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/nsproxy.h>
10 #include <linux/sched/mm.h>
11 #include <linux/slab.h>
12 #include <linux/socket.h>
13 #include <linux/string.h>
15 #include <linux/bio.h>
16 #endif /* CONFIG_BLOCK */
17 #include <linux/dns_resolver.h>
20 #include <linux/ceph/ceph_features.h>
21 #include <linux/ceph/libceph.h>
22 #include <linux/ceph/messenger.h>
23 #include <linux/ceph/decode.h>
24 #include <linux/ceph/pagelist.h>
25 #include <linux/export.h>
28 * Ceph uses the messenger to exchange ceph_msg messages with other
29 * hosts in the system. The messenger provides ordered and reliable
30 * delivery. We tolerate TCP disconnects by reconnecting (with
31 * exponential backoff) in the case of a fault (disconnection, bad
32 * crc, protocol error). Acks allow sent messages to be discarded by
37 * We track the state of the socket on a given connection using
38 * values defined below. The transition to a new socket state is
39 * handled by a function which verifies we aren't coming from an
43 * | NEW* | transient initial state
45 * | con_sock_state_init()
48 * | CLOSED | initialized, but no socket (and no
49 * ---------- TCP connection)
51 * | \ con_sock_state_connecting()
52 * | ----------------------
54 * + con_sock_state_closed() \
55 * |+--------------------------- \
58 * | | CLOSING | socket event; \ \
59 * | ----------- await close \ \
62 * | + con_sock_state_closing() \ |
64 * | / --------------- | |
67 * | / -----------------| CONNECTING | socket created, TCP
68 * | | / -------------- connect initiated
69 * | | | con_sock_state_connected()
72 * | CONNECTED | TCP connection established
75 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
78 #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
79 #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
80 #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
81 #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
82 #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
87 #define CON_STATE_CLOSED 1 /* -> PREOPEN */
88 #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
89 #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
90 #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
91 #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
92 #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
95 * ceph_connection flag bits
97 #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
98 * messages on errors */
99 #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
100 #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
101 #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
102 #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
104 static bool con_flag_valid(unsigned long con_flag)
107 case CON_FLAG_LOSSYTX:
108 case CON_FLAG_KEEPALIVE_PENDING:
109 case CON_FLAG_WRITE_PENDING:
110 case CON_FLAG_SOCK_CLOSED:
111 case CON_FLAG_BACKOFF:
118 static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
120 BUG_ON(!con_flag_valid(con_flag));
122 clear_bit(con_flag, &con->flags);
125 static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
127 BUG_ON(!con_flag_valid(con_flag));
129 set_bit(con_flag, &con->flags);
132 static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
134 BUG_ON(!con_flag_valid(con_flag));
136 return test_bit(con_flag, &con->flags);
139 static bool con_flag_test_and_clear(struct ceph_connection *con,
140 unsigned long con_flag)
142 BUG_ON(!con_flag_valid(con_flag));
144 return test_and_clear_bit(con_flag, &con->flags);
147 static bool con_flag_test_and_set(struct ceph_connection *con,
148 unsigned long con_flag)
150 BUG_ON(!con_flag_valid(con_flag));
152 return test_and_set_bit(con_flag, &con->flags);
155 /* Slab caches for frequently-allocated structures */
157 static struct kmem_cache *ceph_msg_cache;
158 static struct kmem_cache *ceph_msg_data_cache;
160 /* static tag bytes (protocol control messages) */
161 static char tag_msg = CEPH_MSGR_TAG_MSG;
162 static char tag_ack = CEPH_MSGR_TAG_ACK;
163 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
164 static char tag_keepalive2 = CEPH_MSGR_TAG_KEEPALIVE2;
166 #ifdef CONFIG_LOCKDEP
167 static struct lock_class_key socket_class;
171 * When skipping (ignoring) a block of input we read it into a "skip
172 * buffer," which is this many bytes in size.
174 #define SKIP_BUF_SIZE 1024
176 static void queue_con(struct ceph_connection *con);
177 static void cancel_con(struct ceph_connection *con);
178 static void ceph_con_workfn(struct work_struct *);
179 static void con_fault(struct ceph_connection *con);
182 * Nicely render a sockaddr as a string. An array of formatted
183 * strings is used, to approximate reentrancy.
185 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
186 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
187 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
188 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
190 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
191 static atomic_t addr_str_seq = ATOMIC_INIT(0);
193 static struct page *zero_page; /* used in certain error cases */
195 const char *ceph_pr_addr(const struct sockaddr_storage *ss)
199 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
200 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
202 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
205 switch (ss->ss_family) {
207 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
208 ntohs(in4->sin_port));
212 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
213 ntohs(in6->sin6_port));
217 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
223 EXPORT_SYMBOL(ceph_pr_addr);
225 static void encode_my_addr(struct ceph_messenger *msgr)
227 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
228 ceph_encode_addr(&msgr->my_enc_addr);
232 * work queue for all reading and writing to/from the socket.
234 static struct workqueue_struct *ceph_msgr_wq;
236 static int ceph_msgr_slab_init(void)
238 BUG_ON(ceph_msg_cache);
239 ceph_msg_cache = KMEM_CACHE(ceph_msg, 0);
243 BUG_ON(ceph_msg_data_cache);
244 ceph_msg_data_cache = KMEM_CACHE(ceph_msg_data, 0);
245 if (ceph_msg_data_cache)
248 kmem_cache_destroy(ceph_msg_cache);
249 ceph_msg_cache = NULL;
254 static void ceph_msgr_slab_exit(void)
256 BUG_ON(!ceph_msg_data_cache);
257 kmem_cache_destroy(ceph_msg_data_cache);
258 ceph_msg_data_cache = NULL;
260 BUG_ON(!ceph_msg_cache);
261 kmem_cache_destroy(ceph_msg_cache);
262 ceph_msg_cache = NULL;
265 static void _ceph_msgr_exit(void)
268 destroy_workqueue(ceph_msgr_wq);
272 BUG_ON(zero_page == NULL);
276 ceph_msgr_slab_exit();
279 int ceph_msgr_init(void)
281 if (ceph_msgr_slab_init())
284 BUG_ON(zero_page != NULL);
285 zero_page = ZERO_PAGE(0);
289 * The number of active work items is limited by the number of
290 * connections, so leave @max_active at default.
292 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
296 pr_err("msgr_init failed to create workqueue\n");
301 EXPORT_SYMBOL(ceph_msgr_init);
303 void ceph_msgr_exit(void)
305 BUG_ON(ceph_msgr_wq == NULL);
309 EXPORT_SYMBOL(ceph_msgr_exit);
311 void ceph_msgr_flush(void)
313 flush_workqueue(ceph_msgr_wq);
315 EXPORT_SYMBOL(ceph_msgr_flush);
317 /* Connection socket state transition functions */
319 static void con_sock_state_init(struct ceph_connection *con)
323 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
324 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
325 printk("%s: unexpected old state %d\n", __func__, old_state);
326 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
327 CON_SOCK_STATE_CLOSED);
330 static void con_sock_state_connecting(struct ceph_connection *con)
334 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
335 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
336 printk("%s: unexpected old state %d\n", __func__, old_state);
337 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
338 CON_SOCK_STATE_CONNECTING);
341 static void con_sock_state_connected(struct ceph_connection *con)
345 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
346 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
347 printk("%s: unexpected old state %d\n", __func__, old_state);
348 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
349 CON_SOCK_STATE_CONNECTED);
352 static void con_sock_state_closing(struct ceph_connection *con)
356 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
357 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
358 old_state != CON_SOCK_STATE_CONNECTED &&
359 old_state != CON_SOCK_STATE_CLOSING))
360 printk("%s: unexpected old state %d\n", __func__, old_state);
361 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
362 CON_SOCK_STATE_CLOSING);
365 static void con_sock_state_closed(struct ceph_connection *con)
369 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
370 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
371 old_state != CON_SOCK_STATE_CLOSING &&
372 old_state != CON_SOCK_STATE_CONNECTING &&
373 old_state != CON_SOCK_STATE_CLOSED))
374 printk("%s: unexpected old state %d\n", __func__, old_state);
375 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
376 CON_SOCK_STATE_CLOSED);
380 * socket callback functions
383 /* data available on socket, or listen socket received a connect */
384 static void ceph_sock_data_ready(struct sock *sk)
386 struct ceph_connection *con = sk->sk_user_data;
387 if (atomic_read(&con->msgr->stopping)) {
391 if (sk->sk_state != TCP_CLOSE_WAIT) {
392 dout("%s on %p state = %lu, queueing work\n", __func__,
398 /* socket has buffer space for writing */
399 static void ceph_sock_write_space(struct sock *sk)
401 struct ceph_connection *con = sk->sk_user_data;
403 /* only queue to workqueue if there is data we want to write,
404 * and there is sufficient space in the socket buffer to accept
405 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
406 * doesn't get called again until try_write() fills the socket
407 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
408 * and net/core/stream.c:sk_stream_write_space().
410 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
411 if (sk_stream_is_writeable(sk)) {
412 dout("%s %p queueing write work\n", __func__, con);
413 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
417 dout("%s %p nothing to write\n", __func__, con);
421 /* socket's state has changed */
422 static void ceph_sock_state_change(struct sock *sk)
424 struct ceph_connection *con = sk->sk_user_data;
426 dout("%s %p state = %lu sk_state = %u\n", __func__,
427 con, con->state, sk->sk_state);
429 switch (sk->sk_state) {
431 dout("%s TCP_CLOSE\n", __func__);
433 dout("%s TCP_CLOSE_WAIT\n", __func__);
434 con_sock_state_closing(con);
435 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
438 case TCP_ESTABLISHED:
439 dout("%s TCP_ESTABLISHED\n", __func__);
440 con_sock_state_connected(con);
443 default: /* Everything else is uninteresting */
449 * set up socket callbacks
451 static void set_sock_callbacks(struct socket *sock,
452 struct ceph_connection *con)
454 struct sock *sk = sock->sk;
455 sk->sk_user_data = con;
456 sk->sk_data_ready = ceph_sock_data_ready;
457 sk->sk_write_space = ceph_sock_write_space;
458 sk->sk_state_change = ceph_sock_state_change;
467 * initiate connection to a remote socket.
469 static int ceph_tcp_connect(struct ceph_connection *con)
471 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
473 unsigned int noio_flag;
478 /* sock_create_kern() allocates with GFP_KERNEL */
479 noio_flag = memalloc_noio_save();
480 ret = sock_create_kern(read_pnet(&con->msgr->net), paddr->ss_family,
481 SOCK_STREAM, IPPROTO_TCP, &sock);
482 memalloc_noio_restore(noio_flag);
485 sock->sk->sk_allocation = GFP_NOFS;
487 #ifdef CONFIG_LOCKDEP
488 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
491 set_sock_callbacks(sock, con);
493 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
495 con_sock_state_connecting(con);
496 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
498 if (ret == -EINPROGRESS) {
499 dout("connect %s EINPROGRESS sk_state = %u\n",
500 ceph_pr_addr(&con->peer_addr.in_addr),
502 } else if (ret < 0) {
503 pr_err("connect %s error %d\n",
504 ceph_pr_addr(&con->peer_addr.in_addr), ret);
509 if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY)) {
512 ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
513 (char *)&optval, sizeof(optval));
515 pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
523 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
525 struct kvec iov = {buf, len};
526 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
529 iov_iter_kvec(&msg.msg_iter, READ | ITER_KVEC, &iov, 1, len);
530 r = sock_recvmsg(sock, &msg, msg.msg_flags);
536 static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
537 int page_offset, size_t length)
539 struct bio_vec bvec = {
541 .bv_offset = page_offset,
544 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
547 BUG_ON(page_offset + length > PAGE_SIZE);
548 iov_iter_bvec(&msg.msg_iter, READ | ITER_BVEC, &bvec, 1, length);
549 r = sock_recvmsg(sock, &msg, msg.msg_flags);
556 * write something. @more is true if caller will be sending more data
559 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
560 size_t kvlen, size_t len, int more)
562 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
566 msg.msg_flags |= MSG_MORE;
568 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
570 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
576 static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
577 int offset, size_t size, bool more)
579 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
582 ret = kernel_sendpage(sock, page, offset, size, flags);
589 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
590 int offset, size_t size, bool more)
592 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
596 /* sendpage cannot properly handle pages with page_count == 0,
597 * we need to fallback to sendmsg if that's the case */
598 if (page_count(page) >= 1)
599 return __ceph_tcp_sendpage(sock, page, offset, size, more);
602 bvec.bv_offset = offset;
606 msg.msg_flags |= MSG_MORE;
608 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
610 iov_iter_bvec(&msg.msg_iter, WRITE | ITER_BVEC, &bvec, 1, size);
611 ret = sock_sendmsg(sock, &msg);
619 * Shutdown/close the socket for the given connection.
621 static int con_close_socket(struct ceph_connection *con)
625 dout("con_close_socket on %p sock %p\n", con, con->sock);
627 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
628 sock_release(con->sock);
633 * Forcibly clear the SOCK_CLOSED flag. It gets set
634 * independent of the connection mutex, and we could have
635 * received a socket close event before we had the chance to
636 * shut the socket down.
638 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
640 con_sock_state_closed(con);
645 * Reset a connection. Discard all incoming and outgoing messages
646 * and clear *_seq state.
648 static void ceph_msg_remove(struct ceph_msg *msg)
650 list_del_init(&msg->list_head);
654 static void ceph_msg_remove_list(struct list_head *head)
656 while (!list_empty(head)) {
657 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
659 ceph_msg_remove(msg);
663 static void reset_connection(struct ceph_connection *con)
665 /* reset connection, out_queue, msg_ and connect_seq */
666 /* discard existing out_queue and msg_seq */
667 dout("reset_connection %p\n", con);
668 ceph_msg_remove_list(&con->out_queue);
669 ceph_msg_remove_list(&con->out_sent);
672 BUG_ON(con->in_msg->con != con);
673 ceph_msg_put(con->in_msg);
677 con->connect_seq = 0;
680 BUG_ON(con->out_msg->con != con);
681 ceph_msg_put(con->out_msg);
685 con->in_seq_acked = 0;
691 * mark a peer down. drop any open connections.
693 void ceph_con_close(struct ceph_connection *con)
695 mutex_lock(&con->mutex);
696 dout("con_close %p peer %s\n", con,
697 ceph_pr_addr(&con->peer_addr.in_addr));
698 con->state = CON_STATE_CLOSED;
700 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
701 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
702 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
703 con_flag_clear(con, CON_FLAG_BACKOFF);
705 reset_connection(con);
706 con->peer_global_seq = 0;
708 con_close_socket(con);
709 mutex_unlock(&con->mutex);
711 EXPORT_SYMBOL(ceph_con_close);
714 * Reopen a closed connection, with a new peer address.
716 void ceph_con_open(struct ceph_connection *con,
717 __u8 entity_type, __u64 entity_num,
718 struct ceph_entity_addr *addr)
720 mutex_lock(&con->mutex);
721 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
723 WARN_ON(con->state != CON_STATE_CLOSED);
724 con->state = CON_STATE_PREOPEN;
726 con->peer_name.type = (__u8) entity_type;
727 con->peer_name.num = cpu_to_le64(entity_num);
729 memcpy(&con->peer_addr, addr, sizeof(*addr));
730 con->delay = 0; /* reset backoff memory */
731 mutex_unlock(&con->mutex);
734 EXPORT_SYMBOL(ceph_con_open);
737 * return true if this connection ever successfully opened
739 bool ceph_con_opened(struct ceph_connection *con)
741 return con->connect_seq > 0;
745 * initialize a new connection.
747 void ceph_con_init(struct ceph_connection *con, void *private,
748 const struct ceph_connection_operations *ops,
749 struct ceph_messenger *msgr)
751 dout("con_init %p\n", con);
752 memset(con, 0, sizeof(*con));
753 con->private = private;
757 con_sock_state_init(con);
759 mutex_init(&con->mutex);
760 INIT_LIST_HEAD(&con->out_queue);
761 INIT_LIST_HEAD(&con->out_sent);
762 INIT_DELAYED_WORK(&con->work, ceph_con_workfn);
764 con->state = CON_STATE_CLOSED;
766 EXPORT_SYMBOL(ceph_con_init);
770 * We maintain a global counter to order connection attempts. Get
771 * a unique seq greater than @gt.
773 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
777 spin_lock(&msgr->global_seq_lock);
778 if (msgr->global_seq < gt)
779 msgr->global_seq = gt;
780 ret = ++msgr->global_seq;
781 spin_unlock(&msgr->global_seq_lock);
785 static void con_out_kvec_reset(struct ceph_connection *con)
787 BUG_ON(con->out_skip);
789 con->out_kvec_left = 0;
790 con->out_kvec_bytes = 0;
791 con->out_kvec_cur = &con->out_kvec[0];
794 static void con_out_kvec_add(struct ceph_connection *con,
795 size_t size, void *data)
797 int index = con->out_kvec_left;
799 BUG_ON(con->out_skip);
800 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
802 con->out_kvec[index].iov_len = size;
803 con->out_kvec[index].iov_base = data;
804 con->out_kvec_left++;
805 con->out_kvec_bytes += size;
809 * Chop off a kvec from the end. Return residual number of bytes for
810 * that kvec, i.e. how many bytes would have been written if the kvec
813 static int con_out_kvec_skip(struct ceph_connection *con)
815 int off = con->out_kvec_cur - con->out_kvec;
818 if (con->out_kvec_bytes > 0) {
819 skip = con->out_kvec[off + con->out_kvec_left - 1].iov_len;
820 BUG_ON(con->out_kvec_bytes < skip);
821 BUG_ON(!con->out_kvec_left);
822 con->out_kvec_bytes -= skip;
823 con->out_kvec_left--;
832 * For a bio data item, a piece is whatever remains of the next
833 * entry in the current bio iovec, or the first entry in the next
836 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
839 struct ceph_msg_data *data = cursor->data;
842 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
847 cursor->resid = min(length, data->bio_length);
849 cursor->bvec_iter = bio->bi_iter;
851 cursor->resid <= bio_iter_len(bio, cursor->bvec_iter);
854 static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
858 struct ceph_msg_data *data = cursor->data;
860 struct bio_vec bio_vec;
862 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
867 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
869 *page_offset = (size_t) bio_vec.bv_offset;
870 BUG_ON(*page_offset >= PAGE_SIZE);
871 if (cursor->last_piece) /* pagelist offset is always 0 */
872 *length = cursor->resid;
874 *length = (size_t) bio_vec.bv_len;
875 BUG_ON(*length > cursor->resid);
876 BUG_ON(*page_offset + *length > PAGE_SIZE);
878 return bio_vec.bv_page;
881 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
885 struct bio_vec bio_vec;
887 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
892 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
894 /* Advance the cursor offset */
896 BUG_ON(cursor->resid < bytes);
897 cursor->resid -= bytes;
899 bio_advance_iter(bio, &cursor->bvec_iter, bytes);
901 if (bytes < bio_vec.bv_len)
902 return false; /* more bytes to process in this segment */
904 /* Move on to the next segment, and possibly the next bio */
906 if (!cursor->bvec_iter.bi_size) {
910 cursor->bvec_iter = bio->bi_iter;
912 memset(&cursor->bvec_iter, 0,
913 sizeof(cursor->bvec_iter));
916 if (!cursor->last_piece) {
917 BUG_ON(!cursor->resid);
919 /* A short read is OK, so use <= rather than == */
920 if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter))
921 cursor->last_piece = true;
926 #endif /* CONFIG_BLOCK */
929 * For a page array, a piece comes from the first page in the array
930 * that has not already been fully consumed.
932 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
935 struct ceph_msg_data *data = cursor->data;
938 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
940 BUG_ON(!data->pages);
941 BUG_ON(!data->length);
943 cursor->resid = min(length, data->length);
944 page_count = calc_pages_for(data->alignment, (u64)data->length);
945 cursor->page_offset = data->alignment & ~PAGE_MASK;
946 cursor->page_index = 0;
947 BUG_ON(page_count > (int)USHRT_MAX);
948 cursor->page_count = (unsigned short)page_count;
949 BUG_ON(length > SIZE_MAX - cursor->page_offset);
950 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
954 ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
955 size_t *page_offset, size_t *length)
957 struct ceph_msg_data *data = cursor->data;
959 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
961 BUG_ON(cursor->page_index >= cursor->page_count);
962 BUG_ON(cursor->page_offset >= PAGE_SIZE);
964 *page_offset = cursor->page_offset;
965 if (cursor->last_piece)
966 *length = cursor->resid;
968 *length = PAGE_SIZE - *page_offset;
970 return data->pages[cursor->page_index];
973 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
976 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
978 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
980 /* Advance the cursor page offset */
982 cursor->resid -= bytes;
983 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
984 if (!bytes || cursor->page_offset)
985 return false; /* more bytes to process in the current page */
988 return false; /* no more data */
990 /* Move on to the next page; offset is already at 0 */
992 BUG_ON(cursor->page_index >= cursor->page_count);
993 cursor->page_index++;
994 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1000 * For a pagelist, a piece is whatever remains to be consumed in the
1001 * first page in the list, or the front of the next page.
1004 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
1007 struct ceph_msg_data *data = cursor->data;
1008 struct ceph_pagelist *pagelist;
1011 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1013 pagelist = data->pagelist;
1017 return; /* pagelist can be assigned but empty */
1019 BUG_ON(list_empty(&pagelist->head));
1020 page = list_first_entry(&pagelist->head, struct page, lru);
1022 cursor->resid = min(length, pagelist->length);
1023 cursor->page = page;
1025 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1028 static struct page *
1029 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
1030 size_t *page_offset, size_t *length)
1032 struct ceph_msg_data *data = cursor->data;
1033 struct ceph_pagelist *pagelist;
1035 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1037 pagelist = data->pagelist;
1040 BUG_ON(!cursor->page);
1041 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1043 /* offset of first page in pagelist is always 0 */
1044 *page_offset = cursor->offset & ~PAGE_MASK;
1045 if (cursor->last_piece)
1046 *length = cursor->resid;
1048 *length = PAGE_SIZE - *page_offset;
1050 return cursor->page;
1053 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
1056 struct ceph_msg_data *data = cursor->data;
1057 struct ceph_pagelist *pagelist;
1059 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1061 pagelist = data->pagelist;
1064 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1065 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1067 /* Advance the cursor offset */
1069 cursor->resid -= bytes;
1070 cursor->offset += bytes;
1071 /* offset of first page in pagelist is always 0 */
1072 if (!bytes || cursor->offset & ~PAGE_MASK)
1073 return false; /* more bytes to process in the current page */
1076 return false; /* no more data */
1078 /* Move on to the next page */
1080 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1081 cursor->page = list_next_entry(cursor->page, lru);
1082 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1088 * Message data is handled (sent or received) in pieces, where each
1089 * piece resides on a single page. The network layer might not
1090 * consume an entire piece at once. A data item's cursor keeps
1091 * track of which piece is next to process and how much remains to
1092 * be processed in that piece. It also tracks whether the current
1093 * piece is the last one in the data item.
1095 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1097 size_t length = cursor->total_resid;
1099 switch (cursor->data->type) {
1100 case CEPH_MSG_DATA_PAGELIST:
1101 ceph_msg_data_pagelist_cursor_init(cursor, length);
1103 case CEPH_MSG_DATA_PAGES:
1104 ceph_msg_data_pages_cursor_init(cursor, length);
1107 case CEPH_MSG_DATA_BIO:
1108 ceph_msg_data_bio_cursor_init(cursor, length);
1110 #endif /* CONFIG_BLOCK */
1111 case CEPH_MSG_DATA_NONE:
1116 cursor->need_crc = true;
1119 static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1121 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1122 struct ceph_msg_data *data;
1125 BUG_ON(length > msg->data_length);
1126 BUG_ON(list_empty(&msg->data));
1128 cursor->data_head = &msg->data;
1129 cursor->total_resid = length;
1130 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1131 cursor->data = data;
1133 __ceph_msg_data_cursor_init(cursor);
1137 * Return the page containing the next piece to process for a given
1138 * data item, and supply the page offset and length of that piece.
1139 * Indicate whether this is the last piece in this data item.
1141 static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1142 size_t *page_offset, size_t *length,
1147 switch (cursor->data->type) {
1148 case CEPH_MSG_DATA_PAGELIST:
1149 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1151 case CEPH_MSG_DATA_PAGES:
1152 page = ceph_msg_data_pages_next(cursor, page_offset, length);
1155 case CEPH_MSG_DATA_BIO:
1156 page = ceph_msg_data_bio_next(cursor, page_offset, length);
1158 #endif /* CONFIG_BLOCK */
1159 case CEPH_MSG_DATA_NONE:
1165 BUG_ON(*page_offset + *length > PAGE_SIZE);
1168 *last_piece = cursor->last_piece;
1174 * Returns true if the result moves the cursor on to the next piece
1177 static void ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1182 BUG_ON(bytes > cursor->resid);
1183 switch (cursor->data->type) {
1184 case CEPH_MSG_DATA_PAGELIST:
1185 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1187 case CEPH_MSG_DATA_PAGES:
1188 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1191 case CEPH_MSG_DATA_BIO:
1192 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1194 #endif /* CONFIG_BLOCK */
1195 case CEPH_MSG_DATA_NONE:
1200 cursor->total_resid -= bytes;
1202 if (!cursor->resid && cursor->total_resid) {
1203 WARN_ON(!cursor->last_piece);
1204 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1205 cursor->data = list_next_entry(cursor->data, links);
1206 __ceph_msg_data_cursor_init(cursor);
1209 cursor->need_crc = new_piece;
1212 static size_t sizeof_footer(struct ceph_connection *con)
1214 return (con->peer_features & CEPH_FEATURE_MSG_AUTH) ?
1215 sizeof(struct ceph_msg_footer) :
1216 sizeof(struct ceph_msg_footer_old);
1219 static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
1224 /* Initialize data cursor */
1226 ceph_msg_data_cursor_init(msg, (size_t)data_len);
1230 * Prepare footer for currently outgoing message, and finish things
1231 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1233 static void prepare_write_message_footer(struct ceph_connection *con)
1235 struct ceph_msg *m = con->out_msg;
1237 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1239 dout("prepare_write_message_footer %p\n", con);
1240 con_out_kvec_add(con, sizeof_footer(con), &m->footer);
1241 if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
1242 if (con->ops->sign_message)
1243 con->ops->sign_message(m);
1247 m->old_footer.flags = m->footer.flags;
1249 con->out_more = m->more_to_follow;
1250 con->out_msg_done = true;
1254 * Prepare headers for the next outgoing message.
1256 static void prepare_write_message(struct ceph_connection *con)
1261 con_out_kvec_reset(con);
1262 con->out_msg_done = false;
1264 /* Sneak an ack in there first? If we can get it into the same
1265 * TCP packet that's a good thing. */
1266 if (con->in_seq > con->in_seq_acked) {
1267 con->in_seq_acked = con->in_seq;
1268 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1269 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1270 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1271 &con->out_temp_ack);
1274 BUG_ON(list_empty(&con->out_queue));
1275 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1277 BUG_ON(m->con != con);
1279 /* put message on sent list */
1281 list_move_tail(&m->list_head, &con->out_sent);
1284 * only assign outgoing seq # if we haven't sent this message
1285 * yet. if it is requeued, resend with it's original seq.
1287 if (m->needs_out_seq) {
1288 m->hdr.seq = cpu_to_le64(++con->out_seq);
1289 m->needs_out_seq = false;
1291 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
1293 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1294 m, con->out_seq, le16_to_cpu(m->hdr.type),
1295 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1297 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1299 /* tag + hdr + front + middle */
1300 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1301 con_out_kvec_add(con, sizeof(con->out_hdr), &con->out_hdr);
1302 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1305 con_out_kvec_add(con, m->middle->vec.iov_len,
1306 m->middle->vec.iov_base);
1308 /* fill in hdr crc and finalize hdr */
1309 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1310 con->out_msg->hdr.crc = cpu_to_le32(crc);
1311 memcpy(&con->out_hdr, &con->out_msg->hdr, sizeof(con->out_hdr));
1313 /* fill in front and middle crc, footer */
1314 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1315 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1317 crc = crc32c(0, m->middle->vec.iov_base,
1318 m->middle->vec.iov_len);
1319 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1321 con->out_msg->footer.middle_crc = 0;
1322 dout("%s front_crc %u middle_crc %u\n", __func__,
1323 le32_to_cpu(con->out_msg->footer.front_crc),
1324 le32_to_cpu(con->out_msg->footer.middle_crc));
1325 con->out_msg->footer.flags = 0;
1327 /* is there a data payload? */
1328 con->out_msg->footer.data_crc = 0;
1329 if (m->data_length) {
1330 prepare_message_data(con->out_msg, m->data_length);
1331 con->out_more = 1; /* data + footer will follow */
1333 /* no, queue up footer too and be done */
1334 prepare_write_message_footer(con);
1337 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1343 static void prepare_write_ack(struct ceph_connection *con)
1345 dout("prepare_write_ack %p %llu -> %llu\n", con,
1346 con->in_seq_acked, con->in_seq);
1347 con->in_seq_acked = con->in_seq;
1349 con_out_kvec_reset(con);
1351 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1353 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1354 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1355 &con->out_temp_ack);
1357 con->out_more = 1; /* more will follow.. eventually.. */
1358 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1362 * Prepare to share the seq during handshake
1364 static void prepare_write_seq(struct ceph_connection *con)
1366 dout("prepare_write_seq %p %llu -> %llu\n", con,
1367 con->in_seq_acked, con->in_seq);
1368 con->in_seq_acked = con->in_seq;
1370 con_out_kvec_reset(con);
1372 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1373 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1374 &con->out_temp_ack);
1376 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1380 * Prepare to write keepalive byte.
1382 static void prepare_write_keepalive(struct ceph_connection *con)
1384 dout("prepare_write_keepalive %p\n", con);
1385 con_out_kvec_reset(con);
1386 if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
1387 struct timespec now;
1389 ktime_get_real_ts(&now);
1390 con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
1391 ceph_encode_timespec(&con->out_temp_keepalive2, &now);
1392 con_out_kvec_add(con, sizeof(con->out_temp_keepalive2),
1393 &con->out_temp_keepalive2);
1395 con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive);
1397 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1401 * Connection negotiation.
1404 static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1407 struct ceph_auth_handshake *auth;
1409 if (!con->ops->get_authorizer) {
1410 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1411 con->out_connect.authorizer_len = 0;
1415 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
1419 con->auth_reply_buf = auth->authorizer_reply_buf;
1420 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
1425 * We connected to a peer and are saying hello.
1427 static void prepare_write_banner(struct ceph_connection *con)
1429 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1430 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1431 &con->msgr->my_enc_addr);
1434 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1437 static int prepare_write_connect(struct ceph_connection *con)
1439 unsigned int global_seq = get_global_seq(con->msgr, 0);
1442 struct ceph_auth_handshake *auth;
1444 switch (con->peer_name.type) {
1445 case CEPH_ENTITY_TYPE_MON:
1446 proto = CEPH_MONC_PROTOCOL;
1448 case CEPH_ENTITY_TYPE_OSD:
1449 proto = CEPH_OSDC_PROTOCOL;
1451 case CEPH_ENTITY_TYPE_MDS:
1452 proto = CEPH_MDSC_PROTOCOL;
1458 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1459 con->connect_seq, global_seq, proto);
1461 con->out_connect.features =
1462 cpu_to_le64(from_msgr(con->msgr)->supported_features);
1463 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1464 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1465 con->out_connect.global_seq = cpu_to_le32(global_seq);
1466 con->out_connect.protocol_version = cpu_to_le32(proto);
1467 con->out_connect.flags = 0;
1469 auth_proto = CEPH_AUTH_UNKNOWN;
1470 auth = get_connect_authorizer(con, &auth_proto);
1472 return PTR_ERR(auth);
1474 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1475 con->out_connect.authorizer_len = auth ?
1476 cpu_to_le32(auth->authorizer_buf_len) : 0;
1478 con_out_kvec_add(con, sizeof (con->out_connect),
1480 if (auth && auth->authorizer_buf_len)
1481 con_out_kvec_add(con, auth->authorizer_buf_len,
1482 auth->authorizer_buf);
1485 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1491 * write as much of pending kvecs to the socket as we can.
1493 * 0 -> socket full, but more to do
1496 static int write_partial_kvec(struct ceph_connection *con)
1500 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1501 while (con->out_kvec_bytes > 0) {
1502 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1503 con->out_kvec_left, con->out_kvec_bytes,
1507 con->out_kvec_bytes -= ret;
1508 if (con->out_kvec_bytes == 0)
1511 /* account for full iov entries consumed */
1512 while (ret >= con->out_kvec_cur->iov_len) {
1513 BUG_ON(!con->out_kvec_left);
1514 ret -= con->out_kvec_cur->iov_len;
1515 con->out_kvec_cur++;
1516 con->out_kvec_left--;
1518 /* and for a partially-consumed entry */
1520 con->out_kvec_cur->iov_len -= ret;
1521 con->out_kvec_cur->iov_base += ret;
1524 con->out_kvec_left = 0;
1527 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1528 con->out_kvec_bytes, con->out_kvec_left, ret);
1529 return ret; /* done! */
1532 static u32 ceph_crc32c_page(u32 crc, struct page *page,
1533 unsigned int page_offset,
1534 unsigned int length)
1539 BUG_ON(kaddr == NULL);
1540 crc = crc32c(crc, kaddr + page_offset, length);
1546 * Write as much message data payload as we can. If we finish, queue
1548 * 1 -> done, footer is now queued in out_kvec[].
1549 * 0 -> socket full, but more to do
1552 static int write_partial_message_data(struct ceph_connection *con)
1554 struct ceph_msg *msg = con->out_msg;
1555 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1556 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
1559 dout("%s %p msg %p\n", __func__, con, msg);
1561 if (list_empty(&msg->data))
1565 * Iterate through each page that contains data to be
1566 * written, and send as much as possible for each.
1568 * If we are calculating the data crc (the default), we will
1569 * need to map the page. If we have no pages, they have
1570 * been revoked, so use the zero page.
1572 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1573 while (cursor->resid) {
1580 page = ceph_msg_data_next(cursor, &page_offset, &length,
1582 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
1583 length, !last_piece);
1586 msg->footer.data_crc = cpu_to_le32(crc);
1590 if (do_datacrc && cursor->need_crc)
1591 crc = ceph_crc32c_page(crc, page, page_offset, length);
1592 ceph_msg_data_advance(cursor, (size_t)ret);
1595 dout("%s %p msg %p done\n", __func__, con, msg);
1597 /* prepare and queue up footer, too */
1599 msg->footer.data_crc = cpu_to_le32(crc);
1601 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1602 con_out_kvec_reset(con);
1603 prepare_write_message_footer(con);
1605 return 1; /* must return > 0 to indicate success */
1611 static int write_partial_skip(struct ceph_connection *con)
1615 dout("%s %p %d left\n", __func__, con, con->out_skip);
1616 while (con->out_skip > 0) {
1617 size_t size = min(con->out_skip, (int) PAGE_SIZE);
1619 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
1622 con->out_skip -= ret;
1630 * Prepare to read connection handshake, or an ack.
1632 static void prepare_read_banner(struct ceph_connection *con)
1634 dout("prepare_read_banner %p\n", con);
1635 con->in_base_pos = 0;
1638 static void prepare_read_connect(struct ceph_connection *con)
1640 dout("prepare_read_connect %p\n", con);
1641 con->in_base_pos = 0;
1644 static void prepare_read_ack(struct ceph_connection *con)
1646 dout("prepare_read_ack %p\n", con);
1647 con->in_base_pos = 0;
1650 static void prepare_read_seq(struct ceph_connection *con)
1652 dout("prepare_read_seq %p\n", con);
1653 con->in_base_pos = 0;
1654 con->in_tag = CEPH_MSGR_TAG_SEQ;
1657 static void prepare_read_tag(struct ceph_connection *con)
1659 dout("prepare_read_tag %p\n", con);
1660 con->in_base_pos = 0;
1661 con->in_tag = CEPH_MSGR_TAG_READY;
1664 static void prepare_read_keepalive_ack(struct ceph_connection *con)
1666 dout("prepare_read_keepalive_ack %p\n", con);
1667 con->in_base_pos = 0;
1671 * Prepare to read a message.
1673 static int prepare_read_message(struct ceph_connection *con)
1675 dout("prepare_read_message %p\n", con);
1676 BUG_ON(con->in_msg != NULL);
1677 con->in_base_pos = 0;
1678 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1683 static int read_partial(struct ceph_connection *con,
1684 int end, int size, void *object)
1686 while (con->in_base_pos < end) {
1687 int left = end - con->in_base_pos;
1688 int have = size - left;
1689 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1692 con->in_base_pos += ret;
1699 * Read all or part of the connect-side handshake on a new connection
1701 static int read_partial_banner(struct ceph_connection *con)
1707 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1710 size = strlen(CEPH_BANNER);
1712 ret = read_partial(con, end, size, con->in_banner);
1716 size = sizeof (con->actual_peer_addr);
1718 ret = read_partial(con, end, size, &con->actual_peer_addr);
1722 size = sizeof (con->peer_addr_for_me);
1724 ret = read_partial(con, end, size, &con->peer_addr_for_me);
1732 static int read_partial_connect(struct ceph_connection *con)
1738 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1740 size = sizeof (con->in_reply);
1742 ret = read_partial(con, end, size, &con->in_reply);
1746 size = le32_to_cpu(con->in_reply.authorizer_len);
1748 ret = read_partial(con, end, size, con->auth_reply_buf);
1752 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1753 con, (int)con->in_reply.tag,
1754 le32_to_cpu(con->in_reply.connect_seq),
1755 le32_to_cpu(con->in_reply.global_seq));
1762 * Verify the hello banner looks okay.
1764 static int verify_hello(struct ceph_connection *con)
1766 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1767 pr_err("connect to %s got bad banner\n",
1768 ceph_pr_addr(&con->peer_addr.in_addr));
1769 con->error_msg = "protocol error, bad banner";
1775 static bool addr_is_blank(struct sockaddr_storage *ss)
1777 struct in_addr *addr = &((struct sockaddr_in *)ss)->sin_addr;
1778 struct in6_addr *addr6 = &((struct sockaddr_in6 *)ss)->sin6_addr;
1780 switch (ss->ss_family) {
1782 return addr->s_addr == htonl(INADDR_ANY);
1784 return ipv6_addr_any(addr6);
1790 static int addr_port(struct sockaddr_storage *ss)
1792 switch (ss->ss_family) {
1794 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1796 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1801 static void addr_set_port(struct sockaddr_storage *ss, int p)
1803 switch (ss->ss_family) {
1805 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1808 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1814 * Unlike other *_pton function semantics, zero indicates success.
1816 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1817 char delim, const char **ipend)
1819 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1820 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1822 memset(ss, 0, sizeof(*ss));
1824 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1825 ss->ss_family = AF_INET;
1829 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1830 ss->ss_family = AF_INET6;
1838 * Extract hostname string and resolve using kernel DNS facility.
1840 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1841 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1842 struct sockaddr_storage *ss, char delim, const char **ipend)
1844 const char *end, *delim_p;
1845 char *colon_p, *ip_addr = NULL;
1849 * The end of the hostname occurs immediately preceding the delimiter or
1850 * the port marker (':') where the delimiter takes precedence.
1852 delim_p = memchr(name, delim, namelen);
1853 colon_p = memchr(name, ':', namelen);
1855 if (delim_p && colon_p)
1856 end = delim_p < colon_p ? delim_p : colon_p;
1857 else if (!delim_p && colon_p)
1861 if (!end) /* case: hostname:/ */
1862 end = name + namelen;
1868 /* do dns_resolve upcall */
1869 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1871 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1879 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1880 ret, ret ? "failed" : ceph_pr_addr(ss));
1885 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1886 struct sockaddr_storage *ss, char delim, const char **ipend)
1893 * Parse a server name (IP or hostname). If a valid IP address is not found
1894 * then try to extract a hostname to resolve using userspace DNS upcall.
1896 static int ceph_parse_server_name(const char *name, size_t namelen,
1897 struct sockaddr_storage *ss, char delim, const char **ipend)
1901 ret = ceph_pton(name, namelen, ss, delim, ipend);
1903 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1909 * Parse an ip[:port] list into an addr array. Use the default
1910 * monitor port if a port isn't specified.
1912 int ceph_parse_ips(const char *c, const char *end,
1913 struct ceph_entity_addr *addr,
1914 int max_count, int *count)
1916 int i, ret = -EINVAL;
1919 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1920 for (i = 0; i < max_count; i++) {
1922 struct sockaddr_storage *ss = &addr[i].in_addr;
1931 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1940 dout("missing matching ']'\n");
1947 if (p < end && *p == ':') {
1950 while (p < end && *p >= '0' && *p <= '9') {
1951 port = (port * 10) + (*p - '0');
1955 port = CEPH_MON_PORT;
1956 else if (port > 65535)
1959 port = CEPH_MON_PORT;
1962 addr_set_port(ss, port);
1964 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1981 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1984 EXPORT_SYMBOL(ceph_parse_ips);
1986 static int process_banner(struct ceph_connection *con)
1988 dout("process_banner on %p\n", con);
1990 if (verify_hello(con) < 0)
1993 ceph_decode_addr(&con->actual_peer_addr);
1994 ceph_decode_addr(&con->peer_addr_for_me);
1997 * Make sure the other end is who we wanted. note that the other
1998 * end may not yet know their ip address, so if it's 0.0.0.0, give
1999 * them the benefit of the doubt.
2001 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
2002 sizeof(con->peer_addr)) != 0 &&
2003 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
2004 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
2005 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
2006 ceph_pr_addr(&con->peer_addr.in_addr),
2007 (int)le32_to_cpu(con->peer_addr.nonce),
2008 ceph_pr_addr(&con->actual_peer_addr.in_addr),
2009 (int)le32_to_cpu(con->actual_peer_addr.nonce));
2010 con->error_msg = "wrong peer at address";
2015 * did we learn our address?
2017 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
2018 int port = addr_port(&con->msgr->inst.addr.in_addr);
2020 memcpy(&con->msgr->inst.addr.in_addr,
2021 &con->peer_addr_for_me.in_addr,
2022 sizeof(con->peer_addr_for_me.in_addr));
2023 addr_set_port(&con->msgr->inst.addr.in_addr, port);
2024 encode_my_addr(con->msgr);
2025 dout("process_banner learned my addr is %s\n",
2026 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
2032 static int process_connect(struct ceph_connection *con)
2034 u64 sup_feat = from_msgr(con->msgr)->supported_features;
2035 u64 req_feat = from_msgr(con->msgr)->required_features;
2036 u64 server_feat = ceph_sanitize_features(
2037 le64_to_cpu(con->in_reply.features));
2040 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
2042 if (con->auth_reply_buf) {
2044 * Any connection that defines ->get_authorizer()
2045 * should also define ->verify_authorizer_reply().
2046 * See get_connect_authorizer().
2048 ret = con->ops->verify_authorizer_reply(con);
2050 con->error_msg = "bad authorize reply";
2055 switch (con->in_reply.tag) {
2056 case CEPH_MSGR_TAG_FEATURES:
2057 pr_err("%s%lld %s feature set mismatch,"
2058 " my %llx < server's %llx, missing %llx\n",
2059 ENTITY_NAME(con->peer_name),
2060 ceph_pr_addr(&con->peer_addr.in_addr),
2061 sup_feat, server_feat, server_feat & ~sup_feat);
2062 con->error_msg = "missing required protocol features";
2063 reset_connection(con);
2066 case CEPH_MSGR_TAG_BADPROTOVER:
2067 pr_err("%s%lld %s protocol version mismatch,"
2068 " my %d != server's %d\n",
2069 ENTITY_NAME(con->peer_name),
2070 ceph_pr_addr(&con->peer_addr.in_addr),
2071 le32_to_cpu(con->out_connect.protocol_version),
2072 le32_to_cpu(con->in_reply.protocol_version));
2073 con->error_msg = "protocol version mismatch";
2074 reset_connection(con);
2077 case CEPH_MSGR_TAG_BADAUTHORIZER:
2079 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2081 if (con->auth_retry == 2) {
2082 con->error_msg = "connect authorization failure";
2085 con_out_kvec_reset(con);
2086 ret = prepare_write_connect(con);
2089 prepare_read_connect(con);
2092 case CEPH_MSGR_TAG_RESETSESSION:
2094 * If we connected with a large connect_seq but the peer
2095 * has no record of a session with us (no connection, or
2096 * connect_seq == 0), they will send RESETSESION to indicate
2097 * that they must have reset their session, and may have
2100 dout("process_connect got RESET peer seq %u\n",
2101 le32_to_cpu(con->in_reply.connect_seq));
2102 pr_err("%s%lld %s connection reset\n",
2103 ENTITY_NAME(con->peer_name),
2104 ceph_pr_addr(&con->peer_addr.in_addr));
2105 reset_connection(con);
2106 con_out_kvec_reset(con);
2107 ret = prepare_write_connect(con);
2110 prepare_read_connect(con);
2112 /* Tell ceph about it. */
2113 mutex_unlock(&con->mutex);
2114 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2115 if (con->ops->peer_reset)
2116 con->ops->peer_reset(con);
2117 mutex_lock(&con->mutex);
2118 if (con->state != CON_STATE_NEGOTIATING)
2122 case CEPH_MSGR_TAG_RETRY_SESSION:
2124 * If we sent a smaller connect_seq than the peer has, try
2125 * again with a larger value.
2127 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2128 le32_to_cpu(con->out_connect.connect_seq),
2129 le32_to_cpu(con->in_reply.connect_seq));
2130 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2131 con_out_kvec_reset(con);
2132 ret = prepare_write_connect(con);
2135 prepare_read_connect(con);
2138 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2140 * If we sent a smaller global_seq than the peer has, try
2141 * again with a larger value.
2143 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2144 con->peer_global_seq,
2145 le32_to_cpu(con->in_reply.global_seq));
2146 get_global_seq(con->msgr,
2147 le32_to_cpu(con->in_reply.global_seq));
2148 con_out_kvec_reset(con);
2149 ret = prepare_write_connect(con);
2152 prepare_read_connect(con);
2155 case CEPH_MSGR_TAG_SEQ:
2156 case CEPH_MSGR_TAG_READY:
2157 if (req_feat & ~server_feat) {
2158 pr_err("%s%lld %s protocol feature mismatch,"
2159 " my required %llx > server's %llx, need %llx\n",
2160 ENTITY_NAME(con->peer_name),
2161 ceph_pr_addr(&con->peer_addr.in_addr),
2162 req_feat, server_feat, req_feat & ~server_feat);
2163 con->error_msg = "missing required protocol features";
2164 reset_connection(con);
2168 WARN_ON(con->state != CON_STATE_NEGOTIATING);
2169 con->state = CON_STATE_OPEN;
2170 con->auth_retry = 0; /* we authenticated; clear flag */
2171 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2173 con->peer_features = server_feat;
2174 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2175 con->peer_global_seq,
2176 le32_to_cpu(con->in_reply.connect_seq),
2178 WARN_ON(con->connect_seq !=
2179 le32_to_cpu(con->in_reply.connect_seq));
2181 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2182 con_flag_set(con, CON_FLAG_LOSSYTX);
2184 con->delay = 0; /* reset backoff memory */
2186 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2187 prepare_write_seq(con);
2188 prepare_read_seq(con);
2190 prepare_read_tag(con);
2194 case CEPH_MSGR_TAG_WAIT:
2196 * If there is a connection race (we are opening
2197 * connections to each other), one of us may just have
2198 * to WAIT. This shouldn't happen if we are the
2201 con->error_msg = "protocol error, got WAIT as client";
2205 con->error_msg = "protocol error, garbage tag during connect";
2213 * read (part of) an ack
2215 static int read_partial_ack(struct ceph_connection *con)
2217 int size = sizeof (con->in_temp_ack);
2220 return read_partial(con, end, size, &con->in_temp_ack);
2224 * We can finally discard anything that's been acked.
2226 static void process_ack(struct ceph_connection *con)
2229 u64 ack = le64_to_cpu(con->in_temp_ack);
2231 bool reconnect = (con->in_tag == CEPH_MSGR_TAG_SEQ);
2232 struct list_head *list = reconnect ? &con->out_queue : &con->out_sent;
2235 * In the reconnect case, con_fault() has requeued messages
2236 * in out_sent. We should cleanup old messages according to
2237 * the reconnect seq.
2239 while (!list_empty(list)) {
2240 m = list_first_entry(list, struct ceph_msg, list_head);
2241 if (reconnect && m->needs_out_seq)
2243 seq = le64_to_cpu(m->hdr.seq);
2246 dout("got ack for seq %llu type %d at %p\n", seq,
2247 le16_to_cpu(m->hdr.type), m);
2248 m->ack_stamp = jiffies;
2252 prepare_read_tag(con);
2256 static int read_partial_message_section(struct ceph_connection *con,
2257 struct kvec *section,
2258 unsigned int sec_len, u32 *crc)
2264 while (section->iov_len < sec_len) {
2265 BUG_ON(section->iov_base == NULL);
2266 left = sec_len - section->iov_len;
2267 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2268 section->iov_len, left);
2271 section->iov_len += ret;
2273 if (section->iov_len == sec_len)
2274 *crc = crc32c(0, section->iov_base, section->iov_len);
2279 static int read_partial_msg_data(struct ceph_connection *con)
2281 struct ceph_msg *msg = con->in_msg;
2282 struct ceph_msg_data_cursor *cursor = &msg->cursor;
2283 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2291 if (list_empty(&msg->data))
2295 crc = con->in_data_crc;
2296 while (cursor->resid) {
2297 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
2298 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2301 con->in_data_crc = crc;
2307 crc = ceph_crc32c_page(crc, page, page_offset, ret);
2308 ceph_msg_data_advance(cursor, (size_t)ret);
2311 con->in_data_crc = crc;
2313 return 1; /* must return > 0 to indicate success */
2317 * read (part of) a message.
2319 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2321 static int read_partial_message(struct ceph_connection *con)
2323 struct ceph_msg *m = con->in_msg;
2327 unsigned int front_len, middle_len, data_len;
2328 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2329 bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
2333 dout("read_partial_message con %p msg %p\n", con, m);
2336 size = sizeof (con->in_hdr);
2338 ret = read_partial(con, end, size, &con->in_hdr);
2342 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2343 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2344 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
2345 crc, con->in_hdr.crc);
2349 front_len = le32_to_cpu(con->in_hdr.front_len);
2350 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2352 middle_len = le32_to_cpu(con->in_hdr.middle_len);
2353 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2355 data_len = le32_to_cpu(con->in_hdr.data_len);
2356 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2360 seq = le64_to_cpu(con->in_hdr.seq);
2361 if ((s64)seq - (s64)con->in_seq < 1) {
2362 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2363 ENTITY_NAME(con->peer_name),
2364 ceph_pr_addr(&con->peer_addr.in_addr),
2365 seq, con->in_seq + 1);
2366 con->in_base_pos = -front_len - middle_len - data_len -
2368 con->in_tag = CEPH_MSGR_TAG_READY;
2370 } else if ((s64)seq - (s64)con->in_seq > 1) {
2371 pr_err("read_partial_message bad seq %lld expected %lld\n",
2372 seq, con->in_seq + 1);
2373 con->error_msg = "bad message sequence # for incoming message";
2377 /* allocate message? */
2381 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2382 front_len, data_len);
2383 ret = ceph_con_in_msg_alloc(con, &skip);
2387 BUG_ON(!con->in_msg ^ skip);
2389 /* skip this message */
2390 dout("alloc_msg said skip message\n");
2391 con->in_base_pos = -front_len - middle_len - data_len -
2393 con->in_tag = CEPH_MSGR_TAG_READY;
2398 BUG_ON(!con->in_msg);
2399 BUG_ON(con->in_msg->con != con);
2401 m->front.iov_len = 0; /* haven't read it yet */
2403 m->middle->vec.iov_len = 0;
2405 /* prepare for data payload, if any */
2408 prepare_message_data(con->in_msg, data_len);
2412 ret = read_partial_message_section(con, &m->front, front_len,
2413 &con->in_front_crc);
2419 ret = read_partial_message_section(con, &m->middle->vec,
2421 &con->in_middle_crc);
2428 ret = read_partial_msg_data(con);
2434 size = sizeof_footer(con);
2436 ret = read_partial(con, end, size, &m->footer);
2441 m->footer.flags = m->old_footer.flags;
2445 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2446 m, front_len, m->footer.front_crc, middle_len,
2447 m->footer.middle_crc, data_len, m->footer.data_crc);
2450 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2451 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2452 m, con->in_front_crc, m->footer.front_crc);
2455 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2456 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2457 m, con->in_middle_crc, m->footer.middle_crc);
2461 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2462 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2463 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2464 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2468 if (need_sign && con->ops->check_message_signature &&
2469 con->ops->check_message_signature(m)) {
2470 pr_err("read_partial_message %p signature check failed\n", m);
2474 return 1; /* done! */
2478 * Process message. This happens in the worker thread. The callback should
2479 * be careful not to do anything that waits on other incoming messages or it
2482 static void process_message(struct ceph_connection *con)
2484 struct ceph_msg *msg = con->in_msg;
2486 BUG_ON(con->in_msg->con != con);
2489 /* if first message, set peer_name */
2490 if (con->peer_name.type == 0)
2491 con->peer_name = msg->hdr.src;
2494 mutex_unlock(&con->mutex);
2496 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2497 msg, le64_to_cpu(msg->hdr.seq),
2498 ENTITY_NAME(msg->hdr.src),
2499 le16_to_cpu(msg->hdr.type),
2500 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2501 le32_to_cpu(msg->hdr.front_len),
2502 le32_to_cpu(msg->hdr.data_len),
2503 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2504 con->ops->dispatch(con, msg);
2506 mutex_lock(&con->mutex);
2509 static int read_keepalive_ack(struct ceph_connection *con)
2511 struct ceph_timespec ceph_ts;
2512 size_t size = sizeof(ceph_ts);
2513 int ret = read_partial(con, size, size, &ceph_ts);
2516 ceph_decode_timespec(&con->last_keepalive_ack, &ceph_ts);
2517 prepare_read_tag(con);
2522 * Write something to the socket. Called in a worker thread when the
2523 * socket appears to be writeable and we have something ready to send.
2525 static int try_write(struct ceph_connection *con)
2529 dout("try_write start %p state %lu\n", con, con->state);
2532 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2534 /* open the socket first? */
2535 if (con->state == CON_STATE_PREOPEN) {
2537 con->state = CON_STATE_CONNECTING;
2539 con_out_kvec_reset(con);
2540 prepare_write_banner(con);
2541 prepare_read_banner(con);
2543 BUG_ON(con->in_msg);
2544 con->in_tag = CEPH_MSGR_TAG_READY;
2545 dout("try_write initiating connect on %p new state %lu\n",
2547 ret = ceph_tcp_connect(con);
2549 con->error_msg = "connect error";
2555 /* kvec data queued? */
2556 if (con->out_kvec_left) {
2557 ret = write_partial_kvec(con);
2561 if (con->out_skip) {
2562 ret = write_partial_skip(con);
2569 if (con->out_msg_done) {
2570 ceph_msg_put(con->out_msg);
2571 con->out_msg = NULL; /* we're done with this one */
2575 ret = write_partial_message_data(con);
2577 goto more_kvec; /* we need to send the footer, too! */
2581 dout("try_write write_partial_message_data err %d\n",
2588 if (con->state == CON_STATE_OPEN) {
2589 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2590 prepare_write_keepalive(con);
2593 /* is anything else pending? */
2594 if (!list_empty(&con->out_queue)) {
2595 prepare_write_message(con);
2598 if (con->in_seq > con->in_seq_acked) {
2599 prepare_write_ack(con);
2604 /* Nothing to do! */
2605 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2606 dout("try_write nothing else to write.\n");
2609 dout("try_write done on %p ret %d\n", con, ret);
2616 * Read what we can from the socket.
2618 static int try_read(struct ceph_connection *con)
2623 dout("try_read start on %p state %lu\n", con, con->state);
2624 if (con->state != CON_STATE_CONNECTING &&
2625 con->state != CON_STATE_NEGOTIATING &&
2626 con->state != CON_STATE_OPEN)
2631 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2634 if (con->state == CON_STATE_CONNECTING) {
2635 dout("try_read connecting\n");
2636 ret = read_partial_banner(con);
2639 ret = process_banner(con);
2643 con->state = CON_STATE_NEGOTIATING;
2646 * Received banner is good, exchange connection info.
2647 * Do not reset out_kvec, as sending our banner raced
2648 * with receiving peer banner after connect completed.
2650 ret = prepare_write_connect(con);
2653 prepare_read_connect(con);
2655 /* Send connection info before awaiting response */
2659 if (con->state == CON_STATE_NEGOTIATING) {
2660 dout("try_read negotiating\n");
2661 ret = read_partial_connect(con);
2664 ret = process_connect(con);
2670 WARN_ON(con->state != CON_STATE_OPEN);
2672 if (con->in_base_pos < 0) {
2674 * skipping + discarding content.
2676 * FIXME: there must be a better way to do this!
2678 static char buf[SKIP_BUF_SIZE];
2679 int skip = min((int) sizeof (buf), -con->in_base_pos);
2681 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2682 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2685 con->in_base_pos += ret;
2686 if (con->in_base_pos)
2689 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2693 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2696 dout("try_read got tag %d\n", (int)con->in_tag);
2697 switch (con->in_tag) {
2698 case CEPH_MSGR_TAG_MSG:
2699 prepare_read_message(con);
2701 case CEPH_MSGR_TAG_ACK:
2702 prepare_read_ack(con);
2704 case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
2705 prepare_read_keepalive_ack(con);
2707 case CEPH_MSGR_TAG_CLOSE:
2708 con_close_socket(con);
2709 con->state = CON_STATE_CLOSED;
2715 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2716 ret = read_partial_message(con);
2720 con->error_msg = "bad crc/signature";
2726 con->error_msg = "io error";
2731 if (con->in_tag == CEPH_MSGR_TAG_READY)
2733 process_message(con);
2734 if (con->state == CON_STATE_OPEN)
2735 prepare_read_tag(con);
2738 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2739 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2741 * the final handshake seq exchange is semantically
2742 * equivalent to an ACK
2744 ret = read_partial_ack(con);
2750 if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
2751 ret = read_keepalive_ack(con);
2758 dout("try_read done on %p ret %d\n", con, ret);
2762 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2763 con->error_msg = "protocol error, garbage tag";
2770 * Atomically queue work on a connection after the specified delay.
2771 * Bump @con reference to avoid races with connection teardown.
2772 * Returns 0 if work was queued, or an error code otherwise.
2774 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2776 if (!con->ops->get(con)) {
2777 dout("%s %p ref count 0\n", __func__, con);
2781 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2782 dout("%s %p - already queued\n", __func__, con);
2787 dout("%s %p %lu\n", __func__, con, delay);
2791 static void queue_con(struct ceph_connection *con)
2793 (void) queue_con_delay(con, 0);
2796 static void cancel_con(struct ceph_connection *con)
2798 if (cancel_delayed_work(&con->work)) {
2799 dout("%s %p\n", __func__, con);
2804 static bool con_sock_closed(struct ceph_connection *con)
2806 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2810 case CON_STATE_ ## x: \
2811 con->error_msg = "socket closed (con state " #x ")"; \
2814 switch (con->state) {
2822 pr_warn("%s con %p unrecognized state %lu\n",
2823 __func__, con, con->state);
2824 con->error_msg = "unrecognized con state";
2833 static bool con_backoff(struct ceph_connection *con)
2837 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2840 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2842 dout("%s: con %p FAILED to back off %lu\n", __func__,
2844 BUG_ON(ret == -ENOENT);
2845 con_flag_set(con, CON_FLAG_BACKOFF);
2851 /* Finish fault handling; con->mutex must *not* be held here */
2853 static void con_fault_finish(struct ceph_connection *con)
2855 dout("%s %p\n", __func__, con);
2858 * in case we faulted due to authentication, invalidate our
2859 * current tickets so that we can get new ones.
2861 if (con->auth_retry) {
2862 dout("auth_retry %d, invalidating\n", con->auth_retry);
2863 if (con->ops->invalidate_authorizer)
2864 con->ops->invalidate_authorizer(con);
2865 con->auth_retry = 0;
2868 if (con->ops->fault)
2869 con->ops->fault(con);
2873 * Do some work on a connection. Drop a connection ref when we're done.
2875 static void ceph_con_workfn(struct work_struct *work)
2877 struct ceph_connection *con = container_of(work, struct ceph_connection,
2881 mutex_lock(&con->mutex);
2885 if ((fault = con_sock_closed(con))) {
2886 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2889 if (con_backoff(con)) {
2890 dout("%s: con %p BACKOFF\n", __func__, con);
2893 if (con->state == CON_STATE_STANDBY) {
2894 dout("%s: con %p STANDBY\n", __func__, con);
2897 if (con->state == CON_STATE_CLOSED) {
2898 dout("%s: con %p CLOSED\n", __func__, con);
2902 if (con->state == CON_STATE_PREOPEN) {
2903 dout("%s: con %p PREOPEN\n", __func__, con);
2907 ret = try_read(con);
2911 if (!con->error_msg)
2912 con->error_msg = "socket error on read";
2917 ret = try_write(con);
2921 if (!con->error_msg)
2922 con->error_msg = "socket error on write";
2926 break; /* If we make it to here, we're done */
2930 mutex_unlock(&con->mutex);
2933 con_fault_finish(con);
2939 * Generic error/fault handler. A retry mechanism is used with
2940 * exponential backoff
2942 static void con_fault(struct ceph_connection *con)
2944 dout("fault %p state %lu to peer %s\n",
2945 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2947 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2948 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2949 con->error_msg = NULL;
2951 WARN_ON(con->state != CON_STATE_CONNECTING &&
2952 con->state != CON_STATE_NEGOTIATING &&
2953 con->state != CON_STATE_OPEN);
2955 con_close_socket(con);
2957 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2958 dout("fault on LOSSYTX channel, marking CLOSED\n");
2959 con->state = CON_STATE_CLOSED;
2964 BUG_ON(con->in_msg->con != con);
2965 ceph_msg_put(con->in_msg);
2969 /* Requeue anything that hasn't been acked */
2970 list_splice_init(&con->out_sent, &con->out_queue);
2972 /* If there are no messages queued or keepalive pending, place
2973 * the connection in a STANDBY state */
2974 if (list_empty(&con->out_queue) &&
2975 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
2976 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2977 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2978 con->state = CON_STATE_STANDBY;
2980 /* retry after a delay. */
2981 con->state = CON_STATE_PREOPEN;
2982 if (con->delay == 0)
2983 con->delay = BASE_DELAY_INTERVAL;
2984 else if (con->delay < MAX_DELAY_INTERVAL)
2986 con_flag_set(con, CON_FLAG_BACKOFF);
2994 * initialize a new messenger instance
2996 void ceph_messenger_init(struct ceph_messenger *msgr,
2997 struct ceph_entity_addr *myaddr)
2999 spin_lock_init(&msgr->global_seq_lock);
3002 msgr->inst.addr = *myaddr;
3004 /* select a random nonce */
3005 msgr->inst.addr.type = 0;
3006 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
3007 encode_my_addr(msgr);
3009 atomic_set(&msgr->stopping, 0);
3010 write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
3012 dout("%s %p\n", __func__, msgr);
3014 EXPORT_SYMBOL(ceph_messenger_init);
3016 void ceph_messenger_fini(struct ceph_messenger *msgr)
3018 put_net(read_pnet(&msgr->net));
3020 EXPORT_SYMBOL(ceph_messenger_fini);
3022 static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con)
3025 msg->con->ops->put(msg->con);
3027 msg->con = con ? con->ops->get(con) : NULL;
3028 BUG_ON(msg->con != con);
3031 static void clear_standby(struct ceph_connection *con)
3033 /* come back from STANDBY? */
3034 if (con->state == CON_STATE_STANDBY) {
3035 dout("clear_standby %p and ++connect_seq\n", con);
3036 con->state = CON_STATE_PREOPEN;
3038 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
3039 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
3044 * Queue up an outgoing message on the given connection.
3046 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
3049 msg->hdr.src = con->msgr->inst.name;
3050 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
3051 msg->needs_out_seq = true;
3053 mutex_lock(&con->mutex);
3055 if (con->state == CON_STATE_CLOSED) {
3056 dout("con_send %p closed, dropping %p\n", con, msg);
3058 mutex_unlock(&con->mutex);
3062 msg_con_set(msg, con);
3064 BUG_ON(!list_empty(&msg->list_head));
3065 list_add_tail(&msg->list_head, &con->out_queue);
3066 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3067 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3068 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3069 le32_to_cpu(msg->hdr.front_len),
3070 le32_to_cpu(msg->hdr.middle_len),
3071 le32_to_cpu(msg->hdr.data_len));
3074 mutex_unlock(&con->mutex);
3076 /* if there wasn't anything waiting to send before, queue
3078 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3081 EXPORT_SYMBOL(ceph_con_send);
3084 * Revoke a message that was previously queued for send
3086 void ceph_msg_revoke(struct ceph_msg *msg)
3088 struct ceph_connection *con = msg->con;
3091 dout("%s msg %p null con\n", __func__, msg);
3092 return; /* Message not in our possession */
3095 mutex_lock(&con->mutex);
3096 if (!list_empty(&msg->list_head)) {
3097 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
3098 list_del_init(&msg->list_head);
3103 if (con->out_msg == msg) {
3104 BUG_ON(con->out_skip);
3106 if (con->out_msg_done) {
3107 con->out_skip += con_out_kvec_skip(con);
3109 BUG_ON(!msg->data_length);
3110 con->out_skip += sizeof_footer(con);
3112 /* data, middle, front */
3113 if (msg->data_length)
3114 con->out_skip += msg->cursor.total_resid;
3116 con->out_skip += con_out_kvec_skip(con);
3117 con->out_skip += con_out_kvec_skip(con);
3119 dout("%s %p msg %p - was sending, will write %d skip %d\n",
3120 __func__, con, msg, con->out_kvec_bytes, con->out_skip);
3122 con->out_msg = NULL;
3126 mutex_unlock(&con->mutex);
3130 * Revoke a message that we may be reading data into
3132 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
3134 struct ceph_connection *con = msg->con;
3137 dout("%s msg %p null con\n", __func__, msg);
3138 return; /* Message not in our possession */
3141 mutex_lock(&con->mutex);
3142 if (con->in_msg == msg) {
3143 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3144 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3145 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
3147 /* skip rest of message */
3148 dout("%s %p msg %p revoked\n", __func__, con, msg);
3149 con->in_base_pos = con->in_base_pos -
3150 sizeof(struct ceph_msg_header) -
3154 sizeof(struct ceph_msg_footer);
3155 ceph_msg_put(con->in_msg);
3157 con->in_tag = CEPH_MSGR_TAG_READY;
3160 dout("%s %p in_msg %p msg %p no-op\n",
3161 __func__, con, con->in_msg, msg);
3163 mutex_unlock(&con->mutex);
3167 * Queue a keepalive byte to ensure the tcp connection is alive.
3169 void ceph_con_keepalive(struct ceph_connection *con)
3171 dout("con_keepalive %p\n", con);
3172 mutex_lock(&con->mutex);
3174 mutex_unlock(&con->mutex);
3175 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3176 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3179 EXPORT_SYMBOL(ceph_con_keepalive);
3181 bool ceph_con_keepalive_expired(struct ceph_connection *con,
3182 unsigned long interval)
3185 (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
3186 struct timespec now;
3188 ktime_get_real_ts(&now);
3189 jiffies_to_timespec(interval, &ts);
3190 ts = timespec_add(con->last_keepalive_ack, ts);
3191 return timespec_compare(&now, &ts) >= 0;
3196 static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
3198 struct ceph_msg_data *data;
3200 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3203 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
3206 INIT_LIST_HEAD(&data->links);
3211 static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3216 WARN_ON(!list_empty(&data->links));
3217 if (data->type == CEPH_MSG_DATA_PAGELIST)
3218 ceph_pagelist_release(data->pagelist);
3219 kmem_cache_free(ceph_msg_data_cache, data);
3222 void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
3223 size_t length, size_t alignment)
3225 struct ceph_msg_data *data;
3230 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3232 data->pages = pages;
3233 data->length = length;
3234 data->alignment = alignment & ~PAGE_MASK;
3236 list_add_tail(&data->links, &msg->data);
3237 msg->data_length += length;
3239 EXPORT_SYMBOL(ceph_msg_data_add_pages);
3241 void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
3242 struct ceph_pagelist *pagelist)
3244 struct ceph_msg_data *data;
3247 BUG_ON(!pagelist->length);
3249 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3251 data->pagelist = pagelist;
3253 list_add_tail(&data->links, &msg->data);
3254 msg->data_length += pagelist->length;
3256 EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
3259 void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
3262 struct ceph_msg_data *data;
3266 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3269 data->bio_length = length;
3271 list_add_tail(&data->links, &msg->data);
3272 msg->data_length += length;
3274 EXPORT_SYMBOL(ceph_msg_data_add_bio);
3275 #endif /* CONFIG_BLOCK */
3278 * construct a new message with given type, size
3279 * the new msg has a ref count of 1.
3281 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3286 m = kmem_cache_zalloc(ceph_msg_cache, flags);
3290 m->hdr.type = cpu_to_le16(type);
3291 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3292 m->hdr.front_len = cpu_to_le32(front_len);
3294 INIT_LIST_HEAD(&m->list_head);
3295 kref_init(&m->kref);
3296 INIT_LIST_HEAD(&m->data);
3300 m->front.iov_base = ceph_kvmalloc(front_len, flags);
3301 if (m->front.iov_base == NULL) {
3302 dout("ceph_msg_new can't allocate %d bytes\n",
3307 m->front.iov_base = NULL;
3309 m->front_alloc_len = m->front.iov_len = front_len;
3311 dout("ceph_msg_new %p front %d\n", m, front_len);
3318 pr_err("msg_new can't create type %d front %d\n", type,
3322 dout("msg_new can't create type %d front %d\n", type,
3327 EXPORT_SYMBOL(ceph_msg_new);
3330 * Allocate "middle" portion of a message, if it is needed and wasn't
3331 * allocated by alloc_msg. This allows us to read a small fixed-size
3332 * per-type header in the front and then gracefully fail (i.e.,
3333 * propagate the error to the caller based on info in the front) when
3334 * the middle is too large.
3336 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3338 int type = le16_to_cpu(msg->hdr.type);
3339 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3341 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3342 ceph_msg_type_name(type), middle_len);
3343 BUG_ON(!middle_len);
3344 BUG_ON(msg->middle);
3346 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3353 * Allocate a message for receiving an incoming message on a
3354 * connection, and save the result in con->in_msg. Uses the
3355 * connection's private alloc_msg op if available.
3357 * Returns 0 on success, or a negative error code.
3359 * On success, if we set *skip = 1:
3360 * - the next message should be skipped and ignored.
3361 * - con->in_msg == NULL
3362 * or if we set *skip = 0:
3363 * - con->in_msg is non-null.
3364 * On error (ENOMEM, EAGAIN, ...),
3365 * - con->in_msg == NULL
3367 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3369 struct ceph_msg_header *hdr = &con->in_hdr;
3370 int middle_len = le32_to_cpu(hdr->middle_len);
3371 struct ceph_msg *msg;
3374 BUG_ON(con->in_msg != NULL);
3375 BUG_ON(!con->ops->alloc_msg);
3377 mutex_unlock(&con->mutex);
3378 msg = con->ops->alloc_msg(con, hdr, skip);
3379 mutex_lock(&con->mutex);
3380 if (con->state != CON_STATE_OPEN) {
3387 msg_con_set(msg, con);
3391 * Null message pointer means either we should skip
3392 * this message or we couldn't allocate memory. The
3393 * former is not an error.
3398 con->error_msg = "error allocating memory for incoming message";
3401 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3403 if (middle_len && !con->in_msg->middle) {
3404 ret = ceph_alloc_middle(con, con->in_msg);
3406 ceph_msg_put(con->in_msg);
3416 * Free a generically kmalloc'd message.
3418 static void ceph_msg_free(struct ceph_msg *m)
3420 dout("%s %p\n", __func__, m);
3421 kvfree(m->front.iov_base);
3422 kmem_cache_free(ceph_msg_cache, m);
3425 static void ceph_msg_release(struct kref *kref)
3427 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3428 struct ceph_msg_data *data, *next;
3430 dout("%s %p\n", __func__, m);
3431 WARN_ON(!list_empty(&m->list_head));
3433 msg_con_set(m, NULL);
3435 /* drop middle, data, if any */
3437 ceph_buffer_put(m->middle);
3441 list_for_each_entry_safe(data, next, &m->data, links) {
3442 list_del_init(&data->links);
3443 ceph_msg_data_destroy(data);
3448 ceph_msgpool_put(m->pool, m);
3453 struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3455 dout("%s %p (was %d)\n", __func__, msg,
3456 kref_read(&msg->kref));
3457 kref_get(&msg->kref);
3460 EXPORT_SYMBOL(ceph_msg_get);
3462 void ceph_msg_put(struct ceph_msg *msg)
3464 dout("%s %p (was %d)\n", __func__, msg,
3465 kref_read(&msg->kref));
3466 kref_put(&msg->kref, ceph_msg_release);
3468 EXPORT_SYMBOL(ceph_msg_put);
3470 void ceph_msg_dump(struct ceph_msg *msg)
3472 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3473 msg->front_alloc_len, msg->data_length);
3474 print_hex_dump(KERN_DEBUG, "header: ",
3475 DUMP_PREFIX_OFFSET, 16, 1,
3476 &msg->hdr, sizeof(msg->hdr), true);
3477 print_hex_dump(KERN_DEBUG, " front: ",
3478 DUMP_PREFIX_OFFSET, 16, 1,
3479 msg->front.iov_base, msg->front.iov_len, true);
3481 print_hex_dump(KERN_DEBUG, "middle: ",
3482 DUMP_PREFIX_OFFSET, 16, 1,
3483 msg->middle->vec.iov_base,
3484 msg->middle->vec.iov_len, true);
3485 print_hex_dump(KERN_DEBUG, "footer: ",
3486 DUMP_PREFIX_OFFSET, 16, 1,
3487 &msg->footer, sizeof(msg->footer), true);
3489 EXPORT_SYMBOL(ceph_msg_dump);