1 #include <linux/ceph/ceph_debug.h>
3 #include <linux/crc32c.h>
4 #include <linux/ctype.h>
5 #include <linux/highmem.h>
6 #include <linux/inet.h>
7 #include <linux/kthread.h>
9 #include <linux/slab.h>
10 #include <linux/socket.h>
11 #include <linux/string.h>
13 #include <linux/bio.h>
14 #endif /* CONFIG_BLOCK */
15 #include <linux/dns_resolver.h>
18 #include <linux/ceph/ceph_features.h>
19 #include <linux/ceph/libceph.h>
20 #include <linux/ceph/messenger.h>
21 #include <linux/ceph/decode.h>
22 #include <linux/ceph/pagelist.h>
23 #include <linux/export.h>
25 #define list_entry_next(pos, member) \
26 list_entry(pos->member.next, typeof(*pos), member)
29 * Ceph uses the messenger to exchange ceph_msg messages with other
30 * hosts in the system. The messenger provides ordered and reliable
31 * delivery. We tolerate TCP disconnects by reconnecting (with
32 * exponential backoff) in the case of a fault (disconnection, bad
33 * crc, protocol error). Acks allow sent messages to be discarded by
38 * We track the state of the socket on a given connection using
39 * values defined below. The transition to a new socket state is
40 * handled by a function which verifies we aren't coming from an
44 * | NEW* | transient initial state
46 * | con_sock_state_init()
49 * | CLOSED | initialized, but no socket (and no
50 * ---------- TCP connection)
52 * | \ con_sock_state_connecting()
53 * | ----------------------
55 * + con_sock_state_closed() \
56 * |+--------------------------- \
59 * | | CLOSING | socket event; \ \
60 * | ----------- await close \ \
63 * | + con_sock_state_closing() \ |
65 * | / --------------- | |
68 * | / -----------------| CONNECTING | socket created, TCP
69 * | | / -------------- connect initiated
70 * | | | con_sock_state_connected()
73 * | CONNECTED | TCP connection established
76 * State values for ceph_connection->sock_state; NEW is assumed to be 0.
79 #define CON_SOCK_STATE_NEW 0 /* -> CLOSED */
80 #define CON_SOCK_STATE_CLOSED 1 /* -> CONNECTING */
81 #define CON_SOCK_STATE_CONNECTING 2 /* -> CONNECTED or -> CLOSING */
82 #define CON_SOCK_STATE_CONNECTED 3 /* -> CLOSING or -> CLOSED */
83 #define CON_SOCK_STATE_CLOSING 4 /* -> CLOSED */
88 #define CON_STATE_CLOSED 1 /* -> PREOPEN */
89 #define CON_STATE_PREOPEN 2 /* -> CONNECTING, CLOSED */
90 #define CON_STATE_CONNECTING 3 /* -> NEGOTIATING, CLOSED */
91 #define CON_STATE_NEGOTIATING 4 /* -> OPEN, CLOSED */
92 #define CON_STATE_OPEN 5 /* -> STANDBY, CLOSED */
93 #define CON_STATE_STANDBY 6 /* -> PREOPEN, CLOSED */
96 * ceph_connection flag bits
98 #define CON_FLAG_LOSSYTX 0 /* we can close channel or drop
99 * messages on errors */
100 #define CON_FLAG_KEEPALIVE_PENDING 1 /* we need to send a keepalive */
101 #define CON_FLAG_WRITE_PENDING 2 /* we have data ready to send */
102 #define CON_FLAG_SOCK_CLOSED 3 /* socket state changed to closed */
103 #define CON_FLAG_BACKOFF 4 /* need to retry queuing delayed work */
105 static bool con_flag_valid(unsigned long con_flag
)
108 case CON_FLAG_LOSSYTX
:
109 case CON_FLAG_KEEPALIVE_PENDING
:
110 case CON_FLAG_WRITE_PENDING
:
111 case CON_FLAG_SOCK_CLOSED
:
112 case CON_FLAG_BACKOFF
:
119 static void con_flag_clear(struct ceph_connection
*con
, unsigned long con_flag
)
121 BUG_ON(!con_flag_valid(con_flag
));
123 clear_bit(con_flag
, &con
->flags
);
126 static void con_flag_set(struct ceph_connection
*con
, unsigned long con_flag
)
128 BUG_ON(!con_flag_valid(con_flag
));
130 set_bit(con_flag
, &con
->flags
);
133 static bool con_flag_test(struct ceph_connection
*con
, unsigned long con_flag
)
135 BUG_ON(!con_flag_valid(con_flag
));
137 return test_bit(con_flag
, &con
->flags
);
140 static bool con_flag_test_and_clear(struct ceph_connection
*con
,
141 unsigned long con_flag
)
143 BUG_ON(!con_flag_valid(con_flag
));
145 return test_and_clear_bit(con_flag
, &con
->flags
);
148 static bool con_flag_test_and_set(struct ceph_connection
*con
,
149 unsigned long con_flag
)
151 BUG_ON(!con_flag_valid(con_flag
));
153 return test_and_set_bit(con_flag
, &con
->flags
);
156 /* Slab caches for frequently-allocated structures */
158 static struct kmem_cache
*ceph_msg_cache
;
159 static struct kmem_cache
*ceph_msg_data_cache
;
161 /* static tag bytes (protocol control messages) */
162 static char tag_msg
= CEPH_MSGR_TAG_MSG
;
163 static char tag_ack
= CEPH_MSGR_TAG_ACK
;
164 static char tag_keepalive
= CEPH_MSGR_TAG_KEEPALIVE
;
166 #ifdef CONFIG_LOCKDEP
167 static struct lock_class_key socket_class
;
171 * When skipping (ignoring) a block of input we read it into a "skip
172 * buffer," which is this many bytes in size.
174 #define SKIP_BUF_SIZE 1024
176 static void queue_con(struct ceph_connection
*con
);
177 static void con_work(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 ceph_msgr_slab_exit();
279 BUG_ON(zero_page
== NULL
);
281 page_cache_release(zero_page
);
285 int ceph_msgr_init(void)
287 BUG_ON(zero_page
!= NULL
);
288 zero_page
= ZERO_PAGE(0);
289 page_cache_get(zero_page
);
291 if (ceph_msgr_slab_init())
294 ceph_msgr_wq
= alloc_workqueue("ceph-msgr", 0, 0);
298 pr_err("msgr_init failed to create workqueue\n");
303 EXPORT_SYMBOL(ceph_msgr_init
);
305 void ceph_msgr_exit(void)
307 BUG_ON(ceph_msgr_wq
== NULL
);
311 EXPORT_SYMBOL(ceph_msgr_exit
);
313 void ceph_msgr_flush(void)
315 flush_workqueue(ceph_msgr_wq
);
317 EXPORT_SYMBOL(ceph_msgr_flush
);
319 /* Connection socket state transition functions */
321 static void con_sock_state_init(struct ceph_connection
*con
)
325 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSED
);
326 if (WARN_ON(old_state
!= CON_SOCK_STATE_NEW
))
327 printk("%s: unexpected old state %d\n", __func__
, old_state
);
328 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
329 CON_SOCK_STATE_CLOSED
);
332 static void con_sock_state_connecting(struct ceph_connection
*con
)
336 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CONNECTING
);
337 if (WARN_ON(old_state
!= CON_SOCK_STATE_CLOSED
))
338 printk("%s: unexpected old state %d\n", __func__
, old_state
);
339 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
340 CON_SOCK_STATE_CONNECTING
);
343 static void con_sock_state_connected(struct ceph_connection
*con
)
347 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CONNECTED
);
348 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTING
))
349 printk("%s: unexpected old state %d\n", __func__
, old_state
);
350 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
351 CON_SOCK_STATE_CONNECTED
);
354 static void con_sock_state_closing(struct ceph_connection
*con
)
358 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSING
);
359 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTING
&&
360 old_state
!= CON_SOCK_STATE_CONNECTED
&&
361 old_state
!= CON_SOCK_STATE_CLOSING
))
362 printk("%s: unexpected old state %d\n", __func__
, old_state
);
363 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
364 CON_SOCK_STATE_CLOSING
);
367 static void con_sock_state_closed(struct ceph_connection
*con
)
371 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSED
);
372 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTED
&&
373 old_state
!= CON_SOCK_STATE_CLOSING
&&
374 old_state
!= CON_SOCK_STATE_CONNECTING
&&
375 old_state
!= CON_SOCK_STATE_CLOSED
))
376 printk("%s: unexpected old state %d\n", __func__
, old_state
);
377 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
378 CON_SOCK_STATE_CLOSED
);
382 * socket callback functions
385 /* data available on socket, or listen socket received a connect */
386 static void ceph_sock_data_ready(struct sock
*sk
, int count_unused
)
388 struct ceph_connection
*con
= sk
->sk_user_data
;
389 if (atomic_read(&con
->msgr
->stopping
)) {
393 if (sk
->sk_state
!= TCP_CLOSE_WAIT
) {
394 dout("%s on %p state = %lu, queueing work\n", __func__
,
400 /* socket has buffer space for writing */
401 static void ceph_sock_write_space(struct sock
*sk
)
403 struct ceph_connection
*con
= sk
->sk_user_data
;
405 /* only queue to workqueue if there is data we want to write,
406 * and there is sufficient space in the socket buffer to accept
407 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
408 * doesn't get called again until try_write() fills the socket
409 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
410 * and net/core/stream.c:sk_stream_write_space().
412 if (con_flag_test(con
, CON_FLAG_WRITE_PENDING
)) {
413 if (sk_stream_is_writeable(sk
)) {
414 dout("%s %p queueing write work\n", __func__
, con
);
415 clear_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
419 dout("%s %p nothing to write\n", __func__
, con
);
423 /* socket's state has changed */
424 static void ceph_sock_state_change(struct sock
*sk
)
426 struct ceph_connection
*con
= sk
->sk_user_data
;
428 dout("%s %p state = %lu sk_state = %u\n", __func__
,
429 con
, con
->state
, sk
->sk_state
);
431 switch (sk
->sk_state
) {
433 dout("%s TCP_CLOSE\n", __func__
);
435 dout("%s TCP_CLOSE_WAIT\n", __func__
);
436 con_sock_state_closing(con
);
437 con_flag_set(con
, CON_FLAG_SOCK_CLOSED
);
440 case TCP_ESTABLISHED
:
441 dout("%s TCP_ESTABLISHED\n", __func__
);
442 con_sock_state_connected(con
);
445 default: /* Everything else is uninteresting */
451 * set up socket callbacks
453 static void set_sock_callbacks(struct socket
*sock
,
454 struct ceph_connection
*con
)
456 struct sock
*sk
= sock
->sk
;
457 sk
->sk_user_data
= con
;
458 sk
->sk_data_ready
= ceph_sock_data_ready
;
459 sk
->sk_write_space
= ceph_sock_write_space
;
460 sk
->sk_state_change
= ceph_sock_state_change
;
469 * initiate connection to a remote socket.
471 static int ceph_tcp_connect(struct ceph_connection
*con
)
473 struct sockaddr_storage
*paddr
= &con
->peer_addr
.in_addr
;
478 ret
= sock_create_kern(con
->peer_addr
.in_addr
.ss_family
, SOCK_STREAM
,
482 sock
->sk
->sk_allocation
= GFP_NOFS
;
484 #ifdef CONFIG_LOCKDEP
485 lockdep_set_class(&sock
->sk
->sk_lock
, &socket_class
);
488 set_sock_callbacks(sock
, con
);
490 dout("connect %s\n", ceph_pr_addr(&con
->peer_addr
.in_addr
));
492 con_sock_state_connecting(con
);
493 ret
= sock
->ops
->connect(sock
, (struct sockaddr
*)paddr
, sizeof(*paddr
),
495 if (ret
== -EINPROGRESS
) {
496 dout("connect %s EINPROGRESS sk_state = %u\n",
497 ceph_pr_addr(&con
->peer_addr
.in_addr
),
499 } else if (ret
< 0) {
500 pr_err("connect %s error %d\n",
501 ceph_pr_addr(&con
->peer_addr
.in_addr
), ret
);
503 con
->error_msg
= "connect error";
511 static int ceph_tcp_recvmsg(struct socket
*sock
, void *buf
, size_t len
)
513 struct kvec iov
= {buf
, len
};
514 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
517 r
= kernel_recvmsg(sock
, &msg
, &iov
, 1, len
, msg
.msg_flags
);
523 static int ceph_tcp_recvpage(struct socket
*sock
, struct page
*page
,
524 int page_offset
, size_t length
)
529 BUG_ON(page_offset
+ length
> PAGE_SIZE
);
533 ret
= ceph_tcp_recvmsg(sock
, kaddr
+ page_offset
, length
);
540 * write something. @more is true if caller will be sending more data
543 static int ceph_tcp_sendmsg(struct socket
*sock
, struct kvec
*iov
,
544 size_t kvlen
, size_t len
, int more
)
546 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
550 msg
.msg_flags
|= MSG_MORE
;
552 msg
.msg_flags
|= MSG_EOR
; /* superfluous, but what the hell */
554 r
= kernel_sendmsg(sock
, &msg
, iov
, kvlen
, len
);
560 static int ceph_tcp_sendpage(struct socket
*sock
, struct page
*page
,
561 int offset
, size_t size
, bool more
)
563 int flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
| (more
? MSG_MORE
: MSG_EOR
);
566 ret
= kernel_sendpage(sock
, page
, offset
, size
, flags
);
575 * Shutdown/close the socket for the given connection.
577 static int con_close_socket(struct ceph_connection
*con
)
581 dout("con_close_socket on %p sock %p\n", con
, con
->sock
);
583 rc
= con
->sock
->ops
->shutdown(con
->sock
, SHUT_RDWR
);
584 sock_release(con
->sock
);
589 * Forcibly clear the SOCK_CLOSED flag. It gets set
590 * independent of the connection mutex, and we could have
591 * received a socket close event before we had the chance to
592 * shut the socket down.
594 con_flag_clear(con
, CON_FLAG_SOCK_CLOSED
);
596 con_sock_state_closed(con
);
601 * Reset a connection. Discard all incoming and outgoing messages
602 * and clear *_seq state.
604 static void ceph_msg_remove(struct ceph_msg
*msg
)
606 list_del_init(&msg
->list_head
);
607 BUG_ON(msg
->con
== NULL
);
608 msg
->con
->ops
->put(msg
->con
);
613 static void ceph_msg_remove_list(struct list_head
*head
)
615 while (!list_empty(head
)) {
616 struct ceph_msg
*msg
= list_first_entry(head
, struct ceph_msg
,
618 ceph_msg_remove(msg
);
622 static void reset_connection(struct ceph_connection
*con
)
624 /* reset connection, out_queue, msg_ and connect_seq */
625 /* discard existing out_queue and msg_seq */
626 dout("reset_connection %p\n", con
);
627 ceph_msg_remove_list(&con
->out_queue
);
628 ceph_msg_remove_list(&con
->out_sent
);
631 BUG_ON(con
->in_msg
->con
!= con
);
632 con
->in_msg
->con
= NULL
;
633 ceph_msg_put(con
->in_msg
);
638 con
->connect_seq
= 0;
641 ceph_msg_put(con
->out_msg
);
645 con
->in_seq_acked
= 0;
649 * mark a peer down. drop any open connections.
651 void ceph_con_close(struct ceph_connection
*con
)
653 mutex_lock(&con
->mutex
);
654 dout("con_close %p peer %s\n", con
,
655 ceph_pr_addr(&con
->peer_addr
.in_addr
));
656 con
->state
= CON_STATE_CLOSED
;
658 con_flag_clear(con
, CON_FLAG_LOSSYTX
); /* so we retry next connect */
659 con_flag_clear(con
, CON_FLAG_KEEPALIVE_PENDING
);
660 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
661 con_flag_clear(con
, CON_FLAG_BACKOFF
);
663 reset_connection(con
);
664 con
->peer_global_seq
= 0;
665 cancel_delayed_work(&con
->work
);
666 con_close_socket(con
);
667 mutex_unlock(&con
->mutex
);
669 EXPORT_SYMBOL(ceph_con_close
);
672 * Reopen a closed connection, with a new peer address.
674 void ceph_con_open(struct ceph_connection
*con
,
675 __u8 entity_type
, __u64 entity_num
,
676 struct ceph_entity_addr
*addr
)
678 mutex_lock(&con
->mutex
);
679 dout("con_open %p %s\n", con
, ceph_pr_addr(&addr
->in_addr
));
681 WARN_ON(con
->state
!= CON_STATE_CLOSED
);
682 con
->state
= CON_STATE_PREOPEN
;
684 con
->peer_name
.type
= (__u8
) entity_type
;
685 con
->peer_name
.num
= cpu_to_le64(entity_num
);
687 memcpy(&con
->peer_addr
, addr
, sizeof(*addr
));
688 con
->delay
= 0; /* reset backoff memory */
689 mutex_unlock(&con
->mutex
);
692 EXPORT_SYMBOL(ceph_con_open
);
695 * return true if this connection ever successfully opened
697 bool ceph_con_opened(struct ceph_connection
*con
)
699 return con
->connect_seq
> 0;
703 * initialize a new connection.
705 void ceph_con_init(struct ceph_connection
*con
, void *private,
706 const struct ceph_connection_operations
*ops
,
707 struct ceph_messenger
*msgr
)
709 dout("con_init %p\n", con
);
710 memset(con
, 0, sizeof(*con
));
711 con
->private = private;
715 con_sock_state_init(con
);
717 mutex_init(&con
->mutex
);
718 INIT_LIST_HEAD(&con
->out_queue
);
719 INIT_LIST_HEAD(&con
->out_sent
);
720 INIT_DELAYED_WORK(&con
->work
, con_work
);
722 con
->state
= CON_STATE_CLOSED
;
724 EXPORT_SYMBOL(ceph_con_init
);
728 * We maintain a global counter to order connection attempts. Get
729 * a unique seq greater than @gt.
731 static u32
get_global_seq(struct ceph_messenger
*msgr
, u32 gt
)
735 spin_lock(&msgr
->global_seq_lock
);
736 if (msgr
->global_seq
< gt
)
737 msgr
->global_seq
= gt
;
738 ret
= ++msgr
->global_seq
;
739 spin_unlock(&msgr
->global_seq_lock
);
743 static void con_out_kvec_reset(struct ceph_connection
*con
)
745 con
->out_kvec_left
= 0;
746 con
->out_kvec_bytes
= 0;
747 con
->out_kvec_cur
= &con
->out_kvec
[0];
750 static void con_out_kvec_add(struct ceph_connection
*con
,
751 size_t size
, void *data
)
755 index
= con
->out_kvec_left
;
756 BUG_ON(index
>= ARRAY_SIZE(con
->out_kvec
));
758 con
->out_kvec
[index
].iov_len
= size
;
759 con
->out_kvec
[index
].iov_base
= data
;
760 con
->out_kvec_left
++;
761 con
->out_kvec_bytes
+= size
;
767 * For a bio data item, a piece is whatever remains of the next
768 * entry in the current bio iovec, or the first entry in the next
771 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor
*cursor
,
774 struct ceph_msg_data
*data
= cursor
->data
;
777 BUG_ON(data
->type
!= CEPH_MSG_DATA_BIO
);
781 BUG_ON(!bio
->bi_vcnt
);
783 cursor
->resid
= min(length
, data
->bio_length
);
785 cursor
->vector_index
= 0;
786 cursor
->vector_offset
= 0;
787 cursor
->last_piece
= length
<= bio
->bi_io_vec
[0].bv_len
;
790 static struct page
*ceph_msg_data_bio_next(struct ceph_msg_data_cursor
*cursor
,
794 struct ceph_msg_data
*data
= cursor
->data
;
796 struct bio_vec
*bio_vec
;
799 BUG_ON(data
->type
!= CEPH_MSG_DATA_BIO
);
804 index
= cursor
->vector_index
;
805 BUG_ON(index
>= (unsigned int) bio
->bi_vcnt
);
807 bio_vec
= &bio
->bi_io_vec
[index
];
808 BUG_ON(cursor
->vector_offset
>= bio_vec
->bv_len
);
809 *page_offset
= (size_t) (bio_vec
->bv_offset
+ cursor
->vector_offset
);
810 BUG_ON(*page_offset
>= PAGE_SIZE
);
811 if (cursor
->last_piece
) /* pagelist offset is always 0 */
812 *length
= cursor
->resid
;
814 *length
= (size_t) (bio_vec
->bv_len
- cursor
->vector_offset
);
815 BUG_ON(*length
> cursor
->resid
);
816 BUG_ON(*page_offset
+ *length
> PAGE_SIZE
);
818 return bio_vec
->bv_page
;
821 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor
*cursor
,
825 struct bio_vec
*bio_vec
;
828 BUG_ON(cursor
->data
->type
!= CEPH_MSG_DATA_BIO
);
833 index
= cursor
->vector_index
;
834 BUG_ON(index
>= (unsigned int) bio
->bi_vcnt
);
835 bio_vec
= &bio
->bi_io_vec
[index
];
837 /* Advance the cursor offset */
839 BUG_ON(cursor
->resid
< bytes
);
840 cursor
->resid
-= bytes
;
841 cursor
->vector_offset
+= bytes
;
842 if (cursor
->vector_offset
< bio_vec
->bv_len
)
843 return false; /* more bytes to process in this segment */
844 BUG_ON(cursor
->vector_offset
!= bio_vec
->bv_len
);
846 /* Move on to the next segment, and possibly the next bio */
848 if (++index
== (unsigned int) bio
->bi_vcnt
) {
853 cursor
->vector_index
= index
;
854 cursor
->vector_offset
= 0;
856 if (!cursor
->last_piece
) {
857 BUG_ON(!cursor
->resid
);
859 /* A short read is OK, so use <= rather than == */
860 if (cursor
->resid
<= bio
->bi_io_vec
[index
].bv_len
)
861 cursor
->last_piece
= true;
866 #endif /* CONFIG_BLOCK */
869 * For a page array, a piece comes from the first page in the array
870 * that has not already been fully consumed.
872 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor
*cursor
,
875 struct ceph_msg_data
*data
= cursor
->data
;
878 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGES
);
880 BUG_ON(!data
->pages
);
881 BUG_ON(!data
->length
);
883 cursor
->resid
= min(length
, data
->length
);
884 page_count
= calc_pages_for(data
->alignment
, (u64
)data
->length
);
885 cursor
->page_offset
= data
->alignment
& ~PAGE_MASK
;
886 cursor
->page_index
= 0;
887 BUG_ON(page_count
> (int)USHRT_MAX
);
888 cursor
->page_count
= (unsigned short)page_count
;
889 BUG_ON(length
> SIZE_MAX
- cursor
->page_offset
);
890 cursor
->last_piece
= (size_t)cursor
->page_offset
+ length
<= PAGE_SIZE
;
894 ceph_msg_data_pages_next(struct ceph_msg_data_cursor
*cursor
,
895 size_t *page_offset
, size_t *length
)
897 struct ceph_msg_data
*data
= cursor
->data
;
899 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGES
);
901 BUG_ON(cursor
->page_index
>= cursor
->page_count
);
902 BUG_ON(cursor
->page_offset
>= PAGE_SIZE
);
904 *page_offset
= cursor
->page_offset
;
905 if (cursor
->last_piece
)
906 *length
= cursor
->resid
;
908 *length
= PAGE_SIZE
- *page_offset
;
910 return data
->pages
[cursor
->page_index
];
913 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor
*cursor
,
916 BUG_ON(cursor
->data
->type
!= CEPH_MSG_DATA_PAGES
);
918 BUG_ON(cursor
->page_offset
+ bytes
> PAGE_SIZE
);
920 /* Advance the cursor page offset */
922 cursor
->resid
-= bytes
;
923 cursor
->page_offset
= (cursor
->page_offset
+ bytes
) & ~PAGE_MASK
;
924 if (!bytes
|| cursor
->page_offset
)
925 return false; /* more bytes to process in the current page */
927 /* Move on to the next page; offset is already at 0 */
929 BUG_ON(cursor
->page_index
>= cursor
->page_count
);
930 cursor
->page_index
++;
931 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
937 * For a pagelist, a piece is whatever remains to be consumed in the
938 * first page in the list, or the front of the next page.
941 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor
*cursor
,
944 struct ceph_msg_data
*data
= cursor
->data
;
945 struct ceph_pagelist
*pagelist
;
948 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
950 pagelist
= data
->pagelist
;
954 return; /* pagelist can be assigned but empty */
956 BUG_ON(list_empty(&pagelist
->head
));
957 page
= list_first_entry(&pagelist
->head
, struct page
, lru
);
959 cursor
->resid
= min(length
, pagelist
->length
);
962 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
966 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor
*cursor
,
967 size_t *page_offset
, size_t *length
)
969 struct ceph_msg_data
*data
= cursor
->data
;
970 struct ceph_pagelist
*pagelist
;
972 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
974 pagelist
= data
->pagelist
;
977 BUG_ON(!cursor
->page
);
978 BUG_ON(cursor
->offset
+ cursor
->resid
!= pagelist
->length
);
980 /* offset of first page in pagelist is always 0 */
981 *page_offset
= cursor
->offset
& ~PAGE_MASK
;
982 if (cursor
->last_piece
)
983 *length
= cursor
->resid
;
985 *length
= PAGE_SIZE
- *page_offset
;
990 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor
*cursor
,
993 struct ceph_msg_data
*data
= cursor
->data
;
994 struct ceph_pagelist
*pagelist
;
996 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
998 pagelist
= data
->pagelist
;
1001 BUG_ON(cursor
->offset
+ cursor
->resid
!= pagelist
->length
);
1002 BUG_ON((cursor
->offset
& ~PAGE_MASK
) + bytes
> PAGE_SIZE
);
1004 /* Advance the cursor offset */
1006 cursor
->resid
-= bytes
;
1007 cursor
->offset
+= bytes
;
1008 /* offset of first page in pagelist is always 0 */
1009 if (!bytes
|| cursor
->offset
& ~PAGE_MASK
)
1010 return false; /* more bytes to process in the current page */
1012 /* Move on to the next page */
1014 BUG_ON(list_is_last(&cursor
->page
->lru
, &pagelist
->head
));
1015 cursor
->page
= list_entry_next(cursor
->page
, lru
);
1016 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
1022 * Message data is handled (sent or received) in pieces, where each
1023 * piece resides on a single page. The network layer might not
1024 * consume an entire piece at once. A data item's cursor keeps
1025 * track of which piece is next to process and how much remains to
1026 * be processed in that piece. It also tracks whether the current
1027 * piece is the last one in the data item.
1029 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor
*cursor
)
1031 size_t length
= cursor
->total_resid
;
1033 switch (cursor
->data
->type
) {
1034 case CEPH_MSG_DATA_PAGELIST
:
1035 ceph_msg_data_pagelist_cursor_init(cursor
, length
);
1037 case CEPH_MSG_DATA_PAGES
:
1038 ceph_msg_data_pages_cursor_init(cursor
, length
);
1041 case CEPH_MSG_DATA_BIO
:
1042 ceph_msg_data_bio_cursor_init(cursor
, length
);
1044 #endif /* CONFIG_BLOCK */
1045 case CEPH_MSG_DATA_NONE
:
1050 cursor
->need_crc
= true;
1053 static void ceph_msg_data_cursor_init(struct ceph_msg
*msg
, size_t length
)
1055 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
1056 struct ceph_msg_data
*data
;
1059 BUG_ON(length
> msg
->data_length
);
1060 BUG_ON(list_empty(&msg
->data
));
1062 cursor
->data_head
= &msg
->data
;
1063 cursor
->total_resid
= length
;
1064 data
= list_first_entry(&msg
->data
, struct ceph_msg_data
, links
);
1065 cursor
->data
= data
;
1067 __ceph_msg_data_cursor_init(cursor
);
1071 * Return the page containing the next piece to process for a given
1072 * data item, and supply the page offset and length of that piece.
1073 * Indicate whether this is the last piece in this data item.
1075 static struct page
*ceph_msg_data_next(struct ceph_msg_data_cursor
*cursor
,
1076 size_t *page_offset
, size_t *length
,
1081 switch (cursor
->data
->type
) {
1082 case CEPH_MSG_DATA_PAGELIST
:
1083 page
= ceph_msg_data_pagelist_next(cursor
, page_offset
, length
);
1085 case CEPH_MSG_DATA_PAGES
:
1086 page
= ceph_msg_data_pages_next(cursor
, page_offset
, length
);
1089 case CEPH_MSG_DATA_BIO
:
1090 page
= ceph_msg_data_bio_next(cursor
, page_offset
, length
);
1092 #endif /* CONFIG_BLOCK */
1093 case CEPH_MSG_DATA_NONE
:
1099 BUG_ON(*page_offset
+ *length
> PAGE_SIZE
);
1102 *last_piece
= cursor
->last_piece
;
1108 * Returns true if the result moves the cursor on to the next piece
1111 static bool ceph_msg_data_advance(struct ceph_msg_data_cursor
*cursor
,
1116 BUG_ON(bytes
> cursor
->resid
);
1117 switch (cursor
->data
->type
) {
1118 case CEPH_MSG_DATA_PAGELIST
:
1119 new_piece
= ceph_msg_data_pagelist_advance(cursor
, bytes
);
1121 case CEPH_MSG_DATA_PAGES
:
1122 new_piece
= ceph_msg_data_pages_advance(cursor
, bytes
);
1125 case CEPH_MSG_DATA_BIO
:
1126 new_piece
= ceph_msg_data_bio_advance(cursor
, bytes
);
1128 #endif /* CONFIG_BLOCK */
1129 case CEPH_MSG_DATA_NONE
:
1134 cursor
->total_resid
-= bytes
;
1136 if (!cursor
->resid
&& cursor
->total_resid
) {
1137 WARN_ON(!cursor
->last_piece
);
1138 BUG_ON(list_is_last(&cursor
->data
->links
, cursor
->data_head
));
1139 cursor
->data
= list_entry_next(cursor
->data
, links
);
1140 __ceph_msg_data_cursor_init(cursor
);
1143 cursor
->need_crc
= new_piece
;
1148 static void prepare_message_data(struct ceph_msg
*msg
, u32 data_len
)
1153 /* Initialize data cursor */
1155 ceph_msg_data_cursor_init(msg
, (size_t)data_len
);
1159 * Prepare footer for currently outgoing message, and finish things
1160 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1162 static void prepare_write_message_footer(struct ceph_connection
*con
)
1164 struct ceph_msg
*m
= con
->out_msg
;
1165 int v
= con
->out_kvec_left
;
1167 m
->footer
.flags
|= CEPH_MSG_FOOTER_COMPLETE
;
1169 dout("prepare_write_message_footer %p\n", con
);
1170 con
->out_kvec_is_msg
= true;
1171 con
->out_kvec
[v
].iov_base
= &m
->footer
;
1172 con
->out_kvec
[v
].iov_len
= sizeof(m
->footer
);
1173 con
->out_kvec_bytes
+= sizeof(m
->footer
);
1174 con
->out_kvec_left
++;
1175 con
->out_more
= m
->more_to_follow
;
1176 con
->out_msg_done
= true;
1180 * Prepare headers for the next outgoing message.
1182 static void prepare_write_message(struct ceph_connection
*con
)
1187 con_out_kvec_reset(con
);
1188 con
->out_kvec_is_msg
= true;
1189 con
->out_msg_done
= false;
1191 /* Sneak an ack in there first? If we can get it into the same
1192 * TCP packet that's a good thing. */
1193 if (con
->in_seq
> con
->in_seq_acked
) {
1194 con
->in_seq_acked
= con
->in_seq
;
1195 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
1196 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1197 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1198 &con
->out_temp_ack
);
1201 BUG_ON(list_empty(&con
->out_queue
));
1202 m
= list_first_entry(&con
->out_queue
, struct ceph_msg
, list_head
);
1204 BUG_ON(m
->con
!= con
);
1206 /* put message on sent list */
1208 list_move_tail(&m
->list_head
, &con
->out_sent
);
1211 * only assign outgoing seq # if we haven't sent this message
1212 * yet. if it is requeued, resend with it's original seq.
1214 if (m
->needs_out_seq
) {
1215 m
->hdr
.seq
= cpu_to_le64(++con
->out_seq
);
1216 m
->needs_out_seq
= false;
1218 WARN_ON(m
->data_length
!= le32_to_cpu(m
->hdr
.data_len
));
1220 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1221 m
, con
->out_seq
, le16_to_cpu(m
->hdr
.type
),
1222 le32_to_cpu(m
->hdr
.front_len
), le32_to_cpu(m
->hdr
.middle_len
),
1224 BUG_ON(le32_to_cpu(m
->hdr
.front_len
) != m
->front
.iov_len
);
1226 /* tag + hdr + front + middle */
1227 con_out_kvec_add(con
, sizeof (tag_msg
), &tag_msg
);
1228 con_out_kvec_add(con
, sizeof (m
->hdr
), &m
->hdr
);
1229 con_out_kvec_add(con
, m
->front
.iov_len
, m
->front
.iov_base
);
1232 con_out_kvec_add(con
, m
->middle
->vec
.iov_len
,
1233 m
->middle
->vec
.iov_base
);
1235 /* fill in crc (except data pages), footer */
1236 crc
= crc32c(0, &m
->hdr
, offsetof(struct ceph_msg_header
, crc
));
1237 con
->out_msg
->hdr
.crc
= cpu_to_le32(crc
);
1238 con
->out_msg
->footer
.flags
= 0;
1240 crc
= crc32c(0, m
->front
.iov_base
, m
->front
.iov_len
);
1241 con
->out_msg
->footer
.front_crc
= cpu_to_le32(crc
);
1243 crc
= crc32c(0, m
->middle
->vec
.iov_base
,
1244 m
->middle
->vec
.iov_len
);
1245 con
->out_msg
->footer
.middle_crc
= cpu_to_le32(crc
);
1247 con
->out_msg
->footer
.middle_crc
= 0;
1248 dout("%s front_crc %u middle_crc %u\n", __func__
,
1249 le32_to_cpu(con
->out_msg
->footer
.front_crc
),
1250 le32_to_cpu(con
->out_msg
->footer
.middle_crc
));
1252 /* is there a data payload? */
1253 con
->out_msg
->footer
.data_crc
= 0;
1254 if (m
->data_length
) {
1255 prepare_message_data(con
->out_msg
, m
->data_length
);
1256 con
->out_more
= 1; /* data + footer will follow */
1258 /* no, queue up footer too and be done */
1259 prepare_write_message_footer(con
);
1262 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1268 static void prepare_write_ack(struct ceph_connection
*con
)
1270 dout("prepare_write_ack %p %llu -> %llu\n", con
,
1271 con
->in_seq_acked
, con
->in_seq
);
1272 con
->in_seq_acked
= con
->in_seq
;
1274 con_out_kvec_reset(con
);
1276 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
1278 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1279 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1280 &con
->out_temp_ack
);
1282 con
->out_more
= 1; /* more will follow.. eventually.. */
1283 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1287 * Prepare to share the seq during handshake
1289 static void prepare_write_seq(struct ceph_connection
*con
)
1291 dout("prepare_write_seq %p %llu -> %llu\n", con
,
1292 con
->in_seq_acked
, con
->in_seq
);
1293 con
->in_seq_acked
= con
->in_seq
;
1295 con_out_kvec_reset(con
);
1297 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1298 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1299 &con
->out_temp_ack
);
1301 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1305 * Prepare to write keepalive byte.
1307 static void prepare_write_keepalive(struct ceph_connection
*con
)
1309 dout("prepare_write_keepalive %p\n", con
);
1310 con_out_kvec_reset(con
);
1311 con_out_kvec_add(con
, sizeof (tag_keepalive
), &tag_keepalive
);
1312 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1316 * Connection negotiation.
1319 static struct ceph_auth_handshake
*get_connect_authorizer(struct ceph_connection
*con
,
1322 struct ceph_auth_handshake
*auth
;
1324 if (!con
->ops
->get_authorizer
) {
1325 con
->out_connect
.authorizer_protocol
= CEPH_AUTH_UNKNOWN
;
1326 con
->out_connect
.authorizer_len
= 0;
1330 /* Can't hold the mutex while getting authorizer */
1331 mutex_unlock(&con
->mutex
);
1332 auth
= con
->ops
->get_authorizer(con
, auth_proto
, con
->auth_retry
);
1333 mutex_lock(&con
->mutex
);
1337 if (con
->state
!= CON_STATE_NEGOTIATING
)
1338 return ERR_PTR(-EAGAIN
);
1340 con
->auth_reply_buf
= auth
->authorizer_reply_buf
;
1341 con
->auth_reply_buf_len
= auth
->authorizer_reply_buf_len
;
1346 * We connected to a peer and are saying hello.
1348 static void prepare_write_banner(struct ceph_connection
*con
)
1350 con_out_kvec_add(con
, strlen(CEPH_BANNER
), CEPH_BANNER
);
1351 con_out_kvec_add(con
, sizeof (con
->msgr
->my_enc_addr
),
1352 &con
->msgr
->my_enc_addr
);
1355 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1358 static int prepare_write_connect(struct ceph_connection
*con
)
1360 unsigned int global_seq
= get_global_seq(con
->msgr
, 0);
1363 struct ceph_auth_handshake
*auth
;
1365 switch (con
->peer_name
.type
) {
1366 case CEPH_ENTITY_TYPE_MON
:
1367 proto
= CEPH_MONC_PROTOCOL
;
1369 case CEPH_ENTITY_TYPE_OSD
:
1370 proto
= CEPH_OSDC_PROTOCOL
;
1372 case CEPH_ENTITY_TYPE_MDS
:
1373 proto
= CEPH_MDSC_PROTOCOL
;
1379 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con
,
1380 con
->connect_seq
, global_seq
, proto
);
1382 con
->out_connect
.features
= cpu_to_le64(con
->msgr
->supported_features
);
1383 con
->out_connect
.host_type
= cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT
);
1384 con
->out_connect
.connect_seq
= cpu_to_le32(con
->connect_seq
);
1385 con
->out_connect
.global_seq
= cpu_to_le32(global_seq
);
1386 con
->out_connect
.protocol_version
= cpu_to_le32(proto
);
1387 con
->out_connect
.flags
= 0;
1389 auth_proto
= CEPH_AUTH_UNKNOWN
;
1390 auth
= get_connect_authorizer(con
, &auth_proto
);
1392 return PTR_ERR(auth
);
1394 con
->out_connect
.authorizer_protocol
= cpu_to_le32(auth_proto
);
1395 con
->out_connect
.authorizer_len
= auth
?
1396 cpu_to_le32(auth
->authorizer_buf_len
) : 0;
1398 con_out_kvec_add(con
, sizeof (con
->out_connect
),
1400 if (auth
&& auth
->authorizer_buf_len
)
1401 con_out_kvec_add(con
, auth
->authorizer_buf_len
,
1402 auth
->authorizer_buf
);
1405 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1411 * write as much of pending kvecs to the socket as we can.
1413 * 0 -> socket full, but more to do
1416 static int write_partial_kvec(struct ceph_connection
*con
)
1420 dout("write_partial_kvec %p %d left\n", con
, con
->out_kvec_bytes
);
1421 while (con
->out_kvec_bytes
> 0) {
1422 ret
= ceph_tcp_sendmsg(con
->sock
, con
->out_kvec_cur
,
1423 con
->out_kvec_left
, con
->out_kvec_bytes
,
1427 con
->out_kvec_bytes
-= ret
;
1428 if (con
->out_kvec_bytes
== 0)
1431 /* account for full iov entries consumed */
1432 while (ret
>= con
->out_kvec_cur
->iov_len
) {
1433 BUG_ON(!con
->out_kvec_left
);
1434 ret
-= con
->out_kvec_cur
->iov_len
;
1435 con
->out_kvec_cur
++;
1436 con
->out_kvec_left
--;
1438 /* and for a partially-consumed entry */
1440 con
->out_kvec_cur
->iov_len
-= ret
;
1441 con
->out_kvec_cur
->iov_base
+= ret
;
1444 con
->out_kvec_left
= 0;
1445 con
->out_kvec_is_msg
= false;
1448 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con
,
1449 con
->out_kvec_bytes
, con
->out_kvec_left
, ret
);
1450 return ret
; /* done! */
1453 static u32
ceph_crc32c_page(u32 crc
, struct page
*page
,
1454 unsigned int page_offset
,
1455 unsigned int length
)
1460 BUG_ON(kaddr
== NULL
);
1461 crc
= crc32c(crc
, kaddr
+ page_offset
, length
);
1467 * Write as much message data payload as we can. If we finish, queue
1469 * 1 -> done, footer is now queued in out_kvec[].
1470 * 0 -> socket full, but more to do
1473 static int write_partial_message_data(struct ceph_connection
*con
)
1475 struct ceph_msg
*msg
= con
->out_msg
;
1476 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
1477 bool do_datacrc
= !con
->msgr
->nocrc
;
1480 dout("%s %p msg %p\n", __func__
, con
, msg
);
1482 if (list_empty(&msg
->data
))
1486 * Iterate through each page that contains data to be
1487 * written, and send as much as possible for each.
1489 * If we are calculating the data crc (the default), we will
1490 * need to map the page. If we have no pages, they have
1491 * been revoked, so use the zero page.
1493 crc
= do_datacrc
? le32_to_cpu(msg
->footer
.data_crc
) : 0;
1494 while (cursor
->resid
) {
1502 page
= ceph_msg_data_next(&msg
->cursor
, &page_offset
, &length
,
1504 ret
= ceph_tcp_sendpage(con
->sock
, page
, page_offset
,
1505 length
, last_piece
);
1508 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1512 if (do_datacrc
&& cursor
->need_crc
)
1513 crc
= ceph_crc32c_page(crc
, page
, page_offset
, length
);
1514 need_crc
= ceph_msg_data_advance(&msg
->cursor
, (size_t)ret
);
1517 dout("%s %p msg %p done\n", __func__
, con
, msg
);
1519 /* prepare and queue up footer, too */
1521 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1523 msg
->footer
.flags
|= CEPH_MSG_FOOTER_NOCRC
;
1524 con_out_kvec_reset(con
);
1525 prepare_write_message_footer(con
);
1527 return 1; /* must return > 0 to indicate success */
1533 static int write_partial_skip(struct ceph_connection
*con
)
1537 while (con
->out_skip
> 0) {
1538 size_t size
= min(con
->out_skip
, (int) PAGE_CACHE_SIZE
);
1540 ret
= ceph_tcp_sendpage(con
->sock
, zero_page
, 0, size
, true);
1543 con
->out_skip
-= ret
;
1551 * Prepare to read connection handshake, or an ack.
1553 static void prepare_read_banner(struct ceph_connection
*con
)
1555 dout("prepare_read_banner %p\n", con
);
1556 con
->in_base_pos
= 0;
1559 static void prepare_read_connect(struct ceph_connection
*con
)
1561 dout("prepare_read_connect %p\n", con
);
1562 con
->in_base_pos
= 0;
1565 static void prepare_read_ack(struct ceph_connection
*con
)
1567 dout("prepare_read_ack %p\n", con
);
1568 con
->in_base_pos
= 0;
1571 static void prepare_read_seq(struct ceph_connection
*con
)
1573 dout("prepare_read_seq %p\n", con
);
1574 con
->in_base_pos
= 0;
1575 con
->in_tag
= CEPH_MSGR_TAG_SEQ
;
1578 static void prepare_read_tag(struct ceph_connection
*con
)
1580 dout("prepare_read_tag %p\n", con
);
1581 con
->in_base_pos
= 0;
1582 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1586 * Prepare to read a message.
1588 static int prepare_read_message(struct ceph_connection
*con
)
1590 dout("prepare_read_message %p\n", con
);
1591 BUG_ON(con
->in_msg
!= NULL
);
1592 con
->in_base_pos
= 0;
1593 con
->in_front_crc
= con
->in_middle_crc
= con
->in_data_crc
= 0;
1598 static int read_partial(struct ceph_connection
*con
,
1599 int end
, int size
, void *object
)
1601 while (con
->in_base_pos
< end
) {
1602 int left
= end
- con
->in_base_pos
;
1603 int have
= size
- left
;
1604 int ret
= ceph_tcp_recvmsg(con
->sock
, object
+ have
, left
);
1607 con
->in_base_pos
+= ret
;
1614 * Read all or part of the connect-side handshake on a new connection
1616 static int read_partial_banner(struct ceph_connection
*con
)
1622 dout("read_partial_banner %p at %d\n", con
, con
->in_base_pos
);
1625 size
= strlen(CEPH_BANNER
);
1627 ret
= read_partial(con
, end
, size
, con
->in_banner
);
1631 size
= sizeof (con
->actual_peer_addr
);
1633 ret
= read_partial(con
, end
, size
, &con
->actual_peer_addr
);
1637 size
= sizeof (con
->peer_addr_for_me
);
1639 ret
= read_partial(con
, end
, size
, &con
->peer_addr_for_me
);
1647 static int read_partial_connect(struct ceph_connection
*con
)
1653 dout("read_partial_connect %p at %d\n", con
, con
->in_base_pos
);
1655 size
= sizeof (con
->in_reply
);
1657 ret
= read_partial(con
, end
, size
, &con
->in_reply
);
1661 size
= le32_to_cpu(con
->in_reply
.authorizer_len
);
1663 ret
= read_partial(con
, end
, size
, con
->auth_reply_buf
);
1667 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1668 con
, (int)con
->in_reply
.tag
,
1669 le32_to_cpu(con
->in_reply
.connect_seq
),
1670 le32_to_cpu(con
->in_reply
.global_seq
));
1677 * Verify the hello banner looks okay.
1679 static int verify_hello(struct ceph_connection
*con
)
1681 if (memcmp(con
->in_banner
, CEPH_BANNER
, strlen(CEPH_BANNER
))) {
1682 pr_err("connect to %s got bad banner\n",
1683 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1684 con
->error_msg
= "protocol error, bad banner";
1690 static bool addr_is_blank(struct sockaddr_storage
*ss
)
1692 switch (ss
->ss_family
) {
1694 return ((struct sockaddr_in
*)ss
)->sin_addr
.s_addr
== 0;
1697 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[0] == 0 &&
1698 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[1] == 0 &&
1699 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[2] == 0 &&
1700 ((struct sockaddr_in6
*)ss
)->sin6_addr
.s6_addr32
[3] == 0;
1705 static int addr_port(struct sockaddr_storage
*ss
)
1707 switch (ss
->ss_family
) {
1709 return ntohs(((struct sockaddr_in
*)ss
)->sin_port
);
1711 return ntohs(((struct sockaddr_in6
*)ss
)->sin6_port
);
1716 static void addr_set_port(struct sockaddr_storage
*ss
, int p
)
1718 switch (ss
->ss_family
) {
1720 ((struct sockaddr_in
*)ss
)->sin_port
= htons(p
);
1723 ((struct sockaddr_in6
*)ss
)->sin6_port
= htons(p
);
1729 * Unlike other *_pton function semantics, zero indicates success.
1731 static int ceph_pton(const char *str
, size_t len
, struct sockaddr_storage
*ss
,
1732 char delim
, const char **ipend
)
1734 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
1735 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
1737 memset(ss
, 0, sizeof(*ss
));
1739 if (in4_pton(str
, len
, (u8
*)&in4
->sin_addr
.s_addr
, delim
, ipend
)) {
1740 ss
->ss_family
= AF_INET
;
1744 if (in6_pton(str
, len
, (u8
*)&in6
->sin6_addr
.s6_addr
, delim
, ipend
)) {
1745 ss
->ss_family
= AF_INET6
;
1753 * Extract hostname string and resolve using kernel DNS facility.
1755 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1756 static int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1757 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1759 const char *end
, *delim_p
;
1760 char *colon_p
, *ip_addr
= NULL
;
1764 * The end of the hostname occurs immediately preceding the delimiter or
1765 * the port marker (':') where the delimiter takes precedence.
1767 delim_p
= memchr(name
, delim
, namelen
);
1768 colon_p
= memchr(name
, ':', namelen
);
1770 if (delim_p
&& colon_p
)
1771 end
= delim_p
< colon_p
? delim_p
: colon_p
;
1772 else if (!delim_p
&& colon_p
)
1776 if (!end
) /* case: hostname:/ */
1777 end
= name
+ namelen
;
1783 /* do dns_resolve upcall */
1784 ip_len
= dns_query(NULL
, name
, end
- name
, NULL
, &ip_addr
, NULL
);
1786 ret
= ceph_pton(ip_addr
, ip_len
, ss
, -1, NULL
);
1794 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end
- name
), name
,
1795 ret
, ret
? "failed" : ceph_pr_addr(ss
));
1800 static inline int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1801 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1808 * Parse a server name (IP or hostname). If a valid IP address is not found
1809 * then try to extract a hostname to resolve using userspace DNS upcall.
1811 static int ceph_parse_server_name(const char *name
, size_t namelen
,
1812 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1816 ret
= ceph_pton(name
, namelen
, ss
, delim
, ipend
);
1818 ret
= ceph_dns_resolve_name(name
, namelen
, ss
, delim
, ipend
);
1824 * Parse an ip[:port] list into an addr array. Use the default
1825 * monitor port if a port isn't specified.
1827 int ceph_parse_ips(const char *c
, const char *end
,
1828 struct ceph_entity_addr
*addr
,
1829 int max_count
, int *count
)
1831 int i
, ret
= -EINVAL
;
1834 dout("parse_ips on '%.*s'\n", (int)(end
-c
), c
);
1835 for (i
= 0; i
< max_count
; i
++) {
1837 struct sockaddr_storage
*ss
= &addr
[i
].in_addr
;
1846 ret
= ceph_parse_server_name(p
, end
- p
, ss
, delim
, &ipend
);
1855 dout("missing matching ']'\n");
1862 if (p
< end
&& *p
== ':') {
1865 while (p
< end
&& *p
>= '0' && *p
<= '9') {
1866 port
= (port
* 10) + (*p
- '0');
1870 port
= CEPH_MON_PORT
;
1871 else if (port
> 65535)
1874 port
= CEPH_MON_PORT
;
1877 addr_set_port(ss
, port
);
1879 dout("parse_ips got %s\n", ceph_pr_addr(ss
));
1896 pr_err("parse_ips bad ip '%.*s'\n", (int)(end
- c
), c
);
1899 EXPORT_SYMBOL(ceph_parse_ips
);
1901 static int process_banner(struct ceph_connection
*con
)
1903 dout("process_banner on %p\n", con
);
1905 if (verify_hello(con
) < 0)
1908 ceph_decode_addr(&con
->actual_peer_addr
);
1909 ceph_decode_addr(&con
->peer_addr_for_me
);
1912 * Make sure the other end is who we wanted. note that the other
1913 * end may not yet know their ip address, so if it's 0.0.0.0, give
1914 * them the benefit of the doubt.
1916 if (memcmp(&con
->peer_addr
, &con
->actual_peer_addr
,
1917 sizeof(con
->peer_addr
)) != 0 &&
1918 !(addr_is_blank(&con
->actual_peer_addr
.in_addr
) &&
1919 con
->actual_peer_addr
.nonce
== con
->peer_addr
.nonce
)) {
1920 pr_warning("wrong peer, want %s/%d, got %s/%d\n",
1921 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1922 (int)le32_to_cpu(con
->peer_addr
.nonce
),
1923 ceph_pr_addr(&con
->actual_peer_addr
.in_addr
),
1924 (int)le32_to_cpu(con
->actual_peer_addr
.nonce
));
1925 con
->error_msg
= "wrong peer at address";
1930 * did we learn our address?
1932 if (addr_is_blank(&con
->msgr
->inst
.addr
.in_addr
)) {
1933 int port
= addr_port(&con
->msgr
->inst
.addr
.in_addr
);
1935 memcpy(&con
->msgr
->inst
.addr
.in_addr
,
1936 &con
->peer_addr_for_me
.in_addr
,
1937 sizeof(con
->peer_addr_for_me
.in_addr
));
1938 addr_set_port(&con
->msgr
->inst
.addr
.in_addr
, port
);
1939 encode_my_addr(con
->msgr
);
1940 dout("process_banner learned my addr is %s\n",
1941 ceph_pr_addr(&con
->msgr
->inst
.addr
.in_addr
));
1947 static int process_connect(struct ceph_connection
*con
)
1949 u64 sup_feat
= con
->msgr
->supported_features
;
1950 u64 req_feat
= con
->msgr
->required_features
;
1951 u64 server_feat
= ceph_sanitize_features(
1952 le64_to_cpu(con
->in_reply
.features
));
1955 dout("process_connect on %p tag %d\n", con
, (int)con
->in_tag
);
1957 switch (con
->in_reply
.tag
) {
1958 case CEPH_MSGR_TAG_FEATURES
:
1959 pr_err("%s%lld %s feature set mismatch,"
1960 " my %llx < server's %llx, missing %llx\n",
1961 ENTITY_NAME(con
->peer_name
),
1962 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1963 sup_feat
, server_feat
, server_feat
& ~sup_feat
);
1964 con
->error_msg
= "missing required protocol features";
1965 reset_connection(con
);
1968 case CEPH_MSGR_TAG_BADPROTOVER
:
1969 pr_err("%s%lld %s protocol version mismatch,"
1970 " my %d != server's %d\n",
1971 ENTITY_NAME(con
->peer_name
),
1972 ceph_pr_addr(&con
->peer_addr
.in_addr
),
1973 le32_to_cpu(con
->out_connect
.protocol_version
),
1974 le32_to_cpu(con
->in_reply
.protocol_version
));
1975 con
->error_msg
= "protocol version mismatch";
1976 reset_connection(con
);
1979 case CEPH_MSGR_TAG_BADAUTHORIZER
:
1981 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con
,
1983 if (con
->auth_retry
== 2) {
1984 con
->error_msg
= "connect authorization failure";
1987 con_out_kvec_reset(con
);
1988 ret
= prepare_write_connect(con
);
1991 prepare_read_connect(con
);
1994 case CEPH_MSGR_TAG_RESETSESSION
:
1996 * If we connected with a large connect_seq but the peer
1997 * has no record of a session with us (no connection, or
1998 * connect_seq == 0), they will send RESETSESION to indicate
1999 * that they must have reset their session, and may have
2002 dout("process_connect got RESET peer seq %u\n",
2003 le32_to_cpu(con
->in_reply
.connect_seq
));
2004 pr_err("%s%lld %s connection reset\n",
2005 ENTITY_NAME(con
->peer_name
),
2006 ceph_pr_addr(&con
->peer_addr
.in_addr
));
2007 reset_connection(con
);
2008 con_out_kvec_reset(con
);
2009 ret
= prepare_write_connect(con
);
2012 prepare_read_connect(con
);
2014 /* Tell ceph about it. */
2015 mutex_unlock(&con
->mutex
);
2016 pr_info("reset on %s%lld\n", ENTITY_NAME(con
->peer_name
));
2017 if (con
->ops
->peer_reset
)
2018 con
->ops
->peer_reset(con
);
2019 mutex_lock(&con
->mutex
);
2020 if (con
->state
!= CON_STATE_NEGOTIATING
)
2024 case CEPH_MSGR_TAG_RETRY_SESSION
:
2026 * If we sent a smaller connect_seq than the peer has, try
2027 * again with a larger value.
2029 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2030 le32_to_cpu(con
->out_connect
.connect_seq
),
2031 le32_to_cpu(con
->in_reply
.connect_seq
));
2032 con
->connect_seq
= le32_to_cpu(con
->in_reply
.connect_seq
);
2033 con_out_kvec_reset(con
);
2034 ret
= prepare_write_connect(con
);
2037 prepare_read_connect(con
);
2040 case CEPH_MSGR_TAG_RETRY_GLOBAL
:
2042 * If we sent a smaller global_seq than the peer has, try
2043 * again with a larger value.
2045 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2046 con
->peer_global_seq
,
2047 le32_to_cpu(con
->in_reply
.global_seq
));
2048 get_global_seq(con
->msgr
,
2049 le32_to_cpu(con
->in_reply
.global_seq
));
2050 con_out_kvec_reset(con
);
2051 ret
= prepare_write_connect(con
);
2054 prepare_read_connect(con
);
2057 case CEPH_MSGR_TAG_SEQ
:
2058 case CEPH_MSGR_TAG_READY
:
2059 if (req_feat
& ~server_feat
) {
2060 pr_err("%s%lld %s protocol feature mismatch,"
2061 " my required %llx > server's %llx, need %llx\n",
2062 ENTITY_NAME(con
->peer_name
),
2063 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2064 req_feat
, server_feat
, req_feat
& ~server_feat
);
2065 con
->error_msg
= "missing required protocol features";
2066 reset_connection(con
);
2070 WARN_ON(con
->state
!= CON_STATE_NEGOTIATING
);
2071 con
->state
= CON_STATE_OPEN
;
2072 con
->auth_retry
= 0; /* we authenticated; clear flag */
2073 con
->peer_global_seq
= le32_to_cpu(con
->in_reply
.global_seq
);
2075 con
->peer_features
= server_feat
;
2076 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2077 con
->peer_global_seq
,
2078 le32_to_cpu(con
->in_reply
.connect_seq
),
2080 WARN_ON(con
->connect_seq
!=
2081 le32_to_cpu(con
->in_reply
.connect_seq
));
2083 if (con
->in_reply
.flags
& CEPH_MSG_CONNECT_LOSSY
)
2084 con_flag_set(con
, CON_FLAG_LOSSYTX
);
2086 con
->delay
= 0; /* reset backoff memory */
2088 if (con
->in_reply
.tag
== CEPH_MSGR_TAG_SEQ
) {
2089 prepare_write_seq(con
);
2090 prepare_read_seq(con
);
2092 prepare_read_tag(con
);
2096 case CEPH_MSGR_TAG_WAIT
:
2098 * If there is a connection race (we are opening
2099 * connections to each other), one of us may just have
2100 * to WAIT. This shouldn't happen if we are the
2103 pr_err("process_connect got WAIT as client\n");
2104 con
->error_msg
= "protocol error, got WAIT as client";
2108 pr_err("connect protocol error, will retry\n");
2109 con
->error_msg
= "protocol error, garbage tag during connect";
2117 * read (part of) an ack
2119 static int read_partial_ack(struct ceph_connection
*con
)
2121 int size
= sizeof (con
->in_temp_ack
);
2124 return read_partial(con
, end
, size
, &con
->in_temp_ack
);
2128 * We can finally discard anything that's been acked.
2130 static void process_ack(struct ceph_connection
*con
)
2133 u64 ack
= le64_to_cpu(con
->in_temp_ack
);
2136 while (!list_empty(&con
->out_sent
)) {
2137 m
= list_first_entry(&con
->out_sent
, struct ceph_msg
,
2139 seq
= le64_to_cpu(m
->hdr
.seq
);
2142 dout("got ack for seq %llu type %d at %p\n", seq
,
2143 le16_to_cpu(m
->hdr
.type
), m
);
2144 m
->ack_stamp
= jiffies
;
2147 prepare_read_tag(con
);
2151 static int read_partial_message_section(struct ceph_connection
*con
,
2152 struct kvec
*section
,
2153 unsigned int sec_len
, u32
*crc
)
2159 while (section
->iov_len
< sec_len
) {
2160 BUG_ON(section
->iov_base
== NULL
);
2161 left
= sec_len
- section
->iov_len
;
2162 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)section
->iov_base
+
2163 section
->iov_len
, left
);
2166 section
->iov_len
+= ret
;
2168 if (section
->iov_len
== sec_len
)
2169 *crc
= crc32c(0, section
->iov_base
, section
->iov_len
);
2174 static int read_partial_msg_data(struct ceph_connection
*con
)
2176 struct ceph_msg
*msg
= con
->in_msg
;
2177 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
2178 const bool do_datacrc
= !con
->msgr
->nocrc
;
2186 if (list_empty(&msg
->data
))
2190 crc
= con
->in_data_crc
;
2191 while (cursor
->resid
) {
2192 page
= ceph_msg_data_next(&msg
->cursor
, &page_offset
, &length
,
2194 ret
= ceph_tcp_recvpage(con
->sock
, page
, page_offset
, length
);
2197 con
->in_data_crc
= crc
;
2203 crc
= ceph_crc32c_page(crc
, page
, page_offset
, ret
);
2204 (void) ceph_msg_data_advance(&msg
->cursor
, (size_t)ret
);
2207 con
->in_data_crc
= crc
;
2209 return 1; /* must return > 0 to indicate success */
2213 * read (part of) a message.
2215 static int ceph_con_in_msg_alloc(struct ceph_connection
*con
, int *skip
);
2217 static int read_partial_message(struct ceph_connection
*con
)
2219 struct ceph_msg
*m
= con
->in_msg
;
2223 unsigned int front_len
, middle_len
, data_len
;
2224 bool do_datacrc
= !con
->msgr
->nocrc
;
2228 dout("read_partial_message con %p msg %p\n", con
, m
);
2231 size
= sizeof (con
->in_hdr
);
2233 ret
= read_partial(con
, end
, size
, &con
->in_hdr
);
2237 crc
= crc32c(0, &con
->in_hdr
, offsetof(struct ceph_msg_header
, crc
));
2238 if (cpu_to_le32(crc
) != con
->in_hdr
.crc
) {
2239 pr_err("read_partial_message bad hdr "
2240 " crc %u != expected %u\n",
2241 crc
, con
->in_hdr
.crc
);
2245 front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
2246 if (front_len
> CEPH_MSG_MAX_FRONT_LEN
)
2248 middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
2249 if (middle_len
> CEPH_MSG_MAX_MIDDLE_LEN
)
2251 data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
2252 if (data_len
> CEPH_MSG_MAX_DATA_LEN
)
2256 seq
= le64_to_cpu(con
->in_hdr
.seq
);
2257 if ((s64
)seq
- (s64
)con
->in_seq
< 1) {
2258 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2259 ENTITY_NAME(con
->peer_name
),
2260 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2261 seq
, con
->in_seq
+ 1);
2262 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
2264 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2266 } else if ((s64
)seq
- (s64
)con
->in_seq
> 1) {
2267 pr_err("read_partial_message bad seq %lld expected %lld\n",
2268 seq
, con
->in_seq
+ 1);
2269 con
->error_msg
= "bad message sequence # for incoming message";
2273 /* allocate message? */
2277 dout("got hdr type %d front %d data %d\n", con
->in_hdr
.type
,
2278 front_len
, data_len
);
2279 ret
= ceph_con_in_msg_alloc(con
, &skip
);
2283 BUG_ON(!con
->in_msg
^ skip
);
2284 if (con
->in_msg
&& data_len
> con
->in_msg
->data_length
) {
2285 pr_warning("%s skipping long message (%u > %zd)\n",
2286 __func__
, data_len
, con
->in_msg
->data_length
);
2287 ceph_msg_put(con
->in_msg
);
2292 /* skip this message */
2293 dout("alloc_msg said skip message\n");
2294 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
2296 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2301 BUG_ON(!con
->in_msg
);
2302 BUG_ON(con
->in_msg
->con
!= con
);
2304 m
->front
.iov_len
= 0; /* haven't read it yet */
2306 m
->middle
->vec
.iov_len
= 0;
2308 /* prepare for data payload, if any */
2311 prepare_message_data(con
->in_msg
, data_len
);
2315 ret
= read_partial_message_section(con
, &m
->front
, front_len
,
2316 &con
->in_front_crc
);
2322 ret
= read_partial_message_section(con
, &m
->middle
->vec
,
2324 &con
->in_middle_crc
);
2331 ret
= read_partial_msg_data(con
);
2337 size
= sizeof (m
->footer
);
2339 ret
= read_partial(con
, end
, size
, &m
->footer
);
2343 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2344 m
, front_len
, m
->footer
.front_crc
, middle_len
,
2345 m
->footer
.middle_crc
, data_len
, m
->footer
.data_crc
);
2348 if (con
->in_front_crc
!= le32_to_cpu(m
->footer
.front_crc
)) {
2349 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2350 m
, con
->in_front_crc
, m
->footer
.front_crc
);
2353 if (con
->in_middle_crc
!= le32_to_cpu(m
->footer
.middle_crc
)) {
2354 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2355 m
, con
->in_middle_crc
, m
->footer
.middle_crc
);
2359 (m
->footer
.flags
& CEPH_MSG_FOOTER_NOCRC
) == 0 &&
2360 con
->in_data_crc
!= le32_to_cpu(m
->footer
.data_crc
)) {
2361 pr_err("read_partial_message %p data crc %u != exp. %u\n", m
,
2362 con
->in_data_crc
, le32_to_cpu(m
->footer
.data_crc
));
2366 return 1; /* done! */
2370 * Process message. This happens in the worker thread. The callback should
2371 * be careful not to do anything that waits on other incoming messages or it
2374 static void process_message(struct ceph_connection
*con
)
2376 struct ceph_msg
*msg
;
2378 BUG_ON(con
->in_msg
->con
!= con
);
2379 con
->in_msg
->con
= NULL
;
2384 /* if first message, set peer_name */
2385 if (con
->peer_name
.type
== 0)
2386 con
->peer_name
= msg
->hdr
.src
;
2389 mutex_unlock(&con
->mutex
);
2391 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2392 msg
, le64_to_cpu(msg
->hdr
.seq
),
2393 ENTITY_NAME(msg
->hdr
.src
),
2394 le16_to_cpu(msg
->hdr
.type
),
2395 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2396 le32_to_cpu(msg
->hdr
.front_len
),
2397 le32_to_cpu(msg
->hdr
.data_len
),
2398 con
->in_front_crc
, con
->in_middle_crc
, con
->in_data_crc
);
2399 con
->ops
->dispatch(con
, msg
);
2401 mutex_lock(&con
->mutex
);
2406 * Write something to the socket. Called in a worker thread when the
2407 * socket appears to be writeable and we have something ready to send.
2409 static int try_write(struct ceph_connection
*con
)
2413 dout("try_write start %p state %lu\n", con
, con
->state
);
2416 dout("try_write out_kvec_bytes %d\n", con
->out_kvec_bytes
);
2418 /* open the socket first? */
2419 if (con
->state
== CON_STATE_PREOPEN
) {
2421 con
->state
= CON_STATE_CONNECTING
;
2423 con_out_kvec_reset(con
);
2424 prepare_write_banner(con
);
2425 prepare_read_banner(con
);
2427 BUG_ON(con
->in_msg
);
2428 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2429 dout("try_write initiating connect on %p new state %lu\n",
2431 ret
= ceph_tcp_connect(con
);
2433 con
->error_msg
= "connect error";
2439 /* kvec data queued? */
2440 if (con
->out_skip
) {
2441 ret
= write_partial_skip(con
);
2445 if (con
->out_kvec_left
) {
2446 ret
= write_partial_kvec(con
);
2453 if (con
->out_msg_done
) {
2454 ceph_msg_put(con
->out_msg
);
2455 con
->out_msg
= NULL
; /* we're done with this one */
2459 ret
= write_partial_message_data(con
);
2461 goto more_kvec
; /* we need to send the footer, too! */
2465 dout("try_write write_partial_message_data err %d\n",
2472 if (con
->state
== CON_STATE_OPEN
) {
2473 /* is anything else pending? */
2474 if (!list_empty(&con
->out_queue
)) {
2475 prepare_write_message(con
);
2478 if (con
->in_seq
> con
->in_seq_acked
) {
2479 prepare_write_ack(con
);
2482 if (con_flag_test_and_clear(con
, CON_FLAG_KEEPALIVE_PENDING
)) {
2483 prepare_write_keepalive(con
);
2488 /* Nothing to do! */
2489 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
2490 dout("try_write nothing else to write.\n");
2493 dout("try_write done on %p ret %d\n", con
, ret
);
2500 * Read what we can from the socket.
2502 static int try_read(struct ceph_connection
*con
)
2507 dout("try_read start on %p state %lu\n", con
, con
->state
);
2508 if (con
->state
!= CON_STATE_CONNECTING
&&
2509 con
->state
!= CON_STATE_NEGOTIATING
&&
2510 con
->state
!= CON_STATE_OPEN
)
2515 dout("try_read tag %d in_base_pos %d\n", (int)con
->in_tag
,
2518 if (con
->state
== CON_STATE_CONNECTING
) {
2519 dout("try_read connecting\n");
2520 ret
= read_partial_banner(con
);
2523 ret
= process_banner(con
);
2527 con
->state
= CON_STATE_NEGOTIATING
;
2530 * Received banner is good, exchange connection info.
2531 * Do not reset out_kvec, as sending our banner raced
2532 * with receiving peer banner after connect completed.
2534 ret
= prepare_write_connect(con
);
2537 prepare_read_connect(con
);
2539 /* Send connection info before awaiting response */
2543 if (con
->state
== CON_STATE_NEGOTIATING
) {
2544 dout("try_read negotiating\n");
2545 ret
= read_partial_connect(con
);
2548 ret
= process_connect(con
);
2554 WARN_ON(con
->state
!= CON_STATE_OPEN
);
2556 if (con
->in_base_pos
< 0) {
2558 * skipping + discarding content.
2560 * FIXME: there must be a better way to do this!
2562 static char buf
[SKIP_BUF_SIZE
];
2563 int skip
= min((int) sizeof (buf
), -con
->in_base_pos
);
2565 dout("skipping %d / %d bytes\n", skip
, -con
->in_base_pos
);
2566 ret
= ceph_tcp_recvmsg(con
->sock
, buf
, skip
);
2569 con
->in_base_pos
+= ret
;
2570 if (con
->in_base_pos
)
2573 if (con
->in_tag
== CEPH_MSGR_TAG_READY
) {
2577 ret
= ceph_tcp_recvmsg(con
->sock
, &con
->in_tag
, 1);
2580 dout("try_read got tag %d\n", (int)con
->in_tag
);
2581 switch (con
->in_tag
) {
2582 case CEPH_MSGR_TAG_MSG
:
2583 prepare_read_message(con
);
2585 case CEPH_MSGR_TAG_ACK
:
2586 prepare_read_ack(con
);
2588 case CEPH_MSGR_TAG_CLOSE
:
2589 con_close_socket(con
);
2590 con
->state
= CON_STATE_CLOSED
;
2596 if (con
->in_tag
== CEPH_MSGR_TAG_MSG
) {
2597 ret
= read_partial_message(con
);
2601 con
->error_msg
= "bad crc";
2605 con
->error_msg
= "io error";
2610 if (con
->in_tag
== CEPH_MSGR_TAG_READY
)
2612 process_message(con
);
2613 if (con
->state
== CON_STATE_OPEN
)
2614 prepare_read_tag(con
);
2617 if (con
->in_tag
== CEPH_MSGR_TAG_ACK
||
2618 con
->in_tag
== CEPH_MSGR_TAG_SEQ
) {
2620 * the final handshake seq exchange is semantically
2621 * equivalent to an ACK
2623 ret
= read_partial_ack(con
);
2631 dout("try_read done on %p ret %d\n", con
, ret
);
2635 pr_err("try_read bad con->in_tag = %d\n", (int)con
->in_tag
);
2636 con
->error_msg
= "protocol error, garbage tag";
2643 * Atomically queue work on a connection after the specified delay.
2644 * Bump @con reference to avoid races with connection teardown.
2645 * Returns 0 if work was queued, or an error code otherwise.
2647 static int queue_con_delay(struct ceph_connection
*con
, unsigned long delay
)
2649 if (!con
->ops
->get(con
)) {
2650 dout("%s %p ref count 0\n", __func__
, con
);
2655 if (!queue_delayed_work(ceph_msgr_wq
, &con
->work
, delay
)) {
2656 dout("%s %p - already queued\n", __func__
, con
);
2662 dout("%s %p %lu\n", __func__
, con
, delay
);
2667 static void queue_con(struct ceph_connection
*con
)
2669 (void) queue_con_delay(con
, 0);
2672 static bool con_sock_closed(struct ceph_connection
*con
)
2674 if (!con_flag_test_and_clear(con
, CON_FLAG_SOCK_CLOSED
))
2678 case CON_STATE_ ## x: \
2679 con->error_msg = "socket closed (con state " #x ")"; \
2682 switch (con
->state
) {
2690 pr_warning("%s con %p unrecognized state %lu\n",
2691 __func__
, con
, con
->state
);
2692 con
->error_msg
= "unrecognized con state";
2701 static bool con_backoff(struct ceph_connection
*con
)
2705 if (!con_flag_test_and_clear(con
, CON_FLAG_BACKOFF
))
2708 ret
= queue_con_delay(con
, round_jiffies_relative(con
->delay
));
2710 dout("%s: con %p FAILED to back off %lu\n", __func__
,
2712 BUG_ON(ret
== -ENOENT
);
2713 con_flag_set(con
, CON_FLAG_BACKOFF
);
2719 /* Finish fault handling; con->mutex must *not* be held here */
2721 static void con_fault_finish(struct ceph_connection
*con
)
2724 * in case we faulted due to authentication, invalidate our
2725 * current tickets so that we can get new ones.
2727 if (con
->auth_retry
&& con
->ops
->invalidate_authorizer
) {
2728 dout("calling invalidate_authorizer()\n");
2729 con
->ops
->invalidate_authorizer(con
);
2732 if (con
->ops
->fault
)
2733 con
->ops
->fault(con
);
2737 * Do some work on a connection. Drop a connection ref when we're done.
2739 static void con_work(struct work_struct
*work
)
2741 struct ceph_connection
*con
= container_of(work
, struct ceph_connection
,
2745 mutex_lock(&con
->mutex
);
2749 if ((fault
= con_sock_closed(con
))) {
2750 dout("%s: con %p SOCK_CLOSED\n", __func__
, con
);
2753 if (con_backoff(con
)) {
2754 dout("%s: con %p BACKOFF\n", __func__
, con
);
2757 if (con
->state
== CON_STATE_STANDBY
) {
2758 dout("%s: con %p STANDBY\n", __func__
, con
);
2761 if (con
->state
== CON_STATE_CLOSED
) {
2762 dout("%s: con %p CLOSED\n", __func__
, con
);
2766 if (con
->state
== CON_STATE_PREOPEN
) {
2767 dout("%s: con %p PREOPEN\n", __func__
, con
);
2771 ret
= try_read(con
);
2775 con
->error_msg
= "socket error on read";
2780 ret
= try_write(con
);
2784 con
->error_msg
= "socket error on write";
2788 break; /* If we make it to here, we're done */
2792 mutex_unlock(&con
->mutex
);
2795 con_fault_finish(con
);
2801 * Generic error/fault handler. A retry mechanism is used with
2802 * exponential backoff
2804 static void con_fault(struct ceph_connection
*con
)
2806 pr_warning("%s%lld %s %s\n", ENTITY_NAME(con
->peer_name
),
2807 ceph_pr_addr(&con
->peer_addr
.in_addr
), con
->error_msg
);
2808 dout("fault %p state %lu to peer %s\n",
2809 con
, con
->state
, ceph_pr_addr(&con
->peer_addr
.in_addr
));
2811 WARN_ON(con
->state
!= CON_STATE_CONNECTING
&&
2812 con
->state
!= CON_STATE_NEGOTIATING
&&
2813 con
->state
!= CON_STATE_OPEN
);
2815 con_close_socket(con
);
2817 if (con_flag_test(con
, CON_FLAG_LOSSYTX
)) {
2818 dout("fault on LOSSYTX channel, marking CLOSED\n");
2819 con
->state
= CON_STATE_CLOSED
;
2824 BUG_ON(con
->in_msg
->con
!= con
);
2825 con
->in_msg
->con
= NULL
;
2826 ceph_msg_put(con
->in_msg
);
2831 /* Requeue anything that hasn't been acked */
2832 list_splice_init(&con
->out_sent
, &con
->out_queue
);
2834 /* If there are no messages queued or keepalive pending, place
2835 * the connection in a STANDBY state */
2836 if (list_empty(&con
->out_queue
) &&
2837 !con_flag_test(con
, CON_FLAG_KEEPALIVE_PENDING
)) {
2838 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con
);
2839 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
2840 con
->state
= CON_STATE_STANDBY
;
2842 /* retry after a delay. */
2843 con
->state
= CON_STATE_PREOPEN
;
2844 if (con
->delay
== 0)
2845 con
->delay
= BASE_DELAY_INTERVAL
;
2846 else if (con
->delay
< MAX_DELAY_INTERVAL
)
2848 con_flag_set(con
, CON_FLAG_BACKOFF
);
2856 * initialize a new messenger instance
2858 void ceph_messenger_init(struct ceph_messenger
*msgr
,
2859 struct ceph_entity_addr
*myaddr
,
2860 u64 supported_features
,
2861 u64 required_features
,
2864 msgr
->supported_features
= supported_features
;
2865 msgr
->required_features
= required_features
;
2867 spin_lock_init(&msgr
->global_seq_lock
);
2870 msgr
->inst
.addr
= *myaddr
;
2872 /* select a random nonce */
2873 msgr
->inst
.addr
.type
= 0;
2874 get_random_bytes(&msgr
->inst
.addr
.nonce
, sizeof(msgr
->inst
.addr
.nonce
));
2875 encode_my_addr(msgr
);
2876 msgr
->nocrc
= nocrc
;
2878 atomic_set(&msgr
->stopping
, 0);
2880 dout("%s %p\n", __func__
, msgr
);
2882 EXPORT_SYMBOL(ceph_messenger_init
);
2884 static void clear_standby(struct ceph_connection
*con
)
2886 /* come back from STANDBY? */
2887 if (con
->state
== CON_STATE_STANDBY
) {
2888 dout("clear_standby %p and ++connect_seq\n", con
);
2889 con
->state
= CON_STATE_PREOPEN
;
2891 WARN_ON(con_flag_test(con
, CON_FLAG_WRITE_PENDING
));
2892 WARN_ON(con_flag_test(con
, CON_FLAG_KEEPALIVE_PENDING
));
2897 * Queue up an outgoing message on the given connection.
2899 void ceph_con_send(struct ceph_connection
*con
, struct ceph_msg
*msg
)
2902 msg
->hdr
.src
= con
->msgr
->inst
.name
;
2903 BUG_ON(msg
->front
.iov_len
!= le32_to_cpu(msg
->hdr
.front_len
));
2904 msg
->needs_out_seq
= true;
2906 mutex_lock(&con
->mutex
);
2908 if (con
->state
== CON_STATE_CLOSED
) {
2909 dout("con_send %p closed, dropping %p\n", con
, msg
);
2911 mutex_unlock(&con
->mutex
);
2915 BUG_ON(msg
->con
!= NULL
);
2916 msg
->con
= con
->ops
->get(con
);
2917 BUG_ON(msg
->con
== NULL
);
2919 BUG_ON(!list_empty(&msg
->list_head
));
2920 list_add_tail(&msg
->list_head
, &con
->out_queue
);
2921 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg
,
2922 ENTITY_NAME(con
->peer_name
), le16_to_cpu(msg
->hdr
.type
),
2923 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2924 le32_to_cpu(msg
->hdr
.front_len
),
2925 le32_to_cpu(msg
->hdr
.middle_len
),
2926 le32_to_cpu(msg
->hdr
.data_len
));
2929 mutex_unlock(&con
->mutex
);
2931 /* if there wasn't anything waiting to send before, queue
2933 if (con_flag_test_and_set(con
, CON_FLAG_WRITE_PENDING
) == 0)
2936 EXPORT_SYMBOL(ceph_con_send
);
2939 * Revoke a message that was previously queued for send
2941 void ceph_msg_revoke(struct ceph_msg
*msg
)
2943 struct ceph_connection
*con
= msg
->con
;
2946 return; /* Message not in our possession */
2948 mutex_lock(&con
->mutex
);
2949 if (!list_empty(&msg
->list_head
)) {
2950 dout("%s %p msg %p - was on queue\n", __func__
, con
, msg
);
2951 list_del_init(&msg
->list_head
);
2952 BUG_ON(msg
->con
== NULL
);
2953 msg
->con
->ops
->put(msg
->con
);
2959 if (con
->out_msg
== msg
) {
2960 dout("%s %p msg %p - was sending\n", __func__
, con
, msg
);
2961 con
->out_msg
= NULL
;
2962 if (con
->out_kvec_is_msg
) {
2963 con
->out_skip
= con
->out_kvec_bytes
;
2964 con
->out_kvec_is_msg
= false;
2970 mutex_unlock(&con
->mutex
);
2974 * Revoke a message that we may be reading data into
2976 void ceph_msg_revoke_incoming(struct ceph_msg
*msg
)
2978 struct ceph_connection
*con
;
2980 BUG_ON(msg
== NULL
);
2982 dout("%s msg %p null con\n", __func__
, msg
);
2984 return; /* Message not in our possession */
2988 mutex_lock(&con
->mutex
);
2989 if (con
->in_msg
== msg
) {
2990 unsigned int front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
2991 unsigned int middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
2992 unsigned int data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
2994 /* skip rest of message */
2995 dout("%s %p msg %p revoked\n", __func__
, con
, msg
);
2996 con
->in_base_pos
= con
->in_base_pos
-
2997 sizeof(struct ceph_msg_header
) -
3001 sizeof(struct ceph_msg_footer
);
3002 ceph_msg_put(con
->in_msg
);
3004 con
->in_tag
= CEPH_MSGR_TAG_READY
;
3007 dout("%s %p in_msg %p msg %p no-op\n",
3008 __func__
, con
, con
->in_msg
, msg
);
3010 mutex_unlock(&con
->mutex
);
3014 * Queue a keepalive byte to ensure the tcp connection is alive.
3016 void ceph_con_keepalive(struct ceph_connection
*con
)
3018 dout("con_keepalive %p\n", con
);
3019 mutex_lock(&con
->mutex
);
3021 mutex_unlock(&con
->mutex
);
3022 if (con_flag_test_and_set(con
, CON_FLAG_KEEPALIVE_PENDING
) == 0 &&
3023 con_flag_test_and_set(con
, CON_FLAG_WRITE_PENDING
) == 0)
3026 EXPORT_SYMBOL(ceph_con_keepalive
);
3028 static struct ceph_msg_data
*ceph_msg_data_create(enum ceph_msg_data_type type
)
3030 struct ceph_msg_data
*data
;
3032 if (WARN_ON(!ceph_msg_data_type_valid(type
)))
3035 data
= kmem_cache_zalloc(ceph_msg_data_cache
, GFP_NOFS
);
3038 INIT_LIST_HEAD(&data
->links
);
3043 static void ceph_msg_data_destroy(struct ceph_msg_data
*data
)
3048 WARN_ON(!list_empty(&data
->links
));
3049 if (data
->type
== CEPH_MSG_DATA_PAGELIST
) {
3050 ceph_pagelist_release(data
->pagelist
);
3051 kfree(data
->pagelist
);
3053 kmem_cache_free(ceph_msg_data_cache
, data
);
3056 void ceph_msg_data_add_pages(struct ceph_msg
*msg
, struct page
**pages
,
3057 size_t length
, size_t alignment
)
3059 struct ceph_msg_data
*data
;
3064 data
= ceph_msg_data_create(CEPH_MSG_DATA_PAGES
);
3066 data
->pages
= pages
;
3067 data
->length
= length
;
3068 data
->alignment
= alignment
& ~PAGE_MASK
;
3070 list_add_tail(&data
->links
, &msg
->data
);
3071 msg
->data_length
+= length
;
3073 EXPORT_SYMBOL(ceph_msg_data_add_pages
);
3075 void ceph_msg_data_add_pagelist(struct ceph_msg
*msg
,
3076 struct ceph_pagelist
*pagelist
)
3078 struct ceph_msg_data
*data
;
3081 BUG_ON(!pagelist
->length
);
3083 data
= ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST
);
3085 data
->pagelist
= pagelist
;
3087 list_add_tail(&data
->links
, &msg
->data
);
3088 msg
->data_length
+= pagelist
->length
;
3090 EXPORT_SYMBOL(ceph_msg_data_add_pagelist
);
3093 void ceph_msg_data_add_bio(struct ceph_msg
*msg
, struct bio
*bio
,
3096 struct ceph_msg_data
*data
;
3100 data
= ceph_msg_data_create(CEPH_MSG_DATA_BIO
);
3103 data
->bio_length
= length
;
3105 list_add_tail(&data
->links
, &msg
->data
);
3106 msg
->data_length
+= length
;
3108 EXPORT_SYMBOL(ceph_msg_data_add_bio
);
3109 #endif /* CONFIG_BLOCK */
3112 * construct a new message with given type, size
3113 * the new msg has a ref count of 1.
3115 struct ceph_msg
*ceph_msg_new(int type
, int front_len
, gfp_t flags
,
3120 m
= kmem_cache_zalloc(ceph_msg_cache
, flags
);
3124 m
->hdr
.type
= cpu_to_le16(type
);
3125 m
->hdr
.priority
= cpu_to_le16(CEPH_MSG_PRIO_DEFAULT
);
3126 m
->hdr
.front_len
= cpu_to_le32(front_len
);
3128 INIT_LIST_HEAD(&m
->list_head
);
3129 kref_init(&m
->kref
);
3130 INIT_LIST_HEAD(&m
->data
);
3133 m
->front_alloc_len
= front_len
;
3135 if (front_len
> PAGE_CACHE_SIZE
) {
3136 m
->front
.iov_base
= __vmalloc(front_len
, flags
,
3138 m
->front_is_vmalloc
= true;
3140 m
->front
.iov_base
= kmalloc(front_len
, flags
);
3142 if (m
->front
.iov_base
== NULL
) {
3143 dout("ceph_msg_new can't allocate %d bytes\n",
3148 m
->front
.iov_base
= NULL
;
3150 m
->front
.iov_len
= front_len
;
3152 dout("ceph_msg_new %p front %d\n", m
, front_len
);
3159 pr_err("msg_new can't create type %d front %d\n", type
,
3163 dout("msg_new can't create type %d front %d\n", type
,
3168 EXPORT_SYMBOL(ceph_msg_new
);
3171 * Allocate "middle" portion of a message, if it is needed and wasn't
3172 * allocated by alloc_msg. This allows us to read a small fixed-size
3173 * per-type header in the front and then gracefully fail (i.e.,
3174 * propagate the error to the caller based on info in the front) when
3175 * the middle is too large.
3177 static int ceph_alloc_middle(struct ceph_connection
*con
, struct ceph_msg
*msg
)
3179 int type
= le16_to_cpu(msg
->hdr
.type
);
3180 int middle_len
= le32_to_cpu(msg
->hdr
.middle_len
);
3182 dout("alloc_middle %p type %d %s middle_len %d\n", msg
, type
,
3183 ceph_msg_type_name(type
), middle_len
);
3184 BUG_ON(!middle_len
);
3185 BUG_ON(msg
->middle
);
3187 msg
->middle
= ceph_buffer_new(middle_len
, GFP_NOFS
);
3194 * Allocate a message for receiving an incoming message on a
3195 * connection, and save the result in con->in_msg. Uses the
3196 * connection's private alloc_msg op if available.
3198 * Returns 0 on success, or a negative error code.
3200 * On success, if we set *skip = 1:
3201 * - the next message should be skipped and ignored.
3202 * - con->in_msg == NULL
3203 * or if we set *skip = 0:
3204 * - con->in_msg is non-null.
3205 * On error (ENOMEM, EAGAIN, ...),
3206 * - con->in_msg == NULL
3208 static int ceph_con_in_msg_alloc(struct ceph_connection
*con
, int *skip
)
3210 struct ceph_msg_header
*hdr
= &con
->in_hdr
;
3211 int middle_len
= le32_to_cpu(hdr
->middle_len
);
3212 struct ceph_msg
*msg
;
3215 BUG_ON(con
->in_msg
!= NULL
);
3216 BUG_ON(!con
->ops
->alloc_msg
);
3218 mutex_unlock(&con
->mutex
);
3219 msg
= con
->ops
->alloc_msg(con
, hdr
, skip
);
3220 mutex_lock(&con
->mutex
);
3221 if (con
->state
!= CON_STATE_OPEN
) {
3229 con
->in_msg
->con
= con
->ops
->get(con
);
3230 BUG_ON(con
->in_msg
->con
== NULL
);
3233 * Null message pointer means either we should skip
3234 * this message or we couldn't allocate memory. The
3235 * former is not an error.
3239 con
->error_msg
= "error allocating memory for incoming message";
3243 memcpy(&con
->in_msg
->hdr
, &con
->in_hdr
, sizeof(con
->in_hdr
));
3245 if (middle_len
&& !con
->in_msg
->middle
) {
3246 ret
= ceph_alloc_middle(con
, con
->in_msg
);
3248 ceph_msg_put(con
->in_msg
);
3258 * Free a generically kmalloc'd message.
3260 void ceph_msg_kfree(struct ceph_msg
*m
)
3262 dout("msg_kfree %p\n", m
);
3263 if (m
->front_is_vmalloc
)
3264 vfree(m
->front
.iov_base
);
3266 kfree(m
->front
.iov_base
);
3267 kmem_cache_free(ceph_msg_cache
, m
);
3271 * Drop a msg ref. Destroy as needed.
3273 void ceph_msg_last_put(struct kref
*kref
)
3275 struct ceph_msg
*m
= container_of(kref
, struct ceph_msg
, kref
);
3277 struct list_head
*links
;
3278 struct list_head
*next
;
3280 dout("ceph_msg_put last one on %p\n", m
);
3281 WARN_ON(!list_empty(&m
->list_head
));
3283 /* drop middle, data, if any */
3285 ceph_buffer_put(m
->middle
);
3289 list_splice_init(&m
->data
, &data
);
3290 list_for_each_safe(links
, next
, &data
) {
3291 struct ceph_msg_data
*data
;
3293 data
= list_entry(links
, struct ceph_msg_data
, links
);
3294 list_del_init(links
);
3295 ceph_msg_data_destroy(data
);
3300 ceph_msgpool_put(m
->pool
, m
);
3304 EXPORT_SYMBOL(ceph_msg_last_put
);
3306 void ceph_msg_dump(struct ceph_msg
*msg
)
3308 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg
,
3309 msg
->front_alloc_len
, msg
->data_length
);
3310 print_hex_dump(KERN_DEBUG
, "header: ",
3311 DUMP_PREFIX_OFFSET
, 16, 1,
3312 &msg
->hdr
, sizeof(msg
->hdr
), true);
3313 print_hex_dump(KERN_DEBUG
, " front: ",
3314 DUMP_PREFIX_OFFSET
, 16, 1,
3315 msg
->front
.iov_base
, msg
->front
.iov_len
, true);
3317 print_hex_dump(KERN_DEBUG
, "middle: ",
3318 DUMP_PREFIX_OFFSET
, 16, 1,
3319 msg
->middle
->vec
.iov_base
,
3320 msg
->middle
->vec
.iov_len
, true);
3321 print_hex_dump(KERN_DEBUG
, "footer: ",
3322 DUMP_PREFIX_OFFSET
, 16, 1,
3323 &msg
->footer
, sizeof(msg
->footer
), true);
3325 EXPORT_SYMBOL(ceph_msg_dump
);