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>
13 #include <linux/bio.h>
14 #endif /* CONFIG_BLOCK */
15 #include <linux/dns_resolver.h>
18 #include <linux/ceph/libceph.h>
19 #include <linux/ceph/messenger.h>
20 #include <linux/ceph/decode.h>
21 #include <linux/ceph/pagelist.h>
22 #include <linux/export.h>
25 * Ceph uses the messenger to exchange ceph_msg messages with other
26 * hosts in the system. The messenger provides ordered and reliable
27 * delivery. We tolerate TCP disconnects by reconnecting (with
28 * exponential backoff) in the case of a fault (disconnection, bad
29 * crc, protocol error). Acks allow sent messages to be discarded by
34 * We track the state of the socket on a given connection using
35 * values defined below. The transition to a new socket state is
36 * handled by a function which verifies we aren't coming from an
40 * | NEW* | transient initial state
42 * | con_sock_state_init()
45 * | CLOSED | initialized, but no socket (and no
46 * ---------- TCP connection)
48 * | \ con_sock_state_connecting()
49 * | ----------------------
51 * + con_sock_state_closed() \
52 * |+--------------------------- \
55 * | | CLOSING | socket event; \ \
56 * | ----------- await close \ \
59 * | + con_sock_state_closing() \ |
61 * | / --------------- | |
64 * | / -----------------| CONNECTING | socket created, TCP
65 * | | / -------------- connect initiated
66 * | | | con_sock_state_connected()
69 * | CONNECTED | TCP connection established
72 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
75 #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
76 #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
77 #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
78 #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
79 #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
84 #define CON_STATE_CLOSED 1 /* -> PREOPEN */
85 #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
86 #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
87 #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
88 #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
89 #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
92 * ceph_connection flag bits
94 #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
95 * messages on errors */
96 #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
97 #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
98 #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
99 #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
101 static bool con_flag_valid(unsigned long con_flag
)
104 case CON_FLAG_LOSSYTX
:
105 case CON_FLAG_KEEPALIVE_PENDING
:
106 case CON_FLAG_WRITE_PENDING
:
107 case CON_FLAG_SOCK_CLOSED
:
108 case CON_FLAG_BACKOFF
:
115 static void con_flag_clear(struct ceph_connection
*con
, unsigned long con_flag
)
117 BUG_ON(!con_flag_valid(con_flag
));
119 clear_bit(con_flag
, &con
->flags
);
122 static void con_flag_set(struct ceph_connection
*con
, unsigned long con_flag
)
124 BUG_ON(!con_flag_valid(con_flag
));
126 set_bit(con_flag
, &con
->flags
);
129 static bool con_flag_test(struct ceph_connection
*con
, unsigned long con_flag
)
131 BUG_ON(!con_flag_valid(con_flag
));
133 return test_bit(con_flag
, &con
->flags
);
136 static bool con_flag_test_and_clear(struct ceph_connection
*con
,
137 unsigned long con_flag
)
139 BUG_ON(!con_flag_valid(con_flag
));
141 return test_and_clear_bit(con_flag
, &con
->flags
);
144 static bool con_flag_test_and_set(struct ceph_connection
*con
,
145 unsigned long con_flag
)
147 BUG_ON(!con_flag_valid(con_flag
));
149 return test_and_set_bit(con_flag
, &con
->flags
);
152 /* static tag bytes (protocol control messages) */
153 static char tag_msg
= CEPH_MSGR_TAG_MSG
;
154 static char tag_ack
= CEPH_MSGR_TAG_ACK
;
155 static char tag_keepalive
= CEPH_MSGR_TAG_KEEPALIVE
;
157 #ifdef CONFIG_LOCKDEP
158 static struct lock_class_key socket_class
;
162 * When skipping (ignoring) a block of input we read it into a "skip
163 * buffer," which is this many bytes in size.
165 #define SKIP_BUF_SIZE 1024
167 static void queue_con(struct ceph_connection
*con
);
168 static void con_work(struct work_struct
*);
169 static void con_fault(struct ceph_connection
*con
);
172 * Nicely render a sockaddr as a string. An array of formatted
173 * strings is used, to approximate reentrancy.
175 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
176 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
177 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
178 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
180 static char addr_str
[ADDR_STR_COUNT
][MAX_ADDR_STR_LEN
];
181 static atomic_t addr_str_seq
= ATOMIC_INIT(0);
183 static struct page
*zero_page
; /* used in certain error cases */
185 const char *ceph_pr_addr(const struct sockaddr_storage
*ss
)
189 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
190 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
192 i
= atomic_inc_return(&addr_str_seq
) & ADDR_STR_COUNT_MASK
;
195 switch (ss
->ss_family
) {
197 snprintf(s
, MAX_ADDR_STR_LEN
, "%pI4:%hu", &in4
->sin_addr
,
198 ntohs(in4
->sin_port
));
202 snprintf(s
, MAX_ADDR_STR_LEN
, "[%pI6c]:%hu", &in6
->sin6_addr
,
203 ntohs(in6
->sin6_port
));
207 snprintf(s
, MAX_ADDR_STR_LEN
, "(unknown sockaddr family %hu)",
213 EXPORT_SYMBOL(ceph_pr_addr
);
215 static void encode_my_addr(struct ceph_messenger
*msgr
)
217 memcpy(&msgr
->my_enc_addr
, &msgr
->inst
.addr
, sizeof(msgr
->my_enc_addr
));
218 ceph_encode_addr(&msgr
->my_enc_addr
);
222 * work queue for all reading and writing to/from the socket.
224 static struct workqueue_struct
*ceph_msgr_wq
;
226 static void _ceph_msgr_exit(void)
229 destroy_workqueue(ceph_msgr_wq
);
233 BUG_ON(zero_page
== NULL
);
235 page_cache_release(zero_page
);
239 int ceph_msgr_init(void)
241 BUG_ON(zero_page
!= NULL
);
242 zero_page
= ZERO_PAGE(0);
243 page_cache_get(zero_page
);
245 ceph_msgr_wq
= alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT
, 0);
249 pr_err("msgr_init failed to create workqueue\n");
254 EXPORT_SYMBOL(ceph_msgr_init
);
256 void ceph_msgr_exit(void)
258 BUG_ON(ceph_msgr_wq
== NULL
);
262 EXPORT_SYMBOL(ceph_msgr_exit
);
264 void ceph_msgr_flush(void)
266 flush_workqueue(ceph_msgr_wq
);
268 EXPORT_SYMBOL(ceph_msgr_flush
);
270 /* Connection socket state transition functions */
272 static void con_sock_state_init(struct ceph_connection
*con
)
276 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSED
);
277 if (WARN_ON(old_state
!= CON_SOCK_STATE_NEW
))
278 printk("%s: unexpected old state %d\n", __func__
, old_state
);
279 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
280 CON_SOCK_STATE_CLOSED
);
283 static void con_sock_state_connecting(struct ceph_connection
*con
)
287 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CONNECTING
);
288 if (WARN_ON(old_state
!= CON_SOCK_STATE_CLOSED
))
289 printk("%s: unexpected old state %d\n", __func__
, old_state
);
290 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
291 CON_SOCK_STATE_CONNECTING
);
294 static void con_sock_state_connected(struct ceph_connection
*con
)
298 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CONNECTED
);
299 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTING
))
300 printk("%s: unexpected old state %d\n", __func__
, old_state
);
301 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
302 CON_SOCK_STATE_CONNECTED
);
305 static void con_sock_state_closing(struct ceph_connection
*con
)
309 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSING
);
310 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTING
&&
311 old_state
!= CON_SOCK_STATE_CONNECTED
&&
312 old_state
!= CON_SOCK_STATE_CLOSING
))
313 printk("%s: unexpected old state %d\n", __func__
, old_state
);
314 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
315 CON_SOCK_STATE_CLOSING
);
318 static void con_sock_state_closed(struct ceph_connection
*con
)
322 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSED
);
323 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTED
&&
324 old_state
!= CON_SOCK_STATE_CLOSING
&&
325 old_state
!= CON_SOCK_STATE_CONNECTING
&&
326 old_state
!= CON_SOCK_STATE_CLOSED
))
327 printk("%s: unexpected old state %d\n", __func__
, old_state
);
328 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
329 CON_SOCK_STATE_CLOSED
);
333 * socket callback functions
336 /* data available on socket, or listen socket received a connect */
337 static void ceph_sock_data_ready(struct sock
*sk
, int count_unused
)
339 struct ceph_connection
*con
= sk
->sk_user_data
;
340 if (atomic_read(&con
->msgr
->stopping
)) {
344 if (sk
->sk_state
!= TCP_CLOSE_WAIT
) {
345 dout("%s on %p state = %lu, queueing work\n", __func__
,
351 /* socket has buffer space for writing */
352 static void ceph_sock_write_space(struct sock
*sk
)
354 struct ceph_connection
*con
= sk
->sk_user_data
;
356 /* only queue to workqueue if there is data we want to write,
357 * and there is sufficient space in the socket buffer to accept
358 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
359 * doesn't get called again until try_write() fills the socket
360 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
361 * and net/core/stream.c:sk_stream_write_space().
363 if (con_flag_test(con
, CON_FLAG_WRITE_PENDING
)) {
364 if (sk_stream_wspace(sk
) >= sk_stream_min_wspace(sk
)) {
365 dout("%s %p queueing write work\n", __func__
, con
);
366 clear_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
370 dout("%s %p nothing to write\n", __func__
, con
);
374 /* socket's state has changed */
375 static void ceph_sock_state_change(struct sock
*sk
)
377 struct ceph_connection
*con
= sk
->sk_user_data
;
379 dout("%s %p state = %lu sk_state = %u\n", __func__
,
380 con
, con
->state
, sk
->sk_state
);
382 switch (sk
->sk_state
) {
384 dout("%s TCP_CLOSE\n", __func__
);
386 dout("%s TCP_CLOSE_WAIT\n", __func__
);
387 con_sock_state_closing(con
);
388 con_flag_set(con
, CON_FLAG_SOCK_CLOSED
);
391 case TCP_ESTABLISHED
:
392 dout("%s TCP_ESTABLISHED\n", __func__
);
393 con_sock_state_connected(con
);
396 default: /* Everything else is uninteresting */
402 * set up socket callbacks
404 static void set_sock_callbacks(struct socket
*sock
,
405 struct ceph_connection
*con
)
407 struct sock
*sk
= sock
->sk
;
408 sk
->sk_user_data
= con
;
409 sk
->sk_data_ready
= ceph_sock_data_ready
;
410 sk
->sk_write_space
= ceph_sock_write_space
;
411 sk
->sk_state_change
= ceph_sock_state_change
;
420 * initiate connection to a remote socket.
422 static int ceph_tcp_connect(struct ceph_connection
*con
)
424 struct sockaddr_storage
*paddr
= &con
->peer_addr
.in_addr
;
429 ret
= sock_create_kern(con
->peer_addr
.in_addr
.ss_family
, SOCK_STREAM
,
433 sock
->sk
->sk_allocation
= GFP_NOFS
;
435 #ifdef CONFIG_LOCKDEP
436 lockdep_set_class(&sock
->sk
->sk_lock
, &socket_class
);
439 set_sock_callbacks(sock
, con
);
441 dout("connect %s\n", ceph_pr_addr(&con
->peer_addr
.in_addr
));
443 con_sock_state_connecting(con
);
444 ret
= sock
->ops
->connect(sock
, (struct sockaddr
*)paddr
, sizeof(*paddr
),
446 if (ret
== -EINPROGRESS
) {
447 dout("connect %s EINPROGRESS sk_state = %u\n",
448 ceph_pr_addr(&con
->peer_addr
.in_addr
),
450 } else if (ret
< 0) {
451 pr_err("connect %s error %d\n",
452 ceph_pr_addr(&con
->peer_addr
.in_addr
), ret
);
454 con
->error_msg
= "connect error";
462 static int ceph_tcp_recvmsg(struct socket
*sock
, void *buf
, size_t len
)
464 struct kvec iov
= {buf
, len
};
465 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
468 r
= kernel_recvmsg(sock
, &msg
, &iov
, 1, len
, msg
.msg_flags
);
474 static int ceph_tcp_recvpage(struct socket
*sock
, struct page
*page
,
475 int page_offset
, size_t length
)
480 BUG_ON(page_offset
+ length
> PAGE_SIZE
);
484 ret
= ceph_tcp_recvmsg(sock
, kaddr
+ page_offset
, length
);
491 * write something. @more is true if caller will be sending more data
494 static int ceph_tcp_sendmsg(struct socket
*sock
, struct kvec
*iov
,
495 size_t kvlen
, size_t len
, int more
)
497 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
501 msg
.msg_flags
|= MSG_MORE
;
503 msg
.msg_flags
|= MSG_EOR
; /* superfluous, but what the hell */
505 r
= kernel_sendmsg(sock
, &msg
, iov
, kvlen
, len
);
511 static int ceph_tcp_sendpage(struct socket
*sock
, struct page
*page
,
512 int offset
, size_t size
, bool more
)
514 int flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
| (more
? MSG_MORE
: MSG_EOR
);
517 ret
= kernel_sendpage(sock
, page
, offset
, size
, flags
);
526 * Shutdown/close the socket for the given connection.
528 static int con_close_socket(struct ceph_connection
*con
)
532 dout("con_close_socket on %p sock %p\n", con
, con
->sock
);
534 rc
= con
->sock
->ops
->shutdown(con
->sock
, SHUT_RDWR
);
535 sock_release(con
->sock
);
540 * Forcibly clear the SOCK_CLOSED flag. It gets set
541 * independent of the connection mutex, and we could have
542 * received a socket close event before we had the chance to
543 * shut the socket down.
545 con_flag_clear(con
, CON_FLAG_SOCK_CLOSED
);
547 con_sock_state_closed(con
);
552 * Reset a connection. Discard all incoming and outgoing messages
553 * and clear *_seq state.
555 static void ceph_msg_remove(struct ceph_msg
*msg
)
557 list_del_init(&msg
->list_head
);
558 BUG_ON(msg
->con
== NULL
);
559 msg
->con
->ops
->put(msg
->con
);
564 static void ceph_msg_remove_list(struct list_head
*head
)
566 while (!list_empty(head
)) {
567 struct ceph_msg
*msg
= list_first_entry(head
, struct ceph_msg
,
569 ceph_msg_remove(msg
);
573 static void reset_connection(struct ceph_connection
*con
)
575 /* reset connection, out_queue, msg_ and connect_seq */
576 /* discard existing out_queue and msg_seq */
577 dout("reset_connection %p\n", con
);
578 ceph_msg_remove_list(&con
->out_queue
);
579 ceph_msg_remove_list(&con
->out_sent
);
582 BUG_ON(con
->in_msg
->con
!= con
);
583 con
->in_msg
->con
= NULL
;
584 ceph_msg_put(con
->in_msg
);
589 con
->connect_seq
= 0;
592 ceph_msg_put(con
->out_msg
);
596 con
->in_seq_acked
= 0;
600 * mark a peer down. drop any open connections.
602 void ceph_con_close(struct ceph_connection
*con
)
604 mutex_lock(&con
->mutex
);
605 dout("con_close %p peer %s\n", con
,
606 ceph_pr_addr(&con
->peer_addr
.in_addr
));
607 con
->state
= CON_STATE_CLOSED
;
609 con_flag_clear(con
, CON_FLAG_LOSSYTX
); /* so we retry next connect */
610 con_flag_clear(con
, CON_FLAG_KEEPALIVE_PENDING
);
611 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
612 con_flag_clear(con
, CON_FLAG_BACKOFF
);
614 reset_connection(con
);
615 con
->peer_global_seq
= 0;
616 cancel_delayed_work(&con
->work
);
617 con_close_socket(con
);
618 mutex_unlock(&con
->mutex
);
620 EXPORT_SYMBOL(ceph_con_close
);
623 * Reopen a closed connection, with a new peer address.
625 void ceph_con_open(struct ceph_connection
*con
,
626 __u8 entity_type
, __u64 entity_num
,
627 struct ceph_entity_addr
*addr
)
629 mutex_lock(&con
->mutex
);
630 dout("con_open %p %s\n", con
, ceph_pr_addr(&addr
->in_addr
));
632 WARN_ON(con
->state
!= CON_STATE_CLOSED
);
633 con
->state
= CON_STATE_PREOPEN
;
635 con
->peer_name
.type
= (__u8
) entity_type
;
636 con
->peer_name
.num
= cpu_to_le64(entity_num
);
638 memcpy(&con
->peer_addr
, addr
, sizeof(*addr
));
639 con
->delay
= 0; /* reset backoff memory */
640 mutex_unlock(&con
->mutex
);
643 EXPORT_SYMBOL(ceph_con_open
);
646 * return true if this connection ever successfully opened
648 bool ceph_con_opened(struct ceph_connection
*con
)
650 return con
->connect_seq
> 0;
654 * initialize a new connection.
656 void ceph_con_init(struct ceph_connection
*con
, void *private,
657 const struct ceph_connection_operations
*ops
,
658 struct ceph_messenger
*msgr
)
660 dout("con_init %p\n", con
);
661 memset(con
, 0, sizeof(*con
));
662 con
->private = private;
666 con_sock_state_init(con
);
668 mutex_init(&con
->mutex
);
669 INIT_LIST_HEAD(&con
->out_queue
);
670 INIT_LIST_HEAD(&con
->out_sent
);
671 INIT_DELAYED_WORK(&con
->work
, con_work
);
673 con
->state
= CON_STATE_CLOSED
;
675 EXPORT_SYMBOL(ceph_con_init
);
679 * We maintain a global counter to order connection attempts. Get
680 * a unique seq greater than @gt.
682 static u32
get_global_seq(struct ceph_messenger
*msgr
, u32 gt
)
686 spin_lock(&msgr
->global_seq_lock
);
687 if (msgr
->global_seq
< gt
)
688 msgr
->global_seq
= gt
;
689 ret
= ++msgr
->global_seq
;
690 spin_unlock(&msgr
->global_seq_lock
);
694 static void con_out_kvec_reset(struct ceph_connection
*con
)
696 con
->out_kvec_left
= 0;
697 con
->out_kvec_bytes
= 0;
698 con
->out_kvec_cur
= &con
->out_kvec
[0];
701 static void con_out_kvec_add(struct ceph_connection
*con
,
702 size_t size
, void *data
)
706 index
= con
->out_kvec_left
;
707 BUG_ON(index
>= ARRAY_SIZE(con
->out_kvec
));
709 con
->out_kvec
[index
].iov_len
= size
;
710 con
->out_kvec
[index
].iov_base
= data
;
711 con
->out_kvec_left
++;
712 con
->out_kvec_bytes
+= size
;
716 static void init_bio_iter(struct bio
*bio
, struct bio
**bio_iter
,
717 unsigned int *bio_seg
)
725 *bio_seg
= (unsigned int) bio
->bi_idx
;
728 static void iter_bio_next(struct bio
**bio_iter
, unsigned int *seg
)
730 if (*bio_iter
== NULL
)
733 BUG_ON(*seg
>= (*bio_iter
)->bi_vcnt
);
736 if (*seg
== (*bio_iter
)->bi_vcnt
)
737 init_bio_iter((*bio_iter
)->bi_next
, bio_iter
, seg
);
741 static void prepare_message_data(struct ceph_msg
*msg
,
742 struct ceph_msg_pos
*msg_pos
)
745 BUG_ON(!msg
->hdr
.data_len
);
747 /* initialize page iterator */
750 msg_pos
->page_pos
= msg
->page_alignment
;
752 msg_pos
->page_pos
= 0;
755 init_bio_iter(msg
->bio
, &msg
->bio_iter
, &msg
->bio_seg
);
757 msg_pos
->data_pos
= 0;
758 msg_pos
->did_page_crc
= false;
762 * Prepare footer for currently outgoing message, and finish things
763 * off. Assumes out_kvec* are already valid.. we just add on to the end.
765 static void prepare_write_message_footer(struct ceph_connection
*con
)
767 struct ceph_msg
*m
= con
->out_msg
;
768 int v
= con
->out_kvec_left
;
770 m
->footer
.flags
|= CEPH_MSG_FOOTER_COMPLETE
;
772 dout("prepare_write_message_footer %p\n", con
);
773 con
->out_kvec_is_msg
= true;
774 con
->out_kvec
[v
].iov_base
= &m
->footer
;
775 con
->out_kvec
[v
].iov_len
= sizeof(m
->footer
);
776 con
->out_kvec_bytes
+= sizeof(m
->footer
);
777 con
->out_kvec_left
++;
778 con
->out_more
= m
->more_to_follow
;
779 con
->out_msg_done
= true;
783 * Prepare headers for the next outgoing message.
785 static void prepare_write_message(struct ceph_connection
*con
)
790 con_out_kvec_reset(con
);
791 con
->out_kvec_is_msg
= true;
792 con
->out_msg_done
= false;
794 /* Sneak an ack in there first? If we can get it into the same
795 * TCP packet that's a good thing. */
796 if (con
->in_seq
> con
->in_seq_acked
) {
797 con
->in_seq_acked
= con
->in_seq
;
798 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
799 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
800 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
804 BUG_ON(list_empty(&con
->out_queue
));
805 m
= list_first_entry(&con
->out_queue
, struct ceph_msg
, list_head
);
807 BUG_ON(m
->con
!= con
);
809 /* put message on sent list */
811 list_move_tail(&m
->list_head
, &con
->out_sent
);
814 * only assign outgoing seq # if we haven't sent this message
815 * yet. if it is requeued, resend with it's original seq.
817 if (m
->needs_out_seq
) {
818 m
->hdr
.seq
= cpu_to_le64(++con
->out_seq
);
819 m
->needs_out_seq
= false;
822 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d (%zd)\n",
823 m
, con
->out_seq
, le16_to_cpu(m
->hdr
.type
),
824 le32_to_cpu(m
->hdr
.front_len
), le32_to_cpu(m
->hdr
.middle_len
),
825 le32_to_cpu(m
->hdr
.data_len
), m
->length
);
826 BUG_ON(le32_to_cpu(m
->hdr
.front_len
) != m
->front
.iov_len
);
828 /* tag + hdr + front + middle */
829 con_out_kvec_add(con
, sizeof (tag_msg
), &tag_msg
);
830 con_out_kvec_add(con
, sizeof (m
->hdr
), &m
->hdr
);
831 con_out_kvec_add(con
, m
->front
.iov_len
, m
->front
.iov_base
);
834 con_out_kvec_add(con
, m
->middle
->vec
.iov_len
,
835 m
->middle
->vec
.iov_base
);
837 /* fill in crc (except data pages), footer */
838 crc
= crc32c(0, &m
->hdr
, offsetof(struct ceph_msg_header
, crc
));
839 con
->out_msg
->hdr
.crc
= cpu_to_le32(crc
);
840 con
->out_msg
->footer
.flags
= 0;
842 crc
= crc32c(0, m
->front
.iov_base
, m
->front
.iov_len
);
843 con
->out_msg
->footer
.front_crc
= cpu_to_le32(crc
);
845 crc
= crc32c(0, m
->middle
->vec
.iov_base
,
846 m
->middle
->vec
.iov_len
);
847 con
->out_msg
->footer
.middle_crc
= cpu_to_le32(crc
);
849 con
->out_msg
->footer
.middle_crc
= 0;
850 dout("%s front_crc %u middle_crc %u\n", __func__
,
851 le32_to_cpu(con
->out_msg
->footer
.front_crc
),
852 le32_to_cpu(con
->out_msg
->footer
.middle_crc
));
854 /* is there a data payload? */
855 con
->out_msg
->footer
.data_crc
= 0;
856 if (m
->hdr
.data_len
) {
857 prepare_message_data(con
->out_msg
, &con
->out_msg_pos
);
858 con
->out_more
= 1; /* data + footer will follow */
860 /* no, queue up footer too and be done */
861 prepare_write_message_footer(con
);
864 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
870 static void prepare_write_ack(struct ceph_connection
*con
)
872 dout("prepare_write_ack %p %llu -> %llu\n", con
,
873 con
->in_seq_acked
, con
->in_seq
);
874 con
->in_seq_acked
= con
->in_seq
;
876 con_out_kvec_reset(con
);
878 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
880 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
881 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
884 con
->out_more
= 1; /* more will follow.. eventually.. */
885 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
889 * Prepare to write keepalive byte.
891 static void prepare_write_keepalive(struct ceph_connection
*con
)
893 dout("prepare_write_keepalive %p\n", con
);
894 con_out_kvec_reset(con
);
895 con_out_kvec_add(con
, sizeof (tag_keepalive
), &tag_keepalive
);
896 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
900 * Connection negotiation.
903 static struct ceph_auth_handshake
*get_connect_authorizer(struct ceph_connection
*con
,
906 struct ceph_auth_handshake
*auth
;
908 if (!con
->ops
->get_authorizer
) {
909 con
->out_connect
.authorizer_protocol
= CEPH_AUTH_UNKNOWN
;
910 con
->out_connect
.authorizer_len
= 0;
914 /* Can't hold the mutex while getting authorizer */
915 mutex_unlock(&con
->mutex
);
916 auth
= con
->ops
->get_authorizer(con
, auth_proto
, con
->auth_retry
);
917 mutex_lock(&con
->mutex
);
921 if (con
->state
!= CON_STATE_NEGOTIATING
)
922 return ERR_PTR(-EAGAIN
);
924 con
->auth_reply_buf
= auth
->authorizer_reply_buf
;
925 con
->auth_reply_buf_len
= auth
->authorizer_reply_buf_len
;
930 * We connected to a peer and are saying hello.
932 static void prepare_write_banner(struct ceph_connection
*con
)
934 con_out_kvec_add(con
, strlen(CEPH_BANNER
), CEPH_BANNER
);
935 con_out_kvec_add(con
, sizeof (con
->msgr
->my_enc_addr
),
936 &con
->msgr
->my_enc_addr
);
939 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
942 static int prepare_write_connect(struct ceph_connection
*con
)
944 unsigned int global_seq
= get_global_seq(con
->msgr
, 0);
947 struct ceph_auth_handshake
*auth
;
949 switch (con
->peer_name
.type
) {
950 case CEPH_ENTITY_TYPE_MON
:
951 proto
= CEPH_MONC_PROTOCOL
;
953 case CEPH_ENTITY_TYPE_OSD
:
954 proto
= CEPH_OSDC_PROTOCOL
;
956 case CEPH_ENTITY_TYPE_MDS
:
957 proto
= CEPH_MDSC_PROTOCOL
;
963 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con
,
964 con
->connect_seq
, global_seq
, proto
);
966 con
->out_connect
.features
= cpu_to_le64(con
->msgr
->supported_features
);
967 con
->out_connect
.host_type
= cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT
);
968 con
->out_connect
.connect_seq
= cpu_to_le32(con
->connect_seq
);
969 con
->out_connect
.global_seq
= cpu_to_le32(global_seq
);
970 con
->out_connect
.protocol_version
= cpu_to_le32(proto
);
971 con
->out_connect
.flags
= 0;
973 auth_proto
= CEPH_AUTH_UNKNOWN
;
974 auth
= get_connect_authorizer(con
, &auth_proto
);
976 return PTR_ERR(auth
);
978 con
->out_connect
.authorizer_protocol
= cpu_to_le32(auth_proto
);
979 con
->out_connect
.authorizer_len
= auth
?
980 cpu_to_le32(auth
->authorizer_buf_len
) : 0;
982 con_out_kvec_add(con
, sizeof (con
->out_connect
),
984 if (auth
&& auth
->authorizer_buf_len
)
985 con_out_kvec_add(con
, auth
->authorizer_buf_len
,
986 auth
->authorizer_buf
);
989 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
995 * write as much of pending kvecs to the socket as we can.
997 * 0 -> socket full, but more to do
1000 static int write_partial_kvec(struct ceph_connection
*con
)
1004 dout("write_partial_kvec %p %d left\n", con
, con
->out_kvec_bytes
);
1005 while (con
->out_kvec_bytes
> 0) {
1006 ret
= ceph_tcp_sendmsg(con
->sock
, con
->out_kvec_cur
,
1007 con
->out_kvec_left
, con
->out_kvec_bytes
,
1011 con
->out_kvec_bytes
-= ret
;
1012 if (con
->out_kvec_bytes
== 0)
1015 /* account for full iov entries consumed */
1016 while (ret
>= con
->out_kvec_cur
->iov_len
) {
1017 BUG_ON(!con
->out_kvec_left
);
1018 ret
-= con
->out_kvec_cur
->iov_len
;
1019 con
->out_kvec_cur
++;
1020 con
->out_kvec_left
--;
1022 /* and for a partially-consumed entry */
1024 con
->out_kvec_cur
->iov_len
-= ret
;
1025 con
->out_kvec_cur
->iov_base
+= ret
;
1028 con
->out_kvec_left
= 0;
1029 con
->out_kvec_is_msg
= false;
1032 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con
,
1033 con
->out_kvec_bytes
, con
->out_kvec_left
, ret
);
1034 return ret
; /* done! */
1037 static void out_msg_pos_next(struct ceph_connection
*con
, struct page
*page
,
1038 size_t len
, size_t sent
, bool in_trail
)
1040 struct ceph_msg
*msg
= con
->out_msg
;
1041 struct ceph_msg_pos
*msg_pos
= &con
->out_msg_pos
;
1046 msg_pos
->data_pos
+= sent
;
1047 msg_pos
->page_pos
+= sent
;
1051 BUG_ON(sent
!= len
);
1052 msg_pos
->page_pos
= 0;
1054 msg_pos
->did_page_crc
= false;
1056 list_rotate_left(&msg
->trail
->head
);
1057 else if (msg
->pagelist
)
1058 list_rotate_left(&msg
->pagelist
->head
);
1061 iter_bio_next(&msg
->bio_iter
, &msg
->bio_seg
);
1065 static void in_msg_pos_next(struct ceph_connection
*con
, size_t len
,
1068 struct ceph_msg
*msg
= con
->in_msg
;
1069 struct ceph_msg_pos
*msg_pos
= &con
->in_msg_pos
;
1074 msg_pos
->data_pos
+= received
;
1075 msg_pos
->page_pos
+= received
;
1079 BUG_ON(received
!= len
);
1080 msg_pos
->page_pos
= 0;
1084 iter_bio_next(&msg
->bio_iter
, &msg
->bio_seg
);
1085 #endif /* CONFIG_BLOCK */
1089 * Write as much message data payload as we can. If we finish, queue
1091 * 1 -> done, footer is now queued in out_kvec[].
1092 * 0 -> socket full, but more to do
1095 static int write_partial_message_data(struct ceph_connection
*con
)
1097 struct ceph_msg
*msg
= con
->out_msg
;
1098 struct ceph_msg_pos
*msg_pos
= &con
->out_msg_pos
;
1099 unsigned int data_len
= le32_to_cpu(msg
->hdr
.data_len
);
1100 bool do_datacrc
= !con
->msgr
->nocrc
;
1102 int total_max_write
;
1103 bool in_trail
= false;
1104 const size_t trail_len
= (msg
->trail
? msg
->trail
->length
: 0);
1105 const size_t trail_off
= data_len
- trail_len
;
1107 dout("%s %p msg %p page %d offset %d\n", __func__
,
1108 con
, msg
, msg_pos
->page
, msg_pos
->page_pos
);
1111 * Iterate through each page that contains data to be
1112 * written, and send as much as possible for each.
1114 * If we are calculating the data crc (the default), we will
1115 * need to map the page. If we have no pages, they have
1116 * been revoked, so use the zero page.
1118 while (data_len
> msg_pos
->data_pos
) {
1119 struct page
*page
= NULL
;
1122 int max_write
= PAGE_SIZE
;
1125 in_trail
= in_trail
|| msg_pos
->data_pos
>= trail_off
;
1127 total_max_write
= trail_off
- msg_pos
->data_pos
;
1130 total_max_write
= data_len
- msg_pos
->data_pos
;
1132 page
= list_first_entry(&msg
->trail
->head
,
1134 } else if (msg
->pages
) {
1135 page
= msg
->pages
[msg_pos
->page
];
1136 } else if (msg
->pagelist
) {
1137 page
= list_first_entry(&msg
->pagelist
->head
,
1140 } else if (msg
->bio
) {
1143 bv
= bio_iovec_idx(msg
->bio_iter
, msg
->bio_seg
);
1145 bio_offset
= bv
->bv_offset
;
1146 max_write
= bv
->bv_len
;
1151 length
= min_t(int, max_write
- msg_pos
->page_pos
,
1154 page_offset
= msg_pos
->page_pos
+ bio_offset
;
1155 if (do_datacrc
&& !msg_pos
->did_page_crc
) {
1157 u32 crc
= le32_to_cpu(msg
->footer
.data_crc
);
1161 BUG_ON(kaddr
== NULL
);
1162 base
= kaddr
+ page_offset
;
1163 crc
= crc32c(crc
, base
, length
);
1165 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1166 msg_pos
->did_page_crc
= true;
1168 ret
= ceph_tcp_sendpage(con
->sock
, page
, page_offset
,
1173 out_msg_pos_next(con
, page
, length
, (size_t) ret
, in_trail
);
1176 dout("%s %p msg %p done\n", __func__
, con
, msg
);
1178 /* prepare and queue up footer, too */
1180 msg
->footer
.flags
|= CEPH_MSG_FOOTER_NOCRC
;
1181 con_out_kvec_reset(con
);
1182 prepare_write_message_footer(con
);
1191 static int write_partial_skip(struct ceph_connection
*con
)
1195 while (con
->out_skip
> 0) {
1196 size_t size
= min(con
->out_skip
, (int) PAGE_CACHE_SIZE
);
1198 ret
= ceph_tcp_sendpage(con
->sock
, zero_page
, 0, size
, true);
1201 con
->out_skip
-= ret
;
1209 * Prepare to read connection handshake, or an ack.
1211 static void prepare_read_banner(struct ceph_connection
*con
)
1213 dout("prepare_read_banner %p\n", con
);
1214 con
->in_base_pos
= 0;
1217 static void prepare_read_connect(struct ceph_connection
*con
)
1219 dout("prepare_read_connect %p\n", con
);
1220 con
->in_base_pos
= 0;
1223 static void prepare_read_ack(struct ceph_connection
*con
)
1225 dout("prepare_read_ack %p\n", con
);
1226 con
->in_base_pos
= 0;
1229 static void prepare_read_tag(struct ceph_connection
*con
)
1231 dout("prepare_read_tag %p\n", con
);
1232 con
->in_base_pos
= 0;
1233 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1237 * Prepare to read a message.
1239 static int prepare_read_message(struct ceph_connection
*con
)
1241 dout("prepare_read_message %p\n", con
);
1242 BUG_ON(con
->in_msg
!= NULL
);
1243 con
->in_base_pos
= 0;
1244 con
->in_front_crc
= con
->in_middle_crc
= con
->in_data_crc
= 0;
1249 static int read_partial(struct ceph_connection
*con
,
1250 int end
, int size
, void *object
)
1252 while (con
->in_base_pos
< end
) {
1253 int left
= end
- con
->in_base_pos
;
1254 int have
= size
- left
;
1255 int ret
= ceph_tcp_recvmsg(con
->sock
, object
+ have
, left
);
1258 con
->in_base_pos
+= ret
;
1265 * Read all or part of the connect-side handshake on a new connection
1267 static int read_partial_banner(struct ceph_connection
*con
)
1273 dout("read_partial_banner %p at %d\n", con
, con
->in_base_pos
);
1276 size
= strlen(CEPH_BANNER
);
1278 ret
= read_partial(con
, end
, size
, con
->in_banner
);
1282 size
= sizeof (con
->actual_peer_addr
);
1284 ret
= read_partial(con
, end
, size
, &con
->actual_peer_addr
);
1288 size
= sizeof (con
->peer_addr_for_me
);
1290 ret
= read_partial(con
, end
, size
, &con
->peer_addr_for_me
);
1298 static int read_partial_connect(struct ceph_connection
*con
)
1304 dout("read_partial_connect %p at %d\n", con
, con
->in_base_pos
);
1306 size
= sizeof (con
->in_reply
);
1308 ret
= read_partial(con
, end
, size
, &con
->in_reply
);
1312 size
= le32_to_cpu(con
->in_reply
.authorizer_len
);
1314 ret
= read_partial(con
, end
, size
, con
->auth_reply_buf
);
1318 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1319 con
, (int)con
->in_reply
.tag
,
1320 le32_to_cpu(con
->in_reply
.connect_seq
),
1321 le32_to_cpu(con
->in_reply
.global_seq
));
1328 * Verify the hello banner looks okay.
1330 static int verify_hello(struct ceph_connection
*con
)
1332 if (memcmp(con
->in_banner
, CEPH_BANNER
, strlen(CEPH_BANNER
))) {
1333 pr_err("connect to %s got bad banner\n",
1334 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1335 con
->error_msg
= "protocol error, bad banner";
1341 static bool addr_is_blank(struct sockaddr_storage
*ss
)
1343 switch (ss
->ss_family
) {
1345 return ((struct sockaddr_in
*)ss
)->sin_addr
.s_addr
== 0;
1348 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[0] == 0 &&
1349 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[1] == 0 &&
1350 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[2] == 0 &&
1351 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[3] == 0;
1356 static int addr_port(struct sockaddr_storage
*ss
)
1358 switch (ss
->ss_family
) {
1360 return ntohs(((struct sockaddr_in
*)ss
)->sin_port
);
1362 return ntohs(((struct sockaddr_in6
*)ss
)->sin6_port
);
1367 static void addr_set_port(struct sockaddr_storage
*ss
, int p
)
1369 switch (ss
->ss_family
) {
1371 ((struct sockaddr_in
*)ss
)->sin_port
= htons(p
);
1374 ((struct sockaddr_in6
*)ss
)->sin6_port
= htons(p
);
1380 * Unlike other *_pton function semantics, zero indicates success.
1382 static int ceph_pton(const char *str
, size_t len
, struct sockaddr_storage
*ss
,
1383 char delim
, const char **ipend
)
1385 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
1386 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
1388 memset(ss
, 0, sizeof(*ss
));
1390 if (in4_pton(str
, len
, (u8
*)&in4
->sin_addr
.s_addr
, delim
, ipend
)) {
1391 ss
->ss_family
= AF_INET
;
1395 if (in6_pton(str
, len
, (u8
*)&in6
->sin6_addr
.s6_addr
, delim
, ipend
)) {
1396 ss
->ss_family
= AF_INET6
;
1404 * Extract hostname string and resolve using kernel DNS facility.
1406 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1407 static int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1408 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1410 const char *end
, *delim_p
;
1411 char *colon_p
, *ip_addr
= NULL
;
1415 * The end of the hostname occurs immediately preceding the delimiter or
1416 * the port marker (':') where the delimiter takes precedence.
1418 delim_p
= memchr(name
, delim
, namelen
);
1419 colon_p
= memchr(name
, ':', namelen
);
1421 if (delim_p
&& colon_p
)
1422 end
= delim_p
< colon_p
? delim_p
: colon_p
;
1423 else if (!delim_p
&& colon_p
)
1427 if (!end
) /* case: hostname:/ */
1428 end
= name
+ namelen
;
1434 /* do dns_resolve upcall */
1435 ip_len
= dns_query(NULL
, name
, end
- name
, NULL
, &ip_addr
, NULL
);
1437 ret
= ceph_pton(ip_addr
, ip_len
, ss
, -1, NULL
);
1445 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end
- name
), name
,
1446 ret
, ret
? "failed" : ceph_pr_addr(ss
));
1451 static inline int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1452 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1459 * Parse a server name (IP or hostname). If a valid IP address is not found
1460 * then try to extract a hostname to resolve using userspace DNS upcall.
1462 static int ceph_parse_server_name(const char *name
, size_t namelen
,
1463 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1467 ret
= ceph_pton(name
, namelen
, ss
, delim
, ipend
);
1469 ret
= ceph_dns_resolve_name(name
, namelen
, ss
, delim
, ipend
);
1475 * Parse an ip[:port] list into an addr array. Use the default
1476 * monitor port if a port isn't specified.
1478 int ceph_parse_ips(const char *c
, const char *end
,
1479 struct ceph_entity_addr
*addr
,
1480 int max_count
, int *count
)
1482 int i
, ret
= -EINVAL
;
1485 dout("parse_ips on '%.*s'\n", (int)(end
-c
), c
);
1486 for (i
= 0; i
< max_count
; i
++) {
1488 struct sockaddr_storage
*ss
= &addr
[i
].in_addr
;
1497 ret
= ceph_parse_server_name(p
, end
- p
, ss
, delim
, &ipend
);
1506 dout("missing matching ']'\n");
1513 if (p
< end
&& *p
== ':') {
1516 while (p
< end
&& *p
>= '0' && *p
<= '9') {
1517 port
= (port
* 10) + (*p
- '0');
1520 if (port
> 65535 || port
== 0)
1523 port
= CEPH_MON_PORT
;
1526 addr_set_port(ss
, port
);
1528 dout("parse_ips got %s\n", ceph_pr_addr(ss
));
1545 pr_err("parse_ips bad ip '%.*s'\n", (int)(end
- c
), c
);
1548 EXPORT_SYMBOL(ceph_parse_ips
);
1550 static int process_banner(struct ceph_connection
*con
)
1552 dout("process_banner on %p\n", con
);
1554 if (verify_hello(con
) < 0)
1557 ceph_decode_addr(&con
->actual_peer_addr
);
1558 ceph_decode_addr(&con
->peer_addr_for_me
);
1561 * Make sure the other end is who we wanted. note that the other
1562 * end may not yet know their ip address, so if it's 0.0.0.0, give
1563 * them the benefit of the doubt.
1565 if (memcmp(&con
->peer_addr
, &con
->actual_peer_addr
,
1566 sizeof(con
->peer_addr
)) != 0 &&
1567 !(addr_is_blank(&con
->actual_peer_addr
.in_addr
) &&
1568 con
->actual_peer_addr
.nonce
== con
->peer_addr
.nonce
)) {
1569 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1570 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1571 (int)le32_to_cpu(con
->peer_addr
.nonce
),
1572 ceph_pr_addr(&con
->actual_peer_addr
.in_addr
),
1573 (int)le32_to_cpu(con
->actual_peer_addr
.nonce
));
1574 con
->error_msg
= "wrong peer at address";
1579 * did we learn our address?
1581 if (addr_is_blank(&con
->msgr
->inst
.addr
.in_addr
)) {
1582 int port
= addr_port(&con
->msgr
->inst
.addr
.in_addr
);
1584 memcpy(&con
->msgr
->inst
.addr
.in_addr
,
1585 &con
->peer_addr_for_me
.in_addr
,
1586 sizeof(con
->peer_addr_for_me
.in_addr
));
1587 addr_set_port(&con
->msgr
->inst
.addr
.in_addr
, port
);
1588 encode_my_addr(con
->msgr
);
1589 dout("process_banner learned my addr is %s\n",
1590 ceph_pr_addr(&con
->msgr
->inst
.addr
.in_addr
));
1596 static int process_connect(struct ceph_connection
*con
)
1598 u64 sup_feat
= con
->msgr
->supported_features
;
1599 u64 req_feat
= con
->msgr
->required_features
;
1600 u64 server_feat
= le64_to_cpu(con
->in_reply
.features
);
1603 dout("process_connect on %p tag %d\n", con
, (int)con
->in_tag
);
1605 switch (con
->in_reply
.tag
) {
1606 case CEPH_MSGR_TAG_FEATURES
:
1607 pr_err("%s%lld %s feature set mismatch,"
1608 " my %llx < server's %llx, missing %llx\n",
1609 ENTITY_NAME(con
->peer_name
),
1610 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1611 sup_feat
, server_feat
, server_feat
& ~sup_feat
);
1612 con
->error_msg
= "missing required protocol features";
1613 reset_connection(con
);
1616 case CEPH_MSGR_TAG_BADPROTOVER
:
1617 pr_err("%s%lld %s protocol version mismatch,"
1618 " my %d != server's %d\n",
1619 ENTITY_NAME(con
->peer_name
),
1620 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1621 le32_to_cpu(con
->out_connect
.protocol_version
),
1622 le32_to_cpu(con
->in_reply
.protocol_version
));
1623 con
->error_msg
= "protocol version mismatch";
1624 reset_connection(con
);
1627 case CEPH_MSGR_TAG_BADAUTHORIZER
:
1629 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con
,
1631 if (con
->auth_retry
== 2) {
1632 con
->error_msg
= "connect authorization failure";
1635 con
->auth_retry
= 1;
1636 con_out_kvec_reset(con
);
1637 ret
= prepare_write_connect(con
);
1640 prepare_read_connect(con
);
1643 case CEPH_MSGR_TAG_RESETSESSION
:
1645 * If we connected with a large connect_seq but the peer
1646 * has no record of a session with us (no connection, or
1647 * connect_seq == 0), they will send RESETSESION to indicate
1648 * that they must have reset their session, and may have
1651 dout("process_connect got RESET peer seq %u\n",
1652 le32_to_cpu(con
->in_reply
.connect_seq
));
1653 pr_err("%s%lld %s connection reset\n",
1654 ENTITY_NAME(con
->peer_name
),
1655 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1656 reset_connection(con
);
1657 con_out_kvec_reset(con
);
1658 ret
= prepare_write_connect(con
);
1661 prepare_read_connect(con
);
1663 /* Tell ceph about it. */
1664 mutex_unlock(&con
->mutex
);
1665 pr_info("reset on %s%lld\n", ENTITY_NAME(con
->peer_name
));
1666 if (con
->ops
->peer_reset
)
1667 con
->ops
->peer_reset(con
);
1668 mutex_lock(&con
->mutex
);
1669 if (con
->state
!= CON_STATE_NEGOTIATING
)
1673 case CEPH_MSGR_TAG_RETRY_SESSION
:
1675 * If we sent a smaller connect_seq than the peer has, try
1676 * again with a larger value.
1678 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
1679 le32_to_cpu(con
->out_connect
.connect_seq
),
1680 le32_to_cpu(con
->in_reply
.connect_seq
));
1681 con
->connect_seq
= le32_to_cpu(con
->in_reply
.connect_seq
);
1682 con_out_kvec_reset(con
);
1683 ret
= prepare_write_connect(con
);
1686 prepare_read_connect(con
);
1689 case CEPH_MSGR_TAG_RETRY_GLOBAL
:
1691 * If we sent a smaller global_seq than the peer has, try
1692 * again with a larger value.
1694 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1695 con
->peer_global_seq
,
1696 le32_to_cpu(con
->in_reply
.global_seq
));
1697 get_global_seq(con
->msgr
,
1698 le32_to_cpu(con
->in_reply
.global_seq
));
1699 con_out_kvec_reset(con
);
1700 ret
= prepare_write_connect(con
);
1703 prepare_read_connect(con
);
1706 case CEPH_MSGR_TAG_READY
:
1707 if (req_feat
& ~server_feat
) {
1708 pr_err("%s%lld %s protocol feature mismatch,"
1709 " my required %llx > server's %llx, need %llx\n",
1710 ENTITY_NAME(con
->peer_name
),
1711 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1712 req_feat
, server_feat
, req_feat
& ~server_feat
);
1713 con
->error_msg
= "missing required protocol features";
1714 reset_connection(con
);
1718 WARN_ON(con
->state
!= CON_STATE_NEGOTIATING
);
1719 con
->state
= CON_STATE_OPEN
;
1721 con
->peer_global_seq
= le32_to_cpu(con
->in_reply
.global_seq
);
1723 con
->peer_features
= server_feat
;
1724 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1725 con
->peer_global_seq
,
1726 le32_to_cpu(con
->in_reply
.connect_seq
),
1728 WARN_ON(con
->connect_seq
!=
1729 le32_to_cpu(con
->in_reply
.connect_seq
));
1731 if (con
->in_reply
.flags
& CEPH_MSG_CONNECT_LOSSY
)
1732 con_flag_set(con
, CON_FLAG_LOSSYTX
);
1734 con
->delay
= 0; /* reset backoff memory */
1736 prepare_read_tag(con
);
1739 case CEPH_MSGR_TAG_WAIT
:
1741 * If there is a connection race (we are opening
1742 * connections to each other), one of us may just have
1743 * to WAIT. This shouldn't happen if we are the
1746 pr_err("process_connect got WAIT as client\n");
1747 con
->error_msg
= "protocol error, got WAIT as client";
1751 pr_err("connect protocol error, will retry\n");
1752 con
->error_msg
= "protocol error, garbage tag during connect";
1760 * read (part of) an ack
1762 static int read_partial_ack(struct ceph_connection
*con
)
1764 int size
= sizeof (con
->in_temp_ack
);
1767 return read_partial(con
, end
, size
, &con
->in_temp_ack
);
1772 * We can finally discard anything that's been acked.
1774 static void process_ack(struct ceph_connection
*con
)
1777 u64 ack
= le64_to_cpu(con
->in_temp_ack
);
1780 while (!list_empty(&con
->out_sent
)) {
1781 m
= list_first_entry(&con
->out_sent
, struct ceph_msg
,
1783 seq
= le64_to_cpu(m
->hdr
.seq
);
1786 dout("got ack for seq %llu type %d at %p\n", seq
,
1787 le16_to_cpu(m
->hdr
.type
), m
);
1788 m
->ack_stamp
= jiffies
;
1791 prepare_read_tag(con
);
1797 static int read_partial_message_section(struct ceph_connection
*con
,
1798 struct kvec
*section
,
1799 unsigned int sec_len
, u32
*crc
)
1805 while (section
->iov_len
< sec_len
) {
1806 BUG_ON(section
->iov_base
== NULL
);
1807 left
= sec_len
- section
->iov_len
;
1808 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)section
->iov_base
+
1809 section
->iov_len
, left
);
1812 section
->iov_len
+= ret
;
1814 if (section
->iov_len
== sec_len
)
1815 *crc
= crc32c(0, section
->iov_base
, section
->iov_len
);
1820 static int ceph_con_in_msg_alloc(struct ceph_connection
*con
, int *skip
);
1822 static int read_partial_message_pages(struct ceph_connection
*con
,
1823 struct page
**pages
,
1824 unsigned int data_len
, bool do_datacrc
)
1826 struct ceph_msg_pos
*msg_pos
= &con
->in_msg_pos
;
1834 BUG_ON(pages
== NULL
);
1835 page
= pages
[msg_pos
->page
];
1836 page_offset
= msg_pos
->page_pos
;
1837 BUG_ON(msg_pos
->data_pos
>= data_len
);
1838 left
= data_len
- msg_pos
->data_pos
;
1839 BUG_ON(page_offset
>= PAGE_SIZE
);
1840 length
= min_t(unsigned int, PAGE_SIZE
- page_offset
, left
);
1842 ret
= ceph_tcp_recvpage(con
->sock
, page
, page_offset
, length
);
1852 base
= kaddr
+ page_offset
;
1853 con
->in_data_crc
= crc32c(con
->in_data_crc
, base
, ret
);
1857 in_msg_pos_next(con
, length
, ret
);
1863 static int read_partial_message_bio(struct ceph_connection
*con
,
1864 unsigned int data_len
, bool do_datacrc
)
1866 struct ceph_msg
*msg
= con
->in_msg
;
1867 struct ceph_msg_pos
*msg_pos
= &con
->in_msg_pos
;
1876 BUG_ON(!msg
->bio_iter
);
1877 bv
= bio_iovec_idx(msg
->bio_iter
, msg
->bio_seg
);
1879 page_offset
= bv
->bv_offset
+ msg_pos
->page_pos
;
1880 BUG_ON(msg_pos
->data_pos
>= data_len
);
1881 left
= data_len
- msg_pos
->data_pos
;
1882 BUG_ON(msg_pos
->page_pos
>= bv
->bv_len
);
1883 length
= min_t(unsigned int, bv
->bv_len
- msg_pos
->page_pos
, left
);
1885 ret
= ceph_tcp_recvpage(con
->sock
, page
, page_offset
, length
);
1895 base
= kaddr
+ page_offset
;
1896 con
->in_data_crc
= crc32c(con
->in_data_crc
, base
, ret
);
1900 in_msg_pos_next(con
, length
, ret
);
1906 static int read_partial_msg_data(struct ceph_connection
*con
)
1908 struct ceph_msg
*msg
= con
->in_msg
;
1909 struct ceph_msg_pos
*msg_pos
= &con
->in_msg_pos
;
1910 const bool do_datacrc
= !con
->msgr
->nocrc
;
1911 unsigned int data_len
;
1916 data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
1917 while (msg_pos
->data_pos
< data_len
) {
1919 ret
= read_partial_message_pages(con
, msg
->pages
,
1920 data_len
, do_datacrc
);
1924 } else if (msg
->bio
) {
1925 ret
= read_partial_message_bio(con
,
1926 data_len
, do_datacrc
);
1935 return 1; /* must return > 0 to indicate success */
1939 * read (part of) a message.
1941 static int read_partial_message(struct ceph_connection
*con
)
1943 struct ceph_msg
*m
= con
->in_msg
;
1947 unsigned int front_len
, middle_len
, data_len
;
1948 bool do_datacrc
= !con
->msgr
->nocrc
;
1952 dout("read_partial_message con %p msg %p\n", con
, m
);
1955 size
= sizeof (con
->in_hdr
);
1957 ret
= read_partial(con
, end
, size
, &con
->in_hdr
);
1961 crc
= crc32c(0, &con
->in_hdr
, offsetof(struct ceph_msg_header
, crc
));
1962 if (cpu_to_le32(crc
) != con
->in_hdr
.crc
) {
1963 pr_err("read_partial_message bad hdr "
1964 " crc %u != expected %u\n",
1965 crc
, con
->in_hdr
.crc
);
1969 front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
1970 if (front_len
> CEPH_MSG_MAX_FRONT_LEN
)
1972 middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
1973 if (middle_len
> CEPH_MSG_MAX_MIDDLE_LEN
)
1975 data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
1976 if (data_len
> CEPH_MSG_MAX_DATA_LEN
)
1980 seq
= le64_to_cpu(con
->in_hdr
.seq
);
1981 if ((s64
)seq
- (s64
)con
->in_seq
< 1) {
1982 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1983 ENTITY_NAME(con
->peer_name
),
1984 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1985 seq
, con
->in_seq
+ 1);
1986 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1988 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1990 } else if ((s64
)seq
- (s64
)con
->in_seq
> 1) {
1991 pr_err("read_partial_message bad seq %lld expected %lld\n",
1992 seq
, con
->in_seq
+ 1);
1993 con
->error_msg
= "bad message sequence # for incoming message";
1997 /* allocate message? */
2001 dout("got hdr type %d front %d data %d\n", con
->in_hdr
.type
,
2002 front_len
, data_len
);
2003 ret
= ceph_con_in_msg_alloc(con
, &skip
);
2007 /* skip this message */
2008 dout("alloc_msg said skip message\n");
2009 BUG_ON(con
->in_msg
);
2010 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
2012 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2017 BUG_ON(!con
->in_msg
);
2018 BUG_ON(con
->in_msg
->con
!= con
);
2020 m
->front
.iov_len
= 0; /* haven't read it yet */
2022 m
->middle
->vec
.iov_len
= 0;
2024 /* prepare for data payload, if any */
2027 prepare_message_data(con
->in_msg
, &con
->in_msg_pos
);
2031 ret
= read_partial_message_section(con
, &m
->front
, front_len
,
2032 &con
->in_front_crc
);
2038 ret
= read_partial_message_section(con
, &m
->middle
->vec
,
2040 &con
->in_middle_crc
);
2047 ret
= read_partial_msg_data(con
);
2053 size
= sizeof (m
->footer
);
2055 ret
= read_partial(con
, end
, size
, &m
->footer
);
2059 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2060 m
, front_len
, m
->footer
.front_crc
, middle_len
,
2061 m
->footer
.middle_crc
, data_len
, m
->footer
.data_crc
);
2064 if (con
->in_front_crc
!= le32_to_cpu(m
->footer
.front_crc
)) {
2065 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2066 m
, con
->in_front_crc
, m
->footer
.front_crc
);
2069 if (con
->in_middle_crc
!= le32_to_cpu(m
->footer
.middle_crc
)) {
2070 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2071 m
, con
->in_middle_crc
, m
->footer
.middle_crc
);
2075 (m
->footer
.flags
& CEPH_MSG_FOOTER_NOCRC
) == 0 &&
2076 con
->in_data_crc
!= le32_to_cpu(m
->footer
.data_crc
)) {
2077 pr_err("read_partial_message %p data crc %u != exp. %u\n", m
,
2078 con
->in_data_crc
, le32_to_cpu(m
->footer
.data_crc
));
2082 return 1; /* done! */
2086 * Process message. This happens in the worker thread. The callback should
2087 * be careful not to do anything that waits on other incoming messages or it
2090 static void process_message(struct ceph_connection
*con
)
2092 struct ceph_msg
*msg
;
2094 BUG_ON(con
->in_msg
->con
!= con
);
2095 con
->in_msg
->con
= NULL
;
2100 /* if first message, set peer_name */
2101 if (con
->peer_name
.type
== 0)
2102 con
->peer_name
= msg
->hdr
.src
;
2105 mutex_unlock(&con
->mutex
);
2107 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2108 msg
, le64_to_cpu(msg
->hdr
.seq
),
2109 ENTITY_NAME(msg
->hdr
.src
),
2110 le16_to_cpu(msg
->hdr
.type
),
2111 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2112 le32_to_cpu(msg
->hdr
.front_len
),
2113 le32_to_cpu(msg
->hdr
.data_len
),
2114 con
->in_front_crc
, con
->in_middle_crc
, con
->in_data_crc
);
2115 con
->ops
->dispatch(con
, msg
);
2117 mutex_lock(&con
->mutex
);
2122 * Write something to the socket. Called in a worker thread when the
2123 * socket appears to be writeable and we have something ready to send.
2125 static int try_write(struct ceph_connection
*con
)
2129 dout("try_write start %p state %lu\n", con
, con
->state
);
2132 dout("try_write out_kvec_bytes %d\n", con
->out_kvec_bytes
);
2134 /* open the socket first? */
2135 if (con
->state
== CON_STATE_PREOPEN
) {
2137 con
->state
= CON_STATE_CONNECTING
;
2139 con_out_kvec_reset(con
);
2140 prepare_write_banner(con
);
2141 prepare_read_banner(con
);
2143 BUG_ON(con
->in_msg
);
2144 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2145 dout("try_write initiating connect on %p new state %lu\n",
2147 ret
= ceph_tcp_connect(con
);
2149 con
->error_msg
= "connect error";
2155 /* kvec data queued? */
2156 if (con
->out_skip
) {
2157 ret
= write_partial_skip(con
);
2161 if (con
->out_kvec_left
) {
2162 ret
= write_partial_kvec(con
);
2169 if (con
->out_msg_done
) {
2170 ceph_msg_put(con
->out_msg
);
2171 con
->out_msg
= NULL
; /* we're done with this one */
2175 ret
= write_partial_message_data(con
);
2177 goto more_kvec
; /* we need to send the footer, too! */
2181 dout("try_write write_partial_message_data err %d\n",
2188 if (con
->state
== CON_STATE_OPEN
) {
2189 /* is anything else pending? */
2190 if (!list_empty(&con
->out_queue
)) {
2191 prepare_write_message(con
);
2194 if (con
->in_seq
> con
->in_seq_acked
) {
2195 prepare_write_ack(con
);
2198 if (con_flag_test_and_clear(con
, CON_FLAG_KEEPALIVE_PENDING
)) {
2199 prepare_write_keepalive(con
);
2204 /* Nothing to do! */
2205 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
2206 dout("try_write nothing else to write.\n");
2209 dout("try_write done on %p ret %d\n", con
, ret
);
2216 * Read what we can from the socket.
2218 static int try_read(struct ceph_connection
*con
)
2223 dout("try_read start on %p state %lu\n", con
, con
->state
);
2224 if (con
->state
!= CON_STATE_CONNECTING
&&
2225 con
->state
!= CON_STATE_NEGOTIATING
&&
2226 con
->state
!= CON_STATE_OPEN
)
2231 dout("try_read tag %d in_base_pos %d\n", (int)con
->in_tag
,
2234 if (con
->state
== CON_STATE_CONNECTING
) {
2235 dout("try_read connecting\n");
2236 ret
= read_partial_banner(con
);
2239 ret
= process_banner(con
);
2243 con
->state
= CON_STATE_NEGOTIATING
;
2246 * Received banner is good, exchange connection info.
2247 * Do not reset out_kvec, as sending our banner raced
2248 * with receiving peer banner after connect completed.
2250 ret
= prepare_write_connect(con
);
2253 prepare_read_connect(con
);
2255 /* Send connection info before awaiting response */
2259 if (con
->state
== CON_STATE_NEGOTIATING
) {
2260 dout("try_read negotiating\n");
2261 ret
= read_partial_connect(con
);
2264 ret
= process_connect(con
);
2270 WARN_ON(con
->state
!= CON_STATE_OPEN
);
2272 if (con
->in_base_pos
< 0) {
2274 * skipping + discarding content.
2276 * FIXME: there must be a better way to do this!
2278 static char buf
[SKIP_BUF_SIZE
];
2279 int skip
= min((int) sizeof (buf
), -con
->in_base_pos
);
2281 dout("skipping %d / %d bytes\n", skip
, -con
->in_base_pos
);
2282 ret
= ceph_tcp_recvmsg(con
->sock
, buf
, skip
);
2285 con
->in_base_pos
+= ret
;
2286 if (con
->in_base_pos
)
2289 if (con
->in_tag
== CEPH_MSGR_TAG_READY
) {
2293 ret
= ceph_tcp_recvmsg(con
->sock
, &con
->in_tag
, 1);
2296 dout("try_read got tag %d\n", (int)con
->in_tag
);
2297 switch (con
->in_tag
) {
2298 case CEPH_MSGR_TAG_MSG
:
2299 prepare_read_message(con
);
2301 case CEPH_MSGR_TAG_ACK
:
2302 prepare_read_ack(con
);
2304 case CEPH_MSGR_TAG_CLOSE
:
2305 con_close_socket(con
);
2306 con
->state
= CON_STATE_CLOSED
;
2312 if (con
->in_tag
== CEPH_MSGR_TAG_MSG
) {
2313 ret
= read_partial_message(con
);
2317 con
->error_msg
= "bad crc";
2321 con
->error_msg
= "io error";
2326 if (con
->in_tag
== CEPH_MSGR_TAG_READY
)
2328 process_message(con
);
2329 if (con
->state
== CON_STATE_OPEN
)
2330 prepare_read_tag(con
);
2333 if (con
->in_tag
== CEPH_MSGR_TAG_ACK
) {
2334 ret
= read_partial_ack(con
);
2342 dout("try_read done on %p ret %d\n", con
, ret
);
2346 pr_err("try_read bad con->in_tag = %d\n", (int)con
->in_tag
);
2347 con
->error_msg
= "protocol error, garbage tag";
2354 * Atomically queue work on a connection after the specified delay.
2355 * Bump @con reference to avoid races with connection teardown.
2356 * Returns 0 if work was queued, or an error code otherwise.
2358 static int queue_con_delay(struct ceph_connection
*con
, unsigned long delay
)
2360 if (!con
->ops
->get(con
)) {
2361 dout("%s %p ref count 0\n", __func__
, con
);
2366 if (!queue_delayed_work(ceph_msgr_wq
, &con
->work
, delay
)) {
2367 dout("%s %p - already queued\n", __func__
, con
);
2373 dout("%s %p %lu\n", __func__
, con
, delay
);
2378 static void queue_con(struct ceph_connection
*con
)
2380 (void) queue_con_delay(con
, 0);
2383 static bool con_sock_closed(struct ceph_connection
*con
)
2385 if (!con_flag_test_and_clear(con
, CON_FLAG_SOCK_CLOSED
))
2389 case CON_STATE_ ## x: \
2390 con->error_msg = "socket closed (con state " #x ")"; \
2393 switch (con
->state
) {
2401 pr_warning("%s con %p unrecognized state %lu\n",
2402 __func__
, con
, con
->state
);
2403 con
->error_msg
= "unrecognized con state";
2412 static bool con_backoff(struct ceph_connection
*con
)
2416 if (!con_flag_test_and_clear(con
, CON_FLAG_BACKOFF
))
2419 ret
= queue_con_delay(con
, round_jiffies_relative(con
->delay
));
2421 dout("%s: con %p FAILED to back off %lu\n", __func__
,
2423 BUG_ON(ret
== -ENOENT
);
2424 con_flag_set(con
, CON_FLAG_BACKOFF
);
2430 /* Finish fault handling; con->mutex must *not* be held here */
2432 static void con_fault_finish(struct ceph_connection
*con
)
2435 * in case we faulted due to authentication, invalidate our
2436 * current tickets so that we can get new ones.
2438 if (con
->auth_retry
&& con
->ops
->invalidate_authorizer
) {
2439 dout("calling invalidate_authorizer()\n");
2440 con
->ops
->invalidate_authorizer(con
);
2443 if (con
->ops
->fault
)
2444 con
->ops
->fault(con
);
2448 * Do some work on a connection. Drop a connection ref when we're done.
2450 static void con_work(struct work_struct
*work
)
2452 struct ceph_connection
*con
= container_of(work
, struct ceph_connection
,
2456 mutex_lock(&con
->mutex
);
2460 if ((fault
= con_sock_closed(con
))) {
2461 dout("%s: con %p SOCK_CLOSED\n", __func__
, con
);
2464 if (con_backoff(con
)) {
2465 dout("%s: con %p BACKOFF\n", __func__
, con
);
2468 if (con
->state
== CON_STATE_STANDBY
) {
2469 dout("%s: con %p STANDBY\n", __func__
, con
);
2472 if (con
->state
== CON_STATE_CLOSED
) {
2473 dout("%s: con %p CLOSED\n", __func__
, con
);
2477 if (con
->state
== CON_STATE_PREOPEN
) {
2478 dout("%s: con %p PREOPEN\n", __func__
, con
);
2482 ret
= try_read(con
);
2486 con
->error_msg
= "socket error on read";
2491 ret
= try_write(con
);
2495 con
->error_msg
= "socket error on write";
2499 break; /* If we make it to here, we're done */
2503 mutex_unlock(&con
->mutex
);
2506 con_fault_finish(con
);
2512 * Generic error/fault handler. A retry mechanism is used with
2513 * exponential backoff
2515 static void con_fault(struct ceph_connection
*con
)
2517 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con
->peer_name
),
2518 ceph_pr_addr(&con
->peer_addr
.in_addr
), con
->error_msg
);
2519 dout("fault %p state %lu to peer %s\n",
2520 con
, con
->state
, ceph_pr_addr(&con
->peer_addr
.in_addr
));
2522 WARN_ON(con
->state
!= CON_STATE_CONNECTING
&&
2523 con
->state
!= CON_STATE_NEGOTIATING
&&
2524 con
->state
!= CON_STATE_OPEN
);
2526 con_close_socket(con
);
2528 if (con_flag_test(con
, CON_FLAG_LOSSYTX
)) {
2529 dout("fault on LOSSYTX channel, marking CLOSED\n");
2530 con
->state
= CON_STATE_CLOSED
;
2535 BUG_ON(con
->in_msg
->con
!= con
);
2536 con
->in_msg
->con
= NULL
;
2537 ceph_msg_put(con
->in_msg
);
2542 /* Requeue anything that hasn't been acked */
2543 list_splice_init(&con
->out_sent
, &con
->out_queue
);
2545 /* If there are no messages queued or keepalive pending, place
2546 * the connection in a STANDBY state */
2547 if (list_empty(&con
->out_queue
) &&
2548 !con_flag_test(con
, CON_FLAG_KEEPALIVE_PENDING
)) {
2549 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con
);
2550 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
2551 con
->state
= CON_STATE_STANDBY
;
2553 /* retry after a delay. */
2554 con
->state
= CON_STATE_PREOPEN
;
2555 if (con
->delay
== 0)
2556 con
->delay
= BASE_DELAY_INTERVAL
;
2557 else if (con
->delay
< MAX_DELAY_INTERVAL
)
2559 con_flag_set(con
, CON_FLAG_BACKOFF
);
2567 * initialize a new messenger instance
2569 void ceph_messenger_init(struct ceph_messenger
*msgr
,
2570 struct ceph_entity_addr
*myaddr
,
2571 u32 supported_features
,
2572 u32 required_features
,
2575 msgr
->supported_features
= supported_features
;
2576 msgr
->required_features
= required_features
;
2578 spin_lock_init(&msgr
->global_seq_lock
);
2581 msgr
->inst
.addr
= *myaddr
;
2583 /* select a random nonce */
2584 msgr
->inst
.addr
.type
= 0;
2585 get_random_bytes(&msgr
->inst
.addr
.nonce
, sizeof(msgr
->inst
.addr
.nonce
));
2586 encode_my_addr(msgr
);
2587 msgr
->nocrc
= nocrc
;
2589 atomic_set(&msgr
->stopping
, 0);
2591 dout("%s %p\n", __func__
, msgr
);
2593 EXPORT_SYMBOL(ceph_messenger_init
);
2595 static void clear_standby(struct ceph_connection
*con
)
2597 /* come back from STANDBY? */
2598 if (con
->state
== CON_STATE_STANDBY
) {
2599 dout("clear_standby %p and ++connect_seq\n", con
);
2600 con
->state
= CON_STATE_PREOPEN
;
2602 WARN_ON(con_flag_test(con
, CON_FLAG_WRITE_PENDING
));
2603 WARN_ON(con_flag_test(con
, CON_FLAG_KEEPALIVE_PENDING
));
2608 * Queue up an outgoing message on the given connection.
2610 void ceph_con_send(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2613 msg
->hdr
.src
= con
->msgr
->inst
.name
;
2614 BUG_ON(msg
->front
.iov_len
!= le32_to_cpu(msg
->hdr
.front_len
));
2615 msg
->needs_out_seq
= true;
2617 mutex_lock(&con
->mutex
);
2619 if (con
->state
== CON_STATE_CLOSED
) {
2620 dout("con_send %p closed, dropping %p\n", con
, msg
);
2622 mutex_unlock(&con
->mutex
);
2626 BUG_ON(msg
->con
!= NULL
);
2627 msg
->con
= con
->ops
->get(con
);
2628 BUG_ON(msg
->con
== NULL
);
2630 BUG_ON(!list_empty(&msg
->list_head
));
2631 list_add_tail(&msg
->list_head
, &con
->out_queue
);
2632 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg
,
2633 ENTITY_NAME(con
->peer_name
), le16_to_cpu(msg
->hdr
.type
),
2634 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2635 le32_to_cpu(msg
->hdr
.front_len
),
2636 le32_to_cpu(msg
->hdr
.middle_len
),
2637 le32_to_cpu(msg
->hdr
.data_len
));
2640 mutex_unlock(&con
->mutex
);
2642 /* if there wasn't anything waiting to send before, queue
2644 if (con_flag_test_and_set(con
, CON_FLAG_WRITE_PENDING
) == 0)
2647 EXPORT_SYMBOL(ceph_con_send
);
2650 * Revoke a message that was previously queued for send
2652 void ceph_msg_revoke(struct ceph_msg
*msg
)
2654 struct ceph_connection
*con
= msg
->con
;
2657 return; /* Message not in our possession */
2659 mutex_lock(&con
->mutex
);
2660 if (!list_empty(&msg
->list_head
)) {
2661 dout("%s %p msg %p - was on queue\n", __func__
, con
, msg
);
2662 list_del_init(&msg
->list_head
);
2663 BUG_ON(msg
->con
== NULL
);
2664 msg
->con
->ops
->put(msg
->con
);
2670 if (con
->out_msg
== msg
) {
2671 dout("%s %p msg %p - was sending\n", __func__
, con
, msg
);
2672 con
->out_msg
= NULL
;
2673 if (con
->out_kvec_is_msg
) {
2674 con
->out_skip
= con
->out_kvec_bytes
;
2675 con
->out_kvec_is_msg
= false;
2681 mutex_unlock(&con
->mutex
);
2685 * Revoke a message that we may be reading data into
2687 void ceph_msg_revoke_incoming(struct ceph_msg
*msg
)
2689 struct ceph_connection
*con
;
2691 BUG_ON(msg
== NULL
);
2693 dout("%s msg %p null con\n", __func__
, msg
);
2695 return; /* Message not in our possession */
2699 mutex_lock(&con
->mutex
);
2700 if (con
->in_msg
== msg
) {
2701 unsigned int front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
2702 unsigned int middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
2703 unsigned int data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
2705 /* skip rest of message */
2706 dout("%s %p msg %p revoked\n", __func__
, con
, msg
);
2707 con
->in_base_pos
= con
->in_base_pos
-
2708 sizeof(struct ceph_msg_header
) -
2712 sizeof(struct ceph_msg_footer
);
2713 ceph_msg_put(con
->in_msg
);
2715 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2718 dout("%s %p in_msg %p msg %p no-op\n",
2719 __func__
, con
, con
->in_msg
, msg
);
2721 mutex_unlock(&con
->mutex
);
2725 * Queue a keepalive byte to ensure the tcp connection is alive.
2727 void ceph_con_keepalive(struct ceph_connection
*con
)
2729 dout("con_keepalive %p\n", con
);
2730 mutex_lock(&con
->mutex
);
2732 mutex_unlock(&con
->mutex
);
2733 if (con_flag_test_and_set(con
, CON_FLAG_KEEPALIVE_PENDING
) == 0 &&
2734 con_flag_test_and_set(con
, CON_FLAG_WRITE_PENDING
) == 0)
2737 EXPORT_SYMBOL(ceph_con_keepalive
);
2739 void ceph_msg_data_set_pages(struct ceph_msg
*msg
, struct page
**pages
,
2740 size_t length
, size_t alignment
)
2745 BUG_ON(msg
->length
);
2748 msg
->length
= length
;
2749 msg
->page_alignment
= alignment
& ~PAGE_MASK
;
2751 EXPORT_SYMBOL(ceph_msg_data_set_pages
);
2753 void ceph_msg_data_set_pagelist(struct ceph_msg
*msg
,
2754 struct ceph_pagelist
*pagelist
)
2757 BUG_ON(!pagelist
->length
);
2758 BUG_ON(msg
->pagelist
);
2760 msg
->pagelist
= pagelist
;
2762 EXPORT_SYMBOL(ceph_msg_data_set_pagelist
);
2764 void ceph_msg_data_set_bio(struct ceph_msg
*msg
, struct bio
*bio
)
2771 EXPORT_SYMBOL(ceph_msg_data_set_bio
);
2773 void ceph_msg_data_set_trail(struct ceph_msg
*msg
, struct ceph_pagelist
*trail
)
2776 BUG_ON(!trail
->length
);
2781 EXPORT_SYMBOL(ceph_msg_data_set_trail
);
2784 * construct a new message with given type, size
2785 * the new msg has a ref count of 1.
2787 struct ceph_msg
*ceph_msg_new(int type
, int front_len
, gfp_t flags
,
2792 m
= kzalloc(sizeof(*m
), flags
);
2796 m
->hdr
.type
= cpu_to_le16(type
);
2797 m
->hdr
.priority
= cpu_to_le16(CEPH_MSG_PRIO_DEFAULT
);
2798 m
->hdr
.front_len
= cpu_to_le32(front_len
);
2800 INIT_LIST_HEAD(&m
->list_head
);
2801 kref_init(&m
->kref
);
2804 m
->front_max
= front_len
;
2806 if (front_len
> PAGE_CACHE_SIZE
) {
2807 m
->front
.iov_base
= __vmalloc(front_len
, flags
,
2809 m
->front_is_vmalloc
= true;
2811 m
->front
.iov_base
= kmalloc(front_len
, flags
);
2813 if (m
->front
.iov_base
== NULL
) {
2814 dout("ceph_msg_new can't allocate %d bytes\n",
2819 m
->front
.iov_base
= NULL
;
2821 m
->front
.iov_len
= front_len
;
2823 dout("ceph_msg_new %p front %d\n", m
, front_len
);
2830 pr_err("msg_new can't create type %d front %d\n", type
,
2834 dout("msg_new can't create type %d front %d\n", type
,
2839 EXPORT_SYMBOL(ceph_msg_new
);
2842 * Allocate "middle" portion of a message, if it is needed and wasn't
2843 * allocated by alloc_msg. This allows us to read a small fixed-size
2844 * per-type header in the front and then gracefully fail (i.e.,
2845 * propagate the error to the caller based on info in the front) when
2846 * the middle is too large.
2848 static int ceph_alloc_middle(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2850 int type
= le16_to_cpu(msg
->hdr
.type
);
2851 int middle_len
= le32_to_cpu(msg
->hdr
.middle_len
);
2853 dout("alloc_middle %p type %d %s middle_len %d\n", msg
, type
,
2854 ceph_msg_type_name(type
), middle_len
);
2855 BUG_ON(!middle_len
);
2856 BUG_ON(msg
->middle
);
2858 msg
->middle
= ceph_buffer_new(middle_len
, GFP_NOFS
);
2865 * Allocate a message for receiving an incoming message on a
2866 * connection, and save the result in con->in_msg. Uses the
2867 * connection's private alloc_msg op if available.
2869 * Returns 0 on success, or a negative error code.
2871 * On success, if we set *skip = 1:
2872 * - the next message should be skipped and ignored.
2873 * - con->in_msg == NULL
2874 * or if we set *skip = 0:
2875 * - con->in_msg is non-null.
2876 * On error (ENOMEM, EAGAIN, ...),
2877 * - con->in_msg == NULL
2879 static int ceph_con_in_msg_alloc(struct ceph_connection
*con
, int *skip
)
2881 struct ceph_msg_header
*hdr
= &con
->in_hdr
;
2882 int middle_len
= le32_to_cpu(hdr
->middle_len
);
2883 struct ceph_msg
*msg
;
2886 BUG_ON(con
->in_msg
!= NULL
);
2887 BUG_ON(!con
->ops
->alloc_msg
);
2889 mutex_unlock(&con
->mutex
);
2890 msg
= con
->ops
->alloc_msg(con
, hdr
, skip
);
2891 mutex_lock(&con
->mutex
);
2892 if (con
->state
!= CON_STATE_OPEN
) {
2900 con
->in_msg
->con
= con
->ops
->get(con
);
2901 BUG_ON(con
->in_msg
->con
== NULL
);
2904 * Null message pointer means either we should skip
2905 * this message or we couldn't allocate memory. The
2906 * former is not an error.
2910 con
->error_msg
= "error allocating memory for incoming message";
2914 memcpy(&con
->in_msg
->hdr
, &con
->in_hdr
, sizeof(con
->in_hdr
));
2916 if (middle_len
&& !con
->in_msg
->middle
) {
2917 ret
= ceph_alloc_middle(con
, con
->in_msg
);
2919 ceph_msg_put(con
->in_msg
);
2929 * Free a generically kmalloc'd message.
2931 void ceph_msg_kfree(struct ceph_msg
*m
)
2933 dout("msg_kfree %p\n", m
);
2934 if (m
->front_is_vmalloc
)
2935 vfree(m
->front
.iov_base
);
2937 kfree(m
->front
.iov_base
);
2942 * Drop a msg ref. Destroy as needed.
2944 void ceph_msg_last_put(struct kref
*kref
)
2946 struct ceph_msg
*m
= container_of(kref
, struct ceph_msg
, kref
);
2948 dout("ceph_msg_put last one on %p\n", m
);
2949 WARN_ON(!list_empty(&m
->list_head
));
2951 /* drop middle, data, if any */
2953 ceph_buffer_put(m
->middle
);
2960 ceph_pagelist_release(m
->pagelist
);
2968 ceph_msgpool_put(m
->pool
, m
);
2972 EXPORT_SYMBOL(ceph_msg_last_put
);
2974 void ceph_msg_dump(struct ceph_msg
*msg
)
2976 pr_debug("msg_dump %p (front_max %d length %zd)\n", msg
,
2977 msg
->front_max
, msg
->length
);
2978 print_hex_dump(KERN_DEBUG
, "header: ",
2979 DUMP_PREFIX_OFFSET
, 16, 1,
2980 &msg
->hdr
, sizeof(msg
->hdr
), true);
2981 print_hex_dump(KERN_DEBUG
, " front: ",
2982 DUMP_PREFIX_OFFSET
, 16, 1,
2983 msg
->front
.iov_base
, msg
->front
.iov_len
, true);
2985 print_hex_dump(KERN_DEBUG
, "middle: ",
2986 DUMP_PREFIX_OFFSET
, 16, 1,
2987 msg
->middle
->vec
.iov_base
,
2988 msg
->middle
->vec
.iov_len
, true);
2989 print_hex_dump(KERN_DEBUG
, "footer: ",
2990 DUMP_PREFIX_OFFSET
, 16, 1,
2991 &msg
->footer
, sizeof(msg
->footer
), true);
2993 EXPORT_SYMBOL(ceph_msg_dump
);