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
);
475 * write something. @more is true if caller will be sending more data
478 static int ceph_tcp_sendmsg(struct socket
*sock
, struct kvec
*iov
,
479 size_t kvlen
, size_t len
, int more
)
481 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
485 msg
.msg_flags
|= MSG_MORE
;
487 msg
.msg_flags
|= MSG_EOR
; /* superfluous, but what the hell */
489 r
= kernel_sendmsg(sock
, &msg
, iov
, kvlen
, len
);
495 static int ceph_tcp_sendpage(struct socket
*sock
, struct page
*page
,
496 int offset
, size_t size
, bool more
)
498 int flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
| (more
? MSG_MORE
: MSG_EOR
);
501 ret
= kernel_sendpage(sock
, page
, offset
, size
, flags
);
510 * Shutdown/close the socket for the given connection.
512 static int con_close_socket(struct ceph_connection
*con
)
516 dout("con_close_socket on %p sock %p\n", con
, con
->sock
);
518 rc
= con
->sock
->ops
->shutdown(con
->sock
, SHUT_RDWR
);
519 sock_release(con
->sock
);
524 * Forcibly clear the SOCK_CLOSED flag. It gets set
525 * independent of the connection mutex, and we could have
526 * received a socket close event before we had the chance to
527 * shut the socket down.
529 con_flag_clear(con
, CON_FLAG_SOCK_CLOSED
);
531 con_sock_state_closed(con
);
536 * Reset a connection. Discard all incoming and outgoing messages
537 * and clear *_seq state.
539 static void ceph_msg_remove(struct ceph_msg
*msg
)
541 list_del_init(&msg
->list_head
);
542 BUG_ON(msg
->con
== NULL
);
543 msg
->con
->ops
->put(msg
->con
);
548 static void ceph_msg_remove_list(struct list_head
*head
)
550 while (!list_empty(head
)) {
551 struct ceph_msg
*msg
= list_first_entry(head
, struct ceph_msg
,
553 ceph_msg_remove(msg
);
557 static void reset_connection(struct ceph_connection
*con
)
559 /* reset connection, out_queue, msg_ and connect_seq */
560 /* discard existing out_queue and msg_seq */
561 dout("reset_connection %p\n", con
);
562 ceph_msg_remove_list(&con
->out_queue
);
563 ceph_msg_remove_list(&con
->out_sent
);
566 BUG_ON(con
->in_msg
->con
!= con
);
567 con
->in_msg
->con
= NULL
;
568 ceph_msg_put(con
->in_msg
);
573 con
->connect_seq
= 0;
576 ceph_msg_put(con
->out_msg
);
580 con
->in_seq_acked
= 0;
584 * mark a peer down. drop any open connections.
586 void ceph_con_close(struct ceph_connection
*con
)
588 mutex_lock(&con
->mutex
);
589 dout("con_close %p peer %s\n", con
,
590 ceph_pr_addr(&con
->peer_addr
.in_addr
));
591 con
->state
= CON_STATE_CLOSED
;
593 con_flag_clear(con
, CON_FLAG_LOSSYTX
); /* so we retry next connect */
594 con_flag_clear(con
, CON_FLAG_KEEPALIVE_PENDING
);
595 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
596 con_flag_clear(con
, CON_FLAG_BACKOFF
);
598 reset_connection(con
);
599 con
->peer_global_seq
= 0;
600 cancel_delayed_work(&con
->work
);
601 con_close_socket(con
);
602 mutex_unlock(&con
->mutex
);
604 EXPORT_SYMBOL(ceph_con_close
);
607 * Reopen a closed connection, with a new peer address.
609 void ceph_con_open(struct ceph_connection
*con
,
610 __u8 entity_type
, __u64 entity_num
,
611 struct ceph_entity_addr
*addr
)
613 mutex_lock(&con
->mutex
);
614 dout("con_open %p %s\n", con
, ceph_pr_addr(&addr
->in_addr
));
616 WARN_ON(con
->state
!= CON_STATE_CLOSED
);
617 con
->state
= CON_STATE_PREOPEN
;
619 con
->peer_name
.type
= (__u8
) entity_type
;
620 con
->peer_name
.num
= cpu_to_le64(entity_num
);
622 memcpy(&con
->peer_addr
, addr
, sizeof(*addr
));
623 con
->delay
= 0; /* reset backoff memory */
624 mutex_unlock(&con
->mutex
);
627 EXPORT_SYMBOL(ceph_con_open
);
630 * return true if this connection ever successfully opened
632 bool ceph_con_opened(struct ceph_connection
*con
)
634 return con
->connect_seq
> 0;
638 * initialize a new connection.
640 void ceph_con_init(struct ceph_connection
*con
, void *private,
641 const struct ceph_connection_operations
*ops
,
642 struct ceph_messenger
*msgr
)
644 dout("con_init %p\n", con
);
645 memset(con
, 0, sizeof(*con
));
646 con
->private = private;
650 con_sock_state_init(con
);
652 mutex_init(&con
->mutex
);
653 INIT_LIST_HEAD(&con
->out_queue
);
654 INIT_LIST_HEAD(&con
->out_sent
);
655 INIT_DELAYED_WORK(&con
->work
, con_work
);
657 con
->state
= CON_STATE_CLOSED
;
659 EXPORT_SYMBOL(ceph_con_init
);
663 * We maintain a global counter to order connection attempts. Get
664 * a unique seq greater than @gt.
666 static u32
get_global_seq(struct ceph_messenger
*msgr
, u32 gt
)
670 spin_lock(&msgr
->global_seq_lock
);
671 if (msgr
->global_seq
< gt
)
672 msgr
->global_seq
= gt
;
673 ret
= ++msgr
->global_seq
;
674 spin_unlock(&msgr
->global_seq_lock
);
678 static void con_out_kvec_reset(struct ceph_connection
*con
)
680 con
->out_kvec_left
= 0;
681 con
->out_kvec_bytes
= 0;
682 con
->out_kvec_cur
= &con
->out_kvec
[0];
685 static void con_out_kvec_add(struct ceph_connection
*con
,
686 size_t size
, void *data
)
690 index
= con
->out_kvec_left
;
691 BUG_ON(index
>= ARRAY_SIZE(con
->out_kvec
));
693 con
->out_kvec
[index
].iov_len
= size
;
694 con
->out_kvec
[index
].iov_base
= data
;
695 con
->out_kvec_left
++;
696 con
->out_kvec_bytes
+= size
;
700 static void init_bio_iter(struct bio
*bio
, struct bio
**bio_iter
,
701 unsigned int *bio_seg
)
709 *bio_seg
= (unsigned int) bio
->bi_idx
;
712 static void iter_bio_next(struct bio
**bio_iter
, unsigned int *seg
)
714 if (*bio_iter
== NULL
)
717 BUG_ON(*seg
>= (*bio_iter
)->bi_vcnt
);
720 if (*seg
== (*bio_iter
)->bi_vcnt
)
721 init_bio_iter((*bio_iter
)->bi_next
, bio_iter
, seg
);
725 static void prepare_write_message_data(struct ceph_connection
*con
)
727 struct ceph_msg
*msg
= con
->out_msg
;
730 BUG_ON(!msg
->hdr
.data_len
);
732 /* initialize page iterator */
733 con
->out_msg_pos
.page
= 0;
735 con
->out_msg_pos
.page_pos
= msg
->page_alignment
;
737 con
->out_msg_pos
.page_pos
= 0;
740 init_bio_iter(msg
->bio
, &msg
->bio_iter
, &msg
->bio_seg
);
742 con
->out_msg_pos
.data_pos
= 0;
743 con
->out_msg_pos
.did_page_crc
= false;
744 con
->out_more
= 1; /* data + footer will follow */
748 * Prepare footer for currently outgoing message, and finish things
749 * off. Assumes out_kvec* are already valid.. we just add on to the end.
751 static void prepare_write_message_footer(struct ceph_connection
*con
)
753 struct ceph_msg
*m
= con
->out_msg
;
754 int v
= con
->out_kvec_left
;
756 m
->footer
.flags
|= CEPH_MSG_FOOTER_COMPLETE
;
758 dout("prepare_write_message_footer %p\n", con
);
759 con
->out_kvec_is_msg
= true;
760 con
->out_kvec
[v
].iov_base
= &m
->footer
;
761 con
->out_kvec
[v
].iov_len
= sizeof(m
->footer
);
762 con
->out_kvec_bytes
+= sizeof(m
->footer
);
763 con
->out_kvec_left
++;
764 con
->out_more
= m
->more_to_follow
;
765 con
->out_msg_done
= true;
769 * Prepare headers for the next outgoing message.
771 static void prepare_write_message(struct ceph_connection
*con
)
776 con_out_kvec_reset(con
);
777 con
->out_kvec_is_msg
= true;
778 con
->out_msg_done
= false;
780 /* Sneak an ack in there first? If we can get it into the same
781 * TCP packet that's a good thing. */
782 if (con
->in_seq
> con
->in_seq_acked
) {
783 con
->in_seq_acked
= con
->in_seq
;
784 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
785 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
786 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
790 BUG_ON(list_empty(&con
->out_queue
));
791 m
= list_first_entry(&con
->out_queue
, struct ceph_msg
, list_head
);
793 BUG_ON(m
->con
!= con
);
795 /* put message on sent list */
797 list_move_tail(&m
->list_head
, &con
->out_sent
);
800 * only assign outgoing seq # if we haven't sent this message
801 * yet. if it is requeued, resend with it's original seq.
803 if (m
->needs_out_seq
) {
804 m
->hdr
.seq
= cpu_to_le64(++con
->out_seq
);
805 m
->needs_out_seq
= false;
808 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d (%zd)\n",
809 m
, con
->out_seq
, le16_to_cpu(m
->hdr
.type
),
810 le32_to_cpu(m
->hdr
.front_len
), le32_to_cpu(m
->hdr
.middle_len
),
811 le32_to_cpu(m
->hdr
.data_len
), m
->length
);
812 BUG_ON(le32_to_cpu(m
->hdr
.front_len
) != m
->front
.iov_len
);
814 /* tag + hdr + front + middle */
815 con_out_kvec_add(con
, sizeof (tag_msg
), &tag_msg
);
816 con_out_kvec_add(con
, sizeof (m
->hdr
), &m
->hdr
);
817 con_out_kvec_add(con
, m
->front
.iov_len
, m
->front
.iov_base
);
820 con_out_kvec_add(con
, m
->middle
->vec
.iov_len
,
821 m
->middle
->vec
.iov_base
);
823 /* fill in crc (except data pages), footer */
824 crc
= crc32c(0, &m
->hdr
, offsetof(struct ceph_msg_header
, crc
));
825 con
->out_msg
->hdr
.crc
= cpu_to_le32(crc
);
826 con
->out_msg
->footer
.flags
= 0;
828 crc
= crc32c(0, m
->front
.iov_base
, m
->front
.iov_len
);
829 con
->out_msg
->footer
.front_crc
= cpu_to_le32(crc
);
831 crc
= crc32c(0, m
->middle
->vec
.iov_base
,
832 m
->middle
->vec
.iov_len
);
833 con
->out_msg
->footer
.middle_crc
= cpu_to_le32(crc
);
835 con
->out_msg
->footer
.middle_crc
= 0;
836 dout("%s front_crc %u middle_crc %u\n", __func__
,
837 le32_to_cpu(con
->out_msg
->footer
.front_crc
),
838 le32_to_cpu(con
->out_msg
->footer
.middle_crc
));
840 /* is there a data payload? */
841 con
->out_msg
->footer
.data_crc
= 0;
843 prepare_write_message_data(con
);
845 /* no, queue up footer too and be done */
846 prepare_write_message_footer(con
);
848 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
854 static void prepare_write_ack(struct ceph_connection
*con
)
856 dout("prepare_write_ack %p %llu -> %llu\n", con
,
857 con
->in_seq_acked
, con
->in_seq
);
858 con
->in_seq_acked
= con
->in_seq
;
860 con_out_kvec_reset(con
);
862 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
864 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
865 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
868 con
->out_more
= 1; /* more will follow.. eventually.. */
869 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
873 * Prepare to write keepalive byte.
875 static void prepare_write_keepalive(struct ceph_connection
*con
)
877 dout("prepare_write_keepalive %p\n", con
);
878 con_out_kvec_reset(con
);
879 con_out_kvec_add(con
, sizeof (tag_keepalive
), &tag_keepalive
);
880 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
884 * Connection negotiation.
887 static struct ceph_auth_handshake
*get_connect_authorizer(struct ceph_connection
*con
,
890 struct ceph_auth_handshake
*auth
;
892 if (!con
->ops
->get_authorizer
) {
893 con
->out_connect
.authorizer_protocol
= CEPH_AUTH_UNKNOWN
;
894 con
->out_connect
.authorizer_len
= 0;
898 /* Can't hold the mutex while getting authorizer */
899 mutex_unlock(&con
->mutex
);
900 auth
= con
->ops
->get_authorizer(con
, auth_proto
, con
->auth_retry
);
901 mutex_lock(&con
->mutex
);
905 if (con
->state
!= CON_STATE_NEGOTIATING
)
906 return ERR_PTR(-EAGAIN
);
908 con
->auth_reply_buf
= auth
->authorizer_reply_buf
;
909 con
->auth_reply_buf_len
= auth
->authorizer_reply_buf_len
;
914 * We connected to a peer and are saying hello.
916 static void prepare_write_banner(struct ceph_connection
*con
)
918 con_out_kvec_add(con
, strlen(CEPH_BANNER
), CEPH_BANNER
);
919 con_out_kvec_add(con
, sizeof (con
->msgr
->my_enc_addr
),
920 &con
->msgr
->my_enc_addr
);
923 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
926 static int prepare_write_connect(struct ceph_connection
*con
)
928 unsigned int global_seq
= get_global_seq(con
->msgr
, 0);
931 struct ceph_auth_handshake
*auth
;
933 switch (con
->peer_name
.type
) {
934 case CEPH_ENTITY_TYPE_MON
:
935 proto
= CEPH_MONC_PROTOCOL
;
937 case CEPH_ENTITY_TYPE_OSD
:
938 proto
= CEPH_OSDC_PROTOCOL
;
940 case CEPH_ENTITY_TYPE_MDS
:
941 proto
= CEPH_MDSC_PROTOCOL
;
947 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con
,
948 con
->connect_seq
, global_seq
, proto
);
950 con
->out_connect
.features
= cpu_to_le64(con
->msgr
->supported_features
);
951 con
->out_connect
.host_type
= cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT
);
952 con
->out_connect
.connect_seq
= cpu_to_le32(con
->connect_seq
);
953 con
->out_connect
.global_seq
= cpu_to_le32(global_seq
);
954 con
->out_connect
.protocol_version
= cpu_to_le32(proto
);
955 con
->out_connect
.flags
= 0;
957 auth_proto
= CEPH_AUTH_UNKNOWN
;
958 auth
= get_connect_authorizer(con
, &auth_proto
);
960 return PTR_ERR(auth
);
962 con
->out_connect
.authorizer_protocol
= cpu_to_le32(auth_proto
);
963 con
->out_connect
.authorizer_len
= auth
?
964 cpu_to_le32(auth
->authorizer_buf_len
) : 0;
966 con_out_kvec_add(con
, sizeof (con
->out_connect
),
968 if (auth
&& auth
->authorizer_buf_len
)
969 con_out_kvec_add(con
, auth
->authorizer_buf_len
,
970 auth
->authorizer_buf
);
973 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
979 * write as much of pending kvecs to the socket as we can.
981 * 0 -> socket full, but more to do
984 static int write_partial_kvec(struct ceph_connection
*con
)
988 dout("write_partial_kvec %p %d left\n", con
, con
->out_kvec_bytes
);
989 while (con
->out_kvec_bytes
> 0) {
990 ret
= ceph_tcp_sendmsg(con
->sock
, con
->out_kvec_cur
,
991 con
->out_kvec_left
, con
->out_kvec_bytes
,
995 con
->out_kvec_bytes
-= ret
;
996 if (con
->out_kvec_bytes
== 0)
999 /* account for full iov entries consumed */
1000 while (ret
>= con
->out_kvec_cur
->iov_len
) {
1001 BUG_ON(!con
->out_kvec_left
);
1002 ret
-= con
->out_kvec_cur
->iov_len
;
1003 con
->out_kvec_cur
++;
1004 con
->out_kvec_left
--;
1006 /* and for a partially-consumed entry */
1008 con
->out_kvec_cur
->iov_len
-= ret
;
1009 con
->out_kvec_cur
->iov_base
+= ret
;
1012 con
->out_kvec_left
= 0;
1013 con
->out_kvec_is_msg
= false;
1016 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con
,
1017 con
->out_kvec_bytes
, con
->out_kvec_left
, ret
);
1018 return ret
; /* done! */
1021 static void out_msg_pos_next(struct ceph_connection
*con
, struct page
*page
,
1022 size_t len
, size_t sent
, bool in_trail
)
1024 struct ceph_msg
*msg
= con
->out_msg
;
1029 con
->out_msg_pos
.data_pos
+= sent
;
1030 con
->out_msg_pos
.page_pos
+= sent
;
1034 BUG_ON(sent
!= len
);
1035 con
->out_msg_pos
.page_pos
= 0;
1036 con
->out_msg_pos
.page
++;
1037 con
->out_msg_pos
.did_page_crc
= false;
1039 list_rotate_left(&msg
->trail
->head
);
1040 else if (msg
->pagelist
)
1041 list_rotate_left(&msg
->pagelist
->head
);
1044 iter_bio_next(&msg
->bio_iter
, &msg
->bio_seg
);
1048 static void in_msg_pos_next(struct ceph_connection
*con
, size_t len
,
1051 struct ceph_msg
*msg
= con
->in_msg
;
1056 con
->in_msg_pos
.data_pos
+= received
;
1057 con
->in_msg_pos
.page_pos
+= received
;
1061 BUG_ON(received
!= len
);
1062 con
->in_msg_pos
.page_pos
= 0;
1063 con
->in_msg_pos
.page
++;
1066 iter_bio_next(&msg
->bio_iter
, &msg
->bio_seg
);
1067 #endif /* CONFIG_BLOCK */
1071 * Write as much message data payload as we can. If we finish, queue
1073 * 1 -> done, footer is now queued in out_kvec[].
1074 * 0 -> socket full, but more to do
1077 static int write_partial_msg_pages(struct ceph_connection
*con
)
1079 struct ceph_msg
*msg
= con
->out_msg
;
1080 unsigned int data_len
= le32_to_cpu(msg
->hdr
.data_len
);
1082 bool do_datacrc
= !con
->msgr
->nocrc
;
1084 int total_max_write
;
1085 bool in_trail
= false;
1086 const size_t trail_len
= (msg
->trail
? msg
->trail
->length
: 0);
1087 const size_t trail_off
= data_len
- trail_len
;
1089 dout("write_partial_msg_pages %p msg %p page %d offset %d\n",
1090 con
, msg
, con
->out_msg_pos
.page
, con
->out_msg_pos
.page_pos
);
1093 * Iterate through each page that contains data to be
1094 * written, and send as much as possible for each.
1096 * If we are calculating the data crc (the default), we will
1097 * need to map the page. If we have no pages, they have
1098 * been revoked, so use the zero page.
1100 while (data_len
> con
->out_msg_pos
.data_pos
) {
1101 struct page
*page
= NULL
;
1102 int max_write
= PAGE_SIZE
;
1105 in_trail
= in_trail
|| con
->out_msg_pos
.data_pos
>= trail_off
;
1107 total_max_write
= trail_off
- con
->out_msg_pos
.data_pos
;
1110 total_max_write
= data_len
- con
->out_msg_pos
.data_pos
;
1112 page
= list_first_entry(&msg
->trail
->head
,
1114 } else if (msg
->pages
) {
1115 page
= msg
->pages
[con
->out_msg_pos
.page
];
1116 } else if (msg
->pagelist
) {
1117 page
= list_first_entry(&msg
->pagelist
->head
,
1120 } else if (msg
->bio
) {
1123 bv
= bio_iovec_idx(msg
->bio_iter
, msg
->bio_seg
);
1125 bio_offset
= bv
->bv_offset
;
1126 max_write
= bv
->bv_len
;
1131 len
= min_t(int, max_write
- con
->out_msg_pos
.page_pos
,
1134 if (do_datacrc
&& !con
->out_msg_pos
.did_page_crc
) {
1136 u32 crc
= le32_to_cpu(msg
->footer
.data_crc
);
1140 BUG_ON(kaddr
== NULL
);
1141 base
= kaddr
+ con
->out_msg_pos
.page_pos
+ bio_offset
;
1142 crc
= crc32c(crc
, base
, len
);
1144 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1145 con
->out_msg_pos
.did_page_crc
= true;
1147 ret
= ceph_tcp_sendpage(con
->sock
, page
,
1148 con
->out_msg_pos
.page_pos
+ bio_offset
,
1153 out_msg_pos_next(con
, page
, len
, (size_t) ret
, in_trail
);
1156 dout("write_partial_msg_pages %p msg %p done\n", con
, msg
);
1158 /* prepare and queue up footer, too */
1160 msg
->footer
.flags
|= CEPH_MSG_FOOTER_NOCRC
;
1161 con_out_kvec_reset(con
);
1162 prepare_write_message_footer(con
);
1171 static int write_partial_skip(struct ceph_connection
*con
)
1175 while (con
->out_skip
> 0) {
1176 size_t size
= min(con
->out_skip
, (int) PAGE_CACHE_SIZE
);
1178 ret
= ceph_tcp_sendpage(con
->sock
, zero_page
, 0, size
, true);
1181 con
->out_skip
-= ret
;
1189 * Prepare to read connection handshake, or an ack.
1191 static void prepare_read_banner(struct ceph_connection
*con
)
1193 dout("prepare_read_banner %p\n", con
);
1194 con
->in_base_pos
= 0;
1197 static void prepare_read_connect(struct ceph_connection
*con
)
1199 dout("prepare_read_connect %p\n", con
);
1200 con
->in_base_pos
= 0;
1203 static void prepare_read_ack(struct ceph_connection
*con
)
1205 dout("prepare_read_ack %p\n", con
);
1206 con
->in_base_pos
= 0;
1209 static void prepare_read_tag(struct ceph_connection
*con
)
1211 dout("prepare_read_tag %p\n", con
);
1212 con
->in_base_pos
= 0;
1213 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1217 * Prepare to read a message.
1219 static int prepare_read_message(struct ceph_connection
*con
)
1221 dout("prepare_read_message %p\n", con
);
1222 BUG_ON(con
->in_msg
!= NULL
);
1223 con
->in_base_pos
= 0;
1224 con
->in_front_crc
= con
->in_middle_crc
= con
->in_data_crc
= 0;
1229 static int read_partial(struct ceph_connection
*con
,
1230 int end
, int size
, void *object
)
1232 while (con
->in_base_pos
< end
) {
1233 int left
= end
- con
->in_base_pos
;
1234 int have
= size
- left
;
1235 int ret
= ceph_tcp_recvmsg(con
->sock
, object
+ have
, left
);
1238 con
->in_base_pos
+= ret
;
1245 * Read all or part of the connect-side handshake on a new connection
1247 static int read_partial_banner(struct ceph_connection
*con
)
1253 dout("read_partial_banner %p at %d\n", con
, con
->in_base_pos
);
1256 size
= strlen(CEPH_BANNER
);
1258 ret
= read_partial(con
, end
, size
, con
->in_banner
);
1262 size
= sizeof (con
->actual_peer_addr
);
1264 ret
= read_partial(con
, end
, size
, &con
->actual_peer_addr
);
1268 size
= sizeof (con
->peer_addr_for_me
);
1270 ret
= read_partial(con
, end
, size
, &con
->peer_addr_for_me
);
1278 static int read_partial_connect(struct ceph_connection
*con
)
1284 dout("read_partial_connect %p at %d\n", con
, con
->in_base_pos
);
1286 size
= sizeof (con
->in_reply
);
1288 ret
= read_partial(con
, end
, size
, &con
->in_reply
);
1292 size
= le32_to_cpu(con
->in_reply
.authorizer_len
);
1294 ret
= read_partial(con
, end
, size
, con
->auth_reply_buf
);
1298 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1299 con
, (int)con
->in_reply
.tag
,
1300 le32_to_cpu(con
->in_reply
.connect_seq
),
1301 le32_to_cpu(con
->in_reply
.global_seq
));
1308 * Verify the hello banner looks okay.
1310 static int verify_hello(struct ceph_connection
*con
)
1312 if (memcmp(con
->in_banner
, CEPH_BANNER
, strlen(CEPH_BANNER
))) {
1313 pr_err("connect to %s got bad banner\n",
1314 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1315 con
->error_msg
= "protocol error, bad banner";
1321 static bool addr_is_blank(struct sockaddr_storage
*ss
)
1323 switch (ss
->ss_family
) {
1325 return ((struct sockaddr_in
*)ss
)->sin_addr
.s_addr
== 0;
1328 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[0] == 0 &&
1329 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[1] == 0 &&
1330 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[2] == 0 &&
1331 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[3] == 0;
1336 static int addr_port(struct sockaddr_storage
*ss
)
1338 switch (ss
->ss_family
) {
1340 return ntohs(((struct sockaddr_in
*)ss
)->sin_port
);
1342 return ntohs(((struct sockaddr_in6
*)ss
)->sin6_port
);
1347 static void addr_set_port(struct sockaddr_storage
*ss
, int p
)
1349 switch (ss
->ss_family
) {
1351 ((struct sockaddr_in
*)ss
)->sin_port
= htons(p
);
1354 ((struct sockaddr_in6
*)ss
)->sin6_port
= htons(p
);
1360 * Unlike other *_pton function semantics, zero indicates success.
1362 static int ceph_pton(const char *str
, size_t len
, struct sockaddr_storage
*ss
,
1363 char delim
, const char **ipend
)
1365 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
1366 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
1368 memset(ss
, 0, sizeof(*ss
));
1370 if (in4_pton(str
, len
, (u8
*)&in4
->sin_addr
.s_addr
, delim
, ipend
)) {
1371 ss
->ss_family
= AF_INET
;
1375 if (in6_pton(str
, len
, (u8
*)&in6
->sin6_addr
.s6_addr
, delim
, ipend
)) {
1376 ss
->ss_family
= AF_INET6
;
1384 * Extract hostname string and resolve using kernel DNS facility.
1386 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1387 static int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1388 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1390 const char *end
, *delim_p
;
1391 char *colon_p
, *ip_addr
= NULL
;
1395 * The end of the hostname occurs immediately preceding the delimiter or
1396 * the port marker (':') where the delimiter takes precedence.
1398 delim_p
= memchr(name
, delim
, namelen
);
1399 colon_p
= memchr(name
, ':', namelen
);
1401 if (delim_p
&& colon_p
)
1402 end
= delim_p
< colon_p
? delim_p
: colon_p
;
1403 else if (!delim_p
&& colon_p
)
1407 if (!end
) /* case: hostname:/ */
1408 end
= name
+ namelen
;
1414 /* do dns_resolve upcall */
1415 ip_len
= dns_query(NULL
, name
, end
- name
, NULL
, &ip_addr
, NULL
);
1417 ret
= ceph_pton(ip_addr
, ip_len
, ss
, -1, NULL
);
1425 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end
- name
), name
,
1426 ret
, ret
? "failed" : ceph_pr_addr(ss
));
1431 static inline int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1432 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1439 * Parse a server name (IP or hostname). If a valid IP address is not found
1440 * then try to extract a hostname to resolve using userspace DNS upcall.
1442 static int ceph_parse_server_name(const char *name
, size_t namelen
,
1443 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1447 ret
= ceph_pton(name
, namelen
, ss
, delim
, ipend
);
1449 ret
= ceph_dns_resolve_name(name
, namelen
, ss
, delim
, ipend
);
1455 * Parse an ip[:port] list into an addr array. Use the default
1456 * monitor port if a port isn't specified.
1458 int ceph_parse_ips(const char *c
, const char *end
,
1459 struct ceph_entity_addr
*addr
,
1460 int max_count
, int *count
)
1462 int i
, ret
= -EINVAL
;
1465 dout("parse_ips on '%.*s'\n", (int)(end
-c
), c
);
1466 for (i
= 0; i
< max_count
; i
++) {
1468 struct sockaddr_storage
*ss
= &addr
[i
].in_addr
;
1477 ret
= ceph_parse_server_name(p
, end
- p
, ss
, delim
, &ipend
);
1486 dout("missing matching ']'\n");
1493 if (p
< end
&& *p
== ':') {
1496 while (p
< end
&& *p
>= '0' && *p
<= '9') {
1497 port
= (port
* 10) + (*p
- '0');
1500 if (port
> 65535 || port
== 0)
1503 port
= CEPH_MON_PORT
;
1506 addr_set_port(ss
, port
);
1508 dout("parse_ips got %s\n", ceph_pr_addr(ss
));
1525 pr_err("parse_ips bad ip '%.*s'\n", (int)(end
- c
), c
);
1528 EXPORT_SYMBOL(ceph_parse_ips
);
1530 static int process_banner(struct ceph_connection
*con
)
1532 dout("process_banner on %p\n", con
);
1534 if (verify_hello(con
) < 0)
1537 ceph_decode_addr(&con
->actual_peer_addr
);
1538 ceph_decode_addr(&con
->peer_addr_for_me
);
1541 * Make sure the other end is who we wanted. note that the other
1542 * end may not yet know their ip address, so if it's 0.0.0.0, give
1543 * them the benefit of the doubt.
1545 if (memcmp(&con
->peer_addr
, &con
->actual_peer_addr
,
1546 sizeof(con
->peer_addr
)) != 0 &&
1547 !(addr_is_blank(&con
->actual_peer_addr
.in_addr
) &&
1548 con
->actual_peer_addr
.nonce
== con
->peer_addr
.nonce
)) {
1549 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1550 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1551 (int)le32_to_cpu(con
->peer_addr
.nonce
),
1552 ceph_pr_addr(&con
->actual_peer_addr
.in_addr
),
1553 (int)le32_to_cpu(con
->actual_peer_addr
.nonce
));
1554 con
->error_msg
= "wrong peer at address";
1559 * did we learn our address?
1561 if (addr_is_blank(&con
->msgr
->inst
.addr
.in_addr
)) {
1562 int port
= addr_port(&con
->msgr
->inst
.addr
.in_addr
);
1564 memcpy(&con
->msgr
->inst
.addr
.in_addr
,
1565 &con
->peer_addr_for_me
.in_addr
,
1566 sizeof(con
->peer_addr_for_me
.in_addr
));
1567 addr_set_port(&con
->msgr
->inst
.addr
.in_addr
, port
);
1568 encode_my_addr(con
->msgr
);
1569 dout("process_banner learned my addr is %s\n",
1570 ceph_pr_addr(&con
->msgr
->inst
.addr
.in_addr
));
1576 static int process_connect(struct ceph_connection
*con
)
1578 u64 sup_feat
= con
->msgr
->supported_features
;
1579 u64 req_feat
= con
->msgr
->required_features
;
1580 u64 server_feat
= le64_to_cpu(con
->in_reply
.features
);
1583 dout("process_connect on %p tag %d\n", con
, (int)con
->in_tag
);
1585 switch (con
->in_reply
.tag
) {
1586 case CEPH_MSGR_TAG_FEATURES
:
1587 pr_err("%s%lld %s feature set mismatch,"
1588 " my %llx < server's %llx, missing %llx\n",
1589 ENTITY_NAME(con
->peer_name
),
1590 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1591 sup_feat
, server_feat
, server_feat
& ~sup_feat
);
1592 con
->error_msg
= "missing required protocol features";
1593 reset_connection(con
);
1596 case CEPH_MSGR_TAG_BADPROTOVER
:
1597 pr_err("%s%lld %s protocol version mismatch,"
1598 " my %d != server's %d\n",
1599 ENTITY_NAME(con
->peer_name
),
1600 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1601 le32_to_cpu(con
->out_connect
.protocol_version
),
1602 le32_to_cpu(con
->in_reply
.protocol_version
));
1603 con
->error_msg
= "protocol version mismatch";
1604 reset_connection(con
);
1607 case CEPH_MSGR_TAG_BADAUTHORIZER
:
1609 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con
,
1611 if (con
->auth_retry
== 2) {
1612 con
->error_msg
= "connect authorization failure";
1615 con
->auth_retry
= 1;
1616 con_out_kvec_reset(con
);
1617 ret
= prepare_write_connect(con
);
1620 prepare_read_connect(con
);
1623 case CEPH_MSGR_TAG_RESETSESSION
:
1625 * If we connected with a large connect_seq but the peer
1626 * has no record of a session with us (no connection, or
1627 * connect_seq == 0), they will send RESETSESION to indicate
1628 * that they must have reset their session, and may have
1631 dout("process_connect got RESET peer seq %u\n",
1632 le32_to_cpu(con
->in_reply
.connect_seq
));
1633 pr_err("%s%lld %s connection reset\n",
1634 ENTITY_NAME(con
->peer_name
),
1635 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1636 reset_connection(con
);
1637 con_out_kvec_reset(con
);
1638 ret
= prepare_write_connect(con
);
1641 prepare_read_connect(con
);
1643 /* Tell ceph about it. */
1644 mutex_unlock(&con
->mutex
);
1645 pr_info("reset on %s%lld\n", ENTITY_NAME(con
->peer_name
));
1646 if (con
->ops
->peer_reset
)
1647 con
->ops
->peer_reset(con
);
1648 mutex_lock(&con
->mutex
);
1649 if (con
->state
!= CON_STATE_NEGOTIATING
)
1653 case CEPH_MSGR_TAG_RETRY_SESSION
:
1655 * If we sent a smaller connect_seq than the peer has, try
1656 * again with a larger value.
1658 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
1659 le32_to_cpu(con
->out_connect
.connect_seq
),
1660 le32_to_cpu(con
->in_reply
.connect_seq
));
1661 con
->connect_seq
= le32_to_cpu(con
->in_reply
.connect_seq
);
1662 con_out_kvec_reset(con
);
1663 ret
= prepare_write_connect(con
);
1666 prepare_read_connect(con
);
1669 case CEPH_MSGR_TAG_RETRY_GLOBAL
:
1671 * If we sent a smaller global_seq than the peer has, try
1672 * again with a larger value.
1674 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1675 con
->peer_global_seq
,
1676 le32_to_cpu(con
->in_reply
.global_seq
));
1677 get_global_seq(con
->msgr
,
1678 le32_to_cpu(con
->in_reply
.global_seq
));
1679 con_out_kvec_reset(con
);
1680 ret
= prepare_write_connect(con
);
1683 prepare_read_connect(con
);
1686 case CEPH_MSGR_TAG_READY
:
1687 if (req_feat
& ~server_feat
) {
1688 pr_err("%s%lld %s protocol feature mismatch,"
1689 " my required %llx > server's %llx, need %llx\n",
1690 ENTITY_NAME(con
->peer_name
),
1691 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1692 req_feat
, server_feat
, req_feat
& ~server_feat
);
1693 con
->error_msg
= "missing required protocol features";
1694 reset_connection(con
);
1698 WARN_ON(con
->state
!= CON_STATE_NEGOTIATING
);
1699 con
->state
= CON_STATE_OPEN
;
1701 con
->peer_global_seq
= le32_to_cpu(con
->in_reply
.global_seq
);
1703 con
->peer_features
= server_feat
;
1704 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1705 con
->peer_global_seq
,
1706 le32_to_cpu(con
->in_reply
.connect_seq
),
1708 WARN_ON(con
->connect_seq
!=
1709 le32_to_cpu(con
->in_reply
.connect_seq
));
1711 if (con
->in_reply
.flags
& CEPH_MSG_CONNECT_LOSSY
)
1712 con_flag_set(con
, CON_FLAG_LOSSYTX
);
1714 con
->delay
= 0; /* reset backoff memory */
1716 prepare_read_tag(con
);
1719 case CEPH_MSGR_TAG_WAIT
:
1721 * If there is a connection race (we are opening
1722 * connections to each other), one of us may just have
1723 * to WAIT. This shouldn't happen if we are the
1726 pr_err("process_connect got WAIT as client\n");
1727 con
->error_msg
= "protocol error, got WAIT as client";
1731 pr_err("connect protocol error, will retry\n");
1732 con
->error_msg
= "protocol error, garbage tag during connect";
1740 * read (part of) an ack
1742 static int read_partial_ack(struct ceph_connection
*con
)
1744 int size
= sizeof (con
->in_temp_ack
);
1747 return read_partial(con
, end
, size
, &con
->in_temp_ack
);
1752 * We can finally discard anything that's been acked.
1754 static void process_ack(struct ceph_connection
*con
)
1757 u64 ack
= le64_to_cpu(con
->in_temp_ack
);
1760 while (!list_empty(&con
->out_sent
)) {
1761 m
= list_first_entry(&con
->out_sent
, struct ceph_msg
,
1763 seq
= le64_to_cpu(m
->hdr
.seq
);
1766 dout("got ack for seq %llu type %d at %p\n", seq
,
1767 le16_to_cpu(m
->hdr
.type
), m
);
1768 m
->ack_stamp
= jiffies
;
1771 prepare_read_tag(con
);
1777 static int read_partial_message_section(struct ceph_connection
*con
,
1778 struct kvec
*section
,
1779 unsigned int sec_len
, u32
*crc
)
1785 while (section
->iov_len
< sec_len
) {
1786 BUG_ON(section
->iov_base
== NULL
);
1787 left
= sec_len
- section
->iov_len
;
1788 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)section
->iov_base
+
1789 section
->iov_len
, left
);
1792 section
->iov_len
+= ret
;
1794 if (section
->iov_len
== sec_len
)
1795 *crc
= crc32c(0, section
->iov_base
, section
->iov_len
);
1800 static int ceph_con_in_msg_alloc(struct ceph_connection
*con
, int *skip
);
1802 static int read_partial_message_pages(struct ceph_connection
*con
,
1803 struct page
**pages
,
1804 unsigned int data_len
, bool do_datacrc
)
1811 left
= min((int)(data_len
- con
->in_msg_pos
.data_pos
),
1812 (int)(PAGE_SIZE
- con
->in_msg_pos
.page_pos
));
1814 BUG_ON(pages
== NULL
);
1815 page
= pages
[con
->in_msg_pos
.page
];
1817 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ con
->in_msg_pos
.page_pos
, left
);
1818 if (ret
> 0 && do_datacrc
)
1820 crc32c(con
->in_data_crc
,
1821 p
+ con
->in_msg_pos
.page_pos
, ret
);
1826 in_msg_pos_next(con
, left
, ret
);
1832 static int read_partial_message_bio(struct ceph_connection
*con
,
1833 unsigned int data_len
, bool do_datacrc
)
1835 struct ceph_msg
*msg
= con
->in_msg
;
1842 BUG_ON(!msg
->bio_iter
);
1843 bv
= bio_iovec_idx(msg
->bio_iter
, msg
->bio_seg
);
1845 left
= min((int)(data_len
- con
->in_msg_pos
.data_pos
),
1846 (int)(bv
->bv_len
- con
->in_msg_pos
.page_pos
));
1849 p
= kmap(page
) + bv
->bv_offset
;
1851 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ con
->in_msg_pos
.page_pos
, left
);
1852 if (ret
> 0 && do_datacrc
)
1854 crc32c(con
->in_data_crc
,
1855 p
+ con
->in_msg_pos
.page_pos
, ret
);
1860 in_msg_pos_next(con
, left
, ret
);
1867 * read (part of) a message.
1869 static int read_partial_message(struct ceph_connection
*con
)
1871 struct ceph_msg
*m
= con
->in_msg
;
1875 unsigned int front_len
, middle_len
, data_len
;
1876 bool do_datacrc
= !con
->msgr
->nocrc
;
1880 dout("read_partial_message con %p msg %p\n", con
, m
);
1883 size
= sizeof (con
->in_hdr
);
1885 ret
= read_partial(con
, end
, size
, &con
->in_hdr
);
1889 crc
= crc32c(0, &con
->in_hdr
, offsetof(struct ceph_msg_header
, crc
));
1890 if (cpu_to_le32(crc
) != con
->in_hdr
.crc
) {
1891 pr_err("read_partial_message bad hdr "
1892 " crc %u != expected %u\n",
1893 crc
, con
->in_hdr
.crc
);
1897 front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
1898 if (front_len
> CEPH_MSG_MAX_FRONT_LEN
)
1900 middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
1901 if (middle_len
> CEPH_MSG_MAX_MIDDLE_LEN
)
1903 data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
1904 if (data_len
> CEPH_MSG_MAX_DATA_LEN
)
1908 seq
= le64_to_cpu(con
->in_hdr
.seq
);
1909 if ((s64
)seq
- (s64
)con
->in_seq
< 1) {
1910 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1911 ENTITY_NAME(con
->peer_name
),
1912 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1913 seq
, con
->in_seq
+ 1);
1914 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1916 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1918 } else if ((s64
)seq
- (s64
)con
->in_seq
> 1) {
1919 pr_err("read_partial_message bad seq %lld expected %lld\n",
1920 seq
, con
->in_seq
+ 1);
1921 con
->error_msg
= "bad message sequence # for incoming message";
1925 /* allocate message? */
1929 dout("got hdr type %d front %d data %d\n", con
->in_hdr
.type
,
1930 front_len
, data_len
);
1931 ret
= ceph_con_in_msg_alloc(con
, &skip
);
1935 /* skip this message */
1936 dout("alloc_msg said skip message\n");
1937 BUG_ON(con
->in_msg
);
1938 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1940 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1945 BUG_ON(!con
->in_msg
);
1946 BUG_ON(con
->in_msg
->con
!= con
);
1948 m
->front
.iov_len
= 0; /* haven't read it yet */
1950 m
->middle
->vec
.iov_len
= 0;
1952 con
->in_msg_pos
.page
= 0;
1954 con
->in_msg_pos
.page_pos
= m
->page_alignment
;
1956 con
->in_msg_pos
.page_pos
= 0;
1957 con
->in_msg_pos
.data_pos
= 0;
1961 init_bio_iter(m
->bio
, &m
->bio_iter
, &m
->bio_seg
);
1966 ret
= read_partial_message_section(con
, &m
->front
, front_len
,
1967 &con
->in_front_crc
);
1973 ret
= read_partial_message_section(con
, &m
->middle
->vec
,
1975 &con
->in_middle_crc
);
1981 while (con
->in_msg_pos
.data_pos
< data_len
) {
1983 ret
= read_partial_message_pages(con
, m
->pages
,
1984 data_len
, do_datacrc
);
1988 } else if (m
->bio
) {
1989 ret
= read_partial_message_bio(con
,
1990 data_len
, do_datacrc
);
2000 size
= sizeof (m
->footer
);
2002 ret
= read_partial(con
, end
, size
, &m
->footer
);
2006 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2007 m
, front_len
, m
->footer
.front_crc
, middle_len
,
2008 m
->footer
.middle_crc
, data_len
, m
->footer
.data_crc
);
2011 if (con
->in_front_crc
!= le32_to_cpu(m
->footer
.front_crc
)) {
2012 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2013 m
, con
->in_front_crc
, m
->footer
.front_crc
);
2016 if (con
->in_middle_crc
!= le32_to_cpu(m
->footer
.middle_crc
)) {
2017 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2018 m
, con
->in_middle_crc
, m
->footer
.middle_crc
);
2022 (m
->footer
.flags
& CEPH_MSG_FOOTER_NOCRC
) == 0 &&
2023 con
->in_data_crc
!= le32_to_cpu(m
->footer
.data_crc
)) {
2024 pr_err("read_partial_message %p data crc %u != exp. %u\n", m
,
2025 con
->in_data_crc
, le32_to_cpu(m
->footer
.data_crc
));
2029 return 1; /* done! */
2033 * Process message. This happens in the worker thread. The callback should
2034 * be careful not to do anything that waits on other incoming messages or it
2037 static void process_message(struct ceph_connection
*con
)
2039 struct ceph_msg
*msg
;
2041 BUG_ON(con
->in_msg
->con
!= con
);
2042 con
->in_msg
->con
= NULL
;
2047 /* if first message, set peer_name */
2048 if (con
->peer_name
.type
== 0)
2049 con
->peer_name
= msg
->hdr
.src
;
2052 mutex_unlock(&con
->mutex
);
2054 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2055 msg
, le64_to_cpu(msg
->hdr
.seq
),
2056 ENTITY_NAME(msg
->hdr
.src
),
2057 le16_to_cpu(msg
->hdr
.type
),
2058 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2059 le32_to_cpu(msg
->hdr
.front_len
),
2060 le32_to_cpu(msg
->hdr
.data_len
),
2061 con
->in_front_crc
, con
->in_middle_crc
, con
->in_data_crc
);
2062 con
->ops
->dispatch(con
, msg
);
2064 mutex_lock(&con
->mutex
);
2069 * Write something to the socket. Called in a worker thread when the
2070 * socket appears to be writeable and we have something ready to send.
2072 static int try_write(struct ceph_connection
*con
)
2076 dout("try_write start %p state %lu\n", con
, con
->state
);
2079 dout("try_write out_kvec_bytes %d\n", con
->out_kvec_bytes
);
2081 /* open the socket first? */
2082 if (con
->state
== CON_STATE_PREOPEN
) {
2084 con
->state
= CON_STATE_CONNECTING
;
2086 con_out_kvec_reset(con
);
2087 prepare_write_banner(con
);
2088 prepare_read_banner(con
);
2090 BUG_ON(con
->in_msg
);
2091 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2092 dout("try_write initiating connect on %p new state %lu\n",
2094 ret
= ceph_tcp_connect(con
);
2096 con
->error_msg
= "connect error";
2102 /* kvec data queued? */
2103 if (con
->out_skip
) {
2104 ret
= write_partial_skip(con
);
2108 if (con
->out_kvec_left
) {
2109 ret
= write_partial_kvec(con
);
2116 if (con
->out_msg_done
) {
2117 ceph_msg_put(con
->out_msg
);
2118 con
->out_msg
= NULL
; /* we're done with this one */
2122 ret
= write_partial_msg_pages(con
);
2124 goto more_kvec
; /* we need to send the footer, too! */
2128 dout("try_write write_partial_msg_pages err %d\n",
2135 if (con
->state
== CON_STATE_OPEN
) {
2136 /* is anything else pending? */
2137 if (!list_empty(&con
->out_queue
)) {
2138 prepare_write_message(con
);
2141 if (con
->in_seq
> con
->in_seq_acked
) {
2142 prepare_write_ack(con
);
2145 if (con_flag_test_and_clear(con
, CON_FLAG_KEEPALIVE_PENDING
)) {
2146 prepare_write_keepalive(con
);
2151 /* Nothing to do! */
2152 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
2153 dout("try_write nothing else to write.\n");
2156 dout("try_write done on %p ret %d\n", con
, ret
);
2163 * Read what we can from the socket.
2165 static int try_read(struct ceph_connection
*con
)
2170 dout("try_read start on %p state %lu\n", con
, con
->state
);
2171 if (con
->state
!= CON_STATE_CONNECTING
&&
2172 con
->state
!= CON_STATE_NEGOTIATING
&&
2173 con
->state
!= CON_STATE_OPEN
)
2178 dout("try_read tag %d in_base_pos %d\n", (int)con
->in_tag
,
2181 if (con
->state
== CON_STATE_CONNECTING
) {
2182 dout("try_read connecting\n");
2183 ret
= read_partial_banner(con
);
2186 ret
= process_banner(con
);
2190 con
->state
= CON_STATE_NEGOTIATING
;
2193 * Received banner is good, exchange connection info.
2194 * Do not reset out_kvec, as sending our banner raced
2195 * with receiving peer banner after connect completed.
2197 ret
= prepare_write_connect(con
);
2200 prepare_read_connect(con
);
2202 /* Send connection info before awaiting response */
2206 if (con
->state
== CON_STATE_NEGOTIATING
) {
2207 dout("try_read negotiating\n");
2208 ret
= read_partial_connect(con
);
2211 ret
= process_connect(con
);
2217 WARN_ON(con
->state
!= CON_STATE_OPEN
);
2219 if (con
->in_base_pos
< 0) {
2221 * skipping + discarding content.
2223 * FIXME: there must be a better way to do this!
2225 static char buf
[SKIP_BUF_SIZE
];
2226 int skip
= min((int) sizeof (buf
), -con
->in_base_pos
);
2228 dout("skipping %d / %d bytes\n", skip
, -con
->in_base_pos
);
2229 ret
= ceph_tcp_recvmsg(con
->sock
, buf
, skip
);
2232 con
->in_base_pos
+= ret
;
2233 if (con
->in_base_pos
)
2236 if (con
->in_tag
== CEPH_MSGR_TAG_READY
) {
2240 ret
= ceph_tcp_recvmsg(con
->sock
, &con
->in_tag
, 1);
2243 dout("try_read got tag %d\n", (int)con
->in_tag
);
2244 switch (con
->in_tag
) {
2245 case CEPH_MSGR_TAG_MSG
:
2246 prepare_read_message(con
);
2248 case CEPH_MSGR_TAG_ACK
:
2249 prepare_read_ack(con
);
2251 case CEPH_MSGR_TAG_CLOSE
:
2252 con_close_socket(con
);
2253 con
->state
= CON_STATE_CLOSED
;
2259 if (con
->in_tag
== CEPH_MSGR_TAG_MSG
) {
2260 ret
= read_partial_message(con
);
2264 con
->error_msg
= "bad crc";
2268 con
->error_msg
= "io error";
2273 if (con
->in_tag
== CEPH_MSGR_TAG_READY
)
2275 process_message(con
);
2276 if (con
->state
== CON_STATE_OPEN
)
2277 prepare_read_tag(con
);
2280 if (con
->in_tag
== CEPH_MSGR_TAG_ACK
) {
2281 ret
= read_partial_ack(con
);
2289 dout("try_read done on %p ret %d\n", con
, ret
);
2293 pr_err("try_read bad con->in_tag = %d\n", (int)con
->in_tag
);
2294 con
->error_msg
= "protocol error, garbage tag";
2301 * Atomically queue work on a connection after the specified delay.
2302 * Bump @con reference to avoid races with connection teardown.
2303 * Returns 0 if work was queued, or an error code otherwise.
2305 static int queue_con_delay(struct ceph_connection
*con
, unsigned long delay
)
2307 if (!con
->ops
->get(con
)) {
2308 dout("%s %p ref count 0\n", __func__
, con
);
2313 if (!queue_delayed_work(ceph_msgr_wq
, &con
->work
, delay
)) {
2314 dout("%s %p - already queued\n", __func__
, con
);
2320 dout("%s %p %lu\n", __func__
, con
, delay
);
2325 static void queue_con(struct ceph_connection
*con
)
2327 (void) queue_con_delay(con
, 0);
2330 static bool con_sock_closed(struct ceph_connection
*con
)
2332 if (!con_flag_test_and_clear(con
, CON_FLAG_SOCK_CLOSED
))
2336 case CON_STATE_ ## x: \
2337 con->error_msg = "socket closed (con state " #x ")"; \
2340 switch (con
->state
) {
2348 pr_warning("%s con %p unrecognized state %lu\n",
2349 __func__
, con
, con
->state
);
2350 con
->error_msg
= "unrecognized con state";
2359 static bool con_backoff(struct ceph_connection
*con
)
2363 if (!con_flag_test_and_clear(con
, CON_FLAG_BACKOFF
))
2366 ret
= queue_con_delay(con
, round_jiffies_relative(con
->delay
));
2368 dout("%s: con %p FAILED to back off %lu\n", __func__
,
2370 BUG_ON(ret
== -ENOENT
);
2371 con_flag_set(con
, CON_FLAG_BACKOFF
);
2377 /* Finish fault handling; con->mutex must *not* be held here */
2379 static void con_fault_finish(struct ceph_connection
*con
)
2382 * in case we faulted due to authentication, invalidate our
2383 * current tickets so that we can get new ones.
2385 if (con
->auth_retry
&& con
->ops
->invalidate_authorizer
) {
2386 dout("calling invalidate_authorizer()\n");
2387 con
->ops
->invalidate_authorizer(con
);
2390 if (con
->ops
->fault
)
2391 con
->ops
->fault(con
);
2395 * Do some work on a connection. Drop a connection ref when we're done.
2397 static void con_work(struct work_struct
*work
)
2399 struct ceph_connection
*con
= container_of(work
, struct ceph_connection
,
2403 mutex_lock(&con
->mutex
);
2407 if ((fault
= con_sock_closed(con
))) {
2408 dout("%s: con %p SOCK_CLOSED\n", __func__
, con
);
2411 if (con_backoff(con
)) {
2412 dout("%s: con %p BACKOFF\n", __func__
, con
);
2415 if (con
->state
== CON_STATE_STANDBY
) {
2416 dout("%s: con %p STANDBY\n", __func__
, con
);
2419 if (con
->state
== CON_STATE_CLOSED
) {
2420 dout("%s: con %p CLOSED\n", __func__
, con
);
2424 if (con
->state
== CON_STATE_PREOPEN
) {
2425 dout("%s: con %p PREOPEN\n", __func__
, con
);
2429 ret
= try_read(con
);
2433 con
->error_msg
= "socket error on read";
2438 ret
= try_write(con
);
2442 con
->error_msg
= "socket error on write";
2446 break; /* If we make it to here, we're done */
2450 mutex_unlock(&con
->mutex
);
2453 con_fault_finish(con
);
2459 * Generic error/fault handler. A retry mechanism is used with
2460 * exponential backoff
2462 static void con_fault(struct ceph_connection
*con
)
2464 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con
->peer_name
),
2465 ceph_pr_addr(&con
->peer_addr
.in_addr
), con
->error_msg
);
2466 dout("fault %p state %lu to peer %s\n",
2467 con
, con
->state
, ceph_pr_addr(&con
->peer_addr
.in_addr
));
2469 WARN_ON(con
->state
!= CON_STATE_CONNECTING
&&
2470 con
->state
!= CON_STATE_NEGOTIATING
&&
2471 con
->state
!= CON_STATE_OPEN
);
2473 con_close_socket(con
);
2475 if (con_flag_test(con
, CON_FLAG_LOSSYTX
)) {
2476 dout("fault on LOSSYTX channel, marking CLOSED\n");
2477 con
->state
= CON_STATE_CLOSED
;
2482 BUG_ON(con
->in_msg
->con
!= con
);
2483 con
->in_msg
->con
= NULL
;
2484 ceph_msg_put(con
->in_msg
);
2489 /* Requeue anything that hasn't been acked */
2490 list_splice_init(&con
->out_sent
, &con
->out_queue
);
2492 /* If there are no messages queued or keepalive pending, place
2493 * the connection in a STANDBY state */
2494 if (list_empty(&con
->out_queue
) &&
2495 !con_flag_test(con
, CON_FLAG_KEEPALIVE_PENDING
)) {
2496 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con
);
2497 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
2498 con
->state
= CON_STATE_STANDBY
;
2500 /* retry after a delay. */
2501 con
->state
= CON_STATE_PREOPEN
;
2502 if (con
->delay
== 0)
2503 con
->delay
= BASE_DELAY_INTERVAL
;
2504 else if (con
->delay
< MAX_DELAY_INTERVAL
)
2506 con_flag_set(con
, CON_FLAG_BACKOFF
);
2514 * initialize a new messenger instance
2516 void ceph_messenger_init(struct ceph_messenger
*msgr
,
2517 struct ceph_entity_addr
*myaddr
,
2518 u32 supported_features
,
2519 u32 required_features
,
2522 msgr
->supported_features
= supported_features
;
2523 msgr
->required_features
= required_features
;
2525 spin_lock_init(&msgr
->global_seq_lock
);
2528 msgr
->inst
.addr
= *myaddr
;
2530 /* select a random nonce */
2531 msgr
->inst
.addr
.type
= 0;
2532 get_random_bytes(&msgr
->inst
.addr
.nonce
, sizeof(msgr
->inst
.addr
.nonce
));
2533 encode_my_addr(msgr
);
2534 msgr
->nocrc
= nocrc
;
2536 atomic_set(&msgr
->stopping
, 0);
2538 dout("%s %p\n", __func__
, msgr
);
2540 EXPORT_SYMBOL(ceph_messenger_init
);
2542 static void clear_standby(struct ceph_connection
*con
)
2544 /* come back from STANDBY? */
2545 if (con
->state
== CON_STATE_STANDBY
) {
2546 dout("clear_standby %p and ++connect_seq\n", con
);
2547 con
->state
= CON_STATE_PREOPEN
;
2549 WARN_ON(con_flag_test(con
, CON_FLAG_WRITE_PENDING
));
2550 WARN_ON(con_flag_test(con
, CON_FLAG_KEEPALIVE_PENDING
));
2555 * Queue up an outgoing message on the given connection.
2557 void ceph_con_send(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2560 msg
->hdr
.src
= con
->msgr
->inst
.name
;
2561 BUG_ON(msg
->front
.iov_len
!= le32_to_cpu(msg
->hdr
.front_len
));
2562 msg
->needs_out_seq
= true;
2564 mutex_lock(&con
->mutex
);
2566 if (con
->state
== CON_STATE_CLOSED
) {
2567 dout("con_send %p closed, dropping %p\n", con
, msg
);
2569 mutex_unlock(&con
->mutex
);
2573 BUG_ON(msg
->con
!= NULL
);
2574 msg
->con
= con
->ops
->get(con
);
2575 BUG_ON(msg
->con
== NULL
);
2577 BUG_ON(!list_empty(&msg
->list_head
));
2578 list_add_tail(&msg
->list_head
, &con
->out_queue
);
2579 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg
,
2580 ENTITY_NAME(con
->peer_name
), le16_to_cpu(msg
->hdr
.type
),
2581 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2582 le32_to_cpu(msg
->hdr
.front_len
),
2583 le32_to_cpu(msg
->hdr
.middle_len
),
2584 le32_to_cpu(msg
->hdr
.data_len
));
2587 mutex_unlock(&con
->mutex
);
2589 /* if there wasn't anything waiting to send before, queue
2591 if (con_flag_test_and_set(con
, CON_FLAG_WRITE_PENDING
) == 0)
2594 EXPORT_SYMBOL(ceph_con_send
);
2597 * Revoke a message that was previously queued for send
2599 void ceph_msg_revoke(struct ceph_msg
*msg
)
2601 struct ceph_connection
*con
= msg
->con
;
2604 return; /* Message not in our possession */
2606 mutex_lock(&con
->mutex
);
2607 if (!list_empty(&msg
->list_head
)) {
2608 dout("%s %p msg %p - was on queue\n", __func__
, con
, msg
);
2609 list_del_init(&msg
->list_head
);
2610 BUG_ON(msg
->con
== NULL
);
2611 msg
->con
->ops
->put(msg
->con
);
2617 if (con
->out_msg
== msg
) {
2618 dout("%s %p msg %p - was sending\n", __func__
, con
, msg
);
2619 con
->out_msg
= NULL
;
2620 if (con
->out_kvec_is_msg
) {
2621 con
->out_skip
= con
->out_kvec_bytes
;
2622 con
->out_kvec_is_msg
= false;
2628 mutex_unlock(&con
->mutex
);
2632 * Revoke a message that we may be reading data into
2634 void ceph_msg_revoke_incoming(struct ceph_msg
*msg
)
2636 struct ceph_connection
*con
;
2638 BUG_ON(msg
== NULL
);
2640 dout("%s msg %p null con\n", __func__
, msg
);
2642 return; /* Message not in our possession */
2646 mutex_lock(&con
->mutex
);
2647 if (con
->in_msg
== msg
) {
2648 unsigned int front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
2649 unsigned int middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
2650 unsigned int data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
2652 /* skip rest of message */
2653 dout("%s %p msg %p revoked\n", __func__
, con
, msg
);
2654 con
->in_base_pos
= con
->in_base_pos
-
2655 sizeof(struct ceph_msg_header
) -
2659 sizeof(struct ceph_msg_footer
);
2660 ceph_msg_put(con
->in_msg
);
2662 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2665 dout("%s %p in_msg %p msg %p no-op\n",
2666 __func__
, con
, con
->in_msg
, msg
);
2668 mutex_unlock(&con
->mutex
);
2672 * Queue a keepalive byte to ensure the tcp connection is alive.
2674 void ceph_con_keepalive(struct ceph_connection
*con
)
2676 dout("con_keepalive %p\n", con
);
2677 mutex_lock(&con
->mutex
);
2679 mutex_unlock(&con
->mutex
);
2680 if (con_flag_test_and_set(con
, CON_FLAG_KEEPALIVE_PENDING
) == 0 &&
2681 con_flag_test_and_set(con
, CON_FLAG_WRITE_PENDING
) == 0)
2684 EXPORT_SYMBOL(ceph_con_keepalive
);
2686 void ceph_msg_data_set_pages(struct ceph_msg
*msg
, struct page
**pages
,
2687 size_t length
, size_t alignment
)
2692 BUG_ON(msg
->length
);
2695 msg
->length
= length
;
2696 msg
->page_alignment
= alignment
& ~PAGE_MASK
;
2698 EXPORT_SYMBOL(ceph_msg_data_set_pages
);
2700 void ceph_msg_data_set_pagelist(struct ceph_msg
*msg
,
2701 struct ceph_pagelist
*pagelist
)
2704 BUG_ON(!pagelist
->length
);
2705 BUG_ON(msg
->pagelist
);
2707 msg
->pagelist
= pagelist
;
2709 EXPORT_SYMBOL(ceph_msg_data_set_pagelist
);
2711 void ceph_msg_data_set_bio(struct ceph_msg
*msg
, struct bio
*bio
)
2718 EXPORT_SYMBOL(ceph_msg_data_set_bio
);
2720 void ceph_msg_data_set_trail(struct ceph_msg
*msg
, struct ceph_pagelist
*trail
)
2723 BUG_ON(!trail
->length
);
2728 EXPORT_SYMBOL(ceph_msg_data_set_trail
);
2731 * construct a new message with given type, size
2732 * the new msg has a ref count of 1.
2734 struct ceph_msg
*ceph_msg_new(int type
, int front_len
, gfp_t flags
,
2739 m
= kzalloc(sizeof(*m
), flags
);
2743 m
->hdr
.type
= cpu_to_le16(type
);
2744 m
->hdr
.priority
= cpu_to_le16(CEPH_MSG_PRIO_DEFAULT
);
2745 m
->hdr
.front_len
= cpu_to_le32(front_len
);
2747 INIT_LIST_HEAD(&m
->list_head
);
2748 kref_init(&m
->kref
);
2751 m
->front_max
= front_len
;
2753 if (front_len
> PAGE_CACHE_SIZE
) {
2754 m
->front
.iov_base
= __vmalloc(front_len
, flags
,
2756 m
->front_is_vmalloc
= true;
2758 m
->front
.iov_base
= kmalloc(front_len
, flags
);
2760 if (m
->front
.iov_base
== NULL
) {
2761 dout("ceph_msg_new can't allocate %d bytes\n",
2766 m
->front
.iov_base
= NULL
;
2768 m
->front
.iov_len
= front_len
;
2770 dout("ceph_msg_new %p front %d\n", m
, front_len
);
2777 pr_err("msg_new can't create type %d front %d\n", type
,
2781 dout("msg_new can't create type %d front %d\n", type
,
2786 EXPORT_SYMBOL(ceph_msg_new
);
2789 * Allocate "middle" portion of a message, if it is needed and wasn't
2790 * allocated by alloc_msg. This allows us to read a small fixed-size
2791 * per-type header in the front and then gracefully fail (i.e.,
2792 * propagate the error to the caller based on info in the front) when
2793 * the middle is too large.
2795 static int ceph_alloc_middle(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2797 int type
= le16_to_cpu(msg
->hdr
.type
);
2798 int middle_len
= le32_to_cpu(msg
->hdr
.middle_len
);
2800 dout("alloc_middle %p type %d %s middle_len %d\n", msg
, type
,
2801 ceph_msg_type_name(type
), middle_len
);
2802 BUG_ON(!middle_len
);
2803 BUG_ON(msg
->middle
);
2805 msg
->middle
= ceph_buffer_new(middle_len
, GFP_NOFS
);
2812 * Allocate a message for receiving an incoming message on a
2813 * connection, and save the result in con->in_msg. Uses the
2814 * connection's private alloc_msg op if available.
2816 * Returns 0 on success, or a negative error code.
2818 * On success, if we set *skip = 1:
2819 * - the next message should be skipped and ignored.
2820 * - con->in_msg == NULL
2821 * or if we set *skip = 0:
2822 * - con->in_msg is non-null.
2823 * On error (ENOMEM, EAGAIN, ...),
2824 * - con->in_msg == NULL
2826 static int ceph_con_in_msg_alloc(struct ceph_connection
*con
, int *skip
)
2828 struct ceph_msg_header
*hdr
= &con
->in_hdr
;
2829 int middle_len
= le32_to_cpu(hdr
->middle_len
);
2830 struct ceph_msg
*msg
;
2833 BUG_ON(con
->in_msg
!= NULL
);
2834 BUG_ON(!con
->ops
->alloc_msg
);
2836 mutex_unlock(&con
->mutex
);
2837 msg
= con
->ops
->alloc_msg(con
, hdr
, skip
);
2838 mutex_lock(&con
->mutex
);
2839 if (con
->state
!= CON_STATE_OPEN
) {
2847 con
->in_msg
->con
= con
->ops
->get(con
);
2848 BUG_ON(con
->in_msg
->con
== NULL
);
2851 * Null message pointer means either we should skip
2852 * this message or we couldn't allocate memory. The
2853 * former is not an error.
2857 con
->error_msg
= "error allocating memory for incoming message";
2861 memcpy(&con
->in_msg
->hdr
, &con
->in_hdr
, sizeof(con
->in_hdr
));
2863 if (middle_len
&& !con
->in_msg
->middle
) {
2864 ret
= ceph_alloc_middle(con
, con
->in_msg
);
2866 ceph_msg_put(con
->in_msg
);
2876 * Free a generically kmalloc'd message.
2878 void ceph_msg_kfree(struct ceph_msg
*m
)
2880 dout("msg_kfree %p\n", m
);
2881 if (m
->front_is_vmalloc
)
2882 vfree(m
->front
.iov_base
);
2884 kfree(m
->front
.iov_base
);
2889 * Drop a msg ref. Destroy as needed.
2891 void ceph_msg_last_put(struct kref
*kref
)
2893 struct ceph_msg
*m
= container_of(kref
, struct ceph_msg
, kref
);
2895 dout("ceph_msg_put last one on %p\n", m
);
2896 WARN_ON(!list_empty(&m
->list_head
));
2898 /* drop middle, data, if any */
2900 ceph_buffer_put(m
->middle
);
2907 ceph_pagelist_release(m
->pagelist
);
2915 ceph_msgpool_put(m
->pool
, m
);
2919 EXPORT_SYMBOL(ceph_msg_last_put
);
2921 void ceph_msg_dump(struct ceph_msg
*msg
)
2923 pr_debug("msg_dump %p (front_max %d length %zd)\n", msg
,
2924 msg
->front_max
, msg
->length
);
2925 print_hex_dump(KERN_DEBUG
, "header: ",
2926 DUMP_PREFIX_OFFSET
, 16, 1,
2927 &msg
->hdr
, sizeof(msg
->hdr
), true);
2928 print_hex_dump(KERN_DEBUG
, " front: ",
2929 DUMP_PREFIX_OFFSET
, 16, 1,
2930 msg
->front
.iov_base
, msg
->front
.iov_len
, true);
2932 print_hex_dump(KERN_DEBUG
, "middle: ",
2933 DUMP_PREFIX_OFFSET
, 16, 1,
2934 msg
->middle
->vec
.iov_base
,
2935 msg
->middle
->vec
.iov_len
, true);
2936 print_hex_dump(KERN_DEBUG
, "footer: ",
2937 DUMP_PREFIX_OFFSET
, 16, 1,
2938 &msg
->footer
, sizeof(msg
->footer
), true);
2940 EXPORT_SYMBOL(ceph_msg_dump
);