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 */
83 #define CON_STATE_CLOSED 1 /* -> PREOPEN */
84 #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
85 #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
86 #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
87 #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
88 #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
91 * ceph_connection flag bits
93 #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
94 * messages on errors */
95 #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
96 #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
97 #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
98 #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
100 /* static tag bytes (protocol control messages) */
101 static char tag_msg
= CEPH_MSGR_TAG_MSG
;
102 static char tag_ack
= CEPH_MSGR_TAG_ACK
;
103 static char tag_keepalive
= CEPH_MSGR_TAG_KEEPALIVE
;
105 #ifdef CONFIG_LOCKDEP
106 static struct lock_class_key socket_class
;
110 * When skipping (ignoring) a block of input we read it into a "skip
111 * buffer," which is this many bytes in size.
113 #define SKIP_BUF_SIZE 1024
115 static void queue_con(struct ceph_connection
*con
);
116 static void con_work(struct work_struct
*);
117 static void ceph_fault(struct ceph_connection
*con
);
120 * Nicely render a sockaddr as a string. An array of formatted
121 * strings is used, to approximate reentrancy.
123 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
124 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
125 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
126 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
128 static char addr_str
[ADDR_STR_COUNT
][MAX_ADDR_STR_LEN
];
129 static atomic_t addr_str_seq
= ATOMIC_INIT(0);
131 static struct page
*zero_page
; /* used in certain error cases */
133 const char *ceph_pr_addr(const struct sockaddr_storage
*ss
)
137 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
138 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
140 i
= atomic_inc_return(&addr_str_seq
) & ADDR_STR_COUNT_MASK
;
143 switch (ss
->ss_family
) {
145 snprintf(s
, MAX_ADDR_STR_LEN
, "%pI4:%hu", &in4
->sin_addr
,
146 ntohs(in4
->sin_port
));
150 snprintf(s
, MAX_ADDR_STR_LEN
, "[%pI6c]:%hu", &in6
->sin6_addr
,
151 ntohs(in6
->sin6_port
));
155 snprintf(s
, MAX_ADDR_STR_LEN
, "(unknown sockaddr family %hu)",
161 EXPORT_SYMBOL(ceph_pr_addr
);
163 static void encode_my_addr(struct ceph_messenger
*msgr
)
165 memcpy(&msgr
->my_enc_addr
, &msgr
->inst
.addr
, sizeof(msgr
->my_enc_addr
));
166 ceph_encode_addr(&msgr
->my_enc_addr
);
170 * work queue for all reading and writing to/from the socket.
172 static struct workqueue_struct
*ceph_msgr_wq
;
174 void _ceph_msgr_exit(void)
177 destroy_workqueue(ceph_msgr_wq
);
181 BUG_ON(zero_page
== NULL
);
183 page_cache_release(zero_page
);
187 int ceph_msgr_init(void)
189 BUG_ON(zero_page
!= NULL
);
190 zero_page
= ZERO_PAGE(0);
191 page_cache_get(zero_page
);
193 ceph_msgr_wq
= alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT
, 0);
197 pr_err("msgr_init failed to create workqueue\n");
202 EXPORT_SYMBOL(ceph_msgr_init
);
204 void ceph_msgr_exit(void)
206 BUG_ON(ceph_msgr_wq
== NULL
);
210 EXPORT_SYMBOL(ceph_msgr_exit
);
212 void ceph_msgr_flush(void)
214 flush_workqueue(ceph_msgr_wq
);
216 EXPORT_SYMBOL(ceph_msgr_flush
);
218 /* Connection socket state transition functions */
220 static void con_sock_state_init(struct ceph_connection
*con
)
224 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSED
);
225 if (WARN_ON(old_state
!= CON_SOCK_STATE_NEW
))
226 printk("%s: unexpected old state %d\n", __func__
, old_state
);
229 static void con_sock_state_connecting(struct ceph_connection
*con
)
233 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CONNECTING
);
234 if (WARN_ON(old_state
!= CON_SOCK_STATE_CLOSED
))
235 printk("%s: unexpected old state %d\n", __func__
, old_state
);
238 static void con_sock_state_connected(struct ceph_connection
*con
)
242 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CONNECTED
);
243 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTING
))
244 printk("%s: unexpected old state %d\n", __func__
, old_state
);
247 static void con_sock_state_closing(struct ceph_connection
*con
)
251 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSING
);
252 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTING
&&
253 old_state
!= CON_SOCK_STATE_CONNECTED
&&
254 old_state
!= CON_SOCK_STATE_CLOSING
))
255 printk("%s: unexpected old state %d\n", __func__
, old_state
);
258 static void con_sock_state_closed(struct ceph_connection
*con
)
262 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSED
);
263 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTED
&&
264 old_state
!= CON_SOCK_STATE_CLOSING
&&
265 old_state
!= CON_SOCK_STATE_CONNECTING
))
266 printk("%s: unexpected old state %d\n", __func__
, old_state
);
270 * socket callback functions
273 /* data available on socket, or listen socket received a connect */
274 static void ceph_sock_data_ready(struct sock
*sk
, int count_unused
)
276 struct ceph_connection
*con
= sk
->sk_user_data
;
277 if (atomic_read(&con
->msgr
->stopping
)) {
281 if (sk
->sk_state
!= TCP_CLOSE_WAIT
) {
282 dout("%s on %p state = %lu, queueing work\n", __func__
,
288 /* socket has buffer space for writing */
289 static void ceph_sock_write_space(struct sock
*sk
)
291 struct ceph_connection
*con
= sk
->sk_user_data
;
293 /* only queue to workqueue if there is data we want to write,
294 * and there is sufficient space in the socket buffer to accept
295 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
296 * doesn't get called again until try_write() fills the socket
297 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
298 * and net/core/stream.c:sk_stream_write_space().
300 if (test_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
)) {
301 if (sk_stream_wspace(sk
) >= sk_stream_min_wspace(sk
)) {
302 dout("%s %p queueing write work\n", __func__
, con
);
303 clear_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
307 dout("%s %p nothing to write\n", __func__
, con
);
311 /* socket's state has changed */
312 static void ceph_sock_state_change(struct sock
*sk
)
314 struct ceph_connection
*con
= sk
->sk_user_data
;
316 dout("%s %p state = %lu sk_state = %u\n", __func__
,
317 con
, con
->state
, sk
->sk_state
);
319 switch (sk
->sk_state
) {
321 dout("%s TCP_CLOSE\n", __func__
);
323 dout("%s TCP_CLOSE_WAIT\n", __func__
);
324 con_sock_state_closing(con
);
325 set_bit(CON_FLAG_SOCK_CLOSED
, &con
->flags
);
328 case TCP_ESTABLISHED
:
329 dout("%s TCP_ESTABLISHED\n", __func__
);
330 con_sock_state_connected(con
);
333 default: /* Everything else is uninteresting */
339 * set up socket callbacks
341 static void set_sock_callbacks(struct socket
*sock
,
342 struct ceph_connection
*con
)
344 struct sock
*sk
= sock
->sk
;
345 sk
->sk_user_data
= con
;
346 sk
->sk_data_ready
= ceph_sock_data_ready
;
347 sk
->sk_write_space
= ceph_sock_write_space
;
348 sk
->sk_state_change
= ceph_sock_state_change
;
357 * initiate connection to a remote socket.
359 static int ceph_tcp_connect(struct ceph_connection
*con
)
361 struct sockaddr_storage
*paddr
= &con
->peer_addr
.in_addr
;
366 ret
= sock_create_kern(con
->peer_addr
.in_addr
.ss_family
, SOCK_STREAM
,
370 sock
->sk
->sk_allocation
= GFP_NOFS
;
372 #ifdef CONFIG_LOCKDEP
373 lockdep_set_class(&sock
->sk
->sk_lock
, &socket_class
);
376 set_sock_callbacks(sock
, con
);
378 dout("connect %s\n", ceph_pr_addr(&con
->peer_addr
.in_addr
));
380 con_sock_state_connecting(con
);
381 ret
= sock
->ops
->connect(sock
, (struct sockaddr
*)paddr
, sizeof(*paddr
),
383 if (ret
== -EINPROGRESS
) {
384 dout("connect %s EINPROGRESS sk_state = %u\n",
385 ceph_pr_addr(&con
->peer_addr
.in_addr
),
387 } else if (ret
< 0) {
388 pr_err("connect %s error %d\n",
389 ceph_pr_addr(&con
->peer_addr
.in_addr
), ret
);
391 con
->error_msg
= "connect error";
399 static int ceph_tcp_recvmsg(struct socket
*sock
, void *buf
, size_t len
)
401 struct kvec iov
= {buf
, len
};
402 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
405 r
= kernel_recvmsg(sock
, &msg
, &iov
, 1, len
, msg
.msg_flags
);
412 * write something. @more is true if caller will be sending more data
415 static int ceph_tcp_sendmsg(struct socket
*sock
, struct kvec
*iov
,
416 size_t kvlen
, size_t len
, int more
)
418 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
422 msg
.msg_flags
|= MSG_MORE
;
424 msg
.msg_flags
|= MSG_EOR
; /* superfluous, but what the hell */
426 r
= kernel_sendmsg(sock
, &msg
, iov
, kvlen
, len
);
432 static int ceph_tcp_sendpage(struct socket
*sock
, struct page
*page
,
433 int offset
, size_t size
, int more
)
435 int flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
| (more
? MSG_MORE
: MSG_EOR
);
438 ret
= kernel_sendpage(sock
, page
, offset
, size
, flags
);
447 * Shutdown/close the socket for the given connection.
449 static int con_close_socket(struct ceph_connection
*con
)
453 dout("con_close_socket on %p sock %p\n", con
, con
->sock
);
456 rc
= con
->sock
->ops
->shutdown(con
->sock
, SHUT_RDWR
);
457 sock_release(con
->sock
);
461 * Forcibly clear the SOCK_CLOSED flag. It gets set
462 * independent of the connection mutex, and we could have
463 * received a socket close event before we had the chance to
464 * shut the socket down.
466 clear_bit(CON_FLAG_SOCK_CLOSED
, &con
->flags
);
467 con_sock_state_closed(con
);
472 * Reset a connection. Discard all incoming and outgoing messages
473 * and clear *_seq state.
475 static void ceph_msg_remove(struct ceph_msg
*msg
)
477 list_del_init(&msg
->list_head
);
478 BUG_ON(msg
->con
== NULL
);
479 msg
->con
->ops
->put(msg
->con
);
484 static void ceph_msg_remove_list(struct list_head
*head
)
486 while (!list_empty(head
)) {
487 struct ceph_msg
*msg
= list_first_entry(head
, struct ceph_msg
,
489 ceph_msg_remove(msg
);
493 static void reset_connection(struct ceph_connection
*con
)
495 /* reset connection, out_queue, msg_ and connect_seq */
496 /* discard existing out_queue and msg_seq */
497 ceph_msg_remove_list(&con
->out_queue
);
498 ceph_msg_remove_list(&con
->out_sent
);
501 BUG_ON(con
->in_msg
->con
!= con
);
502 con
->in_msg
->con
= NULL
;
503 ceph_msg_put(con
->in_msg
);
508 con
->connect_seq
= 0;
511 ceph_msg_put(con
->out_msg
);
515 con
->in_seq_acked
= 0;
519 * mark a peer down. drop any open connections.
521 void ceph_con_close(struct ceph_connection
*con
)
523 mutex_lock(&con
->mutex
);
524 dout("con_close %p peer %s\n", con
,
525 ceph_pr_addr(&con
->peer_addr
.in_addr
));
526 con
->state
= CON_STATE_CLOSED
;
528 clear_bit(CON_FLAG_LOSSYTX
, &con
->flags
); /* so we retry next connect */
529 clear_bit(CON_FLAG_KEEPALIVE_PENDING
, &con
->flags
);
530 clear_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
);
531 clear_bit(CON_FLAG_KEEPALIVE_PENDING
, &con
->flags
);
532 clear_bit(CON_FLAG_BACKOFF
, &con
->flags
);
534 reset_connection(con
);
535 con
->peer_global_seq
= 0;
536 cancel_delayed_work(&con
->work
);
537 con_close_socket(con
);
538 mutex_unlock(&con
->mutex
);
540 EXPORT_SYMBOL(ceph_con_close
);
543 * Reopen a closed connection, with a new peer address.
545 void ceph_con_open(struct ceph_connection
*con
,
546 __u8 entity_type
, __u64 entity_num
,
547 struct ceph_entity_addr
*addr
)
549 mutex_lock(&con
->mutex
);
550 dout("con_open %p %s\n", con
, ceph_pr_addr(&addr
->in_addr
));
552 BUG_ON(con
->state
!= CON_STATE_CLOSED
);
553 con
->state
= CON_STATE_PREOPEN
;
555 con
->peer_name
.type
= (__u8
) entity_type
;
556 con
->peer_name
.num
= cpu_to_le64(entity_num
);
558 memcpy(&con
->peer_addr
, addr
, sizeof(*addr
));
559 con
->delay
= 0; /* reset backoff memory */
560 mutex_unlock(&con
->mutex
);
563 EXPORT_SYMBOL(ceph_con_open
);
566 * return true if this connection ever successfully opened
568 bool ceph_con_opened(struct ceph_connection
*con
)
570 return con
->connect_seq
> 0;
574 * initialize a new connection.
576 void ceph_con_init(struct ceph_connection
*con
, void *private,
577 const struct ceph_connection_operations
*ops
,
578 struct ceph_messenger
*msgr
)
580 dout("con_init %p\n", con
);
581 memset(con
, 0, sizeof(*con
));
582 con
->private = private;
586 con_sock_state_init(con
);
588 mutex_init(&con
->mutex
);
589 INIT_LIST_HEAD(&con
->out_queue
);
590 INIT_LIST_HEAD(&con
->out_sent
);
591 INIT_DELAYED_WORK(&con
->work
, con_work
);
593 con
->state
= CON_STATE_CLOSED
;
595 EXPORT_SYMBOL(ceph_con_init
);
599 * We maintain a global counter to order connection attempts. Get
600 * a unique seq greater than @gt.
602 static u32
get_global_seq(struct ceph_messenger
*msgr
, u32 gt
)
606 spin_lock(&msgr
->global_seq_lock
);
607 if (msgr
->global_seq
< gt
)
608 msgr
->global_seq
= gt
;
609 ret
= ++msgr
->global_seq
;
610 spin_unlock(&msgr
->global_seq_lock
);
614 static void con_out_kvec_reset(struct ceph_connection
*con
)
616 con
->out_kvec_left
= 0;
617 con
->out_kvec_bytes
= 0;
618 con
->out_kvec_cur
= &con
->out_kvec
[0];
621 static void con_out_kvec_add(struct ceph_connection
*con
,
622 size_t size
, void *data
)
626 index
= con
->out_kvec_left
;
627 BUG_ON(index
>= ARRAY_SIZE(con
->out_kvec
));
629 con
->out_kvec
[index
].iov_len
= size
;
630 con
->out_kvec
[index
].iov_base
= data
;
631 con
->out_kvec_left
++;
632 con
->out_kvec_bytes
+= size
;
636 static void init_bio_iter(struct bio
*bio
, struct bio
**iter
, int *seg
)
647 static void iter_bio_next(struct bio
**bio_iter
, int *seg
)
649 if (*bio_iter
== NULL
)
652 BUG_ON(*seg
>= (*bio_iter
)->bi_vcnt
);
655 if (*seg
== (*bio_iter
)->bi_vcnt
)
656 init_bio_iter((*bio_iter
)->bi_next
, bio_iter
, seg
);
660 static void prepare_write_message_data(struct ceph_connection
*con
)
662 struct ceph_msg
*msg
= con
->out_msg
;
665 BUG_ON(!msg
->hdr
.data_len
);
667 /* initialize page iterator */
668 con
->out_msg_pos
.page
= 0;
670 con
->out_msg_pos
.page_pos
= msg
->page_alignment
;
672 con
->out_msg_pos
.page_pos
= 0;
675 init_bio_iter(msg
->bio
, &msg
->bio_iter
, &msg
->bio_seg
);
677 con
->out_msg_pos
.data_pos
= 0;
678 con
->out_msg_pos
.did_page_crc
= false;
679 con
->out_more
= 1; /* data + footer will follow */
683 * Prepare footer for currently outgoing message, and finish things
684 * off. Assumes out_kvec* are already valid.. we just add on to the end.
686 static void prepare_write_message_footer(struct ceph_connection
*con
)
688 struct ceph_msg
*m
= con
->out_msg
;
689 int v
= con
->out_kvec_left
;
691 m
->footer
.flags
|= CEPH_MSG_FOOTER_COMPLETE
;
693 dout("prepare_write_message_footer %p\n", con
);
694 con
->out_kvec_is_msg
= true;
695 con
->out_kvec
[v
].iov_base
= &m
->footer
;
696 con
->out_kvec
[v
].iov_len
= sizeof(m
->footer
);
697 con
->out_kvec_bytes
+= sizeof(m
->footer
);
698 con
->out_kvec_left
++;
699 con
->out_more
= m
->more_to_follow
;
700 con
->out_msg_done
= true;
704 * Prepare headers for the next outgoing message.
706 static void prepare_write_message(struct ceph_connection
*con
)
711 con_out_kvec_reset(con
);
712 con
->out_kvec_is_msg
= true;
713 con
->out_msg_done
= false;
715 /* Sneak an ack in there first? If we can get it into the same
716 * TCP packet that's a good thing. */
717 if (con
->in_seq
> con
->in_seq_acked
) {
718 con
->in_seq_acked
= con
->in_seq
;
719 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
720 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
721 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
725 BUG_ON(list_empty(&con
->out_queue
));
726 m
= list_first_entry(&con
->out_queue
, struct ceph_msg
, list_head
);
728 BUG_ON(m
->con
!= con
);
730 /* put message on sent list */
732 list_move_tail(&m
->list_head
, &con
->out_sent
);
735 * only assign outgoing seq # if we haven't sent this message
736 * yet. if it is requeued, resend with it's original seq.
738 if (m
->needs_out_seq
) {
739 m
->hdr
.seq
= cpu_to_le64(++con
->out_seq
);
740 m
->needs_out_seq
= false;
743 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n",
744 m
, con
->out_seq
, le16_to_cpu(m
->hdr
.type
),
745 le32_to_cpu(m
->hdr
.front_len
), le32_to_cpu(m
->hdr
.middle_len
),
746 le32_to_cpu(m
->hdr
.data_len
),
748 BUG_ON(le32_to_cpu(m
->hdr
.front_len
) != m
->front
.iov_len
);
750 /* tag + hdr + front + middle */
751 con_out_kvec_add(con
, sizeof (tag_msg
), &tag_msg
);
752 con_out_kvec_add(con
, sizeof (m
->hdr
), &m
->hdr
);
753 con_out_kvec_add(con
, m
->front
.iov_len
, m
->front
.iov_base
);
756 con_out_kvec_add(con
, m
->middle
->vec
.iov_len
,
757 m
->middle
->vec
.iov_base
);
759 /* fill in crc (except data pages), footer */
760 crc
= crc32c(0, &m
->hdr
, offsetof(struct ceph_msg_header
, crc
));
761 con
->out_msg
->hdr
.crc
= cpu_to_le32(crc
);
762 con
->out_msg
->footer
.flags
= 0;
764 crc
= crc32c(0, m
->front
.iov_base
, m
->front
.iov_len
);
765 con
->out_msg
->footer
.front_crc
= cpu_to_le32(crc
);
767 crc
= crc32c(0, m
->middle
->vec
.iov_base
,
768 m
->middle
->vec
.iov_len
);
769 con
->out_msg
->footer
.middle_crc
= cpu_to_le32(crc
);
771 con
->out_msg
->footer
.middle_crc
= 0;
772 dout("%s front_crc %u middle_crc %u\n", __func__
,
773 le32_to_cpu(con
->out_msg
->footer
.front_crc
),
774 le32_to_cpu(con
->out_msg
->footer
.middle_crc
));
776 /* is there a data payload? */
777 con
->out_msg
->footer
.data_crc
= 0;
779 prepare_write_message_data(con
);
781 /* no, queue up footer too and be done */
782 prepare_write_message_footer(con
);
784 set_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
);
790 static void prepare_write_ack(struct ceph_connection
*con
)
792 dout("prepare_write_ack %p %llu -> %llu\n", con
,
793 con
->in_seq_acked
, con
->in_seq
);
794 con
->in_seq_acked
= con
->in_seq
;
796 con_out_kvec_reset(con
);
798 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
800 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
801 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
804 con
->out_more
= 1; /* more will follow.. eventually.. */
805 set_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
);
809 * Prepare to write keepalive byte.
811 static void prepare_write_keepalive(struct ceph_connection
*con
)
813 dout("prepare_write_keepalive %p\n", con
);
814 con_out_kvec_reset(con
);
815 con_out_kvec_add(con
, sizeof (tag_keepalive
), &tag_keepalive
);
816 set_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
);
820 * Connection negotiation.
823 static struct ceph_auth_handshake
*get_connect_authorizer(struct ceph_connection
*con
,
826 struct ceph_auth_handshake
*auth
;
828 if (!con
->ops
->get_authorizer
) {
829 con
->out_connect
.authorizer_protocol
= CEPH_AUTH_UNKNOWN
;
830 con
->out_connect
.authorizer_len
= 0;
834 /* Can't hold the mutex while getting authorizer */
835 mutex_unlock(&con
->mutex
);
836 auth
= con
->ops
->get_authorizer(con
, auth_proto
, con
->auth_retry
);
837 mutex_lock(&con
->mutex
);
841 if (con
->state
!= CON_STATE_NEGOTIATING
)
842 return ERR_PTR(-EAGAIN
);
844 con
->auth_reply_buf
= auth
->authorizer_reply_buf
;
845 con
->auth_reply_buf_len
= auth
->authorizer_reply_buf_len
;
850 * We connected to a peer and are saying hello.
852 static void prepare_write_banner(struct ceph_connection
*con
)
854 con_out_kvec_add(con
, strlen(CEPH_BANNER
), CEPH_BANNER
);
855 con_out_kvec_add(con
, sizeof (con
->msgr
->my_enc_addr
),
856 &con
->msgr
->my_enc_addr
);
859 set_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
);
862 static int prepare_write_connect(struct ceph_connection
*con
)
864 unsigned int global_seq
= get_global_seq(con
->msgr
, 0);
867 struct ceph_auth_handshake
*auth
;
869 switch (con
->peer_name
.type
) {
870 case CEPH_ENTITY_TYPE_MON
:
871 proto
= CEPH_MONC_PROTOCOL
;
873 case CEPH_ENTITY_TYPE_OSD
:
874 proto
= CEPH_OSDC_PROTOCOL
;
876 case CEPH_ENTITY_TYPE_MDS
:
877 proto
= CEPH_MDSC_PROTOCOL
;
883 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con
,
884 con
->connect_seq
, global_seq
, proto
);
886 con
->out_connect
.features
= cpu_to_le64(con
->msgr
->supported_features
);
887 con
->out_connect
.host_type
= cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT
);
888 con
->out_connect
.connect_seq
= cpu_to_le32(con
->connect_seq
);
889 con
->out_connect
.global_seq
= cpu_to_le32(global_seq
);
890 con
->out_connect
.protocol_version
= cpu_to_le32(proto
);
891 con
->out_connect
.flags
= 0;
893 auth_proto
= CEPH_AUTH_UNKNOWN
;
894 auth
= get_connect_authorizer(con
, &auth_proto
);
896 return PTR_ERR(auth
);
898 con
->out_connect
.authorizer_protocol
= cpu_to_le32(auth_proto
);
899 con
->out_connect
.authorizer_len
= auth
?
900 cpu_to_le32(auth
->authorizer_buf_len
) : 0;
902 con_out_kvec_reset(con
);
903 con_out_kvec_add(con
, sizeof (con
->out_connect
),
905 if (auth
&& auth
->authorizer_buf_len
)
906 con_out_kvec_add(con
, auth
->authorizer_buf_len
,
907 auth
->authorizer_buf
);
910 set_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
);
916 * write as much of pending kvecs to the socket as we can.
918 * 0 -> socket full, but more to do
921 static int write_partial_kvec(struct ceph_connection
*con
)
925 dout("write_partial_kvec %p %d left\n", con
, con
->out_kvec_bytes
);
926 while (con
->out_kvec_bytes
> 0) {
927 ret
= ceph_tcp_sendmsg(con
->sock
, con
->out_kvec_cur
,
928 con
->out_kvec_left
, con
->out_kvec_bytes
,
932 con
->out_kvec_bytes
-= ret
;
933 if (con
->out_kvec_bytes
== 0)
936 /* account for full iov entries consumed */
937 while (ret
>= con
->out_kvec_cur
->iov_len
) {
938 BUG_ON(!con
->out_kvec_left
);
939 ret
-= con
->out_kvec_cur
->iov_len
;
941 con
->out_kvec_left
--;
943 /* and for a partially-consumed entry */
945 con
->out_kvec_cur
->iov_len
-= ret
;
946 con
->out_kvec_cur
->iov_base
+= ret
;
949 con
->out_kvec_left
= 0;
950 con
->out_kvec_is_msg
= false;
953 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con
,
954 con
->out_kvec_bytes
, con
->out_kvec_left
, ret
);
955 return ret
; /* done! */
958 static void out_msg_pos_next(struct ceph_connection
*con
, struct page
*page
,
959 size_t len
, size_t sent
, bool in_trail
)
961 struct ceph_msg
*msg
= con
->out_msg
;
966 con
->out_msg_pos
.data_pos
+= sent
;
967 con
->out_msg_pos
.page_pos
+= sent
;
972 con
->out_msg_pos
.page_pos
= 0;
973 con
->out_msg_pos
.page
++;
974 con
->out_msg_pos
.did_page_crc
= false;
976 list_move_tail(&page
->lru
,
978 else if (msg
->pagelist
)
979 list_move_tail(&page
->lru
,
980 &msg
->pagelist
->head
);
983 iter_bio_next(&msg
->bio_iter
, &msg
->bio_seg
);
988 * Write as much message data payload as we can. If we finish, queue
990 * 1 -> done, footer is now queued in out_kvec[].
991 * 0 -> socket full, but more to do
994 static int write_partial_msg_pages(struct ceph_connection
*con
)
996 struct ceph_msg
*msg
= con
->out_msg
;
997 unsigned int data_len
= le32_to_cpu(msg
->hdr
.data_len
);
999 bool do_datacrc
= !con
->msgr
->nocrc
;
1001 int total_max_write
;
1002 bool in_trail
= false;
1003 const size_t trail_len
= (msg
->trail
? msg
->trail
->length
: 0);
1004 const size_t trail_off
= data_len
- trail_len
;
1006 dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
1007 con
, msg
, con
->out_msg_pos
.page
, msg
->nr_pages
,
1008 con
->out_msg_pos
.page_pos
);
1011 * Iterate through each page that contains data to be
1012 * written, and send as much as possible for each.
1014 * If we are calculating the data crc (the default), we will
1015 * need to map the page. If we have no pages, they have
1016 * been revoked, so use the zero page.
1018 while (data_len
> con
->out_msg_pos
.data_pos
) {
1019 struct page
*page
= NULL
;
1020 int max_write
= PAGE_SIZE
;
1023 in_trail
= in_trail
|| con
->out_msg_pos
.data_pos
>= trail_off
;
1025 total_max_write
= trail_off
- con
->out_msg_pos
.data_pos
;
1028 total_max_write
= data_len
- con
->out_msg_pos
.data_pos
;
1030 page
= list_first_entry(&msg
->trail
->head
,
1032 } else if (msg
->pages
) {
1033 page
= msg
->pages
[con
->out_msg_pos
.page
];
1034 } else if (msg
->pagelist
) {
1035 page
= list_first_entry(&msg
->pagelist
->head
,
1038 } else if (msg
->bio
) {
1041 bv
= bio_iovec_idx(msg
->bio_iter
, msg
->bio_seg
);
1043 bio_offset
= bv
->bv_offset
;
1044 max_write
= bv
->bv_len
;
1049 len
= min_t(int, max_write
- con
->out_msg_pos
.page_pos
,
1052 if (do_datacrc
&& !con
->out_msg_pos
.did_page_crc
) {
1054 u32 crc
= le32_to_cpu(msg
->footer
.data_crc
);
1058 BUG_ON(kaddr
== NULL
);
1059 base
= kaddr
+ con
->out_msg_pos
.page_pos
+ bio_offset
;
1060 crc
= crc32c(crc
, base
, len
);
1061 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1062 con
->out_msg_pos
.did_page_crc
= true;
1064 ret
= ceph_tcp_sendpage(con
->sock
, page
,
1065 con
->out_msg_pos
.page_pos
+ bio_offset
,
1074 out_msg_pos_next(con
, page
, len
, (size_t) ret
, in_trail
);
1077 dout("write_partial_msg_pages %p msg %p done\n", con
, msg
);
1079 /* prepare and queue up footer, too */
1081 msg
->footer
.flags
|= CEPH_MSG_FOOTER_NOCRC
;
1082 con_out_kvec_reset(con
);
1083 prepare_write_message_footer(con
);
1092 static int write_partial_skip(struct ceph_connection
*con
)
1096 while (con
->out_skip
> 0) {
1097 size_t size
= min(con
->out_skip
, (int) PAGE_CACHE_SIZE
);
1099 ret
= ceph_tcp_sendpage(con
->sock
, zero_page
, 0, size
, 1);
1102 con
->out_skip
-= ret
;
1110 * Prepare to read connection handshake, or an ack.
1112 static void prepare_read_banner(struct ceph_connection
*con
)
1114 dout("prepare_read_banner %p\n", con
);
1115 con
->in_base_pos
= 0;
1118 static void prepare_read_connect(struct ceph_connection
*con
)
1120 dout("prepare_read_connect %p\n", con
);
1121 con
->in_base_pos
= 0;
1124 static void prepare_read_ack(struct ceph_connection
*con
)
1126 dout("prepare_read_ack %p\n", con
);
1127 con
->in_base_pos
= 0;
1130 static void prepare_read_tag(struct ceph_connection
*con
)
1132 dout("prepare_read_tag %p\n", con
);
1133 con
->in_base_pos
= 0;
1134 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1138 * Prepare to read a message.
1140 static int prepare_read_message(struct ceph_connection
*con
)
1142 dout("prepare_read_message %p\n", con
);
1143 BUG_ON(con
->in_msg
!= NULL
);
1144 con
->in_base_pos
= 0;
1145 con
->in_front_crc
= con
->in_middle_crc
= con
->in_data_crc
= 0;
1150 static int read_partial(struct ceph_connection
*con
,
1151 int end
, int size
, void *object
)
1153 while (con
->in_base_pos
< end
) {
1154 int left
= end
- con
->in_base_pos
;
1155 int have
= size
- left
;
1156 int ret
= ceph_tcp_recvmsg(con
->sock
, object
+ have
, left
);
1159 con
->in_base_pos
+= ret
;
1166 * Read all or part of the connect-side handshake on a new connection
1168 static int read_partial_banner(struct ceph_connection
*con
)
1174 dout("read_partial_banner %p at %d\n", con
, con
->in_base_pos
);
1177 size
= strlen(CEPH_BANNER
);
1179 ret
= read_partial(con
, end
, size
, con
->in_banner
);
1183 size
= sizeof (con
->actual_peer_addr
);
1185 ret
= read_partial(con
, end
, size
, &con
->actual_peer_addr
);
1189 size
= sizeof (con
->peer_addr_for_me
);
1191 ret
= read_partial(con
, end
, size
, &con
->peer_addr_for_me
);
1199 static int read_partial_connect(struct ceph_connection
*con
)
1205 dout("read_partial_connect %p at %d\n", con
, con
->in_base_pos
);
1207 size
= sizeof (con
->in_reply
);
1209 ret
= read_partial(con
, end
, size
, &con
->in_reply
);
1213 size
= le32_to_cpu(con
->in_reply
.authorizer_len
);
1215 ret
= read_partial(con
, end
, size
, con
->auth_reply_buf
);
1219 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1220 con
, (int)con
->in_reply
.tag
,
1221 le32_to_cpu(con
->in_reply
.connect_seq
),
1222 le32_to_cpu(con
->in_reply
.global_seq
));
1229 * Verify the hello banner looks okay.
1231 static int verify_hello(struct ceph_connection
*con
)
1233 if (memcmp(con
->in_banner
, CEPH_BANNER
, strlen(CEPH_BANNER
))) {
1234 pr_err("connect to %s got bad banner\n",
1235 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1236 con
->error_msg
= "protocol error, bad banner";
1242 static bool addr_is_blank(struct sockaddr_storage
*ss
)
1244 switch (ss
->ss_family
) {
1246 return ((struct sockaddr_in
*)ss
)->sin_addr
.s_addr
== 0;
1249 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[0] == 0 &&
1250 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[1] == 0 &&
1251 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[2] == 0 &&
1252 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[3] == 0;
1257 static int addr_port(struct sockaddr_storage
*ss
)
1259 switch (ss
->ss_family
) {
1261 return ntohs(((struct sockaddr_in
*)ss
)->sin_port
);
1263 return ntohs(((struct sockaddr_in6
*)ss
)->sin6_port
);
1268 static void addr_set_port(struct sockaddr_storage
*ss
, int p
)
1270 switch (ss
->ss_family
) {
1272 ((struct sockaddr_in
*)ss
)->sin_port
= htons(p
);
1275 ((struct sockaddr_in6
*)ss
)->sin6_port
= htons(p
);
1281 * Unlike other *_pton function semantics, zero indicates success.
1283 static int ceph_pton(const char *str
, size_t len
, struct sockaddr_storage
*ss
,
1284 char delim
, const char **ipend
)
1286 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
1287 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
1289 memset(ss
, 0, sizeof(*ss
));
1291 if (in4_pton(str
, len
, (u8
*)&in4
->sin_addr
.s_addr
, delim
, ipend
)) {
1292 ss
->ss_family
= AF_INET
;
1296 if (in6_pton(str
, len
, (u8
*)&in6
->sin6_addr
.s6_addr
, delim
, ipend
)) {
1297 ss
->ss_family
= AF_INET6
;
1305 * Extract hostname string and resolve using kernel DNS facility.
1307 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1308 static int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1309 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1311 const char *end
, *delim_p
;
1312 char *colon_p
, *ip_addr
= NULL
;
1316 * The end of the hostname occurs immediately preceding the delimiter or
1317 * the port marker (':') where the delimiter takes precedence.
1319 delim_p
= memchr(name
, delim
, namelen
);
1320 colon_p
= memchr(name
, ':', namelen
);
1322 if (delim_p
&& colon_p
)
1323 end
= delim_p
< colon_p
? delim_p
: colon_p
;
1324 else if (!delim_p
&& colon_p
)
1328 if (!end
) /* case: hostname:/ */
1329 end
= name
+ namelen
;
1335 /* do dns_resolve upcall */
1336 ip_len
= dns_query(NULL
, name
, end
- name
, NULL
, &ip_addr
, NULL
);
1338 ret
= ceph_pton(ip_addr
, ip_len
, ss
, -1, NULL
);
1346 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end
- name
), name
,
1347 ret
, ret
? "failed" : ceph_pr_addr(ss
));
1352 static inline int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1353 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1360 * Parse a server name (IP or hostname). If a valid IP address is not found
1361 * then try to extract a hostname to resolve using userspace DNS upcall.
1363 static int ceph_parse_server_name(const char *name
, size_t namelen
,
1364 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1368 ret
= ceph_pton(name
, namelen
, ss
, delim
, ipend
);
1370 ret
= ceph_dns_resolve_name(name
, namelen
, ss
, delim
, ipend
);
1376 * Parse an ip[:port] list into an addr array. Use the default
1377 * monitor port if a port isn't specified.
1379 int ceph_parse_ips(const char *c
, const char *end
,
1380 struct ceph_entity_addr
*addr
,
1381 int max_count
, int *count
)
1383 int i
, ret
= -EINVAL
;
1386 dout("parse_ips on '%.*s'\n", (int)(end
-c
), c
);
1387 for (i
= 0; i
< max_count
; i
++) {
1389 struct sockaddr_storage
*ss
= &addr
[i
].in_addr
;
1398 ret
= ceph_parse_server_name(p
, end
- p
, ss
, delim
, &ipend
);
1407 dout("missing matching ']'\n");
1414 if (p
< end
&& *p
== ':') {
1417 while (p
< end
&& *p
>= '0' && *p
<= '9') {
1418 port
= (port
* 10) + (*p
- '0');
1421 if (port
> 65535 || port
== 0)
1424 port
= CEPH_MON_PORT
;
1427 addr_set_port(ss
, port
);
1429 dout("parse_ips got %s\n", ceph_pr_addr(ss
));
1446 pr_err("parse_ips bad ip '%.*s'\n", (int)(end
- c
), c
);
1449 EXPORT_SYMBOL(ceph_parse_ips
);
1451 static int process_banner(struct ceph_connection
*con
)
1453 dout("process_banner on %p\n", con
);
1455 if (verify_hello(con
) < 0)
1458 ceph_decode_addr(&con
->actual_peer_addr
);
1459 ceph_decode_addr(&con
->peer_addr_for_me
);
1462 * Make sure the other end is who we wanted. note that the other
1463 * end may not yet know their ip address, so if it's 0.0.0.0, give
1464 * them the benefit of the doubt.
1466 if (memcmp(&con
->peer_addr
, &con
->actual_peer_addr
,
1467 sizeof(con
->peer_addr
)) != 0 &&
1468 !(addr_is_blank(&con
->actual_peer_addr
.in_addr
) &&
1469 con
->actual_peer_addr
.nonce
== con
->peer_addr
.nonce
)) {
1470 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1471 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1472 (int)le32_to_cpu(con
->peer_addr
.nonce
),
1473 ceph_pr_addr(&con
->actual_peer_addr
.in_addr
),
1474 (int)le32_to_cpu(con
->actual_peer_addr
.nonce
));
1475 con
->error_msg
= "wrong peer at address";
1480 * did we learn our address?
1482 if (addr_is_blank(&con
->msgr
->inst
.addr
.in_addr
)) {
1483 int port
= addr_port(&con
->msgr
->inst
.addr
.in_addr
);
1485 memcpy(&con
->msgr
->inst
.addr
.in_addr
,
1486 &con
->peer_addr_for_me
.in_addr
,
1487 sizeof(con
->peer_addr_for_me
.in_addr
));
1488 addr_set_port(&con
->msgr
->inst
.addr
.in_addr
, port
);
1489 encode_my_addr(con
->msgr
);
1490 dout("process_banner learned my addr is %s\n",
1491 ceph_pr_addr(&con
->msgr
->inst
.addr
.in_addr
));
1497 static void fail_protocol(struct ceph_connection
*con
)
1499 reset_connection(con
);
1500 BUG_ON(con
->state
!= CON_STATE_NEGOTIATING
);
1501 con
->state
= CON_STATE_CLOSED
;
1504 static int process_connect(struct ceph_connection
*con
)
1506 u64 sup_feat
= con
->msgr
->supported_features
;
1507 u64 req_feat
= con
->msgr
->required_features
;
1508 u64 server_feat
= le64_to_cpu(con
->in_reply
.features
);
1511 dout("process_connect on %p tag %d\n", con
, (int)con
->in_tag
);
1513 switch (con
->in_reply
.tag
) {
1514 case CEPH_MSGR_TAG_FEATURES
:
1515 pr_err("%s%lld %s feature set mismatch,"
1516 " my %llx < server's %llx, missing %llx\n",
1517 ENTITY_NAME(con
->peer_name
),
1518 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1519 sup_feat
, server_feat
, server_feat
& ~sup_feat
);
1520 con
->error_msg
= "missing required protocol features";
1524 case CEPH_MSGR_TAG_BADPROTOVER
:
1525 pr_err("%s%lld %s protocol version mismatch,"
1526 " my %d != server's %d\n",
1527 ENTITY_NAME(con
->peer_name
),
1528 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1529 le32_to_cpu(con
->out_connect
.protocol_version
),
1530 le32_to_cpu(con
->in_reply
.protocol_version
));
1531 con
->error_msg
= "protocol version mismatch";
1535 case CEPH_MSGR_TAG_BADAUTHORIZER
:
1537 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con
,
1539 if (con
->auth_retry
== 2) {
1540 con
->error_msg
= "connect authorization failure";
1543 con
->auth_retry
= 1;
1544 ret
= prepare_write_connect(con
);
1547 prepare_read_connect(con
);
1550 case CEPH_MSGR_TAG_RESETSESSION
:
1552 * If we connected with a large connect_seq but the peer
1553 * has no record of a session with us (no connection, or
1554 * connect_seq == 0), they will send RESETSESION to indicate
1555 * that they must have reset their session, and may have
1558 dout("process_connect got RESET peer seq %u\n",
1559 le32_to_cpu(con
->in_reply
.connect_seq
));
1560 pr_err("%s%lld %s connection reset\n",
1561 ENTITY_NAME(con
->peer_name
),
1562 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1563 reset_connection(con
);
1564 ret
= prepare_write_connect(con
);
1567 prepare_read_connect(con
);
1569 /* Tell ceph about it. */
1570 mutex_unlock(&con
->mutex
);
1571 pr_info("reset on %s%lld\n", ENTITY_NAME(con
->peer_name
));
1572 if (con
->ops
->peer_reset
)
1573 con
->ops
->peer_reset(con
);
1574 mutex_lock(&con
->mutex
);
1575 if (con
->state
!= CON_STATE_NEGOTIATING
)
1579 case CEPH_MSGR_TAG_RETRY_SESSION
:
1581 * If we sent a smaller connect_seq than the peer has, try
1582 * again with a larger value.
1584 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
1585 le32_to_cpu(con
->out_connect
.connect_seq
),
1586 le32_to_cpu(con
->in_reply
.connect_seq
));
1587 con
->connect_seq
= le32_to_cpu(con
->in_reply
.connect_seq
);
1588 ret
= prepare_write_connect(con
);
1591 prepare_read_connect(con
);
1594 case CEPH_MSGR_TAG_RETRY_GLOBAL
:
1596 * If we sent a smaller global_seq than the peer has, try
1597 * again with a larger value.
1599 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1600 con
->peer_global_seq
,
1601 le32_to_cpu(con
->in_reply
.global_seq
));
1602 get_global_seq(con
->msgr
,
1603 le32_to_cpu(con
->in_reply
.global_seq
));
1604 ret
= prepare_write_connect(con
);
1607 prepare_read_connect(con
);
1610 case CEPH_MSGR_TAG_READY
:
1611 if (req_feat
& ~server_feat
) {
1612 pr_err("%s%lld %s protocol feature mismatch,"
1613 " my required %llx > server's %llx, need %llx\n",
1614 ENTITY_NAME(con
->peer_name
),
1615 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1616 req_feat
, server_feat
, req_feat
& ~server_feat
);
1617 con
->error_msg
= "missing required protocol features";
1622 BUG_ON(con
->state
!= CON_STATE_NEGOTIATING
);
1623 con
->state
= CON_STATE_OPEN
;
1625 con
->peer_global_seq
= le32_to_cpu(con
->in_reply
.global_seq
);
1627 con
->peer_features
= server_feat
;
1628 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1629 con
->peer_global_seq
,
1630 le32_to_cpu(con
->in_reply
.connect_seq
),
1632 WARN_ON(con
->connect_seq
!=
1633 le32_to_cpu(con
->in_reply
.connect_seq
));
1635 if (con
->in_reply
.flags
& CEPH_MSG_CONNECT_LOSSY
)
1636 set_bit(CON_FLAG_LOSSYTX
, &con
->flags
);
1638 con
->delay
= 0; /* reset backoff memory */
1640 prepare_read_tag(con
);
1643 case CEPH_MSGR_TAG_WAIT
:
1645 * If there is a connection race (we are opening
1646 * connections to each other), one of us may just have
1647 * to WAIT. This shouldn't happen if we are the
1650 pr_err("process_connect got WAIT as client\n");
1651 con
->error_msg
= "protocol error, got WAIT as client";
1655 pr_err("connect protocol error, will retry\n");
1656 con
->error_msg
= "protocol error, garbage tag during connect";
1664 * read (part of) an ack
1666 static int read_partial_ack(struct ceph_connection
*con
)
1668 int size
= sizeof (con
->in_temp_ack
);
1671 return read_partial(con
, end
, size
, &con
->in_temp_ack
);
1676 * We can finally discard anything that's been acked.
1678 static void process_ack(struct ceph_connection
*con
)
1681 u64 ack
= le64_to_cpu(con
->in_temp_ack
);
1684 while (!list_empty(&con
->out_sent
)) {
1685 m
= list_first_entry(&con
->out_sent
, struct ceph_msg
,
1687 seq
= le64_to_cpu(m
->hdr
.seq
);
1690 dout("got ack for seq %llu type %d at %p\n", seq
,
1691 le16_to_cpu(m
->hdr
.type
), m
);
1692 m
->ack_stamp
= jiffies
;
1695 prepare_read_tag(con
);
1701 static int read_partial_message_section(struct ceph_connection
*con
,
1702 struct kvec
*section
,
1703 unsigned int sec_len
, u32
*crc
)
1709 while (section
->iov_len
< sec_len
) {
1710 BUG_ON(section
->iov_base
== NULL
);
1711 left
= sec_len
- section
->iov_len
;
1712 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)section
->iov_base
+
1713 section
->iov_len
, left
);
1716 section
->iov_len
+= ret
;
1718 if (section
->iov_len
== sec_len
)
1719 *crc
= crc32c(0, section
->iov_base
, section
->iov_len
);
1724 static bool ceph_con_in_msg_alloc(struct ceph_connection
*con
,
1725 struct ceph_msg_header
*hdr
);
1728 static int read_partial_message_pages(struct ceph_connection
*con
,
1729 struct page
**pages
,
1730 unsigned int data_len
, bool do_datacrc
)
1736 left
= min((int)(data_len
- con
->in_msg_pos
.data_pos
),
1737 (int)(PAGE_SIZE
- con
->in_msg_pos
.page_pos
));
1739 BUG_ON(pages
== NULL
);
1740 p
= kmap(pages
[con
->in_msg_pos
.page
]);
1741 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ con
->in_msg_pos
.page_pos
,
1743 if (ret
> 0 && do_datacrc
)
1745 crc32c(con
->in_data_crc
,
1746 p
+ con
->in_msg_pos
.page_pos
, ret
);
1747 kunmap(pages
[con
->in_msg_pos
.page
]);
1750 con
->in_msg_pos
.data_pos
+= ret
;
1751 con
->in_msg_pos
.page_pos
+= ret
;
1752 if (con
->in_msg_pos
.page_pos
== PAGE_SIZE
) {
1753 con
->in_msg_pos
.page_pos
= 0;
1754 con
->in_msg_pos
.page
++;
1761 static int read_partial_message_bio(struct ceph_connection
*con
,
1762 struct bio
**bio_iter
, int *bio_seg
,
1763 unsigned int data_len
, bool do_datacrc
)
1765 struct bio_vec
*bv
= bio_iovec_idx(*bio_iter
, *bio_seg
);
1769 left
= min((int)(data_len
- con
->in_msg_pos
.data_pos
),
1770 (int)(bv
->bv_len
- con
->in_msg_pos
.page_pos
));
1772 p
= kmap(bv
->bv_page
) + bv
->bv_offset
;
1774 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ con
->in_msg_pos
.page_pos
,
1776 if (ret
> 0 && do_datacrc
)
1778 crc32c(con
->in_data_crc
,
1779 p
+ con
->in_msg_pos
.page_pos
, ret
);
1780 kunmap(bv
->bv_page
);
1783 con
->in_msg_pos
.data_pos
+= ret
;
1784 con
->in_msg_pos
.page_pos
+= ret
;
1785 if (con
->in_msg_pos
.page_pos
== bv
->bv_len
) {
1786 con
->in_msg_pos
.page_pos
= 0;
1787 iter_bio_next(bio_iter
, bio_seg
);
1795 * read (part of) a message.
1797 static int read_partial_message(struct ceph_connection
*con
)
1799 struct ceph_msg
*m
= con
->in_msg
;
1803 unsigned int front_len
, middle_len
, data_len
;
1804 bool do_datacrc
= !con
->msgr
->nocrc
;
1808 dout("read_partial_message con %p msg %p\n", con
, m
);
1811 size
= sizeof (con
->in_hdr
);
1813 ret
= read_partial(con
, end
, size
, &con
->in_hdr
);
1817 crc
= crc32c(0, &con
->in_hdr
, offsetof(struct ceph_msg_header
, crc
));
1818 if (cpu_to_le32(crc
) != con
->in_hdr
.crc
) {
1819 pr_err("read_partial_message bad hdr "
1820 " crc %u != expected %u\n",
1821 crc
, con
->in_hdr
.crc
);
1825 front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
1826 if (front_len
> CEPH_MSG_MAX_FRONT_LEN
)
1828 middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
1829 if (middle_len
> CEPH_MSG_MAX_DATA_LEN
)
1831 data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
1832 if (data_len
> CEPH_MSG_MAX_DATA_LEN
)
1836 seq
= le64_to_cpu(con
->in_hdr
.seq
);
1837 if ((s64
)seq
- (s64
)con
->in_seq
< 1) {
1838 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1839 ENTITY_NAME(con
->peer_name
),
1840 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1841 seq
, con
->in_seq
+ 1);
1842 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1844 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1846 } else if ((s64
)seq
- (s64
)con
->in_seq
> 1) {
1847 pr_err("read_partial_message bad seq %lld expected %lld\n",
1848 seq
, con
->in_seq
+ 1);
1849 con
->error_msg
= "bad message sequence # for incoming message";
1853 /* allocate message? */
1855 dout("got hdr type %d front %d data %d\n", con
->in_hdr
.type
,
1856 con
->in_hdr
.front_len
, con
->in_hdr
.data_len
);
1857 if (ceph_con_in_msg_alloc(con
, &con
->in_hdr
)) {
1858 /* skip this message */
1859 dout("alloc_msg said skip message\n");
1860 BUG_ON(con
->in_msg
);
1861 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1863 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1869 "error allocating memory for incoming message";
1873 BUG_ON(con
->in_msg
->con
!= con
);
1875 m
->front
.iov_len
= 0; /* haven't read it yet */
1877 m
->middle
->vec
.iov_len
= 0;
1879 con
->in_msg_pos
.page
= 0;
1881 con
->in_msg_pos
.page_pos
= m
->page_alignment
;
1883 con
->in_msg_pos
.page_pos
= 0;
1884 con
->in_msg_pos
.data_pos
= 0;
1888 init_bio_iter(m
->bio
, &m
->bio_iter
, &m
->bio_seg
);
1893 ret
= read_partial_message_section(con
, &m
->front
, front_len
,
1894 &con
->in_front_crc
);
1900 ret
= read_partial_message_section(con
, &m
->middle
->vec
,
1902 &con
->in_middle_crc
);
1908 while (con
->in_msg_pos
.data_pos
< data_len
) {
1910 ret
= read_partial_message_pages(con
, m
->pages
,
1911 data_len
, do_datacrc
);
1915 } else if (m
->bio
) {
1916 BUG_ON(!m
->bio_iter
);
1917 ret
= read_partial_message_bio(con
,
1918 &m
->bio_iter
, &m
->bio_seg
,
1919 data_len
, do_datacrc
);
1929 size
= sizeof (m
->footer
);
1931 ret
= read_partial(con
, end
, size
, &m
->footer
);
1935 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1936 m
, front_len
, m
->footer
.front_crc
, middle_len
,
1937 m
->footer
.middle_crc
, data_len
, m
->footer
.data_crc
);
1940 if (con
->in_front_crc
!= le32_to_cpu(m
->footer
.front_crc
)) {
1941 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1942 m
, con
->in_front_crc
, m
->footer
.front_crc
);
1945 if (con
->in_middle_crc
!= le32_to_cpu(m
->footer
.middle_crc
)) {
1946 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1947 m
, con
->in_middle_crc
, m
->footer
.middle_crc
);
1951 (m
->footer
.flags
& CEPH_MSG_FOOTER_NOCRC
) == 0 &&
1952 con
->in_data_crc
!= le32_to_cpu(m
->footer
.data_crc
)) {
1953 pr_err("read_partial_message %p data crc %u != exp. %u\n", m
,
1954 con
->in_data_crc
, le32_to_cpu(m
->footer
.data_crc
));
1958 return 1; /* done! */
1962 * Process message. This happens in the worker thread. The callback should
1963 * be careful not to do anything that waits on other incoming messages or it
1966 static void process_message(struct ceph_connection
*con
)
1968 struct ceph_msg
*msg
;
1970 BUG_ON(con
->in_msg
->con
!= con
);
1971 con
->in_msg
->con
= NULL
;
1976 /* if first message, set peer_name */
1977 if (con
->peer_name
.type
== 0)
1978 con
->peer_name
= msg
->hdr
.src
;
1981 mutex_unlock(&con
->mutex
);
1983 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
1984 msg
, le64_to_cpu(msg
->hdr
.seq
),
1985 ENTITY_NAME(msg
->hdr
.src
),
1986 le16_to_cpu(msg
->hdr
.type
),
1987 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
1988 le32_to_cpu(msg
->hdr
.front_len
),
1989 le32_to_cpu(msg
->hdr
.data_len
),
1990 con
->in_front_crc
, con
->in_middle_crc
, con
->in_data_crc
);
1991 con
->ops
->dispatch(con
, msg
);
1993 mutex_lock(&con
->mutex
);
1994 prepare_read_tag(con
);
1999 * Write something to the socket. Called in a worker thread when the
2000 * socket appears to be writeable and we have something ready to send.
2002 static int try_write(struct ceph_connection
*con
)
2006 dout("try_write start %p state %lu\n", con
, con
->state
);
2009 dout("try_write out_kvec_bytes %d\n", con
->out_kvec_bytes
);
2011 /* open the socket first? */
2012 if (con
->state
== CON_STATE_PREOPEN
) {
2014 con
->state
= CON_STATE_CONNECTING
;
2016 con_out_kvec_reset(con
);
2017 prepare_write_banner(con
);
2018 prepare_read_banner(con
);
2020 BUG_ON(con
->in_msg
);
2021 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2022 dout("try_write initiating connect on %p new state %lu\n",
2024 ret
= ceph_tcp_connect(con
);
2026 con
->error_msg
= "connect error";
2032 /* kvec data queued? */
2033 if (con
->out_skip
) {
2034 ret
= write_partial_skip(con
);
2038 if (con
->out_kvec_left
) {
2039 ret
= write_partial_kvec(con
);
2046 if (con
->out_msg_done
) {
2047 ceph_msg_put(con
->out_msg
);
2048 con
->out_msg
= NULL
; /* we're done with this one */
2052 ret
= write_partial_msg_pages(con
);
2054 goto more_kvec
; /* we need to send the footer, too! */
2058 dout("try_write write_partial_msg_pages err %d\n",
2065 if (con
->state
== CON_STATE_OPEN
) {
2066 /* is anything else pending? */
2067 if (!list_empty(&con
->out_queue
)) {
2068 prepare_write_message(con
);
2071 if (con
->in_seq
> con
->in_seq_acked
) {
2072 prepare_write_ack(con
);
2075 if (test_and_clear_bit(CON_FLAG_KEEPALIVE_PENDING
,
2077 prepare_write_keepalive(con
);
2082 /* Nothing to do! */
2083 clear_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
);
2084 dout("try_write nothing else to write.\n");
2087 dout("try_write done on %p ret %d\n", con
, ret
);
2094 * Read what we can from the socket.
2096 static int try_read(struct ceph_connection
*con
)
2101 dout("try_read start on %p state %lu\n", con
, con
->state
);
2102 if (con
->state
!= CON_STATE_CONNECTING
&&
2103 con
->state
!= CON_STATE_NEGOTIATING
&&
2104 con
->state
!= CON_STATE_OPEN
)
2109 dout("try_read tag %d in_base_pos %d\n", (int)con
->in_tag
,
2112 if (con
->state
== CON_STATE_CONNECTING
) {
2113 dout("try_read connecting\n");
2114 ret
= read_partial_banner(con
);
2117 ret
= process_banner(con
);
2121 BUG_ON(con
->state
!= CON_STATE_CONNECTING
);
2122 con
->state
= CON_STATE_NEGOTIATING
;
2124 /* Banner is good, exchange connection info */
2125 ret
= prepare_write_connect(con
);
2128 prepare_read_connect(con
);
2130 /* Send connection info before awaiting response */
2134 if (con
->state
== CON_STATE_NEGOTIATING
) {
2135 dout("try_read negotiating\n");
2136 ret
= read_partial_connect(con
);
2139 ret
= process_connect(con
);
2145 BUG_ON(con
->state
!= CON_STATE_OPEN
);
2147 if (con
->in_base_pos
< 0) {
2149 * skipping + discarding content.
2151 * FIXME: there must be a better way to do this!
2153 static char buf
[SKIP_BUF_SIZE
];
2154 int skip
= min((int) sizeof (buf
), -con
->in_base_pos
);
2156 dout("skipping %d / %d bytes\n", skip
, -con
->in_base_pos
);
2157 ret
= ceph_tcp_recvmsg(con
->sock
, buf
, skip
);
2160 con
->in_base_pos
+= ret
;
2161 if (con
->in_base_pos
)
2164 if (con
->in_tag
== CEPH_MSGR_TAG_READY
) {
2168 ret
= ceph_tcp_recvmsg(con
->sock
, &con
->in_tag
, 1);
2171 dout("try_read got tag %d\n", (int)con
->in_tag
);
2172 switch (con
->in_tag
) {
2173 case CEPH_MSGR_TAG_MSG
:
2174 prepare_read_message(con
);
2176 case CEPH_MSGR_TAG_ACK
:
2177 prepare_read_ack(con
);
2179 case CEPH_MSGR_TAG_CLOSE
:
2180 con_close_socket(con
);
2181 con
->state
= CON_STATE_CLOSED
;
2187 if (con
->in_tag
== CEPH_MSGR_TAG_MSG
) {
2188 ret
= read_partial_message(con
);
2192 con
->error_msg
= "bad crc";
2196 con
->error_msg
= "io error";
2201 if (con
->in_tag
== CEPH_MSGR_TAG_READY
)
2203 process_message(con
);
2206 if (con
->in_tag
== CEPH_MSGR_TAG_ACK
) {
2207 ret
= read_partial_ack(con
);
2215 dout("try_read done on %p ret %d\n", con
, ret
);
2219 pr_err("try_read bad con->in_tag = %d\n", (int)con
->in_tag
);
2220 con
->error_msg
= "protocol error, garbage tag";
2227 * Atomically queue work on a connection. Bump @con reference to
2228 * avoid races with connection teardown.
2230 static void queue_con(struct ceph_connection
*con
)
2232 if (!con
->ops
->get(con
)) {
2233 dout("queue_con %p ref count 0\n", con
);
2237 if (!queue_delayed_work(ceph_msgr_wq
, &con
->work
, 0)) {
2238 dout("queue_con %p - already queued\n", con
);
2241 dout("queue_con %p\n", con
);
2246 * Do some work on a connection. Drop a connection ref when we're done.
2248 static void con_work(struct work_struct
*work
)
2250 struct ceph_connection
*con
= container_of(work
, struct ceph_connection
,
2254 mutex_lock(&con
->mutex
);
2256 if (test_and_clear_bit(CON_FLAG_SOCK_CLOSED
, &con
->flags
)) {
2257 switch (con
->state
) {
2258 case CON_STATE_CONNECTING
:
2259 con
->error_msg
= "connection failed";
2261 case CON_STATE_NEGOTIATING
:
2262 con
->error_msg
= "negotiation failed";
2264 case CON_STATE_OPEN
:
2265 con
->error_msg
= "socket closed";
2268 dout("unrecognized con state %d\n", (int)con
->state
);
2269 con
->error_msg
= "unrecognized con state";
2275 if (test_and_clear_bit(CON_FLAG_BACKOFF
, &con
->flags
)) {
2276 dout("con_work %p backing off\n", con
);
2277 if (queue_delayed_work(ceph_msgr_wq
, &con
->work
,
2278 round_jiffies_relative(con
->delay
))) {
2279 dout("con_work %p backoff %lu\n", con
, con
->delay
);
2280 mutex_unlock(&con
->mutex
);
2284 dout("con_work %p FAILED to back off %lu\n", con
,
2289 if (con
->state
== CON_STATE_STANDBY
) {
2290 dout("con_work %p STANDBY\n", con
);
2293 if (con
->state
== CON_STATE_CLOSED
) {
2294 dout("con_work %p CLOSED\n", con
);
2298 if (con
->state
== CON_STATE_PREOPEN
) {
2299 dout("con_work OPENING\n");
2303 ret
= try_read(con
);
2307 con
->error_msg
= "socket error on read";
2311 ret
= try_write(con
);
2315 con
->error_msg
= "socket error on write";
2320 mutex_unlock(&con
->mutex
);
2326 mutex_unlock(&con
->mutex
);
2327 ceph_fault(con
); /* error/fault path */
2333 * Generic error/fault handler. A retry mechanism is used with
2334 * exponential backoff
2336 static void ceph_fault(struct ceph_connection
*con
)
2338 mutex_lock(&con
->mutex
);
2340 pr_err("%s%lld %s %s\n", ENTITY_NAME(con
->peer_name
),
2341 ceph_pr_addr(&con
->peer_addr
.in_addr
), con
->error_msg
);
2342 dout("fault %p state %lu to peer %s\n",
2343 con
, con
->state
, ceph_pr_addr(&con
->peer_addr
.in_addr
));
2345 BUG_ON(con
->state
!= CON_STATE_CONNECTING
&&
2346 con
->state
!= CON_STATE_NEGOTIATING
&&
2347 con
->state
!= CON_STATE_OPEN
);
2349 con_close_socket(con
);
2351 if (test_bit(CON_FLAG_LOSSYTX
, &con
->flags
)) {
2352 dout("fault on LOSSYTX channel, marking CLOSED\n");
2353 con
->state
= CON_STATE_CLOSED
;
2358 BUG_ON(con
->in_msg
->con
!= con
);
2359 con
->in_msg
->con
= NULL
;
2360 ceph_msg_put(con
->in_msg
);
2365 /* Requeue anything that hasn't been acked */
2366 list_splice_init(&con
->out_sent
, &con
->out_queue
);
2368 /* If there are no messages queued or keepalive pending, place
2369 * the connection in a STANDBY state */
2370 if (list_empty(&con
->out_queue
) &&
2371 !test_bit(CON_FLAG_KEEPALIVE_PENDING
, &con
->flags
)) {
2372 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con
);
2373 clear_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
);
2374 con
->state
= CON_STATE_STANDBY
;
2376 /* retry after a delay. */
2377 con
->state
= CON_STATE_PREOPEN
;
2378 if (con
->delay
== 0)
2379 con
->delay
= BASE_DELAY_INTERVAL
;
2380 else if (con
->delay
< MAX_DELAY_INTERVAL
)
2383 if (queue_delayed_work(ceph_msgr_wq
, &con
->work
,
2384 round_jiffies_relative(con
->delay
))) {
2385 dout("fault queued %p delay %lu\n", con
, con
->delay
);
2388 dout("fault failed to queue %p delay %lu, backoff\n",
2391 * In many cases we see a socket state change
2392 * while con_work is running and end up
2393 * queuing (non-delayed) work, such that we
2394 * can't backoff with a delay. Set a flag so
2395 * that when con_work restarts we schedule the
2398 set_bit(CON_FLAG_BACKOFF
, &con
->flags
);
2403 mutex_unlock(&con
->mutex
);
2405 * in case we faulted due to authentication, invalidate our
2406 * current tickets so that we can get new ones.
2408 if (con
->auth_retry
&& con
->ops
->invalidate_authorizer
) {
2409 dout("calling invalidate_authorizer()\n");
2410 con
->ops
->invalidate_authorizer(con
);
2413 if (con
->ops
->fault
)
2414 con
->ops
->fault(con
);
2420 * initialize a new messenger instance
2422 void ceph_messenger_init(struct ceph_messenger
*msgr
,
2423 struct ceph_entity_addr
*myaddr
,
2424 u32 supported_features
,
2425 u32 required_features
,
2428 msgr
->supported_features
= supported_features
;
2429 msgr
->required_features
= required_features
;
2431 spin_lock_init(&msgr
->global_seq_lock
);
2434 msgr
->inst
.addr
= *myaddr
;
2436 /* select a random nonce */
2437 msgr
->inst
.addr
.type
= 0;
2438 get_random_bytes(&msgr
->inst
.addr
.nonce
, sizeof(msgr
->inst
.addr
.nonce
));
2439 encode_my_addr(msgr
);
2440 msgr
->nocrc
= nocrc
;
2442 atomic_set(&msgr
->stopping
, 0);
2444 dout("%s %p\n", __func__
, msgr
);
2446 EXPORT_SYMBOL(ceph_messenger_init
);
2448 static void clear_standby(struct ceph_connection
*con
)
2450 /* come back from STANDBY? */
2451 if (con
->state
== CON_STATE_STANDBY
) {
2452 dout("clear_standby %p and ++connect_seq\n", con
);
2453 con
->state
= CON_STATE_PREOPEN
;
2455 WARN_ON(test_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
));
2456 WARN_ON(test_bit(CON_FLAG_KEEPALIVE_PENDING
, &con
->flags
));
2461 * Queue up an outgoing message on the given connection.
2463 void ceph_con_send(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2466 msg
->hdr
.src
= con
->msgr
->inst
.name
;
2467 BUG_ON(msg
->front
.iov_len
!= le32_to_cpu(msg
->hdr
.front_len
));
2468 msg
->needs_out_seq
= true;
2470 mutex_lock(&con
->mutex
);
2472 if (con
->state
== CON_STATE_CLOSED
) {
2473 dout("con_send %p closed, dropping %p\n", con
, msg
);
2475 mutex_unlock(&con
->mutex
);
2479 BUG_ON(msg
->con
!= NULL
);
2480 msg
->con
= con
->ops
->get(con
);
2481 BUG_ON(msg
->con
== NULL
);
2483 BUG_ON(!list_empty(&msg
->list_head
));
2484 list_add_tail(&msg
->list_head
, &con
->out_queue
);
2485 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg
,
2486 ENTITY_NAME(con
->peer_name
), le16_to_cpu(msg
->hdr
.type
),
2487 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2488 le32_to_cpu(msg
->hdr
.front_len
),
2489 le32_to_cpu(msg
->hdr
.middle_len
),
2490 le32_to_cpu(msg
->hdr
.data_len
));
2493 mutex_unlock(&con
->mutex
);
2495 /* if there wasn't anything waiting to send before, queue
2497 if (test_and_set_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
) == 0)
2500 EXPORT_SYMBOL(ceph_con_send
);
2503 * Revoke a message that was previously queued for send
2505 void ceph_msg_revoke(struct ceph_msg
*msg
)
2507 struct ceph_connection
*con
= msg
->con
;
2510 return; /* Message not in our possession */
2512 mutex_lock(&con
->mutex
);
2513 if (!list_empty(&msg
->list_head
)) {
2514 dout("%s %p msg %p - was on queue\n", __func__
, con
, msg
);
2515 list_del_init(&msg
->list_head
);
2516 BUG_ON(msg
->con
== NULL
);
2517 msg
->con
->ops
->put(msg
->con
);
2523 if (con
->out_msg
== msg
) {
2524 dout("%s %p msg %p - was sending\n", __func__
, con
, msg
);
2525 con
->out_msg
= NULL
;
2526 if (con
->out_kvec_is_msg
) {
2527 con
->out_skip
= con
->out_kvec_bytes
;
2528 con
->out_kvec_is_msg
= false;
2534 mutex_unlock(&con
->mutex
);
2538 * Revoke a message that we may be reading data into
2540 void ceph_msg_revoke_incoming(struct ceph_msg
*msg
)
2542 struct ceph_connection
*con
;
2544 BUG_ON(msg
== NULL
);
2546 dout("%s msg %p null con\n", __func__
, msg
);
2548 return; /* Message not in our possession */
2552 mutex_lock(&con
->mutex
);
2553 if (con
->in_msg
== msg
) {
2554 unsigned int front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
2555 unsigned int middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
2556 unsigned int data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
2558 /* skip rest of message */
2559 dout("%s %p msg %p revoked\n", __func__
, con
, msg
);
2560 con
->in_base_pos
= con
->in_base_pos
-
2561 sizeof(struct ceph_msg_header
) -
2565 sizeof(struct ceph_msg_footer
);
2566 ceph_msg_put(con
->in_msg
);
2568 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2571 dout("%s %p in_msg %p msg %p no-op\n",
2572 __func__
, con
, con
->in_msg
, msg
);
2574 mutex_unlock(&con
->mutex
);
2578 * Queue a keepalive byte to ensure the tcp connection is alive.
2580 void ceph_con_keepalive(struct ceph_connection
*con
)
2582 dout("con_keepalive %p\n", con
);
2583 mutex_lock(&con
->mutex
);
2585 mutex_unlock(&con
->mutex
);
2586 if (test_and_set_bit(CON_FLAG_KEEPALIVE_PENDING
, &con
->flags
) == 0 &&
2587 test_and_set_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
) == 0)
2590 EXPORT_SYMBOL(ceph_con_keepalive
);
2594 * construct a new message with given type, size
2595 * the new msg has a ref count of 1.
2597 struct ceph_msg
*ceph_msg_new(int type
, int front_len
, gfp_t flags
,
2602 m
= kmalloc(sizeof(*m
), flags
);
2605 kref_init(&m
->kref
);
2608 INIT_LIST_HEAD(&m
->list_head
);
2611 m
->hdr
.type
= cpu_to_le16(type
);
2612 m
->hdr
.priority
= cpu_to_le16(CEPH_MSG_PRIO_DEFAULT
);
2614 m
->hdr
.front_len
= cpu_to_le32(front_len
);
2615 m
->hdr
.middle_len
= 0;
2616 m
->hdr
.data_len
= 0;
2617 m
->hdr
.data_off
= 0;
2618 m
->hdr
.reserved
= 0;
2619 m
->footer
.front_crc
= 0;
2620 m
->footer
.middle_crc
= 0;
2621 m
->footer
.data_crc
= 0;
2622 m
->footer
.flags
= 0;
2623 m
->front_max
= front_len
;
2624 m
->front_is_vmalloc
= false;
2625 m
->more_to_follow
= false;
2634 m
->page_alignment
= 0;
2644 if (front_len
> PAGE_CACHE_SIZE
) {
2645 m
->front
.iov_base
= __vmalloc(front_len
, flags
,
2647 m
->front_is_vmalloc
= true;
2649 m
->front
.iov_base
= kmalloc(front_len
, flags
);
2651 if (m
->front
.iov_base
== NULL
) {
2652 dout("ceph_msg_new can't allocate %d bytes\n",
2657 m
->front
.iov_base
= NULL
;
2659 m
->front
.iov_len
= front_len
;
2661 dout("ceph_msg_new %p front %d\n", m
, front_len
);
2668 pr_err("msg_new can't create type %d front %d\n", type
,
2672 dout("msg_new can't create type %d front %d\n", type
,
2677 EXPORT_SYMBOL(ceph_msg_new
);
2680 * Allocate "middle" portion of a message, if it is needed and wasn't
2681 * allocated by alloc_msg. This allows us to read a small fixed-size
2682 * per-type header in the front and then gracefully fail (i.e.,
2683 * propagate the error to the caller based on info in the front) when
2684 * the middle is too large.
2686 static int ceph_alloc_middle(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2688 int type
= le16_to_cpu(msg
->hdr
.type
);
2689 int middle_len
= le32_to_cpu(msg
->hdr
.middle_len
);
2691 dout("alloc_middle %p type %d %s middle_len %d\n", msg
, type
,
2692 ceph_msg_type_name(type
), middle_len
);
2693 BUG_ON(!middle_len
);
2694 BUG_ON(msg
->middle
);
2696 msg
->middle
= ceph_buffer_new(middle_len
, GFP_NOFS
);
2703 * Allocate a message for receiving an incoming message on a
2704 * connection, and save the result in con->in_msg. Uses the
2705 * connection's private alloc_msg op if available.
2707 * Returns true if the message should be skipped, false otherwise.
2708 * If true is returned (skip message), con->in_msg will be NULL.
2709 * If false is returned, con->in_msg will contain a pointer to the
2710 * newly-allocated message, or NULL in case of memory exhaustion.
2712 static bool ceph_con_in_msg_alloc(struct ceph_connection
*con
,
2713 struct ceph_msg_header
*hdr
)
2715 int type
= le16_to_cpu(hdr
->type
);
2716 int front_len
= le32_to_cpu(hdr
->front_len
);
2717 int middle_len
= le32_to_cpu(hdr
->middle_len
);
2720 BUG_ON(con
->in_msg
!= NULL
);
2722 if (con
->ops
->alloc_msg
) {
2725 mutex_unlock(&con
->mutex
);
2726 con
->in_msg
= con
->ops
->alloc_msg(con
, hdr
, &skip
);
2727 mutex_lock(&con
->mutex
);
2729 con
->in_msg
->con
= con
->ops
->get(con
);
2730 BUG_ON(con
->in_msg
->con
== NULL
);
2739 con
->in_msg
= ceph_msg_new(type
, front_len
, GFP_NOFS
, false);
2741 pr_err("unable to allocate msg type %d len %d\n",
2745 con
->in_msg
->con
= con
->ops
->get(con
);
2746 BUG_ON(con
->in_msg
->con
== NULL
);
2747 con
->in_msg
->page_alignment
= le16_to_cpu(hdr
->data_off
);
2749 memcpy(&con
->in_msg
->hdr
, &con
->in_hdr
, sizeof(con
->in_hdr
));
2751 if (middle_len
&& !con
->in_msg
->middle
) {
2752 ret
= ceph_alloc_middle(con
, con
->in_msg
);
2754 ceph_msg_put(con
->in_msg
);
2764 * Free a generically kmalloc'd message.
2766 void ceph_msg_kfree(struct ceph_msg
*m
)
2768 dout("msg_kfree %p\n", m
);
2769 if (m
->front_is_vmalloc
)
2770 vfree(m
->front
.iov_base
);
2772 kfree(m
->front
.iov_base
);
2777 * Drop a msg ref. Destroy as needed.
2779 void ceph_msg_last_put(struct kref
*kref
)
2781 struct ceph_msg
*m
= container_of(kref
, struct ceph_msg
, kref
);
2783 dout("ceph_msg_put last one on %p\n", m
);
2784 WARN_ON(!list_empty(&m
->list_head
));
2786 /* drop middle, data, if any */
2788 ceph_buffer_put(m
->middle
);
2795 ceph_pagelist_release(m
->pagelist
);
2803 ceph_msgpool_put(m
->pool
, m
);
2807 EXPORT_SYMBOL(ceph_msg_last_put
);
2809 void ceph_msg_dump(struct ceph_msg
*msg
)
2811 pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg
,
2812 msg
->front_max
, msg
->nr_pages
);
2813 print_hex_dump(KERN_DEBUG
, "header: ",
2814 DUMP_PREFIX_OFFSET
, 16, 1,
2815 &msg
->hdr
, sizeof(msg
->hdr
), true);
2816 print_hex_dump(KERN_DEBUG
, " front: ",
2817 DUMP_PREFIX_OFFSET
, 16, 1,
2818 msg
->front
.iov_base
, msg
->front
.iov_len
, true);
2820 print_hex_dump(KERN_DEBUG
, "middle: ",
2821 DUMP_PREFIX_OFFSET
, 16, 1,
2822 msg
->middle
->vec
.iov_base
,
2823 msg
->middle
->vec
.iov_len
, true);
2824 print_hex_dump(KERN_DEBUG
, "footer: ",
2825 DUMP_PREFIX_OFFSET
, 16, 1,
2826 &msg
->footer
, sizeof(msg
->footer
), true);
2828 EXPORT_SYMBOL(ceph_msg_dump
);