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/ceph_features.h>
19 #include <linux/ceph/libceph.h>
20 #include <linux/ceph/messenger.h>
21 #include <linux/ceph/decode.h>
22 #include <linux/ceph/pagelist.h>
23 #include <linux/export.h>
25 #define list_entry_next(pos, member) \
26 list_entry(pos->member.next, typeof(*pos), member)
29 * Ceph uses the messenger to exchange ceph_msg messages with other
30 * hosts in the system. The messenger provides ordered and reliable
31 * delivery. We tolerate TCP disconnects by reconnecting (with
32 * exponential backoff) in the case of a fault (disconnection, bad
33 * crc, protocol error). Acks allow sent messages to be discarded by
38 * We track the state of the socket on a given connection using
39 * values defined below. The transition to a new socket state is
40 * handled by a function which verifies we aren't coming from an
44 * | NEW* | transient initial state
46 * | con_sock_state_init()
49 * | CLOSED | initialized, but no socket (and no
50 * ---------- TCP connection)
52 * | \ con_sock_state_connecting()
53 * | ----------------------
55 * + con_sock_state_closed() \
56 * |+--------------------------- \
59 * | | CLOSING | socket event; \ \
60 * | ----------- await close \ \
63 * | + con_sock_state_closing() \ |
65 * | / --------------- | |
68 * | / -----------------| CONNECTING | socket created, TCP
69 * | | / -------------- connect initiated
70 * | | | con_sock_state_connected()
73 * | CONNECTED | TCP connection established
76 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
79 #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
80 #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
81 #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
82 #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
83 #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
88 #define CON_STATE_CLOSED 1 /* -> PREOPEN */
89 #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
90 #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
91 #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
92 #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
93 #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
96 * ceph_connection flag bits
98 #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
99 * messages on errors */
100 #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
101 #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
102 #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
103 #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
105 static bool con_flag_valid(unsigned long con_flag
)
108 case CON_FLAG_LOSSYTX
:
109 case CON_FLAG_KEEPALIVE_PENDING
:
110 case CON_FLAG_WRITE_PENDING
:
111 case CON_FLAG_SOCK_CLOSED
:
112 case CON_FLAG_BACKOFF
:
119 static void con_flag_clear(struct ceph_connection
*con
, unsigned long con_flag
)
121 BUG_ON(!con_flag_valid(con_flag
));
123 clear_bit(con_flag
, &con
->flags
);
126 static void con_flag_set(struct ceph_connection
*con
, unsigned long con_flag
)
128 BUG_ON(!con_flag_valid(con_flag
));
130 set_bit(con_flag
, &con
->flags
);
133 static bool con_flag_test(struct ceph_connection
*con
, unsigned long con_flag
)
135 BUG_ON(!con_flag_valid(con_flag
));
137 return test_bit(con_flag
, &con
->flags
);
140 static bool con_flag_test_and_clear(struct ceph_connection
*con
,
141 unsigned long con_flag
)
143 BUG_ON(!con_flag_valid(con_flag
));
145 return test_and_clear_bit(con_flag
, &con
->flags
);
148 static bool con_flag_test_and_set(struct ceph_connection
*con
,
149 unsigned long con_flag
)
151 BUG_ON(!con_flag_valid(con_flag
));
153 return test_and_set_bit(con_flag
, &con
->flags
);
156 /* Slab caches for frequently-allocated structures */
158 static struct kmem_cache
*ceph_msg_cache
;
159 static struct kmem_cache
*ceph_msg_data_cache
;
161 /* static tag bytes (protocol control messages) */
162 static char tag_msg
= CEPH_MSGR_TAG_MSG
;
163 static char tag_ack
= CEPH_MSGR_TAG_ACK
;
164 static char tag_keepalive
= CEPH_MSGR_TAG_KEEPALIVE
;
166 #ifdef CONFIG_LOCKDEP
167 static struct lock_class_key socket_class
;
171 * When skipping (ignoring) a block of input we read it into a "skip
172 * buffer," which is this many bytes in size.
174 #define SKIP_BUF_SIZE 1024
176 static void queue_con(struct ceph_connection
*con
);
177 static void cancel_con(struct ceph_connection
*con
);
178 static void con_work(struct work_struct
*);
179 static void con_fault(struct ceph_connection
*con
);
182 * Nicely render a sockaddr as a string. An array of formatted
183 * strings is used, to approximate reentrancy.
185 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
186 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
187 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
188 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
190 static char addr_str
[ADDR_STR_COUNT
][MAX_ADDR_STR_LEN
];
191 static atomic_t addr_str_seq
= ATOMIC_INIT(0);
193 static struct page
*zero_page
; /* used in certain error cases */
195 const char *ceph_pr_addr(const struct sockaddr_storage
*ss
)
199 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
200 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
202 i
= atomic_inc_return(&addr_str_seq
) & ADDR_STR_COUNT_MASK
;
205 switch (ss
->ss_family
) {
207 snprintf(s
, MAX_ADDR_STR_LEN
, "%pI4:%hu", &in4
->sin_addr
,
208 ntohs(in4
->sin_port
));
212 snprintf(s
, MAX_ADDR_STR_LEN
, "[%pI6c]:%hu", &in6
->sin6_addr
,
213 ntohs(in6
->sin6_port
));
217 snprintf(s
, MAX_ADDR_STR_LEN
, "(unknown sockaddr family %hu)",
223 EXPORT_SYMBOL(ceph_pr_addr
);
225 static void encode_my_addr(struct ceph_messenger
*msgr
)
227 memcpy(&msgr
->my_enc_addr
, &msgr
->inst
.addr
, sizeof(msgr
->my_enc_addr
));
228 ceph_encode_addr(&msgr
->my_enc_addr
);
232 * work queue for all reading and writing to/from the socket.
234 static struct workqueue_struct
*ceph_msgr_wq
;
236 static int ceph_msgr_slab_init(void)
238 BUG_ON(ceph_msg_cache
);
239 ceph_msg_cache
= kmem_cache_create("ceph_msg",
240 sizeof (struct ceph_msg
),
241 __alignof__(struct ceph_msg
), 0, NULL
);
246 BUG_ON(ceph_msg_data_cache
);
247 ceph_msg_data_cache
= kmem_cache_create("ceph_msg_data",
248 sizeof (struct ceph_msg_data
),
249 __alignof__(struct ceph_msg_data
),
251 if (ceph_msg_data_cache
)
254 kmem_cache_destroy(ceph_msg_cache
);
255 ceph_msg_cache
= NULL
;
260 static void ceph_msgr_slab_exit(void)
262 BUG_ON(!ceph_msg_data_cache
);
263 kmem_cache_destroy(ceph_msg_data_cache
);
264 ceph_msg_data_cache
= NULL
;
266 BUG_ON(!ceph_msg_cache
);
267 kmem_cache_destroy(ceph_msg_cache
);
268 ceph_msg_cache
= NULL
;
271 static void _ceph_msgr_exit(void)
274 destroy_workqueue(ceph_msgr_wq
);
278 ceph_msgr_slab_exit();
280 BUG_ON(zero_page
== NULL
);
282 page_cache_release(zero_page
);
286 int ceph_msgr_init(void)
288 BUG_ON(zero_page
!= NULL
);
289 zero_page
= ZERO_PAGE(0);
290 page_cache_get(zero_page
);
292 if (ceph_msgr_slab_init())
296 * The number of active work items is limited by the number of
297 * connections, so leave @max_active at default.
299 ceph_msgr_wq
= alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM
, 0);
303 pr_err("msgr_init failed to create workqueue\n");
308 EXPORT_SYMBOL(ceph_msgr_init
);
310 void ceph_msgr_exit(void)
312 BUG_ON(ceph_msgr_wq
== NULL
);
316 EXPORT_SYMBOL(ceph_msgr_exit
);
318 void ceph_msgr_flush(void)
320 flush_workqueue(ceph_msgr_wq
);
322 EXPORT_SYMBOL(ceph_msgr_flush
);
324 /* Connection socket state transition functions */
326 static void con_sock_state_init(struct ceph_connection
*con
)
330 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSED
);
331 if (WARN_ON(old_state
!= CON_SOCK_STATE_NEW
))
332 printk("%s: unexpected old state %d\n", __func__
, old_state
);
333 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
334 CON_SOCK_STATE_CLOSED
);
337 static void con_sock_state_connecting(struct ceph_connection
*con
)
341 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CONNECTING
);
342 if (WARN_ON(old_state
!= CON_SOCK_STATE_CLOSED
))
343 printk("%s: unexpected old state %d\n", __func__
, old_state
);
344 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
345 CON_SOCK_STATE_CONNECTING
);
348 static void con_sock_state_connected(struct ceph_connection
*con
)
352 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CONNECTED
);
353 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTING
))
354 printk("%s: unexpected old state %d\n", __func__
, old_state
);
355 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
356 CON_SOCK_STATE_CONNECTED
);
359 static void con_sock_state_closing(struct ceph_connection
*con
)
363 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSING
);
364 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTING
&&
365 old_state
!= CON_SOCK_STATE_CONNECTED
&&
366 old_state
!= CON_SOCK_STATE_CLOSING
))
367 printk("%s: unexpected old state %d\n", __func__
, old_state
);
368 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
369 CON_SOCK_STATE_CLOSING
);
372 static void con_sock_state_closed(struct ceph_connection
*con
)
376 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSED
);
377 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTED
&&
378 old_state
!= CON_SOCK_STATE_CLOSING
&&
379 old_state
!= CON_SOCK_STATE_CONNECTING
&&
380 old_state
!= CON_SOCK_STATE_CLOSED
))
381 printk("%s: unexpected old state %d\n", __func__
, old_state
);
382 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
383 CON_SOCK_STATE_CLOSED
);
387 * socket callback functions
390 /* data available on socket, or listen socket received a connect */
391 static void ceph_sock_data_ready(struct sock
*sk
)
393 struct ceph_connection
*con
= sk
->sk_user_data
;
394 if (atomic_read(&con
->msgr
->stopping
)) {
398 if (sk
->sk_state
!= TCP_CLOSE_WAIT
) {
399 dout("%s on %p state = %lu, queueing work\n", __func__
,
405 /* socket has buffer space for writing */
406 static void ceph_sock_write_space(struct sock
*sk
)
408 struct ceph_connection
*con
= sk
->sk_user_data
;
410 /* only queue to workqueue if there is data we want to write,
411 * and there is sufficient space in the socket buffer to accept
412 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
413 * doesn't get called again until try_write() fills the socket
414 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
415 * and net/core/stream.c:sk_stream_write_space().
417 if (con_flag_test(con
, CON_FLAG_WRITE_PENDING
)) {
418 if (sk_stream_is_writeable(sk
)) {
419 dout("%s %p queueing write work\n", __func__
, con
);
420 clear_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
424 dout("%s %p nothing to write\n", __func__
, con
);
428 /* socket's state has changed */
429 static void ceph_sock_state_change(struct sock
*sk
)
431 struct ceph_connection
*con
= sk
->sk_user_data
;
433 dout("%s %p state = %lu sk_state = %u\n", __func__
,
434 con
, con
->state
, sk
->sk_state
);
436 switch (sk
->sk_state
) {
438 dout("%s TCP_CLOSE\n", __func__
);
440 dout("%s TCP_CLOSE_WAIT\n", __func__
);
441 con_sock_state_closing(con
);
442 con_flag_set(con
, CON_FLAG_SOCK_CLOSED
);
445 case TCP_ESTABLISHED
:
446 dout("%s TCP_ESTABLISHED\n", __func__
);
447 con_sock_state_connected(con
);
450 default: /* Everything else is uninteresting */
456 * set up socket callbacks
458 static void set_sock_callbacks(struct socket
*sock
,
459 struct ceph_connection
*con
)
461 struct sock
*sk
= sock
->sk
;
462 sk
->sk_user_data
= con
;
463 sk
->sk_data_ready
= ceph_sock_data_ready
;
464 sk
->sk_write_space
= ceph_sock_write_space
;
465 sk
->sk_state_change
= ceph_sock_state_change
;
474 * initiate connection to a remote socket.
476 static int ceph_tcp_connect(struct ceph_connection
*con
)
478 struct sockaddr_storage
*paddr
= &con
->peer_addr
.in_addr
;
483 ret
= sock_create_kern(con
->peer_addr
.in_addr
.ss_family
, SOCK_STREAM
,
487 sock
->sk
->sk_allocation
= GFP_NOFS
;
489 #ifdef CONFIG_LOCKDEP
490 lockdep_set_class(&sock
->sk
->sk_lock
, &socket_class
);
493 set_sock_callbacks(sock
, con
);
495 dout("connect %s\n", ceph_pr_addr(&con
->peer_addr
.in_addr
));
497 con_sock_state_connecting(con
);
498 ret
= sock
->ops
->connect(sock
, (struct sockaddr
*)paddr
, sizeof(*paddr
),
500 if (ret
== -EINPROGRESS
) {
501 dout("connect %s EINPROGRESS sk_state = %u\n",
502 ceph_pr_addr(&con
->peer_addr
.in_addr
),
504 } else if (ret
< 0) {
505 pr_err("connect %s error %d\n",
506 ceph_pr_addr(&con
->peer_addr
.in_addr
), ret
);
508 con
->error_msg
= "connect error";
513 if (con
->msgr
->tcp_nodelay
) {
516 ret
= kernel_setsockopt(sock
, SOL_TCP
, TCP_NODELAY
,
517 (char *)&optval
, sizeof(optval
));
519 pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
527 static int ceph_tcp_recvmsg(struct socket
*sock
, void *buf
, size_t len
)
529 struct kvec iov
= {buf
, len
};
530 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
533 r
= kernel_recvmsg(sock
, &msg
, &iov
, 1, len
, msg
.msg_flags
);
539 static int ceph_tcp_recvpage(struct socket
*sock
, struct page
*page
,
540 int page_offset
, size_t length
)
545 BUG_ON(page_offset
+ length
> PAGE_SIZE
);
549 ret
= ceph_tcp_recvmsg(sock
, kaddr
+ page_offset
, length
);
556 * write something. @more is true if caller will be sending more data
559 static int ceph_tcp_sendmsg(struct socket
*sock
, struct kvec
*iov
,
560 size_t kvlen
, size_t len
, int more
)
562 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
566 msg
.msg_flags
|= MSG_MORE
;
568 msg
.msg_flags
|= MSG_EOR
; /* superfluous, but what the hell */
570 r
= kernel_sendmsg(sock
, &msg
, iov
, kvlen
, len
);
576 static int __ceph_tcp_sendpage(struct socket
*sock
, struct page
*page
,
577 int offset
, size_t size
, bool more
)
579 int flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
| (more
? MSG_MORE
: MSG_EOR
);
582 ret
= kernel_sendpage(sock
, page
, offset
, size
, flags
);
589 static int ceph_tcp_sendpage(struct socket
*sock
, struct page
*page
,
590 int offset
, size_t size
, bool more
)
595 /* sendpage cannot properly handle pages with page_count == 0,
596 * we need to fallback to sendmsg if that's the case */
597 if (page_count(page
) >= 1)
598 return __ceph_tcp_sendpage(sock
, page
, offset
, size
, more
);
600 iov
.iov_base
= kmap(page
) + offset
;
602 ret
= ceph_tcp_sendmsg(sock
, &iov
, 1, size
, more
);
609 * Shutdown/close the socket for the given connection.
611 static int con_close_socket(struct ceph_connection
*con
)
615 dout("con_close_socket on %p sock %p\n", con
, con
->sock
);
617 rc
= con
->sock
->ops
->shutdown(con
->sock
, SHUT_RDWR
);
618 sock_release(con
->sock
);
623 * Forcibly clear the SOCK_CLOSED flag. It gets set
624 * independent of the connection mutex, and we could have
625 * received a socket close event before we had the chance to
626 * shut the socket down.
628 con_flag_clear(con
, CON_FLAG_SOCK_CLOSED
);
630 con_sock_state_closed(con
);
635 * Reset a connection. Discard all incoming and outgoing messages
636 * and clear *_seq state.
638 static void ceph_msg_remove(struct ceph_msg
*msg
)
640 list_del_init(&msg
->list_head
);
641 BUG_ON(msg
->con
== NULL
);
642 msg
->con
->ops
->put(msg
->con
);
647 static void ceph_msg_remove_list(struct list_head
*head
)
649 while (!list_empty(head
)) {
650 struct ceph_msg
*msg
= list_first_entry(head
, struct ceph_msg
,
652 ceph_msg_remove(msg
);
656 static void reset_connection(struct ceph_connection
*con
)
658 /* reset connection, out_queue, msg_ and connect_seq */
659 /* discard existing out_queue and msg_seq */
660 dout("reset_connection %p\n", con
);
661 ceph_msg_remove_list(&con
->out_queue
);
662 ceph_msg_remove_list(&con
->out_sent
);
665 BUG_ON(con
->in_msg
->con
!= con
);
666 con
->in_msg
->con
= NULL
;
667 ceph_msg_put(con
->in_msg
);
672 con
->connect_seq
= 0;
675 ceph_msg_put(con
->out_msg
);
679 con
->in_seq_acked
= 0;
683 * mark a peer down. drop any open connections.
685 void ceph_con_close(struct ceph_connection
*con
)
687 mutex_lock(&con
->mutex
);
688 dout("con_close %p peer %s\n", con
,
689 ceph_pr_addr(&con
->peer_addr
.in_addr
));
690 con
->state
= CON_STATE_CLOSED
;
692 con_flag_clear(con
, CON_FLAG_LOSSYTX
); /* so we retry next connect */
693 con_flag_clear(con
, CON_FLAG_KEEPALIVE_PENDING
);
694 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
695 con_flag_clear(con
, CON_FLAG_BACKOFF
);
697 reset_connection(con
);
698 con
->peer_global_seq
= 0;
700 con_close_socket(con
);
701 mutex_unlock(&con
->mutex
);
703 EXPORT_SYMBOL(ceph_con_close
);
706 * Reopen a closed connection, with a new peer address.
708 void ceph_con_open(struct ceph_connection
*con
,
709 __u8 entity_type
, __u64 entity_num
,
710 struct ceph_entity_addr
*addr
)
712 mutex_lock(&con
->mutex
);
713 dout("con_open %p %s\n", con
, ceph_pr_addr(&addr
->in_addr
));
715 WARN_ON(con
->state
!= CON_STATE_CLOSED
);
716 con
->state
= CON_STATE_PREOPEN
;
718 con
->peer_name
.type
= (__u8
) entity_type
;
719 con
->peer_name
.num
= cpu_to_le64(entity_num
);
721 memcpy(&con
->peer_addr
, addr
, sizeof(*addr
));
722 con
->delay
= 0; /* reset backoff memory */
723 mutex_unlock(&con
->mutex
);
726 EXPORT_SYMBOL(ceph_con_open
);
729 * return true if this connection ever successfully opened
731 bool ceph_con_opened(struct ceph_connection
*con
)
733 return con
->connect_seq
> 0;
737 * initialize a new connection.
739 void ceph_con_init(struct ceph_connection
*con
, void *private,
740 const struct ceph_connection_operations
*ops
,
741 struct ceph_messenger
*msgr
)
743 dout("con_init %p\n", con
);
744 memset(con
, 0, sizeof(*con
));
745 con
->private = private;
749 con_sock_state_init(con
);
751 mutex_init(&con
->mutex
);
752 INIT_LIST_HEAD(&con
->out_queue
);
753 INIT_LIST_HEAD(&con
->out_sent
);
754 INIT_DELAYED_WORK(&con
->work
, con_work
);
756 con
->state
= CON_STATE_CLOSED
;
758 EXPORT_SYMBOL(ceph_con_init
);
762 * We maintain a global counter to order connection attempts. Get
763 * a unique seq greater than @gt.
765 static u32
get_global_seq(struct ceph_messenger
*msgr
, u32 gt
)
769 spin_lock(&msgr
->global_seq_lock
);
770 if (msgr
->global_seq
< gt
)
771 msgr
->global_seq
= gt
;
772 ret
= ++msgr
->global_seq
;
773 spin_unlock(&msgr
->global_seq_lock
);
777 static void con_out_kvec_reset(struct ceph_connection
*con
)
779 con
->out_kvec_left
= 0;
780 con
->out_kvec_bytes
= 0;
781 con
->out_kvec_cur
= &con
->out_kvec
[0];
784 static void con_out_kvec_add(struct ceph_connection
*con
,
785 size_t size
, void *data
)
789 index
= con
->out_kvec_left
;
790 BUG_ON(index
>= ARRAY_SIZE(con
->out_kvec
));
792 con
->out_kvec
[index
].iov_len
= size
;
793 con
->out_kvec
[index
].iov_base
= data
;
794 con
->out_kvec_left
++;
795 con
->out_kvec_bytes
+= size
;
801 * For a bio data item, a piece is whatever remains of the next
802 * entry in the current bio iovec, or the first entry in the next
805 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor
*cursor
,
808 struct ceph_msg_data
*data
= cursor
->data
;
811 BUG_ON(data
->type
!= CEPH_MSG_DATA_BIO
);
816 cursor
->resid
= min(length
, data
->bio_length
);
818 cursor
->bvec_iter
= bio
->bi_iter
;
820 cursor
->resid
<= bio_iter_len(bio
, cursor
->bvec_iter
);
823 static struct page
*ceph_msg_data_bio_next(struct ceph_msg_data_cursor
*cursor
,
827 struct ceph_msg_data
*data
= cursor
->data
;
829 struct bio_vec bio_vec
;
831 BUG_ON(data
->type
!= CEPH_MSG_DATA_BIO
);
836 bio_vec
= bio_iter_iovec(bio
, cursor
->bvec_iter
);
838 *page_offset
= (size_t) bio_vec
.bv_offset
;
839 BUG_ON(*page_offset
>= PAGE_SIZE
);
840 if (cursor
->last_piece
) /* pagelist offset is always 0 */
841 *length
= cursor
->resid
;
843 *length
= (size_t) bio_vec
.bv_len
;
844 BUG_ON(*length
> cursor
->resid
);
845 BUG_ON(*page_offset
+ *length
> PAGE_SIZE
);
847 return bio_vec
.bv_page
;
850 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor
*cursor
,
854 struct bio_vec bio_vec
;
856 BUG_ON(cursor
->data
->type
!= CEPH_MSG_DATA_BIO
);
861 bio_vec
= bio_iter_iovec(bio
, cursor
->bvec_iter
);
863 /* Advance the cursor offset */
865 BUG_ON(cursor
->resid
< bytes
);
866 cursor
->resid
-= bytes
;
868 bio_advance_iter(bio
, &cursor
->bvec_iter
, bytes
);
870 if (bytes
< bio_vec
.bv_len
)
871 return false; /* more bytes to process in this segment */
873 /* Move on to the next segment, and possibly the next bio */
875 if (!cursor
->bvec_iter
.bi_size
) {
879 cursor
->bvec_iter
= bio
->bi_iter
;
881 memset(&cursor
->bvec_iter
, 0,
882 sizeof(cursor
->bvec_iter
));
885 if (!cursor
->last_piece
) {
886 BUG_ON(!cursor
->resid
);
888 /* A short read is OK, so use <= rather than == */
889 if (cursor
->resid
<= bio_iter_len(bio
, cursor
->bvec_iter
))
890 cursor
->last_piece
= true;
895 #endif /* CONFIG_BLOCK */
898 * For a page array, a piece comes from the first page in the array
899 * that has not already been fully consumed.
901 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor
*cursor
,
904 struct ceph_msg_data
*data
= cursor
->data
;
907 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGES
);
909 BUG_ON(!data
->pages
);
910 BUG_ON(!data
->length
);
912 cursor
->resid
= min(length
, data
->length
);
913 page_count
= calc_pages_for(data
->alignment
, (u64
)data
->length
);
914 cursor
->page_offset
= data
->alignment
& ~PAGE_MASK
;
915 cursor
->page_index
= 0;
916 BUG_ON(page_count
> (int)USHRT_MAX
);
917 cursor
->page_count
= (unsigned short)page_count
;
918 BUG_ON(length
> SIZE_MAX
- cursor
->page_offset
);
919 cursor
->last_piece
= cursor
->page_offset
+ cursor
->resid
<= PAGE_SIZE
;
923 ceph_msg_data_pages_next(struct ceph_msg_data_cursor
*cursor
,
924 size_t *page_offset
, size_t *length
)
926 struct ceph_msg_data
*data
= cursor
->data
;
928 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGES
);
930 BUG_ON(cursor
->page_index
>= cursor
->page_count
);
931 BUG_ON(cursor
->page_offset
>= PAGE_SIZE
);
933 *page_offset
= cursor
->page_offset
;
934 if (cursor
->last_piece
)
935 *length
= cursor
->resid
;
937 *length
= PAGE_SIZE
- *page_offset
;
939 return data
->pages
[cursor
->page_index
];
942 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor
*cursor
,
945 BUG_ON(cursor
->data
->type
!= CEPH_MSG_DATA_PAGES
);
947 BUG_ON(cursor
->page_offset
+ bytes
> PAGE_SIZE
);
949 /* Advance the cursor page offset */
951 cursor
->resid
-= bytes
;
952 cursor
->page_offset
= (cursor
->page_offset
+ bytes
) & ~PAGE_MASK
;
953 if (!bytes
|| cursor
->page_offset
)
954 return false; /* more bytes to process in the current page */
957 return false; /* no more data */
959 /* Move on to the next page; offset is already at 0 */
961 BUG_ON(cursor
->page_index
>= cursor
->page_count
);
962 cursor
->page_index
++;
963 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
969 * For a pagelist, a piece is whatever remains to be consumed in the
970 * first page in the list, or the front of the next page.
973 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor
*cursor
,
976 struct ceph_msg_data
*data
= cursor
->data
;
977 struct ceph_pagelist
*pagelist
;
980 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
982 pagelist
= data
->pagelist
;
986 return; /* pagelist can be assigned but empty */
988 BUG_ON(list_empty(&pagelist
->head
));
989 page
= list_first_entry(&pagelist
->head
, struct page
, lru
);
991 cursor
->resid
= min(length
, pagelist
->length
);
994 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
998 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor
*cursor
,
999 size_t *page_offset
, size_t *length
)
1001 struct ceph_msg_data
*data
= cursor
->data
;
1002 struct ceph_pagelist
*pagelist
;
1004 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
1006 pagelist
= data
->pagelist
;
1009 BUG_ON(!cursor
->page
);
1010 BUG_ON(cursor
->offset
+ cursor
->resid
!= pagelist
->length
);
1012 /* offset of first page in pagelist is always 0 */
1013 *page_offset
= cursor
->offset
& ~PAGE_MASK
;
1014 if (cursor
->last_piece
)
1015 *length
= cursor
->resid
;
1017 *length
= PAGE_SIZE
- *page_offset
;
1019 return cursor
->page
;
1022 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor
*cursor
,
1025 struct ceph_msg_data
*data
= cursor
->data
;
1026 struct ceph_pagelist
*pagelist
;
1028 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
1030 pagelist
= data
->pagelist
;
1033 BUG_ON(cursor
->offset
+ cursor
->resid
!= pagelist
->length
);
1034 BUG_ON((cursor
->offset
& ~PAGE_MASK
) + bytes
> PAGE_SIZE
);
1036 /* Advance the cursor offset */
1038 cursor
->resid
-= bytes
;
1039 cursor
->offset
+= bytes
;
1040 /* offset of first page in pagelist is always 0 */
1041 if (!bytes
|| cursor
->offset
& ~PAGE_MASK
)
1042 return false; /* more bytes to process in the current page */
1045 return false; /* no more data */
1047 /* Move on to the next page */
1049 BUG_ON(list_is_last(&cursor
->page
->lru
, &pagelist
->head
));
1050 cursor
->page
= list_entry_next(cursor
->page
, lru
);
1051 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
1057 * Message data is handled (sent or received) in pieces, where each
1058 * piece resides on a single page. The network layer might not
1059 * consume an entire piece at once. A data item's cursor keeps
1060 * track of which piece is next to process and how much remains to
1061 * be processed in that piece. It also tracks whether the current
1062 * piece is the last one in the data item.
1064 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor
*cursor
)
1066 size_t length
= cursor
->total_resid
;
1068 switch (cursor
->data
->type
) {
1069 case CEPH_MSG_DATA_PAGELIST
:
1070 ceph_msg_data_pagelist_cursor_init(cursor
, length
);
1072 case CEPH_MSG_DATA_PAGES
:
1073 ceph_msg_data_pages_cursor_init(cursor
, length
);
1076 case CEPH_MSG_DATA_BIO
:
1077 ceph_msg_data_bio_cursor_init(cursor
, length
);
1079 #endif /* CONFIG_BLOCK */
1080 case CEPH_MSG_DATA_NONE
:
1085 cursor
->need_crc
= true;
1088 static void ceph_msg_data_cursor_init(struct ceph_msg
*msg
, size_t length
)
1090 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
1091 struct ceph_msg_data
*data
;
1094 BUG_ON(length
> msg
->data_length
);
1095 BUG_ON(list_empty(&msg
->data
));
1097 cursor
->data_head
= &msg
->data
;
1098 cursor
->total_resid
= length
;
1099 data
= list_first_entry(&msg
->data
, struct ceph_msg_data
, links
);
1100 cursor
->data
= data
;
1102 __ceph_msg_data_cursor_init(cursor
);
1106 * Return the page containing the next piece to process for a given
1107 * data item, and supply the page offset and length of that piece.
1108 * Indicate whether this is the last piece in this data item.
1110 static struct page
*ceph_msg_data_next(struct ceph_msg_data_cursor
*cursor
,
1111 size_t *page_offset
, size_t *length
,
1116 switch (cursor
->data
->type
) {
1117 case CEPH_MSG_DATA_PAGELIST
:
1118 page
= ceph_msg_data_pagelist_next(cursor
, page_offset
, length
);
1120 case CEPH_MSG_DATA_PAGES
:
1121 page
= ceph_msg_data_pages_next(cursor
, page_offset
, length
);
1124 case CEPH_MSG_DATA_BIO
:
1125 page
= ceph_msg_data_bio_next(cursor
, page_offset
, length
);
1127 #endif /* CONFIG_BLOCK */
1128 case CEPH_MSG_DATA_NONE
:
1134 BUG_ON(*page_offset
+ *length
> PAGE_SIZE
);
1137 *last_piece
= cursor
->last_piece
;
1143 * Returns true if the result moves the cursor on to the next piece
1146 static bool ceph_msg_data_advance(struct ceph_msg_data_cursor
*cursor
,
1151 BUG_ON(bytes
> cursor
->resid
);
1152 switch (cursor
->data
->type
) {
1153 case CEPH_MSG_DATA_PAGELIST
:
1154 new_piece
= ceph_msg_data_pagelist_advance(cursor
, bytes
);
1156 case CEPH_MSG_DATA_PAGES
:
1157 new_piece
= ceph_msg_data_pages_advance(cursor
, bytes
);
1160 case CEPH_MSG_DATA_BIO
:
1161 new_piece
= ceph_msg_data_bio_advance(cursor
, bytes
);
1163 #endif /* CONFIG_BLOCK */
1164 case CEPH_MSG_DATA_NONE
:
1169 cursor
->total_resid
-= bytes
;
1171 if (!cursor
->resid
&& cursor
->total_resid
) {
1172 WARN_ON(!cursor
->last_piece
);
1173 BUG_ON(list_is_last(&cursor
->data
->links
, cursor
->data_head
));
1174 cursor
->data
= list_entry_next(cursor
->data
, links
);
1175 __ceph_msg_data_cursor_init(cursor
);
1178 cursor
->need_crc
= new_piece
;
1183 static void prepare_message_data(struct ceph_msg
*msg
, u32 data_len
)
1188 /* Initialize data cursor */
1190 ceph_msg_data_cursor_init(msg
, (size_t)data_len
);
1194 * Prepare footer for currently outgoing message, and finish things
1195 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1197 static void prepare_write_message_footer(struct ceph_connection
*con
)
1199 struct ceph_msg
*m
= con
->out_msg
;
1200 int v
= con
->out_kvec_left
;
1202 m
->footer
.flags
|= CEPH_MSG_FOOTER_COMPLETE
;
1204 dout("prepare_write_message_footer %p\n", con
);
1205 con
->out_kvec_is_msg
= true;
1206 con
->out_kvec
[v
].iov_base
= &m
->footer
;
1207 if (con
->peer_features
& CEPH_FEATURE_MSG_AUTH
) {
1208 if (con
->ops
->sign_message
)
1209 con
->ops
->sign_message(con
, m
);
1212 con
->out_kvec
[v
].iov_len
= sizeof(m
->footer
);
1213 con
->out_kvec_bytes
+= sizeof(m
->footer
);
1215 m
->old_footer
.flags
= m
->footer
.flags
;
1216 con
->out_kvec
[v
].iov_len
= sizeof(m
->old_footer
);
1217 con
->out_kvec_bytes
+= sizeof(m
->old_footer
);
1219 con
->out_kvec_left
++;
1220 con
->out_more
= m
->more_to_follow
;
1221 con
->out_msg_done
= true;
1225 * Prepare headers for the next outgoing message.
1227 static void prepare_write_message(struct ceph_connection
*con
)
1232 con_out_kvec_reset(con
);
1233 con
->out_kvec_is_msg
= true;
1234 con
->out_msg_done
= false;
1236 /* Sneak an ack in there first? If we can get it into the same
1237 * TCP packet that's a good thing. */
1238 if (con
->in_seq
> con
->in_seq_acked
) {
1239 con
->in_seq_acked
= con
->in_seq
;
1240 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
1241 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1242 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1243 &con
->out_temp_ack
);
1246 BUG_ON(list_empty(&con
->out_queue
));
1247 m
= list_first_entry(&con
->out_queue
, struct ceph_msg
, list_head
);
1249 BUG_ON(m
->con
!= con
);
1251 /* put message on sent list */
1253 list_move_tail(&m
->list_head
, &con
->out_sent
);
1256 * only assign outgoing seq # if we haven't sent this message
1257 * yet. if it is requeued, resend with it's original seq.
1259 if (m
->needs_out_seq
) {
1260 m
->hdr
.seq
= cpu_to_le64(++con
->out_seq
);
1261 m
->needs_out_seq
= false;
1263 WARN_ON(m
->data_length
!= le32_to_cpu(m
->hdr
.data_len
));
1265 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1266 m
, con
->out_seq
, le16_to_cpu(m
->hdr
.type
),
1267 le32_to_cpu(m
->hdr
.front_len
), le32_to_cpu(m
->hdr
.middle_len
),
1269 BUG_ON(le32_to_cpu(m
->hdr
.front_len
) != m
->front
.iov_len
);
1271 /* tag + hdr + front + middle */
1272 con_out_kvec_add(con
, sizeof (tag_msg
), &tag_msg
);
1273 con_out_kvec_add(con
, sizeof (m
->hdr
), &m
->hdr
);
1274 con_out_kvec_add(con
, m
->front
.iov_len
, m
->front
.iov_base
);
1277 con_out_kvec_add(con
, m
->middle
->vec
.iov_len
,
1278 m
->middle
->vec
.iov_base
);
1280 /* fill in crc (except data pages), footer */
1281 crc
= crc32c(0, &m
->hdr
, offsetof(struct ceph_msg_header
, crc
));
1282 con
->out_msg
->hdr
.crc
= cpu_to_le32(crc
);
1283 con
->out_msg
->footer
.flags
= 0;
1285 crc
= crc32c(0, m
->front
.iov_base
, m
->front
.iov_len
);
1286 con
->out_msg
->footer
.front_crc
= cpu_to_le32(crc
);
1288 crc
= crc32c(0, m
->middle
->vec
.iov_base
,
1289 m
->middle
->vec
.iov_len
);
1290 con
->out_msg
->footer
.middle_crc
= cpu_to_le32(crc
);
1292 con
->out_msg
->footer
.middle_crc
= 0;
1293 dout("%s front_crc %u middle_crc %u\n", __func__
,
1294 le32_to_cpu(con
->out_msg
->footer
.front_crc
),
1295 le32_to_cpu(con
->out_msg
->footer
.middle_crc
));
1297 /* is there a data payload? */
1298 con
->out_msg
->footer
.data_crc
= 0;
1299 if (m
->data_length
) {
1300 prepare_message_data(con
->out_msg
, m
->data_length
);
1301 con
->out_more
= 1; /* data + footer will follow */
1303 /* no, queue up footer too and be done */
1304 prepare_write_message_footer(con
);
1307 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1313 static void prepare_write_ack(struct ceph_connection
*con
)
1315 dout("prepare_write_ack %p %llu -> %llu\n", con
,
1316 con
->in_seq_acked
, con
->in_seq
);
1317 con
->in_seq_acked
= con
->in_seq
;
1319 con_out_kvec_reset(con
);
1321 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
1323 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1324 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1325 &con
->out_temp_ack
);
1327 con
->out_more
= 1; /* more will follow.. eventually.. */
1328 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1332 * Prepare to share the seq during handshake
1334 static void prepare_write_seq(struct ceph_connection
*con
)
1336 dout("prepare_write_seq %p %llu -> %llu\n", con
,
1337 con
->in_seq_acked
, con
->in_seq
);
1338 con
->in_seq_acked
= con
->in_seq
;
1340 con_out_kvec_reset(con
);
1342 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1343 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1344 &con
->out_temp_ack
);
1346 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1350 * Prepare to write keepalive byte.
1352 static void prepare_write_keepalive(struct ceph_connection
*con
)
1354 dout("prepare_write_keepalive %p\n", con
);
1355 con_out_kvec_reset(con
);
1356 con_out_kvec_add(con
, sizeof (tag_keepalive
), &tag_keepalive
);
1357 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1361 * Connection negotiation.
1364 static struct ceph_auth_handshake
*get_connect_authorizer(struct ceph_connection
*con
,
1367 struct ceph_auth_handshake
*auth
;
1369 if (!con
->ops
->get_authorizer
) {
1370 con
->out_connect
.authorizer_protocol
= CEPH_AUTH_UNKNOWN
;
1371 con
->out_connect
.authorizer_len
= 0;
1375 /* Can't hold the mutex while getting authorizer */
1376 mutex_unlock(&con
->mutex
);
1377 auth
= con
->ops
->get_authorizer(con
, auth_proto
, con
->auth_retry
);
1378 mutex_lock(&con
->mutex
);
1382 if (con
->state
!= CON_STATE_NEGOTIATING
)
1383 return ERR_PTR(-EAGAIN
);
1385 con
->auth_reply_buf
= auth
->authorizer_reply_buf
;
1386 con
->auth_reply_buf_len
= auth
->authorizer_reply_buf_len
;
1391 * We connected to a peer and are saying hello.
1393 static void prepare_write_banner(struct ceph_connection
*con
)
1395 con_out_kvec_add(con
, strlen(CEPH_BANNER
), CEPH_BANNER
);
1396 con_out_kvec_add(con
, sizeof (con
->msgr
->my_enc_addr
),
1397 &con
->msgr
->my_enc_addr
);
1400 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1403 static int prepare_write_connect(struct ceph_connection
*con
)
1405 unsigned int global_seq
= get_global_seq(con
->msgr
, 0);
1408 struct ceph_auth_handshake
*auth
;
1410 switch (con
->peer_name
.type
) {
1411 case CEPH_ENTITY_TYPE_MON
:
1412 proto
= CEPH_MONC_PROTOCOL
;
1414 case CEPH_ENTITY_TYPE_OSD
:
1415 proto
= CEPH_OSDC_PROTOCOL
;
1417 case CEPH_ENTITY_TYPE_MDS
:
1418 proto
= CEPH_MDSC_PROTOCOL
;
1424 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con
,
1425 con
->connect_seq
, global_seq
, proto
);
1427 con
->out_connect
.features
= cpu_to_le64(con
->msgr
->supported_features
);
1428 con
->out_connect
.host_type
= cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT
);
1429 con
->out_connect
.connect_seq
= cpu_to_le32(con
->connect_seq
);
1430 con
->out_connect
.global_seq
= cpu_to_le32(global_seq
);
1431 con
->out_connect
.protocol_version
= cpu_to_le32(proto
);
1432 con
->out_connect
.flags
= 0;
1434 auth_proto
= CEPH_AUTH_UNKNOWN
;
1435 auth
= get_connect_authorizer(con
, &auth_proto
);
1437 return PTR_ERR(auth
);
1439 con
->out_connect
.authorizer_protocol
= cpu_to_le32(auth_proto
);
1440 con
->out_connect
.authorizer_len
= auth
?
1441 cpu_to_le32(auth
->authorizer_buf_len
) : 0;
1443 con_out_kvec_add(con
, sizeof (con
->out_connect
),
1445 if (auth
&& auth
->authorizer_buf_len
)
1446 con_out_kvec_add(con
, auth
->authorizer_buf_len
,
1447 auth
->authorizer_buf
);
1450 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1456 * write as much of pending kvecs to the socket as we can.
1458 * 0 -> socket full, but more to do
1461 static int write_partial_kvec(struct ceph_connection
*con
)
1465 dout("write_partial_kvec %p %d left\n", con
, con
->out_kvec_bytes
);
1466 while (con
->out_kvec_bytes
> 0) {
1467 ret
= ceph_tcp_sendmsg(con
->sock
, con
->out_kvec_cur
,
1468 con
->out_kvec_left
, con
->out_kvec_bytes
,
1472 con
->out_kvec_bytes
-= ret
;
1473 if (con
->out_kvec_bytes
== 0)
1476 /* account for full iov entries consumed */
1477 while (ret
>= con
->out_kvec_cur
->iov_len
) {
1478 BUG_ON(!con
->out_kvec_left
);
1479 ret
-= con
->out_kvec_cur
->iov_len
;
1480 con
->out_kvec_cur
++;
1481 con
->out_kvec_left
--;
1483 /* and for a partially-consumed entry */
1485 con
->out_kvec_cur
->iov_len
-= ret
;
1486 con
->out_kvec_cur
->iov_base
+= ret
;
1489 con
->out_kvec_left
= 0;
1490 con
->out_kvec_is_msg
= false;
1493 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con
,
1494 con
->out_kvec_bytes
, con
->out_kvec_left
, ret
);
1495 return ret
; /* done! */
1498 static u32
ceph_crc32c_page(u32 crc
, struct page
*page
,
1499 unsigned int page_offset
,
1500 unsigned int length
)
1505 BUG_ON(kaddr
== NULL
);
1506 crc
= crc32c(crc
, kaddr
+ page_offset
, length
);
1512 * Write as much message data payload as we can. If we finish, queue
1514 * 1 -> done, footer is now queued in out_kvec[].
1515 * 0 -> socket full, but more to do
1518 static int write_partial_message_data(struct ceph_connection
*con
)
1520 struct ceph_msg
*msg
= con
->out_msg
;
1521 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
1522 bool do_datacrc
= !con
->msgr
->nocrc
;
1525 dout("%s %p msg %p\n", __func__
, con
, msg
);
1527 if (list_empty(&msg
->data
))
1531 * Iterate through each page that contains data to be
1532 * written, and send as much as possible for each.
1534 * If we are calculating the data crc (the default), we will
1535 * need to map the page. If we have no pages, they have
1536 * been revoked, so use the zero page.
1538 crc
= do_datacrc
? le32_to_cpu(msg
->footer
.data_crc
) : 0;
1539 while (cursor
->resid
) {
1547 page
= ceph_msg_data_next(&msg
->cursor
, &page_offset
, &length
,
1549 ret
= ceph_tcp_sendpage(con
->sock
, page
, page_offset
,
1550 length
, last_piece
);
1553 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1557 if (do_datacrc
&& cursor
->need_crc
)
1558 crc
= ceph_crc32c_page(crc
, page
, page_offset
, length
);
1559 need_crc
= ceph_msg_data_advance(&msg
->cursor
, (size_t)ret
);
1562 dout("%s %p msg %p done\n", __func__
, con
, msg
);
1564 /* prepare and queue up footer, too */
1566 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1568 msg
->footer
.flags
|= CEPH_MSG_FOOTER_NOCRC
;
1569 con_out_kvec_reset(con
);
1570 prepare_write_message_footer(con
);
1572 return 1; /* must return > 0 to indicate success */
1578 static int write_partial_skip(struct ceph_connection
*con
)
1582 while (con
->out_skip
> 0) {
1583 size_t size
= min(con
->out_skip
, (int) PAGE_CACHE_SIZE
);
1585 ret
= ceph_tcp_sendpage(con
->sock
, zero_page
, 0, size
, true);
1588 con
->out_skip
-= ret
;
1596 * Prepare to read connection handshake, or an ack.
1598 static void prepare_read_banner(struct ceph_connection
*con
)
1600 dout("prepare_read_banner %p\n", con
);
1601 con
->in_base_pos
= 0;
1604 static void prepare_read_connect(struct ceph_connection
*con
)
1606 dout("prepare_read_connect %p\n", con
);
1607 con
->in_base_pos
= 0;
1610 static void prepare_read_ack(struct ceph_connection
*con
)
1612 dout("prepare_read_ack %p\n", con
);
1613 con
->in_base_pos
= 0;
1616 static void prepare_read_seq(struct ceph_connection
*con
)
1618 dout("prepare_read_seq %p\n", con
);
1619 con
->in_base_pos
= 0;
1620 con
->in_tag
= CEPH_MSGR_TAG_SEQ
;
1623 static void prepare_read_tag(struct ceph_connection
*con
)
1625 dout("prepare_read_tag %p\n", con
);
1626 con
->in_base_pos
= 0;
1627 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1631 * Prepare to read a message.
1633 static int prepare_read_message(struct ceph_connection
*con
)
1635 dout("prepare_read_message %p\n", con
);
1636 BUG_ON(con
->in_msg
!= NULL
);
1637 con
->in_base_pos
= 0;
1638 con
->in_front_crc
= con
->in_middle_crc
= con
->in_data_crc
= 0;
1643 static int read_partial(struct ceph_connection
*con
,
1644 int end
, int size
, void *object
)
1646 while (con
->in_base_pos
< end
) {
1647 int left
= end
- con
->in_base_pos
;
1648 int have
= size
- left
;
1649 int ret
= ceph_tcp_recvmsg(con
->sock
, object
+ have
, left
);
1652 con
->in_base_pos
+= ret
;
1659 * Read all or part of the connect-side handshake on a new connection
1661 static int read_partial_banner(struct ceph_connection
*con
)
1667 dout("read_partial_banner %p at %d\n", con
, con
->in_base_pos
);
1670 size
= strlen(CEPH_BANNER
);
1672 ret
= read_partial(con
, end
, size
, con
->in_banner
);
1676 size
= sizeof (con
->actual_peer_addr
);
1678 ret
= read_partial(con
, end
, size
, &con
->actual_peer_addr
);
1682 size
= sizeof (con
->peer_addr_for_me
);
1684 ret
= read_partial(con
, end
, size
, &con
->peer_addr_for_me
);
1692 static int read_partial_connect(struct ceph_connection
*con
)
1698 dout("read_partial_connect %p at %d\n", con
, con
->in_base_pos
);
1700 size
= sizeof (con
->in_reply
);
1702 ret
= read_partial(con
, end
, size
, &con
->in_reply
);
1706 size
= le32_to_cpu(con
->in_reply
.authorizer_len
);
1708 ret
= read_partial(con
, end
, size
, con
->auth_reply_buf
);
1712 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1713 con
, (int)con
->in_reply
.tag
,
1714 le32_to_cpu(con
->in_reply
.connect_seq
),
1715 le32_to_cpu(con
->in_reply
.global_seq
));
1722 * Verify the hello banner looks okay.
1724 static int verify_hello(struct ceph_connection
*con
)
1726 if (memcmp(con
->in_banner
, CEPH_BANNER
, strlen(CEPH_BANNER
))) {
1727 pr_err("connect to %s got bad banner\n",
1728 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1729 con
->error_msg
= "protocol error, bad banner";
1735 static bool addr_is_blank(struct sockaddr_storage
*ss
)
1737 switch (ss
->ss_family
) {
1739 return ((struct sockaddr_in
*)ss
)->sin_addr
.s_addr
== 0;
1742 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[0] == 0 &&
1743 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[1] == 0 &&
1744 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[2] == 0 &&
1745 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[3] == 0;
1750 static int addr_port(struct sockaddr_storage
*ss
)
1752 switch (ss
->ss_family
) {
1754 return ntohs(((struct sockaddr_in
*)ss
)->sin_port
);
1756 return ntohs(((struct sockaddr_in6
*)ss
)->sin6_port
);
1761 static void addr_set_port(struct sockaddr_storage
*ss
, int p
)
1763 switch (ss
->ss_family
) {
1765 ((struct sockaddr_in
*)ss
)->sin_port
= htons(p
);
1768 ((struct sockaddr_in6
*)ss
)->sin6_port
= htons(p
);
1774 * Unlike other *_pton function semantics, zero indicates success.
1776 static int ceph_pton(const char *str
, size_t len
, struct sockaddr_storage
*ss
,
1777 char delim
, const char **ipend
)
1779 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
1780 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
1782 memset(ss
, 0, sizeof(*ss
));
1784 if (in4_pton(str
, len
, (u8
*)&in4
->sin_addr
.s_addr
, delim
, ipend
)) {
1785 ss
->ss_family
= AF_INET
;
1789 if (in6_pton(str
, len
, (u8
*)&in6
->sin6_addr
.s6_addr
, delim
, ipend
)) {
1790 ss
->ss_family
= AF_INET6
;
1798 * Extract hostname string and resolve using kernel DNS facility.
1800 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1801 static int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1802 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1804 const char *end
, *delim_p
;
1805 char *colon_p
, *ip_addr
= NULL
;
1809 * The end of the hostname occurs immediately preceding the delimiter or
1810 * the port marker (':') where the delimiter takes precedence.
1812 delim_p
= memchr(name
, delim
, namelen
);
1813 colon_p
= memchr(name
, ':', namelen
);
1815 if (delim_p
&& colon_p
)
1816 end
= delim_p
< colon_p
? delim_p
: colon_p
;
1817 else if (!delim_p
&& colon_p
)
1821 if (!end
) /* case: hostname:/ */
1822 end
= name
+ namelen
;
1828 /* do dns_resolve upcall */
1829 ip_len
= dns_query(NULL
, name
, end
- name
, NULL
, &ip_addr
, NULL
);
1831 ret
= ceph_pton(ip_addr
, ip_len
, ss
, -1, NULL
);
1839 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end
- name
), name
,
1840 ret
, ret
? "failed" : ceph_pr_addr(ss
));
1845 static inline int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1846 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1853 * Parse a server name (IP or hostname). If a valid IP address is not found
1854 * then try to extract a hostname to resolve using userspace DNS upcall.
1856 static int ceph_parse_server_name(const char *name
, size_t namelen
,
1857 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1861 ret
= ceph_pton(name
, namelen
, ss
, delim
, ipend
);
1863 ret
= ceph_dns_resolve_name(name
, namelen
, ss
, delim
, ipend
);
1869 * Parse an ip[:port] list into an addr array. Use the default
1870 * monitor port if a port isn't specified.
1872 int ceph_parse_ips(const char *c
, const char *end
,
1873 struct ceph_entity_addr
*addr
,
1874 int max_count
, int *count
)
1876 int i
, ret
= -EINVAL
;
1879 dout("parse_ips on '%.*s'\n", (int)(end
-c
), c
);
1880 for (i
= 0; i
< max_count
; i
++) {
1882 struct sockaddr_storage
*ss
= &addr
[i
].in_addr
;
1891 ret
= ceph_parse_server_name(p
, end
- p
, ss
, delim
, &ipend
);
1900 dout("missing matching ']'\n");
1907 if (p
< end
&& *p
== ':') {
1910 while (p
< end
&& *p
>= '0' && *p
<= '9') {
1911 port
= (port
* 10) + (*p
- '0');
1915 port
= CEPH_MON_PORT
;
1916 else if (port
> 65535)
1919 port
= CEPH_MON_PORT
;
1922 addr_set_port(ss
, port
);
1924 dout("parse_ips got %s\n", ceph_pr_addr(ss
));
1941 pr_err("parse_ips bad ip '%.*s'\n", (int)(end
- c
), c
);
1944 EXPORT_SYMBOL(ceph_parse_ips
);
1946 static int process_banner(struct ceph_connection
*con
)
1948 dout("process_banner on %p\n", con
);
1950 if (verify_hello(con
) < 0)
1953 ceph_decode_addr(&con
->actual_peer_addr
);
1954 ceph_decode_addr(&con
->peer_addr_for_me
);
1957 * Make sure the other end is who we wanted. note that the other
1958 * end may not yet know their ip address, so if it's 0.0.0.0, give
1959 * them the benefit of the doubt.
1961 if (memcmp(&con
->peer_addr
, &con
->actual_peer_addr
,
1962 sizeof(con
->peer_addr
)) != 0 &&
1963 !(addr_is_blank(&con
->actual_peer_addr
.in_addr
) &&
1964 con
->actual_peer_addr
.nonce
== con
->peer_addr
.nonce
)) {
1965 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
1966 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1967 (int)le32_to_cpu(con
->peer_addr
.nonce
),
1968 ceph_pr_addr(&con
->actual_peer_addr
.in_addr
),
1969 (int)le32_to_cpu(con
->actual_peer_addr
.nonce
));
1970 con
->error_msg
= "wrong peer at address";
1975 * did we learn our address?
1977 if (addr_is_blank(&con
->msgr
->inst
.addr
.in_addr
)) {
1978 int port
= addr_port(&con
->msgr
->inst
.addr
.in_addr
);
1980 memcpy(&con
->msgr
->inst
.addr
.in_addr
,
1981 &con
->peer_addr_for_me
.in_addr
,
1982 sizeof(con
->peer_addr_for_me
.in_addr
));
1983 addr_set_port(&con
->msgr
->inst
.addr
.in_addr
, port
);
1984 encode_my_addr(con
->msgr
);
1985 dout("process_banner learned my addr is %s\n",
1986 ceph_pr_addr(&con
->msgr
->inst
.addr
.in_addr
));
1992 static int process_connect(struct ceph_connection
*con
)
1994 u64 sup_feat
= con
->msgr
->supported_features
;
1995 u64 req_feat
= con
->msgr
->required_features
;
1996 u64 server_feat
= ceph_sanitize_features(
1997 le64_to_cpu(con
->in_reply
.features
));
2000 dout("process_connect on %p tag %d\n", con
, (int)con
->in_tag
);
2002 switch (con
->in_reply
.tag
) {
2003 case CEPH_MSGR_TAG_FEATURES
:
2004 pr_err("%s%lld %s feature set mismatch,"
2005 " my %llx < server's %llx, missing %llx\n",
2006 ENTITY_NAME(con
->peer_name
),
2007 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2008 sup_feat
, server_feat
, server_feat
& ~sup_feat
);
2009 con
->error_msg
= "missing required protocol features";
2010 reset_connection(con
);
2013 case CEPH_MSGR_TAG_BADPROTOVER
:
2014 pr_err("%s%lld %s protocol version mismatch,"
2015 " my %d != server's %d\n",
2016 ENTITY_NAME(con
->peer_name
),
2017 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2018 le32_to_cpu(con
->out_connect
.protocol_version
),
2019 le32_to_cpu(con
->in_reply
.protocol_version
));
2020 con
->error_msg
= "protocol version mismatch";
2021 reset_connection(con
);
2024 case CEPH_MSGR_TAG_BADAUTHORIZER
:
2026 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con
,
2028 if (con
->auth_retry
== 2) {
2029 con
->error_msg
= "connect authorization failure";
2032 con_out_kvec_reset(con
);
2033 ret
= prepare_write_connect(con
);
2036 prepare_read_connect(con
);
2039 case CEPH_MSGR_TAG_RESETSESSION
:
2041 * If we connected with a large connect_seq but the peer
2042 * has no record of a session with us (no connection, or
2043 * connect_seq == 0), they will send RESETSESION to indicate
2044 * that they must have reset their session, and may have
2047 dout("process_connect got RESET peer seq %u\n",
2048 le32_to_cpu(con
->in_reply
.connect_seq
));
2049 pr_err("%s%lld %s connection reset\n",
2050 ENTITY_NAME(con
->peer_name
),
2051 ceph_pr_addr(&con
->peer_addr
.in_addr
));
2052 reset_connection(con
);
2053 con_out_kvec_reset(con
);
2054 ret
= prepare_write_connect(con
);
2057 prepare_read_connect(con
);
2059 /* Tell ceph about it. */
2060 mutex_unlock(&con
->mutex
);
2061 pr_info("reset on %s%lld\n", ENTITY_NAME(con
->peer_name
));
2062 if (con
->ops
->peer_reset
)
2063 con
->ops
->peer_reset(con
);
2064 mutex_lock(&con
->mutex
);
2065 if (con
->state
!= CON_STATE_NEGOTIATING
)
2069 case CEPH_MSGR_TAG_RETRY_SESSION
:
2071 * If we sent a smaller connect_seq than the peer has, try
2072 * again with a larger value.
2074 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2075 le32_to_cpu(con
->out_connect
.connect_seq
),
2076 le32_to_cpu(con
->in_reply
.connect_seq
));
2077 con
->connect_seq
= le32_to_cpu(con
->in_reply
.connect_seq
);
2078 con_out_kvec_reset(con
);
2079 ret
= prepare_write_connect(con
);
2082 prepare_read_connect(con
);
2085 case CEPH_MSGR_TAG_RETRY_GLOBAL
:
2087 * If we sent a smaller global_seq than the peer has, try
2088 * again with a larger value.
2090 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2091 con
->peer_global_seq
,
2092 le32_to_cpu(con
->in_reply
.global_seq
));
2093 get_global_seq(con
->msgr
,
2094 le32_to_cpu(con
->in_reply
.global_seq
));
2095 con_out_kvec_reset(con
);
2096 ret
= prepare_write_connect(con
);
2099 prepare_read_connect(con
);
2102 case CEPH_MSGR_TAG_SEQ
:
2103 case CEPH_MSGR_TAG_READY
:
2104 if (req_feat
& ~server_feat
) {
2105 pr_err("%s%lld %s protocol feature mismatch,"
2106 " my required %llx > server's %llx, need %llx\n",
2107 ENTITY_NAME(con
->peer_name
),
2108 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2109 req_feat
, server_feat
, req_feat
& ~server_feat
);
2110 con
->error_msg
= "missing required protocol features";
2111 reset_connection(con
);
2115 WARN_ON(con
->state
!= CON_STATE_NEGOTIATING
);
2116 con
->state
= CON_STATE_OPEN
;
2117 con
->auth_retry
= 0; /* we authenticated; clear flag */
2118 con
->peer_global_seq
= le32_to_cpu(con
->in_reply
.global_seq
);
2120 con
->peer_features
= server_feat
;
2121 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2122 con
->peer_global_seq
,
2123 le32_to_cpu(con
->in_reply
.connect_seq
),
2125 WARN_ON(con
->connect_seq
!=
2126 le32_to_cpu(con
->in_reply
.connect_seq
));
2128 if (con
->in_reply
.flags
& CEPH_MSG_CONNECT_LOSSY
)
2129 con_flag_set(con
, CON_FLAG_LOSSYTX
);
2131 con
->delay
= 0; /* reset backoff memory */
2133 if (con
->in_reply
.tag
== CEPH_MSGR_TAG_SEQ
) {
2134 prepare_write_seq(con
);
2135 prepare_read_seq(con
);
2137 prepare_read_tag(con
);
2141 case CEPH_MSGR_TAG_WAIT
:
2143 * If there is a connection race (we are opening
2144 * connections to each other), one of us may just have
2145 * to WAIT. This shouldn't happen if we are the
2148 pr_err("process_connect got WAIT as client\n");
2149 con
->error_msg
= "protocol error, got WAIT as client";
2153 pr_err("connect protocol error, will retry\n");
2154 con
->error_msg
= "protocol error, garbage tag during connect";
2162 * read (part of) an ack
2164 static int read_partial_ack(struct ceph_connection
*con
)
2166 int size
= sizeof (con
->in_temp_ack
);
2169 return read_partial(con
, end
, size
, &con
->in_temp_ack
);
2173 * We can finally discard anything that's been acked.
2175 static void process_ack(struct ceph_connection
*con
)
2178 u64 ack
= le64_to_cpu(con
->in_temp_ack
);
2181 while (!list_empty(&con
->out_sent
)) {
2182 m
= list_first_entry(&con
->out_sent
, struct ceph_msg
,
2184 seq
= le64_to_cpu(m
->hdr
.seq
);
2187 dout("got ack for seq %llu type %d at %p\n", seq
,
2188 le16_to_cpu(m
->hdr
.type
), m
);
2189 m
->ack_stamp
= jiffies
;
2192 prepare_read_tag(con
);
2196 static int read_partial_message_section(struct ceph_connection
*con
,
2197 struct kvec
*section
,
2198 unsigned int sec_len
, u32
*crc
)
2204 while (section
->iov_len
< sec_len
) {
2205 BUG_ON(section
->iov_base
== NULL
);
2206 left
= sec_len
- section
->iov_len
;
2207 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)section
->iov_base
+
2208 section
->iov_len
, left
);
2211 section
->iov_len
+= ret
;
2213 if (section
->iov_len
== sec_len
)
2214 *crc
= crc32c(0, section
->iov_base
, section
->iov_len
);
2219 static int read_partial_msg_data(struct ceph_connection
*con
)
2221 struct ceph_msg
*msg
= con
->in_msg
;
2222 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
2223 const bool do_datacrc
= !con
->msgr
->nocrc
;
2231 if (list_empty(&msg
->data
))
2235 crc
= con
->in_data_crc
;
2236 while (cursor
->resid
) {
2237 page
= ceph_msg_data_next(&msg
->cursor
, &page_offset
, &length
,
2239 ret
= ceph_tcp_recvpage(con
->sock
, page
, page_offset
, length
);
2242 con
->in_data_crc
= crc
;
2248 crc
= ceph_crc32c_page(crc
, page
, page_offset
, ret
);
2249 (void) ceph_msg_data_advance(&msg
->cursor
, (size_t)ret
);
2252 con
->in_data_crc
= crc
;
2254 return 1; /* must return > 0 to indicate success */
2258 * read (part of) a message.
2260 static int ceph_con_in_msg_alloc(struct ceph_connection
*con
, int *skip
);
2262 static int read_partial_message(struct ceph_connection
*con
)
2264 struct ceph_msg
*m
= con
->in_msg
;
2268 unsigned int front_len
, middle_len
, data_len
;
2269 bool do_datacrc
= !con
->msgr
->nocrc
;
2270 bool need_sign
= (con
->peer_features
& CEPH_FEATURE_MSG_AUTH
);
2274 dout("read_partial_message con %p msg %p\n", con
, m
);
2277 size
= sizeof (con
->in_hdr
);
2279 ret
= read_partial(con
, end
, size
, &con
->in_hdr
);
2283 crc
= crc32c(0, &con
->in_hdr
, offsetof(struct ceph_msg_header
, crc
));
2284 if (cpu_to_le32(crc
) != con
->in_hdr
.crc
) {
2285 pr_err("read_partial_message bad hdr "
2286 " crc %u != expected %u\n",
2287 crc
, con
->in_hdr
.crc
);
2291 front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
2292 if (front_len
> CEPH_MSG_MAX_FRONT_LEN
)
2294 middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
2295 if (middle_len
> CEPH_MSG_MAX_MIDDLE_LEN
)
2297 data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
2298 if (data_len
> CEPH_MSG_MAX_DATA_LEN
)
2302 seq
= le64_to_cpu(con
->in_hdr
.seq
);
2303 if ((s64
)seq
- (s64
)con
->in_seq
< 1) {
2304 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2305 ENTITY_NAME(con
->peer_name
),
2306 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2307 seq
, con
->in_seq
+ 1);
2308 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
2310 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2312 } else if ((s64
)seq
- (s64
)con
->in_seq
> 1) {
2313 pr_err("read_partial_message bad seq %lld expected %lld\n",
2314 seq
, con
->in_seq
+ 1);
2315 con
->error_msg
= "bad message sequence # for incoming message";
2319 /* allocate message? */
2323 dout("got hdr type %d front %d data %d\n", con
->in_hdr
.type
,
2324 front_len
, data_len
);
2325 ret
= ceph_con_in_msg_alloc(con
, &skip
);
2329 BUG_ON(!con
->in_msg
^ skip
);
2330 if (con
->in_msg
&& data_len
> con
->in_msg
->data_length
) {
2331 pr_warn("%s skipping long message (%u > %zd)\n",
2332 __func__
, data_len
, con
->in_msg
->data_length
);
2333 ceph_msg_put(con
->in_msg
);
2338 /* skip this message */
2339 dout("alloc_msg said skip message\n");
2340 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
2342 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2347 BUG_ON(!con
->in_msg
);
2348 BUG_ON(con
->in_msg
->con
!= con
);
2350 m
->front
.iov_len
= 0; /* haven't read it yet */
2352 m
->middle
->vec
.iov_len
= 0;
2354 /* prepare for data payload, if any */
2357 prepare_message_data(con
->in_msg
, data_len
);
2361 ret
= read_partial_message_section(con
, &m
->front
, front_len
,
2362 &con
->in_front_crc
);
2368 ret
= read_partial_message_section(con
, &m
->middle
->vec
,
2370 &con
->in_middle_crc
);
2377 ret
= read_partial_msg_data(con
);
2384 size
= sizeof(m
->footer
);
2386 size
= sizeof(m
->old_footer
);
2389 ret
= read_partial(con
, end
, size
, &m
->footer
);
2394 m
->footer
.flags
= m
->old_footer
.flags
;
2398 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2399 m
, front_len
, m
->footer
.front_crc
, middle_len
,
2400 m
->footer
.middle_crc
, data_len
, m
->footer
.data_crc
);
2403 if (con
->in_front_crc
!= le32_to_cpu(m
->footer
.front_crc
)) {
2404 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2405 m
, con
->in_front_crc
, m
->footer
.front_crc
);
2408 if (con
->in_middle_crc
!= le32_to_cpu(m
->footer
.middle_crc
)) {
2409 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2410 m
, con
->in_middle_crc
, m
->footer
.middle_crc
);
2414 (m
->footer
.flags
& CEPH_MSG_FOOTER_NOCRC
) == 0 &&
2415 con
->in_data_crc
!= le32_to_cpu(m
->footer
.data_crc
)) {
2416 pr_err("read_partial_message %p data crc %u != exp. %u\n", m
,
2417 con
->in_data_crc
, le32_to_cpu(m
->footer
.data_crc
));
2421 if (need_sign
&& con
->ops
->check_message_signature
&&
2422 con
->ops
->check_message_signature(con
, m
)) {
2423 pr_err("read_partial_message %p signature check failed\n", m
);
2427 return 1; /* done! */
2431 * Process message. This happens in the worker thread. The callback should
2432 * be careful not to do anything that waits on other incoming messages or it
2435 static void process_message(struct ceph_connection
*con
)
2437 struct ceph_msg
*msg
;
2439 BUG_ON(con
->in_msg
->con
!= con
);
2440 con
->in_msg
->con
= NULL
;
2445 /* if first message, set peer_name */
2446 if (con
->peer_name
.type
== 0)
2447 con
->peer_name
= msg
->hdr
.src
;
2450 mutex_unlock(&con
->mutex
);
2452 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2453 msg
, le64_to_cpu(msg
->hdr
.seq
),
2454 ENTITY_NAME(msg
->hdr
.src
),
2455 le16_to_cpu(msg
->hdr
.type
),
2456 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2457 le32_to_cpu(msg
->hdr
.front_len
),
2458 le32_to_cpu(msg
->hdr
.data_len
),
2459 con
->in_front_crc
, con
->in_middle_crc
, con
->in_data_crc
);
2460 con
->ops
->dispatch(con
, msg
);
2462 mutex_lock(&con
->mutex
);
2467 * Write something to the socket. Called in a worker thread when the
2468 * socket appears to be writeable and we have something ready to send.
2470 static int try_write(struct ceph_connection
*con
)
2474 dout("try_write start %p state %lu\n", con
, con
->state
);
2477 dout("try_write out_kvec_bytes %d\n", con
->out_kvec_bytes
);
2479 /* open the socket first? */
2480 if (con
->state
== CON_STATE_PREOPEN
) {
2482 con
->state
= CON_STATE_CONNECTING
;
2484 con_out_kvec_reset(con
);
2485 prepare_write_banner(con
);
2486 prepare_read_banner(con
);
2488 BUG_ON(con
->in_msg
);
2489 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2490 dout("try_write initiating connect on %p new state %lu\n",
2492 ret
= ceph_tcp_connect(con
);
2494 con
->error_msg
= "connect error";
2500 /* kvec data queued? */
2501 if (con
->out_skip
) {
2502 ret
= write_partial_skip(con
);
2506 if (con
->out_kvec_left
) {
2507 ret
= write_partial_kvec(con
);
2514 if (con
->out_msg_done
) {
2515 ceph_msg_put(con
->out_msg
);
2516 con
->out_msg
= NULL
; /* we're done with this one */
2520 ret
= write_partial_message_data(con
);
2522 goto more_kvec
; /* we need to send the footer, too! */
2526 dout("try_write write_partial_message_data err %d\n",
2533 if (con
->state
== CON_STATE_OPEN
) {
2534 /* is anything else pending? */
2535 if (!list_empty(&con
->out_queue
)) {
2536 prepare_write_message(con
);
2539 if (con
->in_seq
> con
->in_seq_acked
) {
2540 prepare_write_ack(con
);
2543 if (con_flag_test_and_clear(con
, CON_FLAG_KEEPALIVE_PENDING
)) {
2544 prepare_write_keepalive(con
);
2549 /* Nothing to do! */
2550 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
2551 dout("try_write nothing else to write.\n");
2554 dout("try_write done on %p ret %d\n", con
, ret
);
2561 * Read what we can from the socket.
2563 static int try_read(struct ceph_connection
*con
)
2568 dout("try_read start on %p state %lu\n", con
, con
->state
);
2569 if (con
->state
!= CON_STATE_CONNECTING
&&
2570 con
->state
!= CON_STATE_NEGOTIATING
&&
2571 con
->state
!= CON_STATE_OPEN
)
2576 dout("try_read tag %d in_base_pos %d\n", (int)con
->in_tag
,
2579 if (con
->state
== CON_STATE_CONNECTING
) {
2580 dout("try_read connecting\n");
2581 ret
= read_partial_banner(con
);
2584 ret
= process_banner(con
);
2588 con
->state
= CON_STATE_NEGOTIATING
;
2591 * Received banner is good, exchange connection info.
2592 * Do not reset out_kvec, as sending our banner raced
2593 * with receiving peer banner after connect completed.
2595 ret
= prepare_write_connect(con
);
2598 prepare_read_connect(con
);
2600 /* Send connection info before awaiting response */
2604 if (con
->state
== CON_STATE_NEGOTIATING
) {
2605 dout("try_read negotiating\n");
2606 ret
= read_partial_connect(con
);
2609 ret
= process_connect(con
);
2615 WARN_ON(con
->state
!= CON_STATE_OPEN
);
2617 if (con
->in_base_pos
< 0) {
2619 * skipping + discarding content.
2621 * FIXME: there must be a better way to do this!
2623 static char buf
[SKIP_BUF_SIZE
];
2624 int skip
= min((int) sizeof (buf
), -con
->in_base_pos
);
2626 dout("skipping %d / %d bytes\n", skip
, -con
->in_base_pos
);
2627 ret
= ceph_tcp_recvmsg(con
->sock
, buf
, skip
);
2630 con
->in_base_pos
+= ret
;
2631 if (con
->in_base_pos
)
2634 if (con
->in_tag
== CEPH_MSGR_TAG_READY
) {
2638 ret
= ceph_tcp_recvmsg(con
->sock
, &con
->in_tag
, 1);
2641 dout("try_read got tag %d\n", (int)con
->in_tag
);
2642 switch (con
->in_tag
) {
2643 case CEPH_MSGR_TAG_MSG
:
2644 prepare_read_message(con
);
2646 case CEPH_MSGR_TAG_ACK
:
2647 prepare_read_ack(con
);
2649 case CEPH_MSGR_TAG_CLOSE
:
2650 con_close_socket(con
);
2651 con
->state
= CON_STATE_CLOSED
;
2657 if (con
->in_tag
== CEPH_MSGR_TAG_MSG
) {
2658 ret
= read_partial_message(con
);
2662 con
->error_msg
= "bad crc";
2666 con
->error_msg
= "io error";
2671 if (con
->in_tag
== CEPH_MSGR_TAG_READY
)
2673 process_message(con
);
2674 if (con
->state
== CON_STATE_OPEN
)
2675 prepare_read_tag(con
);
2678 if (con
->in_tag
== CEPH_MSGR_TAG_ACK
||
2679 con
->in_tag
== CEPH_MSGR_TAG_SEQ
) {
2681 * the final handshake seq exchange is semantically
2682 * equivalent to an ACK
2684 ret
= read_partial_ack(con
);
2692 dout("try_read done on %p ret %d\n", con
, ret
);
2696 pr_err("try_read bad con->in_tag = %d\n", (int)con
->in_tag
);
2697 con
->error_msg
= "protocol error, garbage tag";
2704 * Atomically queue work on a connection after the specified delay.
2705 * Bump @con reference to avoid races with connection teardown.
2706 * Returns 0 if work was queued, or an error code otherwise.
2708 static int queue_con_delay(struct ceph_connection
*con
, unsigned long delay
)
2710 if (!con
->ops
->get(con
)) {
2711 dout("%s %p ref count 0\n", __func__
, con
);
2715 if (!queue_delayed_work(ceph_msgr_wq
, &con
->work
, delay
)) {
2716 dout("%s %p - already queued\n", __func__
, con
);
2721 dout("%s %p %lu\n", __func__
, con
, delay
);
2725 static void queue_con(struct ceph_connection
*con
)
2727 (void) queue_con_delay(con
, 0);
2730 static void cancel_con(struct ceph_connection
*con
)
2732 if (cancel_delayed_work(&con
->work
)) {
2733 dout("%s %p\n", __func__
, con
);
2738 static bool con_sock_closed(struct ceph_connection
*con
)
2740 if (!con_flag_test_and_clear(con
, CON_FLAG_SOCK_CLOSED
))
2744 case CON_STATE_ ## x: \
2745 con->error_msg = "socket closed (con state " #x ")"; \
2748 switch (con
->state
) {
2756 pr_warn("%s con %p unrecognized state %lu\n",
2757 __func__
, con
, con
->state
);
2758 con
->error_msg
= "unrecognized con state";
2767 static bool con_backoff(struct ceph_connection
*con
)
2771 if (!con_flag_test_and_clear(con
, CON_FLAG_BACKOFF
))
2774 ret
= queue_con_delay(con
, round_jiffies_relative(con
->delay
));
2776 dout("%s: con %p FAILED to back off %lu\n", __func__
,
2778 BUG_ON(ret
== -ENOENT
);
2779 con_flag_set(con
, CON_FLAG_BACKOFF
);
2785 /* Finish fault handling; con->mutex must *not* be held here */
2787 static void con_fault_finish(struct ceph_connection
*con
)
2790 * in case we faulted due to authentication, invalidate our
2791 * current tickets so that we can get new ones.
2793 if (con
->auth_retry
&& con
->ops
->invalidate_authorizer
) {
2794 dout("calling invalidate_authorizer()\n");
2795 con
->ops
->invalidate_authorizer(con
);
2798 if (con
->ops
->fault
)
2799 con
->ops
->fault(con
);
2803 * Do some work on a connection. Drop a connection ref when we're done.
2805 static void con_work(struct work_struct
*work
)
2807 struct ceph_connection
*con
= container_of(work
, struct ceph_connection
,
2811 mutex_lock(&con
->mutex
);
2815 if ((fault
= con_sock_closed(con
))) {
2816 dout("%s: con %p SOCK_CLOSED\n", __func__
, con
);
2819 if (con_backoff(con
)) {
2820 dout("%s: con %p BACKOFF\n", __func__
, con
);
2823 if (con
->state
== CON_STATE_STANDBY
) {
2824 dout("%s: con %p STANDBY\n", __func__
, con
);
2827 if (con
->state
== CON_STATE_CLOSED
) {
2828 dout("%s: con %p CLOSED\n", __func__
, con
);
2832 if (con
->state
== CON_STATE_PREOPEN
) {
2833 dout("%s: con %p PREOPEN\n", __func__
, con
);
2837 ret
= try_read(con
);
2841 con
->error_msg
= "socket error on read";
2846 ret
= try_write(con
);
2850 con
->error_msg
= "socket error on write";
2854 break; /* If we make it to here, we're done */
2858 mutex_unlock(&con
->mutex
);
2861 con_fault_finish(con
);
2867 * Generic error/fault handler. A retry mechanism is used with
2868 * exponential backoff
2870 static void con_fault(struct ceph_connection
*con
)
2872 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con
->peer_name
),
2873 ceph_pr_addr(&con
->peer_addr
.in_addr
), con
->error_msg
);
2874 dout("fault %p state %lu to peer %s\n",
2875 con
, con
->state
, ceph_pr_addr(&con
->peer_addr
.in_addr
));
2877 WARN_ON(con
->state
!= CON_STATE_CONNECTING
&&
2878 con
->state
!= CON_STATE_NEGOTIATING
&&
2879 con
->state
!= CON_STATE_OPEN
);
2881 con_close_socket(con
);
2883 if (con_flag_test(con
, CON_FLAG_LOSSYTX
)) {
2884 dout("fault on LOSSYTX channel, marking CLOSED\n");
2885 con
->state
= CON_STATE_CLOSED
;
2890 BUG_ON(con
->in_msg
->con
!= con
);
2891 con
->in_msg
->con
= NULL
;
2892 ceph_msg_put(con
->in_msg
);
2897 /* Requeue anything that hasn't been acked */
2898 list_splice_init(&con
->out_sent
, &con
->out_queue
);
2900 /* If there are no messages queued or keepalive pending, place
2901 * the connection in a STANDBY state */
2902 if (list_empty(&con
->out_queue
) &&
2903 !con_flag_test(con
, CON_FLAG_KEEPALIVE_PENDING
)) {
2904 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con
);
2905 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
2906 con
->state
= CON_STATE_STANDBY
;
2908 /* retry after a delay. */
2909 con
->state
= CON_STATE_PREOPEN
;
2910 if (con
->delay
== 0)
2911 con
->delay
= BASE_DELAY_INTERVAL
;
2912 else if (con
->delay
< MAX_DELAY_INTERVAL
)
2914 con_flag_set(con
, CON_FLAG_BACKOFF
);
2922 * initialize a new messenger instance
2924 void ceph_messenger_init(struct ceph_messenger
*msgr
,
2925 struct ceph_entity_addr
*myaddr
,
2926 u64 supported_features
,
2927 u64 required_features
,
2931 msgr
->supported_features
= supported_features
;
2932 msgr
->required_features
= required_features
;
2934 spin_lock_init(&msgr
->global_seq_lock
);
2937 msgr
->inst
.addr
= *myaddr
;
2939 /* select a random nonce */
2940 msgr
->inst
.addr
.type
= 0;
2941 get_random_bytes(&msgr
->inst
.addr
.nonce
, sizeof(msgr
->inst
.addr
.nonce
));
2942 encode_my_addr(msgr
);
2943 msgr
->nocrc
= nocrc
;
2944 msgr
->tcp_nodelay
= tcp_nodelay
;
2946 atomic_set(&msgr
->stopping
, 0);
2948 dout("%s %p\n", __func__
, msgr
);
2950 EXPORT_SYMBOL(ceph_messenger_init
);
2952 static void clear_standby(struct ceph_connection
*con
)
2954 /* come back from STANDBY? */
2955 if (con
->state
== CON_STATE_STANDBY
) {
2956 dout("clear_standby %p and ++connect_seq\n", con
);
2957 con
->state
= CON_STATE_PREOPEN
;
2959 WARN_ON(con_flag_test(con
, CON_FLAG_WRITE_PENDING
));
2960 WARN_ON(con_flag_test(con
, CON_FLAG_KEEPALIVE_PENDING
));
2965 * Queue up an outgoing message on the given connection.
2967 void ceph_con_send(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2970 msg
->hdr
.src
= con
->msgr
->inst
.name
;
2971 BUG_ON(msg
->front
.iov_len
!= le32_to_cpu(msg
->hdr
.front_len
));
2972 msg
->needs_out_seq
= true;
2974 mutex_lock(&con
->mutex
);
2976 if (con
->state
== CON_STATE_CLOSED
) {
2977 dout("con_send %p closed, dropping %p\n", con
, msg
);
2979 mutex_unlock(&con
->mutex
);
2983 BUG_ON(msg
->con
!= NULL
);
2984 msg
->con
= con
->ops
->get(con
);
2985 BUG_ON(msg
->con
== NULL
);
2987 BUG_ON(!list_empty(&msg
->list_head
));
2988 list_add_tail(&msg
->list_head
, &con
->out_queue
);
2989 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg
,
2990 ENTITY_NAME(con
->peer_name
), le16_to_cpu(msg
->hdr
.type
),
2991 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2992 le32_to_cpu(msg
->hdr
.front_len
),
2993 le32_to_cpu(msg
->hdr
.middle_len
),
2994 le32_to_cpu(msg
->hdr
.data_len
));
2997 mutex_unlock(&con
->mutex
);
2999 /* if there wasn't anything waiting to send before, queue
3001 if (con_flag_test_and_set(con
, CON_FLAG_WRITE_PENDING
) == 0)
3004 EXPORT_SYMBOL(ceph_con_send
);
3007 * Revoke a message that was previously queued for send
3009 void ceph_msg_revoke(struct ceph_msg
*msg
)
3011 struct ceph_connection
*con
= msg
->con
;
3014 return; /* Message not in our possession */
3016 mutex_lock(&con
->mutex
);
3017 if (!list_empty(&msg
->list_head
)) {
3018 dout("%s %p msg %p - was on queue\n", __func__
, con
, msg
);
3019 list_del_init(&msg
->list_head
);
3020 BUG_ON(msg
->con
== NULL
);
3021 msg
->con
->ops
->put(msg
->con
);
3027 if (con
->out_msg
== msg
) {
3028 dout("%s %p msg %p - was sending\n", __func__
, con
, msg
);
3029 con
->out_msg
= NULL
;
3030 if (con
->out_kvec_is_msg
) {
3031 con
->out_skip
= con
->out_kvec_bytes
;
3032 con
->out_kvec_is_msg
= false;
3038 mutex_unlock(&con
->mutex
);
3042 * Revoke a message that we may be reading data into
3044 void ceph_msg_revoke_incoming(struct ceph_msg
*msg
)
3046 struct ceph_connection
*con
;
3048 BUG_ON(msg
== NULL
);
3050 dout("%s msg %p null con\n", __func__
, msg
);
3052 return; /* Message not in our possession */
3056 mutex_lock(&con
->mutex
);
3057 if (con
->in_msg
== msg
) {
3058 unsigned int front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
3059 unsigned int middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
3060 unsigned int data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
3062 /* skip rest of message */
3063 dout("%s %p msg %p revoked\n", __func__
, con
, msg
);
3064 con
->in_base_pos
= con
->in_base_pos
-
3065 sizeof(struct ceph_msg_header
) -
3069 sizeof(struct ceph_msg_footer
);
3070 ceph_msg_put(con
->in_msg
);
3072 con
->in_tag
= CEPH_MSGR_TAG_READY
;
3075 dout("%s %p in_msg %p msg %p no-op\n",
3076 __func__
, con
, con
->in_msg
, msg
);
3078 mutex_unlock(&con
->mutex
);
3082 * Queue a keepalive byte to ensure the tcp connection is alive.
3084 void ceph_con_keepalive(struct ceph_connection
*con
)
3086 dout("con_keepalive %p\n", con
);
3087 mutex_lock(&con
->mutex
);
3089 mutex_unlock(&con
->mutex
);
3090 if (con_flag_test_and_set(con
, CON_FLAG_KEEPALIVE_PENDING
) == 0 &&
3091 con_flag_test_and_set(con
, CON_FLAG_WRITE_PENDING
) == 0)
3094 EXPORT_SYMBOL(ceph_con_keepalive
);
3096 static struct ceph_msg_data
*ceph_msg_data_create(enum ceph_msg_data_type type
)
3098 struct ceph_msg_data
*data
;
3100 if (WARN_ON(!ceph_msg_data_type_valid(type
)))
3103 data
= kmem_cache_zalloc(ceph_msg_data_cache
, GFP_NOFS
);
3106 INIT_LIST_HEAD(&data
->links
);
3111 static void ceph_msg_data_destroy(struct ceph_msg_data
*data
)
3116 WARN_ON(!list_empty(&data
->links
));
3117 if (data
->type
== CEPH_MSG_DATA_PAGELIST
)
3118 ceph_pagelist_release(data
->pagelist
);
3119 kmem_cache_free(ceph_msg_data_cache
, data
);
3122 void ceph_msg_data_add_pages(struct ceph_msg
*msg
, struct page
**pages
,
3123 size_t length
, size_t alignment
)
3125 struct ceph_msg_data
*data
;
3130 data
= ceph_msg_data_create(CEPH_MSG_DATA_PAGES
);
3132 data
->pages
= pages
;
3133 data
->length
= length
;
3134 data
->alignment
= alignment
& ~PAGE_MASK
;
3136 list_add_tail(&data
->links
, &msg
->data
);
3137 msg
->data_length
+= length
;
3139 EXPORT_SYMBOL(ceph_msg_data_add_pages
);
3141 void ceph_msg_data_add_pagelist(struct ceph_msg
*msg
,
3142 struct ceph_pagelist
*pagelist
)
3144 struct ceph_msg_data
*data
;
3147 BUG_ON(!pagelist
->length
);
3149 data
= ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST
);
3151 data
->pagelist
= pagelist
;
3153 list_add_tail(&data
->links
, &msg
->data
);
3154 msg
->data_length
+= pagelist
->length
;
3156 EXPORT_SYMBOL(ceph_msg_data_add_pagelist
);
3159 void ceph_msg_data_add_bio(struct ceph_msg
*msg
, struct bio
*bio
,
3162 struct ceph_msg_data
*data
;
3166 data
= ceph_msg_data_create(CEPH_MSG_DATA_BIO
);
3169 data
->bio_length
= length
;
3171 list_add_tail(&data
->links
, &msg
->data
);
3172 msg
->data_length
+= length
;
3174 EXPORT_SYMBOL(ceph_msg_data_add_bio
);
3175 #endif /* CONFIG_BLOCK */
3178 * construct a new message with given type, size
3179 * the new msg has a ref count of 1.
3181 struct ceph_msg
*ceph_msg_new(int type
, int front_len
, gfp_t flags
,
3186 m
= kmem_cache_zalloc(ceph_msg_cache
, flags
);
3190 m
->hdr
.type
= cpu_to_le16(type
);
3191 m
->hdr
.priority
= cpu_to_le16(CEPH_MSG_PRIO_DEFAULT
);
3192 m
->hdr
.front_len
= cpu_to_le32(front_len
);
3194 INIT_LIST_HEAD(&m
->list_head
);
3195 kref_init(&m
->kref
);
3196 INIT_LIST_HEAD(&m
->data
);
3200 m
->front
.iov_base
= ceph_kvmalloc(front_len
, flags
);
3201 if (m
->front
.iov_base
== NULL
) {
3202 dout("ceph_msg_new can't allocate %d bytes\n",
3207 m
->front
.iov_base
= NULL
;
3209 m
->front_alloc_len
= m
->front
.iov_len
= front_len
;
3211 dout("ceph_msg_new %p front %d\n", m
, front_len
);
3218 pr_err("msg_new can't create type %d front %d\n", type
,
3222 dout("msg_new can't create type %d front %d\n", type
,
3227 EXPORT_SYMBOL(ceph_msg_new
);
3230 * Allocate "middle" portion of a message, if it is needed and wasn't
3231 * allocated by alloc_msg. This allows us to read a small fixed-size
3232 * per-type header in the front and then gracefully fail (i.e.,
3233 * propagate the error to the caller based on info in the front) when
3234 * the middle is too large.
3236 static int ceph_alloc_middle(struct ceph_connection
*con
, struct ceph_msg
*msg
)
3238 int type
= le16_to_cpu(msg
->hdr
.type
);
3239 int middle_len
= le32_to_cpu(msg
->hdr
.middle_len
);
3241 dout("alloc_middle %p type %d %s middle_len %d\n", msg
, type
,
3242 ceph_msg_type_name(type
), middle_len
);
3243 BUG_ON(!middle_len
);
3244 BUG_ON(msg
->middle
);
3246 msg
->middle
= ceph_buffer_new(middle_len
, GFP_NOFS
);
3253 * Allocate a message for receiving an incoming message on a
3254 * connection, and save the result in con->in_msg. Uses the
3255 * connection's private alloc_msg op if available.
3257 * Returns 0 on success, or a negative error code.
3259 * On success, if we set *skip = 1:
3260 * - the next message should be skipped and ignored.
3261 * - con->in_msg == NULL
3262 * or if we set *skip = 0:
3263 * - con->in_msg is non-null.
3264 * On error (ENOMEM, EAGAIN, ...),
3265 * - con->in_msg == NULL
3267 static int ceph_con_in_msg_alloc(struct ceph_connection
*con
, int *skip
)
3269 struct ceph_msg_header
*hdr
= &con
->in_hdr
;
3270 int middle_len
= le32_to_cpu(hdr
->middle_len
);
3271 struct ceph_msg
*msg
;
3274 BUG_ON(con
->in_msg
!= NULL
);
3275 BUG_ON(!con
->ops
->alloc_msg
);
3277 mutex_unlock(&con
->mutex
);
3278 msg
= con
->ops
->alloc_msg(con
, hdr
, skip
);
3279 mutex_lock(&con
->mutex
);
3280 if (con
->state
!= CON_STATE_OPEN
) {
3288 con
->in_msg
->con
= con
->ops
->get(con
);
3289 BUG_ON(con
->in_msg
->con
== NULL
);
3292 * Null message pointer means either we should skip
3293 * this message or we couldn't allocate memory. The
3294 * former is not an error.
3298 con
->error_msg
= "error allocating memory for incoming message";
3302 memcpy(&con
->in_msg
->hdr
, &con
->in_hdr
, sizeof(con
->in_hdr
));
3304 if (middle_len
&& !con
->in_msg
->middle
) {
3305 ret
= ceph_alloc_middle(con
, con
->in_msg
);
3307 ceph_msg_put(con
->in_msg
);
3317 * Free a generically kmalloc'd message.
3319 static void ceph_msg_free(struct ceph_msg
*m
)
3321 dout("%s %p\n", __func__
, m
);
3322 kvfree(m
->front
.iov_base
);
3323 kmem_cache_free(ceph_msg_cache
, m
);
3326 static void ceph_msg_release(struct kref
*kref
)
3328 struct ceph_msg
*m
= container_of(kref
, struct ceph_msg
, kref
);
3330 struct list_head
*links
;
3331 struct list_head
*next
;
3333 dout("%s %p\n", __func__
, m
);
3334 WARN_ON(!list_empty(&m
->list_head
));
3336 /* drop middle, data, if any */
3338 ceph_buffer_put(m
->middle
);
3342 list_splice_init(&m
->data
, &data
);
3343 list_for_each_safe(links
, next
, &data
) {
3344 struct ceph_msg_data
*data
;
3346 data
= list_entry(links
, struct ceph_msg_data
, links
);
3347 list_del_init(links
);
3348 ceph_msg_data_destroy(data
);
3353 ceph_msgpool_put(m
->pool
, m
);
3358 struct ceph_msg
*ceph_msg_get(struct ceph_msg
*msg
)
3360 dout("%s %p (was %d)\n", __func__
, msg
,
3361 atomic_read(&msg
->kref
.refcount
));
3362 kref_get(&msg
->kref
);
3365 EXPORT_SYMBOL(ceph_msg_get
);
3367 void ceph_msg_put(struct ceph_msg
*msg
)
3369 dout("%s %p (was %d)\n", __func__
, msg
,
3370 atomic_read(&msg
->kref
.refcount
));
3371 kref_put(&msg
->kref
, ceph_msg_release
);
3373 EXPORT_SYMBOL(ceph_msg_put
);
3375 void ceph_msg_dump(struct ceph_msg
*msg
)
3377 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg
,
3378 msg
->front_alloc_len
, msg
->data_length
);
3379 print_hex_dump(KERN_DEBUG
, "header: ",
3380 DUMP_PREFIX_OFFSET
, 16, 1,
3381 &msg
->hdr
, sizeof(msg
->hdr
), true);
3382 print_hex_dump(KERN_DEBUG
, " front: ",
3383 DUMP_PREFIX_OFFSET
, 16, 1,
3384 msg
->front
.iov_base
, msg
->front
.iov_len
, true);
3386 print_hex_dump(KERN_DEBUG
, "middle: ",
3387 DUMP_PREFIX_OFFSET
, 16, 1,
3388 msg
->middle
->vec
.iov_base
,
3389 msg
->middle
->vec
.iov_len
, true);
3390 print_hex_dump(KERN_DEBUG
, "footer: ",
3391 DUMP_PREFIX_OFFSET
, 16, 1,
3392 &msg
->footer
, sizeof(msg
->footer
), true);
3394 EXPORT_SYMBOL(ceph_msg_dump
);