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 con_work(struct work_struct *);
178 static void con_fault(struct ceph_connection *con);
181 * Nicely render a sockaddr as a string. An array of formatted
182 * strings is used, to approximate reentrancy.
184 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
185 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
186 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
187 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
189 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
190 static atomic_t addr_str_seq = ATOMIC_INIT(0);
192 static struct page *zero_page; /* used in certain error cases */
194 const char *ceph_pr_addr(const struct sockaddr_storage *ss)
198 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
199 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
201 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
204 switch (ss->ss_family) {
206 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
207 ntohs(in4->sin_port));
211 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
212 ntohs(in6->sin6_port));
216 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
222 EXPORT_SYMBOL(ceph_pr_addr);
224 static void encode_my_addr(struct ceph_messenger *msgr)
226 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
227 ceph_encode_addr(&msgr->my_enc_addr);
231 * work queue for all reading and writing to/from the socket.
233 static struct workqueue_struct *ceph_msgr_wq;
235 static int ceph_msgr_slab_init(void)
237 BUG_ON(ceph_msg_cache);
238 ceph_msg_cache = kmem_cache_create("ceph_msg",
239 sizeof (struct ceph_msg),
240 __alignof__(struct ceph_msg), 0, NULL);
245 BUG_ON(ceph_msg_data_cache);
246 ceph_msg_data_cache = kmem_cache_create("ceph_msg_data",
247 sizeof (struct ceph_msg_data),
248 __alignof__(struct ceph_msg_data),
250 if (ceph_msg_data_cache)
253 kmem_cache_destroy(ceph_msg_cache);
254 ceph_msg_cache = NULL;
259 static void ceph_msgr_slab_exit(void)
261 BUG_ON(!ceph_msg_data_cache);
262 kmem_cache_destroy(ceph_msg_data_cache);
263 ceph_msg_data_cache = NULL;
265 BUG_ON(!ceph_msg_cache);
266 kmem_cache_destroy(ceph_msg_cache);
267 ceph_msg_cache = NULL;
270 static void _ceph_msgr_exit(void)
273 destroy_workqueue(ceph_msgr_wq);
277 ceph_msgr_slab_exit();
279 BUG_ON(zero_page == NULL);
281 page_cache_release(zero_page);
285 int ceph_msgr_init(void)
287 BUG_ON(zero_page != NULL);
288 zero_page = ZERO_PAGE(0);
289 page_cache_get(zero_page);
291 if (ceph_msgr_slab_init())
294 ceph_msgr_wq = alloc_workqueue("ceph-msgr", 0, 0);
298 pr_err("msgr_init failed to create workqueue\n");
303 EXPORT_SYMBOL(ceph_msgr_init);
305 void ceph_msgr_exit(void)
307 BUG_ON(ceph_msgr_wq == NULL);
311 EXPORT_SYMBOL(ceph_msgr_exit);
313 void ceph_msgr_flush(void)
315 flush_workqueue(ceph_msgr_wq);
317 EXPORT_SYMBOL(ceph_msgr_flush);
319 /* Connection socket state transition functions */
321 static void con_sock_state_init(struct ceph_connection *con)
325 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
326 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
327 printk("%s: unexpected old state %d\n", __func__, old_state);
328 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
329 CON_SOCK_STATE_CLOSED);
332 static void con_sock_state_connecting(struct ceph_connection *con)
336 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
337 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
338 printk("%s: unexpected old state %d\n", __func__, old_state);
339 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
340 CON_SOCK_STATE_CONNECTING);
343 static void con_sock_state_connected(struct ceph_connection *con)
347 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
348 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
349 printk("%s: unexpected old state %d\n", __func__, old_state);
350 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
351 CON_SOCK_STATE_CONNECTED);
354 static void con_sock_state_closing(struct ceph_connection *con)
358 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
359 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
360 old_state != CON_SOCK_STATE_CONNECTED &&
361 old_state != CON_SOCK_STATE_CLOSING))
362 printk("%s: unexpected old state %d\n", __func__, old_state);
363 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
364 CON_SOCK_STATE_CLOSING);
367 static void con_sock_state_closed(struct ceph_connection *con)
371 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
372 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
373 old_state != CON_SOCK_STATE_CLOSING &&
374 old_state != CON_SOCK_STATE_CONNECTING &&
375 old_state != CON_SOCK_STATE_CLOSED))
376 printk("%s: unexpected old state %d\n", __func__, old_state);
377 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
378 CON_SOCK_STATE_CLOSED);
382 * socket callback functions
385 /* data available on socket, or listen socket received a connect */
386 static void ceph_sock_data_ready(struct sock *sk, int count_unused)
388 struct ceph_connection *con = sk->sk_user_data;
389 if (atomic_read(&con->msgr->stopping)) {
393 if (sk->sk_state != TCP_CLOSE_WAIT) {
394 dout("%s on %p state = %lu, queueing work\n", __func__,
400 /* socket has buffer space for writing */
401 static void ceph_sock_write_space(struct sock *sk)
403 struct ceph_connection *con = sk->sk_user_data;
405 /* only queue to workqueue if there is data we want to write,
406 * and there is sufficient space in the socket buffer to accept
407 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
408 * doesn't get called again until try_write() fills the socket
409 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
410 * and net/core/stream.c:sk_stream_write_space().
412 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
413 if (sk_stream_is_writeable(sk)) {
414 dout("%s %p queueing write work\n", __func__, con);
415 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
419 dout("%s %p nothing to write\n", __func__, con);
423 /* socket's state has changed */
424 static void ceph_sock_state_change(struct sock *sk)
426 struct ceph_connection *con = sk->sk_user_data;
428 dout("%s %p state = %lu sk_state = %u\n", __func__,
429 con, con->state, sk->sk_state);
431 switch (sk->sk_state) {
433 dout("%s TCP_CLOSE\n", __func__);
435 dout("%s TCP_CLOSE_WAIT\n", __func__);
436 con_sock_state_closing(con);
437 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
440 case TCP_ESTABLISHED:
441 dout("%s TCP_ESTABLISHED\n", __func__);
442 con_sock_state_connected(con);
445 default: /* Everything else is uninteresting */
451 * set up socket callbacks
453 static void set_sock_callbacks(struct socket *sock,
454 struct ceph_connection *con)
456 struct sock *sk = sock->sk;
457 sk->sk_user_data = con;
458 sk->sk_data_ready = ceph_sock_data_ready;
459 sk->sk_write_space = ceph_sock_write_space;
460 sk->sk_state_change = ceph_sock_state_change;
469 * initiate connection to a remote socket.
471 static int ceph_tcp_connect(struct ceph_connection *con)
473 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
478 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
482 sock->sk->sk_allocation = GFP_NOFS;
484 #ifdef CONFIG_LOCKDEP
485 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
488 set_sock_callbacks(sock, con);
490 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
492 con_sock_state_connecting(con);
493 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
495 if (ret == -EINPROGRESS) {
496 dout("connect %s EINPROGRESS sk_state = %u\n",
497 ceph_pr_addr(&con->peer_addr.in_addr),
499 } else if (ret < 0) {
500 pr_err("connect %s error %d\n",
501 ceph_pr_addr(&con->peer_addr.in_addr), ret);
503 con->error_msg = "connect error";
511 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
513 struct kvec iov = {buf, len};
514 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
517 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
523 static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
524 int page_offset, size_t length)
529 BUG_ON(page_offset + length > PAGE_SIZE);
533 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
540 * write something. @more is true if caller will be sending more data
543 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
544 size_t kvlen, size_t len, int more)
546 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
550 msg.msg_flags |= MSG_MORE;
552 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
554 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
560 static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
561 int offset, size_t size, bool more)
563 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
566 ret = kernel_sendpage(sock, page, offset, size, flags);
573 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
574 int offset, size_t size, bool more)
579 /* sendpage cannot properly handle pages with page_count == 0,
580 * we need to fallback to sendmsg if that's the case */
581 if (page_count(page) >= 1)
582 return __ceph_tcp_sendpage(sock, page, offset, size, more);
584 iov.iov_base = kmap(page) + offset;
586 ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more);
593 * Shutdown/close the socket for the given connection.
595 static int con_close_socket(struct ceph_connection *con)
599 dout("con_close_socket on %p sock %p\n", con, con->sock);
601 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
602 sock_release(con->sock);
607 * Forcibly clear the SOCK_CLOSED flag. It gets set
608 * independent of the connection mutex, and we could have
609 * received a socket close event before we had the chance to
610 * shut the socket down.
612 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
614 con_sock_state_closed(con);
619 * Reset a connection. Discard all incoming and outgoing messages
620 * and clear *_seq state.
622 static void ceph_msg_remove(struct ceph_msg *msg)
624 list_del_init(&msg->list_head);
625 BUG_ON(msg->con == NULL);
626 msg->con->ops->put(msg->con);
631 static void ceph_msg_remove_list(struct list_head *head)
633 while (!list_empty(head)) {
634 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
636 ceph_msg_remove(msg);
640 static void reset_connection(struct ceph_connection *con)
642 /* reset connection, out_queue, msg_ and connect_seq */
643 /* discard existing out_queue and msg_seq */
644 dout("reset_connection %p\n", con);
645 ceph_msg_remove_list(&con->out_queue);
646 ceph_msg_remove_list(&con->out_sent);
649 BUG_ON(con->in_msg->con != con);
650 con->in_msg->con = NULL;
651 ceph_msg_put(con->in_msg);
656 con->connect_seq = 0;
659 ceph_msg_put(con->out_msg);
663 con->in_seq_acked = 0;
667 * mark a peer down. drop any open connections.
669 void ceph_con_close(struct ceph_connection *con)
671 mutex_lock(&con->mutex);
672 dout("con_close %p peer %s\n", con,
673 ceph_pr_addr(&con->peer_addr.in_addr));
674 con->state = CON_STATE_CLOSED;
676 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
677 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
678 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
679 con_flag_clear(con, CON_FLAG_BACKOFF);
681 reset_connection(con);
682 con->peer_global_seq = 0;
683 cancel_delayed_work(&con->work);
684 con_close_socket(con);
685 mutex_unlock(&con->mutex);
687 EXPORT_SYMBOL(ceph_con_close);
690 * Reopen a closed connection, with a new peer address.
692 void ceph_con_open(struct ceph_connection *con,
693 __u8 entity_type, __u64 entity_num,
694 struct ceph_entity_addr *addr)
696 mutex_lock(&con->mutex);
697 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
699 WARN_ON(con->state != CON_STATE_CLOSED);
700 con->state = CON_STATE_PREOPEN;
702 con->peer_name.type = (__u8) entity_type;
703 con->peer_name.num = cpu_to_le64(entity_num);
705 memcpy(&con->peer_addr, addr, sizeof(*addr));
706 con->delay = 0; /* reset backoff memory */
707 mutex_unlock(&con->mutex);
710 EXPORT_SYMBOL(ceph_con_open);
713 * return true if this connection ever successfully opened
715 bool ceph_con_opened(struct ceph_connection *con)
717 return con->connect_seq > 0;
721 * initialize a new connection.
723 void ceph_con_init(struct ceph_connection *con, void *private,
724 const struct ceph_connection_operations *ops,
725 struct ceph_messenger *msgr)
727 dout("con_init %p\n", con);
728 memset(con, 0, sizeof(*con));
729 con->private = private;
733 con_sock_state_init(con);
735 mutex_init(&con->mutex);
736 INIT_LIST_HEAD(&con->out_queue);
737 INIT_LIST_HEAD(&con->out_sent);
738 INIT_DELAYED_WORK(&con->work, con_work);
740 con->state = CON_STATE_CLOSED;
742 EXPORT_SYMBOL(ceph_con_init);
746 * We maintain a global counter to order connection attempts. Get
747 * a unique seq greater than @gt.
749 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
753 spin_lock(&msgr->global_seq_lock);
754 if (msgr->global_seq < gt)
755 msgr->global_seq = gt;
756 ret = ++msgr->global_seq;
757 spin_unlock(&msgr->global_seq_lock);
761 static void con_out_kvec_reset(struct ceph_connection *con)
763 con->out_kvec_left = 0;
764 con->out_kvec_bytes = 0;
765 con->out_kvec_cur = &con->out_kvec[0];
768 static void con_out_kvec_add(struct ceph_connection *con,
769 size_t size, void *data)
773 index = con->out_kvec_left;
774 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
776 con->out_kvec[index].iov_len = size;
777 con->out_kvec[index].iov_base = data;
778 con->out_kvec_left++;
779 con->out_kvec_bytes += size;
785 * For a bio data item, a piece is whatever remains of the next
786 * entry in the current bio iovec, or the first entry in the next
789 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
792 struct ceph_msg_data *data = cursor->data;
795 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
800 cursor->resid = min(length, data->bio_length);
802 cursor->bvec_iter = bio->bi_iter;
804 cursor->resid <= bio_iter_len(bio, cursor->bvec_iter);
807 static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
811 struct ceph_msg_data *data = cursor->data;
813 struct bio_vec bio_vec;
815 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
820 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
822 *page_offset = (size_t) bio_vec.bv_offset;
823 BUG_ON(*page_offset >= PAGE_SIZE);
824 if (cursor->last_piece) /* pagelist offset is always 0 */
825 *length = cursor->resid;
827 *length = (size_t) bio_vec.bv_len;
828 BUG_ON(*length > cursor->resid);
829 BUG_ON(*page_offset + *length > PAGE_SIZE);
831 return bio_vec.bv_page;
834 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
838 struct bio_vec bio_vec;
840 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
845 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
847 /* Advance the cursor offset */
849 BUG_ON(cursor->resid < bytes);
850 cursor->resid -= bytes;
852 bio_advance_iter(bio, &cursor->bvec_iter, bytes);
854 if (bytes < bio_vec.bv_len)
855 return false; /* more bytes to process in this segment */
857 /* Move on to the next segment, and possibly the next bio */
859 if (!cursor->bvec_iter.bi_size) {
863 cursor->bvec_iter = bio->bi_iter;
865 memset(&cursor->bvec_iter, 0,
866 sizeof(cursor->bvec_iter));
869 if (!cursor->last_piece) {
870 BUG_ON(!cursor->resid);
872 /* A short read is OK, so use <= rather than == */
873 if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter))
874 cursor->last_piece = true;
879 #endif /* CONFIG_BLOCK */
882 * For a page array, a piece comes from the first page in the array
883 * that has not already been fully consumed.
885 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
888 struct ceph_msg_data *data = cursor->data;
891 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
893 BUG_ON(!data->pages);
894 BUG_ON(!data->length);
896 cursor->resid = min(length, data->length);
897 page_count = calc_pages_for(data->alignment, (u64)data->length);
898 cursor->page_offset = data->alignment & ~PAGE_MASK;
899 cursor->page_index = 0;
900 BUG_ON(page_count > (int)USHRT_MAX);
901 cursor->page_count = (unsigned short)page_count;
902 BUG_ON(length > SIZE_MAX - cursor->page_offset);
903 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
907 ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
908 size_t *page_offset, size_t *length)
910 struct ceph_msg_data *data = cursor->data;
912 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
914 BUG_ON(cursor->page_index >= cursor->page_count);
915 BUG_ON(cursor->page_offset >= PAGE_SIZE);
917 *page_offset = cursor->page_offset;
918 if (cursor->last_piece)
919 *length = cursor->resid;
921 *length = PAGE_SIZE - *page_offset;
923 return data->pages[cursor->page_index];
926 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
929 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
931 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
933 /* Advance the cursor page offset */
935 cursor->resid -= bytes;
936 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
937 if (!bytes || cursor->page_offset)
938 return false; /* more bytes to process in the current page */
940 /* Move on to the next page; offset is already at 0 */
942 BUG_ON(cursor->page_index >= cursor->page_count);
943 cursor->page_index++;
944 cursor->last_piece = cursor->resid <= PAGE_SIZE;
950 * For a pagelist, a piece is whatever remains to be consumed in the
951 * first page in the list, or the front of the next page.
954 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
957 struct ceph_msg_data *data = cursor->data;
958 struct ceph_pagelist *pagelist;
961 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
963 pagelist = data->pagelist;
967 return; /* pagelist can be assigned but empty */
969 BUG_ON(list_empty(&pagelist->head));
970 page = list_first_entry(&pagelist->head, struct page, lru);
972 cursor->resid = min(length, pagelist->length);
975 cursor->last_piece = cursor->resid <= PAGE_SIZE;
979 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
980 size_t *page_offset, size_t *length)
982 struct ceph_msg_data *data = cursor->data;
983 struct ceph_pagelist *pagelist;
985 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
987 pagelist = data->pagelist;
990 BUG_ON(!cursor->page);
991 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
993 /* offset of first page in pagelist is always 0 */
994 *page_offset = cursor->offset & ~PAGE_MASK;
995 if (cursor->last_piece)
996 *length = cursor->resid;
998 *length = PAGE_SIZE - *page_offset;
1000 return cursor->page;
1003 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
1006 struct ceph_msg_data *data = cursor->data;
1007 struct ceph_pagelist *pagelist;
1009 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1011 pagelist = data->pagelist;
1014 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1015 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1017 /* Advance the cursor offset */
1019 cursor->resid -= bytes;
1020 cursor->offset += bytes;
1021 /* offset of first page in pagelist is always 0 */
1022 if (!bytes || cursor->offset & ~PAGE_MASK)
1023 return false; /* more bytes to process in the current page */
1025 /* Move on to the next page */
1027 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1028 cursor->page = list_entry_next(cursor->page, lru);
1029 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1035 * Message data is handled (sent or received) in pieces, where each
1036 * piece resides on a single page. The network layer might not
1037 * consume an entire piece at once. A data item's cursor keeps
1038 * track of which piece is next to process and how much remains to
1039 * be processed in that piece. It also tracks whether the current
1040 * piece is the last one in the data item.
1042 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1044 size_t length = cursor->total_resid;
1046 switch (cursor->data->type) {
1047 case CEPH_MSG_DATA_PAGELIST:
1048 ceph_msg_data_pagelist_cursor_init(cursor, length);
1050 case CEPH_MSG_DATA_PAGES:
1051 ceph_msg_data_pages_cursor_init(cursor, length);
1054 case CEPH_MSG_DATA_BIO:
1055 ceph_msg_data_bio_cursor_init(cursor, length);
1057 #endif /* CONFIG_BLOCK */
1058 case CEPH_MSG_DATA_NONE:
1063 cursor->need_crc = true;
1066 static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1068 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1069 struct ceph_msg_data *data;
1072 BUG_ON(length > msg->data_length);
1073 BUG_ON(list_empty(&msg->data));
1075 cursor->data_head = &msg->data;
1076 cursor->total_resid = length;
1077 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1078 cursor->data = data;
1080 __ceph_msg_data_cursor_init(cursor);
1084 * Return the page containing the next piece to process for a given
1085 * data item, and supply the page offset and length of that piece.
1086 * Indicate whether this is the last piece in this data item.
1088 static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1089 size_t *page_offset, size_t *length,
1094 switch (cursor->data->type) {
1095 case CEPH_MSG_DATA_PAGELIST:
1096 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1098 case CEPH_MSG_DATA_PAGES:
1099 page = ceph_msg_data_pages_next(cursor, page_offset, length);
1102 case CEPH_MSG_DATA_BIO:
1103 page = ceph_msg_data_bio_next(cursor, page_offset, length);
1105 #endif /* CONFIG_BLOCK */
1106 case CEPH_MSG_DATA_NONE:
1112 BUG_ON(*page_offset + *length > PAGE_SIZE);
1115 *last_piece = cursor->last_piece;
1121 * Returns true if the result moves the cursor on to the next piece
1124 static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1129 BUG_ON(bytes > cursor->resid);
1130 switch (cursor->data->type) {
1131 case CEPH_MSG_DATA_PAGELIST:
1132 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1134 case CEPH_MSG_DATA_PAGES:
1135 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1138 case CEPH_MSG_DATA_BIO:
1139 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1141 #endif /* CONFIG_BLOCK */
1142 case CEPH_MSG_DATA_NONE:
1147 cursor->total_resid -= bytes;
1149 if (!cursor->resid && cursor->total_resid) {
1150 WARN_ON(!cursor->last_piece);
1151 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1152 cursor->data = list_entry_next(cursor->data, links);
1153 __ceph_msg_data_cursor_init(cursor);
1156 cursor->need_crc = new_piece;
1161 static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
1166 /* Initialize data cursor */
1168 ceph_msg_data_cursor_init(msg, (size_t)data_len);
1172 * Prepare footer for currently outgoing message, and finish things
1173 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1175 static void prepare_write_message_footer(struct ceph_connection *con)
1177 struct ceph_msg *m = con->out_msg;
1178 int v = con->out_kvec_left;
1180 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1182 dout("prepare_write_message_footer %p\n", con);
1183 con->out_kvec_is_msg = true;
1184 con->out_kvec[v].iov_base = &m->footer;
1185 con->out_kvec[v].iov_len = sizeof(m->footer);
1186 con->out_kvec_bytes += sizeof(m->footer);
1187 con->out_kvec_left++;
1188 con->out_more = m->more_to_follow;
1189 con->out_msg_done = true;
1193 * Prepare headers for the next outgoing message.
1195 static void prepare_write_message(struct ceph_connection *con)
1200 con_out_kvec_reset(con);
1201 con->out_kvec_is_msg = true;
1202 con->out_msg_done = false;
1204 /* Sneak an ack in there first? If we can get it into the same
1205 * TCP packet that's a good thing. */
1206 if (con->in_seq > con->in_seq_acked) {
1207 con->in_seq_acked = con->in_seq;
1208 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1209 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1210 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1211 &con->out_temp_ack);
1214 BUG_ON(list_empty(&con->out_queue));
1215 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1217 BUG_ON(m->con != con);
1219 /* put message on sent list */
1221 list_move_tail(&m->list_head, &con->out_sent);
1224 * only assign outgoing seq # if we haven't sent this message
1225 * yet. if it is requeued, resend with it's original seq.
1227 if (m->needs_out_seq) {
1228 m->hdr.seq = cpu_to_le64(++con->out_seq);
1229 m->needs_out_seq = false;
1231 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
1233 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1234 m, con->out_seq, le16_to_cpu(m->hdr.type),
1235 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1237 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1239 /* tag + hdr + front + middle */
1240 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1241 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1242 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1245 con_out_kvec_add(con, m->middle->vec.iov_len,
1246 m->middle->vec.iov_base);
1248 /* fill in crc (except data pages), footer */
1249 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1250 con->out_msg->hdr.crc = cpu_to_le32(crc);
1251 con->out_msg->footer.flags = 0;
1253 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1254 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1256 crc = crc32c(0, m->middle->vec.iov_base,
1257 m->middle->vec.iov_len);
1258 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1260 con->out_msg->footer.middle_crc = 0;
1261 dout("%s front_crc %u middle_crc %u\n", __func__,
1262 le32_to_cpu(con->out_msg->footer.front_crc),
1263 le32_to_cpu(con->out_msg->footer.middle_crc));
1265 /* is there a data payload? */
1266 con->out_msg->footer.data_crc = 0;
1267 if (m->data_length) {
1268 prepare_message_data(con->out_msg, m->data_length);
1269 con->out_more = 1; /* data + footer will follow */
1271 /* no, queue up footer too and be done */
1272 prepare_write_message_footer(con);
1275 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1281 static void prepare_write_ack(struct ceph_connection *con)
1283 dout("prepare_write_ack %p %llu -> %llu\n", con,
1284 con->in_seq_acked, con->in_seq);
1285 con->in_seq_acked = con->in_seq;
1287 con_out_kvec_reset(con);
1289 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1291 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1292 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1293 &con->out_temp_ack);
1295 con->out_more = 1; /* more will follow.. eventually.. */
1296 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1300 * Prepare to share the seq during handshake
1302 static void prepare_write_seq(struct ceph_connection *con)
1304 dout("prepare_write_seq %p %llu -> %llu\n", con,
1305 con->in_seq_acked, con->in_seq);
1306 con->in_seq_acked = con->in_seq;
1308 con_out_kvec_reset(con);
1310 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1311 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1312 &con->out_temp_ack);
1314 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1318 * Prepare to write keepalive byte.
1320 static void prepare_write_keepalive(struct ceph_connection *con)
1322 dout("prepare_write_keepalive %p\n", con);
1323 con_out_kvec_reset(con);
1324 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
1325 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1329 * Connection negotiation.
1332 static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1335 struct ceph_auth_handshake *auth;
1337 if (!con->ops->get_authorizer) {
1338 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1339 con->out_connect.authorizer_len = 0;
1343 /* Can't hold the mutex while getting authorizer */
1344 mutex_unlock(&con->mutex);
1345 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
1346 mutex_lock(&con->mutex);
1350 if (con->state != CON_STATE_NEGOTIATING)
1351 return ERR_PTR(-EAGAIN);
1353 con->auth_reply_buf = auth->authorizer_reply_buf;
1354 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
1359 * We connected to a peer and are saying hello.
1361 static void prepare_write_banner(struct ceph_connection *con)
1363 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1364 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1365 &con->msgr->my_enc_addr);
1368 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1371 static int prepare_write_connect(struct ceph_connection *con)
1373 unsigned int global_seq = get_global_seq(con->msgr, 0);
1376 struct ceph_auth_handshake *auth;
1378 switch (con->peer_name.type) {
1379 case CEPH_ENTITY_TYPE_MON:
1380 proto = CEPH_MONC_PROTOCOL;
1382 case CEPH_ENTITY_TYPE_OSD:
1383 proto = CEPH_OSDC_PROTOCOL;
1385 case CEPH_ENTITY_TYPE_MDS:
1386 proto = CEPH_MDSC_PROTOCOL;
1392 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1393 con->connect_seq, global_seq, proto);
1395 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
1396 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1397 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1398 con->out_connect.global_seq = cpu_to_le32(global_seq);
1399 con->out_connect.protocol_version = cpu_to_le32(proto);
1400 con->out_connect.flags = 0;
1402 auth_proto = CEPH_AUTH_UNKNOWN;
1403 auth = get_connect_authorizer(con, &auth_proto);
1405 return PTR_ERR(auth);
1407 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1408 con->out_connect.authorizer_len = auth ?
1409 cpu_to_le32(auth->authorizer_buf_len) : 0;
1411 con_out_kvec_add(con, sizeof (con->out_connect),
1413 if (auth && auth->authorizer_buf_len)
1414 con_out_kvec_add(con, auth->authorizer_buf_len,
1415 auth->authorizer_buf);
1418 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1424 * write as much of pending kvecs to the socket as we can.
1426 * 0 -> socket full, but more to do
1429 static int write_partial_kvec(struct ceph_connection *con)
1433 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1434 while (con->out_kvec_bytes > 0) {
1435 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1436 con->out_kvec_left, con->out_kvec_bytes,
1440 con->out_kvec_bytes -= ret;
1441 if (con->out_kvec_bytes == 0)
1444 /* account for full iov entries consumed */
1445 while (ret >= con->out_kvec_cur->iov_len) {
1446 BUG_ON(!con->out_kvec_left);
1447 ret -= con->out_kvec_cur->iov_len;
1448 con->out_kvec_cur++;
1449 con->out_kvec_left--;
1451 /* and for a partially-consumed entry */
1453 con->out_kvec_cur->iov_len -= ret;
1454 con->out_kvec_cur->iov_base += ret;
1457 con->out_kvec_left = 0;
1458 con->out_kvec_is_msg = false;
1461 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1462 con->out_kvec_bytes, con->out_kvec_left, ret);
1463 return ret; /* done! */
1466 static u32 ceph_crc32c_page(u32 crc, struct page *page,
1467 unsigned int page_offset,
1468 unsigned int length)
1473 BUG_ON(kaddr == NULL);
1474 crc = crc32c(crc, kaddr + page_offset, length);
1480 * Write as much message data payload as we can. If we finish, queue
1482 * 1 -> done, footer is now queued in out_kvec[].
1483 * 0 -> socket full, but more to do
1486 static int write_partial_message_data(struct ceph_connection *con)
1488 struct ceph_msg *msg = con->out_msg;
1489 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1490 bool do_datacrc = !con->msgr->nocrc;
1493 dout("%s %p msg %p\n", __func__, con, msg);
1495 if (list_empty(&msg->data))
1499 * Iterate through each page that contains data to be
1500 * written, and send as much as possible for each.
1502 * If we are calculating the data crc (the default), we will
1503 * need to map the page. If we have no pages, they have
1504 * been revoked, so use the zero page.
1506 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1507 while (cursor->resid) {
1515 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
1517 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
1518 length, last_piece);
1521 msg->footer.data_crc = cpu_to_le32(crc);
1525 if (do_datacrc && cursor->need_crc)
1526 crc = ceph_crc32c_page(crc, page, page_offset, length);
1527 need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret);
1530 dout("%s %p msg %p done\n", __func__, con, msg);
1532 /* prepare and queue up footer, too */
1534 msg->footer.data_crc = cpu_to_le32(crc);
1536 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1537 con_out_kvec_reset(con);
1538 prepare_write_message_footer(con);
1540 return 1; /* must return > 0 to indicate success */
1546 static int write_partial_skip(struct ceph_connection *con)
1550 while (con->out_skip > 0) {
1551 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
1553 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
1556 con->out_skip -= ret;
1564 * Prepare to read connection handshake, or an ack.
1566 static void prepare_read_banner(struct ceph_connection *con)
1568 dout("prepare_read_banner %p\n", con);
1569 con->in_base_pos = 0;
1572 static void prepare_read_connect(struct ceph_connection *con)
1574 dout("prepare_read_connect %p\n", con);
1575 con->in_base_pos = 0;
1578 static void prepare_read_ack(struct ceph_connection *con)
1580 dout("prepare_read_ack %p\n", con);
1581 con->in_base_pos = 0;
1584 static void prepare_read_seq(struct ceph_connection *con)
1586 dout("prepare_read_seq %p\n", con);
1587 con->in_base_pos = 0;
1588 con->in_tag = CEPH_MSGR_TAG_SEQ;
1591 static void prepare_read_tag(struct ceph_connection *con)
1593 dout("prepare_read_tag %p\n", con);
1594 con->in_base_pos = 0;
1595 con->in_tag = CEPH_MSGR_TAG_READY;
1599 * Prepare to read a message.
1601 static int prepare_read_message(struct ceph_connection *con)
1603 dout("prepare_read_message %p\n", con);
1604 BUG_ON(con->in_msg != NULL);
1605 con->in_base_pos = 0;
1606 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1611 static int read_partial(struct ceph_connection *con,
1612 int end, int size, void *object)
1614 while (con->in_base_pos < end) {
1615 int left = end - con->in_base_pos;
1616 int have = size - left;
1617 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1620 con->in_base_pos += ret;
1627 * Read all or part of the connect-side handshake on a new connection
1629 static int read_partial_banner(struct ceph_connection *con)
1635 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1638 size = strlen(CEPH_BANNER);
1640 ret = read_partial(con, end, size, con->in_banner);
1644 size = sizeof (con->actual_peer_addr);
1646 ret = read_partial(con, end, size, &con->actual_peer_addr);
1650 size = sizeof (con->peer_addr_for_me);
1652 ret = read_partial(con, end, size, &con->peer_addr_for_me);
1660 static int read_partial_connect(struct ceph_connection *con)
1666 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1668 size = sizeof (con->in_reply);
1670 ret = read_partial(con, end, size, &con->in_reply);
1674 size = le32_to_cpu(con->in_reply.authorizer_len);
1676 ret = read_partial(con, end, size, con->auth_reply_buf);
1680 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1681 con, (int)con->in_reply.tag,
1682 le32_to_cpu(con->in_reply.connect_seq),
1683 le32_to_cpu(con->in_reply.global_seq));
1690 * Verify the hello banner looks okay.
1692 static int verify_hello(struct ceph_connection *con)
1694 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1695 pr_err("connect to %s got bad banner\n",
1696 ceph_pr_addr(&con->peer_addr.in_addr));
1697 con->error_msg = "protocol error, bad banner";
1703 static bool addr_is_blank(struct sockaddr_storage *ss)
1705 switch (ss->ss_family) {
1707 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1710 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1711 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1712 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1713 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1718 static int addr_port(struct sockaddr_storage *ss)
1720 switch (ss->ss_family) {
1722 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1724 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1729 static void addr_set_port(struct sockaddr_storage *ss, int p)
1731 switch (ss->ss_family) {
1733 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1736 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1742 * Unlike other *_pton function semantics, zero indicates success.
1744 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1745 char delim, const char **ipend)
1747 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1748 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1750 memset(ss, 0, sizeof(*ss));
1752 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1753 ss->ss_family = AF_INET;
1757 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1758 ss->ss_family = AF_INET6;
1766 * Extract hostname string and resolve using kernel DNS facility.
1768 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1769 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1770 struct sockaddr_storage *ss, char delim, const char **ipend)
1772 const char *end, *delim_p;
1773 char *colon_p, *ip_addr = NULL;
1777 * The end of the hostname occurs immediately preceding the delimiter or
1778 * the port marker (':') where the delimiter takes precedence.
1780 delim_p = memchr(name, delim, namelen);
1781 colon_p = memchr(name, ':', namelen);
1783 if (delim_p && colon_p)
1784 end = delim_p < colon_p ? delim_p : colon_p;
1785 else if (!delim_p && colon_p)
1789 if (!end) /* case: hostname:/ */
1790 end = name + namelen;
1796 /* do dns_resolve upcall */
1797 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1799 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1807 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1808 ret, ret ? "failed" : ceph_pr_addr(ss));
1813 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1814 struct sockaddr_storage *ss, char delim, const char **ipend)
1821 * Parse a server name (IP or hostname). If a valid IP address is not found
1822 * then try to extract a hostname to resolve using userspace DNS upcall.
1824 static int ceph_parse_server_name(const char *name, size_t namelen,
1825 struct sockaddr_storage *ss, char delim, const char **ipend)
1829 ret = ceph_pton(name, namelen, ss, delim, ipend);
1831 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1837 * Parse an ip[:port] list into an addr array. Use the default
1838 * monitor port if a port isn't specified.
1840 int ceph_parse_ips(const char *c, const char *end,
1841 struct ceph_entity_addr *addr,
1842 int max_count, int *count)
1844 int i, ret = -EINVAL;
1847 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1848 for (i = 0; i < max_count; i++) {
1850 struct sockaddr_storage *ss = &addr[i].in_addr;
1859 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1868 dout("missing matching ']'\n");
1875 if (p < end && *p == ':') {
1878 while (p < end && *p >= '0' && *p <= '9') {
1879 port = (port * 10) + (*p - '0');
1883 port = CEPH_MON_PORT;
1884 else if (port > 65535)
1887 port = CEPH_MON_PORT;
1890 addr_set_port(ss, port);
1892 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1909 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1912 EXPORT_SYMBOL(ceph_parse_ips);
1914 static int process_banner(struct ceph_connection *con)
1916 dout("process_banner on %p\n", con);
1918 if (verify_hello(con) < 0)
1921 ceph_decode_addr(&con->actual_peer_addr);
1922 ceph_decode_addr(&con->peer_addr_for_me);
1925 * Make sure the other end is who we wanted. note that the other
1926 * end may not yet know their ip address, so if it's 0.0.0.0, give
1927 * them the benefit of the doubt.
1929 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1930 sizeof(con->peer_addr)) != 0 &&
1931 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1932 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1933 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1934 ceph_pr_addr(&con->peer_addr.in_addr),
1935 (int)le32_to_cpu(con->peer_addr.nonce),
1936 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1937 (int)le32_to_cpu(con->actual_peer_addr.nonce));
1938 con->error_msg = "wrong peer at address";
1943 * did we learn our address?
1945 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1946 int port = addr_port(&con->msgr->inst.addr.in_addr);
1948 memcpy(&con->msgr->inst.addr.in_addr,
1949 &con->peer_addr_for_me.in_addr,
1950 sizeof(con->peer_addr_for_me.in_addr));
1951 addr_set_port(&con->msgr->inst.addr.in_addr, port);
1952 encode_my_addr(con->msgr);
1953 dout("process_banner learned my addr is %s\n",
1954 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
1960 static int process_connect(struct ceph_connection *con)
1962 u64 sup_feat = con->msgr->supported_features;
1963 u64 req_feat = con->msgr->required_features;
1964 u64 server_feat = ceph_sanitize_features(
1965 le64_to_cpu(con->in_reply.features));
1968 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1970 switch (con->in_reply.tag) {
1971 case CEPH_MSGR_TAG_FEATURES:
1972 pr_err("%s%lld %s feature set mismatch,"
1973 " my %llx < server's %llx, missing %llx\n",
1974 ENTITY_NAME(con->peer_name),
1975 ceph_pr_addr(&con->peer_addr.in_addr),
1976 sup_feat, server_feat, server_feat & ~sup_feat);
1977 con->error_msg = "missing required protocol features";
1978 reset_connection(con);
1981 case CEPH_MSGR_TAG_BADPROTOVER:
1982 pr_err("%s%lld %s protocol version mismatch,"
1983 " my %d != server's %d\n",
1984 ENTITY_NAME(con->peer_name),
1985 ceph_pr_addr(&con->peer_addr.in_addr),
1986 le32_to_cpu(con->out_connect.protocol_version),
1987 le32_to_cpu(con->in_reply.protocol_version));
1988 con->error_msg = "protocol version mismatch";
1989 reset_connection(con);
1992 case CEPH_MSGR_TAG_BADAUTHORIZER:
1994 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1996 if (con->auth_retry == 2) {
1997 con->error_msg = "connect authorization failure";
2000 con_out_kvec_reset(con);
2001 ret = prepare_write_connect(con);
2004 prepare_read_connect(con);
2007 case CEPH_MSGR_TAG_RESETSESSION:
2009 * If we connected with a large connect_seq but the peer
2010 * has no record of a session with us (no connection, or
2011 * connect_seq == 0), they will send RESETSESION to indicate
2012 * that they must have reset their session, and may have
2015 dout("process_connect got RESET peer seq %u\n",
2016 le32_to_cpu(con->in_reply.connect_seq));
2017 pr_err("%s%lld %s connection reset\n",
2018 ENTITY_NAME(con->peer_name),
2019 ceph_pr_addr(&con->peer_addr.in_addr));
2020 reset_connection(con);
2021 con_out_kvec_reset(con);
2022 ret = prepare_write_connect(con);
2025 prepare_read_connect(con);
2027 /* Tell ceph about it. */
2028 mutex_unlock(&con->mutex);
2029 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2030 if (con->ops->peer_reset)
2031 con->ops->peer_reset(con);
2032 mutex_lock(&con->mutex);
2033 if (con->state != CON_STATE_NEGOTIATING)
2037 case CEPH_MSGR_TAG_RETRY_SESSION:
2039 * If we sent a smaller connect_seq than the peer has, try
2040 * again with a larger value.
2042 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2043 le32_to_cpu(con->out_connect.connect_seq),
2044 le32_to_cpu(con->in_reply.connect_seq));
2045 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2046 con_out_kvec_reset(con);
2047 ret = prepare_write_connect(con);
2050 prepare_read_connect(con);
2053 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2055 * If we sent a smaller global_seq than the peer has, try
2056 * again with a larger value.
2058 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2059 con->peer_global_seq,
2060 le32_to_cpu(con->in_reply.global_seq));
2061 get_global_seq(con->msgr,
2062 le32_to_cpu(con->in_reply.global_seq));
2063 con_out_kvec_reset(con);
2064 ret = prepare_write_connect(con);
2067 prepare_read_connect(con);
2070 case CEPH_MSGR_TAG_SEQ:
2071 case CEPH_MSGR_TAG_READY:
2072 if (req_feat & ~server_feat) {
2073 pr_err("%s%lld %s protocol feature mismatch,"
2074 " my required %llx > server's %llx, need %llx\n",
2075 ENTITY_NAME(con->peer_name),
2076 ceph_pr_addr(&con->peer_addr.in_addr),
2077 req_feat, server_feat, req_feat & ~server_feat);
2078 con->error_msg = "missing required protocol features";
2079 reset_connection(con);
2083 WARN_ON(con->state != CON_STATE_NEGOTIATING);
2084 con->state = CON_STATE_OPEN;
2085 con->auth_retry = 0; /* we authenticated; clear flag */
2086 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2088 con->peer_features = server_feat;
2089 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2090 con->peer_global_seq,
2091 le32_to_cpu(con->in_reply.connect_seq),
2093 WARN_ON(con->connect_seq !=
2094 le32_to_cpu(con->in_reply.connect_seq));
2096 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2097 con_flag_set(con, CON_FLAG_LOSSYTX);
2099 con->delay = 0; /* reset backoff memory */
2101 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2102 prepare_write_seq(con);
2103 prepare_read_seq(con);
2105 prepare_read_tag(con);
2109 case CEPH_MSGR_TAG_WAIT:
2111 * If there is a connection race (we are opening
2112 * connections to each other), one of us may just have
2113 * to WAIT. This shouldn't happen if we are the
2116 pr_err("process_connect got WAIT as client\n");
2117 con->error_msg = "protocol error, got WAIT as client";
2121 pr_err("connect protocol error, will retry\n");
2122 con->error_msg = "protocol error, garbage tag during connect";
2130 * read (part of) an ack
2132 static int read_partial_ack(struct ceph_connection *con)
2134 int size = sizeof (con->in_temp_ack);
2137 return read_partial(con, end, size, &con->in_temp_ack);
2141 * We can finally discard anything that's been acked.
2143 static void process_ack(struct ceph_connection *con)
2146 u64 ack = le64_to_cpu(con->in_temp_ack);
2149 while (!list_empty(&con->out_sent)) {
2150 m = list_first_entry(&con->out_sent, struct ceph_msg,
2152 seq = le64_to_cpu(m->hdr.seq);
2155 dout("got ack for seq %llu type %d at %p\n", seq,
2156 le16_to_cpu(m->hdr.type), m);
2157 m->ack_stamp = jiffies;
2160 prepare_read_tag(con);
2164 static int read_partial_message_section(struct ceph_connection *con,
2165 struct kvec *section,
2166 unsigned int sec_len, u32 *crc)
2172 while (section->iov_len < sec_len) {
2173 BUG_ON(section->iov_base == NULL);
2174 left = sec_len - section->iov_len;
2175 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2176 section->iov_len, left);
2179 section->iov_len += ret;
2181 if (section->iov_len == sec_len)
2182 *crc = crc32c(0, section->iov_base, section->iov_len);
2187 static int read_partial_msg_data(struct ceph_connection *con)
2189 struct ceph_msg *msg = con->in_msg;
2190 struct ceph_msg_data_cursor *cursor = &msg->cursor;
2191 const bool do_datacrc = !con->msgr->nocrc;
2199 if (list_empty(&msg->data))
2203 crc = con->in_data_crc;
2204 while (cursor->resid) {
2205 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
2207 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2210 con->in_data_crc = crc;
2216 crc = ceph_crc32c_page(crc, page, page_offset, ret);
2217 (void) ceph_msg_data_advance(&msg->cursor, (size_t)ret);
2220 con->in_data_crc = crc;
2222 return 1; /* must return > 0 to indicate success */
2226 * read (part of) a message.
2228 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2230 static int read_partial_message(struct ceph_connection *con)
2232 struct ceph_msg *m = con->in_msg;
2236 unsigned int front_len, middle_len, data_len;
2237 bool do_datacrc = !con->msgr->nocrc;
2241 dout("read_partial_message con %p msg %p\n", con, m);
2244 size = sizeof (con->in_hdr);
2246 ret = read_partial(con, end, size, &con->in_hdr);
2250 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2251 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2252 pr_err("read_partial_message bad hdr "
2253 " crc %u != expected %u\n",
2254 crc, con->in_hdr.crc);
2258 front_len = le32_to_cpu(con->in_hdr.front_len);
2259 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2261 middle_len = le32_to_cpu(con->in_hdr.middle_len);
2262 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2264 data_len = le32_to_cpu(con->in_hdr.data_len);
2265 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2269 seq = le64_to_cpu(con->in_hdr.seq);
2270 if ((s64)seq - (s64)con->in_seq < 1) {
2271 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2272 ENTITY_NAME(con->peer_name),
2273 ceph_pr_addr(&con->peer_addr.in_addr),
2274 seq, con->in_seq + 1);
2275 con->in_base_pos = -front_len - middle_len - data_len -
2277 con->in_tag = CEPH_MSGR_TAG_READY;
2279 } else if ((s64)seq - (s64)con->in_seq > 1) {
2280 pr_err("read_partial_message bad seq %lld expected %lld\n",
2281 seq, con->in_seq + 1);
2282 con->error_msg = "bad message sequence # for incoming message";
2286 /* allocate message? */
2290 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2291 front_len, data_len);
2292 ret = ceph_con_in_msg_alloc(con, &skip);
2296 BUG_ON(!con->in_msg ^ skip);
2297 if (con->in_msg && data_len > con->in_msg->data_length) {
2298 pr_warning("%s skipping long message (%u > %zd)\n",
2299 __func__, data_len, con->in_msg->data_length);
2300 ceph_msg_put(con->in_msg);
2305 /* skip this message */
2306 dout("alloc_msg said skip message\n");
2307 con->in_base_pos = -front_len - middle_len - data_len -
2309 con->in_tag = CEPH_MSGR_TAG_READY;
2314 BUG_ON(!con->in_msg);
2315 BUG_ON(con->in_msg->con != con);
2317 m->front.iov_len = 0; /* haven't read it yet */
2319 m->middle->vec.iov_len = 0;
2321 /* prepare for data payload, if any */
2324 prepare_message_data(con->in_msg, data_len);
2328 ret = read_partial_message_section(con, &m->front, front_len,
2329 &con->in_front_crc);
2335 ret = read_partial_message_section(con, &m->middle->vec,
2337 &con->in_middle_crc);
2344 ret = read_partial_msg_data(con);
2350 size = sizeof (m->footer);
2352 ret = read_partial(con, end, size, &m->footer);
2356 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2357 m, front_len, m->footer.front_crc, middle_len,
2358 m->footer.middle_crc, data_len, m->footer.data_crc);
2361 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2362 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2363 m, con->in_front_crc, m->footer.front_crc);
2366 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2367 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2368 m, con->in_middle_crc, m->footer.middle_crc);
2372 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2373 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2374 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2375 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2379 return 1; /* done! */
2383 * Process message. This happens in the worker thread. The callback should
2384 * be careful not to do anything that waits on other incoming messages or it
2387 static void process_message(struct ceph_connection *con)
2389 struct ceph_msg *msg;
2391 BUG_ON(con->in_msg->con != con);
2392 con->in_msg->con = NULL;
2397 /* if first message, set peer_name */
2398 if (con->peer_name.type == 0)
2399 con->peer_name = msg->hdr.src;
2402 mutex_unlock(&con->mutex);
2404 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2405 msg, le64_to_cpu(msg->hdr.seq),
2406 ENTITY_NAME(msg->hdr.src),
2407 le16_to_cpu(msg->hdr.type),
2408 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2409 le32_to_cpu(msg->hdr.front_len),
2410 le32_to_cpu(msg->hdr.data_len),
2411 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2412 con->ops->dispatch(con, msg);
2414 mutex_lock(&con->mutex);
2419 * Write something to the socket. Called in a worker thread when the
2420 * socket appears to be writeable and we have something ready to send.
2422 static int try_write(struct ceph_connection *con)
2426 dout("try_write start %p state %lu\n", con, con->state);
2429 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2431 /* open the socket first? */
2432 if (con->state == CON_STATE_PREOPEN) {
2434 con->state = CON_STATE_CONNECTING;
2436 con_out_kvec_reset(con);
2437 prepare_write_banner(con);
2438 prepare_read_banner(con);
2440 BUG_ON(con->in_msg);
2441 con->in_tag = CEPH_MSGR_TAG_READY;
2442 dout("try_write initiating connect on %p new state %lu\n",
2444 ret = ceph_tcp_connect(con);
2446 con->error_msg = "connect error";
2452 /* kvec data queued? */
2453 if (con->out_skip) {
2454 ret = write_partial_skip(con);
2458 if (con->out_kvec_left) {
2459 ret = write_partial_kvec(con);
2466 if (con->out_msg_done) {
2467 ceph_msg_put(con->out_msg);
2468 con->out_msg = NULL; /* we're done with this one */
2472 ret = write_partial_message_data(con);
2474 goto more_kvec; /* we need to send the footer, too! */
2478 dout("try_write write_partial_message_data err %d\n",
2485 if (con->state == CON_STATE_OPEN) {
2486 /* is anything else pending? */
2487 if (!list_empty(&con->out_queue)) {
2488 prepare_write_message(con);
2491 if (con->in_seq > con->in_seq_acked) {
2492 prepare_write_ack(con);
2495 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2496 prepare_write_keepalive(con);
2501 /* Nothing to do! */
2502 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2503 dout("try_write nothing else to write.\n");
2506 dout("try_write done on %p ret %d\n", con, ret);
2513 * Read what we can from the socket.
2515 static int try_read(struct ceph_connection *con)
2520 dout("try_read start on %p state %lu\n", con, con->state);
2521 if (con->state != CON_STATE_CONNECTING &&
2522 con->state != CON_STATE_NEGOTIATING &&
2523 con->state != CON_STATE_OPEN)
2528 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2531 if (con->state == CON_STATE_CONNECTING) {
2532 dout("try_read connecting\n");
2533 ret = read_partial_banner(con);
2536 ret = process_banner(con);
2540 con->state = CON_STATE_NEGOTIATING;
2543 * Received banner is good, exchange connection info.
2544 * Do not reset out_kvec, as sending our banner raced
2545 * with receiving peer banner after connect completed.
2547 ret = prepare_write_connect(con);
2550 prepare_read_connect(con);
2552 /* Send connection info before awaiting response */
2556 if (con->state == CON_STATE_NEGOTIATING) {
2557 dout("try_read negotiating\n");
2558 ret = read_partial_connect(con);
2561 ret = process_connect(con);
2567 WARN_ON(con->state != CON_STATE_OPEN);
2569 if (con->in_base_pos < 0) {
2571 * skipping + discarding content.
2573 * FIXME: there must be a better way to do this!
2575 static char buf[SKIP_BUF_SIZE];
2576 int skip = min((int) sizeof (buf), -con->in_base_pos);
2578 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2579 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2582 con->in_base_pos += ret;
2583 if (con->in_base_pos)
2586 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2590 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2593 dout("try_read got tag %d\n", (int)con->in_tag);
2594 switch (con->in_tag) {
2595 case CEPH_MSGR_TAG_MSG:
2596 prepare_read_message(con);
2598 case CEPH_MSGR_TAG_ACK:
2599 prepare_read_ack(con);
2601 case CEPH_MSGR_TAG_CLOSE:
2602 con_close_socket(con);
2603 con->state = CON_STATE_CLOSED;
2609 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2610 ret = read_partial_message(con);
2614 con->error_msg = "bad crc";
2618 con->error_msg = "io error";
2623 if (con->in_tag == CEPH_MSGR_TAG_READY)
2625 process_message(con);
2626 if (con->state == CON_STATE_OPEN)
2627 prepare_read_tag(con);
2630 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2631 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2633 * the final handshake seq exchange is semantically
2634 * equivalent to an ACK
2636 ret = read_partial_ack(con);
2644 dout("try_read done on %p ret %d\n", con, ret);
2648 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2649 con->error_msg = "protocol error, garbage tag";
2656 * Atomically queue work on a connection after the specified delay.
2657 * Bump @con reference to avoid races with connection teardown.
2658 * Returns 0 if work was queued, or an error code otherwise.
2660 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2662 if (!con->ops->get(con)) {
2663 dout("%s %p ref count 0\n", __func__, con);
2668 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2669 dout("%s %p - already queued\n", __func__, con);
2675 dout("%s %p %lu\n", __func__, con, delay);
2680 static void queue_con(struct ceph_connection *con)
2682 (void) queue_con_delay(con, 0);
2685 static bool con_sock_closed(struct ceph_connection *con)
2687 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2691 case CON_STATE_ ## x: \
2692 con->error_msg = "socket closed (con state " #x ")"; \
2695 switch (con->state) {
2703 pr_warning("%s con %p unrecognized state %lu\n",
2704 __func__, con, con->state);
2705 con->error_msg = "unrecognized con state";
2714 static bool con_backoff(struct ceph_connection *con)
2718 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2721 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2723 dout("%s: con %p FAILED to back off %lu\n", __func__,
2725 BUG_ON(ret == -ENOENT);
2726 con_flag_set(con, CON_FLAG_BACKOFF);
2732 /* Finish fault handling; con->mutex must *not* be held here */
2734 static void con_fault_finish(struct ceph_connection *con)
2737 * in case we faulted due to authentication, invalidate our
2738 * current tickets so that we can get new ones.
2740 if (con->auth_retry && con->ops->invalidate_authorizer) {
2741 dout("calling invalidate_authorizer()\n");
2742 con->ops->invalidate_authorizer(con);
2745 if (con->ops->fault)
2746 con->ops->fault(con);
2750 * Do some work on a connection. Drop a connection ref when we're done.
2752 static void con_work(struct work_struct *work)
2754 struct ceph_connection *con = container_of(work, struct ceph_connection,
2758 mutex_lock(&con->mutex);
2762 if ((fault = con_sock_closed(con))) {
2763 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2766 if (con_backoff(con)) {
2767 dout("%s: con %p BACKOFF\n", __func__, con);
2770 if (con->state == CON_STATE_STANDBY) {
2771 dout("%s: con %p STANDBY\n", __func__, con);
2774 if (con->state == CON_STATE_CLOSED) {
2775 dout("%s: con %p CLOSED\n", __func__, con);
2779 if (con->state == CON_STATE_PREOPEN) {
2780 dout("%s: con %p PREOPEN\n", __func__, con);
2784 ret = try_read(con);
2788 con->error_msg = "socket error on read";
2793 ret = try_write(con);
2797 con->error_msg = "socket error on write";
2801 break; /* If we make it to here, we're done */
2805 mutex_unlock(&con->mutex);
2808 con_fault_finish(con);
2814 * Generic error/fault handler. A retry mechanism is used with
2815 * exponential backoff
2817 static void con_fault(struct ceph_connection *con)
2819 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2820 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2821 dout("fault %p state %lu to peer %s\n",
2822 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2824 WARN_ON(con->state != CON_STATE_CONNECTING &&
2825 con->state != CON_STATE_NEGOTIATING &&
2826 con->state != CON_STATE_OPEN);
2828 con_close_socket(con);
2830 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2831 dout("fault on LOSSYTX channel, marking CLOSED\n");
2832 con->state = CON_STATE_CLOSED;
2837 BUG_ON(con->in_msg->con != con);
2838 con->in_msg->con = NULL;
2839 ceph_msg_put(con->in_msg);
2844 /* Requeue anything that hasn't been acked */
2845 list_splice_init(&con->out_sent, &con->out_queue);
2847 /* If there are no messages queued or keepalive pending, place
2848 * the connection in a STANDBY state */
2849 if (list_empty(&con->out_queue) &&
2850 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
2851 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2852 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2853 con->state = CON_STATE_STANDBY;
2855 /* retry after a delay. */
2856 con->state = CON_STATE_PREOPEN;
2857 if (con->delay == 0)
2858 con->delay = BASE_DELAY_INTERVAL;
2859 else if (con->delay < MAX_DELAY_INTERVAL)
2861 con_flag_set(con, CON_FLAG_BACKOFF);
2869 * initialize a new messenger instance
2871 void ceph_messenger_init(struct ceph_messenger *msgr,
2872 struct ceph_entity_addr *myaddr,
2873 u64 supported_features,
2874 u64 required_features,
2877 msgr->supported_features = supported_features;
2878 msgr->required_features = required_features;
2880 spin_lock_init(&msgr->global_seq_lock);
2883 msgr->inst.addr = *myaddr;
2885 /* select a random nonce */
2886 msgr->inst.addr.type = 0;
2887 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2888 encode_my_addr(msgr);
2889 msgr->nocrc = nocrc;
2891 atomic_set(&msgr->stopping, 0);
2893 dout("%s %p\n", __func__, msgr);
2895 EXPORT_SYMBOL(ceph_messenger_init);
2897 static void clear_standby(struct ceph_connection *con)
2899 /* come back from STANDBY? */
2900 if (con->state == CON_STATE_STANDBY) {
2901 dout("clear_standby %p and ++connect_seq\n", con);
2902 con->state = CON_STATE_PREOPEN;
2904 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2905 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
2910 * Queue up an outgoing message on the given connection.
2912 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2915 msg->hdr.src = con->msgr->inst.name;
2916 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2917 msg->needs_out_seq = true;
2919 mutex_lock(&con->mutex);
2921 if (con->state == CON_STATE_CLOSED) {
2922 dout("con_send %p closed, dropping %p\n", con, msg);
2924 mutex_unlock(&con->mutex);
2928 BUG_ON(msg->con != NULL);
2929 msg->con = con->ops->get(con);
2930 BUG_ON(msg->con == NULL);
2932 BUG_ON(!list_empty(&msg->list_head));
2933 list_add_tail(&msg->list_head, &con->out_queue);
2934 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2935 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2936 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2937 le32_to_cpu(msg->hdr.front_len),
2938 le32_to_cpu(msg->hdr.middle_len),
2939 le32_to_cpu(msg->hdr.data_len));
2942 mutex_unlock(&con->mutex);
2944 /* if there wasn't anything waiting to send before, queue
2946 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
2949 EXPORT_SYMBOL(ceph_con_send);
2952 * Revoke a message that was previously queued for send
2954 void ceph_msg_revoke(struct ceph_msg *msg)
2956 struct ceph_connection *con = msg->con;
2959 return; /* Message not in our possession */
2961 mutex_lock(&con->mutex);
2962 if (!list_empty(&msg->list_head)) {
2963 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
2964 list_del_init(&msg->list_head);
2965 BUG_ON(msg->con == NULL);
2966 msg->con->ops->put(msg->con);
2972 if (con->out_msg == msg) {
2973 dout("%s %p msg %p - was sending\n", __func__, con, msg);
2974 con->out_msg = NULL;
2975 if (con->out_kvec_is_msg) {
2976 con->out_skip = con->out_kvec_bytes;
2977 con->out_kvec_is_msg = false;
2983 mutex_unlock(&con->mutex);
2987 * Revoke a message that we may be reading data into
2989 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
2991 struct ceph_connection *con;
2993 BUG_ON(msg == NULL);
2995 dout("%s msg %p null con\n", __func__, msg);
2997 return; /* Message not in our possession */
3001 mutex_lock(&con->mutex);
3002 if (con->in_msg == msg) {
3003 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3004 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3005 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
3007 /* skip rest of message */
3008 dout("%s %p msg %p revoked\n", __func__, con, msg);
3009 con->in_base_pos = con->in_base_pos -
3010 sizeof(struct ceph_msg_header) -
3014 sizeof(struct ceph_msg_footer);
3015 ceph_msg_put(con->in_msg);
3017 con->in_tag = CEPH_MSGR_TAG_READY;
3020 dout("%s %p in_msg %p msg %p no-op\n",
3021 __func__, con, con->in_msg, msg);
3023 mutex_unlock(&con->mutex);
3027 * Queue a keepalive byte to ensure the tcp connection is alive.
3029 void ceph_con_keepalive(struct ceph_connection *con)
3031 dout("con_keepalive %p\n", con);
3032 mutex_lock(&con->mutex);
3034 mutex_unlock(&con->mutex);
3035 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3036 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3039 EXPORT_SYMBOL(ceph_con_keepalive);
3041 static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
3043 struct ceph_msg_data *data;
3045 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3048 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
3051 INIT_LIST_HEAD(&data->links);
3056 static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3061 WARN_ON(!list_empty(&data->links));
3062 if (data->type == CEPH_MSG_DATA_PAGELIST) {
3063 ceph_pagelist_release(data->pagelist);
3064 kfree(data->pagelist);
3066 kmem_cache_free(ceph_msg_data_cache, data);
3069 void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
3070 size_t length, size_t alignment)
3072 struct ceph_msg_data *data;
3077 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3079 data->pages = pages;
3080 data->length = length;
3081 data->alignment = alignment & ~PAGE_MASK;
3083 list_add_tail(&data->links, &msg->data);
3084 msg->data_length += length;
3086 EXPORT_SYMBOL(ceph_msg_data_add_pages);
3088 void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
3089 struct ceph_pagelist *pagelist)
3091 struct ceph_msg_data *data;
3094 BUG_ON(!pagelist->length);
3096 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3098 data->pagelist = pagelist;
3100 list_add_tail(&data->links, &msg->data);
3101 msg->data_length += pagelist->length;
3103 EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
3106 void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
3109 struct ceph_msg_data *data;
3113 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3116 data->bio_length = length;
3118 list_add_tail(&data->links, &msg->data);
3119 msg->data_length += length;
3121 EXPORT_SYMBOL(ceph_msg_data_add_bio);
3122 #endif /* CONFIG_BLOCK */
3125 * construct a new message with given type, size
3126 * the new msg has a ref count of 1.
3128 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3133 m = kmem_cache_zalloc(ceph_msg_cache, flags);
3137 m->hdr.type = cpu_to_le16(type);
3138 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3139 m->hdr.front_len = cpu_to_le32(front_len);
3141 INIT_LIST_HEAD(&m->list_head);
3142 kref_init(&m->kref);
3143 INIT_LIST_HEAD(&m->data);
3147 m->front.iov_base = ceph_kvmalloc(front_len, flags);
3148 if (m->front.iov_base == NULL) {
3149 dout("ceph_msg_new can't allocate %d bytes\n",
3154 m->front.iov_base = NULL;
3156 m->front_alloc_len = m->front.iov_len = front_len;
3158 dout("ceph_msg_new %p front %d\n", m, front_len);
3165 pr_err("msg_new can't create type %d front %d\n", type,
3169 dout("msg_new can't create type %d front %d\n", type,
3174 EXPORT_SYMBOL(ceph_msg_new);
3177 * Allocate "middle" portion of a message, if it is needed and wasn't
3178 * allocated by alloc_msg. This allows us to read a small fixed-size
3179 * per-type header in the front and then gracefully fail (i.e.,
3180 * propagate the error to the caller based on info in the front) when
3181 * the middle is too large.
3183 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3185 int type = le16_to_cpu(msg->hdr.type);
3186 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3188 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3189 ceph_msg_type_name(type), middle_len);
3190 BUG_ON(!middle_len);
3191 BUG_ON(msg->middle);
3193 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3200 * Allocate a message for receiving an incoming message on a
3201 * connection, and save the result in con->in_msg. Uses the
3202 * connection's private alloc_msg op if available.
3204 * Returns 0 on success, or a negative error code.
3206 * On success, if we set *skip = 1:
3207 * - the next message should be skipped and ignored.
3208 * - con->in_msg == NULL
3209 * or if we set *skip = 0:
3210 * - con->in_msg is non-null.
3211 * On error (ENOMEM, EAGAIN, ...),
3212 * - con->in_msg == NULL
3214 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3216 struct ceph_msg_header *hdr = &con->in_hdr;
3217 int middle_len = le32_to_cpu(hdr->middle_len);
3218 struct ceph_msg *msg;
3221 BUG_ON(con->in_msg != NULL);
3222 BUG_ON(!con->ops->alloc_msg);
3224 mutex_unlock(&con->mutex);
3225 msg = con->ops->alloc_msg(con, hdr, skip);
3226 mutex_lock(&con->mutex);
3227 if (con->state != CON_STATE_OPEN) {
3235 con->in_msg->con = con->ops->get(con);
3236 BUG_ON(con->in_msg->con == NULL);
3239 * Null message pointer means either we should skip
3240 * this message or we couldn't allocate memory. The
3241 * former is not an error.
3245 con->error_msg = "error allocating memory for incoming message";
3249 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3251 if (middle_len && !con->in_msg->middle) {
3252 ret = ceph_alloc_middle(con, con->in_msg);
3254 ceph_msg_put(con->in_msg);
3264 * Free a generically kmalloc'd message.
3266 void ceph_msg_kfree(struct ceph_msg *m)
3268 dout("msg_kfree %p\n", m);
3269 ceph_kvfree(m->front.iov_base);
3270 kmem_cache_free(ceph_msg_cache, m);
3274 * Drop a msg ref. Destroy as needed.
3276 void ceph_msg_last_put(struct kref *kref)
3278 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3280 struct list_head *links;
3281 struct list_head *next;
3283 dout("ceph_msg_put last one on %p\n", m);
3284 WARN_ON(!list_empty(&m->list_head));
3286 /* drop middle, data, if any */
3288 ceph_buffer_put(m->middle);
3292 list_splice_init(&m->data, &data);
3293 list_for_each_safe(links, next, &data) {
3294 struct ceph_msg_data *data;
3296 data = list_entry(links, struct ceph_msg_data, links);
3297 list_del_init(links);
3298 ceph_msg_data_destroy(data);
3303 ceph_msgpool_put(m->pool, m);
3307 EXPORT_SYMBOL(ceph_msg_last_put);
3309 void ceph_msg_dump(struct ceph_msg *msg)
3311 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3312 msg->front_alloc_len, msg->data_length);
3313 print_hex_dump(KERN_DEBUG, "header: ",
3314 DUMP_PREFIX_OFFSET, 16, 1,
3315 &msg->hdr, sizeof(msg->hdr), true);
3316 print_hex_dump(KERN_DEBUG, " front: ",
3317 DUMP_PREFIX_OFFSET, 16, 1,
3318 msg->front.iov_base, msg->front.iov_len, true);
3320 print_hex_dump(KERN_DEBUG, "middle: ",
3321 DUMP_PREFIX_OFFSET, 16, 1,
3322 msg->middle->vec.iov_base,
3323 msg->middle->vec.iov_len, true);
3324 print_hex_dump(KERN_DEBUG, "footer: ",
3325 DUMP_PREFIX_OFFSET, 16, 1,
3326 &msg->footer, sizeof(msg->footer), true);
3328 EXPORT_SYMBOL(ceph_msg_dump);