1 #include <linux/ceph/ceph_debug.h>
3 #include <linux/crc32c.h>
4 #include <linux/ctype.h>
5 #include <linux/highmem.h>
6 #include <linux/inet.h>
7 #include <linux/kthread.h>
9 #include <linux/slab.h>
10 #include <linux/socket.h>
11 #include <linux/string.h>
12 #include <linux/bio.h>
13 #include <linux/blkdev.h>
14 #include <linux/dns_resolver.h>
17 #include <linux/ceph/libceph.h>
18 #include <linux/ceph/messenger.h>
19 #include <linux/ceph/decode.h>
20 #include <linux/ceph/pagelist.h>
21 #include <linux/export.h>
24 * Ceph uses the messenger to exchange ceph_msg messages with other
25 * hosts in the system. The messenger provides ordered and reliable
26 * delivery. We tolerate TCP disconnects by reconnecting (with
27 * exponential backoff) in the case of a fault (disconnection, bad
28 * crc, protocol error). Acks allow sent messages to be discarded by
33 * We track the state of the socket on a given connection using
34 * values defined below. The transition to a new socket state is
35 * handled by a function which verifies we aren't coming from an
39 * | NEW* | transient initial state
41 * | con_sock_state_init()
44 * | CLOSED | initialized, but no socket (and no
45 * ---------- TCP connection)
47 * | \ con_sock_state_connecting()
48 * | ----------------------
50 * + con_sock_state_closed() \
51 * |+--------------------------- \
54 * | | CLOSING | socket event; \ \
55 * | ----------- await close \ \
58 * | + con_sock_state_closing() \ |
60 * | / --------------- | |
63 * | / -----------------| CONNECTING | socket created, TCP
64 * | | / -------------- connect initiated
65 * | | | con_sock_state_connected()
68 * | CONNECTED | TCP connection established
71 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
74 #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
75 #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
76 #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
77 #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
78 #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
80 /* static tag bytes (protocol control messages) */
81 static char tag_msg
= CEPH_MSGR_TAG_MSG
;
82 static char tag_ack
= CEPH_MSGR_TAG_ACK
;
83 static char tag_keepalive
= CEPH_MSGR_TAG_KEEPALIVE
;
86 static struct lock_class_key socket_class
;
90 * When skipping (ignoring) a block of input we read it into a "skip
91 * buffer," which is this many bytes in size.
93 #define SKIP_BUF_SIZE 1024
95 static void queue_con(struct ceph_connection
*con
);
96 static void con_work(struct work_struct
*);
97 static void ceph_fault(struct ceph_connection
*con
);
100 * Nicely render a sockaddr as a string. An array of formatted
101 * strings is used, to approximate reentrancy.
103 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
104 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
105 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
106 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
108 static char addr_str
[ADDR_STR_COUNT
][MAX_ADDR_STR_LEN
];
109 static atomic_t addr_str_seq
= ATOMIC_INIT(0);
111 static struct page
*zero_page
; /* used in certain error cases */
113 const char *ceph_pr_addr(const struct sockaddr_storage
*ss
)
117 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
118 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
120 i
= atomic_inc_return(&addr_str_seq
) & ADDR_STR_COUNT_MASK
;
123 switch (ss
->ss_family
) {
125 snprintf(s
, MAX_ADDR_STR_LEN
, "%pI4:%hu", &in4
->sin_addr
,
126 ntohs(in4
->sin_port
));
130 snprintf(s
, MAX_ADDR_STR_LEN
, "[%pI6c]:%hu", &in6
->sin6_addr
,
131 ntohs(in6
->sin6_port
));
135 snprintf(s
, MAX_ADDR_STR_LEN
, "(unknown sockaddr family %hu)",
141 EXPORT_SYMBOL(ceph_pr_addr
);
143 static void encode_my_addr(struct ceph_messenger
*msgr
)
145 memcpy(&msgr
->my_enc_addr
, &msgr
->inst
.addr
, sizeof(msgr
->my_enc_addr
));
146 ceph_encode_addr(&msgr
->my_enc_addr
);
150 * work queue for all reading and writing to/from the socket.
152 static struct workqueue_struct
*ceph_msgr_wq
;
154 void _ceph_msgr_exit(void)
157 destroy_workqueue(ceph_msgr_wq
);
161 BUG_ON(zero_page
== NULL
);
163 page_cache_release(zero_page
);
167 int ceph_msgr_init(void)
169 BUG_ON(zero_page
!= NULL
);
170 zero_page
= ZERO_PAGE(0);
171 page_cache_get(zero_page
);
173 ceph_msgr_wq
= alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT
, 0);
177 pr_err("msgr_init failed to create workqueue\n");
182 EXPORT_SYMBOL(ceph_msgr_init
);
184 void ceph_msgr_exit(void)
186 BUG_ON(ceph_msgr_wq
== NULL
);
190 EXPORT_SYMBOL(ceph_msgr_exit
);
192 void ceph_msgr_flush(void)
194 flush_workqueue(ceph_msgr_wq
);
196 EXPORT_SYMBOL(ceph_msgr_flush
);
198 /* Connection socket state transition functions */
200 static void con_sock_state_init(struct ceph_connection
*con
)
204 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSED
);
205 if (WARN_ON(old_state
!= CON_SOCK_STATE_NEW
))
206 printk("%s: unexpected old state %d\n", __func__
, old_state
);
209 static void con_sock_state_connecting(struct ceph_connection
*con
)
213 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CONNECTING
);
214 if (WARN_ON(old_state
!= CON_SOCK_STATE_CLOSED
))
215 printk("%s: unexpected old state %d\n", __func__
, old_state
);
218 static void con_sock_state_connected(struct ceph_connection
*con
)
222 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CONNECTED
);
223 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTING
))
224 printk("%s: unexpected old state %d\n", __func__
, old_state
);
227 static void con_sock_state_closing(struct ceph_connection
*con
)
231 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSING
);
232 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTING
&&
233 old_state
!= CON_SOCK_STATE_CONNECTED
&&
234 old_state
!= CON_SOCK_STATE_CLOSING
))
235 printk("%s: unexpected old state %d\n", __func__
, old_state
);
238 static void con_sock_state_closed(struct ceph_connection
*con
)
242 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSED
);
243 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTED
&&
244 old_state
!= CON_SOCK_STATE_CLOSING
&&
245 old_state
!= CON_SOCK_STATE_CONNECTING
))
246 printk("%s: unexpected old state %d\n", __func__
, old_state
);
250 * socket callback functions
253 /* data available on socket, or listen socket received a connect */
254 static void ceph_sock_data_ready(struct sock
*sk
, int count_unused
)
256 struct ceph_connection
*con
= sk
->sk_user_data
;
257 if (atomic_read(&con
->msgr
->stopping
)) {
261 if (sk
->sk_state
!= TCP_CLOSE_WAIT
) {
262 dout("%s on %p state = %lu, queueing work\n", __func__
,
268 /* socket has buffer space for writing */
269 static void ceph_sock_write_space(struct sock
*sk
)
271 struct ceph_connection
*con
= sk
->sk_user_data
;
273 /* only queue to workqueue if there is data we want to write,
274 * and there is sufficient space in the socket buffer to accept
275 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
276 * doesn't get called again until try_write() fills the socket
277 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
278 * and net/core/stream.c:sk_stream_write_space().
280 if (test_bit(WRITE_PENDING
, &con
->flags
)) {
281 if (sk_stream_wspace(sk
) >= sk_stream_min_wspace(sk
)) {
282 dout("%s %p queueing write work\n", __func__
, con
);
283 clear_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
287 dout("%s %p nothing to write\n", __func__
, con
);
291 /* socket's state has changed */
292 static void ceph_sock_state_change(struct sock
*sk
)
294 struct ceph_connection
*con
= sk
->sk_user_data
;
296 dout("%s %p state = %lu sk_state = %u\n", __func__
,
297 con
, con
->state
, sk
->sk_state
);
299 switch (sk
->sk_state
) {
301 dout("%s TCP_CLOSE\n", __func__
);
303 dout("%s TCP_CLOSE_WAIT\n", __func__
);
304 con_sock_state_closing(con
);
305 set_bit(SOCK_CLOSED
, &con
->flags
);
308 case TCP_ESTABLISHED
:
309 dout("%s TCP_ESTABLISHED\n", __func__
);
310 con_sock_state_connected(con
);
313 default: /* Everything else is uninteresting */
319 * set up socket callbacks
321 static void set_sock_callbacks(struct socket
*sock
,
322 struct ceph_connection
*con
)
324 struct sock
*sk
= sock
->sk
;
325 sk
->sk_user_data
= con
;
326 sk
->sk_data_ready
= ceph_sock_data_ready
;
327 sk
->sk_write_space
= ceph_sock_write_space
;
328 sk
->sk_state_change
= ceph_sock_state_change
;
337 * initiate connection to a remote socket.
339 static int ceph_tcp_connect(struct ceph_connection
*con
)
341 struct sockaddr_storage
*paddr
= &con
->peer_addr
.in_addr
;
346 ret
= sock_create_kern(con
->peer_addr
.in_addr
.ss_family
, SOCK_STREAM
,
350 sock
->sk
->sk_allocation
= GFP_NOFS
;
352 #ifdef CONFIG_LOCKDEP
353 lockdep_set_class(&sock
->sk
->sk_lock
, &socket_class
);
356 set_sock_callbacks(sock
, con
);
358 dout("connect %s\n", ceph_pr_addr(&con
->peer_addr
.in_addr
));
360 con_sock_state_connecting(con
);
361 ret
= sock
->ops
->connect(sock
, (struct sockaddr
*)paddr
, sizeof(*paddr
),
363 if (ret
== -EINPROGRESS
) {
364 dout("connect %s EINPROGRESS sk_state = %u\n",
365 ceph_pr_addr(&con
->peer_addr
.in_addr
),
367 } else if (ret
< 0) {
368 pr_err("connect %s error %d\n",
369 ceph_pr_addr(&con
->peer_addr
.in_addr
), ret
);
371 con
->error_msg
= "connect error";
379 static int ceph_tcp_recvmsg(struct socket
*sock
, void *buf
, size_t len
)
381 struct kvec iov
= {buf
, len
};
382 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
385 r
= kernel_recvmsg(sock
, &msg
, &iov
, 1, len
, msg
.msg_flags
);
392 * write something. @more is true if caller will be sending more data
395 static int ceph_tcp_sendmsg(struct socket
*sock
, struct kvec
*iov
,
396 size_t kvlen
, size_t len
, int more
)
398 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
402 msg
.msg_flags
|= MSG_MORE
;
404 msg
.msg_flags
|= MSG_EOR
; /* superfluous, but what the hell */
406 r
= kernel_sendmsg(sock
, &msg
, iov
, kvlen
, len
);
412 static int ceph_tcp_sendpage(struct socket
*sock
, struct page
*page
,
413 int offset
, size_t size
, int more
)
415 int flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
| (more
? MSG_MORE
: MSG_EOR
);
418 ret
= kernel_sendpage(sock
, page
, offset
, size
, flags
);
427 * Shutdown/close the socket for the given connection.
429 static int con_close_socket(struct ceph_connection
*con
)
433 dout("con_close_socket on %p sock %p\n", con
, con
->sock
);
436 rc
= con
->sock
->ops
->shutdown(con
->sock
, SHUT_RDWR
);
437 sock_release(con
->sock
);
441 * Forcibly clear the SOCK_CLOSE flag. It gets set
442 * independent of the connection mutex, and we could have
443 * received a socket close event before we had the chance to
444 * shut the socket down.
446 clear_bit(SOCK_CLOSED
, &con
->flags
);
447 con_sock_state_closed(con
);
452 * Reset a connection. Discard all incoming and outgoing messages
453 * and clear *_seq state.
455 static void ceph_msg_remove(struct ceph_msg
*msg
)
457 list_del_init(&msg
->list_head
);
458 BUG_ON(msg
->con
== NULL
);
459 msg
->con
->ops
->put(msg
->con
);
464 static void ceph_msg_remove_list(struct list_head
*head
)
466 while (!list_empty(head
)) {
467 struct ceph_msg
*msg
= list_first_entry(head
, struct ceph_msg
,
469 ceph_msg_remove(msg
);
473 static void reset_connection(struct ceph_connection
*con
)
475 /* reset connection, out_queue, msg_ and connect_seq */
476 /* discard existing out_queue and msg_seq */
477 ceph_msg_remove_list(&con
->out_queue
);
478 ceph_msg_remove_list(&con
->out_sent
);
481 BUG_ON(con
->in_msg
->con
!= con
);
482 con
->in_msg
->con
= NULL
;
483 ceph_msg_put(con
->in_msg
);
488 con
->connect_seq
= 0;
491 ceph_msg_put(con
->out_msg
);
495 con
->in_seq_acked
= 0;
499 * mark a peer down. drop any open connections.
501 void ceph_con_close(struct ceph_connection
*con
)
503 mutex_lock(&con
->mutex
);
504 dout("con_close %p peer %s\n", con
,
505 ceph_pr_addr(&con
->peer_addr
.in_addr
));
506 clear_bit(NEGOTIATING
, &con
->state
);
507 clear_bit(CONNECTING
, &con
->state
);
508 clear_bit(CONNECTED
, &con
->state
);
509 clear_bit(STANDBY
, &con
->state
); /* avoid connect_seq bump */
510 set_bit(CLOSED
, &con
->state
);
512 clear_bit(LOSSYTX
, &con
->flags
); /* so we retry next connect */
513 clear_bit(KEEPALIVE_PENDING
, &con
->flags
);
514 clear_bit(WRITE_PENDING
, &con
->flags
);
516 reset_connection(con
);
517 con
->peer_global_seq
= 0;
518 cancel_delayed_work(&con
->work
);
519 con_close_socket(con
);
520 mutex_unlock(&con
->mutex
);
522 EXPORT_SYMBOL(ceph_con_close
);
525 * Reopen a closed connection, with a new peer address.
527 void ceph_con_open(struct ceph_connection
*con
,
528 __u8 entity_type
, __u64 entity_num
,
529 struct ceph_entity_addr
*addr
)
531 mutex_lock(&con
->mutex
);
532 dout("con_open %p %s\n", con
, ceph_pr_addr(&addr
->in_addr
));
533 set_bit(OPENING
, &con
->state
);
534 WARN_ON(!test_and_clear_bit(CLOSED
, &con
->state
));
536 con
->peer_name
.type
= (__u8
) entity_type
;
537 con
->peer_name
.num
= cpu_to_le64(entity_num
);
539 memcpy(&con
->peer_addr
, addr
, sizeof(*addr
));
540 con
->delay
= 0; /* reset backoff memory */
541 mutex_unlock(&con
->mutex
);
544 EXPORT_SYMBOL(ceph_con_open
);
547 * return true if this connection ever successfully opened
549 bool ceph_con_opened(struct ceph_connection
*con
)
551 return con
->connect_seq
> 0;
555 * initialize a new connection.
557 void ceph_con_init(struct ceph_connection
*con
, void *private,
558 const struct ceph_connection_operations
*ops
,
559 struct ceph_messenger
*msgr
)
561 dout("con_init %p\n", con
);
562 memset(con
, 0, sizeof(*con
));
563 con
->private = private;
567 con_sock_state_init(con
);
569 mutex_init(&con
->mutex
);
570 INIT_LIST_HEAD(&con
->out_queue
);
571 INIT_LIST_HEAD(&con
->out_sent
);
572 INIT_DELAYED_WORK(&con
->work
, con_work
);
574 set_bit(CLOSED
, &con
->state
);
576 EXPORT_SYMBOL(ceph_con_init
);
580 * We maintain a global counter to order connection attempts. Get
581 * a unique seq greater than @gt.
583 static u32
get_global_seq(struct ceph_messenger
*msgr
, u32 gt
)
587 spin_lock(&msgr
->global_seq_lock
);
588 if (msgr
->global_seq
< gt
)
589 msgr
->global_seq
= gt
;
590 ret
= ++msgr
->global_seq
;
591 spin_unlock(&msgr
->global_seq_lock
);
595 static void con_out_kvec_reset(struct ceph_connection
*con
)
597 con
->out_kvec_left
= 0;
598 con
->out_kvec_bytes
= 0;
599 con
->out_kvec_cur
= &con
->out_kvec
[0];
602 static void con_out_kvec_add(struct ceph_connection
*con
,
603 size_t size
, void *data
)
607 index
= con
->out_kvec_left
;
608 BUG_ON(index
>= ARRAY_SIZE(con
->out_kvec
));
610 con
->out_kvec
[index
].iov_len
= size
;
611 con
->out_kvec
[index
].iov_base
= data
;
612 con
->out_kvec_left
++;
613 con
->out_kvec_bytes
+= size
;
617 static void init_bio_iter(struct bio
*bio
, struct bio
**iter
, int *seg
)
628 static void iter_bio_next(struct bio
**bio_iter
, int *seg
)
630 if (*bio_iter
== NULL
)
633 BUG_ON(*seg
>= (*bio_iter
)->bi_vcnt
);
636 if (*seg
== (*bio_iter
)->bi_vcnt
)
637 init_bio_iter((*bio_iter
)->bi_next
, bio_iter
, seg
);
641 static void prepare_write_message_data(struct ceph_connection
*con
)
643 struct ceph_msg
*msg
= con
->out_msg
;
646 BUG_ON(!msg
->hdr
.data_len
);
648 /* initialize page iterator */
649 con
->out_msg_pos
.page
= 0;
651 con
->out_msg_pos
.page_pos
= msg
->page_alignment
;
653 con
->out_msg_pos
.page_pos
= 0;
656 init_bio_iter(msg
->bio
, &msg
->bio_iter
, &msg
->bio_seg
);
658 con
->out_msg_pos
.data_pos
= 0;
659 con
->out_msg_pos
.did_page_crc
= false;
660 con
->out_more
= 1; /* data + footer will follow */
664 * Prepare footer for currently outgoing message, and finish things
665 * off. Assumes out_kvec* are already valid.. we just add on to the end.
667 static void prepare_write_message_footer(struct ceph_connection
*con
)
669 struct ceph_msg
*m
= con
->out_msg
;
670 int v
= con
->out_kvec_left
;
672 m
->footer
.flags
|= CEPH_MSG_FOOTER_COMPLETE
;
674 dout("prepare_write_message_footer %p\n", con
);
675 con
->out_kvec_is_msg
= true;
676 con
->out_kvec
[v
].iov_base
= &m
->footer
;
677 con
->out_kvec
[v
].iov_len
= sizeof(m
->footer
);
678 con
->out_kvec_bytes
+= sizeof(m
->footer
);
679 con
->out_kvec_left
++;
680 con
->out_more
= m
->more_to_follow
;
681 con
->out_msg_done
= true;
685 * Prepare headers for the next outgoing message.
687 static void prepare_write_message(struct ceph_connection
*con
)
692 con_out_kvec_reset(con
);
693 con
->out_kvec_is_msg
= true;
694 con
->out_msg_done
= false;
696 /* Sneak an ack in there first? If we can get it into the same
697 * TCP packet that's a good thing. */
698 if (con
->in_seq
> con
->in_seq_acked
) {
699 con
->in_seq_acked
= con
->in_seq
;
700 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
701 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
702 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
706 BUG_ON(list_empty(&con
->out_queue
));
707 m
= list_first_entry(&con
->out_queue
, struct ceph_msg
, list_head
);
709 BUG_ON(m
->con
!= con
);
711 /* put message on sent list */
713 list_move_tail(&m
->list_head
, &con
->out_sent
);
716 * only assign outgoing seq # if we haven't sent this message
717 * yet. if it is requeued, resend with it's original seq.
719 if (m
->needs_out_seq
) {
720 m
->hdr
.seq
= cpu_to_le64(++con
->out_seq
);
721 m
->needs_out_seq
= false;
724 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n",
725 m
, con
->out_seq
, le16_to_cpu(m
->hdr
.type
),
726 le32_to_cpu(m
->hdr
.front_len
), le32_to_cpu(m
->hdr
.middle_len
),
727 le32_to_cpu(m
->hdr
.data_len
),
729 BUG_ON(le32_to_cpu(m
->hdr
.front_len
) != m
->front
.iov_len
);
731 /* tag + hdr + front + middle */
732 con_out_kvec_add(con
, sizeof (tag_msg
), &tag_msg
);
733 con_out_kvec_add(con
, sizeof (m
->hdr
), &m
->hdr
);
734 con_out_kvec_add(con
, m
->front
.iov_len
, m
->front
.iov_base
);
737 con_out_kvec_add(con
, m
->middle
->vec
.iov_len
,
738 m
->middle
->vec
.iov_base
);
740 /* fill in crc (except data pages), footer */
741 crc
= crc32c(0, &m
->hdr
, offsetof(struct ceph_msg_header
, crc
));
742 con
->out_msg
->hdr
.crc
= cpu_to_le32(crc
);
743 con
->out_msg
->footer
.flags
= 0;
745 crc
= crc32c(0, m
->front
.iov_base
, m
->front
.iov_len
);
746 con
->out_msg
->footer
.front_crc
= cpu_to_le32(crc
);
748 crc
= crc32c(0, m
->middle
->vec
.iov_base
,
749 m
->middle
->vec
.iov_len
);
750 con
->out_msg
->footer
.middle_crc
= cpu_to_le32(crc
);
752 con
->out_msg
->footer
.middle_crc
= 0;
753 dout("%s front_crc %u middle_crc %u\n", __func__
,
754 le32_to_cpu(con
->out_msg
->footer
.front_crc
),
755 le32_to_cpu(con
->out_msg
->footer
.middle_crc
));
757 /* is there a data payload? */
758 con
->out_msg
->footer
.data_crc
= 0;
760 prepare_write_message_data(con
);
762 /* no, queue up footer too and be done */
763 prepare_write_message_footer(con
);
765 set_bit(WRITE_PENDING
, &con
->flags
);
771 static void prepare_write_ack(struct ceph_connection
*con
)
773 dout("prepare_write_ack %p %llu -> %llu\n", con
,
774 con
->in_seq_acked
, con
->in_seq
);
775 con
->in_seq_acked
= con
->in_seq
;
777 con_out_kvec_reset(con
);
779 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
781 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
782 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
785 con
->out_more
= 1; /* more will follow.. eventually.. */
786 set_bit(WRITE_PENDING
, &con
->flags
);
790 * Prepare to write keepalive byte.
792 static void prepare_write_keepalive(struct ceph_connection
*con
)
794 dout("prepare_write_keepalive %p\n", con
);
795 con_out_kvec_reset(con
);
796 con_out_kvec_add(con
, sizeof (tag_keepalive
), &tag_keepalive
);
797 set_bit(WRITE_PENDING
, &con
->flags
);
801 * Connection negotiation.
804 static struct ceph_auth_handshake
*get_connect_authorizer(struct ceph_connection
*con
,
807 struct ceph_auth_handshake
*auth
;
809 if (!con
->ops
->get_authorizer
) {
810 con
->out_connect
.authorizer_protocol
= CEPH_AUTH_UNKNOWN
;
811 con
->out_connect
.authorizer_len
= 0;
816 /* Can't hold the mutex while getting authorizer */
818 mutex_unlock(&con
->mutex
);
820 auth
= con
->ops
->get_authorizer(con
, auth_proto
, con
->auth_retry
);
822 mutex_lock(&con
->mutex
);
826 if (test_bit(CLOSED
, &con
->state
) || test_bit(OPENING
, &con
->flags
))
827 return ERR_PTR(-EAGAIN
);
829 con
->auth_reply_buf
= auth
->authorizer_reply_buf
;
830 con
->auth_reply_buf_len
= auth
->authorizer_reply_buf_len
;
837 * We connected to a peer and are saying hello.
839 static void prepare_write_banner(struct ceph_connection
*con
)
841 con_out_kvec_add(con
, strlen(CEPH_BANNER
), CEPH_BANNER
);
842 con_out_kvec_add(con
, sizeof (con
->msgr
->my_enc_addr
),
843 &con
->msgr
->my_enc_addr
);
846 set_bit(WRITE_PENDING
, &con
->flags
);
849 static int prepare_write_connect(struct ceph_connection
*con
)
851 unsigned int global_seq
= get_global_seq(con
->msgr
, 0);
854 struct ceph_auth_handshake
*auth
;
856 switch (con
->peer_name
.type
) {
857 case CEPH_ENTITY_TYPE_MON
:
858 proto
= CEPH_MONC_PROTOCOL
;
860 case CEPH_ENTITY_TYPE_OSD
:
861 proto
= CEPH_OSDC_PROTOCOL
;
863 case CEPH_ENTITY_TYPE_MDS
:
864 proto
= CEPH_MDSC_PROTOCOL
;
870 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con
,
871 con
->connect_seq
, global_seq
, proto
);
873 con
->out_connect
.features
= cpu_to_le64(con
->msgr
->supported_features
);
874 con
->out_connect
.host_type
= cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT
);
875 con
->out_connect
.connect_seq
= cpu_to_le32(con
->connect_seq
);
876 con
->out_connect
.global_seq
= cpu_to_le32(global_seq
);
877 con
->out_connect
.protocol_version
= cpu_to_le32(proto
);
878 con
->out_connect
.flags
= 0;
880 auth_proto
= CEPH_AUTH_UNKNOWN
;
881 auth
= get_connect_authorizer(con
, &auth_proto
);
883 return PTR_ERR(auth
);
885 con
->out_connect
.authorizer_protocol
= cpu_to_le32(auth_proto
);
886 con
->out_connect
.authorizer_len
= auth
?
887 cpu_to_le32(auth
->authorizer_buf_len
) : 0;
889 con_out_kvec_reset(con
);
890 con_out_kvec_add(con
, sizeof (con
->out_connect
),
892 if (auth
&& auth
->authorizer_buf_len
)
893 con_out_kvec_add(con
, auth
->authorizer_buf_len
,
894 auth
->authorizer_buf
);
897 set_bit(WRITE_PENDING
, &con
->flags
);
903 * write as much of pending kvecs to the socket as we can.
905 * 0 -> socket full, but more to do
908 static int write_partial_kvec(struct ceph_connection
*con
)
912 dout("write_partial_kvec %p %d left\n", con
, con
->out_kvec_bytes
);
913 while (con
->out_kvec_bytes
> 0) {
914 ret
= ceph_tcp_sendmsg(con
->sock
, con
->out_kvec_cur
,
915 con
->out_kvec_left
, con
->out_kvec_bytes
,
919 con
->out_kvec_bytes
-= ret
;
920 if (con
->out_kvec_bytes
== 0)
923 /* account for full iov entries consumed */
924 while (ret
>= con
->out_kvec_cur
->iov_len
) {
925 BUG_ON(!con
->out_kvec_left
);
926 ret
-= con
->out_kvec_cur
->iov_len
;
928 con
->out_kvec_left
--;
930 /* and for a partially-consumed entry */
932 con
->out_kvec_cur
->iov_len
-= ret
;
933 con
->out_kvec_cur
->iov_base
+= ret
;
936 con
->out_kvec_left
= 0;
937 con
->out_kvec_is_msg
= false;
940 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con
,
941 con
->out_kvec_bytes
, con
->out_kvec_left
, ret
);
942 return ret
; /* done! */
945 static void out_msg_pos_next(struct ceph_connection
*con
, struct page
*page
,
946 size_t len
, size_t sent
, bool in_trail
)
948 struct ceph_msg
*msg
= con
->out_msg
;
953 con
->out_msg_pos
.data_pos
+= sent
;
954 con
->out_msg_pos
.page_pos
+= sent
;
959 con
->out_msg_pos
.page_pos
= 0;
960 con
->out_msg_pos
.page
++;
961 con
->out_msg_pos
.did_page_crc
= false;
963 list_move_tail(&page
->lru
,
965 else if (msg
->pagelist
)
966 list_move_tail(&page
->lru
,
967 &msg
->pagelist
->head
);
970 iter_bio_next(&msg
->bio_iter
, &msg
->bio_seg
);
975 * Write as much message data payload as we can. If we finish, queue
977 * 1 -> done, footer is now queued in out_kvec[].
978 * 0 -> socket full, but more to do
981 static int write_partial_msg_pages(struct ceph_connection
*con
)
983 struct ceph_msg
*msg
= con
->out_msg
;
984 unsigned int data_len
= le32_to_cpu(msg
->hdr
.data_len
);
986 bool do_datacrc
= !con
->msgr
->nocrc
;
989 bool in_trail
= false;
990 const size_t trail_len
= (msg
->trail
? msg
->trail
->length
: 0);
991 const size_t trail_off
= data_len
- trail_len
;
993 dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
994 con
, msg
, con
->out_msg_pos
.page
, msg
->nr_pages
,
995 con
->out_msg_pos
.page_pos
);
998 * Iterate through each page that contains data to be
999 * written, and send as much as possible for each.
1001 * If we are calculating the data crc (the default), we will
1002 * need to map the page. If we have no pages, they have
1003 * been revoked, so use the zero page.
1005 while (data_len
> con
->out_msg_pos
.data_pos
) {
1006 struct page
*page
= NULL
;
1007 int max_write
= PAGE_SIZE
;
1010 in_trail
= in_trail
|| con
->out_msg_pos
.data_pos
>= trail_off
;
1012 total_max_write
= trail_off
- con
->out_msg_pos
.data_pos
;
1015 total_max_write
= data_len
- con
->out_msg_pos
.data_pos
;
1017 page
= list_first_entry(&msg
->trail
->head
,
1019 } else if (msg
->pages
) {
1020 page
= msg
->pages
[con
->out_msg_pos
.page
];
1021 } else if (msg
->pagelist
) {
1022 page
= list_first_entry(&msg
->pagelist
->head
,
1025 } else if (msg
->bio
) {
1028 bv
= bio_iovec_idx(msg
->bio_iter
, msg
->bio_seg
);
1030 bio_offset
= bv
->bv_offset
;
1031 max_write
= bv
->bv_len
;
1036 len
= min_t(int, max_write
- con
->out_msg_pos
.page_pos
,
1039 if (do_datacrc
&& !con
->out_msg_pos
.did_page_crc
) {
1041 u32 crc
= le32_to_cpu(msg
->footer
.data_crc
);
1045 BUG_ON(kaddr
== NULL
);
1046 base
= kaddr
+ con
->out_msg_pos
.page_pos
+ bio_offset
;
1047 crc
= crc32c(crc
, base
, len
);
1048 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1049 con
->out_msg_pos
.did_page_crc
= true;
1051 ret
= ceph_tcp_sendpage(con
->sock
, page
,
1052 con
->out_msg_pos
.page_pos
+ bio_offset
,
1061 out_msg_pos_next(con
, page
, len
, (size_t) ret
, in_trail
);
1064 dout("write_partial_msg_pages %p msg %p done\n", con
, msg
);
1066 /* prepare and queue up footer, too */
1068 msg
->footer
.flags
|= CEPH_MSG_FOOTER_NOCRC
;
1069 con_out_kvec_reset(con
);
1070 prepare_write_message_footer(con
);
1079 static int write_partial_skip(struct ceph_connection
*con
)
1083 while (con
->out_skip
> 0) {
1084 size_t size
= min(con
->out_skip
, (int) PAGE_CACHE_SIZE
);
1086 ret
= ceph_tcp_sendpage(con
->sock
, zero_page
, 0, size
, 1);
1089 con
->out_skip
-= ret
;
1097 * Prepare to read connection handshake, or an ack.
1099 static void prepare_read_banner(struct ceph_connection
*con
)
1101 dout("prepare_read_banner %p\n", con
);
1102 con
->in_base_pos
= 0;
1105 static void prepare_read_connect(struct ceph_connection
*con
)
1107 dout("prepare_read_connect %p\n", con
);
1108 con
->in_base_pos
= 0;
1111 static void prepare_read_ack(struct ceph_connection
*con
)
1113 dout("prepare_read_ack %p\n", con
);
1114 con
->in_base_pos
= 0;
1117 static void prepare_read_tag(struct ceph_connection
*con
)
1119 dout("prepare_read_tag %p\n", con
);
1120 con
->in_base_pos
= 0;
1121 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1125 * Prepare to read a message.
1127 static int prepare_read_message(struct ceph_connection
*con
)
1129 dout("prepare_read_message %p\n", con
);
1130 BUG_ON(con
->in_msg
!= NULL
);
1131 con
->in_base_pos
= 0;
1132 con
->in_front_crc
= con
->in_middle_crc
= con
->in_data_crc
= 0;
1137 static int read_partial(struct ceph_connection
*con
,
1138 int end
, int size
, void *object
)
1140 while (con
->in_base_pos
< end
) {
1141 int left
= end
- con
->in_base_pos
;
1142 int have
= size
- left
;
1143 int ret
= ceph_tcp_recvmsg(con
->sock
, object
+ have
, left
);
1146 con
->in_base_pos
+= ret
;
1153 * Read all or part of the connect-side handshake on a new connection
1155 static int read_partial_banner(struct ceph_connection
*con
)
1161 dout("read_partial_banner %p at %d\n", con
, con
->in_base_pos
);
1164 size
= strlen(CEPH_BANNER
);
1166 ret
= read_partial(con
, end
, size
, con
->in_banner
);
1170 size
= sizeof (con
->actual_peer_addr
);
1172 ret
= read_partial(con
, end
, size
, &con
->actual_peer_addr
);
1176 size
= sizeof (con
->peer_addr_for_me
);
1178 ret
= read_partial(con
, end
, size
, &con
->peer_addr_for_me
);
1186 static int read_partial_connect(struct ceph_connection
*con
)
1192 dout("read_partial_connect %p at %d\n", con
, con
->in_base_pos
);
1194 size
= sizeof (con
->in_reply
);
1196 ret
= read_partial(con
, end
, size
, &con
->in_reply
);
1200 size
= le32_to_cpu(con
->in_reply
.authorizer_len
);
1202 ret
= read_partial(con
, end
, size
, con
->auth_reply_buf
);
1206 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1207 con
, (int)con
->in_reply
.tag
,
1208 le32_to_cpu(con
->in_reply
.connect_seq
),
1209 le32_to_cpu(con
->in_reply
.global_seq
));
1216 * Verify the hello banner looks okay.
1218 static int verify_hello(struct ceph_connection
*con
)
1220 if (memcmp(con
->in_banner
, CEPH_BANNER
, strlen(CEPH_BANNER
))) {
1221 pr_err("connect to %s got bad banner\n",
1222 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1223 con
->error_msg
= "protocol error, bad banner";
1229 static bool addr_is_blank(struct sockaddr_storage
*ss
)
1231 switch (ss
->ss_family
) {
1233 return ((struct sockaddr_in
*)ss
)->sin_addr
.s_addr
== 0;
1236 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[0] == 0 &&
1237 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[1] == 0 &&
1238 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[2] == 0 &&
1239 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[3] == 0;
1244 static int addr_port(struct sockaddr_storage
*ss
)
1246 switch (ss
->ss_family
) {
1248 return ntohs(((struct sockaddr_in
*)ss
)->sin_port
);
1250 return ntohs(((struct sockaddr_in6
*)ss
)->sin6_port
);
1255 static void addr_set_port(struct sockaddr_storage
*ss
, int p
)
1257 switch (ss
->ss_family
) {
1259 ((struct sockaddr_in
*)ss
)->sin_port
= htons(p
);
1262 ((struct sockaddr_in6
*)ss
)->sin6_port
= htons(p
);
1268 * Unlike other *_pton function semantics, zero indicates success.
1270 static int ceph_pton(const char *str
, size_t len
, struct sockaddr_storage
*ss
,
1271 char delim
, const char **ipend
)
1273 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
1274 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
1276 memset(ss
, 0, sizeof(*ss
));
1278 if (in4_pton(str
, len
, (u8
*)&in4
->sin_addr
.s_addr
, delim
, ipend
)) {
1279 ss
->ss_family
= AF_INET
;
1283 if (in6_pton(str
, len
, (u8
*)&in6
->sin6_addr
.s6_addr
, delim
, ipend
)) {
1284 ss
->ss_family
= AF_INET6
;
1292 * Extract hostname string and resolve using kernel DNS facility.
1294 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1295 static int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1296 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1298 const char *end
, *delim_p
;
1299 char *colon_p
, *ip_addr
= NULL
;
1303 * The end of the hostname occurs immediately preceding the delimiter or
1304 * the port marker (':') where the delimiter takes precedence.
1306 delim_p
= memchr(name
, delim
, namelen
);
1307 colon_p
= memchr(name
, ':', namelen
);
1309 if (delim_p
&& colon_p
)
1310 end
= delim_p
< colon_p
? delim_p
: colon_p
;
1311 else if (!delim_p
&& colon_p
)
1315 if (!end
) /* case: hostname:/ */
1316 end
= name
+ namelen
;
1322 /* do dns_resolve upcall */
1323 ip_len
= dns_query(NULL
, name
, end
- name
, NULL
, &ip_addr
, NULL
);
1325 ret
= ceph_pton(ip_addr
, ip_len
, ss
, -1, NULL
);
1333 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end
- name
), name
,
1334 ret
, ret
? "failed" : ceph_pr_addr(ss
));
1339 static inline int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1340 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1347 * Parse a server name (IP or hostname). If a valid IP address is not found
1348 * then try to extract a hostname to resolve using userspace DNS upcall.
1350 static int ceph_parse_server_name(const char *name
, size_t namelen
,
1351 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1355 ret
= ceph_pton(name
, namelen
, ss
, delim
, ipend
);
1357 ret
= ceph_dns_resolve_name(name
, namelen
, ss
, delim
, ipend
);
1363 * Parse an ip[:port] list into an addr array. Use the default
1364 * monitor port if a port isn't specified.
1366 int ceph_parse_ips(const char *c
, const char *end
,
1367 struct ceph_entity_addr
*addr
,
1368 int max_count
, int *count
)
1370 int i
, ret
= -EINVAL
;
1373 dout("parse_ips on '%.*s'\n", (int)(end
-c
), c
);
1374 for (i
= 0; i
< max_count
; i
++) {
1376 struct sockaddr_storage
*ss
= &addr
[i
].in_addr
;
1385 ret
= ceph_parse_server_name(p
, end
- p
, ss
, delim
, &ipend
);
1394 dout("missing matching ']'\n");
1401 if (p
< end
&& *p
== ':') {
1404 while (p
< end
&& *p
>= '0' && *p
<= '9') {
1405 port
= (port
* 10) + (*p
- '0');
1408 if (port
> 65535 || port
== 0)
1411 port
= CEPH_MON_PORT
;
1414 addr_set_port(ss
, port
);
1416 dout("parse_ips got %s\n", ceph_pr_addr(ss
));
1433 pr_err("parse_ips bad ip '%.*s'\n", (int)(end
- c
), c
);
1436 EXPORT_SYMBOL(ceph_parse_ips
);
1438 static int process_banner(struct ceph_connection
*con
)
1440 dout("process_banner on %p\n", con
);
1442 if (verify_hello(con
) < 0)
1445 ceph_decode_addr(&con
->actual_peer_addr
);
1446 ceph_decode_addr(&con
->peer_addr_for_me
);
1449 * Make sure the other end is who we wanted. note that the other
1450 * end may not yet know their ip address, so if it's 0.0.0.0, give
1451 * them the benefit of the doubt.
1453 if (memcmp(&con
->peer_addr
, &con
->actual_peer_addr
,
1454 sizeof(con
->peer_addr
)) != 0 &&
1455 !(addr_is_blank(&con
->actual_peer_addr
.in_addr
) &&
1456 con
->actual_peer_addr
.nonce
== con
->peer_addr
.nonce
)) {
1457 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1458 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1459 (int)le32_to_cpu(con
->peer_addr
.nonce
),
1460 ceph_pr_addr(&con
->actual_peer_addr
.in_addr
),
1461 (int)le32_to_cpu(con
->actual_peer_addr
.nonce
));
1462 con
->error_msg
= "wrong peer at address";
1467 * did we learn our address?
1469 if (addr_is_blank(&con
->msgr
->inst
.addr
.in_addr
)) {
1470 int port
= addr_port(&con
->msgr
->inst
.addr
.in_addr
);
1472 memcpy(&con
->msgr
->inst
.addr
.in_addr
,
1473 &con
->peer_addr_for_me
.in_addr
,
1474 sizeof(con
->peer_addr_for_me
.in_addr
));
1475 addr_set_port(&con
->msgr
->inst
.addr
.in_addr
, port
);
1476 encode_my_addr(con
->msgr
);
1477 dout("process_banner learned my addr is %s\n",
1478 ceph_pr_addr(&con
->msgr
->inst
.addr
.in_addr
));
1484 static void fail_protocol(struct ceph_connection
*con
)
1486 reset_connection(con
);
1487 set_bit(CLOSED
, &con
->state
); /* in case there's queued work */
1490 static int process_connect(struct ceph_connection
*con
)
1492 u64 sup_feat
= con
->msgr
->supported_features
;
1493 u64 req_feat
= con
->msgr
->required_features
;
1494 u64 server_feat
= le64_to_cpu(con
->in_reply
.features
);
1497 dout("process_connect on %p tag %d\n", con
, (int)con
->in_tag
);
1499 switch (con
->in_reply
.tag
) {
1500 case CEPH_MSGR_TAG_FEATURES
:
1501 pr_err("%s%lld %s feature set mismatch,"
1502 " my %llx < server's %llx, missing %llx\n",
1503 ENTITY_NAME(con
->peer_name
),
1504 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1505 sup_feat
, server_feat
, server_feat
& ~sup_feat
);
1506 con
->error_msg
= "missing required protocol features";
1510 case CEPH_MSGR_TAG_BADPROTOVER
:
1511 pr_err("%s%lld %s protocol version mismatch,"
1512 " my %d != server's %d\n",
1513 ENTITY_NAME(con
->peer_name
),
1514 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1515 le32_to_cpu(con
->out_connect
.protocol_version
),
1516 le32_to_cpu(con
->in_reply
.protocol_version
));
1517 con
->error_msg
= "protocol version mismatch";
1521 case CEPH_MSGR_TAG_BADAUTHORIZER
:
1523 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con
,
1525 if (con
->auth_retry
== 2) {
1526 con
->error_msg
= "connect authorization failure";
1529 con
->auth_retry
= 1;
1530 ret
= prepare_write_connect(con
);
1533 prepare_read_connect(con
);
1536 case CEPH_MSGR_TAG_RESETSESSION
:
1538 * If we connected with a large connect_seq but the peer
1539 * has no record of a session with us (no connection, or
1540 * connect_seq == 0), they will send RESETSESION to indicate
1541 * that they must have reset their session, and may have
1544 dout("process_connect got RESET peer seq %u\n",
1545 le32_to_cpu(con
->in_reply
.connect_seq
));
1546 pr_err("%s%lld %s connection reset\n",
1547 ENTITY_NAME(con
->peer_name
),
1548 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1549 reset_connection(con
);
1550 ret
= prepare_write_connect(con
);
1553 prepare_read_connect(con
);
1555 /* Tell ceph about it. */
1556 mutex_unlock(&con
->mutex
);
1557 pr_info("reset on %s%lld\n", ENTITY_NAME(con
->peer_name
));
1558 if (con
->ops
->peer_reset
)
1559 con
->ops
->peer_reset(con
);
1560 mutex_lock(&con
->mutex
);
1561 if (test_bit(CLOSED
, &con
->state
) ||
1562 test_bit(OPENING
, &con
->state
))
1566 case CEPH_MSGR_TAG_RETRY_SESSION
:
1568 * If we sent a smaller connect_seq than the peer has, try
1569 * again with a larger value.
1571 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
1572 le32_to_cpu(con
->out_connect
.connect_seq
),
1573 le32_to_cpu(con
->in_reply
.connect_seq
));
1574 con
->connect_seq
= le32_to_cpu(con
->in_reply
.connect_seq
);
1575 ret
= prepare_write_connect(con
);
1578 prepare_read_connect(con
);
1581 case CEPH_MSGR_TAG_RETRY_GLOBAL
:
1583 * If we sent a smaller global_seq than the peer has, try
1584 * again with a larger value.
1586 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1587 con
->peer_global_seq
,
1588 le32_to_cpu(con
->in_reply
.global_seq
));
1589 get_global_seq(con
->msgr
,
1590 le32_to_cpu(con
->in_reply
.global_seq
));
1591 ret
= prepare_write_connect(con
);
1594 prepare_read_connect(con
);
1597 case CEPH_MSGR_TAG_READY
:
1598 if (req_feat
& ~server_feat
) {
1599 pr_err("%s%lld %s protocol feature mismatch,"
1600 " my required %llx > server's %llx, need %llx\n",
1601 ENTITY_NAME(con
->peer_name
),
1602 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1603 req_feat
, server_feat
, req_feat
& ~server_feat
);
1604 con
->error_msg
= "missing required protocol features";
1608 clear_bit(NEGOTIATING
, &con
->state
);
1609 set_bit(CONNECTED
, &con
->state
);
1610 con
->peer_global_seq
= le32_to_cpu(con
->in_reply
.global_seq
);
1612 con
->peer_features
= server_feat
;
1613 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1614 con
->peer_global_seq
,
1615 le32_to_cpu(con
->in_reply
.connect_seq
),
1617 WARN_ON(con
->connect_seq
!=
1618 le32_to_cpu(con
->in_reply
.connect_seq
));
1620 if (con
->in_reply
.flags
& CEPH_MSG_CONNECT_LOSSY
)
1621 set_bit(LOSSYTX
, &con
->flags
);
1623 con
->delay
= 0; /* reset backoff memory */
1625 prepare_read_tag(con
);
1628 case CEPH_MSGR_TAG_WAIT
:
1630 * If there is a connection race (we are opening
1631 * connections to each other), one of us may just have
1632 * to WAIT. This shouldn't happen if we are the
1635 pr_err("process_connect got WAIT as client\n");
1636 con
->error_msg
= "protocol error, got WAIT as client";
1640 pr_err("connect protocol error, will retry\n");
1641 con
->error_msg
= "protocol error, garbage tag during connect";
1649 * read (part of) an ack
1651 static int read_partial_ack(struct ceph_connection
*con
)
1653 int size
= sizeof (con
->in_temp_ack
);
1656 return read_partial(con
, end
, size
, &con
->in_temp_ack
);
1661 * We can finally discard anything that's been acked.
1663 static void process_ack(struct ceph_connection
*con
)
1666 u64 ack
= le64_to_cpu(con
->in_temp_ack
);
1669 while (!list_empty(&con
->out_sent
)) {
1670 m
= list_first_entry(&con
->out_sent
, struct ceph_msg
,
1672 seq
= le64_to_cpu(m
->hdr
.seq
);
1675 dout("got ack for seq %llu type %d at %p\n", seq
,
1676 le16_to_cpu(m
->hdr
.type
), m
);
1677 m
->ack_stamp
= jiffies
;
1680 prepare_read_tag(con
);
1686 static int read_partial_message_section(struct ceph_connection
*con
,
1687 struct kvec
*section
,
1688 unsigned int sec_len
, u32
*crc
)
1694 while (section
->iov_len
< sec_len
) {
1695 BUG_ON(section
->iov_base
== NULL
);
1696 left
= sec_len
- section
->iov_len
;
1697 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)section
->iov_base
+
1698 section
->iov_len
, left
);
1701 section
->iov_len
+= ret
;
1703 if (section
->iov_len
== sec_len
)
1704 *crc
= crc32c(0, section
->iov_base
, section
->iov_len
);
1709 static bool ceph_con_in_msg_alloc(struct ceph_connection
*con
,
1710 struct ceph_msg_header
*hdr
);
1713 static int read_partial_message_pages(struct ceph_connection
*con
,
1714 struct page
**pages
,
1715 unsigned int data_len
, bool do_datacrc
)
1721 left
= min((int)(data_len
- con
->in_msg_pos
.data_pos
),
1722 (int)(PAGE_SIZE
- con
->in_msg_pos
.page_pos
));
1724 BUG_ON(pages
== NULL
);
1725 p
= kmap(pages
[con
->in_msg_pos
.page
]);
1726 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ con
->in_msg_pos
.page_pos
,
1728 if (ret
> 0 && do_datacrc
)
1730 crc32c(con
->in_data_crc
,
1731 p
+ con
->in_msg_pos
.page_pos
, ret
);
1732 kunmap(pages
[con
->in_msg_pos
.page
]);
1735 con
->in_msg_pos
.data_pos
+= ret
;
1736 con
->in_msg_pos
.page_pos
+= ret
;
1737 if (con
->in_msg_pos
.page_pos
== PAGE_SIZE
) {
1738 con
->in_msg_pos
.page_pos
= 0;
1739 con
->in_msg_pos
.page
++;
1746 static int read_partial_message_bio(struct ceph_connection
*con
,
1747 struct bio
**bio_iter
, int *bio_seg
,
1748 unsigned int data_len
, bool do_datacrc
)
1750 struct bio_vec
*bv
= bio_iovec_idx(*bio_iter
, *bio_seg
);
1754 left
= min((int)(data_len
- con
->in_msg_pos
.data_pos
),
1755 (int)(bv
->bv_len
- con
->in_msg_pos
.page_pos
));
1757 p
= kmap(bv
->bv_page
) + bv
->bv_offset
;
1759 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ con
->in_msg_pos
.page_pos
,
1761 if (ret
> 0 && do_datacrc
)
1763 crc32c(con
->in_data_crc
,
1764 p
+ con
->in_msg_pos
.page_pos
, ret
);
1765 kunmap(bv
->bv_page
);
1768 con
->in_msg_pos
.data_pos
+= ret
;
1769 con
->in_msg_pos
.page_pos
+= ret
;
1770 if (con
->in_msg_pos
.page_pos
== bv
->bv_len
) {
1771 con
->in_msg_pos
.page_pos
= 0;
1772 iter_bio_next(bio_iter
, bio_seg
);
1780 * read (part of) a message.
1782 static int read_partial_message(struct ceph_connection
*con
)
1784 struct ceph_msg
*m
= con
->in_msg
;
1788 unsigned int front_len
, middle_len
, data_len
;
1789 bool do_datacrc
= !con
->msgr
->nocrc
;
1793 dout("read_partial_message con %p msg %p\n", con
, m
);
1796 size
= sizeof (con
->in_hdr
);
1798 ret
= read_partial(con
, end
, size
, &con
->in_hdr
);
1802 crc
= crc32c(0, &con
->in_hdr
, offsetof(struct ceph_msg_header
, crc
));
1803 if (cpu_to_le32(crc
) != con
->in_hdr
.crc
) {
1804 pr_err("read_partial_message bad hdr "
1805 " crc %u != expected %u\n",
1806 crc
, con
->in_hdr
.crc
);
1810 front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
1811 if (front_len
> CEPH_MSG_MAX_FRONT_LEN
)
1813 middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
1814 if (middle_len
> CEPH_MSG_MAX_DATA_LEN
)
1816 data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
1817 if (data_len
> CEPH_MSG_MAX_DATA_LEN
)
1821 seq
= le64_to_cpu(con
->in_hdr
.seq
);
1822 if ((s64
)seq
- (s64
)con
->in_seq
< 1) {
1823 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1824 ENTITY_NAME(con
->peer_name
),
1825 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1826 seq
, con
->in_seq
+ 1);
1827 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1829 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1831 } else if ((s64
)seq
- (s64
)con
->in_seq
> 1) {
1832 pr_err("read_partial_message bad seq %lld expected %lld\n",
1833 seq
, con
->in_seq
+ 1);
1834 con
->error_msg
= "bad message sequence # for incoming message";
1838 /* allocate message? */
1840 dout("got hdr type %d front %d data %d\n", con
->in_hdr
.type
,
1841 con
->in_hdr
.front_len
, con
->in_hdr
.data_len
);
1842 if (ceph_con_in_msg_alloc(con
, &con
->in_hdr
)) {
1843 /* skip this message */
1844 dout("alloc_msg said skip message\n");
1845 BUG_ON(con
->in_msg
);
1846 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1848 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1854 "error allocating memory for incoming message";
1858 BUG_ON(con
->in_msg
->con
!= con
);
1860 m
->front
.iov_len
= 0; /* haven't read it yet */
1862 m
->middle
->vec
.iov_len
= 0;
1864 con
->in_msg_pos
.page
= 0;
1866 con
->in_msg_pos
.page_pos
= m
->page_alignment
;
1868 con
->in_msg_pos
.page_pos
= 0;
1869 con
->in_msg_pos
.data_pos
= 0;
1873 init_bio_iter(m
->bio
, &m
->bio_iter
, &m
->bio_seg
);
1878 ret
= read_partial_message_section(con
, &m
->front
, front_len
,
1879 &con
->in_front_crc
);
1885 ret
= read_partial_message_section(con
, &m
->middle
->vec
,
1887 &con
->in_middle_crc
);
1893 while (con
->in_msg_pos
.data_pos
< data_len
) {
1895 ret
= read_partial_message_pages(con
, m
->pages
,
1896 data_len
, do_datacrc
);
1900 } else if (m
->bio
) {
1901 BUG_ON(!m
->bio_iter
);
1902 ret
= read_partial_message_bio(con
,
1903 &m
->bio_iter
, &m
->bio_seg
,
1904 data_len
, do_datacrc
);
1914 size
= sizeof (m
->footer
);
1916 ret
= read_partial(con
, end
, size
, &m
->footer
);
1920 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1921 m
, front_len
, m
->footer
.front_crc
, middle_len
,
1922 m
->footer
.middle_crc
, data_len
, m
->footer
.data_crc
);
1925 if (con
->in_front_crc
!= le32_to_cpu(m
->footer
.front_crc
)) {
1926 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1927 m
, con
->in_front_crc
, m
->footer
.front_crc
);
1930 if (con
->in_middle_crc
!= le32_to_cpu(m
->footer
.middle_crc
)) {
1931 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1932 m
, con
->in_middle_crc
, m
->footer
.middle_crc
);
1936 (m
->footer
.flags
& CEPH_MSG_FOOTER_NOCRC
) == 0 &&
1937 con
->in_data_crc
!= le32_to_cpu(m
->footer
.data_crc
)) {
1938 pr_err("read_partial_message %p data crc %u != exp. %u\n", m
,
1939 con
->in_data_crc
, le32_to_cpu(m
->footer
.data_crc
));
1943 return 1; /* done! */
1947 * Process message. This happens in the worker thread. The callback should
1948 * be careful not to do anything that waits on other incoming messages or it
1951 static void process_message(struct ceph_connection
*con
)
1953 struct ceph_msg
*msg
;
1955 BUG_ON(con
->in_msg
->con
!= con
);
1956 con
->in_msg
->con
= NULL
;
1961 /* if first message, set peer_name */
1962 if (con
->peer_name
.type
== 0)
1963 con
->peer_name
= msg
->hdr
.src
;
1966 mutex_unlock(&con
->mutex
);
1968 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
1969 msg
, le64_to_cpu(msg
->hdr
.seq
),
1970 ENTITY_NAME(msg
->hdr
.src
),
1971 le16_to_cpu(msg
->hdr
.type
),
1972 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
1973 le32_to_cpu(msg
->hdr
.front_len
),
1974 le32_to_cpu(msg
->hdr
.data_len
),
1975 con
->in_front_crc
, con
->in_middle_crc
, con
->in_data_crc
);
1976 con
->ops
->dispatch(con
, msg
);
1978 mutex_lock(&con
->mutex
);
1979 prepare_read_tag(con
);
1984 * Write something to the socket. Called in a worker thread when the
1985 * socket appears to be writeable and we have something ready to send.
1987 static int try_write(struct ceph_connection
*con
)
1991 dout("try_write start %p state %lu\n", con
, con
->state
);
1994 dout("try_write out_kvec_bytes %d\n", con
->out_kvec_bytes
);
1996 /* open the socket first? */
1997 if (con
->sock
== NULL
) {
1998 set_bit(CONNECTING
, &con
->state
);
2000 con_out_kvec_reset(con
);
2001 prepare_write_banner(con
);
2002 prepare_read_banner(con
);
2004 BUG_ON(con
->in_msg
);
2005 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2006 dout("try_write initiating connect on %p new state %lu\n",
2008 ret
= ceph_tcp_connect(con
);
2010 con
->error_msg
= "connect error";
2016 /* kvec data queued? */
2017 if (con
->out_skip
) {
2018 ret
= write_partial_skip(con
);
2022 if (con
->out_kvec_left
) {
2023 ret
= write_partial_kvec(con
);
2030 if (con
->out_msg_done
) {
2031 ceph_msg_put(con
->out_msg
);
2032 con
->out_msg
= NULL
; /* we're done with this one */
2036 ret
= write_partial_msg_pages(con
);
2038 goto more_kvec
; /* we need to send the footer, too! */
2042 dout("try_write write_partial_msg_pages err %d\n",
2049 if (!test_bit(CONNECTING
, &con
->state
) &&
2050 !test_bit(NEGOTIATING
, &con
->state
)) {
2051 /* is anything else pending? */
2052 if (!list_empty(&con
->out_queue
)) {
2053 prepare_write_message(con
);
2056 if (con
->in_seq
> con
->in_seq_acked
) {
2057 prepare_write_ack(con
);
2060 if (test_and_clear_bit(KEEPALIVE_PENDING
, &con
->flags
)) {
2061 prepare_write_keepalive(con
);
2066 /* Nothing to do! */
2067 clear_bit(WRITE_PENDING
, &con
->flags
);
2068 dout("try_write nothing else to write.\n");
2071 dout("try_write done on %p ret %d\n", con
, ret
);
2078 * Read what we can from the socket.
2080 static int try_read(struct ceph_connection
*con
)
2087 if (test_bit(STANDBY
, &con
->state
))
2090 dout("try_read start on %p\n", con
);
2093 dout("try_read tag %d in_base_pos %d\n", (int)con
->in_tag
,
2097 * process_connect and process_message drop and re-take
2098 * con->mutex. make sure we handle a racing close or reopen.
2100 if (test_bit(CLOSED
, &con
->state
) ||
2101 test_bit(OPENING
, &con
->state
)) {
2106 if (test_bit(CONNECTING
, &con
->state
)) {
2107 dout("try_read connecting\n");
2108 ret
= read_partial_banner(con
);
2111 ret
= process_banner(con
);
2115 clear_bit(CONNECTING
, &con
->state
);
2116 set_bit(NEGOTIATING
, &con
->state
);
2118 /* Banner is good, exchange connection info */
2119 ret
= prepare_write_connect(con
);
2122 prepare_read_connect(con
);
2124 /* Send connection info before awaiting response */
2128 if (test_bit(NEGOTIATING
, &con
->state
)) {
2129 dout("try_read negotiating\n");
2130 ret
= read_partial_connect(con
);
2133 ret
= process_connect(con
);
2139 if (con
->in_base_pos
< 0) {
2141 * skipping + discarding content.
2143 * FIXME: there must be a better way to do this!
2145 static char buf
[SKIP_BUF_SIZE
];
2146 int skip
= min((int) sizeof (buf
), -con
->in_base_pos
);
2148 dout("skipping %d / %d bytes\n", skip
, -con
->in_base_pos
);
2149 ret
= ceph_tcp_recvmsg(con
->sock
, buf
, skip
);
2152 con
->in_base_pos
+= ret
;
2153 if (con
->in_base_pos
)
2156 if (con
->in_tag
== CEPH_MSGR_TAG_READY
) {
2160 ret
= ceph_tcp_recvmsg(con
->sock
, &con
->in_tag
, 1);
2163 dout("try_read got tag %d\n", (int)con
->in_tag
);
2164 switch (con
->in_tag
) {
2165 case CEPH_MSGR_TAG_MSG
:
2166 prepare_read_message(con
);
2168 case CEPH_MSGR_TAG_ACK
:
2169 prepare_read_ack(con
);
2171 case CEPH_MSGR_TAG_CLOSE
:
2172 clear_bit(CONNECTED
, &con
->state
);
2173 set_bit(CLOSED
, &con
->state
); /* fixme */
2179 if (con
->in_tag
== CEPH_MSGR_TAG_MSG
) {
2180 ret
= read_partial_message(con
);
2184 con
->error_msg
= "bad crc";
2188 con
->error_msg
= "io error";
2193 if (con
->in_tag
== CEPH_MSGR_TAG_READY
)
2195 process_message(con
);
2198 if (con
->in_tag
== CEPH_MSGR_TAG_ACK
) {
2199 ret
= read_partial_ack(con
);
2207 dout("try_read done on %p ret %d\n", con
, ret
);
2211 pr_err("try_read bad con->in_tag = %d\n", (int)con
->in_tag
);
2212 con
->error_msg
= "protocol error, garbage tag";
2219 * Atomically queue work on a connection. Bump @con reference to
2220 * avoid races with connection teardown.
2222 static void queue_con(struct ceph_connection
*con
)
2224 if (!con
->ops
->get(con
)) {
2225 dout("queue_con %p ref count 0\n", con
);
2229 if (!queue_delayed_work(ceph_msgr_wq
, &con
->work
, 0)) {
2230 dout("queue_con %p - already queued\n", con
);
2233 dout("queue_con %p\n", con
);
2238 * Do some work on a connection. Drop a connection ref when we're done.
2240 static void con_work(struct work_struct
*work
)
2242 struct ceph_connection
*con
= container_of(work
, struct ceph_connection
,
2246 mutex_lock(&con
->mutex
);
2248 if (test_and_clear_bit(SOCK_CLOSED
, &con
->flags
)) {
2249 if (test_and_clear_bit(CONNECTED
, &con
->state
))
2250 con
->error_msg
= "socket closed";
2251 else if (test_and_clear_bit(NEGOTIATING
, &con
->state
))
2252 con
->error_msg
= "negotiation failed";
2253 else if (test_and_clear_bit(CONNECTING
, &con
->state
))
2254 con
->error_msg
= "connection failed";
2256 con
->error_msg
= "unrecognized con state";
2260 if (test_and_clear_bit(BACKOFF
, &con
->flags
)) {
2261 dout("con_work %p backing off\n", con
);
2262 if (queue_delayed_work(ceph_msgr_wq
, &con
->work
,
2263 round_jiffies_relative(con
->delay
))) {
2264 dout("con_work %p backoff %lu\n", con
, con
->delay
);
2265 mutex_unlock(&con
->mutex
);
2269 dout("con_work %p FAILED to back off %lu\n", con
,
2274 if (test_bit(STANDBY
, &con
->state
)) {
2275 dout("con_work %p STANDBY\n", con
);
2278 if (test_bit(CLOSED
, &con
->state
)) {
2279 dout("con_work %p CLOSED\n", con
);
2283 if (test_and_clear_bit(OPENING
, &con
->state
)) {
2284 /* reopen w/ new peer */
2285 dout("con_work OPENING\n");
2289 ret
= try_read(con
);
2293 con
->error_msg
= "socket error on read";
2297 ret
= try_write(con
);
2301 con
->error_msg
= "socket error on write";
2306 mutex_unlock(&con
->mutex
);
2312 mutex_unlock(&con
->mutex
);
2313 ceph_fault(con
); /* error/fault path */
2319 * Generic error/fault handler. A retry mechanism is used with
2320 * exponential backoff
2322 static void ceph_fault(struct ceph_connection
*con
)
2324 mutex_lock(&con
->mutex
);
2326 pr_err("%s%lld %s %s\n", ENTITY_NAME(con
->peer_name
),
2327 ceph_pr_addr(&con
->peer_addr
.in_addr
), con
->error_msg
);
2328 dout("fault %p state %lu to peer %s\n",
2329 con
, con
->state
, ceph_pr_addr(&con
->peer_addr
.in_addr
));
2331 if (test_bit(CLOSED
, &con
->state
))
2334 con_close_socket(con
);
2336 if (test_bit(LOSSYTX
, &con
->flags
)) {
2337 dout("fault on LOSSYTX channel\n");
2342 BUG_ON(con
->in_msg
->con
!= con
);
2343 con
->in_msg
->con
= NULL
;
2344 ceph_msg_put(con
->in_msg
);
2349 /* Requeue anything that hasn't been acked */
2350 list_splice_init(&con
->out_sent
, &con
->out_queue
);
2352 /* If there are no messages queued or keepalive pending, place
2353 * the connection in a STANDBY state */
2354 if (list_empty(&con
->out_queue
) &&
2355 !test_bit(KEEPALIVE_PENDING
, &con
->flags
)) {
2356 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con
);
2357 clear_bit(WRITE_PENDING
, &con
->flags
);
2358 set_bit(STANDBY
, &con
->state
);
2360 /* retry after a delay. */
2361 if (con
->delay
== 0)
2362 con
->delay
= BASE_DELAY_INTERVAL
;
2363 else if (con
->delay
< MAX_DELAY_INTERVAL
)
2366 if (queue_delayed_work(ceph_msgr_wq
, &con
->work
,
2367 round_jiffies_relative(con
->delay
))) {
2368 dout("fault queued %p delay %lu\n", con
, con
->delay
);
2371 dout("fault failed to queue %p delay %lu, backoff\n",
2374 * In many cases we see a socket state change
2375 * while con_work is running and end up
2376 * queuing (non-delayed) work, such that we
2377 * can't backoff with a delay. Set a flag so
2378 * that when con_work restarts we schedule the
2381 set_bit(BACKOFF
, &con
->flags
);
2386 mutex_unlock(&con
->mutex
);
2388 * in case we faulted due to authentication, invalidate our
2389 * current tickets so that we can get new ones.
2391 if (con
->auth_retry
&& con
->ops
->invalidate_authorizer
) {
2392 dout("calling invalidate_authorizer()\n");
2393 con
->ops
->invalidate_authorizer(con
);
2396 if (con
->ops
->fault
)
2397 con
->ops
->fault(con
);
2403 * initialize a new messenger instance
2405 void ceph_messenger_init(struct ceph_messenger
*msgr
,
2406 struct ceph_entity_addr
*myaddr
,
2407 u32 supported_features
,
2408 u32 required_features
,
2411 msgr
->supported_features
= supported_features
;
2412 msgr
->required_features
= required_features
;
2414 spin_lock_init(&msgr
->global_seq_lock
);
2417 msgr
->inst
.addr
= *myaddr
;
2419 /* select a random nonce */
2420 msgr
->inst
.addr
.type
= 0;
2421 get_random_bytes(&msgr
->inst
.addr
.nonce
, sizeof(msgr
->inst
.addr
.nonce
));
2422 encode_my_addr(msgr
);
2423 msgr
->nocrc
= nocrc
;
2425 atomic_set(&msgr
->stopping
, 0);
2427 dout("%s %p\n", __func__
, msgr
);
2429 EXPORT_SYMBOL(ceph_messenger_init
);
2431 static void clear_standby(struct ceph_connection
*con
)
2433 /* come back from STANDBY? */
2434 if (test_and_clear_bit(STANDBY
, &con
->state
)) {
2435 dout("clear_standby %p and ++connect_seq\n", con
);
2437 WARN_ON(test_bit(WRITE_PENDING
, &con
->flags
));
2438 WARN_ON(test_bit(KEEPALIVE_PENDING
, &con
->flags
));
2443 * Queue up an outgoing message on the given connection.
2445 void ceph_con_send(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2448 msg
->hdr
.src
= con
->msgr
->inst
.name
;
2449 BUG_ON(msg
->front
.iov_len
!= le32_to_cpu(msg
->hdr
.front_len
));
2450 msg
->needs_out_seq
= true;
2452 mutex_lock(&con
->mutex
);
2454 if (test_bit(CLOSED
, &con
->state
)) {
2455 dout("con_send %p closed, dropping %p\n", con
, msg
);
2457 mutex_unlock(&con
->mutex
);
2461 BUG_ON(msg
->con
!= NULL
);
2462 msg
->con
= con
->ops
->get(con
);
2463 BUG_ON(msg
->con
== NULL
);
2465 BUG_ON(!list_empty(&msg
->list_head
));
2466 list_add_tail(&msg
->list_head
, &con
->out_queue
);
2467 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg
,
2468 ENTITY_NAME(con
->peer_name
), le16_to_cpu(msg
->hdr
.type
),
2469 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2470 le32_to_cpu(msg
->hdr
.front_len
),
2471 le32_to_cpu(msg
->hdr
.middle_len
),
2472 le32_to_cpu(msg
->hdr
.data_len
));
2475 mutex_unlock(&con
->mutex
);
2477 /* if there wasn't anything waiting to send before, queue
2479 if (test_and_set_bit(WRITE_PENDING
, &con
->flags
) == 0)
2482 EXPORT_SYMBOL(ceph_con_send
);
2485 * Revoke a message that was previously queued for send
2487 void ceph_msg_revoke(struct ceph_msg
*msg
)
2489 struct ceph_connection
*con
= msg
->con
;
2492 return; /* Message not in our possession */
2494 mutex_lock(&con
->mutex
);
2495 if (!list_empty(&msg
->list_head
)) {
2496 dout("%s %p msg %p - was on queue\n", __func__
, con
, msg
);
2497 list_del_init(&msg
->list_head
);
2498 BUG_ON(msg
->con
== NULL
);
2499 msg
->con
->ops
->put(msg
->con
);
2505 if (con
->out_msg
== msg
) {
2506 dout("%s %p msg %p - was sending\n", __func__
, con
, msg
);
2507 con
->out_msg
= NULL
;
2508 if (con
->out_kvec_is_msg
) {
2509 con
->out_skip
= con
->out_kvec_bytes
;
2510 con
->out_kvec_is_msg
= false;
2516 mutex_unlock(&con
->mutex
);
2520 * Revoke a message that we may be reading data into
2522 void ceph_msg_revoke_incoming(struct ceph_msg
*msg
)
2524 struct ceph_connection
*con
;
2526 BUG_ON(msg
== NULL
);
2528 dout("%s msg %p null con\n", __func__
, msg
);
2530 return; /* Message not in our possession */
2534 mutex_lock(&con
->mutex
);
2535 if (con
->in_msg
== msg
) {
2536 unsigned int front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
2537 unsigned int middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
2538 unsigned int data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
2540 /* skip rest of message */
2541 dout("%s %p msg %p revoked\n", __func__
, con
, msg
);
2542 con
->in_base_pos
= con
->in_base_pos
-
2543 sizeof(struct ceph_msg_header
) -
2547 sizeof(struct ceph_msg_footer
);
2548 ceph_msg_put(con
->in_msg
);
2550 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2553 dout("%s %p in_msg %p msg %p no-op\n",
2554 __func__
, con
, con
->in_msg
, msg
);
2556 mutex_unlock(&con
->mutex
);
2560 * Queue a keepalive byte to ensure the tcp connection is alive.
2562 void ceph_con_keepalive(struct ceph_connection
*con
)
2564 dout("con_keepalive %p\n", con
);
2565 mutex_lock(&con
->mutex
);
2567 mutex_unlock(&con
->mutex
);
2568 if (test_and_set_bit(KEEPALIVE_PENDING
, &con
->flags
) == 0 &&
2569 test_and_set_bit(WRITE_PENDING
, &con
->flags
) == 0)
2572 EXPORT_SYMBOL(ceph_con_keepalive
);
2576 * construct a new message with given type, size
2577 * the new msg has a ref count of 1.
2579 struct ceph_msg
*ceph_msg_new(int type
, int front_len
, gfp_t flags
,
2584 m
= kmalloc(sizeof(*m
), flags
);
2587 kref_init(&m
->kref
);
2590 INIT_LIST_HEAD(&m
->list_head
);
2593 m
->hdr
.type
= cpu_to_le16(type
);
2594 m
->hdr
.priority
= cpu_to_le16(CEPH_MSG_PRIO_DEFAULT
);
2596 m
->hdr
.front_len
= cpu_to_le32(front_len
);
2597 m
->hdr
.middle_len
= 0;
2598 m
->hdr
.data_len
= 0;
2599 m
->hdr
.data_off
= 0;
2600 m
->hdr
.reserved
= 0;
2601 m
->footer
.front_crc
= 0;
2602 m
->footer
.middle_crc
= 0;
2603 m
->footer
.data_crc
= 0;
2604 m
->footer
.flags
= 0;
2605 m
->front_max
= front_len
;
2606 m
->front_is_vmalloc
= false;
2607 m
->more_to_follow
= false;
2616 m
->page_alignment
= 0;
2626 if (front_len
> PAGE_CACHE_SIZE
) {
2627 m
->front
.iov_base
= __vmalloc(front_len
, flags
,
2629 m
->front_is_vmalloc
= true;
2631 m
->front
.iov_base
= kmalloc(front_len
, flags
);
2633 if (m
->front
.iov_base
== NULL
) {
2634 dout("ceph_msg_new can't allocate %d bytes\n",
2639 m
->front
.iov_base
= NULL
;
2641 m
->front
.iov_len
= front_len
;
2643 dout("ceph_msg_new %p front %d\n", m
, front_len
);
2650 pr_err("msg_new can't create type %d front %d\n", type
,
2654 dout("msg_new can't create type %d front %d\n", type
,
2659 EXPORT_SYMBOL(ceph_msg_new
);
2662 * Allocate "middle" portion of a message, if it is needed and wasn't
2663 * allocated by alloc_msg. This allows us to read a small fixed-size
2664 * per-type header in the front and then gracefully fail (i.e.,
2665 * propagate the error to the caller based on info in the front) when
2666 * the middle is too large.
2668 static int ceph_alloc_middle(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2670 int type
= le16_to_cpu(msg
->hdr
.type
);
2671 int middle_len
= le32_to_cpu(msg
->hdr
.middle_len
);
2673 dout("alloc_middle %p type %d %s middle_len %d\n", msg
, type
,
2674 ceph_msg_type_name(type
), middle_len
);
2675 BUG_ON(!middle_len
);
2676 BUG_ON(msg
->middle
);
2678 msg
->middle
= ceph_buffer_new(middle_len
, GFP_NOFS
);
2685 * Allocate a message for receiving an incoming message on a
2686 * connection, and save the result in con->in_msg. Uses the
2687 * connection's private alloc_msg op if available.
2689 * Returns true if the message should be skipped, false otherwise.
2690 * If true is returned (skip message), con->in_msg will be NULL.
2691 * If false is returned, con->in_msg will contain a pointer to the
2692 * newly-allocated message, or NULL in case of memory exhaustion.
2694 static bool ceph_con_in_msg_alloc(struct ceph_connection
*con
,
2695 struct ceph_msg_header
*hdr
)
2697 int type
= le16_to_cpu(hdr
->type
);
2698 int front_len
= le32_to_cpu(hdr
->front_len
);
2699 int middle_len
= le32_to_cpu(hdr
->middle_len
);
2702 BUG_ON(con
->in_msg
!= NULL
);
2704 if (con
->ops
->alloc_msg
) {
2707 mutex_unlock(&con
->mutex
);
2708 con
->in_msg
= con
->ops
->alloc_msg(con
, hdr
, &skip
);
2709 mutex_lock(&con
->mutex
);
2711 con
->in_msg
->con
= con
->ops
->get(con
);
2712 BUG_ON(con
->in_msg
->con
== NULL
);
2721 con
->in_msg
= ceph_msg_new(type
, front_len
, GFP_NOFS
, false);
2723 pr_err("unable to allocate msg type %d len %d\n",
2727 con
->in_msg
->con
= con
->ops
->get(con
);
2728 BUG_ON(con
->in_msg
->con
== NULL
);
2729 con
->in_msg
->page_alignment
= le16_to_cpu(hdr
->data_off
);
2731 memcpy(&con
->in_msg
->hdr
, &con
->in_hdr
, sizeof(con
->in_hdr
));
2733 if (middle_len
&& !con
->in_msg
->middle
) {
2734 ret
= ceph_alloc_middle(con
, con
->in_msg
);
2736 ceph_msg_put(con
->in_msg
);
2746 * Free a generically kmalloc'd message.
2748 void ceph_msg_kfree(struct ceph_msg
*m
)
2750 dout("msg_kfree %p\n", m
);
2751 if (m
->front_is_vmalloc
)
2752 vfree(m
->front
.iov_base
);
2754 kfree(m
->front
.iov_base
);
2759 * Drop a msg ref. Destroy as needed.
2761 void ceph_msg_last_put(struct kref
*kref
)
2763 struct ceph_msg
*m
= container_of(kref
, struct ceph_msg
, kref
);
2765 dout("ceph_msg_put last one on %p\n", m
);
2766 WARN_ON(!list_empty(&m
->list_head
));
2768 /* drop middle, data, if any */
2770 ceph_buffer_put(m
->middle
);
2777 ceph_pagelist_release(m
->pagelist
);
2785 ceph_msgpool_put(m
->pool
, m
);
2789 EXPORT_SYMBOL(ceph_msg_last_put
);
2791 void ceph_msg_dump(struct ceph_msg
*msg
)
2793 pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg
,
2794 msg
->front_max
, msg
->nr_pages
);
2795 print_hex_dump(KERN_DEBUG
, "header: ",
2796 DUMP_PREFIX_OFFSET
, 16, 1,
2797 &msg
->hdr
, sizeof(msg
->hdr
), true);
2798 print_hex_dump(KERN_DEBUG
, " front: ",
2799 DUMP_PREFIX_OFFSET
, 16, 1,
2800 msg
->front
.iov_base
, msg
->front
.iov_len
, true);
2802 print_hex_dump(KERN_DEBUG
, "middle: ",
2803 DUMP_PREFIX_OFFSET
, 16, 1,
2804 msg
->middle
->vec
.iov_base
,
2805 msg
->middle
->vec
.iov_len
, true);
2806 print_hex_dump(KERN_DEBUG
, "footer: ",
2807 DUMP_PREFIX_OFFSET
, 16, 1,
2808 &msg
->footer
, sizeof(msg
->footer
), true);
2810 EXPORT_SYMBOL(ceph_msg_dump
);