1 #include <linux/ceph/ceph_debug.h>
3 #include <linux/crc32c.h>
4 #include <linux/ctype.h>
5 #include <linux/highmem.h>
6 #include <linux/inet.h>
7 #include <linux/kthread.h>
9 #include <linux/slab.h>
10 #include <linux/socket.h>
11 #include <linux/string.h>
12 #include <linux/bio.h>
13 #include <linux/blkdev.h>
14 #include <linux/dns_resolver.h>
17 #include <linux/ceph/libceph.h>
18 #include <linux/ceph/messenger.h>
19 #include <linux/ceph/decode.h>
20 #include <linux/ceph/pagelist.h>
21 #include <linux/export.h>
24 * Ceph uses the messenger to exchange ceph_msg messages with other
25 * hosts in the system. The messenger provides ordered and reliable
26 * delivery. We tolerate TCP disconnects by reconnecting (with
27 * exponential backoff) in the case of a fault (disconnection, bad
28 * crc, protocol error). Acks allow sent messages to be discarded by
33 * We track the state of the socket on a given connection using
34 * values defined below. The transition to a new socket state is
35 * handled by a function which verifies we aren't coming from an
39 * | NEW* | transient initial state
41 * | con_sock_state_init()
44 * | CLOSED | initialized, but no socket (and no
45 * ---------- TCP connection)
47 * | \ con_sock_state_connecting()
48 * | ----------------------
50 * + con_sock_state_closed() \
51 * |+--------------------------- \
54 * | | CLOSING | socket event; \ \
55 * | ----------- await close \ \
58 * | + con_sock_state_closing() \ |
60 * | / --------------- | |
63 * | / -----------------| CONNECTING | socket created, TCP
64 * | | / -------------- connect initiated
65 * | | | con_sock_state_connected()
68 * | CONNECTED | TCP connection established
71 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
74 #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
75 #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
76 #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
77 #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
78 #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
80 /* static tag bytes (protocol control messages) */
81 static char tag_msg
= CEPH_MSGR_TAG_MSG
;
82 static char tag_ack
= CEPH_MSGR_TAG_ACK
;
83 static char tag_keepalive
= CEPH_MSGR_TAG_KEEPALIVE
;
86 static struct lock_class_key socket_class
;
90 * When skipping (ignoring) a block of input we read it into a "skip
91 * buffer," which is this many bytes in size.
93 #define SKIP_BUF_SIZE 1024
95 static void queue_con(struct ceph_connection
*con
);
96 static void con_work(struct work_struct
*);
97 static void ceph_fault(struct ceph_connection
*con
);
100 * Nicely render a sockaddr as a string. An array of formatted
101 * strings is used, to approximate reentrancy.
103 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
104 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
105 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
106 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
108 static char addr_str
[ADDR_STR_COUNT
][MAX_ADDR_STR_LEN
];
109 static atomic_t addr_str_seq
= ATOMIC_INIT(0);
111 static struct page
*zero_page
; /* used in certain error cases */
113 const char *ceph_pr_addr(const struct sockaddr_storage
*ss
)
117 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
118 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
120 i
= atomic_inc_return(&addr_str_seq
) & ADDR_STR_COUNT_MASK
;
123 switch (ss
->ss_family
) {
125 snprintf(s
, MAX_ADDR_STR_LEN
, "%pI4:%hu", &in4
->sin_addr
,
126 ntohs(in4
->sin_port
));
130 snprintf(s
, MAX_ADDR_STR_LEN
, "[%pI6c]:%hu", &in6
->sin6_addr
,
131 ntohs(in6
->sin6_port
));
135 snprintf(s
, MAX_ADDR_STR_LEN
, "(unknown sockaddr family %hu)",
141 EXPORT_SYMBOL(ceph_pr_addr
);
143 static void encode_my_addr(struct ceph_messenger
*msgr
)
145 memcpy(&msgr
->my_enc_addr
, &msgr
->inst
.addr
, sizeof(msgr
->my_enc_addr
));
146 ceph_encode_addr(&msgr
->my_enc_addr
);
150 * work queue for all reading and writing to/from the socket.
152 static struct workqueue_struct
*ceph_msgr_wq
;
154 void _ceph_msgr_exit(void)
157 destroy_workqueue(ceph_msgr_wq
);
161 BUG_ON(zero_page
== NULL
);
163 page_cache_release(zero_page
);
167 int ceph_msgr_init(void)
169 BUG_ON(zero_page
!= NULL
);
170 zero_page
= ZERO_PAGE(0);
171 page_cache_get(zero_page
);
173 ceph_msgr_wq
= alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT
, 0);
177 pr_err("msgr_init failed to create workqueue\n");
182 EXPORT_SYMBOL(ceph_msgr_init
);
184 void ceph_msgr_exit(void)
186 BUG_ON(ceph_msgr_wq
== NULL
);
190 EXPORT_SYMBOL(ceph_msgr_exit
);
192 void ceph_msgr_flush(void)
194 flush_workqueue(ceph_msgr_wq
);
196 EXPORT_SYMBOL(ceph_msgr_flush
);
198 /* Connection socket state transition functions */
200 static void con_sock_state_init(struct ceph_connection
*con
)
204 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSED
);
205 if (WARN_ON(old_state
!= CON_SOCK_STATE_NEW
))
206 printk("%s: unexpected old state %d\n", __func__
, old_state
);
209 static void con_sock_state_connecting(struct ceph_connection
*con
)
213 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CONNECTING
);
214 if (WARN_ON(old_state
!= CON_SOCK_STATE_CLOSED
))
215 printk("%s: unexpected old state %d\n", __func__
, old_state
);
218 static void con_sock_state_connected(struct ceph_connection
*con
)
222 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CONNECTED
);
223 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTING
))
224 printk("%s: unexpected old state %d\n", __func__
, old_state
);
227 static void con_sock_state_closing(struct ceph_connection
*con
)
231 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSING
);
232 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTING
&&
233 old_state
!= CON_SOCK_STATE_CONNECTED
&&
234 old_state
!= CON_SOCK_STATE_CLOSING
))
235 printk("%s: unexpected old state %d\n", __func__
, old_state
);
238 static void con_sock_state_closed(struct ceph_connection
*con
)
242 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSED
);
243 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTED
&&
244 old_state
!= CON_SOCK_STATE_CLOSING
&&
245 old_state
!= CON_SOCK_STATE_CONNECTING
))
246 printk("%s: unexpected old state %d\n", __func__
, old_state
);
250 * socket callback functions
253 /* data available on socket, or listen socket received a connect */
254 static void ceph_sock_data_ready(struct sock
*sk
, int count_unused
)
256 struct ceph_connection
*con
= sk
->sk_user_data
;
257 if (atomic_read(&con
->msgr
->stopping
)) {
261 if (sk
->sk_state
!= TCP_CLOSE_WAIT
) {
262 dout("%s on %p state = %lu, queueing work\n", __func__
,
268 /* socket has buffer space for writing */
269 static void ceph_sock_write_space(struct sock
*sk
)
271 struct ceph_connection
*con
= sk
->sk_user_data
;
273 /* only queue to workqueue if there is data we want to write,
274 * and there is sufficient space in the socket buffer to accept
275 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
276 * doesn't get called again until try_write() fills the socket
277 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
278 * and net/core/stream.c:sk_stream_write_space().
280 if (test_bit(WRITE_PENDING
, &con
->flags
)) {
281 if (sk_stream_wspace(sk
) >= sk_stream_min_wspace(sk
)) {
282 dout("%s %p queueing write work\n", __func__
, con
);
283 clear_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
287 dout("%s %p nothing to write\n", __func__
, con
);
291 /* socket's state has changed */
292 static void ceph_sock_state_change(struct sock
*sk
)
294 struct ceph_connection
*con
= sk
->sk_user_data
;
296 dout("%s %p state = %lu sk_state = %u\n", __func__
,
297 con
, con
->state
, sk
->sk_state
);
299 if (test_bit(CLOSED
, &con
->state
))
302 switch (sk
->sk_state
) {
304 dout("%s TCP_CLOSE\n", __func__
);
306 dout("%s TCP_CLOSE_WAIT\n", __func__
);
307 con_sock_state_closing(con
);
308 set_bit(SOCK_CLOSED
, &con
->flags
);
311 case TCP_ESTABLISHED
:
312 dout("%s TCP_ESTABLISHED\n", __func__
);
313 con_sock_state_connected(con
);
316 default: /* Everything else is uninteresting */
322 * set up socket callbacks
324 static void set_sock_callbacks(struct socket
*sock
,
325 struct ceph_connection
*con
)
327 struct sock
*sk
= sock
->sk
;
328 sk
->sk_user_data
= con
;
329 sk
->sk_data_ready
= ceph_sock_data_ready
;
330 sk
->sk_write_space
= ceph_sock_write_space
;
331 sk
->sk_state_change
= ceph_sock_state_change
;
340 * initiate connection to a remote socket.
342 static int ceph_tcp_connect(struct ceph_connection
*con
)
344 struct sockaddr_storage
*paddr
= &con
->peer_addr
.in_addr
;
349 ret
= sock_create_kern(con
->peer_addr
.in_addr
.ss_family
, SOCK_STREAM
,
353 sock
->sk
->sk_allocation
= GFP_NOFS
;
355 #ifdef CONFIG_LOCKDEP
356 lockdep_set_class(&sock
->sk
->sk_lock
, &socket_class
);
359 set_sock_callbacks(sock
, con
);
361 dout("connect %s\n", ceph_pr_addr(&con
->peer_addr
.in_addr
));
363 con_sock_state_connecting(con
);
364 ret
= sock
->ops
->connect(sock
, (struct sockaddr
*)paddr
, sizeof(*paddr
),
366 if (ret
== -EINPROGRESS
) {
367 dout("connect %s EINPROGRESS sk_state = %u\n",
368 ceph_pr_addr(&con
->peer_addr
.in_addr
),
370 } else if (ret
< 0) {
371 pr_err("connect %s error %d\n",
372 ceph_pr_addr(&con
->peer_addr
.in_addr
), ret
);
374 con
->error_msg
= "connect error";
382 static int ceph_tcp_recvmsg(struct socket
*sock
, void *buf
, size_t len
)
384 struct kvec iov
= {buf
, len
};
385 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
388 r
= kernel_recvmsg(sock
, &msg
, &iov
, 1, len
, msg
.msg_flags
);
395 * write something. @more is true if caller will be sending more data
398 static int ceph_tcp_sendmsg(struct socket
*sock
, struct kvec
*iov
,
399 size_t kvlen
, size_t len
, int more
)
401 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
405 msg
.msg_flags
|= MSG_MORE
;
407 msg
.msg_flags
|= MSG_EOR
; /* superfluous, but what the hell */
409 r
= kernel_sendmsg(sock
, &msg
, iov
, kvlen
, len
);
415 static int ceph_tcp_sendpage(struct socket
*sock
, struct page
*page
,
416 int offset
, size_t size
, int more
)
418 int flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
| (more
? MSG_MORE
: MSG_EOR
);
421 ret
= kernel_sendpage(sock
, page
, offset
, size
, flags
);
430 * Shutdown/close the socket for the given connection.
432 static int con_close_socket(struct ceph_connection
*con
)
436 dout("con_close_socket on %p sock %p\n", con
, con
->sock
);
439 rc
= con
->sock
->ops
->shutdown(con
->sock
, SHUT_RDWR
);
440 sock_release(con
->sock
);
444 * Forcibly clear the SOCK_CLOSE flag. It gets set
445 * independent of the connection mutex, and we could have
446 * received a socket close event before we had the chance to
447 * shut the socket down.
449 clear_bit(SOCK_CLOSED
, &con
->flags
);
450 con_sock_state_closed(con
);
455 * Reset a connection. Discard all incoming and outgoing messages
456 * and clear *_seq state.
458 static void ceph_msg_remove(struct ceph_msg
*msg
)
460 list_del_init(&msg
->list_head
);
461 BUG_ON(msg
->con
== NULL
);
462 msg
->con
->ops
->put(msg
->con
);
467 static void ceph_msg_remove_list(struct list_head
*head
)
469 while (!list_empty(head
)) {
470 struct ceph_msg
*msg
= list_first_entry(head
, struct ceph_msg
,
472 ceph_msg_remove(msg
);
476 static void reset_connection(struct ceph_connection
*con
)
478 /* reset connection, out_queue, msg_ and connect_seq */
479 /* discard existing out_queue and msg_seq */
480 ceph_msg_remove_list(&con
->out_queue
);
481 ceph_msg_remove_list(&con
->out_sent
);
484 BUG_ON(con
->in_msg
->con
!= con
);
485 con
->in_msg
->con
= NULL
;
486 ceph_msg_put(con
->in_msg
);
491 con
->connect_seq
= 0;
494 ceph_msg_put(con
->out_msg
);
498 con
->in_seq_acked
= 0;
502 * mark a peer down. drop any open connections.
504 void ceph_con_close(struct ceph_connection
*con
)
506 mutex_lock(&con
->mutex
);
507 dout("con_close %p peer %s\n", con
,
508 ceph_pr_addr(&con
->peer_addr
.in_addr
));
509 clear_bit(NEGOTIATING
, &con
->state
);
510 clear_bit(CONNECTING
, &con
->state
);
511 clear_bit(CONNECTED
, &con
->state
);
512 clear_bit(STANDBY
, &con
->state
); /* avoid connect_seq bump */
513 set_bit(CLOSED
, &con
->state
);
515 clear_bit(LOSSYTX
, &con
->flags
); /* so we retry next connect */
516 clear_bit(KEEPALIVE_PENDING
, &con
->flags
);
517 clear_bit(WRITE_PENDING
, &con
->flags
);
519 reset_connection(con
);
520 con
->peer_global_seq
= 0;
521 cancel_delayed_work(&con
->work
);
522 mutex_unlock(&con
->mutex
);
525 * We cannot close the socket directly from here because the
526 * work threads use it without holding the mutex. Instead, let
531 EXPORT_SYMBOL(ceph_con_close
);
534 * Reopen a closed connection, with a new peer address.
536 void ceph_con_open(struct ceph_connection
*con
,
537 __u8 entity_type
, __u64 entity_num
,
538 struct ceph_entity_addr
*addr
)
540 mutex_lock(&con
->mutex
);
541 dout("con_open %p %s\n", con
, ceph_pr_addr(&addr
->in_addr
));
542 set_bit(OPENING
, &con
->state
);
543 WARN_ON(!test_and_clear_bit(CLOSED
, &con
->state
));
545 con
->peer_name
.type
= (__u8
) entity_type
;
546 con
->peer_name
.num
= cpu_to_le64(entity_num
);
548 memcpy(&con
->peer_addr
, addr
, sizeof(*addr
));
549 con
->delay
= 0; /* reset backoff memory */
550 mutex_unlock(&con
->mutex
);
553 EXPORT_SYMBOL(ceph_con_open
);
556 * return true if this connection ever successfully opened
558 bool ceph_con_opened(struct ceph_connection
*con
)
560 return con
->connect_seq
> 0;
564 * initialize a new connection.
566 void ceph_con_init(struct ceph_connection
*con
, void *private,
567 const struct ceph_connection_operations
*ops
,
568 struct ceph_messenger
*msgr
)
570 dout("con_init %p\n", con
);
571 memset(con
, 0, sizeof(*con
));
572 con
->private = private;
576 con_sock_state_init(con
);
578 mutex_init(&con
->mutex
);
579 INIT_LIST_HEAD(&con
->out_queue
);
580 INIT_LIST_HEAD(&con
->out_sent
);
581 INIT_DELAYED_WORK(&con
->work
, con_work
);
583 set_bit(CLOSED
, &con
->state
);
585 EXPORT_SYMBOL(ceph_con_init
);
589 * We maintain a global counter to order connection attempts. Get
590 * a unique seq greater than @gt.
592 static u32
get_global_seq(struct ceph_messenger
*msgr
, u32 gt
)
596 spin_lock(&msgr
->global_seq_lock
);
597 if (msgr
->global_seq
< gt
)
598 msgr
->global_seq
= gt
;
599 ret
= ++msgr
->global_seq
;
600 spin_unlock(&msgr
->global_seq_lock
);
604 static void con_out_kvec_reset(struct ceph_connection
*con
)
606 con
->out_kvec_left
= 0;
607 con
->out_kvec_bytes
= 0;
608 con
->out_kvec_cur
= &con
->out_kvec
[0];
611 static void con_out_kvec_add(struct ceph_connection
*con
,
612 size_t size
, void *data
)
616 index
= con
->out_kvec_left
;
617 BUG_ON(index
>= ARRAY_SIZE(con
->out_kvec
));
619 con
->out_kvec
[index
].iov_len
= size
;
620 con
->out_kvec
[index
].iov_base
= data
;
621 con
->out_kvec_left
++;
622 con
->out_kvec_bytes
+= size
;
626 static void init_bio_iter(struct bio
*bio
, struct bio
**iter
, int *seg
)
637 static void iter_bio_next(struct bio
**bio_iter
, int *seg
)
639 if (*bio_iter
== NULL
)
642 BUG_ON(*seg
>= (*bio_iter
)->bi_vcnt
);
645 if (*seg
== (*bio_iter
)->bi_vcnt
)
646 init_bio_iter((*bio_iter
)->bi_next
, bio_iter
, seg
);
650 static void prepare_write_message_data(struct ceph_connection
*con
)
652 struct ceph_msg
*msg
= con
->out_msg
;
655 BUG_ON(!msg
->hdr
.data_len
);
657 /* initialize page iterator */
658 con
->out_msg_pos
.page
= 0;
660 con
->out_msg_pos
.page_pos
= msg
->page_alignment
;
662 con
->out_msg_pos
.page_pos
= 0;
665 init_bio_iter(msg
->bio
, &msg
->bio_iter
, &msg
->bio_seg
);
667 con
->out_msg_pos
.data_pos
= 0;
668 con
->out_msg_pos
.did_page_crc
= false;
669 con
->out_more
= 1; /* data + footer will follow */
673 * Prepare footer for currently outgoing message, and finish things
674 * off. Assumes out_kvec* are already valid.. we just add on to the end.
676 static void prepare_write_message_footer(struct ceph_connection
*con
)
678 struct ceph_msg
*m
= con
->out_msg
;
679 int v
= con
->out_kvec_left
;
681 m
->footer
.flags
|= CEPH_MSG_FOOTER_COMPLETE
;
683 dout("prepare_write_message_footer %p\n", con
);
684 con
->out_kvec_is_msg
= true;
685 con
->out_kvec
[v
].iov_base
= &m
->footer
;
686 con
->out_kvec
[v
].iov_len
= sizeof(m
->footer
);
687 con
->out_kvec_bytes
+= sizeof(m
->footer
);
688 con
->out_kvec_left
++;
689 con
->out_more
= m
->more_to_follow
;
690 con
->out_msg_done
= true;
694 * Prepare headers for the next outgoing message.
696 static void prepare_write_message(struct ceph_connection
*con
)
701 con_out_kvec_reset(con
);
702 con
->out_kvec_is_msg
= true;
703 con
->out_msg_done
= false;
705 /* Sneak an ack in there first? If we can get it into the same
706 * TCP packet that's a good thing. */
707 if (con
->in_seq
> con
->in_seq_acked
) {
708 con
->in_seq_acked
= con
->in_seq
;
709 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
710 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
711 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
715 BUG_ON(list_empty(&con
->out_queue
));
716 m
= list_first_entry(&con
->out_queue
, struct ceph_msg
, list_head
);
718 BUG_ON(m
->con
!= con
);
720 /* put message on sent list */
722 list_move_tail(&m
->list_head
, &con
->out_sent
);
725 * only assign outgoing seq # if we haven't sent this message
726 * yet. if it is requeued, resend with it's original seq.
728 if (m
->needs_out_seq
) {
729 m
->hdr
.seq
= cpu_to_le64(++con
->out_seq
);
730 m
->needs_out_seq
= false;
733 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n",
734 m
, con
->out_seq
, le16_to_cpu(m
->hdr
.type
),
735 le32_to_cpu(m
->hdr
.front_len
), le32_to_cpu(m
->hdr
.middle_len
),
736 le32_to_cpu(m
->hdr
.data_len
),
738 BUG_ON(le32_to_cpu(m
->hdr
.front_len
) != m
->front
.iov_len
);
740 /* tag + hdr + front + middle */
741 con_out_kvec_add(con
, sizeof (tag_msg
), &tag_msg
);
742 con_out_kvec_add(con
, sizeof (m
->hdr
), &m
->hdr
);
743 con_out_kvec_add(con
, m
->front
.iov_len
, m
->front
.iov_base
);
746 con_out_kvec_add(con
, m
->middle
->vec
.iov_len
,
747 m
->middle
->vec
.iov_base
);
749 /* fill in crc (except data pages), footer */
750 crc
= crc32c(0, &m
->hdr
, offsetof(struct ceph_msg_header
, crc
));
751 con
->out_msg
->hdr
.crc
= cpu_to_le32(crc
);
752 con
->out_msg
->footer
.flags
= 0;
754 crc
= crc32c(0, m
->front
.iov_base
, m
->front
.iov_len
);
755 con
->out_msg
->footer
.front_crc
= cpu_to_le32(crc
);
757 crc
= crc32c(0, m
->middle
->vec
.iov_base
,
758 m
->middle
->vec
.iov_len
);
759 con
->out_msg
->footer
.middle_crc
= cpu_to_le32(crc
);
761 con
->out_msg
->footer
.middle_crc
= 0;
762 dout("%s front_crc %u middle_crc %u\n", __func__
,
763 le32_to_cpu(con
->out_msg
->footer
.front_crc
),
764 le32_to_cpu(con
->out_msg
->footer
.middle_crc
));
766 /* is there a data payload? */
767 con
->out_msg
->footer
.data_crc
= 0;
769 prepare_write_message_data(con
);
771 /* no, queue up footer too and be done */
772 prepare_write_message_footer(con
);
774 set_bit(WRITE_PENDING
, &con
->flags
);
780 static void prepare_write_ack(struct ceph_connection
*con
)
782 dout("prepare_write_ack %p %llu -> %llu\n", con
,
783 con
->in_seq_acked
, con
->in_seq
);
784 con
->in_seq_acked
= con
->in_seq
;
786 con_out_kvec_reset(con
);
788 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
790 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
791 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
794 con
->out_more
= 1; /* more will follow.. eventually.. */
795 set_bit(WRITE_PENDING
, &con
->flags
);
799 * Prepare to write keepalive byte.
801 static void prepare_write_keepalive(struct ceph_connection
*con
)
803 dout("prepare_write_keepalive %p\n", con
);
804 con_out_kvec_reset(con
);
805 con_out_kvec_add(con
, sizeof (tag_keepalive
), &tag_keepalive
);
806 set_bit(WRITE_PENDING
, &con
->flags
);
810 * Connection negotiation.
813 static struct ceph_auth_handshake
*get_connect_authorizer(struct ceph_connection
*con
,
816 struct ceph_auth_handshake
*auth
;
818 if (!con
->ops
->get_authorizer
) {
819 con
->out_connect
.authorizer_protocol
= CEPH_AUTH_UNKNOWN
;
820 con
->out_connect
.authorizer_len
= 0;
825 /* Can't hold the mutex while getting authorizer */
827 mutex_unlock(&con
->mutex
);
829 auth
= con
->ops
->get_authorizer(con
, auth_proto
, con
->auth_retry
);
831 mutex_lock(&con
->mutex
);
835 if (test_bit(CLOSED
, &con
->state
) || test_bit(OPENING
, &con
->flags
))
836 return ERR_PTR(-EAGAIN
);
838 con
->auth_reply_buf
= auth
->authorizer_reply_buf
;
839 con
->auth_reply_buf_len
= auth
->authorizer_reply_buf_len
;
846 * We connected to a peer and are saying hello.
848 static void prepare_write_banner(struct ceph_connection
*con
)
850 con_out_kvec_add(con
, strlen(CEPH_BANNER
), CEPH_BANNER
);
851 con_out_kvec_add(con
, sizeof (con
->msgr
->my_enc_addr
),
852 &con
->msgr
->my_enc_addr
);
855 set_bit(WRITE_PENDING
, &con
->flags
);
858 static int prepare_write_connect(struct ceph_connection
*con
)
860 unsigned int global_seq
= get_global_seq(con
->msgr
, 0);
863 struct ceph_auth_handshake
*auth
;
865 switch (con
->peer_name
.type
) {
866 case CEPH_ENTITY_TYPE_MON
:
867 proto
= CEPH_MONC_PROTOCOL
;
869 case CEPH_ENTITY_TYPE_OSD
:
870 proto
= CEPH_OSDC_PROTOCOL
;
872 case CEPH_ENTITY_TYPE_MDS
:
873 proto
= CEPH_MDSC_PROTOCOL
;
879 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con
,
880 con
->connect_seq
, global_seq
, proto
);
882 con
->out_connect
.features
= cpu_to_le64(con
->msgr
->supported_features
);
883 con
->out_connect
.host_type
= cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT
);
884 con
->out_connect
.connect_seq
= cpu_to_le32(con
->connect_seq
);
885 con
->out_connect
.global_seq
= cpu_to_le32(global_seq
);
886 con
->out_connect
.protocol_version
= cpu_to_le32(proto
);
887 con
->out_connect
.flags
= 0;
889 auth_proto
= CEPH_AUTH_UNKNOWN
;
890 auth
= get_connect_authorizer(con
, &auth_proto
);
892 return PTR_ERR(auth
);
894 con
->out_connect
.authorizer_protocol
= cpu_to_le32(auth_proto
);
895 con
->out_connect
.authorizer_len
= auth
?
896 cpu_to_le32(auth
->authorizer_buf_len
) : 0;
898 con_out_kvec_reset(con
);
899 con_out_kvec_add(con
, sizeof (con
->out_connect
),
901 if (auth
&& auth
->authorizer_buf_len
)
902 con_out_kvec_add(con
, auth
->authorizer_buf_len
,
903 auth
->authorizer_buf
);
906 set_bit(WRITE_PENDING
, &con
->flags
);
912 * write as much of pending kvecs to the socket as we can.
914 * 0 -> socket full, but more to do
917 static int write_partial_kvec(struct ceph_connection
*con
)
921 dout("write_partial_kvec %p %d left\n", con
, con
->out_kvec_bytes
);
922 while (con
->out_kvec_bytes
> 0) {
923 ret
= ceph_tcp_sendmsg(con
->sock
, con
->out_kvec_cur
,
924 con
->out_kvec_left
, con
->out_kvec_bytes
,
928 con
->out_kvec_bytes
-= ret
;
929 if (con
->out_kvec_bytes
== 0)
932 /* account for full iov entries consumed */
933 while (ret
>= con
->out_kvec_cur
->iov_len
) {
934 BUG_ON(!con
->out_kvec_left
);
935 ret
-= con
->out_kvec_cur
->iov_len
;
937 con
->out_kvec_left
--;
939 /* and for a partially-consumed entry */
941 con
->out_kvec_cur
->iov_len
-= ret
;
942 con
->out_kvec_cur
->iov_base
+= ret
;
945 con
->out_kvec_left
= 0;
946 con
->out_kvec_is_msg
= false;
949 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con
,
950 con
->out_kvec_bytes
, con
->out_kvec_left
, ret
);
951 return ret
; /* done! */
954 static void out_msg_pos_next(struct ceph_connection
*con
, struct page
*page
,
955 size_t len
, size_t sent
, bool in_trail
)
957 struct ceph_msg
*msg
= con
->out_msg
;
962 con
->out_msg_pos
.data_pos
+= sent
;
963 con
->out_msg_pos
.page_pos
+= sent
;
968 con
->out_msg_pos
.page_pos
= 0;
969 con
->out_msg_pos
.page
++;
970 con
->out_msg_pos
.did_page_crc
= false;
972 list_move_tail(&page
->lru
,
974 else if (msg
->pagelist
)
975 list_move_tail(&page
->lru
,
976 &msg
->pagelist
->head
);
979 iter_bio_next(&msg
->bio_iter
, &msg
->bio_seg
);
984 * Write as much message data payload as we can. If we finish, queue
986 * 1 -> done, footer is now queued in out_kvec[].
987 * 0 -> socket full, but more to do
990 static int write_partial_msg_pages(struct ceph_connection
*con
)
992 struct ceph_msg
*msg
= con
->out_msg
;
993 unsigned int data_len
= le32_to_cpu(msg
->hdr
.data_len
);
995 bool do_datacrc
= !con
->msgr
->nocrc
;
998 bool in_trail
= false;
999 const size_t trail_len
= (msg
->trail
? msg
->trail
->length
: 0);
1000 const size_t trail_off
= data_len
- trail_len
;
1002 dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
1003 con
, msg
, con
->out_msg_pos
.page
, msg
->nr_pages
,
1004 con
->out_msg_pos
.page_pos
);
1007 * Iterate through each page that contains data to be
1008 * written, and send as much as possible for each.
1010 * If we are calculating the data crc (the default), we will
1011 * need to map the page. If we have no pages, they have
1012 * been revoked, so use the zero page.
1014 while (data_len
> con
->out_msg_pos
.data_pos
) {
1015 struct page
*page
= NULL
;
1016 int max_write
= PAGE_SIZE
;
1019 in_trail
= in_trail
|| con
->out_msg_pos
.data_pos
>= trail_off
;
1021 total_max_write
= trail_off
- con
->out_msg_pos
.data_pos
;
1024 total_max_write
= data_len
- con
->out_msg_pos
.data_pos
;
1026 page
= list_first_entry(&msg
->trail
->head
,
1028 } else if (msg
->pages
) {
1029 page
= msg
->pages
[con
->out_msg_pos
.page
];
1030 } else if (msg
->pagelist
) {
1031 page
= list_first_entry(&msg
->pagelist
->head
,
1034 } else if (msg
->bio
) {
1037 bv
= bio_iovec_idx(msg
->bio_iter
, msg
->bio_seg
);
1039 bio_offset
= bv
->bv_offset
;
1040 max_write
= bv
->bv_len
;
1045 len
= min_t(int, max_write
- con
->out_msg_pos
.page_pos
,
1048 if (do_datacrc
&& !con
->out_msg_pos
.did_page_crc
) {
1050 u32 crc
= le32_to_cpu(msg
->footer
.data_crc
);
1054 BUG_ON(kaddr
== NULL
);
1055 base
= kaddr
+ con
->out_msg_pos
.page_pos
+ bio_offset
;
1056 crc
= crc32c(crc
, base
, len
);
1057 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1058 con
->out_msg_pos
.did_page_crc
= true;
1060 ret
= ceph_tcp_sendpage(con
->sock
, page
,
1061 con
->out_msg_pos
.page_pos
+ bio_offset
,
1070 out_msg_pos_next(con
, page
, len
, (size_t) ret
, in_trail
);
1073 dout("write_partial_msg_pages %p msg %p done\n", con
, msg
);
1075 /* prepare and queue up footer, too */
1077 msg
->footer
.flags
|= CEPH_MSG_FOOTER_NOCRC
;
1078 con_out_kvec_reset(con
);
1079 prepare_write_message_footer(con
);
1088 static int write_partial_skip(struct ceph_connection
*con
)
1092 while (con
->out_skip
> 0) {
1093 size_t size
= min(con
->out_skip
, (int) PAGE_CACHE_SIZE
);
1095 ret
= ceph_tcp_sendpage(con
->sock
, zero_page
, 0, size
, 1);
1098 con
->out_skip
-= ret
;
1106 * Prepare to read connection handshake, or an ack.
1108 static void prepare_read_banner(struct ceph_connection
*con
)
1110 dout("prepare_read_banner %p\n", con
);
1111 con
->in_base_pos
= 0;
1114 static void prepare_read_connect(struct ceph_connection
*con
)
1116 dout("prepare_read_connect %p\n", con
);
1117 con
->in_base_pos
= 0;
1120 static void prepare_read_ack(struct ceph_connection
*con
)
1122 dout("prepare_read_ack %p\n", con
);
1123 con
->in_base_pos
= 0;
1126 static void prepare_read_tag(struct ceph_connection
*con
)
1128 dout("prepare_read_tag %p\n", con
);
1129 con
->in_base_pos
= 0;
1130 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1134 * Prepare to read a message.
1136 static int prepare_read_message(struct ceph_connection
*con
)
1138 dout("prepare_read_message %p\n", con
);
1139 BUG_ON(con
->in_msg
!= NULL
);
1140 con
->in_base_pos
= 0;
1141 con
->in_front_crc
= con
->in_middle_crc
= con
->in_data_crc
= 0;
1146 static int read_partial(struct ceph_connection
*con
,
1147 int end
, int size
, void *object
)
1149 while (con
->in_base_pos
< end
) {
1150 int left
= end
- con
->in_base_pos
;
1151 int have
= size
- left
;
1152 int ret
= ceph_tcp_recvmsg(con
->sock
, object
+ have
, left
);
1155 con
->in_base_pos
+= ret
;
1162 * Read all or part of the connect-side handshake on a new connection
1164 static int read_partial_banner(struct ceph_connection
*con
)
1170 dout("read_partial_banner %p at %d\n", con
, con
->in_base_pos
);
1173 size
= strlen(CEPH_BANNER
);
1175 ret
= read_partial(con
, end
, size
, con
->in_banner
);
1179 size
= sizeof (con
->actual_peer_addr
);
1181 ret
= read_partial(con
, end
, size
, &con
->actual_peer_addr
);
1185 size
= sizeof (con
->peer_addr_for_me
);
1187 ret
= read_partial(con
, end
, size
, &con
->peer_addr_for_me
);
1195 static int read_partial_connect(struct ceph_connection
*con
)
1201 dout("read_partial_connect %p at %d\n", con
, con
->in_base_pos
);
1203 size
= sizeof (con
->in_reply
);
1205 ret
= read_partial(con
, end
, size
, &con
->in_reply
);
1209 size
= le32_to_cpu(con
->in_reply
.authorizer_len
);
1211 ret
= read_partial(con
, end
, size
, con
->auth_reply_buf
);
1215 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1216 con
, (int)con
->in_reply
.tag
,
1217 le32_to_cpu(con
->in_reply
.connect_seq
),
1218 le32_to_cpu(con
->in_reply
.global_seq
));
1225 * Verify the hello banner looks okay.
1227 static int verify_hello(struct ceph_connection
*con
)
1229 if (memcmp(con
->in_banner
, CEPH_BANNER
, strlen(CEPH_BANNER
))) {
1230 pr_err("connect to %s got bad banner\n",
1231 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1232 con
->error_msg
= "protocol error, bad banner";
1238 static bool addr_is_blank(struct sockaddr_storage
*ss
)
1240 switch (ss
->ss_family
) {
1242 return ((struct sockaddr_in
*)ss
)->sin_addr
.s_addr
== 0;
1245 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[0] == 0 &&
1246 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[1] == 0 &&
1247 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[2] == 0 &&
1248 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[3] == 0;
1253 static int addr_port(struct sockaddr_storage
*ss
)
1255 switch (ss
->ss_family
) {
1257 return ntohs(((struct sockaddr_in
*)ss
)->sin_port
);
1259 return ntohs(((struct sockaddr_in6
*)ss
)->sin6_port
);
1264 static void addr_set_port(struct sockaddr_storage
*ss
, int p
)
1266 switch (ss
->ss_family
) {
1268 ((struct sockaddr_in
*)ss
)->sin_port
= htons(p
);
1271 ((struct sockaddr_in6
*)ss
)->sin6_port
= htons(p
);
1277 * Unlike other *_pton function semantics, zero indicates success.
1279 static int ceph_pton(const char *str
, size_t len
, struct sockaddr_storage
*ss
,
1280 char delim
, const char **ipend
)
1282 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
1283 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
1285 memset(ss
, 0, sizeof(*ss
));
1287 if (in4_pton(str
, len
, (u8
*)&in4
->sin_addr
.s_addr
, delim
, ipend
)) {
1288 ss
->ss_family
= AF_INET
;
1292 if (in6_pton(str
, len
, (u8
*)&in6
->sin6_addr
.s6_addr
, delim
, ipend
)) {
1293 ss
->ss_family
= AF_INET6
;
1301 * Extract hostname string and resolve using kernel DNS facility.
1303 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1304 static int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1305 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1307 const char *end
, *delim_p
;
1308 char *colon_p
, *ip_addr
= NULL
;
1312 * The end of the hostname occurs immediately preceding the delimiter or
1313 * the port marker (':') where the delimiter takes precedence.
1315 delim_p
= memchr(name
, delim
, namelen
);
1316 colon_p
= memchr(name
, ':', namelen
);
1318 if (delim_p
&& colon_p
)
1319 end
= delim_p
< colon_p
? delim_p
: colon_p
;
1320 else if (!delim_p
&& colon_p
)
1324 if (!end
) /* case: hostname:/ */
1325 end
= name
+ namelen
;
1331 /* do dns_resolve upcall */
1332 ip_len
= dns_query(NULL
, name
, end
- name
, NULL
, &ip_addr
, NULL
);
1334 ret
= ceph_pton(ip_addr
, ip_len
, ss
, -1, NULL
);
1342 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end
- name
), name
,
1343 ret
, ret
? "failed" : ceph_pr_addr(ss
));
1348 static inline int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1349 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1356 * Parse a server name (IP or hostname). If a valid IP address is not found
1357 * then try to extract a hostname to resolve using userspace DNS upcall.
1359 static int ceph_parse_server_name(const char *name
, size_t namelen
,
1360 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1364 ret
= ceph_pton(name
, namelen
, ss
, delim
, ipend
);
1366 ret
= ceph_dns_resolve_name(name
, namelen
, ss
, delim
, ipend
);
1372 * Parse an ip[:port] list into an addr array. Use the default
1373 * monitor port if a port isn't specified.
1375 int ceph_parse_ips(const char *c
, const char *end
,
1376 struct ceph_entity_addr
*addr
,
1377 int max_count
, int *count
)
1379 int i
, ret
= -EINVAL
;
1382 dout("parse_ips on '%.*s'\n", (int)(end
-c
), c
);
1383 for (i
= 0; i
< max_count
; i
++) {
1385 struct sockaddr_storage
*ss
= &addr
[i
].in_addr
;
1394 ret
= ceph_parse_server_name(p
, end
- p
, ss
, delim
, &ipend
);
1403 dout("missing matching ']'\n");
1410 if (p
< end
&& *p
== ':') {
1413 while (p
< end
&& *p
>= '0' && *p
<= '9') {
1414 port
= (port
* 10) + (*p
- '0');
1417 if (port
> 65535 || port
== 0)
1420 port
= CEPH_MON_PORT
;
1423 addr_set_port(ss
, port
);
1425 dout("parse_ips got %s\n", ceph_pr_addr(ss
));
1442 pr_err("parse_ips bad ip '%.*s'\n", (int)(end
- c
), c
);
1445 EXPORT_SYMBOL(ceph_parse_ips
);
1447 static int process_banner(struct ceph_connection
*con
)
1449 dout("process_banner on %p\n", con
);
1451 if (verify_hello(con
) < 0)
1454 ceph_decode_addr(&con
->actual_peer_addr
);
1455 ceph_decode_addr(&con
->peer_addr_for_me
);
1458 * Make sure the other end is who we wanted. note that the other
1459 * end may not yet know their ip address, so if it's 0.0.0.0, give
1460 * them the benefit of the doubt.
1462 if (memcmp(&con
->peer_addr
, &con
->actual_peer_addr
,
1463 sizeof(con
->peer_addr
)) != 0 &&
1464 !(addr_is_blank(&con
->actual_peer_addr
.in_addr
) &&
1465 con
->actual_peer_addr
.nonce
== con
->peer_addr
.nonce
)) {
1466 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1467 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1468 (int)le32_to_cpu(con
->peer_addr
.nonce
),
1469 ceph_pr_addr(&con
->actual_peer_addr
.in_addr
),
1470 (int)le32_to_cpu(con
->actual_peer_addr
.nonce
));
1471 con
->error_msg
= "wrong peer at address";
1476 * did we learn our address?
1478 if (addr_is_blank(&con
->msgr
->inst
.addr
.in_addr
)) {
1479 int port
= addr_port(&con
->msgr
->inst
.addr
.in_addr
);
1481 memcpy(&con
->msgr
->inst
.addr
.in_addr
,
1482 &con
->peer_addr_for_me
.in_addr
,
1483 sizeof(con
->peer_addr_for_me
.in_addr
));
1484 addr_set_port(&con
->msgr
->inst
.addr
.in_addr
, port
);
1485 encode_my_addr(con
->msgr
);
1486 dout("process_banner learned my addr is %s\n",
1487 ceph_pr_addr(&con
->msgr
->inst
.addr
.in_addr
));
1493 static void fail_protocol(struct ceph_connection
*con
)
1495 reset_connection(con
);
1496 set_bit(CLOSED
, &con
->state
); /* in case there's queued work */
1499 static int process_connect(struct ceph_connection
*con
)
1501 u64 sup_feat
= con
->msgr
->supported_features
;
1502 u64 req_feat
= con
->msgr
->required_features
;
1503 u64 server_feat
= le64_to_cpu(con
->in_reply
.features
);
1506 dout("process_connect on %p tag %d\n", con
, (int)con
->in_tag
);
1508 switch (con
->in_reply
.tag
) {
1509 case CEPH_MSGR_TAG_FEATURES
:
1510 pr_err("%s%lld %s feature set mismatch,"
1511 " my %llx < server's %llx, missing %llx\n",
1512 ENTITY_NAME(con
->peer_name
),
1513 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1514 sup_feat
, server_feat
, server_feat
& ~sup_feat
);
1515 con
->error_msg
= "missing required protocol features";
1519 case CEPH_MSGR_TAG_BADPROTOVER
:
1520 pr_err("%s%lld %s protocol version mismatch,"
1521 " my %d != server's %d\n",
1522 ENTITY_NAME(con
->peer_name
),
1523 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1524 le32_to_cpu(con
->out_connect
.protocol_version
),
1525 le32_to_cpu(con
->in_reply
.protocol_version
));
1526 con
->error_msg
= "protocol version mismatch";
1530 case CEPH_MSGR_TAG_BADAUTHORIZER
:
1532 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con
,
1534 if (con
->auth_retry
== 2) {
1535 con
->error_msg
= "connect authorization failure";
1538 con
->auth_retry
= 1;
1539 ret
= prepare_write_connect(con
);
1542 prepare_read_connect(con
);
1545 case CEPH_MSGR_TAG_RESETSESSION
:
1547 * If we connected with a large connect_seq but the peer
1548 * has no record of a session with us (no connection, or
1549 * connect_seq == 0), they will send RESETSESION to indicate
1550 * that they must have reset their session, and may have
1553 dout("process_connect got RESET peer seq %u\n",
1554 le32_to_cpu(con
->in_reply
.connect_seq
));
1555 pr_err("%s%lld %s connection reset\n",
1556 ENTITY_NAME(con
->peer_name
),
1557 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1558 reset_connection(con
);
1559 ret
= prepare_write_connect(con
);
1562 prepare_read_connect(con
);
1564 /* Tell ceph about it. */
1565 mutex_unlock(&con
->mutex
);
1566 pr_info("reset on %s%lld\n", ENTITY_NAME(con
->peer_name
));
1567 if (con
->ops
->peer_reset
)
1568 con
->ops
->peer_reset(con
);
1569 mutex_lock(&con
->mutex
);
1570 if (test_bit(CLOSED
, &con
->state
) ||
1571 test_bit(OPENING
, &con
->state
))
1575 case CEPH_MSGR_TAG_RETRY_SESSION
:
1577 * If we sent a smaller connect_seq than the peer has, try
1578 * again with a larger value.
1580 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
1581 le32_to_cpu(con
->out_connect
.connect_seq
),
1582 le32_to_cpu(con
->in_reply
.connect_seq
));
1583 con
->connect_seq
= le32_to_cpu(con
->in_reply
.connect_seq
);
1584 ret
= prepare_write_connect(con
);
1587 prepare_read_connect(con
);
1590 case CEPH_MSGR_TAG_RETRY_GLOBAL
:
1592 * If we sent a smaller global_seq than the peer has, try
1593 * again with a larger value.
1595 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1596 con
->peer_global_seq
,
1597 le32_to_cpu(con
->in_reply
.global_seq
));
1598 get_global_seq(con
->msgr
,
1599 le32_to_cpu(con
->in_reply
.global_seq
));
1600 ret
= prepare_write_connect(con
);
1603 prepare_read_connect(con
);
1606 case CEPH_MSGR_TAG_READY
:
1607 if (req_feat
& ~server_feat
) {
1608 pr_err("%s%lld %s protocol feature mismatch,"
1609 " my required %llx > server's %llx, need %llx\n",
1610 ENTITY_NAME(con
->peer_name
),
1611 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1612 req_feat
, server_feat
, req_feat
& ~server_feat
);
1613 con
->error_msg
= "missing required protocol features";
1617 clear_bit(NEGOTIATING
, &con
->state
);
1618 set_bit(CONNECTED
, &con
->state
);
1619 con
->peer_global_seq
= le32_to_cpu(con
->in_reply
.global_seq
);
1621 con
->peer_features
= server_feat
;
1622 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1623 con
->peer_global_seq
,
1624 le32_to_cpu(con
->in_reply
.connect_seq
),
1626 WARN_ON(con
->connect_seq
!=
1627 le32_to_cpu(con
->in_reply
.connect_seq
));
1629 if (con
->in_reply
.flags
& CEPH_MSG_CONNECT_LOSSY
)
1630 set_bit(LOSSYTX
, &con
->flags
);
1632 con
->delay
= 0; /* reset backoff memory */
1634 prepare_read_tag(con
);
1637 case CEPH_MSGR_TAG_WAIT
:
1639 * If there is a connection race (we are opening
1640 * connections to each other), one of us may just have
1641 * to WAIT. This shouldn't happen if we are the
1644 pr_err("process_connect got WAIT as client\n");
1645 con
->error_msg
= "protocol error, got WAIT as client";
1649 pr_err("connect protocol error, will retry\n");
1650 con
->error_msg
= "protocol error, garbage tag during connect";
1658 * read (part of) an ack
1660 static int read_partial_ack(struct ceph_connection
*con
)
1662 int size
= sizeof (con
->in_temp_ack
);
1665 return read_partial(con
, end
, size
, &con
->in_temp_ack
);
1670 * We can finally discard anything that's been acked.
1672 static void process_ack(struct ceph_connection
*con
)
1675 u64 ack
= le64_to_cpu(con
->in_temp_ack
);
1678 while (!list_empty(&con
->out_sent
)) {
1679 m
= list_first_entry(&con
->out_sent
, struct ceph_msg
,
1681 seq
= le64_to_cpu(m
->hdr
.seq
);
1684 dout("got ack for seq %llu type %d at %p\n", seq
,
1685 le16_to_cpu(m
->hdr
.type
), m
);
1686 m
->ack_stamp
= jiffies
;
1689 prepare_read_tag(con
);
1695 static int read_partial_message_section(struct ceph_connection
*con
,
1696 struct kvec
*section
,
1697 unsigned int sec_len
, u32
*crc
)
1703 while (section
->iov_len
< sec_len
) {
1704 BUG_ON(section
->iov_base
== NULL
);
1705 left
= sec_len
- section
->iov_len
;
1706 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)section
->iov_base
+
1707 section
->iov_len
, left
);
1710 section
->iov_len
+= ret
;
1712 if (section
->iov_len
== sec_len
)
1713 *crc
= crc32c(0, section
->iov_base
, section
->iov_len
);
1718 static bool ceph_con_in_msg_alloc(struct ceph_connection
*con
,
1719 struct ceph_msg_header
*hdr
);
1722 static int read_partial_message_pages(struct ceph_connection
*con
,
1723 struct page
**pages
,
1724 unsigned int data_len
, bool do_datacrc
)
1730 left
= min((int)(data_len
- con
->in_msg_pos
.data_pos
),
1731 (int)(PAGE_SIZE
- con
->in_msg_pos
.page_pos
));
1733 BUG_ON(pages
== NULL
);
1734 p
= kmap(pages
[con
->in_msg_pos
.page
]);
1735 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ con
->in_msg_pos
.page_pos
,
1737 if (ret
> 0 && do_datacrc
)
1739 crc32c(con
->in_data_crc
,
1740 p
+ con
->in_msg_pos
.page_pos
, ret
);
1741 kunmap(pages
[con
->in_msg_pos
.page
]);
1744 con
->in_msg_pos
.data_pos
+= ret
;
1745 con
->in_msg_pos
.page_pos
+= ret
;
1746 if (con
->in_msg_pos
.page_pos
== PAGE_SIZE
) {
1747 con
->in_msg_pos
.page_pos
= 0;
1748 con
->in_msg_pos
.page
++;
1755 static int read_partial_message_bio(struct ceph_connection
*con
,
1756 struct bio
**bio_iter
, int *bio_seg
,
1757 unsigned int data_len
, bool do_datacrc
)
1759 struct bio_vec
*bv
= bio_iovec_idx(*bio_iter
, *bio_seg
);
1763 left
= min((int)(data_len
- con
->in_msg_pos
.data_pos
),
1764 (int)(bv
->bv_len
- con
->in_msg_pos
.page_pos
));
1766 p
= kmap(bv
->bv_page
) + bv
->bv_offset
;
1768 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ con
->in_msg_pos
.page_pos
,
1770 if (ret
> 0 && do_datacrc
)
1772 crc32c(con
->in_data_crc
,
1773 p
+ con
->in_msg_pos
.page_pos
, ret
);
1774 kunmap(bv
->bv_page
);
1777 con
->in_msg_pos
.data_pos
+= ret
;
1778 con
->in_msg_pos
.page_pos
+= ret
;
1779 if (con
->in_msg_pos
.page_pos
== bv
->bv_len
) {
1780 con
->in_msg_pos
.page_pos
= 0;
1781 iter_bio_next(bio_iter
, bio_seg
);
1789 * read (part of) a message.
1791 static int read_partial_message(struct ceph_connection
*con
)
1793 struct ceph_msg
*m
= con
->in_msg
;
1797 unsigned int front_len
, middle_len
, data_len
;
1798 bool do_datacrc
= !con
->msgr
->nocrc
;
1802 dout("read_partial_message con %p msg %p\n", con
, m
);
1805 size
= sizeof (con
->in_hdr
);
1807 ret
= read_partial(con
, end
, size
, &con
->in_hdr
);
1811 crc
= crc32c(0, &con
->in_hdr
, offsetof(struct ceph_msg_header
, crc
));
1812 if (cpu_to_le32(crc
) != con
->in_hdr
.crc
) {
1813 pr_err("read_partial_message bad hdr "
1814 " crc %u != expected %u\n",
1815 crc
, con
->in_hdr
.crc
);
1819 front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
1820 if (front_len
> CEPH_MSG_MAX_FRONT_LEN
)
1822 middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
1823 if (middle_len
> CEPH_MSG_MAX_DATA_LEN
)
1825 data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
1826 if (data_len
> CEPH_MSG_MAX_DATA_LEN
)
1830 seq
= le64_to_cpu(con
->in_hdr
.seq
);
1831 if ((s64
)seq
- (s64
)con
->in_seq
< 1) {
1832 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1833 ENTITY_NAME(con
->peer_name
),
1834 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1835 seq
, con
->in_seq
+ 1);
1836 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1838 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1840 } else if ((s64
)seq
- (s64
)con
->in_seq
> 1) {
1841 pr_err("read_partial_message bad seq %lld expected %lld\n",
1842 seq
, con
->in_seq
+ 1);
1843 con
->error_msg
= "bad message sequence # for incoming message";
1847 /* allocate message? */
1849 dout("got hdr type %d front %d data %d\n", con
->in_hdr
.type
,
1850 con
->in_hdr
.front_len
, con
->in_hdr
.data_len
);
1851 if (ceph_con_in_msg_alloc(con
, &con
->in_hdr
)) {
1852 /* skip this message */
1853 dout("alloc_msg said skip message\n");
1854 BUG_ON(con
->in_msg
);
1855 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1857 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1863 "error allocating memory for incoming message";
1867 BUG_ON(con
->in_msg
->con
!= con
);
1869 m
->front
.iov_len
= 0; /* haven't read it yet */
1871 m
->middle
->vec
.iov_len
= 0;
1873 con
->in_msg_pos
.page
= 0;
1875 con
->in_msg_pos
.page_pos
= m
->page_alignment
;
1877 con
->in_msg_pos
.page_pos
= 0;
1878 con
->in_msg_pos
.data_pos
= 0;
1882 init_bio_iter(m
->bio
, &m
->bio_iter
, &m
->bio_seg
);
1887 ret
= read_partial_message_section(con
, &m
->front
, front_len
,
1888 &con
->in_front_crc
);
1894 ret
= read_partial_message_section(con
, &m
->middle
->vec
,
1896 &con
->in_middle_crc
);
1902 while (con
->in_msg_pos
.data_pos
< data_len
) {
1904 ret
= read_partial_message_pages(con
, m
->pages
,
1905 data_len
, do_datacrc
);
1909 } else if (m
->bio
) {
1910 BUG_ON(!m
->bio_iter
);
1911 ret
= read_partial_message_bio(con
,
1912 &m
->bio_iter
, &m
->bio_seg
,
1913 data_len
, do_datacrc
);
1923 size
= sizeof (m
->footer
);
1925 ret
= read_partial(con
, end
, size
, &m
->footer
);
1929 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1930 m
, front_len
, m
->footer
.front_crc
, middle_len
,
1931 m
->footer
.middle_crc
, data_len
, m
->footer
.data_crc
);
1934 if (con
->in_front_crc
!= le32_to_cpu(m
->footer
.front_crc
)) {
1935 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1936 m
, con
->in_front_crc
, m
->footer
.front_crc
);
1939 if (con
->in_middle_crc
!= le32_to_cpu(m
->footer
.middle_crc
)) {
1940 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1941 m
, con
->in_middle_crc
, m
->footer
.middle_crc
);
1945 (m
->footer
.flags
& CEPH_MSG_FOOTER_NOCRC
) == 0 &&
1946 con
->in_data_crc
!= le32_to_cpu(m
->footer
.data_crc
)) {
1947 pr_err("read_partial_message %p data crc %u != exp. %u\n", m
,
1948 con
->in_data_crc
, le32_to_cpu(m
->footer
.data_crc
));
1952 return 1; /* done! */
1956 * Process message. This happens in the worker thread. The callback should
1957 * be careful not to do anything that waits on other incoming messages or it
1960 static void process_message(struct ceph_connection
*con
)
1962 struct ceph_msg
*msg
;
1964 BUG_ON(con
->in_msg
->con
!= con
);
1965 con
->in_msg
->con
= NULL
;
1970 /* if first message, set peer_name */
1971 if (con
->peer_name
.type
== 0)
1972 con
->peer_name
= msg
->hdr
.src
;
1975 mutex_unlock(&con
->mutex
);
1977 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
1978 msg
, le64_to_cpu(msg
->hdr
.seq
),
1979 ENTITY_NAME(msg
->hdr
.src
),
1980 le16_to_cpu(msg
->hdr
.type
),
1981 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
1982 le32_to_cpu(msg
->hdr
.front_len
),
1983 le32_to_cpu(msg
->hdr
.data_len
),
1984 con
->in_front_crc
, con
->in_middle_crc
, con
->in_data_crc
);
1985 con
->ops
->dispatch(con
, msg
);
1987 mutex_lock(&con
->mutex
);
1988 prepare_read_tag(con
);
1993 * Write something to the socket. Called in a worker thread when the
1994 * socket appears to be writeable and we have something ready to send.
1996 static int try_write(struct ceph_connection
*con
)
2000 dout("try_write start %p state %lu\n", con
, con
->state
);
2003 dout("try_write out_kvec_bytes %d\n", con
->out_kvec_bytes
);
2005 /* open the socket first? */
2006 if (con
->sock
== NULL
) {
2007 set_bit(CONNECTING
, &con
->state
);
2009 con_out_kvec_reset(con
);
2010 prepare_write_banner(con
);
2011 prepare_read_banner(con
);
2013 BUG_ON(con
->in_msg
);
2014 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2015 dout("try_write initiating connect on %p new state %lu\n",
2017 ret
= ceph_tcp_connect(con
);
2019 con
->error_msg
= "connect error";
2025 /* kvec data queued? */
2026 if (con
->out_skip
) {
2027 ret
= write_partial_skip(con
);
2031 if (con
->out_kvec_left
) {
2032 ret
= write_partial_kvec(con
);
2039 if (con
->out_msg_done
) {
2040 ceph_msg_put(con
->out_msg
);
2041 con
->out_msg
= NULL
; /* we're done with this one */
2045 ret
= write_partial_msg_pages(con
);
2047 goto more_kvec
; /* we need to send the footer, too! */
2051 dout("try_write write_partial_msg_pages err %d\n",
2058 if (!test_bit(CONNECTING
, &con
->state
) &&
2059 !test_bit(NEGOTIATING
, &con
->state
)) {
2060 /* is anything else pending? */
2061 if (!list_empty(&con
->out_queue
)) {
2062 prepare_write_message(con
);
2065 if (con
->in_seq
> con
->in_seq_acked
) {
2066 prepare_write_ack(con
);
2069 if (test_and_clear_bit(KEEPALIVE_PENDING
, &con
->flags
)) {
2070 prepare_write_keepalive(con
);
2075 /* Nothing to do! */
2076 clear_bit(WRITE_PENDING
, &con
->flags
);
2077 dout("try_write nothing else to write.\n");
2080 dout("try_write done on %p ret %d\n", con
, ret
);
2087 * Read what we can from the socket.
2089 static int try_read(struct ceph_connection
*con
)
2096 if (test_bit(STANDBY
, &con
->state
))
2099 dout("try_read start on %p\n", con
);
2102 dout("try_read tag %d in_base_pos %d\n", (int)con
->in_tag
,
2106 * process_connect and process_message drop and re-take
2107 * con->mutex. make sure we handle a racing close or reopen.
2109 if (test_bit(CLOSED
, &con
->state
) ||
2110 test_bit(OPENING
, &con
->state
)) {
2115 if (test_bit(CONNECTING
, &con
->state
)) {
2116 dout("try_read connecting\n");
2117 ret
= read_partial_banner(con
);
2120 ret
= process_banner(con
);
2124 clear_bit(CONNECTING
, &con
->state
);
2125 set_bit(NEGOTIATING
, &con
->state
);
2127 /* Banner is good, exchange connection info */
2128 ret
= prepare_write_connect(con
);
2131 prepare_read_connect(con
);
2133 /* Send connection info before awaiting response */
2137 if (test_bit(NEGOTIATING
, &con
->state
)) {
2138 dout("try_read negotiating\n");
2139 ret
= read_partial_connect(con
);
2142 ret
= process_connect(con
);
2148 if (con
->in_base_pos
< 0) {
2150 * skipping + discarding content.
2152 * FIXME: there must be a better way to do this!
2154 static char buf
[SKIP_BUF_SIZE
];
2155 int skip
= min((int) sizeof (buf
), -con
->in_base_pos
);
2157 dout("skipping %d / %d bytes\n", skip
, -con
->in_base_pos
);
2158 ret
= ceph_tcp_recvmsg(con
->sock
, buf
, skip
);
2161 con
->in_base_pos
+= ret
;
2162 if (con
->in_base_pos
)
2165 if (con
->in_tag
== CEPH_MSGR_TAG_READY
) {
2169 ret
= ceph_tcp_recvmsg(con
->sock
, &con
->in_tag
, 1);
2172 dout("try_read got tag %d\n", (int)con
->in_tag
);
2173 switch (con
->in_tag
) {
2174 case CEPH_MSGR_TAG_MSG
:
2175 prepare_read_message(con
);
2177 case CEPH_MSGR_TAG_ACK
:
2178 prepare_read_ack(con
);
2180 case CEPH_MSGR_TAG_CLOSE
:
2181 clear_bit(CONNECTED
, &con
->state
);
2182 set_bit(CLOSED
, &con
->state
); /* fixme */
2188 if (con
->in_tag
== CEPH_MSGR_TAG_MSG
) {
2189 ret
= read_partial_message(con
);
2193 con
->error_msg
= "bad crc";
2197 con
->error_msg
= "io error";
2202 if (con
->in_tag
== CEPH_MSGR_TAG_READY
)
2204 process_message(con
);
2207 if (con
->in_tag
== CEPH_MSGR_TAG_ACK
) {
2208 ret
= read_partial_ack(con
);
2216 dout("try_read done on %p ret %d\n", con
, ret
);
2220 pr_err("try_read bad con->in_tag = %d\n", (int)con
->in_tag
);
2221 con
->error_msg
= "protocol error, garbage tag";
2228 * Atomically queue work on a connection. Bump @con reference to
2229 * avoid races with connection teardown.
2231 static void queue_con(struct ceph_connection
*con
)
2233 if (!con
->ops
->get(con
)) {
2234 dout("queue_con %p ref count 0\n", con
);
2238 if (!queue_delayed_work(ceph_msgr_wq
, &con
->work
, 0)) {
2239 dout("queue_con %p - already queued\n", con
);
2242 dout("queue_con %p\n", con
);
2247 * Do some work on a connection. Drop a connection ref when we're done.
2249 static void con_work(struct work_struct
*work
)
2251 struct ceph_connection
*con
= container_of(work
, struct ceph_connection
,
2255 mutex_lock(&con
->mutex
);
2257 if (test_and_clear_bit(SOCK_CLOSED
, &con
->flags
)) {
2258 if (test_and_clear_bit(CONNECTED
, &con
->state
))
2259 con
->error_msg
= "socket closed";
2260 else if (test_and_clear_bit(NEGOTIATING
, &con
->state
))
2261 con
->error_msg
= "negotiation failed";
2262 else if (test_and_clear_bit(CONNECTING
, &con
->state
))
2263 con
->error_msg
= "connection failed";
2265 con
->error_msg
= "unrecognized con state";
2269 if (test_and_clear_bit(BACKOFF
, &con
->flags
)) {
2270 dout("con_work %p backing off\n", con
);
2271 if (queue_delayed_work(ceph_msgr_wq
, &con
->work
,
2272 round_jiffies_relative(con
->delay
))) {
2273 dout("con_work %p backoff %lu\n", con
, con
->delay
);
2274 mutex_unlock(&con
->mutex
);
2278 dout("con_work %p FAILED to back off %lu\n", con
,
2283 if (test_bit(STANDBY
, &con
->state
)) {
2284 dout("con_work %p STANDBY\n", con
);
2287 if (test_bit(CLOSED
, &con
->state
)) {
2288 dout("con_work %p CLOSED\n", con
);
2292 if (test_and_clear_bit(OPENING
, &con
->state
)) {
2293 /* reopen w/ new peer */
2294 dout("con_work OPENING\n");
2298 ret
= try_read(con
);
2302 con
->error_msg
= "socket error on read";
2306 ret
= try_write(con
);
2310 con
->error_msg
= "socket error on write";
2315 mutex_unlock(&con
->mutex
);
2321 mutex_unlock(&con
->mutex
);
2322 ceph_fault(con
); /* error/fault path */
2328 * Generic error/fault handler. A retry mechanism is used with
2329 * exponential backoff
2331 static void ceph_fault(struct ceph_connection
*con
)
2333 mutex_lock(&con
->mutex
);
2335 pr_err("%s%lld %s %s\n", ENTITY_NAME(con
->peer_name
),
2336 ceph_pr_addr(&con
->peer_addr
.in_addr
), con
->error_msg
);
2337 dout("fault %p state %lu to peer %s\n",
2338 con
, con
->state
, ceph_pr_addr(&con
->peer_addr
.in_addr
));
2340 if (test_bit(CLOSED
, &con
->state
))
2343 con_close_socket(con
);
2345 if (test_bit(LOSSYTX
, &con
->flags
)) {
2346 dout("fault on LOSSYTX channel\n");
2351 BUG_ON(con
->in_msg
->con
!= con
);
2352 con
->in_msg
->con
= NULL
;
2353 ceph_msg_put(con
->in_msg
);
2358 /* Requeue anything that hasn't been acked */
2359 list_splice_init(&con
->out_sent
, &con
->out_queue
);
2361 /* If there are no messages queued or keepalive pending, place
2362 * the connection in a STANDBY state */
2363 if (list_empty(&con
->out_queue
) &&
2364 !test_bit(KEEPALIVE_PENDING
, &con
->flags
)) {
2365 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con
);
2366 clear_bit(WRITE_PENDING
, &con
->flags
);
2367 set_bit(STANDBY
, &con
->state
);
2369 /* retry after a delay. */
2370 if (con
->delay
== 0)
2371 con
->delay
= BASE_DELAY_INTERVAL
;
2372 else if (con
->delay
< MAX_DELAY_INTERVAL
)
2375 if (queue_delayed_work(ceph_msgr_wq
, &con
->work
,
2376 round_jiffies_relative(con
->delay
))) {
2377 dout("fault queued %p delay %lu\n", con
, con
->delay
);
2380 dout("fault failed to queue %p delay %lu, backoff\n",
2383 * In many cases we see a socket state change
2384 * while con_work is running and end up
2385 * queuing (non-delayed) work, such that we
2386 * can't backoff with a delay. Set a flag so
2387 * that when con_work restarts we schedule the
2390 set_bit(BACKOFF
, &con
->flags
);
2395 mutex_unlock(&con
->mutex
);
2397 * in case we faulted due to authentication, invalidate our
2398 * current tickets so that we can get new ones.
2400 if (con
->auth_retry
&& con
->ops
->invalidate_authorizer
) {
2401 dout("calling invalidate_authorizer()\n");
2402 con
->ops
->invalidate_authorizer(con
);
2405 if (con
->ops
->fault
)
2406 con
->ops
->fault(con
);
2412 * initialize a new messenger instance
2414 void ceph_messenger_init(struct ceph_messenger
*msgr
,
2415 struct ceph_entity_addr
*myaddr
,
2416 u32 supported_features
,
2417 u32 required_features
,
2420 msgr
->supported_features
= supported_features
;
2421 msgr
->required_features
= required_features
;
2423 spin_lock_init(&msgr
->global_seq_lock
);
2426 msgr
->inst
.addr
= *myaddr
;
2428 /* select a random nonce */
2429 msgr
->inst
.addr
.type
= 0;
2430 get_random_bytes(&msgr
->inst
.addr
.nonce
, sizeof(msgr
->inst
.addr
.nonce
));
2431 encode_my_addr(msgr
);
2432 msgr
->nocrc
= nocrc
;
2434 atomic_set(&msgr
->stopping
, 0);
2436 dout("%s %p\n", __func__
, msgr
);
2438 EXPORT_SYMBOL(ceph_messenger_init
);
2440 static void clear_standby(struct ceph_connection
*con
)
2442 /* come back from STANDBY? */
2443 if (test_and_clear_bit(STANDBY
, &con
->state
)) {
2444 dout("clear_standby %p and ++connect_seq\n", con
);
2446 WARN_ON(test_bit(WRITE_PENDING
, &con
->flags
));
2447 WARN_ON(test_bit(KEEPALIVE_PENDING
, &con
->flags
));
2452 * Queue up an outgoing message on the given connection.
2454 void ceph_con_send(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2457 msg
->hdr
.src
= con
->msgr
->inst
.name
;
2458 BUG_ON(msg
->front
.iov_len
!= le32_to_cpu(msg
->hdr
.front_len
));
2459 msg
->needs_out_seq
= true;
2461 mutex_lock(&con
->mutex
);
2463 if (test_bit(CLOSED
, &con
->state
)) {
2464 dout("con_send %p closed, dropping %p\n", con
, msg
);
2466 mutex_unlock(&con
->mutex
);
2470 BUG_ON(msg
->con
!= NULL
);
2471 msg
->con
= con
->ops
->get(con
);
2472 BUG_ON(msg
->con
== NULL
);
2474 BUG_ON(!list_empty(&msg
->list_head
));
2475 list_add_tail(&msg
->list_head
, &con
->out_queue
);
2476 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg
,
2477 ENTITY_NAME(con
->peer_name
), le16_to_cpu(msg
->hdr
.type
),
2478 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2479 le32_to_cpu(msg
->hdr
.front_len
),
2480 le32_to_cpu(msg
->hdr
.middle_len
),
2481 le32_to_cpu(msg
->hdr
.data_len
));
2484 mutex_unlock(&con
->mutex
);
2486 /* if there wasn't anything waiting to send before, queue
2488 if (test_and_set_bit(WRITE_PENDING
, &con
->flags
) == 0)
2491 EXPORT_SYMBOL(ceph_con_send
);
2494 * Revoke a message that was previously queued for send
2496 void ceph_msg_revoke(struct ceph_msg
*msg
)
2498 struct ceph_connection
*con
= msg
->con
;
2501 return; /* Message not in our possession */
2503 mutex_lock(&con
->mutex
);
2504 if (!list_empty(&msg
->list_head
)) {
2505 dout("%s %p msg %p - was on queue\n", __func__
, con
, msg
);
2506 list_del_init(&msg
->list_head
);
2507 BUG_ON(msg
->con
== NULL
);
2508 msg
->con
->ops
->put(msg
->con
);
2514 if (con
->out_msg
== msg
) {
2515 dout("%s %p msg %p - was sending\n", __func__
, con
, msg
);
2516 con
->out_msg
= NULL
;
2517 if (con
->out_kvec_is_msg
) {
2518 con
->out_skip
= con
->out_kvec_bytes
;
2519 con
->out_kvec_is_msg
= false;
2525 mutex_unlock(&con
->mutex
);
2529 * Revoke a message that we may be reading data into
2531 void ceph_msg_revoke_incoming(struct ceph_msg
*msg
)
2533 struct ceph_connection
*con
;
2535 BUG_ON(msg
== NULL
);
2537 dout("%s msg %p null con\n", __func__
, msg
);
2539 return; /* Message not in our possession */
2543 mutex_lock(&con
->mutex
);
2544 if (con
->in_msg
== msg
) {
2545 unsigned int front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
2546 unsigned int middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
2547 unsigned int data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
2549 /* skip rest of message */
2550 dout("%s %p msg %p revoked\n", __func__
, con
, msg
);
2551 con
->in_base_pos
= con
->in_base_pos
-
2552 sizeof(struct ceph_msg_header
) -
2556 sizeof(struct ceph_msg_footer
);
2557 ceph_msg_put(con
->in_msg
);
2559 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2562 dout("%s %p in_msg %p msg %p no-op\n",
2563 __func__
, con
, con
->in_msg
, msg
);
2565 mutex_unlock(&con
->mutex
);
2569 * Queue a keepalive byte to ensure the tcp connection is alive.
2571 void ceph_con_keepalive(struct ceph_connection
*con
)
2573 dout("con_keepalive %p\n", con
);
2574 mutex_lock(&con
->mutex
);
2576 mutex_unlock(&con
->mutex
);
2577 if (test_and_set_bit(KEEPALIVE_PENDING
, &con
->flags
) == 0 &&
2578 test_and_set_bit(WRITE_PENDING
, &con
->flags
) == 0)
2581 EXPORT_SYMBOL(ceph_con_keepalive
);
2585 * construct a new message with given type, size
2586 * the new msg has a ref count of 1.
2588 struct ceph_msg
*ceph_msg_new(int type
, int front_len
, gfp_t flags
,
2593 m
= kmalloc(sizeof(*m
), flags
);
2596 kref_init(&m
->kref
);
2599 INIT_LIST_HEAD(&m
->list_head
);
2602 m
->hdr
.type
= cpu_to_le16(type
);
2603 m
->hdr
.priority
= cpu_to_le16(CEPH_MSG_PRIO_DEFAULT
);
2605 m
->hdr
.front_len
= cpu_to_le32(front_len
);
2606 m
->hdr
.middle_len
= 0;
2607 m
->hdr
.data_len
= 0;
2608 m
->hdr
.data_off
= 0;
2609 m
->hdr
.reserved
= 0;
2610 m
->footer
.front_crc
= 0;
2611 m
->footer
.middle_crc
= 0;
2612 m
->footer
.data_crc
= 0;
2613 m
->footer
.flags
= 0;
2614 m
->front_max
= front_len
;
2615 m
->front_is_vmalloc
= false;
2616 m
->more_to_follow
= false;
2625 m
->page_alignment
= 0;
2635 if (front_len
> PAGE_CACHE_SIZE
) {
2636 m
->front
.iov_base
= __vmalloc(front_len
, flags
,
2638 m
->front_is_vmalloc
= true;
2640 m
->front
.iov_base
= kmalloc(front_len
, flags
);
2642 if (m
->front
.iov_base
== NULL
) {
2643 dout("ceph_msg_new can't allocate %d bytes\n",
2648 m
->front
.iov_base
= NULL
;
2650 m
->front
.iov_len
= front_len
;
2652 dout("ceph_msg_new %p front %d\n", m
, front_len
);
2659 pr_err("msg_new can't create type %d front %d\n", type
,
2663 dout("msg_new can't create type %d front %d\n", type
,
2668 EXPORT_SYMBOL(ceph_msg_new
);
2671 * Allocate "middle" portion of a message, if it is needed and wasn't
2672 * allocated by alloc_msg. This allows us to read a small fixed-size
2673 * per-type header in the front and then gracefully fail (i.e.,
2674 * propagate the error to the caller based on info in the front) when
2675 * the middle is too large.
2677 static int ceph_alloc_middle(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2679 int type
= le16_to_cpu(msg
->hdr
.type
);
2680 int middle_len
= le32_to_cpu(msg
->hdr
.middle_len
);
2682 dout("alloc_middle %p type %d %s middle_len %d\n", msg
, type
,
2683 ceph_msg_type_name(type
), middle_len
);
2684 BUG_ON(!middle_len
);
2685 BUG_ON(msg
->middle
);
2687 msg
->middle
= ceph_buffer_new(middle_len
, GFP_NOFS
);
2694 * Allocate a message for receiving an incoming message on a
2695 * connection, and save the result in con->in_msg. Uses the
2696 * connection's private alloc_msg op if available.
2698 * Returns true if the message should be skipped, false otherwise.
2699 * If true is returned (skip message), con->in_msg will be NULL.
2700 * If false is returned, con->in_msg will contain a pointer to the
2701 * newly-allocated message, or NULL in case of memory exhaustion.
2703 static bool ceph_con_in_msg_alloc(struct ceph_connection
*con
,
2704 struct ceph_msg_header
*hdr
)
2706 int type
= le16_to_cpu(hdr
->type
);
2707 int front_len
= le32_to_cpu(hdr
->front_len
);
2708 int middle_len
= le32_to_cpu(hdr
->middle_len
);
2711 BUG_ON(con
->in_msg
!= NULL
);
2713 if (con
->ops
->alloc_msg
) {
2716 mutex_unlock(&con
->mutex
);
2717 con
->in_msg
= con
->ops
->alloc_msg(con
, hdr
, &skip
);
2718 mutex_lock(&con
->mutex
);
2720 con
->in_msg
->con
= con
->ops
->get(con
);
2721 BUG_ON(con
->in_msg
->con
== NULL
);
2730 con
->in_msg
= ceph_msg_new(type
, front_len
, GFP_NOFS
, false);
2732 pr_err("unable to allocate msg type %d len %d\n",
2736 con
->in_msg
->con
= con
->ops
->get(con
);
2737 BUG_ON(con
->in_msg
->con
== NULL
);
2738 con
->in_msg
->page_alignment
= le16_to_cpu(hdr
->data_off
);
2740 memcpy(&con
->in_msg
->hdr
, &con
->in_hdr
, sizeof(con
->in_hdr
));
2742 if (middle_len
&& !con
->in_msg
->middle
) {
2743 ret
= ceph_alloc_middle(con
, con
->in_msg
);
2745 ceph_msg_put(con
->in_msg
);
2755 * Free a generically kmalloc'd message.
2757 void ceph_msg_kfree(struct ceph_msg
*m
)
2759 dout("msg_kfree %p\n", m
);
2760 if (m
->front_is_vmalloc
)
2761 vfree(m
->front
.iov_base
);
2763 kfree(m
->front
.iov_base
);
2768 * Drop a msg ref. Destroy as needed.
2770 void ceph_msg_last_put(struct kref
*kref
)
2772 struct ceph_msg
*m
= container_of(kref
, struct ceph_msg
, kref
);
2774 dout("ceph_msg_put last one on %p\n", m
);
2775 WARN_ON(!list_empty(&m
->list_head
));
2777 /* drop middle, data, if any */
2779 ceph_buffer_put(m
->middle
);
2786 ceph_pagelist_release(m
->pagelist
);
2794 ceph_msgpool_put(m
->pool
, m
);
2798 EXPORT_SYMBOL(ceph_msg_last_put
);
2800 void ceph_msg_dump(struct ceph_msg
*msg
)
2802 pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg
,
2803 msg
->front_max
, msg
->nr_pages
);
2804 print_hex_dump(KERN_DEBUG
, "header: ",
2805 DUMP_PREFIX_OFFSET
, 16, 1,
2806 &msg
->hdr
, sizeof(msg
->hdr
), true);
2807 print_hex_dump(KERN_DEBUG
, " front: ",
2808 DUMP_PREFIX_OFFSET
, 16, 1,
2809 msg
->front
.iov_base
, msg
->front
.iov_len
, true);
2811 print_hex_dump(KERN_DEBUG
, "middle: ",
2812 DUMP_PREFIX_OFFSET
, 16, 1,
2813 msg
->middle
->vec
.iov_base
,
2814 msg
->middle
->vec
.iov_len
, true);
2815 print_hex_dump(KERN_DEBUG
, "footer: ",
2816 DUMP_PREFIX_OFFSET
, 16, 1,
2817 &msg
->footer
, sizeof(msg
->footer
), true);
2819 EXPORT_SYMBOL(ceph_msg_dump
);