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
;
728 struct ceph_msg_pos
*msg_pos
= &con
->out_msg_pos
;
731 BUG_ON(!msg
->hdr
.data_len
);
733 /* initialize page iterator */
736 msg_pos
->page_pos
= msg
->page_alignment
;
738 msg_pos
->page_pos
= 0;
741 init_bio_iter(msg
->bio
, &msg
->bio_iter
, &msg
->bio_seg
);
743 msg_pos
->data_pos
= 0;
744 msg_pos
->did_page_crc
= false;
745 con
->out_more
= 1; /* data + footer will follow */
749 * Prepare footer for currently outgoing message, and finish things
750 * off. Assumes out_kvec* are already valid.. we just add on to the end.
752 static void prepare_write_message_footer(struct ceph_connection
*con
)
754 struct ceph_msg
*m
= con
->out_msg
;
755 int v
= con
->out_kvec_left
;
757 m
->footer
.flags
|= CEPH_MSG_FOOTER_COMPLETE
;
759 dout("prepare_write_message_footer %p\n", con
);
760 con
->out_kvec_is_msg
= true;
761 con
->out_kvec
[v
].iov_base
= &m
->footer
;
762 con
->out_kvec
[v
].iov_len
= sizeof(m
->footer
);
763 con
->out_kvec_bytes
+= sizeof(m
->footer
);
764 con
->out_kvec_left
++;
765 con
->out_more
= m
->more_to_follow
;
766 con
->out_msg_done
= true;
770 * Prepare headers for the next outgoing message.
772 static void prepare_write_message(struct ceph_connection
*con
)
777 con_out_kvec_reset(con
);
778 con
->out_kvec_is_msg
= true;
779 con
->out_msg_done
= false;
781 /* Sneak an ack in there first? If we can get it into the same
782 * TCP packet that's a good thing. */
783 if (con
->in_seq
> con
->in_seq_acked
) {
784 con
->in_seq_acked
= con
->in_seq
;
785 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
786 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
787 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
791 BUG_ON(list_empty(&con
->out_queue
));
792 m
= list_first_entry(&con
->out_queue
, struct ceph_msg
, list_head
);
794 BUG_ON(m
->con
!= con
);
796 /* put message on sent list */
798 list_move_tail(&m
->list_head
, &con
->out_sent
);
801 * only assign outgoing seq # if we haven't sent this message
802 * yet. if it is requeued, resend with it's original seq.
804 if (m
->needs_out_seq
) {
805 m
->hdr
.seq
= cpu_to_le64(++con
->out_seq
);
806 m
->needs_out_seq
= false;
809 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d (%zd)\n",
810 m
, con
->out_seq
, le16_to_cpu(m
->hdr
.type
),
811 le32_to_cpu(m
->hdr
.front_len
), le32_to_cpu(m
->hdr
.middle_len
),
812 le32_to_cpu(m
->hdr
.data_len
), m
->length
);
813 BUG_ON(le32_to_cpu(m
->hdr
.front_len
) != m
->front
.iov_len
);
815 /* tag + hdr + front + middle */
816 con_out_kvec_add(con
, sizeof (tag_msg
), &tag_msg
);
817 con_out_kvec_add(con
, sizeof (m
->hdr
), &m
->hdr
);
818 con_out_kvec_add(con
, m
->front
.iov_len
, m
->front
.iov_base
);
821 con_out_kvec_add(con
, m
->middle
->vec
.iov_len
,
822 m
->middle
->vec
.iov_base
);
824 /* fill in crc (except data pages), footer */
825 crc
= crc32c(0, &m
->hdr
, offsetof(struct ceph_msg_header
, crc
));
826 con
->out_msg
->hdr
.crc
= cpu_to_le32(crc
);
827 con
->out_msg
->footer
.flags
= 0;
829 crc
= crc32c(0, m
->front
.iov_base
, m
->front
.iov_len
);
830 con
->out_msg
->footer
.front_crc
= cpu_to_le32(crc
);
832 crc
= crc32c(0, m
->middle
->vec
.iov_base
,
833 m
->middle
->vec
.iov_len
);
834 con
->out_msg
->footer
.middle_crc
= cpu_to_le32(crc
);
836 con
->out_msg
->footer
.middle_crc
= 0;
837 dout("%s front_crc %u middle_crc %u\n", __func__
,
838 le32_to_cpu(con
->out_msg
->footer
.front_crc
),
839 le32_to_cpu(con
->out_msg
->footer
.middle_crc
));
841 /* is there a data payload? */
842 con
->out_msg
->footer
.data_crc
= 0;
844 prepare_write_message_data(con
);
846 /* no, queue up footer too and be done */
847 prepare_write_message_footer(con
);
849 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
855 static void prepare_write_ack(struct ceph_connection
*con
)
857 dout("prepare_write_ack %p %llu -> %llu\n", con
,
858 con
->in_seq_acked
, con
->in_seq
);
859 con
->in_seq_acked
= con
->in_seq
;
861 con_out_kvec_reset(con
);
863 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
865 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
866 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
869 con
->out_more
= 1; /* more will follow.. eventually.. */
870 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
874 * Prepare to write keepalive byte.
876 static void prepare_write_keepalive(struct ceph_connection
*con
)
878 dout("prepare_write_keepalive %p\n", con
);
879 con_out_kvec_reset(con
);
880 con_out_kvec_add(con
, sizeof (tag_keepalive
), &tag_keepalive
);
881 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
885 * Connection negotiation.
888 static struct ceph_auth_handshake
*get_connect_authorizer(struct ceph_connection
*con
,
891 struct ceph_auth_handshake
*auth
;
893 if (!con
->ops
->get_authorizer
) {
894 con
->out_connect
.authorizer_protocol
= CEPH_AUTH_UNKNOWN
;
895 con
->out_connect
.authorizer_len
= 0;
899 /* Can't hold the mutex while getting authorizer */
900 mutex_unlock(&con
->mutex
);
901 auth
= con
->ops
->get_authorizer(con
, auth_proto
, con
->auth_retry
);
902 mutex_lock(&con
->mutex
);
906 if (con
->state
!= CON_STATE_NEGOTIATING
)
907 return ERR_PTR(-EAGAIN
);
909 con
->auth_reply_buf
= auth
->authorizer_reply_buf
;
910 con
->auth_reply_buf_len
= auth
->authorizer_reply_buf_len
;
915 * We connected to a peer and are saying hello.
917 static void prepare_write_banner(struct ceph_connection
*con
)
919 con_out_kvec_add(con
, strlen(CEPH_BANNER
), CEPH_BANNER
);
920 con_out_kvec_add(con
, sizeof (con
->msgr
->my_enc_addr
),
921 &con
->msgr
->my_enc_addr
);
924 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
927 static int prepare_write_connect(struct ceph_connection
*con
)
929 unsigned int global_seq
= get_global_seq(con
->msgr
, 0);
932 struct ceph_auth_handshake
*auth
;
934 switch (con
->peer_name
.type
) {
935 case CEPH_ENTITY_TYPE_MON
:
936 proto
= CEPH_MONC_PROTOCOL
;
938 case CEPH_ENTITY_TYPE_OSD
:
939 proto
= CEPH_OSDC_PROTOCOL
;
941 case CEPH_ENTITY_TYPE_MDS
:
942 proto
= CEPH_MDSC_PROTOCOL
;
948 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con
,
949 con
->connect_seq
, global_seq
, proto
);
951 con
->out_connect
.features
= cpu_to_le64(con
->msgr
->supported_features
);
952 con
->out_connect
.host_type
= cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT
);
953 con
->out_connect
.connect_seq
= cpu_to_le32(con
->connect_seq
);
954 con
->out_connect
.global_seq
= cpu_to_le32(global_seq
);
955 con
->out_connect
.protocol_version
= cpu_to_le32(proto
);
956 con
->out_connect
.flags
= 0;
958 auth_proto
= CEPH_AUTH_UNKNOWN
;
959 auth
= get_connect_authorizer(con
, &auth_proto
);
961 return PTR_ERR(auth
);
963 con
->out_connect
.authorizer_protocol
= cpu_to_le32(auth_proto
);
964 con
->out_connect
.authorizer_len
= auth
?
965 cpu_to_le32(auth
->authorizer_buf_len
) : 0;
967 con_out_kvec_add(con
, sizeof (con
->out_connect
),
969 if (auth
&& auth
->authorizer_buf_len
)
970 con_out_kvec_add(con
, auth
->authorizer_buf_len
,
971 auth
->authorizer_buf
);
974 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
980 * write as much of pending kvecs to the socket as we can.
982 * 0 -> socket full, but more to do
985 static int write_partial_kvec(struct ceph_connection
*con
)
989 dout("write_partial_kvec %p %d left\n", con
, con
->out_kvec_bytes
);
990 while (con
->out_kvec_bytes
> 0) {
991 ret
= ceph_tcp_sendmsg(con
->sock
, con
->out_kvec_cur
,
992 con
->out_kvec_left
, con
->out_kvec_bytes
,
996 con
->out_kvec_bytes
-= ret
;
997 if (con
->out_kvec_bytes
== 0)
1000 /* account for full iov entries consumed */
1001 while (ret
>= con
->out_kvec_cur
->iov_len
) {
1002 BUG_ON(!con
->out_kvec_left
);
1003 ret
-= con
->out_kvec_cur
->iov_len
;
1004 con
->out_kvec_cur
++;
1005 con
->out_kvec_left
--;
1007 /* and for a partially-consumed entry */
1009 con
->out_kvec_cur
->iov_len
-= ret
;
1010 con
->out_kvec_cur
->iov_base
+= ret
;
1013 con
->out_kvec_left
= 0;
1014 con
->out_kvec_is_msg
= false;
1017 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con
,
1018 con
->out_kvec_bytes
, con
->out_kvec_left
, ret
);
1019 return ret
; /* done! */
1022 static void out_msg_pos_next(struct ceph_connection
*con
, struct page
*page
,
1023 size_t len
, size_t sent
, bool in_trail
)
1025 struct ceph_msg
*msg
= con
->out_msg
;
1026 struct ceph_msg_pos
*msg_pos
= &con
->out_msg_pos
;
1031 msg_pos
->data_pos
+= sent
;
1032 msg_pos
->page_pos
+= sent
;
1036 BUG_ON(sent
!= len
);
1037 msg_pos
->page_pos
= 0;
1039 msg_pos
->did_page_crc
= false;
1041 list_rotate_left(&msg
->trail
->head
);
1042 else if (msg
->pagelist
)
1043 list_rotate_left(&msg
->pagelist
->head
);
1046 iter_bio_next(&msg
->bio_iter
, &msg
->bio_seg
);
1050 static void in_msg_pos_next(struct ceph_connection
*con
, size_t len
,
1053 struct ceph_msg
*msg
= con
->in_msg
;
1054 struct ceph_msg_pos
*msg_pos
= &con
->in_msg_pos
;
1059 msg_pos
->data_pos
+= received
;
1060 msg_pos
->page_pos
+= received
;
1064 BUG_ON(received
!= len
);
1065 msg_pos
->page_pos
= 0;
1069 iter_bio_next(&msg
->bio_iter
, &msg
->bio_seg
);
1070 #endif /* CONFIG_BLOCK */
1074 * Write as much message data payload as we can. If we finish, queue
1076 * 1 -> done, footer is now queued in out_kvec[].
1077 * 0 -> socket full, but more to do
1080 static int write_partial_msg_pages(struct ceph_connection
*con
)
1082 struct ceph_msg
*msg
= con
->out_msg
;
1083 struct ceph_msg_pos
*msg_pos
= &con
->out_msg_pos
;
1084 unsigned int data_len
= le32_to_cpu(msg
->hdr
.data_len
);
1086 bool do_datacrc
= !con
->msgr
->nocrc
;
1088 int total_max_write
;
1089 bool in_trail
= false;
1090 const size_t trail_len
= (msg
->trail
? msg
->trail
->length
: 0);
1091 const size_t trail_off
= data_len
- trail_len
;
1093 dout("write_partial_msg_pages %p msg %p page %d offset %d\n",
1094 con
, msg
, msg_pos
->page
, msg_pos
->page_pos
);
1097 * Iterate through each page that contains data to be
1098 * written, and send as much as possible for each.
1100 * If we are calculating the data crc (the default), we will
1101 * need to map the page. If we have no pages, they have
1102 * been revoked, so use the zero page.
1104 while (data_len
> msg_pos
->data_pos
) {
1105 struct page
*page
= NULL
;
1106 int max_write
= PAGE_SIZE
;
1109 in_trail
= in_trail
|| msg_pos
->data_pos
>= trail_off
;
1111 total_max_write
= trail_off
- msg_pos
->data_pos
;
1114 total_max_write
= data_len
- msg_pos
->data_pos
;
1116 page
= list_first_entry(&msg
->trail
->head
,
1118 } else if (msg
->pages
) {
1119 page
= msg
->pages
[msg_pos
->page
];
1120 } else if (msg
->pagelist
) {
1121 page
= list_first_entry(&msg
->pagelist
->head
,
1124 } else if (msg
->bio
) {
1127 bv
= bio_iovec_idx(msg
->bio_iter
, msg
->bio_seg
);
1129 bio_offset
= bv
->bv_offset
;
1130 max_write
= bv
->bv_len
;
1135 len
= min_t(int, max_write
- msg_pos
->page_pos
,
1138 if (do_datacrc
&& !msg_pos
->did_page_crc
) {
1140 u32 crc
= le32_to_cpu(msg
->footer
.data_crc
);
1144 BUG_ON(kaddr
== NULL
);
1145 base
= kaddr
+ msg_pos
->page_pos
+ bio_offset
;
1146 crc
= crc32c(crc
, base
, len
);
1148 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1149 msg_pos
->did_page_crc
= true;
1151 ret
= ceph_tcp_sendpage(con
->sock
, page
,
1152 msg_pos
->page_pos
+ bio_offset
,
1157 out_msg_pos_next(con
, page
, len
, (size_t) ret
, in_trail
);
1160 dout("write_partial_msg_pages %p msg %p done\n", con
, msg
);
1162 /* prepare and queue up footer, too */
1164 msg
->footer
.flags
|= CEPH_MSG_FOOTER_NOCRC
;
1165 con_out_kvec_reset(con
);
1166 prepare_write_message_footer(con
);
1175 static int write_partial_skip(struct ceph_connection
*con
)
1179 while (con
->out_skip
> 0) {
1180 size_t size
= min(con
->out_skip
, (int) PAGE_CACHE_SIZE
);
1182 ret
= ceph_tcp_sendpage(con
->sock
, zero_page
, 0, size
, true);
1185 con
->out_skip
-= ret
;
1193 * Prepare to read connection handshake, or an ack.
1195 static void prepare_read_banner(struct ceph_connection
*con
)
1197 dout("prepare_read_banner %p\n", con
);
1198 con
->in_base_pos
= 0;
1201 static void prepare_read_connect(struct ceph_connection
*con
)
1203 dout("prepare_read_connect %p\n", con
);
1204 con
->in_base_pos
= 0;
1207 static void prepare_read_ack(struct ceph_connection
*con
)
1209 dout("prepare_read_ack %p\n", con
);
1210 con
->in_base_pos
= 0;
1213 static void prepare_read_tag(struct ceph_connection
*con
)
1215 dout("prepare_read_tag %p\n", con
);
1216 con
->in_base_pos
= 0;
1217 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1221 * Prepare to read a message.
1223 static int prepare_read_message(struct ceph_connection
*con
)
1225 dout("prepare_read_message %p\n", con
);
1226 BUG_ON(con
->in_msg
!= NULL
);
1227 con
->in_base_pos
= 0;
1228 con
->in_front_crc
= con
->in_middle_crc
= con
->in_data_crc
= 0;
1233 static int read_partial(struct ceph_connection
*con
,
1234 int end
, int size
, void *object
)
1236 while (con
->in_base_pos
< end
) {
1237 int left
= end
- con
->in_base_pos
;
1238 int have
= size
- left
;
1239 int ret
= ceph_tcp_recvmsg(con
->sock
, object
+ have
, left
);
1242 con
->in_base_pos
+= ret
;
1249 * Read all or part of the connect-side handshake on a new connection
1251 static int read_partial_banner(struct ceph_connection
*con
)
1257 dout("read_partial_banner %p at %d\n", con
, con
->in_base_pos
);
1260 size
= strlen(CEPH_BANNER
);
1262 ret
= read_partial(con
, end
, size
, con
->in_banner
);
1266 size
= sizeof (con
->actual_peer_addr
);
1268 ret
= read_partial(con
, end
, size
, &con
->actual_peer_addr
);
1272 size
= sizeof (con
->peer_addr_for_me
);
1274 ret
= read_partial(con
, end
, size
, &con
->peer_addr_for_me
);
1282 static int read_partial_connect(struct ceph_connection
*con
)
1288 dout("read_partial_connect %p at %d\n", con
, con
->in_base_pos
);
1290 size
= sizeof (con
->in_reply
);
1292 ret
= read_partial(con
, end
, size
, &con
->in_reply
);
1296 size
= le32_to_cpu(con
->in_reply
.authorizer_len
);
1298 ret
= read_partial(con
, end
, size
, con
->auth_reply_buf
);
1302 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1303 con
, (int)con
->in_reply
.tag
,
1304 le32_to_cpu(con
->in_reply
.connect_seq
),
1305 le32_to_cpu(con
->in_reply
.global_seq
));
1312 * Verify the hello banner looks okay.
1314 static int verify_hello(struct ceph_connection
*con
)
1316 if (memcmp(con
->in_banner
, CEPH_BANNER
, strlen(CEPH_BANNER
))) {
1317 pr_err("connect to %s got bad banner\n",
1318 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1319 con
->error_msg
= "protocol error, bad banner";
1325 static bool addr_is_blank(struct sockaddr_storage
*ss
)
1327 switch (ss
->ss_family
) {
1329 return ((struct sockaddr_in
*)ss
)->sin_addr
.s_addr
== 0;
1332 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[0] == 0 &&
1333 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[1] == 0 &&
1334 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[2] == 0 &&
1335 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[3] == 0;
1340 static int addr_port(struct sockaddr_storage
*ss
)
1342 switch (ss
->ss_family
) {
1344 return ntohs(((struct sockaddr_in
*)ss
)->sin_port
);
1346 return ntohs(((struct sockaddr_in6
*)ss
)->sin6_port
);
1351 static void addr_set_port(struct sockaddr_storage
*ss
, int p
)
1353 switch (ss
->ss_family
) {
1355 ((struct sockaddr_in
*)ss
)->sin_port
= htons(p
);
1358 ((struct sockaddr_in6
*)ss
)->sin6_port
= htons(p
);
1364 * Unlike other *_pton function semantics, zero indicates success.
1366 static int ceph_pton(const char *str
, size_t len
, struct sockaddr_storage
*ss
,
1367 char delim
, const char **ipend
)
1369 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
1370 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
1372 memset(ss
, 0, sizeof(*ss
));
1374 if (in4_pton(str
, len
, (u8
*)&in4
->sin_addr
.s_addr
, delim
, ipend
)) {
1375 ss
->ss_family
= AF_INET
;
1379 if (in6_pton(str
, len
, (u8
*)&in6
->sin6_addr
.s6_addr
, delim
, ipend
)) {
1380 ss
->ss_family
= AF_INET6
;
1388 * Extract hostname string and resolve using kernel DNS facility.
1390 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1391 static int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1392 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1394 const char *end
, *delim_p
;
1395 char *colon_p
, *ip_addr
= NULL
;
1399 * The end of the hostname occurs immediately preceding the delimiter or
1400 * the port marker (':') where the delimiter takes precedence.
1402 delim_p
= memchr(name
, delim
, namelen
);
1403 colon_p
= memchr(name
, ':', namelen
);
1405 if (delim_p
&& colon_p
)
1406 end
= delim_p
< colon_p
? delim_p
: colon_p
;
1407 else if (!delim_p
&& colon_p
)
1411 if (!end
) /* case: hostname:/ */
1412 end
= name
+ namelen
;
1418 /* do dns_resolve upcall */
1419 ip_len
= dns_query(NULL
, name
, end
- name
, NULL
, &ip_addr
, NULL
);
1421 ret
= ceph_pton(ip_addr
, ip_len
, ss
, -1, NULL
);
1429 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end
- name
), name
,
1430 ret
, ret
? "failed" : ceph_pr_addr(ss
));
1435 static inline int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1436 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1443 * Parse a server name (IP or hostname). If a valid IP address is not found
1444 * then try to extract a hostname to resolve using userspace DNS upcall.
1446 static int ceph_parse_server_name(const char *name
, size_t namelen
,
1447 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1451 ret
= ceph_pton(name
, namelen
, ss
, delim
, ipend
);
1453 ret
= ceph_dns_resolve_name(name
, namelen
, ss
, delim
, ipend
);
1459 * Parse an ip[:port] list into an addr array. Use the default
1460 * monitor port if a port isn't specified.
1462 int ceph_parse_ips(const char *c
, const char *end
,
1463 struct ceph_entity_addr
*addr
,
1464 int max_count
, int *count
)
1466 int i
, ret
= -EINVAL
;
1469 dout("parse_ips on '%.*s'\n", (int)(end
-c
), c
);
1470 for (i
= 0; i
< max_count
; i
++) {
1472 struct sockaddr_storage
*ss
= &addr
[i
].in_addr
;
1481 ret
= ceph_parse_server_name(p
, end
- p
, ss
, delim
, &ipend
);
1490 dout("missing matching ']'\n");
1497 if (p
< end
&& *p
== ':') {
1500 while (p
< end
&& *p
>= '0' && *p
<= '9') {
1501 port
= (port
* 10) + (*p
- '0');
1504 if (port
> 65535 || port
== 0)
1507 port
= CEPH_MON_PORT
;
1510 addr_set_port(ss
, port
);
1512 dout("parse_ips got %s\n", ceph_pr_addr(ss
));
1529 pr_err("parse_ips bad ip '%.*s'\n", (int)(end
- c
), c
);
1532 EXPORT_SYMBOL(ceph_parse_ips
);
1534 static int process_banner(struct ceph_connection
*con
)
1536 dout("process_banner on %p\n", con
);
1538 if (verify_hello(con
) < 0)
1541 ceph_decode_addr(&con
->actual_peer_addr
);
1542 ceph_decode_addr(&con
->peer_addr_for_me
);
1545 * Make sure the other end is who we wanted. note that the other
1546 * end may not yet know their ip address, so if it's 0.0.0.0, give
1547 * them the benefit of the doubt.
1549 if (memcmp(&con
->peer_addr
, &con
->actual_peer_addr
,
1550 sizeof(con
->peer_addr
)) != 0 &&
1551 !(addr_is_blank(&con
->actual_peer_addr
.in_addr
) &&
1552 con
->actual_peer_addr
.nonce
== con
->peer_addr
.nonce
)) {
1553 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1554 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1555 (int)le32_to_cpu(con
->peer_addr
.nonce
),
1556 ceph_pr_addr(&con
->actual_peer_addr
.in_addr
),
1557 (int)le32_to_cpu(con
->actual_peer_addr
.nonce
));
1558 con
->error_msg
= "wrong peer at address";
1563 * did we learn our address?
1565 if (addr_is_blank(&con
->msgr
->inst
.addr
.in_addr
)) {
1566 int port
= addr_port(&con
->msgr
->inst
.addr
.in_addr
);
1568 memcpy(&con
->msgr
->inst
.addr
.in_addr
,
1569 &con
->peer_addr_for_me
.in_addr
,
1570 sizeof(con
->peer_addr_for_me
.in_addr
));
1571 addr_set_port(&con
->msgr
->inst
.addr
.in_addr
, port
);
1572 encode_my_addr(con
->msgr
);
1573 dout("process_banner learned my addr is %s\n",
1574 ceph_pr_addr(&con
->msgr
->inst
.addr
.in_addr
));
1580 static int process_connect(struct ceph_connection
*con
)
1582 u64 sup_feat
= con
->msgr
->supported_features
;
1583 u64 req_feat
= con
->msgr
->required_features
;
1584 u64 server_feat
= le64_to_cpu(con
->in_reply
.features
);
1587 dout("process_connect on %p tag %d\n", con
, (int)con
->in_tag
);
1589 switch (con
->in_reply
.tag
) {
1590 case CEPH_MSGR_TAG_FEATURES
:
1591 pr_err("%s%lld %s feature set mismatch,"
1592 " my %llx < server's %llx, missing %llx\n",
1593 ENTITY_NAME(con
->peer_name
),
1594 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1595 sup_feat
, server_feat
, server_feat
& ~sup_feat
);
1596 con
->error_msg
= "missing required protocol features";
1597 reset_connection(con
);
1600 case CEPH_MSGR_TAG_BADPROTOVER
:
1601 pr_err("%s%lld %s protocol version mismatch,"
1602 " my %d != server's %d\n",
1603 ENTITY_NAME(con
->peer_name
),
1604 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1605 le32_to_cpu(con
->out_connect
.protocol_version
),
1606 le32_to_cpu(con
->in_reply
.protocol_version
));
1607 con
->error_msg
= "protocol version mismatch";
1608 reset_connection(con
);
1611 case CEPH_MSGR_TAG_BADAUTHORIZER
:
1613 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con
,
1615 if (con
->auth_retry
== 2) {
1616 con
->error_msg
= "connect authorization failure";
1619 con
->auth_retry
= 1;
1620 con_out_kvec_reset(con
);
1621 ret
= prepare_write_connect(con
);
1624 prepare_read_connect(con
);
1627 case CEPH_MSGR_TAG_RESETSESSION
:
1629 * If we connected with a large connect_seq but the peer
1630 * has no record of a session with us (no connection, or
1631 * connect_seq == 0), they will send RESETSESION to indicate
1632 * that they must have reset their session, and may have
1635 dout("process_connect got RESET peer seq %u\n",
1636 le32_to_cpu(con
->in_reply
.connect_seq
));
1637 pr_err("%s%lld %s connection reset\n",
1638 ENTITY_NAME(con
->peer_name
),
1639 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1640 reset_connection(con
);
1641 con_out_kvec_reset(con
);
1642 ret
= prepare_write_connect(con
);
1645 prepare_read_connect(con
);
1647 /* Tell ceph about it. */
1648 mutex_unlock(&con
->mutex
);
1649 pr_info("reset on %s%lld\n", ENTITY_NAME(con
->peer_name
));
1650 if (con
->ops
->peer_reset
)
1651 con
->ops
->peer_reset(con
);
1652 mutex_lock(&con
->mutex
);
1653 if (con
->state
!= CON_STATE_NEGOTIATING
)
1657 case CEPH_MSGR_TAG_RETRY_SESSION
:
1659 * If we sent a smaller connect_seq than the peer has, try
1660 * again with a larger value.
1662 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
1663 le32_to_cpu(con
->out_connect
.connect_seq
),
1664 le32_to_cpu(con
->in_reply
.connect_seq
));
1665 con
->connect_seq
= le32_to_cpu(con
->in_reply
.connect_seq
);
1666 con_out_kvec_reset(con
);
1667 ret
= prepare_write_connect(con
);
1670 prepare_read_connect(con
);
1673 case CEPH_MSGR_TAG_RETRY_GLOBAL
:
1675 * If we sent a smaller global_seq than the peer has, try
1676 * again with a larger value.
1678 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1679 con
->peer_global_seq
,
1680 le32_to_cpu(con
->in_reply
.global_seq
));
1681 get_global_seq(con
->msgr
,
1682 le32_to_cpu(con
->in_reply
.global_seq
));
1683 con_out_kvec_reset(con
);
1684 ret
= prepare_write_connect(con
);
1687 prepare_read_connect(con
);
1690 case CEPH_MSGR_TAG_READY
:
1691 if (req_feat
& ~server_feat
) {
1692 pr_err("%s%lld %s protocol feature mismatch,"
1693 " my required %llx > server's %llx, need %llx\n",
1694 ENTITY_NAME(con
->peer_name
),
1695 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1696 req_feat
, server_feat
, req_feat
& ~server_feat
);
1697 con
->error_msg
= "missing required protocol features";
1698 reset_connection(con
);
1702 WARN_ON(con
->state
!= CON_STATE_NEGOTIATING
);
1703 con
->state
= CON_STATE_OPEN
;
1705 con
->peer_global_seq
= le32_to_cpu(con
->in_reply
.global_seq
);
1707 con
->peer_features
= server_feat
;
1708 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1709 con
->peer_global_seq
,
1710 le32_to_cpu(con
->in_reply
.connect_seq
),
1712 WARN_ON(con
->connect_seq
!=
1713 le32_to_cpu(con
->in_reply
.connect_seq
));
1715 if (con
->in_reply
.flags
& CEPH_MSG_CONNECT_LOSSY
)
1716 con_flag_set(con
, CON_FLAG_LOSSYTX
);
1718 con
->delay
= 0; /* reset backoff memory */
1720 prepare_read_tag(con
);
1723 case CEPH_MSGR_TAG_WAIT
:
1725 * If there is a connection race (we are opening
1726 * connections to each other), one of us may just have
1727 * to WAIT. This shouldn't happen if we are the
1730 pr_err("process_connect got WAIT as client\n");
1731 con
->error_msg
= "protocol error, got WAIT as client";
1735 pr_err("connect protocol error, will retry\n");
1736 con
->error_msg
= "protocol error, garbage tag during connect";
1744 * read (part of) an ack
1746 static int read_partial_ack(struct ceph_connection
*con
)
1748 int size
= sizeof (con
->in_temp_ack
);
1751 return read_partial(con
, end
, size
, &con
->in_temp_ack
);
1756 * We can finally discard anything that's been acked.
1758 static void process_ack(struct ceph_connection
*con
)
1761 u64 ack
= le64_to_cpu(con
->in_temp_ack
);
1764 while (!list_empty(&con
->out_sent
)) {
1765 m
= list_first_entry(&con
->out_sent
, struct ceph_msg
,
1767 seq
= le64_to_cpu(m
->hdr
.seq
);
1770 dout("got ack for seq %llu type %d at %p\n", seq
,
1771 le16_to_cpu(m
->hdr
.type
), m
);
1772 m
->ack_stamp
= jiffies
;
1775 prepare_read_tag(con
);
1781 static int read_partial_message_section(struct ceph_connection
*con
,
1782 struct kvec
*section
,
1783 unsigned int sec_len
, u32
*crc
)
1789 while (section
->iov_len
< sec_len
) {
1790 BUG_ON(section
->iov_base
== NULL
);
1791 left
= sec_len
- section
->iov_len
;
1792 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)section
->iov_base
+
1793 section
->iov_len
, left
);
1796 section
->iov_len
+= ret
;
1798 if (section
->iov_len
== sec_len
)
1799 *crc
= crc32c(0, section
->iov_base
, section
->iov_len
);
1804 static int ceph_con_in_msg_alloc(struct ceph_connection
*con
, int *skip
);
1806 static int read_partial_message_pages(struct ceph_connection
*con
,
1807 struct page
**pages
,
1808 unsigned int data_len
, bool do_datacrc
)
1810 struct ceph_msg_pos
*msg_pos
= &con
->in_msg_pos
;
1816 left
= min((int)(data_len
- msg_pos
->data_pos
),
1817 (int)(PAGE_SIZE
- msg_pos
->page_pos
));
1819 BUG_ON(pages
== NULL
);
1820 page
= pages
[msg_pos
->page
];
1822 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ msg_pos
->page_pos
, left
);
1823 if (ret
> 0 && do_datacrc
)
1825 crc32c(con
->in_data_crc
,
1826 p
+ msg_pos
->page_pos
, ret
);
1831 in_msg_pos_next(con
, left
, ret
);
1837 static int read_partial_message_bio(struct ceph_connection
*con
,
1838 unsigned int data_len
, bool do_datacrc
)
1840 struct ceph_msg
*msg
= con
->in_msg
;
1841 struct ceph_msg_pos
*msg_pos
= &con
->in_msg_pos
;
1848 BUG_ON(!msg
->bio_iter
);
1849 bv
= bio_iovec_idx(msg
->bio_iter
, msg
->bio_seg
);
1851 left
= min((int)(data_len
- msg_pos
->data_pos
),
1852 (int)(bv
->bv_len
- msg_pos
->page_pos
));
1855 p
= kmap(page
) + bv
->bv_offset
;
1857 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ msg_pos
->page_pos
, left
);
1858 if (ret
> 0 && do_datacrc
)
1860 crc32c(con
->in_data_crc
,
1861 p
+ msg_pos
->page_pos
, ret
);
1866 in_msg_pos_next(con
, left
, ret
);
1873 * read (part of) a message.
1875 static int read_partial_message(struct ceph_connection
*con
)
1877 struct ceph_msg
*m
= con
->in_msg
;
1878 struct ceph_msg_pos
*msg_pos
= &con
->in_msg_pos
;
1882 unsigned int front_len
, middle_len
, data_len
;
1883 bool do_datacrc
= !con
->msgr
->nocrc
;
1887 dout("read_partial_message con %p msg %p\n", con
, m
);
1890 size
= sizeof (con
->in_hdr
);
1892 ret
= read_partial(con
, end
, size
, &con
->in_hdr
);
1896 crc
= crc32c(0, &con
->in_hdr
, offsetof(struct ceph_msg_header
, crc
));
1897 if (cpu_to_le32(crc
) != con
->in_hdr
.crc
) {
1898 pr_err("read_partial_message bad hdr "
1899 " crc %u != expected %u\n",
1900 crc
, con
->in_hdr
.crc
);
1904 front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
1905 if (front_len
> CEPH_MSG_MAX_FRONT_LEN
)
1907 middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
1908 if (middle_len
> CEPH_MSG_MAX_MIDDLE_LEN
)
1910 data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
1911 if (data_len
> CEPH_MSG_MAX_DATA_LEN
)
1915 seq
= le64_to_cpu(con
->in_hdr
.seq
);
1916 if ((s64
)seq
- (s64
)con
->in_seq
< 1) {
1917 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1918 ENTITY_NAME(con
->peer_name
),
1919 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1920 seq
, con
->in_seq
+ 1);
1921 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1923 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1925 } else if ((s64
)seq
- (s64
)con
->in_seq
> 1) {
1926 pr_err("read_partial_message bad seq %lld expected %lld\n",
1927 seq
, con
->in_seq
+ 1);
1928 con
->error_msg
= "bad message sequence # for incoming message";
1932 /* allocate message? */
1936 dout("got hdr type %d front %d data %d\n", con
->in_hdr
.type
,
1937 front_len
, data_len
);
1938 ret
= ceph_con_in_msg_alloc(con
, &skip
);
1942 /* skip this message */
1943 dout("alloc_msg said skip message\n");
1944 BUG_ON(con
->in_msg
);
1945 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1947 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1952 BUG_ON(!con
->in_msg
);
1953 BUG_ON(con
->in_msg
->con
!= con
);
1955 m
->front
.iov_len
= 0; /* haven't read it yet */
1957 m
->middle
->vec
.iov_len
= 0;
1961 msg_pos
->page_pos
= m
->page_alignment
;
1963 msg_pos
->page_pos
= 0;
1964 msg_pos
->data_pos
= 0;
1968 init_bio_iter(m
->bio
, &m
->bio_iter
, &m
->bio_seg
);
1973 ret
= read_partial_message_section(con
, &m
->front
, front_len
,
1974 &con
->in_front_crc
);
1980 ret
= read_partial_message_section(con
, &m
->middle
->vec
,
1982 &con
->in_middle_crc
);
1988 while (msg_pos
->data_pos
< data_len
) {
1990 ret
= read_partial_message_pages(con
, m
->pages
,
1991 data_len
, do_datacrc
);
1995 } else if (m
->bio
) {
1996 ret
= read_partial_message_bio(con
,
1997 data_len
, do_datacrc
);
2007 size
= sizeof (m
->footer
);
2009 ret
= read_partial(con
, end
, size
, &m
->footer
);
2013 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2014 m
, front_len
, m
->footer
.front_crc
, middle_len
,
2015 m
->footer
.middle_crc
, data_len
, m
->footer
.data_crc
);
2018 if (con
->in_front_crc
!= le32_to_cpu(m
->footer
.front_crc
)) {
2019 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2020 m
, con
->in_front_crc
, m
->footer
.front_crc
);
2023 if (con
->in_middle_crc
!= le32_to_cpu(m
->footer
.middle_crc
)) {
2024 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2025 m
, con
->in_middle_crc
, m
->footer
.middle_crc
);
2029 (m
->footer
.flags
& CEPH_MSG_FOOTER_NOCRC
) == 0 &&
2030 con
->in_data_crc
!= le32_to_cpu(m
->footer
.data_crc
)) {
2031 pr_err("read_partial_message %p data crc %u != exp. %u\n", m
,
2032 con
->in_data_crc
, le32_to_cpu(m
->footer
.data_crc
));
2036 return 1; /* done! */
2040 * Process message. This happens in the worker thread. The callback should
2041 * be careful not to do anything that waits on other incoming messages or it
2044 static void process_message(struct ceph_connection
*con
)
2046 struct ceph_msg
*msg
;
2048 BUG_ON(con
->in_msg
->con
!= con
);
2049 con
->in_msg
->con
= NULL
;
2054 /* if first message, set peer_name */
2055 if (con
->peer_name
.type
== 0)
2056 con
->peer_name
= msg
->hdr
.src
;
2059 mutex_unlock(&con
->mutex
);
2061 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2062 msg
, le64_to_cpu(msg
->hdr
.seq
),
2063 ENTITY_NAME(msg
->hdr
.src
),
2064 le16_to_cpu(msg
->hdr
.type
),
2065 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2066 le32_to_cpu(msg
->hdr
.front_len
),
2067 le32_to_cpu(msg
->hdr
.data_len
),
2068 con
->in_front_crc
, con
->in_middle_crc
, con
->in_data_crc
);
2069 con
->ops
->dispatch(con
, msg
);
2071 mutex_lock(&con
->mutex
);
2076 * Write something to the socket. Called in a worker thread when the
2077 * socket appears to be writeable and we have something ready to send.
2079 static int try_write(struct ceph_connection
*con
)
2083 dout("try_write start %p state %lu\n", con
, con
->state
);
2086 dout("try_write out_kvec_bytes %d\n", con
->out_kvec_bytes
);
2088 /* open the socket first? */
2089 if (con
->state
== CON_STATE_PREOPEN
) {
2091 con
->state
= CON_STATE_CONNECTING
;
2093 con_out_kvec_reset(con
);
2094 prepare_write_banner(con
);
2095 prepare_read_banner(con
);
2097 BUG_ON(con
->in_msg
);
2098 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2099 dout("try_write initiating connect on %p new state %lu\n",
2101 ret
= ceph_tcp_connect(con
);
2103 con
->error_msg
= "connect error";
2109 /* kvec data queued? */
2110 if (con
->out_skip
) {
2111 ret
= write_partial_skip(con
);
2115 if (con
->out_kvec_left
) {
2116 ret
= write_partial_kvec(con
);
2123 if (con
->out_msg_done
) {
2124 ceph_msg_put(con
->out_msg
);
2125 con
->out_msg
= NULL
; /* we're done with this one */
2129 ret
= write_partial_msg_pages(con
);
2131 goto more_kvec
; /* we need to send the footer, too! */
2135 dout("try_write write_partial_msg_pages err %d\n",
2142 if (con
->state
== CON_STATE_OPEN
) {
2143 /* is anything else pending? */
2144 if (!list_empty(&con
->out_queue
)) {
2145 prepare_write_message(con
);
2148 if (con
->in_seq
> con
->in_seq_acked
) {
2149 prepare_write_ack(con
);
2152 if (con_flag_test_and_clear(con
, CON_FLAG_KEEPALIVE_PENDING
)) {
2153 prepare_write_keepalive(con
);
2158 /* Nothing to do! */
2159 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
2160 dout("try_write nothing else to write.\n");
2163 dout("try_write done on %p ret %d\n", con
, ret
);
2170 * Read what we can from the socket.
2172 static int try_read(struct ceph_connection
*con
)
2177 dout("try_read start on %p state %lu\n", con
, con
->state
);
2178 if (con
->state
!= CON_STATE_CONNECTING
&&
2179 con
->state
!= CON_STATE_NEGOTIATING
&&
2180 con
->state
!= CON_STATE_OPEN
)
2185 dout("try_read tag %d in_base_pos %d\n", (int)con
->in_tag
,
2188 if (con
->state
== CON_STATE_CONNECTING
) {
2189 dout("try_read connecting\n");
2190 ret
= read_partial_banner(con
);
2193 ret
= process_banner(con
);
2197 con
->state
= CON_STATE_NEGOTIATING
;
2200 * Received banner is good, exchange connection info.
2201 * Do not reset out_kvec, as sending our banner raced
2202 * with receiving peer banner after connect completed.
2204 ret
= prepare_write_connect(con
);
2207 prepare_read_connect(con
);
2209 /* Send connection info before awaiting response */
2213 if (con
->state
== CON_STATE_NEGOTIATING
) {
2214 dout("try_read negotiating\n");
2215 ret
= read_partial_connect(con
);
2218 ret
= process_connect(con
);
2224 WARN_ON(con
->state
!= CON_STATE_OPEN
);
2226 if (con
->in_base_pos
< 0) {
2228 * skipping + discarding content.
2230 * FIXME: there must be a better way to do this!
2232 static char buf
[SKIP_BUF_SIZE
];
2233 int skip
= min((int) sizeof (buf
), -con
->in_base_pos
);
2235 dout("skipping %d / %d bytes\n", skip
, -con
->in_base_pos
);
2236 ret
= ceph_tcp_recvmsg(con
->sock
, buf
, skip
);
2239 con
->in_base_pos
+= ret
;
2240 if (con
->in_base_pos
)
2243 if (con
->in_tag
== CEPH_MSGR_TAG_READY
) {
2247 ret
= ceph_tcp_recvmsg(con
->sock
, &con
->in_tag
, 1);
2250 dout("try_read got tag %d\n", (int)con
->in_tag
);
2251 switch (con
->in_tag
) {
2252 case CEPH_MSGR_TAG_MSG
:
2253 prepare_read_message(con
);
2255 case CEPH_MSGR_TAG_ACK
:
2256 prepare_read_ack(con
);
2258 case CEPH_MSGR_TAG_CLOSE
:
2259 con_close_socket(con
);
2260 con
->state
= CON_STATE_CLOSED
;
2266 if (con
->in_tag
== CEPH_MSGR_TAG_MSG
) {
2267 ret
= read_partial_message(con
);
2271 con
->error_msg
= "bad crc";
2275 con
->error_msg
= "io error";
2280 if (con
->in_tag
== CEPH_MSGR_TAG_READY
)
2282 process_message(con
);
2283 if (con
->state
== CON_STATE_OPEN
)
2284 prepare_read_tag(con
);
2287 if (con
->in_tag
== CEPH_MSGR_TAG_ACK
) {
2288 ret
= read_partial_ack(con
);
2296 dout("try_read done on %p ret %d\n", con
, ret
);
2300 pr_err("try_read bad con->in_tag = %d\n", (int)con
->in_tag
);
2301 con
->error_msg
= "protocol error, garbage tag";
2308 * Atomically queue work on a connection after the specified delay.
2309 * Bump @con reference to avoid races with connection teardown.
2310 * Returns 0 if work was queued, or an error code otherwise.
2312 static int queue_con_delay(struct ceph_connection
*con
, unsigned long delay
)
2314 if (!con
->ops
->get(con
)) {
2315 dout("%s %p ref count 0\n", __func__
, con
);
2320 if (!queue_delayed_work(ceph_msgr_wq
, &con
->work
, delay
)) {
2321 dout("%s %p - already queued\n", __func__
, con
);
2327 dout("%s %p %lu\n", __func__
, con
, delay
);
2332 static void queue_con(struct ceph_connection
*con
)
2334 (void) queue_con_delay(con
, 0);
2337 static bool con_sock_closed(struct ceph_connection
*con
)
2339 if (!con_flag_test_and_clear(con
, CON_FLAG_SOCK_CLOSED
))
2343 case CON_STATE_ ## x: \
2344 con->error_msg = "socket closed (con state " #x ")"; \
2347 switch (con
->state
) {
2355 pr_warning("%s con %p unrecognized state %lu\n",
2356 __func__
, con
, con
->state
);
2357 con
->error_msg
= "unrecognized con state";
2366 static bool con_backoff(struct ceph_connection
*con
)
2370 if (!con_flag_test_and_clear(con
, CON_FLAG_BACKOFF
))
2373 ret
= queue_con_delay(con
, round_jiffies_relative(con
->delay
));
2375 dout("%s: con %p FAILED to back off %lu\n", __func__
,
2377 BUG_ON(ret
== -ENOENT
);
2378 con_flag_set(con
, CON_FLAG_BACKOFF
);
2384 /* Finish fault handling; con->mutex must *not* be held here */
2386 static void con_fault_finish(struct ceph_connection
*con
)
2389 * in case we faulted due to authentication, invalidate our
2390 * current tickets so that we can get new ones.
2392 if (con
->auth_retry
&& con
->ops
->invalidate_authorizer
) {
2393 dout("calling invalidate_authorizer()\n");
2394 con
->ops
->invalidate_authorizer(con
);
2397 if (con
->ops
->fault
)
2398 con
->ops
->fault(con
);
2402 * Do some work on a connection. Drop a connection ref when we're done.
2404 static void con_work(struct work_struct
*work
)
2406 struct ceph_connection
*con
= container_of(work
, struct ceph_connection
,
2410 mutex_lock(&con
->mutex
);
2414 if ((fault
= con_sock_closed(con
))) {
2415 dout("%s: con %p SOCK_CLOSED\n", __func__
, con
);
2418 if (con_backoff(con
)) {
2419 dout("%s: con %p BACKOFF\n", __func__
, con
);
2422 if (con
->state
== CON_STATE_STANDBY
) {
2423 dout("%s: con %p STANDBY\n", __func__
, con
);
2426 if (con
->state
== CON_STATE_CLOSED
) {
2427 dout("%s: con %p CLOSED\n", __func__
, con
);
2431 if (con
->state
== CON_STATE_PREOPEN
) {
2432 dout("%s: con %p PREOPEN\n", __func__
, con
);
2436 ret
= try_read(con
);
2440 con
->error_msg
= "socket error on read";
2445 ret
= try_write(con
);
2449 con
->error_msg
= "socket error on write";
2453 break; /* If we make it to here, we're done */
2457 mutex_unlock(&con
->mutex
);
2460 con_fault_finish(con
);
2466 * Generic error/fault handler. A retry mechanism is used with
2467 * exponential backoff
2469 static void con_fault(struct ceph_connection
*con
)
2471 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con
->peer_name
),
2472 ceph_pr_addr(&con
->peer_addr
.in_addr
), con
->error_msg
);
2473 dout("fault %p state %lu to peer %s\n",
2474 con
, con
->state
, ceph_pr_addr(&con
->peer_addr
.in_addr
));
2476 WARN_ON(con
->state
!= CON_STATE_CONNECTING
&&
2477 con
->state
!= CON_STATE_NEGOTIATING
&&
2478 con
->state
!= CON_STATE_OPEN
);
2480 con_close_socket(con
);
2482 if (con_flag_test(con
, CON_FLAG_LOSSYTX
)) {
2483 dout("fault on LOSSYTX channel, marking CLOSED\n");
2484 con
->state
= CON_STATE_CLOSED
;
2489 BUG_ON(con
->in_msg
->con
!= con
);
2490 con
->in_msg
->con
= NULL
;
2491 ceph_msg_put(con
->in_msg
);
2496 /* Requeue anything that hasn't been acked */
2497 list_splice_init(&con
->out_sent
, &con
->out_queue
);
2499 /* If there are no messages queued or keepalive pending, place
2500 * the connection in a STANDBY state */
2501 if (list_empty(&con
->out_queue
) &&
2502 !con_flag_test(con
, CON_FLAG_KEEPALIVE_PENDING
)) {
2503 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con
);
2504 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
2505 con
->state
= CON_STATE_STANDBY
;
2507 /* retry after a delay. */
2508 con
->state
= CON_STATE_PREOPEN
;
2509 if (con
->delay
== 0)
2510 con
->delay
= BASE_DELAY_INTERVAL
;
2511 else if (con
->delay
< MAX_DELAY_INTERVAL
)
2513 con_flag_set(con
, CON_FLAG_BACKOFF
);
2521 * initialize a new messenger instance
2523 void ceph_messenger_init(struct ceph_messenger
*msgr
,
2524 struct ceph_entity_addr
*myaddr
,
2525 u32 supported_features
,
2526 u32 required_features
,
2529 msgr
->supported_features
= supported_features
;
2530 msgr
->required_features
= required_features
;
2532 spin_lock_init(&msgr
->global_seq_lock
);
2535 msgr
->inst
.addr
= *myaddr
;
2537 /* select a random nonce */
2538 msgr
->inst
.addr
.type
= 0;
2539 get_random_bytes(&msgr
->inst
.addr
.nonce
, sizeof(msgr
->inst
.addr
.nonce
));
2540 encode_my_addr(msgr
);
2541 msgr
->nocrc
= nocrc
;
2543 atomic_set(&msgr
->stopping
, 0);
2545 dout("%s %p\n", __func__
, msgr
);
2547 EXPORT_SYMBOL(ceph_messenger_init
);
2549 static void clear_standby(struct ceph_connection
*con
)
2551 /* come back from STANDBY? */
2552 if (con
->state
== CON_STATE_STANDBY
) {
2553 dout("clear_standby %p and ++connect_seq\n", con
);
2554 con
->state
= CON_STATE_PREOPEN
;
2556 WARN_ON(con_flag_test(con
, CON_FLAG_WRITE_PENDING
));
2557 WARN_ON(con_flag_test(con
, CON_FLAG_KEEPALIVE_PENDING
));
2562 * Queue up an outgoing message on the given connection.
2564 void ceph_con_send(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2567 msg
->hdr
.src
= con
->msgr
->inst
.name
;
2568 BUG_ON(msg
->front
.iov_len
!= le32_to_cpu(msg
->hdr
.front_len
));
2569 msg
->needs_out_seq
= true;
2571 mutex_lock(&con
->mutex
);
2573 if (con
->state
== CON_STATE_CLOSED
) {
2574 dout("con_send %p closed, dropping %p\n", con
, msg
);
2576 mutex_unlock(&con
->mutex
);
2580 BUG_ON(msg
->con
!= NULL
);
2581 msg
->con
= con
->ops
->get(con
);
2582 BUG_ON(msg
->con
== NULL
);
2584 BUG_ON(!list_empty(&msg
->list_head
));
2585 list_add_tail(&msg
->list_head
, &con
->out_queue
);
2586 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg
,
2587 ENTITY_NAME(con
->peer_name
), le16_to_cpu(msg
->hdr
.type
),
2588 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2589 le32_to_cpu(msg
->hdr
.front_len
),
2590 le32_to_cpu(msg
->hdr
.middle_len
),
2591 le32_to_cpu(msg
->hdr
.data_len
));
2594 mutex_unlock(&con
->mutex
);
2596 /* if there wasn't anything waiting to send before, queue
2598 if (con_flag_test_and_set(con
, CON_FLAG_WRITE_PENDING
) == 0)
2601 EXPORT_SYMBOL(ceph_con_send
);
2604 * Revoke a message that was previously queued for send
2606 void ceph_msg_revoke(struct ceph_msg
*msg
)
2608 struct ceph_connection
*con
= msg
->con
;
2611 return; /* Message not in our possession */
2613 mutex_lock(&con
->mutex
);
2614 if (!list_empty(&msg
->list_head
)) {
2615 dout("%s %p msg %p - was on queue\n", __func__
, con
, msg
);
2616 list_del_init(&msg
->list_head
);
2617 BUG_ON(msg
->con
== NULL
);
2618 msg
->con
->ops
->put(msg
->con
);
2624 if (con
->out_msg
== msg
) {
2625 dout("%s %p msg %p - was sending\n", __func__
, con
, msg
);
2626 con
->out_msg
= NULL
;
2627 if (con
->out_kvec_is_msg
) {
2628 con
->out_skip
= con
->out_kvec_bytes
;
2629 con
->out_kvec_is_msg
= false;
2635 mutex_unlock(&con
->mutex
);
2639 * Revoke a message that we may be reading data into
2641 void ceph_msg_revoke_incoming(struct ceph_msg
*msg
)
2643 struct ceph_connection
*con
;
2645 BUG_ON(msg
== NULL
);
2647 dout("%s msg %p null con\n", __func__
, msg
);
2649 return; /* Message not in our possession */
2653 mutex_lock(&con
->mutex
);
2654 if (con
->in_msg
== msg
) {
2655 unsigned int front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
2656 unsigned int middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
2657 unsigned int data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
2659 /* skip rest of message */
2660 dout("%s %p msg %p revoked\n", __func__
, con
, msg
);
2661 con
->in_base_pos
= con
->in_base_pos
-
2662 sizeof(struct ceph_msg_header
) -
2666 sizeof(struct ceph_msg_footer
);
2667 ceph_msg_put(con
->in_msg
);
2669 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2672 dout("%s %p in_msg %p msg %p no-op\n",
2673 __func__
, con
, con
->in_msg
, msg
);
2675 mutex_unlock(&con
->mutex
);
2679 * Queue a keepalive byte to ensure the tcp connection is alive.
2681 void ceph_con_keepalive(struct ceph_connection
*con
)
2683 dout("con_keepalive %p\n", con
);
2684 mutex_lock(&con
->mutex
);
2686 mutex_unlock(&con
->mutex
);
2687 if (con_flag_test_and_set(con
, CON_FLAG_KEEPALIVE_PENDING
) == 0 &&
2688 con_flag_test_and_set(con
, CON_FLAG_WRITE_PENDING
) == 0)
2691 EXPORT_SYMBOL(ceph_con_keepalive
);
2693 void ceph_msg_data_set_pages(struct ceph_msg
*msg
, struct page
**pages
,
2694 size_t length
, size_t alignment
)
2699 BUG_ON(msg
->length
);
2702 msg
->length
= length
;
2703 msg
->page_alignment
= alignment
& ~PAGE_MASK
;
2705 EXPORT_SYMBOL(ceph_msg_data_set_pages
);
2707 void ceph_msg_data_set_pagelist(struct ceph_msg
*msg
,
2708 struct ceph_pagelist
*pagelist
)
2711 BUG_ON(!pagelist
->length
);
2712 BUG_ON(msg
->pagelist
);
2714 msg
->pagelist
= pagelist
;
2716 EXPORT_SYMBOL(ceph_msg_data_set_pagelist
);
2718 void ceph_msg_data_set_bio(struct ceph_msg
*msg
, struct bio
*bio
)
2725 EXPORT_SYMBOL(ceph_msg_data_set_bio
);
2727 void ceph_msg_data_set_trail(struct ceph_msg
*msg
, struct ceph_pagelist
*trail
)
2730 BUG_ON(!trail
->length
);
2735 EXPORT_SYMBOL(ceph_msg_data_set_trail
);
2738 * construct a new message with given type, size
2739 * the new msg has a ref count of 1.
2741 struct ceph_msg
*ceph_msg_new(int type
, int front_len
, gfp_t flags
,
2746 m
= kzalloc(sizeof(*m
), flags
);
2750 m
->hdr
.type
= cpu_to_le16(type
);
2751 m
->hdr
.priority
= cpu_to_le16(CEPH_MSG_PRIO_DEFAULT
);
2752 m
->hdr
.front_len
= cpu_to_le32(front_len
);
2754 INIT_LIST_HEAD(&m
->list_head
);
2755 kref_init(&m
->kref
);
2758 m
->front_max
= front_len
;
2760 if (front_len
> PAGE_CACHE_SIZE
) {
2761 m
->front
.iov_base
= __vmalloc(front_len
, flags
,
2763 m
->front_is_vmalloc
= true;
2765 m
->front
.iov_base
= kmalloc(front_len
, flags
);
2767 if (m
->front
.iov_base
== NULL
) {
2768 dout("ceph_msg_new can't allocate %d bytes\n",
2773 m
->front
.iov_base
= NULL
;
2775 m
->front
.iov_len
= front_len
;
2777 dout("ceph_msg_new %p front %d\n", m
, front_len
);
2784 pr_err("msg_new can't create type %d front %d\n", type
,
2788 dout("msg_new can't create type %d front %d\n", type
,
2793 EXPORT_SYMBOL(ceph_msg_new
);
2796 * Allocate "middle" portion of a message, if it is needed and wasn't
2797 * allocated by alloc_msg. This allows us to read a small fixed-size
2798 * per-type header in the front and then gracefully fail (i.e.,
2799 * propagate the error to the caller based on info in the front) when
2800 * the middle is too large.
2802 static int ceph_alloc_middle(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2804 int type
= le16_to_cpu(msg
->hdr
.type
);
2805 int middle_len
= le32_to_cpu(msg
->hdr
.middle_len
);
2807 dout("alloc_middle %p type %d %s middle_len %d\n", msg
, type
,
2808 ceph_msg_type_name(type
), middle_len
);
2809 BUG_ON(!middle_len
);
2810 BUG_ON(msg
->middle
);
2812 msg
->middle
= ceph_buffer_new(middle_len
, GFP_NOFS
);
2819 * Allocate a message for receiving an incoming message on a
2820 * connection, and save the result in con->in_msg. Uses the
2821 * connection's private alloc_msg op if available.
2823 * Returns 0 on success, or a negative error code.
2825 * On success, if we set *skip = 1:
2826 * - the next message should be skipped and ignored.
2827 * - con->in_msg == NULL
2828 * or if we set *skip = 0:
2829 * - con->in_msg is non-null.
2830 * On error (ENOMEM, EAGAIN, ...),
2831 * - con->in_msg == NULL
2833 static int ceph_con_in_msg_alloc(struct ceph_connection
*con
, int *skip
)
2835 struct ceph_msg_header
*hdr
= &con
->in_hdr
;
2836 int middle_len
= le32_to_cpu(hdr
->middle_len
);
2837 struct ceph_msg
*msg
;
2840 BUG_ON(con
->in_msg
!= NULL
);
2841 BUG_ON(!con
->ops
->alloc_msg
);
2843 mutex_unlock(&con
->mutex
);
2844 msg
= con
->ops
->alloc_msg(con
, hdr
, skip
);
2845 mutex_lock(&con
->mutex
);
2846 if (con
->state
!= CON_STATE_OPEN
) {
2854 con
->in_msg
->con
= con
->ops
->get(con
);
2855 BUG_ON(con
->in_msg
->con
== NULL
);
2858 * Null message pointer means either we should skip
2859 * this message or we couldn't allocate memory. The
2860 * former is not an error.
2864 con
->error_msg
= "error allocating memory for incoming message";
2868 memcpy(&con
->in_msg
->hdr
, &con
->in_hdr
, sizeof(con
->in_hdr
));
2870 if (middle_len
&& !con
->in_msg
->middle
) {
2871 ret
= ceph_alloc_middle(con
, con
->in_msg
);
2873 ceph_msg_put(con
->in_msg
);
2883 * Free a generically kmalloc'd message.
2885 void ceph_msg_kfree(struct ceph_msg
*m
)
2887 dout("msg_kfree %p\n", m
);
2888 if (m
->front_is_vmalloc
)
2889 vfree(m
->front
.iov_base
);
2891 kfree(m
->front
.iov_base
);
2896 * Drop a msg ref. Destroy as needed.
2898 void ceph_msg_last_put(struct kref
*kref
)
2900 struct ceph_msg
*m
= container_of(kref
, struct ceph_msg
, kref
);
2902 dout("ceph_msg_put last one on %p\n", m
);
2903 WARN_ON(!list_empty(&m
->list_head
));
2905 /* drop middle, data, if any */
2907 ceph_buffer_put(m
->middle
);
2914 ceph_pagelist_release(m
->pagelist
);
2922 ceph_msgpool_put(m
->pool
, m
);
2926 EXPORT_SYMBOL(ceph_msg_last_put
);
2928 void ceph_msg_dump(struct ceph_msg
*msg
)
2930 pr_debug("msg_dump %p (front_max %d length %zd)\n", msg
,
2931 msg
->front_max
, msg
->length
);
2932 print_hex_dump(KERN_DEBUG
, "header: ",
2933 DUMP_PREFIX_OFFSET
, 16, 1,
2934 &msg
->hdr
, sizeof(msg
->hdr
), true);
2935 print_hex_dump(KERN_DEBUG
, " front: ",
2936 DUMP_PREFIX_OFFSET
, 16, 1,
2937 msg
->front
.iov_base
, msg
->front
.iov_len
, true);
2939 print_hex_dump(KERN_DEBUG
, "middle: ",
2940 DUMP_PREFIX_OFFSET
, 16, 1,
2941 msg
->middle
->vec
.iov_base
,
2942 msg
->middle
->vec
.iov_len
, true);
2943 print_hex_dump(KERN_DEBUG
, "footer: ",
2944 DUMP_PREFIX_OFFSET
, 16, 1,
2945 &msg
->footer
, sizeof(msg
->footer
), true);
2947 EXPORT_SYMBOL(ceph_msg_dump
);