1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
4 #include <linux/crc32c.h>
5 #include <linux/ctype.h>
6 #include <linux/highmem.h>
7 #include <linux/inet.h>
8 #include <linux/kthread.h>
10 #include <linux/nsproxy.h>
11 #include <linux/sched/mm.h>
12 #include <linux/slab.h>
13 #include <linux/socket.h>
14 #include <linux/string.h>
16 #include <linux/bio.h>
17 #endif /* CONFIG_BLOCK */
18 #include <linux/dns_resolver.h>
21 #include <linux/ceph/ceph_features.h>
22 #include <linux/ceph/libceph.h>
23 #include <linux/ceph/messenger.h>
24 #include <linux/ceph/decode.h>
25 #include <linux/ceph/pagelist.h>
26 #include <linux/export.h>
29 * Ceph uses the messenger to exchange ceph_msg messages with other
30 * hosts in the system. The messenger provides ordered and reliable
31 * delivery. We tolerate TCP disconnects by reconnecting (with
32 * exponential backoff) in the case of a fault (disconnection, bad
33 * crc, protocol error). Acks allow sent messages to be discarded by
38 * We track the state of the socket on a given connection using
39 * values defined below. The transition to a new socket state is
40 * handled by a function which verifies we aren't coming from an
44 * | NEW* | transient initial state
46 * | con_sock_state_init()
49 * | CLOSED | initialized, but no socket (and no
50 * ---------- TCP connection)
52 * | \ con_sock_state_connecting()
53 * | ----------------------
55 * + con_sock_state_closed() \
56 * |+--------------------------- \
59 * | | CLOSING | socket event; \ \
60 * | ----------- await close \ \
63 * | + con_sock_state_closing() \ |
65 * | / --------------- | |
68 * | / -----------------| CONNECTING | socket created, TCP
69 * | | / -------------- connect initiated
70 * | | | con_sock_state_connected()
73 * | CONNECTED | TCP connection established
76 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
79 #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
80 #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
81 #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
82 #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
83 #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
88 #define CON_STATE_CLOSED 1 /* -> PREOPEN */
89 #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
90 #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
91 #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
92 #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
93 #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
96 * ceph_connection flag bits
98 #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
99 * messages on errors */
100 #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
101 #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
102 #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
103 #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
105 static bool con_flag_valid(unsigned long con_flag)
108 case CON_FLAG_LOSSYTX:
109 case CON_FLAG_KEEPALIVE_PENDING:
110 case CON_FLAG_WRITE_PENDING:
111 case CON_FLAG_SOCK_CLOSED:
112 case CON_FLAG_BACKOFF:
119 static void con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
121 BUG_ON(!con_flag_valid(con_flag));
123 clear_bit(con_flag, &con->flags);
126 static void con_flag_set(struct ceph_connection *con, unsigned long con_flag)
128 BUG_ON(!con_flag_valid(con_flag));
130 set_bit(con_flag, &con->flags);
133 static bool con_flag_test(struct ceph_connection *con, unsigned long con_flag)
135 BUG_ON(!con_flag_valid(con_flag));
137 return test_bit(con_flag, &con->flags);
140 static bool con_flag_test_and_clear(struct ceph_connection *con,
141 unsigned long con_flag)
143 BUG_ON(!con_flag_valid(con_flag));
145 return test_and_clear_bit(con_flag, &con->flags);
148 static bool con_flag_test_and_set(struct ceph_connection *con,
149 unsigned long con_flag)
151 BUG_ON(!con_flag_valid(con_flag));
153 return test_and_set_bit(con_flag, &con->flags);
156 /* Slab caches for frequently-allocated structures */
158 static struct kmem_cache *ceph_msg_cache;
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;
170 static void queue_con(struct ceph_connection *con);
171 static void cancel_con(struct ceph_connection *con);
172 static void ceph_con_workfn(struct work_struct *);
173 static void con_fault(struct ceph_connection *con);
176 * Nicely render a sockaddr as a string. An array of formatted
177 * strings is used, to approximate reentrancy.
179 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
180 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
181 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
182 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
184 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
185 static atomic_t addr_str_seq = ATOMIC_INIT(0);
187 static struct page *zero_page; /* used in certain error cases */
189 const char *ceph_pr_addr(const struct ceph_entity_addr *addr)
193 struct sockaddr_storage ss = addr->in_addr; /* align */
194 struct sockaddr_in *in4 = (struct sockaddr_in *)&ss;
195 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *)&ss;
197 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
200 switch (ss.ss_family) {
202 snprintf(s, MAX_ADDR_STR_LEN, "(%d)%pI4:%hu",
203 le32_to_cpu(addr->type), &in4->sin_addr,
204 ntohs(in4->sin_port));
208 snprintf(s, MAX_ADDR_STR_LEN, "(%d)[%pI6c]:%hu",
209 le32_to_cpu(addr->type), &in6->sin6_addr,
210 ntohs(in6->sin6_port));
214 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
220 EXPORT_SYMBOL(ceph_pr_addr);
222 static void encode_my_addr(struct ceph_messenger *msgr)
224 memcpy(&msgr->my_enc_addr, &msgr->inst.addr, sizeof(msgr->my_enc_addr));
225 ceph_encode_banner_addr(&msgr->my_enc_addr);
229 * work queue for all reading and writing to/from the socket.
231 static struct workqueue_struct *ceph_msgr_wq;
233 static int ceph_msgr_slab_init(void)
235 BUG_ON(ceph_msg_cache);
236 ceph_msg_cache = KMEM_CACHE(ceph_msg, 0);
243 static void ceph_msgr_slab_exit(void)
245 BUG_ON(!ceph_msg_cache);
246 kmem_cache_destroy(ceph_msg_cache);
247 ceph_msg_cache = NULL;
250 static void _ceph_msgr_exit(void)
253 destroy_workqueue(ceph_msgr_wq);
257 BUG_ON(zero_page == NULL);
261 ceph_msgr_slab_exit();
264 int __init ceph_msgr_init(void)
266 if (ceph_msgr_slab_init())
269 BUG_ON(zero_page != NULL);
270 zero_page = ZERO_PAGE(0);
274 * The number of active work items is limited by the number of
275 * connections, so leave @max_active at default.
277 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
281 pr_err("msgr_init failed to create workqueue\n");
287 void ceph_msgr_exit(void)
289 BUG_ON(ceph_msgr_wq == NULL);
294 void ceph_msgr_flush(void)
296 flush_workqueue(ceph_msgr_wq);
298 EXPORT_SYMBOL(ceph_msgr_flush);
300 /* Connection socket state transition functions */
302 static void con_sock_state_init(struct ceph_connection *con)
306 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
307 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
308 printk("%s: unexpected old state %d\n", __func__, old_state);
309 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
310 CON_SOCK_STATE_CLOSED);
313 static void con_sock_state_connecting(struct ceph_connection *con)
317 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
318 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
319 printk("%s: unexpected old state %d\n", __func__, old_state);
320 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
321 CON_SOCK_STATE_CONNECTING);
324 static void con_sock_state_connected(struct ceph_connection *con)
328 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
329 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
330 printk("%s: unexpected old state %d\n", __func__, old_state);
331 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
332 CON_SOCK_STATE_CONNECTED);
335 static void con_sock_state_closing(struct ceph_connection *con)
339 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
340 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
341 old_state != CON_SOCK_STATE_CONNECTED &&
342 old_state != CON_SOCK_STATE_CLOSING))
343 printk("%s: unexpected old state %d\n", __func__, old_state);
344 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
345 CON_SOCK_STATE_CLOSING);
348 static void con_sock_state_closed(struct ceph_connection *con)
352 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
353 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
354 old_state != CON_SOCK_STATE_CLOSING &&
355 old_state != CON_SOCK_STATE_CONNECTING &&
356 old_state != CON_SOCK_STATE_CLOSED))
357 printk("%s: unexpected old state %d\n", __func__, old_state);
358 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
359 CON_SOCK_STATE_CLOSED);
363 * socket callback functions
366 /* data available on socket, or listen socket received a connect */
367 static void ceph_sock_data_ready(struct sock *sk)
369 struct ceph_connection *con = sk->sk_user_data;
370 if (atomic_read(&con->msgr->stopping)) {
374 if (sk->sk_state != TCP_CLOSE_WAIT) {
375 dout("%s on %p state = %lu, queueing work\n", __func__,
381 /* socket has buffer space for writing */
382 static void ceph_sock_write_space(struct sock *sk)
384 struct ceph_connection *con = sk->sk_user_data;
386 /* only queue to workqueue if there is data we want to write,
387 * and there is sufficient space in the socket buffer to accept
388 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
389 * doesn't get called again until try_write() fills the socket
390 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
391 * and net/core/stream.c:sk_stream_write_space().
393 if (con_flag_test(con, CON_FLAG_WRITE_PENDING)) {
394 if (sk_stream_is_writeable(sk)) {
395 dout("%s %p queueing write work\n", __func__, con);
396 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
400 dout("%s %p nothing to write\n", __func__, con);
404 /* socket's state has changed */
405 static void ceph_sock_state_change(struct sock *sk)
407 struct ceph_connection *con = sk->sk_user_data;
409 dout("%s %p state = %lu sk_state = %u\n", __func__,
410 con, con->state, sk->sk_state);
412 switch (sk->sk_state) {
414 dout("%s TCP_CLOSE\n", __func__);
417 dout("%s TCP_CLOSE_WAIT\n", __func__);
418 con_sock_state_closing(con);
419 con_flag_set(con, CON_FLAG_SOCK_CLOSED);
422 case TCP_ESTABLISHED:
423 dout("%s TCP_ESTABLISHED\n", __func__);
424 con_sock_state_connected(con);
427 default: /* Everything else is uninteresting */
433 * set up socket callbacks
435 static void set_sock_callbacks(struct socket *sock,
436 struct ceph_connection *con)
438 struct sock *sk = sock->sk;
439 sk->sk_user_data = con;
440 sk->sk_data_ready = ceph_sock_data_ready;
441 sk->sk_write_space = ceph_sock_write_space;
442 sk->sk_state_change = ceph_sock_state_change;
451 * initiate connection to a remote socket.
453 static int ceph_tcp_connect(struct ceph_connection *con)
455 struct sockaddr_storage ss = con->peer_addr.in_addr; /* align */
457 unsigned int noio_flag;
462 /* sock_create_kern() allocates with GFP_KERNEL */
463 noio_flag = memalloc_noio_save();
464 ret = sock_create_kern(read_pnet(&con->msgr->net), ss.ss_family,
465 SOCK_STREAM, IPPROTO_TCP, &sock);
466 memalloc_noio_restore(noio_flag);
469 sock->sk->sk_allocation = GFP_NOFS;
471 #ifdef CONFIG_LOCKDEP
472 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
475 set_sock_callbacks(sock, con);
477 dout("connect %s\n", ceph_pr_addr(&con->peer_addr));
479 con_sock_state_connecting(con);
480 ret = sock->ops->connect(sock, (struct sockaddr *)&ss, sizeof(ss),
482 if (ret == -EINPROGRESS) {
483 dout("connect %s EINPROGRESS sk_state = %u\n",
484 ceph_pr_addr(&con->peer_addr),
486 } else if (ret < 0) {
487 pr_err("connect %s error %d\n",
488 ceph_pr_addr(&con->peer_addr), ret);
493 if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY)) {
496 ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
497 (char *)&optval, sizeof(optval));
499 pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
508 * If @buf is NULL, discard up to @len bytes.
510 static int ceph_tcp_recvmsg(struct socket *sock, void *buf, size_t len)
512 struct kvec iov = {buf, len};
513 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
517 msg.msg_flags |= MSG_TRUNC;
519 iov_iter_kvec(&msg.msg_iter, READ, &iov, 1, len);
520 r = sock_recvmsg(sock, &msg, msg.msg_flags);
526 static int ceph_tcp_recvpage(struct socket *sock, struct page *page,
527 int page_offset, size_t length)
529 struct bio_vec bvec = {
531 .bv_offset = page_offset,
534 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
537 BUG_ON(page_offset + length > PAGE_SIZE);
538 iov_iter_bvec(&msg.msg_iter, READ, &bvec, 1, length);
539 r = sock_recvmsg(sock, &msg, msg.msg_flags);
546 * write something. @more is true if caller will be sending more data
549 static int ceph_tcp_sendmsg(struct socket *sock, struct kvec *iov,
550 size_t kvlen, size_t len, bool more)
552 struct msghdr msg = { .msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL };
556 msg.msg_flags |= MSG_MORE;
558 msg.msg_flags |= MSG_EOR; /* superfluous, but what the hell */
560 r = kernel_sendmsg(sock, &msg, iov, kvlen, len);
567 * @more: either or both of MSG_MORE and MSG_SENDPAGE_NOTLAST
569 static int ceph_tcp_sendpage(struct socket *sock, struct page *page,
570 int offset, size_t size, int more)
572 ssize_t (*sendpage)(struct socket *sock, struct page *page,
573 int offset, size_t size, int flags);
574 int flags = MSG_DONTWAIT | MSG_NOSIGNAL | more;
578 * sendpage cannot properly handle pages with page_count == 0,
579 * we need to fall back to sendmsg if that's the case.
581 * Same goes for slab pages: skb_can_coalesce() allows
582 * coalescing neighboring slab objects into a single frag which
583 * triggers one of hardened usercopy checks.
585 if (page_count(page) >= 1 && !PageSlab(page))
586 sendpage = sock->ops->sendpage;
588 sendpage = sock_no_sendpage;
590 ret = sendpage(sock, page, offset, size, flags);
598 * Shutdown/close the socket for the given connection.
600 static int con_close_socket(struct ceph_connection *con)
604 dout("con_close_socket on %p sock %p\n", con, con->sock);
606 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
607 sock_release(con->sock);
612 * Forcibly clear the SOCK_CLOSED flag. It gets set
613 * independent of the connection mutex, and we could have
614 * received a socket close event before we had the chance to
615 * shut the socket down.
617 con_flag_clear(con, CON_FLAG_SOCK_CLOSED);
619 con_sock_state_closed(con);
624 * Reset a connection. Discard all incoming and outgoing messages
625 * and clear *_seq state.
627 static void ceph_msg_remove(struct ceph_msg *msg)
629 list_del_init(&msg->list_head);
633 static void ceph_msg_remove_list(struct list_head *head)
635 while (!list_empty(head)) {
636 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
638 ceph_msg_remove(msg);
642 static void reset_connection(struct ceph_connection *con)
644 /* reset connection, out_queue, msg_ and connect_seq */
645 /* discard existing out_queue and msg_seq */
646 dout("reset_connection %p\n", con);
647 ceph_msg_remove_list(&con->out_queue);
648 ceph_msg_remove_list(&con->out_sent);
651 BUG_ON(con->in_msg->con != con);
652 ceph_msg_put(con->in_msg);
656 con->connect_seq = 0;
659 BUG_ON(con->out_msg->con != con);
660 ceph_msg_put(con->out_msg);
664 con->in_seq_acked = 0;
670 * mark a peer down. drop any open connections.
672 void ceph_con_close(struct ceph_connection *con)
674 mutex_lock(&con->mutex);
675 dout("con_close %p peer %s\n", con, ceph_pr_addr(&con->peer_addr));
676 con->state = CON_STATE_CLOSED;
678 con_flag_clear(con, CON_FLAG_LOSSYTX); /* so we retry next connect */
679 con_flag_clear(con, CON_FLAG_KEEPALIVE_PENDING);
680 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
681 con_flag_clear(con, CON_FLAG_BACKOFF);
683 reset_connection(con);
684 con->peer_global_seq = 0;
686 con_close_socket(con);
687 mutex_unlock(&con->mutex);
689 EXPORT_SYMBOL(ceph_con_close);
692 * Reopen a closed connection, with a new peer address.
694 void ceph_con_open(struct ceph_connection *con,
695 __u8 entity_type, __u64 entity_num,
696 struct ceph_entity_addr *addr)
698 mutex_lock(&con->mutex);
699 dout("con_open %p %s\n", con, ceph_pr_addr(addr));
701 WARN_ON(con->state != CON_STATE_CLOSED);
702 con->state = CON_STATE_PREOPEN;
704 con->peer_name.type = (__u8) entity_type;
705 con->peer_name.num = cpu_to_le64(entity_num);
707 memcpy(&con->peer_addr, addr, sizeof(*addr));
708 con->delay = 0; /* reset backoff memory */
709 mutex_unlock(&con->mutex);
712 EXPORT_SYMBOL(ceph_con_open);
715 * return true if this connection ever successfully opened
717 bool ceph_con_opened(struct ceph_connection *con)
719 return con->connect_seq > 0;
723 * initialize a new connection.
725 void ceph_con_init(struct ceph_connection *con, void *private,
726 const struct ceph_connection_operations *ops,
727 struct ceph_messenger *msgr)
729 dout("con_init %p\n", con);
730 memset(con, 0, sizeof(*con));
731 con->private = private;
735 con_sock_state_init(con);
737 mutex_init(&con->mutex);
738 INIT_LIST_HEAD(&con->out_queue);
739 INIT_LIST_HEAD(&con->out_sent);
740 INIT_DELAYED_WORK(&con->work, ceph_con_workfn);
742 con->state = CON_STATE_CLOSED;
744 EXPORT_SYMBOL(ceph_con_init);
748 * We maintain a global counter to order connection attempts. Get
749 * a unique seq greater than @gt.
751 static u32 get_global_seq(struct ceph_messenger *msgr, u32 gt)
755 spin_lock(&msgr->global_seq_lock);
756 if (msgr->global_seq < gt)
757 msgr->global_seq = gt;
758 ret = ++msgr->global_seq;
759 spin_unlock(&msgr->global_seq_lock);
763 static void con_out_kvec_reset(struct ceph_connection *con)
765 BUG_ON(con->out_skip);
767 con->out_kvec_left = 0;
768 con->out_kvec_bytes = 0;
769 con->out_kvec_cur = &con->out_kvec[0];
772 static void con_out_kvec_add(struct ceph_connection *con,
773 size_t size, void *data)
775 int index = con->out_kvec_left;
777 BUG_ON(con->out_skip);
778 BUG_ON(index >= ARRAY_SIZE(con->out_kvec));
780 con->out_kvec[index].iov_len = size;
781 con->out_kvec[index].iov_base = data;
782 con->out_kvec_left++;
783 con->out_kvec_bytes += size;
787 * Chop off a kvec from the end. Return residual number of bytes for
788 * that kvec, i.e. how many bytes would have been written if the kvec
791 static int con_out_kvec_skip(struct ceph_connection *con)
793 int off = con->out_kvec_cur - con->out_kvec;
796 if (con->out_kvec_bytes > 0) {
797 skip = con->out_kvec[off + con->out_kvec_left - 1].iov_len;
798 BUG_ON(con->out_kvec_bytes < skip);
799 BUG_ON(!con->out_kvec_left);
800 con->out_kvec_bytes -= skip;
801 con->out_kvec_left--;
810 * For a bio data item, a piece is whatever remains of the next
811 * entry in the current bio iovec, or the first entry in the next
814 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
817 struct ceph_msg_data *data = cursor->data;
818 struct ceph_bio_iter *it = &cursor->bio_iter;
820 cursor->resid = min_t(size_t, length, data->bio_length);
822 if (cursor->resid < it->iter.bi_size)
823 it->iter.bi_size = cursor->resid;
825 BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
826 cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter);
829 static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
833 struct bio_vec bv = bio_iter_iovec(cursor->bio_iter.bio,
834 cursor->bio_iter.iter);
836 *page_offset = bv.bv_offset;
841 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
844 struct ceph_bio_iter *it = &cursor->bio_iter;
845 struct page *page = bio_iter_page(it->bio, it->iter);
847 BUG_ON(bytes > cursor->resid);
848 BUG_ON(bytes > bio_iter_len(it->bio, it->iter));
849 cursor->resid -= bytes;
850 bio_advance_iter(it->bio, &it->iter, bytes);
852 if (!cursor->resid) {
853 BUG_ON(!cursor->last_piece);
854 return false; /* no more data */
857 if (!bytes || (it->iter.bi_size && it->iter.bi_bvec_done &&
858 page == bio_iter_page(it->bio, it->iter)))
859 return false; /* more bytes to process in this segment */
861 if (!it->iter.bi_size) {
862 it->bio = it->bio->bi_next;
863 it->iter = it->bio->bi_iter;
864 if (cursor->resid < it->iter.bi_size)
865 it->iter.bi_size = cursor->resid;
868 BUG_ON(cursor->last_piece);
869 BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
870 cursor->last_piece = cursor->resid == bio_iter_len(it->bio, it->iter);
873 #endif /* CONFIG_BLOCK */
875 static void ceph_msg_data_bvecs_cursor_init(struct ceph_msg_data_cursor *cursor,
878 struct ceph_msg_data *data = cursor->data;
879 struct bio_vec *bvecs = data->bvec_pos.bvecs;
881 cursor->resid = min_t(size_t, length, data->bvec_pos.iter.bi_size);
882 cursor->bvec_iter = data->bvec_pos.iter;
883 cursor->bvec_iter.bi_size = cursor->resid;
885 BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
887 cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter);
890 static struct page *ceph_msg_data_bvecs_next(struct ceph_msg_data_cursor *cursor,
894 struct bio_vec bv = bvec_iter_bvec(cursor->data->bvec_pos.bvecs,
897 *page_offset = bv.bv_offset;
902 static bool ceph_msg_data_bvecs_advance(struct ceph_msg_data_cursor *cursor,
905 struct bio_vec *bvecs = cursor->data->bvec_pos.bvecs;
906 struct page *page = bvec_iter_page(bvecs, cursor->bvec_iter);
908 BUG_ON(bytes > cursor->resid);
909 BUG_ON(bytes > bvec_iter_len(bvecs, cursor->bvec_iter));
910 cursor->resid -= bytes;
911 bvec_iter_advance(bvecs, &cursor->bvec_iter, bytes);
913 if (!cursor->resid) {
914 BUG_ON(!cursor->last_piece);
915 return false; /* no more data */
918 if (!bytes || (cursor->bvec_iter.bi_bvec_done &&
919 page == bvec_iter_page(bvecs, cursor->bvec_iter)))
920 return false; /* more bytes to process in this segment */
922 BUG_ON(cursor->last_piece);
923 BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
925 cursor->resid == bvec_iter_len(bvecs, cursor->bvec_iter);
930 * For a page array, a piece comes from the first page in the array
931 * that has not already been fully consumed.
933 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
936 struct ceph_msg_data *data = cursor->data;
939 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
941 BUG_ON(!data->pages);
942 BUG_ON(!data->length);
944 cursor->resid = min(length, data->length);
945 page_count = calc_pages_for(data->alignment, (u64)data->length);
946 cursor->page_offset = data->alignment & ~PAGE_MASK;
947 cursor->page_index = 0;
948 BUG_ON(page_count > (int)USHRT_MAX);
949 cursor->page_count = (unsigned short)page_count;
950 BUG_ON(length > SIZE_MAX - cursor->page_offset);
951 cursor->last_piece = cursor->page_offset + cursor->resid <= PAGE_SIZE;
955 ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
956 size_t *page_offset, size_t *length)
958 struct ceph_msg_data *data = cursor->data;
960 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
962 BUG_ON(cursor->page_index >= cursor->page_count);
963 BUG_ON(cursor->page_offset >= PAGE_SIZE);
965 *page_offset = cursor->page_offset;
966 if (cursor->last_piece)
967 *length = cursor->resid;
969 *length = PAGE_SIZE - *page_offset;
971 return data->pages[cursor->page_index];
974 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
977 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
979 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
981 /* Advance the cursor page offset */
983 cursor->resid -= bytes;
984 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
985 if (!bytes || cursor->page_offset)
986 return false; /* more bytes to process in the current page */
989 return false; /* no more data */
991 /* Move on to the next page; offset is already at 0 */
993 BUG_ON(cursor->page_index >= cursor->page_count);
994 cursor->page_index++;
995 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1001 * For a pagelist, a piece is whatever remains to be consumed in the
1002 * first page in the list, or the front of the next page.
1005 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
1008 struct ceph_msg_data *data = cursor->data;
1009 struct ceph_pagelist *pagelist;
1012 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1014 pagelist = data->pagelist;
1018 return; /* pagelist can be assigned but empty */
1020 BUG_ON(list_empty(&pagelist->head));
1021 page = list_first_entry(&pagelist->head, struct page, lru);
1023 cursor->resid = min(length, pagelist->length);
1024 cursor->page = page;
1026 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1029 static struct page *
1030 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
1031 size_t *page_offset, size_t *length)
1033 struct ceph_msg_data *data = cursor->data;
1034 struct ceph_pagelist *pagelist;
1036 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1038 pagelist = data->pagelist;
1041 BUG_ON(!cursor->page);
1042 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1044 /* offset of first page in pagelist is always 0 */
1045 *page_offset = cursor->offset & ~PAGE_MASK;
1046 if (cursor->last_piece)
1047 *length = cursor->resid;
1049 *length = PAGE_SIZE - *page_offset;
1051 return cursor->page;
1054 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
1057 struct ceph_msg_data *data = cursor->data;
1058 struct ceph_pagelist *pagelist;
1060 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
1062 pagelist = data->pagelist;
1065 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
1066 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
1068 /* Advance the cursor offset */
1070 cursor->resid -= bytes;
1071 cursor->offset += bytes;
1072 /* offset of first page in pagelist is always 0 */
1073 if (!bytes || cursor->offset & ~PAGE_MASK)
1074 return false; /* more bytes to process in the current page */
1077 return false; /* no more data */
1079 /* Move on to the next page */
1081 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
1082 cursor->page = list_next_entry(cursor->page, lru);
1083 cursor->last_piece = cursor->resid <= PAGE_SIZE;
1089 * Message data is handled (sent or received) in pieces, where each
1090 * piece resides on a single page. The network layer might not
1091 * consume an entire piece at once. A data item's cursor keeps
1092 * track of which piece is next to process and how much remains to
1093 * be processed in that piece. It also tracks whether the current
1094 * piece is the last one in the data item.
1096 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
1098 size_t length = cursor->total_resid;
1100 switch (cursor->data->type) {
1101 case CEPH_MSG_DATA_PAGELIST:
1102 ceph_msg_data_pagelist_cursor_init(cursor, length);
1104 case CEPH_MSG_DATA_PAGES:
1105 ceph_msg_data_pages_cursor_init(cursor, length);
1108 case CEPH_MSG_DATA_BIO:
1109 ceph_msg_data_bio_cursor_init(cursor, length);
1111 #endif /* CONFIG_BLOCK */
1112 case CEPH_MSG_DATA_BVECS:
1113 ceph_msg_data_bvecs_cursor_init(cursor, length);
1115 case CEPH_MSG_DATA_NONE:
1120 cursor->need_crc = true;
1123 static void ceph_msg_data_cursor_init(struct ceph_msg *msg, size_t length)
1125 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1128 BUG_ON(length > msg->data_length);
1129 BUG_ON(!msg->num_data_items);
1131 cursor->total_resid = length;
1132 cursor->data = msg->data;
1134 __ceph_msg_data_cursor_init(cursor);
1138 * Return the page containing the next piece to process for a given
1139 * data item, and supply the page offset and length of that piece.
1140 * Indicate whether this is the last piece in this data item.
1142 static struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1143 size_t *page_offset, size_t *length,
1148 switch (cursor->data->type) {
1149 case CEPH_MSG_DATA_PAGELIST:
1150 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1152 case CEPH_MSG_DATA_PAGES:
1153 page = ceph_msg_data_pages_next(cursor, page_offset, length);
1156 case CEPH_MSG_DATA_BIO:
1157 page = ceph_msg_data_bio_next(cursor, page_offset, length);
1159 #endif /* CONFIG_BLOCK */
1160 case CEPH_MSG_DATA_BVECS:
1161 page = ceph_msg_data_bvecs_next(cursor, page_offset, length);
1163 case CEPH_MSG_DATA_NONE:
1170 BUG_ON(*page_offset + *length > PAGE_SIZE);
1172 BUG_ON(*length > cursor->resid);
1174 *last_piece = cursor->last_piece;
1180 * Returns true if the result moves the cursor on to the next piece
1183 static void ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor,
1188 BUG_ON(bytes > cursor->resid);
1189 switch (cursor->data->type) {
1190 case CEPH_MSG_DATA_PAGELIST:
1191 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1193 case CEPH_MSG_DATA_PAGES:
1194 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1197 case CEPH_MSG_DATA_BIO:
1198 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1200 #endif /* CONFIG_BLOCK */
1201 case CEPH_MSG_DATA_BVECS:
1202 new_piece = ceph_msg_data_bvecs_advance(cursor, bytes);
1204 case CEPH_MSG_DATA_NONE:
1209 cursor->total_resid -= bytes;
1211 if (!cursor->resid && cursor->total_resid) {
1212 WARN_ON(!cursor->last_piece);
1214 __ceph_msg_data_cursor_init(cursor);
1217 cursor->need_crc = new_piece;
1220 static size_t sizeof_footer(struct ceph_connection *con)
1222 return (con->peer_features & CEPH_FEATURE_MSG_AUTH) ?
1223 sizeof(struct ceph_msg_footer) :
1224 sizeof(struct ceph_msg_footer_old);
1227 static void prepare_message_data(struct ceph_msg *msg, u32 data_len)
1229 /* Initialize data cursor */
1231 ceph_msg_data_cursor_init(msg, (size_t)data_len);
1235 * Prepare footer for currently outgoing message, and finish things
1236 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1238 static void prepare_write_message_footer(struct ceph_connection *con)
1240 struct ceph_msg *m = con->out_msg;
1242 m->footer.flags |= CEPH_MSG_FOOTER_COMPLETE;
1244 dout("prepare_write_message_footer %p\n", con);
1245 con_out_kvec_add(con, sizeof_footer(con), &m->footer);
1246 if (con->peer_features & CEPH_FEATURE_MSG_AUTH) {
1247 if (con->ops->sign_message)
1248 con->ops->sign_message(m);
1252 m->old_footer.flags = m->footer.flags;
1254 con->out_more = m->more_to_follow;
1255 con->out_msg_done = true;
1259 * Prepare headers for the next outgoing message.
1261 static void prepare_write_message(struct ceph_connection *con)
1266 con_out_kvec_reset(con);
1267 con->out_msg_done = false;
1269 /* Sneak an ack in there first? If we can get it into the same
1270 * TCP packet that's a good thing. */
1271 if (con->in_seq > con->in_seq_acked) {
1272 con->in_seq_acked = con->in_seq;
1273 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1274 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1275 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1276 &con->out_temp_ack);
1279 BUG_ON(list_empty(&con->out_queue));
1280 m = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
1282 BUG_ON(m->con != con);
1284 /* put message on sent list */
1286 list_move_tail(&m->list_head, &con->out_sent);
1289 * only assign outgoing seq # if we haven't sent this message
1290 * yet. if it is requeued, resend with it's original seq.
1292 if (m->needs_out_seq) {
1293 m->hdr.seq = cpu_to_le64(++con->out_seq);
1294 m->needs_out_seq = false;
1296 if (con->ops->reencode_message)
1297 con->ops->reencode_message(m);
1300 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1301 m, con->out_seq, le16_to_cpu(m->hdr.type),
1302 le32_to_cpu(m->hdr.front_len), le32_to_cpu(m->hdr.middle_len),
1304 WARN_ON(m->front.iov_len != le32_to_cpu(m->hdr.front_len));
1305 WARN_ON(m->data_length != le32_to_cpu(m->hdr.data_len));
1307 /* tag + hdr + front + middle */
1308 con_out_kvec_add(con, sizeof (tag_msg), &tag_msg);
1309 con_out_kvec_add(con, sizeof(con->out_hdr), &con->out_hdr);
1310 con_out_kvec_add(con, m->front.iov_len, m->front.iov_base);
1313 con_out_kvec_add(con, m->middle->vec.iov_len,
1314 m->middle->vec.iov_base);
1316 /* fill in hdr crc and finalize hdr */
1317 crc = crc32c(0, &m->hdr, offsetof(struct ceph_msg_header, crc));
1318 con->out_msg->hdr.crc = cpu_to_le32(crc);
1319 memcpy(&con->out_hdr, &con->out_msg->hdr, sizeof(con->out_hdr));
1321 /* fill in front and middle crc, footer */
1322 crc = crc32c(0, m->front.iov_base, m->front.iov_len);
1323 con->out_msg->footer.front_crc = cpu_to_le32(crc);
1325 crc = crc32c(0, m->middle->vec.iov_base,
1326 m->middle->vec.iov_len);
1327 con->out_msg->footer.middle_crc = cpu_to_le32(crc);
1329 con->out_msg->footer.middle_crc = 0;
1330 dout("%s front_crc %u middle_crc %u\n", __func__,
1331 le32_to_cpu(con->out_msg->footer.front_crc),
1332 le32_to_cpu(con->out_msg->footer.middle_crc));
1333 con->out_msg->footer.flags = 0;
1335 /* is there a data payload? */
1336 con->out_msg->footer.data_crc = 0;
1337 if (m->data_length) {
1338 prepare_message_data(con->out_msg, m->data_length);
1339 con->out_more = 1; /* data + footer will follow */
1341 /* no, queue up footer too and be done */
1342 prepare_write_message_footer(con);
1345 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1351 static void prepare_write_ack(struct ceph_connection *con)
1353 dout("prepare_write_ack %p %llu -> %llu\n", con,
1354 con->in_seq_acked, con->in_seq);
1355 con->in_seq_acked = con->in_seq;
1357 con_out_kvec_reset(con);
1359 con_out_kvec_add(con, sizeof (tag_ack), &tag_ack);
1361 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1362 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1363 &con->out_temp_ack);
1365 con->out_more = 1; /* more will follow.. eventually.. */
1366 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1370 * Prepare to share the seq during handshake
1372 static void prepare_write_seq(struct ceph_connection *con)
1374 dout("prepare_write_seq %p %llu -> %llu\n", con,
1375 con->in_seq_acked, con->in_seq);
1376 con->in_seq_acked = con->in_seq;
1378 con_out_kvec_reset(con);
1380 con->out_temp_ack = cpu_to_le64(con->in_seq_acked);
1381 con_out_kvec_add(con, sizeof (con->out_temp_ack),
1382 &con->out_temp_ack);
1384 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1388 * Prepare to write keepalive byte.
1390 static void prepare_write_keepalive(struct ceph_connection *con)
1392 dout("prepare_write_keepalive %p\n", con);
1393 con_out_kvec_reset(con);
1394 if (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2) {
1395 struct timespec64 now;
1397 ktime_get_real_ts64(&now);
1398 con_out_kvec_add(con, sizeof(tag_keepalive2), &tag_keepalive2);
1399 ceph_encode_timespec64(&con->out_temp_keepalive2, &now);
1400 con_out_kvec_add(con, sizeof(con->out_temp_keepalive2),
1401 &con->out_temp_keepalive2);
1403 con_out_kvec_add(con, sizeof(tag_keepalive), &tag_keepalive);
1405 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1409 * Connection negotiation.
1412 static int get_connect_authorizer(struct ceph_connection *con)
1414 struct ceph_auth_handshake *auth;
1417 if (!con->ops->get_authorizer) {
1419 con->out_connect.authorizer_protocol = CEPH_AUTH_UNKNOWN;
1420 con->out_connect.authorizer_len = 0;
1424 auth = con->ops->get_authorizer(con, &auth_proto, con->auth_retry);
1426 return PTR_ERR(auth);
1429 con->out_connect.authorizer_protocol = cpu_to_le32(auth_proto);
1430 con->out_connect.authorizer_len = cpu_to_le32(auth->authorizer_buf_len);
1435 * We connected to a peer and are saying hello.
1437 static void prepare_write_banner(struct ceph_connection *con)
1439 con_out_kvec_add(con, strlen(CEPH_BANNER), CEPH_BANNER);
1440 con_out_kvec_add(con, sizeof (con->msgr->my_enc_addr),
1441 &con->msgr->my_enc_addr);
1444 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1447 static void __prepare_write_connect(struct ceph_connection *con)
1449 con_out_kvec_add(con, sizeof(con->out_connect), &con->out_connect);
1451 con_out_kvec_add(con, con->auth->authorizer_buf_len,
1452 con->auth->authorizer_buf);
1455 con_flag_set(con, CON_FLAG_WRITE_PENDING);
1458 static int prepare_write_connect(struct ceph_connection *con)
1460 unsigned int global_seq = get_global_seq(con->msgr, 0);
1464 switch (con->peer_name.type) {
1465 case CEPH_ENTITY_TYPE_MON:
1466 proto = CEPH_MONC_PROTOCOL;
1468 case CEPH_ENTITY_TYPE_OSD:
1469 proto = CEPH_OSDC_PROTOCOL;
1471 case CEPH_ENTITY_TYPE_MDS:
1472 proto = CEPH_MDSC_PROTOCOL;
1478 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con,
1479 con->connect_seq, global_seq, proto);
1481 con->out_connect.features =
1482 cpu_to_le64(from_msgr(con->msgr)->supported_features);
1483 con->out_connect.host_type = cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT);
1484 con->out_connect.connect_seq = cpu_to_le32(con->connect_seq);
1485 con->out_connect.global_seq = cpu_to_le32(global_seq);
1486 con->out_connect.protocol_version = cpu_to_le32(proto);
1487 con->out_connect.flags = 0;
1489 ret = get_connect_authorizer(con);
1493 __prepare_write_connect(con);
1498 * write as much of pending kvecs to the socket as we can.
1500 * 0 -> socket full, but more to do
1503 static int write_partial_kvec(struct ceph_connection *con)
1507 dout("write_partial_kvec %p %d left\n", con, con->out_kvec_bytes);
1508 while (con->out_kvec_bytes > 0) {
1509 ret = ceph_tcp_sendmsg(con->sock, con->out_kvec_cur,
1510 con->out_kvec_left, con->out_kvec_bytes,
1514 con->out_kvec_bytes -= ret;
1515 if (con->out_kvec_bytes == 0)
1518 /* account for full iov entries consumed */
1519 while (ret >= con->out_kvec_cur->iov_len) {
1520 BUG_ON(!con->out_kvec_left);
1521 ret -= con->out_kvec_cur->iov_len;
1522 con->out_kvec_cur++;
1523 con->out_kvec_left--;
1525 /* and for a partially-consumed entry */
1527 con->out_kvec_cur->iov_len -= ret;
1528 con->out_kvec_cur->iov_base += ret;
1531 con->out_kvec_left = 0;
1534 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con,
1535 con->out_kvec_bytes, con->out_kvec_left, ret);
1536 return ret; /* done! */
1539 static u32 ceph_crc32c_page(u32 crc, struct page *page,
1540 unsigned int page_offset,
1541 unsigned int length)
1546 BUG_ON(kaddr == NULL);
1547 crc = crc32c(crc, kaddr + page_offset, length);
1553 * Write as much message data payload as we can. If we finish, queue
1555 * 1 -> done, footer is now queued in out_kvec[].
1556 * 0 -> socket full, but more to do
1559 static int write_partial_message_data(struct ceph_connection *con)
1561 struct ceph_msg *msg = con->out_msg;
1562 struct ceph_msg_data_cursor *cursor = &msg->cursor;
1563 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
1564 int more = MSG_MORE | MSG_SENDPAGE_NOTLAST;
1567 dout("%s %p msg %p\n", __func__, con, msg);
1569 if (!msg->num_data_items)
1573 * Iterate through each page that contains data to be
1574 * written, and send as much as possible for each.
1576 * If we are calculating the data crc (the default), we will
1577 * need to map the page. If we have no pages, they have
1578 * been revoked, so use the zero page.
1580 crc = do_datacrc ? le32_to_cpu(msg->footer.data_crc) : 0;
1581 while (cursor->total_resid) {
1587 if (!cursor->resid) {
1588 ceph_msg_data_advance(cursor, 0);
1592 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
1593 if (length == cursor->total_resid)
1595 ret = ceph_tcp_sendpage(con->sock, page, page_offset, length,
1599 msg->footer.data_crc = cpu_to_le32(crc);
1603 if (do_datacrc && cursor->need_crc)
1604 crc = ceph_crc32c_page(crc, page, page_offset, length);
1605 ceph_msg_data_advance(cursor, (size_t)ret);
1608 dout("%s %p msg %p done\n", __func__, con, msg);
1610 /* prepare and queue up footer, too */
1612 msg->footer.data_crc = cpu_to_le32(crc);
1614 msg->footer.flags |= CEPH_MSG_FOOTER_NOCRC;
1615 con_out_kvec_reset(con);
1616 prepare_write_message_footer(con);
1618 return 1; /* must return > 0 to indicate success */
1624 static int write_partial_skip(struct ceph_connection *con)
1626 int more = MSG_MORE | MSG_SENDPAGE_NOTLAST;
1629 dout("%s %p %d left\n", __func__, con, con->out_skip);
1630 while (con->out_skip > 0) {
1631 size_t size = min(con->out_skip, (int) PAGE_SIZE);
1633 if (size == con->out_skip)
1635 ret = ceph_tcp_sendpage(con->sock, zero_page, 0, size, more);
1638 con->out_skip -= ret;
1646 * Prepare to read connection handshake, or an ack.
1648 static void prepare_read_banner(struct ceph_connection *con)
1650 dout("prepare_read_banner %p\n", con);
1651 con->in_base_pos = 0;
1654 static void prepare_read_connect(struct ceph_connection *con)
1656 dout("prepare_read_connect %p\n", con);
1657 con->in_base_pos = 0;
1660 static void prepare_read_ack(struct ceph_connection *con)
1662 dout("prepare_read_ack %p\n", con);
1663 con->in_base_pos = 0;
1666 static void prepare_read_seq(struct ceph_connection *con)
1668 dout("prepare_read_seq %p\n", con);
1669 con->in_base_pos = 0;
1670 con->in_tag = CEPH_MSGR_TAG_SEQ;
1673 static void prepare_read_tag(struct ceph_connection *con)
1675 dout("prepare_read_tag %p\n", con);
1676 con->in_base_pos = 0;
1677 con->in_tag = CEPH_MSGR_TAG_READY;
1680 static void prepare_read_keepalive_ack(struct ceph_connection *con)
1682 dout("prepare_read_keepalive_ack %p\n", con);
1683 con->in_base_pos = 0;
1687 * Prepare to read a message.
1689 static int prepare_read_message(struct ceph_connection *con)
1691 dout("prepare_read_message %p\n", con);
1692 BUG_ON(con->in_msg != NULL);
1693 con->in_base_pos = 0;
1694 con->in_front_crc = con->in_middle_crc = con->in_data_crc = 0;
1699 static int read_partial(struct ceph_connection *con,
1700 int end, int size, void *object)
1702 while (con->in_base_pos < end) {
1703 int left = end - con->in_base_pos;
1704 int have = size - left;
1705 int ret = ceph_tcp_recvmsg(con->sock, object + have, left);
1708 con->in_base_pos += ret;
1715 * Read all or part of the connect-side handshake on a new connection
1717 static int read_partial_banner(struct ceph_connection *con)
1723 dout("read_partial_banner %p at %d\n", con, con->in_base_pos);
1726 size = strlen(CEPH_BANNER);
1728 ret = read_partial(con, end, size, con->in_banner);
1732 size = sizeof (con->actual_peer_addr);
1734 ret = read_partial(con, end, size, &con->actual_peer_addr);
1737 ceph_decode_banner_addr(&con->actual_peer_addr);
1739 size = sizeof (con->peer_addr_for_me);
1741 ret = read_partial(con, end, size, &con->peer_addr_for_me);
1744 ceph_decode_banner_addr(&con->peer_addr_for_me);
1750 static int read_partial_connect(struct ceph_connection *con)
1756 dout("read_partial_connect %p at %d\n", con, con->in_base_pos);
1758 size = sizeof (con->in_reply);
1760 ret = read_partial(con, end, size, &con->in_reply);
1765 size = le32_to_cpu(con->in_reply.authorizer_len);
1766 if (size > con->auth->authorizer_reply_buf_len) {
1767 pr_err("authorizer reply too big: %d > %zu\n", size,
1768 con->auth->authorizer_reply_buf_len);
1774 ret = read_partial(con, end, size,
1775 con->auth->authorizer_reply_buf);
1780 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1781 con, (int)con->in_reply.tag,
1782 le32_to_cpu(con->in_reply.connect_seq),
1783 le32_to_cpu(con->in_reply.global_seq));
1789 * Verify the hello banner looks okay.
1791 static int verify_hello(struct ceph_connection *con)
1793 if (memcmp(con->in_banner, CEPH_BANNER, strlen(CEPH_BANNER))) {
1794 pr_err("connect to %s got bad banner\n",
1795 ceph_pr_addr(&con->peer_addr));
1796 con->error_msg = "protocol error, bad banner";
1802 static bool addr_is_blank(struct ceph_entity_addr *addr)
1804 struct sockaddr_storage ss = addr->in_addr; /* align */
1805 struct in_addr *addr4 = &((struct sockaddr_in *)&ss)->sin_addr;
1806 struct in6_addr *addr6 = &((struct sockaddr_in6 *)&ss)->sin6_addr;
1808 switch (ss.ss_family) {
1810 return addr4->s_addr == htonl(INADDR_ANY);
1812 return ipv6_addr_any(addr6);
1818 static int addr_port(struct ceph_entity_addr *addr)
1820 switch (get_unaligned(&addr->in_addr.ss_family)) {
1822 return ntohs(get_unaligned(&((struct sockaddr_in *)&addr->in_addr)->sin_port));
1824 return ntohs(get_unaligned(&((struct sockaddr_in6 *)&addr->in_addr)->sin6_port));
1829 static void addr_set_port(struct ceph_entity_addr *addr, int p)
1831 switch (get_unaligned(&addr->in_addr.ss_family)) {
1833 put_unaligned(htons(p), &((struct sockaddr_in *)&addr->in_addr)->sin_port);
1836 put_unaligned(htons(p), &((struct sockaddr_in6 *)&addr->in_addr)->sin6_port);
1842 * Unlike other *_pton function semantics, zero indicates success.
1844 static int ceph_pton(const char *str, size_t len, struct ceph_entity_addr *addr,
1845 char delim, const char **ipend)
1847 memset(&addr->in_addr, 0, sizeof(addr->in_addr));
1849 if (in4_pton(str, len, (u8 *)&((struct sockaddr_in *)&addr->in_addr)->sin_addr.s_addr, delim, ipend)) {
1850 put_unaligned(AF_INET, &addr->in_addr.ss_family);
1854 if (in6_pton(str, len, (u8 *)&((struct sockaddr_in6 *)&addr->in_addr)->sin6_addr.s6_addr, delim, ipend)) {
1855 put_unaligned(AF_INET6, &addr->in_addr.ss_family);
1863 * Extract hostname string and resolve using kernel DNS facility.
1865 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1866 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1867 struct ceph_entity_addr *addr, char delim, const char **ipend)
1869 const char *end, *delim_p;
1870 char *colon_p, *ip_addr = NULL;
1874 * The end of the hostname occurs immediately preceding the delimiter or
1875 * the port marker (':') where the delimiter takes precedence.
1877 delim_p = memchr(name, delim, namelen);
1878 colon_p = memchr(name, ':', namelen);
1880 if (delim_p && colon_p)
1881 end = delim_p < colon_p ? delim_p : colon_p;
1882 else if (!delim_p && colon_p)
1886 if (!end) /* case: hostname:/ */
1887 end = name + namelen;
1893 /* do dns_resolve upcall */
1894 ip_len = dns_query(current->nsproxy->net_ns,
1895 NULL, name, end - name, NULL, &ip_addr, NULL, false);
1897 ret = ceph_pton(ip_addr, ip_len, addr, -1, NULL);
1905 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1906 ret, ret ? "failed" : ceph_pr_addr(addr));
1911 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1912 struct ceph_entity_addr *addr, char delim, const char **ipend)
1919 * Parse a server name (IP or hostname). If a valid IP address is not found
1920 * then try to extract a hostname to resolve using userspace DNS upcall.
1922 static int ceph_parse_server_name(const char *name, size_t namelen,
1923 struct ceph_entity_addr *addr, char delim, const char **ipend)
1927 ret = ceph_pton(name, namelen, addr, delim, ipend);
1929 ret = ceph_dns_resolve_name(name, namelen, addr, delim, ipend);
1935 * Parse an ip[:port] list into an addr array. Use the default
1936 * monitor port if a port isn't specified.
1938 int ceph_parse_ips(const char *c, const char *end,
1939 struct ceph_entity_addr *addr,
1940 int max_count, int *count)
1942 int i, ret = -EINVAL;
1945 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1946 for (i = 0; i < max_count; i++) {
1956 ret = ceph_parse_server_name(p, end - p, &addr[i], delim, &ipend);
1965 dout("missing matching ']'\n");
1972 if (p < end && *p == ':') {
1975 while (p < end && *p >= '0' && *p <= '9') {
1976 port = (port * 10) + (*p - '0');
1980 port = CEPH_MON_PORT;
1981 else if (port > 65535)
1984 port = CEPH_MON_PORT;
1987 addr_set_port(&addr[i], port);
1988 addr[i].type = CEPH_ENTITY_ADDR_TYPE_LEGACY;
1990 dout("parse_ips got %s\n", ceph_pr_addr(&addr[i]));
2010 static int process_banner(struct ceph_connection *con)
2012 dout("process_banner on %p\n", con);
2014 if (verify_hello(con) < 0)
2018 * Make sure the other end is who we wanted. note that the other
2019 * end may not yet know their ip address, so if it's 0.0.0.0, give
2020 * them the benefit of the doubt.
2022 if (memcmp(&con->peer_addr, &con->actual_peer_addr,
2023 sizeof(con->peer_addr)) != 0 &&
2024 !(addr_is_blank(&con->actual_peer_addr) &&
2025 con->actual_peer_addr.nonce == con->peer_addr.nonce)) {
2026 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
2027 ceph_pr_addr(&con->peer_addr),
2028 (int)le32_to_cpu(con->peer_addr.nonce),
2029 ceph_pr_addr(&con->actual_peer_addr),
2030 (int)le32_to_cpu(con->actual_peer_addr.nonce));
2031 con->error_msg = "wrong peer at address";
2036 * did we learn our address?
2038 if (addr_is_blank(&con->msgr->inst.addr)) {
2039 int port = addr_port(&con->msgr->inst.addr);
2041 memcpy(&con->msgr->inst.addr.in_addr,
2042 &con->peer_addr_for_me.in_addr,
2043 sizeof(con->peer_addr_for_me.in_addr));
2044 addr_set_port(&con->msgr->inst.addr, port);
2045 encode_my_addr(con->msgr);
2046 dout("process_banner learned my addr is %s\n",
2047 ceph_pr_addr(&con->msgr->inst.addr));
2053 static int process_connect(struct ceph_connection *con)
2055 u64 sup_feat = from_msgr(con->msgr)->supported_features;
2056 u64 req_feat = from_msgr(con->msgr)->required_features;
2057 u64 server_feat = le64_to_cpu(con->in_reply.features);
2060 dout("process_connect on %p tag %d\n", con, (int)con->in_tag);
2063 int len = le32_to_cpu(con->in_reply.authorizer_len);
2066 * Any connection that defines ->get_authorizer()
2067 * should also define ->add_authorizer_challenge() and
2068 * ->verify_authorizer_reply().
2070 * See get_connect_authorizer().
2072 if (con->in_reply.tag == CEPH_MSGR_TAG_CHALLENGE_AUTHORIZER) {
2073 ret = con->ops->add_authorizer_challenge(
2074 con, con->auth->authorizer_reply_buf, len);
2078 con_out_kvec_reset(con);
2079 __prepare_write_connect(con);
2080 prepare_read_connect(con);
2085 ret = con->ops->verify_authorizer_reply(con);
2087 con->error_msg = "bad authorize reply";
2093 switch (con->in_reply.tag) {
2094 case CEPH_MSGR_TAG_FEATURES:
2095 pr_err("%s%lld %s feature set mismatch,"
2096 " my %llx < server's %llx, missing %llx\n",
2097 ENTITY_NAME(con->peer_name),
2098 ceph_pr_addr(&con->peer_addr),
2099 sup_feat, server_feat, server_feat & ~sup_feat);
2100 con->error_msg = "missing required protocol features";
2101 reset_connection(con);
2104 case CEPH_MSGR_TAG_BADPROTOVER:
2105 pr_err("%s%lld %s protocol version mismatch,"
2106 " my %d != server's %d\n",
2107 ENTITY_NAME(con->peer_name),
2108 ceph_pr_addr(&con->peer_addr),
2109 le32_to_cpu(con->out_connect.protocol_version),
2110 le32_to_cpu(con->in_reply.protocol_version));
2111 con->error_msg = "protocol version mismatch";
2112 reset_connection(con);
2115 case CEPH_MSGR_TAG_BADAUTHORIZER:
2117 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con,
2119 if (con->auth_retry == 2) {
2120 con->error_msg = "connect authorization failure";
2123 con_out_kvec_reset(con);
2124 ret = prepare_write_connect(con);
2127 prepare_read_connect(con);
2130 case CEPH_MSGR_TAG_RESETSESSION:
2132 * If we connected with a large connect_seq but the peer
2133 * has no record of a session with us (no connection, or
2134 * connect_seq == 0), they will send RESETSESION to indicate
2135 * that they must have reset their session, and may have
2138 dout("process_connect got RESET peer seq %u\n",
2139 le32_to_cpu(con->in_reply.connect_seq));
2140 pr_err("%s%lld %s connection reset\n",
2141 ENTITY_NAME(con->peer_name),
2142 ceph_pr_addr(&con->peer_addr));
2143 reset_connection(con);
2144 con_out_kvec_reset(con);
2145 ret = prepare_write_connect(con);
2148 prepare_read_connect(con);
2150 /* Tell ceph about it. */
2151 mutex_unlock(&con->mutex);
2152 pr_info("reset on %s%lld\n", ENTITY_NAME(con->peer_name));
2153 if (con->ops->peer_reset)
2154 con->ops->peer_reset(con);
2155 mutex_lock(&con->mutex);
2156 if (con->state != CON_STATE_NEGOTIATING)
2160 case CEPH_MSGR_TAG_RETRY_SESSION:
2162 * If we sent a smaller connect_seq than the peer has, try
2163 * again with a larger value.
2165 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2166 le32_to_cpu(con->out_connect.connect_seq),
2167 le32_to_cpu(con->in_reply.connect_seq));
2168 con->connect_seq = le32_to_cpu(con->in_reply.connect_seq);
2169 con_out_kvec_reset(con);
2170 ret = prepare_write_connect(con);
2173 prepare_read_connect(con);
2176 case CEPH_MSGR_TAG_RETRY_GLOBAL:
2178 * If we sent a smaller global_seq than the peer has, try
2179 * again with a larger value.
2181 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2182 con->peer_global_seq,
2183 le32_to_cpu(con->in_reply.global_seq));
2184 get_global_seq(con->msgr,
2185 le32_to_cpu(con->in_reply.global_seq));
2186 con_out_kvec_reset(con);
2187 ret = prepare_write_connect(con);
2190 prepare_read_connect(con);
2193 case CEPH_MSGR_TAG_SEQ:
2194 case CEPH_MSGR_TAG_READY:
2195 if (req_feat & ~server_feat) {
2196 pr_err("%s%lld %s protocol feature mismatch,"
2197 " my required %llx > server's %llx, need %llx\n",
2198 ENTITY_NAME(con->peer_name),
2199 ceph_pr_addr(&con->peer_addr),
2200 req_feat, server_feat, req_feat & ~server_feat);
2201 con->error_msg = "missing required protocol features";
2202 reset_connection(con);
2206 WARN_ON(con->state != CON_STATE_NEGOTIATING);
2207 con->state = CON_STATE_OPEN;
2208 con->auth_retry = 0; /* we authenticated; clear flag */
2209 con->peer_global_seq = le32_to_cpu(con->in_reply.global_seq);
2211 con->peer_features = server_feat;
2212 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2213 con->peer_global_seq,
2214 le32_to_cpu(con->in_reply.connect_seq),
2216 WARN_ON(con->connect_seq !=
2217 le32_to_cpu(con->in_reply.connect_seq));
2219 if (con->in_reply.flags & CEPH_MSG_CONNECT_LOSSY)
2220 con_flag_set(con, CON_FLAG_LOSSYTX);
2222 con->delay = 0; /* reset backoff memory */
2224 if (con->in_reply.tag == CEPH_MSGR_TAG_SEQ) {
2225 prepare_write_seq(con);
2226 prepare_read_seq(con);
2228 prepare_read_tag(con);
2232 case CEPH_MSGR_TAG_WAIT:
2234 * If there is a connection race (we are opening
2235 * connections to each other), one of us may just have
2236 * to WAIT. This shouldn't happen if we are the
2239 con->error_msg = "protocol error, got WAIT as client";
2243 con->error_msg = "protocol error, garbage tag during connect";
2251 * read (part of) an ack
2253 static int read_partial_ack(struct ceph_connection *con)
2255 int size = sizeof (con->in_temp_ack);
2258 return read_partial(con, end, size, &con->in_temp_ack);
2262 * We can finally discard anything that's been acked.
2264 static void process_ack(struct ceph_connection *con)
2267 u64 ack = le64_to_cpu(con->in_temp_ack);
2269 bool reconnect = (con->in_tag == CEPH_MSGR_TAG_SEQ);
2270 struct list_head *list = reconnect ? &con->out_queue : &con->out_sent;
2273 * In the reconnect case, con_fault() has requeued messages
2274 * in out_sent. We should cleanup old messages according to
2275 * the reconnect seq.
2277 while (!list_empty(list)) {
2278 m = list_first_entry(list, struct ceph_msg, list_head);
2279 if (reconnect && m->needs_out_seq)
2281 seq = le64_to_cpu(m->hdr.seq);
2284 dout("got ack for seq %llu type %d at %p\n", seq,
2285 le16_to_cpu(m->hdr.type), m);
2286 m->ack_stamp = jiffies;
2290 prepare_read_tag(con);
2294 static int read_partial_message_section(struct ceph_connection *con,
2295 struct kvec *section,
2296 unsigned int sec_len, u32 *crc)
2302 while (section->iov_len < sec_len) {
2303 BUG_ON(section->iov_base == NULL);
2304 left = sec_len - section->iov_len;
2305 ret = ceph_tcp_recvmsg(con->sock, (char *)section->iov_base +
2306 section->iov_len, left);
2309 section->iov_len += ret;
2311 if (section->iov_len == sec_len)
2312 *crc = crc32c(0, section->iov_base, section->iov_len);
2317 static int read_partial_msg_data(struct ceph_connection *con)
2319 struct ceph_msg *msg = con->in_msg;
2320 struct ceph_msg_data_cursor *cursor = &msg->cursor;
2321 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2328 if (!msg->num_data_items)
2332 crc = con->in_data_crc;
2333 while (cursor->total_resid) {
2334 if (!cursor->resid) {
2335 ceph_msg_data_advance(cursor, 0);
2339 page = ceph_msg_data_next(cursor, &page_offset, &length, NULL);
2340 ret = ceph_tcp_recvpage(con->sock, page, page_offset, length);
2343 con->in_data_crc = crc;
2349 crc = ceph_crc32c_page(crc, page, page_offset, ret);
2350 ceph_msg_data_advance(cursor, (size_t)ret);
2353 con->in_data_crc = crc;
2355 return 1; /* must return > 0 to indicate success */
2359 * read (part of) a message.
2361 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip);
2363 static int read_partial_message(struct ceph_connection *con)
2365 struct ceph_msg *m = con->in_msg;
2369 unsigned int front_len, middle_len, data_len;
2370 bool do_datacrc = !ceph_test_opt(from_msgr(con->msgr), NOCRC);
2371 bool need_sign = (con->peer_features & CEPH_FEATURE_MSG_AUTH);
2375 dout("read_partial_message con %p msg %p\n", con, m);
2378 size = sizeof (con->in_hdr);
2380 ret = read_partial(con, end, size, &con->in_hdr);
2384 crc = crc32c(0, &con->in_hdr, offsetof(struct ceph_msg_header, crc));
2385 if (cpu_to_le32(crc) != con->in_hdr.crc) {
2386 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
2387 crc, con->in_hdr.crc);
2391 front_len = le32_to_cpu(con->in_hdr.front_len);
2392 if (front_len > CEPH_MSG_MAX_FRONT_LEN)
2394 middle_len = le32_to_cpu(con->in_hdr.middle_len);
2395 if (middle_len > CEPH_MSG_MAX_MIDDLE_LEN)
2397 data_len = le32_to_cpu(con->in_hdr.data_len);
2398 if (data_len > CEPH_MSG_MAX_DATA_LEN)
2402 seq = le64_to_cpu(con->in_hdr.seq);
2403 if ((s64)seq - (s64)con->in_seq < 1) {
2404 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2405 ENTITY_NAME(con->peer_name),
2406 ceph_pr_addr(&con->peer_addr),
2407 seq, con->in_seq + 1);
2408 con->in_base_pos = -front_len - middle_len - data_len -
2410 con->in_tag = CEPH_MSGR_TAG_READY;
2412 } else if ((s64)seq - (s64)con->in_seq > 1) {
2413 pr_err("read_partial_message bad seq %lld expected %lld\n",
2414 seq, con->in_seq + 1);
2415 con->error_msg = "bad message sequence # for incoming message";
2419 /* allocate message? */
2423 dout("got hdr type %d front %d data %d\n", con->in_hdr.type,
2424 front_len, data_len);
2425 ret = ceph_con_in_msg_alloc(con, &skip);
2429 BUG_ON(!con->in_msg ^ skip);
2431 /* skip this message */
2432 dout("alloc_msg said skip message\n");
2433 con->in_base_pos = -front_len - middle_len - data_len -
2435 con->in_tag = CEPH_MSGR_TAG_READY;
2440 BUG_ON(!con->in_msg);
2441 BUG_ON(con->in_msg->con != con);
2443 m->front.iov_len = 0; /* haven't read it yet */
2445 m->middle->vec.iov_len = 0;
2447 /* prepare for data payload, if any */
2450 prepare_message_data(con->in_msg, data_len);
2454 ret = read_partial_message_section(con, &m->front, front_len,
2455 &con->in_front_crc);
2461 ret = read_partial_message_section(con, &m->middle->vec,
2463 &con->in_middle_crc);
2470 ret = read_partial_msg_data(con);
2476 size = sizeof_footer(con);
2478 ret = read_partial(con, end, size, &m->footer);
2483 m->footer.flags = m->old_footer.flags;
2487 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2488 m, front_len, m->footer.front_crc, middle_len,
2489 m->footer.middle_crc, data_len, m->footer.data_crc);
2492 if (con->in_front_crc != le32_to_cpu(m->footer.front_crc)) {
2493 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2494 m, con->in_front_crc, m->footer.front_crc);
2497 if (con->in_middle_crc != le32_to_cpu(m->footer.middle_crc)) {
2498 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2499 m, con->in_middle_crc, m->footer.middle_crc);
2503 (m->footer.flags & CEPH_MSG_FOOTER_NOCRC) == 0 &&
2504 con->in_data_crc != le32_to_cpu(m->footer.data_crc)) {
2505 pr_err("read_partial_message %p data crc %u != exp. %u\n", m,
2506 con->in_data_crc, le32_to_cpu(m->footer.data_crc));
2510 if (need_sign && con->ops->check_message_signature &&
2511 con->ops->check_message_signature(m)) {
2512 pr_err("read_partial_message %p signature check failed\n", m);
2516 return 1; /* done! */
2520 * Process message. This happens in the worker thread. The callback should
2521 * be careful not to do anything that waits on other incoming messages or it
2524 static void process_message(struct ceph_connection *con)
2526 struct ceph_msg *msg = con->in_msg;
2528 BUG_ON(con->in_msg->con != con);
2531 /* if first message, set peer_name */
2532 if (con->peer_name.type == 0)
2533 con->peer_name = msg->hdr.src;
2536 mutex_unlock(&con->mutex);
2538 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2539 msg, le64_to_cpu(msg->hdr.seq),
2540 ENTITY_NAME(msg->hdr.src),
2541 le16_to_cpu(msg->hdr.type),
2542 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
2543 le32_to_cpu(msg->hdr.front_len),
2544 le32_to_cpu(msg->hdr.data_len),
2545 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
2546 con->ops->dispatch(con, msg);
2548 mutex_lock(&con->mutex);
2551 static int read_keepalive_ack(struct ceph_connection *con)
2553 struct ceph_timespec ceph_ts;
2554 size_t size = sizeof(ceph_ts);
2555 int ret = read_partial(con, size, size, &ceph_ts);
2558 ceph_decode_timespec64(&con->last_keepalive_ack, &ceph_ts);
2559 prepare_read_tag(con);
2564 * Write something to the socket. Called in a worker thread when the
2565 * socket appears to be writeable and we have something ready to send.
2567 static int try_write(struct ceph_connection *con)
2571 dout("try_write start %p state %lu\n", con, con->state);
2572 if (con->state != CON_STATE_PREOPEN &&
2573 con->state != CON_STATE_CONNECTING &&
2574 con->state != CON_STATE_NEGOTIATING &&
2575 con->state != CON_STATE_OPEN)
2578 /* open the socket first? */
2579 if (con->state == CON_STATE_PREOPEN) {
2581 con->state = CON_STATE_CONNECTING;
2583 con_out_kvec_reset(con);
2584 prepare_write_banner(con);
2585 prepare_read_banner(con);
2587 BUG_ON(con->in_msg);
2588 con->in_tag = CEPH_MSGR_TAG_READY;
2589 dout("try_write initiating connect on %p new state %lu\n",
2591 ret = ceph_tcp_connect(con);
2593 con->error_msg = "connect error";
2599 dout("try_write out_kvec_bytes %d\n", con->out_kvec_bytes);
2602 /* kvec data queued? */
2603 if (con->out_kvec_left) {
2604 ret = write_partial_kvec(con);
2608 if (con->out_skip) {
2609 ret = write_partial_skip(con);
2616 if (con->out_msg_done) {
2617 ceph_msg_put(con->out_msg);
2618 con->out_msg = NULL; /* we're done with this one */
2622 ret = write_partial_message_data(con);
2624 goto more; /* we need to send the footer, too! */
2628 dout("try_write write_partial_message_data err %d\n",
2635 if (con->state == CON_STATE_OPEN) {
2636 if (con_flag_test_and_clear(con, CON_FLAG_KEEPALIVE_PENDING)) {
2637 prepare_write_keepalive(con);
2640 /* is anything else pending? */
2641 if (!list_empty(&con->out_queue)) {
2642 prepare_write_message(con);
2645 if (con->in_seq > con->in_seq_acked) {
2646 prepare_write_ack(con);
2651 /* Nothing to do! */
2652 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
2653 dout("try_write nothing else to write.\n");
2656 dout("try_write done on %p ret %d\n", con, ret);
2661 * Read what we can from the socket.
2663 static int try_read(struct ceph_connection *con)
2668 dout("try_read start on %p state %lu\n", con, con->state);
2669 if (con->state != CON_STATE_CONNECTING &&
2670 con->state != CON_STATE_NEGOTIATING &&
2671 con->state != CON_STATE_OPEN)
2676 dout("try_read tag %d in_base_pos %d\n", (int)con->in_tag,
2679 if (con->state == CON_STATE_CONNECTING) {
2680 dout("try_read connecting\n");
2681 ret = read_partial_banner(con);
2684 ret = process_banner(con);
2688 con->state = CON_STATE_NEGOTIATING;
2691 * Received banner is good, exchange connection info.
2692 * Do not reset out_kvec, as sending our banner raced
2693 * with receiving peer banner after connect completed.
2695 ret = prepare_write_connect(con);
2698 prepare_read_connect(con);
2700 /* Send connection info before awaiting response */
2704 if (con->state == CON_STATE_NEGOTIATING) {
2705 dout("try_read negotiating\n");
2706 ret = read_partial_connect(con);
2709 ret = process_connect(con);
2715 WARN_ON(con->state != CON_STATE_OPEN);
2717 if (con->in_base_pos < 0) {
2719 * skipping + discarding content.
2721 ret = ceph_tcp_recvmsg(con->sock, NULL, -con->in_base_pos);
2724 dout("skipped %d / %d bytes\n", ret, -con->in_base_pos);
2725 con->in_base_pos += ret;
2726 if (con->in_base_pos)
2729 if (con->in_tag == CEPH_MSGR_TAG_READY) {
2733 ret = ceph_tcp_recvmsg(con->sock, &con->in_tag, 1);
2736 dout("try_read got tag %d\n", (int)con->in_tag);
2737 switch (con->in_tag) {
2738 case CEPH_MSGR_TAG_MSG:
2739 prepare_read_message(con);
2741 case CEPH_MSGR_TAG_ACK:
2742 prepare_read_ack(con);
2744 case CEPH_MSGR_TAG_KEEPALIVE2_ACK:
2745 prepare_read_keepalive_ack(con);
2747 case CEPH_MSGR_TAG_CLOSE:
2748 con_close_socket(con);
2749 con->state = CON_STATE_CLOSED;
2755 if (con->in_tag == CEPH_MSGR_TAG_MSG) {
2756 ret = read_partial_message(con);
2760 con->error_msg = "bad crc/signature";
2766 con->error_msg = "io error";
2771 if (con->in_tag == CEPH_MSGR_TAG_READY)
2773 process_message(con);
2774 if (con->state == CON_STATE_OPEN)
2775 prepare_read_tag(con);
2778 if (con->in_tag == CEPH_MSGR_TAG_ACK ||
2779 con->in_tag == CEPH_MSGR_TAG_SEQ) {
2781 * the final handshake seq exchange is semantically
2782 * equivalent to an ACK
2784 ret = read_partial_ack(con);
2790 if (con->in_tag == CEPH_MSGR_TAG_KEEPALIVE2_ACK) {
2791 ret = read_keepalive_ack(con);
2798 dout("try_read done on %p ret %d\n", con, ret);
2802 pr_err("try_read bad con->in_tag = %d\n", (int)con->in_tag);
2803 con->error_msg = "protocol error, garbage tag";
2810 * Atomically queue work on a connection after the specified delay.
2811 * Bump @con reference to avoid races with connection teardown.
2812 * Returns 0 if work was queued, or an error code otherwise.
2814 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
2816 if (!con->ops->get(con)) {
2817 dout("%s %p ref count 0\n", __func__, con);
2821 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
2822 dout("%s %p - already queued\n", __func__, con);
2827 dout("%s %p %lu\n", __func__, con, delay);
2831 static void queue_con(struct ceph_connection *con)
2833 (void) queue_con_delay(con, 0);
2836 static void cancel_con(struct ceph_connection *con)
2838 if (cancel_delayed_work(&con->work)) {
2839 dout("%s %p\n", __func__, con);
2844 static bool con_sock_closed(struct ceph_connection *con)
2846 if (!con_flag_test_and_clear(con, CON_FLAG_SOCK_CLOSED))
2850 case CON_STATE_ ## x: \
2851 con->error_msg = "socket closed (con state " #x ")"; \
2854 switch (con->state) {
2862 pr_warn("%s con %p unrecognized state %lu\n",
2863 __func__, con, con->state);
2864 con->error_msg = "unrecognized con state";
2873 static bool con_backoff(struct ceph_connection *con)
2877 if (!con_flag_test_and_clear(con, CON_FLAG_BACKOFF))
2880 ret = queue_con_delay(con, round_jiffies_relative(con->delay));
2882 dout("%s: con %p FAILED to back off %lu\n", __func__,
2884 BUG_ON(ret == -ENOENT);
2885 con_flag_set(con, CON_FLAG_BACKOFF);
2891 /* Finish fault handling; con->mutex must *not* be held here */
2893 static void con_fault_finish(struct ceph_connection *con)
2895 dout("%s %p\n", __func__, con);
2898 * in case we faulted due to authentication, invalidate our
2899 * current tickets so that we can get new ones.
2901 if (con->auth_retry) {
2902 dout("auth_retry %d, invalidating\n", con->auth_retry);
2903 if (con->ops->invalidate_authorizer)
2904 con->ops->invalidate_authorizer(con);
2905 con->auth_retry = 0;
2908 if (con->ops->fault)
2909 con->ops->fault(con);
2913 * Do some work on a connection. Drop a connection ref when we're done.
2915 static void ceph_con_workfn(struct work_struct *work)
2917 struct ceph_connection *con = container_of(work, struct ceph_connection,
2921 mutex_lock(&con->mutex);
2925 if ((fault = con_sock_closed(con))) {
2926 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
2929 if (con_backoff(con)) {
2930 dout("%s: con %p BACKOFF\n", __func__, con);
2933 if (con->state == CON_STATE_STANDBY) {
2934 dout("%s: con %p STANDBY\n", __func__, con);
2937 if (con->state == CON_STATE_CLOSED) {
2938 dout("%s: con %p CLOSED\n", __func__, con);
2942 if (con->state == CON_STATE_PREOPEN) {
2943 dout("%s: con %p PREOPEN\n", __func__, con);
2947 ret = try_read(con);
2951 if (!con->error_msg)
2952 con->error_msg = "socket error on read";
2957 ret = try_write(con);
2961 if (!con->error_msg)
2962 con->error_msg = "socket error on write";
2966 break; /* If we make it to here, we're done */
2970 mutex_unlock(&con->mutex);
2973 con_fault_finish(con);
2979 * Generic error/fault handler. A retry mechanism is used with
2980 * exponential backoff
2982 static void con_fault(struct ceph_connection *con)
2984 dout("fault %p state %lu to peer %s\n",
2985 con, con->state, ceph_pr_addr(&con->peer_addr));
2987 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
2988 ceph_pr_addr(&con->peer_addr), con->error_msg);
2989 con->error_msg = NULL;
2991 WARN_ON(con->state != CON_STATE_CONNECTING &&
2992 con->state != CON_STATE_NEGOTIATING &&
2993 con->state != CON_STATE_OPEN);
2995 con_close_socket(con);
2997 if (con_flag_test(con, CON_FLAG_LOSSYTX)) {
2998 dout("fault on LOSSYTX channel, marking CLOSED\n");
2999 con->state = CON_STATE_CLOSED;
3004 BUG_ON(con->in_msg->con != con);
3005 ceph_msg_put(con->in_msg);
3009 /* Requeue anything that hasn't been acked */
3010 list_splice_init(&con->out_sent, &con->out_queue);
3012 /* If there are no messages queued or keepalive pending, place
3013 * the connection in a STANDBY state */
3014 if (list_empty(&con->out_queue) &&
3015 !con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING)) {
3016 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
3017 con_flag_clear(con, CON_FLAG_WRITE_PENDING);
3018 con->state = CON_STATE_STANDBY;
3020 /* retry after a delay. */
3021 con->state = CON_STATE_PREOPEN;
3022 if (con->delay == 0)
3023 con->delay = BASE_DELAY_INTERVAL;
3024 else if (con->delay < MAX_DELAY_INTERVAL)
3026 con_flag_set(con, CON_FLAG_BACKOFF);
3032 void ceph_messenger_reset_nonce(struct ceph_messenger *msgr)
3034 u32 nonce = le32_to_cpu(msgr->inst.addr.nonce) + 1000000;
3035 msgr->inst.addr.nonce = cpu_to_le32(nonce);
3036 encode_my_addr(msgr);
3040 * initialize a new messenger instance
3042 void ceph_messenger_init(struct ceph_messenger *msgr,
3043 struct ceph_entity_addr *myaddr)
3045 spin_lock_init(&msgr->global_seq_lock);
3048 msgr->inst.addr = *myaddr;
3050 /* select a random nonce */
3051 msgr->inst.addr.type = 0;
3052 get_random_bytes(&msgr->inst.addr.nonce, sizeof(msgr->inst.addr.nonce));
3053 encode_my_addr(msgr);
3055 atomic_set(&msgr->stopping, 0);
3056 write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
3058 dout("%s %p\n", __func__, msgr);
3060 EXPORT_SYMBOL(ceph_messenger_init);
3062 void ceph_messenger_fini(struct ceph_messenger *msgr)
3064 put_net(read_pnet(&msgr->net));
3066 EXPORT_SYMBOL(ceph_messenger_fini);
3068 static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con)
3071 msg->con->ops->put(msg->con);
3073 msg->con = con ? con->ops->get(con) : NULL;
3074 BUG_ON(msg->con != con);
3077 static void clear_standby(struct ceph_connection *con)
3079 /* come back from STANDBY? */
3080 if (con->state == CON_STATE_STANDBY) {
3081 dout("clear_standby %p and ++connect_seq\n", con);
3082 con->state = CON_STATE_PREOPEN;
3084 WARN_ON(con_flag_test(con, CON_FLAG_WRITE_PENDING));
3085 WARN_ON(con_flag_test(con, CON_FLAG_KEEPALIVE_PENDING));
3090 * Queue up an outgoing message on the given connection.
3092 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
3095 msg->hdr.src = con->msgr->inst.name;
3096 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
3097 msg->needs_out_seq = true;
3099 mutex_lock(&con->mutex);
3101 if (con->state == CON_STATE_CLOSED) {
3102 dout("con_send %p closed, dropping %p\n", con, msg);
3104 mutex_unlock(&con->mutex);
3108 msg_con_set(msg, con);
3110 BUG_ON(!list_empty(&msg->list_head));
3111 list_add_tail(&msg->list_head, &con->out_queue);
3112 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
3113 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
3114 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
3115 le32_to_cpu(msg->hdr.front_len),
3116 le32_to_cpu(msg->hdr.middle_len),
3117 le32_to_cpu(msg->hdr.data_len));
3120 mutex_unlock(&con->mutex);
3122 /* if there wasn't anything waiting to send before, queue
3124 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3127 EXPORT_SYMBOL(ceph_con_send);
3130 * Revoke a message that was previously queued for send
3132 void ceph_msg_revoke(struct ceph_msg *msg)
3134 struct ceph_connection *con = msg->con;
3137 dout("%s msg %p null con\n", __func__, msg);
3138 return; /* Message not in our possession */
3141 mutex_lock(&con->mutex);
3142 if (!list_empty(&msg->list_head)) {
3143 dout("%s %p msg %p - was on queue\n", __func__, con, msg);
3144 list_del_init(&msg->list_head);
3149 if (con->out_msg == msg) {
3150 BUG_ON(con->out_skip);
3152 if (con->out_msg_done) {
3153 con->out_skip += con_out_kvec_skip(con);
3155 BUG_ON(!msg->data_length);
3156 con->out_skip += sizeof_footer(con);
3158 /* data, middle, front */
3159 if (msg->data_length)
3160 con->out_skip += msg->cursor.total_resid;
3162 con->out_skip += con_out_kvec_skip(con);
3163 con->out_skip += con_out_kvec_skip(con);
3165 dout("%s %p msg %p - was sending, will write %d skip %d\n",
3166 __func__, con, msg, con->out_kvec_bytes, con->out_skip);
3168 con->out_msg = NULL;
3172 mutex_unlock(&con->mutex);
3176 * Revoke a message that we may be reading data into
3178 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
3180 struct ceph_connection *con = msg->con;
3183 dout("%s msg %p null con\n", __func__, msg);
3184 return; /* Message not in our possession */
3187 mutex_lock(&con->mutex);
3188 if (con->in_msg == msg) {
3189 unsigned int front_len = le32_to_cpu(con->in_hdr.front_len);
3190 unsigned int middle_len = le32_to_cpu(con->in_hdr.middle_len);
3191 unsigned int data_len = le32_to_cpu(con->in_hdr.data_len);
3193 /* skip rest of message */
3194 dout("%s %p msg %p revoked\n", __func__, con, msg);
3195 con->in_base_pos = con->in_base_pos -
3196 sizeof(struct ceph_msg_header) -
3200 sizeof(struct ceph_msg_footer);
3201 ceph_msg_put(con->in_msg);
3203 con->in_tag = CEPH_MSGR_TAG_READY;
3206 dout("%s %p in_msg %p msg %p no-op\n",
3207 __func__, con, con->in_msg, msg);
3209 mutex_unlock(&con->mutex);
3213 * Queue a keepalive byte to ensure the tcp connection is alive.
3215 void ceph_con_keepalive(struct ceph_connection *con)
3217 dout("con_keepalive %p\n", con);
3218 mutex_lock(&con->mutex);
3220 con_flag_set(con, CON_FLAG_KEEPALIVE_PENDING);
3221 mutex_unlock(&con->mutex);
3223 if (con_flag_test_and_set(con, CON_FLAG_WRITE_PENDING) == 0)
3226 EXPORT_SYMBOL(ceph_con_keepalive);
3228 bool ceph_con_keepalive_expired(struct ceph_connection *con,
3229 unsigned long interval)
3232 (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
3233 struct timespec64 now;
3234 struct timespec64 ts;
3235 ktime_get_real_ts64(&now);
3236 jiffies_to_timespec64(interval, &ts);
3237 ts = timespec64_add(con->last_keepalive_ack, ts);
3238 return timespec64_compare(&now, &ts) >= 0;
3243 static struct ceph_msg_data *ceph_msg_data_add(struct ceph_msg *msg)
3245 BUG_ON(msg->num_data_items >= msg->max_data_items);
3246 return &msg->data[msg->num_data_items++];
3249 static void ceph_msg_data_destroy(struct ceph_msg_data *data)
3251 if (data->type == CEPH_MSG_DATA_PAGES && data->own_pages) {
3252 int num_pages = calc_pages_for(data->alignment, data->length);
3253 ceph_release_page_vector(data->pages, num_pages);
3254 } else if (data->type == CEPH_MSG_DATA_PAGELIST) {
3255 ceph_pagelist_release(data->pagelist);
3259 void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
3260 size_t length, size_t alignment, bool own_pages)
3262 struct ceph_msg_data *data;
3267 data = ceph_msg_data_add(msg);
3268 data->type = CEPH_MSG_DATA_PAGES;
3269 data->pages = pages;
3270 data->length = length;
3271 data->alignment = alignment & ~PAGE_MASK;
3272 data->own_pages = own_pages;
3274 msg->data_length += length;
3276 EXPORT_SYMBOL(ceph_msg_data_add_pages);
3278 void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
3279 struct ceph_pagelist *pagelist)
3281 struct ceph_msg_data *data;
3284 BUG_ON(!pagelist->length);
3286 data = ceph_msg_data_add(msg);
3287 data->type = CEPH_MSG_DATA_PAGELIST;
3288 refcount_inc(&pagelist->refcnt);
3289 data->pagelist = pagelist;
3291 msg->data_length += pagelist->length;
3293 EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
3296 void ceph_msg_data_add_bio(struct ceph_msg *msg, struct ceph_bio_iter *bio_pos,
3299 struct ceph_msg_data *data;
3301 data = ceph_msg_data_add(msg);
3302 data->type = CEPH_MSG_DATA_BIO;
3303 data->bio_pos = *bio_pos;
3304 data->bio_length = length;
3306 msg->data_length += length;
3308 EXPORT_SYMBOL(ceph_msg_data_add_bio);
3309 #endif /* CONFIG_BLOCK */
3311 void ceph_msg_data_add_bvecs(struct ceph_msg *msg,
3312 struct ceph_bvec_iter *bvec_pos)
3314 struct ceph_msg_data *data;
3316 data = ceph_msg_data_add(msg);
3317 data->type = CEPH_MSG_DATA_BVECS;
3318 data->bvec_pos = *bvec_pos;
3320 msg->data_length += bvec_pos->iter.bi_size;
3322 EXPORT_SYMBOL(ceph_msg_data_add_bvecs);
3325 * construct a new message with given type, size
3326 * the new msg has a ref count of 1.
3328 struct ceph_msg *ceph_msg_new2(int type, int front_len, int max_data_items,
3329 gfp_t flags, bool can_fail)
3333 m = kmem_cache_zalloc(ceph_msg_cache, flags);
3337 m->hdr.type = cpu_to_le16(type);
3338 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
3339 m->hdr.front_len = cpu_to_le32(front_len);
3341 INIT_LIST_HEAD(&m->list_head);
3342 kref_init(&m->kref);
3346 m->front.iov_base = ceph_kvmalloc(front_len, flags);
3347 if (m->front.iov_base == NULL) {
3348 dout("ceph_msg_new can't allocate %d bytes\n",
3353 m->front.iov_base = NULL;
3355 m->front_alloc_len = m->front.iov_len = front_len;
3357 if (max_data_items) {
3358 m->data = kmalloc_array(max_data_items, sizeof(*m->data),
3363 m->max_data_items = max_data_items;
3366 dout("ceph_msg_new %p front %d\n", m, front_len);
3373 pr_err("msg_new can't create type %d front %d\n", type,
3377 dout("msg_new can't create type %d front %d\n", type,
3382 EXPORT_SYMBOL(ceph_msg_new2);
3384 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
3387 return ceph_msg_new2(type, front_len, 0, flags, can_fail);
3389 EXPORT_SYMBOL(ceph_msg_new);
3392 * Allocate "middle" portion of a message, if it is needed and wasn't
3393 * allocated by alloc_msg. This allows us to read a small fixed-size
3394 * per-type header in the front and then gracefully fail (i.e.,
3395 * propagate the error to the caller based on info in the front) when
3396 * the middle is too large.
3398 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
3400 int type = le16_to_cpu(msg->hdr.type);
3401 int middle_len = le32_to_cpu(msg->hdr.middle_len);
3403 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
3404 ceph_msg_type_name(type), middle_len);
3405 BUG_ON(!middle_len);
3406 BUG_ON(msg->middle);
3408 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
3415 * Allocate a message for receiving an incoming message on a
3416 * connection, and save the result in con->in_msg. Uses the
3417 * connection's private alloc_msg op if available.
3419 * Returns 0 on success, or a negative error code.
3421 * On success, if we set *skip = 1:
3422 * - the next message should be skipped and ignored.
3423 * - con->in_msg == NULL
3424 * or if we set *skip = 0:
3425 * - con->in_msg is non-null.
3426 * On error (ENOMEM, EAGAIN, ...),
3427 * - con->in_msg == NULL
3429 static int ceph_con_in_msg_alloc(struct ceph_connection *con, int *skip)
3431 struct ceph_msg_header *hdr = &con->in_hdr;
3432 int middle_len = le32_to_cpu(hdr->middle_len);
3433 struct ceph_msg *msg;
3436 BUG_ON(con->in_msg != NULL);
3437 BUG_ON(!con->ops->alloc_msg);
3439 mutex_unlock(&con->mutex);
3440 msg = con->ops->alloc_msg(con, hdr, skip);
3441 mutex_lock(&con->mutex);
3442 if (con->state != CON_STATE_OPEN) {
3449 msg_con_set(msg, con);
3453 * Null message pointer means either we should skip
3454 * this message or we couldn't allocate memory. The
3455 * former is not an error.
3460 con->error_msg = "error allocating memory for incoming message";
3463 memcpy(&con->in_msg->hdr, &con->in_hdr, sizeof(con->in_hdr));
3465 if (middle_len && !con->in_msg->middle) {
3466 ret = ceph_alloc_middle(con, con->in_msg);
3468 ceph_msg_put(con->in_msg);
3478 * Free a generically kmalloc'd message.
3480 static void ceph_msg_free(struct ceph_msg *m)
3482 dout("%s %p\n", __func__, m);
3483 kvfree(m->front.iov_base);
3485 kmem_cache_free(ceph_msg_cache, m);
3488 static void ceph_msg_release(struct kref *kref)
3490 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
3493 dout("%s %p\n", __func__, m);
3494 WARN_ON(!list_empty(&m->list_head));
3496 msg_con_set(m, NULL);
3498 /* drop middle, data, if any */
3500 ceph_buffer_put(m->middle);
3504 for (i = 0; i < m->num_data_items; i++)
3505 ceph_msg_data_destroy(&m->data[i]);
3508 ceph_msgpool_put(m->pool, m);
3513 struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
3515 dout("%s %p (was %d)\n", __func__, msg,
3516 kref_read(&msg->kref));
3517 kref_get(&msg->kref);
3520 EXPORT_SYMBOL(ceph_msg_get);
3522 void ceph_msg_put(struct ceph_msg *msg)
3524 dout("%s %p (was %d)\n", __func__, msg,
3525 kref_read(&msg->kref));
3526 kref_put(&msg->kref, ceph_msg_release);
3528 EXPORT_SYMBOL(ceph_msg_put);
3530 void ceph_msg_dump(struct ceph_msg *msg)
3532 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
3533 msg->front_alloc_len, msg->data_length);
3534 print_hex_dump(KERN_DEBUG, "header: ",
3535 DUMP_PREFIX_OFFSET, 16, 1,
3536 &msg->hdr, sizeof(msg->hdr), true);
3537 print_hex_dump(KERN_DEBUG, " front: ",
3538 DUMP_PREFIX_OFFSET, 16, 1,
3539 msg->front.iov_base, msg->front.iov_len, true);
3541 print_hex_dump(KERN_DEBUG, "middle: ",
3542 DUMP_PREFIX_OFFSET, 16, 1,
3543 msg->middle->vec.iov_base,
3544 msg->middle->vec.iov_len, true);
3545 print_hex_dump(KERN_DEBUG, "footer: ",
3546 DUMP_PREFIX_OFFSET, 16, 1,
3547 &msg->footer, sizeof(msg->footer), true);
3549 EXPORT_SYMBOL(ceph_msg_dump);