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
| __GFP_MEMALLOC
;
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 sk_set_memalloc(sock
->sk
);
519 static int ceph_tcp_recvmsg(struct socket
*sock
, void *buf
, size_t len
)
521 struct kvec iov
= {buf
, len
};
522 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
525 r
= kernel_recvmsg(sock
, &msg
, &iov
, 1, len
, msg
.msg_flags
);
531 static int ceph_tcp_recvpage(struct socket
*sock
, struct page
*page
,
532 int page_offset
, size_t length
)
537 BUG_ON(page_offset
+ length
> PAGE_SIZE
);
541 ret
= ceph_tcp_recvmsg(sock
, kaddr
+ page_offset
, length
);
548 * write something. @more is true if caller will be sending more data
551 static int ceph_tcp_sendmsg(struct socket
*sock
, struct kvec
*iov
,
552 size_t kvlen
, size_t len
, int more
)
554 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
558 msg
.msg_flags
|= MSG_MORE
;
560 msg
.msg_flags
|= MSG_EOR
; /* superfluous, but what the hell */
562 r
= kernel_sendmsg(sock
, &msg
, iov
, kvlen
, len
);
568 static int __ceph_tcp_sendpage(struct socket
*sock
, struct page
*page
,
569 int offset
, size_t size
, bool more
)
571 int flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
| (more
? MSG_MORE
: MSG_EOR
);
574 ret
= kernel_sendpage(sock
, page
, offset
, size
, flags
);
581 static int ceph_tcp_sendpage(struct socket
*sock
, struct page
*page
,
582 int offset
, size_t size
, bool more
)
587 /* sendpage cannot properly handle pages with page_count == 0,
588 * we need to fallback to sendmsg if that's the case */
589 if (page_count(page
) >= 1)
590 return __ceph_tcp_sendpage(sock
, page
, offset
, size
, more
);
592 iov
.iov_base
= kmap(page
) + offset
;
594 ret
= ceph_tcp_sendmsg(sock
, &iov
, 1, size
, more
);
601 * Shutdown/close the socket for the given connection.
603 static int con_close_socket(struct ceph_connection
*con
)
607 dout("con_close_socket on %p sock %p\n", con
, con
->sock
);
609 rc
= con
->sock
->ops
->shutdown(con
->sock
, SHUT_RDWR
);
610 sock_release(con
->sock
);
615 * Forcibly clear the SOCK_CLOSED flag. It gets set
616 * independent of the connection mutex, and we could have
617 * received a socket close event before we had the chance to
618 * shut the socket down.
620 con_flag_clear(con
, CON_FLAG_SOCK_CLOSED
);
622 con_sock_state_closed(con
);
627 * Reset a connection. Discard all incoming and outgoing messages
628 * and clear *_seq state.
630 static void ceph_msg_remove(struct ceph_msg
*msg
)
632 list_del_init(&msg
->list_head
);
633 BUG_ON(msg
->con
== NULL
);
634 msg
->con
->ops
->put(msg
->con
);
639 static void ceph_msg_remove_list(struct list_head
*head
)
641 while (!list_empty(head
)) {
642 struct ceph_msg
*msg
= list_first_entry(head
, struct ceph_msg
,
644 ceph_msg_remove(msg
);
648 static void reset_connection(struct ceph_connection
*con
)
650 /* reset connection, out_queue, msg_ and connect_seq */
651 /* discard existing out_queue and msg_seq */
652 dout("reset_connection %p\n", con
);
653 ceph_msg_remove_list(&con
->out_queue
);
654 ceph_msg_remove_list(&con
->out_sent
);
657 BUG_ON(con
->in_msg
->con
!= con
);
658 con
->in_msg
->con
= NULL
;
659 ceph_msg_put(con
->in_msg
);
664 con
->connect_seq
= 0;
667 ceph_msg_put(con
->out_msg
);
671 con
->in_seq_acked
= 0;
675 * mark a peer down. drop any open connections.
677 void ceph_con_close(struct ceph_connection
*con
)
679 mutex_lock(&con
->mutex
);
680 dout("con_close %p peer %s\n", con
,
681 ceph_pr_addr(&con
->peer_addr
.in_addr
));
682 con
->state
= CON_STATE_CLOSED
;
684 con_flag_clear(con
, CON_FLAG_LOSSYTX
); /* so we retry next connect */
685 con_flag_clear(con
, CON_FLAG_KEEPALIVE_PENDING
);
686 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
687 con_flag_clear(con
, CON_FLAG_BACKOFF
);
689 reset_connection(con
);
690 con
->peer_global_seq
= 0;
692 con_close_socket(con
);
693 mutex_unlock(&con
->mutex
);
695 EXPORT_SYMBOL(ceph_con_close
);
698 * Reopen a closed connection, with a new peer address.
700 void ceph_con_open(struct ceph_connection
*con
,
701 __u8 entity_type
, __u64 entity_num
,
702 struct ceph_entity_addr
*addr
)
704 mutex_lock(&con
->mutex
);
705 dout("con_open %p %s\n", con
, ceph_pr_addr(&addr
->in_addr
));
707 WARN_ON(con
->state
!= CON_STATE_CLOSED
);
708 con
->state
= CON_STATE_PREOPEN
;
710 con
->peer_name
.type
= (__u8
) entity_type
;
711 con
->peer_name
.num
= cpu_to_le64(entity_num
);
713 memcpy(&con
->peer_addr
, addr
, sizeof(*addr
));
714 con
->delay
= 0; /* reset backoff memory */
715 mutex_unlock(&con
->mutex
);
718 EXPORT_SYMBOL(ceph_con_open
);
721 * return true if this connection ever successfully opened
723 bool ceph_con_opened(struct ceph_connection
*con
)
725 return con
->connect_seq
> 0;
729 * initialize a new connection.
731 void ceph_con_init(struct ceph_connection
*con
, void *private,
732 const struct ceph_connection_operations
*ops
,
733 struct ceph_messenger
*msgr
)
735 dout("con_init %p\n", con
);
736 memset(con
, 0, sizeof(*con
));
737 con
->private = private;
741 con_sock_state_init(con
);
743 mutex_init(&con
->mutex
);
744 INIT_LIST_HEAD(&con
->out_queue
);
745 INIT_LIST_HEAD(&con
->out_sent
);
746 INIT_DELAYED_WORK(&con
->work
, con_work
);
748 con
->state
= CON_STATE_CLOSED
;
750 EXPORT_SYMBOL(ceph_con_init
);
754 * We maintain a global counter to order connection attempts. Get
755 * a unique seq greater than @gt.
757 static u32
get_global_seq(struct ceph_messenger
*msgr
, u32 gt
)
761 spin_lock(&msgr
->global_seq_lock
);
762 if (msgr
->global_seq
< gt
)
763 msgr
->global_seq
= gt
;
764 ret
= ++msgr
->global_seq
;
765 spin_unlock(&msgr
->global_seq_lock
);
769 static void con_out_kvec_reset(struct ceph_connection
*con
)
771 con
->out_kvec_left
= 0;
772 con
->out_kvec_bytes
= 0;
773 con
->out_kvec_cur
= &con
->out_kvec
[0];
776 static void con_out_kvec_add(struct ceph_connection
*con
,
777 size_t size
, void *data
)
781 index
= con
->out_kvec_left
;
782 BUG_ON(index
>= ARRAY_SIZE(con
->out_kvec
));
784 con
->out_kvec
[index
].iov_len
= size
;
785 con
->out_kvec
[index
].iov_base
= data
;
786 con
->out_kvec_left
++;
787 con
->out_kvec_bytes
+= size
;
793 * For a bio data item, a piece is whatever remains of the next
794 * entry in the current bio iovec, or the first entry in the next
797 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor
*cursor
,
800 struct ceph_msg_data
*data
= cursor
->data
;
803 BUG_ON(data
->type
!= CEPH_MSG_DATA_BIO
);
808 cursor
->resid
= min(length
, data
->bio_length
);
810 cursor
->bvec_iter
= bio
->bi_iter
;
812 cursor
->resid
<= bio_iter_len(bio
, cursor
->bvec_iter
);
815 static struct page
*ceph_msg_data_bio_next(struct ceph_msg_data_cursor
*cursor
,
819 struct ceph_msg_data
*data
= cursor
->data
;
821 struct bio_vec bio_vec
;
823 BUG_ON(data
->type
!= CEPH_MSG_DATA_BIO
);
828 bio_vec
= bio_iter_iovec(bio
, cursor
->bvec_iter
);
830 *page_offset
= (size_t) bio_vec
.bv_offset
;
831 BUG_ON(*page_offset
>= PAGE_SIZE
);
832 if (cursor
->last_piece
) /* pagelist offset is always 0 */
833 *length
= cursor
->resid
;
835 *length
= (size_t) bio_vec
.bv_len
;
836 BUG_ON(*length
> cursor
->resid
);
837 BUG_ON(*page_offset
+ *length
> PAGE_SIZE
);
839 return bio_vec
.bv_page
;
842 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor
*cursor
,
846 struct bio_vec bio_vec
;
848 BUG_ON(cursor
->data
->type
!= CEPH_MSG_DATA_BIO
);
853 bio_vec
= bio_iter_iovec(bio
, cursor
->bvec_iter
);
855 /* Advance the cursor offset */
857 BUG_ON(cursor
->resid
< bytes
);
858 cursor
->resid
-= bytes
;
860 bio_advance_iter(bio
, &cursor
->bvec_iter
, bytes
);
862 if (bytes
< bio_vec
.bv_len
)
863 return false; /* more bytes to process in this segment */
865 /* Move on to the next segment, and possibly the next bio */
867 if (!cursor
->bvec_iter
.bi_size
) {
871 cursor
->bvec_iter
= bio
->bi_iter
;
873 memset(&cursor
->bvec_iter
, 0,
874 sizeof(cursor
->bvec_iter
));
877 if (!cursor
->last_piece
) {
878 BUG_ON(!cursor
->resid
);
880 /* A short read is OK, so use <= rather than == */
881 if (cursor
->resid
<= bio_iter_len(bio
, cursor
->bvec_iter
))
882 cursor
->last_piece
= true;
887 #endif /* CONFIG_BLOCK */
890 * For a page array, a piece comes from the first page in the array
891 * that has not already been fully consumed.
893 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor
*cursor
,
896 struct ceph_msg_data
*data
= cursor
->data
;
899 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGES
);
901 BUG_ON(!data
->pages
);
902 BUG_ON(!data
->length
);
904 cursor
->resid
= min(length
, data
->length
);
905 page_count
= calc_pages_for(data
->alignment
, (u64
)data
->length
);
906 cursor
->page_offset
= data
->alignment
& ~PAGE_MASK
;
907 cursor
->page_index
= 0;
908 BUG_ON(page_count
> (int)USHRT_MAX
);
909 cursor
->page_count
= (unsigned short)page_count
;
910 BUG_ON(length
> SIZE_MAX
- cursor
->page_offset
);
911 cursor
->last_piece
= cursor
->page_offset
+ cursor
->resid
<= PAGE_SIZE
;
915 ceph_msg_data_pages_next(struct ceph_msg_data_cursor
*cursor
,
916 size_t *page_offset
, size_t *length
)
918 struct ceph_msg_data
*data
= cursor
->data
;
920 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGES
);
922 BUG_ON(cursor
->page_index
>= cursor
->page_count
);
923 BUG_ON(cursor
->page_offset
>= PAGE_SIZE
);
925 *page_offset
= cursor
->page_offset
;
926 if (cursor
->last_piece
)
927 *length
= cursor
->resid
;
929 *length
= PAGE_SIZE
- *page_offset
;
931 return data
->pages
[cursor
->page_index
];
934 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor
*cursor
,
937 BUG_ON(cursor
->data
->type
!= CEPH_MSG_DATA_PAGES
);
939 BUG_ON(cursor
->page_offset
+ bytes
> PAGE_SIZE
);
941 /* Advance the cursor page offset */
943 cursor
->resid
-= bytes
;
944 cursor
->page_offset
= (cursor
->page_offset
+ bytes
) & ~PAGE_MASK
;
945 if (!bytes
|| cursor
->page_offset
)
946 return false; /* more bytes to process in the current page */
949 return false; /* no more data */
951 /* Move on to the next page; offset is already at 0 */
953 BUG_ON(cursor
->page_index
>= cursor
->page_count
);
954 cursor
->page_index
++;
955 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
961 * For a pagelist, a piece is whatever remains to be consumed in the
962 * first page in the list, or the front of the next page.
965 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor
*cursor
,
968 struct ceph_msg_data
*data
= cursor
->data
;
969 struct ceph_pagelist
*pagelist
;
972 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
974 pagelist
= data
->pagelist
;
978 return; /* pagelist can be assigned but empty */
980 BUG_ON(list_empty(&pagelist
->head
));
981 page
= list_first_entry(&pagelist
->head
, struct page
, lru
);
983 cursor
->resid
= min(length
, pagelist
->length
);
986 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
990 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor
*cursor
,
991 size_t *page_offset
, size_t *length
)
993 struct ceph_msg_data
*data
= cursor
->data
;
994 struct ceph_pagelist
*pagelist
;
996 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
998 pagelist
= data
->pagelist
;
1001 BUG_ON(!cursor
->page
);
1002 BUG_ON(cursor
->offset
+ cursor
->resid
!= pagelist
->length
);
1004 /* offset of first page in pagelist is always 0 */
1005 *page_offset
= cursor
->offset
& ~PAGE_MASK
;
1006 if (cursor
->last_piece
)
1007 *length
= cursor
->resid
;
1009 *length
= PAGE_SIZE
- *page_offset
;
1011 return cursor
->page
;
1014 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor
*cursor
,
1017 struct ceph_msg_data
*data
= cursor
->data
;
1018 struct ceph_pagelist
*pagelist
;
1020 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
1022 pagelist
= data
->pagelist
;
1025 BUG_ON(cursor
->offset
+ cursor
->resid
!= pagelist
->length
);
1026 BUG_ON((cursor
->offset
& ~PAGE_MASK
) + bytes
> PAGE_SIZE
);
1028 /* Advance the cursor offset */
1030 cursor
->resid
-= bytes
;
1031 cursor
->offset
+= bytes
;
1032 /* offset of first page in pagelist is always 0 */
1033 if (!bytes
|| cursor
->offset
& ~PAGE_MASK
)
1034 return false; /* more bytes to process in the current page */
1037 return false; /* no more data */
1039 /* Move on to the next page */
1041 BUG_ON(list_is_last(&cursor
->page
->lru
, &pagelist
->head
));
1042 cursor
->page
= list_entry_next(cursor
->page
, lru
);
1043 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
1049 * Message data is handled (sent or received) in pieces, where each
1050 * piece resides on a single page. The network layer might not
1051 * consume an entire piece at once. A data item's cursor keeps
1052 * track of which piece is next to process and how much remains to
1053 * be processed in that piece. It also tracks whether the current
1054 * piece is the last one in the data item.
1056 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor
*cursor
)
1058 size_t length
= cursor
->total_resid
;
1060 switch (cursor
->data
->type
) {
1061 case CEPH_MSG_DATA_PAGELIST
:
1062 ceph_msg_data_pagelist_cursor_init(cursor
, length
);
1064 case CEPH_MSG_DATA_PAGES
:
1065 ceph_msg_data_pages_cursor_init(cursor
, length
);
1068 case CEPH_MSG_DATA_BIO
:
1069 ceph_msg_data_bio_cursor_init(cursor
, length
);
1071 #endif /* CONFIG_BLOCK */
1072 case CEPH_MSG_DATA_NONE
:
1077 cursor
->need_crc
= true;
1080 static void ceph_msg_data_cursor_init(struct ceph_msg
*msg
, size_t length
)
1082 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
1083 struct ceph_msg_data
*data
;
1086 BUG_ON(length
> msg
->data_length
);
1087 BUG_ON(list_empty(&msg
->data
));
1089 cursor
->data_head
= &msg
->data
;
1090 cursor
->total_resid
= length
;
1091 data
= list_first_entry(&msg
->data
, struct ceph_msg_data
, links
);
1092 cursor
->data
= data
;
1094 __ceph_msg_data_cursor_init(cursor
);
1098 * Return the page containing the next piece to process for a given
1099 * data item, and supply the page offset and length of that piece.
1100 * Indicate whether this is the last piece in this data item.
1102 static struct page
*ceph_msg_data_next(struct ceph_msg_data_cursor
*cursor
,
1103 size_t *page_offset
, size_t *length
,
1108 switch (cursor
->data
->type
) {
1109 case CEPH_MSG_DATA_PAGELIST
:
1110 page
= ceph_msg_data_pagelist_next(cursor
, page_offset
, length
);
1112 case CEPH_MSG_DATA_PAGES
:
1113 page
= ceph_msg_data_pages_next(cursor
, page_offset
, length
);
1116 case CEPH_MSG_DATA_BIO
:
1117 page
= ceph_msg_data_bio_next(cursor
, page_offset
, length
);
1119 #endif /* CONFIG_BLOCK */
1120 case CEPH_MSG_DATA_NONE
:
1126 BUG_ON(*page_offset
+ *length
> PAGE_SIZE
);
1129 *last_piece
= cursor
->last_piece
;
1135 * Returns true if the result moves the cursor on to the next piece
1138 static bool ceph_msg_data_advance(struct ceph_msg_data_cursor
*cursor
,
1143 BUG_ON(bytes
> cursor
->resid
);
1144 switch (cursor
->data
->type
) {
1145 case CEPH_MSG_DATA_PAGELIST
:
1146 new_piece
= ceph_msg_data_pagelist_advance(cursor
, bytes
);
1148 case CEPH_MSG_DATA_PAGES
:
1149 new_piece
= ceph_msg_data_pages_advance(cursor
, bytes
);
1152 case CEPH_MSG_DATA_BIO
:
1153 new_piece
= ceph_msg_data_bio_advance(cursor
, bytes
);
1155 #endif /* CONFIG_BLOCK */
1156 case CEPH_MSG_DATA_NONE
:
1161 cursor
->total_resid
-= bytes
;
1163 if (!cursor
->resid
&& cursor
->total_resid
) {
1164 WARN_ON(!cursor
->last_piece
);
1165 BUG_ON(list_is_last(&cursor
->data
->links
, cursor
->data_head
));
1166 cursor
->data
= list_entry_next(cursor
->data
, links
);
1167 __ceph_msg_data_cursor_init(cursor
);
1170 cursor
->need_crc
= new_piece
;
1175 static void prepare_message_data(struct ceph_msg
*msg
, u32 data_len
)
1180 /* Initialize data cursor */
1182 ceph_msg_data_cursor_init(msg
, (size_t)data_len
);
1186 * Prepare footer for currently outgoing message, and finish things
1187 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1189 static void prepare_write_message_footer(struct ceph_connection
*con
)
1191 struct ceph_msg
*m
= con
->out_msg
;
1192 int v
= con
->out_kvec_left
;
1194 m
->footer
.flags
|= CEPH_MSG_FOOTER_COMPLETE
;
1196 dout("prepare_write_message_footer %p\n", con
);
1197 con
->out_kvec_is_msg
= true;
1198 con
->out_kvec
[v
].iov_base
= &m
->footer
;
1199 if (con
->peer_features
& CEPH_FEATURE_MSG_AUTH
) {
1200 if (con
->ops
->sign_message
)
1201 con
->ops
->sign_message(con
, m
);
1204 con
->out_kvec
[v
].iov_len
= sizeof(m
->footer
);
1205 con
->out_kvec_bytes
+= sizeof(m
->footer
);
1207 m
->old_footer
.flags
= m
->footer
.flags
;
1208 con
->out_kvec
[v
].iov_len
= sizeof(m
->old_footer
);
1209 con
->out_kvec_bytes
+= sizeof(m
->old_footer
);
1211 con
->out_kvec_left
++;
1212 con
->out_more
= m
->more_to_follow
;
1213 con
->out_msg_done
= true;
1217 * Prepare headers for the next outgoing message.
1219 static void prepare_write_message(struct ceph_connection
*con
)
1224 con_out_kvec_reset(con
);
1225 con
->out_kvec_is_msg
= true;
1226 con
->out_msg_done
= false;
1228 /* Sneak an ack in there first? If we can get it into the same
1229 * TCP packet that's a good thing. */
1230 if (con
->in_seq
> con
->in_seq_acked
) {
1231 con
->in_seq_acked
= con
->in_seq
;
1232 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
1233 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1234 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1235 &con
->out_temp_ack
);
1238 BUG_ON(list_empty(&con
->out_queue
));
1239 m
= list_first_entry(&con
->out_queue
, struct ceph_msg
, list_head
);
1241 BUG_ON(m
->con
!= con
);
1243 /* put message on sent list */
1245 list_move_tail(&m
->list_head
, &con
->out_sent
);
1248 * only assign outgoing seq # if we haven't sent this message
1249 * yet. if it is requeued, resend with it's original seq.
1251 if (m
->needs_out_seq
) {
1252 m
->hdr
.seq
= cpu_to_le64(++con
->out_seq
);
1253 m
->needs_out_seq
= false;
1255 WARN_ON(m
->data_length
!= le32_to_cpu(m
->hdr
.data_len
));
1257 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1258 m
, con
->out_seq
, le16_to_cpu(m
->hdr
.type
),
1259 le32_to_cpu(m
->hdr
.front_len
), le32_to_cpu(m
->hdr
.middle_len
),
1261 BUG_ON(le32_to_cpu(m
->hdr
.front_len
) != m
->front
.iov_len
);
1263 /* tag + hdr + front + middle */
1264 con_out_kvec_add(con
, sizeof (tag_msg
), &tag_msg
);
1265 con_out_kvec_add(con
, sizeof (m
->hdr
), &m
->hdr
);
1266 con_out_kvec_add(con
, m
->front
.iov_len
, m
->front
.iov_base
);
1269 con_out_kvec_add(con
, m
->middle
->vec
.iov_len
,
1270 m
->middle
->vec
.iov_base
);
1272 /* fill in crc (except data pages), footer */
1273 crc
= crc32c(0, &m
->hdr
, offsetof(struct ceph_msg_header
, crc
));
1274 con
->out_msg
->hdr
.crc
= cpu_to_le32(crc
);
1275 con
->out_msg
->footer
.flags
= 0;
1277 crc
= crc32c(0, m
->front
.iov_base
, m
->front
.iov_len
);
1278 con
->out_msg
->footer
.front_crc
= cpu_to_le32(crc
);
1280 crc
= crc32c(0, m
->middle
->vec
.iov_base
,
1281 m
->middle
->vec
.iov_len
);
1282 con
->out_msg
->footer
.middle_crc
= cpu_to_le32(crc
);
1284 con
->out_msg
->footer
.middle_crc
= 0;
1285 dout("%s front_crc %u middle_crc %u\n", __func__
,
1286 le32_to_cpu(con
->out_msg
->footer
.front_crc
),
1287 le32_to_cpu(con
->out_msg
->footer
.middle_crc
));
1289 /* is there a data payload? */
1290 con
->out_msg
->footer
.data_crc
= 0;
1291 if (m
->data_length
) {
1292 prepare_message_data(con
->out_msg
, m
->data_length
);
1293 con
->out_more
= 1; /* data + footer will follow */
1295 /* no, queue up footer too and be done */
1296 prepare_write_message_footer(con
);
1299 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1305 static void prepare_write_ack(struct ceph_connection
*con
)
1307 dout("prepare_write_ack %p %llu -> %llu\n", con
,
1308 con
->in_seq_acked
, con
->in_seq
);
1309 con
->in_seq_acked
= con
->in_seq
;
1311 con_out_kvec_reset(con
);
1313 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
1315 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1316 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1317 &con
->out_temp_ack
);
1319 con
->out_more
= 1; /* more will follow.. eventually.. */
1320 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1324 * Prepare to share the seq during handshake
1326 static void prepare_write_seq(struct ceph_connection
*con
)
1328 dout("prepare_write_seq %p %llu -> %llu\n", con
,
1329 con
->in_seq_acked
, con
->in_seq
);
1330 con
->in_seq_acked
= con
->in_seq
;
1332 con_out_kvec_reset(con
);
1334 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1335 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1336 &con
->out_temp_ack
);
1338 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1342 * Prepare to write keepalive byte.
1344 static void prepare_write_keepalive(struct ceph_connection
*con
)
1346 dout("prepare_write_keepalive %p\n", con
);
1347 con_out_kvec_reset(con
);
1348 con_out_kvec_add(con
, sizeof (tag_keepalive
), &tag_keepalive
);
1349 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1353 * Connection negotiation.
1356 static struct ceph_auth_handshake
*get_connect_authorizer(struct ceph_connection
*con
,
1359 struct ceph_auth_handshake
*auth
;
1361 if (!con
->ops
->get_authorizer
) {
1362 con
->out_connect
.authorizer_protocol
= CEPH_AUTH_UNKNOWN
;
1363 con
->out_connect
.authorizer_len
= 0;
1367 /* Can't hold the mutex while getting authorizer */
1368 mutex_unlock(&con
->mutex
);
1369 auth
= con
->ops
->get_authorizer(con
, auth_proto
, con
->auth_retry
);
1370 mutex_lock(&con
->mutex
);
1374 if (con
->state
!= CON_STATE_NEGOTIATING
)
1375 return ERR_PTR(-EAGAIN
);
1377 con
->auth_reply_buf
= auth
->authorizer_reply_buf
;
1378 con
->auth_reply_buf_len
= auth
->authorizer_reply_buf_len
;
1383 * We connected to a peer and are saying hello.
1385 static void prepare_write_banner(struct ceph_connection
*con
)
1387 con_out_kvec_add(con
, strlen(CEPH_BANNER
), CEPH_BANNER
);
1388 con_out_kvec_add(con
, sizeof (con
->msgr
->my_enc_addr
),
1389 &con
->msgr
->my_enc_addr
);
1392 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1395 static int prepare_write_connect(struct ceph_connection
*con
)
1397 unsigned int global_seq
= get_global_seq(con
->msgr
, 0);
1400 struct ceph_auth_handshake
*auth
;
1402 switch (con
->peer_name
.type
) {
1403 case CEPH_ENTITY_TYPE_MON
:
1404 proto
= CEPH_MONC_PROTOCOL
;
1406 case CEPH_ENTITY_TYPE_OSD
:
1407 proto
= CEPH_OSDC_PROTOCOL
;
1409 case CEPH_ENTITY_TYPE_MDS
:
1410 proto
= CEPH_MDSC_PROTOCOL
;
1416 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con
,
1417 con
->connect_seq
, global_seq
, proto
);
1419 con
->out_connect
.features
= cpu_to_le64(con
->msgr
->supported_features
);
1420 con
->out_connect
.host_type
= cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT
);
1421 con
->out_connect
.connect_seq
= cpu_to_le32(con
->connect_seq
);
1422 con
->out_connect
.global_seq
= cpu_to_le32(global_seq
);
1423 con
->out_connect
.protocol_version
= cpu_to_le32(proto
);
1424 con
->out_connect
.flags
= 0;
1426 auth_proto
= CEPH_AUTH_UNKNOWN
;
1427 auth
= get_connect_authorizer(con
, &auth_proto
);
1429 return PTR_ERR(auth
);
1431 con
->out_connect
.authorizer_protocol
= cpu_to_le32(auth_proto
);
1432 con
->out_connect
.authorizer_len
= auth
?
1433 cpu_to_le32(auth
->authorizer_buf_len
) : 0;
1435 con_out_kvec_add(con
, sizeof (con
->out_connect
),
1437 if (auth
&& auth
->authorizer_buf_len
)
1438 con_out_kvec_add(con
, auth
->authorizer_buf_len
,
1439 auth
->authorizer_buf
);
1442 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1448 * write as much of pending kvecs to the socket as we can.
1450 * 0 -> socket full, but more to do
1453 static int write_partial_kvec(struct ceph_connection
*con
)
1457 dout("write_partial_kvec %p %d left\n", con
, con
->out_kvec_bytes
);
1458 while (con
->out_kvec_bytes
> 0) {
1459 ret
= ceph_tcp_sendmsg(con
->sock
, con
->out_kvec_cur
,
1460 con
->out_kvec_left
, con
->out_kvec_bytes
,
1464 con
->out_kvec_bytes
-= ret
;
1465 if (con
->out_kvec_bytes
== 0)
1468 /* account for full iov entries consumed */
1469 while (ret
>= con
->out_kvec_cur
->iov_len
) {
1470 BUG_ON(!con
->out_kvec_left
);
1471 ret
-= con
->out_kvec_cur
->iov_len
;
1472 con
->out_kvec_cur
++;
1473 con
->out_kvec_left
--;
1475 /* and for a partially-consumed entry */
1477 con
->out_kvec_cur
->iov_len
-= ret
;
1478 con
->out_kvec_cur
->iov_base
+= ret
;
1481 con
->out_kvec_left
= 0;
1482 con
->out_kvec_is_msg
= false;
1485 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con
,
1486 con
->out_kvec_bytes
, con
->out_kvec_left
, ret
);
1487 return ret
; /* done! */
1490 static u32
ceph_crc32c_page(u32 crc
, struct page
*page
,
1491 unsigned int page_offset
,
1492 unsigned int length
)
1497 BUG_ON(kaddr
== NULL
);
1498 crc
= crc32c(crc
, kaddr
+ page_offset
, length
);
1504 * Write as much message data payload as we can. If we finish, queue
1506 * 1 -> done, footer is now queued in out_kvec[].
1507 * 0 -> socket full, but more to do
1510 static int write_partial_message_data(struct ceph_connection
*con
)
1512 struct ceph_msg
*msg
= con
->out_msg
;
1513 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
1514 bool do_datacrc
= !con
->msgr
->nocrc
;
1517 dout("%s %p msg %p\n", __func__
, con
, msg
);
1519 if (list_empty(&msg
->data
))
1523 * Iterate through each page that contains data to be
1524 * written, and send as much as possible for each.
1526 * If we are calculating the data crc (the default), we will
1527 * need to map the page. If we have no pages, they have
1528 * been revoked, so use the zero page.
1530 crc
= do_datacrc
? le32_to_cpu(msg
->footer
.data_crc
) : 0;
1531 while (cursor
->resid
) {
1539 page
= ceph_msg_data_next(&msg
->cursor
, &page_offset
, &length
,
1541 ret
= ceph_tcp_sendpage(con
->sock
, page
, page_offset
,
1542 length
, last_piece
);
1545 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1549 if (do_datacrc
&& cursor
->need_crc
)
1550 crc
= ceph_crc32c_page(crc
, page
, page_offset
, length
);
1551 need_crc
= ceph_msg_data_advance(&msg
->cursor
, (size_t)ret
);
1554 dout("%s %p msg %p done\n", __func__
, con
, msg
);
1556 /* prepare and queue up footer, too */
1558 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1560 msg
->footer
.flags
|= CEPH_MSG_FOOTER_NOCRC
;
1561 con_out_kvec_reset(con
);
1562 prepare_write_message_footer(con
);
1564 return 1; /* must return > 0 to indicate success */
1570 static int write_partial_skip(struct ceph_connection
*con
)
1574 while (con
->out_skip
> 0) {
1575 size_t size
= min(con
->out_skip
, (int) PAGE_CACHE_SIZE
);
1577 ret
= ceph_tcp_sendpage(con
->sock
, zero_page
, 0, size
, true);
1580 con
->out_skip
-= ret
;
1588 * Prepare to read connection handshake, or an ack.
1590 static void prepare_read_banner(struct ceph_connection
*con
)
1592 dout("prepare_read_banner %p\n", con
);
1593 con
->in_base_pos
= 0;
1596 static void prepare_read_connect(struct ceph_connection
*con
)
1598 dout("prepare_read_connect %p\n", con
);
1599 con
->in_base_pos
= 0;
1602 static void prepare_read_ack(struct ceph_connection
*con
)
1604 dout("prepare_read_ack %p\n", con
);
1605 con
->in_base_pos
= 0;
1608 static void prepare_read_seq(struct ceph_connection
*con
)
1610 dout("prepare_read_seq %p\n", con
);
1611 con
->in_base_pos
= 0;
1612 con
->in_tag
= CEPH_MSGR_TAG_SEQ
;
1615 static void prepare_read_tag(struct ceph_connection
*con
)
1617 dout("prepare_read_tag %p\n", con
);
1618 con
->in_base_pos
= 0;
1619 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1623 * Prepare to read a message.
1625 static int prepare_read_message(struct ceph_connection
*con
)
1627 dout("prepare_read_message %p\n", con
);
1628 BUG_ON(con
->in_msg
!= NULL
);
1629 con
->in_base_pos
= 0;
1630 con
->in_front_crc
= con
->in_middle_crc
= con
->in_data_crc
= 0;
1635 static int read_partial(struct ceph_connection
*con
,
1636 int end
, int size
, void *object
)
1638 while (con
->in_base_pos
< end
) {
1639 int left
= end
- con
->in_base_pos
;
1640 int have
= size
- left
;
1641 int ret
= ceph_tcp_recvmsg(con
->sock
, object
+ have
, left
);
1644 con
->in_base_pos
+= ret
;
1651 * Read all or part of the connect-side handshake on a new connection
1653 static int read_partial_banner(struct ceph_connection
*con
)
1659 dout("read_partial_banner %p at %d\n", con
, con
->in_base_pos
);
1662 size
= strlen(CEPH_BANNER
);
1664 ret
= read_partial(con
, end
, size
, con
->in_banner
);
1668 size
= sizeof (con
->actual_peer_addr
);
1670 ret
= read_partial(con
, end
, size
, &con
->actual_peer_addr
);
1674 size
= sizeof (con
->peer_addr_for_me
);
1676 ret
= read_partial(con
, end
, size
, &con
->peer_addr_for_me
);
1684 static int read_partial_connect(struct ceph_connection
*con
)
1690 dout("read_partial_connect %p at %d\n", con
, con
->in_base_pos
);
1692 size
= sizeof (con
->in_reply
);
1694 ret
= read_partial(con
, end
, size
, &con
->in_reply
);
1698 size
= le32_to_cpu(con
->in_reply
.authorizer_len
);
1700 ret
= read_partial(con
, end
, size
, con
->auth_reply_buf
);
1704 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1705 con
, (int)con
->in_reply
.tag
,
1706 le32_to_cpu(con
->in_reply
.connect_seq
),
1707 le32_to_cpu(con
->in_reply
.global_seq
));
1714 * Verify the hello banner looks okay.
1716 static int verify_hello(struct ceph_connection
*con
)
1718 if (memcmp(con
->in_banner
, CEPH_BANNER
, strlen(CEPH_BANNER
))) {
1719 pr_err("connect to %s got bad banner\n",
1720 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1721 con
->error_msg
= "protocol error, bad banner";
1727 static bool addr_is_blank(struct sockaddr_storage
*ss
)
1729 switch (ss
->ss_family
) {
1731 return ((struct sockaddr_in
*)ss
)->sin_addr
.s_addr
== 0;
1734 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[0] == 0 &&
1735 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[1] == 0 &&
1736 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[2] == 0 &&
1737 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[3] == 0;
1742 static int addr_port(struct sockaddr_storage
*ss
)
1744 switch (ss
->ss_family
) {
1746 return ntohs(((struct sockaddr_in
*)ss
)->sin_port
);
1748 return ntohs(((struct sockaddr_in6
*)ss
)->sin6_port
);
1753 static void addr_set_port(struct sockaddr_storage
*ss
, int p
)
1755 switch (ss
->ss_family
) {
1757 ((struct sockaddr_in
*)ss
)->sin_port
= htons(p
);
1760 ((struct sockaddr_in6
*)ss
)->sin6_port
= htons(p
);
1766 * Unlike other *_pton function semantics, zero indicates success.
1768 static int ceph_pton(const char *str
, size_t len
, struct sockaddr_storage
*ss
,
1769 char delim
, const char **ipend
)
1771 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
1772 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
1774 memset(ss
, 0, sizeof(*ss
));
1776 if (in4_pton(str
, len
, (u8
*)&in4
->sin_addr
.s_addr
, delim
, ipend
)) {
1777 ss
->ss_family
= AF_INET
;
1781 if (in6_pton(str
, len
, (u8
*)&in6
->sin6_addr
.s6_addr
, delim
, ipend
)) {
1782 ss
->ss_family
= AF_INET6
;
1790 * Extract hostname string and resolve using kernel DNS facility.
1792 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1793 static int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1794 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1796 const char *end
, *delim_p
;
1797 char *colon_p
, *ip_addr
= NULL
;
1801 * The end of the hostname occurs immediately preceding the delimiter or
1802 * the port marker (':') where the delimiter takes precedence.
1804 delim_p
= memchr(name
, delim
, namelen
);
1805 colon_p
= memchr(name
, ':', namelen
);
1807 if (delim_p
&& colon_p
)
1808 end
= delim_p
< colon_p
? delim_p
: colon_p
;
1809 else if (!delim_p
&& colon_p
)
1813 if (!end
) /* case: hostname:/ */
1814 end
= name
+ namelen
;
1820 /* do dns_resolve upcall */
1821 ip_len
= dns_query(NULL
, name
, end
- name
, NULL
, &ip_addr
, NULL
);
1823 ret
= ceph_pton(ip_addr
, ip_len
, ss
, -1, NULL
);
1831 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end
- name
), name
,
1832 ret
, ret
? "failed" : ceph_pr_addr(ss
));
1837 static inline int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1838 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1845 * Parse a server name (IP or hostname). If a valid IP address is not found
1846 * then try to extract a hostname to resolve using userspace DNS upcall.
1848 static int ceph_parse_server_name(const char *name
, size_t namelen
,
1849 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1853 ret
= ceph_pton(name
, namelen
, ss
, delim
, ipend
);
1855 ret
= ceph_dns_resolve_name(name
, namelen
, ss
, delim
, ipend
);
1861 * Parse an ip[:port] list into an addr array. Use the default
1862 * monitor port if a port isn't specified.
1864 int ceph_parse_ips(const char *c
, const char *end
,
1865 struct ceph_entity_addr
*addr
,
1866 int max_count
, int *count
)
1868 int i
, ret
= -EINVAL
;
1871 dout("parse_ips on '%.*s'\n", (int)(end
-c
), c
);
1872 for (i
= 0; i
< max_count
; i
++) {
1874 struct sockaddr_storage
*ss
= &addr
[i
].in_addr
;
1883 ret
= ceph_parse_server_name(p
, end
- p
, ss
, delim
, &ipend
);
1892 dout("missing matching ']'\n");
1899 if (p
< end
&& *p
== ':') {
1902 while (p
< end
&& *p
>= '0' && *p
<= '9') {
1903 port
= (port
* 10) + (*p
- '0');
1907 port
= CEPH_MON_PORT
;
1908 else if (port
> 65535)
1911 port
= CEPH_MON_PORT
;
1914 addr_set_port(ss
, port
);
1916 dout("parse_ips got %s\n", ceph_pr_addr(ss
));
1933 pr_err("parse_ips bad ip '%.*s'\n", (int)(end
- c
), c
);
1936 EXPORT_SYMBOL(ceph_parse_ips
);
1938 static int process_banner(struct ceph_connection
*con
)
1940 dout("process_banner on %p\n", con
);
1942 if (verify_hello(con
) < 0)
1945 ceph_decode_addr(&con
->actual_peer_addr
);
1946 ceph_decode_addr(&con
->peer_addr_for_me
);
1949 * Make sure the other end is who we wanted. note that the other
1950 * end may not yet know their ip address, so if it's 0.0.0.0, give
1951 * them the benefit of the doubt.
1953 if (memcmp(&con
->peer_addr
, &con
->actual_peer_addr
,
1954 sizeof(con
->peer_addr
)) != 0 &&
1955 !(addr_is_blank(&con
->actual_peer_addr
.in_addr
) &&
1956 con
->actual_peer_addr
.nonce
== con
->peer_addr
.nonce
)) {
1957 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
1958 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1959 (int)le32_to_cpu(con
->peer_addr
.nonce
),
1960 ceph_pr_addr(&con
->actual_peer_addr
.in_addr
),
1961 (int)le32_to_cpu(con
->actual_peer_addr
.nonce
));
1962 con
->error_msg
= "wrong peer at address";
1967 * did we learn our address?
1969 if (addr_is_blank(&con
->msgr
->inst
.addr
.in_addr
)) {
1970 int port
= addr_port(&con
->msgr
->inst
.addr
.in_addr
);
1972 memcpy(&con
->msgr
->inst
.addr
.in_addr
,
1973 &con
->peer_addr_for_me
.in_addr
,
1974 sizeof(con
->peer_addr_for_me
.in_addr
));
1975 addr_set_port(&con
->msgr
->inst
.addr
.in_addr
, port
);
1976 encode_my_addr(con
->msgr
);
1977 dout("process_banner learned my addr is %s\n",
1978 ceph_pr_addr(&con
->msgr
->inst
.addr
.in_addr
));
1984 static int process_connect(struct ceph_connection
*con
)
1986 u64 sup_feat
= con
->msgr
->supported_features
;
1987 u64 req_feat
= con
->msgr
->required_features
;
1988 u64 server_feat
= ceph_sanitize_features(
1989 le64_to_cpu(con
->in_reply
.features
));
1992 dout("process_connect on %p tag %d\n", con
, (int)con
->in_tag
);
1994 switch (con
->in_reply
.tag
) {
1995 case CEPH_MSGR_TAG_FEATURES
:
1996 pr_err("%s%lld %s feature set mismatch,"
1997 " my %llx < server's %llx, missing %llx\n",
1998 ENTITY_NAME(con
->peer_name
),
1999 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2000 sup_feat
, server_feat
, server_feat
& ~sup_feat
);
2001 con
->error_msg
= "missing required protocol features";
2002 reset_connection(con
);
2005 case CEPH_MSGR_TAG_BADPROTOVER
:
2006 pr_err("%s%lld %s protocol version mismatch,"
2007 " my %d != server's %d\n",
2008 ENTITY_NAME(con
->peer_name
),
2009 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2010 le32_to_cpu(con
->out_connect
.protocol_version
),
2011 le32_to_cpu(con
->in_reply
.protocol_version
));
2012 con
->error_msg
= "protocol version mismatch";
2013 reset_connection(con
);
2016 case CEPH_MSGR_TAG_BADAUTHORIZER
:
2018 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con
,
2020 if (con
->auth_retry
== 2) {
2021 con
->error_msg
= "connect authorization failure";
2024 con_out_kvec_reset(con
);
2025 ret
= prepare_write_connect(con
);
2028 prepare_read_connect(con
);
2031 case CEPH_MSGR_TAG_RESETSESSION
:
2033 * If we connected with a large connect_seq but the peer
2034 * has no record of a session with us (no connection, or
2035 * connect_seq == 0), they will send RESETSESION to indicate
2036 * that they must have reset their session, and may have
2039 dout("process_connect got RESET peer seq %u\n",
2040 le32_to_cpu(con
->in_reply
.connect_seq
));
2041 pr_err("%s%lld %s connection reset\n",
2042 ENTITY_NAME(con
->peer_name
),
2043 ceph_pr_addr(&con
->peer_addr
.in_addr
));
2044 reset_connection(con
);
2045 con_out_kvec_reset(con
);
2046 ret
= prepare_write_connect(con
);
2049 prepare_read_connect(con
);
2051 /* Tell ceph about it. */
2052 mutex_unlock(&con
->mutex
);
2053 pr_info("reset on %s%lld\n", ENTITY_NAME(con
->peer_name
));
2054 if (con
->ops
->peer_reset
)
2055 con
->ops
->peer_reset(con
);
2056 mutex_lock(&con
->mutex
);
2057 if (con
->state
!= CON_STATE_NEGOTIATING
)
2061 case CEPH_MSGR_TAG_RETRY_SESSION
:
2063 * If we sent a smaller connect_seq than the peer has, try
2064 * again with a larger value.
2066 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2067 le32_to_cpu(con
->out_connect
.connect_seq
),
2068 le32_to_cpu(con
->in_reply
.connect_seq
));
2069 con
->connect_seq
= le32_to_cpu(con
->in_reply
.connect_seq
);
2070 con_out_kvec_reset(con
);
2071 ret
= prepare_write_connect(con
);
2074 prepare_read_connect(con
);
2077 case CEPH_MSGR_TAG_RETRY_GLOBAL
:
2079 * If we sent a smaller global_seq than the peer has, try
2080 * again with a larger value.
2082 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2083 con
->peer_global_seq
,
2084 le32_to_cpu(con
->in_reply
.global_seq
));
2085 get_global_seq(con
->msgr
,
2086 le32_to_cpu(con
->in_reply
.global_seq
));
2087 con_out_kvec_reset(con
);
2088 ret
= prepare_write_connect(con
);
2091 prepare_read_connect(con
);
2094 case CEPH_MSGR_TAG_SEQ
:
2095 case CEPH_MSGR_TAG_READY
:
2096 if (req_feat
& ~server_feat
) {
2097 pr_err("%s%lld %s protocol feature mismatch,"
2098 " my required %llx > server's %llx, need %llx\n",
2099 ENTITY_NAME(con
->peer_name
),
2100 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2101 req_feat
, server_feat
, req_feat
& ~server_feat
);
2102 con
->error_msg
= "missing required protocol features";
2103 reset_connection(con
);
2107 WARN_ON(con
->state
!= CON_STATE_NEGOTIATING
);
2108 con
->state
= CON_STATE_OPEN
;
2109 con
->auth_retry
= 0; /* we authenticated; clear flag */
2110 con
->peer_global_seq
= le32_to_cpu(con
->in_reply
.global_seq
);
2112 con
->peer_features
= server_feat
;
2113 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2114 con
->peer_global_seq
,
2115 le32_to_cpu(con
->in_reply
.connect_seq
),
2117 WARN_ON(con
->connect_seq
!=
2118 le32_to_cpu(con
->in_reply
.connect_seq
));
2120 if (con
->in_reply
.flags
& CEPH_MSG_CONNECT_LOSSY
)
2121 con_flag_set(con
, CON_FLAG_LOSSYTX
);
2123 con
->delay
= 0; /* reset backoff memory */
2125 if (con
->in_reply
.tag
== CEPH_MSGR_TAG_SEQ
) {
2126 prepare_write_seq(con
);
2127 prepare_read_seq(con
);
2129 prepare_read_tag(con
);
2133 case CEPH_MSGR_TAG_WAIT
:
2135 * If there is a connection race (we are opening
2136 * connections to each other), one of us may just have
2137 * to WAIT. This shouldn't happen if we are the
2140 pr_err("process_connect got WAIT as client\n");
2141 con
->error_msg
= "protocol error, got WAIT as client";
2145 pr_err("connect protocol error, will retry\n");
2146 con
->error_msg
= "protocol error, garbage tag during connect";
2154 * read (part of) an ack
2156 static int read_partial_ack(struct ceph_connection
*con
)
2158 int size
= sizeof (con
->in_temp_ack
);
2161 return read_partial(con
, end
, size
, &con
->in_temp_ack
);
2165 * We can finally discard anything that's been acked.
2167 static void process_ack(struct ceph_connection
*con
)
2170 u64 ack
= le64_to_cpu(con
->in_temp_ack
);
2173 while (!list_empty(&con
->out_sent
)) {
2174 m
= list_first_entry(&con
->out_sent
, struct ceph_msg
,
2176 seq
= le64_to_cpu(m
->hdr
.seq
);
2179 dout("got ack for seq %llu type %d at %p\n", seq
,
2180 le16_to_cpu(m
->hdr
.type
), m
);
2181 m
->ack_stamp
= jiffies
;
2184 prepare_read_tag(con
);
2188 static int read_partial_message_section(struct ceph_connection
*con
,
2189 struct kvec
*section
,
2190 unsigned int sec_len
, u32
*crc
)
2196 while (section
->iov_len
< sec_len
) {
2197 BUG_ON(section
->iov_base
== NULL
);
2198 left
= sec_len
- section
->iov_len
;
2199 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)section
->iov_base
+
2200 section
->iov_len
, left
);
2203 section
->iov_len
+= ret
;
2205 if (section
->iov_len
== sec_len
)
2206 *crc
= crc32c(0, section
->iov_base
, section
->iov_len
);
2211 static int read_partial_msg_data(struct ceph_connection
*con
)
2213 struct ceph_msg
*msg
= con
->in_msg
;
2214 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
2215 const bool do_datacrc
= !con
->msgr
->nocrc
;
2223 if (list_empty(&msg
->data
))
2227 crc
= con
->in_data_crc
;
2228 while (cursor
->resid
) {
2229 page
= ceph_msg_data_next(&msg
->cursor
, &page_offset
, &length
,
2231 ret
= ceph_tcp_recvpage(con
->sock
, page
, page_offset
, length
);
2234 con
->in_data_crc
= crc
;
2240 crc
= ceph_crc32c_page(crc
, page
, page_offset
, ret
);
2241 (void) ceph_msg_data_advance(&msg
->cursor
, (size_t)ret
);
2244 con
->in_data_crc
= crc
;
2246 return 1; /* must return > 0 to indicate success */
2250 * read (part of) a message.
2252 static int ceph_con_in_msg_alloc(struct ceph_connection
*con
, int *skip
);
2254 static int read_partial_message(struct ceph_connection
*con
)
2256 struct ceph_msg
*m
= con
->in_msg
;
2260 unsigned int front_len
, middle_len
, data_len
;
2261 bool do_datacrc
= !con
->msgr
->nocrc
;
2262 bool need_sign
= (con
->peer_features
& CEPH_FEATURE_MSG_AUTH
);
2266 dout("read_partial_message con %p msg %p\n", con
, m
);
2269 size
= sizeof (con
->in_hdr
);
2271 ret
= read_partial(con
, end
, size
, &con
->in_hdr
);
2275 crc
= crc32c(0, &con
->in_hdr
, offsetof(struct ceph_msg_header
, crc
));
2276 if (cpu_to_le32(crc
) != con
->in_hdr
.crc
) {
2277 pr_err("read_partial_message bad hdr "
2278 " crc %u != expected %u\n",
2279 crc
, con
->in_hdr
.crc
);
2283 front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
2284 if (front_len
> CEPH_MSG_MAX_FRONT_LEN
)
2286 middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
2287 if (middle_len
> CEPH_MSG_MAX_MIDDLE_LEN
)
2289 data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
2290 if (data_len
> CEPH_MSG_MAX_DATA_LEN
)
2294 seq
= le64_to_cpu(con
->in_hdr
.seq
);
2295 if ((s64
)seq
- (s64
)con
->in_seq
< 1) {
2296 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2297 ENTITY_NAME(con
->peer_name
),
2298 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2299 seq
, con
->in_seq
+ 1);
2300 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
2302 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2304 } else if ((s64
)seq
- (s64
)con
->in_seq
> 1) {
2305 pr_err("read_partial_message bad seq %lld expected %lld\n",
2306 seq
, con
->in_seq
+ 1);
2307 con
->error_msg
= "bad message sequence # for incoming message";
2311 /* allocate message? */
2315 dout("got hdr type %d front %d data %d\n", con
->in_hdr
.type
,
2316 front_len
, data_len
);
2317 ret
= ceph_con_in_msg_alloc(con
, &skip
);
2321 BUG_ON(!con
->in_msg
^ skip
);
2322 if (con
->in_msg
&& data_len
> con
->in_msg
->data_length
) {
2323 pr_warn("%s skipping long message (%u > %zd)\n",
2324 __func__
, data_len
, con
->in_msg
->data_length
);
2325 ceph_msg_put(con
->in_msg
);
2330 /* skip this message */
2331 dout("alloc_msg said skip message\n");
2332 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
2334 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2339 BUG_ON(!con
->in_msg
);
2340 BUG_ON(con
->in_msg
->con
!= con
);
2342 m
->front
.iov_len
= 0; /* haven't read it yet */
2344 m
->middle
->vec
.iov_len
= 0;
2346 /* prepare for data payload, if any */
2349 prepare_message_data(con
->in_msg
, data_len
);
2353 ret
= read_partial_message_section(con
, &m
->front
, front_len
,
2354 &con
->in_front_crc
);
2360 ret
= read_partial_message_section(con
, &m
->middle
->vec
,
2362 &con
->in_middle_crc
);
2369 ret
= read_partial_msg_data(con
);
2376 size
= sizeof(m
->footer
);
2378 size
= sizeof(m
->old_footer
);
2381 ret
= read_partial(con
, end
, size
, &m
->footer
);
2386 m
->footer
.flags
= m
->old_footer
.flags
;
2390 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2391 m
, front_len
, m
->footer
.front_crc
, middle_len
,
2392 m
->footer
.middle_crc
, data_len
, m
->footer
.data_crc
);
2395 if (con
->in_front_crc
!= le32_to_cpu(m
->footer
.front_crc
)) {
2396 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2397 m
, con
->in_front_crc
, m
->footer
.front_crc
);
2400 if (con
->in_middle_crc
!= le32_to_cpu(m
->footer
.middle_crc
)) {
2401 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2402 m
, con
->in_middle_crc
, m
->footer
.middle_crc
);
2406 (m
->footer
.flags
& CEPH_MSG_FOOTER_NOCRC
) == 0 &&
2407 con
->in_data_crc
!= le32_to_cpu(m
->footer
.data_crc
)) {
2408 pr_err("read_partial_message %p data crc %u != exp. %u\n", m
,
2409 con
->in_data_crc
, le32_to_cpu(m
->footer
.data_crc
));
2413 if (need_sign
&& con
->ops
->check_message_signature
&&
2414 con
->ops
->check_message_signature(con
, m
)) {
2415 pr_err("read_partial_message %p signature check failed\n", m
);
2419 return 1; /* done! */
2423 * Process message. This happens in the worker thread. The callback should
2424 * be careful not to do anything that waits on other incoming messages or it
2427 static void process_message(struct ceph_connection
*con
)
2429 struct ceph_msg
*msg
;
2431 BUG_ON(con
->in_msg
->con
!= con
);
2432 con
->in_msg
->con
= NULL
;
2437 /* if first message, set peer_name */
2438 if (con
->peer_name
.type
== 0)
2439 con
->peer_name
= msg
->hdr
.src
;
2442 mutex_unlock(&con
->mutex
);
2444 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2445 msg
, le64_to_cpu(msg
->hdr
.seq
),
2446 ENTITY_NAME(msg
->hdr
.src
),
2447 le16_to_cpu(msg
->hdr
.type
),
2448 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2449 le32_to_cpu(msg
->hdr
.front_len
),
2450 le32_to_cpu(msg
->hdr
.data_len
),
2451 con
->in_front_crc
, con
->in_middle_crc
, con
->in_data_crc
);
2452 con
->ops
->dispatch(con
, msg
);
2454 mutex_lock(&con
->mutex
);
2459 * Write something to the socket. Called in a worker thread when the
2460 * socket appears to be writeable and we have something ready to send.
2462 static int try_write(struct ceph_connection
*con
)
2466 dout("try_write start %p state %lu\n", con
, con
->state
);
2469 dout("try_write out_kvec_bytes %d\n", con
->out_kvec_bytes
);
2471 /* open the socket first? */
2472 if (con
->state
== CON_STATE_PREOPEN
) {
2474 con
->state
= CON_STATE_CONNECTING
;
2476 con_out_kvec_reset(con
);
2477 prepare_write_banner(con
);
2478 prepare_read_banner(con
);
2480 BUG_ON(con
->in_msg
);
2481 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2482 dout("try_write initiating connect on %p new state %lu\n",
2484 ret
= ceph_tcp_connect(con
);
2486 con
->error_msg
= "connect error";
2492 /* kvec data queued? */
2493 if (con
->out_skip
) {
2494 ret
= write_partial_skip(con
);
2498 if (con
->out_kvec_left
) {
2499 ret
= write_partial_kvec(con
);
2506 if (con
->out_msg_done
) {
2507 ceph_msg_put(con
->out_msg
);
2508 con
->out_msg
= NULL
; /* we're done with this one */
2512 ret
= write_partial_message_data(con
);
2514 goto more_kvec
; /* we need to send the footer, too! */
2518 dout("try_write write_partial_message_data err %d\n",
2525 if (con
->state
== CON_STATE_OPEN
) {
2526 /* is anything else pending? */
2527 if (!list_empty(&con
->out_queue
)) {
2528 prepare_write_message(con
);
2531 if (con
->in_seq
> con
->in_seq_acked
) {
2532 prepare_write_ack(con
);
2535 if (con_flag_test_and_clear(con
, CON_FLAG_KEEPALIVE_PENDING
)) {
2536 prepare_write_keepalive(con
);
2541 /* Nothing to do! */
2542 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
2543 dout("try_write nothing else to write.\n");
2546 dout("try_write done on %p ret %d\n", con
, ret
);
2553 * Read what we can from the socket.
2555 static int try_read(struct ceph_connection
*con
)
2560 dout("try_read start on %p state %lu\n", con
, con
->state
);
2561 if (con
->state
!= CON_STATE_CONNECTING
&&
2562 con
->state
!= CON_STATE_NEGOTIATING
&&
2563 con
->state
!= CON_STATE_OPEN
)
2568 dout("try_read tag %d in_base_pos %d\n", (int)con
->in_tag
,
2571 if (con
->state
== CON_STATE_CONNECTING
) {
2572 dout("try_read connecting\n");
2573 ret
= read_partial_banner(con
);
2576 ret
= process_banner(con
);
2580 con
->state
= CON_STATE_NEGOTIATING
;
2583 * Received banner is good, exchange connection info.
2584 * Do not reset out_kvec, as sending our banner raced
2585 * with receiving peer banner after connect completed.
2587 ret
= prepare_write_connect(con
);
2590 prepare_read_connect(con
);
2592 /* Send connection info before awaiting response */
2596 if (con
->state
== CON_STATE_NEGOTIATING
) {
2597 dout("try_read negotiating\n");
2598 ret
= read_partial_connect(con
);
2601 ret
= process_connect(con
);
2607 WARN_ON(con
->state
!= CON_STATE_OPEN
);
2609 if (con
->in_base_pos
< 0) {
2611 * skipping + discarding content.
2613 * FIXME: there must be a better way to do this!
2615 static char buf
[SKIP_BUF_SIZE
];
2616 int skip
= min((int) sizeof (buf
), -con
->in_base_pos
);
2618 dout("skipping %d / %d bytes\n", skip
, -con
->in_base_pos
);
2619 ret
= ceph_tcp_recvmsg(con
->sock
, buf
, skip
);
2622 con
->in_base_pos
+= ret
;
2623 if (con
->in_base_pos
)
2626 if (con
->in_tag
== CEPH_MSGR_TAG_READY
) {
2630 ret
= ceph_tcp_recvmsg(con
->sock
, &con
->in_tag
, 1);
2633 dout("try_read got tag %d\n", (int)con
->in_tag
);
2634 switch (con
->in_tag
) {
2635 case CEPH_MSGR_TAG_MSG
:
2636 prepare_read_message(con
);
2638 case CEPH_MSGR_TAG_ACK
:
2639 prepare_read_ack(con
);
2641 case CEPH_MSGR_TAG_CLOSE
:
2642 con_close_socket(con
);
2643 con
->state
= CON_STATE_CLOSED
;
2649 if (con
->in_tag
== CEPH_MSGR_TAG_MSG
) {
2650 ret
= read_partial_message(con
);
2654 con
->error_msg
= "bad crc";
2658 con
->error_msg
= "io error";
2663 if (con
->in_tag
== CEPH_MSGR_TAG_READY
)
2665 process_message(con
);
2666 if (con
->state
== CON_STATE_OPEN
)
2667 prepare_read_tag(con
);
2670 if (con
->in_tag
== CEPH_MSGR_TAG_ACK
||
2671 con
->in_tag
== CEPH_MSGR_TAG_SEQ
) {
2673 * the final handshake seq exchange is semantically
2674 * equivalent to an ACK
2676 ret
= read_partial_ack(con
);
2684 dout("try_read done on %p ret %d\n", con
, ret
);
2688 pr_err("try_read bad con->in_tag = %d\n", (int)con
->in_tag
);
2689 con
->error_msg
= "protocol error, garbage tag";
2696 * Atomically queue work on a connection after the specified delay.
2697 * Bump @con reference to avoid races with connection teardown.
2698 * Returns 0 if work was queued, or an error code otherwise.
2700 static int queue_con_delay(struct ceph_connection
*con
, unsigned long delay
)
2702 if (!con
->ops
->get(con
)) {
2703 dout("%s %p ref count 0\n", __func__
, con
);
2707 if (!queue_delayed_work(ceph_msgr_wq
, &con
->work
, delay
)) {
2708 dout("%s %p - already queued\n", __func__
, con
);
2713 dout("%s %p %lu\n", __func__
, con
, delay
);
2717 static void queue_con(struct ceph_connection
*con
)
2719 (void) queue_con_delay(con
, 0);
2722 static void cancel_con(struct ceph_connection
*con
)
2724 if (cancel_delayed_work(&con
->work
)) {
2725 dout("%s %p\n", __func__
, con
);
2730 static bool con_sock_closed(struct ceph_connection
*con
)
2732 if (!con_flag_test_and_clear(con
, CON_FLAG_SOCK_CLOSED
))
2736 case CON_STATE_ ## x: \
2737 con->error_msg = "socket closed (con state " #x ")"; \
2740 switch (con
->state
) {
2748 pr_warn("%s con %p unrecognized state %lu\n",
2749 __func__
, con
, con
->state
);
2750 con
->error_msg
= "unrecognized con state";
2759 static bool con_backoff(struct ceph_connection
*con
)
2763 if (!con_flag_test_and_clear(con
, CON_FLAG_BACKOFF
))
2766 ret
= queue_con_delay(con
, round_jiffies_relative(con
->delay
));
2768 dout("%s: con %p FAILED to back off %lu\n", __func__
,
2770 BUG_ON(ret
== -ENOENT
);
2771 con_flag_set(con
, CON_FLAG_BACKOFF
);
2777 /* Finish fault handling; con->mutex must *not* be held here */
2779 static void con_fault_finish(struct ceph_connection
*con
)
2782 * in case we faulted due to authentication, invalidate our
2783 * current tickets so that we can get new ones.
2785 if (con
->auth_retry
&& con
->ops
->invalidate_authorizer
) {
2786 dout("calling invalidate_authorizer()\n");
2787 con
->ops
->invalidate_authorizer(con
);
2790 if (con
->ops
->fault
)
2791 con
->ops
->fault(con
);
2795 * Do some work on a connection. Drop a connection ref when we're done.
2797 static void con_work(struct work_struct
*work
)
2799 struct ceph_connection
*con
= container_of(work
, struct ceph_connection
,
2801 unsigned long pflags
= current
->flags
;
2804 current
->flags
|= PF_MEMALLOC
;
2806 mutex_lock(&con
->mutex
);
2810 if ((fault
= con_sock_closed(con
))) {
2811 dout("%s: con %p SOCK_CLOSED\n", __func__
, con
);
2814 if (con_backoff(con
)) {
2815 dout("%s: con %p BACKOFF\n", __func__
, con
);
2818 if (con
->state
== CON_STATE_STANDBY
) {
2819 dout("%s: con %p STANDBY\n", __func__
, con
);
2822 if (con
->state
== CON_STATE_CLOSED
) {
2823 dout("%s: con %p CLOSED\n", __func__
, con
);
2827 if (con
->state
== CON_STATE_PREOPEN
) {
2828 dout("%s: con %p PREOPEN\n", __func__
, con
);
2832 ret
= try_read(con
);
2836 con
->error_msg
= "socket error on read";
2841 ret
= try_write(con
);
2845 con
->error_msg
= "socket error on write";
2849 break; /* If we make it to here, we're done */
2853 mutex_unlock(&con
->mutex
);
2856 con_fault_finish(con
);
2860 tsk_restore_flags(current
, pflags
, PF_MEMALLOC
);
2864 * Generic error/fault handler. A retry mechanism is used with
2865 * exponential backoff
2867 static void con_fault(struct ceph_connection
*con
)
2869 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con
->peer_name
),
2870 ceph_pr_addr(&con
->peer_addr
.in_addr
), con
->error_msg
);
2871 dout("fault %p state %lu to peer %s\n",
2872 con
, con
->state
, ceph_pr_addr(&con
->peer_addr
.in_addr
));
2874 WARN_ON(con
->state
!= CON_STATE_CONNECTING
&&
2875 con
->state
!= CON_STATE_NEGOTIATING
&&
2876 con
->state
!= CON_STATE_OPEN
);
2878 con_close_socket(con
);
2880 if (con_flag_test(con
, CON_FLAG_LOSSYTX
)) {
2881 dout("fault on LOSSYTX channel, marking CLOSED\n");
2882 con
->state
= CON_STATE_CLOSED
;
2887 BUG_ON(con
->in_msg
->con
!= con
);
2888 con
->in_msg
->con
= NULL
;
2889 ceph_msg_put(con
->in_msg
);
2894 /* Requeue anything that hasn't been acked */
2895 list_splice_init(&con
->out_sent
, &con
->out_queue
);
2897 /* If there are no messages queued or keepalive pending, place
2898 * the connection in a STANDBY state */
2899 if (list_empty(&con
->out_queue
) &&
2900 !con_flag_test(con
, CON_FLAG_KEEPALIVE_PENDING
)) {
2901 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con
);
2902 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
2903 con
->state
= CON_STATE_STANDBY
;
2905 /* retry after a delay. */
2906 con
->state
= CON_STATE_PREOPEN
;
2907 if (con
->delay
== 0)
2908 con
->delay
= BASE_DELAY_INTERVAL
;
2909 else if (con
->delay
< MAX_DELAY_INTERVAL
)
2911 con_flag_set(con
, CON_FLAG_BACKOFF
);
2919 * initialize a new messenger instance
2921 void ceph_messenger_init(struct ceph_messenger
*msgr
,
2922 struct ceph_entity_addr
*myaddr
,
2923 u64 supported_features
,
2924 u64 required_features
,
2927 msgr
->supported_features
= supported_features
;
2928 msgr
->required_features
= required_features
;
2930 spin_lock_init(&msgr
->global_seq_lock
);
2933 msgr
->inst
.addr
= *myaddr
;
2935 /* select a random nonce */
2936 msgr
->inst
.addr
.type
= 0;
2937 get_random_bytes(&msgr
->inst
.addr
.nonce
, sizeof(msgr
->inst
.addr
.nonce
));
2938 encode_my_addr(msgr
);
2939 msgr
->nocrc
= nocrc
;
2941 atomic_set(&msgr
->stopping
, 0);
2943 dout("%s %p\n", __func__
, msgr
);
2945 EXPORT_SYMBOL(ceph_messenger_init
);
2947 static void clear_standby(struct ceph_connection
*con
)
2949 /* come back from STANDBY? */
2950 if (con
->state
== CON_STATE_STANDBY
) {
2951 dout("clear_standby %p and ++connect_seq\n", con
);
2952 con
->state
= CON_STATE_PREOPEN
;
2954 WARN_ON(con_flag_test(con
, CON_FLAG_WRITE_PENDING
));
2955 WARN_ON(con_flag_test(con
, CON_FLAG_KEEPALIVE_PENDING
));
2960 * Queue up an outgoing message on the given connection.
2962 void ceph_con_send(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2965 msg
->hdr
.src
= con
->msgr
->inst
.name
;
2966 BUG_ON(msg
->front
.iov_len
!= le32_to_cpu(msg
->hdr
.front_len
));
2967 msg
->needs_out_seq
= true;
2969 mutex_lock(&con
->mutex
);
2971 if (con
->state
== CON_STATE_CLOSED
) {
2972 dout("con_send %p closed, dropping %p\n", con
, msg
);
2974 mutex_unlock(&con
->mutex
);
2978 BUG_ON(msg
->con
!= NULL
);
2979 msg
->con
= con
->ops
->get(con
);
2980 BUG_ON(msg
->con
== NULL
);
2982 BUG_ON(!list_empty(&msg
->list_head
));
2983 list_add_tail(&msg
->list_head
, &con
->out_queue
);
2984 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg
,
2985 ENTITY_NAME(con
->peer_name
), le16_to_cpu(msg
->hdr
.type
),
2986 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2987 le32_to_cpu(msg
->hdr
.front_len
),
2988 le32_to_cpu(msg
->hdr
.middle_len
),
2989 le32_to_cpu(msg
->hdr
.data_len
));
2992 mutex_unlock(&con
->mutex
);
2994 /* if there wasn't anything waiting to send before, queue
2996 if (con_flag_test_and_set(con
, CON_FLAG_WRITE_PENDING
) == 0)
2999 EXPORT_SYMBOL(ceph_con_send
);
3002 * Revoke a message that was previously queued for send
3004 void ceph_msg_revoke(struct ceph_msg
*msg
)
3006 struct ceph_connection
*con
= msg
->con
;
3009 return; /* Message not in our possession */
3011 mutex_lock(&con
->mutex
);
3012 if (!list_empty(&msg
->list_head
)) {
3013 dout("%s %p msg %p - was on queue\n", __func__
, con
, msg
);
3014 list_del_init(&msg
->list_head
);
3015 BUG_ON(msg
->con
== NULL
);
3016 msg
->con
->ops
->put(msg
->con
);
3022 if (con
->out_msg
== msg
) {
3023 dout("%s %p msg %p - was sending\n", __func__
, con
, msg
);
3024 con
->out_msg
= NULL
;
3025 if (con
->out_kvec_is_msg
) {
3026 con
->out_skip
= con
->out_kvec_bytes
;
3027 con
->out_kvec_is_msg
= false;
3033 mutex_unlock(&con
->mutex
);
3037 * Revoke a message that we may be reading data into
3039 void ceph_msg_revoke_incoming(struct ceph_msg
*msg
)
3041 struct ceph_connection
*con
;
3043 BUG_ON(msg
== NULL
);
3045 dout("%s msg %p null con\n", __func__
, msg
);
3047 return; /* Message not in our possession */
3051 mutex_lock(&con
->mutex
);
3052 if (con
->in_msg
== msg
) {
3053 unsigned int front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
3054 unsigned int middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
3055 unsigned int data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
3057 /* skip rest of message */
3058 dout("%s %p msg %p revoked\n", __func__
, con
, msg
);
3059 con
->in_base_pos
= con
->in_base_pos
-
3060 sizeof(struct ceph_msg_header
) -
3064 sizeof(struct ceph_msg_footer
);
3065 ceph_msg_put(con
->in_msg
);
3067 con
->in_tag
= CEPH_MSGR_TAG_READY
;
3070 dout("%s %p in_msg %p msg %p no-op\n",
3071 __func__
, con
, con
->in_msg
, msg
);
3073 mutex_unlock(&con
->mutex
);
3077 * Queue a keepalive byte to ensure the tcp connection is alive.
3079 void ceph_con_keepalive(struct ceph_connection
*con
)
3081 dout("con_keepalive %p\n", con
);
3082 mutex_lock(&con
->mutex
);
3084 mutex_unlock(&con
->mutex
);
3085 if (con_flag_test_and_set(con
, CON_FLAG_KEEPALIVE_PENDING
) == 0 &&
3086 con_flag_test_and_set(con
, CON_FLAG_WRITE_PENDING
) == 0)
3089 EXPORT_SYMBOL(ceph_con_keepalive
);
3091 static struct ceph_msg_data
*ceph_msg_data_create(enum ceph_msg_data_type type
)
3093 struct ceph_msg_data
*data
;
3095 if (WARN_ON(!ceph_msg_data_type_valid(type
)))
3098 data
= kmem_cache_zalloc(ceph_msg_data_cache
, GFP_NOFS
);
3101 INIT_LIST_HEAD(&data
->links
);
3106 static void ceph_msg_data_destroy(struct ceph_msg_data
*data
)
3111 WARN_ON(!list_empty(&data
->links
));
3112 if (data
->type
== CEPH_MSG_DATA_PAGELIST
)
3113 ceph_pagelist_release(data
->pagelist
);
3114 kmem_cache_free(ceph_msg_data_cache
, data
);
3117 void ceph_msg_data_add_pages(struct ceph_msg
*msg
, struct page
**pages
,
3118 size_t length
, size_t alignment
)
3120 struct ceph_msg_data
*data
;
3125 data
= ceph_msg_data_create(CEPH_MSG_DATA_PAGES
);
3127 data
->pages
= pages
;
3128 data
->length
= length
;
3129 data
->alignment
= alignment
& ~PAGE_MASK
;
3131 list_add_tail(&data
->links
, &msg
->data
);
3132 msg
->data_length
+= length
;
3134 EXPORT_SYMBOL(ceph_msg_data_add_pages
);
3136 void ceph_msg_data_add_pagelist(struct ceph_msg
*msg
,
3137 struct ceph_pagelist
*pagelist
)
3139 struct ceph_msg_data
*data
;
3142 BUG_ON(!pagelist
->length
);
3144 data
= ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST
);
3146 data
->pagelist
= pagelist
;
3148 list_add_tail(&data
->links
, &msg
->data
);
3149 msg
->data_length
+= pagelist
->length
;
3151 EXPORT_SYMBOL(ceph_msg_data_add_pagelist
);
3154 void ceph_msg_data_add_bio(struct ceph_msg
*msg
, struct bio
*bio
,
3157 struct ceph_msg_data
*data
;
3161 data
= ceph_msg_data_create(CEPH_MSG_DATA_BIO
);
3164 data
->bio_length
= length
;
3166 list_add_tail(&data
->links
, &msg
->data
);
3167 msg
->data_length
+= length
;
3169 EXPORT_SYMBOL(ceph_msg_data_add_bio
);
3170 #endif /* CONFIG_BLOCK */
3173 * construct a new message with given type, size
3174 * the new msg has a ref count of 1.
3176 struct ceph_msg
*ceph_msg_new(int type
, int front_len
, gfp_t flags
,
3181 m
= kmem_cache_zalloc(ceph_msg_cache
, flags
);
3185 m
->hdr
.type
= cpu_to_le16(type
);
3186 m
->hdr
.priority
= cpu_to_le16(CEPH_MSG_PRIO_DEFAULT
);
3187 m
->hdr
.front_len
= cpu_to_le32(front_len
);
3189 INIT_LIST_HEAD(&m
->list_head
);
3190 kref_init(&m
->kref
);
3191 INIT_LIST_HEAD(&m
->data
);
3195 m
->front
.iov_base
= ceph_kvmalloc(front_len
, flags
);
3196 if (m
->front
.iov_base
== NULL
) {
3197 dout("ceph_msg_new can't allocate %d bytes\n",
3202 m
->front
.iov_base
= NULL
;
3204 m
->front_alloc_len
= m
->front
.iov_len
= front_len
;
3206 dout("ceph_msg_new %p front %d\n", m
, front_len
);
3213 pr_err("msg_new can't create type %d front %d\n", type
,
3217 dout("msg_new can't create type %d front %d\n", type
,
3222 EXPORT_SYMBOL(ceph_msg_new
);
3225 * Allocate "middle" portion of a message, if it is needed and wasn't
3226 * allocated by alloc_msg. This allows us to read a small fixed-size
3227 * per-type header in the front and then gracefully fail (i.e.,
3228 * propagate the error to the caller based on info in the front) when
3229 * the middle is too large.
3231 static int ceph_alloc_middle(struct ceph_connection
*con
, struct ceph_msg
*msg
)
3233 int type
= le16_to_cpu(msg
->hdr
.type
);
3234 int middle_len
= le32_to_cpu(msg
->hdr
.middle_len
);
3236 dout("alloc_middle %p type %d %s middle_len %d\n", msg
, type
,
3237 ceph_msg_type_name(type
), middle_len
);
3238 BUG_ON(!middle_len
);
3239 BUG_ON(msg
->middle
);
3241 msg
->middle
= ceph_buffer_new(middle_len
, GFP_NOFS
);
3248 * Allocate a message for receiving an incoming message on a
3249 * connection, and save the result in con->in_msg. Uses the
3250 * connection's private alloc_msg op if available.
3252 * Returns 0 on success, or a negative error code.
3254 * On success, if we set *skip = 1:
3255 * - the next message should be skipped and ignored.
3256 * - con->in_msg == NULL
3257 * or if we set *skip = 0:
3258 * - con->in_msg is non-null.
3259 * On error (ENOMEM, EAGAIN, ...),
3260 * - con->in_msg == NULL
3262 static int ceph_con_in_msg_alloc(struct ceph_connection
*con
, int *skip
)
3264 struct ceph_msg_header
*hdr
= &con
->in_hdr
;
3265 int middle_len
= le32_to_cpu(hdr
->middle_len
);
3266 struct ceph_msg
*msg
;
3269 BUG_ON(con
->in_msg
!= NULL
);
3270 BUG_ON(!con
->ops
->alloc_msg
);
3272 mutex_unlock(&con
->mutex
);
3273 msg
= con
->ops
->alloc_msg(con
, hdr
, skip
);
3274 mutex_lock(&con
->mutex
);
3275 if (con
->state
!= CON_STATE_OPEN
) {
3283 con
->in_msg
->con
= con
->ops
->get(con
);
3284 BUG_ON(con
->in_msg
->con
== NULL
);
3287 * Null message pointer means either we should skip
3288 * this message or we couldn't allocate memory. The
3289 * former is not an error.
3293 con
->error_msg
= "error allocating memory for incoming message";
3297 memcpy(&con
->in_msg
->hdr
, &con
->in_hdr
, sizeof(con
->in_hdr
));
3299 if (middle_len
&& !con
->in_msg
->middle
) {
3300 ret
= ceph_alloc_middle(con
, con
->in_msg
);
3302 ceph_msg_put(con
->in_msg
);
3312 * Free a generically kmalloc'd message.
3314 static void ceph_msg_free(struct ceph_msg
*m
)
3316 dout("%s %p\n", __func__
, m
);
3317 kvfree(m
->front
.iov_base
);
3318 kmem_cache_free(ceph_msg_cache
, m
);
3321 static void ceph_msg_release(struct kref
*kref
)
3323 struct ceph_msg
*m
= container_of(kref
, struct ceph_msg
, kref
);
3325 struct list_head
*links
;
3326 struct list_head
*next
;
3328 dout("%s %p\n", __func__
, m
);
3329 WARN_ON(!list_empty(&m
->list_head
));
3331 /* drop middle, data, if any */
3333 ceph_buffer_put(m
->middle
);
3337 list_splice_init(&m
->data
, &data
);
3338 list_for_each_safe(links
, next
, &data
) {
3339 struct ceph_msg_data
*data
;
3341 data
= list_entry(links
, struct ceph_msg_data
, links
);
3342 list_del_init(links
);
3343 ceph_msg_data_destroy(data
);
3348 ceph_msgpool_put(m
->pool
, m
);
3353 struct ceph_msg
*ceph_msg_get(struct ceph_msg
*msg
)
3355 dout("%s %p (was %d)\n", __func__
, msg
,
3356 atomic_read(&msg
->kref
.refcount
));
3357 kref_get(&msg
->kref
);
3360 EXPORT_SYMBOL(ceph_msg_get
);
3362 void ceph_msg_put(struct ceph_msg
*msg
)
3364 dout("%s %p (was %d)\n", __func__
, msg
,
3365 atomic_read(&msg
->kref
.refcount
));
3366 kref_put(&msg
->kref
, ceph_msg_release
);
3368 EXPORT_SYMBOL(ceph_msg_put
);
3370 void ceph_msg_dump(struct ceph_msg
*msg
)
3372 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg
,
3373 msg
->front_alloc_len
, msg
->data_length
);
3374 print_hex_dump(KERN_DEBUG
, "header: ",
3375 DUMP_PREFIX_OFFSET
, 16, 1,
3376 &msg
->hdr
, sizeof(msg
->hdr
), true);
3377 print_hex_dump(KERN_DEBUG
, " front: ",
3378 DUMP_PREFIX_OFFSET
, 16, 1,
3379 msg
->front
.iov_base
, msg
->front
.iov_len
, true);
3381 print_hex_dump(KERN_DEBUG
, "middle: ",
3382 DUMP_PREFIX_OFFSET
, 16, 1,
3383 msg
->middle
->vec
.iov_base
,
3384 msg
->middle
->vec
.iov_len
, true);
3385 print_hex_dump(KERN_DEBUG
, "footer: ",
3386 DUMP_PREFIX_OFFSET
, 16, 1,
3387 &msg
->footer
, sizeof(msg
->footer
), true);
3389 EXPORT_SYMBOL(ceph_msg_dump
);