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__
);
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
;
475 unsigned int noio_flag
;
480 /* sock_create_kern() allocates with GFP_KERNEL */
481 noio_flag
= memalloc_noio_save();
482 ret
= sock_create_kern(read_pnet(&con
->msgr
->net
), paddr
->ss_family
,
483 SOCK_STREAM
, IPPROTO_TCP
, &sock
);
484 memalloc_noio_restore(noio_flag
);
487 sock
->sk
->sk_allocation
= GFP_NOFS
;
489 #ifdef CONFIG_LOCKDEP
490 lockdep_set_class(&sock
->sk
->sk_lock
, &socket_class
);
493 set_sock_callbacks(sock
, con
);
495 dout("connect %s\n", ceph_pr_addr(&con
->peer_addr
.in_addr
));
497 con_sock_state_connecting(con
);
498 ret
= sock
->ops
->connect(sock
, (struct sockaddr
*)paddr
, sizeof(*paddr
),
500 if (ret
== -EINPROGRESS
) {
501 dout("connect %s EINPROGRESS sk_state = %u\n",
502 ceph_pr_addr(&con
->peer_addr
.in_addr
),
504 } else if (ret
< 0) {
505 pr_err("connect %s error %d\n",
506 ceph_pr_addr(&con
->peer_addr
.in_addr
), ret
);
511 if (ceph_test_opt(from_msgr(con
->msgr
), TCP_NODELAY
)) {
514 ret
= kernel_setsockopt(sock
, SOL_TCP
, TCP_NODELAY
,
515 (char *)&optval
, sizeof(optval
));
517 pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
525 static int ceph_tcp_recvmsg(struct socket
*sock
, void *buf
, size_t len
)
527 struct kvec iov
= {buf
, len
};
528 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
531 iov_iter_kvec(&msg
.msg_iter
, READ
| ITER_KVEC
, &iov
, 1, len
);
532 r
= sock_recvmsg(sock
, &msg
, msg
.msg_flags
);
538 static int ceph_tcp_recvpage(struct socket
*sock
, struct page
*page
,
539 int page_offset
, size_t length
)
541 struct bio_vec bvec
= {
543 .bv_offset
= page_offset
,
546 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
549 BUG_ON(page_offset
+ length
> PAGE_SIZE
);
550 iov_iter_bvec(&msg
.msg_iter
, READ
| ITER_BVEC
, &bvec
, 1, length
);
551 r
= sock_recvmsg(sock
, &msg
, msg
.msg_flags
);
558 * write something. @more is true if caller will be sending more data
561 static int ceph_tcp_sendmsg(struct socket
*sock
, struct kvec
*iov
,
562 size_t kvlen
, size_t len
, int more
)
564 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
568 msg
.msg_flags
|= MSG_MORE
;
570 msg
.msg_flags
|= MSG_EOR
; /* superfluous, but what the hell */
572 r
= kernel_sendmsg(sock
, &msg
, iov
, kvlen
, len
);
578 static int __ceph_tcp_sendpage(struct socket
*sock
, struct page
*page
,
579 int offset
, size_t size
, bool more
)
581 int flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
| (more
? MSG_MORE
: MSG_EOR
);
584 ret
= kernel_sendpage(sock
, page
, offset
, size
, flags
);
591 static int ceph_tcp_sendpage(struct socket
*sock
, struct page
*page
,
592 int offset
, size_t size
, bool more
)
594 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
599 * sendpage cannot properly handle pages with page_count == 0,
600 * we need to fall back to sendmsg if that's the case.
602 * Same goes for slab pages: skb_can_coalesce() allows
603 * coalescing neighboring slab objects into a single frag which
604 * triggers one of hardened usercopy checks.
606 if (page_count(page
) >= 1 && !PageSlab(page
))
607 return __ceph_tcp_sendpage(sock
, page
, offset
, size
, more
);
610 bvec
.bv_offset
= offset
;
614 msg
.msg_flags
|= MSG_MORE
;
616 msg
.msg_flags
|= MSG_EOR
; /* superfluous, but what the hell */
618 iov_iter_bvec(&msg
.msg_iter
, WRITE
| ITER_BVEC
, &bvec
, 1, size
);
619 ret
= sock_sendmsg(sock
, &msg
);
627 * Shutdown/close the socket for the given connection.
629 static int con_close_socket(struct ceph_connection
*con
)
633 dout("con_close_socket on %p sock %p\n", con
, con
->sock
);
635 rc
= con
->sock
->ops
->shutdown(con
->sock
, SHUT_RDWR
);
636 sock_release(con
->sock
);
641 * Forcibly clear the SOCK_CLOSED flag. It gets set
642 * independent of the connection mutex, and we could have
643 * received a socket close event before we had the chance to
644 * shut the socket down.
646 con_flag_clear(con
, CON_FLAG_SOCK_CLOSED
);
648 con_sock_state_closed(con
);
653 * Reset a connection. Discard all incoming and outgoing messages
654 * and clear *_seq state.
656 static void ceph_msg_remove(struct ceph_msg
*msg
)
658 list_del_init(&msg
->list_head
);
662 static void ceph_msg_remove_list(struct list_head
*head
)
664 while (!list_empty(head
)) {
665 struct ceph_msg
*msg
= list_first_entry(head
, struct ceph_msg
,
667 ceph_msg_remove(msg
);
671 static void reset_connection(struct ceph_connection
*con
)
673 /* reset connection, out_queue, msg_ and connect_seq */
674 /* discard existing out_queue and msg_seq */
675 dout("reset_connection %p\n", con
);
676 ceph_msg_remove_list(&con
->out_queue
);
677 ceph_msg_remove_list(&con
->out_sent
);
680 BUG_ON(con
->in_msg
->con
!= con
);
681 ceph_msg_put(con
->in_msg
);
685 con
->connect_seq
= 0;
688 BUG_ON(con
->out_msg
->con
!= con
);
689 ceph_msg_put(con
->out_msg
);
693 con
->in_seq_acked
= 0;
699 * mark a peer down. drop any open connections.
701 void ceph_con_close(struct ceph_connection
*con
)
703 mutex_lock(&con
->mutex
);
704 dout("con_close %p peer %s\n", con
,
705 ceph_pr_addr(&con
->peer_addr
.in_addr
));
706 con
->state
= CON_STATE_CLOSED
;
708 con_flag_clear(con
, CON_FLAG_LOSSYTX
); /* so we retry next connect */
709 con_flag_clear(con
, CON_FLAG_KEEPALIVE_PENDING
);
710 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
711 con_flag_clear(con
, CON_FLAG_BACKOFF
);
713 reset_connection(con
);
714 con
->peer_global_seq
= 0;
716 con_close_socket(con
);
717 mutex_unlock(&con
->mutex
);
719 EXPORT_SYMBOL(ceph_con_close
);
722 * Reopen a closed connection, with a new peer address.
724 void ceph_con_open(struct ceph_connection
*con
,
725 __u8 entity_type
, __u64 entity_num
,
726 struct ceph_entity_addr
*addr
)
728 mutex_lock(&con
->mutex
);
729 dout("con_open %p %s\n", con
, ceph_pr_addr(&addr
->in_addr
));
731 WARN_ON(con
->state
!= CON_STATE_CLOSED
);
732 con
->state
= CON_STATE_PREOPEN
;
734 con
->peer_name
.type
= (__u8
) entity_type
;
735 con
->peer_name
.num
= cpu_to_le64(entity_num
);
737 memcpy(&con
->peer_addr
, addr
, sizeof(*addr
));
738 con
->delay
= 0; /* reset backoff memory */
739 mutex_unlock(&con
->mutex
);
742 EXPORT_SYMBOL(ceph_con_open
);
745 * return true if this connection ever successfully opened
747 bool ceph_con_opened(struct ceph_connection
*con
)
749 return con
->connect_seq
> 0;
753 * initialize a new connection.
755 void ceph_con_init(struct ceph_connection
*con
, void *private,
756 const struct ceph_connection_operations
*ops
,
757 struct ceph_messenger
*msgr
)
759 dout("con_init %p\n", con
);
760 memset(con
, 0, sizeof(*con
));
761 con
->private = private;
765 con_sock_state_init(con
);
767 mutex_init(&con
->mutex
);
768 INIT_LIST_HEAD(&con
->out_queue
);
769 INIT_LIST_HEAD(&con
->out_sent
);
770 INIT_DELAYED_WORK(&con
->work
, ceph_con_workfn
);
772 con
->state
= CON_STATE_CLOSED
;
774 EXPORT_SYMBOL(ceph_con_init
);
778 * We maintain a global counter to order connection attempts. Get
779 * a unique seq greater than @gt.
781 static u32
get_global_seq(struct ceph_messenger
*msgr
, u32 gt
)
785 spin_lock(&msgr
->global_seq_lock
);
786 if (msgr
->global_seq
< gt
)
787 msgr
->global_seq
= gt
;
788 ret
= ++msgr
->global_seq
;
789 spin_unlock(&msgr
->global_seq_lock
);
793 static void con_out_kvec_reset(struct ceph_connection
*con
)
795 BUG_ON(con
->out_skip
);
797 con
->out_kvec_left
= 0;
798 con
->out_kvec_bytes
= 0;
799 con
->out_kvec_cur
= &con
->out_kvec
[0];
802 static void con_out_kvec_add(struct ceph_connection
*con
,
803 size_t size
, void *data
)
805 int index
= con
->out_kvec_left
;
807 BUG_ON(con
->out_skip
);
808 BUG_ON(index
>= ARRAY_SIZE(con
->out_kvec
));
810 con
->out_kvec
[index
].iov_len
= size
;
811 con
->out_kvec
[index
].iov_base
= data
;
812 con
->out_kvec_left
++;
813 con
->out_kvec_bytes
+= size
;
817 * Chop off a kvec from the end. Return residual number of bytes for
818 * that kvec, i.e. how many bytes would have been written if the kvec
821 static int con_out_kvec_skip(struct ceph_connection
*con
)
823 int off
= con
->out_kvec_cur
- con
->out_kvec
;
826 if (con
->out_kvec_bytes
> 0) {
827 skip
= con
->out_kvec
[off
+ con
->out_kvec_left
- 1].iov_len
;
828 BUG_ON(con
->out_kvec_bytes
< skip
);
829 BUG_ON(!con
->out_kvec_left
);
830 con
->out_kvec_bytes
-= skip
;
831 con
->out_kvec_left
--;
840 * For a bio data item, a piece is whatever remains of the next
841 * entry in the current bio iovec, or the first entry in the next
844 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor
*cursor
,
847 struct ceph_msg_data
*data
= cursor
->data
;
850 BUG_ON(data
->type
!= CEPH_MSG_DATA_BIO
);
855 cursor
->resid
= min(length
, data
->bio_length
);
857 cursor
->bvec_iter
= bio
->bi_iter
;
859 cursor
->resid
<= bio_iter_len(bio
, cursor
->bvec_iter
);
862 static struct page
*ceph_msg_data_bio_next(struct ceph_msg_data_cursor
*cursor
,
866 struct ceph_msg_data
*data
= cursor
->data
;
868 struct bio_vec bio_vec
;
870 BUG_ON(data
->type
!= CEPH_MSG_DATA_BIO
);
875 bio_vec
= bio_iter_iovec(bio
, cursor
->bvec_iter
);
877 *page_offset
= (size_t) bio_vec
.bv_offset
;
878 BUG_ON(*page_offset
>= PAGE_SIZE
);
879 if (cursor
->last_piece
) /* pagelist offset is always 0 */
880 *length
= cursor
->resid
;
882 *length
= (size_t) bio_vec
.bv_len
;
883 BUG_ON(*length
> cursor
->resid
);
884 BUG_ON(*page_offset
+ *length
> PAGE_SIZE
);
886 return bio_vec
.bv_page
;
889 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor
*cursor
,
893 struct bio_vec bio_vec
;
895 BUG_ON(cursor
->data
->type
!= CEPH_MSG_DATA_BIO
);
900 bio_vec
= bio_iter_iovec(bio
, cursor
->bvec_iter
);
902 /* Advance the cursor offset */
904 BUG_ON(cursor
->resid
< bytes
);
905 cursor
->resid
-= bytes
;
907 bio_advance_iter(bio
, &cursor
->bvec_iter
, bytes
);
909 if (bytes
< bio_vec
.bv_len
)
910 return false; /* more bytes to process in this segment */
912 /* Move on to the next segment, and possibly the next bio */
914 if (!cursor
->bvec_iter
.bi_size
) {
918 cursor
->bvec_iter
= bio
->bi_iter
;
920 memset(&cursor
->bvec_iter
, 0,
921 sizeof(cursor
->bvec_iter
));
924 if (!cursor
->last_piece
) {
925 BUG_ON(!cursor
->resid
);
927 /* A short read is OK, so use <= rather than == */
928 if (cursor
->resid
<= bio_iter_len(bio
, cursor
->bvec_iter
))
929 cursor
->last_piece
= true;
934 #endif /* CONFIG_BLOCK */
937 * For a page array, a piece comes from the first page in the array
938 * that has not already been fully consumed.
940 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor
*cursor
,
943 struct ceph_msg_data
*data
= cursor
->data
;
946 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGES
);
948 BUG_ON(!data
->pages
);
949 BUG_ON(!data
->length
);
951 cursor
->resid
= min(length
, data
->length
);
952 page_count
= calc_pages_for(data
->alignment
, (u64
)data
->length
);
953 cursor
->page_offset
= data
->alignment
& ~PAGE_MASK
;
954 cursor
->page_index
= 0;
955 BUG_ON(page_count
> (int)USHRT_MAX
);
956 cursor
->page_count
= (unsigned short)page_count
;
957 BUG_ON(length
> SIZE_MAX
- cursor
->page_offset
);
958 cursor
->last_piece
= cursor
->page_offset
+ cursor
->resid
<= PAGE_SIZE
;
962 ceph_msg_data_pages_next(struct ceph_msg_data_cursor
*cursor
,
963 size_t *page_offset
, size_t *length
)
965 struct ceph_msg_data
*data
= cursor
->data
;
967 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGES
);
969 BUG_ON(cursor
->page_index
>= cursor
->page_count
);
970 BUG_ON(cursor
->page_offset
>= PAGE_SIZE
);
972 *page_offset
= cursor
->page_offset
;
973 if (cursor
->last_piece
)
974 *length
= cursor
->resid
;
976 *length
= PAGE_SIZE
- *page_offset
;
978 return data
->pages
[cursor
->page_index
];
981 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor
*cursor
,
984 BUG_ON(cursor
->data
->type
!= CEPH_MSG_DATA_PAGES
);
986 BUG_ON(cursor
->page_offset
+ bytes
> PAGE_SIZE
);
988 /* Advance the cursor page offset */
990 cursor
->resid
-= bytes
;
991 cursor
->page_offset
= (cursor
->page_offset
+ bytes
) & ~PAGE_MASK
;
992 if (!bytes
|| cursor
->page_offset
)
993 return false; /* more bytes to process in the current page */
996 return false; /* no more data */
998 /* Move on to the next page; offset is already at 0 */
1000 BUG_ON(cursor
->page_index
>= cursor
->page_count
);
1001 cursor
->page_index
++;
1002 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
1008 * For a pagelist, a piece is whatever remains to be consumed in the
1009 * first page in the list, or the front of the next page.
1012 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor
*cursor
,
1015 struct ceph_msg_data
*data
= cursor
->data
;
1016 struct ceph_pagelist
*pagelist
;
1019 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
1021 pagelist
= data
->pagelist
;
1025 return; /* pagelist can be assigned but empty */
1027 BUG_ON(list_empty(&pagelist
->head
));
1028 page
= list_first_entry(&pagelist
->head
, struct page
, lru
);
1030 cursor
->resid
= min(length
, pagelist
->length
);
1031 cursor
->page
= page
;
1033 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
1036 static struct page
*
1037 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor
*cursor
,
1038 size_t *page_offset
, size_t *length
)
1040 struct ceph_msg_data
*data
= cursor
->data
;
1041 struct ceph_pagelist
*pagelist
;
1043 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
1045 pagelist
= data
->pagelist
;
1048 BUG_ON(!cursor
->page
);
1049 BUG_ON(cursor
->offset
+ cursor
->resid
!= pagelist
->length
);
1051 /* offset of first page in pagelist is always 0 */
1052 *page_offset
= cursor
->offset
& ~PAGE_MASK
;
1053 if (cursor
->last_piece
)
1054 *length
= cursor
->resid
;
1056 *length
= PAGE_SIZE
- *page_offset
;
1058 return cursor
->page
;
1061 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor
*cursor
,
1064 struct ceph_msg_data
*data
= cursor
->data
;
1065 struct ceph_pagelist
*pagelist
;
1067 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
1069 pagelist
= data
->pagelist
;
1072 BUG_ON(cursor
->offset
+ cursor
->resid
!= pagelist
->length
);
1073 BUG_ON((cursor
->offset
& ~PAGE_MASK
) + bytes
> PAGE_SIZE
);
1075 /* Advance the cursor offset */
1077 cursor
->resid
-= bytes
;
1078 cursor
->offset
+= bytes
;
1079 /* offset of first page in pagelist is always 0 */
1080 if (!bytes
|| cursor
->offset
& ~PAGE_MASK
)
1081 return false; /* more bytes to process in the current page */
1084 return false; /* no more data */
1086 /* Move on to the next page */
1088 BUG_ON(list_is_last(&cursor
->page
->lru
, &pagelist
->head
));
1089 cursor
->page
= list_next_entry(cursor
->page
, lru
);
1090 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
1096 * Message data is handled (sent or received) in pieces, where each
1097 * piece resides on a single page. The network layer might not
1098 * consume an entire piece at once. A data item's cursor keeps
1099 * track of which piece is next to process and how much remains to
1100 * be processed in that piece. It also tracks whether the current
1101 * piece is the last one in the data item.
1103 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor
*cursor
)
1105 size_t length
= cursor
->total_resid
;
1107 switch (cursor
->data
->type
) {
1108 case CEPH_MSG_DATA_PAGELIST
:
1109 ceph_msg_data_pagelist_cursor_init(cursor
, length
);
1111 case CEPH_MSG_DATA_PAGES
:
1112 ceph_msg_data_pages_cursor_init(cursor
, length
);
1115 case CEPH_MSG_DATA_BIO
:
1116 ceph_msg_data_bio_cursor_init(cursor
, length
);
1118 #endif /* CONFIG_BLOCK */
1119 case CEPH_MSG_DATA_NONE
:
1124 cursor
->need_crc
= true;
1127 static void ceph_msg_data_cursor_init(struct ceph_msg
*msg
, size_t length
)
1129 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
1130 struct ceph_msg_data
*data
;
1133 BUG_ON(length
> msg
->data_length
);
1134 BUG_ON(list_empty(&msg
->data
));
1136 cursor
->data_head
= &msg
->data
;
1137 cursor
->total_resid
= length
;
1138 data
= list_first_entry(&msg
->data
, struct ceph_msg_data
, links
);
1139 cursor
->data
= data
;
1141 __ceph_msg_data_cursor_init(cursor
);
1145 * Return the page containing the next piece to process for a given
1146 * data item, and supply the page offset and length of that piece.
1147 * Indicate whether this is the last piece in this data item.
1149 static struct page
*ceph_msg_data_next(struct ceph_msg_data_cursor
*cursor
,
1150 size_t *page_offset
, size_t *length
,
1155 switch (cursor
->data
->type
) {
1156 case CEPH_MSG_DATA_PAGELIST
:
1157 page
= ceph_msg_data_pagelist_next(cursor
, page_offset
, length
);
1159 case CEPH_MSG_DATA_PAGES
:
1160 page
= ceph_msg_data_pages_next(cursor
, page_offset
, length
);
1163 case CEPH_MSG_DATA_BIO
:
1164 page
= ceph_msg_data_bio_next(cursor
, page_offset
, length
);
1166 #endif /* CONFIG_BLOCK */
1167 case CEPH_MSG_DATA_NONE
:
1173 BUG_ON(*page_offset
+ *length
> PAGE_SIZE
);
1176 *last_piece
= cursor
->last_piece
;
1182 * Returns true if the result moves the cursor on to the next piece
1185 static void ceph_msg_data_advance(struct ceph_msg_data_cursor
*cursor
,
1190 BUG_ON(bytes
> cursor
->resid
);
1191 switch (cursor
->data
->type
) {
1192 case CEPH_MSG_DATA_PAGELIST
:
1193 new_piece
= ceph_msg_data_pagelist_advance(cursor
, bytes
);
1195 case CEPH_MSG_DATA_PAGES
:
1196 new_piece
= ceph_msg_data_pages_advance(cursor
, bytes
);
1199 case CEPH_MSG_DATA_BIO
:
1200 new_piece
= ceph_msg_data_bio_advance(cursor
, bytes
);
1202 #endif /* CONFIG_BLOCK */
1203 case CEPH_MSG_DATA_NONE
:
1208 cursor
->total_resid
-= bytes
;
1210 if (!cursor
->resid
&& cursor
->total_resid
) {
1211 WARN_ON(!cursor
->last_piece
);
1212 BUG_ON(list_is_last(&cursor
->data
->links
, cursor
->data_head
));
1213 cursor
->data
= list_next_entry(cursor
->data
, links
);
1214 __ceph_msg_data_cursor_init(cursor
);
1217 cursor
->need_crc
= new_piece
;
1220 static size_t sizeof_footer(struct ceph_connection
*con
)
1222 return (con
->peer_features
& CEPH_FEATURE_MSG_AUTH
) ?
1223 sizeof(struct ceph_msg_footer
) :
1224 sizeof(struct ceph_msg_footer_old
);
1227 static void prepare_message_data(struct ceph_msg
*msg
, u32 data_len
)
1232 /* Initialize data cursor */
1234 ceph_msg_data_cursor_init(msg
, (size_t)data_len
);
1238 * Prepare footer for currently outgoing message, and finish things
1239 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1241 static void prepare_write_message_footer(struct ceph_connection
*con
)
1243 struct ceph_msg
*m
= con
->out_msg
;
1245 m
->footer
.flags
|= CEPH_MSG_FOOTER_COMPLETE
;
1247 dout("prepare_write_message_footer %p\n", con
);
1248 con_out_kvec_add(con
, sizeof_footer(con
), &m
->footer
);
1249 if (con
->peer_features
& CEPH_FEATURE_MSG_AUTH
) {
1250 if (con
->ops
->sign_message
)
1251 con
->ops
->sign_message(m
);
1255 m
->old_footer
.flags
= m
->footer
.flags
;
1257 con
->out_more
= m
->more_to_follow
;
1258 con
->out_msg_done
= true;
1262 * Prepare headers for the next outgoing message.
1264 static void prepare_write_message(struct ceph_connection
*con
)
1269 con_out_kvec_reset(con
);
1270 con
->out_msg_done
= false;
1272 /* Sneak an ack in there first? If we can get it into the same
1273 * TCP packet that's a good thing. */
1274 if (con
->in_seq
> con
->in_seq_acked
) {
1275 con
->in_seq_acked
= con
->in_seq
;
1276 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
1277 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1278 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1279 &con
->out_temp_ack
);
1282 BUG_ON(list_empty(&con
->out_queue
));
1283 m
= list_first_entry(&con
->out_queue
, struct ceph_msg
, list_head
);
1285 BUG_ON(m
->con
!= con
);
1287 /* put message on sent list */
1289 list_move_tail(&m
->list_head
, &con
->out_sent
);
1292 * only assign outgoing seq # if we haven't sent this message
1293 * yet. if it is requeued, resend with it's original seq.
1295 if (m
->needs_out_seq
) {
1296 m
->hdr
.seq
= cpu_to_le64(++con
->out_seq
);
1297 m
->needs_out_seq
= false;
1299 if (con
->ops
->reencode_message
)
1300 con
->ops
->reencode_message(m
);
1303 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1304 m
, con
->out_seq
, le16_to_cpu(m
->hdr
.type
),
1305 le32_to_cpu(m
->hdr
.front_len
), le32_to_cpu(m
->hdr
.middle_len
),
1307 WARN_ON(m
->front
.iov_len
!= le32_to_cpu(m
->hdr
.front_len
));
1308 WARN_ON(m
->data_length
!= le32_to_cpu(m
->hdr
.data_len
));
1310 /* tag + hdr + front + middle */
1311 con_out_kvec_add(con
, sizeof (tag_msg
), &tag_msg
);
1312 con_out_kvec_add(con
, sizeof(con
->out_hdr
), &con
->out_hdr
);
1313 con_out_kvec_add(con
, m
->front
.iov_len
, m
->front
.iov_base
);
1316 con_out_kvec_add(con
, m
->middle
->vec
.iov_len
,
1317 m
->middle
->vec
.iov_base
);
1319 /* fill in hdr crc and finalize hdr */
1320 crc
= crc32c(0, &m
->hdr
, offsetof(struct ceph_msg_header
, crc
));
1321 con
->out_msg
->hdr
.crc
= cpu_to_le32(crc
);
1322 memcpy(&con
->out_hdr
, &con
->out_msg
->hdr
, sizeof(con
->out_hdr
));
1324 /* fill in front and middle crc, footer */
1325 crc
= crc32c(0, m
->front
.iov_base
, m
->front
.iov_len
);
1326 con
->out_msg
->footer
.front_crc
= cpu_to_le32(crc
);
1328 crc
= crc32c(0, m
->middle
->vec
.iov_base
,
1329 m
->middle
->vec
.iov_len
);
1330 con
->out_msg
->footer
.middle_crc
= cpu_to_le32(crc
);
1332 con
->out_msg
->footer
.middle_crc
= 0;
1333 dout("%s front_crc %u middle_crc %u\n", __func__
,
1334 le32_to_cpu(con
->out_msg
->footer
.front_crc
),
1335 le32_to_cpu(con
->out_msg
->footer
.middle_crc
));
1336 con
->out_msg
->footer
.flags
= 0;
1338 /* is there a data payload? */
1339 con
->out_msg
->footer
.data_crc
= 0;
1340 if (m
->data_length
) {
1341 prepare_message_data(con
->out_msg
, m
->data_length
);
1342 con
->out_more
= 1; /* data + footer will follow */
1344 /* no, queue up footer too and be done */
1345 prepare_write_message_footer(con
);
1348 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1354 static void prepare_write_ack(struct ceph_connection
*con
)
1356 dout("prepare_write_ack %p %llu -> %llu\n", con
,
1357 con
->in_seq_acked
, con
->in_seq
);
1358 con
->in_seq_acked
= con
->in_seq
;
1360 con_out_kvec_reset(con
);
1362 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
1364 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1365 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1366 &con
->out_temp_ack
);
1368 con
->out_more
= 1; /* more will follow.. eventually.. */
1369 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1373 * Prepare to share the seq during handshake
1375 static void prepare_write_seq(struct ceph_connection
*con
)
1377 dout("prepare_write_seq %p %llu -> %llu\n", con
,
1378 con
->in_seq_acked
, con
->in_seq
);
1379 con
->in_seq_acked
= con
->in_seq
;
1381 con_out_kvec_reset(con
);
1383 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1384 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1385 &con
->out_temp_ack
);
1387 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1391 * Prepare to write keepalive byte.
1393 static void prepare_write_keepalive(struct ceph_connection
*con
)
1395 dout("prepare_write_keepalive %p\n", con
);
1396 con_out_kvec_reset(con
);
1397 if (con
->peer_features
& CEPH_FEATURE_MSGR_KEEPALIVE2
) {
1398 struct timespec now
;
1400 ktime_get_real_ts(&now
);
1401 con_out_kvec_add(con
, sizeof(tag_keepalive2
), &tag_keepalive2
);
1402 ceph_encode_timespec(&con
->out_temp_keepalive2
, &now
);
1403 con_out_kvec_add(con
, sizeof(con
->out_temp_keepalive2
),
1404 &con
->out_temp_keepalive2
);
1406 con_out_kvec_add(con
, sizeof(tag_keepalive
), &tag_keepalive
);
1408 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1412 * Connection negotiation.
1415 static int get_connect_authorizer(struct ceph_connection
*con
)
1417 struct ceph_auth_handshake
*auth
;
1420 if (!con
->ops
->get_authorizer
) {
1422 con
->out_connect
.authorizer_protocol
= CEPH_AUTH_UNKNOWN
;
1423 con
->out_connect
.authorizer_len
= 0;
1427 auth
= con
->ops
->get_authorizer(con
, &auth_proto
, con
->auth_retry
);
1429 return PTR_ERR(auth
);
1432 con
->out_connect
.authorizer_protocol
= cpu_to_le32(auth_proto
);
1433 con
->out_connect
.authorizer_len
= cpu_to_le32(auth
->authorizer_buf_len
);
1438 * We connected to a peer and are saying hello.
1440 static void prepare_write_banner(struct ceph_connection
*con
)
1442 con_out_kvec_add(con
, strlen(CEPH_BANNER
), CEPH_BANNER
);
1443 con_out_kvec_add(con
, sizeof (con
->msgr
->my_enc_addr
),
1444 &con
->msgr
->my_enc_addr
);
1447 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1450 static void __prepare_write_connect(struct ceph_connection
*con
)
1452 con_out_kvec_add(con
, sizeof(con
->out_connect
), &con
->out_connect
);
1454 con_out_kvec_add(con
, con
->auth
->authorizer_buf_len
,
1455 con
->auth
->authorizer_buf
);
1458 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1461 static int prepare_write_connect(struct ceph_connection
*con
)
1463 unsigned int global_seq
= get_global_seq(con
->msgr
, 0);
1467 switch (con
->peer_name
.type
) {
1468 case CEPH_ENTITY_TYPE_MON
:
1469 proto
= CEPH_MONC_PROTOCOL
;
1471 case CEPH_ENTITY_TYPE_OSD
:
1472 proto
= CEPH_OSDC_PROTOCOL
;
1474 case CEPH_ENTITY_TYPE_MDS
:
1475 proto
= CEPH_MDSC_PROTOCOL
;
1481 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con
,
1482 con
->connect_seq
, global_seq
, proto
);
1484 con
->out_connect
.features
=
1485 cpu_to_le64(from_msgr(con
->msgr
)->supported_features
);
1486 con
->out_connect
.host_type
= cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT
);
1487 con
->out_connect
.connect_seq
= cpu_to_le32(con
->connect_seq
);
1488 con
->out_connect
.global_seq
= cpu_to_le32(global_seq
);
1489 con
->out_connect
.protocol_version
= cpu_to_le32(proto
);
1490 con
->out_connect
.flags
= 0;
1492 ret
= get_connect_authorizer(con
);
1496 __prepare_write_connect(con
);
1501 * write as much of pending kvecs to the socket as we can.
1503 * 0 -> socket full, but more to do
1506 static int write_partial_kvec(struct ceph_connection
*con
)
1510 dout("write_partial_kvec %p %d left\n", con
, con
->out_kvec_bytes
);
1511 while (con
->out_kvec_bytes
> 0) {
1512 ret
= ceph_tcp_sendmsg(con
->sock
, con
->out_kvec_cur
,
1513 con
->out_kvec_left
, con
->out_kvec_bytes
,
1517 con
->out_kvec_bytes
-= ret
;
1518 if (con
->out_kvec_bytes
== 0)
1521 /* account for full iov entries consumed */
1522 while (ret
>= con
->out_kvec_cur
->iov_len
) {
1523 BUG_ON(!con
->out_kvec_left
);
1524 ret
-= con
->out_kvec_cur
->iov_len
;
1525 con
->out_kvec_cur
++;
1526 con
->out_kvec_left
--;
1528 /* and for a partially-consumed entry */
1530 con
->out_kvec_cur
->iov_len
-= ret
;
1531 con
->out_kvec_cur
->iov_base
+= ret
;
1534 con
->out_kvec_left
= 0;
1537 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con
,
1538 con
->out_kvec_bytes
, con
->out_kvec_left
, ret
);
1539 return ret
; /* done! */
1542 static u32
ceph_crc32c_page(u32 crc
, struct page
*page
,
1543 unsigned int page_offset
,
1544 unsigned int length
)
1549 BUG_ON(kaddr
== NULL
);
1550 crc
= crc32c(crc
, kaddr
+ page_offset
, length
);
1556 * Write as much message data payload as we can. If we finish, queue
1558 * 1 -> done, footer is now queued in out_kvec[].
1559 * 0 -> socket full, but more to do
1562 static int write_partial_message_data(struct ceph_connection
*con
)
1564 struct ceph_msg
*msg
= con
->out_msg
;
1565 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
1566 bool do_datacrc
= !ceph_test_opt(from_msgr(con
->msgr
), NOCRC
);
1569 dout("%s %p msg %p\n", __func__
, con
, msg
);
1571 if (list_empty(&msg
->data
))
1575 * Iterate through each page that contains data to be
1576 * written, and send as much as possible for each.
1578 * If we are calculating the data crc (the default), we will
1579 * need to map the page. If we have no pages, they have
1580 * been revoked, so use the zero page.
1582 crc
= do_datacrc
? le32_to_cpu(msg
->footer
.data_crc
) : 0;
1583 while (cursor
->resid
) {
1590 page
= ceph_msg_data_next(cursor
, &page_offset
, &length
,
1592 ret
= ceph_tcp_sendpage(con
->sock
, page
, page_offset
,
1593 length
, !last_piece
);
1596 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1600 if (do_datacrc
&& cursor
->need_crc
)
1601 crc
= ceph_crc32c_page(crc
, page
, page_offset
, length
);
1602 ceph_msg_data_advance(cursor
, (size_t)ret
);
1605 dout("%s %p msg %p done\n", __func__
, con
, msg
);
1607 /* prepare and queue up footer, too */
1609 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1611 msg
->footer
.flags
|= CEPH_MSG_FOOTER_NOCRC
;
1612 con_out_kvec_reset(con
);
1613 prepare_write_message_footer(con
);
1615 return 1; /* must return > 0 to indicate success */
1621 static int write_partial_skip(struct ceph_connection
*con
)
1625 dout("%s %p %d left\n", __func__
, con
, con
->out_skip
);
1626 while (con
->out_skip
> 0) {
1627 size_t size
= min(con
->out_skip
, (int) PAGE_SIZE
);
1629 ret
= ceph_tcp_sendpage(con
->sock
, zero_page
, 0, size
, true);
1632 con
->out_skip
-= ret
;
1640 * Prepare to read connection handshake, or an ack.
1642 static void prepare_read_banner(struct ceph_connection
*con
)
1644 dout("prepare_read_banner %p\n", con
);
1645 con
->in_base_pos
= 0;
1648 static void prepare_read_connect(struct ceph_connection
*con
)
1650 dout("prepare_read_connect %p\n", con
);
1651 con
->in_base_pos
= 0;
1654 static void prepare_read_ack(struct ceph_connection
*con
)
1656 dout("prepare_read_ack %p\n", con
);
1657 con
->in_base_pos
= 0;
1660 static void prepare_read_seq(struct ceph_connection
*con
)
1662 dout("prepare_read_seq %p\n", con
);
1663 con
->in_base_pos
= 0;
1664 con
->in_tag
= CEPH_MSGR_TAG_SEQ
;
1667 static void prepare_read_tag(struct ceph_connection
*con
)
1669 dout("prepare_read_tag %p\n", con
);
1670 con
->in_base_pos
= 0;
1671 con
->in_tag
= CEPH_MSGR_TAG_READY
;
1674 static void prepare_read_keepalive_ack(struct ceph_connection
*con
)
1676 dout("prepare_read_keepalive_ack %p\n", con
);
1677 con
->in_base_pos
= 0;
1681 * Prepare to read a message.
1683 static int prepare_read_message(struct ceph_connection
*con
)
1685 dout("prepare_read_message %p\n", con
);
1686 BUG_ON(con
->in_msg
!= NULL
);
1687 con
->in_base_pos
= 0;
1688 con
->in_front_crc
= con
->in_middle_crc
= con
->in_data_crc
= 0;
1693 static int read_partial(struct ceph_connection
*con
,
1694 int end
, int size
, void *object
)
1696 while (con
->in_base_pos
< end
) {
1697 int left
= end
- con
->in_base_pos
;
1698 int have
= size
- left
;
1699 int ret
= ceph_tcp_recvmsg(con
->sock
, object
+ have
, left
);
1702 con
->in_base_pos
+= ret
;
1709 * Read all or part of the connect-side handshake on a new connection
1711 static int read_partial_banner(struct ceph_connection
*con
)
1717 dout("read_partial_banner %p at %d\n", con
, con
->in_base_pos
);
1720 size
= strlen(CEPH_BANNER
);
1722 ret
= read_partial(con
, end
, size
, con
->in_banner
);
1726 size
= sizeof (con
->actual_peer_addr
);
1728 ret
= read_partial(con
, end
, size
, &con
->actual_peer_addr
);
1732 size
= sizeof (con
->peer_addr_for_me
);
1734 ret
= read_partial(con
, end
, size
, &con
->peer_addr_for_me
);
1742 static int read_partial_connect(struct ceph_connection
*con
)
1748 dout("read_partial_connect %p at %d\n", con
, con
->in_base_pos
);
1750 size
= sizeof (con
->in_reply
);
1752 ret
= read_partial(con
, end
, size
, &con
->in_reply
);
1757 size
= le32_to_cpu(con
->in_reply
.authorizer_len
);
1759 ret
= read_partial(con
, end
, size
,
1760 con
->auth
->authorizer_reply_buf
);
1765 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1766 con
, (int)con
->in_reply
.tag
,
1767 le32_to_cpu(con
->in_reply
.connect_seq
),
1768 le32_to_cpu(con
->in_reply
.global_seq
));
1774 * Verify the hello banner looks okay.
1776 static int verify_hello(struct ceph_connection
*con
)
1778 if (memcmp(con
->in_banner
, CEPH_BANNER
, strlen(CEPH_BANNER
))) {
1779 pr_err("connect to %s got bad banner\n",
1780 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1781 con
->error_msg
= "protocol error, bad banner";
1787 static bool addr_is_blank(struct sockaddr_storage
*ss
)
1789 struct in_addr
*addr
= &((struct sockaddr_in
*)ss
)->sin_addr
;
1790 struct in6_addr
*addr6
= &((struct sockaddr_in6
*)ss
)->sin6_addr
;
1792 switch (ss
->ss_family
) {
1794 return addr
->s_addr
== htonl(INADDR_ANY
);
1796 return ipv6_addr_any(addr6
);
1802 static int addr_port(struct sockaddr_storage
*ss
)
1804 switch (ss
->ss_family
) {
1806 return ntohs(((struct sockaddr_in
*)ss
)->sin_port
);
1808 return ntohs(((struct sockaddr_in6
*)ss
)->sin6_port
);
1813 static void addr_set_port(struct sockaddr_storage
*ss
, int p
)
1815 switch (ss
->ss_family
) {
1817 ((struct sockaddr_in
*)ss
)->sin_port
= htons(p
);
1820 ((struct sockaddr_in6
*)ss
)->sin6_port
= htons(p
);
1826 * Unlike other *_pton function semantics, zero indicates success.
1828 static int ceph_pton(const char *str
, size_t len
, struct sockaddr_storage
*ss
,
1829 char delim
, const char **ipend
)
1831 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
1832 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
1834 memset(ss
, 0, sizeof(*ss
));
1836 if (in4_pton(str
, len
, (u8
*)&in4
->sin_addr
.s_addr
, delim
, ipend
)) {
1837 ss
->ss_family
= AF_INET
;
1841 if (in6_pton(str
, len
, (u8
*)&in6
->sin6_addr
.s6_addr
, delim
, ipend
)) {
1842 ss
->ss_family
= AF_INET6
;
1850 * Extract hostname string and resolve using kernel DNS facility.
1852 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1853 static int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1854 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1856 const char *end
, *delim_p
;
1857 char *colon_p
, *ip_addr
= NULL
;
1861 * The end of the hostname occurs immediately preceding the delimiter or
1862 * the port marker (':') where the delimiter takes precedence.
1864 delim_p
= memchr(name
, delim
, namelen
);
1865 colon_p
= memchr(name
, ':', namelen
);
1867 if (delim_p
&& colon_p
)
1868 end
= delim_p
< colon_p
? delim_p
: colon_p
;
1869 else if (!delim_p
&& colon_p
)
1873 if (!end
) /* case: hostname:/ */
1874 end
= name
+ namelen
;
1880 /* do dns_resolve upcall */
1881 ip_len
= dns_query(NULL
, name
, end
- name
, NULL
, &ip_addr
, NULL
);
1883 ret
= ceph_pton(ip_addr
, ip_len
, ss
, -1, NULL
);
1891 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end
- name
), name
,
1892 ret
, ret
? "failed" : ceph_pr_addr(ss
));
1897 static inline int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1898 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1905 * Parse a server name (IP or hostname). If a valid IP address is not found
1906 * then try to extract a hostname to resolve using userspace DNS upcall.
1908 static int ceph_parse_server_name(const char *name
, size_t namelen
,
1909 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1913 ret
= ceph_pton(name
, namelen
, ss
, delim
, ipend
);
1915 ret
= ceph_dns_resolve_name(name
, namelen
, ss
, delim
, ipend
);
1921 * Parse an ip[:port] list into an addr array. Use the default
1922 * monitor port if a port isn't specified.
1924 int ceph_parse_ips(const char *c
, const char *end
,
1925 struct ceph_entity_addr
*addr
,
1926 int max_count
, int *count
)
1928 int i
, ret
= -EINVAL
;
1931 dout("parse_ips on '%.*s'\n", (int)(end
-c
), c
);
1932 for (i
= 0; i
< max_count
; i
++) {
1934 struct sockaddr_storage
*ss
= &addr
[i
].in_addr
;
1943 ret
= ceph_parse_server_name(p
, end
- p
, ss
, delim
, &ipend
);
1952 dout("missing matching ']'\n");
1959 if (p
< end
&& *p
== ':') {
1962 while (p
< end
&& *p
>= '0' && *p
<= '9') {
1963 port
= (port
* 10) + (*p
- '0');
1967 port
= CEPH_MON_PORT
;
1968 else if (port
> 65535)
1971 port
= CEPH_MON_PORT
;
1974 addr_set_port(ss
, port
);
1976 dout("parse_ips got %s\n", ceph_pr_addr(ss
));
1993 pr_err("parse_ips bad ip '%.*s'\n", (int)(end
- c
), c
);
1996 EXPORT_SYMBOL(ceph_parse_ips
);
1998 static int process_banner(struct ceph_connection
*con
)
2000 dout("process_banner on %p\n", con
);
2002 if (verify_hello(con
) < 0)
2005 ceph_decode_addr(&con
->actual_peer_addr
);
2006 ceph_decode_addr(&con
->peer_addr_for_me
);
2009 * Make sure the other end is who we wanted. note that the other
2010 * end may not yet know their ip address, so if it's 0.0.0.0, give
2011 * them the benefit of the doubt.
2013 if (memcmp(&con
->peer_addr
, &con
->actual_peer_addr
,
2014 sizeof(con
->peer_addr
)) != 0 &&
2015 !(addr_is_blank(&con
->actual_peer_addr
.in_addr
) &&
2016 con
->actual_peer_addr
.nonce
== con
->peer_addr
.nonce
)) {
2017 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
2018 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2019 (int)le32_to_cpu(con
->peer_addr
.nonce
),
2020 ceph_pr_addr(&con
->actual_peer_addr
.in_addr
),
2021 (int)le32_to_cpu(con
->actual_peer_addr
.nonce
));
2022 con
->error_msg
= "wrong peer at address";
2027 * did we learn our address?
2029 if (addr_is_blank(&con
->msgr
->inst
.addr
.in_addr
)) {
2030 int port
= addr_port(&con
->msgr
->inst
.addr
.in_addr
);
2032 memcpy(&con
->msgr
->inst
.addr
.in_addr
,
2033 &con
->peer_addr_for_me
.in_addr
,
2034 sizeof(con
->peer_addr_for_me
.in_addr
));
2035 addr_set_port(&con
->msgr
->inst
.addr
.in_addr
, port
);
2036 encode_my_addr(con
->msgr
);
2037 dout("process_banner learned my addr is %s\n",
2038 ceph_pr_addr(&con
->msgr
->inst
.addr
.in_addr
));
2044 static int process_connect(struct ceph_connection
*con
)
2046 u64 sup_feat
= from_msgr(con
->msgr
)->supported_features
;
2047 u64 req_feat
= from_msgr(con
->msgr
)->required_features
;
2048 u64 server_feat
= le64_to_cpu(con
->in_reply
.features
);
2051 dout("process_connect on %p tag %d\n", con
, (int)con
->in_tag
);
2055 * Any connection that defines ->get_authorizer()
2056 * should also define ->add_authorizer_challenge() and
2057 * ->verify_authorizer_reply().
2059 * See get_connect_authorizer().
2061 if (con
->in_reply
.tag
== CEPH_MSGR_TAG_CHALLENGE_AUTHORIZER
) {
2062 ret
= con
->ops
->add_authorizer_challenge(
2063 con
, con
->auth
->authorizer_reply_buf
,
2064 le32_to_cpu(con
->in_reply
.authorizer_len
));
2068 con_out_kvec_reset(con
);
2069 __prepare_write_connect(con
);
2070 prepare_read_connect(con
);
2074 ret
= con
->ops
->verify_authorizer_reply(con
);
2076 con
->error_msg
= "bad authorize reply";
2081 switch (con
->in_reply
.tag
) {
2082 case CEPH_MSGR_TAG_FEATURES
:
2083 pr_err("%s%lld %s feature set mismatch,"
2084 " my %llx < server's %llx, missing %llx\n",
2085 ENTITY_NAME(con
->peer_name
),
2086 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2087 sup_feat
, server_feat
, server_feat
& ~sup_feat
);
2088 con
->error_msg
= "missing required protocol features";
2089 reset_connection(con
);
2092 case CEPH_MSGR_TAG_BADPROTOVER
:
2093 pr_err("%s%lld %s protocol version mismatch,"
2094 " my %d != server's %d\n",
2095 ENTITY_NAME(con
->peer_name
),
2096 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2097 le32_to_cpu(con
->out_connect
.protocol_version
),
2098 le32_to_cpu(con
->in_reply
.protocol_version
));
2099 con
->error_msg
= "protocol version mismatch";
2100 reset_connection(con
);
2103 case CEPH_MSGR_TAG_BADAUTHORIZER
:
2105 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con
,
2107 if (con
->auth_retry
== 2) {
2108 con
->error_msg
= "connect authorization failure";
2111 con_out_kvec_reset(con
);
2112 ret
= prepare_write_connect(con
);
2115 prepare_read_connect(con
);
2118 case CEPH_MSGR_TAG_RESETSESSION
:
2120 * If we connected with a large connect_seq but the peer
2121 * has no record of a session with us (no connection, or
2122 * connect_seq == 0), they will send RESETSESION to indicate
2123 * that they must have reset their session, and may have
2126 dout("process_connect got RESET peer seq %u\n",
2127 le32_to_cpu(con
->in_reply
.connect_seq
));
2128 pr_err("%s%lld %s connection reset\n",
2129 ENTITY_NAME(con
->peer_name
),
2130 ceph_pr_addr(&con
->peer_addr
.in_addr
));
2131 reset_connection(con
);
2132 con_out_kvec_reset(con
);
2133 ret
= prepare_write_connect(con
);
2136 prepare_read_connect(con
);
2138 /* Tell ceph about it. */
2139 mutex_unlock(&con
->mutex
);
2140 pr_info("reset on %s%lld\n", ENTITY_NAME(con
->peer_name
));
2141 if (con
->ops
->peer_reset
)
2142 con
->ops
->peer_reset(con
);
2143 mutex_lock(&con
->mutex
);
2144 if (con
->state
!= CON_STATE_NEGOTIATING
)
2148 case CEPH_MSGR_TAG_RETRY_SESSION
:
2150 * If we sent a smaller connect_seq than the peer has, try
2151 * again with a larger value.
2153 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2154 le32_to_cpu(con
->out_connect
.connect_seq
),
2155 le32_to_cpu(con
->in_reply
.connect_seq
));
2156 con
->connect_seq
= le32_to_cpu(con
->in_reply
.connect_seq
);
2157 con_out_kvec_reset(con
);
2158 ret
= prepare_write_connect(con
);
2161 prepare_read_connect(con
);
2164 case CEPH_MSGR_TAG_RETRY_GLOBAL
:
2166 * If we sent a smaller global_seq than the peer has, try
2167 * again with a larger value.
2169 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2170 con
->peer_global_seq
,
2171 le32_to_cpu(con
->in_reply
.global_seq
));
2172 get_global_seq(con
->msgr
,
2173 le32_to_cpu(con
->in_reply
.global_seq
));
2174 con_out_kvec_reset(con
);
2175 ret
= prepare_write_connect(con
);
2178 prepare_read_connect(con
);
2181 case CEPH_MSGR_TAG_SEQ
:
2182 case CEPH_MSGR_TAG_READY
:
2183 if (req_feat
& ~server_feat
) {
2184 pr_err("%s%lld %s protocol feature mismatch,"
2185 " my required %llx > server's %llx, need %llx\n",
2186 ENTITY_NAME(con
->peer_name
),
2187 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2188 req_feat
, server_feat
, req_feat
& ~server_feat
);
2189 con
->error_msg
= "missing required protocol features";
2190 reset_connection(con
);
2194 WARN_ON(con
->state
!= CON_STATE_NEGOTIATING
);
2195 con
->state
= CON_STATE_OPEN
;
2196 con
->auth_retry
= 0; /* we authenticated; clear flag */
2197 con
->peer_global_seq
= le32_to_cpu(con
->in_reply
.global_seq
);
2199 con
->peer_features
= server_feat
;
2200 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2201 con
->peer_global_seq
,
2202 le32_to_cpu(con
->in_reply
.connect_seq
),
2204 WARN_ON(con
->connect_seq
!=
2205 le32_to_cpu(con
->in_reply
.connect_seq
));
2207 if (con
->in_reply
.flags
& CEPH_MSG_CONNECT_LOSSY
)
2208 con_flag_set(con
, CON_FLAG_LOSSYTX
);
2210 con
->delay
= 0; /* reset backoff memory */
2212 if (con
->in_reply
.tag
== CEPH_MSGR_TAG_SEQ
) {
2213 prepare_write_seq(con
);
2214 prepare_read_seq(con
);
2216 prepare_read_tag(con
);
2220 case CEPH_MSGR_TAG_WAIT
:
2222 * If there is a connection race (we are opening
2223 * connections to each other), one of us may just have
2224 * to WAIT. This shouldn't happen if we are the
2227 con
->error_msg
= "protocol error, got WAIT as client";
2231 con
->error_msg
= "protocol error, garbage tag during connect";
2239 * read (part of) an ack
2241 static int read_partial_ack(struct ceph_connection
*con
)
2243 int size
= sizeof (con
->in_temp_ack
);
2246 return read_partial(con
, end
, size
, &con
->in_temp_ack
);
2250 * We can finally discard anything that's been acked.
2252 static void process_ack(struct ceph_connection
*con
)
2255 u64 ack
= le64_to_cpu(con
->in_temp_ack
);
2257 bool reconnect
= (con
->in_tag
== CEPH_MSGR_TAG_SEQ
);
2258 struct list_head
*list
= reconnect
? &con
->out_queue
: &con
->out_sent
;
2261 * In the reconnect case, con_fault() has requeued messages
2262 * in out_sent. We should cleanup old messages according to
2263 * the reconnect seq.
2265 while (!list_empty(list
)) {
2266 m
= list_first_entry(list
, struct ceph_msg
, list_head
);
2267 if (reconnect
&& m
->needs_out_seq
)
2269 seq
= le64_to_cpu(m
->hdr
.seq
);
2272 dout("got ack for seq %llu type %d at %p\n", seq
,
2273 le16_to_cpu(m
->hdr
.type
), m
);
2274 m
->ack_stamp
= jiffies
;
2278 prepare_read_tag(con
);
2282 static int read_partial_message_section(struct ceph_connection
*con
,
2283 struct kvec
*section
,
2284 unsigned int sec_len
, u32
*crc
)
2290 while (section
->iov_len
< sec_len
) {
2291 BUG_ON(section
->iov_base
== NULL
);
2292 left
= sec_len
- section
->iov_len
;
2293 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)section
->iov_base
+
2294 section
->iov_len
, left
);
2297 section
->iov_len
+= ret
;
2299 if (section
->iov_len
== sec_len
)
2300 *crc
= crc32c(0, section
->iov_base
, section
->iov_len
);
2305 static int read_partial_msg_data(struct ceph_connection
*con
)
2307 struct ceph_msg
*msg
= con
->in_msg
;
2308 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
2309 bool do_datacrc
= !ceph_test_opt(from_msgr(con
->msgr
), NOCRC
);
2317 if (list_empty(&msg
->data
))
2321 crc
= con
->in_data_crc
;
2322 while (cursor
->resid
) {
2323 page
= ceph_msg_data_next(cursor
, &page_offset
, &length
, NULL
);
2324 ret
= ceph_tcp_recvpage(con
->sock
, page
, page_offset
, length
);
2327 con
->in_data_crc
= crc
;
2333 crc
= ceph_crc32c_page(crc
, page
, page_offset
, ret
);
2334 ceph_msg_data_advance(cursor
, (size_t)ret
);
2337 con
->in_data_crc
= crc
;
2339 return 1; /* must return > 0 to indicate success */
2343 * read (part of) a message.
2345 static int ceph_con_in_msg_alloc(struct ceph_connection
*con
, int *skip
);
2347 static int read_partial_message(struct ceph_connection
*con
)
2349 struct ceph_msg
*m
= con
->in_msg
;
2353 unsigned int front_len
, middle_len
, data_len
;
2354 bool do_datacrc
= !ceph_test_opt(from_msgr(con
->msgr
), NOCRC
);
2355 bool need_sign
= (con
->peer_features
& CEPH_FEATURE_MSG_AUTH
);
2359 dout("read_partial_message con %p msg %p\n", con
, m
);
2362 size
= sizeof (con
->in_hdr
);
2364 ret
= read_partial(con
, end
, size
, &con
->in_hdr
);
2368 crc
= crc32c(0, &con
->in_hdr
, offsetof(struct ceph_msg_header
, crc
));
2369 if (cpu_to_le32(crc
) != con
->in_hdr
.crc
) {
2370 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
2371 crc
, con
->in_hdr
.crc
);
2375 front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
2376 if (front_len
> CEPH_MSG_MAX_FRONT_LEN
)
2378 middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
2379 if (middle_len
> CEPH_MSG_MAX_MIDDLE_LEN
)
2381 data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
2382 if (data_len
> CEPH_MSG_MAX_DATA_LEN
)
2386 seq
= le64_to_cpu(con
->in_hdr
.seq
);
2387 if ((s64
)seq
- (s64
)con
->in_seq
< 1) {
2388 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2389 ENTITY_NAME(con
->peer_name
),
2390 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2391 seq
, con
->in_seq
+ 1);
2392 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
2394 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2396 } else if ((s64
)seq
- (s64
)con
->in_seq
> 1) {
2397 pr_err("read_partial_message bad seq %lld expected %lld\n",
2398 seq
, con
->in_seq
+ 1);
2399 con
->error_msg
= "bad message sequence # for incoming message";
2403 /* allocate message? */
2407 dout("got hdr type %d front %d data %d\n", con
->in_hdr
.type
,
2408 front_len
, data_len
);
2409 ret
= ceph_con_in_msg_alloc(con
, &skip
);
2413 BUG_ON(!con
->in_msg
^ skip
);
2415 /* skip this message */
2416 dout("alloc_msg said skip message\n");
2417 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
2419 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2424 BUG_ON(!con
->in_msg
);
2425 BUG_ON(con
->in_msg
->con
!= con
);
2427 m
->front
.iov_len
= 0; /* haven't read it yet */
2429 m
->middle
->vec
.iov_len
= 0;
2431 /* prepare for data payload, if any */
2434 prepare_message_data(con
->in_msg
, data_len
);
2438 ret
= read_partial_message_section(con
, &m
->front
, front_len
,
2439 &con
->in_front_crc
);
2445 ret
= read_partial_message_section(con
, &m
->middle
->vec
,
2447 &con
->in_middle_crc
);
2454 ret
= read_partial_msg_data(con
);
2460 size
= sizeof_footer(con
);
2462 ret
= read_partial(con
, end
, size
, &m
->footer
);
2467 m
->footer
.flags
= m
->old_footer
.flags
;
2471 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2472 m
, front_len
, m
->footer
.front_crc
, middle_len
,
2473 m
->footer
.middle_crc
, data_len
, m
->footer
.data_crc
);
2476 if (con
->in_front_crc
!= le32_to_cpu(m
->footer
.front_crc
)) {
2477 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2478 m
, con
->in_front_crc
, m
->footer
.front_crc
);
2481 if (con
->in_middle_crc
!= le32_to_cpu(m
->footer
.middle_crc
)) {
2482 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2483 m
, con
->in_middle_crc
, m
->footer
.middle_crc
);
2487 (m
->footer
.flags
& CEPH_MSG_FOOTER_NOCRC
) == 0 &&
2488 con
->in_data_crc
!= le32_to_cpu(m
->footer
.data_crc
)) {
2489 pr_err("read_partial_message %p data crc %u != exp. %u\n", m
,
2490 con
->in_data_crc
, le32_to_cpu(m
->footer
.data_crc
));
2494 if (need_sign
&& con
->ops
->check_message_signature
&&
2495 con
->ops
->check_message_signature(m
)) {
2496 pr_err("read_partial_message %p signature check failed\n", m
);
2500 return 1; /* done! */
2504 * Process message. This happens in the worker thread. The callback should
2505 * be careful not to do anything that waits on other incoming messages or it
2508 static void process_message(struct ceph_connection
*con
)
2510 struct ceph_msg
*msg
= con
->in_msg
;
2512 BUG_ON(con
->in_msg
->con
!= con
);
2515 /* if first message, set peer_name */
2516 if (con
->peer_name
.type
== 0)
2517 con
->peer_name
= msg
->hdr
.src
;
2520 mutex_unlock(&con
->mutex
);
2522 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2523 msg
, le64_to_cpu(msg
->hdr
.seq
),
2524 ENTITY_NAME(msg
->hdr
.src
),
2525 le16_to_cpu(msg
->hdr
.type
),
2526 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2527 le32_to_cpu(msg
->hdr
.front_len
),
2528 le32_to_cpu(msg
->hdr
.data_len
),
2529 con
->in_front_crc
, con
->in_middle_crc
, con
->in_data_crc
);
2530 con
->ops
->dispatch(con
, msg
);
2532 mutex_lock(&con
->mutex
);
2535 static int read_keepalive_ack(struct ceph_connection
*con
)
2537 struct ceph_timespec ceph_ts
;
2538 size_t size
= sizeof(ceph_ts
);
2539 int ret
= read_partial(con
, size
, size
, &ceph_ts
);
2542 ceph_decode_timespec(&con
->last_keepalive_ack
, &ceph_ts
);
2543 prepare_read_tag(con
);
2548 * Write something to the socket. Called in a worker thread when the
2549 * socket appears to be writeable and we have something ready to send.
2551 static int try_write(struct ceph_connection
*con
)
2555 dout("try_write start %p state %lu\n", con
, con
->state
);
2556 if (con
->state
!= CON_STATE_PREOPEN
&&
2557 con
->state
!= CON_STATE_CONNECTING
&&
2558 con
->state
!= CON_STATE_NEGOTIATING
&&
2559 con
->state
!= CON_STATE_OPEN
)
2563 dout("try_write out_kvec_bytes %d\n", con
->out_kvec_bytes
);
2565 /* open the socket first? */
2566 if (con
->state
== CON_STATE_PREOPEN
) {
2568 con
->state
= CON_STATE_CONNECTING
;
2570 con_out_kvec_reset(con
);
2571 prepare_write_banner(con
);
2572 prepare_read_banner(con
);
2574 BUG_ON(con
->in_msg
);
2575 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2576 dout("try_write initiating connect on %p new state %lu\n",
2578 ret
= ceph_tcp_connect(con
);
2580 con
->error_msg
= "connect error";
2588 /* kvec data queued? */
2589 if (con
->out_kvec_left
) {
2590 ret
= write_partial_kvec(con
);
2594 if (con
->out_skip
) {
2595 ret
= write_partial_skip(con
);
2602 if (con
->out_msg_done
) {
2603 ceph_msg_put(con
->out_msg
);
2604 con
->out_msg
= NULL
; /* we're done with this one */
2608 ret
= write_partial_message_data(con
);
2610 goto more_kvec
; /* we need to send the footer, too! */
2614 dout("try_write write_partial_message_data err %d\n",
2621 if (con
->state
== CON_STATE_OPEN
) {
2622 if (con_flag_test_and_clear(con
, CON_FLAG_KEEPALIVE_PENDING
)) {
2623 prepare_write_keepalive(con
);
2626 /* is anything else pending? */
2627 if (!list_empty(&con
->out_queue
)) {
2628 prepare_write_message(con
);
2631 if (con
->in_seq
> con
->in_seq_acked
) {
2632 prepare_write_ack(con
);
2637 /* Nothing to do! */
2638 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
2639 dout("try_write nothing else to write.\n");
2642 dout("try_write done on %p ret %d\n", con
, ret
);
2649 * Read what we can from the socket.
2651 static int try_read(struct ceph_connection
*con
)
2656 dout("try_read start on %p state %lu\n", con
, con
->state
);
2657 if (con
->state
!= CON_STATE_CONNECTING
&&
2658 con
->state
!= CON_STATE_NEGOTIATING
&&
2659 con
->state
!= CON_STATE_OPEN
)
2664 dout("try_read tag %d in_base_pos %d\n", (int)con
->in_tag
,
2667 if (con
->state
== CON_STATE_CONNECTING
) {
2668 dout("try_read connecting\n");
2669 ret
= read_partial_banner(con
);
2672 ret
= process_banner(con
);
2676 con
->state
= CON_STATE_NEGOTIATING
;
2679 * Received banner is good, exchange connection info.
2680 * Do not reset out_kvec, as sending our banner raced
2681 * with receiving peer banner after connect completed.
2683 ret
= prepare_write_connect(con
);
2686 prepare_read_connect(con
);
2688 /* Send connection info before awaiting response */
2692 if (con
->state
== CON_STATE_NEGOTIATING
) {
2693 dout("try_read negotiating\n");
2694 ret
= read_partial_connect(con
);
2697 ret
= process_connect(con
);
2703 WARN_ON(con
->state
!= CON_STATE_OPEN
);
2705 if (con
->in_base_pos
< 0) {
2707 * skipping + discarding content.
2709 * FIXME: there must be a better way to do this!
2711 static char buf
[SKIP_BUF_SIZE
];
2712 int skip
= min((int) sizeof (buf
), -con
->in_base_pos
);
2714 dout("skipping %d / %d bytes\n", skip
, -con
->in_base_pos
);
2715 ret
= ceph_tcp_recvmsg(con
->sock
, buf
, skip
);
2718 con
->in_base_pos
+= ret
;
2719 if (con
->in_base_pos
)
2722 if (con
->in_tag
== CEPH_MSGR_TAG_READY
) {
2726 ret
= ceph_tcp_recvmsg(con
->sock
, &con
->in_tag
, 1);
2729 dout("try_read got tag %d\n", (int)con
->in_tag
);
2730 switch (con
->in_tag
) {
2731 case CEPH_MSGR_TAG_MSG
:
2732 prepare_read_message(con
);
2734 case CEPH_MSGR_TAG_ACK
:
2735 prepare_read_ack(con
);
2737 case CEPH_MSGR_TAG_KEEPALIVE2_ACK
:
2738 prepare_read_keepalive_ack(con
);
2740 case CEPH_MSGR_TAG_CLOSE
:
2741 con_close_socket(con
);
2742 con
->state
= CON_STATE_CLOSED
;
2748 if (con
->in_tag
== CEPH_MSGR_TAG_MSG
) {
2749 ret
= read_partial_message(con
);
2753 con
->error_msg
= "bad crc/signature";
2759 con
->error_msg
= "io error";
2764 if (con
->in_tag
== CEPH_MSGR_TAG_READY
)
2766 process_message(con
);
2767 if (con
->state
== CON_STATE_OPEN
)
2768 prepare_read_tag(con
);
2771 if (con
->in_tag
== CEPH_MSGR_TAG_ACK
||
2772 con
->in_tag
== CEPH_MSGR_TAG_SEQ
) {
2774 * the final handshake seq exchange is semantically
2775 * equivalent to an ACK
2777 ret
= read_partial_ack(con
);
2783 if (con
->in_tag
== CEPH_MSGR_TAG_KEEPALIVE2_ACK
) {
2784 ret
= read_keepalive_ack(con
);
2791 dout("try_read done on %p ret %d\n", con
, ret
);
2795 pr_err("try_read bad con->in_tag = %d\n", (int)con
->in_tag
);
2796 con
->error_msg
= "protocol error, garbage tag";
2803 * Atomically queue work on a connection after the specified delay.
2804 * Bump @con reference to avoid races with connection teardown.
2805 * Returns 0 if work was queued, or an error code otherwise.
2807 static int queue_con_delay(struct ceph_connection
*con
, unsigned long delay
)
2809 if (!con
->ops
->get(con
)) {
2810 dout("%s %p ref count 0\n", __func__
, con
);
2814 if (!queue_delayed_work(ceph_msgr_wq
, &con
->work
, delay
)) {
2815 dout("%s %p - already queued\n", __func__
, con
);
2820 dout("%s %p %lu\n", __func__
, con
, delay
);
2824 static void queue_con(struct ceph_connection
*con
)
2826 (void) queue_con_delay(con
, 0);
2829 static void cancel_con(struct ceph_connection
*con
)
2831 if (cancel_delayed_work(&con
->work
)) {
2832 dout("%s %p\n", __func__
, con
);
2837 static bool con_sock_closed(struct ceph_connection
*con
)
2839 if (!con_flag_test_and_clear(con
, CON_FLAG_SOCK_CLOSED
))
2843 case CON_STATE_ ## x: \
2844 con->error_msg = "socket closed (con state " #x ")"; \
2847 switch (con
->state
) {
2855 pr_warn("%s con %p unrecognized state %lu\n",
2856 __func__
, con
, con
->state
);
2857 con
->error_msg
= "unrecognized con state";
2866 static bool con_backoff(struct ceph_connection
*con
)
2870 if (!con_flag_test_and_clear(con
, CON_FLAG_BACKOFF
))
2873 ret
= queue_con_delay(con
, round_jiffies_relative(con
->delay
));
2875 dout("%s: con %p FAILED to back off %lu\n", __func__
,
2877 BUG_ON(ret
== -ENOENT
);
2878 con_flag_set(con
, CON_FLAG_BACKOFF
);
2884 /* Finish fault handling; con->mutex must *not* be held here */
2886 static void con_fault_finish(struct ceph_connection
*con
)
2888 dout("%s %p\n", __func__
, con
);
2891 * in case we faulted due to authentication, invalidate our
2892 * current tickets so that we can get new ones.
2894 if (con
->auth_retry
) {
2895 dout("auth_retry %d, invalidating\n", con
->auth_retry
);
2896 if (con
->ops
->invalidate_authorizer
)
2897 con
->ops
->invalidate_authorizer(con
);
2898 con
->auth_retry
= 0;
2901 if (con
->ops
->fault
)
2902 con
->ops
->fault(con
);
2906 * Do some work on a connection. Drop a connection ref when we're done.
2908 static void ceph_con_workfn(struct work_struct
*work
)
2910 struct ceph_connection
*con
= container_of(work
, struct ceph_connection
,
2914 mutex_lock(&con
->mutex
);
2918 if ((fault
= con_sock_closed(con
))) {
2919 dout("%s: con %p SOCK_CLOSED\n", __func__
, con
);
2922 if (con_backoff(con
)) {
2923 dout("%s: con %p BACKOFF\n", __func__
, con
);
2926 if (con
->state
== CON_STATE_STANDBY
) {
2927 dout("%s: con %p STANDBY\n", __func__
, con
);
2930 if (con
->state
== CON_STATE_CLOSED
) {
2931 dout("%s: con %p CLOSED\n", __func__
, con
);
2935 if (con
->state
== CON_STATE_PREOPEN
) {
2936 dout("%s: con %p PREOPEN\n", __func__
, con
);
2940 ret
= try_read(con
);
2944 if (!con
->error_msg
)
2945 con
->error_msg
= "socket error on read";
2950 ret
= try_write(con
);
2954 if (!con
->error_msg
)
2955 con
->error_msg
= "socket error on write";
2959 break; /* If we make it to here, we're done */
2963 mutex_unlock(&con
->mutex
);
2966 con_fault_finish(con
);
2972 * Generic error/fault handler. A retry mechanism is used with
2973 * exponential backoff
2975 static void con_fault(struct ceph_connection
*con
)
2977 dout("fault %p state %lu to peer %s\n",
2978 con
, con
->state
, ceph_pr_addr(&con
->peer_addr
.in_addr
));
2980 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con
->peer_name
),
2981 ceph_pr_addr(&con
->peer_addr
.in_addr
), con
->error_msg
);
2982 con
->error_msg
= NULL
;
2984 WARN_ON(con
->state
!= CON_STATE_CONNECTING
&&
2985 con
->state
!= CON_STATE_NEGOTIATING
&&
2986 con
->state
!= CON_STATE_OPEN
);
2988 con_close_socket(con
);
2990 if (con_flag_test(con
, CON_FLAG_LOSSYTX
)) {
2991 dout("fault on LOSSYTX channel, marking CLOSED\n");
2992 con
->state
= CON_STATE_CLOSED
;
2997 BUG_ON(con
->in_msg
->con
!= con
);
2998 ceph_msg_put(con
->in_msg
);
3002 /* Requeue anything that hasn't been acked */
3003 list_splice_init(&con
->out_sent
, &con
->out_queue
);
3005 /* If there are no messages queued or keepalive pending, place
3006 * the connection in a STANDBY state */
3007 if (list_empty(&con
->out_queue
) &&
3008 !con_flag_test(con
, CON_FLAG_KEEPALIVE_PENDING
)) {
3009 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con
);
3010 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
3011 con
->state
= CON_STATE_STANDBY
;
3013 /* retry after a delay. */
3014 con
->state
= CON_STATE_PREOPEN
;
3015 if (con
->delay
== 0)
3016 con
->delay
= BASE_DELAY_INTERVAL
;
3017 else if (con
->delay
< MAX_DELAY_INTERVAL
)
3019 con_flag_set(con
, CON_FLAG_BACKOFF
);
3027 * initialize a new messenger instance
3029 void ceph_messenger_init(struct ceph_messenger
*msgr
,
3030 struct ceph_entity_addr
*myaddr
)
3032 spin_lock_init(&msgr
->global_seq_lock
);
3035 msgr
->inst
.addr
= *myaddr
;
3037 /* select a random nonce */
3038 msgr
->inst
.addr
.type
= 0;
3039 get_random_bytes(&msgr
->inst
.addr
.nonce
, sizeof(msgr
->inst
.addr
.nonce
));
3040 encode_my_addr(msgr
);
3042 atomic_set(&msgr
->stopping
, 0);
3043 write_pnet(&msgr
->net
, get_net(current
->nsproxy
->net_ns
));
3045 dout("%s %p\n", __func__
, msgr
);
3047 EXPORT_SYMBOL(ceph_messenger_init
);
3049 void ceph_messenger_fini(struct ceph_messenger
*msgr
)
3051 put_net(read_pnet(&msgr
->net
));
3053 EXPORT_SYMBOL(ceph_messenger_fini
);
3055 static void msg_con_set(struct ceph_msg
*msg
, struct ceph_connection
*con
)
3058 msg
->con
->ops
->put(msg
->con
);
3060 msg
->con
= con
? con
->ops
->get(con
) : NULL
;
3061 BUG_ON(msg
->con
!= con
);
3064 static void clear_standby(struct ceph_connection
*con
)
3066 /* come back from STANDBY? */
3067 if (con
->state
== CON_STATE_STANDBY
) {
3068 dout("clear_standby %p and ++connect_seq\n", con
);
3069 con
->state
= CON_STATE_PREOPEN
;
3071 WARN_ON(con_flag_test(con
, CON_FLAG_WRITE_PENDING
));
3072 WARN_ON(con_flag_test(con
, CON_FLAG_KEEPALIVE_PENDING
));
3077 * Queue up an outgoing message on the given connection.
3079 void ceph_con_send(struct ceph_connection
*con
, struct ceph_msg
*msg
)
3082 msg
->hdr
.src
= con
->msgr
->inst
.name
;
3083 BUG_ON(msg
->front
.iov_len
!= le32_to_cpu(msg
->hdr
.front_len
));
3084 msg
->needs_out_seq
= true;
3086 mutex_lock(&con
->mutex
);
3088 if (con
->state
== CON_STATE_CLOSED
) {
3089 dout("con_send %p closed, dropping %p\n", con
, msg
);
3091 mutex_unlock(&con
->mutex
);
3095 msg_con_set(msg
, con
);
3097 BUG_ON(!list_empty(&msg
->list_head
));
3098 list_add_tail(&msg
->list_head
, &con
->out_queue
);
3099 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg
,
3100 ENTITY_NAME(con
->peer_name
), le16_to_cpu(msg
->hdr
.type
),
3101 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
3102 le32_to_cpu(msg
->hdr
.front_len
),
3103 le32_to_cpu(msg
->hdr
.middle_len
),
3104 le32_to_cpu(msg
->hdr
.data_len
));
3107 mutex_unlock(&con
->mutex
);
3109 /* if there wasn't anything waiting to send before, queue
3111 if (con_flag_test_and_set(con
, CON_FLAG_WRITE_PENDING
) == 0)
3114 EXPORT_SYMBOL(ceph_con_send
);
3117 * Revoke a message that was previously queued for send
3119 void ceph_msg_revoke(struct ceph_msg
*msg
)
3121 struct ceph_connection
*con
= msg
->con
;
3124 dout("%s msg %p null con\n", __func__
, msg
);
3125 return; /* Message not in our possession */
3128 mutex_lock(&con
->mutex
);
3129 if (!list_empty(&msg
->list_head
)) {
3130 dout("%s %p msg %p - was on queue\n", __func__
, con
, msg
);
3131 list_del_init(&msg
->list_head
);
3136 if (con
->out_msg
== msg
) {
3137 BUG_ON(con
->out_skip
);
3139 if (con
->out_msg_done
) {
3140 con
->out_skip
+= con_out_kvec_skip(con
);
3142 BUG_ON(!msg
->data_length
);
3143 con
->out_skip
+= sizeof_footer(con
);
3145 /* data, middle, front */
3146 if (msg
->data_length
)
3147 con
->out_skip
+= msg
->cursor
.total_resid
;
3149 con
->out_skip
+= con_out_kvec_skip(con
);
3150 con
->out_skip
+= con_out_kvec_skip(con
);
3152 dout("%s %p msg %p - was sending, will write %d skip %d\n",
3153 __func__
, con
, msg
, con
->out_kvec_bytes
, con
->out_skip
);
3155 con
->out_msg
= NULL
;
3159 mutex_unlock(&con
->mutex
);
3163 * Revoke a message that we may be reading data into
3165 void ceph_msg_revoke_incoming(struct ceph_msg
*msg
)
3167 struct ceph_connection
*con
= msg
->con
;
3170 dout("%s msg %p null con\n", __func__
, msg
);
3171 return; /* Message not in our possession */
3174 mutex_lock(&con
->mutex
);
3175 if (con
->in_msg
== msg
) {
3176 unsigned int front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
3177 unsigned int middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
3178 unsigned int data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
3180 /* skip rest of message */
3181 dout("%s %p msg %p revoked\n", __func__
, con
, msg
);
3182 con
->in_base_pos
= con
->in_base_pos
-
3183 sizeof(struct ceph_msg_header
) -
3187 sizeof(struct ceph_msg_footer
);
3188 ceph_msg_put(con
->in_msg
);
3190 con
->in_tag
= CEPH_MSGR_TAG_READY
;
3193 dout("%s %p in_msg %p msg %p no-op\n",
3194 __func__
, con
, con
->in_msg
, msg
);
3196 mutex_unlock(&con
->mutex
);
3200 * Queue a keepalive byte to ensure the tcp connection is alive.
3202 void ceph_con_keepalive(struct ceph_connection
*con
)
3204 dout("con_keepalive %p\n", con
);
3205 mutex_lock(&con
->mutex
);
3207 mutex_unlock(&con
->mutex
);
3208 if (con_flag_test_and_set(con
, CON_FLAG_KEEPALIVE_PENDING
) == 0 &&
3209 con_flag_test_and_set(con
, CON_FLAG_WRITE_PENDING
) == 0)
3212 EXPORT_SYMBOL(ceph_con_keepalive
);
3214 bool ceph_con_keepalive_expired(struct ceph_connection
*con
,
3215 unsigned long interval
)
3218 (con
->peer_features
& CEPH_FEATURE_MSGR_KEEPALIVE2
)) {
3219 struct timespec now
;
3221 ktime_get_real_ts(&now
);
3222 jiffies_to_timespec(interval
, &ts
);
3223 ts
= timespec_add(con
->last_keepalive_ack
, ts
);
3224 return timespec_compare(&now
, &ts
) >= 0;
3229 static struct ceph_msg_data
*ceph_msg_data_create(enum ceph_msg_data_type type
)
3231 struct ceph_msg_data
*data
;
3233 if (WARN_ON(!ceph_msg_data_type_valid(type
)))
3236 data
= kmem_cache_zalloc(ceph_msg_data_cache
, GFP_NOFS
);
3241 INIT_LIST_HEAD(&data
->links
);
3246 static void ceph_msg_data_destroy(struct ceph_msg_data
*data
)
3251 WARN_ON(!list_empty(&data
->links
));
3252 if (data
->type
== CEPH_MSG_DATA_PAGELIST
)
3253 ceph_pagelist_release(data
->pagelist
);
3254 kmem_cache_free(ceph_msg_data_cache
, data
);
3257 void ceph_msg_data_add_pages(struct ceph_msg
*msg
, struct page
**pages
,
3258 size_t length
, size_t alignment
)
3260 struct ceph_msg_data
*data
;
3265 data
= ceph_msg_data_create(CEPH_MSG_DATA_PAGES
);
3267 data
->pages
= pages
;
3268 data
->length
= length
;
3269 data
->alignment
= alignment
& ~PAGE_MASK
;
3271 list_add_tail(&data
->links
, &msg
->data
);
3272 msg
->data_length
+= length
;
3274 EXPORT_SYMBOL(ceph_msg_data_add_pages
);
3276 void ceph_msg_data_add_pagelist(struct ceph_msg
*msg
,
3277 struct ceph_pagelist
*pagelist
)
3279 struct ceph_msg_data
*data
;
3282 BUG_ON(!pagelist
->length
);
3284 data
= ceph_msg_data_create(CEPH_MSG_DATA_PAGELIST
);
3286 data
->pagelist
= pagelist
;
3288 list_add_tail(&data
->links
, &msg
->data
);
3289 msg
->data_length
+= pagelist
->length
;
3291 EXPORT_SYMBOL(ceph_msg_data_add_pagelist
);
3294 void ceph_msg_data_add_bio(struct ceph_msg
*msg
, struct bio
*bio
,
3297 struct ceph_msg_data
*data
;
3301 data
= ceph_msg_data_create(CEPH_MSG_DATA_BIO
);
3304 data
->bio_length
= length
;
3306 list_add_tail(&data
->links
, &msg
->data
);
3307 msg
->data_length
+= length
;
3309 EXPORT_SYMBOL(ceph_msg_data_add_bio
);
3310 #endif /* CONFIG_BLOCK */
3313 * construct a new message with given type, size
3314 * the new msg has a ref count of 1.
3316 struct ceph_msg
*ceph_msg_new(int type
, int front_len
, gfp_t flags
,
3321 m
= kmem_cache_zalloc(ceph_msg_cache
, flags
);
3325 m
->hdr
.type
= cpu_to_le16(type
);
3326 m
->hdr
.priority
= cpu_to_le16(CEPH_MSG_PRIO_DEFAULT
);
3327 m
->hdr
.front_len
= cpu_to_le32(front_len
);
3329 INIT_LIST_HEAD(&m
->list_head
);
3330 kref_init(&m
->kref
);
3331 INIT_LIST_HEAD(&m
->data
);
3335 m
->front
.iov_base
= ceph_kvmalloc(front_len
, flags
);
3336 if (m
->front
.iov_base
== NULL
) {
3337 dout("ceph_msg_new can't allocate %d bytes\n",
3342 m
->front
.iov_base
= NULL
;
3344 m
->front_alloc_len
= m
->front
.iov_len
= front_len
;
3346 dout("ceph_msg_new %p front %d\n", m
, front_len
);
3353 pr_err("msg_new can't create type %d front %d\n", type
,
3357 dout("msg_new can't create type %d front %d\n", type
,
3362 EXPORT_SYMBOL(ceph_msg_new
);
3365 * Allocate "middle" portion of a message, if it is needed and wasn't
3366 * allocated by alloc_msg. This allows us to read a small fixed-size
3367 * per-type header in the front and then gracefully fail (i.e.,
3368 * propagate the error to the caller based on info in the front) when
3369 * the middle is too large.
3371 static int ceph_alloc_middle(struct ceph_connection
*con
, struct ceph_msg
*msg
)
3373 int type
= le16_to_cpu(msg
->hdr
.type
);
3374 int middle_len
= le32_to_cpu(msg
->hdr
.middle_len
);
3376 dout("alloc_middle %p type %d %s middle_len %d\n", msg
, type
,
3377 ceph_msg_type_name(type
), middle_len
);
3378 BUG_ON(!middle_len
);
3379 BUG_ON(msg
->middle
);
3381 msg
->middle
= ceph_buffer_new(middle_len
, GFP_NOFS
);
3388 * Allocate a message for receiving an incoming message on a
3389 * connection, and save the result in con->in_msg. Uses the
3390 * connection's private alloc_msg op if available.
3392 * Returns 0 on success, or a negative error code.
3394 * On success, if we set *skip = 1:
3395 * - the next message should be skipped and ignored.
3396 * - con->in_msg == NULL
3397 * or if we set *skip = 0:
3398 * - con->in_msg is non-null.
3399 * On error (ENOMEM, EAGAIN, ...),
3400 * - con->in_msg == NULL
3402 static int ceph_con_in_msg_alloc(struct ceph_connection
*con
, int *skip
)
3404 struct ceph_msg_header
*hdr
= &con
->in_hdr
;
3405 int middle_len
= le32_to_cpu(hdr
->middle_len
);
3406 struct ceph_msg
*msg
;
3409 BUG_ON(con
->in_msg
!= NULL
);
3410 BUG_ON(!con
->ops
->alloc_msg
);
3412 mutex_unlock(&con
->mutex
);
3413 msg
= con
->ops
->alloc_msg(con
, hdr
, skip
);
3414 mutex_lock(&con
->mutex
);
3415 if (con
->state
!= CON_STATE_OPEN
) {
3422 msg_con_set(msg
, con
);
3426 * Null message pointer means either we should skip
3427 * this message or we couldn't allocate memory. The
3428 * former is not an error.
3433 con
->error_msg
= "error allocating memory for incoming message";
3436 memcpy(&con
->in_msg
->hdr
, &con
->in_hdr
, sizeof(con
->in_hdr
));
3438 if (middle_len
&& !con
->in_msg
->middle
) {
3439 ret
= ceph_alloc_middle(con
, con
->in_msg
);
3441 ceph_msg_put(con
->in_msg
);
3451 * Free a generically kmalloc'd message.
3453 static void ceph_msg_free(struct ceph_msg
*m
)
3455 dout("%s %p\n", __func__
, m
);
3456 kvfree(m
->front
.iov_base
);
3457 kmem_cache_free(ceph_msg_cache
, m
);
3460 static void ceph_msg_release(struct kref
*kref
)
3462 struct ceph_msg
*m
= container_of(kref
, struct ceph_msg
, kref
);
3463 struct ceph_msg_data
*data
, *next
;
3465 dout("%s %p\n", __func__
, m
);
3466 WARN_ON(!list_empty(&m
->list_head
));
3468 msg_con_set(m
, NULL
);
3470 /* drop middle, data, if any */
3472 ceph_buffer_put(m
->middle
);
3476 list_for_each_entry_safe(data
, next
, &m
->data
, links
) {
3477 list_del_init(&data
->links
);
3478 ceph_msg_data_destroy(data
);
3483 ceph_msgpool_put(m
->pool
, m
);
3488 struct ceph_msg
*ceph_msg_get(struct ceph_msg
*msg
)
3490 dout("%s %p (was %d)\n", __func__
, msg
,
3491 kref_read(&msg
->kref
));
3492 kref_get(&msg
->kref
);
3495 EXPORT_SYMBOL(ceph_msg_get
);
3497 void ceph_msg_put(struct ceph_msg
*msg
)
3499 dout("%s %p (was %d)\n", __func__
, msg
,
3500 kref_read(&msg
->kref
));
3501 kref_put(&msg
->kref
, ceph_msg_release
);
3503 EXPORT_SYMBOL(ceph_msg_put
);
3505 void ceph_msg_dump(struct ceph_msg
*msg
)
3507 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg
,
3508 msg
->front_alloc_len
, msg
->data_length
);
3509 print_hex_dump(KERN_DEBUG
, "header: ",
3510 DUMP_PREFIX_OFFSET
, 16, 1,
3511 &msg
->hdr
, sizeof(msg
->hdr
), true);
3512 print_hex_dump(KERN_DEBUG
, " front: ",
3513 DUMP_PREFIX_OFFSET
, 16, 1,
3514 msg
->front
.iov_base
, msg
->front
.iov_len
, true);
3516 print_hex_dump(KERN_DEBUG
, "middle: ",
3517 DUMP_PREFIX_OFFSET
, 16, 1,
3518 msg
->middle
->vec
.iov_base
,
3519 msg
->middle
->vec
.iov_len
, true);
3520 print_hex_dump(KERN_DEBUG
, "footer: ",
3521 DUMP_PREFIX_OFFSET
, 16, 1,
3522 &msg
->footer
, sizeof(msg
->footer
), true);
3524 EXPORT_SYMBOL(ceph_msg_dump
);