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 if (test_and_set_bit(SOCK_CLOSED
, &con
->flags
) == 0) {
265 if (test_bit(CONNECTING
, &con
->state
))
266 con
->error_msg
= "connection failed";
268 con
->error_msg
= "socket closed";
272 case TCP_ESTABLISHED
:
273 dout("%s TCP_ESTABLISHED\n", __func__
);
274 con_sock_state_connected(con
);
277 default: /* Everything else is uninteresting */
283 * set up socket callbacks
285 static void set_sock_callbacks(struct socket
*sock
,
286 struct ceph_connection
*con
)
288 struct sock
*sk
= sock
->sk
;
289 sk
->sk_user_data
= con
;
290 sk
->sk_data_ready
= ceph_sock_data_ready
;
291 sk
->sk_write_space
= ceph_sock_write_space
;
292 sk
->sk_state_change
= ceph_sock_state_change
;
301 * initiate connection to a remote socket.
303 static int ceph_tcp_connect(struct ceph_connection
*con
)
305 struct sockaddr_storage
*paddr
= &con
->peer_addr
.in_addr
;
310 ret
= sock_create_kern(con
->peer_addr
.in_addr
.ss_family
, SOCK_STREAM
,
314 sock
->sk
->sk_allocation
= GFP_NOFS
;
316 #ifdef CONFIG_LOCKDEP
317 lockdep_set_class(&sock
->sk
->sk_lock
, &socket_class
);
320 set_sock_callbacks(sock
, con
);
322 dout("connect %s\n", ceph_pr_addr(&con
->peer_addr
.in_addr
));
324 ret
= sock
->ops
->connect(sock
, (struct sockaddr
*)paddr
, sizeof(*paddr
),
326 if (ret
== -EINPROGRESS
) {
327 dout("connect %s EINPROGRESS sk_state = %u\n",
328 ceph_pr_addr(&con
->peer_addr
.in_addr
),
330 } else if (ret
< 0) {
331 pr_err("connect %s error %d\n",
332 ceph_pr_addr(&con
->peer_addr
.in_addr
), ret
);
334 con
->error_msg
= "connect error";
339 con_sock_state_connecting(con
);
344 static int ceph_tcp_recvmsg(struct socket
*sock
, void *buf
, size_t len
)
346 struct kvec iov
= {buf
, len
};
347 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
350 r
= kernel_recvmsg(sock
, &msg
, &iov
, 1, len
, msg
.msg_flags
);
357 * write something. @more is true if caller will be sending more data
360 static int ceph_tcp_sendmsg(struct socket
*sock
, struct kvec
*iov
,
361 size_t kvlen
, size_t len
, int more
)
363 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
367 msg
.msg_flags
|= MSG_MORE
;
369 msg
.msg_flags
|= MSG_EOR
; /* superfluous, but what the hell */
371 r
= kernel_sendmsg(sock
, &msg
, iov
, kvlen
, len
);
377 static int ceph_tcp_sendpage(struct socket
*sock
, struct page
*page
,
378 int offset
, size_t size
, int more
)
380 int flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
| (more
? MSG_MORE
: MSG_EOR
);
383 ret
= kernel_sendpage(sock
, page
, offset
, size
, flags
);
392 * Shutdown/close the socket for the given connection.
394 static int con_close_socket(struct ceph_connection
*con
)
398 dout("con_close_socket on %p sock %p\n", con
, con
->sock
);
401 set_bit(SOCK_CLOSED
, &con
->state
);
402 rc
= con
->sock
->ops
->shutdown(con
->sock
, SHUT_RDWR
);
403 sock_release(con
->sock
);
405 clear_bit(SOCK_CLOSED
, &con
->state
);
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
);
419 static void ceph_msg_remove_list(struct list_head
*head
)
421 while (!list_empty(head
)) {
422 struct ceph_msg
*msg
= list_first_entry(head
, struct ceph_msg
,
424 ceph_msg_remove(msg
);
428 static void reset_connection(struct ceph_connection
*con
)
430 /* reset connection, out_queue, msg_ and connect_seq */
431 /* discard existing out_queue and msg_seq */
432 ceph_msg_remove_list(&con
->out_queue
);
433 ceph_msg_remove_list(&con
->out_sent
);
436 ceph_msg_put(con
->in_msg
);
440 con
->connect_seq
= 0;
443 ceph_msg_put(con
->out_msg
);
447 con
->in_seq_acked
= 0;
451 * mark a peer down. drop any open connections.
453 void ceph_con_close(struct ceph_connection
*con
)
455 dout("con_close %p peer %s\n", con
,
456 ceph_pr_addr(&con
->peer_addr
.in_addr
));
457 set_bit(CLOSED
, &con
->state
); /* in case there's queued work */
458 clear_bit(STANDBY
, &con
->state
); /* avoid connect_seq bump */
459 clear_bit(LOSSYTX
, &con
->flags
); /* so we retry next connect */
460 clear_bit(KEEPALIVE_PENDING
, &con
->flags
);
461 clear_bit(WRITE_PENDING
, &con
->flags
);
462 mutex_lock(&con
->mutex
);
463 reset_connection(con
);
464 con
->peer_global_seq
= 0;
465 cancel_delayed_work(&con
->work
);
466 mutex_unlock(&con
->mutex
);
469 EXPORT_SYMBOL(ceph_con_close
);
472 * Reopen a closed connection, with a new peer address.
474 void ceph_con_open(struct ceph_connection
*con
, struct ceph_entity_addr
*addr
)
476 dout("con_open %p %s\n", con
, ceph_pr_addr(&addr
->in_addr
));
477 set_bit(OPENING
, &con
->state
);
478 clear_bit(CLOSED
, &con
->state
);
479 memcpy(&con
->peer_addr
, addr
, sizeof(*addr
));
480 con
->delay
= 0; /* reset backoff memory */
483 EXPORT_SYMBOL(ceph_con_open
);
486 * return true if this connection ever successfully opened
488 bool ceph_con_opened(struct ceph_connection
*con
)
490 return con
->connect_seq
> 0;
496 struct ceph_connection
*ceph_con_get(struct ceph_connection
*con
)
498 int nref
= __atomic_add_unless(&con
->nref
, 1, 0);
500 dout("con_get %p nref = %d -> %d\n", con
, nref
, nref
+ 1);
502 return nref
? con
: NULL
;
505 void ceph_con_put(struct ceph_connection
*con
)
507 int nref
= atomic_dec_return(&con
->nref
);
514 dout("con_put %p nref = %d -> %d\n", con
, nref
+ 1, nref
);
518 * initialize a new connection.
520 void ceph_con_init(struct ceph_messenger
*msgr
, struct ceph_connection
*con
)
522 dout("con_init %p\n", con
);
523 memset(con
, 0, sizeof(*con
));
524 atomic_set(&con
->nref
, 1);
527 con_sock_state_init(con
);
529 mutex_init(&con
->mutex
);
530 INIT_LIST_HEAD(&con
->out_queue
);
531 INIT_LIST_HEAD(&con
->out_sent
);
532 INIT_DELAYED_WORK(&con
->work
, con_work
);
534 EXPORT_SYMBOL(ceph_con_init
);
538 * We maintain a global counter to order connection attempts. Get
539 * a unique seq greater than @gt.
541 static u32
get_global_seq(struct ceph_messenger
*msgr
, u32 gt
)
545 spin_lock(&msgr
->global_seq_lock
);
546 if (msgr
->global_seq
< gt
)
547 msgr
->global_seq
= gt
;
548 ret
= ++msgr
->global_seq
;
549 spin_unlock(&msgr
->global_seq_lock
);
553 static void con_out_kvec_reset(struct ceph_connection
*con
)
555 con
->out_kvec_left
= 0;
556 con
->out_kvec_bytes
= 0;
557 con
->out_kvec_cur
= &con
->out_kvec
[0];
560 static void con_out_kvec_add(struct ceph_connection
*con
,
561 size_t size
, void *data
)
565 index
= con
->out_kvec_left
;
566 BUG_ON(index
>= ARRAY_SIZE(con
->out_kvec
));
568 con
->out_kvec
[index
].iov_len
= size
;
569 con
->out_kvec
[index
].iov_base
= data
;
570 con
->out_kvec_left
++;
571 con
->out_kvec_bytes
+= size
;
575 * Prepare footer for currently outgoing message, and finish things
576 * off. Assumes out_kvec* are already valid.. we just add on to the end.
578 static void prepare_write_message_footer(struct ceph_connection
*con
)
580 struct ceph_msg
*m
= con
->out_msg
;
581 int v
= con
->out_kvec_left
;
583 dout("prepare_write_message_footer %p\n", con
);
584 con
->out_kvec_is_msg
= true;
585 con
->out_kvec
[v
].iov_base
= &m
->footer
;
586 con
->out_kvec
[v
].iov_len
= sizeof(m
->footer
);
587 con
->out_kvec_bytes
+= sizeof(m
->footer
);
588 con
->out_kvec_left
++;
589 con
->out_more
= m
->more_to_follow
;
590 con
->out_msg_done
= true;
594 * Prepare headers for the next outgoing message.
596 static void prepare_write_message(struct ceph_connection
*con
)
601 con_out_kvec_reset(con
);
602 con
->out_kvec_is_msg
= true;
603 con
->out_msg_done
= false;
605 /* Sneak an ack in there first? If we can get it into the same
606 * TCP packet that's a good thing. */
607 if (con
->in_seq
> con
->in_seq_acked
) {
608 con
->in_seq_acked
= con
->in_seq
;
609 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
610 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
611 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
615 m
= list_first_entry(&con
->out_queue
, struct ceph_msg
, list_head
);
618 /* put message on sent list */
620 list_move_tail(&m
->list_head
, &con
->out_sent
);
623 * only assign outgoing seq # if we haven't sent this message
624 * yet. if it is requeued, resend with it's original seq.
626 if (m
->needs_out_seq
) {
627 m
->hdr
.seq
= cpu_to_le64(++con
->out_seq
);
628 m
->needs_out_seq
= false;
631 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n",
632 m
, con
->out_seq
, le16_to_cpu(m
->hdr
.type
),
633 le32_to_cpu(m
->hdr
.front_len
), le32_to_cpu(m
->hdr
.middle_len
),
634 le32_to_cpu(m
->hdr
.data_len
),
636 BUG_ON(le32_to_cpu(m
->hdr
.front_len
) != m
->front
.iov_len
);
638 /* tag + hdr + front + middle */
639 con_out_kvec_add(con
, sizeof (tag_msg
), &tag_msg
);
640 con_out_kvec_add(con
, sizeof (m
->hdr
), &m
->hdr
);
641 con_out_kvec_add(con
, m
->front
.iov_len
, m
->front
.iov_base
);
644 con_out_kvec_add(con
, m
->middle
->vec
.iov_len
,
645 m
->middle
->vec
.iov_base
);
647 /* fill in crc (except data pages), footer */
648 crc
= crc32c(0, &m
->hdr
, offsetof(struct ceph_msg_header
, crc
));
649 con
->out_msg
->hdr
.crc
= cpu_to_le32(crc
);
650 con
->out_msg
->footer
.flags
= CEPH_MSG_FOOTER_COMPLETE
;
652 crc
= crc32c(0, m
->front
.iov_base
, m
->front
.iov_len
);
653 con
->out_msg
->footer
.front_crc
= cpu_to_le32(crc
);
655 crc
= crc32c(0, m
->middle
->vec
.iov_base
,
656 m
->middle
->vec
.iov_len
);
657 con
->out_msg
->footer
.middle_crc
= cpu_to_le32(crc
);
659 con
->out_msg
->footer
.middle_crc
= 0;
660 con
->out_msg
->footer
.data_crc
= 0;
661 dout("prepare_write_message front_crc %u data_crc %u\n",
662 le32_to_cpu(con
->out_msg
->footer
.front_crc
),
663 le32_to_cpu(con
->out_msg
->footer
.middle_crc
));
665 /* is there a data payload? */
666 if (le32_to_cpu(m
->hdr
.data_len
) > 0) {
667 /* initialize page iterator */
668 con
->out_msg_pos
.page
= 0;
670 con
->out_msg_pos
.page_pos
= m
->page_alignment
;
672 con
->out_msg_pos
.page_pos
= 0;
673 con
->out_msg_pos
.data_pos
= 0;
674 con
->out_msg_pos
.did_page_crc
= false;
675 con
->out_more
= 1; /* data + footer will follow */
677 /* no, queue up footer too and be done */
678 prepare_write_message_footer(con
);
681 set_bit(WRITE_PENDING
, &con
->flags
);
687 static void prepare_write_ack(struct ceph_connection
*con
)
689 dout("prepare_write_ack %p %llu -> %llu\n", con
,
690 con
->in_seq_acked
, con
->in_seq
);
691 con
->in_seq_acked
= con
->in_seq
;
693 con_out_kvec_reset(con
);
695 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
697 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
698 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
701 con
->out_more
= 1; /* more will follow.. eventually.. */
702 set_bit(WRITE_PENDING
, &con
->flags
);
706 * Prepare to write keepalive byte.
708 static void prepare_write_keepalive(struct ceph_connection
*con
)
710 dout("prepare_write_keepalive %p\n", con
);
711 con_out_kvec_reset(con
);
712 con_out_kvec_add(con
, sizeof (tag_keepalive
), &tag_keepalive
);
713 set_bit(WRITE_PENDING
, &con
->flags
);
717 * Connection negotiation.
720 static struct ceph_auth_handshake
*get_connect_authorizer(struct ceph_connection
*con
,
723 struct ceph_auth_handshake
*auth
;
725 if (!con
->ops
->get_authorizer
) {
726 con
->out_connect
.authorizer_protocol
= CEPH_AUTH_UNKNOWN
;
727 con
->out_connect
.authorizer_len
= 0;
732 /* Can't hold the mutex while getting authorizer */
734 mutex_unlock(&con
->mutex
);
736 auth
= con
->ops
->get_authorizer(con
, auth_proto
, con
->auth_retry
);
738 mutex_lock(&con
->mutex
);
742 if (test_bit(CLOSED
, &con
->state
) || test_bit(OPENING
, &con
->flags
))
743 return ERR_PTR(-EAGAIN
);
745 con
->auth_reply_buf
= auth
->authorizer_reply_buf
;
746 con
->auth_reply_buf_len
= auth
->authorizer_reply_buf_len
;
753 * We connected to a peer and are saying hello.
755 static void prepare_write_banner(struct ceph_connection
*con
)
757 con_out_kvec_add(con
, strlen(CEPH_BANNER
), CEPH_BANNER
);
758 con_out_kvec_add(con
, sizeof (con
->msgr
->my_enc_addr
),
759 &con
->msgr
->my_enc_addr
);
762 set_bit(WRITE_PENDING
, &con
->flags
);
765 static int prepare_write_connect(struct ceph_connection
*con
)
767 unsigned global_seq
= get_global_seq(con
->msgr
, 0);
770 struct ceph_auth_handshake
*auth
;
772 switch (con
->peer_name
.type
) {
773 case CEPH_ENTITY_TYPE_MON
:
774 proto
= CEPH_MONC_PROTOCOL
;
776 case CEPH_ENTITY_TYPE_OSD
:
777 proto
= CEPH_OSDC_PROTOCOL
;
779 case CEPH_ENTITY_TYPE_MDS
:
780 proto
= CEPH_MDSC_PROTOCOL
;
786 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con
,
787 con
->connect_seq
, global_seq
, proto
);
789 con
->out_connect
.features
= cpu_to_le64(con
->msgr
->supported_features
);
790 con
->out_connect
.host_type
= cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT
);
791 con
->out_connect
.connect_seq
= cpu_to_le32(con
->connect_seq
);
792 con
->out_connect
.global_seq
= cpu_to_le32(global_seq
);
793 con
->out_connect
.protocol_version
= cpu_to_le32(proto
);
794 con
->out_connect
.flags
= 0;
796 auth_proto
= CEPH_AUTH_UNKNOWN
;
797 auth
= get_connect_authorizer(con
, &auth_proto
);
799 return PTR_ERR(auth
);
801 con
->out_connect
.authorizer_protocol
= cpu_to_le32(auth_proto
);
802 con
->out_connect
.authorizer_len
= auth
?
803 cpu_to_le32(auth
->authorizer_buf_len
) : 0;
805 con_out_kvec_add(con
, sizeof (con
->out_connect
),
807 if (auth
&& auth
->authorizer_buf_len
)
808 con_out_kvec_add(con
, auth
->authorizer_buf_len
,
809 auth
->authorizer_buf
);
812 set_bit(WRITE_PENDING
, &con
->flags
);
818 * write as much of pending kvecs to the socket as we can.
820 * 0 -> socket full, but more to do
823 static int write_partial_kvec(struct ceph_connection
*con
)
827 dout("write_partial_kvec %p %d left\n", con
, con
->out_kvec_bytes
);
828 while (con
->out_kvec_bytes
> 0) {
829 ret
= ceph_tcp_sendmsg(con
->sock
, con
->out_kvec_cur
,
830 con
->out_kvec_left
, con
->out_kvec_bytes
,
834 con
->out_kvec_bytes
-= ret
;
835 if (con
->out_kvec_bytes
== 0)
838 /* account for full iov entries consumed */
839 while (ret
>= con
->out_kvec_cur
->iov_len
) {
840 BUG_ON(!con
->out_kvec_left
);
841 ret
-= con
->out_kvec_cur
->iov_len
;
843 con
->out_kvec_left
--;
845 /* and for a partially-consumed entry */
847 con
->out_kvec_cur
->iov_len
-= ret
;
848 con
->out_kvec_cur
->iov_base
+= ret
;
851 con
->out_kvec_left
= 0;
852 con
->out_kvec_is_msg
= false;
855 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con
,
856 con
->out_kvec_bytes
, con
->out_kvec_left
, ret
);
857 return ret
; /* done! */
861 static void init_bio_iter(struct bio
*bio
, struct bio
**iter
, int *seg
)
872 static void iter_bio_next(struct bio
**bio_iter
, int *seg
)
874 if (*bio_iter
== NULL
)
877 BUG_ON(*seg
>= (*bio_iter
)->bi_vcnt
);
880 if (*seg
== (*bio_iter
)->bi_vcnt
)
881 init_bio_iter((*bio_iter
)->bi_next
, bio_iter
, seg
);
886 * Write as much message data payload as we can. If we finish, queue
888 * 1 -> done, footer is now queued in out_kvec[].
889 * 0 -> socket full, but more to do
892 static int write_partial_msg_pages(struct ceph_connection
*con
)
894 struct ceph_msg
*msg
= con
->out_msg
;
895 unsigned data_len
= le32_to_cpu(msg
->hdr
.data_len
);
897 bool do_datacrc
= !con
->msgr
->nocrc
;
901 size_t trail_len
= (msg
->trail
? msg
->trail
->length
: 0);
903 dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
904 con
, con
->out_msg
, con
->out_msg_pos
.page
, con
->out_msg
->nr_pages
,
905 con
->out_msg_pos
.page_pos
);
908 if (msg
->bio
&& !msg
->bio_iter
)
909 init_bio_iter(msg
->bio
, &msg
->bio_iter
, &msg
->bio_seg
);
912 while (data_len
> con
->out_msg_pos
.data_pos
) {
913 struct page
*page
= NULL
;
914 int max_write
= PAGE_SIZE
;
917 total_max_write
= data_len
- trail_len
-
918 con
->out_msg_pos
.data_pos
;
921 * if we are calculating the data crc (the default), we need
922 * to map the page. if our pages[] has been revoked, use the
926 /* have we reached the trail part of the data? */
927 if (con
->out_msg_pos
.data_pos
>= data_len
- trail_len
) {
930 total_max_write
= data_len
- con
->out_msg_pos
.data_pos
;
932 page
= list_first_entry(&msg
->trail
->head
,
934 max_write
= PAGE_SIZE
;
935 } else if (msg
->pages
) {
936 page
= msg
->pages
[con
->out_msg_pos
.page
];
937 } else if (msg
->pagelist
) {
938 page
= list_first_entry(&msg
->pagelist
->head
,
941 } else if (msg
->bio
) {
944 bv
= bio_iovec_idx(msg
->bio_iter
, msg
->bio_seg
);
946 bio_offset
= bv
->bv_offset
;
947 max_write
= bv
->bv_len
;
952 len
= min_t(int, max_write
- con
->out_msg_pos
.page_pos
,
955 if (do_datacrc
&& !con
->out_msg_pos
.did_page_crc
) {
958 u32 tmpcrc
= le32_to_cpu(con
->out_msg
->footer
.data_crc
);
962 BUG_ON(kaddr
== NULL
);
963 base
= kaddr
+ con
->out_msg_pos
.page_pos
+ bio_offset
;
964 crc
= crc32c(tmpcrc
, base
, len
);
965 con
->out_msg
->footer
.data_crc
= cpu_to_le32(crc
);
966 con
->out_msg_pos
.did_page_crc
= true;
968 ret
= ceph_tcp_sendpage(con
->sock
, page
,
969 con
->out_msg_pos
.page_pos
+ bio_offset
,
978 con
->out_msg_pos
.data_pos
+= ret
;
979 con
->out_msg_pos
.page_pos
+= ret
;
981 con
->out_msg_pos
.page_pos
= 0;
982 con
->out_msg_pos
.page
++;
983 con
->out_msg_pos
.did_page_crc
= false;
985 list_move_tail(&page
->lru
,
987 else if (msg
->pagelist
)
988 list_move_tail(&page
->lru
,
989 &msg
->pagelist
->head
);
992 iter_bio_next(&msg
->bio_iter
, &msg
->bio_seg
);
997 dout("write_partial_msg_pages %p msg %p done\n", con
, msg
);
999 /* prepare and queue up footer, too */
1001 con
->out_msg
->footer
.flags
|= CEPH_MSG_FOOTER_NOCRC
;
1002 con_out_kvec_reset(con
);
1003 prepare_write_message_footer(con
);
1012 static int write_partial_skip(struct ceph_connection
*con
)
1016 while (con
->out_skip
> 0) {
1017 size_t size
= min(con
->out_skip
, (int) PAGE_CACHE_SIZE
);
1019 ret
= ceph_tcp_sendpage(con
->sock
, zero_page
, 0, size
, 1);
1022 con
->out_skip
-= ret
;
1030 * Prepare to read connection handshake, or an ack.
1032 static void prepare_read_banner(struct ceph_connection
*con
)
1034 dout("prepare_read_banner %p\n", con
);
1035 con
->in_base_pos
= 0;
1038 static void prepare_read_connect(struct ceph_connection
*con
)
1040 dout("prepare_read_connect %p\n", con
);
1041 con
->in_base_pos
= 0;
1044 static void prepare_read_ack(struct ceph_connection
*con
)
1046 dout("prepare_read_ack %p\n", con
);
1047 con
->in_base_pos
= 0;
1050 static void prepare_read_tag(struct ceph_connection
*con
)
1052 dout("prepare_read_tag %p\n", con
);
1053 con
->in_base_pos
= 0;
1054 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1058 * Prepare to read a message.
1060 static int prepare_read_message(struct ceph_connection
*con
)
1062 dout("prepare_read_message %p\n", con
);
1063 BUG_ON(con
->in_msg
!= NULL
);
1064 con
->in_base_pos
= 0;
1065 con
->in_front_crc
= con
->in_middle_crc
= con
->in_data_crc
= 0;
1070 static int read_partial(struct ceph_connection
*con
,
1071 int end
, int size
, void *object
)
1073 while (con
->in_base_pos
< end
) {
1074 int left
= end
- con
->in_base_pos
;
1075 int have
= size
- left
;
1076 int ret
= ceph_tcp_recvmsg(con
->sock
, object
+ have
, left
);
1079 con
->in_base_pos
+= ret
;
1086 * Read all or part of the connect-side handshake on a new connection
1088 static int read_partial_banner(struct ceph_connection
*con
)
1094 dout("read_partial_banner %p at %d\n", con
, con
->in_base_pos
);
1097 size
= strlen(CEPH_BANNER
);
1099 ret
= read_partial(con
, end
, size
, con
->in_banner
);
1103 size
= sizeof (con
->actual_peer_addr
);
1105 ret
= read_partial(con
, end
, size
, &con
->actual_peer_addr
);
1109 size
= sizeof (con
->peer_addr_for_me
);
1111 ret
= read_partial(con
, end
, size
, &con
->peer_addr_for_me
);
1119 static int read_partial_connect(struct ceph_connection
*con
)
1125 dout("read_partial_connect %p at %d\n", con
, con
->in_base_pos
);
1127 size
= sizeof (con
->in_reply
);
1129 ret
= read_partial(con
, end
, size
, &con
->in_reply
);
1133 size
= le32_to_cpu(con
->in_reply
.authorizer_len
);
1135 ret
= read_partial(con
, end
, size
, con
->auth_reply_buf
);
1139 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1140 con
, (int)con
->in_reply
.tag
,
1141 le32_to_cpu(con
->in_reply
.connect_seq
),
1142 le32_to_cpu(con
->in_reply
.global_seq
));
1149 * Verify the hello banner looks okay.
1151 static int verify_hello(struct ceph_connection
*con
)
1153 if (memcmp(con
->in_banner
, CEPH_BANNER
, strlen(CEPH_BANNER
))) {
1154 pr_err("connect to %s got bad banner\n",
1155 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1156 con
->error_msg
= "protocol error, bad banner";
1162 static bool addr_is_blank(struct sockaddr_storage
*ss
)
1164 switch (ss
->ss_family
) {
1166 return ((struct sockaddr_in
*)ss
)->sin_addr
.s_addr
== 0;
1169 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[0] == 0 &&
1170 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[1] == 0 &&
1171 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[2] == 0 &&
1172 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[3] == 0;
1177 static int addr_port(struct sockaddr_storage
*ss
)
1179 switch (ss
->ss_family
) {
1181 return ntohs(((struct sockaddr_in
*)ss
)->sin_port
);
1183 return ntohs(((struct sockaddr_in6
*)ss
)->sin6_port
);
1188 static void addr_set_port(struct sockaddr_storage
*ss
, int p
)
1190 switch (ss
->ss_family
) {
1192 ((struct sockaddr_in
*)ss
)->sin_port
= htons(p
);
1195 ((struct sockaddr_in6
*)ss
)->sin6_port
= htons(p
);
1201 * Unlike other *_pton function semantics, zero indicates success.
1203 static int ceph_pton(const char *str
, size_t len
, struct sockaddr_storage
*ss
,
1204 char delim
, const char **ipend
)
1206 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
1207 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
1209 memset(ss
, 0, sizeof(*ss
));
1211 if (in4_pton(str
, len
, (u8
*)&in4
->sin_addr
.s_addr
, delim
, ipend
)) {
1212 ss
->ss_family
= AF_INET
;
1216 if (in6_pton(str
, len
, (u8
*)&in6
->sin6_addr
.s6_addr
, delim
, ipend
)) {
1217 ss
->ss_family
= AF_INET6
;
1225 * Extract hostname string and resolve using kernel DNS facility.
1227 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1228 static int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1229 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1231 const char *end
, *delim_p
;
1232 char *colon_p
, *ip_addr
= NULL
;
1236 * The end of the hostname occurs immediately preceding the delimiter or
1237 * the port marker (':') where the delimiter takes precedence.
1239 delim_p
= memchr(name
, delim
, namelen
);
1240 colon_p
= memchr(name
, ':', namelen
);
1242 if (delim_p
&& colon_p
)
1243 end
= delim_p
< colon_p
? delim_p
: colon_p
;
1244 else if (!delim_p
&& colon_p
)
1248 if (!end
) /* case: hostname:/ */
1249 end
= name
+ namelen
;
1255 /* do dns_resolve upcall */
1256 ip_len
= dns_query(NULL
, name
, end
- name
, NULL
, &ip_addr
, NULL
);
1258 ret
= ceph_pton(ip_addr
, ip_len
, ss
, -1, NULL
);
1266 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end
- name
), name
,
1267 ret
, ret
? "failed" : ceph_pr_addr(ss
));
1272 static inline int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1273 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1280 * Parse a server name (IP or hostname). If a valid IP address is not found
1281 * then try to extract a hostname to resolve using userspace DNS upcall.
1283 static int ceph_parse_server_name(const char *name
, size_t namelen
,
1284 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1288 ret
= ceph_pton(name
, namelen
, ss
, delim
, ipend
);
1290 ret
= ceph_dns_resolve_name(name
, namelen
, ss
, delim
, ipend
);
1296 * Parse an ip[:port] list into an addr array. Use the default
1297 * monitor port if a port isn't specified.
1299 int ceph_parse_ips(const char *c
, const char *end
,
1300 struct ceph_entity_addr
*addr
,
1301 int max_count
, int *count
)
1303 int i
, ret
= -EINVAL
;
1306 dout("parse_ips on '%.*s'\n", (int)(end
-c
), c
);
1307 for (i
= 0; i
< max_count
; i
++) {
1309 struct sockaddr_storage
*ss
= &addr
[i
].in_addr
;
1318 ret
= ceph_parse_server_name(p
, end
- p
, ss
, delim
, &ipend
);
1327 dout("missing matching ']'\n");
1334 if (p
< end
&& *p
== ':') {
1337 while (p
< end
&& *p
>= '0' && *p
<= '9') {
1338 port
= (port
* 10) + (*p
- '0');
1341 if (port
> 65535 || port
== 0)
1344 port
= CEPH_MON_PORT
;
1347 addr_set_port(ss
, port
);
1349 dout("parse_ips got %s\n", ceph_pr_addr(ss
));
1366 pr_err("parse_ips bad ip '%.*s'\n", (int)(end
- c
), c
);
1369 EXPORT_SYMBOL(ceph_parse_ips
);
1371 static int process_banner(struct ceph_connection
*con
)
1373 dout("process_banner on %p\n", con
);
1375 if (verify_hello(con
) < 0)
1378 ceph_decode_addr(&con
->actual_peer_addr
);
1379 ceph_decode_addr(&con
->peer_addr_for_me
);
1382 * Make sure the other end is who we wanted. note that the other
1383 * end may not yet know their ip address, so if it's 0.0.0.0, give
1384 * them the benefit of the doubt.
1386 if (memcmp(&con
->peer_addr
, &con
->actual_peer_addr
,
1387 sizeof(con
->peer_addr
)) != 0 &&
1388 !(addr_is_blank(&con
->actual_peer_addr
.in_addr
) &&
1389 con
->actual_peer_addr
.nonce
== con
->peer_addr
.nonce
)) {
1390 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1391 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1392 (int)le32_to_cpu(con
->peer_addr
.nonce
),
1393 ceph_pr_addr(&con
->actual_peer_addr
.in_addr
),
1394 (int)le32_to_cpu(con
->actual_peer_addr
.nonce
));
1395 con
->error_msg
= "wrong peer at address";
1400 * did we learn our address?
1402 if (addr_is_blank(&con
->msgr
->inst
.addr
.in_addr
)) {
1403 int port
= addr_port(&con
->msgr
->inst
.addr
.in_addr
);
1405 memcpy(&con
->msgr
->inst
.addr
.in_addr
,
1406 &con
->peer_addr_for_me
.in_addr
,
1407 sizeof(con
->peer_addr_for_me
.in_addr
));
1408 addr_set_port(&con
->msgr
->inst
.addr
.in_addr
, port
);
1409 encode_my_addr(con
->msgr
);
1410 dout("process_banner learned my addr is %s\n",
1411 ceph_pr_addr(&con
->msgr
->inst
.addr
.in_addr
));
1414 set_bit(NEGOTIATING
, &con
->state
);
1415 prepare_read_connect(con
);
1419 static void fail_protocol(struct ceph_connection
*con
)
1421 reset_connection(con
);
1422 set_bit(CLOSED
, &con
->state
); /* in case there's queued work */
1425 static int process_connect(struct ceph_connection
*con
)
1427 u64 sup_feat
= con
->msgr
->supported_features
;
1428 u64 req_feat
= con
->msgr
->required_features
;
1429 u64 server_feat
= le64_to_cpu(con
->in_reply
.features
);
1432 dout("process_connect on %p tag %d\n", con
, (int)con
->in_tag
);
1434 switch (con
->in_reply
.tag
) {
1435 case CEPH_MSGR_TAG_FEATURES
:
1436 pr_err("%s%lld %s feature set mismatch,"
1437 " my %llx < server's %llx, missing %llx\n",
1438 ENTITY_NAME(con
->peer_name
),
1439 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1440 sup_feat
, server_feat
, server_feat
& ~sup_feat
);
1441 con
->error_msg
= "missing required protocol features";
1445 case CEPH_MSGR_TAG_BADPROTOVER
:
1446 pr_err("%s%lld %s protocol version mismatch,"
1447 " my %d != server's %d\n",
1448 ENTITY_NAME(con
->peer_name
),
1449 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1450 le32_to_cpu(con
->out_connect
.protocol_version
),
1451 le32_to_cpu(con
->in_reply
.protocol_version
));
1452 con
->error_msg
= "protocol version mismatch";
1456 case CEPH_MSGR_TAG_BADAUTHORIZER
:
1458 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con
,
1460 if (con
->auth_retry
== 2) {
1461 con
->error_msg
= "connect authorization failure";
1464 con
->auth_retry
= 1;
1465 con_out_kvec_reset(con
);
1466 ret
= prepare_write_connect(con
);
1469 prepare_read_connect(con
);
1472 case CEPH_MSGR_TAG_RESETSESSION
:
1474 * If we connected with a large connect_seq but the peer
1475 * has no record of a session with us (no connection, or
1476 * connect_seq == 0), they will send RESETSESION to indicate
1477 * that they must have reset their session, and may have
1480 dout("process_connect got RESET peer seq %u\n",
1481 le32_to_cpu(con
->in_connect
.connect_seq
));
1482 pr_err("%s%lld %s connection reset\n",
1483 ENTITY_NAME(con
->peer_name
),
1484 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1485 reset_connection(con
);
1486 con_out_kvec_reset(con
);
1487 ret
= prepare_write_connect(con
);
1490 prepare_read_connect(con
);
1492 /* Tell ceph about it. */
1493 mutex_unlock(&con
->mutex
);
1494 pr_info("reset on %s%lld\n", ENTITY_NAME(con
->peer_name
));
1495 if (con
->ops
->peer_reset
)
1496 con
->ops
->peer_reset(con
);
1497 mutex_lock(&con
->mutex
);
1498 if (test_bit(CLOSED
, &con
->state
) ||
1499 test_bit(OPENING
, &con
->state
))
1503 case CEPH_MSGR_TAG_RETRY_SESSION
:
1505 * If we sent a smaller connect_seq than the peer has, try
1506 * again with a larger value.
1508 dout("process_connect got RETRY my seq = %u, peer_seq = %u\n",
1509 le32_to_cpu(con
->out_connect
.connect_seq
),
1510 le32_to_cpu(con
->in_connect
.connect_seq
));
1511 con
->connect_seq
= le32_to_cpu(con
->in_connect
.connect_seq
);
1512 con_out_kvec_reset(con
);
1513 ret
= prepare_write_connect(con
);
1516 prepare_read_connect(con
);
1519 case CEPH_MSGR_TAG_RETRY_GLOBAL
:
1521 * If we sent a smaller global_seq than the peer has, try
1522 * again with a larger value.
1524 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1525 con
->peer_global_seq
,
1526 le32_to_cpu(con
->in_connect
.global_seq
));
1527 get_global_seq(con
->msgr
,
1528 le32_to_cpu(con
->in_connect
.global_seq
));
1529 con_out_kvec_reset(con
);
1530 ret
= prepare_write_connect(con
);
1533 prepare_read_connect(con
);
1536 case CEPH_MSGR_TAG_READY
:
1537 if (req_feat
& ~server_feat
) {
1538 pr_err("%s%lld %s protocol feature mismatch,"
1539 " my required %llx > server's %llx, need %llx\n",
1540 ENTITY_NAME(con
->peer_name
),
1541 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1542 req_feat
, server_feat
, req_feat
& ~server_feat
);
1543 con
->error_msg
= "missing required protocol features";
1547 clear_bit(CONNECTING
, &con
->state
);
1548 con
->peer_global_seq
= le32_to_cpu(con
->in_reply
.global_seq
);
1550 con
->peer_features
= server_feat
;
1551 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1552 con
->peer_global_seq
,
1553 le32_to_cpu(con
->in_reply
.connect_seq
),
1555 WARN_ON(con
->connect_seq
!=
1556 le32_to_cpu(con
->in_reply
.connect_seq
));
1558 if (con
->in_reply
.flags
& CEPH_MSG_CONNECT_LOSSY
)
1559 set_bit(LOSSYTX
, &con
->flags
);
1561 prepare_read_tag(con
);
1564 case CEPH_MSGR_TAG_WAIT
:
1566 * If there is a connection race (we are opening
1567 * connections to each other), one of us may just have
1568 * to WAIT. This shouldn't happen if we are the
1571 pr_err("process_connect got WAIT as client\n");
1572 con
->error_msg
= "protocol error, got WAIT as client";
1576 pr_err("connect protocol error, will retry\n");
1577 con
->error_msg
= "protocol error, garbage tag during connect";
1585 * read (part of) an ack
1587 static int read_partial_ack(struct ceph_connection
*con
)
1589 int size
= sizeof (con
->in_temp_ack
);
1592 return read_partial(con
, end
, size
, &con
->in_temp_ack
);
1597 * We can finally discard anything that's been acked.
1599 static void process_ack(struct ceph_connection
*con
)
1602 u64 ack
= le64_to_cpu(con
->in_temp_ack
);
1605 while (!list_empty(&con
->out_sent
)) {
1606 m
= list_first_entry(&con
->out_sent
, struct ceph_msg
,
1608 seq
= le64_to_cpu(m
->hdr
.seq
);
1611 dout("got ack for seq %llu type %d at %p\n", seq
,
1612 le16_to_cpu(m
->hdr
.type
), m
);
1613 m
->ack_stamp
= jiffies
;
1616 prepare_read_tag(con
);
1622 static int read_partial_message_section(struct ceph_connection
*con
,
1623 struct kvec
*section
,
1624 unsigned int sec_len
, u32
*crc
)
1630 while (section
->iov_len
< sec_len
) {
1631 BUG_ON(section
->iov_base
== NULL
);
1632 left
= sec_len
- section
->iov_len
;
1633 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)section
->iov_base
+
1634 section
->iov_len
, left
);
1637 section
->iov_len
+= ret
;
1639 if (section
->iov_len
== sec_len
)
1640 *crc
= crc32c(0, section
->iov_base
, section
->iov_len
);
1645 static struct ceph_msg
*ceph_alloc_msg(struct ceph_connection
*con
,
1646 struct ceph_msg_header
*hdr
,
1650 static int read_partial_message_pages(struct ceph_connection
*con
,
1651 struct page
**pages
,
1652 unsigned data_len
, bool do_datacrc
)
1658 left
= min((int)(data_len
- con
->in_msg_pos
.data_pos
),
1659 (int)(PAGE_SIZE
- con
->in_msg_pos
.page_pos
));
1661 BUG_ON(pages
== NULL
);
1662 p
= kmap(pages
[con
->in_msg_pos
.page
]);
1663 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ con
->in_msg_pos
.page_pos
,
1665 if (ret
> 0 && do_datacrc
)
1667 crc32c(con
->in_data_crc
,
1668 p
+ con
->in_msg_pos
.page_pos
, ret
);
1669 kunmap(pages
[con
->in_msg_pos
.page
]);
1672 con
->in_msg_pos
.data_pos
+= ret
;
1673 con
->in_msg_pos
.page_pos
+= ret
;
1674 if (con
->in_msg_pos
.page_pos
== PAGE_SIZE
) {
1675 con
->in_msg_pos
.page_pos
= 0;
1676 con
->in_msg_pos
.page
++;
1683 static int read_partial_message_bio(struct ceph_connection
*con
,
1684 struct bio
**bio_iter
, int *bio_seg
,
1685 unsigned data_len
, bool do_datacrc
)
1687 struct bio_vec
*bv
= bio_iovec_idx(*bio_iter
, *bio_seg
);
1694 left
= min((int)(data_len
- con
->in_msg_pos
.data_pos
),
1695 (int)(bv
->bv_len
- con
->in_msg_pos
.page_pos
));
1697 p
= kmap(bv
->bv_page
) + bv
->bv_offset
;
1699 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ con
->in_msg_pos
.page_pos
,
1701 if (ret
> 0 && do_datacrc
)
1703 crc32c(con
->in_data_crc
,
1704 p
+ con
->in_msg_pos
.page_pos
, ret
);
1705 kunmap(bv
->bv_page
);
1708 con
->in_msg_pos
.data_pos
+= ret
;
1709 con
->in_msg_pos
.page_pos
+= ret
;
1710 if (con
->in_msg_pos
.page_pos
== bv
->bv_len
) {
1711 con
->in_msg_pos
.page_pos
= 0;
1712 iter_bio_next(bio_iter
, bio_seg
);
1720 * read (part of) a message.
1722 static int read_partial_message(struct ceph_connection
*con
)
1724 struct ceph_msg
*m
= con
->in_msg
;
1728 unsigned front_len
, middle_len
, data_len
;
1729 bool do_datacrc
= !con
->msgr
->nocrc
;
1734 dout("read_partial_message con %p msg %p\n", con
, m
);
1737 size
= sizeof (con
->in_hdr
);
1739 ret
= read_partial(con
, end
, size
, &con
->in_hdr
);
1743 crc
= crc32c(0, &con
->in_hdr
, offsetof(struct ceph_msg_header
, crc
));
1744 if (cpu_to_le32(crc
) != con
->in_hdr
.crc
) {
1745 pr_err("read_partial_message bad hdr "
1746 " crc %u != expected %u\n",
1747 crc
, con
->in_hdr
.crc
);
1751 front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
1752 if (front_len
> CEPH_MSG_MAX_FRONT_LEN
)
1754 middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
1755 if (middle_len
> CEPH_MSG_MAX_DATA_LEN
)
1757 data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
1758 if (data_len
> CEPH_MSG_MAX_DATA_LEN
)
1762 seq
= le64_to_cpu(con
->in_hdr
.seq
);
1763 if ((s64
)seq
- (s64
)con
->in_seq
< 1) {
1764 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1765 ENTITY_NAME(con
->peer_name
),
1766 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1767 seq
, con
->in_seq
+ 1);
1768 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1770 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1772 } else if ((s64
)seq
- (s64
)con
->in_seq
> 1) {
1773 pr_err("read_partial_message bad seq %lld expected %lld\n",
1774 seq
, con
->in_seq
+ 1);
1775 con
->error_msg
= "bad message sequence # for incoming message";
1779 /* allocate message? */
1781 dout("got hdr type %d front %d data %d\n", con
->in_hdr
.type
,
1782 con
->in_hdr
.front_len
, con
->in_hdr
.data_len
);
1784 con
->in_msg
= ceph_alloc_msg(con
, &con
->in_hdr
, &skip
);
1786 /* skip this message */
1787 dout("alloc_msg said skip message\n");
1788 BUG_ON(con
->in_msg
);
1789 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1791 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1797 "error allocating memory for incoming message";
1801 m
->front
.iov_len
= 0; /* haven't read it yet */
1803 m
->middle
->vec
.iov_len
= 0;
1805 con
->in_msg_pos
.page
= 0;
1807 con
->in_msg_pos
.page_pos
= m
->page_alignment
;
1809 con
->in_msg_pos
.page_pos
= 0;
1810 con
->in_msg_pos
.data_pos
= 0;
1814 ret
= read_partial_message_section(con
, &m
->front
, front_len
,
1815 &con
->in_front_crc
);
1821 ret
= read_partial_message_section(con
, &m
->middle
->vec
,
1823 &con
->in_middle_crc
);
1828 if (m
->bio
&& !m
->bio_iter
)
1829 init_bio_iter(m
->bio
, &m
->bio_iter
, &m
->bio_seg
);
1833 while (con
->in_msg_pos
.data_pos
< data_len
) {
1835 ret
= read_partial_message_pages(con
, m
->pages
,
1836 data_len
, do_datacrc
);
1840 } else if (m
->bio
) {
1842 ret
= read_partial_message_bio(con
,
1843 &m
->bio_iter
, &m
->bio_seg
,
1844 data_len
, do_datacrc
);
1854 size
= sizeof (m
->footer
);
1856 ret
= read_partial(con
, end
, size
, &m
->footer
);
1860 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1861 m
, front_len
, m
->footer
.front_crc
, middle_len
,
1862 m
->footer
.middle_crc
, data_len
, m
->footer
.data_crc
);
1865 if (con
->in_front_crc
!= le32_to_cpu(m
->footer
.front_crc
)) {
1866 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1867 m
, con
->in_front_crc
, m
->footer
.front_crc
);
1870 if (con
->in_middle_crc
!= le32_to_cpu(m
->footer
.middle_crc
)) {
1871 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1872 m
, con
->in_middle_crc
, m
->footer
.middle_crc
);
1876 (m
->footer
.flags
& CEPH_MSG_FOOTER_NOCRC
) == 0 &&
1877 con
->in_data_crc
!= le32_to_cpu(m
->footer
.data_crc
)) {
1878 pr_err("read_partial_message %p data crc %u != exp. %u\n", m
,
1879 con
->in_data_crc
, le32_to_cpu(m
->footer
.data_crc
));
1883 return 1; /* done! */
1887 * Process message. This happens in the worker thread. The callback should
1888 * be careful not to do anything that waits on other incoming messages or it
1891 static void process_message(struct ceph_connection
*con
)
1893 struct ceph_msg
*msg
;
1898 /* if first message, set peer_name */
1899 if (con
->peer_name
.type
== 0)
1900 con
->peer_name
= msg
->hdr
.src
;
1903 mutex_unlock(&con
->mutex
);
1905 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
1906 msg
, le64_to_cpu(msg
->hdr
.seq
),
1907 ENTITY_NAME(msg
->hdr
.src
),
1908 le16_to_cpu(msg
->hdr
.type
),
1909 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
1910 le32_to_cpu(msg
->hdr
.front_len
),
1911 le32_to_cpu(msg
->hdr
.data_len
),
1912 con
->in_front_crc
, con
->in_middle_crc
, con
->in_data_crc
);
1913 con
->ops
->dispatch(con
, msg
);
1915 mutex_lock(&con
->mutex
);
1916 prepare_read_tag(con
);
1921 * Write something to the socket. Called in a worker thread when the
1922 * socket appears to be writeable and we have something ready to send.
1924 static int try_write(struct ceph_connection
*con
)
1928 dout("try_write start %p state %lu nref %d\n", con
, con
->state
,
1929 atomic_read(&con
->nref
));
1932 dout("try_write out_kvec_bytes %d\n", con
->out_kvec_bytes
);
1934 /* open the socket first? */
1935 if (con
->sock
== NULL
) {
1936 con_out_kvec_reset(con
);
1937 prepare_write_banner(con
);
1938 ret
= prepare_write_connect(con
);
1941 prepare_read_banner(con
);
1942 set_bit(CONNECTING
, &con
->state
);
1943 clear_bit(NEGOTIATING
, &con
->state
);
1945 BUG_ON(con
->in_msg
);
1946 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1947 dout("try_write initiating connect on %p new state %lu\n",
1949 ret
= ceph_tcp_connect(con
);
1951 con
->error_msg
= "connect error";
1957 /* kvec data queued? */
1958 if (con
->out_skip
) {
1959 ret
= write_partial_skip(con
);
1963 if (con
->out_kvec_left
) {
1964 ret
= write_partial_kvec(con
);
1971 if (con
->out_msg_done
) {
1972 ceph_msg_put(con
->out_msg
);
1973 con
->out_msg
= NULL
; /* we're done with this one */
1977 ret
= write_partial_msg_pages(con
);
1979 goto more_kvec
; /* we need to send the footer, too! */
1983 dout("try_write write_partial_msg_pages err %d\n",
1990 if (!test_bit(CONNECTING
, &con
->state
)) {
1991 /* is anything else pending? */
1992 if (!list_empty(&con
->out_queue
)) {
1993 prepare_write_message(con
);
1996 if (con
->in_seq
> con
->in_seq_acked
) {
1997 prepare_write_ack(con
);
2000 if (test_and_clear_bit(KEEPALIVE_PENDING
, &con
->flags
)) {
2001 prepare_write_keepalive(con
);
2006 /* Nothing to do! */
2007 clear_bit(WRITE_PENDING
, &con
->flags
);
2008 dout("try_write nothing else to write.\n");
2011 dout("try_write done on %p ret %d\n", con
, ret
);
2018 * Read what we can from the socket.
2020 static int try_read(struct ceph_connection
*con
)
2027 if (test_bit(STANDBY
, &con
->state
))
2030 dout("try_read start on %p\n", con
);
2033 dout("try_read tag %d in_base_pos %d\n", (int)con
->in_tag
,
2037 * process_connect and process_message drop and re-take
2038 * con->mutex. make sure we handle a racing close or reopen.
2040 if (test_bit(CLOSED
, &con
->state
) ||
2041 test_bit(OPENING
, &con
->state
)) {
2046 if (test_bit(CONNECTING
, &con
->state
)) {
2047 if (!test_bit(NEGOTIATING
, &con
->state
)) {
2048 dout("try_read connecting\n");
2049 ret
= read_partial_banner(con
);
2052 ret
= process_banner(con
);
2056 ret
= read_partial_connect(con
);
2059 ret
= process_connect(con
);
2065 if (con
->in_base_pos
< 0) {
2067 * skipping + discarding content.
2069 * FIXME: there must be a better way to do this!
2071 static char buf
[SKIP_BUF_SIZE
];
2072 int skip
= min((int) sizeof (buf
), -con
->in_base_pos
);
2074 dout("skipping %d / %d bytes\n", skip
, -con
->in_base_pos
);
2075 ret
= ceph_tcp_recvmsg(con
->sock
, buf
, skip
);
2078 con
->in_base_pos
+= ret
;
2079 if (con
->in_base_pos
)
2082 if (con
->in_tag
== CEPH_MSGR_TAG_READY
) {
2086 ret
= ceph_tcp_recvmsg(con
->sock
, &con
->in_tag
, 1);
2089 dout("try_read got tag %d\n", (int)con
->in_tag
);
2090 switch (con
->in_tag
) {
2091 case CEPH_MSGR_TAG_MSG
:
2092 prepare_read_message(con
);
2094 case CEPH_MSGR_TAG_ACK
:
2095 prepare_read_ack(con
);
2097 case CEPH_MSGR_TAG_CLOSE
:
2098 set_bit(CLOSED
, &con
->state
); /* fixme */
2104 if (con
->in_tag
== CEPH_MSGR_TAG_MSG
) {
2105 ret
= read_partial_message(con
);
2109 con
->error_msg
= "bad crc";
2113 con
->error_msg
= "io error";
2118 if (con
->in_tag
== CEPH_MSGR_TAG_READY
)
2120 process_message(con
);
2123 if (con
->in_tag
== CEPH_MSGR_TAG_ACK
) {
2124 ret
= read_partial_ack(con
);
2132 dout("try_read done on %p ret %d\n", con
, ret
);
2136 pr_err("try_read bad con->in_tag = %d\n", (int)con
->in_tag
);
2137 con
->error_msg
= "protocol error, garbage tag";
2144 * Atomically queue work on a connection. Bump @con reference to
2145 * avoid races with connection teardown.
2147 static void queue_con(struct ceph_connection
*con
)
2149 if (!con
->ops
->get(con
)) {
2150 dout("queue_con %p ref count 0\n", con
);
2154 if (!queue_delayed_work(ceph_msgr_wq
, &con
->work
, 0)) {
2155 dout("queue_con %p - already queued\n", con
);
2158 dout("queue_con %p\n", con
);
2163 * Do some work on a connection. Drop a connection ref when we're done.
2165 static void con_work(struct work_struct
*work
)
2167 struct ceph_connection
*con
= container_of(work
, struct ceph_connection
,
2171 mutex_lock(&con
->mutex
);
2173 if (test_and_clear_bit(BACKOFF
, &con
->flags
)) {
2174 dout("con_work %p backing off\n", con
);
2175 if (queue_delayed_work(ceph_msgr_wq
, &con
->work
,
2176 round_jiffies_relative(con
->delay
))) {
2177 dout("con_work %p backoff %lu\n", con
, con
->delay
);
2178 mutex_unlock(&con
->mutex
);
2182 dout("con_work %p FAILED to back off %lu\n", con
,
2187 if (test_bit(STANDBY
, &con
->state
)) {
2188 dout("con_work %p STANDBY\n", con
);
2191 if (test_bit(CLOSED
, &con
->state
)) { /* e.g. if we are replaced */
2192 dout("con_work CLOSED\n");
2193 con_close_socket(con
);
2196 if (test_and_clear_bit(OPENING
, &con
->state
)) {
2197 /* reopen w/ new peer */
2198 dout("con_work OPENING\n");
2199 con_close_socket(con
);
2202 if (test_and_clear_bit(SOCK_CLOSED
, &con
->flags
))
2205 ret
= try_read(con
);
2211 ret
= try_write(con
);
2218 mutex_unlock(&con
->mutex
);
2224 mutex_unlock(&con
->mutex
);
2225 ceph_fault(con
); /* error/fault path */
2231 * Generic error/fault handler. A retry mechanism is used with
2232 * exponential backoff
2234 static void ceph_fault(struct ceph_connection
*con
)
2236 pr_err("%s%lld %s %s\n", ENTITY_NAME(con
->peer_name
),
2237 ceph_pr_addr(&con
->peer_addr
.in_addr
), con
->error_msg
);
2238 dout("fault %p state %lu to peer %s\n",
2239 con
, con
->state
, ceph_pr_addr(&con
->peer_addr
.in_addr
));
2241 if (test_bit(LOSSYTX
, &con
->flags
)) {
2242 dout("fault on LOSSYTX channel\n");
2246 mutex_lock(&con
->mutex
);
2247 if (test_bit(CLOSED
, &con
->state
))
2250 con_close_socket(con
);
2253 ceph_msg_put(con
->in_msg
);
2257 /* Requeue anything that hasn't been acked */
2258 list_splice_init(&con
->out_sent
, &con
->out_queue
);
2260 /* If there are no messages queued or keepalive pending, place
2261 * the connection in a STANDBY state */
2262 if (list_empty(&con
->out_queue
) &&
2263 !test_bit(KEEPALIVE_PENDING
, &con
->flags
)) {
2264 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con
);
2265 clear_bit(WRITE_PENDING
, &con
->flags
);
2266 set_bit(STANDBY
, &con
->state
);
2268 /* retry after a delay. */
2269 if (con
->delay
== 0)
2270 con
->delay
= BASE_DELAY_INTERVAL
;
2271 else if (con
->delay
< MAX_DELAY_INTERVAL
)
2274 if (queue_delayed_work(ceph_msgr_wq
, &con
->work
,
2275 round_jiffies_relative(con
->delay
))) {
2276 dout("fault queued %p delay %lu\n", con
, con
->delay
);
2279 dout("fault failed to queue %p delay %lu, backoff\n",
2282 * In many cases we see a socket state change
2283 * while con_work is running and end up
2284 * queuing (non-delayed) work, such that we
2285 * can't backoff with a delay. Set a flag so
2286 * that when con_work restarts we schedule the
2289 set_bit(BACKOFF
, &con
->flags
);
2294 mutex_unlock(&con
->mutex
);
2297 * in case we faulted due to authentication, invalidate our
2298 * current tickets so that we can get new ones.
2300 if (con
->auth_retry
&& con
->ops
->invalidate_authorizer
) {
2301 dout("calling invalidate_authorizer()\n");
2302 con
->ops
->invalidate_authorizer(con
);
2305 if (con
->ops
->fault
)
2306 con
->ops
->fault(con
);
2312 * initialize a new messenger instance
2314 void ceph_messenger_init(struct ceph_messenger
*msgr
,
2315 struct ceph_entity_addr
*myaddr
,
2316 u32 supported_features
,
2317 u32 required_features
,
2320 msgr
->supported_features
= supported_features
;
2321 msgr
->required_features
= required_features
;
2323 spin_lock_init(&msgr
->global_seq_lock
);
2326 msgr
->inst
.addr
= *myaddr
;
2328 /* select a random nonce */
2329 msgr
->inst
.addr
.type
= 0;
2330 get_random_bytes(&msgr
->inst
.addr
.nonce
, sizeof(msgr
->inst
.addr
.nonce
));
2331 encode_my_addr(msgr
);
2332 msgr
->nocrc
= nocrc
;
2334 dout("%s %p\n", __func__
, msgr
);
2336 EXPORT_SYMBOL(ceph_messenger_init
);
2338 static void clear_standby(struct ceph_connection
*con
)
2340 /* come back from STANDBY? */
2341 if (test_and_clear_bit(STANDBY
, &con
->state
)) {
2342 mutex_lock(&con
->mutex
);
2343 dout("clear_standby %p and ++connect_seq\n", con
);
2345 WARN_ON(test_bit(WRITE_PENDING
, &con
->flags
));
2346 WARN_ON(test_bit(KEEPALIVE_PENDING
, &con
->flags
));
2347 mutex_unlock(&con
->mutex
);
2352 * Queue up an outgoing message on the given connection.
2354 void ceph_con_send(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2356 if (test_bit(CLOSED
, &con
->state
)) {
2357 dout("con_send %p closed, dropping %p\n", con
, msg
);
2363 msg
->hdr
.src
= con
->msgr
->inst
.name
;
2365 BUG_ON(msg
->front
.iov_len
!= le32_to_cpu(msg
->hdr
.front_len
));
2367 msg
->needs_out_seq
= true;
2370 mutex_lock(&con
->mutex
);
2371 BUG_ON(!list_empty(&msg
->list_head
));
2372 list_add_tail(&msg
->list_head
, &con
->out_queue
);
2373 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg
,
2374 ENTITY_NAME(con
->peer_name
), le16_to_cpu(msg
->hdr
.type
),
2375 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2376 le32_to_cpu(msg
->hdr
.front_len
),
2377 le32_to_cpu(msg
->hdr
.middle_len
),
2378 le32_to_cpu(msg
->hdr
.data_len
));
2379 mutex_unlock(&con
->mutex
);
2381 /* if there wasn't anything waiting to send before, queue
2384 if (test_and_set_bit(WRITE_PENDING
, &con
->flags
) == 0)
2387 EXPORT_SYMBOL(ceph_con_send
);
2390 * Revoke a message that was previously queued for send
2392 void ceph_con_revoke(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2394 mutex_lock(&con
->mutex
);
2395 if (!list_empty(&msg
->list_head
)) {
2396 dout("con_revoke %p msg %p - was on queue\n", con
, msg
);
2397 list_del_init(&msg
->list_head
);
2401 if (con
->out_msg
== msg
) {
2402 dout("con_revoke %p msg %p - was sending\n", con
, msg
);
2403 con
->out_msg
= NULL
;
2404 if (con
->out_kvec_is_msg
) {
2405 con
->out_skip
= con
->out_kvec_bytes
;
2406 con
->out_kvec_is_msg
= false;
2411 mutex_unlock(&con
->mutex
);
2415 * Revoke a message that we may be reading data into
2417 void ceph_con_revoke_message(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2419 mutex_lock(&con
->mutex
);
2420 if (con
->in_msg
&& con
->in_msg
== msg
) {
2421 unsigned front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
2422 unsigned middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
2423 unsigned data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
2425 /* skip rest of message */
2426 dout("con_revoke_pages %p msg %p revoked\n", con
, msg
);
2427 con
->in_base_pos
= con
->in_base_pos
-
2428 sizeof(struct ceph_msg_header
) -
2432 sizeof(struct ceph_msg_footer
);
2433 ceph_msg_put(con
->in_msg
);
2435 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2438 dout("con_revoke_pages %p msg %p pages %p no-op\n",
2439 con
, con
->in_msg
, msg
);
2441 mutex_unlock(&con
->mutex
);
2445 * Queue a keepalive byte to ensure the tcp connection is alive.
2447 void ceph_con_keepalive(struct ceph_connection
*con
)
2449 dout("con_keepalive %p\n", con
);
2451 if (test_and_set_bit(KEEPALIVE_PENDING
, &con
->flags
) == 0 &&
2452 test_and_set_bit(WRITE_PENDING
, &con
->flags
) == 0)
2455 EXPORT_SYMBOL(ceph_con_keepalive
);
2459 * construct a new message with given type, size
2460 * the new msg has a ref count of 1.
2462 struct ceph_msg
*ceph_msg_new(int type
, int front_len
, gfp_t flags
,
2467 m
= kmalloc(sizeof(*m
), flags
);
2470 kref_init(&m
->kref
);
2471 INIT_LIST_HEAD(&m
->list_head
);
2474 m
->hdr
.type
= cpu_to_le16(type
);
2475 m
->hdr
.priority
= cpu_to_le16(CEPH_MSG_PRIO_DEFAULT
);
2477 m
->hdr
.front_len
= cpu_to_le32(front_len
);
2478 m
->hdr
.middle_len
= 0;
2479 m
->hdr
.data_len
= 0;
2480 m
->hdr
.data_off
= 0;
2481 m
->hdr
.reserved
= 0;
2482 m
->footer
.front_crc
= 0;
2483 m
->footer
.middle_crc
= 0;
2484 m
->footer
.data_crc
= 0;
2485 m
->footer
.flags
= 0;
2486 m
->front_max
= front_len
;
2487 m
->front_is_vmalloc
= false;
2488 m
->more_to_follow
= false;
2497 m
->page_alignment
= 0;
2507 if (front_len
> PAGE_CACHE_SIZE
) {
2508 m
->front
.iov_base
= __vmalloc(front_len
, flags
,
2510 m
->front_is_vmalloc
= true;
2512 m
->front
.iov_base
= kmalloc(front_len
, flags
);
2514 if (m
->front
.iov_base
== NULL
) {
2515 dout("ceph_msg_new can't allocate %d bytes\n",
2520 m
->front
.iov_base
= NULL
;
2522 m
->front
.iov_len
= front_len
;
2524 dout("ceph_msg_new %p front %d\n", m
, front_len
);
2531 pr_err("msg_new can't create type %d front %d\n", type
,
2535 dout("msg_new can't create type %d front %d\n", type
,
2540 EXPORT_SYMBOL(ceph_msg_new
);
2543 * Allocate "middle" portion of a message, if it is needed and wasn't
2544 * allocated by alloc_msg. This allows us to read a small fixed-size
2545 * per-type header in the front and then gracefully fail (i.e.,
2546 * propagate the error to the caller based on info in the front) when
2547 * the middle is too large.
2549 static int ceph_alloc_middle(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2551 int type
= le16_to_cpu(msg
->hdr
.type
);
2552 int middle_len
= le32_to_cpu(msg
->hdr
.middle_len
);
2554 dout("alloc_middle %p type %d %s middle_len %d\n", msg
, type
,
2555 ceph_msg_type_name(type
), middle_len
);
2556 BUG_ON(!middle_len
);
2557 BUG_ON(msg
->middle
);
2559 msg
->middle
= ceph_buffer_new(middle_len
, GFP_NOFS
);
2566 * Generic message allocator, for incoming messages.
2568 static struct ceph_msg
*ceph_alloc_msg(struct ceph_connection
*con
,
2569 struct ceph_msg_header
*hdr
,
2572 int type
= le16_to_cpu(hdr
->type
);
2573 int front_len
= le32_to_cpu(hdr
->front_len
);
2574 int middle_len
= le32_to_cpu(hdr
->middle_len
);
2575 struct ceph_msg
*msg
= NULL
;
2578 if (con
->ops
->alloc_msg
) {
2579 mutex_unlock(&con
->mutex
);
2580 msg
= con
->ops
->alloc_msg(con
, hdr
, skip
);
2581 mutex_lock(&con
->mutex
);
2587 msg
= ceph_msg_new(type
, front_len
, GFP_NOFS
, false);
2589 pr_err("unable to allocate msg type %d len %d\n",
2593 msg
->page_alignment
= le16_to_cpu(hdr
->data_off
);
2595 memcpy(&msg
->hdr
, &con
->in_hdr
, sizeof(con
->in_hdr
));
2597 if (middle_len
&& !msg
->middle
) {
2598 ret
= ceph_alloc_middle(con
, msg
);
2610 * Free a generically kmalloc'd message.
2612 void ceph_msg_kfree(struct ceph_msg
*m
)
2614 dout("msg_kfree %p\n", m
);
2615 if (m
->front_is_vmalloc
)
2616 vfree(m
->front
.iov_base
);
2618 kfree(m
->front
.iov_base
);
2623 * Drop a msg ref. Destroy as needed.
2625 void ceph_msg_last_put(struct kref
*kref
)
2627 struct ceph_msg
*m
= container_of(kref
, struct ceph_msg
, kref
);
2629 dout("ceph_msg_put last one on %p\n", m
);
2630 WARN_ON(!list_empty(&m
->list_head
));
2632 /* drop middle, data, if any */
2634 ceph_buffer_put(m
->middle
);
2641 ceph_pagelist_release(m
->pagelist
);
2649 ceph_msgpool_put(m
->pool
, m
);
2653 EXPORT_SYMBOL(ceph_msg_last_put
);
2655 void ceph_msg_dump(struct ceph_msg
*msg
)
2657 pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg
,
2658 msg
->front_max
, msg
->nr_pages
);
2659 print_hex_dump(KERN_DEBUG
, "header: ",
2660 DUMP_PREFIX_OFFSET
, 16, 1,
2661 &msg
->hdr
, sizeof(msg
->hdr
), true);
2662 print_hex_dump(KERN_DEBUG
, " front: ",
2663 DUMP_PREFIX_OFFSET
, 16, 1,
2664 msg
->front
.iov_base
, msg
->front
.iov_len
, true);
2666 print_hex_dump(KERN_DEBUG
, "middle: ",
2667 DUMP_PREFIX_OFFSET
, 16, 1,
2668 msg
->middle
->vec
.iov_base
,
2669 msg
->middle
->vec
.iov_len
, true);
2670 print_hex_dump(KERN_DEBUG
, "footer: ",
2671 DUMP_PREFIX_OFFSET
, 16, 1,
2672 &msg
->footer
, sizeof(msg
->footer
), true);
2674 EXPORT_SYMBOL(ceph_msg_dump
);