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/nsproxy.h>
10 #include <linux/slab.h>
11 #include <linux/socket.h>
12 #include <linux/string.h>
14 #include <linux/bio.h>
15 #endif /* CONFIG_BLOCK */
16 #include <linux/dns_resolver.h>
19 #include <linux/ceph/ceph_features.h>
20 #include <linux/ceph/libceph.h>
21 #include <linux/ceph/messenger.h>
22 #include <linux/ceph/decode.h>
23 #include <linux/ceph/pagelist.h>
24 #include <linux/export.h>
27 * Ceph uses the messenger to exchange ceph_msg messages with other
28 * hosts in the system. The messenger provides ordered and reliable
29 * delivery. We tolerate TCP disconnects by reconnecting (with
30 * exponential backoff) in the case of a fault (disconnection, bad
31 * crc, protocol error). Acks allow sent messages to be discarded by
36 * We track the state of the socket on a given connection using
37 * values defined below. The transition to a new socket state is
38 * handled by a function which verifies we aren't coming from an
42 * | NEW* | transient initial state
44 * | con_sock_state_init()
47 * | CLOSED | initialized, but no socket (and no
48 * ---------- TCP connection)
50 * | \ con_sock_state_connecting()
51 * | ----------------------
53 * + con_sock_state_closed() \
54 * |+--------------------------- \
57 * | | CLOSING | socket event; \ \
58 * | ----------- await close \ \
61 * | + con_sock_state_closing() \ |
63 * | / --------------- | |
66 * | / -----------------| CONNECTING | socket created, TCP
67 * | | / -------------- connect initiated
68 * | | | con_sock_state_connected()
71 * | CONNECTED | TCP connection established
74 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
77 #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
78 #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
79 #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
80 #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
81 #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
86 #define CON_STATE_CLOSED 1 /* -> PREOPEN */
87 #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
88 #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
89 #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
90 #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
91 #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
94 * ceph_connection flag bits
96 #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
97 * messages on errors */
98 #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
99 #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
100 #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
101 #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
103 static bool con_flag_valid(unsigned long con_flag
)
106 case CON_FLAG_LOSSYTX
:
107 case CON_FLAG_KEEPALIVE_PENDING
:
108 case CON_FLAG_WRITE_PENDING
:
109 case CON_FLAG_SOCK_CLOSED
:
110 case CON_FLAG_BACKOFF
:
117 static void con_flag_clear(struct ceph_connection
*con
, unsigned long con_flag
)
119 BUG_ON(!con_flag_valid(con_flag
));
121 clear_bit(con_flag
, &con
->flags
);
124 static void con_flag_set(struct ceph_connection
*con
, unsigned long con_flag
)
126 BUG_ON(!con_flag_valid(con_flag
));
128 set_bit(con_flag
, &con
->flags
);
131 static bool con_flag_test(struct ceph_connection
*con
, unsigned long con_flag
)
133 BUG_ON(!con_flag_valid(con_flag
));
135 return test_bit(con_flag
, &con
->flags
);
138 static bool con_flag_test_and_clear(struct ceph_connection
*con
,
139 unsigned long con_flag
)
141 BUG_ON(!con_flag_valid(con_flag
));
143 return test_and_clear_bit(con_flag
, &con
->flags
);
146 static bool con_flag_test_and_set(struct ceph_connection
*con
,
147 unsigned long con_flag
)
149 BUG_ON(!con_flag_valid(con_flag
));
151 return test_and_set_bit(con_flag
, &con
->flags
);
154 /* Slab caches for frequently-allocated structures */
156 static struct kmem_cache
*ceph_msg_cache
;
157 static struct kmem_cache
*ceph_msg_data_cache
;
159 /* static tag bytes (protocol control messages) */
160 static char tag_msg
= CEPH_MSGR_TAG_MSG
;
161 static char tag_ack
= CEPH_MSGR_TAG_ACK
;
162 static char tag_keepalive
= CEPH_MSGR_TAG_KEEPALIVE
;
163 static char tag_keepalive2
= CEPH_MSGR_TAG_KEEPALIVE2
;
165 #ifdef CONFIG_LOCKDEP
166 static struct lock_class_key socket_class
;
170 * When skipping (ignoring) a block of input we read it into a "skip
171 * buffer," which is this many bytes in size.
173 #define SKIP_BUF_SIZE 1024
175 static void queue_con(struct ceph_connection
*con
);
176 static void cancel_con(struct ceph_connection
*con
);
177 static void ceph_con_workfn(struct work_struct
*);
178 static void con_fault(struct ceph_connection
*con
);
181 * Nicely render a sockaddr as a string. An array of formatted
182 * strings is used, to approximate reentrancy.
184 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
185 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
186 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
187 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
189 static char addr_str
[ADDR_STR_COUNT
][MAX_ADDR_STR_LEN
];
190 static atomic_t addr_str_seq
= ATOMIC_INIT(0);
192 static struct page
*zero_page
; /* used in certain error cases */
194 const char *ceph_pr_addr(const struct sockaddr_storage
*ss
)
198 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
199 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
201 i
= atomic_inc_return(&addr_str_seq
) & ADDR_STR_COUNT_MASK
;
204 switch (ss
->ss_family
) {
206 snprintf(s
, MAX_ADDR_STR_LEN
, "%pI4:%hu", &in4
->sin_addr
,
207 ntohs(in4
->sin_port
));
211 snprintf(s
, MAX_ADDR_STR_LEN
, "[%pI6c]:%hu", &in6
->sin6_addr
,
212 ntohs(in6
->sin6_port
));
216 snprintf(s
, MAX_ADDR_STR_LEN
, "(unknown sockaddr family %hu)",
222 EXPORT_SYMBOL(ceph_pr_addr
);
224 static void encode_my_addr(struct ceph_messenger
*msgr
)
226 memcpy(&msgr
->my_enc_addr
, &msgr
->inst
.addr
, sizeof(msgr
->my_enc_addr
));
227 ceph_encode_addr(&msgr
->my_enc_addr
);
231 * work queue for all reading and writing to/from the socket.
233 static struct workqueue_struct
*ceph_msgr_wq
;
235 static int ceph_msgr_slab_init(void)
237 BUG_ON(ceph_msg_cache
);
238 ceph_msg_cache
= KMEM_CACHE(ceph_msg
, 0);
242 BUG_ON(ceph_msg_data_cache
);
243 ceph_msg_data_cache
= KMEM_CACHE(ceph_msg_data
, 0);
244 if (ceph_msg_data_cache
)
247 kmem_cache_destroy(ceph_msg_cache
);
248 ceph_msg_cache
= NULL
;
253 static void ceph_msgr_slab_exit(void)
255 BUG_ON(!ceph_msg_data_cache
);
256 kmem_cache_destroy(ceph_msg_data_cache
);
257 ceph_msg_data_cache
= NULL
;
259 BUG_ON(!ceph_msg_cache
);
260 kmem_cache_destroy(ceph_msg_cache
);
261 ceph_msg_cache
= NULL
;
264 static void _ceph_msgr_exit(void)
267 destroy_workqueue(ceph_msgr_wq
);
271 BUG_ON(zero_page
== NULL
);
275 ceph_msgr_slab_exit();
278 int ceph_msgr_init(void)
280 if (ceph_msgr_slab_init())
283 BUG_ON(zero_page
!= NULL
);
284 zero_page
= ZERO_PAGE(0);
288 * The number of active work items is limited by the number of
289 * connections, so leave @max_active at default.
291 ceph_msgr_wq
= alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM
, 0);
295 pr_err("msgr_init failed to create workqueue\n");
300 EXPORT_SYMBOL(ceph_msgr_init
);
302 void ceph_msgr_exit(void)
304 BUG_ON(ceph_msgr_wq
== NULL
);
308 EXPORT_SYMBOL(ceph_msgr_exit
);
310 void ceph_msgr_flush(void)
312 flush_workqueue(ceph_msgr_wq
);
314 EXPORT_SYMBOL(ceph_msgr_flush
);
316 /* Connection socket state transition functions */
318 static void con_sock_state_init(struct ceph_connection
*con
)
322 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSED
);
323 if (WARN_ON(old_state
!= CON_SOCK_STATE_NEW
))
324 printk("%s: unexpected old state %d\n", __func__
, old_state
);
325 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
326 CON_SOCK_STATE_CLOSED
);
329 static void con_sock_state_connecting(struct ceph_connection
*con
)
333 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CONNECTING
);
334 if (WARN_ON(old_state
!= CON_SOCK_STATE_CLOSED
))
335 printk("%s: unexpected old state %d\n", __func__
, old_state
);
336 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
337 CON_SOCK_STATE_CONNECTING
);
340 static void con_sock_state_connected(struct ceph_connection
*con
)
344 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CONNECTED
);
345 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTING
))
346 printk("%s: unexpected old state %d\n", __func__
, old_state
);
347 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
348 CON_SOCK_STATE_CONNECTED
);
351 static void con_sock_state_closing(struct ceph_connection
*con
)
355 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSING
);
356 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTING
&&
357 old_state
!= CON_SOCK_STATE_CONNECTED
&&
358 old_state
!= CON_SOCK_STATE_CLOSING
))
359 printk("%s: unexpected old state %d\n", __func__
, old_state
);
360 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
361 CON_SOCK_STATE_CLOSING
);
364 static void con_sock_state_closed(struct ceph_connection
*con
)
368 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSED
);
369 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTED
&&
370 old_state
!= CON_SOCK_STATE_CLOSING
&&
371 old_state
!= CON_SOCK_STATE_CONNECTING
&&
372 old_state
!= CON_SOCK_STATE_CLOSED
))
373 printk("%s: unexpected old state %d\n", __func__
, old_state
);
374 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
375 CON_SOCK_STATE_CLOSED
);
379 * socket callback functions
382 /* data available on socket, or listen socket received a connect */
383 static void ceph_sock_data_ready(struct sock
*sk
)
385 struct ceph_connection
*con
= sk
->sk_user_data
;
386 if (atomic_read(&con
->msgr
->stopping
)) {
390 if (sk
->sk_state
!= TCP_CLOSE_WAIT
) {
391 dout("%s on %p state = %lu, queueing work\n", __func__
,
397 /* socket has buffer space for writing */
398 static void ceph_sock_write_space(struct sock
*sk
)
400 struct ceph_connection
*con
= sk
->sk_user_data
;
402 /* only queue to workqueue if there is data we want to write,
403 * and there is sufficient space in the socket buffer to accept
404 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
405 * doesn't get called again until try_write() fills the socket
406 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
407 * and net/core/stream.c:sk_stream_write_space().
409 if (con_flag_test(con
, CON_FLAG_WRITE_PENDING
)) {
410 if (sk_stream_is_writeable(sk
)) {
411 dout("%s %p queueing write work\n", __func__
, con
);
412 clear_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
416 dout("%s %p nothing to write\n", __func__
, con
);
420 /* socket's state has changed */
421 static void ceph_sock_state_change(struct sock
*sk
)
423 struct ceph_connection
*con
= sk
->sk_user_data
;
425 dout("%s %p state = %lu sk_state = %u\n", __func__
,
426 con
, con
->state
, sk
->sk_state
);
428 switch (sk
->sk_state
) {
430 dout("%s TCP_CLOSE\n", __func__
);
432 dout("%s TCP_CLOSE_WAIT\n", __func__
);
433 con_sock_state_closing(con
);
434 con_flag_set(con
, CON_FLAG_SOCK_CLOSED
);
437 case TCP_ESTABLISHED
:
438 dout("%s TCP_ESTABLISHED\n", __func__
);
439 con_sock_state_connected(con
);
442 default: /* Everything else is uninteresting */
448 * set up socket callbacks
450 static void set_sock_callbacks(struct socket
*sock
,
451 struct ceph_connection
*con
)
453 struct sock
*sk
= sock
->sk
;
454 sk
->sk_user_data
= con
;
455 sk
->sk_data_ready
= ceph_sock_data_ready
;
456 sk
->sk_write_space
= ceph_sock_write_space
;
457 sk
->sk_state_change
= ceph_sock_state_change
;
466 * initiate connection to a remote socket.
468 static int ceph_tcp_connect(struct ceph_connection
*con
)
470 struct sockaddr_storage
*paddr
= &con
->peer_addr
.in_addr
;
475 ret
= sock_create_kern(read_pnet(&con
->msgr
->net
), paddr
->ss_family
,
476 SOCK_STREAM
, IPPROTO_TCP
, &sock
);
479 sock
->sk
->sk_allocation
= GFP_NOFS
;
481 #ifdef CONFIG_LOCKDEP
482 lockdep_set_class(&sock
->sk
->sk_lock
, &socket_class
);
485 set_sock_callbacks(sock
, con
);
487 dout("connect %s\n", ceph_pr_addr(&con
->peer_addr
.in_addr
));
489 con_sock_state_connecting(con
);
490 ret
= sock
->ops
->connect(sock
, (struct sockaddr
*)paddr
, sizeof(*paddr
),
492 if (ret
== -EINPROGRESS
) {
493 dout("connect %s EINPROGRESS sk_state = %u\n",
494 ceph_pr_addr(&con
->peer_addr
.in_addr
),
496 } else if (ret
< 0) {
497 pr_err("connect %s error %d\n",
498 ceph_pr_addr(&con
->peer_addr
.in_addr
), ret
);
503 if (ceph_test_opt(from_msgr(con
->msgr
), TCP_NODELAY
)) {
506 ret
= kernel_setsockopt(sock
, SOL_TCP
, TCP_NODELAY
,
507 (char *)&optval
, sizeof(optval
));
509 pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
517 static int ceph_tcp_recvmsg(struct socket
*sock
, void *buf
, size_t len
)
519 struct kvec iov
= {buf
, len
};
520 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
523 iov_iter_kvec(&msg
.msg_iter
, READ
| ITER_KVEC
, &iov
, 1, len
);
524 r
= sock_recvmsg(sock
, &msg
, msg
.msg_flags
);
530 static int ceph_tcp_recvpage(struct socket
*sock
, struct page
*page
,
531 int page_offset
, size_t length
)
533 struct bio_vec bvec
= {
535 .bv_offset
= page_offset
,
538 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
541 BUG_ON(page_offset
+ length
> PAGE_SIZE
);
542 iov_iter_bvec(&msg
.msg_iter
, READ
| ITER_BVEC
, &bvec
, 1, length
);
543 r
= sock_recvmsg(sock
, &msg
, msg
.msg_flags
);
550 * write something. @more is true if caller will be sending more data
553 static int ceph_tcp_sendmsg(struct socket
*sock
, struct kvec
*iov
,
554 size_t kvlen
, size_t len
, int more
)
556 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
560 msg
.msg_flags
|= MSG_MORE
;
562 msg
.msg_flags
|= MSG_EOR
; /* superfluous, but what the hell */
564 r
= kernel_sendmsg(sock
, &msg
, iov
, kvlen
, len
);
570 static int __ceph_tcp_sendpage(struct socket
*sock
, struct page
*page
,
571 int offset
, size_t size
, bool more
)
573 int flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
| (more
? MSG_MORE
: MSG_EOR
);
576 ret
= kernel_sendpage(sock
, page
, offset
, size
, flags
);
583 static int ceph_tcp_sendpage(struct socket
*sock
, struct page
*page
,
584 int offset
, size_t size
, bool more
)
586 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
590 /* sendpage cannot properly handle pages with page_count == 0,
591 * we need to fallback to sendmsg if that's the case */
592 if (page_count(page
) >= 1)
593 return __ceph_tcp_sendpage(sock
, page
, offset
, size
, more
);
596 bvec
.bv_offset
= offset
;
600 msg
.msg_flags
|= MSG_MORE
;
602 msg
.msg_flags
|= MSG_EOR
; /* superfluous, but what the hell */
604 iov_iter_bvec(&msg
.msg_iter
, WRITE
| ITER_BVEC
, &bvec
, 1, size
);
605 ret
= sock_sendmsg(sock
, &msg
);
613 * Shutdown/close the socket for the given connection.
615 static int con_close_socket(struct ceph_connection
*con
)
619 dout("con_close_socket on %p sock %p\n", con
, con
->sock
);
621 rc
= con
->sock
->ops
->shutdown(con
->sock
, SHUT_RDWR
);
622 sock_release(con
->sock
);
627 * Forcibly clear the SOCK_CLOSED flag. It gets set
628 * independent of the connection mutex, and we could have
629 * received a socket close event before we had the chance to
630 * shut the socket down.
632 con_flag_clear(con
, CON_FLAG_SOCK_CLOSED
);
634 con_sock_state_closed(con
);
639 * Reset a connection. Discard all incoming and outgoing messages
640 * and clear *_seq state.
642 static void ceph_msg_remove(struct ceph_msg
*msg
)
644 list_del_init(&msg
->list_head
);
648 static void ceph_msg_remove_list(struct list_head
*head
)
650 while (!list_empty(head
)) {
651 struct ceph_msg
*msg
= list_first_entry(head
, struct ceph_msg
,
653 ceph_msg_remove(msg
);
657 static void reset_connection(struct ceph_connection
*con
)
659 /* reset connection, out_queue, msg_ and connect_seq */
660 /* discard existing out_queue and msg_seq */
661 dout("reset_connection %p\n", con
);
662 ceph_msg_remove_list(&con
->out_queue
);
663 ceph_msg_remove_list(&con
->out_sent
);
666 BUG_ON(con
->in_msg
->con
!= con
);
667 ceph_msg_put(con
->in_msg
);
671 con
->connect_seq
= 0;
674 BUG_ON(con
->out_msg
->con
!= con
);
675 ceph_msg_put(con
->out_msg
);
679 con
->in_seq_acked
= 0;
685 * mark a peer down. drop any open connections.
687 void ceph_con_close(struct ceph_connection
*con
)
689 mutex_lock(&con
->mutex
);
690 dout("con_close %p peer %s\n", con
,
691 ceph_pr_addr(&con
->peer_addr
.in_addr
));
692 con
->state
= CON_STATE_CLOSED
;
694 con_flag_clear(con
, CON_FLAG_LOSSYTX
); /* so we retry next connect */
695 con_flag_clear(con
, CON_FLAG_KEEPALIVE_PENDING
);
696 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
697 con_flag_clear(con
, CON_FLAG_BACKOFF
);
699 reset_connection(con
);
700 con
->peer_global_seq
= 0;
702 con_close_socket(con
);
703 mutex_unlock(&con
->mutex
);
705 EXPORT_SYMBOL(ceph_con_close
);
708 * Reopen a closed connection, with a new peer address.
710 void ceph_con_open(struct ceph_connection
*con
,
711 __u8 entity_type
, __u64 entity_num
,
712 struct ceph_entity_addr
*addr
)
714 mutex_lock(&con
->mutex
);
715 dout("con_open %p %s\n", con
, ceph_pr_addr(&addr
->in_addr
));
717 WARN_ON(con
->state
!= CON_STATE_CLOSED
);
718 con
->state
= CON_STATE_PREOPEN
;
720 con
->peer_name
.type
= (__u8
) entity_type
;
721 con
->peer_name
.num
= cpu_to_le64(entity_num
);
723 memcpy(&con
->peer_addr
, addr
, sizeof(*addr
));
724 con
->delay
= 0; /* reset backoff memory */
725 mutex_unlock(&con
->mutex
);
728 EXPORT_SYMBOL(ceph_con_open
);
731 * return true if this connection ever successfully opened
733 bool ceph_con_opened(struct ceph_connection
*con
)
735 return con
->connect_seq
> 0;
739 * initialize a new connection.
741 void ceph_con_init(struct ceph_connection
*con
, void *private,
742 const struct ceph_connection_operations
*ops
,
743 struct ceph_messenger
*msgr
)
745 dout("con_init %p\n", con
);
746 memset(con
, 0, sizeof(*con
));
747 con
->private = private;
751 con_sock_state_init(con
);
753 mutex_init(&con
->mutex
);
754 INIT_LIST_HEAD(&con
->out_queue
);
755 INIT_LIST_HEAD(&con
->out_sent
);
756 INIT_DELAYED_WORK(&con
->work
, ceph_con_workfn
);
758 con
->state
= CON_STATE_CLOSED
;
760 EXPORT_SYMBOL(ceph_con_init
);
764 * We maintain a global counter to order connection attempts. Get
765 * a unique seq greater than @gt.
767 static u32
get_global_seq(struct ceph_messenger
*msgr
, u32 gt
)
771 spin_lock(&msgr
->global_seq_lock
);
772 if (msgr
->global_seq
< gt
)
773 msgr
->global_seq
= gt
;
774 ret
= ++msgr
->global_seq
;
775 spin_unlock(&msgr
->global_seq_lock
);
779 static void con_out_kvec_reset(struct ceph_connection
*con
)
781 BUG_ON(con
->out_skip
);
783 con
->out_kvec_left
= 0;
784 con
->out_kvec_bytes
= 0;
785 con
->out_kvec_cur
= &con
->out_kvec
[0];
788 static void con_out_kvec_add(struct ceph_connection
*con
,
789 size_t size
, void *data
)
791 int index
= con
->out_kvec_left
;
793 BUG_ON(con
->out_skip
);
794 BUG_ON(index
>= ARRAY_SIZE(con
->out_kvec
));
796 con
->out_kvec
[index
].iov_len
= size
;
797 con
->out_kvec
[index
].iov_base
= data
;
798 con
->out_kvec_left
++;
799 con
->out_kvec_bytes
+= size
;
803 * Chop off a kvec from the end. Return residual number of bytes for
804 * that kvec, i.e. how many bytes would have been written if the kvec
807 static int con_out_kvec_skip(struct ceph_connection
*con
)
809 int off
= con
->out_kvec_cur
- con
->out_kvec
;
812 if (con
->out_kvec_bytes
> 0) {
813 skip
= con
->out_kvec
[off
+ con
->out_kvec_left
- 1].iov_len
;
814 BUG_ON(con
->out_kvec_bytes
< skip
);
815 BUG_ON(!con
->out_kvec_left
);
816 con
->out_kvec_bytes
-= skip
;
817 con
->out_kvec_left
--;
826 * For a bio data item, a piece is whatever remains of the next
827 * entry in the current bio iovec, or the first entry in the next
830 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor
*cursor
,
833 struct ceph_msg_data
*data
= cursor
->data
;
836 BUG_ON(data
->type
!= CEPH_MSG_DATA_BIO
);
841 cursor
->resid
= min(length
, data
->bio_length
);
843 cursor
->bvec_iter
= bio
->bi_iter
;
845 cursor
->resid
<= bio_iter_len(bio
, cursor
->bvec_iter
);
848 static struct page
*ceph_msg_data_bio_next(struct ceph_msg_data_cursor
*cursor
,
852 struct ceph_msg_data
*data
= cursor
->data
;
854 struct bio_vec bio_vec
;
856 BUG_ON(data
->type
!= CEPH_MSG_DATA_BIO
);
861 bio_vec
= bio_iter_iovec(bio
, cursor
->bvec_iter
);
863 *page_offset
= (size_t) bio_vec
.bv_offset
;
864 BUG_ON(*page_offset
>= PAGE_SIZE
);
865 if (cursor
->last_piece
) /* pagelist offset is always 0 */
866 *length
= cursor
->resid
;
868 *length
= (size_t) bio_vec
.bv_len
;
869 BUG_ON(*length
> cursor
->resid
);
870 BUG_ON(*page_offset
+ *length
> PAGE_SIZE
);
872 return bio_vec
.bv_page
;
875 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor
*cursor
,
879 struct bio_vec bio_vec
;
881 BUG_ON(cursor
->data
->type
!= CEPH_MSG_DATA_BIO
);
886 bio_vec
= bio_iter_iovec(bio
, cursor
->bvec_iter
);
888 /* Advance the cursor offset */
890 BUG_ON(cursor
->resid
< bytes
);
891 cursor
->resid
-= bytes
;
893 bio_advance_iter(bio
, &cursor
->bvec_iter
, bytes
);
895 if (bytes
< bio_vec
.bv_len
)
896 return false; /* more bytes to process in this segment */
898 /* Move on to the next segment, and possibly the next bio */
900 if (!cursor
->bvec_iter
.bi_size
) {
904 cursor
->bvec_iter
= bio
->bi_iter
;
906 memset(&cursor
->bvec_iter
, 0,
907 sizeof(cursor
->bvec_iter
));
910 if (!cursor
->last_piece
) {
911 BUG_ON(!cursor
->resid
);
913 /* A short read is OK, so use <= rather than == */
914 if (cursor
->resid
<= bio_iter_len(bio
, cursor
->bvec_iter
))
915 cursor
->last_piece
= true;
920 #endif /* CONFIG_BLOCK */
923 * For a page array, a piece comes from the first page in the array
924 * that has not already been fully consumed.
926 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor
*cursor
,
929 struct ceph_msg_data
*data
= cursor
->data
;
932 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGES
);
934 BUG_ON(!data
->pages
);
935 BUG_ON(!data
->length
);
937 cursor
->resid
= min(length
, data
->length
);
938 page_count
= calc_pages_for(data
->alignment
, (u64
)data
->length
);
939 cursor
->page_offset
= data
->alignment
& ~PAGE_MASK
;
940 cursor
->page_index
= 0;
941 BUG_ON(page_count
> (int)USHRT_MAX
);
942 cursor
->page_count
= (unsigned short)page_count
;
943 BUG_ON(length
> SIZE_MAX
- cursor
->page_offset
);
944 cursor
->last_piece
= cursor
->page_offset
+ cursor
->resid
<= PAGE_SIZE
;
948 ceph_msg_data_pages_next(struct ceph_msg_data_cursor
*cursor
,
949 size_t *page_offset
, size_t *length
)
951 struct ceph_msg_data
*data
= cursor
->data
;
953 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGES
);
955 BUG_ON(cursor
->page_index
>= cursor
->page_count
);
956 BUG_ON(cursor
->page_offset
>= PAGE_SIZE
);
958 *page_offset
= cursor
->page_offset
;
959 if (cursor
->last_piece
)
960 *length
= cursor
->resid
;
962 *length
= PAGE_SIZE
- *page_offset
;
964 return data
->pages
[cursor
->page_index
];
967 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor
*cursor
,
970 BUG_ON(cursor
->data
->type
!= CEPH_MSG_DATA_PAGES
);
972 BUG_ON(cursor
->page_offset
+ bytes
> PAGE_SIZE
);
974 /* Advance the cursor page offset */
976 cursor
->resid
-= bytes
;
977 cursor
->page_offset
= (cursor
->page_offset
+ bytes
) & ~PAGE_MASK
;
978 if (!bytes
|| cursor
->page_offset
)
979 return false; /* more bytes to process in the current page */
982 return false; /* no more data */
984 /* Move on to the next page; offset is already at 0 */
986 BUG_ON(cursor
->page_index
>= cursor
->page_count
);
987 cursor
->page_index
++;
988 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
994 * For a pagelist, a piece is whatever remains to be consumed in the
995 * first page in the list, or the front of the next page.
998 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor
*cursor
,
1001 struct ceph_msg_data
*data
= cursor
->data
;
1002 struct ceph_pagelist
*pagelist
;
1005 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
1007 pagelist
= data
->pagelist
;
1011 return; /* pagelist can be assigned but empty */
1013 BUG_ON(list_empty(&pagelist
->head
));
1014 page
= list_first_entry(&pagelist
->head
, struct page
, lru
);
1016 cursor
->resid
= min(length
, pagelist
->length
);
1017 cursor
->page
= page
;
1019 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
1022 static struct page
*
1023 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor
*cursor
,
1024 size_t *page_offset
, size_t *length
)
1026 struct ceph_msg_data
*data
= cursor
->data
;
1027 struct ceph_pagelist
*pagelist
;
1029 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
1031 pagelist
= data
->pagelist
;
1034 BUG_ON(!cursor
->page
);
1035 BUG_ON(cursor
->offset
+ cursor
->resid
!= pagelist
->length
);
1037 /* offset of first page in pagelist is always 0 */
1038 *page_offset
= cursor
->offset
& ~PAGE_MASK
;
1039 if (cursor
->last_piece
)
1040 *length
= cursor
->resid
;
1042 *length
= PAGE_SIZE
- *page_offset
;
1044 return cursor
->page
;
1047 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor
*cursor
,
1050 struct ceph_msg_data
*data
= cursor
->data
;
1051 struct ceph_pagelist
*pagelist
;
1053 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
1055 pagelist
= data
->pagelist
;
1058 BUG_ON(cursor
->offset
+ cursor
->resid
!= pagelist
->length
);
1059 BUG_ON((cursor
->offset
& ~PAGE_MASK
) + bytes
> PAGE_SIZE
);
1061 /* Advance the cursor offset */
1063 cursor
->resid
-= bytes
;
1064 cursor
->offset
+= bytes
;
1065 /* offset of first page in pagelist is always 0 */
1066 if (!bytes
|| cursor
->offset
& ~PAGE_MASK
)
1067 return false; /* more bytes to process in the current page */
1070 return false; /* no more data */
1072 /* Move on to the next page */
1074 BUG_ON(list_is_last(&cursor
->page
->lru
, &pagelist
->head
));
1075 cursor
->page
= list_next_entry(cursor
->page
, lru
);
1076 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
1082 * Message data is handled (sent or received) in pieces, where each
1083 * piece resides on a single page. The network layer might not
1084 * consume an entire piece at once. A data item's cursor keeps
1085 * track of which piece is next to process and how much remains to
1086 * be processed in that piece. It also tracks whether the current
1087 * piece is the last one in the data item.
1089 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor
*cursor
)
1091 size_t length
= cursor
->total_resid
;
1093 switch (cursor
->data
->type
) {
1094 case CEPH_MSG_DATA_PAGELIST
:
1095 ceph_msg_data_pagelist_cursor_init(cursor
, length
);
1097 case CEPH_MSG_DATA_PAGES
:
1098 ceph_msg_data_pages_cursor_init(cursor
, length
);
1101 case CEPH_MSG_DATA_BIO
:
1102 ceph_msg_data_bio_cursor_init(cursor
, length
);
1104 #endif /* CONFIG_BLOCK */
1105 case CEPH_MSG_DATA_NONE
:
1110 cursor
->need_crc
= true;
1113 static void ceph_msg_data_cursor_init(struct ceph_msg
*msg
, size_t length
)
1115 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
1116 struct ceph_msg_data
*data
;
1119 BUG_ON(length
> msg
->data_length
);
1120 BUG_ON(list_empty(&msg
->data
));
1122 cursor
->data_head
= &msg
->data
;
1123 cursor
->total_resid
= length
;
1124 data
= list_first_entry(&msg
->data
, struct ceph_msg_data
, links
);
1125 cursor
->data
= data
;
1127 __ceph_msg_data_cursor_init(cursor
);
1131 * Return the page containing the next piece to process for a given
1132 * data item, and supply the page offset and length of that piece.
1133 * Indicate whether this is the last piece in this data item.
1135 static struct page
*ceph_msg_data_next(struct ceph_msg_data_cursor
*cursor
,
1136 size_t *page_offset
, size_t *length
,
1141 switch (cursor
->data
->type
) {
1142 case CEPH_MSG_DATA_PAGELIST
:
1143 page
= ceph_msg_data_pagelist_next(cursor
, page_offset
, length
);
1145 case CEPH_MSG_DATA_PAGES
:
1146 page
= ceph_msg_data_pages_next(cursor
, page_offset
, length
);
1149 case CEPH_MSG_DATA_BIO
:
1150 page
= ceph_msg_data_bio_next(cursor
, page_offset
, length
);
1152 #endif /* CONFIG_BLOCK */
1153 case CEPH_MSG_DATA_NONE
:
1159 BUG_ON(*page_offset
+ *length
> PAGE_SIZE
);
1162 *last_piece
= cursor
->last_piece
;
1168 * Returns true if the result moves the cursor on to the next piece
1171 static bool ceph_msg_data_advance(struct ceph_msg_data_cursor
*cursor
,
1176 BUG_ON(bytes
> cursor
->resid
);
1177 switch (cursor
->data
->type
) {
1178 case CEPH_MSG_DATA_PAGELIST
:
1179 new_piece
= ceph_msg_data_pagelist_advance(cursor
, bytes
);
1181 case CEPH_MSG_DATA_PAGES
:
1182 new_piece
= ceph_msg_data_pages_advance(cursor
, bytes
);
1185 case CEPH_MSG_DATA_BIO
:
1186 new_piece
= ceph_msg_data_bio_advance(cursor
, bytes
);
1188 #endif /* CONFIG_BLOCK */
1189 case CEPH_MSG_DATA_NONE
:
1194 cursor
->total_resid
-= bytes
;
1196 if (!cursor
->resid
&& cursor
->total_resid
) {
1197 WARN_ON(!cursor
->last_piece
);
1198 BUG_ON(list_is_last(&cursor
->data
->links
, cursor
->data_head
));
1199 cursor
->data
= list_next_entry(cursor
->data
, links
);
1200 __ceph_msg_data_cursor_init(cursor
);
1203 cursor
->need_crc
= new_piece
;
1208 static size_t sizeof_footer(struct ceph_connection
*con
)
1210 return (con
->peer_features
& CEPH_FEATURE_MSG_AUTH
) ?
1211 sizeof(struct ceph_msg_footer
) :
1212 sizeof(struct ceph_msg_footer_old
);
1215 static void prepare_message_data(struct ceph_msg
*msg
, u32 data_len
)
1220 /* Initialize data cursor */
1222 ceph_msg_data_cursor_init(msg
, (size_t)data_len
);
1226 * Prepare footer for currently outgoing message, and finish things
1227 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1229 static void prepare_write_message_footer(struct ceph_connection
*con
)
1231 struct ceph_msg
*m
= con
->out_msg
;
1233 m
->footer
.flags
|= CEPH_MSG_FOOTER_COMPLETE
;
1235 dout("prepare_write_message_footer %p\n", con
);
1236 con_out_kvec_add(con
, sizeof_footer(con
), &m
->footer
);
1237 if (con
->peer_features
& CEPH_FEATURE_MSG_AUTH
) {
1238 if (con
->ops
->sign_message
)
1239 con
->ops
->sign_message(m
);
1243 m
->old_footer
.flags
= m
->footer
.flags
;
1245 con
->out_more
= m
->more_to_follow
;
1246 con
->out_msg_done
= true;
1250 * Prepare headers for the next outgoing message.
1252 static void prepare_write_message(struct ceph_connection
*con
)
1257 con_out_kvec_reset(con
);
1258 con
->out_msg_done
= false;
1260 /* Sneak an ack in there first? If we can get it into the same
1261 * TCP packet that's a good thing. */
1262 if (con
->in_seq
> con
->in_seq_acked
) {
1263 con
->in_seq_acked
= con
->in_seq
;
1264 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
1265 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1266 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1267 &con
->out_temp_ack
);
1270 BUG_ON(list_empty(&con
->out_queue
));
1271 m
= list_first_entry(&con
->out_queue
, struct ceph_msg
, list_head
);
1273 BUG_ON(m
->con
!= con
);
1275 /* put message on sent list */
1277 list_move_tail(&m
->list_head
, &con
->out_sent
);
1280 * only assign outgoing seq # if we haven't sent this message
1281 * yet. if it is requeued, resend with it's original seq.
1283 if (m
->needs_out_seq
) {
1284 m
->hdr
.seq
= cpu_to_le64(++con
->out_seq
);
1285 m
->needs_out_seq
= false;
1287 WARN_ON(m
->data_length
!= le32_to_cpu(m
->hdr
.data_len
));
1289 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1290 m
, con
->out_seq
, le16_to_cpu(m
->hdr
.type
),
1291 le32_to_cpu(m
->hdr
.front_len
), le32_to_cpu(m
->hdr
.middle_len
),
1293 BUG_ON(le32_to_cpu(m
->hdr
.front_len
) != m
->front
.iov_len
);
1295 /* tag + hdr + front + middle */
1296 con_out_kvec_add(con
, sizeof (tag_msg
), &tag_msg
);
1297 con_out_kvec_add(con
, sizeof(con
->out_hdr
), &con
->out_hdr
);
1298 con_out_kvec_add(con
, m
->front
.iov_len
, m
->front
.iov_base
);
1301 con_out_kvec_add(con
, m
->middle
->vec
.iov_len
,
1302 m
->middle
->vec
.iov_base
);
1304 /* fill in hdr crc and finalize hdr */
1305 crc
= crc32c(0, &m
->hdr
, offsetof(struct ceph_msg_header
, crc
));
1306 con
->out_msg
->hdr
.crc
= cpu_to_le32(crc
);
1307 memcpy(&con
->out_hdr
, &con
->out_msg
->hdr
, sizeof(con
->out_hdr
));
1309 /* fill in front and middle crc, footer */
1310 crc
= crc32c(0, m
->front
.iov_base
, m
->front
.iov_len
);
1311 con
->out_msg
->footer
.front_crc
= cpu_to_le32(crc
);
1313 crc
= crc32c(0, m
->middle
->vec
.iov_base
,
1314 m
->middle
->vec
.iov_len
);
1315 con
->out_msg
->footer
.middle_crc
= cpu_to_le32(crc
);
1317 con
->out_msg
->footer
.middle_crc
= 0;
1318 dout("%s front_crc %u middle_crc %u\n", __func__
,
1319 le32_to_cpu(con
->out_msg
->footer
.front_crc
),
1320 le32_to_cpu(con
->out_msg
->footer
.middle_crc
));
1321 con
->out_msg
->footer
.flags
= 0;
1323 /* is there a data payload? */
1324 con
->out_msg
->footer
.data_crc
= 0;
1325 if (m
->data_length
) {
1326 prepare_message_data(con
->out_msg
, m
->data_length
);
1327 con
->out_more
= 1; /* data + footer will follow */
1329 /* no, queue up footer too and be done */
1330 prepare_write_message_footer(con
);
1333 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1339 static void prepare_write_ack(struct ceph_connection
*con
)
1341 dout("prepare_write_ack %p %llu -> %llu\n", con
,
1342 con
->in_seq_acked
, con
->in_seq
);
1343 con
->in_seq_acked
= con
->in_seq
;
1345 con_out_kvec_reset(con
);
1347 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
1349 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1350 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1351 &con
->out_temp_ack
);
1353 con
->out_more
= 1; /* more will follow.. eventually.. */
1354 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1358 * Prepare to share the seq during handshake
1360 static void prepare_write_seq(struct ceph_connection
*con
)
1362 dout("prepare_write_seq %p %llu -> %llu\n", con
,
1363 con
->in_seq_acked
, con
->in_seq
);
1364 con
->in_seq_acked
= con
->in_seq
;
1366 con_out_kvec_reset(con
);
1368 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1369 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1370 &con
->out_temp_ack
);
1372 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1376 * Prepare to write keepalive byte.
1378 static void prepare_write_keepalive(struct ceph_connection
*con
)
1380 dout("prepare_write_keepalive %p\n", con
);
1381 con_out_kvec_reset(con
);
1382 if (con
->peer_features
& CEPH_FEATURE_MSGR_KEEPALIVE2
) {
1383 struct timespec now
= CURRENT_TIME
;
1385 con_out_kvec_add(con
, sizeof(tag_keepalive2
), &tag_keepalive2
);
1386 ceph_encode_timespec(&con
->out_temp_keepalive2
, &now
);
1387 con_out_kvec_add(con
, sizeof(con
->out_temp_keepalive2
),
1388 &con
->out_temp_keepalive2
);
1390 con_out_kvec_add(con
, sizeof(tag_keepalive
), &tag_keepalive
);
1392 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1396 * Connection negotiation.
1399 static struct ceph_auth_handshake
*get_connect_authorizer(struct ceph_connection
*con
,
1402 struct ceph_auth_handshake
*auth
;
1404 if (!con
->ops
->get_authorizer
) {
1405 con
->out_connect
.authorizer_protocol
= CEPH_AUTH_UNKNOWN
;
1406 con
->out_connect
.authorizer_len
= 0;
1410 auth
= con
->ops
->get_authorizer(con
, auth_proto
, con
->auth_retry
);
1414 con
->auth_reply_buf
= auth
->authorizer_reply_buf
;
1415 con
->auth_reply_buf_len
= auth
->authorizer_reply_buf_len
;
1420 * We connected to a peer and are saying hello.
1422 static void prepare_write_banner(struct ceph_connection
*con
)
1424 con_out_kvec_add(con
, strlen(CEPH_BANNER
), CEPH_BANNER
);
1425 con_out_kvec_add(con
, sizeof (con
->msgr
->my_enc_addr
),
1426 &con
->msgr
->my_enc_addr
);
1429 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1432 static int prepare_write_connect(struct ceph_connection
*con
)
1434 unsigned int global_seq
= get_global_seq(con
->msgr
, 0);
1437 struct ceph_auth_handshake
*auth
;
1439 switch (con
->peer_name
.type
) {
1440 case CEPH_ENTITY_TYPE_MON
:
1441 proto
= CEPH_MONC_PROTOCOL
;
1443 case CEPH_ENTITY_TYPE_OSD
:
1444 proto
= CEPH_OSDC_PROTOCOL
;
1446 case CEPH_ENTITY_TYPE_MDS
:
1447 proto
= CEPH_MDSC_PROTOCOL
;
1453 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con
,
1454 con
->connect_seq
, global_seq
, proto
);
1456 con
->out_connect
.features
=
1457 cpu_to_le64(from_msgr(con
->msgr
)->supported_features
);
1458 con
->out_connect
.host_type
= cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT
);
1459 con
->out_connect
.connect_seq
= cpu_to_le32(con
->connect_seq
);
1460 con
->out_connect
.global_seq
= cpu_to_le32(global_seq
);
1461 con
->out_connect
.protocol_version
= cpu_to_le32(proto
);
1462 con
->out_connect
.flags
= 0;
1464 auth_proto
= CEPH_AUTH_UNKNOWN
;
1465 auth
= get_connect_authorizer(con
, &auth_proto
);
1467 return PTR_ERR(auth
);
1469 con
->out_connect
.authorizer_protocol
= cpu_to_le32(auth_proto
);
1470 con
->out_connect
.authorizer_len
= auth
?
1471 cpu_to_le32(auth
->authorizer_buf_len
) : 0;
1473 con_out_kvec_add(con
, sizeof (con
->out_connect
),
1475 if (auth
&& auth
->authorizer_buf_len
)
1476 con_out_kvec_add(con
, auth
->authorizer_buf_len
,
1477 auth
->authorizer_buf
);
1480 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1486 * write as much of pending kvecs to the socket as we can.
1488 * 0 -> socket full, but more to do
1491 static int write_partial_kvec(struct ceph_connection
*con
)
1495 dout("write_partial_kvec %p %d left\n", con
, con
->out_kvec_bytes
);
1496 while (con
->out_kvec_bytes
> 0) {
1497 ret
= ceph_tcp_sendmsg(con
->sock
, con
->out_kvec_cur
,
1498 con
->out_kvec_left
, con
->out_kvec_bytes
,
1502 con
->out_kvec_bytes
-= ret
;
1503 if (con
->out_kvec_bytes
== 0)
1506 /* account for full iov entries consumed */
1507 while (ret
>= con
->out_kvec_cur
->iov_len
) {
1508 BUG_ON(!con
->out_kvec_left
);
1509 ret
-= con
->out_kvec_cur
->iov_len
;
1510 con
->out_kvec_cur
++;
1511 con
->out_kvec_left
--;
1513 /* and for a partially-consumed entry */
1515 con
->out_kvec_cur
->iov_len
-= ret
;
1516 con
->out_kvec_cur
->iov_base
+= ret
;
1519 con
->out_kvec_left
= 0;
1522 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con
,
1523 con
->out_kvec_bytes
, con
->out_kvec_left
, ret
);
1524 return ret
; /* done! */
1527 static u32
ceph_crc32c_page(u32 crc
, struct page
*page
,
1528 unsigned int page_offset
,
1529 unsigned int length
)
1534 BUG_ON(kaddr
== NULL
);
1535 crc
= crc32c(crc
, kaddr
+ page_offset
, length
);
1541 * Write as much message data payload as we can. If we finish, queue
1543 * 1 -> done, footer is now queued in out_kvec[].
1544 * 0 -> socket full, but more to do
1547 static int write_partial_message_data(struct ceph_connection
*con
)
1549 struct ceph_msg
*msg
= con
->out_msg
;
1550 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
1551 bool do_datacrc
= !ceph_test_opt(from_msgr(con
->msgr
), NOCRC
);
1554 dout("%s %p msg %p\n", __func__
, con
, msg
);
1556 if (list_empty(&msg
->data
))
1560 * Iterate through each page that contains data to be
1561 * written, and send as much as possible for each.
1563 * If we are calculating the data crc (the default), we will
1564 * need to map the page. If we have no pages, they have
1565 * been revoked, so use the zero page.
1567 crc
= do_datacrc
? le32_to_cpu(msg
->footer
.data_crc
) : 0;
1568 while (cursor
->resid
) {
1576 page
= ceph_msg_data_next(cursor
, &page_offset
, &length
,
1578 ret
= ceph_tcp_sendpage(con
->sock
, page
, page_offset
,
1579 length
, !last_piece
);
1582 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1586 if (do_datacrc
&& cursor
->need_crc
)
1587 crc
= ceph_crc32c_page(crc
, page
, page_offset
, length
);
1588 need_crc
= ceph_msg_data_advance(cursor
, (size_t)ret
);
1591 dout("%s %p msg %p done\n", __func__
, con
, msg
);
1593 /* prepare and queue up footer, too */
1595 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1597 msg
->footer
.flags
|= CEPH_MSG_FOOTER_NOCRC
;
1598 con_out_kvec_reset(con
);
1599 prepare_write_message_footer(con
);
1601 return 1; /* must return > 0 to indicate success */
1607 static int write_partial_skip(struct ceph_connection
*con
)
1611 dout("%s %p %d left\n", __func__
, con
, con
->out_skip
);
1612 while (con
->out_skip
> 0) {
1613 size_t size
= min(con
->out_skip
, (int) PAGE_SIZE
);
1615 ret
= ceph_tcp_sendpage(con
->sock
, zero_page
, 0, size
, true);
1618 con
->out_skip
-= ret
;
1626 * Prepare to read connection handshake, or an ack.
1628 static void prepare_read_banner(struct ceph_connection
*con
)
1630 dout("prepare_read_banner %p\n", con
);
1631 con
->in_base_pos
= 0;
1634 static void prepare_read_connect(struct ceph_connection
*con
)
1636 dout("prepare_read_connect %p\n", con
);
1637 con
->in_base_pos
= 0;
1640 static void prepare_read_ack(struct ceph_connection
*con
)
1642 dout("prepare_read_ack %p\n", con
);
1643 con
->in_base_pos
= 0;
1646 static void prepare_read_seq(struct ceph_connection
*con
)
1648 dout("prepare_read_seq %p\n", con
);
1649 con
->in_base_pos
= 0;
1650 con
->in_tag
= CEPH_MSGR_TAG_SEQ
;
1653 static void prepare_read_tag(struct ceph_connection
*con
)
1655 dout("prepare_read_tag %p\n", con
);
1656 con
->in_base_pos
= 0;
1657 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1660 static void prepare_read_keepalive_ack(struct ceph_connection
*con
)
1662 dout("prepare_read_keepalive_ack %p\n", con
);
1663 con
->in_base_pos
= 0;
1667 * Prepare to read a message.
1669 static int prepare_read_message(struct ceph_connection
*con
)
1671 dout("prepare_read_message %p\n", con
);
1672 BUG_ON(con
->in_msg
!= NULL
);
1673 con
->in_base_pos
= 0;
1674 con
->in_front_crc
= con
->in_middle_crc
= con
->in_data_crc
= 0;
1679 static int read_partial(struct ceph_connection
*con
,
1680 int end
, int size
, void *object
)
1682 while (con
->in_base_pos
< end
) {
1683 int left
= end
- con
->in_base_pos
;
1684 int have
= size
- left
;
1685 int ret
= ceph_tcp_recvmsg(con
->sock
, object
+ have
, left
);
1688 con
->in_base_pos
+= ret
;
1695 * Read all or part of the connect-side handshake on a new connection
1697 static int read_partial_banner(struct ceph_connection
*con
)
1703 dout("read_partial_banner %p at %d\n", con
, con
->in_base_pos
);
1706 size
= strlen(CEPH_BANNER
);
1708 ret
= read_partial(con
, end
, size
, con
->in_banner
);
1712 size
= sizeof (con
->actual_peer_addr
);
1714 ret
= read_partial(con
, end
, size
, &con
->actual_peer_addr
);
1718 size
= sizeof (con
->peer_addr_for_me
);
1720 ret
= read_partial(con
, end
, size
, &con
->peer_addr_for_me
);
1728 static int read_partial_connect(struct ceph_connection
*con
)
1734 dout("read_partial_connect %p at %d\n", con
, con
->in_base_pos
);
1736 size
= sizeof (con
->in_reply
);
1738 ret
= read_partial(con
, end
, size
, &con
->in_reply
);
1742 size
= le32_to_cpu(con
->in_reply
.authorizer_len
);
1744 ret
= read_partial(con
, end
, size
, con
->auth_reply_buf
);
1748 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1749 con
, (int)con
->in_reply
.tag
,
1750 le32_to_cpu(con
->in_reply
.connect_seq
),
1751 le32_to_cpu(con
->in_reply
.global_seq
));
1758 * Verify the hello banner looks okay.
1760 static int verify_hello(struct ceph_connection
*con
)
1762 if (memcmp(con
->in_banner
, CEPH_BANNER
, strlen(CEPH_BANNER
))) {
1763 pr_err("connect to %s got bad banner\n",
1764 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1765 con
->error_msg
= "protocol error, bad banner";
1771 static bool addr_is_blank(struct sockaddr_storage
*ss
)
1773 struct in_addr
*addr
= &((struct sockaddr_in
*)ss
)->sin_addr
;
1774 struct in6_addr
*addr6
= &((struct sockaddr_in6
*)ss
)->sin6_addr
;
1776 switch (ss
->ss_family
) {
1778 return addr
->s_addr
== htonl(INADDR_ANY
);
1780 return ipv6_addr_any(addr6
);
1786 static int addr_port(struct sockaddr_storage
*ss
)
1788 switch (ss
->ss_family
) {
1790 return ntohs(((struct sockaddr_in
*)ss
)->sin_port
);
1792 return ntohs(((struct sockaddr_in6
*)ss
)->sin6_port
);
1797 static void addr_set_port(struct sockaddr_storage
*ss
, int p
)
1799 switch (ss
->ss_family
) {
1801 ((struct sockaddr_in
*)ss
)->sin_port
= htons(p
);
1804 ((struct sockaddr_in6
*)ss
)->sin6_port
= htons(p
);
1810 * Unlike other *_pton function semantics, zero indicates success.
1812 static int ceph_pton(const char *str
, size_t len
, struct sockaddr_storage
*ss
,
1813 char delim
, const char **ipend
)
1815 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
1816 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
1818 memset(ss
, 0, sizeof(*ss
));
1820 if (in4_pton(str
, len
, (u8
*)&in4
->sin_addr
.s_addr
, delim
, ipend
)) {
1821 ss
->ss_family
= AF_INET
;
1825 if (in6_pton(str
, len
, (u8
*)&in6
->sin6_addr
.s6_addr
, delim
, ipend
)) {
1826 ss
->ss_family
= AF_INET6
;
1834 * Extract hostname string and resolve using kernel DNS facility.
1836 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1837 static int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1838 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1840 const char *end
, *delim_p
;
1841 char *colon_p
, *ip_addr
= NULL
;
1845 * The end of the hostname occurs immediately preceding the delimiter or
1846 * the port marker (':') where the delimiter takes precedence.
1848 delim_p
= memchr(name
, delim
, namelen
);
1849 colon_p
= memchr(name
, ':', namelen
);
1851 if (delim_p
&& colon_p
)
1852 end
= delim_p
< colon_p
? delim_p
: colon_p
;
1853 else if (!delim_p
&& colon_p
)
1857 if (!end
) /* case: hostname:/ */
1858 end
= name
+ namelen
;
1864 /* do dns_resolve upcall */
1865 ip_len
= dns_query(NULL
, name
, end
- name
, NULL
, &ip_addr
, NULL
);
1867 ret
= ceph_pton(ip_addr
, ip_len
, ss
, -1, NULL
);
1875 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end
- name
), name
,
1876 ret
, ret
? "failed" : ceph_pr_addr(ss
));
1881 static inline int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1882 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1889 * Parse a server name (IP or hostname). If a valid IP address is not found
1890 * then try to extract a hostname to resolve using userspace DNS upcall.
1892 static int ceph_parse_server_name(const char *name
, size_t namelen
,
1893 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1897 ret
= ceph_pton(name
, namelen
, ss
, delim
, ipend
);
1899 ret
= ceph_dns_resolve_name(name
, namelen
, ss
, delim
, ipend
);
1905 * Parse an ip[:port] list into an addr array. Use the default
1906 * monitor port if a port isn't specified.
1908 int ceph_parse_ips(const char *c
, const char *end
,
1909 struct ceph_entity_addr
*addr
,
1910 int max_count
, int *count
)
1912 int i
, ret
= -EINVAL
;
1915 dout("parse_ips on '%.*s'\n", (int)(end
-c
), c
);
1916 for (i
= 0; i
< max_count
; i
++) {
1918 struct sockaddr_storage
*ss
= &addr
[i
].in_addr
;
1927 ret
= ceph_parse_server_name(p
, end
- p
, ss
, delim
, &ipend
);
1936 dout("missing matching ']'\n");
1943 if (p
< end
&& *p
== ':') {
1946 while (p
< end
&& *p
>= '0' && *p
<= '9') {
1947 port
= (port
* 10) + (*p
- '0');
1951 port
= CEPH_MON_PORT
;
1952 else if (port
> 65535)
1955 port
= CEPH_MON_PORT
;
1958 addr_set_port(ss
, port
);
1960 dout("parse_ips got %s\n", ceph_pr_addr(ss
));
1977 pr_err("parse_ips bad ip '%.*s'\n", (int)(end
- c
), c
);
1980 EXPORT_SYMBOL(ceph_parse_ips
);
1982 static int process_banner(struct ceph_connection
*con
)
1984 dout("process_banner on %p\n", con
);
1986 if (verify_hello(con
) < 0)
1989 ceph_decode_addr(&con
->actual_peer_addr
);
1990 ceph_decode_addr(&con
->peer_addr_for_me
);
1993 * Make sure the other end is who we wanted. note that the other
1994 * end may not yet know their ip address, so if it's 0.0.0.0, give
1995 * them the benefit of the doubt.
1997 if (memcmp(&con
->peer_addr
, &con
->actual_peer_addr
,
1998 sizeof(con
->peer_addr
)) != 0 &&
1999 !(addr_is_blank(&con
->actual_peer_addr
.in_addr
) &&
2000 con
->actual_peer_addr
.nonce
== con
->peer_addr
.nonce
)) {
2001 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
2002 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2003 (int)le32_to_cpu(con
->peer_addr
.nonce
),
2004 ceph_pr_addr(&con
->actual_peer_addr
.in_addr
),
2005 (int)le32_to_cpu(con
->actual_peer_addr
.nonce
));
2006 con
->error_msg
= "wrong peer at address";
2011 * did we learn our address?
2013 if (addr_is_blank(&con
->msgr
->inst
.addr
.in_addr
)) {
2014 int port
= addr_port(&con
->msgr
->inst
.addr
.in_addr
);
2016 memcpy(&con
->msgr
->inst
.addr
.in_addr
,
2017 &con
->peer_addr_for_me
.in_addr
,
2018 sizeof(con
->peer_addr_for_me
.in_addr
));
2019 addr_set_port(&con
->msgr
->inst
.addr
.in_addr
, port
);
2020 encode_my_addr(con
->msgr
);
2021 dout("process_banner learned my addr is %s\n",
2022 ceph_pr_addr(&con
->msgr
->inst
.addr
.in_addr
));
2028 static int process_connect(struct ceph_connection
*con
)
2030 u64 sup_feat
= from_msgr(con
->msgr
)->supported_features
;
2031 u64 req_feat
= from_msgr(con
->msgr
)->required_features
;
2032 u64 server_feat
= ceph_sanitize_features(
2033 le64_to_cpu(con
->in_reply
.features
));
2036 dout("process_connect on %p tag %d\n", con
, (int)con
->in_tag
);
2038 if (con
->auth_reply_buf
) {
2040 * Any connection that defines ->get_authorizer()
2041 * should also define ->verify_authorizer_reply().
2042 * See get_connect_authorizer().
2044 ret
= con
->ops
->verify_authorizer_reply(con
);
2046 con
->error_msg
= "bad authorize reply";
2051 switch (con
->in_reply
.tag
) {
2052 case CEPH_MSGR_TAG_FEATURES
:
2053 pr_err("%s%lld %s feature set mismatch,"
2054 " my %llx < server's %llx, missing %llx\n",
2055 ENTITY_NAME(con
->peer_name
),
2056 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2057 sup_feat
, server_feat
, server_feat
& ~sup_feat
);
2058 con
->error_msg
= "missing required protocol features";
2059 reset_connection(con
);
2062 case CEPH_MSGR_TAG_BADPROTOVER
:
2063 pr_err("%s%lld %s protocol version mismatch,"
2064 " my %d != server's %d\n",
2065 ENTITY_NAME(con
->peer_name
),
2066 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2067 le32_to_cpu(con
->out_connect
.protocol_version
),
2068 le32_to_cpu(con
->in_reply
.protocol_version
));
2069 con
->error_msg
= "protocol version mismatch";
2070 reset_connection(con
);
2073 case CEPH_MSGR_TAG_BADAUTHORIZER
:
2075 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con
,
2077 if (con
->auth_retry
== 2) {
2078 con
->error_msg
= "connect authorization failure";
2081 con_out_kvec_reset(con
);
2082 ret
= prepare_write_connect(con
);
2085 prepare_read_connect(con
);
2088 case CEPH_MSGR_TAG_RESETSESSION
:
2090 * If we connected with a large connect_seq but the peer
2091 * has no record of a session with us (no connection, or
2092 * connect_seq == 0), they will send RESETSESION to indicate
2093 * that they must have reset their session, and may have
2096 dout("process_connect got RESET peer seq %u\n",
2097 le32_to_cpu(con
->in_reply
.connect_seq
));
2098 pr_err("%s%lld %s connection reset\n",
2099 ENTITY_NAME(con
->peer_name
),
2100 ceph_pr_addr(&con
->peer_addr
.in_addr
));
2101 reset_connection(con
);
2102 con_out_kvec_reset(con
);
2103 ret
= prepare_write_connect(con
);
2106 prepare_read_connect(con
);
2108 /* Tell ceph about it. */
2109 mutex_unlock(&con
->mutex
);
2110 pr_info("reset on %s%lld\n", ENTITY_NAME(con
->peer_name
));
2111 if (con
->ops
->peer_reset
)
2112 con
->ops
->peer_reset(con
);
2113 mutex_lock(&con
->mutex
);
2114 if (con
->state
!= CON_STATE_NEGOTIATING
)
2118 case CEPH_MSGR_TAG_RETRY_SESSION
:
2120 * If we sent a smaller connect_seq than the peer has, try
2121 * again with a larger value.
2123 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2124 le32_to_cpu(con
->out_connect
.connect_seq
),
2125 le32_to_cpu(con
->in_reply
.connect_seq
));
2126 con
->connect_seq
= le32_to_cpu(con
->in_reply
.connect_seq
);
2127 con_out_kvec_reset(con
);
2128 ret
= prepare_write_connect(con
);
2131 prepare_read_connect(con
);
2134 case CEPH_MSGR_TAG_RETRY_GLOBAL
:
2136 * If we sent a smaller global_seq than the peer has, try
2137 * again with a larger value.
2139 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2140 con
->peer_global_seq
,
2141 le32_to_cpu(con
->in_reply
.global_seq
));
2142 get_global_seq(con
->msgr
,
2143 le32_to_cpu(con
->in_reply
.global_seq
));
2144 con_out_kvec_reset(con
);
2145 ret
= prepare_write_connect(con
);
2148 prepare_read_connect(con
);
2151 case CEPH_MSGR_TAG_SEQ
:
2152 case CEPH_MSGR_TAG_READY
:
2153 if (req_feat
& ~server_feat
) {
2154 pr_err("%s%lld %s protocol feature mismatch,"
2155 " my required %llx > server's %llx, need %llx\n",
2156 ENTITY_NAME(con
->peer_name
),
2157 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2158 req_feat
, server_feat
, req_feat
& ~server_feat
);
2159 con
->error_msg
= "missing required protocol features";
2160 reset_connection(con
);
2164 WARN_ON(con
->state
!= CON_STATE_NEGOTIATING
);
2165 con
->state
= CON_STATE_OPEN
;
2166 con
->auth_retry
= 0; /* we authenticated; clear flag */
2167 con
->peer_global_seq
= le32_to_cpu(con
->in_reply
.global_seq
);
2169 con
->peer_features
= server_feat
;
2170 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2171 con
->peer_global_seq
,
2172 le32_to_cpu(con
->in_reply
.connect_seq
),
2174 WARN_ON(con
->connect_seq
!=
2175 le32_to_cpu(con
->in_reply
.connect_seq
));
2177 if (con
->in_reply
.flags
& CEPH_MSG_CONNECT_LOSSY
)
2178 con_flag_set(con
, CON_FLAG_LOSSYTX
);
2180 con
->delay
= 0; /* reset backoff memory */
2182 if (con
->in_reply
.tag
== CEPH_MSGR_TAG_SEQ
) {
2183 prepare_write_seq(con
);
2184 prepare_read_seq(con
);
2186 prepare_read_tag(con
);
2190 case CEPH_MSGR_TAG_WAIT
:
2192 * If there is a connection race (we are opening
2193 * connections to each other), one of us may just have
2194 * to WAIT. This shouldn't happen if we are the
2197 con
->error_msg
= "protocol error, got WAIT as client";
2201 con
->error_msg
= "protocol error, garbage tag during connect";
2209 * read (part of) an ack
2211 static int read_partial_ack(struct ceph_connection
*con
)
2213 int size
= sizeof (con
->in_temp_ack
);
2216 return read_partial(con
, end
, size
, &con
->in_temp_ack
);
2220 * We can finally discard anything that's been acked.
2222 static void process_ack(struct ceph_connection
*con
)
2225 u64 ack
= le64_to_cpu(con
->in_temp_ack
);
2228 while (!list_empty(&con
->out_sent
)) {
2229 m
= list_first_entry(&con
->out_sent
, struct ceph_msg
,
2231 seq
= le64_to_cpu(m
->hdr
.seq
);
2234 dout("got ack for seq %llu type %d at %p\n", seq
,
2235 le16_to_cpu(m
->hdr
.type
), m
);
2236 m
->ack_stamp
= jiffies
;
2239 prepare_read_tag(con
);
2243 static int read_partial_message_section(struct ceph_connection
*con
,
2244 struct kvec
*section
,
2245 unsigned int sec_len
, u32
*crc
)
2251 while (section
->iov_len
< sec_len
) {
2252 BUG_ON(section
->iov_base
== NULL
);
2253 left
= sec_len
- section
->iov_len
;
2254 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)section
->iov_base
+
2255 section
->iov_len
, left
);
2258 section
->iov_len
+= ret
;
2260 if (section
->iov_len
== sec_len
)
2261 *crc
= crc32c(0, section
->iov_base
, section
->iov_len
);
2266 static int read_partial_msg_data(struct ceph_connection
*con
)
2268 struct ceph_msg
*msg
= con
->in_msg
;
2269 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
2270 bool do_datacrc
= !ceph_test_opt(from_msgr(con
->msgr
), NOCRC
);
2278 if (list_empty(&msg
->data
))
2282 crc
= con
->in_data_crc
;
2283 while (cursor
->resid
) {
2284 page
= ceph_msg_data_next(cursor
, &page_offset
, &length
, NULL
);
2285 ret
= ceph_tcp_recvpage(con
->sock
, page
, page_offset
, length
);
2288 con
->in_data_crc
= crc
;
2294 crc
= ceph_crc32c_page(crc
, page
, page_offset
, ret
);
2295 (void) ceph_msg_data_advance(cursor
, (size_t)ret
);
2298 con
->in_data_crc
= crc
;
2300 return 1; /* must return > 0 to indicate success */
2304 * read (part of) a message.
2306 static int ceph_con_in_msg_alloc(struct ceph_connection
*con
, int *skip
);
2308 static int read_partial_message(struct ceph_connection
*con
)
2310 struct ceph_msg
*m
= con
->in_msg
;
2314 unsigned int front_len
, middle_len
, data_len
;
2315 bool do_datacrc
= !ceph_test_opt(from_msgr(con
->msgr
), NOCRC
);
2316 bool need_sign
= (con
->peer_features
& CEPH_FEATURE_MSG_AUTH
);
2320 dout("read_partial_message con %p msg %p\n", con
, m
);
2323 size
= sizeof (con
->in_hdr
);
2325 ret
= read_partial(con
, end
, size
, &con
->in_hdr
);
2329 crc
= crc32c(0, &con
->in_hdr
, offsetof(struct ceph_msg_header
, crc
));
2330 if (cpu_to_le32(crc
) != con
->in_hdr
.crc
) {
2331 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
2332 crc
, con
->in_hdr
.crc
);
2336 front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
2337 if (front_len
> CEPH_MSG_MAX_FRONT_LEN
)
2339 middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
2340 if (middle_len
> CEPH_MSG_MAX_MIDDLE_LEN
)
2342 data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
2343 if (data_len
> CEPH_MSG_MAX_DATA_LEN
)
2347 seq
= le64_to_cpu(con
->in_hdr
.seq
);
2348 if ((s64
)seq
- (s64
)con
->in_seq
< 1) {
2349 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2350 ENTITY_NAME(con
->peer_name
),
2351 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2352 seq
, con
->in_seq
+ 1);
2353 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
2355 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2357 } else if ((s64
)seq
- (s64
)con
->in_seq
> 1) {
2358 pr_err("read_partial_message bad seq %lld expected %lld\n",
2359 seq
, con
->in_seq
+ 1);
2360 con
->error_msg
= "bad message sequence # for incoming message";
2364 /* allocate message? */
2368 dout("got hdr type %d front %d data %d\n", con
->in_hdr
.type
,
2369 front_len
, data_len
);
2370 ret
= ceph_con_in_msg_alloc(con
, &skip
);
2374 BUG_ON(!con
->in_msg
^ skip
);
2376 /* skip this message */
2377 dout("alloc_msg said skip message\n");
2378 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
2380 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2385 BUG_ON(!con
->in_msg
);
2386 BUG_ON(con
->in_msg
->con
!= con
);
2388 m
->front
.iov_len
= 0; /* haven't read it yet */
2390 m
->middle
->vec
.iov_len
= 0;
2392 /* prepare for data payload, if any */
2395 prepare_message_data(con
->in_msg
, data_len
);
2399 ret
= read_partial_message_section(con
, &m
->front
, front_len
,
2400 &con
->in_front_crc
);
2406 ret
= read_partial_message_section(con
, &m
->middle
->vec
,
2408 &con
->in_middle_crc
);
2415 ret
= read_partial_msg_data(con
);
2421 size
= sizeof_footer(con
);
2423 ret
= read_partial(con
, end
, size
, &m
->footer
);
2428 m
->footer
.flags
= m
->old_footer
.flags
;
2432 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2433 m
, front_len
, m
->footer
.front_crc
, middle_len
,
2434 m
->footer
.middle_crc
, data_len
, m
->footer
.data_crc
);
2437 if (con
->in_front_crc
!= le32_to_cpu(m
->footer
.front_crc
)) {
2438 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2439 m
, con
->in_front_crc
, m
->footer
.front_crc
);
2442 if (con
->in_middle_crc
!= le32_to_cpu(m
->footer
.middle_crc
)) {
2443 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2444 m
, con
->in_middle_crc
, m
->footer
.middle_crc
);
2448 (m
->footer
.flags
& CEPH_MSG_FOOTER_NOCRC
) == 0 &&
2449 con
->in_data_crc
!= le32_to_cpu(m
->footer
.data_crc
)) {
2450 pr_err("read_partial_message %p data crc %u != exp. %u\n", m
,
2451 con
->in_data_crc
, le32_to_cpu(m
->footer
.data_crc
));
2455 if (need_sign
&& con
->ops
->check_message_signature
&&
2456 con
->ops
->check_message_signature(m
)) {
2457 pr_err("read_partial_message %p signature check failed\n", m
);
2461 return 1; /* done! */
2465 * Process message. This happens in the worker thread. The callback should
2466 * be careful not to do anything that waits on other incoming messages or it
2469 static void process_message(struct ceph_connection
*con
)
2471 struct ceph_msg
*msg
= con
->in_msg
;
2473 BUG_ON(con
->in_msg
->con
!= con
);
2476 /* if first message, set peer_name */
2477 if (con
->peer_name
.type
== 0)
2478 con
->peer_name
= msg
->hdr
.src
;
2481 mutex_unlock(&con
->mutex
);
2483 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2484 msg
, le64_to_cpu(msg
->hdr
.seq
),
2485 ENTITY_NAME(msg
->hdr
.src
),
2486 le16_to_cpu(msg
->hdr
.type
),
2487 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2488 le32_to_cpu(msg
->hdr
.front_len
),
2489 le32_to_cpu(msg
->hdr
.data_len
),
2490 con
->in_front_crc
, con
->in_middle_crc
, con
->in_data_crc
);
2491 con
->ops
->dispatch(con
, msg
);
2493 mutex_lock(&con
->mutex
);
2496 static int read_keepalive_ack(struct ceph_connection
*con
)
2498 struct ceph_timespec ceph_ts
;
2499 size_t size
= sizeof(ceph_ts
);
2500 int ret
= read_partial(con
, size
, size
, &ceph_ts
);
2503 ceph_decode_timespec(&con
->last_keepalive_ack
, &ceph_ts
);
2504 prepare_read_tag(con
);
2509 * Write something to the socket. Called in a worker thread when the
2510 * socket appears to be writeable and we have something ready to send.
2512 static int try_write(struct ceph_connection
*con
)
2516 dout("try_write start %p state %lu\n", con
, con
->state
);
2519 dout("try_write out_kvec_bytes %d\n", con
->out_kvec_bytes
);
2521 /* open the socket first? */
2522 if (con
->state
== CON_STATE_PREOPEN
) {
2524 con
->state
= CON_STATE_CONNECTING
;
2526 con_out_kvec_reset(con
);
2527 prepare_write_banner(con
);
2528 prepare_read_banner(con
);
2530 BUG_ON(con
->in_msg
);
2531 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2532 dout("try_write initiating connect on %p new state %lu\n",
2534 ret
= ceph_tcp_connect(con
);
2536 con
->error_msg
= "connect error";
2542 /* kvec data queued? */
2543 if (con
->out_kvec_left
) {
2544 ret
= write_partial_kvec(con
);
2548 if (con
->out_skip
) {
2549 ret
= write_partial_skip(con
);
2556 if (con
->out_msg_done
) {
2557 ceph_msg_put(con
->out_msg
);
2558 con
->out_msg
= NULL
; /* we're done with this one */
2562 ret
= write_partial_message_data(con
);
2564 goto more_kvec
; /* we need to send the footer, too! */
2568 dout("try_write write_partial_message_data err %d\n",
2575 if (con
->state
== CON_STATE_OPEN
) {
2576 if (con_flag_test_and_clear(con
, CON_FLAG_KEEPALIVE_PENDING
)) {
2577 prepare_write_keepalive(con
);
2580 /* is anything else pending? */
2581 if (!list_empty(&con
->out_queue
)) {
2582 prepare_write_message(con
);
2585 if (con
->in_seq
> con
->in_seq_acked
) {
2586 prepare_write_ack(con
);
2591 /* Nothing to do! */
2592 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
2593 dout("try_write nothing else to write.\n");
2596 dout("try_write done on %p ret %d\n", con
, ret
);
2603 * Read what we can from the socket.
2605 static int try_read(struct ceph_connection
*con
)
2610 dout("try_read start on %p state %lu\n", con
, con
->state
);
2611 if (con
->state
!= CON_STATE_CONNECTING
&&
2612 con
->state
!= CON_STATE_NEGOTIATING
&&
2613 con
->state
!= CON_STATE_OPEN
)
2618 dout("try_read tag %d in_base_pos %d\n", (int)con
->in_tag
,
2621 if (con
->state
== CON_STATE_CONNECTING
) {
2622 dout("try_read connecting\n");
2623 ret
= read_partial_banner(con
);
2626 ret
= process_banner(con
);
2630 con
->state
= CON_STATE_NEGOTIATING
;
2633 * Received banner is good, exchange connection info.
2634 * Do not reset out_kvec, as sending our banner raced
2635 * with receiving peer banner after connect completed.
2637 ret
= prepare_write_connect(con
);
2640 prepare_read_connect(con
);
2642 /* Send connection info before awaiting response */
2646 if (con
->state
== CON_STATE_NEGOTIATING
) {
2647 dout("try_read negotiating\n");
2648 ret
= read_partial_connect(con
);
2651 ret
= process_connect(con
);
2657 WARN_ON(con
->state
!= CON_STATE_OPEN
);
2659 if (con
->in_base_pos
< 0) {
2661 * skipping + discarding content.
2663 * FIXME: there must be a better way to do this!
2665 static char buf
[SKIP_BUF_SIZE
];
2666 int skip
= min((int) sizeof (buf
), -con
->in_base_pos
);
2668 dout("skipping %d / %d bytes\n", skip
, -con
->in_base_pos
);
2669 ret
= ceph_tcp_recvmsg(con
->sock
, buf
, skip
);
2672 con
->in_base_pos
+= ret
;
2673 if (con
->in_base_pos
)
2676 if (con
->in_tag
== CEPH_MSGR_TAG_READY
) {
2680 ret
= ceph_tcp_recvmsg(con
->sock
, &con
->in_tag
, 1);
2683 dout("try_read got tag %d\n", (int)con
->in_tag
);
2684 switch (con
->in_tag
) {
2685 case CEPH_MSGR_TAG_MSG
:
2686 prepare_read_message(con
);
2688 case CEPH_MSGR_TAG_ACK
:
2689 prepare_read_ack(con
);
2691 case CEPH_MSGR_TAG_KEEPALIVE2_ACK
:
2692 prepare_read_keepalive_ack(con
);
2694 case CEPH_MSGR_TAG_CLOSE
:
2695 con_close_socket(con
);
2696 con
->state
= CON_STATE_CLOSED
;
2702 if (con
->in_tag
== CEPH_MSGR_TAG_MSG
) {
2703 ret
= read_partial_message(con
);
2707 con
->error_msg
= "bad crc/signature";
2713 con
->error_msg
= "io error";
2718 if (con
->in_tag
== CEPH_MSGR_TAG_READY
)
2720 process_message(con
);
2721 if (con
->state
== CON_STATE_OPEN
)
2722 prepare_read_tag(con
);
2725 if (con
->in_tag
== CEPH_MSGR_TAG_ACK
||
2726 con
->in_tag
== CEPH_MSGR_TAG_SEQ
) {
2728 * the final handshake seq exchange is semantically
2729 * equivalent to an ACK
2731 ret
= read_partial_ack(con
);
2737 if (con
->in_tag
== CEPH_MSGR_TAG_KEEPALIVE2_ACK
) {
2738 ret
= read_keepalive_ack(con
);
2745 dout("try_read done on %p ret %d\n", con
, ret
);
2749 pr_err("try_read bad con->in_tag = %d\n", (int)con
->in_tag
);
2750 con
->error_msg
= "protocol error, garbage tag";
2757 * Atomically queue work on a connection after the specified delay.
2758 * Bump @con reference to avoid races with connection teardown.
2759 * Returns 0 if work was queued, or an error code otherwise.
2761 static int queue_con_delay(struct ceph_connection
*con
, unsigned long delay
)
2763 if (!con
->ops
->get(con
)) {
2764 dout("%s %p ref count 0\n", __func__
, con
);
2768 if (!queue_delayed_work(ceph_msgr_wq
, &con
->work
, delay
)) {
2769 dout("%s %p - already queued\n", __func__
, con
);
2774 dout("%s %p %lu\n", __func__
, con
, delay
);
2778 static void queue_con(struct ceph_connection
*con
)
2780 (void) queue_con_delay(con
, 0);
2783 static void cancel_con(struct ceph_connection
*con
)
2785 if (cancel_delayed_work(&con
->work
)) {
2786 dout("%s %p\n", __func__
, con
);
2791 static bool con_sock_closed(struct ceph_connection
*con
)
2793 if (!con_flag_test_and_clear(con
, CON_FLAG_SOCK_CLOSED
))
2797 case CON_STATE_ ## x: \
2798 con->error_msg = "socket closed (con state " #x ")"; \
2801 switch (con
->state
) {
2809 pr_warn("%s con %p unrecognized state %lu\n",
2810 __func__
, con
, con
->state
);
2811 con
->error_msg
= "unrecognized con state";
2820 static bool con_backoff(struct ceph_connection
*con
)
2824 if (!con_flag_test_and_clear(con
, CON_FLAG_BACKOFF
))
2827 ret
= queue_con_delay(con
, round_jiffies_relative(con
->delay
));
2829 dout("%s: con %p FAILED to back off %lu\n", __func__
,
2831 BUG_ON(ret
== -ENOENT
);
2832 con_flag_set(con
, CON_FLAG_BACKOFF
);
2838 /* Finish fault handling; con->mutex must *not* be held here */
2840 static void con_fault_finish(struct ceph_connection
*con
)
2842 dout("%s %p\n", __func__
, con
);
2845 * in case we faulted due to authentication, invalidate our
2846 * current tickets so that we can get new ones.
2848 if (con
->auth_retry
) {
2849 dout("auth_retry %d, invalidating\n", con
->auth_retry
);
2850 if (con
->ops
->invalidate_authorizer
)
2851 con
->ops
->invalidate_authorizer(con
);
2852 con
->auth_retry
= 0;
2855 if (con
->ops
->fault
)
2856 con
->ops
->fault(con
);
2860 * Do some work on a connection. Drop a connection ref when we're done.
2862 static void ceph_con_workfn(struct work_struct
*work
)
2864 struct ceph_connection
*con
= container_of(work
, struct ceph_connection
,
2868 mutex_lock(&con
->mutex
);
2872 if ((fault
= con_sock_closed(con
))) {
2873 dout("%s: con %p SOCK_CLOSED\n", __func__
, con
);
2876 if (con_backoff(con
)) {
2877 dout("%s: con %p BACKOFF\n", __func__
, con
);
2880 if (con
->state
== CON_STATE_STANDBY
) {
2881 dout("%s: con %p STANDBY\n", __func__
, con
);
2884 if (con
->state
== CON_STATE_CLOSED
) {
2885 dout("%s: con %p CLOSED\n", __func__
, con
);
2889 if (con
->state
== CON_STATE_PREOPEN
) {
2890 dout("%s: con %p PREOPEN\n", __func__
, con
);
2894 ret
= try_read(con
);
2898 if (!con
->error_msg
)
2899 con
->error_msg
= "socket error on read";
2904 ret
= try_write(con
);
2908 if (!con
->error_msg
)
2909 con
->error_msg
= "socket error on write";
2913 break; /* If we make it to here, we're done */
2917 mutex_unlock(&con
->mutex
);
2920 con_fault_finish(con
);
2926 * Generic error/fault handler. A retry mechanism is used with
2927 * exponential backoff
2929 static void con_fault(struct ceph_connection
*con
)
2931 dout("fault %p state %lu to peer %s\n",
2932 con
, con
->state
, ceph_pr_addr(&con
->peer_addr
.in_addr
));
2934 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con
->peer_name
),
2935 ceph_pr_addr(&con
->peer_addr
.in_addr
), con
->error_msg
);
2936 con
->error_msg
= NULL
;
2938 WARN_ON(con
->state
!= CON_STATE_CONNECTING
&&
2939 con
->state
!= CON_STATE_NEGOTIATING
&&
2940 con
->state
!= CON_STATE_OPEN
);
2942 con_close_socket(con
);
2944 if (con_flag_test(con
, CON_FLAG_LOSSYTX
)) {
2945 dout("fault on LOSSYTX channel, marking CLOSED\n");
2946 con
->state
= CON_STATE_CLOSED
;
2951 BUG_ON(con
->in_msg
->con
!= con
);
2952 ceph_msg_put(con
->in_msg
);
2956 /* Requeue anything that hasn't been acked */
2957 list_splice_init(&con
->out_sent
, &con
->out_queue
);
2959 /* If there are no messages queued or keepalive pending, place
2960 * the connection in a STANDBY state */
2961 if (list_empty(&con
->out_queue
) &&
2962 !con_flag_test(con
, CON_FLAG_KEEPALIVE_PENDING
)) {
2963 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con
);
2964 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
2965 con
->state
= CON_STATE_STANDBY
;
2967 /* retry after a delay. */
2968 con
->state
= CON_STATE_PREOPEN
;
2969 if (con
->delay
== 0)
2970 con
->delay
= BASE_DELAY_INTERVAL
;
2971 else if (con
->delay
< MAX_DELAY_INTERVAL
)
2973 con_flag_set(con
, CON_FLAG_BACKOFF
);
2981 * initialize a new messenger instance
2983 void ceph_messenger_init(struct ceph_messenger
*msgr
,
2984 struct ceph_entity_addr
*myaddr
)
2986 spin_lock_init(&msgr
->global_seq_lock
);
2989 msgr
->inst
.addr
= *myaddr
;
2991 /* select a random nonce */
2992 msgr
->inst
.addr
.type
= 0;
2993 get_random_bytes(&msgr
->inst
.addr
.nonce
, sizeof(msgr
->inst
.addr
.nonce
));
2994 encode_my_addr(msgr
);
2996 atomic_set(&msgr
->stopping
, 0);
2997 write_pnet(&msgr
->net
, get_net(current
->nsproxy
->net_ns
));
2999 dout("%s %p\n", __func__
, msgr
);
3001 EXPORT_SYMBOL(ceph_messenger_init
);
3003 void ceph_messenger_fini(struct ceph_messenger
*msgr
)
3005 put_net(read_pnet(&msgr
->net
));
3007 EXPORT_SYMBOL(ceph_messenger_fini
);
3009 static void msg_con_set(struct ceph_msg
*msg
, struct ceph_connection
*con
)
3012 msg
->con
->ops
->put(msg
->con
);
3014 msg
->con
= con
? con
->ops
->get(con
) : NULL
;
3015 BUG_ON(msg
->con
!= con
);
3018 static void clear_standby(struct ceph_connection
*con
)
3020 /* come back from STANDBY? */
3021 if (con
->state
== CON_STATE_STANDBY
) {
3022 dout("clear_standby %p and ++connect_seq\n", con
);
3023 con
->state
= CON_STATE_PREOPEN
;
3025 WARN_ON(con_flag_test(con
, CON_FLAG_WRITE_PENDING
));
3026 WARN_ON(con_flag_test(con
, CON_FLAG_KEEPALIVE_PENDING
));
3031 * Queue up an outgoing message on the given connection.
3033 void ceph_con_send(struct ceph_connection
*con
, struct ceph_msg
*msg
)
3036 msg
->hdr
.src
= con
->msgr
->inst
.name
;
3037 BUG_ON(msg
->front
.iov_len
!= le32_to_cpu(msg
->hdr
.front_len
));
3038 msg
->needs_out_seq
= true;
3040 mutex_lock(&con
->mutex
);
3042 if (con
->state
== CON_STATE_CLOSED
) {
3043 dout("con_send %p closed, dropping %p\n", con
, msg
);
3045 mutex_unlock(&con
->mutex
);
3049 msg_con_set(msg
, con
);
3051 BUG_ON(!list_empty(&msg
->list_head
));
3052 list_add_tail(&msg
->list_head
, &con
->out_queue
);
3053 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg
,
3054 ENTITY_NAME(con
->peer_name
), le16_to_cpu(msg
->hdr
.type
),
3055 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
3056 le32_to_cpu(msg
->hdr
.front_len
),
3057 le32_to_cpu(msg
->hdr
.middle_len
),
3058 le32_to_cpu(msg
->hdr
.data_len
));
3061 mutex_unlock(&con
->mutex
);
3063 /* if there wasn't anything waiting to send before, queue
3065 if (con_flag_test_and_set(con
, CON_FLAG_WRITE_PENDING
) == 0)
3068 EXPORT_SYMBOL(ceph_con_send
);
3071 * Revoke a message that was previously queued for send
3073 void ceph_msg_revoke(struct ceph_msg
*msg
)
3075 struct ceph_connection
*con
= msg
->con
;
3078 dout("%s msg %p null con\n", __func__
, msg
);
3079 return; /* Message not in our possession */
3082 mutex_lock(&con
->mutex
);
3083 if (!list_empty(&msg
->list_head
)) {
3084 dout("%s %p msg %p - was on queue\n", __func__
, con
, msg
);
3085 list_del_init(&msg
->list_head
);
3090 if (con
->out_msg
== msg
) {
3091 BUG_ON(con
->out_skip
);
3093 if (con
->out_msg_done
) {
3094 con
->out_skip
+= con_out_kvec_skip(con
);
3096 BUG_ON(!msg
->data_length
);
3097 con
->out_skip
+= sizeof_footer(con
);
3099 /* data, middle, front */
3100 if (msg
->data_length
)
3101 con
->out_skip
+= msg
->cursor
.total_resid
;
3103 con
->out_skip
+= con_out_kvec_skip(con
);
3104 con
->out_skip
+= con_out_kvec_skip(con
);
3106 dout("%s %p msg %p - was sending, will write %d skip %d\n",
3107 __func__
, con
, msg
, con
->out_kvec_bytes
, con
->out_skip
);
3109 con
->out_msg
= NULL
;
3113 mutex_unlock(&con
->mutex
);
3117 * Revoke a message that we may be reading data into
3119 void ceph_msg_revoke_incoming(struct ceph_msg
*msg
)
3121 struct ceph_connection
*con
= msg
->con
;
3124 dout("%s msg %p null con\n", __func__
, msg
);
3125 return; /* Message not in our possession */
3128 mutex_lock(&con
->mutex
);
3129 if (con
->in_msg
== msg
) {
3130 unsigned int front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
3131 unsigned int middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
3132 unsigned int data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
3134 /* skip rest of message */
3135 dout("%s %p msg %p revoked\n", __func__
, con
, msg
);
3136 con
->in_base_pos
= con
->in_base_pos
-
3137 sizeof(struct ceph_msg_header
) -
3141 sizeof(struct ceph_msg_footer
);
3142 ceph_msg_put(con
->in_msg
);
3144 con
->in_tag
= CEPH_MSGR_TAG_READY
;
3147 dout("%s %p in_msg %p msg %p no-op\n",
3148 __func__
, con
, con
->in_msg
, msg
);
3150 mutex_unlock(&con
->mutex
);
3154 * Queue a keepalive byte to ensure the tcp connection is alive.
3156 void ceph_con_keepalive(struct ceph_connection
*con
)
3158 dout("con_keepalive %p\n", con
);
3159 mutex_lock(&con
->mutex
);
3161 mutex_unlock(&con
->mutex
);
3162 if (con_flag_test_and_set(con
, CON_FLAG_KEEPALIVE_PENDING
) == 0 &&
3163 con_flag_test_and_set(con
, CON_FLAG_WRITE_PENDING
) == 0)
3166 EXPORT_SYMBOL(ceph_con_keepalive
);
3168 bool ceph_con_keepalive_expired(struct ceph_connection
*con
,
3169 unsigned long interval
)
3172 (con
->peer_features
& CEPH_FEATURE_MSGR_KEEPALIVE2
)) {
3173 struct timespec now
= CURRENT_TIME
;
3175 jiffies_to_timespec(interval
, &ts
);
3176 ts
= timespec_add(con
->last_keepalive_ack
, ts
);
3177 return timespec_compare(&now
, &ts
) >= 0;
3182 static struct ceph_msg_data
*ceph_msg_data_create(enum ceph_msg_data_type type
)
3184 struct ceph_msg_data
*data
;
3186 if (WARN_ON(!ceph_msg_data_type_valid(type
)))
3189 data
= kmem_cache_zalloc(ceph_msg_data_cache
, GFP_NOFS
);
3192 INIT_LIST_HEAD(&data
->links
);
3197 static void ceph_msg_data_destroy(struct ceph_msg_data
*data
)
3202 WARN_ON(!list_empty(&data
->links
));
3203 if (data
->type
== CEPH_MSG_DATA_PAGELIST
)
3204 ceph_pagelist_release(data
->pagelist
);
3205 kmem_cache_free(ceph_msg_data_cache
, data
);
3208 void ceph_msg_data_add_pages(struct ceph_msg
*msg
, struct page
**pages
,
3209 size_t length
, size_t alignment
)
3211 struct ceph_msg_data
*data
;
3216 data
= ceph_msg_data_create(CEPH_MSG_DATA_PAGES
);
3218 data
->pages
= pages
;
3219 data
->length
= length
;
3220 data
->alignment
= alignment
& ~PAGE_MASK
;
3222 list_add_tail(&data
->links
, &msg
->data
);
3223 msg
->data_length
+= length
;
3225 EXPORT_SYMBOL(ceph_msg_data_add_pages
);
3227 void ceph_msg_data_add_pagelist(struct ceph_msg
*msg
,
3228 struct ceph_pagelist
*pagelist
)
3230 struct ceph_msg_data
*data
;
3233 BUG_ON(!pagelist
->length
);
3235 data
= ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST
);
3237 data
->pagelist
= pagelist
;
3239 list_add_tail(&data
->links
, &msg
->data
);
3240 msg
->data_length
+= pagelist
->length
;
3242 EXPORT_SYMBOL(ceph_msg_data_add_pagelist
);
3245 void ceph_msg_data_add_bio(struct ceph_msg
*msg
, struct bio
*bio
,
3248 struct ceph_msg_data
*data
;
3252 data
= ceph_msg_data_create(CEPH_MSG_DATA_BIO
);
3255 data
->bio_length
= length
;
3257 list_add_tail(&data
->links
, &msg
->data
);
3258 msg
->data_length
+= length
;
3260 EXPORT_SYMBOL(ceph_msg_data_add_bio
);
3261 #endif /* CONFIG_BLOCK */
3264 * construct a new message with given type, size
3265 * the new msg has a ref count of 1.
3267 struct ceph_msg
*ceph_msg_new(int type
, int front_len
, gfp_t flags
,
3272 m
= kmem_cache_zalloc(ceph_msg_cache
, flags
);
3276 m
->hdr
.type
= cpu_to_le16(type
);
3277 m
->hdr
.priority
= cpu_to_le16(CEPH_MSG_PRIO_DEFAULT
);
3278 m
->hdr
.front_len
= cpu_to_le32(front_len
);
3280 INIT_LIST_HEAD(&m
->list_head
);
3281 kref_init(&m
->kref
);
3282 INIT_LIST_HEAD(&m
->data
);
3286 m
->front
.iov_base
= ceph_kvmalloc(front_len
, flags
);
3287 if (m
->front
.iov_base
== NULL
) {
3288 dout("ceph_msg_new can't allocate %d bytes\n",
3293 m
->front
.iov_base
= NULL
;
3295 m
->front_alloc_len
= m
->front
.iov_len
= front_len
;
3297 dout("ceph_msg_new %p front %d\n", m
, front_len
);
3304 pr_err("msg_new can't create type %d front %d\n", type
,
3308 dout("msg_new can't create type %d front %d\n", type
,
3313 EXPORT_SYMBOL(ceph_msg_new
);
3316 * Allocate "middle" portion of a message, if it is needed and wasn't
3317 * allocated by alloc_msg. This allows us to read a small fixed-size
3318 * per-type header in the front and then gracefully fail (i.e.,
3319 * propagate the error to the caller based on info in the front) when
3320 * the middle is too large.
3322 static int ceph_alloc_middle(struct ceph_connection
*con
, struct ceph_msg
*msg
)
3324 int type
= le16_to_cpu(msg
->hdr
.type
);
3325 int middle_len
= le32_to_cpu(msg
->hdr
.middle_len
);
3327 dout("alloc_middle %p type %d %s middle_len %d\n", msg
, type
,
3328 ceph_msg_type_name(type
), middle_len
);
3329 BUG_ON(!middle_len
);
3330 BUG_ON(msg
->middle
);
3332 msg
->middle
= ceph_buffer_new(middle_len
, GFP_NOFS
);
3339 * Allocate a message for receiving an incoming message on a
3340 * connection, and save the result in con->in_msg. Uses the
3341 * connection's private alloc_msg op if available.
3343 * Returns 0 on success, or a negative error code.
3345 * On success, if we set *skip = 1:
3346 * - the next message should be skipped and ignored.
3347 * - con->in_msg == NULL
3348 * or if we set *skip = 0:
3349 * - con->in_msg is non-null.
3350 * On error (ENOMEM, EAGAIN, ...),
3351 * - con->in_msg == NULL
3353 static int ceph_con_in_msg_alloc(struct ceph_connection
*con
, int *skip
)
3355 struct ceph_msg_header
*hdr
= &con
->in_hdr
;
3356 int middle_len
= le32_to_cpu(hdr
->middle_len
);
3357 struct ceph_msg
*msg
;
3360 BUG_ON(con
->in_msg
!= NULL
);
3361 BUG_ON(!con
->ops
->alloc_msg
);
3363 mutex_unlock(&con
->mutex
);
3364 msg
= con
->ops
->alloc_msg(con
, hdr
, skip
);
3365 mutex_lock(&con
->mutex
);
3366 if (con
->state
!= CON_STATE_OPEN
) {
3373 msg_con_set(msg
, con
);
3377 * Null message pointer means either we should skip
3378 * this message or we couldn't allocate memory. The
3379 * former is not an error.
3384 con
->error_msg
= "error allocating memory for incoming message";
3387 memcpy(&con
->in_msg
->hdr
, &con
->in_hdr
, sizeof(con
->in_hdr
));
3389 if (middle_len
&& !con
->in_msg
->middle
) {
3390 ret
= ceph_alloc_middle(con
, con
->in_msg
);
3392 ceph_msg_put(con
->in_msg
);
3402 * Free a generically kmalloc'd message.
3404 static void ceph_msg_free(struct ceph_msg
*m
)
3406 dout("%s %p\n", __func__
, m
);
3407 kvfree(m
->front
.iov_base
);
3408 kmem_cache_free(ceph_msg_cache
, m
);
3411 static void ceph_msg_release(struct kref
*kref
)
3413 struct ceph_msg
*m
= container_of(kref
, struct ceph_msg
, kref
);
3414 struct ceph_msg_data
*data
, *next
;
3416 dout("%s %p\n", __func__
, m
);
3417 WARN_ON(!list_empty(&m
->list_head
));
3419 msg_con_set(m
, NULL
);
3421 /* drop middle, data, if any */
3423 ceph_buffer_put(m
->middle
);
3427 list_for_each_entry_safe(data
, next
, &m
->data
, links
) {
3428 list_del_init(&data
->links
);
3429 ceph_msg_data_destroy(data
);
3434 ceph_msgpool_put(m
->pool
, m
);
3439 struct ceph_msg
*ceph_msg_get(struct ceph_msg
*msg
)
3441 dout("%s %p (was %d)\n", __func__
, msg
,
3442 kref_read(&msg
->kref
));
3443 kref_get(&msg
->kref
);
3446 EXPORT_SYMBOL(ceph_msg_get
);
3448 void ceph_msg_put(struct ceph_msg
*msg
)
3450 dout("%s %p (was %d)\n", __func__
, msg
,
3451 kref_read(&msg
->kref
));
3452 kref_put(&msg
->kref
, ceph_msg_release
);
3454 EXPORT_SYMBOL(ceph_msg_put
);
3456 void ceph_msg_dump(struct ceph_msg
*msg
)
3458 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg
,
3459 msg
->front_alloc_len
, msg
->data_length
);
3460 print_hex_dump(KERN_DEBUG
, "header: ",
3461 DUMP_PREFIX_OFFSET
, 16, 1,
3462 &msg
->hdr
, sizeof(msg
->hdr
), true);
3463 print_hex_dump(KERN_DEBUG
, " front: ",
3464 DUMP_PREFIX_OFFSET
, 16, 1,
3465 msg
->front
.iov_base
, msg
->front
.iov_len
, true);
3467 print_hex_dump(KERN_DEBUG
, "middle: ",
3468 DUMP_PREFIX_OFFSET
, 16, 1,
3469 msg
->middle
->vec
.iov_base
,
3470 msg
->middle
->vec
.iov_len
, true);
3471 print_hex_dump(KERN_DEBUG
, "footer: ",
3472 DUMP_PREFIX_OFFSET
, 16, 1,
3473 &msg
->footer
, sizeof(msg
->footer
), true);
3475 EXPORT_SYMBOL(ceph_msg_dump
);