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/libceph.h>
19 #include <linux/ceph/messenger.h>
20 #include <linux/ceph/decode.h>
21 #include <linux/ceph/pagelist.h>
22 #include <linux/export.h>
25 * Ceph uses the messenger to exchange ceph_msg messages with other
26 * hosts in the system. The messenger provides ordered and reliable
27 * delivery. We tolerate TCP disconnects by reconnecting (with
28 * exponential backoff) in the case of a fault (disconnection, bad
29 * crc, protocol error). Acks allow sent messages to be discarded by
34 * We track the state of the socket on a given connection using
35 * values defined below. The transition to a new socket state is
36 * handled by a function which verifies we aren't coming from an
40 * | NEW* | transient initial state
42 * | con_sock_state_init()
45 * | CLOSED | initialized, but no socket (and no
46 * ---------- TCP connection)
48 * | \ con_sock_state_connecting()
49 * | ----------------------
51 * + con_sock_state_closed() \
52 * |+--------------------------- \
55 * | | CLOSING | socket event; \ \
56 * | ----------- await close \ \
59 * | + con_sock_state_closing() \ |
61 * | / --------------- | |
64 * | / -----------------| CONNECTING | socket created, TCP
65 * | | / -------------- connect initiated
66 * | | | con_sock_state_connected()
69 * | CONNECTED | TCP connection established
72 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
75 #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
76 #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
77 #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
78 #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
79 #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
84 #define CON_STATE_CLOSED 1 /* -> PREOPEN */
85 #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
86 #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
87 #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
88 #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
89 #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
92 * ceph_connection flag bits
94 #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
95 * messages on errors */
96 #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
97 #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
98 #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
99 #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
101 static bool con_flag_valid(unsigned long con_flag)
104 case CON_FLAG_LOSSYTX:
105 case CON_FLAG_KEEPALIVE_PENDING:
106 case CON_FLAG_WRITE_PENDING:
107 case CON_FLAG_SOCK_CLOSED:
108 case CON_FLAG_BACKOFF:
115 static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
117 BUG_ON(!con_flag_valid(con_flag));
119 clear_bit(con_flag, &con->flags);
122 static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
124 BUG_ON(!con_flag_valid(con_flag));
126 set_bit(con_flag, &con->flags);
129 static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
131 BUG_ON(!con_flag_valid(con_flag));
133 return test_bit(con_flag, &con->flags);
136 static bool con_flag_test_and_clear(struct ceph_connection *con,
137 unsigned long con_flag)
139 BUG_ON(!con_flag_valid(con_flag));
141 return test_and_clear_bit(con_flag, &con->flags);
144 static bool con_flag_test_and_set(struct ceph_connection *con,
145 unsigned long con_flag)
147 BUG_ON(!con_flag_valid(con_flag));
149 return test_and_set_bit(con_flag, &con->flags);
152 /* static tag bytes (protocol control messages) */
153 static char tag_msg = CEPH_MSGR_TAG_MSG;
154 static char tag_ack = CEPH_MSGR_TAG_ACK;
155 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
157 #ifdef CONFIG_LOCKDEP
158 static struct lock_class_key socket_class;
162 * When skipping (ignoring) a block of input we read it into a "skip
163 * buffer," which is this many bytes in size.
165 #define SKIP_BUF_SIZE 1024
167 static void queue_con(struct ceph_connection *con);
168 static void con_work(struct work_struct *);
169 static void con_fault(struct ceph_connection *con);
172 * Nicely render a sockaddr as a string. An array of formatted
173 * strings is used, to approximate reentrancy.
175 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
176 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
177 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
178 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
180 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
181 static atomic_t addr_str_seq = ATOMIC_INIT(0);
183 static struct page *zero_page; /* used in certain error cases */
185 const char *ceph_pr_addr(const struct sockaddr_storage *ss)
189 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
190 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
192 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
195 switch (ss->ss_family) {
197 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
198 ntohs(in4->sin_port));
202 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
203 ntohs(in6->sin6_port));
207 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
213 EXPORT_SYMBOL(ceph_pr_addr);
215 static void encode_my_addr(struct ceph_messenger *msgr)
217 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
218 ceph_encode_addr(&msgr->my_enc_addr);
222 * work queue for all reading and writing to/from the socket.
224 static struct workqueue_struct *ceph_msgr_wq;
226 static void _ceph_msgr_exit(void)
229 destroy_workqueue(ceph_msgr_wq);
233 BUG_ON(zero_page == NULL);
235 page_cache_release(zero_page);
239 int ceph_msgr_init(void)
241 BUG_ON(zero_page != NULL);
242 zero_page = ZERO_PAGE(0);
243 page_cache_get(zero_page);
245 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT, 0);
249 pr_err("msgr_init failed to create workqueue\n");
254 EXPORT_SYMBOL(ceph_msgr_init);
256 void ceph_msgr_exit(void)
258 BUG_ON(ceph_msgr_wq == NULL);
262 EXPORT_SYMBOL(ceph_msgr_exit);
264 void ceph_msgr_flush(void)
266 flush_workqueue(ceph_msgr_wq);
268 EXPORT_SYMBOL(ceph_msgr_flush);
270 /* Connection socket state transition functions */
272 static void con_sock_state_init(struct ceph_connection *con)
276 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
277 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
278 printk("%s: unexpected old state %d\n", __func__, old_state);
279 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
280 CON_SOCK_STATE_CLOSED);
283 static void con_sock_state_connecting(struct ceph_connection *con)
287 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
288 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
289 printk("%s: unexpected old state %d\n", __func__, old_state);
290 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
291 CON_SOCK_STATE_CONNECTING);
294 static void con_sock_state_connected(struct ceph_connection *con)
298 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
299 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
300 printk("%s: unexpected old state %d\n", __func__, old_state);
301 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
302 CON_SOCK_STATE_CONNECTED);
305 static void con_sock_state_closing(struct ceph_connection *con)
309 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
310 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
311 old_state != CON_SOCK_STATE_CONNECTED &&
312 old_state != CON_SOCK_STATE_CLOSING))
313 printk("%s: unexpected old state %d\n", __func__, old_state);
314 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
315 CON_SOCK_STATE_CLOSING);
318 static void con_sock_state_closed(struct ceph_connection *con)
322 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
323 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
324 old_state != CON_SOCK_STATE_CLOSING &&
325 old_state != CON_SOCK_STATE_CONNECTING &&
326 old_state != CON_SOCK_STATE_CLOSED))
327 printk("%s: unexpected old state %d\n", __func__, old_state);
328 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
329 CON_SOCK_STATE_CLOSED);
333 * socket callback functions
336 /* data available on socket, or listen socket received a connect */
337 static void ceph_sock_data_ready(struct sock *sk, int count_unused)
339 struct ceph_connection *con = sk->sk_user_data;
340 if (atomic_read(&con->msgr->stopping)) {
344 if (sk->sk_state != TCP_CLOSE_WAIT) {
345 dout("%s on %p state = %lu, queueing work\n", __func__,
351 /* socket has buffer space for writing */
352 static void ceph_sock_write_space(struct sock *sk)
354 struct ceph_connection *con = sk->sk_user_data;
356 /* only queue to workqueue if there is data we want to write,
357 * and there is sufficient space in the socket buffer to accept
358 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
359 * doesn't get called again until try_write() fills the socket
360 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
361 * and net/core/stream.c:sk_stream_write_space().
363 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
364 if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
365 dout("%s %p queueing write work\n", __func__, con);
366 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
370 dout("%s %p nothing to write\n", __func__, con);
374 /* socket's state has changed */
375 static void ceph_sock_state_change(struct sock *sk)
377 struct ceph_connection *con = sk->sk_user_data;
379 dout("%s %p state = %lu sk_state = %u\n", __func__,
380 con, con->state, sk->sk_state);
382 switch (sk->sk_state) {
384 dout("%s TCP_CLOSE\n", __func__);
386 dout("%s TCP_CLOSE_WAIT\n", __func__);
387 con_sock_state_closing(con);
388 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
391 case TCP_ESTABLISHED:
392 dout("%s TCP_ESTABLISHED\n", __func__);
393 con_sock_state_connected(con);
396 default: /* Everything else is uninteresting */
402 * set up socket callbacks
404 static void set_sock_callbacks(struct socket *sock,
405 struct ceph_connection *con)
407 struct sock *sk = sock->sk;
408 sk->sk_user_data = con;
409 sk->sk_data_ready = ceph_sock_data_ready;
410 sk->sk_write_space = ceph_sock_write_space;
411 sk->sk_state_change = ceph_sock_state_change;
420 * initiate connection to a remote socket.
422 static int ceph_tcp_connect(struct ceph_connection *con)
424 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
429 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
433 sock->sk->sk_allocation = GFP_NOFS;
435 #ifdef CONFIG_LOCKDEP
436 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
439 set_sock_callbacks(sock, con);
441 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
443 con_sock_state_connecting(con);
444 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
446 if (ret == -EINPROGRESS) {
447 dout("connect %s EINPROGRESS sk_state = %u\n",
448 ceph_pr_addr(&con->peer_addr.in_addr),
450 } else if (ret < 0) {
451 pr_err("connect %s error %d\n",
452 ceph_pr_addr(&con->peer_addr.in_addr), ret);
454 con->error_msg = "connect error";
462 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
464 struct kvec iov = {buf, len};
465 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
468 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
475 * write something. @more is true if caller will be sending more data
478 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
479 size_t kvlen, size_t len, int more)
481 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
485 msg.msg_flags |= MSG_MORE;
487 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
489 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
495 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
496 int offset, size_t size, int more)
498 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
501 ret = kernel_sendpage(sock, page, offset, size, flags);
510 * Shutdown/close the socket for the given connection.
512 static int con_close_socket(struct ceph_connection *con)
516 dout("con_close_socket on %p sock %p\n", con, con->sock);
518 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
519 sock_release(con->sock);
524 * Forcibly clear the SOCK_CLOSED flag. It gets set
525 * independent of the connection mutex, and we could have
526 * received a socket close event before we had the chance to
527 * shut the socket down.
529 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
531 con_sock_state_closed(con);
536 * Reset a connection. Discard all incoming and outgoing messages
537 * and clear *_seq state.
539 static void ceph_msg_remove(struct ceph_msg *msg)
541 list_del_init(&msg->list_head);
542 BUG_ON(msg->con == NULL);
543 msg->con->ops->put(msg->con);
548 static void ceph_msg_remove_list(struct list_head *head)
550 while (!list_empty(head)) {
551 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
553 ceph_msg_remove(msg);
557 static void reset_connection(struct ceph_connection *con)
559 /* reset connection, out_queue, msg_ and connect_seq */
560 /* discard existing out_queue and msg_seq */
561 dout("reset_connection %p\n", con);
562 ceph_msg_remove_list(&con->out_queue);
563 ceph_msg_remove_list(&con->out_sent);
566 BUG_ON(con->in_msg->con != con);
567 con->in_msg->con = NULL;
568 ceph_msg_put(con->in_msg);
573 con->connect_seq = 0;
576 ceph_msg_put(con->out_msg);
580 con->in_seq_acked = 0;
584 * mark a peer down. drop any open connections.
586 void ceph_con_close(struct ceph_connection *con)
588 mutex_lock(&con->mutex);
589 dout("con_close %p peer %s\n", con,
590 ceph_pr_addr(&con->peer_addr.in_addr));
591 con->state = CON_STATE_CLOSED;
593 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
594 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
595 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
596 con_flag_clear(con, CON_FLAG_BACKOFF);
598 reset_connection(con);
599 con->peer_global_seq = 0;
600 cancel_delayed_work(&con->work);
601 con_close_socket(con);
602 mutex_unlock(&con->mutex);
604 EXPORT_SYMBOL(ceph_con_close);
607 * Reopen a closed connection, with a new peer address.
609 void ceph_con_open(struct ceph_connection *con,
610 __u8 entity_type, __u64 entity_num,
611 struct ceph_entity_addr *addr)
613 mutex_lock(&con->mutex);
614 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
616 WARN_ON(con->state != CON_STATE_CLOSED);
617 con->state = CON_STATE_PREOPEN;
619 con->peer_name.type = (__u8) entity_type;
620 con->peer_name.num = cpu_to_le64(entity_num);
622 memcpy(&con->peer_addr, addr, sizeof(*addr));
623 con->delay = 0; /* reset backoff memory */
624 mutex_unlock(&con->mutex);
627 EXPORT_SYMBOL(ceph_con_open);
630 * return true if this connection ever successfully opened
632 bool ceph_con_opened(struct ceph_connection *con)
634 return con->connect_seq > 0;
638 * initialize a new connection.
640 void ceph_con_init(struct ceph_connection *con, void *private,
641 const struct ceph_connection_operations *ops,
642 struct ceph_messenger *msgr)
644 dout("con_init %p\n", con);
645 memset(con, 0, sizeof(*con));
646 con->private = private;
650 con_sock_state_init(con);
652 mutex_init(&con->mutex);
653 INIT_LIST_HEAD(&con->out_queue);
654 INIT_LIST_HEAD(&con->out_sent);
655 INIT_DELAYED_WORK(&con->work, con_work);
657 con->state = CON_STATE_CLOSED;
659 EXPORT_SYMBOL(ceph_con_init);
663 * We maintain a global counter to order connection attempts. Get
664 * a unique seq greater than @gt.
666 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
670 spin_lock(&msgr->global_seq_lock);
671 if (msgr->global_seq < gt)
672 msgr->global_seq = gt;
673 ret = ++msgr->global_seq;
674 spin_unlock(&msgr->global_seq_lock);
678 static void con_out_kvec_reset(struct ceph_connection *con)
680 con->out_kvec_left = 0;
681 con->out_kvec_bytes = 0;
682 con->out_kvec_cur = &con->out_kvec[0];
685 static void con_out_kvec_add(struct ceph_connection *con,
686 size_t size, void *data)
690 index = con->out_kvec_left;
691 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
693 con->out_kvec[index].iov_len = size;
694 con->out_kvec[index].iov_base = data;
695 con->out_kvec_left++;
696 con->out_kvec_bytes += size;
700 static void init_bio_iter(struct bio *bio, struct bio **iter, int *seg)
711 static void iter_bio_next(struct bio **bio_iter, int *seg)
713 if (*bio_iter == NULL)
716 BUG_ON(*seg >= (*bio_iter)->bi_vcnt);
719 if (*seg == (*bio_iter)->bi_vcnt)
720 init_bio_iter((*bio_iter)->bi_next, bio_iter, seg);
724 static void prepare_write_message_data(struct ceph_connection *con)
726 struct ceph_msg *msg = con->out_msg;
729 BUG_ON(!msg->hdr.data_len);
731 /* initialize page iterator */
732 con->out_msg_pos.page = 0;
734 con->out_msg_pos.page_pos = msg->page_alignment;
736 con->out_msg_pos.page_pos = 0;
739 init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg);
741 con->out_msg_pos.data_pos = 0;
742 con->out_msg_pos.did_page_crc = false;
743 con->out_more = 1; /* data + footer will follow */
747 * Prepare footer for currently outgoing message, and finish things
748 * off. Assumes out_kvec* are already valid.. we just add on to the end.
750 static void prepare_write_message_footer(struct ceph_connection *con)
752 struct ceph_msg *m = con->out_msg;
753 int v = con->out_kvec_left;
755 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
757 dout("prepare_write_message_footer %p\n", con);
758 con->out_kvec_is_msg = true;
759 con->out_kvec[v].iov_base = &m->footer;
760 con->out_kvec[v].iov_len = sizeof(m->footer);
761 con->out_kvec_bytes += sizeof(m->footer);
762 con->out_kvec_left++;
763 con->out_more = m->more_to_follow;
764 con->out_msg_done = true;
768 * Prepare headers for the next outgoing message.
770 static void prepare_write_message(struct ceph_connection *con)
775 con_out_kvec_reset(con);
776 con->out_kvec_is_msg = true;
777 con->out_msg_done = false;
779 /* Sneak an ack in there first? If we can get it into the same
780 * TCP packet that's a good thing. */
781 if (con->in_seq > con->in_seq_acked) {
782 con->in_seq_acked = con->in_seq;
783 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
784 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
785 con_out_kvec_add(con, sizeof (con->out_temp_ack),
789 BUG_ON(list_empty(&con->out_queue));
790 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
792 BUG_ON(m->con != con);
794 /* put message on sent list */
796 list_move_tail(&m->list_head, &con->out_sent);
799 * only assign outgoing seq # if we haven't sent this message
800 * yet. if it is requeued, resend with it's original seq.
802 if (m->needs_out_seq) {
803 m->hdr.seq = cpu_to_le64(++con->out_seq);
804 m->needs_out_seq = false;
811 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n",
812 m, con->out_seq, le16_to_cpu(m->hdr.type),
813 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
814 le32_to_cpu(m->hdr.data_len),
816 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
818 /* tag + hdr + front + middle */
819 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
820 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
821 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
824 con_out_kvec_add(con, m->middle->vec.iov_len,
825 m->middle->vec.iov_base);
827 /* fill in crc (except data pages), footer */
828 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
829 con->out_msg->hdr.crc = cpu_to_le32(crc);
830 con->out_msg->footer.flags = 0;
832 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
833 con->out_msg->footer.front_crc = cpu_to_le32(crc);
835 crc = crc32c(0, m->middle->vec.iov_base,
836 m->middle->vec.iov_len);
837 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
839 con->out_msg->footer.middle_crc = 0;
840 dout("%s front_crc %u middle_crc %u\n", __func__,
841 le32_to_cpu(con->out_msg->footer.front_crc),
842 le32_to_cpu(con->out_msg->footer.middle_crc));
844 /* is there a data payload? */
845 con->out_msg->footer.data_crc = 0;
847 prepare_write_message_data(con);
849 /* no, queue up footer too and be done */
850 prepare_write_message_footer(con);
852 con_flag_set(con, CON_FLAG_WRITE_PENDING);
858 static void prepare_write_ack(struct ceph_connection *con)
860 dout("prepare_write_ack %p %llu -> %llu\n", con,
861 con->in_seq_acked, con->in_seq);
862 con->in_seq_acked = con->in_seq;
864 con_out_kvec_reset(con);
866 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
868 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
869 con_out_kvec_add(con, sizeof (con->out_temp_ack),
872 con->out_more = 1; /* more will follow.. eventually.. */
873 con_flag_set(con, CON_FLAG_WRITE_PENDING);
877 * Prepare to write keepalive byte.
879 static void prepare_write_keepalive(struct ceph_connection *con)
881 dout("prepare_write_keepalive %p\n", con);
882 con_out_kvec_reset(con);
883 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
884 con_flag_set(con, CON_FLAG_WRITE_PENDING);
888 * Connection negotiation.
891 static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
894 struct ceph_auth_handshake *auth;
896 if (!con->ops->get_authorizer) {
897 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
898 con->out_connect.authorizer_len = 0;
902 /* Can't hold the mutex while getting authorizer */
903 mutex_unlock(&con->mutex);
904 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
905 mutex_lock(&con->mutex);
909 if (con->state != CON_STATE_NEGOTIATING)
910 return ERR_PTR(-EAGAIN);
912 con->auth_reply_buf = auth->authorizer_reply_buf;
913 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
918 * We connected to a peer and are saying hello.
920 static void prepare_write_banner(struct ceph_connection *con)
922 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
923 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
924 &con->msgr->my_enc_addr);
927 con_flag_set(con, CON_FLAG_WRITE_PENDING);
930 static int prepare_write_connect(struct ceph_connection *con)
932 unsigned int global_seq = get_global_seq(con->msgr, 0);
935 struct ceph_auth_handshake *auth;
937 switch (con->peer_name.type) {
938 case CEPH_ENTITY_TYPE_MON:
939 proto = CEPH_MONC_PROTOCOL;
941 case CEPH_ENTITY_TYPE_OSD:
942 proto = CEPH_OSDC_PROTOCOL;
944 case CEPH_ENTITY_TYPE_MDS:
945 proto = CEPH_MDSC_PROTOCOL;
951 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
952 con->connect_seq, global_seq, proto);
954 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
955 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
956 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
957 con->out_connect.global_seq = cpu_to_le32(global_seq);
958 con->out_connect.protocol_version = cpu_to_le32(proto);
959 con->out_connect.flags = 0;
961 auth_proto = CEPH_AUTH_UNKNOWN;
962 auth = get_connect_authorizer(con, &auth_proto);
964 return PTR_ERR(auth);
966 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
967 con->out_connect.authorizer_len = auth ?
968 cpu_to_le32(auth->authorizer_buf_len) : 0;
970 con_out_kvec_add(con, sizeof (con->out_connect),
972 if (auth && auth->authorizer_buf_len)
973 con_out_kvec_add(con, auth->authorizer_buf_len,
974 auth->authorizer_buf);
977 con_flag_set(con, CON_FLAG_WRITE_PENDING);
983 * write as much of pending kvecs to the socket as we can.
985 * 0 -> socket full, but more to do
988 static int write_partial_kvec(struct ceph_connection *con)
992 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
993 while (con->out_kvec_bytes > 0) {
994 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
995 con->out_kvec_left, con->out_kvec_bytes,
999 con->out_kvec_bytes -= ret;
1000 if (con->out_kvec_bytes == 0)
1003 /* account for full iov entries consumed */
1004 while (ret >= con->out_kvec_cur->iov_len) {
1005 BUG_ON(!con->out_kvec_left);
1006 ret -= con->out_kvec_cur->iov_len;
1007 con->out_kvec_cur++;
1008 con->out_kvec_left--;
1010 /* and for a partially-consumed entry */
1012 con->out_kvec_cur->iov_len -= ret;
1013 con->out_kvec_cur->iov_base += ret;
1016 con->out_kvec_left = 0;
1017 con->out_kvec_is_msg = false;
1020 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1021 con->out_kvec_bytes, con->out_kvec_left, ret);
1022 return ret; /* done! */
1025 static void out_msg_pos_next(struct ceph_connection *con, struct page *page,
1026 size_t len, size_t sent, bool in_trail)
1028 struct ceph_msg *msg = con->out_msg;
1033 con->out_msg_pos.data_pos += sent;
1034 con->out_msg_pos.page_pos += sent;
1038 BUG_ON(sent != len);
1039 con->out_msg_pos.page_pos = 0;
1040 con->out_msg_pos.page++;
1041 con->out_msg_pos.did_page_crc = false;
1043 list_move_tail(&page->lru,
1045 else if (msg->pagelist)
1046 list_move_tail(&page->lru,
1047 &msg->pagelist->head);
1050 iter_bio_next(&msg->bio_iter, &msg->bio_seg);
1055 * Write as much message data payload as we can. If we finish, queue
1057 * 1 -> done, footer is now queued in out_kvec[].
1058 * 0 -> socket full, but more to do
1061 static int write_partial_msg_pages(struct ceph_connection *con)
1063 struct ceph_msg *msg = con->out_msg;
1064 unsigned int data_len = le32_to_cpu(msg->hdr.data_len);
1066 bool do_datacrc = !con->msgr->nocrc;
1068 int total_max_write;
1069 bool in_trail = false;
1070 const size_t trail_len = (msg->trail ? msg->trail->length : 0);
1071 const size_t trail_off = data_len - trail_len;
1073 dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
1074 con, msg, con->out_msg_pos.page, msg->nr_pages,
1075 con->out_msg_pos.page_pos);
1078 * Iterate through each page that contains data to be
1079 * written, and send as much as possible for each.
1081 * If we are calculating the data crc (the default), we will
1082 * need to map the page. If we have no pages, they have
1083 * been revoked, so use the zero page.
1085 while (data_len > con->out_msg_pos.data_pos) {
1086 struct page *page = NULL;
1087 int max_write = PAGE_SIZE;
1090 in_trail = in_trail || con->out_msg_pos.data_pos >= trail_off;
1092 total_max_write = trail_off - con->out_msg_pos.data_pos;
1095 total_max_write = data_len - con->out_msg_pos.data_pos;
1097 page = list_first_entry(&msg->trail->head,
1099 } else if (msg->pages) {
1100 page = msg->pages[con->out_msg_pos.page];
1101 } else if (msg->pagelist) {
1102 page = list_first_entry(&msg->pagelist->head,
1105 } else if (msg->bio) {
1108 bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg);
1110 bio_offset = bv->bv_offset;
1111 max_write = bv->bv_len;
1116 len = min_t(int, max_write - con->out_msg_pos.page_pos,
1119 if (do_datacrc && !con->out_msg_pos.did_page_crc) {
1121 u32 crc = le32_to_cpu(msg->footer.data_crc);
1125 BUG_ON(kaddr == NULL);
1126 base = kaddr + con->out_msg_pos.page_pos + bio_offset;
1127 crc = crc32c(crc, base, len);
1129 msg->footer.data_crc = cpu_to_le32(crc);
1130 con->out_msg_pos.did_page_crc = true;
1132 ret = ceph_tcp_sendpage(con->sock, page,
1133 con->out_msg_pos.page_pos + bio_offset,
1138 out_msg_pos_next(con, page, len, (size_t) ret, in_trail);
1141 dout("write_partial_msg_pages %p msg %p done\n", con, msg);
1143 /* prepare and queue up footer, too */
1145 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1146 con_out_kvec_reset(con);
1147 prepare_write_message_footer(con);
1156 static int write_partial_skip(struct ceph_connection *con)
1160 while (con->out_skip > 0) {
1161 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
1163 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, 1);
1166 con->out_skip -= ret;
1174 * Prepare to read connection handshake, or an ack.
1176 static void prepare_read_banner(struct ceph_connection *con)
1178 dout("prepare_read_banner %p\n", con);
1179 con->in_base_pos = 0;
1182 static void prepare_read_connect(struct ceph_connection *con)
1184 dout("prepare_read_connect %p\n", con);
1185 con->in_base_pos = 0;
1188 static void prepare_read_ack(struct ceph_connection *con)
1190 dout("prepare_read_ack %p\n", con);
1191 con->in_base_pos = 0;
1194 static void prepare_read_tag(struct ceph_connection *con)
1196 dout("prepare_read_tag %p\n", con);
1197 con->in_base_pos = 0;
1198 con->in_tag = CEPH_MSGR_TAG_READY;
1202 * Prepare to read a message.
1204 static int prepare_read_message(struct ceph_connection *con)
1206 dout("prepare_read_message %p\n", con);
1207 BUG_ON(con->in_msg != NULL);
1208 con->in_base_pos = 0;
1209 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1214 static int read_partial(struct ceph_connection *con,
1215 int end, int size, void *object)
1217 while (con->in_base_pos < end) {
1218 int left = end - con->in_base_pos;
1219 int have = size - left;
1220 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1223 con->in_base_pos += ret;
1230 * Read all or part of the connect-side handshake on a new connection
1232 static int read_partial_banner(struct ceph_connection *con)
1238 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1241 size = strlen(CEPH_BANNER);
1243 ret = read_partial(con, end, size, con->in_banner);
1247 size = sizeof (con->actual_peer_addr);
1249 ret = read_partial(con, end, size, &con->actual_peer_addr);
1253 size = sizeof (con->peer_addr_for_me);
1255 ret = read_partial(con, end, size, &con->peer_addr_for_me);
1263 static int read_partial_connect(struct ceph_connection *con)
1269 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1271 size = sizeof (con->in_reply);
1273 ret = read_partial(con, end, size, &con->in_reply);
1277 size = le32_to_cpu(con->in_reply.authorizer_len);
1279 ret = read_partial(con, end, size, con->auth_reply_buf);
1283 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1284 con, (int)con->in_reply.tag,
1285 le32_to_cpu(con->in_reply.connect_seq),
1286 le32_to_cpu(con->in_reply.global_seq));
1293 * Verify the hello banner looks okay.
1295 static int verify_hello(struct ceph_connection *con)
1297 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1298 pr_err("connect to %s got bad banner\n",
1299 ceph_pr_addr(&con->peer_addr.in_addr));
1300 con->error_msg = "protocol error, bad banner";
1306 static bool addr_is_blank(struct sockaddr_storage *ss)
1308 switch (ss->ss_family) {
1310 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1313 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1314 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1315 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1316 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1321 static int addr_port(struct sockaddr_storage *ss)
1323 switch (ss->ss_family) {
1325 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1327 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1332 static void addr_set_port(struct sockaddr_storage *ss, int p)
1334 switch (ss->ss_family) {
1336 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1339 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1345 * Unlike other *_pton function semantics, zero indicates success.
1347 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1348 char delim, const char **ipend)
1350 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1351 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1353 memset(ss, 0, sizeof(*ss));
1355 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1356 ss->ss_family = AF_INET;
1360 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1361 ss->ss_family = AF_INET6;
1369 * Extract hostname string and resolve using kernel DNS facility.
1371 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1372 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1373 struct sockaddr_storage *ss, char delim, const char **ipend)
1375 const char *end, *delim_p;
1376 char *colon_p, *ip_addr = NULL;
1380 * The end of the hostname occurs immediately preceding the delimiter or
1381 * the port marker (':') where the delimiter takes precedence.
1383 delim_p = memchr(name, delim, namelen);
1384 colon_p = memchr(name, ':', namelen);
1386 if (delim_p && colon_p)
1387 end = delim_p < colon_p ? delim_p : colon_p;
1388 else if (!delim_p && colon_p)
1392 if (!end) /* case: hostname:/ */
1393 end = name + namelen;
1399 /* do dns_resolve upcall */
1400 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1402 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1410 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1411 ret, ret ? "failed" : ceph_pr_addr(ss));
1416 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1417 struct sockaddr_storage *ss, char delim, const char **ipend)
1424 * Parse a server name (IP or hostname). If a valid IP address is not found
1425 * then try to extract a hostname to resolve using userspace DNS upcall.
1427 static int ceph_parse_server_name(const char *name, size_t namelen,
1428 struct sockaddr_storage *ss, char delim, const char **ipend)
1432 ret = ceph_pton(name, namelen, ss, delim, ipend);
1434 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1440 * Parse an ip[:port] list into an addr array. Use the default
1441 * monitor port if a port isn't specified.
1443 int ceph_parse_ips(const char *c, const char *end,
1444 struct ceph_entity_addr *addr,
1445 int max_count, int *count)
1447 int i, ret = -EINVAL;
1450 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1451 for (i = 0; i < max_count; i++) {
1453 struct sockaddr_storage *ss = &addr[i].in_addr;
1462 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1471 dout("missing matching ']'\n");
1478 if (p < end && *p == ':') {
1481 while (p < end && *p >= '0' && *p <= '9') {
1482 port = (port * 10) + (*p - '0');
1485 if (port > 65535 || port == 0)
1488 port = CEPH_MON_PORT;
1491 addr_set_port(ss, port);
1493 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1510 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1513 EXPORT_SYMBOL(ceph_parse_ips);
1515 static int process_banner(struct ceph_connection *con)
1517 dout("process_banner on %p\n", con);
1519 if (verify_hello(con) < 0)
1522 ceph_decode_addr(&con->actual_peer_addr);
1523 ceph_decode_addr(&con->peer_addr_for_me);
1526 * Make sure the other end is who we wanted. note that the other
1527 * end may not yet know their ip address, so if it's 0.0.0.0, give
1528 * them the benefit of the doubt.
1530 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1531 sizeof(con->peer_addr)) != 0 &&
1532 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1533 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1534 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1535 ceph_pr_addr(&con->peer_addr.in_addr),
1536 (int)le32_to_cpu(con->peer_addr.nonce),
1537 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1538 (int)le32_to_cpu(con->actual_peer_addr.nonce));
1539 con->error_msg = "wrong peer at address";
1544 * did we learn our address?
1546 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1547 int port = addr_port(&con->msgr->inst.addr.in_addr);
1549 memcpy(&con->msgr->inst.addr.in_addr,
1550 &con->peer_addr_for_me.in_addr,
1551 sizeof(con->peer_addr_for_me.in_addr));
1552 addr_set_port(&con->msgr->inst.addr.in_addr, port);
1553 encode_my_addr(con->msgr);
1554 dout("process_banner learned my addr is %s\n",
1555 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
1561 static int process_connect(struct ceph_connection *con)
1563 u64 sup_feat = con->msgr->supported_features;
1564 u64 req_feat = con->msgr->required_features;
1565 u64 server_feat = le64_to_cpu(con->in_reply.features);
1568 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1570 switch (con->in_reply.tag) {
1571 case CEPH_MSGR_TAG_FEATURES:
1572 pr_err("%s%lld %s feature set mismatch,"
1573 " my %llx < server's %llx, missing %llx\n",
1574 ENTITY_NAME(con->peer_name),
1575 ceph_pr_addr(&con->peer_addr.in_addr),
1576 sup_feat, server_feat, server_feat & ~sup_feat);
1577 con->error_msg = "missing required protocol features";
1578 reset_connection(con);
1581 case CEPH_MSGR_TAG_BADPROTOVER:
1582 pr_err("%s%lld %s protocol version mismatch,"
1583 " my %d != server's %d\n",
1584 ENTITY_NAME(con->peer_name),
1585 ceph_pr_addr(&con->peer_addr.in_addr),
1586 le32_to_cpu(con->out_connect.protocol_version),
1587 le32_to_cpu(con->in_reply.protocol_version));
1588 con->error_msg = "protocol version mismatch";
1589 reset_connection(con);
1592 case CEPH_MSGR_TAG_BADAUTHORIZER:
1594 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1596 if (con->auth_retry == 2) {
1597 con->error_msg = "connect authorization failure";
1600 con->auth_retry = 1;
1601 con_out_kvec_reset(con);
1602 ret = prepare_write_connect(con);
1605 prepare_read_connect(con);
1608 case CEPH_MSGR_TAG_RESETSESSION:
1610 * If we connected with a large connect_seq but the peer
1611 * has no record of a session with us (no connection, or
1612 * connect_seq == 0), they will send RESETSESION to indicate
1613 * that they must have reset their session, and may have
1616 dout("process_connect got RESET peer seq %u\n",
1617 le32_to_cpu(con->in_reply.connect_seq));
1618 pr_err("%s%lld %s connection reset\n",
1619 ENTITY_NAME(con->peer_name),
1620 ceph_pr_addr(&con->peer_addr.in_addr));
1621 reset_connection(con);
1622 con_out_kvec_reset(con);
1623 ret = prepare_write_connect(con);
1626 prepare_read_connect(con);
1628 /* Tell ceph about it. */
1629 mutex_unlock(&con->mutex);
1630 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
1631 if (con->ops->peer_reset)
1632 con->ops->peer_reset(con);
1633 mutex_lock(&con->mutex);
1634 if (con->state != CON_STATE_NEGOTIATING)
1638 case CEPH_MSGR_TAG_RETRY_SESSION:
1640 * If we sent a smaller connect_seq than the peer has, try
1641 * again with a larger value.
1643 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
1644 le32_to_cpu(con->out_connect.connect_seq),
1645 le32_to_cpu(con->in_reply.connect_seq));
1646 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
1647 con_out_kvec_reset(con);
1648 ret = prepare_write_connect(con);
1651 prepare_read_connect(con);
1654 case CEPH_MSGR_TAG_RETRY_GLOBAL:
1656 * If we sent a smaller global_seq than the peer has, try
1657 * again with a larger value.
1659 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1660 con->peer_global_seq,
1661 le32_to_cpu(con->in_reply.global_seq));
1662 get_global_seq(con->msgr,
1663 le32_to_cpu(con->in_reply.global_seq));
1664 con_out_kvec_reset(con);
1665 ret = prepare_write_connect(con);
1668 prepare_read_connect(con);
1671 case CEPH_MSGR_TAG_READY:
1672 if (req_feat & ~server_feat) {
1673 pr_err("%s%lld %s protocol feature mismatch,"
1674 " my required %llx > server's %llx, need %llx\n",
1675 ENTITY_NAME(con->peer_name),
1676 ceph_pr_addr(&con->peer_addr.in_addr),
1677 req_feat, server_feat, req_feat & ~server_feat);
1678 con->error_msg = "missing required protocol features";
1679 reset_connection(con);
1683 WARN_ON(con->state != CON_STATE_NEGOTIATING);
1684 con->state = CON_STATE_OPEN;
1686 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
1688 con->peer_features = server_feat;
1689 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1690 con->peer_global_seq,
1691 le32_to_cpu(con->in_reply.connect_seq),
1693 WARN_ON(con->connect_seq !=
1694 le32_to_cpu(con->in_reply.connect_seq));
1696 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
1697 con_flag_set(con, CON_FLAG_LOSSYTX);
1699 con->delay = 0; /* reset backoff memory */
1701 prepare_read_tag(con);
1704 case CEPH_MSGR_TAG_WAIT:
1706 * If there is a connection race (we are opening
1707 * connections to each other), one of us may just have
1708 * to WAIT. This shouldn't happen if we are the
1711 pr_err("process_connect got WAIT as client\n");
1712 con->error_msg = "protocol error, got WAIT as client";
1716 pr_err("connect protocol error, will retry\n");
1717 con->error_msg = "protocol error, garbage tag during connect";
1725 * read (part of) an ack
1727 static int read_partial_ack(struct ceph_connection *con)
1729 int size = sizeof (con->in_temp_ack);
1732 return read_partial(con, end, size, &con->in_temp_ack);
1737 * We can finally discard anything that's been acked.
1739 static void process_ack(struct ceph_connection *con)
1742 u64 ack = le64_to_cpu(con->in_temp_ack);
1745 while (!list_empty(&con->out_sent)) {
1746 m = list_first_entry(&con->out_sent, struct ceph_msg,
1748 seq = le64_to_cpu(m->hdr.seq);
1751 dout("got ack for seq %llu type %d at %p\n", seq,
1752 le16_to_cpu(m->hdr.type), m);
1753 m->ack_stamp = jiffies;
1756 prepare_read_tag(con);
1762 static int read_partial_message_section(struct ceph_connection *con,
1763 struct kvec *section,
1764 unsigned int sec_len, u32 *crc)
1770 while (section->iov_len < sec_len) {
1771 BUG_ON(section->iov_base == NULL);
1772 left = sec_len - section->iov_len;
1773 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
1774 section->iov_len, left);
1777 section->iov_len += ret;
1779 if (section->iov_len == sec_len)
1780 *crc = crc32c(0, section->iov_base, section->iov_len);
1785 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
1787 static int read_partial_message_pages(struct ceph_connection *con,
1788 struct page **pages,
1789 unsigned int data_len, bool do_datacrc)
1795 left = min((int)(data_len - con->in_msg_pos.data_pos),
1796 (int)(PAGE_SIZE - con->in_msg_pos.page_pos));
1798 BUG_ON(pages == NULL);
1799 p = kmap(pages[con->in_msg_pos.page]);
1800 ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos,
1802 if (ret > 0 && do_datacrc)
1804 crc32c(con->in_data_crc,
1805 p + con->in_msg_pos.page_pos, ret);
1806 kunmap(pages[con->in_msg_pos.page]);
1809 con->in_msg_pos.data_pos += ret;
1810 con->in_msg_pos.page_pos += ret;
1811 if (con->in_msg_pos.page_pos == PAGE_SIZE) {
1812 con->in_msg_pos.page_pos = 0;
1813 con->in_msg_pos.page++;
1820 static int read_partial_message_bio(struct ceph_connection *con,
1821 struct bio **bio_iter, int *bio_seg,
1822 unsigned int data_len, bool do_datacrc)
1824 struct bio_vec *bv = bio_iovec_idx(*bio_iter, *bio_seg);
1828 left = min((int)(data_len - con->in_msg_pos.data_pos),
1829 (int)(bv->bv_len - con->in_msg_pos.page_pos));
1831 p = kmap(bv->bv_page) + bv->bv_offset;
1833 ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos,
1835 if (ret > 0 && do_datacrc)
1837 crc32c(con->in_data_crc,
1838 p + con->in_msg_pos.page_pos, ret);
1839 kunmap(bv->bv_page);
1842 con->in_msg_pos.data_pos += ret;
1843 con->in_msg_pos.page_pos += ret;
1844 if (con->in_msg_pos.page_pos == bv->bv_len) {
1845 con->in_msg_pos.page_pos = 0;
1846 iter_bio_next(bio_iter, bio_seg);
1854 * read (part of) a message.
1856 static int read_partial_message(struct ceph_connection *con)
1858 struct ceph_msg *m = con->in_msg;
1862 unsigned int front_len, middle_len, data_len;
1863 bool do_datacrc = !con->msgr->nocrc;
1867 dout("read_partial_message con %p msg %p\n", con, m);
1870 size = sizeof (con->in_hdr);
1872 ret = read_partial(con, end, size, &con->in_hdr);
1876 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
1877 if (cpu_to_le32(crc) != con->in_hdr.crc) {
1878 pr_err("read_partial_message bad hdr "
1879 " crc %u != expected %u\n",
1880 crc, con->in_hdr.crc);
1884 front_len = le32_to_cpu(con->in_hdr.front_len);
1885 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
1887 middle_len = le32_to_cpu(con->in_hdr.middle_len);
1888 if (middle_len > CEPH_MSG_MAX_DATA_LEN)
1890 data_len = le32_to_cpu(con->in_hdr.data_len);
1891 if (data_len > CEPH_MSG_MAX_DATA_LEN)
1895 seq = le64_to_cpu(con->in_hdr.seq);
1896 if ((s64)seq - (s64)con->in_seq < 1) {
1897 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1898 ENTITY_NAME(con->peer_name),
1899 ceph_pr_addr(&con->peer_addr.in_addr),
1900 seq, con->in_seq + 1);
1901 con->in_base_pos = -front_len - middle_len - data_len -
1903 con->in_tag = CEPH_MSGR_TAG_READY;
1905 } else if ((s64)seq - (s64)con->in_seq > 1) {
1906 pr_err("read_partial_message bad seq %lld expected %lld\n",
1907 seq, con->in_seq + 1);
1908 con->error_msg = "bad message sequence # for incoming message";
1912 /* allocate message? */
1916 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
1917 con->in_hdr.front_len, con->in_hdr.data_len);
1918 ret = ceph_con_in_msg_alloc(con, &skip);
1922 /* skip this message */
1923 dout("alloc_msg said skip message\n");
1924 BUG_ON(con->in_msg);
1925 con->in_base_pos = -front_len - middle_len - data_len -
1927 con->in_tag = CEPH_MSGR_TAG_READY;
1932 BUG_ON(!con->in_msg);
1933 BUG_ON(con->in_msg->con != con);
1935 m->front.iov_len = 0; /* haven't read it yet */
1937 m->middle->vec.iov_len = 0;
1939 con->in_msg_pos.page = 0;
1941 con->in_msg_pos.page_pos = m->page_alignment;
1943 con->in_msg_pos.page_pos = 0;
1944 con->in_msg_pos.data_pos = 0;
1948 init_bio_iter(m->bio, &m->bio_iter, &m->bio_seg);
1953 ret = read_partial_message_section(con, &m->front, front_len,
1954 &con->in_front_crc);
1960 ret = read_partial_message_section(con, &m->middle->vec,
1962 &con->in_middle_crc);
1968 while (con->in_msg_pos.data_pos < data_len) {
1970 ret = read_partial_message_pages(con, m->pages,
1971 data_len, do_datacrc);
1975 } else if (m->bio) {
1976 BUG_ON(!m->bio_iter);
1977 ret = read_partial_message_bio(con,
1978 &m->bio_iter, &m->bio_seg,
1979 data_len, do_datacrc);
1989 size = sizeof (m->footer);
1991 ret = read_partial(con, end, size, &m->footer);
1995 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1996 m, front_len, m->footer.front_crc, middle_len,
1997 m->footer.middle_crc, data_len, m->footer.data_crc);
2000 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2001 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2002 m, con->in_front_crc, m->footer.front_crc);
2005 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2006 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2007 m, con->in_middle_crc, m->footer.middle_crc);
2011 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2012 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2013 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2014 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2018 return 1; /* done! */
2022 * Process message. This happens in the worker thread. The callback should
2023 * be careful not to do anything that waits on other incoming messages or it
2026 static void process_message(struct ceph_connection *con)
2028 struct ceph_msg *msg;
2030 BUG_ON(con->in_msg->con != con);
2031 con->in_msg->con = NULL;
2036 /* if first message, set peer_name */
2037 if (con->peer_name.type == 0)
2038 con->peer_name = msg->hdr.src;
2041 mutex_unlock(&con->mutex);
2043 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2044 msg, le64_to_cpu(msg->hdr.seq),
2045 ENTITY_NAME(msg->hdr.src),
2046 le16_to_cpu(msg->hdr.type),
2047 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2048 le32_to_cpu(msg->hdr.front_len),
2049 le32_to_cpu(msg->hdr.data_len),
2050 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2051 con->ops->dispatch(con, msg);
2053 mutex_lock(&con->mutex);
2058 * Write something to the socket. Called in a worker thread when the
2059 * socket appears to be writeable and we have something ready to send.
2061 static int try_write(struct ceph_connection *con)
2065 dout("try_write start %p state %lu\n", con, con->state);
2068 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2070 /* open the socket first? */
2071 if (con->state == CON_STATE_PREOPEN) {
2073 con->state = CON_STATE_CONNECTING;
2075 con_out_kvec_reset(con);
2076 prepare_write_banner(con);
2077 prepare_read_banner(con);
2079 BUG_ON(con->in_msg);
2080 con->in_tag = CEPH_MSGR_TAG_READY;
2081 dout("try_write initiating connect on %p new state %lu\n",
2083 ret = ceph_tcp_connect(con);
2085 con->error_msg = "connect error";
2091 /* kvec data queued? */
2092 if (con->out_skip) {
2093 ret = write_partial_skip(con);
2097 if (con->out_kvec_left) {
2098 ret = write_partial_kvec(con);
2105 if (con->out_msg_done) {
2106 ceph_msg_put(con->out_msg);
2107 con->out_msg = NULL; /* we're done with this one */
2111 ret = write_partial_msg_pages(con);
2113 goto more_kvec; /* we need to send the footer, too! */
2117 dout("try_write write_partial_msg_pages err %d\n",
2124 if (con->state == CON_STATE_OPEN) {
2125 /* is anything else pending? */
2126 if (!list_empty(&con->out_queue)) {
2127 prepare_write_message(con);
2130 if (con->in_seq > con->in_seq_acked) {
2131 prepare_write_ack(con);
2134 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2135 prepare_write_keepalive(con);
2140 /* Nothing to do! */
2141 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2142 dout("try_write nothing else to write.\n");
2145 dout("try_write done on %p ret %d\n", con, ret);
2152 * Read what we can from the socket.
2154 static int try_read(struct ceph_connection *con)
2159 dout("try_read start on %p state %lu\n", con, con->state);
2160 if (con->state != CON_STATE_CONNECTING &&
2161 con->state != CON_STATE_NEGOTIATING &&
2162 con->state != CON_STATE_OPEN)
2167 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2170 if (con->state == CON_STATE_CONNECTING) {
2171 dout("try_read connecting\n");
2172 ret = read_partial_banner(con);
2175 ret = process_banner(con);
2179 con->state = CON_STATE_NEGOTIATING;
2182 * Received banner is good, exchange connection info.
2183 * Do not reset out_kvec, as sending our banner raced
2184 * with receiving peer banner after connect completed.
2186 ret = prepare_write_connect(con);
2189 prepare_read_connect(con);
2191 /* Send connection info before awaiting response */
2195 if (con->state == CON_STATE_NEGOTIATING) {
2196 dout("try_read negotiating\n");
2197 ret = read_partial_connect(con);
2200 ret = process_connect(con);
2206 WARN_ON(con->state != CON_STATE_OPEN);
2208 if (con->in_base_pos < 0) {
2210 * skipping + discarding content.
2212 * FIXME: there must be a better way to do this!
2214 static char buf[SKIP_BUF_SIZE];
2215 int skip = min((int) sizeof (buf), -con->in_base_pos);
2217 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2218 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2221 con->in_base_pos += ret;
2222 if (con->in_base_pos)
2225 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2229 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2232 dout("try_read got tag %d\n", (int)con->in_tag);
2233 switch (con->in_tag) {
2234 case CEPH_MSGR_TAG_MSG:
2235 prepare_read_message(con);
2237 case CEPH_MSGR_TAG_ACK:
2238 prepare_read_ack(con);
2240 case CEPH_MSGR_TAG_CLOSE:
2241 con_close_socket(con);
2242 con->state = CON_STATE_CLOSED;
2248 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2249 ret = read_partial_message(con);
2253 con->error_msg = "bad crc";
2257 con->error_msg = "io error";
2262 if (con->in_tag == CEPH_MSGR_TAG_READY)
2264 process_message(con);
2265 if (con->state == CON_STATE_OPEN)
2266 prepare_read_tag(con);
2269 if (con->in_tag == CEPH_MSGR_TAG_ACK) {
2270 ret = read_partial_ack(con);
2278 dout("try_read done on %p ret %d\n", con, ret);
2282 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2283 con->error_msg = "protocol error, garbage tag";
2290 * Atomically queue work on a connection after the specified delay.
2291 * Bump @con reference to avoid races with connection teardown.
2292 * Returns 0 if work was queued, or an error code otherwise.
2294 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2296 if (!con->ops->get(con)) {
2297 dout("%s %p ref count 0\n", __func__, con);
2302 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2303 dout("%s %p - already queued\n", __func__, con);
2309 dout("%s %p %lu\n", __func__, con, delay);
2314 static void queue_con(struct ceph_connection *con)
2316 (void) queue_con_delay(con, 0);
2319 static bool con_sock_closed(struct ceph_connection *con)
2321 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2325 case CON_STATE_ ## x: \
2326 con->error_msg = "socket closed (con state " #x ")"; \
2329 switch (con->state) {
2337 pr_warning("%s con %p unrecognized state %lu\n",
2338 __func__, con, con->state);
2339 con->error_msg = "unrecognized con state";
2348 static bool con_backoff(struct ceph_connection *con)
2352 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2355 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2357 dout("%s: con %p FAILED to back off %lu\n", __func__,
2359 BUG_ON(ret == -ENOENT);
2360 con_flag_set(con, CON_FLAG_BACKOFF);
2366 /* Finish fault handling; con->mutex must *not* be held here */
2368 static void con_fault_finish(struct ceph_connection *con)
2371 * in case we faulted due to authentication, invalidate our
2372 * current tickets so that we can get new ones.
2374 if (con->auth_retry && con->ops->invalidate_authorizer) {
2375 dout("calling invalidate_authorizer()\n");
2376 con->ops->invalidate_authorizer(con);
2379 if (con->ops->fault)
2380 con->ops->fault(con);
2384 * Do some work on a connection. Drop a connection ref when we're done.
2386 static void con_work(struct work_struct *work)
2388 struct ceph_connection *con = container_of(work, struct ceph_connection,
2392 mutex_lock(&con->mutex);
2396 if ((fault = con_sock_closed(con))) {
2397 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2400 if (con_backoff(con)) {
2401 dout("%s: con %p BACKOFF\n", __func__, con);
2404 if (con->state == CON_STATE_STANDBY) {
2405 dout("%s: con %p STANDBY\n", __func__, con);
2408 if (con->state == CON_STATE_CLOSED) {
2409 dout("%s: con %p CLOSED\n", __func__, con);
2413 if (con->state == CON_STATE_PREOPEN) {
2414 dout("%s: con %p PREOPEN\n", __func__, con);
2418 ret = try_read(con);
2422 con->error_msg = "socket error on read";
2427 ret = try_write(con);
2431 con->error_msg = "socket error on write";
2435 break; /* If we make it to here, we're done */
2439 mutex_unlock(&con->mutex);
2442 con_fault_finish(con);
2448 * Generic error/fault handler. A retry mechanism is used with
2449 * exponential backoff
2451 static void con_fault(struct ceph_connection *con)
2453 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2454 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2455 dout("fault %p state %lu to peer %s\n",
2456 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2458 WARN_ON(con->state != CON_STATE_CONNECTING &&
2459 con->state != CON_STATE_NEGOTIATING &&
2460 con->state != CON_STATE_OPEN);
2462 con_close_socket(con);
2464 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2465 dout("fault on LOSSYTX channel, marking CLOSED\n");
2466 con->state = CON_STATE_CLOSED;
2471 BUG_ON(con->in_msg->con != con);
2472 con->in_msg->con = NULL;
2473 ceph_msg_put(con->in_msg);
2478 /* Requeue anything that hasn't been acked */
2479 list_splice_init(&con->out_sent, &con->out_queue);
2481 /* If there are no messages queued or keepalive pending, place
2482 * the connection in a STANDBY state */
2483 if (list_empty(&con->out_queue) &&
2484 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
2485 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2486 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2487 con->state = CON_STATE_STANDBY;
2489 /* retry after a delay. */
2490 con->state = CON_STATE_PREOPEN;
2491 if (con->delay == 0)
2492 con->delay = BASE_DELAY_INTERVAL;
2493 else if (con->delay < MAX_DELAY_INTERVAL)
2495 con_flag_set(con, CON_FLAG_BACKOFF);
2503 * initialize a new messenger instance
2505 void ceph_messenger_init(struct ceph_messenger *msgr,
2506 struct ceph_entity_addr *myaddr,
2507 u32 supported_features,
2508 u32 required_features,
2511 msgr->supported_features = supported_features;
2512 msgr->required_features = required_features;
2514 spin_lock_init(&msgr->global_seq_lock);
2517 msgr->inst.addr = *myaddr;
2519 /* select a random nonce */
2520 msgr->inst.addr.type = 0;
2521 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2522 encode_my_addr(msgr);
2523 msgr->nocrc = nocrc;
2525 atomic_set(&msgr->stopping, 0);
2527 dout("%s %p\n", __func__, msgr);
2529 EXPORT_SYMBOL(ceph_messenger_init);
2531 static void clear_standby(struct ceph_connection *con)
2533 /* come back from STANDBY? */
2534 if (con->state == CON_STATE_STANDBY) {
2535 dout("clear_standby %p and ++connect_seq\n", con);
2536 con->state = CON_STATE_PREOPEN;
2538 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2539 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
2544 * Queue up an outgoing message on the given connection.
2546 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2549 msg->hdr.src = con->msgr->inst.name;
2550 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2551 msg->needs_out_seq = true;
2553 mutex_lock(&con->mutex);
2555 if (con->state == CON_STATE_CLOSED) {
2556 dout("con_send %p closed, dropping %p\n", con, msg);
2558 mutex_unlock(&con->mutex);
2562 BUG_ON(msg->con != NULL);
2563 msg->con = con->ops->get(con);
2564 BUG_ON(msg->con == NULL);
2566 BUG_ON(!list_empty(&msg->list_head));
2567 list_add_tail(&msg->list_head, &con->out_queue);
2568 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2569 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2570 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2571 le32_to_cpu(msg->hdr.front_len),
2572 le32_to_cpu(msg->hdr.middle_len),
2573 le32_to_cpu(msg->hdr.data_len));
2576 mutex_unlock(&con->mutex);
2578 /* if there wasn't anything waiting to send before, queue
2580 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
2583 EXPORT_SYMBOL(ceph_con_send);
2586 * Revoke a message that was previously queued for send
2588 void ceph_msg_revoke(struct ceph_msg *msg)
2590 struct ceph_connection *con = msg->con;
2593 return; /* Message not in our possession */
2595 mutex_lock(&con->mutex);
2596 if (!list_empty(&msg->list_head)) {
2597 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
2598 list_del_init(&msg->list_head);
2599 BUG_ON(msg->con == NULL);
2600 msg->con->ops->put(msg->con);
2606 if (con->out_msg == msg) {
2607 dout("%s %p msg %p - was sending\n", __func__, con, msg);
2608 con->out_msg = NULL;
2609 if (con->out_kvec_is_msg) {
2610 con->out_skip = con->out_kvec_bytes;
2611 con->out_kvec_is_msg = false;
2617 mutex_unlock(&con->mutex);
2621 * Revoke a message that we may be reading data into
2623 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
2625 struct ceph_connection *con;
2627 BUG_ON(msg == NULL);
2629 dout("%s msg %p null con\n", __func__, msg);
2631 return; /* Message not in our possession */
2635 mutex_lock(&con->mutex);
2636 if (con->in_msg == msg) {
2637 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
2638 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
2639 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
2641 /* skip rest of message */
2642 dout("%s %p msg %p revoked\n", __func__, con, msg);
2643 con->in_base_pos = con->in_base_pos -
2644 sizeof(struct ceph_msg_header) -
2648 sizeof(struct ceph_msg_footer);
2649 ceph_msg_put(con->in_msg);
2651 con->in_tag = CEPH_MSGR_TAG_READY;
2654 dout("%s %p in_msg %p msg %p no-op\n",
2655 __func__, con, con->in_msg, msg);
2657 mutex_unlock(&con->mutex);
2661 * Queue a keepalive byte to ensure the tcp connection is alive.
2663 void ceph_con_keepalive(struct ceph_connection *con)
2665 dout("con_keepalive %p\n", con);
2666 mutex_lock(&con->mutex);
2668 mutex_unlock(&con->mutex);
2669 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
2670 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
2673 EXPORT_SYMBOL(ceph_con_keepalive);
2677 * construct a new message with given type, size
2678 * the new msg has a ref count of 1.
2680 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
2685 m = kmalloc(sizeof(*m), flags);
2688 kref_init(&m->kref);
2691 INIT_LIST_HEAD(&m->list_head);
2694 m->hdr.type = cpu_to_le16(type);
2695 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
2697 m->hdr.front_len = cpu_to_le32(front_len);
2698 m->hdr.middle_len = 0;
2699 m->hdr.data_len = 0;
2700 m->hdr.data_off = 0;
2701 m->hdr.reserved = 0;
2702 m->footer.front_crc = 0;
2703 m->footer.middle_crc = 0;
2704 m->footer.data_crc = 0;
2705 m->footer.flags = 0;
2706 m->front_max = front_len;
2707 m->front_is_vmalloc = false;
2708 m->more_to_follow = false;
2717 m->page_alignment = 0;
2724 #endif /* CONFIG_BLOCK */
2729 if (front_len > PAGE_CACHE_SIZE) {
2730 m->front.iov_base = __vmalloc(front_len, flags,
2732 m->front_is_vmalloc = true;
2734 m->front.iov_base = kmalloc(front_len, flags);
2736 if (m->front.iov_base == NULL) {
2737 dout("ceph_msg_new can't allocate %d bytes\n",
2742 m->front.iov_base = NULL;
2744 m->front.iov_len = front_len;
2746 dout("ceph_msg_new %p front %d\n", m, front_len);
2753 pr_err("msg_new can't create type %d front %d\n", type,
2757 dout("msg_new can't create type %d front %d\n", type,
2762 EXPORT_SYMBOL(ceph_msg_new);
2765 * Allocate "middle" portion of a message, if it is needed and wasn't
2766 * allocated by alloc_msg. This allows us to read a small fixed-size
2767 * per-type header in the front and then gracefully fail (i.e.,
2768 * propagate the error to the caller based on info in the front) when
2769 * the middle is too large.
2771 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
2773 int type = le16_to_cpu(msg->hdr.type);
2774 int middle_len = le32_to_cpu(msg->hdr.middle_len);
2776 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
2777 ceph_msg_type_name(type), middle_len);
2778 BUG_ON(!middle_len);
2779 BUG_ON(msg->middle);
2781 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
2788 * Allocate a message for receiving an incoming message on a
2789 * connection, and save the result in con->in_msg. Uses the
2790 * connection's private alloc_msg op if available.
2792 * Returns 0 on success, or a negative error code.
2794 * On success, if we set *skip = 1:
2795 * - the next message should be skipped and ignored.
2796 * - con->in_msg == NULL
2797 * or if we set *skip = 0:
2798 * - con->in_msg is non-null.
2799 * On error (ENOMEM, EAGAIN, ...),
2800 * - con->in_msg == NULL
2802 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
2804 struct ceph_msg_header *hdr = &con->in_hdr;
2805 int type = le16_to_cpu(hdr->type);
2806 int front_len = le32_to_cpu(hdr->front_len);
2807 int middle_len = le32_to_cpu(hdr->middle_len);
2810 BUG_ON(con->in_msg != NULL);
2812 if (con->ops->alloc_msg) {
2813 struct ceph_msg *msg;
2815 mutex_unlock(&con->mutex);
2816 msg = con->ops->alloc_msg(con, hdr, skip);
2817 mutex_lock(&con->mutex);
2818 if (con->state != CON_STATE_OPEN) {
2825 con->in_msg->con = con->ops->get(con);
2826 BUG_ON(con->in_msg->con == NULL);
2834 "error allocating memory for incoming message";
2839 con->in_msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
2841 pr_err("unable to allocate msg type %d len %d\n",
2845 con->in_msg->con = con->ops->get(con);
2846 BUG_ON(con->in_msg->con == NULL);
2847 con->in_msg->page_alignment = le16_to_cpu(hdr->data_off);
2849 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2851 if (middle_len && !con->in_msg->middle) {
2852 ret = ceph_alloc_middle(con, con->in_msg);
2854 ceph_msg_put(con->in_msg);
2864 * Free a generically kmalloc'd message.
2866 void ceph_msg_kfree(struct ceph_msg *m)
2868 dout("msg_kfree %p\n", m);
2869 if (m->front_is_vmalloc)
2870 vfree(m->front.iov_base);
2872 kfree(m->front.iov_base);
2877 * Drop a msg ref. Destroy as needed.
2879 void ceph_msg_last_put(struct kref *kref)
2881 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
2883 dout("ceph_msg_put last one on %p\n", m);
2884 WARN_ON(!list_empty(&m->list_head));
2886 /* drop middle, data, if any */
2888 ceph_buffer_put(m->middle);
2895 ceph_pagelist_release(m->pagelist);
2903 ceph_msgpool_put(m->pool, m);
2907 EXPORT_SYMBOL(ceph_msg_last_put);
2909 void ceph_msg_dump(struct ceph_msg *msg)
2911 pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg,
2912 msg->front_max, msg->nr_pages);
2913 print_hex_dump(KERN_DEBUG, "header: ",
2914 DUMP_PREFIX_OFFSET, 16, 1,
2915 &msg->hdr, sizeof(msg->hdr), true);
2916 print_hex_dump(KERN_DEBUG, " front: ",
2917 DUMP_PREFIX_OFFSET, 16, 1,
2918 msg->front.iov_base, msg->front.iov_len, true);
2920 print_hex_dump(KERN_DEBUG, "middle: ",
2921 DUMP_PREFIX_OFFSET, 16, 1,
2922 msg->middle->vec.iov_base,
2923 msg->middle->vec.iov_len, true);
2924 print_hex_dump(KERN_DEBUG, "footer: ",
2925 DUMP_PREFIX_OFFSET, 16, 1,
2926 &msg->footer, sizeof(msg->footer), true);
2928 EXPORT_SYMBOL(ceph_msg_dump);