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 */
85 static bool con_flag_valid(unsigned long con_flag)
88 case CEPH_CON_F_LOSSYTX:
89 case CEPH_CON_F_KEEPALIVE_PENDING:
90 case CEPH_CON_F_WRITE_PENDING:
91 case CEPH_CON_F_SOCK_CLOSED:
92 case CEPH_CON_F_BACKOFF:
99 void ceph_con_flag_clear(struct ceph_connection *con, unsigned long con_flag)
101 BUG_ON(!con_flag_valid(con_flag));
103 clear_bit(con_flag, &con->flags);
106 void ceph_con_flag_set(struct ceph_connection *con, unsigned long con_flag)
108 BUG_ON(!con_flag_valid(con_flag));
110 set_bit(con_flag, &con->flags);
113 bool ceph_con_flag_test(struct ceph_connection *con, unsigned long con_flag)
115 BUG_ON(!con_flag_valid(con_flag));
117 return test_bit(con_flag, &con->flags);
120 bool ceph_con_flag_test_and_clear(struct ceph_connection *con,
121 unsigned long con_flag)
123 BUG_ON(!con_flag_valid(con_flag));
125 return test_and_clear_bit(con_flag, &con->flags);
128 bool ceph_con_flag_test_and_set(struct ceph_connection *con,
129 unsigned long con_flag)
131 BUG_ON(!con_flag_valid(con_flag));
133 return test_and_set_bit(con_flag, &con->flags);
136 /* Slab caches for frequently-allocated structures */
138 static struct kmem_cache *ceph_msg_cache;
140 #ifdef CONFIG_LOCKDEP
141 static struct lock_class_key socket_class;
144 static void queue_con(struct ceph_connection *con);
145 static void cancel_con(struct ceph_connection *con);
146 static void ceph_con_workfn(struct work_struct *);
147 static void con_fault(struct ceph_connection *con);
150 * Nicely render a sockaddr as a string. An array of formatted
151 * strings is used, to approximate reentrancy.
153 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
154 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
155 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
156 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
158 static char addr_str[ADDR_STR_COUNT][MAX_ADDR_STR_LEN];
159 static atomic_t addr_str_seq = ATOMIC_INIT(0);
161 struct page *ceph_zero_page; /* used in certain error cases */
163 const char *ceph_pr_addr(const struct ceph_entity_addr *addr)
167 struct sockaddr_storage ss = addr->in_addr; /* align */
168 struct sockaddr_in *in4 = (struct sockaddr_in *)&ss;
169 struct sockaddr_in6 *in6 = (struct sockaddr_in6 *)&ss;
171 i = atomic_inc_return(&addr_str_seq) & ADDR_STR_COUNT_MASK;
174 switch (ss.ss_family) {
176 snprintf(s, MAX_ADDR_STR_LEN, "(%d)%pI4:%hu",
177 le32_to_cpu(addr->type), &in4->sin_addr,
178 ntohs(in4->sin_port));
182 snprintf(s, MAX_ADDR_STR_LEN, "(%d)[%pI6c]:%hu",
183 le32_to_cpu(addr->type), &in6->sin6_addr,
184 ntohs(in6->sin6_port));
188 snprintf(s, MAX_ADDR_STR_LEN, "(unknown sockaddr family %hu)",
194 EXPORT_SYMBOL(ceph_pr_addr);
196 void ceph_encode_my_addr(struct ceph_messenger *msgr)
198 if (!ceph_msgr2(from_msgr(msgr))) {
199 memcpy(&msgr->my_enc_addr, &msgr->inst.addr,
200 sizeof(msgr->my_enc_addr));
201 ceph_encode_banner_addr(&msgr->my_enc_addr);
206 * work queue for all reading and writing to/from the socket.
208 static struct workqueue_struct *ceph_msgr_wq;
210 static int ceph_msgr_slab_init(void)
212 BUG_ON(ceph_msg_cache);
213 ceph_msg_cache = KMEM_CACHE(ceph_msg, 0);
220 static void ceph_msgr_slab_exit(void)
222 BUG_ON(!ceph_msg_cache);
223 kmem_cache_destroy(ceph_msg_cache);
224 ceph_msg_cache = NULL;
227 static void _ceph_msgr_exit(void)
230 destroy_workqueue(ceph_msgr_wq);
234 BUG_ON(!ceph_zero_page);
235 put_page(ceph_zero_page);
236 ceph_zero_page = NULL;
238 ceph_msgr_slab_exit();
241 int __init ceph_msgr_init(void)
243 if (ceph_msgr_slab_init())
246 BUG_ON(ceph_zero_page);
247 ceph_zero_page = ZERO_PAGE(0);
248 get_page(ceph_zero_page);
251 * The number of active work items is limited by the number of
252 * connections, so leave @max_active at default.
254 ceph_msgr_wq = alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM, 0);
258 pr_err("msgr_init failed to create workqueue\n");
264 void ceph_msgr_exit(void)
266 BUG_ON(ceph_msgr_wq == NULL);
271 void ceph_msgr_flush(void)
273 flush_workqueue(ceph_msgr_wq);
275 EXPORT_SYMBOL(ceph_msgr_flush);
277 /* Connection socket state transition functions */
279 static void con_sock_state_init(struct ceph_connection *con)
283 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
284 if (WARN_ON(old_state != CON_SOCK_STATE_NEW))
285 printk("%s: unexpected old state %d\n", __func__, old_state);
286 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
287 CON_SOCK_STATE_CLOSED);
290 static void con_sock_state_connecting(struct ceph_connection *con)
294 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTING);
295 if (WARN_ON(old_state != CON_SOCK_STATE_CLOSED))
296 printk("%s: unexpected old state %d\n", __func__, old_state);
297 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
298 CON_SOCK_STATE_CONNECTING);
301 static void con_sock_state_connected(struct ceph_connection *con)
305 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CONNECTED);
306 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING))
307 printk("%s: unexpected old state %d\n", __func__, old_state);
308 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
309 CON_SOCK_STATE_CONNECTED);
312 static void con_sock_state_closing(struct ceph_connection *con)
316 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSING);
317 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTING &&
318 old_state != CON_SOCK_STATE_CONNECTED &&
319 old_state != CON_SOCK_STATE_CLOSING))
320 printk("%s: unexpected old state %d\n", __func__, old_state);
321 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
322 CON_SOCK_STATE_CLOSING);
325 static void con_sock_state_closed(struct ceph_connection *con)
329 old_state = atomic_xchg(&con->sock_state, CON_SOCK_STATE_CLOSED);
330 if (WARN_ON(old_state != CON_SOCK_STATE_CONNECTED &&
331 old_state != CON_SOCK_STATE_CLOSING &&
332 old_state != CON_SOCK_STATE_CONNECTING &&
333 old_state != CON_SOCK_STATE_CLOSED))
334 printk("%s: unexpected old state %d\n", __func__, old_state);
335 dout("%s con %p sock %d -> %d\n", __func__, con, old_state,
336 CON_SOCK_STATE_CLOSED);
340 * socket callback functions
343 /* data available on socket, or listen socket received a connect */
344 static void ceph_sock_data_ready(struct sock *sk)
346 struct ceph_connection *con = sk->sk_user_data;
347 if (atomic_read(&con->msgr->stopping)) {
351 if (sk->sk_state != TCP_CLOSE_WAIT) {
352 dout("%s %p state = %d, queueing work\n", __func__,
358 /* socket has buffer space for writing */
359 static void ceph_sock_write_space(struct sock *sk)
361 struct ceph_connection *con = sk->sk_user_data;
363 /* only queue to workqueue if there is data we want to write,
364 * and there is sufficient space in the socket buffer to accept
365 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
366 * doesn't get called again until try_write() fills the socket
367 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
368 * and net/core/stream.c:sk_stream_write_space().
370 if (ceph_con_flag_test(con, CEPH_CON_F_WRITE_PENDING)) {
371 if (sk_stream_is_writeable(sk)) {
372 dout("%s %p queueing write work\n", __func__, con);
373 clear_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
377 dout("%s %p nothing to write\n", __func__, con);
381 /* socket's state has changed */
382 static void ceph_sock_state_change(struct sock *sk)
384 struct ceph_connection *con = sk->sk_user_data;
386 dout("%s %p state = %d sk_state = %u\n", __func__,
387 con, con->state, sk->sk_state);
389 switch (sk->sk_state) {
391 dout("%s TCP_CLOSE\n", __func__);
394 dout("%s TCP_CLOSE_WAIT\n", __func__);
395 con_sock_state_closing(con);
396 ceph_con_flag_set(con, CEPH_CON_F_SOCK_CLOSED);
399 case TCP_ESTABLISHED:
400 dout("%s TCP_ESTABLISHED\n", __func__);
401 con_sock_state_connected(con);
404 default: /* Everything else is uninteresting */
410 * set up socket callbacks
412 static void set_sock_callbacks(struct socket *sock,
413 struct ceph_connection *con)
415 struct sock *sk = sock->sk;
416 sk->sk_user_data = con;
417 sk->sk_data_ready = ceph_sock_data_ready;
418 sk->sk_write_space = ceph_sock_write_space;
419 sk->sk_state_change = ceph_sock_state_change;
428 * initiate connection to a remote socket.
430 int ceph_tcp_connect(struct ceph_connection *con)
432 struct sockaddr_storage ss = con->peer_addr.in_addr; /* align */
434 unsigned int noio_flag;
437 dout("%s con %p peer_addr %s\n", __func__, con,
438 ceph_pr_addr(&con->peer_addr));
441 /* sock_create_kern() allocates with GFP_KERNEL */
442 noio_flag = memalloc_noio_save();
443 ret = sock_create_kern(read_pnet(&con->msgr->net), ss.ss_family,
444 SOCK_STREAM, IPPROTO_TCP, &sock);
445 memalloc_noio_restore(noio_flag);
448 sock->sk->sk_allocation = GFP_NOFS;
450 #ifdef CONFIG_LOCKDEP
451 lockdep_set_class(&sock->sk->sk_lock, &socket_class);
454 set_sock_callbacks(sock, con);
456 con_sock_state_connecting(con);
457 ret = sock->ops->connect(sock, (struct sockaddr *)&ss, sizeof(ss),
459 if (ret == -EINPROGRESS) {
460 dout("connect %s EINPROGRESS sk_state = %u\n",
461 ceph_pr_addr(&con->peer_addr),
463 } else if (ret < 0) {
464 pr_err("connect %s error %d\n",
465 ceph_pr_addr(&con->peer_addr), ret);
470 if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY))
471 tcp_sock_set_nodelay(sock->sk);
478 * Shutdown/close the socket for the given connection.
480 int ceph_con_close_socket(struct ceph_connection *con)
484 dout("%s con %p sock %p\n", __func__, con, con->sock);
486 rc = con->sock->ops->shutdown(con->sock, SHUT_RDWR);
487 sock_release(con->sock);
492 * Forcibly clear the SOCK_CLOSED flag. It gets set
493 * independent of the connection mutex, and we could have
494 * received a socket close event before we had the chance to
495 * shut the socket down.
497 ceph_con_flag_clear(con, CEPH_CON_F_SOCK_CLOSED);
499 con_sock_state_closed(con);
503 static void ceph_con_reset_protocol(struct ceph_connection *con)
505 dout("%s con %p\n", __func__, con);
507 ceph_con_close_socket(con);
509 WARN_ON(con->in_msg->con != con);
510 ceph_msg_put(con->in_msg);
514 WARN_ON(con->out_msg->con != con);
515 ceph_msg_put(con->out_msg);
518 if (con->bounce_page) {
519 __free_page(con->bounce_page);
520 con->bounce_page = NULL;
523 if (ceph_msgr2(from_msgr(con->msgr)))
524 ceph_con_v2_reset_protocol(con);
526 ceph_con_v1_reset_protocol(con);
530 * Reset a connection. Discard all incoming and outgoing messages
531 * and clear *_seq state.
533 static void ceph_msg_remove(struct ceph_msg *msg)
535 list_del_init(&msg->list_head);
540 static void ceph_msg_remove_list(struct list_head *head)
542 while (!list_empty(head)) {
543 struct ceph_msg *msg = list_first_entry(head, struct ceph_msg,
545 ceph_msg_remove(msg);
549 void ceph_con_reset_session(struct ceph_connection *con)
551 dout("%s con %p\n", __func__, con);
553 WARN_ON(con->in_msg);
554 WARN_ON(con->out_msg);
555 ceph_msg_remove_list(&con->out_queue);
556 ceph_msg_remove_list(&con->out_sent);
559 con->in_seq_acked = 0;
561 if (ceph_msgr2(from_msgr(con->msgr)))
562 ceph_con_v2_reset_session(con);
564 ceph_con_v1_reset_session(con);
568 * mark a peer down. drop any open connections.
570 void ceph_con_close(struct ceph_connection *con)
572 mutex_lock(&con->mutex);
573 dout("con_close %p peer %s\n", con, ceph_pr_addr(&con->peer_addr));
574 con->state = CEPH_CON_S_CLOSED;
576 ceph_con_flag_clear(con, CEPH_CON_F_LOSSYTX); /* so we retry next
578 ceph_con_flag_clear(con, CEPH_CON_F_KEEPALIVE_PENDING);
579 ceph_con_flag_clear(con, CEPH_CON_F_WRITE_PENDING);
580 ceph_con_flag_clear(con, CEPH_CON_F_BACKOFF);
582 ceph_con_reset_protocol(con);
583 ceph_con_reset_session(con);
585 mutex_unlock(&con->mutex);
587 EXPORT_SYMBOL(ceph_con_close);
590 * Reopen a closed connection, with a new peer address.
592 void ceph_con_open(struct ceph_connection *con,
593 __u8 entity_type, __u64 entity_num,
594 struct ceph_entity_addr *addr)
596 mutex_lock(&con->mutex);
597 dout("con_open %p %s\n", con, ceph_pr_addr(addr));
599 WARN_ON(con->state != CEPH_CON_S_CLOSED);
600 con->state = CEPH_CON_S_PREOPEN;
602 con->peer_name.type = (__u8) entity_type;
603 con->peer_name.num = cpu_to_le64(entity_num);
605 memcpy(&con->peer_addr, addr, sizeof(*addr));
606 con->delay = 0; /* reset backoff memory */
607 mutex_unlock(&con->mutex);
610 EXPORT_SYMBOL(ceph_con_open);
613 * return true if this connection ever successfully opened
615 bool ceph_con_opened(struct ceph_connection *con)
617 if (ceph_msgr2(from_msgr(con->msgr)))
618 return ceph_con_v2_opened(con);
620 return ceph_con_v1_opened(con);
624 * initialize a new connection.
626 void ceph_con_init(struct ceph_connection *con, void *private,
627 const struct ceph_connection_operations *ops,
628 struct ceph_messenger *msgr)
630 dout("con_init %p\n", con);
631 memset(con, 0, sizeof(*con));
632 con->private = private;
636 con_sock_state_init(con);
638 mutex_init(&con->mutex);
639 INIT_LIST_HEAD(&con->out_queue);
640 INIT_LIST_HEAD(&con->out_sent);
641 INIT_DELAYED_WORK(&con->work, ceph_con_workfn);
643 con->state = CEPH_CON_S_CLOSED;
645 EXPORT_SYMBOL(ceph_con_init);
648 * We maintain a global counter to order connection attempts. Get
649 * a unique seq greater than @gt.
651 u32 ceph_get_global_seq(struct ceph_messenger *msgr, u32 gt)
655 spin_lock(&msgr->global_seq_lock);
656 if (msgr->global_seq < gt)
657 msgr->global_seq = gt;
658 ret = ++msgr->global_seq;
659 spin_unlock(&msgr->global_seq_lock);
664 * Discard messages that have been acked by the server.
666 void ceph_con_discard_sent(struct ceph_connection *con, u64 ack_seq)
668 struct ceph_msg *msg;
671 dout("%s con %p ack_seq %llu\n", __func__, con, ack_seq);
672 while (!list_empty(&con->out_sent)) {
673 msg = list_first_entry(&con->out_sent, struct ceph_msg,
675 WARN_ON(msg->needs_out_seq);
676 seq = le64_to_cpu(msg->hdr.seq);
680 dout("%s con %p discarding msg %p seq %llu\n", __func__, con,
682 ceph_msg_remove(msg);
687 * Discard messages that have been requeued in con_fault(), up to
688 * reconnect_seq. This avoids gratuitously resending messages that
689 * the server had received and handled prior to reconnect.
691 void ceph_con_discard_requeued(struct ceph_connection *con, u64 reconnect_seq)
693 struct ceph_msg *msg;
696 dout("%s con %p reconnect_seq %llu\n", __func__, con, reconnect_seq);
697 while (!list_empty(&con->out_queue)) {
698 msg = list_first_entry(&con->out_queue, struct ceph_msg,
700 if (msg->needs_out_seq)
702 seq = le64_to_cpu(msg->hdr.seq);
703 if (seq > reconnect_seq)
706 dout("%s con %p discarding msg %p seq %llu\n", __func__, con,
708 ceph_msg_remove(msg);
715 * For a bio data item, a piece is whatever remains of the next
716 * entry in the current bio iovec, or the first entry in the next
719 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor *cursor,
722 struct ceph_msg_data *data = cursor->data;
723 struct ceph_bio_iter *it = &cursor->bio_iter;
725 cursor->resid = min_t(size_t, length, data->bio_length);
727 if (cursor->resid < it->iter.bi_size)
728 it->iter.bi_size = cursor->resid;
730 BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
733 static struct page *ceph_msg_data_bio_next(struct ceph_msg_data_cursor *cursor,
737 struct bio_vec bv = bio_iter_iovec(cursor->bio_iter.bio,
738 cursor->bio_iter.iter);
740 *page_offset = bv.bv_offset;
745 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor *cursor,
748 struct ceph_bio_iter *it = &cursor->bio_iter;
749 struct page *page = bio_iter_page(it->bio, it->iter);
751 BUG_ON(bytes > cursor->resid);
752 BUG_ON(bytes > bio_iter_len(it->bio, it->iter));
753 cursor->resid -= bytes;
754 bio_advance_iter(it->bio, &it->iter, bytes);
757 return false; /* no more data */
759 if (!bytes || (it->iter.bi_size && it->iter.bi_bvec_done &&
760 page == bio_iter_page(it->bio, it->iter)))
761 return false; /* more bytes to process in this segment */
763 if (!it->iter.bi_size) {
764 it->bio = it->bio->bi_next;
765 it->iter = it->bio->bi_iter;
766 if (cursor->resid < it->iter.bi_size)
767 it->iter.bi_size = cursor->resid;
770 BUG_ON(cursor->resid < bio_iter_len(it->bio, it->iter));
773 #endif /* CONFIG_BLOCK */
775 static void ceph_msg_data_bvecs_cursor_init(struct ceph_msg_data_cursor *cursor,
778 struct ceph_msg_data *data = cursor->data;
779 struct bio_vec *bvecs = data->bvec_pos.bvecs;
781 cursor->resid = min_t(size_t, length, data->bvec_pos.iter.bi_size);
782 cursor->bvec_iter = data->bvec_pos.iter;
783 cursor->bvec_iter.bi_size = cursor->resid;
785 BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
788 static struct page *ceph_msg_data_bvecs_next(struct ceph_msg_data_cursor *cursor,
792 struct bio_vec bv = bvec_iter_bvec(cursor->data->bvec_pos.bvecs,
795 *page_offset = bv.bv_offset;
800 static bool ceph_msg_data_bvecs_advance(struct ceph_msg_data_cursor *cursor,
803 struct bio_vec *bvecs = cursor->data->bvec_pos.bvecs;
804 struct page *page = bvec_iter_page(bvecs, cursor->bvec_iter);
806 BUG_ON(bytes > cursor->resid);
807 BUG_ON(bytes > bvec_iter_len(bvecs, cursor->bvec_iter));
808 cursor->resid -= bytes;
809 bvec_iter_advance(bvecs, &cursor->bvec_iter, bytes);
812 return false; /* no more data */
814 if (!bytes || (cursor->bvec_iter.bi_bvec_done &&
815 page == bvec_iter_page(bvecs, cursor->bvec_iter)))
816 return false; /* more bytes to process in this segment */
818 BUG_ON(cursor->resid < bvec_iter_len(bvecs, cursor->bvec_iter));
823 * For a page array, a piece comes from the first page in the array
824 * that has not already been fully consumed.
826 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor *cursor,
829 struct ceph_msg_data *data = cursor->data;
832 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
834 BUG_ON(!data->pages);
835 BUG_ON(!data->length);
837 cursor->resid = min(length, data->length);
838 page_count = calc_pages_for(data->alignment, (u64)data->length);
839 cursor->page_offset = data->alignment & ~PAGE_MASK;
840 cursor->page_index = 0;
841 BUG_ON(page_count > (int)USHRT_MAX);
842 cursor->page_count = (unsigned short)page_count;
843 BUG_ON(length > SIZE_MAX - cursor->page_offset);
847 ceph_msg_data_pages_next(struct ceph_msg_data_cursor *cursor,
848 size_t *page_offset, size_t *length)
850 struct ceph_msg_data *data = cursor->data;
852 BUG_ON(data->type != CEPH_MSG_DATA_PAGES);
854 BUG_ON(cursor->page_index >= cursor->page_count);
855 BUG_ON(cursor->page_offset >= PAGE_SIZE);
857 *page_offset = cursor->page_offset;
858 *length = min_t(size_t, cursor->resid, PAGE_SIZE - *page_offset);
859 return data->pages[cursor->page_index];
862 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor *cursor,
865 BUG_ON(cursor->data->type != CEPH_MSG_DATA_PAGES);
867 BUG_ON(cursor->page_offset + bytes > PAGE_SIZE);
869 /* Advance the cursor page offset */
871 cursor->resid -= bytes;
872 cursor->page_offset = (cursor->page_offset + bytes) & ~PAGE_MASK;
873 if (!bytes || cursor->page_offset)
874 return false; /* more bytes to process in the current page */
877 return false; /* no more data */
879 /* Move on to the next page; offset is already at 0 */
881 BUG_ON(cursor->page_index >= cursor->page_count);
882 cursor->page_index++;
887 * For a pagelist, a piece is whatever remains to be consumed in the
888 * first page in the list, or the front of the next page.
891 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor *cursor,
894 struct ceph_msg_data *data = cursor->data;
895 struct ceph_pagelist *pagelist;
898 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
900 pagelist = data->pagelist;
904 return; /* pagelist can be assigned but empty */
906 BUG_ON(list_empty(&pagelist->head));
907 page = list_first_entry(&pagelist->head, struct page, lru);
909 cursor->resid = min(length, pagelist->length);
915 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor *cursor,
916 size_t *page_offset, size_t *length)
918 struct ceph_msg_data *data = cursor->data;
919 struct ceph_pagelist *pagelist;
921 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
923 pagelist = data->pagelist;
926 BUG_ON(!cursor->page);
927 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
929 /* offset of first page in pagelist is always 0 */
930 *page_offset = cursor->offset & ~PAGE_MASK;
931 *length = min_t(size_t, cursor->resid, PAGE_SIZE - *page_offset);
935 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor *cursor,
938 struct ceph_msg_data *data = cursor->data;
939 struct ceph_pagelist *pagelist;
941 BUG_ON(data->type != CEPH_MSG_DATA_PAGELIST);
943 pagelist = data->pagelist;
946 BUG_ON(cursor->offset + cursor->resid != pagelist->length);
947 BUG_ON((cursor->offset & ~PAGE_MASK) + bytes > PAGE_SIZE);
949 /* Advance the cursor offset */
951 cursor->resid -= bytes;
952 cursor->offset += bytes;
953 /* offset of first page in pagelist is always 0 */
954 if (!bytes || cursor->offset & ~PAGE_MASK)
955 return false; /* more bytes to process in the current page */
958 return false; /* no more data */
960 /* Move on to the next page */
962 BUG_ON(list_is_last(&cursor->page->lru, &pagelist->head));
963 cursor->page = list_next_entry(cursor->page, lru);
968 * Message data is handled (sent or received) in pieces, where each
969 * piece resides on a single page. The network layer might not
970 * consume an entire piece at once. A data item's cursor keeps
971 * track of which piece is next to process and how much remains to
972 * be processed in that piece. It also tracks whether the current
973 * piece is the last one in the data item.
975 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor)
977 size_t length = cursor->total_resid;
979 switch (cursor->data->type) {
980 case CEPH_MSG_DATA_PAGELIST:
981 ceph_msg_data_pagelist_cursor_init(cursor, length);
983 case CEPH_MSG_DATA_PAGES:
984 ceph_msg_data_pages_cursor_init(cursor, length);
987 case CEPH_MSG_DATA_BIO:
988 ceph_msg_data_bio_cursor_init(cursor, length);
990 #endif /* CONFIG_BLOCK */
991 case CEPH_MSG_DATA_BVECS:
992 ceph_msg_data_bvecs_cursor_init(cursor, length);
994 case CEPH_MSG_DATA_NONE:
999 cursor->need_crc = true;
1002 void ceph_msg_data_cursor_init(struct ceph_msg_data_cursor *cursor,
1003 struct ceph_msg *msg, size_t length)
1006 BUG_ON(length > msg->data_length);
1007 BUG_ON(!msg->num_data_items);
1009 cursor->total_resid = length;
1010 cursor->data = msg->data;
1012 __ceph_msg_data_cursor_init(cursor);
1016 * Return the page containing the next piece to process for a given
1017 * data item, and supply the page offset and length of that piece.
1018 * Indicate whether this is the last piece in this data item.
1020 struct page *ceph_msg_data_next(struct ceph_msg_data_cursor *cursor,
1021 size_t *page_offset, size_t *length)
1025 switch (cursor->data->type) {
1026 case CEPH_MSG_DATA_PAGELIST:
1027 page = ceph_msg_data_pagelist_next(cursor, page_offset, length);
1029 case CEPH_MSG_DATA_PAGES:
1030 page = ceph_msg_data_pages_next(cursor, page_offset, length);
1033 case CEPH_MSG_DATA_BIO:
1034 page = ceph_msg_data_bio_next(cursor, page_offset, length);
1036 #endif /* CONFIG_BLOCK */
1037 case CEPH_MSG_DATA_BVECS:
1038 page = ceph_msg_data_bvecs_next(cursor, page_offset, length);
1040 case CEPH_MSG_DATA_NONE:
1047 BUG_ON(*page_offset + *length > PAGE_SIZE);
1049 BUG_ON(*length > cursor->resid);
1055 * Returns true if the result moves the cursor on to the next piece
1058 void ceph_msg_data_advance(struct ceph_msg_data_cursor *cursor, size_t bytes)
1062 BUG_ON(bytes > cursor->resid);
1063 switch (cursor->data->type) {
1064 case CEPH_MSG_DATA_PAGELIST:
1065 new_piece = ceph_msg_data_pagelist_advance(cursor, bytes);
1067 case CEPH_MSG_DATA_PAGES:
1068 new_piece = ceph_msg_data_pages_advance(cursor, bytes);
1071 case CEPH_MSG_DATA_BIO:
1072 new_piece = ceph_msg_data_bio_advance(cursor, bytes);
1074 #endif /* CONFIG_BLOCK */
1075 case CEPH_MSG_DATA_BVECS:
1076 new_piece = ceph_msg_data_bvecs_advance(cursor, bytes);
1078 case CEPH_MSG_DATA_NONE:
1083 cursor->total_resid -= bytes;
1085 if (!cursor->resid && cursor->total_resid) {
1087 __ceph_msg_data_cursor_init(cursor);
1090 cursor->need_crc = new_piece;
1093 u32 ceph_crc32c_page(u32 crc, struct page *page, unsigned int page_offset,
1094 unsigned int length)
1099 BUG_ON(kaddr == NULL);
1100 crc = crc32c(crc, kaddr + page_offset, length);
1106 bool ceph_addr_is_blank(const struct ceph_entity_addr *addr)
1108 struct sockaddr_storage ss = addr->in_addr; /* align */
1109 struct in_addr *addr4 = &((struct sockaddr_in *)&ss)->sin_addr;
1110 struct in6_addr *addr6 = &((struct sockaddr_in6 *)&ss)->sin6_addr;
1112 switch (ss.ss_family) {
1114 return addr4->s_addr == htonl(INADDR_ANY);
1116 return ipv6_addr_any(addr6);
1122 int ceph_addr_port(const struct ceph_entity_addr *addr)
1124 switch (get_unaligned(&addr->in_addr.ss_family)) {
1126 return ntohs(get_unaligned(&((struct sockaddr_in *)&addr->in_addr)->sin_port));
1128 return ntohs(get_unaligned(&((struct sockaddr_in6 *)&addr->in_addr)->sin6_port));
1133 void ceph_addr_set_port(struct ceph_entity_addr *addr, int p)
1135 switch (get_unaligned(&addr->in_addr.ss_family)) {
1137 put_unaligned(htons(p), &((struct sockaddr_in *)&addr->in_addr)->sin_port);
1140 put_unaligned(htons(p), &((struct sockaddr_in6 *)&addr->in_addr)->sin6_port);
1146 * Unlike other *_pton function semantics, zero indicates success.
1148 static int ceph_pton(const char *str, size_t len, struct ceph_entity_addr *addr,
1149 char delim, const char **ipend)
1151 memset(&addr->in_addr, 0, sizeof(addr->in_addr));
1153 if (in4_pton(str, len, (u8 *)&((struct sockaddr_in *)&addr->in_addr)->sin_addr.s_addr, delim, ipend)) {
1154 put_unaligned(AF_INET, &addr->in_addr.ss_family);
1158 if (in6_pton(str, len, (u8 *)&((struct sockaddr_in6 *)&addr->in_addr)->sin6_addr.s6_addr, delim, ipend)) {
1159 put_unaligned(AF_INET6, &addr->in_addr.ss_family);
1167 * Extract hostname string and resolve using kernel DNS facility.
1169 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1170 static int ceph_dns_resolve_name(const char *name, size_t namelen,
1171 struct ceph_entity_addr *addr, char delim, const char **ipend)
1173 const char *end, *delim_p;
1174 char *colon_p, *ip_addr = NULL;
1178 * The end of the hostname occurs immediately preceding the delimiter or
1179 * the port marker (':') where the delimiter takes precedence.
1181 delim_p = memchr(name, delim, namelen);
1182 colon_p = memchr(name, ':', namelen);
1184 if (delim_p && colon_p)
1185 end = delim_p < colon_p ? delim_p : colon_p;
1186 else if (!delim_p && colon_p)
1190 if (!end) /* case: hostname:/ */
1191 end = name + namelen;
1197 /* do dns_resolve upcall */
1198 ip_len = dns_query(current->nsproxy->net_ns,
1199 NULL, name, end - name, NULL, &ip_addr, NULL, false);
1201 ret = ceph_pton(ip_addr, ip_len, addr, -1, NULL);
1209 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end - name), name,
1210 ret, ret ? "failed" : ceph_pr_addr(addr));
1215 static inline int ceph_dns_resolve_name(const char *name, size_t namelen,
1216 struct ceph_entity_addr *addr, char delim, const char **ipend)
1223 * Parse a server name (IP or hostname). If a valid IP address is not found
1224 * then try to extract a hostname to resolve using userspace DNS upcall.
1226 static int ceph_parse_server_name(const char *name, size_t namelen,
1227 struct ceph_entity_addr *addr, char delim, const char **ipend)
1231 ret = ceph_pton(name, namelen, addr, delim, ipend);
1233 ret = ceph_dns_resolve_name(name, namelen, addr, delim, ipend);
1239 * Parse an ip[:port] list into an addr array. Use the default
1240 * monitor port if a port isn't specified.
1242 int ceph_parse_ips(const char *c, const char *end,
1243 struct ceph_entity_addr *addr,
1244 int max_count, int *count, char delim)
1246 int i, ret = -EINVAL;
1249 dout("parse_ips on '%.*s'\n", (int)(end-c), c);
1250 for (i = 0; i < max_count; i++) {
1251 char cur_delim = delim;
1260 ret = ceph_parse_server_name(p, end - p, &addr[i], cur_delim,
1268 if (cur_delim == ']') {
1270 dout("missing matching ']'\n");
1277 if (p < end && *p == ':') {
1280 while (p < end && *p >= '0' && *p <= '9') {
1281 port = (port * 10) + (*p - '0');
1285 port = CEPH_MON_PORT;
1286 else if (port > 65535)
1289 port = CEPH_MON_PORT;
1292 ceph_addr_set_port(&addr[i], port);
1294 * We want the type to be set according to ms_mode
1295 * option, but options are normally parsed after mon
1296 * addresses. Rather than complicating parsing, set
1297 * to LEGACY and override in build_initial_monmap()
1298 * for mon addresses and ceph_messenger_init() for
1301 addr[i].type = CEPH_ENTITY_ADDR_TYPE_LEGACY;
1304 dout("%s got %s\n", __func__, ceph_pr_addr(&addr[i]));
1325 * Process message. This happens in the worker thread. The callback should
1326 * be careful not to do anything that waits on other incoming messages or it
1329 void ceph_con_process_message(struct ceph_connection *con)
1331 struct ceph_msg *msg = con->in_msg;
1333 BUG_ON(con->in_msg->con != con);
1336 /* if first message, set peer_name */
1337 if (con->peer_name.type == 0)
1338 con->peer_name = msg->hdr.src;
1341 mutex_unlock(&con->mutex);
1343 dout("===== %p %llu from %s%lld %d=%s len %d+%d+%d (%u %u %u) =====\n",
1344 msg, le64_to_cpu(msg->hdr.seq),
1345 ENTITY_NAME(msg->hdr.src),
1346 le16_to_cpu(msg->hdr.type),
1347 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
1348 le32_to_cpu(msg->hdr.front_len),
1349 le32_to_cpu(msg->hdr.middle_len),
1350 le32_to_cpu(msg->hdr.data_len),
1351 con->in_front_crc, con->in_middle_crc, con->in_data_crc);
1352 con->ops->dispatch(con, msg);
1354 mutex_lock(&con->mutex);
1358 * Atomically queue work on a connection after the specified delay.
1359 * Bump @con reference to avoid races with connection teardown.
1360 * Returns 0 if work was queued, or an error code otherwise.
1362 static int queue_con_delay(struct ceph_connection *con, unsigned long delay)
1364 if (!con->ops->get(con)) {
1365 dout("%s %p ref count 0\n", __func__, con);
1370 delay = round_jiffies_relative(delay);
1372 dout("%s %p %lu\n", __func__, con, delay);
1373 if (!queue_delayed_work(ceph_msgr_wq, &con->work, delay)) {
1374 dout("%s %p - already queued\n", __func__, con);
1382 static void queue_con(struct ceph_connection *con)
1384 (void) queue_con_delay(con, 0);
1387 static void cancel_con(struct ceph_connection *con)
1389 if (cancel_delayed_work(&con->work)) {
1390 dout("%s %p\n", __func__, con);
1395 static bool con_sock_closed(struct ceph_connection *con)
1397 if (!ceph_con_flag_test_and_clear(con, CEPH_CON_F_SOCK_CLOSED))
1401 case CEPH_CON_S_ ## x: \
1402 con->error_msg = "socket closed (con state " #x ")"; \
1405 switch (con->state) {
1409 CASE(V1_CONNECT_MSG);
1410 CASE(V2_BANNER_PREFIX);
1411 CASE(V2_BANNER_PAYLOAD);
1414 CASE(V2_AUTH_SIGNATURE);
1415 CASE(V2_SESSION_CONNECT);
1416 CASE(V2_SESSION_RECONNECT);
1427 static bool con_backoff(struct ceph_connection *con)
1431 if (!ceph_con_flag_test_and_clear(con, CEPH_CON_F_BACKOFF))
1434 ret = queue_con_delay(con, con->delay);
1436 dout("%s: con %p FAILED to back off %lu\n", __func__,
1438 BUG_ON(ret == -ENOENT);
1439 ceph_con_flag_set(con, CEPH_CON_F_BACKOFF);
1445 /* Finish fault handling; con->mutex must *not* be held here */
1447 static void con_fault_finish(struct ceph_connection *con)
1449 dout("%s %p\n", __func__, con);
1452 * in case we faulted due to authentication, invalidate our
1453 * current tickets so that we can get new ones.
1455 if (con->v1.auth_retry) {
1456 dout("auth_retry %d, invalidating\n", con->v1.auth_retry);
1457 if (con->ops->invalidate_authorizer)
1458 con->ops->invalidate_authorizer(con);
1459 con->v1.auth_retry = 0;
1462 if (con->ops->fault)
1463 con->ops->fault(con);
1467 * Do some work on a connection. Drop a connection ref when we're done.
1469 static void ceph_con_workfn(struct work_struct *work)
1471 struct ceph_connection *con = container_of(work, struct ceph_connection,
1475 mutex_lock(&con->mutex);
1479 if ((fault = con_sock_closed(con))) {
1480 dout("%s: con %p SOCK_CLOSED\n", __func__, con);
1483 if (con_backoff(con)) {
1484 dout("%s: con %p BACKOFF\n", __func__, con);
1487 if (con->state == CEPH_CON_S_STANDBY) {
1488 dout("%s: con %p STANDBY\n", __func__, con);
1491 if (con->state == CEPH_CON_S_CLOSED) {
1492 dout("%s: con %p CLOSED\n", __func__, con);
1496 if (con->state == CEPH_CON_S_PREOPEN) {
1497 dout("%s: con %p PREOPEN\n", __func__, con);
1501 if (ceph_msgr2(from_msgr(con->msgr)))
1502 ret = ceph_con_v2_try_read(con);
1504 ret = ceph_con_v1_try_read(con);
1508 if (!con->error_msg)
1509 con->error_msg = "socket error on read";
1514 if (ceph_msgr2(from_msgr(con->msgr)))
1515 ret = ceph_con_v2_try_write(con);
1517 ret = ceph_con_v1_try_write(con);
1521 if (!con->error_msg)
1522 con->error_msg = "socket error on write";
1526 break; /* If we make it to here, we're done */
1530 mutex_unlock(&con->mutex);
1533 con_fault_finish(con);
1539 * Generic error/fault handler. A retry mechanism is used with
1540 * exponential backoff
1542 static void con_fault(struct ceph_connection *con)
1544 dout("fault %p state %d to peer %s\n",
1545 con, con->state, ceph_pr_addr(&con->peer_addr));
1547 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con->peer_name),
1548 ceph_pr_addr(&con->peer_addr), con->error_msg);
1549 con->error_msg = NULL;
1551 WARN_ON(con->state == CEPH_CON_S_STANDBY ||
1552 con->state == CEPH_CON_S_CLOSED);
1554 ceph_con_reset_protocol(con);
1556 if (ceph_con_flag_test(con, CEPH_CON_F_LOSSYTX)) {
1557 dout("fault on LOSSYTX channel, marking CLOSED\n");
1558 con->state = CEPH_CON_S_CLOSED;
1562 /* Requeue anything that hasn't been acked */
1563 list_splice_init(&con->out_sent, &con->out_queue);
1565 /* If there are no messages queued or keepalive pending, place
1566 * the connection in a STANDBY state */
1567 if (list_empty(&con->out_queue) &&
1568 !ceph_con_flag_test(con, CEPH_CON_F_KEEPALIVE_PENDING)) {
1569 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con);
1570 ceph_con_flag_clear(con, CEPH_CON_F_WRITE_PENDING);
1571 con->state = CEPH_CON_S_STANDBY;
1573 /* retry after a delay. */
1574 con->state = CEPH_CON_S_PREOPEN;
1576 con->delay = BASE_DELAY_INTERVAL;
1577 } else if (con->delay < MAX_DELAY_INTERVAL) {
1579 if (con->delay > MAX_DELAY_INTERVAL)
1580 con->delay = MAX_DELAY_INTERVAL;
1582 ceph_con_flag_set(con, CEPH_CON_F_BACKOFF);
1587 void ceph_messenger_reset_nonce(struct ceph_messenger *msgr)
1589 u32 nonce = le32_to_cpu(msgr->inst.addr.nonce) + 1000000;
1590 msgr->inst.addr.nonce = cpu_to_le32(nonce);
1591 ceph_encode_my_addr(msgr);
1595 * initialize a new messenger instance
1597 void ceph_messenger_init(struct ceph_messenger *msgr,
1598 struct ceph_entity_addr *myaddr)
1600 spin_lock_init(&msgr->global_seq_lock);
1603 memcpy(&msgr->inst.addr.in_addr, &myaddr->in_addr,
1604 sizeof(msgr->inst.addr.in_addr));
1605 ceph_addr_set_port(&msgr->inst.addr, 0);
1609 * Since nautilus, clients are identified using type ANY.
1610 * For msgr1, ceph_encode_banner_addr() munges it to NONE.
1612 msgr->inst.addr.type = CEPH_ENTITY_ADDR_TYPE_ANY;
1614 /* generate a random non-zero nonce */
1616 get_random_bytes(&msgr->inst.addr.nonce,
1617 sizeof(msgr->inst.addr.nonce));
1618 } while (!msgr->inst.addr.nonce);
1619 ceph_encode_my_addr(msgr);
1621 atomic_set(&msgr->stopping, 0);
1622 write_pnet(&msgr->net, get_net(current->nsproxy->net_ns));
1624 dout("%s %p\n", __func__, msgr);
1627 void ceph_messenger_fini(struct ceph_messenger *msgr)
1629 put_net(read_pnet(&msgr->net));
1632 static void msg_con_set(struct ceph_msg *msg, struct ceph_connection *con)
1635 msg->con->ops->put(msg->con);
1637 msg->con = con ? con->ops->get(con) : NULL;
1638 BUG_ON(msg->con != con);
1641 static void clear_standby(struct ceph_connection *con)
1643 /* come back from STANDBY? */
1644 if (con->state == CEPH_CON_S_STANDBY) {
1645 dout("clear_standby %p and ++connect_seq\n", con);
1646 con->state = CEPH_CON_S_PREOPEN;
1647 con->v1.connect_seq++;
1648 WARN_ON(ceph_con_flag_test(con, CEPH_CON_F_WRITE_PENDING));
1649 WARN_ON(ceph_con_flag_test(con, CEPH_CON_F_KEEPALIVE_PENDING));
1654 * Queue up an outgoing message on the given connection.
1656 * Consumes a ref on @msg.
1658 void ceph_con_send(struct ceph_connection *con, struct ceph_msg *msg)
1661 msg->hdr.src = con->msgr->inst.name;
1662 BUG_ON(msg->front.iov_len != le32_to_cpu(msg->hdr.front_len));
1663 msg->needs_out_seq = true;
1665 mutex_lock(&con->mutex);
1667 if (con->state == CEPH_CON_S_CLOSED) {
1668 dout("con_send %p closed, dropping %p\n", con, msg);
1670 mutex_unlock(&con->mutex);
1674 msg_con_set(msg, con);
1676 BUG_ON(!list_empty(&msg->list_head));
1677 list_add_tail(&msg->list_head, &con->out_queue);
1678 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg,
1679 ENTITY_NAME(con->peer_name), le16_to_cpu(msg->hdr.type),
1680 ceph_msg_type_name(le16_to_cpu(msg->hdr.type)),
1681 le32_to_cpu(msg->hdr.front_len),
1682 le32_to_cpu(msg->hdr.middle_len),
1683 le32_to_cpu(msg->hdr.data_len));
1686 mutex_unlock(&con->mutex);
1688 /* if there wasn't anything waiting to send before, queue
1690 if (!ceph_con_flag_test_and_set(con, CEPH_CON_F_WRITE_PENDING))
1693 EXPORT_SYMBOL(ceph_con_send);
1696 * Revoke a message that was previously queued for send
1698 void ceph_msg_revoke(struct ceph_msg *msg)
1700 struct ceph_connection *con = msg->con;
1703 dout("%s msg %p null con\n", __func__, msg);
1704 return; /* Message not in our possession */
1707 mutex_lock(&con->mutex);
1708 if (list_empty(&msg->list_head)) {
1709 WARN_ON(con->out_msg == msg);
1710 dout("%s con %p msg %p not linked\n", __func__, con, msg);
1711 mutex_unlock(&con->mutex);
1715 dout("%s con %p msg %p was linked\n", __func__, con, msg);
1717 ceph_msg_remove(msg);
1719 if (con->out_msg == msg) {
1720 WARN_ON(con->state != CEPH_CON_S_OPEN);
1721 dout("%s con %p msg %p was sending\n", __func__, con, msg);
1722 if (ceph_msgr2(from_msgr(con->msgr)))
1723 ceph_con_v2_revoke(con);
1725 ceph_con_v1_revoke(con);
1726 ceph_msg_put(con->out_msg);
1727 con->out_msg = NULL;
1729 dout("%s con %p msg %p not current, out_msg %p\n", __func__,
1730 con, msg, con->out_msg);
1732 mutex_unlock(&con->mutex);
1736 * Revoke a message that we may be reading data into
1738 void ceph_msg_revoke_incoming(struct ceph_msg *msg)
1740 struct ceph_connection *con = msg->con;
1743 dout("%s msg %p null con\n", __func__, msg);
1744 return; /* Message not in our possession */
1747 mutex_lock(&con->mutex);
1748 if (con->in_msg == msg) {
1749 WARN_ON(con->state != CEPH_CON_S_OPEN);
1750 dout("%s con %p msg %p was recving\n", __func__, con, msg);
1751 if (ceph_msgr2(from_msgr(con->msgr)))
1752 ceph_con_v2_revoke_incoming(con);
1754 ceph_con_v1_revoke_incoming(con);
1755 ceph_msg_put(con->in_msg);
1758 dout("%s con %p msg %p not current, in_msg %p\n", __func__,
1759 con, msg, con->in_msg);
1761 mutex_unlock(&con->mutex);
1765 * Queue a keepalive byte to ensure the tcp connection is alive.
1767 void ceph_con_keepalive(struct ceph_connection *con)
1769 dout("con_keepalive %p\n", con);
1770 mutex_lock(&con->mutex);
1772 ceph_con_flag_set(con, CEPH_CON_F_KEEPALIVE_PENDING);
1773 mutex_unlock(&con->mutex);
1775 if (!ceph_con_flag_test_and_set(con, CEPH_CON_F_WRITE_PENDING))
1778 EXPORT_SYMBOL(ceph_con_keepalive);
1780 bool ceph_con_keepalive_expired(struct ceph_connection *con,
1781 unsigned long interval)
1784 (con->peer_features & CEPH_FEATURE_MSGR_KEEPALIVE2)) {
1785 struct timespec64 now;
1786 struct timespec64 ts;
1787 ktime_get_real_ts64(&now);
1788 jiffies_to_timespec64(interval, &ts);
1789 ts = timespec64_add(con->last_keepalive_ack, ts);
1790 return timespec64_compare(&now, &ts) >= 0;
1795 static struct ceph_msg_data *ceph_msg_data_add(struct ceph_msg *msg)
1797 BUG_ON(msg->num_data_items >= msg->max_data_items);
1798 return &msg->data[msg->num_data_items++];
1801 static void ceph_msg_data_destroy(struct ceph_msg_data *data)
1803 if (data->type == CEPH_MSG_DATA_PAGES && data->own_pages) {
1804 int num_pages = calc_pages_for(data->alignment, data->length);
1805 ceph_release_page_vector(data->pages, num_pages);
1806 } else if (data->type == CEPH_MSG_DATA_PAGELIST) {
1807 ceph_pagelist_release(data->pagelist);
1811 void ceph_msg_data_add_pages(struct ceph_msg *msg, struct page **pages,
1812 size_t length, size_t alignment, bool own_pages)
1814 struct ceph_msg_data *data;
1819 data = ceph_msg_data_add(msg);
1820 data->type = CEPH_MSG_DATA_PAGES;
1821 data->pages = pages;
1822 data->length = length;
1823 data->alignment = alignment & ~PAGE_MASK;
1824 data->own_pages = own_pages;
1826 msg->data_length += length;
1828 EXPORT_SYMBOL(ceph_msg_data_add_pages);
1830 void ceph_msg_data_add_pagelist(struct ceph_msg *msg,
1831 struct ceph_pagelist *pagelist)
1833 struct ceph_msg_data *data;
1836 BUG_ON(!pagelist->length);
1838 data = ceph_msg_data_add(msg);
1839 data->type = CEPH_MSG_DATA_PAGELIST;
1840 refcount_inc(&pagelist->refcnt);
1841 data->pagelist = pagelist;
1843 msg->data_length += pagelist->length;
1845 EXPORT_SYMBOL(ceph_msg_data_add_pagelist);
1848 void ceph_msg_data_add_bio(struct ceph_msg *msg, struct ceph_bio_iter *bio_pos,
1851 struct ceph_msg_data *data;
1853 data = ceph_msg_data_add(msg);
1854 data->type = CEPH_MSG_DATA_BIO;
1855 data->bio_pos = *bio_pos;
1856 data->bio_length = length;
1858 msg->data_length += length;
1860 EXPORT_SYMBOL(ceph_msg_data_add_bio);
1861 #endif /* CONFIG_BLOCK */
1863 void ceph_msg_data_add_bvecs(struct ceph_msg *msg,
1864 struct ceph_bvec_iter *bvec_pos)
1866 struct ceph_msg_data *data;
1868 data = ceph_msg_data_add(msg);
1869 data->type = CEPH_MSG_DATA_BVECS;
1870 data->bvec_pos = *bvec_pos;
1872 msg->data_length += bvec_pos->iter.bi_size;
1874 EXPORT_SYMBOL(ceph_msg_data_add_bvecs);
1877 * construct a new message with given type, size
1878 * the new msg has a ref count of 1.
1880 struct ceph_msg *ceph_msg_new2(int type, int front_len, int max_data_items,
1881 gfp_t flags, bool can_fail)
1885 m = kmem_cache_zalloc(ceph_msg_cache, flags);
1889 m->hdr.type = cpu_to_le16(type);
1890 m->hdr.priority = cpu_to_le16(CEPH_MSG_PRIO_DEFAULT);
1891 m->hdr.front_len = cpu_to_le32(front_len);
1893 INIT_LIST_HEAD(&m->list_head);
1894 kref_init(&m->kref);
1898 m->front.iov_base = kvmalloc(front_len, flags);
1899 if (m->front.iov_base == NULL) {
1900 dout("ceph_msg_new can't allocate %d bytes\n",
1905 m->front.iov_base = NULL;
1907 m->front_alloc_len = m->front.iov_len = front_len;
1909 if (max_data_items) {
1910 m->data = kmalloc_array(max_data_items, sizeof(*m->data),
1915 m->max_data_items = max_data_items;
1918 dout("ceph_msg_new %p front %d\n", m, front_len);
1925 pr_err("msg_new can't create type %d front %d\n", type,
1929 dout("msg_new can't create type %d front %d\n", type,
1934 EXPORT_SYMBOL(ceph_msg_new2);
1936 struct ceph_msg *ceph_msg_new(int type, int front_len, gfp_t flags,
1939 return ceph_msg_new2(type, front_len, 0, flags, can_fail);
1941 EXPORT_SYMBOL(ceph_msg_new);
1944 * Allocate "middle" portion of a message, if it is needed and wasn't
1945 * allocated by alloc_msg. This allows us to read a small fixed-size
1946 * per-type header in the front and then gracefully fail (i.e.,
1947 * propagate the error to the caller based on info in the front) when
1948 * the middle is too large.
1950 static int ceph_alloc_middle(struct ceph_connection *con, struct ceph_msg *msg)
1952 int type = le16_to_cpu(msg->hdr.type);
1953 int middle_len = le32_to_cpu(msg->hdr.middle_len);
1955 dout("alloc_middle %p type %d %s middle_len %d\n", msg, type,
1956 ceph_msg_type_name(type), middle_len);
1957 BUG_ON(!middle_len);
1958 BUG_ON(msg->middle);
1960 msg->middle = ceph_buffer_new(middle_len, GFP_NOFS);
1967 * Allocate a message for receiving an incoming message on a
1968 * connection, and save the result in con->in_msg. Uses the
1969 * connection's private alloc_msg op if available.
1971 * Returns 0 on success, or a negative error code.
1973 * On success, if we set *skip = 1:
1974 * - the next message should be skipped and ignored.
1975 * - con->in_msg == NULL
1976 * or if we set *skip = 0:
1977 * - con->in_msg is non-null.
1978 * On error (ENOMEM, EAGAIN, ...),
1979 * - con->in_msg == NULL
1981 int ceph_con_in_msg_alloc(struct ceph_connection *con,
1982 struct ceph_msg_header *hdr, int *skip)
1984 int middle_len = le32_to_cpu(hdr->middle_len);
1985 struct ceph_msg *msg;
1988 BUG_ON(con->in_msg != NULL);
1989 BUG_ON(!con->ops->alloc_msg);
1991 mutex_unlock(&con->mutex);
1992 msg = con->ops->alloc_msg(con, hdr, skip);
1993 mutex_lock(&con->mutex);
1994 if (con->state != CEPH_CON_S_OPEN) {
2001 msg_con_set(msg, con);
2005 * Null message pointer means either we should skip
2006 * this message or we couldn't allocate memory. The
2007 * former is not an error.
2012 con->error_msg = "error allocating memory for incoming message";
2015 memcpy(&con->in_msg->hdr, hdr, sizeof(*hdr));
2017 if (middle_len && !con->in_msg->middle) {
2018 ret = ceph_alloc_middle(con, con->in_msg);
2020 ceph_msg_put(con->in_msg);
2028 void ceph_con_get_out_msg(struct ceph_connection *con)
2030 struct ceph_msg *msg;
2032 BUG_ON(list_empty(&con->out_queue));
2033 msg = list_first_entry(&con->out_queue, struct ceph_msg, list_head);
2034 WARN_ON(msg->con != con);
2037 * Put the message on "sent" list using a ref from ceph_con_send().
2038 * It is put when the message is acked or revoked.
2040 list_move_tail(&msg->list_head, &con->out_sent);
2043 * Only assign outgoing seq # if we haven't sent this message
2044 * yet. If it is requeued, resend with it's original seq.
2046 if (msg->needs_out_seq) {
2047 msg->hdr.seq = cpu_to_le64(++con->out_seq);
2048 msg->needs_out_seq = false;
2050 if (con->ops->reencode_message)
2051 con->ops->reencode_message(msg);
2055 * Get a ref for out_msg. It is put when we are done sending the
2056 * message or in case of a fault.
2058 WARN_ON(con->out_msg);
2059 con->out_msg = ceph_msg_get(msg);
2063 * Free a generically kmalloc'd message.
2065 static void ceph_msg_free(struct ceph_msg *m)
2067 dout("%s %p\n", __func__, m);
2068 kvfree(m->front.iov_base);
2070 kmem_cache_free(ceph_msg_cache, m);
2073 static void ceph_msg_release(struct kref *kref)
2075 struct ceph_msg *m = container_of(kref, struct ceph_msg, kref);
2078 dout("%s %p\n", __func__, m);
2079 WARN_ON(!list_empty(&m->list_head));
2081 msg_con_set(m, NULL);
2083 /* drop middle, data, if any */
2085 ceph_buffer_put(m->middle);
2089 for (i = 0; i < m->num_data_items; i++)
2090 ceph_msg_data_destroy(&m->data[i]);
2093 ceph_msgpool_put(m->pool, m);
2098 struct ceph_msg *ceph_msg_get(struct ceph_msg *msg)
2100 dout("%s %p (was %d)\n", __func__, msg,
2101 kref_read(&msg->kref));
2102 kref_get(&msg->kref);
2105 EXPORT_SYMBOL(ceph_msg_get);
2107 void ceph_msg_put(struct ceph_msg *msg)
2109 dout("%s %p (was %d)\n", __func__, msg,
2110 kref_read(&msg->kref));
2111 kref_put(&msg->kref, ceph_msg_release);
2113 EXPORT_SYMBOL(ceph_msg_put);
2115 void ceph_msg_dump(struct ceph_msg *msg)
2117 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg,
2118 msg->front_alloc_len, msg->data_length);
2119 print_hex_dump(KERN_DEBUG, "header: ",
2120 DUMP_PREFIX_OFFSET, 16, 1,
2121 &msg->hdr, sizeof(msg->hdr), true);
2122 print_hex_dump(KERN_DEBUG, " front: ",
2123 DUMP_PREFIX_OFFSET, 16, 1,
2124 msg->front.iov_base, msg->front.iov_len, true);
2126 print_hex_dump(KERN_DEBUG, "middle: ",
2127 DUMP_PREFIX_OFFSET, 16, 1,
2128 msg->middle->vec.iov_base,
2129 msg->middle->vec.iov_len, true);
2130 print_hex_dump(KERN_DEBUG, "footer: ",
2131 DUMP_PREFIX_OFFSET, 16, 1,
2132 &msg->footer, sizeof(msg->footer), true);
2134 EXPORT_SYMBOL(ceph_msg_dump);