __func__, con, msg);
3065 con->out_msg = NULL;
3066 if (con->out_kvec_is_msg) {
3067 con->out_skip = con->out_kvec_bytes;
3068 con->out_kvec_is_msg = false;
3074 mutex_unlock(&con->mutex);
3078 * Revoke a message that we may be reading data into
3080 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
3082 struct ceph_connection *con;
3084 BUG_ON(msg == NULL);
3086 dout("%s msg %p null con\n", __func__, msg);
3088 return; /* Message not in our possession */
3092 mutex_lock(&con->mutex);
3093 if (con->in_msg == msg) {
3094 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3095 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3096 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
3098 /* skip rest of message */
3099 dout("%s %p msg %p revoked\n", __func__, con, msg);
3100 con->in_base_pos = con->in_base_pos -
3101 sizeof(struct ceph_msg_header) -
3105 sizeof(struct ceph_msg_footer);
3106 ceph_msg_put(con->in_msg);
3108 con->in_tag = CEPH_MSGR_TAG_READY;
3111 dout("%s %p in_msg %p msg %p no-op\n",
3112 __func__, con, con->in_msg, msg);
3114 mutex_unlock(&con->mutex);
3118 * Queue a keepalive byte to ensure the tcp connection is alive.
3120 void ceph_con_keepalive(struct ceph_connection *con)
3122 dout("con_keepalive %p\n", con);
3123 mutex_lock(&con->mutex);
3125 mutex_unlock(&con->mutex);
3126 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3127 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3130 EXPORT_SYMBOL(ceph_con_keepalive);
3132 bool ceph_con_keepalive_expired(struct ceph_connection *con,
3133 unsigned long interval)
3136 (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
3137 struct timespec now = CURRENT_TIME;
3139 jiffies_to_timespec(interval, &ts);
3140 ts = timespec_add(con->last_keepalive_ack, ts);
3141 return timespec_compare(&now, &ts) >= 0;
3146 static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
3148 struct ceph_msg_data *data;
3150 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3153 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
3156 INIT_LIST_HEAD(&data->links);
3161 static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3166 WARN_ON(!list_empty(&data->links));
3167 if (data->type == CEPH_MSG_DATA_PAGELIST)
3168 ceph_pagelist_release(data->pagelist);
3169 kmem_cache_free(ceph_msg_data_cache, data);
3172 void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
3173 size_t length, size_t alignment)
3175 struct ceph_msg_data *data;
3180 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3182 data->pages = pages;
3183 data->length = length;
3184 data->alignment = alignment & ~PAGE_MASK;
3186 list_add_tail(&data->links, &msg->data);
3187 msg->data_length += length;
3189 EXPORT_SYMBOL(ceph_msg_data_add_pages);
3191 void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
3192 struct ceph_pagelist *pagelist)
3194 struct ceph_msg_data *data;
3197 BUG_ON(!pagelist->length);
3199 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3201 data->pagelist = pagelist;
3203 list_add_tail(&data->links, &msg->data);
3204 msg->data_length += pagelist->length;
3206 EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
3209 void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
3212 struct ceph_msg_data *data;
3216 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3219 data->bio_length = length;
3221 list_add_tail(&data->links, &msg->data);
3222 msg->data_length += length;
3224 EXPORT_SYMBOL(ceph_msg_data_add_bio);
3225 #endif /* CONFIG_BLOCK */
3228 * construct a new message with given type, size
3229 * the new msg has a ref count of 1.
3231 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3236 m = kmem_cache_zalloc(ceph_msg_cache, flags);
3240 m->hdr.type = cpu_to_le16(type);
3241 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3242 m->hdr.front_len = cpu_to_le32(front_len);
3244 INIT_LIST_HEAD(&m->list_head);
3245 kref_init(&m->kref);
3246 INIT_LIST_HEAD(&m->data);
3250 m->front.iov_base = ceph_kvmalloc(front_len, flags);
3251 if (m->front.iov_base == NULL) {
3252 dout("ceph_msg_new can't allocate %d bytes\n",
3257 m->front.iov_base = NULL;
3259 m->front_alloc_len = m->front.iov_len = front_len;
3261 dout("ceph_msg_new %p front %d\n", m, front_len);
3268 pr_err("msg_new can't create type %d front %d\n", type,
3272 dout("msg_new can't create type %d front %d\n", type,
3277 EXPORT_SYMBOL(ceph_msg_new);
3280 * Allocate "middle" portion of a message, if it is needed and wasn't
3281 * allocated by alloc_msg. This allows us to read a small fixed-size
3282 * per-type header in the front and then gracefully fail (i.e.,
3283 * propagate the error to the caller based on info in the front) when
3284 * the middle is too large.
3286 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3288 int type = le16_to_cpu(msg->hdr.type);
3289 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3291 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3292 ceph_msg_type_name(type), middle_len);
3293 BUG_ON(!middle_len);
3294 BUG_ON(msg->middle);
3296 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3303 * Allocate a message for receiving an incoming message on a
3304 * connection, and save the result in con->in_msg. Uses the
3305 * connection's private alloc_msg op if available.
3307 * Returns 0 on success, or a negative error code.
3309 * On success, if we set *skip = 1:
3310 * - the next message should be skipped and ignored.
3311 * - con->in_msg == NULL
3312 * or if we set *skip = 0:
3313 * - con->in_msg is non-null.
3314 * On error (ENOMEM, EAGAIN, ...),
3315 * - con->in_msg == NULL
3317 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3319 struct ceph_msg_header *hdr = &con->in_hdr;
3320 int middle_len = le32_to_cpu(hdr->middle_len);
3321 struct ceph_msg *msg;
3324 BUG_ON(con->in_msg != NULL);
3325 BUG_ON(!con->ops->alloc_msg);
3327 mutex_unlock(&con->mutex);
3328 msg = con->ops->alloc_msg(con, hdr, skip);
3329 mutex_lock(&con->mutex);
3330 if (con->state != CON_STATE_OPEN) {
3338 con->in_msg->con = con->ops->get(con);
3339 BUG_ON(con->in_msg->con == NULL);
3342 * Null message pointer means either we should skip
3343 * this message or we couldn't allocate memory. The
3344 * former is not an error.
3349 con->error_msg = "error allocating memory for incoming message";
3352 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3354 if (middle_len && !con->in_msg->middle) {
3355 ret = ceph_alloc_middle(con, con->in_msg);
3357 ceph_msg_put(con->in_msg);
3367 * Free a generically kmalloc'd message.
3369 static void ceph_msg_free(struct ceph_msg *m)
3371 dout("%s %p\n", __func__, m);
3372 kvfree(m->front.iov_base);
3373 kmem_cache_free(ceph_msg_cache, m);
3376 static void ceph_msg_release(struct kref *kref)
3378 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3380 struct list_head *links;
3381 struct list_head *next;
3383 dout("%s %p\n", __func__, m);
3384 WARN_ON(!list_empty(&m->list_head));
3386 /* drop middle, data, if any */
3388 ceph_buffer_put(m->middle);
3392 list_splice_init(&m->data, &data);
3393 list_for_each_safe(links, next, &data) {
3394 struct ceph_msg_data *data;
3396 data = list_entry(links, struct ceph_msg_data, links);
3397 list_del_init(links);
3398 ceph_msg_data_destroy(data);
3403 ceph_msgpool_put(m->pool, m);
3408 struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3410 dout("%s %p (was %d)\n", __func__, msg,
3411 atomic_read(&msg->kref.refcount));
3412 kref_get(&msg->kref);
3415 EXPORT_SYMBOL(ceph_msg_get);
3417 void ceph_msg_put(struct ceph_msg *msg)
3419 dout("%s %p (was %d)\n", __func__, msg,
3420 atomic_read(&msg->kref.refcount));
3421 kref_put(&msg->kref, ceph_msg_release);
3423 EXPORT_SYMBOL(ceph_msg_put);
3425 void ceph_msg_dump(struct ceph_msg *msg)
3427 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3428 msg->front_alloc_len, msg->data_length);
3429 print_hex_dump(KERN_DEBUG, "header: ",
3430 DUMP_PREFIX_OFFSET, 16, 1,
3431 &msg->hdr, sizeof(msg->hdr), true);
3432 print_hex_dump(KERN_DEBUG, " front: ",
3433 DUMP_PREFIX_OFFSET, 16, 1,
3434 msg->front.iov_base, msg->front.iov_len, true);
3436 print_hex_dump(KERN_DEBUG, "middle: ",
3437 DUMP_PREFIX_OFFSET, 16, 1,
3438 msg->middle->vec.iov_base,
3439 msg->middle->vec.iov_len, true);
3440 print_hex_dump(KERN_DEBUG, "footer: ",
3441 DUMP_PREFIX_OFFSET, 16, 1,
3442 &msg->footer, sizeof(msg->footer), true);
3444 EXPORT_SYMBOL(ceph_msg_dump);