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);
282 page_cache_release(zero_page);
286 int ceph_msgr_init(void)
288 BUG_ON(zero_page != NULL);
289 zero_page = ZERO_PAGE(0);
290 page_cache_get(zero_page);
292 if (ceph_msgr_slab_init())
295 ceph_msgr_wq = alloc_workqueue("ceph-msgr", 0, 0);
299 pr_err("msgr_init failed to create workqueue\n");
304 EXPORT_SYMBOL(ceph_msgr_init);
306 void ceph_msgr_exit(void)
308 BUG_ON(ceph_msgr_wq == NULL);
312 EXPORT_SYMBOL(ceph_msgr_exit);
314 void ceph_msgr_flush(void)
316 flush_workqueue(ceph_msgr_wq);
318 EXPORT_SYMBOL(ceph_msgr_flush);
320 /* Connection socket state transition functions */
322 static void con_sock_state_init(struct ceph_connection *con)
326 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
327 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
328 printk("%s: unexpected old state %d\n", __func__, old_state);
329 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
330 CON_SOCK_STATE_CLOSED);
333 static void con_sock_state_connecting(struct ceph_connection *con)
337 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
338 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
339 printk("%s: unexpected old state %d\n", __func__, old_state);
340 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
341 CON_SOCK_STATE_CONNECTING);
344 static void con_sock_state_connected(struct ceph_connection *con)
348 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
349 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
350 printk("%s: unexpected old state %d\n", __func__, old_state);
351 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
352 CON_SOCK_STATE_CONNECTED);
355 static void con_sock_state_closing(struct ceph_connection *con)
359 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
360 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
361 old_state != CON_SOCK_STATE_CONNECTED &&
362 old_state != CON_SOCK_STATE_CLOSING))
363 printk("%s: unexpected old state %d\n", __func__, old_state);
364 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
365 CON_SOCK_STATE_CLOSING);
368 static void con_sock_state_closed(struct ceph_connection *con)
372 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
373 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
374 old_state != CON_SOCK_STATE_CLOSING &&
375 old_state != CON_SOCK_STATE_CONNECTING &&
376 old_state != CON_SOCK_STATE_CLOSED))
377 printk("%s: unexpected old state %d\n", __func__, old_state);
378 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
379 CON_SOCK_STATE_CLOSED);
383 * socket callback functions
386 /* data available on socket, or listen socket received a connect */
387 static void ceph_sock_data_ready(struct sock *sk)
389 struct ceph_connection *con = sk->sk_user_data;
390 if (atomic_read(&con->msgr->stopping)) {
394 if (sk->sk_state != TCP_CLOSE_WAIT) {
395 dout("%s on %p state = %lu, queueing work\n", __func__,
401 /* socket has buffer space for writing */
402 static void ceph_sock_write_space(struct sock *sk)
404 struct ceph_connection *con = sk->sk_user_data;
406 /* only queue to workqueue if there is data we want to write,
407 * and there is sufficient space in the socket buffer to accept
408 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
409 * doesn't get called again until try_write() fills the socket
410 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
411 * and net/core/stream.c:sk_stream_write_space().
413 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
414 if (sk_stream_is_writeable(sk)) {
415 dout("%s %p queueing write work\n", __func__, con);
416 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
420 dout("%s %p nothing to write\n", __func__, con);
424 /* socket's state has changed */
425 static void ceph_sock_state_change(struct sock *sk)
427 struct ceph_connection *con = sk->sk_user_data;
429 dout("%s %p state = %lu sk_state = %u\n", __func__,
430 con, con->state, sk->sk_state);
432 switch (sk->sk_state) {
434 dout("%s TCP_CLOSE\n", __func__);
436 dout("%s TCP_CLOSE_WAIT\n", __func__);
437 con_sock_state_closing(con);
438 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
441 case TCP_ESTABLISHED:
442 dout("%s TCP_ESTABLISHED\n", __func__);
443 con_sock_state_connected(con);
446 default: /* Everything else is uninteresting */
452 * set up socket callbacks
454 static void set_sock_callbacks(struct socket *sock,
455 struct ceph_connection *con)
457 struct sock *sk = sock->sk;
458 sk->sk_user_data = con;
459 sk->sk_data_ready = ceph_sock_data_ready;
460 sk->sk_write_space = ceph_sock_write_space;
461 sk->sk_state_change = ceph_sock_state_change;
470 * initiate connection to a remote socket.
472 static int ceph_tcp_connect(struct ceph_connection *con)
474 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
479 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
483 sock->sk->sk_allocation = GFP_NOFS;
485 #ifdef CONFIG_LOCKDEP
486 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
489 set_sock_callbacks(sock, con);
491 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
493 con_sock_state_connecting(con);
494 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
496 if (ret == -EINPROGRESS) {
497 dout("connect %s EINPROGRESS sk_state = %u\n",
498 ceph_pr_addr(&con->peer_addr.in_addr),
500 } else if (ret < 0) {
501 pr_err("connect %s error %d\n",
502 ceph_pr_addr(&con->peer_addr.in_addr), ret);
504 con->error_msg = "connect error";
512 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
514 struct kvec iov = {buf, len};
515 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
518 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
524 static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
525 int page_offset, size_t length)
530 BUG_ON(page_offset + length > PAGE_SIZE);
534 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
541 * write something. @more is true if caller will be sending more data
544 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
545 size_t kvlen, size_t len, int more)
547 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
551 msg.msg_flags |= MSG_MORE;
553 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
555 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
561 static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
562 int offset, size_t size, bool more)
564 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
567 ret = kernel_sendpage(sock, page, offset, size, flags);
574 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
575 int offset, size_t size, bool more)
580 /* sendpage cannot properly handle pages with page_count == 0,
581 * we need to fallback to sendmsg if that's the case */
582 if (page_count(page) >= 1)
583 return __ceph_tcp_sendpage(sock, page, offset, size, more);
585 iov.iov_base = kmap(page) + offset;
587 ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more);
594 * Shutdown/close the socket for the given connection.
596 static int con_close_socket(struct ceph_connection *con)
600 dout("con_close_socket on %p sock %p\n", con, con->sock);
602 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
603 sock_release(con->sock);
608 * Forcibly clear the SOCK_CLOSED flag. It gets set
609 * independent of the connection mutex, and we could have
610 * received a socket close event before we had the chance to
611 * shut the socket down.
613 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
615 con_sock_state_closed(con);
620 * Reset a connection. Discard all incoming and outgoing messages
621 * and clear *_seq state.
623 static void ceph_msg_remove(struct ceph_msg *msg)
625 list_del_init(&msg->list_head);
626 BUG_ON(msg->con == NULL);
627 msg->con->ops->put(msg->con);
632 static void ceph_msg_remove_list(struct list_head *head)
634 while (!list_empty(head)) {
635 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
637 ceph_msg_remove(msg);
641 static void reset_connection(struct ceph_connection *con)
643 /* reset connection, out_queue, msg_ and connect_seq */
644 /* discard existing out_queue and msg_seq */
645 dout("reset_connection %p\n", con);
646 ceph_msg_remove_list(&con->out_queue);
647 ceph_msg_remove_list(&con->out_sent);
650 BUG_ON(con->in_msg->con != con);
651 con->in_msg->con = NULL;
652 ceph_msg_put(con->in_msg);
657 con->connect_seq = 0;
660 ceph_msg_put(con->out_msg);
664 con->in_seq_acked = 0;
668 * mark a peer down. drop any open connections.
670 void ceph_con_close(struct ceph_connection *con)
672 mutex_lock(&con->mutex);
673 dout("con_close %p peer %s\n", con,
674 ceph_pr_addr(&con->peer_addr.in_addr));
675 con->state = CON_STATE_CLOSED;
677 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
678 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
679 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
680 con_flag_clear(con, CON_FLAG_BACKOFF);
682 reset_connection(con);
683 con->peer_global_seq = 0;
685 con_close_socket(con);
686 mutex_unlock(&con->mutex);
688 EXPORT_SYMBOL(ceph_con_close);
691 * Reopen a closed connection, with a new peer address.
693 void ceph_con_open(struct ceph_connection *con,
694 __u8 entity_type, __u64 entity_num,
695 struct ceph_entity_addr *addr)
697 mutex_lock(&con->mutex);
698 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
700 WARN_ON(con->state != CON_STATE_CLOSED);
701 con->state = CON_STATE_PREOPEN;
703 con->peer_name.type = (__u8) entity_type;
704 con->peer_name.num = cpu_to_le64(entity_num);
706 memcpy(&con->peer_addr, addr, sizeof(*addr));
707 con->delay = 0; /* reset backoff memory */
708 mutex_unlock(&con->mutex);
711 EXPORT_SYMBOL(ceph_con_open);
714 * return true if this connection ever successfully opened
716 bool ceph_con_opened(struct ceph_connection *con)
718 return con->connect_seq > 0;
722 * initialize a new connection.
724 void ceph_con_init(struct ceph_connection *con, void *private,
725 const struct ceph_connection_operations *ops,
726 struct ceph_messenger *msgr)
728 dout("con_init %p\n", con);
729 memset(con, 0, sizeof(*con));
730 con->private = private;
734 con_sock_state_init(con);
736 mutex_init(&con->mutex);
737 INIT_LIST_HEAD(&con->out_queue);
738 INIT_LIST_HEAD(&con->out_sent);
739 INIT_DELAYED_WORK(&con->work, con_work);
741 con->state = CON_STATE_CLOSED;
743 EXPORT_SYMBOL(ceph_con_init);
747 * We maintain a global counter to order connection attempts. Get
748 * a unique seq greater than @gt.
750 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
754 spin_lock(&msgr->global_seq_lock);
755 if (msgr->global_seq < gt)
756 msgr->global_seq = gt;
757 ret = ++msgr->global_seq;
758 spin_unlock(&msgr->global_seq_lock);
762 static void con_out_kvec_reset(struct ceph_connection *con)
764 con->out_kvec_left = 0;
765 con->out_kvec_bytes = 0;
766 con->out_kvec_cur = &con->out_kvec[0];
769 static void con_out_kvec_add(struct ceph_connection *con,
770 size_t size, void *data)
774 index = con->out_kvec_left;
775 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
777 con->out_kvec[index].iov_len = size;
778 con->out_kvec[index].iov_base = data;
779 con->out_kvec_left++;
780 con->out_kvec_bytes += size;
786 * For a bio data item, a piece is whatever remains of the next
787 * entry in the current bio iovec, or the first entry in the next
790 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
793 struct ceph_msg_data *data = cursor->data;
796 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
801 cursor->resid = min(length, data->bio_length);
803 cursor->bvec_iter = bio->bi_iter;
805 cursor->resid <= bio_iter_len(bio, cursor->bvec_iter);
808 static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
812 struct ceph_msg_data *data = cursor->data;
814 struct bio_vec bio_vec;
816 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
821 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
823 *page_offset = (size_t) bio_vec.bv_offset;
824 BUG_ON(*page_offset >= PAGE_SIZE);
825 if (cursor->last_piece) /* pagelist offset is always 0 */
826 *length = cursor->resid;
828 *length = (size_t) bio_vec.bv_len;
829 BUG_ON(*length > cursor->resid);
830 BUG_ON(*page_offset + *length > PAGE_SIZE);
832 return bio_vec.bv_page;
835 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
839 struct bio_vec bio_vec;
841 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
846 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
848 /* Advance the cursor offset */
850 BUG_ON(cursor->resid < bytes);
851 cursor->resid -= bytes;
853 bio_advance_iter(bio, &cursor->bvec_iter, bytes);
855 if (bytes < bio_vec.bv_len)
856 return false; /* more bytes to process in this segment */
858 /* Move on to the next segment, and possibly the next bio */
860 if (!cursor->bvec_iter.bi_size) {
864 cursor->bvec_iter = bio->bi_iter;
866 memset(&cursor->bvec_iter, 0,
867 sizeof(cursor->bvec_iter));
870 if (!cursor->last_piece) {
871 BUG_ON(!cursor->resid);
873 /* A short read is OK, so use <= rather than == */
874 if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter))
875 cursor->last_piece = true;
880 #endif /* CONFIG_BLOCK */
883 * For a page array, a piece comes from the first page in the array
884 * that has not already been fully consumed.
886 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
889 struct ceph_msg_data *data = cursor->data;
892 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
894 BUG_ON(!data->pages);
895 BUG_ON(!data->length);
897 cursor->resid = min(length, data->length);
898 page_count = calc_pages_for(data->alignment, (u64)data->length);
899 cursor->page_offset = data->alignment & ~PAGE_MASK;
900 cursor->page_index = 0;
901 BUG_ON(page_count > (int)USHRT_MAX);
902 cursor->page_count = (unsigned short)page_count;
903 BUG_ON(length > SIZE_MAX - cursor->page_offset);
904 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
908 ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
909 size_t *page_offset, size_t *length)
911 struct ceph_msg_data *data = cursor->data;
913 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
915 BUG_ON(cursor->page_index >= cursor->page_count);
916 BUG_ON(cursor->page_offset >= PAGE_SIZE);
918 *page_offset = cursor->page_offset;
919 if (cursor->last_piece)
920 *length = cursor->resid;
922 *length = PAGE_SIZE - *page_offset;
924 return data->pages[cursor->page_index];
927 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
930 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
932 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
934 /* Advance the cursor page offset */
936 cursor->resid -= bytes;
937 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
938 if (!bytes || cursor->page_offset)
939 return false; /* more bytes to process in the current page */
942 return false; /* no more data */
944 /* Move on to the next page; offset is already at 0 */
946 BUG_ON(cursor->page_index >= cursor->page_count);
947 cursor->page_index++;
948 cursor->last_piece = cursor->resid <= PAGE_SIZE;
954 * For a pagelist, a piece is whatever remains to be consumed in the
955 * first page in the list, or the front of the next page.
958 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
961 struct ceph_msg_data *data = cursor->data;
962 struct ceph_pagelist *pagelist;
965 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
967 pagelist = data->pagelist;
971 return; /* pagelist can be assigned but empty */
973 BUG_ON(list_empty(&pagelist->head));
974 page = list_first_entry(&pagelist->head, struct page, lru);
976 cursor->resid = min(length, pagelist->length);
979 cursor->last_piece = cursor->resid <= PAGE_SIZE;
983 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
984 size_t *page_offset, size_t *length)
986 struct ceph_msg_data *data = cursor->data;
987 struct ceph_pagelist *pagelist;
989 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
991 pagelist = data->pagelist;
994 BUG_ON(!cursor->page);
995 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
997 /* offset of first page in pagelist is always 0 */
998 *page_offset = cursor->offset & ~PAGE_MASK;
999 if (cursor->last_piece)
1000 *length = cursor->resid;
1002 *length = PAGE_SIZE - *page_offset;
1004 return cursor->page;
1007 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
1010 struct ceph_msg_data *data = cursor->data;
1011 struct ceph_pagelist *pagelist;
1013 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1015 pagelist = data->pagelist;
1018 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1019 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1021 /* Advance the cursor offset */
1023 cursor->resid -= bytes;
1024 cursor->offset += bytes;
1025 /* offset of first page in pagelist is always 0 */
1026 if (!bytes || cursor->offset & ~PAGE_MASK)
1027 return false; /* more bytes to process in the current page */
1030 return false; /* no more data */
1032 /* Move on to the next page */
1034 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1035 cursor->page = list_entry_next(cursor->page, lru);
1036 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1042 * Message data is handled (sent or received) in pieces, where each
1043 * piece resides on a single page. The network layer might not
1044 * consume an entire piece at once. A data item's cursor keeps
1045 * track of which piece is next to process and how much remains to
1046 * be processed in that piece. It also tracks whether the current
1047 * piece is the last one in the data item.
1049 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1051 size_t length = cursor->total_resid;
1053 switch (cursor->data->type) {
1054 case CEPH_MSG_DATA_PAGELIST:
1055 ceph_msg_data_pagelist_cursor_init(cursor, length);
1057 case CEPH_MSG_DATA_PAGES:
1058 ceph_msg_data_pages_cursor_init(cursor, length);
1061 case CEPH_MSG_DATA_BIO:
1062 ceph_msg_data_bio_cursor_init(cursor, length);
1064 #endif /* CONFIG_BLOCK */
1065 case CEPH_MSG_DATA_NONE:
1070 cursor->need_crc = true;
1073 static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1075 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1076 struct ceph_msg_data *data;
1079 BUG_ON(length > msg->data_length);
1080 BUG_ON(list_empty(&msg->data));
1082 cursor->data_head = &msg->data;
1083 cursor->total_resid = length;
1084 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1085 cursor->data = data;
1087 __ceph_msg_data_cursor_init(cursor);
1091 * Return the page containing the next piece to process for a given
1092 * data item, and supply the page offset and length of that piece.
1093 * Indicate whether this is the last piece in this data item.
1095 static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1096 size_t *page_offset, size_t *length,
1101 switch (cursor->data->type) {
1102 case CEPH_MSG_DATA_PAGELIST:
1103 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1105 case CEPH_MSG_DATA_PAGES:
1106 page = ceph_msg_data_pages_next(cursor, page_offset, length);
1109 case CEPH_MSG_DATA_BIO:
1110 page = ceph_msg_data_bio_next(cursor, page_offset, length);
1112 #endif /* CONFIG_BLOCK */
1113 case CEPH_MSG_DATA_NONE:
1119 BUG_ON(*page_offset + *length > PAGE_SIZE);
1122 *last_piece = cursor->last_piece;
1128 * Returns true if the result moves the cursor on to the next piece
1131 static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1136 BUG_ON(bytes > cursor->resid);
1137 switch (cursor->data->type) {
1138 case CEPH_MSG_DATA_PAGELIST:
1139 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1141 case CEPH_MSG_DATA_PAGES:
1142 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1145 case CEPH_MSG_DATA_BIO:
1146 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1148 #endif /* CONFIG_BLOCK */
1149 case CEPH_MSG_DATA_NONE:
1154 cursor->total_resid -= bytes;
1156 if (!cursor->resid && cursor->total_resid) {
1157 WARN_ON(!cursor->last_piece);
1158 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1159 cursor->data = list_entry_next(cursor->data, links);
1160 __ceph_msg_data_cursor_init(cursor);
1163 cursor->need_crc = new_piece;
1168 static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
1173 /* Initialize data cursor */
1175 ceph_msg_data_cursor_init(msg, (size_t)data_len);
1179 * Prepare footer for currently outgoing message, and finish things
1180 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1182 static void prepare_write_message_footer(struct ceph_connection *con)
1184 struct ceph_msg *m = con->out_msg;
1185 int v = con->out_kvec_left;
1187 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1189 dout("prepare_write_message_footer %p\n", con);
1190 con->out_kvec_is_msg = true;
1191 con->out_kvec[v].iov_base = &m->footer;
1192 con->out_kvec[v].iov_len = sizeof(m->footer);
1193 con->out_kvec_bytes += sizeof(m->footer);
1194 con->out_kvec_left++;
1195 con->out_more = m->more_to_follow;
1196 con->out_msg_done = true;
1200 * Prepare headers for the next outgoing message.
1202 static void prepare_write_message(struct ceph_connection *con)
1207 con_out_kvec_reset(con);
1208 con->out_kvec_is_msg = true;
1209 con->out_msg_done = false;
1211 /* Sneak an ack in there first? If we can get it into the same
1212 * TCP packet that's a good thing. */
1213 if (con->in_seq > con->in_seq_acked) {
1214 con->in_seq_acked = con->in_seq;
1215 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1216 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1217 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1218 &con->out_temp_ack);
1221 BUG_ON(list_empty(&con->out_queue));
1222 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1224 BUG_ON(m->con != con);
1226 /* put message on sent list */
1228 list_move_tail(&m->list_head, &con->out_sent);
1231 * only assign outgoing seq # if we haven't sent this message
1232 * yet. if it is requeued, resend with it's original seq.
1234 if (m->needs_out_seq) {
1235 m->hdr.seq = cpu_to_le64(++con->out_seq);
1236 m->needs_out_seq = false;
1238 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
1240 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1241 m, con->out_seq, le16_to_cpu(m->hdr.type),
1242 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1244 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1246 /* tag + hdr + front + middle */
1247 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1248 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1249 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1252 con_out_kvec_add(con, m->middle->vec.iov_len,
1253 m->middle->vec.iov_base);
1255 /* fill in crc (except data pages), footer */
1256 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1257 con->out_msg->hdr.crc = cpu_to_le32(crc);
1258 con->out_msg->footer.flags = 0;
1260 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1261 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1263 crc = crc32c(0, m->middle->vec.iov_base,
1264 m->middle->vec.iov_len);
1265 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1267 con->out_msg->footer.middle_crc = 0;
1268 dout("%s front_crc %u middle_crc %u\n", __func__,
1269 le32_to_cpu(con->out_msg->footer.front_crc),
1270 le32_to_cpu(con->out_msg->footer.middle_crc));
1272 /* is there a data payload? */
1273 con->out_msg->footer.data_crc = 0;
1274 if (m->data_length) {
1275 prepare_message_data(con->out_msg, m->data_length);
1276 con->out_more = 1; /* data + footer will follow */
1278 /* no, queue up footer too and be done */
1279 prepare_write_message_footer(con);
1282 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1288 static void prepare_write_ack(struct ceph_connection *con)
1290 dout("prepare_write_ack %p %llu -> %llu\n", con,
1291 con->in_seq_acked, con->in_seq);
1292 con->in_seq_acked = con->in_seq;
1294 con_out_kvec_reset(con);
1296 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1298 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1299 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1300 &con->out_temp_ack);
1302 con->out_more = 1; /* more will follow.. eventually.. */
1303 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1307 * Prepare to share the seq during handshake
1309 static void prepare_write_seq(struct ceph_connection *con)
1311 dout("prepare_write_seq %p %llu -> %llu\n", con,
1312 con->in_seq_acked, con->in_seq);
1313 con->in_seq_acked = con->in_seq;
1315 con_out_kvec_reset(con);
1317 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1318 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1319 &con->out_temp_ack);
1321 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1325 * Prepare to write keepalive byte.
1327 static void prepare_write_keepalive(struct ceph_connection *con)
1329 dout("prepare_write_keepalive %p\n", con);
1330 con_out_kvec_reset(con);
1331 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
1332 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1336 * Connection negotiation.
1339 static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1342 struct ceph_auth_handshake *auth;
1344 if (!con->ops->get_authorizer) {
1345 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1346 con->out_connect.authorizer_len = 0;
1350 /* Can't hold the mutex while getting authorizer */
1351 mutex_unlock(&con->mutex);
1352 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
1353 mutex_lock(&con->mutex);
1357 if (con->state != CON_STATE_NEGOTIATING)
1358 return ERR_PTR(-EAGAIN);
1360 con->auth_reply_buf = auth->authorizer_reply_buf;
1361 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
1366 * We connected to a peer and are saying hello.
1368 static void prepare_write_banner(struct ceph_connection *con)
1370 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1371 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1372 &con->msgr->my_enc_addr);
1375 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1378 static int prepare_write_connect(struct ceph_connection *con)
1380 unsigned int global_seq = get_global_seq(con->msgr, 0);
1383 struct ceph_auth_handshake *auth;
1385 switch (con->peer_name.type) {
1386 case CEPH_ENTITY_TYPE_MON:
1387 proto = CEPH_MONC_PROTOCOL;
1389 case CEPH_ENTITY_TYPE_OSD:
1390 proto = CEPH_OSDC_PROTOCOL;
1392 case CEPH_ENTITY_TYPE_MDS:
1393 proto = CEPH_MDSC_PROTOCOL;
1399 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1400 con->connect_seq, global_seq, proto);
1402 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
1403 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1404 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1405 con->out_connect.global_seq = cpu_to_le32(global_seq);
1406 con->out_connect.protocol_version = cpu_to_le32(proto);
1407 con->out_connect.flags = 0;
1409 auth_proto = CEPH_AUTH_UNKNOWN;
1410 auth = get_connect_authorizer(con, &auth_proto);
1412 return PTR_ERR(auth);
1414 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1415 con->out_connect.authorizer_len = auth ?
1416 cpu_to_le32(auth->authorizer_buf_len) : 0;
1418 con_out_kvec_add(con, sizeof (con->out_connect),
1420 if (auth && auth->authorizer_buf_len)
1421 con_out_kvec_add(con, auth->authorizer_buf_len,
1422 auth->authorizer_buf);
1425 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1431 * write as much of pending kvecs to the socket as we can.
1433 * 0 -> socket full, but more to do
1436 static int write_partial_kvec(struct ceph_connection *con)
1440 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1441 while (con->out_kvec_bytes > 0) {
1442 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1443 con->out_kvec_left, con->out_kvec_bytes,
1447 con->out_kvec_bytes -= ret;
1448 if (con->out_kvec_bytes == 0)
1451 /* account for full iov entries consumed */
1452 while (ret >= con->out_kvec_cur->iov_len) {
1453 BUG_ON(!con->out_kvec_left);
1454 ret -= con->out_kvec_cur->iov_len;
1455 con->out_kvec_cur++;
1456 con->out_kvec_left--;
1458 /* and for a partially-consumed entry */
1460 con->out_kvec_cur->iov_len -= ret;
1461 con->out_kvec_cur->iov_base += ret;
1464 con->out_kvec_left = 0;
1465 con->out_kvec_is_msg = false;
1468 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1469 con->out_kvec_bytes, con->out_kvec_left, ret);
1470 return ret; /* done! */
1473 static u32 ceph_crc32c_page(u32 crc, struct page *page,
1474 unsigned int page_offset,
1475 unsigned int length)
1480 BUG_ON(kaddr == NULL);
1481 crc = crc32c(crc, kaddr + page_offset, length);
1487 * Write as much message data payload as we can. If we finish, queue
1489 * 1 -> done, footer is now queued in out_kvec[].
1490 * 0 -> socket full, but more to do
1493 static int write_partial_message_data(struct ceph_connection *con)
1495 struct ceph_msg *msg = con->out_msg;
1496 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1497 bool do_datacrc = !con->msgr->nocrc;
1500 dout("%s %p msg %p\n", __func__, con, msg);
1502 if (list_empty(&msg->data))
1506 * Iterate through each page that contains data to be
1507 * written, and send as much as possible for each.
1509 * If we are calculating the data crc (the default), we will
1510 * need to map the page. If we have no pages, they have
1511 * been revoked, so use the zero page.
1513 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1514 while (cursor->resid) {
1522 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
1524 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
1525 length, last_piece);
1528 msg->footer.data_crc = cpu_to_le32(crc);
1532 if (do_datacrc && cursor->need_crc)
1533 crc = ceph_crc32c_page(crc, page, page_offset, length);
1534 need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret);
1537 dout("%s %p msg %p done\n", __func__, con, msg);
1539 /* prepare and queue up footer, too */
1541 msg->footer.data_crc = cpu_to_le32(crc);
1543 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1544 con_out_kvec_reset(con);
1545 prepare_write_message_footer(con);
1547 return 1; /* must return > 0 to indicate success */
1553 static int write_partial_skip(struct ceph_connection *con)
1557 while (con->out_skip > 0) {
1558 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
1560 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
1563 con->out_skip -= ret;
1571 * Prepare to read connection handshake, or an ack.
1573 static void prepare_read_banner(struct ceph_connection *con)
1575 dout("prepare_read_banner %p\n", con);
1576 con->in_base_pos = 0;
1579 static void prepare_read_connect(struct ceph_connection *con)
1581 dout("prepare_read_connect %p\n", con);
1582 con->in_base_pos = 0;
1585 static void prepare_read_ack(struct ceph_connection *con)
1587 dout("prepare_read_ack %p\n", con);
1588 con->in_base_pos = 0;
1591 static void prepare_read_seq(struct ceph_connection *con)
1593 dout("prepare_read_seq %p\n", con);
1594 con->in_base_pos = 0;
1595 con->in_tag = CEPH_MSGR_TAG_SEQ;
1598 static void prepare_read_tag(struct ceph_connection *con)
1600 dout("prepare_read_tag %p\n", con);
1601 con->in_base_pos = 0;
1602 con->in_tag = CEPH_MSGR_TAG_READY;
1606 * Prepare to read a message.
1608 static int prepare_read_message(struct ceph_connection *con)
1610 dout("prepare_read_message %p\n", con);
1611 BUG_ON(con->in_msg != NULL);
1612 con->in_base_pos = 0;
1613 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1618 static int read_partial(struct ceph_connection *con,
1619 int end, int size, void *object)
1621 while (con->in_base_pos < end) {
1622 int left = end - con->in_base_pos;
1623 int have = size - left;
1624 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1627 con->in_base_pos += ret;
1634 * Read all or part of the connect-side handshake on a new connection
1636 static int read_partial_banner(struct ceph_connection *con)
1642 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1645 size = strlen(CEPH_BANNER);
1647 ret = read_partial(con, end, size, con->in_banner);
1651 size = sizeof (con->actual_peer_addr);
1653 ret = read_partial(con, end, size, &con->actual_peer_addr);
1657 size = sizeof (con->peer_addr_for_me);
1659 ret = read_partial(con, end, size, &con->peer_addr_for_me);
1667 static int read_partial_connect(struct ceph_connection *con)
1673 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1675 size = sizeof (con->in_reply);
1677 ret = read_partial(con, end, size, &con->in_reply);
1681 size = le32_to_cpu(con->in_reply.authorizer_len);
1683 ret = read_partial(con, end, size, con->auth_reply_buf);
1687 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1688 con, (int)con->in_reply.tag,
1689 le32_to_cpu(con->in_reply.connect_seq),
1690 le32_to_cpu(con->in_reply.global_seq));
1697 * Verify the hello banner looks okay.
1699 static int verify_hello(struct ceph_connection *con)
1701 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1702 pr_err("connect to %s got bad banner\n",
1703 ceph_pr_addr(&con->peer_addr.in_addr));
1704 con->error_msg = "protocol error, bad banner";
1710 static bool addr_is_blank(struct sockaddr_storage *ss)
1712 switch (ss->ss_family) {
1714 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1717 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1718 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1719 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1720 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1725 static int addr_port(struct sockaddr_storage *ss)
1727 switch (ss->ss_family) {
1729 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1731 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1736 static void addr_set_port(struct sockaddr_storage *ss, int p)
1738 switch (ss->ss_family) {
1740 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1743 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1749 * Unlike other *_pton function semantics, zero indicates success.
1751 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1752 char delim, const char **ipend)
1754 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1755 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1757 memset(ss, 0, sizeof(*ss));
1759 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1760 ss->ss_family = AF_INET;
1764 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1765 ss->ss_family = AF_INET6;
1773 * Extract hostname string and resolve using kernel DNS facility.
1775 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1776 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1777 struct sockaddr_storage *ss, char delim, const char **ipend)
1779 const char *end, *delim_p;
1780 char *colon_p, *ip_addr = NULL;
1784 * The end of the hostname occurs immediately preceding the delimiter or
1785 * the port marker (':') where the delimiter takes precedence.
1787 delim_p = memchr(name, delim, namelen);
1788 colon_p = memchr(name, ':', namelen);
1790 if (delim_p && colon_p)
1791 end = delim_p < colon_p ? delim_p : colon_p;
1792 else if (!delim_p && colon_p)
1796 if (!end) /* case: hostname:/ */
1797 end = name + namelen;
1803 /* do dns_resolve upcall */
1804 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1806 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1814 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1815 ret, ret ? "failed" : ceph_pr_addr(ss));
1820 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1821 struct sockaddr_storage *ss, char delim, const char **ipend)
1828 * Parse a server name (IP or hostname). If a valid IP address is not found
1829 * then try to extract a hostname to resolve using userspace DNS upcall.
1831 static int ceph_parse_server_name(const char *name, size_t namelen,
1832 struct sockaddr_storage *ss, char delim, const char **ipend)
1836 ret = ceph_pton(name, namelen, ss, delim, ipend);
1838 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1844 * Parse an ip[:port] list into an addr array. Use the default
1845 * monitor port if a port isn't specified.
1847 int ceph_parse_ips(const char *c, const char *end,
1848 struct ceph_entity_addr *addr,
1849 int max_count, int *count)
1851 int i, ret = -EINVAL;
1854 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1855 for (i = 0; i < max_count; i++) {
1857 struct sockaddr_storage *ss = &addr[i].in_addr;
1866 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1875 dout("missing matching ']'\n");
1882 if (p < end && *p == ':') {
1885 while (p < end && *p >= '0' && *p <= '9') {
1886 port = (port * 10) + (*p - '0');
1890 port = CEPH_MON_PORT;
1891 else if (port > 65535)
1894 port = CEPH_MON_PORT;
1897 addr_set_port(ss, port);
1899 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1916 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1919 EXPORT_SYMBOL(ceph_parse_ips);
1921 static int process_banner(struct ceph_connection *con)
1923 dout("process_banner on %p\n", con);
1925 if (verify_hello(con) < 0)
1928 ceph_decode_addr(&con->actual_peer_addr);
1929 ceph_decode_addr(&con->peer_addr_for_me);
1932 * Make sure the other end is who we wanted. note that the other
1933 * end may not yet know their ip address, so if it's 0.0.0.0, give
1934 * them the benefit of the doubt.
1936 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1937 sizeof(con->peer_addr)) != 0 &&
1938 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1939 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1940 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1941 ceph_pr_addr(&con->peer_addr.in_addr),
1942 (int)le32_to_cpu(con->peer_addr.nonce),
1943 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1944 (int)le32_to_cpu(con->actual_peer_addr.nonce));
1945 con->error_msg = "wrong peer at address";
1950 * did we learn our address?
1952 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1953 int port = addr_port(&con->msgr->inst.addr.in_addr);
1955 memcpy(&con->msgr->inst.addr.in_addr,
1956 &con->peer_addr_for_me.in_addr,
1957 sizeof(con->peer_addr_for_me.in_addr));
1958 addr_set_port(&con->msgr->inst.addr.in_addr, port);
1959 encode_my_addr(con->msgr);
1960 dout("process_banner learned my addr is %s\n",
1961 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
1967 static int process_connect(struct ceph_connection *con)
1969 u64 sup_feat = con->msgr->supported_features;
1970 u64 req_feat = con->msgr->required_features;
1971 u64 server_feat = ceph_sanitize_features(
1972 le64_to_cpu(con->in_reply.features));
1975 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1977 switch (con->in_reply.tag) {
1978 case CEPH_MSGR_TAG_FEATURES:
1979 pr_err("%s%lld %s feature set mismatch,"
1980 " my %llx < server's %llx, missing %llx\n",
1981 ENTITY_NAME(con->peer_name),
1982 ceph_pr_addr(&con->peer_addr.in_addr),
1983 sup_feat, server_feat, server_feat & ~sup_feat);
1984 con->error_msg = "missing required protocol features";
1985 reset_connection(con);
1988 case CEPH_MSGR_TAG_BADPROTOVER:
1989 pr_err("%s%lld %s protocol version mismatch,"
1990 " my %d != server's %d\n",
1991 ENTITY_NAME(con->peer_name),
1992 ceph_pr_addr(&con->peer_addr.in_addr),
1993 le32_to_cpu(con->out_connect.protocol_version),
1994 le32_to_cpu(con->in_reply.protocol_version));
1995 con->error_msg = "protocol version mismatch";
1996 reset_connection(con);
1999 case CEPH_MSGR_TAG_BADAUTHORIZER:
2001 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2003 if (con->auth_retry == 2) {
2004 con->error_msg = "connect authorization failure";
2007 con_out_kvec_reset(con);
2008 ret = prepare_write_connect(con);
2011 prepare_read_connect(con);
2014 case CEPH_MSGR_TAG_RESETSESSION:
2016 * If we connected with a large connect_seq but the peer
2017 * has no record of a session with us (no connection, or
2018 * connect_seq == 0), they will send RESETSESION to indicate
2019 * that they must have reset their session, and may have
2022 dout("process_connect got RESET peer seq %u\n",
2023 le32_to_cpu(con->in_reply.connect_seq));
2024 pr_err("%s%lld %s connection reset\n",
2025 ENTITY_NAME(con->peer_name),
2026 ceph_pr_addr(&con->peer_addr.in_addr));
2027 reset_connection(con);
2028 con_out_kvec_reset(con);
2029 ret = prepare_write_connect(con);
2032 prepare_read_connect(con);
2034 /* Tell ceph about it. */
2035 mutex_unlock(&con->mutex);
2036 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2037 if (con->ops->peer_reset)
2038 con->ops->peer_reset(con);
2039 mutex_lock(&con->mutex);
2040 if (con->state != CON_STATE_NEGOTIATING)
2044 case CEPH_MSGR_TAG_RETRY_SESSION:
2046 * If we sent a smaller connect_seq than the peer has, try
2047 * again with a larger value.
2049 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2050 le32_to_cpu(con->out_connect.connect_seq),
2051 le32_to_cpu(con->in_reply.connect_seq));
2052 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2053 con_out_kvec_reset(con);
2054 ret = prepare_write_connect(con);
2057 prepare_read_connect(con);
2060 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2062 * If we sent a smaller global_seq than the peer has, try
2063 * again with a larger value.
2065 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2066 con->peer_global_seq,
2067 le32_to_cpu(con->in_reply.global_seq));
2068 get_global_seq(con->msgr,
2069 le32_to_cpu(con->in_reply.global_seq));
2070 con_out_kvec_reset(con);
2071 ret = prepare_write_connect(con);
2074 prepare_read_connect(con);
2077 case CEPH_MSGR_TAG_SEQ:
2078 case CEPH_MSGR_TAG_READY:
2079 if (req_feat & ~server_feat) {
2080 pr_err("%s%lld %s protocol feature mismatch,"
2081 " my required %llx > server's %llx, need %llx\n",
2082 ENTITY_NAME(con->peer_name),
2083 ceph_pr_addr(&con->peer_addr.in_addr),
2084 req_feat, server_feat, req_feat & ~server_feat);
2085 con->error_msg = "missing required protocol features";
2086 reset_connection(con);
2090 WARN_ON(con->state != CON_STATE_NEGOTIATING);
2091 con->state = CON_STATE_OPEN;
2092 con->auth_retry = 0; /* we authenticated; clear flag */
2093 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2095 con->peer_features = server_feat;
2096 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2097 con->peer_global_seq,
2098 le32_to_cpu(con->in_reply.connect_seq),
2100 WARN_ON(con->connect_seq !=
2101 le32_to_cpu(con->in_reply.connect_seq));
2103 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2104 con_flag_set(con, CON_FLAG_LOSSYTX);
2106 con->delay = 0; /* reset backoff memory */
2108 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2109 prepare_write_seq(con);
2110 prepare_read_seq(con);
2112 prepare_read_tag(con);
2116 case CEPH_MSGR_TAG_WAIT:
2118 * If there is a connection race (we are opening
2119 * connections to each other), one of us may just have
2120 * to WAIT. This shouldn't happen if we are the
2123 pr_err("process_connect got WAIT as client\n");
2124 con->error_msg = "protocol error, got WAIT as client";
2128 pr_err("connect protocol error, will retry\n");
2129 con->error_msg = "protocol error, garbage tag during connect";
2137 * read (part of) an ack
2139 static int read_partial_ack(struct ceph_connection *con)
2141 int size = sizeof (con->in_temp_ack);
2144 return read_partial(con, end, size, &con->in_temp_ack);
2148 * We can finally discard anything that's been acked.
2150 static void process_ack(struct ceph_connection *con)
2153 u64 ack = le64_to_cpu(con->in_temp_ack);
2156 while (!list_empty(&con->out_sent)) {
2157 m = list_first_entry(&con->out_sent, struct ceph_msg,
2159 seq = le64_to_cpu(m->hdr.seq);
2162 dout("got ack for seq %llu type %d at %p\n", seq,
2163 le16_to_cpu(m->hdr.type), m);
2164 m->ack_stamp = jiffies;
2167 prepare_read_tag(con);
2171 static int read_partial_message_section(struct ceph_connection *con,
2172 struct kvec *section,
2173 unsigned int sec_len, u32 *crc)
2179 while (section->iov_len < sec_len) {
2180 BUG_ON(section->iov_base == NULL);
2181 left = sec_len - section->iov_len;
2182 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2183 section->iov_len, left);
2186 section->iov_len += ret;
2188 if (section->iov_len == sec_len)
2189 *crc = crc32c(0, section->iov_base, section->iov_len);
2194 static int read_partial_msg_data(struct ceph_connection *con)
2196 struct ceph_msg *msg = con->in_msg;
2197 struct ceph_msg_data_cursor *cursor = &msg->cursor;
2198 const bool do_datacrc = !con->msgr->nocrc;
2206 if (list_empty(&msg->data))
2210 crc = con->in_data_crc;
2211 while (cursor->resid) {
2212 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
2214 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2217 con->in_data_crc = crc;
2223 crc = ceph_crc32c_page(crc, page, page_offset, ret);
2224 (void) ceph_msg_data_advance(&msg->cursor, (size_t)ret);
2227 con->in_data_crc = crc;
2229 return 1; /* must return > 0 to indicate success */
2233 * read (part of) a message.
2235 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2237 static int read_partial_message(struct ceph_connection *con)
2239 struct ceph_msg *m = con->in_msg;
2243 unsigned int front_len, middle_len, data_len;
2244 bool do_datacrc = !con->msgr->nocrc;
2248 dout("read_partial_message con %p msg %p\n", con, m);
2251 size = sizeof (con->in_hdr);
2253 ret = read_partial(con, end, size, &con->in_hdr);
2257 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2258 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2259 pr_err("read_partial_message bad hdr "
2260 " crc %u != expected %u\n",
2261 crc, con->in_hdr.crc);
2265 front_len = le32_to_cpu(con->in_hdr.front_len);
2266 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2268 middle_len = le32_to_cpu(con->in_hdr.middle_len);
2269 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2271 data_len = le32_to_cpu(con->in_hdr.data_len);
2272 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2276 seq = le64_to_cpu(con->in_hdr.seq);
2277 if ((s64)seq - (s64)con->in_seq < 1) {
2278 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2279 ENTITY_NAME(con->peer_name),
2280 ceph_pr_addr(&con->peer_addr.in_addr),
2281 seq, con->in_seq + 1);
2282 con->in_base_pos = -front_len - middle_len - data_len -
2284 con->in_tag = CEPH_MSGR_TAG_READY;
2286 } else if ((s64)seq - (s64)con->in_seq > 1) {
2287 pr_err("read_partial_message bad seq %lld expected %lld\n",
2288 seq, con->in_seq + 1);
2289 con->error_msg = "bad message sequence # for incoming message";
2293 /* allocate message? */
2297 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2298 front_len, data_len);
2299 ret = ceph_con_in_msg_alloc(con, &skip);
2303 BUG_ON(!con->in_msg ^ skip);
2304 if (con->in_msg && data_len > con->in_msg->data_length) {
2305 pr_warning("%s skipping long message (%u > %zd)\n",
2306 __func__, data_len, con->in_msg->data_length);
2307 ceph_msg_put(con->in_msg);
2312 /* skip this message */
2313 dout("alloc_msg said skip message\n");
2314 con->in_base_pos = -front_len - middle_len - data_len -
2316 con->in_tag = CEPH_MSGR_TAG_READY;
2321 BUG_ON(!con->in_msg);
2322 BUG_ON(con->in_msg->con != con);
2324 m->front.iov_len = 0; /* haven't read it yet */
2326 m->middle->vec.iov_len = 0;
2328 /* prepare for data payload, if any */
2331 prepare_message_data(con->in_msg, data_len);
2335 ret = read_partial_message_section(con, &m->front, front_len,
2336 &con->in_front_crc);
2342 ret = read_partial_message_section(con, &m->middle->vec,
2344 &con->in_middle_crc);
2351 ret = read_partial_msg_data(con);
2357 size = sizeof (m->footer);
2359 ret = read_partial(con, end, size, &m->footer);
2363 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2364 m, front_len, m->footer.front_crc, middle_len,
2365 m->footer.middle_crc, data_len, m->footer.data_crc);
2368 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2369 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2370 m, con->in_front_crc, m->footer.front_crc);
2373 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2374 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2375 m, con->in_middle_crc, m->footer.middle_crc);
2379 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2380 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2381 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2382 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2386 return 1; /* done! */
2390 * Process message. This happens in the worker thread. The callback should
2391 * be careful not to do anything that waits on other incoming messages or it
2394 static void process_message(struct ceph_connection *con)
2396 struct ceph_msg *msg;
2398 BUG_ON(con->in_msg->con != con);
2399 con->in_msg->con = NULL;
2404 /* if first message, set peer_name */
2405 if (con->peer_name.type == 0)
2406 con->peer_name = msg->hdr.src;
2409 mutex_unlock(&con->mutex);
2411 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2412 msg, le64_to_cpu(msg->hdr.seq),
2413 ENTITY_NAME(msg->hdr.src),
2414 le16_to_cpu(msg->hdr.type),
2415 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2416 le32_to_cpu(msg->hdr.front_len),
2417 le32_to_cpu(msg->hdr.data_len),
2418 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2419 con->ops->dispatch(con, msg);
2421 mutex_lock(&con->mutex);
2426 * Write something to the socket. Called in a worker thread when the
2427 * socket appears to be writeable and we have something ready to send.
2429 static int try_write(struct ceph_connection *con)
2433 dout("try_write start %p state %lu\n", con, con->state);
2436 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2438 /* open the socket first? */
2439 if (con->state == CON_STATE_PREOPEN) {
2441 con->state = CON_STATE_CONNECTING;
2443 con_out_kvec_reset(con);
2444 prepare_write_banner(con);
2445 prepare_read_banner(con);
2447 BUG_ON(con->in_msg);
2448 con->in_tag = CEPH_MSGR_TAG_READY;
2449 dout("try_write initiating connect on %p new state %lu\n",
2451 ret = ceph_tcp_connect(con);
2453 con->error_msg = "connect error";
2459 /* kvec data queued? */
2460 if (con->out_skip) {
2461 ret = write_partial_skip(con);
2465 if (con->out_kvec_left) {
2466 ret = write_partial_kvec(con);
2473 if (con->out_msg_done) {
2474 ceph_msg_put(con->out_msg);
2475 con->out_msg = NULL; /* we're done with this one */
2479 ret = write_partial_message_data(con);
2481 goto more_kvec; /* we need to send the footer, too! */
2485 dout("try_write write_partial_message_data err %d\n",
2492 if (con->state == CON_STATE_OPEN) {
2493 /* is anything else pending? */
2494 if (!list_empty(&con->out_queue)) {
2495 prepare_write_message(con);
2498 if (con->in_seq > con->in_seq_acked) {
2499 prepare_write_ack(con);
2502 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2503 prepare_write_keepalive(con);
2508 /* Nothing to do! */
2509 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2510 dout("try_write nothing else to write.\n");
2513 dout("try_write done on %p ret %d\n", con, ret);
2520 * Read what we can from the socket.
2522 static int try_read(struct ceph_connection *con)
2527 dout("try_read start on %p state %lu\n", con, con->state);
2528 if (con->state != CON_STATE_CONNECTING &&
2529 con->state != CON_STATE_NEGOTIATING &&
2530 con->state != CON_STATE_OPEN)
2535 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2538 if (con->state == CON_STATE_CONNECTING) {
2539 dout("try_read connecting\n");
2540 ret = read_partial_banner(con);
2543 ret = process_banner(con);
2547 con->state = CON_STATE_NEGOTIATING;
2550 * Received banner is good, exchange connection info.
2551 * Do not reset out_kvec, as sending our banner raced
2552 * with receiving peer banner after connect completed.
2554 ret = prepare_write_connect(con);
2557 prepare_read_connect(con);
2559 /* Send connection info before awaiting response */
2563 if (con->state == CON_STATE_NEGOTIATING) {
2564 dout("try_read negotiating\n");
2565 ret = read_partial_connect(con);
2568 ret = process_connect(con);
2574 WARN_ON(con->state != CON_STATE_OPEN);
2576 if (con->in_base_pos < 0) {
2578 * skipping + discarding content.
2580 * FIXME: there must be a better way to do this!
2582 static char buf[SKIP_BUF_SIZE];
2583 int skip = min((int) sizeof (buf), -con->in_base_pos);
2585 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2586 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2589 con->in_base_pos += ret;
2590 if (con->in_base_pos)
2593 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2597 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2600 dout("try_read got tag %d\n", (int)con->in_tag);
2601 switch (con->in_tag) {
2602 case CEPH_MSGR_TAG_MSG:
2603 prepare_read_message(con);
2605 case CEPH_MSGR_TAG_ACK:
2606 prepare_read_ack(con);
2608 case CEPH_MSGR_TAG_CLOSE:
2609 con_close_socket(con);
2610 con->state = CON_STATE_CLOSED;
2616 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2617 ret = read_partial_message(con);
2621 con->error_msg = "bad crc";
2625 con->error_msg = "io error";
2630 if (con->in_tag == CEPH_MSGR_TAG_READY)
2632 process_message(con);
2633 if (con->state == CON_STATE_OPEN)
2634 prepare_read_tag(con);
2637 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2638 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2640 * the final handshake seq exchange is semantically
2641 * equivalent to an ACK
2643 ret = read_partial_ack(con);
2651 dout("try_read done on %p ret %d\n", con, ret);
2655 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2656 con->error_msg = "protocol error, garbage tag";
2663 * Atomically queue work on a connection after the specified delay.
2664 * Bump @con reference to avoid races with connection teardown.
2665 * Returns 0 if work was queued, or an error code otherwise.
2667 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2669 if (!con->ops->get(con)) {
2670 dout("%s %p ref count 0\n", __func__, con);
2674 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2675 dout("%s %p - already queued\n", __func__, con);
2680 dout("%s %p %lu\n", __func__, con, delay);
2684 static void queue_con(struct ceph_connection *con)
2686 (void) queue_con_delay(con, 0);
2689 static void cancel_con(struct ceph_connection *con)
2691 if (cancel_delayed_work(&con->work)) {
2692 dout("%s %p\n", __func__, con);
2697 static bool con_sock_closed(struct ceph_connection *con)
2699 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2703 case CON_STATE_ ## x: \
2704 con->error_msg = "socket closed (con state " #x ")"; \
2707 switch (con->state) {
2715 pr_warning("%s con %p unrecognized state %lu\n",
2716 __func__, con, con->state);
2717 con->error_msg = "unrecognized con state";
2726 static bool con_backoff(struct ceph_connection *con)
2730 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2733 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2735 dout("%s: con %p FAILED to back off %lu\n", __func__,
2737 BUG_ON(ret == -ENOENT);
2738 con_flag_set(con, CON_FLAG_BACKOFF);
2744 /* Finish fault handling; con->mutex must *not* be held here */
2746 static void con_fault_finish(struct ceph_connection *con)
2749 * in case we faulted due to authentication, invalidate our
2750 * current tickets so that we can get new ones.
2752 if (con->auth_retry && con->ops->invalidate_authorizer) {
2753 dout("calling invalidate_authorizer()\n");
2754 con->ops->invalidate_authorizer(con);
2757 if (con->ops->fault)
2758 con->ops->fault(con);
2762 * Do some work on a connection. Drop a connection ref when we're done.
2764 static void con_work(struct work_struct *work)
2766 struct ceph_connection *con = container_of(work, struct ceph_connection,
2770 mutex_lock(&con->mutex);
2774 if ((fault = con_sock_closed(con))) {
2775 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2778 if (con_backoff(con)) {
2779 dout("%s: con %p BACKOFF\n", __func__, con);
2782 if (con->state == CON_STATE_STANDBY) {
2783 dout("%s: con %p STANDBY\n", __func__, con);
2786 if (con->state == CON_STATE_CLOSED) {
2787 dout("%s: con %p CLOSED\n", __func__, con);
2791 if (con->state == CON_STATE_PREOPEN) {
2792 dout("%s: con %p PREOPEN\n", __func__, con);
2796 ret = try_read(con);
2800 con->error_msg = "socket error on read";
2805 ret = try_write(con);
2809 con->error_msg = "socket error on write";
2813 break; /* If we make it to here, we're done */
2817 mutex_unlock(&con->mutex);
2820 con_fault_finish(con);
2826 * Generic error/fault handler. A retry mechanism is used with
2827 * exponential backoff
2829 static void con_fault(struct ceph_connection *con)
2831 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2832 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2833 dout("fault %p state %lu to peer %s\n",
2834 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2836 WARN_ON(con->state != CON_STATE_CONNECTING &&
2837 con->state != CON_STATE_NEGOTIATING &&
2838 con->state != CON_STATE_OPEN);
2840 con_close_socket(con);
2842 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2843 dout("fault on LOSSYTX channel, marking CLOSED\n");
2844 con->state = CON_STATE_CLOSED;
2849 BUG_ON(con->in_msg->con != con);
2850 con->in_msg->con = NULL;
2851 ceph_msg_put(con->in_msg);
2856 /* Requeue anything that hasn't been acked */
2857 list_splice_init(&con->out_sent, &con->out_queue);
2859 /* If there are no messages queued or keepalive pending, place
2860 * the connection in a STANDBY state */
2861 if (list_empty(&con->out_queue) &&
2862 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
2863 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2864 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2865 con->state = CON_STATE_STANDBY;
2867 /* retry after a delay. */
2868 con->state = CON_STATE_PREOPEN;
2869 if (con->delay == 0)
2870 con->delay = BASE_DELAY_INTERVAL;
2871 else if (con->delay < MAX_DELAY_INTERVAL)
2873 con_flag_set(con, CON_FLAG_BACKOFF);
2881 * initialize a new messenger instance
2883 void ceph_messenger_init(struct ceph_messenger *msgr,
2884 struct ceph_entity_addr *myaddr,
2885 u64 supported_features,
2886 u64 required_features,
2889 msgr->supported_features = supported_features;
2890 msgr->required_features = required_features;
2892 spin_lock_init(&msgr->global_seq_lock);
2895 msgr->inst.addr = *myaddr;
2897 /* select a random nonce */
2898 msgr->inst.addr.type = 0;
2899 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2900 encode_my_addr(msgr);
2901 msgr->nocrc = nocrc;
2903 atomic_set(&msgr->stopping, 0);
2905 dout("%s %p\n", __func__, msgr);
2907 EXPORT_SYMBOL(ceph_messenger_init);
2909 static void clear_standby(struct ceph_connection *con)
2911 /* come back from STANDBY? */
2912 if (con->state == CON_STATE_STANDBY) {
2913 dout("clear_standby %p and ++connect_seq\n", con);
2914 con->state = CON_STATE_PREOPEN;
2916 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2917 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
2922 * Queue up an outgoing message on the given connection.
2924 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2927 msg->hdr.src = con->msgr->inst.name;
2928 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2929 msg->needs_out_seq = true;
2931 mutex_lock(&con->mutex);
2933 if (con->state == CON_STATE_CLOSED) {
2934 dout("con_send %p closed, dropping %p\n", con, msg);
2936 mutex_unlock(&con->mutex);
2940 BUG_ON(msg->con != NULL);
2941 msg->con = con->ops->get(con);
2942 BUG_ON(msg->con == NULL);
2944 BUG_ON(!list_empty(&msg->list_head));
2945 list_add_tail(&msg->list_head, &con->out_queue);
2946 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2947 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2948 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2949 le32_to_cpu(msg->hdr.front_len),
2950 le32_to_cpu(msg->hdr.middle_len),
2951 le32_to_cpu(msg->hdr.data_len));
2954 mutex_unlock(&con->mutex);
2956 /* if there wasn't anything waiting to send before, queue
2958 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
2961 EXPORT_SYMBOL(ceph_con_send);
2964 * Revoke a message that was previously queued for send
2966 void ceph_msg_revoke(struct ceph_msg *msg)
2968 struct ceph_connection *con = msg->con;
2971 return; /* Message not in our possession */
2973 mutex_lock(&con->mutex);
2974 if (!list_empty(&msg->list_head)) {
2975 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
2976 list_del_init(&msg->list_head);
2977 BUG_ON(msg->con == NULL);
2978 msg->con->ops->put(msg->con);
2984 if (con->out_msg == msg) {
2985 dout("%s %p msg %p - was sending\n", __func__, con, msg);
2986 con->out_msg = NULL;
2987 if (con->out_kvec_is_msg) {
2988 con->out_skip = con->out_kvec_bytes;
2989 con->out_kvec_is_msg = false;
2995 mutex_unlock(&con->mutex);
2999 * Revoke a message that we may be reading data into
3001 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
3003 struct ceph_connection *con;
3005 BUG_ON(msg == NULL);
3007 dout("%s msg %p null con\n", __func__, msg);
3009 return; /* Message not in our possession */
3013 mutex_lock(&con->mutex);
3014 if (con->in_msg == msg) {
3015 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3016 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3017 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
3019 /* skip rest of message */
3020 dout("%s %p msg %p revoked\n", __func__, con, msg);
3021 con->in_base_pos = con->in_base_pos -
3022 sizeof(struct ceph_msg_header) -
3026 sizeof(struct ceph_msg_footer);
3027 ceph_msg_put(con->in_msg);
3029 con->in_tag = CEPH_MSGR_TAG_READY;
3032 dout("%s %p in_msg %p msg %p no-op\n",
3033 __func__, con, con->in_msg, msg);
3035 mutex_unlock(&con->mutex);
3039 * Queue a keepalive byte to ensure the tcp connection is alive.
3041 void ceph_con_keepalive(struct ceph_connection *con)
3043 dout("con_keepalive %p\n", con);
3044 mutex_lock(&con->mutex);
3046 mutex_unlock(&con->mutex);
3047 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3048 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3051 EXPORT_SYMBOL(ceph_con_keepalive);
3053 static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
3055 struct ceph_msg_data *data;
3057 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3060 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
3063 INIT_LIST_HEAD(&data->links);
3068 static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3073 WARN_ON(!list_empty(&data->links));
3074 if (data->type == CEPH_MSG_DATA_PAGELIST) {
3075 ceph_pagelist_release(data->pagelist);
3076 kfree(data->pagelist);
3078 kmem_cache_free(ceph_msg_data_cache, data);
3081 void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
3082 size_t length, size_t alignment)
3084 struct ceph_msg_data *data;
3089 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3091 data->pages = pages;
3092 data->length = length;
3093 data->alignment = alignment & ~PAGE_MASK;
3095 list_add_tail(&data->links, &msg->data);
3096 msg->data_length += length;
3098 EXPORT_SYMBOL(ceph_msg_data_add_pages);
3100 void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
3101 struct ceph_pagelist *pagelist)
3103 struct ceph_msg_data *data;
3106 BUG_ON(!pagelist->length);
3108 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3110 data->pagelist = pagelist;
3112 list_add_tail(&data->links, &msg->data);
3113 msg->data_length += pagelist->length;
3115 EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
3118 void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
3121 struct ceph_msg_data *data;
3125 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3128 data->bio_length = length;
3130 list_add_tail(&data->links, &msg->data);
3131 msg->data_length += length;
3133 EXPORT_SYMBOL(ceph_msg_data_add_bio);
3134 #endif /* CONFIG_BLOCK */
3137 * construct a new message with given type, size
3138 * the new msg has a ref count of 1.
3140 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3145 m = kmem_cache_zalloc(ceph_msg_cache, flags);
3149 m->hdr.type = cpu_to_le16(type);
3150 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3151 m->hdr.front_len = cpu_to_le32(front_len);
3153 INIT_LIST_HEAD(&m->list_head);
3154 kref_init(&m->kref);
3155 INIT_LIST_HEAD(&m->data);
3159 m->front.iov_base = ceph_kvmalloc(front_len, flags);
3160 if (m->front.iov_base == NULL) {
3161 dout("ceph_msg_new can't allocate %d bytes\n",
3166 m->front.iov_base = NULL;
3168 m->front_alloc_len = m->front.iov_len = front_len;
3170 dout("ceph_msg_new %p front %d\n", m, front_len);
3177 pr_err("msg_new can't create type %d front %d\n", type,
3181 dout("msg_new can't create type %d front %d\n", type,
3186 EXPORT_SYMBOL(ceph_msg_new);
3189 * Allocate "middle" portion of a message, if it is needed and wasn't
3190 * allocated by alloc_msg. This allows us to read a small fixed-size
3191 * per-type header in the front and then gracefully fail (i.e.,
3192 * propagate the error to the caller based on info in the front) when
3193 * the middle is too large.
3195 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3197 int type = le16_to_cpu(msg->hdr.type);
3198 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3200 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3201 ceph_msg_type_name(type), middle_len);
3202 BUG_ON(!middle_len);
3203 BUG_ON(msg->middle);
3205 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3212 * Allocate a message for receiving an incoming message on a
3213 * connection, and save the result in con->in_msg. Uses the
3214 * connection's private alloc_msg op if available.
3216 * Returns 0 on success, or a negative error code.
3218 * On success, if we set *skip = 1:
3219 * - the next message should be skipped and ignored.
3220 * - con->in_msg == NULL
3221 * or if we set *skip = 0:
3222 * - con->in_msg is non-null.
3223 * On error (ENOMEM, EAGAIN, ...),
3224 * - con->in_msg == NULL
3226 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3228 struct ceph_msg_header *hdr = &con->in_hdr;
3229 int middle_len = le32_to_cpu(hdr->middle_len);
3230 struct ceph_msg *msg;
3233 BUG_ON(con->in_msg != NULL);
3234 BUG_ON(!con->ops->alloc_msg);
3236 mutex_unlock(&con->mutex);
3237 msg = con->ops->alloc_msg(con, hdr, skip);
3238 mutex_lock(&con->mutex);
3239 if (con->state != CON_STATE_OPEN) {
3247 con->in_msg->con = con->ops->get(con);
3248 BUG_ON(con->in_msg->con == NULL);
3251 * Null message pointer means either we should skip
3252 * this message or we couldn't allocate memory. The
3253 * former is not an error.
3257 con->error_msg = "error allocating memory for incoming message";
3261 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3263 if (middle_len && !con->in_msg->middle) {
3264 ret = ceph_alloc_middle(con, con->in_msg);
3266 ceph_msg_put(con->in_msg);
3276 * Free a generically kmalloc'd message.
3278 static void ceph_msg_free(struct ceph_msg *m)
3280 dout("%s %p\n", __func__, m);
3281 ceph_kvfree(m->front.iov_base);
3282 kmem_cache_free(ceph_msg_cache, m);
3285 static void ceph_msg_release(struct kref *kref)
3287 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3289 struct list_head *links;
3290 struct list_head *next;
3292 dout("%s %p\n", __func__, m);
3293 WARN_ON(!list_empty(&m->list_head));
3295 /* drop middle, data, if any */
3297 ceph_buffer_put(m->middle);
3301 list_splice_init(&m->data, &data);
3302 list_for_each_safe(links, next, &data) {
3303 struct ceph_msg_data *data;
3305 data = list_entry(links, struct ceph_msg_data, links);
3306 list_del_init(links);
3307 ceph_msg_data_destroy(data);
3312 ceph_msgpool_put(m->pool, m);
3317 struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3319 dout("%s %p (was %d)\n", __func__, msg,
3320 atomic_read(&msg->kref.refcount));
3321 kref_get(&msg->kref);
3324 EXPORT_SYMBOL(ceph_msg_get);
3326 void ceph_msg_put(struct ceph_msg *msg)
3328 dout("%s %p (was %d)\n", __func__, msg,
3329 atomic_read(&msg->kref.refcount));
3330 kref_put(&msg->kref, ceph_msg_release);
3332 EXPORT_SYMBOL(ceph_msg_put);
3334 void ceph_msg_dump(struct ceph_msg *msg)
3336 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3337 msg->front_alloc_len, msg->data_length);
3338 print_hex_dump(KERN_DEBUG, "header: ",
3339 DUMP_PREFIX_OFFSET, 16, 1,
3340 &msg->hdr, sizeof(msg->hdr), true);
3341 print_hex_dump(KERN_DEBUG, " front: ",
3342 DUMP_PREFIX_OFFSET, 16, 1,
3343 msg->front.iov_base, msg->front.iov_len, true);
3345 print_hex_dump(KERN_DEBUG, "middle: ",
3346 DUMP_PREFIX_OFFSET, 16, 1,
3347 msg->middle->vec.iov_base,
3348 msg->middle->vec.iov_len, true);
3349 print_hex_dump(KERN_DEBUG, "footer: ",
3350 DUMP_PREFIX_OFFSET, 16, 1,
3351 &msg->footer, sizeof(msg->footer), true);
3353 EXPORT_SYMBOL(ceph_msg_dump);