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
;
160 /* static tag bytes (protocol control messages) */
161 static char tag_msg
= CEPH_MSGR_TAG_MSG
;
162 static char tag_ack
= CEPH_MSGR_TAG_ACK
;
163 static char tag_keepalive
= CEPH_MSGR_TAG_KEEPALIVE
;
164 static char tag_keepalive2
= CEPH_MSGR_TAG_KEEPALIVE2
;
166 #ifdef CONFIG_LOCKDEP
167 static struct lock_class_key socket_class
;
170 static void queue_con(struct ceph_connection
*con
);
171 static void cancel_con(struct ceph_connection
*con
);
172 static void ceph_con_workfn(struct work_struct
*);
173 static void con_fault(struct ceph_connection
*con
);
176 * Nicely render a sockaddr as a string. An array of formatted
177 * strings is used, to approximate reentrancy.
179 #define ADDR_STR_COUNT_LOG 5 /* log2(# address strings in array) */
180 #define ADDR_STR_COUNT (1 << ADDR_STR_COUNT_LOG)
181 #define ADDR_STR_COUNT_MASK (ADDR_STR_COUNT - 1)
182 #define MAX_ADDR_STR_LEN 64 /* 54 is enough */
184 static char addr_str
[ADDR_STR_COUNT
][MAX_ADDR_STR_LEN
];
185 static atomic_t addr_str_seq
= ATOMIC_INIT(0);
187 static struct page
*zero_page
; /* used in certain error cases */
189 const char *ceph_pr_addr(const struct sockaddr_storage
*ss
)
193 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
194 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
196 i
= atomic_inc_return(&addr_str_seq
) & ADDR_STR_COUNT_MASK
;
199 switch (ss
->ss_family
) {
201 snprintf(s
, MAX_ADDR_STR_LEN
, "%pI4:%hu", &in4
->sin_addr
,
202 ntohs(in4
->sin_port
));
206 snprintf(s
, MAX_ADDR_STR_LEN
, "[%pI6c]:%hu", &in6
->sin6_addr
,
207 ntohs(in6
->sin6_port
));
211 snprintf(s
, MAX_ADDR_STR_LEN
, "(unknown sockaddr family %hu)",
217 EXPORT_SYMBOL(ceph_pr_addr
);
219 static void encode_my_addr(struct ceph_messenger
*msgr
)
221 memcpy(&msgr
->my_enc_addr
, &msgr
->inst
.addr
, sizeof(msgr
->my_enc_addr
));
222 ceph_encode_addr(&msgr
->my_enc_addr
);
226 * work queue for all reading and writing to/from the socket.
228 static struct workqueue_struct
*ceph_msgr_wq
;
230 static int ceph_msgr_slab_init(void)
232 BUG_ON(ceph_msg_cache
);
233 ceph_msg_cache
= KMEM_CACHE(ceph_msg
, 0);
240 static void ceph_msgr_slab_exit(void)
242 BUG_ON(!ceph_msg_cache
);
243 kmem_cache_destroy(ceph_msg_cache
);
244 ceph_msg_cache
= NULL
;
247 static void _ceph_msgr_exit(void)
250 destroy_workqueue(ceph_msgr_wq
);
254 BUG_ON(zero_page
== NULL
);
258 ceph_msgr_slab_exit();
261 int __init
ceph_msgr_init(void)
263 if (ceph_msgr_slab_init())
266 BUG_ON(zero_page
!= NULL
);
267 zero_page
= ZERO_PAGE(0);
271 * The number of active work items is limited by the number of
272 * connections, so leave @max_active at default.
274 ceph_msgr_wq
= alloc_workqueue("ceph-msgr", WQ_MEM_RECLAIM
, 0);
278 pr_err("msgr_init failed to create workqueue\n");
284 void ceph_msgr_exit(void)
286 BUG_ON(ceph_msgr_wq
== NULL
);
291 void ceph_msgr_flush(void)
293 flush_workqueue(ceph_msgr_wq
);
295 EXPORT_SYMBOL(ceph_msgr_flush
);
297 /* Connection socket state transition functions */
299 static void con_sock_state_init(struct ceph_connection
*con
)
303 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSED
);
304 if (WARN_ON(old_state
!= CON_SOCK_STATE_NEW
))
305 printk("%s: unexpected old state %d\n", __func__
, old_state
);
306 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
307 CON_SOCK_STATE_CLOSED
);
310 static void con_sock_state_connecting(struct ceph_connection
*con
)
314 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CONNECTING
);
315 if (WARN_ON(old_state
!= CON_SOCK_STATE_CLOSED
))
316 printk("%s: unexpected old state %d\n", __func__
, old_state
);
317 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
318 CON_SOCK_STATE_CONNECTING
);
321 static void con_sock_state_connected(struct ceph_connection
*con
)
325 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CONNECTED
);
326 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTING
))
327 printk("%s: unexpected old state %d\n", __func__
, old_state
);
328 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
329 CON_SOCK_STATE_CONNECTED
);
332 static void con_sock_state_closing(struct ceph_connection
*con
)
336 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSING
);
337 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTING
&&
338 old_state
!= CON_SOCK_STATE_CONNECTED
&&
339 old_state
!= CON_SOCK_STATE_CLOSING
))
340 printk("%s: unexpected old state %d\n", __func__
, old_state
);
341 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
342 CON_SOCK_STATE_CLOSING
);
345 static void con_sock_state_closed(struct ceph_connection
*con
)
349 old_state
= atomic_xchg(&con
->sock_state
, CON_SOCK_STATE_CLOSED
);
350 if (WARN_ON(old_state
!= CON_SOCK_STATE_CONNECTED
&&
351 old_state
!= CON_SOCK_STATE_CLOSING
&&
352 old_state
!= CON_SOCK_STATE_CONNECTING
&&
353 old_state
!= CON_SOCK_STATE_CLOSED
))
354 printk("%s: unexpected old state %d\n", __func__
, old_state
);
355 dout("%s con %p sock %d -> %d\n", __func__
, con
, old_state
,
356 CON_SOCK_STATE_CLOSED
);
360 * socket callback functions
363 /* data available on socket, or listen socket received a connect */
364 static void ceph_sock_data_ready(struct sock
*sk
)
366 struct ceph_connection
*con
= sk
->sk_user_data
;
367 if (atomic_read(&con
->msgr
->stopping
)) {
371 if (sk
->sk_state
!= TCP_CLOSE_WAIT
) {
372 dout("%s on %p state = %lu, queueing work\n", __func__
,
378 /* socket has buffer space for writing */
379 static void ceph_sock_write_space(struct sock
*sk
)
381 struct ceph_connection
*con
= sk
->sk_user_data
;
383 /* only queue to workqueue if there is data we want to write,
384 * and there is sufficient space in the socket buffer to accept
385 * more data. clear SOCK_NOSPACE so that ceph_sock_write_space()
386 * doesn't get called again until try_write() fills the socket
387 * buffer. See net/ipv4/tcp_input.c:tcp_check_space()
388 * and net/core/stream.c:sk_stream_write_space().
390 if (con_flag_test(con
, CON_FLAG_WRITE_PENDING
)) {
391 if (sk_stream_is_writeable(sk
)) {
392 dout("%s %p queueing write work\n", __func__
, con
);
393 clear_bit(SOCK_NOSPACE
, &sk
->sk_socket
->flags
);
397 dout("%s %p nothing to write\n", __func__
, con
);
401 /* socket's state has changed */
402 static void ceph_sock_state_change(struct sock
*sk
)
404 struct ceph_connection
*con
= sk
->sk_user_data
;
406 dout("%s %p state = %lu sk_state = %u\n", __func__
,
407 con
, con
->state
, sk
->sk_state
);
409 switch (sk
->sk_state
) {
411 dout("%s TCP_CLOSE\n", __func__
);
414 dout("%s TCP_CLOSE_WAIT\n", __func__
);
415 con_sock_state_closing(con
);
416 con_flag_set(con
, CON_FLAG_SOCK_CLOSED
);
419 case TCP_ESTABLISHED
:
420 dout("%s TCP_ESTABLISHED\n", __func__
);
421 con_sock_state_connected(con
);
424 default: /* Everything else is uninteresting */
430 * set up socket callbacks
432 static void set_sock_callbacks(struct socket
*sock
,
433 struct ceph_connection
*con
)
435 struct sock
*sk
= sock
->sk
;
436 sk
->sk_user_data
= con
;
437 sk
->sk_data_ready
= ceph_sock_data_ready
;
438 sk
->sk_write_space
= ceph_sock_write_space
;
439 sk
->sk_state_change
= ceph_sock_state_change
;
448 * initiate connection to a remote socket.
450 static int ceph_tcp_connect(struct ceph_connection
*con
)
452 struct sockaddr_storage
*paddr
= &con
->peer_addr
.in_addr
;
454 unsigned int noio_flag
;
459 /* sock_create_kern() allocates with GFP_KERNEL */
460 noio_flag
= memalloc_noio_save();
461 ret
= sock_create_kern(read_pnet(&con
->msgr
->net
), paddr
->ss_family
,
462 SOCK_STREAM
, IPPROTO_TCP
, &sock
);
463 memalloc_noio_restore(noio_flag
);
466 sock
->sk
->sk_allocation
= GFP_NOFS
;
468 #ifdef CONFIG_LOCKDEP
469 lockdep_set_class(&sock
->sk
->sk_lock
, &socket_class
);
472 set_sock_callbacks(sock
, con
);
474 dout("connect %s\n", ceph_pr_addr(&con
->peer_addr
.in_addr
));
476 con_sock_state_connecting(con
);
477 ret
= sock
->ops
->connect(sock
, (struct sockaddr
*)paddr
, sizeof(*paddr
),
479 if (ret
== -EINPROGRESS
) {
480 dout("connect %s EINPROGRESS sk_state = %u\n",
481 ceph_pr_addr(&con
->peer_addr
.in_addr
),
483 } else if (ret
< 0) {
484 pr_err("connect %s error %d\n",
485 ceph_pr_addr(&con
->peer_addr
.in_addr
), ret
);
490 if (ceph_test_opt(from_msgr(con
->msgr
), TCP_NODELAY
)) {
493 ret
= kernel_setsockopt(sock
, SOL_TCP
, TCP_NODELAY
,
494 (char *)&optval
, sizeof(optval
));
496 pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
505 * If @buf is NULL, discard up to @len bytes.
507 static int ceph_tcp_recvmsg(struct socket
*sock
, void *buf
, size_t len
)
509 struct kvec iov
= {buf
, len
};
510 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
514 msg
.msg_flags
|= MSG_TRUNC
;
516 iov_iter_kvec(&msg
.msg_iter
, READ
, &iov
, 1, len
);
517 r
= sock_recvmsg(sock
, &msg
, msg
.msg_flags
);
523 static int ceph_tcp_recvpage(struct socket
*sock
, struct page
*page
,
524 int page_offset
, size_t length
)
526 struct bio_vec bvec
= {
528 .bv_offset
= page_offset
,
531 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
534 BUG_ON(page_offset
+ length
> PAGE_SIZE
);
535 iov_iter_bvec(&msg
.msg_iter
, READ
, &bvec
, 1, length
);
536 r
= sock_recvmsg(sock
, &msg
, msg
.msg_flags
);
543 * write something. @more is true if caller will be sending more data
546 static int ceph_tcp_sendmsg(struct socket
*sock
, struct kvec
*iov
,
547 size_t kvlen
, size_t len
, bool more
)
549 struct msghdr msg
= { .msg_flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
};
553 msg
.msg_flags
|= MSG_MORE
;
555 msg
.msg_flags
|= MSG_EOR
; /* superfluous, but what the hell */
557 r
= kernel_sendmsg(sock
, &msg
, iov
, kvlen
, len
);
564 * @more: either or both of MSG_MORE and MSG_SENDPAGE_NOTLAST
566 static int ceph_tcp_sendpage(struct socket
*sock
, struct page
*page
,
567 int offset
, size_t size
, int more
)
569 ssize_t (*sendpage
)(struct socket
*sock
, struct page
*page
,
570 int offset
, size_t size
, int flags
);
571 int flags
= MSG_DONTWAIT
| MSG_NOSIGNAL
| more
;
575 * sendpage cannot properly handle pages with page_count == 0,
576 * we need to fall back to sendmsg if that's the case.
578 * Same goes for slab pages: skb_can_coalesce() allows
579 * coalescing neighboring slab objects into a single frag which
580 * triggers one of hardened usercopy checks.
582 if (page_count(page
) >= 1 && !PageSlab(page
))
583 sendpage
= sock
->ops
->sendpage
;
585 sendpage
= sock_no_sendpage
;
587 ret
= sendpage(sock
, page
, offset
, size
, flags
);
595 * Shutdown/close the socket for the given connection.
597 static int con_close_socket(struct ceph_connection
*con
)
601 dout("con_close_socket on %p sock %p\n", con
, con
->sock
);
603 rc
= con
->sock
->ops
->shutdown(con
->sock
, SHUT_RDWR
);
604 sock_release(con
->sock
);
609 * Forcibly clear the SOCK_CLOSED flag. It gets set
610 * independent of the connection mutex, and we could have
611 * received a socket close event before we had the chance to
612 * shut the socket down.
614 con_flag_clear(con
, CON_FLAG_SOCK_CLOSED
);
616 con_sock_state_closed(con
);
621 * Reset a connection. Discard all incoming and outgoing messages
622 * and clear *_seq state.
624 static void ceph_msg_remove(struct ceph_msg
*msg
)
626 list_del_init(&msg
->list_head
);
630 static void ceph_msg_remove_list(struct list_head
*head
)
632 while (!list_empty(head
)) {
633 struct ceph_msg
*msg
= list_first_entry(head
, struct ceph_msg
,
635 ceph_msg_remove(msg
);
639 static void reset_connection(struct ceph_connection
*con
)
641 /* reset connection, out_queue, msg_ and connect_seq */
642 /* discard existing out_queue and msg_seq */
643 dout("reset_connection %p\n", con
);
644 ceph_msg_remove_list(&con
->out_queue
);
645 ceph_msg_remove_list(&con
->out_sent
);
648 BUG_ON(con
->in_msg
->con
!= con
);
649 ceph_msg_put(con
->in_msg
);
653 con
->connect_seq
= 0;
656 BUG_ON(con
->out_msg
->con
!= con
);
657 ceph_msg_put(con
->out_msg
);
661 con
->in_seq_acked
= 0;
667 * mark a peer down. drop any open connections.
669 void ceph_con_close(struct ceph_connection
*con
)
671 mutex_lock(&con
->mutex
);
672 dout("con_close %p peer %s\n", con
,
673 ceph_pr_addr(&con
->peer_addr
.in_addr
));
674 con
->state
= CON_STATE_CLOSED
;
676 con_flag_clear(con
, CON_FLAG_LOSSYTX
); /* so we retry next connect */
677 con_flag_clear(con
, CON_FLAG_KEEPALIVE_PENDING
);
678 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
679 con_flag_clear(con
, CON_FLAG_BACKOFF
);
681 reset_connection(con
);
682 con
->peer_global_seq
= 0;
684 con_close_socket(con
);
685 mutex_unlock(&con
->mutex
);
687 EXPORT_SYMBOL(ceph_con_close
);
690 * Reopen a closed connection, with a new peer address.
692 void ceph_con_open(struct ceph_connection
*con
,
693 __u8 entity_type
, __u64 entity_num
,
694 struct ceph_entity_addr
*addr
)
696 mutex_lock(&con
->mutex
);
697 dout("con_open %p %s\n", con
, ceph_pr_addr(&addr
->in_addr
));
699 WARN_ON(con
->state
!= CON_STATE_CLOSED
);
700 con
->state
= CON_STATE_PREOPEN
;
702 con
->peer_name
.type
= (__u8
) entity_type
;
703 con
->peer_name
.num
= cpu_to_le64(entity_num
);
705 memcpy(&con
->peer_addr
, addr
, sizeof(*addr
));
706 con
->delay
= 0; /* reset backoff memory */
707 mutex_unlock(&con
->mutex
);
710 EXPORT_SYMBOL(ceph_con_open
);
713 * return true if this connection ever successfully opened
715 bool ceph_con_opened(struct ceph_connection
*con
)
717 return con
->connect_seq
> 0;
721 * initialize a new connection.
723 void ceph_con_init(struct ceph_connection
*con
, void *private,
724 const struct ceph_connection_operations
*ops
,
725 struct ceph_messenger
*msgr
)
727 dout("con_init %p\n", con
);
728 memset(con
, 0, sizeof(*con
));
729 con
->private = private;
733 con_sock_state_init(con
);
735 mutex_init(&con
->mutex
);
736 INIT_LIST_HEAD(&con
->out_queue
);
737 INIT_LIST_HEAD(&con
->out_sent
);
738 INIT_DELAYED_WORK(&con
->work
, ceph_con_workfn
);
740 con
->state
= CON_STATE_CLOSED
;
742 EXPORT_SYMBOL(ceph_con_init
);
746 * We maintain a global counter to order connection attempts. Get
747 * a unique seq greater than @gt.
749 static u32
get_global_seq(struct ceph_messenger
*msgr
, u32 gt
)
753 spin_lock(&msgr
->global_seq_lock
);
754 if (msgr
->global_seq
< gt
)
755 msgr
->global_seq
= gt
;
756 ret
= ++msgr
->global_seq
;
757 spin_unlock(&msgr
->global_seq_lock
);
761 static void con_out_kvec_reset(struct ceph_connection
*con
)
763 BUG_ON(con
->out_skip
);
765 con
->out_kvec_left
= 0;
766 con
->out_kvec_bytes
= 0;
767 con
->out_kvec_cur
= &con
->out_kvec
[0];
770 static void con_out_kvec_add(struct ceph_connection
*con
,
771 size_t size
, void *data
)
773 int index
= con
->out_kvec_left
;
775 BUG_ON(con
->out_skip
);
776 BUG_ON(index
>= ARRAY_SIZE(con
->out_kvec
));
778 con
->out_kvec
[index
].iov_len
= size
;
779 con
->out_kvec
[index
].iov_base
= data
;
780 con
->out_kvec_left
++;
781 con
->out_kvec_bytes
+= size
;
785 * Chop off a kvec from the end. Return residual number of bytes for
786 * that kvec, i.e. how many bytes would have been written if the kvec
789 static int con_out_kvec_skip(struct ceph_connection
*con
)
791 int off
= con
->out_kvec_cur
- con
->out_kvec
;
794 if (con
->out_kvec_bytes
> 0) {
795 skip
= con
->out_kvec
[off
+ con
->out_kvec_left
- 1].iov_len
;
796 BUG_ON(con
->out_kvec_bytes
< skip
);
797 BUG_ON(!con
->out_kvec_left
);
798 con
->out_kvec_bytes
-= skip
;
799 con
->out_kvec_left
--;
808 * For a bio data item, a piece is whatever remains of the next
809 * entry in the current bio iovec, or the first entry in the next
812 static void ceph_msg_data_bio_cursor_init(struct ceph_msg_data_cursor
*cursor
,
815 struct ceph_msg_data
*data
= cursor
->data
;
816 struct ceph_bio_iter
*it
= &cursor
->bio_iter
;
818 cursor
->resid
= min_t(size_t, length
, data
->bio_length
);
820 if (cursor
->resid
< it
->iter
.bi_size
)
821 it
->iter
.bi_size
= cursor
->resid
;
823 BUG_ON(cursor
->resid
< bio_iter_len(it
->bio
, it
->iter
));
824 cursor
->last_piece
= cursor
->resid
== bio_iter_len(it
->bio
, it
->iter
);
827 static struct page
*ceph_msg_data_bio_next(struct ceph_msg_data_cursor
*cursor
,
831 struct bio_vec bv
= bio_iter_iovec(cursor
->bio_iter
.bio
,
832 cursor
->bio_iter
.iter
);
834 *page_offset
= bv
.bv_offset
;
839 static bool ceph_msg_data_bio_advance(struct ceph_msg_data_cursor
*cursor
,
842 struct ceph_bio_iter
*it
= &cursor
->bio_iter
;
844 BUG_ON(bytes
> cursor
->resid
);
845 BUG_ON(bytes
> bio_iter_len(it
->bio
, it
->iter
));
846 cursor
->resid
-= bytes
;
847 bio_advance_iter(it
->bio
, &it
->iter
, bytes
);
849 if (!cursor
->resid
) {
850 BUG_ON(!cursor
->last_piece
);
851 return false; /* no more data */
854 if (!bytes
|| (it
->iter
.bi_size
&& it
->iter
.bi_bvec_done
))
855 return false; /* more bytes to process in this segment */
857 if (!it
->iter
.bi_size
) {
858 it
->bio
= it
->bio
->bi_next
;
859 it
->iter
= it
->bio
->bi_iter
;
860 if (cursor
->resid
< it
->iter
.bi_size
)
861 it
->iter
.bi_size
= cursor
->resid
;
864 BUG_ON(cursor
->last_piece
);
865 BUG_ON(cursor
->resid
< bio_iter_len(it
->bio
, it
->iter
));
866 cursor
->last_piece
= cursor
->resid
== bio_iter_len(it
->bio
, it
->iter
);
869 #endif /* CONFIG_BLOCK */
871 static void ceph_msg_data_bvecs_cursor_init(struct ceph_msg_data_cursor
*cursor
,
874 struct ceph_msg_data
*data
= cursor
->data
;
875 struct bio_vec
*bvecs
= data
->bvec_pos
.bvecs
;
877 cursor
->resid
= min_t(size_t, length
, data
->bvec_pos
.iter
.bi_size
);
878 cursor
->bvec_iter
= data
->bvec_pos
.iter
;
879 cursor
->bvec_iter
.bi_size
= cursor
->resid
;
881 BUG_ON(cursor
->resid
< bvec_iter_len(bvecs
, cursor
->bvec_iter
));
883 cursor
->resid
== bvec_iter_len(bvecs
, cursor
->bvec_iter
);
886 static struct page
*ceph_msg_data_bvecs_next(struct ceph_msg_data_cursor
*cursor
,
890 struct bio_vec bv
= bvec_iter_bvec(cursor
->data
->bvec_pos
.bvecs
,
893 *page_offset
= bv
.bv_offset
;
898 static bool ceph_msg_data_bvecs_advance(struct ceph_msg_data_cursor
*cursor
,
901 struct bio_vec
*bvecs
= cursor
->data
->bvec_pos
.bvecs
;
903 BUG_ON(bytes
> cursor
->resid
);
904 BUG_ON(bytes
> bvec_iter_len(bvecs
, cursor
->bvec_iter
));
905 cursor
->resid
-= bytes
;
906 bvec_iter_advance(bvecs
, &cursor
->bvec_iter
, bytes
);
908 if (!cursor
->resid
) {
909 BUG_ON(!cursor
->last_piece
);
910 return false; /* no more data */
913 if (!bytes
|| cursor
->bvec_iter
.bi_bvec_done
)
914 return false; /* more bytes to process in this segment */
916 BUG_ON(cursor
->last_piece
);
917 BUG_ON(cursor
->resid
< bvec_iter_len(bvecs
, cursor
->bvec_iter
));
919 cursor
->resid
== bvec_iter_len(bvecs
, cursor
->bvec_iter
);
924 * For a page array, a piece comes from the first page in the array
925 * that has not already been fully consumed.
927 static void ceph_msg_data_pages_cursor_init(struct ceph_msg_data_cursor
*cursor
,
930 struct ceph_msg_data
*data
= cursor
->data
;
933 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGES
);
935 BUG_ON(!data
->pages
);
936 BUG_ON(!data
->length
);
938 cursor
->resid
= min(length
, data
->length
);
939 page_count
= calc_pages_for(data
->alignment
, (u64
)data
->length
);
940 cursor
->page_offset
= data
->alignment
& ~PAGE_MASK
;
941 cursor
->page_index
= 0;
942 BUG_ON(page_count
> (int)USHRT_MAX
);
943 cursor
->page_count
= (unsigned short)page_count
;
944 BUG_ON(length
> SIZE_MAX
- cursor
->page_offset
);
945 cursor
->last_piece
= cursor
->page_offset
+ cursor
->resid
<= PAGE_SIZE
;
949 ceph_msg_data_pages_next(struct ceph_msg_data_cursor
*cursor
,
950 size_t *page_offset
, size_t *length
)
952 struct ceph_msg_data
*data
= cursor
->data
;
954 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGES
);
956 BUG_ON(cursor
->page_index
>= cursor
->page_count
);
957 BUG_ON(cursor
->page_offset
>= PAGE_SIZE
);
959 *page_offset
= cursor
->page_offset
;
960 if (cursor
->last_piece
)
961 *length
= cursor
->resid
;
963 *length
= PAGE_SIZE
- *page_offset
;
965 return data
->pages
[cursor
->page_index
];
968 static bool ceph_msg_data_pages_advance(struct ceph_msg_data_cursor
*cursor
,
971 BUG_ON(cursor
->data
->type
!= CEPH_MSG_DATA_PAGES
);
973 BUG_ON(cursor
->page_offset
+ bytes
> PAGE_SIZE
);
975 /* Advance the cursor page offset */
977 cursor
->resid
-= bytes
;
978 cursor
->page_offset
= (cursor
->page_offset
+ bytes
) & ~PAGE_MASK
;
979 if (!bytes
|| cursor
->page_offset
)
980 return false; /* more bytes to process in the current page */
983 return false; /* no more data */
985 /* Move on to the next page; offset is already at 0 */
987 BUG_ON(cursor
->page_index
>= cursor
->page_count
);
988 cursor
->page_index
++;
989 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
995 * For a pagelist, a piece is whatever remains to be consumed in the
996 * first page in the list, or the front of the next page.
999 ceph_msg_data_pagelist_cursor_init(struct ceph_msg_data_cursor
*cursor
,
1002 struct ceph_msg_data
*data
= cursor
->data
;
1003 struct ceph_pagelist
*pagelist
;
1006 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
1008 pagelist
= data
->pagelist
;
1012 return; /* pagelist can be assigned but empty */
1014 BUG_ON(list_empty(&pagelist
->head
));
1015 page
= list_first_entry(&pagelist
->head
, struct page
, lru
);
1017 cursor
->resid
= min(length
, pagelist
->length
);
1018 cursor
->page
= page
;
1020 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
1023 static struct page
*
1024 ceph_msg_data_pagelist_next(struct ceph_msg_data_cursor
*cursor
,
1025 size_t *page_offset
, size_t *length
)
1027 struct ceph_msg_data
*data
= cursor
->data
;
1028 struct ceph_pagelist
*pagelist
;
1030 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
1032 pagelist
= data
->pagelist
;
1035 BUG_ON(!cursor
->page
);
1036 BUG_ON(cursor
->offset
+ cursor
->resid
!= pagelist
->length
);
1038 /* offset of first page in pagelist is always 0 */
1039 *page_offset
= cursor
->offset
& ~PAGE_MASK
;
1040 if (cursor
->last_piece
)
1041 *length
= cursor
->resid
;
1043 *length
= PAGE_SIZE
- *page_offset
;
1045 return cursor
->page
;
1048 static bool ceph_msg_data_pagelist_advance(struct ceph_msg_data_cursor
*cursor
,
1051 struct ceph_msg_data
*data
= cursor
->data
;
1052 struct ceph_pagelist
*pagelist
;
1054 BUG_ON(data
->type
!= CEPH_MSG_DATA_PAGELIST
);
1056 pagelist
= data
->pagelist
;
1059 BUG_ON(cursor
->offset
+ cursor
->resid
!= pagelist
->length
);
1060 BUG_ON((cursor
->offset
& ~PAGE_MASK
) + bytes
> PAGE_SIZE
);
1062 /* Advance the cursor offset */
1064 cursor
->resid
-= bytes
;
1065 cursor
->offset
+= bytes
;
1066 /* offset of first page in pagelist is always 0 */
1067 if (!bytes
|| cursor
->offset
& ~PAGE_MASK
)
1068 return false; /* more bytes to process in the current page */
1071 return false; /* no more data */
1073 /* Move on to the next page */
1075 BUG_ON(list_is_last(&cursor
->page
->lru
, &pagelist
->head
));
1076 cursor
->page
= list_next_entry(cursor
->page
, lru
);
1077 cursor
->last_piece
= cursor
->resid
<= PAGE_SIZE
;
1083 * Message data is handled (sent or received) in pieces, where each
1084 * piece resides on a single page. The network layer might not
1085 * consume an entire piece at once. A data item's cursor keeps
1086 * track of which piece is next to process and how much remains to
1087 * be processed in that piece. It also tracks whether the current
1088 * piece is the last one in the data item.
1090 static void __ceph_msg_data_cursor_init(struct ceph_msg_data_cursor
*cursor
)
1092 size_t length
= cursor
->total_resid
;
1094 switch (cursor
->data
->type
) {
1095 case CEPH_MSG_DATA_PAGELIST
:
1096 ceph_msg_data_pagelist_cursor_init(cursor
, length
);
1098 case CEPH_MSG_DATA_PAGES
:
1099 ceph_msg_data_pages_cursor_init(cursor
, length
);
1102 case CEPH_MSG_DATA_BIO
:
1103 ceph_msg_data_bio_cursor_init(cursor
, length
);
1105 #endif /* CONFIG_BLOCK */
1106 case CEPH_MSG_DATA_BVECS
:
1107 ceph_msg_data_bvecs_cursor_init(cursor
, length
);
1109 case CEPH_MSG_DATA_NONE
:
1114 cursor
->need_crc
= true;
1117 static void ceph_msg_data_cursor_init(struct ceph_msg
*msg
, size_t length
)
1119 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
1122 BUG_ON(length
> msg
->data_length
);
1123 BUG_ON(!msg
->num_data_items
);
1125 cursor
->total_resid
= length
;
1126 cursor
->data
= msg
->data
;
1128 __ceph_msg_data_cursor_init(cursor
);
1132 * Return the page containing the next piece to process for a given
1133 * data item, and supply the page offset and length of that piece.
1134 * Indicate whether this is the last piece in this data item.
1136 static struct page
*ceph_msg_data_next(struct ceph_msg_data_cursor
*cursor
,
1137 size_t *page_offset
, size_t *length
,
1142 switch (cursor
->data
->type
) {
1143 case CEPH_MSG_DATA_PAGELIST
:
1144 page
= ceph_msg_data_pagelist_next(cursor
, page_offset
, length
);
1146 case CEPH_MSG_DATA_PAGES
:
1147 page
= ceph_msg_data_pages_next(cursor
, page_offset
, length
);
1150 case CEPH_MSG_DATA_BIO
:
1151 page
= ceph_msg_data_bio_next(cursor
, page_offset
, length
);
1153 #endif /* CONFIG_BLOCK */
1154 case CEPH_MSG_DATA_BVECS
:
1155 page
= ceph_msg_data_bvecs_next(cursor
, page_offset
, length
);
1157 case CEPH_MSG_DATA_NONE
:
1164 BUG_ON(*page_offset
+ *length
> PAGE_SIZE
);
1166 BUG_ON(*length
> cursor
->resid
);
1168 *last_piece
= cursor
->last_piece
;
1174 * Returns true if the result moves the cursor on to the next piece
1177 static void ceph_msg_data_advance(struct ceph_msg_data_cursor
*cursor
,
1182 BUG_ON(bytes
> cursor
->resid
);
1183 switch (cursor
->data
->type
) {
1184 case CEPH_MSG_DATA_PAGELIST
:
1185 new_piece
= ceph_msg_data_pagelist_advance(cursor
, bytes
);
1187 case CEPH_MSG_DATA_PAGES
:
1188 new_piece
= ceph_msg_data_pages_advance(cursor
, bytes
);
1191 case CEPH_MSG_DATA_BIO
:
1192 new_piece
= ceph_msg_data_bio_advance(cursor
, bytes
);
1194 #endif /* CONFIG_BLOCK */
1195 case CEPH_MSG_DATA_BVECS
:
1196 new_piece
= ceph_msg_data_bvecs_advance(cursor
, bytes
);
1198 case CEPH_MSG_DATA_NONE
:
1203 cursor
->total_resid
-= bytes
;
1205 if (!cursor
->resid
&& cursor
->total_resid
) {
1206 WARN_ON(!cursor
->last_piece
);
1208 __ceph_msg_data_cursor_init(cursor
);
1211 cursor
->need_crc
= new_piece
;
1214 static size_t sizeof_footer(struct ceph_connection
*con
)
1216 return (con
->peer_features
& CEPH_FEATURE_MSG_AUTH
) ?
1217 sizeof(struct ceph_msg_footer
) :
1218 sizeof(struct ceph_msg_footer_old
);
1221 static void prepare_message_data(struct ceph_msg
*msg
, u32 data_len
)
1223 /* Initialize data cursor */
1225 ceph_msg_data_cursor_init(msg
, (size_t)data_len
);
1229 * Prepare footer for currently outgoing message, and finish things
1230 * off. Assumes out_kvec* are already valid.. we just add on to the end.
1232 static void prepare_write_message_footer(struct ceph_connection
*con
)
1234 struct ceph_msg
*m
= con
->out_msg
;
1236 m
->footer
.flags
|= CEPH_MSG_FOOTER_COMPLETE
;
1238 dout("prepare_write_message_footer %p\n", con
);
1239 con_out_kvec_add(con
, sizeof_footer(con
), &m
->footer
);
1240 if (con
->peer_features
& CEPH_FEATURE_MSG_AUTH
) {
1241 if (con
->ops
->sign_message
)
1242 con
->ops
->sign_message(m
);
1246 m
->old_footer
.flags
= m
->footer
.flags
;
1248 con
->out_more
= m
->more_to_follow
;
1249 con
->out_msg_done
= true;
1253 * Prepare headers for the next outgoing message.
1255 static void prepare_write_message(struct ceph_connection
*con
)
1260 con_out_kvec_reset(con
);
1261 con
->out_msg_done
= false;
1263 /* Sneak an ack in there first? If we can get it into the same
1264 * TCP packet that's a good thing. */
1265 if (con
->in_seq
> con
->in_seq_acked
) {
1266 con
->in_seq_acked
= con
->in_seq
;
1267 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
1268 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1269 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1270 &con
->out_temp_ack
);
1273 BUG_ON(list_empty(&con
->out_queue
));
1274 m
= list_first_entry(&con
->out_queue
, struct ceph_msg
, list_head
);
1276 BUG_ON(m
->con
!= con
);
1278 /* put message on sent list */
1280 list_move_tail(&m
->list_head
, &con
->out_sent
);
1283 * only assign outgoing seq # if we haven't sent this message
1284 * yet. if it is requeued, resend with it's original seq.
1286 if (m
->needs_out_seq
) {
1287 m
->hdr
.seq
= cpu_to_le64(++con
->out_seq
);
1288 m
->needs_out_seq
= false;
1290 if (con
->ops
->reencode_message
)
1291 con
->ops
->reencode_message(m
);
1294 dout("prepare_write_message %p seq %lld type %d len %d+%d+%zd\n",
1295 m
, con
->out_seq
, le16_to_cpu(m
->hdr
.type
),
1296 le32_to_cpu(m
->hdr
.front_len
), le32_to_cpu(m
->hdr
.middle_len
),
1298 WARN_ON(m
->front
.iov_len
!= le32_to_cpu(m
->hdr
.front_len
));
1299 WARN_ON(m
->data_length
!= le32_to_cpu(m
->hdr
.data_len
));
1301 /* tag + hdr + front + middle */
1302 con_out_kvec_add(con
, sizeof (tag_msg
), &tag_msg
);
1303 con_out_kvec_add(con
, sizeof(con
->out_hdr
), &con
->out_hdr
);
1304 con_out_kvec_add(con
, m
->front
.iov_len
, m
->front
.iov_base
);
1307 con_out_kvec_add(con
, m
->middle
->vec
.iov_len
,
1308 m
->middle
->vec
.iov_base
);
1310 /* fill in hdr crc and finalize hdr */
1311 crc
= crc32c(0, &m
->hdr
, offsetof(struct ceph_msg_header
, crc
));
1312 con
->out_msg
->hdr
.crc
= cpu_to_le32(crc
);
1313 memcpy(&con
->out_hdr
, &con
->out_msg
->hdr
, sizeof(con
->out_hdr
));
1315 /* fill in front and middle crc, footer */
1316 crc
= crc32c(0, m
->front
.iov_base
, m
->front
.iov_len
);
1317 con
->out_msg
->footer
.front_crc
= cpu_to_le32(crc
);
1319 crc
= crc32c(0, m
->middle
->vec
.iov_base
,
1320 m
->middle
->vec
.iov_len
);
1321 con
->out_msg
->footer
.middle_crc
= cpu_to_le32(crc
);
1323 con
->out_msg
->footer
.middle_crc
= 0;
1324 dout("%s front_crc %u middle_crc %u\n", __func__
,
1325 le32_to_cpu(con
->out_msg
->footer
.front_crc
),
1326 le32_to_cpu(con
->out_msg
->footer
.middle_crc
));
1327 con
->out_msg
->footer
.flags
= 0;
1329 /* is there a data payload? */
1330 con
->out_msg
->footer
.data_crc
= 0;
1331 if (m
->data_length
) {
1332 prepare_message_data(con
->out_msg
, m
->data_length
);
1333 con
->out_more
= 1; /* data + footer will follow */
1335 /* no, queue up footer too and be done */
1336 prepare_write_message_footer(con
);
1339 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1345 static void prepare_write_ack(struct ceph_connection
*con
)
1347 dout("prepare_write_ack %p %llu -> %llu\n", con
,
1348 con
->in_seq_acked
, con
->in_seq
);
1349 con
->in_seq_acked
= con
->in_seq
;
1351 con_out_kvec_reset(con
);
1353 con_out_kvec_add(con
, sizeof (tag_ack
), &tag_ack
);
1355 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1356 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1357 &con
->out_temp_ack
);
1359 con
->out_more
= 1; /* more will follow.. eventually.. */
1360 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1364 * Prepare to share the seq during handshake
1366 static void prepare_write_seq(struct ceph_connection
*con
)
1368 dout("prepare_write_seq %p %llu -> %llu\n", con
,
1369 con
->in_seq_acked
, con
->in_seq
);
1370 con
->in_seq_acked
= con
->in_seq
;
1372 con_out_kvec_reset(con
);
1374 con
->out_temp_ack
= cpu_to_le64(con
->in_seq_acked
);
1375 con_out_kvec_add(con
, sizeof (con
->out_temp_ack
),
1376 &con
->out_temp_ack
);
1378 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1382 * Prepare to write keepalive byte.
1384 static void prepare_write_keepalive(struct ceph_connection
*con
)
1386 dout("prepare_write_keepalive %p\n", con
);
1387 con_out_kvec_reset(con
);
1388 if (con
->peer_features
& CEPH_FEATURE_MSGR_KEEPALIVE2
) {
1389 struct timespec64 now
;
1391 ktime_get_real_ts64(&now
);
1392 con_out_kvec_add(con
, sizeof(tag_keepalive2
), &tag_keepalive2
);
1393 ceph_encode_timespec64(&con
->out_temp_keepalive2
, &now
);
1394 con_out_kvec_add(con
, sizeof(con
->out_temp_keepalive2
),
1395 &con
->out_temp_keepalive2
);
1397 con_out_kvec_add(con
, sizeof(tag_keepalive
), &tag_keepalive
);
1399 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1403 * Connection negotiation.
1406 static int get_connect_authorizer(struct ceph_connection
*con
)
1408 struct ceph_auth_handshake
*auth
;
1411 if (!con
->ops
->get_authorizer
) {
1413 con
->out_connect
.authorizer_protocol
= CEPH_AUTH_UNKNOWN
;
1414 con
->out_connect
.authorizer_len
= 0;
1418 auth
= con
->ops
->get_authorizer(con
, &auth_proto
, con
->auth_retry
);
1420 return PTR_ERR(auth
);
1423 con
->out_connect
.authorizer_protocol
= cpu_to_le32(auth_proto
);
1424 con
->out_connect
.authorizer_len
= cpu_to_le32(auth
->authorizer_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 void __prepare_write_connect(struct ceph_connection
*con
)
1443 con_out_kvec_add(con
, sizeof(con
->out_connect
), &con
->out_connect
);
1445 con_out_kvec_add(con
, con
->auth
->authorizer_buf_len
,
1446 con
->auth
->authorizer_buf
);
1449 con_flag_set(con
, CON_FLAG_WRITE_PENDING
);
1452 static int prepare_write_connect(struct ceph_connection
*con
)
1454 unsigned int global_seq
= get_global_seq(con
->msgr
, 0);
1458 switch (con
->peer_name
.type
) {
1459 case CEPH_ENTITY_TYPE_MON
:
1460 proto
= CEPH_MONC_PROTOCOL
;
1462 case CEPH_ENTITY_TYPE_OSD
:
1463 proto
= CEPH_OSDC_PROTOCOL
;
1465 case CEPH_ENTITY_TYPE_MDS
:
1466 proto
= CEPH_MDSC_PROTOCOL
;
1472 dout("prepare_write_connect %p cseq=%d gseq=%d proto=%d\n", con
,
1473 con
->connect_seq
, global_seq
, proto
);
1475 con
->out_connect
.features
=
1476 cpu_to_le64(from_msgr(con
->msgr
)->supported_features
);
1477 con
->out_connect
.host_type
= cpu_to_le32(CEPH_ENTITY_TYPE_CLIENT
);
1478 con
->out_connect
.connect_seq
= cpu_to_le32(con
->connect_seq
);
1479 con
->out_connect
.global_seq
= cpu_to_le32(global_seq
);
1480 con
->out_connect
.protocol_version
= cpu_to_le32(proto
);
1481 con
->out_connect
.flags
= 0;
1483 ret
= get_connect_authorizer(con
);
1487 __prepare_write_connect(con
);
1492 * write as much of pending kvecs to the socket as we can.
1494 * 0 -> socket full, but more to do
1497 static int write_partial_kvec(struct ceph_connection
*con
)
1501 dout("write_partial_kvec %p %d left\n", con
, con
->out_kvec_bytes
);
1502 while (con
->out_kvec_bytes
> 0) {
1503 ret
= ceph_tcp_sendmsg(con
->sock
, con
->out_kvec_cur
,
1504 con
->out_kvec_left
, con
->out_kvec_bytes
,
1508 con
->out_kvec_bytes
-= ret
;
1509 if (con
->out_kvec_bytes
== 0)
1512 /* account for full iov entries consumed */
1513 while (ret
>= con
->out_kvec_cur
->iov_len
) {
1514 BUG_ON(!con
->out_kvec_left
);
1515 ret
-= con
->out_kvec_cur
->iov_len
;
1516 con
->out_kvec_cur
++;
1517 con
->out_kvec_left
--;
1519 /* and for a partially-consumed entry */
1521 con
->out_kvec_cur
->iov_len
-= ret
;
1522 con
->out_kvec_cur
->iov_base
+= ret
;
1525 con
->out_kvec_left
= 0;
1528 dout("write_partial_kvec %p %d left in %d kvecs ret = %d\n", con
,
1529 con
->out_kvec_bytes
, con
->out_kvec_left
, ret
);
1530 return ret
; /* done! */
1533 static u32
ceph_crc32c_page(u32 crc
, struct page
*page
,
1534 unsigned int page_offset
,
1535 unsigned int length
)
1540 BUG_ON(kaddr
== NULL
);
1541 crc
= crc32c(crc
, kaddr
+ page_offset
, length
);
1547 * Write as much message data payload as we can. If we finish, queue
1549 * 1 -> done, footer is now queued in out_kvec[].
1550 * 0 -> socket full, but more to do
1553 static int write_partial_message_data(struct ceph_connection
*con
)
1555 struct ceph_msg
*msg
= con
->out_msg
;
1556 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
1557 bool do_datacrc
= !ceph_test_opt(from_msgr(con
->msgr
), NOCRC
);
1558 int more
= MSG_MORE
| MSG_SENDPAGE_NOTLAST
;
1561 dout("%s %p msg %p\n", __func__
, con
, msg
);
1563 if (!msg
->num_data_items
)
1567 * Iterate through each page that contains data to be
1568 * written, and send as much as possible for each.
1570 * If we are calculating the data crc (the default), we will
1571 * need to map the page. If we have no pages, they have
1572 * been revoked, so use the zero page.
1574 crc
= do_datacrc
? le32_to_cpu(msg
->footer
.data_crc
) : 0;
1575 while (cursor
->total_resid
) {
1581 if (!cursor
->resid
) {
1582 ceph_msg_data_advance(cursor
, 0);
1586 page
= ceph_msg_data_next(cursor
, &page_offset
, &length
, NULL
);
1587 if (length
== cursor
->total_resid
)
1589 ret
= ceph_tcp_sendpage(con
->sock
, page
, page_offset
, length
,
1593 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1597 if (do_datacrc
&& cursor
->need_crc
)
1598 crc
= ceph_crc32c_page(crc
, page
, page_offset
, length
);
1599 ceph_msg_data_advance(cursor
, (size_t)ret
);
1602 dout("%s %p msg %p done\n", __func__
, con
, msg
);
1604 /* prepare and queue up footer, too */
1606 msg
->footer
.data_crc
= cpu_to_le32(crc
);
1608 msg
->footer
.flags
|= CEPH_MSG_FOOTER_NOCRC
;
1609 con_out_kvec_reset(con
);
1610 prepare_write_message_footer(con
);
1612 return 1; /* must return > 0 to indicate success */
1618 static int write_partial_skip(struct ceph_connection
*con
)
1620 int more
= MSG_MORE
| MSG_SENDPAGE_NOTLAST
;
1623 dout("%s %p %d left\n", __func__
, con
, con
->out_skip
);
1624 while (con
->out_skip
> 0) {
1625 size_t size
= min(con
->out_skip
, (int) PAGE_SIZE
);
1627 if (size
== con
->out_skip
)
1629 ret
= ceph_tcp_sendpage(con
->sock
, zero_page
, 0, size
, more
);
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
);
1758 if (size
> con
->auth
->authorizer_reply_buf_len
) {
1759 pr_err("authorizer reply too big: %d > %zu\n", size
,
1760 con
->auth
->authorizer_reply_buf_len
);
1766 ret
= read_partial(con
, end
, size
,
1767 con
->auth
->authorizer_reply_buf
);
1772 dout("read_partial_connect %p tag %d, con_seq = %u, g_seq = %u\n",
1773 con
, (int)con
->in_reply
.tag
,
1774 le32_to_cpu(con
->in_reply
.connect_seq
),
1775 le32_to_cpu(con
->in_reply
.global_seq
));
1781 * Verify the hello banner looks okay.
1783 static int verify_hello(struct ceph_connection
*con
)
1785 if (memcmp(con
->in_banner
, CEPH_BANNER
, strlen(CEPH_BANNER
))) {
1786 pr_err("connect to %s got bad banner\n",
1787 ceph_pr_addr(&con
->peer_addr
.in_addr
));
1788 con
->error_msg
= "protocol error, bad banner";
1794 static bool addr_is_blank(struct sockaddr_storage
*ss
)
1796 struct in_addr
*addr
= &((struct sockaddr_in
*)ss
)->sin_addr
;
1797 struct in6_addr
*addr6
= &((struct sockaddr_in6
*)ss
)->sin6_addr
;
1799 switch (ss
->ss_family
) {
1801 return addr
->s_addr
== htonl(INADDR_ANY
);
1803 return ipv6_addr_any(addr6
);
1809 static int addr_port(struct sockaddr_storage
*ss
)
1811 switch (ss
->ss_family
) {
1813 return ntohs(((struct sockaddr_in
*)ss
)->sin_port
);
1815 return ntohs(((struct sockaddr_in6
*)ss
)->sin6_port
);
1820 static void addr_set_port(struct sockaddr_storage
*ss
, int p
)
1822 switch (ss
->ss_family
) {
1824 ((struct sockaddr_in
*)ss
)->sin_port
= htons(p
);
1827 ((struct sockaddr_in6
*)ss
)->sin6_port
= htons(p
);
1833 * Unlike other *_pton function semantics, zero indicates success.
1835 static int ceph_pton(const char *str
, size_t len
, struct sockaddr_storage
*ss
,
1836 char delim
, const char **ipend
)
1838 struct sockaddr_in
*in4
= (struct sockaddr_in
*) ss
;
1839 struct sockaddr_in6
*in6
= (struct sockaddr_in6
*) ss
;
1841 memset(ss
, 0, sizeof(*ss
));
1843 if (in4_pton(str
, len
, (u8
*)&in4
->sin_addr
.s_addr
, delim
, ipend
)) {
1844 ss
->ss_family
= AF_INET
;
1848 if (in6_pton(str
, len
, (u8
*)&in6
->sin6_addr
.s6_addr
, delim
, ipend
)) {
1849 ss
->ss_family
= AF_INET6
;
1857 * Extract hostname string and resolve using kernel DNS facility.
1859 #ifdef CONFIG_CEPH_LIB_USE_DNS_RESOLVER
1860 static int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1861 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1863 const char *end
, *delim_p
;
1864 char *colon_p
, *ip_addr
= NULL
;
1868 * The end of the hostname occurs immediately preceding the delimiter or
1869 * the port marker (':') where the delimiter takes precedence.
1871 delim_p
= memchr(name
, delim
, namelen
);
1872 colon_p
= memchr(name
, ':', namelen
);
1874 if (delim_p
&& colon_p
)
1875 end
= delim_p
< colon_p
? delim_p
: colon_p
;
1876 else if (!delim_p
&& colon_p
)
1880 if (!end
) /* case: hostname:/ */
1881 end
= name
+ namelen
;
1887 /* do dns_resolve upcall */
1888 ip_len
= dns_query(NULL
, name
, end
- name
, NULL
, &ip_addr
, NULL
);
1890 ret
= ceph_pton(ip_addr
, ip_len
, ss
, -1, NULL
);
1898 pr_info("resolve '%.*s' (ret=%d): %s\n", (int)(end
- name
), name
,
1899 ret
, ret
? "failed" : ceph_pr_addr(ss
));
1904 static inline int ceph_dns_resolve_name(const char *name
, size_t namelen
,
1905 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1912 * Parse a server name (IP or hostname). If a valid IP address is not found
1913 * then try to extract a hostname to resolve using userspace DNS upcall.
1915 static int ceph_parse_server_name(const char *name
, size_t namelen
,
1916 struct sockaddr_storage
*ss
, char delim
, const char **ipend
)
1920 ret
= ceph_pton(name
, namelen
, ss
, delim
, ipend
);
1922 ret
= ceph_dns_resolve_name(name
, namelen
, ss
, delim
, ipend
);
1928 * Parse an ip[:port] list into an addr array. Use the default
1929 * monitor port if a port isn't specified.
1931 int ceph_parse_ips(const char *c
, const char *end
,
1932 struct ceph_entity_addr
*addr
,
1933 int max_count
, int *count
)
1935 int i
, ret
= -EINVAL
;
1938 dout("parse_ips on '%.*s'\n", (int)(end
-c
), c
);
1939 for (i
= 0; i
< max_count
; i
++) {
1941 struct sockaddr_storage
*ss
= &addr
[i
].in_addr
;
1950 ret
= ceph_parse_server_name(p
, end
- p
, ss
, delim
, &ipend
);
1959 dout("missing matching ']'\n");
1966 if (p
< end
&& *p
== ':') {
1969 while (p
< end
&& *p
>= '0' && *p
<= '9') {
1970 port
= (port
* 10) + (*p
- '0');
1974 port
= CEPH_MON_PORT
;
1975 else if (port
> 65535)
1978 port
= CEPH_MON_PORT
;
1981 addr_set_port(ss
, port
);
1983 dout("parse_ips got %s\n", ceph_pr_addr(ss
));
2000 pr_err("parse_ips bad ip '%.*s'\n", (int)(end
- c
), c
);
2003 EXPORT_SYMBOL(ceph_parse_ips
);
2005 static int process_banner(struct ceph_connection
*con
)
2007 dout("process_banner on %p\n", con
);
2009 if (verify_hello(con
) < 0)
2012 ceph_decode_addr(&con
->actual_peer_addr
);
2013 ceph_decode_addr(&con
->peer_addr_for_me
);
2016 * Make sure the other end is who we wanted. note that the other
2017 * end may not yet know their ip address, so if it's 0.0.0.0, give
2018 * them the benefit of the doubt.
2020 if (memcmp(&con
->peer_addr
, &con
->actual_peer_addr
,
2021 sizeof(con
->peer_addr
)) != 0 &&
2022 !(addr_is_blank(&con
->actual_peer_addr
.in_addr
) &&
2023 con
->actual_peer_addr
.nonce
== con
->peer_addr
.nonce
)) {
2024 pr_warn("wrong peer, want %s/%d, got %s/%d\n",
2025 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2026 (int)le32_to_cpu(con
->peer_addr
.nonce
),
2027 ceph_pr_addr(&con
->actual_peer_addr
.in_addr
),
2028 (int)le32_to_cpu(con
->actual_peer_addr
.nonce
));
2029 con
->error_msg
= "wrong peer at address";
2034 * did we learn our address?
2036 if (addr_is_blank(&con
->msgr
->inst
.addr
.in_addr
)) {
2037 int port
= addr_port(&con
->msgr
->inst
.addr
.in_addr
);
2039 memcpy(&con
->msgr
->inst
.addr
.in_addr
,
2040 &con
->peer_addr_for_me
.in_addr
,
2041 sizeof(con
->peer_addr_for_me
.in_addr
));
2042 addr_set_port(&con
->msgr
->inst
.addr
.in_addr
, port
);
2043 encode_my_addr(con
->msgr
);
2044 dout("process_banner learned my addr is %s\n",
2045 ceph_pr_addr(&con
->msgr
->inst
.addr
.in_addr
));
2051 static int process_connect(struct ceph_connection
*con
)
2053 u64 sup_feat
= from_msgr(con
->msgr
)->supported_features
;
2054 u64 req_feat
= from_msgr(con
->msgr
)->required_features
;
2055 u64 server_feat
= le64_to_cpu(con
->in_reply
.features
);
2058 dout("process_connect on %p tag %d\n", con
, (int)con
->in_tag
);
2062 * Any connection that defines ->get_authorizer()
2063 * should also define ->add_authorizer_challenge() and
2064 * ->verify_authorizer_reply().
2066 * See get_connect_authorizer().
2068 if (con
->in_reply
.tag
== CEPH_MSGR_TAG_CHALLENGE_AUTHORIZER
) {
2069 ret
= con
->ops
->add_authorizer_challenge(
2070 con
, con
->auth
->authorizer_reply_buf
,
2071 le32_to_cpu(con
->in_reply
.authorizer_len
));
2075 con_out_kvec_reset(con
);
2076 __prepare_write_connect(con
);
2077 prepare_read_connect(con
);
2081 ret
= con
->ops
->verify_authorizer_reply(con
);
2083 con
->error_msg
= "bad authorize reply";
2088 switch (con
->in_reply
.tag
) {
2089 case CEPH_MSGR_TAG_FEATURES
:
2090 pr_err("%s%lld %s feature set mismatch,"
2091 " my %llx < server's %llx, missing %llx\n",
2092 ENTITY_NAME(con
->peer_name
),
2093 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2094 sup_feat
, server_feat
, server_feat
& ~sup_feat
);
2095 con
->error_msg
= "missing required protocol features";
2096 reset_connection(con
);
2099 case CEPH_MSGR_TAG_BADPROTOVER
:
2100 pr_err("%s%lld %s protocol version mismatch,"
2101 " my %d != server's %d\n",
2102 ENTITY_NAME(con
->peer_name
),
2103 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2104 le32_to_cpu(con
->out_connect
.protocol_version
),
2105 le32_to_cpu(con
->in_reply
.protocol_version
));
2106 con
->error_msg
= "protocol version mismatch";
2107 reset_connection(con
);
2110 case CEPH_MSGR_TAG_BADAUTHORIZER
:
2112 dout("process_connect %p got BADAUTHORIZER attempt %d\n", con
,
2114 if (con
->auth_retry
== 2) {
2115 con
->error_msg
= "connect authorization failure";
2118 con_out_kvec_reset(con
);
2119 ret
= prepare_write_connect(con
);
2122 prepare_read_connect(con
);
2125 case CEPH_MSGR_TAG_RESETSESSION
:
2127 * If we connected with a large connect_seq but the peer
2128 * has no record of a session with us (no connection, or
2129 * connect_seq == 0), they will send RESETSESION to indicate
2130 * that they must have reset their session, and may have
2133 dout("process_connect got RESET peer seq %u\n",
2134 le32_to_cpu(con
->in_reply
.connect_seq
));
2135 pr_err("%s%lld %s connection reset\n",
2136 ENTITY_NAME(con
->peer_name
),
2137 ceph_pr_addr(&con
->peer_addr
.in_addr
));
2138 reset_connection(con
);
2139 con_out_kvec_reset(con
);
2140 ret
= prepare_write_connect(con
);
2143 prepare_read_connect(con
);
2145 /* Tell ceph about it. */
2146 mutex_unlock(&con
->mutex
);
2147 pr_info("reset on %s%lld\n", ENTITY_NAME(con
->peer_name
));
2148 if (con
->ops
->peer_reset
)
2149 con
->ops
->peer_reset(con
);
2150 mutex_lock(&con
->mutex
);
2151 if (con
->state
!= CON_STATE_NEGOTIATING
)
2155 case CEPH_MSGR_TAG_RETRY_SESSION
:
2157 * If we sent a smaller connect_seq than the peer has, try
2158 * again with a larger value.
2160 dout("process_connect got RETRY_SESSION my seq %u, peer %u\n",
2161 le32_to_cpu(con
->out_connect
.connect_seq
),
2162 le32_to_cpu(con
->in_reply
.connect_seq
));
2163 con
->connect_seq
= le32_to_cpu(con
->in_reply
.connect_seq
);
2164 con_out_kvec_reset(con
);
2165 ret
= prepare_write_connect(con
);
2168 prepare_read_connect(con
);
2171 case CEPH_MSGR_TAG_RETRY_GLOBAL
:
2173 * If we sent a smaller global_seq than the peer has, try
2174 * again with a larger value.
2176 dout("process_connect got RETRY_GLOBAL my %u peer_gseq %u\n",
2177 con
->peer_global_seq
,
2178 le32_to_cpu(con
->in_reply
.global_seq
));
2179 get_global_seq(con
->msgr
,
2180 le32_to_cpu(con
->in_reply
.global_seq
));
2181 con_out_kvec_reset(con
);
2182 ret
= prepare_write_connect(con
);
2185 prepare_read_connect(con
);
2188 case CEPH_MSGR_TAG_SEQ
:
2189 case CEPH_MSGR_TAG_READY
:
2190 if (req_feat
& ~server_feat
) {
2191 pr_err("%s%lld %s protocol feature mismatch,"
2192 " my required %llx > server's %llx, need %llx\n",
2193 ENTITY_NAME(con
->peer_name
),
2194 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2195 req_feat
, server_feat
, req_feat
& ~server_feat
);
2196 con
->error_msg
= "missing required protocol features";
2197 reset_connection(con
);
2201 WARN_ON(con
->state
!= CON_STATE_NEGOTIATING
);
2202 con
->state
= CON_STATE_OPEN
;
2203 con
->auth_retry
= 0; /* we authenticated; clear flag */
2204 con
->peer_global_seq
= le32_to_cpu(con
->in_reply
.global_seq
);
2206 con
->peer_features
= server_feat
;
2207 dout("process_connect got READY gseq %d cseq %d (%d)\n",
2208 con
->peer_global_seq
,
2209 le32_to_cpu(con
->in_reply
.connect_seq
),
2211 WARN_ON(con
->connect_seq
!=
2212 le32_to_cpu(con
->in_reply
.connect_seq
));
2214 if (con
->in_reply
.flags
& CEPH_MSG_CONNECT_LOSSY
)
2215 con_flag_set(con
, CON_FLAG_LOSSYTX
);
2217 con
->delay
= 0; /* reset backoff memory */
2219 if (con
->in_reply
.tag
== CEPH_MSGR_TAG_SEQ
) {
2220 prepare_write_seq(con
);
2221 prepare_read_seq(con
);
2223 prepare_read_tag(con
);
2227 case CEPH_MSGR_TAG_WAIT
:
2229 * If there is a connection race (we are opening
2230 * connections to each other), one of us may just have
2231 * to WAIT. This shouldn't happen if we are the
2234 con
->error_msg
= "protocol error, got WAIT as client";
2238 con
->error_msg
= "protocol error, garbage tag during connect";
2246 * read (part of) an ack
2248 static int read_partial_ack(struct ceph_connection
*con
)
2250 int size
= sizeof (con
->in_temp_ack
);
2253 return read_partial(con
, end
, size
, &con
->in_temp_ack
);
2257 * We can finally discard anything that's been acked.
2259 static void process_ack(struct ceph_connection
*con
)
2262 u64 ack
= le64_to_cpu(con
->in_temp_ack
);
2264 bool reconnect
= (con
->in_tag
== CEPH_MSGR_TAG_SEQ
);
2265 struct list_head
*list
= reconnect
? &con
->out_queue
: &con
->out_sent
;
2268 * In the reconnect case, con_fault() has requeued messages
2269 * in out_sent. We should cleanup old messages according to
2270 * the reconnect seq.
2272 while (!list_empty(list
)) {
2273 m
= list_first_entry(list
, struct ceph_msg
, list_head
);
2274 if (reconnect
&& m
->needs_out_seq
)
2276 seq
= le64_to_cpu(m
->hdr
.seq
);
2279 dout("got ack for seq %llu type %d at %p\n", seq
,
2280 le16_to_cpu(m
->hdr
.type
), m
);
2281 m
->ack_stamp
= jiffies
;
2285 prepare_read_tag(con
);
2289 static int read_partial_message_section(struct ceph_connection
*con
,
2290 struct kvec
*section
,
2291 unsigned int sec_len
, u32
*crc
)
2297 while (section
->iov_len
< sec_len
) {
2298 BUG_ON(section
->iov_base
== NULL
);
2299 left
= sec_len
- section
->iov_len
;
2300 ret
= ceph_tcp_recvmsg(con
->sock
, (char *)section
->iov_base
+
2301 section
->iov_len
, left
);
2304 section
->iov_len
+= ret
;
2306 if (section
->iov_len
== sec_len
)
2307 *crc
= crc32c(0, section
->iov_base
, section
->iov_len
);
2312 static int read_partial_msg_data(struct ceph_connection
*con
)
2314 struct ceph_msg
*msg
= con
->in_msg
;
2315 struct ceph_msg_data_cursor
*cursor
= &msg
->cursor
;
2316 bool do_datacrc
= !ceph_test_opt(from_msgr(con
->msgr
), NOCRC
);
2323 if (!msg
->num_data_items
)
2327 crc
= con
->in_data_crc
;
2328 while (cursor
->total_resid
) {
2329 if (!cursor
->resid
) {
2330 ceph_msg_data_advance(cursor
, 0);
2334 page
= ceph_msg_data_next(cursor
, &page_offset
, &length
, NULL
);
2335 ret
= ceph_tcp_recvpage(con
->sock
, page
, page_offset
, length
);
2338 con
->in_data_crc
= crc
;
2344 crc
= ceph_crc32c_page(crc
, page
, page_offset
, ret
);
2345 ceph_msg_data_advance(cursor
, (size_t)ret
);
2348 con
->in_data_crc
= crc
;
2350 return 1; /* must return > 0 to indicate success */
2354 * read (part of) a message.
2356 static int ceph_con_in_msg_alloc(struct ceph_connection
*con
, int *skip
);
2358 static int read_partial_message(struct ceph_connection
*con
)
2360 struct ceph_msg
*m
= con
->in_msg
;
2364 unsigned int front_len
, middle_len
, data_len
;
2365 bool do_datacrc
= !ceph_test_opt(from_msgr(con
->msgr
), NOCRC
);
2366 bool need_sign
= (con
->peer_features
& CEPH_FEATURE_MSG_AUTH
);
2370 dout("read_partial_message con %p msg %p\n", con
, m
);
2373 size
= sizeof (con
->in_hdr
);
2375 ret
= read_partial(con
, end
, size
, &con
->in_hdr
);
2379 crc
= crc32c(0, &con
->in_hdr
, offsetof(struct ceph_msg_header
, crc
));
2380 if (cpu_to_le32(crc
) != con
->in_hdr
.crc
) {
2381 pr_err("read_partial_message bad hdr crc %u != expected %u\n",
2382 crc
, con
->in_hdr
.crc
);
2386 front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
2387 if (front_len
> CEPH_MSG_MAX_FRONT_LEN
)
2389 middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
2390 if (middle_len
> CEPH_MSG_MAX_MIDDLE_LEN
)
2392 data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
2393 if (data_len
> CEPH_MSG_MAX_DATA_LEN
)
2397 seq
= le64_to_cpu(con
->in_hdr
.seq
);
2398 if ((s64
)seq
- (s64
)con
->in_seq
< 1) {
2399 pr_info("skipping %s%lld %s seq %lld expected %lld\n",
2400 ENTITY_NAME(con
->peer_name
),
2401 ceph_pr_addr(&con
->peer_addr
.in_addr
),
2402 seq
, con
->in_seq
+ 1);
2403 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
2405 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2407 } else if ((s64
)seq
- (s64
)con
->in_seq
> 1) {
2408 pr_err("read_partial_message bad seq %lld expected %lld\n",
2409 seq
, con
->in_seq
+ 1);
2410 con
->error_msg
= "bad message sequence # for incoming message";
2414 /* allocate message? */
2418 dout("got hdr type %d front %d data %d\n", con
->in_hdr
.type
,
2419 front_len
, data_len
);
2420 ret
= ceph_con_in_msg_alloc(con
, &skip
);
2424 BUG_ON(!con
->in_msg
^ skip
);
2426 /* skip this message */
2427 dout("alloc_msg said skip message\n");
2428 con
->in_base_pos
= -front_len
- middle_len
- data_len
-
2430 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2435 BUG_ON(!con
->in_msg
);
2436 BUG_ON(con
->in_msg
->con
!= con
);
2438 m
->front
.iov_len
= 0; /* haven't read it yet */
2440 m
->middle
->vec
.iov_len
= 0;
2442 /* prepare for data payload, if any */
2445 prepare_message_data(con
->in_msg
, data_len
);
2449 ret
= read_partial_message_section(con
, &m
->front
, front_len
,
2450 &con
->in_front_crc
);
2456 ret
= read_partial_message_section(con
, &m
->middle
->vec
,
2458 &con
->in_middle_crc
);
2465 ret
= read_partial_msg_data(con
);
2471 size
= sizeof_footer(con
);
2473 ret
= read_partial(con
, end
, size
, &m
->footer
);
2478 m
->footer
.flags
= m
->old_footer
.flags
;
2482 dout("read_partial_message got msg %p %d (%u) + %d (%u) + %d (%u)\n",
2483 m
, front_len
, m
->footer
.front_crc
, middle_len
,
2484 m
->footer
.middle_crc
, data_len
, m
->footer
.data_crc
);
2487 if (con
->in_front_crc
!= le32_to_cpu(m
->footer
.front_crc
)) {
2488 pr_err("read_partial_message %p front crc %u != exp. %u\n",
2489 m
, con
->in_front_crc
, m
->footer
.front_crc
);
2492 if (con
->in_middle_crc
!= le32_to_cpu(m
->footer
.middle_crc
)) {
2493 pr_err("read_partial_message %p middle crc %u != exp %u\n",
2494 m
, con
->in_middle_crc
, m
->footer
.middle_crc
);
2498 (m
->footer
.flags
& CEPH_MSG_FOOTER_NOCRC
) == 0 &&
2499 con
->in_data_crc
!= le32_to_cpu(m
->footer
.data_crc
)) {
2500 pr_err("read_partial_message %p data crc %u != exp. %u\n", m
,
2501 con
->in_data_crc
, le32_to_cpu(m
->footer
.data_crc
));
2505 if (need_sign
&& con
->ops
->check_message_signature
&&
2506 con
->ops
->check_message_signature(m
)) {
2507 pr_err("read_partial_message %p signature check failed\n", m
);
2511 return 1; /* done! */
2515 * Process message. This happens in the worker thread. The callback should
2516 * be careful not to do anything that waits on other incoming messages or it
2519 static void process_message(struct ceph_connection
*con
)
2521 struct ceph_msg
*msg
= con
->in_msg
;
2523 BUG_ON(con
->in_msg
->con
!= con
);
2526 /* if first message, set peer_name */
2527 if (con
->peer_name
.type
== 0)
2528 con
->peer_name
= msg
->hdr
.src
;
2531 mutex_unlock(&con
->mutex
);
2533 dout("===== %p %llu from %s%lld %d=%s len %d+%d (%u %u %u) =====\n",
2534 msg
, le64_to_cpu(msg
->hdr
.seq
),
2535 ENTITY_NAME(msg
->hdr
.src
),
2536 le16_to_cpu(msg
->hdr
.type
),
2537 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
2538 le32_to_cpu(msg
->hdr
.front_len
),
2539 le32_to_cpu(msg
->hdr
.data_len
),
2540 con
->in_front_crc
, con
->in_middle_crc
, con
->in_data_crc
);
2541 con
->ops
->dispatch(con
, msg
);
2543 mutex_lock(&con
->mutex
);
2546 static int read_keepalive_ack(struct ceph_connection
*con
)
2548 struct ceph_timespec ceph_ts
;
2549 size_t size
= sizeof(ceph_ts
);
2550 int ret
= read_partial(con
, size
, size
, &ceph_ts
);
2553 ceph_decode_timespec64(&con
->last_keepalive_ack
, &ceph_ts
);
2554 prepare_read_tag(con
);
2559 * Write something to the socket. Called in a worker thread when the
2560 * socket appears to be writeable and we have something ready to send.
2562 static int try_write(struct ceph_connection
*con
)
2566 dout("try_write start %p state %lu\n", con
, con
->state
);
2567 if (con
->state
!= CON_STATE_PREOPEN
&&
2568 con
->state
!= CON_STATE_CONNECTING
&&
2569 con
->state
!= CON_STATE_NEGOTIATING
&&
2570 con
->state
!= CON_STATE_OPEN
)
2573 /* open the socket first? */
2574 if (con
->state
== CON_STATE_PREOPEN
) {
2576 con
->state
= CON_STATE_CONNECTING
;
2578 con_out_kvec_reset(con
);
2579 prepare_write_banner(con
);
2580 prepare_read_banner(con
);
2582 BUG_ON(con
->in_msg
);
2583 con
->in_tag
= CEPH_MSGR_TAG_READY
;
2584 dout("try_write initiating connect on %p new state %lu\n",
2586 ret
= ceph_tcp_connect(con
);
2588 con
->error_msg
= "connect error";
2594 dout("try_write out_kvec_bytes %d\n", con
->out_kvec_bytes
);
2597 /* kvec data queued? */
2598 if (con
->out_kvec_left
) {
2599 ret
= write_partial_kvec(con
);
2603 if (con
->out_skip
) {
2604 ret
= write_partial_skip(con
);
2611 if (con
->out_msg_done
) {
2612 ceph_msg_put(con
->out_msg
);
2613 con
->out_msg
= NULL
; /* we're done with this one */
2617 ret
= write_partial_message_data(con
);
2619 goto more
; /* we need to send the footer, too! */
2623 dout("try_write write_partial_message_data err %d\n",
2630 if (con
->state
== CON_STATE_OPEN
) {
2631 if (con_flag_test_and_clear(con
, CON_FLAG_KEEPALIVE_PENDING
)) {
2632 prepare_write_keepalive(con
);
2635 /* is anything else pending? */
2636 if (!list_empty(&con
->out_queue
)) {
2637 prepare_write_message(con
);
2640 if (con
->in_seq
> con
->in_seq_acked
) {
2641 prepare_write_ack(con
);
2646 /* Nothing to do! */
2647 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
2648 dout("try_write nothing else to write.\n");
2651 dout("try_write done on %p ret %d\n", con
, ret
);
2656 * Read what we can from the socket.
2658 static int try_read(struct ceph_connection
*con
)
2663 dout("try_read start on %p state %lu\n", con
, con
->state
);
2664 if (con
->state
!= CON_STATE_CONNECTING
&&
2665 con
->state
!= CON_STATE_NEGOTIATING
&&
2666 con
->state
!= CON_STATE_OPEN
)
2671 dout("try_read tag %d in_base_pos %d\n", (int)con
->in_tag
,
2674 if (con
->state
== CON_STATE_CONNECTING
) {
2675 dout("try_read connecting\n");
2676 ret
= read_partial_banner(con
);
2679 ret
= process_banner(con
);
2683 con
->state
= CON_STATE_NEGOTIATING
;
2686 * Received banner is good, exchange connection info.
2687 * Do not reset out_kvec, as sending our banner raced
2688 * with receiving peer banner after connect completed.
2690 ret
= prepare_write_connect(con
);
2693 prepare_read_connect(con
);
2695 /* Send connection info before awaiting response */
2699 if (con
->state
== CON_STATE_NEGOTIATING
) {
2700 dout("try_read negotiating\n");
2701 ret
= read_partial_connect(con
);
2704 ret
= process_connect(con
);
2710 WARN_ON(con
->state
!= CON_STATE_OPEN
);
2712 if (con
->in_base_pos
< 0) {
2714 * skipping + discarding content.
2716 ret
= ceph_tcp_recvmsg(con
->sock
, NULL
, -con
->in_base_pos
);
2719 dout("skipped %d / %d bytes\n", ret
, -con
->in_base_pos
);
2720 con
->in_base_pos
+= ret
;
2721 if (con
->in_base_pos
)
2724 if (con
->in_tag
== CEPH_MSGR_TAG_READY
) {
2728 ret
= ceph_tcp_recvmsg(con
->sock
, &con
->in_tag
, 1);
2731 dout("try_read got tag %d\n", (int)con
->in_tag
);
2732 switch (con
->in_tag
) {
2733 case CEPH_MSGR_TAG_MSG
:
2734 prepare_read_message(con
);
2736 case CEPH_MSGR_TAG_ACK
:
2737 prepare_read_ack(con
);
2739 case CEPH_MSGR_TAG_KEEPALIVE2_ACK
:
2740 prepare_read_keepalive_ack(con
);
2742 case CEPH_MSGR_TAG_CLOSE
:
2743 con_close_socket(con
);
2744 con
->state
= CON_STATE_CLOSED
;
2750 if (con
->in_tag
== CEPH_MSGR_TAG_MSG
) {
2751 ret
= read_partial_message(con
);
2755 con
->error_msg
= "bad crc/signature";
2761 con
->error_msg
= "io error";
2766 if (con
->in_tag
== CEPH_MSGR_TAG_READY
)
2768 process_message(con
);
2769 if (con
->state
== CON_STATE_OPEN
)
2770 prepare_read_tag(con
);
2773 if (con
->in_tag
== CEPH_MSGR_TAG_ACK
||
2774 con
->in_tag
== CEPH_MSGR_TAG_SEQ
) {
2776 * the final handshake seq exchange is semantically
2777 * equivalent to an ACK
2779 ret
= read_partial_ack(con
);
2785 if (con
->in_tag
== CEPH_MSGR_TAG_KEEPALIVE2_ACK
) {
2786 ret
= read_keepalive_ack(con
);
2793 dout("try_read done on %p ret %d\n", con
, ret
);
2797 pr_err("try_read bad con->in_tag = %d\n", (int)con
->in_tag
);
2798 con
->error_msg
= "protocol error, garbage tag";
2805 * Atomically queue work on a connection after the specified delay.
2806 * Bump @con reference to avoid races with connection teardown.
2807 * Returns 0 if work was queued, or an error code otherwise.
2809 static int queue_con_delay(struct ceph_connection
*con
, unsigned long delay
)
2811 if (!con
->ops
->get(con
)) {
2812 dout("%s %p ref count 0\n", __func__
, con
);
2816 if (!queue_delayed_work(ceph_msgr_wq
, &con
->work
, delay
)) {
2817 dout("%s %p - already queued\n", __func__
, con
);
2822 dout("%s %p %lu\n", __func__
, con
, delay
);
2826 static void queue_con(struct ceph_connection
*con
)
2828 (void) queue_con_delay(con
, 0);
2831 static void cancel_con(struct ceph_connection
*con
)
2833 if (cancel_delayed_work(&con
->work
)) {
2834 dout("%s %p\n", __func__
, con
);
2839 static bool con_sock_closed(struct ceph_connection
*con
)
2841 if (!con_flag_test_and_clear(con
, CON_FLAG_SOCK_CLOSED
))
2845 case CON_STATE_ ## x: \
2846 con->error_msg = "socket closed (con state " #x ")"; \
2849 switch (con
->state
) {
2857 pr_warn("%s con %p unrecognized state %lu\n",
2858 __func__
, con
, con
->state
);
2859 con
->error_msg
= "unrecognized con state";
2868 static bool con_backoff(struct ceph_connection
*con
)
2872 if (!con_flag_test_and_clear(con
, CON_FLAG_BACKOFF
))
2875 ret
= queue_con_delay(con
, round_jiffies_relative(con
->delay
));
2877 dout("%s: con %p FAILED to back off %lu\n", __func__
,
2879 BUG_ON(ret
== -ENOENT
);
2880 con_flag_set(con
, CON_FLAG_BACKOFF
);
2886 /* Finish fault handling; con->mutex must *not* be held here */
2888 static void con_fault_finish(struct ceph_connection
*con
)
2890 dout("%s %p\n", __func__
, con
);
2893 * in case we faulted due to authentication, invalidate our
2894 * current tickets so that we can get new ones.
2896 if (con
->auth_retry
) {
2897 dout("auth_retry %d, invalidating\n", con
->auth_retry
);
2898 if (con
->ops
->invalidate_authorizer
)
2899 con
->ops
->invalidate_authorizer(con
);
2900 con
->auth_retry
= 0;
2903 if (con
->ops
->fault
)
2904 con
->ops
->fault(con
);
2908 * Do some work on a connection. Drop a connection ref when we're done.
2910 static void ceph_con_workfn(struct work_struct
*work
)
2912 struct ceph_connection
*con
= container_of(work
, struct ceph_connection
,
2916 mutex_lock(&con
->mutex
);
2920 if ((fault
= con_sock_closed(con
))) {
2921 dout("%s: con %p SOCK_CLOSED\n", __func__
, con
);
2924 if (con_backoff(con
)) {
2925 dout("%s: con %p BACKOFF\n", __func__
, con
);
2928 if (con
->state
== CON_STATE_STANDBY
) {
2929 dout("%s: con %p STANDBY\n", __func__
, con
);
2932 if (con
->state
== CON_STATE_CLOSED
) {
2933 dout("%s: con %p CLOSED\n", __func__
, con
);
2937 if (con
->state
== CON_STATE_PREOPEN
) {
2938 dout("%s: con %p PREOPEN\n", __func__
, con
);
2942 ret
= try_read(con
);
2946 if (!con
->error_msg
)
2947 con
->error_msg
= "socket error on read";
2952 ret
= try_write(con
);
2956 if (!con
->error_msg
)
2957 con
->error_msg
= "socket error on write";
2961 break; /* If we make it to here, we're done */
2965 mutex_unlock(&con
->mutex
);
2968 con_fault_finish(con
);
2974 * Generic error/fault handler. A retry mechanism is used with
2975 * exponential backoff
2977 static void con_fault(struct ceph_connection
*con
)
2979 dout("fault %p state %lu to peer %s\n",
2980 con
, con
->state
, ceph_pr_addr(&con
->peer_addr
.in_addr
));
2982 pr_warn("%s%lld %s %s\n", ENTITY_NAME(con
->peer_name
),
2983 ceph_pr_addr(&con
->peer_addr
.in_addr
), con
->error_msg
);
2984 con
->error_msg
= NULL
;
2986 WARN_ON(con
->state
!= CON_STATE_CONNECTING
&&
2987 con
->state
!= CON_STATE_NEGOTIATING
&&
2988 con
->state
!= CON_STATE_OPEN
);
2990 con_close_socket(con
);
2992 if (con_flag_test(con
, CON_FLAG_LOSSYTX
)) {
2993 dout("fault on LOSSYTX channel, marking CLOSED\n");
2994 con
->state
= CON_STATE_CLOSED
;
2999 BUG_ON(con
->in_msg
->con
!= con
);
3000 ceph_msg_put(con
->in_msg
);
3004 /* Requeue anything that hasn't been acked */
3005 list_splice_init(&con
->out_sent
, &con
->out_queue
);
3007 /* If there are no messages queued or keepalive pending, place
3008 * the connection in a STANDBY state */
3009 if (list_empty(&con
->out_queue
) &&
3010 !con_flag_test(con
, CON_FLAG_KEEPALIVE_PENDING
)) {
3011 dout("fault %p setting STANDBY clearing WRITE_PENDING\n", con
);
3012 con_flag_clear(con
, CON_FLAG_WRITE_PENDING
);
3013 con
->state
= CON_STATE_STANDBY
;
3015 /* retry after a delay. */
3016 con
->state
= CON_STATE_PREOPEN
;
3017 if (con
->delay
== 0)
3018 con
->delay
= BASE_DELAY_INTERVAL
;
3019 else if (con
->delay
< MAX_DELAY_INTERVAL
)
3021 con_flag_set(con
, CON_FLAG_BACKOFF
);
3029 * initialize a new messenger instance
3031 void ceph_messenger_init(struct ceph_messenger
*msgr
,
3032 struct ceph_entity_addr
*myaddr
)
3034 spin_lock_init(&msgr
->global_seq_lock
);
3037 msgr
->inst
.addr
= *myaddr
;
3039 /* select a random nonce */
3040 msgr
->inst
.addr
.type
= 0;
3041 get_random_bytes(&msgr
->inst
.addr
.nonce
, sizeof(msgr
->inst
.addr
.nonce
));
3042 encode_my_addr(msgr
);
3044 atomic_set(&msgr
->stopping
, 0);
3045 write_pnet(&msgr
->net
, get_net(current
->nsproxy
->net_ns
));
3047 dout("%s %p\n", __func__
, msgr
);
3049 EXPORT_SYMBOL(ceph_messenger_init
);
3051 void ceph_messenger_fini(struct ceph_messenger
*msgr
)
3053 put_net(read_pnet(&msgr
->net
));
3055 EXPORT_SYMBOL(ceph_messenger_fini
);
3057 static void msg_con_set(struct ceph_msg
*msg
, struct ceph_connection
*con
)
3060 msg
->con
->ops
->put(msg
->con
);
3062 msg
->con
= con
? con
->ops
->get(con
) : NULL
;
3063 BUG_ON(msg
->con
!= con
);
3066 static void clear_standby(struct ceph_connection
*con
)
3068 /* come back from STANDBY? */
3069 if (con
->state
== CON_STATE_STANDBY
) {
3070 dout("clear_standby %p and ++connect_seq\n", con
);
3071 con
->state
= CON_STATE_PREOPEN
;
3073 WARN_ON(con_flag_test(con
, CON_FLAG_WRITE_PENDING
));
3074 WARN_ON(con_flag_test(con
, CON_FLAG_KEEPALIVE_PENDING
));
3079 * Queue up an outgoing message on the given connection.
3081 void ceph_con_send(struct ceph_connection
*con
, struct ceph_msg
*msg
)
3084 msg
->hdr
.src
= con
->msgr
->inst
.name
;
3085 BUG_ON(msg
->front
.iov_len
!= le32_to_cpu(msg
->hdr
.front_len
));
3086 msg
->needs_out_seq
= true;
3088 mutex_lock(&con
->mutex
);
3090 if (con
->state
== CON_STATE_CLOSED
) {
3091 dout("con_send %p closed, dropping %p\n", con
, msg
);
3093 mutex_unlock(&con
->mutex
);
3097 msg_con_set(msg
, con
);
3099 BUG_ON(!list_empty(&msg
->list_head
));
3100 list_add_tail(&msg
->list_head
, &con
->out_queue
);
3101 dout("----- %p to %s%lld %d=%s len %d+%d+%d -----\n", msg
,
3102 ENTITY_NAME(con
->peer_name
), le16_to_cpu(msg
->hdr
.type
),
3103 ceph_msg_type_name(le16_to_cpu(msg
->hdr
.type
)),
3104 le32_to_cpu(msg
->hdr
.front_len
),
3105 le32_to_cpu(msg
->hdr
.middle_len
),
3106 le32_to_cpu(msg
->hdr
.data_len
));
3109 mutex_unlock(&con
->mutex
);
3111 /* if there wasn't anything waiting to send before, queue
3113 if (con_flag_test_and_set(con
, CON_FLAG_WRITE_PENDING
) == 0)
3116 EXPORT_SYMBOL(ceph_con_send
);
3119 * Revoke a message that was previously queued for send
3121 void ceph_msg_revoke(struct ceph_msg
*msg
)
3123 struct ceph_connection
*con
= msg
->con
;
3126 dout("%s msg %p null con\n", __func__
, msg
);
3127 return; /* Message not in our possession */
3130 mutex_lock(&con
->mutex
);
3131 if (!list_empty(&msg
->list_head
)) {
3132 dout("%s %p msg %p - was on queue\n", __func__
, con
, msg
);
3133 list_del_init(&msg
->list_head
);
3138 if (con
->out_msg
== msg
) {
3139 BUG_ON(con
->out_skip
);
3141 if (con
->out_msg_done
) {
3142 con
->out_skip
+= con_out_kvec_skip(con
);
3144 BUG_ON(!msg
->data_length
);
3145 con
->out_skip
+= sizeof_footer(con
);
3147 /* data, middle, front */
3148 if (msg
->data_length
)
3149 con
->out_skip
+= msg
->cursor
.total_resid
;
3151 con
->out_skip
+= con_out_kvec_skip(con
);
3152 con
->out_skip
+= con_out_kvec_skip(con
);
3154 dout("%s %p msg %p - was sending, will write %d skip %d\n",
3155 __func__
, con
, msg
, con
->out_kvec_bytes
, con
->out_skip
);
3157 con
->out_msg
= NULL
;
3161 mutex_unlock(&con
->mutex
);
3165 * Revoke a message that we may be reading data into
3167 void ceph_msg_revoke_incoming(struct ceph_msg
*msg
)
3169 struct ceph_connection
*con
= msg
->con
;
3172 dout("%s msg %p null con\n", __func__
, msg
);
3173 return; /* Message not in our possession */
3176 mutex_lock(&con
->mutex
);
3177 if (con
->in_msg
== msg
) {
3178 unsigned int front_len
= le32_to_cpu(con
->in_hdr
.front_len
);
3179 unsigned int middle_len
= le32_to_cpu(con
->in_hdr
.middle_len
);
3180 unsigned int data_len
= le32_to_cpu(con
->in_hdr
.data_len
);
3182 /* skip rest of message */
3183 dout("%s %p msg %p revoked\n", __func__
, con
, msg
);
3184 con
->in_base_pos
= con
->in_base_pos
-
3185 sizeof(struct ceph_msg_header
) -
3189 sizeof(struct ceph_msg_footer
);
3190 ceph_msg_put(con
->in_msg
);
3192 con
->in_tag
= CEPH_MSGR_TAG_READY
;
3195 dout("%s %p in_msg %p msg %p no-op\n",
3196 __func__
, con
, con
->in_msg
, msg
);
3198 mutex_unlock(&con
->mutex
);
3202 * Queue a keepalive byte to ensure the tcp connection is alive.
3204 void ceph_con_keepalive(struct ceph_connection
*con
)
3206 dout("con_keepalive %p\n", con
);
3207 mutex_lock(&con
->mutex
);
3209 mutex_unlock(&con
->mutex
);
3210 if (con_flag_test_and_set(con
, CON_FLAG_KEEPALIVE_PENDING
) == 0 &&
3211 con_flag_test_and_set(con
, CON_FLAG_WRITE_PENDING
) == 0)
3214 EXPORT_SYMBOL(ceph_con_keepalive
);
3216 bool ceph_con_keepalive_expired(struct ceph_connection
*con
,
3217 unsigned long interval
)
3220 (con
->peer_features
& CEPH_FEATURE_MSGR_KEEPALIVE2
)) {
3221 struct timespec64 now
;
3222 struct timespec64 ts
;
3223 ktime_get_real_ts64(&now
);
3224 jiffies_to_timespec64(interval
, &ts
);
3225 ts
= timespec64_add(con
->last_keepalive_ack
, ts
);
3226 return timespec64_compare(&now
, &ts
) >= 0;
3231 static struct ceph_msg_data
*ceph_msg_data_add(struct ceph_msg
*msg
)
3233 BUG_ON(msg
->num_data_items
>= msg
->max_data_items
);
3234 return &msg
->data
[msg
->num_data_items
++];
3237 static void ceph_msg_data_destroy(struct ceph_msg_data
*data
)
3239 if (data
->type
== CEPH_MSG_DATA_PAGELIST
)
3240 ceph_pagelist_release(data
->pagelist
);
3243 void ceph_msg_data_add_pages(struct ceph_msg
*msg
, struct page
**pages
,
3244 size_t length
, size_t alignment
)
3246 struct ceph_msg_data
*data
;
3251 data
= ceph_msg_data_add(msg
);
3252 data
->type
= CEPH_MSG_DATA_PAGES
;
3253 data
->pages
= pages
;
3254 data
->length
= length
;
3255 data
->alignment
= alignment
& ~PAGE_MASK
;
3257 msg
->data_length
+= length
;
3259 EXPORT_SYMBOL(ceph_msg_data_add_pages
);
3261 void ceph_msg_data_add_pagelist(struct ceph_msg
*msg
,
3262 struct ceph_pagelist
*pagelist
)
3264 struct ceph_msg_data
*data
;
3267 BUG_ON(!pagelist
->length
);
3269 data
= ceph_msg_data_add(msg
);
3270 data
->type
= CEPH_MSG_DATA_PAGELIST
;
3271 refcount_inc(&pagelist
->refcnt
);
3272 data
->pagelist
= pagelist
;
3274 msg
->data_length
+= pagelist
->length
;
3276 EXPORT_SYMBOL(ceph_msg_data_add_pagelist
);
3279 void ceph_msg_data_add_bio(struct ceph_msg
*msg
, struct ceph_bio_iter
*bio_pos
,
3282 struct ceph_msg_data
*data
;
3284 data
= ceph_msg_data_add(msg
);
3285 data
->type
= CEPH_MSG_DATA_BIO
;
3286 data
->bio_pos
= *bio_pos
;
3287 data
->bio_length
= length
;
3289 msg
->data_length
+= length
;
3291 EXPORT_SYMBOL(ceph_msg_data_add_bio
);
3292 #endif /* CONFIG_BLOCK */
3294 void ceph_msg_data_add_bvecs(struct ceph_msg
*msg
,
3295 struct ceph_bvec_iter
*bvec_pos
)
3297 struct ceph_msg_data
*data
;
3299 data
= ceph_msg_data_add(msg
);
3300 data
->type
= CEPH_MSG_DATA_BVECS
;
3301 data
->bvec_pos
= *bvec_pos
;
3303 msg
->data_length
+= bvec_pos
->iter
.bi_size
;
3305 EXPORT_SYMBOL(ceph_msg_data_add_bvecs
);
3308 * construct a new message with given type, size
3309 * the new msg has a ref count of 1.
3311 struct ceph_msg
*ceph_msg_new2(int type
, int front_len
, int max_data_items
,
3312 gfp_t flags
, bool can_fail
)
3316 m
= kmem_cache_zalloc(ceph_msg_cache
, flags
);
3320 m
->hdr
.type
= cpu_to_le16(type
);
3321 m
->hdr
.priority
= cpu_to_le16(CEPH_MSG_PRIO_DEFAULT
);
3322 m
->hdr
.front_len
= cpu_to_le32(front_len
);
3324 INIT_LIST_HEAD(&m
->list_head
);
3325 kref_init(&m
->kref
);
3329 m
->front
.iov_base
= ceph_kvmalloc(front_len
, flags
);
3330 if (m
->front
.iov_base
== NULL
) {
3331 dout("ceph_msg_new can't allocate %d bytes\n",
3336 m
->front
.iov_base
= NULL
;
3338 m
->front_alloc_len
= m
->front
.iov_len
= front_len
;
3340 if (max_data_items
) {
3341 m
->data
= kmalloc_array(max_data_items
, sizeof(*m
->data
),
3346 m
->max_data_items
= max_data_items
;
3349 dout("ceph_msg_new %p front %d\n", m
, front_len
);
3356 pr_err("msg_new can't create type %d front %d\n", type
,
3360 dout("msg_new can't create type %d front %d\n", type
,
3365 EXPORT_SYMBOL(ceph_msg_new2
);
3367 struct ceph_msg
*ceph_msg_new(int type
, int front_len
, gfp_t flags
,
3370 return ceph_msg_new2(type
, front_len
, 0, flags
, can_fail
);
3372 EXPORT_SYMBOL(ceph_msg_new
);
3375 * Allocate "middle" portion of a message, if it is needed and wasn't
3376 * allocated by alloc_msg. This allows us to read a small fixed-size
3377 * per-type header in the front and then gracefully fail (i.e.,
3378 * propagate the error to the caller based on info in the front) when
3379 * the middle is too large.
3381 static int ceph_alloc_middle(struct ceph_connection
*con
, struct ceph_msg
*msg
)
3383 int type
= le16_to_cpu(msg
->hdr
.type
);
3384 int middle_len
= le32_to_cpu(msg
->hdr
.middle_len
);
3386 dout("alloc_middle %p type %d %s middle_len %d\n", msg
, type
,
3387 ceph_msg_type_name(type
), middle_len
);
3388 BUG_ON(!middle_len
);
3389 BUG_ON(msg
->middle
);
3391 msg
->middle
= ceph_buffer_new(middle_len
, GFP_NOFS
);
3398 * Allocate a message for receiving an incoming message on a
3399 * connection, and save the result in con->in_msg. Uses the
3400 * connection's private alloc_msg op if available.
3402 * Returns 0 on success, or a negative error code.
3404 * On success, if we set *skip = 1:
3405 * - the next message should be skipped and ignored.
3406 * - con->in_msg == NULL
3407 * or if we set *skip = 0:
3408 * - con->in_msg is non-null.
3409 * On error (ENOMEM, EAGAIN, ...),
3410 * - con->in_msg == NULL
3412 static int ceph_con_in_msg_alloc(struct ceph_connection
*con
, int *skip
)
3414 struct ceph_msg_header
*hdr
= &con
->in_hdr
;
3415 int middle_len
= le32_to_cpu(hdr
->middle_len
);
3416 struct ceph_msg
*msg
;
3419 BUG_ON(con
->in_msg
!= NULL
);
3420 BUG_ON(!con
->ops
->alloc_msg
);
3422 mutex_unlock(&con
->mutex
);
3423 msg
= con
->ops
->alloc_msg(con
, hdr
, skip
);
3424 mutex_lock(&con
->mutex
);
3425 if (con
->state
!= CON_STATE_OPEN
) {
3432 msg_con_set(msg
, con
);
3436 * Null message pointer means either we should skip
3437 * this message or we couldn't allocate memory. The
3438 * former is not an error.
3443 con
->error_msg
= "error allocating memory for incoming message";
3446 memcpy(&con
->in_msg
->hdr
, &con
->in_hdr
, sizeof(con
->in_hdr
));
3448 if (middle_len
&& !con
->in_msg
->middle
) {
3449 ret
= ceph_alloc_middle(con
, con
->in_msg
);
3451 ceph_msg_put(con
->in_msg
);
3461 * Free a generically kmalloc'd message.
3463 static void ceph_msg_free(struct ceph_msg
*m
)
3465 dout("%s %p\n", __func__
, m
);
3466 kvfree(m
->front
.iov_base
);
3468 kmem_cache_free(ceph_msg_cache
, m
);
3471 static void ceph_msg_release(struct kref
*kref
)
3473 struct ceph_msg
*m
= container_of(kref
, struct ceph_msg
, kref
);
3476 dout("%s %p\n", __func__
, m
);
3477 WARN_ON(!list_empty(&m
->list_head
));
3479 msg_con_set(m
, NULL
);
3481 /* drop middle, data, if any */
3483 ceph_buffer_put(m
->middle
);
3487 for (i
= 0; i
< m
->num_data_items
; i
++)
3488 ceph_msg_data_destroy(&m
->data
[i
]);
3491 ceph_msgpool_put(m
->pool
, m
);
3496 struct ceph_msg
*ceph_msg_get(struct ceph_msg
*msg
)
3498 dout("%s %p (was %d)\n", __func__
, msg
,
3499 kref_read(&msg
->kref
));
3500 kref_get(&msg
->kref
);
3503 EXPORT_SYMBOL(ceph_msg_get
);
3505 void ceph_msg_put(struct ceph_msg
*msg
)
3507 dout("%s %p (was %d)\n", __func__
, msg
,
3508 kref_read(&msg
->kref
));
3509 kref_put(&msg
->kref
, ceph_msg_release
);
3511 EXPORT_SYMBOL(ceph_msg_put
);
3513 void ceph_msg_dump(struct ceph_msg
*msg
)
3515 pr_debug("msg_dump %p (front_alloc_len %d length %zd)\n", msg
,
3516 msg
->front_alloc_len
, msg
->data_length
);
3517 print_hex_dump(KERN_DEBUG
, "header: ",
3518 DUMP_PREFIX_OFFSET
, 16, 1,
3519 &msg
->hdr
, sizeof(msg
->hdr
), true);
3520 print_hex_dump(KERN_DEBUG
, " front: ",
3521 DUMP_PREFIX_OFFSET
, 16, 1,
3522 msg
->front
.iov_base
, msg
->front
.iov_len
, true);
3524 print_hex_dump(KERN_DEBUG
, "middle: ",
3525 DUMP_PREFIX_OFFSET
, 16, 1,
3526 msg
->middle
->vec
.iov_base
,
3527 msg
->middle
->vec
.iov_len
, true);
3528 print_hex_dump(KERN_DEBUG
, "footer: ",
3529 DUMP_PREFIX_OFFSET
, 16, 1,
3530 &msg
->footer
, sizeof(msg
->footer
), true);
3532 EXPORT_SYMBOL(ceph_msg_dump
);