1 #include "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/socket.h>
10 #include <linux/string.h>
14 #include "messenger.h"
18 * Ceph uses the messenger to exchange ceph_msg messages with other
19 * hosts in the system. The messenger provides ordered and reliable
20 * delivery. We tolerate TCP disconnects by reconnecting (with
21 * exponential backoff) in the case of a fault (disconnection, bad
22 * crc, protocol error). Acks allow sent messages to be discarded by
26 /* static tag bytes (protocol control messages) */
27 static char tag_msg
= CEPH_MSGR_TAG_MSG
;
28 static char tag_ack
= CEPH_MSGR_TAG_ACK
;
29 static char tag_keepalive
= CEPH_MSGR_TAG_KEEPALIVE
;
32 static void queue_con(struct ceph_connection
*con
);
33 static void con_work(struct work_struct
*);
34 static void ceph_fault(struct ceph_connection
*con
);
36 const char *ceph_name_type_str(int t
)
39 case CEPH_ENTITY_TYPE_MON
: return "mon";
40 case CEPH_ENTITY_TYPE_MDS
: return "mds";
41 case CEPH_ENTITY_TYPE_OSD
: return "osd";
42 case CEPH_ENTITY_TYPE_CLIENT
: return "client";
43 case CEPH_ENTITY_TYPE_ADMIN
: return "admin";
44 default: return "???";
49 * nicely render a sockaddr as a string.
51 #define MAX_ADDR_STR 20
52 static char addr_str
[MAX_ADDR_STR
][40];
53 static DEFINE_SPINLOCK(addr_str_lock
);
54 static int last_addr_str
;
56 const char *pr_addr(const struct sockaddr_storage
*ss
)
60 struct sockaddr_in
*in4
= (void *)ss
;
61 unsigned char *quad
= (void *)&in4
->sin_addr
.s_addr
;
62 struct sockaddr_in6
*in6
= (void *)ss
;
64 spin_lock(&addr_str_lock
);
66 if (last_addr_str
== MAX_ADDR_STR
)
68 spin_unlock(&addr_str_lock
);
71 switch (ss
->ss_family
) {
73 sprintf(s
, "%u.%u.%u.%u:%u",
74 (unsigned int)quad
[0],
75 (unsigned int)quad
[1],
76 (unsigned int)quad
[2],
77 (unsigned int)quad
[3],
78 (unsigned int)ntohs(in4
->sin_port
));
82 sprintf(s
, "%04x:%04x:%04x:%04x:%04x:%04x:%04x:%04x:%u",
83 in6
->sin6_addr
.s6_addr16
[0],
84 in6
->sin6_addr
.s6_addr16
[1],
85 in6
->sin6_addr
.s6_addr16
[2],
86 in6
->sin6_addr
.s6_addr16
[3],
87 in6
->sin6_addr
.s6_addr16
[4],
88 in6
->sin6_addr
.s6_addr16
[5],
89 in6
->sin6_addr
.s6_addr16
[6],
90 in6
->sin6_addr
.s6_addr16
[7],
91 (unsigned int)ntohs(in6
->sin6_port
));
95 sprintf(s
, "(unknown sockaddr family %d)", (int)ss
->ss_family
);
101 static void encode_my_addr(struct ceph_messenger
*msgr
)
103 memcpy(&msgr
->my_enc_addr
, &msgr
->inst
.addr
, sizeof(msgr
->my_enc_addr
));
104 ceph_encode_addr(&msgr
->my_enc_addr
);
108 * work queue for all reading and writing to/from the socket.
110 struct workqueue_struct
*ceph_msgr_wq
;
112 int __init
ceph_msgr_init(void)
114 ceph_msgr_wq
= create_workqueue("ceph-msgr");
115 if (IS_ERR(ceph_msgr_wq
)) {
116 int ret
= PTR_ERR(ceph_msgr_wq
);
117 pr_err("msgr_init failed to create workqueue: %d\n", ret
);
124 void ceph_msgr_exit(void)
126 destroy_workqueue(ceph_msgr_wq
);
130 * socket callback functions
133 /* data available on socket, or listen socket received a connect */
134 static void ceph_data_ready(struct sock
*sk
, int count_unused
)
136 struct ceph_connection
*con
=
137 (struct ceph_connection
*)sk
->sk_user_data
;
138 if (sk
->sk_state
!= TCP_CLOSE_WAIT
) {
139 dout("ceph_data_ready on %p state = %lu, queueing work\n",
145 /* socket has buffer space for writing */
146 static void ceph_write_space(struct sock
*sk
)
148 struct ceph_connection
*con
=
149 (struct ceph_connection
*)sk
->sk_user_data
;
151 /* only queue to workqueue if there is data we want to write. */
152 if (test_bit(WRITE_PENDING
, &con
->state
)) {
153 dout("ceph_write_space %p queueing write work\n", con
);
156 dout("ceph_write_space %p nothing to write\n", con
);
159 /* since we have our own write_space, clear the SOCK_NOSPACE flag */
160 clear_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
163 /* socket's state has changed */
164 static void ceph_state_change(struct sock
*sk
)
166 struct ceph_connection
*con
=
167 (struct ceph_connection
*)sk
->sk_user_data
;
169 dout("ceph_state_change %p state = %lu sk_state = %u\n",
170 con
, con
->state
, sk
->sk_state
);
172 if (test_bit(CLOSED
, &con
->state
))
175 switch (sk
->sk_state
) {
177 dout("ceph_state_change TCP_CLOSE\n");
179 dout("ceph_state_change TCP_CLOSE_WAIT\n");
180 if (test_and_set_bit(SOCK_CLOSED
, &con
->state
) == 0) {
181 if (test_bit(CONNECTING
, &con
->state
))
182 con
->error_msg
= "connection failed";
184 con
->error_msg
= "socket closed";
188 case TCP_ESTABLISHED
:
189 dout("ceph_state_change TCP_ESTABLISHED\n");
196 * set up socket callbacks
198 static void set_sock_callbacks(struct socket
*sock
,
199 struct ceph_connection
*con
)
201 struct sock
*sk
= sock
->sk
;
202 sk
->sk_user_data
= (void *)con
;
203 sk
->sk_data_ready
= ceph_data_ready
;
204 sk
->sk_write_space
= ceph_write_space
;
205 sk
->sk_state_change
= ceph_state_change
;
214 * initiate connection to a remote socket.
216 static struct socket
*ceph_tcp_connect(struct ceph_connection
*con
)
218 struct sockaddr
*paddr
= (struct sockaddr
*)&con
->peer_addr
.in_addr
;
223 ret
= sock_create_kern(AF_INET
, SOCK_STREAM
, IPPROTO_TCP
, &sock
);
227 sock
->sk
->sk_allocation
= GFP_NOFS
;
229 set_sock_callbacks(sock
, con
);
231 dout("connect %s\n", pr_addr(&con
->peer_addr
.in_addr
));
233 ret
= sock
->ops
->connect(sock
, paddr
, sizeof(*paddr
), O_NONBLOCK
);
234 if (ret
== -EINPROGRESS
) {
235 dout("connect %s EINPROGRESS sk_state = %u\n",
236 pr_addr(&con
->peer_addr
.in_addr
),
241 pr_err("connect %s error %d\n",
242 pr_addr(&con
->peer_addr
.in_addr
), ret
);
245 con
->error_msg
= "connect error";
253 static int ceph_tcp_recvmsg(struct socket
*sock
, void *buf
, size_t len
)
255 struct kvec iov
= {buf
, len
};
256 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
258 return kernel_recvmsg(sock
, &msg
, &iov
, 1, len
, msg
.msg_flags
);
262 * write something. @more is true if caller will be sending more data
265 static int ceph_tcp_sendmsg(struct socket
*sock
, struct kvec
*iov
,
266 size_t kvlen
, size_t len
, int more
)
268 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
271 msg
.msg_flags
|= MSG_MORE
;
273 msg
.msg_flags
|= MSG_EOR
; /* superfluous, but what the hell */
275 return kernel_sendmsg(sock
, &msg
, iov
, kvlen
, len
);
280 * Shutdown/close the socket for the given connection.
282 static int con_close_socket(struct ceph_connection
*con
)
286 dout("con_close_socket on %p sock %p\n", con
, con
->sock
);
289 set_bit(SOCK_CLOSED
, &con
->state
);
290 rc
= con
->sock
->ops
->shutdown(con
->sock
, SHUT_RDWR
);
291 sock_release(con
->sock
);
293 clear_bit(SOCK_CLOSED
, &con
->state
);
298 * Reset a connection. Discard all incoming and outgoing messages
299 * and clear *_seq state.
301 static void ceph_msg_remove(struct ceph_msg
*msg
)
303 list_del_init(&msg
->list_head
);
306 static void ceph_msg_remove_list(struct list_head
*head
)
308 while (!list_empty(head
)) {
309 struct ceph_msg
*msg
= list_first_entry(head
, struct ceph_msg
,
311 ceph_msg_remove(msg
);
315 static void reset_connection(struct ceph_connection
*con
)
317 /* reset connection, out_queue, msg_ and connect_seq */
318 /* discard existing out_queue and msg_seq */
319 mutex_lock(&con
->out_mutex
);
320 ceph_msg_remove_list(&con
->out_queue
);
321 ceph_msg_remove_list(&con
->out_sent
);
323 con
->connect_seq
= 0;
327 mutex_unlock(&con
->out_mutex
);
331 * mark a peer down. drop any open connections.
333 void ceph_con_close(struct ceph_connection
*con
)
335 dout("con_close %p peer %s\n", con
, pr_addr(&con
->peer_addr
.in_addr
));
336 set_bit(CLOSED
, &con
->state
); /* in case there's queued work */
337 clear_bit(STANDBY
, &con
->state
); /* avoid connect_seq bump */
338 reset_connection(con
);
343 * clean up connection state
345 void ceph_con_shutdown(struct ceph_connection
*con
)
347 dout("con_shutdown %p\n", con
);
348 reset_connection(con
);
349 set_bit(DEAD
, &con
->state
);
350 con_close_socket(con
); /* silently ignore errors */
354 * Reopen a closed connection, with a new peer address.
356 void ceph_con_open(struct ceph_connection
*con
, struct ceph_entity_addr
*addr
)
358 dout("con_open %p %s\n", con
, pr_addr(&addr
->in_addr
));
359 set_bit(OPENING
, &con
->state
);
360 clear_bit(CLOSED
, &con
->state
);
361 memcpy(&con
->peer_addr
, addr
, sizeof(*addr
));
368 struct ceph_connection
*ceph_con_get(struct ceph_connection
*con
)
370 dout("con_get %p nref = %d -> %d\n", con
,
371 atomic_read(&con
->nref
), atomic_read(&con
->nref
) + 1);
372 if (atomic_inc_not_zero(&con
->nref
))
377 void ceph_con_put(struct ceph_connection
*con
)
379 dout("con_put %p nref = %d -> %d\n", con
,
380 atomic_read(&con
->nref
), atomic_read(&con
->nref
) - 1);
381 BUG_ON(atomic_read(&con
->nref
) == 0);
382 if (atomic_dec_and_test(&con
->nref
)) {
383 ceph_con_shutdown(con
);
389 * initialize a new connection.
391 void ceph_con_init(struct ceph_messenger
*msgr
, struct ceph_connection
*con
)
393 dout("con_init %p\n", con
);
394 memset(con
, 0, sizeof(*con
));
395 atomic_set(&con
->nref
, 1);
397 mutex_init(&con
->out_mutex
);
398 INIT_LIST_HEAD(&con
->out_queue
);
399 INIT_LIST_HEAD(&con
->out_sent
);
400 INIT_DELAYED_WORK(&con
->work
, con_work
);
405 * We maintain a global counter to order connection attempts. Get
406 * a unique seq greater than @gt.
408 static u32
get_global_seq(struct ceph_messenger
*msgr
, u32 gt
)
412 spin_lock(&msgr
->global_seq_lock
);
413 if (msgr
->global_seq
< gt
)
414 msgr
->global_seq
= gt
;
415 ret
= ++msgr
->global_seq
;
416 spin_unlock(&msgr
->global_seq_lock
);
422 * Prepare footer for currently outgoing message, and finish things
423 * off. Assumes out_kvec* are already valid.. we just add on to the end.
425 static void prepare_write_message_footer(struct ceph_connection
*con
, int v
)
427 struct ceph_msg
*m
= con
->out_msg
;
429 dout("prepare_write_message_footer %p\n", con
);
430 con
->out_kvec_is_msg
= true;
431 con
->out_kvec
[v
].iov_base
= &m
->footer
;
432 con
->out_kvec
[v
].iov_len
= sizeof(m
->footer
);
433 con
->out_kvec_bytes
+= sizeof(m
->footer
);
434 con
->out_kvec_left
++;
435 con
->out_more
= m
->more_to_follow
;
436 con
->out_msg
= NULL
; /* we're done with this one */
440 * Prepare headers for the next outgoing message.
442 static void prepare_write_message(struct ceph_connection
*con
)
447 con
->out_kvec_bytes
= 0;
448 con
->out_kvec_is_msg
= true;
450 /* Sneak an ack in there first? If we can get it into the same
451 * TCP packet that's a good thing. */
452 if (con
->in_seq
> con
->in_seq_acked
) {
453 con
->in_seq_acked
= con
->in_seq
;
454 con
->out_kvec
[v
].iov_base
= &tag_ack
;
455 con
->out_kvec
[v
++].iov_len
= 1;
456 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
457 con
->out_kvec
[v
].iov_base
= &con
->out_temp_ack
;
458 con
->out_kvec
[v
++].iov_len
= sizeof(con
->out_temp_ack
);
459 con
->out_kvec_bytes
= 1 + sizeof(con
->out_temp_ack
);
462 /* move message to sending/sent list */
463 m
= list_first_entry(&con
->out_queue
,
464 struct ceph_msg
, list_head
);
465 list_move_tail(&m
->list_head
, &con
->out_sent
);
466 con
->out_msg
= m
; /* we don't bother taking a reference here. */
468 m
->hdr
.seq
= cpu_to_le64(++con
->out_seq
);
470 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n",
471 m
, con
->out_seq
, le16_to_cpu(m
->hdr
.type
),
472 le32_to_cpu(m
->hdr
.front_len
), le32_to_cpu(m
->hdr
.middle_len
),
473 le32_to_cpu(m
->hdr
.data_len
),
475 BUG_ON(le32_to_cpu(m
->hdr
.front_len
) != m
->front
.iov_len
);
477 /* tag + hdr + front + middle */
478 con
->out_kvec
[v
].iov_base
= &tag_msg
;
479 con
->out_kvec
[v
++].iov_len
= 1;
480 con
->out_kvec
[v
].iov_base
= &m
->hdr
;
481 con
->out_kvec
[v
++].iov_len
= sizeof(m
->hdr
);
482 con
->out_kvec
[v
++] = m
->front
;
484 con
->out_kvec
[v
++] = m
->middle
->vec
;
485 con
->out_kvec_left
= v
;
486 con
->out_kvec_bytes
+= 1 + sizeof(m
->hdr
) + m
->front
.iov_len
+
487 (m
->middle
? m
->middle
->vec
.iov_len
: 0);
488 con
->out_kvec_cur
= con
->out_kvec
;
490 /* fill in crc (except data pages), footer */
491 con
->out_msg
->hdr
.crc
=
492 cpu_to_le32(crc32c(0, (void *)&m
->hdr
,
493 sizeof(m
->hdr
) - sizeof(m
->hdr
.crc
)));
494 con
->out_msg
->footer
.flags
= CEPH_MSG_FOOTER_COMPLETE
;
495 con
->out_msg
->footer
.front_crc
=
496 cpu_to_le32(crc32c(0, m
->front
.iov_base
, m
->front
.iov_len
));
498 con
->out_msg
->footer
.middle_crc
=
499 cpu_to_le32(crc32c(0, m
->middle
->vec
.iov_base
,
500 m
->middle
->vec
.iov_len
));
502 con
->out_msg
->footer
.middle_crc
= 0;
503 con
->out_msg
->footer
.data_crc
= 0;
504 dout("prepare_write_message front_crc %u data_crc %u\n",
505 le32_to_cpu(con
->out_msg
->footer
.front_crc
),
506 le32_to_cpu(con
->out_msg
->footer
.middle_crc
));
508 /* is there a data payload? */
509 if (le32_to_cpu(m
->hdr
.data_len
) > 0) {
510 /* initialize page iterator */
511 con
->out_msg_pos
.page
= 0;
512 con
->out_msg_pos
.page_pos
=
513 le16_to_cpu(m
->hdr
.data_off
) & ~PAGE_MASK
;
514 con
->out_msg_pos
.data_pos
= 0;
515 con
->out_msg_pos
.did_page_crc
= 0;
516 con
->out_more
= 1; /* data + footer will follow */
518 /* no, queue up footer too and be done */
519 prepare_write_message_footer(con
, v
);
522 set_bit(WRITE_PENDING
, &con
->state
);
528 static void prepare_write_ack(struct ceph_connection
*con
)
530 dout("prepare_write_ack %p %llu -> %llu\n", con
,
531 con
->in_seq_acked
, con
->in_seq
);
532 con
->in_seq_acked
= con
->in_seq
;
534 con
->out_kvec
[0].iov_base
= &tag_ack
;
535 con
->out_kvec
[0].iov_len
= 1;
536 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
537 con
->out_kvec
[1].iov_base
= &con
->out_temp_ack
;
538 con
->out_kvec
[1].iov_len
= sizeof(con
->out_temp_ack
);
539 con
->out_kvec_left
= 2;
540 con
->out_kvec_bytes
= 1 + sizeof(con
->out_temp_ack
);
541 con
->out_kvec_cur
= con
->out_kvec
;
542 con
->out_more
= 1; /* more will follow.. eventually.. */
543 set_bit(WRITE_PENDING
, &con
->state
);
547 * Prepare to write keepalive byte.
549 static void prepare_write_keepalive(struct ceph_connection
*con
)
551 dout("prepare_write_keepalive %p\n", con
);
552 con
->out_kvec
[0].iov_base
= &tag_keepalive
;
553 con
->out_kvec
[0].iov_len
= 1;
554 con
->out_kvec_left
= 1;
555 con
->out_kvec_bytes
= 1;
556 con
->out_kvec_cur
= con
->out_kvec
;
557 set_bit(WRITE_PENDING
, &con
->state
);
561 * Connection negotiation.
565 * We connected to a peer and are saying hello.
567 static void prepare_write_connect(struct ceph_messenger
*msgr
,
568 struct ceph_connection
*con
)
570 int len
= strlen(CEPH_BANNER
);
571 unsigned global_seq
= get_global_seq(con
->msgr
, 0);
574 switch (con
->peer_name
.type
) {
575 case CEPH_ENTITY_TYPE_MON
:
576 proto
= CEPH_MONC_PROTOCOL
;
578 case CEPH_ENTITY_TYPE_OSD
:
579 proto
= CEPH_OSDC_PROTOCOL
;
581 case CEPH_ENTITY_TYPE_MDS
:
582 proto
= CEPH_MDSC_PROTOCOL
;
588 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con
,
589 con
->connect_seq
, global_seq
, proto
);
590 con
->out_connect
.host_type
= cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT
);
591 con
->out_connect
.connect_seq
= cpu_to_le32(con
->connect_seq
);
592 con
->out_connect
.global_seq
= cpu_to_le32(global_seq
);
593 con
->out_connect
.protocol_version
= cpu_to_le32(proto
);
594 con
->out_connect
.flags
= 0;
595 if (test_bit(LOSSYTX
, &con
->state
))
596 con
->out_connect
.flags
= CEPH_MSG_CONNECT_LOSSY
;
598 con
->out_kvec
[0].iov_base
= CEPH_BANNER
;
599 con
->out_kvec
[0].iov_len
= len
;
600 con
->out_kvec
[1].iov_base
= &msgr
->my_enc_addr
;
601 con
->out_kvec
[1].iov_len
= sizeof(msgr
->my_enc_addr
);
602 con
->out_kvec
[2].iov_base
= &con
->out_connect
;
603 con
->out_kvec
[2].iov_len
= sizeof(con
->out_connect
);
604 con
->out_kvec_left
= 3;
605 con
->out_kvec_bytes
= len
+ sizeof(msgr
->my_enc_addr
) +
606 sizeof(con
->out_connect
);
607 con
->out_kvec_cur
= con
->out_kvec
;
609 set_bit(WRITE_PENDING
, &con
->state
);
612 static void prepare_write_connect_retry(struct ceph_messenger
*msgr
,
613 struct ceph_connection
*con
)
615 dout("prepare_write_connect_retry %p\n", con
);
616 con
->out_connect
.connect_seq
= cpu_to_le32(con
->connect_seq
);
617 con
->out_connect
.global_seq
=
618 cpu_to_le32(get_global_seq(con
->msgr
, 0));
620 con
->out_kvec
[0].iov_base
= &con
->out_connect
;
621 con
->out_kvec
[0].iov_len
= sizeof(con
->out_connect
);
622 con
->out_kvec_left
= 1;
623 con
->out_kvec_bytes
= sizeof(con
->out_connect
);
624 con
->out_kvec_cur
= con
->out_kvec
;
626 set_bit(WRITE_PENDING
, &con
->state
);
631 * write as much of pending kvecs to the socket as we can.
633 * 0 -> socket full, but more to do
636 static int write_partial_kvec(struct ceph_connection
*con
)
640 dout("write_partial_kvec %p %d left\n", con
, con
->out_kvec_bytes
);
641 while (con
->out_kvec_bytes
> 0) {
642 ret
= ceph_tcp_sendmsg(con
->sock
, con
->out_kvec_cur
,
643 con
->out_kvec_left
, con
->out_kvec_bytes
,
647 con
->out_kvec_bytes
-= ret
;
648 if (con
->out_kvec_bytes
== 0)
651 if (ret
>= con
->out_kvec_cur
->iov_len
) {
652 ret
-= con
->out_kvec_cur
->iov_len
;
654 con
->out_kvec_left
--;
656 con
->out_kvec_cur
->iov_len
-= ret
;
657 con
->out_kvec_cur
->iov_base
+= ret
;
663 con
->out_kvec_left
= 0;
664 con
->out_kvec_is_msg
= false;
667 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con
,
668 con
->out_kvec_bytes
, con
->out_kvec_left
, ret
);
669 return ret
; /* done! */
673 * Write as much message data payload as we can. If we finish, queue
675 * 1 -> done, footer is now queued in out_kvec[].
676 * 0 -> socket full, but more to do
679 static int write_partial_msg_pages(struct ceph_connection
*con
)
681 struct ceph_msg
*msg
= con
->out_msg
;
682 unsigned data_len
= le32_to_cpu(msg
->hdr
.data_len
);
684 int crc
= con
->msgr
->nocrc
;
687 dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
688 con
, con
->out_msg
, con
->out_msg_pos
.page
, con
->out_msg
->nr_pages
,
689 con
->out_msg_pos
.page_pos
);
691 while (con
->out_msg_pos
.page
< con
->out_msg
->nr_pages
) {
692 struct page
*page
= NULL
;
696 * if we are calculating the data crc (the default), we need
697 * to map the page. if our pages[] has been revoked, use the
701 page
= msg
->pages
[con
->out_msg_pos
.page
];
705 page
= con
->msgr
->zero_page
;
707 kaddr
= page_address(con
->msgr
->zero_page
);
709 len
= min((int)(PAGE_SIZE
- con
->out_msg_pos
.page_pos
),
710 (int)(data_len
- con
->out_msg_pos
.data_pos
));
711 if (crc
&& !con
->out_msg_pos
.did_page_crc
) {
712 void *base
= kaddr
+ con
->out_msg_pos
.page_pos
;
713 u32 tmpcrc
= le32_to_cpu(con
->out_msg
->footer
.data_crc
);
715 BUG_ON(kaddr
== NULL
);
716 con
->out_msg
->footer
.data_crc
=
717 cpu_to_le32(crc32c(tmpcrc
, base
, len
));
718 con
->out_msg_pos
.did_page_crc
= 1;
721 ret
= kernel_sendpage(con
->sock
, page
,
722 con
->out_msg_pos
.page_pos
, len
,
723 MSG_DONTWAIT
| MSG_NOSIGNAL
|
726 if (crc
&& msg
->pages
)
732 con
->out_msg_pos
.data_pos
+= ret
;
733 con
->out_msg_pos
.page_pos
+= ret
;
735 con
->out_msg_pos
.page_pos
= 0;
736 con
->out_msg_pos
.page
++;
737 con
->out_msg_pos
.did_page_crc
= 0;
741 dout("write_partial_msg_pages %p msg %p done\n", con
, msg
);
743 /* prepare and queue up footer, too */
745 con
->out_msg
->footer
.flags
|= CEPH_MSG_FOOTER_NOCRC
;
746 con
->out_kvec_bytes
= 0;
747 con
->out_kvec_left
= 0;
748 con
->out_kvec_cur
= con
->out_kvec
;
749 prepare_write_message_footer(con
, 0);
758 static int write_partial_skip(struct ceph_connection
*con
)
762 while (con
->out_skip
> 0) {
764 .iov_base
= page_address(con
->msgr
->zero_page
),
765 .iov_len
= min(con
->out_skip
, (int)PAGE_CACHE_SIZE
)
768 ret
= ceph_tcp_sendmsg(con
->sock
, &iov
, 1, iov
.iov_len
, 1);
771 con
->out_skip
-= ret
;
779 * Prepare to read connection handshake, or an ack.
781 static void prepare_read_connect(struct ceph_connection
*con
)
783 dout("prepare_read_connect %p\n", con
);
784 con
->in_base_pos
= 0;
787 static void prepare_read_ack(struct ceph_connection
*con
)
789 dout("prepare_read_ack %p\n", con
);
790 con
->in_base_pos
= 0;
793 static void prepare_read_tag(struct ceph_connection
*con
)
795 dout("prepare_read_tag %p\n", con
);
796 con
->in_base_pos
= 0;
797 con
->in_tag
= CEPH_MSGR_TAG_READY
;
801 * Prepare to read a message.
803 static int prepare_read_message(struct ceph_connection
*con
)
805 dout("prepare_read_message %p\n", con
);
806 BUG_ON(con
->in_msg
!= NULL
);
807 con
->in_base_pos
= 0;
808 con
->in_front_crc
= con
->in_middle_crc
= con
->in_data_crc
= 0;
813 static int read_partial(struct ceph_connection
*con
,
814 int *to
, int size
, void *object
)
817 while (con
->in_base_pos
< *to
) {
818 int left
= *to
- con
->in_base_pos
;
819 int have
= size
- left
;
820 int ret
= ceph_tcp_recvmsg(con
->sock
, object
+ have
, left
);
823 con
->in_base_pos
+= ret
;
830 * Read all or part of the connect-side handshake on a new connection
832 static int read_partial_connect(struct ceph_connection
*con
)
836 dout("read_partial_connect %p at %d\n", con
, con
->in_base_pos
);
839 ret
= read_partial(con
, &to
, strlen(CEPH_BANNER
), con
->in_banner
);
842 ret
= read_partial(con
, &to
, sizeof(con
->actual_peer_addr
),
843 &con
->actual_peer_addr
);
846 ret
= read_partial(con
, &to
, sizeof(con
->peer_addr_for_me
),
847 &con
->peer_addr_for_me
);
850 ret
= read_partial(con
, &to
, sizeof(con
->in_reply
), &con
->in_reply
);
854 dout("read_partial_connect %p connect_seq = %u, global_seq = %u\n",
855 con
, le32_to_cpu(con
->in_reply
.connect_seq
),
856 le32_to_cpu(con
->in_reply
.global_seq
));
862 * Verify the hello banner looks okay.
864 static int verify_hello(struct ceph_connection
*con
)
866 if (memcmp(con
->in_banner
, CEPH_BANNER
, strlen(CEPH_BANNER
))) {
867 pr_err("connect to %s got bad banner\n",
868 pr_addr(&con
->peer_addr
.in_addr
));
869 con
->error_msg
= "protocol error, bad banner";
875 static bool addr_is_blank(struct sockaddr_storage
*ss
)
877 switch (ss
->ss_family
) {
879 return ((struct sockaddr_in
*)ss
)->sin_addr
.s_addr
== 0;
882 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[0] == 0 &&
883 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[1] == 0 &&
884 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[2] == 0 &&
885 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[3] == 0;
890 static int addr_port(struct sockaddr_storage
*ss
)
892 switch (ss
->ss_family
) {
894 return ((struct sockaddr_in
*)ss
)->sin_port
;
896 return ((struct sockaddr_in6
*)ss
)->sin6_port
;
901 static void addr_set_port(struct sockaddr_storage
*ss
, int p
)
903 switch (ss
->ss_family
) {
905 ((struct sockaddr_in
*)ss
)->sin_port
= htons(p
);
907 ((struct sockaddr_in6
*)ss
)->sin6_port
= htons(p
);
912 * Parse an ip[:port] list into an addr array. Use the default
913 * monitor port if a port isn't specified.
915 int ceph_parse_ips(const char *c
, const char *end
,
916 struct ceph_entity_addr
*addr
,
917 int max_count
, int *count
)
922 dout("parse_ips on '%.*s'\n", (int)(end
-c
), c
);
923 for (i
= 0; i
< max_count
; i
++) {
925 struct sockaddr_storage
*ss
= &addr
[i
].in_addr
;
926 struct sockaddr_in
*in4
= (void *)ss
;
927 struct sockaddr_in6
*in6
= (void *)ss
;
930 memset(ss
, 0, sizeof(*ss
));
931 if (in4_pton(p
, end
- p
, (u8
*)&in4
->sin_addr
.s_addr
,
933 ss
->ss_family
= AF_INET
;
934 } else if (in6_pton(p
, end
- p
, (u8
*)&in6
->sin6_addr
.s6_addr
,
936 ss
->ss_family
= AF_INET6
;
943 if (p
< end
&& *p
== ':') {
946 while (p
< end
&& *p
>= '0' && *p
<= '9') {
947 port
= (port
* 10) + (*p
- '0');
950 if (port
> 65535 || port
== 0)
953 port
= CEPH_MON_PORT
;
956 addr_set_port(ss
, port
);
958 dout("parse_ips got %s\n", pr_addr(ss
));
975 pr_err("parse_ips bad ip '%s'\n", c
);
979 static int process_connect(struct ceph_connection
*con
)
981 dout("process_connect on %p tag %d\n", con
, (int)con
->in_tag
);
983 if (verify_hello(con
) < 0)
986 ceph_decode_addr(&con
->actual_peer_addr
);
987 ceph_decode_addr(&con
->peer_addr_for_me
);
990 * Make sure the other end is who we wanted. note that the other
991 * end may not yet know their ip address, so if it's 0.0.0.0, give
992 * them the benefit of the doubt.
994 if (!ceph_entity_addr_is_local(&con
->peer_addr
,
995 &con
->actual_peer_addr
) &&
996 !(addr_is_blank(&con
->actual_peer_addr
.in_addr
) &&
997 con
->actual_peer_addr
.nonce
== con
->peer_addr
.nonce
)) {
998 pr_err("wrong peer, want %s/%d, "
1000 pr_addr(&con
->peer_addr
.in_addr
),
1001 con
->peer_addr
.nonce
,
1002 pr_addr(&con
->actual_peer_addr
.in_addr
),
1003 con
->actual_peer_addr
.nonce
);
1004 con
->error_msg
= "protocol error, wrong peer";
1009 * did we learn our address?
1011 if (addr_is_blank(&con
->msgr
->inst
.addr
.in_addr
)) {
1012 int port
= addr_port(&con
->msgr
->inst
.addr
.in_addr
);
1014 memcpy(&con
->msgr
->inst
.addr
.in_addr
,
1015 &con
->peer_addr_for_me
.in_addr
,
1016 sizeof(con
->peer_addr_for_me
.in_addr
));
1017 addr_set_port(&con
->msgr
->inst
.addr
.in_addr
, port
);
1018 encode_my_addr(con
->msgr
);
1019 dout("process_connect learned my addr is %s\n",
1020 pr_addr(&con
->msgr
->inst
.addr
.in_addr
));
1023 switch (con
->in_reply
.tag
) {
1024 case CEPH_MSGR_TAG_BADPROTOVER
:
1025 dout("process_connect got BADPROTOVER my %d != their %d\n",
1026 le32_to_cpu(con
->out_connect
.protocol_version
),
1027 le32_to_cpu(con
->in_reply
.protocol_version
));
1028 pr_err("%s%lld %s protocol version mismatch,"
1029 " my %d != server's %d\n",
1030 ENTITY_NAME(con
->peer_name
),
1031 pr_addr(&con
->peer_addr
.in_addr
),
1032 le32_to_cpu(con
->out_connect
.protocol_version
),
1033 le32_to_cpu(con
->in_reply
.protocol_version
));
1034 con
->error_msg
= "protocol version mismatch";
1035 if (con
->ops
->bad_proto
)
1036 con
->ops
->bad_proto(con
);
1037 reset_connection(con
);
1038 set_bit(CLOSED
, &con
->state
); /* in case there's queued work */
1042 case CEPH_MSGR_TAG_RESETSESSION
:
1044 * If we connected with a large connect_seq but the peer
1045 * has no record of a session with us (no connection, or
1046 * connect_seq == 0), they will send RESETSESION to indicate
1047 * that they must have reset their session, and may have
1050 dout("process_connect got RESET peer seq %u\n",
1051 le32_to_cpu(con
->in_connect
.connect_seq
));
1052 pr_err("%s%lld %s connection reset\n",
1053 ENTITY_NAME(con
->peer_name
),
1054 pr_addr(&con
->peer_addr
.in_addr
));
1055 reset_connection(con
);
1056 prepare_write_connect_retry(con
->msgr
, con
);
1057 prepare_read_connect(con
);
1059 /* Tell ceph about it. */
1060 pr_info("reset on %s%lld\n", ENTITY_NAME(con
->peer_name
));
1061 if (con
->ops
->peer_reset
)
1062 con
->ops
->peer_reset(con
);
1065 case CEPH_MSGR_TAG_RETRY_SESSION
:
1067 * If we sent a smaller connect_seq than the peer has, try
1068 * again with a larger value.
1070 dout("process_connect got RETRY my seq = %u, peer_seq = %u\n",
1071 le32_to_cpu(con
->out_connect
.connect_seq
),
1072 le32_to_cpu(con
->in_connect
.connect_seq
));
1073 con
->connect_seq
= le32_to_cpu(con
->in_connect
.connect_seq
);
1074 prepare_write_connect_retry(con
->msgr
, con
);
1075 prepare_read_connect(con
);
1078 case CEPH_MSGR_TAG_RETRY_GLOBAL
:
1080 * If we sent a smaller global_seq than the peer has, try
1081 * again with a larger value.
1083 dout("process_connect got RETRY_GLOBAL my %u, peer_gseq = %u\n",
1084 con
->peer_global_seq
,
1085 le32_to_cpu(con
->in_connect
.global_seq
));
1086 get_global_seq(con
->msgr
,
1087 le32_to_cpu(con
->in_connect
.global_seq
));
1088 prepare_write_connect_retry(con
->msgr
, con
);
1089 prepare_read_connect(con
);
1092 case CEPH_MSGR_TAG_READY
:
1093 clear_bit(CONNECTING
, &con
->state
);
1094 if (con
->in_reply
.flags
& CEPH_MSG_CONNECT_LOSSY
)
1095 set_bit(LOSSYRX
, &con
->state
);
1096 con
->peer_global_seq
= le32_to_cpu(con
->in_reply
.global_seq
);
1098 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1099 con
->peer_global_seq
,
1100 le32_to_cpu(con
->in_reply
.connect_seq
),
1102 WARN_ON(con
->connect_seq
!=
1103 le32_to_cpu(con
->in_reply
.connect_seq
));
1105 con
->delay
= 0; /* reset backoff memory */
1106 prepare_read_tag(con
);
1109 case CEPH_MSGR_TAG_WAIT
:
1111 * If there is a connection race (we are opening
1112 * connections to each other), one of us may just have
1113 * to WAIT. This shouldn't happen if we are the
1116 pr_err("process_connect peer connecting WAIT\n");
1119 pr_err("connect protocol error, will retry\n");
1120 con
->error_msg
= "protocol error, garbage tag during connect";
1128 * read (part of) an ack
1130 static int read_partial_ack(struct ceph_connection
*con
)
1134 return read_partial(con
, &to
, sizeof(con
->in_temp_ack
),
1140 * We can finally discard anything that's been acked.
1142 static void process_ack(struct ceph_connection
*con
)
1145 u64 ack
= le64_to_cpu(con
->in_temp_ack
);
1148 mutex_lock(&con
->out_mutex
);
1149 while (!list_empty(&con
->out_sent
)) {
1150 m
= list_first_entry(&con
->out_sent
, struct ceph_msg
,
1152 seq
= le64_to_cpu(m
->hdr
.seq
);
1155 dout("got ack for seq %llu type %d at %p\n", seq
,
1156 le16_to_cpu(m
->hdr
.type
), m
);
1159 mutex_unlock(&con
->out_mutex
);
1160 prepare_read_tag(con
);
1169 * read (part of) a message.
1171 static int read_partial_message(struct ceph_connection
*con
)
1173 struct ceph_msg
*m
= con
->in_msg
;
1177 unsigned front_len
, middle_len
, data_len
, data_off
;
1178 int datacrc
= con
->msgr
->nocrc
;
1180 dout("read_partial_message con %p msg %p\n", con
, m
);
1183 while (con
->in_base_pos
< sizeof(con
->in_hdr
)) {
1184 left
= sizeof(con
->in_hdr
) - con
->in_base_pos
;
1185 ret
= ceph_tcp_recvmsg(con
->sock
,
1186 (char *)&con
->in_hdr
+ con
->in_base_pos
,
1190 con
->in_base_pos
+= ret
;
1191 if (con
->in_base_pos
== sizeof(con
->in_hdr
)) {
1192 u32 crc
= crc32c(0, (void *)&con
->in_hdr
,
1193 sizeof(con
->in_hdr
) - sizeof(con
->in_hdr
.crc
));
1194 if (crc
!= le32_to_cpu(con
->in_hdr
.crc
)) {
1195 pr_err("read_partial_message bad hdr "
1196 " crc %u != expected %u\n",
1197 crc
, con
->in_hdr
.crc
);
1203 front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
1204 if (front_len
> CEPH_MSG_MAX_FRONT_LEN
)
1206 middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
1207 if (middle_len
> CEPH_MSG_MAX_DATA_LEN
)
1209 data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
1210 if (data_len
> CEPH_MSG_MAX_DATA_LEN
)
1213 /* allocate message? */
1215 dout("got hdr type %d front %d data %d\n", con
->in_hdr
.type
,
1216 con
->in_hdr
.front_len
, con
->in_hdr
.data_len
);
1217 con
->in_msg
= con
->ops
->alloc_msg(con
, &con
->in_hdr
);
1219 /* skip this message */
1220 dout("alloc_msg returned NULL, skipping message\n");
1221 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1223 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1226 if (IS_ERR(con
->in_msg
)) {
1227 ret
= PTR_ERR(con
->in_msg
);
1229 con
->error_msg
= "out of memory for incoming message";
1233 m
->front
.iov_len
= 0; /* haven't read it yet */
1234 memcpy(&m
->hdr
, &con
->in_hdr
, sizeof(con
->in_hdr
));
1238 while (m
->front
.iov_len
< front_len
) {
1239 BUG_ON(m
->front
.iov_base
== NULL
);
1240 left
= front_len
- m
->front
.iov_len
;
1241 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)m
->front
.iov_base
+
1242 m
->front
.iov_len
, left
);
1245 m
->front
.iov_len
+= ret
;
1246 if (m
->front
.iov_len
== front_len
)
1247 con
->in_front_crc
= crc32c(0, m
->front
.iov_base
,
1252 while (middle_len
> 0 && (!m
->middle
||
1253 m
->middle
->vec
.iov_len
< middle_len
)) {
1254 if (m
->middle
== NULL
) {
1256 if (con
->ops
->alloc_middle
)
1257 ret
= con
->ops
->alloc_middle(con
, m
);
1259 dout("alloc_middle failed, skipping payload\n");
1260 con
->in_base_pos
= -middle_len
- data_len
1261 - sizeof(m
->footer
);
1262 ceph_msg_put(con
->in_msg
);
1264 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1267 m
->middle
->vec
.iov_len
= 0;
1269 left
= middle_len
- m
->middle
->vec
.iov_len
;
1270 ret
= ceph_tcp_recvmsg(con
->sock
,
1271 (char *)m
->middle
->vec
.iov_base
+
1272 m
->middle
->vec
.iov_len
, left
);
1275 m
->middle
->vec
.iov_len
+= ret
;
1276 if (m
->middle
->vec
.iov_len
== middle_len
)
1277 con
->in_middle_crc
= crc32c(0, m
->middle
->vec
.iov_base
,
1278 m
->middle
->vec
.iov_len
);
1282 data_off
= le16_to_cpu(m
->hdr
.data_off
);
1286 if (m
->nr_pages
== 0) {
1287 con
->in_msg_pos
.page
= 0;
1288 con
->in_msg_pos
.page_pos
= data_off
& ~PAGE_MASK
;
1289 con
->in_msg_pos
.data_pos
= 0;
1290 /* find pages for data payload */
1291 want
= calc_pages_for(data_off
& ~PAGE_MASK
, data_len
);
1293 if (con
->ops
->prepare_pages
)
1294 ret
= con
->ops
->prepare_pages(con
, m
, want
);
1296 dout("%p prepare_pages failed, skipping payload\n", m
);
1297 con
->in_base_pos
= -data_len
- sizeof(m
->footer
);
1298 ceph_msg_put(con
->in_msg
);
1300 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1303 BUG_ON(m
->nr_pages
< want
);
1305 while (con
->in_msg_pos
.data_pos
< data_len
) {
1306 left
= min((int)(data_len
- con
->in_msg_pos
.data_pos
),
1307 (int)(PAGE_SIZE
- con
->in_msg_pos
.page_pos
));
1308 BUG_ON(m
->pages
== NULL
);
1309 p
= kmap(m
->pages
[con
->in_msg_pos
.page
]);
1310 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ con
->in_msg_pos
.page_pos
,
1312 if (ret
> 0 && datacrc
)
1314 crc32c(con
->in_data_crc
,
1315 p
+ con
->in_msg_pos
.page_pos
, ret
);
1316 kunmap(m
->pages
[con
->in_msg_pos
.page
]);
1319 con
->in_msg_pos
.data_pos
+= ret
;
1320 con
->in_msg_pos
.page_pos
+= ret
;
1321 if (con
->in_msg_pos
.page_pos
== PAGE_SIZE
) {
1322 con
->in_msg_pos
.page_pos
= 0;
1323 con
->in_msg_pos
.page
++;
1329 to
= sizeof(m
->hdr
) + sizeof(m
->footer
);
1330 while (con
->in_base_pos
< to
) {
1331 left
= to
- con
->in_base_pos
;
1332 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)&m
->footer
+
1333 (con
->in_base_pos
- sizeof(m
->hdr
)),
1337 con
->in_base_pos
+= ret
;
1339 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1340 m
, front_len
, m
->footer
.front_crc
, middle_len
,
1341 m
->footer
.middle_crc
, data_len
, m
->footer
.data_crc
);
1344 if (con
->in_front_crc
!= le32_to_cpu(m
->footer
.front_crc
)) {
1345 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1346 m
, con
->in_front_crc
, m
->footer
.front_crc
);
1349 if (con
->in_middle_crc
!= le32_to_cpu(m
->footer
.middle_crc
)) {
1350 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1351 m
, con
->in_middle_crc
, m
->footer
.middle_crc
);
1355 (m
->footer
.flags
& CEPH_MSG_FOOTER_NOCRC
) == 0 &&
1356 con
->in_data_crc
!= le32_to_cpu(m
->footer
.data_crc
)) {
1357 pr_err("read_partial_message %p data crc %u != exp. %u\n", m
,
1358 con
->in_data_crc
, le32_to_cpu(m
->footer
.data_crc
));
1362 return 1; /* done! */
1366 * Process message. This happens in the worker thread. The callback should
1367 * be careful not to do anything that waits on other incoming messages or it
1370 static void process_message(struct ceph_connection
*con
)
1372 struct ceph_msg
*msg
= con
->in_msg
;
1376 /* if first message, set peer_name */
1377 if (con
->peer_name
.type
== 0)
1378 con
->peer_name
= msg
->hdr
.src
.name
;
1380 mutex_lock(&con
->out_mutex
);
1382 mutex_unlock(&con
->out_mutex
);
1384 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
1385 msg
, le64_to_cpu(msg
->hdr
.seq
),
1386 ENTITY_NAME(msg
->hdr
.src
.name
),
1387 le16_to_cpu(msg
->hdr
.type
),
1388 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
1389 le32_to_cpu(msg
->hdr
.front_len
),
1390 le32_to_cpu(msg
->hdr
.data_len
),
1391 con
->in_front_crc
, con
->in_middle_crc
, con
->in_data_crc
);
1392 con
->ops
->dispatch(con
, msg
);
1393 prepare_read_tag(con
);
1398 * Write something to the socket. Called in a worker thread when the
1399 * socket appears to be writeable and we have something ready to send.
1401 static int try_write(struct ceph_connection
*con
)
1403 struct ceph_messenger
*msgr
= con
->msgr
;
1406 dout("try_write start %p state %lu nref %d\n", con
, con
->state
,
1407 atomic_read(&con
->nref
));
1409 mutex_lock(&con
->out_mutex
);
1411 dout("try_write out_kvec_bytes %d\n", con
->out_kvec_bytes
);
1413 /* open the socket first? */
1414 if (con
->sock
== NULL
) {
1416 * if we were STANDBY and are reconnecting _this_
1417 * connection, bump connect_seq now. Always bump
1420 if (test_and_clear_bit(STANDBY
, &con
->state
))
1423 prepare_write_connect(msgr
, con
);
1424 prepare_read_connect(con
);
1425 set_bit(CONNECTING
, &con
->state
);
1427 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1428 dout("try_write initiating connect on %p new state %lu\n",
1430 con
->sock
= ceph_tcp_connect(con
);
1431 if (IS_ERR(con
->sock
)) {
1433 con
->error_msg
= "connect error";
1440 /* kvec data queued? */
1441 if (con
->out_skip
) {
1442 ret
= write_partial_skip(con
);
1446 dout("try_write write_partial_skip err %d\n", ret
);
1450 if (con
->out_kvec_left
) {
1451 ret
= write_partial_kvec(con
);
1455 dout("try_write write_partial_kvec err %d\n", ret
);
1462 ret
= write_partial_msg_pages(con
);
1464 goto more_kvec
; /* we need to send the footer, too! */
1468 dout("try_write write_partial_msg_pages err %d\n",
1474 if (!test_bit(CONNECTING
, &con
->state
)) {
1475 /* is anything else pending? */
1476 if (!list_empty(&con
->out_queue
)) {
1477 prepare_write_message(con
);
1480 if (con
->in_seq
> con
->in_seq_acked
) {
1481 prepare_write_ack(con
);
1484 if (test_and_clear_bit(KEEPALIVE_PENDING
, &con
->state
)) {
1485 prepare_write_keepalive(con
);
1490 /* Nothing to do! */
1491 clear_bit(WRITE_PENDING
, &con
->state
);
1492 dout("try_write nothing else to write.\n");
1496 mutex_unlock(&con
->out_mutex
);
1497 dout("try_write done on %p\n", con
);
1504 * Read what we can from the socket.
1506 static int try_read(struct ceph_connection
*con
)
1508 struct ceph_messenger
*msgr
;
1514 if (test_bit(STANDBY
, &con
->state
))
1517 dout("try_read start on %p\n", con
);
1521 dout("try_read tag %d in_base_pos %d\n", (int)con
->in_tag
,
1523 if (test_bit(CONNECTING
, &con
->state
)) {
1524 dout("try_read connecting\n");
1525 ret
= read_partial_connect(con
);
1528 if (process_connect(con
) < 0) {
1535 if (con
->in_base_pos
< 0) {
1537 * skipping + discarding content.
1539 * FIXME: there must be a better way to do this!
1541 static char buf
[1024];
1542 int skip
= min(1024, -con
->in_base_pos
);
1543 dout("skipping %d / %d bytes\n", skip
, -con
->in_base_pos
);
1544 ret
= ceph_tcp_recvmsg(con
->sock
, buf
, skip
);
1547 con
->in_base_pos
+= ret
;
1548 if (con
->in_base_pos
)
1551 if (con
->in_tag
== CEPH_MSGR_TAG_READY
) {
1555 ret
= ceph_tcp_recvmsg(con
->sock
, &con
->in_tag
, 1);
1558 dout("try_read got tag %d\n", (int)con
->in_tag
);
1559 switch (con
->in_tag
) {
1560 case CEPH_MSGR_TAG_MSG
:
1561 prepare_read_message(con
);
1563 case CEPH_MSGR_TAG_ACK
:
1564 prepare_read_ack(con
);
1566 case CEPH_MSGR_TAG_CLOSE
:
1567 set_bit(CLOSED
, &con
->state
); /* fixme */
1573 if (con
->in_tag
== CEPH_MSGR_TAG_MSG
) {
1574 ret
= read_partial_message(con
);
1578 con
->error_msg
= "bad crc";
1582 con
->error_msg
= "io error";
1588 if (con
->in_tag
== CEPH_MSGR_TAG_READY
)
1590 process_message(con
);
1593 if (con
->in_tag
== CEPH_MSGR_TAG_ACK
) {
1594 ret
= read_partial_ack(con
);
1604 dout("try_read done on %p\n", con
);
1608 pr_err("try_read bad con->in_tag = %d\n", (int)con
->in_tag
);
1609 con
->error_msg
= "protocol error, garbage tag";
1616 * Atomically queue work on a connection. Bump @con reference to
1617 * avoid races with connection teardown.
1619 * There is some trickery going on with QUEUED and BUSY because we
1620 * only want a _single_ thread operating on each connection at any
1621 * point in time, but we want to use all available CPUs.
1623 * The worker thread only proceeds if it can atomically set BUSY. It
1624 * clears QUEUED and does it's thing. When it thinks it's done, it
1625 * clears BUSY, then rechecks QUEUED.. if it's set again, it loops
1626 * (tries again to set BUSY).
1628 * To queue work, we first set QUEUED, _then_ if BUSY isn't set, we
1629 * try to queue work. If that fails (work is already queued, or BUSY)
1630 * we give up (work also already being done or is queued) but leave QUEUED
1631 * set so that the worker thread will loop if necessary.
1633 static void queue_con(struct ceph_connection
*con
)
1635 if (test_bit(DEAD
, &con
->state
)) {
1636 dout("queue_con %p ignoring: DEAD\n",
1641 if (!con
->ops
->get(con
)) {
1642 dout("queue_con %p ref count 0\n", con
);
1646 set_bit(QUEUED
, &con
->state
);
1647 if (test_bit(BUSY
, &con
->state
)) {
1648 dout("queue_con %p - already BUSY\n", con
);
1650 } else if (!queue_work(ceph_msgr_wq
, &con
->work
.work
)) {
1651 dout("queue_con %p - already queued\n", con
);
1654 dout("queue_con %p\n", con
);
1659 * Do some work on a connection. Drop a connection ref when we're done.
1661 static void con_work(struct work_struct
*work
)
1663 struct ceph_connection
*con
= container_of(work
, struct ceph_connection
,
1668 if (test_and_set_bit(BUSY
, &con
->state
) != 0) {
1669 dout("con_work %p BUSY already set\n", con
);
1672 dout("con_work %p start, clearing QUEUED\n", con
);
1673 clear_bit(QUEUED
, &con
->state
);
1675 if (test_bit(CLOSED
, &con
->state
)) { /* e.g. if we are replaced */
1676 dout("con_work CLOSED\n");
1677 con_close_socket(con
);
1680 if (test_and_clear_bit(OPENING
, &con
->state
)) {
1681 /* reopen w/ new peer */
1682 dout("con_work OPENING\n");
1683 con_close_socket(con
);
1686 if (test_and_clear_bit(SOCK_CLOSED
, &con
->state
) ||
1687 try_read(con
) < 0 ||
1688 try_write(con
) < 0) {
1690 ceph_fault(con
); /* error/fault path */
1694 clear_bit(BUSY
, &con
->state
);
1695 dout("con->state=%lu\n", con
->state
);
1696 if (test_bit(QUEUED
, &con
->state
)) {
1698 dout("con_work %p QUEUED reset, looping\n", con
);
1701 dout("con_work %p QUEUED reset, but just faulted\n", con
);
1702 clear_bit(QUEUED
, &con
->state
);
1704 dout("con_work %p done\n", con
);
1712 * Generic error/fault handler. A retry mechanism is used with
1713 * exponential backoff
1715 static void ceph_fault(struct ceph_connection
*con
)
1717 pr_err("%s%lld %s %s\n", ENTITY_NAME(con
->peer_name
),
1718 pr_addr(&con
->peer_addr
.in_addr
), con
->error_msg
);
1719 dout("fault %p state %lu to peer %s\n",
1720 con
, con
->state
, pr_addr(&con
->peer_addr
.in_addr
));
1722 if (test_bit(LOSSYTX
, &con
->state
)) {
1723 dout("fault on LOSSYTX channel\n");
1727 clear_bit(BUSY
, &con
->state
); /* to avoid an improbable race */
1729 con_close_socket(con
);
1732 /* If there are no messages in the queue, place the connection
1733 * in a STANDBY state (i.e., don't try to reconnect just yet). */
1734 mutex_lock(&con
->out_mutex
);
1735 if (list_empty(&con
->out_queue
) && !con
->out_keepalive_pending
) {
1736 dout("fault setting STANDBY\n");
1737 set_bit(STANDBY
, &con
->state
);
1738 mutex_unlock(&con
->out_mutex
);
1742 /* Requeue anything that hasn't been acked, and retry after a
1744 list_splice_init(&con
->out_sent
, &con
->out_queue
);
1745 mutex_unlock(&con
->out_mutex
);
1747 if (con
->delay
== 0)
1748 con
->delay
= BASE_DELAY_INTERVAL
;
1749 else if (con
->delay
< MAX_DELAY_INTERVAL
)
1752 /* explicitly schedule work to try to reconnect again later. */
1753 dout("fault queueing %p delay %lu\n", con
, con
->delay
);
1755 if (queue_delayed_work(ceph_msgr_wq
, &con
->work
,
1756 round_jiffies_relative(con
->delay
)) == 0)
1760 if (con
->ops
->fault
)
1761 con
->ops
->fault(con
);
1767 * create a new messenger instance
1769 struct ceph_messenger
*ceph_messenger_create(struct ceph_entity_addr
*myaddr
)
1771 struct ceph_messenger
*msgr
;
1773 msgr
= kzalloc(sizeof(*msgr
), GFP_KERNEL
);
1775 return ERR_PTR(-ENOMEM
);
1777 spin_lock_init(&msgr
->global_seq_lock
);
1779 /* the zero page is needed if a request is "canceled" while the message
1780 * is being written over the socket */
1781 msgr
->zero_page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
1782 if (!msgr
->zero_page
) {
1784 return ERR_PTR(-ENOMEM
);
1786 kmap(msgr
->zero_page
);
1789 msgr
->inst
.addr
= *myaddr
;
1791 /* select a random nonce */
1792 get_random_bytes(&msgr
->inst
.addr
.nonce
,
1793 sizeof(msgr
->inst
.addr
.nonce
));
1794 encode_my_addr(msgr
);
1796 dout("messenger_create %p\n", msgr
);
1800 void ceph_messenger_destroy(struct ceph_messenger
*msgr
)
1802 dout("destroy %p\n", msgr
);
1803 kunmap(msgr
->zero_page
);
1804 __free_page(msgr
->zero_page
);
1806 dout("destroyed messenger %p\n", msgr
);
1810 * Queue up an outgoing message on the given connection.
1812 void ceph_con_send(struct ceph_connection
*con
, struct ceph_msg
*msg
)
1814 if (test_bit(CLOSED
, &con
->state
)) {
1815 dout("con_send %p closed, dropping %p\n", con
, msg
);
1821 msg
->hdr
.src
.name
= con
->msgr
->inst
.name
;
1822 msg
->hdr
.src
.addr
= con
->msgr
->my_enc_addr
;
1823 msg
->hdr
.orig_src
= msg
->hdr
.src
;
1824 msg
->hdr
.dst_erank
= con
->peer_addr
.erank
;
1827 mutex_lock(&con
->out_mutex
);
1828 BUG_ON(!list_empty(&msg
->list_head
));
1829 list_add_tail(&msg
->list_head
, &con
->out_queue
);
1830 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg
,
1831 ENTITY_NAME(con
->peer_name
), le16_to_cpu(msg
->hdr
.type
),
1832 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
1833 le32_to_cpu(msg
->hdr
.front_len
),
1834 le32_to_cpu(msg
->hdr
.middle_len
),
1835 le32_to_cpu(msg
->hdr
.data_len
));
1836 mutex_unlock(&con
->out_mutex
);
1838 /* if there wasn't anything waiting to send before, queue
1840 if (test_and_set_bit(WRITE_PENDING
, &con
->state
) == 0)
1845 * Revoke a message that was previously queued for send
1847 void ceph_con_revoke(struct ceph_connection
*con
, struct ceph_msg
*msg
)
1849 mutex_lock(&con
->out_mutex
);
1850 if (!list_empty(&msg
->list_head
)) {
1851 dout("con_revoke %p msg %p\n", con
, msg
);
1852 list_del_init(&msg
->list_head
);
1855 if (con
->out_msg
== msg
)
1856 con
->out_msg
= NULL
;
1857 if (con
->out_kvec_is_msg
) {
1858 con
->out_skip
= con
->out_kvec_bytes
;
1859 con
->out_kvec_is_msg
= false;
1862 dout("con_revoke %p msg %p - not queued (sent?)\n", con
, msg
);
1864 mutex_unlock(&con
->out_mutex
);
1868 * Queue a keepalive byte to ensure the tcp connection is alive.
1870 void ceph_con_keepalive(struct ceph_connection
*con
)
1872 if (test_and_set_bit(KEEPALIVE_PENDING
, &con
->state
) == 0 &&
1873 test_and_set_bit(WRITE_PENDING
, &con
->state
) == 0)
1879 * construct a new message with given type, size
1880 * the new msg has a ref count of 1.
1882 struct ceph_msg
*ceph_msg_new(int type
, int front_len
,
1883 int page_len
, int page_off
, struct page
**pages
)
1887 m
= kmalloc(sizeof(*m
), GFP_NOFS
);
1890 atomic_set(&m
->nref
, 1);
1891 INIT_LIST_HEAD(&m
->list_head
);
1893 m
->hdr
.type
= cpu_to_le16(type
);
1894 m
->hdr
.front_len
= cpu_to_le32(front_len
);
1895 m
->hdr
.middle_len
= 0;
1896 m
->hdr
.data_len
= cpu_to_le32(page_len
);
1897 m
->hdr
.data_off
= cpu_to_le16(page_off
);
1898 m
->hdr
.priority
= cpu_to_le16(CEPH_MSG_PRIO_DEFAULT
);
1899 m
->footer
.front_crc
= 0;
1900 m
->footer
.middle_crc
= 0;
1901 m
->footer
.data_crc
= 0;
1902 m
->front_max
= front_len
;
1903 m
->front_is_vmalloc
= false;
1904 m
->more_to_follow
= false;
1909 if (front_len
> PAGE_CACHE_SIZE
) {
1910 m
->front
.iov_base
= __vmalloc(front_len
, GFP_NOFS
,
1912 m
->front_is_vmalloc
= true;
1914 m
->front
.iov_base
= kmalloc(front_len
, GFP_NOFS
);
1916 if (m
->front
.iov_base
== NULL
) {
1917 pr_err("msg_new can't allocate %d bytes\n",
1922 m
->front
.iov_base
= NULL
;
1924 m
->front
.iov_len
= front_len
;
1930 m
->nr_pages
= calc_pages_for(page_off
, page_len
);
1933 dout("ceph_msg_new %p page %d~%d -> %d\n", m
, page_off
, page_len
,
1940 pr_err("msg_new can't create type %d len %d\n", type
, front_len
);
1941 return ERR_PTR(-ENOMEM
);
1945 * Generic message allocator, for incoming messages.
1947 struct ceph_msg
*ceph_alloc_msg(struct ceph_connection
*con
,
1948 struct ceph_msg_header
*hdr
)
1950 int type
= le16_to_cpu(hdr
->type
);
1951 int front_len
= le32_to_cpu(hdr
->front_len
);
1952 struct ceph_msg
*msg
= ceph_msg_new(type
, front_len
, 0, 0, NULL
);
1955 pr_err("unable to allocate msg type %d len %d\n",
1957 return ERR_PTR(-ENOMEM
);
1963 * Allocate "middle" portion of a message, if it is needed and wasn't
1964 * allocated by alloc_msg. This allows us to read a small fixed-size
1965 * per-type header in the front and then gracefully fail (i.e.,
1966 * propagate the error to the caller based on info in the front) when
1967 * the middle is too large.
1969 int ceph_alloc_middle(struct ceph_connection
*con
, struct ceph_msg
*msg
)
1971 int type
= le16_to_cpu(msg
->hdr
.type
);
1972 int middle_len
= le32_to_cpu(msg
->hdr
.middle_len
);
1974 dout("alloc_middle %p type %d %s middle_len %d\n", msg
, type
,
1975 ceph_msg_type_name(type
), middle_len
);
1976 BUG_ON(!middle_len
);
1977 BUG_ON(msg
->middle
);
1979 msg
->middle
= ceph_buffer_new_alloc(middle_len
, GFP_NOFS
);
1987 * Free a generically kmalloc'd message.
1989 void ceph_msg_kfree(struct ceph_msg
*m
)
1991 dout("msg_kfree %p\n", m
);
1992 if (m
->front_is_vmalloc
)
1993 vfree(m
->front
.iov_base
);
1995 kfree(m
->front
.iov_base
);
2000 * Drop a msg ref. Destroy as needed.
2002 void ceph_msg_put(struct ceph_msg
*m
)
2004 dout("ceph_msg_put %p %d -> %d\n", m
, atomic_read(&m
->nref
),
2005 atomic_read(&m
->nref
)-1);
2006 if (atomic_read(&m
->nref
) <= 0) {
2007 pr_err("bad ceph_msg_put on %p %llu %d=%s %d+%d\n",
2008 m
, le64_to_cpu(m
->hdr
.seq
),
2009 le16_to_cpu(m
->hdr
.type
),
2010 ceph_msg_type_name(le16_to_cpu(m
->hdr
.type
)),
2011 le32_to_cpu(m
->hdr
.front_len
),
2012 le32_to_cpu(m
->hdr
.data_len
));
2015 if (atomic_dec_and_test(&m
->nref
)) {
2016 dout("ceph_msg_put last one on %p\n", m
);
2017 WARN_ON(!list_empty(&m
->list_head
));
2019 /* drop middle, data, if any */
2021 ceph_buffer_put(m
->middle
);
2028 ceph_msgpool_put(m
->pool
, m
);