1 #include <linux/ceph/ceph_debug.h>
3 #include <linux/crc32c.h>
4 #include <linux/ctype.h>
5 #include <linux/highmem.h>
6 #include <linux/inet.h>
7 #include <linux/kthread.h>
9 #include <linux/slab.h>
10 #include <linux/socket.h>
11 #include <linux/string.h>
12 #include <linux/bio.h>
13 #include <linux/blkdev.h>
14 #include <linux/dns_resolver.h>
17 #include <linux/ceph/libceph.h>
18 #include <linux/ceph/messenger.h>
19 #include <linux/ceph/decode.h>
20 #include <linux/ceph/pagelist.h>
21 #include <linux/export.h>
24 * Ceph uses the messenger to exchange ceph_msg messages with other
25 * hosts in the system. The messenger provides ordered and reliable
26 * delivery. We tolerate TCP disconnects by reconnecting (with
27 * exponential backoff) in the case of a fault (disconnection, bad
28 * crc, protocol error). Acks allow sent messages to be discarded by
32 /* static tag bytes (protocol control messages) */
33 static char tag_msg
= CEPH_MSGR_TAG_MSG
;
34 static char tag_ack
= CEPH_MSGR_TAG_ACK
;
35 static char tag_keepalive
= CEPH_MSGR_TAG_KEEPALIVE
;
38 static struct lock_class_key socket_class
;
42 * When skipping (ignoring) a block of input we read it into a "skip
43 * buffer," which is this many bytes in size.
45 #define SKIP_BUF_SIZE 1024
47 static void queue_con(struct ceph_connection
*con
);
48 static void con_work(struct work_struct
*);
49 static void ceph_fault(struct ceph_connection
*con
);
52 * Nicely render a sockaddr as a string. An array of formatted
53 * strings is used, to approximate reentrancy.
55 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
56 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
57 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
58 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
60 static char addr_str
[ADDR_STR_COUNT
][MAX_ADDR_STR_LEN
];
61 static atomic_t addr_str_seq
= ATOMIC_INIT(0);
63 static struct page
*zero_page
; /* used in certain error cases */
64 static void *zero_page_address
; /* kernel virtual addr of zero_page */
66 const char *ceph_pr_addr(const struct sockaddr_storage
*ss
)
70 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
71 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
73 i
= atomic_inc_return(&addr_str_seq
) & ADDR_STR_COUNT_MASK
;
76 switch (ss
->ss_family
) {
78 snprintf(s
, MAX_ADDR_STR_LEN
, "%pI4:%hu", &in4
->sin_addr
,
79 ntohs(in4
->sin_port
));
83 snprintf(s
, MAX_ADDR_STR_LEN
, "[%pI6c]:%hu", &in6
->sin6_addr
,
84 ntohs(in6
->sin6_port
));
88 snprintf(s
, MAX_ADDR_STR_LEN
, "(unknown sockaddr family %hu)",
94 EXPORT_SYMBOL(ceph_pr_addr
);
96 static void encode_my_addr(struct ceph_messenger
*msgr
)
98 memcpy(&msgr
->my_enc_addr
, &msgr
->inst
.addr
, sizeof(msgr
->my_enc_addr
));
99 ceph_encode_addr(&msgr
->my_enc_addr
);
103 * work queue for all reading and writing to/from the socket.
105 static struct workqueue_struct
*ceph_msgr_wq
;
107 void _ceph_msgr_exit(void)
110 destroy_workqueue(ceph_msgr_wq
);
114 BUG_ON(zero_page_address
== NULL
);
115 zero_page_address
= NULL
;
117 BUG_ON(zero_page
== NULL
);
119 page_cache_release(zero_page
);
123 int ceph_msgr_init(void)
125 BUG_ON(zero_page
!= NULL
);
126 zero_page
= ZERO_PAGE(0);
127 page_cache_get(zero_page
);
129 BUG_ON(zero_page_address
!= NULL
);
130 zero_page_address
= kmap(zero_page
);
132 ceph_msgr_wq
= alloc_workqueue("ceph-msgr", WQ_NON_REENTRANT
, 0);
136 pr_err("msgr_init failed to create workqueue\n");
141 EXPORT_SYMBOL(ceph_msgr_init
);
143 void ceph_msgr_exit(void)
145 BUG_ON(ceph_msgr_wq
== NULL
);
149 EXPORT_SYMBOL(ceph_msgr_exit
);
151 void ceph_msgr_flush(void)
153 flush_workqueue(ceph_msgr_wq
);
155 EXPORT_SYMBOL(ceph_msgr_flush
);
159 * socket callback functions
162 /* data available on socket, or listen socket received a connect */
163 static void ceph_data_ready(struct sock
*sk
, int count_unused
)
165 struct ceph_connection
*con
= sk
->sk_user_data
;
167 if (sk
->sk_state
!= TCP_CLOSE_WAIT
) {
168 dout("ceph_data_ready on %p state = %lu, queueing work\n",
174 /* socket has buffer space for writing */
175 static void ceph_write_space(struct sock
*sk
)
177 struct ceph_connection
*con
= sk
->sk_user_data
;
179 /* only queue to workqueue if there is data we want to write,
180 * and there is sufficient space in the socket buffer to accept
181 * more data. clear SOCK_NOSPACE so that ceph_write_space()
182 * doesn't get called again until try_write() fills the socket
183 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
184 * and net/core/stream.c:sk_stream_write_space().
186 if (test_bit(WRITE_PENDING
, &con
->state
)) {
187 if (sk_stream_wspace(sk
) >= sk_stream_min_wspace(sk
)) {
188 dout("ceph_write_space %p queueing write work\n", con
);
189 clear_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
193 dout("ceph_write_space %p nothing to write\n", con
);
197 /* socket's state has changed */
198 static void ceph_state_change(struct sock
*sk
)
200 struct ceph_connection
*con
= sk
->sk_user_data
;
202 dout("ceph_state_change %p state = %lu sk_state = %u\n",
203 con
, con
->state
, sk
->sk_state
);
205 if (test_bit(CLOSED
, &con
->state
))
208 switch (sk
->sk_state
) {
210 dout("ceph_state_change TCP_CLOSE\n");
212 dout("ceph_state_change TCP_CLOSE_WAIT\n");
213 if (test_and_set_bit(SOCK_CLOSED
, &con
->state
) == 0) {
214 if (test_bit(CONNECTING
, &con
->state
))
215 con
->error_msg
= "connection failed";
217 con
->error_msg
= "socket closed";
221 case TCP_ESTABLISHED
:
222 dout("ceph_state_change TCP_ESTABLISHED\n");
225 default: /* Everything else is uninteresting */
231 * set up socket callbacks
233 static void set_sock_callbacks(struct socket
*sock
,
234 struct ceph_connection
*con
)
236 struct sock
*sk
= sock
->sk
;
237 sk
->sk_user_data
= con
;
238 sk
->sk_data_ready
= ceph_data_ready
;
239 sk
->sk_write_space
= ceph_write_space
;
240 sk
->sk_state_change
= ceph_state_change
;
249 * initiate connection to a remote socket.
251 static int ceph_tcp_connect(struct ceph_connection
*con
)
253 struct sockaddr_storage
*paddr
= &con
->peer_addr
.in_addr
;
258 ret
= sock_create_kern(con
->peer_addr
.in_addr
.ss_family
, SOCK_STREAM
,
262 sock
->sk
->sk_allocation
= GFP_NOFS
;
264 #ifdef CONFIG_LOCKDEP
265 lockdep_set_class(&sock
->sk
->sk_lock
, &socket_class
);
268 set_sock_callbacks(sock
, con
);
270 dout("connect %s\n", ceph_pr_addr(&con
->peer_addr
.in_addr
));
272 ret
= sock
->ops
->connect(sock
, (struct sockaddr
*)paddr
, sizeof(*paddr
),
274 if (ret
== -EINPROGRESS
) {
275 dout("connect %s EINPROGRESS sk_state = %u\n",
276 ceph_pr_addr(&con
->peer_addr
.in_addr
),
278 } else if (ret
< 0) {
279 pr_err("connect %s error %d\n",
280 ceph_pr_addr(&con
->peer_addr
.in_addr
), ret
);
282 con
->error_msg
= "connect error";
291 static int ceph_tcp_recvmsg(struct socket
*sock
, void *buf
, size_t len
)
293 struct kvec iov
= {buf
, len
};
294 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
297 r
= kernel_recvmsg(sock
, &msg
, &iov
, 1, len
, msg
.msg_flags
);
304 * write something. @more is true if caller will be sending more data
307 static int ceph_tcp_sendmsg(struct socket
*sock
, struct kvec
*iov
,
308 size_t kvlen
, size_t len
, int more
)
310 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
314 msg
.msg_flags
|= MSG_MORE
;
316 msg
.msg_flags
|= MSG_EOR
; /* superfluous, but what the hell */
318 r
= kernel_sendmsg(sock
, &msg
, iov
, kvlen
, len
);
326 * Shutdown/close the socket for the given connection.
328 static int con_close_socket(struct ceph_connection
*con
)
332 dout("con_close_socket on %p sock %p\n", con
, con
->sock
);
335 set_bit(SOCK_CLOSED
, &con
->state
);
336 rc
= con
->sock
->ops
->shutdown(con
->sock
, SHUT_RDWR
);
337 sock_release(con
->sock
);
339 clear_bit(SOCK_CLOSED
, &con
->state
);
344 * Reset a connection. Discard all incoming and outgoing messages
345 * and clear *_seq state.
347 static void ceph_msg_remove(struct ceph_msg
*msg
)
349 list_del_init(&msg
->list_head
);
352 static void ceph_msg_remove_list(struct list_head
*head
)
354 while (!list_empty(head
)) {
355 struct ceph_msg
*msg
= list_first_entry(head
, struct ceph_msg
,
357 ceph_msg_remove(msg
);
361 static void reset_connection(struct ceph_connection
*con
)
363 /* reset connection, out_queue, msg_ and connect_seq */
364 /* discard existing out_queue and msg_seq */
365 ceph_msg_remove_list(&con
->out_queue
);
366 ceph_msg_remove_list(&con
->out_sent
);
369 ceph_msg_put(con
->in_msg
);
373 con
->connect_seq
= 0;
376 ceph_msg_put(con
->out_msg
);
380 con
->in_seq_acked
= 0;
384 * mark a peer down. drop any open connections.
386 void ceph_con_close(struct ceph_connection
*con
)
388 dout("con_close %p peer %s\n", con
,
389 ceph_pr_addr(&con
->peer_addr
.in_addr
));
390 set_bit(CLOSED
, &con
->state
); /* in case there's queued work */
391 clear_bit(STANDBY
, &con
->state
); /* avoid connect_seq bump */
392 clear_bit(LOSSYTX
, &con
->state
); /* so we retry next connect */
393 clear_bit(KEEPALIVE_PENDING
, &con
->state
);
394 clear_bit(WRITE_PENDING
, &con
->state
);
395 mutex_lock(&con
->mutex
);
396 reset_connection(con
);
397 con
->peer_global_seq
= 0;
398 cancel_delayed_work(&con
->work
);
399 mutex_unlock(&con
->mutex
);
402 EXPORT_SYMBOL(ceph_con_close
);
405 * Reopen a closed connection, with a new peer address.
407 void ceph_con_open(struct ceph_connection
*con
, struct ceph_entity_addr
*addr
)
409 dout("con_open %p %s\n", con
, ceph_pr_addr(&addr
->in_addr
));
410 set_bit(OPENING
, &con
->state
);
411 clear_bit(CLOSED
, &con
->state
);
412 memcpy(&con
->peer_addr
, addr
, sizeof(*addr
));
413 con
->delay
= 0; /* reset backoff memory */
416 EXPORT_SYMBOL(ceph_con_open
);
419 * return true if this connection ever successfully opened
421 bool ceph_con_opened(struct ceph_connection
*con
)
423 return con
->connect_seq
> 0;
429 struct ceph_connection
*ceph_con_get(struct ceph_connection
*con
)
431 int nref
= __atomic_add_unless(&con
->nref
, 1, 0);
433 dout("con_get %p nref = %d -> %d\n", con
, nref
, nref
+ 1);
435 return nref
? con
: NULL
;
438 void ceph_con_put(struct ceph_connection
*con
)
440 int nref
= atomic_dec_return(&con
->nref
);
447 dout("con_put %p nref = %d -> %d\n", con
, nref
+ 1, nref
);
451 * initialize a new connection.
453 void ceph_con_init(struct ceph_messenger
*msgr
, struct ceph_connection
*con
)
455 dout("con_init %p\n", con
);
456 memset(con
, 0, sizeof(*con
));
457 atomic_set(&con
->nref
, 1);
459 mutex_init(&con
->mutex
);
460 INIT_LIST_HEAD(&con
->out_queue
);
461 INIT_LIST_HEAD(&con
->out_sent
);
462 INIT_DELAYED_WORK(&con
->work
, con_work
);
464 EXPORT_SYMBOL(ceph_con_init
);
468 * We maintain a global counter to order connection attempts. Get
469 * a unique seq greater than @gt.
471 static u32
get_global_seq(struct ceph_messenger
*msgr
, u32 gt
)
475 spin_lock(&msgr
->global_seq_lock
);
476 if (msgr
->global_seq
< gt
)
477 msgr
->global_seq
= gt
;
478 ret
= ++msgr
->global_seq
;
479 spin_unlock(&msgr
->global_seq_lock
);
483 static void ceph_con_out_kvec_reset(struct ceph_connection
*con
)
485 con
->out_kvec_left
= 0;
486 con
->out_kvec_bytes
= 0;
487 con
->out_kvec_cur
= &con
->out_kvec
[0];
490 static void ceph_con_out_kvec_add(struct ceph_connection
*con
,
491 size_t size
, void *data
)
495 index
= con
->out_kvec_left
;
496 BUG_ON(index
>= ARRAY_SIZE(con
->out_kvec
));
498 con
->out_kvec
[index
].iov_len
= size
;
499 con
->out_kvec
[index
].iov_base
= data
;
500 con
->out_kvec_left
++;
501 con
->out_kvec_bytes
+= size
;
505 * Prepare footer for currently outgoing message, and finish things
506 * off. Assumes out_kvec* are already valid.. we just add on to the end.
508 static void prepare_write_message_footer(struct ceph_connection
*con
)
510 struct ceph_msg
*m
= con
->out_msg
;
511 int v
= con
->out_kvec_left
;
513 dout("prepare_write_message_footer %p\n", con
);
514 con
->out_kvec_is_msg
= true;
515 con
->out_kvec
[v
].iov_base
= &m
->footer
;
516 con
->out_kvec
[v
].iov_len
= sizeof(m
->footer
);
517 con
->out_kvec_bytes
+= sizeof(m
->footer
);
518 con
->out_kvec_left
++;
519 con
->out_more
= m
->more_to_follow
;
520 con
->out_msg_done
= true;
524 * Prepare headers for the next outgoing message.
526 static void prepare_write_message(struct ceph_connection
*con
)
531 ceph_con_out_kvec_reset(con
);
532 con
->out_kvec_is_msg
= true;
533 con
->out_msg_done
= false;
535 /* Sneak an ack in there first? If we can get it into the same
536 * TCP packet that's a good thing. */
537 if (con
->in_seq
> con
->in_seq_acked
) {
538 con
->in_seq_acked
= con
->in_seq
;
539 ceph_con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
540 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
541 ceph_con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
545 m
= list_first_entry(&con
->out_queue
, struct ceph_msg
, list_head
);
548 /* put message on sent list */
550 list_move_tail(&m
->list_head
, &con
->out_sent
);
553 * only assign outgoing seq # if we haven't sent this message
554 * yet. if it is requeued, resend with it's original seq.
556 if (m
->needs_out_seq
) {
557 m
->hdr
.seq
= cpu_to_le64(++con
->out_seq
);
558 m
->needs_out_seq
= false;
561 dout("prepare_write_message %p seq %lld type %d len %d+%d+%d %d pgs\n",
562 m
, con
->out_seq
, le16_to_cpu(m
->hdr
.type
),
563 le32_to_cpu(m
->hdr
.front_len
), le32_to_cpu(m
->hdr
.middle_len
),
564 le32_to_cpu(m
->hdr
.data_len
),
566 BUG_ON(le32_to_cpu(m
->hdr
.front_len
) != m
->front
.iov_len
);
568 /* tag + hdr + front + middle */
569 ceph_con_out_kvec_add(con
, sizeof (tag_msg
), &tag_msg
);
570 ceph_con_out_kvec_add(con
, sizeof (m
->hdr
), &m
->hdr
);
571 ceph_con_out_kvec_add(con
, m
->front
.iov_len
, m
->front
.iov_base
);
574 ceph_con_out_kvec_add(con
, m
->middle
->vec
.iov_len
,
575 m
->middle
->vec
.iov_base
);
577 /* fill in crc (except data pages), footer */
578 crc
= crc32c(0, &m
->hdr
, offsetof(struct ceph_msg_header
, crc
));
579 con
->out_msg
->hdr
.crc
= cpu_to_le32(crc
);
580 con
->out_msg
->footer
.flags
= CEPH_MSG_FOOTER_COMPLETE
;
582 crc
= crc32c(0, m
->front
.iov_base
, m
->front
.iov_len
);
583 con
->out_msg
->footer
.front_crc
= cpu_to_le32(crc
);
585 crc
= crc32c(0, m
->middle
->vec
.iov_base
,
586 m
->middle
->vec
.iov_len
);
587 con
->out_msg
->footer
.middle_crc
= cpu_to_le32(crc
);
589 con
->out_msg
->footer
.middle_crc
= 0;
590 con
->out_msg
->footer
.data_crc
= 0;
591 dout("prepare_write_message front_crc %u data_crc %u\n",
592 le32_to_cpu(con
->out_msg
->footer
.front_crc
),
593 le32_to_cpu(con
->out_msg
->footer
.middle_crc
));
595 /* is there a data payload? */
596 if (le32_to_cpu(m
->hdr
.data_len
) > 0) {
597 /* initialize page iterator */
598 con
->out_msg_pos
.page
= 0;
600 con
->out_msg_pos
.page_pos
= m
->page_alignment
;
602 con
->out_msg_pos
.page_pos
= 0;
603 con
->out_msg_pos
.data_pos
= 0;
604 con
->out_msg_pos
.did_page_crc
= false;
605 con
->out_more
= 1; /* data + footer will follow */
607 /* no, queue up footer too and be done */
608 prepare_write_message_footer(con
);
611 set_bit(WRITE_PENDING
, &con
->state
);
617 static void prepare_write_ack(struct ceph_connection
*con
)
619 dout("prepare_write_ack %p %llu -> %llu\n", con
,
620 con
->in_seq_acked
, con
->in_seq
);
621 con
->in_seq_acked
= con
->in_seq
;
623 ceph_con_out_kvec_reset(con
);
625 ceph_con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
627 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
628 ceph_con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
631 con
->out_more
= 1; /* more will follow.. eventually.. */
632 set_bit(WRITE_PENDING
, &con
->state
);
636 * Prepare to write keepalive byte.
638 static void prepare_write_keepalive(struct ceph_connection
*con
)
640 dout("prepare_write_keepalive %p\n", con
);
641 ceph_con_out_kvec_reset(con
);
642 ceph_con_out_kvec_add(con
, sizeof (tag_keepalive
), &tag_keepalive
);
643 set_bit(WRITE_PENDING
, &con
->state
);
647 * Connection negotiation.
650 static int prepare_connect_authorizer(struct ceph_connection
*con
)
654 int auth_protocol
= 0;
656 mutex_unlock(&con
->mutex
);
657 if (con
->ops
->get_authorizer
)
658 con
->ops
->get_authorizer(con
, &auth_buf
, &auth_len
,
659 &auth_protocol
, &con
->auth_reply_buf
,
660 &con
->auth_reply_buf_len
,
662 mutex_lock(&con
->mutex
);
664 if (test_bit(CLOSED
, &con
->state
) ||
665 test_bit(OPENING
, &con
->state
))
668 con
->out_connect
.authorizer_protocol
= cpu_to_le32(auth_protocol
);
669 con
->out_connect
.authorizer_len
= cpu_to_le32(auth_len
);
672 ceph_con_out_kvec_add(con
, auth_len
, auth_buf
);
678 * We connected to a peer and are saying hello.
680 static void prepare_write_banner(struct ceph_messenger
*msgr
,
681 struct ceph_connection
*con
)
683 ceph_con_out_kvec_reset(con
);
684 ceph_con_out_kvec_add(con
, strlen(CEPH_BANNER
), CEPH_BANNER
);
685 ceph_con_out_kvec_add(con
, sizeof (msgr
->my_enc_addr
),
689 set_bit(WRITE_PENDING
, &con
->state
);
692 static int prepare_write_connect(struct ceph_messenger
*msgr
,
693 struct ceph_connection
*con
,
696 unsigned global_seq
= get_global_seq(con
->msgr
, 0);
699 switch (con
->peer_name
.type
) {
700 case CEPH_ENTITY_TYPE_MON
:
701 proto
= CEPH_MONC_PROTOCOL
;
703 case CEPH_ENTITY_TYPE_OSD
:
704 proto
= CEPH_OSDC_PROTOCOL
;
706 case CEPH_ENTITY_TYPE_MDS
:
707 proto
= CEPH_MDSC_PROTOCOL
;
713 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con
,
714 con
->connect_seq
, global_seq
, proto
);
716 con
->out_connect
.features
= cpu_to_le64(msgr
->supported_features
);
717 con
->out_connect
.host_type
= cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT
);
718 con
->out_connect
.connect_seq
= cpu_to_le32(con
->connect_seq
);
719 con
->out_connect
.global_seq
= cpu_to_le32(global_seq
);
720 con
->out_connect
.protocol_version
= cpu_to_le32(proto
);
721 con
->out_connect
.flags
= 0;
724 prepare_write_banner(msgr
, con
);
726 ceph_con_out_kvec_reset(con
);
727 ceph_con_out_kvec_add(con
, sizeof (con
->out_connect
), &con
->out_connect
);
730 set_bit(WRITE_PENDING
, &con
->state
);
732 return prepare_connect_authorizer(con
);
736 * write as much of pending kvecs to the socket as we can.
738 * 0 -> socket full, but more to do
741 static int write_partial_kvec(struct ceph_connection
*con
)
745 dout("write_partial_kvec %p %d left\n", con
, con
->out_kvec_bytes
);
746 while (con
->out_kvec_bytes
> 0) {
747 ret
= ceph_tcp_sendmsg(con
->sock
, con
->out_kvec_cur
,
748 con
->out_kvec_left
, con
->out_kvec_bytes
,
752 con
->out_kvec_bytes
-= ret
;
753 if (con
->out_kvec_bytes
== 0)
756 /* account for full iov entries consumed */
757 while (ret
>= con
->out_kvec_cur
->iov_len
) {
758 BUG_ON(!con
->out_kvec_left
);
759 ret
-= con
->out_kvec_cur
->iov_len
;
761 con
->out_kvec_left
--;
763 /* and for a partially-consumed entry */
765 con
->out_kvec_cur
->iov_len
-= ret
;
766 con
->out_kvec_cur
->iov_base
+= ret
;
769 con
->out_kvec_left
= 0;
770 con
->out_kvec_is_msg
= false;
773 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con
,
774 con
->out_kvec_bytes
, con
->out_kvec_left
, ret
);
775 return ret
; /* done! */
779 static void init_bio_iter(struct bio
*bio
, struct bio
**iter
, int *seg
)
790 static void iter_bio_next(struct bio
**bio_iter
, int *seg
)
792 if (*bio_iter
== NULL
)
795 BUG_ON(*seg
>= (*bio_iter
)->bi_vcnt
);
798 if (*seg
== (*bio_iter
)->bi_vcnt
)
799 init_bio_iter((*bio_iter
)->bi_next
, bio_iter
, seg
);
804 * Write as much message data payload as we can. If we finish, queue
806 * 1 -> done, footer is now queued in out_kvec[].
807 * 0 -> socket full, but more to do
810 static int write_partial_msg_pages(struct ceph_connection
*con
)
812 struct ceph_msg
*msg
= con
->out_msg
;
813 unsigned data_len
= le32_to_cpu(msg
->hdr
.data_len
);
815 bool do_crc
= con
->msgr
->nocrc
;
819 size_t trail_len
= (msg
->trail
? msg
->trail
->length
: 0);
821 dout("write_partial_msg_pages %p msg %p page %d/%d offset %d\n",
822 con
, con
->out_msg
, con
->out_msg_pos
.page
, con
->out_msg
->nr_pages
,
823 con
->out_msg_pos
.page_pos
);
826 if (msg
->bio
&& !msg
->bio_iter
)
827 init_bio_iter(msg
->bio
, &msg
->bio_iter
, &msg
->bio_seg
);
830 while (data_len
> con
->out_msg_pos
.data_pos
) {
831 struct page
*page
= NULL
;
833 int max_write
= PAGE_SIZE
;
836 total_max_write
= data_len
- trail_len
-
837 con
->out_msg_pos
.data_pos
;
840 * if we are calculating the data crc (the default), we need
841 * to map the page. if our pages[] has been revoked, use the
845 /* have we reached the trail part of the data? */
846 if (con
->out_msg_pos
.data_pos
>= data_len
- trail_len
) {
849 total_max_write
= data_len
- con
->out_msg_pos
.data_pos
;
851 page
= list_first_entry(&msg
->trail
->head
,
855 max_write
= PAGE_SIZE
;
856 } else if (msg
->pages
) {
857 page
= msg
->pages
[con
->out_msg_pos
.page
];
860 } else if (msg
->pagelist
) {
861 page
= list_first_entry(&msg
->pagelist
->head
,
866 } else if (msg
->bio
) {
869 bv
= bio_iovec_idx(msg
->bio_iter
, msg
->bio_seg
);
871 page_shift
= bv
->bv_offset
;
873 kaddr
= kmap(page
) + page_shift
;
874 max_write
= bv
->bv_len
;
879 kaddr
= zero_page_address
;
881 len
= min_t(int, max_write
- con
->out_msg_pos
.page_pos
,
884 if (do_crc
&& !con
->out_msg_pos
.did_page_crc
) {
886 void *base
= kaddr
+ con
->out_msg_pos
.page_pos
;
887 u32 tmpcrc
= le32_to_cpu(con
->out_msg
->footer
.data_crc
);
889 BUG_ON(kaddr
== NULL
);
890 crc
= crc32c(tmpcrc
, base
, len
);
891 con
->out_msg
->footer
.data_crc
= cpu_to_le32(crc
);
892 con
->out_msg_pos
.did_page_crc
= true;
894 ret
= kernel_sendpage(con
->sock
, page
,
895 con
->out_msg_pos
.page_pos
+ page_shift
,
897 MSG_DONTWAIT
| MSG_NOSIGNAL
|
900 if (do_crc
&& kaddr
!= zero_page_address
)
908 con
->out_msg_pos
.data_pos
+= ret
;
909 con
->out_msg_pos
.page_pos
+= ret
;
911 con
->out_msg_pos
.page_pos
= 0;
912 con
->out_msg_pos
.page
++;
913 con
->out_msg_pos
.did_page_crc
= false;
915 list_move_tail(&page
->lru
,
917 else if (msg
->pagelist
)
918 list_move_tail(&page
->lru
,
919 &msg
->pagelist
->head
);
922 iter_bio_next(&msg
->bio_iter
, &msg
->bio_seg
);
927 dout("write_partial_msg_pages %p msg %p done\n", con
, msg
);
929 /* prepare and queue up footer, too */
931 con
->out_msg
->footer
.flags
|= CEPH_MSG_FOOTER_NOCRC
;
932 ceph_con_out_kvec_reset(con
);
933 prepare_write_message_footer(con
);
942 static int write_partial_skip(struct ceph_connection
*con
)
946 while (con
->out_skip
> 0) {
948 .iov_base
= zero_page_address
,
949 .iov_len
= min(con
->out_skip
, (int)PAGE_CACHE_SIZE
)
952 ret
= ceph_tcp_sendmsg(con
->sock
, &iov
, 1, iov
.iov_len
, 1);
955 con
->out_skip
-= ret
;
963 * Prepare to read connection handshake, or an ack.
965 static void prepare_read_banner(struct ceph_connection
*con
)
967 dout("prepare_read_banner %p\n", con
);
968 con
->in_base_pos
= 0;
971 static void prepare_read_connect(struct ceph_connection
*con
)
973 dout("prepare_read_connect %p\n", con
);
974 con
->in_base_pos
= 0;
977 static void prepare_read_ack(struct ceph_connection
*con
)
979 dout("prepare_read_ack %p\n", con
);
980 con
->in_base_pos
= 0;
983 static void prepare_read_tag(struct ceph_connection
*con
)
985 dout("prepare_read_tag %p\n", con
);
986 con
->in_base_pos
= 0;
987 con
->in_tag
= CEPH_MSGR_TAG_READY
;
991 * Prepare to read a message.
993 static int prepare_read_message(struct ceph_connection
*con
)
995 dout("prepare_read_message %p\n", con
);
996 BUG_ON(con
->in_msg
!= NULL
);
997 con
->in_base_pos
= 0;
998 con
->in_front_crc
= con
->in_middle_crc
= con
->in_data_crc
= 0;
1003 static int read_partial(struct ceph_connection
*con
,
1004 int *to
, int size
, void *object
)
1007 while (con
->in_base_pos
< *to
) {
1008 int left
= *to
- con
->in_base_pos
;
1009 int have
= size
- left
;
1010 int ret
= ceph_tcp_recvmsg(con
->sock
, object
+ have
, left
);
1013 con
->in_base_pos
+= ret
;
1020 * Read all or part of the connect-side handshake on a new connection
1022 static int read_partial_banner(struct ceph_connection
*con
)
1026 dout("read_partial_banner %p at %d\n", con
, con
->in_base_pos
);
1029 ret
= read_partial(con
, &to
, strlen(CEPH_BANNER
), con
->in_banner
);
1032 ret
= read_partial(con
, &to
, sizeof(con
->actual_peer_addr
),
1033 &con
->actual_peer_addr
);
1036 ret
= read_partial(con
, &to
, sizeof(con
->peer_addr_for_me
),
1037 &con
->peer_addr_for_me
);
1044 static int read_partial_connect(struct ceph_connection
*con
)
1048 dout("read_partial_connect %p at %d\n", con
, con
->in_base_pos
);
1050 ret
= read_partial(con
, &to
, sizeof(con
->in_reply
), &con
->in_reply
);
1053 ret
= read_partial(con
, &to
, le32_to_cpu(con
->in_reply
.authorizer_len
),
1054 con
->auth_reply_buf
);
1058 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1059 con
, (int)con
->in_reply
.tag
,
1060 le32_to_cpu(con
->in_reply
.connect_seq
),
1061 le32_to_cpu(con
->in_reply
.global_seq
));
1068 * Verify the hello banner looks okay.
1070 static int verify_hello(struct ceph_connection
*con
)
1072 if (memcmp(con
->in_banner
, CEPH_BANNER
, strlen(CEPH_BANNER
))) {
1073 pr_err("connect to %s got bad banner\n",
1074 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1075 con
->error_msg
= "protocol error, bad banner";
1081 static bool addr_is_blank(struct sockaddr_storage
*ss
)
1083 switch (ss
->ss_family
) {
1085 return ((struct sockaddr_in
*)ss
)->sin_addr
.s_addr
== 0;
1088 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[0] == 0 &&
1089 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[1] == 0 &&
1090 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[2] == 0 &&
1091 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[3] == 0;
1096 static int addr_port(struct sockaddr_storage
*ss
)
1098 switch (ss
->ss_family
) {
1100 return ntohs(((struct sockaddr_in
*)ss
)->sin_port
);
1102 return ntohs(((struct sockaddr_in6
*)ss
)->sin6_port
);
1107 static void addr_set_port(struct sockaddr_storage
*ss
, int p
)
1109 switch (ss
->ss_family
) {
1111 ((struct sockaddr_in
*)ss
)->sin_port
= htons(p
);
1114 ((struct sockaddr_in6
*)ss
)->sin6_port
= htons(p
);
1120 * Unlike other *_pton function semantics, zero indicates success.
1122 static int ceph_pton(const char *str
, size_t len
, struct sockaddr_storage
*ss
,
1123 char delim
, const char **ipend
)
1125 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
1126 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
1128 memset(ss
, 0, sizeof(*ss
));
1130 if (in4_pton(str
, len
, (u8
*)&in4
->sin_addr
.s_addr
, delim
, ipend
)) {
1131 ss
->ss_family
= AF_INET
;
1135 if (in6_pton(str
, len
, (u8
*)&in6
->sin6_addr
.s6_addr
, delim
, ipend
)) {
1136 ss
->ss_family
= AF_INET6
;
1144 * Extract hostname string and resolve using kernel DNS facility.
1146 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1147 static int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1148 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1150 const char *end
, *delim_p
;
1151 char *colon_p
, *ip_addr
= NULL
;
1155 * The end of the hostname occurs immediately preceding the delimiter or
1156 * the port marker (':') where the delimiter takes precedence.
1158 delim_p
= memchr(name
, delim
, namelen
);
1159 colon_p
= memchr(name
, ':', namelen
);
1161 if (delim_p
&& colon_p
)
1162 end
= delim_p
< colon_p
? delim_p
: colon_p
;
1163 else if (!delim_p
&& colon_p
)
1167 if (!end
) /* case: hostname:/ */
1168 end
= name
+ namelen
;
1174 /* do dns_resolve upcall */
1175 ip_len
= dns_query(NULL
, name
, end
- name
, NULL
, &ip_addr
, NULL
);
1177 ret
= ceph_pton(ip_addr
, ip_len
, ss
, -1, NULL
);
1185 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end
- name
), name
,
1186 ret
, ret
? "failed" : ceph_pr_addr(ss
));
1191 static inline int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1192 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1199 * Parse a server name (IP or hostname). If a valid IP address is not found
1200 * then try to extract a hostname to resolve using userspace DNS upcall.
1202 static int ceph_parse_server_name(const char *name
, size_t namelen
,
1203 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1207 ret
= ceph_pton(name
, namelen
, ss
, delim
, ipend
);
1209 ret
= ceph_dns_resolve_name(name
, namelen
, ss
, delim
, ipend
);
1215 * Parse an ip[:port] list into an addr array. Use the default
1216 * monitor port if a port isn't specified.
1218 int ceph_parse_ips(const char *c
, const char *end
,
1219 struct ceph_entity_addr
*addr
,
1220 int max_count
, int *count
)
1222 int i
, ret
= -EINVAL
;
1225 dout("parse_ips on '%.*s'\n", (int)(end
-c
), c
);
1226 for (i
= 0; i
< max_count
; i
++) {
1228 struct sockaddr_storage
*ss
= &addr
[i
].in_addr
;
1237 ret
= ceph_parse_server_name(p
, end
- p
, ss
, delim
, &ipend
);
1246 dout("missing matching ']'\n");
1253 if (p
< end
&& *p
== ':') {
1256 while (p
< end
&& *p
>= '0' && *p
<= '9') {
1257 port
= (port
* 10) + (*p
- '0');
1260 if (port
> 65535 || port
== 0)
1263 port
= CEPH_MON_PORT
;
1266 addr_set_port(ss
, port
);
1268 dout("parse_ips got %s\n", ceph_pr_addr(ss
));
1285 pr_err("parse_ips bad ip '%.*s'\n", (int)(end
- c
), c
);
1288 EXPORT_SYMBOL(ceph_parse_ips
);
1290 static int process_banner(struct ceph_connection
*con
)
1292 dout("process_banner on %p\n", con
);
1294 if (verify_hello(con
) < 0)
1297 ceph_decode_addr(&con
->actual_peer_addr
);
1298 ceph_decode_addr(&con
->peer_addr_for_me
);
1301 * Make sure the other end is who we wanted. note that the other
1302 * end may not yet know their ip address, so if it's 0.0.0.0, give
1303 * them the benefit of the doubt.
1305 if (memcmp(&con
->peer_addr
, &con
->actual_peer_addr
,
1306 sizeof(con
->peer_addr
)) != 0 &&
1307 !(addr_is_blank(&con
->actual_peer_addr
.in_addr
) &&
1308 con
->actual_peer_addr
.nonce
== con
->peer_addr
.nonce
)) {
1309 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1310 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1311 (int)le32_to_cpu(con
->peer_addr
.nonce
),
1312 ceph_pr_addr(&con
->actual_peer_addr
.in_addr
),
1313 (int)le32_to_cpu(con
->actual_peer_addr
.nonce
));
1314 con
->error_msg
= "wrong peer at address";
1319 * did we learn our address?
1321 if (addr_is_blank(&con
->msgr
->inst
.addr
.in_addr
)) {
1322 int port
= addr_port(&con
->msgr
->inst
.addr
.in_addr
);
1324 memcpy(&con
->msgr
->inst
.addr
.in_addr
,
1325 &con
->peer_addr_for_me
.in_addr
,
1326 sizeof(con
->peer_addr_for_me
.in_addr
));
1327 addr_set_port(&con
->msgr
->inst
.addr
.in_addr
, port
);
1328 encode_my_addr(con
->msgr
);
1329 dout("process_banner learned my addr is %s\n",
1330 ceph_pr_addr(&con
->msgr
->inst
.addr
.in_addr
));
1333 set_bit(NEGOTIATING
, &con
->state
);
1334 prepare_read_connect(con
);
1338 static void fail_protocol(struct ceph_connection
*con
)
1340 reset_connection(con
);
1341 set_bit(CLOSED
, &con
->state
); /* in case there's queued work */
1343 mutex_unlock(&con
->mutex
);
1344 if (con
->ops
->bad_proto
)
1345 con
->ops
->bad_proto(con
);
1346 mutex_lock(&con
->mutex
);
1349 static int process_connect(struct ceph_connection
*con
)
1351 u64 sup_feat
= con
->msgr
->supported_features
;
1352 u64 req_feat
= con
->msgr
->required_features
;
1353 u64 server_feat
= le64_to_cpu(con
->in_reply
.features
);
1356 dout("process_connect on %p tag %d\n", con
, (int)con
->in_tag
);
1358 switch (con
->in_reply
.tag
) {
1359 case CEPH_MSGR_TAG_FEATURES
:
1360 pr_err("%s%lld %s feature set mismatch,"
1361 " my %llx < server's %llx, missing %llx\n",
1362 ENTITY_NAME(con
->peer_name
),
1363 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1364 sup_feat
, server_feat
, server_feat
& ~sup_feat
);
1365 con
->error_msg
= "missing required protocol features";
1369 case CEPH_MSGR_TAG_BADPROTOVER
:
1370 pr_err("%s%lld %s protocol version mismatch,"
1371 " my %d != server's %d\n",
1372 ENTITY_NAME(con
->peer_name
),
1373 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1374 le32_to_cpu(con
->out_connect
.protocol_version
),
1375 le32_to_cpu(con
->in_reply
.protocol_version
));
1376 con
->error_msg
= "protocol version mismatch";
1380 case CEPH_MSGR_TAG_BADAUTHORIZER
:
1382 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con
,
1384 if (con
->auth_retry
== 2) {
1385 con
->error_msg
= "connect authorization failure";
1388 con
->auth_retry
= 1;
1389 ret
= prepare_write_connect(con
->msgr
, con
, 0);
1392 prepare_read_connect(con
);
1395 case CEPH_MSGR_TAG_RESETSESSION
:
1397 * If we connected with a large connect_seq but the peer
1398 * has no record of a session with us (no connection, or
1399 * connect_seq == 0), they will send RESETSESION to indicate
1400 * that they must have reset their session, and may have
1403 dout("process_connect got RESET peer seq %u\n",
1404 le32_to_cpu(con
->in_connect
.connect_seq
));
1405 pr_err("%s%lld %s connection reset\n",
1406 ENTITY_NAME(con
->peer_name
),
1407 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1408 reset_connection(con
);
1409 prepare_write_connect(con
->msgr
, con
, 0);
1410 prepare_read_connect(con
);
1412 /* Tell ceph about it. */
1413 mutex_unlock(&con
->mutex
);
1414 pr_info("reset on %s%lld\n", ENTITY_NAME(con
->peer_name
));
1415 if (con
->ops
->peer_reset
)
1416 con
->ops
->peer_reset(con
);
1417 mutex_lock(&con
->mutex
);
1418 if (test_bit(CLOSED
, &con
->state
) ||
1419 test_bit(OPENING
, &con
->state
))
1423 case CEPH_MSGR_TAG_RETRY_SESSION
:
1425 * If we sent a smaller connect_seq than the peer has, try
1426 * again with a larger value.
1428 dout("process_connect got RETRY my seq = %u, peer_seq = %u\n",
1429 le32_to_cpu(con
->out_connect
.connect_seq
),
1430 le32_to_cpu(con
->in_connect
.connect_seq
));
1431 con
->connect_seq
= le32_to_cpu(con
->in_connect
.connect_seq
);
1432 prepare_write_connect(con
->msgr
, con
, 0);
1433 prepare_read_connect(con
);
1436 case CEPH_MSGR_TAG_RETRY_GLOBAL
:
1438 * If we sent a smaller global_seq than the peer has, try
1439 * again with a larger value.
1441 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
1442 con
->peer_global_seq
,
1443 le32_to_cpu(con
->in_connect
.global_seq
));
1444 get_global_seq(con
->msgr
,
1445 le32_to_cpu(con
->in_connect
.global_seq
));
1446 prepare_write_connect(con
->msgr
, con
, 0);
1447 prepare_read_connect(con
);
1450 case CEPH_MSGR_TAG_READY
:
1451 if (req_feat
& ~server_feat
) {
1452 pr_err("%s%lld %s protocol feature mismatch,"
1453 " my required %llx > server's %llx, need %llx\n",
1454 ENTITY_NAME(con
->peer_name
),
1455 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1456 req_feat
, server_feat
, req_feat
& ~server_feat
);
1457 con
->error_msg
= "missing required protocol features";
1461 clear_bit(CONNECTING
, &con
->state
);
1462 con
->peer_global_seq
= le32_to_cpu(con
->in_reply
.global_seq
);
1464 con
->peer_features
= server_feat
;
1465 dout("process_connect got READY gseq %d cseq %d (%d)\n",
1466 con
->peer_global_seq
,
1467 le32_to_cpu(con
->in_reply
.connect_seq
),
1469 WARN_ON(con
->connect_seq
!=
1470 le32_to_cpu(con
->in_reply
.connect_seq
));
1472 if (con
->in_reply
.flags
& CEPH_MSG_CONNECT_LOSSY
)
1473 set_bit(LOSSYTX
, &con
->state
);
1475 prepare_read_tag(con
);
1478 case CEPH_MSGR_TAG_WAIT
:
1480 * If there is a connection race (we are opening
1481 * connections to each other), one of us may just have
1482 * to WAIT. This shouldn't happen if we are the
1485 pr_err("process_connect got WAIT as client\n");
1486 con
->error_msg
= "protocol error, got WAIT as client";
1490 pr_err("connect protocol error, will retry\n");
1491 con
->error_msg
= "protocol error, garbage tag during connect";
1499 * read (part of) an ack
1501 static int read_partial_ack(struct ceph_connection
*con
)
1505 return read_partial(con
, &to
, sizeof(con
->in_temp_ack
),
1511 * We can finally discard anything that's been acked.
1513 static void process_ack(struct ceph_connection
*con
)
1516 u64 ack
= le64_to_cpu(con
->in_temp_ack
);
1519 while (!list_empty(&con
->out_sent
)) {
1520 m
= list_first_entry(&con
->out_sent
, struct ceph_msg
,
1522 seq
= le64_to_cpu(m
->hdr
.seq
);
1525 dout("got ack for seq %llu type %d at %p\n", seq
,
1526 le16_to_cpu(m
->hdr
.type
), m
);
1527 m
->ack_stamp
= jiffies
;
1530 prepare_read_tag(con
);
1536 static int read_partial_message_section(struct ceph_connection
*con
,
1537 struct kvec
*section
,
1538 unsigned int sec_len
, u32
*crc
)
1544 while (section
->iov_len
< sec_len
) {
1545 BUG_ON(section
->iov_base
== NULL
);
1546 left
= sec_len
- section
->iov_len
;
1547 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)section
->iov_base
+
1548 section
->iov_len
, left
);
1551 section
->iov_len
+= ret
;
1553 if (section
->iov_len
== sec_len
)
1554 *crc
= crc32c(0, section
->iov_base
, section
->iov_len
);
1559 static struct ceph_msg
*ceph_alloc_msg(struct ceph_connection
*con
,
1560 struct ceph_msg_header
*hdr
,
1564 static int read_partial_message_pages(struct ceph_connection
*con
,
1565 struct page
**pages
,
1566 unsigned data_len
, bool do_datacrc
)
1572 left
= min((int)(data_len
- con
->in_msg_pos
.data_pos
),
1573 (int)(PAGE_SIZE
- con
->in_msg_pos
.page_pos
));
1575 BUG_ON(pages
== NULL
);
1576 p
= kmap(pages
[con
->in_msg_pos
.page
]);
1577 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ con
->in_msg_pos
.page_pos
,
1579 if (ret
> 0 && do_datacrc
)
1581 crc32c(con
->in_data_crc
,
1582 p
+ con
->in_msg_pos
.page_pos
, ret
);
1583 kunmap(pages
[con
->in_msg_pos
.page
]);
1586 con
->in_msg_pos
.data_pos
+= ret
;
1587 con
->in_msg_pos
.page_pos
+= ret
;
1588 if (con
->in_msg_pos
.page_pos
== PAGE_SIZE
) {
1589 con
->in_msg_pos
.page_pos
= 0;
1590 con
->in_msg_pos
.page
++;
1597 static int read_partial_message_bio(struct ceph_connection
*con
,
1598 struct bio
**bio_iter
, int *bio_seg
,
1599 unsigned data_len
, bool do_datacrc
)
1601 struct bio_vec
*bv
= bio_iovec_idx(*bio_iter
, *bio_seg
);
1608 left
= min((int)(data_len
- con
->in_msg_pos
.data_pos
),
1609 (int)(bv
->bv_len
- con
->in_msg_pos
.page_pos
));
1611 p
= kmap(bv
->bv_page
) + bv
->bv_offset
;
1613 ret
= ceph_tcp_recvmsg(con
->sock
, p
+ con
->in_msg_pos
.page_pos
,
1615 if (ret
> 0 && do_datacrc
)
1617 crc32c(con
->in_data_crc
,
1618 p
+ con
->in_msg_pos
.page_pos
, ret
);
1619 kunmap(bv
->bv_page
);
1622 con
->in_msg_pos
.data_pos
+= ret
;
1623 con
->in_msg_pos
.page_pos
+= ret
;
1624 if (con
->in_msg_pos
.page_pos
== bv
->bv_len
) {
1625 con
->in_msg_pos
.page_pos
= 0;
1626 iter_bio_next(bio_iter
, bio_seg
);
1634 * read (part of) a message.
1636 static int read_partial_message(struct ceph_connection
*con
)
1638 struct ceph_msg
*m
= con
->in_msg
;
1641 unsigned front_len
, middle_len
, data_len
;
1642 bool do_datacrc
= con
->msgr
->nocrc
;
1647 dout("read_partial_message con %p msg %p\n", con
, m
);
1650 while (con
->in_base_pos
< sizeof(con
->in_hdr
)) {
1651 left
= sizeof(con
->in_hdr
) - con
->in_base_pos
;
1652 ret
= ceph_tcp_recvmsg(con
->sock
,
1653 (char *)&con
->in_hdr
+ con
->in_base_pos
,
1657 con
->in_base_pos
+= ret
;
1660 crc
= crc32c(0, &con
->in_hdr
, offsetof(struct ceph_msg_header
, crc
));
1661 if (cpu_to_le32(crc
) != con
->in_hdr
.crc
) {
1662 pr_err("read_partial_message bad hdr "
1663 " crc %u != expected %u\n",
1664 crc
, con
->in_hdr
.crc
);
1668 front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
1669 if (front_len
> CEPH_MSG_MAX_FRONT_LEN
)
1671 middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
1672 if (middle_len
> CEPH_MSG_MAX_DATA_LEN
)
1674 data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
1675 if (data_len
> CEPH_MSG_MAX_DATA_LEN
)
1679 seq
= le64_to_cpu(con
->in_hdr
.seq
);
1680 if ((s64
)seq
- (s64
)con
->in_seq
< 1) {
1681 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
1682 ENTITY_NAME(con
->peer_name
),
1683 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1684 seq
, con
->in_seq
+ 1);
1685 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1687 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1689 } else if ((s64
)seq
- (s64
)con
->in_seq
> 1) {
1690 pr_err("read_partial_message bad seq %lld expected %lld\n",
1691 seq
, con
->in_seq
+ 1);
1692 con
->error_msg
= "bad message sequence # for incoming message";
1696 /* allocate message? */
1698 dout("got hdr type %d front %d data %d\n", con
->in_hdr
.type
,
1699 con
->in_hdr
.front_len
, con
->in_hdr
.data_len
);
1701 con
->in_msg
= ceph_alloc_msg(con
, &con
->in_hdr
, &skip
);
1703 /* skip this message */
1704 dout("alloc_msg said skip message\n");
1705 BUG_ON(con
->in_msg
);
1706 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
1708 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1714 "error allocating memory for incoming message";
1718 m
->front
.iov_len
= 0; /* haven't read it yet */
1720 m
->middle
->vec
.iov_len
= 0;
1722 con
->in_msg_pos
.page
= 0;
1724 con
->in_msg_pos
.page_pos
= m
->page_alignment
;
1726 con
->in_msg_pos
.page_pos
= 0;
1727 con
->in_msg_pos
.data_pos
= 0;
1731 ret
= read_partial_message_section(con
, &m
->front
, front_len
,
1732 &con
->in_front_crc
);
1738 ret
= read_partial_message_section(con
, &m
->middle
->vec
,
1740 &con
->in_middle_crc
);
1745 if (m
->bio
&& !m
->bio_iter
)
1746 init_bio_iter(m
->bio
, &m
->bio_iter
, &m
->bio_seg
);
1750 while (con
->in_msg_pos
.data_pos
< data_len
) {
1752 ret
= read_partial_message_pages(con
, m
->pages
,
1753 data_len
, do_datacrc
);
1757 } else if (m
->bio
) {
1759 ret
= read_partial_message_bio(con
,
1760 &m
->bio_iter
, &m
->bio_seg
,
1761 data_len
, do_datacrc
);
1771 to
= sizeof(m
->hdr
) + sizeof(m
->footer
);
1772 while (con
->in_base_pos
< to
) {
1773 left
= to
- con
->in_base_pos
;
1774 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)&m
->footer
+
1775 (con
->in_base_pos
- sizeof(m
->hdr
)),
1779 con
->in_base_pos
+= ret
;
1781 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
1782 m
, front_len
, m
->footer
.front_crc
, middle_len
,
1783 m
->footer
.middle_crc
, data_len
, m
->footer
.data_crc
);
1786 if (con
->in_front_crc
!= le32_to_cpu(m
->footer
.front_crc
)) {
1787 pr_err("read_partial_message %p front crc %u != exp. %u\n",
1788 m
, con
->in_front_crc
, m
->footer
.front_crc
);
1791 if (con
->in_middle_crc
!= le32_to_cpu(m
->footer
.middle_crc
)) {
1792 pr_err("read_partial_message %p middle crc %u != exp %u\n",
1793 m
, con
->in_middle_crc
, m
->footer
.middle_crc
);
1797 (m
->footer
.flags
& CEPH_MSG_FOOTER_NOCRC
) == 0 &&
1798 con
->in_data_crc
!= le32_to_cpu(m
->footer
.data_crc
)) {
1799 pr_err("read_partial_message %p data crc %u != exp. %u\n", m
,
1800 con
->in_data_crc
, le32_to_cpu(m
->footer
.data_crc
));
1804 return 1; /* done! */
1808 * Process message. This happens in the worker thread. The callback should
1809 * be careful not to do anything that waits on other incoming messages or it
1812 static void process_message(struct ceph_connection
*con
)
1814 struct ceph_msg
*msg
;
1819 /* if first message, set peer_name */
1820 if (con
->peer_name
.type
== 0)
1821 con
->peer_name
= msg
->hdr
.src
;
1824 mutex_unlock(&con
->mutex
);
1826 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
1827 msg
, le64_to_cpu(msg
->hdr
.seq
),
1828 ENTITY_NAME(msg
->hdr
.src
),
1829 le16_to_cpu(msg
->hdr
.type
),
1830 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
1831 le32_to_cpu(msg
->hdr
.front_len
),
1832 le32_to_cpu(msg
->hdr
.data_len
),
1833 con
->in_front_crc
, con
->in_middle_crc
, con
->in_data_crc
);
1834 con
->ops
->dispatch(con
, msg
);
1836 mutex_lock(&con
->mutex
);
1837 prepare_read_tag(con
);
1842 * Write something to the socket. Called in a worker thread when the
1843 * socket appears to be writeable and we have something ready to send.
1845 static int try_write(struct ceph_connection
*con
)
1847 struct ceph_messenger
*msgr
= con
->msgr
;
1850 dout("try_write start %p state %lu nref %d\n", con
, con
->state
,
1851 atomic_read(&con
->nref
));
1854 dout("try_write out_kvec_bytes %d\n", con
->out_kvec_bytes
);
1856 /* open the socket first? */
1857 if (con
->sock
== NULL
) {
1858 prepare_write_connect(msgr
, con
, 1);
1859 prepare_read_banner(con
);
1860 set_bit(CONNECTING
, &con
->state
);
1861 clear_bit(NEGOTIATING
, &con
->state
);
1863 BUG_ON(con
->in_msg
);
1864 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1865 dout("try_write initiating connect on %p new state %lu\n",
1867 ret
= ceph_tcp_connect(con
);
1869 con
->error_msg
= "connect error";
1875 /* kvec data queued? */
1876 if (con
->out_skip
) {
1877 ret
= write_partial_skip(con
);
1881 if (con
->out_kvec_left
) {
1882 ret
= write_partial_kvec(con
);
1889 if (con
->out_msg_done
) {
1890 ceph_msg_put(con
->out_msg
);
1891 con
->out_msg
= NULL
; /* we're done with this one */
1895 ret
= write_partial_msg_pages(con
);
1897 goto more_kvec
; /* we need to send the footer, too! */
1901 dout("try_write write_partial_msg_pages err %d\n",
1908 if (!test_bit(CONNECTING
, &con
->state
)) {
1909 /* is anything else pending? */
1910 if (!list_empty(&con
->out_queue
)) {
1911 prepare_write_message(con
);
1914 if (con
->in_seq
> con
->in_seq_acked
) {
1915 prepare_write_ack(con
);
1918 if (test_and_clear_bit(KEEPALIVE_PENDING
, &con
->state
)) {
1919 prepare_write_keepalive(con
);
1924 /* Nothing to do! */
1925 clear_bit(WRITE_PENDING
, &con
->state
);
1926 dout("try_write nothing else to write.\n");
1929 dout("try_write done on %p ret %d\n", con
, ret
);
1936 * Read what we can from the socket.
1938 static int try_read(struct ceph_connection
*con
)
1945 if (test_bit(STANDBY
, &con
->state
))
1948 dout("try_read start on %p\n", con
);
1951 dout("try_read tag %d in_base_pos %d\n", (int)con
->in_tag
,
1955 * process_connect and process_message drop and re-take
1956 * con->mutex. make sure we handle a racing close or reopen.
1958 if (test_bit(CLOSED
, &con
->state
) ||
1959 test_bit(OPENING
, &con
->state
)) {
1964 if (test_bit(CONNECTING
, &con
->state
)) {
1965 if (!test_bit(NEGOTIATING
, &con
->state
)) {
1966 dout("try_read connecting\n");
1967 ret
= read_partial_banner(con
);
1970 ret
= process_banner(con
);
1974 ret
= read_partial_connect(con
);
1977 ret
= process_connect(con
);
1983 if (con
->in_base_pos
< 0) {
1985 * skipping + discarding content.
1987 * FIXME: there must be a better way to do this!
1989 static char buf
[SKIP_BUF_SIZE
];
1990 int skip
= min((int) sizeof (buf
), -con
->in_base_pos
);
1992 dout("skipping %d / %d bytes\n", skip
, -con
->in_base_pos
);
1993 ret
= ceph_tcp_recvmsg(con
->sock
, buf
, skip
);
1996 con
->in_base_pos
+= ret
;
1997 if (con
->in_base_pos
)
2000 if (con
->in_tag
== CEPH_MSGR_TAG_READY
) {
2004 ret
= ceph_tcp_recvmsg(con
->sock
, &con
->in_tag
, 1);
2007 dout("try_read got tag %d\n", (int)con
->in_tag
);
2008 switch (con
->in_tag
) {
2009 case CEPH_MSGR_TAG_MSG
:
2010 prepare_read_message(con
);
2012 case CEPH_MSGR_TAG_ACK
:
2013 prepare_read_ack(con
);
2015 case CEPH_MSGR_TAG_CLOSE
:
2016 set_bit(CLOSED
, &con
->state
); /* fixme */
2022 if (con
->in_tag
== CEPH_MSGR_TAG_MSG
) {
2023 ret
= read_partial_message(con
);
2027 con
->error_msg
= "bad crc";
2031 con
->error_msg
= "io error";
2036 if (con
->in_tag
== CEPH_MSGR_TAG_READY
)
2038 process_message(con
);
2041 if (con
->in_tag
== CEPH_MSGR_TAG_ACK
) {
2042 ret
= read_partial_ack(con
);
2050 dout("try_read done on %p ret %d\n", con
, ret
);
2054 pr_err("try_read bad con->in_tag = %d\n", (int)con
->in_tag
);
2055 con
->error_msg
= "protocol error, garbage tag";
2062 * Atomically queue work on a connection. Bump @con reference to
2063 * avoid races with connection teardown.
2065 static void queue_con(struct ceph_connection
*con
)
2067 if (test_bit(DEAD
, &con
->state
)) {
2068 dout("queue_con %p ignoring: DEAD\n",
2073 if (!con
->ops
->get(con
)) {
2074 dout("queue_con %p ref count 0\n", con
);
2078 if (!queue_delayed_work(ceph_msgr_wq
, &con
->work
, 0)) {
2079 dout("queue_con %p - already queued\n", con
);
2082 dout("queue_con %p\n", con
);
2087 * Do some work on a connection. Drop a connection ref when we're done.
2089 static void con_work(struct work_struct
*work
)
2091 struct ceph_connection
*con
= container_of(work
, struct ceph_connection
,
2095 mutex_lock(&con
->mutex
);
2097 if (test_and_clear_bit(BACKOFF
, &con
->state
)) {
2098 dout("con_work %p backing off\n", con
);
2099 if (queue_delayed_work(ceph_msgr_wq
, &con
->work
,
2100 round_jiffies_relative(con
->delay
))) {
2101 dout("con_work %p backoff %lu\n", con
, con
->delay
);
2102 mutex_unlock(&con
->mutex
);
2106 dout("con_work %p FAILED to back off %lu\n", con
,
2111 if (test_bit(STANDBY
, &con
->state
)) {
2112 dout("con_work %p STANDBY\n", con
);
2115 if (test_bit(CLOSED
, &con
->state
)) { /* e.g. if we are replaced */
2116 dout("con_work CLOSED\n");
2117 con_close_socket(con
);
2120 if (test_and_clear_bit(OPENING
, &con
->state
)) {
2121 /* reopen w/ new peer */
2122 dout("con_work OPENING\n");
2123 con_close_socket(con
);
2126 if (test_and_clear_bit(SOCK_CLOSED
, &con
->state
))
2129 ret
= try_read(con
);
2135 ret
= try_write(con
);
2142 mutex_unlock(&con
->mutex
);
2148 mutex_unlock(&con
->mutex
);
2149 ceph_fault(con
); /* error/fault path */
2155 * Generic error/fault handler. A retry mechanism is used with
2156 * exponential backoff
2158 static void ceph_fault(struct ceph_connection
*con
)
2160 pr_err("%s%lld %s %s\n", ENTITY_NAME(con
->peer_name
),
2161 ceph_pr_addr(&con
->peer_addr
.in_addr
), con
->error_msg
);
2162 dout("fault %p state %lu to peer %s\n",
2163 con
, con
->state
, ceph_pr_addr(&con
->peer_addr
.in_addr
));
2165 if (test_bit(LOSSYTX
, &con
->state
)) {
2166 dout("fault on LOSSYTX channel\n");
2170 mutex_lock(&con
->mutex
);
2171 if (test_bit(CLOSED
, &con
->state
))
2174 con_close_socket(con
);
2177 ceph_msg_put(con
->in_msg
);
2181 /* Requeue anything that hasn't been acked */
2182 list_splice_init(&con
->out_sent
, &con
->out_queue
);
2184 /* If there are no messages queued or keepalive pending, place
2185 * the connection in a STANDBY state */
2186 if (list_empty(&con
->out_queue
) &&
2187 !test_bit(KEEPALIVE_PENDING
, &con
->state
)) {
2188 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con
);
2189 clear_bit(WRITE_PENDING
, &con
->state
);
2190 set_bit(STANDBY
, &con
->state
);
2192 /* retry after a delay. */
2193 if (con
->delay
== 0)
2194 con
->delay
= BASE_DELAY_INTERVAL
;
2195 else if (con
->delay
< MAX_DELAY_INTERVAL
)
2198 if (queue_delayed_work(ceph_msgr_wq
, &con
->work
,
2199 round_jiffies_relative(con
->delay
))) {
2200 dout("fault queued %p delay %lu\n", con
, con
->delay
);
2203 dout("fault failed to queue %p delay %lu, backoff\n",
2206 * In many cases we see a socket state change
2207 * while con_work is running and end up
2208 * queuing (non-delayed) work, such that we
2209 * can't backoff with a delay. Set a flag so
2210 * that when con_work restarts we schedule the
2213 set_bit(BACKOFF
, &con
->state
);
2218 mutex_unlock(&con
->mutex
);
2221 * in case we faulted due to authentication, invalidate our
2222 * current tickets so that we can get new ones.
2224 if (con
->auth_retry
&& con
->ops
->invalidate_authorizer
) {
2225 dout("calling invalidate_authorizer()\n");
2226 con
->ops
->invalidate_authorizer(con
);
2229 if (con
->ops
->fault
)
2230 con
->ops
->fault(con
);
2236 * create a new messenger instance
2238 struct ceph_messenger
*ceph_messenger_create(struct ceph_entity_addr
*myaddr
,
2239 u32 supported_features
,
2240 u32 required_features
)
2242 struct ceph_messenger
*msgr
;
2244 msgr
= kzalloc(sizeof(*msgr
), GFP_KERNEL
);
2246 return ERR_PTR(-ENOMEM
);
2248 msgr
->supported_features
= supported_features
;
2249 msgr
->required_features
= required_features
;
2251 spin_lock_init(&msgr
->global_seq_lock
);
2254 msgr
->inst
.addr
= *myaddr
;
2256 /* select a random nonce */
2257 msgr
->inst
.addr
.type
= 0;
2258 get_random_bytes(&msgr
->inst
.addr
.nonce
, sizeof(msgr
->inst
.addr
.nonce
));
2259 encode_my_addr(msgr
);
2261 dout("messenger_create %p\n", msgr
);
2264 EXPORT_SYMBOL(ceph_messenger_create
);
2266 void ceph_messenger_destroy(struct ceph_messenger
*msgr
)
2268 dout("destroy %p\n", msgr
);
2270 dout("destroyed messenger %p\n", msgr
);
2272 EXPORT_SYMBOL(ceph_messenger_destroy
);
2274 static void clear_standby(struct ceph_connection
*con
)
2276 /* come back from STANDBY? */
2277 if (test_and_clear_bit(STANDBY
, &con
->state
)) {
2278 mutex_lock(&con
->mutex
);
2279 dout("clear_standby %p and ++connect_seq\n", con
);
2281 WARN_ON(test_bit(WRITE_PENDING
, &con
->state
));
2282 WARN_ON(test_bit(KEEPALIVE_PENDING
, &con
->state
));
2283 mutex_unlock(&con
->mutex
);
2288 * Queue up an outgoing message on the given connection.
2290 void ceph_con_send(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2292 if (test_bit(CLOSED
, &con
->state
)) {
2293 dout("con_send %p closed, dropping %p\n", con
, msg
);
2299 msg
->hdr
.src
= con
->msgr
->inst
.name
;
2301 BUG_ON(msg
->front
.iov_len
!= le32_to_cpu(msg
->hdr
.front_len
));
2303 msg
->needs_out_seq
= true;
2306 mutex_lock(&con
->mutex
);
2307 BUG_ON(!list_empty(&msg
->list_head
));
2308 list_add_tail(&msg
->list_head
, &con
->out_queue
);
2309 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg
,
2310 ENTITY_NAME(con
->peer_name
), le16_to_cpu(msg
->hdr
.type
),
2311 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2312 le32_to_cpu(msg
->hdr
.front_len
),
2313 le32_to_cpu(msg
->hdr
.middle_len
),
2314 le32_to_cpu(msg
->hdr
.data_len
));
2315 mutex_unlock(&con
->mutex
);
2317 /* if there wasn't anything waiting to send before, queue
2320 if (test_and_set_bit(WRITE_PENDING
, &con
->state
) == 0)
2323 EXPORT_SYMBOL(ceph_con_send
);
2326 * Revoke a message that was previously queued for send
2328 void ceph_con_revoke(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2330 mutex_lock(&con
->mutex
);
2331 if (!list_empty(&msg
->list_head
)) {
2332 dout("con_revoke %p msg %p - was on queue\n", con
, msg
);
2333 list_del_init(&msg
->list_head
);
2337 if (con
->out_msg
== msg
) {
2338 dout("con_revoke %p msg %p - was sending\n", con
, msg
);
2339 con
->out_msg
= NULL
;
2340 if (con
->out_kvec_is_msg
) {
2341 con
->out_skip
= con
->out_kvec_bytes
;
2342 con
->out_kvec_is_msg
= false;
2347 mutex_unlock(&con
->mutex
);
2351 * Revoke a message that we may be reading data into
2353 void ceph_con_revoke_message(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2355 mutex_lock(&con
->mutex
);
2356 if (con
->in_msg
&& con
->in_msg
== msg
) {
2357 unsigned front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
2358 unsigned middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
2359 unsigned data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
2361 /* skip rest of message */
2362 dout("con_revoke_pages %p msg %p revoked\n", con
, msg
);
2363 con
->in_base_pos
= con
->in_base_pos
-
2364 sizeof(struct ceph_msg_header
) -
2368 sizeof(struct ceph_msg_footer
);
2369 ceph_msg_put(con
->in_msg
);
2371 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2374 dout("con_revoke_pages %p msg %p pages %p no-op\n",
2375 con
, con
->in_msg
, msg
);
2377 mutex_unlock(&con
->mutex
);
2381 * Queue a keepalive byte to ensure the tcp connection is alive.
2383 void ceph_con_keepalive(struct ceph_connection
*con
)
2385 dout("con_keepalive %p\n", con
);
2387 if (test_and_set_bit(KEEPALIVE_PENDING
, &con
->state
) == 0 &&
2388 test_and_set_bit(WRITE_PENDING
, &con
->state
) == 0)
2391 EXPORT_SYMBOL(ceph_con_keepalive
);
2395 * construct a new message with given type, size
2396 * the new msg has a ref count of 1.
2398 struct ceph_msg
*ceph_msg_new(int type
, int front_len
, gfp_t flags
,
2403 m
= kmalloc(sizeof(*m
), flags
);
2406 kref_init(&m
->kref
);
2407 INIT_LIST_HEAD(&m
->list_head
);
2410 m
->hdr
.type
= cpu_to_le16(type
);
2411 m
->hdr
.priority
= cpu_to_le16(CEPH_MSG_PRIO_DEFAULT
);
2413 m
->hdr
.front_len
= cpu_to_le32(front_len
);
2414 m
->hdr
.middle_len
= 0;
2415 m
->hdr
.data_len
= 0;
2416 m
->hdr
.data_off
= 0;
2417 m
->hdr
.reserved
= 0;
2418 m
->footer
.front_crc
= 0;
2419 m
->footer
.middle_crc
= 0;
2420 m
->footer
.data_crc
= 0;
2421 m
->footer
.flags
= 0;
2422 m
->front_max
= front_len
;
2423 m
->front_is_vmalloc
= false;
2424 m
->more_to_follow
= false;
2433 m
->page_alignment
= 0;
2443 if (front_len
> PAGE_CACHE_SIZE
) {
2444 m
->front
.iov_base
= __vmalloc(front_len
, flags
,
2446 m
->front_is_vmalloc
= true;
2448 m
->front
.iov_base
= kmalloc(front_len
, flags
);
2450 if (m
->front
.iov_base
== NULL
) {
2451 dout("ceph_msg_new can't allocate %d bytes\n",
2456 m
->front
.iov_base
= NULL
;
2458 m
->front
.iov_len
= front_len
;
2460 dout("ceph_msg_new %p front %d\n", m
, front_len
);
2467 pr_err("msg_new can't create type %d front %d\n", type
,
2471 dout("msg_new can't create type %d front %d\n", type
,
2476 EXPORT_SYMBOL(ceph_msg_new
);
2479 * Allocate "middle" portion of a message, if it is needed and wasn't
2480 * allocated by alloc_msg. This allows us to read a small fixed-size
2481 * per-type header in the front and then gracefully fail (i.e.,
2482 * propagate the error to the caller based on info in the front) when
2483 * the middle is too large.
2485 static int ceph_alloc_middle(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2487 int type
= le16_to_cpu(msg
->hdr
.type
);
2488 int middle_len
= le32_to_cpu(msg
->hdr
.middle_len
);
2490 dout("alloc_middle %p type %d %s middle_len %d\n", msg
, type
,
2491 ceph_msg_type_name(type
), middle_len
);
2492 BUG_ON(!middle_len
);
2493 BUG_ON(msg
->middle
);
2495 msg
->middle
= ceph_buffer_new(middle_len
, GFP_NOFS
);
2502 * Generic message allocator, for incoming messages.
2504 static struct ceph_msg
*ceph_alloc_msg(struct ceph_connection
*con
,
2505 struct ceph_msg_header
*hdr
,
2508 int type
= le16_to_cpu(hdr
->type
);
2509 int front_len
= le32_to_cpu(hdr
->front_len
);
2510 int middle_len
= le32_to_cpu(hdr
->middle_len
);
2511 struct ceph_msg
*msg
= NULL
;
2514 if (con
->ops
->alloc_msg
) {
2515 mutex_unlock(&con
->mutex
);
2516 msg
= con
->ops
->alloc_msg(con
, hdr
, skip
);
2517 mutex_lock(&con
->mutex
);
2523 msg
= ceph_msg_new(type
, front_len
, GFP_NOFS
, false);
2525 pr_err("unable to allocate msg type %d len %d\n",
2529 msg
->page_alignment
= le16_to_cpu(hdr
->data_off
);
2531 memcpy(&msg
->hdr
, &con
->in_hdr
, sizeof(con
->in_hdr
));
2533 if (middle_len
&& !msg
->middle
) {
2534 ret
= ceph_alloc_middle(con
, msg
);
2546 * Free a generically kmalloc'd message.
2548 void ceph_msg_kfree(struct ceph_msg
*m
)
2550 dout("msg_kfree %p\n", m
);
2551 if (m
->front_is_vmalloc
)
2552 vfree(m
->front
.iov_base
);
2554 kfree(m
->front
.iov_base
);
2559 * Drop a msg ref. Destroy as needed.
2561 void ceph_msg_last_put(struct kref
*kref
)
2563 struct ceph_msg
*m
= container_of(kref
, struct ceph_msg
, kref
);
2565 dout("ceph_msg_put last one on %p\n", m
);
2566 WARN_ON(!list_empty(&m
->list_head
));
2568 /* drop middle, data, if any */
2570 ceph_buffer_put(m
->middle
);
2577 ceph_pagelist_release(m
->pagelist
);
2585 ceph_msgpool_put(m
->pool
, m
);
2589 EXPORT_SYMBOL(ceph_msg_last_put
);
2591 void ceph_msg_dump(struct ceph_msg
*msg
)
2593 pr_debug("msg_dump %p (front_max %d nr_pages %d)\n", msg
,
2594 msg
->front_max
, msg
->nr_pages
);
2595 print_hex_dump(KERN_DEBUG
, "header: ",
2596 DUMP_PREFIX_OFFSET
, 16, 1,
2597 &msg
->hdr
, sizeof(msg
->hdr
), true);
2598 print_hex_dump(KERN_DEBUG
, " front: ",
2599 DUMP_PREFIX_OFFSET
, 16, 1,
2600 msg
->front
.iov_base
, msg
->front
.iov_len
, true);
2602 print_hex_dump(KERN_DEBUG
, "middle: ",
2603 DUMP_PREFIX_OFFSET
, 16, 1,
2604 msg
->middle
->vec
.iov_base
,
2605 msg
->middle
->vec
.iov_len
, true);
2606 print_hex_dump(KERN_DEBUG
, "footer: ",
2607 DUMP_PREFIX_OFFSET
, 16, 1,
2608 &msg
->footer
, sizeof(msg
->footer
), true);
2610 EXPORT_SYMBOL(ceph_msg_dump
);