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())
295 * The number of active work items is limited by the number of
296 * connections, so leave @max_active at default.
298 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
302 pr_err("msgr_init failed to create workqueue\n");
307 EXPORT_SYMBOL(ceph_msgr_init);
309 void ceph_msgr_exit(void)
311 BUG_ON(ceph_msgr_wq == NULL);
315 EXPORT_SYMBOL(ceph_msgr_exit);
317 void ceph_msgr_flush(void)
319 flush_workqueue(ceph_msgr_wq);
321 EXPORT_SYMBOL(ceph_msgr_flush);
323 /* Connection socket state transition functions */
325 static void con_sock_state_init(struct ceph_connection *con)
329 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
330 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
331 printk("%s: unexpected old state %d\n", __func__, old_state);
332 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
333 CON_SOCK_STATE_CLOSED);
336 static void con_sock_state_connecting(struct ceph_connection *con)
340 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
341 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
342 printk("%s: unexpected old state %d\n", __func__, old_state);
343 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
344 CON_SOCK_STATE_CONNECTING);
347 static void con_sock_state_connected(struct ceph_connection *con)
351 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
352 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
353 printk("%s: unexpected old state %d\n", __func__, old_state);
354 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
355 CON_SOCK_STATE_CONNECTED);
358 static void con_sock_state_closing(struct ceph_connection *con)
362 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
363 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
364 old_state != CON_SOCK_STATE_CONNECTED &&
365 old_state != CON_SOCK_STATE_CLOSING))
366 printk("%s: unexpected old state %d\n", __func__, old_state);
367 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
368 CON_SOCK_STATE_CLOSING);
371 static void con_sock_state_closed(struct ceph_connection *con)
375 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
376 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
377 old_state != CON_SOCK_STATE_CLOSING &&
378 old_state != CON_SOCK_STATE_CONNECTING &&
379 old_state != CON_SOCK_STATE_CLOSED))
380 printk("%s: unexpected old state %d\n", __func__, old_state);
381 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
382 CON_SOCK_STATE_CLOSED);
386 * socket callback functions
389 /* data available on socket, or listen socket received a connect */
390 static void ceph_sock_data_ready(struct sock *sk, int count_unused)
392 struct ceph_connection *con = sk->sk_user_data;
393 if (atomic_read(&con->msgr->stopping)) {
397 if (sk->sk_state != TCP_CLOSE_WAIT) {
398 dout("%s on %p state = %lu, queueing work\n", __func__,
404 /* socket has buffer space for writing */
405 static void ceph_sock_write_space(struct sock *sk)
407 struct ceph_connection *con = sk->sk_user_data;
409 /* only queue to workqueue if there is data we want to write,
410 * and there is sufficient space in the socket buffer to accept
411 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
412 * doesn't get called again until try_write() fills the socket
413 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
414 * and net/core/stream.c:sk_stream_write_space().
416 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
417 if (sk_stream_is_writeable(sk)) {
418 dout("%s %p queueing write work\n", __func__, con);
419 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
423 dout("%s %p nothing to write\n", __func__, con);
427 /* socket's state has changed */
428 static void ceph_sock_state_change(struct sock *sk)
430 struct ceph_connection *con = sk->sk_user_data;
432 dout("%s %p state = %lu sk_state = %u\n", __func__,
433 con, con->state, sk->sk_state);
435 switch (sk->sk_state) {
437 dout("%s TCP_CLOSE\n", __func__);
439 dout("%s TCP_CLOSE_WAIT\n", __func__);
440 con_sock_state_closing(con);
441 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
444 case TCP_ESTABLISHED:
445 dout("%s TCP_ESTABLISHED\n", __func__);
446 con_sock_state_connected(con);
449 default: /* Everything else is uninteresting */
455 * set up socket callbacks
457 static void set_sock_callbacks(struct socket *sock,
458 struct ceph_connection *con)
460 struct sock *sk = sock->sk;
461 sk->sk_user_data = con;
462 sk->sk_data_ready = ceph_sock_data_ready;
463 sk->sk_write_space = ceph_sock_write_space;
464 sk->sk_state_change = ceph_sock_state_change;
473 * initiate connection to a remote socket.
475 static int ceph_tcp_connect(struct ceph_connection *con)
477 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
482 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
486 sock->sk->sk_allocation = GFP_NOFS;
488 #ifdef CONFIG_LOCKDEP
489 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
492 set_sock_callbacks(sock, con);
494 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
496 con_sock_state_connecting(con);
497 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
499 if (ret == -EINPROGRESS) {
500 dout("connect %s EINPROGRESS sk_state = %u\n",
501 ceph_pr_addr(&con->peer_addr.in_addr),
503 } else if (ret < 0) {
504 pr_err("connect %s error %d\n",
505 ceph_pr_addr(&con->peer_addr.in_addr), ret);
507 con->error_msg = "connect error";
515 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
517 struct kvec iov = {buf, len};
518 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
521 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
527 static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
528 int page_offset, size_t length)
533 BUG_ON(page_offset + length > PAGE_SIZE);
537 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
544 * write something. @more is true if caller will be sending more data
547 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
548 size_t kvlen, size_t len, int more)
550 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
554 msg.msg_flags |= MSG_MORE;
556 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
558 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
564 static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
565 int offset, size_t size, bool more)
567 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
570 ret = kernel_sendpage(sock, page, offset, size, flags);
577 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
578 int offset, size_t size, bool more)
583 /* sendpage cannot properly handle pages with page_count == 0,
584 * we need to fallback to sendmsg if that's the case */
585 if (page_count(page) >= 1)
586 return __ceph_tcp_sendpage(sock, page, offset, size, more);
588 iov.iov_base = kmap(page) + offset;
590 ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more);
597 * Shutdown/close the socket for the given connection.
599 static int con_close_socket(struct ceph_connection *con)
603 dout("con_close_socket on %p sock %p\n", con, con->sock);
605 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
606 sock_release(con->sock);
611 * Forcibly clear the SOCK_CLOSED flag. It gets set
612 * independent of the connection mutex, and we could have
613 * received a socket close event before we had the chance to
614 * shut the socket down.
616 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
618 con_sock_state_closed(con);
623 * Reset a connection. Discard all incoming and outgoing messages
624 * and clear *_seq state.
626 static void ceph_msg_remove(struct ceph_msg *msg)
628 list_del_init(&msg->list_head);
629 BUG_ON(msg->con == NULL);
630 msg->con->ops->put(msg->con);
635 static void ceph_msg_remove_list(struct list_head *head)
637 while (!list_empty(head)) {
638 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
640 ceph_msg_remove(msg);
644 static void reset_connection(struct ceph_connection *con)
646 /* reset connection, out_queue, msg_ and connect_seq */
647 /* discard existing out_queue and msg_seq */
648 dout("reset_connection %p\n", con);
649 ceph_msg_remove_list(&con->out_queue);
650 ceph_msg_remove_list(&con->out_sent);
653 BUG_ON(con->in_msg->con != con);
654 con->in_msg->con = NULL;
655 ceph_msg_put(con->in_msg);
660 con->connect_seq = 0;
663 ceph_msg_put(con->out_msg);
667 con->in_seq_acked = 0;
671 * mark a peer down. drop any open connections.
673 void ceph_con_close(struct ceph_connection *con)
675 mutex_lock(&con->mutex);
676 dout("con_close %p peer %s\n", con,
677 ceph_pr_addr(&con->peer_addr.in_addr));
678 con->state = CON_STATE_CLOSED;
680 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
681 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
682 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
683 con_flag_clear(con, CON_FLAG_BACKOFF);
685 reset_connection(con);
686 con->peer_global_seq = 0;
687 cancel_delayed_work(&con->work);
688 con_close_socket(con);
689 mutex_unlock(&con->mutex);
691 EXPORT_SYMBOL(ceph_con_close);
694 * Reopen a closed connection, with a new peer address.
696 void ceph_con_open(struct ceph_connection *con,
697 __u8 entity_type, __u64 entity_num,
698 struct ceph_entity_addr *addr)
700 mutex_lock(&con->mutex);
701 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
703 WARN_ON(con->state != CON_STATE_CLOSED);
704 con->state = CON_STATE_PREOPEN;
706 con->peer_name.type = (__u8) entity_type;
707 con->peer_name.num = cpu_to_le64(entity_num);
709 memcpy(&con->peer_addr, addr, sizeof(*addr));
710 con->delay = 0; /* reset backoff memory */
711 mutex_unlock(&con->mutex);
714 EXPORT_SYMBOL(ceph_con_open);
717 * return true if this connection ever successfully opened
719 bool ceph_con_opened(struct ceph_connection *con)
721 return con->connect_seq > 0;
725 * initialize a new connection.
727 void ceph_con_init(struct ceph_connection *con, void *private,
728 const struct ceph_connection_operations *ops,
729 struct ceph_messenger *msgr)
731 dout("con_init %p\n", con);
732 memset(con, 0, sizeof(*con));
733 con->private = private;
737 con_sock_state_init(con);
739 mutex_init(&con->mutex);
740 INIT_LIST_HEAD(&con->out_queue);
741 INIT_LIST_HEAD(&con->out_sent);
742 INIT_DELAYED_WORK(&con->work, con_work);
744 con->state = CON_STATE_CLOSED;
746 EXPORT_SYMBOL(ceph_con_init);
750 * We maintain a global counter to order connection attempts. Get
751 * a unique seq greater than @gt.
753 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
757 spin_lock(&msgr->global_seq_lock);
758 if (msgr->global_seq < gt)
759 msgr->global_seq = gt;
760 ret = ++msgr->global_seq;
761 spin_unlock(&msgr->global_seq_lock);
765 static void con_out_kvec_reset(struct ceph_connection *con)
767 con->out_kvec_left = 0;
768 con->out_kvec_bytes = 0;
769 con->out_kvec_cur = &con->out_kvec[0];
772 static void con_out_kvec_add(struct ceph_connection *con,
773 size_t size, void *data)
777 index = con->out_kvec_left;
778 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
780 con->out_kvec[index].iov_len = size;
781 con->out_kvec[index].iov_base = data;
782 con->out_kvec_left++;
783 con->out_kvec_bytes += size;
789 * For a bio data item, a piece is whatever remains of the next
790 * entry in the current bio iovec, or the first entry in the next
793 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
796 struct ceph_msg_data *data = cursor->data;
799 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
804 cursor->resid = min(length, data->bio_length);
806 cursor->bvec_iter = bio->bi_iter;
808 cursor->resid <= bio_iter_len(bio, cursor->bvec_iter);
811 static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
815 struct ceph_msg_data *data = cursor->data;
817 struct bio_vec bio_vec;
819 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
824 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
826 *page_offset = (size_t) bio_vec.bv_offset;
827 BUG_ON(*page_offset >= PAGE_SIZE);
828 if (cursor->last_piece) /* pagelist offset is always 0 */
829 *length = cursor->resid;
831 *length = (size_t) bio_vec.bv_len;
832 BUG_ON(*length > cursor->resid);
833 BUG_ON(*page_offset + *length > PAGE_SIZE);
835 return bio_vec.bv_page;
838 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
842 struct bio_vec bio_vec;
844 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
849 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
851 /* Advance the cursor offset */
853 BUG_ON(cursor->resid < bytes);
854 cursor->resid -= bytes;
856 bio_advance_iter(bio, &cursor->bvec_iter, bytes);
858 if (bytes < bio_vec.bv_len)
859 return false; /* more bytes to process in this segment */
861 /* Move on to the next segment, and possibly the next bio */
863 if (!cursor->bvec_iter.bi_size) {
867 cursor->bvec_iter = bio->bi_iter;
869 memset(&cursor->bvec_iter, 0,
870 sizeof(cursor->bvec_iter));
873 if (!cursor->last_piece) {
874 BUG_ON(!cursor->resid);
876 /* A short read is OK, so use <= rather than == */
877 if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter))
878 cursor->last_piece = true;
883 #endif /* CONFIG_BLOCK */
886 * For a page array, a piece comes from the first page in the array
887 * that has not already been fully consumed.
889 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
892 struct ceph_msg_data *data = cursor->data;
895 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
897 BUG_ON(!data->pages);
898 BUG_ON(!data->length);
900 cursor->resid = min(length, data->length);
901 page_count = calc_pages_for(data->alignment, (u64)data->length);
902 cursor->page_offset = data->alignment & ~PAGE_MASK;
903 cursor->page_index = 0;
904 BUG_ON(page_count > (int)USHRT_MAX);
905 cursor->page_count = (unsigned short)page_count;
906 BUG_ON(length > SIZE_MAX - cursor->page_offset);
907 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
911 ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
912 size_t *page_offset, size_t *length)
914 struct ceph_msg_data *data = cursor->data;
916 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
918 BUG_ON(cursor->page_index >= cursor->page_count);
919 BUG_ON(cursor->page_offset >= PAGE_SIZE);
921 *page_offset = cursor->page_offset;
922 if (cursor->last_piece)
923 *length = cursor->resid;
925 *length = PAGE_SIZE - *page_offset;
927 return data->pages[cursor->page_index];
930 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
933 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
935 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
937 /* Advance the cursor page offset */
939 cursor->resid -= bytes;
940 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
941 if (!bytes || cursor->page_offset)
942 return false; /* more bytes to process in the current page */
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 */
1029 /* Move on to the next page */
1031 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1032 cursor->page = list_entry_next(cursor->page, lru);
1033 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1039 * Message data is handled (sent or received) in pieces, where each
1040 * piece resides on a single page. The network layer might not
1041 * consume an entire piece at once. A data item's cursor keeps
1042 * track of which piece is next to process and how much remains to
1043 * be processed in that piece. It also tracks whether the current
1044 * piece is the last one in the data item.
1046 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1048 size_t length = cursor->total_resid;
1050 switch (cursor->data->type) {
1051 case CEPH_MSG_DATA_PAGELIST:
1052 ceph_msg_data_pagelist_cursor_init(cursor, length);
1054 case CEPH_MSG_DATA_PAGES:
1055 ceph_msg_data_pages_cursor_init(cursor, length);
1058 case CEPH_MSG_DATA_BIO:
1059 ceph_msg_data_bio_cursor_init(cursor, length);
1061 #endif /* CONFIG_BLOCK */
1062 case CEPH_MSG_DATA_NONE:
1067 cursor->need_crc = true;
1070 static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1072 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1073 struct ceph_msg_data *data;
1076 BUG_ON(length > msg->data_length);
1077 BUG_ON(list_empty(&msg->data));
1079 cursor->data_head = &msg->data;
1080 cursor->total_resid = length;
1081 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1082 cursor->data = data;
1084 __ceph_msg_data_cursor_init(cursor);
1088 * Return the page containing the next piece to process for a given
1089 * data item, and supply the page offset and length of that piece.
1090 * Indicate whether this is the last piece in this data item.
1092 static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1093 size_t *page_offset, size_t *length,
1098 switch (cursor->data->type) {
1099 case CEPH_MSG_DATA_PAGELIST:
1100 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1102 case CEPH_MSG_DATA_PAGES:
1103 page = ceph_msg_data_pages_next(cursor, page_offset, length);
1106 case CEPH_MSG_DATA_BIO:
1107 page = ceph_msg_data_bio_next(cursor, page_offset, length);
1109 #endif /* CONFIG_BLOCK */
1110 case CEPH_MSG_DATA_NONE:
1116 BUG_ON(*page_offset + *length > PAGE_SIZE);
1119 *last_piece = cursor->last_piece;
1125 * Returns true if the result moves the cursor on to the next piece
1128 static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1133 BUG_ON(bytes > cursor->resid);
1134 switch (cursor->data->type) {
1135 case CEPH_MSG_DATA_PAGELIST:
1136 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1138 case CEPH_MSG_DATA_PAGES:
1139 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1142 case CEPH_MSG_DATA_BIO:
1143 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1145 #endif /* CONFIG_BLOCK */
1146 case CEPH_MSG_DATA_NONE:
1151 cursor->total_resid -= bytes;
1153 if (!cursor->resid && cursor->total_resid) {
1154 WARN_ON(!cursor->last_piece);
1155 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1156 cursor->data = list_entry_next(cursor->data, links);
1157 __ceph_msg_data_cursor_init(cursor);
1160 cursor->need_crc = new_piece;
1165 static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
1170 /* Initialize data cursor */
1172 ceph_msg_data_cursor_init(msg, (size_t)data_len);
1176 * Prepare footer for currently outgoing message, and finish things
1177 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1179 static void prepare_write_message_footer(struct ceph_connection *con)
1181 struct ceph_msg *m = con->out_msg;
1182 int v = con->out_kvec_left;
1184 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1186 dout("prepare_write_message_footer %p\n", con);
1187 con->out_kvec_is_msg = true;
1188 con->out_kvec[v].iov_base = &m->footer;
1189 con->out_kvec[v].iov_len = sizeof(m->footer);
1190 con->out_kvec_bytes += sizeof(m->footer);
1191 con->out_kvec_left++;
1192 con->out_more = m->more_to_follow;
1193 con->out_msg_done = true;
1197 * Prepare headers for the next outgoing message.
1199 static void prepare_write_message(struct ceph_connection *con)
1204 con_out_kvec_reset(con);
1205 con->out_kvec_is_msg = true;
1206 con->out_msg_done = false;
1208 /* Sneak an ack in there first? If we can get it into the same
1209 * TCP packet that's a good thing. */
1210 if (con->in_seq > con->in_seq_acked) {
1211 con->in_seq_acked = con->in_seq;
1212 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1213 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1214 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1215 &con->out_temp_ack);
1218 BUG_ON(list_empty(&con->out_queue));
1219 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1221 BUG_ON(m->con != con);
1223 /* put message on sent list */
1225 list_move_tail(&m->list_head, &con->out_sent);
1228 * only assign outgoing seq # if we haven't sent this message
1229 * yet. if it is requeued, resend with it's original seq.
1231 if (m->needs_out_seq) {
1232 m->hdr.seq = cpu_to_le64(++con->out_seq);
1233 m->needs_out_seq = false;
1235 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
1237 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1238 m, con->out_seq, le16_to_cpu(m->hdr.type),
1239 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1241 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1243 /* tag + hdr + front + middle */
1244 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1245 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1246 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1249 con_out_kvec_add(con, m->middle->vec.iov_len,
1250 m->middle->vec.iov_base);
1252 /* fill in crc (except data pages), footer */
1253 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1254 con->out_msg->hdr.crc = cpu_to_le32(crc);
1255 con->out_msg->footer.flags = 0;
1257 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1258 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1260 crc = crc32c(0, m->middle->vec.iov_base,
1261 m->middle->vec.iov_len);
1262 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1264 con->out_msg->footer.middle_crc = 0;
1265 dout("%s front_crc %u middle_crc %u\n", __func__,
1266 le32_to_cpu(con->out_msg->footer.front_crc),
1267 le32_to_cpu(con->out_msg->footer.middle_crc));
1269 /* is there a data payload? */
1270 con->out_msg->footer.data_crc = 0;
1271 if (m->data_length) {
1272 prepare_message_data(con->out_msg, m->data_length);
1273 con->out_more = 1; /* data + footer will follow */
1275 /* no, queue up footer too and be done */
1276 prepare_write_message_footer(con);
1279 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1285 static void prepare_write_ack(struct ceph_connection *con)
1287 dout("prepare_write_ack %p %llu -> %llu\n", con,
1288 con->in_seq_acked, con->in_seq);
1289 con->in_seq_acked = con->in_seq;
1291 con_out_kvec_reset(con);
1293 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1295 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1296 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1297 &con->out_temp_ack);
1299 con->out_more = 1; /* more will follow.. eventually.. */
1300 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1304 * Prepare to share the seq during handshake
1306 static void prepare_write_seq(struct ceph_connection *con)
1308 dout("prepare_write_seq %p %llu -> %llu\n", con,
1309 con->in_seq_acked, con->in_seq);
1310 con->in_seq_acked = con->in_seq;
1312 con_out_kvec_reset(con);
1314 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1315 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1316 &con->out_temp_ack);
1318 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1322 * Prepare to write keepalive byte.
1324 static void prepare_write_keepalive(struct ceph_connection *con)
1326 dout("prepare_write_keepalive %p\n", con);
1327 con_out_kvec_reset(con);
1328 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
1329 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1333 * Connection negotiation.
1336 static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1339 struct ceph_auth_handshake *auth;
1341 if (!con->ops->get_authorizer) {
1342 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1343 con->out_connect.authorizer_len = 0;
1347 /* Can't hold the mutex while getting authorizer */
1348 mutex_unlock(&con->mutex);
1349 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
1350 mutex_lock(&con->mutex);
1354 if (con->state != CON_STATE_NEGOTIATING)
1355 return ERR_PTR(-EAGAIN);
1357 con->auth_reply_buf = auth->authorizer_reply_buf;
1358 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
1363 * We connected to a peer and are saying hello.
1365 static void prepare_write_banner(struct ceph_connection *con)
1367 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1368 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1369 &con->msgr->my_enc_addr);
1372 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1375 static int prepare_write_connect(struct ceph_connection *con)
1377 unsigned int global_seq = get_global_seq(con->msgr, 0);
1380 struct ceph_auth_handshake *auth;
1382 switch (con->peer_name.type) {
1383 case CEPH_ENTITY_TYPE_MON:
1384 proto = CEPH_MONC_PROTOCOL;
1386 case CEPH_ENTITY_TYPE_OSD:
1387 proto = CEPH_OSDC_PROTOCOL;
1389 case CEPH_ENTITY_TYPE_MDS:
1390 proto = CEPH_MDSC_PROTOCOL;
1396 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1397 con->connect_seq, global_seq, proto);
1399 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
1400 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1401 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1402 con->out_connect.global_seq = cpu_to_le32(global_seq);
1403 con->out_connect.protocol_version = cpu_to_le32(proto);
1404 con->out_connect.flags = 0;
1406 auth_proto = CEPH_AUTH_UNKNOWN;
1407 auth = get_connect_authorizer(con, &auth_proto);
1409 return PTR_ERR(auth);
1411 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1412 con->out_connect.authorizer_len = auth ?
1413 cpu_to_le32(auth->authorizer_buf_len) : 0;
1415 con_out_kvec_add(con, sizeof (con->out_connect),
1417 if (auth && auth->authorizer_buf_len)
1418 con_out_kvec_add(con, auth->authorizer_buf_len,
1419 auth->authorizer_buf);
1422 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1428 * write as much of pending kvecs to the socket as we can.
1430 * 0 -> socket full, but more to do
1433 static int write_partial_kvec(struct ceph_connection *con)
1437 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1438 while (con->out_kvec_bytes > 0) {
1439 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1440 con->out_kvec_left, con->out_kvec_bytes,
1444 con->out_kvec_bytes -= ret;
1445 if (con->out_kvec_bytes == 0)
1448 /* account for full iov entries consumed */
1449 while (ret >= con->out_kvec_cur->iov_len) {
1450 BUG_ON(!con->out_kvec_left);
1451 ret -= con->out_kvec_cur->iov_len;
1452 con->out_kvec_cur++;
1453 con->out_kvec_left--;
1455 /* and for a partially-consumed entry */
1457 con->out_kvec_cur->iov_len -= ret;
1458 con->out_kvec_cur->iov_base += ret;
1461 con->out_kvec_left = 0;
1462 con->out_kvec_is_msg = false;
1465 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1466 con->out_kvec_bytes, con->out_kvec_left, ret);
1467 return ret; /* done! */
1470 static u32 ceph_crc32c_page(u32 crc, struct page *page,
1471 unsigned int page_offset,
1472 unsigned int length)
1477 BUG_ON(kaddr == NULL);
1478 crc = crc32c(crc, kaddr + page_offset, length);
1484 * Write as much message data payload as we can. If we finish, queue
1486 * 1 -> done, footer is now queued in out_kvec[].
1487 * 0 -> socket full, but more to do
1490 static int write_partial_message_data(struct ceph_connection *con)
1492 struct ceph_msg *msg = con->out_msg;
1493 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1494 bool do_datacrc = !con->msgr->nocrc;
1497 dout("%s %p msg %p\n", __func__, con, msg);
1499 if (list_empty(&msg->data))
1503 * Iterate through each page that contains data to be
1504 * written, and send as much as possible for each.
1506 * If we are calculating the data crc (the default), we will
1507 * need to map the page. If we have no pages, they have
1508 * been revoked, so use the zero page.
1510 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1511 while (cursor->resid) {
1519 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
1521 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
1522 length, last_piece);
1525 msg->footer.data_crc = cpu_to_le32(crc);
1529 if (do_datacrc && cursor->need_crc)
1530 crc = ceph_crc32c_page(crc, page, page_offset, length);
1531 need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret);
1534 dout("%s %p msg %p done\n", __func__, con, msg);
1536 /* prepare and queue up footer, too */
1538 msg->footer.data_crc = cpu_to_le32(crc);
1540 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1541 con_out_kvec_reset(con);
1542 prepare_write_message_footer(con);
1544 return 1; /* must return > 0 to indicate success */
1550 static int write_partial_skip(struct ceph_connection *con)
1554 while (con->out_skip > 0) {
1555 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
1557 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
1560 con->out_skip -= ret;
1568 * Prepare to read connection handshake, or an ack.
1570 static void prepare_read_banner(struct ceph_connection *con)
1572 dout("prepare_read_banner %p\n", con);
1573 con->in_base_pos = 0;
1576 static void prepare_read_connect(struct ceph_connection *con)
1578 dout("prepare_read_connect %p\n", con);
1579 con->in_base_pos = 0;
1582 static void prepare_read_ack(struct ceph_connection *con)
1584 dout("prepare_read_ack %p\n", con);
1585 con->in_base_pos = 0;
1588 static void prepare_read_seq(struct ceph_connection *con)
1590 dout("prepare_read_seq %p\n", con);
1591 con->in_base_pos = 0;
1592 con->in_tag = CEPH_MSGR_TAG_SEQ;
1595 static void prepare_read_tag(struct ceph_connection *con)
1597 dout("prepare_read_tag %p\n", con);
1598 con->in_base_pos = 0;
1599 con->in_tag = CEPH_MSGR_TAG_READY;
1603 * Prepare to read a message.
1605 static int prepare_read_message(struct ceph_connection *con)
1607 dout("prepare_read_message %p\n", con);
1608 BUG_ON(con->in_msg != NULL);
1609 con->in_base_pos = 0;
1610 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1615 static int read_partial(struct ceph_connection *con,
1616 int end, int size, void *object)
1618 while (con->in_base_pos < end) {
1619 int left = end - con->in_base_pos;
1620 int have = size - left;
1621 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1624 con->in_base_pos += ret;
1631 * Read all or part of the connect-side handshake on a new connection
1633 static int read_partial_banner(struct ceph_connection *con)
1639 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1642 size = strlen(CEPH_BANNER);
1644 ret = read_partial(con, end, size, con->in_banner);
1648 size = sizeof (con->actual_peer_addr);
1650 ret = read_partial(con, end, size, &con->actual_peer_addr);
1654 size = sizeof (con->peer_addr_for_me);
1656 ret = read_partial(con, end, size, &con->peer_addr_for_me);
1664 static int read_partial_connect(struct ceph_connection *con)
1670 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1672 size = sizeof (con->in_reply);
1674 ret = read_partial(con, end, size, &con->in_reply);
1678 size = le32_to_cpu(con->in_reply.authorizer_len);
1680 ret = read_partial(con, end, size, con->auth_reply_buf);
1684 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1685 con, (int)con->in_reply.tag,
1686 le32_to_cpu(con->in_reply.connect_seq),
1687 le32_to_cpu(con->in_reply.global_seq));
1694 * Verify the hello banner looks okay.
1696 static int verify_hello(struct ceph_connection *con)
1698 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1699 pr_err("connect to %s got bad banner\n",
1700 ceph_pr_addr(&con->peer_addr.in_addr));
1701 con->error_msg = "protocol error, bad banner";
1707 static bool addr_is_blank(struct sockaddr_storage *ss)
1709 switch (ss->ss_family) {
1711 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1714 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1715 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1716 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1717 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1722 static int addr_port(struct sockaddr_storage *ss)
1724 switch (ss->ss_family) {
1726 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1728 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1733 static void addr_set_port(struct sockaddr_storage *ss, int p)
1735 switch (ss->ss_family) {
1737 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1740 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1746 * Unlike other *_pton function semantics, zero indicates success.
1748 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1749 char delim, const char **ipend)
1751 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1752 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1754 memset(ss, 0, sizeof(*ss));
1756 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1757 ss->ss_family = AF_INET;
1761 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1762 ss->ss_family = AF_INET6;
1770 * Extract hostname string and resolve using kernel DNS facility.
1772 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1773 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1774 struct sockaddr_storage *ss, char delim, const char **ipend)
1776 const char *end, *delim_p;
1777 char *colon_p, *ip_addr = NULL;
1781 * The end of the hostname occurs immediately preceding the delimiter or
1782 * the port marker (':') where the delimiter takes precedence.
1784 delim_p = memchr(name, delim, namelen);
1785 colon_p = memchr(name, ':', namelen);
1787 if (delim_p && colon_p)
1788 end = delim_p < colon_p ? delim_p : colon_p;
1789 else if (!delim_p && colon_p)
1793 if (!end) /* case: hostname:/ */
1794 end = name + namelen;
1800 /* do dns_resolve upcall */
1801 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1803 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1811 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1812 ret, ret ? "failed" : ceph_pr_addr(ss));
1817 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1818 struct sockaddr_storage *ss, char delim, const char **ipend)
1825 * Parse a server name (IP or hostname). If a valid IP address is not found
1826 * then try to extract a hostname to resolve using userspace DNS upcall.
1828 static int ceph_parse_server_name(const char *name, size_t namelen,
1829 struct sockaddr_storage *ss, char delim, const char **ipend)
1833 ret = ceph_pton(name, namelen, ss, delim, ipend);
1835 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1841 * Parse an ip[:port] list into an addr array. Use the default
1842 * monitor port if a port isn't specified.
1844 int ceph_parse_ips(const char *c, const char *end,
1845 struct ceph_entity_addr *addr,
1846 int max_count, int *count)
1848 int i, ret = -EINVAL;
1851 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1852 for (i = 0; i < max_count; i++) {
1854 struct sockaddr_storage *ss = &addr[i].in_addr;
1863 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1872 dout("missing matching ']'\n");
1879 if (p < end && *p == ':') {
1882 while (p < end && *p >= '0' && *p <= '9') {
1883 port = (port * 10) + (*p - '0');
1887 port = CEPH_MON_PORT;
1888 else if (port > 65535)
1891 port = CEPH_MON_PORT;
1894 addr_set_port(ss, port);
1896 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1913 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1916 EXPORT_SYMBOL(ceph_parse_ips);
1918 static int process_banner(struct ceph_connection *con)
1920 dout("process_banner on %p\n", con);
1922 if (verify_hello(con) < 0)
1925 ceph_decode_addr(&con->actual_peer_addr);
1926 ceph_decode_addr(&con->peer_addr_for_me);
1929 * Make sure the other end is who we wanted. note that the other
1930 * end may not yet know their ip address, so if it's 0.0.0.0, give
1931 * them the benefit of the doubt.
1933 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1934 sizeof(con->peer_addr)) != 0 &&
1935 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1936 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1937 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1938 ceph_pr_addr(&con->peer_addr.in_addr),
1939 (int)le32_to_cpu(con->peer_addr.nonce),
1940 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1941 (int)le32_to_cpu(con->actual_peer_addr.nonce));
1942 con->error_msg = "wrong peer at address";
1947 * did we learn our address?
1949 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1950 int port = addr_port(&con->msgr->inst.addr.in_addr);
1952 memcpy(&con->msgr->inst.addr.in_addr,
1953 &con->peer_addr_for_me.in_addr,
1954 sizeof(con->peer_addr_for_me.in_addr));
1955 addr_set_port(&con->msgr->inst.addr.in_addr, port);
1956 encode_my_addr(con->msgr);
1957 dout("process_banner learned my addr is %s\n",
1958 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
1964 static int process_connect(struct ceph_connection *con)
1966 u64 sup_feat = con->msgr->supported_features;
1967 u64 req_feat = con->msgr->required_features;
1968 u64 server_feat = ceph_sanitize_features(
1969 le64_to_cpu(con->in_reply.features));
1972 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1974 switch (con->in_reply.tag) {
1975 case CEPH_MSGR_TAG_FEATURES:
1976 pr_err("%s%lld %s feature set mismatch,"
1977 " my %llx < server's %llx, missing %llx\n",
1978 ENTITY_NAME(con->peer_name),
1979 ceph_pr_addr(&con->peer_addr.in_addr),
1980 sup_feat, server_feat, server_feat & ~sup_feat);
1981 con->error_msg = "missing required protocol features";
1982 reset_connection(con);
1985 case CEPH_MSGR_TAG_BADPROTOVER:
1986 pr_err("%s%lld %s protocol version mismatch,"
1987 " my %d != server's %d\n",
1988 ENTITY_NAME(con->peer_name),
1989 ceph_pr_addr(&con->peer_addr.in_addr),
1990 le32_to_cpu(con->out_connect.protocol_version),
1991 le32_to_cpu(con->in_reply.protocol_version));
1992 con->error_msg = "protocol version mismatch";
1993 reset_connection(con);
1996 case CEPH_MSGR_TAG_BADAUTHORIZER:
1998 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2000 if (con->auth_retry == 2) {
2001 con->error_msg = "connect authorization failure";
2004 con_out_kvec_reset(con);
2005 ret = prepare_write_connect(con);
2008 prepare_read_connect(con);
2011 case CEPH_MSGR_TAG_RESETSESSION:
2013 * If we connected with a large connect_seq but the peer
2014 * has no record of a session with us (no connection, or
2015 * connect_seq == 0), they will send RESETSESION to indicate
2016 * that they must have reset their session, and may have
2019 dout("process_connect got RESET peer seq %u\n",
2020 le32_to_cpu(con->in_reply.connect_seq));
2021 pr_err("%s%lld %s connection reset\n",
2022 ENTITY_NAME(con->peer_name),
2023 ceph_pr_addr(&con->peer_addr.in_addr));
2024 reset_connection(con);
2025 con_out_kvec_reset(con);
2026 ret = prepare_write_connect(con);
2029 prepare_read_connect(con);
2031 /* Tell ceph about it. */
2032 mutex_unlock(&con->mutex);
2033 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2034 if (con->ops->peer_reset)
2035 con->ops->peer_reset(con);
2036 mutex_lock(&con->mutex);
2037 if (con->state != CON_STATE_NEGOTIATING)
2041 case CEPH_MSGR_TAG_RETRY_SESSION:
2043 * If we sent a smaller connect_seq than the peer has, try
2044 * again with a larger value.
2046 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2047 le32_to_cpu(con->out_connect.connect_seq),
2048 le32_to_cpu(con->in_reply.connect_seq));
2049 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2050 con_out_kvec_reset(con);
2051 ret = prepare_write_connect(con);
2054 prepare_read_connect(con);
2057 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2059 * If we sent a smaller global_seq than the peer has, try
2060 * again with a larger value.
2062 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2063 con->peer_global_seq,
2064 le32_to_cpu(con->in_reply.global_seq));
2065 get_global_seq(con->msgr,
2066 le32_to_cpu(con->in_reply.global_seq));
2067 con_out_kvec_reset(con);
2068 ret = prepare_write_connect(con);
2071 prepare_read_connect(con);
2074 case CEPH_MSGR_TAG_SEQ:
2075 case CEPH_MSGR_TAG_READY:
2076 if (req_feat & ~server_feat) {
2077 pr_err("%s%lld %s protocol feature mismatch,"
2078 " my required %llx > server's %llx, need %llx\n",
2079 ENTITY_NAME(con->peer_name),
2080 ceph_pr_addr(&con->peer_addr.in_addr),
2081 req_feat, server_feat, req_feat & ~server_feat);
2082 con->error_msg = "missing required protocol features";
2083 reset_connection(con);
2087 WARN_ON(con->state != CON_STATE_NEGOTIATING);
2088 con->state = CON_STATE_OPEN;
2089 con->auth_retry = 0; /* we authenticated; clear flag */
2090 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2092 con->peer_features = server_feat;
2093 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2094 con->peer_global_seq,
2095 le32_to_cpu(con->in_reply.connect_seq),
2097 WARN_ON(con->connect_seq !=
2098 le32_to_cpu(con->in_reply.connect_seq));
2100 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2101 con_flag_set(con, CON_FLAG_LOSSYTX);
2103 con->delay = 0; /* reset backoff memory */
2105 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2106 prepare_write_seq(con);
2107 prepare_read_seq(con);
2109 prepare_read_tag(con);
2113 case CEPH_MSGR_TAG_WAIT:
2115 * If there is a connection race (we are opening
2116 * connections to each other), one of us may just have
2117 * to WAIT. This shouldn't happen if we are the
2120 pr_err("process_connect got WAIT as client\n");
2121 con->error_msg = "protocol error, got WAIT as client";
2125 pr_err("connect protocol error, will retry\n");
2126 con->error_msg = "protocol error, garbage tag during connect";
2134 * read (part of) an ack
2136 static int read_partial_ack(struct ceph_connection *con)
2138 int size = sizeof (con->in_temp_ack);
2141 return read_partial(con, end, size, &con->in_temp_ack);
2145 * We can finally discard anything that's been acked.
2147 static void process_ack(struct ceph_connection *con)
2150 u64 ack = le64_to_cpu(con->in_temp_ack);
2153 while (!list_empty(&con->out_sent)) {
2154 m = list_first_entry(&con->out_sent, struct ceph_msg,
2156 seq = le64_to_cpu(m->hdr.seq);
2159 dout("got ack for seq %llu type %d at %p\n", seq,
2160 le16_to_cpu(m->hdr.type), m);
2161 m->ack_stamp = jiffies;
2164 prepare_read_tag(con);
2168 static int read_partial_message_section(struct ceph_connection *con,
2169 struct kvec *section,
2170 unsigned int sec_len, u32 *crc)
2176 while (section->iov_len < sec_len) {
2177 BUG_ON(section->iov_base == NULL);
2178 left = sec_len - section->iov_len;
2179 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2180 section->iov_len, left);
2183 section->iov_len += ret;
2185 if (section->iov_len == sec_len)
2186 *crc = crc32c(0, section->iov_base, section->iov_len);
2191 static int read_partial_msg_data(struct ceph_connection *con)
2193 struct ceph_msg *msg = con->in_msg;
2194 struct ceph_msg_data_cursor *cursor = &msg->cursor;
2195 const bool do_datacrc = !con->msgr->nocrc;
2203 if (list_empty(&msg->data))
2207 crc = con->in_data_crc;
2208 while (cursor->resid) {
2209 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
2211 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2214 con->in_data_crc = crc;
2220 crc = ceph_crc32c_page(crc, page, page_offset, ret);
2221 (void) ceph_msg_data_advance(&msg->cursor, (size_t)ret);
2224 con->in_data_crc = crc;
2226 return 1; /* must return > 0 to indicate success */
2230 * read (part of) a message.
2232 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2234 static int read_partial_message(struct ceph_connection *con)
2236 struct ceph_msg *m = con->in_msg;
2240 unsigned int front_len, middle_len, data_len;
2241 bool do_datacrc = !con->msgr->nocrc;
2245 dout("read_partial_message con %p msg %p\n", con, m);
2248 size = sizeof (con->in_hdr);
2250 ret = read_partial(con, end, size, &con->in_hdr);
2254 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2255 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2256 pr_err("read_partial_message bad hdr "
2257 " crc %u != expected %u\n",
2258 crc, con->in_hdr.crc);
2262 front_len = le32_to_cpu(con->in_hdr.front_len);
2263 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2265 middle_len = le32_to_cpu(con->in_hdr.middle_len);
2266 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2268 data_len = le32_to_cpu(con->in_hdr.data_len);
2269 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2273 seq = le64_to_cpu(con->in_hdr.seq);
2274 if ((s64)seq - (s64)con->in_seq < 1) {
2275 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2276 ENTITY_NAME(con->peer_name),
2277 ceph_pr_addr(&con->peer_addr.in_addr),
2278 seq, con->in_seq + 1);
2279 con->in_base_pos = -front_len - middle_len - data_len -
2281 con->in_tag = CEPH_MSGR_TAG_READY;
2283 } else if ((s64)seq - (s64)con->in_seq > 1) {
2284 pr_err("read_partial_message bad seq %lld expected %lld\n",
2285 seq, con->in_seq + 1);
2286 con->error_msg = "bad message sequence # for incoming message";
2290 /* allocate message? */
2294 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2295 front_len, data_len);
2296 ret = ceph_con_in_msg_alloc(con, &skip);
2300 BUG_ON(!con->in_msg ^ skip);
2301 if (con->in_msg && data_len > con->in_msg->data_length) {
2302 pr_warning("%s skipping long message (%u > %zd)\n",
2303 __func__, data_len, con->in_msg->data_length);
2304 ceph_msg_put(con->in_msg);
2309 /* skip this message */
2310 dout("alloc_msg said skip message\n");
2311 con->in_base_pos = -front_len - middle_len - data_len -
2313 con->in_tag = CEPH_MSGR_TAG_READY;
2318 BUG_ON(!con->in_msg);
2319 BUG_ON(con->in_msg->con != con);
2321 m->front.iov_len = 0; /* haven't read it yet */
2323 m->middle->vec.iov_len = 0;
2325 /* prepare for data payload, if any */
2328 prepare_message_data(con->in_msg, data_len);
2332 ret = read_partial_message_section(con, &m->front, front_len,
2333 &con->in_front_crc);
2339 ret = read_partial_message_section(con, &m->middle->vec,
2341 &con->in_middle_crc);
2348 ret = read_partial_msg_data(con);
2354 size = sizeof (m->footer);
2356 ret = read_partial(con, end, size, &m->footer);
2360 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2361 m, front_len, m->footer.front_crc, middle_len,
2362 m->footer.middle_crc, data_len, m->footer.data_crc);
2365 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2366 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2367 m, con->in_front_crc, m->footer.front_crc);
2370 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2371 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2372 m, con->in_middle_crc, m->footer.middle_crc);
2376 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2377 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2378 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2379 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2383 return 1; /* done! */
2387 * Process message. This happens in the worker thread. The callback should
2388 * be careful not to do anything that waits on other incoming messages or it
2391 static void process_message(struct ceph_connection *con)
2393 struct ceph_msg *msg;
2395 BUG_ON(con->in_msg->con != con);
2396 con->in_msg->con = NULL;
2401 /* if first message, set peer_name */
2402 if (con->peer_name.type == 0)
2403 con->peer_name = msg->hdr.src;
2406 mutex_unlock(&con->mutex);
2408 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2409 msg, le64_to_cpu(msg->hdr.seq),
2410 ENTITY_NAME(msg->hdr.src),
2411 le16_to_cpu(msg->hdr.type),
2412 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2413 le32_to_cpu(msg->hdr.front_len),
2414 le32_to_cpu(msg->hdr.data_len),
2415 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2416 con->ops->dispatch(con, msg);
2418 mutex_lock(&con->mutex);
2423 * Write something to the socket. Called in a worker thread when the
2424 * socket appears to be writeable and we have something ready to send.
2426 static int try_write(struct ceph_connection *con)
2430 dout("try_write start %p state %lu\n", con, con->state);
2433 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2435 /* open the socket first? */
2436 if (con->state == CON_STATE_PREOPEN) {
2438 con->state = CON_STATE_CONNECTING;
2440 con_out_kvec_reset(con);
2441 prepare_write_banner(con);
2442 prepare_read_banner(con);
2444 BUG_ON(con->in_msg);
2445 con->in_tag = CEPH_MSGR_TAG_READY;
2446 dout("try_write initiating connect on %p new state %lu\n",
2448 ret = ceph_tcp_connect(con);
2450 con->error_msg = "connect error";
2456 /* kvec data queued? */
2457 if (con->out_skip) {
2458 ret = write_partial_skip(con);
2462 if (con->out_kvec_left) {
2463 ret = write_partial_kvec(con);
2470 if (con->out_msg_done) {
2471 ceph_msg_put(con->out_msg);
2472 con->out_msg = NULL; /* we're done with this one */
2476 ret = write_partial_message_data(con);
2478 goto more_kvec; /* we need to send the footer, too! */
2482 dout("try_write write_partial_message_data err %d\n",
2489 if (con->state == CON_STATE_OPEN) {
2490 /* is anything else pending? */
2491 if (!list_empty(&con->out_queue)) {
2492 prepare_write_message(con);
2495 if (con->in_seq > con->in_seq_acked) {
2496 prepare_write_ack(con);
2499 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2500 prepare_write_keepalive(con);
2505 /* Nothing to do! */
2506 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2507 dout("try_write nothing else to write.\n");
2510 dout("try_write done on %p ret %d\n", con, ret);
2517 * Read what we can from the socket.
2519 static int try_read(struct ceph_connection *con)
2524 dout("try_read start on %p state %lu\n", con, con->state);
2525 if (con->state != CON_STATE_CONNECTING &&
2526 con->state != CON_STATE_NEGOTIATING &&
2527 con->state != CON_STATE_OPEN)
2532 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2535 if (con->state == CON_STATE_CONNECTING) {
2536 dout("try_read connecting\n");
2537 ret = read_partial_banner(con);
2540 ret = process_banner(con);
2544 con->state = CON_STATE_NEGOTIATING;
2547 * Received banner is good, exchange connection info.
2548 * Do not reset out_kvec, as sending our banner raced
2549 * with receiving peer banner after connect completed.
2551 ret = prepare_write_connect(con);
2554 prepare_read_connect(con);
2556 /* Send connection info before awaiting response */
2560 if (con->state == CON_STATE_NEGOTIATING) {
2561 dout("try_read negotiating\n");
2562 ret = read_partial_connect(con);
2565 ret = process_connect(con);
2571 WARN_ON(con->state != CON_STATE_OPEN);
2573 if (con->in_base_pos < 0) {
2575 * skipping + discarding content.
2577 * FIXME: there must be a better way to do this!
2579 static char buf[SKIP_BUF_SIZE];
2580 int skip = min((int) sizeof (buf), -con->in_base_pos);
2582 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2583 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2586 con->in_base_pos += ret;
2587 if (con->in_base_pos)
2590 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2594 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2597 dout("try_read got tag %d\n", (int)con->in_tag);
2598 switch (con->in_tag) {
2599 case CEPH_MSGR_TAG_MSG:
2600 prepare_read_message(con);
2602 case CEPH_MSGR_TAG_ACK:
2603 prepare_read_ack(con);
2605 case CEPH_MSGR_TAG_CLOSE:
2606 con_close_socket(con);
2607 con->state = CON_STATE_CLOSED;
2613 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2614 ret = read_partial_message(con);
2618 con->error_msg = "bad crc";
2622 con->error_msg = "io error";
2627 if (con->in_tag == CEPH_MSGR_TAG_READY)
2629 process_message(con);
2630 if (con->state == CON_STATE_OPEN)
2631 prepare_read_tag(con);
2634 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2635 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2637 * the final handshake seq exchange is semantically
2638 * equivalent to an ACK
2640 ret = read_partial_ack(con);
2648 dout("try_read done on %p ret %d\n", con, ret);
2652 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2653 con->error_msg = "protocol error, garbage tag";
2660 * Atomically queue work on a connection after the specified delay.
2661 * Bump @con reference to avoid races with connection teardown.
2662 * Returns 0 if work was queued, or an error code otherwise.
2664 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2666 if (!con->ops->get(con)) {
2667 dout("%s %p ref count 0\n", __func__, con);
2672 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2673 dout("%s %p - already queued\n", __func__, con);
2679 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 bool con_sock_closed(struct ceph_connection *con)
2691 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2695 case CON_STATE_ ## x: \
2696 con->error_msg = "socket closed (con state " #x ")"; \
2699 switch (con->state) {
2707 pr_warning("%s con %p unrecognized state %lu\n",
2708 __func__, con, con->state);
2709 con->error_msg = "unrecognized con state";
2718 static bool con_backoff(struct ceph_connection *con)
2722 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2725 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2727 dout("%s: con %p FAILED to back off %lu\n", __func__,
2729 BUG_ON(ret == -ENOENT);
2730 con_flag_set(con, CON_FLAG_BACKOFF);
2736 /* Finish fault handling; con->mutex must *not* be held here */
2738 static void con_fault_finish(struct ceph_connection *con)
2741 * in case we faulted due to authentication, invalidate our
2742 * current tickets so that we can get new ones.
2744 if (con->auth_retry && con->ops->invalidate_authorizer) {
2745 dout("calling invalidate_authorizer()\n");
2746 con->ops->invalidate_authorizer(con);
2749 if (con->ops->fault)
2750 con->ops->fault(con);
2754 * Do some work on a connection. Drop a connection ref when we're done.
2756 static void con_work(struct work_struct *work)
2758 struct ceph_connection *con = container_of(work, struct ceph_connection,
2762 mutex_lock(&con->mutex);
2766 if ((fault = con_sock_closed(con))) {
2767 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2770 if (con_backoff(con)) {
2771 dout("%s: con %p BACKOFF\n", __func__, con);
2774 if (con->state == CON_STATE_STANDBY) {
2775 dout("%s: con %p STANDBY\n", __func__, con);
2778 if (con->state == CON_STATE_CLOSED) {
2779 dout("%s: con %p CLOSED\n", __func__, con);
2783 if (con->state == CON_STATE_PREOPEN) {
2784 dout("%s: con %p PREOPEN\n", __func__, con);
2788 ret = try_read(con);
2792 con->error_msg = "socket error on read";
2797 ret = try_write(con);
2801 con->error_msg = "socket error on write";
2805 break; /* If we make it to here, we're done */
2809 mutex_unlock(&con->mutex);
2812 con_fault_finish(con);
2818 * Generic error/fault handler. A retry mechanism is used with
2819 * exponential backoff
2821 static void con_fault(struct ceph_connection *con)
2823 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2824 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2825 dout("fault %p state %lu to peer %s\n",
2826 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2828 WARN_ON(con->state != CON_STATE_CONNECTING &&
2829 con->state != CON_STATE_NEGOTIATING &&
2830 con->state != CON_STATE_OPEN);
2832 con_close_socket(con);
2834 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2835 dout("fault on LOSSYTX channel, marking CLOSED\n");
2836 con->state = CON_STATE_CLOSED;
2841 BUG_ON(con->in_msg->con != con);
2842 con->in_msg->con = NULL;
2843 ceph_msg_put(con->in_msg);
2848 /* Requeue anything that hasn't been acked */
2849 list_splice_init(&con->out_sent, &con->out_queue);
2851 /* If there are no messages queued or keepalive pending, place
2852 * the connection in a STANDBY state */
2853 if (list_empty(&con->out_queue) &&
2854 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
2855 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2856 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2857 con->state = CON_STATE_STANDBY;
2859 /* retry after a delay. */
2860 con->state = CON_STATE_PREOPEN;
2861 if (con->delay == 0)
2862 con->delay = BASE_DELAY_INTERVAL;
2863 else if (con->delay < MAX_DELAY_INTERVAL)
2865 con_flag_set(con, CON_FLAG_BACKOFF);
2873 * initialize a new messenger instance
2875 void ceph_messenger_init(struct ceph_messenger *msgr,
2876 struct ceph_entity_addr *myaddr,
2877 u64 supported_features,
2878 u64 required_features,
2881 msgr->supported_features = supported_features;
2882 msgr->required_features = required_features;
2884 spin_lock_init(&msgr->global_seq_lock);
2887 msgr->inst.addr = *myaddr;
2889 /* select a random nonce */
2890 msgr->inst.addr.type = 0;
2891 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2892 encode_my_addr(msgr);
2893 msgr->nocrc = nocrc;
2895 atomic_set(&msgr->stopping, 0);
2897 dout("%s %p\n", __func__, msgr);
2899 EXPORT_SYMBOL(ceph_messenger_init);
2901 static void clear_standby(struct ceph_connection *con)
2903 /* come back from STANDBY? */
2904 if (con->state == CON_STATE_STANDBY) {
2905 dout("clear_standby %p and ++connect_seq\n", con);
2906 con->state = CON_STATE_PREOPEN;
2908 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2909 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
2914 * Queue up an outgoing message on the given connection.
2916 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2919 msg->hdr.src = con->msgr->inst.name;
2920 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2921 msg->needs_out_seq = true;
2923 mutex_lock(&con->mutex);
2925 if (con->state == CON_STATE_CLOSED) {
2926 dout("con_send %p closed, dropping %p\n", con, msg);
2928 mutex_unlock(&con->mutex);
2932 BUG_ON(msg->con != NULL);
2933 msg->con = con->ops->get(con);
2934 BUG_ON(msg->con == NULL);
2936 BUG_ON(!list_empty(&msg->list_head));
2937 list_add_tail(&msg->list_head, &con->out_queue);
2938 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2939 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2940 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2941 le32_to_cpu(msg->hdr.front_len),
2942 le32_to_cpu(msg->hdr.middle_len),
2943 le32_to_cpu(msg->hdr.data_len));
2946 mutex_unlock(&con->mutex);
2948 /* if there wasn't anything waiting to send before, queue
2950 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
2953 EXPORT_SYMBOL(ceph_con_send);
2956 * Revoke a message that was previously queued for send
2958 void ceph_msg_revoke(struct ceph_msg *msg)
2960 struct ceph_connection *con = msg->con;
2963 return; /* Message not in our possession */
2965 mutex_lock(&con->mutex);
2966 if (!list_empty(&msg->list_head)) {
2967 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
2968 list_del_init(&msg->list_head);
2969 BUG_ON(msg->con == NULL);
2970 msg->con->ops->put(msg->con);
2976 if (con->out_msg == msg) {
2977 dout("%s %p msg %p - was sending\n", __func__, con, msg);
2978 con->out_msg = NULL;
2979 if (con->out_kvec_is_msg) {
2980 con->out_skip = con->out_kvec_bytes;
2981 con->out_kvec_is_msg = false;
2987 mutex_unlock(&con->mutex);
2991 * Revoke a message that we may be reading data into
2993 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
2995 struct ceph_connection *con;
2997 BUG_ON(msg == NULL);
2999 dout("%s msg %p null con\n", __func__, msg);
3001 return; /* Message not in our possession */
3005 mutex_lock(&con->mutex);
3006 if (con->in_msg == msg) {
3007 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3008 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3009 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
3011 /* skip rest of message */
3012 dout("%s %p msg %p revoked\n", __func__, con, msg);
3013 con->in_base_pos = con->in_base_pos -
3014 sizeof(struct ceph_msg_header) -
3018 sizeof(struct ceph_msg_footer);
3019 ceph_msg_put(con->in_msg);
3021 con->in_tag = CEPH_MSGR_TAG_READY;
3024 dout("%s %p in_msg %p msg %p no-op\n",
3025 __func__, con, con->in_msg, msg);
3027 mutex_unlock(&con->mutex);
3031 * Queue a keepalive byte to ensure the tcp connection is alive.
3033 void ceph_con_keepalive(struct ceph_connection *con)
3035 dout("con_keepalive %p\n", con);
3036 mutex_lock(&con->mutex);
3038 mutex_unlock(&con->mutex);
3039 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3040 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3043 EXPORT_SYMBOL(ceph_con_keepalive);
3045 static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
3047 struct ceph_msg_data *data;
3049 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3052 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
3055 INIT_LIST_HEAD(&data->links);
3060 static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3065 WARN_ON(!list_empty(&data->links));
3066 if (data->type == CEPH_MSG_DATA_PAGELIST) {
3067 ceph_pagelist_release(data->pagelist);
3068 kfree(data->pagelist);
3070 kmem_cache_free(ceph_msg_data_cache, data);
3073 void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
3074 size_t length, size_t alignment)
3076 struct ceph_msg_data *data;
3081 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3083 data->pages = pages;
3084 data->length = length;
3085 data->alignment = alignment & ~PAGE_MASK;
3087 list_add_tail(&data->links, &msg->data);
3088 msg->data_length += length;
3090 EXPORT_SYMBOL(ceph_msg_data_add_pages);
3092 void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
3093 struct ceph_pagelist *pagelist)
3095 struct ceph_msg_data *data;
3098 BUG_ON(!pagelist->length);
3100 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3102 data->pagelist = pagelist;
3104 list_add_tail(&data->links, &msg->data);
3105 msg->data_length += pagelist->length;
3107 EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
3110 void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
3113 struct ceph_msg_data *data;
3117 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3120 data->bio_length = length;
3122 list_add_tail(&data->links, &msg->data);
3123 msg->data_length += length;
3125 EXPORT_SYMBOL(ceph_msg_data_add_bio);
3126 #endif /* CONFIG_BLOCK */
3129 * construct a new message with given type, size
3130 * the new msg has a ref count of 1.
3132 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3137 m = kmem_cache_zalloc(ceph_msg_cache, flags);
3141 m->hdr.type = cpu_to_le16(type);
3142 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3143 m->hdr.front_len = cpu_to_le32(front_len);
3145 INIT_LIST_HEAD(&m->list_head);
3146 kref_init(&m->kref);
3147 INIT_LIST_HEAD(&m->data);
3151 m->front.iov_base = ceph_kvmalloc(front_len, flags);
3152 if (m->front.iov_base == NULL) {
3153 dout("ceph_msg_new can't allocate %d bytes\n",
3158 m->front.iov_base = NULL;
3160 m->front_alloc_len = m->front.iov_len = front_len;
3162 dout("ceph_msg_new %p front %d\n", m, front_len);
3169 pr_err("msg_new can't create type %d front %d\n", type,
3173 dout("msg_new can't create type %d front %d\n", type,
3178 EXPORT_SYMBOL(ceph_msg_new);
3181 * Allocate "middle" portion of a message, if it is needed and wasn't
3182 * allocated by alloc_msg. This allows us to read a small fixed-size
3183 * per-type header in the front and then gracefully fail (i.e.,
3184 * propagate the error to the caller based on info in the front) when
3185 * the middle is too large.
3187 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3189 int type = le16_to_cpu(msg->hdr.type);
3190 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3192 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3193 ceph_msg_type_name(type), middle_len);
3194 BUG_ON(!middle_len);
3195 BUG_ON(msg->middle);
3197 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3204 * Allocate a message for receiving an incoming message on a
3205 * connection, and save the result in con->in_msg. Uses the
3206 * connection's private alloc_msg op if available.
3208 * Returns 0 on success, or a negative error code.
3210 * On success, if we set *skip = 1:
3211 * - the next message should be skipped and ignored.
3212 * - con->in_msg == NULL
3213 * or if we set *skip = 0:
3214 * - con->in_msg is non-null.
3215 * On error (ENOMEM, EAGAIN, ...),
3216 * - con->in_msg == NULL
3218 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3220 struct ceph_msg_header *hdr = &con->in_hdr;
3221 int middle_len = le32_to_cpu(hdr->middle_len);
3222 struct ceph_msg *msg;
3225 BUG_ON(con->in_msg != NULL);
3226 BUG_ON(!con->ops->alloc_msg);
3228 mutex_unlock(&con->mutex);
3229 msg = con->ops->alloc_msg(con, hdr, skip);
3230 mutex_lock(&con->mutex);
3231 if (con->state != CON_STATE_OPEN) {
3239 con->in_msg->con = con->ops->get(con);
3240 BUG_ON(con->in_msg->con == NULL);
3243 * Null message pointer means either we should skip
3244 * this message or we couldn't allocate memory. The
3245 * former is not an error.
3249 con->error_msg = "error allocating memory for incoming message";
3253 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3255 if (middle_len && !con->in_msg->middle) {
3256 ret = ceph_alloc_middle(con, con->in_msg);
3258 ceph_msg_put(con->in_msg);
3268 * Free a generically kmalloc'd message.
3270 void ceph_msg_kfree(struct ceph_msg *m)
3272 dout("msg_kfree %p\n", m);
3273 ceph_kvfree(m->front.iov_base);
3274 kmem_cache_free(ceph_msg_cache, m);
3278 * Drop a msg ref. Destroy as needed.
3280 void ceph_msg_last_put(struct kref *kref)
3282 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3284 struct list_head *links;
3285 struct list_head *next;
3287 dout("ceph_msg_put last one on %p\n", m);
3288 WARN_ON(!list_empty(&m->list_head));
3290 /* drop middle, data, if any */
3292 ceph_buffer_put(m->middle);
3296 list_splice_init(&m->data, &data);
3297 list_for_each_safe(links, next, &data) {
3298 struct ceph_msg_data *data;
3300 data = list_entry(links, struct ceph_msg_data, links);
3301 list_del_init(links);
3302 ceph_msg_data_destroy(data);
3307 ceph_msgpool_put(m->pool, m);
3311 EXPORT_SYMBOL(ceph_msg_last_put);
3313 void ceph_msg_dump(struct ceph_msg *msg)
3315 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3316 msg->front_alloc_len, msg->data_length);
3317 print_hex_dump(KERN_DEBUG, "header: ",
3318 DUMP_PREFIX_OFFSET, 16, 1,
3319 &msg->hdr, sizeof(msg->hdr), true);
3320 print_hex_dump(KERN_DEBUG, " front: ",
3321 DUMP_PREFIX_OFFSET, 16, 1,
3322 msg->front.iov_base, msg->front.iov_len, true);
3324 print_hex_dump(KERN_DEBUG, "middle: ",
3325 DUMP_PREFIX_OFFSET, 16, 1,
3326 msg->middle->vec.iov_base,
3327 msg->middle->vec.iov_len, true);
3328 print_hex_dump(KERN_DEBUG, "footer: ",
3329 DUMP_PREFIX_OFFSET, 16, 1,
3330 &msg->footer, sizeof(msg->footer), true);
3332 EXPORT_SYMBOL(ceph_msg_dump);