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
;
258 if (sk
->sk_state
!= TCP_CLOSE_WAIT
) {
259 dout("%s on %p state = %lu, queueing work\n", __func__
,
265 /* socket has buffer space for writing */
266 static void ceph_sock_write_space(struct sock
*sk
)
268 struct ceph_connection
*con
= sk
->sk_user_data
;
270 /* only queue to workqueue if there is data we want to write,
271 * and there is sufficient space in the socket buffer to accept
272 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
273 * doesn't get called again until try_write() fills the socket
274 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
275 * and net/core/stream.c:sk_stream_write_space().
277 if (test_bit(WRITE_PENDING
, &con
->flags
)) {
278 if (sk_stream_wspace(sk
) >= sk_stream_min_wspace(sk
)) {
279 dout("%s %p queueing write work\n", __func__
, con
);
280 clear_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
284 dout("%s %p nothing to write\n", __func__
, con
);
288 /* socket's state has changed */
289 static void ceph_sock_state_change(struct sock
*sk
)
291 struct ceph_connection
*con
= sk
->sk_user_data
;
293 dout("%s %p state = %lu sk_state = %u\n", __func__
,
294 con
, con
->state
, sk
->sk_state
);
296 if (test_bit(CLOSED
, &con
->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 dout("con_close %p peer %s\n", con
,
504 ceph_pr_addr(&con
->peer_addr
.in_addr
));
505 clear_bit(NEGOTIATING
, &con
->state
);
506 clear_bit(CONNECTING
, &con
->state
);
507 clear_bit(CONNECTED
, &con
->state
);
508 clear_bit(STANDBY
, &con
->state
); /* avoid connect_seq bump */
509 set_bit(CLOSED
, &con
->state
);
511 clear_bit(LOSSYTX
, &con
->flags
); /* so we retry next connect */
512 clear_bit(KEEPALIVE_PENDING
, &con
->flags
);
513 clear_bit(WRITE_PENDING
, &con
->flags
);
515 mutex_lock(&con
->mutex
);
516 reset_connection(con
);
517 con
->peer_global_seq
= 0;
518 cancel_delayed_work(&con
->work
);
519 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 dout("con_open %p %s\n", con
, ceph_pr_addr(&addr
->in_addr
));
532 set_bit(OPENING
, &con
->state
);
533 WARN_ON(!test_and_clear_bit(CLOSED
, &con
->state
));
535 con
->peer_name
.type
= (__u8
) entity_type
;
536 con
->peer_name
.num
= cpu_to_le64(entity_num
);
538 memcpy(&con
->peer_addr
, addr
, sizeof(*addr
));
539 con
->delay
= 0; /* reset backoff memory */
542 EXPORT_SYMBOL(ceph_con_open
);
545 * return true if this connection ever successfully opened
547 bool ceph_con_opened(struct ceph_connection
*con
)
549 return con
->connect_seq
> 0;
553 * initialize a new connection.
555 void ceph_con_init(struct ceph_connection
*con
, void *private,
556 const struct ceph_connection_operations
*ops
,
557 struct ceph_messenger
*msgr
)
559 dout("con_init %p\n", con
);
560 memset(con
, 0, sizeof(*con
));
561 con
->private = private;
565 con_sock_state_init(con
);
567 mutex_init(&con
->mutex
);
568 INIT_LIST_HEAD(&con
->out_queue
);
569 INIT_LIST_HEAD(&con
->out_sent
);
570 INIT_DELAYED_WORK(&con
->work
, con_work
);
572 set_bit(CLOSED
, &con
->state
);
574 EXPORT_SYMBOL(ceph_con_init
);
578 * We maintain a global counter to order connection attempts. Get
579 * a unique seq greater than @gt.
581 static u32
get_global_seq(struct ceph_messenger
*msgr
, u32 gt
)
585 spin_lock(&msgr
->global_seq_lock
);
586 if (msgr
->global_seq
< gt
)
587 msgr
->global_seq
= gt
;
588 ret
= ++msgr
->global_seq
;
589 spin_unlock(&msgr
->global_seq_lock
);
593 static void con_out_kvec_reset(struct ceph_connection
*con
)
595 con
->out_kvec_left
= 0;
596 con
->out_kvec_bytes
= 0;
597 con
->out_kvec_cur
= &con
->out_kvec
[0];
600 static void con_out_kvec_add(struct ceph_connection
*con
,
601 size_t size
, void *data
)
605 index
= con
->out_kvec_left
;
606 BUG_ON(index
>= ARRAY_SIZE(con
->out_kvec
));
608 con
->out_kvec
[index
].iov_len
= size
;
609 con
->out_kvec
[index
].iov_base
= data
;
610 con
->out_kvec_left
++;
611 con
->out_kvec_bytes
+= size
;
615 static void init_bio_iter(struct bio
*bio
, struct bio
**iter
, int *seg
)
626 static void iter_bio_next(struct bio
**bio_iter
, int *seg
)
628 if (*bio_iter
== NULL
)
631 BUG_ON(*seg
>= (*bio_iter
)->bi_vcnt
);
634 if (*seg
== (*bio_iter
)->bi_vcnt
)
635 init_bio_iter((*bio_iter
)->bi_next
, bio_iter
, seg
);
639 static void prepare_write_message_data(struct ceph_connection
*con
)
641 struct ceph_msg
*msg
= con
->out_msg
;
644 BUG_ON(!msg
->hdr
.data_len
);
646 /* initialize page iterator */
647 con
->out_msg_pos
.page
= 0;
649 con
->out_msg_pos
.page_pos
= msg
->page_alignment
;
651 con
->out_msg_pos
.page_pos
= 0;
654 init_bio_iter(msg
->bio
, &msg
->bio_iter
, &msg
->bio_seg
);
656 con
->out_msg_pos
.data_pos
= 0;
657 con
->out_msg_pos
.did_page_crc
= false;
658 con
->out_more
= 1; /* data + footer will follow */
662 * Prepare footer for currently outgoing message, and finish things
663 * off. Assumes out_kvec* are already valid.. we just add on to the end.
665 static void prepare_write_message_footer(struct ceph_connection
*con
)
667 struct ceph_msg
*m
= con
->out_msg
;
668 int v
= con
->out_kvec_left
;
670 m
->footer
.flags
|= CEPH_MSG_FOOTER_COMPLETE
;
672 dout("prepare_write_message_footer %p\n", con
);
673 con
->out_kvec_is_msg
= true;
674 con
->out_kvec
[v
].iov_base
= &m
->footer
;
675 con
->out_kvec
[v
].iov_len
= sizeof(m
->footer
);
676 con
->out_kvec_bytes
+= sizeof(m
->footer
);
677 con
->out_kvec_left
++;
678 con
->out_more
= m
->more_to_follow
;
679 con
->out_msg_done
= true;
683 * Prepare headers for the next outgoing message.
685 static void prepare_write_message(struct ceph_connection
*con
)
690 con_out_kvec_reset(con
);
691 con
->out_kvec_is_msg
= true;
692 con
->out_msg_done
= false;
694 /* Sneak an ack in there first? If we can get it into the same
695 * TCP packet that's a good thing. */
696 if (con
->in_seq
> con
->in_seq_acked
) {
697 con
->in_seq_acked
= con
->in_seq
;
698 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
699 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
700 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
704 BUG_ON(list_empty(&con
->out_queue
));
705 m
= list_first_entry(&con
->out_queue
, struct ceph_msg
, list_head
);
707 BUG_ON(m
->con
!= con
);
709 /* put message on sent list */
711 list_move_tail(&m
->list_head
, &con
->out_sent
);
714 * only assign outgoing seq # if we haven't sent this message
715 * yet. if it is requeued, resend with it's original seq.
717 if (m
->needs_out_seq
) {
718 m
->hdr
.seq
= cpu_to_le64(++con
->out_seq
);
719 m
->needs_out_seq
= false;
722 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n",
723 m
, con
->out_seq
, le16_to_cpu(m
->hdr
.type
),
724 le32_to_cpu(m
->hdr
.front_len
), le32_to_cpu(m
->hdr
.middle_len
),
725 le32_to_cpu(m
->hdr
.data_len
),
727 BUG_ON(le32_to_cpu(m
->hdr
.front_len
) != m
->front
.iov_len
);
729 /* tag + hdr + front + middle */
730 con_out_kvec_add(con
, sizeof (tag_msg
), &tag_msg
);
731 con_out_kvec_add(con
, sizeof (m
->hdr
), &m
->hdr
);
732 con_out_kvec_add(con
, m
->front
.iov_len
, m
->front
.iov_base
);
735 con_out_kvec_add(con
, m
->middle
->vec
.iov_len
,
736 m
->middle
->vec
.iov_base
);
738 /* fill in crc (except data pages), footer */
739 crc
= crc32c(0, &m
->hdr
, offsetof(struct ceph_msg_header
, crc
));
740 con
->out_msg
->hdr
.crc
= cpu_to_le32(crc
);
741 con
->out_msg
->footer
.flags
= 0;
743 crc
= crc32c(0, m
->front
.iov_base
, m
->front
.iov_len
);
744 con
->out_msg
->footer
.front_crc
= cpu_to_le32(crc
);
746 crc
= crc32c(0, m
->middle
->vec
.iov_base
,
747 m
->middle
->vec
.iov_len
);
748 con
->out_msg
->footer
.middle_crc
= cpu_to_le32(crc
);
750 con
->out_msg
->footer
.middle_crc
= 0;
751 dout("%s front_crc %u middle_crc %u\n", __func__
,
752 le32_to_cpu(con
->out_msg
->footer
.front_crc
),
753 le32_to_cpu(con
->out_msg
->footer
.middle_crc
));
755 /* is there a data payload? */
756 con
->out_msg
->footer
.data_crc
= 0;
758 prepare_write_message_data(con
);
760 /* no, queue up footer too and be done */
761 prepare_write_message_footer(con
);
763 set_bit(WRITE_PENDING
, &con
->flags
);
769 static void prepare_write_ack(struct ceph_connection
*con
)
771 dout("prepare_write_ack %p %llu -> %llu\n", con
,
772 con
->in_seq_acked
, con
->in_seq
);
773 con
->in_seq_acked
= con
->in_seq
;
775 con_out_kvec_reset(con
);
777 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
779 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
780 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
783 con
->out_more
= 1; /* more will follow.. eventually.. */
784 set_bit(WRITE_PENDING
, &con
->flags
);
788 * Prepare to write keepalive byte.
790 static void prepare_write_keepalive(struct ceph_connection
*con
)
792 dout("prepare_write_keepalive %p\n", con
);
793 con_out_kvec_reset(con
);
794 con_out_kvec_add(con
, sizeof (tag_keepalive
), &tag_keepalive
);
795 set_bit(WRITE_PENDING
, &con
->flags
);
799 * Connection negotiation.
802 static struct ceph_auth_handshake
*get_connect_authorizer(struct ceph_connection
*con
,
805 struct ceph_auth_handshake
*auth
;
807 if (!con
->ops
->get_authorizer
) {
808 con
->out_connect
.authorizer_protocol
= CEPH_AUTH_UNKNOWN
;
809 con
->out_connect
.authorizer_len
= 0;
814 /* Can't hold the mutex while getting authorizer */
816 mutex_unlock(&con
->mutex
);
818 auth
= con
->ops
->get_authorizer(con
, auth_proto
, con
->auth_retry
);
820 mutex_lock(&con
->mutex
);
824 if (test_bit(CLOSED
, &con
->state
) || test_bit(OPENING
, &con
->flags
))
825 return ERR_PTR(-EAGAIN
);
827 con
->auth_reply_buf
= auth
->authorizer_reply_buf
;
828 con
->auth_reply_buf_len
= auth
->authorizer_reply_buf_len
;
835 * We connected to a peer and are saying hello.
837 static void prepare_write_banner(struct ceph_connection
*con
)
839 con_out_kvec_add(con
, strlen(CEPH_BANNER
), CEPH_BANNER
);
840 con_out_kvec_add(con
, sizeof (con
->msgr
->my_enc_addr
),
841 &con
->msgr
->my_enc_addr
);
844 set_bit(WRITE_PENDING
, &con
->flags
);
847 static int prepare_write_connect(struct ceph_connection
*con
)
849 unsigned int global_seq
= get_global_seq(con
->msgr
, 0);
852 struct ceph_auth_handshake
*auth
;
854 switch (con
->peer_name
.type
) {
855 case CEPH_ENTITY_TYPE_MON
:
856 proto
= CEPH_MONC_PROTOCOL
;
858 case CEPH_ENTITY_TYPE_OSD
:
859 proto
= CEPH_OSDC_PROTOCOL
;
861 case CEPH_ENTITY_TYPE_MDS
:
862 proto
= CEPH_MDSC_PROTOCOL
;
868 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con
,
869 con
->connect_seq
, global_seq
, proto
);
871 con
->out_connect
.features
= cpu_to_le64(con
->msgr
->supported_features
);
872 con
->out_connect
.host_type
= cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT
);
873 con
->out_connect
.connect_seq
= cpu_to_le32(con
->connect_seq
);
874 con
->out_connect
.global_seq
= cpu_to_le32(global_seq
);
875 con
->out_connect
.protocol_version
= cpu_to_le32(proto
);
876 con
->out_connect
.flags
= 0;
878 auth_proto
= CEPH_AUTH_UNKNOWN
;
879 auth
= get_connect_authorizer(con
, &auth_proto
);
881 return PTR_ERR(auth
);
883 con
->out_connect
.authorizer_protocol
= cpu_to_le32(auth_proto
);
884 con
->out_connect
.authorizer_len
= auth
?
885 cpu_to_le32(auth
->authorizer_buf_len
) : 0;
887 con_out_kvec_reset(con
);
888 con_out_kvec_add(con
, sizeof (con
->out_connect
),
890 if (auth
&& auth
->authorizer_buf_len
)
891 con_out_kvec_add(con
, auth
->authorizer_buf_len
,
892 auth
->authorizer_buf
);
895 set_bit(WRITE_PENDING
, &con
->flags
);
901 * write as much of pending kvecs to the socket as we can.
903 * 0 -> socket full, but more to do
906 static int write_partial_kvec(struct ceph_connection
*con
)
910 dout("write_partial_kvec %p %d left\n", con
, con
->out_kvec_bytes
);
911 while (con
->out_kvec_bytes
> 0) {
912 ret
= ceph_tcp_sendmsg(con
->sock
, con
->out_kvec_cur
,
913 con
->out_kvec_left
, con
->out_kvec_bytes
,
917 con
->out_kvec_bytes
-= ret
;
918 if (con
->out_kvec_bytes
== 0)
921 /* account for full iov entries consumed */
922 while (ret
>= con
->out_kvec_cur
->iov_len
) {
923 BUG_ON(!con
->out_kvec_left
);
924 ret
-= con
->out_kvec_cur
->iov_len
;
926 con
->out_kvec_left
--;
928 /* and for a partially-consumed entry */
930 con
->out_kvec_cur
->iov_len
-= ret
;
931 con
->out_kvec_cur
->iov_base
+= ret
;
934 con
->out_kvec_left
= 0;
935 con
->out_kvec_is_msg
= false;
938 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con
,
939 con
->out_kvec_bytes
, con
->out_kvec_left
, ret
);
940 return ret
; /* done! */
943 static void out_msg_pos_next(struct ceph_connection
*con
, struct page
*page
,
944 size_t len
, size_t sent
, bool in_trail
)
946 struct ceph_msg
*msg
= con
->out_msg
;
951 con
->out_msg_pos
.data_pos
+= sent
;
952 con
->out_msg_pos
.page_pos
+= sent
;
957 con
->out_msg_pos
.page_pos
= 0;
958 con
->out_msg_pos
.page
++;
959 con
->out_msg_pos
.did_page_crc
= false;
961 list_move_tail(&page
->lru
,
963 else if (msg
->pagelist
)
964 list_move_tail(&page
->lru
,
965 &msg
->pagelist
->head
);
968 iter_bio_next(&msg
->bio_iter
, &msg
->bio_seg
);
973 * Write as much message data payload as we can. If we finish, queue
975 * 1 -> done, footer is now queued in out_kvec[].
976 * 0 -> socket full, but more to do
979 static int write_partial_msg_pages(struct ceph_connection
*con
)
981 struct ceph_msg
*msg
= con
->out_msg
;
982 unsigned int data_len
= le32_to_cpu(msg
->hdr
.data_len
);
984 bool do_datacrc
= !con
->msgr
->nocrc
;
987 bool in_trail
= false;
988 const size_t trail_len
= (msg
->trail
? msg
->trail
->length
: 0);
989 const size_t trail_off
= data_len
- trail_len
;
991 dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
992 con
, msg
, con
->out_msg_pos
.page
, msg
->nr_pages
,
993 con
->out_msg_pos
.page_pos
);
996 * Iterate through each page that contains data to be
997 * written, and send as much as possible for each.
999 * If we are calculating the data crc (the default), we will
1000 * need to map the page. If we have no pages, they have
1001 * been revoked, so use the zero page.
1003 while (data_len
> con
->out_msg_pos
.data_pos
) {
1004 struct page
*page
= NULL
;
1005 int max_write
= PAGE_SIZE
;
1008 in_trail
= in_trail
|| con
->out_msg_pos
.data_pos
>= trail_off
;
1010 total_max_write
= trail_off
- con
->out_msg_pos
.data_pos
;
1013 total_max_write
= data_len
- con
->out_msg_pos
.data_pos
;
1015 page
= list_first_entry(&msg
->trail
->head
,
1017 } else if (msg
->pages
) {
1018 page
= msg
->pages
[con
->out_msg_pos
.page
];
1019 } else if (msg
->pagelist
) {
1020 page
= list_first_entry(&msg
->pagelist
->head
,
1023 } else if (msg
->bio
) {
1026 bv
= bio_iovec_idx(msg
->bio_iter
, msg
->bio_seg
);
1028 bio_offset
= bv
->bv_offset
;
1029 max_write
= bv
->bv_len
;
1034 len
= min_t(int, max_write
- con
->out_msg_pos
.page_pos
,
1037 if (do_datacrc
&& !con
->out_msg_pos
.did_page_crc
) {
1039 u32 crc
= le32_to_cpu(msg
->footer
.data_crc
);
1043 BUG_ON(kaddr
== NULL
);
1044 base
= kaddr
+ con
->out_msg_pos
.page_pos
+ bio_offset
;
1045 crc
= crc32c(crc
, base
, len
);
1046 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1047 con
->out_msg_pos
.did_page_crc
= true;
1049 ret
= ceph_tcp_sendpage(con
->sock
, page
,
1050 con
->out_msg_pos
.page_pos
+ bio_offset
,
1059 out_msg_pos_next(con
, page
, len
, (size_t) ret
, in_trail
);
1062 dout("write_partial_msg_pages %p msg %p done\n", con
, msg
);
1064 /* prepare and queue up footer, too */
1066 msg
->footer
.flags
|= CEPH_MSG_FOOTER_NOCRC
;
1067 con_out_kvec_reset(con
);
1068 prepare_write_message_footer(con
);
1077 static int write_partial_skip(struct ceph_connection
*con
)
1081 while (con
->out_skip
> 0) {
1082 size_t size
= min(con
->out_skip
, (int) PAGE_CACHE_SIZE
);
1084 ret
= ceph_tcp_sendpage(con
->sock
, zero_page
, 0, size
, 1);
1087 con
->out_skip
-= ret
;
1095 * Prepare to read connection handshake, or an ack.
1097 static void prepare_read_banner(struct ceph_connection
*con
)
1099 dout("prepare_read_banner %p\n", con
);
1100 con
->in_base_pos
= 0;
1103 static void prepare_read_connect(struct ceph_connection
*con
)
1105 dout("prepare_read_connect %p\n", con
);
1106 con
->in_base_pos
= 0;
1109 static void prepare_read_ack(struct ceph_connection
*con
)
1111 dout("prepare_read_ack %p\n", con
);
1112 con
->in_base_pos
= 0;
1115 static void prepare_read_tag(struct ceph_connection
*con
)
1117 dout("prepare_read_tag %p\n", con
);
1118 con
->in_base_pos
= 0;
1119 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1123 * Prepare to read a message.
1125 static int prepare_read_message(struct ceph_connection
*con
)
1127 dout("prepare_read_message %p\n", con
);
1128 BUG_ON(con
->in_msg
!= NULL
);
1129 con
->in_base_pos
= 0;
1130 con
->in_front_crc
= con
->in_middle_crc
= con
->in_data_crc
= 0;
1135 static int read_partial(struct ceph_connection
*con
,
1136 int end
, int size
, void *object
)
1138 while (con
->in_base_pos
< end
) {
1139 int left
= end
- con
->in_base_pos
;
1140 int have
= size
- left
;
1141 int ret
= ceph_tcp_recvmsg(con
->sock
, object
+ have
, left
);
1144 con
->in_base_pos
+= ret
;
1151 * Read all or part of the connect-side handshake on a new connection
1153 static int read_partial_banner(struct ceph_connection
*con
)
1159 dout("read_partial_banner %p at %d\n", con
, con
->in_base_pos
);
1162 size
= strlen(CEPH_BANNER
);
1164 ret
= read_partial(con
, end
, size
, con
->in_banner
);
1168 size
= sizeof (con
->actual_peer_addr
);
1170 ret
= read_partial(con
, end
, size
, &con
->actual_peer_addr
);
1174 size
= sizeof (con
->peer_addr_for_me
);
1176 ret
= read_partial(con
, end
, size
, &con
->peer_addr_for_me
);
1184 static int read_partial_connect(struct ceph_connection
*con
)
1190 dout("read_partial_connect %p at %d\n", con
, con
->in_base_pos
);
1192 size
= sizeof (con
->in_reply
);
1194 ret
= read_partial(con
, end
, size
, &con
->in_reply
);
1198 size
= le32_to_cpu(con
->in_reply
.authorizer_len
);
1200 ret
= read_partial(con
, end
, size
, con
->auth_reply_buf
);
1204 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1205 con
, (int)con
->in_reply
.tag
,
1206 le32_to_cpu(con
->in_reply
.connect_seq
),
1207 le32_to_cpu(con
->in_reply
.global_seq
));
1214 * Verify the hello banner looks okay.
1216 static int verify_hello(struct ceph_connection
*con
)
1218 if (memcmp(con
->in_banner
, CEPH_BANNER
, strlen(CEPH_BANNER
))) {
1219 pr_err("connect to %s got bad banner\n",
1220 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1221 con
->error_msg
= "protocol error, bad banner";
1227 static bool addr_is_blank(struct sockaddr_storage
*ss
)
1229 switch (ss
->ss_family
) {
1231 return ((struct sockaddr_in
*)ss
)->sin_addr
.s_addr
== 0;
1234 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[0] == 0 &&
1235 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[1] == 0 &&
1236 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[2] == 0 &&
1237 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[3] == 0;
1242 static int addr_port(struct sockaddr_storage
*ss
)
1244 switch (ss
->ss_family
) {
1246 return ntohs(((struct sockaddr_in
*)ss
)->sin_port
);
1248 return ntohs(((struct sockaddr_in6
*)ss
)->sin6_port
);
1253 static void addr_set_port(struct sockaddr_storage
*ss
, int p
)
1255 switch (ss
->ss_family
) {
1257 ((struct sockaddr_in
*)ss
)->sin_port
= htons(p
);
1260 ((struct sockaddr_in6
*)ss
)->sin6_port
= htons(p
);
1266 * Unlike other *_pton function semantics, zero indicates success.
1268 static int ceph_pton(const char *str
, size_t len
, struct sockaddr_storage
*ss
,
1269 char delim
, const char **ipend
)
1271 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
1272 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
1274 memset(ss
, 0, sizeof(*ss
));
1276 if (in4_pton(str
, len
, (u8
*)&in4
->sin_addr
.s_addr
, delim
, ipend
)) {
1277 ss
->ss_family
= AF_INET
;
1281 if (in6_pton(str
, len
, (u8
*)&in6
->sin6_addr
.s6_addr
, delim
, ipend
)) {
1282 ss
->ss_family
= AF_INET6
;
1290 * Extract hostname string and resolve using kernel DNS facility.
1292 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1293 static int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1294 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1296 const char *end
, *delim_p
;
1297 char *colon_p
, *ip_addr
= NULL
;
1301 * The end of the hostname occurs immediately preceding the delimiter or
1302 * the port marker (':') where the delimiter takes precedence.
1304 delim_p
= memchr(name
, delim
, namelen
);
1305 colon_p
= memchr(name
, ':', namelen
);
1307 if (delim_p
&& colon_p
)
1308 end
= delim_p
< colon_p
? delim_p
: colon_p
;
1309 else if (!delim_p
&& colon_p
)
1313 if (!end
) /* case: hostname:/ */
1314 end
= name
+ namelen
;
1320 /* do dns_resolve upcall */
1321 ip_len
= dns_query(NULL
, name
, end
- name
, NULL
, &ip_addr
, NULL
);
1323 ret
= ceph_pton(ip_addr
, ip_len
, ss
, -1, NULL
);
1331 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end
- name
), name
,
1332 ret
, ret
? "failed" : ceph_pr_addr(ss
));
1337 static inline int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1338 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1345 * Parse a server name (IP or hostname). If a valid IP address is not found
1346 * then try to extract a hostname to resolve using userspace DNS upcall.
1348 static int ceph_parse_server_name(const char *name
, size_t namelen
,
1349 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1353 ret
= ceph_pton(name
, namelen
, ss
, delim
, ipend
);
1355 ret
= ceph_dns_resolve_name(name
, namelen
, ss
, delim
, ipend
);
1361 * Parse an ip[:port] list into an addr array. Use the default
1362 * monitor port if a port isn't specified.
1364 int ceph_parse_ips(const char *c
, const char *end
,
1365 struct ceph_entity_addr
*addr
,
1366 int max_count
, int *count
)
1368 int i
, ret
= -EINVAL
;
1371 dout("parse_ips on '%.*s'\n", (int)(end
-c
), c
);
1372 for (i
= 0; i
< max_count
; i
++) {
1374 struct sockaddr_storage
*ss
= &addr
[i
].in_addr
;
1383 ret
= ceph_parse_server_name(p
, end
- p
, ss
, delim
, &ipend
);
1392 dout("missing matching ']'\n");
1399 if (p
< end
&& *p
== ':') {
1402 while (p
< end
&& *p
>= '0' && *p
<= '9') {
1403 port
= (port
* 10) + (*p
- '0');
1406 if (port
> 65535 || port
== 0)
1409 port
= CEPH_MON_PORT
;
1412 addr_set_port(ss
, port
);
1414 dout("parse_ips got %s\n", ceph_pr_addr(ss
));
1431 pr_err("parse_ips bad ip '%.*s'\n", (int)(end
- c
), c
);
1434 EXPORT_SYMBOL(ceph_parse_ips
);
1436 static int process_banner(struct ceph_connection
*con
)
1438 dout("process_banner on %p\n", con
);
1440 if (verify_hello(con
) < 0)
1443 ceph_decode_addr(&con
->actual_peer_addr
);
1444 ceph_decode_addr(&con
->peer_addr_for_me
);
1447 * Make sure the other end is who we wanted. note that the other
1448 * end may not yet know their ip address, so if it's 0.0.0.0, give
1449 * them the benefit of the doubt.
1451 if (memcmp(&con
->peer_addr
, &con
->actual_peer_addr
,
1452 sizeof(con
->peer_addr
)) != 0 &&
1453 !(addr_is_blank(&con
->actual_peer_addr
.in_addr
) &&
1454 con
->actual_peer_addr
.nonce
== con
->peer_addr
.nonce
)) {
1455 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1456 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1457 (int)le32_to_cpu(con
->peer_addr
.nonce
),
1458 ceph_pr_addr(&con
->actual_peer_addr
.in_addr
),
1459 (int)le32_to_cpu(con
->actual_peer_addr
.nonce
));
1460 con
->error_msg
= "wrong peer at address";
1465 * did we learn our address?
1467 if (addr_is_blank(&con
->msgr
->inst
.addr
.in_addr
)) {
1468 int port
= addr_port(&con
->msgr
->inst
.addr
.in_addr
);
1470 memcpy(&con
->msgr
->inst
.addr
.in_addr
,
1471 &con
->peer_addr_for_me
.in_addr
,
1472 sizeof(con
->peer_addr_for_me
.in_addr
));
1473 addr_set_port(&con
->msgr
->inst
.addr
.in_addr
, port
);
1474 encode_my_addr(con
->msgr
);
1475 dout("process_banner learned my addr is %s\n",
1476 ceph_pr_addr(&con
->msgr
->inst
.addr
.in_addr
));
1482 static void fail_protocol(struct ceph_connection
*con
)
1484 reset_connection(con
);
1485 set_bit(CLOSED
, &con
->state
); /* in case there's queued work */
1488 static int process_connect(struct ceph_connection
*con
)
1490 u64 sup_feat
= con
->msgr
->supported_features
;
1491 u64 req_feat
= con
->msgr
->required_features
;
1492 u64 server_feat
= le64_to_cpu(con
->in_reply
.features
);
1495 dout("process_connect on %p tag %d\n", con
, (int)con
->in_tag
);
1497 switch (con
->in_reply
.tag
) {
1498 case CEPH_MSGR_TAG_FEATURES
:
1499 pr_err("%s%lld %s feature set mismatch,"
1500 " my %llx < server's %llx, missing %llx\n",
1501 ENTITY_NAME(con
->peer_name
),
1502 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1503 sup_feat
, server_feat
, server_feat
& ~sup_feat
);
1504 con
->error_msg
= "missing required protocol features";
1508 case CEPH_MSGR_TAG_BADPROTOVER
:
1509 pr_err("%s%lld %s protocol version mismatch,"
1510 " my %d != server's %d\n",
1511 ENTITY_NAME(con
->peer_name
),
1512 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1513 le32_to_cpu(con
->out_connect
.protocol_version
),
1514 le32_to_cpu(con
->in_reply
.protocol_version
));
1515 con
->error_msg
= "protocol version mismatch";
1519 case CEPH_MSGR_TAG_BADAUTHORIZER
:
1521 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con
,
1523 if (con
->auth_retry
== 2) {
1524 con
->error_msg
= "connect authorization failure";
1527 con
->auth_retry
= 1;
1528 ret
= prepare_write_connect(con
);
1531 prepare_read_connect(con
);
1534 case CEPH_MSGR_TAG_RESETSESSION
:
1536 * If we connected with a large connect_seq but the peer
1537 * has no record of a session with us (no connection, or
1538 * connect_seq == 0), they will send RESETSESION to indicate
1539 * that they must have reset their session, and may have
1542 dout("process_connect got RESET peer seq %u\n",
1543 le32_to_cpu(con
->in_connect
.connect_seq
));
1544 pr_err("%s%lld %s connection reset\n",
1545 ENTITY_NAME(con
->peer_name
),
1546 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1547 reset_connection(con
);
1548 ret
= prepare_write_connect(con
);
1551 prepare_read_connect(con
);
1553 /* Tell ceph about it. */
1554 mutex_unlock(&con
->mutex
);
1555 pr_info("reset on %s%lld\n", ENTITY_NAME(con
->peer_name
));
1556 if (con
->ops
->peer_reset
)
1557 con
->ops
->peer_reset(con
);
1558 mutex_lock(&con
->mutex
);
1559 if (test_bit(CLOSED
, &con
->state
) ||
1560 test_bit(OPENING
, &con
->state
))
1564 case CEPH_MSGR_TAG_RETRY_SESSION
:
1566 * If we sent a smaller connect_seq than the peer has, try
1567 * again with a larger value.
1569 dout("process_connect got RETRY my seq = %u, peer_seq = %u\n",
1570 le32_to_cpu(con
->out_connect
.connect_seq
),
1571 le32_to_cpu(con
->in_connect
.connect_seq
));
1572 con
->connect_seq
= le32_to_cpu(con
->in_connect
.connect_seq
);
1573 ret
= prepare_write_connect(con
);
1576 prepare_read_connect(con
);
1579 case CEPH_MSGR_TAG_RETRY_GLOBAL
:
1581 * If we sent a smaller global_seq than the peer has, try
1582 * again with a larger value.
1584 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1585 con
->peer_global_seq
,
1586 le32_to_cpu(con
->in_connect
.global_seq
));
1587 get_global_seq(con
->msgr
,
1588 le32_to_cpu(con
->in_connect
.global_seq
));
1589 ret
= prepare_write_connect(con
);
1592 prepare_read_connect(con
);
1595 case CEPH_MSGR_TAG_READY
:
1596 if (req_feat
& ~server_feat
) {
1597 pr_err("%s%lld %s protocol feature mismatch,"
1598 " my required %llx > server's %llx, need %llx\n",
1599 ENTITY_NAME(con
->peer_name
),
1600 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1601 req_feat
, server_feat
, req_feat
& ~server_feat
);
1602 con
->error_msg
= "missing required protocol features";
1606 clear_bit(NEGOTIATING
, &con
->state
);
1607 set_bit(CONNECTED
, &con
->state
);
1608 con
->peer_global_seq
= le32_to_cpu(con
->in_reply
.global_seq
);
1610 con
->peer_features
= server_feat
;
1611 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1612 con
->peer_global_seq
,
1613 le32_to_cpu(con
->in_reply
.connect_seq
),
1615 WARN_ON(con
->connect_seq
!=
1616 le32_to_cpu(con
->in_reply
.connect_seq
));
1618 if (con
->in_reply
.flags
& CEPH_MSG_CONNECT_LOSSY
)
1619 set_bit(LOSSYTX
, &con
->flags
);
1621 prepare_read_tag(con
);
1624 case CEPH_MSGR_TAG_WAIT
:
1626 * If there is a connection race (we are opening
1627 * connections to each other), one of us may just have
1628 * to WAIT. This shouldn't happen if we are the
1631 pr_err("process_connect got WAIT as client\n");
1632 con
->error_msg
= "protocol error, got WAIT as client";
1636 pr_err("connect protocol error, will retry\n");
1637 con
->error_msg
= "protocol error, garbage tag during connect";
1645 * read (part of) an ack
1647 static int read_partial_ack(struct ceph_connection
*con
)
1649 int size
= sizeof (con
->in_temp_ack
);
1652 return read_partial(con
, end
, size
, &con
->in_temp_ack
);
1657 * We can finally discard anything that's been acked.
1659 static void process_ack(struct ceph_connection
*con
)
1662 u64 ack
= le64_to_cpu(con
->in_temp_ack
);
1665 while (!list_empty(&con
->out_sent
)) {
1666 m
= list_first_entry(&con
->out_sent
, struct ceph_msg
,
1668 seq
= le64_to_cpu(m
->hdr
.seq
);
1671 dout("got ack for seq %llu type %d at %p\n", seq
,
1672 le16_to_cpu(m
->hdr
.type
), m
);
1673 m
->ack_stamp
= jiffies
;
1676 prepare_read_tag(con
);
1682 static int read_partial_message_section(struct ceph_connection
*con
,
1683 struct kvec
*section
,
1684 unsigned int sec_len
, u32
*crc
)
1690 while (section
->iov_len
< sec_len
) {
1691 BUG_ON(section
->iov_base
== NULL
);
1692 left
= sec_len
- section
->iov_len
;
1693 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)section
->iov_base
+
1694 section
->iov_len
, left
);
1697 section
->iov_len
+= ret
;
1699 if (section
->iov_len
== sec_len
)
1700 *crc
= crc32c(0, section
->iov_base
, section
->iov_len
);
1705 static bool ceph_con_in_msg_alloc(struct ceph_connection
*con
,
1706 struct ceph_msg_header
*hdr
);
1709 static int read_partial_message_pages(struct ceph_connection
*con
,
1710 struct page
**pages
,
1711 unsigned int data_len
, bool do_datacrc
)
1717 left
= min((int)(data_len
- con
->in_msg_pos
.data_pos
),
1718 (int)(PAGE_SIZE
- con
->in_msg_pos
.page_pos
));
1720 BUG_ON(pages
== NULL
);
1721 p
= kmap(pages
[con
->in_msg_pos
.page
]);
1722 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ con
->in_msg_pos
.page_pos
,
1724 if (ret
> 0 && do_datacrc
)
1726 crc32c(con
->in_data_crc
,
1727 p
+ con
->in_msg_pos
.page_pos
, ret
);
1728 kunmap(pages
[con
->in_msg_pos
.page
]);
1731 con
->in_msg_pos
.data_pos
+= ret
;
1732 con
->in_msg_pos
.page_pos
+= ret
;
1733 if (con
->in_msg_pos
.page_pos
== PAGE_SIZE
) {
1734 con
->in_msg_pos
.page_pos
= 0;
1735 con
->in_msg_pos
.page
++;
1742 static int read_partial_message_bio(struct ceph_connection
*con
,
1743 struct bio
**bio_iter
, int *bio_seg
,
1744 unsigned int data_len
, bool do_datacrc
)
1746 struct bio_vec
*bv
= bio_iovec_idx(*bio_iter
, *bio_seg
);
1750 left
= min((int)(data_len
- con
->in_msg_pos
.data_pos
),
1751 (int)(bv
->bv_len
- con
->in_msg_pos
.page_pos
));
1753 p
= kmap(bv
->bv_page
) + bv
->bv_offset
;
1755 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ con
->in_msg_pos
.page_pos
,
1757 if (ret
> 0 && do_datacrc
)
1759 crc32c(con
->in_data_crc
,
1760 p
+ con
->in_msg_pos
.page_pos
, ret
);
1761 kunmap(bv
->bv_page
);
1764 con
->in_msg_pos
.data_pos
+= ret
;
1765 con
->in_msg_pos
.page_pos
+= ret
;
1766 if (con
->in_msg_pos
.page_pos
== bv
->bv_len
) {
1767 con
->in_msg_pos
.page_pos
= 0;
1768 iter_bio_next(bio_iter
, bio_seg
);
1776 * read (part of) a message.
1778 static int read_partial_message(struct ceph_connection
*con
)
1780 struct ceph_msg
*m
= con
->in_msg
;
1784 unsigned int front_len
, middle_len
, data_len
;
1785 bool do_datacrc
= !con
->msgr
->nocrc
;
1789 dout("read_partial_message con %p msg %p\n", con
, m
);
1792 size
= sizeof (con
->in_hdr
);
1794 ret
= read_partial(con
, end
, size
, &con
->in_hdr
);
1798 crc
= crc32c(0, &con
->in_hdr
, offsetof(struct ceph_msg_header
, crc
));
1799 if (cpu_to_le32(crc
) != con
->in_hdr
.crc
) {
1800 pr_err("read_partial_message bad hdr "
1801 " crc %u != expected %u\n",
1802 crc
, con
->in_hdr
.crc
);
1806 front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
1807 if (front_len
> CEPH_MSG_MAX_FRONT_LEN
)
1809 middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
1810 if (middle_len
> CEPH_MSG_MAX_DATA_LEN
)
1812 data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
1813 if (data_len
> CEPH_MSG_MAX_DATA_LEN
)
1817 seq
= le64_to_cpu(con
->in_hdr
.seq
);
1818 if ((s64
)seq
- (s64
)con
->in_seq
< 1) {
1819 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1820 ENTITY_NAME(con
->peer_name
),
1821 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1822 seq
, con
->in_seq
+ 1);
1823 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1825 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1827 } else if ((s64
)seq
- (s64
)con
->in_seq
> 1) {
1828 pr_err("read_partial_message bad seq %lld expected %lld\n",
1829 seq
, con
->in_seq
+ 1);
1830 con
->error_msg
= "bad message sequence # for incoming message";
1834 /* allocate message? */
1836 dout("got hdr type %d front %d data %d\n", con
->in_hdr
.type
,
1837 con
->in_hdr
.front_len
, con
->in_hdr
.data_len
);
1838 if (ceph_con_in_msg_alloc(con
, &con
->in_hdr
)) {
1839 /* skip this message */
1840 dout("alloc_msg said skip message\n");
1841 BUG_ON(con
->in_msg
);
1842 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1844 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1850 "error allocating memory for incoming message";
1854 BUG_ON(con
->in_msg
->con
!= con
);
1856 m
->front
.iov_len
= 0; /* haven't read it yet */
1858 m
->middle
->vec
.iov_len
= 0;
1860 con
->in_msg_pos
.page
= 0;
1862 con
->in_msg_pos
.page_pos
= m
->page_alignment
;
1864 con
->in_msg_pos
.page_pos
= 0;
1865 con
->in_msg_pos
.data_pos
= 0;
1869 ret
= read_partial_message_section(con
, &m
->front
, front_len
,
1870 &con
->in_front_crc
);
1876 ret
= read_partial_message_section(con
, &m
->middle
->vec
,
1878 &con
->in_middle_crc
);
1883 if (m
->bio
&& !m
->bio_iter
)
1884 init_bio_iter(m
->bio
, &m
->bio_iter
, &m
->bio_seg
);
1888 while (con
->in_msg_pos
.data_pos
< data_len
) {
1890 ret
= read_partial_message_pages(con
, m
->pages
,
1891 data_len
, do_datacrc
);
1895 } else if (m
->bio
) {
1897 ret
= read_partial_message_bio(con
,
1898 &m
->bio_iter
, &m
->bio_seg
,
1899 data_len
, do_datacrc
);
1909 size
= sizeof (m
->footer
);
1911 ret
= read_partial(con
, end
, size
, &m
->footer
);
1915 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1916 m
, front_len
, m
->footer
.front_crc
, middle_len
,
1917 m
->footer
.middle_crc
, data_len
, m
->footer
.data_crc
);
1920 if (con
->in_front_crc
!= le32_to_cpu(m
->footer
.front_crc
)) {
1921 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1922 m
, con
->in_front_crc
, m
->footer
.front_crc
);
1925 if (con
->in_middle_crc
!= le32_to_cpu(m
->footer
.middle_crc
)) {
1926 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1927 m
, con
->in_middle_crc
, m
->footer
.middle_crc
);
1931 (m
->footer
.flags
& CEPH_MSG_FOOTER_NOCRC
) == 0 &&
1932 con
->in_data_crc
!= le32_to_cpu(m
->footer
.data_crc
)) {
1933 pr_err("read_partial_message %p data crc %u != exp. %u\n", m
,
1934 con
->in_data_crc
, le32_to_cpu(m
->footer
.data_crc
));
1938 return 1; /* done! */
1942 * Process message. This happens in the worker thread. The callback should
1943 * be careful not to do anything that waits on other incoming messages or it
1946 static void process_message(struct ceph_connection
*con
)
1948 struct ceph_msg
*msg
;
1950 BUG_ON(con
->in_msg
->con
!= con
);
1951 con
->in_msg
->con
= NULL
;
1956 /* if first message, set peer_name */
1957 if (con
->peer_name
.type
== 0)
1958 con
->peer_name
= msg
->hdr
.src
;
1961 mutex_unlock(&con
->mutex
);
1963 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
1964 msg
, le64_to_cpu(msg
->hdr
.seq
),
1965 ENTITY_NAME(msg
->hdr
.src
),
1966 le16_to_cpu(msg
->hdr
.type
),
1967 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
1968 le32_to_cpu(msg
->hdr
.front_len
),
1969 le32_to_cpu(msg
->hdr
.data_len
),
1970 con
->in_front_crc
, con
->in_middle_crc
, con
->in_data_crc
);
1971 con
->ops
->dispatch(con
, msg
);
1973 mutex_lock(&con
->mutex
);
1974 prepare_read_tag(con
);
1979 * Write something to the socket. Called in a worker thread when the
1980 * socket appears to be writeable and we have something ready to send.
1982 static int try_write(struct ceph_connection
*con
)
1986 dout("try_write start %p state %lu\n", con
, con
->state
);
1989 dout("try_write out_kvec_bytes %d\n", con
->out_kvec_bytes
);
1991 /* open the socket first? */
1992 if (con
->sock
== NULL
) {
1993 set_bit(CONNECTING
, &con
->state
);
1995 con_out_kvec_reset(con
);
1996 prepare_write_banner(con
);
1997 prepare_read_banner(con
);
1999 BUG_ON(con
->in_msg
);
2000 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2001 dout("try_write initiating connect on %p new state %lu\n",
2003 ret
= ceph_tcp_connect(con
);
2005 con
->error_msg
= "connect error";
2011 /* kvec data queued? */
2012 if (con
->out_skip
) {
2013 ret
= write_partial_skip(con
);
2017 if (con
->out_kvec_left
) {
2018 ret
= write_partial_kvec(con
);
2025 if (con
->out_msg_done
) {
2026 ceph_msg_put(con
->out_msg
);
2027 con
->out_msg
= NULL
; /* we're done with this one */
2031 ret
= write_partial_msg_pages(con
);
2033 goto more_kvec
; /* we need to send the footer, too! */
2037 dout("try_write write_partial_msg_pages err %d\n",
2044 if (!test_bit(CONNECTING
, &con
->state
) &&
2045 !test_bit(NEGOTIATING
, &con
->state
)) {
2046 /* is anything else pending? */
2047 if (!list_empty(&con
->out_queue
)) {
2048 prepare_write_message(con
);
2051 if (con
->in_seq
> con
->in_seq_acked
) {
2052 prepare_write_ack(con
);
2055 if (test_and_clear_bit(KEEPALIVE_PENDING
, &con
->flags
)) {
2056 prepare_write_keepalive(con
);
2061 /* Nothing to do! */
2062 clear_bit(WRITE_PENDING
, &con
->flags
);
2063 dout("try_write nothing else to write.\n");
2066 dout("try_write done on %p ret %d\n", con
, ret
);
2073 * Read what we can from the socket.
2075 static int try_read(struct ceph_connection
*con
)
2082 if (test_bit(STANDBY
, &con
->state
))
2085 dout("try_read start on %p\n", con
);
2088 dout("try_read tag %d in_base_pos %d\n", (int)con
->in_tag
,
2092 * process_connect and process_message drop and re-take
2093 * con->mutex. make sure we handle a racing close or reopen.
2095 if (test_bit(CLOSED
, &con
->state
) ||
2096 test_bit(OPENING
, &con
->state
)) {
2101 if (test_bit(CONNECTING
, &con
->state
)) {
2102 dout("try_read connecting\n");
2103 ret
= read_partial_banner(con
);
2106 ret
= process_banner(con
);
2110 clear_bit(CONNECTING
, &con
->state
);
2111 set_bit(NEGOTIATING
, &con
->state
);
2113 /* Banner is good, exchange connection info */
2114 ret
= prepare_write_connect(con
);
2117 prepare_read_connect(con
);
2119 /* Send connection info before awaiting response */
2123 if (test_bit(NEGOTIATING
, &con
->state
)) {
2124 dout("try_read negotiating\n");
2125 ret
= read_partial_connect(con
);
2128 ret
= process_connect(con
);
2134 if (con
->in_base_pos
< 0) {
2136 * skipping + discarding content.
2138 * FIXME: there must be a better way to do this!
2140 static char buf
[SKIP_BUF_SIZE
];
2141 int skip
= min((int) sizeof (buf
), -con
->in_base_pos
);
2143 dout("skipping %d / %d bytes\n", skip
, -con
->in_base_pos
);
2144 ret
= ceph_tcp_recvmsg(con
->sock
, buf
, skip
);
2147 con
->in_base_pos
+= ret
;
2148 if (con
->in_base_pos
)
2151 if (con
->in_tag
== CEPH_MSGR_TAG_READY
) {
2155 ret
= ceph_tcp_recvmsg(con
->sock
, &con
->in_tag
, 1);
2158 dout("try_read got tag %d\n", (int)con
->in_tag
);
2159 switch (con
->in_tag
) {
2160 case CEPH_MSGR_TAG_MSG
:
2161 prepare_read_message(con
);
2163 case CEPH_MSGR_TAG_ACK
:
2164 prepare_read_ack(con
);
2166 case CEPH_MSGR_TAG_CLOSE
:
2167 clear_bit(CONNECTED
, &con
->state
);
2168 set_bit(CLOSED
, &con
->state
); /* fixme */
2174 if (con
->in_tag
== CEPH_MSGR_TAG_MSG
) {
2175 ret
= read_partial_message(con
);
2179 con
->error_msg
= "bad crc";
2183 con
->error_msg
= "io error";
2188 if (con
->in_tag
== CEPH_MSGR_TAG_READY
)
2190 process_message(con
);
2193 if (con
->in_tag
== CEPH_MSGR_TAG_ACK
) {
2194 ret
= read_partial_ack(con
);
2202 dout("try_read done on %p ret %d\n", con
, ret
);
2206 pr_err("try_read bad con->in_tag = %d\n", (int)con
->in_tag
);
2207 con
->error_msg
= "protocol error, garbage tag";
2214 * Atomically queue work on a connection. Bump @con reference to
2215 * avoid races with connection teardown.
2217 static void queue_con(struct ceph_connection
*con
)
2219 if (!con
->ops
->get(con
)) {
2220 dout("queue_con %p ref count 0\n", con
);
2224 if (!queue_delayed_work(ceph_msgr_wq
, &con
->work
, 0)) {
2225 dout("queue_con %p - already queued\n", con
);
2228 dout("queue_con %p\n", con
);
2233 * Do some work on a connection. Drop a connection ref when we're done.
2235 static void con_work(struct work_struct
*work
)
2237 struct ceph_connection
*con
= container_of(work
, struct ceph_connection
,
2241 mutex_lock(&con
->mutex
);
2243 if (test_and_clear_bit(SOCK_CLOSED
, &con
->flags
)) {
2244 if (test_and_clear_bit(CONNECTED
, &con
->state
))
2245 con
->error_msg
= "socket closed";
2246 else if (test_and_clear_bit(NEGOTIATING
, &con
->state
))
2247 con
->error_msg
= "negotiation failed";
2248 else if (test_and_clear_bit(CONNECTING
, &con
->state
))
2249 con
->error_msg
= "connection failed";
2251 con
->error_msg
= "unrecognized con state";
2255 if (test_and_clear_bit(BACKOFF
, &con
->flags
)) {
2256 dout("con_work %p backing off\n", con
);
2257 if (queue_delayed_work(ceph_msgr_wq
, &con
->work
,
2258 round_jiffies_relative(con
->delay
))) {
2259 dout("con_work %p backoff %lu\n", con
, con
->delay
);
2260 mutex_unlock(&con
->mutex
);
2264 dout("con_work %p FAILED to back off %lu\n", con
,
2269 if (test_bit(STANDBY
, &con
->state
)) {
2270 dout("con_work %p STANDBY\n", con
);
2273 if (test_bit(CLOSED
, &con
->state
)) { /* e.g. if we are replaced */
2274 dout("con_work CLOSED\n");
2275 con_close_socket(con
);
2278 if (test_and_clear_bit(OPENING
, &con
->state
)) {
2279 /* reopen w/ new peer */
2280 dout("con_work OPENING\n");
2281 con_close_socket(con
);
2284 ret
= try_read(con
);
2290 ret
= try_write(con
);
2297 mutex_unlock(&con
->mutex
);
2303 mutex_unlock(&con
->mutex
);
2304 ceph_fault(con
); /* error/fault path */
2310 * Generic error/fault handler. A retry mechanism is used with
2311 * exponential backoff
2313 static void ceph_fault(struct ceph_connection
*con
)
2315 pr_err("%s%lld %s %s\n", ENTITY_NAME(con
->peer_name
),
2316 ceph_pr_addr(&con
->peer_addr
.in_addr
), con
->error_msg
);
2317 dout("fault %p state %lu to peer %s\n",
2318 con
, con
->state
, ceph_pr_addr(&con
->peer_addr
.in_addr
));
2320 if (test_bit(LOSSYTX
, &con
->flags
)) {
2321 dout("fault on LOSSYTX channel\n");
2325 mutex_lock(&con
->mutex
);
2326 if (test_bit(CLOSED
, &con
->state
))
2329 con_close_socket(con
);
2332 BUG_ON(con
->in_msg
->con
!= con
);
2333 con
->in_msg
->con
= NULL
;
2334 ceph_msg_put(con
->in_msg
);
2339 /* Requeue anything that hasn't been acked */
2340 list_splice_init(&con
->out_sent
, &con
->out_queue
);
2342 /* If there are no messages queued or keepalive pending, place
2343 * the connection in a STANDBY state */
2344 if (list_empty(&con
->out_queue
) &&
2345 !test_bit(KEEPALIVE_PENDING
, &con
->flags
)) {
2346 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con
);
2347 clear_bit(WRITE_PENDING
, &con
->flags
);
2348 set_bit(STANDBY
, &con
->state
);
2350 /* retry after a delay. */
2351 if (con
->delay
== 0)
2352 con
->delay
= BASE_DELAY_INTERVAL
;
2353 else if (con
->delay
< MAX_DELAY_INTERVAL
)
2356 if (queue_delayed_work(ceph_msgr_wq
, &con
->work
,
2357 round_jiffies_relative(con
->delay
))) {
2358 dout("fault queued %p delay %lu\n", con
, con
->delay
);
2361 dout("fault failed to queue %p delay %lu, backoff\n",
2364 * In many cases we see a socket state change
2365 * while con_work is running and end up
2366 * queuing (non-delayed) work, such that we
2367 * can't backoff with a delay. Set a flag so
2368 * that when con_work restarts we schedule the
2371 set_bit(BACKOFF
, &con
->flags
);
2376 mutex_unlock(&con
->mutex
);
2379 * in case we faulted due to authentication, invalidate our
2380 * current tickets so that we can get new ones.
2382 if (con
->auth_retry
&& con
->ops
->invalidate_authorizer
) {
2383 dout("calling invalidate_authorizer()\n");
2384 con
->ops
->invalidate_authorizer(con
);
2387 if (con
->ops
->fault
)
2388 con
->ops
->fault(con
);
2394 * initialize a new messenger instance
2396 void ceph_messenger_init(struct ceph_messenger
*msgr
,
2397 struct ceph_entity_addr
*myaddr
,
2398 u32 supported_features
,
2399 u32 required_features
,
2402 msgr
->supported_features
= supported_features
;
2403 msgr
->required_features
= required_features
;
2405 spin_lock_init(&msgr
->global_seq_lock
);
2408 msgr
->inst
.addr
= *myaddr
;
2410 /* select a random nonce */
2411 msgr
->inst
.addr
.type
= 0;
2412 get_random_bytes(&msgr
->inst
.addr
.nonce
, sizeof(msgr
->inst
.addr
.nonce
));
2413 encode_my_addr(msgr
);
2414 msgr
->nocrc
= nocrc
;
2416 dout("%s %p\n", __func__
, msgr
);
2418 EXPORT_SYMBOL(ceph_messenger_init
);
2420 static void clear_standby(struct ceph_connection
*con
)
2422 /* come back from STANDBY? */
2423 if (test_and_clear_bit(STANDBY
, &con
->state
)) {
2424 mutex_lock(&con
->mutex
);
2425 dout("clear_standby %p and ++connect_seq\n", con
);
2427 WARN_ON(test_bit(WRITE_PENDING
, &con
->flags
));
2428 WARN_ON(test_bit(KEEPALIVE_PENDING
, &con
->flags
));
2429 mutex_unlock(&con
->mutex
);
2434 * Queue up an outgoing message on the given connection.
2436 void ceph_con_send(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2438 if (test_bit(CLOSED
, &con
->state
)) {
2439 dout("con_send %p closed, dropping %p\n", con
, msg
);
2445 msg
->hdr
.src
= con
->msgr
->inst
.name
;
2447 BUG_ON(msg
->front
.iov_len
!= le32_to_cpu(msg
->hdr
.front_len
));
2449 msg
->needs_out_seq
= true;
2452 mutex_lock(&con
->mutex
);
2454 BUG_ON(msg
->con
!= NULL
);
2455 msg
->con
= con
->ops
->get(con
);
2456 BUG_ON(msg
->con
== NULL
);
2458 BUG_ON(!list_empty(&msg
->list_head
));
2459 list_add_tail(&msg
->list_head
, &con
->out_queue
);
2460 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg
,
2461 ENTITY_NAME(con
->peer_name
), le16_to_cpu(msg
->hdr
.type
),
2462 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2463 le32_to_cpu(msg
->hdr
.front_len
),
2464 le32_to_cpu(msg
->hdr
.middle_len
),
2465 le32_to_cpu(msg
->hdr
.data_len
));
2466 mutex_unlock(&con
->mutex
);
2468 /* if there wasn't anything waiting to send before, queue
2471 if (test_and_set_bit(WRITE_PENDING
, &con
->flags
) == 0)
2474 EXPORT_SYMBOL(ceph_con_send
);
2477 * Revoke a message that was previously queued for send
2479 void ceph_msg_revoke(struct ceph_msg
*msg
)
2481 struct ceph_connection
*con
= msg
->con
;
2484 return; /* Message not in our possession */
2486 mutex_lock(&con
->mutex
);
2487 if (!list_empty(&msg
->list_head
)) {
2488 dout("%s %p msg %p - was on queue\n", __func__
, con
, msg
);
2489 list_del_init(&msg
->list_head
);
2490 BUG_ON(msg
->con
== NULL
);
2491 msg
->con
->ops
->put(msg
->con
);
2497 if (con
->out_msg
== msg
) {
2498 dout("%s %p msg %p - was sending\n", __func__
, con
, msg
);
2499 con
->out_msg
= NULL
;
2500 if (con
->out_kvec_is_msg
) {
2501 con
->out_skip
= con
->out_kvec_bytes
;
2502 con
->out_kvec_is_msg
= false;
2508 mutex_unlock(&con
->mutex
);
2512 * Revoke a message that we may be reading data into
2514 void ceph_msg_revoke_incoming(struct ceph_msg
*msg
)
2516 struct ceph_connection
*con
;
2518 BUG_ON(msg
== NULL
);
2520 dout("%s msg %p null con\n", __func__
, msg
);
2522 return; /* Message not in our possession */
2526 mutex_lock(&con
->mutex
);
2527 if (con
->in_msg
== msg
) {
2528 unsigned int front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
2529 unsigned int middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
2530 unsigned int data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
2532 /* skip rest of message */
2533 dout("%s %p msg %p revoked\n", __func__
, con
, msg
);
2534 con
->in_base_pos
= con
->in_base_pos
-
2535 sizeof(struct ceph_msg_header
) -
2539 sizeof(struct ceph_msg_footer
);
2540 ceph_msg_put(con
->in_msg
);
2542 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2545 dout("%s %p in_msg %p msg %p no-op\n",
2546 __func__
, con
, con
->in_msg
, msg
);
2548 mutex_unlock(&con
->mutex
);
2552 * Queue a keepalive byte to ensure the tcp connection is alive.
2554 void ceph_con_keepalive(struct ceph_connection
*con
)
2556 dout("con_keepalive %p\n", con
);
2558 if (test_and_set_bit(KEEPALIVE_PENDING
, &con
->flags
) == 0 &&
2559 test_and_set_bit(WRITE_PENDING
, &con
->flags
) == 0)
2562 EXPORT_SYMBOL(ceph_con_keepalive
);
2566 * construct a new message with given type, size
2567 * the new msg has a ref count of 1.
2569 struct ceph_msg
*ceph_msg_new(int type
, int front_len
, gfp_t flags
,
2574 m
= kmalloc(sizeof(*m
), flags
);
2577 kref_init(&m
->kref
);
2580 INIT_LIST_HEAD(&m
->list_head
);
2583 m
->hdr
.type
= cpu_to_le16(type
);
2584 m
->hdr
.priority
= cpu_to_le16(CEPH_MSG_PRIO_DEFAULT
);
2586 m
->hdr
.front_len
= cpu_to_le32(front_len
);
2587 m
->hdr
.middle_len
= 0;
2588 m
->hdr
.data_len
= 0;
2589 m
->hdr
.data_off
= 0;
2590 m
->hdr
.reserved
= 0;
2591 m
->footer
.front_crc
= 0;
2592 m
->footer
.middle_crc
= 0;
2593 m
->footer
.data_crc
= 0;
2594 m
->footer
.flags
= 0;
2595 m
->front_max
= front_len
;
2596 m
->front_is_vmalloc
= false;
2597 m
->more_to_follow
= false;
2606 m
->page_alignment
= 0;
2616 if (front_len
> PAGE_CACHE_SIZE
) {
2617 m
->front
.iov_base
= __vmalloc(front_len
, flags
,
2619 m
->front_is_vmalloc
= true;
2621 m
->front
.iov_base
= kmalloc(front_len
, flags
);
2623 if (m
->front
.iov_base
== NULL
) {
2624 dout("ceph_msg_new can't allocate %d bytes\n",
2629 m
->front
.iov_base
= NULL
;
2631 m
->front
.iov_len
= front_len
;
2633 dout("ceph_msg_new %p front %d\n", m
, front_len
);
2640 pr_err("msg_new can't create type %d front %d\n", type
,
2644 dout("msg_new can't create type %d front %d\n", type
,
2649 EXPORT_SYMBOL(ceph_msg_new
);
2652 * Allocate "middle" portion of a message, if it is needed and wasn't
2653 * allocated by alloc_msg. This allows us to read a small fixed-size
2654 * per-type header in the front and then gracefully fail (i.e.,
2655 * propagate the error to the caller based on info in the front) when
2656 * the middle is too large.
2658 static int ceph_alloc_middle(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2660 int type
= le16_to_cpu(msg
->hdr
.type
);
2661 int middle_len
= le32_to_cpu(msg
->hdr
.middle_len
);
2663 dout("alloc_middle %p type %d %s middle_len %d\n", msg
, type
,
2664 ceph_msg_type_name(type
), middle_len
);
2665 BUG_ON(!middle_len
);
2666 BUG_ON(msg
->middle
);
2668 msg
->middle
= ceph_buffer_new(middle_len
, GFP_NOFS
);
2675 * Allocate a message for receiving an incoming message on a
2676 * connection, and save the result in con->in_msg. Uses the
2677 * connection's private alloc_msg op if available.
2679 * Returns true if the message should be skipped, false otherwise.
2680 * If true is returned (skip message), con->in_msg will be NULL.
2681 * If false is returned, con->in_msg will contain a pointer to the
2682 * newly-allocated message, or NULL in case of memory exhaustion.
2684 static bool ceph_con_in_msg_alloc(struct ceph_connection
*con
,
2685 struct ceph_msg_header
*hdr
)
2687 int type
= le16_to_cpu(hdr
->type
);
2688 int front_len
= le32_to_cpu(hdr
->front_len
);
2689 int middle_len
= le32_to_cpu(hdr
->middle_len
);
2692 BUG_ON(con
->in_msg
!= NULL
);
2694 if (con
->ops
->alloc_msg
) {
2697 mutex_unlock(&con
->mutex
);
2698 con
->in_msg
= con
->ops
->alloc_msg(con
, hdr
, &skip
);
2699 mutex_lock(&con
->mutex
);
2701 con
->in_msg
->con
= con
->ops
->get(con
);
2702 BUG_ON(con
->in_msg
->con
== NULL
);
2711 con
->in_msg
= ceph_msg_new(type
, front_len
, GFP_NOFS
, false);
2713 pr_err("unable to allocate msg type %d len %d\n",
2717 con
->in_msg
->con
= con
->ops
->get(con
);
2718 BUG_ON(con
->in_msg
->con
== NULL
);
2719 con
->in_msg
->page_alignment
= le16_to_cpu(hdr
->data_off
);
2721 memcpy(&con
->in_msg
->hdr
, &con
->in_hdr
, sizeof(con
->in_hdr
));
2723 if (middle_len
&& !con
->in_msg
->middle
) {
2724 ret
= ceph_alloc_middle(con
, con
->in_msg
);
2726 ceph_msg_put(con
->in_msg
);
2736 * Free a generically kmalloc'd message.
2738 void ceph_msg_kfree(struct ceph_msg
*m
)
2740 dout("msg_kfree %p\n", m
);
2741 if (m
->front_is_vmalloc
)
2742 vfree(m
->front
.iov_base
);
2744 kfree(m
->front
.iov_base
);
2749 * Drop a msg ref. Destroy as needed.
2751 void ceph_msg_last_put(struct kref
*kref
)
2753 struct ceph_msg
*m
= container_of(kref
, struct ceph_msg
, kref
);
2755 dout("ceph_msg_put last one on %p\n", m
);
2756 WARN_ON(!list_empty(&m
->list_head
));
2758 /* drop middle, data, if any */
2760 ceph_buffer_put(m
->middle
);
2767 ceph_pagelist_release(m
->pagelist
);
2775 ceph_msgpool_put(m
->pool
, m
);
2779 EXPORT_SYMBOL(ceph_msg_last_put
);
2781 void ceph_msg_dump(struct ceph_msg
*msg
)
2783 pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg
,
2784 msg
->front_max
, msg
->nr_pages
);
2785 print_hex_dump(KERN_DEBUG
, "header: ",
2786 DUMP_PREFIX_OFFSET
, 16, 1,
2787 &msg
->hdr
, sizeof(msg
->hdr
), true);
2788 print_hex_dump(KERN_DEBUG
, " front: ",
2789 DUMP_PREFIX_OFFSET
, 16, 1,
2790 msg
->front
.iov_base
, msg
->front
.iov_len
, true);
2792 print_hex_dump(KERN_DEBUG
, "middle: ",
2793 DUMP_PREFIX_OFFSET
, 16, 1,
2794 msg
->middle
->vec
.iov_base
,
2795 msg
->middle
->vec
.iov_len
, true);
2796 print_hex_dump(KERN_DEBUG
, "footer: ",
2797 DUMP_PREFIX_OFFSET
, 16, 1,
2798 &msg
->footer
, sizeof(msg
->footer
), true);
2800 EXPORT_SYMBOL(ceph_msg_dump
);