1 #include <linux/ceph/ceph_debug.h>
3 #include <linux/crc32c.h>
4 #include <linux/ctype.h>
5 #include <linux/highmem.h>
6 #include <linux/inet.h>
7 #include <linux/kthread.h>
9 #include <linux/slab.h>
10 #include <linux/socket.h>
11 #include <linux/string.h>
12 #include <linux/bio.h>
13 #include <linux/blkdev.h>
14 #include <linux/dns_resolver.h>
17 #include <linux/ceph/libceph.h>
18 #include <linux/ceph/messenger.h>
19 #include <linux/ceph/decode.h>
20 #include <linux/ceph/pagelist.h>
21 #include <linux/export.h>
24 * Ceph uses the messenger to exchange ceph_msg messages with other
25 * hosts in the system. The messenger provides ordered and reliable
26 * delivery. We tolerate TCP disconnects by reconnecting (with
27 * exponential backoff) in the case of a fault (disconnection, bad
28 * crc, protocol error). Acks allow sent messages to be discarded by
32 /* State values for ceph_connection->sock_state; NEW is assumed to be 0 */
34 #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
35 #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
36 #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
37 #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
38 #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
40 /* static tag bytes (protocol control messages) */
41 static char tag_msg
= CEPH_MSGR_TAG_MSG
;
42 static char tag_ack
= CEPH_MSGR_TAG_ACK
;
43 static char tag_keepalive
= CEPH_MSGR_TAG_KEEPALIVE
;
46 static struct lock_class_key socket_class
;
50 * When skipping (ignoring) a block of input we read it into a "skip
51 * buffer," which is this many bytes in size.
53 #define SKIP_BUF_SIZE 1024
55 static void queue_con(struct ceph_connection
*con
);
56 static void con_work(struct work_struct
*);
57 static void ceph_fault(struct ceph_connection
*con
);
60 * Nicely render a sockaddr as a string. An array of formatted
61 * strings is used, to approximate reentrancy.
63 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
64 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
65 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
66 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
68 static char addr_str
[ADDR_STR_COUNT
][MAX_ADDR_STR_LEN
];
69 static atomic_t addr_str_seq
= ATOMIC_INIT(0);
71 static struct page
*zero_page
; /* used in certain error cases */
73 const char *ceph_pr_addr(const struct sockaddr_storage
*ss
)
77 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
78 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
80 i
= atomic_inc_return(&addr_str_seq
) & ADDR_STR_COUNT_MASK
;
83 switch (ss
->ss_family
) {
85 snprintf(s
, MAX_ADDR_STR_LEN
, "%pI4:%hu", &in4
->sin_addr
,
86 ntohs(in4
->sin_port
));
90 snprintf(s
, MAX_ADDR_STR_LEN
, "[%pI6c]:%hu", &in6
->sin6_addr
,
91 ntohs(in6
->sin6_port
));
95 snprintf(s
, MAX_ADDR_STR_LEN
, "(unknown sockaddr family %hu)",
101 EXPORT_SYMBOL(ceph_pr_addr
);
103 static void encode_my_addr(struct ceph_messenger
*msgr
)
105 memcpy(&msgr
->my_enc_addr
, &msgr
->inst
.addr
, sizeof(msgr
->my_enc_addr
));
106 ceph_encode_addr(&msgr
->my_enc_addr
);
110 * work queue for all reading and writing to/from the socket.
112 static struct workqueue_struct
*ceph_msgr_wq
;
114 void _ceph_msgr_exit(void)
117 destroy_workqueue(ceph_msgr_wq
);
121 BUG_ON(zero_page
== NULL
);
123 page_cache_release(zero_page
);
127 int ceph_msgr_init(void)
129 BUG_ON(zero_page
!= NULL
);
130 zero_page
= ZERO_PAGE(0);
131 page_cache_get(zero_page
);
133 ceph_msgr_wq
= alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT
, 0);
137 pr_err("msgr_init failed to create workqueue\n");
142 EXPORT_SYMBOL(ceph_msgr_init
);
144 void ceph_msgr_exit(void)
146 BUG_ON(ceph_msgr_wq
== NULL
);
150 EXPORT_SYMBOL(ceph_msgr_exit
);
152 void ceph_msgr_flush(void)
154 flush_workqueue(ceph_msgr_wq
);
156 EXPORT_SYMBOL(ceph_msgr_flush
);
158 /* Connection socket state transition functions */
160 static void con_sock_state_init(struct ceph_connection
*con
)
164 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSED
);
165 if (WARN_ON(old_state
!= CON_SOCK_STATE_NEW
))
166 printk("%s: unexpected old state %d\n", __func__
, old_state
);
169 static void con_sock_state_connecting(struct ceph_connection
*con
)
173 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CONNECTING
);
174 if (WARN_ON(old_state
!= CON_SOCK_STATE_CLOSED
))
175 printk("%s: unexpected old state %d\n", __func__
, old_state
);
178 static void con_sock_state_connected(struct ceph_connection
*con
)
182 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CONNECTED
);
183 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTING
))
184 printk("%s: unexpected old state %d\n", __func__
, old_state
);
187 static void con_sock_state_closing(struct ceph_connection
*con
)
191 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSING
);
192 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTING
&&
193 old_state
!= CON_SOCK_STATE_CONNECTED
&&
194 old_state
!= CON_SOCK_STATE_CLOSING
))
195 printk("%s: unexpected old state %d\n", __func__
, old_state
);
198 static void con_sock_state_closed(struct ceph_connection
*con
)
202 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSED
);
203 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTED
&&
204 old_state
!= CON_SOCK_STATE_CLOSING
))
205 printk("%s: unexpected old state %d\n", __func__
, old_state
);
209 * socket callback functions
212 /* data available on socket, or listen socket received a connect */
213 static void ceph_sock_data_ready(struct sock
*sk
, int count_unused
)
215 struct ceph_connection
*con
= sk
->sk_user_data
;
217 if (sk
->sk_state
!= TCP_CLOSE_WAIT
) {
218 dout("%s on %p state = %lu, queueing work\n", __func__
,
224 /* socket has buffer space for writing */
225 static void ceph_sock_write_space(struct sock
*sk
)
227 struct ceph_connection
*con
= sk
->sk_user_data
;
229 /* only queue to workqueue if there is data we want to write,
230 * and there is sufficient space in the socket buffer to accept
231 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
232 * doesn't get called again until try_write() fills the socket
233 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
234 * and net/core/stream.c:sk_stream_write_space().
236 if (test_bit(WRITE_PENDING
, &con
->flags
)) {
237 if (sk_stream_wspace(sk
) >= sk_stream_min_wspace(sk
)) {
238 dout("%s %p queueing write work\n", __func__
, con
);
239 clear_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
243 dout("%s %p nothing to write\n", __func__
, con
);
247 /* socket's state has changed */
248 static void ceph_sock_state_change(struct sock
*sk
)
250 struct ceph_connection
*con
= sk
->sk_user_data
;
252 dout("%s %p state = %lu sk_state = %u\n", __func__
,
253 con
, con
->state
, sk
->sk_state
);
255 if (test_bit(CLOSED
, &con
->state
))
258 switch (sk
->sk_state
) {
260 dout("%s TCP_CLOSE\n", __func__
);
262 dout("%s TCP_CLOSE_WAIT\n", __func__
);
263 con_sock_state_closing(con
);
264 set_bit(SOCK_CLOSED
, &con
->flags
);
267 case TCP_ESTABLISHED
:
268 dout("%s TCP_ESTABLISHED\n", __func__
);
269 con_sock_state_connected(con
);
272 default: /* Everything else is uninteresting */
278 * set up socket callbacks
280 static void set_sock_callbacks(struct socket
*sock
,
281 struct ceph_connection
*con
)
283 struct sock
*sk
= sock
->sk
;
284 sk
->sk_user_data
= con
;
285 sk
->sk_data_ready
= ceph_sock_data_ready
;
286 sk
->sk_write_space
= ceph_sock_write_space
;
287 sk
->sk_state_change
= ceph_sock_state_change
;
296 * initiate connection to a remote socket.
298 static int ceph_tcp_connect(struct ceph_connection
*con
)
300 struct sockaddr_storage
*paddr
= &con
->peer_addr
.in_addr
;
305 ret
= sock_create_kern(con
->peer_addr
.in_addr
.ss_family
, SOCK_STREAM
,
309 sock
->sk
->sk_allocation
= GFP_NOFS
;
311 #ifdef CONFIG_LOCKDEP
312 lockdep_set_class(&sock
->sk
->sk_lock
, &socket_class
);
315 set_sock_callbacks(sock
, con
);
317 dout("connect %s\n", ceph_pr_addr(&con
->peer_addr
.in_addr
));
319 con_sock_state_connecting(con
);
320 ret
= sock
->ops
->connect(sock
, (struct sockaddr
*)paddr
, sizeof(*paddr
),
322 if (ret
== -EINPROGRESS
) {
323 dout("connect %s EINPROGRESS sk_state = %u\n",
324 ceph_pr_addr(&con
->peer_addr
.in_addr
),
326 } else if (ret
< 0) {
327 pr_err("connect %s error %d\n",
328 ceph_pr_addr(&con
->peer_addr
.in_addr
), ret
);
330 con
->error_msg
= "connect error";
338 static int ceph_tcp_recvmsg(struct socket
*sock
, void *buf
, size_t len
)
340 struct kvec iov
= {buf
, len
};
341 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
344 r
= kernel_recvmsg(sock
, &msg
, &iov
, 1, len
, msg
.msg_flags
);
351 * write something. @more is true if caller will be sending more data
354 static int ceph_tcp_sendmsg(struct socket
*sock
, struct kvec
*iov
,
355 size_t kvlen
, size_t len
, int more
)
357 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
361 msg
.msg_flags
|= MSG_MORE
;
363 msg
.msg_flags
|= MSG_EOR
; /* superfluous, but what the hell */
365 r
= kernel_sendmsg(sock
, &msg
, iov
, kvlen
, len
);
371 static int ceph_tcp_sendpage(struct socket
*sock
, struct page
*page
,
372 int offset
, size_t size
, int more
)
374 int flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
| (more
? MSG_MORE
: MSG_EOR
);
377 ret
= kernel_sendpage(sock
, page
, offset
, size
, flags
);
386 * Shutdown/close the socket for the given connection.
388 static int con_close_socket(struct ceph_connection
*con
)
392 dout("con_close_socket on %p sock %p\n", con
, con
->sock
);
395 rc
= con
->sock
->ops
->shutdown(con
->sock
, SHUT_RDWR
);
396 sock_release(con
->sock
);
400 * Forcibly clear the SOCK_CLOSE flag. It gets set
401 * independent of the connection mutex, and we could have
402 * received a socket close event before we had the chance to
403 * shut the socket down.
405 clear_bit(SOCK_CLOSED
, &con
->flags
);
406 con_sock_state_closed(con
);
411 * Reset a connection. Discard all incoming and outgoing messages
412 * and clear *_seq state.
414 static void ceph_msg_remove(struct ceph_msg
*msg
)
416 list_del_init(&msg
->list_head
);
417 BUG_ON(msg
->con
== NULL
);
418 msg
->con
->ops
->put(msg
->con
);
423 static void ceph_msg_remove_list(struct list_head
*head
)
425 while (!list_empty(head
)) {
426 struct ceph_msg
*msg
= list_first_entry(head
, struct ceph_msg
,
428 ceph_msg_remove(msg
);
432 static void reset_connection(struct ceph_connection
*con
)
434 /* reset connection, out_queue, msg_ and connect_seq */
435 /* discard existing out_queue and msg_seq */
436 ceph_msg_remove_list(&con
->out_queue
);
437 ceph_msg_remove_list(&con
->out_sent
);
440 BUG_ON(con
->in_msg
->con
!= con
);
441 con
->in_msg
->con
= NULL
;
442 ceph_msg_put(con
->in_msg
);
447 con
->connect_seq
= 0;
450 ceph_msg_put(con
->out_msg
);
454 con
->in_seq_acked
= 0;
458 * mark a peer down. drop any open connections.
460 void ceph_con_close(struct ceph_connection
*con
)
462 dout("con_close %p peer %s\n", con
,
463 ceph_pr_addr(&con
->peer_addr
.in_addr
));
464 clear_bit(NEGOTIATING
, &con
->state
);
465 clear_bit(CONNECTING
, &con
->state
);
466 clear_bit(CONNECTED
, &con
->state
);
467 clear_bit(STANDBY
, &con
->state
); /* avoid connect_seq bump */
468 set_bit(CLOSED
, &con
->state
);
470 clear_bit(LOSSYTX
, &con
->flags
); /* so we retry next connect */
471 clear_bit(KEEPALIVE_PENDING
, &con
->flags
);
472 clear_bit(WRITE_PENDING
, &con
->flags
);
474 mutex_lock(&con
->mutex
);
475 reset_connection(con
);
476 con
->peer_global_seq
= 0;
477 cancel_delayed_work(&con
->work
);
478 mutex_unlock(&con
->mutex
);
481 EXPORT_SYMBOL(ceph_con_close
);
484 * Reopen a closed connection, with a new peer address.
486 void ceph_con_open(struct ceph_connection
*con
, struct ceph_entity_addr
*addr
)
488 dout("con_open %p %s\n", con
, ceph_pr_addr(&addr
->in_addr
));
489 set_bit(OPENING
, &con
->state
);
490 WARN_ON(!test_and_clear_bit(CLOSED
, &con
->state
));
492 memcpy(&con
->peer_addr
, addr
, sizeof(*addr
));
493 con
->delay
= 0; /* reset backoff memory */
496 EXPORT_SYMBOL(ceph_con_open
);
499 * return true if this connection ever successfully opened
501 bool ceph_con_opened(struct ceph_connection
*con
)
503 return con
->connect_seq
> 0;
507 * initialize a new connection.
509 void ceph_con_init(struct ceph_connection
*con
, void *private,
510 const struct ceph_connection_operations
*ops
,
511 struct ceph_messenger
*msgr
, __u8 entity_type
, __u64 entity_num
)
513 dout("con_init %p\n", con
);
514 memset(con
, 0, sizeof(*con
));
515 con
->private = private;
519 con_sock_state_init(con
);
521 con
->peer_name
.type
= (__u8
) entity_type
;
522 con
->peer_name
.num
= cpu_to_le64(entity_num
);
524 mutex_init(&con
->mutex
);
525 INIT_LIST_HEAD(&con
->out_queue
);
526 INIT_LIST_HEAD(&con
->out_sent
);
527 INIT_DELAYED_WORK(&con
->work
, con_work
);
529 set_bit(CLOSED
, &con
->state
);
531 EXPORT_SYMBOL(ceph_con_init
);
535 * We maintain a global counter to order connection attempts. Get
536 * a unique seq greater than @gt.
538 static u32
get_global_seq(struct ceph_messenger
*msgr
, u32 gt
)
542 spin_lock(&msgr
->global_seq_lock
);
543 if (msgr
->global_seq
< gt
)
544 msgr
->global_seq
= gt
;
545 ret
= ++msgr
->global_seq
;
546 spin_unlock(&msgr
->global_seq_lock
);
550 static void con_out_kvec_reset(struct ceph_connection
*con
)
552 con
->out_kvec_left
= 0;
553 con
->out_kvec_bytes
= 0;
554 con
->out_kvec_cur
= &con
->out_kvec
[0];
557 static void con_out_kvec_add(struct ceph_connection
*con
,
558 size_t size
, void *data
)
562 index
= con
->out_kvec_left
;
563 BUG_ON(index
>= ARRAY_SIZE(con
->out_kvec
));
565 con
->out_kvec
[index
].iov_len
= size
;
566 con
->out_kvec
[index
].iov_base
= data
;
567 con
->out_kvec_left
++;
568 con
->out_kvec_bytes
+= size
;
572 static void init_bio_iter(struct bio
*bio
, struct bio
**iter
, int *seg
)
583 static void iter_bio_next(struct bio
**bio_iter
, int *seg
)
585 if (*bio_iter
== NULL
)
588 BUG_ON(*seg
>= (*bio_iter
)->bi_vcnt
);
591 if (*seg
== (*bio_iter
)->bi_vcnt
)
592 init_bio_iter((*bio_iter
)->bi_next
, bio_iter
, seg
);
596 static void prepare_write_message_data(struct ceph_connection
*con
)
598 struct ceph_msg
*msg
= con
->out_msg
;
601 BUG_ON(!msg
->hdr
.data_len
);
603 /* initialize page iterator */
604 con
->out_msg_pos
.page
= 0;
606 con
->out_msg_pos
.page_pos
= msg
->page_alignment
;
608 con
->out_msg_pos
.page_pos
= 0;
611 init_bio_iter(msg
->bio
, &msg
->bio_iter
, &msg
->bio_seg
);
613 con
->out_msg_pos
.data_pos
= 0;
614 con
->out_msg_pos
.did_page_crc
= false;
615 con
->out_more
= 1; /* data + footer will follow */
619 * Prepare footer for currently outgoing message, and finish things
620 * off. Assumes out_kvec* are already valid.. we just add on to the end.
622 static void prepare_write_message_footer(struct ceph_connection
*con
)
624 struct ceph_msg
*m
= con
->out_msg
;
625 int v
= con
->out_kvec_left
;
627 m
->footer
.flags
|= CEPH_MSG_FOOTER_COMPLETE
;
629 dout("prepare_write_message_footer %p\n", con
);
630 con
->out_kvec_is_msg
= true;
631 con
->out_kvec
[v
].iov_base
= &m
->footer
;
632 con
->out_kvec
[v
].iov_len
= sizeof(m
->footer
);
633 con
->out_kvec_bytes
+= sizeof(m
->footer
);
634 con
->out_kvec_left
++;
635 con
->out_more
= m
->more_to_follow
;
636 con
->out_msg_done
= true;
640 * Prepare headers for the next outgoing message.
642 static void prepare_write_message(struct ceph_connection
*con
)
647 con_out_kvec_reset(con
);
648 con
->out_kvec_is_msg
= true;
649 con
->out_msg_done
= false;
651 /* Sneak an ack in there first? If we can get it into the same
652 * TCP packet that's a good thing. */
653 if (con
->in_seq
> con
->in_seq_acked
) {
654 con
->in_seq_acked
= con
->in_seq
;
655 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
656 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
657 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
661 BUG_ON(list_empty(&con
->out_queue
));
662 m
= list_first_entry(&con
->out_queue
, struct ceph_msg
, list_head
);
664 BUG_ON(m
->con
!= con
);
666 /* put message on sent list */
668 list_move_tail(&m
->list_head
, &con
->out_sent
);
671 * only assign outgoing seq # if we haven't sent this message
672 * yet. if it is requeued, resend with it's original seq.
674 if (m
->needs_out_seq
) {
675 m
->hdr
.seq
= cpu_to_le64(++con
->out_seq
);
676 m
->needs_out_seq
= false;
679 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n",
680 m
, con
->out_seq
, le16_to_cpu(m
->hdr
.type
),
681 le32_to_cpu(m
->hdr
.front_len
), le32_to_cpu(m
->hdr
.middle_len
),
682 le32_to_cpu(m
->hdr
.data_len
),
684 BUG_ON(le32_to_cpu(m
->hdr
.front_len
) != m
->front
.iov_len
);
686 /* tag + hdr + front + middle */
687 con_out_kvec_add(con
, sizeof (tag_msg
), &tag_msg
);
688 con_out_kvec_add(con
, sizeof (m
->hdr
), &m
->hdr
);
689 con_out_kvec_add(con
, m
->front
.iov_len
, m
->front
.iov_base
);
692 con_out_kvec_add(con
, m
->middle
->vec
.iov_len
,
693 m
->middle
->vec
.iov_base
);
695 /* fill in crc (except data pages), footer */
696 crc
= crc32c(0, &m
->hdr
, offsetof(struct ceph_msg_header
, crc
));
697 con
->out_msg
->hdr
.crc
= cpu_to_le32(crc
);
698 con
->out_msg
->footer
.flags
= 0;
700 crc
= crc32c(0, m
->front
.iov_base
, m
->front
.iov_len
);
701 con
->out_msg
->footer
.front_crc
= cpu_to_le32(crc
);
703 crc
= crc32c(0, m
->middle
->vec
.iov_base
,
704 m
->middle
->vec
.iov_len
);
705 con
->out_msg
->footer
.middle_crc
= cpu_to_le32(crc
);
707 con
->out_msg
->footer
.middle_crc
= 0;
708 dout("%s front_crc %u middle_crc %u\n", __func__
,
709 le32_to_cpu(con
->out_msg
->footer
.front_crc
),
710 le32_to_cpu(con
->out_msg
->footer
.middle_crc
));
712 /* is there a data payload? */
713 con
->out_msg
->footer
.data_crc
= 0;
715 prepare_write_message_data(con
);
717 /* no, queue up footer too and be done */
718 prepare_write_message_footer(con
);
720 set_bit(WRITE_PENDING
, &con
->flags
);
726 static void prepare_write_ack(struct ceph_connection
*con
)
728 dout("prepare_write_ack %p %llu -> %llu\n", con
,
729 con
->in_seq_acked
, con
->in_seq
);
730 con
->in_seq_acked
= con
->in_seq
;
732 con_out_kvec_reset(con
);
734 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
736 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
737 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
740 con
->out_more
= 1; /* more will follow.. eventually.. */
741 set_bit(WRITE_PENDING
, &con
->flags
);
745 * Prepare to write keepalive byte.
747 static void prepare_write_keepalive(struct ceph_connection
*con
)
749 dout("prepare_write_keepalive %p\n", con
);
750 con_out_kvec_reset(con
);
751 con_out_kvec_add(con
, sizeof (tag_keepalive
), &tag_keepalive
);
752 set_bit(WRITE_PENDING
, &con
->flags
);
756 * Connection negotiation.
759 static struct ceph_auth_handshake
*get_connect_authorizer(struct ceph_connection
*con
,
762 struct ceph_auth_handshake
*auth
;
764 if (!con
->ops
->get_authorizer
) {
765 con
->out_connect
.authorizer_protocol
= CEPH_AUTH_UNKNOWN
;
766 con
->out_connect
.authorizer_len
= 0;
771 /* Can't hold the mutex while getting authorizer */
773 mutex_unlock(&con
->mutex
);
775 auth
= con
->ops
->get_authorizer(con
, auth_proto
, con
->auth_retry
);
777 mutex_lock(&con
->mutex
);
781 if (test_bit(CLOSED
, &con
->state
) || test_bit(OPENING
, &con
->flags
))
782 return ERR_PTR(-EAGAIN
);
784 con
->auth_reply_buf
= auth
->authorizer_reply_buf
;
785 con
->auth_reply_buf_len
= auth
->authorizer_reply_buf_len
;
792 * We connected to a peer and are saying hello.
794 static void prepare_write_banner(struct ceph_connection
*con
)
796 con_out_kvec_add(con
, strlen(CEPH_BANNER
), CEPH_BANNER
);
797 con_out_kvec_add(con
, sizeof (con
->msgr
->my_enc_addr
),
798 &con
->msgr
->my_enc_addr
);
801 set_bit(WRITE_PENDING
, &con
->flags
);
804 static int prepare_write_connect(struct ceph_connection
*con
)
806 unsigned int global_seq
= get_global_seq(con
->msgr
, 0);
809 struct ceph_auth_handshake
*auth
;
811 switch (con
->peer_name
.type
) {
812 case CEPH_ENTITY_TYPE_MON
:
813 proto
= CEPH_MONC_PROTOCOL
;
815 case CEPH_ENTITY_TYPE_OSD
:
816 proto
= CEPH_OSDC_PROTOCOL
;
818 case CEPH_ENTITY_TYPE_MDS
:
819 proto
= CEPH_MDSC_PROTOCOL
;
825 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con
,
826 con
->connect_seq
, global_seq
, proto
);
828 con
->out_connect
.features
= cpu_to_le64(con
->msgr
->supported_features
);
829 con
->out_connect
.host_type
= cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT
);
830 con
->out_connect
.connect_seq
= cpu_to_le32(con
->connect_seq
);
831 con
->out_connect
.global_seq
= cpu_to_le32(global_seq
);
832 con
->out_connect
.protocol_version
= cpu_to_le32(proto
);
833 con
->out_connect
.flags
= 0;
835 auth_proto
= CEPH_AUTH_UNKNOWN
;
836 auth
= get_connect_authorizer(con
, &auth_proto
);
838 return PTR_ERR(auth
);
840 con
->out_connect
.authorizer_protocol
= cpu_to_le32(auth_proto
);
841 con
->out_connect
.authorizer_len
= auth
?
842 cpu_to_le32(auth
->authorizer_buf_len
) : 0;
844 con_out_kvec_reset(con
);
845 con_out_kvec_add(con
, sizeof (con
->out_connect
),
847 if (auth
&& auth
->authorizer_buf_len
)
848 con_out_kvec_add(con
, auth
->authorizer_buf_len
,
849 auth
->authorizer_buf
);
852 set_bit(WRITE_PENDING
, &con
->flags
);
858 * write as much of pending kvecs to the socket as we can.
860 * 0 -> socket full, but more to do
863 static int write_partial_kvec(struct ceph_connection
*con
)
867 dout("write_partial_kvec %p %d left\n", con
, con
->out_kvec_bytes
);
868 while (con
->out_kvec_bytes
> 0) {
869 ret
= ceph_tcp_sendmsg(con
->sock
, con
->out_kvec_cur
,
870 con
->out_kvec_left
, con
->out_kvec_bytes
,
874 con
->out_kvec_bytes
-= ret
;
875 if (con
->out_kvec_bytes
== 0)
878 /* account for full iov entries consumed */
879 while (ret
>= con
->out_kvec_cur
->iov_len
) {
880 BUG_ON(!con
->out_kvec_left
);
881 ret
-= con
->out_kvec_cur
->iov_len
;
883 con
->out_kvec_left
--;
885 /* and for a partially-consumed entry */
887 con
->out_kvec_cur
->iov_len
-= ret
;
888 con
->out_kvec_cur
->iov_base
+= ret
;
891 con
->out_kvec_left
= 0;
892 con
->out_kvec_is_msg
= false;
895 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con
,
896 con
->out_kvec_bytes
, con
->out_kvec_left
, ret
);
897 return ret
; /* done! */
900 static void out_msg_pos_next(struct ceph_connection
*con
, struct page
*page
,
901 size_t len
, size_t sent
, bool in_trail
)
903 struct ceph_msg
*msg
= con
->out_msg
;
908 con
->out_msg_pos
.data_pos
+= sent
;
909 con
->out_msg_pos
.page_pos
+= sent
;
911 con
->out_msg_pos
.page_pos
= 0;
912 con
->out_msg_pos
.page
++;
913 con
->out_msg_pos
.did_page_crc
= false;
915 list_move_tail(&page
->lru
,
917 else if (msg
->pagelist
)
918 list_move_tail(&page
->lru
,
919 &msg
->pagelist
->head
);
922 iter_bio_next(&msg
->bio_iter
, &msg
->bio_seg
);
928 * Write as much message data payload as we can. If we finish, queue
930 * 1 -> done, footer is now queued in out_kvec[].
931 * 0 -> socket full, but more to do
934 static int write_partial_msg_pages(struct ceph_connection
*con
)
936 struct ceph_msg
*msg
= con
->out_msg
;
937 unsigned int data_len
= le32_to_cpu(msg
->hdr
.data_len
);
939 bool do_datacrc
= !con
->msgr
->nocrc
;
942 bool in_trail
= false;
943 size_t trail_len
= (msg
->trail
? msg
->trail
->length
: 0);
945 dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
946 con
, msg
, con
->out_msg_pos
.page
, msg
->nr_pages
,
947 con
->out_msg_pos
.page_pos
);
949 while (data_len
> con
->out_msg_pos
.data_pos
) {
950 struct page
*page
= NULL
;
951 int max_write
= PAGE_SIZE
;
954 total_max_write
= data_len
- trail_len
-
955 con
->out_msg_pos
.data_pos
;
958 * if we are calculating the data crc (the default), we need
959 * to map the page. if our pages[] has been revoked, use the
963 /* have we reached the trail part of the data? */
964 if (con
->out_msg_pos
.data_pos
>= data_len
- trail_len
) {
967 total_max_write
= data_len
- con
->out_msg_pos
.data_pos
;
969 page
= list_first_entry(&msg
->trail
->head
,
971 } else if (msg
->pages
) {
972 page
= msg
->pages
[con
->out_msg_pos
.page
];
973 } else if (msg
->pagelist
) {
974 page
= list_first_entry(&msg
->pagelist
->head
,
977 } else if (msg
->bio
) {
980 bv
= bio_iovec_idx(msg
->bio_iter
, msg
->bio_seg
);
982 bio_offset
= bv
->bv_offset
;
983 max_write
= bv
->bv_len
;
988 len
= min_t(int, max_write
- con
->out_msg_pos
.page_pos
,
991 if (do_datacrc
&& !con
->out_msg_pos
.did_page_crc
) {
994 u32 tmpcrc
= le32_to_cpu(msg
->footer
.data_crc
);
998 BUG_ON(kaddr
== NULL
);
999 base
= kaddr
+ con
->out_msg_pos
.page_pos
+ bio_offset
;
1000 crc
= crc32c(tmpcrc
, base
, len
);
1001 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1002 con
->out_msg_pos
.did_page_crc
= true;
1004 ret
= ceph_tcp_sendpage(con
->sock
, page
,
1005 con
->out_msg_pos
.page_pos
+ bio_offset
,
1014 out_msg_pos_next(con
, page
, len
, (size_t) ret
, in_trail
);
1017 dout("write_partial_msg_pages %p msg %p done\n", con
, msg
);
1019 /* prepare and queue up footer, too */
1021 msg
->footer
.flags
|= CEPH_MSG_FOOTER_NOCRC
;
1022 con_out_kvec_reset(con
);
1023 prepare_write_message_footer(con
);
1032 static int write_partial_skip(struct ceph_connection
*con
)
1036 while (con
->out_skip
> 0) {
1037 size_t size
= min(con
->out_skip
, (int) PAGE_CACHE_SIZE
);
1039 ret
= ceph_tcp_sendpage(con
->sock
, zero_page
, 0, size
, 1);
1042 con
->out_skip
-= ret
;
1050 * Prepare to read connection handshake, or an ack.
1052 static void prepare_read_banner(struct ceph_connection
*con
)
1054 dout("prepare_read_banner %p\n", con
);
1055 con
->in_base_pos
= 0;
1058 static void prepare_read_connect(struct ceph_connection
*con
)
1060 dout("prepare_read_connect %p\n", con
);
1061 con
->in_base_pos
= 0;
1064 static void prepare_read_ack(struct ceph_connection
*con
)
1066 dout("prepare_read_ack %p\n", con
);
1067 con
->in_base_pos
= 0;
1070 static void prepare_read_tag(struct ceph_connection
*con
)
1072 dout("prepare_read_tag %p\n", con
);
1073 con
->in_base_pos
= 0;
1074 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1078 * Prepare to read a message.
1080 static int prepare_read_message(struct ceph_connection
*con
)
1082 dout("prepare_read_message %p\n", con
);
1083 BUG_ON(con
->in_msg
!= NULL
);
1084 con
->in_base_pos
= 0;
1085 con
->in_front_crc
= con
->in_middle_crc
= con
->in_data_crc
= 0;
1090 static int read_partial(struct ceph_connection
*con
,
1091 int end
, int size
, void *object
)
1093 while (con
->in_base_pos
< end
) {
1094 int left
= end
- con
->in_base_pos
;
1095 int have
= size
- left
;
1096 int ret
= ceph_tcp_recvmsg(con
->sock
, object
+ have
, left
);
1099 con
->in_base_pos
+= ret
;
1106 * Read all or part of the connect-side handshake on a new connection
1108 static int read_partial_banner(struct ceph_connection
*con
)
1114 dout("read_partial_banner %p at %d\n", con
, con
->in_base_pos
);
1117 size
= strlen(CEPH_BANNER
);
1119 ret
= read_partial(con
, end
, size
, con
->in_banner
);
1123 size
= sizeof (con
->actual_peer_addr
);
1125 ret
= read_partial(con
, end
, size
, &con
->actual_peer_addr
);
1129 size
= sizeof (con
->peer_addr_for_me
);
1131 ret
= read_partial(con
, end
, size
, &con
->peer_addr_for_me
);
1139 static int read_partial_connect(struct ceph_connection
*con
)
1145 dout("read_partial_connect %p at %d\n", con
, con
->in_base_pos
);
1147 size
= sizeof (con
->in_reply
);
1149 ret
= read_partial(con
, end
, size
, &con
->in_reply
);
1153 size
= le32_to_cpu(con
->in_reply
.authorizer_len
);
1155 ret
= read_partial(con
, end
, size
, con
->auth_reply_buf
);
1159 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1160 con
, (int)con
->in_reply
.tag
,
1161 le32_to_cpu(con
->in_reply
.connect_seq
),
1162 le32_to_cpu(con
->in_reply
.global_seq
));
1169 * Verify the hello banner looks okay.
1171 static int verify_hello(struct ceph_connection
*con
)
1173 if (memcmp(con
->in_banner
, CEPH_BANNER
, strlen(CEPH_BANNER
))) {
1174 pr_err("connect to %s got bad banner\n",
1175 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1176 con
->error_msg
= "protocol error, bad banner";
1182 static bool addr_is_blank(struct sockaddr_storage
*ss
)
1184 switch (ss
->ss_family
) {
1186 return ((struct sockaddr_in
*)ss
)->sin_addr
.s_addr
== 0;
1189 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[0] == 0 &&
1190 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[1] == 0 &&
1191 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[2] == 0 &&
1192 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[3] == 0;
1197 static int addr_port(struct sockaddr_storage
*ss
)
1199 switch (ss
->ss_family
) {
1201 return ntohs(((struct sockaddr_in
*)ss
)->sin_port
);
1203 return ntohs(((struct sockaddr_in6
*)ss
)->sin6_port
);
1208 static void addr_set_port(struct sockaddr_storage
*ss
, int p
)
1210 switch (ss
->ss_family
) {
1212 ((struct sockaddr_in
*)ss
)->sin_port
= htons(p
);
1215 ((struct sockaddr_in6
*)ss
)->sin6_port
= htons(p
);
1221 * Unlike other *_pton function semantics, zero indicates success.
1223 static int ceph_pton(const char *str
, size_t len
, struct sockaddr_storage
*ss
,
1224 char delim
, const char **ipend
)
1226 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
1227 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
1229 memset(ss
, 0, sizeof(*ss
));
1231 if (in4_pton(str
, len
, (u8
*)&in4
->sin_addr
.s_addr
, delim
, ipend
)) {
1232 ss
->ss_family
= AF_INET
;
1236 if (in6_pton(str
, len
, (u8
*)&in6
->sin6_addr
.s6_addr
, delim
, ipend
)) {
1237 ss
->ss_family
= AF_INET6
;
1245 * Extract hostname string and resolve using kernel DNS facility.
1247 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1248 static int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1249 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1251 const char *end
, *delim_p
;
1252 char *colon_p
, *ip_addr
= NULL
;
1256 * The end of the hostname occurs immediately preceding the delimiter or
1257 * the port marker (':') where the delimiter takes precedence.
1259 delim_p
= memchr(name
, delim
, namelen
);
1260 colon_p
= memchr(name
, ':', namelen
);
1262 if (delim_p
&& colon_p
)
1263 end
= delim_p
< colon_p
? delim_p
: colon_p
;
1264 else if (!delim_p
&& colon_p
)
1268 if (!end
) /* case: hostname:/ */
1269 end
= name
+ namelen
;
1275 /* do dns_resolve upcall */
1276 ip_len
= dns_query(NULL
, name
, end
- name
, NULL
, &ip_addr
, NULL
);
1278 ret
= ceph_pton(ip_addr
, ip_len
, ss
, -1, NULL
);
1286 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end
- name
), name
,
1287 ret
, ret
? "failed" : ceph_pr_addr(ss
));
1292 static inline int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1293 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1300 * Parse a server name (IP or hostname). If a valid IP address is not found
1301 * then try to extract a hostname to resolve using userspace DNS upcall.
1303 static int ceph_parse_server_name(const char *name
, size_t namelen
,
1304 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1308 ret
= ceph_pton(name
, namelen
, ss
, delim
, ipend
);
1310 ret
= ceph_dns_resolve_name(name
, namelen
, ss
, delim
, ipend
);
1316 * Parse an ip[:port] list into an addr array. Use the default
1317 * monitor port if a port isn't specified.
1319 int ceph_parse_ips(const char *c
, const char *end
,
1320 struct ceph_entity_addr
*addr
,
1321 int max_count
, int *count
)
1323 int i
, ret
= -EINVAL
;
1326 dout("parse_ips on '%.*s'\n", (int)(end
-c
), c
);
1327 for (i
= 0; i
< max_count
; i
++) {
1329 struct sockaddr_storage
*ss
= &addr
[i
].in_addr
;
1338 ret
= ceph_parse_server_name(p
, end
- p
, ss
, delim
, &ipend
);
1347 dout("missing matching ']'\n");
1354 if (p
< end
&& *p
== ':') {
1357 while (p
< end
&& *p
>= '0' && *p
<= '9') {
1358 port
= (port
* 10) + (*p
- '0');
1361 if (port
> 65535 || port
== 0)
1364 port
= CEPH_MON_PORT
;
1367 addr_set_port(ss
, port
);
1369 dout("parse_ips got %s\n", ceph_pr_addr(ss
));
1386 pr_err("parse_ips bad ip '%.*s'\n", (int)(end
- c
), c
);
1389 EXPORT_SYMBOL(ceph_parse_ips
);
1391 static int process_banner(struct ceph_connection
*con
)
1393 dout("process_banner on %p\n", con
);
1395 if (verify_hello(con
) < 0)
1398 ceph_decode_addr(&con
->actual_peer_addr
);
1399 ceph_decode_addr(&con
->peer_addr_for_me
);
1402 * Make sure the other end is who we wanted. note that the other
1403 * end may not yet know their ip address, so if it's 0.0.0.0, give
1404 * them the benefit of the doubt.
1406 if (memcmp(&con
->peer_addr
, &con
->actual_peer_addr
,
1407 sizeof(con
->peer_addr
)) != 0 &&
1408 !(addr_is_blank(&con
->actual_peer_addr
.in_addr
) &&
1409 con
->actual_peer_addr
.nonce
== con
->peer_addr
.nonce
)) {
1410 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1411 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1412 (int)le32_to_cpu(con
->peer_addr
.nonce
),
1413 ceph_pr_addr(&con
->actual_peer_addr
.in_addr
),
1414 (int)le32_to_cpu(con
->actual_peer_addr
.nonce
));
1415 con
->error_msg
= "wrong peer at address";
1420 * did we learn our address?
1422 if (addr_is_blank(&con
->msgr
->inst
.addr
.in_addr
)) {
1423 int port
= addr_port(&con
->msgr
->inst
.addr
.in_addr
);
1425 memcpy(&con
->msgr
->inst
.addr
.in_addr
,
1426 &con
->peer_addr_for_me
.in_addr
,
1427 sizeof(con
->peer_addr_for_me
.in_addr
));
1428 addr_set_port(&con
->msgr
->inst
.addr
.in_addr
, port
);
1429 encode_my_addr(con
->msgr
);
1430 dout("process_banner learned my addr is %s\n",
1431 ceph_pr_addr(&con
->msgr
->inst
.addr
.in_addr
));
1437 static void fail_protocol(struct ceph_connection
*con
)
1439 reset_connection(con
);
1440 set_bit(CLOSED
, &con
->state
); /* in case there's queued work */
1443 static int process_connect(struct ceph_connection
*con
)
1445 u64 sup_feat
= con
->msgr
->supported_features
;
1446 u64 req_feat
= con
->msgr
->required_features
;
1447 u64 server_feat
= le64_to_cpu(con
->in_reply
.features
);
1450 dout("process_connect on %p tag %d\n", con
, (int)con
->in_tag
);
1452 switch (con
->in_reply
.tag
) {
1453 case CEPH_MSGR_TAG_FEATURES
:
1454 pr_err("%s%lld %s feature set mismatch,"
1455 " my %llx < server's %llx, missing %llx\n",
1456 ENTITY_NAME(con
->peer_name
),
1457 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1458 sup_feat
, server_feat
, server_feat
& ~sup_feat
);
1459 con
->error_msg
= "missing required protocol features";
1463 case CEPH_MSGR_TAG_BADPROTOVER
:
1464 pr_err("%s%lld %s protocol version mismatch,"
1465 " my %d != server's %d\n",
1466 ENTITY_NAME(con
->peer_name
),
1467 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1468 le32_to_cpu(con
->out_connect
.protocol_version
),
1469 le32_to_cpu(con
->in_reply
.protocol_version
));
1470 con
->error_msg
= "protocol version mismatch";
1474 case CEPH_MSGR_TAG_BADAUTHORIZER
:
1476 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con
,
1478 if (con
->auth_retry
== 2) {
1479 con
->error_msg
= "connect authorization failure";
1482 con
->auth_retry
= 1;
1483 ret
= prepare_write_connect(con
);
1486 prepare_read_connect(con
);
1489 case CEPH_MSGR_TAG_RESETSESSION
:
1491 * If we connected with a large connect_seq but the peer
1492 * has no record of a session with us (no connection, or
1493 * connect_seq == 0), they will send RESETSESION to indicate
1494 * that they must have reset their session, and may have
1497 dout("process_connect got RESET peer seq %u\n",
1498 le32_to_cpu(con
->in_connect
.connect_seq
));
1499 pr_err("%s%lld %s connection reset\n",
1500 ENTITY_NAME(con
->peer_name
),
1501 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1502 reset_connection(con
);
1503 ret
= prepare_write_connect(con
);
1506 prepare_read_connect(con
);
1508 /* Tell ceph about it. */
1509 mutex_unlock(&con
->mutex
);
1510 pr_info("reset on %s%lld\n", ENTITY_NAME(con
->peer_name
));
1511 if (con
->ops
->peer_reset
)
1512 con
->ops
->peer_reset(con
);
1513 mutex_lock(&con
->mutex
);
1514 if (test_bit(CLOSED
, &con
->state
) ||
1515 test_bit(OPENING
, &con
->state
))
1519 case CEPH_MSGR_TAG_RETRY_SESSION
:
1521 * If we sent a smaller connect_seq than the peer has, try
1522 * again with a larger value.
1524 dout("process_connect got RETRY my seq = %u, peer_seq = %u\n",
1525 le32_to_cpu(con
->out_connect
.connect_seq
),
1526 le32_to_cpu(con
->in_connect
.connect_seq
));
1527 con
->connect_seq
= le32_to_cpu(con
->in_connect
.connect_seq
);
1528 ret
= prepare_write_connect(con
);
1531 prepare_read_connect(con
);
1534 case CEPH_MSGR_TAG_RETRY_GLOBAL
:
1536 * If we sent a smaller global_seq than the peer has, try
1537 * again with a larger value.
1539 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1540 con
->peer_global_seq
,
1541 le32_to_cpu(con
->in_connect
.global_seq
));
1542 get_global_seq(con
->msgr
,
1543 le32_to_cpu(con
->in_connect
.global_seq
));
1544 ret
= prepare_write_connect(con
);
1547 prepare_read_connect(con
);
1550 case CEPH_MSGR_TAG_READY
:
1551 if (req_feat
& ~server_feat
) {
1552 pr_err("%s%lld %s protocol feature mismatch,"
1553 " my required %llx > server's %llx, need %llx\n",
1554 ENTITY_NAME(con
->peer_name
),
1555 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1556 req_feat
, server_feat
, req_feat
& ~server_feat
);
1557 con
->error_msg
= "missing required protocol features";
1561 clear_bit(NEGOTIATING
, &con
->state
);
1562 clear_bit(CONNECTING
, &con
->state
);
1563 set_bit(CONNECTED
, &con
->state
);
1564 con
->peer_global_seq
= le32_to_cpu(con
->in_reply
.global_seq
);
1566 con
->peer_features
= server_feat
;
1567 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1568 con
->peer_global_seq
,
1569 le32_to_cpu(con
->in_reply
.connect_seq
),
1571 WARN_ON(con
->connect_seq
!=
1572 le32_to_cpu(con
->in_reply
.connect_seq
));
1574 if (con
->in_reply
.flags
& CEPH_MSG_CONNECT_LOSSY
)
1575 set_bit(LOSSYTX
, &con
->flags
);
1577 prepare_read_tag(con
);
1580 case CEPH_MSGR_TAG_WAIT
:
1582 * If there is a connection race (we are opening
1583 * connections to each other), one of us may just have
1584 * to WAIT. This shouldn't happen if we are the
1587 pr_err("process_connect got WAIT as client\n");
1588 con
->error_msg
= "protocol error, got WAIT as client";
1592 pr_err("connect protocol error, will retry\n");
1593 con
->error_msg
= "protocol error, garbage tag during connect";
1601 * read (part of) an ack
1603 static int read_partial_ack(struct ceph_connection
*con
)
1605 int size
= sizeof (con
->in_temp_ack
);
1608 return read_partial(con
, end
, size
, &con
->in_temp_ack
);
1613 * We can finally discard anything that's been acked.
1615 static void process_ack(struct ceph_connection
*con
)
1618 u64 ack
= le64_to_cpu(con
->in_temp_ack
);
1621 while (!list_empty(&con
->out_sent
)) {
1622 m
= list_first_entry(&con
->out_sent
, struct ceph_msg
,
1624 seq
= le64_to_cpu(m
->hdr
.seq
);
1627 dout("got ack for seq %llu type %d at %p\n", seq
,
1628 le16_to_cpu(m
->hdr
.type
), m
);
1629 m
->ack_stamp
= jiffies
;
1632 prepare_read_tag(con
);
1638 static int read_partial_message_section(struct ceph_connection
*con
,
1639 struct kvec
*section
,
1640 unsigned int sec_len
, u32
*crc
)
1646 while (section
->iov_len
< sec_len
) {
1647 BUG_ON(section
->iov_base
== NULL
);
1648 left
= sec_len
- section
->iov_len
;
1649 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)section
->iov_base
+
1650 section
->iov_len
, left
);
1653 section
->iov_len
+= ret
;
1655 if (section
->iov_len
== sec_len
)
1656 *crc
= crc32c(0, section
->iov_base
, section
->iov_len
);
1661 static bool ceph_con_in_msg_alloc(struct ceph_connection
*con
,
1662 struct ceph_msg_header
*hdr
);
1665 static int read_partial_message_pages(struct ceph_connection
*con
,
1666 struct page
**pages
,
1667 unsigned int data_len
, bool do_datacrc
)
1673 left
= min((int)(data_len
- con
->in_msg_pos
.data_pos
),
1674 (int)(PAGE_SIZE
- con
->in_msg_pos
.page_pos
));
1676 BUG_ON(pages
== NULL
);
1677 p
= kmap(pages
[con
->in_msg_pos
.page
]);
1678 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ con
->in_msg_pos
.page_pos
,
1680 if (ret
> 0 && do_datacrc
)
1682 crc32c(con
->in_data_crc
,
1683 p
+ con
->in_msg_pos
.page_pos
, ret
);
1684 kunmap(pages
[con
->in_msg_pos
.page
]);
1687 con
->in_msg_pos
.data_pos
+= ret
;
1688 con
->in_msg_pos
.page_pos
+= ret
;
1689 if (con
->in_msg_pos
.page_pos
== PAGE_SIZE
) {
1690 con
->in_msg_pos
.page_pos
= 0;
1691 con
->in_msg_pos
.page
++;
1698 static int read_partial_message_bio(struct ceph_connection
*con
,
1699 struct bio
**bio_iter
, int *bio_seg
,
1700 unsigned int data_len
, bool do_datacrc
)
1702 struct bio_vec
*bv
= bio_iovec_idx(*bio_iter
, *bio_seg
);
1709 left
= min((int)(data_len
- con
->in_msg_pos
.data_pos
),
1710 (int)(bv
->bv_len
- con
->in_msg_pos
.page_pos
));
1712 p
= kmap(bv
->bv_page
) + bv
->bv_offset
;
1714 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ con
->in_msg_pos
.page_pos
,
1716 if (ret
> 0 && do_datacrc
)
1718 crc32c(con
->in_data_crc
,
1719 p
+ con
->in_msg_pos
.page_pos
, ret
);
1720 kunmap(bv
->bv_page
);
1723 con
->in_msg_pos
.data_pos
+= ret
;
1724 con
->in_msg_pos
.page_pos
+= ret
;
1725 if (con
->in_msg_pos
.page_pos
== bv
->bv_len
) {
1726 con
->in_msg_pos
.page_pos
= 0;
1727 iter_bio_next(bio_iter
, bio_seg
);
1735 * read (part of) a message.
1737 static int read_partial_message(struct ceph_connection
*con
)
1739 struct ceph_msg
*m
= con
->in_msg
;
1743 unsigned int front_len
, middle_len
, data_len
;
1744 bool do_datacrc
= !con
->msgr
->nocrc
;
1748 dout("read_partial_message con %p msg %p\n", con
, m
);
1751 size
= sizeof (con
->in_hdr
);
1753 ret
= read_partial(con
, end
, size
, &con
->in_hdr
);
1757 crc
= crc32c(0, &con
->in_hdr
, offsetof(struct ceph_msg_header
, crc
));
1758 if (cpu_to_le32(crc
) != con
->in_hdr
.crc
) {
1759 pr_err("read_partial_message bad hdr "
1760 " crc %u != expected %u\n",
1761 crc
, con
->in_hdr
.crc
);
1765 front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
1766 if (front_len
> CEPH_MSG_MAX_FRONT_LEN
)
1768 middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
1769 if (middle_len
> CEPH_MSG_MAX_DATA_LEN
)
1771 data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
1772 if (data_len
> CEPH_MSG_MAX_DATA_LEN
)
1776 seq
= le64_to_cpu(con
->in_hdr
.seq
);
1777 if ((s64
)seq
- (s64
)con
->in_seq
< 1) {
1778 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1779 ENTITY_NAME(con
->peer_name
),
1780 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1781 seq
, con
->in_seq
+ 1);
1782 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1784 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1786 } else if ((s64
)seq
- (s64
)con
->in_seq
> 1) {
1787 pr_err("read_partial_message bad seq %lld expected %lld\n",
1788 seq
, con
->in_seq
+ 1);
1789 con
->error_msg
= "bad message sequence # for incoming message";
1793 /* allocate message? */
1795 dout("got hdr type %d front %d data %d\n", con
->in_hdr
.type
,
1796 con
->in_hdr
.front_len
, con
->in_hdr
.data_len
);
1797 if (ceph_con_in_msg_alloc(con
, &con
->in_hdr
)) {
1798 /* skip this message */
1799 dout("alloc_msg said skip message\n");
1800 BUG_ON(con
->in_msg
);
1801 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1803 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1809 "error allocating memory for incoming message";
1813 BUG_ON(con
->in_msg
->con
!= con
);
1815 m
->front
.iov_len
= 0; /* haven't read it yet */
1817 m
->middle
->vec
.iov_len
= 0;
1819 con
->in_msg_pos
.page
= 0;
1821 con
->in_msg_pos
.page_pos
= m
->page_alignment
;
1823 con
->in_msg_pos
.page_pos
= 0;
1824 con
->in_msg_pos
.data_pos
= 0;
1828 ret
= read_partial_message_section(con
, &m
->front
, front_len
,
1829 &con
->in_front_crc
);
1835 ret
= read_partial_message_section(con
, &m
->middle
->vec
,
1837 &con
->in_middle_crc
);
1842 if (m
->bio
&& !m
->bio_iter
)
1843 init_bio_iter(m
->bio
, &m
->bio_iter
, &m
->bio_seg
);
1847 while (con
->in_msg_pos
.data_pos
< data_len
) {
1849 ret
= read_partial_message_pages(con
, m
->pages
,
1850 data_len
, do_datacrc
);
1854 } else if (m
->bio
) {
1856 ret
= read_partial_message_bio(con
,
1857 &m
->bio_iter
, &m
->bio_seg
,
1858 data_len
, do_datacrc
);
1868 size
= sizeof (m
->footer
);
1870 ret
= read_partial(con
, end
, size
, &m
->footer
);
1874 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1875 m
, front_len
, m
->footer
.front_crc
, middle_len
,
1876 m
->footer
.middle_crc
, data_len
, m
->footer
.data_crc
);
1879 if (con
->in_front_crc
!= le32_to_cpu(m
->footer
.front_crc
)) {
1880 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1881 m
, con
->in_front_crc
, m
->footer
.front_crc
);
1884 if (con
->in_middle_crc
!= le32_to_cpu(m
->footer
.middle_crc
)) {
1885 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1886 m
, con
->in_middle_crc
, m
->footer
.middle_crc
);
1890 (m
->footer
.flags
& CEPH_MSG_FOOTER_NOCRC
) == 0 &&
1891 con
->in_data_crc
!= le32_to_cpu(m
->footer
.data_crc
)) {
1892 pr_err("read_partial_message %p data crc %u != exp. %u\n", m
,
1893 con
->in_data_crc
, le32_to_cpu(m
->footer
.data_crc
));
1897 return 1; /* done! */
1901 * Process message. This happens in the worker thread. The callback should
1902 * be careful not to do anything that waits on other incoming messages or it
1905 static void process_message(struct ceph_connection
*con
)
1907 struct ceph_msg
*msg
;
1909 BUG_ON(con
->in_msg
->con
!= con
);
1910 con
->in_msg
->con
= NULL
;
1915 /* if first message, set peer_name */
1916 if (con
->peer_name
.type
== 0)
1917 con
->peer_name
= msg
->hdr
.src
;
1920 mutex_unlock(&con
->mutex
);
1922 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
1923 msg
, le64_to_cpu(msg
->hdr
.seq
),
1924 ENTITY_NAME(msg
->hdr
.src
),
1925 le16_to_cpu(msg
->hdr
.type
),
1926 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
1927 le32_to_cpu(msg
->hdr
.front_len
),
1928 le32_to_cpu(msg
->hdr
.data_len
),
1929 con
->in_front_crc
, con
->in_middle_crc
, con
->in_data_crc
);
1930 con
->ops
->dispatch(con
, msg
);
1932 mutex_lock(&con
->mutex
);
1933 prepare_read_tag(con
);
1938 * Write something to the socket. Called in a worker thread when the
1939 * socket appears to be writeable and we have something ready to send.
1941 static int try_write(struct ceph_connection
*con
)
1945 dout("try_write start %p state %lu\n", con
, con
->state
);
1948 dout("try_write out_kvec_bytes %d\n", con
->out_kvec_bytes
);
1950 /* open the socket first? */
1951 if (con
->sock
== NULL
) {
1952 set_bit(CONNECTING
, &con
->state
);
1954 con_out_kvec_reset(con
);
1955 prepare_write_banner(con
);
1956 prepare_read_banner(con
);
1958 BUG_ON(con
->in_msg
);
1959 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1960 dout("try_write initiating connect on %p new state %lu\n",
1962 ret
= ceph_tcp_connect(con
);
1964 con
->error_msg
= "connect error";
1970 /* kvec data queued? */
1971 if (con
->out_skip
) {
1972 ret
= write_partial_skip(con
);
1976 if (con
->out_kvec_left
) {
1977 ret
= write_partial_kvec(con
);
1984 if (con
->out_msg_done
) {
1985 ceph_msg_put(con
->out_msg
);
1986 con
->out_msg
= NULL
; /* we're done with this one */
1990 ret
= write_partial_msg_pages(con
);
1992 goto more_kvec
; /* we need to send the footer, too! */
1996 dout("try_write write_partial_msg_pages err %d\n",
2003 if (!test_bit(CONNECTING
, &con
->state
)) {
2004 /* is anything else pending? */
2005 if (!list_empty(&con
->out_queue
)) {
2006 prepare_write_message(con
);
2009 if (con
->in_seq
> con
->in_seq_acked
) {
2010 prepare_write_ack(con
);
2013 if (test_and_clear_bit(KEEPALIVE_PENDING
, &con
->flags
)) {
2014 prepare_write_keepalive(con
);
2019 /* Nothing to do! */
2020 clear_bit(WRITE_PENDING
, &con
->flags
);
2021 dout("try_write nothing else to write.\n");
2024 dout("try_write done on %p ret %d\n", con
, ret
);
2031 * Read what we can from the socket.
2033 static int try_read(struct ceph_connection
*con
)
2040 if (test_bit(STANDBY
, &con
->state
))
2043 dout("try_read start on %p\n", con
);
2046 dout("try_read tag %d in_base_pos %d\n", (int)con
->in_tag
,
2050 * process_connect and process_message drop and re-take
2051 * con->mutex. make sure we handle a racing close or reopen.
2053 if (test_bit(CLOSED
, &con
->state
) ||
2054 test_bit(OPENING
, &con
->state
)) {
2059 if (test_bit(CONNECTING
, &con
->state
)) {
2060 if (!test_bit(NEGOTIATING
, &con
->state
)) {
2061 dout("try_read connecting\n");
2062 ret
= read_partial_banner(con
);
2065 ret
= process_banner(con
);
2069 /* Banner is good, exchange connection info */
2070 ret
= prepare_write_connect(con
);
2073 prepare_read_connect(con
);
2074 set_bit(NEGOTIATING
, &con
->state
);
2076 /* Send connection info before awaiting response */
2079 ret
= read_partial_connect(con
);
2082 ret
= process_connect(con
);
2088 if (con
->in_base_pos
< 0) {
2090 * skipping + discarding content.
2092 * FIXME: there must be a better way to do this!
2094 static char buf
[SKIP_BUF_SIZE
];
2095 int skip
= min((int) sizeof (buf
), -con
->in_base_pos
);
2097 dout("skipping %d / %d bytes\n", skip
, -con
->in_base_pos
);
2098 ret
= ceph_tcp_recvmsg(con
->sock
, buf
, skip
);
2101 con
->in_base_pos
+= ret
;
2102 if (con
->in_base_pos
)
2105 if (con
->in_tag
== CEPH_MSGR_TAG_READY
) {
2109 ret
= ceph_tcp_recvmsg(con
->sock
, &con
->in_tag
, 1);
2112 dout("try_read got tag %d\n", (int)con
->in_tag
);
2113 switch (con
->in_tag
) {
2114 case CEPH_MSGR_TAG_MSG
:
2115 prepare_read_message(con
);
2117 case CEPH_MSGR_TAG_ACK
:
2118 prepare_read_ack(con
);
2120 case CEPH_MSGR_TAG_CLOSE
:
2121 clear_bit(CONNECTED
, &con
->state
);
2122 set_bit(CLOSED
, &con
->state
); /* fixme */
2128 if (con
->in_tag
== CEPH_MSGR_TAG_MSG
) {
2129 ret
= read_partial_message(con
);
2133 con
->error_msg
= "bad crc";
2137 con
->error_msg
= "io error";
2142 if (con
->in_tag
== CEPH_MSGR_TAG_READY
)
2144 process_message(con
);
2147 if (con
->in_tag
== CEPH_MSGR_TAG_ACK
) {
2148 ret
= read_partial_ack(con
);
2156 dout("try_read done on %p ret %d\n", con
, ret
);
2160 pr_err("try_read bad con->in_tag = %d\n", (int)con
->in_tag
);
2161 con
->error_msg
= "protocol error, garbage tag";
2168 * Atomically queue work on a connection. Bump @con reference to
2169 * avoid races with connection teardown.
2171 static void queue_con(struct ceph_connection
*con
)
2173 if (!con
->ops
->get(con
)) {
2174 dout("queue_con %p ref count 0\n", con
);
2178 if (!queue_delayed_work(ceph_msgr_wq
, &con
->work
, 0)) {
2179 dout("queue_con %p - already queued\n", con
);
2182 dout("queue_con %p\n", con
);
2187 * Do some work on a connection. Drop a connection ref when we're done.
2189 static void con_work(struct work_struct
*work
)
2191 struct ceph_connection
*con
= container_of(work
, struct ceph_connection
,
2195 mutex_lock(&con
->mutex
);
2197 if (test_and_clear_bit(SOCK_CLOSED
, &con
->flags
)) {
2198 if (test_and_clear_bit(CONNECTED
, &con
->state
))
2199 con
->error_msg
= "socket closed";
2200 else if (test_and_clear_bit(CONNECTING
, &con
->state
)) {
2201 clear_bit(NEGOTIATING
, &con
->state
);
2202 con
->error_msg
= "connection failed";
2204 con
->error_msg
= "unrecognized con state";
2209 if (test_and_clear_bit(BACKOFF
, &con
->flags
)) {
2210 dout("con_work %p backing off\n", con
);
2211 if (queue_delayed_work(ceph_msgr_wq
, &con
->work
,
2212 round_jiffies_relative(con
->delay
))) {
2213 dout("con_work %p backoff %lu\n", con
, con
->delay
);
2214 mutex_unlock(&con
->mutex
);
2218 dout("con_work %p FAILED to back off %lu\n", con
,
2223 if (test_bit(STANDBY
, &con
->state
)) {
2224 dout("con_work %p STANDBY\n", con
);
2227 if (test_bit(CLOSED
, &con
->state
)) { /* e.g. if we are replaced */
2228 dout("con_work CLOSED\n");
2229 con_close_socket(con
);
2232 if (test_and_clear_bit(OPENING
, &con
->state
)) {
2233 /* reopen w/ new peer */
2234 dout("con_work OPENING\n");
2235 con_close_socket(con
);
2238 ret
= try_read(con
);
2244 ret
= try_write(con
);
2251 mutex_unlock(&con
->mutex
);
2257 mutex_unlock(&con
->mutex
);
2258 ceph_fault(con
); /* error/fault path */
2264 * Generic error/fault handler. A retry mechanism is used with
2265 * exponential backoff
2267 static void ceph_fault(struct ceph_connection
*con
)
2269 pr_err("%s%lld %s %s\n", ENTITY_NAME(con
->peer_name
),
2270 ceph_pr_addr(&con
->peer_addr
.in_addr
), con
->error_msg
);
2271 dout("fault %p state %lu to peer %s\n",
2272 con
, con
->state
, ceph_pr_addr(&con
->peer_addr
.in_addr
));
2274 if (test_bit(LOSSYTX
, &con
->flags
)) {
2275 dout("fault on LOSSYTX channel\n");
2279 mutex_lock(&con
->mutex
);
2280 if (test_bit(CLOSED
, &con
->state
))
2283 con_close_socket(con
);
2286 BUG_ON(con
->in_msg
->con
!= con
);
2287 con
->in_msg
->con
= NULL
;
2288 ceph_msg_put(con
->in_msg
);
2293 /* Requeue anything that hasn't been acked */
2294 list_splice_init(&con
->out_sent
, &con
->out_queue
);
2296 /* If there are no messages queued or keepalive pending, place
2297 * the connection in a STANDBY state */
2298 if (list_empty(&con
->out_queue
) &&
2299 !test_bit(KEEPALIVE_PENDING
, &con
->flags
)) {
2300 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con
);
2301 clear_bit(WRITE_PENDING
, &con
->flags
);
2302 set_bit(STANDBY
, &con
->state
);
2304 /* retry after a delay. */
2305 if (con
->delay
== 0)
2306 con
->delay
= BASE_DELAY_INTERVAL
;
2307 else if (con
->delay
< MAX_DELAY_INTERVAL
)
2310 if (queue_delayed_work(ceph_msgr_wq
, &con
->work
,
2311 round_jiffies_relative(con
->delay
))) {
2312 dout("fault queued %p delay %lu\n", con
, con
->delay
);
2315 dout("fault failed to queue %p delay %lu, backoff\n",
2318 * In many cases we see a socket state change
2319 * while con_work is running and end up
2320 * queuing (non-delayed) work, such that we
2321 * can't backoff with a delay. Set a flag so
2322 * that when con_work restarts we schedule the
2325 set_bit(BACKOFF
, &con
->flags
);
2330 mutex_unlock(&con
->mutex
);
2333 * in case we faulted due to authentication, invalidate our
2334 * current tickets so that we can get new ones.
2336 if (con
->auth_retry
&& con
->ops
->invalidate_authorizer
) {
2337 dout("calling invalidate_authorizer()\n");
2338 con
->ops
->invalidate_authorizer(con
);
2341 if (con
->ops
->fault
)
2342 con
->ops
->fault(con
);
2348 * initialize a new messenger instance
2350 void ceph_messenger_init(struct ceph_messenger
*msgr
,
2351 struct ceph_entity_addr
*myaddr
,
2352 u32 supported_features
,
2353 u32 required_features
,
2356 msgr
->supported_features
= supported_features
;
2357 msgr
->required_features
= required_features
;
2359 spin_lock_init(&msgr
->global_seq_lock
);
2362 msgr
->inst
.addr
= *myaddr
;
2364 /* select a random nonce */
2365 msgr
->inst
.addr
.type
= 0;
2366 get_random_bytes(&msgr
->inst
.addr
.nonce
, sizeof(msgr
->inst
.addr
.nonce
));
2367 encode_my_addr(msgr
);
2368 msgr
->nocrc
= nocrc
;
2370 dout("%s %p\n", __func__
, msgr
);
2372 EXPORT_SYMBOL(ceph_messenger_init
);
2374 static void clear_standby(struct ceph_connection
*con
)
2376 /* come back from STANDBY? */
2377 if (test_and_clear_bit(STANDBY
, &con
->state
)) {
2378 mutex_lock(&con
->mutex
);
2379 dout("clear_standby %p and ++connect_seq\n", con
);
2381 WARN_ON(test_bit(WRITE_PENDING
, &con
->flags
));
2382 WARN_ON(test_bit(KEEPALIVE_PENDING
, &con
->flags
));
2383 mutex_unlock(&con
->mutex
);
2388 * Queue up an outgoing message on the given connection.
2390 void ceph_con_send(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2392 if (test_bit(CLOSED
, &con
->state
)) {
2393 dout("con_send %p closed, dropping %p\n", con
, msg
);
2399 msg
->hdr
.src
= con
->msgr
->inst
.name
;
2401 BUG_ON(msg
->front
.iov_len
!= le32_to_cpu(msg
->hdr
.front_len
));
2403 msg
->needs_out_seq
= true;
2406 mutex_lock(&con
->mutex
);
2408 BUG_ON(msg
->con
!= NULL
);
2409 msg
->con
= con
->ops
->get(con
);
2410 BUG_ON(msg
->con
== NULL
);
2412 BUG_ON(!list_empty(&msg
->list_head
));
2413 list_add_tail(&msg
->list_head
, &con
->out_queue
);
2414 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg
,
2415 ENTITY_NAME(con
->peer_name
), le16_to_cpu(msg
->hdr
.type
),
2416 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2417 le32_to_cpu(msg
->hdr
.front_len
),
2418 le32_to_cpu(msg
->hdr
.middle_len
),
2419 le32_to_cpu(msg
->hdr
.data_len
));
2420 mutex_unlock(&con
->mutex
);
2422 /* if there wasn't anything waiting to send before, queue
2425 if (test_and_set_bit(WRITE_PENDING
, &con
->flags
) == 0)
2428 EXPORT_SYMBOL(ceph_con_send
);
2431 * Revoke a message that was previously queued for send
2433 void ceph_msg_revoke(struct ceph_msg
*msg
)
2435 struct ceph_connection
*con
= msg
->con
;
2438 return; /* Message not in our possession */
2440 mutex_lock(&con
->mutex
);
2441 if (!list_empty(&msg
->list_head
)) {
2442 dout("%s %p msg %p - was on queue\n", __func__
, con
, msg
);
2443 list_del_init(&msg
->list_head
);
2444 BUG_ON(msg
->con
== NULL
);
2445 msg
->con
->ops
->put(msg
->con
);
2451 if (con
->out_msg
== msg
) {
2452 dout("%s %p msg %p - was sending\n", __func__
, con
, msg
);
2453 con
->out_msg
= NULL
;
2454 if (con
->out_kvec_is_msg
) {
2455 con
->out_skip
= con
->out_kvec_bytes
;
2456 con
->out_kvec_is_msg
= false;
2462 mutex_unlock(&con
->mutex
);
2466 * Revoke a message that we may be reading data into
2468 void ceph_msg_revoke_incoming(struct ceph_msg
*msg
)
2470 struct ceph_connection
*con
;
2472 BUG_ON(msg
== NULL
);
2474 dout("%s msg %p null con\n", __func__
, msg
);
2476 return; /* Message not in our possession */
2480 mutex_lock(&con
->mutex
);
2481 if (con
->in_msg
== msg
) {
2482 unsigned int front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
2483 unsigned int middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
2484 unsigned int data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
2486 /* skip rest of message */
2487 dout("%s %p msg %p revoked\n", __func__
, con
, msg
);
2488 con
->in_base_pos
= con
->in_base_pos
-
2489 sizeof(struct ceph_msg_header
) -
2493 sizeof(struct ceph_msg_footer
);
2494 ceph_msg_put(con
->in_msg
);
2496 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2499 dout("%s %p in_msg %p msg %p no-op\n",
2500 __func__
, con
, con
->in_msg
, msg
);
2502 mutex_unlock(&con
->mutex
);
2506 * Queue a keepalive byte to ensure the tcp connection is alive.
2508 void ceph_con_keepalive(struct ceph_connection
*con
)
2510 dout("con_keepalive %p\n", con
);
2512 if (test_and_set_bit(KEEPALIVE_PENDING
, &con
->flags
) == 0 &&
2513 test_and_set_bit(WRITE_PENDING
, &con
->flags
) == 0)
2516 EXPORT_SYMBOL(ceph_con_keepalive
);
2520 * construct a new message with given type, size
2521 * the new msg has a ref count of 1.
2523 struct ceph_msg
*ceph_msg_new(int type
, int front_len
, gfp_t flags
,
2528 m
= kmalloc(sizeof(*m
), flags
);
2531 kref_init(&m
->kref
);
2534 INIT_LIST_HEAD(&m
->list_head
);
2537 m
->hdr
.type
= cpu_to_le16(type
);
2538 m
->hdr
.priority
= cpu_to_le16(CEPH_MSG_PRIO_DEFAULT
);
2540 m
->hdr
.front_len
= cpu_to_le32(front_len
);
2541 m
->hdr
.middle_len
= 0;
2542 m
->hdr
.data_len
= 0;
2543 m
->hdr
.data_off
= 0;
2544 m
->hdr
.reserved
= 0;
2545 m
->footer
.front_crc
= 0;
2546 m
->footer
.middle_crc
= 0;
2547 m
->footer
.data_crc
= 0;
2548 m
->footer
.flags
= 0;
2549 m
->front_max
= front_len
;
2550 m
->front_is_vmalloc
= false;
2551 m
->more_to_follow
= false;
2560 m
->page_alignment
= 0;
2570 if (front_len
> PAGE_CACHE_SIZE
) {
2571 m
->front
.iov_base
= __vmalloc(front_len
, flags
,
2573 m
->front_is_vmalloc
= true;
2575 m
->front
.iov_base
= kmalloc(front_len
, flags
);
2577 if (m
->front
.iov_base
== NULL
) {
2578 dout("ceph_msg_new can't allocate %d bytes\n",
2583 m
->front
.iov_base
= NULL
;
2585 m
->front
.iov_len
= front_len
;
2587 dout("ceph_msg_new %p front %d\n", m
, front_len
);
2594 pr_err("msg_new can't create type %d front %d\n", type
,
2598 dout("msg_new can't create type %d front %d\n", type
,
2603 EXPORT_SYMBOL(ceph_msg_new
);
2606 * Allocate "middle" portion of a message, if it is needed and wasn't
2607 * allocated by alloc_msg. This allows us to read a small fixed-size
2608 * per-type header in the front and then gracefully fail (i.e.,
2609 * propagate the error to the caller based on info in the front) when
2610 * the middle is too large.
2612 static int ceph_alloc_middle(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2614 int type
= le16_to_cpu(msg
->hdr
.type
);
2615 int middle_len
= le32_to_cpu(msg
->hdr
.middle_len
);
2617 dout("alloc_middle %p type %d %s middle_len %d\n", msg
, type
,
2618 ceph_msg_type_name(type
), middle_len
);
2619 BUG_ON(!middle_len
);
2620 BUG_ON(msg
->middle
);
2622 msg
->middle
= ceph_buffer_new(middle_len
, GFP_NOFS
);
2629 * Allocate a message for receiving an incoming message on a
2630 * connection, and save the result in con->in_msg. Uses the
2631 * connection's private alloc_msg op if available.
2633 * Returns true if the message should be skipped, false otherwise.
2634 * If true is returned (skip message), con->in_msg will be NULL.
2635 * If false is returned, con->in_msg will contain a pointer to the
2636 * newly-allocated message, or NULL in case of memory exhaustion.
2638 static bool ceph_con_in_msg_alloc(struct ceph_connection
*con
,
2639 struct ceph_msg_header
*hdr
)
2641 int type
= le16_to_cpu(hdr
->type
);
2642 int front_len
= le32_to_cpu(hdr
->front_len
);
2643 int middle_len
= le32_to_cpu(hdr
->middle_len
);
2646 BUG_ON(con
->in_msg
!= NULL
);
2648 if (con
->ops
->alloc_msg
) {
2651 mutex_unlock(&con
->mutex
);
2652 con
->in_msg
= con
->ops
->alloc_msg(con
, hdr
, &skip
);
2653 mutex_lock(&con
->mutex
);
2655 con
->in_msg
->con
= con
->ops
->get(con
);
2656 BUG_ON(con
->in_msg
->con
== NULL
);
2665 con
->in_msg
= ceph_msg_new(type
, front_len
, GFP_NOFS
, false);
2667 pr_err("unable to allocate msg type %d len %d\n",
2671 con
->in_msg
->con
= con
->ops
->get(con
);
2672 BUG_ON(con
->in_msg
->con
== NULL
);
2673 con
->in_msg
->page_alignment
= le16_to_cpu(hdr
->data_off
);
2675 memcpy(&con
->in_msg
->hdr
, &con
->in_hdr
, sizeof(con
->in_hdr
));
2677 if (middle_len
&& !con
->in_msg
->middle
) {
2678 ret
= ceph_alloc_middle(con
, con
->in_msg
);
2680 ceph_msg_put(con
->in_msg
);
2690 * Free a generically kmalloc'd message.
2692 void ceph_msg_kfree(struct ceph_msg
*m
)
2694 dout("msg_kfree %p\n", m
);
2695 if (m
->front_is_vmalloc
)
2696 vfree(m
->front
.iov_base
);
2698 kfree(m
->front
.iov_base
);
2703 * Drop a msg ref. Destroy as needed.
2705 void ceph_msg_last_put(struct kref
*kref
)
2707 struct ceph_msg
*m
= container_of(kref
, struct ceph_msg
, kref
);
2709 dout("ceph_msg_put last one on %p\n", m
);
2710 WARN_ON(!list_empty(&m
->list_head
));
2712 /* drop middle, data, if any */
2714 ceph_buffer_put(m
->middle
);
2721 ceph_pagelist_release(m
->pagelist
);
2729 ceph_msgpool_put(m
->pool
, m
);
2733 EXPORT_SYMBOL(ceph_msg_last_put
);
2735 void ceph_msg_dump(struct ceph_msg
*msg
)
2737 pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg
,
2738 msg
->front_max
, msg
->nr_pages
);
2739 print_hex_dump(KERN_DEBUG
, "header: ",
2740 DUMP_PREFIX_OFFSET
, 16, 1,
2741 &msg
->hdr
, sizeof(msg
->hdr
), true);
2742 print_hex_dump(KERN_DEBUG
, " front: ",
2743 DUMP_PREFIX_OFFSET
, 16, 1,
2744 msg
->front
.iov_base
, msg
->front
.iov_len
, true);
2746 print_hex_dump(KERN_DEBUG
, "middle: ",
2747 DUMP_PREFIX_OFFSET
, 16, 1,
2748 msg
->middle
->vec
.iov_base
,
2749 msg
->middle
->vec
.iov_len
, true);
2750 print_hex_dump(KERN_DEBUG
, "footer: ",
2751 DUMP_PREFIX_OFFSET
, 16, 1,
2752 &msg
->footer
, sizeof(msg
->footer
), true);
2754 EXPORT_SYMBOL(ceph_msg_dump
);