1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
4 #include <linux/crc32c.h>
5 #include <linux/ctype.h>
6 #include <linux/highmem.h>
7 #include <linux/inet.h>
8 #include <linux/kthread.h>
10 #include <linux/nsproxy.h>
11 #include <linux/sched/mm.h>
12 #include <linux/slab.h>
13 #include <linux/socket.h>
14 #include <linux/string.h>
16 #include <linux/bio.h>
17 #endif /* CONFIG_BLOCK */
18 #include <linux/dns_resolver.h>
21 #include <linux/ceph/ceph_features.h>
22 #include <linux/ceph/libceph.h>
23 #include <linux/ceph/messenger.h>
24 #include <linux/ceph/decode.h>
25 #include <linux/ceph/pagelist.h>
26 #include <linux/export.h>
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
;
165 static char tag_keepalive2
= CEPH_MSGR_TAG_KEEPALIVE2
;
167 #ifdef CONFIG_LOCKDEP
168 static struct lock_class_key socket_class
;
172 * When skipping (ignoring) a block of input we read it into a "skip
173 * buffer," which is this many bytes in size.
175 #define SKIP_BUF_SIZE 1024
177 static void queue_con(struct ceph_connection
*con
);
178 static void cancel_con(struct ceph_connection
*con
);
179 static void ceph_con_workfn(struct work_struct
*);
180 static void con_fault(struct ceph_connection
*con
);
183 * Nicely render a sockaddr as a string. An array of formatted
184 * strings is used, to approximate reentrancy.
186 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
187 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
188 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
189 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
191 static char addr_str
[ADDR_STR_COUNT
][MAX_ADDR_STR_LEN
];
192 static atomic_t addr_str_seq
= ATOMIC_INIT(0);
194 static struct page
*zero_page
; /* used in certain error cases */
196 const char *ceph_pr_addr(const struct sockaddr_storage
*ss
)
200 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
201 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
203 i
= atomic_inc_return(&addr_str_seq
) & ADDR_STR_COUNT_MASK
;
206 switch (ss
->ss_family
) {
208 snprintf(s
, MAX_ADDR_STR_LEN
, "%pI4:%hu", &in4
->sin_addr
,
209 ntohs(in4
->sin_port
));
213 snprintf(s
, MAX_ADDR_STR_LEN
, "[%pI6c]:%hu", &in6
->sin6_addr
,
214 ntohs(in6
->sin6_port
));
218 snprintf(s
, MAX_ADDR_STR_LEN
, "(unknown sockaddr family %hu)",
224 EXPORT_SYMBOL(ceph_pr_addr
);
226 static void encode_my_addr(struct ceph_messenger
*msgr
)
228 memcpy(&msgr
->my_enc_addr
, &msgr
->inst
.addr
, sizeof(msgr
->my_enc_addr
));
229 ceph_encode_addr(&msgr
->my_enc_addr
);
233 * work queue for all reading and writing to/from the socket.
235 static struct workqueue_struct
*ceph_msgr_wq
;
237 static int ceph_msgr_slab_init(void)
239 BUG_ON(ceph_msg_cache
);
240 ceph_msg_cache
= KMEM_CACHE(ceph_msg
, 0);
244 BUG_ON(ceph_msg_data_cache
);
245 ceph_msg_data_cache
= KMEM_CACHE(ceph_msg_data
, 0);
246 if (ceph_msg_data_cache
)
249 kmem_cache_destroy(ceph_msg_cache
);
250 ceph_msg_cache
= NULL
;
255 static void ceph_msgr_slab_exit(void)
257 BUG_ON(!ceph_msg_data_cache
);
258 kmem_cache_destroy(ceph_msg_data_cache
);
259 ceph_msg_data_cache
= NULL
;
261 BUG_ON(!ceph_msg_cache
);
262 kmem_cache_destroy(ceph_msg_cache
);
263 ceph_msg_cache
= NULL
;
266 static void _ceph_msgr_exit(void)
269 destroy_workqueue(ceph_msgr_wq
);
273 BUG_ON(zero_page
== NULL
);
277 ceph_msgr_slab_exit();
280 int ceph_msgr_init(void)
282 if (ceph_msgr_slab_init())
285 BUG_ON(zero_page
!= NULL
);
286 zero_page
= ZERO_PAGE(0);
290 * The number of active work items is limited by the number of
291 * connections, so leave @max_active at default.
293 ceph_msgr_wq
= alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM
, 0);
297 pr_err("msgr_init failed to create workqueue\n");
302 EXPORT_SYMBOL(ceph_msgr_init
);
304 void ceph_msgr_exit(void)
306 BUG_ON(ceph_msgr_wq
== NULL
);
310 EXPORT_SYMBOL(ceph_msgr_exit
);
312 void ceph_msgr_flush(void)
314 flush_workqueue(ceph_msgr_wq
);
316 EXPORT_SYMBOL(ceph_msgr_flush
);
318 /* Connection socket state transition functions */
320 static void con_sock_state_init(struct ceph_connection
*con
)
324 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSED
);
325 if (WARN_ON(old_state
!= CON_SOCK_STATE_NEW
))
326 printk("%s: unexpected old state %d\n", __func__
, old_state
);
327 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
328 CON_SOCK_STATE_CLOSED
);
331 static void con_sock_state_connecting(struct ceph_connection
*con
)
335 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CONNECTING
);
336 if (WARN_ON(old_state
!= CON_SOCK_STATE_CLOSED
))
337 printk("%s: unexpected old state %d\n", __func__
, old_state
);
338 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
339 CON_SOCK_STATE_CONNECTING
);
342 static void con_sock_state_connected(struct ceph_connection
*con
)
346 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CONNECTED
);
347 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTING
))
348 printk("%s: unexpected old state %d\n", __func__
, old_state
);
349 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
350 CON_SOCK_STATE_CONNECTED
);
353 static void con_sock_state_closing(struct ceph_connection
*con
)
357 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSING
);
358 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTING
&&
359 old_state
!= CON_SOCK_STATE_CONNECTED
&&
360 old_state
!= CON_SOCK_STATE_CLOSING
))
361 printk("%s: unexpected old state %d\n", __func__
, old_state
);
362 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
363 CON_SOCK_STATE_CLOSING
);
366 static void con_sock_state_closed(struct ceph_connection
*con
)
370 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSED
);
371 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTED
&&
372 old_state
!= CON_SOCK_STATE_CLOSING
&&
373 old_state
!= CON_SOCK_STATE_CONNECTING
&&
374 old_state
!= CON_SOCK_STATE_CLOSED
))
375 printk("%s: unexpected old state %d\n", __func__
, old_state
);
376 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
377 CON_SOCK_STATE_CLOSED
);
381 * socket callback functions
384 /* data available on socket, or listen socket received a connect */
385 static void ceph_sock_data_ready(struct sock
*sk
)
387 struct ceph_connection
*con
= sk
->sk_user_data
;
388 if (atomic_read(&con
->msgr
->stopping
)) {
392 if (sk
->sk_state
!= TCP_CLOSE_WAIT
) {
393 dout("%s on %p state = %lu, queueing work\n", __func__
,
399 /* socket has buffer space for writing */
400 static void ceph_sock_write_space(struct sock
*sk
)
402 struct ceph_connection
*con
= sk
->sk_user_data
;
404 /* only queue to workqueue if there is data we want to write,
405 * and there is sufficient space in the socket buffer to accept
406 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
407 * doesn't get called again until try_write() fills the socket
408 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
409 * and net/core/stream.c:sk_stream_write_space().
411 if (con_flag_test(con
, CON_FLAG_WRITE_PENDING
)) {
412 if (sk_stream_is_writeable(sk
)) {
413 dout("%s %p queueing write work\n", __func__
, con
);
414 clear_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
418 dout("%s %p nothing to write\n", __func__
, con
);
422 /* socket's state has changed */
423 static void ceph_sock_state_change(struct sock
*sk
)
425 struct ceph_connection
*con
= sk
->sk_user_data
;
427 dout("%s %p state = %lu sk_state = %u\n", __func__
,
428 con
, con
->state
, sk
->sk_state
);
430 switch (sk
->sk_state
) {
432 dout("%s TCP_CLOSE\n", __func__
);
434 dout("%s TCP_CLOSE_WAIT\n", __func__
);
435 con_sock_state_closing(con
);
436 con_flag_set(con
, CON_FLAG_SOCK_CLOSED
);
439 case TCP_ESTABLISHED
:
440 dout("%s TCP_ESTABLISHED\n", __func__
);
441 con_sock_state_connected(con
);
444 default: /* Everything else is uninteresting */
450 * set up socket callbacks
452 static void set_sock_callbacks(struct socket
*sock
,
453 struct ceph_connection
*con
)
455 struct sock
*sk
= sock
->sk
;
456 sk
->sk_user_data
= con
;
457 sk
->sk_data_ready
= ceph_sock_data_ready
;
458 sk
->sk_write_space
= ceph_sock_write_space
;
459 sk
->sk_state_change
= ceph_sock_state_change
;
468 * initiate connection to a remote socket.
470 static int ceph_tcp_connect(struct ceph_connection
*con
)
472 struct sockaddr_storage
*paddr
= &con
->peer_addr
.in_addr
;
474 unsigned int noio_flag
;
479 /* sock_create_kern() allocates with GFP_KERNEL */
480 noio_flag
= memalloc_noio_save();
481 ret
= sock_create_kern(read_pnet(&con
->msgr
->net
), paddr
->ss_family
,
482 SOCK_STREAM
, IPPROTO_TCP
, &sock
);
483 memalloc_noio_restore(noio_flag
);
486 sock
->sk
->sk_allocation
= GFP_NOFS
;
488 #ifdef CONFIG_LOCKDEP
489 lockdep_set_class(&sock
->sk
->sk_lock
, &socket_class
);
492 set_sock_callbacks(sock
, con
);
494 dout("connect %s\n", ceph_pr_addr(&con
->peer_addr
.in_addr
));
496 con_sock_state_connecting(con
);
497 ret
= sock
->ops
->connect(sock
, (struct sockaddr
*)paddr
, sizeof(*paddr
),
499 if (ret
== -EINPROGRESS
) {
500 dout("connect %s EINPROGRESS sk_state = %u\n",
501 ceph_pr_addr(&con
->peer_addr
.in_addr
),
503 } else if (ret
< 0) {
504 pr_err("connect %s error %d\n",
505 ceph_pr_addr(&con
->peer_addr
.in_addr
), ret
);
510 if (ceph_test_opt(from_msgr(con
->msgr
), TCP_NODELAY
)) {
513 ret
= kernel_setsockopt(sock
, SOL_TCP
, TCP_NODELAY
,
514 (char *)&optval
, sizeof(optval
));
516 pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
524 static int ceph_tcp_recvmsg(struct socket
*sock
, void *buf
, size_t len
)
526 struct kvec iov
= {buf
, len
};
527 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
530 iov_iter_kvec(&msg
.msg_iter
, READ
| ITER_KVEC
, &iov
, 1, len
);
531 r
= sock_recvmsg(sock
, &msg
, msg
.msg_flags
);
537 static int ceph_tcp_recvpage(struct socket
*sock
, struct page
*page
,
538 int page_offset
, size_t length
)
540 struct bio_vec bvec
= {
542 .bv_offset
= page_offset
,
545 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
548 BUG_ON(page_offset
+ length
> PAGE_SIZE
);
549 iov_iter_bvec(&msg
.msg_iter
, READ
| ITER_BVEC
, &bvec
, 1, length
);
550 r
= sock_recvmsg(sock
, &msg
, msg
.msg_flags
);
557 * write something. @more is true if caller will be sending more data
560 static int ceph_tcp_sendmsg(struct socket
*sock
, struct kvec
*iov
,
561 size_t kvlen
, size_t len
, int more
)
563 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
567 msg
.msg_flags
|= MSG_MORE
;
569 msg
.msg_flags
|= MSG_EOR
; /* superfluous, but what the hell */
571 r
= kernel_sendmsg(sock
, &msg
, iov
, kvlen
, len
);
577 static int __ceph_tcp_sendpage(struct socket
*sock
, struct page
*page
,
578 int offset
, size_t size
, bool more
)
580 int flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
| (more
? MSG_MORE
: MSG_EOR
);
583 ret
= kernel_sendpage(sock
, page
, offset
, size
, flags
);
590 static int ceph_tcp_sendpage(struct socket
*sock
, struct page
*page
,
591 int offset
, size_t size
, bool more
)
593 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
597 /* sendpage cannot properly handle pages with page_count == 0,
598 * we need to fallback to sendmsg if that's the case */
599 if (page_count(page
) >= 1)
600 return __ceph_tcp_sendpage(sock
, page
, offset
, size
, more
);
603 bvec
.bv_offset
= offset
;
607 msg
.msg_flags
|= MSG_MORE
;
609 msg
.msg_flags
|= MSG_EOR
; /* superfluous, but what the hell */
611 iov_iter_bvec(&msg
.msg_iter
, WRITE
| ITER_BVEC
, &bvec
, 1, size
);
612 ret
= sock_sendmsg(sock
, &msg
);
620 * Shutdown/close the socket for the given connection.
622 static int con_close_socket(struct ceph_connection
*con
)
626 dout("con_close_socket on %p sock %p\n", con
, con
->sock
);
628 rc
= con
->sock
->ops
->shutdown(con
->sock
, SHUT_RDWR
);
629 sock_release(con
->sock
);
634 * Forcibly clear the SOCK_CLOSED flag. It gets set
635 * independent of the connection mutex, and we could have
636 * received a socket close event before we had the chance to
637 * shut the socket down.
639 con_flag_clear(con
, CON_FLAG_SOCK_CLOSED
);
641 con_sock_state_closed(con
);
646 * Reset a connection. Discard all incoming and outgoing messages
647 * and clear *_seq state.
649 static void ceph_msg_remove(struct ceph_msg
*msg
)
651 list_del_init(&msg
->list_head
);
655 static void ceph_msg_remove_list(struct list_head
*head
)
657 while (!list_empty(head
)) {
658 struct ceph_msg
*msg
= list_first_entry(head
, struct ceph_msg
,
660 ceph_msg_remove(msg
);
664 static void reset_connection(struct ceph_connection
*con
)
666 /* reset connection, out_queue, msg_ and connect_seq */
667 /* discard existing out_queue and msg_seq */
668 dout("reset_connection %p\n", con
);
669 ceph_msg_remove_list(&con
->out_queue
);
670 ceph_msg_remove_list(&con
->out_sent
);
673 BUG_ON(con
->in_msg
->con
!= con
);
674 ceph_msg_put(con
->in_msg
);
678 con
->connect_seq
= 0;
681 BUG_ON(con
->out_msg
->con
!= con
);
682 ceph_msg_put(con
->out_msg
);
686 con
->in_seq_acked
= 0;
692 * mark a peer down. drop any open connections.
694 void ceph_con_close(struct ceph_connection
*con
)
696 mutex_lock(&con
->mutex
);
697 dout("con_close %p peer %s\n", con
,
698 ceph_pr_addr(&con
->peer_addr
.in_addr
));
699 con
->state
= CON_STATE_CLOSED
;
701 con_flag_clear(con
, CON_FLAG_LOSSYTX
); /* so we retry next connect */
702 con_flag_clear(con
, CON_FLAG_KEEPALIVE_PENDING
);
703 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
704 con_flag_clear(con
, CON_FLAG_BACKOFF
);
706 reset_connection(con
);
707 con
->peer_global_seq
= 0;
709 con_close_socket(con
);
710 mutex_unlock(&con
->mutex
);
712 EXPORT_SYMBOL(ceph_con_close
);
715 * Reopen a closed connection, with a new peer address.
717 void ceph_con_open(struct ceph_connection
*con
,
718 __u8 entity_type
, __u64 entity_num
,
719 struct ceph_entity_addr
*addr
)
721 mutex_lock(&con
->mutex
);
722 dout("con_open %p %s\n", con
, ceph_pr_addr(&addr
->in_addr
));
724 WARN_ON(con
->state
!= CON_STATE_CLOSED
);
725 con
->state
= CON_STATE_PREOPEN
;
727 con
->peer_name
.type
= (__u8
) entity_type
;
728 con
->peer_name
.num
= cpu_to_le64(entity_num
);
730 memcpy(&con
->peer_addr
, addr
, sizeof(*addr
));
731 con
->delay
= 0; /* reset backoff memory */
732 mutex_unlock(&con
->mutex
);
735 EXPORT_SYMBOL(ceph_con_open
);
738 * return true if this connection ever successfully opened
740 bool ceph_con_opened(struct ceph_connection
*con
)
742 return con
->connect_seq
> 0;
746 * initialize a new connection.
748 void ceph_con_init(struct ceph_connection
*con
, void *private,
749 const struct ceph_connection_operations
*ops
,
750 struct ceph_messenger
*msgr
)
752 dout("con_init %p\n", con
);
753 memset(con
, 0, sizeof(*con
));
754 con
->private = private;
758 con_sock_state_init(con
);
760 mutex_init(&con
->mutex
);
761 INIT_LIST_HEAD(&con
->out_queue
);
762 INIT_LIST_HEAD(&con
->out_sent
);
763 INIT_DELAYED_WORK(&con
->work
, ceph_con_workfn
);
765 con
->state
= CON_STATE_CLOSED
;
767 EXPORT_SYMBOL(ceph_con_init
);
771 * We maintain a global counter to order connection attempts. Get
772 * a unique seq greater than @gt.
774 static u32
get_global_seq(struct ceph_messenger
*msgr
, u32 gt
)
778 spin_lock(&msgr
->global_seq_lock
);
779 if (msgr
->global_seq
< gt
)
780 msgr
->global_seq
= gt
;
781 ret
= ++msgr
->global_seq
;
782 spin_unlock(&msgr
->global_seq_lock
);
786 static void con_out_kvec_reset(struct ceph_connection
*con
)
788 BUG_ON(con
->out_skip
);
790 con
->out_kvec_left
= 0;
791 con
->out_kvec_bytes
= 0;
792 con
->out_kvec_cur
= &con
->out_kvec
[0];
795 static void con_out_kvec_add(struct ceph_connection
*con
,
796 size_t size
, void *data
)
798 int index
= con
->out_kvec_left
;
800 BUG_ON(con
->out_skip
);
801 BUG_ON(index
>= ARRAY_SIZE(con
->out_kvec
));
803 con
->out_kvec
[index
].iov_len
= size
;
804 con
->out_kvec
[index
].iov_base
= data
;
805 con
->out_kvec_left
++;
806 con
->out_kvec_bytes
+= size
;
810 * Chop off a kvec from the end. Return residual number of bytes for
811 * that kvec, i.e. how many bytes would have been written if the kvec
814 static int con_out_kvec_skip(struct ceph_connection
*con
)
816 int off
= con
->out_kvec_cur
- con
->out_kvec
;
819 if (con
->out_kvec_bytes
> 0) {
820 skip
= con
->out_kvec
[off
+ con
->out_kvec_left
- 1].iov_len
;
821 BUG_ON(con
->out_kvec_bytes
< skip
);
822 BUG_ON(!con
->out_kvec_left
);
823 con
->out_kvec_bytes
-= skip
;
824 con
->out_kvec_left
--;
833 * For a bio data item, a piece is whatever remains of the next
834 * entry in the current bio iovec, or the first entry in the next
837 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor
*cursor
,
840 struct ceph_msg_data
*data
= cursor
->data
;
843 BUG_ON(data
->type
!= CEPH_MSG_DATA_BIO
);
848 cursor
->resid
= min(length
, data
->bio_length
);
850 cursor
->bvec_iter
= bio
->bi_iter
;
852 cursor
->resid
<= bio_iter_len(bio
, cursor
->bvec_iter
);
855 static struct page
*ceph_msg_data_bio_next(struct ceph_msg_data_cursor
*cursor
,
859 struct ceph_msg_data
*data
= cursor
->data
;
861 struct bio_vec bio_vec
;
863 BUG_ON(data
->type
!= CEPH_MSG_DATA_BIO
);
868 bio_vec
= bio_iter_iovec(bio
, cursor
->bvec_iter
);
870 *page_offset
= (size_t) bio_vec
.bv_offset
;
871 BUG_ON(*page_offset
>= PAGE_SIZE
);
872 if (cursor
->last_piece
) /* pagelist offset is always 0 */
873 *length
= cursor
->resid
;
875 *length
= (size_t) bio_vec
.bv_len
;
876 BUG_ON(*length
> cursor
->resid
);
877 BUG_ON(*page_offset
+ *length
> PAGE_SIZE
);
879 return bio_vec
.bv_page
;
882 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor
*cursor
,
886 struct bio_vec bio_vec
;
888 BUG_ON(cursor
->data
->type
!= CEPH_MSG_DATA_BIO
);
893 bio_vec
= bio_iter_iovec(bio
, cursor
->bvec_iter
);
895 /* Advance the cursor offset */
897 BUG_ON(cursor
->resid
< bytes
);
898 cursor
->resid
-= bytes
;
900 bio_advance_iter(bio
, &cursor
->bvec_iter
, bytes
);
902 if (bytes
< bio_vec
.bv_len
)
903 return false; /* more bytes to process in this segment */
905 /* Move on to the next segment, and possibly the next bio */
907 if (!cursor
->bvec_iter
.bi_size
) {
911 cursor
->bvec_iter
= bio
->bi_iter
;
913 memset(&cursor
->bvec_iter
, 0,
914 sizeof(cursor
->bvec_iter
));
917 if (!cursor
->last_piece
) {
918 BUG_ON(!cursor
->resid
);
920 /* A short read is OK, so use <= rather than == */
921 if (cursor
->resid
<= bio_iter_len(bio
, cursor
->bvec_iter
))
922 cursor
->last_piece
= true;
927 #endif /* CONFIG_BLOCK */
930 * For a page array, a piece comes from the first page in the array
931 * that has not already been fully consumed.
933 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor
*cursor
,
936 struct ceph_msg_data
*data
= cursor
->data
;
939 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGES
);
941 BUG_ON(!data
->pages
);
942 BUG_ON(!data
->length
);
944 cursor
->resid
= min(length
, data
->length
);
945 page_count
= calc_pages_for(data
->alignment
, (u64
)data
->length
);
946 cursor
->page_offset
= data
->alignment
& ~PAGE_MASK
;
947 cursor
->page_index
= 0;
948 BUG_ON(page_count
> (int)USHRT_MAX
);
949 cursor
->page_count
= (unsigned short)page_count
;
950 BUG_ON(length
> SIZE_MAX
- cursor
->page_offset
);
951 cursor
->last_piece
= cursor
->page_offset
+ cursor
->resid
<= PAGE_SIZE
;
955 ceph_msg_data_pages_next(struct ceph_msg_data_cursor
*cursor
,
956 size_t *page_offset
, size_t *length
)
958 struct ceph_msg_data
*data
= cursor
->data
;
960 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGES
);
962 BUG_ON(cursor
->page_index
>= cursor
->page_count
);
963 BUG_ON(cursor
->page_offset
>= PAGE_SIZE
);
965 *page_offset
= cursor
->page_offset
;
966 if (cursor
->last_piece
)
967 *length
= cursor
->resid
;
969 *length
= PAGE_SIZE
- *page_offset
;
971 return data
->pages
[cursor
->page_index
];
974 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor
*cursor
,
977 BUG_ON(cursor
->data
->type
!= CEPH_MSG_DATA_PAGES
);
979 BUG_ON(cursor
->page_offset
+ bytes
> PAGE_SIZE
);
981 /* Advance the cursor page offset */
983 cursor
->resid
-= bytes
;
984 cursor
->page_offset
= (cursor
->page_offset
+ bytes
) & ~PAGE_MASK
;
985 if (!bytes
|| cursor
->page_offset
)
986 return false; /* more bytes to process in the current page */
989 return false; /* no more data */
991 /* Move on to the next page; offset is already at 0 */
993 BUG_ON(cursor
->page_index
>= cursor
->page_count
);
994 cursor
->page_index
++;
995 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
1001 * For a pagelist, a piece is whatever remains to be consumed in the
1002 * first page in the list, or the front of the next page.
1005 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor
*cursor
,
1008 struct ceph_msg_data
*data
= cursor
->data
;
1009 struct ceph_pagelist
*pagelist
;
1012 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
1014 pagelist
= data
->pagelist
;
1018 return; /* pagelist can be assigned but empty */
1020 BUG_ON(list_empty(&pagelist
->head
));
1021 page
= list_first_entry(&pagelist
->head
, struct page
, lru
);
1023 cursor
->resid
= min(length
, pagelist
->length
);
1024 cursor
->page
= page
;
1026 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
1029 static struct page
*
1030 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor
*cursor
,
1031 size_t *page_offset
, size_t *length
)
1033 struct ceph_msg_data
*data
= cursor
->data
;
1034 struct ceph_pagelist
*pagelist
;
1036 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
1038 pagelist
= data
->pagelist
;
1041 BUG_ON(!cursor
->page
);
1042 BUG_ON(cursor
->offset
+ cursor
->resid
!= pagelist
->length
);
1044 /* offset of first page in pagelist is always 0 */
1045 *page_offset
= cursor
->offset
& ~PAGE_MASK
;
1046 if (cursor
->last_piece
)
1047 *length
= cursor
->resid
;
1049 *length
= PAGE_SIZE
- *page_offset
;
1051 return cursor
->page
;
1054 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor
*cursor
,
1057 struct ceph_msg_data
*data
= cursor
->data
;
1058 struct ceph_pagelist
*pagelist
;
1060 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
1062 pagelist
= data
->pagelist
;
1065 BUG_ON(cursor
->offset
+ cursor
->resid
!= pagelist
->length
);
1066 BUG_ON((cursor
->offset
& ~PAGE_MASK
) + bytes
> PAGE_SIZE
);
1068 /* Advance the cursor offset */
1070 cursor
->resid
-= bytes
;
1071 cursor
->offset
+= bytes
;
1072 /* offset of first page in pagelist is always 0 */
1073 if (!bytes
|| cursor
->offset
& ~PAGE_MASK
)
1074 return false; /* more bytes to process in the current page */
1077 return false; /* no more data */
1079 /* Move on to the next page */
1081 BUG_ON(list_is_last(&cursor
->page
->lru
, &pagelist
->head
));
1082 cursor
->page
= list_next_entry(cursor
->page
, lru
);
1083 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
1089 * Message data is handled (sent or received) in pieces, where each
1090 * piece resides on a single page. The network layer might not
1091 * consume an entire piece at once. A data item's cursor keeps
1092 * track of which piece is next to process and how much remains to
1093 * be processed in that piece. It also tracks whether the current
1094 * piece is the last one in the data item.
1096 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor
*cursor
)
1098 size_t length
= cursor
->total_resid
;
1100 switch (cursor
->data
->type
) {
1101 case CEPH_MSG_DATA_PAGELIST
:
1102 ceph_msg_data_pagelist_cursor_init(cursor
, length
);
1104 case CEPH_MSG_DATA_PAGES
:
1105 ceph_msg_data_pages_cursor_init(cursor
, length
);
1108 case CEPH_MSG_DATA_BIO
:
1109 ceph_msg_data_bio_cursor_init(cursor
, length
);
1111 #endif /* CONFIG_BLOCK */
1112 case CEPH_MSG_DATA_NONE
:
1117 cursor
->need_crc
= true;
1120 static void ceph_msg_data_cursor_init(struct ceph_msg
*msg
, size_t length
)
1122 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
1123 struct ceph_msg_data
*data
;
1126 BUG_ON(length
> msg
->data_length
);
1127 BUG_ON(list_empty(&msg
->data
));
1129 cursor
->data_head
= &msg
->data
;
1130 cursor
->total_resid
= length
;
1131 data
= list_first_entry(&msg
->data
, struct ceph_msg_data
, links
);
1132 cursor
->data
= data
;
1134 __ceph_msg_data_cursor_init(cursor
);
1138 * Return the page containing the next piece to process for a given
1139 * data item, and supply the page offset and length of that piece.
1140 * Indicate whether this is the last piece in this data item.
1142 static struct page
*ceph_msg_data_next(struct ceph_msg_data_cursor
*cursor
,
1143 size_t *page_offset
, size_t *length
,
1148 switch (cursor
->data
->type
) {
1149 case CEPH_MSG_DATA_PAGELIST
:
1150 page
= ceph_msg_data_pagelist_next(cursor
, page_offset
, length
);
1152 case CEPH_MSG_DATA_PAGES
:
1153 page
= ceph_msg_data_pages_next(cursor
, page_offset
, length
);
1156 case CEPH_MSG_DATA_BIO
:
1157 page
= ceph_msg_data_bio_next(cursor
, page_offset
, length
);
1159 #endif /* CONFIG_BLOCK */
1160 case CEPH_MSG_DATA_NONE
:
1166 BUG_ON(*page_offset
+ *length
> PAGE_SIZE
);
1169 *last_piece
= cursor
->last_piece
;
1175 * Returns true if the result moves the cursor on to the next piece
1178 static void ceph_msg_data_advance(struct ceph_msg_data_cursor
*cursor
,
1183 BUG_ON(bytes
> cursor
->resid
);
1184 switch (cursor
->data
->type
) {
1185 case CEPH_MSG_DATA_PAGELIST
:
1186 new_piece
= ceph_msg_data_pagelist_advance(cursor
, bytes
);
1188 case CEPH_MSG_DATA_PAGES
:
1189 new_piece
= ceph_msg_data_pages_advance(cursor
, bytes
);
1192 case CEPH_MSG_DATA_BIO
:
1193 new_piece
= ceph_msg_data_bio_advance(cursor
, bytes
);
1195 #endif /* CONFIG_BLOCK */
1196 case CEPH_MSG_DATA_NONE
:
1201 cursor
->total_resid
-= bytes
;
1203 if (!cursor
->resid
&& cursor
->total_resid
) {
1204 WARN_ON(!cursor
->last_piece
);
1205 BUG_ON(list_is_last(&cursor
->data
->links
, cursor
->data_head
));
1206 cursor
->data
= list_next_entry(cursor
->data
, links
);
1207 __ceph_msg_data_cursor_init(cursor
);
1210 cursor
->need_crc
= new_piece
;
1213 static size_t sizeof_footer(struct ceph_connection
*con
)
1215 return (con
->peer_features
& CEPH_FEATURE_MSG_AUTH
) ?
1216 sizeof(struct ceph_msg_footer
) :
1217 sizeof(struct ceph_msg_footer_old
);
1220 static void prepare_message_data(struct ceph_msg
*msg
, u32 data_len
)
1225 /* Initialize data cursor */
1227 ceph_msg_data_cursor_init(msg
, (size_t)data_len
);
1231 * Prepare footer for currently outgoing message, and finish things
1232 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1234 static void prepare_write_message_footer(struct ceph_connection
*con
)
1236 struct ceph_msg
*m
= con
->out_msg
;
1238 m
->footer
.flags
|= CEPH_MSG_FOOTER_COMPLETE
;
1240 dout("prepare_write_message_footer %p\n", con
);
1241 con_out_kvec_add(con
, sizeof_footer(con
), &m
->footer
);
1242 if (con
->peer_features
& CEPH_FEATURE_MSG_AUTH
) {
1243 if (con
->ops
->sign_message
)
1244 con
->ops
->sign_message(m
);
1248 m
->old_footer
.flags
= m
->footer
.flags
;
1250 con
->out_more
= m
->more_to_follow
;
1251 con
->out_msg_done
= true;
1255 * Prepare headers for the next outgoing message.
1257 static void prepare_write_message(struct ceph_connection
*con
)
1262 con_out_kvec_reset(con
);
1263 con
->out_msg_done
= false;
1265 /* Sneak an ack in there first? If we can get it into the same
1266 * TCP packet that's a good thing. */
1267 if (con
->in_seq
> con
->in_seq_acked
) {
1268 con
->in_seq_acked
= con
->in_seq
;
1269 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
1270 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1271 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1272 &con
->out_temp_ack
);
1275 BUG_ON(list_empty(&con
->out_queue
));
1276 m
= list_first_entry(&con
->out_queue
, struct ceph_msg
, list_head
);
1278 BUG_ON(m
->con
!= con
);
1280 /* put message on sent list */
1282 list_move_tail(&m
->list_head
, &con
->out_sent
);
1285 * only assign outgoing seq # if we haven't sent this message
1286 * yet. if it is requeued, resend with it's original seq.
1288 if (m
->needs_out_seq
) {
1289 m
->hdr
.seq
= cpu_to_le64(++con
->out_seq
);
1290 m
->needs_out_seq
= false;
1292 if (con
->ops
->reencode_message
)
1293 con
->ops
->reencode_message(m
);
1296 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1297 m
, con
->out_seq
, le16_to_cpu(m
->hdr
.type
),
1298 le32_to_cpu(m
->hdr
.front_len
), le32_to_cpu(m
->hdr
.middle_len
),
1300 WARN_ON(m
->front
.iov_len
!= le32_to_cpu(m
->hdr
.front_len
));
1301 WARN_ON(m
->data_length
!= le32_to_cpu(m
->hdr
.data_len
));
1303 /* tag + hdr + front + middle */
1304 con_out_kvec_add(con
, sizeof (tag_msg
), &tag_msg
);
1305 con_out_kvec_add(con
, sizeof(con
->out_hdr
), &con
->out_hdr
);
1306 con_out_kvec_add(con
, m
->front
.iov_len
, m
->front
.iov_base
);
1309 con_out_kvec_add(con
, m
->middle
->vec
.iov_len
,
1310 m
->middle
->vec
.iov_base
);
1312 /* fill in hdr crc and finalize hdr */
1313 crc
= crc32c(0, &m
->hdr
, offsetof(struct ceph_msg_header
, crc
));
1314 con
->out_msg
->hdr
.crc
= cpu_to_le32(crc
);
1315 memcpy(&con
->out_hdr
, &con
->out_msg
->hdr
, sizeof(con
->out_hdr
));
1317 /* fill in front and middle crc, footer */
1318 crc
= crc32c(0, m
->front
.iov_base
, m
->front
.iov_len
);
1319 con
->out_msg
->footer
.front_crc
= cpu_to_le32(crc
);
1321 crc
= crc32c(0, m
->middle
->vec
.iov_base
,
1322 m
->middle
->vec
.iov_len
);
1323 con
->out_msg
->footer
.middle_crc
= cpu_to_le32(crc
);
1325 con
->out_msg
->footer
.middle_crc
= 0;
1326 dout("%s front_crc %u middle_crc %u\n", __func__
,
1327 le32_to_cpu(con
->out_msg
->footer
.front_crc
),
1328 le32_to_cpu(con
->out_msg
->footer
.middle_crc
));
1329 con
->out_msg
->footer
.flags
= 0;
1331 /* is there a data payload? */
1332 con
->out_msg
->footer
.data_crc
= 0;
1333 if (m
->data_length
) {
1334 prepare_message_data(con
->out_msg
, m
->data_length
);
1335 con
->out_more
= 1; /* data + footer will follow */
1337 /* no, queue up footer too and be done */
1338 prepare_write_message_footer(con
);
1341 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1347 static void prepare_write_ack(struct ceph_connection
*con
)
1349 dout("prepare_write_ack %p %llu -> %llu\n", con
,
1350 con
->in_seq_acked
, con
->in_seq
);
1351 con
->in_seq_acked
= con
->in_seq
;
1353 con_out_kvec_reset(con
);
1355 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
1357 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1358 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1359 &con
->out_temp_ack
);
1361 con
->out_more
= 1; /* more will follow.. eventually.. */
1362 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1366 * Prepare to share the seq during handshake
1368 static void prepare_write_seq(struct ceph_connection
*con
)
1370 dout("prepare_write_seq %p %llu -> %llu\n", con
,
1371 con
->in_seq_acked
, con
->in_seq
);
1372 con
->in_seq_acked
= con
->in_seq
;
1374 con_out_kvec_reset(con
);
1376 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1377 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1378 &con
->out_temp_ack
);
1380 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1384 * Prepare to write keepalive byte.
1386 static void prepare_write_keepalive(struct ceph_connection
*con
)
1388 dout("prepare_write_keepalive %p\n", con
);
1389 con_out_kvec_reset(con
);
1390 if (con
->peer_features
& CEPH_FEATURE_MSGR_KEEPALIVE2
) {
1391 struct timespec now
;
1393 ktime_get_real_ts(&now
);
1394 con_out_kvec_add(con
, sizeof(tag_keepalive2
), &tag_keepalive2
);
1395 ceph_encode_timespec(&con
->out_temp_keepalive2
, &now
);
1396 con_out_kvec_add(con
, sizeof(con
->out_temp_keepalive2
),
1397 &con
->out_temp_keepalive2
);
1399 con_out_kvec_add(con
, sizeof(tag_keepalive
), &tag_keepalive
);
1401 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1405 * Connection negotiation.
1408 static struct ceph_auth_handshake
*get_connect_authorizer(struct ceph_connection
*con
,
1411 struct ceph_auth_handshake
*auth
;
1413 if (!con
->ops
->get_authorizer
) {
1414 con
->out_connect
.authorizer_protocol
= CEPH_AUTH_UNKNOWN
;
1415 con
->out_connect
.authorizer_len
= 0;
1419 auth
= con
->ops
->get_authorizer(con
, auth_proto
, con
->auth_retry
);
1423 con
->auth_reply_buf
= auth
->authorizer_reply_buf
;
1424 con
->auth_reply_buf_len
= auth
->authorizer_reply_buf_len
;
1429 * We connected to a peer and are saying hello.
1431 static void prepare_write_banner(struct ceph_connection
*con
)
1433 con_out_kvec_add(con
, strlen(CEPH_BANNER
), CEPH_BANNER
);
1434 con_out_kvec_add(con
, sizeof (con
->msgr
->my_enc_addr
),
1435 &con
->msgr
->my_enc_addr
);
1438 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1441 static int prepare_write_connect(struct ceph_connection
*con
)
1443 unsigned int global_seq
= get_global_seq(con
->msgr
, 0);
1446 struct ceph_auth_handshake
*auth
;
1448 switch (con
->peer_name
.type
) {
1449 case CEPH_ENTITY_TYPE_MON
:
1450 proto
= CEPH_MONC_PROTOCOL
;
1452 case CEPH_ENTITY_TYPE_OSD
:
1453 proto
= CEPH_OSDC_PROTOCOL
;
1455 case CEPH_ENTITY_TYPE_MDS
:
1456 proto
= CEPH_MDSC_PROTOCOL
;
1462 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con
,
1463 con
->connect_seq
, global_seq
, proto
);
1465 con
->out_connect
.features
=
1466 cpu_to_le64(from_msgr(con
->msgr
)->supported_features
);
1467 con
->out_connect
.host_type
= cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT
);
1468 con
->out_connect
.connect_seq
= cpu_to_le32(con
->connect_seq
);
1469 con
->out_connect
.global_seq
= cpu_to_le32(global_seq
);
1470 con
->out_connect
.protocol_version
= cpu_to_le32(proto
);
1471 con
->out_connect
.flags
= 0;
1473 auth_proto
= CEPH_AUTH_UNKNOWN
;
1474 auth
= get_connect_authorizer(con
, &auth_proto
);
1476 return PTR_ERR(auth
);
1478 con
->out_connect
.authorizer_protocol
= cpu_to_le32(auth_proto
);
1479 con
->out_connect
.authorizer_len
= auth
?
1480 cpu_to_le32(auth
->authorizer_buf_len
) : 0;
1482 con_out_kvec_add(con
, sizeof (con
->out_connect
),
1484 if (auth
&& auth
->authorizer_buf_len
)
1485 con_out_kvec_add(con
, auth
->authorizer_buf_len
,
1486 auth
->authorizer_buf
);
1489 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1495 * write as much of pending kvecs to the socket as we can.
1497 * 0 -> socket full, but more to do
1500 static int write_partial_kvec(struct ceph_connection
*con
)
1504 dout("write_partial_kvec %p %d left\n", con
, con
->out_kvec_bytes
);
1505 while (con
->out_kvec_bytes
> 0) {
1506 ret
= ceph_tcp_sendmsg(con
->sock
, con
->out_kvec_cur
,
1507 con
->out_kvec_left
, con
->out_kvec_bytes
,
1511 con
->out_kvec_bytes
-= ret
;
1512 if (con
->out_kvec_bytes
== 0)
1515 /* account for full iov entries consumed */
1516 while (ret
>= con
->out_kvec_cur
->iov_len
) {
1517 BUG_ON(!con
->out_kvec_left
);
1518 ret
-= con
->out_kvec_cur
->iov_len
;
1519 con
->out_kvec_cur
++;
1520 con
->out_kvec_left
--;
1522 /* and for a partially-consumed entry */
1524 con
->out_kvec_cur
->iov_len
-= ret
;
1525 con
->out_kvec_cur
->iov_base
+= ret
;
1528 con
->out_kvec_left
= 0;
1531 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con
,
1532 con
->out_kvec_bytes
, con
->out_kvec_left
, ret
);
1533 return ret
; /* done! */
1536 static u32
ceph_crc32c_page(u32 crc
, struct page
*page
,
1537 unsigned int page_offset
,
1538 unsigned int length
)
1543 BUG_ON(kaddr
== NULL
);
1544 crc
= crc32c(crc
, kaddr
+ page_offset
, length
);
1550 * Write as much message data payload as we can. If we finish, queue
1552 * 1 -> done, footer is now queued in out_kvec[].
1553 * 0 -> socket full, but more to do
1556 static int write_partial_message_data(struct ceph_connection
*con
)
1558 struct ceph_msg
*msg
= con
->out_msg
;
1559 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
1560 bool do_datacrc
= !ceph_test_opt(from_msgr(con
->msgr
), NOCRC
);
1563 dout("%s %p msg %p\n", __func__
, con
, msg
);
1565 if (list_empty(&msg
->data
))
1569 * Iterate through each page that contains data to be
1570 * written, and send as much as possible for each.
1572 * If we are calculating the data crc (the default), we will
1573 * need to map the page. If we have no pages, they have
1574 * been revoked, so use the zero page.
1576 crc
= do_datacrc
? le32_to_cpu(msg
->footer
.data_crc
) : 0;
1577 while (cursor
->resid
) {
1584 page
= ceph_msg_data_next(cursor
, &page_offset
, &length
,
1586 ret
= ceph_tcp_sendpage(con
->sock
, page
, page_offset
,
1587 length
, !last_piece
);
1590 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1594 if (do_datacrc
&& cursor
->need_crc
)
1595 crc
= ceph_crc32c_page(crc
, page
, page_offset
, length
);
1596 ceph_msg_data_advance(cursor
, (size_t)ret
);
1599 dout("%s %p msg %p done\n", __func__
, con
, msg
);
1601 /* prepare and queue up footer, too */
1603 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1605 msg
->footer
.flags
|= CEPH_MSG_FOOTER_NOCRC
;
1606 con_out_kvec_reset(con
);
1607 prepare_write_message_footer(con
);
1609 return 1; /* must return > 0 to indicate success */
1615 static int write_partial_skip(struct ceph_connection
*con
)
1619 dout("%s %p %d left\n", __func__
, con
, con
->out_skip
);
1620 while (con
->out_skip
> 0) {
1621 size_t size
= min(con
->out_skip
, (int) PAGE_SIZE
);
1623 ret
= ceph_tcp_sendpage(con
->sock
, zero_page
, 0, size
, true);
1626 con
->out_skip
-= ret
;
1634 * Prepare to read connection handshake, or an ack.
1636 static void prepare_read_banner(struct ceph_connection
*con
)
1638 dout("prepare_read_banner %p\n", con
);
1639 con
->in_base_pos
= 0;
1642 static void prepare_read_connect(struct ceph_connection
*con
)
1644 dout("prepare_read_connect %p\n", con
);
1645 con
->in_base_pos
= 0;
1648 static void prepare_read_ack(struct ceph_connection
*con
)
1650 dout("prepare_read_ack %p\n", con
);
1651 con
->in_base_pos
= 0;
1654 static void prepare_read_seq(struct ceph_connection
*con
)
1656 dout("prepare_read_seq %p\n", con
);
1657 con
->in_base_pos
= 0;
1658 con
->in_tag
= CEPH_MSGR_TAG_SEQ
;
1661 static void prepare_read_tag(struct ceph_connection
*con
)
1663 dout("prepare_read_tag %p\n", con
);
1664 con
->in_base_pos
= 0;
1665 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1668 static void prepare_read_keepalive_ack(struct ceph_connection
*con
)
1670 dout("prepare_read_keepalive_ack %p\n", con
);
1671 con
->in_base_pos
= 0;
1675 * Prepare to read a message.
1677 static int prepare_read_message(struct ceph_connection
*con
)
1679 dout("prepare_read_message %p\n", con
);
1680 BUG_ON(con
->in_msg
!= NULL
);
1681 con
->in_base_pos
= 0;
1682 con
->in_front_crc
= con
->in_middle_crc
= con
->in_data_crc
= 0;
1687 static int read_partial(struct ceph_connection
*con
,
1688 int end
, int size
, void *object
)
1690 while (con
->in_base_pos
< end
) {
1691 int left
= end
- con
->in_base_pos
;
1692 int have
= size
- left
;
1693 int ret
= ceph_tcp_recvmsg(con
->sock
, object
+ have
, left
);
1696 con
->in_base_pos
+= ret
;
1703 * Read all or part of the connect-side handshake on a new connection
1705 static int read_partial_banner(struct ceph_connection
*con
)
1711 dout("read_partial_banner %p at %d\n", con
, con
->in_base_pos
);
1714 size
= strlen(CEPH_BANNER
);
1716 ret
= read_partial(con
, end
, size
, con
->in_banner
);
1720 size
= sizeof (con
->actual_peer_addr
);
1722 ret
= read_partial(con
, end
, size
, &con
->actual_peer_addr
);
1726 size
= sizeof (con
->peer_addr_for_me
);
1728 ret
= read_partial(con
, end
, size
, &con
->peer_addr_for_me
);
1736 static int read_partial_connect(struct ceph_connection
*con
)
1742 dout("read_partial_connect %p at %d\n", con
, con
->in_base_pos
);
1744 size
= sizeof (con
->in_reply
);
1746 ret
= read_partial(con
, end
, size
, &con
->in_reply
);
1750 size
= le32_to_cpu(con
->in_reply
.authorizer_len
);
1752 ret
= read_partial(con
, end
, size
, con
->auth_reply_buf
);
1756 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1757 con
, (int)con
->in_reply
.tag
,
1758 le32_to_cpu(con
->in_reply
.connect_seq
),
1759 le32_to_cpu(con
->in_reply
.global_seq
));
1766 * Verify the hello banner looks okay.
1768 static int verify_hello(struct ceph_connection
*con
)
1770 if (memcmp(con
->in_banner
, CEPH_BANNER
, strlen(CEPH_BANNER
))) {
1771 pr_err("connect to %s got bad banner\n",
1772 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1773 con
->error_msg
= "protocol error, bad banner";
1779 static bool addr_is_blank(struct sockaddr_storage
*ss
)
1781 struct in_addr
*addr
= &((struct sockaddr_in
*)ss
)->sin_addr
;
1782 struct in6_addr
*addr6
= &((struct sockaddr_in6
*)ss
)->sin6_addr
;
1784 switch (ss
->ss_family
) {
1786 return addr
->s_addr
== htonl(INADDR_ANY
);
1788 return ipv6_addr_any(addr6
);
1794 static int addr_port(struct sockaddr_storage
*ss
)
1796 switch (ss
->ss_family
) {
1798 return ntohs(((struct sockaddr_in
*)ss
)->sin_port
);
1800 return ntohs(((struct sockaddr_in6
*)ss
)->sin6_port
);
1805 static void addr_set_port(struct sockaddr_storage
*ss
, int p
)
1807 switch (ss
->ss_family
) {
1809 ((struct sockaddr_in
*)ss
)->sin_port
= htons(p
);
1812 ((struct sockaddr_in6
*)ss
)->sin6_port
= htons(p
);
1818 * Unlike other *_pton function semantics, zero indicates success.
1820 static int ceph_pton(const char *str
, size_t len
, struct sockaddr_storage
*ss
,
1821 char delim
, const char **ipend
)
1823 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
1824 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
1826 memset(ss
, 0, sizeof(*ss
));
1828 if (in4_pton(str
, len
, (u8
*)&in4
->sin_addr
.s_addr
, delim
, ipend
)) {
1829 ss
->ss_family
= AF_INET
;
1833 if (in6_pton(str
, len
, (u8
*)&in6
->sin6_addr
.s6_addr
, delim
, ipend
)) {
1834 ss
->ss_family
= AF_INET6
;
1842 * Extract hostname string and resolve using kernel DNS facility.
1844 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1845 static int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1846 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1848 const char *end
, *delim_p
;
1849 char *colon_p
, *ip_addr
= NULL
;
1853 * The end of the hostname occurs immediately preceding the delimiter or
1854 * the port marker (':') where the delimiter takes precedence.
1856 delim_p
= memchr(name
, delim
, namelen
);
1857 colon_p
= memchr(name
, ':', namelen
);
1859 if (delim_p
&& colon_p
)
1860 end
= delim_p
< colon_p
? delim_p
: colon_p
;
1861 else if (!delim_p
&& colon_p
)
1865 if (!end
) /* case: hostname:/ */
1866 end
= name
+ namelen
;
1872 /* do dns_resolve upcall */
1873 ip_len
= dns_query(NULL
, name
, end
- name
, NULL
, &ip_addr
, NULL
);
1875 ret
= ceph_pton(ip_addr
, ip_len
, ss
, -1, NULL
);
1883 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end
- name
), name
,
1884 ret
, ret
? "failed" : ceph_pr_addr(ss
));
1889 static inline int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1890 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1897 * Parse a server name (IP or hostname). If a valid IP address is not found
1898 * then try to extract a hostname to resolve using userspace DNS upcall.
1900 static int ceph_parse_server_name(const char *name
, size_t namelen
,
1901 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1905 ret
= ceph_pton(name
, namelen
, ss
, delim
, ipend
);
1907 ret
= ceph_dns_resolve_name(name
, namelen
, ss
, delim
, ipend
);
1913 * Parse an ip[:port] list into an addr array. Use the default
1914 * monitor port if a port isn't specified.
1916 int ceph_parse_ips(const char *c
, const char *end
,
1917 struct ceph_entity_addr
*addr
,
1918 int max_count
, int *count
)
1920 int i
, ret
= -EINVAL
;
1923 dout("parse_ips on '%.*s'\n", (int)(end
-c
), c
);
1924 for (i
= 0; i
< max_count
; i
++) {
1926 struct sockaddr_storage
*ss
= &addr
[i
].in_addr
;
1935 ret
= ceph_parse_server_name(p
, end
- p
, ss
, delim
, &ipend
);
1944 dout("missing matching ']'\n");
1951 if (p
< end
&& *p
== ':') {
1954 while (p
< end
&& *p
>= '0' && *p
<= '9') {
1955 port
= (port
* 10) + (*p
- '0');
1959 port
= CEPH_MON_PORT
;
1960 else if (port
> 65535)
1963 port
= CEPH_MON_PORT
;
1966 addr_set_port(ss
, port
);
1968 dout("parse_ips got %s\n", ceph_pr_addr(ss
));
1985 pr_err("parse_ips bad ip '%.*s'\n", (int)(end
- c
), c
);
1988 EXPORT_SYMBOL(ceph_parse_ips
);
1990 static int process_banner(struct ceph_connection
*con
)
1992 dout("process_banner on %p\n", con
);
1994 if (verify_hello(con
) < 0)
1997 ceph_decode_addr(&con
->actual_peer_addr
);
1998 ceph_decode_addr(&con
->peer_addr_for_me
);
2001 * Make sure the other end is who we wanted. note that the other
2002 * end may not yet know their ip address, so if it's 0.0.0.0, give
2003 * them the benefit of the doubt.
2005 if (memcmp(&con
->peer_addr
, &con
->actual_peer_addr
,
2006 sizeof(con
->peer_addr
)) != 0 &&
2007 !(addr_is_blank(&con
->actual_peer_addr
.in_addr
) &&
2008 con
->actual_peer_addr
.nonce
== con
->peer_addr
.nonce
)) {
2009 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
2010 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2011 (int)le32_to_cpu(con
->peer_addr
.nonce
),
2012 ceph_pr_addr(&con
->actual_peer_addr
.in_addr
),
2013 (int)le32_to_cpu(con
->actual_peer_addr
.nonce
));
2014 con
->error_msg
= "wrong peer at address";
2019 * did we learn our address?
2021 if (addr_is_blank(&con
->msgr
->inst
.addr
.in_addr
)) {
2022 int port
= addr_port(&con
->msgr
->inst
.addr
.in_addr
);
2024 memcpy(&con
->msgr
->inst
.addr
.in_addr
,
2025 &con
->peer_addr_for_me
.in_addr
,
2026 sizeof(con
->peer_addr_for_me
.in_addr
));
2027 addr_set_port(&con
->msgr
->inst
.addr
.in_addr
, port
);
2028 encode_my_addr(con
->msgr
);
2029 dout("process_banner learned my addr is %s\n",
2030 ceph_pr_addr(&con
->msgr
->inst
.addr
.in_addr
));
2036 static int process_connect(struct ceph_connection
*con
)
2038 u64 sup_feat
= from_msgr(con
->msgr
)->supported_features
;
2039 u64 req_feat
= from_msgr(con
->msgr
)->required_features
;
2040 u64 server_feat
= le64_to_cpu(con
->in_reply
.features
);
2043 dout("process_connect on %p tag %d\n", con
, (int)con
->in_tag
);
2045 if (con
->auth_reply_buf
) {
2047 * Any connection that defines ->get_authorizer()
2048 * should also define ->verify_authorizer_reply().
2049 * See get_connect_authorizer().
2051 ret
= con
->ops
->verify_authorizer_reply(con
);
2053 con
->error_msg
= "bad authorize reply";
2058 switch (con
->in_reply
.tag
) {
2059 case CEPH_MSGR_TAG_FEATURES
:
2060 pr_err("%s%lld %s feature set mismatch,"
2061 " my %llx < server's %llx, missing %llx\n",
2062 ENTITY_NAME(con
->peer_name
),
2063 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2064 sup_feat
, server_feat
, server_feat
& ~sup_feat
);
2065 con
->error_msg
= "missing required protocol features";
2066 reset_connection(con
);
2069 case CEPH_MSGR_TAG_BADPROTOVER
:
2070 pr_err("%s%lld %s protocol version mismatch,"
2071 " my %d != server's %d\n",
2072 ENTITY_NAME(con
->peer_name
),
2073 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2074 le32_to_cpu(con
->out_connect
.protocol_version
),
2075 le32_to_cpu(con
->in_reply
.protocol_version
));
2076 con
->error_msg
= "protocol version mismatch";
2077 reset_connection(con
);
2080 case CEPH_MSGR_TAG_BADAUTHORIZER
:
2082 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con
,
2084 if (con
->auth_retry
== 2) {
2085 con
->error_msg
= "connect authorization failure";
2088 con_out_kvec_reset(con
);
2089 ret
= prepare_write_connect(con
);
2092 prepare_read_connect(con
);
2095 case CEPH_MSGR_TAG_RESETSESSION
:
2097 * If we connected with a large connect_seq but the peer
2098 * has no record of a session with us (no connection, or
2099 * connect_seq == 0), they will send RESETSESION to indicate
2100 * that they must have reset their session, and may have
2103 dout("process_connect got RESET peer seq %u\n",
2104 le32_to_cpu(con
->in_reply
.connect_seq
));
2105 pr_err("%s%lld %s connection reset\n",
2106 ENTITY_NAME(con
->peer_name
),
2107 ceph_pr_addr(&con
->peer_addr
.in_addr
));
2108 reset_connection(con
);
2109 con_out_kvec_reset(con
);
2110 ret
= prepare_write_connect(con
);
2113 prepare_read_connect(con
);
2115 /* Tell ceph about it. */
2116 mutex_unlock(&con
->mutex
);
2117 pr_info("reset on %s%lld\n", ENTITY_NAME(con
->peer_name
));
2118 if (con
->ops
->peer_reset
)
2119 con
->ops
->peer_reset(con
);
2120 mutex_lock(&con
->mutex
);
2121 if (con
->state
!= CON_STATE_NEGOTIATING
)
2125 case CEPH_MSGR_TAG_RETRY_SESSION
:
2127 * If we sent a smaller connect_seq than the peer has, try
2128 * again with a larger value.
2130 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2131 le32_to_cpu(con
->out_connect
.connect_seq
),
2132 le32_to_cpu(con
->in_reply
.connect_seq
));
2133 con
->connect_seq
= le32_to_cpu(con
->in_reply
.connect_seq
);
2134 con_out_kvec_reset(con
);
2135 ret
= prepare_write_connect(con
);
2138 prepare_read_connect(con
);
2141 case CEPH_MSGR_TAG_RETRY_GLOBAL
:
2143 * If we sent a smaller global_seq than the peer has, try
2144 * again with a larger value.
2146 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2147 con
->peer_global_seq
,
2148 le32_to_cpu(con
->in_reply
.global_seq
));
2149 get_global_seq(con
->msgr
,
2150 le32_to_cpu(con
->in_reply
.global_seq
));
2151 con_out_kvec_reset(con
);
2152 ret
= prepare_write_connect(con
);
2155 prepare_read_connect(con
);
2158 case CEPH_MSGR_TAG_SEQ
:
2159 case CEPH_MSGR_TAG_READY
:
2160 if (req_feat
& ~server_feat
) {
2161 pr_err("%s%lld %s protocol feature mismatch,"
2162 " my required %llx > server's %llx, need %llx\n",
2163 ENTITY_NAME(con
->peer_name
),
2164 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2165 req_feat
, server_feat
, req_feat
& ~server_feat
);
2166 con
->error_msg
= "missing required protocol features";
2167 reset_connection(con
);
2171 WARN_ON(con
->state
!= CON_STATE_NEGOTIATING
);
2172 con
->state
= CON_STATE_OPEN
;
2173 con
->auth_retry
= 0; /* we authenticated; clear flag */
2174 con
->peer_global_seq
= le32_to_cpu(con
->in_reply
.global_seq
);
2176 con
->peer_features
= server_feat
;
2177 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2178 con
->peer_global_seq
,
2179 le32_to_cpu(con
->in_reply
.connect_seq
),
2181 WARN_ON(con
->connect_seq
!=
2182 le32_to_cpu(con
->in_reply
.connect_seq
));
2184 if (con
->in_reply
.flags
& CEPH_MSG_CONNECT_LOSSY
)
2185 con_flag_set(con
, CON_FLAG_LOSSYTX
);
2187 con
->delay
= 0; /* reset backoff memory */
2189 if (con
->in_reply
.tag
== CEPH_MSGR_TAG_SEQ
) {
2190 prepare_write_seq(con
);
2191 prepare_read_seq(con
);
2193 prepare_read_tag(con
);
2197 case CEPH_MSGR_TAG_WAIT
:
2199 * If there is a connection race (we are opening
2200 * connections to each other), one of us may just have
2201 * to WAIT. This shouldn't happen if we are the
2204 con
->error_msg
= "protocol error, got WAIT as client";
2208 con
->error_msg
= "protocol error, garbage tag during connect";
2216 * read (part of) an ack
2218 static int read_partial_ack(struct ceph_connection
*con
)
2220 int size
= sizeof (con
->in_temp_ack
);
2223 return read_partial(con
, end
, size
, &con
->in_temp_ack
);
2227 * We can finally discard anything that's been acked.
2229 static void process_ack(struct ceph_connection
*con
)
2232 u64 ack
= le64_to_cpu(con
->in_temp_ack
);
2234 bool reconnect
= (con
->in_tag
== CEPH_MSGR_TAG_SEQ
);
2235 struct list_head
*list
= reconnect
? &con
->out_queue
: &con
->out_sent
;
2238 * In the reconnect case, con_fault() has requeued messages
2239 * in out_sent. We should cleanup old messages according to
2240 * the reconnect seq.
2242 while (!list_empty(list
)) {
2243 m
= list_first_entry(list
, struct ceph_msg
, list_head
);
2244 if (reconnect
&& m
->needs_out_seq
)
2246 seq
= le64_to_cpu(m
->hdr
.seq
);
2249 dout("got ack for seq %llu type %d at %p\n", seq
,
2250 le16_to_cpu(m
->hdr
.type
), m
);
2251 m
->ack_stamp
= jiffies
;
2255 prepare_read_tag(con
);
2259 static int read_partial_message_section(struct ceph_connection
*con
,
2260 struct kvec
*section
,
2261 unsigned int sec_len
, u32
*crc
)
2267 while (section
->iov_len
< sec_len
) {
2268 BUG_ON(section
->iov_base
== NULL
);
2269 left
= sec_len
- section
->iov_len
;
2270 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)section
->iov_base
+
2271 section
->iov_len
, left
);
2274 section
->iov_len
+= ret
;
2276 if (section
->iov_len
== sec_len
)
2277 *crc
= crc32c(0, section
->iov_base
, section
->iov_len
);
2282 static int read_partial_msg_data(struct ceph_connection
*con
)
2284 struct ceph_msg
*msg
= con
->in_msg
;
2285 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
2286 bool do_datacrc
= !ceph_test_opt(from_msgr(con
->msgr
), NOCRC
);
2294 if (list_empty(&msg
->data
))
2298 crc
= con
->in_data_crc
;
2299 while (cursor
->resid
) {
2300 page
= ceph_msg_data_next(cursor
, &page_offset
, &length
, NULL
);
2301 ret
= ceph_tcp_recvpage(con
->sock
, page
, page_offset
, length
);
2304 con
->in_data_crc
= crc
;
2310 crc
= ceph_crc32c_page(crc
, page
, page_offset
, ret
);
2311 ceph_msg_data_advance(cursor
, (size_t)ret
);
2314 con
->in_data_crc
= crc
;
2316 return 1; /* must return > 0 to indicate success */
2320 * read (part of) a message.
2322 static int ceph_con_in_msg_alloc(struct ceph_connection
*con
, int *skip
);
2324 static int read_partial_message(struct ceph_connection
*con
)
2326 struct ceph_msg
*m
= con
->in_msg
;
2330 unsigned int front_len
, middle_len
, data_len
;
2331 bool do_datacrc
= !ceph_test_opt(from_msgr(con
->msgr
), NOCRC
);
2332 bool need_sign
= (con
->peer_features
& CEPH_FEATURE_MSG_AUTH
);
2336 dout("read_partial_message con %p msg %p\n", con
, m
);
2339 size
= sizeof (con
->in_hdr
);
2341 ret
= read_partial(con
, end
, size
, &con
->in_hdr
);
2345 crc
= crc32c(0, &con
->in_hdr
, offsetof(struct ceph_msg_header
, crc
));
2346 if (cpu_to_le32(crc
) != con
->in_hdr
.crc
) {
2347 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
2348 crc
, con
->in_hdr
.crc
);
2352 front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
2353 if (front_len
> CEPH_MSG_MAX_FRONT_LEN
)
2355 middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
2356 if (middle_len
> CEPH_MSG_MAX_MIDDLE_LEN
)
2358 data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
2359 if (data_len
> CEPH_MSG_MAX_DATA_LEN
)
2363 seq
= le64_to_cpu(con
->in_hdr
.seq
);
2364 if ((s64
)seq
- (s64
)con
->in_seq
< 1) {
2365 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2366 ENTITY_NAME(con
->peer_name
),
2367 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2368 seq
, con
->in_seq
+ 1);
2369 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
2371 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2373 } else if ((s64
)seq
- (s64
)con
->in_seq
> 1) {
2374 pr_err("read_partial_message bad seq %lld expected %lld\n",
2375 seq
, con
->in_seq
+ 1);
2376 con
->error_msg
= "bad message sequence # for incoming message";
2380 /* allocate message? */
2384 dout("got hdr type %d front %d data %d\n", con
->in_hdr
.type
,
2385 front_len
, data_len
);
2386 ret
= ceph_con_in_msg_alloc(con
, &skip
);
2390 BUG_ON(!con
->in_msg
^ skip
);
2392 /* skip this message */
2393 dout("alloc_msg said skip message\n");
2394 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
2396 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2401 BUG_ON(!con
->in_msg
);
2402 BUG_ON(con
->in_msg
->con
!= con
);
2404 m
->front
.iov_len
= 0; /* haven't read it yet */
2406 m
->middle
->vec
.iov_len
= 0;
2408 /* prepare for data payload, if any */
2411 prepare_message_data(con
->in_msg
, data_len
);
2415 ret
= read_partial_message_section(con
, &m
->front
, front_len
,
2416 &con
->in_front_crc
);
2422 ret
= read_partial_message_section(con
, &m
->middle
->vec
,
2424 &con
->in_middle_crc
);
2431 ret
= read_partial_msg_data(con
);
2437 size
= sizeof_footer(con
);
2439 ret
= read_partial(con
, end
, size
, &m
->footer
);
2444 m
->footer
.flags
= m
->old_footer
.flags
;
2448 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2449 m
, front_len
, m
->footer
.front_crc
, middle_len
,
2450 m
->footer
.middle_crc
, data_len
, m
->footer
.data_crc
);
2453 if (con
->in_front_crc
!= le32_to_cpu(m
->footer
.front_crc
)) {
2454 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2455 m
, con
->in_front_crc
, m
->footer
.front_crc
);
2458 if (con
->in_middle_crc
!= le32_to_cpu(m
->footer
.middle_crc
)) {
2459 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2460 m
, con
->in_middle_crc
, m
->footer
.middle_crc
);
2464 (m
->footer
.flags
& CEPH_MSG_FOOTER_NOCRC
) == 0 &&
2465 con
->in_data_crc
!= le32_to_cpu(m
->footer
.data_crc
)) {
2466 pr_err("read_partial_message %p data crc %u != exp. %u\n", m
,
2467 con
->in_data_crc
, le32_to_cpu(m
->footer
.data_crc
));
2471 if (need_sign
&& con
->ops
->check_message_signature
&&
2472 con
->ops
->check_message_signature(m
)) {
2473 pr_err("read_partial_message %p signature check failed\n", m
);
2477 return 1; /* done! */
2481 * Process message. This happens in the worker thread. The callback should
2482 * be careful not to do anything that waits on other incoming messages or it
2485 static void process_message(struct ceph_connection
*con
)
2487 struct ceph_msg
*msg
= con
->in_msg
;
2489 BUG_ON(con
->in_msg
->con
!= con
);
2492 /* if first message, set peer_name */
2493 if (con
->peer_name
.type
== 0)
2494 con
->peer_name
= msg
->hdr
.src
;
2497 mutex_unlock(&con
->mutex
);
2499 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2500 msg
, le64_to_cpu(msg
->hdr
.seq
),
2501 ENTITY_NAME(msg
->hdr
.src
),
2502 le16_to_cpu(msg
->hdr
.type
),
2503 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2504 le32_to_cpu(msg
->hdr
.front_len
),
2505 le32_to_cpu(msg
->hdr
.data_len
),
2506 con
->in_front_crc
, con
->in_middle_crc
, con
->in_data_crc
);
2507 con
->ops
->dispatch(con
, msg
);
2509 mutex_lock(&con
->mutex
);
2512 static int read_keepalive_ack(struct ceph_connection
*con
)
2514 struct ceph_timespec ceph_ts
;
2515 size_t size
= sizeof(ceph_ts
);
2516 int ret
= read_partial(con
, size
, size
, &ceph_ts
);
2519 ceph_decode_timespec(&con
->last_keepalive_ack
, &ceph_ts
);
2520 prepare_read_tag(con
);
2525 * Write something to the socket. Called in a worker thread when the
2526 * socket appears to be writeable and we have something ready to send.
2528 static int try_write(struct ceph_connection
*con
)
2532 dout("try_write start %p state %lu\n", con
, con
->state
);
2535 dout("try_write out_kvec_bytes %d\n", con
->out_kvec_bytes
);
2537 /* open the socket first? */
2538 if (con
->state
== CON_STATE_PREOPEN
) {
2540 con
->state
= CON_STATE_CONNECTING
;
2542 con_out_kvec_reset(con
);
2543 prepare_write_banner(con
);
2544 prepare_read_banner(con
);
2546 BUG_ON(con
->in_msg
);
2547 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2548 dout("try_write initiating connect on %p new state %lu\n",
2550 ret
= ceph_tcp_connect(con
);
2552 con
->error_msg
= "connect error";
2558 /* kvec data queued? */
2559 if (con
->out_kvec_left
) {
2560 ret
= write_partial_kvec(con
);
2564 if (con
->out_skip
) {
2565 ret
= write_partial_skip(con
);
2572 if (con
->out_msg_done
) {
2573 ceph_msg_put(con
->out_msg
);
2574 con
->out_msg
= NULL
; /* we're done with this one */
2578 ret
= write_partial_message_data(con
);
2580 goto more_kvec
; /* we need to send the footer, too! */
2584 dout("try_write write_partial_message_data err %d\n",
2591 if (con
->state
== CON_STATE_OPEN
) {
2592 if (con_flag_test_and_clear(con
, CON_FLAG_KEEPALIVE_PENDING
)) {
2593 prepare_write_keepalive(con
);
2596 /* is anything else pending? */
2597 if (!list_empty(&con
->out_queue
)) {
2598 prepare_write_message(con
);
2601 if (con
->in_seq
> con
->in_seq_acked
) {
2602 prepare_write_ack(con
);
2607 /* Nothing to do! */
2608 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
2609 dout("try_write nothing else to write.\n");
2612 dout("try_write done on %p ret %d\n", con
, ret
);
2619 * Read what we can from the socket.
2621 static int try_read(struct ceph_connection
*con
)
2626 dout("try_read start on %p state %lu\n", con
, con
->state
);
2627 if (con
->state
!= CON_STATE_CONNECTING
&&
2628 con
->state
!= CON_STATE_NEGOTIATING
&&
2629 con
->state
!= CON_STATE_OPEN
)
2634 dout("try_read tag %d in_base_pos %d\n", (int)con
->in_tag
,
2637 if (con
->state
== CON_STATE_CONNECTING
) {
2638 dout("try_read connecting\n");
2639 ret
= read_partial_banner(con
);
2642 ret
= process_banner(con
);
2646 con
->state
= CON_STATE_NEGOTIATING
;
2649 * Received banner is good, exchange connection info.
2650 * Do not reset out_kvec, as sending our banner raced
2651 * with receiving peer banner after connect completed.
2653 ret
= prepare_write_connect(con
);
2656 prepare_read_connect(con
);
2658 /* Send connection info before awaiting response */
2662 if (con
->state
== CON_STATE_NEGOTIATING
) {
2663 dout("try_read negotiating\n");
2664 ret
= read_partial_connect(con
);
2667 ret
= process_connect(con
);
2673 WARN_ON(con
->state
!= CON_STATE_OPEN
);
2675 if (con
->in_base_pos
< 0) {
2677 * skipping + discarding content.
2679 * FIXME: there must be a better way to do this!
2681 static char buf
[SKIP_BUF_SIZE
];
2682 int skip
= min((int) sizeof (buf
), -con
->in_base_pos
);
2684 dout("skipping %d / %d bytes\n", skip
, -con
->in_base_pos
);
2685 ret
= ceph_tcp_recvmsg(con
->sock
, buf
, skip
);
2688 con
->in_base_pos
+= ret
;
2689 if (con
->in_base_pos
)
2692 if (con
->in_tag
== CEPH_MSGR_TAG_READY
) {
2696 ret
= ceph_tcp_recvmsg(con
->sock
, &con
->in_tag
, 1);
2699 dout("try_read got tag %d\n", (int)con
->in_tag
);
2700 switch (con
->in_tag
) {
2701 case CEPH_MSGR_TAG_MSG
:
2702 prepare_read_message(con
);
2704 case CEPH_MSGR_TAG_ACK
:
2705 prepare_read_ack(con
);
2707 case CEPH_MSGR_TAG_KEEPALIVE2_ACK
:
2708 prepare_read_keepalive_ack(con
);
2710 case CEPH_MSGR_TAG_CLOSE
:
2711 con_close_socket(con
);
2712 con
->state
= CON_STATE_CLOSED
;
2718 if (con
->in_tag
== CEPH_MSGR_TAG_MSG
) {
2719 ret
= read_partial_message(con
);
2723 con
->error_msg
= "bad crc/signature";
2729 con
->error_msg
= "io error";
2734 if (con
->in_tag
== CEPH_MSGR_TAG_READY
)
2736 process_message(con
);
2737 if (con
->state
== CON_STATE_OPEN
)
2738 prepare_read_tag(con
);
2741 if (con
->in_tag
== CEPH_MSGR_TAG_ACK
||
2742 con
->in_tag
== CEPH_MSGR_TAG_SEQ
) {
2744 * the final handshake seq exchange is semantically
2745 * equivalent to an ACK
2747 ret
= read_partial_ack(con
);
2753 if (con
->in_tag
== CEPH_MSGR_TAG_KEEPALIVE2_ACK
) {
2754 ret
= read_keepalive_ack(con
);
2761 dout("try_read done on %p ret %d\n", con
, ret
);
2765 pr_err("try_read bad con->in_tag = %d\n", (int)con
->in_tag
);
2766 con
->error_msg
= "protocol error, garbage tag";
2773 * Atomically queue work on a connection after the specified delay.
2774 * Bump @con reference to avoid races with connection teardown.
2775 * Returns 0 if work was queued, or an error code otherwise.
2777 static int queue_con_delay(struct ceph_connection
*con
, unsigned long delay
)
2779 if (!con
->ops
->get(con
)) {
2780 dout("%s %p ref count 0\n", __func__
, con
);
2784 if (!queue_delayed_work(ceph_msgr_wq
, &con
->work
, delay
)) {
2785 dout("%s %p - already queued\n", __func__
, con
);
2790 dout("%s %p %lu\n", __func__
, con
, delay
);
2794 static void queue_con(struct ceph_connection
*con
)
2796 (void) queue_con_delay(con
, 0);
2799 static void cancel_con(struct ceph_connection
*con
)
2801 if (cancel_delayed_work(&con
->work
)) {
2802 dout("%s %p\n", __func__
, con
);
2807 static bool con_sock_closed(struct ceph_connection
*con
)
2809 if (!con_flag_test_and_clear(con
, CON_FLAG_SOCK_CLOSED
))
2813 case CON_STATE_ ## x: \
2814 con->error_msg = "socket closed (con state " #x ")"; \
2817 switch (con
->state
) {
2825 pr_warn("%s con %p unrecognized state %lu\n",
2826 __func__
, con
, con
->state
);
2827 con
->error_msg
= "unrecognized con state";
2836 static bool con_backoff(struct ceph_connection
*con
)
2840 if (!con_flag_test_and_clear(con
, CON_FLAG_BACKOFF
))
2843 ret
= queue_con_delay(con
, round_jiffies_relative(con
->delay
));
2845 dout("%s: con %p FAILED to back off %lu\n", __func__
,
2847 BUG_ON(ret
== -ENOENT
);
2848 con_flag_set(con
, CON_FLAG_BACKOFF
);
2854 /* Finish fault handling; con->mutex must *not* be held here */
2856 static void con_fault_finish(struct ceph_connection
*con
)
2858 dout("%s %p\n", __func__
, con
);
2861 * in case we faulted due to authentication, invalidate our
2862 * current tickets so that we can get new ones.
2864 if (con
->auth_retry
) {
2865 dout("auth_retry %d, invalidating\n", con
->auth_retry
);
2866 if (con
->ops
->invalidate_authorizer
)
2867 con
->ops
->invalidate_authorizer(con
);
2868 con
->auth_retry
= 0;
2871 if (con
->ops
->fault
)
2872 con
->ops
->fault(con
);
2876 * Do some work on a connection. Drop a connection ref when we're done.
2878 static void ceph_con_workfn(struct work_struct
*work
)
2880 struct ceph_connection
*con
= container_of(work
, struct ceph_connection
,
2884 mutex_lock(&con
->mutex
);
2888 if ((fault
= con_sock_closed(con
))) {
2889 dout("%s: con %p SOCK_CLOSED\n", __func__
, con
);
2892 if (con_backoff(con
)) {
2893 dout("%s: con %p BACKOFF\n", __func__
, con
);
2896 if (con
->state
== CON_STATE_STANDBY
) {
2897 dout("%s: con %p STANDBY\n", __func__
, con
);
2900 if (con
->state
== CON_STATE_CLOSED
) {
2901 dout("%s: con %p CLOSED\n", __func__
, con
);
2905 if (con
->state
== CON_STATE_PREOPEN
) {
2906 dout("%s: con %p PREOPEN\n", __func__
, con
);
2910 ret
= try_read(con
);
2914 if (!con
->error_msg
)
2915 con
->error_msg
= "socket error on read";
2920 ret
= try_write(con
);
2924 if (!con
->error_msg
)
2925 con
->error_msg
= "socket error on write";
2929 break; /* If we make it to here, we're done */
2933 mutex_unlock(&con
->mutex
);
2936 con_fault_finish(con
);
2942 * Generic error/fault handler. A retry mechanism is used with
2943 * exponential backoff
2945 static void con_fault(struct ceph_connection
*con
)
2947 dout("fault %p state %lu to peer %s\n",
2948 con
, con
->state
, ceph_pr_addr(&con
->peer_addr
.in_addr
));
2950 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con
->peer_name
),
2951 ceph_pr_addr(&con
->peer_addr
.in_addr
), con
->error_msg
);
2952 con
->error_msg
= NULL
;
2954 WARN_ON(con
->state
!= CON_STATE_CONNECTING
&&
2955 con
->state
!= CON_STATE_NEGOTIATING
&&
2956 con
->state
!= CON_STATE_OPEN
);
2958 con_close_socket(con
);
2960 if (con_flag_test(con
, CON_FLAG_LOSSYTX
)) {
2961 dout("fault on LOSSYTX channel, marking CLOSED\n");
2962 con
->state
= CON_STATE_CLOSED
;
2967 BUG_ON(con
->in_msg
->con
!= con
);
2968 ceph_msg_put(con
->in_msg
);
2972 /* Requeue anything that hasn't been acked */
2973 list_splice_init(&con
->out_sent
, &con
->out_queue
);
2975 /* If there are no messages queued or keepalive pending, place
2976 * the connection in a STANDBY state */
2977 if (list_empty(&con
->out_queue
) &&
2978 !con_flag_test(con
, CON_FLAG_KEEPALIVE_PENDING
)) {
2979 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con
);
2980 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
2981 con
->state
= CON_STATE_STANDBY
;
2983 /* retry after a delay. */
2984 con
->state
= CON_STATE_PREOPEN
;
2985 if (con
->delay
== 0)
2986 con
->delay
= BASE_DELAY_INTERVAL
;
2987 else if (con
->delay
< MAX_DELAY_INTERVAL
)
2989 con_flag_set(con
, CON_FLAG_BACKOFF
);
2997 * initialize a new messenger instance
2999 void ceph_messenger_init(struct ceph_messenger
*msgr
,
3000 struct ceph_entity_addr
*myaddr
)
3002 spin_lock_init(&msgr
->global_seq_lock
);
3005 msgr
->inst
.addr
= *myaddr
;
3007 /* select a random nonce */
3008 msgr
->inst
.addr
.type
= 0;
3009 get_random_bytes(&msgr
->inst
.addr
.nonce
, sizeof(msgr
->inst
.addr
.nonce
));
3010 encode_my_addr(msgr
);
3012 atomic_set(&msgr
->stopping
, 0);
3013 write_pnet(&msgr
->net
, get_net(current
->nsproxy
->net_ns
));
3015 dout("%s %p\n", __func__
, msgr
);
3017 EXPORT_SYMBOL(ceph_messenger_init
);
3019 void ceph_messenger_fini(struct ceph_messenger
*msgr
)
3021 put_net(read_pnet(&msgr
->net
));
3023 EXPORT_SYMBOL(ceph_messenger_fini
);
3025 static void msg_con_set(struct ceph_msg
*msg
, struct ceph_connection
*con
)
3028 msg
->con
->ops
->put(msg
->con
);
3030 msg
->con
= con
? con
->ops
->get(con
) : NULL
;
3031 BUG_ON(msg
->con
!= con
);
3034 static void clear_standby(struct ceph_connection
*con
)
3036 /* come back from STANDBY? */
3037 if (con
->state
== CON_STATE_STANDBY
) {
3038 dout("clear_standby %p and ++connect_seq\n", con
);
3039 con
->state
= CON_STATE_PREOPEN
;
3041 WARN_ON(con_flag_test(con
, CON_FLAG_WRITE_PENDING
));
3042 WARN_ON(con_flag_test(con
, CON_FLAG_KEEPALIVE_PENDING
));
3047 * Queue up an outgoing message on the given connection.
3049 void ceph_con_send(struct ceph_connection
*con
, struct ceph_msg
*msg
)
3052 msg
->hdr
.src
= con
->msgr
->inst
.name
;
3053 BUG_ON(msg
->front
.iov_len
!= le32_to_cpu(msg
->hdr
.front_len
));
3054 msg
->needs_out_seq
= true;
3056 mutex_lock(&con
->mutex
);
3058 if (con
->state
== CON_STATE_CLOSED
) {
3059 dout("con_send %p closed, dropping %p\n", con
, msg
);
3061 mutex_unlock(&con
->mutex
);
3065 msg_con_set(msg
, con
);
3067 BUG_ON(!list_empty(&msg
->list_head
));
3068 list_add_tail(&msg
->list_head
, &con
->out_queue
);
3069 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg
,
3070 ENTITY_NAME(con
->peer_name
), le16_to_cpu(msg
->hdr
.type
),
3071 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
3072 le32_to_cpu(msg
->hdr
.front_len
),
3073 le32_to_cpu(msg
->hdr
.middle_len
),
3074 le32_to_cpu(msg
->hdr
.data_len
));
3077 mutex_unlock(&con
->mutex
);
3079 /* if there wasn't anything waiting to send before, queue
3081 if (con_flag_test_and_set(con
, CON_FLAG_WRITE_PENDING
) == 0)
3084 EXPORT_SYMBOL(ceph_con_send
);
3087 * Revoke a message that was previously queued for send
3089 void ceph_msg_revoke(struct ceph_msg
*msg
)
3091 struct ceph_connection
*con
= msg
->con
;
3094 dout("%s msg %p null con\n", __func__
, msg
);
3095 return; /* Message not in our possession */
3098 mutex_lock(&con
->mutex
);
3099 if (!list_empty(&msg
->list_head
)) {
3100 dout("%s %p msg %p - was on queue\n", __func__
, con
, msg
);
3101 list_del_init(&msg
->list_head
);
3106 if (con
->out_msg
== msg
) {
3107 BUG_ON(con
->out_skip
);
3109 if (con
->out_msg_done
) {
3110 con
->out_skip
+= con_out_kvec_skip(con
);
3112 BUG_ON(!msg
->data_length
);
3113 con
->out_skip
+= sizeof_footer(con
);
3115 /* data, middle, front */
3116 if (msg
->data_length
)
3117 con
->out_skip
+= msg
->cursor
.total_resid
;
3119 con
->out_skip
+= con_out_kvec_skip(con
);
3120 con
->out_skip
+= con_out_kvec_skip(con
);
3122 dout("%s %p msg %p - was sending, will write %d skip %d\n",
3123 __func__
, con
, msg
, con
->out_kvec_bytes
, con
->out_skip
);
3125 con
->out_msg
= NULL
;
3129 mutex_unlock(&con
->mutex
);
3133 * Revoke a message that we may be reading data into
3135 void ceph_msg_revoke_incoming(struct ceph_msg
*msg
)
3137 struct ceph_connection
*con
= msg
->con
;
3140 dout("%s msg %p null con\n", __func__
, msg
);
3141 return; /* Message not in our possession */
3144 mutex_lock(&con
->mutex
);
3145 if (con
->in_msg
== msg
) {
3146 unsigned int front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
3147 unsigned int middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
3148 unsigned int data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
3150 /* skip rest of message */
3151 dout("%s %p msg %p revoked\n", __func__
, con
, msg
);
3152 con
->in_base_pos
= con
->in_base_pos
-
3153 sizeof(struct ceph_msg_header
) -
3157 sizeof(struct ceph_msg_footer
);
3158 ceph_msg_put(con
->in_msg
);
3160 con
->in_tag
= CEPH_MSGR_TAG_READY
;
3163 dout("%s %p in_msg %p msg %p no-op\n",
3164 __func__
, con
, con
->in_msg
, msg
);
3166 mutex_unlock(&con
->mutex
);
3170 * Queue a keepalive byte to ensure the tcp connection is alive.
3172 void ceph_con_keepalive(struct ceph_connection
*con
)
3174 dout("con_keepalive %p\n", con
);
3175 mutex_lock(&con
->mutex
);
3177 mutex_unlock(&con
->mutex
);
3178 if (con_flag_test_and_set(con
, CON_FLAG_KEEPALIVE_PENDING
) == 0 &&
3179 con_flag_test_and_set(con
, CON_FLAG_WRITE_PENDING
) == 0)
3182 EXPORT_SYMBOL(ceph_con_keepalive
);
3184 bool ceph_con_keepalive_expired(struct ceph_connection
*con
,
3185 unsigned long interval
)
3188 (con
->peer_features
& CEPH_FEATURE_MSGR_KEEPALIVE2
)) {
3189 struct timespec now
;
3191 ktime_get_real_ts(&now
);
3192 jiffies_to_timespec(interval
, &ts
);
3193 ts
= timespec_add(con
->last_keepalive_ack
, ts
);
3194 return timespec_compare(&now
, &ts
) >= 0;
3199 static struct ceph_msg_data
*ceph_msg_data_create(enum ceph_msg_data_type type
)
3201 struct ceph_msg_data
*data
;
3203 if (WARN_ON(!ceph_msg_data_type_valid(type
)))
3206 data
= kmem_cache_zalloc(ceph_msg_data_cache
, GFP_NOFS
);
3211 INIT_LIST_HEAD(&data
->links
);
3216 static void ceph_msg_data_destroy(struct ceph_msg_data
*data
)
3221 WARN_ON(!list_empty(&data
->links
));
3222 if (data
->type
== CEPH_MSG_DATA_PAGELIST
)
3223 ceph_pagelist_release(data
->pagelist
);
3224 kmem_cache_free(ceph_msg_data_cache
, data
);
3227 void ceph_msg_data_add_pages(struct ceph_msg
*msg
, struct page
**pages
,
3228 size_t length
, size_t alignment
)
3230 struct ceph_msg_data
*data
;
3235 data
= ceph_msg_data_create(CEPH_MSG_DATA_PAGES
);
3237 data
->pages
= pages
;
3238 data
->length
= length
;
3239 data
->alignment
= alignment
& ~PAGE_MASK
;
3241 list_add_tail(&data
->links
, &msg
->data
);
3242 msg
->data_length
+= length
;
3244 EXPORT_SYMBOL(ceph_msg_data_add_pages
);
3246 void ceph_msg_data_add_pagelist(struct ceph_msg
*msg
,
3247 struct ceph_pagelist
*pagelist
)
3249 struct ceph_msg_data
*data
;
3252 BUG_ON(!pagelist
->length
);
3254 data
= ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST
);
3256 data
->pagelist
= pagelist
;
3258 list_add_tail(&data
->links
, &msg
->data
);
3259 msg
->data_length
+= pagelist
->length
;
3261 EXPORT_SYMBOL(ceph_msg_data_add_pagelist
);
3264 void ceph_msg_data_add_bio(struct ceph_msg
*msg
, struct bio
*bio
,
3267 struct ceph_msg_data
*data
;
3271 data
= ceph_msg_data_create(CEPH_MSG_DATA_BIO
);
3274 data
->bio_length
= length
;
3276 list_add_tail(&data
->links
, &msg
->data
);
3277 msg
->data_length
+= length
;
3279 EXPORT_SYMBOL(ceph_msg_data_add_bio
);
3280 #endif /* CONFIG_BLOCK */
3283 * construct a new message with given type, size
3284 * the new msg has a ref count of 1.
3286 struct ceph_msg
*ceph_msg_new(int type
, int front_len
, gfp_t flags
,
3291 m
= kmem_cache_zalloc(ceph_msg_cache
, flags
);
3295 m
->hdr
.type
= cpu_to_le16(type
);
3296 m
->hdr
.priority
= cpu_to_le16(CEPH_MSG_PRIO_DEFAULT
);
3297 m
->hdr
.front_len
= cpu_to_le32(front_len
);
3299 INIT_LIST_HEAD(&m
->list_head
);
3300 kref_init(&m
->kref
);
3301 INIT_LIST_HEAD(&m
->data
);
3305 m
->front
.iov_base
= ceph_kvmalloc(front_len
, flags
);
3306 if (m
->front
.iov_base
== NULL
) {
3307 dout("ceph_msg_new can't allocate %d bytes\n",
3312 m
->front
.iov_base
= NULL
;
3314 m
->front_alloc_len
= m
->front
.iov_len
= front_len
;
3316 dout("ceph_msg_new %p front %d\n", m
, front_len
);
3323 pr_err("msg_new can't create type %d front %d\n", type
,
3327 dout("msg_new can't create type %d front %d\n", type
,
3332 EXPORT_SYMBOL(ceph_msg_new
);
3335 * Allocate "middle" portion of a message, if it is needed and wasn't
3336 * allocated by alloc_msg. This allows us to read a small fixed-size
3337 * per-type header in the front and then gracefully fail (i.e.,
3338 * propagate the error to the caller based on info in the front) when
3339 * the middle is too large.
3341 static int ceph_alloc_middle(struct ceph_connection
*con
, struct ceph_msg
*msg
)
3343 int type
= le16_to_cpu(msg
->hdr
.type
);
3344 int middle_len
= le32_to_cpu(msg
->hdr
.middle_len
);
3346 dout("alloc_middle %p type %d %s middle_len %d\n", msg
, type
,
3347 ceph_msg_type_name(type
), middle_len
);
3348 BUG_ON(!middle_len
);
3349 BUG_ON(msg
->middle
);
3351 msg
->middle
= ceph_buffer_new(middle_len
, GFP_NOFS
);
3358 * Allocate a message for receiving an incoming message on a
3359 * connection, and save the result in con->in_msg. Uses the
3360 * connection's private alloc_msg op if available.
3362 * Returns 0 on success, or a negative error code.
3364 * On success, if we set *skip = 1:
3365 * - the next message should be skipped and ignored.
3366 * - con->in_msg == NULL
3367 * or if we set *skip = 0:
3368 * - con->in_msg is non-null.
3369 * On error (ENOMEM, EAGAIN, ...),
3370 * - con->in_msg == NULL
3372 static int ceph_con_in_msg_alloc(struct ceph_connection
*con
, int *skip
)
3374 struct ceph_msg_header
*hdr
= &con
->in_hdr
;
3375 int middle_len
= le32_to_cpu(hdr
->middle_len
);
3376 struct ceph_msg
*msg
;
3379 BUG_ON(con
->in_msg
!= NULL
);
3380 BUG_ON(!con
->ops
->alloc_msg
);
3382 mutex_unlock(&con
->mutex
);
3383 msg
= con
->ops
->alloc_msg(con
, hdr
, skip
);
3384 mutex_lock(&con
->mutex
);
3385 if (con
->state
!= CON_STATE_OPEN
) {
3392 msg_con_set(msg
, con
);
3396 * Null message pointer means either we should skip
3397 * this message or we couldn't allocate memory. The
3398 * former is not an error.
3403 con
->error_msg
= "error allocating memory for incoming message";
3406 memcpy(&con
->in_msg
->hdr
, &con
->in_hdr
, sizeof(con
->in_hdr
));
3408 if (middle_len
&& !con
->in_msg
->middle
) {
3409 ret
= ceph_alloc_middle(con
, con
->in_msg
);
3411 ceph_msg_put(con
->in_msg
);
3421 * Free a generically kmalloc'd message.
3423 static void ceph_msg_free(struct ceph_msg
*m
)
3425 dout("%s %p\n", __func__
, m
);
3426 kvfree(m
->front
.iov_base
);
3427 kmem_cache_free(ceph_msg_cache
, m
);
3430 static void ceph_msg_release(struct kref
*kref
)
3432 struct ceph_msg
*m
= container_of(kref
, struct ceph_msg
, kref
);
3433 struct ceph_msg_data
*data
, *next
;
3435 dout("%s %p\n", __func__
, m
);
3436 WARN_ON(!list_empty(&m
->list_head
));
3438 msg_con_set(m
, NULL
);
3440 /* drop middle, data, if any */
3442 ceph_buffer_put(m
->middle
);
3446 list_for_each_entry_safe(data
, next
, &m
->data
, links
) {
3447 list_del_init(&data
->links
);
3448 ceph_msg_data_destroy(data
);
3453 ceph_msgpool_put(m
->pool
, m
);
3458 struct ceph_msg
*ceph_msg_get(struct ceph_msg
*msg
)
3460 dout("%s %p (was %d)\n", __func__
, msg
,
3461 kref_read(&msg
->kref
));
3462 kref_get(&msg
->kref
);
3465 EXPORT_SYMBOL(ceph_msg_get
);
3467 void ceph_msg_put(struct ceph_msg
*msg
)
3469 dout("%s %p (was %d)\n", __func__
, msg
,
3470 kref_read(&msg
->kref
));
3471 kref_put(&msg
->kref
, ceph_msg_release
);
3473 EXPORT_SYMBOL(ceph_msg_put
);
3475 void ceph_msg_dump(struct ceph_msg
*msg
)
3477 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg
,
3478 msg
->front_alloc_len
, msg
->data_length
);
3479 print_hex_dump(KERN_DEBUG
, "header: ",
3480 DUMP_PREFIX_OFFSET
, 16, 1,
3481 &msg
->hdr
, sizeof(msg
->hdr
), true);
3482 print_hex_dump(KERN_DEBUG
, " front: ",
3483 DUMP_PREFIX_OFFSET
, 16, 1,
3484 msg
->front
.iov_base
, msg
->front
.iov_len
, true);
3486 print_hex_dump(KERN_DEBUG
, "middle: ",
3487 DUMP_PREFIX_OFFSET
, 16, 1,
3488 msg
->middle
->vec
.iov_base
,
3489 msg
->middle
->vec
.iov_len
, true);
3490 print_hex_dump(KERN_DEBUG
, "footer: ",
3491 DUMP_PREFIX_OFFSET
, 16, 1,
3492 &msg
->footer
, sizeof(msg
->footer
), true);
3494 EXPORT_SYMBOL(ceph_msg_dump
);