1 #include <linux/ceph/ceph_debug.h>
3 #include <linux/crc32c.h>
4 #include <linux/ctype.h>
5 #include <linux/highmem.h>
6 #include <linux/inet.h>
7 #include <linux/kthread.h>
9 #include <linux/slab.h>
10 #include <linux/socket.h>
11 #include <linux/string.h>
13 #include <linux/bio.h>
14 #endif /* CONFIG_BLOCK */
15 #include <linux/dns_resolver.h>
18 #include <linux/ceph/ceph_features.h>
19 #include <linux/ceph/libceph.h>
20 #include <linux/ceph/messenger.h>
21 #include <linux/ceph/decode.h>
22 #include <linux/ceph/pagelist.h>
23 #include <linux/export.h>
25 #define list_entry_next(pos, member) \
26 list_entry(pos->member.next, typeof(*pos), member)
29 * Ceph uses the messenger to exchange ceph_msg messages with other
30 * hosts in the system. The messenger provides ordered and reliable
31 * delivery. We tolerate TCP disconnects by reconnecting (with
32 * exponential backoff) in the case of a fault (disconnection, bad
33 * crc, protocol error). Acks allow sent messages to be discarded by
38 * We track the state of the socket on a given connection using
39 * values defined below. The transition to a new socket state is
40 * handled by a function which verifies we aren't coming from an
44 * | NEW* | transient initial state
46 * | con_sock_state_init()
49 * | CLOSED | initialized, but no socket (and no
50 * ---------- TCP connection)
52 * | \ con_sock_state_connecting()
53 * | ----------------------
55 * + con_sock_state_closed() \
56 * |+--------------------------- \
59 * | | CLOSING | socket event; \ \
60 * | ----------- await close \ \
63 * | + con_sock_state_closing() \ |
65 * | / --------------- | |
68 * | / -----------------| CONNECTING | socket created, TCP
69 * | | / -------------- connect initiated
70 * | | | con_sock_state_connected()
73 * | CONNECTED | TCP connection established
76 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
79 #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
80 #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
81 #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
82 #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
83 #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
88 #define CON_STATE_CLOSED 1 /* -> PREOPEN */
89 #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
90 #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
91 #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
92 #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
93 #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
96 * ceph_connection flag bits
98 #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
99 * messages on errors */
100 #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
101 #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
102 #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
103 #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
105 static bool con_flag_valid(unsigned long con_flag)
108 case CON_FLAG_LOSSYTX:
109 case CON_FLAG_KEEPALIVE_PENDING:
110 case CON_FLAG_WRITE_PENDING:
111 case CON_FLAG_SOCK_CLOSED:
112 case CON_FLAG_BACKOFF:
119 static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
121 BUG_ON(!con_flag_valid(con_flag));
123 clear_bit(con_flag, &con->flags);
126 static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
128 BUG_ON(!con_flag_valid(con_flag));
130 set_bit(con_flag, &con->flags);
133 static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
135 BUG_ON(!con_flag_valid(con_flag));
137 return test_bit(con_flag, &con->flags);
140 static bool con_flag_test_and_clear(struct ceph_connection *con,
141 unsigned long con_flag)
143 BUG_ON(!con_flag_valid(con_flag));
145 return test_and_clear_bit(con_flag, &con->flags);
148 static bool con_flag_test_and_set(struct ceph_connection *con,
149 unsigned long con_flag)
151 BUG_ON(!con_flag_valid(con_flag));
153 return test_and_set_bit(con_flag, &con->flags);
156 /* Slab caches for frequently-allocated structures */
158 static struct kmem_cache *ceph_msg_cache;
159 static struct kmem_cache *ceph_msg_data_cache;
161 /* static tag bytes (protocol control messages) */
162 static char tag_msg = CEPH_MSGR_TAG_MSG;
163 static char tag_ack = CEPH_MSGR_TAG_ACK;
164 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
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 con_work(struct work_struct *);
178 static void con_fault(struct ceph_connection *con);
181 * Nicely render a sockaddr as a string. An array of formatted
182 * strings is used, to approximate reentrancy.
184 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
185 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
186 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
187 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
189 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
190 static atomic_t addr_str_seq = ATOMIC_INIT(0);
192 static struct page *zero_page; /* used in certain error cases */
194 const char *ceph_pr_addr(const struct sockaddr_storage *ss)
198 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
199 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
201 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
204 switch (ss->ss_family) {
206 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
207 ntohs(in4->sin_port));
211 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
212 ntohs(in6->sin6_port));
216 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
222 EXPORT_SYMBOL(ceph_pr_addr);
224 static void encode_my_addr(struct ceph_messenger *msgr)
226 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
227 ceph_encode_addr(&msgr->my_enc_addr);
231 * work queue for all reading and writing to/from the socket.
233 static struct workqueue_struct *ceph_msgr_wq;
235 static int ceph_msgr_slab_init(void)
237 BUG_ON(ceph_msg_cache);
238 ceph_msg_cache = kmem_cache_create("ceph_msg",
239 sizeof (struct ceph_msg),
240 __alignof__(struct ceph_msg), 0, NULL);
245 BUG_ON(ceph_msg_data_cache);
246 ceph_msg_data_cache = kmem_cache_create("ceph_msg_data",
247 sizeof (struct ceph_msg_data),
248 __alignof__(struct ceph_msg_data),
250 if (ceph_msg_data_cache)
253 kmem_cache_destroy(ceph_msg_cache);
254 ceph_msg_cache = NULL;
259 static void ceph_msgr_slab_exit(void)
261 BUG_ON(!ceph_msg_data_cache);
262 kmem_cache_destroy(ceph_msg_data_cache);
263 ceph_msg_data_cache = NULL;
265 BUG_ON(!ceph_msg_cache);
266 kmem_cache_destroy(ceph_msg_cache);
267 ceph_msg_cache = NULL;
270 static void _ceph_msgr_exit(void)
273 destroy_workqueue(ceph_msgr_wq);
277 ceph_msgr_slab_exit();
279 BUG_ON(zero_page == NULL);
281 page_cache_release(zero_page);
285 int ceph_msgr_init(void)
287 BUG_ON(zero_page != NULL);
288 zero_page = ZERO_PAGE(0);
289 page_cache_get(zero_page);
291 if (ceph_msgr_slab_init())
294 ceph_msgr_wq = alloc_workqueue("ceph-msgr", 0, 0);
298 pr_err("msgr_init failed to create workqueue\n");
303 EXPORT_SYMBOL(ceph_msgr_init);
305 void ceph_msgr_exit(void)
307 BUG_ON(ceph_msgr_wq == NULL);
311 EXPORT_SYMBOL(ceph_msgr_exit);
313 void ceph_msgr_flush(void)
315 flush_workqueue(ceph_msgr_wq);
317 EXPORT_SYMBOL(ceph_msgr_flush);
319 /* Connection socket state transition functions */
321 static void con_sock_state_init(struct ceph_connection *con)
325 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
326 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
327 printk("%s: unexpected old state %d\n", __func__, old_state);
328 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
329 CON_SOCK_STATE_CLOSED);
332 static void con_sock_state_connecting(struct ceph_connection *con)
336 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
337 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
338 printk("%s: unexpected old state %d\n", __func__, old_state);
339 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
340 CON_SOCK_STATE_CONNECTING);
343 static void con_sock_state_connected(struct ceph_connection *con)
347 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
348 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
349 printk("%s: unexpected old state %d\n", __func__, old_state);
350 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
351 CON_SOCK_STATE_CONNECTED);
354 static void con_sock_state_closing(struct ceph_connection *con)
358 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
359 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
360 old_state != CON_SOCK_STATE_CONNECTED &&
361 old_state != CON_SOCK_STATE_CLOSING))
362 printk("%s: unexpected old state %d\n", __func__, old_state);
363 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
364 CON_SOCK_STATE_CLOSING);
367 static void con_sock_state_closed(struct ceph_connection *con)
371 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
372 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
373 old_state != CON_SOCK_STATE_CLOSING &&
374 old_state != CON_SOCK_STATE_CONNECTING &&
375 old_state != CON_SOCK_STATE_CLOSED))
376 printk("%s: unexpected old state %d\n", __func__, old_state);
377 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
378 CON_SOCK_STATE_CLOSED);
382 * socket callback functions
385 /* data available on socket, or listen socket received a connect */
386 static void ceph_sock_data_ready(struct sock *sk, int count_unused)
388 struct ceph_connection *con = sk->sk_user_data;
389 if (atomic_read(&con->msgr->stopping)) {
393 if (sk->sk_state != TCP_CLOSE_WAIT) {
394 dout("%s on %p state = %lu, queueing work\n", __func__,
400 /* socket has buffer space for writing */
401 static void ceph_sock_write_space(struct sock *sk)
403 struct ceph_connection *con = sk->sk_user_data;
405 /* only queue to workqueue if there is data we want to write,
406 * and there is sufficient space in the socket buffer to accept
407 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
408 * doesn't get called again until try_write() fills the socket
409 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
410 * and net/core/stream.c:sk_stream_write_space().
412 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
413 if (sk_stream_is_writeable(sk)) {
414 dout("%s %p queueing write work\n", __func__, con);
415 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
419 dout("%s %p nothing to write\n", __func__, con);
423 /* socket's state has changed */
424 static void ceph_sock_state_change(struct sock *sk)
426 struct ceph_connection *con = sk->sk_user_data;
428 dout("%s %p state = %lu sk_state = %u\n", __func__,
429 con, con->state, sk->sk_state);
431 switch (sk->sk_state) {
433 dout("%s TCP_CLOSE\n", __func__);
435 dout("%s TCP_CLOSE_WAIT\n", __func__);
436 con_sock_state_closing(con);
437 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
440 case TCP_ESTABLISHED:
441 dout("%s TCP_ESTABLISHED\n", __func__);
442 con_sock_state_connected(con);
445 default: /* Everything else is uninteresting */
451 * set up socket callbacks
453 static void set_sock_callbacks(struct socket *sock,
454 struct ceph_connection *con)
456 struct sock *sk = sock->sk;
457 sk->sk_user_data = con;
458 sk->sk_data_ready = ceph_sock_data_ready;
459 sk->sk_write_space = ceph_sock_write_space;
460 sk->sk_state_change = ceph_sock_state_change;
469 * initiate connection to a remote socket.
471 static int ceph_tcp_connect(struct ceph_connection *con)
473 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
478 ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM,
482 sock->sk->sk_allocation = GFP_NOFS;
484 #ifdef CONFIG_LOCKDEP
485 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
488 set_sock_callbacks(sock, con);
490 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
492 con_sock_state_connecting(con);
493 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
495 if (ret == -EINPROGRESS) {
496 dout("connect %s EINPROGRESS sk_state = %u\n",
497 ceph_pr_addr(&con->peer_addr.in_addr),
499 } else if (ret < 0) {
500 pr_err("connect %s error %d\n",
501 ceph_pr_addr(&con->peer_addr.in_addr), ret);
503 con->error_msg = "connect error";
511 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
513 struct kvec iov = {buf, len};
514 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
517 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
523 static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
524 int page_offset, size_t length)
529 BUG_ON(page_offset + length > PAGE_SIZE);
533 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
540 * write something. @more is true if caller will be sending more data
543 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
544 size_t kvlen, size_t len, int more)
546 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
550 msg.msg_flags |= MSG_MORE;
552 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
554 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
560 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
561 int offset, size_t size, bool more)
563 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
566 ret = kernel_sendpage(sock, page, offset, size, flags);
575 * Shutdown/close the socket for the given connection.
577 static int con_close_socket(struct ceph_connection *con)
581 dout("con_close_socket on %p sock %p\n", con, con->sock);
583 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
584 sock_release(con->sock);
589 * Forcibly clear the SOCK_CLOSED flag. It gets set
590 * independent of the connection mutex, and we could have
591 * received a socket close event before we had the chance to
592 * shut the socket down.
594 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
596 con_sock_state_closed(con);
601 * Reset a connection. Discard all incoming and outgoing messages
602 * and clear *_seq state.
604 static void ceph_msg_remove(struct ceph_msg *msg)
606 list_del_init(&msg->list_head);
607 BUG_ON(msg->con == NULL);
608 msg->con->ops->put(msg->con);
613 static void ceph_msg_remove_list(struct list_head *head)
615 while (!list_empty(head)) {
616 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
618 ceph_msg_remove(msg);
622 static void reset_connection(struct ceph_connection *con)
624 /* reset connection, out_queue, msg_ and connect_seq */
625 /* discard existing out_queue and msg_seq */
626 dout("reset_connection %p\n", con);
627 ceph_msg_remove_list(&con->out_queue);
628 ceph_msg_remove_list(&con->out_sent);
631 BUG_ON(con->in_msg->con != con);
632 con->in_msg->con = NULL;
633 ceph_msg_put(con->in_msg);
638 con->connect_seq = 0;
641 ceph_msg_put(con->out_msg);
645 con->in_seq_acked = 0;
649 * mark a peer down. drop any open connections.
651 void ceph_con_close(struct ceph_connection *con)
653 mutex_lock(&con->mutex);
654 dout("con_close %p peer %s\n", con,
655 ceph_pr_addr(&con->peer_addr.in_addr));
656 con->state = CON_STATE_CLOSED;
658 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
659 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
660 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
661 con_flag_clear(con, CON_FLAG_BACKOFF);
663 reset_connection(con);
664 con->peer_global_seq = 0;
665 cancel_delayed_work(&con->work);
666 con_close_socket(con);
667 mutex_unlock(&con->mutex);
669 EXPORT_SYMBOL(ceph_con_close);
672 * Reopen a closed connection, with a new peer address.
674 void ceph_con_open(struct ceph_connection *con,
675 __u8 entity_type, __u64 entity_num,
676 struct ceph_entity_addr *addr)
678 mutex_lock(&con->mutex);
679 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
681 WARN_ON(con->state != CON_STATE_CLOSED);
682 con->state = CON_STATE_PREOPEN;
684 con->peer_name.type = (__u8) entity_type;
685 con->peer_name.num = cpu_to_le64(entity_num);
687 memcpy(&con->peer_addr, addr, sizeof(*addr));
688 con->delay = 0; /* reset backoff memory */
689 mutex_unlock(&con->mutex);
692 EXPORT_SYMBOL(ceph_con_open);
695 * return true if this connection ever successfully opened
697 bool ceph_con_opened(struct ceph_connection *con)
699 return con->connect_seq > 0;
703 * initialize a new connection.
705 void ceph_con_init(struct ceph_connection *con, void *private,
706 const struct ceph_connection_operations *ops,
707 struct ceph_messenger *msgr)
709 dout("con_init %p\n", con);
710 memset(con, 0, sizeof(*con));
711 con->private = private;
715 con_sock_state_init(con);
717 mutex_init(&con->mutex);
718 INIT_LIST_HEAD(&con->out_queue);
719 INIT_LIST_HEAD(&con->out_sent);
720 INIT_DELAYED_WORK(&con->work, con_work);
722 con->state = CON_STATE_CLOSED;
724 EXPORT_SYMBOL(ceph_con_init);
728 * We maintain a global counter to order connection attempts. Get
729 * a unique seq greater than @gt.
731 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
735 spin_lock(&msgr->global_seq_lock);
736 if (msgr->global_seq < gt)
737 msgr->global_seq = gt;
738 ret = ++msgr->global_seq;
739 spin_unlock(&msgr->global_seq_lock);
743 static void con_out_kvec_reset(struct ceph_connection *con)
745 con->out_kvec_left = 0;
746 con->out_kvec_bytes = 0;
747 con->out_kvec_cur = &con->out_kvec[0];
750 static void con_out_kvec_add(struct ceph_connection *con,
751 size_t size, void *data)
755 index = con->out_kvec_left;
756 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
758 con->out_kvec[index].iov_len = size;
759 con->out_kvec[index].iov_base = data;
760 con->out_kvec_left++;
761 con->out_kvec_bytes += size;
767 * For a bio data item, a piece is whatever remains of the next
768 * entry in the current bio iovec, or the first entry in the next
771 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
774 struct ceph_msg_data *data = cursor->data;
777 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
782 cursor->resid = min(length, data->bio_length);
784 cursor->bvec_iter = bio->bi_iter;
786 cursor->resid <= bio_iter_len(bio, cursor->bvec_iter);
789 static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
793 struct ceph_msg_data *data = cursor->data;
795 struct bio_vec bio_vec;
797 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
802 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
804 *page_offset = (size_t) bio_vec.bv_offset;
805 BUG_ON(*page_offset >= PAGE_SIZE);
806 if (cursor->last_piece) /* pagelist offset is always 0 */
807 *length = cursor->resid;
809 *length = (size_t) bio_vec.bv_len;
810 BUG_ON(*length > cursor->resid);
811 BUG_ON(*page_offset + *length > PAGE_SIZE);
813 return bio_vec.bv_page;
816 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
820 struct bio_vec bio_vec;
822 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
827 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
829 /* Advance the cursor offset */
831 BUG_ON(cursor->resid < bytes);
832 cursor->resid -= bytes;
834 bio_advance_iter(bio, &cursor->bvec_iter, bytes);
836 if (bytes < bio_vec.bv_len)
837 return false; /* more bytes to process in this segment */
839 /* Move on to the next segment, and possibly the next bio */
841 if (!cursor->bvec_iter.bi_size) {
843 cursor->bvec_iter = bio->bi_iter;
847 if (!cursor->last_piece) {
848 BUG_ON(!cursor->resid);
850 /* A short read is OK, so use <= rather than == */
851 if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter))
852 cursor->last_piece = true;
857 #endif /* CONFIG_BLOCK */
860 * For a page array, a piece comes from the first page in the array
861 * that has not already been fully consumed.
863 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
866 struct ceph_msg_data *data = cursor->data;
869 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
871 BUG_ON(!data->pages);
872 BUG_ON(!data->length);
874 cursor->resid = min(length, data->length);
875 page_count = calc_pages_for(data->alignment, (u64)data->length);
876 cursor->page_offset = data->alignment & ~PAGE_MASK;
877 cursor->page_index = 0;
878 BUG_ON(page_count > (int)USHRT_MAX);
879 cursor->page_count = (unsigned short)page_count;
880 BUG_ON(length > SIZE_MAX - cursor->page_offset);
881 cursor->last_piece = (size_t)cursor->page_offset + length <= PAGE_SIZE;
885 ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
886 size_t *page_offset, size_t *length)
888 struct ceph_msg_data *data = cursor->data;
890 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
892 BUG_ON(cursor->page_index >= cursor->page_count);
893 BUG_ON(cursor->page_offset >= PAGE_SIZE);
895 *page_offset = cursor->page_offset;
896 if (cursor->last_piece)
897 *length = cursor->resid;
899 *length = PAGE_SIZE - *page_offset;
901 return data->pages[cursor->page_index];
904 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
907 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
909 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
911 /* Advance the cursor page offset */
913 cursor->resid -= bytes;
914 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
915 if (!bytes || cursor->page_offset)
916 return false; /* more bytes to process in the current page */
918 /* Move on to the next page; offset is already at 0 */
920 BUG_ON(cursor->page_index >= cursor->page_count);
921 cursor->page_index++;
922 cursor->last_piece = cursor->resid <= PAGE_SIZE;
928 * For a pagelist, a piece is whatever remains to be consumed in the
929 * first page in the list, or the front of the next page.
932 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
935 struct ceph_msg_data *data = cursor->data;
936 struct ceph_pagelist *pagelist;
939 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
941 pagelist = data->pagelist;
945 return; /* pagelist can be assigned but empty */
947 BUG_ON(list_empty(&pagelist->head));
948 page = list_first_entry(&pagelist->head, struct page, lru);
950 cursor->resid = min(length, pagelist->length);
953 cursor->last_piece = cursor->resid <= PAGE_SIZE;
957 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
958 size_t *page_offset, size_t *length)
960 struct ceph_msg_data *data = cursor->data;
961 struct ceph_pagelist *pagelist;
963 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
965 pagelist = data->pagelist;
968 BUG_ON(!cursor->page);
969 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
971 /* offset of first page in pagelist is always 0 */
972 *page_offset = cursor->offset & ~PAGE_MASK;
973 if (cursor->last_piece)
974 *length = cursor->resid;
976 *length = PAGE_SIZE - *page_offset;
981 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
984 struct ceph_msg_data *data = cursor->data;
985 struct ceph_pagelist *pagelist;
987 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
989 pagelist = data->pagelist;
992 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
993 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
995 /* Advance the cursor offset */
997 cursor->resid -= bytes;
998 cursor->offset += bytes;
999 /* offset of first page in pagelist is always 0 */
1000 if (!bytes || cursor->offset & ~PAGE_MASK)
1001 return false; /* more bytes to process in the current page */
1003 /* Move on to the next page */
1005 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1006 cursor->page = list_entry_next(cursor->page, lru);
1007 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1013 * Message data is handled (sent or received) in pieces, where each
1014 * piece resides on a single page. The network layer might not
1015 * consume an entire piece at once. A data item's cursor keeps
1016 * track of which piece is next to process and how much remains to
1017 * be processed in that piece. It also tracks whether the current
1018 * piece is the last one in the data item.
1020 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1022 size_t length = cursor->total_resid;
1024 switch (cursor->data->type) {
1025 case CEPH_MSG_DATA_PAGELIST:
1026 ceph_msg_data_pagelist_cursor_init(cursor, length);
1028 case CEPH_MSG_DATA_PAGES:
1029 ceph_msg_data_pages_cursor_init(cursor, length);
1032 case CEPH_MSG_DATA_BIO:
1033 ceph_msg_data_bio_cursor_init(cursor, length);
1035 #endif /* CONFIG_BLOCK */
1036 case CEPH_MSG_DATA_NONE:
1041 cursor->need_crc = true;
1044 static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1046 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1047 struct ceph_msg_data *data;
1050 BUG_ON(length > msg->data_length);
1051 BUG_ON(list_empty(&msg->data));
1053 cursor->data_head = &msg->data;
1054 cursor->total_resid = length;
1055 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1056 cursor->data = data;
1058 __ceph_msg_data_cursor_init(cursor);
1062 * Return the page containing the next piece to process for a given
1063 * data item, and supply the page offset and length of that piece.
1064 * Indicate whether this is the last piece in this data item.
1066 static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1067 size_t *page_offset, size_t *length,
1072 switch (cursor->data->type) {
1073 case CEPH_MSG_DATA_PAGELIST:
1074 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1076 case CEPH_MSG_DATA_PAGES:
1077 page = ceph_msg_data_pages_next(cursor, page_offset, length);
1080 case CEPH_MSG_DATA_BIO:
1081 page = ceph_msg_data_bio_next(cursor, page_offset, length);
1083 #endif /* CONFIG_BLOCK */
1084 case CEPH_MSG_DATA_NONE:
1090 BUG_ON(*page_offset + *length > PAGE_SIZE);
1093 *last_piece = cursor->last_piece;
1099 * Returns true if the result moves the cursor on to the next piece
1102 static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1107 BUG_ON(bytes > cursor->resid);
1108 switch (cursor->data->type) {
1109 case CEPH_MSG_DATA_PAGELIST:
1110 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1112 case CEPH_MSG_DATA_PAGES:
1113 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1116 case CEPH_MSG_DATA_BIO:
1117 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1119 #endif /* CONFIG_BLOCK */
1120 case CEPH_MSG_DATA_NONE:
1125 cursor->total_resid -= bytes;
1127 if (!cursor->resid && cursor->total_resid) {
1128 WARN_ON(!cursor->last_piece);
1129 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1130 cursor->data = list_entry_next(cursor->data, links);
1131 __ceph_msg_data_cursor_init(cursor);
1134 cursor->need_crc = new_piece;
1139 static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
1144 /* Initialize data cursor */
1146 ceph_msg_data_cursor_init(msg, (size_t)data_len);
1150 * Prepare footer for currently outgoing message, and finish things
1151 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1153 static void prepare_write_message_footer(struct ceph_connection *con)
1155 struct ceph_msg *m = con->out_msg;
1156 int v = con->out_kvec_left;
1158 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1160 dout("prepare_write_message_footer %p\n", con);
1161 con->out_kvec_is_msg = true;
1162 con->out_kvec[v].iov_base = &m->footer;
1163 con->out_kvec[v].iov_len = sizeof(m->footer);
1164 con->out_kvec_bytes += sizeof(m->footer);
1165 con->out_kvec_left++;
1166 con->out_more = m->more_to_follow;
1167 con->out_msg_done = true;
1171 * Prepare headers for the next outgoing message.
1173 static void prepare_write_message(struct ceph_connection *con)
1178 con_out_kvec_reset(con);
1179 con->out_kvec_is_msg = true;
1180 con->out_msg_done = false;
1182 /* Sneak an ack in there first? If we can get it into the same
1183 * TCP packet that's a good thing. */
1184 if (con->in_seq > con->in_seq_acked) {
1185 con->in_seq_acked = con->in_seq;
1186 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1187 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1188 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1189 &con->out_temp_ack);
1192 BUG_ON(list_empty(&con->out_queue));
1193 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1195 BUG_ON(m->con != con);
1197 /* put message on sent list */
1199 list_move_tail(&m->list_head, &con->out_sent);
1202 * only assign outgoing seq # if we haven't sent this message
1203 * yet. if it is requeued, resend with it's original seq.
1205 if (m->needs_out_seq) {
1206 m->hdr.seq = cpu_to_le64(++con->out_seq);
1207 m->needs_out_seq = false;
1209 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
1211 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1212 m, con->out_seq, le16_to_cpu(m->hdr.type),
1213 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1215 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1217 /* tag + hdr + front + middle */
1218 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1219 con_out_kvec_add(con, sizeof (m->hdr), &m->hdr);
1220 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1223 con_out_kvec_add(con, m->middle->vec.iov_len,
1224 m->middle->vec.iov_base);
1226 /* fill in crc (except data pages), footer */
1227 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1228 con->out_msg->hdr.crc = cpu_to_le32(crc);
1229 con->out_msg->footer.flags = 0;
1231 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1232 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1234 crc = crc32c(0, m->middle->vec.iov_base,
1235 m->middle->vec.iov_len);
1236 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1238 con->out_msg->footer.middle_crc = 0;
1239 dout("%s front_crc %u middle_crc %u\n", __func__,
1240 le32_to_cpu(con->out_msg->footer.front_crc),
1241 le32_to_cpu(con->out_msg->footer.middle_crc));
1243 /* is there a data payload? */
1244 con->out_msg->footer.data_crc = 0;
1245 if (m->data_length) {
1246 prepare_message_data(con->out_msg, m->data_length);
1247 con->out_more = 1; /* data + footer will follow */
1249 /* no, queue up footer too and be done */
1250 prepare_write_message_footer(con);
1253 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1259 static void prepare_write_ack(struct ceph_connection *con)
1261 dout("prepare_write_ack %p %llu -> %llu\n", con,
1262 con->in_seq_acked, con->in_seq);
1263 con->in_seq_acked = con->in_seq;
1265 con_out_kvec_reset(con);
1267 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);
1273 con->out_more = 1; /* more will follow.. eventually.. */
1274 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1278 * Prepare to share the seq during handshake
1280 static void prepare_write_seq(struct ceph_connection *con)
1282 dout("prepare_write_seq %p %llu -> %llu\n", con,
1283 con->in_seq_acked, con->in_seq);
1284 con->in_seq_acked = con->in_seq;
1286 con_out_kvec_reset(con);
1288 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1289 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1290 &con->out_temp_ack);
1292 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1296 * Prepare to write keepalive byte.
1298 static void prepare_write_keepalive(struct ceph_connection *con)
1300 dout("prepare_write_keepalive %p\n", con);
1301 con_out_kvec_reset(con);
1302 con_out_kvec_add(con, sizeof (tag_keepalive), &tag_keepalive);
1303 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1307 * Connection negotiation.
1310 static struct ceph_auth_handshake *get_connect_authorizer(struct ceph_connection *con,
1313 struct ceph_auth_handshake *auth;
1315 if (!con->ops->get_authorizer) {
1316 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1317 con->out_connect.authorizer_len = 0;
1321 /* Can't hold the mutex while getting authorizer */
1322 mutex_unlock(&con->mutex);
1323 auth = con->ops->get_authorizer(con, auth_proto, con->auth_retry);
1324 mutex_lock(&con->mutex);
1328 if (con->state != CON_STATE_NEGOTIATING)
1329 return ERR_PTR(-EAGAIN);
1331 con->auth_reply_buf = auth->authorizer_reply_buf;
1332 con->auth_reply_buf_len = auth->authorizer_reply_buf_len;
1337 * We connected to a peer and are saying hello.
1339 static void prepare_write_banner(struct ceph_connection *con)
1341 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1342 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1343 &con->msgr->my_enc_addr);
1346 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1349 static int prepare_write_connect(struct ceph_connection *con)
1351 unsigned int global_seq = get_global_seq(con->msgr, 0);
1354 struct ceph_auth_handshake *auth;
1356 switch (con->peer_name.type) {
1357 case CEPH_ENTITY_TYPE_MON:
1358 proto = CEPH_MONC_PROTOCOL;
1360 case CEPH_ENTITY_TYPE_OSD:
1361 proto = CEPH_OSDC_PROTOCOL;
1363 case CEPH_ENTITY_TYPE_MDS:
1364 proto = CEPH_MDSC_PROTOCOL;
1370 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1371 con->connect_seq, global_seq, proto);
1373 con->out_connect.features = cpu_to_le64(con->msgr->supported_features);
1374 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1375 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1376 con->out_connect.global_seq = cpu_to_le32(global_seq);
1377 con->out_connect.protocol_version = cpu_to_le32(proto);
1378 con->out_connect.flags = 0;
1380 auth_proto = CEPH_AUTH_UNKNOWN;
1381 auth = get_connect_authorizer(con, &auth_proto);
1383 return PTR_ERR(auth);
1385 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1386 con->out_connect.authorizer_len = auth ?
1387 cpu_to_le32(auth->authorizer_buf_len) : 0;
1389 con_out_kvec_add(con, sizeof (con->out_connect),
1391 if (auth && auth->authorizer_buf_len)
1392 con_out_kvec_add(con, auth->authorizer_buf_len,
1393 auth->authorizer_buf);
1396 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1402 * write as much of pending kvecs to the socket as we can.
1404 * 0 -> socket full, but more to do
1407 static int write_partial_kvec(struct ceph_connection *con)
1411 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1412 while (con->out_kvec_bytes > 0) {
1413 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1414 con->out_kvec_left, con->out_kvec_bytes,
1418 con->out_kvec_bytes -= ret;
1419 if (con->out_kvec_bytes == 0)
1422 /* account for full iov entries consumed */
1423 while (ret >= con->out_kvec_cur->iov_len) {
1424 BUG_ON(!con->out_kvec_left);
1425 ret -= con->out_kvec_cur->iov_len;
1426 con->out_kvec_cur++;
1427 con->out_kvec_left--;
1429 /* and for a partially-consumed entry */
1431 con->out_kvec_cur->iov_len -= ret;
1432 con->out_kvec_cur->iov_base += ret;
1435 con->out_kvec_left = 0;
1436 con->out_kvec_is_msg = false;
1439 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1440 con->out_kvec_bytes, con->out_kvec_left, ret);
1441 return ret; /* done! */
1444 static u32 ceph_crc32c_page(u32 crc, struct page *page,
1445 unsigned int page_offset,
1446 unsigned int length)
1451 BUG_ON(kaddr == NULL);
1452 crc = crc32c(crc, kaddr + page_offset, length);
1458 * Write as much message data payload as we can. If we finish, queue
1460 * 1 -> done, footer is now queued in out_kvec[].
1461 * 0 -> socket full, but more to do
1464 static int write_partial_message_data(struct ceph_connection *con)
1466 struct ceph_msg *msg = con->out_msg;
1467 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1468 bool do_datacrc = !con->msgr->nocrc;
1471 dout("%s %p msg %p\n", __func__, con, msg);
1473 if (list_empty(&msg->data))
1477 * Iterate through each page that contains data to be
1478 * written, and send as much as possible for each.
1480 * If we are calculating the data crc (the default), we will
1481 * need to map the page. If we have no pages, they have
1482 * been revoked, so use the zero page.
1484 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1485 while (cursor->resid) {
1493 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
1495 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
1496 length, last_piece);
1499 msg->footer.data_crc = cpu_to_le32(crc);
1503 if (do_datacrc && cursor->need_crc)
1504 crc = ceph_crc32c_page(crc, page, page_offset, length);
1505 need_crc = ceph_msg_data_advance(&msg->cursor, (size_t)ret);
1508 dout("%s %p msg %p done\n", __func__, con, msg);
1510 /* prepare and queue up footer, too */
1512 msg->footer.data_crc = cpu_to_le32(crc);
1514 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1515 con_out_kvec_reset(con);
1516 prepare_write_message_footer(con);
1518 return 1; /* must return > 0 to indicate success */
1524 static int write_partial_skip(struct ceph_connection *con)
1528 while (con->out_skip > 0) {
1529 size_t size = min(con->out_skip, (int) PAGE_CACHE_SIZE);
1531 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
1534 con->out_skip -= ret;
1542 * Prepare to read connection handshake, or an ack.
1544 static void prepare_read_banner(struct ceph_connection *con)
1546 dout("prepare_read_banner %p\n", con);
1547 con->in_base_pos = 0;
1550 static void prepare_read_connect(struct ceph_connection *con)
1552 dout("prepare_read_connect %p\n", con);
1553 con->in_base_pos = 0;
1556 static void prepare_read_ack(struct ceph_connection *con)
1558 dout("prepare_read_ack %p\n", con);
1559 con->in_base_pos = 0;
1562 static void prepare_read_seq(struct ceph_connection *con)
1564 dout("prepare_read_seq %p\n", con);
1565 con->in_base_pos = 0;
1566 con->in_tag = CEPH_MSGR_TAG_SEQ;
1569 static void prepare_read_tag(struct ceph_connection *con)
1571 dout("prepare_read_tag %p\n", con);
1572 con->in_base_pos = 0;
1573 con->in_tag = CEPH_MSGR_TAG_READY;
1577 * Prepare to read a message.
1579 static int prepare_read_message(struct ceph_connection *con)
1581 dout("prepare_read_message %p\n", con);
1582 BUG_ON(con->in_msg != NULL);
1583 con->in_base_pos = 0;
1584 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1589 static int read_partial(struct ceph_connection *con,
1590 int end, int size, void *object)
1592 while (con->in_base_pos < end) {
1593 int left = end - con->in_base_pos;
1594 int have = size - left;
1595 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1598 con->in_base_pos += ret;
1605 * Read all or part of the connect-side handshake on a new connection
1607 static int read_partial_banner(struct ceph_connection *con)
1613 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1616 size = strlen(CEPH_BANNER);
1618 ret = read_partial(con, end, size, con->in_banner);
1622 size = sizeof (con->actual_peer_addr);
1624 ret = read_partial(con, end, size, &con->actual_peer_addr);
1628 size = sizeof (con->peer_addr_for_me);
1630 ret = read_partial(con, end, size, &con->peer_addr_for_me);
1638 static int read_partial_connect(struct ceph_connection *con)
1644 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1646 size = sizeof (con->in_reply);
1648 ret = read_partial(con, end, size, &con->in_reply);
1652 size = le32_to_cpu(con->in_reply.authorizer_len);
1654 ret = read_partial(con, end, size, con->auth_reply_buf);
1658 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1659 con, (int)con->in_reply.tag,
1660 le32_to_cpu(con->in_reply.connect_seq),
1661 le32_to_cpu(con->in_reply.global_seq));
1668 * Verify the hello banner looks okay.
1670 static int verify_hello(struct ceph_connection *con)
1672 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1673 pr_err("connect to %s got bad banner\n",
1674 ceph_pr_addr(&con->peer_addr.in_addr));
1675 con->error_msg = "protocol error, bad banner";
1681 static bool addr_is_blank(struct sockaddr_storage *ss)
1683 switch (ss->ss_family) {
1685 return ((struct sockaddr_in *)ss)->sin_addr.s_addr == 0;
1688 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[0] == 0 &&
1689 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[1] == 0 &&
1690 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[2] == 0 &&
1691 ((struct sockaddr_in6 *)ss)->sin6_addr.s6_addr32[3] == 0;
1696 static int addr_port(struct sockaddr_storage *ss)
1698 switch (ss->ss_family) {
1700 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1702 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1707 static void addr_set_port(struct sockaddr_storage *ss, int p)
1709 switch (ss->ss_family) {
1711 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1714 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1720 * Unlike other *_pton function semantics, zero indicates success.
1722 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1723 char delim, const char **ipend)
1725 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1726 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1728 memset(ss, 0, sizeof(*ss));
1730 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1731 ss->ss_family = AF_INET;
1735 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1736 ss->ss_family = AF_INET6;
1744 * Extract hostname string and resolve using kernel DNS facility.
1746 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1747 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1748 struct sockaddr_storage *ss, char delim, const char **ipend)
1750 const char *end, *delim_p;
1751 char *colon_p, *ip_addr = NULL;
1755 * The end of the hostname occurs immediately preceding the delimiter or
1756 * the port marker (':') where the delimiter takes precedence.
1758 delim_p = memchr(name, delim, namelen);
1759 colon_p = memchr(name, ':', namelen);
1761 if (delim_p && colon_p)
1762 end = delim_p < colon_p ? delim_p : colon_p;
1763 else if (!delim_p && colon_p)
1767 if (!end) /* case: hostname:/ */
1768 end = name + namelen;
1774 /* do dns_resolve upcall */
1775 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1777 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1785 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1786 ret, ret ? "failed" : ceph_pr_addr(ss));
1791 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1792 struct sockaddr_storage *ss, char delim, const char **ipend)
1799 * Parse a server name (IP or hostname). If a valid IP address is not found
1800 * then try to extract a hostname to resolve using userspace DNS upcall.
1802 static int ceph_parse_server_name(const char *name, size_t namelen,
1803 struct sockaddr_storage *ss, char delim, const char **ipend)
1807 ret = ceph_pton(name, namelen, ss, delim, ipend);
1809 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1815 * Parse an ip[:port] list into an addr array. Use the default
1816 * monitor port if a port isn't specified.
1818 int ceph_parse_ips(const char *c, const char *end,
1819 struct ceph_entity_addr *addr,
1820 int max_count, int *count)
1822 int i, ret = -EINVAL;
1825 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1826 for (i = 0; i < max_count; i++) {
1828 struct sockaddr_storage *ss = &addr[i].in_addr;
1837 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1846 dout("missing matching ']'\n");
1853 if (p < end && *p == ':') {
1856 while (p < end && *p >= '0' && *p <= '9') {
1857 port = (port * 10) + (*p - '0');
1861 port = CEPH_MON_PORT;
1862 else if (port > 65535)
1865 port = CEPH_MON_PORT;
1868 addr_set_port(ss, port);
1870 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1887 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1890 EXPORT_SYMBOL(ceph_parse_ips);
1892 static int process_banner(struct ceph_connection *con)
1894 dout("process_banner on %p\n", con);
1896 if (verify_hello(con) < 0)
1899 ceph_decode_addr(&con->actual_peer_addr);
1900 ceph_decode_addr(&con->peer_addr_for_me);
1903 * Make sure the other end is who we wanted. note that the other
1904 * end may not yet know their ip address, so if it's 0.0.0.0, give
1905 * them the benefit of the doubt.
1907 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
1908 sizeof(con->peer_addr)) != 0 &&
1909 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
1910 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
1911 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1912 ceph_pr_addr(&con->peer_addr.in_addr),
1913 (int)le32_to_cpu(con->peer_addr.nonce),
1914 ceph_pr_addr(&con->actual_peer_addr.in_addr),
1915 (int)le32_to_cpu(con->actual_peer_addr.nonce));
1916 con->error_msg = "wrong peer at address";
1921 * did we learn our address?
1923 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
1924 int port = addr_port(&con->msgr->inst.addr.in_addr);
1926 memcpy(&con->msgr->inst.addr.in_addr,
1927 &con->peer_addr_for_me.in_addr,
1928 sizeof(con->peer_addr_for_me.in_addr));
1929 addr_set_port(&con->msgr->inst.addr.in_addr, port);
1930 encode_my_addr(con->msgr);
1931 dout("process_banner learned my addr is %s\n",
1932 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
1938 static int process_connect(struct ceph_connection *con)
1940 u64 sup_feat = con->msgr->supported_features;
1941 u64 req_feat = con->msgr->required_features;
1942 u64 server_feat = ceph_sanitize_features(
1943 le64_to_cpu(con->in_reply.features));
1946 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
1948 switch (con->in_reply.tag) {
1949 case CEPH_MSGR_TAG_FEATURES:
1950 pr_err("%s%lld %s feature set mismatch,"
1951 " my %llx < server's %llx, missing %llx\n",
1952 ENTITY_NAME(con->peer_name),
1953 ceph_pr_addr(&con->peer_addr.in_addr),
1954 sup_feat, server_feat, server_feat & ~sup_feat);
1955 con->error_msg = "missing required protocol features";
1956 reset_connection(con);
1959 case CEPH_MSGR_TAG_BADPROTOVER:
1960 pr_err("%s%lld %s protocol version mismatch,"
1961 " my %d != server's %d\n",
1962 ENTITY_NAME(con->peer_name),
1963 ceph_pr_addr(&con->peer_addr.in_addr),
1964 le32_to_cpu(con->out_connect.protocol_version),
1965 le32_to_cpu(con->in_reply.protocol_version));
1966 con->error_msg = "protocol version mismatch";
1967 reset_connection(con);
1970 case CEPH_MSGR_TAG_BADAUTHORIZER:
1972 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
1974 if (con->auth_retry == 2) {
1975 con->error_msg = "connect authorization failure";
1978 con_out_kvec_reset(con);
1979 ret = prepare_write_connect(con);
1982 prepare_read_connect(con);
1985 case CEPH_MSGR_TAG_RESETSESSION:
1987 * If we connected with a large connect_seq but the peer
1988 * has no record of a session with us (no connection, or
1989 * connect_seq == 0), they will send RESETSESION to indicate
1990 * that they must have reset their session, and may have
1993 dout("process_connect got RESET peer seq %u\n",
1994 le32_to_cpu(con->in_reply.connect_seq));
1995 pr_err("%s%lld %s connection reset\n",
1996 ENTITY_NAME(con->peer_name),
1997 ceph_pr_addr(&con->peer_addr.in_addr));
1998 reset_connection(con);
1999 con_out_kvec_reset(con);
2000 ret = prepare_write_connect(con);
2003 prepare_read_connect(con);
2005 /* Tell ceph about it. */
2006 mutex_unlock(&con->mutex);
2007 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2008 if (con->ops->peer_reset)
2009 con->ops->peer_reset(con);
2010 mutex_lock(&con->mutex);
2011 if (con->state != CON_STATE_NEGOTIATING)
2015 case CEPH_MSGR_TAG_RETRY_SESSION:
2017 * If we sent a smaller connect_seq than the peer has, try
2018 * again with a larger value.
2020 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2021 le32_to_cpu(con->out_connect.connect_seq),
2022 le32_to_cpu(con->in_reply.connect_seq));
2023 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2024 con_out_kvec_reset(con);
2025 ret = prepare_write_connect(con);
2028 prepare_read_connect(con);
2031 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2033 * If we sent a smaller global_seq than the peer has, try
2034 * again with a larger value.
2036 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2037 con->peer_global_seq,
2038 le32_to_cpu(con->in_reply.global_seq));
2039 get_global_seq(con->msgr,
2040 le32_to_cpu(con->in_reply.global_seq));
2041 con_out_kvec_reset(con);
2042 ret = prepare_write_connect(con);
2045 prepare_read_connect(con);
2048 case CEPH_MSGR_TAG_SEQ:
2049 case CEPH_MSGR_TAG_READY:
2050 if (req_feat & ~server_feat) {
2051 pr_err("%s%lld %s protocol feature mismatch,"
2052 " my required %llx > server's %llx, need %llx\n",
2053 ENTITY_NAME(con->peer_name),
2054 ceph_pr_addr(&con->peer_addr.in_addr),
2055 req_feat, server_feat, req_feat & ~server_feat);
2056 con->error_msg = "missing required protocol features";
2057 reset_connection(con);
2061 WARN_ON(con->state != CON_STATE_NEGOTIATING);
2062 con->state = CON_STATE_OPEN;
2063 con->auth_retry = 0; /* we authenticated; clear flag */
2064 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2066 con->peer_features = server_feat;
2067 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2068 con->peer_global_seq,
2069 le32_to_cpu(con->in_reply.connect_seq),
2071 WARN_ON(con->connect_seq !=
2072 le32_to_cpu(con->in_reply.connect_seq));
2074 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2075 con_flag_set(con, CON_FLAG_LOSSYTX);
2077 con->delay = 0; /* reset backoff memory */
2079 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2080 prepare_write_seq(con);
2081 prepare_read_seq(con);
2083 prepare_read_tag(con);
2087 case CEPH_MSGR_TAG_WAIT:
2089 * If there is a connection race (we are opening
2090 * connections to each other), one of us may just have
2091 * to WAIT. This shouldn't happen if we are the
2094 pr_err("process_connect got WAIT as client\n");
2095 con->error_msg = "protocol error, got WAIT as client";
2099 pr_err("connect protocol error, will retry\n");
2100 con->error_msg = "protocol error, garbage tag during connect";
2108 * read (part of) an ack
2110 static int read_partial_ack(struct ceph_connection *con)
2112 int size = sizeof (con->in_temp_ack);
2115 return read_partial(con, end, size, &con->in_temp_ack);
2119 * We can finally discard anything that's been acked.
2121 static void process_ack(struct ceph_connection *con)
2124 u64 ack = le64_to_cpu(con->in_temp_ack);
2127 while (!list_empty(&con->out_sent)) {
2128 m = list_first_entry(&con->out_sent, struct ceph_msg,
2130 seq = le64_to_cpu(m->hdr.seq);
2133 dout("got ack for seq %llu type %d at %p\n", seq,
2134 le16_to_cpu(m->hdr.type), m);
2135 m->ack_stamp = jiffies;
2138 prepare_read_tag(con);
2142 static int read_partial_message_section(struct ceph_connection *con,
2143 struct kvec *section,
2144 unsigned int sec_len, u32 *crc)
2150 while (section->iov_len < sec_len) {
2151 BUG_ON(section->iov_base == NULL);
2152 left = sec_len - section->iov_len;
2153 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2154 section->iov_len, left);
2157 section->iov_len += ret;
2159 if (section->iov_len == sec_len)
2160 *crc = crc32c(0, section->iov_base, section->iov_len);
2165 static int read_partial_msg_data(struct ceph_connection *con)
2167 struct ceph_msg *msg = con->in_msg;
2168 struct ceph_msg_data_cursor *cursor = &msg->cursor;
2169 const bool do_datacrc = !con->msgr->nocrc;
2177 if (list_empty(&msg->data))
2181 crc = con->in_data_crc;
2182 while (cursor->resid) {
2183 page = ceph_msg_data_next(&msg->cursor, &page_offset, &length,
2185 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2188 con->in_data_crc = crc;
2194 crc = ceph_crc32c_page(crc, page, page_offset, ret);
2195 (void) ceph_msg_data_advance(&msg->cursor, (size_t)ret);
2198 con->in_data_crc = crc;
2200 return 1; /* must return > 0 to indicate success */
2204 * read (part of) a message.
2206 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2208 static int read_partial_message(struct ceph_connection *con)
2210 struct ceph_msg *m = con->in_msg;
2214 unsigned int front_len, middle_len, data_len;
2215 bool do_datacrc = !con->msgr->nocrc;
2219 dout("read_partial_message con %p msg %p\n", con, m);
2222 size = sizeof (con->in_hdr);
2224 ret = read_partial(con, end, size, &con->in_hdr);
2228 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2229 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2230 pr_err("read_partial_message bad hdr "
2231 " crc %u != expected %u\n",
2232 crc, con->in_hdr.crc);
2236 front_len = le32_to_cpu(con->in_hdr.front_len);
2237 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2239 middle_len = le32_to_cpu(con->in_hdr.middle_len);
2240 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2242 data_len = le32_to_cpu(con->in_hdr.data_len);
2243 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2247 seq = le64_to_cpu(con->in_hdr.seq);
2248 if ((s64)seq - (s64)con->in_seq < 1) {
2249 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2250 ENTITY_NAME(con->peer_name),
2251 ceph_pr_addr(&con->peer_addr.in_addr),
2252 seq, con->in_seq + 1);
2253 con->in_base_pos = -front_len - middle_len - data_len -
2255 con->in_tag = CEPH_MSGR_TAG_READY;
2257 } else if ((s64)seq - (s64)con->in_seq > 1) {
2258 pr_err("read_partial_message bad seq %lld expected %lld\n",
2259 seq, con->in_seq + 1);
2260 con->error_msg = "bad message sequence # for incoming message";
2264 /* allocate message? */
2268 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2269 front_len, data_len);
2270 ret = ceph_con_in_msg_alloc(con, &skip);
2274 BUG_ON(!con->in_msg ^ skip);
2275 if (con->in_msg && data_len > con->in_msg->data_length) {
2276 pr_warning("%s skipping long message (%u > %zd)\n",
2277 __func__, data_len, con->in_msg->data_length);
2278 ceph_msg_put(con->in_msg);
2283 /* skip this message */
2284 dout("alloc_msg said skip message\n");
2285 con->in_base_pos = -front_len - middle_len - data_len -
2287 con->in_tag = CEPH_MSGR_TAG_READY;
2292 BUG_ON(!con->in_msg);
2293 BUG_ON(con->in_msg->con != con);
2295 m->front.iov_len = 0; /* haven't read it yet */
2297 m->middle->vec.iov_len = 0;
2299 /* prepare for data payload, if any */
2302 prepare_message_data(con->in_msg, data_len);
2306 ret = read_partial_message_section(con, &m->front, front_len,
2307 &con->in_front_crc);
2313 ret = read_partial_message_section(con, &m->middle->vec,
2315 &con->in_middle_crc);
2322 ret = read_partial_msg_data(con);
2328 size = sizeof (m->footer);
2330 ret = read_partial(con, end, size, &m->footer);
2334 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2335 m, front_len, m->footer.front_crc, middle_len,
2336 m->footer.middle_crc, data_len, m->footer.data_crc);
2339 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2340 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2341 m, con->in_front_crc, m->footer.front_crc);
2344 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2345 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2346 m, con->in_middle_crc, m->footer.middle_crc);
2350 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2351 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2352 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2353 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2357 return 1; /* done! */
2361 * Process message. This happens in the worker thread. The callback should
2362 * be careful not to do anything that waits on other incoming messages or it
2365 static void process_message(struct ceph_connection *con)
2367 struct ceph_msg *msg;
2369 BUG_ON(con->in_msg->con != con);
2370 con->in_msg->con = NULL;
2375 /* if first message, set peer_name */
2376 if (con->peer_name.type == 0)
2377 con->peer_name = msg->hdr.src;
2380 mutex_unlock(&con->mutex);
2382 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2383 msg, le64_to_cpu(msg->hdr.seq),
2384 ENTITY_NAME(msg->hdr.src),
2385 le16_to_cpu(msg->hdr.type),
2386 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2387 le32_to_cpu(msg->hdr.front_len),
2388 le32_to_cpu(msg->hdr.data_len),
2389 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2390 con->ops->dispatch(con, msg);
2392 mutex_lock(&con->mutex);
2397 * Write something to the socket. Called in a worker thread when the
2398 * socket appears to be writeable and we have something ready to send.
2400 static int try_write(struct ceph_connection *con)
2404 dout("try_write start %p state %lu\n", con, con->state);
2407 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2409 /* open the socket first? */
2410 if (con->state == CON_STATE_PREOPEN) {
2412 con->state = CON_STATE_CONNECTING;
2414 con_out_kvec_reset(con);
2415 prepare_write_banner(con);
2416 prepare_read_banner(con);
2418 BUG_ON(con->in_msg);
2419 con->in_tag = CEPH_MSGR_TAG_READY;
2420 dout("try_write initiating connect on %p new state %lu\n",
2422 ret = ceph_tcp_connect(con);
2424 con->error_msg = "connect error";
2430 /* kvec data queued? */
2431 if (con->out_skip) {
2432 ret = write_partial_skip(con);
2436 if (con->out_kvec_left) {
2437 ret = write_partial_kvec(con);
2444 if (con->out_msg_done) {
2445 ceph_msg_put(con->out_msg);
2446 con->out_msg = NULL; /* we're done with this one */
2450 ret = write_partial_message_data(con);
2452 goto more_kvec; /* we need to send the footer, too! */
2456 dout("try_write write_partial_message_data err %d\n",
2463 if (con->state == CON_STATE_OPEN) {
2464 /* is anything else pending? */
2465 if (!list_empty(&con->out_queue)) {
2466 prepare_write_message(con);
2469 if (con->in_seq > con->in_seq_acked) {
2470 prepare_write_ack(con);
2473 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2474 prepare_write_keepalive(con);
2479 /* Nothing to do! */
2480 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2481 dout("try_write nothing else to write.\n");
2484 dout("try_write done on %p ret %d\n", con, ret);
2491 * Read what we can from the socket.
2493 static int try_read(struct ceph_connection *con)
2498 dout("try_read start on %p state %lu\n", con, con->state);
2499 if (con->state != CON_STATE_CONNECTING &&
2500 con->state != CON_STATE_NEGOTIATING &&
2501 con->state != CON_STATE_OPEN)
2506 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2509 if (con->state == CON_STATE_CONNECTING) {
2510 dout("try_read connecting\n");
2511 ret = read_partial_banner(con);
2514 ret = process_banner(con);
2518 con->state = CON_STATE_NEGOTIATING;
2521 * Received banner is good, exchange connection info.
2522 * Do not reset out_kvec, as sending our banner raced
2523 * with receiving peer banner after connect completed.
2525 ret = prepare_write_connect(con);
2528 prepare_read_connect(con);
2530 /* Send connection info before awaiting response */
2534 if (con->state == CON_STATE_NEGOTIATING) {
2535 dout("try_read negotiating\n");
2536 ret = read_partial_connect(con);
2539 ret = process_connect(con);
2545 WARN_ON(con->state != CON_STATE_OPEN);
2547 if (con->in_base_pos < 0) {
2549 * skipping + discarding content.
2551 * FIXME: there must be a better way to do this!
2553 static char buf[SKIP_BUF_SIZE];
2554 int skip = min((int) sizeof (buf), -con->in_base_pos);
2556 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2557 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2560 con->in_base_pos += ret;
2561 if (con->in_base_pos)
2564 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2568 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2571 dout("try_read got tag %d\n", (int)con->in_tag);
2572 switch (con->in_tag) {
2573 case CEPH_MSGR_TAG_MSG:
2574 prepare_read_message(con);
2576 case CEPH_MSGR_TAG_ACK:
2577 prepare_read_ack(con);
2579 case CEPH_MSGR_TAG_CLOSE:
2580 con_close_socket(con);
2581 con->state = CON_STATE_CLOSED;
2587 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2588 ret = read_partial_message(con);
2592 con->error_msg = "bad crc";
2596 con->error_msg = "io error";
2601 if (con->in_tag == CEPH_MSGR_TAG_READY)
2603 process_message(con);
2604 if (con->state == CON_STATE_OPEN)
2605 prepare_read_tag(con);
2608 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2609 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2611 * the final handshake seq exchange is semantically
2612 * equivalent to an ACK
2614 ret = read_partial_ack(con);
2622 dout("try_read done on %p ret %d\n", con, ret);
2626 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2627 con->error_msg = "protocol error, garbage tag";
2634 * Atomically queue work on a connection after the specified delay.
2635 * Bump @con reference to avoid races with connection teardown.
2636 * Returns 0 if work was queued, or an error code otherwise.
2638 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2640 if (!con->ops->get(con)) {
2641 dout("%s %p ref count 0\n", __func__, con);
2646 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2647 dout("%s %p - already queued\n", __func__, con);
2653 dout("%s %p %lu\n", __func__, con, delay);
2658 static void queue_con(struct ceph_connection *con)
2660 (void) queue_con_delay(con, 0);
2663 static bool con_sock_closed(struct ceph_connection *con)
2665 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2669 case CON_STATE_ ## x: \
2670 con->error_msg = "socket closed (con state " #x ")"; \
2673 switch (con->state) {
2681 pr_warning("%s con %p unrecognized state %lu\n",
2682 __func__, con, con->state);
2683 con->error_msg = "unrecognized con state";
2692 static bool con_backoff(struct ceph_connection *con)
2696 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2699 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2701 dout("%s: con %p FAILED to back off %lu\n", __func__,
2703 BUG_ON(ret == -ENOENT);
2704 con_flag_set(con, CON_FLAG_BACKOFF);
2710 /* Finish fault handling; con->mutex must *not* be held here */
2712 static void con_fault_finish(struct ceph_connection *con)
2715 * in case we faulted due to authentication, invalidate our
2716 * current tickets so that we can get new ones.
2718 if (con->auth_retry && con->ops->invalidate_authorizer) {
2719 dout("calling invalidate_authorizer()\n");
2720 con->ops->invalidate_authorizer(con);
2723 if (con->ops->fault)
2724 con->ops->fault(con);
2728 * Do some work on a connection. Drop a connection ref when we're done.
2730 static void con_work(struct work_struct *work)
2732 struct ceph_connection *con = container_of(work, struct ceph_connection,
2736 mutex_lock(&con->mutex);
2740 if ((fault = con_sock_closed(con))) {
2741 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2744 if (con_backoff(con)) {
2745 dout("%s: con %p BACKOFF\n", __func__, con);
2748 if (con->state == CON_STATE_STANDBY) {
2749 dout("%s: con %p STANDBY\n", __func__, con);
2752 if (con->state == CON_STATE_CLOSED) {
2753 dout("%s: con %p CLOSED\n", __func__, con);
2757 if (con->state == CON_STATE_PREOPEN) {
2758 dout("%s: con %p PREOPEN\n", __func__, con);
2762 ret = try_read(con);
2766 con->error_msg = "socket error on read";
2771 ret = try_write(con);
2775 con->error_msg = "socket error on write";
2779 break; /* If we make it to here, we're done */
2783 mutex_unlock(&con->mutex);
2786 con_fault_finish(con);
2792 * Generic error/fault handler. A retry mechanism is used with
2793 * exponential backoff
2795 static void con_fault(struct ceph_connection *con)
2797 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2798 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2799 dout("fault %p state %lu to peer %s\n",
2800 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2802 WARN_ON(con->state != CON_STATE_CONNECTING &&
2803 con->state != CON_STATE_NEGOTIATING &&
2804 con->state != CON_STATE_OPEN);
2806 con_close_socket(con);
2808 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2809 dout("fault on LOSSYTX channel, marking CLOSED\n");
2810 con->state = CON_STATE_CLOSED;
2815 BUG_ON(con->in_msg->con != con);
2816 con->in_msg->con = NULL;
2817 ceph_msg_put(con->in_msg);
2822 /* Requeue anything that hasn't been acked */
2823 list_splice_init(&con->out_sent, &con->out_queue);
2825 /* If there are no messages queued or keepalive pending, place
2826 * the connection in a STANDBY state */
2827 if (list_empty(&con->out_queue) &&
2828 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
2829 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
2830 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2831 con->state = CON_STATE_STANDBY;
2833 /* retry after a delay. */
2834 con->state = CON_STATE_PREOPEN;
2835 if (con->delay == 0)
2836 con->delay = BASE_DELAY_INTERVAL;
2837 else if (con->delay < MAX_DELAY_INTERVAL)
2839 con_flag_set(con, CON_FLAG_BACKOFF);
2847 * initialize a new messenger instance
2849 void ceph_messenger_init(struct ceph_messenger *msgr,
2850 struct ceph_entity_addr *myaddr,
2851 u64 supported_features,
2852 u64 required_features,
2855 msgr->supported_features = supported_features;
2856 msgr->required_features = required_features;
2858 spin_lock_init(&msgr->global_seq_lock);
2861 msgr->inst.addr = *myaddr;
2863 /* select a random nonce */
2864 msgr->inst.addr.type = 0;
2865 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
2866 encode_my_addr(msgr);
2867 msgr->nocrc = nocrc;
2869 atomic_set(&msgr->stopping, 0);
2871 dout("%s %p\n", __func__, msgr);
2873 EXPORT_SYMBOL(ceph_messenger_init);
2875 static void clear_standby(struct ceph_connection *con)
2877 /* come back from STANDBY? */
2878 if (con->state == CON_STATE_STANDBY) {
2879 dout("clear_standby %p and ++connect_seq\n", con);
2880 con->state = CON_STATE_PREOPEN;
2882 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
2883 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
2888 * Queue up an outgoing message on the given connection.
2890 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
2893 msg->hdr.src = con->msgr->inst.name;
2894 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
2895 msg->needs_out_seq = true;
2897 mutex_lock(&con->mutex);
2899 if (con->state == CON_STATE_CLOSED) {
2900 dout("con_send %p closed, dropping %p\n", con, msg);
2902 mutex_unlock(&con->mutex);
2906 BUG_ON(msg->con != NULL);
2907 msg->con = con->ops->get(con);
2908 BUG_ON(msg->con == NULL);
2910 BUG_ON(!list_empty(&msg->list_head));
2911 list_add_tail(&msg->list_head, &con->out_queue);
2912 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
2913 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
2914 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2915 le32_to_cpu(msg->hdr.front_len),
2916 le32_to_cpu(msg->hdr.middle_len),
2917 le32_to_cpu(msg->hdr.data_len));
2920 mutex_unlock(&con->mutex);
2922 /* if there wasn't anything waiting to send before, queue
2924 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
2927 EXPORT_SYMBOL(ceph_con_send);
2930 * Revoke a message that was previously queued for send
2932 void ceph_msg_revoke(struct ceph_msg *msg)
2934 struct ceph_connection *con = msg->con;
2937 return; /* Message not in our possession */
2939 mutex_lock(&con->mutex);
2940 if (!list_empty(&msg->list_head)) {
2941 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
2942 list_del_init(&msg->list_head);
2943 BUG_ON(msg->con == NULL);
2944 msg->con->ops->put(msg->con);
2950 if (con->out_msg == msg) {
2951 dout("%s %p msg %p - was sending\n", __func__, con, msg);
2952 con->out_msg = NULL;
2953 if (con->out_kvec_is_msg) {
2954 con->out_skip = con->out_kvec_bytes;
2955 con->out_kvec_is_msg = false;
2961 mutex_unlock(&con->mutex);
2965 * Revoke a message that we may be reading data into
2967 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
2969 struct ceph_connection *con;
2971 BUG_ON(msg == NULL);
2973 dout("%s msg %p null con\n", __func__, msg);
2975 return; /* Message not in our possession */
2979 mutex_lock(&con->mutex);
2980 if (con->in_msg == msg) {
2981 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
2982 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
2983 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
2985 /* skip rest of message */
2986 dout("%s %p msg %p revoked\n", __func__, con, msg);
2987 con->in_base_pos = con->in_base_pos -
2988 sizeof(struct ceph_msg_header) -
2992 sizeof(struct ceph_msg_footer);
2993 ceph_msg_put(con->in_msg);
2995 con->in_tag = CEPH_MSGR_TAG_READY;
2998 dout("%s %p in_msg %p msg %p no-op\n",
2999 __func__, con, con->in_msg, msg);
3001 mutex_unlock(&con->mutex);
3005 * Queue a keepalive byte to ensure the tcp connection is alive.
3007 void ceph_con_keepalive(struct ceph_connection *con)
3009 dout("con_keepalive %p\n", con);
3010 mutex_lock(&con->mutex);
3012 mutex_unlock(&con->mutex);
3013 if (con_flag_test_and_set(con, CON_FLAG_KEEPALIVE_PENDING) == 0 &&
3014 con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3017 EXPORT_SYMBOL(ceph_con_keepalive);
3019 static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
3021 struct ceph_msg_data *data;
3023 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3026 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
3029 INIT_LIST_HEAD(&data->links);
3034 static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3039 WARN_ON(!list_empty(&data->links));
3040 if (data->type == CEPH_MSG_DATA_PAGELIST) {
3041 ceph_pagelist_release(data->pagelist);
3042 kfree(data->pagelist);
3044 kmem_cache_free(ceph_msg_data_cache, data);
3047 void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
3048 size_t length, size_t alignment)
3050 struct ceph_msg_data *data;
3055 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3057 data->pages = pages;
3058 data->length = length;
3059 data->alignment = alignment & ~PAGE_MASK;
3061 list_add_tail(&data->links, &msg->data);
3062 msg->data_length += length;
3064 EXPORT_SYMBOL(ceph_msg_data_add_pages);
3066 void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
3067 struct ceph_pagelist *pagelist)
3069 struct ceph_msg_data *data;
3072 BUG_ON(!pagelist->length);
3074 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3076 data->pagelist = pagelist;
3078 list_add_tail(&data->links, &msg->data);
3079 msg->data_length += pagelist->length;
3081 EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
3084 void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
3087 struct ceph_msg_data *data;
3091 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3094 data->bio_length = length;
3096 list_add_tail(&data->links, &msg->data);
3097 msg->data_length += length;
3099 EXPORT_SYMBOL(ceph_msg_data_add_bio);
3100 #endif /* CONFIG_BLOCK */
3103 * construct a new message with given type, size
3104 * the new msg has a ref count of 1.
3106 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3111 m = kmem_cache_zalloc(ceph_msg_cache, flags);
3115 m->hdr.type = cpu_to_le16(type);
3116 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3117 m->hdr.front_len = cpu_to_le32(front_len);
3119 INIT_LIST_HEAD(&m->list_head);
3120 kref_init(&m->kref);
3121 INIT_LIST_HEAD(&m->data);
3125 m->front.iov_base = ceph_kvmalloc(front_len, flags);
3126 if (m->front.iov_base == NULL) {
3127 dout("ceph_msg_new can't allocate %d bytes\n",
3132 m->front.iov_base = NULL;
3134 m->front_alloc_len = m->front.iov_len = front_len;
3136 dout("ceph_msg_new %p front %d\n", m, front_len);
3143 pr_err("msg_new can't create type %d front %d\n", type,
3147 dout("msg_new can't create type %d front %d\n", type,
3152 EXPORT_SYMBOL(ceph_msg_new);
3155 * Allocate "middle" portion of a message, if it is needed and wasn't
3156 * allocated by alloc_msg. This allows us to read a small fixed-size
3157 * per-type header in the front and then gracefully fail (i.e.,
3158 * propagate the error to the caller based on info in the front) when
3159 * the middle is too large.
3161 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3163 int type = le16_to_cpu(msg->hdr.type);
3164 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3166 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3167 ceph_msg_type_name(type), middle_len);
3168 BUG_ON(!middle_len);
3169 BUG_ON(msg->middle);
3171 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3178 * Allocate a message for receiving an incoming message on a
3179 * connection, and save the result in con->in_msg. Uses the
3180 * connection's private alloc_msg op if available.
3182 * Returns 0 on success, or a negative error code.
3184 * On success, if we set *skip = 1:
3185 * - the next message should be skipped and ignored.
3186 * - con->in_msg == NULL
3187 * or if we set *skip = 0:
3188 * - con->in_msg is non-null.
3189 * On error (ENOMEM, EAGAIN, ...),
3190 * - con->in_msg == NULL
3192 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3194 struct ceph_msg_header *hdr = &con->in_hdr;
3195 int middle_len = le32_to_cpu(hdr->middle_len);
3196 struct ceph_msg *msg;
3199 BUG_ON(con->in_msg != NULL);
3200 BUG_ON(!con->ops->alloc_msg);
3202 mutex_unlock(&con->mutex);
3203 msg = con->ops->alloc_msg(con, hdr, skip);
3204 mutex_lock(&con->mutex);
3205 if (con->state != CON_STATE_OPEN) {
3213 con->in_msg->con = con->ops->get(con);
3214 BUG_ON(con->in_msg->con == NULL);
3217 * Null message pointer means either we should skip
3218 * this message or we couldn't allocate memory. The
3219 * former is not an error.
3223 con->error_msg = "error allocating memory for incoming message";
3227 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3229 if (middle_len && !con->in_msg->middle) {
3230 ret = ceph_alloc_middle(con, con->in_msg);
3232 ceph_msg_put(con->in_msg);
3242 * Free a generically kmalloc'd message.
3244 void ceph_msg_kfree(struct ceph_msg *m)
3246 dout("msg_kfree %p\n", m);
3247 ceph_kvfree(m->front.iov_base);
3248 kmem_cache_free(ceph_msg_cache, m);
3252 * Drop a msg ref. Destroy as needed.
3254 void ceph_msg_last_put(struct kref *kref)
3256 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3258 struct list_head *links;
3259 struct list_head *next;
3261 dout("ceph_msg_put last one on %p\n", m);
3262 WARN_ON(!list_empty(&m->list_head));
3264 /* drop middle, data, if any */
3266 ceph_buffer_put(m->middle);
3270 list_splice_init(&m->data, &data);
3271 list_for_each_safe(links, next, &data) {
3272 struct ceph_msg_data *data;
3274 data = list_entry(links, struct ceph_msg_data, links);
3275 list_del_init(links);
3276 ceph_msg_data_destroy(data);
3281 ceph_msgpool_put(m->pool, m);
3285 EXPORT_SYMBOL(ceph_msg_last_put);
3287 void ceph_msg_dump(struct ceph_msg *msg)
3289 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3290 msg->front_alloc_len, msg->data_length);
3291 print_hex_dump(KERN_DEBUG, "header: ",
3292 DUMP_PREFIX_OFFSET, 16, 1,
3293 &msg->hdr, sizeof(msg->hdr), true);
3294 print_hex_dump(KERN_DEBUG, " front: ",
3295 DUMP_PREFIX_OFFSET, 16, 1,
3296 msg->front.iov_base, msg->front.iov_len, true);
3298 print_hex_dump(KERN_DEBUG, "middle: ",
3299 DUMP_PREFIX_OFFSET, 16, 1,
3300 msg->middle->vec.iov_base,
3301 msg->middle->vec.iov_len, true);
3302 print_hex_dump(KERN_DEBUG, "footer: ",
3303 DUMP_PREFIX_OFFSET, 16, 1,
3304 &msg->footer, sizeof(msg->footer), true);
3306 EXPORT_SYMBOL(ceph_msg_dump);