1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
4 #include <linux/crc32c.h>
5 #include <linux/ctype.h>
6 #include <linux/highmem.h>
7 #include <linux/inet.h>
8 #include <linux/kthread.h>
10 #include <linux/nsproxy.h>
11 #include <linux/sched/mm.h>
12 #include <linux/slab.h>
13 #include <linux/socket.h>
14 #include <linux/string.h>
16 #include <linux/bio.h>
17 #endif /* CONFIG_BLOCK */
18 #include <linux/dns_resolver.h>
21 #include <linux/ceph/ceph_features.h>
22 #include <linux/ceph/libceph.h>
23 #include <linux/ceph/messenger.h>
24 #include <linux/ceph/decode.h>
25 #include <linux/ceph/pagelist.h>
26 #include <linux/export.h>
29 * Ceph uses the messenger to exchange ceph_msg messages with other
30 * hosts in the system. The messenger provides ordered and reliable
31 * delivery. We tolerate TCP disconnects by reconnecting (with
32 * exponential backoff) in the case of a fault (disconnection, bad
33 * crc, protocol error). Acks allow sent messages to be discarded by
38 * We track the state of the socket on a given connection using
39 * values defined below. The transition to a new socket state is
40 * handled by a function which verifies we aren't coming from an
44 * | NEW* | transient initial state
46 * | con_sock_state_init()
49 * | CLOSED | initialized, but no socket (and no
50 * ---------- TCP connection)
52 * | \ con_sock_state_connecting()
53 * | ----------------------
55 * + con_sock_state_closed() \
56 * |+--------------------------- \
59 * | | CLOSING | socket event; \ \
60 * | ----------- await close \ \
63 * | + con_sock_state_closing() \ |
65 * | / --------------- | |
68 * | / -----------------| CONNECTING | socket created, TCP
69 * | | / -------------- connect initiated
70 * | | | con_sock_state_connected()
73 * | CONNECTED | TCP connection established
76 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
79 #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
80 #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
81 #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
82 #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
83 #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
88 #define CON_STATE_CLOSED 1 /* -> PREOPEN */
89 #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
90 #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
91 #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
92 #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
93 #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
96 * ceph_connection flag bits
98 #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
99 * messages on errors */
100 #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
101 #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
102 #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
103 #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
105 static bool con_flag_valid(unsigned long con_flag)
108 case CON_FLAG_LOSSYTX:
109 case CON_FLAG_KEEPALIVE_PENDING:
110 case CON_FLAG_WRITE_PENDING:
111 case CON_FLAG_SOCK_CLOSED:
112 case CON_FLAG_BACKOFF:
119 static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
121 BUG_ON(!con_flag_valid(con_flag));
123 clear_bit(con_flag, &con->flags);
126 static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
128 BUG_ON(!con_flag_valid(con_flag));
130 set_bit(con_flag, &con->flags);
133 static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
135 BUG_ON(!con_flag_valid(con_flag));
137 return test_bit(con_flag, &con->flags);
140 static bool con_flag_test_and_clear(struct ceph_connection *con,
141 unsigned long con_flag)
143 BUG_ON(!con_flag_valid(con_flag));
145 return test_and_clear_bit(con_flag, &con->flags);
148 static bool con_flag_test_and_set(struct ceph_connection *con,
149 unsigned long con_flag)
151 BUG_ON(!con_flag_valid(con_flag));
153 return test_and_set_bit(con_flag, &con->flags);
156 /* Slab caches for frequently-allocated structures */
158 static struct kmem_cache *ceph_msg_cache;
159 static struct kmem_cache *ceph_msg_data_cache;
161 /* static tag bytes (protocol control messages) */
162 static char tag_msg = CEPH_MSGR_TAG_MSG;
163 static char tag_ack = CEPH_MSGR_TAG_ACK;
164 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
165 static char tag_keepalive2 = CEPH_MSGR_TAG_KEEPALIVE2;
167 #ifdef CONFIG_LOCKDEP
168 static struct lock_class_key socket_class;
172 * When skipping (ignoring) a block of input we read it into a "skip
173 * buffer," which is this many bytes in size.
175 #define SKIP_BUF_SIZE 1024
177 static void queue_con(struct ceph_connection *con);
178 static void cancel_con(struct ceph_connection *con);
179 static void ceph_con_workfn(struct work_struct *);
180 static void con_fault(struct ceph_connection *con);
183 * Nicely render a sockaddr as a string. An array of formatted
184 * strings is used, to approximate reentrancy.
186 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
187 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
188 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
189 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
191 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
192 static atomic_t addr_str_seq = ATOMIC_INIT(0);
194 static struct page *zero_page; /* used in certain error cases */
196 const char *ceph_pr_addr(const struct sockaddr_storage *ss)
200 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
201 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
203 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
206 switch (ss->ss_family) {
208 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
209 ntohs(in4->sin_port));
213 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
214 ntohs(in6->sin6_port));
218 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
224 EXPORT_SYMBOL(ceph_pr_addr);
226 static void encode_my_addr(struct ceph_messenger *msgr)
228 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
229 ceph_encode_addr(&msgr->my_enc_addr);
233 * work queue for all reading and writing to/from the socket.
235 static struct workqueue_struct *ceph_msgr_wq;
237 static int ceph_msgr_slab_init(void)
239 BUG_ON(ceph_msg_cache);
240 ceph_msg_cache = KMEM_CACHE(ceph_msg, 0);
244 BUG_ON(ceph_msg_data_cache);
245 ceph_msg_data_cache = KMEM_CACHE(ceph_msg_data, 0);
246 if (ceph_msg_data_cache)
249 kmem_cache_destroy(ceph_msg_cache);
250 ceph_msg_cache = NULL;
255 static void ceph_msgr_slab_exit(void)
257 BUG_ON(!ceph_msg_data_cache);
258 kmem_cache_destroy(ceph_msg_data_cache);
259 ceph_msg_data_cache = NULL;
261 BUG_ON(!ceph_msg_cache);
262 kmem_cache_destroy(ceph_msg_cache);
263 ceph_msg_cache = NULL;
266 static void _ceph_msgr_exit(void)
269 destroy_workqueue(ceph_msgr_wq);
273 BUG_ON(zero_page == NULL);
277 ceph_msgr_slab_exit();
280 int ceph_msgr_init(void)
282 if (ceph_msgr_slab_init())
285 BUG_ON(zero_page != NULL);
286 zero_page = ZERO_PAGE(0);
290 * The number of active work items is limited by the number of
291 * connections, so leave @max_active at default.
293 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
297 pr_err("msgr_init failed to create workqueue\n");
302 EXPORT_SYMBOL(ceph_msgr_init);
304 void ceph_msgr_exit(void)
306 BUG_ON(ceph_msgr_wq == NULL);
310 EXPORT_SYMBOL(ceph_msgr_exit);
312 void ceph_msgr_flush(void)
314 flush_workqueue(ceph_msgr_wq);
316 EXPORT_SYMBOL(ceph_msgr_flush);
318 /* Connection socket state transition functions */
320 static void con_sock_state_init(struct ceph_connection *con)
324 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
325 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
326 printk("%s: unexpected old state %d\n", __func__, old_state);
327 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
328 CON_SOCK_STATE_CLOSED);
331 static void con_sock_state_connecting(struct ceph_connection *con)
335 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
336 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
337 printk("%s: unexpected old state %d\n", __func__, old_state);
338 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
339 CON_SOCK_STATE_CONNECTING);
342 static void con_sock_state_connected(struct ceph_connection *con)
346 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
347 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
348 printk("%s: unexpected old state %d\n", __func__, old_state);
349 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
350 CON_SOCK_STATE_CONNECTED);
353 static void con_sock_state_closing(struct ceph_connection *con)
357 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
358 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
359 old_state != CON_SOCK_STATE_CONNECTED &&
360 old_state != CON_SOCK_STATE_CLOSING))
361 printk("%s: unexpected old state %d\n", __func__, old_state);
362 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
363 CON_SOCK_STATE_CLOSING);
366 static void con_sock_state_closed(struct ceph_connection *con)
370 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
371 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
372 old_state != CON_SOCK_STATE_CLOSING &&
373 old_state != CON_SOCK_STATE_CONNECTING &&
374 old_state != CON_SOCK_STATE_CLOSED))
375 printk("%s: unexpected old state %d\n", __func__, old_state);
376 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
377 CON_SOCK_STATE_CLOSED);
381 * socket callback functions
384 /* data available on socket, or listen socket received a connect */
385 static void ceph_sock_data_ready(struct sock *sk)
387 struct ceph_connection *con = sk->sk_user_data;
388 if (atomic_read(&con->msgr->stopping)) {
392 if (sk->sk_state != TCP_CLOSE_WAIT) {
393 dout("%s on %p state = %lu, queueing work\n", __func__,
399 /* socket has buffer space for writing */
400 static void ceph_sock_write_space(struct sock *sk)
402 struct ceph_connection *con = sk->sk_user_data;
404 /* only queue to workqueue if there is data we want to write,
405 * and there is sufficient space in the socket buffer to accept
406 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
407 * doesn't get called again until try_write() fills the socket
408 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
409 * and net/core/stream.c:sk_stream_write_space().
411 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
412 if (sk_stream_is_writeable(sk)) {
413 dout("%s %p queueing write work\n", __func__, con);
414 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
418 dout("%s %p nothing to write\n", __func__, con);
422 /* socket's state has changed */
423 static void ceph_sock_state_change(struct sock *sk)
425 struct ceph_connection *con = sk->sk_user_data;
427 dout("%s %p state = %lu sk_state = %u\n", __func__,
428 con, con->state, sk->sk_state);
430 switch (sk->sk_state) {
432 dout("%s TCP_CLOSE\n", __func__);
434 dout("%s TCP_CLOSE_WAIT\n", __func__);
435 con_sock_state_closing(con);
436 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
439 case TCP_ESTABLISHED:
440 dout("%s TCP_ESTABLISHED\n", __func__);
441 con_sock_state_connected(con);
444 default: /* Everything else is uninteresting */
450 * set up socket callbacks
452 static void set_sock_callbacks(struct socket *sock,
453 struct ceph_connection *con)
455 struct sock *sk = sock->sk;
456 sk->sk_user_data = con;
457 sk->sk_data_ready = ceph_sock_data_ready;
458 sk->sk_write_space = ceph_sock_write_space;
459 sk->sk_state_change = ceph_sock_state_change;
468 * initiate connection to a remote socket.
470 static int ceph_tcp_connect(struct ceph_connection *con)
472 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
474 unsigned int noio_flag;
479 /* sock_create_kern() allocates with GFP_KERNEL */
480 noio_flag = memalloc_noio_save();
481 ret = sock_create_kern(read_pnet(&con->msgr->net), paddr->ss_family,
482 SOCK_STREAM, IPPROTO_TCP, &sock);
483 memalloc_noio_restore(noio_flag);
486 sock->sk->sk_allocation = GFP_NOFS;
488 #ifdef CONFIG_LOCKDEP
489 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
492 set_sock_callbacks(sock, con);
494 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
496 con_sock_state_connecting(con);
497 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
499 if (ret == -EINPROGRESS) {
500 dout("connect %s EINPROGRESS sk_state = %u\n",
501 ceph_pr_addr(&con->peer_addr.in_addr),
503 } else if (ret < 0) {
504 pr_err("connect %s error %d\n",
505 ceph_pr_addr(&con->peer_addr.in_addr), ret);
510 if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY)) {
513 ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
514 (char *)&optval, sizeof(optval));
516 pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
524 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
526 struct kvec iov = {buf, len};
527 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
530 iov_iter_kvec(&msg.msg_iter, READ | ITER_KVEC, &iov, 1, len);
531 r = sock_recvmsg(sock, &msg, msg.msg_flags);
537 static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
538 int page_offset, size_t length)
540 struct bio_vec bvec = {
542 .bv_offset = page_offset,
545 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
548 BUG_ON(page_offset + length > PAGE_SIZE);
549 iov_iter_bvec(&msg.msg_iter, READ | ITER_BVEC, &bvec, 1, length);
550 r = sock_recvmsg(sock, &msg, msg.msg_flags);
557 * write something. @more is true if caller will be sending more data
560 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
561 size_t kvlen, size_t len, int more)
563 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
567 msg.msg_flags |= MSG_MORE;
569 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
571 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
577 static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
578 int offset, size_t size, bool more)
580 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
583 ret = kernel_sendpage(sock, page, offset, size, flags);
590 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
591 int offset, size_t size, bool more)
593 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
598 * sendpage cannot properly handle pages with page_count == 0,
599 * we need to fall back to sendmsg if that's the case.
601 * Same goes for slab pages: skb_can_coalesce() allows
602 * coalescing neighboring slab objects into a single frag which
603 * triggers one of hardened usercopy checks.
605 if (page_count(page) >= 1 && !PageSlab(page))
606 return __ceph_tcp_sendpage(sock, page, offset, size, more);
609 bvec.bv_offset = offset;
613 msg.msg_flags |= MSG_MORE;
615 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
617 iov_iter_bvec(&msg.msg_iter, WRITE | ITER_BVEC, &bvec, 1, size);
618 ret = sock_sendmsg(sock, &msg);
626 * Shutdown/close the socket for the given connection.
628 static int con_close_socket(struct ceph_connection *con)
632 dout("con_close_socket on %p sock %p\n", con, con->sock);
634 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
635 sock_release(con->sock);
640 * Forcibly clear the SOCK_CLOSED flag. It gets set
641 * independent of the connection mutex, and we could have
642 * received a socket close event before we had the chance to
643 * shut the socket down.
645 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
647 con_sock_state_closed(con);
652 * Reset a connection. Discard all incoming and outgoing messages
653 * and clear *_seq state.
655 static void ceph_msg_remove(struct ceph_msg *msg)
657 list_del_init(&msg->list_head);
661 static void ceph_msg_remove_list(struct list_head *head)
663 while (!list_empty(head)) {
664 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
666 ceph_msg_remove(msg);
670 static void reset_connection(struct ceph_connection *con)
672 /* reset connection, out_queue, msg_ and connect_seq */
673 /* discard existing out_queue and msg_seq */
674 dout("reset_connection %p\n", con);
675 ceph_msg_remove_list(&con->out_queue);
676 ceph_msg_remove_list(&con->out_sent);
679 BUG_ON(con->in_msg->con != con);
680 ceph_msg_put(con->in_msg);
684 con->connect_seq = 0;
687 BUG_ON(con->out_msg->con != con);
688 ceph_msg_put(con->out_msg);
692 con->in_seq_acked = 0;
698 * mark a peer down. drop any open connections.
700 void ceph_con_close(struct ceph_connection *con)
702 mutex_lock(&con->mutex);
703 dout("con_close %p peer %s\n", con,
704 ceph_pr_addr(&con->peer_addr.in_addr));
705 con->state = CON_STATE_CLOSED;
707 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
708 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
709 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
710 con_flag_clear(con, CON_FLAG_BACKOFF);
712 reset_connection(con);
713 con->peer_global_seq = 0;
715 con_close_socket(con);
716 mutex_unlock(&con->mutex);
718 EXPORT_SYMBOL(ceph_con_close);
721 * Reopen a closed connection, with a new peer address.
723 void ceph_con_open(struct ceph_connection *con,
724 __u8 entity_type, __u64 entity_num,
725 struct ceph_entity_addr *addr)
727 mutex_lock(&con->mutex);
728 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
730 WARN_ON(con->state != CON_STATE_CLOSED);
731 con->state = CON_STATE_PREOPEN;
733 con->peer_name.type = (__u8) entity_type;
734 con->peer_name.num = cpu_to_le64(entity_num);
736 memcpy(&con->peer_addr, addr, sizeof(*addr));
737 con->delay = 0; /* reset backoff memory */
738 mutex_unlock(&con->mutex);
741 EXPORT_SYMBOL(ceph_con_open);
744 * return true if this connection ever successfully opened
746 bool ceph_con_opened(struct ceph_connection *con)
748 return con->connect_seq > 0;
752 * initialize a new connection.
754 void ceph_con_init(struct ceph_connection *con, void *private,
755 const struct ceph_connection_operations *ops,
756 struct ceph_messenger *msgr)
758 dout("con_init %p\n", con);
759 memset(con, 0, sizeof(*con));
760 con->private = private;
764 con_sock_state_init(con);
766 mutex_init(&con->mutex);
767 INIT_LIST_HEAD(&con->out_queue);
768 INIT_LIST_HEAD(&con->out_sent);
769 INIT_DELAYED_WORK(&con->work, ceph_con_workfn);
771 con->state = CON_STATE_CLOSED;
773 EXPORT_SYMBOL(ceph_con_init);
777 * We maintain a global counter to order connection attempts. Get
778 * a unique seq greater than @gt.
780 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
784 spin_lock(&msgr->global_seq_lock);
785 if (msgr->global_seq < gt)
786 msgr->global_seq = gt;
787 ret = ++msgr->global_seq;
788 spin_unlock(&msgr->global_seq_lock);
792 static void con_out_kvec_reset(struct ceph_connection *con)
794 BUG_ON(con->out_skip);
796 con->out_kvec_left = 0;
797 con->out_kvec_bytes = 0;
798 con->out_kvec_cur = &con->out_kvec[0];
801 static void con_out_kvec_add(struct ceph_connection *con,
802 size_t size, void *data)
804 int index = con->out_kvec_left;
806 BUG_ON(con->out_skip);
807 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
809 con->out_kvec[index].iov_len = size;
810 con->out_kvec[index].iov_base = data;
811 con->out_kvec_left++;
812 con->out_kvec_bytes += size;
816 * Chop off a kvec from the end. Return residual number of bytes for
817 * that kvec, i.e. how many bytes would have been written if the kvec
820 static int con_out_kvec_skip(struct ceph_connection *con)
822 int off = con->out_kvec_cur - con->out_kvec;
825 if (con->out_kvec_bytes > 0) {
826 skip = con->out_kvec[off + con->out_kvec_left - 1].iov_len;
827 BUG_ON(con->out_kvec_bytes < skip);
828 BUG_ON(!con->out_kvec_left);
829 con->out_kvec_bytes -= skip;
830 con->out_kvec_left--;
839 * For a bio data item, a piece is whatever remains of the next
840 * entry in the current bio iovec, or the first entry in the next
843 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
846 struct ceph_msg_data *data = cursor->data;
849 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
854 cursor->resid = min(length, data->bio_length);
856 cursor->bvec_iter = bio->bi_iter;
858 cursor->resid <= bio_iter_len(bio, cursor->bvec_iter);
861 static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
865 struct ceph_msg_data *data = cursor->data;
867 struct bio_vec bio_vec;
869 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
874 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
876 *page_offset = (size_t) bio_vec.bv_offset;
877 BUG_ON(*page_offset >= PAGE_SIZE);
878 if (cursor->last_piece) /* pagelist offset is always 0 */
879 *length = cursor->resid;
881 *length = (size_t) bio_vec.bv_len;
882 BUG_ON(*length > cursor->resid);
883 BUG_ON(*page_offset + *length > PAGE_SIZE);
885 return bio_vec.bv_page;
888 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
892 struct bio_vec bio_vec;
894 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
899 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
901 /* Advance the cursor offset */
903 BUG_ON(cursor->resid < bytes);
904 cursor->resid -= bytes;
906 bio_advance_iter(bio, &cursor->bvec_iter, bytes);
908 if (bytes < bio_vec.bv_len)
909 return false; /* more bytes to process in this segment */
911 /* Move on to the next segment, and possibly the next bio */
913 if (!cursor->bvec_iter.bi_size) {
917 cursor->bvec_iter = bio->bi_iter;
919 memset(&cursor->bvec_iter, 0,
920 sizeof(cursor->bvec_iter));
923 if (!cursor->last_piece) {
924 BUG_ON(!cursor->resid);
926 /* A short read is OK, so use <= rather than == */
927 if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter))
928 cursor->last_piece = true;
933 #endif /* CONFIG_BLOCK */
936 * For a page array, a piece comes from the first page in the array
937 * that has not already been fully consumed.
939 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
942 struct ceph_msg_data *data = cursor->data;
945 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
947 BUG_ON(!data->pages);
948 BUG_ON(!data->length);
950 cursor->resid = min(length, data->length);
951 page_count = calc_pages_for(data->alignment, (u64)data->length);
952 cursor->page_offset = data->alignment & ~PAGE_MASK;
953 cursor->page_index = 0;
954 BUG_ON(page_count > (int)USHRT_MAX);
955 cursor->page_count = (unsigned short)page_count;
956 BUG_ON(length > SIZE_MAX - cursor->page_offset);
957 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
961 ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
962 size_t *page_offset, size_t *length)
964 struct ceph_msg_data *data = cursor->data;
966 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
968 BUG_ON(cursor->page_index >= cursor->page_count);
969 BUG_ON(cursor->page_offset >= PAGE_SIZE);
971 *page_offset = cursor->page_offset;
972 if (cursor->last_piece)
973 *length = cursor->resid;
975 *length = PAGE_SIZE - *page_offset;
977 return data->pages[cursor->page_index];
980 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
983 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
985 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
987 /* Advance the cursor page offset */
989 cursor->resid -= bytes;
990 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
991 if (!bytes || cursor->page_offset)
992 return false; /* more bytes to process in the current page */
995 return false; /* no more data */
997 /* Move on to the next page; offset is already at 0 */
999 BUG_ON(cursor->page_index >= cursor->page_count);
1000 cursor->page_index++;
1001 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1007 * For a pagelist, a piece is whatever remains to be consumed in the
1008 * first page in the list, or the front of the next page.
1011 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
1014 struct ceph_msg_data *data = cursor->data;
1015 struct ceph_pagelist *pagelist;
1018 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1020 pagelist = data->pagelist;
1024 return; /* pagelist can be assigned but empty */
1026 BUG_ON(list_empty(&pagelist->head));
1027 page = list_first_entry(&pagelist->head, struct page, lru);
1029 cursor->resid = min(length, pagelist->length);
1030 cursor->page = page;
1032 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1035 static struct page *
1036 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
1037 size_t *page_offset, size_t *length)
1039 struct ceph_msg_data *data = cursor->data;
1040 struct ceph_pagelist *pagelist;
1042 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1044 pagelist = data->pagelist;
1047 BUG_ON(!cursor->page);
1048 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1050 /* offset of first page in pagelist is always 0 */
1051 *page_offset = cursor->offset & ~PAGE_MASK;
1052 if (cursor->last_piece)
1053 *length = cursor->resid;
1055 *length = PAGE_SIZE - *page_offset;
1057 return cursor->page;
1060 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
1063 struct ceph_msg_data *data = cursor->data;
1064 struct ceph_pagelist *pagelist;
1066 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1068 pagelist = data->pagelist;
1071 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1072 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1074 /* Advance the cursor offset */
1076 cursor->resid -= bytes;
1077 cursor->offset += bytes;
1078 /* offset of first page in pagelist is always 0 */
1079 if (!bytes || cursor->offset & ~PAGE_MASK)
1080 return false; /* more bytes to process in the current page */
1083 return false; /* no more data */
1085 /* Move on to the next page */
1087 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1088 cursor->page = list_next_entry(cursor->page, lru);
1089 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1095 * Message data is handled (sent or received) in pieces, where each
1096 * piece resides on a single page. The network layer might not
1097 * consume an entire piece at once. A data item's cursor keeps
1098 * track of which piece is next to process and how much remains to
1099 * be processed in that piece. It also tracks whether the current
1100 * piece is the last one in the data item.
1102 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1104 size_t length = cursor->total_resid;
1106 switch (cursor->data->type) {
1107 case CEPH_MSG_DATA_PAGELIST:
1108 ceph_msg_data_pagelist_cursor_init(cursor, length);
1110 case CEPH_MSG_DATA_PAGES:
1111 ceph_msg_data_pages_cursor_init(cursor, length);
1114 case CEPH_MSG_DATA_BIO:
1115 ceph_msg_data_bio_cursor_init(cursor, length);
1117 #endif /* CONFIG_BLOCK */
1118 case CEPH_MSG_DATA_NONE:
1123 cursor->need_crc = true;
1126 static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1128 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1129 struct ceph_msg_data *data;
1132 BUG_ON(length > msg->data_length);
1133 BUG_ON(list_empty(&msg->data));
1135 cursor->data_head = &msg->data;
1136 cursor->total_resid = length;
1137 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1138 cursor->data = data;
1140 __ceph_msg_data_cursor_init(cursor);
1144 * Return the page containing the next piece to process for a given
1145 * data item, and supply the page offset and length of that piece.
1146 * Indicate whether this is the last piece in this data item.
1148 static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1149 size_t *page_offset, size_t *length,
1154 switch (cursor->data->type) {
1155 case CEPH_MSG_DATA_PAGELIST:
1156 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1158 case CEPH_MSG_DATA_PAGES:
1159 page = ceph_msg_data_pages_next(cursor, page_offset, length);
1162 case CEPH_MSG_DATA_BIO:
1163 page = ceph_msg_data_bio_next(cursor, page_offset, length);
1165 #endif /* CONFIG_BLOCK */
1166 case CEPH_MSG_DATA_NONE:
1172 BUG_ON(*page_offset + *length > PAGE_SIZE);
1175 *last_piece = cursor->last_piece;
1181 * Returns true if the result moves the cursor on to the next piece
1184 static void ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1189 BUG_ON(bytes > cursor->resid);
1190 switch (cursor->data->type) {
1191 case CEPH_MSG_DATA_PAGELIST:
1192 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1194 case CEPH_MSG_DATA_PAGES:
1195 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1198 case CEPH_MSG_DATA_BIO:
1199 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1201 #endif /* CONFIG_BLOCK */
1202 case CEPH_MSG_DATA_NONE:
1207 cursor->total_resid -= bytes;
1209 if (!cursor->resid && cursor->total_resid) {
1210 WARN_ON(!cursor->last_piece);
1211 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1212 cursor->data = list_next_entry(cursor->data, links);
1213 __ceph_msg_data_cursor_init(cursor);
1216 cursor->need_crc = new_piece;
1219 static size_t sizeof_footer(struct ceph_connection *con)
1221 return (con->peer_features & CEPH_FEATURE_MSG_AUTH) ?
1222 sizeof(struct ceph_msg_footer) :
1223 sizeof(struct ceph_msg_footer_old);
1226 static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
1231 /* Initialize data cursor */
1233 ceph_msg_data_cursor_init(msg, (size_t)data_len);
1237 * Prepare footer for currently outgoing message, and finish things
1238 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1240 static void prepare_write_message_footer(struct ceph_connection *con)
1242 struct ceph_msg *m = con->out_msg;
1244 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1246 dout("prepare_write_message_footer %p\n", con);
1247 con_out_kvec_add(con, sizeof_footer(con), &m->footer);
1248 if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
1249 if (con->ops->sign_message)
1250 con->ops->sign_message(m);
1254 m->old_footer.flags = m->footer.flags;
1256 con->out_more = m->more_to_follow;
1257 con->out_msg_done = true;
1261 * Prepare headers for the next outgoing message.
1263 static void prepare_write_message(struct ceph_connection *con)
1268 con_out_kvec_reset(con);
1269 con->out_msg_done = false;
1271 /* Sneak an ack in there first? If we can get it into the same
1272 * TCP packet that's a good thing. */
1273 if (con->in_seq > con->in_seq_acked) {
1274 con->in_seq_acked = con->in_seq;
1275 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1276 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1277 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1278 &con->out_temp_ack);
1281 BUG_ON(list_empty(&con->out_queue));
1282 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1284 BUG_ON(m->con != con);
1286 /* put message on sent list */
1288 list_move_tail(&m->list_head, &con->out_sent);
1291 * only assign outgoing seq # if we haven't sent this message
1292 * yet. if it is requeued, resend with it's original seq.
1294 if (m->needs_out_seq) {
1295 m->hdr.seq = cpu_to_le64(++con->out_seq);
1296 m->needs_out_seq = false;
1298 if (con->ops->reencode_message)
1299 con->ops->reencode_message(m);
1302 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1303 m, con->out_seq, le16_to_cpu(m->hdr.type),
1304 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1306 WARN_ON(m->front.iov_len != le32_to_cpu(m->hdr.front_len));
1307 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
1309 /* tag + hdr + front + middle */
1310 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1311 con_out_kvec_add(con, sizeof(con->out_hdr), &con->out_hdr);
1312 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1315 con_out_kvec_add(con, m->middle->vec.iov_len,
1316 m->middle->vec.iov_base);
1318 /* fill in hdr crc and finalize hdr */
1319 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1320 con->out_msg->hdr.crc = cpu_to_le32(crc);
1321 memcpy(&con->out_hdr, &con->out_msg->hdr, sizeof(con->out_hdr));
1323 /* fill in front and middle crc, footer */
1324 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1325 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1327 crc = crc32c(0, m->middle->vec.iov_base,
1328 m->middle->vec.iov_len);
1329 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1331 con->out_msg->footer.middle_crc = 0;
1332 dout("%s front_crc %u middle_crc %u\n", __func__,
1333 le32_to_cpu(con->out_msg->footer.front_crc),
1334 le32_to_cpu(con->out_msg->footer.middle_crc));
1335 con->out_msg->footer.flags = 0;
1337 /* is there a data payload? */
1338 con->out_msg->footer.data_crc = 0;
1339 if (m->data_length) {
1340 prepare_message_data(con->out_msg, m->data_length);
1341 con->out_more = 1; /* data + footer will follow */
1343 /* no, queue up footer too and be done */
1344 prepare_write_message_footer(con);
1347 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1353 static void prepare_write_ack(struct ceph_connection *con)
1355 dout("prepare_write_ack %p %llu -> %llu\n", con,
1356 con->in_seq_acked, con->in_seq);
1357 con->in_seq_acked = con->in_seq;
1359 con_out_kvec_reset(con);
1361 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1363 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1364 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1365 &con->out_temp_ack);
1367 con->out_more = 1; /* more will follow.. eventually.. */
1368 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1372 * Prepare to share the seq during handshake
1374 static void prepare_write_seq(struct ceph_connection *con)
1376 dout("prepare_write_seq %p %llu -> %llu\n", con,
1377 con->in_seq_acked, con->in_seq);
1378 con->in_seq_acked = con->in_seq;
1380 con_out_kvec_reset(con);
1382 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1383 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1384 &con->out_temp_ack);
1386 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1390 * Prepare to write keepalive byte.
1392 static void prepare_write_keepalive(struct ceph_connection *con)
1394 dout("prepare_write_keepalive %p\n", con);
1395 con_out_kvec_reset(con);
1396 if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
1397 struct timespec now;
1399 ktime_get_real_ts(&now);
1400 con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
1401 ceph_encode_timespec(&con->out_temp_keepalive2, &now);
1402 con_out_kvec_add(con, sizeof(con->out_temp_keepalive2),
1403 &con->out_temp_keepalive2);
1405 con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive);
1407 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1411 * Connection negotiation.
1414 static int get_connect_authorizer(struct ceph_connection *con)
1416 struct ceph_auth_handshake *auth;
1419 if (!con->ops->get_authorizer) {
1421 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1422 con->out_connect.authorizer_len = 0;
1426 auth = con->ops->get_authorizer(con, &auth_proto, con->auth_retry);
1428 return PTR_ERR(auth);
1431 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1432 con->out_connect.authorizer_len = cpu_to_le32(auth->authorizer_buf_len);
1437 * We connected to a peer and are saying hello.
1439 static void prepare_write_banner(struct ceph_connection *con)
1441 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1442 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1443 &con->msgr->my_enc_addr);
1446 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1449 static void __prepare_write_connect(struct ceph_connection *con)
1451 con_out_kvec_add(con, sizeof(con->out_connect), &con->out_connect);
1453 con_out_kvec_add(con, con->auth->authorizer_buf_len,
1454 con->auth->authorizer_buf);
1457 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1460 static int prepare_write_connect(struct ceph_connection *con)
1462 unsigned int global_seq = get_global_seq(con->msgr, 0);
1466 switch (con->peer_name.type) {
1467 case CEPH_ENTITY_TYPE_MON:
1468 proto = CEPH_MONC_PROTOCOL;
1470 case CEPH_ENTITY_TYPE_OSD:
1471 proto = CEPH_OSDC_PROTOCOL;
1473 case CEPH_ENTITY_TYPE_MDS:
1474 proto = CEPH_MDSC_PROTOCOL;
1480 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1481 con->connect_seq, global_seq, proto);
1483 con->out_connect.features =
1484 cpu_to_le64(from_msgr(con->msgr)->supported_features);
1485 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1486 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1487 con->out_connect.global_seq = cpu_to_le32(global_seq);
1488 con->out_connect.protocol_version = cpu_to_le32(proto);
1489 con->out_connect.flags = 0;
1491 ret = get_connect_authorizer(con);
1495 __prepare_write_connect(con);
1500 * write as much of pending kvecs to the socket as we can.
1502 * 0 -> socket full, but more to do
1505 static int write_partial_kvec(struct ceph_connection *con)
1509 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1510 while (con->out_kvec_bytes > 0) {
1511 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1512 con->out_kvec_left, con->out_kvec_bytes,
1516 con->out_kvec_bytes -= ret;
1517 if (con->out_kvec_bytes == 0)
1520 /* account for full iov entries consumed */
1521 while (ret >= con->out_kvec_cur->iov_len) {
1522 BUG_ON(!con->out_kvec_left);
1523 ret -= con->out_kvec_cur->iov_len;
1524 con->out_kvec_cur++;
1525 con->out_kvec_left--;
1527 /* and for a partially-consumed entry */
1529 con->out_kvec_cur->iov_len -= ret;
1530 con->out_kvec_cur->iov_base += ret;
1533 con->out_kvec_left = 0;
1536 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1537 con->out_kvec_bytes, con->out_kvec_left, ret);
1538 return ret; /* done! */
1541 static u32 ceph_crc32c_page(u32 crc, struct page *page,
1542 unsigned int page_offset,
1543 unsigned int length)
1548 BUG_ON(kaddr == NULL);
1549 crc = crc32c(crc, kaddr + page_offset, length);
1555 * Write as much message data payload as we can. If we finish, queue
1557 * 1 -> done, footer is now queued in out_kvec[].
1558 * 0 -> socket full, but more to do
1561 static int write_partial_message_data(struct ceph_connection *con)
1563 struct ceph_msg *msg = con->out_msg;
1564 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1565 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
1568 dout("%s %p msg %p\n", __func__, con, msg);
1570 if (list_empty(&msg->data))
1574 * Iterate through each page that contains data to be
1575 * written, and send as much as possible for each.
1577 * If we are calculating the data crc (the default), we will
1578 * need to map the page. If we have no pages, they have
1579 * been revoked, so use the zero page.
1581 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1582 while (cursor->resid) {
1589 page = ceph_msg_data_next(cursor, &page_offset, &length,
1591 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
1592 length, !last_piece);
1595 msg->footer.data_crc = cpu_to_le32(crc);
1599 if (do_datacrc && cursor->need_crc)
1600 crc = ceph_crc32c_page(crc, page, page_offset, length);
1601 ceph_msg_data_advance(cursor, (size_t)ret);
1604 dout("%s %p msg %p done\n", __func__, con, msg);
1606 /* prepare and queue up footer, too */
1608 msg->footer.data_crc = cpu_to_le32(crc);
1610 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1611 con_out_kvec_reset(con);
1612 prepare_write_message_footer(con);
1614 return 1; /* must return > 0 to indicate success */
1620 static int write_partial_skip(struct ceph_connection *con)
1624 dout("%s %p %d left\n", __func__, con, con->out_skip);
1625 while (con->out_skip > 0) {
1626 size_t size = min(con->out_skip, (int) PAGE_SIZE);
1628 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
1631 con->out_skip -= ret;
1639 * Prepare to read connection handshake, or an ack.
1641 static void prepare_read_banner(struct ceph_connection *con)
1643 dout("prepare_read_banner %p\n", con);
1644 con->in_base_pos = 0;
1647 static void prepare_read_connect(struct ceph_connection *con)
1649 dout("prepare_read_connect %p\n", con);
1650 con->in_base_pos = 0;
1653 static void prepare_read_ack(struct ceph_connection *con)
1655 dout("prepare_read_ack %p\n", con);
1656 con->in_base_pos = 0;
1659 static void prepare_read_seq(struct ceph_connection *con)
1661 dout("prepare_read_seq %p\n", con);
1662 con->in_base_pos = 0;
1663 con->in_tag = CEPH_MSGR_TAG_SEQ;
1666 static void prepare_read_tag(struct ceph_connection *con)
1668 dout("prepare_read_tag %p\n", con);
1669 con->in_base_pos = 0;
1670 con->in_tag = CEPH_MSGR_TAG_READY;
1673 static void prepare_read_keepalive_ack(struct ceph_connection *con)
1675 dout("prepare_read_keepalive_ack %p\n", con);
1676 con->in_base_pos = 0;
1680 * Prepare to read a message.
1682 static int prepare_read_message(struct ceph_connection *con)
1684 dout("prepare_read_message %p\n", con);
1685 BUG_ON(con->in_msg != NULL);
1686 con->in_base_pos = 0;
1687 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1692 static int read_partial(struct ceph_connection *con,
1693 int end, int size, void *object)
1695 while (con->in_base_pos < end) {
1696 int left = end - con->in_base_pos;
1697 int have = size - left;
1698 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1701 con->in_base_pos += ret;
1708 * Read all or part of the connect-side handshake on a new connection
1710 static int read_partial_banner(struct ceph_connection *con)
1716 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1719 size = strlen(CEPH_BANNER);
1721 ret = read_partial(con, end, size, con->in_banner);
1725 size = sizeof (con->actual_peer_addr);
1727 ret = read_partial(con, end, size, &con->actual_peer_addr);
1731 size = sizeof (con->peer_addr_for_me);
1733 ret = read_partial(con, end, size, &con->peer_addr_for_me);
1741 static int read_partial_connect(struct ceph_connection *con)
1747 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1749 size = sizeof (con->in_reply);
1751 ret = read_partial(con, end, size, &con->in_reply);
1756 size = le32_to_cpu(con->in_reply.authorizer_len);
1757 if (size > con->auth->authorizer_reply_buf_len) {
1758 pr_err("authorizer reply too big: %d > %zu\n", size,
1759 con->auth->authorizer_reply_buf_len);
1765 ret = read_partial(con, end, size,
1766 con->auth->authorizer_reply_buf);
1771 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1772 con, (int)con->in_reply.tag,
1773 le32_to_cpu(con->in_reply.connect_seq),
1774 le32_to_cpu(con->in_reply.global_seq));
1780 * Verify the hello banner looks okay.
1782 static int verify_hello(struct ceph_connection *con)
1784 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1785 pr_err("connect to %s got bad banner\n",
1786 ceph_pr_addr(&con->peer_addr.in_addr));
1787 con->error_msg = "protocol error, bad banner";
1793 static bool addr_is_blank(struct sockaddr_storage *ss)
1795 struct in_addr *addr = &((struct sockaddr_in *)ss)->sin_addr;
1796 struct in6_addr *addr6 = &((struct sockaddr_in6 *)ss)->sin6_addr;
1798 switch (ss->ss_family) {
1800 return addr->s_addr == htonl(INADDR_ANY);
1802 return ipv6_addr_any(addr6);
1808 static int addr_port(struct sockaddr_storage *ss)
1810 switch (ss->ss_family) {
1812 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1814 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1819 static void addr_set_port(struct sockaddr_storage *ss, int p)
1821 switch (ss->ss_family) {
1823 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1826 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1832 * Unlike other *_pton function semantics, zero indicates success.
1834 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1835 char delim, const char **ipend)
1837 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1838 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1840 memset(ss, 0, sizeof(*ss));
1842 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1843 ss->ss_family = AF_INET;
1847 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1848 ss->ss_family = AF_INET6;
1856 * Extract hostname string and resolve using kernel DNS facility.
1858 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1859 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1860 struct sockaddr_storage *ss, char delim, const char **ipend)
1862 const char *end, *delim_p;
1863 char *colon_p, *ip_addr = NULL;
1867 * The end of the hostname occurs immediately preceding the delimiter or
1868 * the port marker (':') where the delimiter takes precedence.
1870 delim_p = memchr(name, delim, namelen);
1871 colon_p = memchr(name, ':', namelen);
1873 if (delim_p && colon_p)
1874 end = delim_p < colon_p ? delim_p : colon_p;
1875 else if (!delim_p && colon_p)
1879 if (!end) /* case: hostname:/ */
1880 end = name + namelen;
1886 /* do dns_resolve upcall */
1887 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1889 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1897 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1898 ret, ret ? "failed" : ceph_pr_addr(ss));
1903 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1904 struct sockaddr_storage *ss, char delim, const char **ipend)
1911 * Parse a server name (IP or hostname). If a valid IP address is not found
1912 * then try to extract a hostname to resolve using userspace DNS upcall.
1914 static int ceph_parse_server_name(const char *name, size_t namelen,
1915 struct sockaddr_storage *ss, char delim, const char **ipend)
1919 ret = ceph_pton(name, namelen, ss, delim, ipend);
1921 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1927 * Parse an ip[:port] list into an addr array. Use the default
1928 * monitor port if a port isn't specified.
1930 int ceph_parse_ips(const char *c, const char *end,
1931 struct ceph_entity_addr *addr,
1932 int max_count, int *count)
1934 int i, ret = -EINVAL;
1937 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1938 for (i = 0; i < max_count; i++) {
1940 struct sockaddr_storage *ss = &addr[i].in_addr;
1949 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1958 dout("missing matching ']'\n");
1965 if (p < end && *p == ':') {
1968 while (p < end && *p >= '0' && *p <= '9') {
1969 port = (port * 10) + (*p - '0');
1973 port = CEPH_MON_PORT;
1974 else if (port > 65535)
1977 port = CEPH_MON_PORT;
1980 addr_set_port(ss, port);
1982 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1999 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
2002 EXPORT_SYMBOL(ceph_parse_ips);
2004 static int process_banner(struct ceph_connection *con)
2006 dout("process_banner on %p\n", con);
2008 if (verify_hello(con) < 0)
2011 ceph_decode_addr(&con->actual_peer_addr);
2012 ceph_decode_addr(&con->peer_addr_for_me);
2015 * Make sure the other end is who we wanted. note that the other
2016 * end may not yet know their ip address, so if it's 0.0.0.0, give
2017 * them the benefit of the doubt.
2019 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
2020 sizeof(con->peer_addr)) != 0 &&
2021 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
2022 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
2023 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
2024 ceph_pr_addr(&con->peer_addr.in_addr),
2025 (int)le32_to_cpu(con->peer_addr.nonce),
2026 ceph_pr_addr(&con->actual_peer_addr.in_addr),
2027 (int)le32_to_cpu(con->actual_peer_addr.nonce));
2028 con->error_msg = "wrong peer at address";
2033 * did we learn our address?
2035 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
2036 int port = addr_port(&con->msgr->inst.addr.in_addr);
2038 memcpy(&con->msgr->inst.addr.in_addr,
2039 &con->peer_addr_for_me.in_addr,
2040 sizeof(con->peer_addr_for_me.in_addr));
2041 addr_set_port(&con->msgr->inst.addr.in_addr, port);
2042 encode_my_addr(con->msgr);
2043 dout("process_banner learned my addr is %s\n",
2044 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
2050 static int process_connect(struct ceph_connection *con)
2052 u64 sup_feat = from_msgr(con->msgr)->supported_features;
2053 u64 req_feat = from_msgr(con->msgr)->required_features;
2054 u64 server_feat = le64_to_cpu(con->in_reply.features);
2057 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
2060 int len = le32_to_cpu(con->in_reply.authorizer_len);
2063 * Any connection that defines ->get_authorizer()
2064 * should also define ->add_authorizer_challenge() and
2065 * ->verify_authorizer_reply().
2067 * See get_connect_authorizer().
2069 if (con->in_reply.tag == CEPH_MSGR_TAG_CHALLENGE_AUTHORIZER) {
2070 ret = con->ops->add_authorizer_challenge(
2071 con, con->auth->authorizer_reply_buf, len);
2075 con_out_kvec_reset(con);
2076 __prepare_write_connect(con);
2077 prepare_read_connect(con);
2082 ret = con->ops->verify_authorizer_reply(con);
2084 con->error_msg = "bad authorize reply";
2090 switch (con->in_reply.tag) {
2091 case CEPH_MSGR_TAG_FEATURES:
2092 pr_err("%s%lld %s feature set mismatch,"
2093 " my %llx < server's %llx, missing %llx\n",
2094 ENTITY_NAME(con->peer_name),
2095 ceph_pr_addr(&con->peer_addr.in_addr),
2096 sup_feat, server_feat, server_feat & ~sup_feat);
2097 con->error_msg = "missing required protocol features";
2098 reset_connection(con);
2101 case CEPH_MSGR_TAG_BADPROTOVER:
2102 pr_err("%s%lld %s protocol version mismatch,"
2103 " my %d != server's %d\n",
2104 ENTITY_NAME(con->peer_name),
2105 ceph_pr_addr(&con->peer_addr.in_addr),
2106 le32_to_cpu(con->out_connect.protocol_version),
2107 le32_to_cpu(con->in_reply.protocol_version));
2108 con->error_msg = "protocol version mismatch";
2109 reset_connection(con);
2112 case CEPH_MSGR_TAG_BADAUTHORIZER:
2114 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2116 if (con->auth_retry == 2) {
2117 con->error_msg = "connect authorization failure";
2120 con_out_kvec_reset(con);
2121 ret = prepare_write_connect(con);
2124 prepare_read_connect(con);
2127 case CEPH_MSGR_TAG_RESETSESSION:
2129 * If we connected with a large connect_seq but the peer
2130 * has no record of a session with us (no connection, or
2131 * connect_seq == 0), they will send RESETSESION to indicate
2132 * that they must have reset their session, and may have
2135 dout("process_connect got RESET peer seq %u\n",
2136 le32_to_cpu(con->in_reply.connect_seq));
2137 pr_err("%s%lld %s connection reset\n",
2138 ENTITY_NAME(con->peer_name),
2139 ceph_pr_addr(&con->peer_addr.in_addr));
2140 reset_connection(con);
2141 con_out_kvec_reset(con);
2142 ret = prepare_write_connect(con);
2145 prepare_read_connect(con);
2147 /* Tell ceph about it. */
2148 mutex_unlock(&con->mutex);
2149 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2150 if (con->ops->peer_reset)
2151 con->ops->peer_reset(con);
2152 mutex_lock(&con->mutex);
2153 if (con->state != CON_STATE_NEGOTIATING)
2157 case CEPH_MSGR_TAG_RETRY_SESSION:
2159 * If we sent a smaller connect_seq than the peer has, try
2160 * again with a larger value.
2162 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2163 le32_to_cpu(con->out_connect.connect_seq),
2164 le32_to_cpu(con->in_reply.connect_seq));
2165 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2166 con_out_kvec_reset(con);
2167 ret = prepare_write_connect(con);
2170 prepare_read_connect(con);
2173 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2175 * If we sent a smaller global_seq than the peer has, try
2176 * again with a larger value.
2178 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2179 con->peer_global_seq,
2180 le32_to_cpu(con->in_reply.global_seq));
2181 get_global_seq(con->msgr,
2182 le32_to_cpu(con->in_reply.global_seq));
2183 con_out_kvec_reset(con);
2184 ret = prepare_write_connect(con);
2187 prepare_read_connect(con);
2190 case CEPH_MSGR_TAG_SEQ:
2191 case CEPH_MSGR_TAG_READY:
2192 if (req_feat & ~server_feat) {
2193 pr_err("%s%lld %s protocol feature mismatch,"
2194 " my required %llx > server's %llx, need %llx\n",
2195 ENTITY_NAME(con->peer_name),
2196 ceph_pr_addr(&con->peer_addr.in_addr),
2197 req_feat, server_feat, req_feat & ~server_feat);
2198 con->error_msg = "missing required protocol features";
2199 reset_connection(con);
2203 WARN_ON(con->state != CON_STATE_NEGOTIATING);
2204 con->state = CON_STATE_OPEN;
2205 con->auth_retry = 0; /* we authenticated; clear flag */
2206 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2208 con->peer_features = server_feat;
2209 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2210 con->peer_global_seq,
2211 le32_to_cpu(con->in_reply.connect_seq),
2213 WARN_ON(con->connect_seq !=
2214 le32_to_cpu(con->in_reply.connect_seq));
2216 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2217 con_flag_set(con, CON_FLAG_LOSSYTX);
2219 con->delay = 0; /* reset backoff memory */
2221 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2222 prepare_write_seq(con);
2223 prepare_read_seq(con);
2225 prepare_read_tag(con);
2229 case CEPH_MSGR_TAG_WAIT:
2231 * If there is a connection race (we are opening
2232 * connections to each other), one of us may just have
2233 * to WAIT. This shouldn't happen if we are the
2236 con->error_msg = "protocol error, got WAIT as client";
2240 con->error_msg = "protocol error, garbage tag during connect";
2248 * read (part of) an ack
2250 static int read_partial_ack(struct ceph_connection *con)
2252 int size = sizeof (con->in_temp_ack);
2255 return read_partial(con, end, size, &con->in_temp_ack);
2259 * We can finally discard anything that's been acked.
2261 static void process_ack(struct ceph_connection *con)
2264 u64 ack = le64_to_cpu(con->in_temp_ack);
2266 bool reconnect = (con->in_tag == CEPH_MSGR_TAG_SEQ);
2267 struct list_head *list = reconnect ? &con->out_queue : &con->out_sent;
2270 * In the reconnect case, con_fault() has requeued messages
2271 * in out_sent. We should cleanup old messages according to
2272 * the reconnect seq.
2274 while (!list_empty(list)) {
2275 m = list_first_entry(list, struct ceph_msg, list_head);
2276 if (reconnect && m->needs_out_seq)
2278 seq = le64_to_cpu(m->hdr.seq);
2281 dout("got ack for seq %llu type %d at %p\n", seq,
2282 le16_to_cpu(m->hdr.type), m);
2283 m->ack_stamp = jiffies;
2287 prepare_read_tag(con);
2291 static int read_partial_message_section(struct ceph_connection *con,
2292 struct kvec *section,
2293 unsigned int sec_len, u32 *crc)
2299 while (section->iov_len < sec_len) {
2300 BUG_ON(section->iov_base == NULL);
2301 left = sec_len - section->iov_len;
2302 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2303 section->iov_len, left);
2306 section->iov_len += ret;
2308 if (section->iov_len == sec_len)
2309 *crc = crc32c(0, section->iov_base, section->iov_len);
2314 static int read_partial_msg_data(struct ceph_connection *con)
2316 struct ceph_msg *msg = con->in_msg;
2317 struct ceph_msg_data_cursor *cursor = &msg->cursor;
2318 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2326 if (list_empty(&msg->data))
2330 crc = con->in_data_crc;
2331 while (cursor->resid) {
2332 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
2333 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2336 con->in_data_crc = crc;
2342 crc = ceph_crc32c_page(crc, page, page_offset, ret);
2343 ceph_msg_data_advance(cursor, (size_t)ret);
2346 con->in_data_crc = crc;
2348 return 1; /* must return > 0 to indicate success */
2352 * read (part of) a message.
2354 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2356 static int read_partial_message(struct ceph_connection *con)
2358 struct ceph_msg *m = con->in_msg;
2362 unsigned int front_len, middle_len, data_len;
2363 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2364 bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
2368 dout("read_partial_message con %p msg %p\n", con, m);
2371 size = sizeof (con->in_hdr);
2373 ret = read_partial(con, end, size, &con->in_hdr);
2377 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2378 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2379 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
2380 crc, con->in_hdr.crc);
2384 front_len = le32_to_cpu(con->in_hdr.front_len);
2385 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2387 middle_len = le32_to_cpu(con->in_hdr.middle_len);
2388 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2390 data_len = le32_to_cpu(con->in_hdr.data_len);
2391 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2395 seq = le64_to_cpu(con->in_hdr.seq);
2396 if ((s64)seq - (s64)con->in_seq < 1) {
2397 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2398 ENTITY_NAME(con->peer_name),
2399 ceph_pr_addr(&con->peer_addr.in_addr),
2400 seq, con->in_seq + 1);
2401 con->in_base_pos = -front_len - middle_len - data_len -
2403 con->in_tag = CEPH_MSGR_TAG_READY;
2405 } else if ((s64)seq - (s64)con->in_seq > 1) {
2406 pr_err("read_partial_message bad seq %lld expected %lld\n",
2407 seq, con->in_seq + 1);
2408 con->error_msg = "bad message sequence # for incoming message";
2412 /* allocate message? */
2416 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2417 front_len, data_len);
2418 ret = ceph_con_in_msg_alloc(con, &skip);
2422 BUG_ON(!con->in_msg ^ skip);
2424 /* skip this message */
2425 dout("alloc_msg said skip message\n");
2426 con->in_base_pos = -front_len - middle_len - data_len -
2428 con->in_tag = CEPH_MSGR_TAG_READY;
2433 BUG_ON(!con->in_msg);
2434 BUG_ON(con->in_msg->con != con);
2436 m->front.iov_len = 0; /* haven't read it yet */
2438 m->middle->vec.iov_len = 0;
2440 /* prepare for data payload, if any */
2443 prepare_message_data(con->in_msg, data_len);
2447 ret = read_partial_message_section(con, &m->front, front_len,
2448 &con->in_front_crc);
2454 ret = read_partial_message_section(con, &m->middle->vec,
2456 &con->in_middle_crc);
2463 ret = read_partial_msg_data(con);
2469 size = sizeof_footer(con);
2471 ret = read_partial(con, end, size, &m->footer);
2476 m->footer.flags = m->old_footer.flags;
2480 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2481 m, front_len, m->footer.front_crc, middle_len,
2482 m->footer.middle_crc, data_len, m->footer.data_crc);
2485 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2486 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2487 m, con->in_front_crc, m->footer.front_crc);
2490 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2491 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2492 m, con->in_middle_crc, m->footer.middle_crc);
2496 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2497 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2498 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2499 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2503 if (need_sign && con->ops->check_message_signature &&
2504 con->ops->check_message_signature(m)) {
2505 pr_err("read_partial_message %p signature check failed\n", m);
2509 return 1; /* done! */
2513 * Process message. This happens in the worker thread. The callback should
2514 * be careful not to do anything that waits on other incoming messages or it
2517 static void process_message(struct ceph_connection *con)
2519 struct ceph_msg *msg = con->in_msg;
2521 BUG_ON(con->in_msg->con != con);
2524 /* if first message, set peer_name */
2525 if (con->peer_name.type == 0)
2526 con->peer_name = msg->hdr.src;
2529 mutex_unlock(&con->mutex);
2531 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2532 msg, le64_to_cpu(msg->hdr.seq),
2533 ENTITY_NAME(msg->hdr.src),
2534 le16_to_cpu(msg->hdr.type),
2535 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2536 le32_to_cpu(msg->hdr.front_len),
2537 le32_to_cpu(msg->hdr.data_len),
2538 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2539 con->ops->dispatch(con, msg);
2541 mutex_lock(&con->mutex);
2544 static int read_keepalive_ack(struct ceph_connection *con)
2546 struct ceph_timespec ceph_ts;
2547 size_t size = sizeof(ceph_ts);
2548 int ret = read_partial(con, size, size, &ceph_ts);
2551 ceph_decode_timespec(&con->last_keepalive_ack, &ceph_ts);
2552 prepare_read_tag(con);
2557 * Write something to the socket. Called in a worker thread when the
2558 * socket appears to be writeable and we have something ready to send.
2560 static int try_write(struct ceph_connection *con)
2564 dout("try_write start %p state %lu\n", con, con->state);
2565 if (con->state != CON_STATE_PREOPEN &&
2566 con->state != CON_STATE_CONNECTING &&
2567 con->state != CON_STATE_NEGOTIATING &&
2568 con->state != CON_STATE_OPEN)
2572 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2574 /* open the socket first? */
2575 if (con->state == CON_STATE_PREOPEN) {
2577 con->state = CON_STATE_CONNECTING;
2579 con_out_kvec_reset(con);
2580 prepare_write_banner(con);
2581 prepare_read_banner(con);
2583 BUG_ON(con->in_msg);
2584 con->in_tag = CEPH_MSGR_TAG_READY;
2585 dout("try_write initiating connect on %p new state %lu\n",
2587 ret = ceph_tcp_connect(con);
2589 con->error_msg = "connect error";
2597 /* kvec data queued? */
2598 if (con->out_kvec_left) {
2599 ret = write_partial_kvec(con);
2603 if (con->out_skip) {
2604 ret = write_partial_skip(con);
2611 if (con->out_msg_done) {
2612 ceph_msg_put(con->out_msg);
2613 con->out_msg = NULL; /* we're done with this one */
2617 ret = write_partial_message_data(con);
2619 goto more_kvec; /* we need to send the footer, too! */
2623 dout("try_write write_partial_message_data err %d\n",
2630 if (con->state == CON_STATE_OPEN) {
2631 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2632 prepare_write_keepalive(con);
2635 /* is anything else pending? */
2636 if (!list_empty(&con->out_queue)) {
2637 prepare_write_message(con);
2640 if (con->in_seq > con->in_seq_acked) {
2641 prepare_write_ack(con);
2646 /* Nothing to do! */
2647 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2648 dout("try_write nothing else to write.\n");
2651 dout("try_write done on %p ret %d\n", con, ret);
2658 * Read what we can from the socket.
2660 static int try_read(struct ceph_connection *con)
2665 dout("try_read start on %p state %lu\n", con, con->state);
2666 if (con->state != CON_STATE_CONNECTING &&
2667 con->state != CON_STATE_NEGOTIATING &&
2668 con->state != CON_STATE_OPEN)
2673 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2676 if (con->state == CON_STATE_CONNECTING) {
2677 dout("try_read connecting\n");
2678 ret = read_partial_banner(con);
2681 ret = process_banner(con);
2685 con->state = CON_STATE_NEGOTIATING;
2688 * Received banner is good, exchange connection info.
2689 * Do not reset out_kvec, as sending our banner raced
2690 * with receiving peer banner after connect completed.
2692 ret = prepare_write_connect(con);
2695 prepare_read_connect(con);
2697 /* Send connection info before awaiting response */
2701 if (con->state == CON_STATE_NEGOTIATING) {
2702 dout("try_read negotiating\n");
2703 ret = read_partial_connect(con);
2706 ret = process_connect(con);
2712 WARN_ON(con->state != CON_STATE_OPEN);
2714 if (con->in_base_pos < 0) {
2716 * skipping + discarding content.
2718 * FIXME: there must be a better way to do this!
2720 static char buf[SKIP_BUF_SIZE];
2721 int skip = min((int) sizeof (buf), -con->in_base_pos);
2723 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2724 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2727 con->in_base_pos += ret;
2728 if (con->in_base_pos)
2731 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2735 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2738 dout("try_read got tag %d\n", (int)con->in_tag);
2739 switch (con->in_tag) {
2740 case CEPH_MSGR_TAG_MSG:
2741 prepare_read_message(con);
2743 case CEPH_MSGR_TAG_ACK:
2744 prepare_read_ack(con);
2746 case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
2747 prepare_read_keepalive_ack(con);
2749 case CEPH_MSGR_TAG_CLOSE:
2750 con_close_socket(con);
2751 con->state = CON_STATE_CLOSED;
2757 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2758 ret = read_partial_message(con);
2762 con->error_msg = "bad crc/signature";
2768 con->error_msg = "io error";
2773 if (con->in_tag == CEPH_MSGR_TAG_READY)
2775 process_message(con);
2776 if (con->state == CON_STATE_OPEN)
2777 prepare_read_tag(con);
2780 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2781 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2783 * the final handshake seq exchange is semantically
2784 * equivalent to an ACK
2786 ret = read_partial_ack(con);
2792 if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
2793 ret = read_keepalive_ack(con);
2800 dout("try_read done on %p ret %d\n", con, ret);
2804 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2805 con->error_msg = "protocol error, garbage tag";
2812 * Atomically queue work on a connection after the specified delay.
2813 * Bump @con reference to avoid races with connection teardown.
2814 * Returns 0 if work was queued, or an error code otherwise.
2816 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2818 if (!con->ops->get(con)) {
2819 dout("%s %p ref count 0\n", __func__, con);
2823 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2824 dout("%s %p - already queued\n", __func__, con);
2829 dout("%s %p %lu\n", __func__, con, delay);
2833 static void queue_con(struct ceph_connection *con)
2835 (void) queue_con_delay(con, 0);
2838 static void cancel_con(struct ceph_connection *con)
2840 if (cancel_delayed_work(&con->work)) {
2841 dout("%s %p\n", __func__, con);
2846 static bool con_sock_closed(struct ceph_connection *con)
2848 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2852 case CON_STATE_ ## x: \
2853 con->error_msg = "socket closed (con state " #x ")"; \
2856 switch (con->state) {
2864 pr_warn("%s con %p unrecognized state %lu\n",
2865 __func__, con, con->state);
2866 con->error_msg = "unrecognized con state";
2875 static bool con_backoff(struct ceph_connection *con)
2879 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2882 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2884 dout("%s: con %p FAILED to back off %lu\n", __func__,
2886 BUG_ON(ret == -ENOENT);
2887 con_flag_set(con, CON_FLAG_BACKOFF);
2893 /* Finish fault handling; con->mutex must *not* be held here */
2895 static void con_fault_finish(struct ceph_connection *con)
2897 dout("%s %p\n", __func__, con);
2900 * in case we faulted due to authentication, invalidate our
2901 * current tickets so that we can get new ones.
2903 if (con->auth_retry) {
2904 dout("auth_retry %d, invalidating\n", con->auth_retry);
2905 if (con->ops->invalidate_authorizer)
2906 con->ops->invalidate_authorizer(con);
2907 con->auth_retry = 0;
2910 if (con->ops->fault)
2911 con->ops->fault(con);
2915 * Do some work on a connection. Drop a connection ref when we're done.
2917 static void ceph_con_workfn(struct work_struct *work)
2919 struct ceph_connection *con = container_of(work, struct ceph_connection,
2923 mutex_lock(&con->mutex);
2927 if ((fault = con_sock_closed(con))) {
2928 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2931 if (con_backoff(con)) {
2932 dout("%s: con %p BACKOFF\n", __func__, con);
2935 if (con->state == CON_STATE_STANDBY) {
2936 dout("%s: con %p STANDBY\n", __func__, con);
2939 if (con->state == CON_STATE_CLOSED) {
2940 dout("%s: con %p CLOSED\n", __func__, con);
2944 if (con->state == CON_STATE_PREOPEN) {
2945 dout("%s: con %p PREOPEN\n", __func__, con);
2949 ret = try_read(con);
2953 if (!con->error_msg)
2954 con->error_msg = "socket error on read";
2959 ret = try_write(con);
2963 if (!con->error_msg)
2964 con->error_msg = "socket error on write";
2968 break; /* If we make it to here, we're done */
2972 mutex_unlock(&con->mutex);
2975 con_fault_finish(con);
2981 * Generic error/fault handler. A retry mechanism is used with
2982 * exponential backoff
2984 static void con_fault(struct ceph_connection *con)
2986 dout("fault %p state %lu to peer %s\n",
2987 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2989 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2990 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2991 con->error_msg = NULL;
2993 WARN_ON(con->state != CON_STATE_CONNECTING &&
2994 con->state != CON_STATE_NEGOTIATING &&
2995 con->state != CON_STATE_OPEN);
2997 con_close_socket(con);
2999 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
3000 dout("fault on LOSSYTX channel, marking CLOSED\n");
3001 con->state = CON_STATE_CLOSED;
3006 BUG_ON(con->in_msg->con != con);
3007 ceph_msg_put(con->in_msg);
3011 BUG_ON(con->out_msg->con != con);
3012 ceph_msg_put(con->out_msg);
3013 con->out_msg = NULL;
3016 /* Requeue anything that hasn't been acked */
3017 list_splice_init(&con->out_sent, &con->out_queue);
3019 /* If there are no messages queued or keepalive pending, place
3020 * the connection in a STANDBY state */
3021 if (list_empty(&con->out_queue) &&
3022 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
3023 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
3024 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
3025 con->state = CON_STATE_STANDBY;
3027 /* retry after a delay. */
3028 con->state = CON_STATE_PREOPEN;
3029 if (con->delay == 0)
3030 con->delay = BASE_DELAY_INTERVAL;
3031 else if (con->delay < MAX_DELAY_INTERVAL)
3033 con_flag_set(con, CON_FLAG_BACKOFF);
3041 * initialize a new messenger instance
3043 void ceph_messenger_init(struct ceph_messenger *msgr,
3044 struct ceph_entity_addr *myaddr)
3046 spin_lock_init(&msgr->global_seq_lock);
3049 msgr->inst.addr = *myaddr;
3051 /* select a random nonce */
3052 msgr->inst.addr.type = 0;
3053 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
3054 encode_my_addr(msgr);
3056 atomic_set(&msgr->stopping, 0);
3057 write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
3059 dout("%s %p\n", __func__, msgr);
3061 EXPORT_SYMBOL(ceph_messenger_init);
3063 void ceph_messenger_fini(struct ceph_messenger *msgr)
3065 put_net(read_pnet(&msgr->net));
3067 EXPORT_SYMBOL(ceph_messenger_fini);
3069 static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con)
3072 msg->con->ops->put(msg->con);
3074 msg->con = con ? con->ops->get(con) : NULL;
3075 BUG_ON(msg->con != con);
3078 static void clear_standby(struct ceph_connection *con)
3080 /* come back from STANDBY? */
3081 if (con->state == CON_STATE_STANDBY) {
3082 dout("clear_standby %p and ++connect_seq\n", con);
3083 con->state = CON_STATE_PREOPEN;
3085 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
3086 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
3091 * Queue up an outgoing message on the given connection.
3093 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
3096 msg->hdr.src = con->msgr->inst.name;
3097 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
3098 msg->needs_out_seq = true;
3100 mutex_lock(&con->mutex);
3102 if (con->state == CON_STATE_CLOSED) {
3103 dout("con_send %p closed, dropping %p\n", con, msg);
3105 mutex_unlock(&con->mutex);
3109 msg_con_set(msg, con);
3111 BUG_ON(!list_empty(&msg->list_head));
3112 list_add_tail(&msg->list_head, &con->out_queue);
3113 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3114 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3115 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3116 le32_to_cpu(msg->hdr.front_len),
3117 le32_to_cpu(msg->hdr.middle_len),
3118 le32_to_cpu(msg->hdr.data_len));
3121 mutex_unlock(&con->mutex);
3123 /* if there wasn't anything waiting to send before, queue
3125 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3128 EXPORT_SYMBOL(ceph_con_send);
3131 * Revoke a message that was previously queued for send
3133 void ceph_msg_revoke(struct ceph_msg *msg)
3135 struct ceph_connection *con = msg->con;
3138 dout("%s msg %p null con\n", __func__, msg);
3139 return; /* Message not in our possession */
3142 mutex_lock(&con->mutex);
3143 if (!list_empty(&msg->list_head)) {
3144 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
3145 list_del_init(&msg->list_head);
3150 if (con->out_msg == msg) {
3151 BUG_ON(con->out_skip);
3153 if (con->out_msg_done) {
3154 con->out_skip += con_out_kvec_skip(con);
3156 BUG_ON(!msg->data_length);
3157 con->out_skip += sizeof_footer(con);
3159 /* data, middle, front */
3160 if (msg->data_length)
3161 con->out_skip += msg->cursor.total_resid;
3163 con->out_skip += con_out_kvec_skip(con);
3164 con->out_skip += con_out_kvec_skip(con);
3166 dout("%s %p msg %p - was sending, will write %d skip %d\n",
3167 __func__, con, msg, con->out_kvec_bytes, con->out_skip);
3169 con->out_msg = NULL;
3173 mutex_unlock(&con->mutex);
3177 * Revoke a message that we may be reading data into
3179 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
3181 struct ceph_connection *con = msg->con;
3184 dout("%s msg %p null con\n", __func__, msg);
3185 return; /* Message not in our possession */
3188 mutex_lock(&con->mutex);
3189 if (con->in_msg == msg) {
3190 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3191 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3192 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
3194 /* skip rest of message */
3195 dout("%s %p msg %p revoked\n", __func__, con, msg);
3196 con->in_base_pos = con->in_base_pos -
3197 sizeof(struct ceph_msg_header) -
3201 sizeof(struct ceph_msg_footer);
3202 ceph_msg_put(con->in_msg);
3204 con->in_tag = CEPH_MSGR_TAG_READY;
3207 dout("%s %p in_msg %p msg %p no-op\n",
3208 __func__, con, con->in_msg, msg);
3210 mutex_unlock(&con->mutex);
3214 * Queue a keepalive byte to ensure the tcp connection is alive.
3216 void ceph_con_keepalive(struct ceph_connection *con)
3218 dout("con_keepalive %p\n", con);
3219 mutex_lock(&con->mutex);
3221 con_flag_set(con, CON_FLAG_KEEPALIVE_PENDING);
3222 mutex_unlock(&con->mutex);
3224 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3227 EXPORT_SYMBOL(ceph_con_keepalive);
3229 bool ceph_con_keepalive_expired(struct ceph_connection *con,
3230 unsigned long interval)
3233 (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
3234 struct timespec now;
3236 ktime_get_real_ts(&now);
3237 jiffies_to_timespec(interval, &ts);
3238 ts = timespec_add(con->last_keepalive_ack, ts);
3239 return timespec_compare(&now, &ts) >= 0;
3244 static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
3246 struct ceph_msg_data *data;
3248 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3251 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
3256 INIT_LIST_HEAD(&data->links);
3261 static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3266 WARN_ON(!list_empty(&data->links));
3267 if (data->type == CEPH_MSG_DATA_PAGELIST)
3268 ceph_pagelist_release(data->pagelist);
3269 kmem_cache_free(ceph_msg_data_cache, data);
3272 void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
3273 size_t length, size_t alignment)
3275 struct ceph_msg_data *data;
3280 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3282 data->pages = pages;
3283 data->length = length;
3284 data->alignment = alignment & ~PAGE_MASK;
3286 list_add_tail(&data->links, &msg->data);
3287 msg->data_length += length;
3289 EXPORT_SYMBOL(ceph_msg_data_add_pages);
3291 void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
3292 struct ceph_pagelist *pagelist)
3294 struct ceph_msg_data *data;
3297 BUG_ON(!pagelist->length);
3299 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3301 data->pagelist = pagelist;
3303 list_add_tail(&data->links, &msg->data);
3304 msg->data_length += pagelist->length;
3306 EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
3309 void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
3312 struct ceph_msg_data *data;
3316 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3319 data->bio_length = length;
3321 list_add_tail(&data->links, &msg->data);
3322 msg->data_length += length;
3324 EXPORT_SYMBOL(ceph_msg_data_add_bio);
3325 #endif /* CONFIG_BLOCK */
3328 * construct a new message with given type, size
3329 * the new msg has a ref count of 1.
3331 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3336 m = kmem_cache_zalloc(ceph_msg_cache, flags);
3340 m->hdr.type = cpu_to_le16(type);
3341 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3342 m->hdr.front_len = cpu_to_le32(front_len);
3344 INIT_LIST_HEAD(&m->list_head);
3345 kref_init(&m->kref);
3346 INIT_LIST_HEAD(&m->data);
3350 m->front.iov_base = ceph_kvmalloc(front_len, flags);
3351 if (m->front.iov_base == NULL) {
3352 dout("ceph_msg_new can't allocate %d bytes\n",
3357 m->front.iov_base = NULL;
3359 m->front_alloc_len = m->front.iov_len = front_len;
3361 dout("ceph_msg_new %p front %d\n", m, front_len);
3368 pr_err("msg_new can't create type %d front %d\n", type,
3372 dout("msg_new can't create type %d front %d\n", type,
3377 EXPORT_SYMBOL(ceph_msg_new);
3380 * Allocate "middle" portion of a message, if it is needed and wasn't
3381 * allocated by alloc_msg. This allows us to read a small fixed-size
3382 * per-type header in the front and then gracefully fail (i.e.,
3383 * propagate the error to the caller based on info in the front) when
3384 * the middle is too large.
3386 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3388 int type = le16_to_cpu(msg->hdr.type);
3389 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3391 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3392 ceph_msg_type_name(type), middle_len);
3393 BUG_ON(!middle_len);
3394 BUG_ON(msg->middle);
3396 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3403 * Allocate a message for receiving an incoming message on a
3404 * connection, and save the result in con->in_msg. Uses the
3405 * connection's private alloc_msg op if available.
3407 * Returns 0 on success, or a negative error code.
3409 * On success, if we set *skip = 1:
3410 * - the next message should be skipped and ignored.
3411 * - con->in_msg == NULL
3412 * or if we set *skip = 0:
3413 * - con->in_msg is non-null.
3414 * On error (ENOMEM, EAGAIN, ...),
3415 * - con->in_msg == NULL
3417 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3419 struct ceph_msg_header *hdr = &con->in_hdr;
3420 int middle_len = le32_to_cpu(hdr->middle_len);
3421 struct ceph_msg *msg;
3424 BUG_ON(con->in_msg != NULL);
3425 BUG_ON(!con->ops->alloc_msg);
3427 mutex_unlock(&con->mutex);
3428 msg = con->ops->alloc_msg(con, hdr, skip);
3429 mutex_lock(&con->mutex);
3430 if (con->state != CON_STATE_OPEN) {
3437 msg_con_set(msg, con);
3441 * Null message pointer means either we should skip
3442 * this message or we couldn't allocate memory. The
3443 * former is not an error.
3448 con->error_msg = "error allocating memory for incoming message";
3451 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3453 if (middle_len && !con->in_msg->middle) {
3454 ret = ceph_alloc_middle(con, con->in_msg);
3456 ceph_msg_put(con->in_msg);
3466 * Free a generically kmalloc'd message.
3468 static void ceph_msg_free(struct ceph_msg *m)
3470 dout("%s %p\n", __func__, m);
3471 kvfree(m->front.iov_base);
3472 kmem_cache_free(ceph_msg_cache, m);
3475 static void ceph_msg_release(struct kref *kref)
3477 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3478 struct ceph_msg_data *data, *next;
3480 dout("%s %p\n", __func__, m);
3481 WARN_ON(!list_empty(&m->list_head));
3483 msg_con_set(m, NULL);
3485 /* drop middle, data, if any */
3487 ceph_buffer_put(m->middle);
3491 list_for_each_entry_safe(data, next, &m->data, links) {
3492 list_del_init(&data->links);
3493 ceph_msg_data_destroy(data);
3498 ceph_msgpool_put(m->pool, m);
3503 struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3505 dout("%s %p (was %d)\n", __func__, msg,
3506 kref_read(&msg->kref));
3507 kref_get(&msg->kref);
3510 EXPORT_SYMBOL(ceph_msg_get);
3512 void ceph_msg_put(struct ceph_msg *msg)
3514 dout("%s %p (was %d)\n", __func__, msg,
3515 kref_read(&msg->kref));
3516 kref_put(&msg->kref, ceph_msg_release);
3518 EXPORT_SYMBOL(ceph_msg_put);
3520 void ceph_msg_dump(struct ceph_msg *msg)
3522 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3523 msg->front_alloc_len, msg->data_length);
3524 print_hex_dump(KERN_DEBUG, "header: ",
3525 DUMP_PREFIX_OFFSET, 16, 1,
3526 &msg->hdr, sizeof(msg->hdr), true);
3527 print_hex_dump(KERN_DEBUG, " front: ",
3528 DUMP_PREFIX_OFFSET, 16, 1,
3529 msg->front.iov_base, msg->front.iov_len, true);
3531 print_hex_dump(KERN_DEBUG, "middle: ",
3532 DUMP_PREFIX_OFFSET, 16, 1,
3533 msg->middle->vec.iov_base,
3534 msg->middle->vec.iov_len, true);
3535 print_hex_dump(KERN_DEBUG, "footer: ",
3536 DUMP_PREFIX_OFFSET, 16, 1,
3537 &msg->footer, sizeof(msg->footer), true);
3539 EXPORT_SYMBOL(ceph_msg_dump);