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
);
532 reset_connection(con
);
533 con
->peer_global_seq
= 0;
534 cancel_delayed_work(&con
->work
);
535 con_close_socket(con
);
536 mutex_unlock(&con
->mutex
);
538 EXPORT_SYMBOL(ceph_con_close
);
541 * Reopen a closed connection, with a new peer address.
543 void ceph_con_open(struct ceph_connection
*con
,
544 __u8 entity_type
, __u64 entity_num
,
545 struct ceph_entity_addr
*addr
)
547 mutex_lock(&con
->mutex
);
548 dout("con_open %p %s\n", con
, ceph_pr_addr(&addr
->in_addr
));
550 BUG_ON(con
->state
!= CON_STATE_CLOSED
);
551 con
->state
= CON_STATE_PREOPEN
;
553 con
->peer_name
.type
= (__u8
) entity_type
;
554 con
->peer_name
.num
= cpu_to_le64(entity_num
);
556 memcpy(&con
->peer_addr
, addr
, sizeof(*addr
));
557 con
->delay
= 0; /* reset backoff memory */
558 mutex_unlock(&con
->mutex
);
561 EXPORT_SYMBOL(ceph_con_open
);
564 * return true if this connection ever successfully opened
566 bool ceph_con_opened(struct ceph_connection
*con
)
568 return con
->connect_seq
> 0;
572 * initialize a new connection.
574 void ceph_con_init(struct ceph_connection
*con
, void *private,
575 const struct ceph_connection_operations
*ops
,
576 struct ceph_messenger
*msgr
)
578 dout("con_init %p\n", con
);
579 memset(con
, 0, sizeof(*con
));
580 con
->private = private;
584 con_sock_state_init(con
);
586 mutex_init(&con
->mutex
);
587 INIT_LIST_HEAD(&con
->out_queue
);
588 INIT_LIST_HEAD(&con
->out_sent
);
589 INIT_DELAYED_WORK(&con
->work
, con_work
);
591 con
->state
= CON_STATE_CLOSED
;
593 EXPORT_SYMBOL(ceph_con_init
);
597 * We maintain a global counter to order connection attempts. Get
598 * a unique seq greater than @gt.
600 static u32
get_global_seq(struct ceph_messenger
*msgr
, u32 gt
)
604 spin_lock(&msgr
->global_seq_lock
);
605 if (msgr
->global_seq
< gt
)
606 msgr
->global_seq
= gt
;
607 ret
= ++msgr
->global_seq
;
608 spin_unlock(&msgr
->global_seq_lock
);
612 static void con_out_kvec_reset(struct ceph_connection
*con
)
614 con
->out_kvec_left
= 0;
615 con
->out_kvec_bytes
= 0;
616 con
->out_kvec_cur
= &con
->out_kvec
[0];
619 static void con_out_kvec_add(struct ceph_connection
*con
,
620 size_t size
, void *data
)
624 index
= con
->out_kvec_left
;
625 BUG_ON(index
>= ARRAY_SIZE(con
->out_kvec
));
627 con
->out_kvec
[index
].iov_len
= size
;
628 con
->out_kvec
[index
].iov_base
= data
;
629 con
->out_kvec_left
++;
630 con
->out_kvec_bytes
+= size
;
634 static void init_bio_iter(struct bio
*bio
, struct bio
**iter
, int *seg
)
645 static void iter_bio_next(struct bio
**bio_iter
, int *seg
)
647 if (*bio_iter
== NULL
)
650 BUG_ON(*seg
>= (*bio_iter
)->bi_vcnt
);
653 if (*seg
== (*bio_iter
)->bi_vcnt
)
654 init_bio_iter((*bio_iter
)->bi_next
, bio_iter
, seg
);
658 static void prepare_write_message_data(struct ceph_connection
*con
)
660 struct ceph_msg
*msg
= con
->out_msg
;
663 BUG_ON(!msg
->hdr
.data_len
);
665 /* initialize page iterator */
666 con
->out_msg_pos
.page
= 0;
668 con
->out_msg_pos
.page_pos
= msg
->page_alignment
;
670 con
->out_msg_pos
.page_pos
= 0;
673 init_bio_iter(msg
->bio
, &msg
->bio_iter
, &msg
->bio_seg
);
675 con
->out_msg_pos
.data_pos
= 0;
676 con
->out_msg_pos
.did_page_crc
= false;
677 con
->out_more
= 1; /* data + footer will follow */
681 * Prepare footer for currently outgoing message, and finish things
682 * off. Assumes out_kvec* are already valid.. we just add on to the end.
684 static void prepare_write_message_footer(struct ceph_connection
*con
)
686 struct ceph_msg
*m
= con
->out_msg
;
687 int v
= con
->out_kvec_left
;
689 m
->footer
.flags
|= CEPH_MSG_FOOTER_COMPLETE
;
691 dout("prepare_write_message_footer %p\n", con
);
692 con
->out_kvec_is_msg
= true;
693 con
->out_kvec
[v
].iov_base
= &m
->footer
;
694 con
->out_kvec
[v
].iov_len
= sizeof(m
->footer
);
695 con
->out_kvec_bytes
+= sizeof(m
->footer
);
696 con
->out_kvec_left
++;
697 con
->out_more
= m
->more_to_follow
;
698 con
->out_msg_done
= true;
702 * Prepare headers for the next outgoing message.
704 static void prepare_write_message(struct ceph_connection
*con
)
709 con_out_kvec_reset(con
);
710 con
->out_kvec_is_msg
= true;
711 con
->out_msg_done
= false;
713 /* Sneak an ack in there first? If we can get it into the same
714 * TCP packet that's a good thing. */
715 if (con
->in_seq
> con
->in_seq_acked
) {
716 con
->in_seq_acked
= con
->in_seq
;
717 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
718 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
719 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
723 BUG_ON(list_empty(&con
->out_queue
));
724 m
= list_first_entry(&con
->out_queue
, struct ceph_msg
, list_head
);
726 BUG_ON(m
->con
!= con
);
728 /* put message on sent list */
730 list_move_tail(&m
->list_head
, &con
->out_sent
);
733 * only assign outgoing seq # if we haven't sent this message
734 * yet. if it is requeued, resend with it's original seq.
736 if (m
->needs_out_seq
) {
737 m
->hdr
.seq
= cpu_to_le64(++con
->out_seq
);
738 m
->needs_out_seq
= false;
741 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n",
742 m
, con
->out_seq
, le16_to_cpu(m
->hdr
.type
),
743 le32_to_cpu(m
->hdr
.front_len
), le32_to_cpu(m
->hdr
.middle_len
),
744 le32_to_cpu(m
->hdr
.data_len
),
746 BUG_ON(le32_to_cpu(m
->hdr
.front_len
) != m
->front
.iov_len
);
748 /* tag + hdr + front + middle */
749 con_out_kvec_add(con
, sizeof (tag_msg
), &tag_msg
);
750 con_out_kvec_add(con
, sizeof (m
->hdr
), &m
->hdr
);
751 con_out_kvec_add(con
, m
->front
.iov_len
, m
->front
.iov_base
);
754 con_out_kvec_add(con
, m
->middle
->vec
.iov_len
,
755 m
->middle
->vec
.iov_base
);
757 /* fill in crc (except data pages), footer */
758 crc
= crc32c(0, &m
->hdr
, offsetof(struct ceph_msg_header
, crc
));
759 con
->out_msg
->hdr
.crc
= cpu_to_le32(crc
);
760 con
->out_msg
->footer
.flags
= 0;
762 crc
= crc32c(0, m
->front
.iov_base
, m
->front
.iov_len
);
763 con
->out_msg
->footer
.front_crc
= cpu_to_le32(crc
);
765 crc
= crc32c(0, m
->middle
->vec
.iov_base
,
766 m
->middle
->vec
.iov_len
);
767 con
->out_msg
->footer
.middle_crc
= cpu_to_le32(crc
);
769 con
->out_msg
->footer
.middle_crc
= 0;
770 dout("%s front_crc %u middle_crc %u\n", __func__
,
771 le32_to_cpu(con
->out_msg
->footer
.front_crc
),
772 le32_to_cpu(con
->out_msg
->footer
.middle_crc
));
774 /* is there a data payload? */
775 con
->out_msg
->footer
.data_crc
= 0;
777 prepare_write_message_data(con
);
779 /* no, queue up footer too and be done */
780 prepare_write_message_footer(con
);
782 set_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
);
788 static void prepare_write_ack(struct ceph_connection
*con
)
790 dout("prepare_write_ack %p %llu -> %llu\n", con
,
791 con
->in_seq_acked
, con
->in_seq
);
792 con
->in_seq_acked
= con
->in_seq
;
794 con_out_kvec_reset(con
);
796 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
798 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
799 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
802 con
->out_more
= 1; /* more will follow.. eventually.. */
803 set_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
);
807 * Prepare to write keepalive byte.
809 static void prepare_write_keepalive(struct ceph_connection
*con
)
811 dout("prepare_write_keepalive %p\n", con
);
812 con_out_kvec_reset(con
);
813 con_out_kvec_add(con
, sizeof (tag_keepalive
), &tag_keepalive
);
814 set_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
);
818 * Connection negotiation.
821 static struct ceph_auth_handshake
*get_connect_authorizer(struct ceph_connection
*con
,
824 struct ceph_auth_handshake
*auth
;
826 if (!con
->ops
->get_authorizer
) {
827 con
->out_connect
.authorizer_protocol
= CEPH_AUTH_UNKNOWN
;
828 con
->out_connect
.authorizer_len
= 0;
832 /* Can't hold the mutex while getting authorizer */
833 mutex_unlock(&con
->mutex
);
834 auth
= con
->ops
->get_authorizer(con
, auth_proto
, con
->auth_retry
);
835 mutex_lock(&con
->mutex
);
839 if (con
->state
!= CON_STATE_NEGOTIATING
)
840 return ERR_PTR(-EAGAIN
);
842 con
->auth_reply_buf
= auth
->authorizer_reply_buf
;
843 con
->auth_reply_buf_len
= auth
->authorizer_reply_buf_len
;
848 * We connected to a peer and are saying hello.
850 static void prepare_write_banner(struct ceph_connection
*con
)
852 con_out_kvec_add(con
, strlen(CEPH_BANNER
), CEPH_BANNER
);
853 con_out_kvec_add(con
, sizeof (con
->msgr
->my_enc_addr
),
854 &con
->msgr
->my_enc_addr
);
857 set_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
);
860 static int prepare_write_connect(struct ceph_connection
*con
)
862 unsigned int global_seq
= get_global_seq(con
->msgr
, 0);
865 struct ceph_auth_handshake
*auth
;
867 switch (con
->peer_name
.type
) {
868 case CEPH_ENTITY_TYPE_MON
:
869 proto
= CEPH_MONC_PROTOCOL
;
871 case CEPH_ENTITY_TYPE_OSD
:
872 proto
= CEPH_OSDC_PROTOCOL
;
874 case CEPH_ENTITY_TYPE_MDS
:
875 proto
= CEPH_MDSC_PROTOCOL
;
881 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con
,
882 con
->connect_seq
, global_seq
, proto
);
884 con
->out_connect
.features
= cpu_to_le64(con
->msgr
->supported_features
);
885 con
->out_connect
.host_type
= cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT
);
886 con
->out_connect
.connect_seq
= cpu_to_le32(con
->connect_seq
);
887 con
->out_connect
.global_seq
= cpu_to_le32(global_seq
);
888 con
->out_connect
.protocol_version
= cpu_to_le32(proto
);
889 con
->out_connect
.flags
= 0;
891 auth_proto
= CEPH_AUTH_UNKNOWN
;
892 auth
= get_connect_authorizer(con
, &auth_proto
);
894 return PTR_ERR(auth
);
896 con
->out_connect
.authorizer_protocol
= cpu_to_le32(auth_proto
);
897 con
->out_connect
.authorizer_len
= auth
?
898 cpu_to_le32(auth
->authorizer_buf_len
) : 0;
900 con_out_kvec_reset(con
);
901 con_out_kvec_add(con
, sizeof (con
->out_connect
),
903 if (auth
&& auth
->authorizer_buf_len
)
904 con_out_kvec_add(con
, auth
->authorizer_buf_len
,
905 auth
->authorizer_buf
);
908 set_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
);
914 * write as much of pending kvecs to the socket as we can.
916 * 0 -> socket full, but more to do
919 static int write_partial_kvec(struct ceph_connection
*con
)
923 dout("write_partial_kvec %p %d left\n", con
, con
->out_kvec_bytes
);
924 while (con
->out_kvec_bytes
> 0) {
925 ret
= ceph_tcp_sendmsg(con
->sock
, con
->out_kvec_cur
,
926 con
->out_kvec_left
, con
->out_kvec_bytes
,
930 con
->out_kvec_bytes
-= ret
;
931 if (con
->out_kvec_bytes
== 0)
934 /* account for full iov entries consumed */
935 while (ret
>= con
->out_kvec_cur
->iov_len
) {
936 BUG_ON(!con
->out_kvec_left
);
937 ret
-= con
->out_kvec_cur
->iov_len
;
939 con
->out_kvec_left
--;
941 /* and for a partially-consumed entry */
943 con
->out_kvec_cur
->iov_len
-= ret
;
944 con
->out_kvec_cur
->iov_base
+= ret
;
947 con
->out_kvec_left
= 0;
948 con
->out_kvec_is_msg
= false;
951 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con
,
952 con
->out_kvec_bytes
, con
->out_kvec_left
, ret
);
953 return ret
; /* done! */
956 static void out_msg_pos_next(struct ceph_connection
*con
, struct page
*page
,
957 size_t len
, size_t sent
, bool in_trail
)
959 struct ceph_msg
*msg
= con
->out_msg
;
964 con
->out_msg_pos
.data_pos
+= sent
;
965 con
->out_msg_pos
.page_pos
+= sent
;
970 con
->out_msg_pos
.page_pos
= 0;
971 con
->out_msg_pos
.page
++;
972 con
->out_msg_pos
.did_page_crc
= false;
974 list_move_tail(&page
->lru
,
976 else if (msg
->pagelist
)
977 list_move_tail(&page
->lru
,
978 &msg
->pagelist
->head
);
981 iter_bio_next(&msg
->bio_iter
, &msg
->bio_seg
);
986 * Write as much message data payload as we can. If we finish, queue
988 * 1 -> done, footer is now queued in out_kvec[].
989 * 0 -> socket full, but more to do
992 static int write_partial_msg_pages(struct ceph_connection
*con
)
994 struct ceph_msg
*msg
= con
->out_msg
;
995 unsigned int data_len
= le32_to_cpu(msg
->hdr
.data_len
);
997 bool do_datacrc
= !con
->msgr
->nocrc
;
1000 bool in_trail
= false;
1001 const size_t trail_len
= (msg
->trail
? msg
->trail
->length
: 0);
1002 const size_t trail_off
= data_len
- trail_len
;
1004 dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
1005 con
, msg
, con
->out_msg_pos
.page
, msg
->nr_pages
,
1006 con
->out_msg_pos
.page_pos
);
1009 * Iterate through each page that contains data to be
1010 * written, and send as much as possible for each.
1012 * If we are calculating the data crc (the default), we will
1013 * need to map the page. If we have no pages, they have
1014 * been revoked, so use the zero page.
1016 while (data_len
> con
->out_msg_pos
.data_pos
) {
1017 struct page
*page
= NULL
;
1018 int max_write
= PAGE_SIZE
;
1021 in_trail
= in_trail
|| con
->out_msg_pos
.data_pos
>= trail_off
;
1023 total_max_write
= trail_off
- con
->out_msg_pos
.data_pos
;
1026 total_max_write
= data_len
- con
->out_msg_pos
.data_pos
;
1028 page
= list_first_entry(&msg
->trail
->head
,
1030 } else if (msg
->pages
) {
1031 page
= msg
->pages
[con
->out_msg_pos
.page
];
1032 } else if (msg
->pagelist
) {
1033 page
= list_first_entry(&msg
->pagelist
->head
,
1036 } else if (msg
->bio
) {
1039 bv
= bio_iovec_idx(msg
->bio_iter
, msg
->bio_seg
);
1041 bio_offset
= bv
->bv_offset
;
1042 max_write
= bv
->bv_len
;
1047 len
= min_t(int, max_write
- con
->out_msg_pos
.page_pos
,
1050 if (do_datacrc
&& !con
->out_msg_pos
.did_page_crc
) {
1052 u32 crc
= le32_to_cpu(msg
->footer
.data_crc
);
1056 BUG_ON(kaddr
== NULL
);
1057 base
= kaddr
+ con
->out_msg_pos
.page_pos
+ bio_offset
;
1058 crc
= crc32c(crc
, base
, len
);
1059 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1060 con
->out_msg_pos
.did_page_crc
= true;
1062 ret
= ceph_tcp_sendpage(con
->sock
, page
,
1063 con
->out_msg_pos
.page_pos
+ bio_offset
,
1072 out_msg_pos_next(con
, page
, len
, (size_t) ret
, in_trail
);
1075 dout("write_partial_msg_pages %p msg %p done\n", con
, msg
);
1077 /* prepare and queue up footer, too */
1079 msg
->footer
.flags
|= CEPH_MSG_FOOTER_NOCRC
;
1080 con_out_kvec_reset(con
);
1081 prepare_write_message_footer(con
);
1090 static int write_partial_skip(struct ceph_connection
*con
)
1094 while (con
->out_skip
> 0) {
1095 size_t size
= min(con
->out_skip
, (int) PAGE_CACHE_SIZE
);
1097 ret
= ceph_tcp_sendpage(con
->sock
, zero_page
, 0, size
, 1);
1100 con
->out_skip
-= ret
;
1108 * Prepare to read connection handshake, or an ack.
1110 static void prepare_read_banner(struct ceph_connection
*con
)
1112 dout("prepare_read_banner %p\n", con
);
1113 con
->in_base_pos
= 0;
1116 static void prepare_read_connect(struct ceph_connection
*con
)
1118 dout("prepare_read_connect %p\n", con
);
1119 con
->in_base_pos
= 0;
1122 static void prepare_read_ack(struct ceph_connection
*con
)
1124 dout("prepare_read_ack %p\n", con
);
1125 con
->in_base_pos
= 0;
1128 static void prepare_read_tag(struct ceph_connection
*con
)
1130 dout("prepare_read_tag %p\n", con
);
1131 con
->in_base_pos
= 0;
1132 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1136 * Prepare to read a message.
1138 static int prepare_read_message(struct ceph_connection
*con
)
1140 dout("prepare_read_message %p\n", con
);
1141 BUG_ON(con
->in_msg
!= NULL
);
1142 con
->in_base_pos
= 0;
1143 con
->in_front_crc
= con
->in_middle_crc
= con
->in_data_crc
= 0;
1148 static int read_partial(struct ceph_connection
*con
,
1149 int end
, int size
, void *object
)
1151 while (con
->in_base_pos
< end
) {
1152 int left
= end
- con
->in_base_pos
;
1153 int have
= size
- left
;
1154 int ret
= ceph_tcp_recvmsg(con
->sock
, object
+ have
, left
);
1157 con
->in_base_pos
+= ret
;
1164 * Read all or part of the connect-side handshake on a new connection
1166 static int read_partial_banner(struct ceph_connection
*con
)
1172 dout("read_partial_banner %p at %d\n", con
, con
->in_base_pos
);
1175 size
= strlen(CEPH_BANNER
);
1177 ret
= read_partial(con
, end
, size
, con
->in_banner
);
1181 size
= sizeof (con
->actual_peer_addr
);
1183 ret
= read_partial(con
, end
, size
, &con
->actual_peer_addr
);
1187 size
= sizeof (con
->peer_addr_for_me
);
1189 ret
= read_partial(con
, end
, size
, &con
->peer_addr_for_me
);
1197 static int read_partial_connect(struct ceph_connection
*con
)
1203 dout("read_partial_connect %p at %d\n", con
, con
->in_base_pos
);
1205 size
= sizeof (con
->in_reply
);
1207 ret
= read_partial(con
, end
, size
, &con
->in_reply
);
1211 size
= le32_to_cpu(con
->in_reply
.authorizer_len
);
1213 ret
= read_partial(con
, end
, size
, con
->auth_reply_buf
);
1217 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1218 con
, (int)con
->in_reply
.tag
,
1219 le32_to_cpu(con
->in_reply
.connect_seq
),
1220 le32_to_cpu(con
->in_reply
.global_seq
));
1227 * Verify the hello banner looks okay.
1229 static int verify_hello(struct ceph_connection
*con
)
1231 if (memcmp(con
->in_banner
, CEPH_BANNER
, strlen(CEPH_BANNER
))) {
1232 pr_err("connect to %s got bad banner\n",
1233 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1234 con
->error_msg
= "protocol error, bad banner";
1240 static bool addr_is_blank(struct sockaddr_storage
*ss
)
1242 switch (ss
->ss_family
) {
1244 return ((struct sockaddr_in
*)ss
)->sin_addr
.s_addr
== 0;
1247 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[0] == 0 &&
1248 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[1] == 0 &&
1249 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[2] == 0 &&
1250 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[3] == 0;
1255 static int addr_port(struct sockaddr_storage
*ss
)
1257 switch (ss
->ss_family
) {
1259 return ntohs(((struct sockaddr_in
*)ss
)->sin_port
);
1261 return ntohs(((struct sockaddr_in6
*)ss
)->sin6_port
);
1266 static void addr_set_port(struct sockaddr_storage
*ss
, int p
)
1268 switch (ss
->ss_family
) {
1270 ((struct sockaddr_in
*)ss
)->sin_port
= htons(p
);
1273 ((struct sockaddr_in6
*)ss
)->sin6_port
= htons(p
);
1279 * Unlike other *_pton function semantics, zero indicates success.
1281 static int ceph_pton(const char *str
, size_t len
, struct sockaddr_storage
*ss
,
1282 char delim
, const char **ipend
)
1284 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
1285 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
1287 memset(ss
, 0, sizeof(*ss
));
1289 if (in4_pton(str
, len
, (u8
*)&in4
->sin_addr
.s_addr
, delim
, ipend
)) {
1290 ss
->ss_family
= AF_INET
;
1294 if (in6_pton(str
, len
, (u8
*)&in6
->sin6_addr
.s6_addr
, delim
, ipend
)) {
1295 ss
->ss_family
= AF_INET6
;
1303 * Extract hostname string and resolve using kernel DNS facility.
1305 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1306 static int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1307 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1309 const char *end
, *delim_p
;
1310 char *colon_p
, *ip_addr
= NULL
;
1314 * The end of the hostname occurs immediately preceding the delimiter or
1315 * the port marker (':') where the delimiter takes precedence.
1317 delim_p
= memchr(name
, delim
, namelen
);
1318 colon_p
= memchr(name
, ':', namelen
);
1320 if (delim_p
&& colon_p
)
1321 end
= delim_p
< colon_p
? delim_p
: colon_p
;
1322 else if (!delim_p
&& colon_p
)
1326 if (!end
) /* case: hostname:/ */
1327 end
= name
+ namelen
;
1333 /* do dns_resolve upcall */
1334 ip_len
= dns_query(NULL
, name
, end
- name
, NULL
, &ip_addr
, NULL
);
1336 ret
= ceph_pton(ip_addr
, ip_len
, ss
, -1, NULL
);
1344 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end
- name
), name
,
1345 ret
, ret
? "failed" : ceph_pr_addr(ss
));
1350 static inline int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1351 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1358 * Parse a server name (IP or hostname). If a valid IP address is not found
1359 * then try to extract a hostname to resolve using userspace DNS upcall.
1361 static int ceph_parse_server_name(const char *name
, size_t namelen
,
1362 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1366 ret
= ceph_pton(name
, namelen
, ss
, delim
, ipend
);
1368 ret
= ceph_dns_resolve_name(name
, namelen
, ss
, delim
, ipend
);
1374 * Parse an ip[:port] list into an addr array. Use the default
1375 * monitor port if a port isn't specified.
1377 int ceph_parse_ips(const char *c
, const char *end
,
1378 struct ceph_entity_addr
*addr
,
1379 int max_count
, int *count
)
1381 int i
, ret
= -EINVAL
;
1384 dout("parse_ips on '%.*s'\n", (int)(end
-c
), c
);
1385 for (i
= 0; i
< max_count
; i
++) {
1387 struct sockaddr_storage
*ss
= &addr
[i
].in_addr
;
1396 ret
= ceph_parse_server_name(p
, end
- p
, ss
, delim
, &ipend
);
1405 dout("missing matching ']'\n");
1412 if (p
< end
&& *p
== ':') {
1415 while (p
< end
&& *p
>= '0' && *p
<= '9') {
1416 port
= (port
* 10) + (*p
- '0');
1419 if (port
> 65535 || port
== 0)
1422 port
= CEPH_MON_PORT
;
1425 addr_set_port(ss
, port
);
1427 dout("parse_ips got %s\n", ceph_pr_addr(ss
));
1444 pr_err("parse_ips bad ip '%.*s'\n", (int)(end
- c
), c
);
1447 EXPORT_SYMBOL(ceph_parse_ips
);
1449 static int process_banner(struct ceph_connection
*con
)
1451 dout("process_banner on %p\n", con
);
1453 if (verify_hello(con
) < 0)
1456 ceph_decode_addr(&con
->actual_peer_addr
);
1457 ceph_decode_addr(&con
->peer_addr_for_me
);
1460 * Make sure the other end is who we wanted. note that the other
1461 * end may not yet know their ip address, so if it's 0.0.0.0, give
1462 * them the benefit of the doubt.
1464 if (memcmp(&con
->peer_addr
, &con
->actual_peer_addr
,
1465 sizeof(con
->peer_addr
)) != 0 &&
1466 !(addr_is_blank(&con
->actual_peer_addr
.in_addr
) &&
1467 con
->actual_peer_addr
.nonce
== con
->peer_addr
.nonce
)) {
1468 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1469 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1470 (int)le32_to_cpu(con
->peer_addr
.nonce
),
1471 ceph_pr_addr(&con
->actual_peer_addr
.in_addr
),
1472 (int)le32_to_cpu(con
->actual_peer_addr
.nonce
));
1473 con
->error_msg
= "wrong peer at address";
1478 * did we learn our address?
1480 if (addr_is_blank(&con
->msgr
->inst
.addr
.in_addr
)) {
1481 int port
= addr_port(&con
->msgr
->inst
.addr
.in_addr
);
1483 memcpy(&con
->msgr
->inst
.addr
.in_addr
,
1484 &con
->peer_addr_for_me
.in_addr
,
1485 sizeof(con
->peer_addr_for_me
.in_addr
));
1486 addr_set_port(&con
->msgr
->inst
.addr
.in_addr
, port
);
1487 encode_my_addr(con
->msgr
);
1488 dout("process_banner learned my addr is %s\n",
1489 ceph_pr_addr(&con
->msgr
->inst
.addr
.in_addr
));
1495 static void fail_protocol(struct ceph_connection
*con
)
1497 reset_connection(con
);
1498 BUG_ON(con
->state
!= CON_STATE_NEGOTIATING
);
1499 con
->state
= CON_STATE_CLOSED
;
1502 static int process_connect(struct ceph_connection
*con
)
1504 u64 sup_feat
= con
->msgr
->supported_features
;
1505 u64 req_feat
= con
->msgr
->required_features
;
1506 u64 server_feat
= le64_to_cpu(con
->in_reply
.features
);
1509 dout("process_connect on %p tag %d\n", con
, (int)con
->in_tag
);
1511 switch (con
->in_reply
.tag
) {
1512 case CEPH_MSGR_TAG_FEATURES
:
1513 pr_err("%s%lld %s feature set mismatch,"
1514 " my %llx < server's %llx, missing %llx\n",
1515 ENTITY_NAME(con
->peer_name
),
1516 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1517 sup_feat
, server_feat
, server_feat
& ~sup_feat
);
1518 con
->error_msg
= "missing required protocol features";
1522 case CEPH_MSGR_TAG_BADPROTOVER
:
1523 pr_err("%s%lld %s protocol version mismatch,"
1524 " my %d != server's %d\n",
1525 ENTITY_NAME(con
->peer_name
),
1526 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1527 le32_to_cpu(con
->out_connect
.protocol_version
),
1528 le32_to_cpu(con
->in_reply
.protocol_version
));
1529 con
->error_msg
= "protocol version mismatch";
1533 case CEPH_MSGR_TAG_BADAUTHORIZER
:
1535 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con
,
1537 if (con
->auth_retry
== 2) {
1538 con
->error_msg
= "connect authorization failure";
1541 con
->auth_retry
= 1;
1542 ret
= prepare_write_connect(con
);
1545 prepare_read_connect(con
);
1548 case CEPH_MSGR_TAG_RESETSESSION
:
1550 * If we connected with a large connect_seq but the peer
1551 * has no record of a session with us (no connection, or
1552 * connect_seq == 0), they will send RESETSESION to indicate
1553 * that they must have reset their session, and may have
1556 dout("process_connect got RESET peer seq %u\n",
1557 le32_to_cpu(con
->in_reply
.connect_seq
));
1558 pr_err("%s%lld %s connection reset\n",
1559 ENTITY_NAME(con
->peer_name
),
1560 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1561 reset_connection(con
);
1562 ret
= prepare_write_connect(con
);
1565 prepare_read_connect(con
);
1567 /* Tell ceph about it. */
1568 mutex_unlock(&con
->mutex
);
1569 pr_info("reset on %s%lld\n", ENTITY_NAME(con
->peer_name
));
1570 if (con
->ops
->peer_reset
)
1571 con
->ops
->peer_reset(con
);
1572 mutex_lock(&con
->mutex
);
1573 if (con
->state
!= CON_STATE_NEGOTIATING
)
1577 case CEPH_MSGR_TAG_RETRY_SESSION
:
1579 * If we sent a smaller connect_seq than the peer has, try
1580 * again with a larger value.
1582 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
1583 le32_to_cpu(con
->out_connect
.connect_seq
),
1584 le32_to_cpu(con
->in_reply
.connect_seq
));
1585 con
->connect_seq
= le32_to_cpu(con
->in_reply
.connect_seq
);
1586 ret
= prepare_write_connect(con
);
1589 prepare_read_connect(con
);
1592 case CEPH_MSGR_TAG_RETRY_GLOBAL
:
1594 * If we sent a smaller global_seq than the peer has, try
1595 * again with a larger value.
1597 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1598 con
->peer_global_seq
,
1599 le32_to_cpu(con
->in_reply
.global_seq
));
1600 get_global_seq(con
->msgr
,
1601 le32_to_cpu(con
->in_reply
.global_seq
));
1602 ret
= prepare_write_connect(con
);
1605 prepare_read_connect(con
);
1608 case CEPH_MSGR_TAG_READY
:
1609 if (req_feat
& ~server_feat
) {
1610 pr_err("%s%lld %s protocol feature mismatch,"
1611 " my required %llx > server's %llx, need %llx\n",
1612 ENTITY_NAME(con
->peer_name
),
1613 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1614 req_feat
, server_feat
, req_feat
& ~server_feat
);
1615 con
->error_msg
= "missing required protocol features";
1620 BUG_ON(con
->state
!= CON_STATE_NEGOTIATING
);
1621 con
->state
= CON_STATE_OPEN
;
1623 con
->peer_global_seq
= le32_to_cpu(con
->in_reply
.global_seq
);
1625 con
->peer_features
= server_feat
;
1626 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1627 con
->peer_global_seq
,
1628 le32_to_cpu(con
->in_reply
.connect_seq
),
1630 WARN_ON(con
->connect_seq
!=
1631 le32_to_cpu(con
->in_reply
.connect_seq
));
1633 if (con
->in_reply
.flags
& CEPH_MSG_CONNECT_LOSSY
)
1634 set_bit(CON_FLAG_LOSSYTX
, &con
->flags
);
1636 con
->delay
= 0; /* reset backoff memory */
1638 prepare_read_tag(con
);
1641 case CEPH_MSGR_TAG_WAIT
:
1643 * If there is a connection race (we are opening
1644 * connections to each other), one of us may just have
1645 * to WAIT. This shouldn't happen if we are the
1648 pr_err("process_connect got WAIT as client\n");
1649 con
->error_msg
= "protocol error, got WAIT as client";
1653 pr_err("connect protocol error, will retry\n");
1654 con
->error_msg
= "protocol error, garbage tag during connect";
1662 * read (part of) an ack
1664 static int read_partial_ack(struct ceph_connection
*con
)
1666 int size
= sizeof (con
->in_temp_ack
);
1669 return read_partial(con
, end
, size
, &con
->in_temp_ack
);
1674 * We can finally discard anything that's been acked.
1676 static void process_ack(struct ceph_connection
*con
)
1679 u64 ack
= le64_to_cpu(con
->in_temp_ack
);
1682 while (!list_empty(&con
->out_sent
)) {
1683 m
= list_first_entry(&con
->out_sent
, struct ceph_msg
,
1685 seq
= le64_to_cpu(m
->hdr
.seq
);
1688 dout("got ack for seq %llu type %d at %p\n", seq
,
1689 le16_to_cpu(m
->hdr
.type
), m
);
1690 m
->ack_stamp
= jiffies
;
1693 prepare_read_tag(con
);
1699 static int read_partial_message_section(struct ceph_connection
*con
,
1700 struct kvec
*section
,
1701 unsigned int sec_len
, u32
*crc
)
1707 while (section
->iov_len
< sec_len
) {
1708 BUG_ON(section
->iov_base
== NULL
);
1709 left
= sec_len
- section
->iov_len
;
1710 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)section
->iov_base
+
1711 section
->iov_len
, left
);
1714 section
->iov_len
+= ret
;
1716 if (section
->iov_len
== sec_len
)
1717 *crc
= crc32c(0, section
->iov_base
, section
->iov_len
);
1722 static bool ceph_con_in_msg_alloc(struct ceph_connection
*con
,
1723 struct ceph_msg_header
*hdr
);
1726 static int read_partial_message_pages(struct ceph_connection
*con
,
1727 struct page
**pages
,
1728 unsigned int data_len
, bool do_datacrc
)
1734 left
= min((int)(data_len
- con
->in_msg_pos
.data_pos
),
1735 (int)(PAGE_SIZE
- con
->in_msg_pos
.page_pos
));
1737 BUG_ON(pages
== NULL
);
1738 p
= kmap(pages
[con
->in_msg_pos
.page
]);
1739 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ con
->in_msg_pos
.page_pos
,
1741 if (ret
> 0 && do_datacrc
)
1743 crc32c(con
->in_data_crc
,
1744 p
+ con
->in_msg_pos
.page_pos
, ret
);
1745 kunmap(pages
[con
->in_msg_pos
.page
]);
1748 con
->in_msg_pos
.data_pos
+= ret
;
1749 con
->in_msg_pos
.page_pos
+= ret
;
1750 if (con
->in_msg_pos
.page_pos
== PAGE_SIZE
) {
1751 con
->in_msg_pos
.page_pos
= 0;
1752 con
->in_msg_pos
.page
++;
1759 static int read_partial_message_bio(struct ceph_connection
*con
,
1760 struct bio
**bio_iter
, int *bio_seg
,
1761 unsigned int data_len
, bool do_datacrc
)
1763 struct bio_vec
*bv
= bio_iovec_idx(*bio_iter
, *bio_seg
);
1767 left
= min((int)(data_len
- con
->in_msg_pos
.data_pos
),
1768 (int)(bv
->bv_len
- con
->in_msg_pos
.page_pos
));
1770 p
= kmap(bv
->bv_page
) + bv
->bv_offset
;
1772 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ con
->in_msg_pos
.page_pos
,
1774 if (ret
> 0 && do_datacrc
)
1776 crc32c(con
->in_data_crc
,
1777 p
+ con
->in_msg_pos
.page_pos
, ret
);
1778 kunmap(bv
->bv_page
);
1781 con
->in_msg_pos
.data_pos
+= ret
;
1782 con
->in_msg_pos
.page_pos
+= ret
;
1783 if (con
->in_msg_pos
.page_pos
== bv
->bv_len
) {
1784 con
->in_msg_pos
.page_pos
= 0;
1785 iter_bio_next(bio_iter
, bio_seg
);
1793 * read (part of) a message.
1795 static int read_partial_message(struct ceph_connection
*con
)
1797 struct ceph_msg
*m
= con
->in_msg
;
1801 unsigned int front_len
, middle_len
, data_len
;
1802 bool do_datacrc
= !con
->msgr
->nocrc
;
1806 dout("read_partial_message con %p msg %p\n", con
, m
);
1809 size
= sizeof (con
->in_hdr
);
1811 ret
= read_partial(con
, end
, size
, &con
->in_hdr
);
1815 crc
= crc32c(0, &con
->in_hdr
, offsetof(struct ceph_msg_header
, crc
));
1816 if (cpu_to_le32(crc
) != con
->in_hdr
.crc
) {
1817 pr_err("read_partial_message bad hdr "
1818 " crc %u != expected %u\n",
1819 crc
, con
->in_hdr
.crc
);
1823 front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
1824 if (front_len
> CEPH_MSG_MAX_FRONT_LEN
)
1826 middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
1827 if (middle_len
> CEPH_MSG_MAX_DATA_LEN
)
1829 data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
1830 if (data_len
> CEPH_MSG_MAX_DATA_LEN
)
1834 seq
= le64_to_cpu(con
->in_hdr
.seq
);
1835 if ((s64
)seq
- (s64
)con
->in_seq
< 1) {
1836 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1837 ENTITY_NAME(con
->peer_name
),
1838 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1839 seq
, con
->in_seq
+ 1);
1840 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1842 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1844 } else if ((s64
)seq
- (s64
)con
->in_seq
> 1) {
1845 pr_err("read_partial_message bad seq %lld expected %lld\n",
1846 seq
, con
->in_seq
+ 1);
1847 con
->error_msg
= "bad message sequence # for incoming message";
1851 /* allocate message? */
1853 dout("got hdr type %d front %d data %d\n", con
->in_hdr
.type
,
1854 con
->in_hdr
.front_len
, con
->in_hdr
.data_len
);
1855 if (ceph_con_in_msg_alloc(con
, &con
->in_hdr
)) {
1856 /* skip this message */
1857 dout("alloc_msg said skip message\n");
1858 BUG_ON(con
->in_msg
);
1859 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1861 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1867 "error allocating memory for incoming message";
1871 BUG_ON(con
->in_msg
->con
!= con
);
1873 m
->front
.iov_len
= 0; /* haven't read it yet */
1875 m
->middle
->vec
.iov_len
= 0;
1877 con
->in_msg_pos
.page
= 0;
1879 con
->in_msg_pos
.page_pos
= m
->page_alignment
;
1881 con
->in_msg_pos
.page_pos
= 0;
1882 con
->in_msg_pos
.data_pos
= 0;
1886 init_bio_iter(m
->bio
, &m
->bio_iter
, &m
->bio_seg
);
1891 ret
= read_partial_message_section(con
, &m
->front
, front_len
,
1892 &con
->in_front_crc
);
1898 ret
= read_partial_message_section(con
, &m
->middle
->vec
,
1900 &con
->in_middle_crc
);
1906 while (con
->in_msg_pos
.data_pos
< data_len
) {
1908 ret
= read_partial_message_pages(con
, m
->pages
,
1909 data_len
, do_datacrc
);
1913 } else if (m
->bio
) {
1914 BUG_ON(!m
->bio_iter
);
1915 ret
= read_partial_message_bio(con
,
1916 &m
->bio_iter
, &m
->bio_seg
,
1917 data_len
, do_datacrc
);
1927 size
= sizeof (m
->footer
);
1929 ret
= read_partial(con
, end
, size
, &m
->footer
);
1933 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1934 m
, front_len
, m
->footer
.front_crc
, middle_len
,
1935 m
->footer
.middle_crc
, data_len
, m
->footer
.data_crc
);
1938 if (con
->in_front_crc
!= le32_to_cpu(m
->footer
.front_crc
)) {
1939 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1940 m
, con
->in_front_crc
, m
->footer
.front_crc
);
1943 if (con
->in_middle_crc
!= le32_to_cpu(m
->footer
.middle_crc
)) {
1944 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1945 m
, con
->in_middle_crc
, m
->footer
.middle_crc
);
1949 (m
->footer
.flags
& CEPH_MSG_FOOTER_NOCRC
) == 0 &&
1950 con
->in_data_crc
!= le32_to_cpu(m
->footer
.data_crc
)) {
1951 pr_err("read_partial_message %p data crc %u != exp. %u\n", m
,
1952 con
->in_data_crc
, le32_to_cpu(m
->footer
.data_crc
));
1956 return 1; /* done! */
1960 * Process message. This happens in the worker thread. The callback should
1961 * be careful not to do anything that waits on other incoming messages or it
1964 static void process_message(struct ceph_connection
*con
)
1966 struct ceph_msg
*msg
;
1968 BUG_ON(con
->in_msg
->con
!= con
);
1969 con
->in_msg
->con
= NULL
;
1974 /* if first message, set peer_name */
1975 if (con
->peer_name
.type
== 0)
1976 con
->peer_name
= msg
->hdr
.src
;
1979 mutex_unlock(&con
->mutex
);
1981 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
1982 msg
, le64_to_cpu(msg
->hdr
.seq
),
1983 ENTITY_NAME(msg
->hdr
.src
),
1984 le16_to_cpu(msg
->hdr
.type
),
1985 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
1986 le32_to_cpu(msg
->hdr
.front_len
),
1987 le32_to_cpu(msg
->hdr
.data_len
),
1988 con
->in_front_crc
, con
->in_middle_crc
, con
->in_data_crc
);
1989 con
->ops
->dispatch(con
, msg
);
1991 mutex_lock(&con
->mutex
);
1992 prepare_read_tag(con
);
1997 * Write something to the socket. Called in a worker thread when the
1998 * socket appears to be writeable and we have something ready to send.
2000 static int try_write(struct ceph_connection
*con
)
2004 dout("try_write start %p state %lu\n", con
, con
->state
);
2007 dout("try_write out_kvec_bytes %d\n", con
->out_kvec_bytes
);
2009 /* open the socket first? */
2010 if (con
->state
== CON_STATE_PREOPEN
) {
2012 con
->state
= CON_STATE_CONNECTING
;
2014 con_out_kvec_reset(con
);
2015 prepare_write_banner(con
);
2016 prepare_read_banner(con
);
2018 BUG_ON(con
->in_msg
);
2019 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2020 dout("try_write initiating connect on %p new state %lu\n",
2022 ret
= ceph_tcp_connect(con
);
2024 con
->error_msg
= "connect error";
2030 /* kvec data queued? */
2031 if (con
->out_skip
) {
2032 ret
= write_partial_skip(con
);
2036 if (con
->out_kvec_left
) {
2037 ret
= write_partial_kvec(con
);
2044 if (con
->out_msg_done
) {
2045 ceph_msg_put(con
->out_msg
);
2046 con
->out_msg
= NULL
; /* we're done with this one */
2050 ret
= write_partial_msg_pages(con
);
2052 goto more_kvec
; /* we need to send the footer, too! */
2056 dout("try_write write_partial_msg_pages err %d\n",
2063 if (con
->state
== CON_STATE_OPEN
) {
2064 /* is anything else pending? */
2065 if (!list_empty(&con
->out_queue
)) {
2066 prepare_write_message(con
);
2069 if (con
->in_seq
> con
->in_seq_acked
) {
2070 prepare_write_ack(con
);
2073 if (test_and_clear_bit(CON_FLAG_KEEPALIVE_PENDING
,
2075 prepare_write_keepalive(con
);
2080 /* Nothing to do! */
2081 clear_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
);
2082 dout("try_write nothing else to write.\n");
2085 dout("try_write done on %p ret %d\n", con
, ret
);
2092 * Read what we can from the socket.
2094 static int try_read(struct ceph_connection
*con
)
2099 dout("try_read start on %p state %lu\n", con
, con
->state
);
2100 if (con
->state
!= CON_STATE_CONNECTING
&&
2101 con
->state
!= CON_STATE_NEGOTIATING
&&
2102 con
->state
!= CON_STATE_OPEN
)
2107 dout("try_read tag %d in_base_pos %d\n", (int)con
->in_tag
,
2110 if (con
->state
== CON_STATE_CONNECTING
) {
2111 dout("try_read connecting\n");
2112 ret
= read_partial_banner(con
);
2115 ret
= process_banner(con
);
2119 BUG_ON(con
->state
!= CON_STATE_CONNECTING
);
2120 con
->state
= CON_STATE_NEGOTIATING
;
2122 /* Banner is good, exchange connection info */
2123 ret
= prepare_write_connect(con
);
2126 prepare_read_connect(con
);
2128 /* Send connection info before awaiting response */
2132 if (con
->state
== CON_STATE_NEGOTIATING
) {
2133 dout("try_read negotiating\n");
2134 ret
= read_partial_connect(con
);
2137 ret
= process_connect(con
);
2143 BUG_ON(con
->state
!= CON_STATE_OPEN
);
2145 if (con
->in_base_pos
< 0) {
2147 * skipping + discarding content.
2149 * FIXME: there must be a better way to do this!
2151 static char buf
[SKIP_BUF_SIZE
];
2152 int skip
= min((int) sizeof (buf
), -con
->in_base_pos
);
2154 dout("skipping %d / %d bytes\n", skip
, -con
->in_base_pos
);
2155 ret
= ceph_tcp_recvmsg(con
->sock
, buf
, skip
);
2158 con
->in_base_pos
+= ret
;
2159 if (con
->in_base_pos
)
2162 if (con
->in_tag
== CEPH_MSGR_TAG_READY
) {
2166 ret
= ceph_tcp_recvmsg(con
->sock
, &con
->in_tag
, 1);
2169 dout("try_read got tag %d\n", (int)con
->in_tag
);
2170 switch (con
->in_tag
) {
2171 case CEPH_MSGR_TAG_MSG
:
2172 prepare_read_message(con
);
2174 case CEPH_MSGR_TAG_ACK
:
2175 prepare_read_ack(con
);
2177 case CEPH_MSGR_TAG_CLOSE
:
2178 con_close_socket(con
);
2179 con
->state
= CON_STATE_CLOSED
;
2185 if (con
->in_tag
== CEPH_MSGR_TAG_MSG
) {
2186 ret
= read_partial_message(con
);
2190 con
->error_msg
= "bad crc";
2194 con
->error_msg
= "io error";
2199 if (con
->in_tag
== CEPH_MSGR_TAG_READY
)
2201 process_message(con
);
2204 if (con
->in_tag
== CEPH_MSGR_TAG_ACK
) {
2205 ret
= read_partial_ack(con
);
2213 dout("try_read done on %p ret %d\n", con
, ret
);
2217 pr_err("try_read bad con->in_tag = %d\n", (int)con
->in_tag
);
2218 con
->error_msg
= "protocol error, garbage tag";
2225 * Atomically queue work on a connection. Bump @con reference to
2226 * avoid races with connection teardown.
2228 static void queue_con(struct ceph_connection
*con
)
2230 if (!con
->ops
->get(con
)) {
2231 dout("queue_con %p ref count 0\n", con
);
2235 if (!queue_delayed_work(ceph_msgr_wq
, &con
->work
, 0)) {
2236 dout("queue_con %p - already queued\n", con
);
2239 dout("queue_con %p\n", con
);
2244 * Do some work on a connection. Drop a connection ref when we're done.
2246 static void con_work(struct work_struct
*work
)
2248 struct ceph_connection
*con
= container_of(work
, struct ceph_connection
,
2252 mutex_lock(&con
->mutex
);
2254 if (test_and_clear_bit(CON_FLAG_SOCK_CLOSED
, &con
->flags
)) {
2255 switch (con
->state
) {
2256 case CON_STATE_CONNECTING
:
2257 con
->error_msg
= "connection failed";
2259 case CON_STATE_NEGOTIATING
:
2260 con
->error_msg
= "negotiation failed";
2262 case CON_STATE_OPEN
:
2263 con
->error_msg
= "socket closed";
2266 dout("unrecognized con state %d\n", (int)con
->state
);
2267 con
->error_msg
= "unrecognized con state";
2273 if (test_and_clear_bit(CON_FLAG_BACKOFF
, &con
->flags
)) {
2274 dout("con_work %p backing off\n", con
);
2275 if (queue_delayed_work(ceph_msgr_wq
, &con
->work
,
2276 round_jiffies_relative(con
->delay
))) {
2277 dout("con_work %p backoff %lu\n", con
, con
->delay
);
2278 mutex_unlock(&con
->mutex
);
2282 dout("con_work %p FAILED to back off %lu\n", con
,
2287 if (con
->state
== CON_STATE_STANDBY
) {
2288 dout("con_work %p STANDBY\n", con
);
2291 if (con
->state
== CON_STATE_CLOSED
) {
2292 dout("con_work %p CLOSED\n", con
);
2296 if (con
->state
== CON_STATE_PREOPEN
) {
2297 dout("con_work OPENING\n");
2301 ret
= try_read(con
);
2305 con
->error_msg
= "socket error on read";
2309 ret
= try_write(con
);
2313 con
->error_msg
= "socket error on write";
2318 mutex_unlock(&con
->mutex
);
2324 mutex_unlock(&con
->mutex
);
2325 ceph_fault(con
); /* error/fault path */
2331 * Generic error/fault handler. A retry mechanism is used with
2332 * exponential backoff
2334 static void ceph_fault(struct ceph_connection
*con
)
2336 mutex_lock(&con
->mutex
);
2338 pr_err("%s%lld %s %s\n", ENTITY_NAME(con
->peer_name
),
2339 ceph_pr_addr(&con
->peer_addr
.in_addr
), con
->error_msg
);
2340 dout("fault %p state %lu to peer %s\n",
2341 con
, con
->state
, ceph_pr_addr(&con
->peer_addr
.in_addr
));
2343 BUG_ON(con
->state
!= CON_STATE_CONNECTING
&&
2344 con
->state
!= CON_STATE_NEGOTIATING
&&
2345 con
->state
!= CON_STATE_OPEN
);
2347 con_close_socket(con
);
2349 if (test_bit(CON_FLAG_LOSSYTX
, &con
->flags
)) {
2350 dout("fault on LOSSYTX channel, marking CLOSED\n");
2351 con
->state
= CON_STATE_CLOSED
;
2356 BUG_ON(con
->in_msg
->con
!= con
);
2357 con
->in_msg
->con
= NULL
;
2358 ceph_msg_put(con
->in_msg
);
2363 /* Requeue anything that hasn't been acked */
2364 list_splice_init(&con
->out_sent
, &con
->out_queue
);
2366 /* If there are no messages queued or keepalive pending, place
2367 * the connection in a STANDBY state */
2368 if (list_empty(&con
->out_queue
) &&
2369 !test_bit(CON_FLAG_KEEPALIVE_PENDING
, &con
->flags
)) {
2370 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con
);
2371 clear_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
);
2372 con
->state
= CON_STATE_STANDBY
;
2374 /* retry after a delay. */
2375 con
->state
= CON_STATE_PREOPEN
;
2376 if (con
->delay
== 0)
2377 con
->delay
= BASE_DELAY_INTERVAL
;
2378 else if (con
->delay
< MAX_DELAY_INTERVAL
)
2381 if (queue_delayed_work(ceph_msgr_wq
, &con
->work
,
2382 round_jiffies_relative(con
->delay
))) {
2383 dout("fault queued %p delay %lu\n", con
, con
->delay
);
2386 dout("fault failed to queue %p delay %lu, backoff\n",
2389 * In many cases we see a socket state change
2390 * while con_work is running and end up
2391 * queuing (non-delayed) work, such that we
2392 * can't backoff with a delay. Set a flag so
2393 * that when con_work restarts we schedule the
2396 set_bit(CON_FLAG_BACKOFF
, &con
->flags
);
2401 mutex_unlock(&con
->mutex
);
2403 * in case we faulted due to authentication, invalidate our
2404 * current tickets so that we can get new ones.
2406 if (con
->auth_retry
&& con
->ops
->invalidate_authorizer
) {
2407 dout("calling invalidate_authorizer()\n");
2408 con
->ops
->invalidate_authorizer(con
);
2411 if (con
->ops
->fault
)
2412 con
->ops
->fault(con
);
2418 * initialize a new messenger instance
2420 void ceph_messenger_init(struct ceph_messenger
*msgr
,
2421 struct ceph_entity_addr
*myaddr
,
2422 u32 supported_features
,
2423 u32 required_features
,
2426 msgr
->supported_features
= supported_features
;
2427 msgr
->required_features
= required_features
;
2429 spin_lock_init(&msgr
->global_seq_lock
);
2432 msgr
->inst
.addr
= *myaddr
;
2434 /* select a random nonce */
2435 msgr
->inst
.addr
.type
= 0;
2436 get_random_bytes(&msgr
->inst
.addr
.nonce
, sizeof(msgr
->inst
.addr
.nonce
));
2437 encode_my_addr(msgr
);
2438 msgr
->nocrc
= nocrc
;
2440 atomic_set(&msgr
->stopping
, 0);
2442 dout("%s %p\n", __func__
, msgr
);
2444 EXPORT_SYMBOL(ceph_messenger_init
);
2446 static void clear_standby(struct ceph_connection
*con
)
2448 /* come back from STANDBY? */
2449 if (con
->state
== CON_STATE_STANDBY
) {
2450 dout("clear_standby %p and ++connect_seq\n", con
);
2451 con
->state
= CON_STATE_PREOPEN
;
2453 WARN_ON(test_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
));
2454 WARN_ON(test_bit(CON_FLAG_KEEPALIVE_PENDING
, &con
->flags
));
2459 * Queue up an outgoing message on the given connection.
2461 void ceph_con_send(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2464 msg
->hdr
.src
= con
->msgr
->inst
.name
;
2465 BUG_ON(msg
->front
.iov_len
!= le32_to_cpu(msg
->hdr
.front_len
));
2466 msg
->needs_out_seq
= true;
2468 mutex_lock(&con
->mutex
);
2470 if (con
->state
== CON_STATE_CLOSED
) {
2471 dout("con_send %p closed, dropping %p\n", con
, msg
);
2473 mutex_unlock(&con
->mutex
);
2477 BUG_ON(msg
->con
!= NULL
);
2478 msg
->con
= con
->ops
->get(con
);
2479 BUG_ON(msg
->con
== NULL
);
2481 BUG_ON(!list_empty(&msg
->list_head
));
2482 list_add_tail(&msg
->list_head
, &con
->out_queue
);
2483 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg
,
2484 ENTITY_NAME(con
->peer_name
), le16_to_cpu(msg
->hdr
.type
),
2485 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2486 le32_to_cpu(msg
->hdr
.front_len
),
2487 le32_to_cpu(msg
->hdr
.middle_len
),
2488 le32_to_cpu(msg
->hdr
.data_len
));
2491 mutex_unlock(&con
->mutex
);
2493 /* if there wasn't anything waiting to send before, queue
2495 if (test_and_set_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
) == 0)
2498 EXPORT_SYMBOL(ceph_con_send
);
2501 * Revoke a message that was previously queued for send
2503 void ceph_msg_revoke(struct ceph_msg
*msg
)
2505 struct ceph_connection
*con
= msg
->con
;
2508 return; /* Message not in our possession */
2510 mutex_lock(&con
->mutex
);
2511 if (!list_empty(&msg
->list_head
)) {
2512 dout("%s %p msg %p - was on queue\n", __func__
, con
, msg
);
2513 list_del_init(&msg
->list_head
);
2514 BUG_ON(msg
->con
== NULL
);
2515 msg
->con
->ops
->put(msg
->con
);
2521 if (con
->out_msg
== msg
) {
2522 dout("%s %p msg %p - was sending\n", __func__
, con
, msg
);
2523 con
->out_msg
= NULL
;
2524 if (con
->out_kvec_is_msg
) {
2525 con
->out_skip
= con
->out_kvec_bytes
;
2526 con
->out_kvec_is_msg
= false;
2532 mutex_unlock(&con
->mutex
);
2536 * Revoke a message that we may be reading data into
2538 void ceph_msg_revoke_incoming(struct ceph_msg
*msg
)
2540 struct ceph_connection
*con
;
2542 BUG_ON(msg
== NULL
);
2544 dout("%s msg %p null con\n", __func__
, msg
);
2546 return; /* Message not in our possession */
2550 mutex_lock(&con
->mutex
);
2551 if (con
->in_msg
== msg
) {
2552 unsigned int front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
2553 unsigned int middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
2554 unsigned int data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
2556 /* skip rest of message */
2557 dout("%s %p msg %p revoked\n", __func__
, con
, msg
);
2558 con
->in_base_pos
= con
->in_base_pos
-
2559 sizeof(struct ceph_msg_header
) -
2563 sizeof(struct ceph_msg_footer
);
2564 ceph_msg_put(con
->in_msg
);
2566 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2569 dout("%s %p in_msg %p msg %p no-op\n",
2570 __func__
, con
, con
->in_msg
, msg
);
2572 mutex_unlock(&con
->mutex
);
2576 * Queue a keepalive byte to ensure the tcp connection is alive.
2578 void ceph_con_keepalive(struct ceph_connection
*con
)
2580 dout("con_keepalive %p\n", con
);
2581 mutex_lock(&con
->mutex
);
2583 mutex_unlock(&con
->mutex
);
2584 if (test_and_set_bit(CON_FLAG_KEEPALIVE_PENDING
, &con
->flags
) == 0 &&
2585 test_and_set_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
) == 0)
2588 EXPORT_SYMBOL(ceph_con_keepalive
);
2592 * construct a new message with given type, size
2593 * the new msg has a ref count of 1.
2595 struct ceph_msg
*ceph_msg_new(int type
, int front_len
, gfp_t flags
,
2600 m
= kmalloc(sizeof(*m
), flags
);
2603 kref_init(&m
->kref
);
2606 INIT_LIST_HEAD(&m
->list_head
);
2609 m
->hdr
.type
= cpu_to_le16(type
);
2610 m
->hdr
.priority
= cpu_to_le16(CEPH_MSG_PRIO_DEFAULT
);
2612 m
->hdr
.front_len
= cpu_to_le32(front_len
);
2613 m
->hdr
.middle_len
= 0;
2614 m
->hdr
.data_len
= 0;
2615 m
->hdr
.data_off
= 0;
2616 m
->hdr
.reserved
= 0;
2617 m
->footer
.front_crc
= 0;
2618 m
->footer
.middle_crc
= 0;
2619 m
->footer
.data_crc
= 0;
2620 m
->footer
.flags
= 0;
2621 m
->front_max
= front_len
;
2622 m
->front_is_vmalloc
= false;
2623 m
->more_to_follow
= false;
2632 m
->page_alignment
= 0;
2642 if (front_len
> PAGE_CACHE_SIZE
) {
2643 m
->front
.iov_base
= __vmalloc(front_len
, flags
,
2645 m
->front_is_vmalloc
= true;
2647 m
->front
.iov_base
= kmalloc(front_len
, flags
);
2649 if (m
->front
.iov_base
== NULL
) {
2650 dout("ceph_msg_new can't allocate %d bytes\n",
2655 m
->front
.iov_base
= NULL
;
2657 m
->front
.iov_len
= front_len
;
2659 dout("ceph_msg_new %p front %d\n", m
, front_len
);
2666 pr_err("msg_new can't create type %d front %d\n", type
,
2670 dout("msg_new can't create type %d front %d\n", type
,
2675 EXPORT_SYMBOL(ceph_msg_new
);
2678 * Allocate "middle" portion of a message, if it is needed and wasn't
2679 * allocated by alloc_msg. This allows us to read a small fixed-size
2680 * per-type header in the front and then gracefully fail (i.e.,
2681 * propagate the error to the caller based on info in the front) when
2682 * the middle is too large.
2684 static int ceph_alloc_middle(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2686 int type
= le16_to_cpu(msg
->hdr
.type
);
2687 int middle_len
= le32_to_cpu(msg
->hdr
.middle_len
);
2689 dout("alloc_middle %p type %d %s middle_len %d\n", msg
, type
,
2690 ceph_msg_type_name(type
), middle_len
);
2691 BUG_ON(!middle_len
);
2692 BUG_ON(msg
->middle
);
2694 msg
->middle
= ceph_buffer_new(middle_len
, GFP_NOFS
);
2701 * Allocate a message for receiving an incoming message on a
2702 * connection, and save the result in con->in_msg. Uses the
2703 * connection's private alloc_msg op if available.
2705 * Returns true if the message should be skipped, false otherwise.
2706 * If true is returned (skip message), con->in_msg will be NULL.
2707 * If false is returned, con->in_msg will contain a pointer to the
2708 * newly-allocated message, or NULL in case of memory exhaustion.
2710 static bool ceph_con_in_msg_alloc(struct ceph_connection
*con
,
2711 struct ceph_msg_header
*hdr
)
2713 int type
= le16_to_cpu(hdr
->type
);
2714 int front_len
= le32_to_cpu(hdr
->front_len
);
2715 int middle_len
= le32_to_cpu(hdr
->middle_len
);
2718 BUG_ON(con
->in_msg
!= NULL
);
2720 if (con
->ops
->alloc_msg
) {
2723 mutex_unlock(&con
->mutex
);
2724 con
->in_msg
= con
->ops
->alloc_msg(con
, hdr
, &skip
);
2725 mutex_lock(&con
->mutex
);
2727 con
->in_msg
->con
= con
->ops
->get(con
);
2728 BUG_ON(con
->in_msg
->con
== NULL
);
2737 con
->in_msg
= ceph_msg_new(type
, front_len
, GFP_NOFS
, false);
2739 pr_err("unable to allocate msg type %d len %d\n",
2743 con
->in_msg
->con
= con
->ops
->get(con
);
2744 BUG_ON(con
->in_msg
->con
== NULL
);
2745 con
->in_msg
->page_alignment
= le16_to_cpu(hdr
->data_off
);
2747 memcpy(&con
->in_msg
->hdr
, &con
->in_hdr
, sizeof(con
->in_hdr
));
2749 if (middle_len
&& !con
->in_msg
->middle
) {
2750 ret
= ceph_alloc_middle(con
, con
->in_msg
);
2752 ceph_msg_put(con
->in_msg
);
2762 * Free a generically kmalloc'd message.
2764 void ceph_msg_kfree(struct ceph_msg
*m
)
2766 dout("msg_kfree %p\n", m
);
2767 if (m
->front_is_vmalloc
)
2768 vfree(m
->front
.iov_base
);
2770 kfree(m
->front
.iov_base
);
2775 * Drop a msg ref. Destroy as needed.
2777 void ceph_msg_last_put(struct kref
*kref
)
2779 struct ceph_msg
*m
= container_of(kref
, struct ceph_msg
, kref
);
2781 dout("ceph_msg_put last one on %p\n", m
);
2782 WARN_ON(!list_empty(&m
->list_head
));
2784 /* drop middle, data, if any */
2786 ceph_buffer_put(m
->middle
);
2793 ceph_pagelist_release(m
->pagelist
);
2801 ceph_msgpool_put(m
->pool
, m
);
2805 EXPORT_SYMBOL(ceph_msg_last_put
);
2807 void ceph_msg_dump(struct ceph_msg
*msg
)
2809 pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg
,
2810 msg
->front_max
, msg
->nr_pages
);
2811 print_hex_dump(KERN_DEBUG
, "header: ",
2812 DUMP_PREFIX_OFFSET
, 16, 1,
2813 &msg
->hdr
, sizeof(msg
->hdr
), true);
2814 print_hex_dump(KERN_DEBUG
, " front: ",
2815 DUMP_PREFIX_OFFSET
, 16, 1,
2816 msg
->front
.iov_base
, msg
->front
.iov_len
, true);
2818 print_hex_dump(KERN_DEBUG
, "middle: ",
2819 DUMP_PREFIX_OFFSET
, 16, 1,
2820 msg
->middle
->vec
.iov_base
,
2821 msg
->middle
->vec
.iov_len
, true);
2822 print_hex_dump(KERN_DEBUG
, "footer: ",
2823 DUMP_PREFIX_OFFSET
, 16, 1,
2824 &msg
->footer
, sizeof(msg
->footer
), true);
2826 EXPORT_SYMBOL(ceph_msg_dump
);