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
);
227 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
228 CON_SOCK_STATE_CLOSED
);
231 static void con_sock_state_connecting(struct ceph_connection
*con
)
235 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CONNECTING
);
236 if (WARN_ON(old_state
!= CON_SOCK_STATE_CLOSED
))
237 printk("%s: unexpected old state %d\n", __func__
, old_state
);
238 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
239 CON_SOCK_STATE_CONNECTING
);
242 static void con_sock_state_connected(struct ceph_connection
*con
)
246 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CONNECTED
);
247 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTING
))
248 printk("%s: unexpected old state %d\n", __func__
, old_state
);
249 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
250 CON_SOCK_STATE_CONNECTED
);
253 static void con_sock_state_closing(struct ceph_connection
*con
)
257 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSING
);
258 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTING
&&
259 old_state
!= CON_SOCK_STATE_CONNECTED
&&
260 old_state
!= CON_SOCK_STATE_CLOSING
))
261 printk("%s: unexpected old state %d\n", __func__
, old_state
);
262 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
263 CON_SOCK_STATE_CLOSING
);
266 static void con_sock_state_closed(struct ceph_connection
*con
)
270 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSED
);
271 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTED
&&
272 old_state
!= CON_SOCK_STATE_CLOSING
&&
273 old_state
!= CON_SOCK_STATE_CONNECTING
&&
274 old_state
!= CON_SOCK_STATE_CLOSED
))
275 printk("%s: unexpected old state %d\n", __func__
, old_state
);
276 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
277 CON_SOCK_STATE_CLOSED
);
281 * socket callback functions
284 /* data available on socket, or listen socket received a connect */
285 static void ceph_sock_data_ready(struct sock
*sk
, int count_unused
)
287 struct ceph_connection
*con
= sk
->sk_user_data
;
288 if (atomic_read(&con
->msgr
->stopping
)) {
292 if (sk
->sk_state
!= TCP_CLOSE_WAIT
) {
293 dout("%s on %p state = %lu, queueing work\n", __func__
,
299 /* socket has buffer space for writing */
300 static void ceph_sock_write_space(struct sock
*sk
)
302 struct ceph_connection
*con
= sk
->sk_user_data
;
304 /* only queue to workqueue if there is data we want to write,
305 * and there is sufficient space in the socket buffer to accept
306 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
307 * doesn't get called again until try_write() fills the socket
308 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
309 * and net/core/stream.c:sk_stream_write_space().
311 if (test_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
)) {
312 if (sk_stream_wspace(sk
) >= sk_stream_min_wspace(sk
)) {
313 dout("%s %p queueing write work\n", __func__
, con
);
314 clear_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
318 dout("%s %p nothing to write\n", __func__
, con
);
322 /* socket's state has changed */
323 static void ceph_sock_state_change(struct sock
*sk
)
325 struct ceph_connection
*con
= sk
->sk_user_data
;
327 dout("%s %p state = %lu sk_state = %u\n", __func__
,
328 con
, con
->state
, sk
->sk_state
);
330 switch (sk
->sk_state
) {
332 dout("%s TCP_CLOSE\n", __func__
);
334 dout("%s TCP_CLOSE_WAIT\n", __func__
);
335 con_sock_state_closing(con
);
336 set_bit(CON_FLAG_SOCK_CLOSED
, &con
->flags
);
339 case TCP_ESTABLISHED
:
340 dout("%s TCP_ESTABLISHED\n", __func__
);
341 con_sock_state_connected(con
);
344 default: /* Everything else is uninteresting */
350 * set up socket callbacks
352 static void set_sock_callbacks(struct socket
*sock
,
353 struct ceph_connection
*con
)
355 struct sock
*sk
= sock
->sk
;
356 sk
->sk_user_data
= con
;
357 sk
->sk_data_ready
= ceph_sock_data_ready
;
358 sk
->sk_write_space
= ceph_sock_write_space
;
359 sk
->sk_state_change
= ceph_sock_state_change
;
368 * initiate connection to a remote socket.
370 static int ceph_tcp_connect(struct ceph_connection
*con
)
372 struct sockaddr_storage
*paddr
= &con
->peer_addr
.in_addr
;
377 ret
= sock_create_kern(con
->peer_addr
.in_addr
.ss_family
, SOCK_STREAM
,
381 sock
->sk
->sk_allocation
= GFP_NOFS
;
383 #ifdef CONFIG_LOCKDEP
384 lockdep_set_class(&sock
->sk
->sk_lock
, &socket_class
);
387 set_sock_callbacks(sock
, con
);
389 dout("connect %s\n", ceph_pr_addr(&con
->peer_addr
.in_addr
));
391 con_sock_state_connecting(con
);
392 ret
= sock
->ops
->connect(sock
, (struct sockaddr
*)paddr
, sizeof(*paddr
),
394 if (ret
== -EINPROGRESS
) {
395 dout("connect %s EINPROGRESS sk_state = %u\n",
396 ceph_pr_addr(&con
->peer_addr
.in_addr
),
398 } else if (ret
< 0) {
399 pr_err("connect %s error %d\n",
400 ceph_pr_addr(&con
->peer_addr
.in_addr
), ret
);
402 con
->error_msg
= "connect error";
410 static int ceph_tcp_recvmsg(struct socket
*sock
, void *buf
, size_t len
)
412 struct kvec iov
= {buf
, len
};
413 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
416 r
= kernel_recvmsg(sock
, &msg
, &iov
, 1, len
, msg
.msg_flags
);
423 * write something. @more is true if caller will be sending more data
426 static int ceph_tcp_sendmsg(struct socket
*sock
, struct kvec
*iov
,
427 size_t kvlen
, size_t len
, int more
)
429 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
433 msg
.msg_flags
|= MSG_MORE
;
435 msg
.msg_flags
|= MSG_EOR
; /* superfluous, but what the hell */
437 r
= kernel_sendmsg(sock
, &msg
, iov
, kvlen
, len
);
443 static int ceph_tcp_sendpage(struct socket
*sock
, struct page
*page
,
444 int offset
, size_t size
, int more
)
446 int flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
| (more
? MSG_MORE
: MSG_EOR
);
449 ret
= kernel_sendpage(sock
, page
, offset
, size
, flags
);
458 * Shutdown/close the socket for the given connection.
460 static int con_close_socket(struct ceph_connection
*con
)
464 dout("con_close_socket on %p sock %p\n", con
, con
->sock
);
466 rc
= con
->sock
->ops
->shutdown(con
->sock
, SHUT_RDWR
);
467 sock_release(con
->sock
);
472 * Forcibly clear the SOCK_CLOSED flag. It gets set
473 * independent of the connection mutex, and we could have
474 * received a socket close event before we had the chance to
475 * shut the socket down.
477 clear_bit(CON_FLAG_SOCK_CLOSED
, &con
->flags
);
479 con_sock_state_closed(con
);
484 * Reset a connection. Discard all incoming and outgoing messages
485 * and clear *_seq state.
487 static void ceph_msg_remove(struct ceph_msg
*msg
)
489 list_del_init(&msg
->list_head
);
490 BUG_ON(msg
->con
== NULL
);
491 msg
->con
->ops
->put(msg
->con
);
496 static void ceph_msg_remove_list(struct list_head
*head
)
498 while (!list_empty(head
)) {
499 struct ceph_msg
*msg
= list_first_entry(head
, struct ceph_msg
,
501 ceph_msg_remove(msg
);
505 static void reset_connection(struct ceph_connection
*con
)
507 /* reset connection, out_queue, msg_ and connect_seq */
508 /* discard existing out_queue and msg_seq */
509 ceph_msg_remove_list(&con
->out_queue
);
510 ceph_msg_remove_list(&con
->out_sent
);
513 BUG_ON(con
->in_msg
->con
!= con
);
514 con
->in_msg
->con
= NULL
;
515 ceph_msg_put(con
->in_msg
);
520 con
->connect_seq
= 0;
523 ceph_msg_put(con
->out_msg
);
527 con
->in_seq_acked
= 0;
531 * mark a peer down. drop any open connections.
533 void ceph_con_close(struct ceph_connection
*con
)
535 mutex_lock(&con
->mutex
);
536 dout("con_close %p peer %s\n", con
,
537 ceph_pr_addr(&con
->peer_addr
.in_addr
));
538 con
->state
= CON_STATE_CLOSED
;
540 clear_bit(CON_FLAG_LOSSYTX
, &con
->flags
); /* so we retry next connect */
541 clear_bit(CON_FLAG_KEEPALIVE_PENDING
, &con
->flags
);
542 clear_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
);
543 clear_bit(CON_FLAG_KEEPALIVE_PENDING
, &con
->flags
);
544 clear_bit(CON_FLAG_BACKOFF
, &con
->flags
);
546 reset_connection(con
);
547 con
->peer_global_seq
= 0;
548 cancel_delayed_work(&con
->work
);
549 con_close_socket(con
);
550 mutex_unlock(&con
->mutex
);
552 EXPORT_SYMBOL(ceph_con_close
);
555 * Reopen a closed connection, with a new peer address.
557 void ceph_con_open(struct ceph_connection
*con
,
558 __u8 entity_type
, __u64 entity_num
,
559 struct ceph_entity_addr
*addr
)
561 mutex_lock(&con
->mutex
);
562 dout("con_open %p %s\n", con
, ceph_pr_addr(&addr
->in_addr
));
564 BUG_ON(con
->state
!= CON_STATE_CLOSED
);
565 con
->state
= CON_STATE_PREOPEN
;
567 con
->peer_name
.type
= (__u8
) entity_type
;
568 con
->peer_name
.num
= cpu_to_le64(entity_num
);
570 memcpy(&con
->peer_addr
, addr
, sizeof(*addr
));
571 con
->delay
= 0; /* reset backoff memory */
572 mutex_unlock(&con
->mutex
);
575 EXPORT_SYMBOL(ceph_con_open
);
578 * return true if this connection ever successfully opened
580 bool ceph_con_opened(struct ceph_connection
*con
)
582 return con
->connect_seq
> 0;
586 * initialize a new connection.
588 void ceph_con_init(struct ceph_connection
*con
, void *private,
589 const struct ceph_connection_operations
*ops
,
590 struct ceph_messenger
*msgr
)
592 dout("con_init %p\n", con
);
593 memset(con
, 0, sizeof(*con
));
594 con
->private = private;
598 con_sock_state_init(con
);
600 mutex_init(&con
->mutex
);
601 INIT_LIST_HEAD(&con
->out_queue
);
602 INIT_LIST_HEAD(&con
->out_sent
);
603 INIT_DELAYED_WORK(&con
->work
, con_work
);
605 con
->state
= CON_STATE_CLOSED
;
607 EXPORT_SYMBOL(ceph_con_init
);
611 * We maintain a global counter to order connection attempts. Get
612 * a unique seq greater than @gt.
614 static u32
get_global_seq(struct ceph_messenger
*msgr
, u32 gt
)
618 spin_lock(&msgr
->global_seq_lock
);
619 if (msgr
->global_seq
< gt
)
620 msgr
->global_seq
= gt
;
621 ret
= ++msgr
->global_seq
;
622 spin_unlock(&msgr
->global_seq_lock
);
626 static void con_out_kvec_reset(struct ceph_connection
*con
)
628 con
->out_kvec_left
= 0;
629 con
->out_kvec_bytes
= 0;
630 con
->out_kvec_cur
= &con
->out_kvec
[0];
633 static void con_out_kvec_add(struct ceph_connection
*con
,
634 size_t size
, void *data
)
638 index
= con
->out_kvec_left
;
639 BUG_ON(index
>= ARRAY_SIZE(con
->out_kvec
));
641 con
->out_kvec
[index
].iov_len
= size
;
642 con
->out_kvec
[index
].iov_base
= data
;
643 con
->out_kvec_left
++;
644 con
->out_kvec_bytes
+= size
;
648 static void init_bio_iter(struct bio
*bio
, struct bio
**iter
, int *seg
)
659 static void iter_bio_next(struct bio
**bio_iter
, int *seg
)
661 if (*bio_iter
== NULL
)
664 BUG_ON(*seg
>= (*bio_iter
)->bi_vcnt
);
667 if (*seg
== (*bio_iter
)->bi_vcnt
)
668 init_bio_iter((*bio_iter
)->bi_next
, bio_iter
, seg
);
672 static void prepare_write_message_data(struct ceph_connection
*con
)
674 struct ceph_msg
*msg
= con
->out_msg
;
677 BUG_ON(!msg
->hdr
.data_len
);
679 /* initialize page iterator */
680 con
->out_msg_pos
.page
= 0;
682 con
->out_msg_pos
.page_pos
= msg
->page_alignment
;
684 con
->out_msg_pos
.page_pos
= 0;
687 init_bio_iter(msg
->bio
, &msg
->bio_iter
, &msg
->bio_seg
);
689 con
->out_msg_pos
.data_pos
= 0;
690 con
->out_msg_pos
.did_page_crc
= false;
691 con
->out_more
= 1; /* data + footer will follow */
695 * Prepare footer for currently outgoing message, and finish things
696 * off. Assumes out_kvec* are already valid.. we just add on to the end.
698 static void prepare_write_message_footer(struct ceph_connection
*con
)
700 struct ceph_msg
*m
= con
->out_msg
;
701 int v
= con
->out_kvec_left
;
703 m
->footer
.flags
|= CEPH_MSG_FOOTER_COMPLETE
;
705 dout("prepare_write_message_footer %p\n", con
);
706 con
->out_kvec_is_msg
= true;
707 con
->out_kvec
[v
].iov_base
= &m
->footer
;
708 con
->out_kvec
[v
].iov_len
= sizeof(m
->footer
);
709 con
->out_kvec_bytes
+= sizeof(m
->footer
);
710 con
->out_kvec_left
++;
711 con
->out_more
= m
->more_to_follow
;
712 con
->out_msg_done
= true;
716 * Prepare headers for the next outgoing message.
718 static void prepare_write_message(struct ceph_connection
*con
)
723 con_out_kvec_reset(con
);
724 con
->out_kvec_is_msg
= true;
725 con
->out_msg_done
= false;
727 /* Sneak an ack in there first? If we can get it into the same
728 * TCP packet that's a good thing. */
729 if (con
->in_seq
> con
->in_seq_acked
) {
730 con
->in_seq_acked
= con
->in_seq
;
731 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
732 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
733 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
737 BUG_ON(list_empty(&con
->out_queue
));
738 m
= list_first_entry(&con
->out_queue
, struct ceph_msg
, list_head
);
740 BUG_ON(m
->con
!= con
);
742 /* put message on sent list */
744 list_move_tail(&m
->list_head
, &con
->out_sent
);
747 * only assign outgoing seq # if we haven't sent this message
748 * yet. if it is requeued, resend with it's original seq.
750 if (m
->needs_out_seq
) {
751 m
->hdr
.seq
= cpu_to_le64(++con
->out_seq
);
752 m
->needs_out_seq
= false;
759 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n",
760 m
, con
->out_seq
, le16_to_cpu(m
->hdr
.type
),
761 le32_to_cpu(m
->hdr
.front_len
), le32_to_cpu(m
->hdr
.middle_len
),
762 le32_to_cpu(m
->hdr
.data_len
),
764 BUG_ON(le32_to_cpu(m
->hdr
.front_len
) != m
->front
.iov_len
);
766 /* tag + hdr + front + middle */
767 con_out_kvec_add(con
, sizeof (tag_msg
), &tag_msg
);
768 con_out_kvec_add(con
, sizeof (m
->hdr
), &m
->hdr
);
769 con_out_kvec_add(con
, m
->front
.iov_len
, m
->front
.iov_base
);
772 con_out_kvec_add(con
, m
->middle
->vec
.iov_len
,
773 m
->middle
->vec
.iov_base
);
775 /* fill in crc (except data pages), footer */
776 crc
= crc32c(0, &m
->hdr
, offsetof(struct ceph_msg_header
, crc
));
777 con
->out_msg
->hdr
.crc
= cpu_to_le32(crc
);
778 con
->out_msg
->footer
.flags
= 0;
780 crc
= crc32c(0, m
->front
.iov_base
, m
->front
.iov_len
);
781 con
->out_msg
->footer
.front_crc
= cpu_to_le32(crc
);
783 crc
= crc32c(0, m
->middle
->vec
.iov_base
,
784 m
->middle
->vec
.iov_len
);
785 con
->out_msg
->footer
.middle_crc
= cpu_to_le32(crc
);
787 con
->out_msg
->footer
.middle_crc
= 0;
788 dout("%s front_crc %u middle_crc %u\n", __func__
,
789 le32_to_cpu(con
->out_msg
->footer
.front_crc
),
790 le32_to_cpu(con
->out_msg
->footer
.middle_crc
));
792 /* is there a data payload? */
793 con
->out_msg
->footer
.data_crc
= 0;
795 prepare_write_message_data(con
);
797 /* no, queue up footer too and be done */
798 prepare_write_message_footer(con
);
800 set_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
);
806 static void prepare_write_ack(struct ceph_connection
*con
)
808 dout("prepare_write_ack %p %llu -> %llu\n", con
,
809 con
->in_seq_acked
, con
->in_seq
);
810 con
->in_seq_acked
= con
->in_seq
;
812 con_out_kvec_reset(con
);
814 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
816 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
817 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
820 con
->out_more
= 1; /* more will follow.. eventually.. */
821 set_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
);
825 * Prepare to write keepalive byte.
827 static void prepare_write_keepalive(struct ceph_connection
*con
)
829 dout("prepare_write_keepalive %p\n", con
);
830 con_out_kvec_reset(con
);
831 con_out_kvec_add(con
, sizeof (tag_keepalive
), &tag_keepalive
);
832 set_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
);
836 * Connection negotiation.
839 static struct ceph_auth_handshake
*get_connect_authorizer(struct ceph_connection
*con
,
842 struct ceph_auth_handshake
*auth
;
844 if (!con
->ops
->get_authorizer
) {
845 con
->out_connect
.authorizer_protocol
= CEPH_AUTH_UNKNOWN
;
846 con
->out_connect
.authorizer_len
= 0;
850 /* Can't hold the mutex while getting authorizer */
851 mutex_unlock(&con
->mutex
);
852 auth
= con
->ops
->get_authorizer(con
, auth_proto
, con
->auth_retry
);
853 mutex_lock(&con
->mutex
);
857 if (con
->state
!= CON_STATE_NEGOTIATING
)
858 return ERR_PTR(-EAGAIN
);
860 con
->auth_reply_buf
= auth
->authorizer_reply_buf
;
861 con
->auth_reply_buf_len
= auth
->authorizer_reply_buf_len
;
866 * We connected to a peer and are saying hello.
868 static void prepare_write_banner(struct ceph_connection
*con
)
870 con_out_kvec_add(con
, strlen(CEPH_BANNER
), CEPH_BANNER
);
871 con_out_kvec_add(con
, sizeof (con
->msgr
->my_enc_addr
),
872 &con
->msgr
->my_enc_addr
);
875 set_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
);
878 static int prepare_write_connect(struct ceph_connection
*con
)
880 unsigned int global_seq
= get_global_seq(con
->msgr
, 0);
883 struct ceph_auth_handshake
*auth
;
885 switch (con
->peer_name
.type
) {
886 case CEPH_ENTITY_TYPE_MON
:
887 proto
= CEPH_MONC_PROTOCOL
;
889 case CEPH_ENTITY_TYPE_OSD
:
890 proto
= CEPH_OSDC_PROTOCOL
;
892 case CEPH_ENTITY_TYPE_MDS
:
893 proto
= CEPH_MDSC_PROTOCOL
;
899 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con
,
900 con
->connect_seq
, global_seq
, proto
);
902 con
->out_connect
.features
= cpu_to_le64(con
->msgr
->supported_features
);
903 con
->out_connect
.host_type
= cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT
);
904 con
->out_connect
.connect_seq
= cpu_to_le32(con
->connect_seq
);
905 con
->out_connect
.global_seq
= cpu_to_le32(global_seq
);
906 con
->out_connect
.protocol_version
= cpu_to_le32(proto
);
907 con
->out_connect
.flags
= 0;
909 auth_proto
= CEPH_AUTH_UNKNOWN
;
910 auth
= get_connect_authorizer(con
, &auth_proto
);
912 return PTR_ERR(auth
);
914 con
->out_connect
.authorizer_protocol
= cpu_to_le32(auth_proto
);
915 con
->out_connect
.authorizer_len
= auth
?
916 cpu_to_le32(auth
->authorizer_buf_len
) : 0;
918 con_out_kvec_add(con
, sizeof (con
->out_connect
),
920 if (auth
&& auth
->authorizer_buf_len
)
921 con_out_kvec_add(con
, auth
->authorizer_buf_len
,
922 auth
->authorizer_buf
);
925 set_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
);
931 * write as much of pending kvecs to the socket as we can.
933 * 0 -> socket full, but more to do
936 static int write_partial_kvec(struct ceph_connection
*con
)
940 dout("write_partial_kvec %p %d left\n", con
, con
->out_kvec_bytes
);
941 while (con
->out_kvec_bytes
> 0) {
942 ret
= ceph_tcp_sendmsg(con
->sock
, con
->out_kvec_cur
,
943 con
->out_kvec_left
, con
->out_kvec_bytes
,
947 con
->out_kvec_bytes
-= ret
;
948 if (con
->out_kvec_bytes
== 0)
951 /* account for full iov entries consumed */
952 while (ret
>= con
->out_kvec_cur
->iov_len
) {
953 BUG_ON(!con
->out_kvec_left
);
954 ret
-= con
->out_kvec_cur
->iov_len
;
956 con
->out_kvec_left
--;
958 /* and for a partially-consumed entry */
960 con
->out_kvec_cur
->iov_len
-= ret
;
961 con
->out_kvec_cur
->iov_base
+= ret
;
964 con
->out_kvec_left
= 0;
965 con
->out_kvec_is_msg
= false;
968 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con
,
969 con
->out_kvec_bytes
, con
->out_kvec_left
, ret
);
970 return ret
; /* done! */
973 static void out_msg_pos_next(struct ceph_connection
*con
, struct page
*page
,
974 size_t len
, size_t sent
, bool in_trail
)
976 struct ceph_msg
*msg
= con
->out_msg
;
981 con
->out_msg_pos
.data_pos
+= sent
;
982 con
->out_msg_pos
.page_pos
+= sent
;
987 con
->out_msg_pos
.page_pos
= 0;
988 con
->out_msg_pos
.page
++;
989 con
->out_msg_pos
.did_page_crc
= false;
991 list_move_tail(&page
->lru
,
993 else if (msg
->pagelist
)
994 list_move_tail(&page
->lru
,
995 &msg
->pagelist
->head
);
998 iter_bio_next(&msg
->bio_iter
, &msg
->bio_seg
);
1003 * Write as much message data payload as we can. If we finish, queue
1005 * 1 -> done, footer is now queued in out_kvec[].
1006 * 0 -> socket full, but more to do
1009 static int write_partial_msg_pages(struct ceph_connection
*con
)
1011 struct ceph_msg
*msg
= con
->out_msg
;
1012 unsigned int data_len
= le32_to_cpu(msg
->hdr
.data_len
);
1014 bool do_datacrc
= !con
->msgr
->nocrc
;
1016 int total_max_write
;
1017 bool in_trail
= false;
1018 const size_t trail_len
= (msg
->trail
? msg
->trail
->length
: 0);
1019 const size_t trail_off
= data_len
- trail_len
;
1021 dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
1022 con
, msg
, con
->out_msg_pos
.page
, msg
->nr_pages
,
1023 con
->out_msg_pos
.page_pos
);
1026 * Iterate through each page that contains data to be
1027 * written, and send as much as possible for each.
1029 * If we are calculating the data crc (the default), we will
1030 * need to map the page. If we have no pages, they have
1031 * been revoked, so use the zero page.
1033 while (data_len
> con
->out_msg_pos
.data_pos
) {
1034 struct page
*page
= NULL
;
1035 int max_write
= PAGE_SIZE
;
1038 in_trail
= in_trail
|| con
->out_msg_pos
.data_pos
>= trail_off
;
1040 total_max_write
= trail_off
- con
->out_msg_pos
.data_pos
;
1043 total_max_write
= data_len
- con
->out_msg_pos
.data_pos
;
1045 page
= list_first_entry(&msg
->trail
->head
,
1047 } else if (msg
->pages
) {
1048 page
= msg
->pages
[con
->out_msg_pos
.page
];
1049 } else if (msg
->pagelist
) {
1050 page
= list_first_entry(&msg
->pagelist
->head
,
1053 } else if (msg
->bio
) {
1056 bv
= bio_iovec_idx(msg
->bio_iter
, msg
->bio_seg
);
1058 bio_offset
= bv
->bv_offset
;
1059 max_write
= bv
->bv_len
;
1064 len
= min_t(int, max_write
- con
->out_msg_pos
.page_pos
,
1067 if (do_datacrc
&& !con
->out_msg_pos
.did_page_crc
) {
1069 u32 crc
= le32_to_cpu(msg
->footer
.data_crc
);
1073 BUG_ON(kaddr
== NULL
);
1074 base
= kaddr
+ con
->out_msg_pos
.page_pos
+ bio_offset
;
1075 crc
= crc32c(crc
, base
, len
);
1076 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1077 con
->out_msg_pos
.did_page_crc
= true;
1079 ret
= ceph_tcp_sendpage(con
->sock
, page
,
1080 con
->out_msg_pos
.page_pos
+ bio_offset
,
1089 out_msg_pos_next(con
, page
, len
, (size_t) ret
, in_trail
);
1092 dout("write_partial_msg_pages %p msg %p done\n", con
, msg
);
1094 /* prepare and queue up footer, too */
1096 msg
->footer
.flags
|= CEPH_MSG_FOOTER_NOCRC
;
1097 con_out_kvec_reset(con
);
1098 prepare_write_message_footer(con
);
1107 static int write_partial_skip(struct ceph_connection
*con
)
1111 while (con
->out_skip
> 0) {
1112 size_t size
= min(con
->out_skip
, (int) PAGE_CACHE_SIZE
);
1114 ret
= ceph_tcp_sendpage(con
->sock
, zero_page
, 0, size
, 1);
1117 con
->out_skip
-= ret
;
1125 * Prepare to read connection handshake, or an ack.
1127 static void prepare_read_banner(struct ceph_connection
*con
)
1129 dout("prepare_read_banner %p\n", con
);
1130 con
->in_base_pos
= 0;
1133 static void prepare_read_connect(struct ceph_connection
*con
)
1135 dout("prepare_read_connect %p\n", con
);
1136 con
->in_base_pos
= 0;
1139 static void prepare_read_ack(struct ceph_connection
*con
)
1141 dout("prepare_read_ack %p\n", con
);
1142 con
->in_base_pos
= 0;
1145 static void prepare_read_tag(struct ceph_connection
*con
)
1147 dout("prepare_read_tag %p\n", con
);
1148 con
->in_base_pos
= 0;
1149 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1153 * Prepare to read a message.
1155 static int prepare_read_message(struct ceph_connection
*con
)
1157 dout("prepare_read_message %p\n", con
);
1158 BUG_ON(con
->in_msg
!= NULL
);
1159 con
->in_base_pos
= 0;
1160 con
->in_front_crc
= con
->in_middle_crc
= con
->in_data_crc
= 0;
1165 static int read_partial(struct ceph_connection
*con
,
1166 int end
, int size
, void *object
)
1168 while (con
->in_base_pos
< end
) {
1169 int left
= end
- con
->in_base_pos
;
1170 int have
= size
- left
;
1171 int ret
= ceph_tcp_recvmsg(con
->sock
, object
+ have
, left
);
1174 con
->in_base_pos
+= ret
;
1181 * Read all or part of the connect-side handshake on a new connection
1183 static int read_partial_banner(struct ceph_connection
*con
)
1189 dout("read_partial_banner %p at %d\n", con
, con
->in_base_pos
);
1192 size
= strlen(CEPH_BANNER
);
1194 ret
= read_partial(con
, end
, size
, con
->in_banner
);
1198 size
= sizeof (con
->actual_peer_addr
);
1200 ret
= read_partial(con
, end
, size
, &con
->actual_peer_addr
);
1204 size
= sizeof (con
->peer_addr_for_me
);
1206 ret
= read_partial(con
, end
, size
, &con
->peer_addr_for_me
);
1214 static int read_partial_connect(struct ceph_connection
*con
)
1220 dout("read_partial_connect %p at %d\n", con
, con
->in_base_pos
);
1222 size
= sizeof (con
->in_reply
);
1224 ret
= read_partial(con
, end
, size
, &con
->in_reply
);
1228 size
= le32_to_cpu(con
->in_reply
.authorizer_len
);
1230 ret
= read_partial(con
, end
, size
, con
->auth_reply_buf
);
1234 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1235 con
, (int)con
->in_reply
.tag
,
1236 le32_to_cpu(con
->in_reply
.connect_seq
),
1237 le32_to_cpu(con
->in_reply
.global_seq
));
1244 * Verify the hello banner looks okay.
1246 static int verify_hello(struct ceph_connection
*con
)
1248 if (memcmp(con
->in_banner
, CEPH_BANNER
, strlen(CEPH_BANNER
))) {
1249 pr_err("connect to %s got bad banner\n",
1250 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1251 con
->error_msg
= "protocol error, bad banner";
1257 static bool addr_is_blank(struct sockaddr_storage
*ss
)
1259 switch (ss
->ss_family
) {
1261 return ((struct sockaddr_in
*)ss
)->sin_addr
.s_addr
== 0;
1264 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[0] == 0 &&
1265 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[1] == 0 &&
1266 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[2] == 0 &&
1267 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[3] == 0;
1272 static int addr_port(struct sockaddr_storage
*ss
)
1274 switch (ss
->ss_family
) {
1276 return ntohs(((struct sockaddr_in
*)ss
)->sin_port
);
1278 return ntohs(((struct sockaddr_in6
*)ss
)->sin6_port
);
1283 static void addr_set_port(struct sockaddr_storage
*ss
, int p
)
1285 switch (ss
->ss_family
) {
1287 ((struct sockaddr_in
*)ss
)->sin_port
= htons(p
);
1290 ((struct sockaddr_in6
*)ss
)->sin6_port
= htons(p
);
1296 * Unlike other *_pton function semantics, zero indicates success.
1298 static int ceph_pton(const char *str
, size_t len
, struct sockaddr_storage
*ss
,
1299 char delim
, const char **ipend
)
1301 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
1302 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
1304 memset(ss
, 0, sizeof(*ss
));
1306 if (in4_pton(str
, len
, (u8
*)&in4
->sin_addr
.s_addr
, delim
, ipend
)) {
1307 ss
->ss_family
= AF_INET
;
1311 if (in6_pton(str
, len
, (u8
*)&in6
->sin6_addr
.s6_addr
, delim
, ipend
)) {
1312 ss
->ss_family
= AF_INET6
;
1320 * Extract hostname string and resolve using kernel DNS facility.
1322 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1323 static int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1324 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1326 const char *end
, *delim_p
;
1327 char *colon_p
, *ip_addr
= NULL
;
1331 * The end of the hostname occurs immediately preceding the delimiter or
1332 * the port marker (':') where the delimiter takes precedence.
1334 delim_p
= memchr(name
, delim
, namelen
);
1335 colon_p
= memchr(name
, ':', namelen
);
1337 if (delim_p
&& colon_p
)
1338 end
= delim_p
< colon_p
? delim_p
: colon_p
;
1339 else if (!delim_p
&& colon_p
)
1343 if (!end
) /* case: hostname:/ */
1344 end
= name
+ namelen
;
1350 /* do dns_resolve upcall */
1351 ip_len
= dns_query(NULL
, name
, end
- name
, NULL
, &ip_addr
, NULL
);
1353 ret
= ceph_pton(ip_addr
, ip_len
, ss
, -1, NULL
);
1361 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end
- name
), name
,
1362 ret
, ret
? "failed" : ceph_pr_addr(ss
));
1367 static inline int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1368 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1375 * Parse a server name (IP or hostname). If a valid IP address is not found
1376 * then try to extract a hostname to resolve using userspace DNS upcall.
1378 static int ceph_parse_server_name(const char *name
, size_t namelen
,
1379 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1383 ret
= ceph_pton(name
, namelen
, ss
, delim
, ipend
);
1385 ret
= ceph_dns_resolve_name(name
, namelen
, ss
, delim
, ipend
);
1391 * Parse an ip[:port] list into an addr array. Use the default
1392 * monitor port if a port isn't specified.
1394 int ceph_parse_ips(const char *c
, const char *end
,
1395 struct ceph_entity_addr
*addr
,
1396 int max_count
, int *count
)
1398 int i
, ret
= -EINVAL
;
1401 dout("parse_ips on '%.*s'\n", (int)(end
-c
), c
);
1402 for (i
= 0; i
< max_count
; i
++) {
1404 struct sockaddr_storage
*ss
= &addr
[i
].in_addr
;
1413 ret
= ceph_parse_server_name(p
, end
- p
, ss
, delim
, &ipend
);
1422 dout("missing matching ']'\n");
1429 if (p
< end
&& *p
== ':') {
1432 while (p
< end
&& *p
>= '0' && *p
<= '9') {
1433 port
= (port
* 10) + (*p
- '0');
1436 if (port
> 65535 || port
== 0)
1439 port
= CEPH_MON_PORT
;
1442 addr_set_port(ss
, port
);
1444 dout("parse_ips got %s\n", ceph_pr_addr(ss
));
1461 pr_err("parse_ips bad ip '%.*s'\n", (int)(end
- c
), c
);
1464 EXPORT_SYMBOL(ceph_parse_ips
);
1466 static int process_banner(struct ceph_connection
*con
)
1468 dout("process_banner on %p\n", con
);
1470 if (verify_hello(con
) < 0)
1473 ceph_decode_addr(&con
->actual_peer_addr
);
1474 ceph_decode_addr(&con
->peer_addr_for_me
);
1477 * Make sure the other end is who we wanted. note that the other
1478 * end may not yet know their ip address, so if it's 0.0.0.0, give
1479 * them the benefit of the doubt.
1481 if (memcmp(&con
->peer_addr
, &con
->actual_peer_addr
,
1482 sizeof(con
->peer_addr
)) != 0 &&
1483 !(addr_is_blank(&con
->actual_peer_addr
.in_addr
) &&
1484 con
->actual_peer_addr
.nonce
== con
->peer_addr
.nonce
)) {
1485 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1486 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1487 (int)le32_to_cpu(con
->peer_addr
.nonce
),
1488 ceph_pr_addr(&con
->actual_peer_addr
.in_addr
),
1489 (int)le32_to_cpu(con
->actual_peer_addr
.nonce
));
1490 con
->error_msg
= "wrong peer at address";
1495 * did we learn our address?
1497 if (addr_is_blank(&con
->msgr
->inst
.addr
.in_addr
)) {
1498 int port
= addr_port(&con
->msgr
->inst
.addr
.in_addr
);
1500 memcpy(&con
->msgr
->inst
.addr
.in_addr
,
1501 &con
->peer_addr_for_me
.in_addr
,
1502 sizeof(con
->peer_addr_for_me
.in_addr
));
1503 addr_set_port(&con
->msgr
->inst
.addr
.in_addr
, port
);
1504 encode_my_addr(con
->msgr
);
1505 dout("process_banner learned my addr is %s\n",
1506 ceph_pr_addr(&con
->msgr
->inst
.addr
.in_addr
));
1512 static void fail_protocol(struct ceph_connection
*con
)
1514 reset_connection(con
);
1515 BUG_ON(con
->state
!= CON_STATE_NEGOTIATING
);
1516 con
->state
= CON_STATE_CLOSED
;
1519 static int process_connect(struct ceph_connection
*con
)
1521 u64 sup_feat
= con
->msgr
->supported_features
;
1522 u64 req_feat
= con
->msgr
->required_features
;
1523 u64 server_feat
= le64_to_cpu(con
->in_reply
.features
);
1526 dout("process_connect on %p tag %d\n", con
, (int)con
->in_tag
);
1528 switch (con
->in_reply
.tag
) {
1529 case CEPH_MSGR_TAG_FEATURES
:
1530 pr_err("%s%lld %s feature set mismatch,"
1531 " my %llx < server's %llx, missing %llx\n",
1532 ENTITY_NAME(con
->peer_name
),
1533 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1534 sup_feat
, server_feat
, server_feat
& ~sup_feat
);
1535 con
->error_msg
= "missing required protocol features";
1539 case CEPH_MSGR_TAG_BADPROTOVER
:
1540 pr_err("%s%lld %s protocol version mismatch,"
1541 " my %d != server's %d\n",
1542 ENTITY_NAME(con
->peer_name
),
1543 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1544 le32_to_cpu(con
->out_connect
.protocol_version
),
1545 le32_to_cpu(con
->in_reply
.protocol_version
));
1546 con
->error_msg
= "protocol version mismatch";
1550 case CEPH_MSGR_TAG_BADAUTHORIZER
:
1552 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con
,
1554 if (con
->auth_retry
== 2) {
1555 con
->error_msg
= "connect authorization failure";
1558 con
->auth_retry
= 1;
1559 con_out_kvec_reset(con
);
1560 ret
= prepare_write_connect(con
);
1563 prepare_read_connect(con
);
1566 case CEPH_MSGR_TAG_RESETSESSION
:
1568 * If we connected with a large connect_seq but the peer
1569 * has no record of a session with us (no connection, or
1570 * connect_seq == 0), they will send RESETSESION to indicate
1571 * that they must have reset their session, and may have
1574 dout("process_connect got RESET peer seq %u\n",
1575 le32_to_cpu(con
->in_reply
.connect_seq
));
1576 pr_err("%s%lld %s connection reset\n",
1577 ENTITY_NAME(con
->peer_name
),
1578 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1579 reset_connection(con
);
1580 con_out_kvec_reset(con
);
1581 ret
= prepare_write_connect(con
);
1584 prepare_read_connect(con
);
1586 /* Tell ceph about it. */
1587 mutex_unlock(&con
->mutex
);
1588 pr_info("reset on %s%lld\n", ENTITY_NAME(con
->peer_name
));
1589 if (con
->ops
->peer_reset
)
1590 con
->ops
->peer_reset(con
);
1591 mutex_lock(&con
->mutex
);
1592 if (con
->state
!= CON_STATE_NEGOTIATING
)
1596 case CEPH_MSGR_TAG_RETRY_SESSION
:
1598 * If we sent a smaller connect_seq than the peer has, try
1599 * again with a larger value.
1601 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
1602 le32_to_cpu(con
->out_connect
.connect_seq
),
1603 le32_to_cpu(con
->in_reply
.connect_seq
));
1604 con
->connect_seq
= le32_to_cpu(con
->in_reply
.connect_seq
);
1605 con_out_kvec_reset(con
);
1606 ret
= prepare_write_connect(con
);
1609 prepare_read_connect(con
);
1612 case CEPH_MSGR_TAG_RETRY_GLOBAL
:
1614 * If we sent a smaller global_seq than the peer has, try
1615 * again with a larger value.
1617 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1618 con
->peer_global_seq
,
1619 le32_to_cpu(con
->in_reply
.global_seq
));
1620 get_global_seq(con
->msgr
,
1621 le32_to_cpu(con
->in_reply
.global_seq
));
1622 con_out_kvec_reset(con
);
1623 ret
= prepare_write_connect(con
);
1626 prepare_read_connect(con
);
1629 case CEPH_MSGR_TAG_READY
:
1630 if (req_feat
& ~server_feat
) {
1631 pr_err("%s%lld %s protocol feature mismatch,"
1632 " my required %llx > server's %llx, need %llx\n",
1633 ENTITY_NAME(con
->peer_name
),
1634 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1635 req_feat
, server_feat
, req_feat
& ~server_feat
);
1636 con
->error_msg
= "missing required protocol features";
1641 BUG_ON(con
->state
!= CON_STATE_NEGOTIATING
);
1642 con
->state
= CON_STATE_OPEN
;
1644 con
->peer_global_seq
= le32_to_cpu(con
->in_reply
.global_seq
);
1646 con
->peer_features
= server_feat
;
1647 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1648 con
->peer_global_seq
,
1649 le32_to_cpu(con
->in_reply
.connect_seq
),
1651 WARN_ON(con
->connect_seq
!=
1652 le32_to_cpu(con
->in_reply
.connect_seq
));
1654 if (con
->in_reply
.flags
& CEPH_MSG_CONNECT_LOSSY
)
1655 set_bit(CON_FLAG_LOSSYTX
, &con
->flags
);
1657 con
->delay
= 0; /* reset backoff memory */
1659 prepare_read_tag(con
);
1662 case CEPH_MSGR_TAG_WAIT
:
1664 * If there is a connection race (we are opening
1665 * connections to each other), one of us may just have
1666 * to WAIT. This shouldn't happen if we are the
1669 pr_err("process_connect got WAIT as client\n");
1670 con
->error_msg
= "protocol error, got WAIT as client";
1674 pr_err("connect protocol error, will retry\n");
1675 con
->error_msg
= "protocol error, garbage tag during connect";
1683 * read (part of) an ack
1685 static int read_partial_ack(struct ceph_connection
*con
)
1687 int size
= sizeof (con
->in_temp_ack
);
1690 return read_partial(con
, end
, size
, &con
->in_temp_ack
);
1695 * We can finally discard anything that's been acked.
1697 static void process_ack(struct ceph_connection
*con
)
1700 u64 ack
= le64_to_cpu(con
->in_temp_ack
);
1703 while (!list_empty(&con
->out_sent
)) {
1704 m
= list_first_entry(&con
->out_sent
, struct ceph_msg
,
1706 seq
= le64_to_cpu(m
->hdr
.seq
);
1709 dout("got ack for seq %llu type %d at %p\n", seq
,
1710 le16_to_cpu(m
->hdr
.type
), m
);
1711 m
->ack_stamp
= jiffies
;
1714 prepare_read_tag(con
);
1720 static int read_partial_message_section(struct ceph_connection
*con
,
1721 struct kvec
*section
,
1722 unsigned int sec_len
, u32
*crc
)
1728 while (section
->iov_len
< sec_len
) {
1729 BUG_ON(section
->iov_base
== NULL
);
1730 left
= sec_len
- section
->iov_len
;
1731 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)section
->iov_base
+
1732 section
->iov_len
, left
);
1735 section
->iov_len
+= ret
;
1737 if (section
->iov_len
== sec_len
)
1738 *crc
= crc32c(0, section
->iov_base
, section
->iov_len
);
1743 static int ceph_con_in_msg_alloc(struct ceph_connection
*con
, int *skip
);
1745 static int read_partial_message_pages(struct ceph_connection
*con
,
1746 struct page
**pages
,
1747 unsigned int data_len
, bool do_datacrc
)
1753 left
= min((int)(data_len
- con
->in_msg_pos
.data_pos
),
1754 (int)(PAGE_SIZE
- con
->in_msg_pos
.page_pos
));
1756 BUG_ON(pages
== NULL
);
1757 p
= kmap(pages
[con
->in_msg_pos
.page
]);
1758 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ con
->in_msg_pos
.page_pos
,
1760 if (ret
> 0 && do_datacrc
)
1762 crc32c(con
->in_data_crc
,
1763 p
+ con
->in_msg_pos
.page_pos
, ret
);
1764 kunmap(pages
[con
->in_msg_pos
.page
]);
1767 con
->in_msg_pos
.data_pos
+= ret
;
1768 con
->in_msg_pos
.page_pos
+= ret
;
1769 if (con
->in_msg_pos
.page_pos
== PAGE_SIZE
) {
1770 con
->in_msg_pos
.page_pos
= 0;
1771 con
->in_msg_pos
.page
++;
1778 static int read_partial_message_bio(struct ceph_connection
*con
,
1779 struct bio
**bio_iter
, int *bio_seg
,
1780 unsigned int data_len
, bool do_datacrc
)
1782 struct bio_vec
*bv
= bio_iovec_idx(*bio_iter
, *bio_seg
);
1786 left
= min((int)(data_len
- con
->in_msg_pos
.data_pos
),
1787 (int)(bv
->bv_len
- con
->in_msg_pos
.page_pos
));
1789 p
= kmap(bv
->bv_page
) + bv
->bv_offset
;
1791 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ con
->in_msg_pos
.page_pos
,
1793 if (ret
> 0 && do_datacrc
)
1795 crc32c(con
->in_data_crc
,
1796 p
+ con
->in_msg_pos
.page_pos
, ret
);
1797 kunmap(bv
->bv_page
);
1800 con
->in_msg_pos
.data_pos
+= ret
;
1801 con
->in_msg_pos
.page_pos
+= ret
;
1802 if (con
->in_msg_pos
.page_pos
== bv
->bv_len
) {
1803 con
->in_msg_pos
.page_pos
= 0;
1804 iter_bio_next(bio_iter
, bio_seg
);
1812 * read (part of) a message.
1814 static int read_partial_message(struct ceph_connection
*con
)
1816 struct ceph_msg
*m
= con
->in_msg
;
1820 unsigned int front_len
, middle_len
, data_len
;
1821 bool do_datacrc
= !con
->msgr
->nocrc
;
1825 dout("read_partial_message con %p msg %p\n", con
, m
);
1828 size
= sizeof (con
->in_hdr
);
1830 ret
= read_partial(con
, end
, size
, &con
->in_hdr
);
1834 crc
= crc32c(0, &con
->in_hdr
, offsetof(struct ceph_msg_header
, crc
));
1835 if (cpu_to_le32(crc
) != con
->in_hdr
.crc
) {
1836 pr_err("read_partial_message bad hdr "
1837 " crc %u != expected %u\n",
1838 crc
, con
->in_hdr
.crc
);
1842 front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
1843 if (front_len
> CEPH_MSG_MAX_FRONT_LEN
)
1845 middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
1846 if (middle_len
> CEPH_MSG_MAX_DATA_LEN
)
1848 data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
1849 if (data_len
> CEPH_MSG_MAX_DATA_LEN
)
1853 seq
= le64_to_cpu(con
->in_hdr
.seq
);
1854 if ((s64
)seq
- (s64
)con
->in_seq
< 1) {
1855 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1856 ENTITY_NAME(con
->peer_name
),
1857 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1858 seq
, con
->in_seq
+ 1);
1859 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1861 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1863 } else if ((s64
)seq
- (s64
)con
->in_seq
> 1) {
1864 pr_err("read_partial_message bad seq %lld expected %lld\n",
1865 seq
, con
->in_seq
+ 1);
1866 con
->error_msg
= "bad message sequence # for incoming message";
1870 /* allocate message? */
1874 dout("got hdr type %d front %d data %d\n", con
->in_hdr
.type
,
1875 con
->in_hdr
.front_len
, con
->in_hdr
.data_len
);
1876 ret
= ceph_con_in_msg_alloc(con
, &skip
);
1880 /* skip this message */
1881 dout("alloc_msg said skip message\n");
1882 BUG_ON(con
->in_msg
);
1883 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1885 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1890 BUG_ON(!con
->in_msg
);
1891 BUG_ON(con
->in_msg
->con
!= con
);
1893 m
->front
.iov_len
= 0; /* haven't read it yet */
1895 m
->middle
->vec
.iov_len
= 0;
1897 con
->in_msg_pos
.page
= 0;
1899 con
->in_msg_pos
.page_pos
= m
->page_alignment
;
1901 con
->in_msg_pos
.page_pos
= 0;
1902 con
->in_msg_pos
.data_pos
= 0;
1906 init_bio_iter(m
->bio
, &m
->bio_iter
, &m
->bio_seg
);
1911 ret
= read_partial_message_section(con
, &m
->front
, front_len
,
1912 &con
->in_front_crc
);
1918 ret
= read_partial_message_section(con
, &m
->middle
->vec
,
1920 &con
->in_middle_crc
);
1926 while (con
->in_msg_pos
.data_pos
< data_len
) {
1928 ret
= read_partial_message_pages(con
, m
->pages
,
1929 data_len
, do_datacrc
);
1933 } else if (m
->bio
) {
1934 BUG_ON(!m
->bio_iter
);
1935 ret
= read_partial_message_bio(con
,
1936 &m
->bio_iter
, &m
->bio_seg
,
1937 data_len
, do_datacrc
);
1947 size
= sizeof (m
->footer
);
1949 ret
= read_partial(con
, end
, size
, &m
->footer
);
1953 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1954 m
, front_len
, m
->footer
.front_crc
, middle_len
,
1955 m
->footer
.middle_crc
, data_len
, m
->footer
.data_crc
);
1958 if (con
->in_front_crc
!= le32_to_cpu(m
->footer
.front_crc
)) {
1959 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1960 m
, con
->in_front_crc
, m
->footer
.front_crc
);
1963 if (con
->in_middle_crc
!= le32_to_cpu(m
->footer
.middle_crc
)) {
1964 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1965 m
, con
->in_middle_crc
, m
->footer
.middle_crc
);
1969 (m
->footer
.flags
& CEPH_MSG_FOOTER_NOCRC
) == 0 &&
1970 con
->in_data_crc
!= le32_to_cpu(m
->footer
.data_crc
)) {
1971 pr_err("read_partial_message %p data crc %u != exp. %u\n", m
,
1972 con
->in_data_crc
, le32_to_cpu(m
->footer
.data_crc
));
1976 return 1; /* done! */
1980 * Process message. This happens in the worker thread. The callback should
1981 * be careful not to do anything that waits on other incoming messages or it
1984 static void process_message(struct ceph_connection
*con
)
1986 struct ceph_msg
*msg
;
1988 BUG_ON(con
->in_msg
->con
!= con
);
1989 con
->in_msg
->con
= NULL
;
1994 /* if first message, set peer_name */
1995 if (con
->peer_name
.type
== 0)
1996 con
->peer_name
= msg
->hdr
.src
;
1999 mutex_unlock(&con
->mutex
);
2001 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2002 msg
, le64_to_cpu(msg
->hdr
.seq
),
2003 ENTITY_NAME(msg
->hdr
.src
),
2004 le16_to_cpu(msg
->hdr
.type
),
2005 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2006 le32_to_cpu(msg
->hdr
.front_len
),
2007 le32_to_cpu(msg
->hdr
.data_len
),
2008 con
->in_front_crc
, con
->in_middle_crc
, con
->in_data_crc
);
2009 con
->ops
->dispatch(con
, msg
);
2011 mutex_lock(&con
->mutex
);
2016 * Write something to the socket. Called in a worker thread when the
2017 * socket appears to be writeable and we have something ready to send.
2019 static int try_write(struct ceph_connection
*con
)
2023 dout("try_write start %p state %lu\n", con
, con
->state
);
2026 dout("try_write out_kvec_bytes %d\n", con
->out_kvec_bytes
);
2028 /* open the socket first? */
2029 if (con
->state
== CON_STATE_PREOPEN
) {
2031 con
->state
= CON_STATE_CONNECTING
;
2033 con_out_kvec_reset(con
);
2034 prepare_write_banner(con
);
2035 prepare_read_banner(con
);
2037 BUG_ON(con
->in_msg
);
2038 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2039 dout("try_write initiating connect on %p new state %lu\n",
2041 ret
= ceph_tcp_connect(con
);
2043 con
->error_msg
= "connect error";
2049 /* kvec data queued? */
2050 if (con
->out_skip
) {
2051 ret
= write_partial_skip(con
);
2055 if (con
->out_kvec_left
) {
2056 ret
= write_partial_kvec(con
);
2063 if (con
->out_msg_done
) {
2064 ceph_msg_put(con
->out_msg
);
2065 con
->out_msg
= NULL
; /* we're done with this one */
2069 ret
= write_partial_msg_pages(con
);
2071 goto more_kvec
; /* we need to send the footer, too! */
2075 dout("try_write write_partial_msg_pages err %d\n",
2082 if (con
->state
== CON_STATE_OPEN
) {
2083 /* is anything else pending? */
2084 if (!list_empty(&con
->out_queue
)) {
2085 prepare_write_message(con
);
2088 if (con
->in_seq
> con
->in_seq_acked
) {
2089 prepare_write_ack(con
);
2092 if (test_and_clear_bit(CON_FLAG_KEEPALIVE_PENDING
,
2094 prepare_write_keepalive(con
);
2099 /* Nothing to do! */
2100 clear_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
);
2101 dout("try_write nothing else to write.\n");
2104 dout("try_write done on %p ret %d\n", con
, ret
);
2111 * Read what we can from the socket.
2113 static int try_read(struct ceph_connection
*con
)
2118 dout("try_read start on %p state %lu\n", con
, con
->state
);
2119 if (con
->state
!= CON_STATE_CONNECTING
&&
2120 con
->state
!= CON_STATE_NEGOTIATING
&&
2121 con
->state
!= CON_STATE_OPEN
)
2126 dout("try_read tag %d in_base_pos %d\n", (int)con
->in_tag
,
2129 if (con
->state
== CON_STATE_CONNECTING
) {
2130 dout("try_read connecting\n");
2131 ret
= read_partial_banner(con
);
2134 ret
= process_banner(con
);
2138 BUG_ON(con
->state
!= CON_STATE_CONNECTING
);
2139 con
->state
= CON_STATE_NEGOTIATING
;
2142 * Received banner is good, exchange connection info.
2143 * Do not reset out_kvec, as sending our banner raced
2144 * with receiving peer banner after connect completed.
2146 ret
= prepare_write_connect(con
);
2149 prepare_read_connect(con
);
2151 /* Send connection info before awaiting response */
2155 if (con
->state
== CON_STATE_NEGOTIATING
) {
2156 dout("try_read negotiating\n");
2157 ret
= read_partial_connect(con
);
2160 ret
= process_connect(con
);
2166 BUG_ON(con
->state
!= CON_STATE_OPEN
);
2168 if (con
->in_base_pos
< 0) {
2170 * skipping + discarding content.
2172 * FIXME: there must be a better way to do this!
2174 static char buf
[SKIP_BUF_SIZE
];
2175 int skip
= min((int) sizeof (buf
), -con
->in_base_pos
);
2177 dout("skipping %d / %d bytes\n", skip
, -con
->in_base_pos
);
2178 ret
= ceph_tcp_recvmsg(con
->sock
, buf
, skip
);
2181 con
->in_base_pos
+= ret
;
2182 if (con
->in_base_pos
)
2185 if (con
->in_tag
== CEPH_MSGR_TAG_READY
) {
2189 ret
= ceph_tcp_recvmsg(con
->sock
, &con
->in_tag
, 1);
2192 dout("try_read got tag %d\n", (int)con
->in_tag
);
2193 switch (con
->in_tag
) {
2194 case CEPH_MSGR_TAG_MSG
:
2195 prepare_read_message(con
);
2197 case CEPH_MSGR_TAG_ACK
:
2198 prepare_read_ack(con
);
2200 case CEPH_MSGR_TAG_CLOSE
:
2201 con_close_socket(con
);
2202 con
->state
= CON_STATE_CLOSED
;
2208 if (con
->in_tag
== CEPH_MSGR_TAG_MSG
) {
2209 ret
= read_partial_message(con
);
2213 con
->error_msg
= "bad crc";
2217 con
->error_msg
= "io error";
2222 if (con
->in_tag
== CEPH_MSGR_TAG_READY
)
2224 process_message(con
);
2225 if (con
->state
== CON_STATE_OPEN
)
2226 prepare_read_tag(con
);
2229 if (con
->in_tag
== CEPH_MSGR_TAG_ACK
) {
2230 ret
= read_partial_ack(con
);
2238 dout("try_read done on %p ret %d\n", con
, ret
);
2242 pr_err("try_read bad con->in_tag = %d\n", (int)con
->in_tag
);
2243 con
->error_msg
= "protocol error, garbage tag";
2250 * Atomically queue work on a connection. Bump @con reference to
2251 * avoid races with connection teardown.
2253 static void queue_con(struct ceph_connection
*con
)
2255 if (!con
->ops
->get(con
)) {
2256 dout("queue_con %p ref count 0\n", con
);
2260 if (!queue_delayed_work(ceph_msgr_wq
, &con
->work
, 0)) {
2261 dout("queue_con %p - already queued\n", con
);
2264 dout("queue_con %p\n", con
);
2269 * Do some work on a connection. Drop a connection ref when we're done.
2271 static void con_work(struct work_struct
*work
)
2273 struct ceph_connection
*con
= container_of(work
, struct ceph_connection
,
2277 mutex_lock(&con
->mutex
);
2279 if (test_and_clear_bit(CON_FLAG_SOCK_CLOSED
, &con
->flags
)) {
2280 switch (con
->state
) {
2281 case CON_STATE_CONNECTING
:
2282 con
->error_msg
= "connection failed";
2284 case CON_STATE_NEGOTIATING
:
2285 con
->error_msg
= "negotiation failed";
2287 case CON_STATE_OPEN
:
2288 con
->error_msg
= "socket closed";
2291 dout("unrecognized con state %d\n", (int)con
->state
);
2292 con
->error_msg
= "unrecognized con state";
2298 if (test_and_clear_bit(CON_FLAG_BACKOFF
, &con
->flags
)) {
2299 dout("con_work %p backing off\n", con
);
2300 if (queue_delayed_work(ceph_msgr_wq
, &con
->work
,
2301 round_jiffies_relative(con
->delay
))) {
2302 dout("con_work %p backoff %lu\n", con
, con
->delay
);
2303 mutex_unlock(&con
->mutex
);
2307 dout("con_work %p FAILED to back off %lu\n", con
,
2312 if (con
->state
== CON_STATE_STANDBY
) {
2313 dout("con_work %p STANDBY\n", con
);
2316 if (con
->state
== CON_STATE_CLOSED
) {
2317 dout("con_work %p CLOSED\n", con
);
2321 if (con
->state
== CON_STATE_PREOPEN
) {
2322 dout("con_work OPENING\n");
2326 ret
= try_read(con
);
2330 con
->error_msg
= "socket error on read";
2334 ret
= try_write(con
);
2338 con
->error_msg
= "socket error on write";
2343 mutex_unlock(&con
->mutex
);
2349 ceph_fault(con
); /* error/fault path */
2355 * Generic error/fault handler. A retry mechanism is used with
2356 * exponential backoff
2358 static void ceph_fault(struct ceph_connection
*con
)
2359 __releases(con
->mutex
)
2361 pr_err("%s%lld %s %s\n", ENTITY_NAME(con
->peer_name
),
2362 ceph_pr_addr(&con
->peer_addr
.in_addr
), con
->error_msg
);
2363 dout("fault %p state %lu to peer %s\n",
2364 con
, con
->state
, ceph_pr_addr(&con
->peer_addr
.in_addr
));
2366 BUG_ON(con
->state
!= CON_STATE_CONNECTING
&&
2367 con
->state
!= CON_STATE_NEGOTIATING
&&
2368 con
->state
!= CON_STATE_OPEN
);
2370 con_close_socket(con
);
2372 if (test_bit(CON_FLAG_LOSSYTX
, &con
->flags
)) {
2373 dout("fault on LOSSYTX channel, marking CLOSED\n");
2374 con
->state
= CON_STATE_CLOSED
;
2379 BUG_ON(con
->in_msg
->con
!= con
);
2380 con
->in_msg
->con
= NULL
;
2381 ceph_msg_put(con
->in_msg
);
2386 /* Requeue anything that hasn't been acked */
2387 list_splice_init(&con
->out_sent
, &con
->out_queue
);
2389 /* If there are no messages queued or keepalive pending, place
2390 * the connection in a STANDBY state */
2391 if (list_empty(&con
->out_queue
) &&
2392 !test_bit(CON_FLAG_KEEPALIVE_PENDING
, &con
->flags
)) {
2393 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con
);
2394 clear_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
);
2395 con
->state
= CON_STATE_STANDBY
;
2397 /* retry after a delay. */
2398 con
->state
= CON_STATE_PREOPEN
;
2399 if (con
->delay
== 0)
2400 con
->delay
= BASE_DELAY_INTERVAL
;
2401 else if (con
->delay
< MAX_DELAY_INTERVAL
)
2404 if (queue_delayed_work(ceph_msgr_wq
, &con
->work
,
2405 round_jiffies_relative(con
->delay
))) {
2406 dout("fault queued %p delay %lu\n", con
, con
->delay
);
2409 dout("fault failed to queue %p delay %lu, backoff\n",
2412 * In many cases we see a socket state change
2413 * while con_work is running and end up
2414 * queuing (non-delayed) work, such that we
2415 * can't backoff with a delay. Set a flag so
2416 * that when con_work restarts we schedule the
2419 set_bit(CON_FLAG_BACKOFF
, &con
->flags
);
2424 mutex_unlock(&con
->mutex
);
2426 * in case we faulted due to authentication, invalidate our
2427 * current tickets so that we can get new ones.
2429 if (con
->auth_retry
&& con
->ops
->invalidate_authorizer
) {
2430 dout("calling invalidate_authorizer()\n");
2431 con
->ops
->invalidate_authorizer(con
);
2434 if (con
->ops
->fault
)
2435 con
->ops
->fault(con
);
2441 * initialize a new messenger instance
2443 void ceph_messenger_init(struct ceph_messenger
*msgr
,
2444 struct ceph_entity_addr
*myaddr
,
2445 u32 supported_features
,
2446 u32 required_features
,
2449 msgr
->supported_features
= supported_features
;
2450 msgr
->required_features
= required_features
;
2452 spin_lock_init(&msgr
->global_seq_lock
);
2455 msgr
->inst
.addr
= *myaddr
;
2457 /* select a random nonce */
2458 msgr
->inst
.addr
.type
= 0;
2459 get_random_bytes(&msgr
->inst
.addr
.nonce
, sizeof(msgr
->inst
.addr
.nonce
));
2460 encode_my_addr(msgr
);
2461 msgr
->nocrc
= nocrc
;
2463 atomic_set(&msgr
->stopping
, 0);
2465 dout("%s %p\n", __func__
, msgr
);
2467 EXPORT_SYMBOL(ceph_messenger_init
);
2469 static void clear_standby(struct ceph_connection
*con
)
2471 /* come back from STANDBY? */
2472 if (con
->state
== CON_STATE_STANDBY
) {
2473 dout("clear_standby %p and ++connect_seq\n", con
);
2474 con
->state
= CON_STATE_PREOPEN
;
2476 WARN_ON(test_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
));
2477 WARN_ON(test_bit(CON_FLAG_KEEPALIVE_PENDING
, &con
->flags
));
2482 * Queue up an outgoing message on the given connection.
2484 void ceph_con_send(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2487 msg
->hdr
.src
= con
->msgr
->inst
.name
;
2488 BUG_ON(msg
->front
.iov_len
!= le32_to_cpu(msg
->hdr
.front_len
));
2489 msg
->needs_out_seq
= true;
2491 mutex_lock(&con
->mutex
);
2493 if (con
->state
== CON_STATE_CLOSED
) {
2494 dout("con_send %p closed, dropping %p\n", con
, msg
);
2496 mutex_unlock(&con
->mutex
);
2500 BUG_ON(msg
->con
!= NULL
);
2501 msg
->con
= con
->ops
->get(con
);
2502 BUG_ON(msg
->con
== NULL
);
2504 BUG_ON(!list_empty(&msg
->list_head
));
2505 list_add_tail(&msg
->list_head
, &con
->out_queue
);
2506 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg
,
2507 ENTITY_NAME(con
->peer_name
), le16_to_cpu(msg
->hdr
.type
),
2508 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2509 le32_to_cpu(msg
->hdr
.front_len
),
2510 le32_to_cpu(msg
->hdr
.middle_len
),
2511 le32_to_cpu(msg
->hdr
.data_len
));
2514 mutex_unlock(&con
->mutex
);
2516 /* if there wasn't anything waiting to send before, queue
2518 if (test_and_set_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
) == 0)
2521 EXPORT_SYMBOL(ceph_con_send
);
2524 * Revoke a message that was previously queued for send
2526 void ceph_msg_revoke(struct ceph_msg
*msg
)
2528 struct ceph_connection
*con
= msg
->con
;
2531 return; /* Message not in our possession */
2533 mutex_lock(&con
->mutex
);
2534 if (!list_empty(&msg
->list_head
)) {
2535 dout("%s %p msg %p - was on queue\n", __func__
, con
, msg
);
2536 list_del_init(&msg
->list_head
);
2537 BUG_ON(msg
->con
== NULL
);
2538 msg
->con
->ops
->put(msg
->con
);
2544 if (con
->out_msg
== msg
) {
2545 dout("%s %p msg %p - was sending\n", __func__
, con
, msg
);
2546 con
->out_msg
= NULL
;
2547 if (con
->out_kvec_is_msg
) {
2548 con
->out_skip
= con
->out_kvec_bytes
;
2549 con
->out_kvec_is_msg
= false;
2555 mutex_unlock(&con
->mutex
);
2559 * Revoke a message that we may be reading data into
2561 void ceph_msg_revoke_incoming(struct ceph_msg
*msg
)
2563 struct ceph_connection
*con
;
2565 BUG_ON(msg
== NULL
);
2567 dout("%s msg %p null con\n", __func__
, msg
);
2569 return; /* Message not in our possession */
2573 mutex_lock(&con
->mutex
);
2574 if (con
->in_msg
== msg
) {
2575 unsigned int front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
2576 unsigned int middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
2577 unsigned int data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
2579 /* skip rest of message */
2580 dout("%s %p msg %p revoked\n", __func__
, con
, msg
);
2581 con
->in_base_pos
= con
->in_base_pos
-
2582 sizeof(struct ceph_msg_header
) -
2586 sizeof(struct ceph_msg_footer
);
2587 ceph_msg_put(con
->in_msg
);
2589 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2592 dout("%s %p in_msg %p msg %p no-op\n",
2593 __func__
, con
, con
->in_msg
, msg
);
2595 mutex_unlock(&con
->mutex
);
2599 * Queue a keepalive byte to ensure the tcp connection is alive.
2601 void ceph_con_keepalive(struct ceph_connection
*con
)
2603 dout("con_keepalive %p\n", con
);
2604 mutex_lock(&con
->mutex
);
2606 mutex_unlock(&con
->mutex
);
2607 if (test_and_set_bit(CON_FLAG_KEEPALIVE_PENDING
, &con
->flags
) == 0 &&
2608 test_and_set_bit(CON_FLAG_WRITE_PENDING
, &con
->flags
) == 0)
2611 EXPORT_SYMBOL(ceph_con_keepalive
);
2615 * construct a new message with given type, size
2616 * the new msg has a ref count of 1.
2618 struct ceph_msg
*ceph_msg_new(int type
, int front_len
, gfp_t flags
,
2623 m
= kmalloc(sizeof(*m
), flags
);
2626 kref_init(&m
->kref
);
2629 INIT_LIST_HEAD(&m
->list_head
);
2632 m
->hdr
.type
= cpu_to_le16(type
);
2633 m
->hdr
.priority
= cpu_to_le16(CEPH_MSG_PRIO_DEFAULT
);
2635 m
->hdr
.front_len
= cpu_to_le32(front_len
);
2636 m
->hdr
.middle_len
= 0;
2637 m
->hdr
.data_len
= 0;
2638 m
->hdr
.data_off
= 0;
2639 m
->hdr
.reserved
= 0;
2640 m
->footer
.front_crc
= 0;
2641 m
->footer
.middle_crc
= 0;
2642 m
->footer
.data_crc
= 0;
2643 m
->footer
.flags
= 0;
2644 m
->front_max
= front_len
;
2645 m
->front_is_vmalloc
= false;
2646 m
->more_to_follow
= false;
2655 m
->page_alignment
= 0;
2665 if (front_len
> PAGE_CACHE_SIZE
) {
2666 m
->front
.iov_base
= __vmalloc(front_len
, flags
,
2668 m
->front_is_vmalloc
= true;
2670 m
->front
.iov_base
= kmalloc(front_len
, flags
);
2672 if (m
->front
.iov_base
== NULL
) {
2673 dout("ceph_msg_new can't allocate %d bytes\n",
2678 m
->front
.iov_base
= NULL
;
2680 m
->front
.iov_len
= front_len
;
2682 dout("ceph_msg_new %p front %d\n", m
, front_len
);
2689 pr_err("msg_new can't create type %d front %d\n", type
,
2693 dout("msg_new can't create type %d front %d\n", type
,
2698 EXPORT_SYMBOL(ceph_msg_new
);
2701 * Allocate "middle" portion of a message, if it is needed and wasn't
2702 * allocated by alloc_msg. This allows us to read a small fixed-size
2703 * per-type header in the front and then gracefully fail (i.e.,
2704 * propagate the error to the caller based on info in the front) when
2705 * the middle is too large.
2707 static int ceph_alloc_middle(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2709 int type
= le16_to_cpu(msg
->hdr
.type
);
2710 int middle_len
= le32_to_cpu(msg
->hdr
.middle_len
);
2712 dout("alloc_middle %p type %d %s middle_len %d\n", msg
, type
,
2713 ceph_msg_type_name(type
), middle_len
);
2714 BUG_ON(!middle_len
);
2715 BUG_ON(msg
->middle
);
2717 msg
->middle
= ceph_buffer_new(middle_len
, GFP_NOFS
);
2724 * Allocate a message for receiving an incoming message on a
2725 * connection, and save the result in con->in_msg. Uses the
2726 * connection's private alloc_msg op if available.
2728 * Returns 0 on success, or a negative error code.
2730 * On success, if we set *skip = 1:
2731 * - the next message should be skipped and ignored.
2732 * - con->in_msg == NULL
2733 * or if we set *skip = 0:
2734 * - con->in_msg is non-null.
2735 * On error (ENOMEM, EAGAIN, ...),
2736 * - con->in_msg == NULL
2738 static int ceph_con_in_msg_alloc(struct ceph_connection
*con
, int *skip
)
2740 struct ceph_msg_header
*hdr
= &con
->in_hdr
;
2741 int type
= le16_to_cpu(hdr
->type
);
2742 int front_len
= le32_to_cpu(hdr
->front_len
);
2743 int middle_len
= le32_to_cpu(hdr
->middle_len
);
2746 BUG_ON(con
->in_msg
!= NULL
);
2748 if (con
->ops
->alloc_msg
) {
2749 struct ceph_msg
*msg
;
2751 mutex_unlock(&con
->mutex
);
2752 msg
= con
->ops
->alloc_msg(con
, hdr
, skip
);
2753 mutex_lock(&con
->mutex
);
2754 if (con
->state
!= CON_STATE_OPEN
) {
2760 con
->in_msg
->con
= con
->ops
->get(con
);
2761 BUG_ON(con
->in_msg
->con
== NULL
);
2769 "error allocating memory for incoming message";
2774 con
->in_msg
= ceph_msg_new(type
, front_len
, GFP_NOFS
, false);
2776 pr_err("unable to allocate msg type %d len %d\n",
2780 con
->in_msg
->con
= con
->ops
->get(con
);
2781 BUG_ON(con
->in_msg
->con
== NULL
);
2782 con
->in_msg
->page_alignment
= le16_to_cpu(hdr
->data_off
);
2784 memcpy(&con
->in_msg
->hdr
, &con
->in_hdr
, sizeof(con
->in_hdr
));
2786 if (middle_len
&& !con
->in_msg
->middle
) {
2787 ret
= ceph_alloc_middle(con
, con
->in_msg
);
2789 ceph_msg_put(con
->in_msg
);
2799 * Free a generically kmalloc'd message.
2801 void ceph_msg_kfree(struct ceph_msg
*m
)
2803 dout("msg_kfree %p\n", m
);
2804 if (m
->front_is_vmalloc
)
2805 vfree(m
->front
.iov_base
);
2807 kfree(m
->front
.iov_base
);
2812 * Drop a msg ref. Destroy as needed.
2814 void ceph_msg_last_put(struct kref
*kref
)
2816 struct ceph_msg
*m
= container_of(kref
, struct ceph_msg
, kref
);
2818 dout("ceph_msg_put last one on %p\n", m
);
2819 WARN_ON(!list_empty(&m
->list_head
));
2821 /* drop middle, data, if any */
2823 ceph_buffer_put(m
->middle
);
2830 ceph_pagelist_release(m
->pagelist
);
2838 ceph_msgpool_put(m
->pool
, m
);
2842 EXPORT_SYMBOL(ceph_msg_last_put
);
2844 void ceph_msg_dump(struct ceph_msg
*msg
)
2846 pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg
,
2847 msg
->front_max
, msg
->nr_pages
);
2848 print_hex_dump(KERN_DEBUG
, "header: ",
2849 DUMP_PREFIX_OFFSET
, 16, 1,
2850 &msg
->hdr
, sizeof(msg
->hdr
), true);
2851 print_hex_dump(KERN_DEBUG
, " front: ",
2852 DUMP_PREFIX_OFFSET
, 16, 1,
2853 msg
->front
.iov_base
, msg
->front
.iov_len
, true);
2855 print_hex_dump(KERN_DEBUG
, "middle: ",
2856 DUMP_PREFIX_OFFSET
, 16, 1,
2857 msg
->middle
->vec
.iov_base
,
2858 msg
->middle
->vec
.iov_len
, true);
2859 print_hex_dump(KERN_DEBUG
, "footer: ",
2860 DUMP_PREFIX_OFFSET
, 16, 1,
2861 &msg
->footer
, sizeof(msg
->footer
), true);
2863 EXPORT_SYMBOL(ceph_msg_dump
);