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
)
589 /* sendpage cannot properly handle pages with page_count == 0,
590 * we need to fallback to sendmsg if that's the case */
591 if (page_count(page
) >= 1)
592 return __ceph_tcp_sendpage(sock
, page
, offset
, size
, more
);
594 iov
.iov_base
= kmap(page
) + offset
;
596 ret
= ceph_tcp_sendmsg(sock
, &iov
, 1, size
, more
);
603 * Shutdown/close the socket for the given connection.
605 static int con_close_socket(struct ceph_connection
*con
)
609 dout("con_close_socket on %p sock %p\n", con
, con
->sock
);
611 rc
= con
->sock
->ops
->shutdown(con
->sock
, SHUT_RDWR
);
612 sock_release(con
->sock
);
617 * Forcibly clear the SOCK_CLOSED flag. It gets set
618 * independent of the connection mutex, and we could have
619 * received a socket close event before we had the chance to
620 * shut the socket down.
622 con_flag_clear(con
, CON_FLAG_SOCK_CLOSED
);
624 con_sock_state_closed(con
);
629 * Reset a connection. Discard all incoming and outgoing messages
630 * and clear *_seq state.
632 static void ceph_msg_remove(struct ceph_msg
*msg
)
634 list_del_init(&msg
->list_head
);
638 static void ceph_msg_remove_list(struct list_head
*head
)
640 while (!list_empty(head
)) {
641 struct ceph_msg
*msg
= list_first_entry(head
, struct ceph_msg
,
643 ceph_msg_remove(msg
);
647 static void reset_connection(struct ceph_connection
*con
)
649 /* reset connection, out_queue, msg_ and connect_seq */
650 /* discard existing out_queue and msg_seq */
651 dout("reset_connection %p\n", con
);
652 ceph_msg_remove_list(&con
->out_queue
);
653 ceph_msg_remove_list(&con
->out_sent
);
656 BUG_ON(con
->in_msg
->con
!= con
);
657 ceph_msg_put(con
->in_msg
);
661 con
->connect_seq
= 0;
664 BUG_ON(con
->out_msg
->con
!= con
);
665 ceph_msg_put(con
->out_msg
);
669 con
->in_seq_acked
= 0;
675 * mark a peer down. drop any open connections.
677 void ceph_con_close(struct ceph_connection
*con
)
679 mutex_lock(&con
->mutex
);
680 dout("con_close %p peer %s\n", con
,
681 ceph_pr_addr(&con
->peer_addr
.in_addr
));
682 con
->state
= CON_STATE_CLOSED
;
684 con_flag_clear(con
, CON_FLAG_LOSSYTX
); /* so we retry next connect */
685 con_flag_clear(con
, CON_FLAG_KEEPALIVE_PENDING
);
686 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
687 con_flag_clear(con
, CON_FLAG_BACKOFF
);
689 reset_connection(con
);
690 con
->peer_global_seq
= 0;
692 con_close_socket(con
);
693 mutex_unlock(&con
->mutex
);
695 EXPORT_SYMBOL(ceph_con_close
);
698 * Reopen a closed connection, with a new peer address.
700 void ceph_con_open(struct ceph_connection
*con
,
701 __u8 entity_type
, __u64 entity_num
,
702 struct ceph_entity_addr
*addr
)
704 mutex_lock(&con
->mutex
);
705 dout("con_open %p %s\n", con
, ceph_pr_addr(&addr
->in_addr
));
707 WARN_ON(con
->state
!= CON_STATE_CLOSED
);
708 con
->state
= CON_STATE_PREOPEN
;
710 con
->peer_name
.type
= (__u8
) entity_type
;
711 con
->peer_name
.num
= cpu_to_le64(entity_num
);
713 memcpy(&con
->peer_addr
, addr
, sizeof(*addr
));
714 con
->delay
= 0; /* reset backoff memory */
715 mutex_unlock(&con
->mutex
);
718 EXPORT_SYMBOL(ceph_con_open
);
721 * return true if this connection ever successfully opened
723 bool ceph_con_opened(struct ceph_connection
*con
)
725 return con
->connect_seq
> 0;
729 * initialize a new connection.
731 void ceph_con_init(struct ceph_connection
*con
, void *private,
732 const struct ceph_connection_operations
*ops
,
733 struct ceph_messenger
*msgr
)
735 dout("con_init %p\n", con
);
736 memset(con
, 0, sizeof(*con
));
737 con
->private = private;
741 con_sock_state_init(con
);
743 mutex_init(&con
->mutex
);
744 INIT_LIST_HEAD(&con
->out_queue
);
745 INIT_LIST_HEAD(&con
->out_sent
);
746 INIT_DELAYED_WORK(&con
->work
, ceph_con_workfn
);
748 con
->state
= CON_STATE_CLOSED
;
750 EXPORT_SYMBOL(ceph_con_init
);
754 * We maintain a global counter to order connection attempts. Get
755 * a unique seq greater than @gt.
757 static u32
get_global_seq(struct ceph_messenger
*msgr
, u32 gt
)
761 spin_lock(&msgr
->global_seq_lock
);
762 if (msgr
->global_seq
< gt
)
763 msgr
->global_seq
= gt
;
764 ret
= ++msgr
->global_seq
;
765 spin_unlock(&msgr
->global_seq_lock
);
769 static void con_out_kvec_reset(struct ceph_connection
*con
)
771 BUG_ON(con
->out_skip
);
773 con
->out_kvec_left
= 0;
774 con
->out_kvec_bytes
= 0;
775 con
->out_kvec_cur
= &con
->out_kvec
[0];
778 static void con_out_kvec_add(struct ceph_connection
*con
,
779 size_t size
, void *data
)
781 int index
= con
->out_kvec_left
;
783 BUG_ON(con
->out_skip
);
784 BUG_ON(index
>= ARRAY_SIZE(con
->out_kvec
));
786 con
->out_kvec
[index
].iov_len
= size
;
787 con
->out_kvec
[index
].iov_base
= data
;
788 con
->out_kvec_left
++;
789 con
->out_kvec_bytes
+= size
;
793 * Chop off a kvec from the end. Return residual number of bytes for
794 * that kvec, i.e. how many bytes would have been written if the kvec
797 static int con_out_kvec_skip(struct ceph_connection
*con
)
799 int off
= con
->out_kvec_cur
- con
->out_kvec
;
802 if (con
->out_kvec_bytes
> 0) {
803 skip
= con
->out_kvec
[off
+ con
->out_kvec_left
- 1].iov_len
;
804 BUG_ON(con
->out_kvec_bytes
< skip
);
805 BUG_ON(!con
->out_kvec_left
);
806 con
->out_kvec_bytes
-= skip
;
807 con
->out_kvec_left
--;
816 * For a bio data item, a piece is whatever remains of the next
817 * entry in the current bio iovec, or the first entry in the next
820 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor
*cursor
,
823 struct ceph_msg_data
*data
= cursor
->data
;
826 BUG_ON(data
->type
!= CEPH_MSG_DATA_BIO
);
831 cursor
->resid
= min(length
, data
->bio_length
);
833 cursor
->bvec_iter
= bio
->bi_iter
;
835 cursor
->resid
<= bio_iter_len(bio
, cursor
->bvec_iter
);
838 static struct page
*ceph_msg_data_bio_next(struct ceph_msg_data_cursor
*cursor
,
842 struct ceph_msg_data
*data
= cursor
->data
;
844 struct bio_vec bio_vec
;
846 BUG_ON(data
->type
!= CEPH_MSG_DATA_BIO
);
851 bio_vec
= bio_iter_iovec(bio
, cursor
->bvec_iter
);
853 *page_offset
= (size_t) bio_vec
.bv_offset
;
854 BUG_ON(*page_offset
>= PAGE_SIZE
);
855 if (cursor
->last_piece
) /* pagelist offset is always 0 */
856 *length
= cursor
->resid
;
858 *length
= (size_t) bio_vec
.bv_len
;
859 BUG_ON(*length
> cursor
->resid
);
860 BUG_ON(*page_offset
+ *length
> PAGE_SIZE
);
862 return bio_vec
.bv_page
;
865 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor
*cursor
,
869 struct bio_vec bio_vec
;
871 BUG_ON(cursor
->data
->type
!= CEPH_MSG_DATA_BIO
);
876 bio_vec
= bio_iter_iovec(bio
, cursor
->bvec_iter
);
878 /* Advance the cursor offset */
880 BUG_ON(cursor
->resid
< bytes
);
881 cursor
->resid
-= bytes
;
883 bio_advance_iter(bio
, &cursor
->bvec_iter
, bytes
);
885 if (bytes
< bio_vec
.bv_len
)
886 return false; /* more bytes to process in this segment */
888 /* Move on to the next segment, and possibly the next bio */
890 if (!cursor
->bvec_iter
.bi_size
) {
894 cursor
->bvec_iter
= bio
->bi_iter
;
896 memset(&cursor
->bvec_iter
, 0,
897 sizeof(cursor
->bvec_iter
));
900 if (!cursor
->last_piece
) {
901 BUG_ON(!cursor
->resid
);
903 /* A short read is OK, so use <= rather than == */
904 if (cursor
->resid
<= bio_iter_len(bio
, cursor
->bvec_iter
))
905 cursor
->last_piece
= true;
910 #endif /* CONFIG_BLOCK */
913 * For a page array, a piece comes from the first page in the array
914 * that has not already been fully consumed.
916 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor
*cursor
,
919 struct ceph_msg_data
*data
= cursor
->data
;
922 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGES
);
924 BUG_ON(!data
->pages
);
925 BUG_ON(!data
->length
);
927 cursor
->resid
= min(length
, data
->length
);
928 page_count
= calc_pages_for(data
->alignment
, (u64
)data
->length
);
929 cursor
->page_offset
= data
->alignment
& ~PAGE_MASK
;
930 cursor
->page_index
= 0;
931 BUG_ON(page_count
> (int)USHRT_MAX
);
932 cursor
->page_count
= (unsigned short)page_count
;
933 BUG_ON(length
> SIZE_MAX
- cursor
->page_offset
);
934 cursor
->last_piece
= cursor
->page_offset
+ cursor
->resid
<= PAGE_SIZE
;
938 ceph_msg_data_pages_next(struct ceph_msg_data_cursor
*cursor
,
939 size_t *page_offset
, size_t *length
)
941 struct ceph_msg_data
*data
= cursor
->data
;
943 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGES
);
945 BUG_ON(cursor
->page_index
>= cursor
->page_count
);
946 BUG_ON(cursor
->page_offset
>= PAGE_SIZE
);
948 *page_offset
= cursor
->page_offset
;
949 if (cursor
->last_piece
)
950 *length
= cursor
->resid
;
952 *length
= PAGE_SIZE
- *page_offset
;
954 return data
->pages
[cursor
->page_index
];
957 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor
*cursor
,
960 BUG_ON(cursor
->data
->type
!= CEPH_MSG_DATA_PAGES
);
962 BUG_ON(cursor
->page_offset
+ bytes
> PAGE_SIZE
);
964 /* Advance the cursor page offset */
966 cursor
->resid
-= bytes
;
967 cursor
->page_offset
= (cursor
->page_offset
+ bytes
) & ~PAGE_MASK
;
968 if (!bytes
|| cursor
->page_offset
)
969 return false; /* more bytes to process in the current page */
972 return false; /* no more data */
974 /* Move on to the next page; offset is already at 0 */
976 BUG_ON(cursor
->page_index
>= cursor
->page_count
);
977 cursor
->page_index
++;
978 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
984 * For a pagelist, a piece is whatever remains to be consumed in the
985 * first page in the list, or the front of the next page.
988 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor
*cursor
,
991 struct ceph_msg_data
*data
= cursor
->data
;
992 struct ceph_pagelist
*pagelist
;
995 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
997 pagelist
= data
->pagelist
;
1001 return; /* pagelist can be assigned but empty */
1003 BUG_ON(list_empty(&pagelist
->head
));
1004 page
= list_first_entry(&pagelist
->head
, struct page
, lru
);
1006 cursor
->resid
= min(length
, pagelist
->length
);
1007 cursor
->page
= page
;
1009 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
1012 static struct page
*
1013 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor
*cursor
,
1014 size_t *page_offset
, size_t *length
)
1016 struct ceph_msg_data
*data
= cursor
->data
;
1017 struct ceph_pagelist
*pagelist
;
1019 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
1021 pagelist
= data
->pagelist
;
1024 BUG_ON(!cursor
->page
);
1025 BUG_ON(cursor
->offset
+ cursor
->resid
!= pagelist
->length
);
1027 /* offset of first page in pagelist is always 0 */
1028 *page_offset
= cursor
->offset
& ~PAGE_MASK
;
1029 if (cursor
->last_piece
)
1030 *length
= cursor
->resid
;
1032 *length
= PAGE_SIZE
- *page_offset
;
1034 return cursor
->page
;
1037 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor
*cursor
,
1040 struct ceph_msg_data
*data
= cursor
->data
;
1041 struct ceph_pagelist
*pagelist
;
1043 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
1045 pagelist
= data
->pagelist
;
1048 BUG_ON(cursor
->offset
+ cursor
->resid
!= pagelist
->length
);
1049 BUG_ON((cursor
->offset
& ~PAGE_MASK
) + bytes
> PAGE_SIZE
);
1051 /* Advance the cursor offset */
1053 cursor
->resid
-= bytes
;
1054 cursor
->offset
+= bytes
;
1055 /* offset of first page in pagelist is always 0 */
1056 if (!bytes
|| cursor
->offset
& ~PAGE_MASK
)
1057 return false; /* more bytes to process in the current page */
1060 return false; /* no more data */
1062 /* Move on to the next page */
1064 BUG_ON(list_is_last(&cursor
->page
->lru
, &pagelist
->head
));
1065 cursor
->page
= list_next_entry(cursor
->page
, lru
);
1066 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
1072 * Message data is handled (sent or received) in pieces, where each
1073 * piece resides on a single page. The network layer might not
1074 * consume an entire piece at once. A data item's cursor keeps
1075 * track of which piece is next to process and how much remains to
1076 * be processed in that piece. It also tracks whether the current
1077 * piece is the last one in the data item.
1079 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor
*cursor
)
1081 size_t length
= cursor
->total_resid
;
1083 switch (cursor
->data
->type
) {
1084 case CEPH_MSG_DATA_PAGELIST
:
1085 ceph_msg_data_pagelist_cursor_init(cursor
, length
);
1087 case CEPH_MSG_DATA_PAGES
:
1088 ceph_msg_data_pages_cursor_init(cursor
, length
);
1091 case CEPH_MSG_DATA_BIO
:
1092 ceph_msg_data_bio_cursor_init(cursor
, length
);
1094 #endif /* CONFIG_BLOCK */
1095 case CEPH_MSG_DATA_NONE
:
1100 cursor
->need_crc
= true;
1103 static void ceph_msg_data_cursor_init(struct ceph_msg
*msg
, size_t length
)
1105 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
1106 struct ceph_msg_data
*data
;
1109 BUG_ON(length
> msg
->data_length
);
1110 BUG_ON(list_empty(&msg
->data
));
1112 cursor
->data_head
= &msg
->data
;
1113 cursor
->total_resid
= length
;
1114 data
= list_first_entry(&msg
->data
, struct ceph_msg_data
, links
);
1115 cursor
->data
= data
;
1117 __ceph_msg_data_cursor_init(cursor
);
1121 * Return the page containing the next piece to process for a given
1122 * data item, and supply the page offset and length of that piece.
1123 * Indicate whether this is the last piece in this data item.
1125 static struct page
*ceph_msg_data_next(struct ceph_msg_data_cursor
*cursor
,
1126 size_t *page_offset
, size_t *length
,
1131 switch (cursor
->data
->type
) {
1132 case CEPH_MSG_DATA_PAGELIST
:
1133 page
= ceph_msg_data_pagelist_next(cursor
, page_offset
, length
);
1135 case CEPH_MSG_DATA_PAGES
:
1136 page
= ceph_msg_data_pages_next(cursor
, page_offset
, length
);
1139 case CEPH_MSG_DATA_BIO
:
1140 page
= ceph_msg_data_bio_next(cursor
, page_offset
, length
);
1142 #endif /* CONFIG_BLOCK */
1143 case CEPH_MSG_DATA_NONE
:
1149 BUG_ON(*page_offset
+ *length
> PAGE_SIZE
);
1152 *last_piece
= cursor
->last_piece
;
1158 * Returns true if the result moves the cursor on to the next piece
1161 static bool ceph_msg_data_advance(struct ceph_msg_data_cursor
*cursor
,
1166 BUG_ON(bytes
> cursor
->resid
);
1167 switch (cursor
->data
->type
) {
1168 case CEPH_MSG_DATA_PAGELIST
:
1169 new_piece
= ceph_msg_data_pagelist_advance(cursor
, bytes
);
1171 case CEPH_MSG_DATA_PAGES
:
1172 new_piece
= ceph_msg_data_pages_advance(cursor
, bytes
);
1175 case CEPH_MSG_DATA_BIO
:
1176 new_piece
= ceph_msg_data_bio_advance(cursor
, bytes
);
1178 #endif /* CONFIG_BLOCK */
1179 case CEPH_MSG_DATA_NONE
:
1184 cursor
->total_resid
-= bytes
;
1186 if (!cursor
->resid
&& cursor
->total_resid
) {
1187 WARN_ON(!cursor
->last_piece
);
1188 BUG_ON(list_is_last(&cursor
->data
->links
, cursor
->data_head
));
1189 cursor
->data
= list_next_entry(cursor
->data
, links
);
1190 __ceph_msg_data_cursor_init(cursor
);
1193 cursor
->need_crc
= new_piece
;
1198 static size_t sizeof_footer(struct ceph_connection
*con
)
1200 return (con
->peer_features
& CEPH_FEATURE_MSG_AUTH
) ?
1201 sizeof(struct ceph_msg_footer
) :
1202 sizeof(struct ceph_msg_footer_old
);
1205 static void prepare_message_data(struct ceph_msg
*msg
, u32 data_len
)
1210 /* Initialize data cursor */
1212 ceph_msg_data_cursor_init(msg
, (size_t)data_len
);
1216 * Prepare footer for currently outgoing message, and finish things
1217 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1219 static void prepare_write_message_footer(struct ceph_connection
*con
)
1221 struct ceph_msg
*m
= con
->out_msg
;
1223 m
->footer
.flags
|= CEPH_MSG_FOOTER_COMPLETE
;
1225 dout("prepare_write_message_footer %p\n", con
);
1226 con_out_kvec_add(con
, sizeof_footer(con
), &m
->footer
);
1227 if (con
->peer_features
& CEPH_FEATURE_MSG_AUTH
) {
1228 if (con
->ops
->sign_message
)
1229 con
->ops
->sign_message(m
);
1233 m
->old_footer
.flags
= m
->footer
.flags
;
1235 con
->out_more
= m
->more_to_follow
;
1236 con
->out_msg_done
= true;
1240 * Prepare headers for the next outgoing message.
1242 static void prepare_write_message(struct ceph_connection
*con
)
1247 con_out_kvec_reset(con
);
1248 con
->out_msg_done
= false;
1250 /* Sneak an ack in there first? If we can get it into the same
1251 * TCP packet that's a good thing. */
1252 if (con
->in_seq
> con
->in_seq_acked
) {
1253 con
->in_seq_acked
= con
->in_seq
;
1254 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
1255 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1256 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1257 &con
->out_temp_ack
);
1260 BUG_ON(list_empty(&con
->out_queue
));
1261 m
= list_first_entry(&con
->out_queue
, struct ceph_msg
, list_head
);
1263 BUG_ON(m
->con
!= con
);
1265 /* put message on sent list */
1267 list_move_tail(&m
->list_head
, &con
->out_sent
);
1270 * only assign outgoing seq # if we haven't sent this message
1271 * yet. if it is requeued, resend with it's original seq.
1273 if (m
->needs_out_seq
) {
1274 m
->hdr
.seq
= cpu_to_le64(++con
->out_seq
);
1275 m
->needs_out_seq
= false;
1277 WARN_ON(m
->data_length
!= le32_to_cpu(m
->hdr
.data_len
));
1279 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1280 m
, con
->out_seq
, le16_to_cpu(m
->hdr
.type
),
1281 le32_to_cpu(m
->hdr
.front_len
), le32_to_cpu(m
->hdr
.middle_len
),
1283 BUG_ON(le32_to_cpu(m
->hdr
.front_len
) != m
->front
.iov_len
);
1285 /* tag + hdr + front + middle */
1286 con_out_kvec_add(con
, sizeof (tag_msg
), &tag_msg
);
1287 con_out_kvec_add(con
, sizeof(con
->out_hdr
), &con
->out_hdr
);
1288 con_out_kvec_add(con
, m
->front
.iov_len
, m
->front
.iov_base
);
1291 con_out_kvec_add(con
, m
->middle
->vec
.iov_len
,
1292 m
->middle
->vec
.iov_base
);
1294 /* fill in hdr crc and finalize hdr */
1295 crc
= crc32c(0, &m
->hdr
, offsetof(struct ceph_msg_header
, crc
));
1296 con
->out_msg
->hdr
.crc
= cpu_to_le32(crc
);
1297 memcpy(&con
->out_hdr
, &con
->out_msg
->hdr
, sizeof(con
->out_hdr
));
1299 /* fill in front and middle crc, footer */
1300 crc
= crc32c(0, m
->front
.iov_base
, m
->front
.iov_len
);
1301 con
->out_msg
->footer
.front_crc
= cpu_to_le32(crc
);
1303 crc
= crc32c(0, m
->middle
->vec
.iov_base
,
1304 m
->middle
->vec
.iov_len
);
1305 con
->out_msg
->footer
.middle_crc
= cpu_to_le32(crc
);
1307 con
->out_msg
->footer
.middle_crc
= 0;
1308 dout("%s front_crc %u middle_crc %u\n", __func__
,
1309 le32_to_cpu(con
->out_msg
->footer
.front_crc
),
1310 le32_to_cpu(con
->out_msg
->footer
.middle_crc
));
1311 con
->out_msg
->footer
.flags
= 0;
1313 /* is there a data payload? */
1314 con
->out_msg
->footer
.data_crc
= 0;
1315 if (m
->data_length
) {
1316 prepare_message_data(con
->out_msg
, m
->data_length
);
1317 con
->out_more
= 1; /* data + footer will follow */
1319 /* no, queue up footer too and be done */
1320 prepare_write_message_footer(con
);
1323 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1329 static void prepare_write_ack(struct ceph_connection
*con
)
1331 dout("prepare_write_ack %p %llu -> %llu\n", con
,
1332 con
->in_seq_acked
, con
->in_seq
);
1333 con
->in_seq_acked
= con
->in_seq
;
1335 con_out_kvec_reset(con
);
1337 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
1339 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1340 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1341 &con
->out_temp_ack
);
1343 con
->out_more
= 1; /* more will follow.. eventually.. */
1344 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1348 * Prepare to share the seq during handshake
1350 static void prepare_write_seq(struct ceph_connection
*con
)
1352 dout("prepare_write_seq %p %llu -> %llu\n", con
,
1353 con
->in_seq_acked
, con
->in_seq
);
1354 con
->in_seq_acked
= con
->in_seq
;
1356 con_out_kvec_reset(con
);
1358 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1359 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1360 &con
->out_temp_ack
);
1362 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1366 * Prepare to write keepalive byte.
1368 static void prepare_write_keepalive(struct ceph_connection
*con
)
1370 dout("prepare_write_keepalive %p\n", con
);
1371 con_out_kvec_reset(con
);
1372 if (con
->peer_features
& CEPH_FEATURE_MSGR_KEEPALIVE2
) {
1373 struct timespec now
= CURRENT_TIME
;
1375 con_out_kvec_add(con
, sizeof(tag_keepalive2
), &tag_keepalive2
);
1376 ceph_encode_timespec(&con
->out_temp_keepalive2
, &now
);
1377 con_out_kvec_add(con
, sizeof(con
->out_temp_keepalive2
),
1378 &con
->out_temp_keepalive2
);
1380 con_out_kvec_add(con
, sizeof(tag_keepalive
), &tag_keepalive
);
1382 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1386 * Connection negotiation.
1389 static struct ceph_auth_handshake
*get_connect_authorizer(struct ceph_connection
*con
,
1392 struct ceph_auth_handshake
*auth
;
1394 if (!con
->ops
->get_authorizer
) {
1395 con
->out_connect
.authorizer_protocol
= CEPH_AUTH_UNKNOWN
;
1396 con
->out_connect
.authorizer_len
= 0;
1400 auth
= con
->ops
->get_authorizer(con
, auth_proto
, con
->auth_retry
);
1404 con
->auth_reply_buf
= auth
->authorizer_reply_buf
;
1405 con
->auth_reply_buf_len
= auth
->authorizer_reply_buf_len
;
1410 * We connected to a peer and are saying hello.
1412 static void prepare_write_banner(struct ceph_connection
*con
)
1414 con_out_kvec_add(con
, strlen(CEPH_BANNER
), CEPH_BANNER
);
1415 con_out_kvec_add(con
, sizeof (con
->msgr
->my_enc_addr
),
1416 &con
->msgr
->my_enc_addr
);
1419 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1422 static int prepare_write_connect(struct ceph_connection
*con
)
1424 unsigned int global_seq
= get_global_seq(con
->msgr
, 0);
1427 struct ceph_auth_handshake
*auth
;
1429 switch (con
->peer_name
.type
) {
1430 case CEPH_ENTITY_TYPE_MON
:
1431 proto
= CEPH_MONC_PROTOCOL
;
1433 case CEPH_ENTITY_TYPE_OSD
:
1434 proto
= CEPH_OSDC_PROTOCOL
;
1436 case CEPH_ENTITY_TYPE_MDS
:
1437 proto
= CEPH_MDSC_PROTOCOL
;
1443 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con
,
1444 con
->connect_seq
, global_seq
, proto
);
1446 con
->out_connect
.features
=
1447 cpu_to_le64(from_msgr(con
->msgr
)->supported_features
);
1448 con
->out_connect
.host_type
= cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT
);
1449 con
->out_connect
.connect_seq
= cpu_to_le32(con
->connect_seq
);
1450 con
->out_connect
.global_seq
= cpu_to_le32(global_seq
);
1451 con
->out_connect
.protocol_version
= cpu_to_le32(proto
);
1452 con
->out_connect
.flags
= 0;
1454 auth_proto
= CEPH_AUTH_UNKNOWN
;
1455 auth
= get_connect_authorizer(con
, &auth_proto
);
1457 return PTR_ERR(auth
);
1459 con
->out_connect
.authorizer_protocol
= cpu_to_le32(auth_proto
);
1460 con
->out_connect
.authorizer_len
= auth
?
1461 cpu_to_le32(auth
->authorizer_buf_len
) : 0;
1463 con_out_kvec_add(con
, sizeof (con
->out_connect
),
1465 if (auth
&& auth
->authorizer_buf_len
)
1466 con_out_kvec_add(con
, auth
->authorizer_buf_len
,
1467 auth
->authorizer_buf
);
1470 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1476 * write as much of pending kvecs to the socket as we can.
1478 * 0 -> socket full, but more to do
1481 static int write_partial_kvec(struct ceph_connection
*con
)
1485 dout("write_partial_kvec %p %d left\n", con
, con
->out_kvec_bytes
);
1486 while (con
->out_kvec_bytes
> 0) {
1487 ret
= ceph_tcp_sendmsg(con
->sock
, con
->out_kvec_cur
,
1488 con
->out_kvec_left
, con
->out_kvec_bytes
,
1492 con
->out_kvec_bytes
-= ret
;
1493 if (con
->out_kvec_bytes
== 0)
1496 /* account for full iov entries consumed */
1497 while (ret
>= con
->out_kvec_cur
->iov_len
) {
1498 BUG_ON(!con
->out_kvec_left
);
1499 ret
-= con
->out_kvec_cur
->iov_len
;
1500 con
->out_kvec_cur
++;
1501 con
->out_kvec_left
--;
1503 /* and for a partially-consumed entry */
1505 con
->out_kvec_cur
->iov_len
-= ret
;
1506 con
->out_kvec_cur
->iov_base
+= ret
;
1509 con
->out_kvec_left
= 0;
1512 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con
,
1513 con
->out_kvec_bytes
, con
->out_kvec_left
, ret
);
1514 return ret
; /* done! */
1517 static u32
ceph_crc32c_page(u32 crc
, struct page
*page
,
1518 unsigned int page_offset
,
1519 unsigned int length
)
1524 BUG_ON(kaddr
== NULL
);
1525 crc
= crc32c(crc
, kaddr
+ page_offset
, length
);
1531 * Write as much message data payload as we can. If we finish, queue
1533 * 1 -> done, footer is now queued in out_kvec[].
1534 * 0 -> socket full, but more to do
1537 static int write_partial_message_data(struct ceph_connection
*con
)
1539 struct ceph_msg
*msg
= con
->out_msg
;
1540 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
1541 bool do_datacrc
= !ceph_test_opt(from_msgr(con
->msgr
), NOCRC
);
1544 dout("%s %p msg %p\n", __func__
, con
, msg
);
1546 if (list_empty(&msg
->data
))
1550 * Iterate through each page that contains data to be
1551 * written, and send as much as possible for each.
1553 * If we are calculating the data crc (the default), we will
1554 * need to map the page. If we have no pages, they have
1555 * been revoked, so use the zero page.
1557 crc
= do_datacrc
? le32_to_cpu(msg
->footer
.data_crc
) : 0;
1558 while (cursor
->resid
) {
1566 page
= ceph_msg_data_next(cursor
, &page_offset
, &length
,
1568 ret
= ceph_tcp_sendpage(con
->sock
, page
, page_offset
,
1569 length
, !last_piece
);
1572 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1576 if (do_datacrc
&& cursor
->need_crc
)
1577 crc
= ceph_crc32c_page(crc
, page
, page_offset
, length
);
1578 need_crc
= ceph_msg_data_advance(cursor
, (size_t)ret
);
1581 dout("%s %p msg %p done\n", __func__
, con
, msg
);
1583 /* prepare and queue up footer, too */
1585 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1587 msg
->footer
.flags
|= CEPH_MSG_FOOTER_NOCRC
;
1588 con_out_kvec_reset(con
);
1589 prepare_write_message_footer(con
);
1591 return 1; /* must return > 0 to indicate success */
1597 static int write_partial_skip(struct ceph_connection
*con
)
1601 dout("%s %p %d left\n", __func__
, con
, con
->out_skip
);
1602 while (con
->out_skip
> 0) {
1603 size_t size
= min(con
->out_skip
, (int) PAGE_SIZE
);
1605 ret
= ceph_tcp_sendpage(con
->sock
, zero_page
, 0, size
, true);
1608 con
->out_skip
-= ret
;
1616 * Prepare to read connection handshake, or an ack.
1618 static void prepare_read_banner(struct ceph_connection
*con
)
1620 dout("prepare_read_banner %p\n", con
);
1621 con
->in_base_pos
= 0;
1624 static void prepare_read_connect(struct ceph_connection
*con
)
1626 dout("prepare_read_connect %p\n", con
);
1627 con
->in_base_pos
= 0;
1630 static void prepare_read_ack(struct ceph_connection
*con
)
1632 dout("prepare_read_ack %p\n", con
);
1633 con
->in_base_pos
= 0;
1636 static void prepare_read_seq(struct ceph_connection
*con
)
1638 dout("prepare_read_seq %p\n", con
);
1639 con
->in_base_pos
= 0;
1640 con
->in_tag
= CEPH_MSGR_TAG_SEQ
;
1643 static void prepare_read_tag(struct ceph_connection
*con
)
1645 dout("prepare_read_tag %p\n", con
);
1646 con
->in_base_pos
= 0;
1647 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1650 static void prepare_read_keepalive_ack(struct ceph_connection
*con
)
1652 dout("prepare_read_keepalive_ack %p\n", con
);
1653 con
->in_base_pos
= 0;
1657 * Prepare to read a message.
1659 static int prepare_read_message(struct ceph_connection
*con
)
1661 dout("prepare_read_message %p\n", con
);
1662 BUG_ON(con
->in_msg
!= NULL
);
1663 con
->in_base_pos
= 0;
1664 con
->in_front_crc
= con
->in_middle_crc
= con
->in_data_crc
= 0;
1669 static int read_partial(struct ceph_connection
*con
,
1670 int end
, int size
, void *object
)
1672 while (con
->in_base_pos
< end
) {
1673 int left
= end
- con
->in_base_pos
;
1674 int have
= size
- left
;
1675 int ret
= ceph_tcp_recvmsg(con
->sock
, object
+ have
, left
);
1678 con
->in_base_pos
+= ret
;
1685 * Read all or part of the connect-side handshake on a new connection
1687 static int read_partial_banner(struct ceph_connection
*con
)
1693 dout("read_partial_banner %p at %d\n", con
, con
->in_base_pos
);
1696 size
= strlen(CEPH_BANNER
);
1698 ret
= read_partial(con
, end
, size
, con
->in_banner
);
1702 size
= sizeof (con
->actual_peer_addr
);
1704 ret
= read_partial(con
, end
, size
, &con
->actual_peer_addr
);
1708 size
= sizeof (con
->peer_addr_for_me
);
1710 ret
= read_partial(con
, end
, size
, &con
->peer_addr_for_me
);
1718 static int read_partial_connect(struct ceph_connection
*con
)
1724 dout("read_partial_connect %p at %d\n", con
, con
->in_base_pos
);
1726 size
= sizeof (con
->in_reply
);
1728 ret
= read_partial(con
, end
, size
, &con
->in_reply
);
1732 size
= le32_to_cpu(con
->in_reply
.authorizer_len
);
1734 ret
= read_partial(con
, end
, size
, con
->auth_reply_buf
);
1738 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1739 con
, (int)con
->in_reply
.tag
,
1740 le32_to_cpu(con
->in_reply
.connect_seq
),
1741 le32_to_cpu(con
->in_reply
.global_seq
));
1748 * Verify the hello banner looks okay.
1750 static int verify_hello(struct ceph_connection
*con
)
1752 if (memcmp(con
->in_banner
, CEPH_BANNER
, strlen(CEPH_BANNER
))) {
1753 pr_err("connect to %s got bad banner\n",
1754 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1755 con
->error_msg
= "protocol error, bad banner";
1761 static bool addr_is_blank(struct sockaddr_storage
*ss
)
1763 struct in_addr
*addr
= &((struct sockaddr_in
*)ss
)->sin_addr
;
1764 struct in6_addr
*addr6
= &((struct sockaddr_in6
*)ss
)->sin6_addr
;
1766 switch (ss
->ss_family
) {
1768 return addr
->s_addr
== htonl(INADDR_ANY
);
1770 return ipv6_addr_any(addr6
);
1776 static int addr_port(struct sockaddr_storage
*ss
)
1778 switch (ss
->ss_family
) {
1780 return ntohs(((struct sockaddr_in
*)ss
)->sin_port
);
1782 return ntohs(((struct sockaddr_in6
*)ss
)->sin6_port
);
1787 static void addr_set_port(struct sockaddr_storage
*ss
, int p
)
1789 switch (ss
->ss_family
) {
1791 ((struct sockaddr_in
*)ss
)->sin_port
= htons(p
);
1794 ((struct sockaddr_in6
*)ss
)->sin6_port
= htons(p
);
1800 * Unlike other *_pton function semantics, zero indicates success.
1802 static int ceph_pton(const char *str
, size_t len
, struct sockaddr_storage
*ss
,
1803 char delim
, const char **ipend
)
1805 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
1806 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
1808 memset(ss
, 0, sizeof(*ss
));
1810 if (in4_pton(str
, len
, (u8
*)&in4
->sin_addr
.s_addr
, delim
, ipend
)) {
1811 ss
->ss_family
= AF_INET
;
1815 if (in6_pton(str
, len
, (u8
*)&in6
->sin6_addr
.s6_addr
, delim
, ipend
)) {
1816 ss
->ss_family
= AF_INET6
;
1824 * Extract hostname string and resolve using kernel DNS facility.
1826 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1827 static int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1828 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1830 const char *end
, *delim_p
;
1831 char *colon_p
, *ip_addr
= NULL
;
1835 * The end of the hostname occurs immediately preceding the delimiter or
1836 * the port marker (':') where the delimiter takes precedence.
1838 delim_p
= memchr(name
, delim
, namelen
);
1839 colon_p
= memchr(name
, ':', namelen
);
1841 if (delim_p
&& colon_p
)
1842 end
= delim_p
< colon_p
? delim_p
: colon_p
;
1843 else if (!delim_p
&& colon_p
)
1847 if (!end
) /* case: hostname:/ */
1848 end
= name
+ namelen
;
1854 /* do dns_resolve upcall */
1855 ip_len
= dns_query(NULL
, name
, end
- name
, NULL
, &ip_addr
, NULL
);
1857 ret
= ceph_pton(ip_addr
, ip_len
, ss
, -1, NULL
);
1865 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end
- name
), name
,
1866 ret
, ret
? "failed" : ceph_pr_addr(ss
));
1871 static inline int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1872 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1879 * Parse a server name (IP or hostname). If a valid IP address is not found
1880 * then try to extract a hostname to resolve using userspace DNS upcall.
1882 static int ceph_parse_server_name(const char *name
, size_t namelen
,
1883 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1887 ret
= ceph_pton(name
, namelen
, ss
, delim
, ipend
);
1889 ret
= ceph_dns_resolve_name(name
, namelen
, ss
, delim
, ipend
);
1895 * Parse an ip[:port] list into an addr array. Use the default
1896 * monitor port if a port isn't specified.
1898 int ceph_parse_ips(const char *c
, const char *end
,
1899 struct ceph_entity_addr
*addr
,
1900 int max_count
, int *count
)
1902 int i
, ret
= -EINVAL
;
1905 dout("parse_ips on '%.*s'\n", (int)(end
-c
), c
);
1906 for (i
= 0; i
< max_count
; i
++) {
1908 struct sockaddr_storage
*ss
= &addr
[i
].in_addr
;
1917 ret
= ceph_parse_server_name(p
, end
- p
, ss
, delim
, &ipend
);
1926 dout("missing matching ']'\n");
1933 if (p
< end
&& *p
== ':') {
1936 while (p
< end
&& *p
>= '0' && *p
<= '9') {
1937 port
= (port
* 10) + (*p
- '0');
1941 port
= CEPH_MON_PORT
;
1942 else if (port
> 65535)
1945 port
= CEPH_MON_PORT
;
1948 addr_set_port(ss
, port
);
1950 dout("parse_ips got %s\n", ceph_pr_addr(ss
));
1967 pr_err("parse_ips bad ip '%.*s'\n", (int)(end
- c
), c
);
1970 EXPORT_SYMBOL(ceph_parse_ips
);
1972 static int process_banner(struct ceph_connection
*con
)
1974 dout("process_banner on %p\n", con
);
1976 if (verify_hello(con
) < 0)
1979 ceph_decode_addr(&con
->actual_peer_addr
);
1980 ceph_decode_addr(&con
->peer_addr_for_me
);
1983 * Make sure the other end is who we wanted. note that the other
1984 * end may not yet know their ip address, so if it's 0.0.0.0, give
1985 * them the benefit of the doubt.
1987 if (memcmp(&con
->peer_addr
, &con
->actual_peer_addr
,
1988 sizeof(con
->peer_addr
)) != 0 &&
1989 !(addr_is_blank(&con
->actual_peer_addr
.in_addr
) &&
1990 con
->actual_peer_addr
.nonce
== con
->peer_addr
.nonce
)) {
1991 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
1992 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1993 (int)le32_to_cpu(con
->peer_addr
.nonce
),
1994 ceph_pr_addr(&con
->actual_peer_addr
.in_addr
),
1995 (int)le32_to_cpu(con
->actual_peer_addr
.nonce
));
1996 con
->error_msg
= "wrong peer at address";
2001 * did we learn our address?
2003 if (addr_is_blank(&con
->msgr
->inst
.addr
.in_addr
)) {
2004 int port
= addr_port(&con
->msgr
->inst
.addr
.in_addr
);
2006 memcpy(&con
->msgr
->inst
.addr
.in_addr
,
2007 &con
->peer_addr_for_me
.in_addr
,
2008 sizeof(con
->peer_addr_for_me
.in_addr
));
2009 addr_set_port(&con
->msgr
->inst
.addr
.in_addr
, port
);
2010 encode_my_addr(con
->msgr
);
2011 dout("process_banner learned my addr is %s\n",
2012 ceph_pr_addr(&con
->msgr
->inst
.addr
.in_addr
));
2018 static int process_connect(struct ceph_connection
*con
)
2020 u64 sup_feat
= from_msgr(con
->msgr
)->supported_features
;
2021 u64 req_feat
= from_msgr(con
->msgr
)->required_features
;
2022 u64 server_feat
= ceph_sanitize_features(
2023 le64_to_cpu(con
->in_reply
.features
));
2026 dout("process_connect on %p tag %d\n", con
, (int)con
->in_tag
);
2028 if (con
->auth_reply_buf
) {
2030 * Any connection that defines ->get_authorizer()
2031 * should also define ->verify_authorizer_reply().
2032 * See get_connect_authorizer().
2034 ret
= con
->ops
->verify_authorizer_reply(con
);
2036 con
->error_msg
= "bad authorize reply";
2041 switch (con
->in_reply
.tag
) {
2042 case CEPH_MSGR_TAG_FEATURES
:
2043 pr_err("%s%lld %s feature set mismatch,"
2044 " my %llx < server's %llx, missing %llx\n",
2045 ENTITY_NAME(con
->peer_name
),
2046 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2047 sup_feat
, server_feat
, server_feat
& ~sup_feat
);
2048 con
->error_msg
= "missing required protocol features";
2049 reset_connection(con
);
2052 case CEPH_MSGR_TAG_BADPROTOVER
:
2053 pr_err("%s%lld %s protocol version mismatch,"
2054 " my %d != server's %d\n",
2055 ENTITY_NAME(con
->peer_name
),
2056 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2057 le32_to_cpu(con
->out_connect
.protocol_version
),
2058 le32_to_cpu(con
->in_reply
.protocol_version
));
2059 con
->error_msg
= "protocol version mismatch";
2060 reset_connection(con
);
2063 case CEPH_MSGR_TAG_BADAUTHORIZER
:
2065 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con
,
2067 if (con
->auth_retry
== 2) {
2068 con
->error_msg
= "connect authorization failure";
2071 con_out_kvec_reset(con
);
2072 ret
= prepare_write_connect(con
);
2075 prepare_read_connect(con
);
2078 case CEPH_MSGR_TAG_RESETSESSION
:
2080 * If we connected with a large connect_seq but the peer
2081 * has no record of a session with us (no connection, or
2082 * connect_seq == 0), they will send RESETSESION to indicate
2083 * that they must have reset their session, and may have
2086 dout("process_connect got RESET peer seq %u\n",
2087 le32_to_cpu(con
->in_reply
.connect_seq
));
2088 pr_err("%s%lld %s connection reset\n",
2089 ENTITY_NAME(con
->peer_name
),
2090 ceph_pr_addr(&con
->peer_addr
.in_addr
));
2091 reset_connection(con
);
2092 con_out_kvec_reset(con
);
2093 ret
= prepare_write_connect(con
);
2096 prepare_read_connect(con
);
2098 /* Tell ceph about it. */
2099 mutex_unlock(&con
->mutex
);
2100 pr_info("reset on %s%lld\n", ENTITY_NAME(con
->peer_name
));
2101 if (con
->ops
->peer_reset
)
2102 con
->ops
->peer_reset(con
);
2103 mutex_lock(&con
->mutex
);
2104 if (con
->state
!= CON_STATE_NEGOTIATING
)
2108 case CEPH_MSGR_TAG_RETRY_SESSION
:
2110 * If we sent a smaller connect_seq than the peer has, try
2111 * again with a larger value.
2113 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2114 le32_to_cpu(con
->out_connect
.connect_seq
),
2115 le32_to_cpu(con
->in_reply
.connect_seq
));
2116 con
->connect_seq
= le32_to_cpu(con
->in_reply
.connect_seq
);
2117 con_out_kvec_reset(con
);
2118 ret
= prepare_write_connect(con
);
2121 prepare_read_connect(con
);
2124 case CEPH_MSGR_TAG_RETRY_GLOBAL
:
2126 * If we sent a smaller global_seq than the peer has, try
2127 * again with a larger value.
2129 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2130 con
->peer_global_seq
,
2131 le32_to_cpu(con
->in_reply
.global_seq
));
2132 get_global_seq(con
->msgr
,
2133 le32_to_cpu(con
->in_reply
.global_seq
));
2134 con_out_kvec_reset(con
);
2135 ret
= prepare_write_connect(con
);
2138 prepare_read_connect(con
);
2141 case CEPH_MSGR_TAG_SEQ
:
2142 case CEPH_MSGR_TAG_READY
:
2143 if (req_feat
& ~server_feat
) {
2144 pr_err("%s%lld %s protocol feature mismatch,"
2145 " my required %llx > server's %llx, need %llx\n",
2146 ENTITY_NAME(con
->peer_name
),
2147 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2148 req_feat
, server_feat
, req_feat
& ~server_feat
);
2149 con
->error_msg
= "missing required protocol features";
2150 reset_connection(con
);
2154 WARN_ON(con
->state
!= CON_STATE_NEGOTIATING
);
2155 con
->state
= CON_STATE_OPEN
;
2156 con
->auth_retry
= 0; /* we authenticated; clear flag */
2157 con
->peer_global_seq
= le32_to_cpu(con
->in_reply
.global_seq
);
2159 con
->peer_features
= server_feat
;
2160 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2161 con
->peer_global_seq
,
2162 le32_to_cpu(con
->in_reply
.connect_seq
),
2164 WARN_ON(con
->connect_seq
!=
2165 le32_to_cpu(con
->in_reply
.connect_seq
));
2167 if (con
->in_reply
.flags
& CEPH_MSG_CONNECT_LOSSY
)
2168 con_flag_set(con
, CON_FLAG_LOSSYTX
);
2170 con
->delay
= 0; /* reset backoff memory */
2172 if (con
->in_reply
.tag
== CEPH_MSGR_TAG_SEQ
) {
2173 prepare_write_seq(con
);
2174 prepare_read_seq(con
);
2176 prepare_read_tag(con
);
2180 case CEPH_MSGR_TAG_WAIT
:
2182 * If there is a connection race (we are opening
2183 * connections to each other), one of us may just have
2184 * to WAIT. This shouldn't happen if we are the
2187 con
->error_msg
= "protocol error, got WAIT as client";
2191 con
->error_msg
= "protocol error, garbage tag during connect";
2199 * read (part of) an ack
2201 static int read_partial_ack(struct ceph_connection
*con
)
2203 int size
= sizeof (con
->in_temp_ack
);
2206 return read_partial(con
, end
, size
, &con
->in_temp_ack
);
2210 * We can finally discard anything that's been acked.
2212 static void process_ack(struct ceph_connection
*con
)
2215 u64 ack
= le64_to_cpu(con
->in_temp_ack
);
2218 while (!list_empty(&con
->out_sent
)) {
2219 m
= list_first_entry(&con
->out_sent
, struct ceph_msg
,
2221 seq
= le64_to_cpu(m
->hdr
.seq
);
2224 dout("got ack for seq %llu type %d at %p\n", seq
,
2225 le16_to_cpu(m
->hdr
.type
), m
);
2226 m
->ack_stamp
= jiffies
;
2229 prepare_read_tag(con
);
2233 static int read_partial_message_section(struct ceph_connection
*con
,
2234 struct kvec
*section
,
2235 unsigned int sec_len
, u32
*crc
)
2241 while (section
->iov_len
< sec_len
) {
2242 BUG_ON(section
->iov_base
== NULL
);
2243 left
= sec_len
- section
->iov_len
;
2244 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)section
->iov_base
+
2245 section
->iov_len
, left
);
2248 section
->iov_len
+= ret
;
2250 if (section
->iov_len
== sec_len
)
2251 *crc
= crc32c(0, section
->iov_base
, section
->iov_len
);
2256 static int read_partial_msg_data(struct ceph_connection
*con
)
2258 struct ceph_msg
*msg
= con
->in_msg
;
2259 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
2260 bool do_datacrc
= !ceph_test_opt(from_msgr(con
->msgr
), NOCRC
);
2268 if (list_empty(&msg
->data
))
2272 crc
= con
->in_data_crc
;
2273 while (cursor
->resid
) {
2274 page
= ceph_msg_data_next(cursor
, &page_offset
, &length
, NULL
);
2275 ret
= ceph_tcp_recvpage(con
->sock
, page
, page_offset
, length
);
2278 con
->in_data_crc
= crc
;
2284 crc
= ceph_crc32c_page(crc
, page
, page_offset
, ret
);
2285 (void) ceph_msg_data_advance(cursor
, (size_t)ret
);
2288 con
->in_data_crc
= crc
;
2290 return 1; /* must return > 0 to indicate success */
2294 * read (part of) a message.
2296 static int ceph_con_in_msg_alloc(struct ceph_connection
*con
, int *skip
);
2298 static int read_partial_message(struct ceph_connection
*con
)
2300 struct ceph_msg
*m
= con
->in_msg
;
2304 unsigned int front_len
, middle_len
, data_len
;
2305 bool do_datacrc
= !ceph_test_opt(from_msgr(con
->msgr
), NOCRC
);
2306 bool need_sign
= (con
->peer_features
& CEPH_FEATURE_MSG_AUTH
);
2310 dout("read_partial_message con %p msg %p\n", con
, m
);
2313 size
= sizeof (con
->in_hdr
);
2315 ret
= read_partial(con
, end
, size
, &con
->in_hdr
);
2319 crc
= crc32c(0, &con
->in_hdr
, offsetof(struct ceph_msg_header
, crc
));
2320 if (cpu_to_le32(crc
) != con
->in_hdr
.crc
) {
2321 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
2322 crc
, con
->in_hdr
.crc
);
2326 front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
2327 if (front_len
> CEPH_MSG_MAX_FRONT_LEN
)
2329 middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
2330 if (middle_len
> CEPH_MSG_MAX_MIDDLE_LEN
)
2332 data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
2333 if (data_len
> CEPH_MSG_MAX_DATA_LEN
)
2337 seq
= le64_to_cpu(con
->in_hdr
.seq
);
2338 if ((s64
)seq
- (s64
)con
->in_seq
< 1) {
2339 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2340 ENTITY_NAME(con
->peer_name
),
2341 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2342 seq
, con
->in_seq
+ 1);
2343 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
2345 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2347 } else if ((s64
)seq
- (s64
)con
->in_seq
> 1) {
2348 pr_err("read_partial_message bad seq %lld expected %lld\n",
2349 seq
, con
->in_seq
+ 1);
2350 con
->error_msg
= "bad message sequence # for incoming message";
2354 /* allocate message? */
2358 dout("got hdr type %d front %d data %d\n", con
->in_hdr
.type
,
2359 front_len
, data_len
);
2360 ret
= ceph_con_in_msg_alloc(con
, &skip
);
2364 BUG_ON(!con
->in_msg
^ skip
);
2366 /* skip this message */
2367 dout("alloc_msg said skip message\n");
2368 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
2370 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2375 BUG_ON(!con
->in_msg
);
2376 BUG_ON(con
->in_msg
->con
!= con
);
2378 m
->front
.iov_len
= 0; /* haven't read it yet */
2380 m
->middle
->vec
.iov_len
= 0;
2382 /* prepare for data payload, if any */
2385 prepare_message_data(con
->in_msg
, data_len
);
2389 ret
= read_partial_message_section(con
, &m
->front
, front_len
,
2390 &con
->in_front_crc
);
2396 ret
= read_partial_message_section(con
, &m
->middle
->vec
,
2398 &con
->in_middle_crc
);
2405 ret
= read_partial_msg_data(con
);
2411 size
= sizeof_footer(con
);
2413 ret
= read_partial(con
, end
, size
, &m
->footer
);
2418 m
->footer
.flags
= m
->old_footer
.flags
;
2422 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2423 m
, front_len
, m
->footer
.front_crc
, middle_len
,
2424 m
->footer
.middle_crc
, data_len
, m
->footer
.data_crc
);
2427 if (con
->in_front_crc
!= le32_to_cpu(m
->footer
.front_crc
)) {
2428 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2429 m
, con
->in_front_crc
, m
->footer
.front_crc
);
2432 if (con
->in_middle_crc
!= le32_to_cpu(m
->footer
.middle_crc
)) {
2433 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2434 m
, con
->in_middle_crc
, m
->footer
.middle_crc
);
2438 (m
->footer
.flags
& CEPH_MSG_FOOTER_NOCRC
) == 0 &&
2439 con
->in_data_crc
!= le32_to_cpu(m
->footer
.data_crc
)) {
2440 pr_err("read_partial_message %p data crc %u != exp. %u\n", m
,
2441 con
->in_data_crc
, le32_to_cpu(m
->footer
.data_crc
));
2445 if (need_sign
&& con
->ops
->check_message_signature
&&
2446 con
->ops
->check_message_signature(m
)) {
2447 pr_err("read_partial_message %p signature check failed\n", m
);
2451 return 1; /* done! */
2455 * Process message. This happens in the worker thread. The callback should
2456 * be careful not to do anything that waits on other incoming messages or it
2459 static void process_message(struct ceph_connection
*con
)
2461 struct ceph_msg
*msg
= con
->in_msg
;
2463 BUG_ON(con
->in_msg
->con
!= con
);
2466 /* if first message, set peer_name */
2467 if (con
->peer_name
.type
== 0)
2468 con
->peer_name
= msg
->hdr
.src
;
2471 mutex_unlock(&con
->mutex
);
2473 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2474 msg
, le64_to_cpu(msg
->hdr
.seq
),
2475 ENTITY_NAME(msg
->hdr
.src
),
2476 le16_to_cpu(msg
->hdr
.type
),
2477 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2478 le32_to_cpu(msg
->hdr
.front_len
),
2479 le32_to_cpu(msg
->hdr
.data_len
),
2480 con
->in_front_crc
, con
->in_middle_crc
, con
->in_data_crc
);
2481 con
->ops
->dispatch(con
, msg
);
2483 mutex_lock(&con
->mutex
);
2486 static int read_keepalive_ack(struct ceph_connection
*con
)
2488 struct ceph_timespec ceph_ts
;
2489 size_t size
= sizeof(ceph_ts
);
2490 int ret
= read_partial(con
, size
, size
, &ceph_ts
);
2493 ceph_decode_timespec(&con
->last_keepalive_ack
, &ceph_ts
);
2494 prepare_read_tag(con
);
2499 * Write something to the socket. Called in a worker thread when the
2500 * socket appears to be writeable and we have something ready to send.
2502 static int try_write(struct ceph_connection
*con
)
2506 dout("try_write start %p state %lu\n", con
, con
->state
);
2509 dout("try_write out_kvec_bytes %d\n", con
->out_kvec_bytes
);
2511 /* open the socket first? */
2512 if (con
->state
== CON_STATE_PREOPEN
) {
2514 con
->state
= CON_STATE_CONNECTING
;
2516 con_out_kvec_reset(con
);
2517 prepare_write_banner(con
);
2518 prepare_read_banner(con
);
2520 BUG_ON(con
->in_msg
);
2521 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2522 dout("try_write initiating connect on %p new state %lu\n",
2524 ret
= ceph_tcp_connect(con
);
2526 con
->error_msg
= "connect error";
2532 /* kvec data queued? */
2533 if (con
->out_kvec_left
) {
2534 ret
= write_partial_kvec(con
);
2538 if (con
->out_skip
) {
2539 ret
= write_partial_skip(con
);
2546 if (con
->out_msg_done
) {
2547 ceph_msg_put(con
->out_msg
);
2548 con
->out_msg
= NULL
; /* we're done with this one */
2552 ret
= write_partial_message_data(con
);
2554 goto more_kvec
; /* we need to send the footer, too! */
2558 dout("try_write write_partial_message_data err %d\n",
2565 if (con
->state
== CON_STATE_OPEN
) {
2566 if (con_flag_test_and_clear(con
, CON_FLAG_KEEPALIVE_PENDING
)) {
2567 prepare_write_keepalive(con
);
2570 /* is anything else pending? */
2571 if (!list_empty(&con
->out_queue
)) {
2572 prepare_write_message(con
);
2575 if (con
->in_seq
> con
->in_seq_acked
) {
2576 prepare_write_ack(con
);
2581 /* Nothing to do! */
2582 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
2583 dout("try_write nothing else to write.\n");
2586 dout("try_write done on %p ret %d\n", con
, ret
);
2593 * Read what we can from the socket.
2595 static int try_read(struct ceph_connection
*con
)
2600 dout("try_read start on %p state %lu\n", con
, con
->state
);
2601 if (con
->state
!= CON_STATE_CONNECTING
&&
2602 con
->state
!= CON_STATE_NEGOTIATING
&&
2603 con
->state
!= CON_STATE_OPEN
)
2608 dout("try_read tag %d in_base_pos %d\n", (int)con
->in_tag
,
2611 if (con
->state
== CON_STATE_CONNECTING
) {
2612 dout("try_read connecting\n");
2613 ret
= read_partial_banner(con
);
2616 ret
= process_banner(con
);
2620 con
->state
= CON_STATE_NEGOTIATING
;
2623 * Received banner is good, exchange connection info.
2624 * Do not reset out_kvec, as sending our banner raced
2625 * with receiving peer banner after connect completed.
2627 ret
= prepare_write_connect(con
);
2630 prepare_read_connect(con
);
2632 /* Send connection info before awaiting response */
2636 if (con
->state
== CON_STATE_NEGOTIATING
) {
2637 dout("try_read negotiating\n");
2638 ret
= read_partial_connect(con
);
2641 ret
= process_connect(con
);
2647 WARN_ON(con
->state
!= CON_STATE_OPEN
);
2649 if (con
->in_base_pos
< 0) {
2651 * skipping + discarding content.
2653 * FIXME: there must be a better way to do this!
2655 static char buf
[SKIP_BUF_SIZE
];
2656 int skip
= min((int) sizeof (buf
), -con
->in_base_pos
);
2658 dout("skipping %d / %d bytes\n", skip
, -con
->in_base_pos
);
2659 ret
= ceph_tcp_recvmsg(con
->sock
, buf
, skip
);
2662 con
->in_base_pos
+= ret
;
2663 if (con
->in_base_pos
)
2666 if (con
->in_tag
== CEPH_MSGR_TAG_READY
) {
2670 ret
= ceph_tcp_recvmsg(con
->sock
, &con
->in_tag
, 1);
2673 dout("try_read got tag %d\n", (int)con
->in_tag
);
2674 switch (con
->in_tag
) {
2675 case CEPH_MSGR_TAG_MSG
:
2676 prepare_read_message(con
);
2678 case CEPH_MSGR_TAG_ACK
:
2679 prepare_read_ack(con
);
2681 case CEPH_MSGR_TAG_KEEPALIVE2_ACK
:
2682 prepare_read_keepalive_ack(con
);
2684 case CEPH_MSGR_TAG_CLOSE
:
2685 con_close_socket(con
);
2686 con
->state
= CON_STATE_CLOSED
;
2692 if (con
->in_tag
== CEPH_MSGR_TAG_MSG
) {
2693 ret
= read_partial_message(con
);
2697 con
->error_msg
= "bad crc/signature";
2703 con
->error_msg
= "io error";
2708 if (con
->in_tag
== CEPH_MSGR_TAG_READY
)
2710 process_message(con
);
2711 if (con
->state
== CON_STATE_OPEN
)
2712 prepare_read_tag(con
);
2715 if (con
->in_tag
== CEPH_MSGR_TAG_ACK
||
2716 con
->in_tag
== CEPH_MSGR_TAG_SEQ
) {
2718 * the final handshake seq exchange is semantically
2719 * equivalent to an ACK
2721 ret
= read_partial_ack(con
);
2727 if (con
->in_tag
== CEPH_MSGR_TAG_KEEPALIVE2_ACK
) {
2728 ret
= read_keepalive_ack(con
);
2735 dout("try_read done on %p ret %d\n", con
, ret
);
2739 pr_err("try_read bad con->in_tag = %d\n", (int)con
->in_tag
);
2740 con
->error_msg
= "protocol error, garbage tag";
2747 * Atomically queue work on a connection after the specified delay.
2748 * Bump @con reference to avoid races with connection teardown.
2749 * Returns 0 if work was queued, or an error code otherwise.
2751 static int queue_con_delay(struct ceph_connection
*con
, unsigned long delay
)
2753 if (!con
->ops
->get(con
)) {
2754 dout("%s %p ref count 0\n", __func__
, con
);
2758 if (!queue_delayed_work(ceph_msgr_wq
, &con
->work
, delay
)) {
2759 dout("%s %p - already queued\n", __func__
, con
);
2764 dout("%s %p %lu\n", __func__
, con
, delay
);
2768 static void queue_con(struct ceph_connection
*con
)
2770 (void) queue_con_delay(con
, 0);
2773 static void cancel_con(struct ceph_connection
*con
)
2775 if (cancel_delayed_work(&con
->work
)) {
2776 dout("%s %p\n", __func__
, con
);
2781 static bool con_sock_closed(struct ceph_connection
*con
)
2783 if (!con_flag_test_and_clear(con
, CON_FLAG_SOCK_CLOSED
))
2787 case CON_STATE_ ## x: \
2788 con->error_msg = "socket closed (con state " #x ")"; \
2791 switch (con
->state
) {
2799 pr_warn("%s con %p unrecognized state %lu\n",
2800 __func__
, con
, con
->state
);
2801 con
->error_msg
= "unrecognized con state";
2810 static bool con_backoff(struct ceph_connection
*con
)
2814 if (!con_flag_test_and_clear(con
, CON_FLAG_BACKOFF
))
2817 ret
= queue_con_delay(con
, round_jiffies_relative(con
->delay
));
2819 dout("%s: con %p FAILED to back off %lu\n", __func__
,
2821 BUG_ON(ret
== -ENOENT
);
2822 con_flag_set(con
, CON_FLAG_BACKOFF
);
2828 /* Finish fault handling; con->mutex must *not* be held here */
2830 static void con_fault_finish(struct ceph_connection
*con
)
2832 dout("%s %p\n", __func__
, con
);
2835 * in case we faulted due to authentication, invalidate our
2836 * current tickets so that we can get new ones.
2838 if (con
->auth_retry
) {
2839 dout("auth_retry %d, invalidating\n", con
->auth_retry
);
2840 if (con
->ops
->invalidate_authorizer
)
2841 con
->ops
->invalidate_authorizer(con
);
2842 con
->auth_retry
= 0;
2845 if (con
->ops
->fault
)
2846 con
->ops
->fault(con
);
2850 * Do some work on a connection. Drop a connection ref when we're done.
2852 static void ceph_con_workfn(struct work_struct
*work
)
2854 struct ceph_connection
*con
= container_of(work
, struct ceph_connection
,
2858 mutex_lock(&con
->mutex
);
2862 if ((fault
= con_sock_closed(con
))) {
2863 dout("%s: con %p SOCK_CLOSED\n", __func__
, con
);
2866 if (con_backoff(con
)) {
2867 dout("%s: con %p BACKOFF\n", __func__
, con
);
2870 if (con
->state
== CON_STATE_STANDBY
) {
2871 dout("%s: con %p STANDBY\n", __func__
, con
);
2874 if (con
->state
== CON_STATE_CLOSED
) {
2875 dout("%s: con %p CLOSED\n", __func__
, con
);
2879 if (con
->state
== CON_STATE_PREOPEN
) {
2880 dout("%s: con %p PREOPEN\n", __func__
, con
);
2884 ret
= try_read(con
);
2888 if (!con
->error_msg
)
2889 con
->error_msg
= "socket error on read";
2894 ret
= try_write(con
);
2898 if (!con
->error_msg
)
2899 con
->error_msg
= "socket error on write";
2903 break; /* If we make it to here, we're done */
2907 mutex_unlock(&con
->mutex
);
2910 con_fault_finish(con
);
2916 * Generic error/fault handler. A retry mechanism is used with
2917 * exponential backoff
2919 static void con_fault(struct ceph_connection
*con
)
2921 dout("fault %p state %lu to peer %s\n",
2922 con
, con
->state
, ceph_pr_addr(&con
->peer_addr
.in_addr
));
2924 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con
->peer_name
),
2925 ceph_pr_addr(&con
->peer_addr
.in_addr
), con
->error_msg
);
2926 con
->error_msg
= NULL
;
2928 WARN_ON(con
->state
!= CON_STATE_CONNECTING
&&
2929 con
->state
!= CON_STATE_NEGOTIATING
&&
2930 con
->state
!= CON_STATE_OPEN
);
2932 con_close_socket(con
);
2934 if (con_flag_test(con
, CON_FLAG_LOSSYTX
)) {
2935 dout("fault on LOSSYTX channel, marking CLOSED\n");
2936 con
->state
= CON_STATE_CLOSED
;
2941 BUG_ON(con
->in_msg
->con
!= con
);
2942 ceph_msg_put(con
->in_msg
);
2946 /* Requeue anything that hasn't been acked */
2947 list_splice_init(&con
->out_sent
, &con
->out_queue
);
2949 /* If there are no messages queued or keepalive pending, place
2950 * the connection in a STANDBY state */
2951 if (list_empty(&con
->out_queue
) &&
2952 !con_flag_test(con
, CON_FLAG_KEEPALIVE_PENDING
)) {
2953 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con
);
2954 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
2955 con
->state
= CON_STATE_STANDBY
;
2957 /* retry after a delay. */
2958 con
->state
= CON_STATE_PREOPEN
;
2959 if (con
->delay
== 0)
2960 con
->delay
= BASE_DELAY_INTERVAL
;
2961 else if (con
->delay
< MAX_DELAY_INTERVAL
)
2963 con_flag_set(con
, CON_FLAG_BACKOFF
);
2971 * initialize a new messenger instance
2973 void ceph_messenger_init(struct ceph_messenger
*msgr
,
2974 struct ceph_entity_addr
*myaddr
)
2976 spin_lock_init(&msgr
->global_seq_lock
);
2979 msgr
->inst
.addr
= *myaddr
;
2981 /* select a random nonce */
2982 msgr
->inst
.addr
.type
= 0;
2983 get_random_bytes(&msgr
->inst
.addr
.nonce
, sizeof(msgr
->inst
.addr
.nonce
));
2984 encode_my_addr(msgr
);
2986 atomic_set(&msgr
->stopping
, 0);
2987 write_pnet(&msgr
->net
, get_net(current
->nsproxy
->net_ns
));
2989 dout("%s %p\n", __func__
, msgr
);
2991 EXPORT_SYMBOL(ceph_messenger_init
);
2993 void ceph_messenger_fini(struct ceph_messenger
*msgr
)
2995 put_net(read_pnet(&msgr
->net
));
2997 EXPORT_SYMBOL(ceph_messenger_fini
);
2999 static void msg_con_set(struct ceph_msg
*msg
, struct ceph_connection
*con
)
3002 msg
->con
->ops
->put(msg
->con
);
3004 msg
->con
= con
? con
->ops
->get(con
) : NULL
;
3005 BUG_ON(msg
->con
!= con
);
3008 static void clear_standby(struct ceph_connection
*con
)
3010 /* come back from STANDBY? */
3011 if (con
->state
== CON_STATE_STANDBY
) {
3012 dout("clear_standby %p and ++connect_seq\n", con
);
3013 con
->state
= CON_STATE_PREOPEN
;
3015 WARN_ON(con_flag_test(con
, CON_FLAG_WRITE_PENDING
));
3016 WARN_ON(con_flag_test(con
, CON_FLAG_KEEPALIVE_PENDING
));
3021 * Queue up an outgoing message on the given connection.
3023 void ceph_con_send(struct ceph_connection
*con
, struct ceph_msg
*msg
)
3026 msg
->hdr
.src
= con
->msgr
->inst
.name
;
3027 BUG_ON(msg
->front
.iov_len
!= le32_to_cpu(msg
->hdr
.front_len
));
3028 msg
->needs_out_seq
= true;
3030 mutex_lock(&con
->mutex
);
3032 if (con
->state
== CON_STATE_CLOSED
) {
3033 dout("con_send %p closed, dropping %p\n", con
, msg
);
3035 mutex_unlock(&con
->mutex
);
3039 msg_con_set(msg
, con
);
3041 BUG_ON(!list_empty(&msg
->list_head
));
3042 list_add_tail(&msg
->list_head
, &con
->out_queue
);
3043 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg
,
3044 ENTITY_NAME(con
->peer_name
), le16_to_cpu(msg
->hdr
.type
),
3045 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
3046 le32_to_cpu(msg
->hdr
.front_len
),
3047 le32_to_cpu(msg
->hdr
.middle_len
),
3048 le32_to_cpu(msg
->hdr
.data_len
));
3051 mutex_unlock(&con
->mutex
);
3053 /* if there wasn't anything waiting to send before, queue
3055 if (con_flag_test_and_set(con
, CON_FLAG_WRITE_PENDING
) == 0)
3058 EXPORT_SYMBOL(ceph_con_send
);
3061 * Revoke a message that was previously queued for send
3063 void ceph_msg_revoke(struct ceph_msg
*msg
)
3065 struct ceph_connection
*con
= msg
->con
;
3068 dout("%s msg %p null con\n", __func__
, msg
);
3069 return; /* Message not in our possession */
3072 mutex_lock(&con
->mutex
);
3073 if (!list_empty(&msg
->list_head
)) {
3074 dout("%s %p msg %p - was on queue\n", __func__
, con
, msg
);
3075 list_del_init(&msg
->list_head
);
3080 if (con
->out_msg
== msg
) {
3081 BUG_ON(con
->out_skip
);
3083 if (con
->out_msg_done
) {
3084 con
->out_skip
+= con_out_kvec_skip(con
);
3086 BUG_ON(!msg
->data_length
);
3087 con
->out_skip
+= sizeof_footer(con
);
3089 /* data, middle, front */
3090 if (msg
->data_length
)
3091 con
->out_skip
+= msg
->cursor
.total_resid
;
3093 con
->out_skip
+= con_out_kvec_skip(con
);
3094 con
->out_skip
+= con_out_kvec_skip(con
);
3096 dout("%s %p msg %p - was sending, will write %d skip %d\n",
3097 __func__
, con
, msg
, con
->out_kvec_bytes
, con
->out_skip
);
3099 con
->out_msg
= NULL
;
3103 mutex_unlock(&con
->mutex
);
3107 * Revoke a message that we may be reading data into
3109 void ceph_msg_revoke_incoming(struct ceph_msg
*msg
)
3111 struct ceph_connection
*con
= msg
->con
;
3114 dout("%s msg %p null con\n", __func__
, msg
);
3115 return; /* Message not in our possession */
3118 mutex_lock(&con
->mutex
);
3119 if (con
->in_msg
== msg
) {
3120 unsigned int front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
3121 unsigned int middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
3122 unsigned int data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
3124 /* skip rest of message */
3125 dout("%s %p msg %p revoked\n", __func__
, con
, msg
);
3126 con
->in_base_pos
= con
->in_base_pos
-
3127 sizeof(struct ceph_msg_header
) -
3131 sizeof(struct ceph_msg_footer
);
3132 ceph_msg_put(con
->in_msg
);
3134 con
->in_tag
= CEPH_MSGR_TAG_READY
;
3137 dout("%s %p in_msg %p msg %p no-op\n",
3138 __func__
, con
, con
->in_msg
, msg
);
3140 mutex_unlock(&con
->mutex
);
3144 * Queue a keepalive byte to ensure the tcp connection is alive.
3146 void ceph_con_keepalive(struct ceph_connection
*con
)
3148 dout("con_keepalive %p\n", con
);
3149 mutex_lock(&con
->mutex
);
3151 mutex_unlock(&con
->mutex
);
3152 if (con_flag_test_and_set(con
, CON_FLAG_KEEPALIVE_PENDING
) == 0 &&
3153 con_flag_test_and_set(con
, CON_FLAG_WRITE_PENDING
) == 0)
3156 EXPORT_SYMBOL(ceph_con_keepalive
);
3158 bool ceph_con_keepalive_expired(struct ceph_connection
*con
,
3159 unsigned long interval
)
3162 (con
->peer_features
& CEPH_FEATURE_MSGR_KEEPALIVE2
)) {
3163 struct timespec now
= CURRENT_TIME
;
3165 jiffies_to_timespec(interval
, &ts
);
3166 ts
= timespec_add(con
->last_keepalive_ack
, ts
);
3167 return timespec_compare(&now
, &ts
) >= 0;
3172 static struct ceph_msg_data
*ceph_msg_data_create(enum ceph_msg_data_type type
)
3174 struct ceph_msg_data
*data
;
3176 if (WARN_ON(!ceph_msg_data_type_valid(type
)))
3179 data
= kmem_cache_zalloc(ceph_msg_data_cache
, GFP_NOFS
);
3182 INIT_LIST_HEAD(&data
->links
);
3187 static void ceph_msg_data_destroy(struct ceph_msg_data
*data
)
3192 WARN_ON(!list_empty(&data
->links
));
3193 if (data
->type
== CEPH_MSG_DATA_PAGELIST
)
3194 ceph_pagelist_release(data
->pagelist
);
3195 kmem_cache_free(ceph_msg_data_cache
, data
);
3198 void ceph_msg_data_add_pages(struct ceph_msg
*msg
, struct page
**pages
,
3199 size_t length
, size_t alignment
)
3201 struct ceph_msg_data
*data
;
3206 data
= ceph_msg_data_create(CEPH_MSG_DATA_PAGES
);
3208 data
->pages
= pages
;
3209 data
->length
= length
;
3210 data
->alignment
= alignment
& ~PAGE_MASK
;
3212 list_add_tail(&data
->links
, &msg
->data
);
3213 msg
->data_length
+= length
;
3215 EXPORT_SYMBOL(ceph_msg_data_add_pages
);
3217 void ceph_msg_data_add_pagelist(struct ceph_msg
*msg
,
3218 struct ceph_pagelist
*pagelist
)
3220 struct ceph_msg_data
*data
;
3223 BUG_ON(!pagelist
->length
);
3225 data
= ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST
);
3227 data
->pagelist
= pagelist
;
3229 list_add_tail(&data
->links
, &msg
->data
);
3230 msg
->data_length
+= pagelist
->length
;
3232 EXPORT_SYMBOL(ceph_msg_data_add_pagelist
);
3235 void ceph_msg_data_add_bio(struct ceph_msg
*msg
, struct bio
*bio
,
3238 struct ceph_msg_data
*data
;
3242 data
= ceph_msg_data_create(CEPH_MSG_DATA_BIO
);
3245 data
->bio_length
= length
;
3247 list_add_tail(&data
->links
, &msg
->data
);
3248 msg
->data_length
+= length
;
3250 EXPORT_SYMBOL(ceph_msg_data_add_bio
);
3251 #endif /* CONFIG_BLOCK */
3254 * construct a new message with given type, size
3255 * the new msg has a ref count of 1.
3257 struct ceph_msg
*ceph_msg_new(int type
, int front_len
, gfp_t flags
,
3262 m
= kmem_cache_zalloc(ceph_msg_cache
, flags
);
3266 m
->hdr
.type
= cpu_to_le16(type
);
3267 m
->hdr
.priority
= cpu_to_le16(CEPH_MSG_PRIO_DEFAULT
);
3268 m
->hdr
.front_len
= cpu_to_le32(front_len
);
3270 INIT_LIST_HEAD(&m
->list_head
);
3271 kref_init(&m
->kref
);
3272 INIT_LIST_HEAD(&m
->data
);
3276 m
->front
.iov_base
= ceph_kvmalloc(front_len
, flags
);
3277 if (m
->front
.iov_base
== NULL
) {
3278 dout("ceph_msg_new can't allocate %d bytes\n",
3283 m
->front
.iov_base
= NULL
;
3285 m
->front_alloc_len
= m
->front
.iov_len
= front_len
;
3287 dout("ceph_msg_new %p front %d\n", m
, front_len
);
3294 pr_err("msg_new can't create type %d front %d\n", type
,
3298 dout("msg_new can't create type %d front %d\n", type
,
3303 EXPORT_SYMBOL(ceph_msg_new
);
3306 * Allocate "middle" portion of a message, if it is needed and wasn't
3307 * allocated by alloc_msg. This allows us to read a small fixed-size
3308 * per-type header in the front and then gracefully fail (i.e.,
3309 * propagate the error to the caller based on info in the front) when
3310 * the middle is too large.
3312 static int ceph_alloc_middle(struct ceph_connection
*con
, struct ceph_msg
*msg
)
3314 int type
= le16_to_cpu(msg
->hdr
.type
);
3315 int middle_len
= le32_to_cpu(msg
->hdr
.middle_len
);
3317 dout("alloc_middle %p type %d %s middle_len %d\n", msg
, type
,
3318 ceph_msg_type_name(type
), middle_len
);
3319 BUG_ON(!middle_len
);
3320 BUG_ON(msg
->middle
);
3322 msg
->middle
= ceph_buffer_new(middle_len
, GFP_NOFS
);
3329 * Allocate a message for receiving an incoming message on a
3330 * connection, and save the result in con->in_msg. Uses the
3331 * connection's private alloc_msg op if available.
3333 * Returns 0 on success, or a negative error code.
3335 * On success, if we set *skip = 1:
3336 * - the next message should be skipped and ignored.
3337 * - con->in_msg == NULL
3338 * or if we set *skip = 0:
3339 * - con->in_msg is non-null.
3340 * On error (ENOMEM, EAGAIN, ...),
3341 * - con->in_msg == NULL
3343 static int ceph_con_in_msg_alloc(struct ceph_connection
*con
, int *skip
)
3345 struct ceph_msg_header
*hdr
= &con
->in_hdr
;
3346 int middle_len
= le32_to_cpu(hdr
->middle_len
);
3347 struct ceph_msg
*msg
;
3350 BUG_ON(con
->in_msg
!= NULL
);
3351 BUG_ON(!con
->ops
->alloc_msg
);
3353 mutex_unlock(&con
->mutex
);
3354 msg
= con
->ops
->alloc_msg(con
, hdr
, skip
);
3355 mutex_lock(&con
->mutex
);
3356 if (con
->state
!= CON_STATE_OPEN
) {
3363 msg_con_set(msg
, con
);
3367 * Null message pointer means either we should skip
3368 * this message or we couldn't allocate memory. The
3369 * former is not an error.
3374 con
->error_msg
= "error allocating memory for incoming message";
3377 memcpy(&con
->in_msg
->hdr
, &con
->in_hdr
, sizeof(con
->in_hdr
));
3379 if (middle_len
&& !con
->in_msg
->middle
) {
3380 ret
= ceph_alloc_middle(con
, con
->in_msg
);
3382 ceph_msg_put(con
->in_msg
);
3392 * Free a generically kmalloc'd message.
3394 static void ceph_msg_free(struct ceph_msg
*m
)
3396 dout("%s %p\n", __func__
, m
);
3397 kvfree(m
->front
.iov_base
);
3398 kmem_cache_free(ceph_msg_cache
, m
);
3401 static void ceph_msg_release(struct kref
*kref
)
3403 struct ceph_msg
*m
= container_of(kref
, struct ceph_msg
, kref
);
3404 struct ceph_msg_data
*data
, *next
;
3406 dout("%s %p\n", __func__
, m
);
3407 WARN_ON(!list_empty(&m
->list_head
));
3409 msg_con_set(m
, NULL
);
3411 /* drop middle, data, if any */
3413 ceph_buffer_put(m
->middle
);
3417 list_for_each_entry_safe(data
, next
, &m
->data
, links
) {
3418 list_del_init(&data
->links
);
3419 ceph_msg_data_destroy(data
);
3424 ceph_msgpool_put(m
->pool
, m
);
3429 struct ceph_msg
*ceph_msg_get(struct ceph_msg
*msg
)
3431 dout("%s %p (was %d)\n", __func__
, msg
,
3432 atomic_read(&msg
->kref
.refcount
));
3433 kref_get(&msg
->kref
);
3436 EXPORT_SYMBOL(ceph_msg_get
);
3438 void ceph_msg_put(struct ceph_msg
*msg
)
3440 dout("%s %p (was %d)\n", __func__
, msg
,
3441 atomic_read(&msg
->kref
.refcount
));
3442 kref_put(&msg
->kref
, ceph_msg_release
);
3444 EXPORT_SYMBOL(ceph_msg_put
);
3446 void ceph_msg_dump(struct ceph_msg
*msg
)
3448 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg
,
3449 msg
->front_alloc_len
, msg
->data_length
);
3450 print_hex_dump(KERN_DEBUG
, "header: ",
3451 DUMP_PREFIX_OFFSET
, 16, 1,
3452 &msg
->hdr
, sizeof(msg
->hdr
), true);
3453 print_hex_dump(KERN_DEBUG
, " front: ",
3454 DUMP_PREFIX_OFFSET
, 16, 1,
3455 msg
->front
.iov_base
, msg
->front
.iov_len
, true);
3457 print_hex_dump(KERN_DEBUG
, "middle: ",
3458 DUMP_PREFIX_OFFSET
, 16, 1,
3459 msg
->middle
->vec
.iov_base
,
3460 msg
->middle
->vec
.iov_len
, true);
3461 print_hex_dump(KERN_DEBUG
, "footer: ",
3462 DUMP_PREFIX_OFFSET
, 16, 1,
3463 &msg
->footer
, sizeof(msg
->footer
), true);
3465 EXPORT_SYMBOL(ceph_msg_dump
);