1 #include <linux/ceph/ceph_debug.h>
3 #include <linux/crc32c.h>
4 #include <linux/ctype.h>
5 #include <linux/highmem.h>
6 #include <linux/inet.h>
7 #include <linux/kthread.h>
9 #include <linux/nsproxy.h>
10 #include <linux/sched.h>
11 #include <linux/slab.h>
12 #include <linux/socket.h>
13 #include <linux/string.h>
15 #include <linux/bio.h>
16 #endif /* CONFIG_BLOCK */
17 #include <linux/dns_resolver.h>
20 #include <linux/ceph/ceph_features.h>
21 #include <linux/ceph/libceph.h>
22 #include <linux/ceph/messenger.h>
23 #include <linux/ceph/decode.h>
24 #include <linux/ceph/pagelist.h>
25 #include <linux/export.h>
28 * Ceph uses the messenger to exchange ceph_msg messages with other
29 * hosts in the system. The messenger provides ordered and reliable
30 * delivery. We tolerate TCP disconnects by reconnecting (with
31 * exponential backoff) in the case of a fault (disconnection, bad
32 * crc, protocol error). Acks allow sent messages to be discarded by
37 * We track the state of the socket on a given connection using
38 * values defined below. The transition to a new socket state is
39 * handled by a function which verifies we aren't coming from an
43 * | NEW* | transient initial state
45 * | con_sock_state_init()
48 * | CLOSED | initialized, but no socket (and no
49 * ---------- TCP connection)
51 * | \ con_sock_state_connecting()
52 * | ----------------------
54 * + con_sock_state_closed() \
55 * |+--------------------------- \
58 * | | CLOSING | socket event; \ \
59 * | ----------- await close \ \
62 * | + con_sock_state_closing() \ |
64 * | / --------------- | |
67 * | / -----------------| CONNECTING | socket created, TCP
68 * | | / -------------- connect initiated
69 * | | | con_sock_state_connected()
72 * | CONNECTED | TCP connection established
75 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
78 #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
79 #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
80 #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
81 #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
82 #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
87 #define CON_STATE_CLOSED 1 /* -> PREOPEN */
88 #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
89 #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
90 #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
91 #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
92 #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
95 * ceph_connection flag bits
97 #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
98 * messages on errors */
99 #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
100 #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
101 #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
102 #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
104 static bool con_flag_valid(unsigned long con_flag)
107 case CON_FLAG_LOSSYTX:
108 case CON_FLAG_KEEPALIVE_PENDING:
109 case CON_FLAG_WRITE_PENDING:
110 case CON_FLAG_SOCK_CLOSED:
111 case CON_FLAG_BACKOFF:
118 static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
120 BUG_ON(!con_flag_valid(con_flag));
122 clear_bit(con_flag, &con->flags);
125 static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
127 BUG_ON(!con_flag_valid(con_flag));
129 set_bit(con_flag, &con->flags);
132 static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
134 BUG_ON(!con_flag_valid(con_flag));
136 return test_bit(con_flag, &con->flags);
139 static bool con_flag_test_and_clear(struct ceph_connection *con,
140 unsigned long con_flag)
142 BUG_ON(!con_flag_valid(con_flag));
144 return test_and_clear_bit(con_flag, &con->flags);
147 static bool con_flag_test_and_set(struct ceph_connection *con,
148 unsigned long con_flag)
150 BUG_ON(!con_flag_valid(con_flag));
152 return test_and_set_bit(con_flag, &con->flags);
155 /* Slab caches for frequently-allocated structures */
157 static struct kmem_cache *ceph_msg_cache;
158 static struct kmem_cache *ceph_msg_data_cache;
160 /* static tag bytes (protocol control messages) */
161 static char tag_msg = CEPH_MSGR_TAG_MSG;
162 static char tag_ack = CEPH_MSGR_TAG_ACK;
163 static char tag_keepalive = CEPH_MSGR_TAG_KEEPALIVE;
164 static char tag_keepalive2 = CEPH_MSGR_TAG_KEEPALIVE2;
166 #ifdef CONFIG_LOCKDEP
167 static struct lock_class_key socket_class;
171 * When skipping (ignoring) a block of input we read it into a "skip
172 * buffer," which is this many bytes in size.
174 #define SKIP_BUF_SIZE 1024
176 static void queue_con(struct ceph_connection *con);
177 static void cancel_con(struct ceph_connection *con);
178 static void ceph_con_workfn(struct work_struct *);
179 static void con_fault(struct ceph_connection *con);
182 * Nicely render a sockaddr as a string. An array of formatted
183 * strings is used, to approximate reentrancy.
185 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
186 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
187 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
188 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
190 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
191 static atomic_t addr_str_seq = ATOMIC_INIT(0);
193 static struct page *zero_page; /* used in certain error cases */
195 const char *ceph_pr_addr(const struct sockaddr_storage *ss)
199 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
200 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
202 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
205 switch (ss->ss_family) {
207 snprintf(s, MAX_ADDR_STR_LEN, "%pI4:%hu", &in4->sin_addr,
208 ntohs(in4->sin_port));
212 snprintf(s, MAX_ADDR_STR_LEN, "[%pI6c]:%hu", &in6->sin6_addr,
213 ntohs(in6->sin6_port));
217 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
223 EXPORT_SYMBOL(ceph_pr_addr);
225 static void encode_my_addr(struct ceph_messenger *msgr)
227 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
228 ceph_encode_addr(&msgr->my_enc_addr);
232 * work queue for all reading and writing to/from the socket.
234 static struct workqueue_struct *ceph_msgr_wq;
236 static int ceph_msgr_slab_init(void)
238 BUG_ON(ceph_msg_cache);
239 ceph_msg_cache = KMEM_CACHE(ceph_msg, 0);
243 BUG_ON(ceph_msg_data_cache);
244 ceph_msg_data_cache = KMEM_CACHE(ceph_msg_data, 0);
245 if (ceph_msg_data_cache)
248 kmem_cache_destroy(ceph_msg_cache);
249 ceph_msg_cache = NULL;
254 static void ceph_msgr_slab_exit(void)
256 BUG_ON(!ceph_msg_data_cache);
257 kmem_cache_destroy(ceph_msg_data_cache);
258 ceph_msg_data_cache = NULL;
260 BUG_ON(!ceph_msg_cache);
261 kmem_cache_destroy(ceph_msg_cache);
262 ceph_msg_cache = NULL;
265 static void _ceph_msgr_exit(void)
268 destroy_workqueue(ceph_msgr_wq);
272 BUG_ON(zero_page == NULL);
276 ceph_msgr_slab_exit();
279 int ceph_msgr_init(void)
281 if (ceph_msgr_slab_init())
284 BUG_ON(zero_page != NULL);
285 zero_page = ZERO_PAGE(0);
289 * The number of active work items is limited by the number of
290 * connections, so leave @max_active at default.
292 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
296 pr_err("msgr_init failed to create workqueue\n");
301 EXPORT_SYMBOL(ceph_msgr_init);
303 void ceph_msgr_exit(void)
305 BUG_ON(ceph_msgr_wq == NULL);
309 EXPORT_SYMBOL(ceph_msgr_exit);
311 void ceph_msgr_flush(void)
313 flush_workqueue(ceph_msgr_wq);
315 EXPORT_SYMBOL(ceph_msgr_flush);
317 /* Connection socket state transition functions */
319 static void con_sock_state_init(struct ceph_connection *con)
323 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
324 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
325 printk("%s: unexpected old state %d\n", __func__, old_state);
326 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
327 CON_SOCK_STATE_CLOSED);
330 static void con_sock_state_connecting(struct ceph_connection *con)
334 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
335 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
336 printk("%s: unexpected old state %d\n", __func__, old_state);
337 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
338 CON_SOCK_STATE_CONNECTING);
341 static void con_sock_state_connected(struct ceph_connection *con)
345 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
346 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
347 printk("%s: unexpected old state %d\n", __func__, old_state);
348 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
349 CON_SOCK_STATE_CONNECTED);
352 static void con_sock_state_closing(struct ceph_connection *con)
356 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
357 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
358 old_state != CON_SOCK_STATE_CONNECTED &&
359 old_state != CON_SOCK_STATE_CLOSING))
360 printk("%s: unexpected old state %d\n", __func__, old_state);
361 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
362 CON_SOCK_STATE_CLOSING);
365 static void con_sock_state_closed(struct ceph_connection *con)
369 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
370 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
371 old_state != CON_SOCK_STATE_CLOSING &&
372 old_state != CON_SOCK_STATE_CONNECTING &&
373 old_state != CON_SOCK_STATE_CLOSED))
374 printk("%s: unexpected old state %d\n", __func__, old_state);
375 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
376 CON_SOCK_STATE_CLOSED);
380 * socket callback functions
383 /* data available on socket, or listen socket received a connect */
384 static void ceph_sock_data_ready(struct sock *sk)
386 struct ceph_connection *con = sk->sk_user_data;
387 if (atomic_read(&con->msgr->stopping)) {
391 if (sk->sk_state != TCP_CLOSE_WAIT) {
392 dout("%s on %p state = %lu, queueing work\n", __func__,
398 /* socket has buffer space for writing */
399 static void ceph_sock_write_space(struct sock *sk)
401 struct ceph_connection *con = sk->sk_user_data;
403 /* only queue to workqueue if there is data we want to write,
404 * and there is sufficient space in the socket buffer to accept
405 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
406 * doesn't get called again until try_write() fills the socket
407 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
408 * and net/core/stream.c:sk_stream_write_space().
410 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
411 if (sk_stream_is_writeable(sk)) {
412 dout("%s %p queueing write work\n", __func__, con);
413 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
417 dout("%s %p nothing to write\n", __func__, con);
421 /* socket's state has changed */
422 static void ceph_sock_state_change(struct sock *sk)
424 struct ceph_connection *con = sk->sk_user_data;
426 dout("%s %p state = %lu sk_state = %u\n", __func__,
427 con, con->state, sk->sk_state);
429 switch (sk->sk_state) {
431 dout("%s TCP_CLOSE\n", __func__);
433 dout("%s TCP_CLOSE_WAIT\n", __func__);
434 con_sock_state_closing(con);
435 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
438 case TCP_ESTABLISHED:
439 dout("%s TCP_ESTABLISHED\n", __func__);
440 con_sock_state_connected(con);
443 default: /* Everything else is uninteresting */
449 * set up socket callbacks
451 static void set_sock_callbacks(struct socket *sock,
452 struct ceph_connection *con)
454 struct sock *sk = sock->sk;
455 sk->sk_user_data = con;
456 sk->sk_data_ready = ceph_sock_data_ready;
457 sk->sk_write_space = ceph_sock_write_space;
458 sk->sk_state_change = ceph_sock_state_change;
467 * initiate connection to a remote socket.
469 static int ceph_tcp_connect(struct ceph_connection *con)
471 struct sockaddr_storage *paddr = &con->peer_addr.in_addr;
473 unsigned int noio_flag;
478 /* sock_create_kern() allocates with GFP_KERNEL */
479 noio_flag = memalloc_noio_save();
480 ret = sock_create_kern(read_pnet(&con->msgr->net), paddr->ss_family,
481 SOCK_STREAM, IPPROTO_TCP, &sock);
482 memalloc_noio_restore(noio_flag);
485 sock->sk->sk_allocation = GFP_NOFS;
487 #ifdef CONFIG_LOCKDEP
488 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
491 set_sock_callbacks(sock, con);
493 dout("connect %s\n", ceph_pr_addr(&con->peer_addr.in_addr));
495 con_sock_state_connecting(con);
496 ret = sock->ops->connect(sock, (struct sockaddr *)paddr, sizeof(*paddr),
498 if (ret == -EINPROGRESS) {
499 dout("connect %s EINPROGRESS sk_state = %u\n",
500 ceph_pr_addr(&con->peer_addr.in_addr),
502 } else if (ret < 0) {
503 pr_err("connect %s error %d\n",
504 ceph_pr_addr(&con->peer_addr.in_addr), ret);
509 if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY)) {
512 ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
513 (char *)&optval, sizeof(optval));
515 pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
523 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
525 struct kvec iov = {buf, len};
526 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
529 r = kernel_recvmsg(sock, &msg, &iov, 1, len, msg.msg_flags);
535 static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
536 int page_offset, size_t length)
541 BUG_ON(page_offset + length > PAGE_SIZE);
545 ret = ceph_tcp_recvmsg(sock, kaddr + page_offset, length);
552 * write something. @more is true if caller will be sending more data
555 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
556 size_t kvlen, size_t len, int more)
558 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
562 msg.msg_flags |= MSG_MORE;
564 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
566 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
572 static int __ceph_tcp_sendpage(struct socket *sock, struct page *page,
573 int offset, size_t size, bool more)
575 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | (more ? MSG_MORE : MSG_EOR);
578 ret = kernel_sendpage(sock, page, offset, size, flags);
585 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
586 int offset, size_t size, bool more)
592 * sendpage cannot properly handle pages with page_count == 0,
593 * we need to fall back to sendmsg if that's the case.
595 * Same goes for slab pages: skb_can_coalesce() allows
596 * coalescing neighboring slab objects into a single frag which
597 * triggers one of hardened usercopy checks.
599 if (page_count(page) >= 1 && !PageSlab(page))
600 return __ceph_tcp_sendpage(sock, page, offset, size, more);
602 iov.iov_base = kmap(page) + offset;
604 ret = ceph_tcp_sendmsg(sock, &iov, 1, size, more);
611 * Shutdown/close the socket for the given connection.
613 static int con_close_socket(struct ceph_connection *con)
617 dout("con_close_socket on %p sock %p\n", con, con->sock);
619 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
620 sock_release(con->sock);
625 * Forcibly clear the SOCK_CLOSED flag. It gets set
626 * independent of the connection mutex, and we could have
627 * received a socket close event before we had the chance to
628 * shut the socket down.
630 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
632 con_sock_state_closed(con);
637 * Reset a connection. Discard all incoming and outgoing messages
638 * and clear *_seq state.
640 static void ceph_msg_remove(struct ceph_msg *msg)
642 list_del_init(&msg->list_head);
646 static void ceph_msg_remove_list(struct list_head *head)
648 while (!list_empty(head)) {
649 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
651 ceph_msg_remove(msg);
655 static void reset_connection(struct ceph_connection *con)
657 /* reset connection, out_queue, msg_ and connect_seq */
658 /* discard existing out_queue and msg_seq */
659 dout("reset_connection %p\n", con);
660 ceph_msg_remove_list(&con->out_queue);
661 ceph_msg_remove_list(&con->out_sent);
664 BUG_ON(con->in_msg->con != con);
665 ceph_msg_put(con->in_msg);
669 con->connect_seq = 0;
672 BUG_ON(con->out_msg->con != con);
673 ceph_msg_put(con->out_msg);
677 con->in_seq_acked = 0;
683 * mark a peer down. drop any open connections.
685 void ceph_con_close(struct ceph_connection *con)
687 mutex_lock(&con->mutex);
688 dout("con_close %p peer %s\n", con,
689 ceph_pr_addr(&con->peer_addr.in_addr));
690 con->state = CON_STATE_CLOSED;
692 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
693 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
694 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
695 con_flag_clear(con, CON_FLAG_BACKOFF);
697 reset_connection(con);
698 con->peer_global_seq = 0;
700 con_close_socket(con);
701 mutex_unlock(&con->mutex);
703 EXPORT_SYMBOL(ceph_con_close);
706 * Reopen a closed connection, with a new peer address.
708 void ceph_con_open(struct ceph_connection *con,
709 __u8 entity_type, __u64 entity_num,
710 struct ceph_entity_addr *addr)
712 mutex_lock(&con->mutex);
713 dout("con_open %p %s\n", con, ceph_pr_addr(&addr->in_addr));
715 WARN_ON(con->state != CON_STATE_CLOSED);
716 con->state = CON_STATE_PREOPEN;
718 con->peer_name.type = (__u8) entity_type;
719 con->peer_name.num = cpu_to_le64(entity_num);
721 memcpy(&con->peer_addr, addr, sizeof(*addr));
722 con->delay = 0; /* reset backoff memory */
723 mutex_unlock(&con->mutex);
726 EXPORT_SYMBOL(ceph_con_open);
729 * return true if this connection ever successfully opened
731 bool ceph_con_opened(struct ceph_connection *con)
733 return con->connect_seq > 0;
737 * initialize a new connection.
739 void ceph_con_init(struct ceph_connection *con, void *private,
740 const struct ceph_connection_operations *ops,
741 struct ceph_messenger *msgr)
743 dout("con_init %p\n", con);
744 memset(con, 0, sizeof(*con));
745 con->private = private;
749 con_sock_state_init(con);
751 mutex_init(&con->mutex);
752 INIT_LIST_HEAD(&con->out_queue);
753 INIT_LIST_HEAD(&con->out_sent);
754 INIT_DELAYED_WORK(&con->work, ceph_con_workfn);
756 con->state = CON_STATE_CLOSED;
758 EXPORT_SYMBOL(ceph_con_init);
762 * We maintain a global counter to order connection attempts. Get
763 * a unique seq greater than @gt.
765 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
769 spin_lock(&msgr->global_seq_lock);
770 if (msgr->global_seq < gt)
771 msgr->global_seq = gt;
772 ret = ++msgr->global_seq;
773 spin_unlock(&msgr->global_seq_lock);
777 static void con_out_kvec_reset(struct ceph_connection *con)
779 BUG_ON(con->out_skip);
781 con->out_kvec_left = 0;
782 con->out_kvec_bytes = 0;
783 con->out_kvec_cur = &con->out_kvec[0];
786 static void con_out_kvec_add(struct ceph_connection *con,
787 size_t size, void *data)
789 int index = con->out_kvec_left;
791 BUG_ON(con->out_skip);
792 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
794 con->out_kvec[index].iov_len = size;
795 con->out_kvec[index].iov_base = data;
796 con->out_kvec_left++;
797 con->out_kvec_bytes += size;
801 * Chop off a kvec from the end. Return residual number of bytes for
802 * that kvec, i.e. how many bytes would have been written if the kvec
805 static int con_out_kvec_skip(struct ceph_connection *con)
807 int off = con->out_kvec_cur - con->out_kvec;
810 if (con->out_kvec_bytes > 0) {
811 skip = con->out_kvec[off + con->out_kvec_left - 1].iov_len;
812 BUG_ON(con->out_kvec_bytes < skip);
813 BUG_ON(!con->out_kvec_left);
814 con->out_kvec_bytes -= skip;
815 con->out_kvec_left--;
824 * For a bio data item, a piece is whatever remains of the next
825 * entry in the current bio iovec, or the first entry in the next
828 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
831 struct ceph_msg_data *data = cursor->data;
834 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
839 cursor->resid = min(length, data->bio_length);
841 cursor->bvec_iter = bio->bi_iter;
843 cursor->resid <= bio_iter_len(bio, cursor->bvec_iter);
846 static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
850 struct ceph_msg_data *data = cursor->data;
852 struct bio_vec bio_vec;
854 BUG_ON(data->type != CEPH_MSG_DATA_BIO);
859 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
861 *page_offset = (size_t) bio_vec.bv_offset;
862 BUG_ON(*page_offset >= PAGE_SIZE);
863 if (cursor->last_piece) /* pagelist offset is always 0 */
864 *length = cursor->resid;
866 *length = (size_t) bio_vec.bv_len;
867 BUG_ON(*length > cursor->resid);
868 BUG_ON(*page_offset + *length > PAGE_SIZE);
870 return bio_vec.bv_page;
873 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
877 struct bio_vec bio_vec;
879 BUG_ON(cursor->data->type != CEPH_MSG_DATA_BIO);
884 bio_vec = bio_iter_iovec(bio, cursor->bvec_iter);
886 /* Advance the cursor offset */
888 BUG_ON(cursor->resid < bytes);
889 cursor->resid -= bytes;
891 bio_advance_iter(bio, &cursor->bvec_iter, bytes);
893 if (bytes < bio_vec.bv_len)
894 return false; /* more bytes to process in this segment */
896 /* Move on to the next segment, and possibly the next bio */
898 if (!cursor->bvec_iter.bi_size) {
902 cursor->bvec_iter = bio->bi_iter;
904 memset(&cursor->bvec_iter, 0,
905 sizeof(cursor->bvec_iter));
908 if (!cursor->last_piece) {
909 BUG_ON(!cursor->resid);
911 /* A short read is OK, so use <= rather than == */
912 if (cursor->resid <= bio_iter_len(bio, cursor->bvec_iter))
913 cursor->last_piece = true;
918 #endif /* CONFIG_BLOCK */
921 * For a page array, a piece comes from the first page in the array
922 * that has not already been fully consumed.
924 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
927 struct ceph_msg_data *data = cursor->data;
930 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
932 BUG_ON(!data->pages);
933 BUG_ON(!data->length);
935 cursor->resid = min(length, data->length);
936 page_count = calc_pages_for(data->alignment, (u64)data->length);
937 cursor->page_offset = data->alignment & ~PAGE_MASK;
938 cursor->page_index = 0;
939 BUG_ON(page_count > (int)USHRT_MAX);
940 cursor->page_count = (unsigned short)page_count;
941 BUG_ON(length > SIZE_MAX - cursor->page_offset);
942 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
946 ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
947 size_t *page_offset, size_t *length)
949 struct ceph_msg_data *data = cursor->data;
951 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
953 BUG_ON(cursor->page_index >= cursor->page_count);
954 BUG_ON(cursor->page_offset >= PAGE_SIZE);
956 *page_offset = cursor->page_offset;
957 if (cursor->last_piece)
958 *length = cursor->resid;
960 *length = PAGE_SIZE - *page_offset;
962 return data->pages[cursor->page_index];
965 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
968 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
970 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
972 /* Advance the cursor page offset */
974 cursor->resid -= bytes;
975 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
976 if (!bytes || cursor->page_offset)
977 return false; /* more bytes to process in the current page */
980 return false; /* no more data */
982 /* Move on to the next page; offset is already at 0 */
984 BUG_ON(cursor->page_index >= cursor->page_count);
985 cursor->page_index++;
986 cursor->last_piece = cursor->resid <= PAGE_SIZE;
992 * For a pagelist, a piece is whatever remains to be consumed in the
993 * first page in the list, or the front of the next page.
996 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
999 struct ceph_msg_data *data = cursor->data;
1000 struct ceph_pagelist *pagelist;
1003 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1005 pagelist = data->pagelist;
1009 return; /* pagelist can be assigned but empty */
1011 BUG_ON(list_empty(&pagelist->head));
1012 page = list_first_entry(&pagelist->head, struct page, lru);
1014 cursor->resid = min(length, pagelist->length);
1015 cursor->page = page;
1017 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1020 static struct page *
1021 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
1022 size_t *page_offset, size_t *length)
1024 struct ceph_msg_data *data = cursor->data;
1025 struct ceph_pagelist *pagelist;
1027 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1029 pagelist = data->pagelist;
1032 BUG_ON(!cursor->page);
1033 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1035 /* offset of first page in pagelist is always 0 */
1036 *page_offset = cursor->offset & ~PAGE_MASK;
1037 if (cursor->last_piece)
1038 *length = cursor->resid;
1040 *length = PAGE_SIZE - *page_offset;
1042 return cursor->page;
1045 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
1048 struct ceph_msg_data *data = cursor->data;
1049 struct ceph_pagelist *pagelist;
1051 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1053 pagelist = data->pagelist;
1056 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1057 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1059 /* Advance the cursor offset */
1061 cursor->resid -= bytes;
1062 cursor->offset += bytes;
1063 /* offset of first page in pagelist is always 0 */
1064 if (!bytes || cursor->offset & ~PAGE_MASK)
1065 return false; /* more bytes to process in the current page */
1068 return false; /* no more data */
1070 /* Move on to the next page */
1072 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1073 cursor->page = list_next_entry(cursor->page, lru);
1074 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1080 * Message data is handled (sent or received) in pieces, where each
1081 * piece resides on a single page. The network layer might not
1082 * consume an entire piece at once. A data item's cursor keeps
1083 * track of which piece is next to process and how much remains to
1084 * be processed in that piece. It also tracks whether the current
1085 * piece is the last one in the data item.
1087 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1089 size_t length = cursor->total_resid;
1091 switch (cursor->data->type) {
1092 case CEPH_MSG_DATA_PAGELIST:
1093 ceph_msg_data_pagelist_cursor_init(cursor, length);
1095 case CEPH_MSG_DATA_PAGES:
1096 ceph_msg_data_pages_cursor_init(cursor, length);
1099 case CEPH_MSG_DATA_BIO:
1100 ceph_msg_data_bio_cursor_init(cursor, length);
1102 #endif /* CONFIG_BLOCK */
1103 case CEPH_MSG_DATA_NONE:
1108 cursor->need_crc = true;
1111 static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1113 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1114 struct ceph_msg_data *data;
1117 BUG_ON(length > msg->data_length);
1118 BUG_ON(list_empty(&msg->data));
1120 cursor->data_head = &msg->data;
1121 cursor->total_resid = length;
1122 data = list_first_entry(&msg->data, struct ceph_msg_data, links);
1123 cursor->data = data;
1125 __ceph_msg_data_cursor_init(cursor);
1129 * Return the page containing the next piece to process for a given
1130 * data item, and supply the page offset and length of that piece.
1131 * Indicate whether this is the last piece in this data item.
1133 static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1134 size_t *page_offset, size_t *length,
1139 switch (cursor->data->type) {
1140 case CEPH_MSG_DATA_PAGELIST:
1141 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1143 case CEPH_MSG_DATA_PAGES:
1144 page = ceph_msg_data_pages_next(cursor, page_offset, length);
1147 case CEPH_MSG_DATA_BIO:
1148 page = ceph_msg_data_bio_next(cursor, page_offset, length);
1150 #endif /* CONFIG_BLOCK */
1151 case CEPH_MSG_DATA_NONE:
1157 BUG_ON(*page_offset + *length > PAGE_SIZE);
1160 *last_piece = cursor->last_piece;
1166 * Returns true if the result moves the cursor on to the next piece
1169 static bool ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1174 BUG_ON(bytes > cursor->resid);
1175 switch (cursor->data->type) {
1176 case CEPH_MSG_DATA_PAGELIST:
1177 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1179 case CEPH_MSG_DATA_PAGES:
1180 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1183 case CEPH_MSG_DATA_BIO:
1184 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1186 #endif /* CONFIG_BLOCK */
1187 case CEPH_MSG_DATA_NONE:
1192 cursor->total_resid -= bytes;
1194 if (!cursor->resid && cursor->total_resid) {
1195 WARN_ON(!cursor->last_piece);
1196 BUG_ON(list_is_last(&cursor->data->links, cursor->data_head));
1197 cursor->data = list_next_entry(cursor->data, links);
1198 __ceph_msg_data_cursor_init(cursor);
1201 cursor->need_crc = new_piece;
1206 static size_t sizeof_footer(struct ceph_connection *con)
1208 return (con->peer_features & CEPH_FEATURE_MSG_AUTH) ?
1209 sizeof(struct ceph_msg_footer) :
1210 sizeof(struct ceph_msg_footer_old);
1213 static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
1218 /* Initialize data cursor */
1220 ceph_msg_data_cursor_init(msg, (size_t)data_len);
1224 * Prepare footer for currently outgoing message, and finish things
1225 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1227 static void prepare_write_message_footer(struct ceph_connection *con)
1229 struct ceph_msg *m = con->out_msg;
1231 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1233 dout("prepare_write_message_footer %p\n", con);
1234 con_out_kvec_add(con, sizeof_footer(con), &m->footer);
1235 if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
1236 if (con->ops->sign_message)
1237 con->ops->sign_message(m);
1241 m->old_footer.flags = m->footer.flags;
1243 con->out_more = m->more_to_follow;
1244 con->out_msg_done = true;
1248 * Prepare headers for the next outgoing message.
1250 static void prepare_write_message(struct ceph_connection *con)
1255 con_out_kvec_reset(con);
1256 con->out_msg_done = false;
1258 /* Sneak an ack in there first? If we can get it into the same
1259 * TCP packet that's a good thing. */
1260 if (con->in_seq > con->in_seq_acked) {
1261 con->in_seq_acked = con->in_seq;
1262 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1263 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1264 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1265 &con->out_temp_ack);
1268 BUG_ON(list_empty(&con->out_queue));
1269 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1271 BUG_ON(m->con != con);
1273 /* put message on sent list */
1275 list_move_tail(&m->list_head, &con->out_sent);
1278 * only assign outgoing seq # if we haven't sent this message
1279 * yet. if it is requeued, resend with it's original seq.
1281 if (m->needs_out_seq) {
1282 m->hdr.seq = cpu_to_le64(++con->out_seq);
1283 m->needs_out_seq = false;
1285 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
1287 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1288 m, con->out_seq, le16_to_cpu(m->hdr.type),
1289 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1291 BUG_ON(le32_to_cpu(m->hdr.front_len) != m->front.iov_len);
1293 /* tag + hdr + front + middle */
1294 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1295 con_out_kvec_add(con, sizeof(con->out_hdr), &con->out_hdr);
1296 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1299 con_out_kvec_add(con, m->middle->vec.iov_len,
1300 m->middle->vec.iov_base);
1302 /* fill in hdr crc and finalize hdr */
1303 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1304 con->out_msg->hdr.crc = cpu_to_le32(crc);
1305 memcpy(&con->out_hdr, &con->out_msg->hdr, sizeof(con->out_hdr));
1307 /* fill in front and middle crc, footer */
1308 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1309 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1311 crc = crc32c(0, m->middle->vec.iov_base,
1312 m->middle->vec.iov_len);
1313 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1315 con->out_msg->footer.middle_crc = 0;
1316 dout("%s front_crc %u middle_crc %u\n", __func__,
1317 le32_to_cpu(con->out_msg->footer.front_crc),
1318 le32_to_cpu(con->out_msg->footer.middle_crc));
1319 con->out_msg->footer.flags = 0;
1321 /* is there a data payload? */
1322 con->out_msg->footer.data_crc = 0;
1323 if (m->data_length) {
1324 prepare_message_data(con->out_msg, m->data_length);
1325 con->out_more = 1; /* data + footer will follow */
1327 /* no, queue up footer too and be done */
1328 prepare_write_message_footer(con);
1331 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1337 static void prepare_write_ack(struct ceph_connection *con)
1339 dout("prepare_write_ack %p %llu -> %llu\n", con,
1340 con->in_seq_acked, con->in_seq);
1341 con->in_seq_acked = con->in_seq;
1343 con_out_kvec_reset(con);
1345 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1347 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1348 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1349 &con->out_temp_ack);
1351 con->out_more = 1; /* more will follow.. eventually.. */
1352 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1356 * Prepare to share the seq during handshake
1358 static void prepare_write_seq(struct ceph_connection *con)
1360 dout("prepare_write_seq %p %llu -> %llu\n", con,
1361 con->in_seq_acked, con->in_seq);
1362 con->in_seq_acked = con->in_seq;
1364 con_out_kvec_reset(con);
1366 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1367 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1368 &con->out_temp_ack);
1370 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1374 * Prepare to write keepalive byte.
1376 static void prepare_write_keepalive(struct ceph_connection *con)
1378 dout("prepare_write_keepalive %p\n", con);
1379 con_out_kvec_reset(con);
1380 if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
1381 struct timespec now = CURRENT_TIME;
1383 con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
1384 ceph_encode_timespec(&con->out_temp_keepalive2, &now);
1385 con_out_kvec_add(con, sizeof(con->out_temp_keepalive2),
1386 &con->out_temp_keepalive2);
1388 con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive);
1390 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1394 * Connection negotiation.
1397 static int get_connect_authorizer(struct ceph_connection *con)
1399 struct ceph_auth_handshake *auth;
1402 if (!con->ops->get_authorizer) {
1404 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1405 con->out_connect.authorizer_len = 0;
1409 auth = con->ops->get_authorizer(con, &auth_proto, con->auth_retry);
1411 return PTR_ERR(auth);
1414 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1415 con->out_connect.authorizer_len = cpu_to_le32(auth->authorizer_buf_len);
1420 * We connected to a peer and are saying hello.
1422 static void prepare_write_banner(struct ceph_connection *con)
1424 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1425 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1426 &con->msgr->my_enc_addr);
1429 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1432 static void __prepare_write_connect(struct ceph_connection *con)
1434 con_out_kvec_add(con, sizeof(con->out_connect), &con->out_connect);
1436 con_out_kvec_add(con, con->auth->authorizer_buf_len,
1437 con->auth->authorizer_buf);
1440 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1443 static int prepare_write_connect(struct ceph_connection *con)
1445 unsigned int global_seq = get_global_seq(con->msgr, 0);
1449 switch (con->peer_name.type) {
1450 case CEPH_ENTITY_TYPE_MON:
1451 proto = CEPH_MONC_PROTOCOL;
1453 case CEPH_ENTITY_TYPE_OSD:
1454 proto = CEPH_OSDC_PROTOCOL;
1456 case CEPH_ENTITY_TYPE_MDS:
1457 proto = CEPH_MDSC_PROTOCOL;
1463 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1464 con->connect_seq, global_seq, proto);
1466 con->out_connect.features =
1467 cpu_to_le64(from_msgr(con->msgr)->supported_features);
1468 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1469 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1470 con->out_connect.global_seq = cpu_to_le32(global_seq);
1471 con->out_connect.protocol_version = cpu_to_le32(proto);
1472 con->out_connect.flags = 0;
1474 ret = get_connect_authorizer(con);
1478 __prepare_write_connect(con);
1483 * write as much of pending kvecs to the socket as we can.
1485 * 0 -> socket full, but more to do
1488 static int write_partial_kvec(struct ceph_connection *con)
1492 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1493 while (con->out_kvec_bytes > 0) {
1494 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1495 con->out_kvec_left, con->out_kvec_bytes,
1499 con->out_kvec_bytes -= ret;
1500 if (con->out_kvec_bytes == 0)
1503 /* account for full iov entries consumed */
1504 while (ret >= con->out_kvec_cur->iov_len) {
1505 BUG_ON(!con->out_kvec_left);
1506 ret -= con->out_kvec_cur->iov_len;
1507 con->out_kvec_cur++;
1508 con->out_kvec_left--;
1510 /* and for a partially-consumed entry */
1512 con->out_kvec_cur->iov_len -= ret;
1513 con->out_kvec_cur->iov_base += ret;
1516 con->out_kvec_left = 0;
1519 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1520 con->out_kvec_bytes, con->out_kvec_left, ret);
1521 return ret; /* done! */
1524 static u32 ceph_crc32c_page(u32 crc, struct page *page,
1525 unsigned int page_offset,
1526 unsigned int length)
1531 BUG_ON(kaddr == NULL);
1532 crc = crc32c(crc, kaddr + page_offset, length);
1538 * Write as much message data payload as we can. If we finish, queue
1540 * 1 -> done, footer is now queued in out_kvec[].
1541 * 0 -> socket full, but more to do
1544 static int write_partial_message_data(struct ceph_connection *con)
1546 struct ceph_msg *msg = con->out_msg;
1547 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1548 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
1551 dout("%s %p msg %p\n", __func__, con, msg);
1553 if (list_empty(&msg->data))
1557 * Iterate through each page that contains data to be
1558 * written, and send as much as possible for each.
1560 * If we are calculating the data crc (the default), we will
1561 * need to map the page. If we have no pages, they have
1562 * been revoked, so use the zero page.
1564 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1565 while (cursor->resid) {
1573 page = ceph_msg_data_next(cursor, &page_offset, &length,
1575 ret = ceph_tcp_sendpage(con->sock, page, page_offset,
1576 length, !last_piece);
1579 msg->footer.data_crc = cpu_to_le32(crc);
1583 if (do_datacrc && cursor->need_crc)
1584 crc = ceph_crc32c_page(crc, page, page_offset, length);
1585 need_crc = ceph_msg_data_advance(cursor, (size_t)ret);
1588 dout("%s %p msg %p done\n", __func__, con, msg);
1590 /* prepare and queue up footer, too */
1592 msg->footer.data_crc = cpu_to_le32(crc);
1594 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1595 con_out_kvec_reset(con);
1596 prepare_write_message_footer(con);
1598 return 1; /* must return > 0 to indicate success */
1604 static int write_partial_skip(struct ceph_connection *con)
1608 dout("%s %p %d left\n", __func__, con, con->out_skip);
1609 while (con->out_skip > 0) {
1610 size_t size = min(con->out_skip, (int) PAGE_SIZE);
1612 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, true);
1615 con->out_skip -= ret;
1623 * Prepare to read connection handshake, or an ack.
1625 static void prepare_read_banner(struct ceph_connection *con)
1627 dout("prepare_read_banner %p\n", con);
1628 con->in_base_pos = 0;
1631 static void prepare_read_connect(struct ceph_connection *con)
1633 dout("prepare_read_connect %p\n", con);
1634 con->in_base_pos = 0;
1637 static void prepare_read_ack(struct ceph_connection *con)
1639 dout("prepare_read_ack %p\n", con);
1640 con->in_base_pos = 0;
1643 static void prepare_read_seq(struct ceph_connection *con)
1645 dout("prepare_read_seq %p\n", con);
1646 con->in_base_pos = 0;
1647 con->in_tag = CEPH_MSGR_TAG_SEQ;
1650 static void prepare_read_tag(struct ceph_connection *con)
1652 dout("prepare_read_tag %p\n", con);
1653 con->in_base_pos = 0;
1654 con->in_tag = CEPH_MSGR_TAG_READY;
1657 static void prepare_read_keepalive_ack(struct ceph_connection *con)
1659 dout("prepare_read_keepalive_ack %p\n", con);
1660 con->in_base_pos = 0;
1664 * Prepare to read a message.
1666 static int prepare_read_message(struct ceph_connection *con)
1668 dout("prepare_read_message %p\n", con);
1669 BUG_ON(con->in_msg != NULL);
1670 con->in_base_pos = 0;
1671 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1676 static int read_partial(struct ceph_connection *con,
1677 int end, int size, void *object)
1679 while (con->in_base_pos < end) {
1680 int left = end - con->in_base_pos;
1681 int have = size - left;
1682 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1685 con->in_base_pos += ret;
1692 * Read all or part of the connect-side handshake on a new connection
1694 static int read_partial_banner(struct ceph_connection *con)
1700 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1703 size = strlen(CEPH_BANNER);
1705 ret = read_partial(con, end, size, con->in_banner);
1709 size = sizeof (con->actual_peer_addr);
1711 ret = read_partial(con, end, size, &con->actual_peer_addr);
1715 size = sizeof (con->peer_addr_for_me);
1717 ret = read_partial(con, end, size, &con->peer_addr_for_me);
1725 static int read_partial_connect(struct ceph_connection *con)
1731 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1733 size = sizeof (con->in_reply);
1735 ret = read_partial(con, end, size, &con->in_reply);
1740 size = le32_to_cpu(con->in_reply.authorizer_len);
1741 if (size > con->auth->authorizer_reply_buf_len) {
1742 pr_err("authorizer reply too big: %d > %zu\n", size,
1743 con->auth->authorizer_reply_buf_len);
1749 ret = read_partial(con, end, size,
1750 con->auth->authorizer_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));
1764 * Verify the hello banner looks okay.
1766 static int verify_hello(struct ceph_connection *con)
1768 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1769 pr_err("connect to %s got bad banner\n",
1770 ceph_pr_addr(&con->peer_addr.in_addr));
1771 con->error_msg = "protocol error, bad banner";
1777 static bool addr_is_blank(struct sockaddr_storage *ss)
1779 struct in_addr *addr = &((struct sockaddr_in *)ss)->sin_addr;
1780 struct in6_addr *addr6 = &((struct sockaddr_in6 *)ss)->sin6_addr;
1782 switch (ss->ss_family) {
1784 return addr->s_addr == htonl(INADDR_ANY);
1786 return ipv6_addr_any(addr6);
1792 static int addr_port(struct sockaddr_storage *ss)
1794 switch (ss->ss_family) {
1796 return ntohs(((struct sockaddr_in *)ss)->sin_port);
1798 return ntohs(((struct sockaddr_in6 *)ss)->sin6_port);
1803 static void addr_set_port(struct sockaddr_storage *ss, int p)
1805 switch (ss->ss_family) {
1807 ((struct sockaddr_in *)ss)->sin_port = htons(p);
1810 ((struct sockaddr_in6 *)ss)->sin6_port = htons(p);
1816 * Unlike other *_pton function semantics, zero indicates success.
1818 static int ceph_pton(const char *str, size_t len, struct sockaddr_storage *ss,
1819 char delim, const char **ipend)
1821 struct sockaddr_in *in4 = (struct sockaddr_in *) ss;
1822 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *) ss;
1824 memset(ss, 0, sizeof(*ss));
1826 if (in4_pton(str, len, (u8 *)&in4->sin_addr.s_addr, delim, ipend)) {
1827 ss->ss_family = AF_INET;
1831 if (in6_pton(str, len, (u8 *)&in6->sin6_addr.s6_addr, delim, ipend)) {
1832 ss->ss_family = AF_INET6;
1840 * Extract hostname string and resolve using kernel DNS facility.
1842 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1843 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1844 struct sockaddr_storage *ss, char delim, const char **ipend)
1846 const char *end, *delim_p;
1847 char *colon_p, *ip_addr = NULL;
1851 * The end of the hostname occurs immediately preceding the delimiter or
1852 * the port marker (':') where the delimiter takes precedence.
1854 delim_p = memchr(name, delim, namelen);
1855 colon_p = memchr(name, ':', namelen);
1857 if (delim_p && colon_p)
1858 end = delim_p < colon_p ? delim_p : colon_p;
1859 else if (!delim_p && colon_p)
1863 if (!end) /* case: hostname:/ */
1864 end = name + namelen;
1870 /* do dns_resolve upcall */
1871 ip_len = dns_query(NULL, name, end - name, NULL, &ip_addr, NULL);
1873 ret = ceph_pton(ip_addr, ip_len, ss, -1, NULL);
1881 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1882 ret, ret ? "failed" : ceph_pr_addr(ss));
1887 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1888 struct sockaddr_storage *ss, char delim, const char **ipend)
1895 * Parse a server name (IP or hostname). If a valid IP address is not found
1896 * then try to extract a hostname to resolve using userspace DNS upcall.
1898 static int ceph_parse_server_name(const char *name, size_t namelen,
1899 struct sockaddr_storage *ss, char delim, const char **ipend)
1903 ret = ceph_pton(name, namelen, ss, delim, ipend);
1905 ret = ceph_dns_resolve_name(name, namelen, ss, delim, ipend);
1911 * Parse an ip[:port] list into an addr array. Use the default
1912 * monitor port if a port isn't specified.
1914 int ceph_parse_ips(const char *c, const char *end,
1915 struct ceph_entity_addr *addr,
1916 int max_count, int *count)
1918 int i, ret = -EINVAL;
1921 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1922 for (i = 0; i < max_count; i++) {
1924 struct sockaddr_storage *ss = &addr[i].in_addr;
1933 ret = ceph_parse_server_name(p, end - p, ss, delim, &ipend);
1942 dout("missing matching ']'\n");
1949 if (p < end && *p == ':') {
1952 while (p < end && *p >= '0' && *p <= '9') {
1953 port = (port * 10) + (*p - '0');
1957 port = CEPH_MON_PORT;
1958 else if (port > 65535)
1961 port = CEPH_MON_PORT;
1964 addr_set_port(ss, port);
1966 dout("parse_ips got %s\n", ceph_pr_addr(ss));
1983 pr_err("parse_ips bad ip '%.*s'\n", (int)(end - c), c);
1986 EXPORT_SYMBOL(ceph_parse_ips);
1988 static int process_banner(struct ceph_connection *con)
1990 dout("process_banner on %p\n", con);
1992 if (verify_hello(con) < 0)
1995 ceph_decode_addr(&con->actual_peer_addr);
1996 ceph_decode_addr(&con->peer_addr_for_me);
1999 * Make sure the other end is who we wanted. note that the other
2000 * end may not yet know their ip address, so if it's 0.0.0.0, give
2001 * them the benefit of the doubt.
2003 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
2004 sizeof(con->peer_addr)) != 0 &&
2005 !(addr_is_blank(&con->actual_peer_addr.in_addr) &&
2006 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
2007 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
2008 ceph_pr_addr(&con->peer_addr.in_addr),
2009 (int)le32_to_cpu(con->peer_addr.nonce),
2010 ceph_pr_addr(&con->actual_peer_addr.in_addr),
2011 (int)le32_to_cpu(con->actual_peer_addr.nonce));
2012 con->error_msg = "wrong peer at address";
2017 * did we learn our address?
2019 if (addr_is_blank(&con->msgr->inst.addr.in_addr)) {
2020 int port = addr_port(&con->msgr->inst.addr.in_addr);
2022 memcpy(&con->msgr->inst.addr.in_addr,
2023 &con->peer_addr_for_me.in_addr,
2024 sizeof(con->peer_addr_for_me.in_addr));
2025 addr_set_port(&con->msgr->inst.addr.in_addr, port);
2026 encode_my_addr(con->msgr);
2027 dout("process_banner learned my addr is %s\n",
2028 ceph_pr_addr(&con->msgr->inst.addr.in_addr));
2034 static int process_connect(struct ceph_connection *con)
2036 u64 sup_feat = from_msgr(con->msgr)->supported_features;
2037 u64 req_feat = from_msgr(con->msgr)->required_features;
2038 u64 server_feat = ceph_sanitize_features(
2039 le64_to_cpu(con->in_reply.features));
2042 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
2045 int len = le32_to_cpu(con->in_reply.authorizer_len);
2048 * Any connection that defines ->get_authorizer()
2049 * should also define ->add_authorizer_challenge() and
2050 * ->verify_authorizer_reply().
2052 * See get_connect_authorizer().
2054 if (con->in_reply.tag == CEPH_MSGR_TAG_CHALLENGE_AUTHORIZER) {
2055 ret = con->ops->add_authorizer_challenge(
2056 con, con->auth->authorizer_reply_buf, len);
2060 con_out_kvec_reset(con);
2061 __prepare_write_connect(con);
2062 prepare_read_connect(con);
2067 ret = con->ops->verify_authorizer_reply(con);
2069 con->error_msg = "bad authorize reply";
2075 switch (con->in_reply.tag) {
2076 case CEPH_MSGR_TAG_FEATURES:
2077 pr_err("%s%lld %s feature set mismatch,"
2078 " my %llx < server's %llx, missing %llx\n",
2079 ENTITY_NAME(con->peer_name),
2080 ceph_pr_addr(&con->peer_addr.in_addr),
2081 sup_feat, server_feat, server_feat & ~sup_feat);
2082 con->error_msg = "missing required protocol features";
2083 reset_connection(con);
2086 case CEPH_MSGR_TAG_BADPROTOVER:
2087 pr_err("%s%lld %s protocol version mismatch,"
2088 " my %d != server's %d\n",
2089 ENTITY_NAME(con->peer_name),
2090 ceph_pr_addr(&con->peer_addr.in_addr),
2091 le32_to_cpu(con->out_connect.protocol_version),
2092 le32_to_cpu(con->in_reply.protocol_version));
2093 con->error_msg = "protocol version mismatch";
2094 reset_connection(con);
2097 case CEPH_MSGR_TAG_BADAUTHORIZER:
2099 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2101 if (con->auth_retry == 2) {
2102 con->error_msg = "connect authorization failure";
2105 con_out_kvec_reset(con);
2106 ret = prepare_write_connect(con);
2109 prepare_read_connect(con);
2112 case CEPH_MSGR_TAG_RESETSESSION:
2114 * If we connected with a large connect_seq but the peer
2115 * has no record of a session with us (no connection, or
2116 * connect_seq == 0), they will send RESETSESION to indicate
2117 * that they must have reset their session, and may have
2120 dout("process_connect got RESET peer seq %u\n",
2121 le32_to_cpu(con->in_reply.connect_seq));
2122 pr_err("%s%lld %s connection reset\n",
2123 ENTITY_NAME(con->peer_name),
2124 ceph_pr_addr(&con->peer_addr.in_addr));
2125 reset_connection(con);
2126 con_out_kvec_reset(con);
2127 ret = prepare_write_connect(con);
2130 prepare_read_connect(con);
2132 /* Tell ceph about it. */
2133 mutex_unlock(&con->mutex);
2134 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2135 if (con->ops->peer_reset)
2136 con->ops->peer_reset(con);
2137 mutex_lock(&con->mutex);
2138 if (con->state != CON_STATE_NEGOTIATING)
2142 case CEPH_MSGR_TAG_RETRY_SESSION:
2144 * If we sent a smaller connect_seq than the peer has, try
2145 * again with a larger value.
2147 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2148 le32_to_cpu(con->out_connect.connect_seq),
2149 le32_to_cpu(con->in_reply.connect_seq));
2150 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2151 con_out_kvec_reset(con);
2152 ret = prepare_write_connect(con);
2155 prepare_read_connect(con);
2158 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2160 * If we sent a smaller global_seq than the peer has, try
2161 * again with a larger value.
2163 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2164 con->peer_global_seq,
2165 le32_to_cpu(con->in_reply.global_seq));
2166 get_global_seq(con->msgr,
2167 le32_to_cpu(con->in_reply.global_seq));
2168 con_out_kvec_reset(con);
2169 ret = prepare_write_connect(con);
2172 prepare_read_connect(con);
2175 case CEPH_MSGR_TAG_SEQ:
2176 case CEPH_MSGR_TAG_READY:
2177 if (req_feat & ~server_feat) {
2178 pr_err("%s%lld %s protocol feature mismatch,"
2179 " my required %llx > server's %llx, need %llx\n",
2180 ENTITY_NAME(con->peer_name),
2181 ceph_pr_addr(&con->peer_addr.in_addr),
2182 req_feat, server_feat, req_feat & ~server_feat);
2183 con->error_msg = "missing required protocol features";
2184 reset_connection(con);
2188 WARN_ON(con->state != CON_STATE_NEGOTIATING);
2189 con->state = CON_STATE_OPEN;
2190 con->auth_retry = 0; /* we authenticated; clear flag */
2191 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2193 con->peer_features = server_feat;
2194 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2195 con->peer_global_seq,
2196 le32_to_cpu(con->in_reply.connect_seq),
2198 WARN_ON(con->connect_seq !=
2199 le32_to_cpu(con->in_reply.connect_seq));
2201 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2202 con_flag_set(con, CON_FLAG_LOSSYTX);
2204 con->delay = 0; /* reset backoff memory */
2206 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2207 prepare_write_seq(con);
2208 prepare_read_seq(con);
2210 prepare_read_tag(con);
2214 case CEPH_MSGR_TAG_WAIT:
2216 * If there is a connection race (we are opening
2217 * connections to each other), one of us may just have
2218 * to WAIT. This shouldn't happen if we are the
2221 con->error_msg = "protocol error, got WAIT as client";
2225 con->error_msg = "protocol error, garbage tag during connect";
2233 * read (part of) an ack
2235 static int read_partial_ack(struct ceph_connection *con)
2237 int size = sizeof (con->in_temp_ack);
2240 return read_partial(con, end, size, &con->in_temp_ack);
2244 * We can finally discard anything that's been acked.
2246 static void process_ack(struct ceph_connection *con)
2249 u64 ack = le64_to_cpu(con->in_temp_ack);
2252 while (!list_empty(&con->out_sent)) {
2253 m = list_first_entry(&con->out_sent, struct ceph_msg,
2255 seq = le64_to_cpu(m->hdr.seq);
2258 dout("got ack for seq %llu type %d at %p\n", seq,
2259 le16_to_cpu(m->hdr.type), m);
2260 m->ack_stamp = jiffies;
2263 prepare_read_tag(con);
2267 static int read_partial_message_section(struct ceph_connection *con,
2268 struct kvec *section,
2269 unsigned int sec_len, u32 *crc)
2275 while (section->iov_len < sec_len) {
2276 BUG_ON(section->iov_base == NULL);
2277 left = sec_len - section->iov_len;
2278 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2279 section->iov_len, left);
2282 section->iov_len += ret;
2284 if (section->iov_len == sec_len)
2285 *crc = crc32c(0, section->iov_base, section->iov_len);
2290 static int read_partial_msg_data(struct ceph_connection *con)
2292 struct ceph_msg *msg = con->in_msg;
2293 struct ceph_msg_data_cursor *cursor = &msg->cursor;
2294 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2302 if (list_empty(&msg->data))
2306 crc = con->in_data_crc;
2307 while (cursor->resid) {
2308 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
2309 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2312 con->in_data_crc = crc;
2318 crc = ceph_crc32c_page(crc, page, page_offset, ret);
2319 (void) ceph_msg_data_advance(cursor, (size_t)ret);
2322 con->in_data_crc = crc;
2324 return 1; /* must return > 0 to indicate success */
2328 * read (part of) a message.
2330 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2332 static int read_partial_message(struct ceph_connection *con)
2334 struct ceph_msg *m = con->in_msg;
2338 unsigned int front_len, middle_len, data_len;
2339 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2340 bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
2344 dout("read_partial_message con %p msg %p\n", con, m);
2347 size = sizeof (con->in_hdr);
2349 ret = read_partial(con, end, size, &con->in_hdr);
2353 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2354 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2355 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
2356 crc, con->in_hdr.crc);
2360 front_len = le32_to_cpu(con->in_hdr.front_len);
2361 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2363 middle_len = le32_to_cpu(con->in_hdr.middle_len);
2364 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2366 data_len = le32_to_cpu(con->in_hdr.data_len);
2367 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2371 seq = le64_to_cpu(con->in_hdr.seq);
2372 if ((s64)seq - (s64)con->in_seq < 1) {
2373 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2374 ENTITY_NAME(con->peer_name),
2375 ceph_pr_addr(&con->peer_addr.in_addr),
2376 seq, con->in_seq + 1);
2377 con->in_base_pos = -front_len - middle_len - data_len -
2379 con->in_tag = CEPH_MSGR_TAG_READY;
2381 } else if ((s64)seq - (s64)con->in_seq > 1) {
2382 pr_err("read_partial_message bad seq %lld expected %lld\n",
2383 seq, con->in_seq + 1);
2384 con->error_msg = "bad message sequence # for incoming message";
2388 /* allocate message? */
2392 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2393 front_len, data_len);
2394 ret = ceph_con_in_msg_alloc(con, &skip);
2398 BUG_ON(!con->in_msg ^ skip);
2400 /* skip this message */
2401 dout("alloc_msg said skip message\n");
2402 con->in_base_pos = -front_len - middle_len - data_len -
2404 con->in_tag = CEPH_MSGR_TAG_READY;
2409 BUG_ON(!con->in_msg);
2410 BUG_ON(con->in_msg->con != con);
2412 m->front.iov_len = 0; /* haven't read it yet */
2414 m->middle->vec.iov_len = 0;
2416 /* prepare for data payload, if any */
2419 prepare_message_data(con->in_msg, data_len);
2423 ret = read_partial_message_section(con, &m->front, front_len,
2424 &con->in_front_crc);
2430 ret = read_partial_message_section(con, &m->middle->vec,
2432 &con->in_middle_crc);
2439 ret = read_partial_msg_data(con);
2445 size = sizeof_footer(con);
2447 ret = read_partial(con, end, size, &m->footer);
2452 m->footer.flags = m->old_footer.flags;
2456 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2457 m, front_len, m->footer.front_crc, middle_len,
2458 m->footer.middle_crc, data_len, m->footer.data_crc);
2461 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2462 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2463 m, con->in_front_crc, m->footer.front_crc);
2466 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2467 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2468 m, con->in_middle_crc, m->footer.middle_crc);
2472 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2473 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2474 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2475 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2479 if (need_sign && con->ops->check_message_signature &&
2480 con->ops->check_message_signature(m)) {
2481 pr_err("read_partial_message %p signature check failed\n", m);
2485 return 1; /* done! */
2489 * Process message. This happens in the worker thread. The callback should
2490 * be careful not to do anything that waits on other incoming messages or it
2493 static void process_message(struct ceph_connection *con)
2495 struct ceph_msg *msg = con->in_msg;
2497 BUG_ON(con->in_msg->con != con);
2500 /* if first message, set peer_name */
2501 if (con->peer_name.type == 0)
2502 con->peer_name = msg->hdr.src;
2505 mutex_unlock(&con->mutex);
2507 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2508 msg, le64_to_cpu(msg->hdr.seq),
2509 ENTITY_NAME(msg->hdr.src),
2510 le16_to_cpu(msg->hdr.type),
2511 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2512 le32_to_cpu(msg->hdr.front_len),
2513 le32_to_cpu(msg->hdr.data_len),
2514 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2515 con->ops->dispatch(con, msg);
2517 mutex_lock(&con->mutex);
2520 static int read_keepalive_ack(struct ceph_connection *con)
2522 struct ceph_timespec ceph_ts;
2523 size_t size = sizeof(ceph_ts);
2524 int ret = read_partial(con, size, size, &ceph_ts);
2527 ceph_decode_timespec(&con->last_keepalive_ack, &ceph_ts);
2528 prepare_read_tag(con);
2533 * Write something to the socket. Called in a worker thread when the
2534 * socket appears to be writeable and we have something ready to send.
2536 static int try_write(struct ceph_connection *con)
2540 dout("try_write start %p state %lu\n", con, con->state);
2541 if (con->state != CON_STATE_PREOPEN &&
2542 con->state != CON_STATE_CONNECTING &&
2543 con->state != CON_STATE_NEGOTIATING &&
2544 con->state != CON_STATE_OPEN)
2548 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2550 /* open the socket first? */
2551 if (con->state == CON_STATE_PREOPEN) {
2553 con->state = CON_STATE_CONNECTING;
2555 con_out_kvec_reset(con);
2556 prepare_write_banner(con);
2557 prepare_read_banner(con);
2559 BUG_ON(con->in_msg);
2560 con->in_tag = CEPH_MSGR_TAG_READY;
2561 dout("try_write initiating connect on %p new state %lu\n",
2563 ret = ceph_tcp_connect(con);
2565 con->error_msg = "connect error";
2573 /* kvec data queued? */
2574 if (con->out_kvec_left) {
2575 ret = write_partial_kvec(con);
2579 if (con->out_skip) {
2580 ret = write_partial_skip(con);
2587 if (con->out_msg_done) {
2588 ceph_msg_put(con->out_msg);
2589 con->out_msg = NULL; /* we're done with this one */
2593 ret = write_partial_message_data(con);
2595 goto more_kvec; /* we need to send the footer, too! */
2599 dout("try_write write_partial_message_data err %d\n",
2606 if (con->state == CON_STATE_OPEN) {
2607 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2608 prepare_write_keepalive(con);
2611 /* is anything else pending? */
2612 if (!list_empty(&con->out_queue)) {
2613 prepare_write_message(con);
2616 if (con->in_seq > con->in_seq_acked) {
2617 prepare_write_ack(con);
2622 /* Nothing to do! */
2623 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2624 dout("try_write nothing else to write.\n");
2627 dout("try_write done on %p ret %d\n", con, ret);
2634 * Read what we can from the socket.
2636 static int try_read(struct ceph_connection *con)
2641 dout("try_read start on %p state %lu\n", con, con->state);
2642 if (con->state != CON_STATE_CONNECTING &&
2643 con->state != CON_STATE_NEGOTIATING &&
2644 con->state != CON_STATE_OPEN)
2649 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2652 if (con->state == CON_STATE_CONNECTING) {
2653 dout("try_read connecting\n");
2654 ret = read_partial_banner(con);
2657 ret = process_banner(con);
2661 con->state = CON_STATE_NEGOTIATING;
2664 * Received banner is good, exchange connection info.
2665 * Do not reset out_kvec, as sending our banner raced
2666 * with receiving peer banner after connect completed.
2668 ret = prepare_write_connect(con);
2671 prepare_read_connect(con);
2673 /* Send connection info before awaiting response */
2677 if (con->state == CON_STATE_NEGOTIATING) {
2678 dout("try_read negotiating\n");
2679 ret = read_partial_connect(con);
2682 ret = process_connect(con);
2688 WARN_ON(con->state != CON_STATE_OPEN);
2690 if (con->in_base_pos < 0) {
2692 * skipping + discarding content.
2694 * FIXME: there must be a better way to do this!
2696 static char buf[SKIP_BUF_SIZE];
2697 int skip = min((int) sizeof (buf), -con->in_base_pos);
2699 dout("skipping %d / %d bytes\n", skip, -con->in_base_pos);
2700 ret = ceph_tcp_recvmsg(con->sock, buf, skip);
2703 con->in_base_pos += ret;
2704 if (con->in_base_pos)
2707 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2711 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2714 dout("try_read got tag %d\n", (int)con->in_tag);
2715 switch (con->in_tag) {
2716 case CEPH_MSGR_TAG_MSG:
2717 prepare_read_message(con);
2719 case CEPH_MSGR_TAG_ACK:
2720 prepare_read_ack(con);
2722 case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
2723 prepare_read_keepalive_ack(con);
2725 case CEPH_MSGR_TAG_CLOSE:
2726 con_close_socket(con);
2727 con->state = CON_STATE_CLOSED;
2733 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2734 ret = read_partial_message(con);
2738 con->error_msg = "bad crc/signature";
2744 con->error_msg = "io error";
2749 if (con->in_tag == CEPH_MSGR_TAG_READY)
2751 process_message(con);
2752 if (con->state == CON_STATE_OPEN)
2753 prepare_read_tag(con);
2756 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2757 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2759 * the final handshake seq exchange is semantically
2760 * equivalent to an ACK
2762 ret = read_partial_ack(con);
2768 if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
2769 ret = read_keepalive_ack(con);
2776 dout("try_read done on %p ret %d\n", con, ret);
2780 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2781 con->error_msg = "protocol error, garbage tag";
2788 * Atomically queue work on a connection after the specified delay.
2789 * Bump @con reference to avoid races with connection teardown.
2790 * Returns 0 if work was queued, or an error code otherwise.
2792 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2794 if (!con->ops->get(con)) {
2795 dout("%s %p ref count 0\n", __func__, con);
2799 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2800 dout("%s %p - already queued\n", __func__, con);
2805 dout("%s %p %lu\n", __func__, con, delay);
2809 static void queue_con(struct ceph_connection *con)
2811 (void) queue_con_delay(con, 0);
2814 static void cancel_con(struct ceph_connection *con)
2816 if (cancel_delayed_work(&con->work)) {
2817 dout("%s %p\n", __func__, con);
2822 static bool con_sock_closed(struct ceph_connection *con)
2824 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2828 case CON_STATE_ ## x: \
2829 con->error_msg = "socket closed (con state " #x ")"; \
2832 switch (con->state) {
2840 pr_warn("%s con %p unrecognized state %lu\n",
2841 __func__, con, con->state);
2842 con->error_msg = "unrecognized con state";
2851 static bool con_backoff(struct ceph_connection *con)
2855 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2858 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2860 dout("%s: con %p FAILED to back off %lu\n", __func__,
2862 BUG_ON(ret == -ENOENT);
2863 con_flag_set(con, CON_FLAG_BACKOFF);
2869 /* Finish fault handling; con->mutex must *not* be held here */
2871 static void con_fault_finish(struct ceph_connection *con)
2873 dout("%s %p\n", __func__, con);
2876 * in case we faulted due to authentication, invalidate our
2877 * current tickets so that we can get new ones.
2879 if (con->auth_retry) {
2880 dout("auth_retry %d, invalidating\n", con->auth_retry);
2881 if (con->ops->invalidate_authorizer)
2882 con->ops->invalidate_authorizer(con);
2883 con->auth_retry = 0;
2886 if (con->ops->fault)
2887 con->ops->fault(con);
2891 * Do some work on a connection. Drop a connection ref when we're done.
2893 static void ceph_con_workfn(struct work_struct *work)
2895 struct ceph_connection *con = container_of(work, struct ceph_connection,
2899 mutex_lock(&con->mutex);
2903 if ((fault = con_sock_closed(con))) {
2904 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2907 if (con_backoff(con)) {
2908 dout("%s: con %p BACKOFF\n", __func__, con);
2911 if (con->state == CON_STATE_STANDBY) {
2912 dout("%s: con %p STANDBY\n", __func__, con);
2915 if (con->state == CON_STATE_CLOSED) {
2916 dout("%s: con %p CLOSED\n", __func__, con);
2920 if (con->state == CON_STATE_PREOPEN) {
2921 dout("%s: con %p PREOPEN\n", __func__, con);
2925 ret = try_read(con);
2929 if (!con->error_msg)
2930 con->error_msg = "socket error on read";
2935 ret = try_write(con);
2939 if (!con->error_msg)
2940 con->error_msg = "socket error on write";
2944 break; /* If we make it to here, we're done */
2948 mutex_unlock(&con->mutex);
2951 con_fault_finish(con);
2957 * Generic error/fault handler. A retry mechanism is used with
2958 * exponential backoff
2960 static void con_fault(struct ceph_connection *con)
2962 dout("fault %p state %lu to peer %s\n",
2963 con, con->state, ceph_pr_addr(&con->peer_addr.in_addr));
2965 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2966 ceph_pr_addr(&con->peer_addr.in_addr), con->error_msg);
2967 con->error_msg = NULL;
2969 WARN_ON(con->state != CON_STATE_CONNECTING &&
2970 con->state != CON_STATE_NEGOTIATING &&
2971 con->state != CON_STATE_OPEN);
2973 con_close_socket(con);
2975 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2976 dout("fault on LOSSYTX channel, marking CLOSED\n");
2977 con->state = CON_STATE_CLOSED;
2982 BUG_ON(con->in_msg->con != con);
2983 ceph_msg_put(con->in_msg);
2987 BUG_ON(con->out_msg->con != con);
2988 ceph_msg_put(con->out_msg);
2989 con->out_msg = NULL;
2992 /* Requeue anything that hasn't been acked */
2993 list_splice_init(&con->out_sent, &con->out_queue);
2995 /* If there are no messages queued or keepalive pending, place
2996 * the connection in a STANDBY state */
2997 if (list_empty(&con->out_queue) &&
2998 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
2999 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
3000 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
3001 con->state = CON_STATE_STANDBY;
3003 /* retry after a delay. */
3004 con->state = CON_STATE_PREOPEN;
3005 if (con->delay == 0)
3006 con->delay = BASE_DELAY_INTERVAL;
3007 else if (con->delay < MAX_DELAY_INTERVAL)
3009 con_flag_set(con, CON_FLAG_BACKOFF);
3017 * initialize a new messenger instance
3019 void ceph_messenger_init(struct ceph_messenger *msgr,
3020 struct ceph_entity_addr *myaddr)
3022 spin_lock_init(&msgr->global_seq_lock);
3025 msgr->inst.addr = *myaddr;
3027 /* select a random nonce */
3028 msgr->inst.addr.type = 0;
3029 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
3030 encode_my_addr(msgr);
3032 atomic_set(&msgr->stopping, 0);
3033 write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
3035 dout("%s %p\n", __func__, msgr);
3037 EXPORT_SYMBOL(ceph_messenger_init);
3039 void ceph_messenger_fini(struct ceph_messenger *msgr)
3041 put_net(read_pnet(&msgr->net));
3043 EXPORT_SYMBOL(ceph_messenger_fini);
3045 static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con)
3048 msg->con->ops->put(msg->con);
3050 msg->con = con ? con->ops->get(con) : NULL;
3051 BUG_ON(msg->con != con);
3054 static void clear_standby(struct ceph_connection *con)
3056 /* come back from STANDBY? */
3057 if (con->state == CON_STATE_STANDBY) {
3058 dout("clear_standby %p and ++connect_seq\n", con);
3059 con->state = CON_STATE_PREOPEN;
3061 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
3062 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
3067 * Queue up an outgoing message on the given connection.
3069 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
3072 msg->hdr.src = con->msgr->inst.name;
3073 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
3074 msg->needs_out_seq = true;
3076 mutex_lock(&con->mutex);
3078 if (con->state == CON_STATE_CLOSED) {
3079 dout("con_send %p closed, dropping %p\n", con, msg);
3081 mutex_unlock(&con->mutex);
3085 msg_con_set(msg, con);
3087 BUG_ON(!list_empty(&msg->list_head));
3088 list_add_tail(&msg->list_head, &con->out_queue);
3089 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3090 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3091 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3092 le32_to_cpu(msg->hdr.front_len),
3093 le32_to_cpu(msg->hdr.middle_len),
3094 le32_to_cpu(msg->hdr.data_len));
3097 mutex_unlock(&con->mutex);
3099 /* if there wasn't anything waiting to send before, queue
3101 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3104 EXPORT_SYMBOL(ceph_con_send);
3107 * Revoke a message that was previously queued for send
3109 void ceph_msg_revoke(struct ceph_msg *msg)
3111 struct ceph_connection *con = msg->con;
3114 dout("%s msg %p null con\n", __func__, msg);
3115 return; /* Message not in our possession */
3118 mutex_lock(&con->mutex);
3119 if (!list_empty(&msg->list_head)) {
3120 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
3121 list_del_init(&msg->list_head);
3126 if (con->out_msg == msg) {
3127 BUG_ON(con->out_skip);
3129 if (con->out_msg_done) {
3130 con->out_skip += con_out_kvec_skip(con);
3132 BUG_ON(!msg->data_length);
3133 con->out_skip += sizeof_footer(con);
3135 /* data, middle, front */
3136 if (msg->data_length)
3137 con->out_skip += msg->cursor.total_resid;
3139 con->out_skip += con_out_kvec_skip(con);
3140 con->out_skip += con_out_kvec_skip(con);
3142 dout("%s %p msg %p - was sending, will write %d skip %d\n",
3143 __func__, con, msg, con->out_kvec_bytes, con->out_skip);
3145 con->out_msg = NULL;
3149 mutex_unlock(&con->mutex);
3153 * Revoke a message that we may be reading data into
3155 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
3157 struct ceph_connection *con = msg->con;
3160 dout("%s msg %p null con\n", __func__, msg);
3161 return; /* Message not in our possession */
3164 mutex_lock(&con->mutex);
3165 if (con->in_msg == msg) {
3166 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3167 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3168 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
3170 /* skip rest of message */
3171 dout("%s %p msg %p revoked\n", __func__, con, msg);
3172 con->in_base_pos = con->in_base_pos -
3173 sizeof(struct ceph_msg_header) -
3177 sizeof(struct ceph_msg_footer);
3178 ceph_msg_put(con->in_msg);
3180 con->in_tag = CEPH_MSGR_TAG_READY;
3183 dout("%s %p in_msg %p msg %p no-op\n",
3184 __func__, con, con->in_msg, msg);
3186 mutex_unlock(&con->mutex);
3190 * Queue a keepalive byte to ensure the tcp connection is alive.
3192 void ceph_con_keepalive(struct ceph_connection *con)
3194 dout("con_keepalive %p\n", con);
3195 mutex_lock(&con->mutex);
3197 con_flag_set(con, CON_FLAG_KEEPALIVE_PENDING);
3198 mutex_unlock(&con->mutex);
3200 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3203 EXPORT_SYMBOL(ceph_con_keepalive);
3205 bool ceph_con_keepalive_expired(struct ceph_connection *con,
3206 unsigned long interval)
3209 (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
3210 struct timespec now = CURRENT_TIME;
3212 jiffies_to_timespec(interval, &ts);
3213 ts = timespec_add(con->last_keepalive_ack, ts);
3214 return timespec_compare(&now, &ts) >= 0;
3219 static struct ceph_msg_data *ceph_msg_data_create(enum ceph_msg_data_type type)
3221 struct ceph_msg_data *data;
3223 if (WARN_ON(!ceph_msg_data_type_valid(type)))
3226 data = kmem_cache_zalloc(ceph_msg_data_cache, GFP_NOFS);
3229 INIT_LIST_HEAD(&data->links);
3234 static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3239 WARN_ON(!list_empty(&data->links));
3240 if (data->type == CEPH_MSG_DATA_PAGELIST)
3241 ceph_pagelist_release(data->pagelist);
3242 kmem_cache_free(ceph_msg_data_cache, data);
3245 void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
3246 size_t length, size_t alignment)
3248 struct ceph_msg_data *data;
3253 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGES);
3255 data->pages = pages;
3256 data->length = length;
3257 data->alignment = alignment & ~PAGE_MASK;
3259 list_add_tail(&data->links, &msg->data);
3260 msg->data_length += length;
3262 EXPORT_SYMBOL(ceph_msg_data_add_pages);
3264 void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
3265 struct ceph_pagelist *pagelist)
3267 struct ceph_msg_data *data;
3270 BUG_ON(!pagelist->length);
3272 data = ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST);
3274 data->pagelist = pagelist;
3276 list_add_tail(&data->links, &msg->data);
3277 msg->data_length += pagelist->length;
3279 EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
3282 void ceph_msg_data_add_bio(struct ceph_msg *msg, struct bio *bio,
3285 struct ceph_msg_data *data;
3289 data = ceph_msg_data_create(CEPH_MSG_DATA_BIO);
3292 data->bio_length = length;
3294 list_add_tail(&data->links, &msg->data);
3295 msg->data_length += length;
3297 EXPORT_SYMBOL(ceph_msg_data_add_bio);
3298 #endif /* CONFIG_BLOCK */
3301 * construct a new message with given type, size
3302 * the new msg has a ref count of 1.
3304 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3309 m = kmem_cache_zalloc(ceph_msg_cache, flags);
3313 m->hdr.type = cpu_to_le16(type);
3314 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3315 m->hdr.front_len = cpu_to_le32(front_len);
3317 INIT_LIST_HEAD(&m->list_head);
3318 kref_init(&m->kref);
3319 INIT_LIST_HEAD(&m->data);
3323 m->front.iov_base = ceph_kvmalloc(front_len, flags);
3324 if (m->front.iov_base == NULL) {
3325 dout("ceph_msg_new can't allocate %d bytes\n",
3330 m->front.iov_base = NULL;
3332 m->front_alloc_len = m->front.iov_len = front_len;
3334 dout("ceph_msg_new %p front %d\n", m, front_len);
3341 pr_err("msg_new can't create type %d front %d\n", type,
3345 dout("msg_new can't create type %d front %d\n", type,
3350 EXPORT_SYMBOL(ceph_msg_new);
3353 * Allocate "middle" portion of a message, if it is needed and wasn't
3354 * allocated by alloc_msg. This allows us to read a small fixed-size
3355 * per-type header in the front and then gracefully fail (i.e.,
3356 * propagate the error to the caller based on info in the front) when
3357 * the middle is too large.
3359 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3361 int type = le16_to_cpu(msg->hdr.type);
3362 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3364 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3365 ceph_msg_type_name(type), middle_len);
3366 BUG_ON(!middle_len);
3367 BUG_ON(msg->middle);
3369 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3376 * Allocate a message for receiving an incoming message on a
3377 * connection, and save the result in con->in_msg. Uses the
3378 * connection's private alloc_msg op if available.
3380 * Returns 0 on success, or a negative error code.
3382 * On success, if we set *skip = 1:
3383 * - the next message should be skipped and ignored.
3384 * - con->in_msg == NULL
3385 * or if we set *skip = 0:
3386 * - con->in_msg is non-null.
3387 * On error (ENOMEM, EAGAIN, ...),
3388 * - con->in_msg == NULL
3390 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3392 struct ceph_msg_header *hdr = &con->in_hdr;
3393 int middle_len = le32_to_cpu(hdr->middle_len);
3394 struct ceph_msg *msg;
3397 BUG_ON(con->in_msg != NULL);
3398 BUG_ON(!con->ops->alloc_msg);
3400 mutex_unlock(&con->mutex);
3401 msg = con->ops->alloc_msg(con, hdr, skip);
3402 mutex_lock(&con->mutex);
3403 if (con->state != CON_STATE_OPEN) {
3410 msg_con_set(msg, con);
3414 * Null message pointer means either we should skip
3415 * this message or we couldn't allocate memory. The
3416 * former is not an error.
3421 con->error_msg = "error allocating memory for incoming message";
3424 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3426 if (middle_len && !con->in_msg->middle) {
3427 ret = ceph_alloc_middle(con, con->in_msg);
3429 ceph_msg_put(con->in_msg);
3439 * Free a generically kmalloc'd message.
3441 static void ceph_msg_free(struct ceph_msg *m)
3443 dout("%s %p\n", __func__, m);
3444 kvfree(m->front.iov_base);
3445 kmem_cache_free(ceph_msg_cache, m);
3448 static void ceph_msg_release(struct kref *kref)
3450 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3451 struct ceph_msg_data *data, *next;
3453 dout("%s %p\n", __func__, m);
3454 WARN_ON(!list_empty(&m->list_head));
3456 msg_con_set(m, NULL);
3458 /* drop middle, data, if any */
3460 ceph_buffer_put(m->middle);
3464 list_for_each_entry_safe(data, next, &m->data, links) {
3465 list_del_init(&data->links);
3466 ceph_msg_data_destroy(data);
3471 ceph_msgpool_put(m->pool, m);
3476 struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3478 dout("%s %p (was %d)\n", __func__, msg,
3479 atomic_read(&msg->kref.refcount));
3480 kref_get(&msg->kref);
3483 EXPORT_SYMBOL(ceph_msg_get);
3485 void ceph_msg_put(struct ceph_msg *msg)
3487 dout("%s %p (was %d)\n", __func__, msg,
3488 atomic_read(&msg->kref.refcount));
3489 kref_put(&msg->kref, ceph_msg_release);
3491 EXPORT_SYMBOL(ceph_msg_put);
3493 void ceph_msg_dump(struct ceph_msg *msg)
3495 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3496 msg->front_alloc_len, msg->data_length);
3497 print_hex_dump(KERN_DEBUG, "header: ",
3498 DUMP_PREFIX_OFFSET, 16, 1,
3499 &msg->hdr, sizeof(msg->hdr), true);
3500 print_hex_dump(KERN_DEBUG, " front: ",
3501 DUMP_PREFIX_OFFSET, 16, 1,
3502 msg->front.iov_base, msg->front.iov_len, true);
3504 print_hex_dump(KERN_DEBUG, "middle: ",
3505 DUMP_PREFIX_OFFSET, 16, 1,
3506 msg->middle->vec.iov_base,
3507 msg->middle->vec.iov_len, true);
3508 print_hex_dump(KERN_DEBUG, "footer: ",
3509 DUMP_PREFIX_OFFSET, 16, 1,
3510 &msg->footer, sizeof(msg->footer), true);
3512 EXPORT_SYMBOL(ceph_msg_dump);