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, bool 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 **bio_iter,
701 unsigned int *bio_seg)
709 *bio_seg = (unsigned int) bio->bi_idx;
712 static void iter_bio_next(struct bio **bio_iter, unsigned int *seg)
714 if (*bio_iter == NULL)
717 BUG_ON(*seg >= (*bio_iter)->bi_vcnt);
720 if (*seg == (*bio_iter)->bi_vcnt)
721 init_bio_iter((*bio_iter)->bi_next, bio_iter, seg);
725 static void prepare_write_message_data(struct ceph_connection *con)
727 struct ceph_msg *msg = con->out_msg;
730 BUG_ON(!msg->hdr.data_len);
732 /* initialize page iterator */
733 con->out_msg_pos.page = 0;
735 con->out_msg_pos.page_pos = msg->page_alignment;
737 con->out_msg_pos.page_pos = 0;
740 init_bio_iter(msg->bio, &msg->bio_iter, &msg->bio_seg);
742 con->out_msg_pos.data_pos = 0;
743 con->out_msg_pos.did_page_crc = false;
744 con->out_more = 1; /* data + footer will follow */
748 * Prepare footer for currently outgoing message, and finish things
749 * off. Assumes out_kvec* are already valid.. we just add on to the end.
751 static void prepare_write_message_footer(struct ceph_connection *con)
753 struct ceph_msg *m = con->out_msg;
754 int v = con->out_kvec_left;
756 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
758 dout("prepare_write_message_footer %p\n", con);
759 con->out_kvec_is_msg = true;
760 con->out_kvec[v].iov_base = &m->footer;
761 con->out_kvec[v].iov_len = sizeof(m->footer);
762 con->out_kvec_bytes += sizeof(m->footer);
763 con->out_kvec_left++;
764 con->out_more = m->more_to_follow;
765 con->out_msg_done = true;
769 * Prepare headers for the next outgoing message.
771 static void prepare_write_message(struct ceph_connection *con)
776 con_out_kvec_reset(con);
777 con->out_kvec_is_msg = true;
778 con->out_msg_done = false;
780 /* Sneak an ack in there first? If we can get it into the same
781 * TCP packet that's a good thing. */
782 if (con->in_seq > con->in_seq_acked) {
783 con->in_seq_acked = con->in_seq;
784 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
785 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
786 con_out_kvec_add(con, sizeof (con->out_temp_ack),
790 BUG_ON(list_empty(&con->out_queue));
791 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
793 BUG_ON(m->con != con);
795 /* put message on sent list */
797 list_move_tail(&m->list_head, &con->out_sent);
800 * only assign outgoing seq # if we haven't sent this message
801 * yet. if it is requeued, resend with it's original seq.
803 if (m->needs_out_seq) {
804 m->hdr.seq = cpu_to_le64(++con->out_seq);
805 m->needs_out_seq = false;
812 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n",
813 m, con->out_seq, le16_to_cpu(m->hdr.type),
814 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
815 le32_to_cpu(m->hdr.data_len),
817 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
819 /* tag + hdr + front + middle */
820 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
821 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
822 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
825 con_out_kvec_add(con, m->middle->vec.iov_len,
826 m->middle->vec.iov_base);
828 /* fill in crc (except data pages), footer */
829 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
830 con->out_msg->hdr.crc = cpu_to_le32(crc);
831 con->out_msg->footer.flags = 0;
833 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
834 con->out_msg->footer.front_crc = cpu_to_le32(crc);
836 crc = crc32c(0, m->middle->vec.iov_base,
837 m->middle->vec.iov_len);
838 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
840 con->out_msg->footer.middle_crc = 0;
841 dout("%s front_crc %u middle_crc %u\n", __func__,
842 le32_to_cpu(con->out_msg->footer.front_crc),
843 le32_to_cpu(con->out_msg->footer.middle_crc));
845 /* is there a data payload? */
846 con->out_msg->footer.data_crc = 0;
848 prepare_write_message_data(con);
850 /* no, queue up footer too and be done */
851 prepare_write_message_footer(con);
853 con_flag_set(con, CON_FLAG_WRITE_PENDING);
859 static void prepare_write_ack(struct ceph_connection *con)
861 dout("prepare_write_ack %p %llu -> %llu\n", con,
862 con->in_seq_acked, con->in_seq);
863 con->in_seq_acked = con->in_seq;
865 con_out_kvec_reset(con);
867 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
869 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
870 con_out_kvec_add(con, sizeof (con->out_temp_ack),
873 con->out_more = 1; /* more will follow.. eventually.. */
874 con_flag_set(con, CON_FLAG_WRITE_PENDING);
878 * Prepare to write keepalive byte.
880 static void prepare_write_keepalive(struct ceph_connection *con)
882 dout("prepare_write_keepalive %p\n", con);
883 con_out_kvec_reset(con);
884 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
885 con_flag_set(con, CON_FLAG_WRITE_PENDING);
889 * Connection negotiation.
892 static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
895 struct ceph_auth_handshake *auth;
897 if (!con->ops->get_authorizer) {
898 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
899 con->out_connect.authorizer_len = 0;
903 /* Can't hold the mutex while getting authorizer */
904 mutex_unlock(&con->mutex);
905 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
906 mutex_lock(&con->mutex);
910 if (con->state != CON_STATE_NEGOTIATING)
911 return ERR_PTR(-EAGAIN);
913 con->auth_reply_buf = auth->authorizer_reply_buf;
914 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
919 * We connected to a peer and are saying hello.
921 static void prepare_write_banner(struct ceph_connection *con)
923 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
924 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
925 &con->msgr->my_enc_addr);
928 con_flag_set(con, CON_FLAG_WRITE_PENDING);
931 static int prepare_write_connect(struct ceph_connection *con)
933 unsigned int global_seq = get_global_seq(con->msgr, 0);
936 struct ceph_auth_handshake *auth;
938 switch (con->peer_name.type) {
939 case CEPH_ENTITY_TYPE_MON:
940 proto = CEPH_MONC_PROTOCOL;
942 case CEPH_ENTITY_TYPE_OSD:
943 proto = CEPH_OSDC_PROTOCOL;
945 case CEPH_ENTITY_TYPE_MDS:
946 proto = CEPH_MDSC_PROTOCOL;
952 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
953 con->connect_seq, global_seq, proto);
955 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
956 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
957 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
958 con->out_connect.global_seq = cpu_to_le32(global_seq);
959 con->out_connect.protocol_version = cpu_to_le32(proto);
960 con->out_connect.flags = 0;
962 auth_proto = CEPH_AUTH_UNKNOWN;
963 auth = get_connect_authorizer(con, &auth_proto);
965 return PTR_ERR(auth);
967 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
968 con->out_connect.authorizer_len = auth ?
969 cpu_to_le32(auth->authorizer_buf_len) : 0;
971 con_out_kvec_add(con, sizeof (con->out_connect),
973 if (auth && auth->authorizer_buf_len)
974 con_out_kvec_add(con, auth->authorizer_buf_len,
975 auth->authorizer_buf);
978 con_flag_set(con, CON_FLAG_WRITE_PENDING);
984 * write as much of pending kvecs to the socket as we can.
986 * 0 -> socket full, but more to do
989 static int write_partial_kvec(struct ceph_connection *con)
993 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
994 while (con->out_kvec_bytes > 0) {
995 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
996 con->out_kvec_left, con->out_kvec_bytes,
1000 con->out_kvec_bytes -= ret;
1001 if (con->out_kvec_bytes == 0)
1004 /* account for full iov entries consumed */
1005 while (ret >= con->out_kvec_cur->iov_len) {
1006 BUG_ON(!con->out_kvec_left);
1007 ret -= con->out_kvec_cur->iov_len;
1008 con->out_kvec_cur++;
1009 con->out_kvec_left--;
1011 /* and for a partially-consumed entry */
1013 con->out_kvec_cur->iov_len -= ret;
1014 con->out_kvec_cur->iov_base += ret;
1017 con->out_kvec_left = 0;
1018 con->out_kvec_is_msg = false;
1021 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1022 con->out_kvec_bytes, con->out_kvec_left, ret);
1023 return ret; /* done! */
1026 static void out_msg_pos_next(struct ceph_connection *con, struct page *page,
1027 size_t len, size_t sent, bool in_trail)
1029 struct ceph_msg *msg = con->out_msg;
1034 con->out_msg_pos.data_pos += sent;
1035 con->out_msg_pos.page_pos += sent;
1039 BUG_ON(sent != len);
1040 con->out_msg_pos.page_pos = 0;
1041 con->out_msg_pos.page++;
1042 con->out_msg_pos.did_page_crc = false;
1044 list_rotate_left(&msg->trail->head);
1045 else if (msg->pagelist)
1046 list_rotate_left(&msg->pagelist->head);
1049 iter_bio_next(&msg->bio_iter, &msg->bio_seg);
1053 static void in_msg_pos_next(struct ceph_connection *con, size_t len,
1056 struct ceph_msg *msg = con->in_msg;
1061 con->in_msg_pos.data_pos += received;
1062 con->in_msg_pos.page_pos += received;
1066 BUG_ON(received != len);
1067 con->in_msg_pos.page_pos = 0;
1068 con->in_msg_pos.page++;
1071 iter_bio_next(&msg->bio_iter, &msg->bio_seg);
1072 #endif /* CONFIG_BLOCK */
1076 * Write as much message data payload as we can. If we finish, queue
1078 * 1 -> done, footer is now queued in out_kvec[].
1079 * 0 -> socket full, but more to do
1082 static int write_partial_msg_pages(struct ceph_connection *con)
1084 struct ceph_msg *msg = con->out_msg;
1085 unsigned int data_len = le32_to_cpu(msg->hdr.data_len);
1087 bool do_datacrc = !con->msgr->nocrc;
1089 int total_max_write;
1090 bool in_trail = false;
1091 const size_t trail_len = (msg->trail ? msg->trail->length : 0);
1092 const size_t trail_off = data_len - trail_len;
1094 dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
1095 con, msg, con->out_msg_pos.page, msg->page_count,
1096 con->out_msg_pos.page_pos);
1099 * Iterate through each page that contains data to be
1100 * written, and send as much as possible for each.
1102 * If we are calculating the data crc (the default), we will
1103 * need to map the page. If we have no pages, they have
1104 * been revoked, so use the zero page.
1106 while (data_len > con->out_msg_pos.data_pos) {
1107 struct page *page = NULL;
1108 int max_write = PAGE_SIZE;
1111 in_trail = in_trail || con->out_msg_pos.data_pos >= trail_off;
1113 total_max_write = trail_off - con->out_msg_pos.data_pos;
1116 total_max_write = data_len - con->out_msg_pos.data_pos;
1118 page = list_first_entry(&msg->trail->head,
1120 } else if (msg->pages) {
1121 page = msg->pages[con->out_msg_pos.page];
1122 } else if (msg->pagelist) {
1123 page = list_first_entry(&msg->pagelist->head,
1126 } else if (msg->bio) {
1129 bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg);
1131 bio_offset = bv->bv_offset;
1132 max_write = bv->bv_len;
1137 len = min_t(int, max_write - con->out_msg_pos.page_pos,
1140 if (do_datacrc && !con->out_msg_pos.did_page_crc) {
1142 u32 crc = le32_to_cpu(msg->footer.data_crc);
1146 BUG_ON(kaddr == NULL);
1147 base = kaddr + con->out_msg_pos.page_pos + bio_offset;
1148 crc = crc32c(crc, base, len);
1150 msg->footer.data_crc = cpu_to_le32(crc);
1151 con->out_msg_pos.did_page_crc = true;
1153 ret = ceph_tcp_sendpage(con->sock, page,
1154 con->out_msg_pos.page_pos + bio_offset,
1159 out_msg_pos_next(con, page, len, (size_t) ret, in_trail);
1162 dout("write_partial_msg_pages %p msg %p done\n", con, msg);
1164 /* prepare and queue up footer, too */
1166 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1167 con_out_kvec_reset(con);
1168 prepare_write_message_footer(con);
1177 static int write_partial_skip(struct ceph_connection *con)
1181 while (con->out_skip > 0) {
1182 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
1184 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
1187 con->out_skip -= ret;
1195 * Prepare to read connection handshake, or an ack.
1197 static void prepare_read_banner(struct ceph_connection *con)
1199 dout("prepare_read_banner %p\n", con);
1200 con->in_base_pos = 0;
1203 static void prepare_read_connect(struct ceph_connection *con)
1205 dout("prepare_read_connect %p\n", con);
1206 con->in_base_pos = 0;
1209 static void prepare_read_ack(struct ceph_connection *con)
1211 dout("prepare_read_ack %p\n", con);
1212 con->in_base_pos = 0;
1215 static void prepare_read_tag(struct ceph_connection *con)
1217 dout("prepare_read_tag %p\n", con);
1218 con->in_base_pos = 0;
1219 con->in_tag = CEPH_MSGR_TAG_READY;
1223 * Prepare to read a message.
1225 static int prepare_read_message(struct ceph_connection *con)
1227 dout("prepare_read_message %p\n", con);
1228 BUG_ON(con->in_msg != NULL);
1229 con->in_base_pos = 0;
1230 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1235 static int read_partial(struct ceph_connection *con,
1236 int end, int size, void *object)
1238 while (con->in_base_pos < end) {
1239 int left = end - con->in_base_pos;
1240 int have = size - left;
1241 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1244 con->in_base_pos += ret;
1251 * Read all or part of the connect-side handshake on a new connection
1253 static int read_partial_banner(struct ceph_connection *con)
1259 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1262 size = strlen(CEPH_BANNER);
1264 ret = read_partial(con, end, size, con->in_banner);
1268 size = sizeof (con->actual_peer_addr);
1270 ret = read_partial(con, end, size, &con->actual_peer_addr);
1274 size = sizeof (con->peer_addr_for_me);
1276 ret = read_partial(con, end, size, &con->peer_addr_for_me);
1284 static int read_partial_connect(struct ceph_connection *con)
1290 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1292 size = sizeof (con->in_reply);
1294 ret = read_partial(con, end, size, &con->in_reply);
1298 size = le32_to_cpu(con->in_reply.authorizer_len);
1300 ret = read_partial(con, end, size, con->auth_reply_buf);
1304 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1305 con, (int)con->in_reply.tag,
1306 le32_to_cpu(con->in_reply.connect_seq),
1307 le32_to_cpu(con->in_reply.global_seq));
1314 * Verify the hello banner looks okay.
1316 static int verify_hello(struct ceph_connection *con)
1318 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1319 pr_err("connect to %s got bad banner\n",
1320 ceph_pr_addr(&con->peer_addr.in_addr));
1321 con->error_msg = "protocol error, bad banner";
1327 static bool addr_is_blank(struct sockaddr_storage *ss)
1329 switch (ss->ss_family) {
1331 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1334 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1335 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1336 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1337 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1342 static int addr_port(struct sockaddr_storage *ss)
1344 switch (ss->ss_family) {
1346 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1348 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1353 static void addr_set_port(struct sockaddr_storage *ss, int p)
1355 switch (ss->ss_family) {
1357 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1360 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1366 * Unlike other *_pton function semantics, zero indicates success.
1368 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1369 char delim, const char **ipend)
1371 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1372 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1374 memset(ss, 0, sizeof(*ss));
1376 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1377 ss->ss_family = AF_INET;
1381 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1382 ss->ss_family = AF_INET6;
1390 * Extract hostname string and resolve using kernel DNS facility.
1392 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1393 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1394 struct sockaddr_storage *ss, char delim, const char **ipend)
1396 const char *end, *delim_p;
1397 char *colon_p, *ip_addr = NULL;
1401 * The end of the hostname occurs immediately preceding the delimiter or
1402 * the port marker (':') where the delimiter takes precedence.
1404 delim_p = memchr(name, delim, namelen);
1405 colon_p = memchr(name, ':', namelen);
1407 if (delim_p && colon_p)
1408 end = delim_p < colon_p ? delim_p : colon_p;
1409 else if (!delim_p && colon_p)
1413 if (!end) /* case: hostname:/ */
1414 end = name + namelen;
1420 /* do dns_resolve upcall */
1421 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1423 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1431 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1432 ret, ret ? "failed" : ceph_pr_addr(ss));
1437 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1438 struct sockaddr_storage *ss, char delim, const char **ipend)
1445 * Parse a server name (IP or hostname). If a valid IP address is not found
1446 * then try to extract a hostname to resolve using userspace DNS upcall.
1448 static int ceph_parse_server_name(const char *name, size_t namelen,
1449 struct sockaddr_storage *ss, char delim, const char **ipend)
1453 ret = ceph_pton(name, namelen, ss, delim, ipend);
1455 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1461 * Parse an ip[:port] list into an addr array. Use the default
1462 * monitor port if a port isn't specified.
1464 int ceph_parse_ips(const char *c, const char *end,
1465 struct ceph_entity_addr *addr,
1466 int max_count, int *count)
1468 int i, ret = -EINVAL;
1471 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1472 for (i = 0; i < max_count; i++) {
1474 struct sockaddr_storage *ss = &addr[i].in_addr;
1483 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1492 dout("missing matching ']'\n");
1499 if (p < end && *p == ':') {
1502 while (p < end && *p >= '0' && *p <= '9') {
1503 port = (port * 10) + (*p - '0');
1506 if (port > 65535 || port == 0)
1509 port = CEPH_MON_PORT;
1512 addr_set_port(ss, port);
1514 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1531 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1534 EXPORT_SYMBOL(ceph_parse_ips);
1536 static int process_banner(struct ceph_connection *con)
1538 dout("process_banner on %p\n", con);
1540 if (verify_hello(con) < 0)
1543 ceph_decode_addr(&con->actual_peer_addr);
1544 ceph_decode_addr(&con->peer_addr_for_me);
1547 * Make sure the other end is who we wanted. note that the other
1548 * end may not yet know their ip address, so if it's 0.0.0.0, give
1549 * them the benefit of the doubt.
1551 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1552 sizeof(con->peer_addr)) != 0 &&
1553 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1554 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1555 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1556 ceph_pr_addr(&con->peer_addr.in_addr),
1557 (int)le32_to_cpu(con->peer_addr.nonce),
1558 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1559 (int)le32_to_cpu(con->actual_peer_addr.nonce));
1560 con->error_msg = "wrong peer at address";
1565 * did we learn our address?
1567 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1568 int port = addr_port(&con->msgr->inst.addr.in_addr);
1570 memcpy(&con->msgr->inst.addr.in_addr,
1571 &con->peer_addr_for_me.in_addr,
1572 sizeof(con->peer_addr_for_me.in_addr));
1573 addr_set_port(&con->msgr->inst.addr.in_addr, port);
1574 encode_my_addr(con->msgr);
1575 dout("process_banner learned my addr is %s\n",
1576 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
1582 static int process_connect(struct ceph_connection *con)
1584 u64 sup_feat = con->msgr->supported_features;
1585 u64 req_feat = con->msgr->required_features;
1586 u64 server_feat = le64_to_cpu(con->in_reply.features);
1589 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1591 switch (con->in_reply.tag) {
1592 case CEPH_MSGR_TAG_FEATURES:
1593 pr_err("%s%lld %s feature set mismatch,"
1594 " my %llx < server's %llx, missing %llx\n",
1595 ENTITY_NAME(con->peer_name),
1596 ceph_pr_addr(&con->peer_addr.in_addr),
1597 sup_feat, server_feat, server_feat & ~sup_feat);
1598 con->error_msg = "missing required protocol features";
1599 reset_connection(con);
1602 case CEPH_MSGR_TAG_BADPROTOVER:
1603 pr_err("%s%lld %s protocol version mismatch,"
1604 " my %d != server's %d\n",
1605 ENTITY_NAME(con->peer_name),
1606 ceph_pr_addr(&con->peer_addr.in_addr),
1607 le32_to_cpu(con->out_connect.protocol_version),
1608 le32_to_cpu(con->in_reply.protocol_version));
1609 con->error_msg = "protocol version mismatch";
1610 reset_connection(con);
1613 case CEPH_MSGR_TAG_BADAUTHORIZER:
1615 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1617 if (con->auth_retry == 2) {
1618 con->error_msg = "connect authorization failure";
1621 con->auth_retry = 1;
1622 con_out_kvec_reset(con);
1623 ret = prepare_write_connect(con);
1626 prepare_read_connect(con);
1629 case CEPH_MSGR_TAG_RESETSESSION:
1631 * If we connected with a large connect_seq but the peer
1632 * has no record of a session with us (no connection, or
1633 * connect_seq == 0), they will send RESETSESION to indicate
1634 * that they must have reset their session, and may have
1637 dout("process_connect got RESET peer seq %u\n",
1638 le32_to_cpu(con->in_reply.connect_seq));
1639 pr_err("%s%lld %s connection reset\n",
1640 ENTITY_NAME(con->peer_name),
1641 ceph_pr_addr(&con->peer_addr.in_addr));
1642 reset_connection(con);
1643 con_out_kvec_reset(con);
1644 ret = prepare_write_connect(con);
1647 prepare_read_connect(con);
1649 /* Tell ceph about it. */
1650 mutex_unlock(&con->mutex);
1651 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
1652 if (con->ops->peer_reset)
1653 con->ops->peer_reset(con);
1654 mutex_lock(&con->mutex);
1655 if (con->state != CON_STATE_NEGOTIATING)
1659 case CEPH_MSGR_TAG_RETRY_SESSION:
1661 * If we sent a smaller connect_seq than the peer has, try
1662 * again with a larger value.
1664 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
1665 le32_to_cpu(con->out_connect.connect_seq),
1666 le32_to_cpu(con->in_reply.connect_seq));
1667 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
1668 con_out_kvec_reset(con);
1669 ret = prepare_write_connect(con);
1672 prepare_read_connect(con);
1675 case CEPH_MSGR_TAG_RETRY_GLOBAL:
1677 * If we sent a smaller global_seq than the peer has, try
1678 * again with a larger value.
1680 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1681 con->peer_global_seq,
1682 le32_to_cpu(con->in_reply.global_seq));
1683 get_global_seq(con->msgr,
1684 le32_to_cpu(con->in_reply.global_seq));
1685 con_out_kvec_reset(con);
1686 ret = prepare_write_connect(con);
1689 prepare_read_connect(con);
1692 case CEPH_MSGR_TAG_READY:
1693 if (req_feat & ~server_feat) {
1694 pr_err("%s%lld %s protocol feature mismatch,"
1695 " my required %llx > server's %llx, need %llx\n",
1696 ENTITY_NAME(con->peer_name),
1697 ceph_pr_addr(&con->peer_addr.in_addr),
1698 req_feat, server_feat, req_feat & ~server_feat);
1699 con->error_msg = "missing required protocol features";
1700 reset_connection(con);
1704 WARN_ON(con->state != CON_STATE_NEGOTIATING);
1705 con->state = CON_STATE_OPEN;
1707 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
1709 con->peer_features = server_feat;
1710 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1711 con->peer_global_seq,
1712 le32_to_cpu(con->in_reply.connect_seq),
1714 WARN_ON(con->connect_seq !=
1715 le32_to_cpu(con->in_reply.connect_seq));
1717 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
1718 con_flag_set(con, CON_FLAG_LOSSYTX);
1720 con->delay = 0; /* reset backoff memory */
1722 prepare_read_tag(con);
1725 case CEPH_MSGR_TAG_WAIT:
1727 * If there is a connection race (we are opening
1728 * connections to each other), one of us may just have
1729 * to WAIT. This shouldn't happen if we are the
1732 pr_err("process_connect got WAIT as client\n");
1733 con->error_msg = "protocol error, got WAIT as client";
1737 pr_err("connect protocol error, will retry\n");
1738 con->error_msg = "protocol error, garbage tag during connect";
1746 * read (part of) an ack
1748 static int read_partial_ack(struct ceph_connection *con)
1750 int size = sizeof (con->in_temp_ack);
1753 return read_partial(con, end, size, &con->in_temp_ack);
1758 * We can finally discard anything that's been acked.
1760 static void process_ack(struct ceph_connection *con)
1763 u64 ack = le64_to_cpu(con->in_temp_ack);
1766 while (!list_empty(&con->out_sent)) {
1767 m = list_first_entry(&con->out_sent, struct ceph_msg,
1769 seq = le64_to_cpu(m->hdr.seq);
1772 dout("got ack for seq %llu type %d at %p\n", seq,
1773 le16_to_cpu(m->hdr.type), m);
1774 m->ack_stamp = jiffies;
1777 prepare_read_tag(con);
1783 static int read_partial_message_section(struct ceph_connection *con,
1784 struct kvec *section,
1785 unsigned int sec_len, u32 *crc)
1791 while (section->iov_len < sec_len) {
1792 BUG_ON(section->iov_base == NULL);
1793 left = sec_len - section->iov_len;
1794 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
1795 section->iov_len, left);
1798 section->iov_len += ret;
1800 if (section->iov_len == sec_len)
1801 *crc = crc32c(0, section->iov_base, section->iov_len);
1806 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
1808 static int read_partial_message_pages(struct ceph_connection *con,
1809 struct page **pages,
1810 unsigned int data_len, bool do_datacrc)
1817 left = min((int)(data_len - con->in_msg_pos.data_pos),
1818 (int)(PAGE_SIZE - con->in_msg_pos.page_pos));
1820 BUG_ON(pages == NULL);
1821 page = pages[con->in_msg_pos.page];
1823 ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, left);
1824 if (ret > 0 && do_datacrc)
1826 crc32c(con->in_data_crc,
1827 p + con->in_msg_pos.page_pos, ret);
1832 in_msg_pos_next(con, left, ret);
1838 static int read_partial_message_bio(struct ceph_connection *con,
1839 unsigned int data_len, bool do_datacrc)
1841 struct ceph_msg *msg = con->in_msg;
1848 BUG_ON(!msg->bio_iter);
1849 bv = bio_iovec_idx(msg->bio_iter, msg->bio_seg);
1851 left = min((int)(data_len - con->in_msg_pos.data_pos),
1852 (int)(bv->bv_len - con->in_msg_pos.page_pos));
1855 p = kmap(page) + bv->bv_offset;
1857 ret = ceph_tcp_recvmsg(con->sock, p + con->in_msg_pos.page_pos, left);
1858 if (ret > 0 && do_datacrc)
1860 crc32c(con->in_data_crc,
1861 p + con->in_msg_pos.page_pos, ret);
1866 in_msg_pos_next(con, left, ret);
1873 * read (part of) a message.
1875 static int read_partial_message(struct ceph_connection *con)
1877 struct ceph_msg *m = con->in_msg;
1881 unsigned int front_len, middle_len, data_len;
1882 bool do_datacrc = !con->msgr->nocrc;
1886 dout("read_partial_message con %p msg %p\n", con, m);
1889 size = sizeof (con->in_hdr);
1891 ret = read_partial(con, end, size, &con->in_hdr);
1895 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
1896 if (cpu_to_le32(crc) != con->in_hdr.crc) {
1897 pr_err("read_partial_message bad hdr "
1898 " crc %u != expected %u\n",
1899 crc, con->in_hdr.crc);
1903 front_len = le32_to_cpu(con->in_hdr.front_len);
1904 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
1906 middle_len = le32_to_cpu(con->in_hdr.middle_len);
1907 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
1909 data_len = le32_to_cpu(con->in_hdr.data_len);
1910 if (data_len > CEPH_MSG_MAX_DATA_LEN)
1914 seq = le64_to_cpu(con->in_hdr.seq);
1915 if ((s64)seq - (s64)con->in_seq < 1) {
1916 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1917 ENTITY_NAME(con->peer_name),
1918 ceph_pr_addr(&con->peer_addr.in_addr),
1919 seq, con->in_seq + 1);
1920 con->in_base_pos = -front_len - middle_len - data_len -
1922 con->in_tag = CEPH_MSGR_TAG_READY;
1924 } else if ((s64)seq - (s64)con->in_seq > 1) {
1925 pr_err("read_partial_message bad seq %lld expected %lld\n",
1926 seq, con->in_seq + 1);
1927 con->error_msg = "bad message sequence # for incoming message";
1931 /* allocate message? */
1935 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
1936 front_len, data_len);
1937 ret = ceph_con_in_msg_alloc(con, &skip);
1941 /* skip this message */
1942 dout("alloc_msg said skip message\n");
1943 BUG_ON(con->in_msg);
1944 con->in_base_pos = -front_len - middle_len - data_len -
1946 con->in_tag = CEPH_MSGR_TAG_READY;
1951 BUG_ON(!con->in_msg);
1952 BUG_ON(con->in_msg->con != con);
1954 m->front.iov_len = 0; /* haven't read it yet */
1956 m->middle->vec.iov_len = 0;
1958 con->in_msg_pos.page = 0;
1960 con->in_msg_pos.page_pos = m->page_alignment;
1962 con->in_msg_pos.page_pos = 0;
1963 con->in_msg_pos.data_pos = 0;
1967 init_bio_iter(m->bio, &m->bio_iter, &m->bio_seg);
1972 ret = read_partial_message_section(con, &m->front, front_len,
1973 &con->in_front_crc);
1979 ret = read_partial_message_section(con, &m->middle->vec,
1981 &con->in_middle_crc);
1987 while (con->in_msg_pos.data_pos < data_len) {
1989 ret = read_partial_message_pages(con, m->pages,
1990 data_len, do_datacrc);
1994 } else if (m->bio) {
1995 ret = read_partial_message_bio(con,
1996 data_len, do_datacrc);
2006 size = sizeof (m->footer);
2008 ret = read_partial(con, end, size, &m->footer);
2012 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2013 m, front_len, m->footer.front_crc, middle_len,
2014 m->footer.middle_crc, data_len, m->footer.data_crc);
2017 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2018 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2019 m, con->in_front_crc, m->footer.front_crc);
2022 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2023 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2024 m, con->in_middle_crc, m->footer.middle_crc);
2028 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2029 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2030 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2031 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2035 return 1; /* done! */
2039 * Process message. This happens in the worker thread. The callback should
2040 * be careful not to do anything that waits on other incoming messages or it
2043 static void process_message(struct ceph_connection *con)
2045 struct ceph_msg *msg;
2047 BUG_ON(con->in_msg->con != con);
2048 con->in_msg->con = NULL;
2053 /* if first message, set peer_name */
2054 if (con->peer_name.type == 0)
2055 con->peer_name = msg->hdr.src;
2058 mutex_unlock(&con->mutex);
2060 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2061 msg, le64_to_cpu(msg->hdr.seq),
2062 ENTITY_NAME(msg->hdr.src),
2063 le16_to_cpu(msg->hdr.type),
2064 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2065 le32_to_cpu(msg->hdr.front_len),
2066 le32_to_cpu(msg->hdr.data_len),
2067 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2068 con->ops->dispatch(con, msg);
2070 mutex_lock(&con->mutex);
2075 * Write something to the socket. Called in a worker thread when the
2076 * socket appears to be writeable and we have something ready to send.
2078 static int try_write(struct ceph_connection *con)
2082 dout("try_write start %p state %lu\n", con, con->state);
2085 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2087 /* open the socket first? */
2088 if (con->state == CON_STATE_PREOPEN) {
2090 con->state = CON_STATE_CONNECTING;
2092 con_out_kvec_reset(con);
2093 prepare_write_banner(con);
2094 prepare_read_banner(con);
2096 BUG_ON(con->in_msg);
2097 con->in_tag = CEPH_MSGR_TAG_READY;
2098 dout("try_write initiating connect on %p new state %lu\n",
2100 ret = ceph_tcp_connect(con);
2102 con->error_msg = "connect error";
2108 /* kvec data queued? */
2109 if (con->out_skip) {
2110 ret = write_partial_skip(con);
2114 if (con->out_kvec_left) {
2115 ret = write_partial_kvec(con);
2122 if (con->out_msg_done) {
2123 ceph_msg_put(con->out_msg);
2124 con->out_msg = NULL; /* we're done with this one */
2128 ret = write_partial_msg_pages(con);
2130 goto more_kvec; /* we need to send the footer, too! */
2134 dout("try_write write_partial_msg_pages err %d\n",
2141 if (con->state == CON_STATE_OPEN) {
2142 /* is anything else pending? */
2143 if (!list_empty(&con->out_queue)) {
2144 prepare_write_message(con);
2147 if (con->in_seq > con->in_seq_acked) {
2148 prepare_write_ack(con);
2151 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2152 prepare_write_keepalive(con);
2157 /* Nothing to do! */
2158 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2159 dout("try_write nothing else to write.\n");
2162 dout("try_write done on %p ret %d\n", con, ret);
2169 * Read what we can from the socket.
2171 static int try_read(struct ceph_connection *con)
2176 dout("try_read start on %p state %lu\n", con, con->state);
2177 if (con->state != CON_STATE_CONNECTING &&
2178 con->state != CON_STATE_NEGOTIATING &&
2179 con->state != CON_STATE_OPEN)
2184 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2187 if (con->state == CON_STATE_CONNECTING) {
2188 dout("try_read connecting\n");
2189 ret = read_partial_banner(con);
2192 ret = process_banner(con);
2196 con->state = CON_STATE_NEGOTIATING;
2199 * Received banner is good, exchange connection info.
2200 * Do not reset out_kvec, as sending our banner raced
2201 * with receiving peer banner after connect completed.
2203 ret = prepare_write_connect(con);
2206 prepare_read_connect(con);
2208 /* Send connection info before awaiting response */
2212 if (con->state == CON_STATE_NEGOTIATING) {
2213 dout("try_read negotiating\n");
2214 ret = read_partial_connect(con);
2217 ret = process_connect(con);
2223 WARN_ON(con->state != CON_STATE_OPEN);
2225 if (con->in_base_pos < 0) {
2227 * skipping + discarding content.
2229 * FIXME: there must be a better way to do this!
2231 static char buf[SKIP_BUF_SIZE];
2232 int skip = min((int) sizeof (buf), -con->in_base_pos);
2234 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2235 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2238 con->in_base_pos += ret;
2239 if (con->in_base_pos)
2242 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2246 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2249 dout("try_read got tag %d\n", (int)con->in_tag);
2250 switch (con->in_tag) {
2251 case CEPH_MSGR_TAG_MSG:
2252 prepare_read_message(con);
2254 case CEPH_MSGR_TAG_ACK:
2255 prepare_read_ack(con);
2257 case CEPH_MSGR_TAG_CLOSE:
2258 con_close_socket(con);
2259 con->state = CON_STATE_CLOSED;
2265 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2266 ret = read_partial_message(con);
2270 con->error_msg = "bad crc";
2274 con->error_msg = "io error";
2279 if (con->in_tag == CEPH_MSGR_TAG_READY)
2281 process_message(con);
2282 if (con->state == CON_STATE_OPEN)
2283 prepare_read_tag(con);
2286 if (con->in_tag == CEPH_MSGR_TAG_ACK) {
2287 ret = read_partial_ack(con);
2295 dout("try_read done on %p ret %d\n", con, ret);
2299 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2300 con->error_msg = "protocol error, garbage tag";
2307 * Atomically queue work on a connection after the specified delay.
2308 * Bump @con reference to avoid races with connection teardown.
2309 * Returns 0 if work was queued, or an error code otherwise.
2311 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2313 if (!con->ops->get(con)) {
2314 dout("%s %p ref count 0\n", __func__, con);
2319 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2320 dout("%s %p - already queued\n", __func__, con);
2326 dout("%s %p %lu\n", __func__, con, delay);
2331 static void queue_con(struct ceph_connection *con)
2333 (void) queue_con_delay(con, 0);
2336 static bool con_sock_closed(struct ceph_connection *con)
2338 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2342 case CON_STATE_ ## x: \
2343 con->error_msg = "socket closed (con state " #x ")"; \
2346 switch (con->state) {
2354 pr_warning("%s con %p unrecognized state %lu\n",
2355 __func__, con, con->state);
2356 con->error_msg = "unrecognized con state";
2365 static bool con_backoff(struct ceph_connection *con)
2369 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2372 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2374 dout("%s: con %p FAILED to back off %lu\n", __func__,
2376 BUG_ON(ret == -ENOENT);
2377 con_flag_set(con, CON_FLAG_BACKOFF);
2383 /* Finish fault handling; con->mutex must *not* be held here */
2385 static void con_fault_finish(struct ceph_connection *con)
2388 * in case we faulted due to authentication, invalidate our
2389 * current tickets so that we can get new ones.
2391 if (con->auth_retry && con->ops->invalidate_authorizer) {
2392 dout("calling invalidate_authorizer()\n");
2393 con->ops->invalidate_authorizer(con);
2396 if (con->ops->fault)
2397 con->ops->fault(con);
2401 * Do some work on a connection. Drop a connection ref when we're done.
2403 static void con_work(struct work_struct *work)
2405 struct ceph_connection *con = container_of(work, struct ceph_connection,
2409 mutex_lock(&con->mutex);
2413 if ((fault = con_sock_closed(con))) {
2414 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2417 if (con_backoff(con)) {
2418 dout("%s: con %p BACKOFF\n", __func__, con);
2421 if (con->state == CON_STATE_STANDBY) {
2422 dout("%s: con %p STANDBY\n", __func__, con);
2425 if (con->state == CON_STATE_CLOSED) {
2426 dout("%s: con %p CLOSED\n", __func__, con);
2430 if (con->state == CON_STATE_PREOPEN) {
2431 dout("%s: con %p PREOPEN\n", __func__, con);
2435 ret = try_read(con);
2439 con->error_msg = "socket error on read";
2444 ret = try_write(con);
2448 con->error_msg = "socket error on write";
2452 break; /* If we make it to here, we're done */
2456 mutex_unlock(&con->mutex);
2459 con_fault_finish(con);
2465 * Generic error/fault handler. A retry mechanism is used with
2466 * exponential backoff
2468 static void con_fault(struct ceph_connection *con)
2470 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2471 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2472 dout("fault %p state %lu to peer %s\n",
2473 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2475 WARN_ON(con->state != CON_STATE_CONNECTING &&
2476 con->state != CON_STATE_NEGOTIATING &&
2477 con->state != CON_STATE_OPEN);
2479 con_close_socket(con);
2481 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2482 dout("fault on LOSSYTX channel, marking CLOSED\n");
2483 con->state = CON_STATE_CLOSED;
2488 BUG_ON(con->in_msg->con != con);
2489 con->in_msg->con = NULL;
2490 ceph_msg_put(con->in_msg);
2495 /* Requeue anything that hasn't been acked */
2496 list_splice_init(&con->out_sent, &con->out_queue);
2498 /* If there are no messages queued or keepalive pending, place
2499 * the connection in a STANDBY state */
2500 if (list_empty(&con->out_queue) &&
2501 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
2502 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2503 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2504 con->state = CON_STATE_STANDBY;
2506 /* retry after a delay. */
2507 con->state = CON_STATE_PREOPEN;
2508 if (con->delay == 0)
2509 con->delay = BASE_DELAY_INTERVAL;
2510 else if (con->delay < MAX_DELAY_INTERVAL)
2512 con_flag_set(con, CON_FLAG_BACKOFF);
2520 * initialize a new messenger instance
2522 void ceph_messenger_init(struct ceph_messenger *msgr,
2523 struct ceph_entity_addr *myaddr,
2524 u32 supported_features,
2525 u32 required_features,
2528 msgr->supported_features = supported_features;
2529 msgr->required_features = required_features;
2531 spin_lock_init(&msgr->global_seq_lock);
2534 msgr->inst.addr = *myaddr;
2536 /* select a random nonce */
2537 msgr->inst.addr.type = 0;
2538 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2539 encode_my_addr(msgr);
2540 msgr->nocrc = nocrc;
2542 atomic_set(&msgr->stopping, 0);
2544 dout("%s %p\n", __func__, msgr);
2546 EXPORT_SYMBOL(ceph_messenger_init);
2548 static void clear_standby(struct ceph_connection *con)
2550 /* come back from STANDBY? */
2551 if (con->state == CON_STATE_STANDBY) {
2552 dout("clear_standby %p and ++connect_seq\n", con);
2553 con->state = CON_STATE_PREOPEN;
2555 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2556 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
2561 * Queue up an outgoing message on the given connection.
2563 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2566 msg->hdr.src = con->msgr->inst.name;
2567 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2568 msg->needs_out_seq = true;
2570 mutex_lock(&con->mutex);
2572 if (con->state == CON_STATE_CLOSED) {
2573 dout("con_send %p closed, dropping %p\n", con, msg);
2575 mutex_unlock(&con->mutex);
2579 BUG_ON(msg->con != NULL);
2580 msg->con = con->ops->get(con);
2581 BUG_ON(msg->con == NULL);
2583 BUG_ON(!list_empty(&msg->list_head));
2584 list_add_tail(&msg->list_head, &con->out_queue);
2585 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2586 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2587 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2588 le32_to_cpu(msg->hdr.front_len),
2589 le32_to_cpu(msg->hdr.middle_len),
2590 le32_to_cpu(msg->hdr.data_len));
2593 mutex_unlock(&con->mutex);
2595 /* if there wasn't anything waiting to send before, queue
2597 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
2600 EXPORT_SYMBOL(ceph_con_send);
2603 * Revoke a message that was previously queued for send
2605 void ceph_msg_revoke(struct ceph_msg *msg)
2607 struct ceph_connection *con = msg->con;
2610 return; /* Message not in our possession */
2612 mutex_lock(&con->mutex);
2613 if (!list_empty(&msg->list_head)) {
2614 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
2615 list_del_init(&msg->list_head);
2616 BUG_ON(msg->con == NULL);
2617 msg->con->ops->put(msg->con);
2623 if (con->out_msg == msg) {
2624 dout("%s %p msg %p - was sending\n", __func__, con, msg);
2625 con->out_msg = NULL;
2626 if (con->out_kvec_is_msg) {
2627 con->out_skip = con->out_kvec_bytes;
2628 con->out_kvec_is_msg = false;
2634 mutex_unlock(&con->mutex);
2638 * Revoke a message that we may be reading data into
2640 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
2642 struct ceph_connection *con;
2644 BUG_ON(msg == NULL);
2646 dout("%s msg %p null con\n", __func__, msg);
2648 return; /* Message not in our possession */
2652 mutex_lock(&con->mutex);
2653 if (con->in_msg == msg) {
2654 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
2655 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
2656 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
2658 /* skip rest of message */
2659 dout("%s %p msg %p revoked\n", __func__, con, msg);
2660 con->in_base_pos = con->in_base_pos -
2661 sizeof(struct ceph_msg_header) -
2665 sizeof(struct ceph_msg_footer);
2666 ceph_msg_put(con->in_msg);
2668 con->in_tag = CEPH_MSGR_TAG_READY;
2671 dout("%s %p in_msg %p msg %p no-op\n",
2672 __func__, con, con->in_msg, msg);
2674 mutex_unlock(&con->mutex);
2678 * Queue a keepalive byte to ensure the tcp connection is alive.
2680 void ceph_con_keepalive(struct ceph_connection *con)
2682 dout("con_keepalive %p\n", con);
2683 mutex_lock(&con->mutex);
2685 mutex_unlock(&con->mutex);
2686 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
2687 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
2690 EXPORT_SYMBOL(ceph_con_keepalive);
2694 * construct a new message with given type, size
2695 * the new msg has a ref count of 1.
2697 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
2702 m = kzalloc(sizeof(*m), flags);
2706 m->hdr.type = cpu_to_le16(type);
2707 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
2708 m->hdr.front_len = cpu_to_le32(front_len);
2710 INIT_LIST_HEAD(&m->list_head);
2711 kref_init(&m->kref);
2714 m->front_max = front_len;
2716 if (front_len > PAGE_CACHE_SIZE) {
2717 m->front.iov_base = __vmalloc(front_len, flags,
2719 m->front_is_vmalloc = true;
2721 m->front.iov_base = kmalloc(front_len, flags);
2723 if (m->front.iov_base == NULL) {
2724 dout("ceph_msg_new can't allocate %d bytes\n",
2729 m->front.iov_base = NULL;
2731 m->front.iov_len = front_len;
2733 dout("ceph_msg_new %p front %d\n", m, front_len);
2740 pr_err("msg_new can't create type %d front %d\n", type,
2744 dout("msg_new can't create type %d front %d\n", type,
2749 EXPORT_SYMBOL(ceph_msg_new);
2752 * Allocate "middle" portion of a message, if it is needed and wasn't
2753 * allocated by alloc_msg. This allows us to read a small fixed-size
2754 * per-type header in the front and then gracefully fail (i.e.,
2755 * propagate the error to the caller based on info in the front) when
2756 * the middle is too large.
2758 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
2760 int type = le16_to_cpu(msg->hdr.type);
2761 int middle_len = le32_to_cpu(msg->hdr.middle_len);
2763 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
2764 ceph_msg_type_name(type), middle_len);
2765 BUG_ON(!middle_len);
2766 BUG_ON(msg->middle);
2768 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
2775 * Allocate a message for receiving an incoming message on a
2776 * connection, and save the result in con->in_msg. Uses the
2777 * connection's private alloc_msg op if available.
2779 * Returns 0 on success, or a negative error code.
2781 * On success, if we set *skip = 1:
2782 * - the next message should be skipped and ignored.
2783 * - con->in_msg == NULL
2784 * or if we set *skip = 0:
2785 * - con->in_msg is non-null.
2786 * On error (ENOMEM, EAGAIN, ...),
2787 * - con->in_msg == NULL
2789 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
2791 struct ceph_msg_header *hdr = &con->in_hdr;
2792 int middle_len = le32_to_cpu(hdr->middle_len);
2793 struct ceph_msg *msg;
2796 BUG_ON(con->in_msg != NULL);
2797 BUG_ON(!con->ops->alloc_msg);
2799 mutex_unlock(&con->mutex);
2800 msg = con->ops->alloc_msg(con, hdr, skip);
2801 mutex_lock(&con->mutex);
2802 if (con->state != CON_STATE_OPEN) {
2810 con->in_msg->con = con->ops->get(con);
2811 BUG_ON(con->in_msg->con == NULL);
2814 * Null message pointer means either we should skip
2815 * this message or we couldn't allocate memory. The
2816 * former is not an error.
2820 con->error_msg = "error allocating memory for incoming message";
2824 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
2826 if (middle_len && !con->in_msg->middle) {
2827 ret = ceph_alloc_middle(con, con->in_msg);
2829 ceph_msg_put(con->in_msg);
2839 * Free a generically kmalloc'd message.
2841 void ceph_msg_kfree(struct ceph_msg *m)
2843 dout("msg_kfree %p\n", m);
2844 if (m->front_is_vmalloc)
2845 vfree(m->front.iov_base);
2847 kfree(m->front.iov_base);
2852 * Drop a msg ref. Destroy as needed.
2854 void ceph_msg_last_put(struct kref *kref)
2856 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
2858 dout("ceph_msg_put last one on %p\n", m);
2859 WARN_ON(!list_empty(&m->list_head));
2861 /* drop middle, data, if any */
2863 ceph_buffer_put(m->middle);
2870 ceph_pagelist_release(m->pagelist);
2878 ceph_msgpool_put(m->pool, m);
2882 EXPORT_SYMBOL(ceph_msg_last_put);
2884 void ceph_msg_dump(struct ceph_msg *msg)
2886 pr_debug("msg_dump %p (front_max %d page_count %d)\n", msg,
2887 msg->front_max, msg->page_count);
2888 print_hex_dump(KERN_DEBUG, "header: ",
2889 DUMP_PREFIX_OFFSET, 16, 1,
2890 &msg->hdr, sizeof(msg->hdr), true);
2891 print_hex_dump(KERN_DEBUG, " front: ",
2892 DUMP_PREFIX_OFFSET, 16, 1,
2893 msg->front.iov_base, msg->front.iov_len, true);
2895 print_hex_dump(KERN_DEBUG, "middle: ",
2896 DUMP_PREFIX_OFFSET, 16, 1,
2897 msg->middle->vec.iov_base,
2898 msg->middle->vec.iov_len, true);
2899 print_hex_dump(KERN_DEBUG, "footer: ",
2900 DUMP_PREFIX_OFFSET, 16, 1,
2901 &msg->footer, sizeof(msg->footer), true);
2903 EXPORT_SYMBOL(ceph_msg_dump);