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 set_bit(SOCK_CLOSED
, &con
->flags
);
396 rc
= con
->sock
->ops
->shutdown(con
->sock
, SHUT_RDWR
);
397 sock_release(con
->sock
);
399 clear_bit(SOCK_CLOSED
, &con
->flags
);
400 con_sock_state_closed(con
);
405 * Reset a connection. Discard all incoming and outgoing messages
406 * and clear *_seq state.
408 static void ceph_msg_remove(struct ceph_msg
*msg
)
410 list_del_init(&msg
->list_head
);
411 BUG_ON(msg
->con
== NULL
);
412 msg
->con
->ops
->put(msg
->con
);
417 static void ceph_msg_remove_list(struct list_head
*head
)
419 while (!list_empty(head
)) {
420 struct ceph_msg
*msg
= list_first_entry(head
, struct ceph_msg
,
422 ceph_msg_remove(msg
);
426 static void reset_connection(struct ceph_connection
*con
)
428 /* reset connection, out_queue, msg_ and connect_seq */
429 /* discard existing out_queue and msg_seq */
430 ceph_msg_remove_list(&con
->out_queue
);
431 ceph_msg_remove_list(&con
->out_sent
);
434 BUG_ON(con
->in_msg
->con
!= con
);
435 con
->in_msg
->con
= NULL
;
436 ceph_msg_put(con
->in_msg
);
441 con
->connect_seq
= 0;
444 ceph_msg_put(con
->out_msg
);
448 con
->in_seq_acked
= 0;
452 * mark a peer down. drop any open connections.
454 void ceph_con_close(struct ceph_connection
*con
)
456 dout("con_close %p peer %s\n", con
,
457 ceph_pr_addr(&con
->peer_addr
.in_addr
));
458 clear_bit(NEGOTIATING
, &con
->state
);
459 clear_bit(STANDBY
, &con
->state
); /* avoid connect_seq bump */
460 set_bit(CLOSED
, &con
->state
);
462 clear_bit(LOSSYTX
, &con
->flags
); /* so we retry next connect */
463 clear_bit(KEEPALIVE_PENDING
, &con
->flags
);
464 clear_bit(WRITE_PENDING
, &con
->flags
);
466 mutex_lock(&con
->mutex
);
467 reset_connection(con
);
468 con
->peer_global_seq
= 0;
469 cancel_delayed_work(&con
->work
);
470 mutex_unlock(&con
->mutex
);
473 EXPORT_SYMBOL(ceph_con_close
);
476 * Reopen a closed connection, with a new peer address.
478 void ceph_con_open(struct ceph_connection
*con
, struct ceph_entity_addr
*addr
)
480 dout("con_open %p %s\n", con
, ceph_pr_addr(&addr
->in_addr
));
481 set_bit(OPENING
, &con
->state
);
482 WARN_ON(!test_and_clear_bit(CLOSED
, &con
->state
));
484 memcpy(&con
->peer_addr
, addr
, sizeof(*addr
));
485 con
->delay
= 0; /* reset backoff memory */
488 EXPORT_SYMBOL(ceph_con_open
);
491 * return true if this connection ever successfully opened
493 bool ceph_con_opened(struct ceph_connection
*con
)
495 return con
->connect_seq
> 0;
499 * initialize a new connection.
501 void ceph_con_init(struct ceph_connection
*con
, void *private,
502 const struct ceph_connection_operations
*ops
,
503 struct ceph_messenger
*msgr
, __u8 entity_type
, __u64 entity_num
)
505 dout("con_init %p\n", con
);
506 memset(con
, 0, sizeof(*con
));
507 con
->private = private;
511 con_sock_state_init(con
);
513 con
->peer_name
.type
= (__u8
) entity_type
;
514 con
->peer_name
.num
= cpu_to_le64(entity_num
);
516 mutex_init(&con
->mutex
);
517 INIT_LIST_HEAD(&con
->out_queue
);
518 INIT_LIST_HEAD(&con
->out_sent
);
519 INIT_DELAYED_WORK(&con
->work
, con_work
);
521 set_bit(CLOSED
, &con
->state
);
523 EXPORT_SYMBOL(ceph_con_init
);
527 * We maintain a global counter to order connection attempts. Get
528 * a unique seq greater than @gt.
530 static u32
get_global_seq(struct ceph_messenger
*msgr
, u32 gt
)
534 spin_lock(&msgr
->global_seq_lock
);
535 if (msgr
->global_seq
< gt
)
536 msgr
->global_seq
= gt
;
537 ret
= ++msgr
->global_seq
;
538 spin_unlock(&msgr
->global_seq_lock
);
542 static void con_out_kvec_reset(struct ceph_connection
*con
)
544 con
->out_kvec_left
= 0;
545 con
->out_kvec_bytes
= 0;
546 con
->out_kvec_cur
= &con
->out_kvec
[0];
549 static void con_out_kvec_add(struct ceph_connection
*con
,
550 size_t size
, void *data
)
554 index
= con
->out_kvec_left
;
555 BUG_ON(index
>= ARRAY_SIZE(con
->out_kvec
));
557 con
->out_kvec
[index
].iov_len
= size
;
558 con
->out_kvec
[index
].iov_base
= data
;
559 con
->out_kvec_left
++;
560 con
->out_kvec_bytes
+= size
;
564 static void init_bio_iter(struct bio
*bio
, struct bio
**iter
, int *seg
)
575 static void iter_bio_next(struct bio
**bio_iter
, int *seg
)
577 if (*bio_iter
== NULL
)
580 BUG_ON(*seg
>= (*bio_iter
)->bi_vcnt
);
583 if (*seg
== (*bio_iter
)->bi_vcnt
)
584 init_bio_iter((*bio_iter
)->bi_next
, bio_iter
, seg
);
588 static void prepare_write_message_data(struct ceph_connection
*con
)
590 struct ceph_msg
*msg
= con
->out_msg
;
593 BUG_ON(!msg
->hdr
.data_len
);
595 /* initialize page iterator */
596 con
->out_msg_pos
.page
= 0;
598 con
->out_msg_pos
.page_pos
= msg
->page_alignment
;
600 con
->out_msg_pos
.page_pos
= 0;
603 init_bio_iter(msg
->bio
, &msg
->bio_iter
, &msg
->bio_seg
);
605 con
->out_msg_pos
.data_pos
= 0;
606 con
->out_msg_pos
.did_page_crc
= false;
607 con
->out_more
= 1; /* data + footer will follow */
611 * Prepare footer for currently outgoing message, and finish things
612 * off. Assumes out_kvec* are already valid.. we just add on to the end.
614 static void prepare_write_message_footer(struct ceph_connection
*con
)
616 struct ceph_msg
*m
= con
->out_msg
;
617 int v
= con
->out_kvec_left
;
619 m
->footer
.flags
|= CEPH_MSG_FOOTER_COMPLETE
;
621 dout("prepare_write_message_footer %p\n", con
);
622 con
->out_kvec_is_msg
= true;
623 con
->out_kvec
[v
].iov_base
= &m
->footer
;
624 con
->out_kvec
[v
].iov_len
= sizeof(m
->footer
);
625 con
->out_kvec_bytes
+= sizeof(m
->footer
);
626 con
->out_kvec_left
++;
627 con
->out_more
= m
->more_to_follow
;
628 con
->out_msg_done
= true;
632 * Prepare headers for the next outgoing message.
634 static void prepare_write_message(struct ceph_connection
*con
)
639 con_out_kvec_reset(con
);
640 con
->out_kvec_is_msg
= true;
641 con
->out_msg_done
= false;
643 /* Sneak an ack in there first? If we can get it into the same
644 * TCP packet that's a good thing. */
645 if (con
->in_seq
> con
->in_seq_acked
) {
646 con
->in_seq_acked
= con
->in_seq
;
647 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
648 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
649 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
653 BUG_ON(list_empty(&con
->out_queue
));
654 m
= list_first_entry(&con
->out_queue
, struct ceph_msg
, list_head
);
656 BUG_ON(m
->con
!= con
);
658 /* put message on sent list */
660 list_move_tail(&m
->list_head
, &con
->out_sent
);
663 * only assign outgoing seq # if we haven't sent this message
664 * yet. if it is requeued, resend with it's original seq.
666 if (m
->needs_out_seq
) {
667 m
->hdr
.seq
= cpu_to_le64(++con
->out_seq
);
668 m
->needs_out_seq
= false;
671 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n",
672 m
, con
->out_seq
, le16_to_cpu(m
->hdr
.type
),
673 le32_to_cpu(m
->hdr
.front_len
), le32_to_cpu(m
->hdr
.middle_len
),
674 le32_to_cpu(m
->hdr
.data_len
),
676 BUG_ON(le32_to_cpu(m
->hdr
.front_len
) != m
->front
.iov_len
);
678 /* tag + hdr + front + middle */
679 con_out_kvec_add(con
, sizeof (tag_msg
), &tag_msg
);
680 con_out_kvec_add(con
, sizeof (m
->hdr
), &m
->hdr
);
681 con_out_kvec_add(con
, m
->front
.iov_len
, m
->front
.iov_base
);
684 con_out_kvec_add(con
, m
->middle
->vec
.iov_len
,
685 m
->middle
->vec
.iov_base
);
687 /* fill in crc (except data pages), footer */
688 crc
= crc32c(0, &m
->hdr
, offsetof(struct ceph_msg_header
, crc
));
689 con
->out_msg
->hdr
.crc
= cpu_to_le32(crc
);
690 con
->out_msg
->footer
.flags
= 0;
692 crc
= crc32c(0, m
->front
.iov_base
, m
->front
.iov_len
);
693 con
->out_msg
->footer
.front_crc
= cpu_to_le32(crc
);
695 crc
= crc32c(0, m
->middle
->vec
.iov_base
,
696 m
->middle
->vec
.iov_len
);
697 con
->out_msg
->footer
.middle_crc
= cpu_to_le32(crc
);
699 con
->out_msg
->footer
.middle_crc
= 0;
700 dout("%s front_crc %u middle_crc %u\n", __func__
,
701 le32_to_cpu(con
->out_msg
->footer
.front_crc
),
702 le32_to_cpu(con
->out_msg
->footer
.middle_crc
));
704 /* is there a data payload? */
705 con
->out_msg
->footer
.data_crc
= 0;
707 prepare_write_message_data(con
);
709 /* no, queue up footer too and be done */
710 prepare_write_message_footer(con
);
712 set_bit(WRITE_PENDING
, &con
->flags
);
718 static void prepare_write_ack(struct ceph_connection
*con
)
720 dout("prepare_write_ack %p %llu -> %llu\n", con
,
721 con
->in_seq_acked
, con
->in_seq
);
722 con
->in_seq_acked
= con
->in_seq
;
724 con_out_kvec_reset(con
);
726 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
728 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
729 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
732 con
->out_more
= 1; /* more will follow.. eventually.. */
733 set_bit(WRITE_PENDING
, &con
->flags
);
737 * Prepare to write keepalive byte.
739 static void prepare_write_keepalive(struct ceph_connection
*con
)
741 dout("prepare_write_keepalive %p\n", con
);
742 con_out_kvec_reset(con
);
743 con_out_kvec_add(con
, sizeof (tag_keepalive
), &tag_keepalive
);
744 set_bit(WRITE_PENDING
, &con
->flags
);
748 * Connection negotiation.
751 static struct ceph_auth_handshake
*get_connect_authorizer(struct ceph_connection
*con
,
754 struct ceph_auth_handshake
*auth
;
756 if (!con
->ops
->get_authorizer
) {
757 con
->out_connect
.authorizer_protocol
= CEPH_AUTH_UNKNOWN
;
758 con
->out_connect
.authorizer_len
= 0;
763 /* Can't hold the mutex while getting authorizer */
765 mutex_unlock(&con
->mutex
);
767 auth
= con
->ops
->get_authorizer(con
, auth_proto
, con
->auth_retry
);
769 mutex_lock(&con
->mutex
);
773 if (test_bit(CLOSED
, &con
->state
) || test_bit(OPENING
, &con
->flags
))
774 return ERR_PTR(-EAGAIN
);
776 con
->auth_reply_buf
= auth
->authorizer_reply_buf
;
777 con
->auth_reply_buf_len
= auth
->authorizer_reply_buf_len
;
784 * We connected to a peer and are saying hello.
786 static void prepare_write_banner(struct ceph_connection
*con
)
788 con_out_kvec_add(con
, strlen(CEPH_BANNER
), CEPH_BANNER
);
789 con_out_kvec_add(con
, sizeof (con
->msgr
->my_enc_addr
),
790 &con
->msgr
->my_enc_addr
);
793 set_bit(WRITE_PENDING
, &con
->flags
);
796 static int prepare_write_connect(struct ceph_connection
*con
)
798 unsigned int global_seq
= get_global_seq(con
->msgr
, 0);
801 struct ceph_auth_handshake
*auth
;
803 switch (con
->peer_name
.type
) {
804 case CEPH_ENTITY_TYPE_MON
:
805 proto
= CEPH_MONC_PROTOCOL
;
807 case CEPH_ENTITY_TYPE_OSD
:
808 proto
= CEPH_OSDC_PROTOCOL
;
810 case CEPH_ENTITY_TYPE_MDS
:
811 proto
= CEPH_MDSC_PROTOCOL
;
817 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con
,
818 con
->connect_seq
, global_seq
, proto
);
820 con
->out_connect
.features
= cpu_to_le64(con
->msgr
->supported_features
);
821 con
->out_connect
.host_type
= cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT
);
822 con
->out_connect
.connect_seq
= cpu_to_le32(con
->connect_seq
);
823 con
->out_connect
.global_seq
= cpu_to_le32(global_seq
);
824 con
->out_connect
.protocol_version
= cpu_to_le32(proto
);
825 con
->out_connect
.flags
= 0;
827 auth_proto
= CEPH_AUTH_UNKNOWN
;
828 auth
= get_connect_authorizer(con
, &auth_proto
);
830 return PTR_ERR(auth
);
832 con
->out_connect
.authorizer_protocol
= cpu_to_le32(auth_proto
);
833 con
->out_connect
.authorizer_len
= auth
?
834 cpu_to_le32(auth
->authorizer_buf_len
) : 0;
836 con_out_kvec_add(con
, sizeof (con
->out_connect
),
838 if (auth
&& auth
->authorizer_buf_len
)
839 con_out_kvec_add(con
, auth
->authorizer_buf_len
,
840 auth
->authorizer_buf
);
843 set_bit(WRITE_PENDING
, &con
->flags
);
849 * write as much of pending kvecs to the socket as we can.
851 * 0 -> socket full, but more to do
854 static int write_partial_kvec(struct ceph_connection
*con
)
858 dout("write_partial_kvec %p %d left\n", con
, con
->out_kvec_bytes
);
859 while (con
->out_kvec_bytes
> 0) {
860 ret
= ceph_tcp_sendmsg(con
->sock
, con
->out_kvec_cur
,
861 con
->out_kvec_left
, con
->out_kvec_bytes
,
865 con
->out_kvec_bytes
-= ret
;
866 if (con
->out_kvec_bytes
== 0)
869 /* account for full iov entries consumed */
870 while (ret
>= con
->out_kvec_cur
->iov_len
) {
871 BUG_ON(!con
->out_kvec_left
);
872 ret
-= con
->out_kvec_cur
->iov_len
;
874 con
->out_kvec_left
--;
876 /* and for a partially-consumed entry */
878 con
->out_kvec_cur
->iov_len
-= ret
;
879 con
->out_kvec_cur
->iov_base
+= ret
;
882 con
->out_kvec_left
= 0;
883 con
->out_kvec_is_msg
= false;
886 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con
,
887 con
->out_kvec_bytes
, con
->out_kvec_left
, ret
);
888 return ret
; /* done! */
891 static void out_msg_pos_next(struct ceph_connection
*con
, struct page
*page
,
892 size_t len
, size_t sent
, bool in_trail
)
894 struct ceph_msg
*msg
= con
->out_msg
;
899 con
->out_msg_pos
.data_pos
+= sent
;
900 con
->out_msg_pos
.page_pos
+= sent
;
902 con
->out_msg_pos
.page_pos
= 0;
903 con
->out_msg_pos
.page
++;
904 con
->out_msg_pos
.did_page_crc
= false;
906 list_move_tail(&page
->lru
,
908 else if (msg
->pagelist
)
909 list_move_tail(&page
->lru
,
910 &msg
->pagelist
->head
);
913 iter_bio_next(&msg
->bio_iter
, &msg
->bio_seg
);
919 * Write as much message data payload as we can. If we finish, queue
921 * 1 -> done, footer is now queued in out_kvec[].
922 * 0 -> socket full, but more to do
925 static int write_partial_msg_pages(struct ceph_connection
*con
)
927 struct ceph_msg
*msg
= con
->out_msg
;
928 unsigned int data_len
= le32_to_cpu(msg
->hdr
.data_len
);
930 bool do_datacrc
= !con
->msgr
->nocrc
;
933 bool in_trail
= false;
934 size_t trail_len
= (msg
->trail
? msg
->trail
->length
: 0);
936 dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
937 con
, msg
, con
->out_msg_pos
.page
, msg
->nr_pages
,
938 con
->out_msg_pos
.page_pos
);
940 while (data_len
> con
->out_msg_pos
.data_pos
) {
941 struct page
*page
= NULL
;
942 int max_write
= PAGE_SIZE
;
945 total_max_write
= data_len
- trail_len
-
946 con
->out_msg_pos
.data_pos
;
949 * if we are calculating the data crc (the default), we need
950 * to map the page. if our pages[] has been revoked, use the
954 /* have we reached the trail part of the data? */
955 if (con
->out_msg_pos
.data_pos
>= data_len
- trail_len
) {
958 total_max_write
= data_len
- con
->out_msg_pos
.data_pos
;
960 page
= list_first_entry(&msg
->trail
->head
,
962 } else if (msg
->pages
) {
963 page
= msg
->pages
[con
->out_msg_pos
.page
];
964 } else if (msg
->pagelist
) {
965 page
= list_first_entry(&msg
->pagelist
->head
,
968 } else if (msg
->bio
) {
971 bv
= bio_iovec_idx(msg
->bio_iter
, msg
->bio_seg
);
973 bio_offset
= bv
->bv_offset
;
974 max_write
= bv
->bv_len
;
979 len
= min_t(int, max_write
- con
->out_msg_pos
.page_pos
,
982 if (do_datacrc
&& !con
->out_msg_pos
.did_page_crc
) {
985 u32 tmpcrc
= le32_to_cpu(msg
->footer
.data_crc
);
989 BUG_ON(kaddr
== NULL
);
990 base
= kaddr
+ con
->out_msg_pos
.page_pos
+ bio_offset
;
991 crc
= crc32c(tmpcrc
, base
, len
);
992 msg
->footer
.data_crc
= cpu_to_le32(crc
);
993 con
->out_msg_pos
.did_page_crc
= true;
995 ret
= ceph_tcp_sendpage(con
->sock
, page
,
996 con
->out_msg_pos
.page_pos
+ bio_offset
,
1005 out_msg_pos_next(con
, page
, len
, (size_t) ret
, in_trail
);
1008 dout("write_partial_msg_pages %p msg %p done\n", con
, msg
);
1010 /* prepare and queue up footer, too */
1012 msg
->footer
.flags
|= CEPH_MSG_FOOTER_NOCRC
;
1013 con_out_kvec_reset(con
);
1014 prepare_write_message_footer(con
);
1023 static int write_partial_skip(struct ceph_connection
*con
)
1027 while (con
->out_skip
> 0) {
1028 size_t size
= min(con
->out_skip
, (int) PAGE_CACHE_SIZE
);
1030 ret
= ceph_tcp_sendpage(con
->sock
, zero_page
, 0, size
, 1);
1033 con
->out_skip
-= ret
;
1041 * Prepare to read connection handshake, or an ack.
1043 static void prepare_read_banner(struct ceph_connection
*con
)
1045 dout("prepare_read_banner %p\n", con
);
1046 con
->in_base_pos
= 0;
1049 static void prepare_read_connect(struct ceph_connection
*con
)
1051 dout("prepare_read_connect %p\n", con
);
1052 con
->in_base_pos
= 0;
1055 static void prepare_read_ack(struct ceph_connection
*con
)
1057 dout("prepare_read_ack %p\n", con
);
1058 con
->in_base_pos
= 0;
1061 static void prepare_read_tag(struct ceph_connection
*con
)
1063 dout("prepare_read_tag %p\n", con
);
1064 con
->in_base_pos
= 0;
1065 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1069 * Prepare to read a message.
1071 static int prepare_read_message(struct ceph_connection
*con
)
1073 dout("prepare_read_message %p\n", con
);
1074 BUG_ON(con
->in_msg
!= NULL
);
1075 con
->in_base_pos
= 0;
1076 con
->in_front_crc
= con
->in_middle_crc
= con
->in_data_crc
= 0;
1081 static int read_partial(struct ceph_connection
*con
,
1082 int end
, int size
, void *object
)
1084 while (con
->in_base_pos
< end
) {
1085 int left
= end
- con
->in_base_pos
;
1086 int have
= size
- left
;
1087 int ret
= ceph_tcp_recvmsg(con
->sock
, object
+ have
, left
);
1090 con
->in_base_pos
+= ret
;
1097 * Read all or part of the connect-side handshake on a new connection
1099 static int read_partial_banner(struct ceph_connection
*con
)
1105 dout("read_partial_banner %p at %d\n", con
, con
->in_base_pos
);
1108 size
= strlen(CEPH_BANNER
);
1110 ret
= read_partial(con
, end
, size
, con
->in_banner
);
1114 size
= sizeof (con
->actual_peer_addr
);
1116 ret
= read_partial(con
, end
, size
, &con
->actual_peer_addr
);
1120 size
= sizeof (con
->peer_addr_for_me
);
1122 ret
= read_partial(con
, end
, size
, &con
->peer_addr_for_me
);
1130 static int read_partial_connect(struct ceph_connection
*con
)
1136 dout("read_partial_connect %p at %d\n", con
, con
->in_base_pos
);
1138 size
= sizeof (con
->in_reply
);
1140 ret
= read_partial(con
, end
, size
, &con
->in_reply
);
1144 size
= le32_to_cpu(con
->in_reply
.authorizer_len
);
1146 ret
= read_partial(con
, end
, size
, con
->auth_reply_buf
);
1150 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1151 con
, (int)con
->in_reply
.tag
,
1152 le32_to_cpu(con
->in_reply
.connect_seq
),
1153 le32_to_cpu(con
->in_reply
.global_seq
));
1160 * Verify the hello banner looks okay.
1162 static int verify_hello(struct ceph_connection
*con
)
1164 if (memcmp(con
->in_banner
, CEPH_BANNER
, strlen(CEPH_BANNER
))) {
1165 pr_err("connect to %s got bad banner\n",
1166 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1167 con
->error_msg
= "protocol error, bad banner";
1173 static bool addr_is_blank(struct sockaddr_storage
*ss
)
1175 switch (ss
->ss_family
) {
1177 return ((struct sockaddr_in
*)ss
)->sin_addr
.s_addr
== 0;
1180 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[0] == 0 &&
1181 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[1] == 0 &&
1182 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[2] == 0 &&
1183 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[3] == 0;
1188 static int addr_port(struct sockaddr_storage
*ss
)
1190 switch (ss
->ss_family
) {
1192 return ntohs(((struct sockaddr_in
*)ss
)->sin_port
);
1194 return ntohs(((struct sockaddr_in6
*)ss
)->sin6_port
);
1199 static void addr_set_port(struct sockaddr_storage
*ss
, int p
)
1201 switch (ss
->ss_family
) {
1203 ((struct sockaddr_in
*)ss
)->sin_port
= htons(p
);
1206 ((struct sockaddr_in6
*)ss
)->sin6_port
= htons(p
);
1212 * Unlike other *_pton function semantics, zero indicates success.
1214 static int ceph_pton(const char *str
, size_t len
, struct sockaddr_storage
*ss
,
1215 char delim
, const char **ipend
)
1217 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
1218 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
1220 memset(ss
, 0, sizeof(*ss
));
1222 if (in4_pton(str
, len
, (u8
*)&in4
->sin_addr
.s_addr
, delim
, ipend
)) {
1223 ss
->ss_family
= AF_INET
;
1227 if (in6_pton(str
, len
, (u8
*)&in6
->sin6_addr
.s6_addr
, delim
, ipend
)) {
1228 ss
->ss_family
= AF_INET6
;
1236 * Extract hostname string and resolve using kernel DNS facility.
1238 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1239 static int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1240 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1242 const char *end
, *delim_p
;
1243 char *colon_p
, *ip_addr
= NULL
;
1247 * The end of the hostname occurs immediately preceding the delimiter or
1248 * the port marker (':') where the delimiter takes precedence.
1250 delim_p
= memchr(name
, delim
, namelen
);
1251 colon_p
= memchr(name
, ':', namelen
);
1253 if (delim_p
&& colon_p
)
1254 end
= delim_p
< colon_p
? delim_p
: colon_p
;
1255 else if (!delim_p
&& colon_p
)
1259 if (!end
) /* case: hostname:/ */
1260 end
= name
+ namelen
;
1266 /* do dns_resolve upcall */
1267 ip_len
= dns_query(NULL
, name
, end
- name
, NULL
, &ip_addr
, NULL
);
1269 ret
= ceph_pton(ip_addr
, ip_len
, ss
, -1, NULL
);
1277 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end
- name
), name
,
1278 ret
, ret
? "failed" : ceph_pr_addr(ss
));
1283 static inline int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1284 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1291 * Parse a server name (IP or hostname). If a valid IP address is not found
1292 * then try to extract a hostname to resolve using userspace DNS upcall.
1294 static int ceph_parse_server_name(const char *name
, size_t namelen
,
1295 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1299 ret
= ceph_pton(name
, namelen
, ss
, delim
, ipend
);
1301 ret
= ceph_dns_resolve_name(name
, namelen
, ss
, delim
, ipend
);
1307 * Parse an ip[:port] list into an addr array. Use the default
1308 * monitor port if a port isn't specified.
1310 int ceph_parse_ips(const char *c
, const char *end
,
1311 struct ceph_entity_addr
*addr
,
1312 int max_count
, int *count
)
1314 int i
, ret
= -EINVAL
;
1317 dout("parse_ips on '%.*s'\n", (int)(end
-c
), c
);
1318 for (i
= 0; i
< max_count
; i
++) {
1320 struct sockaddr_storage
*ss
= &addr
[i
].in_addr
;
1329 ret
= ceph_parse_server_name(p
, end
- p
, ss
, delim
, &ipend
);
1338 dout("missing matching ']'\n");
1345 if (p
< end
&& *p
== ':') {
1348 while (p
< end
&& *p
>= '0' && *p
<= '9') {
1349 port
= (port
* 10) + (*p
- '0');
1352 if (port
> 65535 || port
== 0)
1355 port
= CEPH_MON_PORT
;
1358 addr_set_port(ss
, port
);
1360 dout("parse_ips got %s\n", ceph_pr_addr(ss
));
1377 pr_err("parse_ips bad ip '%.*s'\n", (int)(end
- c
), c
);
1380 EXPORT_SYMBOL(ceph_parse_ips
);
1382 static int process_banner(struct ceph_connection
*con
)
1384 dout("process_banner on %p\n", con
);
1386 if (verify_hello(con
) < 0)
1389 ceph_decode_addr(&con
->actual_peer_addr
);
1390 ceph_decode_addr(&con
->peer_addr_for_me
);
1393 * Make sure the other end is who we wanted. note that the other
1394 * end may not yet know their ip address, so if it's 0.0.0.0, give
1395 * them the benefit of the doubt.
1397 if (memcmp(&con
->peer_addr
, &con
->actual_peer_addr
,
1398 sizeof(con
->peer_addr
)) != 0 &&
1399 !(addr_is_blank(&con
->actual_peer_addr
.in_addr
) &&
1400 con
->actual_peer_addr
.nonce
== con
->peer_addr
.nonce
)) {
1401 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1402 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1403 (int)le32_to_cpu(con
->peer_addr
.nonce
),
1404 ceph_pr_addr(&con
->actual_peer_addr
.in_addr
),
1405 (int)le32_to_cpu(con
->actual_peer_addr
.nonce
));
1406 con
->error_msg
= "wrong peer at address";
1411 * did we learn our address?
1413 if (addr_is_blank(&con
->msgr
->inst
.addr
.in_addr
)) {
1414 int port
= addr_port(&con
->msgr
->inst
.addr
.in_addr
);
1416 memcpy(&con
->msgr
->inst
.addr
.in_addr
,
1417 &con
->peer_addr_for_me
.in_addr
,
1418 sizeof(con
->peer_addr_for_me
.in_addr
));
1419 addr_set_port(&con
->msgr
->inst
.addr
.in_addr
, port
);
1420 encode_my_addr(con
->msgr
);
1421 dout("process_banner learned my addr is %s\n",
1422 ceph_pr_addr(&con
->msgr
->inst
.addr
.in_addr
));
1425 set_bit(NEGOTIATING
, &con
->state
);
1426 prepare_read_connect(con
);
1430 static void fail_protocol(struct ceph_connection
*con
)
1432 reset_connection(con
);
1433 set_bit(CLOSED
, &con
->state
); /* in case there's queued work */
1436 static int process_connect(struct ceph_connection
*con
)
1438 u64 sup_feat
= con
->msgr
->supported_features
;
1439 u64 req_feat
= con
->msgr
->required_features
;
1440 u64 server_feat
= le64_to_cpu(con
->in_reply
.features
);
1443 dout("process_connect on %p tag %d\n", con
, (int)con
->in_tag
);
1445 switch (con
->in_reply
.tag
) {
1446 case CEPH_MSGR_TAG_FEATURES
:
1447 pr_err("%s%lld %s feature set mismatch,"
1448 " my %llx < server's %llx, missing %llx\n",
1449 ENTITY_NAME(con
->peer_name
),
1450 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1451 sup_feat
, server_feat
, server_feat
& ~sup_feat
);
1452 con
->error_msg
= "missing required protocol features";
1456 case CEPH_MSGR_TAG_BADPROTOVER
:
1457 pr_err("%s%lld %s protocol version mismatch,"
1458 " my %d != server's %d\n",
1459 ENTITY_NAME(con
->peer_name
),
1460 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1461 le32_to_cpu(con
->out_connect
.protocol_version
),
1462 le32_to_cpu(con
->in_reply
.protocol_version
));
1463 con
->error_msg
= "protocol version mismatch";
1467 case CEPH_MSGR_TAG_BADAUTHORIZER
:
1469 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con
,
1471 if (con
->auth_retry
== 2) {
1472 con
->error_msg
= "connect authorization failure";
1475 con
->auth_retry
= 1;
1476 con_out_kvec_reset(con
);
1477 ret
= prepare_write_connect(con
);
1480 prepare_read_connect(con
);
1483 case CEPH_MSGR_TAG_RESETSESSION
:
1485 * If we connected with a large connect_seq but the peer
1486 * has no record of a session with us (no connection, or
1487 * connect_seq == 0), they will send RESETSESION to indicate
1488 * that they must have reset their session, and may have
1491 dout("process_connect got RESET peer seq %u\n",
1492 le32_to_cpu(con
->in_connect
.connect_seq
));
1493 pr_err("%s%lld %s connection reset\n",
1494 ENTITY_NAME(con
->peer_name
),
1495 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1496 reset_connection(con
);
1497 con_out_kvec_reset(con
);
1498 ret
= prepare_write_connect(con
);
1501 prepare_read_connect(con
);
1503 /* Tell ceph about it. */
1504 mutex_unlock(&con
->mutex
);
1505 pr_info("reset on %s%lld\n", ENTITY_NAME(con
->peer_name
));
1506 if (con
->ops
->peer_reset
)
1507 con
->ops
->peer_reset(con
);
1508 mutex_lock(&con
->mutex
);
1509 if (test_bit(CLOSED
, &con
->state
) ||
1510 test_bit(OPENING
, &con
->state
))
1514 case CEPH_MSGR_TAG_RETRY_SESSION
:
1516 * If we sent a smaller connect_seq than the peer has, try
1517 * again with a larger value.
1519 dout("process_connect got RETRY my seq = %u, peer_seq = %u\n",
1520 le32_to_cpu(con
->out_connect
.connect_seq
),
1521 le32_to_cpu(con
->in_connect
.connect_seq
));
1522 con
->connect_seq
= le32_to_cpu(con
->in_connect
.connect_seq
);
1523 con_out_kvec_reset(con
);
1524 ret
= prepare_write_connect(con
);
1527 prepare_read_connect(con
);
1530 case CEPH_MSGR_TAG_RETRY_GLOBAL
:
1532 * If we sent a smaller global_seq than the peer has, try
1533 * again with a larger value.
1535 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1536 con
->peer_global_seq
,
1537 le32_to_cpu(con
->in_connect
.global_seq
));
1538 get_global_seq(con
->msgr
,
1539 le32_to_cpu(con
->in_connect
.global_seq
));
1540 con_out_kvec_reset(con
);
1541 ret
= prepare_write_connect(con
);
1544 prepare_read_connect(con
);
1547 case CEPH_MSGR_TAG_READY
:
1548 if (req_feat
& ~server_feat
) {
1549 pr_err("%s%lld %s protocol feature mismatch,"
1550 " my required %llx > server's %llx, need %llx\n",
1551 ENTITY_NAME(con
->peer_name
),
1552 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1553 req_feat
, server_feat
, req_feat
& ~server_feat
);
1554 con
->error_msg
= "missing required protocol features";
1558 clear_bit(CONNECTING
, &con
->state
);
1559 con
->peer_global_seq
= le32_to_cpu(con
->in_reply
.global_seq
);
1561 con
->peer_features
= server_feat
;
1562 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1563 con
->peer_global_seq
,
1564 le32_to_cpu(con
->in_reply
.connect_seq
),
1566 WARN_ON(con
->connect_seq
!=
1567 le32_to_cpu(con
->in_reply
.connect_seq
));
1569 if (con
->in_reply
.flags
& CEPH_MSG_CONNECT_LOSSY
)
1570 set_bit(LOSSYTX
, &con
->flags
);
1572 prepare_read_tag(con
);
1575 case CEPH_MSGR_TAG_WAIT
:
1577 * If there is a connection race (we are opening
1578 * connections to each other), one of us may just have
1579 * to WAIT. This shouldn't happen if we are the
1582 pr_err("process_connect got WAIT as client\n");
1583 con
->error_msg
= "protocol error, got WAIT as client";
1587 pr_err("connect protocol error, will retry\n");
1588 con
->error_msg
= "protocol error, garbage tag during connect";
1596 * read (part of) an ack
1598 static int read_partial_ack(struct ceph_connection
*con
)
1600 int size
= sizeof (con
->in_temp_ack
);
1603 return read_partial(con
, end
, size
, &con
->in_temp_ack
);
1608 * We can finally discard anything that's been acked.
1610 static void process_ack(struct ceph_connection
*con
)
1613 u64 ack
= le64_to_cpu(con
->in_temp_ack
);
1616 while (!list_empty(&con
->out_sent
)) {
1617 m
= list_first_entry(&con
->out_sent
, struct ceph_msg
,
1619 seq
= le64_to_cpu(m
->hdr
.seq
);
1622 dout("got ack for seq %llu type %d at %p\n", seq
,
1623 le16_to_cpu(m
->hdr
.type
), m
);
1624 m
->ack_stamp
= jiffies
;
1627 prepare_read_tag(con
);
1633 static int read_partial_message_section(struct ceph_connection
*con
,
1634 struct kvec
*section
,
1635 unsigned int sec_len
, u32
*crc
)
1641 while (section
->iov_len
< sec_len
) {
1642 BUG_ON(section
->iov_base
== NULL
);
1643 left
= sec_len
- section
->iov_len
;
1644 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)section
->iov_base
+
1645 section
->iov_len
, left
);
1648 section
->iov_len
+= ret
;
1650 if (section
->iov_len
== sec_len
)
1651 *crc
= crc32c(0, section
->iov_base
, section
->iov_len
);
1656 static bool ceph_con_in_msg_alloc(struct ceph_connection
*con
,
1657 struct ceph_msg_header
*hdr
);
1660 static int read_partial_message_pages(struct ceph_connection
*con
,
1661 struct page
**pages
,
1662 unsigned int data_len
, bool do_datacrc
)
1668 left
= min((int)(data_len
- con
->in_msg_pos
.data_pos
),
1669 (int)(PAGE_SIZE
- con
->in_msg_pos
.page_pos
));
1671 BUG_ON(pages
== NULL
);
1672 p
= kmap(pages
[con
->in_msg_pos
.page
]);
1673 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ con
->in_msg_pos
.page_pos
,
1675 if (ret
> 0 && do_datacrc
)
1677 crc32c(con
->in_data_crc
,
1678 p
+ con
->in_msg_pos
.page_pos
, ret
);
1679 kunmap(pages
[con
->in_msg_pos
.page
]);
1682 con
->in_msg_pos
.data_pos
+= ret
;
1683 con
->in_msg_pos
.page_pos
+= ret
;
1684 if (con
->in_msg_pos
.page_pos
== PAGE_SIZE
) {
1685 con
->in_msg_pos
.page_pos
= 0;
1686 con
->in_msg_pos
.page
++;
1693 static int read_partial_message_bio(struct ceph_connection
*con
,
1694 struct bio
**bio_iter
, int *bio_seg
,
1695 unsigned int data_len
, bool do_datacrc
)
1697 struct bio_vec
*bv
= bio_iovec_idx(*bio_iter
, *bio_seg
);
1704 left
= min((int)(data_len
- con
->in_msg_pos
.data_pos
),
1705 (int)(bv
->bv_len
- con
->in_msg_pos
.page_pos
));
1707 p
= kmap(bv
->bv_page
) + bv
->bv_offset
;
1709 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ con
->in_msg_pos
.page_pos
,
1711 if (ret
> 0 && do_datacrc
)
1713 crc32c(con
->in_data_crc
,
1714 p
+ con
->in_msg_pos
.page_pos
, ret
);
1715 kunmap(bv
->bv_page
);
1718 con
->in_msg_pos
.data_pos
+= ret
;
1719 con
->in_msg_pos
.page_pos
+= ret
;
1720 if (con
->in_msg_pos
.page_pos
== bv
->bv_len
) {
1721 con
->in_msg_pos
.page_pos
= 0;
1722 iter_bio_next(bio_iter
, bio_seg
);
1730 * read (part of) a message.
1732 static int read_partial_message(struct ceph_connection
*con
)
1734 struct ceph_msg
*m
= con
->in_msg
;
1738 unsigned int front_len
, middle_len
, data_len
;
1739 bool do_datacrc
= !con
->msgr
->nocrc
;
1743 dout("read_partial_message con %p msg %p\n", con
, m
);
1746 size
= sizeof (con
->in_hdr
);
1748 ret
= read_partial(con
, end
, size
, &con
->in_hdr
);
1752 crc
= crc32c(0, &con
->in_hdr
, offsetof(struct ceph_msg_header
, crc
));
1753 if (cpu_to_le32(crc
) != con
->in_hdr
.crc
) {
1754 pr_err("read_partial_message bad hdr "
1755 " crc %u != expected %u\n",
1756 crc
, con
->in_hdr
.crc
);
1760 front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
1761 if (front_len
> CEPH_MSG_MAX_FRONT_LEN
)
1763 middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
1764 if (middle_len
> CEPH_MSG_MAX_DATA_LEN
)
1766 data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
1767 if (data_len
> CEPH_MSG_MAX_DATA_LEN
)
1771 seq
= le64_to_cpu(con
->in_hdr
.seq
);
1772 if ((s64
)seq
- (s64
)con
->in_seq
< 1) {
1773 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1774 ENTITY_NAME(con
->peer_name
),
1775 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1776 seq
, con
->in_seq
+ 1);
1777 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1779 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1781 } else if ((s64
)seq
- (s64
)con
->in_seq
> 1) {
1782 pr_err("read_partial_message bad seq %lld expected %lld\n",
1783 seq
, con
->in_seq
+ 1);
1784 con
->error_msg
= "bad message sequence # for incoming message";
1788 /* allocate message? */
1790 dout("got hdr type %d front %d data %d\n", con
->in_hdr
.type
,
1791 con
->in_hdr
.front_len
, con
->in_hdr
.data_len
);
1792 if (ceph_con_in_msg_alloc(con
, &con
->in_hdr
)) {
1793 /* skip this message */
1794 dout("alloc_msg said skip message\n");
1795 BUG_ON(con
->in_msg
);
1796 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1798 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1804 "error allocating memory for incoming message";
1808 BUG_ON(con
->in_msg
->con
!= con
);
1810 m
->front
.iov_len
= 0; /* haven't read it yet */
1812 m
->middle
->vec
.iov_len
= 0;
1814 con
->in_msg_pos
.page
= 0;
1816 con
->in_msg_pos
.page_pos
= m
->page_alignment
;
1818 con
->in_msg_pos
.page_pos
= 0;
1819 con
->in_msg_pos
.data_pos
= 0;
1823 ret
= read_partial_message_section(con
, &m
->front
, front_len
,
1824 &con
->in_front_crc
);
1830 ret
= read_partial_message_section(con
, &m
->middle
->vec
,
1832 &con
->in_middle_crc
);
1837 if (m
->bio
&& !m
->bio_iter
)
1838 init_bio_iter(m
->bio
, &m
->bio_iter
, &m
->bio_seg
);
1842 while (con
->in_msg_pos
.data_pos
< data_len
) {
1844 ret
= read_partial_message_pages(con
, m
->pages
,
1845 data_len
, do_datacrc
);
1849 } else if (m
->bio
) {
1851 ret
= read_partial_message_bio(con
,
1852 &m
->bio_iter
, &m
->bio_seg
,
1853 data_len
, do_datacrc
);
1863 size
= sizeof (m
->footer
);
1865 ret
= read_partial(con
, end
, size
, &m
->footer
);
1869 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1870 m
, front_len
, m
->footer
.front_crc
, middle_len
,
1871 m
->footer
.middle_crc
, data_len
, m
->footer
.data_crc
);
1874 if (con
->in_front_crc
!= le32_to_cpu(m
->footer
.front_crc
)) {
1875 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1876 m
, con
->in_front_crc
, m
->footer
.front_crc
);
1879 if (con
->in_middle_crc
!= le32_to_cpu(m
->footer
.middle_crc
)) {
1880 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1881 m
, con
->in_middle_crc
, m
->footer
.middle_crc
);
1885 (m
->footer
.flags
& CEPH_MSG_FOOTER_NOCRC
) == 0 &&
1886 con
->in_data_crc
!= le32_to_cpu(m
->footer
.data_crc
)) {
1887 pr_err("read_partial_message %p data crc %u != exp. %u\n", m
,
1888 con
->in_data_crc
, le32_to_cpu(m
->footer
.data_crc
));
1892 return 1; /* done! */
1896 * Process message. This happens in the worker thread. The callback should
1897 * be careful not to do anything that waits on other incoming messages or it
1900 static void process_message(struct ceph_connection
*con
)
1902 struct ceph_msg
*msg
;
1904 BUG_ON(con
->in_msg
->con
!= con
);
1905 con
->in_msg
->con
= NULL
;
1910 /* if first message, set peer_name */
1911 if (con
->peer_name
.type
== 0)
1912 con
->peer_name
= msg
->hdr
.src
;
1915 mutex_unlock(&con
->mutex
);
1917 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
1918 msg
, le64_to_cpu(msg
->hdr
.seq
),
1919 ENTITY_NAME(msg
->hdr
.src
),
1920 le16_to_cpu(msg
->hdr
.type
),
1921 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
1922 le32_to_cpu(msg
->hdr
.front_len
),
1923 le32_to_cpu(msg
->hdr
.data_len
),
1924 con
->in_front_crc
, con
->in_middle_crc
, con
->in_data_crc
);
1925 con
->ops
->dispatch(con
, msg
);
1927 mutex_lock(&con
->mutex
);
1928 prepare_read_tag(con
);
1933 * Write something to the socket. Called in a worker thread when the
1934 * socket appears to be writeable and we have something ready to send.
1936 static int try_write(struct ceph_connection
*con
)
1940 dout("try_write start %p state %lu\n", con
, con
->state
);
1943 dout("try_write out_kvec_bytes %d\n", con
->out_kvec_bytes
);
1945 /* open the socket first? */
1946 if (con
->sock
== NULL
) {
1947 clear_bit(NEGOTIATING
, &con
->state
);
1948 set_bit(CONNECTING
, &con
->state
);
1950 con_out_kvec_reset(con
);
1951 prepare_write_banner(con
);
1952 ret
= prepare_write_connect(con
);
1955 prepare_read_banner(con
);
1957 BUG_ON(con
->in_msg
);
1958 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1959 dout("try_write initiating connect on %p new state %lu\n",
1961 ret
= ceph_tcp_connect(con
);
1963 con
->error_msg
= "connect error";
1969 /* kvec data queued? */
1970 if (con
->out_skip
) {
1971 ret
= write_partial_skip(con
);
1975 if (con
->out_kvec_left
) {
1976 ret
= write_partial_kvec(con
);
1983 if (con
->out_msg_done
) {
1984 ceph_msg_put(con
->out_msg
);
1985 con
->out_msg
= NULL
; /* we're done with this one */
1989 ret
= write_partial_msg_pages(con
);
1991 goto more_kvec
; /* we need to send the footer, too! */
1995 dout("try_write write_partial_msg_pages err %d\n",
2002 if (!test_bit(CONNECTING
, &con
->state
)) {
2003 /* is anything else pending? */
2004 if (!list_empty(&con
->out_queue
)) {
2005 prepare_write_message(con
);
2008 if (con
->in_seq
> con
->in_seq_acked
) {
2009 prepare_write_ack(con
);
2012 if (test_and_clear_bit(KEEPALIVE_PENDING
, &con
->flags
)) {
2013 prepare_write_keepalive(con
);
2018 /* Nothing to do! */
2019 clear_bit(WRITE_PENDING
, &con
->flags
);
2020 dout("try_write nothing else to write.\n");
2023 dout("try_write done on %p ret %d\n", con
, ret
);
2030 * Read what we can from the socket.
2032 static int try_read(struct ceph_connection
*con
)
2039 if (test_bit(STANDBY
, &con
->state
))
2042 dout("try_read start on %p\n", con
);
2045 dout("try_read tag %d in_base_pos %d\n", (int)con
->in_tag
,
2049 * process_connect and process_message drop and re-take
2050 * con->mutex. make sure we handle a racing close or reopen.
2052 if (test_bit(CLOSED
, &con
->state
) ||
2053 test_bit(OPENING
, &con
->state
)) {
2058 if (test_bit(CONNECTING
, &con
->state
)) {
2059 if (!test_bit(NEGOTIATING
, &con
->state
)) {
2060 dout("try_read connecting\n");
2061 ret
= read_partial_banner(con
);
2064 ret
= process_banner(con
);
2068 ret
= read_partial_connect(con
);
2071 ret
= process_connect(con
);
2077 if (con
->in_base_pos
< 0) {
2079 * skipping + discarding content.
2081 * FIXME: there must be a better way to do this!
2083 static char buf
[SKIP_BUF_SIZE
];
2084 int skip
= min((int) sizeof (buf
), -con
->in_base_pos
);
2086 dout("skipping %d / %d bytes\n", skip
, -con
->in_base_pos
);
2087 ret
= ceph_tcp_recvmsg(con
->sock
, buf
, skip
);
2090 con
->in_base_pos
+= ret
;
2091 if (con
->in_base_pos
)
2094 if (con
->in_tag
== CEPH_MSGR_TAG_READY
) {
2098 ret
= ceph_tcp_recvmsg(con
->sock
, &con
->in_tag
, 1);
2101 dout("try_read got tag %d\n", (int)con
->in_tag
);
2102 switch (con
->in_tag
) {
2103 case CEPH_MSGR_TAG_MSG
:
2104 prepare_read_message(con
);
2106 case CEPH_MSGR_TAG_ACK
:
2107 prepare_read_ack(con
);
2109 case CEPH_MSGR_TAG_CLOSE
:
2110 set_bit(CLOSED
, &con
->state
); /* fixme */
2116 if (con
->in_tag
== CEPH_MSGR_TAG_MSG
) {
2117 ret
= read_partial_message(con
);
2121 con
->error_msg
= "bad crc";
2125 con
->error_msg
= "io error";
2130 if (con
->in_tag
== CEPH_MSGR_TAG_READY
)
2132 process_message(con
);
2135 if (con
->in_tag
== CEPH_MSGR_TAG_ACK
) {
2136 ret
= read_partial_ack(con
);
2144 dout("try_read done on %p ret %d\n", con
, ret
);
2148 pr_err("try_read bad con->in_tag = %d\n", (int)con
->in_tag
);
2149 con
->error_msg
= "protocol error, garbage tag";
2156 * Atomically queue work on a connection. Bump @con reference to
2157 * avoid races with connection teardown.
2159 static void queue_con(struct ceph_connection
*con
)
2161 if (!con
->ops
->get(con
)) {
2162 dout("queue_con %p ref count 0\n", con
);
2166 if (!queue_delayed_work(ceph_msgr_wq
, &con
->work
, 0)) {
2167 dout("queue_con %p - already queued\n", con
);
2170 dout("queue_con %p\n", con
);
2175 * Do some work on a connection. Drop a connection ref when we're done.
2177 static void con_work(struct work_struct
*work
)
2179 struct ceph_connection
*con
= container_of(work
, struct ceph_connection
,
2183 mutex_lock(&con
->mutex
);
2185 if (test_and_clear_bit(SOCK_CLOSED
, &con
->flags
)) {
2186 if (test_bit(CONNECTING
, &con
->state
))
2187 con
->error_msg
= "connection failed";
2189 con
->error_msg
= "socket closed";
2193 if (test_and_clear_bit(BACKOFF
, &con
->flags
)) {
2194 dout("con_work %p backing off\n", con
);
2195 if (queue_delayed_work(ceph_msgr_wq
, &con
->work
,
2196 round_jiffies_relative(con
->delay
))) {
2197 dout("con_work %p backoff %lu\n", con
, con
->delay
);
2198 mutex_unlock(&con
->mutex
);
2202 dout("con_work %p FAILED to back off %lu\n", con
,
2207 if (test_bit(STANDBY
, &con
->state
)) {
2208 dout("con_work %p STANDBY\n", con
);
2211 if (test_bit(CLOSED
, &con
->state
)) { /* e.g. if we are replaced */
2212 dout("con_work CLOSED\n");
2213 con_close_socket(con
);
2216 if (test_and_clear_bit(OPENING
, &con
->state
)) {
2217 /* reopen w/ new peer */
2218 dout("con_work OPENING\n");
2219 con_close_socket(con
);
2222 ret
= try_read(con
);
2228 ret
= try_write(con
);
2235 mutex_unlock(&con
->mutex
);
2241 mutex_unlock(&con
->mutex
);
2242 ceph_fault(con
); /* error/fault path */
2248 * Generic error/fault handler. A retry mechanism is used with
2249 * exponential backoff
2251 static void ceph_fault(struct ceph_connection
*con
)
2253 pr_err("%s%lld %s %s\n", ENTITY_NAME(con
->peer_name
),
2254 ceph_pr_addr(&con
->peer_addr
.in_addr
), con
->error_msg
);
2255 dout("fault %p state %lu to peer %s\n",
2256 con
, con
->state
, ceph_pr_addr(&con
->peer_addr
.in_addr
));
2258 if (test_bit(LOSSYTX
, &con
->flags
)) {
2259 dout("fault on LOSSYTX channel\n");
2263 mutex_lock(&con
->mutex
);
2264 if (test_bit(CLOSED
, &con
->state
))
2267 con_close_socket(con
);
2270 BUG_ON(con
->in_msg
->con
!= con
);
2271 con
->in_msg
->con
= NULL
;
2272 ceph_msg_put(con
->in_msg
);
2277 /* Requeue anything that hasn't been acked */
2278 list_splice_init(&con
->out_sent
, &con
->out_queue
);
2280 /* If there are no messages queued or keepalive pending, place
2281 * the connection in a STANDBY state */
2282 if (list_empty(&con
->out_queue
) &&
2283 !test_bit(KEEPALIVE_PENDING
, &con
->flags
)) {
2284 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con
);
2285 clear_bit(WRITE_PENDING
, &con
->flags
);
2286 set_bit(STANDBY
, &con
->state
);
2288 /* retry after a delay. */
2289 if (con
->delay
== 0)
2290 con
->delay
= BASE_DELAY_INTERVAL
;
2291 else if (con
->delay
< MAX_DELAY_INTERVAL
)
2294 if (queue_delayed_work(ceph_msgr_wq
, &con
->work
,
2295 round_jiffies_relative(con
->delay
))) {
2296 dout("fault queued %p delay %lu\n", con
, con
->delay
);
2299 dout("fault failed to queue %p delay %lu, backoff\n",
2302 * In many cases we see a socket state change
2303 * while con_work is running and end up
2304 * queuing (non-delayed) work, such that we
2305 * can't backoff with a delay. Set a flag so
2306 * that when con_work restarts we schedule the
2309 set_bit(BACKOFF
, &con
->flags
);
2314 mutex_unlock(&con
->mutex
);
2317 * in case we faulted due to authentication, invalidate our
2318 * current tickets so that we can get new ones.
2320 if (con
->auth_retry
&& con
->ops
->invalidate_authorizer
) {
2321 dout("calling invalidate_authorizer()\n");
2322 con
->ops
->invalidate_authorizer(con
);
2325 if (con
->ops
->fault
)
2326 con
->ops
->fault(con
);
2332 * initialize a new messenger instance
2334 void ceph_messenger_init(struct ceph_messenger
*msgr
,
2335 struct ceph_entity_addr
*myaddr
,
2336 u32 supported_features
,
2337 u32 required_features
,
2340 msgr
->supported_features
= supported_features
;
2341 msgr
->required_features
= required_features
;
2343 spin_lock_init(&msgr
->global_seq_lock
);
2346 msgr
->inst
.addr
= *myaddr
;
2348 /* select a random nonce */
2349 msgr
->inst
.addr
.type
= 0;
2350 get_random_bytes(&msgr
->inst
.addr
.nonce
, sizeof(msgr
->inst
.addr
.nonce
));
2351 encode_my_addr(msgr
);
2352 msgr
->nocrc
= nocrc
;
2354 dout("%s %p\n", __func__
, msgr
);
2356 EXPORT_SYMBOL(ceph_messenger_init
);
2358 static void clear_standby(struct ceph_connection
*con
)
2360 /* come back from STANDBY? */
2361 if (test_and_clear_bit(STANDBY
, &con
->state
)) {
2362 mutex_lock(&con
->mutex
);
2363 dout("clear_standby %p and ++connect_seq\n", con
);
2365 WARN_ON(test_bit(WRITE_PENDING
, &con
->flags
));
2366 WARN_ON(test_bit(KEEPALIVE_PENDING
, &con
->flags
));
2367 mutex_unlock(&con
->mutex
);
2372 * Queue up an outgoing message on the given connection.
2374 void ceph_con_send(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2376 if (test_bit(CLOSED
, &con
->state
)) {
2377 dout("con_send %p closed, dropping %p\n", con
, msg
);
2383 msg
->hdr
.src
= con
->msgr
->inst
.name
;
2385 BUG_ON(msg
->front
.iov_len
!= le32_to_cpu(msg
->hdr
.front_len
));
2387 msg
->needs_out_seq
= true;
2390 mutex_lock(&con
->mutex
);
2392 BUG_ON(msg
->con
!= NULL
);
2393 msg
->con
= con
->ops
->get(con
);
2394 BUG_ON(msg
->con
== NULL
);
2396 BUG_ON(!list_empty(&msg
->list_head
));
2397 list_add_tail(&msg
->list_head
, &con
->out_queue
);
2398 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg
,
2399 ENTITY_NAME(con
->peer_name
), le16_to_cpu(msg
->hdr
.type
),
2400 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2401 le32_to_cpu(msg
->hdr
.front_len
),
2402 le32_to_cpu(msg
->hdr
.middle_len
),
2403 le32_to_cpu(msg
->hdr
.data_len
));
2404 mutex_unlock(&con
->mutex
);
2406 /* if there wasn't anything waiting to send before, queue
2409 if (test_and_set_bit(WRITE_PENDING
, &con
->flags
) == 0)
2412 EXPORT_SYMBOL(ceph_con_send
);
2415 * Revoke a message that was previously queued for send
2417 void ceph_msg_revoke(struct ceph_msg
*msg
)
2419 struct ceph_connection
*con
= msg
->con
;
2422 return; /* Message not in our possession */
2424 mutex_lock(&con
->mutex
);
2425 if (!list_empty(&msg
->list_head
)) {
2426 dout("%s %p msg %p - was on queue\n", __func__
, con
, msg
);
2427 list_del_init(&msg
->list_head
);
2428 BUG_ON(msg
->con
== NULL
);
2429 msg
->con
->ops
->put(msg
->con
);
2435 if (con
->out_msg
== msg
) {
2436 dout("%s %p msg %p - was sending\n", __func__
, con
, msg
);
2437 con
->out_msg
= NULL
;
2438 if (con
->out_kvec_is_msg
) {
2439 con
->out_skip
= con
->out_kvec_bytes
;
2440 con
->out_kvec_is_msg
= false;
2446 mutex_unlock(&con
->mutex
);
2450 * Revoke a message that we may be reading data into
2452 void ceph_msg_revoke_incoming(struct ceph_msg
*msg
)
2454 struct ceph_connection
*con
;
2456 BUG_ON(msg
== NULL
);
2458 dout("%s msg %p null con\n", __func__
, msg
);
2460 return; /* Message not in our possession */
2464 mutex_lock(&con
->mutex
);
2465 if (con
->in_msg
== msg
) {
2466 unsigned int front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
2467 unsigned int middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
2468 unsigned int data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
2470 /* skip rest of message */
2471 dout("%s %p msg %p revoked\n", __func__
, con
, msg
);
2472 con
->in_base_pos
= con
->in_base_pos
-
2473 sizeof(struct ceph_msg_header
) -
2477 sizeof(struct ceph_msg_footer
);
2478 ceph_msg_put(con
->in_msg
);
2480 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2483 dout("%s %p in_msg %p msg %p no-op\n",
2484 __func__
, con
, con
->in_msg
, msg
);
2486 mutex_unlock(&con
->mutex
);
2490 * Queue a keepalive byte to ensure the tcp connection is alive.
2492 void ceph_con_keepalive(struct ceph_connection
*con
)
2494 dout("con_keepalive %p\n", con
);
2496 if (test_and_set_bit(KEEPALIVE_PENDING
, &con
->flags
) == 0 &&
2497 test_and_set_bit(WRITE_PENDING
, &con
->flags
) == 0)
2500 EXPORT_SYMBOL(ceph_con_keepalive
);
2504 * construct a new message with given type, size
2505 * the new msg has a ref count of 1.
2507 struct ceph_msg
*ceph_msg_new(int type
, int front_len
, gfp_t flags
,
2512 m
= kmalloc(sizeof(*m
), flags
);
2515 kref_init(&m
->kref
);
2518 INIT_LIST_HEAD(&m
->list_head
);
2521 m
->hdr
.type
= cpu_to_le16(type
);
2522 m
->hdr
.priority
= cpu_to_le16(CEPH_MSG_PRIO_DEFAULT
);
2524 m
->hdr
.front_len
= cpu_to_le32(front_len
);
2525 m
->hdr
.middle_len
= 0;
2526 m
->hdr
.data_len
= 0;
2527 m
->hdr
.data_off
= 0;
2528 m
->hdr
.reserved
= 0;
2529 m
->footer
.front_crc
= 0;
2530 m
->footer
.middle_crc
= 0;
2531 m
->footer
.data_crc
= 0;
2532 m
->footer
.flags
= 0;
2533 m
->front_max
= front_len
;
2534 m
->front_is_vmalloc
= false;
2535 m
->more_to_follow
= false;
2544 m
->page_alignment
= 0;
2554 if (front_len
> PAGE_CACHE_SIZE
) {
2555 m
->front
.iov_base
= __vmalloc(front_len
, flags
,
2557 m
->front_is_vmalloc
= true;
2559 m
->front
.iov_base
= kmalloc(front_len
, flags
);
2561 if (m
->front
.iov_base
== NULL
) {
2562 dout("ceph_msg_new can't allocate %d bytes\n",
2567 m
->front
.iov_base
= NULL
;
2569 m
->front
.iov_len
= front_len
;
2571 dout("ceph_msg_new %p front %d\n", m
, front_len
);
2578 pr_err("msg_new can't create type %d front %d\n", type
,
2582 dout("msg_new can't create type %d front %d\n", type
,
2587 EXPORT_SYMBOL(ceph_msg_new
);
2590 * Allocate "middle" portion of a message, if it is needed and wasn't
2591 * allocated by alloc_msg. This allows us to read a small fixed-size
2592 * per-type header in the front and then gracefully fail (i.e.,
2593 * propagate the error to the caller based on info in the front) when
2594 * the middle is too large.
2596 static int ceph_alloc_middle(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2598 int type
= le16_to_cpu(msg
->hdr
.type
);
2599 int middle_len
= le32_to_cpu(msg
->hdr
.middle_len
);
2601 dout("alloc_middle %p type %d %s middle_len %d\n", msg
, type
,
2602 ceph_msg_type_name(type
), middle_len
);
2603 BUG_ON(!middle_len
);
2604 BUG_ON(msg
->middle
);
2606 msg
->middle
= ceph_buffer_new(middle_len
, GFP_NOFS
);
2613 * Allocate a message for receiving an incoming message on a
2614 * connection, and save the result in con->in_msg. Uses the
2615 * connection's private alloc_msg op if available.
2617 * Returns true if the message should be skipped, false otherwise.
2618 * If true is returned (skip message), con->in_msg will be NULL.
2619 * If false is returned, con->in_msg will contain a pointer to the
2620 * newly-allocated message, or NULL in case of memory exhaustion.
2622 static bool ceph_con_in_msg_alloc(struct ceph_connection
*con
,
2623 struct ceph_msg_header
*hdr
)
2625 int type
= le16_to_cpu(hdr
->type
);
2626 int front_len
= le32_to_cpu(hdr
->front_len
);
2627 int middle_len
= le32_to_cpu(hdr
->middle_len
);
2630 BUG_ON(con
->in_msg
!= NULL
);
2632 if (con
->ops
->alloc_msg
) {
2635 mutex_unlock(&con
->mutex
);
2636 con
->in_msg
= con
->ops
->alloc_msg(con
, hdr
, &skip
);
2637 mutex_lock(&con
->mutex
);
2639 con
->in_msg
->con
= con
->ops
->get(con
);
2640 BUG_ON(con
->in_msg
->con
== NULL
);
2649 con
->in_msg
= ceph_msg_new(type
, front_len
, GFP_NOFS
, false);
2651 pr_err("unable to allocate msg type %d len %d\n",
2655 con
->in_msg
->con
= con
->ops
->get(con
);
2656 BUG_ON(con
->in_msg
->con
== NULL
);
2657 con
->in_msg
->page_alignment
= le16_to_cpu(hdr
->data_off
);
2659 memcpy(&con
->in_msg
->hdr
, &con
->in_hdr
, sizeof(con
->in_hdr
));
2661 if (middle_len
&& !con
->in_msg
->middle
) {
2662 ret
= ceph_alloc_middle(con
, con
->in_msg
);
2664 ceph_msg_put(con
->in_msg
);
2674 * Free a generically kmalloc'd message.
2676 void ceph_msg_kfree(struct ceph_msg
*m
)
2678 dout("msg_kfree %p\n", m
);
2679 if (m
->front_is_vmalloc
)
2680 vfree(m
->front
.iov_base
);
2682 kfree(m
->front
.iov_base
);
2687 * Drop a msg ref. Destroy as needed.
2689 void ceph_msg_last_put(struct kref
*kref
)
2691 struct ceph_msg
*m
= container_of(kref
, struct ceph_msg
, kref
);
2693 dout("ceph_msg_put last one on %p\n", m
);
2694 WARN_ON(!list_empty(&m
->list_head
));
2696 /* drop middle, data, if any */
2698 ceph_buffer_put(m
->middle
);
2705 ceph_pagelist_release(m
->pagelist
);
2713 ceph_msgpool_put(m
->pool
, m
);
2717 EXPORT_SYMBOL(ceph_msg_last_put
);
2719 void ceph_msg_dump(struct ceph_msg
*msg
)
2721 pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg
,
2722 msg
->front_max
, msg
->nr_pages
);
2723 print_hex_dump(KERN_DEBUG
, "header: ",
2724 DUMP_PREFIX_OFFSET
, 16, 1,
2725 &msg
->hdr
, sizeof(msg
->hdr
), true);
2726 print_hex_dump(KERN_DEBUG
, " front: ",
2727 DUMP_PREFIX_OFFSET
, 16, 1,
2728 msg
->front
.iov_base
, msg
->front
.iov_len
, true);
2730 print_hex_dump(KERN_DEBUG
, "middle: ",
2731 DUMP_PREFIX_OFFSET
, 16, 1,
2732 msg
->middle
->vec
.iov_base
,
2733 msg
->middle
->vec
.iov_len
, true);
2734 print_hex_dump(KERN_DEBUG
, "footer: ",
2735 DUMP_PREFIX_OFFSET
, 16, 1,
2736 &msg
->footer
, sizeof(msg
->footer
), true);
2738 EXPORT_SYMBOL(ceph_msg_dump
);