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_create("ceph_msg",
239 sizeof (struct ceph_msg
),
240 __alignof__(struct ceph_msg
), 0, NULL
);
245 BUG_ON(ceph_msg_data_cache
);
246 ceph_msg_data_cache
= kmem_cache_create("ceph_msg_data",
247 sizeof (struct ceph_msg_data
),
248 __alignof__(struct ceph_msg_data
),
250 if (ceph_msg_data_cache
)
253 kmem_cache_destroy(ceph_msg_cache
);
254 ceph_msg_cache
= NULL
;
259 static void ceph_msgr_slab_exit(void)
261 BUG_ON(!ceph_msg_data_cache
);
262 kmem_cache_destroy(ceph_msg_data_cache
);
263 ceph_msg_data_cache
= NULL
;
265 BUG_ON(!ceph_msg_cache
);
266 kmem_cache_destroy(ceph_msg_cache
);
267 ceph_msg_cache
= NULL
;
270 static void _ceph_msgr_exit(void)
273 destroy_workqueue(ceph_msgr_wq
);
277 BUG_ON(zero_page
== NULL
);
278 page_cache_release(zero_page
);
281 ceph_msgr_slab_exit();
284 int ceph_msgr_init(void)
286 if (ceph_msgr_slab_init())
289 BUG_ON(zero_page
!= NULL
);
290 zero_page
= ZERO_PAGE(0);
291 page_cache_get(zero_page
);
294 * The number of active work items is limited by the number of
295 * connections, so leave @max_active at default.
297 ceph_msgr_wq
= alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM
, 0);
301 pr_err("msgr_init failed to create workqueue\n");
306 EXPORT_SYMBOL(ceph_msgr_init
);
308 void ceph_msgr_exit(void)
310 BUG_ON(ceph_msgr_wq
== NULL
);
314 EXPORT_SYMBOL(ceph_msgr_exit
);
316 void ceph_msgr_flush(void)
318 flush_workqueue(ceph_msgr_wq
);
320 EXPORT_SYMBOL(ceph_msgr_flush
);
322 /* Connection socket state transition functions */
324 static void con_sock_state_init(struct ceph_connection
*con
)
328 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSED
);
329 if (WARN_ON(old_state
!= CON_SOCK_STATE_NEW
))
330 printk("%s: unexpected old state %d\n", __func__
, old_state
);
331 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
332 CON_SOCK_STATE_CLOSED
);
335 static void con_sock_state_connecting(struct ceph_connection
*con
)
339 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CONNECTING
);
340 if (WARN_ON(old_state
!= CON_SOCK_STATE_CLOSED
))
341 printk("%s: unexpected old state %d\n", __func__
, old_state
);
342 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
343 CON_SOCK_STATE_CONNECTING
);
346 static void con_sock_state_connected(struct ceph_connection
*con
)
350 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CONNECTED
);
351 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTING
))
352 printk("%s: unexpected old state %d\n", __func__
, old_state
);
353 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
354 CON_SOCK_STATE_CONNECTED
);
357 static void con_sock_state_closing(struct ceph_connection
*con
)
361 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSING
);
362 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTING
&&
363 old_state
!= CON_SOCK_STATE_CONNECTED
&&
364 old_state
!= CON_SOCK_STATE_CLOSING
))
365 printk("%s: unexpected old state %d\n", __func__
, old_state
);
366 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
367 CON_SOCK_STATE_CLOSING
);
370 static void con_sock_state_closed(struct ceph_connection
*con
)
374 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSED
);
375 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTED
&&
376 old_state
!= CON_SOCK_STATE_CLOSING
&&
377 old_state
!= CON_SOCK_STATE_CONNECTING
&&
378 old_state
!= CON_SOCK_STATE_CLOSED
))
379 printk("%s: unexpected old state %d\n", __func__
, old_state
);
380 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
381 CON_SOCK_STATE_CLOSED
);
385 * socket callback functions
388 /* data available on socket, or listen socket received a connect */
389 static void ceph_sock_data_ready(struct sock
*sk
)
391 struct ceph_connection
*con
= sk
->sk_user_data
;
392 if (atomic_read(&con
->msgr
->stopping
)) {
396 if (sk
->sk_state
!= TCP_CLOSE_WAIT
) {
397 dout("%s on %p state = %lu, queueing work\n", __func__
,
403 /* socket has buffer space for writing */
404 static void ceph_sock_write_space(struct sock
*sk
)
406 struct ceph_connection
*con
= sk
->sk_user_data
;
408 /* only queue to workqueue if there is data we want to write,
409 * and there is sufficient space in the socket buffer to accept
410 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
411 * doesn't get called again until try_write() fills the socket
412 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
413 * and net/core/stream.c:sk_stream_write_space().
415 if (con_flag_test(con
, CON_FLAG_WRITE_PENDING
)) {
416 if (sk_stream_is_writeable(sk
)) {
417 dout("%s %p queueing write work\n", __func__
, con
);
418 clear_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
422 dout("%s %p nothing to write\n", __func__
, con
);
426 /* socket's state has changed */
427 static void ceph_sock_state_change(struct sock
*sk
)
429 struct ceph_connection
*con
= sk
->sk_user_data
;
431 dout("%s %p state = %lu sk_state = %u\n", __func__
,
432 con
, con
->state
, sk
->sk_state
);
434 switch (sk
->sk_state
) {
436 dout("%s TCP_CLOSE\n", __func__
);
438 dout("%s TCP_CLOSE_WAIT\n", __func__
);
439 con_sock_state_closing(con
);
440 con_flag_set(con
, CON_FLAG_SOCK_CLOSED
);
443 case TCP_ESTABLISHED
:
444 dout("%s TCP_ESTABLISHED\n", __func__
);
445 con_sock_state_connected(con
);
448 default: /* Everything else is uninteresting */
454 * set up socket callbacks
456 static void set_sock_callbacks(struct socket
*sock
,
457 struct ceph_connection
*con
)
459 struct sock
*sk
= sock
->sk
;
460 sk
->sk_user_data
= con
;
461 sk
->sk_data_ready
= ceph_sock_data_ready
;
462 sk
->sk_write_space
= ceph_sock_write_space
;
463 sk
->sk_state_change
= ceph_sock_state_change
;
472 * initiate connection to a remote socket.
474 static int ceph_tcp_connect(struct ceph_connection
*con
)
476 struct sockaddr_storage
*paddr
= &con
->peer_addr
.in_addr
;
481 ret
= sock_create_kern(read_pnet(&con
->msgr
->net
), paddr
->ss_family
,
482 SOCK_STREAM
, IPPROTO_TCP
, &sock
);
485 sock
->sk
->sk_allocation
= GFP_NOFS
;
487 #ifdef CONFIG_LOCKDEP
488 lockdep_set_class(&sock
->sk
->sk_lock
, &socket_class
);
491 set_sock_callbacks(sock
, con
);
493 dout("connect %s\n", ceph_pr_addr(&con
->peer_addr
.in_addr
));
495 con_sock_state_connecting(con
);
496 ret
= sock
->ops
->connect(sock
, (struct sockaddr
*)paddr
, sizeof(*paddr
),
498 if (ret
== -EINPROGRESS
) {
499 dout("connect %s EINPROGRESS sk_state = %u\n",
500 ceph_pr_addr(&con
->peer_addr
.in_addr
),
502 } else if (ret
< 0) {
503 pr_err("connect %s error %d\n",
504 ceph_pr_addr(&con
->peer_addr
.in_addr
), ret
);
509 if (ceph_test_opt(from_msgr(con
->msgr
), TCP_NODELAY
)) {
512 ret
= kernel_setsockopt(sock
, SOL_TCP
, TCP_NODELAY
,
513 (char *)&optval
, sizeof(optval
));
515 pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
523 static int ceph_tcp_recvmsg(struct socket
*sock
, void *buf
, size_t len
)
525 struct kvec iov
= {buf
, len
};
526 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
529 r
= kernel_recvmsg(sock
, &msg
, &iov
, 1, len
, msg
.msg_flags
);
535 static int ceph_tcp_recvpage(struct socket
*sock
, struct page
*page
,
536 int page_offset
, size_t length
)
541 BUG_ON(page_offset
+ length
> PAGE_SIZE
);
545 ret
= ceph_tcp_recvmsg(sock
, kaddr
+ page_offset
, length
);
552 * write something. @more is true if caller will be sending more data
555 static int ceph_tcp_sendmsg(struct socket
*sock
, struct kvec
*iov
,
556 size_t kvlen
, size_t len
, int more
)
558 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
562 msg
.msg_flags
|= MSG_MORE
;
564 msg
.msg_flags
|= MSG_EOR
; /* superfluous, but what the hell */
566 r
= kernel_sendmsg(sock
, &msg
, iov
, kvlen
, len
);
572 static int __ceph_tcp_sendpage(struct socket
*sock
, struct page
*page
,
573 int offset
, size_t size
, bool more
)
575 int flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
| (more
? MSG_MORE
: MSG_EOR
);
578 ret
= kernel_sendpage(sock
, page
, offset
, size
, flags
);
585 static int ceph_tcp_sendpage(struct socket
*sock
, struct page
*page
,
586 int offset
, size_t size
, bool more
)
591 /* sendpage cannot properly handle pages with page_count == 0,
592 * we need to fallback to sendmsg if that's the case */
593 if (page_count(page
) >= 1)
594 return __ceph_tcp_sendpage(sock
, page
, offset
, size
, more
);
596 iov
.iov_base
= kmap(page
) + offset
;
598 ret
= ceph_tcp_sendmsg(sock
, &iov
, 1, size
, more
);
605 * Shutdown/close the socket for the given connection.
607 static int con_close_socket(struct ceph_connection
*con
)
611 dout("con_close_socket on %p sock %p\n", con
, con
->sock
);
613 rc
= con
->sock
->ops
->shutdown(con
->sock
, SHUT_RDWR
);
614 sock_release(con
->sock
);
619 * Forcibly clear the SOCK_CLOSED flag. It gets set
620 * independent of the connection mutex, and we could have
621 * received a socket close event before we had the chance to
622 * shut the socket down.
624 con_flag_clear(con
, CON_FLAG_SOCK_CLOSED
);
626 con_sock_state_closed(con
);
631 * Reset a connection. Discard all incoming and outgoing messages
632 * and clear *_seq state.
634 static void ceph_msg_remove(struct ceph_msg
*msg
)
636 list_del_init(&msg
->list_head
);
640 static void ceph_msg_remove_list(struct list_head
*head
)
642 while (!list_empty(head
)) {
643 struct ceph_msg
*msg
= list_first_entry(head
, struct ceph_msg
,
645 ceph_msg_remove(msg
);
649 static void reset_connection(struct ceph_connection
*con
)
651 /* reset connection, out_queue, msg_ and connect_seq */
652 /* discard existing out_queue and msg_seq */
653 dout("reset_connection %p\n", con
);
654 ceph_msg_remove_list(&con
->out_queue
);
655 ceph_msg_remove_list(&con
->out_sent
);
658 BUG_ON(con
->in_msg
->con
!= con
);
659 ceph_msg_put(con
->in_msg
);
663 con
->connect_seq
= 0;
666 BUG_ON(con
->out_msg
->con
!= con
);
667 ceph_msg_put(con
->out_msg
);
671 con
->in_seq_acked
= 0;
677 * mark a peer down. drop any open connections.
679 void ceph_con_close(struct ceph_connection
*con
)
681 mutex_lock(&con
->mutex
);
682 dout("con_close %p peer %s\n", con
,
683 ceph_pr_addr(&con
->peer_addr
.in_addr
));
684 con
->state
= CON_STATE_CLOSED
;
686 con_flag_clear(con
, CON_FLAG_LOSSYTX
); /* so we retry next connect */
687 con_flag_clear(con
, CON_FLAG_KEEPALIVE_PENDING
);
688 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
689 con_flag_clear(con
, CON_FLAG_BACKOFF
);
691 reset_connection(con
);
692 con
->peer_global_seq
= 0;
694 con_close_socket(con
);
695 mutex_unlock(&con
->mutex
);
697 EXPORT_SYMBOL(ceph_con_close
);
700 * Reopen a closed connection, with a new peer address.
702 void ceph_con_open(struct ceph_connection
*con
,
703 __u8 entity_type
, __u64 entity_num
,
704 struct ceph_entity_addr
*addr
)
706 mutex_lock(&con
->mutex
);
707 dout("con_open %p %s\n", con
, ceph_pr_addr(&addr
->in_addr
));
709 WARN_ON(con
->state
!= CON_STATE_CLOSED
);
710 con
->state
= CON_STATE_PREOPEN
;
712 con
->peer_name
.type
= (__u8
) entity_type
;
713 con
->peer_name
.num
= cpu_to_le64(entity_num
);
715 memcpy(&con
->peer_addr
, addr
, sizeof(*addr
));
716 con
->delay
= 0; /* reset backoff memory */
717 mutex_unlock(&con
->mutex
);
720 EXPORT_SYMBOL(ceph_con_open
);
723 * return true if this connection ever successfully opened
725 bool ceph_con_opened(struct ceph_connection
*con
)
727 return con
->connect_seq
> 0;
731 * initialize a new connection.
733 void ceph_con_init(struct ceph_connection
*con
, void *private,
734 const struct ceph_connection_operations
*ops
,
735 struct ceph_messenger
*msgr
)
737 dout("con_init %p\n", con
);
738 memset(con
, 0, sizeof(*con
));
739 con
->private = private;
743 con_sock_state_init(con
);
745 mutex_init(&con
->mutex
);
746 INIT_LIST_HEAD(&con
->out_queue
);
747 INIT_LIST_HEAD(&con
->out_sent
);
748 INIT_DELAYED_WORK(&con
->work
, ceph_con_workfn
);
750 con
->state
= CON_STATE_CLOSED
;
752 EXPORT_SYMBOL(ceph_con_init
);
756 * We maintain a global counter to order connection attempts. Get
757 * a unique seq greater than @gt.
759 static u32
get_global_seq(struct ceph_messenger
*msgr
, u32 gt
)
763 spin_lock(&msgr
->global_seq_lock
);
764 if (msgr
->global_seq
< gt
)
765 msgr
->global_seq
= gt
;
766 ret
= ++msgr
->global_seq
;
767 spin_unlock(&msgr
->global_seq_lock
);
771 static void con_out_kvec_reset(struct ceph_connection
*con
)
773 BUG_ON(con
->out_skip
);
775 con
->out_kvec_left
= 0;
776 con
->out_kvec_bytes
= 0;
777 con
->out_kvec_cur
= &con
->out_kvec
[0];
780 static void con_out_kvec_add(struct ceph_connection
*con
,
781 size_t size
, void *data
)
783 int index
= con
->out_kvec_left
;
785 BUG_ON(con
->out_skip
);
786 BUG_ON(index
>= ARRAY_SIZE(con
->out_kvec
));
788 con
->out_kvec
[index
].iov_len
= size
;
789 con
->out_kvec
[index
].iov_base
= data
;
790 con
->out_kvec_left
++;
791 con
->out_kvec_bytes
+= size
;
795 * Chop off a kvec from the end. Return residual number of bytes for
796 * that kvec, i.e. how many bytes would have been written if the kvec
799 static int con_out_kvec_skip(struct ceph_connection
*con
)
801 int off
= con
->out_kvec_cur
- con
->out_kvec
;
804 if (con
->out_kvec_bytes
> 0) {
805 skip
= con
->out_kvec
[off
+ con
->out_kvec_left
- 1].iov_len
;
806 BUG_ON(con
->out_kvec_bytes
< skip
);
807 BUG_ON(!con
->out_kvec_left
);
808 con
->out_kvec_bytes
-= skip
;
809 con
->out_kvec_left
--;
818 * For a bio data item, a piece is whatever remains of the next
819 * entry in the current bio iovec, or the first entry in the next
822 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor
*cursor
,
825 struct ceph_msg_data
*data
= cursor
->data
;
828 BUG_ON(data
->type
!= CEPH_MSG_DATA_BIO
);
833 cursor
->resid
= min(length
, data
->bio_length
);
835 cursor
->bvec_iter
= bio
->bi_iter
;
837 cursor
->resid
<= bio_iter_len(bio
, cursor
->bvec_iter
);
840 static struct page
*ceph_msg_data_bio_next(struct ceph_msg_data_cursor
*cursor
,
844 struct ceph_msg_data
*data
= cursor
->data
;
846 struct bio_vec bio_vec
;
848 BUG_ON(data
->type
!= CEPH_MSG_DATA_BIO
);
853 bio_vec
= bio_iter_iovec(bio
, cursor
->bvec_iter
);
855 *page_offset
= (size_t) bio_vec
.bv_offset
;
856 BUG_ON(*page_offset
>= PAGE_SIZE
);
857 if (cursor
->last_piece
) /* pagelist offset is always 0 */
858 *length
= cursor
->resid
;
860 *length
= (size_t) bio_vec
.bv_len
;
861 BUG_ON(*length
> cursor
->resid
);
862 BUG_ON(*page_offset
+ *length
> PAGE_SIZE
);
864 return bio_vec
.bv_page
;
867 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor
*cursor
,
871 struct bio_vec bio_vec
;
873 BUG_ON(cursor
->data
->type
!= CEPH_MSG_DATA_BIO
);
878 bio_vec
= bio_iter_iovec(bio
, cursor
->bvec_iter
);
880 /* Advance the cursor offset */
882 BUG_ON(cursor
->resid
< bytes
);
883 cursor
->resid
-= bytes
;
885 bio_advance_iter(bio
, &cursor
->bvec_iter
, bytes
);
887 if (bytes
< bio_vec
.bv_len
)
888 return false; /* more bytes to process in this segment */
890 /* Move on to the next segment, and possibly the next bio */
892 if (!cursor
->bvec_iter
.bi_size
) {
896 cursor
->bvec_iter
= bio
->bi_iter
;
898 memset(&cursor
->bvec_iter
, 0,
899 sizeof(cursor
->bvec_iter
));
902 if (!cursor
->last_piece
) {
903 BUG_ON(!cursor
->resid
);
905 /* A short read is OK, so use <= rather than == */
906 if (cursor
->resid
<= bio_iter_len(bio
, cursor
->bvec_iter
))
907 cursor
->last_piece
= true;
912 #endif /* CONFIG_BLOCK */
915 * For a page array, a piece comes from the first page in the array
916 * that has not already been fully consumed.
918 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor
*cursor
,
921 struct ceph_msg_data
*data
= cursor
->data
;
924 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGES
);
926 BUG_ON(!data
->pages
);
927 BUG_ON(!data
->length
);
929 cursor
->resid
= min(length
, data
->length
);
930 page_count
= calc_pages_for(data
->alignment
, (u64
)data
->length
);
931 cursor
->page_offset
= data
->alignment
& ~PAGE_MASK
;
932 cursor
->page_index
= 0;
933 BUG_ON(page_count
> (int)USHRT_MAX
);
934 cursor
->page_count
= (unsigned short)page_count
;
935 BUG_ON(length
> SIZE_MAX
- cursor
->page_offset
);
936 cursor
->last_piece
= cursor
->page_offset
+ cursor
->resid
<= PAGE_SIZE
;
940 ceph_msg_data_pages_next(struct ceph_msg_data_cursor
*cursor
,
941 size_t *page_offset
, size_t *length
)
943 struct ceph_msg_data
*data
= cursor
->data
;
945 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGES
);
947 BUG_ON(cursor
->page_index
>= cursor
->page_count
);
948 BUG_ON(cursor
->page_offset
>= PAGE_SIZE
);
950 *page_offset
= cursor
->page_offset
;
951 if (cursor
->last_piece
)
952 *length
= cursor
->resid
;
954 *length
= PAGE_SIZE
- *page_offset
;
956 return data
->pages
[cursor
->page_index
];
959 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor
*cursor
,
962 BUG_ON(cursor
->data
->type
!= CEPH_MSG_DATA_PAGES
);
964 BUG_ON(cursor
->page_offset
+ bytes
> PAGE_SIZE
);
966 /* Advance the cursor page offset */
968 cursor
->resid
-= bytes
;
969 cursor
->page_offset
= (cursor
->page_offset
+ bytes
) & ~PAGE_MASK
;
970 if (!bytes
|| cursor
->page_offset
)
971 return false; /* more bytes to process in the current page */
974 return false; /* no more data */
976 /* Move on to the next page; offset is already at 0 */
978 BUG_ON(cursor
->page_index
>= cursor
->page_count
);
979 cursor
->page_index
++;
980 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
986 * For a pagelist, a piece is whatever remains to be consumed in the
987 * first page in the list, or the front of the next page.
990 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor
*cursor
,
993 struct ceph_msg_data
*data
= cursor
->data
;
994 struct ceph_pagelist
*pagelist
;
997 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
999 pagelist
= data
->pagelist
;
1003 return; /* pagelist can be assigned but empty */
1005 BUG_ON(list_empty(&pagelist
->head
));
1006 page
= list_first_entry(&pagelist
->head
, struct page
, lru
);
1008 cursor
->resid
= min(length
, pagelist
->length
);
1009 cursor
->page
= page
;
1011 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
1014 static struct page
*
1015 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor
*cursor
,
1016 size_t *page_offset
, size_t *length
)
1018 struct ceph_msg_data
*data
= cursor
->data
;
1019 struct ceph_pagelist
*pagelist
;
1021 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
1023 pagelist
= data
->pagelist
;
1026 BUG_ON(!cursor
->page
);
1027 BUG_ON(cursor
->offset
+ cursor
->resid
!= pagelist
->length
);
1029 /* offset of first page in pagelist is always 0 */
1030 *page_offset
= cursor
->offset
& ~PAGE_MASK
;
1031 if (cursor
->last_piece
)
1032 *length
= cursor
->resid
;
1034 *length
= PAGE_SIZE
- *page_offset
;
1036 return cursor
->page
;
1039 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor
*cursor
,
1042 struct ceph_msg_data
*data
= cursor
->data
;
1043 struct ceph_pagelist
*pagelist
;
1045 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
1047 pagelist
= data
->pagelist
;
1050 BUG_ON(cursor
->offset
+ cursor
->resid
!= pagelist
->length
);
1051 BUG_ON((cursor
->offset
& ~PAGE_MASK
) + bytes
> PAGE_SIZE
);
1053 /* Advance the cursor offset */
1055 cursor
->resid
-= bytes
;
1056 cursor
->offset
+= bytes
;
1057 /* offset of first page in pagelist is always 0 */
1058 if (!bytes
|| cursor
->offset
& ~PAGE_MASK
)
1059 return false; /* more bytes to process in the current page */
1062 return false; /* no more data */
1064 /* Move on to the next page */
1066 BUG_ON(list_is_last(&cursor
->page
->lru
, &pagelist
->head
));
1067 cursor
->page
= list_next_entry(cursor
->page
, lru
);
1068 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
1074 * Message data is handled (sent or received) in pieces, where each
1075 * piece resides on a single page. The network layer might not
1076 * consume an entire piece at once. A data item's cursor keeps
1077 * track of which piece is next to process and how much remains to
1078 * be processed in that piece. It also tracks whether the current
1079 * piece is the last one in the data item.
1081 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor
*cursor
)
1083 size_t length
= cursor
->total_resid
;
1085 switch (cursor
->data
->type
) {
1086 case CEPH_MSG_DATA_PAGELIST
:
1087 ceph_msg_data_pagelist_cursor_init(cursor
, length
);
1089 case CEPH_MSG_DATA_PAGES
:
1090 ceph_msg_data_pages_cursor_init(cursor
, length
);
1093 case CEPH_MSG_DATA_BIO
:
1094 ceph_msg_data_bio_cursor_init(cursor
, length
);
1096 #endif /* CONFIG_BLOCK */
1097 case CEPH_MSG_DATA_NONE
:
1102 cursor
->need_crc
= true;
1105 static void ceph_msg_data_cursor_init(struct ceph_msg
*msg
, size_t length
)
1107 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
1108 struct ceph_msg_data
*data
;
1111 BUG_ON(length
> msg
->data_length
);
1112 BUG_ON(list_empty(&msg
->data
));
1114 cursor
->data_head
= &msg
->data
;
1115 cursor
->total_resid
= length
;
1116 data
= list_first_entry(&msg
->data
, struct ceph_msg_data
, links
);
1117 cursor
->data
= data
;
1119 __ceph_msg_data_cursor_init(cursor
);
1123 * Return the page containing the next piece to process for a given
1124 * data item, and supply the page offset and length of that piece.
1125 * Indicate whether this is the last piece in this data item.
1127 static struct page
*ceph_msg_data_next(struct ceph_msg_data_cursor
*cursor
,
1128 size_t *page_offset
, size_t *length
,
1133 switch (cursor
->data
->type
) {
1134 case CEPH_MSG_DATA_PAGELIST
:
1135 page
= ceph_msg_data_pagelist_next(cursor
, page_offset
, length
);
1137 case CEPH_MSG_DATA_PAGES
:
1138 page
= ceph_msg_data_pages_next(cursor
, page_offset
, length
);
1141 case CEPH_MSG_DATA_BIO
:
1142 page
= ceph_msg_data_bio_next(cursor
, page_offset
, length
);
1144 #endif /* CONFIG_BLOCK */
1145 case CEPH_MSG_DATA_NONE
:
1151 BUG_ON(*page_offset
+ *length
> PAGE_SIZE
);
1154 *last_piece
= cursor
->last_piece
;
1160 * Returns true if the result moves the cursor on to the next piece
1163 static bool ceph_msg_data_advance(struct ceph_msg_data_cursor
*cursor
,
1168 BUG_ON(bytes
> cursor
->resid
);
1169 switch (cursor
->data
->type
) {
1170 case CEPH_MSG_DATA_PAGELIST
:
1171 new_piece
= ceph_msg_data_pagelist_advance(cursor
, bytes
);
1173 case CEPH_MSG_DATA_PAGES
:
1174 new_piece
= ceph_msg_data_pages_advance(cursor
, bytes
);
1177 case CEPH_MSG_DATA_BIO
:
1178 new_piece
= ceph_msg_data_bio_advance(cursor
, bytes
);
1180 #endif /* CONFIG_BLOCK */
1181 case CEPH_MSG_DATA_NONE
:
1186 cursor
->total_resid
-= bytes
;
1188 if (!cursor
->resid
&& cursor
->total_resid
) {
1189 WARN_ON(!cursor
->last_piece
);
1190 BUG_ON(list_is_last(&cursor
->data
->links
, cursor
->data_head
));
1191 cursor
->data
= list_next_entry(cursor
->data
, links
);
1192 __ceph_msg_data_cursor_init(cursor
);
1195 cursor
->need_crc
= new_piece
;
1200 static void prepare_message_data(struct ceph_msg
*msg
, u32 data_len
)
1205 /* Initialize data cursor */
1207 ceph_msg_data_cursor_init(msg
, (size_t)data_len
);
1211 * Prepare footer for currently outgoing message, and finish things
1212 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1214 static void prepare_write_message_footer(struct ceph_connection
*con
)
1216 struct ceph_msg
*m
= con
->out_msg
;
1217 int v
= con
->out_kvec_left
;
1219 m
->footer
.flags
|= CEPH_MSG_FOOTER_COMPLETE
;
1221 dout("prepare_write_message_footer %p\n", con
);
1222 con
->out_kvec
[v
].iov_base
= &m
->footer
;
1223 if (con
->peer_features
& CEPH_FEATURE_MSG_AUTH
) {
1224 if (con
->ops
->sign_message
)
1225 con
->ops
->sign_message(m
);
1228 con
->out_kvec
[v
].iov_len
= sizeof(m
->footer
);
1229 con
->out_kvec_bytes
+= sizeof(m
->footer
);
1231 m
->old_footer
.flags
= m
->footer
.flags
;
1232 con
->out_kvec
[v
].iov_len
= sizeof(m
->old_footer
);
1233 con
->out_kvec_bytes
+= sizeof(m
->old_footer
);
1235 con
->out_kvec_left
++;
1236 con
->out_more
= m
->more_to_follow
;
1237 con
->out_msg_done
= true;
1241 * Prepare headers for the next outgoing message.
1243 static void prepare_write_message(struct ceph_connection
*con
)
1248 con_out_kvec_reset(con
);
1249 con
->out_msg_done
= false;
1251 /* Sneak an ack in there first? If we can get it into the same
1252 * TCP packet that's a good thing. */
1253 if (con
->in_seq
> con
->in_seq_acked
) {
1254 con
->in_seq_acked
= con
->in_seq
;
1255 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
1256 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1257 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1258 &con
->out_temp_ack
);
1261 BUG_ON(list_empty(&con
->out_queue
));
1262 m
= list_first_entry(&con
->out_queue
, struct ceph_msg
, list_head
);
1264 BUG_ON(m
->con
!= con
);
1266 /* put message on sent list */
1268 list_move_tail(&m
->list_head
, &con
->out_sent
);
1271 * only assign outgoing seq # if we haven't sent this message
1272 * yet. if it is requeued, resend with it's original seq.
1274 if (m
->needs_out_seq
) {
1275 m
->hdr
.seq
= cpu_to_le64(++con
->out_seq
);
1276 m
->needs_out_seq
= false;
1278 WARN_ON(m
->data_length
!= le32_to_cpu(m
->hdr
.data_len
));
1280 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1281 m
, con
->out_seq
, le16_to_cpu(m
->hdr
.type
),
1282 le32_to_cpu(m
->hdr
.front_len
), le32_to_cpu(m
->hdr
.middle_len
),
1284 BUG_ON(le32_to_cpu(m
->hdr
.front_len
) != m
->front
.iov_len
);
1286 /* tag + hdr + front + middle */
1287 con_out_kvec_add(con
, sizeof (tag_msg
), &tag_msg
);
1288 con_out_kvec_add(con
, sizeof(con
->out_hdr
), &con
->out_hdr
);
1289 con_out_kvec_add(con
, m
->front
.iov_len
, m
->front
.iov_base
);
1292 con_out_kvec_add(con
, m
->middle
->vec
.iov_len
,
1293 m
->middle
->vec
.iov_base
);
1295 /* fill in hdr crc and finalize hdr */
1296 crc
= crc32c(0, &m
->hdr
, offsetof(struct ceph_msg_header
, crc
));
1297 con
->out_msg
->hdr
.crc
= cpu_to_le32(crc
);
1298 memcpy(&con
->out_hdr
, &con
->out_msg
->hdr
, sizeof(con
->out_hdr
));
1300 /* fill in front and middle crc, footer */
1301 crc
= crc32c(0, m
->front
.iov_base
, m
->front
.iov_len
);
1302 con
->out_msg
->footer
.front_crc
= cpu_to_le32(crc
);
1304 crc
= crc32c(0, m
->middle
->vec
.iov_base
,
1305 m
->middle
->vec
.iov_len
);
1306 con
->out_msg
->footer
.middle_crc
= cpu_to_le32(crc
);
1308 con
->out_msg
->footer
.middle_crc
= 0;
1309 dout("%s front_crc %u middle_crc %u\n", __func__
,
1310 le32_to_cpu(con
->out_msg
->footer
.front_crc
),
1311 le32_to_cpu(con
->out_msg
->footer
.middle_crc
));
1312 con
->out_msg
->footer
.flags
= 0;
1314 /* is there a data payload? */
1315 con
->out_msg
->footer
.data_crc
= 0;
1316 if (m
->data_length
) {
1317 prepare_message_data(con
->out_msg
, m
->data_length
);
1318 con
->out_more
= 1; /* data + footer will follow */
1320 /* no, queue up footer too and be done */
1321 prepare_write_message_footer(con
);
1324 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1330 static void prepare_write_ack(struct ceph_connection
*con
)
1332 dout("prepare_write_ack %p %llu -> %llu\n", con
,
1333 con
->in_seq_acked
, con
->in_seq
);
1334 con
->in_seq_acked
= con
->in_seq
;
1336 con_out_kvec_reset(con
);
1338 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
1340 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1341 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1342 &con
->out_temp_ack
);
1344 con
->out_more
= 1; /* more will follow.. eventually.. */
1345 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1349 * Prepare to share the seq during handshake
1351 static void prepare_write_seq(struct ceph_connection
*con
)
1353 dout("prepare_write_seq %p %llu -> %llu\n", con
,
1354 con
->in_seq_acked
, con
->in_seq
);
1355 con
->in_seq_acked
= con
->in_seq
;
1357 con_out_kvec_reset(con
);
1359 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1360 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1361 &con
->out_temp_ack
);
1363 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1367 * Prepare to write keepalive byte.
1369 static void prepare_write_keepalive(struct ceph_connection
*con
)
1371 dout("prepare_write_keepalive %p\n", con
);
1372 con_out_kvec_reset(con
);
1373 if (con
->peer_features
& CEPH_FEATURE_MSGR_KEEPALIVE2
) {
1374 struct timespec now
= CURRENT_TIME
;
1376 con_out_kvec_add(con
, sizeof(tag_keepalive2
), &tag_keepalive2
);
1377 ceph_encode_timespec(&con
->out_temp_keepalive2
, &now
);
1378 con_out_kvec_add(con
, sizeof(con
->out_temp_keepalive2
),
1379 &con
->out_temp_keepalive2
);
1381 con_out_kvec_add(con
, sizeof(tag_keepalive
), &tag_keepalive
);
1383 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1387 * Connection negotiation.
1390 static struct ceph_auth_handshake
*get_connect_authorizer(struct ceph_connection
*con
,
1393 struct ceph_auth_handshake
*auth
;
1395 if (!con
->ops
->get_authorizer
) {
1396 con
->out_connect
.authorizer_protocol
= CEPH_AUTH_UNKNOWN
;
1397 con
->out_connect
.authorizer_len
= 0;
1401 /* Can't hold the mutex while getting authorizer */
1402 mutex_unlock(&con
->mutex
);
1403 auth
= con
->ops
->get_authorizer(con
, auth_proto
, con
->auth_retry
);
1404 mutex_lock(&con
->mutex
);
1408 if (con
->state
!= CON_STATE_NEGOTIATING
)
1409 return ERR_PTR(-EAGAIN
);
1411 con
->auth_reply_buf
= auth
->authorizer_reply_buf
;
1412 con
->auth_reply_buf_len
= auth
->authorizer_reply_buf_len
;
1417 * We connected to a peer and are saying hello.
1419 static void prepare_write_banner(struct ceph_connection
*con
)
1421 con_out_kvec_add(con
, strlen(CEPH_BANNER
), CEPH_BANNER
);
1422 con_out_kvec_add(con
, sizeof (con
->msgr
->my_enc_addr
),
1423 &con
->msgr
->my_enc_addr
);
1426 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1429 static int prepare_write_connect(struct ceph_connection
*con
)
1431 unsigned int global_seq
= get_global_seq(con
->msgr
, 0);
1434 struct ceph_auth_handshake
*auth
;
1436 switch (con
->peer_name
.type
) {
1437 case CEPH_ENTITY_TYPE_MON
:
1438 proto
= CEPH_MONC_PROTOCOL
;
1440 case CEPH_ENTITY_TYPE_OSD
:
1441 proto
= CEPH_OSDC_PROTOCOL
;
1443 case CEPH_ENTITY_TYPE_MDS
:
1444 proto
= CEPH_MDSC_PROTOCOL
;
1450 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con
,
1451 con
->connect_seq
, global_seq
, proto
);
1453 con
->out_connect
.features
=
1454 cpu_to_le64(from_msgr(con
->msgr
)->supported_features
);
1455 con
->out_connect
.host_type
= cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT
);
1456 con
->out_connect
.connect_seq
= cpu_to_le32(con
->connect_seq
);
1457 con
->out_connect
.global_seq
= cpu_to_le32(global_seq
);
1458 con
->out_connect
.protocol_version
= cpu_to_le32(proto
);
1459 con
->out_connect
.flags
= 0;
1461 auth_proto
= CEPH_AUTH_UNKNOWN
;
1462 auth
= get_connect_authorizer(con
, &auth_proto
);
1464 return PTR_ERR(auth
);
1466 con
->out_connect
.authorizer_protocol
= cpu_to_le32(auth_proto
);
1467 con
->out_connect
.authorizer_len
= auth
?
1468 cpu_to_le32(auth
->authorizer_buf_len
) : 0;
1470 con_out_kvec_add(con
, sizeof (con
->out_connect
),
1472 if (auth
&& auth
->authorizer_buf_len
)
1473 con_out_kvec_add(con
, auth
->authorizer_buf_len
,
1474 auth
->authorizer_buf
);
1477 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1483 * write as much of pending kvecs to the socket as we can.
1485 * 0 -> socket full, but more to do
1488 static int write_partial_kvec(struct ceph_connection
*con
)
1492 dout("write_partial_kvec %p %d left\n", con
, con
->out_kvec_bytes
);
1493 while (con
->out_kvec_bytes
> 0) {
1494 ret
= ceph_tcp_sendmsg(con
->sock
, con
->out_kvec_cur
,
1495 con
->out_kvec_left
, con
->out_kvec_bytes
,
1499 con
->out_kvec_bytes
-= ret
;
1500 if (con
->out_kvec_bytes
== 0)
1503 /* account for full iov entries consumed */
1504 while (ret
>= con
->out_kvec_cur
->iov_len
) {
1505 BUG_ON(!con
->out_kvec_left
);
1506 ret
-= con
->out_kvec_cur
->iov_len
;
1507 con
->out_kvec_cur
++;
1508 con
->out_kvec_left
--;
1510 /* and for a partially-consumed entry */
1512 con
->out_kvec_cur
->iov_len
-= ret
;
1513 con
->out_kvec_cur
->iov_base
+= ret
;
1516 con
->out_kvec_left
= 0;
1519 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con
,
1520 con
->out_kvec_bytes
, con
->out_kvec_left
, ret
);
1521 return ret
; /* done! */
1524 static u32
ceph_crc32c_page(u32 crc
, struct page
*page
,
1525 unsigned int page_offset
,
1526 unsigned int length
)
1531 BUG_ON(kaddr
== NULL
);
1532 crc
= crc32c(crc
, kaddr
+ page_offset
, length
);
1538 * Write as much message data payload as we can. If we finish, queue
1540 * 1 -> done, footer is now queued in out_kvec[].
1541 * 0 -> socket full, but more to do
1544 static int write_partial_message_data(struct ceph_connection
*con
)
1546 struct ceph_msg
*msg
= con
->out_msg
;
1547 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
1548 bool do_datacrc
= !ceph_test_opt(from_msgr(con
->msgr
), NOCRC
);
1551 dout("%s %p msg %p\n", __func__
, con
, msg
);
1553 if (list_empty(&msg
->data
))
1557 * Iterate through each page that contains data to be
1558 * written, and send as much as possible for each.
1560 * If we are calculating the data crc (the default), we will
1561 * need to map the page. If we have no pages, they have
1562 * been revoked, so use the zero page.
1564 crc
= do_datacrc
? le32_to_cpu(msg
->footer
.data_crc
) : 0;
1565 while (cursor
->resid
) {
1573 page
= ceph_msg_data_next(cursor
, &page_offset
, &length
,
1575 ret
= ceph_tcp_sendpage(con
->sock
, page
, page_offset
,
1576 length
, !last_piece
);
1579 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1583 if (do_datacrc
&& cursor
->need_crc
)
1584 crc
= ceph_crc32c_page(crc
, page
, page_offset
, length
);
1585 need_crc
= ceph_msg_data_advance(cursor
, (size_t)ret
);
1588 dout("%s %p msg %p done\n", __func__
, con
, msg
);
1590 /* prepare and queue up footer, too */
1592 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1594 msg
->footer
.flags
|= CEPH_MSG_FOOTER_NOCRC
;
1595 con_out_kvec_reset(con
);
1596 prepare_write_message_footer(con
);
1598 return 1; /* must return > 0 to indicate success */
1604 static int write_partial_skip(struct ceph_connection
*con
)
1608 dout("%s %p %d left\n", __func__
, con
, con
->out_skip
);
1609 while (con
->out_skip
> 0) {
1610 size_t size
= min(con
->out_skip
, (int) PAGE_CACHE_SIZE
);
1612 ret
= ceph_tcp_sendpage(con
->sock
, zero_page
, 0, size
, true);
1615 con
->out_skip
-= ret
;
1623 * Prepare to read connection handshake, or an ack.
1625 static void prepare_read_banner(struct ceph_connection
*con
)
1627 dout("prepare_read_banner %p\n", con
);
1628 con
->in_base_pos
= 0;
1631 static void prepare_read_connect(struct ceph_connection
*con
)
1633 dout("prepare_read_connect %p\n", con
);
1634 con
->in_base_pos
= 0;
1637 static void prepare_read_ack(struct ceph_connection
*con
)
1639 dout("prepare_read_ack %p\n", con
);
1640 con
->in_base_pos
= 0;
1643 static void prepare_read_seq(struct ceph_connection
*con
)
1645 dout("prepare_read_seq %p\n", con
);
1646 con
->in_base_pos
= 0;
1647 con
->in_tag
= CEPH_MSGR_TAG_SEQ
;
1650 static void prepare_read_tag(struct ceph_connection
*con
)
1652 dout("prepare_read_tag %p\n", con
);
1653 con
->in_base_pos
= 0;
1654 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1657 static void prepare_read_keepalive_ack(struct ceph_connection
*con
)
1659 dout("prepare_read_keepalive_ack %p\n", con
);
1660 con
->in_base_pos
= 0;
1664 * Prepare to read a message.
1666 static int prepare_read_message(struct ceph_connection
*con
)
1668 dout("prepare_read_message %p\n", con
);
1669 BUG_ON(con
->in_msg
!= NULL
);
1670 con
->in_base_pos
= 0;
1671 con
->in_front_crc
= con
->in_middle_crc
= con
->in_data_crc
= 0;
1676 static int read_partial(struct ceph_connection
*con
,
1677 int end
, int size
, void *object
)
1679 while (con
->in_base_pos
< end
) {
1680 int left
= end
- con
->in_base_pos
;
1681 int have
= size
- left
;
1682 int ret
= ceph_tcp_recvmsg(con
->sock
, object
+ have
, left
);
1685 con
->in_base_pos
+= ret
;
1692 * Read all or part of the connect-side handshake on a new connection
1694 static int read_partial_banner(struct ceph_connection
*con
)
1700 dout("read_partial_banner %p at %d\n", con
, con
->in_base_pos
);
1703 size
= strlen(CEPH_BANNER
);
1705 ret
= read_partial(con
, end
, size
, con
->in_banner
);
1709 size
= sizeof (con
->actual_peer_addr
);
1711 ret
= read_partial(con
, end
, size
, &con
->actual_peer_addr
);
1715 size
= sizeof (con
->peer_addr_for_me
);
1717 ret
= read_partial(con
, end
, size
, &con
->peer_addr_for_me
);
1725 static int read_partial_connect(struct ceph_connection
*con
)
1731 dout("read_partial_connect %p at %d\n", con
, con
->in_base_pos
);
1733 size
= sizeof (con
->in_reply
);
1735 ret
= read_partial(con
, end
, size
, &con
->in_reply
);
1739 size
= le32_to_cpu(con
->in_reply
.authorizer_len
);
1741 ret
= read_partial(con
, end
, size
, con
->auth_reply_buf
);
1745 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1746 con
, (int)con
->in_reply
.tag
,
1747 le32_to_cpu(con
->in_reply
.connect_seq
),
1748 le32_to_cpu(con
->in_reply
.global_seq
));
1755 * Verify the hello banner looks okay.
1757 static int verify_hello(struct ceph_connection
*con
)
1759 if (memcmp(con
->in_banner
, CEPH_BANNER
, strlen(CEPH_BANNER
))) {
1760 pr_err("connect to %s got bad banner\n",
1761 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1762 con
->error_msg
= "protocol error, bad banner";
1768 static bool addr_is_blank(struct sockaddr_storage
*ss
)
1770 struct in_addr
*addr
= &((struct sockaddr_in
*)ss
)->sin_addr
;
1771 struct in6_addr
*addr6
= &((struct sockaddr_in6
*)ss
)->sin6_addr
;
1773 switch (ss
->ss_family
) {
1775 return addr
->s_addr
== htonl(INADDR_ANY
);
1777 return ipv6_addr_any(addr6
);
1783 static int addr_port(struct sockaddr_storage
*ss
)
1785 switch (ss
->ss_family
) {
1787 return ntohs(((struct sockaddr_in
*)ss
)->sin_port
);
1789 return ntohs(((struct sockaddr_in6
*)ss
)->sin6_port
);
1794 static void addr_set_port(struct sockaddr_storage
*ss
, int p
)
1796 switch (ss
->ss_family
) {
1798 ((struct sockaddr_in
*)ss
)->sin_port
= htons(p
);
1801 ((struct sockaddr_in6
*)ss
)->sin6_port
= htons(p
);
1807 * Unlike other *_pton function semantics, zero indicates success.
1809 static int ceph_pton(const char *str
, size_t len
, struct sockaddr_storage
*ss
,
1810 char delim
, const char **ipend
)
1812 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
1813 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
1815 memset(ss
, 0, sizeof(*ss
));
1817 if (in4_pton(str
, len
, (u8
*)&in4
->sin_addr
.s_addr
, delim
, ipend
)) {
1818 ss
->ss_family
= AF_INET
;
1822 if (in6_pton(str
, len
, (u8
*)&in6
->sin6_addr
.s6_addr
, delim
, ipend
)) {
1823 ss
->ss_family
= AF_INET6
;
1831 * Extract hostname string and resolve using kernel DNS facility.
1833 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1834 static int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1835 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1837 const char *end
, *delim_p
;
1838 char *colon_p
, *ip_addr
= NULL
;
1842 * The end of the hostname occurs immediately preceding the delimiter or
1843 * the port marker (':') where the delimiter takes precedence.
1845 delim_p
= memchr(name
, delim
, namelen
);
1846 colon_p
= memchr(name
, ':', namelen
);
1848 if (delim_p
&& colon_p
)
1849 end
= delim_p
< colon_p
? delim_p
: colon_p
;
1850 else if (!delim_p
&& colon_p
)
1854 if (!end
) /* case: hostname:/ */
1855 end
= name
+ namelen
;
1861 /* do dns_resolve upcall */
1862 ip_len
= dns_query(NULL
, name
, end
- name
, NULL
, &ip_addr
, NULL
);
1864 ret
= ceph_pton(ip_addr
, ip_len
, ss
, -1, NULL
);
1872 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end
- name
), name
,
1873 ret
, ret
? "failed" : ceph_pr_addr(ss
));
1878 static inline int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1879 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1886 * Parse a server name (IP or hostname). If a valid IP address is not found
1887 * then try to extract a hostname to resolve using userspace DNS upcall.
1889 static int ceph_parse_server_name(const char *name
, size_t namelen
,
1890 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1894 ret
= ceph_pton(name
, namelen
, ss
, delim
, ipend
);
1896 ret
= ceph_dns_resolve_name(name
, namelen
, ss
, delim
, ipend
);
1902 * Parse an ip[:port] list into an addr array. Use the default
1903 * monitor port if a port isn't specified.
1905 int ceph_parse_ips(const char *c
, const char *end
,
1906 struct ceph_entity_addr
*addr
,
1907 int max_count
, int *count
)
1909 int i
, ret
= -EINVAL
;
1912 dout("parse_ips on '%.*s'\n", (int)(end
-c
), c
);
1913 for (i
= 0; i
< max_count
; i
++) {
1915 struct sockaddr_storage
*ss
= &addr
[i
].in_addr
;
1924 ret
= ceph_parse_server_name(p
, end
- p
, ss
, delim
, &ipend
);
1933 dout("missing matching ']'\n");
1940 if (p
< end
&& *p
== ':') {
1943 while (p
< end
&& *p
>= '0' && *p
<= '9') {
1944 port
= (port
* 10) + (*p
- '0');
1948 port
= CEPH_MON_PORT
;
1949 else if (port
> 65535)
1952 port
= CEPH_MON_PORT
;
1955 addr_set_port(ss
, port
);
1957 dout("parse_ips got %s\n", ceph_pr_addr(ss
));
1974 pr_err("parse_ips bad ip '%.*s'\n", (int)(end
- c
), c
);
1977 EXPORT_SYMBOL(ceph_parse_ips
);
1979 static int process_banner(struct ceph_connection
*con
)
1981 dout("process_banner on %p\n", con
);
1983 if (verify_hello(con
) < 0)
1986 ceph_decode_addr(&con
->actual_peer_addr
);
1987 ceph_decode_addr(&con
->peer_addr_for_me
);
1990 * Make sure the other end is who we wanted. note that the other
1991 * end may not yet know their ip address, so if it's 0.0.0.0, give
1992 * them the benefit of the doubt.
1994 if (memcmp(&con
->peer_addr
, &con
->actual_peer_addr
,
1995 sizeof(con
->peer_addr
)) != 0 &&
1996 !(addr_is_blank(&con
->actual_peer_addr
.in_addr
) &&
1997 con
->actual_peer_addr
.nonce
== con
->peer_addr
.nonce
)) {
1998 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
1999 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2000 (int)le32_to_cpu(con
->peer_addr
.nonce
),
2001 ceph_pr_addr(&con
->actual_peer_addr
.in_addr
),
2002 (int)le32_to_cpu(con
->actual_peer_addr
.nonce
));
2003 con
->error_msg
= "wrong peer at address";
2008 * did we learn our address?
2010 if (addr_is_blank(&con
->msgr
->inst
.addr
.in_addr
)) {
2011 int port
= addr_port(&con
->msgr
->inst
.addr
.in_addr
);
2013 memcpy(&con
->msgr
->inst
.addr
.in_addr
,
2014 &con
->peer_addr_for_me
.in_addr
,
2015 sizeof(con
->peer_addr_for_me
.in_addr
));
2016 addr_set_port(&con
->msgr
->inst
.addr
.in_addr
, port
);
2017 encode_my_addr(con
->msgr
);
2018 dout("process_banner learned my addr is %s\n",
2019 ceph_pr_addr(&con
->msgr
->inst
.addr
.in_addr
));
2025 static int process_connect(struct ceph_connection
*con
)
2027 u64 sup_feat
= from_msgr(con
->msgr
)->supported_features
;
2028 u64 req_feat
= from_msgr(con
->msgr
)->required_features
;
2029 u64 server_feat
= ceph_sanitize_features(
2030 le64_to_cpu(con
->in_reply
.features
));
2033 dout("process_connect on %p tag %d\n", con
, (int)con
->in_tag
);
2035 switch (con
->in_reply
.tag
) {
2036 case CEPH_MSGR_TAG_FEATURES
:
2037 pr_err("%s%lld %s feature set mismatch,"
2038 " my %llx < server's %llx, missing %llx\n",
2039 ENTITY_NAME(con
->peer_name
),
2040 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2041 sup_feat
, server_feat
, server_feat
& ~sup_feat
);
2042 con
->error_msg
= "missing required protocol features";
2043 reset_connection(con
);
2046 case CEPH_MSGR_TAG_BADPROTOVER
:
2047 pr_err("%s%lld %s protocol version mismatch,"
2048 " my %d != server's %d\n",
2049 ENTITY_NAME(con
->peer_name
),
2050 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2051 le32_to_cpu(con
->out_connect
.protocol_version
),
2052 le32_to_cpu(con
->in_reply
.protocol_version
));
2053 con
->error_msg
= "protocol version mismatch";
2054 reset_connection(con
);
2057 case CEPH_MSGR_TAG_BADAUTHORIZER
:
2059 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con
,
2061 if (con
->auth_retry
== 2) {
2062 con
->error_msg
= "connect authorization failure";
2065 con_out_kvec_reset(con
);
2066 ret
= prepare_write_connect(con
);
2069 prepare_read_connect(con
);
2072 case CEPH_MSGR_TAG_RESETSESSION
:
2074 * If we connected with a large connect_seq but the peer
2075 * has no record of a session with us (no connection, or
2076 * connect_seq == 0), they will send RESETSESION to indicate
2077 * that they must have reset their session, and may have
2080 dout("process_connect got RESET peer seq %u\n",
2081 le32_to_cpu(con
->in_reply
.connect_seq
));
2082 pr_err("%s%lld %s connection reset\n",
2083 ENTITY_NAME(con
->peer_name
),
2084 ceph_pr_addr(&con
->peer_addr
.in_addr
));
2085 reset_connection(con
);
2086 con_out_kvec_reset(con
);
2087 ret
= prepare_write_connect(con
);
2090 prepare_read_connect(con
);
2092 /* Tell ceph about it. */
2093 mutex_unlock(&con
->mutex
);
2094 pr_info("reset on %s%lld\n", ENTITY_NAME(con
->peer_name
));
2095 if (con
->ops
->peer_reset
)
2096 con
->ops
->peer_reset(con
);
2097 mutex_lock(&con
->mutex
);
2098 if (con
->state
!= CON_STATE_NEGOTIATING
)
2102 case CEPH_MSGR_TAG_RETRY_SESSION
:
2104 * If we sent a smaller connect_seq than the peer has, try
2105 * again with a larger value.
2107 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2108 le32_to_cpu(con
->out_connect
.connect_seq
),
2109 le32_to_cpu(con
->in_reply
.connect_seq
));
2110 con
->connect_seq
= le32_to_cpu(con
->in_reply
.connect_seq
);
2111 con_out_kvec_reset(con
);
2112 ret
= prepare_write_connect(con
);
2115 prepare_read_connect(con
);
2118 case CEPH_MSGR_TAG_RETRY_GLOBAL
:
2120 * If we sent a smaller global_seq than the peer has, try
2121 * again with a larger value.
2123 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2124 con
->peer_global_seq
,
2125 le32_to_cpu(con
->in_reply
.global_seq
));
2126 get_global_seq(con
->msgr
,
2127 le32_to_cpu(con
->in_reply
.global_seq
));
2128 con_out_kvec_reset(con
);
2129 ret
= prepare_write_connect(con
);
2132 prepare_read_connect(con
);
2135 case CEPH_MSGR_TAG_SEQ
:
2136 case CEPH_MSGR_TAG_READY
:
2137 if (req_feat
& ~server_feat
) {
2138 pr_err("%s%lld %s protocol feature mismatch,"
2139 " my required %llx > server's %llx, need %llx\n",
2140 ENTITY_NAME(con
->peer_name
),
2141 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2142 req_feat
, server_feat
, req_feat
& ~server_feat
);
2143 con
->error_msg
= "missing required protocol features";
2144 reset_connection(con
);
2148 WARN_ON(con
->state
!= CON_STATE_NEGOTIATING
);
2149 con
->state
= CON_STATE_OPEN
;
2150 con
->auth_retry
= 0; /* we authenticated; clear flag */
2151 con
->peer_global_seq
= le32_to_cpu(con
->in_reply
.global_seq
);
2153 con
->peer_features
= server_feat
;
2154 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2155 con
->peer_global_seq
,
2156 le32_to_cpu(con
->in_reply
.connect_seq
),
2158 WARN_ON(con
->connect_seq
!=
2159 le32_to_cpu(con
->in_reply
.connect_seq
));
2161 if (con
->in_reply
.flags
& CEPH_MSG_CONNECT_LOSSY
)
2162 con_flag_set(con
, CON_FLAG_LOSSYTX
);
2164 con
->delay
= 0; /* reset backoff memory */
2166 if (con
->in_reply
.tag
== CEPH_MSGR_TAG_SEQ
) {
2167 prepare_write_seq(con
);
2168 prepare_read_seq(con
);
2170 prepare_read_tag(con
);
2174 case CEPH_MSGR_TAG_WAIT
:
2176 * If there is a connection race (we are opening
2177 * connections to each other), one of us may just have
2178 * to WAIT. This shouldn't happen if we are the
2181 con
->error_msg
= "protocol error, got WAIT as client";
2185 con
->error_msg
= "protocol error, garbage tag during connect";
2193 * read (part of) an ack
2195 static int read_partial_ack(struct ceph_connection
*con
)
2197 int size
= sizeof (con
->in_temp_ack
);
2200 return read_partial(con
, end
, size
, &con
->in_temp_ack
);
2204 * We can finally discard anything that's been acked.
2206 static void process_ack(struct ceph_connection
*con
)
2209 u64 ack
= le64_to_cpu(con
->in_temp_ack
);
2212 while (!list_empty(&con
->out_sent
)) {
2213 m
= list_first_entry(&con
->out_sent
, struct ceph_msg
,
2215 seq
= le64_to_cpu(m
->hdr
.seq
);
2218 dout("got ack for seq %llu type %d at %p\n", seq
,
2219 le16_to_cpu(m
->hdr
.type
), m
);
2220 m
->ack_stamp
= jiffies
;
2223 prepare_read_tag(con
);
2227 static int read_partial_message_section(struct ceph_connection
*con
,
2228 struct kvec
*section
,
2229 unsigned int sec_len
, u32
*crc
)
2235 while (section
->iov_len
< sec_len
) {
2236 BUG_ON(section
->iov_base
== NULL
);
2237 left
= sec_len
- section
->iov_len
;
2238 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)section
->iov_base
+
2239 section
->iov_len
, left
);
2242 section
->iov_len
+= ret
;
2244 if (section
->iov_len
== sec_len
)
2245 *crc
= crc32c(0, section
->iov_base
, section
->iov_len
);
2250 static int read_partial_msg_data(struct ceph_connection
*con
)
2252 struct ceph_msg
*msg
= con
->in_msg
;
2253 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
2254 bool do_datacrc
= !ceph_test_opt(from_msgr(con
->msgr
), NOCRC
);
2262 if (list_empty(&msg
->data
))
2266 crc
= con
->in_data_crc
;
2267 while (cursor
->resid
) {
2268 page
= ceph_msg_data_next(cursor
, &page_offset
, &length
, NULL
);
2269 ret
= ceph_tcp_recvpage(con
->sock
, page
, page_offset
, length
);
2272 con
->in_data_crc
= crc
;
2278 crc
= ceph_crc32c_page(crc
, page
, page_offset
, ret
);
2279 (void) ceph_msg_data_advance(cursor
, (size_t)ret
);
2282 con
->in_data_crc
= crc
;
2284 return 1; /* must return > 0 to indicate success */
2288 * read (part of) a message.
2290 static int ceph_con_in_msg_alloc(struct ceph_connection
*con
, int *skip
);
2292 static int read_partial_message(struct ceph_connection
*con
)
2294 struct ceph_msg
*m
= con
->in_msg
;
2298 unsigned int front_len
, middle_len
, data_len
;
2299 bool do_datacrc
= !ceph_test_opt(from_msgr(con
->msgr
), NOCRC
);
2300 bool need_sign
= (con
->peer_features
& CEPH_FEATURE_MSG_AUTH
);
2304 dout("read_partial_message con %p msg %p\n", con
, m
);
2307 size
= sizeof (con
->in_hdr
);
2309 ret
= read_partial(con
, end
, size
, &con
->in_hdr
);
2313 crc
= crc32c(0, &con
->in_hdr
, offsetof(struct ceph_msg_header
, crc
));
2314 if (cpu_to_le32(crc
) != con
->in_hdr
.crc
) {
2315 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
2316 crc
, con
->in_hdr
.crc
);
2320 front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
2321 if (front_len
> CEPH_MSG_MAX_FRONT_LEN
)
2323 middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
2324 if (middle_len
> CEPH_MSG_MAX_MIDDLE_LEN
)
2326 data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
2327 if (data_len
> CEPH_MSG_MAX_DATA_LEN
)
2331 seq
= le64_to_cpu(con
->in_hdr
.seq
);
2332 if ((s64
)seq
- (s64
)con
->in_seq
< 1) {
2333 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2334 ENTITY_NAME(con
->peer_name
),
2335 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2336 seq
, con
->in_seq
+ 1);
2337 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
2339 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2341 } else if ((s64
)seq
- (s64
)con
->in_seq
> 1) {
2342 pr_err("read_partial_message bad seq %lld expected %lld\n",
2343 seq
, con
->in_seq
+ 1);
2344 con
->error_msg
= "bad message sequence # for incoming message";
2348 /* allocate message? */
2352 dout("got hdr type %d front %d data %d\n", con
->in_hdr
.type
,
2353 front_len
, data_len
);
2354 ret
= ceph_con_in_msg_alloc(con
, &skip
);
2358 BUG_ON(!con
->in_msg
^ skip
);
2360 /* skip this message */
2361 dout("alloc_msg said skip message\n");
2362 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
2364 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2369 BUG_ON(!con
->in_msg
);
2370 BUG_ON(con
->in_msg
->con
!= con
);
2372 m
->front
.iov_len
= 0; /* haven't read it yet */
2374 m
->middle
->vec
.iov_len
= 0;
2376 /* prepare for data payload, if any */
2379 prepare_message_data(con
->in_msg
, data_len
);
2383 ret
= read_partial_message_section(con
, &m
->front
, front_len
,
2384 &con
->in_front_crc
);
2390 ret
= read_partial_message_section(con
, &m
->middle
->vec
,
2392 &con
->in_middle_crc
);
2399 ret
= read_partial_msg_data(con
);
2406 size
= sizeof(m
->footer
);
2408 size
= sizeof(m
->old_footer
);
2411 ret
= read_partial(con
, end
, size
, &m
->footer
);
2416 m
->footer
.flags
= m
->old_footer
.flags
;
2420 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2421 m
, front_len
, m
->footer
.front_crc
, middle_len
,
2422 m
->footer
.middle_crc
, data_len
, m
->footer
.data_crc
);
2425 if (con
->in_front_crc
!= le32_to_cpu(m
->footer
.front_crc
)) {
2426 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2427 m
, con
->in_front_crc
, m
->footer
.front_crc
);
2430 if (con
->in_middle_crc
!= le32_to_cpu(m
->footer
.middle_crc
)) {
2431 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2432 m
, con
->in_middle_crc
, m
->footer
.middle_crc
);
2436 (m
->footer
.flags
& CEPH_MSG_FOOTER_NOCRC
) == 0 &&
2437 con
->in_data_crc
!= le32_to_cpu(m
->footer
.data_crc
)) {
2438 pr_err("read_partial_message %p data crc %u != exp. %u\n", m
,
2439 con
->in_data_crc
, le32_to_cpu(m
->footer
.data_crc
));
2443 if (need_sign
&& con
->ops
->check_message_signature
&&
2444 con
->ops
->check_message_signature(m
)) {
2445 pr_err("read_partial_message %p signature check failed\n", m
);
2449 return 1; /* done! */
2453 * Process message. This happens in the worker thread. The callback should
2454 * be careful not to do anything that waits on other incoming messages or it
2457 static void process_message(struct ceph_connection
*con
)
2459 struct ceph_msg
*msg
= con
->in_msg
;
2461 BUG_ON(con
->in_msg
->con
!= con
);
2464 /* if first message, set peer_name */
2465 if (con
->peer_name
.type
== 0)
2466 con
->peer_name
= msg
->hdr
.src
;
2469 mutex_unlock(&con
->mutex
);
2471 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2472 msg
, le64_to_cpu(msg
->hdr
.seq
),
2473 ENTITY_NAME(msg
->hdr
.src
),
2474 le16_to_cpu(msg
->hdr
.type
),
2475 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2476 le32_to_cpu(msg
->hdr
.front_len
),
2477 le32_to_cpu(msg
->hdr
.data_len
),
2478 con
->in_front_crc
, con
->in_middle_crc
, con
->in_data_crc
);
2479 con
->ops
->dispatch(con
, msg
);
2481 mutex_lock(&con
->mutex
);
2484 static int read_keepalive_ack(struct ceph_connection
*con
)
2486 struct ceph_timespec ceph_ts
;
2487 size_t size
= sizeof(ceph_ts
);
2488 int ret
= read_partial(con
, size
, size
, &ceph_ts
);
2491 ceph_decode_timespec(&con
->last_keepalive_ack
, &ceph_ts
);
2492 prepare_read_tag(con
);
2497 * Write something to the socket. Called in a worker thread when the
2498 * socket appears to be writeable and we have something ready to send.
2500 static int try_write(struct ceph_connection
*con
)
2504 dout("try_write start %p state %lu\n", con
, con
->state
);
2507 dout("try_write out_kvec_bytes %d\n", con
->out_kvec_bytes
);
2509 /* open the socket first? */
2510 if (con
->state
== CON_STATE_PREOPEN
) {
2512 con
->state
= CON_STATE_CONNECTING
;
2514 con_out_kvec_reset(con
);
2515 prepare_write_banner(con
);
2516 prepare_read_banner(con
);
2518 BUG_ON(con
->in_msg
);
2519 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2520 dout("try_write initiating connect on %p new state %lu\n",
2522 ret
= ceph_tcp_connect(con
);
2524 con
->error_msg
= "connect error";
2530 /* kvec data queued? */
2531 if (con
->out_kvec_left
) {
2532 ret
= write_partial_kvec(con
);
2536 if (con
->out_skip
) {
2537 ret
= write_partial_skip(con
);
2544 if (con
->out_msg_done
) {
2545 ceph_msg_put(con
->out_msg
);
2546 con
->out_msg
= NULL
; /* we're done with this one */
2550 ret
= write_partial_message_data(con
);
2552 goto more_kvec
; /* we need to send the footer, too! */
2556 dout("try_write write_partial_message_data err %d\n",
2563 if (con
->state
== CON_STATE_OPEN
) {
2564 if (con_flag_test_and_clear(con
, CON_FLAG_KEEPALIVE_PENDING
)) {
2565 prepare_write_keepalive(con
);
2568 /* is anything else pending? */
2569 if (!list_empty(&con
->out_queue
)) {
2570 prepare_write_message(con
);
2573 if (con
->in_seq
> con
->in_seq_acked
) {
2574 prepare_write_ack(con
);
2579 /* Nothing to do! */
2580 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
2581 dout("try_write nothing else to write.\n");
2584 dout("try_write done on %p ret %d\n", con
, ret
);
2591 * Read what we can from the socket.
2593 static int try_read(struct ceph_connection
*con
)
2598 dout("try_read start on %p state %lu\n", con
, con
->state
);
2599 if (con
->state
!= CON_STATE_CONNECTING
&&
2600 con
->state
!= CON_STATE_NEGOTIATING
&&
2601 con
->state
!= CON_STATE_OPEN
)
2606 dout("try_read tag %d in_base_pos %d\n", (int)con
->in_tag
,
2609 if (con
->state
== CON_STATE_CONNECTING
) {
2610 dout("try_read connecting\n");
2611 ret
= read_partial_banner(con
);
2614 ret
= process_banner(con
);
2618 con
->state
= CON_STATE_NEGOTIATING
;
2621 * Received banner is good, exchange connection info.
2622 * Do not reset out_kvec, as sending our banner raced
2623 * with receiving peer banner after connect completed.
2625 ret
= prepare_write_connect(con
);
2628 prepare_read_connect(con
);
2630 /* Send connection info before awaiting response */
2634 if (con
->state
== CON_STATE_NEGOTIATING
) {
2635 dout("try_read negotiating\n");
2636 ret
= read_partial_connect(con
);
2639 ret
= process_connect(con
);
2645 WARN_ON(con
->state
!= CON_STATE_OPEN
);
2647 if (con
->in_base_pos
< 0) {
2649 * skipping + discarding content.
2651 * FIXME: there must be a better way to do this!
2653 static char buf
[SKIP_BUF_SIZE
];
2654 int skip
= min((int) sizeof (buf
), -con
->in_base_pos
);
2656 dout("skipping %d / %d bytes\n", skip
, -con
->in_base_pos
);
2657 ret
= ceph_tcp_recvmsg(con
->sock
, buf
, skip
);
2660 con
->in_base_pos
+= ret
;
2661 if (con
->in_base_pos
)
2664 if (con
->in_tag
== CEPH_MSGR_TAG_READY
) {
2668 ret
= ceph_tcp_recvmsg(con
->sock
, &con
->in_tag
, 1);
2671 dout("try_read got tag %d\n", (int)con
->in_tag
);
2672 switch (con
->in_tag
) {
2673 case CEPH_MSGR_TAG_MSG
:
2674 prepare_read_message(con
);
2676 case CEPH_MSGR_TAG_ACK
:
2677 prepare_read_ack(con
);
2679 case CEPH_MSGR_TAG_KEEPALIVE2_ACK
:
2680 prepare_read_keepalive_ack(con
);
2682 case CEPH_MSGR_TAG_CLOSE
:
2683 con_close_socket(con
);
2684 con
->state
= CON_STATE_CLOSED
;
2690 if (con
->in_tag
== CEPH_MSGR_TAG_MSG
) {
2691 ret
= read_partial_message(con
);
2695 con
->error_msg
= "bad crc/signature";
2701 con
->error_msg
= "io error";
2706 if (con
->in_tag
== CEPH_MSGR_TAG_READY
)
2708 process_message(con
);
2709 if (con
->state
== CON_STATE_OPEN
)
2710 prepare_read_tag(con
);
2713 if (con
->in_tag
== CEPH_MSGR_TAG_ACK
||
2714 con
->in_tag
== CEPH_MSGR_TAG_SEQ
) {
2716 * the final handshake seq exchange is semantically
2717 * equivalent to an ACK
2719 ret
= read_partial_ack(con
);
2725 if (con
->in_tag
== CEPH_MSGR_TAG_KEEPALIVE2_ACK
) {
2726 ret
= read_keepalive_ack(con
);
2733 dout("try_read done on %p ret %d\n", con
, ret
);
2737 pr_err("try_read bad con->in_tag = %d\n", (int)con
->in_tag
);
2738 con
->error_msg
= "protocol error, garbage tag";
2745 * Atomically queue work on a connection after the specified delay.
2746 * Bump @con reference to avoid races with connection teardown.
2747 * Returns 0 if work was queued, or an error code otherwise.
2749 static int queue_con_delay(struct ceph_connection
*con
, unsigned long delay
)
2751 if (!con
->ops
->get(con
)) {
2752 dout("%s %p ref count 0\n", __func__
, con
);
2756 if (!queue_delayed_work(ceph_msgr_wq
, &con
->work
, delay
)) {
2757 dout("%s %p - already queued\n", __func__
, con
);
2762 dout("%s %p %lu\n", __func__
, con
, delay
);
2766 static void queue_con(struct ceph_connection
*con
)
2768 (void) queue_con_delay(con
, 0);
2771 static void cancel_con(struct ceph_connection
*con
)
2773 if (cancel_delayed_work(&con
->work
)) {
2774 dout("%s %p\n", __func__
, con
);
2779 static bool con_sock_closed(struct ceph_connection
*con
)
2781 if (!con_flag_test_and_clear(con
, CON_FLAG_SOCK_CLOSED
))
2785 case CON_STATE_ ## x: \
2786 con->error_msg = "socket closed (con state " #x ")"; \
2789 switch (con
->state
) {
2797 pr_warn("%s con %p unrecognized state %lu\n",
2798 __func__
, con
, con
->state
);
2799 con
->error_msg
= "unrecognized con state";
2808 static bool con_backoff(struct ceph_connection
*con
)
2812 if (!con_flag_test_and_clear(con
, CON_FLAG_BACKOFF
))
2815 ret
= queue_con_delay(con
, round_jiffies_relative(con
->delay
));
2817 dout("%s: con %p FAILED to back off %lu\n", __func__
,
2819 BUG_ON(ret
== -ENOENT
);
2820 con_flag_set(con
, CON_FLAG_BACKOFF
);
2826 /* Finish fault handling; con->mutex must *not* be held here */
2828 static void con_fault_finish(struct ceph_connection
*con
)
2830 dout("%s %p\n", __func__
, con
);
2833 * in case we faulted due to authentication, invalidate our
2834 * current tickets so that we can get new ones.
2836 if (con
->auth_retry
) {
2837 dout("auth_retry %d, invalidating\n", con
->auth_retry
);
2838 if (con
->ops
->invalidate_authorizer
)
2839 con
->ops
->invalidate_authorizer(con
);
2840 con
->auth_retry
= 0;
2843 if (con
->ops
->fault
)
2844 con
->ops
->fault(con
);
2848 * Do some work on a connection. Drop a connection ref when we're done.
2850 static void ceph_con_workfn(struct work_struct
*work
)
2852 struct ceph_connection
*con
= container_of(work
, struct ceph_connection
,
2856 mutex_lock(&con
->mutex
);
2860 if ((fault
= con_sock_closed(con
))) {
2861 dout("%s: con %p SOCK_CLOSED\n", __func__
, con
);
2864 if (con_backoff(con
)) {
2865 dout("%s: con %p BACKOFF\n", __func__
, con
);
2868 if (con
->state
== CON_STATE_STANDBY
) {
2869 dout("%s: con %p STANDBY\n", __func__
, con
);
2872 if (con
->state
== CON_STATE_CLOSED
) {
2873 dout("%s: con %p CLOSED\n", __func__
, con
);
2877 if (con
->state
== CON_STATE_PREOPEN
) {
2878 dout("%s: con %p PREOPEN\n", __func__
, con
);
2882 ret
= try_read(con
);
2886 if (!con
->error_msg
)
2887 con
->error_msg
= "socket error on read";
2892 ret
= try_write(con
);
2896 if (!con
->error_msg
)
2897 con
->error_msg
= "socket error on write";
2901 break; /* If we make it to here, we're done */
2905 mutex_unlock(&con
->mutex
);
2908 con_fault_finish(con
);
2914 * Generic error/fault handler. A retry mechanism is used with
2915 * exponential backoff
2917 static void con_fault(struct ceph_connection
*con
)
2919 dout("fault %p state %lu to peer %s\n",
2920 con
, con
->state
, ceph_pr_addr(&con
->peer_addr
.in_addr
));
2922 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con
->peer_name
),
2923 ceph_pr_addr(&con
->peer_addr
.in_addr
), con
->error_msg
);
2924 con
->error_msg
= NULL
;
2926 WARN_ON(con
->state
!= CON_STATE_CONNECTING
&&
2927 con
->state
!= CON_STATE_NEGOTIATING
&&
2928 con
->state
!= CON_STATE_OPEN
);
2930 con_close_socket(con
);
2932 if (con_flag_test(con
, CON_FLAG_LOSSYTX
)) {
2933 dout("fault on LOSSYTX channel, marking CLOSED\n");
2934 con
->state
= CON_STATE_CLOSED
;
2939 BUG_ON(con
->in_msg
->con
!= con
);
2940 ceph_msg_put(con
->in_msg
);
2944 /* Requeue anything that hasn't been acked */
2945 list_splice_init(&con
->out_sent
, &con
->out_queue
);
2947 /* If there are no messages queued or keepalive pending, place
2948 * the connection in a STANDBY state */
2949 if (list_empty(&con
->out_queue
) &&
2950 !con_flag_test(con
, CON_FLAG_KEEPALIVE_PENDING
)) {
2951 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con
);
2952 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
2953 con
->state
= CON_STATE_STANDBY
;
2955 /* retry after a delay. */
2956 con
->state
= CON_STATE_PREOPEN
;
2957 if (con
->delay
== 0)
2958 con
->delay
= BASE_DELAY_INTERVAL
;
2959 else if (con
->delay
< MAX_DELAY_INTERVAL
)
2961 con_flag_set(con
, CON_FLAG_BACKOFF
);
2969 * initialize a new messenger instance
2971 void ceph_messenger_init(struct ceph_messenger
*msgr
,
2972 struct ceph_entity_addr
*myaddr
)
2974 spin_lock_init(&msgr
->global_seq_lock
);
2977 msgr
->inst
.addr
= *myaddr
;
2979 /* select a random nonce */
2980 msgr
->inst
.addr
.type
= 0;
2981 get_random_bytes(&msgr
->inst
.addr
.nonce
, sizeof(msgr
->inst
.addr
.nonce
));
2982 encode_my_addr(msgr
);
2984 atomic_set(&msgr
->stopping
, 0);
2985 write_pnet(&msgr
->net
, get_net(current
->nsproxy
->net_ns
));
2987 dout("%s %p\n", __func__
, msgr
);
2989 EXPORT_SYMBOL(ceph_messenger_init
);
2991 void ceph_messenger_fini(struct ceph_messenger
*msgr
)
2993 put_net(read_pnet(&msgr
->net
));
2995 EXPORT_SYMBOL(ceph_messenger_fini
);
2997 static void msg_con_set(struct ceph_msg
*msg
, struct ceph_connection
*con
)
3000 msg
->con
->ops
->put(msg
->con
);
3002 msg
->con
= con
? con
->ops
->get(con
) : NULL
;
3003 BUG_ON(msg
->con
!= con
);
3006 static void clear_standby(struct ceph_connection
*con
)
3008 /* come back from STANDBY? */
3009 if (con
->state
== CON_STATE_STANDBY
) {
3010 dout("clear_standby %p and ++connect_seq\n", con
);
3011 con
->state
= CON_STATE_PREOPEN
;
3013 WARN_ON(con_flag_test(con
, CON_FLAG_WRITE_PENDING
));
3014 WARN_ON(con_flag_test(con
, CON_FLAG_KEEPALIVE_PENDING
));
3019 * Queue up an outgoing message on the given connection.
3021 void ceph_con_send(struct ceph_connection
*con
, struct ceph_msg
*msg
)
3024 msg
->hdr
.src
= con
->msgr
->inst
.name
;
3025 BUG_ON(msg
->front
.iov_len
!= le32_to_cpu(msg
->hdr
.front_len
));
3026 msg
->needs_out_seq
= true;
3028 mutex_lock(&con
->mutex
);
3030 if (con
->state
== CON_STATE_CLOSED
) {
3031 dout("con_send %p closed, dropping %p\n", con
, msg
);
3033 mutex_unlock(&con
->mutex
);
3037 msg_con_set(msg
, con
);
3039 BUG_ON(!list_empty(&msg
->list_head
));
3040 list_add_tail(&msg
->list_head
, &con
->out_queue
);
3041 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg
,
3042 ENTITY_NAME(con
->peer_name
), le16_to_cpu(msg
->hdr
.type
),
3043 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
3044 le32_to_cpu(msg
->hdr
.front_len
),
3045 le32_to_cpu(msg
->hdr
.middle_len
),
3046 le32_to_cpu(msg
->hdr
.data_len
));
3049 mutex_unlock(&con
->mutex
);
3051 /* if there wasn't anything waiting to send before, queue
3053 if (con_flag_test_and_set(con
, CON_FLAG_WRITE_PENDING
) == 0)
3056 EXPORT_SYMBOL(ceph_con_send
);
3059 * Revoke a message that was previously queued for send
3061 void ceph_msg_revoke(struct ceph_msg
*msg
)
3063 struct ceph_connection
*con
= msg
->con
;
3066 dout("%s msg %p null con\n", __func__
, msg
);
3067 return; /* Message not in our possession */
3070 mutex_lock(&con
->mutex
);
3071 if (!list_empty(&msg
->list_head
)) {
3072 dout("%s %p msg %p - was on queue\n", __func__
, con
, msg
);
3073 list_del_init(&msg
->list_head
);
3078 if (con
->out_msg
== msg
) {
3079 BUG_ON(con
->out_skip
);
3081 if (con
->out_msg_done
) {
3082 con
->out_skip
+= con_out_kvec_skip(con
);
3084 BUG_ON(!msg
->data_length
);
3085 if (con
->peer_features
& CEPH_FEATURE_MSG_AUTH
)
3086 con
->out_skip
+= sizeof(msg
->footer
);
3088 con
->out_skip
+= sizeof(msg
->old_footer
);
3090 /* data, middle, front */
3091 if (msg
->data_length
)
3092 con
->out_skip
+= msg
->cursor
.total_resid
;
3094 con
->out_skip
+= con_out_kvec_skip(con
);
3095 con
->out_skip
+= con_out_kvec_skip(con
);
3097 dout("%s %p msg %p - was sending, will write %d skip %d\n",
3098 __func__
, con
, msg
, con
->out_kvec_bytes
, con
->out_skip
);
3100 con
->out_msg
= NULL
;
3104 mutex_unlock(&con
->mutex
);
3108 * Revoke a message that we may be reading data into
3110 void ceph_msg_revoke_incoming(struct ceph_msg
*msg
)
3112 struct ceph_connection
*con
= msg
->con
;
3115 dout("%s msg %p null con\n", __func__
, msg
);
3116 return; /* Message not in our possession */
3119 mutex_lock(&con
->mutex
);
3120 if (con
->in_msg
== msg
) {
3121 unsigned int front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
3122 unsigned int middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
3123 unsigned int data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
3125 /* skip rest of message */
3126 dout("%s %p msg %p revoked\n", __func__
, con
, msg
);
3127 con
->in_base_pos
= con
->in_base_pos
-
3128 sizeof(struct ceph_msg_header
) -
3132 sizeof(struct ceph_msg_footer
);
3133 ceph_msg_put(con
->in_msg
);
3135 con
->in_tag
= CEPH_MSGR_TAG_READY
;
3138 dout("%s %p in_msg %p msg %p no-op\n",
3139 __func__
, con
, con
->in_msg
, msg
);
3141 mutex_unlock(&con
->mutex
);
3145 * Queue a keepalive byte to ensure the tcp connection is alive.
3147 void ceph_con_keepalive(struct ceph_connection
*con
)
3149 dout("con_keepalive %p\n", con
);
3150 mutex_lock(&con
->mutex
);
3152 mutex_unlock(&con
->mutex
);
3153 if (con_flag_test_and_set(con
, CON_FLAG_KEEPALIVE_PENDING
) == 0 &&
3154 con_flag_test_and_set(con
, CON_FLAG_WRITE_PENDING
) == 0)
3157 EXPORT_SYMBOL(ceph_con_keepalive
);
3159 bool ceph_con_keepalive_expired(struct ceph_connection
*con
,
3160 unsigned long interval
)
3163 (con
->peer_features
& CEPH_FEATURE_MSGR_KEEPALIVE2
)) {
3164 struct timespec now
= CURRENT_TIME
;
3166 jiffies_to_timespec(interval
, &ts
);
3167 ts
= timespec_add(con
->last_keepalive_ack
, ts
);
3168 return timespec_compare(&now
, &ts
) >= 0;
3173 static struct ceph_msg_data
*ceph_msg_data_create(enum ceph_msg_data_type type
)
3175 struct ceph_msg_data
*data
;
3177 if (WARN_ON(!ceph_msg_data_type_valid(type
)))
3180 data
= kmem_cache_zalloc(ceph_msg_data_cache
, GFP_NOFS
);
3183 INIT_LIST_HEAD(&data
->links
);
3188 static void ceph_msg_data_destroy(struct ceph_msg_data
*data
)
3193 WARN_ON(!list_empty(&data
->links
));
3194 if (data
->type
== CEPH_MSG_DATA_PAGELIST
)
3195 ceph_pagelist_release(data
->pagelist
);
3196 kmem_cache_free(ceph_msg_data_cache
, data
);
3199 void ceph_msg_data_add_pages(struct ceph_msg
*msg
, struct page
**pages
,
3200 size_t length
, size_t alignment
)
3202 struct ceph_msg_data
*data
;
3207 data
= ceph_msg_data_create(CEPH_MSG_DATA_PAGES
);
3209 data
->pages
= pages
;
3210 data
->length
= length
;
3211 data
->alignment
= alignment
& ~PAGE_MASK
;
3213 list_add_tail(&data
->links
, &msg
->data
);
3214 msg
->data_length
+= length
;
3216 EXPORT_SYMBOL(ceph_msg_data_add_pages
);
3218 void ceph_msg_data_add_pagelist(struct ceph_msg
*msg
,
3219 struct ceph_pagelist
*pagelist
)
3221 struct ceph_msg_data
*data
;
3224 BUG_ON(!pagelist
->length
);
3226 data
= ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST
);
3228 data
->pagelist
= pagelist
;
3230 list_add_tail(&data
->links
, &msg
->data
);
3231 msg
->data_length
+= pagelist
->length
;
3233 EXPORT_SYMBOL(ceph_msg_data_add_pagelist
);
3236 void ceph_msg_data_add_bio(struct ceph_msg
*msg
, struct bio
*bio
,
3239 struct ceph_msg_data
*data
;
3243 data
= ceph_msg_data_create(CEPH_MSG_DATA_BIO
);
3246 data
->bio_length
= length
;
3248 list_add_tail(&data
->links
, &msg
->data
);
3249 msg
->data_length
+= length
;
3251 EXPORT_SYMBOL(ceph_msg_data_add_bio
);
3252 #endif /* CONFIG_BLOCK */
3255 * construct a new message with given type, size
3256 * the new msg has a ref count of 1.
3258 struct ceph_msg
*ceph_msg_new(int type
, int front_len
, gfp_t flags
,
3263 m
= kmem_cache_zalloc(ceph_msg_cache
, flags
);
3267 m
->hdr
.type
= cpu_to_le16(type
);
3268 m
->hdr
.priority
= cpu_to_le16(CEPH_MSG_PRIO_DEFAULT
);
3269 m
->hdr
.front_len
= cpu_to_le32(front_len
);
3271 INIT_LIST_HEAD(&m
->list_head
);
3272 kref_init(&m
->kref
);
3273 INIT_LIST_HEAD(&m
->data
);
3277 m
->front
.iov_base
= ceph_kvmalloc(front_len
, flags
);
3278 if (m
->front
.iov_base
== NULL
) {
3279 dout("ceph_msg_new can't allocate %d bytes\n",
3284 m
->front
.iov_base
= NULL
;
3286 m
->front_alloc_len
= m
->front
.iov_len
= front_len
;
3288 dout("ceph_msg_new %p front %d\n", m
, front_len
);
3295 pr_err("msg_new can't create type %d front %d\n", type
,
3299 dout("msg_new can't create type %d front %d\n", type
,
3304 EXPORT_SYMBOL(ceph_msg_new
);
3307 * Allocate "middle" portion of a message, if it is needed and wasn't
3308 * allocated by alloc_msg. This allows us to read a small fixed-size
3309 * per-type header in the front and then gracefully fail (i.e.,
3310 * propagate the error to the caller based on info in the front) when
3311 * the middle is too large.
3313 static int ceph_alloc_middle(struct ceph_connection
*con
, struct ceph_msg
*msg
)
3315 int type
= le16_to_cpu(msg
->hdr
.type
);
3316 int middle_len
= le32_to_cpu(msg
->hdr
.middle_len
);
3318 dout("alloc_middle %p type %d %s middle_len %d\n", msg
, type
,
3319 ceph_msg_type_name(type
), middle_len
);
3320 BUG_ON(!middle_len
);
3321 BUG_ON(msg
->middle
);
3323 msg
->middle
= ceph_buffer_new(middle_len
, GFP_NOFS
);
3330 * Allocate a message for receiving an incoming message on a
3331 * connection, and save the result in con->in_msg. Uses the
3332 * connection's private alloc_msg op if available.
3334 * Returns 0 on success, or a negative error code.
3336 * On success, if we set *skip = 1:
3337 * - the next message should be skipped and ignored.
3338 * - con->in_msg == NULL
3339 * or if we set *skip = 0:
3340 * - con->in_msg is non-null.
3341 * On error (ENOMEM, EAGAIN, ...),
3342 * - con->in_msg == NULL
3344 static int ceph_con_in_msg_alloc(struct ceph_connection
*con
, int *skip
)
3346 struct ceph_msg_header
*hdr
= &con
->in_hdr
;
3347 int middle_len
= le32_to_cpu(hdr
->middle_len
);
3348 struct ceph_msg
*msg
;
3351 BUG_ON(con
->in_msg
!= NULL
);
3352 BUG_ON(!con
->ops
->alloc_msg
);
3354 mutex_unlock(&con
->mutex
);
3355 msg
= con
->ops
->alloc_msg(con
, hdr
, skip
);
3356 mutex_lock(&con
->mutex
);
3357 if (con
->state
!= CON_STATE_OPEN
) {
3364 msg_con_set(msg
, con
);
3368 * Null message pointer means either we should skip
3369 * this message or we couldn't allocate memory. The
3370 * former is not an error.
3375 con
->error_msg
= "error allocating memory for incoming message";
3378 memcpy(&con
->in_msg
->hdr
, &con
->in_hdr
, sizeof(con
->in_hdr
));
3380 if (middle_len
&& !con
->in_msg
->middle
) {
3381 ret
= ceph_alloc_middle(con
, con
->in_msg
);
3383 ceph_msg_put(con
->in_msg
);
3393 * Free a generically kmalloc'd message.
3395 static void ceph_msg_free(struct ceph_msg
*m
)
3397 dout("%s %p\n", __func__
, m
);
3398 kvfree(m
->front
.iov_base
);
3399 kmem_cache_free(ceph_msg_cache
, m
);
3402 static void ceph_msg_release(struct kref
*kref
)
3404 struct ceph_msg
*m
= container_of(kref
, struct ceph_msg
, kref
);
3405 struct ceph_msg_data
*data
, *next
;
3407 dout("%s %p\n", __func__
, m
);
3408 WARN_ON(!list_empty(&m
->list_head
));
3410 msg_con_set(m
, NULL
);
3412 /* drop middle, data, if any */
3414 ceph_buffer_put(m
->middle
);
3418 list_for_each_entry_safe(data
, next
, &m
->data
, links
) {
3419 list_del_init(&data
->links
);
3420 ceph_msg_data_destroy(data
);
3425 ceph_msgpool_put(m
->pool
, m
);
3430 struct ceph_msg
*ceph_msg_get(struct ceph_msg
*msg
)
3432 dout("%s %p (was %d)\n", __func__
, msg
,
3433 atomic_read(&msg
->kref
.refcount
));
3434 kref_get(&msg
->kref
);
3437 EXPORT_SYMBOL(ceph_msg_get
);
3439 void ceph_msg_put(struct ceph_msg
*msg
)
3441 dout("%s %p (was %d)\n", __func__
, msg
,
3442 atomic_read(&msg
->kref
.refcount
));
3443 kref_put(&msg
->kref
, ceph_msg_release
);
3445 EXPORT_SYMBOL(ceph_msg_put
);
3447 void ceph_msg_dump(struct ceph_msg
*msg
)
3449 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg
,
3450 msg
->front_alloc_len
, msg
->data_length
);
3451 print_hex_dump(KERN_DEBUG
, "header: ",
3452 DUMP_PREFIX_OFFSET
, 16, 1,
3453 &msg
->hdr
, sizeof(msg
->hdr
), true);
3454 print_hex_dump(KERN_DEBUG
, " front: ",
3455 DUMP_PREFIX_OFFSET
, 16, 1,
3456 msg
->front
.iov_base
, msg
->front
.iov_len
, true);
3458 print_hex_dump(KERN_DEBUG
, "middle: ",
3459 DUMP_PREFIX_OFFSET
, 16, 1,
3460 msg
->middle
->vec
.iov_base
,
3461 msg
->middle
->vec
.iov_len
, true);
3462 print_hex_dump(KERN_DEBUG
, "footer: ",
3463 DUMP_PREFIX_OFFSET
, 16, 1,
3464 &msg
->footer
, sizeof(msg
->footer
), true);
3466 EXPORT_SYMBOL(ceph_msg_dump
);