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>
12 #include <linux/bio.h>
13 #include <linux/blkdev.h>
14 #include <linux/dns_resolver.h>
17 #include <linux/ceph/libceph.h>
18 #include <linux/ceph/messenger.h>
19 #include <linux/ceph/decode.h>
20 #include <linux/ceph/pagelist.h>
21 #include <linux/export.h>
24 * Ceph uses the messenger to exchange ceph_msg messages with other
25 * hosts in the system. The messenger provides ordered and reliable
26 * delivery. We tolerate TCP disconnects by reconnecting (with
27 * exponential backoff) in the case of a fault (disconnection, bad
28 * crc, protocol error). Acks allow sent messages to be discarded by
33 * We track the state of the socket on a given connection using
34 * values defined below. The transition to a new socket state is
35 * handled by a function which verifies we aren't coming from an
39 * | NEW* | transient initial state
41 * | con_sock_state_init()
44 * | CLOSED | initialized, but no socket (and no
45 * ---------- TCP connection)
47 * | \ con_sock_state_connecting()
48 * | ----------------------
50 * + con_sock_state_closed() \
51 * |+--------------------------- \
54 * | | CLOSING | socket event; \ \
55 * | ----------- await close \ \
58 * | + con_sock_state_closing() \ |
60 * | / --------------- | |
63 * | / -----------------| CONNECTING | socket created, TCP
64 * | | / -------------- connect initiated
65 * | | | con_sock_state_connected()
68 * | CONNECTED | TCP connection established
71 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
74 #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
75 #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
76 #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
77 #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
78 #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
83 #define CON_STATE_CLOSED 1 /* -> PREOPEN */
84 #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
85 #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
86 #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
87 #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
88 #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
91 * ceph_connection flag bits
93 #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
94 * messages on errors */
95 #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
96 #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
97 #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
98 #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
100 /* static tag bytes (protocol control messages) */
101 static char tag_msg = CEPH_MSGR_TAG_MSG;
102 static char tag_ack = CEPH_MSGR_TAG_ACK;
103 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
105 #ifdef CONFIG_LOCKDEP
106 static struct lock_class_key socket_class;
110 * When skipping (ignoring) a block of input we read it into a "skip
111 * buffer," which is this many bytes in size.
113 #define SKIP_BUF_SIZE 1024
115 static void queue_con(struct ceph_connection *con);
116 static void con_work(struct work_struct *);
117 static void ceph_fault(struct ceph_connection *con);
120 * Nicely render a sockaddr as a string. An array of formatted
121 * strings is used, to approximate reentrancy.
123 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
124 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
125 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
126 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
128 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
129 static atomic_t addr_str_seq = ATOMIC_INIT(0);
131 static struct page *zero_page; /* used in certain error cases */
133 const char *ceph_pr_addr(const struct sockaddr_storage *ss)
137 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
138 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
140 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
143 switch (ss->ss_family) {
145 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
146 ntohs(in4->sin_port));
150 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
151 ntohs(in6->sin6_port));
155 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
161 EXPORT_SYMBOL(ceph_pr_addr);
163 static void encode_my_addr(struct ceph_messenger *msgr)
165 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
166 ceph_encode_addr(&msgr->my_enc_addr);
170 * work queue for all reading and writing to/from the socket.
172 static struct workqueue_struct *ceph_msgr_wq;
174 void _ceph_msgr_exit(void)
177 destroy_workqueue(ceph_msgr_wq);
181 BUG_ON(zero_page == NULL);
183 page_cache_release(zero_page);
187 int ceph_msgr_init(void)
189 BUG_ON(zero_page != NULL);
190 zero_page = ZERO_PAGE(0);
191 page_cache_get(zero_page);
193 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0);
197 pr_err("msgr_init failed to create workqueue\n");
202 EXPORT_SYMBOL(ceph_msgr_init);
204 void ceph_msgr_exit(void)
206 BUG_ON(ceph_msgr_wq == NULL);
210 EXPORT_SYMBOL(ceph_msgr_exit);
212 void ceph_msgr_flush(void)
214 flush_workqueue(ceph_msgr_wq);
216 EXPORT_SYMBOL(ceph_msgr_flush);
218 /* Connection socket state transition functions */
220 static void con_sock_state_init(struct ceph_connection *con)
224 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
225 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
226 printk("%s: unexpected old state %d\n", __func__, old_state);
227 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
228 CON_SOCK_STATE_CLOSED);
231 static void con_sock_state_connecting(struct ceph_connection *con)
235 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
236 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
237 printk("%s: unexpected old state %d\n", __func__, old_state);
238 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
239 CON_SOCK_STATE_CONNECTING);
242 static void con_sock_state_connected(struct ceph_connection *con)
246 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
247 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
248 printk("%s: unexpected old state %d\n", __func__, old_state);
249 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
250 CON_SOCK_STATE_CONNECTED);
253 static void con_sock_state_closing(struct ceph_connection *con)
257 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
258 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
259 old_state != CON_SOCK_STATE_CONNECTED &&
260 old_state != CON_SOCK_STATE_CLOSING))
261 printk("%s: unexpected old state %d\n", __func__, old_state);
262 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
263 CON_SOCK_STATE_CLOSING);
266 static void con_sock_state_closed(struct ceph_connection *con)
270 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
271 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
272 old_state != CON_SOCK_STATE_CLOSING &&
273 old_state != CON_SOCK_STATE_CONNECTING &&
274 old_state != CON_SOCK_STATE_CLOSED))
275 printk("%s: unexpected old state %d\n", __func__, old_state);
276 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
277 CON_SOCK_STATE_CLOSED);
281 * socket callback functions
284 /* data available on socket, or listen socket received a connect */
285 static void ceph_sock_data_ready(struct sock *sk, int count_unused)
287 struct ceph_connection *con = sk->sk_user_data;
288 if (atomic_read(&con->msgr->stopping)) {
292 if (sk->sk_state != TCP_CLOSE_WAIT) {
293 dout("%s on %p state = %lu, queueing work\n", __func__,
299 /* socket has buffer space for writing */
300 static void ceph_sock_write_space(struct sock *sk)
302 struct ceph_connection *con = sk->sk_user_data;
304 /* only queue to workqueue if there is data we want to write,
305 * and there is sufficient space in the socket buffer to accept
306 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
307 * doesn't get called again until try_write() fills the socket
308 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
309 * and net/core/stream.c:sk_stream_write_space().
311 if (test_bit(CON_FLAG_WRITE_PENDING, &con->flags)) {
312 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
313 dout("%s %p queueing write work\n", __func__, con);
314 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
318 dout("%s %p nothing to write\n", __func__, con);
322 /* socket's state has changed */
323 static void ceph_sock_state_change(struct sock *sk)
325 struct ceph_connection *con = sk->sk_user_data;
327 dout("%s %p state = %lu sk_state = %u\n", __func__,
328 con, con->state, sk->sk_state);
330 switch (sk->sk_state) {
332 dout("%s TCP_CLOSE\n", __func__);
334 dout("%s TCP_CLOSE_WAIT\n", __func__);
335 con_sock_state_closing(con);
336 set_bit(CON_FLAG_SOCK_CLOSED, &con->flags);
339 case TCP_ESTABLISHED:
340 dout("%s TCP_ESTABLISHED\n", __func__);
341 con_sock_state_connected(con);
344 default: /* Everything else is uninteresting */
350 * set up socket callbacks
352 static void set_sock_callbacks(struct socket *sock,
353 struct ceph_connection *con)
355 struct sock *sk = sock->sk;
356 sk->sk_user_data = con;
357 sk->sk_data_ready = ceph_sock_data_ready;
358 sk->sk_write_space = ceph_sock_write_space;
359 sk->sk_state_change = ceph_sock_state_change;
368 * initiate connection to a remote socket.
370 static int ceph_tcp_connect(struct ceph_connection *con)
372 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
377 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
381 sock->sk->sk_allocation = GFP_NOFS;
383 #ifdef CONFIG_LOCKDEP
384 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
387 set_sock_callbacks(sock, con);
389 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
391 con_sock_state_connecting(con);
392 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
394 if (ret == -EINPROGRESS) {
395 dout("connect %s EINPROGRESS sk_state = %u\n",
396 ceph_pr_addr(&con->peer_addr.in_addr),
398 } else if (ret < 0) {
399 pr_err("connect %s error %d\n",
400 ceph_pr_addr(&con->peer_addr.in_addr), ret);
402 con->error_msg = "connect error";
410 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
412 struct kvec iov = {buf, len};
413 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
416 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
423 * write something. @more is true if caller will be sending more data
426 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
427 size_t kvlen, size_t len, int more)
429 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
433 msg.msg_flags |= MSG_MORE;
435 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
437 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
443 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
444 int offset, size_t size, int more)
446 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
449 ret = kernel_sendpage(sock, page, offset, size, flags);
458 * Shutdown/close the socket for the given connection.
460 static int con_close_socket(struct ceph_connection *con)
464 dout("con_close_socket on %p sock %p\n", con, con->sock);
466 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
467 sock_release(con->sock);
472 * Forcibly clear the SOCK_CLOSED flag. It gets set
473 * independent of the connection mutex, and we could have
474 * received a socket close event before we had the chance to
475 * shut the socket down.
477 clear_bit(CON_FLAG_SOCK_CLOSED, &con->flags);
479 con_sock_state_closed(con);
484 * Reset a connection. Discard all incoming and outgoing messages
485 * and clear *_seq state.
487 static void ceph_msg_remove(struct ceph_msg *msg)
489 list_del_init(&msg->list_head);
490 BUG_ON(msg->con == NULL);
491 msg->con->ops->put(msg->con);
496 static void ceph_msg_remove_list(struct list_head *head)
498 while (!list_empty(head)) {
499 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
501 ceph_msg_remove(msg);
505 static void reset_connection(struct ceph_connection *con)
507 /* reset connection, out_queue, msg_ and connect_seq */
508 /* discard existing out_queue and msg_seq */
509 dout("reset_connection %p\n", con);
510 ceph_msg_remove_list(&con->out_queue);
511 ceph_msg_remove_list(&con->out_sent);
514 BUG_ON(con->in_msg->con != con);
515 con->in_msg->con = NULL;
516 ceph_msg_put(con->in_msg);
521 con->connect_seq = 0;
524 ceph_msg_put(con->out_msg);
528 con->in_seq_acked = 0;
532 * mark a peer down. drop any open connections.
534 void ceph_con_close(struct ceph_connection *con)
536 mutex_lock(&con->mutex);
537 dout("con_close %p peer %s\n", con,
538 ceph_pr_addr(&con->peer_addr.in_addr));
539 con->state = CON_STATE_CLOSED;
541 clear_bit(CON_FLAG_LOSSYTX, &con->flags); /* so we retry next connect */
542 clear_bit(CON_FLAG_KEEPALIVE_PENDING, &con->flags);
543 clear_bit(CON_FLAG_WRITE_PENDING, &con->flags);
544 clear_bit(CON_FLAG_KEEPALIVE_PENDING, &con->flags);
545 clear_bit(CON_FLAG_BACKOFF, &con->flags);
547 reset_connection(con);
548 con->peer_global_seq = 0;
549 cancel_delayed_work(&con->work);
550 con_close_socket(con);
551 mutex_unlock(&con->mutex);
553 EXPORT_SYMBOL(ceph_con_close);
556 * Reopen a closed connection, with a new peer address.
558 void ceph_con_open(struct ceph_connection *con,
559 __u8 entity_type, __u64 entity_num,
560 struct ceph_entity_addr *addr)
562 mutex_lock(&con->mutex);
563 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
565 WARN_ON(con->state != CON_STATE_CLOSED);
566 con->state = CON_STATE_PREOPEN;
568 con->peer_name.type = (__u8) entity_type;
569 con->peer_name.num = cpu_to_le64(entity_num);
571 memcpy(&con->peer_addr, addr, sizeof(*addr));
572 con->delay = 0; /* reset backoff memory */
573 mutex_unlock(&con->mutex);
576 EXPORT_SYMBOL(ceph_con_open);
579 * return true if this connection ever successfully opened
581 bool ceph_con_opened(struct ceph_connection *con)
583 return con->connect_seq > 0;
587 * initialize a new connection.
589 void ceph_con_init(struct ceph_connection *con, void *private,
590 const struct ceph_connection_operations *ops,
591 struct ceph_messenger *msgr)
593 dout("con_init %p\n", con);
594 memset(con, 0, sizeof(*con));
595 con->private = private;
599 con_sock_state_init(con);
601 mutex_init(&con->mutex);
602 INIT_LIST_HEAD(&con->out_queue);
603 INIT_LIST_HEAD(&con->out_sent);
604 INIT_DELAYED_WORK(&con->work, con_work);
606 con->state = CON_STATE_CLOSED;
608 EXPORT_SYMBOL(ceph_con_init);
612 * We maintain a global counter to order connection attempts. Get
613 * a unique seq greater than @gt.
615 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
619 spin_lock(&msgr->global_seq_lock);
620 if (msgr->global_seq < gt)
621 msgr->global_seq = gt;
622 ret = ++msgr->global_seq;
623 spin_unlock(&msgr->global_seq_lock);
627 static void con_out_kvec_reset(struct ceph_connection *con)
629 con->out_kvec_left = 0;
630 con->out_kvec_bytes = 0;
631 con->out_kvec_cur = &con->out_kvec[0];
634 static void con_out_kvec_add(struct ceph_connection *con,
635 size_t size, void *data)
639 index = con->out_kvec_left;
640 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
642 con->out_kvec[index].iov_len = size;
643 con->out_kvec[index].iov_base = data;
644 con->out_kvec_left++;
645 con->out_kvec_bytes += size;
649 static void init_bio_iter(struct bio *bio, struct bio **iter, int *seg)
660 static void iter_bio_next(struct bio **bio_iter, int *seg)
662 if (*bio_iter == NULL)
665 BUG_ON(*seg >= (*bio_iter)->bi_vcnt);
668 if (*seg == (*bio_iter)->bi_vcnt)
669 init_bio_iter((*bio_iter)->bi_next, bio_iter, seg);
673 static void prepare_write_message_data(struct ceph_connection *con)
675 struct ceph_msg *msg = con->out_msg;
678 BUG_ON(!msg->hdr.data_len);
680 /* initialize page iterator */
681 con->out_msg_pos.page = 0;
683 con->out_msg_pos.page_pos = msg->page_alignment;
685 con->out_msg_pos.page_pos = 0;
688 init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg);
690 con->out_msg_pos.data_pos = 0;
691 con->out_msg_pos.did_page_crc = false;
692 con->out_more = 1; /* data + footer will follow */
696 * Prepare footer for currently outgoing message, and finish things
697 * off. Assumes out_kvec* are already valid.. we just add on to the end.
699 static void prepare_write_message_footer(struct ceph_connection *con)
701 struct ceph_msg *m = con->out_msg;
702 int v = con->out_kvec_left;
704 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
706 dout("prepare_write_message_footer %p\n", con);
707 con->out_kvec_is_msg = true;
708 con->out_kvec[v].iov_base = &m->footer;
709 con->out_kvec[v].iov_len = sizeof(m->footer);
710 con->out_kvec_bytes += sizeof(m->footer);
711 con->out_kvec_left++;
712 con->out_more = m->more_to_follow;
713 con->out_msg_done = true;
717 * Prepare headers for the next outgoing message.
719 static void prepare_write_message(struct ceph_connection *con)
724 con_out_kvec_reset(con);
725 con->out_kvec_is_msg = true;
726 con->out_msg_done = false;
728 /* Sneak an ack in there first? If we can get it into the same
729 * TCP packet that's a good thing. */
730 if (con->in_seq > con->in_seq_acked) {
731 con->in_seq_acked = con->in_seq;
732 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
733 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
734 con_out_kvec_add(con, sizeof (con->out_temp_ack),
738 BUG_ON(list_empty(&con->out_queue));
739 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
741 BUG_ON(m->con != con);
743 /* put message on sent list */
745 list_move_tail(&m->list_head, &con->out_sent);
748 * only assign outgoing seq # if we haven't sent this message
749 * yet. if it is requeued, resend with it's original seq.
751 if (m->needs_out_seq) {
752 m->hdr.seq = cpu_to_le64(++con->out_seq);
753 m->needs_out_seq = false;
760 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n",
761 m, con->out_seq, le16_to_cpu(m->hdr.type),
762 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
763 le32_to_cpu(m->hdr.data_len),
765 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
767 /* tag + hdr + front + middle */
768 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
769 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
770 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
773 con_out_kvec_add(con, m->middle->vec.iov_len,
774 m->middle->vec.iov_base);
776 /* fill in crc (except data pages), footer */
777 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
778 con->out_msg->hdr.crc = cpu_to_le32(crc);
779 con->out_msg->footer.flags = 0;
781 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
782 con->out_msg->footer.front_crc = cpu_to_le32(crc);
784 crc = crc32c(0, m->middle->vec.iov_base,
785 m->middle->vec.iov_len);
786 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
788 con->out_msg->footer.middle_crc = 0;
789 dout("%s front_crc %u middle_crc %u\n", __func__,
790 le32_to_cpu(con->out_msg->footer.front_crc),
791 le32_to_cpu(con->out_msg->footer.middle_crc));
793 /* is there a data payload? */
794 con->out_msg->footer.data_crc = 0;
796 prepare_write_message_data(con);
798 /* no, queue up footer too and be done */
799 prepare_write_message_footer(con);
801 set_bit(CON_FLAG_WRITE_PENDING, &con->flags);
807 static void prepare_write_ack(struct ceph_connection *con)
809 dout("prepare_write_ack %p %llu -> %llu\n", con,
810 con->in_seq_acked, con->in_seq);
811 con->in_seq_acked = con->in_seq;
813 con_out_kvec_reset(con);
815 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
817 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
818 con_out_kvec_add(con, sizeof (con->out_temp_ack),
821 con->out_more = 1; /* more will follow.. eventually.. */
822 set_bit(CON_FLAG_WRITE_PENDING, &con->flags);
826 * Prepare to write keepalive byte.
828 static void prepare_write_keepalive(struct ceph_connection *con)
830 dout("prepare_write_keepalive %p\n", con);
831 con_out_kvec_reset(con);
832 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
833 set_bit(CON_FLAG_WRITE_PENDING, &con->flags);
837 * Connection negotiation.
840 static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
843 struct ceph_auth_handshake *auth;
845 if (!con->ops->get_authorizer) {
846 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
847 con->out_connect.authorizer_len = 0;
851 /* Can't hold the mutex while getting authorizer */
852 mutex_unlock(&con->mutex);
853 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
854 mutex_lock(&con->mutex);
858 if (con->state != CON_STATE_NEGOTIATING)
859 return ERR_PTR(-EAGAIN);
861 con->auth_reply_buf = auth->authorizer_reply_buf;
862 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
867 * We connected to a peer and are saying hello.
869 static void prepare_write_banner(struct ceph_connection *con)
871 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
872 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
873 &con->msgr->my_enc_addr);
876 set_bit(CON_FLAG_WRITE_PENDING, &con->flags);
879 static int prepare_write_connect(struct ceph_connection *con)
881 unsigned int global_seq = get_global_seq(con->msgr, 0);
884 struct ceph_auth_handshake *auth;
886 switch (con->peer_name.type) {
887 case CEPH_ENTITY_TYPE_MON:
888 proto = CEPH_MONC_PROTOCOL;
890 case CEPH_ENTITY_TYPE_OSD:
891 proto = CEPH_OSDC_PROTOCOL;
893 case CEPH_ENTITY_TYPE_MDS:
894 proto = CEPH_MDSC_PROTOCOL;
900 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
901 con->connect_seq, global_seq, proto);
903 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
904 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
905 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
906 con->out_connect.global_seq = cpu_to_le32(global_seq);
907 con->out_connect.protocol_version = cpu_to_le32(proto);
908 con->out_connect.flags = 0;
910 auth_proto = CEPH_AUTH_UNKNOWN;
911 auth = get_connect_authorizer(con, &auth_proto);
913 return PTR_ERR(auth);
915 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
916 con->out_connect.authorizer_len = auth ?
917 cpu_to_le32(auth->authorizer_buf_len) : 0;
919 con_out_kvec_add(con, sizeof (con->out_connect),
921 if (auth && auth->authorizer_buf_len)
922 con_out_kvec_add(con, auth->authorizer_buf_len,
923 auth->authorizer_buf);
926 set_bit(CON_FLAG_WRITE_PENDING, &con->flags);
932 * write as much of pending kvecs to the socket as we can.
934 * 0 -> socket full, but more to do
937 static int write_partial_kvec(struct ceph_connection *con)
941 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
942 while (con->out_kvec_bytes > 0) {
943 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
944 con->out_kvec_left, con->out_kvec_bytes,
948 con->out_kvec_bytes -= ret;
949 if (con->out_kvec_bytes == 0)
952 /* account for full iov entries consumed */
953 while (ret >= con->out_kvec_cur->iov_len) {
954 BUG_ON(!con->out_kvec_left);
955 ret -= con->out_kvec_cur->iov_len;
957 con->out_kvec_left--;
959 /* and for a partially-consumed entry */
961 con->out_kvec_cur->iov_len -= ret;
962 con->out_kvec_cur->iov_base += ret;
965 con->out_kvec_left = 0;
966 con->out_kvec_is_msg = false;
969 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
970 con->out_kvec_bytes, con->out_kvec_left, ret);
971 return ret; /* done! */
974 static void out_msg_pos_next(struct ceph_connection *con, struct page *page,
975 size_t len, size_t sent, bool in_trail)
977 struct ceph_msg *msg = con->out_msg;
982 con->out_msg_pos.data_pos += sent;
983 con->out_msg_pos.page_pos += sent;
988 con->out_msg_pos.page_pos = 0;
989 con->out_msg_pos.page++;
990 con->out_msg_pos.did_page_crc = false;
992 list_move_tail(&page->lru,
994 else if (msg->pagelist)
995 list_move_tail(&page->lru,
996 &msg->pagelist->head);
999 iter_bio_next(&msg->bio_iter, &msg->bio_seg);
1004 * Write as much message data payload as we can. If we finish, queue
1006 * 1 -> done, footer is now queued in out_kvec[].
1007 * 0 -> socket full, but more to do
1010 static int write_partial_msg_pages(struct ceph_connection *con)
1012 struct ceph_msg *msg = con->out_msg;
1013 unsigned int data_len = le32_to_cpu(msg->hdr.data_len);
1015 bool do_datacrc = !con->msgr->nocrc;
1017 int total_max_write;
1018 bool in_trail = false;
1019 const size_t trail_len = (msg->trail ? msg->trail->length : 0);
1020 const size_t trail_off = data_len - trail_len;
1022 dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
1023 con, msg, con->out_msg_pos.page, msg->nr_pages,
1024 con->out_msg_pos.page_pos);
1027 * Iterate through each page that contains data to be
1028 * written, and send as much as possible for each.
1030 * If we are calculating the data crc (the default), we will
1031 * need to map the page. If we have no pages, they have
1032 * been revoked, so use the zero page.
1034 while (data_len > con->out_msg_pos.data_pos) {
1035 struct page *page = NULL;
1036 int max_write = PAGE_SIZE;
1039 in_trail = in_trail || con->out_msg_pos.data_pos >= trail_off;
1041 total_max_write = trail_off - con->out_msg_pos.data_pos;
1044 total_max_write = data_len - con->out_msg_pos.data_pos;
1046 page = list_first_entry(&msg->trail->head,
1048 } else if (msg->pages) {
1049 page = msg->pages[con->out_msg_pos.page];
1050 } else if (msg->pagelist) {
1051 page = list_first_entry(&msg->pagelist->head,
1054 } else if (msg->bio) {
1057 bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg);
1059 bio_offset = bv->bv_offset;
1060 max_write = bv->bv_len;
1065 len = min_t(int, max_write - con->out_msg_pos.page_pos,
1068 if (do_datacrc && !con->out_msg_pos.did_page_crc) {
1070 u32 crc = le32_to_cpu(msg->footer.data_crc);
1074 BUG_ON(kaddr == NULL);
1075 base = kaddr + con->out_msg_pos.page_pos + bio_offset;
1076 crc = crc32c(crc, base, len);
1078 msg->footer.data_crc = cpu_to_le32(crc);
1079 con->out_msg_pos.did_page_crc = true;
1081 ret = ceph_tcp_sendpage(con->sock, page,
1082 con->out_msg_pos.page_pos + bio_offset,
1087 out_msg_pos_next(con, page, len, (size_t) ret, in_trail);
1090 dout("write_partial_msg_pages %p msg %p done\n", con, msg);
1092 /* prepare and queue up footer, too */
1094 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1095 con_out_kvec_reset(con);
1096 prepare_write_message_footer(con);
1105 static int write_partial_skip(struct ceph_connection *con)
1109 while (con->out_skip > 0) {
1110 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
1112 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, 1);
1115 con->out_skip -= ret;
1123 * Prepare to read connection handshake, or an ack.
1125 static void prepare_read_banner(struct ceph_connection *con)
1127 dout("prepare_read_banner %p\n", con);
1128 con->in_base_pos = 0;
1131 static void prepare_read_connect(struct ceph_connection *con)
1133 dout("prepare_read_connect %p\n", con);
1134 con->in_base_pos = 0;
1137 static void prepare_read_ack(struct ceph_connection *con)
1139 dout("prepare_read_ack %p\n", con);
1140 con->in_base_pos = 0;
1143 static void prepare_read_tag(struct ceph_connection *con)
1145 dout("prepare_read_tag %p\n", con);
1146 con->in_base_pos = 0;
1147 con->in_tag = CEPH_MSGR_TAG_READY;
1151 * Prepare to read a message.
1153 static int prepare_read_message(struct ceph_connection *con)
1155 dout("prepare_read_message %p\n", con);
1156 BUG_ON(con->in_msg != NULL);
1157 con->in_base_pos = 0;
1158 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1163 static int read_partial(struct ceph_connection *con,
1164 int end, int size, void *object)
1166 while (con->in_base_pos < end) {
1167 int left = end - con->in_base_pos;
1168 int have = size - left;
1169 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1172 con->in_base_pos += ret;
1179 * Read all or part of the connect-side handshake on a new connection
1181 static int read_partial_banner(struct ceph_connection *con)
1187 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1190 size = strlen(CEPH_BANNER);
1192 ret = read_partial(con, end, size, con->in_banner);
1196 size = sizeof (con->actual_peer_addr);
1198 ret = read_partial(con, end, size, &con->actual_peer_addr);
1202 size = sizeof (con->peer_addr_for_me);
1204 ret = read_partial(con, end, size, &con->peer_addr_for_me);
1212 static int read_partial_connect(struct ceph_connection *con)
1218 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1220 size = sizeof (con->in_reply);
1222 ret = read_partial(con, end, size, &con->in_reply);
1226 size = le32_to_cpu(con->in_reply.authorizer_len);
1228 ret = read_partial(con, end, size, con->auth_reply_buf);
1232 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1233 con, (int)con->in_reply.tag,
1234 le32_to_cpu(con->in_reply.connect_seq),
1235 le32_to_cpu(con->in_reply.global_seq));
1242 * Verify the hello banner looks okay.
1244 static int verify_hello(struct ceph_connection *con)
1246 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1247 pr_err("connect to %s got bad banner\n",
1248 ceph_pr_addr(&con->peer_addr.in_addr));
1249 con->error_msg = "protocol error, bad banner";
1255 static bool addr_is_blank(struct sockaddr_storage *ss)
1257 switch (ss->ss_family) {
1259 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1262 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1263 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1264 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1265 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1270 static int addr_port(struct sockaddr_storage *ss)
1272 switch (ss->ss_family) {
1274 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1276 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1281 static void addr_set_port(struct sockaddr_storage *ss, int p)
1283 switch (ss->ss_family) {
1285 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1288 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1294 * Unlike other *_pton function semantics, zero indicates success.
1296 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1297 char delim, const char **ipend)
1299 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1300 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1302 memset(ss, 0, sizeof(*ss));
1304 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1305 ss->ss_family = AF_INET;
1309 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1310 ss->ss_family = AF_INET6;
1318 * Extract hostname string and resolve using kernel DNS facility.
1320 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1321 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1322 struct sockaddr_storage *ss, char delim, const char **ipend)
1324 const char *end, *delim_p;
1325 char *colon_p, *ip_addr = NULL;
1329 * The end of the hostname occurs immediately preceding the delimiter or
1330 * the port marker (':') where the delimiter takes precedence.
1332 delim_p = memchr(name, delim, namelen);
1333 colon_p = memchr(name, ':', namelen);
1335 if (delim_p && colon_p)
1336 end = delim_p < colon_p ? delim_p : colon_p;
1337 else if (!delim_p && colon_p)
1341 if (!end) /* case: hostname:/ */
1342 end = name + namelen;
1348 /* do dns_resolve upcall */
1349 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1351 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1359 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1360 ret, ret ? "failed" : ceph_pr_addr(ss));
1365 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1366 struct sockaddr_storage *ss, char delim, const char **ipend)
1373 * Parse a server name (IP or hostname). If a valid IP address is not found
1374 * then try to extract a hostname to resolve using userspace DNS upcall.
1376 static int ceph_parse_server_name(const char *name, size_t namelen,
1377 struct sockaddr_storage *ss, char delim, const char **ipend)
1381 ret = ceph_pton(name, namelen, ss, delim, ipend);
1383 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1389 * Parse an ip[:port] list into an addr array. Use the default
1390 * monitor port if a port isn't specified.
1392 int ceph_parse_ips(const char *c, const char *end,
1393 struct ceph_entity_addr *addr,
1394 int max_count, int *count)
1396 int i, ret = -EINVAL;
1399 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1400 for (i = 0; i < max_count; i++) {
1402 struct sockaddr_storage *ss = &addr[i].in_addr;
1411 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1420 dout("missing matching ']'\n");
1427 if (p < end && *p == ':') {
1430 while (p < end && *p >= '0' && *p <= '9') {
1431 port = (port * 10) + (*p - '0');
1434 if (port > 65535 || port == 0)
1437 port = CEPH_MON_PORT;
1440 addr_set_port(ss, port);
1442 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1459 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1462 EXPORT_SYMBOL(ceph_parse_ips);
1464 static int process_banner(struct ceph_connection *con)
1466 dout("process_banner on %p\n", con);
1468 if (verify_hello(con) < 0)
1471 ceph_decode_addr(&con->actual_peer_addr);
1472 ceph_decode_addr(&con->peer_addr_for_me);
1475 * Make sure the other end is who we wanted. note that the other
1476 * end may not yet know their ip address, so if it's 0.0.0.0, give
1477 * them the benefit of the doubt.
1479 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1480 sizeof(con->peer_addr)) != 0 &&
1481 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1482 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1483 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1484 ceph_pr_addr(&con->peer_addr.in_addr),
1485 (int)le32_to_cpu(con->peer_addr.nonce),
1486 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1487 (int)le32_to_cpu(con->actual_peer_addr.nonce));
1488 con->error_msg = "wrong peer at address";
1493 * did we learn our address?
1495 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1496 int port = addr_port(&con->msgr->inst.addr.in_addr);
1498 memcpy(&con->msgr->inst.addr.in_addr,
1499 &con->peer_addr_for_me.in_addr,
1500 sizeof(con->peer_addr_for_me.in_addr));
1501 addr_set_port(&con->msgr->inst.addr.in_addr, port);
1502 encode_my_addr(con->msgr);
1503 dout("process_banner learned my addr is %s\n",
1504 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
1510 static int process_connect(struct ceph_connection *con)
1512 u64 sup_feat = con->msgr->supported_features;
1513 u64 req_feat = con->msgr->required_features;
1514 u64 server_feat = le64_to_cpu(con->in_reply.features);
1517 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1519 switch (con->in_reply.tag) {
1520 case CEPH_MSGR_TAG_FEATURES:
1521 pr_err("%s%lld %s feature set mismatch,"
1522 " my %llx < server's %llx, missing %llx\n",
1523 ENTITY_NAME(con->peer_name),
1524 ceph_pr_addr(&con->peer_addr.in_addr),
1525 sup_feat, server_feat, server_feat & ~sup_feat);
1526 con->error_msg = "missing required protocol features";
1527 reset_connection(con);
1530 case CEPH_MSGR_TAG_BADPROTOVER:
1531 pr_err("%s%lld %s protocol version mismatch,"
1532 " my %d != server's %d\n",
1533 ENTITY_NAME(con->peer_name),
1534 ceph_pr_addr(&con->peer_addr.in_addr),
1535 le32_to_cpu(con->out_connect.protocol_version),
1536 le32_to_cpu(con->in_reply.protocol_version));
1537 con->error_msg = "protocol version mismatch";
1538 reset_connection(con);
1541 case CEPH_MSGR_TAG_BADAUTHORIZER:
1543 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1545 if (con->auth_retry == 2) {
1546 con->error_msg = "connect authorization failure";
1549 con->auth_retry = 1;
1550 con_out_kvec_reset(con);
1551 ret = prepare_write_connect(con);
1554 prepare_read_connect(con);
1557 case CEPH_MSGR_TAG_RESETSESSION:
1559 * If we connected with a large connect_seq but the peer
1560 * has no record of a session with us (no connection, or
1561 * connect_seq == 0), they will send RESETSESION to indicate
1562 * that they must have reset their session, and may have
1565 dout("process_connect got RESET peer seq %u\n",
1566 le32_to_cpu(con->in_reply.connect_seq));
1567 pr_err("%s%lld %s connection reset\n",
1568 ENTITY_NAME(con->peer_name),
1569 ceph_pr_addr(&con->peer_addr.in_addr));
1570 reset_connection(con);
1571 con_out_kvec_reset(con);
1572 ret = prepare_write_connect(con);
1575 prepare_read_connect(con);
1577 /* Tell ceph about it. */
1578 mutex_unlock(&con->mutex);
1579 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
1580 if (con->ops->peer_reset)
1581 con->ops->peer_reset(con);
1582 mutex_lock(&con->mutex);
1583 if (con->state != CON_STATE_NEGOTIATING)
1587 case CEPH_MSGR_TAG_RETRY_SESSION:
1589 * If we sent a smaller connect_seq than the peer has, try
1590 * again with a larger value.
1592 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
1593 le32_to_cpu(con->out_connect.connect_seq),
1594 le32_to_cpu(con->in_reply.connect_seq));
1595 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
1596 con_out_kvec_reset(con);
1597 ret = prepare_write_connect(con);
1600 prepare_read_connect(con);
1603 case CEPH_MSGR_TAG_RETRY_GLOBAL:
1605 * If we sent a smaller global_seq than the peer has, try
1606 * again with a larger value.
1608 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1609 con->peer_global_seq,
1610 le32_to_cpu(con->in_reply.global_seq));
1611 get_global_seq(con->msgr,
1612 le32_to_cpu(con->in_reply.global_seq));
1613 con_out_kvec_reset(con);
1614 ret = prepare_write_connect(con);
1617 prepare_read_connect(con);
1620 case CEPH_MSGR_TAG_READY:
1621 if (req_feat & ~server_feat) {
1622 pr_err("%s%lld %s protocol feature mismatch,"
1623 " my required %llx > server's %llx, need %llx\n",
1624 ENTITY_NAME(con->peer_name),
1625 ceph_pr_addr(&con->peer_addr.in_addr),
1626 req_feat, server_feat, req_feat & ~server_feat);
1627 con->error_msg = "missing required protocol features";
1628 reset_connection(con);
1632 WARN_ON(con->state != CON_STATE_NEGOTIATING);
1633 con->state = CON_STATE_OPEN;
1635 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
1637 con->peer_features = server_feat;
1638 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1639 con->peer_global_seq,
1640 le32_to_cpu(con->in_reply.connect_seq),
1642 WARN_ON(con->connect_seq !=
1643 le32_to_cpu(con->in_reply.connect_seq));
1645 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
1646 set_bit(CON_FLAG_LOSSYTX, &con->flags);
1648 con->delay = 0; /* reset backoff memory */
1650 prepare_read_tag(con);
1653 case CEPH_MSGR_TAG_WAIT:
1655 * If there is a connection race (we are opening
1656 * connections to each other), one of us may just have
1657 * to WAIT. This shouldn't happen if we are the
1660 pr_err("process_connect got WAIT as client\n");
1661 con->error_msg = "protocol error, got WAIT as client";
1665 pr_err("connect protocol error, will retry\n");
1666 con->error_msg = "protocol error, garbage tag during connect";
1674 * read (part of) an ack
1676 static int read_partial_ack(struct ceph_connection *con)
1678 int size = sizeof (con->in_temp_ack);
1681 return read_partial(con, end, size, &con->in_temp_ack);
1686 * We can finally discard anything that's been acked.
1688 static void process_ack(struct ceph_connection *con)
1691 u64 ack = le64_to_cpu(con->in_temp_ack);
1694 while (!list_empty(&con->out_sent)) {
1695 m = list_first_entry(&con->out_sent, struct ceph_msg,
1697 seq = le64_to_cpu(m->hdr.seq);
1700 dout("got ack for seq %llu type %d at %p\n", seq,
1701 le16_to_cpu(m->hdr.type), m);
1702 m->ack_stamp = jiffies;
1705 prepare_read_tag(con);
1711 static int read_partial_message_section(struct ceph_connection *con,
1712 struct kvec *section,
1713 unsigned int sec_len, u32 *crc)
1719 while (section->iov_len < sec_len) {
1720 BUG_ON(section->iov_base == NULL);
1721 left = sec_len - section->iov_len;
1722 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
1723 section->iov_len, left);
1726 section->iov_len += ret;
1728 if (section->iov_len == sec_len)
1729 *crc = crc32c(0, section->iov_base, section->iov_len);
1734 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
1736 static int read_partial_message_pages(struct ceph_connection *con,
1737 struct page **pages,
1738 unsigned int data_len, bool do_datacrc)
1744 left = min((int)(data_len - con->in_msg_pos.data_pos),
1745 (int)(PAGE_SIZE - con->in_msg_pos.page_pos));
1747 BUG_ON(pages == NULL);
1748 p = kmap(pages[con->in_msg_pos.page]);
1749 ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos,
1751 if (ret > 0 && do_datacrc)
1753 crc32c(con->in_data_crc,
1754 p + con->in_msg_pos.page_pos, ret);
1755 kunmap(pages[con->in_msg_pos.page]);
1758 con->in_msg_pos.data_pos += ret;
1759 con->in_msg_pos.page_pos += ret;
1760 if (con->in_msg_pos.page_pos == PAGE_SIZE) {
1761 con->in_msg_pos.page_pos = 0;
1762 con->in_msg_pos.page++;
1769 static int read_partial_message_bio(struct ceph_connection *con,
1770 struct bio **bio_iter, int *bio_seg,
1771 unsigned int data_len, bool do_datacrc)
1773 struct bio_vec *bv = bio_iovec_idx(*bio_iter, *bio_seg);
1777 left = min((int)(data_len - con->in_msg_pos.data_pos),
1778 (int)(bv->bv_len - con->in_msg_pos.page_pos));
1780 p = kmap(bv->bv_page) + bv->bv_offset;
1782 ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos,
1784 if (ret > 0 && do_datacrc)
1786 crc32c(con->in_data_crc,
1787 p + con->in_msg_pos.page_pos, ret);
1788 kunmap(bv->bv_page);
1791 con->in_msg_pos.data_pos += ret;
1792 con->in_msg_pos.page_pos += ret;
1793 if (con->in_msg_pos.page_pos == bv->bv_len) {
1794 con->in_msg_pos.page_pos = 0;
1795 iter_bio_next(bio_iter, bio_seg);
1803 * read (part of) a message.
1805 static int read_partial_message(struct ceph_connection *con)
1807 struct ceph_msg *m = con->in_msg;
1811 unsigned int front_len, middle_len, data_len;
1812 bool do_datacrc = !con->msgr->nocrc;
1816 dout("read_partial_message con %p msg %p\n", con, m);
1819 size = sizeof (con->in_hdr);
1821 ret = read_partial(con, end, size, &con->in_hdr);
1825 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
1826 if (cpu_to_le32(crc) != con->in_hdr.crc) {
1827 pr_err("read_partial_message bad hdr "
1828 " crc %u != expected %u\n",
1829 crc, con->in_hdr.crc);
1833 front_len = le32_to_cpu(con->in_hdr.front_len);
1834 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
1836 middle_len = le32_to_cpu(con->in_hdr.middle_len);
1837 if (middle_len > CEPH_MSG_MAX_DATA_LEN)
1839 data_len = le32_to_cpu(con->in_hdr.data_len);
1840 if (data_len > CEPH_MSG_MAX_DATA_LEN)
1844 seq = le64_to_cpu(con->in_hdr.seq);
1845 if ((s64)seq - (s64)con->in_seq < 1) {
1846 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1847 ENTITY_NAME(con->peer_name),
1848 ceph_pr_addr(&con->peer_addr.in_addr),
1849 seq, con->in_seq + 1);
1850 con->in_base_pos = -front_len - middle_len - data_len -
1852 con->in_tag = CEPH_MSGR_TAG_READY;
1854 } else if ((s64)seq - (s64)con->in_seq > 1) {
1855 pr_err("read_partial_message bad seq %lld expected %lld\n",
1856 seq, con->in_seq + 1);
1857 con->error_msg = "bad message sequence # for incoming message";
1861 /* allocate message? */
1865 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
1866 con->in_hdr.front_len, con->in_hdr.data_len);
1867 ret = ceph_con_in_msg_alloc(con, &skip);
1871 /* skip this message */
1872 dout("alloc_msg said skip message\n");
1873 BUG_ON(con->in_msg);
1874 con->in_base_pos = -front_len - middle_len - data_len -
1876 con->in_tag = CEPH_MSGR_TAG_READY;
1881 BUG_ON(!con->in_msg);
1882 BUG_ON(con->in_msg->con != con);
1884 m->front.iov_len = 0; /* haven't read it yet */
1886 m->middle->vec.iov_len = 0;
1888 con->in_msg_pos.page = 0;
1890 con->in_msg_pos.page_pos = m->page_alignment;
1892 con->in_msg_pos.page_pos = 0;
1893 con->in_msg_pos.data_pos = 0;
1897 init_bio_iter(m->bio, &m->bio_iter, &m->bio_seg);
1902 ret = read_partial_message_section(con, &m->front, front_len,
1903 &con->in_front_crc);
1909 ret = read_partial_message_section(con, &m->middle->vec,
1911 &con->in_middle_crc);
1917 while (con->in_msg_pos.data_pos < data_len) {
1919 ret = read_partial_message_pages(con, m->pages,
1920 data_len, do_datacrc);
1924 } else if (m->bio) {
1925 BUG_ON(!m->bio_iter);
1926 ret = read_partial_message_bio(con,
1927 &m->bio_iter, &m->bio_seg,
1928 data_len, do_datacrc);
1938 size = sizeof (m->footer);
1940 ret = read_partial(con, end, size, &m->footer);
1944 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1945 m, front_len, m->footer.front_crc, middle_len,
1946 m->footer.middle_crc, data_len, m->footer.data_crc);
1949 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
1950 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1951 m, con->in_front_crc, m->footer.front_crc);
1954 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
1955 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1956 m, con->in_middle_crc, m->footer.middle_crc);
1960 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
1961 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
1962 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
1963 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
1967 return 1; /* done! */
1971 * Process message. This happens in the worker thread. The callback should
1972 * be careful not to do anything that waits on other incoming messages or it
1975 static void process_message(struct ceph_connection *con)
1977 struct ceph_msg *msg;
1979 BUG_ON(con->in_msg->con != con);
1980 con->in_msg->con = NULL;
1985 /* if first message, set peer_name */
1986 if (con->peer_name.type == 0)
1987 con->peer_name = msg->hdr.src;
1990 mutex_unlock(&con->mutex);
1992 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
1993 msg, le64_to_cpu(msg->hdr.seq),
1994 ENTITY_NAME(msg->hdr.src),
1995 le16_to_cpu(msg->hdr.type),
1996 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
1997 le32_to_cpu(msg->hdr.front_len),
1998 le32_to_cpu(msg->hdr.data_len),
1999 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2000 con->ops->dispatch(con, msg);
2002 mutex_lock(&con->mutex);
2007 * Write something to the socket. Called in a worker thread when the
2008 * socket appears to be writeable and we have something ready to send.
2010 static int try_write(struct ceph_connection *con)
2014 dout("try_write start %p state %lu\n", con, con->state);
2017 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2019 /* open the socket first? */
2020 if (con->state == CON_STATE_PREOPEN) {
2022 con->state = CON_STATE_CONNECTING;
2024 con_out_kvec_reset(con);
2025 prepare_write_banner(con);
2026 prepare_read_banner(con);
2028 BUG_ON(con->in_msg);
2029 con->in_tag = CEPH_MSGR_TAG_READY;
2030 dout("try_write initiating connect on %p new state %lu\n",
2032 ret = ceph_tcp_connect(con);
2034 con->error_msg = "connect error";
2040 /* kvec data queued? */
2041 if (con->out_skip) {
2042 ret = write_partial_skip(con);
2046 if (con->out_kvec_left) {
2047 ret = write_partial_kvec(con);
2054 if (con->out_msg_done) {
2055 ceph_msg_put(con->out_msg);
2056 con->out_msg = NULL; /* we're done with this one */
2060 ret = write_partial_msg_pages(con);
2062 goto more_kvec; /* we need to send the footer, too! */
2066 dout("try_write write_partial_msg_pages err %d\n",
2073 if (con->state == CON_STATE_OPEN) {
2074 /* is anything else pending? */
2075 if (!list_empty(&con->out_queue)) {
2076 prepare_write_message(con);
2079 if (con->in_seq > con->in_seq_acked) {
2080 prepare_write_ack(con);
2083 if (test_and_clear_bit(CON_FLAG_KEEPALIVE_PENDING,
2085 prepare_write_keepalive(con);
2090 /* Nothing to do! */
2091 clear_bit(CON_FLAG_WRITE_PENDING, &con->flags);
2092 dout("try_write nothing else to write.\n");
2095 dout("try_write done on %p ret %d\n", con, ret);
2102 * Read what we can from the socket.
2104 static int try_read(struct ceph_connection *con)
2109 dout("try_read start on %p state %lu\n", con, con->state);
2110 if (con->state != CON_STATE_CONNECTING &&
2111 con->state != CON_STATE_NEGOTIATING &&
2112 con->state != CON_STATE_OPEN)
2117 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2120 if (con->state == CON_STATE_CONNECTING) {
2121 dout("try_read connecting\n");
2122 ret = read_partial_banner(con);
2125 ret = process_banner(con);
2129 con->state = CON_STATE_NEGOTIATING;
2132 * Received banner is good, exchange connection info.
2133 * Do not reset out_kvec, as sending our banner raced
2134 * with receiving peer banner after connect completed.
2136 ret = prepare_write_connect(con);
2139 prepare_read_connect(con);
2141 /* Send connection info before awaiting response */
2145 if (con->state == CON_STATE_NEGOTIATING) {
2146 dout("try_read negotiating\n");
2147 ret = read_partial_connect(con);
2150 ret = process_connect(con);
2156 WARN_ON(con->state != CON_STATE_OPEN);
2158 if (con->in_base_pos < 0) {
2160 * skipping + discarding content.
2162 * FIXME: there must be a better way to do this!
2164 static char buf[SKIP_BUF_SIZE];
2165 int skip = min((int) sizeof (buf), -con->in_base_pos);
2167 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2168 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2171 con->in_base_pos += ret;
2172 if (con->in_base_pos)
2175 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2179 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2182 dout("try_read got tag %d\n", (int)con->in_tag);
2183 switch (con->in_tag) {
2184 case CEPH_MSGR_TAG_MSG:
2185 prepare_read_message(con);
2187 case CEPH_MSGR_TAG_ACK:
2188 prepare_read_ack(con);
2190 case CEPH_MSGR_TAG_CLOSE:
2191 con_close_socket(con);
2192 con->state = CON_STATE_CLOSED;
2198 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2199 ret = read_partial_message(con);
2203 con->error_msg = "bad crc";
2207 con->error_msg = "io error";
2212 if (con->in_tag == CEPH_MSGR_TAG_READY)
2214 process_message(con);
2215 if (con->state == CON_STATE_OPEN)
2216 prepare_read_tag(con);
2219 if (con->in_tag == CEPH_MSGR_TAG_ACK) {
2220 ret = read_partial_ack(con);
2228 dout("try_read done on %p ret %d\n", con, ret);
2232 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2233 con->error_msg = "protocol error, garbage tag";
2240 * Atomically queue work on a connection after the specified delay.
2241 * Bump @con reference to avoid races with connection teardown.
2242 * Returns 0 if work was queued, or an error code otherwise.
2244 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2246 if (!con->ops->get(con)) {
2247 dout("%s %p ref count 0\n", __func__, con);
2252 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2253 dout("%s %p - already queued\n", __func__, con);
2259 dout("%s %p %lu\n", __func__, con, delay);
2264 static void queue_con(struct ceph_connection *con)
2266 (void) queue_con_delay(con, 0);
2269 static bool con_sock_closed(struct ceph_connection *con)
2271 if (!test_and_clear_bit(CON_FLAG_SOCK_CLOSED, &con->flags))
2275 case CON_STATE_ ## x: \
2276 con->error_msg = "socket closed (con state " #x ")"; \
2279 switch (con->state) {
2287 pr_warning("%s con %p unrecognized state %lu\n",
2288 __func__, con, con->state);
2289 con->error_msg = "unrecognized con state";
2299 * Do some work on a connection. Drop a connection ref when we're done.
2301 static void con_work(struct work_struct *work)
2303 struct ceph_connection *con = container_of(work, struct ceph_connection,
2307 mutex_lock(&con->mutex);
2309 if (con_sock_closed(con))
2312 if (test_and_clear_bit(CON_FLAG_BACKOFF, &con->flags)) {
2313 dout("con_work %p backing off\n", con);
2314 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2316 dout("con_work %p FAILED to back off %lu\n", con,
2318 BUG_ON(ret == -ENOENT);
2319 set_bit(CON_FLAG_BACKOFF, &con->flags);
2324 if (con->state == CON_STATE_STANDBY) {
2325 dout("con_work %p STANDBY\n", con);
2328 if (con->state == CON_STATE_CLOSED) {
2329 dout("con_work %p CLOSED\n", con);
2333 if (con->state == CON_STATE_PREOPEN) {
2334 dout("con_work OPENING\n");
2338 ret = try_read(con);
2342 con->error_msg = "socket error on read";
2346 ret = try_write(con);
2350 con->error_msg = "socket error on write";
2355 mutex_unlock(&con->mutex);
2361 ceph_fault(con); /* error/fault path */
2367 * Generic error/fault handler. A retry mechanism is used with
2368 * exponential backoff
2370 static void ceph_fault(struct ceph_connection *con)
2371 __releases(con->mutex)
2373 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2374 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2375 dout("fault %p state %lu to peer %s\n",
2376 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2378 WARN_ON(con->state != CON_STATE_CONNECTING &&
2379 con->state != CON_STATE_NEGOTIATING &&
2380 con->state != CON_STATE_OPEN);
2382 con_close_socket(con);
2384 if (test_bit(CON_FLAG_LOSSYTX, &con->flags)) {
2385 dout("fault on LOSSYTX channel, marking CLOSED\n");
2386 con->state = CON_STATE_CLOSED;
2391 BUG_ON(con->in_msg->con != con);
2392 con->in_msg->con = NULL;
2393 ceph_msg_put(con->in_msg);
2398 /* Requeue anything that hasn't been acked */
2399 list_splice_init(&con->out_sent, &con->out_queue);
2401 /* If there are no messages queued or keepalive pending, place
2402 * the connection in a STANDBY state */
2403 if (list_empty(&con->out_queue) &&
2404 !test_bit(CON_FLAG_KEEPALIVE_PENDING, &con->flags)) {
2405 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2406 clear_bit(CON_FLAG_WRITE_PENDING, &con->flags);
2407 con->state = CON_STATE_STANDBY;
2409 /* retry after a delay. */
2410 con->state = CON_STATE_PREOPEN;
2411 if (con->delay == 0)
2412 con->delay = BASE_DELAY_INTERVAL;
2413 else if (con->delay < MAX_DELAY_INTERVAL)
2415 set_bit(CON_FLAG_BACKOFF, &con->flags);
2420 mutex_unlock(&con->mutex);
2422 * in case we faulted due to authentication, invalidate our
2423 * current tickets so that we can get new ones.
2425 if (con->auth_retry && con->ops->invalidate_authorizer) {
2426 dout("calling invalidate_authorizer()\n");
2427 con->ops->invalidate_authorizer(con);
2430 if (con->ops->fault)
2431 con->ops->fault(con);
2437 * initialize a new messenger instance
2439 void ceph_messenger_init(struct ceph_messenger *msgr,
2440 struct ceph_entity_addr *myaddr,
2441 u32 supported_features,
2442 u32 required_features,
2445 msgr->supported_features = supported_features;
2446 msgr->required_features = required_features;
2448 spin_lock_init(&msgr->global_seq_lock);
2451 msgr->inst.addr = *myaddr;
2453 /* select a random nonce */
2454 msgr->inst.addr.type = 0;
2455 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2456 encode_my_addr(msgr);
2457 msgr->nocrc = nocrc;
2459 atomic_set(&msgr->stopping, 0);
2461 dout("%s %p\n", __func__, msgr);
2463 EXPORT_SYMBOL(ceph_messenger_init);
2465 static void clear_standby(struct ceph_connection *con)
2467 /* come back from STANDBY? */
2468 if (con->state == CON_STATE_STANDBY) {
2469 dout("clear_standby %p and ++connect_seq\n", con);
2470 con->state = CON_STATE_PREOPEN;
2472 WARN_ON(test_bit(CON_FLAG_WRITE_PENDING, &con->flags));
2473 WARN_ON(test_bit(CON_FLAG_KEEPALIVE_PENDING, &con->flags));
2478 * Queue up an outgoing message on the given connection.
2480 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2483 msg->hdr.src = con->msgr->inst.name;
2484 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2485 msg->needs_out_seq = true;
2487 mutex_lock(&con->mutex);
2489 if (con->state == CON_STATE_CLOSED) {
2490 dout("con_send %p closed, dropping %p\n", con, msg);
2492 mutex_unlock(&con->mutex);
2496 BUG_ON(msg->con != NULL);
2497 msg->con = con->ops->get(con);
2498 BUG_ON(msg->con == NULL);
2500 BUG_ON(!list_empty(&msg->list_head));
2501 list_add_tail(&msg->list_head, &con->out_queue);
2502 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2503 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2504 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2505 le32_to_cpu(msg->hdr.front_len),
2506 le32_to_cpu(msg->hdr.middle_len),
2507 le32_to_cpu(msg->hdr.data_len));
2510 mutex_unlock(&con->mutex);
2512 /* if there wasn't anything waiting to send before, queue
2514 if (test_and_set_bit(CON_FLAG_WRITE_PENDING, &con->flags) == 0)
2517 EXPORT_SYMBOL(ceph_con_send);
2520 * Revoke a message that was previously queued for send
2522 void ceph_msg_revoke(struct ceph_msg *msg)
2524 struct ceph_connection *con = msg->con;
2527 return; /* Message not in our possession */
2529 mutex_lock(&con->mutex);
2530 if (!list_empty(&msg->list_head)) {
2531 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
2532 list_del_init(&msg->list_head);
2533 BUG_ON(msg->con == NULL);
2534 msg->con->ops->put(msg->con);
2540 if (con->out_msg == msg) {
2541 dout("%s %p msg %p - was sending\n", __func__, con, msg);
2542 con->out_msg = NULL;
2543 if (con->out_kvec_is_msg) {
2544 con->out_skip = con->out_kvec_bytes;
2545 con->out_kvec_is_msg = false;
2551 mutex_unlock(&con->mutex);
2555 * Revoke a message that we may be reading data into
2557 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
2559 struct ceph_connection *con;
2561 BUG_ON(msg == NULL);
2563 dout("%s msg %p null con\n", __func__, msg);
2565 return; /* Message not in our possession */
2569 mutex_lock(&con->mutex);
2570 if (con->in_msg == msg) {
2571 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
2572 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
2573 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
2575 /* skip rest of message */
2576 dout("%s %p msg %p revoked\n", __func__, con, msg);
2577 con->in_base_pos = con->in_base_pos -
2578 sizeof(struct ceph_msg_header) -
2582 sizeof(struct ceph_msg_footer);
2583 ceph_msg_put(con->in_msg);
2585 con->in_tag = CEPH_MSGR_TAG_READY;
2588 dout("%s %p in_msg %p msg %p no-op\n",
2589 __func__, con, con->in_msg, msg);
2591 mutex_unlock(&con->mutex);
2595 * Queue a keepalive byte to ensure the tcp connection is alive.
2597 void ceph_con_keepalive(struct ceph_connection *con)
2599 dout("con_keepalive %p\n", con);
2600 mutex_lock(&con->mutex);
2602 mutex_unlock(&con->mutex);
2603 if (test_and_set_bit(CON_FLAG_KEEPALIVE_PENDING, &con->flags) == 0 &&
2604 test_and_set_bit(CON_FLAG_WRITE_PENDING, &con->flags) == 0)
2607 EXPORT_SYMBOL(ceph_con_keepalive);
2611 * construct a new message with given type, size
2612 * the new msg has a ref count of 1.
2614 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
2619 m = kmalloc(sizeof(*m), flags);
2622 kref_init(&m->kref);
2625 INIT_LIST_HEAD(&m->list_head);
2628 m->hdr.type = cpu_to_le16(type);
2629 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
2631 m->hdr.front_len = cpu_to_le32(front_len);
2632 m->hdr.middle_len = 0;
2633 m->hdr.data_len = 0;
2634 m->hdr.data_off = 0;
2635 m->hdr.reserved = 0;
2636 m->footer.front_crc = 0;
2637 m->footer.middle_crc = 0;
2638 m->footer.data_crc = 0;
2639 m->footer.flags = 0;
2640 m->front_max = front_len;
2641 m->front_is_vmalloc = false;
2642 m->more_to_follow = false;
2651 m->page_alignment = 0;
2661 if (front_len > PAGE_CACHE_SIZE) {
2662 m->front.iov_base = __vmalloc(front_len, flags,
2664 m->front_is_vmalloc = true;
2666 m->front.iov_base = kmalloc(front_len, flags);
2668 if (m->front.iov_base == NULL) {
2669 dout("ceph_msg_new can't allocate %d bytes\n",
2674 m->front.iov_base = NULL;
2676 m->front.iov_len = front_len;
2678 dout("ceph_msg_new %p front %d\n", m, front_len);
2685 pr_err("msg_new can't create type %d front %d\n", type,
2689 dout("msg_new can't create type %d front %d\n", type,
2694 EXPORT_SYMBOL(ceph_msg_new);
2697 * Allocate "middle" portion of a message, if it is needed and wasn't
2698 * allocated by alloc_msg. This allows us to read a small fixed-size
2699 * per-type header in the front and then gracefully fail (i.e.,
2700 * propagate the error to the caller based on info in the front) when
2701 * the middle is too large.
2703 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
2705 int type = le16_to_cpu(msg->hdr.type);
2706 int middle_len = le32_to_cpu(msg->hdr.middle_len);
2708 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
2709 ceph_msg_type_name(type), middle_len);
2710 BUG_ON(!middle_len);
2711 BUG_ON(msg->middle);
2713 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
2720 * Allocate a message for receiving an incoming message on a
2721 * connection, and save the result in con->in_msg. Uses the
2722 * connection's private alloc_msg op if available.
2724 * Returns 0 on success, or a negative error code.
2726 * On success, if we set *skip = 1:
2727 * - the next message should be skipped and ignored.
2728 * - con->in_msg == NULL
2729 * or if we set *skip = 0:
2730 * - con->in_msg is non-null.
2731 * On error (ENOMEM, EAGAIN, ...),
2732 * - con->in_msg == NULL
2734 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
2736 struct ceph_msg_header *hdr = &con->in_hdr;
2737 int type = le16_to_cpu(hdr->type);
2738 int front_len = le32_to_cpu(hdr->front_len);
2739 int middle_len = le32_to_cpu(hdr->middle_len);
2742 BUG_ON(con->in_msg != NULL);
2744 if (con->ops->alloc_msg) {
2745 struct ceph_msg *msg;
2747 mutex_unlock(&con->mutex);
2748 msg = con->ops->alloc_msg(con, hdr, skip);
2749 mutex_lock(&con->mutex);
2750 if (con->state != CON_STATE_OPEN) {
2757 con->in_msg->con = con->ops->get(con);
2758 BUG_ON(con->in_msg->con == NULL);
2766 "error allocating memory for incoming message";
2771 con->in_msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
2773 pr_err("unable to allocate msg type %d len %d\n",
2777 con->in_msg->con = con->ops->get(con);
2778 BUG_ON(con->in_msg->con == NULL);
2779 con->in_msg->page_alignment = le16_to_cpu(hdr->data_off);
2781 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2783 if (middle_len && !con->in_msg->middle) {
2784 ret = ceph_alloc_middle(con, con->in_msg);
2786 ceph_msg_put(con->in_msg);
2796 * Free a generically kmalloc'd message.
2798 void ceph_msg_kfree(struct ceph_msg *m)
2800 dout("msg_kfree %p\n", m);
2801 if (m->front_is_vmalloc)
2802 vfree(m->front.iov_base);
2804 kfree(m->front.iov_base);
2809 * Drop a msg ref. Destroy as needed.
2811 void ceph_msg_last_put(struct kref *kref)
2813 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
2815 dout("ceph_msg_put last one on %p\n", m);
2816 WARN_ON(!list_empty(&m->list_head));
2818 /* drop middle, data, if any */
2820 ceph_buffer_put(m->middle);
2827 ceph_pagelist_release(m->pagelist);
2835 ceph_msgpool_put(m->pool, m);
2839 EXPORT_SYMBOL(ceph_msg_last_put);
2841 void ceph_msg_dump(struct ceph_msg *msg)
2843 pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg,
2844 msg->front_max, msg->nr_pages);
2845 print_hex_dump(KERN_DEBUG, "header: ",
2846 DUMP_PREFIX_OFFSET, 16, 1,
2847 &msg->hdr, sizeof(msg->hdr), true);
2848 print_hex_dump(KERN_DEBUG, " front: ",
2849 DUMP_PREFIX_OFFSET, 16, 1,
2850 msg->front.iov_base, msg->front.iov_len, true);
2852 print_hex_dump(KERN_DEBUG, "middle: ",
2853 DUMP_PREFIX_OFFSET, 16, 1,
2854 msg->middle->vec.iov_base,
2855 msg->middle->vec.iov_len, true);
2856 print_hex_dump(KERN_DEBUG, "footer: ",
2857 DUMP_PREFIX_OFFSET, 16, 1,
2858 &msg->footer, sizeof(msg->footer), true);
2860 EXPORT_SYMBOL(ceph_msg_dump);