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>
16 #include <linux/ceph/libceph.h>
17 #include <linux/ceph/messenger.h>
18 #include <linux/ceph/decode.h>
19 #include <linux/ceph/pagelist.h>
22 * Ceph uses the messenger to exchange ceph_msg messages with other
23 * hosts in the system. The messenger provides ordered and reliable
24 * delivery. We tolerate TCP disconnects by reconnecting (with
25 * exponential backoff) in the case of a fault (disconnection, bad
26 * crc, protocol error). Acks allow sent messages to be discarded by
30 /* static tag bytes (protocol control messages) */
31 static char tag_msg
= CEPH_MSGR_TAG_MSG
;
32 static char tag_ack
= CEPH_MSGR_TAG_ACK
;
33 static char tag_keepalive
= CEPH_MSGR_TAG_KEEPALIVE
;
36 static struct lock_class_key socket_class
;
40 static void queue_con(struct ceph_connection
*con
);
41 static void con_work(struct work_struct
*);
42 static void ceph_fault(struct ceph_connection
*con
);
45 * nicely render a sockaddr as a string.
47 #define MAX_ADDR_STR 20
48 #define MAX_ADDR_STR_LEN 60
49 static char addr_str
[MAX_ADDR_STR
][MAX_ADDR_STR_LEN
];
50 static DEFINE_SPINLOCK(addr_str_lock
);
51 static int last_addr_str
;
53 const char *ceph_pr_addr(const struct sockaddr_storage
*ss
)
57 struct sockaddr_in
*in4
= (void *)ss
;
58 struct sockaddr_in6
*in6
= (void *)ss
;
60 spin_lock(&addr_str_lock
);
62 if (last_addr_str
== MAX_ADDR_STR
)
64 spin_unlock(&addr_str_lock
);
67 switch (ss
->ss_family
) {
69 snprintf(s
, MAX_ADDR_STR_LEN
, "%pI4:%u", &in4
->sin_addr
,
70 (unsigned int)ntohs(in4
->sin_port
));
74 snprintf(s
, MAX_ADDR_STR_LEN
, "[%pI6c]:%u", &in6
->sin6_addr
,
75 (unsigned int)ntohs(in6
->sin6_port
));
79 sprintf(s
, "(unknown sockaddr family %d)", (int)ss
->ss_family
);
84 EXPORT_SYMBOL(ceph_pr_addr
);
86 static void encode_my_addr(struct ceph_messenger
*msgr
)
88 memcpy(&msgr
->my_enc_addr
, &msgr
->inst
.addr
, sizeof(msgr
->my_enc_addr
));
89 ceph_encode_addr(&msgr
->my_enc_addr
);
93 * work queue for all reading and writing to/from the socket.
95 struct workqueue_struct
*ceph_msgr_wq
;
97 int ceph_msgr_init(void)
99 ceph_msgr_wq
= alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT
, 0);
101 pr_err("msgr_init failed to create workqueue\n");
106 EXPORT_SYMBOL(ceph_msgr_init
);
108 void ceph_msgr_exit(void)
110 destroy_workqueue(ceph_msgr_wq
);
112 EXPORT_SYMBOL(ceph_msgr_exit
);
114 void ceph_msgr_flush(void)
116 flush_workqueue(ceph_msgr_wq
);
118 EXPORT_SYMBOL(ceph_msgr_flush
);
122 * socket callback functions
125 /* data available on socket, or listen socket received a connect */
126 static void ceph_data_ready(struct sock
*sk
, int count_unused
)
128 struct ceph_connection
*con
=
129 (struct ceph_connection
*)sk
->sk_user_data
;
130 if (sk
->sk_state
!= TCP_CLOSE_WAIT
) {
131 dout("ceph_data_ready on %p state = %lu, queueing work\n",
137 /* socket has buffer space for writing */
138 static void ceph_write_space(struct sock
*sk
)
140 struct ceph_connection
*con
=
141 (struct ceph_connection
*)sk
->sk_user_data
;
143 /* only queue to workqueue if there is data we want to write. */
144 if (test_bit(WRITE_PENDING
, &con
->state
)) {
145 dout("ceph_write_space %p queueing write work\n", con
);
148 dout("ceph_write_space %p nothing to write\n", con
);
151 /* since we have our own write_space, clear the SOCK_NOSPACE flag */
152 clear_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
155 /* socket's state has changed */
156 static void ceph_state_change(struct sock
*sk
)
158 struct ceph_connection
*con
=
159 (struct ceph_connection
*)sk
->sk_user_data
;
161 dout("ceph_state_change %p state = %lu sk_state = %u\n",
162 con
, con
->state
, sk
->sk_state
);
164 if (test_bit(CLOSED
, &con
->state
))
167 switch (sk
->sk_state
) {
169 dout("ceph_state_change TCP_CLOSE\n");
171 dout("ceph_state_change TCP_CLOSE_WAIT\n");
172 if (test_and_set_bit(SOCK_CLOSED
, &con
->state
) == 0) {
173 if (test_bit(CONNECTING
, &con
->state
))
174 con
->error_msg
= "connection failed";
176 con
->error_msg
= "socket closed";
180 case TCP_ESTABLISHED
:
181 dout("ceph_state_change TCP_ESTABLISHED\n");
188 * set up socket callbacks
190 static void set_sock_callbacks(struct socket
*sock
,
191 struct ceph_connection
*con
)
193 struct sock
*sk
= sock
->sk
;
194 sk
->sk_user_data
= (void *)con
;
195 sk
->sk_data_ready
= ceph_data_ready
;
196 sk
->sk_write_space
= ceph_write_space
;
197 sk
->sk_state_change
= ceph_state_change
;
206 * initiate connection to a remote socket.
208 static struct socket
*ceph_tcp_connect(struct ceph_connection
*con
)
210 struct sockaddr_storage
*paddr
= &con
->peer_addr
.in_addr
;
215 ret
= sock_create_kern(con
->peer_addr
.in_addr
.ss_family
, SOCK_STREAM
,
220 sock
->sk
->sk_allocation
= GFP_NOFS
;
222 #ifdef CONFIG_LOCKDEP
223 lockdep_set_class(&sock
->sk
->sk_lock
, &socket_class
);
226 set_sock_callbacks(sock
, con
);
228 dout("connect %s\n", ceph_pr_addr(&con
->peer_addr
.in_addr
));
230 ret
= sock
->ops
->connect(sock
, (struct sockaddr
*)paddr
, sizeof(*paddr
),
232 if (ret
== -EINPROGRESS
) {
233 dout("connect %s EINPROGRESS sk_state = %u\n",
234 ceph_pr_addr(&con
->peer_addr
.in_addr
),
239 pr_err("connect %s error %d\n",
240 ceph_pr_addr(&con
->peer_addr
.in_addr
), ret
);
243 con
->error_msg
= "connect error";
251 static int ceph_tcp_recvmsg(struct socket
*sock
, void *buf
, size_t len
)
253 struct kvec iov
= {buf
, len
};
254 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
257 r
= kernel_recvmsg(sock
, &msg
, &iov
, 1, len
, msg
.msg_flags
);
264 * write something. @more is true if caller will be sending more data
267 static int ceph_tcp_sendmsg(struct socket
*sock
, struct kvec
*iov
,
268 size_t kvlen
, size_t len
, int more
)
270 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
274 msg
.msg_flags
|= MSG_MORE
;
276 msg
.msg_flags
|= MSG_EOR
; /* superfluous, but what the hell */
278 r
= kernel_sendmsg(sock
, &msg
, iov
, kvlen
, len
);
286 * Shutdown/close the socket for the given connection.
288 static int con_close_socket(struct ceph_connection
*con
)
292 dout("con_close_socket on %p sock %p\n", con
, con
->sock
);
295 set_bit(SOCK_CLOSED
, &con
->state
);
296 rc
= con
->sock
->ops
->shutdown(con
->sock
, SHUT_RDWR
);
297 sock_release(con
->sock
);
299 clear_bit(SOCK_CLOSED
, &con
->state
);
304 * Reset a connection. Discard all incoming and outgoing messages
305 * and clear *_seq state.
307 static void ceph_msg_remove(struct ceph_msg
*msg
)
309 list_del_init(&msg
->list_head
);
312 static void ceph_msg_remove_list(struct list_head
*head
)
314 while (!list_empty(head
)) {
315 struct ceph_msg
*msg
= list_first_entry(head
, struct ceph_msg
,
317 ceph_msg_remove(msg
);
321 static void reset_connection(struct ceph_connection
*con
)
323 /* reset connection, out_queue, msg_ and connect_seq */
324 /* discard existing out_queue and msg_seq */
325 ceph_msg_remove_list(&con
->out_queue
);
326 ceph_msg_remove_list(&con
->out_sent
);
329 ceph_msg_put(con
->in_msg
);
333 con
->connect_seq
= 0;
336 ceph_msg_put(con
->out_msg
);
339 con
->out_keepalive_pending
= false;
341 con
->in_seq_acked
= 0;
345 * mark a peer down. drop any open connections.
347 void ceph_con_close(struct ceph_connection
*con
)
349 dout("con_close %p peer %s\n", con
,
350 ceph_pr_addr(&con
->peer_addr
.in_addr
));
351 set_bit(CLOSED
, &con
->state
); /* in case there's queued work */
352 clear_bit(STANDBY
, &con
->state
); /* avoid connect_seq bump */
353 clear_bit(LOSSYTX
, &con
->state
); /* so we retry next connect */
354 clear_bit(KEEPALIVE_PENDING
, &con
->state
);
355 clear_bit(WRITE_PENDING
, &con
->state
);
356 mutex_lock(&con
->mutex
);
357 reset_connection(con
);
358 con
->peer_global_seq
= 0;
359 cancel_delayed_work(&con
->work
);
360 mutex_unlock(&con
->mutex
);
363 EXPORT_SYMBOL(ceph_con_close
);
366 * Reopen a closed connection, with a new peer address.
368 void ceph_con_open(struct ceph_connection
*con
, struct ceph_entity_addr
*addr
)
370 dout("con_open %p %s\n", con
, ceph_pr_addr(&addr
->in_addr
));
371 set_bit(OPENING
, &con
->state
);
372 clear_bit(CLOSED
, &con
->state
);
373 memcpy(&con
->peer_addr
, addr
, sizeof(*addr
));
374 con
->delay
= 0; /* reset backoff memory */
377 EXPORT_SYMBOL(ceph_con_open
);
380 * return true if this connection ever successfully opened
382 bool ceph_con_opened(struct ceph_connection
*con
)
384 return con
->connect_seq
> 0;
390 struct ceph_connection
*ceph_con_get(struct ceph_connection
*con
)
392 dout("con_get %p nref = %d -> %d\n", con
,
393 atomic_read(&con
->nref
), atomic_read(&con
->nref
) + 1);
394 if (atomic_inc_not_zero(&con
->nref
))
399 void ceph_con_put(struct ceph_connection
*con
)
401 dout("con_put %p nref = %d -> %d\n", con
,
402 atomic_read(&con
->nref
), atomic_read(&con
->nref
) - 1);
403 BUG_ON(atomic_read(&con
->nref
) == 0);
404 if (atomic_dec_and_test(&con
->nref
)) {
411 * initialize a new connection.
413 void ceph_con_init(struct ceph_messenger
*msgr
, struct ceph_connection
*con
)
415 dout("con_init %p\n", con
);
416 memset(con
, 0, sizeof(*con
));
417 atomic_set(&con
->nref
, 1);
419 mutex_init(&con
->mutex
);
420 INIT_LIST_HEAD(&con
->out_queue
);
421 INIT_LIST_HEAD(&con
->out_sent
);
422 INIT_DELAYED_WORK(&con
->work
, con_work
);
424 EXPORT_SYMBOL(ceph_con_init
);
428 * We maintain a global counter to order connection attempts. Get
429 * a unique seq greater than @gt.
431 static u32
get_global_seq(struct ceph_messenger
*msgr
, u32 gt
)
435 spin_lock(&msgr
->global_seq_lock
);
436 if (msgr
->global_seq
< gt
)
437 msgr
->global_seq
= gt
;
438 ret
= ++msgr
->global_seq
;
439 spin_unlock(&msgr
->global_seq_lock
);
445 * Prepare footer for currently outgoing message, and finish things
446 * off. Assumes out_kvec* are already valid.. we just add on to the end.
448 static void prepare_write_message_footer(struct ceph_connection
*con
, int v
)
450 struct ceph_msg
*m
= con
->out_msg
;
452 dout("prepare_write_message_footer %p\n", con
);
453 con
->out_kvec_is_msg
= true;
454 con
->out_kvec
[v
].iov_base
= &m
->footer
;
455 con
->out_kvec
[v
].iov_len
= sizeof(m
->footer
);
456 con
->out_kvec_bytes
+= sizeof(m
->footer
);
457 con
->out_kvec_left
++;
458 con
->out_more
= m
->more_to_follow
;
459 con
->out_msg_done
= true;
463 * Prepare headers for the next outgoing message.
465 static void prepare_write_message(struct ceph_connection
*con
)
470 con
->out_kvec_bytes
= 0;
471 con
->out_kvec_is_msg
= true;
472 con
->out_msg_done
= false;
474 /* Sneak an ack in there first? If we can get it into the same
475 * TCP packet that's a good thing. */
476 if (con
->in_seq
> con
->in_seq_acked
) {
477 con
->in_seq_acked
= con
->in_seq
;
478 con
->out_kvec
[v
].iov_base
= &tag_ack
;
479 con
->out_kvec
[v
++].iov_len
= 1;
480 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
481 con
->out_kvec
[v
].iov_base
= &con
->out_temp_ack
;
482 con
->out_kvec
[v
++].iov_len
= sizeof(con
->out_temp_ack
);
483 con
->out_kvec_bytes
= 1 + sizeof(con
->out_temp_ack
);
486 m
= list_first_entry(&con
->out_queue
,
487 struct ceph_msg
, list_head
);
489 if (test_bit(LOSSYTX
, &con
->state
)) {
490 list_del_init(&m
->list_head
);
492 /* put message on sent list */
494 list_move_tail(&m
->list_head
, &con
->out_sent
);
498 * only assign outgoing seq # if we haven't sent this message
499 * yet. if it is requeued, resend with it's original seq.
501 if (m
->needs_out_seq
) {
502 m
->hdr
.seq
= cpu_to_le64(++con
->out_seq
);
503 m
->needs_out_seq
= false;
506 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n",
507 m
, con
->out_seq
, le16_to_cpu(m
->hdr
.type
),
508 le32_to_cpu(m
->hdr
.front_len
), le32_to_cpu(m
->hdr
.middle_len
),
509 le32_to_cpu(m
->hdr
.data_len
),
511 BUG_ON(le32_to_cpu(m
->hdr
.front_len
) != m
->front
.iov_len
);
513 /* tag + hdr + front + middle */
514 con
->out_kvec
[v
].iov_base
= &tag_msg
;
515 con
->out_kvec
[v
++].iov_len
= 1;
516 con
->out_kvec
[v
].iov_base
= &m
->hdr
;
517 con
->out_kvec
[v
++].iov_len
= sizeof(m
->hdr
);
518 con
->out_kvec
[v
++] = m
->front
;
520 con
->out_kvec
[v
++] = m
->middle
->vec
;
521 con
->out_kvec_left
= v
;
522 con
->out_kvec_bytes
+= 1 + sizeof(m
->hdr
) + m
->front
.iov_len
+
523 (m
->middle
? m
->middle
->vec
.iov_len
: 0);
524 con
->out_kvec_cur
= con
->out_kvec
;
526 /* fill in crc (except data pages), footer */
527 con
->out_msg
->hdr
.crc
=
528 cpu_to_le32(crc32c(0, (void *)&m
->hdr
,
529 sizeof(m
->hdr
) - sizeof(m
->hdr
.crc
)));
530 con
->out_msg
->footer
.flags
= CEPH_MSG_FOOTER_COMPLETE
;
531 con
->out_msg
->footer
.front_crc
=
532 cpu_to_le32(crc32c(0, m
->front
.iov_base
, m
->front
.iov_len
));
534 con
->out_msg
->footer
.middle_crc
=
535 cpu_to_le32(crc32c(0, m
->middle
->vec
.iov_base
,
536 m
->middle
->vec
.iov_len
));
538 con
->out_msg
->footer
.middle_crc
= 0;
539 con
->out_msg
->footer
.data_crc
= 0;
540 dout("prepare_write_message front_crc %u data_crc %u\n",
541 le32_to_cpu(con
->out_msg
->footer
.front_crc
),
542 le32_to_cpu(con
->out_msg
->footer
.middle_crc
));
544 /* is there a data payload? */
545 if (le32_to_cpu(m
->hdr
.data_len
) > 0) {
546 /* initialize page iterator */
547 con
->out_msg_pos
.page
= 0;
549 con
->out_msg_pos
.page_pos
= m
->page_alignment
;
551 con
->out_msg_pos
.page_pos
= 0;
552 con
->out_msg_pos
.data_pos
= 0;
553 con
->out_msg_pos
.did_page_crc
= 0;
554 con
->out_more
= 1; /* data + footer will follow */
556 /* no, queue up footer too and be done */
557 prepare_write_message_footer(con
, v
);
560 set_bit(WRITE_PENDING
, &con
->state
);
566 static void prepare_write_ack(struct ceph_connection
*con
)
568 dout("prepare_write_ack %p %llu -> %llu\n", con
,
569 con
->in_seq_acked
, con
->in_seq
);
570 con
->in_seq_acked
= con
->in_seq
;
572 con
->out_kvec
[0].iov_base
= &tag_ack
;
573 con
->out_kvec
[0].iov_len
= 1;
574 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
575 con
->out_kvec
[1].iov_base
= &con
->out_temp_ack
;
576 con
->out_kvec
[1].iov_len
= sizeof(con
->out_temp_ack
);
577 con
->out_kvec_left
= 2;
578 con
->out_kvec_bytes
= 1 + sizeof(con
->out_temp_ack
);
579 con
->out_kvec_cur
= con
->out_kvec
;
580 con
->out_more
= 1; /* more will follow.. eventually.. */
581 set_bit(WRITE_PENDING
, &con
->state
);
585 * Prepare to write keepalive byte.
587 static void prepare_write_keepalive(struct ceph_connection
*con
)
589 dout("prepare_write_keepalive %p\n", con
);
590 con
->out_kvec
[0].iov_base
= &tag_keepalive
;
591 con
->out_kvec
[0].iov_len
= 1;
592 con
->out_kvec_left
= 1;
593 con
->out_kvec_bytes
= 1;
594 con
->out_kvec_cur
= con
->out_kvec
;
595 set_bit(WRITE_PENDING
, &con
->state
);
599 * Connection negotiation.
602 static void prepare_connect_authorizer(struct ceph_connection
*con
)
606 int auth_protocol
= 0;
608 mutex_unlock(&con
->mutex
);
609 if (con
->ops
->get_authorizer
)
610 con
->ops
->get_authorizer(con
, &auth_buf
, &auth_len
,
611 &auth_protocol
, &con
->auth_reply_buf
,
612 &con
->auth_reply_buf_len
,
614 mutex_lock(&con
->mutex
);
616 con
->out_connect
.authorizer_protocol
= cpu_to_le32(auth_protocol
);
617 con
->out_connect
.authorizer_len
= cpu_to_le32(auth_len
);
619 con
->out_kvec
[con
->out_kvec_left
].iov_base
= auth_buf
;
620 con
->out_kvec
[con
->out_kvec_left
].iov_len
= auth_len
;
621 con
->out_kvec_left
++;
622 con
->out_kvec_bytes
+= auth_len
;
626 * We connected to a peer and are saying hello.
628 static void prepare_write_banner(struct ceph_messenger
*msgr
,
629 struct ceph_connection
*con
)
631 int len
= strlen(CEPH_BANNER
);
633 con
->out_kvec
[0].iov_base
= CEPH_BANNER
;
634 con
->out_kvec
[0].iov_len
= len
;
635 con
->out_kvec
[1].iov_base
= &msgr
->my_enc_addr
;
636 con
->out_kvec
[1].iov_len
= sizeof(msgr
->my_enc_addr
);
637 con
->out_kvec_left
= 2;
638 con
->out_kvec_bytes
= len
+ sizeof(msgr
->my_enc_addr
);
639 con
->out_kvec_cur
= con
->out_kvec
;
641 set_bit(WRITE_PENDING
, &con
->state
);
644 static void prepare_write_connect(struct ceph_messenger
*msgr
,
645 struct ceph_connection
*con
,
648 unsigned global_seq
= get_global_seq(con
->msgr
, 0);
651 switch (con
->peer_name
.type
) {
652 case CEPH_ENTITY_TYPE_MON
:
653 proto
= CEPH_MONC_PROTOCOL
;
655 case CEPH_ENTITY_TYPE_OSD
:
656 proto
= CEPH_OSDC_PROTOCOL
;
658 case CEPH_ENTITY_TYPE_MDS
:
659 proto
= CEPH_MDSC_PROTOCOL
;
665 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con
,
666 con
->connect_seq
, global_seq
, proto
);
668 con
->out_connect
.features
= cpu_to_le64(msgr
->supported_features
);
669 con
->out_connect
.host_type
= cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT
);
670 con
->out_connect
.connect_seq
= cpu_to_le32(con
->connect_seq
);
671 con
->out_connect
.global_seq
= cpu_to_le32(global_seq
);
672 con
->out_connect
.protocol_version
= cpu_to_le32(proto
);
673 con
->out_connect
.flags
= 0;
676 con
->out_kvec_left
= 0;
677 con
->out_kvec_bytes
= 0;
679 con
->out_kvec
[con
->out_kvec_left
].iov_base
= &con
->out_connect
;
680 con
->out_kvec
[con
->out_kvec_left
].iov_len
= sizeof(con
->out_connect
);
681 con
->out_kvec_left
++;
682 con
->out_kvec_bytes
+= sizeof(con
->out_connect
);
683 con
->out_kvec_cur
= con
->out_kvec
;
685 set_bit(WRITE_PENDING
, &con
->state
);
687 prepare_connect_authorizer(con
);
692 * write as much of pending kvecs to the socket as we can.
694 * 0 -> socket full, but more to do
697 static int write_partial_kvec(struct ceph_connection
*con
)
701 dout("write_partial_kvec %p %d left\n", con
, con
->out_kvec_bytes
);
702 while (con
->out_kvec_bytes
> 0) {
703 ret
= ceph_tcp_sendmsg(con
->sock
, con
->out_kvec_cur
,
704 con
->out_kvec_left
, con
->out_kvec_bytes
,
708 con
->out_kvec_bytes
-= ret
;
709 if (con
->out_kvec_bytes
== 0)
712 if (ret
>= con
->out_kvec_cur
->iov_len
) {
713 ret
-= con
->out_kvec_cur
->iov_len
;
715 con
->out_kvec_left
--;
717 con
->out_kvec_cur
->iov_len
-= ret
;
718 con
->out_kvec_cur
->iov_base
+= ret
;
724 con
->out_kvec_left
= 0;
725 con
->out_kvec_is_msg
= false;
728 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con
,
729 con
->out_kvec_bytes
, con
->out_kvec_left
, ret
);
730 return ret
; /* done! */
734 static void init_bio_iter(struct bio
*bio
, struct bio
**iter
, int *seg
)
745 static void iter_bio_next(struct bio
**bio_iter
, int *seg
)
747 if (*bio_iter
== NULL
)
750 BUG_ON(*seg
>= (*bio_iter
)->bi_vcnt
);
753 if (*seg
== (*bio_iter
)->bi_vcnt
)
754 init_bio_iter((*bio_iter
)->bi_next
, bio_iter
, seg
);
759 * Write as much message data payload as we can. If we finish, queue
761 * 1 -> done, footer is now queued in out_kvec[].
762 * 0 -> socket full, but more to do
765 static int write_partial_msg_pages(struct ceph_connection
*con
)
767 struct ceph_msg
*msg
= con
->out_msg
;
768 unsigned data_len
= le32_to_cpu(msg
->hdr
.data_len
);
770 int crc
= con
->msgr
->nocrc
;
774 size_t trail_len
= (msg
->trail
? msg
->trail
->length
: 0);
776 dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
777 con
, con
->out_msg
, con
->out_msg_pos
.page
, con
->out_msg
->nr_pages
,
778 con
->out_msg_pos
.page_pos
);
781 if (msg
->bio
&& !msg
->bio_iter
)
782 init_bio_iter(msg
->bio
, &msg
->bio_iter
, &msg
->bio_seg
);
785 while (data_len
> con
->out_msg_pos
.data_pos
) {
786 struct page
*page
= NULL
;
788 int max_write
= PAGE_SIZE
;
791 total_max_write
= data_len
- trail_len
-
792 con
->out_msg_pos
.data_pos
;
795 * if we are calculating the data crc (the default), we need
796 * to map the page. if our pages[] has been revoked, use the
800 /* have we reached the trail part of the data? */
801 if (con
->out_msg_pos
.data_pos
>= data_len
- trail_len
) {
804 total_max_write
= data_len
- con
->out_msg_pos
.data_pos
;
806 page
= list_first_entry(&msg
->trail
->head
,
810 max_write
= PAGE_SIZE
;
811 } else if (msg
->pages
) {
812 page
= msg
->pages
[con
->out_msg_pos
.page
];
815 } else if (msg
->pagelist
) {
816 page
= list_first_entry(&msg
->pagelist
->head
,
821 } else if (msg
->bio
) {
824 bv
= bio_iovec_idx(msg
->bio_iter
, msg
->bio_seg
);
826 page_shift
= bv
->bv_offset
;
828 kaddr
= kmap(page
) + page_shift
;
829 max_write
= bv
->bv_len
;
832 page
= con
->msgr
->zero_page
;
834 kaddr
= page_address(con
->msgr
->zero_page
);
836 len
= min_t(int, max_write
- con
->out_msg_pos
.page_pos
,
839 if (crc
&& !con
->out_msg_pos
.did_page_crc
) {
840 void *base
= kaddr
+ con
->out_msg_pos
.page_pos
;
841 u32 tmpcrc
= le32_to_cpu(con
->out_msg
->footer
.data_crc
);
843 BUG_ON(kaddr
== NULL
);
844 con
->out_msg
->footer
.data_crc
=
845 cpu_to_le32(crc32c(tmpcrc
, base
, len
));
846 con
->out_msg_pos
.did_page_crc
= 1;
848 ret
= kernel_sendpage(con
->sock
, page
,
849 con
->out_msg_pos
.page_pos
+ page_shift
,
851 MSG_DONTWAIT
| MSG_NOSIGNAL
|
855 (msg
->pages
|| msg
->pagelist
|| msg
->bio
|| in_trail
))
863 con
->out_msg_pos
.data_pos
+= ret
;
864 con
->out_msg_pos
.page_pos
+= ret
;
866 con
->out_msg_pos
.page_pos
= 0;
867 con
->out_msg_pos
.page
++;
868 con
->out_msg_pos
.did_page_crc
= 0;
870 list_move_tail(&page
->lru
,
872 else if (msg
->pagelist
)
873 list_move_tail(&page
->lru
,
874 &msg
->pagelist
->head
);
877 iter_bio_next(&msg
->bio_iter
, &msg
->bio_seg
);
882 dout("write_partial_msg_pages %p msg %p done\n", con
, msg
);
884 /* prepare and queue up footer, too */
886 con
->out_msg
->footer
.flags
|= CEPH_MSG_FOOTER_NOCRC
;
887 con
->out_kvec_bytes
= 0;
888 con
->out_kvec_left
= 0;
889 con
->out_kvec_cur
= con
->out_kvec
;
890 prepare_write_message_footer(con
, 0);
899 static int write_partial_skip(struct ceph_connection
*con
)
903 while (con
->out_skip
> 0) {
905 .iov_base
= page_address(con
->msgr
->zero_page
),
906 .iov_len
= min(con
->out_skip
, (int)PAGE_CACHE_SIZE
)
909 ret
= ceph_tcp_sendmsg(con
->sock
, &iov
, 1, iov
.iov_len
, 1);
912 con
->out_skip
-= ret
;
920 * Prepare to read connection handshake, or an ack.
922 static void prepare_read_banner(struct ceph_connection
*con
)
924 dout("prepare_read_banner %p\n", con
);
925 con
->in_base_pos
= 0;
928 static void prepare_read_connect(struct ceph_connection
*con
)
930 dout("prepare_read_connect %p\n", con
);
931 con
->in_base_pos
= 0;
934 static void prepare_read_ack(struct ceph_connection
*con
)
936 dout("prepare_read_ack %p\n", con
);
937 con
->in_base_pos
= 0;
940 static void prepare_read_tag(struct ceph_connection
*con
)
942 dout("prepare_read_tag %p\n", con
);
943 con
->in_base_pos
= 0;
944 con
->in_tag
= CEPH_MSGR_TAG_READY
;
948 * Prepare to read a message.
950 static int prepare_read_message(struct ceph_connection
*con
)
952 dout("prepare_read_message %p\n", con
);
953 BUG_ON(con
->in_msg
!= NULL
);
954 con
->in_base_pos
= 0;
955 con
->in_front_crc
= con
->in_middle_crc
= con
->in_data_crc
= 0;
960 static int read_partial(struct ceph_connection
*con
,
961 int *to
, int size
, void *object
)
964 while (con
->in_base_pos
< *to
) {
965 int left
= *to
- con
->in_base_pos
;
966 int have
= size
- left
;
967 int ret
= ceph_tcp_recvmsg(con
->sock
, object
+ have
, left
);
970 con
->in_base_pos
+= ret
;
977 * Read all or part of the connect-side handshake on a new connection
979 static int read_partial_banner(struct ceph_connection
*con
)
983 dout("read_partial_banner %p at %d\n", con
, con
->in_base_pos
);
986 ret
= read_partial(con
, &to
, strlen(CEPH_BANNER
), con
->in_banner
);
989 ret
= read_partial(con
, &to
, sizeof(con
->actual_peer_addr
),
990 &con
->actual_peer_addr
);
993 ret
= read_partial(con
, &to
, sizeof(con
->peer_addr_for_me
),
994 &con
->peer_addr_for_me
);
1001 static int read_partial_connect(struct ceph_connection
*con
)
1005 dout("read_partial_connect %p at %d\n", con
, con
->in_base_pos
);
1007 ret
= read_partial(con
, &to
, sizeof(con
->in_reply
), &con
->in_reply
);
1010 ret
= read_partial(con
, &to
, le32_to_cpu(con
->in_reply
.authorizer_len
),
1011 con
->auth_reply_buf
);
1015 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1016 con
, (int)con
->in_reply
.tag
,
1017 le32_to_cpu(con
->in_reply
.connect_seq
),
1018 le32_to_cpu(con
->in_reply
.global_seq
));
1025 * Verify the hello banner looks okay.
1027 static int verify_hello(struct ceph_connection
*con
)
1029 if (memcmp(con
->in_banner
, CEPH_BANNER
, strlen(CEPH_BANNER
))) {
1030 pr_err("connect to %s got bad banner\n",
1031 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1032 con
->error_msg
= "protocol error, bad banner";
1038 static bool addr_is_blank(struct sockaddr_storage
*ss
)
1040 switch (ss
->ss_family
) {
1042 return ((struct sockaddr_in
*)ss
)->sin_addr
.s_addr
== 0;
1045 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[0] == 0 &&
1046 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[1] == 0 &&
1047 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[2] == 0 &&
1048 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[3] == 0;
1053 static int addr_port(struct sockaddr_storage
*ss
)
1055 switch (ss
->ss_family
) {
1057 return ntohs(((struct sockaddr_in
*)ss
)->sin_port
);
1059 return ntohs(((struct sockaddr_in6
*)ss
)->sin6_port
);
1064 static void addr_set_port(struct sockaddr_storage
*ss
, int p
)
1066 switch (ss
->ss_family
) {
1068 ((struct sockaddr_in
*)ss
)->sin_port
= htons(p
);
1070 ((struct sockaddr_in6
*)ss
)->sin6_port
= htons(p
);
1075 * Parse an ip[:port] list into an addr array. Use the default
1076 * monitor port if a port isn't specified.
1078 int ceph_parse_ips(const char *c
, const char *end
,
1079 struct ceph_entity_addr
*addr
,
1080 int max_count
, int *count
)
1085 dout("parse_ips on '%.*s'\n", (int)(end
-c
), c
);
1086 for (i
= 0; i
< max_count
; i
++) {
1088 struct sockaddr_storage
*ss
= &addr
[i
].in_addr
;
1089 struct sockaddr_in
*in4
= (void *)ss
;
1090 struct sockaddr_in6
*in6
= (void *)ss
;
1099 memset(ss
, 0, sizeof(*ss
));
1100 if (in4_pton(p
, end
- p
, (u8
*)&in4
->sin_addr
.s_addr
,
1102 ss
->ss_family
= AF_INET
;
1103 else if (in6_pton(p
, end
- p
, (u8
*)&in6
->sin6_addr
.s6_addr
,
1105 ss
->ss_family
= AF_INET6
;
1112 dout("missing matching ']'\n");
1119 if (p
< end
&& *p
== ':') {
1122 while (p
< end
&& *p
>= '0' && *p
<= '9') {
1123 port
= (port
* 10) + (*p
- '0');
1126 if (port
> 65535 || port
== 0)
1129 port
= CEPH_MON_PORT
;
1132 addr_set_port(ss
, port
);
1134 dout("parse_ips got %s\n", ceph_pr_addr(ss
));
1151 pr_err("parse_ips bad ip '%.*s'\n", (int)(end
- c
), c
);
1154 EXPORT_SYMBOL(ceph_parse_ips
);
1156 static int process_banner(struct ceph_connection
*con
)
1158 dout("process_banner on %p\n", con
);
1160 if (verify_hello(con
) < 0)
1163 ceph_decode_addr(&con
->actual_peer_addr
);
1164 ceph_decode_addr(&con
->peer_addr_for_me
);
1167 * Make sure the other end is who we wanted. note that the other
1168 * end may not yet know their ip address, so if it's 0.0.0.0, give
1169 * them the benefit of the doubt.
1171 if (memcmp(&con
->peer_addr
, &con
->actual_peer_addr
,
1172 sizeof(con
->peer_addr
)) != 0 &&
1173 !(addr_is_blank(&con
->actual_peer_addr
.in_addr
) &&
1174 con
->actual_peer_addr
.nonce
== con
->peer_addr
.nonce
)) {
1175 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1176 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1177 (int)le32_to_cpu(con
->peer_addr
.nonce
),
1178 ceph_pr_addr(&con
->actual_peer_addr
.in_addr
),
1179 (int)le32_to_cpu(con
->actual_peer_addr
.nonce
));
1180 con
->error_msg
= "wrong peer at address";
1185 * did we learn our address?
1187 if (addr_is_blank(&con
->msgr
->inst
.addr
.in_addr
)) {
1188 int port
= addr_port(&con
->msgr
->inst
.addr
.in_addr
);
1190 memcpy(&con
->msgr
->inst
.addr
.in_addr
,
1191 &con
->peer_addr_for_me
.in_addr
,
1192 sizeof(con
->peer_addr_for_me
.in_addr
));
1193 addr_set_port(&con
->msgr
->inst
.addr
.in_addr
, port
);
1194 encode_my_addr(con
->msgr
);
1195 dout("process_banner learned my addr is %s\n",
1196 ceph_pr_addr(&con
->msgr
->inst
.addr
.in_addr
));
1199 set_bit(NEGOTIATING
, &con
->state
);
1200 prepare_read_connect(con
);
1204 static void fail_protocol(struct ceph_connection
*con
)
1206 reset_connection(con
);
1207 set_bit(CLOSED
, &con
->state
); /* in case there's queued work */
1209 mutex_unlock(&con
->mutex
);
1210 if (con
->ops
->bad_proto
)
1211 con
->ops
->bad_proto(con
);
1212 mutex_lock(&con
->mutex
);
1215 static int process_connect(struct ceph_connection
*con
)
1217 u64 sup_feat
= con
->msgr
->supported_features
;
1218 u64 req_feat
= con
->msgr
->required_features
;
1219 u64 server_feat
= le64_to_cpu(con
->in_reply
.features
);
1221 dout("process_connect on %p tag %d\n", con
, (int)con
->in_tag
);
1223 switch (con
->in_reply
.tag
) {
1224 case CEPH_MSGR_TAG_FEATURES
:
1225 pr_err("%s%lld %s feature set mismatch,"
1226 " my %llx < server's %llx, missing %llx\n",
1227 ENTITY_NAME(con
->peer_name
),
1228 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1229 sup_feat
, server_feat
, server_feat
& ~sup_feat
);
1230 con
->error_msg
= "missing required protocol features";
1234 case CEPH_MSGR_TAG_BADPROTOVER
:
1235 pr_err("%s%lld %s protocol version mismatch,"
1236 " my %d != server's %d\n",
1237 ENTITY_NAME(con
->peer_name
),
1238 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1239 le32_to_cpu(con
->out_connect
.protocol_version
),
1240 le32_to_cpu(con
->in_reply
.protocol_version
));
1241 con
->error_msg
= "protocol version mismatch";
1245 case CEPH_MSGR_TAG_BADAUTHORIZER
:
1247 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con
,
1249 if (con
->auth_retry
== 2) {
1250 con
->error_msg
= "connect authorization failure";
1251 reset_connection(con
);
1252 set_bit(CLOSED
, &con
->state
);
1255 con
->auth_retry
= 1;
1256 prepare_write_connect(con
->msgr
, con
, 0);
1257 prepare_read_connect(con
);
1260 case CEPH_MSGR_TAG_RESETSESSION
:
1262 * If we connected with a large connect_seq but the peer
1263 * has no record of a session with us (no connection, or
1264 * connect_seq == 0), they will send RESETSESION to indicate
1265 * that they must have reset their session, and may have
1268 dout("process_connect got RESET peer seq %u\n",
1269 le32_to_cpu(con
->in_connect
.connect_seq
));
1270 pr_err("%s%lld %s connection reset\n",
1271 ENTITY_NAME(con
->peer_name
),
1272 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1273 reset_connection(con
);
1274 prepare_write_connect(con
->msgr
, con
, 0);
1275 prepare_read_connect(con
);
1277 /* Tell ceph about it. */
1278 mutex_unlock(&con
->mutex
);
1279 pr_info("reset on %s%lld\n", ENTITY_NAME(con
->peer_name
));
1280 if (con
->ops
->peer_reset
)
1281 con
->ops
->peer_reset(con
);
1282 mutex_lock(&con
->mutex
);
1285 case CEPH_MSGR_TAG_RETRY_SESSION
:
1287 * If we sent a smaller connect_seq than the peer has, try
1288 * again with a larger value.
1290 dout("process_connect got RETRY my seq = %u, peer_seq = %u\n",
1291 le32_to_cpu(con
->out_connect
.connect_seq
),
1292 le32_to_cpu(con
->in_connect
.connect_seq
));
1293 con
->connect_seq
= le32_to_cpu(con
->in_connect
.connect_seq
);
1294 prepare_write_connect(con
->msgr
, con
, 0);
1295 prepare_read_connect(con
);
1298 case CEPH_MSGR_TAG_RETRY_GLOBAL
:
1300 * If we sent a smaller global_seq than the peer has, try
1301 * again with a larger value.
1303 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1304 con
->peer_global_seq
,
1305 le32_to_cpu(con
->in_connect
.global_seq
));
1306 get_global_seq(con
->msgr
,
1307 le32_to_cpu(con
->in_connect
.global_seq
));
1308 prepare_write_connect(con
->msgr
, con
, 0);
1309 prepare_read_connect(con
);
1312 case CEPH_MSGR_TAG_READY
:
1313 if (req_feat
& ~server_feat
) {
1314 pr_err("%s%lld %s protocol feature mismatch,"
1315 " my required %llx > server's %llx, need %llx\n",
1316 ENTITY_NAME(con
->peer_name
),
1317 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1318 req_feat
, server_feat
, req_feat
& ~server_feat
);
1319 con
->error_msg
= "missing required protocol features";
1323 clear_bit(CONNECTING
, &con
->state
);
1324 con
->peer_global_seq
= le32_to_cpu(con
->in_reply
.global_seq
);
1326 con
->peer_features
= server_feat
;
1327 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1328 con
->peer_global_seq
,
1329 le32_to_cpu(con
->in_reply
.connect_seq
),
1331 WARN_ON(con
->connect_seq
!=
1332 le32_to_cpu(con
->in_reply
.connect_seq
));
1334 if (con
->in_reply
.flags
& CEPH_MSG_CONNECT_LOSSY
)
1335 set_bit(LOSSYTX
, &con
->state
);
1337 prepare_read_tag(con
);
1340 case CEPH_MSGR_TAG_WAIT
:
1342 * If there is a connection race (we are opening
1343 * connections to each other), one of us may just have
1344 * to WAIT. This shouldn't happen if we are the
1347 pr_err("process_connect peer connecting WAIT\n");
1350 pr_err("connect protocol error, will retry\n");
1351 con
->error_msg
= "protocol error, garbage tag during connect";
1359 * read (part of) an ack
1361 static int read_partial_ack(struct ceph_connection
*con
)
1365 return read_partial(con
, &to
, sizeof(con
->in_temp_ack
),
1371 * We can finally discard anything that's been acked.
1373 static void process_ack(struct ceph_connection
*con
)
1376 u64 ack
= le64_to_cpu(con
->in_temp_ack
);
1379 while (!list_empty(&con
->out_sent
)) {
1380 m
= list_first_entry(&con
->out_sent
, struct ceph_msg
,
1382 seq
= le64_to_cpu(m
->hdr
.seq
);
1385 dout("got ack for seq %llu type %d at %p\n", seq
,
1386 le16_to_cpu(m
->hdr
.type
), m
);
1389 prepare_read_tag(con
);
1395 static int read_partial_message_section(struct ceph_connection
*con
,
1396 struct kvec
*section
,
1397 unsigned int sec_len
, u32
*crc
)
1403 while (section
->iov_len
< sec_len
) {
1404 BUG_ON(section
->iov_base
== NULL
);
1405 left
= sec_len
- section
->iov_len
;
1406 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)section
->iov_base
+
1407 section
->iov_len
, left
);
1410 section
->iov_len
+= ret
;
1411 if (section
->iov_len
== sec_len
)
1412 *crc
= crc32c(0, section
->iov_base
,
1419 static struct ceph_msg
*ceph_alloc_msg(struct ceph_connection
*con
,
1420 struct ceph_msg_header
*hdr
,
1424 static int read_partial_message_pages(struct ceph_connection
*con
,
1425 struct page
**pages
,
1426 unsigned data_len
, int datacrc
)
1432 left
= min((int)(data_len
- con
->in_msg_pos
.data_pos
),
1433 (int)(PAGE_SIZE
- con
->in_msg_pos
.page_pos
));
1435 BUG_ON(pages
== NULL
);
1436 p
= kmap(pages
[con
->in_msg_pos
.page
]);
1437 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ con
->in_msg_pos
.page_pos
,
1439 if (ret
> 0 && datacrc
)
1441 crc32c(con
->in_data_crc
,
1442 p
+ con
->in_msg_pos
.page_pos
, ret
);
1443 kunmap(pages
[con
->in_msg_pos
.page
]);
1446 con
->in_msg_pos
.data_pos
+= ret
;
1447 con
->in_msg_pos
.page_pos
+= ret
;
1448 if (con
->in_msg_pos
.page_pos
== PAGE_SIZE
) {
1449 con
->in_msg_pos
.page_pos
= 0;
1450 con
->in_msg_pos
.page
++;
1457 static int read_partial_message_bio(struct ceph_connection
*con
,
1458 struct bio
**bio_iter
, int *bio_seg
,
1459 unsigned data_len
, int datacrc
)
1461 struct bio_vec
*bv
= bio_iovec_idx(*bio_iter
, *bio_seg
);
1468 left
= min((int)(data_len
- con
->in_msg_pos
.data_pos
),
1469 (int)(bv
->bv_len
- con
->in_msg_pos
.page_pos
));
1471 p
= kmap(bv
->bv_page
) + bv
->bv_offset
;
1473 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ con
->in_msg_pos
.page_pos
,
1475 if (ret
> 0 && datacrc
)
1477 crc32c(con
->in_data_crc
,
1478 p
+ con
->in_msg_pos
.page_pos
, ret
);
1479 kunmap(bv
->bv_page
);
1482 con
->in_msg_pos
.data_pos
+= ret
;
1483 con
->in_msg_pos
.page_pos
+= ret
;
1484 if (con
->in_msg_pos
.page_pos
== bv
->bv_len
) {
1485 con
->in_msg_pos
.page_pos
= 0;
1486 iter_bio_next(bio_iter
, bio_seg
);
1494 * read (part of) a message.
1496 static int read_partial_message(struct ceph_connection
*con
)
1498 struct ceph_msg
*m
= con
->in_msg
;
1501 unsigned front_len
, middle_len
, data_len
;
1502 int datacrc
= con
->msgr
->nocrc
;
1506 dout("read_partial_message con %p msg %p\n", con
, m
);
1509 while (con
->in_base_pos
< sizeof(con
->in_hdr
)) {
1510 left
= sizeof(con
->in_hdr
) - con
->in_base_pos
;
1511 ret
= ceph_tcp_recvmsg(con
->sock
,
1512 (char *)&con
->in_hdr
+ con
->in_base_pos
,
1516 con
->in_base_pos
+= ret
;
1517 if (con
->in_base_pos
== sizeof(con
->in_hdr
)) {
1518 u32 crc
= crc32c(0, (void *)&con
->in_hdr
,
1519 sizeof(con
->in_hdr
) - sizeof(con
->in_hdr
.crc
));
1520 if (crc
!= le32_to_cpu(con
->in_hdr
.crc
)) {
1521 pr_err("read_partial_message bad hdr "
1522 " crc %u != expected %u\n",
1523 crc
, con
->in_hdr
.crc
);
1528 front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
1529 if (front_len
> CEPH_MSG_MAX_FRONT_LEN
)
1531 middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
1532 if (middle_len
> CEPH_MSG_MAX_DATA_LEN
)
1534 data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
1535 if (data_len
> CEPH_MSG_MAX_DATA_LEN
)
1539 seq
= le64_to_cpu(con
->in_hdr
.seq
);
1540 if ((s64
)seq
- (s64
)con
->in_seq
< 1) {
1541 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1542 ENTITY_NAME(con
->peer_name
),
1543 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1544 seq
, con
->in_seq
+ 1);
1545 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1547 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1549 } else if ((s64
)seq
- (s64
)con
->in_seq
> 1) {
1550 pr_err("read_partial_message bad seq %lld expected %lld\n",
1551 seq
, con
->in_seq
+ 1);
1552 con
->error_msg
= "bad message sequence # for incoming message";
1556 /* allocate message? */
1558 dout("got hdr type %d front %d data %d\n", con
->in_hdr
.type
,
1559 con
->in_hdr
.front_len
, con
->in_hdr
.data_len
);
1561 con
->in_msg
= ceph_alloc_msg(con
, &con
->in_hdr
, &skip
);
1563 /* skip this message */
1564 dout("alloc_msg said skip message\n");
1565 BUG_ON(con
->in_msg
);
1566 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1568 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1574 "error allocating memory for incoming message";
1578 m
->front
.iov_len
= 0; /* haven't read it yet */
1580 m
->middle
->vec
.iov_len
= 0;
1582 con
->in_msg_pos
.page
= 0;
1584 con
->in_msg_pos
.page_pos
= m
->page_alignment
;
1586 con
->in_msg_pos
.page_pos
= 0;
1587 con
->in_msg_pos
.data_pos
= 0;
1591 ret
= read_partial_message_section(con
, &m
->front
, front_len
,
1592 &con
->in_front_crc
);
1598 ret
= read_partial_message_section(con
, &m
->middle
->vec
,
1600 &con
->in_middle_crc
);
1605 if (m
->bio
&& !m
->bio_iter
)
1606 init_bio_iter(m
->bio
, &m
->bio_iter
, &m
->bio_seg
);
1610 while (con
->in_msg_pos
.data_pos
< data_len
) {
1612 ret
= read_partial_message_pages(con
, m
->pages
,
1617 } else if (m
->bio
) {
1619 ret
= read_partial_message_bio(con
,
1620 &m
->bio_iter
, &m
->bio_seg
,
1631 to
= sizeof(m
->hdr
) + sizeof(m
->footer
);
1632 while (con
->in_base_pos
< to
) {
1633 left
= to
- con
->in_base_pos
;
1634 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)&m
->footer
+
1635 (con
->in_base_pos
- sizeof(m
->hdr
)),
1639 con
->in_base_pos
+= ret
;
1641 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1642 m
, front_len
, m
->footer
.front_crc
, middle_len
,
1643 m
->footer
.middle_crc
, data_len
, m
->footer
.data_crc
);
1646 if (con
->in_front_crc
!= le32_to_cpu(m
->footer
.front_crc
)) {
1647 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1648 m
, con
->in_front_crc
, m
->footer
.front_crc
);
1651 if (con
->in_middle_crc
!= le32_to_cpu(m
->footer
.middle_crc
)) {
1652 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1653 m
, con
->in_middle_crc
, m
->footer
.middle_crc
);
1657 (m
->footer
.flags
& CEPH_MSG_FOOTER_NOCRC
) == 0 &&
1658 con
->in_data_crc
!= le32_to_cpu(m
->footer
.data_crc
)) {
1659 pr_err("read_partial_message %p data crc %u != exp. %u\n", m
,
1660 con
->in_data_crc
, le32_to_cpu(m
->footer
.data_crc
));
1664 return 1; /* done! */
1668 * Process message. This happens in the worker thread. The callback should
1669 * be careful not to do anything that waits on other incoming messages or it
1672 static void process_message(struct ceph_connection
*con
)
1674 struct ceph_msg
*msg
;
1679 /* if first message, set peer_name */
1680 if (con
->peer_name
.type
== 0)
1681 con
->peer_name
= msg
->hdr
.src
;
1684 mutex_unlock(&con
->mutex
);
1686 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
1687 msg
, le64_to_cpu(msg
->hdr
.seq
),
1688 ENTITY_NAME(msg
->hdr
.src
),
1689 le16_to_cpu(msg
->hdr
.type
),
1690 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
1691 le32_to_cpu(msg
->hdr
.front_len
),
1692 le32_to_cpu(msg
->hdr
.data_len
),
1693 con
->in_front_crc
, con
->in_middle_crc
, con
->in_data_crc
);
1694 con
->ops
->dispatch(con
, msg
);
1696 mutex_lock(&con
->mutex
);
1697 prepare_read_tag(con
);
1702 * Write something to the socket. Called in a worker thread when the
1703 * socket appears to be writeable and we have something ready to send.
1705 static int try_write(struct ceph_connection
*con
)
1707 struct ceph_messenger
*msgr
= con
->msgr
;
1710 dout("try_write start %p state %lu nref %d\n", con
, con
->state
,
1711 atomic_read(&con
->nref
));
1714 dout("try_write out_kvec_bytes %d\n", con
->out_kvec_bytes
);
1716 /* open the socket first? */
1717 if (con
->sock
== NULL
) {
1719 * if we were STANDBY and are reconnecting _this_
1720 * connection, bump connect_seq now. Always bump
1723 if (test_and_clear_bit(STANDBY
, &con
->state
))
1726 prepare_write_banner(msgr
, con
);
1727 prepare_write_connect(msgr
, con
, 1);
1728 prepare_read_banner(con
);
1729 set_bit(CONNECTING
, &con
->state
);
1730 clear_bit(NEGOTIATING
, &con
->state
);
1732 BUG_ON(con
->in_msg
);
1733 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1734 dout("try_write initiating connect on %p new state %lu\n",
1736 con
->sock
= ceph_tcp_connect(con
);
1737 if (IS_ERR(con
->sock
)) {
1739 con
->error_msg
= "connect error";
1746 /* kvec data queued? */
1747 if (con
->out_skip
) {
1748 ret
= write_partial_skip(con
);
1752 if (con
->out_kvec_left
) {
1753 ret
= write_partial_kvec(con
);
1760 if (con
->out_msg_done
) {
1761 ceph_msg_put(con
->out_msg
);
1762 con
->out_msg
= NULL
; /* we're done with this one */
1766 ret
= write_partial_msg_pages(con
);
1768 goto more_kvec
; /* we need to send the footer, too! */
1772 dout("try_write write_partial_msg_pages err %d\n",
1779 if (!test_bit(CONNECTING
, &con
->state
)) {
1780 /* is anything else pending? */
1781 if (!list_empty(&con
->out_queue
)) {
1782 prepare_write_message(con
);
1785 if (con
->in_seq
> con
->in_seq_acked
) {
1786 prepare_write_ack(con
);
1789 if (test_and_clear_bit(KEEPALIVE_PENDING
, &con
->state
)) {
1790 prepare_write_keepalive(con
);
1795 /* Nothing to do! */
1796 clear_bit(WRITE_PENDING
, &con
->state
);
1797 dout("try_write nothing else to write.\n");
1800 dout("try_write done on %p ret %d\n", con
, ret
);
1807 * Read what we can from the socket.
1809 static int try_read(struct ceph_connection
*con
)
1816 if (test_bit(STANDBY
, &con
->state
))
1819 dout("try_read start on %p\n", con
);
1822 dout("try_read tag %d in_base_pos %d\n", (int)con
->in_tag
,
1824 if (test_bit(CONNECTING
, &con
->state
)) {
1825 if (!test_bit(NEGOTIATING
, &con
->state
)) {
1826 dout("try_read connecting\n");
1827 ret
= read_partial_banner(con
);
1830 ret
= process_banner(con
);
1834 ret
= read_partial_connect(con
);
1837 ret
= process_connect(con
);
1843 if (con
->in_base_pos
< 0) {
1845 * skipping + discarding content.
1847 * FIXME: there must be a better way to do this!
1849 static char buf
[1024];
1850 int skip
= min(1024, -con
->in_base_pos
);
1851 dout("skipping %d / %d bytes\n", skip
, -con
->in_base_pos
);
1852 ret
= ceph_tcp_recvmsg(con
->sock
, buf
, skip
);
1855 con
->in_base_pos
+= ret
;
1856 if (con
->in_base_pos
)
1859 if (con
->in_tag
== CEPH_MSGR_TAG_READY
) {
1863 ret
= ceph_tcp_recvmsg(con
->sock
, &con
->in_tag
, 1);
1866 dout("try_read got tag %d\n", (int)con
->in_tag
);
1867 switch (con
->in_tag
) {
1868 case CEPH_MSGR_TAG_MSG
:
1869 prepare_read_message(con
);
1871 case CEPH_MSGR_TAG_ACK
:
1872 prepare_read_ack(con
);
1874 case CEPH_MSGR_TAG_CLOSE
:
1875 set_bit(CLOSED
, &con
->state
); /* fixme */
1881 if (con
->in_tag
== CEPH_MSGR_TAG_MSG
) {
1882 ret
= read_partial_message(con
);
1886 con
->error_msg
= "bad crc";
1890 con
->error_msg
= "io error";
1895 if (con
->in_tag
== CEPH_MSGR_TAG_READY
)
1897 process_message(con
);
1900 if (con
->in_tag
== CEPH_MSGR_TAG_ACK
) {
1901 ret
= read_partial_ack(con
);
1909 dout("try_read done on %p ret %d\n", con
, ret
);
1913 pr_err("try_read bad con->in_tag = %d\n", (int)con
->in_tag
);
1914 con
->error_msg
= "protocol error, garbage tag";
1921 * Atomically queue work on a connection. Bump @con reference to
1922 * avoid races with connection teardown.
1924 static void queue_con(struct ceph_connection
*con
)
1926 if (test_bit(DEAD
, &con
->state
)) {
1927 dout("queue_con %p ignoring: DEAD\n",
1932 if (!con
->ops
->get(con
)) {
1933 dout("queue_con %p ref count 0\n", con
);
1937 if (!queue_delayed_work(ceph_msgr_wq
, &con
->work
, 0)) {
1938 dout("queue_con %p - already queued\n", con
);
1941 dout("queue_con %p\n", con
);
1946 * Do some work on a connection. Drop a connection ref when we're done.
1948 static void con_work(struct work_struct
*work
)
1950 struct ceph_connection
*con
= container_of(work
, struct ceph_connection
,
1953 mutex_lock(&con
->mutex
);
1955 if (test_bit(CLOSED
, &con
->state
)) { /* e.g. if we are replaced */
1956 dout("con_work CLOSED\n");
1957 con_close_socket(con
);
1960 if (test_and_clear_bit(OPENING
, &con
->state
)) {
1961 /* reopen w/ new peer */
1962 dout("con_work OPENING\n");
1963 con_close_socket(con
);
1966 if (test_and_clear_bit(SOCK_CLOSED
, &con
->state
) ||
1967 try_read(con
) < 0 ||
1968 try_write(con
) < 0) {
1969 mutex_unlock(&con
->mutex
);
1970 ceph_fault(con
); /* error/fault path */
1975 mutex_unlock(&con
->mutex
);
1982 * Generic error/fault handler. A retry mechanism is used with
1983 * exponential backoff
1985 static void ceph_fault(struct ceph_connection
*con
)
1987 pr_err("%s%lld %s %s\n", ENTITY_NAME(con
->peer_name
),
1988 ceph_pr_addr(&con
->peer_addr
.in_addr
), con
->error_msg
);
1989 dout("fault %p state %lu to peer %s\n",
1990 con
, con
->state
, ceph_pr_addr(&con
->peer_addr
.in_addr
));
1992 if (test_bit(LOSSYTX
, &con
->state
)) {
1993 dout("fault on LOSSYTX channel\n");
1997 mutex_lock(&con
->mutex
);
1998 if (test_bit(CLOSED
, &con
->state
))
2001 con_close_socket(con
);
2004 ceph_msg_put(con
->in_msg
);
2008 /* Requeue anything that hasn't been acked */
2009 list_splice_init(&con
->out_sent
, &con
->out_queue
);
2011 /* If there are no messages in the queue, place the connection
2012 * in a STANDBY state (i.e., don't try to reconnect just yet). */
2013 if (list_empty(&con
->out_queue
) && !con
->out_keepalive_pending
) {
2014 dout("fault setting STANDBY\n");
2015 set_bit(STANDBY
, &con
->state
);
2017 /* retry after a delay. */
2018 if (con
->delay
== 0)
2019 con
->delay
= BASE_DELAY_INTERVAL
;
2020 else if (con
->delay
< MAX_DELAY_INTERVAL
)
2022 dout("fault queueing %p delay %lu\n", con
, con
->delay
);
2024 if (queue_delayed_work(ceph_msgr_wq
, &con
->work
,
2025 round_jiffies_relative(con
->delay
)) == 0)
2030 mutex_unlock(&con
->mutex
);
2033 * in case we faulted due to authentication, invalidate our
2034 * current tickets so that we can get new ones.
2036 if (con
->auth_retry
&& con
->ops
->invalidate_authorizer
) {
2037 dout("calling invalidate_authorizer()\n");
2038 con
->ops
->invalidate_authorizer(con
);
2041 if (con
->ops
->fault
)
2042 con
->ops
->fault(con
);
2048 * create a new messenger instance
2050 struct ceph_messenger
*ceph_messenger_create(struct ceph_entity_addr
*myaddr
,
2051 u32 supported_features
,
2052 u32 required_features
)
2054 struct ceph_messenger
*msgr
;
2056 msgr
= kzalloc(sizeof(*msgr
), GFP_KERNEL
);
2058 return ERR_PTR(-ENOMEM
);
2060 msgr
->supported_features
= supported_features
;
2061 msgr
->required_features
= required_features
;
2063 spin_lock_init(&msgr
->global_seq_lock
);
2065 /* the zero page is needed if a request is "canceled" while the message
2066 * is being written over the socket */
2067 msgr
->zero_page
= __page_cache_alloc(GFP_KERNEL
| __GFP_ZERO
);
2068 if (!msgr
->zero_page
) {
2070 return ERR_PTR(-ENOMEM
);
2072 kmap(msgr
->zero_page
);
2075 msgr
->inst
.addr
= *myaddr
;
2077 /* select a random nonce */
2078 msgr
->inst
.addr
.type
= 0;
2079 get_random_bytes(&msgr
->inst
.addr
.nonce
, sizeof(msgr
->inst
.addr
.nonce
));
2080 encode_my_addr(msgr
);
2082 dout("messenger_create %p\n", msgr
);
2085 EXPORT_SYMBOL(ceph_messenger_create
);
2087 void ceph_messenger_destroy(struct ceph_messenger
*msgr
)
2089 dout("destroy %p\n", msgr
);
2090 kunmap(msgr
->zero_page
);
2091 __free_page(msgr
->zero_page
);
2093 dout("destroyed messenger %p\n", msgr
);
2095 EXPORT_SYMBOL(ceph_messenger_destroy
);
2098 * Queue up an outgoing message on the given connection.
2100 void ceph_con_send(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2102 if (test_bit(CLOSED
, &con
->state
)) {
2103 dout("con_send %p closed, dropping %p\n", con
, msg
);
2109 msg
->hdr
.src
= con
->msgr
->inst
.name
;
2111 BUG_ON(msg
->front
.iov_len
!= le32_to_cpu(msg
->hdr
.front_len
));
2113 msg
->needs_out_seq
= true;
2116 mutex_lock(&con
->mutex
);
2117 BUG_ON(!list_empty(&msg
->list_head
));
2118 list_add_tail(&msg
->list_head
, &con
->out_queue
);
2119 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg
,
2120 ENTITY_NAME(con
->peer_name
), le16_to_cpu(msg
->hdr
.type
),
2121 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2122 le32_to_cpu(msg
->hdr
.front_len
),
2123 le32_to_cpu(msg
->hdr
.middle_len
),
2124 le32_to_cpu(msg
->hdr
.data_len
));
2125 mutex_unlock(&con
->mutex
);
2127 /* if there wasn't anything waiting to send before, queue
2129 if (test_and_set_bit(WRITE_PENDING
, &con
->state
) == 0)
2132 EXPORT_SYMBOL(ceph_con_send
);
2135 * Revoke a message that was previously queued for send
2137 void ceph_con_revoke(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2139 mutex_lock(&con
->mutex
);
2140 if (!list_empty(&msg
->list_head
)) {
2141 dout("con_revoke %p msg %p - was on queue\n", con
, msg
);
2142 list_del_init(&msg
->list_head
);
2146 if (con
->out_msg
== msg
) {
2147 dout("con_revoke %p msg %p - was sending\n", con
, msg
);
2148 con
->out_msg
= NULL
;
2149 if (con
->out_kvec_is_msg
) {
2150 con
->out_skip
= con
->out_kvec_bytes
;
2151 con
->out_kvec_is_msg
= false;
2156 mutex_unlock(&con
->mutex
);
2160 * Revoke a message that we may be reading data into
2162 void ceph_con_revoke_message(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2164 mutex_lock(&con
->mutex
);
2165 if (con
->in_msg
&& con
->in_msg
== msg
) {
2166 unsigned front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
2167 unsigned middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
2168 unsigned data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
2170 /* skip rest of message */
2171 dout("con_revoke_pages %p msg %p revoked\n", con
, msg
);
2172 con
->in_base_pos
= con
->in_base_pos
-
2173 sizeof(struct ceph_msg_header
) -
2177 sizeof(struct ceph_msg_footer
);
2178 ceph_msg_put(con
->in_msg
);
2180 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2183 dout("con_revoke_pages %p msg %p pages %p no-op\n",
2184 con
, con
->in_msg
, msg
);
2186 mutex_unlock(&con
->mutex
);
2190 * Queue a keepalive byte to ensure the tcp connection is alive.
2192 void ceph_con_keepalive(struct ceph_connection
*con
)
2194 if (test_and_set_bit(KEEPALIVE_PENDING
, &con
->state
) == 0 &&
2195 test_and_set_bit(WRITE_PENDING
, &con
->state
) == 0)
2198 EXPORT_SYMBOL(ceph_con_keepalive
);
2202 * construct a new message with given type, size
2203 * the new msg has a ref count of 1.
2205 struct ceph_msg
*ceph_msg_new(int type
, int front_len
, gfp_t flags
)
2209 m
= kmalloc(sizeof(*m
), flags
);
2212 kref_init(&m
->kref
);
2213 INIT_LIST_HEAD(&m
->list_head
);
2216 m
->hdr
.type
= cpu_to_le16(type
);
2217 m
->hdr
.priority
= cpu_to_le16(CEPH_MSG_PRIO_DEFAULT
);
2219 m
->hdr
.front_len
= cpu_to_le32(front_len
);
2220 m
->hdr
.middle_len
= 0;
2221 m
->hdr
.data_len
= 0;
2222 m
->hdr
.data_off
= 0;
2223 m
->hdr
.reserved
= 0;
2224 m
->footer
.front_crc
= 0;
2225 m
->footer
.middle_crc
= 0;
2226 m
->footer
.data_crc
= 0;
2227 m
->footer
.flags
= 0;
2228 m
->front_max
= front_len
;
2229 m
->front_is_vmalloc
= false;
2230 m
->more_to_follow
= false;
2235 if (front_len
> PAGE_CACHE_SIZE
) {
2236 m
->front
.iov_base
= __vmalloc(front_len
, flags
,
2238 m
->front_is_vmalloc
= true;
2240 m
->front
.iov_base
= kmalloc(front_len
, flags
);
2242 if (m
->front
.iov_base
== NULL
) {
2243 pr_err("msg_new can't allocate %d bytes\n",
2248 m
->front
.iov_base
= NULL
;
2250 m
->front
.iov_len
= front_len
;
2257 m
->page_alignment
= 0;
2265 dout("ceph_msg_new %p front %d\n", m
, front_len
);
2271 pr_err("msg_new can't create type %d front %d\n", type
, front_len
);
2274 EXPORT_SYMBOL(ceph_msg_new
);
2277 * Allocate "middle" portion of a message, if it is needed and wasn't
2278 * allocated by alloc_msg. This allows us to read a small fixed-size
2279 * per-type header in the front and then gracefully fail (i.e.,
2280 * propagate the error to the caller based on info in the front) when
2281 * the middle is too large.
2283 static int ceph_alloc_middle(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2285 int type
= le16_to_cpu(msg
->hdr
.type
);
2286 int middle_len
= le32_to_cpu(msg
->hdr
.middle_len
);
2288 dout("alloc_middle %p type %d %s middle_len %d\n", msg
, type
,
2289 ceph_msg_type_name(type
), middle_len
);
2290 BUG_ON(!middle_len
);
2291 BUG_ON(msg
->middle
);
2293 msg
->middle
= ceph_buffer_new(middle_len
, GFP_NOFS
);
2300 * Generic message allocator, for incoming messages.
2302 static struct ceph_msg
*ceph_alloc_msg(struct ceph_connection
*con
,
2303 struct ceph_msg_header
*hdr
,
2306 int type
= le16_to_cpu(hdr
->type
);
2307 int front_len
= le32_to_cpu(hdr
->front_len
);
2308 int middle_len
= le32_to_cpu(hdr
->middle_len
);
2309 struct ceph_msg
*msg
= NULL
;
2312 if (con
->ops
->alloc_msg
) {
2313 mutex_unlock(&con
->mutex
);
2314 msg
= con
->ops
->alloc_msg(con
, hdr
, skip
);
2315 mutex_lock(&con
->mutex
);
2321 msg
= ceph_msg_new(type
, front_len
, GFP_NOFS
);
2323 pr_err("unable to allocate msg type %d len %d\n",
2327 msg
->page_alignment
= le16_to_cpu(hdr
->data_off
);
2329 memcpy(&msg
->hdr
, &con
->in_hdr
, sizeof(con
->in_hdr
));
2331 if (middle_len
&& !msg
->middle
) {
2332 ret
= ceph_alloc_middle(con
, msg
);
2344 * Free a generically kmalloc'd message.
2346 void ceph_msg_kfree(struct ceph_msg
*m
)
2348 dout("msg_kfree %p\n", m
);
2349 if (m
->front_is_vmalloc
)
2350 vfree(m
->front
.iov_base
);
2352 kfree(m
->front
.iov_base
);
2357 * Drop a msg ref. Destroy as needed.
2359 void ceph_msg_last_put(struct kref
*kref
)
2361 struct ceph_msg
*m
= container_of(kref
, struct ceph_msg
, kref
);
2363 dout("ceph_msg_put last one on %p\n", m
);
2364 WARN_ON(!list_empty(&m
->list_head
));
2366 /* drop middle, data, if any */
2368 ceph_buffer_put(m
->middle
);
2375 ceph_pagelist_release(m
->pagelist
);
2383 ceph_msgpool_put(m
->pool
, m
);
2387 EXPORT_SYMBOL(ceph_msg_last_put
);
2389 void ceph_msg_dump(struct ceph_msg
*msg
)
2391 pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg
,
2392 msg
->front_max
, msg
->nr_pages
);
2393 print_hex_dump(KERN_DEBUG
, "header: ",
2394 DUMP_PREFIX_OFFSET
, 16, 1,
2395 &msg
->hdr
, sizeof(msg
->hdr
), true);
2396 print_hex_dump(KERN_DEBUG
, " front: ",
2397 DUMP_PREFIX_OFFSET
, 16, 1,
2398 msg
->front
.iov_base
, msg
->front
.iov_len
, true);
2400 print_hex_dump(KERN_DEBUG
, "middle: ",
2401 DUMP_PREFIX_OFFSET
, 16, 1,
2402 msg
->middle
->vec
.iov_base
,
2403 msg
->middle
->vec
.iov_len
, true);
2404 print_hex_dump(KERN_DEBUG
, "footer: ",
2405 DUMP_PREFIX_OFFSET
, 16, 1,
2406 &msg
->footer
, sizeof(msg
->footer
), true);
2408 EXPORT_SYMBOL(ceph_msg_dump
);