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/slab.h>
11 #include <linux/socket.h>
12 #include <linux/string.h>
14 #include <linux/bio.h>
15 #endif /* CONFIG_BLOCK */
16 #include <linux/dns_resolver.h>
19 #include <linux/ceph/ceph_features.h>
20 #include <linux/ceph/libceph.h>
21 #include <linux/ceph/messenger.h>
22 #include <linux/ceph/decode.h>
23 #include <linux/ceph/pagelist.h>
24 #include <linux/export.h>
26 #define list_entry_next(pos, member) \
27 list_entry(pos->member.next, typeof(*pos), member)
30 * Ceph uses the messenger to exchange ceph_msg messages with other
31 * hosts in the system. The messenger provides ordered and reliable
32 * delivery. We tolerate TCP disconnects by reconnecting (with
33 * exponential backoff) in the case of a fault (disconnection, bad
34 * crc, protocol error). Acks allow sent messages to be discarded by
39 * We track the state of the socket on a given connection using
40 * values defined below. The transition to a new socket state is
41 * handled by a function which verifies we aren't coming from an
45 * | NEW* | transient initial state
47 * | con_sock_state_init()
50 * | CLOSED | initialized, but no socket (and no
51 * ---------- TCP connection)
53 * | \ con_sock_state_connecting()
54 * | ----------------------
56 * + con_sock_state_closed() \
57 * |+--------------------------- \
60 * | | CLOSING | socket event; \ \
61 * | ----------- await close \ \
64 * | + con_sock_state_closing() \ |
66 * | / --------------- | |
69 * | / -----------------| CONNECTING | socket created, TCP
70 * | | / -------------- connect initiated
71 * | | | con_sock_state_connected()
74 * | CONNECTED | TCP connection established
77 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
80 #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
81 #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
82 #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
83 #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
84 #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
89 #define CON_STATE_CLOSED 1 /* -> PREOPEN */
90 #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
91 #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
92 #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
93 #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
94 #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
97 * ceph_connection flag bits
99 #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
100 * messages on errors */
101 #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
102 #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
103 #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
104 #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
106 static bool con_flag_valid(unsigned long con_flag)
109 case CON_FLAG_LOSSYTX:
110 case CON_FLAG_KEEPALIVE_PENDING:
111 case CON_FLAG_WRITE_PENDING:
112 case CON_FLAG_SOCK_CLOSED:
113 case CON_FLAG_BACKOFF:
120 static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
122 BUG_ON(!con_flag_valid(con_flag));
124 clear_bit(con_flag, &con->flags);
127 static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
129 BUG_ON(!con_flag_valid(con_flag));
131 set_bit(con_flag, &con->flags);
134 static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
136 BUG_ON(!con_flag_valid(con_flag));
138 return test_bit(con_flag, &con->flags);
141 static bool con_flag_test_and_clear(struct ceph_connection *con,
142 unsigned long con_flag)
144 BUG_ON(!con_flag_valid(con_flag));
146 return test_and_clear_bit(con_flag, &con->flags);
149 static bool con_flag_test_and_set(struct ceph_connection *con,
150 unsigned long con_flag)
152 BUG_ON(!con_flag_valid(con_flag));
154 return test_and_set_bit(con_flag, &con->flags);
157 /* Slab caches for frequently-allocated structures */
159 static struct kmem_cache *ceph_msg_cache;
160 static struct kmem_cache *ceph_msg_data_cache;
162 /* static tag bytes (protocol control messages) */
163 static char tag_msg = CEPH_MSGR_TAG_MSG;
164 static char tag_ack = CEPH_MSGR_TAG_ACK;
165 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
166 static char tag_keepalive2 = CEPH_MSGR_TAG_KEEPALIVE2;
168 #ifdef CONFIG_LOCKDEP
169 static struct lock_class_key socket_class;
173 * When skipping (ignoring) a block of input we read it into a "skip
174 * buffer," which is this many bytes in size.
176 #define SKIP_BUF_SIZE 1024
178 static void queue_con(struct ceph_connection *con);
179 static void cancel_con(struct ceph_connection *con);
180 static void ceph_con_workfn(struct work_struct *);
181 static void con_fault(struct ceph_connection *con);
184 * Nicely render a sockaddr as a string. An array of formatted
185 * strings is used, to approximate reentrancy.
187 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
188 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
189 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
190 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
192 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
193 static atomic_t addr_str_seq = ATOMIC_INIT(0);
195 static struct page *zero_page; /* used in certain error cases */
197 const char *ceph_pr_addr(const struct sockaddr_storage *ss)
201 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
202 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
204 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
207 switch (ss->ss_family) {
209 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
210 ntohs(in4->sin_port));
214 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
215 ntohs(in6->sin6_port));
219 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
225 EXPORT_SYMBOL(ceph_pr_addr);
227 static void encode_my_addr(struct ceph_messenger *msgr)
229 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
230 ceph_encode_addr(&msgr->my_enc_addr);
234 * work queue for all reading and writing to/from the socket.
236 static struct workqueue_struct *ceph_msgr_wq;
238 static int ceph_msgr_slab_init(void)
240 BUG_ON(ceph_msg_cache);
241 ceph_msg_cache = kmem_cache_create("ceph_msg",
242 sizeof (struct ceph_msg),
243 __alignof__(struct ceph_msg), 0, NULL);
248 BUG_ON(ceph_msg_data_cache);
249 ceph_msg_data_cache = kmem_cache_create("ceph_msg_data",
250 sizeof (struct ceph_msg_data),
251 __alignof__(struct ceph_msg_data),
253 if (ceph_msg_data_cache)
256 kmem_cache_destroy(ceph_msg_cache);
257 ceph_msg_cache = NULL;
262 static void ceph_msgr_slab_exit(void)
264 BUG_ON(!ceph_msg_data_cache);
265 kmem_cache_destroy(ceph_msg_data_cache);
266 ceph_msg_data_cache = NULL;
268 BUG_ON(!ceph_msg_cache);
269 kmem_cache_destroy(ceph_msg_cache);
270 ceph_msg_cache = NULL;
273 static void _ceph_msgr_exit(void)
276 destroy_workqueue(ceph_msgr_wq);
280 BUG_ON(zero_page == NULL);
281 page_cache_release(zero_page);
284 ceph_msgr_slab_exit();
287 int ceph_msgr_init(void)
289 if (ceph_msgr_slab_init())
292 BUG_ON(zero_page != NULL);
293 zero_page = ZERO_PAGE(0);
294 page_cache_get(zero_page);
297 * The number of active work items is limited by the number of
298 * connections, so leave @max_active at default.
300 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
304 pr_err("msgr_init failed to create workqueue\n");
309 EXPORT_SYMBOL(ceph_msgr_init);
311 void ceph_msgr_exit(void)
313 BUG_ON(ceph_msgr_wq == NULL);
317 EXPORT_SYMBOL(ceph_msgr_exit);
319 void ceph_msgr_flush(void)
321 flush_workqueue(ceph_msgr_wq);
323 EXPORT_SYMBOL(ceph_msgr_flush);
325 /* Connection socket state transition functions */
327 static void con_sock_state_init(struct ceph_connection *con)
331 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
332 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
333 printk("%s: unexpected old state %d\n", __func__, old_state);
334 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
335 CON_SOCK_STATE_CLOSED);
338 static void con_sock_state_connecting(struct ceph_connection *con)
342 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
343 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
344 printk("%s: unexpected old state %d\n", __func__, old_state);
345 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
346 CON_SOCK_STATE_CONNECTING);
349 static void con_sock_state_connected(struct ceph_connection *con)
353 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
354 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
355 printk("%s: unexpected old state %d\n", __func__, old_state);
356 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
357 CON_SOCK_STATE_CONNECTED);
360 static void con_sock_state_closing(struct ceph_connection *con)
364 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
365 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
366 old_state != CON_SOCK_STATE_CONNECTED &&
367 old_state != CON_SOCK_STATE_CLOSING))
368 printk("%s: unexpected old state %d\n", __func__, old_state);
369 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
370 CON_SOCK_STATE_CLOSING);
373 static void con_sock_state_closed(struct ceph_connection *con)
377 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
378 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
379 old_state != CON_SOCK_STATE_CLOSING &&
380 old_state != CON_SOCK_STATE_CONNECTING &&
381 old_state != CON_SOCK_STATE_CLOSED))
382 printk("%s: unexpected old state %d\n", __func__, old_state);
383 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
384 CON_SOCK_STATE_CLOSED);
388 * socket callback functions
391 /* data available on socket, or listen socket received a connect */
392 static void ceph_sock_data_ready(struct sock *sk)
394 struct ceph_connection *con = sk->sk_user_data;
395 if (atomic_read(&con->msgr->stopping)) {
399 if (sk->sk_state != TCP_CLOSE_WAIT) {
400 dout("%s on %p state = %lu, queueing work\n", __func__,
406 /* socket has buffer space for writing */
407 static void ceph_sock_write_space(struct sock *sk)
409 struct ceph_connection *con = sk->sk_user_data;
411 /* only queue to workqueue if there is data we want to write,
412 * and there is sufficient space in the socket buffer to accept
413 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
414 * doesn't get called again until try_write() fills the socket
415 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
416 * and net/core/stream.c:sk_stream_write_space().
418 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
419 if (sk_stream_is_writeable(sk)) {
420 dout("%s %p queueing write work\n", __func__, con);
421 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
425 dout("%s %p nothing to write\n", __func__, con);
429 /* socket's state has changed */
430 static void ceph_sock_state_change(struct sock *sk)
432 struct ceph_connection *con = sk->sk_user_data;
434 dout("%s %p state = %lu sk_state = %u\n", __func__,
435 con, con->state, sk->sk_state);
437 switch (sk->sk_state) {
439 dout("%s TCP_CLOSE\n", __func__);
441 dout("%s TCP_CLOSE_WAIT\n", __func__);
442 con_sock_state_closing(con);
443 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
446 case TCP_ESTABLISHED:
447 dout("%s TCP_ESTABLISHED\n", __func__);
448 con_sock_state_connected(con);
451 default: /* Everything else is uninteresting */
457 * set up socket callbacks
459 static void set_sock_callbacks(struct socket *sock,
460 struct ceph_connection *con)
462 struct sock *sk = sock->sk;
463 sk->sk_user_data = con;
464 sk->sk_data_ready = ceph_sock_data_ready;
465 sk->sk_write_space = ceph_sock_write_space;
466 sk->sk_state_change = ceph_sock_state_change;
475 * initiate connection to a remote socket.
477 static int ceph_tcp_connect(struct ceph_connection *con)
479 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
484 ret = sock_create_kern(read_pnet(&con->msgr->net), paddr->ss_family,
485 SOCK_STREAM, IPPROTO_TCP, &sock);
488 sock->sk->sk_allocation = GFP_NOFS;
490 #ifdef CONFIG_LOCKDEP
491 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
494 set_sock_callbacks(sock, con);
496 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
498 con_sock_state_connecting(con);
499 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
501 if (ret == -EINPROGRESS) {
502 dout("connect %s EINPROGRESS sk_state = %u\n",
503 ceph_pr_addr(&con->peer_addr.in_addr),
505 } else if (ret < 0) {
506 pr_err("connect %s error %d\n",
507 ceph_pr_addr(&con->peer_addr.in_addr), ret);
512 if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY)) {
515 ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
516 (char *)&optval, sizeof(optval));
518 pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
526 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
528 struct kvec iov = {buf, len};
529 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
532 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
538 static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
539 int page_offset, size_t length)
544 BUG_ON(page_offset + length > PAGE_SIZE);
548 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
555 * write something. @more is true if caller will be sending more data
558 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
559 size_t kvlen, size_t len, int more)
561 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
565 msg.msg_flags |= MSG_MORE;
567 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
569 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
575 static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
576 int offset, size_t size, bool more)
578 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
581 ret = kernel_sendpage(sock, page, offset, size, flags);
588 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
589 int offset, size_t size, bool more)
594 /* sendpage cannot properly handle pages with page_count == 0,
595 * we need to fallback to sendmsg if that's the case */
596 if (page_count(page) >= 1)
597 return __ceph_tcp_sendpage(sock, page, offset, size, more);
599 iov.iov_base = kmap(page) + offset;
601 ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more);
608 * Shutdown/close the socket for the given connection.
610 static int con_close_socket(struct ceph_connection *con)
614 dout("con_close_socket on %p sock %p\n", con, con->sock);
616 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
617 sock_release(con->sock);
622 * Forcibly clear the SOCK_CLOSED flag. It gets set
623 * independent of the connection mutex, and we could have
624 * received a socket close event before we had the chance to
625 * shut the socket down.
627 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
629 con_sock_state_closed(con);
634 * Reset a connection. Discard all incoming and outgoing messages
635 * and clear *_seq state.
637 static void ceph_msg_remove(struct ceph_msg *msg)
639 list_del_init(&msg->list_head);
643 static void ceph_msg_remove_list(struct list_head *head)
645 while (!list_empty(head)) {
646 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
648 ceph_msg_remove(msg);
652 static void reset_connection(struct ceph_connection *con)
654 /* reset connection, out_queue, msg_ and connect_seq */
655 /* discard existing out_queue and msg_seq */
656 dout("reset_connection %p\n", con);
657 ceph_msg_remove_list(&con->out_queue);
658 ceph_msg_remove_list(&con->out_sent);
661 BUG_ON(con->in_msg->con != con);
662 ceph_msg_put(con->in_msg);
666 con->connect_seq = 0;
669 BUG_ON(con->out_msg->con != con);
670 ceph_msg_put(con->out_msg);
674 con->in_seq_acked = 0;
680 * mark a peer down. drop any open connections.
682 void ceph_con_close(struct ceph_connection *con)
684 mutex_lock(&con->mutex);
685 dout("con_close %p peer %s\n", con,
686 ceph_pr_addr(&con->peer_addr.in_addr));
687 con->state = CON_STATE_CLOSED;
689 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
690 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
691 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
692 con_flag_clear(con, CON_FLAG_BACKOFF);
694 reset_connection(con);
695 con->peer_global_seq = 0;
697 con_close_socket(con);
698 mutex_unlock(&con->mutex);
700 EXPORT_SYMBOL(ceph_con_close);
703 * Reopen a closed connection, with a new peer address.
705 void ceph_con_open(struct ceph_connection *con,
706 __u8 entity_type, __u64 entity_num,
707 struct ceph_entity_addr *addr)
709 mutex_lock(&con->mutex);
710 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
712 WARN_ON(con->state != CON_STATE_CLOSED);
713 con->state = CON_STATE_PREOPEN;
715 con->peer_name.type = (__u8) entity_type;
716 con->peer_name.num = cpu_to_le64(entity_num);
718 memcpy(&con->peer_addr, addr, sizeof(*addr));
719 con->delay = 0; /* reset backoff memory */
720 mutex_unlock(&con->mutex);
723 EXPORT_SYMBOL(ceph_con_open);
726 * return true if this connection ever successfully opened
728 bool ceph_con_opened(struct ceph_connection *con)
730 return con->connect_seq > 0;
734 * initialize a new connection.
736 void ceph_con_init(struct ceph_connection *con, void *private,
737 const struct ceph_connection_operations *ops,
738 struct ceph_messenger *msgr)
740 dout("con_init %p\n", con);
741 memset(con, 0, sizeof(*con));
742 con->private = private;
746 con_sock_state_init(con);
748 mutex_init(&con->mutex);
749 INIT_LIST_HEAD(&con->out_queue);
750 INIT_LIST_HEAD(&con->out_sent);
751 INIT_DELAYED_WORK(&con->work, ceph_con_workfn);
753 con->state = CON_STATE_CLOSED;
755 EXPORT_SYMBOL(ceph_con_init);
759 * We maintain a global counter to order connection attempts. Get
760 * a unique seq greater than @gt.
762 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
766 spin_lock(&msgr->global_seq_lock);
767 if (msgr->global_seq < gt)
768 msgr->global_seq = gt;
769 ret = ++msgr->global_seq;
770 spin_unlock(&msgr->global_seq_lock);
774 static void con_out_kvec_reset(struct ceph_connection *con)
776 BUG_ON(con->out_skip);
778 con->out_kvec_left = 0;
779 con->out_kvec_bytes = 0;
780 con->out_kvec_cur = &con->out_kvec[0];
783 static void con_out_kvec_add(struct ceph_connection *con,
784 size_t size, void *data)
786 int index = con->out_kvec_left;
788 BUG_ON(con->out_skip);
789 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
791 con->out_kvec[index].iov_len = size;
792 con->out_kvec[index].iov_base = data;
793 con->out_kvec_left++;
794 con->out_kvec_bytes += size;
798 * Chop off a kvec from the end. Return residual number of bytes for
799 * that kvec, i.e. how many bytes would have been written if the kvec
802 static int con_out_kvec_skip(struct ceph_connection *con)
804 int off = con->out_kvec_cur - con->out_kvec;
807 if (con->out_kvec_bytes > 0) {
808 skip = con->out_kvec[off + con->out_kvec_left - 1].iov_len;
809 BUG_ON(con->out_kvec_bytes < skip);
810 BUG_ON(!con->out_kvec_left);
811 con->out_kvec_bytes -= skip;
812 con->out_kvec_left--;
821 * For a bio data item, a piece is whatever remains of the next
822 * entry in the current bio iovec, or the first entry in the next
825 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
828 struct ceph_msg_data *data = cursor->data;
831 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
836 cursor->resid = min(length, data->bio_length);
838 cursor->bvec_iter = bio->bi_iter;
840 cursor->resid <= bio_iter_len(bio, cursor->bvec_iter);
843 static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
847 struct ceph_msg_data *data = cursor->data;
849 struct bio_vec bio_vec;
851 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
856 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
858 *page_offset = (size_t) bio_vec.bv_offset;
859 BUG_ON(*page_offset >= PAGE_SIZE);
860 if (cursor->last_piece) /* pagelist offset is always 0 */
861 *length = cursor->resid;
863 *length = (size_t) bio_vec.bv_len;
864 BUG_ON(*length > cursor->resid);
865 BUG_ON(*page_offset + *length > PAGE_SIZE);
867 return bio_vec.bv_page;
870 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
874 struct bio_vec bio_vec;
876 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
881 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
883 /* Advance the cursor offset */
885 BUG_ON(cursor->resid < bytes);
886 cursor->resid -= bytes;
888 bio_advance_iter(bio, &cursor->bvec_iter, bytes);
890 if (bytes < bio_vec.bv_len)
891 return false; /* more bytes to process in this segment */
893 /* Move on to the next segment, and possibly the next bio */
895 if (!cursor->bvec_iter.bi_size) {
899 cursor->bvec_iter = bio->bi_iter;
901 memset(&cursor->bvec_iter, 0,
902 sizeof(cursor->bvec_iter));
905 if (!cursor->last_piece) {
906 BUG_ON(!cursor->resid);
908 /* A short read is OK, so use <= rather than == */
909 if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter))
910 cursor->last_piece = true;
915 #endif /* CONFIG_BLOCK */
918 * For a page array, a piece comes from the first page in the array
919 * that has not already been fully consumed.
921 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
924 struct ceph_msg_data *data = cursor->data;
927 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
929 BUG_ON(!data->pages);
930 BUG_ON(!data->length);
932 cursor->resid = min(length, data->length);
933 page_count = calc_pages_for(data->alignment, (u64)data->length);
934 cursor->page_offset = data->alignment & ~PAGE_MASK;
935 cursor->page_index = 0;
936 BUG_ON(page_count > (int)USHRT_MAX);
937 cursor->page_count = (unsigned short)page_count;
938 BUG_ON(length > SIZE_MAX - cursor->page_offset);
939 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
943 ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
944 size_t *page_offset, size_t *length)
946 struct ceph_msg_data *data = cursor->data;
948 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
950 BUG_ON(cursor->page_index >= cursor->page_count);
951 BUG_ON(cursor->page_offset >= PAGE_SIZE);
953 *page_offset = cursor->page_offset;
954 if (cursor->last_piece)
955 *length = cursor->resid;
957 *length = PAGE_SIZE - *page_offset;
959 return data->pages[cursor->page_index];
962 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
965 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
967 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
969 /* Advance the cursor page offset */
971 cursor->resid -= bytes;
972 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
973 if (!bytes || cursor->page_offset)
974 return false; /* more bytes to process in the current page */
977 return false; /* no more data */
979 /* Move on to the next page; offset is already at 0 */
981 BUG_ON(cursor->page_index >= cursor->page_count);
982 cursor->page_index++;
983 cursor->last_piece = cursor->resid <= PAGE_SIZE;
989 * For a pagelist, a piece is whatever remains to be consumed in the
990 * first page in the list, or the front of the next page.
993 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
996 struct ceph_msg_data *data = cursor->data;
997 struct ceph_pagelist *pagelist;
1000 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1002 pagelist = data->pagelist;
1006 return; /* pagelist can be assigned but empty */
1008 BUG_ON(list_empty(&pagelist->head));
1009 page = list_first_entry(&pagelist->head, struct page, lru);
1011 cursor->resid = min(length, pagelist->length);
1012 cursor->page = page;
1014 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1017 static struct page *
1018 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
1019 size_t *page_offset, size_t *length)
1021 struct ceph_msg_data *data = cursor->data;
1022 struct ceph_pagelist *pagelist;
1024 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1026 pagelist = data->pagelist;
1029 BUG_ON(!cursor->page);
1030 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1032 /* offset of first page in pagelist is always 0 */
1033 *page_offset = cursor->offset & ~PAGE_MASK;
1034 if (cursor->last_piece)
1035 *length = cursor->resid;
1037 *length = PAGE_SIZE - *page_offset;
1039 return cursor->page;
1042 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
1045 struct ceph_msg_data *data = cursor->data;
1046 struct ceph_pagelist *pagelist;
1048 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1050 pagelist = data->pagelist;
1053 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1054 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1056 /* Advance the cursor offset */
1058 cursor->resid -= bytes;
1059 cursor->offset += bytes;
1060 /* offset of first page in pagelist is always 0 */
1061 if (!bytes || cursor->offset & ~PAGE_MASK)
1062 return false; /* more bytes to process in the current page */
1065 return false; /* no more data */
1067 /* Move on to the next page */
1069 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1070 cursor->page = list_entry_next(cursor->page, lru);
1071 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1077 * Message data is handled (sent or received) in pieces, where each
1078 * piece resides on a single page. The network layer might not
1079 * consume an entire piece at once. A data item's cursor keeps
1080 * track of which piece is next to process and how much remains to
1081 * be processed in that piece. It also tracks whether the current
1082 * piece is the last one in the data item.
1084 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1086 size_t length = cursor->total_resid;
1088 switch (cursor->data->type) {
1089 case CEPH_MSG_DATA_PAGELIST:
1090 ceph_msg_data_pagelist_cursor_init(cursor, length);
1092 case CEPH_MSG_DATA_PAGES:
1093 ceph_msg_data_pages_cursor_init(cursor, length);
1096 case CEPH_MSG_DATA_BIO:
1097 ceph_msg_data_bio_cursor_init(cursor, length);
1099 #endif /* CONFIG_BLOCK */
1100 case CEPH_MSG_DATA_NONE:
1105 cursor->need_crc = true;
1108 static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1110 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1111 struct ceph_msg_data *data;
1114 BUG_ON(length > msg->data_length);
1115 BUG_ON(list_empty(&msg->data));
1117 cursor->data_head = &msg->data;
1118 cursor->total_resid = length;
1119 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1120 cursor->data = data;
1122 __ceph_msg_data_cursor_init(cursor);
1126 * Return the page containing the next piece to process for a given
1127 * data item, and supply the page offset and length of that piece.
1128 * Indicate whether this is the last piece in this data item.
1130 static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1131 size_t *page_offset, size_t *length,
1136 switch (cursor->data->type) {
1137 case CEPH_MSG_DATA_PAGELIST:
1138 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1140 case CEPH_MSG_DATA_PAGES:
1141 page = ceph_msg_data_pages_next(cursor, page_offset, length);
1144 case CEPH_MSG_DATA_BIO:
1145 page = ceph_msg_data_bio_next(cursor, page_offset, length);
1147 #endif /* CONFIG_BLOCK */
1148 case CEPH_MSG_DATA_NONE:
1154 BUG_ON(*page_offset + *length > PAGE_SIZE);
1157 *last_piece = cursor->last_piece;
1163 * Returns true if the result moves the cursor on to the next piece
1166 static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1171 BUG_ON(bytes > cursor->resid);
1172 switch (cursor->data->type) {
1173 case CEPH_MSG_DATA_PAGELIST:
1174 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1176 case CEPH_MSG_DATA_PAGES:
1177 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1180 case CEPH_MSG_DATA_BIO:
1181 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1183 #endif /* CONFIG_BLOCK */
1184 case CEPH_MSG_DATA_NONE:
1189 cursor->total_resid -= bytes;
1191 if (!cursor->resid && cursor->total_resid) {
1192 WARN_ON(!cursor->last_piece);
1193 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1194 cursor->data = list_entry_next(cursor->data, links);
1195 __ceph_msg_data_cursor_init(cursor);
1198 cursor->need_crc = new_piece;
1203 static size_t sizeof_footer(struct ceph_connection *con)
1205 return (con->peer_features & CEPH_FEATURE_MSG_AUTH) ?
1206 sizeof(struct ceph_msg_footer) :
1207 sizeof(struct ceph_msg_footer_old);
1210 static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
1215 /* Initialize data cursor */
1217 ceph_msg_data_cursor_init(msg, (size_t)data_len);
1221 * Prepare footer for currently outgoing message, and finish things
1222 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1224 static void prepare_write_message_footer(struct ceph_connection *con)
1226 struct ceph_msg *m = con->out_msg;
1227 int v = con->out_kvec_left;
1229 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1231 dout("prepare_write_message_footer %p\n", con);
1232 con->out_kvec[v].iov_base = &m->footer;
1233 if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
1234 if (con->ops->sign_message)
1235 con->ops->sign_message(m);
1238 con->out_kvec[v].iov_len = sizeof(m->footer);
1239 con->out_kvec_bytes += sizeof(m->footer);
1241 m->old_footer.flags = m->footer.flags;
1242 con->out_kvec[v].iov_len = sizeof(m->old_footer);
1243 con->out_kvec_bytes += sizeof(m->old_footer);
1245 con->out_kvec_left++;
1246 con->out_more = m->more_to_follow;
1247 con->out_msg_done = true;
1251 * Prepare headers for the next outgoing message.
1253 static void prepare_write_message(struct ceph_connection *con)
1258 con_out_kvec_reset(con);
1259 con->out_msg_done = false;
1261 /* Sneak an ack in there first? If we can get it into the same
1262 * TCP packet that's a good thing. */
1263 if (con->in_seq > con->in_seq_acked) {
1264 con->in_seq_acked = con->in_seq;
1265 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1266 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1267 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1268 &con->out_temp_ack);
1271 BUG_ON(list_empty(&con->out_queue));
1272 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1274 BUG_ON(m->con != con);
1276 /* put message on sent list */
1278 list_move_tail(&m->list_head, &con->out_sent);
1281 * only assign outgoing seq # if we haven't sent this message
1282 * yet. if it is requeued, resend with it's original seq.
1284 if (m->needs_out_seq) {
1285 m->hdr.seq = cpu_to_le64(++con->out_seq);
1286 m->needs_out_seq = false;
1288 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
1290 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1291 m, con->out_seq, le16_to_cpu(m->hdr.type),
1292 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1294 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1296 /* tag + hdr + front + middle */
1297 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1298 con_out_kvec_add(con, sizeof(con->out_hdr), &con->out_hdr);
1299 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1302 con_out_kvec_add(con, m->middle->vec.iov_len,
1303 m->middle->vec.iov_base);
1305 /* fill in hdr crc and finalize hdr */
1306 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1307 con->out_msg->hdr.crc = cpu_to_le32(crc);
1308 memcpy(&con->out_hdr, &con->out_msg->hdr, sizeof(con->out_hdr));
1310 /* fill in front and middle crc, footer */
1311 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1312 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1314 crc = crc32c(0, m->middle->vec.iov_base,
1315 m->middle->vec.iov_len);
1316 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1318 con->out_msg->footer.middle_crc = 0;
1319 dout("%s front_crc %u middle_crc %u\n", __func__,
1320 le32_to_cpu(con->out_msg->footer.front_crc),
1321 le32_to_cpu(con->out_msg->footer.middle_crc));
1322 con->out_msg->footer.flags = 0;
1324 /* is there a data payload? */
1325 con->out_msg->footer.data_crc = 0;
1326 if (m->data_length) {
1327 prepare_message_data(con->out_msg, m->data_length);
1328 con->out_more = 1; /* data + footer will follow */
1330 /* no, queue up footer too and be done */
1331 prepare_write_message_footer(con);
1334 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1340 static void prepare_write_ack(struct ceph_connection *con)
1342 dout("prepare_write_ack %p %llu -> %llu\n", con,
1343 con->in_seq_acked, con->in_seq);
1344 con->in_seq_acked = con->in_seq;
1346 con_out_kvec_reset(con);
1348 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1350 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1351 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1352 &con->out_temp_ack);
1354 con->out_more = 1; /* more will follow.. eventually.. */
1355 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1359 * Prepare to share the seq during handshake
1361 static void prepare_write_seq(struct ceph_connection *con)
1363 dout("prepare_write_seq %p %llu -> %llu\n", con,
1364 con->in_seq_acked, con->in_seq);
1365 con->in_seq_acked = con->in_seq;
1367 con_out_kvec_reset(con);
1369 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1370 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1371 &con->out_temp_ack);
1373 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1377 * Prepare to write keepalive byte.
1379 static void prepare_write_keepalive(struct ceph_connection *con)
1381 dout("prepare_write_keepalive %p\n", con);
1382 con_out_kvec_reset(con);
1383 if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
1384 struct timespec now = CURRENT_TIME;
1386 con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
1387 ceph_encode_timespec(&con->out_temp_keepalive2, &now);
1388 con_out_kvec_add(con, sizeof(con->out_temp_keepalive2),
1389 &con->out_temp_keepalive2);
1391 con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive);
1393 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1397 * Connection negotiation.
1400 static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1403 struct ceph_auth_handshake *auth;
1405 if (!con->ops->get_authorizer) {
1406 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1407 con->out_connect.authorizer_len = 0;
1411 /* Can't hold the mutex while getting authorizer */
1412 mutex_unlock(&con->mutex);
1413 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
1414 mutex_lock(&con->mutex);
1418 if (con->state != CON_STATE_NEGOTIATING)
1419 return ERR_PTR(-EAGAIN);
1421 con->auth_reply_buf = auth->authorizer_reply_buf;
1422 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
1427 * We connected to a peer and are saying hello.
1429 static void prepare_write_banner(struct ceph_connection *con)
1431 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1432 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1433 &con->msgr->my_enc_addr);
1436 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1439 static int prepare_write_connect(struct ceph_connection *con)
1441 unsigned int global_seq = get_global_seq(con->msgr, 0);
1444 struct ceph_auth_handshake *auth;
1446 switch (con->peer_name.type) {
1447 case CEPH_ENTITY_TYPE_MON:
1448 proto = CEPH_MONC_PROTOCOL;
1450 case CEPH_ENTITY_TYPE_OSD:
1451 proto = CEPH_OSDC_PROTOCOL;
1453 case CEPH_ENTITY_TYPE_MDS:
1454 proto = CEPH_MDSC_PROTOCOL;
1460 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1461 con->connect_seq, global_seq, proto);
1463 con->out_connect.features =
1464 cpu_to_le64(from_msgr(con->msgr)->supported_features);
1465 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1466 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1467 con->out_connect.global_seq = cpu_to_le32(global_seq);
1468 con->out_connect.protocol_version = cpu_to_le32(proto);
1469 con->out_connect.flags = 0;
1471 auth_proto = CEPH_AUTH_UNKNOWN;
1472 auth = get_connect_authorizer(con, &auth_proto);
1474 return PTR_ERR(auth);
1476 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1477 con->out_connect.authorizer_len = auth ?
1478 cpu_to_le32(auth->authorizer_buf_len) : 0;
1480 con_out_kvec_add(con, sizeof (con->out_connect),
1482 if (auth && auth->authorizer_buf_len)
1483 con_out_kvec_add(con, auth->authorizer_buf_len,
1484 auth->authorizer_buf);
1487 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1493 * write as much of pending kvecs to the socket as we can.
1495 * 0 -> socket full, but more to do
1498 static int write_partial_kvec(struct ceph_connection *con)
1502 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1503 while (con->out_kvec_bytes > 0) {
1504 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1505 con->out_kvec_left, con->out_kvec_bytes,
1509 con->out_kvec_bytes -= ret;
1510 if (con->out_kvec_bytes == 0)
1513 /* account for full iov entries consumed */
1514 while (ret >= con->out_kvec_cur->iov_len) {
1515 BUG_ON(!con->out_kvec_left);
1516 ret -= con->out_kvec_cur->iov_len;
1517 con->out_kvec_cur++;
1518 con->out_kvec_left--;
1520 /* and for a partially-consumed entry */
1522 con->out_kvec_cur->iov_len -= ret;
1523 con->out_kvec_cur->iov_base += ret;
1526 con->out_kvec_left = 0;
1529 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1530 con->out_kvec_bytes, con->out_kvec_left, ret);
1531 return ret; /* done! */
1534 static u32 ceph_crc32c_page(u32 crc, struct page *page,
1535 unsigned int page_offset,
1536 unsigned int length)
1541 BUG_ON(kaddr == NULL);
1542 crc = crc32c(crc, kaddr + page_offset, length);
1548 * Write as much message data payload as we can. If we finish, queue
1550 * 1 -> done, footer is now queued in out_kvec[].
1551 * 0 -> socket full, but more to do
1554 static int write_partial_message_data(struct ceph_connection *con)
1556 struct ceph_msg *msg = con->out_msg;
1557 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1558 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
1561 dout("%s %p msg %p\n", __func__, con, msg);
1563 if (list_empty(&msg->data))
1567 * Iterate through each page that contains data to be
1568 * written, and send as much as possible for each.
1570 * If we are calculating the data crc (the default), we will
1571 * need to map the page. If we have no pages, they have
1572 * been revoked, so use the zero page.
1574 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1575 while (cursor->resid) {
1583 page = ceph_msg_data_next(cursor, &page_offset, &length,
1585 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
1586 length, !last_piece);
1589 msg->footer.data_crc = cpu_to_le32(crc);
1593 if (do_datacrc && cursor->need_crc)
1594 crc = ceph_crc32c_page(crc, page, page_offset, length);
1595 need_crc = ceph_msg_data_advance(cursor, (size_t)ret);
1598 dout("%s %p msg %p done\n", __func__, con, msg);
1600 /* prepare and queue up footer, too */
1602 msg->footer.data_crc = cpu_to_le32(crc);
1604 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1605 con_out_kvec_reset(con);
1606 prepare_write_message_footer(con);
1608 return 1; /* must return > 0 to indicate success */
1614 static int write_partial_skip(struct ceph_connection *con)
1618 dout("%s %p %d left\n", __func__, con, con->out_skip);
1619 while (con->out_skip > 0) {
1620 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
1622 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
1625 con->out_skip -= ret;
1633 * Prepare to read connection handshake, or an ack.
1635 static void prepare_read_banner(struct ceph_connection *con)
1637 dout("prepare_read_banner %p\n", con);
1638 con->in_base_pos = 0;
1641 static void prepare_read_connect(struct ceph_connection *con)
1643 dout("prepare_read_connect %p\n", con);
1644 con->in_base_pos = 0;
1647 static void prepare_read_ack(struct ceph_connection *con)
1649 dout("prepare_read_ack %p\n", con);
1650 con->in_base_pos = 0;
1653 static void prepare_read_seq(struct ceph_connection *con)
1655 dout("prepare_read_seq %p\n", con);
1656 con->in_base_pos = 0;
1657 con->in_tag = CEPH_MSGR_TAG_SEQ;
1660 static void prepare_read_tag(struct ceph_connection *con)
1662 dout("prepare_read_tag %p\n", con);
1663 con->in_base_pos = 0;
1664 con->in_tag = CEPH_MSGR_TAG_READY;
1667 static void prepare_read_keepalive_ack(struct ceph_connection *con)
1669 dout("prepare_read_keepalive_ack %p\n", con);
1670 con->in_base_pos = 0;
1674 * Prepare to read a message.
1676 static int prepare_read_message(struct ceph_connection *con)
1678 dout("prepare_read_message %p\n", con);
1679 BUG_ON(con->in_msg != NULL);
1680 con->in_base_pos = 0;
1681 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1686 static int read_partial(struct ceph_connection *con,
1687 int end, int size, void *object)
1689 while (con->in_base_pos < end) {
1690 int left = end - con->in_base_pos;
1691 int have = size - left;
1692 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1695 con->in_base_pos += ret;
1702 * Read all or part of the connect-side handshake on a new connection
1704 static int read_partial_banner(struct ceph_connection *con)
1710 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1713 size = strlen(CEPH_BANNER);
1715 ret = read_partial(con, end, size, con->in_banner);
1719 size = sizeof (con->actual_peer_addr);
1721 ret = read_partial(con, end, size, &con->actual_peer_addr);
1725 size = sizeof (con->peer_addr_for_me);
1727 ret = read_partial(con, end, size, &con->peer_addr_for_me);
1735 static int read_partial_connect(struct ceph_connection *con)
1741 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1743 size = sizeof (con->in_reply);
1745 ret = read_partial(con, end, size, &con->in_reply);
1749 size = le32_to_cpu(con->in_reply.authorizer_len);
1751 ret = read_partial(con, end, size, con->auth_reply_buf);
1755 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1756 con, (int)con->in_reply.tag,
1757 le32_to_cpu(con->in_reply.connect_seq),
1758 le32_to_cpu(con->in_reply.global_seq));
1765 * Verify the hello banner looks okay.
1767 static int verify_hello(struct ceph_connection *con)
1769 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1770 pr_err("connect to %s got bad banner\n",
1771 ceph_pr_addr(&con->peer_addr.in_addr));
1772 con->error_msg = "protocol error, bad banner";
1778 static bool addr_is_blank(struct sockaddr_storage *ss)
1780 struct in_addr *addr = &((struct sockaddr_in *)ss)->sin_addr;
1781 struct in6_addr *addr6 = &((struct sockaddr_in6 *)ss)->sin6_addr;
1783 switch (ss->ss_family) {
1785 return addr->s_addr == htonl(INADDR_ANY);
1787 return ipv6_addr_any(addr6);
1793 static int addr_port(struct sockaddr_storage *ss)
1795 switch (ss->ss_family) {
1797 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1799 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1804 static void addr_set_port(struct sockaddr_storage *ss, int p)
1806 switch (ss->ss_family) {
1808 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1811 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1817 * Unlike other *_pton function semantics, zero indicates success.
1819 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1820 char delim, const char **ipend)
1822 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1823 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1825 memset(ss, 0, sizeof(*ss));
1827 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1828 ss->ss_family = AF_INET;
1832 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1833 ss->ss_family = AF_INET6;
1841 * Extract hostname string and resolve using kernel DNS facility.
1843 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1844 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1845 struct sockaddr_storage *ss, char delim, const char **ipend)
1847 const char *end, *delim_p;
1848 char *colon_p, *ip_addr = NULL;
1852 * The end of the hostname occurs immediately preceding the delimiter or
1853 * the port marker (':') where the delimiter takes precedence.
1855 delim_p = memchr(name, delim, namelen);
1856 colon_p = memchr(name, ':', namelen);
1858 if (delim_p && colon_p)
1859 end = delim_p < colon_p ? delim_p : colon_p;
1860 else if (!delim_p && colon_p)
1864 if (!end) /* case: hostname:/ */
1865 end = name + namelen;
1871 /* do dns_resolve upcall */
1872 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1874 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1882 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1883 ret, ret ? "failed" : ceph_pr_addr(ss));
1888 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1889 struct sockaddr_storage *ss, char delim, const char **ipend)
1896 * Parse a server name (IP or hostname). If a valid IP address is not found
1897 * then try to extract a hostname to resolve using userspace DNS upcall.
1899 static int ceph_parse_server_name(const char *name, size_t namelen,
1900 struct sockaddr_storage *ss, char delim, const char **ipend)
1904 ret = ceph_pton(name, namelen, ss, delim, ipend);
1906 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1912 * Parse an ip[:port] list into an addr array. Use the default
1913 * monitor port if a port isn't specified.
1915 int ceph_parse_ips(const char *c, const char *end,
1916 struct ceph_entity_addr *addr,
1917 int max_count, int *count)
1919 int i, ret = -EINVAL;
1922 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1923 for (i = 0; i < max_count; i++) {
1925 struct sockaddr_storage *ss = &addr[i].in_addr;
1934 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1943 dout("missing matching ']'\n");
1950 if (p < end && *p == ':') {
1953 while (p < end && *p >= '0' && *p <= '9') {
1954 port = (port * 10) + (*p - '0');
1958 port = CEPH_MON_PORT;
1959 else if (port > 65535)
1962 port = CEPH_MON_PORT;
1965 addr_set_port(ss, port);
1967 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1984 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1987 EXPORT_SYMBOL(ceph_parse_ips);
1989 static int process_banner(struct ceph_connection *con)
1991 dout("process_banner on %p\n", con);
1993 if (verify_hello(con) < 0)
1996 ceph_decode_addr(&con->actual_peer_addr);
1997 ceph_decode_addr(&con->peer_addr_for_me);
2000 * Make sure the other end is who we wanted. note that the other
2001 * end may not yet know their ip address, so if it's 0.0.0.0, give
2002 * them the benefit of the doubt.
2004 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
2005 sizeof(con->peer_addr)) != 0 &&
2006 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
2007 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
2008 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
2009 ceph_pr_addr(&con->peer_addr.in_addr),
2010 (int)le32_to_cpu(con->peer_addr.nonce),
2011 ceph_pr_addr(&con->actual_peer_addr.in_addr),
2012 (int)le32_to_cpu(con->actual_peer_addr.nonce));
2013 con->error_msg = "wrong peer at address";
2018 * did we learn our address?
2020 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
2021 int port = addr_port(&con->msgr->inst.addr.in_addr);
2023 memcpy(&con->msgr->inst.addr.in_addr,
2024 &con->peer_addr_for_me.in_addr,
2025 sizeof(con->peer_addr_for_me.in_addr));
2026 addr_set_port(&con->msgr->inst.addr.in_addr, port);
2027 encode_my_addr(con->msgr);
2028 dout("process_banner learned my addr is %s\n",
2029 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
2035 static int process_connect(struct ceph_connection *con)
2037 u64 sup_feat = from_msgr(con->msgr)->supported_features;
2038 u64 req_feat = from_msgr(con->msgr)->required_features;
2039 u64 server_feat = ceph_sanitize_features(
2040 le64_to_cpu(con->in_reply.features));
2043 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
2045 if (con->auth_reply_buf) {
2047 * Any connection that defines ->get_authorizer()
2048 * should also define ->verify_authorizer_reply().
2049 * See get_connect_authorizer().
2051 ret = con->ops->verify_authorizer_reply(con, 0);
2053 con->error_msg = "bad authorize reply";
2058 switch (con->in_reply.tag) {
2059 case CEPH_MSGR_TAG_FEATURES:
2060 pr_err("%s%lld %s feature set mismatch,"
2061 " my %llx < server's %llx, missing %llx\n",
2062 ENTITY_NAME(con->peer_name),
2063 ceph_pr_addr(&con->peer_addr.in_addr),
2064 sup_feat, server_feat, server_feat & ~sup_feat);
2065 con->error_msg = "missing required protocol features";
2066 reset_connection(con);
2069 case CEPH_MSGR_TAG_BADPROTOVER:
2070 pr_err("%s%lld %s protocol version mismatch,"
2071 " my %d != server's %d\n",
2072 ENTITY_NAME(con->peer_name),
2073 ceph_pr_addr(&con->peer_addr.in_addr),
2074 le32_to_cpu(con->out_connect.protocol_version),
2075 le32_to_cpu(con->in_reply.protocol_version));
2076 con->error_msg = "protocol version mismatch";
2077 reset_connection(con);
2080 case CEPH_MSGR_TAG_BADAUTHORIZER:
2082 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2084 if (con->auth_retry == 2) {
2085 con->error_msg = "connect authorization failure";
2088 con_out_kvec_reset(con);
2089 ret = prepare_write_connect(con);
2092 prepare_read_connect(con);
2095 case CEPH_MSGR_TAG_RESETSESSION:
2097 * If we connected with a large connect_seq but the peer
2098 * has no record of a session with us (no connection, or
2099 * connect_seq == 0), they will send RESETSESION to indicate
2100 * that they must have reset their session, and may have
2103 dout("process_connect got RESET peer seq %u\n",
2104 le32_to_cpu(con->in_reply.connect_seq));
2105 pr_err("%s%lld %s connection reset\n",
2106 ENTITY_NAME(con->peer_name),
2107 ceph_pr_addr(&con->peer_addr.in_addr));
2108 reset_connection(con);
2109 con_out_kvec_reset(con);
2110 ret = prepare_write_connect(con);
2113 prepare_read_connect(con);
2115 /* Tell ceph about it. */
2116 mutex_unlock(&con->mutex);
2117 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2118 if (con->ops->peer_reset)
2119 con->ops->peer_reset(con);
2120 mutex_lock(&con->mutex);
2121 if (con->state != CON_STATE_NEGOTIATING)
2125 case CEPH_MSGR_TAG_RETRY_SESSION:
2127 * If we sent a smaller connect_seq than the peer has, try
2128 * again with a larger value.
2130 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2131 le32_to_cpu(con->out_connect.connect_seq),
2132 le32_to_cpu(con->in_reply.connect_seq));
2133 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2134 con_out_kvec_reset(con);
2135 ret = prepare_write_connect(con);
2138 prepare_read_connect(con);
2141 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2143 * If we sent a smaller global_seq than the peer has, try
2144 * again with a larger value.
2146 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2147 con->peer_global_seq,
2148 le32_to_cpu(con->in_reply.global_seq));
2149 get_global_seq(con->msgr,
2150 le32_to_cpu(con->in_reply.global_seq));
2151 con_out_kvec_reset(con);
2152 ret = prepare_write_connect(con);
2155 prepare_read_connect(con);
2158 case CEPH_MSGR_TAG_SEQ:
2159 case CEPH_MSGR_TAG_READY:
2160 if (req_feat & ~server_feat) {
2161 pr_err("%s%lld %s protocol feature mismatch,"
2162 " my required %llx > server's %llx, need %llx\n",
2163 ENTITY_NAME(con->peer_name),
2164 ceph_pr_addr(&con->peer_addr.in_addr),
2165 req_feat, server_feat, req_feat & ~server_feat);
2166 con->error_msg = "missing required protocol features";
2167 reset_connection(con);
2171 WARN_ON(con->state != CON_STATE_NEGOTIATING);
2172 con->state = CON_STATE_OPEN;
2173 con->auth_retry = 0; /* we authenticated; clear flag */
2174 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2176 con->peer_features = server_feat;
2177 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2178 con->peer_global_seq,
2179 le32_to_cpu(con->in_reply.connect_seq),
2181 WARN_ON(con->connect_seq !=
2182 le32_to_cpu(con->in_reply.connect_seq));
2184 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2185 con_flag_set(con, CON_FLAG_LOSSYTX);
2187 con->delay = 0; /* reset backoff memory */
2189 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2190 prepare_write_seq(con);
2191 prepare_read_seq(con);
2193 prepare_read_tag(con);
2197 case CEPH_MSGR_TAG_WAIT:
2199 * If there is a connection race (we are opening
2200 * connections to each other), one of us may just have
2201 * to WAIT. This shouldn't happen if we are the
2204 con->error_msg = "protocol error, got WAIT as client";
2208 con->error_msg = "protocol error, garbage tag during connect";
2216 * read (part of) an ack
2218 static int read_partial_ack(struct ceph_connection *con)
2220 int size = sizeof (con->in_temp_ack);
2223 return read_partial(con, end, size, &con->in_temp_ack);
2227 * We can finally discard anything that's been acked.
2229 static void process_ack(struct ceph_connection *con)
2232 u64 ack = le64_to_cpu(con->in_temp_ack);
2235 while (!list_empty(&con->out_sent)) {
2236 m = list_first_entry(&con->out_sent, struct ceph_msg,
2238 seq = le64_to_cpu(m->hdr.seq);
2241 dout("got ack for seq %llu type %d at %p\n", seq,
2242 le16_to_cpu(m->hdr.type), m);
2243 m->ack_stamp = jiffies;
2246 prepare_read_tag(con);
2250 static int read_partial_message_section(struct ceph_connection *con,
2251 struct kvec *section,
2252 unsigned int sec_len, u32 *crc)
2258 while (section->iov_len < sec_len) {
2259 BUG_ON(section->iov_base == NULL);
2260 left = sec_len - section->iov_len;
2261 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2262 section->iov_len, left);
2265 section->iov_len += ret;
2267 if (section->iov_len == sec_len)
2268 *crc = crc32c(0, section->iov_base, section->iov_len);
2273 static int read_partial_msg_data(struct ceph_connection *con)
2275 struct ceph_msg *msg = con->in_msg;
2276 struct ceph_msg_data_cursor *cursor = &msg->cursor;
2277 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2285 if (list_empty(&msg->data))
2289 crc = con->in_data_crc;
2290 while (cursor->resid) {
2291 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
2292 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2295 con->in_data_crc = crc;
2301 crc = ceph_crc32c_page(crc, page, page_offset, ret);
2302 (void) ceph_msg_data_advance(cursor, (size_t)ret);
2305 con->in_data_crc = crc;
2307 return 1; /* must return > 0 to indicate success */
2311 * read (part of) a message.
2313 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2315 static int read_partial_message(struct ceph_connection *con)
2317 struct ceph_msg *m = con->in_msg;
2321 unsigned int front_len, middle_len, data_len;
2322 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2323 bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
2327 dout("read_partial_message con %p msg %p\n", con, m);
2330 size = sizeof (con->in_hdr);
2332 ret = read_partial(con, end, size, &con->in_hdr);
2336 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2337 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2338 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
2339 crc, con->in_hdr.crc);
2343 front_len = le32_to_cpu(con->in_hdr.front_len);
2344 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2346 middle_len = le32_to_cpu(con->in_hdr.middle_len);
2347 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2349 data_len = le32_to_cpu(con->in_hdr.data_len);
2350 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2354 seq = le64_to_cpu(con->in_hdr.seq);
2355 if ((s64)seq - (s64)con->in_seq < 1) {
2356 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2357 ENTITY_NAME(con->peer_name),
2358 ceph_pr_addr(&con->peer_addr.in_addr),
2359 seq, con->in_seq + 1);
2360 con->in_base_pos = -front_len - middle_len - data_len -
2362 con->in_tag = CEPH_MSGR_TAG_READY;
2364 } else if ((s64)seq - (s64)con->in_seq > 1) {
2365 pr_err("read_partial_message bad seq %lld expected %lld\n",
2366 seq, con->in_seq + 1);
2367 con->error_msg = "bad message sequence # for incoming message";
2371 /* allocate message? */
2375 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2376 front_len, data_len);
2377 ret = ceph_con_in_msg_alloc(con, &skip);
2381 BUG_ON(!con->in_msg ^ skip);
2383 /* skip this message */
2384 dout("alloc_msg said skip message\n");
2385 con->in_base_pos = -front_len - middle_len - data_len -
2387 con->in_tag = CEPH_MSGR_TAG_READY;
2392 BUG_ON(!con->in_msg);
2393 BUG_ON(con->in_msg->con != con);
2395 m->front.iov_len = 0; /* haven't read it yet */
2397 m->middle->vec.iov_len = 0;
2399 /* prepare for data payload, if any */
2402 prepare_message_data(con->in_msg, data_len);
2406 ret = read_partial_message_section(con, &m->front, front_len,
2407 &con->in_front_crc);
2413 ret = read_partial_message_section(con, &m->middle->vec,
2415 &con->in_middle_crc);
2422 ret = read_partial_msg_data(con);
2429 size = sizeof(m->footer);
2431 size = sizeof(m->old_footer);
2434 ret = read_partial(con, end, size, &m->footer);
2439 m->footer.flags = m->old_footer.flags;
2443 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2444 m, front_len, m->footer.front_crc, middle_len,
2445 m->footer.middle_crc, data_len, m->footer.data_crc);
2448 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2449 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2450 m, con->in_front_crc, m->footer.front_crc);
2453 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2454 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2455 m, con->in_middle_crc, m->footer.middle_crc);
2459 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2460 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2461 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2462 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2466 if (need_sign && con->ops->check_message_signature &&
2467 con->ops->check_message_signature(m)) {
2468 pr_err("read_partial_message %p signature check failed\n", m);
2472 return 1; /* done! */
2476 * Process message. This happens in the worker thread. The callback should
2477 * be careful not to do anything that waits on other incoming messages or it
2480 static void process_message(struct ceph_connection *con)
2482 struct ceph_msg *msg = con->in_msg;
2484 BUG_ON(con->in_msg->con != con);
2487 /* if first message, set peer_name */
2488 if (con->peer_name.type == 0)
2489 con->peer_name = msg->hdr.src;
2492 mutex_unlock(&con->mutex);
2494 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2495 msg, le64_to_cpu(msg->hdr.seq),
2496 ENTITY_NAME(msg->hdr.src),
2497 le16_to_cpu(msg->hdr.type),
2498 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2499 le32_to_cpu(msg->hdr.front_len),
2500 le32_to_cpu(msg->hdr.data_len),
2501 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2502 con->ops->dispatch(con, msg);
2504 mutex_lock(&con->mutex);
2507 static int read_keepalive_ack(struct ceph_connection *con)
2509 struct ceph_timespec ceph_ts;
2510 size_t size = sizeof(ceph_ts);
2511 int ret = read_partial(con, size, size, &ceph_ts);
2514 ceph_decode_timespec(&con->last_keepalive_ack, &ceph_ts);
2515 prepare_read_tag(con);
2520 * Write something to the socket. Called in a worker thread when the
2521 * socket appears to be writeable and we have something ready to send.
2523 static int try_write(struct ceph_connection *con)
2527 dout("try_write start %p state %lu\n", con, con->state);
2530 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2532 /* open the socket first? */
2533 if (con->state == CON_STATE_PREOPEN) {
2535 con->state = CON_STATE_CONNECTING;
2537 con_out_kvec_reset(con);
2538 prepare_write_banner(con);
2539 prepare_read_banner(con);
2541 BUG_ON(con->in_msg);
2542 con->in_tag = CEPH_MSGR_TAG_READY;
2543 dout("try_write initiating connect on %p new state %lu\n",
2545 ret = ceph_tcp_connect(con);
2547 con->error_msg = "connect error";
2553 /* kvec data queued? */
2554 if (con->out_kvec_left) {
2555 ret = write_partial_kvec(con);
2559 if (con->out_skip) {
2560 ret = write_partial_skip(con);
2567 if (con->out_msg_done) {
2568 ceph_msg_put(con->out_msg);
2569 con->out_msg = NULL; /* we're done with this one */
2573 ret = write_partial_message_data(con);
2575 goto more_kvec; /* we need to send the footer, too! */
2579 dout("try_write write_partial_message_data err %d\n",
2586 if (con->state == CON_STATE_OPEN) {
2587 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2588 prepare_write_keepalive(con);
2591 /* is anything else pending? */
2592 if (!list_empty(&con->out_queue)) {
2593 prepare_write_message(con);
2596 if (con->in_seq > con->in_seq_acked) {
2597 prepare_write_ack(con);
2602 /* Nothing to do! */
2603 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2604 dout("try_write nothing else to write.\n");
2607 dout("try_write done on %p ret %d\n", con, ret);
2614 * Read what we can from the socket.
2616 static int try_read(struct ceph_connection *con)
2621 dout("try_read start on %p state %lu\n", con, con->state);
2622 if (con->state != CON_STATE_CONNECTING &&
2623 con->state != CON_STATE_NEGOTIATING &&
2624 con->state != CON_STATE_OPEN)
2629 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2632 if (con->state == CON_STATE_CONNECTING) {
2633 dout("try_read connecting\n");
2634 ret = read_partial_banner(con);
2637 ret = process_banner(con);
2641 con->state = CON_STATE_NEGOTIATING;
2644 * Received banner is good, exchange connection info.
2645 * Do not reset out_kvec, as sending our banner raced
2646 * with receiving peer banner after connect completed.
2648 ret = prepare_write_connect(con);
2651 prepare_read_connect(con);
2653 /* Send connection info before awaiting response */
2657 if (con->state == CON_STATE_NEGOTIATING) {
2658 dout("try_read negotiating\n");
2659 ret = read_partial_connect(con);
2662 ret = process_connect(con);
2668 WARN_ON(con->state != CON_STATE_OPEN);
2670 if (con->in_base_pos < 0) {
2672 * skipping + discarding content.
2674 * FIXME: there must be a better way to do this!
2676 static char buf[SKIP_BUF_SIZE];
2677 int skip = min((int) sizeof (buf), -con->in_base_pos);
2679 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2680 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2683 con->in_base_pos += ret;
2684 if (con->in_base_pos)
2687 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2691 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2694 dout("try_read got tag %d\n", (int)con->in_tag);
2695 switch (con->in_tag) {
2696 case CEPH_MSGR_TAG_MSG:
2697 prepare_read_message(con);
2699 case CEPH_MSGR_TAG_ACK:
2700 prepare_read_ack(con);
2702 case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
2703 prepare_read_keepalive_ack(con);
2705 case CEPH_MSGR_TAG_CLOSE:
2706 con_close_socket(con);
2707 con->state = CON_STATE_CLOSED;
2713 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2714 ret = read_partial_message(con);
2718 con->error_msg = "bad crc/signature";
2724 con->error_msg = "io error";
2729 if (con->in_tag == CEPH_MSGR_TAG_READY)
2731 process_message(con);
2732 if (con->state == CON_STATE_OPEN)
2733 prepare_read_tag(con);
2736 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2737 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2739 * the final handshake seq exchange is semantically
2740 * equivalent to an ACK
2742 ret = read_partial_ack(con);
2748 if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
2749 ret = read_keepalive_ack(con);
2756 dout("try_read done on %p ret %d\n", con, ret);
2760 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2761 con->error_msg = "protocol error, garbage tag";
2768 * Atomically queue work on a connection after the specified delay.
2769 * Bump @con reference to avoid races with connection teardown.
2770 * Returns 0 if work was queued, or an error code otherwise.
2772 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2774 if (!con->ops->get(con)) {
2775 dout("%s %p ref count 0\n", __func__, con);
2779 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2780 dout("%s %p - already queued\n", __func__, con);
2785 dout("%s %p %lu\n", __func__, con, delay);
2789 static void queue_con(struct ceph_connection *con)
2791 (void) queue_con_delay(con, 0);
2794 static void cancel_con(struct ceph_connection *con)
2796 if (cancel_delayed_work(&con->work)) {
2797 dout("%s %p\n", __func__, con);
2802 static bool con_sock_closed(struct ceph_connection *con)
2804 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2808 case CON_STATE_ ## x: \
2809 con->error_msg = "socket closed (con state " #x ")"; \
2812 switch (con->state) {
2820 pr_warn("%s con %p unrecognized state %lu\n",
2821 __func__, con, con->state);
2822 con->error_msg = "unrecognized con state";
2831 static bool con_backoff(struct ceph_connection *con)
2835 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2838 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2840 dout("%s: con %p FAILED to back off %lu\n", __func__,
2842 BUG_ON(ret == -ENOENT);
2843 con_flag_set(con, CON_FLAG_BACKOFF);
2849 /* Finish fault handling; con->mutex must *not* be held here */
2851 static void con_fault_finish(struct ceph_connection *con)
2854 * in case we faulted due to authentication, invalidate our
2855 * current tickets so that we can get new ones.
2857 if (con->auth_retry && con->ops->invalidate_authorizer) {
2858 dout("calling invalidate_authorizer()\n");
2859 con->ops->invalidate_authorizer(con);
2862 if (con->ops->fault)
2863 con->ops->fault(con);
2867 * Do some work on a connection. Drop a connection ref when we're done.
2869 static void ceph_con_workfn(struct work_struct *work)
2871 struct ceph_connection *con = container_of(work, struct ceph_connection,
2875 mutex_lock(&con->mutex);
2879 if ((fault = con_sock_closed(con))) {
2880 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2883 if (con_backoff(con)) {
2884 dout("%s: con %p BACKOFF\n", __func__, con);
2887 if (con->state == CON_STATE_STANDBY) {
2888 dout("%s: con %p STANDBY\n", __func__, con);
2891 if (con->state == CON_STATE_CLOSED) {
2892 dout("%s: con %p CLOSED\n", __func__, con);
2896 if (con->state == CON_STATE_PREOPEN) {
2897 dout("%s: con %p PREOPEN\n", __func__, con);
2901 ret = try_read(con);
2905 if (!con->error_msg)
2906 con->error_msg = "socket error on read";
2911 ret = try_write(con);
2915 if (!con->error_msg)
2916 con->error_msg = "socket error on write";
2920 break; /* If we make it to here, we're done */
2924 mutex_unlock(&con->mutex);
2927 con_fault_finish(con);
2933 * Generic error/fault handler. A retry mechanism is used with
2934 * exponential backoff
2936 static void con_fault(struct ceph_connection *con)
2938 dout("fault %p state %lu to peer %s\n",
2939 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2941 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2942 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2943 con->error_msg = NULL;
2945 WARN_ON(con->state != CON_STATE_CONNECTING &&
2946 con->state != CON_STATE_NEGOTIATING &&
2947 con->state != CON_STATE_OPEN);
2949 con_close_socket(con);
2951 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2952 dout("fault on LOSSYTX channel, marking CLOSED\n");
2953 con->state = CON_STATE_CLOSED;
2958 BUG_ON(con->in_msg->con != con);
2959 ceph_msg_put(con->in_msg);
2963 /* Requeue anything that hasn't been acked */
2964 list_splice_init(&con->out_sent, &con->out_queue);
2966 /* If there are no messages queued or keepalive pending, place
2967 * the connection in a STANDBY state */
2968 if (list_empty(&con->out_queue) &&
2969 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
2970 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2971 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2972 con->state = CON_STATE_STANDBY;
2974 /* retry after a delay. */
2975 con->state = CON_STATE_PREOPEN;
2976 if (con->delay == 0)
2977 con->delay = BASE_DELAY_INTERVAL;
2978 else if (con->delay < MAX_DELAY_INTERVAL)
2980 con_flag_set(con, CON_FLAG_BACKOFF);
2988 * initialize a new messenger instance
2990 void ceph_messenger_init(struct ceph_messenger *msgr,
2991 struct ceph_entity_addr *myaddr)
2993 spin_lock_init(&msgr->global_seq_lock);
2996 msgr->inst.addr = *myaddr;
2998 /* select a random nonce */
2999 msgr->inst.addr.type = 0;
3000 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
3001 encode_my_addr(msgr);
3003 atomic_set(&msgr->stopping, 0);
3004 write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
3006 dout("%s %p\n", __func__, msgr);
3008 EXPORT_SYMBOL(ceph_messenger_init);
3010 void ceph_messenger_fini(struct ceph_messenger *msgr)
3012 put_net(read_pnet(&msgr->net));
3014 EXPORT_SYMBOL(ceph_messenger_fini);
3016 static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con)
3019 msg->con->ops->put(msg->con);
3021 msg->con = con ? con->ops->get(con) : NULL;
3022 BUG_ON(msg->con != con);
3025 static void clear_standby(struct ceph_connection *con)
3027 /* come back from STANDBY? */
3028 if (con->state == CON_STATE_STANDBY) {
3029 dout("clear_standby %p and ++connect_seq\n", con);
3030 con->state = CON_STATE_PREOPEN;
3032 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
3033 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
3038 * Queue up an outgoing message on the given connection.
3040 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
3043 msg->hdr.src = con->msgr->inst.name;
3044 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
3045 msg->needs_out_seq = true;
3047 mutex_lock(&con->mutex);
3049 if (con->state == CON_STATE_CLOSED) {
3050 dout("con_send %p closed, dropping %p\n", con, msg);
3052 mutex_unlock(&con->mutex);
3056 msg_con_set(msg, con);
3058 BUG_ON(!list_empty(&msg->list_head));
3059 list_add_tail(&msg->list_head, &con->out_queue);
3060 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3061 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3062 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3063 le32_to_cpu(msg->hdr.front_len),
3064 le32_to_cpu(msg->hdr.middle_len),
3065 le32_to_cpu(msg->hdr.data_len));
3068 mutex_unlock(&con->mutex);
3070 /* if there wasn't anything waiting to send before, queue
3072 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3075 EXPORT_SYMBOL(ceph_con_send);
3078 * Revoke a message that was previously queued for send
3080 void ceph_msg_revoke(struct ceph_msg *msg)
3082 struct ceph_connection *con = msg->con;
3085 dout("%s msg %p null con\n", __func__, msg);
3086 return; /* Message not in our possession */
3089 mutex_lock(&con->mutex);
3090 if (!list_empty(&msg->list_head)) {
3091 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
3092 list_del_init(&msg->list_head);
3097 if (con->out_msg == msg) {
3098 BUG_ON(con->out_skip);
3100 if (con->out_msg_done) {
3101 con->out_skip += con_out_kvec_skip(con);
3103 BUG_ON(!msg->data_length);
3104 if (con->peer_features & CEPH_FEATURE_MSG_AUTH)
3105 con->out_skip += sizeof(msg->footer);
3107 con->out_skip += sizeof(msg->old_footer);
3109 /* data, middle, front */
3110 if (msg->data_length)
3111 con->out_skip += msg->cursor.total_resid;
3113 con->out_skip += con_out_kvec_skip(con);
3114 con->out_skip += con_out_kvec_skip(con);
3116 dout("%s %p msg %p - was sending, will write %d skip %d\n",
3117 __func__, con, msg, con->out_kvec_bytes, con->out_skip);
3119 con->out_msg = NULL;
3123 mutex_unlock(&con->mutex);
3127 * Revoke a message that we may be reading data into
3129 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
3131 struct ceph_connection *con = msg->con;
3134 dout("%s msg %p null con\n", __func__, msg);
3135 return; /* Message not in our possession */
3138 mutex_lock(&con->mutex);
3139 if (con->in_msg == msg) {
3140 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3141 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3142 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
3144 /* skip rest of message */
3145 dout("%s %p msg %p revoked\n", __func__, con, msg);
3146 con->in_base_pos = con->in_base_pos -
3147 sizeof(struct ceph_msg_header) -
3151 sizeof(struct ceph_msg_footer);
3152 ceph_msg_put(con->in_msg);
3154 con->in_tag = CEPH_MSGR_TAG_READY;
3157 dout("%s %p in_msg %p msg %p no-op\n",
3158 __func__, con, con->in_msg, msg);
3160 mutex_unlock(&con->mutex);
3164 * Queue a keepalive byte to ensure the tcp connection is alive.
3166 void ceph_con_keepalive(struct ceph_connection *con)
3168 dout("con_keepalive %p\n", con);
3169 mutex_lock(&con->mutex);
3171 mutex_unlock(&con->mutex);
3172 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3173 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3176 EXPORT_SYMBOL(ceph_con_keepalive);
3178 bool ceph_con_keepalive_expired(struct ceph_connection *con,
3179 unsigned long interval)
3182 (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
3183 struct timespec now = CURRENT_TIME;
3185 jiffies_to_timespec(interval, &ts);
3186 ts = timespec_add(con->last_keepalive_ack, ts);
3187 return timespec_compare(&now, &ts) >= 0;
3192 static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
3194 struct ceph_msg_data *data;
3196 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3199 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
3202 INIT_LIST_HEAD(&data->links);
3207 static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3212 WARN_ON(!list_empty(&data->links));
3213 if (data->type == CEPH_MSG_DATA_PAGELIST)
3214 ceph_pagelist_release(data->pagelist);
3215 kmem_cache_free(ceph_msg_data_cache, data);
3218 void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
3219 size_t length, size_t alignment)
3221 struct ceph_msg_data *data;
3226 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3228 data->pages = pages;
3229 data->length = length;
3230 data->alignment = alignment & ~PAGE_MASK;
3232 list_add_tail(&data->links, &msg->data);
3233 msg->data_length += length;
3235 EXPORT_SYMBOL(ceph_msg_data_add_pages);
3237 void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
3238 struct ceph_pagelist *pagelist)
3240 struct ceph_msg_data *data;
3243 BUG_ON(!pagelist->length);
3245 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3247 data->pagelist = pagelist;
3249 list_add_tail(&data->links, &msg->data);
3250 msg->data_length += pagelist->length;
3252 EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
3255 void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
3258 struct ceph_msg_data *data;
3262 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3265 data->bio_length = length;
3267 list_add_tail(&data->links, &msg->data);
3268 msg->data_length += length;
3270 EXPORT_SYMBOL(ceph_msg_data_add_bio);
3271 #endif /* CONFIG_BLOCK */
3274 * construct a new message with given type, size
3275 * the new msg has a ref count of 1.
3277 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3282 m = kmem_cache_zalloc(ceph_msg_cache, flags);
3286 m->hdr.type = cpu_to_le16(type);
3287 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3288 m->hdr.front_len = cpu_to_le32(front_len);
3290 INIT_LIST_HEAD(&m->list_head);
3291 kref_init(&m->kref);
3292 INIT_LIST_HEAD(&m->data);
3296 m->front.iov_base = ceph_kvmalloc(front_len, flags);
3297 if (m->front.iov_base == NULL) {
3298 dout("ceph_msg_new can't allocate %d bytes\n",
3303 m->front.iov_base = NULL;
3305 m->front_alloc_len = m->front.iov_len = front_len;
3307 dout("ceph_msg_new %p front %d\n", m, front_len);
3314 pr_err("msg_new can't create type %d front %d\n", type,
3318 dout("msg_new can't create type %d front %d\n", type,
3323 EXPORT_SYMBOL(ceph_msg_new);
3326 * Allocate "middle" portion of a message, if it is needed and wasn't
3327 * allocated by alloc_msg. This allows us to read a small fixed-size
3328 * per-type header in the front and then gracefully fail (i.e.,
3329 * propagate the error to the caller based on info in the front) when
3330 * the middle is too large.
3332 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3334 int type = le16_to_cpu(msg->hdr.type);
3335 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3337 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3338 ceph_msg_type_name(type), middle_len);
3339 BUG_ON(!middle_len);
3340 BUG_ON(msg->middle);
3342 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3349 * Allocate a message for receiving an incoming message on a
3350 * connection, and save the result in con->in_msg. Uses the
3351 * connection's private alloc_msg op if available.
3353 * Returns 0 on success, or a negative error code.
3355 * On success, if we set *skip = 1:
3356 * - the next message should be skipped and ignored.
3357 * - con->in_msg == NULL
3358 * or if we set *skip = 0:
3359 * - con->in_msg is non-null.
3360 * On error (ENOMEM, EAGAIN, ...),
3361 * - con->in_msg == NULL
3363 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3365 struct ceph_msg_header *hdr = &con->in_hdr;
3366 int middle_len = le32_to_cpu(hdr->middle_len);
3367 struct ceph_msg *msg;
3370 BUG_ON(con->in_msg != NULL);
3371 BUG_ON(!con->ops->alloc_msg);
3373 mutex_unlock(&con->mutex);
3374 msg = con->ops->alloc_msg(con, hdr, skip);
3375 mutex_lock(&con->mutex);
3376 if (con->state != CON_STATE_OPEN) {
3383 msg_con_set(msg, con);
3387 * Null message pointer means either we should skip
3388 * this message or we couldn't allocate memory. The
3389 * former is not an error.
3394 con->error_msg = "error allocating memory for incoming message";
3397 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3399 if (middle_len && !con->in_msg->middle) {
3400 ret = ceph_alloc_middle(con, con->in_msg);
3402 ceph_msg_put(con->in_msg);
3412 * Free a generically kmalloc'd message.
3414 static void ceph_msg_free(struct ceph_msg *m)
3416 dout("%s %p\n", __func__, m);
3417 kvfree(m->front.iov_base);
3418 kmem_cache_free(ceph_msg_cache, m);
3421 static void ceph_msg_release(struct kref *kref)
3423 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3425 struct list_head *links;
3426 struct list_head *next;
3428 dout("%s %p\n", __func__, m);
3429 WARN_ON(!list_empty(&m->list_head));
3431 msg_con_set(m, NULL);
3433 /* drop middle, data, if any */
3435 ceph_buffer_put(m->middle);
3439 list_splice_init(&m->data, &data);
3440 list_for_each_safe(links, next, &data) {
3441 struct ceph_msg_data *data;
3443 data = list_entry(links, struct ceph_msg_data, links);
3444 list_del_init(links);
3445 ceph_msg_data_destroy(data);
3450 ceph_msgpool_put(m->pool, m);
3455 struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3457 dout("%s %p (was %d)\n", __func__, msg,
3458 atomic_read(&msg->kref.refcount));
3459 kref_get(&msg->kref);
3462 EXPORT_SYMBOL(ceph_msg_get);
3464 void ceph_msg_put(struct ceph_msg *msg)
3466 dout("%s %p (was %d)\n", __func__, msg,
3467 atomic_read(&msg->kref.refcount));
3468 kref_put(&msg->kref, ceph_msg_release);
3470 EXPORT_SYMBOL(ceph_msg_put);
3472 void ceph_msg_dump(struct ceph_msg *msg)
3474 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3475 msg->front_alloc_len, msg->data_length);
3476 print_hex_dump(KERN_DEBUG, "header: ",
3477 DUMP_PREFIX_OFFSET, 16, 1,
3478 &msg->hdr, sizeof(msg->hdr), true);
3479 print_hex_dump(KERN_DEBUG, " front: ",
3480 DUMP_PREFIX_OFFSET, 16, 1,
3481 msg->front.iov_base, msg->front.iov_len, true);
3483 print_hex_dump(KERN_DEBUG, "middle: ",
3484 DUMP_PREFIX_OFFSET, 16, 1,
3485 msg->middle->vec.iov_base,
3486 msg->middle->vec.iov_len, true);
3487 print_hex_dump(KERN_DEBUG, "footer: ",
3488 DUMP_PREFIX_OFFSET, 16, 1,
3489 &msg->footer, sizeof(msg->footer), true);
3491 EXPORT_SYMBOL(ceph_msg_dump);